]>
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> |
f361bf4a | 19 | #include <linux/sched/signal.h> |
edbd8d4e | 20 | #include <linux/pagemap.h> |
ec44a35c | 21 | #include <linux/writeback.h> |
21af804c | 22 | #include <linux/blkdev.h> |
b7a9f29f | 23 | #include <linux/sort.h> |
4184ea7f | 24 | #include <linux/rcupdate.h> |
817d52f8 | 25 | #include <linux/kthread.h> |
5a0e3ad6 | 26 | #include <linux/slab.h> |
dff51cd1 | 27 | #include <linux/ratelimit.h> |
b150a4f1 | 28 | #include <linux/percpu_counter.h> |
74493f7a | 29 | #include "hash.h" |
995946dd | 30 | #include "tree-log.h" |
fec577fb CM |
31 | #include "disk-io.h" |
32 | #include "print-tree.h" | |
0b86a832 | 33 | #include "volumes.h" |
53b381b3 | 34 | #include "raid56.h" |
925baedd | 35 | #include "locking.h" |
fa9c0d79 | 36 | #include "free-space-cache.h" |
1e144fb8 | 37 | #include "free-space-tree.h" |
3fed40cc | 38 | #include "math.h" |
6ab0a202 | 39 | #include "sysfs.h" |
fcebe456 | 40 | #include "qgroup.h" |
fec577fb | 41 | |
709c0486 AJ |
42 | #undef SCRAMBLE_DELAYED_REFS |
43 | ||
9e622d6b MX |
44 | /* |
45 | * control flags for do_chunk_alloc's force field | |
0e4f8f88 CM |
46 | * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk |
47 | * if we really need one. | |
48 | * | |
0e4f8f88 CM |
49 | * CHUNK_ALLOC_LIMITED means to only try and allocate one |
50 | * if we have very few chunks already allocated. This is | |
51 | * used as part of the clustering code to help make sure | |
52 | * we have a good pool of storage to cluster in, without | |
53 | * filling the FS with empty chunks | |
54 | * | |
9e622d6b MX |
55 | * CHUNK_ALLOC_FORCE means it must try to allocate one |
56 | * | |
0e4f8f88 CM |
57 | */ |
58 | enum { | |
59 | CHUNK_ALLOC_NO_FORCE = 0, | |
9e622d6b MX |
60 | CHUNK_ALLOC_LIMITED = 1, |
61 | CHUNK_ALLOC_FORCE = 2, | |
0e4f8f88 CM |
62 | }; |
63 | ||
ce93ec54 | 64 | static int update_block_group(struct btrfs_trans_handle *trans, |
6202df69 | 65 | struct btrfs_fs_info *fs_info, u64 bytenr, |
ce93ec54 | 66 | u64 num_bytes, int alloc); |
5d4f98a2 | 67 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
2ff7e61e | 68 | struct btrfs_fs_info *fs_info, |
c682f9b3 | 69 | struct btrfs_delayed_ref_node *node, u64 parent, |
5d4f98a2 YZ |
70 | u64 root_objectid, u64 owner_objectid, |
71 | u64 owner_offset, int refs_to_drop, | |
c682f9b3 | 72 | struct btrfs_delayed_extent_op *extra_op); |
5d4f98a2 YZ |
73 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, |
74 | struct extent_buffer *leaf, | |
75 | struct btrfs_extent_item *ei); | |
76 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
2ff7e61e | 77 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
78 | u64 parent, u64 root_objectid, |
79 | u64 flags, u64 owner, u64 offset, | |
80 | struct btrfs_key *ins, int ref_mod); | |
81 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, | |
2ff7e61e | 82 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
83 | u64 parent, u64 root_objectid, |
84 | u64 flags, struct btrfs_disk_key *key, | |
b06c4bf5 | 85 | int level, struct btrfs_key *ins); |
6a63209f | 86 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
2ff7e61e | 87 | struct btrfs_fs_info *fs_info, u64 flags, |
698d0082 | 88 | int force); |
11833d66 YZ |
89 | static int find_next_key(struct btrfs_path *path, int level, |
90 | struct btrfs_key *key); | |
ab8d0fc4 JM |
91 | static void dump_space_info(struct btrfs_fs_info *fs_info, |
92 | struct btrfs_space_info *info, u64 bytes, | |
9ed74f2d | 93 | int dump_block_groups); |
4824f1f4 | 94 | static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache *cache, |
18513091 | 95 | u64 ram_bytes, u64 num_bytes, int delalloc); |
4824f1f4 WX |
96 | static int btrfs_free_reserved_bytes(struct btrfs_block_group_cache *cache, |
97 | u64 num_bytes, int delalloc); | |
5d80366e JB |
98 | static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, |
99 | u64 num_bytes); | |
c1c4919b | 100 | static int __reserve_metadata_bytes(struct btrfs_fs_info *fs_info, |
957780eb JB |
101 | struct btrfs_space_info *space_info, |
102 | u64 orig_bytes, | |
c1c4919b JM |
103 | enum btrfs_reserve_flush_enum flush, |
104 | bool system_chunk); | |
957780eb JB |
105 | static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info, |
106 | struct btrfs_space_info *space_info, | |
107 | u64 num_bytes); | |
108 | static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info, | |
109 | struct btrfs_space_info *space_info, | |
110 | u64 num_bytes); | |
6a63209f | 111 | |
817d52f8 JB |
112 | static noinline int |
113 | block_group_cache_done(struct btrfs_block_group_cache *cache) | |
114 | { | |
115 | smp_mb(); | |
36cce922 JB |
116 | return cache->cached == BTRFS_CACHE_FINISHED || |
117 | cache->cached == BTRFS_CACHE_ERROR; | |
817d52f8 JB |
118 | } |
119 | ||
0f9dd46c JB |
120 | static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits) |
121 | { | |
122 | return (cache->flags & bits) == bits; | |
123 | } | |
124 | ||
758f2dfc | 125 | void btrfs_get_block_group(struct btrfs_block_group_cache *cache) |
11dfe35a JB |
126 | { |
127 | atomic_inc(&cache->count); | |
128 | } | |
129 | ||
130 | void btrfs_put_block_group(struct btrfs_block_group_cache *cache) | |
131 | { | |
f0486c68 YZ |
132 | if (atomic_dec_and_test(&cache->count)) { |
133 | WARN_ON(cache->pinned > 0); | |
134 | WARN_ON(cache->reserved > 0); | |
0966a7b1 QW |
135 | |
136 | /* | |
137 | * If not empty, someone is still holding mutex of | |
138 | * full_stripe_lock, which can only be released by caller. | |
139 | * And it will definitely cause use-after-free when caller | |
140 | * tries to release full stripe lock. | |
141 | * | |
142 | * No better way to resolve, but only to warn. | |
143 | */ | |
144 | WARN_ON(!RB_EMPTY_ROOT(&cache->full_stripe_locks_root.root)); | |
34d52cb6 | 145 | kfree(cache->free_space_ctl); |
11dfe35a | 146 | kfree(cache); |
f0486c68 | 147 | } |
11dfe35a JB |
148 | } |
149 | ||
0f9dd46c JB |
150 | /* |
151 | * this adds the block group to the fs_info rb tree for the block group | |
152 | * cache | |
153 | */ | |
b2950863 | 154 | static int btrfs_add_block_group_cache(struct btrfs_fs_info *info, |
0f9dd46c JB |
155 | struct btrfs_block_group_cache *block_group) |
156 | { | |
157 | struct rb_node **p; | |
158 | struct rb_node *parent = NULL; | |
159 | struct btrfs_block_group_cache *cache; | |
160 | ||
161 | spin_lock(&info->block_group_cache_lock); | |
162 | p = &info->block_group_cache_tree.rb_node; | |
163 | ||
164 | while (*p) { | |
165 | parent = *p; | |
166 | cache = rb_entry(parent, struct btrfs_block_group_cache, | |
167 | cache_node); | |
168 | if (block_group->key.objectid < cache->key.objectid) { | |
169 | p = &(*p)->rb_left; | |
170 | } else if (block_group->key.objectid > cache->key.objectid) { | |
171 | p = &(*p)->rb_right; | |
172 | } else { | |
173 | spin_unlock(&info->block_group_cache_lock); | |
174 | return -EEXIST; | |
175 | } | |
176 | } | |
177 | ||
178 | rb_link_node(&block_group->cache_node, parent, p); | |
179 | rb_insert_color(&block_group->cache_node, | |
180 | &info->block_group_cache_tree); | |
a1897fdd LB |
181 | |
182 | if (info->first_logical_byte > block_group->key.objectid) | |
183 | info->first_logical_byte = block_group->key.objectid; | |
184 | ||
0f9dd46c JB |
185 | spin_unlock(&info->block_group_cache_lock); |
186 | ||
187 | return 0; | |
188 | } | |
189 | ||
190 | /* | |
191 | * This will return the block group at or after bytenr if contains is 0, else | |
192 | * it will return the block group that contains the bytenr | |
193 | */ | |
194 | static struct btrfs_block_group_cache * | |
195 | block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr, | |
196 | int contains) | |
197 | { | |
198 | struct btrfs_block_group_cache *cache, *ret = NULL; | |
199 | struct rb_node *n; | |
200 | u64 end, start; | |
201 | ||
202 | spin_lock(&info->block_group_cache_lock); | |
203 | n = info->block_group_cache_tree.rb_node; | |
204 | ||
205 | while (n) { | |
206 | cache = rb_entry(n, struct btrfs_block_group_cache, | |
207 | cache_node); | |
208 | end = cache->key.objectid + cache->key.offset - 1; | |
209 | start = cache->key.objectid; | |
210 | ||
211 | if (bytenr < start) { | |
212 | if (!contains && (!ret || start < ret->key.objectid)) | |
213 | ret = cache; | |
214 | n = n->rb_left; | |
215 | } else if (bytenr > start) { | |
216 | if (contains && bytenr <= end) { | |
217 | ret = cache; | |
218 | break; | |
219 | } | |
220 | n = n->rb_right; | |
221 | } else { | |
222 | ret = cache; | |
223 | break; | |
224 | } | |
225 | } | |
a1897fdd | 226 | if (ret) { |
11dfe35a | 227 | btrfs_get_block_group(ret); |
a1897fdd LB |
228 | if (bytenr == 0 && info->first_logical_byte > ret->key.objectid) |
229 | info->first_logical_byte = ret->key.objectid; | |
230 | } | |
0f9dd46c JB |
231 | spin_unlock(&info->block_group_cache_lock); |
232 | ||
233 | return ret; | |
234 | } | |
235 | ||
2ff7e61e | 236 | static int add_excluded_extent(struct btrfs_fs_info *fs_info, |
11833d66 | 237 | u64 start, u64 num_bytes) |
817d52f8 | 238 | { |
11833d66 | 239 | u64 end = start + num_bytes - 1; |
0b246afa | 240 | set_extent_bits(&fs_info->freed_extents[0], |
ceeb0ae7 | 241 | start, end, EXTENT_UPTODATE); |
0b246afa | 242 | set_extent_bits(&fs_info->freed_extents[1], |
ceeb0ae7 | 243 | start, end, EXTENT_UPTODATE); |
11833d66 YZ |
244 | return 0; |
245 | } | |
817d52f8 | 246 | |
2ff7e61e | 247 | static void free_excluded_extents(struct btrfs_fs_info *fs_info, |
11833d66 YZ |
248 | struct btrfs_block_group_cache *cache) |
249 | { | |
250 | u64 start, end; | |
817d52f8 | 251 | |
11833d66 YZ |
252 | start = cache->key.objectid; |
253 | end = start + cache->key.offset - 1; | |
254 | ||
0b246afa | 255 | clear_extent_bits(&fs_info->freed_extents[0], |
91166212 | 256 | start, end, EXTENT_UPTODATE); |
0b246afa | 257 | clear_extent_bits(&fs_info->freed_extents[1], |
91166212 | 258 | start, end, EXTENT_UPTODATE); |
817d52f8 JB |
259 | } |
260 | ||
2ff7e61e | 261 | static int exclude_super_stripes(struct btrfs_fs_info *fs_info, |
11833d66 | 262 | struct btrfs_block_group_cache *cache) |
817d52f8 | 263 | { |
817d52f8 JB |
264 | u64 bytenr; |
265 | u64 *logical; | |
266 | int stripe_len; | |
267 | int i, nr, ret; | |
268 | ||
06b2331f YZ |
269 | if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) { |
270 | stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid; | |
271 | cache->bytes_super += stripe_len; | |
2ff7e61e | 272 | ret = add_excluded_extent(fs_info, cache->key.objectid, |
06b2331f | 273 | stripe_len); |
835d974f JB |
274 | if (ret) |
275 | return ret; | |
06b2331f YZ |
276 | } |
277 | ||
817d52f8 JB |
278 | for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { |
279 | bytenr = btrfs_sb_offset(i); | |
0b246afa | 280 | ret = btrfs_rmap_block(fs_info, cache->key.objectid, |
ab8d0fc4 | 281 | bytenr, 0, &logical, &nr, &stripe_len); |
835d974f JB |
282 | if (ret) |
283 | return ret; | |
11833d66 | 284 | |
817d52f8 | 285 | while (nr--) { |
51bf5f0b JB |
286 | u64 start, len; |
287 | ||
288 | if (logical[nr] > cache->key.objectid + | |
289 | cache->key.offset) | |
290 | continue; | |
291 | ||
292 | if (logical[nr] + stripe_len <= cache->key.objectid) | |
293 | continue; | |
294 | ||
295 | start = logical[nr]; | |
296 | if (start < cache->key.objectid) { | |
297 | start = cache->key.objectid; | |
298 | len = (logical[nr] + stripe_len) - start; | |
299 | } else { | |
300 | len = min_t(u64, stripe_len, | |
301 | cache->key.objectid + | |
302 | cache->key.offset - start); | |
303 | } | |
304 | ||
305 | cache->bytes_super += len; | |
2ff7e61e | 306 | ret = add_excluded_extent(fs_info, start, len); |
835d974f JB |
307 | if (ret) { |
308 | kfree(logical); | |
309 | return ret; | |
310 | } | |
817d52f8 | 311 | } |
11833d66 | 312 | |
817d52f8 JB |
313 | kfree(logical); |
314 | } | |
817d52f8 JB |
315 | return 0; |
316 | } | |
317 | ||
11833d66 YZ |
318 | static struct btrfs_caching_control * |
319 | get_caching_control(struct btrfs_block_group_cache *cache) | |
320 | { | |
321 | struct btrfs_caching_control *ctl; | |
322 | ||
323 | spin_lock(&cache->lock); | |
dde5abee JB |
324 | if (!cache->caching_ctl) { |
325 | spin_unlock(&cache->lock); | |
11833d66 YZ |
326 | return NULL; |
327 | } | |
328 | ||
329 | ctl = cache->caching_ctl; | |
1e4f4714 | 330 | refcount_inc(&ctl->count); |
11833d66 YZ |
331 | spin_unlock(&cache->lock); |
332 | return ctl; | |
333 | } | |
334 | ||
335 | static void put_caching_control(struct btrfs_caching_control *ctl) | |
336 | { | |
1e4f4714 | 337 | if (refcount_dec_and_test(&ctl->count)) |
11833d66 YZ |
338 | kfree(ctl); |
339 | } | |
340 | ||
d0bd4560 | 341 | #ifdef CONFIG_BTRFS_DEBUG |
2ff7e61e | 342 | static void fragment_free_space(struct btrfs_block_group_cache *block_group) |
d0bd4560 | 343 | { |
2ff7e61e | 344 | struct btrfs_fs_info *fs_info = block_group->fs_info; |
d0bd4560 JB |
345 | u64 start = block_group->key.objectid; |
346 | u64 len = block_group->key.offset; | |
347 | u64 chunk = block_group->flags & BTRFS_BLOCK_GROUP_METADATA ? | |
0b246afa | 348 | fs_info->nodesize : fs_info->sectorsize; |
d0bd4560 JB |
349 | u64 step = chunk << 1; |
350 | ||
351 | while (len > chunk) { | |
352 | btrfs_remove_free_space(block_group, start, chunk); | |
353 | start += step; | |
354 | if (len < step) | |
355 | len = 0; | |
356 | else | |
357 | len -= step; | |
358 | } | |
359 | } | |
360 | #endif | |
361 | ||
0f9dd46c JB |
362 | /* |
363 | * this is only called by cache_block_group, since we could have freed extents | |
364 | * we need to check the pinned_extents for any extents that can't be used yet | |
365 | * since their free space will be released as soon as the transaction commits. | |
366 | */ | |
a5ed9182 OS |
367 | u64 add_new_free_space(struct btrfs_block_group_cache *block_group, |
368 | struct btrfs_fs_info *info, u64 start, u64 end) | |
0f9dd46c | 369 | { |
817d52f8 | 370 | u64 extent_start, extent_end, size, total_added = 0; |
0f9dd46c JB |
371 | int ret; |
372 | ||
373 | while (start < end) { | |
11833d66 | 374 | ret = find_first_extent_bit(info->pinned_extents, start, |
0f9dd46c | 375 | &extent_start, &extent_end, |
e6138876 JB |
376 | EXTENT_DIRTY | EXTENT_UPTODATE, |
377 | NULL); | |
0f9dd46c JB |
378 | if (ret) |
379 | break; | |
380 | ||
06b2331f | 381 | if (extent_start <= start) { |
0f9dd46c JB |
382 | start = extent_end + 1; |
383 | } else if (extent_start > start && extent_start < end) { | |
384 | size = extent_start - start; | |
817d52f8 | 385 | total_added += size; |
ea6a478e JB |
386 | ret = btrfs_add_free_space(block_group, start, |
387 | size); | |
79787eaa | 388 | BUG_ON(ret); /* -ENOMEM or logic error */ |
0f9dd46c JB |
389 | start = extent_end + 1; |
390 | } else { | |
391 | break; | |
392 | } | |
393 | } | |
394 | ||
395 | if (start < end) { | |
396 | size = end - start; | |
817d52f8 | 397 | total_added += size; |
ea6a478e | 398 | ret = btrfs_add_free_space(block_group, start, size); |
79787eaa | 399 | BUG_ON(ret); /* -ENOMEM or logic error */ |
0f9dd46c JB |
400 | } |
401 | ||
817d52f8 | 402 | return total_added; |
0f9dd46c JB |
403 | } |
404 | ||
73fa48b6 | 405 | static int load_extent_tree_free(struct btrfs_caching_control *caching_ctl) |
e37c9e69 | 406 | { |
0b246afa JM |
407 | struct btrfs_block_group_cache *block_group = caching_ctl->block_group; |
408 | struct btrfs_fs_info *fs_info = block_group->fs_info; | |
409 | struct btrfs_root *extent_root = fs_info->extent_root; | |
e37c9e69 | 410 | struct btrfs_path *path; |
5f39d397 | 411 | struct extent_buffer *leaf; |
11833d66 | 412 | struct btrfs_key key; |
817d52f8 | 413 | u64 total_found = 0; |
11833d66 YZ |
414 | u64 last = 0; |
415 | u32 nritems; | |
73fa48b6 | 416 | int ret; |
d0bd4560 | 417 | bool wakeup = true; |
f510cfec | 418 | |
e37c9e69 CM |
419 | path = btrfs_alloc_path(); |
420 | if (!path) | |
73fa48b6 | 421 | return -ENOMEM; |
7d7d6068 | 422 | |
817d52f8 | 423 | last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET); |
11833d66 | 424 | |
d0bd4560 JB |
425 | #ifdef CONFIG_BTRFS_DEBUG |
426 | /* | |
427 | * If we're fragmenting we don't want to make anybody think we can | |
428 | * allocate from this block group until we've had a chance to fragment | |
429 | * the free space. | |
430 | */ | |
2ff7e61e | 431 | if (btrfs_should_fragment_free_space(block_group)) |
d0bd4560 JB |
432 | wakeup = false; |
433 | #endif | |
5cd57b2c | 434 | /* |
817d52f8 JB |
435 | * We don't want to deadlock with somebody trying to allocate a new |
436 | * extent for the extent root while also trying to search the extent | |
437 | * root to add free space. So we skip locking and search the commit | |
438 | * root, since its read-only | |
5cd57b2c CM |
439 | */ |
440 | path->skip_locking = 1; | |
817d52f8 | 441 | path->search_commit_root = 1; |
e4058b54 | 442 | path->reada = READA_FORWARD; |
817d52f8 | 443 | |
e4404d6e | 444 | key.objectid = last; |
e37c9e69 | 445 | key.offset = 0; |
11833d66 | 446 | key.type = BTRFS_EXTENT_ITEM_KEY; |
013f1b12 | 447 | |
52ee28d2 | 448 | next: |
11833d66 | 449 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
e37c9e69 | 450 | if (ret < 0) |
73fa48b6 | 451 | goto out; |
a512bbf8 | 452 | |
11833d66 YZ |
453 | leaf = path->nodes[0]; |
454 | nritems = btrfs_header_nritems(leaf); | |
455 | ||
d397712b | 456 | while (1) { |
7841cb28 | 457 | if (btrfs_fs_closing(fs_info) > 1) { |
f25784b3 | 458 | last = (u64)-1; |
817d52f8 | 459 | break; |
f25784b3 | 460 | } |
817d52f8 | 461 | |
11833d66 YZ |
462 | if (path->slots[0] < nritems) { |
463 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
464 | } else { | |
465 | ret = find_next_key(path, 0, &key); | |
466 | if (ret) | |
e37c9e69 | 467 | break; |
817d52f8 | 468 | |
c9ea7b24 | 469 | if (need_resched() || |
9e351cc8 | 470 | rwsem_is_contended(&fs_info->commit_root_sem)) { |
d0bd4560 JB |
471 | if (wakeup) |
472 | caching_ctl->progress = last; | |
ff5714cc | 473 | btrfs_release_path(path); |
9e351cc8 | 474 | up_read(&fs_info->commit_root_sem); |
589d8ade | 475 | mutex_unlock(&caching_ctl->mutex); |
11833d66 | 476 | cond_resched(); |
73fa48b6 OS |
477 | mutex_lock(&caching_ctl->mutex); |
478 | down_read(&fs_info->commit_root_sem); | |
479 | goto next; | |
589d8ade | 480 | } |
0a3896d0 JB |
481 | |
482 | ret = btrfs_next_leaf(extent_root, path); | |
483 | if (ret < 0) | |
73fa48b6 | 484 | goto out; |
0a3896d0 JB |
485 | if (ret) |
486 | break; | |
589d8ade JB |
487 | leaf = path->nodes[0]; |
488 | nritems = btrfs_header_nritems(leaf); | |
489 | continue; | |
11833d66 | 490 | } |
817d52f8 | 491 | |
52ee28d2 LB |
492 | if (key.objectid < last) { |
493 | key.objectid = last; | |
494 | key.offset = 0; | |
495 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
496 | ||
d0bd4560 JB |
497 | if (wakeup) |
498 | caching_ctl->progress = last; | |
52ee28d2 LB |
499 | btrfs_release_path(path); |
500 | goto next; | |
501 | } | |
502 | ||
11833d66 YZ |
503 | if (key.objectid < block_group->key.objectid) { |
504 | path->slots[0]++; | |
817d52f8 | 505 | continue; |
e37c9e69 | 506 | } |
0f9dd46c | 507 | |
e37c9e69 | 508 | if (key.objectid >= block_group->key.objectid + |
0f9dd46c | 509 | block_group->key.offset) |
e37c9e69 | 510 | break; |
7d7d6068 | 511 | |
3173a18f JB |
512 | if (key.type == BTRFS_EXTENT_ITEM_KEY || |
513 | key.type == BTRFS_METADATA_ITEM_KEY) { | |
817d52f8 JB |
514 | total_found += add_new_free_space(block_group, |
515 | fs_info, last, | |
516 | key.objectid); | |
3173a18f JB |
517 | if (key.type == BTRFS_METADATA_ITEM_KEY) |
518 | last = key.objectid + | |
da17066c | 519 | fs_info->nodesize; |
3173a18f JB |
520 | else |
521 | last = key.objectid + key.offset; | |
817d52f8 | 522 | |
73fa48b6 | 523 | if (total_found > CACHING_CTL_WAKE_UP) { |
11833d66 | 524 | total_found = 0; |
d0bd4560 JB |
525 | if (wakeup) |
526 | wake_up(&caching_ctl->wait); | |
11833d66 | 527 | } |
817d52f8 | 528 | } |
e37c9e69 CM |
529 | path->slots[0]++; |
530 | } | |
817d52f8 | 531 | ret = 0; |
e37c9e69 | 532 | |
817d52f8 JB |
533 | total_found += add_new_free_space(block_group, fs_info, last, |
534 | block_group->key.objectid + | |
535 | block_group->key.offset); | |
11833d66 | 536 | caching_ctl->progress = (u64)-1; |
817d52f8 | 537 | |
73fa48b6 OS |
538 | out: |
539 | btrfs_free_path(path); | |
540 | return ret; | |
541 | } | |
542 | ||
543 | static noinline void caching_thread(struct btrfs_work *work) | |
544 | { | |
545 | struct btrfs_block_group_cache *block_group; | |
546 | struct btrfs_fs_info *fs_info; | |
547 | struct btrfs_caching_control *caching_ctl; | |
b4570aa9 | 548 | struct btrfs_root *extent_root; |
73fa48b6 OS |
549 | int ret; |
550 | ||
551 | caching_ctl = container_of(work, struct btrfs_caching_control, work); | |
552 | block_group = caching_ctl->block_group; | |
553 | fs_info = block_group->fs_info; | |
b4570aa9 | 554 | extent_root = fs_info->extent_root; |
73fa48b6 OS |
555 | |
556 | mutex_lock(&caching_ctl->mutex); | |
557 | down_read(&fs_info->commit_root_sem); | |
558 | ||
1e144fb8 OS |
559 | if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) |
560 | ret = load_free_space_tree(caching_ctl); | |
561 | else | |
562 | ret = load_extent_tree_free(caching_ctl); | |
73fa48b6 | 563 | |
817d52f8 | 564 | spin_lock(&block_group->lock); |
11833d66 | 565 | block_group->caching_ctl = NULL; |
73fa48b6 | 566 | block_group->cached = ret ? BTRFS_CACHE_ERROR : BTRFS_CACHE_FINISHED; |
817d52f8 | 567 | spin_unlock(&block_group->lock); |
0f9dd46c | 568 | |
d0bd4560 | 569 | #ifdef CONFIG_BTRFS_DEBUG |
2ff7e61e | 570 | if (btrfs_should_fragment_free_space(block_group)) { |
d0bd4560 JB |
571 | u64 bytes_used; |
572 | ||
573 | spin_lock(&block_group->space_info->lock); | |
574 | spin_lock(&block_group->lock); | |
575 | bytes_used = block_group->key.offset - | |
576 | btrfs_block_group_used(&block_group->item); | |
577 | block_group->space_info->bytes_used += bytes_used >> 1; | |
578 | spin_unlock(&block_group->lock); | |
579 | spin_unlock(&block_group->space_info->lock); | |
2ff7e61e | 580 | fragment_free_space(block_group); |
d0bd4560 JB |
581 | } |
582 | #endif | |
583 | ||
584 | caching_ctl->progress = (u64)-1; | |
11833d66 | 585 | |
9e351cc8 | 586 | up_read(&fs_info->commit_root_sem); |
2ff7e61e | 587 | free_excluded_extents(fs_info, block_group); |
11833d66 | 588 | mutex_unlock(&caching_ctl->mutex); |
73fa48b6 | 589 | |
11833d66 YZ |
590 | wake_up(&caching_ctl->wait); |
591 | ||
592 | put_caching_control(caching_ctl); | |
11dfe35a | 593 | btrfs_put_block_group(block_group); |
817d52f8 JB |
594 | } |
595 | ||
9d66e233 | 596 | static int cache_block_group(struct btrfs_block_group_cache *cache, |
9d66e233 | 597 | int load_cache_only) |
817d52f8 | 598 | { |
291c7d2f | 599 | DEFINE_WAIT(wait); |
11833d66 YZ |
600 | struct btrfs_fs_info *fs_info = cache->fs_info; |
601 | struct btrfs_caching_control *caching_ctl; | |
817d52f8 JB |
602 | int ret = 0; |
603 | ||
291c7d2f | 604 | caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS); |
79787eaa JM |
605 | if (!caching_ctl) |
606 | return -ENOMEM; | |
291c7d2f JB |
607 | |
608 | INIT_LIST_HEAD(&caching_ctl->list); | |
609 | mutex_init(&caching_ctl->mutex); | |
610 | init_waitqueue_head(&caching_ctl->wait); | |
611 | caching_ctl->block_group = cache; | |
612 | caching_ctl->progress = cache->key.objectid; | |
1e4f4714 | 613 | refcount_set(&caching_ctl->count, 1); |
9e0af237 LB |
614 | btrfs_init_work(&caching_ctl->work, btrfs_cache_helper, |
615 | caching_thread, NULL, NULL); | |
291c7d2f JB |
616 | |
617 | spin_lock(&cache->lock); | |
618 | /* | |
619 | * This should be a rare occasion, but this could happen I think in the | |
620 | * case where one thread starts to load the space cache info, and then | |
621 | * some other thread starts a transaction commit which tries to do an | |
622 | * allocation while the other thread is still loading the space cache | |
623 | * info. The previous loop should have kept us from choosing this block | |
624 | * group, but if we've moved to the state where we will wait on caching | |
625 | * block groups we need to first check if we're doing a fast load here, | |
626 | * so we can wait for it to finish, otherwise we could end up allocating | |
627 | * from a block group who's cache gets evicted for one reason or | |
628 | * another. | |
629 | */ | |
630 | while (cache->cached == BTRFS_CACHE_FAST) { | |
631 | struct btrfs_caching_control *ctl; | |
632 | ||
633 | ctl = cache->caching_ctl; | |
1e4f4714 | 634 | refcount_inc(&ctl->count); |
291c7d2f JB |
635 | prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE); |
636 | spin_unlock(&cache->lock); | |
637 | ||
638 | schedule(); | |
639 | ||
640 | finish_wait(&ctl->wait, &wait); | |
641 | put_caching_control(ctl); | |
642 | spin_lock(&cache->lock); | |
643 | } | |
644 | ||
645 | if (cache->cached != BTRFS_CACHE_NO) { | |
646 | spin_unlock(&cache->lock); | |
647 | kfree(caching_ctl); | |
11833d66 | 648 | return 0; |
291c7d2f JB |
649 | } |
650 | WARN_ON(cache->caching_ctl); | |
651 | cache->caching_ctl = caching_ctl; | |
652 | cache->cached = BTRFS_CACHE_FAST; | |
653 | spin_unlock(&cache->lock); | |
11833d66 | 654 | |
d53ba474 | 655 | if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) { |
cb83b7b8 | 656 | mutex_lock(&caching_ctl->mutex); |
9d66e233 JB |
657 | ret = load_free_space_cache(fs_info, cache); |
658 | ||
659 | spin_lock(&cache->lock); | |
660 | if (ret == 1) { | |
291c7d2f | 661 | cache->caching_ctl = NULL; |
9d66e233 JB |
662 | cache->cached = BTRFS_CACHE_FINISHED; |
663 | cache->last_byte_to_unpin = (u64)-1; | |
cb83b7b8 | 664 | caching_ctl->progress = (u64)-1; |
9d66e233 | 665 | } else { |
291c7d2f JB |
666 | if (load_cache_only) { |
667 | cache->caching_ctl = NULL; | |
668 | cache->cached = BTRFS_CACHE_NO; | |
669 | } else { | |
670 | cache->cached = BTRFS_CACHE_STARTED; | |
4f69cb98 | 671 | cache->has_caching_ctl = 1; |
291c7d2f | 672 | } |
9d66e233 JB |
673 | } |
674 | spin_unlock(&cache->lock); | |
d0bd4560 JB |
675 | #ifdef CONFIG_BTRFS_DEBUG |
676 | if (ret == 1 && | |
2ff7e61e | 677 | btrfs_should_fragment_free_space(cache)) { |
d0bd4560 JB |
678 | u64 bytes_used; |
679 | ||
680 | spin_lock(&cache->space_info->lock); | |
681 | spin_lock(&cache->lock); | |
682 | bytes_used = cache->key.offset - | |
683 | btrfs_block_group_used(&cache->item); | |
684 | cache->space_info->bytes_used += bytes_used >> 1; | |
685 | spin_unlock(&cache->lock); | |
686 | spin_unlock(&cache->space_info->lock); | |
2ff7e61e | 687 | fragment_free_space(cache); |
d0bd4560 JB |
688 | } |
689 | #endif | |
cb83b7b8 JB |
690 | mutex_unlock(&caching_ctl->mutex); |
691 | ||
291c7d2f | 692 | wake_up(&caching_ctl->wait); |
3c14874a | 693 | if (ret == 1) { |
291c7d2f | 694 | put_caching_control(caching_ctl); |
2ff7e61e | 695 | free_excluded_extents(fs_info, cache); |
9d66e233 | 696 | return 0; |
3c14874a | 697 | } |
291c7d2f JB |
698 | } else { |
699 | /* | |
1e144fb8 OS |
700 | * We're either using the free space tree or no caching at all. |
701 | * Set cached to the appropriate value and wakeup any waiters. | |
291c7d2f JB |
702 | */ |
703 | spin_lock(&cache->lock); | |
704 | if (load_cache_only) { | |
705 | cache->caching_ctl = NULL; | |
706 | cache->cached = BTRFS_CACHE_NO; | |
707 | } else { | |
708 | cache->cached = BTRFS_CACHE_STARTED; | |
4f69cb98 | 709 | cache->has_caching_ctl = 1; |
291c7d2f JB |
710 | } |
711 | spin_unlock(&cache->lock); | |
712 | wake_up(&caching_ctl->wait); | |
9d66e233 JB |
713 | } |
714 | ||
291c7d2f JB |
715 | if (load_cache_only) { |
716 | put_caching_control(caching_ctl); | |
11833d66 | 717 | return 0; |
817d52f8 | 718 | } |
817d52f8 | 719 | |
9e351cc8 | 720 | down_write(&fs_info->commit_root_sem); |
1e4f4714 | 721 | refcount_inc(&caching_ctl->count); |
11833d66 | 722 | list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups); |
9e351cc8 | 723 | up_write(&fs_info->commit_root_sem); |
11833d66 | 724 | |
11dfe35a | 725 | btrfs_get_block_group(cache); |
11833d66 | 726 | |
e66f0bb1 | 727 | btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work); |
817d52f8 | 728 | |
ef8bbdfe | 729 | return ret; |
e37c9e69 CM |
730 | } |
731 | ||
0f9dd46c JB |
732 | /* |
733 | * return the block group that starts at or after bytenr | |
734 | */ | |
d397712b CM |
735 | static struct btrfs_block_group_cache * |
736 | btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr) | |
0ef3e66b | 737 | { |
e2c89907 | 738 | return block_group_cache_tree_search(info, bytenr, 0); |
0ef3e66b CM |
739 | } |
740 | ||
0f9dd46c | 741 | /* |
9f55684c | 742 | * return the block group that contains the given bytenr |
0f9dd46c | 743 | */ |
d397712b CM |
744 | struct btrfs_block_group_cache *btrfs_lookup_block_group( |
745 | struct btrfs_fs_info *info, | |
746 | u64 bytenr) | |
be744175 | 747 | { |
e2c89907 | 748 | return block_group_cache_tree_search(info, bytenr, 1); |
be744175 | 749 | } |
0b86a832 | 750 | |
0f9dd46c JB |
751 | static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info, |
752 | u64 flags) | |
6324fbf3 | 753 | { |
0f9dd46c | 754 | struct list_head *head = &info->space_info; |
0f9dd46c | 755 | struct btrfs_space_info *found; |
4184ea7f | 756 | |
52ba6929 | 757 | flags &= BTRFS_BLOCK_GROUP_TYPE_MASK; |
b742bb82 | 758 | |
4184ea7f CM |
759 | rcu_read_lock(); |
760 | list_for_each_entry_rcu(found, head, list) { | |
67377734 | 761 | if (found->flags & flags) { |
4184ea7f | 762 | rcu_read_unlock(); |
0f9dd46c | 763 | return found; |
4184ea7f | 764 | } |
0f9dd46c | 765 | } |
4184ea7f | 766 | rcu_read_unlock(); |
0f9dd46c | 767 | return NULL; |
6324fbf3 CM |
768 | } |
769 | ||
0d9f824d OS |
770 | static void add_pinned_bytes(struct btrfs_fs_info *fs_info, s64 num_bytes, |
771 | u64 owner, u64 root_objectid) | |
772 | { | |
773 | struct btrfs_space_info *space_info; | |
774 | u64 flags; | |
775 | ||
776 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
777 | if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID) | |
778 | flags = BTRFS_BLOCK_GROUP_SYSTEM; | |
779 | else | |
780 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
781 | } else { | |
782 | flags = BTRFS_BLOCK_GROUP_DATA; | |
783 | } | |
784 | ||
785 | space_info = __find_space_info(fs_info, flags); | |
55e8196a | 786 | ASSERT(space_info); |
0d9f824d OS |
787 | percpu_counter_add(&space_info->total_bytes_pinned, num_bytes); |
788 | } | |
789 | ||
4184ea7f CM |
790 | /* |
791 | * after adding space to the filesystem, we need to clear the full flags | |
792 | * on all the space infos. | |
793 | */ | |
794 | void btrfs_clear_space_info_full(struct btrfs_fs_info *info) | |
795 | { | |
796 | struct list_head *head = &info->space_info; | |
797 | struct btrfs_space_info *found; | |
798 | ||
799 | rcu_read_lock(); | |
800 | list_for_each_entry_rcu(found, head, list) | |
801 | found->full = 0; | |
802 | rcu_read_unlock(); | |
803 | } | |
804 | ||
1a4ed8fd | 805 | /* simple helper to search for an existing data extent at a given offset */ |
2ff7e61e | 806 | int btrfs_lookup_data_extent(struct btrfs_fs_info *fs_info, u64 start, u64 len) |
e02119d5 CM |
807 | { |
808 | int ret; | |
809 | struct btrfs_key key; | |
31840ae1 | 810 | struct btrfs_path *path; |
e02119d5 | 811 | |
31840ae1 | 812 | path = btrfs_alloc_path(); |
d8926bb3 MF |
813 | if (!path) |
814 | return -ENOMEM; | |
815 | ||
e02119d5 CM |
816 | key.objectid = start; |
817 | key.offset = len; | |
3173a18f | 818 | key.type = BTRFS_EXTENT_ITEM_KEY; |
0b246afa | 819 | ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0); |
31840ae1 | 820 | btrfs_free_path(path); |
7bb86316 CM |
821 | return ret; |
822 | } | |
823 | ||
a22285a6 | 824 | /* |
3173a18f | 825 | * helper function to lookup reference count and flags of a tree block. |
a22285a6 YZ |
826 | * |
827 | * the head node for delayed ref is used to store the sum of all the | |
828 | * reference count modifications queued up in the rbtree. the head | |
829 | * node may also store the extent flags to set. This way you can check | |
830 | * to see what the reference count and extent flags would be if all of | |
831 | * the delayed refs are not processed. | |
832 | */ | |
833 | int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans, | |
2ff7e61e | 834 | struct btrfs_fs_info *fs_info, u64 bytenr, |
3173a18f | 835 | u64 offset, int metadata, u64 *refs, u64 *flags) |
a22285a6 YZ |
836 | { |
837 | struct btrfs_delayed_ref_head *head; | |
838 | struct btrfs_delayed_ref_root *delayed_refs; | |
839 | struct btrfs_path *path; | |
840 | struct btrfs_extent_item *ei; | |
841 | struct extent_buffer *leaf; | |
842 | struct btrfs_key key; | |
843 | u32 item_size; | |
844 | u64 num_refs; | |
845 | u64 extent_flags; | |
846 | int ret; | |
847 | ||
3173a18f JB |
848 | /* |
849 | * If we don't have skinny metadata, don't bother doing anything | |
850 | * different | |
851 | */ | |
0b246afa JM |
852 | if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) { |
853 | offset = fs_info->nodesize; | |
3173a18f JB |
854 | metadata = 0; |
855 | } | |
856 | ||
a22285a6 YZ |
857 | path = btrfs_alloc_path(); |
858 | if (!path) | |
859 | return -ENOMEM; | |
860 | ||
a22285a6 YZ |
861 | if (!trans) { |
862 | path->skip_locking = 1; | |
863 | path->search_commit_root = 1; | |
864 | } | |
639eefc8 FDBM |
865 | |
866 | search_again: | |
867 | key.objectid = bytenr; | |
868 | key.offset = offset; | |
869 | if (metadata) | |
870 | key.type = BTRFS_METADATA_ITEM_KEY; | |
871 | else | |
872 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
873 | ||
0b246afa | 874 | ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0); |
a22285a6 YZ |
875 | if (ret < 0) |
876 | goto out_free; | |
877 | ||
3173a18f | 878 | if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) { |
74be9510 FDBM |
879 | if (path->slots[0]) { |
880 | path->slots[0]--; | |
881 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
882 | path->slots[0]); | |
883 | if (key.objectid == bytenr && | |
884 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
0b246afa | 885 | key.offset == fs_info->nodesize) |
74be9510 FDBM |
886 | ret = 0; |
887 | } | |
3173a18f JB |
888 | } |
889 | ||
a22285a6 YZ |
890 | if (ret == 0) { |
891 | leaf = path->nodes[0]; | |
892 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
893 | if (item_size >= sizeof(*ei)) { | |
894 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
895 | struct btrfs_extent_item); | |
896 | num_refs = btrfs_extent_refs(leaf, ei); | |
897 | extent_flags = btrfs_extent_flags(leaf, ei); | |
898 | } else { | |
899 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
900 | struct btrfs_extent_item_v0 *ei0; | |
901 | BUG_ON(item_size != sizeof(*ei0)); | |
902 | ei0 = btrfs_item_ptr(leaf, path->slots[0], | |
903 | struct btrfs_extent_item_v0); | |
904 | num_refs = btrfs_extent_refs_v0(leaf, ei0); | |
905 | /* FIXME: this isn't correct for data */ | |
906 | extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
907 | #else | |
908 | BUG(); | |
909 | #endif | |
910 | } | |
911 | BUG_ON(num_refs == 0); | |
912 | } else { | |
913 | num_refs = 0; | |
914 | extent_flags = 0; | |
915 | ret = 0; | |
916 | } | |
917 | ||
918 | if (!trans) | |
919 | goto out; | |
920 | ||
921 | delayed_refs = &trans->transaction->delayed_refs; | |
922 | spin_lock(&delayed_refs->lock); | |
f72ad18e | 923 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
a22285a6 YZ |
924 | if (head) { |
925 | if (!mutex_trylock(&head->mutex)) { | |
6df8cdf5 | 926 | refcount_inc(&head->node.refs); |
a22285a6 YZ |
927 | spin_unlock(&delayed_refs->lock); |
928 | ||
b3b4aa74 | 929 | btrfs_release_path(path); |
a22285a6 | 930 | |
8cc33e5c DS |
931 | /* |
932 | * Mutex was contended, block until it's released and try | |
933 | * again | |
934 | */ | |
a22285a6 YZ |
935 | mutex_lock(&head->mutex); |
936 | mutex_unlock(&head->mutex); | |
937 | btrfs_put_delayed_ref(&head->node); | |
639eefc8 | 938 | goto search_again; |
a22285a6 | 939 | } |
d7df2c79 | 940 | spin_lock(&head->lock); |
a22285a6 YZ |
941 | if (head->extent_op && head->extent_op->update_flags) |
942 | extent_flags |= head->extent_op->flags_to_set; | |
943 | else | |
944 | BUG_ON(num_refs == 0); | |
945 | ||
946 | num_refs += head->node.ref_mod; | |
d7df2c79 | 947 | spin_unlock(&head->lock); |
a22285a6 YZ |
948 | mutex_unlock(&head->mutex); |
949 | } | |
950 | spin_unlock(&delayed_refs->lock); | |
951 | out: | |
952 | WARN_ON(num_refs == 0); | |
953 | if (refs) | |
954 | *refs = num_refs; | |
955 | if (flags) | |
956 | *flags = extent_flags; | |
957 | out_free: | |
958 | btrfs_free_path(path); | |
959 | return ret; | |
960 | } | |
961 | ||
d8d5f3e1 CM |
962 | /* |
963 | * Back reference rules. Back refs have three main goals: | |
964 | * | |
965 | * 1) differentiate between all holders of references to an extent so that | |
966 | * when a reference is dropped we can make sure it was a valid reference | |
967 | * before freeing the extent. | |
968 | * | |
969 | * 2) Provide enough information to quickly find the holders of an extent | |
970 | * if we notice a given block is corrupted or bad. | |
971 | * | |
972 | * 3) Make it easy to migrate blocks for FS shrinking or storage pool | |
973 | * maintenance. This is actually the same as #2, but with a slightly | |
974 | * different use case. | |
975 | * | |
5d4f98a2 YZ |
976 | * There are two kinds of back refs. The implicit back refs is optimized |
977 | * for pointers in non-shared tree blocks. For a given pointer in a block, | |
978 | * back refs of this kind provide information about the block's owner tree | |
979 | * and the pointer's key. These information allow us to find the block by | |
980 | * b-tree searching. The full back refs is for pointers in tree blocks not | |
981 | * referenced by their owner trees. The location of tree block is recorded | |
982 | * in the back refs. Actually the full back refs is generic, and can be | |
983 | * used in all cases the implicit back refs is used. The major shortcoming | |
984 | * of the full back refs is its overhead. Every time a tree block gets | |
985 | * COWed, we have to update back refs entry for all pointers in it. | |
986 | * | |
987 | * For a newly allocated tree block, we use implicit back refs for | |
988 | * pointers in it. This means most tree related operations only involve | |
989 | * implicit back refs. For a tree block created in old transaction, the | |
990 | * only way to drop a reference to it is COW it. So we can detect the | |
991 | * event that tree block loses its owner tree's reference and do the | |
992 | * back refs conversion. | |
993 | * | |
01327610 | 994 | * When a tree block is COWed through a tree, there are four cases: |
5d4f98a2 YZ |
995 | * |
996 | * The reference count of the block is one and the tree is the block's | |
997 | * owner tree. Nothing to do in this case. | |
998 | * | |
999 | * The reference count of the block is one and the tree is not the | |
1000 | * block's owner tree. In this case, full back refs is used for pointers | |
1001 | * in the block. Remove these full back refs, add implicit back refs for | |
1002 | * every pointers in the new block. | |
1003 | * | |
1004 | * The reference count of the block is greater than one and the tree is | |
1005 | * the block's owner tree. In this case, implicit back refs is used for | |
1006 | * pointers in the block. Add full back refs for every pointers in the | |
1007 | * block, increase lower level extents' reference counts. The original | |
1008 | * implicit back refs are entailed to the new block. | |
1009 | * | |
1010 | * The reference count of the block is greater than one and the tree is | |
1011 | * not the block's owner tree. Add implicit back refs for every pointer in | |
1012 | * the new block, increase lower level extents' reference count. | |
1013 | * | |
1014 | * Back Reference Key composing: | |
1015 | * | |
1016 | * The key objectid corresponds to the first byte in the extent, | |
1017 | * The key type is used to differentiate between types of back refs. | |
1018 | * There are different meanings of the key offset for different types | |
1019 | * of back refs. | |
1020 | * | |
d8d5f3e1 CM |
1021 | * File extents can be referenced by: |
1022 | * | |
1023 | * - multiple snapshots, subvolumes, or different generations in one subvol | |
31840ae1 | 1024 | * - different files inside a single subvolume |
d8d5f3e1 CM |
1025 | * - different offsets inside a file (bookend extents in file.c) |
1026 | * | |
5d4f98a2 | 1027 | * The extent ref structure for the implicit back refs has fields for: |
d8d5f3e1 CM |
1028 | * |
1029 | * - Objectid of the subvolume root | |
d8d5f3e1 | 1030 | * - objectid of the file holding the reference |
5d4f98a2 YZ |
1031 | * - original offset in the file |
1032 | * - how many bookend extents | |
d8d5f3e1 | 1033 | * |
5d4f98a2 YZ |
1034 | * The key offset for the implicit back refs is hash of the first |
1035 | * three fields. | |
d8d5f3e1 | 1036 | * |
5d4f98a2 | 1037 | * The extent ref structure for the full back refs has field for: |
d8d5f3e1 | 1038 | * |
5d4f98a2 | 1039 | * - number of pointers in the tree leaf |
d8d5f3e1 | 1040 | * |
5d4f98a2 YZ |
1041 | * The key offset for the implicit back refs is the first byte of |
1042 | * the tree leaf | |
d8d5f3e1 | 1043 | * |
5d4f98a2 YZ |
1044 | * When a file extent is allocated, The implicit back refs is used. |
1045 | * the fields are filled in: | |
d8d5f3e1 | 1046 | * |
5d4f98a2 | 1047 | * (root_key.objectid, inode objectid, offset in file, 1) |
d8d5f3e1 | 1048 | * |
5d4f98a2 YZ |
1049 | * When a file extent is removed file truncation, we find the |
1050 | * corresponding implicit back refs and check the following fields: | |
d8d5f3e1 | 1051 | * |
5d4f98a2 | 1052 | * (btrfs_header_owner(leaf), inode objectid, offset in file) |
d8d5f3e1 | 1053 | * |
5d4f98a2 | 1054 | * Btree extents can be referenced by: |
d8d5f3e1 | 1055 | * |
5d4f98a2 | 1056 | * - Different subvolumes |
d8d5f3e1 | 1057 | * |
5d4f98a2 YZ |
1058 | * Both the implicit back refs and the full back refs for tree blocks |
1059 | * only consist of key. The key offset for the implicit back refs is | |
1060 | * objectid of block's owner tree. The key offset for the full back refs | |
1061 | * is the first byte of parent block. | |
d8d5f3e1 | 1062 | * |
5d4f98a2 YZ |
1063 | * When implicit back refs is used, information about the lowest key and |
1064 | * level of the tree block are required. These information are stored in | |
1065 | * tree block info structure. | |
d8d5f3e1 | 1066 | */ |
31840ae1 | 1067 | |
5d4f98a2 YZ |
1068 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
1069 | static int convert_extent_item_v0(struct btrfs_trans_handle *trans, | |
87bde3cd | 1070 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1071 | struct btrfs_path *path, |
1072 | u64 owner, u32 extra_size) | |
7bb86316 | 1073 | { |
87bde3cd | 1074 | struct btrfs_root *root = fs_info->extent_root; |
5d4f98a2 YZ |
1075 | struct btrfs_extent_item *item; |
1076 | struct btrfs_extent_item_v0 *ei0; | |
1077 | struct btrfs_extent_ref_v0 *ref0; | |
1078 | struct btrfs_tree_block_info *bi; | |
1079 | struct extent_buffer *leaf; | |
7bb86316 | 1080 | struct btrfs_key key; |
5d4f98a2 YZ |
1081 | struct btrfs_key found_key; |
1082 | u32 new_size = sizeof(*item); | |
1083 | u64 refs; | |
1084 | int ret; | |
1085 | ||
1086 | leaf = path->nodes[0]; | |
1087 | BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0)); | |
1088 | ||
1089 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1090 | ei0 = btrfs_item_ptr(leaf, path->slots[0], | |
1091 | struct btrfs_extent_item_v0); | |
1092 | refs = btrfs_extent_refs_v0(leaf, ei0); | |
1093 | ||
1094 | if (owner == (u64)-1) { | |
1095 | while (1) { | |
1096 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1097 | ret = btrfs_next_leaf(root, path); | |
1098 | if (ret < 0) | |
1099 | return ret; | |
79787eaa | 1100 | BUG_ON(ret > 0); /* Corruption */ |
5d4f98a2 YZ |
1101 | leaf = path->nodes[0]; |
1102 | } | |
1103 | btrfs_item_key_to_cpu(leaf, &found_key, | |
1104 | path->slots[0]); | |
1105 | BUG_ON(key.objectid != found_key.objectid); | |
1106 | if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) { | |
1107 | path->slots[0]++; | |
1108 | continue; | |
1109 | } | |
1110 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1111 | struct btrfs_extent_ref_v0); | |
1112 | owner = btrfs_ref_objectid_v0(leaf, ref0); | |
1113 | break; | |
1114 | } | |
1115 | } | |
b3b4aa74 | 1116 | btrfs_release_path(path); |
5d4f98a2 YZ |
1117 | |
1118 | if (owner < BTRFS_FIRST_FREE_OBJECTID) | |
1119 | new_size += sizeof(*bi); | |
1120 | ||
1121 | new_size -= sizeof(*ei0); | |
1122 | ret = btrfs_search_slot(trans, root, &key, path, | |
1123 | new_size + extra_size, 1); | |
1124 | if (ret < 0) | |
1125 | return ret; | |
79787eaa | 1126 | BUG_ON(ret); /* Corruption */ |
5d4f98a2 | 1127 | |
87bde3cd | 1128 | btrfs_extend_item(fs_info, path, new_size); |
5d4f98a2 YZ |
1129 | |
1130 | leaf = path->nodes[0]; | |
1131 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1132 | btrfs_set_extent_refs(leaf, item, refs); | |
1133 | /* FIXME: get real generation */ | |
1134 | btrfs_set_extent_generation(leaf, item, 0); | |
1135 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1136 | btrfs_set_extent_flags(leaf, item, | |
1137 | BTRFS_EXTENT_FLAG_TREE_BLOCK | | |
1138 | BTRFS_BLOCK_FLAG_FULL_BACKREF); | |
1139 | bi = (struct btrfs_tree_block_info *)(item + 1); | |
1140 | /* FIXME: get first key of the block */ | |
b159fa28 | 1141 | memzero_extent_buffer(leaf, (unsigned long)bi, sizeof(*bi)); |
5d4f98a2 YZ |
1142 | btrfs_set_tree_block_level(leaf, bi, (int)owner); |
1143 | } else { | |
1144 | btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA); | |
1145 | } | |
1146 | btrfs_mark_buffer_dirty(leaf); | |
1147 | return 0; | |
1148 | } | |
1149 | #endif | |
1150 | ||
1151 | static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset) | |
1152 | { | |
1153 | u32 high_crc = ~(u32)0; | |
1154 | u32 low_crc = ~(u32)0; | |
1155 | __le64 lenum; | |
1156 | ||
1157 | lenum = cpu_to_le64(root_objectid); | |
14a958e6 | 1158 | high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 1159 | lenum = cpu_to_le64(owner); |
14a958e6 | 1160 | low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 1161 | lenum = cpu_to_le64(offset); |
14a958e6 | 1162 | low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 YZ |
1163 | |
1164 | return ((u64)high_crc << 31) ^ (u64)low_crc; | |
1165 | } | |
1166 | ||
1167 | static u64 hash_extent_data_ref_item(struct extent_buffer *leaf, | |
1168 | struct btrfs_extent_data_ref *ref) | |
1169 | { | |
1170 | return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref), | |
1171 | btrfs_extent_data_ref_objectid(leaf, ref), | |
1172 | btrfs_extent_data_ref_offset(leaf, ref)); | |
1173 | } | |
1174 | ||
1175 | static int match_extent_data_ref(struct extent_buffer *leaf, | |
1176 | struct btrfs_extent_data_ref *ref, | |
1177 | u64 root_objectid, u64 owner, u64 offset) | |
1178 | { | |
1179 | if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid || | |
1180 | btrfs_extent_data_ref_objectid(leaf, ref) != owner || | |
1181 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
1182 | return 0; | |
1183 | return 1; | |
1184 | } | |
1185 | ||
1186 | static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans, | |
87bde3cd | 1187 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1188 | struct btrfs_path *path, |
1189 | u64 bytenr, u64 parent, | |
1190 | u64 root_objectid, | |
1191 | u64 owner, u64 offset) | |
1192 | { | |
87bde3cd | 1193 | struct btrfs_root *root = fs_info->extent_root; |
5d4f98a2 YZ |
1194 | struct btrfs_key key; |
1195 | struct btrfs_extent_data_ref *ref; | |
31840ae1 | 1196 | struct extent_buffer *leaf; |
5d4f98a2 | 1197 | u32 nritems; |
74493f7a | 1198 | int ret; |
5d4f98a2 YZ |
1199 | int recow; |
1200 | int err = -ENOENT; | |
74493f7a | 1201 | |
31840ae1 | 1202 | key.objectid = bytenr; |
5d4f98a2 YZ |
1203 | if (parent) { |
1204 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1205 | key.offset = parent; | |
1206 | } else { | |
1207 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1208 | key.offset = hash_extent_data_ref(root_objectid, | |
1209 | owner, offset); | |
1210 | } | |
1211 | again: | |
1212 | recow = 0; | |
1213 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1214 | if (ret < 0) { | |
1215 | err = ret; | |
1216 | goto fail; | |
1217 | } | |
31840ae1 | 1218 | |
5d4f98a2 YZ |
1219 | if (parent) { |
1220 | if (!ret) | |
1221 | return 0; | |
1222 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1223 | key.type = BTRFS_EXTENT_REF_V0_KEY; | |
b3b4aa74 | 1224 | btrfs_release_path(path); |
5d4f98a2 YZ |
1225 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1226 | if (ret < 0) { | |
1227 | err = ret; | |
1228 | goto fail; | |
1229 | } | |
1230 | if (!ret) | |
1231 | return 0; | |
1232 | #endif | |
1233 | goto fail; | |
31840ae1 ZY |
1234 | } |
1235 | ||
1236 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1237 | nritems = btrfs_header_nritems(leaf); |
1238 | while (1) { | |
1239 | if (path->slots[0] >= nritems) { | |
1240 | ret = btrfs_next_leaf(root, path); | |
1241 | if (ret < 0) | |
1242 | err = ret; | |
1243 | if (ret) | |
1244 | goto fail; | |
1245 | ||
1246 | leaf = path->nodes[0]; | |
1247 | nritems = btrfs_header_nritems(leaf); | |
1248 | recow = 1; | |
1249 | } | |
1250 | ||
1251 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1252 | if (key.objectid != bytenr || | |
1253 | key.type != BTRFS_EXTENT_DATA_REF_KEY) | |
1254 | goto fail; | |
1255 | ||
1256 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1257 | struct btrfs_extent_data_ref); | |
1258 | ||
1259 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1260 | owner, offset)) { | |
1261 | if (recow) { | |
b3b4aa74 | 1262 | btrfs_release_path(path); |
5d4f98a2 YZ |
1263 | goto again; |
1264 | } | |
1265 | err = 0; | |
1266 | break; | |
1267 | } | |
1268 | path->slots[0]++; | |
31840ae1 | 1269 | } |
5d4f98a2 YZ |
1270 | fail: |
1271 | return err; | |
31840ae1 ZY |
1272 | } |
1273 | ||
5d4f98a2 | 1274 | static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans, |
87bde3cd | 1275 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1276 | struct btrfs_path *path, |
1277 | u64 bytenr, u64 parent, | |
1278 | u64 root_objectid, u64 owner, | |
1279 | u64 offset, int refs_to_add) | |
31840ae1 | 1280 | { |
87bde3cd | 1281 | struct btrfs_root *root = fs_info->extent_root; |
31840ae1 ZY |
1282 | struct btrfs_key key; |
1283 | struct extent_buffer *leaf; | |
5d4f98a2 | 1284 | u32 size; |
31840ae1 ZY |
1285 | u32 num_refs; |
1286 | int ret; | |
74493f7a | 1287 | |
74493f7a | 1288 | key.objectid = bytenr; |
5d4f98a2 YZ |
1289 | if (parent) { |
1290 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1291 | key.offset = parent; | |
1292 | size = sizeof(struct btrfs_shared_data_ref); | |
1293 | } else { | |
1294 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1295 | key.offset = hash_extent_data_ref(root_objectid, | |
1296 | owner, offset); | |
1297 | size = sizeof(struct btrfs_extent_data_ref); | |
1298 | } | |
74493f7a | 1299 | |
5d4f98a2 YZ |
1300 | ret = btrfs_insert_empty_item(trans, root, path, &key, size); |
1301 | if (ret && ret != -EEXIST) | |
1302 | goto fail; | |
1303 | ||
1304 | leaf = path->nodes[0]; | |
1305 | if (parent) { | |
1306 | struct btrfs_shared_data_ref *ref; | |
31840ae1 | 1307 | ref = btrfs_item_ptr(leaf, path->slots[0], |
5d4f98a2 YZ |
1308 | struct btrfs_shared_data_ref); |
1309 | if (ret == 0) { | |
1310 | btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add); | |
1311 | } else { | |
1312 | num_refs = btrfs_shared_data_ref_count(leaf, ref); | |
1313 | num_refs += refs_to_add; | |
1314 | btrfs_set_shared_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1315 | } |
5d4f98a2 YZ |
1316 | } else { |
1317 | struct btrfs_extent_data_ref *ref; | |
1318 | while (ret == -EEXIST) { | |
1319 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1320 | struct btrfs_extent_data_ref); | |
1321 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1322 | owner, offset)) | |
1323 | break; | |
b3b4aa74 | 1324 | btrfs_release_path(path); |
5d4f98a2 YZ |
1325 | key.offset++; |
1326 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1327 | size); | |
1328 | if (ret && ret != -EEXIST) | |
1329 | goto fail; | |
31840ae1 | 1330 | |
5d4f98a2 YZ |
1331 | leaf = path->nodes[0]; |
1332 | } | |
1333 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1334 | struct btrfs_extent_data_ref); | |
1335 | if (ret == 0) { | |
1336 | btrfs_set_extent_data_ref_root(leaf, ref, | |
1337 | root_objectid); | |
1338 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
1339 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
1340 | btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add); | |
1341 | } else { | |
1342 | num_refs = btrfs_extent_data_ref_count(leaf, ref); | |
1343 | num_refs += refs_to_add; | |
1344 | btrfs_set_extent_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1345 | } |
31840ae1 | 1346 | } |
5d4f98a2 YZ |
1347 | btrfs_mark_buffer_dirty(leaf); |
1348 | ret = 0; | |
1349 | fail: | |
b3b4aa74 | 1350 | btrfs_release_path(path); |
7bb86316 | 1351 | return ret; |
74493f7a CM |
1352 | } |
1353 | ||
5d4f98a2 | 1354 | static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans, |
87bde3cd | 1355 | struct btrfs_fs_info *fs_info, |
5d4f98a2 | 1356 | struct btrfs_path *path, |
fcebe456 | 1357 | int refs_to_drop, int *last_ref) |
31840ae1 | 1358 | { |
5d4f98a2 YZ |
1359 | struct btrfs_key key; |
1360 | struct btrfs_extent_data_ref *ref1 = NULL; | |
1361 | struct btrfs_shared_data_ref *ref2 = NULL; | |
31840ae1 | 1362 | struct extent_buffer *leaf; |
5d4f98a2 | 1363 | u32 num_refs = 0; |
31840ae1 ZY |
1364 | int ret = 0; |
1365 | ||
1366 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1367 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
1368 | ||
1369 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1370 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1371 | struct btrfs_extent_data_ref); | |
1372 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1373 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1374 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1375 | struct btrfs_shared_data_ref); | |
1376 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1377 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1378 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
1379 | struct btrfs_extent_ref_v0 *ref0; | |
1380 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1381 | struct btrfs_extent_ref_v0); | |
1382 | num_refs = btrfs_ref_count_v0(leaf, ref0); | |
1383 | #endif | |
1384 | } else { | |
1385 | BUG(); | |
1386 | } | |
1387 | ||
56bec294 CM |
1388 | BUG_ON(num_refs < refs_to_drop); |
1389 | num_refs -= refs_to_drop; | |
5d4f98a2 | 1390 | |
31840ae1 | 1391 | if (num_refs == 0) { |
87bde3cd | 1392 | ret = btrfs_del_item(trans, fs_info->extent_root, path); |
fcebe456 | 1393 | *last_ref = 1; |
31840ae1 | 1394 | } else { |
5d4f98a2 YZ |
1395 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) |
1396 | btrfs_set_extent_data_ref_count(leaf, ref1, num_refs); | |
1397 | else if (key.type == BTRFS_SHARED_DATA_REF_KEY) | |
1398 | btrfs_set_shared_data_ref_count(leaf, ref2, num_refs); | |
1399 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1400 | else { | |
1401 | struct btrfs_extent_ref_v0 *ref0; | |
1402 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1403 | struct btrfs_extent_ref_v0); | |
1404 | btrfs_set_ref_count_v0(leaf, ref0, num_refs); | |
1405 | } | |
1406 | #endif | |
31840ae1 ZY |
1407 | btrfs_mark_buffer_dirty(leaf); |
1408 | } | |
31840ae1 ZY |
1409 | return ret; |
1410 | } | |
1411 | ||
9ed0dea0 | 1412 | static noinline u32 extent_data_ref_count(struct btrfs_path *path, |
5d4f98a2 | 1413 | struct btrfs_extent_inline_ref *iref) |
15916de8 | 1414 | { |
5d4f98a2 YZ |
1415 | struct btrfs_key key; |
1416 | struct extent_buffer *leaf; | |
1417 | struct btrfs_extent_data_ref *ref1; | |
1418 | struct btrfs_shared_data_ref *ref2; | |
1419 | u32 num_refs = 0; | |
1420 | ||
1421 | leaf = path->nodes[0]; | |
1422 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1423 | if (iref) { | |
1424 | if (btrfs_extent_inline_ref_type(leaf, iref) == | |
1425 | BTRFS_EXTENT_DATA_REF_KEY) { | |
1426 | ref1 = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1427 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1428 | } else { | |
1429 | ref2 = (struct btrfs_shared_data_ref *)(iref + 1); | |
1430 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1431 | } | |
1432 | } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1433 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1434 | struct btrfs_extent_data_ref); | |
1435 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1436 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1437 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1438 | struct btrfs_shared_data_ref); | |
1439 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1440 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1441 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
1442 | struct btrfs_extent_ref_v0 *ref0; | |
1443 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1444 | struct btrfs_extent_ref_v0); | |
1445 | num_refs = btrfs_ref_count_v0(leaf, ref0); | |
4b4e25f2 | 1446 | #endif |
5d4f98a2 YZ |
1447 | } else { |
1448 | WARN_ON(1); | |
1449 | } | |
1450 | return num_refs; | |
1451 | } | |
15916de8 | 1452 | |
5d4f98a2 | 1453 | static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans, |
87bde3cd | 1454 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1455 | struct btrfs_path *path, |
1456 | u64 bytenr, u64 parent, | |
1457 | u64 root_objectid) | |
1f3c79a2 | 1458 | { |
87bde3cd | 1459 | struct btrfs_root *root = fs_info->extent_root; |
5d4f98a2 | 1460 | struct btrfs_key key; |
1f3c79a2 | 1461 | int ret; |
1f3c79a2 | 1462 | |
5d4f98a2 YZ |
1463 | key.objectid = bytenr; |
1464 | if (parent) { | |
1465 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1466 | key.offset = parent; | |
1467 | } else { | |
1468 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1469 | key.offset = root_objectid; | |
1f3c79a2 LH |
1470 | } |
1471 | ||
5d4f98a2 YZ |
1472 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1473 | if (ret > 0) | |
1474 | ret = -ENOENT; | |
1475 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1476 | if (ret == -ENOENT && parent) { | |
b3b4aa74 | 1477 | btrfs_release_path(path); |
5d4f98a2 YZ |
1478 | key.type = BTRFS_EXTENT_REF_V0_KEY; |
1479 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1480 | if (ret > 0) | |
1481 | ret = -ENOENT; | |
1482 | } | |
1f3c79a2 | 1483 | #endif |
5d4f98a2 | 1484 | return ret; |
1f3c79a2 LH |
1485 | } |
1486 | ||
5d4f98a2 | 1487 | static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans, |
87bde3cd | 1488 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1489 | struct btrfs_path *path, |
1490 | u64 bytenr, u64 parent, | |
1491 | u64 root_objectid) | |
31840ae1 | 1492 | { |
5d4f98a2 | 1493 | struct btrfs_key key; |
31840ae1 | 1494 | int ret; |
31840ae1 | 1495 | |
5d4f98a2 YZ |
1496 | key.objectid = bytenr; |
1497 | if (parent) { | |
1498 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1499 | key.offset = parent; | |
1500 | } else { | |
1501 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1502 | key.offset = root_objectid; | |
1503 | } | |
1504 | ||
87bde3cd JM |
1505 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, |
1506 | path, &key, 0); | |
b3b4aa74 | 1507 | btrfs_release_path(path); |
31840ae1 ZY |
1508 | return ret; |
1509 | } | |
1510 | ||
5d4f98a2 | 1511 | static inline int extent_ref_type(u64 parent, u64 owner) |
31840ae1 | 1512 | { |
5d4f98a2 YZ |
1513 | int type; |
1514 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1515 | if (parent > 0) | |
1516 | type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1517 | else | |
1518 | type = BTRFS_TREE_BLOCK_REF_KEY; | |
1519 | } else { | |
1520 | if (parent > 0) | |
1521 | type = BTRFS_SHARED_DATA_REF_KEY; | |
1522 | else | |
1523 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
1524 | } | |
1525 | return type; | |
31840ae1 | 1526 | } |
56bec294 | 1527 | |
2c47e605 YZ |
1528 | static int find_next_key(struct btrfs_path *path, int level, |
1529 | struct btrfs_key *key) | |
56bec294 | 1530 | |
02217ed2 | 1531 | { |
2c47e605 | 1532 | for (; level < BTRFS_MAX_LEVEL; level++) { |
5d4f98a2 YZ |
1533 | if (!path->nodes[level]) |
1534 | break; | |
5d4f98a2 YZ |
1535 | if (path->slots[level] + 1 >= |
1536 | btrfs_header_nritems(path->nodes[level])) | |
1537 | continue; | |
1538 | if (level == 0) | |
1539 | btrfs_item_key_to_cpu(path->nodes[level], key, | |
1540 | path->slots[level] + 1); | |
1541 | else | |
1542 | btrfs_node_key_to_cpu(path->nodes[level], key, | |
1543 | path->slots[level] + 1); | |
1544 | return 0; | |
1545 | } | |
1546 | return 1; | |
1547 | } | |
037e6390 | 1548 | |
5d4f98a2 YZ |
1549 | /* |
1550 | * look for inline back ref. if back ref is found, *ref_ret is set | |
1551 | * to the address of inline back ref, and 0 is returned. | |
1552 | * | |
1553 | * if back ref isn't found, *ref_ret is set to the address where it | |
1554 | * should be inserted, and -ENOENT is returned. | |
1555 | * | |
1556 | * if insert is true and there are too many inline back refs, the path | |
1557 | * points to the extent item, and -EAGAIN is returned. | |
1558 | * | |
1559 | * NOTE: inline back refs are ordered in the same way that back ref | |
1560 | * items in the tree are ordered. | |
1561 | */ | |
1562 | static noinline_for_stack | |
1563 | int lookup_inline_extent_backref(struct btrfs_trans_handle *trans, | |
87bde3cd | 1564 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1565 | struct btrfs_path *path, |
1566 | struct btrfs_extent_inline_ref **ref_ret, | |
1567 | u64 bytenr, u64 num_bytes, | |
1568 | u64 parent, u64 root_objectid, | |
1569 | u64 owner, u64 offset, int insert) | |
1570 | { | |
87bde3cd | 1571 | struct btrfs_root *root = fs_info->extent_root; |
5d4f98a2 YZ |
1572 | struct btrfs_key key; |
1573 | struct extent_buffer *leaf; | |
1574 | struct btrfs_extent_item *ei; | |
1575 | struct btrfs_extent_inline_ref *iref; | |
1576 | u64 flags; | |
1577 | u64 item_size; | |
1578 | unsigned long ptr; | |
1579 | unsigned long end; | |
1580 | int extra_size; | |
1581 | int type; | |
1582 | int want; | |
1583 | int ret; | |
1584 | int err = 0; | |
0b246afa | 1585 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); |
26b8003f | 1586 | |
db94535d | 1587 | key.objectid = bytenr; |
31840ae1 | 1588 | key.type = BTRFS_EXTENT_ITEM_KEY; |
56bec294 | 1589 | key.offset = num_bytes; |
31840ae1 | 1590 | |
5d4f98a2 YZ |
1591 | want = extent_ref_type(parent, owner); |
1592 | if (insert) { | |
1593 | extra_size = btrfs_extent_inline_ref_size(want); | |
85d4198e | 1594 | path->keep_locks = 1; |
5d4f98a2 YZ |
1595 | } else |
1596 | extra_size = -1; | |
3173a18f JB |
1597 | |
1598 | /* | |
1599 | * Owner is our parent level, so we can just add one to get the level | |
1600 | * for the block we are interested in. | |
1601 | */ | |
1602 | if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1603 | key.type = BTRFS_METADATA_ITEM_KEY; | |
1604 | key.offset = owner; | |
1605 | } | |
1606 | ||
1607 | again: | |
5d4f98a2 | 1608 | ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1); |
b9473439 | 1609 | if (ret < 0) { |
5d4f98a2 YZ |
1610 | err = ret; |
1611 | goto out; | |
1612 | } | |
3173a18f JB |
1613 | |
1614 | /* | |
1615 | * We may be a newly converted file system which still has the old fat | |
1616 | * extent entries for metadata, so try and see if we have one of those. | |
1617 | */ | |
1618 | if (ret > 0 && skinny_metadata) { | |
1619 | skinny_metadata = false; | |
1620 | if (path->slots[0]) { | |
1621 | path->slots[0]--; | |
1622 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
1623 | path->slots[0]); | |
1624 | if (key.objectid == bytenr && | |
1625 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
1626 | key.offset == num_bytes) | |
1627 | ret = 0; | |
1628 | } | |
1629 | if (ret) { | |
9ce49a0b | 1630 | key.objectid = bytenr; |
3173a18f JB |
1631 | key.type = BTRFS_EXTENT_ITEM_KEY; |
1632 | key.offset = num_bytes; | |
1633 | btrfs_release_path(path); | |
1634 | goto again; | |
1635 | } | |
1636 | } | |
1637 | ||
79787eaa JM |
1638 | if (ret && !insert) { |
1639 | err = -ENOENT; | |
1640 | goto out; | |
fae7f21c | 1641 | } else if (WARN_ON(ret)) { |
492104c8 | 1642 | err = -EIO; |
492104c8 | 1643 | goto out; |
79787eaa | 1644 | } |
5d4f98a2 YZ |
1645 | |
1646 | leaf = path->nodes[0]; | |
1647 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1648 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1649 | if (item_size < sizeof(*ei)) { | |
1650 | if (!insert) { | |
1651 | err = -ENOENT; | |
1652 | goto out; | |
1653 | } | |
87bde3cd | 1654 | ret = convert_extent_item_v0(trans, fs_info, path, owner, |
5d4f98a2 YZ |
1655 | extra_size); |
1656 | if (ret < 0) { | |
1657 | err = ret; | |
1658 | goto out; | |
1659 | } | |
1660 | leaf = path->nodes[0]; | |
1661 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1662 | } | |
1663 | #endif | |
1664 | BUG_ON(item_size < sizeof(*ei)); | |
1665 | ||
5d4f98a2 YZ |
1666 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1667 | flags = btrfs_extent_flags(leaf, ei); | |
1668 | ||
1669 | ptr = (unsigned long)(ei + 1); | |
1670 | end = (unsigned long)ei + item_size; | |
1671 | ||
3173a18f | 1672 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) { |
5d4f98a2 YZ |
1673 | ptr += sizeof(struct btrfs_tree_block_info); |
1674 | BUG_ON(ptr > end); | |
5d4f98a2 YZ |
1675 | } |
1676 | ||
1677 | err = -ENOENT; | |
1678 | while (1) { | |
1679 | if (ptr >= end) { | |
1680 | WARN_ON(ptr > end); | |
1681 | break; | |
1682 | } | |
1683 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1684 | type = btrfs_extent_inline_ref_type(leaf, iref); | |
1685 | if (want < type) | |
1686 | break; | |
1687 | if (want > type) { | |
1688 | ptr += btrfs_extent_inline_ref_size(type); | |
1689 | continue; | |
1690 | } | |
1691 | ||
1692 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1693 | struct btrfs_extent_data_ref *dref; | |
1694 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1695 | if (match_extent_data_ref(leaf, dref, root_objectid, | |
1696 | owner, offset)) { | |
1697 | err = 0; | |
1698 | break; | |
1699 | } | |
1700 | if (hash_extent_data_ref_item(leaf, dref) < | |
1701 | hash_extent_data_ref(root_objectid, owner, offset)) | |
1702 | break; | |
1703 | } else { | |
1704 | u64 ref_offset; | |
1705 | ref_offset = btrfs_extent_inline_ref_offset(leaf, iref); | |
1706 | if (parent > 0) { | |
1707 | if (parent == ref_offset) { | |
1708 | err = 0; | |
1709 | break; | |
1710 | } | |
1711 | if (ref_offset < parent) | |
1712 | break; | |
1713 | } else { | |
1714 | if (root_objectid == ref_offset) { | |
1715 | err = 0; | |
1716 | break; | |
1717 | } | |
1718 | if (ref_offset < root_objectid) | |
1719 | break; | |
1720 | } | |
1721 | } | |
1722 | ptr += btrfs_extent_inline_ref_size(type); | |
1723 | } | |
1724 | if (err == -ENOENT && insert) { | |
1725 | if (item_size + extra_size >= | |
1726 | BTRFS_MAX_EXTENT_ITEM_SIZE(root)) { | |
1727 | err = -EAGAIN; | |
1728 | goto out; | |
1729 | } | |
1730 | /* | |
1731 | * To add new inline back ref, we have to make sure | |
1732 | * there is no corresponding back ref item. | |
1733 | * For simplicity, we just do not add new inline back | |
1734 | * ref if there is any kind of item for this block | |
1735 | */ | |
2c47e605 YZ |
1736 | if (find_next_key(path, 0, &key) == 0 && |
1737 | key.objectid == bytenr && | |
85d4198e | 1738 | key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) { |
5d4f98a2 YZ |
1739 | err = -EAGAIN; |
1740 | goto out; | |
1741 | } | |
1742 | } | |
1743 | *ref_ret = (struct btrfs_extent_inline_ref *)ptr; | |
1744 | out: | |
85d4198e | 1745 | if (insert) { |
5d4f98a2 YZ |
1746 | path->keep_locks = 0; |
1747 | btrfs_unlock_up_safe(path, 1); | |
1748 | } | |
1749 | return err; | |
1750 | } | |
1751 | ||
1752 | /* | |
1753 | * helper to add new inline back ref | |
1754 | */ | |
1755 | static noinline_for_stack | |
87bde3cd | 1756 | void setup_inline_extent_backref(struct btrfs_fs_info *fs_info, |
143bede5 JM |
1757 | struct btrfs_path *path, |
1758 | struct btrfs_extent_inline_ref *iref, | |
1759 | u64 parent, u64 root_objectid, | |
1760 | u64 owner, u64 offset, int refs_to_add, | |
1761 | struct btrfs_delayed_extent_op *extent_op) | |
5d4f98a2 YZ |
1762 | { |
1763 | struct extent_buffer *leaf; | |
1764 | struct btrfs_extent_item *ei; | |
1765 | unsigned long ptr; | |
1766 | unsigned long end; | |
1767 | unsigned long item_offset; | |
1768 | u64 refs; | |
1769 | int size; | |
1770 | int type; | |
5d4f98a2 YZ |
1771 | |
1772 | leaf = path->nodes[0]; | |
1773 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1774 | item_offset = (unsigned long)iref - (unsigned long)ei; | |
1775 | ||
1776 | type = extent_ref_type(parent, owner); | |
1777 | size = btrfs_extent_inline_ref_size(type); | |
1778 | ||
87bde3cd | 1779 | btrfs_extend_item(fs_info, path, size); |
5d4f98a2 YZ |
1780 | |
1781 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1782 | refs = btrfs_extent_refs(leaf, ei); | |
1783 | refs += refs_to_add; | |
1784 | btrfs_set_extent_refs(leaf, ei, refs); | |
1785 | if (extent_op) | |
1786 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1787 | ||
1788 | ptr = (unsigned long)ei + item_offset; | |
1789 | end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]); | |
1790 | if (ptr < end - size) | |
1791 | memmove_extent_buffer(leaf, ptr + size, ptr, | |
1792 | end - size - ptr); | |
1793 | ||
1794 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1795 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
1796 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1797 | struct btrfs_extent_data_ref *dref; | |
1798 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1799 | btrfs_set_extent_data_ref_root(leaf, dref, root_objectid); | |
1800 | btrfs_set_extent_data_ref_objectid(leaf, dref, owner); | |
1801 | btrfs_set_extent_data_ref_offset(leaf, dref, offset); | |
1802 | btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add); | |
1803 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1804 | struct btrfs_shared_data_ref *sref; | |
1805 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1806 | btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add); | |
1807 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1808 | } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) { | |
1809 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1810 | } else { | |
1811 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
1812 | } | |
1813 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1814 | } |
1815 | ||
1816 | static int lookup_extent_backref(struct btrfs_trans_handle *trans, | |
87bde3cd | 1817 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1818 | struct btrfs_path *path, |
1819 | struct btrfs_extent_inline_ref **ref_ret, | |
1820 | u64 bytenr, u64 num_bytes, u64 parent, | |
1821 | u64 root_objectid, u64 owner, u64 offset) | |
1822 | { | |
1823 | int ret; | |
1824 | ||
87bde3cd | 1825 | ret = lookup_inline_extent_backref(trans, fs_info, path, ref_ret, |
5d4f98a2 YZ |
1826 | bytenr, num_bytes, parent, |
1827 | root_objectid, owner, offset, 0); | |
1828 | if (ret != -ENOENT) | |
54aa1f4d | 1829 | return ret; |
5d4f98a2 | 1830 | |
b3b4aa74 | 1831 | btrfs_release_path(path); |
5d4f98a2 YZ |
1832 | *ref_ret = NULL; |
1833 | ||
1834 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
87bde3cd JM |
1835 | ret = lookup_tree_block_ref(trans, fs_info, path, bytenr, |
1836 | parent, root_objectid); | |
5d4f98a2 | 1837 | } else { |
87bde3cd JM |
1838 | ret = lookup_extent_data_ref(trans, fs_info, path, bytenr, |
1839 | parent, root_objectid, owner, | |
1840 | offset); | |
b9473439 | 1841 | } |
5d4f98a2 YZ |
1842 | return ret; |
1843 | } | |
31840ae1 | 1844 | |
5d4f98a2 YZ |
1845 | /* |
1846 | * helper to update/remove inline back ref | |
1847 | */ | |
1848 | static noinline_for_stack | |
87bde3cd | 1849 | void update_inline_extent_backref(struct btrfs_fs_info *fs_info, |
143bede5 JM |
1850 | struct btrfs_path *path, |
1851 | struct btrfs_extent_inline_ref *iref, | |
1852 | int refs_to_mod, | |
fcebe456 JB |
1853 | struct btrfs_delayed_extent_op *extent_op, |
1854 | int *last_ref) | |
5d4f98a2 YZ |
1855 | { |
1856 | struct extent_buffer *leaf; | |
1857 | struct btrfs_extent_item *ei; | |
1858 | struct btrfs_extent_data_ref *dref = NULL; | |
1859 | struct btrfs_shared_data_ref *sref = NULL; | |
1860 | unsigned long ptr; | |
1861 | unsigned long end; | |
1862 | u32 item_size; | |
1863 | int size; | |
1864 | int type; | |
5d4f98a2 YZ |
1865 | u64 refs; |
1866 | ||
1867 | leaf = path->nodes[0]; | |
1868 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1869 | refs = btrfs_extent_refs(leaf, ei); | |
1870 | WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0); | |
1871 | refs += refs_to_mod; | |
1872 | btrfs_set_extent_refs(leaf, ei, refs); | |
1873 | if (extent_op) | |
1874 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1875 | ||
1876 | type = btrfs_extent_inline_ref_type(leaf, iref); | |
1877 | ||
1878 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1879 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1880 | refs = btrfs_extent_data_ref_count(leaf, dref); | |
1881 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1882 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1883 | refs = btrfs_shared_data_ref_count(leaf, sref); | |
1884 | } else { | |
1885 | refs = 1; | |
1886 | BUG_ON(refs_to_mod != -1); | |
56bec294 | 1887 | } |
31840ae1 | 1888 | |
5d4f98a2 YZ |
1889 | BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod); |
1890 | refs += refs_to_mod; | |
1891 | ||
1892 | if (refs > 0) { | |
1893 | if (type == BTRFS_EXTENT_DATA_REF_KEY) | |
1894 | btrfs_set_extent_data_ref_count(leaf, dref, refs); | |
1895 | else | |
1896 | btrfs_set_shared_data_ref_count(leaf, sref, refs); | |
1897 | } else { | |
fcebe456 | 1898 | *last_ref = 1; |
5d4f98a2 YZ |
1899 | size = btrfs_extent_inline_ref_size(type); |
1900 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1901 | ptr = (unsigned long)iref; | |
1902 | end = (unsigned long)ei + item_size; | |
1903 | if (ptr + size < end) | |
1904 | memmove_extent_buffer(leaf, ptr, ptr + size, | |
1905 | end - ptr - size); | |
1906 | item_size -= size; | |
87bde3cd | 1907 | btrfs_truncate_item(fs_info, path, item_size, 1); |
5d4f98a2 YZ |
1908 | } |
1909 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1910 | } |
1911 | ||
1912 | static noinline_for_stack | |
1913 | int insert_inline_extent_backref(struct btrfs_trans_handle *trans, | |
87bde3cd | 1914 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1915 | struct btrfs_path *path, |
1916 | u64 bytenr, u64 num_bytes, u64 parent, | |
1917 | u64 root_objectid, u64 owner, | |
1918 | u64 offset, int refs_to_add, | |
1919 | struct btrfs_delayed_extent_op *extent_op) | |
1920 | { | |
1921 | struct btrfs_extent_inline_ref *iref; | |
1922 | int ret; | |
1923 | ||
87bde3cd | 1924 | ret = lookup_inline_extent_backref(trans, fs_info, path, &iref, |
5d4f98a2 YZ |
1925 | bytenr, num_bytes, parent, |
1926 | root_objectid, owner, offset, 1); | |
1927 | if (ret == 0) { | |
1928 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID); | |
87bde3cd | 1929 | update_inline_extent_backref(fs_info, path, iref, |
fcebe456 | 1930 | refs_to_add, extent_op, NULL); |
5d4f98a2 | 1931 | } else if (ret == -ENOENT) { |
87bde3cd | 1932 | setup_inline_extent_backref(fs_info, path, iref, parent, |
143bede5 JM |
1933 | root_objectid, owner, offset, |
1934 | refs_to_add, extent_op); | |
1935 | ret = 0; | |
771ed689 | 1936 | } |
5d4f98a2 YZ |
1937 | return ret; |
1938 | } | |
31840ae1 | 1939 | |
5d4f98a2 | 1940 | static int insert_extent_backref(struct btrfs_trans_handle *trans, |
87bde3cd | 1941 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1942 | struct btrfs_path *path, |
1943 | u64 bytenr, u64 parent, u64 root_objectid, | |
1944 | u64 owner, u64 offset, int refs_to_add) | |
1945 | { | |
1946 | int ret; | |
1947 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1948 | BUG_ON(refs_to_add != 1); | |
87bde3cd | 1949 | ret = insert_tree_block_ref(trans, fs_info, path, bytenr, |
5d4f98a2 YZ |
1950 | parent, root_objectid); |
1951 | } else { | |
87bde3cd | 1952 | ret = insert_extent_data_ref(trans, fs_info, path, bytenr, |
5d4f98a2 YZ |
1953 | parent, root_objectid, |
1954 | owner, offset, refs_to_add); | |
1955 | } | |
1956 | return ret; | |
1957 | } | |
56bec294 | 1958 | |
5d4f98a2 | 1959 | static int remove_extent_backref(struct btrfs_trans_handle *trans, |
87bde3cd | 1960 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1961 | struct btrfs_path *path, |
1962 | struct btrfs_extent_inline_ref *iref, | |
fcebe456 | 1963 | int refs_to_drop, int is_data, int *last_ref) |
5d4f98a2 | 1964 | { |
143bede5 | 1965 | int ret = 0; |
b9473439 | 1966 | |
5d4f98a2 YZ |
1967 | BUG_ON(!is_data && refs_to_drop != 1); |
1968 | if (iref) { | |
87bde3cd | 1969 | update_inline_extent_backref(fs_info, path, iref, |
fcebe456 | 1970 | -refs_to_drop, NULL, last_ref); |
5d4f98a2 | 1971 | } else if (is_data) { |
87bde3cd | 1972 | ret = remove_extent_data_ref(trans, fs_info, path, refs_to_drop, |
fcebe456 | 1973 | last_ref); |
5d4f98a2 | 1974 | } else { |
fcebe456 | 1975 | *last_ref = 1; |
87bde3cd | 1976 | ret = btrfs_del_item(trans, fs_info->extent_root, path); |
5d4f98a2 YZ |
1977 | } |
1978 | return ret; | |
1979 | } | |
1980 | ||
86557861 | 1981 | #define in_range(b, first, len) ((b) >= (first) && (b) < (first) + (len)) |
d04c6b88 JM |
1982 | static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len, |
1983 | u64 *discarded_bytes) | |
5d4f98a2 | 1984 | { |
86557861 JM |
1985 | int j, ret = 0; |
1986 | u64 bytes_left, end; | |
4d89d377 | 1987 | u64 aligned_start = ALIGN(start, 1 << 9); |
d04c6b88 | 1988 | |
4d89d377 JM |
1989 | if (WARN_ON(start != aligned_start)) { |
1990 | len -= aligned_start - start; | |
1991 | len = round_down(len, 1 << 9); | |
1992 | start = aligned_start; | |
1993 | } | |
d04c6b88 | 1994 | |
4d89d377 | 1995 | *discarded_bytes = 0; |
86557861 JM |
1996 | |
1997 | if (!len) | |
1998 | return 0; | |
1999 | ||
2000 | end = start + len; | |
2001 | bytes_left = len; | |
2002 | ||
2003 | /* Skip any superblocks on this device. */ | |
2004 | for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) { | |
2005 | u64 sb_start = btrfs_sb_offset(j); | |
2006 | u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE; | |
2007 | u64 size = sb_start - start; | |
2008 | ||
2009 | if (!in_range(sb_start, start, bytes_left) && | |
2010 | !in_range(sb_end, start, bytes_left) && | |
2011 | !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE)) | |
2012 | continue; | |
2013 | ||
2014 | /* | |
2015 | * Superblock spans beginning of range. Adjust start and | |
2016 | * try again. | |
2017 | */ | |
2018 | if (sb_start <= start) { | |
2019 | start += sb_end - start; | |
2020 | if (start > end) { | |
2021 | bytes_left = 0; | |
2022 | break; | |
2023 | } | |
2024 | bytes_left = end - start; | |
2025 | continue; | |
2026 | } | |
2027 | ||
2028 | if (size) { | |
2029 | ret = blkdev_issue_discard(bdev, start >> 9, size >> 9, | |
2030 | GFP_NOFS, 0); | |
2031 | if (!ret) | |
2032 | *discarded_bytes += size; | |
2033 | else if (ret != -EOPNOTSUPP) | |
2034 | return ret; | |
2035 | } | |
2036 | ||
2037 | start = sb_end; | |
2038 | if (start > end) { | |
2039 | bytes_left = 0; | |
2040 | break; | |
2041 | } | |
2042 | bytes_left = end - start; | |
2043 | } | |
2044 | ||
2045 | if (bytes_left) { | |
2046 | ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9, | |
4d89d377 JM |
2047 | GFP_NOFS, 0); |
2048 | if (!ret) | |
86557861 | 2049 | *discarded_bytes += bytes_left; |
4d89d377 | 2050 | } |
d04c6b88 | 2051 | return ret; |
5d4f98a2 | 2052 | } |
5d4f98a2 | 2053 | |
2ff7e61e | 2054 | int btrfs_discard_extent(struct btrfs_fs_info *fs_info, u64 bytenr, |
1edb647b | 2055 | u64 num_bytes, u64 *actual_bytes) |
5d4f98a2 | 2056 | { |
5d4f98a2 | 2057 | int ret; |
5378e607 | 2058 | u64 discarded_bytes = 0; |
a1d3c478 | 2059 | struct btrfs_bio *bbio = NULL; |
5d4f98a2 | 2060 | |
e244a0ae | 2061 | |
2999241d FM |
2062 | /* |
2063 | * Avoid races with device replace and make sure our bbio has devices | |
2064 | * associated to its stripes that don't go away while we are discarding. | |
2065 | */ | |
0b246afa | 2066 | btrfs_bio_counter_inc_blocked(fs_info); |
5d4f98a2 | 2067 | /* Tell the block device(s) that the sectors can be discarded */ |
0b246afa JM |
2068 | ret = btrfs_map_block(fs_info, BTRFS_MAP_DISCARD, bytenr, &num_bytes, |
2069 | &bbio, 0); | |
79787eaa | 2070 | /* Error condition is -ENOMEM */ |
5d4f98a2 | 2071 | if (!ret) { |
a1d3c478 | 2072 | struct btrfs_bio_stripe *stripe = bbio->stripes; |
5d4f98a2 YZ |
2073 | int i; |
2074 | ||
5d4f98a2 | 2075 | |
a1d3c478 | 2076 | for (i = 0; i < bbio->num_stripes; i++, stripe++) { |
d04c6b88 | 2077 | u64 bytes; |
d5e2003c JB |
2078 | if (!stripe->dev->can_discard) |
2079 | continue; | |
2080 | ||
5378e607 LD |
2081 | ret = btrfs_issue_discard(stripe->dev->bdev, |
2082 | stripe->physical, | |
d04c6b88 JM |
2083 | stripe->length, |
2084 | &bytes); | |
5378e607 | 2085 | if (!ret) |
d04c6b88 | 2086 | discarded_bytes += bytes; |
5378e607 | 2087 | else if (ret != -EOPNOTSUPP) |
79787eaa | 2088 | break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */ |
d5e2003c JB |
2089 | |
2090 | /* | |
2091 | * Just in case we get back EOPNOTSUPP for some reason, | |
2092 | * just ignore the return value so we don't screw up | |
2093 | * people calling discard_extent. | |
2094 | */ | |
2095 | ret = 0; | |
5d4f98a2 | 2096 | } |
6e9606d2 | 2097 | btrfs_put_bbio(bbio); |
5d4f98a2 | 2098 | } |
0b246afa | 2099 | btrfs_bio_counter_dec(fs_info); |
5378e607 LD |
2100 | |
2101 | if (actual_bytes) | |
2102 | *actual_bytes = discarded_bytes; | |
2103 | ||
5d4f98a2 | 2104 | |
53b381b3 DW |
2105 | if (ret == -EOPNOTSUPP) |
2106 | ret = 0; | |
5d4f98a2 | 2107 | return ret; |
5d4f98a2 YZ |
2108 | } |
2109 | ||
79787eaa | 2110 | /* Can return -ENOMEM */ |
5d4f98a2 | 2111 | int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
2ff7e61e | 2112 | struct btrfs_fs_info *fs_info, |
5d4f98a2 | 2113 | u64 bytenr, u64 num_bytes, u64 parent, |
b06c4bf5 | 2114 | u64 root_objectid, u64 owner, u64 offset) |
5d4f98a2 | 2115 | { |
d7eae340 | 2116 | int old_ref_mod, new_ref_mod; |
5d4f98a2 | 2117 | int ret; |
66d7e7f0 | 2118 | |
5d4f98a2 YZ |
2119 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID && |
2120 | root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
2121 | ||
2122 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
66d7e7f0 | 2123 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr, |
7be07912 OS |
2124 | num_bytes, parent, |
2125 | root_objectid, (int)owner, | |
2126 | BTRFS_ADD_DELAYED_REF, NULL, | |
d7eae340 | 2127 | &old_ref_mod, &new_ref_mod); |
5d4f98a2 | 2128 | } else { |
66d7e7f0 | 2129 | ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr, |
7be07912 OS |
2130 | num_bytes, parent, |
2131 | root_objectid, owner, offset, | |
d7eae340 OS |
2132 | 0, BTRFS_ADD_DELAYED_REF, |
2133 | &old_ref_mod, &new_ref_mod); | |
5d4f98a2 | 2134 | } |
d7eae340 OS |
2135 | |
2136 | if (ret == 0 && old_ref_mod < 0 && new_ref_mod >= 0) | |
2137 | add_pinned_bytes(fs_info, -num_bytes, owner, root_objectid); | |
2138 | ||
5d4f98a2 YZ |
2139 | return ret; |
2140 | } | |
2141 | ||
2142 | static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, | |
2ff7e61e | 2143 | struct btrfs_fs_info *fs_info, |
c682f9b3 | 2144 | struct btrfs_delayed_ref_node *node, |
5d4f98a2 YZ |
2145 | u64 parent, u64 root_objectid, |
2146 | u64 owner, u64 offset, int refs_to_add, | |
2147 | struct btrfs_delayed_extent_op *extent_op) | |
2148 | { | |
2149 | struct btrfs_path *path; | |
2150 | struct extent_buffer *leaf; | |
2151 | struct btrfs_extent_item *item; | |
fcebe456 | 2152 | struct btrfs_key key; |
c682f9b3 QW |
2153 | u64 bytenr = node->bytenr; |
2154 | u64 num_bytes = node->num_bytes; | |
5d4f98a2 YZ |
2155 | u64 refs; |
2156 | int ret; | |
5d4f98a2 YZ |
2157 | |
2158 | path = btrfs_alloc_path(); | |
2159 | if (!path) | |
2160 | return -ENOMEM; | |
2161 | ||
e4058b54 | 2162 | path->reada = READA_FORWARD; |
5d4f98a2 YZ |
2163 | path->leave_spinning = 1; |
2164 | /* this will setup the path even if it fails to insert the back ref */ | |
87bde3cd JM |
2165 | ret = insert_inline_extent_backref(trans, fs_info, path, bytenr, |
2166 | num_bytes, parent, root_objectid, | |
2167 | owner, offset, | |
5d4f98a2 | 2168 | refs_to_add, extent_op); |
0ed4792a | 2169 | if ((ret < 0 && ret != -EAGAIN) || !ret) |
5d4f98a2 | 2170 | goto out; |
fcebe456 JB |
2171 | |
2172 | /* | |
2173 | * Ok we had -EAGAIN which means we didn't have space to insert and | |
2174 | * inline extent ref, so just update the reference count and add a | |
2175 | * normal backref. | |
2176 | */ | |
5d4f98a2 | 2177 | leaf = path->nodes[0]; |
fcebe456 | 2178 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
5d4f98a2 YZ |
2179 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
2180 | refs = btrfs_extent_refs(leaf, item); | |
2181 | btrfs_set_extent_refs(leaf, item, refs + refs_to_add); | |
2182 | if (extent_op) | |
2183 | __run_delayed_extent_op(extent_op, leaf, item); | |
56bec294 | 2184 | |
5d4f98a2 | 2185 | btrfs_mark_buffer_dirty(leaf); |
b3b4aa74 | 2186 | btrfs_release_path(path); |
56bec294 | 2187 | |
e4058b54 | 2188 | path->reada = READA_FORWARD; |
b9473439 | 2189 | path->leave_spinning = 1; |
56bec294 | 2190 | /* now insert the actual backref */ |
87bde3cd JM |
2191 | ret = insert_extent_backref(trans, fs_info, path, bytenr, parent, |
2192 | root_objectid, owner, offset, refs_to_add); | |
79787eaa | 2193 | if (ret) |
66642832 | 2194 | btrfs_abort_transaction(trans, ret); |
5d4f98a2 | 2195 | out: |
56bec294 | 2196 | btrfs_free_path(path); |
30d133fc | 2197 | return ret; |
56bec294 CM |
2198 | } |
2199 | ||
5d4f98a2 | 2200 | static int run_delayed_data_ref(struct btrfs_trans_handle *trans, |
2ff7e61e | 2201 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
2202 | struct btrfs_delayed_ref_node *node, |
2203 | struct btrfs_delayed_extent_op *extent_op, | |
2204 | int insert_reserved) | |
56bec294 | 2205 | { |
5d4f98a2 YZ |
2206 | int ret = 0; |
2207 | struct btrfs_delayed_data_ref *ref; | |
2208 | struct btrfs_key ins; | |
2209 | u64 parent = 0; | |
2210 | u64 ref_root = 0; | |
2211 | u64 flags = 0; | |
2212 | ||
2213 | ins.objectid = node->bytenr; | |
2214 | ins.offset = node->num_bytes; | |
2215 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
2216 | ||
2217 | ref = btrfs_delayed_node_to_data_ref(node); | |
0b246afa | 2218 | trace_run_delayed_data_ref(fs_info, node, ref, node->action); |
599c75ec | 2219 | |
5d4f98a2 YZ |
2220 | if (node->type == BTRFS_SHARED_DATA_REF_KEY) |
2221 | parent = ref->parent; | |
fcebe456 | 2222 | ref_root = ref->root; |
5d4f98a2 YZ |
2223 | |
2224 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
3173a18f | 2225 | if (extent_op) |
5d4f98a2 | 2226 | flags |= extent_op->flags_to_set; |
2ff7e61e | 2227 | ret = alloc_reserved_file_extent(trans, fs_info, |
5d4f98a2 YZ |
2228 | parent, ref_root, flags, |
2229 | ref->objectid, ref->offset, | |
2230 | &ins, node->ref_mod); | |
5d4f98a2 | 2231 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2ff7e61e | 2232 | ret = __btrfs_inc_extent_ref(trans, fs_info, node, parent, |
5d4f98a2 YZ |
2233 | ref_root, ref->objectid, |
2234 | ref->offset, node->ref_mod, | |
c682f9b3 | 2235 | extent_op); |
5d4f98a2 | 2236 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
2ff7e61e | 2237 | ret = __btrfs_free_extent(trans, fs_info, node, parent, |
5d4f98a2 YZ |
2238 | ref_root, ref->objectid, |
2239 | ref->offset, node->ref_mod, | |
c682f9b3 | 2240 | extent_op); |
5d4f98a2 YZ |
2241 | } else { |
2242 | BUG(); | |
2243 | } | |
2244 | return ret; | |
2245 | } | |
2246 | ||
2247 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
2248 | struct extent_buffer *leaf, | |
2249 | struct btrfs_extent_item *ei) | |
2250 | { | |
2251 | u64 flags = btrfs_extent_flags(leaf, ei); | |
2252 | if (extent_op->update_flags) { | |
2253 | flags |= extent_op->flags_to_set; | |
2254 | btrfs_set_extent_flags(leaf, ei, flags); | |
2255 | } | |
2256 | ||
2257 | if (extent_op->update_key) { | |
2258 | struct btrfs_tree_block_info *bi; | |
2259 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)); | |
2260 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
2261 | btrfs_set_tree_block_key(leaf, bi, &extent_op->key); | |
2262 | } | |
2263 | } | |
2264 | ||
2265 | static int run_delayed_extent_op(struct btrfs_trans_handle *trans, | |
2ff7e61e | 2266 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
2267 | struct btrfs_delayed_ref_node *node, |
2268 | struct btrfs_delayed_extent_op *extent_op) | |
2269 | { | |
2270 | struct btrfs_key key; | |
2271 | struct btrfs_path *path; | |
2272 | struct btrfs_extent_item *ei; | |
2273 | struct extent_buffer *leaf; | |
2274 | u32 item_size; | |
56bec294 | 2275 | int ret; |
5d4f98a2 | 2276 | int err = 0; |
b1c79e09 | 2277 | int metadata = !extent_op->is_data; |
5d4f98a2 | 2278 | |
79787eaa JM |
2279 | if (trans->aborted) |
2280 | return 0; | |
2281 | ||
0b246afa | 2282 | if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) |
3173a18f JB |
2283 | metadata = 0; |
2284 | ||
5d4f98a2 YZ |
2285 | path = btrfs_alloc_path(); |
2286 | if (!path) | |
2287 | return -ENOMEM; | |
2288 | ||
2289 | key.objectid = node->bytenr; | |
5d4f98a2 | 2290 | |
3173a18f | 2291 | if (metadata) { |
3173a18f | 2292 | key.type = BTRFS_METADATA_ITEM_KEY; |
b1c79e09 | 2293 | key.offset = extent_op->level; |
3173a18f JB |
2294 | } else { |
2295 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
2296 | key.offset = node->num_bytes; | |
2297 | } | |
2298 | ||
2299 | again: | |
e4058b54 | 2300 | path->reada = READA_FORWARD; |
5d4f98a2 | 2301 | path->leave_spinning = 1; |
0b246afa | 2302 | ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 1); |
5d4f98a2 YZ |
2303 | if (ret < 0) { |
2304 | err = ret; | |
2305 | goto out; | |
2306 | } | |
2307 | if (ret > 0) { | |
3173a18f | 2308 | if (metadata) { |
55994887 FDBM |
2309 | if (path->slots[0] > 0) { |
2310 | path->slots[0]--; | |
2311 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
2312 | path->slots[0]); | |
2313 | if (key.objectid == node->bytenr && | |
2314 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
2315 | key.offset == node->num_bytes) | |
2316 | ret = 0; | |
2317 | } | |
2318 | if (ret > 0) { | |
2319 | btrfs_release_path(path); | |
2320 | metadata = 0; | |
3173a18f | 2321 | |
55994887 FDBM |
2322 | key.objectid = node->bytenr; |
2323 | key.offset = node->num_bytes; | |
2324 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
2325 | goto again; | |
2326 | } | |
2327 | } else { | |
2328 | err = -EIO; | |
2329 | goto out; | |
3173a18f | 2330 | } |
5d4f98a2 YZ |
2331 | } |
2332 | ||
2333 | leaf = path->nodes[0]; | |
2334 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2335 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
2336 | if (item_size < sizeof(*ei)) { | |
87bde3cd | 2337 | ret = convert_extent_item_v0(trans, fs_info, path, (u64)-1, 0); |
5d4f98a2 YZ |
2338 | if (ret < 0) { |
2339 | err = ret; | |
2340 | goto out; | |
2341 | } | |
2342 | leaf = path->nodes[0]; | |
2343 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2344 | } | |
2345 | #endif | |
2346 | BUG_ON(item_size < sizeof(*ei)); | |
2347 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
2348 | __run_delayed_extent_op(extent_op, leaf, ei); | |
56bec294 | 2349 | |
5d4f98a2 YZ |
2350 | btrfs_mark_buffer_dirty(leaf); |
2351 | out: | |
2352 | btrfs_free_path(path); | |
2353 | return err; | |
56bec294 CM |
2354 | } |
2355 | ||
5d4f98a2 | 2356 | static int run_delayed_tree_ref(struct btrfs_trans_handle *trans, |
2ff7e61e | 2357 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
2358 | struct btrfs_delayed_ref_node *node, |
2359 | struct btrfs_delayed_extent_op *extent_op, | |
2360 | int insert_reserved) | |
56bec294 CM |
2361 | { |
2362 | int ret = 0; | |
5d4f98a2 YZ |
2363 | struct btrfs_delayed_tree_ref *ref; |
2364 | struct btrfs_key ins; | |
2365 | u64 parent = 0; | |
2366 | u64 ref_root = 0; | |
0b246afa | 2367 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); |
56bec294 | 2368 | |
5d4f98a2 | 2369 | ref = btrfs_delayed_node_to_tree_ref(node); |
0b246afa | 2370 | trace_run_delayed_tree_ref(fs_info, node, ref, node->action); |
599c75ec | 2371 | |
5d4f98a2 YZ |
2372 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) |
2373 | parent = ref->parent; | |
fcebe456 | 2374 | ref_root = ref->root; |
5d4f98a2 | 2375 | |
3173a18f JB |
2376 | ins.objectid = node->bytenr; |
2377 | if (skinny_metadata) { | |
2378 | ins.offset = ref->level; | |
2379 | ins.type = BTRFS_METADATA_ITEM_KEY; | |
2380 | } else { | |
2381 | ins.offset = node->num_bytes; | |
2382 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
2383 | } | |
2384 | ||
02794222 | 2385 | if (node->ref_mod != 1) { |
2ff7e61e | 2386 | btrfs_err(fs_info, |
02794222 LB |
2387 | "btree block(%llu) has %d references rather than 1: action %d ref_root %llu parent %llu", |
2388 | node->bytenr, node->ref_mod, node->action, ref_root, | |
2389 | parent); | |
2390 | return -EIO; | |
2391 | } | |
5d4f98a2 | 2392 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { |
3173a18f | 2393 | BUG_ON(!extent_op || !extent_op->update_flags); |
2ff7e61e | 2394 | ret = alloc_reserved_tree_block(trans, fs_info, |
5d4f98a2 YZ |
2395 | parent, ref_root, |
2396 | extent_op->flags_to_set, | |
2397 | &extent_op->key, | |
b06c4bf5 | 2398 | ref->level, &ins); |
5d4f98a2 | 2399 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2ff7e61e | 2400 | ret = __btrfs_inc_extent_ref(trans, fs_info, node, |
c682f9b3 QW |
2401 | parent, ref_root, |
2402 | ref->level, 0, 1, | |
fcebe456 | 2403 | extent_op); |
5d4f98a2 | 2404 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
2ff7e61e | 2405 | ret = __btrfs_free_extent(trans, fs_info, node, |
c682f9b3 QW |
2406 | parent, ref_root, |
2407 | ref->level, 0, 1, extent_op); | |
5d4f98a2 YZ |
2408 | } else { |
2409 | BUG(); | |
2410 | } | |
56bec294 CM |
2411 | return ret; |
2412 | } | |
2413 | ||
2414 | /* helper function to actually process a single delayed ref entry */ | |
5d4f98a2 | 2415 | static int run_one_delayed_ref(struct btrfs_trans_handle *trans, |
2ff7e61e | 2416 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
2417 | struct btrfs_delayed_ref_node *node, |
2418 | struct btrfs_delayed_extent_op *extent_op, | |
2419 | int insert_reserved) | |
56bec294 | 2420 | { |
79787eaa JM |
2421 | int ret = 0; |
2422 | ||
857cc2fc JB |
2423 | if (trans->aborted) { |
2424 | if (insert_reserved) | |
2ff7e61e | 2425 | btrfs_pin_extent(fs_info, node->bytenr, |
857cc2fc | 2426 | node->num_bytes, 1); |
79787eaa | 2427 | return 0; |
857cc2fc | 2428 | } |
79787eaa | 2429 | |
5d4f98a2 | 2430 | if (btrfs_delayed_ref_is_head(node)) { |
56bec294 CM |
2431 | struct btrfs_delayed_ref_head *head; |
2432 | /* | |
2433 | * we've hit the end of the chain and we were supposed | |
2434 | * to insert this extent into the tree. But, it got | |
2435 | * deleted before we ever needed to insert it, so all | |
2436 | * we have to do is clean up the accounting | |
2437 | */ | |
5d4f98a2 YZ |
2438 | BUG_ON(extent_op); |
2439 | head = btrfs_delayed_node_to_head(node); | |
0b246afa | 2440 | trace_run_delayed_ref_head(fs_info, node, head, node->action); |
599c75ec | 2441 | |
d7eae340 OS |
2442 | if (head->total_ref_mod < 0) { |
2443 | struct btrfs_block_group_cache *cache; | |
2444 | ||
2445 | cache = btrfs_lookup_block_group(fs_info, node->bytenr); | |
2446 | ASSERT(cache); | |
2447 | percpu_counter_add(&cache->space_info->total_bytes_pinned, | |
2448 | -node->num_bytes); | |
2449 | btrfs_put_block_group(cache); | |
2450 | } | |
2451 | ||
56bec294 | 2452 | if (insert_reserved) { |
2ff7e61e | 2453 | btrfs_pin_extent(fs_info, node->bytenr, |
f0486c68 | 2454 | node->num_bytes, 1); |
5d4f98a2 | 2455 | if (head->is_data) { |
0b246afa | 2456 | ret = btrfs_del_csums(trans, fs_info, |
5d4f98a2 YZ |
2457 | node->bytenr, |
2458 | node->num_bytes); | |
5d4f98a2 | 2459 | } |
56bec294 | 2460 | } |
297d750b QW |
2461 | |
2462 | /* Also free its reserved qgroup space */ | |
0b246afa | 2463 | btrfs_qgroup_free_delayed_ref(fs_info, head->qgroup_ref_root, |
297d750b | 2464 | head->qgroup_reserved); |
79787eaa | 2465 | return ret; |
56bec294 CM |
2466 | } |
2467 | ||
5d4f98a2 YZ |
2468 | if (node->type == BTRFS_TREE_BLOCK_REF_KEY || |
2469 | node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
2ff7e61e | 2470 | ret = run_delayed_tree_ref(trans, fs_info, node, extent_op, |
5d4f98a2 YZ |
2471 | insert_reserved); |
2472 | else if (node->type == BTRFS_EXTENT_DATA_REF_KEY || | |
2473 | node->type == BTRFS_SHARED_DATA_REF_KEY) | |
2ff7e61e | 2474 | ret = run_delayed_data_ref(trans, fs_info, node, extent_op, |
5d4f98a2 YZ |
2475 | insert_reserved); |
2476 | else | |
2477 | BUG(); | |
2478 | return ret; | |
56bec294 CM |
2479 | } |
2480 | ||
c6fc2454 | 2481 | static inline struct btrfs_delayed_ref_node * |
56bec294 CM |
2482 | select_delayed_ref(struct btrfs_delayed_ref_head *head) |
2483 | { | |
cffc3374 FM |
2484 | struct btrfs_delayed_ref_node *ref; |
2485 | ||
c6fc2454 QW |
2486 | if (list_empty(&head->ref_list)) |
2487 | return NULL; | |
d7df2c79 | 2488 | |
cffc3374 FM |
2489 | /* |
2490 | * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first. | |
2491 | * This is to prevent a ref count from going down to zero, which deletes | |
2492 | * the extent item from the extent tree, when there still are references | |
2493 | * to add, which would fail because they would not find the extent item. | |
2494 | */ | |
1d57ee94 WX |
2495 | if (!list_empty(&head->ref_add_list)) |
2496 | return list_first_entry(&head->ref_add_list, | |
2497 | struct btrfs_delayed_ref_node, add_list); | |
2498 | ||
2499 | ref = list_first_entry(&head->ref_list, struct btrfs_delayed_ref_node, | |
2500 | list); | |
2501 | ASSERT(list_empty(&ref->add_list)); | |
2502 | return ref; | |
56bec294 CM |
2503 | } |
2504 | ||
79787eaa JM |
2505 | /* |
2506 | * Returns 0 on success or if called with an already aborted transaction. | |
2507 | * Returns -ENOMEM or -EIO on failure and will abort the transaction. | |
2508 | */ | |
d7df2c79 | 2509 | static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, |
2ff7e61e | 2510 | struct btrfs_fs_info *fs_info, |
d7df2c79 | 2511 | unsigned long nr) |
56bec294 | 2512 | { |
56bec294 CM |
2513 | struct btrfs_delayed_ref_root *delayed_refs; |
2514 | struct btrfs_delayed_ref_node *ref; | |
2515 | struct btrfs_delayed_ref_head *locked_ref = NULL; | |
5d4f98a2 | 2516 | struct btrfs_delayed_extent_op *extent_op; |
0a2b2a84 | 2517 | ktime_t start = ktime_get(); |
56bec294 | 2518 | int ret; |
d7df2c79 | 2519 | unsigned long count = 0; |
0a2b2a84 | 2520 | unsigned long actual_count = 0; |
56bec294 | 2521 | int must_insert_reserved = 0; |
56bec294 CM |
2522 | |
2523 | delayed_refs = &trans->transaction->delayed_refs; | |
56bec294 CM |
2524 | while (1) { |
2525 | if (!locked_ref) { | |
d7df2c79 | 2526 | if (count >= nr) |
56bec294 | 2527 | break; |
56bec294 | 2528 | |
d7df2c79 JB |
2529 | spin_lock(&delayed_refs->lock); |
2530 | locked_ref = btrfs_select_ref_head(trans); | |
2531 | if (!locked_ref) { | |
2532 | spin_unlock(&delayed_refs->lock); | |
2533 | break; | |
2534 | } | |
c3e69d58 CM |
2535 | |
2536 | /* grab the lock that says we are going to process | |
2537 | * all the refs for this head */ | |
2538 | ret = btrfs_delayed_ref_lock(trans, locked_ref); | |
d7df2c79 | 2539 | spin_unlock(&delayed_refs->lock); |
c3e69d58 CM |
2540 | /* |
2541 | * we may have dropped the spin lock to get the head | |
2542 | * mutex lock, and that might have given someone else | |
2543 | * time to free the head. If that's true, it has been | |
2544 | * removed from our list and we can move on. | |
2545 | */ | |
2546 | if (ret == -EAGAIN) { | |
2547 | locked_ref = NULL; | |
2548 | count++; | |
2549 | continue; | |
56bec294 CM |
2550 | } |
2551 | } | |
a28ec197 | 2552 | |
2c3cf7d5 FM |
2553 | /* |
2554 | * We need to try and merge add/drops of the same ref since we | |
2555 | * can run into issues with relocate dropping the implicit ref | |
2556 | * and then it being added back again before the drop can | |
2557 | * finish. If we merged anything we need to re-loop so we can | |
2558 | * get a good ref. | |
2559 | * Or we can get node references of the same type that weren't | |
2560 | * merged when created due to bumps in the tree mod seq, and | |
2561 | * we need to merge them to prevent adding an inline extent | |
2562 | * backref before dropping it (triggering a BUG_ON at | |
2563 | * insert_inline_extent_backref()). | |
2564 | */ | |
d7df2c79 | 2565 | spin_lock(&locked_ref->lock); |
2c3cf7d5 FM |
2566 | btrfs_merge_delayed_refs(trans, fs_info, delayed_refs, |
2567 | locked_ref); | |
ae1e206b | 2568 | |
d1270cd9 AJ |
2569 | /* |
2570 | * locked_ref is the head node, so we have to go one | |
2571 | * node back for any delayed ref updates | |
2572 | */ | |
2573 | ref = select_delayed_ref(locked_ref); | |
2574 | ||
2575 | if (ref && ref->seq && | |
097b8a7c | 2576 | btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) { |
d7df2c79 | 2577 | spin_unlock(&locked_ref->lock); |
d7df2c79 JB |
2578 | spin_lock(&delayed_refs->lock); |
2579 | locked_ref->processing = 0; | |
d1270cd9 AJ |
2580 | delayed_refs->num_heads_ready++; |
2581 | spin_unlock(&delayed_refs->lock); | |
d0280996 | 2582 | btrfs_delayed_ref_unlock(locked_ref); |
d7df2c79 | 2583 | locked_ref = NULL; |
d1270cd9 | 2584 | cond_resched(); |
27a377db | 2585 | count++; |
d1270cd9 AJ |
2586 | continue; |
2587 | } | |
2588 | ||
56bec294 CM |
2589 | /* |
2590 | * record the must insert reserved flag before we | |
2591 | * drop the spin lock. | |
2592 | */ | |
2593 | must_insert_reserved = locked_ref->must_insert_reserved; | |
2594 | locked_ref->must_insert_reserved = 0; | |
7bb86316 | 2595 | |
5d4f98a2 YZ |
2596 | extent_op = locked_ref->extent_op; |
2597 | locked_ref->extent_op = NULL; | |
2598 | ||
56bec294 | 2599 | if (!ref) { |
d7df2c79 JB |
2600 | |
2601 | ||
56bec294 CM |
2602 | /* All delayed refs have been processed, Go ahead |
2603 | * and send the head node to run_one_delayed_ref, | |
2604 | * so that any accounting fixes can happen | |
2605 | */ | |
2606 | ref = &locked_ref->node; | |
5d4f98a2 YZ |
2607 | |
2608 | if (extent_op && must_insert_reserved) { | |
78a6184a | 2609 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 YZ |
2610 | extent_op = NULL; |
2611 | } | |
2612 | ||
2613 | if (extent_op) { | |
d7df2c79 | 2614 | spin_unlock(&locked_ref->lock); |
2ff7e61e | 2615 | ret = run_delayed_extent_op(trans, fs_info, |
5d4f98a2 | 2616 | ref, extent_op); |
78a6184a | 2617 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 | 2618 | |
79787eaa | 2619 | if (ret) { |
857cc2fc JB |
2620 | /* |
2621 | * Need to reset must_insert_reserved if | |
2622 | * there was an error so the abort stuff | |
2623 | * can cleanup the reserved space | |
2624 | * properly. | |
2625 | */ | |
2626 | if (must_insert_reserved) | |
2627 | locked_ref->must_insert_reserved = 1; | |
aa7c8da3 | 2628 | spin_lock(&delayed_refs->lock); |
d7df2c79 | 2629 | locked_ref->processing = 0; |
aa7c8da3 JM |
2630 | delayed_refs->num_heads_ready++; |
2631 | spin_unlock(&delayed_refs->lock); | |
5d163e0e JM |
2632 | btrfs_debug(fs_info, |
2633 | "run_delayed_extent_op returned %d", | |
2634 | ret); | |
093486c4 | 2635 | btrfs_delayed_ref_unlock(locked_ref); |
79787eaa JM |
2636 | return ret; |
2637 | } | |
d7df2c79 | 2638 | continue; |
5d4f98a2 | 2639 | } |
02217ed2 | 2640 | |
d7df2c79 | 2641 | /* |
01327610 | 2642 | * Need to drop our head ref lock and re-acquire the |
d7df2c79 JB |
2643 | * delayed ref lock and then re-check to make sure |
2644 | * nobody got added. | |
2645 | */ | |
2646 | spin_unlock(&locked_ref->lock); | |
2647 | spin_lock(&delayed_refs->lock); | |
2648 | spin_lock(&locked_ref->lock); | |
c6fc2454 | 2649 | if (!list_empty(&locked_ref->ref_list) || |
573a0755 | 2650 | locked_ref->extent_op) { |
d7df2c79 JB |
2651 | spin_unlock(&locked_ref->lock); |
2652 | spin_unlock(&delayed_refs->lock); | |
2653 | continue; | |
2654 | } | |
2655 | ref->in_tree = 0; | |
2656 | delayed_refs->num_heads--; | |
c46effa6 LB |
2657 | rb_erase(&locked_ref->href_node, |
2658 | &delayed_refs->href_root); | |
d7df2c79 JB |
2659 | spin_unlock(&delayed_refs->lock); |
2660 | } else { | |
0a2b2a84 | 2661 | actual_count++; |
d7df2c79 | 2662 | ref->in_tree = 0; |
c6fc2454 | 2663 | list_del(&ref->list); |
1d57ee94 WX |
2664 | if (!list_empty(&ref->add_list)) |
2665 | list_del(&ref->add_list); | |
c46effa6 | 2666 | } |
d7df2c79 JB |
2667 | atomic_dec(&delayed_refs->num_entries); |
2668 | ||
093486c4 | 2669 | if (!btrfs_delayed_ref_is_head(ref)) { |
22cd2e7d AJ |
2670 | /* |
2671 | * when we play the delayed ref, also correct the | |
2672 | * ref_mod on head | |
2673 | */ | |
2674 | switch (ref->action) { | |
2675 | case BTRFS_ADD_DELAYED_REF: | |
2676 | case BTRFS_ADD_DELAYED_EXTENT: | |
2677 | locked_ref->node.ref_mod -= ref->ref_mod; | |
2678 | break; | |
2679 | case BTRFS_DROP_DELAYED_REF: | |
2680 | locked_ref->node.ref_mod += ref->ref_mod; | |
2681 | break; | |
2682 | default: | |
2683 | WARN_ON(1); | |
2684 | } | |
2685 | } | |
d7df2c79 | 2686 | spin_unlock(&locked_ref->lock); |
925baedd | 2687 | |
2ff7e61e | 2688 | ret = run_one_delayed_ref(trans, fs_info, ref, extent_op, |
56bec294 | 2689 | must_insert_reserved); |
eb099670 | 2690 | |
78a6184a | 2691 | btrfs_free_delayed_extent_op(extent_op); |
79787eaa | 2692 | if (ret) { |
9d1032cc | 2693 | spin_lock(&delayed_refs->lock); |
d7df2c79 | 2694 | locked_ref->processing = 0; |
9d1032cc WX |
2695 | delayed_refs->num_heads_ready++; |
2696 | spin_unlock(&delayed_refs->lock); | |
093486c4 MX |
2697 | btrfs_delayed_ref_unlock(locked_ref); |
2698 | btrfs_put_delayed_ref(ref); | |
5d163e0e JM |
2699 | btrfs_debug(fs_info, "run_one_delayed_ref returned %d", |
2700 | ret); | |
79787eaa JM |
2701 | return ret; |
2702 | } | |
2703 | ||
093486c4 MX |
2704 | /* |
2705 | * If this node is a head, that means all the refs in this head | |
2706 | * have been dealt with, and we will pick the next head to deal | |
2707 | * with, so we must unlock the head and drop it from the cluster | |
2708 | * list before we release it. | |
2709 | */ | |
2710 | if (btrfs_delayed_ref_is_head(ref)) { | |
1262133b JB |
2711 | if (locked_ref->is_data && |
2712 | locked_ref->total_ref_mod < 0) { | |
2713 | spin_lock(&delayed_refs->lock); | |
2714 | delayed_refs->pending_csums -= ref->num_bytes; | |
2715 | spin_unlock(&delayed_refs->lock); | |
2716 | } | |
093486c4 MX |
2717 | btrfs_delayed_ref_unlock(locked_ref); |
2718 | locked_ref = NULL; | |
2719 | } | |
2720 | btrfs_put_delayed_ref(ref); | |
2721 | count++; | |
c3e69d58 | 2722 | cond_resched(); |
c3e69d58 | 2723 | } |
0a2b2a84 JB |
2724 | |
2725 | /* | |
2726 | * We don't want to include ref heads since we can have empty ref heads | |
2727 | * and those will drastically skew our runtime down since we just do | |
2728 | * accounting, no actual extent tree updates. | |
2729 | */ | |
2730 | if (actual_count > 0) { | |
2731 | u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start)); | |
2732 | u64 avg; | |
2733 | ||
2734 | /* | |
2735 | * We weigh the current average higher than our current runtime | |
2736 | * to avoid large swings in the average. | |
2737 | */ | |
2738 | spin_lock(&delayed_refs->lock); | |
2739 | avg = fs_info->avg_delayed_ref_runtime * 3 + runtime; | |
f8c269d7 | 2740 | fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */ |
0a2b2a84 JB |
2741 | spin_unlock(&delayed_refs->lock); |
2742 | } | |
d7df2c79 | 2743 | return 0; |
c3e69d58 CM |
2744 | } |
2745 | ||
709c0486 AJ |
2746 | #ifdef SCRAMBLE_DELAYED_REFS |
2747 | /* | |
2748 | * Normally delayed refs get processed in ascending bytenr order. This | |
2749 | * correlates in most cases to the order added. To expose dependencies on this | |
2750 | * order, we start to process the tree in the middle instead of the beginning | |
2751 | */ | |
2752 | static u64 find_middle(struct rb_root *root) | |
2753 | { | |
2754 | struct rb_node *n = root->rb_node; | |
2755 | struct btrfs_delayed_ref_node *entry; | |
2756 | int alt = 1; | |
2757 | u64 middle; | |
2758 | u64 first = 0, last = 0; | |
2759 | ||
2760 | n = rb_first(root); | |
2761 | if (n) { | |
2762 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2763 | first = entry->bytenr; | |
2764 | } | |
2765 | n = rb_last(root); | |
2766 | if (n) { | |
2767 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2768 | last = entry->bytenr; | |
2769 | } | |
2770 | n = root->rb_node; | |
2771 | ||
2772 | while (n) { | |
2773 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2774 | WARN_ON(!entry->in_tree); | |
2775 | ||
2776 | middle = entry->bytenr; | |
2777 | ||
2778 | if (alt) | |
2779 | n = n->rb_left; | |
2780 | else | |
2781 | n = n->rb_right; | |
2782 | ||
2783 | alt = 1 - alt; | |
2784 | } | |
2785 | return middle; | |
2786 | } | |
2787 | #endif | |
2788 | ||
2ff7e61e | 2789 | static inline u64 heads_to_leaves(struct btrfs_fs_info *fs_info, u64 heads) |
1be41b78 JB |
2790 | { |
2791 | u64 num_bytes; | |
2792 | ||
2793 | num_bytes = heads * (sizeof(struct btrfs_extent_item) + | |
2794 | sizeof(struct btrfs_extent_inline_ref)); | |
0b246afa | 2795 | if (!btrfs_fs_incompat(fs_info, SKINNY_METADATA)) |
1be41b78 JB |
2796 | num_bytes += heads * sizeof(struct btrfs_tree_block_info); |
2797 | ||
2798 | /* | |
2799 | * We don't ever fill up leaves all the way so multiply by 2 just to be | |
01327610 | 2800 | * closer to what we're really going to want to use. |
1be41b78 | 2801 | */ |
0b246afa | 2802 | return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(fs_info)); |
1be41b78 JB |
2803 | } |
2804 | ||
1262133b JB |
2805 | /* |
2806 | * Takes the number of bytes to be csumm'ed and figures out how many leaves it | |
2807 | * would require to store the csums for that many bytes. | |
2808 | */ | |
2ff7e61e | 2809 | u64 btrfs_csum_bytes_to_leaves(struct btrfs_fs_info *fs_info, u64 csum_bytes) |
1262133b JB |
2810 | { |
2811 | u64 csum_size; | |
2812 | u64 num_csums_per_leaf; | |
2813 | u64 num_csums; | |
2814 | ||
0b246afa | 2815 | csum_size = BTRFS_MAX_ITEM_SIZE(fs_info); |
1262133b | 2816 | num_csums_per_leaf = div64_u64(csum_size, |
0b246afa JM |
2817 | (u64)btrfs_super_csum_size(fs_info->super_copy)); |
2818 | num_csums = div64_u64(csum_bytes, fs_info->sectorsize); | |
1262133b JB |
2819 | num_csums += num_csums_per_leaf - 1; |
2820 | num_csums = div64_u64(num_csums, num_csums_per_leaf); | |
2821 | return num_csums; | |
2822 | } | |
2823 | ||
0a2b2a84 | 2824 | int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans, |
2ff7e61e | 2825 | struct btrfs_fs_info *fs_info) |
1be41b78 JB |
2826 | { |
2827 | struct btrfs_block_rsv *global_rsv; | |
2828 | u64 num_heads = trans->transaction->delayed_refs.num_heads_ready; | |
1262133b | 2829 | u64 csum_bytes = trans->transaction->delayed_refs.pending_csums; |
cb723e49 JB |
2830 | u64 num_dirty_bgs = trans->transaction->num_dirty_bgs; |
2831 | u64 num_bytes, num_dirty_bgs_bytes; | |
1be41b78 JB |
2832 | int ret = 0; |
2833 | ||
0b246afa | 2834 | num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1); |
2ff7e61e | 2835 | num_heads = heads_to_leaves(fs_info, num_heads); |
1be41b78 | 2836 | if (num_heads > 1) |
0b246afa | 2837 | num_bytes += (num_heads - 1) * fs_info->nodesize; |
1be41b78 | 2838 | num_bytes <<= 1; |
2ff7e61e JM |
2839 | num_bytes += btrfs_csum_bytes_to_leaves(fs_info, csum_bytes) * |
2840 | fs_info->nodesize; | |
0b246afa | 2841 | num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(fs_info, |
cb723e49 | 2842 | num_dirty_bgs); |
0b246afa | 2843 | global_rsv = &fs_info->global_block_rsv; |
1be41b78 JB |
2844 | |
2845 | /* | |
2846 | * If we can't allocate any more chunks lets make sure we have _lots_ of | |
2847 | * wiggle room since running delayed refs can create more delayed refs. | |
2848 | */ | |
cb723e49 JB |
2849 | if (global_rsv->space_info->full) { |
2850 | num_dirty_bgs_bytes <<= 1; | |
1be41b78 | 2851 | num_bytes <<= 1; |
cb723e49 | 2852 | } |
1be41b78 JB |
2853 | |
2854 | spin_lock(&global_rsv->lock); | |
cb723e49 | 2855 | if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes) |
1be41b78 JB |
2856 | ret = 1; |
2857 | spin_unlock(&global_rsv->lock); | |
2858 | return ret; | |
2859 | } | |
2860 | ||
0a2b2a84 | 2861 | int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans, |
2ff7e61e | 2862 | struct btrfs_fs_info *fs_info) |
0a2b2a84 | 2863 | { |
0a2b2a84 JB |
2864 | u64 num_entries = |
2865 | atomic_read(&trans->transaction->delayed_refs.num_entries); | |
2866 | u64 avg_runtime; | |
a79b7d4b | 2867 | u64 val; |
0a2b2a84 JB |
2868 | |
2869 | smp_mb(); | |
2870 | avg_runtime = fs_info->avg_delayed_ref_runtime; | |
a79b7d4b | 2871 | val = num_entries * avg_runtime; |
dc1a90c6 | 2872 | if (val >= NSEC_PER_SEC) |
0a2b2a84 | 2873 | return 1; |
a79b7d4b CM |
2874 | if (val >= NSEC_PER_SEC / 2) |
2875 | return 2; | |
0a2b2a84 | 2876 | |
2ff7e61e | 2877 | return btrfs_check_space_for_delayed_refs(trans, fs_info); |
0a2b2a84 JB |
2878 | } |
2879 | ||
a79b7d4b CM |
2880 | struct async_delayed_refs { |
2881 | struct btrfs_root *root; | |
31b9655f | 2882 | u64 transid; |
a79b7d4b CM |
2883 | int count; |
2884 | int error; | |
2885 | int sync; | |
2886 | struct completion wait; | |
2887 | struct btrfs_work work; | |
2888 | }; | |
2889 | ||
2ff7e61e JM |
2890 | static inline struct async_delayed_refs * |
2891 | to_async_delayed_refs(struct btrfs_work *work) | |
2892 | { | |
2893 | return container_of(work, struct async_delayed_refs, work); | |
2894 | } | |
2895 | ||
a79b7d4b CM |
2896 | static void delayed_ref_async_start(struct btrfs_work *work) |
2897 | { | |
2ff7e61e | 2898 | struct async_delayed_refs *async = to_async_delayed_refs(work); |
a79b7d4b | 2899 | struct btrfs_trans_handle *trans; |
2ff7e61e | 2900 | struct btrfs_fs_info *fs_info = async->root->fs_info; |
a79b7d4b CM |
2901 | int ret; |
2902 | ||
0f873eca | 2903 | /* if the commit is already started, we don't need to wait here */ |
2ff7e61e | 2904 | if (btrfs_transaction_blocked(fs_info)) |
31b9655f | 2905 | goto done; |
31b9655f | 2906 | |
0f873eca CM |
2907 | trans = btrfs_join_transaction(async->root); |
2908 | if (IS_ERR(trans)) { | |
2909 | async->error = PTR_ERR(trans); | |
a79b7d4b CM |
2910 | goto done; |
2911 | } | |
2912 | ||
2913 | /* | |
01327610 | 2914 | * trans->sync means that when we call end_transaction, we won't |
a79b7d4b CM |
2915 | * wait on delayed refs |
2916 | */ | |
2917 | trans->sync = true; | |
0f873eca CM |
2918 | |
2919 | /* Don't bother flushing if we got into a different transaction */ | |
2920 | if (trans->transid > async->transid) | |
2921 | goto end; | |
2922 | ||
2ff7e61e | 2923 | ret = btrfs_run_delayed_refs(trans, fs_info, async->count); |
a79b7d4b CM |
2924 | if (ret) |
2925 | async->error = ret; | |
0f873eca | 2926 | end: |
3a45bb20 | 2927 | ret = btrfs_end_transaction(trans); |
a79b7d4b CM |
2928 | if (ret && !async->error) |
2929 | async->error = ret; | |
2930 | done: | |
2931 | if (async->sync) | |
2932 | complete(&async->wait); | |
2933 | else | |
2934 | kfree(async); | |
2935 | } | |
2936 | ||
2ff7e61e | 2937 | int btrfs_async_run_delayed_refs(struct btrfs_fs_info *fs_info, |
31b9655f | 2938 | unsigned long count, u64 transid, int wait) |
a79b7d4b CM |
2939 | { |
2940 | struct async_delayed_refs *async; | |
2941 | int ret; | |
2942 | ||
2943 | async = kmalloc(sizeof(*async), GFP_NOFS); | |
2944 | if (!async) | |
2945 | return -ENOMEM; | |
2946 | ||
0b246afa | 2947 | async->root = fs_info->tree_root; |
a79b7d4b CM |
2948 | async->count = count; |
2949 | async->error = 0; | |
31b9655f | 2950 | async->transid = transid; |
a79b7d4b CM |
2951 | if (wait) |
2952 | async->sync = 1; | |
2953 | else | |
2954 | async->sync = 0; | |
2955 | init_completion(&async->wait); | |
2956 | ||
9e0af237 LB |
2957 | btrfs_init_work(&async->work, btrfs_extent_refs_helper, |
2958 | delayed_ref_async_start, NULL, NULL); | |
a79b7d4b | 2959 | |
0b246afa | 2960 | btrfs_queue_work(fs_info->extent_workers, &async->work); |
a79b7d4b CM |
2961 | |
2962 | if (wait) { | |
2963 | wait_for_completion(&async->wait); | |
2964 | ret = async->error; | |
2965 | kfree(async); | |
2966 | return ret; | |
2967 | } | |
2968 | return 0; | |
2969 | } | |
2970 | ||
c3e69d58 CM |
2971 | /* |
2972 | * this starts processing the delayed reference count updates and | |
2973 | * extent insertions we have queued up so far. count can be | |
2974 | * 0, which means to process everything in the tree at the start | |
2975 | * of the run (but not newly added entries), or it can be some target | |
2976 | * number you'd like to process. | |
79787eaa JM |
2977 | * |
2978 | * Returns 0 on success or if called with an aborted transaction | |
2979 | * Returns <0 on error and aborts the transaction | |
c3e69d58 CM |
2980 | */ |
2981 | int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, | |
2ff7e61e | 2982 | struct btrfs_fs_info *fs_info, unsigned long count) |
c3e69d58 CM |
2983 | { |
2984 | struct rb_node *node; | |
2985 | struct btrfs_delayed_ref_root *delayed_refs; | |
c46effa6 | 2986 | struct btrfs_delayed_ref_head *head; |
c3e69d58 CM |
2987 | int ret; |
2988 | int run_all = count == (unsigned long)-1; | |
d9a0540a | 2989 | bool can_flush_pending_bgs = trans->can_flush_pending_bgs; |
c3e69d58 | 2990 | |
79787eaa JM |
2991 | /* We'll clean this up in btrfs_cleanup_transaction */ |
2992 | if (trans->aborted) | |
2993 | return 0; | |
2994 | ||
0b246afa | 2995 | if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags)) |
511711af CM |
2996 | return 0; |
2997 | ||
c3e69d58 | 2998 | delayed_refs = &trans->transaction->delayed_refs; |
26455d33 | 2999 | if (count == 0) |
d7df2c79 | 3000 | count = atomic_read(&delayed_refs->num_entries) * 2; |
bb721703 | 3001 | |
c3e69d58 | 3002 | again: |
709c0486 AJ |
3003 | #ifdef SCRAMBLE_DELAYED_REFS |
3004 | delayed_refs->run_delayed_start = find_middle(&delayed_refs->root); | |
3005 | #endif | |
d9a0540a | 3006 | trans->can_flush_pending_bgs = false; |
2ff7e61e | 3007 | ret = __btrfs_run_delayed_refs(trans, fs_info, count); |
d7df2c79 | 3008 | if (ret < 0) { |
66642832 | 3009 | btrfs_abort_transaction(trans, ret); |
d7df2c79 | 3010 | return ret; |
eb099670 | 3011 | } |
c3e69d58 | 3012 | |
56bec294 | 3013 | if (run_all) { |
d7df2c79 | 3014 | if (!list_empty(&trans->new_bgs)) |
2ff7e61e | 3015 | btrfs_create_pending_block_groups(trans, fs_info); |
ea658bad | 3016 | |
d7df2c79 | 3017 | spin_lock(&delayed_refs->lock); |
c46effa6 | 3018 | node = rb_first(&delayed_refs->href_root); |
d7df2c79 JB |
3019 | if (!node) { |
3020 | spin_unlock(&delayed_refs->lock); | |
56bec294 | 3021 | goto out; |
d7df2c79 | 3022 | } |
e9d0b13b | 3023 | |
56bec294 | 3024 | while (node) { |
c46effa6 LB |
3025 | head = rb_entry(node, struct btrfs_delayed_ref_head, |
3026 | href_node); | |
3027 | if (btrfs_delayed_ref_is_head(&head->node)) { | |
3028 | struct btrfs_delayed_ref_node *ref; | |
5caf2a00 | 3029 | |
c46effa6 | 3030 | ref = &head->node; |
6df8cdf5 | 3031 | refcount_inc(&ref->refs); |
56bec294 CM |
3032 | |
3033 | spin_unlock(&delayed_refs->lock); | |
8cc33e5c DS |
3034 | /* |
3035 | * Mutex was contended, block until it's | |
3036 | * released and try again | |
3037 | */ | |
56bec294 CM |
3038 | mutex_lock(&head->mutex); |
3039 | mutex_unlock(&head->mutex); | |
3040 | ||
3041 | btrfs_put_delayed_ref(ref); | |
1887be66 | 3042 | cond_resched(); |
56bec294 | 3043 | goto again; |
c46effa6 LB |
3044 | } else { |
3045 | WARN_ON(1); | |
56bec294 CM |
3046 | } |
3047 | node = rb_next(node); | |
3048 | } | |
3049 | spin_unlock(&delayed_refs->lock); | |
d7df2c79 | 3050 | cond_resched(); |
56bec294 | 3051 | goto again; |
5f39d397 | 3052 | } |
54aa1f4d | 3053 | out: |
d9a0540a | 3054 | trans->can_flush_pending_bgs = can_flush_pending_bgs; |
a28ec197 CM |
3055 | return 0; |
3056 | } | |
3057 | ||
5d4f98a2 | 3058 | int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans, |
2ff7e61e | 3059 | struct btrfs_fs_info *fs_info, |
5d4f98a2 | 3060 | u64 bytenr, u64 num_bytes, u64 flags, |
b1c79e09 | 3061 | int level, int is_data) |
5d4f98a2 YZ |
3062 | { |
3063 | struct btrfs_delayed_extent_op *extent_op; | |
3064 | int ret; | |
3065 | ||
78a6184a | 3066 | extent_op = btrfs_alloc_delayed_extent_op(); |
5d4f98a2 YZ |
3067 | if (!extent_op) |
3068 | return -ENOMEM; | |
3069 | ||
3070 | extent_op->flags_to_set = flags; | |
35b3ad50 DS |
3071 | extent_op->update_flags = true; |
3072 | extent_op->update_key = false; | |
3073 | extent_op->is_data = is_data ? true : false; | |
b1c79e09 | 3074 | extent_op->level = level; |
5d4f98a2 | 3075 | |
0b246afa | 3076 | ret = btrfs_add_delayed_extent_op(fs_info, trans, bytenr, |
66d7e7f0 | 3077 | num_bytes, extent_op); |
5d4f98a2 | 3078 | if (ret) |
78a6184a | 3079 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 YZ |
3080 | return ret; |
3081 | } | |
3082 | ||
e4c3b2dc | 3083 | static noinline int check_delayed_ref(struct btrfs_root *root, |
5d4f98a2 YZ |
3084 | struct btrfs_path *path, |
3085 | u64 objectid, u64 offset, u64 bytenr) | |
3086 | { | |
3087 | struct btrfs_delayed_ref_head *head; | |
3088 | struct btrfs_delayed_ref_node *ref; | |
3089 | struct btrfs_delayed_data_ref *data_ref; | |
3090 | struct btrfs_delayed_ref_root *delayed_refs; | |
e4c3b2dc | 3091 | struct btrfs_transaction *cur_trans; |
5d4f98a2 YZ |
3092 | int ret = 0; |
3093 | ||
e4c3b2dc LB |
3094 | cur_trans = root->fs_info->running_transaction; |
3095 | if (!cur_trans) | |
3096 | return 0; | |
3097 | ||
3098 | delayed_refs = &cur_trans->delayed_refs; | |
5d4f98a2 | 3099 | spin_lock(&delayed_refs->lock); |
f72ad18e | 3100 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
d7df2c79 JB |
3101 | if (!head) { |
3102 | spin_unlock(&delayed_refs->lock); | |
3103 | return 0; | |
3104 | } | |
5d4f98a2 YZ |
3105 | |
3106 | if (!mutex_trylock(&head->mutex)) { | |
6df8cdf5 | 3107 | refcount_inc(&head->node.refs); |
5d4f98a2 YZ |
3108 | spin_unlock(&delayed_refs->lock); |
3109 | ||
b3b4aa74 | 3110 | btrfs_release_path(path); |
5d4f98a2 | 3111 | |
8cc33e5c DS |
3112 | /* |
3113 | * Mutex was contended, block until it's released and let | |
3114 | * caller try again | |
3115 | */ | |
5d4f98a2 YZ |
3116 | mutex_lock(&head->mutex); |
3117 | mutex_unlock(&head->mutex); | |
3118 | btrfs_put_delayed_ref(&head->node); | |
3119 | return -EAGAIN; | |
3120 | } | |
d7df2c79 | 3121 | spin_unlock(&delayed_refs->lock); |
5d4f98a2 | 3122 | |
d7df2c79 | 3123 | spin_lock(&head->lock); |
c6fc2454 | 3124 | list_for_each_entry(ref, &head->ref_list, list) { |
d7df2c79 JB |
3125 | /* If it's a shared ref we know a cross reference exists */ |
3126 | if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) { | |
3127 | ret = 1; | |
3128 | break; | |
3129 | } | |
5d4f98a2 | 3130 | |
d7df2c79 | 3131 | data_ref = btrfs_delayed_node_to_data_ref(ref); |
5d4f98a2 | 3132 | |
d7df2c79 JB |
3133 | /* |
3134 | * If our ref doesn't match the one we're currently looking at | |
3135 | * then we have a cross reference. | |
3136 | */ | |
3137 | if (data_ref->root != root->root_key.objectid || | |
3138 | data_ref->objectid != objectid || | |
3139 | data_ref->offset != offset) { | |
3140 | ret = 1; | |
3141 | break; | |
3142 | } | |
5d4f98a2 | 3143 | } |
d7df2c79 | 3144 | spin_unlock(&head->lock); |
5d4f98a2 | 3145 | mutex_unlock(&head->mutex); |
5d4f98a2 YZ |
3146 | return ret; |
3147 | } | |
3148 | ||
e4c3b2dc | 3149 | static noinline int check_committed_ref(struct btrfs_root *root, |
5d4f98a2 YZ |
3150 | struct btrfs_path *path, |
3151 | u64 objectid, u64 offset, u64 bytenr) | |
be20aa9d | 3152 | { |
0b246afa JM |
3153 | struct btrfs_fs_info *fs_info = root->fs_info; |
3154 | struct btrfs_root *extent_root = fs_info->extent_root; | |
f321e491 | 3155 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
3156 | struct btrfs_extent_data_ref *ref; |
3157 | struct btrfs_extent_inline_ref *iref; | |
3158 | struct btrfs_extent_item *ei; | |
f321e491 | 3159 | struct btrfs_key key; |
5d4f98a2 | 3160 | u32 item_size; |
be20aa9d | 3161 | int ret; |
925baedd | 3162 | |
be20aa9d | 3163 | key.objectid = bytenr; |
31840ae1 | 3164 | key.offset = (u64)-1; |
f321e491 | 3165 | key.type = BTRFS_EXTENT_ITEM_KEY; |
be20aa9d | 3166 | |
be20aa9d CM |
3167 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
3168 | if (ret < 0) | |
3169 | goto out; | |
79787eaa | 3170 | BUG_ON(ret == 0); /* Corruption */ |
80ff3856 YZ |
3171 | |
3172 | ret = -ENOENT; | |
3173 | if (path->slots[0] == 0) | |
31840ae1 | 3174 | goto out; |
be20aa9d | 3175 | |
31840ae1 | 3176 | path->slots[0]--; |
f321e491 | 3177 | leaf = path->nodes[0]; |
5d4f98a2 | 3178 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
be20aa9d | 3179 | |
5d4f98a2 | 3180 | if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY) |
be20aa9d | 3181 | goto out; |
f321e491 | 3182 | |
5d4f98a2 YZ |
3183 | ret = 1; |
3184 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
3185 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
3186 | if (item_size < sizeof(*ei)) { | |
3187 | WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0)); | |
3188 | goto out; | |
3189 | } | |
3190 | #endif | |
3191 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
bd09835d | 3192 | |
5d4f98a2 YZ |
3193 | if (item_size != sizeof(*ei) + |
3194 | btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY)) | |
3195 | goto out; | |
be20aa9d | 3196 | |
5d4f98a2 YZ |
3197 | if (btrfs_extent_generation(leaf, ei) <= |
3198 | btrfs_root_last_snapshot(&root->root_item)) | |
3199 | goto out; | |
3200 | ||
3201 | iref = (struct btrfs_extent_inline_ref *)(ei + 1); | |
3202 | if (btrfs_extent_inline_ref_type(leaf, iref) != | |
3203 | BTRFS_EXTENT_DATA_REF_KEY) | |
3204 | goto out; | |
3205 | ||
3206 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
3207 | if (btrfs_extent_refs(leaf, ei) != | |
3208 | btrfs_extent_data_ref_count(leaf, ref) || | |
3209 | btrfs_extent_data_ref_root(leaf, ref) != | |
3210 | root->root_key.objectid || | |
3211 | btrfs_extent_data_ref_objectid(leaf, ref) != objectid || | |
3212 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
3213 | goto out; | |
3214 | ||
3215 | ret = 0; | |
3216 | out: | |
3217 | return ret; | |
3218 | } | |
3219 | ||
e4c3b2dc LB |
3220 | int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset, |
3221 | u64 bytenr) | |
5d4f98a2 YZ |
3222 | { |
3223 | struct btrfs_path *path; | |
3224 | int ret; | |
3225 | int ret2; | |
3226 | ||
3227 | path = btrfs_alloc_path(); | |
3228 | if (!path) | |
3229 | return -ENOENT; | |
3230 | ||
3231 | do { | |
e4c3b2dc | 3232 | ret = check_committed_ref(root, path, objectid, |
5d4f98a2 YZ |
3233 | offset, bytenr); |
3234 | if (ret && ret != -ENOENT) | |
f321e491 | 3235 | goto out; |
80ff3856 | 3236 | |
e4c3b2dc | 3237 | ret2 = check_delayed_ref(root, path, objectid, |
5d4f98a2 YZ |
3238 | offset, bytenr); |
3239 | } while (ret2 == -EAGAIN); | |
3240 | ||
3241 | if (ret2 && ret2 != -ENOENT) { | |
3242 | ret = ret2; | |
3243 | goto out; | |
f321e491 | 3244 | } |
5d4f98a2 YZ |
3245 | |
3246 | if (ret != -ENOENT || ret2 != -ENOENT) | |
3247 | ret = 0; | |
be20aa9d | 3248 | out: |
80ff3856 | 3249 | btrfs_free_path(path); |
f0486c68 YZ |
3250 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) |
3251 | WARN_ON(ret > 0); | |
f321e491 | 3252 | return ret; |
be20aa9d | 3253 | } |
c5739bba | 3254 | |
5d4f98a2 | 3255 | static int __btrfs_mod_ref(struct btrfs_trans_handle *trans, |
b7a9f29f | 3256 | struct btrfs_root *root, |
5d4f98a2 | 3257 | struct extent_buffer *buf, |
e339a6b0 | 3258 | int full_backref, int inc) |
31840ae1 | 3259 | { |
0b246afa | 3260 | struct btrfs_fs_info *fs_info = root->fs_info; |
31840ae1 | 3261 | u64 bytenr; |
5d4f98a2 YZ |
3262 | u64 num_bytes; |
3263 | u64 parent; | |
31840ae1 | 3264 | u64 ref_root; |
31840ae1 | 3265 | u32 nritems; |
31840ae1 ZY |
3266 | struct btrfs_key key; |
3267 | struct btrfs_file_extent_item *fi; | |
3268 | int i; | |
3269 | int level; | |
3270 | int ret = 0; | |
2ff7e61e JM |
3271 | int (*process_func)(struct btrfs_trans_handle *, |
3272 | struct btrfs_fs_info *, | |
b06c4bf5 | 3273 | u64, u64, u64, u64, u64, u64); |
31840ae1 | 3274 | |
fccb84c9 | 3275 | |
0b246afa | 3276 | if (btrfs_is_testing(fs_info)) |
faa2dbf0 | 3277 | return 0; |
fccb84c9 | 3278 | |
31840ae1 | 3279 | ref_root = btrfs_header_owner(buf); |
31840ae1 ZY |
3280 | nritems = btrfs_header_nritems(buf); |
3281 | level = btrfs_header_level(buf); | |
3282 | ||
27cdeb70 | 3283 | if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0) |
5d4f98a2 | 3284 | return 0; |
31840ae1 | 3285 | |
5d4f98a2 YZ |
3286 | if (inc) |
3287 | process_func = btrfs_inc_extent_ref; | |
3288 | else | |
3289 | process_func = btrfs_free_extent; | |
31840ae1 | 3290 | |
5d4f98a2 YZ |
3291 | if (full_backref) |
3292 | parent = buf->start; | |
3293 | else | |
3294 | parent = 0; | |
3295 | ||
3296 | for (i = 0; i < nritems; i++) { | |
31840ae1 | 3297 | if (level == 0) { |
5d4f98a2 | 3298 | btrfs_item_key_to_cpu(buf, &key, i); |
962a298f | 3299 | if (key.type != BTRFS_EXTENT_DATA_KEY) |
31840ae1 | 3300 | continue; |
5d4f98a2 | 3301 | fi = btrfs_item_ptr(buf, i, |
31840ae1 ZY |
3302 | struct btrfs_file_extent_item); |
3303 | if (btrfs_file_extent_type(buf, fi) == | |
3304 | BTRFS_FILE_EXTENT_INLINE) | |
3305 | continue; | |
3306 | bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
3307 | if (bytenr == 0) | |
3308 | continue; | |
5d4f98a2 YZ |
3309 | |
3310 | num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi); | |
3311 | key.offset -= btrfs_file_extent_offset(buf, fi); | |
2ff7e61e | 3312 | ret = process_func(trans, fs_info, bytenr, num_bytes, |
5d4f98a2 | 3313 | parent, ref_root, key.objectid, |
b06c4bf5 | 3314 | key.offset); |
31840ae1 ZY |
3315 | if (ret) |
3316 | goto fail; | |
3317 | } else { | |
5d4f98a2 | 3318 | bytenr = btrfs_node_blockptr(buf, i); |
0b246afa | 3319 | num_bytes = fs_info->nodesize; |
2ff7e61e | 3320 | ret = process_func(trans, fs_info, bytenr, num_bytes, |
b06c4bf5 | 3321 | parent, ref_root, level - 1, 0); |
31840ae1 ZY |
3322 | if (ret) |
3323 | goto fail; | |
3324 | } | |
3325 | } | |
3326 | return 0; | |
3327 | fail: | |
5d4f98a2 YZ |
3328 | return ret; |
3329 | } | |
3330 | ||
3331 | int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
e339a6b0 | 3332 | struct extent_buffer *buf, int full_backref) |
5d4f98a2 | 3333 | { |
e339a6b0 | 3334 | return __btrfs_mod_ref(trans, root, buf, full_backref, 1); |
5d4f98a2 YZ |
3335 | } |
3336 | ||
3337 | int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
e339a6b0 | 3338 | struct extent_buffer *buf, int full_backref) |
5d4f98a2 | 3339 | { |
e339a6b0 | 3340 | return __btrfs_mod_ref(trans, root, buf, full_backref, 0); |
31840ae1 ZY |
3341 | } |
3342 | ||
9078a3e1 | 3343 | static int write_one_cache_group(struct btrfs_trans_handle *trans, |
2ff7e61e | 3344 | struct btrfs_fs_info *fs_info, |
9078a3e1 CM |
3345 | struct btrfs_path *path, |
3346 | struct btrfs_block_group_cache *cache) | |
3347 | { | |
3348 | int ret; | |
0b246afa | 3349 | struct btrfs_root *extent_root = fs_info->extent_root; |
5f39d397 CM |
3350 | unsigned long bi; |
3351 | struct extent_buffer *leaf; | |
9078a3e1 | 3352 | |
9078a3e1 | 3353 | ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1); |
df95e7f0 JB |
3354 | if (ret) { |
3355 | if (ret > 0) | |
3356 | ret = -ENOENT; | |
54aa1f4d | 3357 | goto fail; |
df95e7f0 | 3358 | } |
5f39d397 CM |
3359 | |
3360 | leaf = path->nodes[0]; | |
3361 | bi = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
3362 | write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item)); | |
3363 | btrfs_mark_buffer_dirty(leaf); | |
54aa1f4d | 3364 | fail: |
24b89d08 | 3365 | btrfs_release_path(path); |
df95e7f0 | 3366 | return ret; |
9078a3e1 CM |
3367 | |
3368 | } | |
3369 | ||
4a8c9a62 | 3370 | static struct btrfs_block_group_cache * |
2ff7e61e | 3371 | next_block_group(struct btrfs_fs_info *fs_info, |
4a8c9a62 YZ |
3372 | struct btrfs_block_group_cache *cache) |
3373 | { | |
3374 | struct rb_node *node; | |
292cbd51 | 3375 | |
0b246afa | 3376 | spin_lock(&fs_info->block_group_cache_lock); |
292cbd51 FM |
3377 | |
3378 | /* If our block group was removed, we need a full search. */ | |
3379 | if (RB_EMPTY_NODE(&cache->cache_node)) { | |
3380 | const u64 next_bytenr = cache->key.objectid + cache->key.offset; | |
3381 | ||
0b246afa | 3382 | spin_unlock(&fs_info->block_group_cache_lock); |
292cbd51 | 3383 | btrfs_put_block_group(cache); |
0b246afa | 3384 | cache = btrfs_lookup_first_block_group(fs_info, next_bytenr); return cache; |
292cbd51 | 3385 | } |
4a8c9a62 YZ |
3386 | node = rb_next(&cache->cache_node); |
3387 | btrfs_put_block_group(cache); | |
3388 | if (node) { | |
3389 | cache = rb_entry(node, struct btrfs_block_group_cache, | |
3390 | cache_node); | |
11dfe35a | 3391 | btrfs_get_block_group(cache); |
4a8c9a62 YZ |
3392 | } else |
3393 | cache = NULL; | |
0b246afa | 3394 | spin_unlock(&fs_info->block_group_cache_lock); |
4a8c9a62 YZ |
3395 | return cache; |
3396 | } | |
3397 | ||
0af3d00b JB |
3398 | static int cache_save_setup(struct btrfs_block_group_cache *block_group, |
3399 | struct btrfs_trans_handle *trans, | |
3400 | struct btrfs_path *path) | |
3401 | { | |
0b246afa JM |
3402 | struct btrfs_fs_info *fs_info = block_group->fs_info; |
3403 | struct btrfs_root *root = fs_info->tree_root; | |
0af3d00b | 3404 | struct inode *inode = NULL; |
364ecf36 | 3405 | struct extent_changeset *data_reserved = NULL; |
0af3d00b | 3406 | u64 alloc_hint = 0; |
2b20982e | 3407 | int dcs = BTRFS_DC_ERROR; |
f8c269d7 | 3408 | u64 num_pages = 0; |
0af3d00b JB |
3409 | int retries = 0; |
3410 | int ret = 0; | |
3411 | ||
3412 | /* | |
3413 | * If this block group is smaller than 100 megs don't bother caching the | |
3414 | * block group. | |
3415 | */ | |
ee22184b | 3416 | if (block_group->key.offset < (100 * SZ_1M)) { |
0af3d00b JB |
3417 | spin_lock(&block_group->lock); |
3418 | block_group->disk_cache_state = BTRFS_DC_WRITTEN; | |
3419 | spin_unlock(&block_group->lock); | |
3420 | return 0; | |
3421 | } | |
3422 | ||
0c0ef4bc JB |
3423 | if (trans->aborted) |
3424 | return 0; | |
0af3d00b | 3425 | again: |
77ab86bf | 3426 | inode = lookup_free_space_inode(fs_info, block_group, path); |
0af3d00b JB |
3427 | if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) { |
3428 | ret = PTR_ERR(inode); | |
b3b4aa74 | 3429 | btrfs_release_path(path); |
0af3d00b JB |
3430 | goto out; |
3431 | } | |
3432 | ||
3433 | if (IS_ERR(inode)) { | |
3434 | BUG_ON(retries); | |
3435 | retries++; | |
3436 | ||
3437 | if (block_group->ro) | |
3438 | goto out_free; | |
3439 | ||
77ab86bf JM |
3440 | ret = create_free_space_inode(fs_info, trans, block_group, |
3441 | path); | |
0af3d00b JB |
3442 | if (ret) |
3443 | goto out_free; | |
3444 | goto again; | |
3445 | } | |
3446 | ||
5b0e95bf JB |
3447 | /* We've already setup this transaction, go ahead and exit */ |
3448 | if (block_group->cache_generation == trans->transid && | |
3449 | i_size_read(inode)) { | |
3450 | dcs = BTRFS_DC_SETUP; | |
3451 | goto out_put; | |
3452 | } | |
3453 | ||
0af3d00b JB |
3454 | /* |
3455 | * We want to set the generation to 0, that way if anything goes wrong | |
3456 | * from here on out we know not to trust this cache when we load up next | |
3457 | * time. | |
3458 | */ | |
3459 | BTRFS_I(inode)->generation = 0; | |
3460 | ret = btrfs_update_inode(trans, root, inode); | |
0c0ef4bc JB |
3461 | if (ret) { |
3462 | /* | |
3463 | * So theoretically we could recover from this, simply set the | |
3464 | * super cache generation to 0 so we know to invalidate the | |
3465 | * cache, but then we'd have to keep track of the block groups | |
3466 | * that fail this way so we know we _have_ to reset this cache | |
3467 | * before the next commit or risk reading stale cache. So to | |
3468 | * limit our exposure to horrible edge cases lets just abort the | |
3469 | * transaction, this only happens in really bad situations | |
3470 | * anyway. | |
3471 | */ | |
66642832 | 3472 | btrfs_abort_transaction(trans, ret); |
0c0ef4bc JB |
3473 | goto out_put; |
3474 | } | |
0af3d00b JB |
3475 | WARN_ON(ret); |
3476 | ||
3477 | if (i_size_read(inode) > 0) { | |
2ff7e61e | 3478 | ret = btrfs_check_trunc_cache_free_space(fs_info, |
0b246afa | 3479 | &fs_info->global_block_rsv); |
7b61cd92 MX |
3480 | if (ret) |
3481 | goto out_put; | |
3482 | ||
77ab86bf | 3483 | ret = btrfs_truncate_free_space_cache(trans, NULL, inode); |
0af3d00b JB |
3484 | if (ret) |
3485 | goto out_put; | |
3486 | } | |
3487 | ||
3488 | spin_lock(&block_group->lock); | |
cf7c1ef6 | 3489 | if (block_group->cached != BTRFS_CACHE_FINISHED || |
0b246afa | 3490 | !btrfs_test_opt(fs_info, SPACE_CACHE)) { |
cf7c1ef6 LB |
3491 | /* |
3492 | * don't bother trying to write stuff out _if_ | |
3493 | * a) we're not cached, | |
1a79c1f2 LB |
3494 | * b) we're with nospace_cache mount option, |
3495 | * c) we're with v2 space_cache (FREE_SPACE_TREE). | |
cf7c1ef6 | 3496 | */ |
2b20982e | 3497 | dcs = BTRFS_DC_WRITTEN; |
0af3d00b JB |
3498 | spin_unlock(&block_group->lock); |
3499 | goto out_put; | |
3500 | } | |
3501 | spin_unlock(&block_group->lock); | |
3502 | ||
2968b1f4 JB |
3503 | /* |
3504 | * We hit an ENOSPC when setting up the cache in this transaction, just | |
3505 | * skip doing the setup, we've already cleared the cache so we're safe. | |
3506 | */ | |
3507 | if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) { | |
3508 | ret = -ENOSPC; | |
3509 | goto out_put; | |
3510 | } | |
3511 | ||
6fc823b1 JB |
3512 | /* |
3513 | * Try to preallocate enough space based on how big the block group is. | |
3514 | * Keep in mind this has to include any pinned space which could end up | |
3515 | * taking up quite a bit since it's not folded into the other space | |
3516 | * cache. | |
3517 | */ | |
ee22184b | 3518 | num_pages = div_u64(block_group->key.offset, SZ_256M); |
0af3d00b JB |
3519 | if (!num_pages) |
3520 | num_pages = 1; | |
3521 | ||
0af3d00b | 3522 | num_pages *= 16; |
09cbfeaf | 3523 | num_pages *= PAGE_SIZE; |
0af3d00b | 3524 | |
364ecf36 | 3525 | ret = btrfs_check_data_free_space(inode, &data_reserved, 0, num_pages); |
0af3d00b JB |
3526 | if (ret) |
3527 | goto out_put; | |
3528 | ||
3529 | ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages, | |
3530 | num_pages, num_pages, | |
3531 | &alloc_hint); | |
2968b1f4 JB |
3532 | /* |
3533 | * Our cache requires contiguous chunks so that we don't modify a bunch | |
3534 | * of metadata or split extents when writing the cache out, which means | |
3535 | * we can enospc if we are heavily fragmented in addition to just normal | |
3536 | * out of space conditions. So if we hit this just skip setting up any | |
3537 | * other block groups for this transaction, maybe we'll unpin enough | |
3538 | * space the next time around. | |
3539 | */ | |
2b20982e JB |
3540 | if (!ret) |
3541 | dcs = BTRFS_DC_SETUP; | |
2968b1f4 JB |
3542 | else if (ret == -ENOSPC) |
3543 | set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags); | |
c09544e0 | 3544 | |
0af3d00b JB |
3545 | out_put: |
3546 | iput(inode); | |
3547 | out_free: | |
b3b4aa74 | 3548 | btrfs_release_path(path); |
0af3d00b JB |
3549 | out: |
3550 | spin_lock(&block_group->lock); | |
e65cbb94 | 3551 | if (!ret && dcs == BTRFS_DC_SETUP) |
5b0e95bf | 3552 | block_group->cache_generation = trans->transid; |
2b20982e | 3553 | block_group->disk_cache_state = dcs; |
0af3d00b JB |
3554 | spin_unlock(&block_group->lock); |
3555 | ||
364ecf36 | 3556 | extent_changeset_free(data_reserved); |
0af3d00b JB |
3557 | return ret; |
3558 | } | |
3559 | ||
dcdf7f6d | 3560 | int btrfs_setup_space_cache(struct btrfs_trans_handle *trans, |
2ff7e61e | 3561 | struct btrfs_fs_info *fs_info) |
dcdf7f6d JB |
3562 | { |
3563 | struct btrfs_block_group_cache *cache, *tmp; | |
3564 | struct btrfs_transaction *cur_trans = trans->transaction; | |
3565 | struct btrfs_path *path; | |
3566 | ||
3567 | if (list_empty(&cur_trans->dirty_bgs) || | |
0b246afa | 3568 | !btrfs_test_opt(fs_info, SPACE_CACHE)) |
dcdf7f6d JB |
3569 | return 0; |
3570 | ||
3571 | path = btrfs_alloc_path(); | |
3572 | if (!path) | |
3573 | return -ENOMEM; | |
3574 | ||
3575 | /* Could add new block groups, use _safe just in case */ | |
3576 | list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs, | |
3577 | dirty_list) { | |
3578 | if (cache->disk_cache_state == BTRFS_DC_CLEAR) | |
3579 | cache_save_setup(cache, trans, path); | |
3580 | } | |
3581 | ||
3582 | btrfs_free_path(path); | |
3583 | return 0; | |
3584 | } | |
3585 | ||
1bbc621e CM |
3586 | /* |
3587 | * transaction commit does final block group cache writeback during a | |
3588 | * critical section where nothing is allowed to change the FS. This is | |
3589 | * required in order for the cache to actually match the block group, | |
3590 | * but can introduce a lot of latency into the commit. | |
3591 | * | |
3592 | * So, btrfs_start_dirty_block_groups is here to kick off block group | |
3593 | * cache IO. There's a chance we'll have to redo some of it if the | |
3594 | * block group changes again during the commit, but it greatly reduces | |
3595 | * the commit latency by getting rid of the easy block groups while | |
3596 | * we're still allowing others to join the commit. | |
3597 | */ | |
3598 | int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans, | |
2ff7e61e | 3599 | struct btrfs_fs_info *fs_info) |
9078a3e1 | 3600 | { |
4a8c9a62 | 3601 | struct btrfs_block_group_cache *cache; |
ce93ec54 JB |
3602 | struct btrfs_transaction *cur_trans = trans->transaction; |
3603 | int ret = 0; | |
c9dc4c65 | 3604 | int should_put; |
1bbc621e CM |
3605 | struct btrfs_path *path = NULL; |
3606 | LIST_HEAD(dirty); | |
3607 | struct list_head *io = &cur_trans->io_bgs; | |
c9dc4c65 | 3608 | int num_started = 0; |
1bbc621e CM |
3609 | int loops = 0; |
3610 | ||
3611 | spin_lock(&cur_trans->dirty_bgs_lock); | |
b58d1a9e FM |
3612 | if (list_empty(&cur_trans->dirty_bgs)) { |
3613 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3614 | return 0; | |
1bbc621e | 3615 | } |
b58d1a9e | 3616 | list_splice_init(&cur_trans->dirty_bgs, &dirty); |
1bbc621e | 3617 | spin_unlock(&cur_trans->dirty_bgs_lock); |
ce93ec54 | 3618 | |
1bbc621e | 3619 | again: |
1bbc621e CM |
3620 | /* |
3621 | * make sure all the block groups on our dirty list actually | |
3622 | * exist | |
3623 | */ | |
2ff7e61e | 3624 | btrfs_create_pending_block_groups(trans, fs_info); |
1bbc621e CM |
3625 | |
3626 | if (!path) { | |
3627 | path = btrfs_alloc_path(); | |
3628 | if (!path) | |
3629 | return -ENOMEM; | |
3630 | } | |
3631 | ||
b58d1a9e FM |
3632 | /* |
3633 | * cache_write_mutex is here only to save us from balance or automatic | |
3634 | * removal of empty block groups deleting this block group while we are | |
3635 | * writing out the cache | |
3636 | */ | |
3637 | mutex_lock(&trans->transaction->cache_write_mutex); | |
1bbc621e CM |
3638 | while (!list_empty(&dirty)) { |
3639 | cache = list_first_entry(&dirty, | |
3640 | struct btrfs_block_group_cache, | |
3641 | dirty_list); | |
1bbc621e CM |
3642 | /* |
3643 | * this can happen if something re-dirties a block | |
3644 | * group that is already under IO. Just wait for it to | |
3645 | * finish and then do it all again | |
3646 | */ | |
3647 | if (!list_empty(&cache->io_list)) { | |
3648 | list_del_init(&cache->io_list); | |
afdb5718 | 3649 | btrfs_wait_cache_io(trans, cache, path); |
1bbc621e CM |
3650 | btrfs_put_block_group(cache); |
3651 | } | |
3652 | ||
3653 | ||
3654 | /* | |
3655 | * btrfs_wait_cache_io uses the cache->dirty_list to decide | |
3656 | * if it should update the cache_state. Don't delete | |
3657 | * until after we wait. | |
3658 | * | |
3659 | * Since we're not running in the commit critical section | |
3660 | * we need the dirty_bgs_lock to protect from update_block_group | |
3661 | */ | |
3662 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3663 | list_del_init(&cache->dirty_list); | |
3664 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3665 | ||
3666 | should_put = 1; | |
3667 | ||
3668 | cache_save_setup(cache, trans, path); | |
3669 | ||
3670 | if (cache->disk_cache_state == BTRFS_DC_SETUP) { | |
3671 | cache->io_ctl.inode = NULL; | |
0b246afa | 3672 | ret = btrfs_write_out_cache(fs_info, trans, |
5b4aacef | 3673 | cache, path); |
1bbc621e CM |
3674 | if (ret == 0 && cache->io_ctl.inode) { |
3675 | num_started++; | |
3676 | should_put = 0; | |
3677 | ||
3678 | /* | |
3679 | * the cache_write_mutex is protecting | |
3680 | * the io_list | |
3681 | */ | |
3682 | list_add_tail(&cache->io_list, io); | |
3683 | } else { | |
3684 | /* | |
3685 | * if we failed to write the cache, the | |
3686 | * generation will be bad and life goes on | |
3687 | */ | |
3688 | ret = 0; | |
3689 | } | |
3690 | } | |
ff1f8250 | 3691 | if (!ret) { |
2ff7e61e JM |
3692 | ret = write_one_cache_group(trans, fs_info, |
3693 | path, cache); | |
ff1f8250 FM |
3694 | /* |
3695 | * Our block group might still be attached to the list | |
3696 | * of new block groups in the transaction handle of some | |
3697 | * other task (struct btrfs_trans_handle->new_bgs). This | |
3698 | * means its block group item isn't yet in the extent | |
3699 | * tree. If this happens ignore the error, as we will | |
3700 | * try again later in the critical section of the | |
3701 | * transaction commit. | |
3702 | */ | |
3703 | if (ret == -ENOENT) { | |
3704 | ret = 0; | |
3705 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3706 | if (list_empty(&cache->dirty_list)) { | |
3707 | list_add_tail(&cache->dirty_list, | |
3708 | &cur_trans->dirty_bgs); | |
3709 | btrfs_get_block_group(cache); | |
3710 | } | |
3711 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3712 | } else if (ret) { | |
66642832 | 3713 | btrfs_abort_transaction(trans, ret); |
ff1f8250 FM |
3714 | } |
3715 | } | |
1bbc621e CM |
3716 | |
3717 | /* if its not on the io list, we need to put the block group */ | |
3718 | if (should_put) | |
3719 | btrfs_put_block_group(cache); | |
3720 | ||
3721 | if (ret) | |
3722 | break; | |
b58d1a9e FM |
3723 | |
3724 | /* | |
3725 | * Avoid blocking other tasks for too long. It might even save | |
3726 | * us from writing caches for block groups that are going to be | |
3727 | * removed. | |
3728 | */ | |
3729 | mutex_unlock(&trans->transaction->cache_write_mutex); | |
3730 | mutex_lock(&trans->transaction->cache_write_mutex); | |
1bbc621e | 3731 | } |
b58d1a9e | 3732 | mutex_unlock(&trans->transaction->cache_write_mutex); |
1bbc621e CM |
3733 | |
3734 | /* | |
3735 | * go through delayed refs for all the stuff we've just kicked off | |
3736 | * and then loop back (just once) | |
3737 | */ | |
2ff7e61e | 3738 | ret = btrfs_run_delayed_refs(trans, fs_info, 0); |
1bbc621e CM |
3739 | if (!ret && loops == 0) { |
3740 | loops++; | |
3741 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3742 | list_splice_init(&cur_trans->dirty_bgs, &dirty); | |
b58d1a9e FM |
3743 | /* |
3744 | * dirty_bgs_lock protects us from concurrent block group | |
3745 | * deletes too (not just cache_write_mutex). | |
3746 | */ | |
3747 | if (!list_empty(&dirty)) { | |
3748 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3749 | goto again; | |
3750 | } | |
1bbc621e | 3751 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c79a1751 | 3752 | } else if (ret < 0) { |
2ff7e61e | 3753 | btrfs_cleanup_dirty_bgs(cur_trans, fs_info); |
1bbc621e CM |
3754 | } |
3755 | ||
3756 | btrfs_free_path(path); | |
3757 | return ret; | |
3758 | } | |
3759 | ||
3760 | int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans, | |
2ff7e61e | 3761 | struct btrfs_fs_info *fs_info) |
1bbc621e CM |
3762 | { |
3763 | struct btrfs_block_group_cache *cache; | |
3764 | struct btrfs_transaction *cur_trans = trans->transaction; | |
3765 | int ret = 0; | |
3766 | int should_put; | |
3767 | struct btrfs_path *path; | |
3768 | struct list_head *io = &cur_trans->io_bgs; | |
3769 | int num_started = 0; | |
9078a3e1 CM |
3770 | |
3771 | path = btrfs_alloc_path(); | |
3772 | if (!path) | |
3773 | return -ENOMEM; | |
3774 | ||
ce93ec54 | 3775 | /* |
e44081ef FM |
3776 | * Even though we are in the critical section of the transaction commit, |
3777 | * we can still have concurrent tasks adding elements to this | |
3778 | * transaction's list of dirty block groups. These tasks correspond to | |
3779 | * endio free space workers started when writeback finishes for a | |
3780 | * space cache, which run inode.c:btrfs_finish_ordered_io(), and can | |
3781 | * allocate new block groups as a result of COWing nodes of the root | |
3782 | * tree when updating the free space inode. The writeback for the space | |
3783 | * caches is triggered by an earlier call to | |
3784 | * btrfs_start_dirty_block_groups() and iterations of the following | |
3785 | * loop. | |
3786 | * Also we want to do the cache_save_setup first and then run the | |
ce93ec54 JB |
3787 | * delayed refs to make sure we have the best chance at doing this all |
3788 | * in one shot. | |
3789 | */ | |
e44081ef | 3790 | spin_lock(&cur_trans->dirty_bgs_lock); |
ce93ec54 JB |
3791 | while (!list_empty(&cur_trans->dirty_bgs)) { |
3792 | cache = list_first_entry(&cur_trans->dirty_bgs, | |
3793 | struct btrfs_block_group_cache, | |
3794 | dirty_list); | |
c9dc4c65 CM |
3795 | |
3796 | /* | |
3797 | * this can happen if cache_save_setup re-dirties a block | |
3798 | * group that is already under IO. Just wait for it to | |
3799 | * finish and then do it all again | |
3800 | */ | |
3801 | if (!list_empty(&cache->io_list)) { | |
e44081ef | 3802 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 | 3803 | list_del_init(&cache->io_list); |
afdb5718 | 3804 | btrfs_wait_cache_io(trans, cache, path); |
c9dc4c65 | 3805 | btrfs_put_block_group(cache); |
e44081ef | 3806 | spin_lock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 CM |
3807 | } |
3808 | ||
1bbc621e CM |
3809 | /* |
3810 | * don't remove from the dirty list until after we've waited | |
3811 | * on any pending IO | |
3812 | */ | |
ce93ec54 | 3813 | list_del_init(&cache->dirty_list); |
e44081ef | 3814 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 CM |
3815 | should_put = 1; |
3816 | ||
1bbc621e | 3817 | cache_save_setup(cache, trans, path); |
c9dc4c65 | 3818 | |
ce93ec54 | 3819 | if (!ret) |
2ff7e61e JM |
3820 | ret = btrfs_run_delayed_refs(trans, fs_info, |
3821 | (unsigned long) -1); | |
c9dc4c65 CM |
3822 | |
3823 | if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) { | |
3824 | cache->io_ctl.inode = NULL; | |
0b246afa | 3825 | ret = btrfs_write_out_cache(fs_info, trans, |
5b4aacef | 3826 | cache, path); |
c9dc4c65 CM |
3827 | if (ret == 0 && cache->io_ctl.inode) { |
3828 | num_started++; | |
3829 | should_put = 0; | |
1bbc621e | 3830 | list_add_tail(&cache->io_list, io); |
c9dc4c65 CM |
3831 | } else { |
3832 | /* | |
3833 | * if we failed to write the cache, the | |
3834 | * generation will be bad and life goes on | |
3835 | */ | |
3836 | ret = 0; | |
3837 | } | |
3838 | } | |
ff1f8250 | 3839 | if (!ret) { |
2ff7e61e JM |
3840 | ret = write_one_cache_group(trans, fs_info, |
3841 | path, cache); | |
2bc0bb5f FM |
3842 | /* |
3843 | * One of the free space endio workers might have | |
3844 | * created a new block group while updating a free space | |
3845 | * cache's inode (at inode.c:btrfs_finish_ordered_io()) | |
3846 | * and hasn't released its transaction handle yet, in | |
3847 | * which case the new block group is still attached to | |
3848 | * its transaction handle and its creation has not | |
3849 | * finished yet (no block group item in the extent tree | |
3850 | * yet, etc). If this is the case, wait for all free | |
3851 | * space endio workers to finish and retry. This is a | |
3852 | * a very rare case so no need for a more efficient and | |
3853 | * complex approach. | |
3854 | */ | |
3855 | if (ret == -ENOENT) { | |
3856 | wait_event(cur_trans->writer_wait, | |
3857 | atomic_read(&cur_trans->num_writers) == 1); | |
2ff7e61e JM |
3858 | ret = write_one_cache_group(trans, fs_info, |
3859 | path, cache); | |
2bc0bb5f | 3860 | } |
ff1f8250 | 3861 | if (ret) |
66642832 | 3862 | btrfs_abort_transaction(trans, ret); |
ff1f8250 | 3863 | } |
c9dc4c65 CM |
3864 | |
3865 | /* if its not on the io list, we need to put the block group */ | |
3866 | if (should_put) | |
3867 | btrfs_put_block_group(cache); | |
e44081ef | 3868 | spin_lock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 | 3869 | } |
e44081ef | 3870 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 | 3871 | |
1bbc621e CM |
3872 | while (!list_empty(io)) { |
3873 | cache = list_first_entry(io, struct btrfs_block_group_cache, | |
c9dc4c65 CM |
3874 | io_list); |
3875 | list_del_init(&cache->io_list); | |
afdb5718 | 3876 | btrfs_wait_cache_io(trans, cache, path); |
0cb59c99 JB |
3877 | btrfs_put_block_group(cache); |
3878 | } | |
3879 | ||
9078a3e1 | 3880 | btrfs_free_path(path); |
ce93ec54 | 3881 | return ret; |
9078a3e1 CM |
3882 | } |
3883 | ||
2ff7e61e | 3884 | int btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr) |
d2fb3437 YZ |
3885 | { |
3886 | struct btrfs_block_group_cache *block_group; | |
3887 | int readonly = 0; | |
3888 | ||
0b246afa | 3889 | block_group = btrfs_lookup_block_group(fs_info, bytenr); |
d2fb3437 YZ |
3890 | if (!block_group || block_group->ro) |
3891 | readonly = 1; | |
3892 | if (block_group) | |
fa9c0d79 | 3893 | btrfs_put_block_group(block_group); |
d2fb3437 YZ |
3894 | return readonly; |
3895 | } | |
3896 | ||
f78c436c FM |
3897 | bool btrfs_inc_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr) |
3898 | { | |
3899 | struct btrfs_block_group_cache *bg; | |
3900 | bool ret = true; | |
3901 | ||
3902 | bg = btrfs_lookup_block_group(fs_info, bytenr); | |
3903 | if (!bg) | |
3904 | return false; | |
3905 | ||
3906 | spin_lock(&bg->lock); | |
3907 | if (bg->ro) | |
3908 | ret = false; | |
3909 | else | |
3910 | atomic_inc(&bg->nocow_writers); | |
3911 | spin_unlock(&bg->lock); | |
3912 | ||
3913 | /* no put on block group, done by btrfs_dec_nocow_writers */ | |
3914 | if (!ret) | |
3915 | btrfs_put_block_group(bg); | |
3916 | ||
3917 | return ret; | |
3918 | ||
3919 | } | |
3920 | ||
3921 | void btrfs_dec_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr) | |
3922 | { | |
3923 | struct btrfs_block_group_cache *bg; | |
3924 | ||
3925 | bg = btrfs_lookup_block_group(fs_info, bytenr); | |
3926 | ASSERT(bg); | |
3927 | if (atomic_dec_and_test(&bg->nocow_writers)) | |
3928 | wake_up_atomic_t(&bg->nocow_writers); | |
3929 | /* | |
3930 | * Once for our lookup and once for the lookup done by a previous call | |
3931 | * to btrfs_inc_nocow_writers() | |
3932 | */ | |
3933 | btrfs_put_block_group(bg); | |
3934 | btrfs_put_block_group(bg); | |
3935 | } | |
3936 | ||
3937 | static int btrfs_wait_nocow_writers_atomic_t(atomic_t *a) | |
3938 | { | |
3939 | schedule(); | |
3940 | return 0; | |
3941 | } | |
3942 | ||
3943 | void btrfs_wait_nocow_writers(struct btrfs_block_group_cache *bg) | |
3944 | { | |
3945 | wait_on_atomic_t(&bg->nocow_writers, | |
3946 | btrfs_wait_nocow_writers_atomic_t, | |
3947 | TASK_UNINTERRUPTIBLE); | |
3948 | } | |
3949 | ||
6ab0a202 JM |
3950 | static const char *alloc_name(u64 flags) |
3951 | { | |
3952 | switch (flags) { | |
3953 | case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA: | |
3954 | return "mixed"; | |
3955 | case BTRFS_BLOCK_GROUP_METADATA: | |
3956 | return "metadata"; | |
3957 | case BTRFS_BLOCK_GROUP_DATA: | |
3958 | return "data"; | |
3959 | case BTRFS_BLOCK_GROUP_SYSTEM: | |
3960 | return "system"; | |
3961 | default: | |
3962 | WARN_ON(1); | |
3963 | return "invalid-combination"; | |
3964 | }; | |
3965 | } | |
3966 | ||
2be12ef7 NB |
3967 | static int create_space_info(struct btrfs_fs_info *info, u64 flags, |
3968 | struct btrfs_space_info **new) | |
3969 | { | |
3970 | ||
3971 | struct btrfs_space_info *space_info; | |
3972 | int i; | |
3973 | int ret; | |
3974 | ||
3975 | space_info = kzalloc(sizeof(*space_info), GFP_NOFS); | |
3976 | if (!space_info) | |
3977 | return -ENOMEM; | |
3978 | ||
3979 | ret = percpu_counter_init(&space_info->total_bytes_pinned, 0, | |
3980 | GFP_KERNEL); | |
3981 | if (ret) { | |
3982 | kfree(space_info); | |
3983 | return ret; | |
3984 | } | |
3985 | ||
3986 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) | |
3987 | INIT_LIST_HEAD(&space_info->block_groups[i]); | |
3988 | init_rwsem(&space_info->groups_sem); | |
3989 | spin_lock_init(&space_info->lock); | |
3990 | space_info->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK; | |
3991 | space_info->force_alloc = CHUNK_ALLOC_NO_FORCE; | |
3992 | init_waitqueue_head(&space_info->wait); | |
3993 | INIT_LIST_HEAD(&space_info->ro_bgs); | |
3994 | INIT_LIST_HEAD(&space_info->tickets); | |
3995 | INIT_LIST_HEAD(&space_info->priority_tickets); | |
3996 | ||
3997 | ret = kobject_init_and_add(&space_info->kobj, &space_info_ktype, | |
3998 | info->space_info_kobj, "%s", | |
3999 | alloc_name(space_info->flags)); | |
4000 | if (ret) { | |
4001 | percpu_counter_destroy(&space_info->total_bytes_pinned); | |
4002 | kfree(space_info); | |
4003 | return ret; | |
4004 | } | |
4005 | ||
4006 | *new = space_info; | |
4007 | list_add_rcu(&space_info->list, &info->space_info); | |
4008 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
4009 | info->data_sinfo = space_info; | |
4010 | ||
4011 | return ret; | |
4012 | } | |
4013 | ||
d2006e6d | 4014 | static void update_space_info(struct btrfs_fs_info *info, u64 flags, |
593060d7 | 4015 | u64 total_bytes, u64 bytes_used, |
e40edf2d | 4016 | u64 bytes_readonly, |
593060d7 CM |
4017 | struct btrfs_space_info **space_info) |
4018 | { | |
4019 | struct btrfs_space_info *found; | |
b742bb82 YZ |
4020 | int factor; |
4021 | ||
4022 | if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
4023 | BTRFS_BLOCK_GROUP_RAID10)) | |
4024 | factor = 2; | |
4025 | else | |
4026 | factor = 1; | |
593060d7 CM |
4027 | |
4028 | found = __find_space_info(info, flags); | |
d2006e6d NB |
4029 | ASSERT(found); |
4030 | spin_lock(&found->lock); | |
4031 | found->total_bytes += total_bytes; | |
4032 | found->disk_total += total_bytes * factor; | |
4033 | found->bytes_used += bytes_used; | |
4034 | found->disk_used += bytes_used * factor; | |
4035 | found->bytes_readonly += bytes_readonly; | |
4036 | if (total_bytes > 0) | |
4037 | found->full = 0; | |
4038 | space_info_add_new_bytes(info, found, total_bytes - | |
4039 | bytes_used - bytes_readonly); | |
4040 | spin_unlock(&found->lock); | |
4041 | *space_info = found; | |
593060d7 CM |
4042 | } |
4043 | ||
8790d502 CM |
4044 | static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
4045 | { | |
899c81ea ID |
4046 | u64 extra_flags = chunk_to_extended(flags) & |
4047 | BTRFS_EXTENDED_PROFILE_MASK; | |
a46d11a8 | 4048 | |
de98ced9 | 4049 | write_seqlock(&fs_info->profiles_lock); |
a46d11a8 ID |
4050 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
4051 | fs_info->avail_data_alloc_bits |= extra_flags; | |
4052 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
4053 | fs_info->avail_metadata_alloc_bits |= extra_flags; | |
4054 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
4055 | fs_info->avail_system_alloc_bits |= extra_flags; | |
de98ced9 | 4056 | write_sequnlock(&fs_info->profiles_lock); |
8790d502 | 4057 | } |
593060d7 | 4058 | |
fc67c450 ID |
4059 | /* |
4060 | * returns target flags in extended format or 0 if restripe for this | |
4061 | * chunk_type is not in progress | |
c6664b42 ID |
4062 | * |
4063 | * should be called with either volume_mutex or balance_lock held | |
fc67c450 ID |
4064 | */ |
4065 | static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags) | |
4066 | { | |
4067 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
4068 | u64 target = 0; | |
4069 | ||
fc67c450 ID |
4070 | if (!bctl) |
4071 | return 0; | |
4072 | ||
4073 | if (flags & BTRFS_BLOCK_GROUP_DATA && | |
4074 | bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
4075 | target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target; | |
4076 | } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM && | |
4077 | bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
4078 | target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target; | |
4079 | } else if (flags & BTRFS_BLOCK_GROUP_METADATA && | |
4080 | bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
4081 | target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target; | |
4082 | } | |
4083 | ||
4084 | return target; | |
4085 | } | |
4086 | ||
a46d11a8 ID |
4087 | /* |
4088 | * @flags: available profiles in extended format (see ctree.h) | |
4089 | * | |
e4d8ec0f ID |
4090 | * Returns reduced profile in chunk format. If profile changing is in |
4091 | * progress (either running or paused) picks the target profile (if it's | |
4092 | * already available), otherwise falls back to plain reducing. | |
a46d11a8 | 4093 | */ |
2ff7e61e | 4094 | static u64 btrfs_reduce_alloc_profile(struct btrfs_fs_info *fs_info, u64 flags) |
ec44a35c | 4095 | { |
0b246afa | 4096 | u64 num_devices = fs_info->fs_devices->rw_devices; |
fc67c450 | 4097 | u64 target; |
9c170b26 ZL |
4098 | u64 raid_type; |
4099 | u64 allowed = 0; | |
a061fc8d | 4100 | |
fc67c450 ID |
4101 | /* |
4102 | * see if restripe for this chunk_type is in progress, if so | |
4103 | * try to reduce to the target profile | |
4104 | */ | |
0b246afa JM |
4105 | spin_lock(&fs_info->balance_lock); |
4106 | target = get_restripe_target(fs_info, flags); | |
fc67c450 ID |
4107 | if (target) { |
4108 | /* pick target profile only if it's already available */ | |
4109 | if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) { | |
0b246afa | 4110 | spin_unlock(&fs_info->balance_lock); |
fc67c450 | 4111 | return extended_to_chunk(target); |
e4d8ec0f ID |
4112 | } |
4113 | } | |
0b246afa | 4114 | spin_unlock(&fs_info->balance_lock); |
e4d8ec0f | 4115 | |
53b381b3 | 4116 | /* First, mask out the RAID levels which aren't possible */ |
9c170b26 ZL |
4117 | for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) { |
4118 | if (num_devices >= btrfs_raid_array[raid_type].devs_min) | |
4119 | allowed |= btrfs_raid_group[raid_type]; | |
4120 | } | |
4121 | allowed &= flags; | |
4122 | ||
4123 | if (allowed & BTRFS_BLOCK_GROUP_RAID6) | |
4124 | allowed = BTRFS_BLOCK_GROUP_RAID6; | |
4125 | else if (allowed & BTRFS_BLOCK_GROUP_RAID5) | |
4126 | allowed = BTRFS_BLOCK_GROUP_RAID5; | |
4127 | else if (allowed & BTRFS_BLOCK_GROUP_RAID10) | |
4128 | allowed = BTRFS_BLOCK_GROUP_RAID10; | |
4129 | else if (allowed & BTRFS_BLOCK_GROUP_RAID1) | |
4130 | allowed = BTRFS_BLOCK_GROUP_RAID1; | |
4131 | else if (allowed & BTRFS_BLOCK_GROUP_RAID0) | |
4132 | allowed = BTRFS_BLOCK_GROUP_RAID0; | |
4133 | ||
4134 | flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK; | |
4135 | ||
4136 | return extended_to_chunk(flags | allowed); | |
ec44a35c CM |
4137 | } |
4138 | ||
2ff7e61e | 4139 | static u64 get_alloc_profile(struct btrfs_fs_info *fs_info, u64 orig_flags) |
6a63209f | 4140 | { |
de98ced9 | 4141 | unsigned seq; |
f8213bdc | 4142 | u64 flags; |
de98ced9 MX |
4143 | |
4144 | do { | |
f8213bdc | 4145 | flags = orig_flags; |
0b246afa | 4146 | seq = read_seqbegin(&fs_info->profiles_lock); |
de98ced9 MX |
4147 | |
4148 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
0b246afa | 4149 | flags |= fs_info->avail_data_alloc_bits; |
de98ced9 | 4150 | else if (flags & BTRFS_BLOCK_GROUP_SYSTEM) |
0b246afa | 4151 | flags |= fs_info->avail_system_alloc_bits; |
de98ced9 | 4152 | else if (flags & BTRFS_BLOCK_GROUP_METADATA) |
0b246afa JM |
4153 | flags |= fs_info->avail_metadata_alloc_bits; |
4154 | } while (read_seqretry(&fs_info->profiles_lock, seq)); | |
6fef8df1 | 4155 | |
2ff7e61e | 4156 | return btrfs_reduce_alloc_profile(fs_info, flags); |
6a63209f JB |
4157 | } |
4158 | ||
1b86826d | 4159 | static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data) |
9ed74f2d | 4160 | { |
0b246afa | 4161 | struct btrfs_fs_info *fs_info = root->fs_info; |
b742bb82 | 4162 | u64 flags; |
53b381b3 | 4163 | u64 ret; |
9ed74f2d | 4164 | |
b742bb82 YZ |
4165 | if (data) |
4166 | flags = BTRFS_BLOCK_GROUP_DATA; | |
0b246afa | 4167 | else if (root == fs_info->chunk_root) |
b742bb82 | 4168 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
9ed74f2d | 4169 | else |
b742bb82 | 4170 | flags = BTRFS_BLOCK_GROUP_METADATA; |
9ed74f2d | 4171 | |
2ff7e61e | 4172 | ret = get_alloc_profile(fs_info, flags); |
53b381b3 | 4173 | return ret; |
6a63209f | 4174 | } |
9ed74f2d | 4175 | |
1b86826d JM |
4176 | u64 btrfs_data_alloc_profile(struct btrfs_fs_info *fs_info) |
4177 | { | |
4178 | return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_DATA); | |
4179 | } | |
4180 | ||
4181 | u64 btrfs_metadata_alloc_profile(struct btrfs_fs_info *fs_info) | |
4182 | { | |
4183 | return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_METADATA); | |
4184 | } | |
4185 | ||
4186 | u64 btrfs_system_alloc_profile(struct btrfs_fs_info *fs_info) | |
4187 | { | |
4188 | return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); | |
4189 | } | |
4190 | ||
4136135b LB |
4191 | static u64 btrfs_space_info_used(struct btrfs_space_info *s_info, |
4192 | bool may_use_included) | |
4193 | { | |
4194 | ASSERT(s_info); | |
4195 | return s_info->bytes_used + s_info->bytes_reserved + | |
4196 | s_info->bytes_pinned + s_info->bytes_readonly + | |
4197 | (may_use_included ? s_info->bytes_may_use : 0); | |
4198 | } | |
4199 | ||
04f4f916 | 4200 | int btrfs_alloc_data_chunk_ondemand(struct btrfs_inode *inode, u64 bytes) |
6a63209f | 4201 | { |
6a63209f | 4202 | struct btrfs_space_info *data_sinfo; |
04f4f916 | 4203 | struct btrfs_root *root = inode->root; |
b4d7c3c9 | 4204 | struct btrfs_fs_info *fs_info = root->fs_info; |
ab6e2410 | 4205 | u64 used; |
94b947b2 | 4206 | int ret = 0; |
c99f1b0c ZL |
4207 | int need_commit = 2; |
4208 | int have_pinned_space; | |
6a63209f | 4209 | |
6a63209f | 4210 | /* make sure bytes are sectorsize aligned */ |
0b246afa | 4211 | bytes = ALIGN(bytes, fs_info->sectorsize); |
6a63209f | 4212 | |
9dced186 | 4213 | if (btrfs_is_free_space_inode(inode)) { |
c99f1b0c | 4214 | need_commit = 0; |
9dced186 | 4215 | ASSERT(current->journal_info); |
0af3d00b JB |
4216 | } |
4217 | ||
b4d7c3c9 | 4218 | data_sinfo = fs_info->data_sinfo; |
33b4d47f CM |
4219 | if (!data_sinfo) |
4220 | goto alloc; | |
9ed74f2d | 4221 | |
6a63209f JB |
4222 | again: |
4223 | /* make sure we have enough space to handle the data first */ | |
4224 | spin_lock(&data_sinfo->lock); | |
4136135b | 4225 | used = btrfs_space_info_used(data_sinfo, true); |
ab6e2410 JB |
4226 | |
4227 | if (used + bytes > data_sinfo->total_bytes) { | |
4e06bdd6 | 4228 | struct btrfs_trans_handle *trans; |
9ed74f2d | 4229 | |
6a63209f JB |
4230 | /* |
4231 | * if we don't have enough free bytes in this space then we need | |
4232 | * to alloc a new chunk. | |
4233 | */ | |
b9fd47cd | 4234 | if (!data_sinfo->full) { |
6a63209f | 4235 | u64 alloc_target; |
9ed74f2d | 4236 | |
0e4f8f88 | 4237 | data_sinfo->force_alloc = CHUNK_ALLOC_FORCE; |
6a63209f | 4238 | spin_unlock(&data_sinfo->lock); |
33b4d47f | 4239 | alloc: |
1b86826d | 4240 | alloc_target = btrfs_data_alloc_profile(fs_info); |
9dced186 MX |
4241 | /* |
4242 | * It is ugly that we don't call nolock join | |
4243 | * transaction for the free space inode case here. | |
4244 | * But it is safe because we only do the data space | |
4245 | * reservation for the free space cache in the | |
4246 | * transaction context, the common join transaction | |
4247 | * just increase the counter of the current transaction | |
4248 | * handler, doesn't try to acquire the trans_lock of | |
4249 | * the fs. | |
4250 | */ | |
7a7eaa40 | 4251 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
4252 | if (IS_ERR(trans)) |
4253 | return PTR_ERR(trans); | |
9ed74f2d | 4254 | |
2ff7e61e | 4255 | ret = do_chunk_alloc(trans, fs_info, alloc_target, |
0e4f8f88 | 4256 | CHUNK_ALLOC_NO_FORCE); |
3a45bb20 | 4257 | btrfs_end_transaction(trans); |
d52a5b5f MX |
4258 | if (ret < 0) { |
4259 | if (ret != -ENOSPC) | |
4260 | return ret; | |
c99f1b0c ZL |
4261 | else { |
4262 | have_pinned_space = 1; | |
d52a5b5f | 4263 | goto commit_trans; |
c99f1b0c | 4264 | } |
d52a5b5f | 4265 | } |
9ed74f2d | 4266 | |
b4d7c3c9 LZ |
4267 | if (!data_sinfo) |
4268 | data_sinfo = fs_info->data_sinfo; | |
4269 | ||
6a63209f JB |
4270 | goto again; |
4271 | } | |
f2bb8f5c JB |
4272 | |
4273 | /* | |
b150a4f1 | 4274 | * If we don't have enough pinned space to deal with this |
94b947b2 ZL |
4275 | * allocation, and no removed chunk in current transaction, |
4276 | * don't bother committing the transaction. | |
f2bb8f5c | 4277 | */ |
c99f1b0c ZL |
4278 | have_pinned_space = percpu_counter_compare( |
4279 | &data_sinfo->total_bytes_pinned, | |
4280 | used + bytes - data_sinfo->total_bytes); | |
6a63209f | 4281 | spin_unlock(&data_sinfo->lock); |
6a63209f | 4282 | |
4e06bdd6 | 4283 | /* commit the current transaction and try again */ |
d52a5b5f | 4284 | commit_trans: |
c99f1b0c | 4285 | if (need_commit && |
0b246afa | 4286 | !atomic_read(&fs_info->open_ioctl_trans)) { |
c99f1b0c | 4287 | need_commit--; |
b150a4f1 | 4288 | |
e1746e83 ZL |
4289 | if (need_commit > 0) { |
4290 | btrfs_start_delalloc_roots(fs_info, 0, -1); | |
6374e57a | 4291 | btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, |
0b246afa | 4292 | (u64)-1); |
e1746e83 | 4293 | } |
9a4e7276 | 4294 | |
7a7eaa40 | 4295 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
4296 | if (IS_ERR(trans)) |
4297 | return PTR_ERR(trans); | |
c99f1b0c | 4298 | if (have_pinned_space >= 0 || |
3204d33c JB |
4299 | test_bit(BTRFS_TRANS_HAVE_FREE_BGS, |
4300 | &trans->transaction->flags) || | |
c99f1b0c | 4301 | need_commit > 0) { |
3a45bb20 | 4302 | ret = btrfs_commit_transaction(trans); |
94b947b2 ZL |
4303 | if (ret) |
4304 | return ret; | |
d7c15171 | 4305 | /* |
c2d6cb16 FM |
4306 | * The cleaner kthread might still be doing iput |
4307 | * operations. Wait for it to finish so that | |
4308 | * more space is released. | |
d7c15171 | 4309 | */ |
0b246afa JM |
4310 | mutex_lock(&fs_info->cleaner_delayed_iput_mutex); |
4311 | mutex_unlock(&fs_info->cleaner_delayed_iput_mutex); | |
94b947b2 ZL |
4312 | goto again; |
4313 | } else { | |
3a45bb20 | 4314 | btrfs_end_transaction(trans); |
94b947b2 | 4315 | } |
4e06bdd6 | 4316 | } |
9ed74f2d | 4317 | |
0b246afa | 4318 | trace_btrfs_space_reservation(fs_info, |
cab45e22 JM |
4319 | "space_info:enospc", |
4320 | data_sinfo->flags, bytes, 1); | |
6a63209f JB |
4321 | return -ENOSPC; |
4322 | } | |
4323 | data_sinfo->bytes_may_use += bytes; | |
0b246afa | 4324 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 4325 | data_sinfo->flags, bytes, 1); |
6a63209f | 4326 | spin_unlock(&data_sinfo->lock); |
6a63209f | 4327 | |
237c0e9f | 4328 | return ret; |
9ed74f2d | 4329 | } |
6a63209f | 4330 | |
364ecf36 QW |
4331 | int btrfs_check_data_free_space(struct inode *inode, |
4332 | struct extent_changeset **reserved, u64 start, u64 len) | |
4ceff079 | 4333 | { |
0b246afa | 4334 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4ceff079 QW |
4335 | int ret; |
4336 | ||
4337 | /* align the range */ | |
0b246afa JM |
4338 | len = round_up(start + len, fs_info->sectorsize) - |
4339 | round_down(start, fs_info->sectorsize); | |
4340 | start = round_down(start, fs_info->sectorsize); | |
4ceff079 | 4341 | |
04f4f916 | 4342 | ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode), len); |
4ceff079 QW |
4343 | if (ret < 0) |
4344 | return ret; | |
4345 | ||
1e5ec2e7 | 4346 | /* Use new btrfs_qgroup_reserve_data to reserve precious data space. */ |
364ecf36 | 4347 | ret = btrfs_qgroup_reserve_data(inode, reserved, start, len); |
7bc329c1 | 4348 | if (ret < 0) |
1e5ec2e7 | 4349 | btrfs_free_reserved_data_space_noquota(inode, start, len); |
364ecf36 QW |
4350 | else |
4351 | ret = 0; | |
4ceff079 QW |
4352 | return ret; |
4353 | } | |
4354 | ||
4ceff079 QW |
4355 | /* |
4356 | * Called if we need to clear a data reservation for this inode | |
4357 | * Normally in a error case. | |
4358 | * | |
51773bec QW |
4359 | * This one will *NOT* use accurate qgroup reserved space API, just for case |
4360 | * which we can't sleep and is sure it won't affect qgroup reserved space. | |
4361 | * Like clear_bit_hook(). | |
4ceff079 | 4362 | */ |
51773bec QW |
4363 | void btrfs_free_reserved_data_space_noquota(struct inode *inode, u64 start, |
4364 | u64 len) | |
4ceff079 | 4365 | { |
0b246afa | 4366 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4ceff079 QW |
4367 | struct btrfs_space_info *data_sinfo; |
4368 | ||
4369 | /* Make sure the range is aligned to sectorsize */ | |
0b246afa JM |
4370 | len = round_up(start + len, fs_info->sectorsize) - |
4371 | round_down(start, fs_info->sectorsize); | |
4372 | start = round_down(start, fs_info->sectorsize); | |
4ceff079 | 4373 | |
0b246afa | 4374 | data_sinfo = fs_info->data_sinfo; |
4ceff079 QW |
4375 | spin_lock(&data_sinfo->lock); |
4376 | if (WARN_ON(data_sinfo->bytes_may_use < len)) | |
4377 | data_sinfo->bytes_may_use = 0; | |
4378 | else | |
4379 | data_sinfo->bytes_may_use -= len; | |
0b246afa | 4380 | trace_btrfs_space_reservation(fs_info, "space_info", |
4ceff079 QW |
4381 | data_sinfo->flags, len, 0); |
4382 | spin_unlock(&data_sinfo->lock); | |
4383 | } | |
4384 | ||
51773bec QW |
4385 | /* |
4386 | * Called if we need to clear a data reservation for this inode | |
4387 | * Normally in a error case. | |
4388 | * | |
01327610 | 4389 | * This one will handle the per-inode data rsv map for accurate reserved |
51773bec QW |
4390 | * space framework. |
4391 | */ | |
bc42bda2 QW |
4392 | void btrfs_free_reserved_data_space(struct inode *inode, |
4393 | struct extent_changeset *reserved, u64 start, u64 len) | |
51773bec | 4394 | { |
0c476a5d JM |
4395 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4396 | ||
4397 | /* Make sure the range is aligned to sectorsize */ | |
da17066c JM |
4398 | len = round_up(start + len, root->fs_info->sectorsize) - |
4399 | round_down(start, root->fs_info->sectorsize); | |
4400 | start = round_down(start, root->fs_info->sectorsize); | |
0c476a5d | 4401 | |
51773bec | 4402 | btrfs_free_reserved_data_space_noquota(inode, start, len); |
bc42bda2 | 4403 | btrfs_qgroup_free_data(inode, reserved, start, len); |
51773bec QW |
4404 | } |
4405 | ||
97e728d4 | 4406 | static void force_metadata_allocation(struct btrfs_fs_info *info) |
e3ccfa98 | 4407 | { |
97e728d4 JB |
4408 | struct list_head *head = &info->space_info; |
4409 | struct btrfs_space_info *found; | |
e3ccfa98 | 4410 | |
97e728d4 JB |
4411 | rcu_read_lock(); |
4412 | list_for_each_entry_rcu(found, head, list) { | |
4413 | if (found->flags & BTRFS_BLOCK_GROUP_METADATA) | |
0e4f8f88 | 4414 | found->force_alloc = CHUNK_ALLOC_FORCE; |
e3ccfa98 | 4415 | } |
97e728d4 | 4416 | rcu_read_unlock(); |
e3ccfa98 JB |
4417 | } |
4418 | ||
3c76cd84 MX |
4419 | static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global) |
4420 | { | |
4421 | return (global->size << 1); | |
4422 | } | |
4423 | ||
2ff7e61e | 4424 | static int should_alloc_chunk(struct btrfs_fs_info *fs_info, |
698d0082 | 4425 | struct btrfs_space_info *sinfo, int force) |
32c00aff | 4426 | { |
0b246afa | 4427 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; |
424499db | 4428 | u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly; |
0e4f8f88 | 4429 | u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved; |
e5bc2458 | 4430 | u64 thresh; |
e3ccfa98 | 4431 | |
0e4f8f88 CM |
4432 | if (force == CHUNK_ALLOC_FORCE) |
4433 | return 1; | |
4434 | ||
fb25e914 JB |
4435 | /* |
4436 | * We need to take into account the global rsv because for all intents | |
4437 | * and purposes it's used space. Don't worry about locking the | |
4438 | * global_rsv, it doesn't change except when the transaction commits. | |
4439 | */ | |
54338b5c | 4440 | if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA) |
3c76cd84 | 4441 | num_allocated += calc_global_rsv_need_space(global_rsv); |
fb25e914 | 4442 | |
0e4f8f88 CM |
4443 | /* |
4444 | * in limited mode, we want to have some free space up to | |
4445 | * about 1% of the FS size. | |
4446 | */ | |
4447 | if (force == CHUNK_ALLOC_LIMITED) { | |
0b246afa | 4448 | thresh = btrfs_super_total_bytes(fs_info->super_copy); |
ee22184b | 4449 | thresh = max_t(u64, SZ_64M, div_factor_fine(thresh, 1)); |
0e4f8f88 CM |
4450 | |
4451 | if (num_bytes - num_allocated < thresh) | |
4452 | return 1; | |
4453 | } | |
0e4f8f88 | 4454 | |
ee22184b | 4455 | if (num_allocated + SZ_2M < div_factor(num_bytes, 8)) |
14ed0ca6 | 4456 | return 0; |
424499db | 4457 | return 1; |
32c00aff JB |
4458 | } |
4459 | ||
2ff7e61e | 4460 | static u64 get_profile_num_devs(struct btrfs_fs_info *fs_info, u64 type) |
15d1ff81 LB |
4461 | { |
4462 | u64 num_dev; | |
4463 | ||
53b381b3 DW |
4464 | if (type & (BTRFS_BLOCK_GROUP_RAID10 | |
4465 | BTRFS_BLOCK_GROUP_RAID0 | | |
4466 | BTRFS_BLOCK_GROUP_RAID5 | | |
4467 | BTRFS_BLOCK_GROUP_RAID6)) | |
0b246afa | 4468 | num_dev = fs_info->fs_devices->rw_devices; |
15d1ff81 LB |
4469 | else if (type & BTRFS_BLOCK_GROUP_RAID1) |
4470 | num_dev = 2; | |
4471 | else | |
4472 | num_dev = 1; /* DUP or single */ | |
4473 | ||
39c2d7fa | 4474 | return num_dev; |
15d1ff81 LB |
4475 | } |
4476 | ||
39c2d7fa FM |
4477 | /* |
4478 | * If @is_allocation is true, reserve space in the system space info necessary | |
4479 | * for allocating a chunk, otherwise if it's false, reserve space necessary for | |
4480 | * removing a chunk. | |
4481 | */ | |
4482 | void check_system_chunk(struct btrfs_trans_handle *trans, | |
2ff7e61e | 4483 | struct btrfs_fs_info *fs_info, u64 type) |
15d1ff81 LB |
4484 | { |
4485 | struct btrfs_space_info *info; | |
4486 | u64 left; | |
4487 | u64 thresh; | |
4fbcdf66 | 4488 | int ret = 0; |
39c2d7fa | 4489 | u64 num_devs; |
4fbcdf66 FM |
4490 | |
4491 | /* | |
4492 | * Needed because we can end up allocating a system chunk and for an | |
4493 | * atomic and race free space reservation in the chunk block reserve. | |
4494 | */ | |
0b246afa | 4495 | ASSERT(mutex_is_locked(&fs_info->chunk_mutex)); |
15d1ff81 | 4496 | |
0b246afa | 4497 | info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); |
15d1ff81 | 4498 | spin_lock(&info->lock); |
4136135b | 4499 | left = info->total_bytes - btrfs_space_info_used(info, true); |
15d1ff81 LB |
4500 | spin_unlock(&info->lock); |
4501 | ||
2ff7e61e | 4502 | num_devs = get_profile_num_devs(fs_info, type); |
39c2d7fa FM |
4503 | |
4504 | /* num_devs device items to update and 1 chunk item to add or remove */ | |
0b246afa JM |
4505 | thresh = btrfs_calc_trunc_metadata_size(fs_info, num_devs) + |
4506 | btrfs_calc_trans_metadata_size(fs_info, 1); | |
39c2d7fa | 4507 | |
0b246afa JM |
4508 | if (left < thresh && btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
4509 | btrfs_info(fs_info, "left=%llu, need=%llu, flags=%llu", | |
4510 | left, thresh, type); | |
4511 | dump_space_info(fs_info, info, 0, 0); | |
15d1ff81 LB |
4512 | } |
4513 | ||
4514 | if (left < thresh) { | |
1b86826d | 4515 | u64 flags = btrfs_system_alloc_profile(fs_info); |
15d1ff81 | 4516 | |
4fbcdf66 FM |
4517 | /* |
4518 | * Ignore failure to create system chunk. We might end up not | |
4519 | * needing it, as we might not need to COW all nodes/leafs from | |
4520 | * the paths we visit in the chunk tree (they were already COWed | |
4521 | * or created in the current transaction for example). | |
4522 | */ | |
2ff7e61e | 4523 | ret = btrfs_alloc_chunk(trans, fs_info, flags); |
4fbcdf66 FM |
4524 | } |
4525 | ||
4526 | if (!ret) { | |
0b246afa JM |
4527 | ret = btrfs_block_rsv_add(fs_info->chunk_root, |
4528 | &fs_info->chunk_block_rsv, | |
4fbcdf66 FM |
4529 | thresh, BTRFS_RESERVE_NO_FLUSH); |
4530 | if (!ret) | |
4531 | trans->chunk_bytes_reserved += thresh; | |
15d1ff81 LB |
4532 | } |
4533 | } | |
4534 | ||
28b737f6 LB |
4535 | /* |
4536 | * If force is CHUNK_ALLOC_FORCE: | |
4537 | * - return 1 if it successfully allocates a chunk, | |
4538 | * - return errors including -ENOSPC otherwise. | |
4539 | * If force is NOT CHUNK_ALLOC_FORCE: | |
4540 | * - return 0 if it doesn't need to allocate a new chunk, | |
4541 | * - return 1 if it successfully allocates a chunk, | |
4542 | * - return errors including -ENOSPC otherwise. | |
4543 | */ | |
6324fbf3 | 4544 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
2ff7e61e | 4545 | struct btrfs_fs_info *fs_info, u64 flags, int force) |
9ed74f2d | 4546 | { |
6324fbf3 | 4547 | struct btrfs_space_info *space_info; |
6d74119f | 4548 | int wait_for_alloc = 0; |
9ed74f2d | 4549 | int ret = 0; |
9ed74f2d | 4550 | |
c6b305a8 JB |
4551 | /* Don't re-enter if we're already allocating a chunk */ |
4552 | if (trans->allocating_chunk) | |
4553 | return -ENOSPC; | |
4554 | ||
0b246afa | 4555 | space_info = __find_space_info(fs_info, flags); |
593060d7 | 4556 | if (!space_info) { |
2be12ef7 NB |
4557 | ret = create_space_info(fs_info, flags, &space_info); |
4558 | if (ret) | |
4559 | return ret; | |
9ed74f2d JB |
4560 | } |
4561 | ||
6d74119f | 4562 | again: |
25179201 | 4563 | spin_lock(&space_info->lock); |
9e622d6b | 4564 | if (force < space_info->force_alloc) |
0e4f8f88 | 4565 | force = space_info->force_alloc; |
25179201 | 4566 | if (space_info->full) { |
2ff7e61e | 4567 | if (should_alloc_chunk(fs_info, space_info, force)) |
09fb99a6 FDBM |
4568 | ret = -ENOSPC; |
4569 | else | |
4570 | ret = 0; | |
25179201 | 4571 | spin_unlock(&space_info->lock); |
09fb99a6 | 4572 | return ret; |
9ed74f2d JB |
4573 | } |
4574 | ||
2ff7e61e | 4575 | if (!should_alloc_chunk(fs_info, space_info, force)) { |
25179201 | 4576 | spin_unlock(&space_info->lock); |
6d74119f JB |
4577 | return 0; |
4578 | } else if (space_info->chunk_alloc) { | |
4579 | wait_for_alloc = 1; | |
4580 | } else { | |
4581 | space_info->chunk_alloc = 1; | |
9ed74f2d | 4582 | } |
0e4f8f88 | 4583 | |
25179201 | 4584 | spin_unlock(&space_info->lock); |
9ed74f2d | 4585 | |
6d74119f JB |
4586 | mutex_lock(&fs_info->chunk_mutex); |
4587 | ||
4588 | /* | |
4589 | * The chunk_mutex is held throughout the entirety of a chunk | |
4590 | * allocation, so once we've acquired the chunk_mutex we know that the | |
4591 | * other guy is done and we need to recheck and see if we should | |
4592 | * allocate. | |
4593 | */ | |
4594 | if (wait_for_alloc) { | |
4595 | mutex_unlock(&fs_info->chunk_mutex); | |
4596 | wait_for_alloc = 0; | |
4597 | goto again; | |
4598 | } | |
4599 | ||
c6b305a8 JB |
4600 | trans->allocating_chunk = true; |
4601 | ||
67377734 JB |
4602 | /* |
4603 | * If we have mixed data/metadata chunks we want to make sure we keep | |
4604 | * allocating mixed chunks instead of individual chunks. | |
4605 | */ | |
4606 | if (btrfs_mixed_space_info(space_info)) | |
4607 | flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA); | |
4608 | ||
97e728d4 JB |
4609 | /* |
4610 | * if we're doing a data chunk, go ahead and make sure that | |
4611 | * we keep a reasonable number of metadata chunks allocated in the | |
4612 | * FS as well. | |
4613 | */ | |
9ed74f2d | 4614 | if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) { |
97e728d4 JB |
4615 | fs_info->data_chunk_allocations++; |
4616 | if (!(fs_info->data_chunk_allocations % | |
4617 | fs_info->metadata_ratio)) | |
4618 | force_metadata_allocation(fs_info); | |
9ed74f2d JB |
4619 | } |
4620 | ||
15d1ff81 LB |
4621 | /* |
4622 | * Check if we have enough space in SYSTEM chunk because we may need | |
4623 | * to update devices. | |
4624 | */ | |
2ff7e61e | 4625 | check_system_chunk(trans, fs_info, flags); |
15d1ff81 | 4626 | |
2ff7e61e | 4627 | ret = btrfs_alloc_chunk(trans, fs_info, flags); |
c6b305a8 | 4628 | trans->allocating_chunk = false; |
92b8e897 | 4629 | |
9ed74f2d | 4630 | spin_lock(&space_info->lock); |
a81cb9a2 AO |
4631 | if (ret < 0 && ret != -ENOSPC) |
4632 | goto out; | |
9ed74f2d | 4633 | if (ret) |
6324fbf3 | 4634 | space_info->full = 1; |
424499db YZ |
4635 | else |
4636 | ret = 1; | |
6d74119f | 4637 | |
0e4f8f88 | 4638 | space_info->force_alloc = CHUNK_ALLOC_NO_FORCE; |
a81cb9a2 | 4639 | out: |
6d74119f | 4640 | space_info->chunk_alloc = 0; |
9ed74f2d | 4641 | spin_unlock(&space_info->lock); |
a25c75d5 | 4642 | mutex_unlock(&fs_info->chunk_mutex); |
00d80e34 FM |
4643 | /* |
4644 | * When we allocate a new chunk we reserve space in the chunk block | |
4645 | * reserve to make sure we can COW nodes/leafs in the chunk tree or | |
4646 | * add new nodes/leafs to it if we end up needing to do it when | |
4647 | * inserting the chunk item and updating device items as part of the | |
4648 | * second phase of chunk allocation, performed by | |
4649 | * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a | |
4650 | * large number of new block groups to create in our transaction | |
4651 | * handle's new_bgs list to avoid exhausting the chunk block reserve | |
4652 | * in extreme cases - like having a single transaction create many new | |
4653 | * block groups when starting to write out the free space caches of all | |
4654 | * the block groups that were made dirty during the lifetime of the | |
4655 | * transaction. | |
4656 | */ | |
d9a0540a | 4657 | if (trans->can_flush_pending_bgs && |
ee22184b | 4658 | trans->chunk_bytes_reserved >= (u64)SZ_2M) { |
2ff7e61e | 4659 | btrfs_create_pending_block_groups(trans, fs_info); |
00d80e34 FM |
4660 | btrfs_trans_release_chunk_metadata(trans); |
4661 | } | |
0f9dd46c | 4662 | return ret; |
6324fbf3 | 4663 | } |
9ed74f2d | 4664 | |
c1c4919b | 4665 | static int can_overcommit(struct btrfs_fs_info *fs_info, |
a80c8dcf | 4666 | struct btrfs_space_info *space_info, u64 bytes, |
c1c4919b JM |
4667 | enum btrfs_reserve_flush_enum flush, |
4668 | bool system_chunk) | |
a80c8dcf | 4669 | { |
0b246afa | 4670 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; |
957780eb | 4671 | u64 profile; |
3c76cd84 | 4672 | u64 space_size; |
a80c8dcf JB |
4673 | u64 avail; |
4674 | u64 used; | |
4675 | ||
957780eb JB |
4676 | /* Don't overcommit when in mixed mode. */ |
4677 | if (space_info->flags & BTRFS_BLOCK_GROUP_DATA) | |
4678 | return 0; | |
4679 | ||
c1c4919b JM |
4680 | if (system_chunk) |
4681 | profile = btrfs_system_alloc_profile(fs_info); | |
4682 | else | |
4683 | profile = btrfs_metadata_alloc_profile(fs_info); | |
4684 | ||
4136135b | 4685 | used = btrfs_space_info_used(space_info, false); |
96f1bb57 | 4686 | |
96f1bb57 JB |
4687 | /* |
4688 | * We only want to allow over committing if we have lots of actual space | |
4689 | * free, but if we don't have enough space to handle the global reserve | |
4690 | * space then we could end up having a real enospc problem when trying | |
4691 | * to allocate a chunk or some other such important allocation. | |
4692 | */ | |
3c76cd84 MX |
4693 | spin_lock(&global_rsv->lock); |
4694 | space_size = calc_global_rsv_need_space(global_rsv); | |
4695 | spin_unlock(&global_rsv->lock); | |
4696 | if (used + space_size >= space_info->total_bytes) | |
96f1bb57 JB |
4697 | return 0; |
4698 | ||
4699 | used += space_info->bytes_may_use; | |
a80c8dcf | 4700 | |
a5ed45f8 | 4701 | avail = atomic64_read(&fs_info->free_chunk_space); |
a80c8dcf JB |
4702 | |
4703 | /* | |
4704 | * If we have dup, raid1 or raid10 then only half of the free | |
53b381b3 DW |
4705 | * space is actually useable. For raid56, the space info used |
4706 | * doesn't include the parity drive, so we don't have to | |
4707 | * change the math | |
a80c8dcf JB |
4708 | */ |
4709 | if (profile & (BTRFS_BLOCK_GROUP_DUP | | |
4710 | BTRFS_BLOCK_GROUP_RAID1 | | |
4711 | BTRFS_BLOCK_GROUP_RAID10)) | |
4712 | avail >>= 1; | |
4713 | ||
4714 | /* | |
561c294d MX |
4715 | * If we aren't flushing all things, let us overcommit up to |
4716 | * 1/2th of the space. If we can flush, don't let us overcommit | |
4717 | * too much, let it overcommit up to 1/8 of the space. | |
a80c8dcf | 4718 | */ |
08e007d2 | 4719 | if (flush == BTRFS_RESERVE_FLUSH_ALL) |
14575aef | 4720 | avail >>= 3; |
a80c8dcf | 4721 | else |
14575aef | 4722 | avail >>= 1; |
a80c8dcf | 4723 | |
14575aef | 4724 | if (used + bytes < space_info->total_bytes + avail) |
a80c8dcf JB |
4725 | return 1; |
4726 | return 0; | |
4727 | } | |
4728 | ||
2ff7e61e | 4729 | static void btrfs_writeback_inodes_sb_nr(struct btrfs_fs_info *fs_info, |
6c255e67 | 4730 | unsigned long nr_pages, int nr_items) |
da633a42 | 4731 | { |
0b246afa | 4732 | struct super_block *sb = fs_info->sb; |
da633a42 | 4733 | |
925a6efb JB |
4734 | if (down_read_trylock(&sb->s_umount)) { |
4735 | writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE); | |
4736 | up_read(&sb->s_umount); | |
4737 | } else { | |
da633a42 MX |
4738 | /* |
4739 | * We needn't worry the filesystem going from r/w to r/o though | |
4740 | * we don't acquire ->s_umount mutex, because the filesystem | |
4741 | * should guarantee the delalloc inodes list be empty after | |
4742 | * the filesystem is readonly(all dirty pages are written to | |
4743 | * the disk). | |
4744 | */ | |
0b246afa | 4745 | btrfs_start_delalloc_roots(fs_info, 0, nr_items); |
98ad69cf | 4746 | if (!current->journal_info) |
0b246afa | 4747 | btrfs_wait_ordered_roots(fs_info, nr_items, 0, (u64)-1); |
da633a42 MX |
4748 | } |
4749 | } | |
4750 | ||
6374e57a | 4751 | static inline u64 calc_reclaim_items_nr(struct btrfs_fs_info *fs_info, |
2ff7e61e | 4752 | u64 to_reclaim) |
18cd8ea6 MX |
4753 | { |
4754 | u64 bytes; | |
6374e57a | 4755 | u64 nr; |
18cd8ea6 | 4756 | |
2ff7e61e | 4757 | bytes = btrfs_calc_trans_metadata_size(fs_info, 1); |
6374e57a | 4758 | nr = div64_u64(to_reclaim, bytes); |
18cd8ea6 MX |
4759 | if (!nr) |
4760 | nr = 1; | |
4761 | return nr; | |
4762 | } | |
4763 | ||
ee22184b | 4764 | #define EXTENT_SIZE_PER_ITEM SZ_256K |
c61a16a7 | 4765 | |
9ed74f2d | 4766 | /* |
5da9d01b | 4767 | * shrink metadata reservation for delalloc |
9ed74f2d | 4768 | */ |
c1c4919b JM |
4769 | static void shrink_delalloc(struct btrfs_fs_info *fs_info, u64 to_reclaim, |
4770 | u64 orig, bool wait_ordered) | |
5da9d01b | 4771 | { |
0ca1f7ce | 4772 | struct btrfs_block_rsv *block_rsv; |
0019f10d | 4773 | struct btrfs_space_info *space_info; |
663350ac | 4774 | struct btrfs_trans_handle *trans; |
f4c738c2 | 4775 | u64 delalloc_bytes; |
5da9d01b | 4776 | u64 max_reclaim; |
6374e57a | 4777 | u64 items; |
b1953bce | 4778 | long time_left; |
d3ee29e3 MX |
4779 | unsigned long nr_pages; |
4780 | int loops; | |
08e007d2 | 4781 | enum btrfs_reserve_flush_enum flush; |
5da9d01b | 4782 | |
c61a16a7 | 4783 | /* Calc the number of the pages we need flush for space reservation */ |
2ff7e61e | 4784 | items = calc_reclaim_items_nr(fs_info, to_reclaim); |
6374e57a | 4785 | to_reclaim = items * EXTENT_SIZE_PER_ITEM; |
c61a16a7 | 4786 | |
663350ac | 4787 | trans = (struct btrfs_trans_handle *)current->journal_info; |
0b246afa | 4788 | block_rsv = &fs_info->delalloc_block_rsv; |
0019f10d | 4789 | space_info = block_rsv->space_info; |
bf9022e0 | 4790 | |
963d678b | 4791 | delalloc_bytes = percpu_counter_sum_positive( |
0b246afa | 4792 | &fs_info->delalloc_bytes); |
f4c738c2 | 4793 | if (delalloc_bytes == 0) { |
fdb5effd | 4794 | if (trans) |
f4c738c2 | 4795 | return; |
38c135af | 4796 | if (wait_ordered) |
0b246afa | 4797 | btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1); |
f4c738c2 | 4798 | return; |
fdb5effd JB |
4799 | } |
4800 | ||
d3ee29e3 | 4801 | loops = 0; |
f4c738c2 JB |
4802 | while (delalloc_bytes && loops < 3) { |
4803 | max_reclaim = min(delalloc_bytes, to_reclaim); | |
09cbfeaf | 4804 | nr_pages = max_reclaim >> PAGE_SHIFT; |
2ff7e61e | 4805 | btrfs_writeback_inodes_sb_nr(fs_info, nr_pages, items); |
dea31f52 JB |
4806 | /* |
4807 | * We need to wait for the async pages to actually start before | |
4808 | * we do anything. | |
4809 | */ | |
0b246afa | 4810 | max_reclaim = atomic_read(&fs_info->async_delalloc_pages); |
9f3a074d MX |
4811 | if (!max_reclaim) |
4812 | goto skip_async; | |
4813 | ||
4814 | if (max_reclaim <= nr_pages) | |
4815 | max_reclaim = 0; | |
4816 | else | |
4817 | max_reclaim -= nr_pages; | |
dea31f52 | 4818 | |
0b246afa JM |
4819 | wait_event(fs_info->async_submit_wait, |
4820 | atomic_read(&fs_info->async_delalloc_pages) <= | |
9f3a074d MX |
4821 | (int)max_reclaim); |
4822 | skip_async: | |
08e007d2 MX |
4823 | if (!trans) |
4824 | flush = BTRFS_RESERVE_FLUSH_ALL; | |
4825 | else | |
4826 | flush = BTRFS_RESERVE_NO_FLUSH; | |
0019f10d | 4827 | spin_lock(&space_info->lock); |
957780eb JB |
4828 | if (list_empty(&space_info->tickets) && |
4829 | list_empty(&space_info->priority_tickets)) { | |
4830 | spin_unlock(&space_info->lock); | |
4831 | break; | |
4832 | } | |
0019f10d | 4833 | spin_unlock(&space_info->lock); |
5da9d01b | 4834 | |
36e39c40 | 4835 | loops++; |
f104d044 | 4836 | if (wait_ordered && !trans) { |
0b246afa | 4837 | btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1); |
f104d044 | 4838 | } else { |
f4c738c2 | 4839 | time_left = schedule_timeout_killable(1); |
f104d044 JB |
4840 | if (time_left) |
4841 | break; | |
4842 | } | |
963d678b | 4843 | delalloc_bytes = percpu_counter_sum_positive( |
0b246afa | 4844 | &fs_info->delalloc_bytes); |
5da9d01b | 4845 | } |
5da9d01b YZ |
4846 | } |
4847 | ||
663350ac JB |
4848 | /** |
4849 | * maybe_commit_transaction - possibly commit the transaction if its ok to | |
4850 | * @root - the root we're allocating for | |
4851 | * @bytes - the number of bytes we want to reserve | |
4852 | * @force - force the commit | |
8bb8ab2e | 4853 | * |
663350ac JB |
4854 | * This will check to make sure that committing the transaction will actually |
4855 | * get us somewhere and then commit the transaction if it does. Otherwise it | |
4856 | * will return -ENOSPC. | |
8bb8ab2e | 4857 | */ |
0c9ab349 | 4858 | static int may_commit_transaction(struct btrfs_fs_info *fs_info, |
663350ac JB |
4859 | struct btrfs_space_info *space_info, |
4860 | u64 bytes, int force) | |
4861 | { | |
0b246afa | 4862 | struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_block_rsv; |
663350ac JB |
4863 | struct btrfs_trans_handle *trans; |
4864 | ||
4865 | trans = (struct btrfs_trans_handle *)current->journal_info; | |
4866 | if (trans) | |
4867 | return -EAGAIN; | |
4868 | ||
4869 | if (force) | |
4870 | goto commit; | |
4871 | ||
4872 | /* See if there is enough pinned space to make this reservation */ | |
b150a4f1 | 4873 | if (percpu_counter_compare(&space_info->total_bytes_pinned, |
0424c548 | 4874 | bytes) >= 0) |
663350ac | 4875 | goto commit; |
663350ac JB |
4876 | |
4877 | /* | |
4878 | * See if there is some space in the delayed insertion reservation for | |
4879 | * this reservation. | |
4880 | */ | |
4881 | if (space_info != delayed_rsv->space_info) | |
4882 | return -ENOSPC; | |
4883 | ||
4884 | spin_lock(&delayed_rsv->lock); | |
b150a4f1 | 4885 | if (percpu_counter_compare(&space_info->total_bytes_pinned, |
28785f70 | 4886 | bytes - delayed_rsv->size) < 0) { |
663350ac JB |
4887 | spin_unlock(&delayed_rsv->lock); |
4888 | return -ENOSPC; | |
4889 | } | |
4890 | spin_unlock(&delayed_rsv->lock); | |
4891 | ||
4892 | commit: | |
a9b3311e | 4893 | trans = btrfs_join_transaction(fs_info->extent_root); |
663350ac JB |
4894 | if (IS_ERR(trans)) |
4895 | return -ENOSPC; | |
4896 | ||
3a45bb20 | 4897 | return btrfs_commit_transaction(trans); |
663350ac JB |
4898 | } |
4899 | ||
957780eb JB |
4900 | struct reserve_ticket { |
4901 | u64 bytes; | |
4902 | int error; | |
4903 | struct list_head list; | |
4904 | wait_queue_head_t wait; | |
96c3f433 JB |
4905 | }; |
4906 | ||
0c9ab349 | 4907 | static int flush_space(struct btrfs_fs_info *fs_info, |
96c3f433 JB |
4908 | struct btrfs_space_info *space_info, u64 num_bytes, |
4909 | u64 orig_bytes, int state) | |
4910 | { | |
a9b3311e | 4911 | struct btrfs_root *root = fs_info->extent_root; |
96c3f433 JB |
4912 | struct btrfs_trans_handle *trans; |
4913 | int nr; | |
f4c738c2 | 4914 | int ret = 0; |
96c3f433 JB |
4915 | |
4916 | switch (state) { | |
96c3f433 JB |
4917 | case FLUSH_DELAYED_ITEMS_NR: |
4918 | case FLUSH_DELAYED_ITEMS: | |
18cd8ea6 | 4919 | if (state == FLUSH_DELAYED_ITEMS_NR) |
2ff7e61e | 4920 | nr = calc_reclaim_items_nr(fs_info, num_bytes) * 2; |
18cd8ea6 | 4921 | else |
96c3f433 | 4922 | nr = -1; |
18cd8ea6 | 4923 | |
96c3f433 JB |
4924 | trans = btrfs_join_transaction(root); |
4925 | if (IS_ERR(trans)) { | |
4926 | ret = PTR_ERR(trans); | |
4927 | break; | |
4928 | } | |
2ff7e61e | 4929 | ret = btrfs_run_delayed_items_nr(trans, fs_info, nr); |
3a45bb20 | 4930 | btrfs_end_transaction(trans); |
96c3f433 | 4931 | break; |
67b0fd63 JB |
4932 | case FLUSH_DELALLOC: |
4933 | case FLUSH_DELALLOC_WAIT: | |
c1c4919b | 4934 | shrink_delalloc(fs_info, num_bytes * 2, orig_bytes, |
67b0fd63 JB |
4935 | state == FLUSH_DELALLOC_WAIT); |
4936 | break; | |
ea658bad JB |
4937 | case ALLOC_CHUNK: |
4938 | trans = btrfs_join_transaction(root); | |
4939 | if (IS_ERR(trans)) { | |
4940 | ret = PTR_ERR(trans); | |
4941 | break; | |
4942 | } | |
2ff7e61e | 4943 | ret = do_chunk_alloc(trans, fs_info, |
1b86826d | 4944 | btrfs_metadata_alloc_profile(fs_info), |
ea658bad | 4945 | CHUNK_ALLOC_NO_FORCE); |
3a45bb20 | 4946 | btrfs_end_transaction(trans); |
eecba891 | 4947 | if (ret > 0 || ret == -ENOSPC) |
ea658bad JB |
4948 | ret = 0; |
4949 | break; | |
96c3f433 | 4950 | case COMMIT_TRANS: |
0c9ab349 JM |
4951 | ret = may_commit_transaction(fs_info, space_info, |
4952 | orig_bytes, 0); | |
96c3f433 JB |
4953 | break; |
4954 | default: | |
4955 | ret = -ENOSPC; | |
4956 | break; | |
4957 | } | |
4958 | ||
0b246afa | 4959 | trace_btrfs_flush_space(fs_info, space_info->flags, num_bytes, |
f376df2b | 4960 | orig_bytes, state, ret); |
96c3f433 JB |
4961 | return ret; |
4962 | } | |
21c7e756 MX |
4963 | |
4964 | static inline u64 | |
c1c4919b JM |
4965 | btrfs_calc_reclaim_metadata_size(struct btrfs_fs_info *fs_info, |
4966 | struct btrfs_space_info *space_info, | |
4967 | bool system_chunk) | |
21c7e756 | 4968 | { |
957780eb | 4969 | struct reserve_ticket *ticket; |
21c7e756 MX |
4970 | u64 used; |
4971 | u64 expected; | |
957780eb | 4972 | u64 to_reclaim = 0; |
21c7e756 | 4973 | |
957780eb JB |
4974 | list_for_each_entry(ticket, &space_info->tickets, list) |
4975 | to_reclaim += ticket->bytes; | |
4976 | list_for_each_entry(ticket, &space_info->priority_tickets, list) | |
4977 | to_reclaim += ticket->bytes; | |
4978 | if (to_reclaim) | |
4979 | return to_reclaim; | |
21c7e756 | 4980 | |
e0af2484 | 4981 | to_reclaim = min_t(u64, num_online_cpus() * SZ_1M, SZ_16M); |
c1c4919b JM |
4982 | if (can_overcommit(fs_info, space_info, to_reclaim, |
4983 | BTRFS_RESERVE_FLUSH_ALL, system_chunk)) | |
e0af2484 WX |
4984 | return 0; |
4985 | ||
0eee8a49 NB |
4986 | used = btrfs_space_info_used(space_info, true); |
4987 | ||
c1c4919b JM |
4988 | if (can_overcommit(fs_info, space_info, SZ_1M, |
4989 | BTRFS_RESERVE_FLUSH_ALL, system_chunk)) | |
21c7e756 MX |
4990 | expected = div_factor_fine(space_info->total_bytes, 95); |
4991 | else | |
4992 | expected = div_factor_fine(space_info->total_bytes, 90); | |
4993 | ||
4994 | if (used > expected) | |
4995 | to_reclaim = used - expected; | |
4996 | else | |
4997 | to_reclaim = 0; | |
4998 | to_reclaim = min(to_reclaim, space_info->bytes_may_use + | |
4999 | space_info->bytes_reserved); | |
21c7e756 MX |
5000 | return to_reclaim; |
5001 | } | |
5002 | ||
c1c4919b JM |
5003 | static inline int need_do_async_reclaim(struct btrfs_fs_info *fs_info, |
5004 | struct btrfs_space_info *space_info, | |
5005 | u64 used, bool system_chunk) | |
21c7e756 | 5006 | { |
365c5313 JB |
5007 | u64 thresh = div_factor_fine(space_info->total_bytes, 98); |
5008 | ||
5009 | /* If we're just plain full then async reclaim just slows us down. */ | |
baee8790 | 5010 | if ((space_info->bytes_used + space_info->bytes_reserved) >= thresh) |
365c5313 JB |
5011 | return 0; |
5012 | ||
c1c4919b JM |
5013 | if (!btrfs_calc_reclaim_metadata_size(fs_info, space_info, |
5014 | system_chunk)) | |
d38b349c JB |
5015 | return 0; |
5016 | ||
0b246afa JM |
5017 | return (used >= thresh && !btrfs_fs_closing(fs_info) && |
5018 | !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state)); | |
21c7e756 MX |
5019 | } |
5020 | ||
957780eb | 5021 | static void wake_all_tickets(struct list_head *head) |
21c7e756 | 5022 | { |
957780eb | 5023 | struct reserve_ticket *ticket; |
25ce459c | 5024 | |
957780eb JB |
5025 | while (!list_empty(head)) { |
5026 | ticket = list_first_entry(head, struct reserve_ticket, list); | |
5027 | list_del_init(&ticket->list); | |
5028 | ticket->error = -ENOSPC; | |
5029 | wake_up(&ticket->wait); | |
21c7e756 | 5030 | } |
21c7e756 MX |
5031 | } |
5032 | ||
957780eb JB |
5033 | /* |
5034 | * This is for normal flushers, we can wait all goddamned day if we want to. We | |
5035 | * will loop and continuously try to flush as long as we are making progress. | |
5036 | * We count progress as clearing off tickets each time we have to loop. | |
5037 | */ | |
21c7e756 MX |
5038 | static void btrfs_async_reclaim_metadata_space(struct work_struct *work) |
5039 | { | |
5040 | struct btrfs_fs_info *fs_info; | |
5041 | struct btrfs_space_info *space_info; | |
5042 | u64 to_reclaim; | |
5043 | int flush_state; | |
957780eb | 5044 | int commit_cycles = 0; |
ce129655 | 5045 | u64 last_tickets_id; |
21c7e756 MX |
5046 | |
5047 | fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work); | |
5048 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); | |
5049 | ||
957780eb | 5050 | spin_lock(&space_info->lock); |
c1c4919b JM |
5051 | to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info, |
5052 | false); | |
957780eb JB |
5053 | if (!to_reclaim) { |
5054 | space_info->flush = 0; | |
5055 | spin_unlock(&space_info->lock); | |
21c7e756 | 5056 | return; |
957780eb | 5057 | } |
ce129655 | 5058 | last_tickets_id = space_info->tickets_id; |
957780eb | 5059 | spin_unlock(&space_info->lock); |
21c7e756 MX |
5060 | |
5061 | flush_state = FLUSH_DELAYED_ITEMS_NR; | |
957780eb JB |
5062 | do { |
5063 | struct reserve_ticket *ticket; | |
5064 | int ret; | |
5065 | ||
0c9ab349 JM |
5066 | ret = flush_space(fs_info, space_info, to_reclaim, to_reclaim, |
5067 | flush_state); | |
957780eb JB |
5068 | spin_lock(&space_info->lock); |
5069 | if (list_empty(&space_info->tickets)) { | |
5070 | space_info->flush = 0; | |
5071 | spin_unlock(&space_info->lock); | |
5072 | return; | |
5073 | } | |
c1c4919b JM |
5074 | to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, |
5075 | space_info, | |
5076 | false); | |
957780eb JB |
5077 | ticket = list_first_entry(&space_info->tickets, |
5078 | struct reserve_ticket, list); | |
ce129655 | 5079 | if (last_tickets_id == space_info->tickets_id) { |
957780eb JB |
5080 | flush_state++; |
5081 | } else { | |
ce129655 | 5082 | last_tickets_id = space_info->tickets_id; |
957780eb JB |
5083 | flush_state = FLUSH_DELAYED_ITEMS_NR; |
5084 | if (commit_cycles) | |
5085 | commit_cycles--; | |
5086 | } | |
5087 | ||
5088 | if (flush_state > COMMIT_TRANS) { | |
5089 | commit_cycles++; | |
5090 | if (commit_cycles > 2) { | |
5091 | wake_all_tickets(&space_info->tickets); | |
5092 | space_info->flush = 0; | |
5093 | } else { | |
5094 | flush_state = FLUSH_DELAYED_ITEMS_NR; | |
5095 | } | |
5096 | } | |
5097 | spin_unlock(&space_info->lock); | |
5098 | } while (flush_state <= COMMIT_TRANS); | |
5099 | } | |
5100 | ||
5101 | void btrfs_init_async_reclaim_work(struct work_struct *work) | |
5102 | { | |
5103 | INIT_WORK(work, btrfs_async_reclaim_metadata_space); | |
5104 | } | |
5105 | ||
5106 | static void priority_reclaim_metadata_space(struct btrfs_fs_info *fs_info, | |
5107 | struct btrfs_space_info *space_info, | |
5108 | struct reserve_ticket *ticket) | |
5109 | { | |
5110 | u64 to_reclaim; | |
5111 | int flush_state = FLUSH_DELAYED_ITEMS_NR; | |
5112 | ||
5113 | spin_lock(&space_info->lock); | |
c1c4919b JM |
5114 | to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info, |
5115 | false); | |
957780eb JB |
5116 | if (!to_reclaim) { |
5117 | spin_unlock(&space_info->lock); | |
5118 | return; | |
5119 | } | |
5120 | spin_unlock(&space_info->lock); | |
5121 | ||
21c7e756 | 5122 | do { |
0c9ab349 JM |
5123 | flush_space(fs_info, space_info, to_reclaim, to_reclaim, |
5124 | flush_state); | |
21c7e756 | 5125 | flush_state++; |
957780eb JB |
5126 | spin_lock(&space_info->lock); |
5127 | if (ticket->bytes == 0) { | |
5128 | spin_unlock(&space_info->lock); | |
21c7e756 | 5129 | return; |
957780eb JB |
5130 | } |
5131 | spin_unlock(&space_info->lock); | |
5132 | ||
5133 | /* | |
5134 | * Priority flushers can't wait on delalloc without | |
5135 | * deadlocking. | |
5136 | */ | |
5137 | if (flush_state == FLUSH_DELALLOC || | |
5138 | flush_state == FLUSH_DELALLOC_WAIT) | |
5139 | flush_state = ALLOC_CHUNK; | |
365c5313 | 5140 | } while (flush_state < COMMIT_TRANS); |
21c7e756 MX |
5141 | } |
5142 | ||
957780eb JB |
5143 | static int wait_reserve_ticket(struct btrfs_fs_info *fs_info, |
5144 | struct btrfs_space_info *space_info, | |
5145 | struct reserve_ticket *ticket, u64 orig_bytes) | |
5146 | ||
21c7e756 | 5147 | { |
957780eb JB |
5148 | DEFINE_WAIT(wait); |
5149 | int ret = 0; | |
5150 | ||
5151 | spin_lock(&space_info->lock); | |
5152 | while (ticket->bytes > 0 && ticket->error == 0) { | |
5153 | ret = prepare_to_wait_event(&ticket->wait, &wait, TASK_KILLABLE); | |
5154 | if (ret) { | |
5155 | ret = -EINTR; | |
5156 | break; | |
5157 | } | |
5158 | spin_unlock(&space_info->lock); | |
5159 | ||
5160 | schedule(); | |
5161 | ||
5162 | finish_wait(&ticket->wait, &wait); | |
5163 | spin_lock(&space_info->lock); | |
5164 | } | |
5165 | if (!ret) | |
5166 | ret = ticket->error; | |
5167 | if (!list_empty(&ticket->list)) | |
5168 | list_del_init(&ticket->list); | |
5169 | if (ticket->bytes && ticket->bytes < orig_bytes) { | |
5170 | u64 num_bytes = orig_bytes - ticket->bytes; | |
5171 | space_info->bytes_may_use -= num_bytes; | |
5172 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5173 | space_info->flags, num_bytes, 0); | |
5174 | } | |
5175 | spin_unlock(&space_info->lock); | |
5176 | ||
5177 | return ret; | |
21c7e756 MX |
5178 | } |
5179 | ||
4a92b1b8 JB |
5180 | /** |
5181 | * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space | |
5182 | * @root - the root we're allocating for | |
957780eb | 5183 | * @space_info - the space info we want to allocate from |
4a92b1b8 | 5184 | * @orig_bytes - the number of bytes we want |
48fc7f7e | 5185 | * @flush - whether or not we can flush to make our reservation |
8bb8ab2e | 5186 | * |
01327610 | 5187 | * This will reserve orig_bytes number of bytes from the space info associated |
4a92b1b8 JB |
5188 | * with the block_rsv. If there is not enough space it will make an attempt to |
5189 | * flush out space to make room. It will do this by flushing delalloc if | |
5190 | * possible or committing the transaction. If flush is 0 then no attempts to | |
5191 | * regain reservations will be made and this will fail if there is not enough | |
5192 | * space already. | |
8bb8ab2e | 5193 | */ |
c1c4919b | 5194 | static int __reserve_metadata_bytes(struct btrfs_fs_info *fs_info, |
957780eb JB |
5195 | struct btrfs_space_info *space_info, |
5196 | u64 orig_bytes, | |
c1c4919b JM |
5197 | enum btrfs_reserve_flush_enum flush, |
5198 | bool system_chunk) | |
9ed74f2d | 5199 | { |
957780eb | 5200 | struct reserve_ticket ticket; |
2bf64758 | 5201 | u64 used; |
8bb8ab2e | 5202 | int ret = 0; |
9ed74f2d | 5203 | |
957780eb | 5204 | ASSERT(orig_bytes); |
8ca17f0f | 5205 | ASSERT(!current->journal_info || flush != BTRFS_RESERVE_FLUSH_ALL); |
fdb5effd | 5206 | |
8bb8ab2e | 5207 | spin_lock(&space_info->lock); |
fdb5effd | 5208 | ret = -ENOSPC; |
4136135b | 5209 | used = btrfs_space_info_used(space_info, true); |
9ed74f2d | 5210 | |
8bb8ab2e | 5211 | /* |
957780eb JB |
5212 | * If we have enough space then hooray, make our reservation and carry |
5213 | * on. If not see if we can overcommit, and if we can, hooray carry on. | |
5214 | * If not things get more complicated. | |
8bb8ab2e | 5215 | */ |
957780eb JB |
5216 | if (used + orig_bytes <= space_info->total_bytes) { |
5217 | space_info->bytes_may_use += orig_bytes; | |
0b246afa JM |
5218 | trace_btrfs_space_reservation(fs_info, "space_info", |
5219 | space_info->flags, orig_bytes, 1); | |
957780eb | 5220 | ret = 0; |
c1c4919b JM |
5221 | } else if (can_overcommit(fs_info, space_info, orig_bytes, flush, |
5222 | system_chunk)) { | |
44734ed1 | 5223 | space_info->bytes_may_use += orig_bytes; |
0b246afa JM |
5224 | trace_btrfs_space_reservation(fs_info, "space_info", |
5225 | space_info->flags, orig_bytes, 1); | |
44734ed1 | 5226 | ret = 0; |
2bf64758 JB |
5227 | } |
5228 | ||
8bb8ab2e | 5229 | /* |
957780eb JB |
5230 | * If we couldn't make a reservation then setup our reservation ticket |
5231 | * and kick the async worker if it's not already running. | |
08e007d2 | 5232 | * |
957780eb JB |
5233 | * If we are a priority flusher then we just need to add our ticket to |
5234 | * the list and we will do our own flushing further down. | |
8bb8ab2e | 5235 | */ |
72bcd99d | 5236 | if (ret && flush != BTRFS_RESERVE_NO_FLUSH) { |
957780eb JB |
5237 | ticket.bytes = orig_bytes; |
5238 | ticket.error = 0; | |
5239 | init_waitqueue_head(&ticket.wait); | |
5240 | if (flush == BTRFS_RESERVE_FLUSH_ALL) { | |
5241 | list_add_tail(&ticket.list, &space_info->tickets); | |
5242 | if (!space_info->flush) { | |
5243 | space_info->flush = 1; | |
0b246afa | 5244 | trace_btrfs_trigger_flush(fs_info, |
f376df2b JB |
5245 | space_info->flags, |
5246 | orig_bytes, flush, | |
5247 | "enospc"); | |
957780eb | 5248 | queue_work(system_unbound_wq, |
c1c4919b | 5249 | &fs_info->async_reclaim_work); |
957780eb JB |
5250 | } |
5251 | } else { | |
5252 | list_add_tail(&ticket.list, | |
5253 | &space_info->priority_tickets); | |
5254 | } | |
21c7e756 MX |
5255 | } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) { |
5256 | used += orig_bytes; | |
f6acfd50 JB |
5257 | /* |
5258 | * We will do the space reservation dance during log replay, | |
5259 | * which means we won't have fs_info->fs_root set, so don't do | |
5260 | * the async reclaim as we will panic. | |
5261 | */ | |
0b246afa | 5262 | if (!test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags) && |
c1c4919b JM |
5263 | need_do_async_reclaim(fs_info, space_info, |
5264 | used, system_chunk) && | |
0b246afa JM |
5265 | !work_busy(&fs_info->async_reclaim_work)) { |
5266 | trace_btrfs_trigger_flush(fs_info, space_info->flags, | |
5267 | orig_bytes, flush, "preempt"); | |
21c7e756 | 5268 | queue_work(system_unbound_wq, |
0b246afa | 5269 | &fs_info->async_reclaim_work); |
f376df2b | 5270 | } |
8bb8ab2e | 5271 | } |
f0486c68 | 5272 | spin_unlock(&space_info->lock); |
08e007d2 | 5273 | if (!ret || flush == BTRFS_RESERVE_NO_FLUSH) |
957780eb | 5274 | return ret; |
f0486c68 | 5275 | |
957780eb | 5276 | if (flush == BTRFS_RESERVE_FLUSH_ALL) |
0b246afa | 5277 | return wait_reserve_ticket(fs_info, space_info, &ticket, |
957780eb | 5278 | orig_bytes); |
08e007d2 | 5279 | |
957780eb | 5280 | ret = 0; |
0b246afa | 5281 | priority_reclaim_metadata_space(fs_info, space_info, &ticket); |
957780eb JB |
5282 | spin_lock(&space_info->lock); |
5283 | if (ticket.bytes) { | |
5284 | if (ticket.bytes < orig_bytes) { | |
5285 | u64 num_bytes = orig_bytes - ticket.bytes; | |
5286 | space_info->bytes_may_use -= num_bytes; | |
0b246afa JM |
5287 | trace_btrfs_space_reservation(fs_info, "space_info", |
5288 | space_info->flags, | |
5289 | num_bytes, 0); | |
08e007d2 | 5290 | |
957780eb JB |
5291 | } |
5292 | list_del_init(&ticket.list); | |
5293 | ret = -ENOSPC; | |
5294 | } | |
5295 | spin_unlock(&space_info->lock); | |
5296 | ASSERT(list_empty(&ticket.list)); | |
5297 | return ret; | |
5298 | } | |
8bb8ab2e | 5299 | |
957780eb JB |
5300 | /** |
5301 | * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space | |
5302 | * @root - the root we're allocating for | |
5303 | * @block_rsv - the block_rsv we're allocating for | |
5304 | * @orig_bytes - the number of bytes we want | |
5305 | * @flush - whether or not we can flush to make our reservation | |
5306 | * | |
5307 | * This will reserve orgi_bytes number of bytes from the space info associated | |
5308 | * with the block_rsv. If there is not enough space it will make an attempt to | |
5309 | * flush out space to make room. It will do this by flushing delalloc if | |
5310 | * possible or committing the transaction. If flush is 0 then no attempts to | |
5311 | * regain reservations will be made and this will fail if there is not enough | |
5312 | * space already. | |
5313 | */ | |
5314 | static int reserve_metadata_bytes(struct btrfs_root *root, | |
5315 | struct btrfs_block_rsv *block_rsv, | |
5316 | u64 orig_bytes, | |
5317 | enum btrfs_reserve_flush_enum flush) | |
5318 | { | |
0b246afa JM |
5319 | struct btrfs_fs_info *fs_info = root->fs_info; |
5320 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
957780eb | 5321 | int ret; |
c1c4919b | 5322 | bool system_chunk = (root == fs_info->chunk_root); |
957780eb | 5323 | |
c1c4919b JM |
5324 | ret = __reserve_metadata_bytes(fs_info, block_rsv->space_info, |
5325 | orig_bytes, flush, system_chunk); | |
5d80366e JB |
5326 | if (ret == -ENOSPC && |
5327 | unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) { | |
5d80366e JB |
5328 | if (block_rsv != global_rsv && |
5329 | !block_rsv_use_bytes(global_rsv, orig_bytes)) | |
5330 | ret = 0; | |
5331 | } | |
cab45e22 | 5332 | if (ret == -ENOSPC) |
0b246afa | 5333 | trace_btrfs_space_reservation(fs_info, "space_info:enospc", |
957780eb JB |
5334 | block_rsv->space_info->flags, |
5335 | orig_bytes, 1); | |
f0486c68 YZ |
5336 | return ret; |
5337 | } | |
5338 | ||
79787eaa JM |
5339 | static struct btrfs_block_rsv *get_block_rsv( |
5340 | const struct btrfs_trans_handle *trans, | |
5341 | const struct btrfs_root *root) | |
f0486c68 | 5342 | { |
0b246afa | 5343 | struct btrfs_fs_info *fs_info = root->fs_info; |
4c13d758 JB |
5344 | struct btrfs_block_rsv *block_rsv = NULL; |
5345 | ||
e9cf439f | 5346 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) || |
0b246afa JM |
5347 | (root == fs_info->csum_root && trans->adding_csums) || |
5348 | (root == fs_info->uuid_root)) | |
f7a81ea4 SB |
5349 | block_rsv = trans->block_rsv; |
5350 | ||
4c13d758 | 5351 | if (!block_rsv) |
f0486c68 YZ |
5352 | block_rsv = root->block_rsv; |
5353 | ||
5354 | if (!block_rsv) | |
0b246afa | 5355 | block_rsv = &fs_info->empty_block_rsv; |
f0486c68 YZ |
5356 | |
5357 | return block_rsv; | |
5358 | } | |
5359 | ||
5360 | static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, | |
5361 | u64 num_bytes) | |
5362 | { | |
5363 | int ret = -ENOSPC; | |
5364 | spin_lock(&block_rsv->lock); | |
5365 | if (block_rsv->reserved >= num_bytes) { | |
5366 | block_rsv->reserved -= num_bytes; | |
5367 | if (block_rsv->reserved < block_rsv->size) | |
5368 | block_rsv->full = 0; | |
5369 | ret = 0; | |
5370 | } | |
5371 | spin_unlock(&block_rsv->lock); | |
5372 | return ret; | |
5373 | } | |
5374 | ||
5375 | static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv, | |
5376 | u64 num_bytes, int update_size) | |
5377 | { | |
5378 | spin_lock(&block_rsv->lock); | |
5379 | block_rsv->reserved += num_bytes; | |
5380 | if (update_size) | |
5381 | block_rsv->size += num_bytes; | |
5382 | else if (block_rsv->reserved >= block_rsv->size) | |
5383 | block_rsv->full = 1; | |
5384 | spin_unlock(&block_rsv->lock); | |
5385 | } | |
5386 | ||
d52be818 JB |
5387 | int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info, |
5388 | struct btrfs_block_rsv *dest, u64 num_bytes, | |
5389 | int min_factor) | |
5390 | { | |
5391 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
5392 | u64 min_bytes; | |
5393 | ||
5394 | if (global_rsv->space_info != dest->space_info) | |
5395 | return -ENOSPC; | |
5396 | ||
5397 | spin_lock(&global_rsv->lock); | |
5398 | min_bytes = div_factor(global_rsv->size, min_factor); | |
5399 | if (global_rsv->reserved < min_bytes + num_bytes) { | |
5400 | spin_unlock(&global_rsv->lock); | |
5401 | return -ENOSPC; | |
5402 | } | |
5403 | global_rsv->reserved -= num_bytes; | |
5404 | if (global_rsv->reserved < global_rsv->size) | |
5405 | global_rsv->full = 0; | |
5406 | spin_unlock(&global_rsv->lock); | |
5407 | ||
5408 | block_rsv_add_bytes(dest, num_bytes, 1); | |
5409 | return 0; | |
5410 | } | |
5411 | ||
957780eb JB |
5412 | /* |
5413 | * This is for space we already have accounted in space_info->bytes_may_use, so | |
5414 | * basically when we're returning space from block_rsv's. | |
5415 | */ | |
5416 | static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info, | |
5417 | struct btrfs_space_info *space_info, | |
5418 | u64 num_bytes) | |
5419 | { | |
5420 | struct reserve_ticket *ticket; | |
5421 | struct list_head *head; | |
5422 | u64 used; | |
5423 | enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_NO_FLUSH; | |
5424 | bool check_overcommit = false; | |
5425 | ||
5426 | spin_lock(&space_info->lock); | |
5427 | head = &space_info->priority_tickets; | |
5428 | ||
5429 | /* | |
5430 | * If we are over our limit then we need to check and see if we can | |
5431 | * overcommit, and if we can't then we just need to free up our space | |
5432 | * and not satisfy any requests. | |
5433 | */ | |
0eee8a49 | 5434 | used = btrfs_space_info_used(space_info, true); |
957780eb JB |
5435 | if (used - num_bytes >= space_info->total_bytes) |
5436 | check_overcommit = true; | |
5437 | again: | |
5438 | while (!list_empty(head) && num_bytes) { | |
5439 | ticket = list_first_entry(head, struct reserve_ticket, | |
5440 | list); | |
5441 | /* | |
5442 | * We use 0 bytes because this space is already reserved, so | |
5443 | * adding the ticket space would be a double count. | |
5444 | */ | |
5445 | if (check_overcommit && | |
c1c4919b | 5446 | !can_overcommit(fs_info, space_info, 0, flush, false)) |
957780eb JB |
5447 | break; |
5448 | if (num_bytes >= ticket->bytes) { | |
5449 | list_del_init(&ticket->list); | |
5450 | num_bytes -= ticket->bytes; | |
5451 | ticket->bytes = 0; | |
ce129655 | 5452 | space_info->tickets_id++; |
957780eb JB |
5453 | wake_up(&ticket->wait); |
5454 | } else { | |
5455 | ticket->bytes -= num_bytes; | |
5456 | num_bytes = 0; | |
5457 | } | |
5458 | } | |
5459 | ||
5460 | if (num_bytes && head == &space_info->priority_tickets) { | |
5461 | head = &space_info->tickets; | |
5462 | flush = BTRFS_RESERVE_FLUSH_ALL; | |
5463 | goto again; | |
5464 | } | |
5465 | space_info->bytes_may_use -= num_bytes; | |
5466 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5467 | space_info->flags, num_bytes, 0); | |
5468 | spin_unlock(&space_info->lock); | |
5469 | } | |
5470 | ||
5471 | /* | |
5472 | * This is for newly allocated space that isn't accounted in | |
5473 | * space_info->bytes_may_use yet. So if we allocate a chunk or unpin an extent | |
5474 | * we use this helper. | |
5475 | */ | |
5476 | static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info, | |
5477 | struct btrfs_space_info *space_info, | |
5478 | u64 num_bytes) | |
5479 | { | |
5480 | struct reserve_ticket *ticket; | |
5481 | struct list_head *head = &space_info->priority_tickets; | |
5482 | ||
5483 | again: | |
5484 | while (!list_empty(head) && num_bytes) { | |
5485 | ticket = list_first_entry(head, struct reserve_ticket, | |
5486 | list); | |
5487 | if (num_bytes >= ticket->bytes) { | |
5488 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5489 | space_info->flags, | |
5490 | ticket->bytes, 1); | |
5491 | list_del_init(&ticket->list); | |
5492 | num_bytes -= ticket->bytes; | |
5493 | space_info->bytes_may_use += ticket->bytes; | |
5494 | ticket->bytes = 0; | |
ce129655 | 5495 | space_info->tickets_id++; |
957780eb JB |
5496 | wake_up(&ticket->wait); |
5497 | } else { | |
5498 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5499 | space_info->flags, | |
5500 | num_bytes, 1); | |
5501 | space_info->bytes_may_use += num_bytes; | |
5502 | ticket->bytes -= num_bytes; | |
5503 | num_bytes = 0; | |
5504 | } | |
5505 | } | |
5506 | ||
5507 | if (num_bytes && head == &space_info->priority_tickets) { | |
5508 | head = &space_info->tickets; | |
5509 | goto again; | |
5510 | } | |
5511 | } | |
5512 | ||
8c2a3ca2 JB |
5513 | static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info, |
5514 | struct btrfs_block_rsv *block_rsv, | |
62a45b60 | 5515 | struct btrfs_block_rsv *dest, u64 num_bytes) |
f0486c68 YZ |
5516 | { |
5517 | struct btrfs_space_info *space_info = block_rsv->space_info; | |
5518 | ||
5519 | spin_lock(&block_rsv->lock); | |
5520 | if (num_bytes == (u64)-1) | |
5521 | num_bytes = block_rsv->size; | |
5522 | block_rsv->size -= num_bytes; | |
5523 | if (block_rsv->reserved >= block_rsv->size) { | |
5524 | num_bytes = block_rsv->reserved - block_rsv->size; | |
5525 | block_rsv->reserved = block_rsv->size; | |
5526 | block_rsv->full = 1; | |
5527 | } else { | |
5528 | num_bytes = 0; | |
5529 | } | |
5530 | spin_unlock(&block_rsv->lock); | |
5531 | ||
5532 | if (num_bytes > 0) { | |
5533 | if (dest) { | |
e9e22899 JB |
5534 | spin_lock(&dest->lock); |
5535 | if (!dest->full) { | |
5536 | u64 bytes_to_add; | |
5537 | ||
5538 | bytes_to_add = dest->size - dest->reserved; | |
5539 | bytes_to_add = min(num_bytes, bytes_to_add); | |
5540 | dest->reserved += bytes_to_add; | |
5541 | if (dest->reserved >= dest->size) | |
5542 | dest->full = 1; | |
5543 | num_bytes -= bytes_to_add; | |
5544 | } | |
5545 | spin_unlock(&dest->lock); | |
5546 | } | |
957780eb JB |
5547 | if (num_bytes) |
5548 | space_info_add_old_bytes(fs_info, space_info, | |
5549 | num_bytes); | |
9ed74f2d | 5550 | } |
f0486c68 | 5551 | } |
4e06bdd6 | 5552 | |
25d609f8 JB |
5553 | int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src, |
5554 | struct btrfs_block_rsv *dst, u64 num_bytes, | |
5555 | int update_size) | |
f0486c68 YZ |
5556 | { |
5557 | int ret; | |
9ed74f2d | 5558 | |
f0486c68 YZ |
5559 | ret = block_rsv_use_bytes(src, num_bytes); |
5560 | if (ret) | |
5561 | return ret; | |
9ed74f2d | 5562 | |
25d609f8 | 5563 | block_rsv_add_bytes(dst, num_bytes, update_size); |
9ed74f2d JB |
5564 | return 0; |
5565 | } | |
5566 | ||
66d8f3dd | 5567 | void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type) |
9ed74f2d | 5568 | { |
f0486c68 YZ |
5569 | memset(rsv, 0, sizeof(*rsv)); |
5570 | spin_lock_init(&rsv->lock); | |
66d8f3dd | 5571 | rsv->type = type; |
f0486c68 YZ |
5572 | } |
5573 | ||
2ff7e61e | 5574 | struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_fs_info *fs_info, |
66d8f3dd | 5575 | unsigned short type) |
f0486c68 YZ |
5576 | { |
5577 | struct btrfs_block_rsv *block_rsv; | |
9ed74f2d | 5578 | |
f0486c68 YZ |
5579 | block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS); |
5580 | if (!block_rsv) | |
5581 | return NULL; | |
9ed74f2d | 5582 | |
66d8f3dd | 5583 | btrfs_init_block_rsv(block_rsv, type); |
f0486c68 YZ |
5584 | block_rsv->space_info = __find_space_info(fs_info, |
5585 | BTRFS_BLOCK_GROUP_METADATA); | |
f0486c68 YZ |
5586 | return block_rsv; |
5587 | } | |
9ed74f2d | 5588 | |
2ff7e61e | 5589 | void btrfs_free_block_rsv(struct btrfs_fs_info *fs_info, |
f0486c68 YZ |
5590 | struct btrfs_block_rsv *rsv) |
5591 | { | |
2aaa6655 JB |
5592 | if (!rsv) |
5593 | return; | |
2ff7e61e | 5594 | btrfs_block_rsv_release(fs_info, rsv, (u64)-1); |
dabdb640 | 5595 | kfree(rsv); |
9ed74f2d JB |
5596 | } |
5597 | ||
cdfb080e CM |
5598 | void __btrfs_free_block_rsv(struct btrfs_block_rsv *rsv) |
5599 | { | |
5600 | kfree(rsv); | |
5601 | } | |
5602 | ||
08e007d2 MX |
5603 | int btrfs_block_rsv_add(struct btrfs_root *root, |
5604 | struct btrfs_block_rsv *block_rsv, u64 num_bytes, | |
5605 | enum btrfs_reserve_flush_enum flush) | |
9ed74f2d | 5606 | { |
f0486c68 | 5607 | int ret; |
9ed74f2d | 5608 | |
f0486c68 YZ |
5609 | if (num_bytes == 0) |
5610 | return 0; | |
8bb8ab2e | 5611 | |
61b520a9 | 5612 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
f0486c68 YZ |
5613 | if (!ret) { |
5614 | block_rsv_add_bytes(block_rsv, num_bytes, 1); | |
5615 | return 0; | |
5616 | } | |
9ed74f2d | 5617 | |
f0486c68 | 5618 | return ret; |
f0486c68 | 5619 | } |
9ed74f2d | 5620 | |
2ff7e61e | 5621 | int btrfs_block_rsv_check(struct btrfs_block_rsv *block_rsv, int min_factor) |
f0486c68 YZ |
5622 | { |
5623 | u64 num_bytes = 0; | |
f0486c68 | 5624 | int ret = -ENOSPC; |
9ed74f2d | 5625 | |
f0486c68 YZ |
5626 | if (!block_rsv) |
5627 | return 0; | |
9ed74f2d | 5628 | |
f0486c68 | 5629 | spin_lock(&block_rsv->lock); |
36ba022a JB |
5630 | num_bytes = div_factor(block_rsv->size, min_factor); |
5631 | if (block_rsv->reserved >= num_bytes) | |
5632 | ret = 0; | |
5633 | spin_unlock(&block_rsv->lock); | |
9ed74f2d | 5634 | |
36ba022a JB |
5635 | return ret; |
5636 | } | |
5637 | ||
08e007d2 MX |
5638 | int btrfs_block_rsv_refill(struct btrfs_root *root, |
5639 | struct btrfs_block_rsv *block_rsv, u64 min_reserved, | |
5640 | enum btrfs_reserve_flush_enum flush) | |
36ba022a JB |
5641 | { |
5642 | u64 num_bytes = 0; | |
5643 | int ret = -ENOSPC; | |
5644 | ||
5645 | if (!block_rsv) | |
5646 | return 0; | |
5647 | ||
5648 | spin_lock(&block_rsv->lock); | |
5649 | num_bytes = min_reserved; | |
13553e52 | 5650 | if (block_rsv->reserved >= num_bytes) |
f0486c68 | 5651 | ret = 0; |
13553e52 | 5652 | else |
f0486c68 | 5653 | num_bytes -= block_rsv->reserved; |
f0486c68 | 5654 | spin_unlock(&block_rsv->lock); |
13553e52 | 5655 | |
f0486c68 YZ |
5656 | if (!ret) |
5657 | return 0; | |
5658 | ||
aa38a711 | 5659 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
dabdb640 JB |
5660 | if (!ret) { |
5661 | block_rsv_add_bytes(block_rsv, num_bytes, 0); | |
f0486c68 | 5662 | return 0; |
6a63209f | 5663 | } |
9ed74f2d | 5664 | |
13553e52 | 5665 | return ret; |
f0486c68 YZ |
5666 | } |
5667 | ||
2ff7e61e | 5668 | void btrfs_block_rsv_release(struct btrfs_fs_info *fs_info, |
f0486c68 YZ |
5669 | struct btrfs_block_rsv *block_rsv, |
5670 | u64 num_bytes) | |
5671 | { | |
0b246afa JM |
5672 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; |
5673 | ||
17504584 | 5674 | if (global_rsv == block_rsv || |
f0486c68 YZ |
5675 | block_rsv->space_info != global_rsv->space_info) |
5676 | global_rsv = NULL; | |
0b246afa | 5677 | block_rsv_release_bytes(fs_info, block_rsv, global_rsv, num_bytes); |
6a63209f JB |
5678 | } |
5679 | ||
8929ecfa YZ |
5680 | static void update_global_block_rsv(struct btrfs_fs_info *fs_info) |
5681 | { | |
5682 | struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv; | |
5683 | struct btrfs_space_info *sinfo = block_rsv->space_info; | |
5684 | u64 num_bytes; | |
6a63209f | 5685 | |
ae2e4728 JB |
5686 | /* |
5687 | * The global block rsv is based on the size of the extent tree, the | |
5688 | * checksum tree and the root tree. If the fs is empty we want to set | |
5689 | * it to a minimal amount for safety. | |
5690 | */ | |
5691 | num_bytes = btrfs_root_used(&fs_info->extent_root->root_item) + | |
5692 | btrfs_root_used(&fs_info->csum_root->root_item) + | |
5693 | btrfs_root_used(&fs_info->tree_root->root_item); | |
5694 | num_bytes = max_t(u64, num_bytes, SZ_16M); | |
33b4d47f | 5695 | |
8929ecfa | 5696 | spin_lock(&sinfo->lock); |
1f699d38 | 5697 | spin_lock(&block_rsv->lock); |
4e06bdd6 | 5698 | |
ee22184b | 5699 | block_rsv->size = min_t(u64, num_bytes, SZ_512M); |
4e06bdd6 | 5700 | |
fb4b10e5 | 5701 | if (block_rsv->reserved < block_rsv->size) { |
4136135b | 5702 | num_bytes = btrfs_space_info_used(sinfo, true); |
fb4b10e5 JB |
5703 | if (sinfo->total_bytes > num_bytes) { |
5704 | num_bytes = sinfo->total_bytes - num_bytes; | |
5705 | num_bytes = min(num_bytes, | |
5706 | block_rsv->size - block_rsv->reserved); | |
5707 | block_rsv->reserved += num_bytes; | |
5708 | sinfo->bytes_may_use += num_bytes; | |
5709 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5710 | sinfo->flags, num_bytes, | |
5711 | 1); | |
5712 | } | |
5713 | } else if (block_rsv->reserved > block_rsv->size) { | |
8929ecfa | 5714 | num_bytes = block_rsv->reserved - block_rsv->size; |
fb25e914 | 5715 | sinfo->bytes_may_use -= num_bytes; |
8c2a3ca2 | 5716 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 5717 | sinfo->flags, num_bytes, 0); |
8929ecfa | 5718 | block_rsv->reserved = block_rsv->size; |
8929ecfa | 5719 | } |
182608c8 | 5720 | |
fb4b10e5 JB |
5721 | if (block_rsv->reserved == block_rsv->size) |
5722 | block_rsv->full = 1; | |
5723 | else | |
5724 | block_rsv->full = 0; | |
5725 | ||
8929ecfa | 5726 | spin_unlock(&block_rsv->lock); |
1f699d38 | 5727 | spin_unlock(&sinfo->lock); |
6a63209f JB |
5728 | } |
5729 | ||
f0486c68 | 5730 | static void init_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 5731 | { |
f0486c68 | 5732 | struct btrfs_space_info *space_info; |
6a63209f | 5733 | |
f0486c68 YZ |
5734 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); |
5735 | fs_info->chunk_block_rsv.space_info = space_info; | |
6a63209f | 5736 | |
f0486c68 | 5737 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
8929ecfa | 5738 | fs_info->global_block_rsv.space_info = space_info; |
8929ecfa | 5739 | fs_info->delalloc_block_rsv.space_info = space_info; |
f0486c68 YZ |
5740 | fs_info->trans_block_rsv.space_info = space_info; |
5741 | fs_info->empty_block_rsv.space_info = space_info; | |
6d668dda | 5742 | fs_info->delayed_block_rsv.space_info = space_info; |
f0486c68 | 5743 | |
8929ecfa YZ |
5744 | fs_info->extent_root->block_rsv = &fs_info->global_block_rsv; |
5745 | fs_info->csum_root->block_rsv = &fs_info->global_block_rsv; | |
5746 | fs_info->dev_root->block_rsv = &fs_info->global_block_rsv; | |
5747 | fs_info->tree_root->block_rsv = &fs_info->global_block_rsv; | |
3a6cad90 SB |
5748 | if (fs_info->quota_root) |
5749 | fs_info->quota_root->block_rsv = &fs_info->global_block_rsv; | |
f0486c68 | 5750 | fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv; |
8929ecfa | 5751 | |
8929ecfa | 5752 | update_global_block_rsv(fs_info); |
6a63209f JB |
5753 | } |
5754 | ||
8929ecfa | 5755 | static void release_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 5756 | { |
8c2a3ca2 JB |
5757 | block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL, |
5758 | (u64)-1); | |
8929ecfa YZ |
5759 | WARN_ON(fs_info->delalloc_block_rsv.size > 0); |
5760 | WARN_ON(fs_info->delalloc_block_rsv.reserved > 0); | |
5761 | WARN_ON(fs_info->trans_block_rsv.size > 0); | |
5762 | WARN_ON(fs_info->trans_block_rsv.reserved > 0); | |
5763 | WARN_ON(fs_info->chunk_block_rsv.size > 0); | |
5764 | WARN_ON(fs_info->chunk_block_rsv.reserved > 0); | |
6d668dda JB |
5765 | WARN_ON(fs_info->delayed_block_rsv.size > 0); |
5766 | WARN_ON(fs_info->delayed_block_rsv.reserved > 0); | |
fcb80c2a JB |
5767 | } |
5768 | ||
a22285a6 | 5769 | void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans, |
2ff7e61e | 5770 | struct btrfs_fs_info *fs_info) |
6a63209f | 5771 | { |
0e721106 JB |
5772 | if (!trans->block_rsv) |
5773 | return; | |
5774 | ||
a22285a6 YZ |
5775 | if (!trans->bytes_reserved) |
5776 | return; | |
6a63209f | 5777 | |
0b246afa | 5778 | trace_btrfs_space_reservation(fs_info, "transaction", |
2bcc0328 | 5779 | trans->transid, trans->bytes_reserved, 0); |
2ff7e61e JM |
5780 | btrfs_block_rsv_release(fs_info, trans->block_rsv, |
5781 | trans->bytes_reserved); | |
a22285a6 YZ |
5782 | trans->bytes_reserved = 0; |
5783 | } | |
6a63209f | 5784 | |
4fbcdf66 FM |
5785 | /* |
5786 | * To be called after all the new block groups attached to the transaction | |
5787 | * handle have been created (btrfs_create_pending_block_groups()). | |
5788 | */ | |
5789 | void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans) | |
5790 | { | |
64b63580 | 5791 | struct btrfs_fs_info *fs_info = trans->fs_info; |
4fbcdf66 FM |
5792 | |
5793 | if (!trans->chunk_bytes_reserved) | |
5794 | return; | |
5795 | ||
5796 | WARN_ON_ONCE(!list_empty(&trans->new_bgs)); | |
5797 | ||
5798 | block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL, | |
5799 | trans->chunk_bytes_reserved); | |
5800 | trans->chunk_bytes_reserved = 0; | |
5801 | } | |
5802 | ||
79787eaa | 5803 | /* Can only return 0 or -ENOSPC */ |
d68fc57b | 5804 | int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans, |
8ed7a2a0 | 5805 | struct btrfs_inode *inode) |
d68fc57b | 5806 | { |
8ed7a2a0 NB |
5807 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
5808 | struct btrfs_root *root = inode->root; | |
40acc3ee JB |
5809 | /* |
5810 | * We always use trans->block_rsv here as we will have reserved space | |
5811 | * for our orphan when starting the transaction, using get_block_rsv() | |
5812 | * here will sometimes make us choose the wrong block rsv as we could be | |
5813 | * doing a reloc inode for a non refcounted root. | |
5814 | */ | |
5815 | struct btrfs_block_rsv *src_rsv = trans->block_rsv; | |
d68fc57b YZ |
5816 | struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv; |
5817 | ||
5818 | /* | |
fcb80c2a JB |
5819 | * We need to hold space in order to delete our orphan item once we've |
5820 | * added it, so this takes the reservation so we can release it later | |
5821 | * when we are truly done with the orphan item. | |
d68fc57b | 5822 | */ |
0b246afa JM |
5823 | u64 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1); |
5824 | ||
8ed7a2a0 NB |
5825 | trace_btrfs_space_reservation(fs_info, "orphan", btrfs_ino(inode), |
5826 | num_bytes, 1); | |
25d609f8 | 5827 | return btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, 1); |
6a63209f JB |
5828 | } |
5829 | ||
703b391a | 5830 | void btrfs_orphan_release_metadata(struct btrfs_inode *inode) |
97e728d4 | 5831 | { |
703b391a NB |
5832 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
5833 | struct btrfs_root *root = inode->root; | |
0b246afa JM |
5834 | u64 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1); |
5835 | ||
703b391a NB |
5836 | trace_btrfs_space_reservation(fs_info, "orphan", btrfs_ino(inode), |
5837 | num_bytes, 0); | |
2ff7e61e | 5838 | btrfs_block_rsv_release(fs_info, root->orphan_block_rsv, num_bytes); |
d68fc57b | 5839 | } |
97e728d4 | 5840 | |
d5c12070 MX |
5841 | /* |
5842 | * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation | |
5843 | * root: the root of the parent directory | |
5844 | * rsv: block reservation | |
5845 | * items: the number of items that we need do reservation | |
5846 | * qgroup_reserved: used to return the reserved size in qgroup | |
5847 | * | |
5848 | * This function is used to reserve the space for snapshot/subvolume | |
5849 | * creation and deletion. Those operations are different with the | |
5850 | * common file/directory operations, they change two fs/file trees | |
5851 | * and root tree, the number of items that the qgroup reserves is | |
5852 | * different with the free space reservation. So we can not use | |
01327610 | 5853 | * the space reservation mechanism in start_transaction(). |
d5c12070 MX |
5854 | */ |
5855 | int btrfs_subvolume_reserve_metadata(struct btrfs_root *root, | |
5856 | struct btrfs_block_rsv *rsv, | |
5857 | int items, | |
ee3441b4 JM |
5858 | u64 *qgroup_reserved, |
5859 | bool use_global_rsv) | |
a22285a6 | 5860 | { |
d5c12070 MX |
5861 | u64 num_bytes; |
5862 | int ret; | |
0b246afa JM |
5863 | struct btrfs_fs_info *fs_info = root->fs_info; |
5864 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
d5c12070 | 5865 | |
0b246afa | 5866 | if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) { |
d5c12070 | 5867 | /* One for parent inode, two for dir entries */ |
0b246afa | 5868 | num_bytes = 3 * fs_info->nodesize; |
003d7c59 | 5869 | ret = btrfs_qgroup_reserve_meta(root, num_bytes, true); |
d5c12070 MX |
5870 | if (ret) |
5871 | return ret; | |
5872 | } else { | |
5873 | num_bytes = 0; | |
5874 | } | |
5875 | ||
5876 | *qgroup_reserved = num_bytes; | |
5877 | ||
0b246afa JM |
5878 | num_bytes = btrfs_calc_trans_metadata_size(fs_info, items); |
5879 | rsv->space_info = __find_space_info(fs_info, | |
d5c12070 MX |
5880 | BTRFS_BLOCK_GROUP_METADATA); |
5881 | ret = btrfs_block_rsv_add(root, rsv, num_bytes, | |
5882 | BTRFS_RESERVE_FLUSH_ALL); | |
ee3441b4 JM |
5883 | |
5884 | if (ret == -ENOSPC && use_global_rsv) | |
25d609f8 | 5885 | ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes, 1); |
ee3441b4 | 5886 | |
7174109c QW |
5887 | if (ret && *qgroup_reserved) |
5888 | btrfs_qgroup_free_meta(root, *qgroup_reserved); | |
d5c12070 MX |
5889 | |
5890 | return ret; | |
5891 | } | |
5892 | ||
2ff7e61e | 5893 | void btrfs_subvolume_release_metadata(struct btrfs_fs_info *fs_info, |
7775c818 | 5894 | struct btrfs_block_rsv *rsv) |
d5c12070 | 5895 | { |
2ff7e61e | 5896 | btrfs_block_rsv_release(fs_info, rsv, (u64)-1); |
97e728d4 JB |
5897 | } |
5898 | ||
7709cde3 JB |
5899 | /** |
5900 | * drop_outstanding_extent - drop an outstanding extent | |
5901 | * @inode: the inode we're dropping the extent for | |
01327610 | 5902 | * @num_bytes: the number of bytes we're releasing. |
7709cde3 JB |
5903 | * |
5904 | * This is called when we are freeing up an outstanding extent, either called | |
5905 | * after an error or after an extent is written. This will return the number of | |
5906 | * reserved extents that need to be freed. This must be called with | |
5907 | * BTRFS_I(inode)->lock held. | |
5908 | */ | |
baa3ba39 NB |
5909 | static unsigned drop_outstanding_extent(struct btrfs_inode *inode, |
5910 | u64 num_bytes) | |
9e0baf60 | 5911 | { |
7fd2ae21 | 5912 | unsigned drop_inode_space = 0; |
9e0baf60 | 5913 | unsigned dropped_extents = 0; |
823bb20a | 5914 | unsigned num_extents; |
9e0baf60 | 5915 | |
823bb20a | 5916 | num_extents = count_max_extents(num_bytes); |
dcab6a3b | 5917 | ASSERT(num_extents); |
baa3ba39 NB |
5918 | ASSERT(inode->outstanding_extents >= num_extents); |
5919 | inode->outstanding_extents -= num_extents; | |
9e0baf60 | 5920 | |
baa3ba39 | 5921 | if (inode->outstanding_extents == 0 && |
72ac3c0d | 5922 | test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED, |
baa3ba39 | 5923 | &inode->runtime_flags)) |
7fd2ae21 | 5924 | drop_inode_space = 1; |
7fd2ae21 | 5925 | |
9e0baf60 | 5926 | /* |
01327610 | 5927 | * If we have more or the same amount of outstanding extents than we have |
9e0baf60 JB |
5928 | * reserved then we need to leave the reserved extents count alone. |
5929 | */ | |
baa3ba39 | 5930 | if (inode->outstanding_extents >= inode->reserved_extents) |
7fd2ae21 | 5931 | return drop_inode_space; |
9e0baf60 | 5932 | |
baa3ba39 NB |
5933 | dropped_extents = inode->reserved_extents - inode->outstanding_extents; |
5934 | inode->reserved_extents -= dropped_extents; | |
7fd2ae21 | 5935 | return dropped_extents + drop_inode_space; |
9e0baf60 JB |
5936 | } |
5937 | ||
7709cde3 | 5938 | /** |
01327610 NS |
5939 | * calc_csum_metadata_size - return the amount of metadata space that must be |
5940 | * reserved/freed for the given bytes. | |
7709cde3 JB |
5941 | * @inode: the inode we're manipulating |
5942 | * @num_bytes: the number of bytes in question | |
5943 | * @reserve: 1 if we are reserving space, 0 if we are freeing space | |
5944 | * | |
5945 | * This adjusts the number of csum_bytes in the inode and then returns the | |
5946 | * correct amount of metadata that must either be reserved or freed. We | |
5947 | * calculate how many checksums we can fit into one leaf and then divide the | |
5948 | * number of bytes that will need to be checksumed by this value to figure out | |
5949 | * how many checksums will be required. If we are adding bytes then the number | |
5950 | * may go up and we will return the number of additional bytes that must be | |
5951 | * reserved. If it is going down we will return the number of bytes that must | |
5952 | * be freed. | |
5953 | * | |
5954 | * This must be called with BTRFS_I(inode)->lock held. | |
5955 | */ | |
0e6bf9b1 | 5956 | static u64 calc_csum_metadata_size(struct btrfs_inode *inode, u64 num_bytes, |
7709cde3 | 5957 | int reserve) |
6324fbf3 | 5958 | { |
0e6bf9b1 | 5959 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
1262133b | 5960 | u64 old_csums, num_csums; |
7709cde3 | 5961 | |
0e6bf9b1 | 5962 | if (inode->flags & BTRFS_INODE_NODATASUM && inode->csum_bytes == 0) |
7709cde3 JB |
5963 | return 0; |
5964 | ||
0e6bf9b1 | 5965 | old_csums = btrfs_csum_bytes_to_leaves(fs_info, inode->csum_bytes); |
7709cde3 | 5966 | if (reserve) |
0e6bf9b1 | 5967 | inode->csum_bytes += num_bytes; |
7709cde3 | 5968 | else |
0e6bf9b1 NB |
5969 | inode->csum_bytes -= num_bytes; |
5970 | num_csums = btrfs_csum_bytes_to_leaves(fs_info, inode->csum_bytes); | |
7709cde3 JB |
5971 | |
5972 | /* No change, no need to reserve more */ | |
5973 | if (old_csums == num_csums) | |
5974 | return 0; | |
5975 | ||
5976 | if (reserve) | |
0b246afa | 5977 | return btrfs_calc_trans_metadata_size(fs_info, |
7709cde3 JB |
5978 | num_csums - old_csums); |
5979 | ||
0b246afa | 5980 | return btrfs_calc_trans_metadata_size(fs_info, old_csums - num_csums); |
0ca1f7ce | 5981 | } |
c146afad | 5982 | |
9f3db423 | 5983 | int btrfs_delalloc_reserve_metadata(struct btrfs_inode *inode, u64 num_bytes) |
0ca1f7ce | 5984 | { |
9f3db423 NB |
5985 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
5986 | struct btrfs_root *root = inode->root; | |
0b246afa | 5987 | struct btrfs_block_rsv *block_rsv = &fs_info->delalloc_block_rsv; |
9e0baf60 | 5988 | u64 to_reserve = 0; |
660d3f6c | 5989 | u64 csum_bytes; |
823bb20a | 5990 | unsigned nr_extents; |
08e007d2 | 5991 | enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL; |
eb6b88d9 | 5992 | int ret = 0; |
c64c2bd8 | 5993 | bool delalloc_lock = true; |
88e081bf WS |
5994 | u64 to_free = 0; |
5995 | unsigned dropped; | |
48c3d480 | 5996 | bool release_extra = false; |
6324fbf3 | 5997 | |
c64c2bd8 JB |
5998 | /* If we are a free space inode we need to not flush since we will be in |
5999 | * the middle of a transaction commit. We also don't need the delalloc | |
6000 | * mutex since we won't race with anybody. We need this mostly to make | |
6001 | * lockdep shut its filthy mouth. | |
bac357dc JB |
6002 | * |
6003 | * If we have a transaction open (can happen if we call truncate_block | |
6004 | * from truncate), then we need FLUSH_LIMIT so we don't deadlock. | |
c64c2bd8 JB |
6005 | */ |
6006 | if (btrfs_is_free_space_inode(inode)) { | |
08e007d2 | 6007 | flush = BTRFS_RESERVE_NO_FLUSH; |
c64c2bd8 | 6008 | delalloc_lock = false; |
bac357dc JB |
6009 | } else if (current->journal_info) { |
6010 | flush = BTRFS_RESERVE_FLUSH_LIMIT; | |
c64c2bd8 | 6011 | } |
c09544e0 | 6012 | |
08e007d2 | 6013 | if (flush != BTRFS_RESERVE_NO_FLUSH && |
0b246afa | 6014 | btrfs_transaction_in_commit(fs_info)) |
0ca1f7ce | 6015 | schedule_timeout(1); |
ec44a35c | 6016 | |
c64c2bd8 | 6017 | if (delalloc_lock) |
9f3db423 | 6018 | mutex_lock(&inode->delalloc_mutex); |
c64c2bd8 | 6019 | |
0b246afa | 6020 | num_bytes = ALIGN(num_bytes, fs_info->sectorsize); |
8bb8ab2e | 6021 | |
9f3db423 | 6022 | spin_lock(&inode->lock); |
823bb20a | 6023 | nr_extents = count_max_extents(num_bytes); |
9f3db423 | 6024 | inode->outstanding_extents += nr_extents; |
9e0baf60 | 6025 | |
48c3d480 | 6026 | nr_extents = 0; |
9f3db423 NB |
6027 | if (inode->outstanding_extents > inode->reserved_extents) |
6028 | nr_extents += inode->outstanding_extents - | |
6029 | inode->reserved_extents; | |
57a45ced | 6030 | |
48c3d480 | 6031 | /* We always want to reserve a slot for updating the inode. */ |
0b246afa | 6032 | to_reserve = btrfs_calc_trans_metadata_size(fs_info, nr_extents + 1); |
7709cde3 | 6033 | to_reserve += calc_csum_metadata_size(inode, num_bytes, 1); |
9f3db423 NB |
6034 | csum_bytes = inode->csum_bytes; |
6035 | spin_unlock(&inode->lock); | |
57a45ced | 6036 | |
0b246afa | 6037 | if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) { |
7174109c | 6038 | ret = btrfs_qgroup_reserve_meta(root, |
003d7c59 | 6039 | nr_extents * fs_info->nodesize, true); |
88e081bf WS |
6040 | if (ret) |
6041 | goto out_fail; | |
6042 | } | |
c5567237 | 6043 | |
48c3d480 | 6044 | ret = btrfs_block_rsv_add(root, block_rsv, to_reserve, flush); |
88e081bf | 6045 | if (unlikely(ret)) { |
da17066c | 6046 | btrfs_qgroup_free_meta(root, |
0b246afa | 6047 | nr_extents * fs_info->nodesize); |
88e081bf | 6048 | goto out_fail; |
9e0baf60 | 6049 | } |
25179201 | 6050 | |
9f3db423 | 6051 | spin_lock(&inode->lock); |
48c3d480 | 6052 | if (test_and_set_bit(BTRFS_INODE_DELALLOC_META_RESERVED, |
9f3db423 | 6053 | &inode->runtime_flags)) { |
0b246afa | 6054 | to_reserve -= btrfs_calc_trans_metadata_size(fs_info, 1); |
48c3d480 | 6055 | release_extra = true; |
660d3f6c | 6056 | } |
9f3db423 NB |
6057 | inode->reserved_extents += nr_extents; |
6058 | spin_unlock(&inode->lock); | |
c64c2bd8 JB |
6059 | |
6060 | if (delalloc_lock) | |
9f3db423 | 6061 | mutex_unlock(&inode->delalloc_mutex); |
660d3f6c | 6062 | |
8c2a3ca2 | 6063 | if (to_reserve) |
0b246afa | 6064 | trace_btrfs_space_reservation(fs_info, "delalloc", |
9f3db423 | 6065 | btrfs_ino(inode), to_reserve, 1); |
48c3d480 | 6066 | if (release_extra) |
2ff7e61e | 6067 | btrfs_block_rsv_release(fs_info, block_rsv, |
0b246afa | 6068 | btrfs_calc_trans_metadata_size(fs_info, 1)); |
0ca1f7ce | 6069 | return 0; |
88e081bf WS |
6070 | |
6071 | out_fail: | |
9f3db423 | 6072 | spin_lock(&inode->lock); |
dcab6a3b | 6073 | dropped = drop_outstanding_extent(inode, num_bytes); |
88e081bf WS |
6074 | /* |
6075 | * If the inodes csum_bytes is the same as the original | |
6076 | * csum_bytes then we know we haven't raced with any free()ers | |
6077 | * so we can just reduce our inodes csum bytes and carry on. | |
88e081bf | 6078 | */ |
9f3db423 | 6079 | if (inode->csum_bytes == csum_bytes) { |
88e081bf | 6080 | calc_csum_metadata_size(inode, num_bytes, 0); |
f4881bc7 | 6081 | } else { |
9f3db423 | 6082 | u64 orig_csum_bytes = inode->csum_bytes; |
f4881bc7 JB |
6083 | u64 bytes; |
6084 | ||
6085 | /* | |
6086 | * This is tricky, but first we need to figure out how much we | |
01327610 | 6087 | * freed from any free-ers that occurred during this |
f4881bc7 JB |
6088 | * reservation, so we reset ->csum_bytes to the csum_bytes |
6089 | * before we dropped our lock, and then call the free for the | |
6090 | * number of bytes that were freed while we were trying our | |
6091 | * reservation. | |
6092 | */ | |
9f3db423 NB |
6093 | bytes = csum_bytes - inode->csum_bytes; |
6094 | inode->csum_bytes = csum_bytes; | |
f4881bc7 JB |
6095 | to_free = calc_csum_metadata_size(inode, bytes, 0); |
6096 | ||
6097 | ||
6098 | /* | |
6099 | * Now we need to see how much we would have freed had we not | |
6100 | * been making this reservation and our ->csum_bytes were not | |
6101 | * artificially inflated. | |
6102 | */ | |
9f3db423 | 6103 | inode->csum_bytes = csum_bytes - num_bytes; |
f4881bc7 JB |
6104 | bytes = csum_bytes - orig_csum_bytes; |
6105 | bytes = calc_csum_metadata_size(inode, bytes, 0); | |
6106 | ||
6107 | /* | |
6108 | * Now reset ->csum_bytes to what it should be. If bytes is | |
01327610 | 6109 | * more than to_free then we would have freed more space had we |
f4881bc7 JB |
6110 | * not had an artificially high ->csum_bytes, so we need to free |
6111 | * the remainder. If bytes is the same or less then we don't | |
6112 | * need to do anything, the other free-ers did the correct | |
6113 | * thing. | |
6114 | */ | |
9f3db423 | 6115 | inode->csum_bytes = orig_csum_bytes - num_bytes; |
f4881bc7 JB |
6116 | if (bytes > to_free) |
6117 | to_free = bytes - to_free; | |
6118 | else | |
6119 | to_free = 0; | |
6120 | } | |
9f3db423 | 6121 | spin_unlock(&inode->lock); |
e2d1f923 | 6122 | if (dropped) |
0b246afa | 6123 | to_free += btrfs_calc_trans_metadata_size(fs_info, dropped); |
88e081bf WS |
6124 | |
6125 | if (to_free) { | |
2ff7e61e | 6126 | btrfs_block_rsv_release(fs_info, block_rsv, to_free); |
0b246afa | 6127 | trace_btrfs_space_reservation(fs_info, "delalloc", |
9f3db423 | 6128 | btrfs_ino(inode), to_free, 0); |
88e081bf WS |
6129 | } |
6130 | if (delalloc_lock) | |
9f3db423 | 6131 | mutex_unlock(&inode->delalloc_mutex); |
88e081bf | 6132 | return ret; |
0ca1f7ce YZ |
6133 | } |
6134 | ||
7709cde3 JB |
6135 | /** |
6136 | * btrfs_delalloc_release_metadata - release a metadata reservation for an inode | |
6137 | * @inode: the inode to release the reservation for | |
6138 | * @num_bytes: the number of bytes we're releasing | |
6139 | * | |
6140 | * This will release the metadata reservation for an inode. This can be called | |
6141 | * once we complete IO for a given set of bytes to release their metadata | |
6142 | * reservations. | |
6143 | */ | |
691fa059 | 6144 | void btrfs_delalloc_release_metadata(struct btrfs_inode *inode, u64 num_bytes) |
0ca1f7ce | 6145 | { |
691fa059 | 6146 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
9e0baf60 JB |
6147 | u64 to_free = 0; |
6148 | unsigned dropped; | |
0ca1f7ce | 6149 | |
0b246afa | 6150 | num_bytes = ALIGN(num_bytes, fs_info->sectorsize); |
691fa059 | 6151 | spin_lock(&inode->lock); |
dcab6a3b | 6152 | dropped = drop_outstanding_extent(inode, num_bytes); |
97e728d4 | 6153 | |
0934856d MX |
6154 | if (num_bytes) |
6155 | to_free = calc_csum_metadata_size(inode, num_bytes, 0); | |
691fa059 | 6156 | spin_unlock(&inode->lock); |
9e0baf60 | 6157 | if (dropped > 0) |
0b246afa | 6158 | to_free += btrfs_calc_trans_metadata_size(fs_info, dropped); |
0ca1f7ce | 6159 | |
0b246afa | 6160 | if (btrfs_is_testing(fs_info)) |
6a3891c5 JB |
6161 | return; |
6162 | ||
691fa059 NB |
6163 | trace_btrfs_space_reservation(fs_info, "delalloc", btrfs_ino(inode), |
6164 | to_free, 0); | |
c5567237 | 6165 | |
2ff7e61e | 6166 | btrfs_block_rsv_release(fs_info, &fs_info->delalloc_block_rsv, to_free); |
0ca1f7ce YZ |
6167 | } |
6168 | ||
1ada3a62 | 6169 | /** |
7cf5b976 | 6170 | * btrfs_delalloc_reserve_space - reserve data and metadata space for |
1ada3a62 QW |
6171 | * delalloc |
6172 | * @inode: inode we're writing to | |
6173 | * @start: start range we are writing to | |
6174 | * @len: how long the range we are writing to | |
364ecf36 QW |
6175 | * @reserved: mandatory parameter, record actually reserved qgroup ranges of |
6176 | * current reservation. | |
1ada3a62 | 6177 | * |
1ada3a62 QW |
6178 | * This will do the following things |
6179 | * | |
6180 | * o reserve space in data space info for num bytes | |
6181 | * and reserve precious corresponding qgroup space | |
6182 | * (Done in check_data_free_space) | |
6183 | * | |
6184 | * o reserve space for metadata space, based on the number of outstanding | |
6185 | * extents and how much csums will be needed | |
6186 | * also reserve metadata space in a per root over-reserve method. | |
6187 | * o add to the inodes->delalloc_bytes | |
6188 | * o add it to the fs_info's delalloc inodes list. | |
6189 | * (Above 3 all done in delalloc_reserve_metadata) | |
6190 | * | |
6191 | * Return 0 for success | |
6192 | * Return <0 for error(-ENOSPC or -EQUOT) | |
6193 | */ | |
364ecf36 QW |
6194 | int btrfs_delalloc_reserve_space(struct inode *inode, |
6195 | struct extent_changeset **reserved, u64 start, u64 len) | |
1ada3a62 QW |
6196 | { |
6197 | int ret; | |
6198 | ||
364ecf36 | 6199 | ret = btrfs_check_data_free_space(inode, reserved, start, len); |
1ada3a62 QW |
6200 | if (ret < 0) |
6201 | return ret; | |
9f3db423 | 6202 | ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len); |
1ada3a62 | 6203 | if (ret < 0) |
bc42bda2 | 6204 | btrfs_free_reserved_data_space(inode, *reserved, start, len); |
1ada3a62 QW |
6205 | return ret; |
6206 | } | |
6207 | ||
7709cde3 | 6208 | /** |
7cf5b976 | 6209 | * btrfs_delalloc_release_space - release data and metadata space for delalloc |
1ada3a62 QW |
6210 | * @inode: inode we're releasing space for |
6211 | * @start: start position of the space already reserved | |
6212 | * @len: the len of the space already reserved | |
6213 | * | |
6214 | * This must be matched with a call to btrfs_delalloc_reserve_space. This is | |
6215 | * called in the case that we don't need the metadata AND data reservations | |
6216 | * anymore. So if there is an error or we insert an inline extent. | |
6217 | * | |
6218 | * This function will release the metadata space that was not used and will | |
6219 | * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes | |
6220 | * list if there are no delalloc bytes left. | |
6221 | * Also it will handle the qgroup reserved space. | |
6222 | */ | |
bc42bda2 QW |
6223 | void btrfs_delalloc_release_space(struct inode *inode, |
6224 | struct extent_changeset *reserved, u64 start, u64 len) | |
1ada3a62 | 6225 | { |
691fa059 | 6226 | btrfs_delalloc_release_metadata(BTRFS_I(inode), len); |
bc42bda2 | 6227 | btrfs_free_reserved_data_space(inode, reserved, start, len); |
6324fbf3 CM |
6228 | } |
6229 | ||
ce93ec54 | 6230 | static int update_block_group(struct btrfs_trans_handle *trans, |
6202df69 | 6231 | struct btrfs_fs_info *info, u64 bytenr, |
ce93ec54 | 6232 | u64 num_bytes, int alloc) |
9078a3e1 | 6233 | { |
0af3d00b | 6234 | struct btrfs_block_group_cache *cache = NULL; |
db94535d | 6235 | u64 total = num_bytes; |
9078a3e1 | 6236 | u64 old_val; |
db94535d | 6237 | u64 byte_in_group; |
0af3d00b | 6238 | int factor; |
3e1ad54f | 6239 | |
5d4f98a2 | 6240 | /* block accounting for super block */ |
eb73c1b7 | 6241 | spin_lock(&info->delalloc_root_lock); |
6c41761f | 6242 | old_val = btrfs_super_bytes_used(info->super_copy); |
5d4f98a2 YZ |
6243 | if (alloc) |
6244 | old_val += num_bytes; | |
6245 | else | |
6246 | old_val -= num_bytes; | |
6c41761f | 6247 | btrfs_set_super_bytes_used(info->super_copy, old_val); |
eb73c1b7 | 6248 | spin_unlock(&info->delalloc_root_lock); |
5d4f98a2 | 6249 | |
d397712b | 6250 | while (total) { |
db94535d | 6251 | cache = btrfs_lookup_block_group(info, bytenr); |
f3465ca4 | 6252 | if (!cache) |
79787eaa | 6253 | return -ENOENT; |
b742bb82 YZ |
6254 | if (cache->flags & (BTRFS_BLOCK_GROUP_DUP | |
6255 | BTRFS_BLOCK_GROUP_RAID1 | | |
6256 | BTRFS_BLOCK_GROUP_RAID10)) | |
6257 | factor = 2; | |
6258 | else | |
6259 | factor = 1; | |
9d66e233 JB |
6260 | /* |
6261 | * If this block group has free space cache written out, we | |
6262 | * need to make sure to load it if we are removing space. This | |
6263 | * is because we need the unpinning stage to actually add the | |
6264 | * space back to the block group, otherwise we will leak space. | |
6265 | */ | |
6266 | if (!alloc && cache->cached == BTRFS_CACHE_NO) | |
f6373bf3 | 6267 | cache_block_group(cache, 1); |
0af3d00b | 6268 | |
db94535d CM |
6269 | byte_in_group = bytenr - cache->key.objectid; |
6270 | WARN_ON(byte_in_group > cache->key.offset); | |
9078a3e1 | 6271 | |
25179201 | 6272 | spin_lock(&cache->space_info->lock); |
c286ac48 | 6273 | spin_lock(&cache->lock); |
0af3d00b | 6274 | |
6202df69 | 6275 | if (btrfs_test_opt(info, SPACE_CACHE) && |
0af3d00b JB |
6276 | cache->disk_cache_state < BTRFS_DC_CLEAR) |
6277 | cache->disk_cache_state = BTRFS_DC_CLEAR; | |
6278 | ||
9078a3e1 | 6279 | old_val = btrfs_block_group_used(&cache->item); |
db94535d | 6280 | num_bytes = min(total, cache->key.offset - byte_in_group); |
cd1bc465 | 6281 | if (alloc) { |
db94535d | 6282 | old_val += num_bytes; |
11833d66 YZ |
6283 | btrfs_set_block_group_used(&cache->item, old_val); |
6284 | cache->reserved -= num_bytes; | |
11833d66 | 6285 | cache->space_info->bytes_reserved -= num_bytes; |
b742bb82 YZ |
6286 | cache->space_info->bytes_used += num_bytes; |
6287 | cache->space_info->disk_used += num_bytes * factor; | |
c286ac48 | 6288 | spin_unlock(&cache->lock); |
25179201 | 6289 | spin_unlock(&cache->space_info->lock); |
cd1bc465 | 6290 | } else { |
db94535d | 6291 | old_val -= num_bytes; |
ae0ab003 FM |
6292 | btrfs_set_block_group_used(&cache->item, old_val); |
6293 | cache->pinned += num_bytes; | |
6294 | cache->space_info->bytes_pinned += num_bytes; | |
6295 | cache->space_info->bytes_used -= num_bytes; | |
6296 | cache->space_info->disk_used -= num_bytes * factor; | |
6297 | spin_unlock(&cache->lock); | |
6298 | spin_unlock(&cache->space_info->lock); | |
47ab2a6c | 6299 | |
0b246afa | 6300 | trace_btrfs_space_reservation(info, "pinned", |
c51e7bb1 JB |
6301 | cache->space_info->flags, |
6302 | num_bytes, 1); | |
d7eae340 OS |
6303 | percpu_counter_add(&cache->space_info->total_bytes_pinned, |
6304 | num_bytes); | |
ae0ab003 FM |
6305 | set_extent_dirty(info->pinned_extents, |
6306 | bytenr, bytenr + num_bytes - 1, | |
6307 | GFP_NOFS | __GFP_NOFAIL); | |
cd1bc465 | 6308 | } |
1bbc621e CM |
6309 | |
6310 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
6311 | if (list_empty(&cache->dirty_list)) { | |
6312 | list_add_tail(&cache->dirty_list, | |
6313 | &trans->transaction->dirty_bgs); | |
6314 | trans->transaction->num_dirty_bgs++; | |
6315 | btrfs_get_block_group(cache); | |
6316 | } | |
6317 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
6318 | ||
036a9348 FM |
6319 | /* |
6320 | * No longer have used bytes in this block group, queue it for | |
6321 | * deletion. We do this after adding the block group to the | |
6322 | * dirty list to avoid races between cleaner kthread and space | |
6323 | * cache writeout. | |
6324 | */ | |
6325 | if (!alloc && old_val == 0) { | |
6326 | spin_lock(&info->unused_bgs_lock); | |
6327 | if (list_empty(&cache->bg_list)) { | |
6328 | btrfs_get_block_group(cache); | |
6329 | list_add_tail(&cache->bg_list, | |
6330 | &info->unused_bgs); | |
6331 | } | |
6332 | spin_unlock(&info->unused_bgs_lock); | |
6333 | } | |
6334 | ||
fa9c0d79 | 6335 | btrfs_put_block_group(cache); |
db94535d CM |
6336 | total -= num_bytes; |
6337 | bytenr += num_bytes; | |
9078a3e1 CM |
6338 | } |
6339 | return 0; | |
6340 | } | |
6324fbf3 | 6341 | |
2ff7e61e | 6342 | static u64 first_logical_byte(struct btrfs_fs_info *fs_info, u64 search_start) |
a061fc8d | 6343 | { |
0f9dd46c | 6344 | struct btrfs_block_group_cache *cache; |
d2fb3437 | 6345 | u64 bytenr; |
0f9dd46c | 6346 | |
0b246afa JM |
6347 | spin_lock(&fs_info->block_group_cache_lock); |
6348 | bytenr = fs_info->first_logical_byte; | |
6349 | spin_unlock(&fs_info->block_group_cache_lock); | |
a1897fdd LB |
6350 | |
6351 | if (bytenr < (u64)-1) | |
6352 | return bytenr; | |
6353 | ||
0b246afa | 6354 | cache = btrfs_lookup_first_block_group(fs_info, search_start); |
0f9dd46c | 6355 | if (!cache) |
a061fc8d | 6356 | return 0; |
0f9dd46c | 6357 | |
d2fb3437 | 6358 | bytenr = cache->key.objectid; |
fa9c0d79 | 6359 | btrfs_put_block_group(cache); |
d2fb3437 YZ |
6360 | |
6361 | return bytenr; | |
a061fc8d CM |
6362 | } |
6363 | ||
2ff7e61e | 6364 | static int pin_down_extent(struct btrfs_fs_info *fs_info, |
f0486c68 YZ |
6365 | struct btrfs_block_group_cache *cache, |
6366 | u64 bytenr, u64 num_bytes, int reserved) | |
324ae4df | 6367 | { |
11833d66 YZ |
6368 | spin_lock(&cache->space_info->lock); |
6369 | spin_lock(&cache->lock); | |
6370 | cache->pinned += num_bytes; | |
6371 | cache->space_info->bytes_pinned += num_bytes; | |
6372 | if (reserved) { | |
6373 | cache->reserved -= num_bytes; | |
6374 | cache->space_info->bytes_reserved -= num_bytes; | |
6375 | } | |
6376 | spin_unlock(&cache->lock); | |
6377 | spin_unlock(&cache->space_info->lock); | |
68b38550 | 6378 | |
0b246afa | 6379 | trace_btrfs_space_reservation(fs_info, "pinned", |
c51e7bb1 | 6380 | cache->space_info->flags, num_bytes, 1); |
4da8b76d | 6381 | percpu_counter_add(&cache->space_info->total_bytes_pinned, num_bytes); |
0b246afa | 6382 | set_extent_dirty(fs_info->pinned_extents, bytenr, |
f0486c68 YZ |
6383 | bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL); |
6384 | return 0; | |
6385 | } | |
68b38550 | 6386 | |
f0486c68 YZ |
6387 | /* |
6388 | * this function must be called within transaction | |
6389 | */ | |
2ff7e61e | 6390 | int btrfs_pin_extent(struct btrfs_fs_info *fs_info, |
f0486c68 YZ |
6391 | u64 bytenr, u64 num_bytes, int reserved) |
6392 | { | |
6393 | struct btrfs_block_group_cache *cache; | |
68b38550 | 6394 | |
0b246afa | 6395 | cache = btrfs_lookup_block_group(fs_info, bytenr); |
79787eaa | 6396 | BUG_ON(!cache); /* Logic error */ |
f0486c68 | 6397 | |
2ff7e61e | 6398 | pin_down_extent(fs_info, cache, bytenr, num_bytes, reserved); |
f0486c68 YZ |
6399 | |
6400 | btrfs_put_block_group(cache); | |
11833d66 YZ |
6401 | return 0; |
6402 | } | |
6403 | ||
f0486c68 | 6404 | /* |
e688b725 CM |
6405 | * this function must be called within transaction |
6406 | */ | |
2ff7e61e | 6407 | int btrfs_pin_extent_for_log_replay(struct btrfs_fs_info *fs_info, |
e688b725 CM |
6408 | u64 bytenr, u64 num_bytes) |
6409 | { | |
6410 | struct btrfs_block_group_cache *cache; | |
b50c6e25 | 6411 | int ret; |
e688b725 | 6412 | |
0b246afa | 6413 | cache = btrfs_lookup_block_group(fs_info, bytenr); |
b50c6e25 JB |
6414 | if (!cache) |
6415 | return -EINVAL; | |
e688b725 CM |
6416 | |
6417 | /* | |
6418 | * pull in the free space cache (if any) so that our pin | |
6419 | * removes the free space from the cache. We have load_only set | |
6420 | * to one because the slow code to read in the free extents does check | |
6421 | * the pinned extents. | |
6422 | */ | |
f6373bf3 | 6423 | cache_block_group(cache, 1); |
e688b725 | 6424 | |
2ff7e61e | 6425 | pin_down_extent(fs_info, cache, bytenr, num_bytes, 0); |
e688b725 CM |
6426 | |
6427 | /* remove us from the free space cache (if we're there at all) */ | |
b50c6e25 | 6428 | ret = btrfs_remove_free_space(cache, bytenr, num_bytes); |
e688b725 | 6429 | btrfs_put_block_group(cache); |
b50c6e25 | 6430 | return ret; |
e688b725 CM |
6431 | } |
6432 | ||
2ff7e61e JM |
6433 | static int __exclude_logged_extent(struct btrfs_fs_info *fs_info, |
6434 | u64 start, u64 num_bytes) | |
8c2a1a30 JB |
6435 | { |
6436 | int ret; | |
6437 | struct btrfs_block_group_cache *block_group; | |
6438 | struct btrfs_caching_control *caching_ctl; | |
6439 | ||
0b246afa | 6440 | block_group = btrfs_lookup_block_group(fs_info, start); |
8c2a1a30 JB |
6441 | if (!block_group) |
6442 | return -EINVAL; | |
6443 | ||
6444 | cache_block_group(block_group, 0); | |
6445 | caching_ctl = get_caching_control(block_group); | |
6446 | ||
6447 | if (!caching_ctl) { | |
6448 | /* Logic error */ | |
6449 | BUG_ON(!block_group_cache_done(block_group)); | |
6450 | ret = btrfs_remove_free_space(block_group, start, num_bytes); | |
6451 | } else { | |
6452 | mutex_lock(&caching_ctl->mutex); | |
6453 | ||
6454 | if (start >= caching_ctl->progress) { | |
2ff7e61e | 6455 | ret = add_excluded_extent(fs_info, start, num_bytes); |
8c2a1a30 JB |
6456 | } else if (start + num_bytes <= caching_ctl->progress) { |
6457 | ret = btrfs_remove_free_space(block_group, | |
6458 | start, num_bytes); | |
6459 | } else { | |
6460 | num_bytes = caching_ctl->progress - start; | |
6461 | ret = btrfs_remove_free_space(block_group, | |
6462 | start, num_bytes); | |
6463 | if (ret) | |
6464 | goto out_lock; | |
6465 | ||
6466 | num_bytes = (start + num_bytes) - | |
6467 | caching_ctl->progress; | |
6468 | start = caching_ctl->progress; | |
2ff7e61e | 6469 | ret = add_excluded_extent(fs_info, start, num_bytes); |
8c2a1a30 JB |
6470 | } |
6471 | out_lock: | |
6472 | mutex_unlock(&caching_ctl->mutex); | |
6473 | put_caching_control(caching_ctl); | |
6474 | } | |
6475 | btrfs_put_block_group(block_group); | |
6476 | return ret; | |
6477 | } | |
6478 | ||
2ff7e61e | 6479 | int btrfs_exclude_logged_extents(struct btrfs_fs_info *fs_info, |
8c2a1a30 JB |
6480 | struct extent_buffer *eb) |
6481 | { | |
6482 | struct btrfs_file_extent_item *item; | |
6483 | struct btrfs_key key; | |
6484 | int found_type; | |
6485 | int i; | |
6486 | ||
2ff7e61e | 6487 | if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) |
8c2a1a30 JB |
6488 | return 0; |
6489 | ||
6490 | for (i = 0; i < btrfs_header_nritems(eb); i++) { | |
6491 | btrfs_item_key_to_cpu(eb, &key, i); | |
6492 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
6493 | continue; | |
6494 | item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item); | |
6495 | found_type = btrfs_file_extent_type(eb, item); | |
6496 | if (found_type == BTRFS_FILE_EXTENT_INLINE) | |
6497 | continue; | |
6498 | if (btrfs_file_extent_disk_bytenr(eb, item) == 0) | |
6499 | continue; | |
6500 | key.objectid = btrfs_file_extent_disk_bytenr(eb, item); | |
6501 | key.offset = btrfs_file_extent_disk_num_bytes(eb, item); | |
2ff7e61e | 6502 | __exclude_logged_extent(fs_info, key.objectid, key.offset); |
8c2a1a30 JB |
6503 | } |
6504 | ||
6505 | return 0; | |
6506 | } | |
6507 | ||
9cfa3e34 FM |
6508 | static void |
6509 | btrfs_inc_block_group_reservations(struct btrfs_block_group_cache *bg) | |
6510 | { | |
6511 | atomic_inc(&bg->reservations); | |
6512 | } | |
6513 | ||
6514 | void btrfs_dec_block_group_reservations(struct btrfs_fs_info *fs_info, | |
6515 | const u64 start) | |
6516 | { | |
6517 | struct btrfs_block_group_cache *bg; | |
6518 | ||
6519 | bg = btrfs_lookup_block_group(fs_info, start); | |
6520 | ASSERT(bg); | |
6521 | if (atomic_dec_and_test(&bg->reservations)) | |
6522 | wake_up_atomic_t(&bg->reservations); | |
6523 | btrfs_put_block_group(bg); | |
6524 | } | |
6525 | ||
6526 | static int btrfs_wait_bg_reservations_atomic_t(atomic_t *a) | |
6527 | { | |
6528 | schedule(); | |
6529 | return 0; | |
6530 | } | |
6531 | ||
6532 | void btrfs_wait_block_group_reservations(struct btrfs_block_group_cache *bg) | |
6533 | { | |
6534 | struct btrfs_space_info *space_info = bg->space_info; | |
6535 | ||
6536 | ASSERT(bg->ro); | |
6537 | ||
6538 | if (!(bg->flags & BTRFS_BLOCK_GROUP_DATA)) | |
6539 | return; | |
6540 | ||
6541 | /* | |
6542 | * Our block group is read only but before we set it to read only, | |
6543 | * some task might have had allocated an extent from it already, but it | |
6544 | * has not yet created a respective ordered extent (and added it to a | |
6545 | * root's list of ordered extents). | |
6546 | * Therefore wait for any task currently allocating extents, since the | |
6547 | * block group's reservations counter is incremented while a read lock | |
6548 | * on the groups' semaphore is held and decremented after releasing | |
6549 | * the read access on that semaphore and creating the ordered extent. | |
6550 | */ | |
6551 | down_write(&space_info->groups_sem); | |
6552 | up_write(&space_info->groups_sem); | |
6553 | ||
6554 | wait_on_atomic_t(&bg->reservations, | |
6555 | btrfs_wait_bg_reservations_atomic_t, | |
6556 | TASK_UNINTERRUPTIBLE); | |
6557 | } | |
6558 | ||
fb25e914 | 6559 | /** |
4824f1f4 | 6560 | * btrfs_add_reserved_bytes - update the block_group and space info counters |
fb25e914 | 6561 | * @cache: The cache we are manipulating |
18513091 WX |
6562 | * @ram_bytes: The number of bytes of file content, and will be same to |
6563 | * @num_bytes except for the compress path. | |
fb25e914 | 6564 | * @num_bytes: The number of bytes in question |
e570fd27 | 6565 | * @delalloc: The blocks are allocated for the delalloc write |
fb25e914 | 6566 | * |
745699ef XW |
6567 | * This is called by the allocator when it reserves space. If this is a |
6568 | * reservation and the block group has become read only we cannot make the | |
6569 | * reservation and return -EAGAIN, otherwise this function always succeeds. | |
f0486c68 | 6570 | */ |
4824f1f4 | 6571 | static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache *cache, |
18513091 | 6572 | u64 ram_bytes, u64 num_bytes, int delalloc) |
11833d66 | 6573 | { |
fb25e914 | 6574 | struct btrfs_space_info *space_info = cache->space_info; |
f0486c68 | 6575 | int ret = 0; |
79787eaa | 6576 | |
fb25e914 JB |
6577 | spin_lock(&space_info->lock); |
6578 | spin_lock(&cache->lock); | |
4824f1f4 WX |
6579 | if (cache->ro) { |
6580 | ret = -EAGAIN; | |
fb25e914 | 6581 | } else { |
4824f1f4 WX |
6582 | cache->reserved += num_bytes; |
6583 | space_info->bytes_reserved += num_bytes; | |
e570fd27 | 6584 | |
18513091 WX |
6585 | trace_btrfs_space_reservation(cache->fs_info, |
6586 | "space_info", space_info->flags, | |
6587 | ram_bytes, 0); | |
6588 | space_info->bytes_may_use -= ram_bytes; | |
e570fd27 | 6589 | if (delalloc) |
4824f1f4 | 6590 | cache->delalloc_bytes += num_bytes; |
324ae4df | 6591 | } |
fb25e914 JB |
6592 | spin_unlock(&cache->lock); |
6593 | spin_unlock(&space_info->lock); | |
f0486c68 | 6594 | return ret; |
324ae4df | 6595 | } |
9078a3e1 | 6596 | |
4824f1f4 WX |
6597 | /** |
6598 | * btrfs_free_reserved_bytes - update the block_group and space info counters | |
6599 | * @cache: The cache we are manipulating | |
6600 | * @num_bytes: The number of bytes in question | |
6601 | * @delalloc: The blocks are allocated for the delalloc write | |
6602 | * | |
6603 | * This is called by somebody who is freeing space that was never actually used | |
6604 | * on disk. For example if you reserve some space for a new leaf in transaction | |
6605 | * A and before transaction A commits you free that leaf, you call this with | |
6606 | * reserve set to 0 in order to clear the reservation. | |
6607 | */ | |
6608 | ||
6609 | static int btrfs_free_reserved_bytes(struct btrfs_block_group_cache *cache, | |
6610 | u64 num_bytes, int delalloc) | |
6611 | { | |
6612 | struct btrfs_space_info *space_info = cache->space_info; | |
6613 | int ret = 0; | |
6614 | ||
6615 | spin_lock(&space_info->lock); | |
6616 | spin_lock(&cache->lock); | |
6617 | if (cache->ro) | |
6618 | space_info->bytes_readonly += num_bytes; | |
6619 | cache->reserved -= num_bytes; | |
6620 | space_info->bytes_reserved -= num_bytes; | |
6621 | ||
6622 | if (delalloc) | |
6623 | cache->delalloc_bytes -= num_bytes; | |
6624 | spin_unlock(&cache->lock); | |
6625 | spin_unlock(&space_info->lock); | |
6626 | return ret; | |
6627 | } | |
8b74c03e | 6628 | void btrfs_prepare_extent_commit(struct btrfs_fs_info *fs_info) |
e8569813 | 6629 | { |
11833d66 YZ |
6630 | struct btrfs_caching_control *next; |
6631 | struct btrfs_caching_control *caching_ctl; | |
6632 | struct btrfs_block_group_cache *cache; | |
e8569813 | 6633 | |
9e351cc8 | 6634 | down_write(&fs_info->commit_root_sem); |
25179201 | 6635 | |
11833d66 YZ |
6636 | list_for_each_entry_safe(caching_ctl, next, |
6637 | &fs_info->caching_block_groups, list) { | |
6638 | cache = caching_ctl->block_group; | |
6639 | if (block_group_cache_done(cache)) { | |
6640 | cache->last_byte_to_unpin = (u64)-1; | |
6641 | list_del_init(&caching_ctl->list); | |
6642 | put_caching_control(caching_ctl); | |
e8569813 | 6643 | } else { |
11833d66 | 6644 | cache->last_byte_to_unpin = caching_ctl->progress; |
e8569813 | 6645 | } |
e8569813 | 6646 | } |
11833d66 YZ |
6647 | |
6648 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) | |
6649 | fs_info->pinned_extents = &fs_info->freed_extents[1]; | |
6650 | else | |
6651 | fs_info->pinned_extents = &fs_info->freed_extents[0]; | |
6652 | ||
9e351cc8 | 6653 | up_write(&fs_info->commit_root_sem); |
8929ecfa YZ |
6654 | |
6655 | update_global_block_rsv(fs_info); | |
e8569813 ZY |
6656 | } |
6657 | ||
c759c4e1 JB |
6658 | /* |
6659 | * Returns the free cluster for the given space info and sets empty_cluster to | |
6660 | * what it should be based on the mount options. | |
6661 | */ | |
6662 | static struct btrfs_free_cluster * | |
2ff7e61e JM |
6663 | fetch_cluster_info(struct btrfs_fs_info *fs_info, |
6664 | struct btrfs_space_info *space_info, u64 *empty_cluster) | |
c759c4e1 JB |
6665 | { |
6666 | struct btrfs_free_cluster *ret = NULL; | |
0b246afa | 6667 | bool ssd = btrfs_test_opt(fs_info, SSD); |
c759c4e1 JB |
6668 | |
6669 | *empty_cluster = 0; | |
6670 | if (btrfs_mixed_space_info(space_info)) | |
6671 | return ret; | |
6672 | ||
6673 | if (ssd) | |
ee22184b | 6674 | *empty_cluster = SZ_2M; |
c759c4e1 | 6675 | if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) { |
0b246afa | 6676 | ret = &fs_info->meta_alloc_cluster; |
c759c4e1 | 6677 | if (!ssd) |
ee22184b | 6678 | *empty_cluster = SZ_64K; |
c759c4e1 | 6679 | } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) && ssd) { |
0b246afa | 6680 | ret = &fs_info->data_alloc_cluster; |
c759c4e1 JB |
6681 | } |
6682 | ||
6683 | return ret; | |
6684 | } | |
6685 | ||
2ff7e61e JM |
6686 | static int unpin_extent_range(struct btrfs_fs_info *fs_info, |
6687 | u64 start, u64 end, | |
678886bd | 6688 | const bool return_free_space) |
ccd467d6 | 6689 | { |
11833d66 | 6690 | struct btrfs_block_group_cache *cache = NULL; |
7b398f8e JB |
6691 | struct btrfs_space_info *space_info; |
6692 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
c759c4e1 | 6693 | struct btrfs_free_cluster *cluster = NULL; |
11833d66 | 6694 | u64 len; |
c759c4e1 JB |
6695 | u64 total_unpinned = 0; |
6696 | u64 empty_cluster = 0; | |
7b398f8e | 6697 | bool readonly; |
ccd467d6 | 6698 | |
11833d66 | 6699 | while (start <= end) { |
7b398f8e | 6700 | readonly = false; |
11833d66 YZ |
6701 | if (!cache || |
6702 | start >= cache->key.objectid + cache->key.offset) { | |
6703 | if (cache) | |
6704 | btrfs_put_block_group(cache); | |
c759c4e1 | 6705 | total_unpinned = 0; |
11833d66 | 6706 | cache = btrfs_lookup_block_group(fs_info, start); |
79787eaa | 6707 | BUG_ON(!cache); /* Logic error */ |
c759c4e1 | 6708 | |
2ff7e61e | 6709 | cluster = fetch_cluster_info(fs_info, |
c759c4e1 JB |
6710 | cache->space_info, |
6711 | &empty_cluster); | |
6712 | empty_cluster <<= 1; | |
11833d66 YZ |
6713 | } |
6714 | ||
6715 | len = cache->key.objectid + cache->key.offset - start; | |
6716 | len = min(len, end + 1 - start); | |
6717 | ||
6718 | if (start < cache->last_byte_to_unpin) { | |
6719 | len = min(len, cache->last_byte_to_unpin - start); | |
678886bd FM |
6720 | if (return_free_space) |
6721 | btrfs_add_free_space(cache, start, len); | |
11833d66 YZ |
6722 | } |
6723 | ||
f0486c68 | 6724 | start += len; |
c759c4e1 | 6725 | total_unpinned += len; |
7b398f8e | 6726 | space_info = cache->space_info; |
f0486c68 | 6727 | |
c759c4e1 JB |
6728 | /* |
6729 | * If this space cluster has been marked as fragmented and we've | |
6730 | * unpinned enough in this block group to potentially allow a | |
6731 | * cluster to be created inside of it go ahead and clear the | |
6732 | * fragmented check. | |
6733 | */ | |
6734 | if (cluster && cluster->fragmented && | |
6735 | total_unpinned > empty_cluster) { | |
6736 | spin_lock(&cluster->lock); | |
6737 | cluster->fragmented = 0; | |
6738 | spin_unlock(&cluster->lock); | |
6739 | } | |
6740 | ||
7b398f8e | 6741 | spin_lock(&space_info->lock); |
11833d66 YZ |
6742 | spin_lock(&cache->lock); |
6743 | cache->pinned -= len; | |
7b398f8e | 6744 | space_info->bytes_pinned -= len; |
c51e7bb1 JB |
6745 | |
6746 | trace_btrfs_space_reservation(fs_info, "pinned", | |
6747 | space_info->flags, len, 0); | |
4f4db217 | 6748 | space_info->max_extent_size = 0; |
d288db5d | 6749 | percpu_counter_add(&space_info->total_bytes_pinned, -len); |
7b398f8e JB |
6750 | if (cache->ro) { |
6751 | space_info->bytes_readonly += len; | |
6752 | readonly = true; | |
6753 | } | |
11833d66 | 6754 | spin_unlock(&cache->lock); |
957780eb JB |
6755 | if (!readonly && return_free_space && |
6756 | global_rsv->space_info == space_info) { | |
6757 | u64 to_add = len; | |
6758 | WARN_ON(!return_free_space); | |
7b398f8e JB |
6759 | spin_lock(&global_rsv->lock); |
6760 | if (!global_rsv->full) { | |
957780eb JB |
6761 | to_add = min(len, global_rsv->size - |
6762 | global_rsv->reserved); | |
6763 | global_rsv->reserved += to_add; | |
6764 | space_info->bytes_may_use += to_add; | |
7b398f8e JB |
6765 | if (global_rsv->reserved >= global_rsv->size) |
6766 | global_rsv->full = 1; | |
957780eb JB |
6767 | trace_btrfs_space_reservation(fs_info, |
6768 | "space_info", | |
6769 | space_info->flags, | |
6770 | to_add, 1); | |
6771 | len -= to_add; | |
7b398f8e JB |
6772 | } |
6773 | spin_unlock(&global_rsv->lock); | |
957780eb JB |
6774 | /* Add to any tickets we may have */ |
6775 | if (len) | |
6776 | space_info_add_new_bytes(fs_info, space_info, | |
6777 | len); | |
7b398f8e JB |
6778 | } |
6779 | spin_unlock(&space_info->lock); | |
ccd467d6 | 6780 | } |
11833d66 YZ |
6781 | |
6782 | if (cache) | |
6783 | btrfs_put_block_group(cache); | |
ccd467d6 CM |
6784 | return 0; |
6785 | } | |
6786 | ||
6787 | int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans, | |
2ff7e61e | 6788 | struct btrfs_fs_info *fs_info) |
a28ec197 | 6789 | { |
e33e17ee JM |
6790 | struct btrfs_block_group_cache *block_group, *tmp; |
6791 | struct list_head *deleted_bgs; | |
11833d66 | 6792 | struct extent_io_tree *unpin; |
1a5bc167 CM |
6793 | u64 start; |
6794 | u64 end; | |
a28ec197 | 6795 | int ret; |
a28ec197 | 6796 | |
11833d66 YZ |
6797 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) |
6798 | unpin = &fs_info->freed_extents[1]; | |
6799 | else | |
6800 | unpin = &fs_info->freed_extents[0]; | |
6801 | ||
e33e17ee | 6802 | while (!trans->aborted) { |
d4b450cd | 6803 | mutex_lock(&fs_info->unused_bg_unpin_mutex); |
1a5bc167 | 6804 | ret = find_first_extent_bit(unpin, 0, &start, &end, |
e6138876 | 6805 | EXTENT_DIRTY, NULL); |
d4b450cd FM |
6806 | if (ret) { |
6807 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); | |
a28ec197 | 6808 | break; |
d4b450cd | 6809 | } |
1f3c79a2 | 6810 | |
0b246afa | 6811 | if (btrfs_test_opt(fs_info, DISCARD)) |
2ff7e61e | 6812 | ret = btrfs_discard_extent(fs_info, start, |
5378e607 | 6813 | end + 1 - start, NULL); |
1f3c79a2 | 6814 | |
af6f8f60 | 6815 | clear_extent_dirty(unpin, start, end); |
2ff7e61e | 6816 | unpin_extent_range(fs_info, start, end, true); |
d4b450cd | 6817 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
b9473439 | 6818 | cond_resched(); |
a28ec197 | 6819 | } |
817d52f8 | 6820 | |
e33e17ee JM |
6821 | /* |
6822 | * Transaction is finished. We don't need the lock anymore. We | |
6823 | * do need to clean up the block groups in case of a transaction | |
6824 | * abort. | |
6825 | */ | |
6826 | deleted_bgs = &trans->transaction->deleted_bgs; | |
6827 | list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) { | |
6828 | u64 trimmed = 0; | |
6829 | ||
6830 | ret = -EROFS; | |
6831 | if (!trans->aborted) | |
2ff7e61e | 6832 | ret = btrfs_discard_extent(fs_info, |
e33e17ee JM |
6833 | block_group->key.objectid, |
6834 | block_group->key.offset, | |
6835 | &trimmed); | |
6836 | ||
6837 | list_del_init(&block_group->bg_list); | |
6838 | btrfs_put_block_group_trimming(block_group); | |
6839 | btrfs_put_block_group(block_group); | |
6840 | ||
6841 | if (ret) { | |
6842 | const char *errstr = btrfs_decode_error(ret); | |
6843 | btrfs_warn(fs_info, | |
6844 | "Discard failed while removing blockgroup: errno=%d %s\n", | |
6845 | ret, errstr); | |
6846 | } | |
6847 | } | |
6848 | ||
e20d96d6 CM |
6849 | return 0; |
6850 | } | |
6851 | ||
5d4f98a2 | 6852 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
2ff7e61e | 6853 | struct btrfs_fs_info *info, |
c682f9b3 | 6854 | struct btrfs_delayed_ref_node *node, u64 parent, |
5d4f98a2 YZ |
6855 | u64 root_objectid, u64 owner_objectid, |
6856 | u64 owner_offset, int refs_to_drop, | |
c682f9b3 | 6857 | struct btrfs_delayed_extent_op *extent_op) |
a28ec197 | 6858 | { |
e2fa7227 | 6859 | struct btrfs_key key; |
5d4f98a2 | 6860 | struct btrfs_path *path; |
1261ec42 | 6861 | struct btrfs_root *extent_root = info->extent_root; |
5f39d397 | 6862 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
6863 | struct btrfs_extent_item *ei; |
6864 | struct btrfs_extent_inline_ref *iref; | |
a28ec197 | 6865 | int ret; |
5d4f98a2 | 6866 | int is_data; |
952fccac CM |
6867 | int extent_slot = 0; |
6868 | int found_extent = 0; | |
6869 | int num_to_del = 1; | |
5d4f98a2 YZ |
6870 | u32 item_size; |
6871 | u64 refs; | |
c682f9b3 QW |
6872 | u64 bytenr = node->bytenr; |
6873 | u64 num_bytes = node->num_bytes; | |
fcebe456 | 6874 | int last_ref = 0; |
0b246afa | 6875 | bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA); |
037e6390 | 6876 | |
5caf2a00 | 6877 | path = btrfs_alloc_path(); |
54aa1f4d CM |
6878 | if (!path) |
6879 | return -ENOMEM; | |
5f26f772 | 6880 | |
e4058b54 | 6881 | path->reada = READA_FORWARD; |
b9473439 | 6882 | path->leave_spinning = 1; |
5d4f98a2 YZ |
6883 | |
6884 | is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID; | |
6885 | BUG_ON(!is_data && refs_to_drop != 1); | |
6886 | ||
3173a18f JB |
6887 | if (is_data) |
6888 | skinny_metadata = 0; | |
6889 | ||
87bde3cd | 6890 | ret = lookup_extent_backref(trans, info, path, &iref, |
5d4f98a2 YZ |
6891 | bytenr, num_bytes, parent, |
6892 | root_objectid, owner_objectid, | |
6893 | owner_offset); | |
7bb86316 | 6894 | if (ret == 0) { |
952fccac | 6895 | extent_slot = path->slots[0]; |
5d4f98a2 YZ |
6896 | while (extent_slot >= 0) { |
6897 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
952fccac | 6898 | extent_slot); |
5d4f98a2 | 6899 | if (key.objectid != bytenr) |
952fccac | 6900 | break; |
5d4f98a2 YZ |
6901 | if (key.type == BTRFS_EXTENT_ITEM_KEY && |
6902 | key.offset == num_bytes) { | |
952fccac CM |
6903 | found_extent = 1; |
6904 | break; | |
6905 | } | |
3173a18f JB |
6906 | if (key.type == BTRFS_METADATA_ITEM_KEY && |
6907 | key.offset == owner_objectid) { | |
6908 | found_extent = 1; | |
6909 | break; | |
6910 | } | |
952fccac CM |
6911 | if (path->slots[0] - extent_slot > 5) |
6912 | break; | |
5d4f98a2 | 6913 | extent_slot--; |
952fccac | 6914 | } |
5d4f98a2 YZ |
6915 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
6916 | item_size = btrfs_item_size_nr(path->nodes[0], extent_slot); | |
6917 | if (found_extent && item_size < sizeof(*ei)) | |
6918 | found_extent = 0; | |
6919 | #endif | |
31840ae1 | 6920 | if (!found_extent) { |
5d4f98a2 | 6921 | BUG_ON(iref); |
87bde3cd JM |
6922 | ret = remove_extent_backref(trans, info, path, NULL, |
6923 | refs_to_drop, | |
fcebe456 | 6924 | is_data, &last_ref); |
005d6427 | 6925 | if (ret) { |
66642832 | 6926 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
6927 | goto out; |
6928 | } | |
b3b4aa74 | 6929 | btrfs_release_path(path); |
b9473439 | 6930 | path->leave_spinning = 1; |
5d4f98a2 YZ |
6931 | |
6932 | key.objectid = bytenr; | |
6933 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
6934 | key.offset = num_bytes; | |
6935 | ||
3173a18f JB |
6936 | if (!is_data && skinny_metadata) { |
6937 | key.type = BTRFS_METADATA_ITEM_KEY; | |
6938 | key.offset = owner_objectid; | |
6939 | } | |
6940 | ||
31840ae1 ZY |
6941 | ret = btrfs_search_slot(trans, extent_root, |
6942 | &key, path, -1, 1); | |
3173a18f JB |
6943 | if (ret > 0 && skinny_metadata && path->slots[0]) { |
6944 | /* | |
6945 | * Couldn't find our skinny metadata item, | |
6946 | * see if we have ye olde extent item. | |
6947 | */ | |
6948 | path->slots[0]--; | |
6949 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
6950 | path->slots[0]); | |
6951 | if (key.objectid == bytenr && | |
6952 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
6953 | key.offset == num_bytes) | |
6954 | ret = 0; | |
6955 | } | |
6956 | ||
6957 | if (ret > 0 && skinny_metadata) { | |
6958 | skinny_metadata = false; | |
9ce49a0b | 6959 | key.objectid = bytenr; |
3173a18f JB |
6960 | key.type = BTRFS_EXTENT_ITEM_KEY; |
6961 | key.offset = num_bytes; | |
6962 | btrfs_release_path(path); | |
6963 | ret = btrfs_search_slot(trans, extent_root, | |
6964 | &key, path, -1, 1); | |
6965 | } | |
6966 | ||
f3465ca4 | 6967 | if (ret) { |
5d163e0e JM |
6968 | btrfs_err(info, |
6969 | "umm, got %d back from search, was looking for %llu", | |
6970 | ret, bytenr); | |
b783e62d | 6971 | if (ret > 0) |
2ff7e61e | 6972 | btrfs_print_leaf(info, path->nodes[0]); |
f3465ca4 | 6973 | } |
005d6427 | 6974 | if (ret < 0) { |
66642832 | 6975 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
6976 | goto out; |
6977 | } | |
31840ae1 ZY |
6978 | extent_slot = path->slots[0]; |
6979 | } | |
fae7f21c | 6980 | } else if (WARN_ON(ret == -ENOENT)) { |
2ff7e61e | 6981 | btrfs_print_leaf(info, path->nodes[0]); |
c2cf52eb SK |
6982 | btrfs_err(info, |
6983 | "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu", | |
c1c9ff7c GU |
6984 | bytenr, parent, root_objectid, owner_objectid, |
6985 | owner_offset); | |
66642832 | 6986 | btrfs_abort_transaction(trans, ret); |
c4a050bb | 6987 | goto out; |
79787eaa | 6988 | } else { |
66642832 | 6989 | btrfs_abort_transaction(trans, ret); |
005d6427 | 6990 | goto out; |
7bb86316 | 6991 | } |
5f39d397 CM |
6992 | |
6993 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
6994 | item_size = btrfs_item_size_nr(leaf, extent_slot); |
6995 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
6996 | if (item_size < sizeof(*ei)) { | |
6997 | BUG_ON(found_extent || extent_slot != path->slots[0]); | |
87bde3cd JM |
6998 | ret = convert_extent_item_v0(trans, info, path, owner_objectid, |
6999 | 0); | |
005d6427 | 7000 | if (ret < 0) { |
66642832 | 7001 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7002 | goto out; |
7003 | } | |
5d4f98a2 | 7004 | |
b3b4aa74 | 7005 | btrfs_release_path(path); |
5d4f98a2 YZ |
7006 | path->leave_spinning = 1; |
7007 | ||
7008 | key.objectid = bytenr; | |
7009 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
7010 | key.offset = num_bytes; | |
7011 | ||
7012 | ret = btrfs_search_slot(trans, extent_root, &key, path, | |
7013 | -1, 1); | |
7014 | if (ret) { | |
5d163e0e JM |
7015 | btrfs_err(info, |
7016 | "umm, got %d back from search, was looking for %llu", | |
c1c9ff7c | 7017 | ret, bytenr); |
2ff7e61e | 7018 | btrfs_print_leaf(info, path->nodes[0]); |
5d4f98a2 | 7019 | } |
005d6427 | 7020 | if (ret < 0) { |
66642832 | 7021 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7022 | goto out; |
7023 | } | |
7024 | ||
5d4f98a2 YZ |
7025 | extent_slot = path->slots[0]; |
7026 | leaf = path->nodes[0]; | |
7027 | item_size = btrfs_item_size_nr(leaf, extent_slot); | |
7028 | } | |
7029 | #endif | |
7030 | BUG_ON(item_size < sizeof(*ei)); | |
952fccac | 7031 | ei = btrfs_item_ptr(leaf, extent_slot, |
123abc88 | 7032 | struct btrfs_extent_item); |
3173a18f JB |
7033 | if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID && |
7034 | key.type == BTRFS_EXTENT_ITEM_KEY) { | |
5d4f98a2 YZ |
7035 | struct btrfs_tree_block_info *bi; |
7036 | BUG_ON(item_size < sizeof(*ei) + sizeof(*bi)); | |
7037 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
7038 | WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi)); | |
7039 | } | |
56bec294 | 7040 | |
5d4f98a2 | 7041 | refs = btrfs_extent_refs(leaf, ei); |
32b02538 | 7042 | if (refs < refs_to_drop) { |
5d163e0e JM |
7043 | btrfs_err(info, |
7044 | "trying to drop %d refs but we only have %Lu for bytenr %Lu", | |
7045 | refs_to_drop, refs, bytenr); | |
32b02538 | 7046 | ret = -EINVAL; |
66642832 | 7047 | btrfs_abort_transaction(trans, ret); |
32b02538 JB |
7048 | goto out; |
7049 | } | |
56bec294 | 7050 | refs -= refs_to_drop; |
5f39d397 | 7051 | |
5d4f98a2 YZ |
7052 | if (refs > 0) { |
7053 | if (extent_op) | |
7054 | __run_delayed_extent_op(extent_op, leaf, ei); | |
7055 | /* | |
7056 | * In the case of inline back ref, reference count will | |
7057 | * be updated by remove_extent_backref | |
952fccac | 7058 | */ |
5d4f98a2 YZ |
7059 | if (iref) { |
7060 | BUG_ON(!found_extent); | |
7061 | } else { | |
7062 | btrfs_set_extent_refs(leaf, ei, refs); | |
7063 | btrfs_mark_buffer_dirty(leaf); | |
7064 | } | |
7065 | if (found_extent) { | |
87bde3cd | 7066 | ret = remove_extent_backref(trans, info, path, |
5d4f98a2 | 7067 | iref, refs_to_drop, |
fcebe456 | 7068 | is_data, &last_ref); |
005d6427 | 7069 | if (ret) { |
66642832 | 7070 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7071 | goto out; |
7072 | } | |
952fccac | 7073 | } |
5d4f98a2 | 7074 | } else { |
5d4f98a2 YZ |
7075 | if (found_extent) { |
7076 | BUG_ON(is_data && refs_to_drop != | |
9ed0dea0 | 7077 | extent_data_ref_count(path, iref)); |
5d4f98a2 YZ |
7078 | if (iref) { |
7079 | BUG_ON(path->slots[0] != extent_slot); | |
7080 | } else { | |
7081 | BUG_ON(path->slots[0] != extent_slot + 1); | |
7082 | path->slots[0] = extent_slot; | |
7083 | num_to_del = 2; | |
7084 | } | |
78fae27e | 7085 | } |
b9473439 | 7086 | |
fcebe456 | 7087 | last_ref = 1; |
952fccac CM |
7088 | ret = btrfs_del_items(trans, extent_root, path, path->slots[0], |
7089 | num_to_del); | |
005d6427 | 7090 | if (ret) { |
66642832 | 7091 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7092 | goto out; |
7093 | } | |
b3b4aa74 | 7094 | btrfs_release_path(path); |
21af804c | 7095 | |
5d4f98a2 | 7096 | if (is_data) { |
5b4aacef | 7097 | ret = btrfs_del_csums(trans, info, bytenr, num_bytes); |
005d6427 | 7098 | if (ret) { |
66642832 | 7099 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7100 | goto out; |
7101 | } | |
459931ec CM |
7102 | } |
7103 | ||
0b246afa | 7104 | ret = add_to_free_space_tree(trans, info, bytenr, num_bytes); |
1e144fb8 | 7105 | if (ret) { |
66642832 | 7106 | btrfs_abort_transaction(trans, ret); |
1e144fb8 OS |
7107 | goto out; |
7108 | } | |
7109 | ||
0b246afa | 7110 | ret = update_block_group(trans, info, bytenr, num_bytes, 0); |
005d6427 | 7111 | if (ret) { |
66642832 | 7112 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7113 | goto out; |
7114 | } | |
a28ec197 | 7115 | } |
fcebe456 JB |
7116 | btrfs_release_path(path); |
7117 | ||
79787eaa | 7118 | out: |
5caf2a00 | 7119 | btrfs_free_path(path); |
a28ec197 CM |
7120 | return ret; |
7121 | } | |
7122 | ||
1887be66 | 7123 | /* |
f0486c68 | 7124 | * when we free an block, it is possible (and likely) that we free the last |
1887be66 CM |
7125 | * delayed ref for that extent as well. This searches the delayed ref tree for |
7126 | * a given extent, and if there are no other delayed refs to be processed, it | |
7127 | * removes it from the tree. | |
7128 | */ | |
7129 | static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans, | |
2ff7e61e | 7130 | u64 bytenr) |
1887be66 CM |
7131 | { |
7132 | struct btrfs_delayed_ref_head *head; | |
7133 | struct btrfs_delayed_ref_root *delayed_refs; | |
f0486c68 | 7134 | int ret = 0; |
1887be66 CM |
7135 | |
7136 | delayed_refs = &trans->transaction->delayed_refs; | |
7137 | spin_lock(&delayed_refs->lock); | |
f72ad18e | 7138 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
1887be66 | 7139 | if (!head) |
cf93da7b | 7140 | goto out_delayed_unlock; |
1887be66 | 7141 | |
d7df2c79 | 7142 | spin_lock(&head->lock); |
c6fc2454 | 7143 | if (!list_empty(&head->ref_list)) |
1887be66 CM |
7144 | goto out; |
7145 | ||
5d4f98a2 YZ |
7146 | if (head->extent_op) { |
7147 | if (!head->must_insert_reserved) | |
7148 | goto out; | |
78a6184a | 7149 | btrfs_free_delayed_extent_op(head->extent_op); |
5d4f98a2 YZ |
7150 | head->extent_op = NULL; |
7151 | } | |
7152 | ||
1887be66 CM |
7153 | /* |
7154 | * waiting for the lock here would deadlock. If someone else has it | |
7155 | * locked they are already in the process of dropping it anyway | |
7156 | */ | |
7157 | if (!mutex_trylock(&head->mutex)) | |
7158 | goto out; | |
7159 | ||
7160 | /* | |
7161 | * at this point we have a head with no other entries. Go | |
7162 | * ahead and process it. | |
7163 | */ | |
7164 | head->node.in_tree = 0; | |
c46effa6 | 7165 | rb_erase(&head->href_node, &delayed_refs->href_root); |
c3e69d58 | 7166 | |
d7df2c79 | 7167 | atomic_dec(&delayed_refs->num_entries); |
1887be66 CM |
7168 | |
7169 | /* | |
7170 | * we don't take a ref on the node because we're removing it from the | |
7171 | * tree, so we just steal the ref the tree was holding. | |
7172 | */ | |
c3e69d58 | 7173 | delayed_refs->num_heads--; |
d7df2c79 | 7174 | if (head->processing == 0) |
c3e69d58 | 7175 | delayed_refs->num_heads_ready--; |
d7df2c79 JB |
7176 | head->processing = 0; |
7177 | spin_unlock(&head->lock); | |
1887be66 CM |
7178 | spin_unlock(&delayed_refs->lock); |
7179 | ||
f0486c68 YZ |
7180 | BUG_ON(head->extent_op); |
7181 | if (head->must_insert_reserved) | |
7182 | ret = 1; | |
7183 | ||
7184 | mutex_unlock(&head->mutex); | |
1887be66 | 7185 | btrfs_put_delayed_ref(&head->node); |
f0486c68 | 7186 | return ret; |
1887be66 | 7187 | out: |
d7df2c79 | 7188 | spin_unlock(&head->lock); |
cf93da7b CM |
7189 | |
7190 | out_delayed_unlock: | |
1887be66 CM |
7191 | spin_unlock(&delayed_refs->lock); |
7192 | return 0; | |
7193 | } | |
7194 | ||
f0486c68 YZ |
7195 | void btrfs_free_tree_block(struct btrfs_trans_handle *trans, |
7196 | struct btrfs_root *root, | |
7197 | struct extent_buffer *buf, | |
5581a51a | 7198 | u64 parent, int last_ref) |
f0486c68 | 7199 | { |
0b246afa | 7200 | struct btrfs_fs_info *fs_info = root->fs_info; |
b150a4f1 | 7201 | int pin = 1; |
f0486c68 YZ |
7202 | int ret; |
7203 | ||
7204 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { | |
d7eae340 OS |
7205 | int old_ref_mod, new_ref_mod; |
7206 | ||
7be07912 OS |
7207 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, buf->start, |
7208 | buf->len, parent, | |
0b246afa JM |
7209 | root->root_key.objectid, |
7210 | btrfs_header_level(buf), | |
7be07912 | 7211 | BTRFS_DROP_DELAYED_REF, NULL, |
d7eae340 | 7212 | &old_ref_mod, &new_ref_mod); |
79787eaa | 7213 | BUG_ON(ret); /* -ENOMEM */ |
d7eae340 | 7214 | pin = old_ref_mod >= 0 && new_ref_mod < 0; |
f0486c68 YZ |
7215 | } |
7216 | ||
0a16c7d7 | 7217 | if (last_ref && btrfs_header_generation(buf) == trans->transid) { |
6219872d FM |
7218 | struct btrfs_block_group_cache *cache; |
7219 | ||
f0486c68 | 7220 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { |
2ff7e61e | 7221 | ret = check_ref_cleanup(trans, buf->start); |
f0486c68 | 7222 | if (!ret) |
37be25bc | 7223 | goto out; |
f0486c68 YZ |
7224 | } |
7225 | ||
4da8b76d | 7226 | pin = 0; |
0b246afa | 7227 | cache = btrfs_lookup_block_group(fs_info, buf->start); |
6219872d | 7228 | |
f0486c68 | 7229 | if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) { |
2ff7e61e JM |
7230 | pin_down_extent(fs_info, cache, buf->start, |
7231 | buf->len, 1); | |
6219872d | 7232 | btrfs_put_block_group(cache); |
37be25bc | 7233 | goto out; |
f0486c68 YZ |
7234 | } |
7235 | ||
7236 | WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)); | |
7237 | ||
7238 | btrfs_add_free_space(cache, buf->start, buf->len); | |
4824f1f4 | 7239 | btrfs_free_reserved_bytes(cache, buf->len, 0); |
6219872d | 7240 | btrfs_put_block_group(cache); |
71ff6437 | 7241 | trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len); |
f0486c68 YZ |
7242 | } |
7243 | out: | |
b150a4f1 | 7244 | if (pin) |
0b246afa | 7245 | add_pinned_bytes(fs_info, buf->len, btrfs_header_level(buf), |
b150a4f1 JB |
7246 | root->root_key.objectid); |
7247 | ||
0a16c7d7 OS |
7248 | if (last_ref) { |
7249 | /* | |
7250 | * Deleting the buffer, clear the corrupt flag since it doesn't | |
7251 | * matter anymore. | |
7252 | */ | |
7253 | clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags); | |
7254 | } | |
f0486c68 YZ |
7255 | } |
7256 | ||
79787eaa | 7257 | /* Can return -ENOMEM */ |
2ff7e61e JM |
7258 | int btrfs_free_extent(struct btrfs_trans_handle *trans, |
7259 | struct btrfs_fs_info *fs_info, | |
66d7e7f0 | 7260 | u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid, |
b06c4bf5 | 7261 | u64 owner, u64 offset) |
925baedd | 7262 | { |
d7eae340 | 7263 | int old_ref_mod, new_ref_mod; |
925baedd CM |
7264 | int ret; |
7265 | ||
f5ee5c9a | 7266 | if (btrfs_is_testing(fs_info)) |
faa2dbf0 | 7267 | return 0; |
fccb84c9 | 7268 | |
b150a4f1 | 7269 | |
56bec294 CM |
7270 | /* |
7271 | * tree log blocks never actually go into the extent allocation | |
7272 | * tree, just update pinning info and exit early. | |
56bec294 | 7273 | */ |
5d4f98a2 YZ |
7274 | if (root_objectid == BTRFS_TREE_LOG_OBJECTID) { |
7275 | WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID); | |
b9473439 | 7276 | /* unlocks the pinned mutex */ |
2ff7e61e | 7277 | btrfs_pin_extent(fs_info, bytenr, num_bytes, 1); |
d7eae340 | 7278 | old_ref_mod = new_ref_mod = 0; |
56bec294 | 7279 | ret = 0; |
5d4f98a2 | 7280 | } else if (owner < BTRFS_FIRST_FREE_OBJECTID) { |
66d7e7f0 | 7281 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr, |
7be07912 OS |
7282 | num_bytes, parent, |
7283 | root_objectid, (int)owner, | |
7284 | BTRFS_DROP_DELAYED_REF, NULL, | |
d7eae340 | 7285 | &old_ref_mod, &new_ref_mod); |
5d4f98a2 | 7286 | } else { |
66d7e7f0 | 7287 | ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr, |
7be07912 OS |
7288 | num_bytes, parent, |
7289 | root_objectid, owner, offset, | |
7290 | 0, BTRFS_DROP_DELAYED_REF, | |
d7eae340 | 7291 | &old_ref_mod, &new_ref_mod); |
56bec294 | 7292 | } |
d7eae340 OS |
7293 | |
7294 | if (ret == 0 && old_ref_mod >= 0 && new_ref_mod < 0) | |
7295 | add_pinned_bytes(fs_info, num_bytes, owner, root_objectid); | |
7296 | ||
925baedd CM |
7297 | return ret; |
7298 | } | |
7299 | ||
817d52f8 JB |
7300 | /* |
7301 | * when we wait for progress in the block group caching, its because | |
7302 | * our allocation attempt failed at least once. So, we must sleep | |
7303 | * and let some progress happen before we try again. | |
7304 | * | |
7305 | * This function will sleep at least once waiting for new free space to | |
7306 | * show up, and then it will check the block group free space numbers | |
7307 | * for our min num_bytes. Another option is to have it go ahead | |
7308 | * and look in the rbtree for a free extent of a given size, but this | |
7309 | * is a good start. | |
36cce922 JB |
7310 | * |
7311 | * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using | |
7312 | * any of the information in this block group. | |
817d52f8 | 7313 | */ |
36cce922 | 7314 | static noinline void |
817d52f8 JB |
7315 | wait_block_group_cache_progress(struct btrfs_block_group_cache *cache, |
7316 | u64 num_bytes) | |
7317 | { | |
11833d66 | 7318 | struct btrfs_caching_control *caching_ctl; |
817d52f8 | 7319 | |
11833d66 YZ |
7320 | caching_ctl = get_caching_control(cache); |
7321 | if (!caching_ctl) | |
36cce922 | 7322 | return; |
817d52f8 | 7323 | |
11833d66 | 7324 | wait_event(caching_ctl->wait, block_group_cache_done(cache) || |
34d52cb6 | 7325 | (cache->free_space_ctl->free_space >= num_bytes)); |
11833d66 YZ |
7326 | |
7327 | put_caching_control(caching_ctl); | |
11833d66 YZ |
7328 | } |
7329 | ||
7330 | static noinline int | |
7331 | wait_block_group_cache_done(struct btrfs_block_group_cache *cache) | |
7332 | { | |
7333 | struct btrfs_caching_control *caching_ctl; | |
36cce922 | 7334 | int ret = 0; |
11833d66 YZ |
7335 | |
7336 | caching_ctl = get_caching_control(cache); | |
7337 | if (!caching_ctl) | |
36cce922 | 7338 | return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0; |
11833d66 YZ |
7339 | |
7340 | wait_event(caching_ctl->wait, block_group_cache_done(cache)); | |
36cce922 JB |
7341 | if (cache->cached == BTRFS_CACHE_ERROR) |
7342 | ret = -EIO; | |
11833d66 | 7343 | put_caching_control(caching_ctl); |
36cce922 | 7344 | return ret; |
817d52f8 JB |
7345 | } |
7346 | ||
31e50229 | 7347 | int __get_raid_index(u64 flags) |
b742bb82 | 7348 | { |
7738a53a | 7349 | if (flags & BTRFS_BLOCK_GROUP_RAID10) |
e6ec716f | 7350 | return BTRFS_RAID_RAID10; |
7738a53a | 7351 | else if (flags & BTRFS_BLOCK_GROUP_RAID1) |
e6ec716f | 7352 | return BTRFS_RAID_RAID1; |
7738a53a | 7353 | else if (flags & BTRFS_BLOCK_GROUP_DUP) |
e6ec716f | 7354 | return BTRFS_RAID_DUP; |
7738a53a | 7355 | else if (flags & BTRFS_BLOCK_GROUP_RAID0) |
e6ec716f | 7356 | return BTRFS_RAID_RAID0; |
53b381b3 | 7357 | else if (flags & BTRFS_BLOCK_GROUP_RAID5) |
e942f883 | 7358 | return BTRFS_RAID_RAID5; |
53b381b3 | 7359 | else if (flags & BTRFS_BLOCK_GROUP_RAID6) |
e942f883 | 7360 | return BTRFS_RAID_RAID6; |
7738a53a | 7361 | |
e942f883 | 7362 | return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */ |
b742bb82 YZ |
7363 | } |
7364 | ||
6ab0a202 | 7365 | int get_block_group_index(struct btrfs_block_group_cache *cache) |
7738a53a | 7366 | { |
31e50229 | 7367 | return __get_raid_index(cache->flags); |
7738a53a ID |
7368 | } |
7369 | ||
6ab0a202 JM |
7370 | static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = { |
7371 | [BTRFS_RAID_RAID10] = "raid10", | |
7372 | [BTRFS_RAID_RAID1] = "raid1", | |
7373 | [BTRFS_RAID_DUP] = "dup", | |
7374 | [BTRFS_RAID_RAID0] = "raid0", | |
7375 | [BTRFS_RAID_SINGLE] = "single", | |
7376 | [BTRFS_RAID_RAID5] = "raid5", | |
7377 | [BTRFS_RAID_RAID6] = "raid6", | |
7378 | }; | |
7379 | ||
1b8e5df6 | 7380 | static const char *get_raid_name(enum btrfs_raid_types type) |
6ab0a202 JM |
7381 | { |
7382 | if (type >= BTRFS_NR_RAID_TYPES) | |
7383 | return NULL; | |
7384 | ||
7385 | return btrfs_raid_type_names[type]; | |
7386 | } | |
7387 | ||
817d52f8 | 7388 | enum btrfs_loop_type { |
285ff5af JB |
7389 | LOOP_CACHING_NOWAIT = 0, |
7390 | LOOP_CACHING_WAIT = 1, | |
7391 | LOOP_ALLOC_CHUNK = 2, | |
7392 | LOOP_NO_EMPTY_SIZE = 3, | |
817d52f8 JB |
7393 | }; |
7394 | ||
e570fd27 MX |
7395 | static inline void |
7396 | btrfs_lock_block_group(struct btrfs_block_group_cache *cache, | |
7397 | int delalloc) | |
7398 | { | |
7399 | if (delalloc) | |
7400 | down_read(&cache->data_rwsem); | |
7401 | } | |
7402 | ||
7403 | static inline void | |
7404 | btrfs_grab_block_group(struct btrfs_block_group_cache *cache, | |
7405 | int delalloc) | |
7406 | { | |
7407 | btrfs_get_block_group(cache); | |
7408 | if (delalloc) | |
7409 | down_read(&cache->data_rwsem); | |
7410 | } | |
7411 | ||
7412 | static struct btrfs_block_group_cache * | |
7413 | btrfs_lock_cluster(struct btrfs_block_group_cache *block_group, | |
7414 | struct btrfs_free_cluster *cluster, | |
7415 | int delalloc) | |
7416 | { | |
89771cc9 | 7417 | struct btrfs_block_group_cache *used_bg = NULL; |
6719afdc | 7418 | |
e570fd27 | 7419 | spin_lock(&cluster->refill_lock); |
6719afdc GU |
7420 | while (1) { |
7421 | used_bg = cluster->block_group; | |
7422 | if (!used_bg) | |
7423 | return NULL; | |
7424 | ||
7425 | if (used_bg == block_group) | |
e570fd27 MX |
7426 | return used_bg; |
7427 | ||
6719afdc | 7428 | btrfs_get_block_group(used_bg); |
e570fd27 | 7429 | |
6719afdc GU |
7430 | if (!delalloc) |
7431 | return used_bg; | |
e570fd27 | 7432 | |
6719afdc GU |
7433 | if (down_read_trylock(&used_bg->data_rwsem)) |
7434 | return used_bg; | |
e570fd27 | 7435 | |
6719afdc | 7436 | spin_unlock(&cluster->refill_lock); |
e570fd27 | 7437 | |
e321f8a8 LB |
7438 | /* We should only have one-level nested. */ |
7439 | down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING); | |
e570fd27 | 7440 | |
6719afdc GU |
7441 | spin_lock(&cluster->refill_lock); |
7442 | if (used_bg == cluster->block_group) | |
7443 | return used_bg; | |
e570fd27 | 7444 | |
6719afdc GU |
7445 | up_read(&used_bg->data_rwsem); |
7446 | btrfs_put_block_group(used_bg); | |
7447 | } | |
e570fd27 MX |
7448 | } |
7449 | ||
7450 | static inline void | |
7451 | btrfs_release_block_group(struct btrfs_block_group_cache *cache, | |
7452 | int delalloc) | |
7453 | { | |
7454 | if (delalloc) | |
7455 | up_read(&cache->data_rwsem); | |
7456 | btrfs_put_block_group(cache); | |
7457 | } | |
7458 | ||
fec577fb CM |
7459 | /* |
7460 | * walks the btree of allocated extents and find a hole of a given size. | |
7461 | * The key ins is changed to record the hole: | |
a4820398 | 7462 | * ins->objectid == start position |
62e2749e | 7463 | * ins->flags = BTRFS_EXTENT_ITEM_KEY |
a4820398 | 7464 | * ins->offset == the size of the hole. |
fec577fb | 7465 | * Any available blocks before search_start are skipped. |
a4820398 MX |
7466 | * |
7467 | * If there is no suitable free space, we will record the max size of | |
7468 | * the free space extent currently. | |
fec577fb | 7469 | */ |
87bde3cd | 7470 | static noinline int find_free_extent(struct btrfs_fs_info *fs_info, |
18513091 WX |
7471 | u64 ram_bytes, u64 num_bytes, u64 empty_size, |
7472 | u64 hint_byte, struct btrfs_key *ins, | |
7473 | u64 flags, int delalloc) | |
fec577fb | 7474 | { |
80eb234a | 7475 | int ret = 0; |
0b246afa | 7476 | struct btrfs_root *root = fs_info->extent_root; |
fa9c0d79 | 7477 | struct btrfs_free_cluster *last_ptr = NULL; |
80eb234a | 7478 | struct btrfs_block_group_cache *block_group = NULL; |
81c9ad23 | 7479 | u64 search_start = 0; |
a4820398 | 7480 | u64 max_extent_size = 0; |
c759c4e1 | 7481 | u64 empty_cluster = 0; |
80eb234a | 7482 | struct btrfs_space_info *space_info; |
fa9c0d79 | 7483 | int loop = 0; |
b6919a58 | 7484 | int index = __get_raid_index(flags); |
0a24325e | 7485 | bool failed_cluster_refill = false; |
1cdda9b8 | 7486 | bool failed_alloc = false; |
67377734 | 7487 | bool use_cluster = true; |
60d2adbb | 7488 | bool have_caching_bg = false; |
13a0db5a | 7489 | bool orig_have_caching_bg = false; |
a5e681d9 | 7490 | bool full_search = false; |
fec577fb | 7491 | |
0b246afa | 7492 | WARN_ON(num_bytes < fs_info->sectorsize); |
962a298f | 7493 | ins->type = BTRFS_EXTENT_ITEM_KEY; |
80eb234a JB |
7494 | ins->objectid = 0; |
7495 | ins->offset = 0; | |
b1a4d965 | 7496 | |
71ff6437 | 7497 | trace_find_free_extent(fs_info, num_bytes, empty_size, flags); |
3f7de037 | 7498 | |
0b246afa | 7499 | space_info = __find_space_info(fs_info, flags); |
1b1d1f66 | 7500 | if (!space_info) { |
0b246afa | 7501 | btrfs_err(fs_info, "No space info for %llu", flags); |
1b1d1f66 JB |
7502 | return -ENOSPC; |
7503 | } | |
2552d17e | 7504 | |
67377734 | 7505 | /* |
4f4db217 JB |
7506 | * If our free space is heavily fragmented we may not be able to make |
7507 | * big contiguous allocations, so instead of doing the expensive search | |
7508 | * for free space, simply return ENOSPC with our max_extent_size so we | |
7509 | * can go ahead and search for a more manageable chunk. | |
7510 | * | |
7511 | * If our max_extent_size is large enough for our allocation simply | |
7512 | * disable clustering since we will likely not be able to find enough | |
7513 | * space to create a cluster and induce latency trying. | |
67377734 | 7514 | */ |
4f4db217 JB |
7515 | if (unlikely(space_info->max_extent_size)) { |
7516 | spin_lock(&space_info->lock); | |
7517 | if (space_info->max_extent_size && | |
7518 | num_bytes > space_info->max_extent_size) { | |
7519 | ins->offset = space_info->max_extent_size; | |
7520 | spin_unlock(&space_info->lock); | |
7521 | return -ENOSPC; | |
7522 | } else if (space_info->max_extent_size) { | |
7523 | use_cluster = false; | |
7524 | } | |
7525 | spin_unlock(&space_info->lock); | |
fa9c0d79 | 7526 | } |
0f9dd46c | 7527 | |
2ff7e61e | 7528 | last_ptr = fetch_cluster_info(fs_info, space_info, &empty_cluster); |
239b14b3 | 7529 | if (last_ptr) { |
fa9c0d79 CM |
7530 | spin_lock(&last_ptr->lock); |
7531 | if (last_ptr->block_group) | |
7532 | hint_byte = last_ptr->window_start; | |
c759c4e1 JB |
7533 | if (last_ptr->fragmented) { |
7534 | /* | |
7535 | * We still set window_start so we can keep track of the | |
7536 | * last place we found an allocation to try and save | |
7537 | * some time. | |
7538 | */ | |
7539 | hint_byte = last_ptr->window_start; | |
7540 | use_cluster = false; | |
7541 | } | |
fa9c0d79 | 7542 | spin_unlock(&last_ptr->lock); |
239b14b3 | 7543 | } |
fa9c0d79 | 7544 | |
2ff7e61e | 7545 | search_start = max(search_start, first_logical_byte(fs_info, 0)); |
239b14b3 | 7546 | search_start = max(search_start, hint_byte); |
2552d17e | 7547 | if (search_start == hint_byte) { |
0b246afa | 7548 | block_group = btrfs_lookup_block_group(fs_info, search_start); |
817d52f8 JB |
7549 | /* |
7550 | * we don't want to use the block group if it doesn't match our | |
7551 | * allocation bits, or if its not cached. | |
ccf0e725 JB |
7552 | * |
7553 | * However if we are re-searching with an ideal block group | |
7554 | * picked out then we don't care that the block group is cached. | |
817d52f8 | 7555 | */ |
b6919a58 | 7556 | if (block_group && block_group_bits(block_group, flags) && |
285ff5af | 7557 | block_group->cached != BTRFS_CACHE_NO) { |
2552d17e | 7558 | down_read(&space_info->groups_sem); |
44fb5511 CM |
7559 | if (list_empty(&block_group->list) || |
7560 | block_group->ro) { | |
7561 | /* | |
7562 | * someone is removing this block group, | |
7563 | * we can't jump into the have_block_group | |
7564 | * target because our list pointers are not | |
7565 | * valid | |
7566 | */ | |
7567 | btrfs_put_block_group(block_group); | |
7568 | up_read(&space_info->groups_sem); | |
ccf0e725 | 7569 | } else { |
b742bb82 | 7570 | index = get_block_group_index(block_group); |
e570fd27 | 7571 | btrfs_lock_block_group(block_group, delalloc); |
44fb5511 | 7572 | goto have_block_group; |
ccf0e725 | 7573 | } |
2552d17e | 7574 | } else if (block_group) { |
fa9c0d79 | 7575 | btrfs_put_block_group(block_group); |
2552d17e | 7576 | } |
42e70e7a | 7577 | } |
2552d17e | 7578 | search: |
60d2adbb | 7579 | have_caching_bg = false; |
a5e681d9 JB |
7580 | if (index == 0 || index == __get_raid_index(flags)) |
7581 | full_search = true; | |
80eb234a | 7582 | down_read(&space_info->groups_sem); |
b742bb82 YZ |
7583 | list_for_each_entry(block_group, &space_info->block_groups[index], |
7584 | list) { | |
6226cb0a | 7585 | u64 offset; |
817d52f8 | 7586 | int cached; |
8a1413a2 | 7587 | |
14443937 JM |
7588 | /* If the block group is read-only, we can skip it entirely. */ |
7589 | if (unlikely(block_group->ro)) | |
7590 | continue; | |
7591 | ||
e570fd27 | 7592 | btrfs_grab_block_group(block_group, delalloc); |
2552d17e | 7593 | search_start = block_group->key.objectid; |
42e70e7a | 7594 | |
83a50de9 CM |
7595 | /* |
7596 | * this can happen if we end up cycling through all the | |
7597 | * raid types, but we want to make sure we only allocate | |
7598 | * for the proper type. | |
7599 | */ | |
b6919a58 | 7600 | if (!block_group_bits(block_group, flags)) { |
83a50de9 CM |
7601 | u64 extra = BTRFS_BLOCK_GROUP_DUP | |
7602 | BTRFS_BLOCK_GROUP_RAID1 | | |
53b381b3 DW |
7603 | BTRFS_BLOCK_GROUP_RAID5 | |
7604 | BTRFS_BLOCK_GROUP_RAID6 | | |
83a50de9 CM |
7605 | BTRFS_BLOCK_GROUP_RAID10; |
7606 | ||
7607 | /* | |
7608 | * if they asked for extra copies and this block group | |
7609 | * doesn't provide them, bail. This does allow us to | |
7610 | * fill raid0 from raid1. | |
7611 | */ | |
b6919a58 | 7612 | if ((flags & extra) && !(block_group->flags & extra)) |
83a50de9 CM |
7613 | goto loop; |
7614 | } | |
7615 | ||
2552d17e | 7616 | have_block_group: |
291c7d2f JB |
7617 | cached = block_group_cache_done(block_group); |
7618 | if (unlikely(!cached)) { | |
a5e681d9 | 7619 | have_caching_bg = true; |
f6373bf3 | 7620 | ret = cache_block_group(block_group, 0); |
1d4284bd CM |
7621 | BUG_ON(ret < 0); |
7622 | ret = 0; | |
817d52f8 JB |
7623 | } |
7624 | ||
36cce922 JB |
7625 | if (unlikely(block_group->cached == BTRFS_CACHE_ERROR)) |
7626 | goto loop; | |
0f9dd46c | 7627 | |
0a24325e | 7628 | /* |
062c05c4 AO |
7629 | * Ok we want to try and use the cluster allocator, so |
7630 | * lets look there | |
0a24325e | 7631 | */ |
c759c4e1 | 7632 | if (last_ptr && use_cluster) { |
215a63d1 | 7633 | struct btrfs_block_group_cache *used_block_group; |
8de972b4 | 7634 | unsigned long aligned_cluster; |
fa9c0d79 CM |
7635 | /* |
7636 | * the refill lock keeps out other | |
7637 | * people trying to start a new cluster | |
7638 | */ | |
e570fd27 MX |
7639 | used_block_group = btrfs_lock_cluster(block_group, |
7640 | last_ptr, | |
7641 | delalloc); | |
7642 | if (!used_block_group) | |
44fb5511 | 7643 | goto refill_cluster; |
274bd4fb | 7644 | |
e570fd27 MX |
7645 | if (used_block_group != block_group && |
7646 | (used_block_group->ro || | |
7647 | !block_group_bits(used_block_group, flags))) | |
7648 | goto release_cluster; | |
44fb5511 | 7649 | |
274bd4fb | 7650 | offset = btrfs_alloc_from_cluster(used_block_group, |
a4820398 MX |
7651 | last_ptr, |
7652 | num_bytes, | |
7653 | used_block_group->key.objectid, | |
7654 | &max_extent_size); | |
fa9c0d79 CM |
7655 | if (offset) { |
7656 | /* we have a block, we're done */ | |
7657 | spin_unlock(&last_ptr->refill_lock); | |
71ff6437 | 7658 | trace_btrfs_reserve_extent_cluster(fs_info, |
89d4346a MX |
7659 | used_block_group, |
7660 | search_start, num_bytes); | |
215a63d1 | 7661 | if (used_block_group != block_group) { |
e570fd27 MX |
7662 | btrfs_release_block_group(block_group, |
7663 | delalloc); | |
215a63d1 MX |
7664 | block_group = used_block_group; |
7665 | } | |
fa9c0d79 CM |
7666 | goto checks; |
7667 | } | |
7668 | ||
274bd4fb | 7669 | WARN_ON(last_ptr->block_group != used_block_group); |
e570fd27 | 7670 | release_cluster: |
062c05c4 AO |
7671 | /* If we are on LOOP_NO_EMPTY_SIZE, we can't |
7672 | * set up a new clusters, so lets just skip it | |
7673 | * and let the allocator find whatever block | |
7674 | * it can find. If we reach this point, we | |
7675 | * will have tried the cluster allocator | |
7676 | * plenty of times and not have found | |
7677 | * anything, so we are likely way too | |
7678 | * fragmented for the clustering stuff to find | |
a5f6f719 AO |
7679 | * anything. |
7680 | * | |
7681 | * However, if the cluster is taken from the | |
7682 | * current block group, release the cluster | |
7683 | * first, so that we stand a better chance of | |
7684 | * succeeding in the unclustered | |
7685 | * allocation. */ | |
7686 | if (loop >= LOOP_NO_EMPTY_SIZE && | |
e570fd27 | 7687 | used_block_group != block_group) { |
062c05c4 | 7688 | spin_unlock(&last_ptr->refill_lock); |
e570fd27 MX |
7689 | btrfs_release_block_group(used_block_group, |
7690 | delalloc); | |
062c05c4 AO |
7691 | goto unclustered_alloc; |
7692 | } | |
7693 | ||
fa9c0d79 CM |
7694 | /* |
7695 | * this cluster didn't work out, free it and | |
7696 | * start over | |
7697 | */ | |
7698 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
7699 | ||
e570fd27 MX |
7700 | if (used_block_group != block_group) |
7701 | btrfs_release_block_group(used_block_group, | |
7702 | delalloc); | |
7703 | refill_cluster: | |
a5f6f719 AO |
7704 | if (loop >= LOOP_NO_EMPTY_SIZE) { |
7705 | spin_unlock(&last_ptr->refill_lock); | |
7706 | goto unclustered_alloc; | |
7707 | } | |
7708 | ||
8de972b4 CM |
7709 | aligned_cluster = max_t(unsigned long, |
7710 | empty_cluster + empty_size, | |
7711 | block_group->full_stripe_len); | |
7712 | ||
fa9c0d79 | 7713 | /* allocate a cluster in this block group */ |
2ff7e61e | 7714 | ret = btrfs_find_space_cluster(fs_info, block_group, |
00361589 JB |
7715 | last_ptr, search_start, |
7716 | num_bytes, | |
7717 | aligned_cluster); | |
fa9c0d79 CM |
7718 | if (ret == 0) { |
7719 | /* | |
7720 | * now pull our allocation out of this | |
7721 | * cluster | |
7722 | */ | |
7723 | offset = btrfs_alloc_from_cluster(block_group, | |
a4820398 MX |
7724 | last_ptr, |
7725 | num_bytes, | |
7726 | search_start, | |
7727 | &max_extent_size); | |
fa9c0d79 CM |
7728 | if (offset) { |
7729 | /* we found one, proceed */ | |
7730 | spin_unlock(&last_ptr->refill_lock); | |
71ff6437 | 7731 | trace_btrfs_reserve_extent_cluster(fs_info, |
3f7de037 JB |
7732 | block_group, search_start, |
7733 | num_bytes); | |
fa9c0d79 CM |
7734 | goto checks; |
7735 | } | |
0a24325e JB |
7736 | } else if (!cached && loop > LOOP_CACHING_NOWAIT |
7737 | && !failed_cluster_refill) { | |
817d52f8 JB |
7738 | spin_unlock(&last_ptr->refill_lock); |
7739 | ||
0a24325e | 7740 | failed_cluster_refill = true; |
817d52f8 JB |
7741 | wait_block_group_cache_progress(block_group, |
7742 | num_bytes + empty_cluster + empty_size); | |
7743 | goto have_block_group; | |
fa9c0d79 | 7744 | } |
817d52f8 | 7745 | |
fa9c0d79 CM |
7746 | /* |
7747 | * at this point we either didn't find a cluster | |
7748 | * or we weren't able to allocate a block from our | |
7749 | * cluster. Free the cluster we've been trying | |
7750 | * to use, and go to the next block group | |
7751 | */ | |
0a24325e | 7752 | btrfs_return_cluster_to_free_space(NULL, last_ptr); |
fa9c0d79 | 7753 | spin_unlock(&last_ptr->refill_lock); |
0a24325e | 7754 | goto loop; |
fa9c0d79 CM |
7755 | } |
7756 | ||
062c05c4 | 7757 | unclustered_alloc: |
c759c4e1 JB |
7758 | /* |
7759 | * We are doing an unclustered alloc, set the fragmented flag so | |
7760 | * we don't bother trying to setup a cluster again until we get | |
7761 | * more space. | |
7762 | */ | |
7763 | if (unlikely(last_ptr)) { | |
7764 | spin_lock(&last_ptr->lock); | |
7765 | last_ptr->fragmented = 1; | |
7766 | spin_unlock(&last_ptr->lock); | |
7767 | } | |
0c9b36e0 LB |
7768 | if (cached) { |
7769 | struct btrfs_free_space_ctl *ctl = | |
7770 | block_group->free_space_ctl; | |
7771 | ||
7772 | spin_lock(&ctl->tree_lock); | |
7773 | if (ctl->free_space < | |
7774 | num_bytes + empty_cluster + empty_size) { | |
7775 | if (ctl->free_space > max_extent_size) | |
7776 | max_extent_size = ctl->free_space; | |
7777 | spin_unlock(&ctl->tree_lock); | |
7778 | goto loop; | |
7779 | } | |
7780 | spin_unlock(&ctl->tree_lock); | |
a5f6f719 | 7781 | } |
a5f6f719 | 7782 | |
6226cb0a | 7783 | offset = btrfs_find_space_for_alloc(block_group, search_start, |
a4820398 MX |
7784 | num_bytes, empty_size, |
7785 | &max_extent_size); | |
1cdda9b8 JB |
7786 | /* |
7787 | * If we didn't find a chunk, and we haven't failed on this | |
7788 | * block group before, and this block group is in the middle of | |
7789 | * caching and we are ok with waiting, then go ahead and wait | |
7790 | * for progress to be made, and set failed_alloc to true. | |
7791 | * | |
7792 | * If failed_alloc is true then we've already waited on this | |
7793 | * block group once and should move on to the next block group. | |
7794 | */ | |
7795 | if (!offset && !failed_alloc && !cached && | |
7796 | loop > LOOP_CACHING_NOWAIT) { | |
817d52f8 | 7797 | wait_block_group_cache_progress(block_group, |
1cdda9b8 JB |
7798 | num_bytes + empty_size); |
7799 | failed_alloc = true; | |
817d52f8 | 7800 | goto have_block_group; |
1cdda9b8 JB |
7801 | } else if (!offset) { |
7802 | goto loop; | |
817d52f8 | 7803 | } |
fa9c0d79 | 7804 | checks: |
0b246afa | 7805 | search_start = ALIGN(offset, fs_info->stripesize); |
25179201 | 7806 | |
2552d17e JB |
7807 | /* move on to the next group */ |
7808 | if (search_start + num_bytes > | |
215a63d1 MX |
7809 | block_group->key.objectid + block_group->key.offset) { |
7810 | btrfs_add_free_space(block_group, offset, num_bytes); | |
2552d17e | 7811 | goto loop; |
6226cb0a | 7812 | } |
f5a31e16 | 7813 | |
f0486c68 | 7814 | if (offset < search_start) |
215a63d1 | 7815 | btrfs_add_free_space(block_group, offset, |
f0486c68 YZ |
7816 | search_start - offset); |
7817 | BUG_ON(offset > search_start); | |
2552d17e | 7818 | |
18513091 WX |
7819 | ret = btrfs_add_reserved_bytes(block_group, ram_bytes, |
7820 | num_bytes, delalloc); | |
f0486c68 | 7821 | if (ret == -EAGAIN) { |
215a63d1 | 7822 | btrfs_add_free_space(block_group, offset, num_bytes); |
2552d17e | 7823 | goto loop; |
0f9dd46c | 7824 | } |
9cfa3e34 | 7825 | btrfs_inc_block_group_reservations(block_group); |
0b86a832 | 7826 | |
f0486c68 | 7827 | /* we are all good, lets return */ |
2552d17e JB |
7828 | ins->objectid = search_start; |
7829 | ins->offset = num_bytes; | |
d2fb3437 | 7830 | |
71ff6437 | 7831 | trace_btrfs_reserve_extent(fs_info, block_group, |
3f7de037 | 7832 | search_start, num_bytes); |
e570fd27 | 7833 | btrfs_release_block_group(block_group, delalloc); |
2552d17e JB |
7834 | break; |
7835 | loop: | |
0a24325e | 7836 | failed_cluster_refill = false; |
1cdda9b8 | 7837 | failed_alloc = false; |
b742bb82 | 7838 | BUG_ON(index != get_block_group_index(block_group)); |
e570fd27 | 7839 | btrfs_release_block_group(block_group, delalloc); |
14443937 | 7840 | cond_resched(); |
2552d17e JB |
7841 | } |
7842 | up_read(&space_info->groups_sem); | |
7843 | ||
13a0db5a | 7844 | if ((loop == LOOP_CACHING_NOWAIT) && have_caching_bg |
7845 | && !orig_have_caching_bg) | |
7846 | orig_have_caching_bg = true; | |
7847 | ||
60d2adbb MX |
7848 | if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg) |
7849 | goto search; | |
7850 | ||
b742bb82 YZ |
7851 | if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES) |
7852 | goto search; | |
7853 | ||
285ff5af | 7854 | /* |
ccf0e725 JB |
7855 | * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking |
7856 | * caching kthreads as we move along | |
817d52f8 JB |
7857 | * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching |
7858 | * LOOP_ALLOC_CHUNK, force a chunk allocation and try again | |
7859 | * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try | |
7860 | * again | |
fa9c0d79 | 7861 | */ |
723bda20 | 7862 | if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) { |
b742bb82 | 7863 | index = 0; |
a5e681d9 JB |
7864 | if (loop == LOOP_CACHING_NOWAIT) { |
7865 | /* | |
7866 | * We want to skip the LOOP_CACHING_WAIT step if we | |
01327610 | 7867 | * don't have any uncached bgs and we've already done a |
a5e681d9 JB |
7868 | * full search through. |
7869 | */ | |
13a0db5a | 7870 | if (orig_have_caching_bg || !full_search) |
a5e681d9 JB |
7871 | loop = LOOP_CACHING_WAIT; |
7872 | else | |
7873 | loop = LOOP_ALLOC_CHUNK; | |
7874 | } else { | |
7875 | loop++; | |
7876 | } | |
7877 | ||
817d52f8 | 7878 | if (loop == LOOP_ALLOC_CHUNK) { |
00361589 | 7879 | struct btrfs_trans_handle *trans; |
f017f15f WS |
7880 | int exist = 0; |
7881 | ||
7882 | trans = current->journal_info; | |
7883 | if (trans) | |
7884 | exist = 1; | |
7885 | else | |
7886 | trans = btrfs_join_transaction(root); | |
00361589 | 7887 | |
00361589 JB |
7888 | if (IS_ERR(trans)) { |
7889 | ret = PTR_ERR(trans); | |
7890 | goto out; | |
7891 | } | |
7892 | ||
2ff7e61e | 7893 | ret = do_chunk_alloc(trans, fs_info, flags, |
ea658bad | 7894 | CHUNK_ALLOC_FORCE); |
a5e681d9 JB |
7895 | |
7896 | /* | |
7897 | * If we can't allocate a new chunk we've already looped | |
7898 | * through at least once, move on to the NO_EMPTY_SIZE | |
7899 | * case. | |
7900 | */ | |
7901 | if (ret == -ENOSPC) | |
7902 | loop = LOOP_NO_EMPTY_SIZE; | |
7903 | ||
ea658bad JB |
7904 | /* |
7905 | * Do not bail out on ENOSPC since we | |
7906 | * can do more things. | |
7907 | */ | |
00361589 | 7908 | if (ret < 0 && ret != -ENOSPC) |
66642832 | 7909 | btrfs_abort_transaction(trans, ret); |
00361589 JB |
7910 | else |
7911 | ret = 0; | |
f017f15f | 7912 | if (!exist) |
3a45bb20 | 7913 | btrfs_end_transaction(trans); |
00361589 | 7914 | if (ret) |
ea658bad | 7915 | goto out; |
2552d17e JB |
7916 | } |
7917 | ||
723bda20 | 7918 | if (loop == LOOP_NO_EMPTY_SIZE) { |
a5e681d9 JB |
7919 | /* |
7920 | * Don't loop again if we already have no empty_size and | |
7921 | * no empty_cluster. | |
7922 | */ | |
7923 | if (empty_size == 0 && | |
7924 | empty_cluster == 0) { | |
7925 | ret = -ENOSPC; | |
7926 | goto out; | |
7927 | } | |
723bda20 JB |
7928 | empty_size = 0; |
7929 | empty_cluster = 0; | |
fa9c0d79 | 7930 | } |
723bda20 JB |
7931 | |
7932 | goto search; | |
2552d17e JB |
7933 | } else if (!ins->objectid) { |
7934 | ret = -ENOSPC; | |
d82a6f1d | 7935 | } else if (ins->objectid) { |
c759c4e1 JB |
7936 | if (!use_cluster && last_ptr) { |
7937 | spin_lock(&last_ptr->lock); | |
7938 | last_ptr->window_start = ins->objectid; | |
7939 | spin_unlock(&last_ptr->lock); | |
7940 | } | |
80eb234a | 7941 | ret = 0; |
be744175 | 7942 | } |
79787eaa | 7943 | out: |
4f4db217 JB |
7944 | if (ret == -ENOSPC) { |
7945 | spin_lock(&space_info->lock); | |
7946 | space_info->max_extent_size = max_extent_size; | |
7947 | spin_unlock(&space_info->lock); | |
a4820398 | 7948 | ins->offset = max_extent_size; |
4f4db217 | 7949 | } |
0f70abe2 | 7950 | return ret; |
fec577fb | 7951 | } |
ec44a35c | 7952 | |
ab8d0fc4 JM |
7953 | static void dump_space_info(struct btrfs_fs_info *fs_info, |
7954 | struct btrfs_space_info *info, u64 bytes, | |
9ed74f2d | 7955 | int dump_block_groups) |
0f9dd46c JB |
7956 | { |
7957 | struct btrfs_block_group_cache *cache; | |
b742bb82 | 7958 | int index = 0; |
0f9dd46c | 7959 | |
9ed74f2d | 7960 | spin_lock(&info->lock); |
ab8d0fc4 JM |
7961 | btrfs_info(fs_info, "space_info %llu has %llu free, is %sfull", |
7962 | info->flags, | |
4136135b LB |
7963 | info->total_bytes - btrfs_space_info_used(info, true), |
7964 | info->full ? "" : "not "); | |
ab8d0fc4 JM |
7965 | btrfs_info(fs_info, |
7966 | "space_info total=%llu, used=%llu, pinned=%llu, reserved=%llu, may_use=%llu, readonly=%llu", | |
7967 | info->total_bytes, info->bytes_used, info->bytes_pinned, | |
7968 | info->bytes_reserved, info->bytes_may_use, | |
7969 | info->bytes_readonly); | |
9ed74f2d JB |
7970 | spin_unlock(&info->lock); |
7971 | ||
7972 | if (!dump_block_groups) | |
7973 | return; | |
0f9dd46c | 7974 | |
80eb234a | 7975 | down_read(&info->groups_sem); |
b742bb82 YZ |
7976 | again: |
7977 | list_for_each_entry(cache, &info->block_groups[index], list) { | |
0f9dd46c | 7978 | spin_lock(&cache->lock); |
ab8d0fc4 JM |
7979 | btrfs_info(fs_info, |
7980 | "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s", | |
7981 | cache->key.objectid, cache->key.offset, | |
7982 | btrfs_block_group_used(&cache->item), cache->pinned, | |
7983 | cache->reserved, cache->ro ? "[readonly]" : ""); | |
0f9dd46c JB |
7984 | btrfs_dump_free_space(cache, bytes); |
7985 | spin_unlock(&cache->lock); | |
7986 | } | |
b742bb82 YZ |
7987 | if (++index < BTRFS_NR_RAID_TYPES) |
7988 | goto again; | |
80eb234a | 7989 | up_read(&info->groups_sem); |
0f9dd46c | 7990 | } |
e8569813 | 7991 | |
18513091 | 7992 | int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes, |
11833d66 YZ |
7993 | u64 num_bytes, u64 min_alloc_size, |
7994 | u64 empty_size, u64 hint_byte, | |
e570fd27 | 7995 | struct btrfs_key *ins, int is_data, int delalloc) |
fec577fb | 7996 | { |
ab8d0fc4 | 7997 | struct btrfs_fs_info *fs_info = root->fs_info; |
36af4e07 | 7998 | bool final_tried = num_bytes == min_alloc_size; |
b6919a58 | 7999 | u64 flags; |
fec577fb | 8000 | int ret; |
925baedd | 8001 | |
1b86826d | 8002 | flags = get_alloc_profile_by_root(root, is_data); |
98d20f67 | 8003 | again: |
0b246afa | 8004 | WARN_ON(num_bytes < fs_info->sectorsize); |
87bde3cd | 8005 | ret = find_free_extent(fs_info, ram_bytes, num_bytes, empty_size, |
18513091 | 8006 | hint_byte, ins, flags, delalloc); |
9cfa3e34 | 8007 | if (!ret && !is_data) { |
ab8d0fc4 | 8008 | btrfs_dec_block_group_reservations(fs_info, ins->objectid); |
9cfa3e34 | 8009 | } else if (ret == -ENOSPC) { |
a4820398 MX |
8010 | if (!final_tried && ins->offset) { |
8011 | num_bytes = min(num_bytes >> 1, ins->offset); | |
da17066c | 8012 | num_bytes = round_down(num_bytes, |
0b246afa | 8013 | fs_info->sectorsize); |
9e622d6b | 8014 | num_bytes = max(num_bytes, min_alloc_size); |
18513091 | 8015 | ram_bytes = num_bytes; |
9e622d6b MX |
8016 | if (num_bytes == min_alloc_size) |
8017 | final_tried = true; | |
8018 | goto again; | |
ab8d0fc4 | 8019 | } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
9e622d6b MX |
8020 | struct btrfs_space_info *sinfo; |
8021 | ||
ab8d0fc4 | 8022 | sinfo = __find_space_info(fs_info, flags); |
0b246afa | 8023 | btrfs_err(fs_info, |
5d163e0e JM |
8024 | "allocation failed flags %llu, wanted %llu", |
8025 | flags, num_bytes); | |
53804280 | 8026 | if (sinfo) |
ab8d0fc4 | 8027 | dump_space_info(fs_info, sinfo, num_bytes, 1); |
9e622d6b | 8028 | } |
925baedd | 8029 | } |
0f9dd46c JB |
8030 | |
8031 | return ret; | |
e6dcd2dc CM |
8032 | } |
8033 | ||
2ff7e61e | 8034 | static int __btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info, |
e570fd27 MX |
8035 | u64 start, u64 len, |
8036 | int pin, int delalloc) | |
65b51a00 | 8037 | { |
0f9dd46c | 8038 | struct btrfs_block_group_cache *cache; |
1f3c79a2 | 8039 | int ret = 0; |
0f9dd46c | 8040 | |
0b246afa | 8041 | cache = btrfs_lookup_block_group(fs_info, start); |
0f9dd46c | 8042 | if (!cache) { |
0b246afa JM |
8043 | btrfs_err(fs_info, "Unable to find block group for %llu", |
8044 | start); | |
0f9dd46c JB |
8045 | return -ENOSPC; |
8046 | } | |
1f3c79a2 | 8047 | |
e688b725 | 8048 | if (pin) |
2ff7e61e | 8049 | pin_down_extent(fs_info, cache, start, len, 1); |
e688b725 | 8050 | else { |
0b246afa | 8051 | if (btrfs_test_opt(fs_info, DISCARD)) |
2ff7e61e | 8052 | ret = btrfs_discard_extent(fs_info, start, len, NULL); |
e688b725 | 8053 | btrfs_add_free_space(cache, start, len); |
4824f1f4 | 8054 | btrfs_free_reserved_bytes(cache, len, delalloc); |
71ff6437 | 8055 | trace_btrfs_reserved_extent_free(fs_info, start, len); |
e688b725 | 8056 | } |
31193213 | 8057 | |
fa9c0d79 | 8058 | btrfs_put_block_group(cache); |
e6dcd2dc CM |
8059 | return ret; |
8060 | } | |
8061 | ||
2ff7e61e | 8062 | int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info, |
e570fd27 | 8063 | u64 start, u64 len, int delalloc) |
e688b725 | 8064 | { |
2ff7e61e | 8065 | return __btrfs_free_reserved_extent(fs_info, start, len, 0, delalloc); |
e688b725 CM |
8066 | } |
8067 | ||
2ff7e61e | 8068 | int btrfs_free_and_pin_reserved_extent(struct btrfs_fs_info *fs_info, |
e688b725 CM |
8069 | u64 start, u64 len) |
8070 | { | |
2ff7e61e | 8071 | return __btrfs_free_reserved_extent(fs_info, start, len, 1, 0); |
e688b725 CM |
8072 | } |
8073 | ||
5d4f98a2 | 8074 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, |
2ff7e61e | 8075 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
8076 | u64 parent, u64 root_objectid, |
8077 | u64 flags, u64 owner, u64 offset, | |
8078 | struct btrfs_key *ins, int ref_mod) | |
e6dcd2dc CM |
8079 | { |
8080 | int ret; | |
e6dcd2dc | 8081 | struct btrfs_extent_item *extent_item; |
5d4f98a2 | 8082 | struct btrfs_extent_inline_ref *iref; |
e6dcd2dc | 8083 | struct btrfs_path *path; |
5d4f98a2 YZ |
8084 | struct extent_buffer *leaf; |
8085 | int type; | |
8086 | u32 size; | |
26b8003f | 8087 | |
5d4f98a2 YZ |
8088 | if (parent > 0) |
8089 | type = BTRFS_SHARED_DATA_REF_KEY; | |
8090 | else | |
8091 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
58176a96 | 8092 | |
5d4f98a2 | 8093 | size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type); |
7bb86316 CM |
8094 | |
8095 | path = btrfs_alloc_path(); | |
db5b493a TI |
8096 | if (!path) |
8097 | return -ENOMEM; | |
47e4bb98 | 8098 | |
b9473439 | 8099 | path->leave_spinning = 1; |
5d4f98a2 YZ |
8100 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, |
8101 | ins, size); | |
79787eaa JM |
8102 | if (ret) { |
8103 | btrfs_free_path(path); | |
8104 | return ret; | |
8105 | } | |
0f9dd46c | 8106 | |
5d4f98a2 YZ |
8107 | leaf = path->nodes[0]; |
8108 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
47e4bb98 | 8109 | struct btrfs_extent_item); |
5d4f98a2 YZ |
8110 | btrfs_set_extent_refs(leaf, extent_item, ref_mod); |
8111 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
8112 | btrfs_set_extent_flags(leaf, extent_item, | |
8113 | flags | BTRFS_EXTENT_FLAG_DATA); | |
8114 | ||
8115 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
8116 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
8117 | if (parent > 0) { | |
8118 | struct btrfs_shared_data_ref *ref; | |
8119 | ref = (struct btrfs_shared_data_ref *)(iref + 1); | |
8120 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
8121 | btrfs_set_shared_data_ref_count(leaf, ref, ref_mod); | |
8122 | } else { | |
8123 | struct btrfs_extent_data_ref *ref; | |
8124 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
8125 | btrfs_set_extent_data_ref_root(leaf, ref, root_objectid); | |
8126 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
8127 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
8128 | btrfs_set_extent_data_ref_count(leaf, ref, ref_mod); | |
8129 | } | |
47e4bb98 CM |
8130 | |
8131 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
7bb86316 | 8132 | btrfs_free_path(path); |
f510cfec | 8133 | |
1e144fb8 OS |
8134 | ret = remove_from_free_space_tree(trans, fs_info, ins->objectid, |
8135 | ins->offset); | |
8136 | if (ret) | |
8137 | return ret; | |
8138 | ||
6202df69 | 8139 | ret = update_block_group(trans, fs_info, ins->objectid, ins->offset, 1); |
79787eaa | 8140 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb | 8141 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
c1c9ff7c | 8142 | ins->objectid, ins->offset); |
f5947066 CM |
8143 | BUG(); |
8144 | } | |
71ff6437 | 8145 | trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid, ins->offset); |
e6dcd2dc CM |
8146 | return ret; |
8147 | } | |
8148 | ||
5d4f98a2 | 8149 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, |
2ff7e61e | 8150 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
8151 | u64 parent, u64 root_objectid, |
8152 | u64 flags, struct btrfs_disk_key *key, | |
b06c4bf5 | 8153 | int level, struct btrfs_key *ins) |
e6dcd2dc CM |
8154 | { |
8155 | int ret; | |
5d4f98a2 YZ |
8156 | struct btrfs_extent_item *extent_item; |
8157 | struct btrfs_tree_block_info *block_info; | |
8158 | struct btrfs_extent_inline_ref *iref; | |
8159 | struct btrfs_path *path; | |
8160 | struct extent_buffer *leaf; | |
3173a18f | 8161 | u32 size = sizeof(*extent_item) + sizeof(*iref); |
fcebe456 | 8162 | u64 num_bytes = ins->offset; |
0b246afa | 8163 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); |
3173a18f JB |
8164 | |
8165 | if (!skinny_metadata) | |
8166 | size += sizeof(*block_info); | |
1c2308f8 | 8167 | |
5d4f98a2 | 8168 | path = btrfs_alloc_path(); |
857cc2fc | 8169 | if (!path) { |
2ff7e61e | 8170 | btrfs_free_and_pin_reserved_extent(fs_info, ins->objectid, |
0b246afa | 8171 | fs_info->nodesize); |
d8926bb3 | 8172 | return -ENOMEM; |
857cc2fc | 8173 | } |
56bec294 | 8174 | |
5d4f98a2 YZ |
8175 | path->leave_spinning = 1; |
8176 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, | |
8177 | ins, size); | |
79787eaa | 8178 | if (ret) { |
dd825259 | 8179 | btrfs_free_path(path); |
2ff7e61e | 8180 | btrfs_free_and_pin_reserved_extent(fs_info, ins->objectid, |
0b246afa | 8181 | fs_info->nodesize); |
79787eaa JM |
8182 | return ret; |
8183 | } | |
5d4f98a2 YZ |
8184 | |
8185 | leaf = path->nodes[0]; | |
8186 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
8187 | struct btrfs_extent_item); | |
8188 | btrfs_set_extent_refs(leaf, extent_item, 1); | |
8189 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
8190 | btrfs_set_extent_flags(leaf, extent_item, | |
8191 | flags | BTRFS_EXTENT_FLAG_TREE_BLOCK); | |
5d4f98a2 | 8192 | |
3173a18f JB |
8193 | if (skinny_metadata) { |
8194 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
0b246afa | 8195 | num_bytes = fs_info->nodesize; |
3173a18f JB |
8196 | } else { |
8197 | block_info = (struct btrfs_tree_block_info *)(extent_item + 1); | |
8198 | btrfs_set_tree_block_key(leaf, block_info, key); | |
8199 | btrfs_set_tree_block_level(leaf, block_info, level); | |
8200 | iref = (struct btrfs_extent_inline_ref *)(block_info + 1); | |
8201 | } | |
5d4f98a2 | 8202 | |
5d4f98a2 YZ |
8203 | if (parent > 0) { |
8204 | BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); | |
8205 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
8206 | BTRFS_SHARED_BLOCK_REF_KEY); | |
8207 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
8208 | } else { | |
8209 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
8210 | BTRFS_TREE_BLOCK_REF_KEY); | |
8211 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
8212 | } | |
8213 | ||
8214 | btrfs_mark_buffer_dirty(leaf); | |
8215 | btrfs_free_path(path); | |
8216 | ||
1e144fb8 OS |
8217 | ret = remove_from_free_space_tree(trans, fs_info, ins->objectid, |
8218 | num_bytes); | |
8219 | if (ret) | |
8220 | return ret; | |
8221 | ||
6202df69 JM |
8222 | ret = update_block_group(trans, fs_info, ins->objectid, |
8223 | fs_info->nodesize, 1); | |
79787eaa | 8224 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb | 8225 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
c1c9ff7c | 8226 | ins->objectid, ins->offset); |
5d4f98a2 YZ |
8227 | BUG(); |
8228 | } | |
0be5dc67 | 8229 | |
71ff6437 | 8230 | trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid, |
0b246afa | 8231 | fs_info->nodesize); |
5d4f98a2 YZ |
8232 | return ret; |
8233 | } | |
8234 | ||
8235 | int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
5d4f98a2 | 8236 | u64 root_objectid, u64 owner, |
5846a3c2 QW |
8237 | u64 offset, u64 ram_bytes, |
8238 | struct btrfs_key *ins) | |
5d4f98a2 | 8239 | { |
2ff7e61e | 8240 | struct btrfs_fs_info *fs_info = trans->fs_info; |
5d4f98a2 YZ |
8241 | int ret; |
8242 | ||
8243 | BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
8244 | ||
0b246afa | 8245 | ret = btrfs_add_delayed_data_ref(fs_info, trans, ins->objectid, |
7be07912 OS |
8246 | ins->offset, 0, root_objectid, owner, |
8247 | offset, ram_bytes, | |
8248 | BTRFS_ADD_DELAYED_EXTENT, NULL, NULL); | |
e6dcd2dc CM |
8249 | return ret; |
8250 | } | |
e02119d5 CM |
8251 | |
8252 | /* | |
8253 | * this is used by the tree logging recovery code. It records that | |
8254 | * an extent has been allocated and makes sure to clear the free | |
8255 | * space cache bits as well | |
8256 | */ | |
5d4f98a2 | 8257 | int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans, |
2ff7e61e | 8258 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
8259 | u64 root_objectid, u64 owner, u64 offset, |
8260 | struct btrfs_key *ins) | |
e02119d5 CM |
8261 | { |
8262 | int ret; | |
8263 | struct btrfs_block_group_cache *block_group; | |
ed7a6948 | 8264 | struct btrfs_space_info *space_info; |
11833d66 | 8265 | |
8c2a1a30 JB |
8266 | /* |
8267 | * Mixed block groups will exclude before processing the log so we only | |
01327610 | 8268 | * need to do the exclude dance if this fs isn't mixed. |
8c2a1a30 | 8269 | */ |
0b246afa | 8270 | if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) { |
2ff7e61e JM |
8271 | ret = __exclude_logged_extent(fs_info, ins->objectid, |
8272 | ins->offset); | |
b50c6e25 | 8273 | if (ret) |
8c2a1a30 | 8274 | return ret; |
11833d66 YZ |
8275 | } |
8276 | ||
0b246afa | 8277 | block_group = btrfs_lookup_block_group(fs_info, ins->objectid); |
8c2a1a30 JB |
8278 | if (!block_group) |
8279 | return -EINVAL; | |
8280 | ||
ed7a6948 WX |
8281 | space_info = block_group->space_info; |
8282 | spin_lock(&space_info->lock); | |
8283 | spin_lock(&block_group->lock); | |
8284 | space_info->bytes_reserved += ins->offset; | |
8285 | block_group->reserved += ins->offset; | |
8286 | spin_unlock(&block_group->lock); | |
8287 | spin_unlock(&space_info->lock); | |
8288 | ||
2ff7e61e | 8289 | ret = alloc_reserved_file_extent(trans, fs_info, 0, root_objectid, |
5d4f98a2 | 8290 | 0, owner, offset, ins, 1); |
b50c6e25 | 8291 | btrfs_put_block_group(block_group); |
e02119d5 CM |
8292 | return ret; |
8293 | } | |
8294 | ||
48a3b636 ES |
8295 | static struct extent_buffer * |
8296 | btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
fe864576 | 8297 | u64 bytenr, int level) |
65b51a00 | 8298 | { |
0b246afa | 8299 | struct btrfs_fs_info *fs_info = root->fs_info; |
65b51a00 CM |
8300 | struct extent_buffer *buf; |
8301 | ||
2ff7e61e | 8302 | buf = btrfs_find_create_tree_block(fs_info, bytenr); |
c871b0f2 LB |
8303 | if (IS_ERR(buf)) |
8304 | return buf; | |
8305 | ||
65b51a00 | 8306 | btrfs_set_header_generation(buf, trans->transid); |
85d4e461 | 8307 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level); |
65b51a00 | 8308 | btrfs_tree_lock(buf); |
7c302b49 | 8309 | clean_tree_block(fs_info, buf); |
3083ee2e | 8310 | clear_bit(EXTENT_BUFFER_STALE, &buf->bflags); |
b4ce94de CM |
8311 | |
8312 | btrfs_set_lock_blocking(buf); | |
4db8c528 | 8313 | set_extent_buffer_uptodate(buf); |
b4ce94de | 8314 | |
d0c803c4 | 8315 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { |
656f30db | 8316 | buf->log_index = root->log_transid % 2; |
8cef4e16 YZ |
8317 | /* |
8318 | * we allow two log transactions at a time, use different | |
8319 | * EXENT bit to differentiate dirty pages. | |
8320 | */ | |
656f30db | 8321 | if (buf->log_index == 0) |
8cef4e16 YZ |
8322 | set_extent_dirty(&root->dirty_log_pages, buf->start, |
8323 | buf->start + buf->len - 1, GFP_NOFS); | |
8324 | else | |
8325 | set_extent_new(&root->dirty_log_pages, buf->start, | |
3744dbeb | 8326 | buf->start + buf->len - 1); |
d0c803c4 | 8327 | } else { |
656f30db | 8328 | buf->log_index = -1; |
d0c803c4 | 8329 | set_extent_dirty(&trans->transaction->dirty_pages, buf->start, |
65b51a00 | 8330 | buf->start + buf->len - 1, GFP_NOFS); |
d0c803c4 | 8331 | } |
64c12921 | 8332 | trans->dirty = true; |
b4ce94de | 8333 | /* this returns a buffer locked for blocking */ |
65b51a00 CM |
8334 | return buf; |
8335 | } | |
8336 | ||
f0486c68 YZ |
8337 | static struct btrfs_block_rsv * |
8338 | use_block_rsv(struct btrfs_trans_handle *trans, | |
8339 | struct btrfs_root *root, u32 blocksize) | |
8340 | { | |
0b246afa | 8341 | struct btrfs_fs_info *fs_info = root->fs_info; |
f0486c68 | 8342 | struct btrfs_block_rsv *block_rsv; |
0b246afa | 8343 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; |
f0486c68 | 8344 | int ret; |
d88033db | 8345 | bool global_updated = false; |
f0486c68 YZ |
8346 | |
8347 | block_rsv = get_block_rsv(trans, root); | |
8348 | ||
b586b323 MX |
8349 | if (unlikely(block_rsv->size == 0)) |
8350 | goto try_reserve; | |
d88033db | 8351 | again: |
f0486c68 YZ |
8352 | ret = block_rsv_use_bytes(block_rsv, blocksize); |
8353 | if (!ret) | |
8354 | return block_rsv; | |
8355 | ||
b586b323 MX |
8356 | if (block_rsv->failfast) |
8357 | return ERR_PTR(ret); | |
8358 | ||
d88033db MX |
8359 | if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) { |
8360 | global_updated = true; | |
0b246afa | 8361 | update_global_block_rsv(fs_info); |
d88033db MX |
8362 | goto again; |
8363 | } | |
8364 | ||
0b246afa | 8365 | if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
b586b323 MX |
8366 | static DEFINE_RATELIMIT_STATE(_rs, |
8367 | DEFAULT_RATELIMIT_INTERVAL * 10, | |
8368 | /*DEFAULT_RATELIMIT_BURST*/ 1); | |
8369 | if (__ratelimit(&_rs)) | |
8370 | WARN(1, KERN_DEBUG | |
efe120a0 | 8371 | "BTRFS: block rsv returned %d\n", ret); |
b586b323 MX |
8372 | } |
8373 | try_reserve: | |
8374 | ret = reserve_metadata_bytes(root, block_rsv, blocksize, | |
8375 | BTRFS_RESERVE_NO_FLUSH); | |
8376 | if (!ret) | |
8377 | return block_rsv; | |
8378 | /* | |
8379 | * If we couldn't reserve metadata bytes try and use some from | |
5881cfc9 MX |
8380 | * the global reserve if its space type is the same as the global |
8381 | * reservation. | |
b586b323 | 8382 | */ |
5881cfc9 MX |
8383 | if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL && |
8384 | block_rsv->space_info == global_rsv->space_info) { | |
b586b323 MX |
8385 | ret = block_rsv_use_bytes(global_rsv, blocksize); |
8386 | if (!ret) | |
8387 | return global_rsv; | |
8388 | } | |
8389 | return ERR_PTR(ret); | |
f0486c68 YZ |
8390 | } |
8391 | ||
8c2a3ca2 JB |
8392 | static void unuse_block_rsv(struct btrfs_fs_info *fs_info, |
8393 | struct btrfs_block_rsv *block_rsv, u32 blocksize) | |
f0486c68 YZ |
8394 | { |
8395 | block_rsv_add_bytes(block_rsv, blocksize, 0); | |
8c2a3ca2 | 8396 | block_rsv_release_bytes(fs_info, block_rsv, NULL, 0); |
f0486c68 YZ |
8397 | } |
8398 | ||
fec577fb | 8399 | /* |
f0486c68 | 8400 | * finds a free extent and does all the dirty work required for allocation |
67b7859e | 8401 | * returns the tree buffer or an ERR_PTR on error. |
fec577fb | 8402 | */ |
4d75f8a9 | 8403 | struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans, |
310712b2 OS |
8404 | struct btrfs_root *root, |
8405 | u64 parent, u64 root_objectid, | |
8406 | const struct btrfs_disk_key *key, | |
8407 | int level, u64 hint, | |
8408 | u64 empty_size) | |
fec577fb | 8409 | { |
0b246afa | 8410 | struct btrfs_fs_info *fs_info = root->fs_info; |
e2fa7227 | 8411 | struct btrfs_key ins; |
f0486c68 | 8412 | struct btrfs_block_rsv *block_rsv; |
5f39d397 | 8413 | struct extent_buffer *buf; |
67b7859e | 8414 | struct btrfs_delayed_extent_op *extent_op; |
f0486c68 YZ |
8415 | u64 flags = 0; |
8416 | int ret; | |
0b246afa JM |
8417 | u32 blocksize = fs_info->nodesize; |
8418 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); | |
fec577fb | 8419 | |
05653ef3 | 8420 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
0b246afa | 8421 | if (btrfs_is_testing(fs_info)) { |
faa2dbf0 | 8422 | buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr, |
fe864576 | 8423 | level); |
faa2dbf0 JB |
8424 | if (!IS_ERR(buf)) |
8425 | root->alloc_bytenr += blocksize; | |
8426 | return buf; | |
8427 | } | |
05653ef3 | 8428 | #endif |
fccb84c9 | 8429 | |
f0486c68 YZ |
8430 | block_rsv = use_block_rsv(trans, root, blocksize); |
8431 | if (IS_ERR(block_rsv)) | |
8432 | return ERR_CAST(block_rsv); | |
8433 | ||
18513091 | 8434 | ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize, |
e570fd27 | 8435 | empty_size, hint, &ins, 0, 0); |
67b7859e OS |
8436 | if (ret) |
8437 | goto out_unuse; | |
55c69072 | 8438 | |
fe864576 | 8439 | buf = btrfs_init_new_buffer(trans, root, ins.objectid, level); |
67b7859e OS |
8440 | if (IS_ERR(buf)) { |
8441 | ret = PTR_ERR(buf); | |
8442 | goto out_free_reserved; | |
8443 | } | |
f0486c68 YZ |
8444 | |
8445 | if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
8446 | if (parent == 0) | |
8447 | parent = ins.objectid; | |
8448 | flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
8449 | } else | |
8450 | BUG_ON(parent > 0); | |
8451 | ||
8452 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) { | |
78a6184a | 8453 | extent_op = btrfs_alloc_delayed_extent_op(); |
67b7859e OS |
8454 | if (!extent_op) { |
8455 | ret = -ENOMEM; | |
8456 | goto out_free_buf; | |
8457 | } | |
f0486c68 YZ |
8458 | if (key) |
8459 | memcpy(&extent_op->key, key, sizeof(extent_op->key)); | |
8460 | else | |
8461 | memset(&extent_op->key, 0, sizeof(extent_op->key)); | |
8462 | extent_op->flags_to_set = flags; | |
35b3ad50 DS |
8463 | extent_op->update_key = skinny_metadata ? false : true; |
8464 | extent_op->update_flags = true; | |
8465 | extent_op->is_data = false; | |
b1c79e09 | 8466 | extent_op->level = level; |
f0486c68 | 8467 | |
7be07912 OS |
8468 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, ins.objectid, |
8469 | ins.offset, parent, | |
8470 | root_objectid, level, | |
67b7859e | 8471 | BTRFS_ADD_DELAYED_EXTENT, |
7be07912 | 8472 | extent_op, NULL, NULL); |
67b7859e OS |
8473 | if (ret) |
8474 | goto out_free_delayed; | |
f0486c68 | 8475 | } |
fec577fb | 8476 | return buf; |
67b7859e OS |
8477 | |
8478 | out_free_delayed: | |
8479 | btrfs_free_delayed_extent_op(extent_op); | |
8480 | out_free_buf: | |
8481 | free_extent_buffer(buf); | |
8482 | out_free_reserved: | |
2ff7e61e | 8483 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0); |
67b7859e | 8484 | out_unuse: |
0b246afa | 8485 | unuse_block_rsv(fs_info, block_rsv, blocksize); |
67b7859e | 8486 | return ERR_PTR(ret); |
fec577fb | 8487 | } |
a28ec197 | 8488 | |
2c47e605 YZ |
8489 | struct walk_control { |
8490 | u64 refs[BTRFS_MAX_LEVEL]; | |
8491 | u64 flags[BTRFS_MAX_LEVEL]; | |
8492 | struct btrfs_key update_progress; | |
8493 | int stage; | |
8494 | int level; | |
8495 | int shared_level; | |
8496 | int update_ref; | |
8497 | int keep_locks; | |
1c4850e2 YZ |
8498 | int reada_slot; |
8499 | int reada_count; | |
66d7e7f0 | 8500 | int for_reloc; |
2c47e605 YZ |
8501 | }; |
8502 | ||
8503 | #define DROP_REFERENCE 1 | |
8504 | #define UPDATE_BACKREF 2 | |
8505 | ||
1c4850e2 YZ |
8506 | static noinline void reada_walk_down(struct btrfs_trans_handle *trans, |
8507 | struct btrfs_root *root, | |
8508 | struct walk_control *wc, | |
8509 | struct btrfs_path *path) | |
6407bf6d | 8510 | { |
0b246afa | 8511 | struct btrfs_fs_info *fs_info = root->fs_info; |
1c4850e2 YZ |
8512 | u64 bytenr; |
8513 | u64 generation; | |
8514 | u64 refs; | |
94fcca9f | 8515 | u64 flags; |
5d4f98a2 | 8516 | u32 nritems; |
1c4850e2 YZ |
8517 | struct btrfs_key key; |
8518 | struct extent_buffer *eb; | |
6407bf6d | 8519 | int ret; |
1c4850e2 YZ |
8520 | int slot; |
8521 | int nread = 0; | |
6407bf6d | 8522 | |
1c4850e2 YZ |
8523 | if (path->slots[wc->level] < wc->reada_slot) { |
8524 | wc->reada_count = wc->reada_count * 2 / 3; | |
8525 | wc->reada_count = max(wc->reada_count, 2); | |
8526 | } else { | |
8527 | wc->reada_count = wc->reada_count * 3 / 2; | |
8528 | wc->reada_count = min_t(int, wc->reada_count, | |
0b246afa | 8529 | BTRFS_NODEPTRS_PER_BLOCK(fs_info)); |
1c4850e2 | 8530 | } |
7bb86316 | 8531 | |
1c4850e2 YZ |
8532 | eb = path->nodes[wc->level]; |
8533 | nritems = btrfs_header_nritems(eb); | |
bd56b302 | 8534 | |
1c4850e2 YZ |
8535 | for (slot = path->slots[wc->level]; slot < nritems; slot++) { |
8536 | if (nread >= wc->reada_count) | |
8537 | break; | |
bd56b302 | 8538 | |
2dd3e67b | 8539 | cond_resched(); |
1c4850e2 YZ |
8540 | bytenr = btrfs_node_blockptr(eb, slot); |
8541 | generation = btrfs_node_ptr_generation(eb, slot); | |
2dd3e67b | 8542 | |
1c4850e2 YZ |
8543 | if (slot == path->slots[wc->level]) |
8544 | goto reada; | |
5d4f98a2 | 8545 | |
1c4850e2 YZ |
8546 | if (wc->stage == UPDATE_BACKREF && |
8547 | generation <= root->root_key.offset) | |
bd56b302 CM |
8548 | continue; |
8549 | ||
94fcca9f | 8550 | /* We don't lock the tree block, it's OK to be racy here */ |
2ff7e61e | 8551 | ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, |
3173a18f JB |
8552 | wc->level - 1, 1, &refs, |
8553 | &flags); | |
79787eaa JM |
8554 | /* We don't care about errors in readahead. */ |
8555 | if (ret < 0) | |
8556 | continue; | |
94fcca9f YZ |
8557 | BUG_ON(refs == 0); |
8558 | ||
1c4850e2 | 8559 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 YZ |
8560 | if (refs == 1) |
8561 | goto reada; | |
bd56b302 | 8562 | |
94fcca9f YZ |
8563 | if (wc->level == 1 && |
8564 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8565 | continue; | |
1c4850e2 YZ |
8566 | if (!wc->update_ref || |
8567 | generation <= root->root_key.offset) | |
8568 | continue; | |
8569 | btrfs_node_key_to_cpu(eb, &key, slot); | |
8570 | ret = btrfs_comp_cpu_keys(&key, | |
8571 | &wc->update_progress); | |
8572 | if (ret < 0) | |
8573 | continue; | |
94fcca9f YZ |
8574 | } else { |
8575 | if (wc->level == 1 && | |
8576 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8577 | continue; | |
6407bf6d | 8578 | } |
1c4850e2 | 8579 | reada: |
2ff7e61e | 8580 | readahead_tree_block(fs_info, bytenr); |
1c4850e2 | 8581 | nread++; |
20524f02 | 8582 | } |
1c4850e2 | 8583 | wc->reada_slot = slot; |
20524f02 | 8584 | } |
2c47e605 | 8585 | |
f82d02d9 | 8586 | /* |
2c016dc2 | 8587 | * helper to process tree block while walking down the tree. |
2c47e605 | 8588 | * |
2c47e605 YZ |
8589 | * when wc->stage == UPDATE_BACKREF, this function updates |
8590 | * back refs for pointers in the block. | |
8591 | * | |
8592 | * NOTE: return value 1 means we should stop walking down. | |
f82d02d9 | 8593 | */ |
2c47e605 | 8594 | static noinline int walk_down_proc(struct btrfs_trans_handle *trans, |
5d4f98a2 | 8595 | struct btrfs_root *root, |
2c47e605 | 8596 | struct btrfs_path *path, |
94fcca9f | 8597 | struct walk_control *wc, int lookup_info) |
f82d02d9 | 8598 | { |
2ff7e61e | 8599 | struct btrfs_fs_info *fs_info = root->fs_info; |
2c47e605 YZ |
8600 | int level = wc->level; |
8601 | struct extent_buffer *eb = path->nodes[level]; | |
2c47e605 | 8602 | u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF; |
f82d02d9 YZ |
8603 | int ret; |
8604 | ||
2c47e605 YZ |
8605 | if (wc->stage == UPDATE_BACKREF && |
8606 | btrfs_header_owner(eb) != root->root_key.objectid) | |
8607 | return 1; | |
f82d02d9 | 8608 | |
2c47e605 YZ |
8609 | /* |
8610 | * when reference count of tree block is 1, it won't increase | |
8611 | * again. once full backref flag is set, we never clear it. | |
8612 | */ | |
94fcca9f YZ |
8613 | if (lookup_info && |
8614 | ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) || | |
8615 | (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) { | |
2c47e605 | 8616 | BUG_ON(!path->locks[level]); |
2ff7e61e | 8617 | ret = btrfs_lookup_extent_info(trans, fs_info, |
3173a18f | 8618 | eb->start, level, 1, |
2c47e605 YZ |
8619 | &wc->refs[level], |
8620 | &wc->flags[level]); | |
79787eaa JM |
8621 | BUG_ON(ret == -ENOMEM); |
8622 | if (ret) | |
8623 | return ret; | |
2c47e605 YZ |
8624 | BUG_ON(wc->refs[level] == 0); |
8625 | } | |
5d4f98a2 | 8626 | |
2c47e605 YZ |
8627 | if (wc->stage == DROP_REFERENCE) { |
8628 | if (wc->refs[level] > 1) | |
8629 | return 1; | |
f82d02d9 | 8630 | |
2c47e605 | 8631 | if (path->locks[level] && !wc->keep_locks) { |
bd681513 | 8632 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
8633 | path->locks[level] = 0; |
8634 | } | |
8635 | return 0; | |
8636 | } | |
f82d02d9 | 8637 | |
2c47e605 YZ |
8638 | /* wc->stage == UPDATE_BACKREF */ |
8639 | if (!(wc->flags[level] & flag)) { | |
8640 | BUG_ON(!path->locks[level]); | |
e339a6b0 | 8641 | ret = btrfs_inc_ref(trans, root, eb, 1); |
79787eaa | 8642 | BUG_ON(ret); /* -ENOMEM */ |
e339a6b0 | 8643 | ret = btrfs_dec_ref(trans, root, eb, 0); |
79787eaa | 8644 | BUG_ON(ret); /* -ENOMEM */ |
2ff7e61e | 8645 | ret = btrfs_set_disk_extent_flags(trans, fs_info, eb->start, |
b1c79e09 JB |
8646 | eb->len, flag, |
8647 | btrfs_header_level(eb), 0); | |
79787eaa | 8648 | BUG_ON(ret); /* -ENOMEM */ |
2c47e605 YZ |
8649 | wc->flags[level] |= flag; |
8650 | } | |
8651 | ||
8652 | /* | |
8653 | * the block is shared by multiple trees, so it's not good to | |
8654 | * keep the tree lock | |
8655 | */ | |
8656 | if (path->locks[level] && level > 0) { | |
bd681513 | 8657 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
8658 | path->locks[level] = 0; |
8659 | } | |
8660 | return 0; | |
8661 | } | |
8662 | ||
1c4850e2 | 8663 | /* |
2c016dc2 | 8664 | * helper to process tree block pointer. |
1c4850e2 YZ |
8665 | * |
8666 | * when wc->stage == DROP_REFERENCE, this function checks | |
8667 | * reference count of the block pointed to. if the block | |
8668 | * is shared and we need update back refs for the subtree | |
8669 | * rooted at the block, this function changes wc->stage to | |
8670 | * UPDATE_BACKREF. if the block is shared and there is no | |
8671 | * need to update back, this function drops the reference | |
8672 | * to the block. | |
8673 | * | |
8674 | * NOTE: return value 1 means we should stop walking down. | |
8675 | */ | |
8676 | static noinline int do_walk_down(struct btrfs_trans_handle *trans, | |
8677 | struct btrfs_root *root, | |
8678 | struct btrfs_path *path, | |
94fcca9f | 8679 | struct walk_control *wc, int *lookup_info) |
1c4850e2 | 8680 | { |
0b246afa | 8681 | struct btrfs_fs_info *fs_info = root->fs_info; |
1c4850e2 YZ |
8682 | u64 bytenr; |
8683 | u64 generation; | |
8684 | u64 parent; | |
8685 | u32 blocksize; | |
8686 | struct btrfs_key key; | |
8687 | struct extent_buffer *next; | |
8688 | int level = wc->level; | |
8689 | int reada = 0; | |
8690 | int ret = 0; | |
1152651a | 8691 | bool need_account = false; |
1c4850e2 YZ |
8692 | |
8693 | generation = btrfs_node_ptr_generation(path->nodes[level], | |
8694 | path->slots[level]); | |
8695 | /* | |
8696 | * if the lower level block was created before the snapshot | |
8697 | * was created, we know there is no need to update back refs | |
8698 | * for the subtree | |
8699 | */ | |
8700 | if (wc->stage == UPDATE_BACKREF && | |
94fcca9f YZ |
8701 | generation <= root->root_key.offset) { |
8702 | *lookup_info = 1; | |
1c4850e2 | 8703 | return 1; |
94fcca9f | 8704 | } |
1c4850e2 YZ |
8705 | |
8706 | bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]); | |
0b246afa | 8707 | blocksize = fs_info->nodesize; |
1c4850e2 | 8708 | |
0b246afa | 8709 | next = find_extent_buffer(fs_info, bytenr); |
1c4850e2 | 8710 | if (!next) { |
2ff7e61e | 8711 | next = btrfs_find_create_tree_block(fs_info, bytenr); |
c871b0f2 LB |
8712 | if (IS_ERR(next)) |
8713 | return PTR_ERR(next); | |
8714 | ||
b2aaaa3b JB |
8715 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, next, |
8716 | level - 1); | |
1c4850e2 YZ |
8717 | reada = 1; |
8718 | } | |
8719 | btrfs_tree_lock(next); | |
8720 | btrfs_set_lock_blocking(next); | |
8721 | ||
2ff7e61e | 8722 | ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1, |
94fcca9f YZ |
8723 | &wc->refs[level - 1], |
8724 | &wc->flags[level - 1]); | |
4867268c JB |
8725 | if (ret < 0) |
8726 | goto out_unlock; | |
79787eaa | 8727 | |
c2cf52eb | 8728 | if (unlikely(wc->refs[level - 1] == 0)) { |
0b246afa | 8729 | btrfs_err(fs_info, "Missing references."); |
4867268c JB |
8730 | ret = -EIO; |
8731 | goto out_unlock; | |
c2cf52eb | 8732 | } |
94fcca9f | 8733 | *lookup_info = 0; |
1c4850e2 | 8734 | |
94fcca9f | 8735 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 | 8736 | if (wc->refs[level - 1] > 1) { |
1152651a | 8737 | need_account = true; |
94fcca9f YZ |
8738 | if (level == 1 && |
8739 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8740 | goto skip; | |
8741 | ||
1c4850e2 YZ |
8742 | if (!wc->update_ref || |
8743 | generation <= root->root_key.offset) | |
8744 | goto skip; | |
8745 | ||
8746 | btrfs_node_key_to_cpu(path->nodes[level], &key, | |
8747 | path->slots[level]); | |
8748 | ret = btrfs_comp_cpu_keys(&key, &wc->update_progress); | |
8749 | if (ret < 0) | |
8750 | goto skip; | |
8751 | ||
8752 | wc->stage = UPDATE_BACKREF; | |
8753 | wc->shared_level = level - 1; | |
8754 | } | |
94fcca9f YZ |
8755 | } else { |
8756 | if (level == 1 && | |
8757 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8758 | goto skip; | |
1c4850e2 YZ |
8759 | } |
8760 | ||
b9fab919 | 8761 | if (!btrfs_buffer_uptodate(next, generation, 0)) { |
1c4850e2 YZ |
8762 | btrfs_tree_unlock(next); |
8763 | free_extent_buffer(next); | |
8764 | next = NULL; | |
94fcca9f | 8765 | *lookup_info = 1; |
1c4850e2 YZ |
8766 | } |
8767 | ||
8768 | if (!next) { | |
8769 | if (reada && level == 1) | |
8770 | reada_walk_down(trans, root, wc, path); | |
2ff7e61e | 8771 | next = read_tree_block(fs_info, bytenr, generation); |
64c043de LB |
8772 | if (IS_ERR(next)) { |
8773 | return PTR_ERR(next); | |
8774 | } else if (!extent_buffer_uptodate(next)) { | |
416bc658 | 8775 | free_extent_buffer(next); |
97d9a8a4 | 8776 | return -EIO; |
416bc658 | 8777 | } |
1c4850e2 YZ |
8778 | btrfs_tree_lock(next); |
8779 | btrfs_set_lock_blocking(next); | |
8780 | } | |
8781 | ||
8782 | level--; | |
4867268c JB |
8783 | ASSERT(level == btrfs_header_level(next)); |
8784 | if (level != btrfs_header_level(next)) { | |
8785 | btrfs_err(root->fs_info, "mismatched level"); | |
8786 | ret = -EIO; | |
8787 | goto out_unlock; | |
8788 | } | |
1c4850e2 YZ |
8789 | path->nodes[level] = next; |
8790 | path->slots[level] = 0; | |
bd681513 | 8791 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
1c4850e2 YZ |
8792 | wc->level = level; |
8793 | if (wc->level == 1) | |
8794 | wc->reada_slot = 0; | |
8795 | return 0; | |
8796 | skip: | |
8797 | wc->refs[level - 1] = 0; | |
8798 | wc->flags[level - 1] = 0; | |
94fcca9f YZ |
8799 | if (wc->stage == DROP_REFERENCE) { |
8800 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
8801 | parent = path->nodes[level]->start; | |
8802 | } else { | |
4867268c | 8803 | ASSERT(root->root_key.objectid == |
94fcca9f | 8804 | btrfs_header_owner(path->nodes[level])); |
4867268c JB |
8805 | if (root->root_key.objectid != |
8806 | btrfs_header_owner(path->nodes[level])) { | |
8807 | btrfs_err(root->fs_info, | |
8808 | "mismatched block owner"); | |
8809 | ret = -EIO; | |
8810 | goto out_unlock; | |
8811 | } | |
94fcca9f YZ |
8812 | parent = 0; |
8813 | } | |
1c4850e2 | 8814 | |
1152651a | 8815 | if (need_account) { |
33d1f05c QW |
8816 | ret = btrfs_qgroup_trace_subtree(trans, root, next, |
8817 | generation, level - 1); | |
1152651a | 8818 | if (ret) { |
0b246afa | 8819 | btrfs_err_rl(fs_info, |
5d163e0e JM |
8820 | "Error %d accounting shared subtree. Quota is out of sync, rescan required.", |
8821 | ret); | |
1152651a MF |
8822 | } |
8823 | } | |
2ff7e61e JM |
8824 | ret = btrfs_free_extent(trans, fs_info, bytenr, blocksize, |
8825 | parent, root->root_key.objectid, | |
8826 | level - 1, 0); | |
4867268c JB |
8827 | if (ret) |
8828 | goto out_unlock; | |
1c4850e2 | 8829 | } |
4867268c JB |
8830 | |
8831 | *lookup_info = 1; | |
8832 | ret = 1; | |
8833 | ||
8834 | out_unlock: | |
1c4850e2 YZ |
8835 | btrfs_tree_unlock(next); |
8836 | free_extent_buffer(next); | |
4867268c JB |
8837 | |
8838 | return ret; | |
1c4850e2 YZ |
8839 | } |
8840 | ||
2c47e605 | 8841 | /* |
2c016dc2 | 8842 | * helper to process tree block while walking up the tree. |
2c47e605 YZ |
8843 | * |
8844 | * when wc->stage == DROP_REFERENCE, this function drops | |
8845 | * reference count on the block. | |
8846 | * | |
8847 | * when wc->stage == UPDATE_BACKREF, this function changes | |
8848 | * wc->stage back to DROP_REFERENCE if we changed wc->stage | |
8849 | * to UPDATE_BACKREF previously while processing the block. | |
8850 | * | |
8851 | * NOTE: return value 1 means we should stop walking up. | |
8852 | */ | |
8853 | static noinline int walk_up_proc(struct btrfs_trans_handle *trans, | |
8854 | struct btrfs_root *root, | |
8855 | struct btrfs_path *path, | |
8856 | struct walk_control *wc) | |
8857 | { | |
0b246afa | 8858 | struct btrfs_fs_info *fs_info = root->fs_info; |
f0486c68 | 8859 | int ret; |
2c47e605 YZ |
8860 | int level = wc->level; |
8861 | struct extent_buffer *eb = path->nodes[level]; | |
8862 | u64 parent = 0; | |
8863 | ||
8864 | if (wc->stage == UPDATE_BACKREF) { | |
8865 | BUG_ON(wc->shared_level < level); | |
8866 | if (level < wc->shared_level) | |
8867 | goto out; | |
8868 | ||
2c47e605 YZ |
8869 | ret = find_next_key(path, level + 1, &wc->update_progress); |
8870 | if (ret > 0) | |
8871 | wc->update_ref = 0; | |
8872 | ||
8873 | wc->stage = DROP_REFERENCE; | |
8874 | wc->shared_level = -1; | |
8875 | path->slots[level] = 0; | |
8876 | ||
8877 | /* | |
8878 | * check reference count again if the block isn't locked. | |
8879 | * we should start walking down the tree again if reference | |
8880 | * count is one. | |
8881 | */ | |
8882 | if (!path->locks[level]) { | |
8883 | BUG_ON(level == 0); | |
8884 | btrfs_tree_lock(eb); | |
8885 | btrfs_set_lock_blocking(eb); | |
bd681513 | 8886 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 | 8887 | |
2ff7e61e | 8888 | ret = btrfs_lookup_extent_info(trans, fs_info, |
3173a18f | 8889 | eb->start, level, 1, |
2c47e605 YZ |
8890 | &wc->refs[level], |
8891 | &wc->flags[level]); | |
79787eaa JM |
8892 | if (ret < 0) { |
8893 | btrfs_tree_unlock_rw(eb, path->locks[level]); | |
3268a246 | 8894 | path->locks[level] = 0; |
79787eaa JM |
8895 | return ret; |
8896 | } | |
2c47e605 YZ |
8897 | BUG_ON(wc->refs[level] == 0); |
8898 | if (wc->refs[level] == 1) { | |
bd681513 | 8899 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
3268a246 | 8900 | path->locks[level] = 0; |
2c47e605 YZ |
8901 | return 1; |
8902 | } | |
f82d02d9 | 8903 | } |
2c47e605 | 8904 | } |
f82d02d9 | 8905 | |
2c47e605 YZ |
8906 | /* wc->stage == DROP_REFERENCE */ |
8907 | BUG_ON(wc->refs[level] > 1 && !path->locks[level]); | |
5d4f98a2 | 8908 | |
2c47e605 YZ |
8909 | if (wc->refs[level] == 1) { |
8910 | if (level == 0) { | |
8911 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
e339a6b0 | 8912 | ret = btrfs_dec_ref(trans, root, eb, 1); |
2c47e605 | 8913 | else |
e339a6b0 | 8914 | ret = btrfs_dec_ref(trans, root, eb, 0); |
79787eaa | 8915 | BUG_ON(ret); /* -ENOMEM */ |
2ff7e61e | 8916 | ret = btrfs_qgroup_trace_leaf_items(trans, fs_info, eb); |
1152651a | 8917 | if (ret) { |
0b246afa | 8918 | btrfs_err_rl(fs_info, |
5d163e0e JM |
8919 | "error %d accounting leaf items. Quota is out of sync, rescan required.", |
8920 | ret); | |
1152651a | 8921 | } |
2c47e605 YZ |
8922 | } |
8923 | /* make block locked assertion in clean_tree_block happy */ | |
8924 | if (!path->locks[level] && | |
8925 | btrfs_header_generation(eb) == trans->transid) { | |
8926 | btrfs_tree_lock(eb); | |
8927 | btrfs_set_lock_blocking(eb); | |
bd681513 | 8928 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 | 8929 | } |
7c302b49 | 8930 | clean_tree_block(fs_info, eb); |
2c47e605 YZ |
8931 | } |
8932 | ||
8933 | if (eb == root->node) { | |
8934 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
8935 | parent = eb->start; | |
8936 | else | |
8937 | BUG_ON(root->root_key.objectid != | |
8938 | btrfs_header_owner(eb)); | |
8939 | } else { | |
8940 | if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
8941 | parent = path->nodes[level + 1]->start; | |
8942 | else | |
8943 | BUG_ON(root->root_key.objectid != | |
8944 | btrfs_header_owner(path->nodes[level + 1])); | |
f82d02d9 | 8945 | } |
f82d02d9 | 8946 | |
5581a51a | 8947 | btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1); |
2c47e605 YZ |
8948 | out: |
8949 | wc->refs[level] = 0; | |
8950 | wc->flags[level] = 0; | |
f0486c68 | 8951 | return 0; |
2c47e605 YZ |
8952 | } |
8953 | ||
8954 | static noinline int walk_down_tree(struct btrfs_trans_handle *trans, | |
8955 | struct btrfs_root *root, | |
8956 | struct btrfs_path *path, | |
8957 | struct walk_control *wc) | |
8958 | { | |
2c47e605 | 8959 | int level = wc->level; |
94fcca9f | 8960 | int lookup_info = 1; |
2c47e605 YZ |
8961 | int ret; |
8962 | ||
8963 | while (level >= 0) { | |
94fcca9f | 8964 | ret = walk_down_proc(trans, root, path, wc, lookup_info); |
2c47e605 YZ |
8965 | if (ret > 0) |
8966 | break; | |
8967 | ||
8968 | if (level == 0) | |
8969 | break; | |
8970 | ||
7a7965f8 YZ |
8971 | if (path->slots[level] >= |
8972 | btrfs_header_nritems(path->nodes[level])) | |
8973 | break; | |
8974 | ||
94fcca9f | 8975 | ret = do_walk_down(trans, root, path, wc, &lookup_info); |
1c4850e2 YZ |
8976 | if (ret > 0) { |
8977 | path->slots[level]++; | |
8978 | continue; | |
90d2c51d MX |
8979 | } else if (ret < 0) |
8980 | return ret; | |
1c4850e2 | 8981 | level = wc->level; |
f82d02d9 | 8982 | } |
f82d02d9 YZ |
8983 | return 0; |
8984 | } | |
8985 | ||
d397712b | 8986 | static noinline int walk_up_tree(struct btrfs_trans_handle *trans, |
98ed5174 | 8987 | struct btrfs_root *root, |
f82d02d9 | 8988 | struct btrfs_path *path, |
2c47e605 | 8989 | struct walk_control *wc, int max_level) |
20524f02 | 8990 | { |
2c47e605 | 8991 | int level = wc->level; |
20524f02 | 8992 | int ret; |
9f3a7427 | 8993 | |
2c47e605 YZ |
8994 | path->slots[level] = btrfs_header_nritems(path->nodes[level]); |
8995 | while (level < max_level && path->nodes[level]) { | |
8996 | wc->level = level; | |
8997 | if (path->slots[level] + 1 < | |
8998 | btrfs_header_nritems(path->nodes[level])) { | |
8999 | path->slots[level]++; | |
20524f02 CM |
9000 | return 0; |
9001 | } else { | |
2c47e605 YZ |
9002 | ret = walk_up_proc(trans, root, path, wc); |
9003 | if (ret > 0) | |
9004 | return 0; | |
bd56b302 | 9005 | |
2c47e605 | 9006 | if (path->locks[level]) { |
bd681513 CM |
9007 | btrfs_tree_unlock_rw(path->nodes[level], |
9008 | path->locks[level]); | |
2c47e605 | 9009 | path->locks[level] = 0; |
f82d02d9 | 9010 | } |
2c47e605 YZ |
9011 | free_extent_buffer(path->nodes[level]); |
9012 | path->nodes[level] = NULL; | |
9013 | level++; | |
20524f02 CM |
9014 | } |
9015 | } | |
9016 | return 1; | |
9017 | } | |
9018 | ||
9aca1d51 | 9019 | /* |
2c47e605 YZ |
9020 | * drop a subvolume tree. |
9021 | * | |
9022 | * this function traverses the tree freeing any blocks that only | |
9023 | * referenced by the tree. | |
9024 | * | |
9025 | * when a shared tree block is found. this function decreases its | |
9026 | * reference count by one. if update_ref is true, this function | |
9027 | * also make sure backrefs for the shared block and all lower level | |
9028 | * blocks are properly updated. | |
9d1a2a3a DS |
9029 | * |
9030 | * If called with for_reloc == 0, may exit early with -EAGAIN | |
9aca1d51 | 9031 | */ |
2c536799 | 9032 | int btrfs_drop_snapshot(struct btrfs_root *root, |
66d7e7f0 AJ |
9033 | struct btrfs_block_rsv *block_rsv, int update_ref, |
9034 | int for_reloc) | |
20524f02 | 9035 | { |
ab8d0fc4 | 9036 | struct btrfs_fs_info *fs_info = root->fs_info; |
5caf2a00 | 9037 | struct btrfs_path *path; |
2c47e605 | 9038 | struct btrfs_trans_handle *trans; |
ab8d0fc4 | 9039 | struct btrfs_root *tree_root = fs_info->tree_root; |
9f3a7427 | 9040 | struct btrfs_root_item *root_item = &root->root_item; |
2c47e605 YZ |
9041 | struct walk_control *wc; |
9042 | struct btrfs_key key; | |
9043 | int err = 0; | |
9044 | int ret; | |
9045 | int level; | |
d29a9f62 | 9046 | bool root_dropped = false; |
20524f02 | 9047 | |
ab8d0fc4 | 9048 | btrfs_debug(fs_info, "Drop subvolume %llu", root->objectid); |
1152651a | 9049 | |
5caf2a00 | 9050 | path = btrfs_alloc_path(); |
cb1b69f4 TI |
9051 | if (!path) { |
9052 | err = -ENOMEM; | |
9053 | goto out; | |
9054 | } | |
20524f02 | 9055 | |
2c47e605 | 9056 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
38a1a919 MF |
9057 | if (!wc) { |
9058 | btrfs_free_path(path); | |
cb1b69f4 TI |
9059 | err = -ENOMEM; |
9060 | goto out; | |
38a1a919 | 9061 | } |
2c47e605 | 9062 | |
a22285a6 | 9063 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
9064 | if (IS_ERR(trans)) { |
9065 | err = PTR_ERR(trans); | |
9066 | goto out_free; | |
9067 | } | |
98d5dc13 | 9068 | |
3fd0a558 YZ |
9069 | if (block_rsv) |
9070 | trans->block_rsv = block_rsv; | |
2c47e605 | 9071 | |
9f3a7427 | 9072 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { |
2c47e605 | 9073 | level = btrfs_header_level(root->node); |
5d4f98a2 YZ |
9074 | path->nodes[level] = btrfs_lock_root_node(root); |
9075 | btrfs_set_lock_blocking(path->nodes[level]); | |
9f3a7427 | 9076 | path->slots[level] = 0; |
bd681513 | 9077 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
9078 | memset(&wc->update_progress, 0, |
9079 | sizeof(wc->update_progress)); | |
9f3a7427 | 9080 | } else { |
9f3a7427 | 9081 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); |
2c47e605 YZ |
9082 | memcpy(&wc->update_progress, &key, |
9083 | sizeof(wc->update_progress)); | |
9084 | ||
6702ed49 | 9085 | level = root_item->drop_level; |
2c47e605 | 9086 | BUG_ON(level == 0); |
6702ed49 | 9087 | path->lowest_level = level; |
2c47e605 YZ |
9088 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
9089 | path->lowest_level = 0; | |
9090 | if (ret < 0) { | |
9091 | err = ret; | |
79787eaa | 9092 | goto out_end_trans; |
9f3a7427 | 9093 | } |
1c4850e2 | 9094 | WARN_ON(ret > 0); |
2c47e605 | 9095 | |
7d9eb12c CM |
9096 | /* |
9097 | * unlock our path, this is safe because only this | |
9098 | * function is allowed to delete this snapshot | |
9099 | */ | |
5d4f98a2 | 9100 | btrfs_unlock_up_safe(path, 0); |
2c47e605 YZ |
9101 | |
9102 | level = btrfs_header_level(root->node); | |
9103 | while (1) { | |
9104 | btrfs_tree_lock(path->nodes[level]); | |
9105 | btrfs_set_lock_blocking(path->nodes[level]); | |
fec386ac | 9106 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 | 9107 | |
2ff7e61e | 9108 | ret = btrfs_lookup_extent_info(trans, fs_info, |
2c47e605 | 9109 | path->nodes[level]->start, |
3173a18f | 9110 | level, 1, &wc->refs[level], |
2c47e605 | 9111 | &wc->flags[level]); |
79787eaa JM |
9112 | if (ret < 0) { |
9113 | err = ret; | |
9114 | goto out_end_trans; | |
9115 | } | |
2c47e605 YZ |
9116 | BUG_ON(wc->refs[level] == 0); |
9117 | ||
9118 | if (level == root_item->drop_level) | |
9119 | break; | |
9120 | ||
9121 | btrfs_tree_unlock(path->nodes[level]); | |
fec386ac | 9122 | path->locks[level] = 0; |
2c47e605 YZ |
9123 | WARN_ON(wc->refs[level] != 1); |
9124 | level--; | |
9125 | } | |
9f3a7427 | 9126 | } |
2c47e605 YZ |
9127 | |
9128 | wc->level = level; | |
9129 | wc->shared_level = -1; | |
9130 | wc->stage = DROP_REFERENCE; | |
9131 | wc->update_ref = update_ref; | |
9132 | wc->keep_locks = 0; | |
66d7e7f0 | 9133 | wc->for_reloc = for_reloc; |
0b246afa | 9134 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info); |
2c47e605 | 9135 | |
d397712b | 9136 | while (1) { |
9d1a2a3a | 9137 | |
2c47e605 YZ |
9138 | ret = walk_down_tree(trans, root, path, wc); |
9139 | if (ret < 0) { | |
9140 | err = ret; | |
20524f02 | 9141 | break; |
2c47e605 | 9142 | } |
9aca1d51 | 9143 | |
2c47e605 YZ |
9144 | ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL); |
9145 | if (ret < 0) { | |
9146 | err = ret; | |
20524f02 | 9147 | break; |
2c47e605 YZ |
9148 | } |
9149 | ||
9150 | if (ret > 0) { | |
9151 | BUG_ON(wc->stage != DROP_REFERENCE); | |
e7a84565 CM |
9152 | break; |
9153 | } | |
2c47e605 YZ |
9154 | |
9155 | if (wc->stage == DROP_REFERENCE) { | |
9156 | level = wc->level; | |
9157 | btrfs_node_key(path->nodes[level], | |
9158 | &root_item->drop_progress, | |
9159 | path->slots[level]); | |
9160 | root_item->drop_level = level; | |
9161 | } | |
9162 | ||
9163 | BUG_ON(wc->level == 0); | |
3a45bb20 | 9164 | if (btrfs_should_end_transaction(trans) || |
2ff7e61e | 9165 | (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) { |
2c47e605 YZ |
9166 | ret = btrfs_update_root(trans, tree_root, |
9167 | &root->root_key, | |
9168 | root_item); | |
79787eaa | 9169 | if (ret) { |
66642832 | 9170 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9171 | err = ret; |
9172 | goto out_end_trans; | |
9173 | } | |
2c47e605 | 9174 | |
3a45bb20 | 9175 | btrfs_end_transaction_throttle(trans); |
2ff7e61e | 9176 | if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) { |
ab8d0fc4 JM |
9177 | btrfs_debug(fs_info, |
9178 | "drop snapshot early exit"); | |
3c8f2422 JB |
9179 | err = -EAGAIN; |
9180 | goto out_free; | |
9181 | } | |
9182 | ||
a22285a6 | 9183 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
9184 | if (IS_ERR(trans)) { |
9185 | err = PTR_ERR(trans); | |
9186 | goto out_free; | |
9187 | } | |
3fd0a558 YZ |
9188 | if (block_rsv) |
9189 | trans->block_rsv = block_rsv; | |
c3e69d58 | 9190 | } |
20524f02 | 9191 | } |
b3b4aa74 | 9192 | btrfs_release_path(path); |
79787eaa JM |
9193 | if (err) |
9194 | goto out_end_trans; | |
2c47e605 YZ |
9195 | |
9196 | ret = btrfs_del_root(trans, tree_root, &root->root_key); | |
79787eaa | 9197 | if (ret) { |
66642832 | 9198 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9199 | goto out_end_trans; |
9200 | } | |
2c47e605 | 9201 | |
76dda93c | 9202 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { |
cb517eab MX |
9203 | ret = btrfs_find_root(tree_root, &root->root_key, path, |
9204 | NULL, NULL); | |
79787eaa | 9205 | if (ret < 0) { |
66642832 | 9206 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9207 | err = ret; |
9208 | goto out_end_trans; | |
9209 | } else if (ret > 0) { | |
84cd948c JB |
9210 | /* if we fail to delete the orphan item this time |
9211 | * around, it'll get picked up the next time. | |
9212 | * | |
9213 | * The most common failure here is just -ENOENT. | |
9214 | */ | |
9215 | btrfs_del_orphan_item(trans, tree_root, | |
9216 | root->root_key.objectid); | |
76dda93c YZ |
9217 | } |
9218 | } | |
9219 | ||
27cdeb70 | 9220 | if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) { |
2b9dbef2 | 9221 | btrfs_add_dropped_root(trans, root); |
76dda93c YZ |
9222 | } else { |
9223 | free_extent_buffer(root->node); | |
9224 | free_extent_buffer(root->commit_root); | |
b0feb9d9 | 9225 | btrfs_put_fs_root(root); |
76dda93c | 9226 | } |
d29a9f62 | 9227 | root_dropped = true; |
79787eaa | 9228 | out_end_trans: |
3a45bb20 | 9229 | btrfs_end_transaction_throttle(trans); |
79787eaa | 9230 | out_free: |
2c47e605 | 9231 | kfree(wc); |
5caf2a00 | 9232 | btrfs_free_path(path); |
cb1b69f4 | 9233 | out: |
d29a9f62 JB |
9234 | /* |
9235 | * So if we need to stop dropping the snapshot for whatever reason we | |
9236 | * need to make sure to add it back to the dead root list so that we | |
9237 | * keep trying to do the work later. This also cleans up roots if we | |
9238 | * don't have it in the radix (like when we recover after a power fail | |
9239 | * or unmount) so we don't leak memory. | |
9240 | */ | |
b37b39cd | 9241 | if (!for_reloc && root_dropped == false) |
d29a9f62 | 9242 | btrfs_add_dead_root(root); |
90515e7f | 9243 | if (err && err != -EAGAIN) |
ab8d0fc4 | 9244 | btrfs_handle_fs_error(fs_info, err, NULL); |
2c536799 | 9245 | return err; |
20524f02 | 9246 | } |
9078a3e1 | 9247 | |
2c47e605 YZ |
9248 | /* |
9249 | * drop subtree rooted at tree block 'node'. | |
9250 | * | |
9251 | * NOTE: this function will unlock and release tree block 'node' | |
66d7e7f0 | 9252 | * only used by relocation code |
2c47e605 | 9253 | */ |
f82d02d9 YZ |
9254 | int btrfs_drop_subtree(struct btrfs_trans_handle *trans, |
9255 | struct btrfs_root *root, | |
9256 | struct extent_buffer *node, | |
9257 | struct extent_buffer *parent) | |
9258 | { | |
0b246afa | 9259 | struct btrfs_fs_info *fs_info = root->fs_info; |
f82d02d9 | 9260 | struct btrfs_path *path; |
2c47e605 | 9261 | struct walk_control *wc; |
f82d02d9 YZ |
9262 | int level; |
9263 | int parent_level; | |
9264 | int ret = 0; | |
9265 | int wret; | |
9266 | ||
2c47e605 YZ |
9267 | BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); |
9268 | ||
f82d02d9 | 9269 | path = btrfs_alloc_path(); |
db5b493a TI |
9270 | if (!path) |
9271 | return -ENOMEM; | |
f82d02d9 | 9272 | |
2c47e605 | 9273 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
db5b493a TI |
9274 | if (!wc) { |
9275 | btrfs_free_path(path); | |
9276 | return -ENOMEM; | |
9277 | } | |
2c47e605 | 9278 | |
b9447ef8 | 9279 | btrfs_assert_tree_locked(parent); |
f82d02d9 YZ |
9280 | parent_level = btrfs_header_level(parent); |
9281 | extent_buffer_get(parent); | |
9282 | path->nodes[parent_level] = parent; | |
9283 | path->slots[parent_level] = btrfs_header_nritems(parent); | |
9284 | ||
b9447ef8 | 9285 | btrfs_assert_tree_locked(node); |
f82d02d9 | 9286 | level = btrfs_header_level(node); |
f82d02d9 YZ |
9287 | path->nodes[level] = node; |
9288 | path->slots[level] = 0; | |
bd681513 | 9289 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
9290 | |
9291 | wc->refs[parent_level] = 1; | |
9292 | wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
9293 | wc->level = level; | |
9294 | wc->shared_level = -1; | |
9295 | wc->stage = DROP_REFERENCE; | |
9296 | wc->update_ref = 0; | |
9297 | wc->keep_locks = 1; | |
66d7e7f0 | 9298 | wc->for_reloc = 1; |
0b246afa | 9299 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info); |
f82d02d9 YZ |
9300 | |
9301 | while (1) { | |
2c47e605 YZ |
9302 | wret = walk_down_tree(trans, root, path, wc); |
9303 | if (wret < 0) { | |
f82d02d9 | 9304 | ret = wret; |
f82d02d9 | 9305 | break; |
2c47e605 | 9306 | } |
f82d02d9 | 9307 | |
2c47e605 | 9308 | wret = walk_up_tree(trans, root, path, wc, parent_level); |
f82d02d9 YZ |
9309 | if (wret < 0) |
9310 | ret = wret; | |
9311 | if (wret != 0) | |
9312 | break; | |
9313 | } | |
9314 | ||
2c47e605 | 9315 | kfree(wc); |
f82d02d9 YZ |
9316 | btrfs_free_path(path); |
9317 | return ret; | |
9318 | } | |
9319 | ||
6202df69 | 9320 | static u64 update_block_group_flags(struct btrfs_fs_info *fs_info, u64 flags) |
ec44a35c CM |
9321 | { |
9322 | u64 num_devices; | |
fc67c450 | 9323 | u64 stripped; |
e4d8ec0f | 9324 | |
fc67c450 ID |
9325 | /* |
9326 | * if restripe for this chunk_type is on pick target profile and | |
9327 | * return, otherwise do the usual balance | |
9328 | */ | |
6202df69 | 9329 | stripped = get_restripe_target(fs_info, flags); |
fc67c450 ID |
9330 | if (stripped) |
9331 | return extended_to_chunk(stripped); | |
e4d8ec0f | 9332 | |
6202df69 | 9333 | num_devices = fs_info->fs_devices->rw_devices; |
cd02dca5 | 9334 | |
fc67c450 | 9335 | stripped = BTRFS_BLOCK_GROUP_RAID0 | |
53b381b3 | 9336 | BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 | |
fc67c450 ID |
9337 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10; |
9338 | ||
ec44a35c CM |
9339 | if (num_devices == 1) { |
9340 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
9341 | stripped = flags & ~stripped; | |
9342 | ||
9343 | /* turn raid0 into single device chunks */ | |
9344 | if (flags & BTRFS_BLOCK_GROUP_RAID0) | |
9345 | return stripped; | |
9346 | ||
9347 | /* turn mirroring into duplication */ | |
9348 | if (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
9349 | BTRFS_BLOCK_GROUP_RAID10)) | |
9350 | return stripped | BTRFS_BLOCK_GROUP_DUP; | |
ec44a35c CM |
9351 | } else { |
9352 | /* they already had raid on here, just return */ | |
ec44a35c CM |
9353 | if (flags & stripped) |
9354 | return flags; | |
9355 | ||
9356 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
9357 | stripped = flags & ~stripped; | |
9358 | ||
9359 | /* switch duplicated blocks with raid1 */ | |
9360 | if (flags & BTRFS_BLOCK_GROUP_DUP) | |
9361 | return stripped | BTRFS_BLOCK_GROUP_RAID1; | |
9362 | ||
e3176ca2 | 9363 | /* this is drive concat, leave it alone */ |
ec44a35c | 9364 | } |
e3176ca2 | 9365 | |
ec44a35c CM |
9366 | return flags; |
9367 | } | |
9368 | ||
868f401a | 9369 | static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force) |
0ef3e66b | 9370 | { |
f0486c68 YZ |
9371 | struct btrfs_space_info *sinfo = cache->space_info; |
9372 | u64 num_bytes; | |
199c36ea | 9373 | u64 min_allocable_bytes; |
f0486c68 | 9374 | int ret = -ENOSPC; |
0ef3e66b | 9375 | |
199c36ea MX |
9376 | /* |
9377 | * We need some metadata space and system metadata space for | |
9378 | * allocating chunks in some corner cases until we force to set | |
9379 | * it to be readonly. | |
9380 | */ | |
9381 | if ((sinfo->flags & | |
9382 | (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) && | |
9383 | !force) | |
ee22184b | 9384 | min_allocable_bytes = SZ_1M; |
199c36ea MX |
9385 | else |
9386 | min_allocable_bytes = 0; | |
9387 | ||
f0486c68 YZ |
9388 | spin_lock(&sinfo->lock); |
9389 | spin_lock(&cache->lock); | |
61cfea9b W |
9390 | |
9391 | if (cache->ro) { | |
868f401a | 9392 | cache->ro++; |
61cfea9b W |
9393 | ret = 0; |
9394 | goto out; | |
9395 | } | |
9396 | ||
f0486c68 YZ |
9397 | num_bytes = cache->key.offset - cache->reserved - cache->pinned - |
9398 | cache->bytes_super - btrfs_block_group_used(&cache->item); | |
9399 | ||
4136135b | 9400 | if (btrfs_space_info_used(sinfo, true) + num_bytes + |
37be25bc | 9401 | min_allocable_bytes <= sinfo->total_bytes) { |
f0486c68 | 9402 | sinfo->bytes_readonly += num_bytes; |
868f401a | 9403 | cache->ro++; |
633c0aad | 9404 | list_add_tail(&cache->ro_list, &sinfo->ro_bgs); |
f0486c68 YZ |
9405 | ret = 0; |
9406 | } | |
61cfea9b | 9407 | out: |
f0486c68 YZ |
9408 | spin_unlock(&cache->lock); |
9409 | spin_unlock(&sinfo->lock); | |
9410 | return ret; | |
9411 | } | |
7d9eb12c | 9412 | |
5e00f193 | 9413 | int btrfs_inc_block_group_ro(struct btrfs_fs_info *fs_info, |
f0486c68 | 9414 | struct btrfs_block_group_cache *cache) |
c286ac48 | 9415 | |
f0486c68 YZ |
9416 | { |
9417 | struct btrfs_trans_handle *trans; | |
9418 | u64 alloc_flags; | |
9419 | int ret; | |
7d9eb12c | 9420 | |
1bbc621e | 9421 | again: |
5e00f193 | 9422 | trans = btrfs_join_transaction(fs_info->extent_root); |
79787eaa JM |
9423 | if (IS_ERR(trans)) |
9424 | return PTR_ERR(trans); | |
5d4f98a2 | 9425 | |
1bbc621e CM |
9426 | /* |
9427 | * we're not allowed to set block groups readonly after the dirty | |
9428 | * block groups cache has started writing. If it already started, | |
9429 | * back off and let this transaction commit | |
9430 | */ | |
0b246afa | 9431 | mutex_lock(&fs_info->ro_block_group_mutex); |
3204d33c | 9432 | if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) { |
1bbc621e CM |
9433 | u64 transid = trans->transid; |
9434 | ||
0b246afa | 9435 | mutex_unlock(&fs_info->ro_block_group_mutex); |
3a45bb20 | 9436 | btrfs_end_transaction(trans); |
1bbc621e | 9437 | |
2ff7e61e | 9438 | ret = btrfs_wait_for_commit(fs_info, transid); |
1bbc621e CM |
9439 | if (ret) |
9440 | return ret; | |
9441 | goto again; | |
9442 | } | |
9443 | ||
153c35b6 CM |
9444 | /* |
9445 | * if we are changing raid levels, try to allocate a corresponding | |
9446 | * block group with the new raid level. | |
9447 | */ | |
0b246afa | 9448 | alloc_flags = update_block_group_flags(fs_info, cache->flags); |
153c35b6 | 9449 | if (alloc_flags != cache->flags) { |
2ff7e61e | 9450 | ret = do_chunk_alloc(trans, fs_info, alloc_flags, |
153c35b6 CM |
9451 | CHUNK_ALLOC_FORCE); |
9452 | /* | |
9453 | * ENOSPC is allowed here, we may have enough space | |
9454 | * already allocated at the new raid level to | |
9455 | * carry on | |
9456 | */ | |
9457 | if (ret == -ENOSPC) | |
9458 | ret = 0; | |
9459 | if (ret < 0) | |
9460 | goto out; | |
9461 | } | |
1bbc621e | 9462 | |
868f401a | 9463 | ret = inc_block_group_ro(cache, 0); |
f0486c68 YZ |
9464 | if (!ret) |
9465 | goto out; | |
2ff7e61e JM |
9466 | alloc_flags = get_alloc_profile(fs_info, cache->space_info->flags); |
9467 | ret = do_chunk_alloc(trans, fs_info, alloc_flags, | |
0e4f8f88 | 9468 | CHUNK_ALLOC_FORCE); |
f0486c68 YZ |
9469 | if (ret < 0) |
9470 | goto out; | |
868f401a | 9471 | ret = inc_block_group_ro(cache, 0); |
f0486c68 | 9472 | out: |
2f081088 | 9473 | if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) { |
0b246afa | 9474 | alloc_flags = update_block_group_flags(fs_info, cache->flags); |
34441361 | 9475 | mutex_lock(&fs_info->chunk_mutex); |
2ff7e61e | 9476 | check_system_chunk(trans, fs_info, alloc_flags); |
34441361 | 9477 | mutex_unlock(&fs_info->chunk_mutex); |
2f081088 | 9478 | } |
0b246afa | 9479 | mutex_unlock(&fs_info->ro_block_group_mutex); |
2f081088 | 9480 | |
3a45bb20 | 9481 | btrfs_end_transaction(trans); |
f0486c68 YZ |
9482 | return ret; |
9483 | } | |
5d4f98a2 | 9484 | |
c87f08ca | 9485 | int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans, |
2ff7e61e | 9486 | struct btrfs_fs_info *fs_info, u64 type) |
c87f08ca | 9487 | { |
2ff7e61e JM |
9488 | u64 alloc_flags = get_alloc_profile(fs_info, type); |
9489 | ||
9490 | return do_chunk_alloc(trans, fs_info, alloc_flags, CHUNK_ALLOC_FORCE); | |
c87f08ca CM |
9491 | } |
9492 | ||
6d07bcec MX |
9493 | /* |
9494 | * helper to account the unused space of all the readonly block group in the | |
633c0aad | 9495 | * space_info. takes mirrors into account. |
6d07bcec | 9496 | */ |
633c0aad | 9497 | u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo) |
6d07bcec MX |
9498 | { |
9499 | struct btrfs_block_group_cache *block_group; | |
9500 | u64 free_bytes = 0; | |
9501 | int factor; | |
9502 | ||
01327610 | 9503 | /* It's df, we don't care if it's racy */ |
633c0aad JB |
9504 | if (list_empty(&sinfo->ro_bgs)) |
9505 | return 0; | |
9506 | ||
9507 | spin_lock(&sinfo->lock); | |
9508 | list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) { | |
6d07bcec MX |
9509 | spin_lock(&block_group->lock); |
9510 | ||
9511 | if (!block_group->ro) { | |
9512 | spin_unlock(&block_group->lock); | |
9513 | continue; | |
9514 | } | |
9515 | ||
9516 | if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
9517 | BTRFS_BLOCK_GROUP_RAID10 | | |
9518 | BTRFS_BLOCK_GROUP_DUP)) | |
9519 | factor = 2; | |
9520 | else | |
9521 | factor = 1; | |
9522 | ||
9523 | free_bytes += (block_group->key.offset - | |
9524 | btrfs_block_group_used(&block_group->item)) * | |
9525 | factor; | |
9526 | ||
9527 | spin_unlock(&block_group->lock); | |
9528 | } | |
6d07bcec MX |
9529 | spin_unlock(&sinfo->lock); |
9530 | ||
9531 | return free_bytes; | |
9532 | } | |
9533 | ||
2ff7e61e | 9534 | void btrfs_dec_block_group_ro(struct btrfs_block_group_cache *cache) |
5d4f98a2 | 9535 | { |
f0486c68 YZ |
9536 | struct btrfs_space_info *sinfo = cache->space_info; |
9537 | u64 num_bytes; | |
9538 | ||
9539 | BUG_ON(!cache->ro); | |
9540 | ||
9541 | spin_lock(&sinfo->lock); | |
9542 | spin_lock(&cache->lock); | |
868f401a Z |
9543 | if (!--cache->ro) { |
9544 | num_bytes = cache->key.offset - cache->reserved - | |
9545 | cache->pinned - cache->bytes_super - | |
9546 | btrfs_block_group_used(&cache->item); | |
9547 | sinfo->bytes_readonly -= num_bytes; | |
9548 | list_del_init(&cache->ro_list); | |
9549 | } | |
f0486c68 YZ |
9550 | spin_unlock(&cache->lock); |
9551 | spin_unlock(&sinfo->lock); | |
5d4f98a2 YZ |
9552 | } |
9553 | ||
ba1bf481 JB |
9554 | /* |
9555 | * checks to see if its even possible to relocate this block group. | |
9556 | * | |
9557 | * @return - -1 if it's not a good idea to relocate this block group, 0 if its | |
9558 | * ok to go ahead and try. | |
9559 | */ | |
6bccf3ab | 9560 | int btrfs_can_relocate(struct btrfs_fs_info *fs_info, u64 bytenr) |
1a40e23b | 9561 | { |
6bccf3ab | 9562 | struct btrfs_root *root = fs_info->extent_root; |
ba1bf481 JB |
9563 | struct btrfs_block_group_cache *block_group; |
9564 | struct btrfs_space_info *space_info; | |
0b246afa | 9565 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
ba1bf481 | 9566 | struct btrfs_device *device; |
6df9a95e | 9567 | struct btrfs_trans_handle *trans; |
cdcb725c | 9568 | u64 min_free; |
6719db6a JB |
9569 | u64 dev_min = 1; |
9570 | u64 dev_nr = 0; | |
4a5e98f5 | 9571 | u64 target; |
0305bc27 | 9572 | int debug; |
cdcb725c | 9573 | int index; |
ba1bf481 JB |
9574 | int full = 0; |
9575 | int ret = 0; | |
1a40e23b | 9576 | |
0b246afa | 9577 | debug = btrfs_test_opt(fs_info, ENOSPC_DEBUG); |
0305bc27 | 9578 | |
0b246afa | 9579 | block_group = btrfs_lookup_block_group(fs_info, bytenr); |
1a40e23b | 9580 | |
ba1bf481 | 9581 | /* odd, couldn't find the block group, leave it alone */ |
0305bc27 QW |
9582 | if (!block_group) { |
9583 | if (debug) | |
0b246afa | 9584 | btrfs_warn(fs_info, |
0305bc27 QW |
9585 | "can't find block group for bytenr %llu", |
9586 | bytenr); | |
ba1bf481 | 9587 | return -1; |
0305bc27 | 9588 | } |
1a40e23b | 9589 | |
cdcb725c | 9590 | min_free = btrfs_block_group_used(&block_group->item); |
9591 | ||
ba1bf481 | 9592 | /* no bytes used, we're good */ |
cdcb725c | 9593 | if (!min_free) |
1a40e23b ZY |
9594 | goto out; |
9595 | ||
ba1bf481 JB |
9596 | space_info = block_group->space_info; |
9597 | spin_lock(&space_info->lock); | |
17d217fe | 9598 | |
ba1bf481 | 9599 | full = space_info->full; |
17d217fe | 9600 | |
ba1bf481 JB |
9601 | /* |
9602 | * if this is the last block group we have in this space, we can't | |
7ce618db CM |
9603 | * relocate it unless we're able to allocate a new chunk below. |
9604 | * | |
9605 | * Otherwise, we need to make sure we have room in the space to handle | |
9606 | * all of the extents from this block group. If we can, we're good | |
ba1bf481 | 9607 | */ |
7ce618db | 9608 | if ((space_info->total_bytes != block_group->key.offset) && |
4136135b LB |
9609 | (btrfs_space_info_used(space_info, false) + min_free < |
9610 | space_info->total_bytes)) { | |
ba1bf481 JB |
9611 | spin_unlock(&space_info->lock); |
9612 | goto out; | |
17d217fe | 9613 | } |
ba1bf481 | 9614 | spin_unlock(&space_info->lock); |
ea8c2819 | 9615 | |
ba1bf481 JB |
9616 | /* |
9617 | * ok we don't have enough space, but maybe we have free space on our | |
9618 | * devices to allocate new chunks for relocation, so loop through our | |
4a5e98f5 ID |
9619 | * alloc devices and guess if we have enough space. if this block |
9620 | * group is going to be restriped, run checks against the target | |
9621 | * profile instead of the current one. | |
ba1bf481 JB |
9622 | */ |
9623 | ret = -1; | |
ea8c2819 | 9624 | |
cdcb725c | 9625 | /* |
9626 | * index: | |
9627 | * 0: raid10 | |
9628 | * 1: raid1 | |
9629 | * 2: dup | |
9630 | * 3: raid0 | |
9631 | * 4: single | |
9632 | */ | |
0b246afa | 9633 | target = get_restripe_target(fs_info, block_group->flags); |
4a5e98f5 | 9634 | if (target) { |
31e50229 | 9635 | index = __get_raid_index(extended_to_chunk(target)); |
4a5e98f5 ID |
9636 | } else { |
9637 | /* | |
9638 | * this is just a balance, so if we were marked as full | |
9639 | * we know there is no space for a new chunk | |
9640 | */ | |
0305bc27 QW |
9641 | if (full) { |
9642 | if (debug) | |
0b246afa JM |
9643 | btrfs_warn(fs_info, |
9644 | "no space to alloc new chunk for block group %llu", | |
9645 | block_group->key.objectid); | |
4a5e98f5 | 9646 | goto out; |
0305bc27 | 9647 | } |
4a5e98f5 ID |
9648 | |
9649 | index = get_block_group_index(block_group); | |
9650 | } | |
9651 | ||
e6ec716f | 9652 | if (index == BTRFS_RAID_RAID10) { |
cdcb725c | 9653 | dev_min = 4; |
6719db6a JB |
9654 | /* Divide by 2 */ |
9655 | min_free >>= 1; | |
e6ec716f | 9656 | } else if (index == BTRFS_RAID_RAID1) { |
cdcb725c | 9657 | dev_min = 2; |
e6ec716f | 9658 | } else if (index == BTRFS_RAID_DUP) { |
6719db6a JB |
9659 | /* Multiply by 2 */ |
9660 | min_free <<= 1; | |
e6ec716f | 9661 | } else if (index == BTRFS_RAID_RAID0) { |
cdcb725c | 9662 | dev_min = fs_devices->rw_devices; |
47c5713f | 9663 | min_free = div64_u64(min_free, dev_min); |
cdcb725c | 9664 | } |
9665 | ||
6df9a95e JB |
9666 | /* We need to do this so that we can look at pending chunks */ |
9667 | trans = btrfs_join_transaction(root); | |
9668 | if (IS_ERR(trans)) { | |
9669 | ret = PTR_ERR(trans); | |
9670 | goto out; | |
9671 | } | |
9672 | ||
0b246afa | 9673 | mutex_lock(&fs_info->chunk_mutex); |
ba1bf481 | 9674 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { |
7bfc837d | 9675 | u64 dev_offset; |
56bec294 | 9676 | |
ba1bf481 JB |
9677 | /* |
9678 | * check to make sure we can actually find a chunk with enough | |
9679 | * space to fit our block group in. | |
9680 | */ | |
63a212ab SB |
9681 | if (device->total_bytes > device->bytes_used + min_free && |
9682 | !device->is_tgtdev_for_dev_replace) { | |
6df9a95e | 9683 | ret = find_free_dev_extent(trans, device, min_free, |
7bfc837d | 9684 | &dev_offset, NULL); |
ba1bf481 | 9685 | if (!ret) |
cdcb725c | 9686 | dev_nr++; |
9687 | ||
9688 | if (dev_nr >= dev_min) | |
73e48b27 | 9689 | break; |
cdcb725c | 9690 | |
ba1bf481 | 9691 | ret = -1; |
725c8463 | 9692 | } |
edbd8d4e | 9693 | } |
0305bc27 | 9694 | if (debug && ret == -1) |
0b246afa JM |
9695 | btrfs_warn(fs_info, |
9696 | "no space to allocate a new chunk for block group %llu", | |
9697 | block_group->key.objectid); | |
9698 | mutex_unlock(&fs_info->chunk_mutex); | |
3a45bb20 | 9699 | btrfs_end_transaction(trans); |
edbd8d4e | 9700 | out: |
ba1bf481 | 9701 | btrfs_put_block_group(block_group); |
edbd8d4e CM |
9702 | return ret; |
9703 | } | |
9704 | ||
6bccf3ab JM |
9705 | static int find_first_block_group(struct btrfs_fs_info *fs_info, |
9706 | struct btrfs_path *path, | |
9707 | struct btrfs_key *key) | |
0b86a832 | 9708 | { |
6bccf3ab | 9709 | struct btrfs_root *root = fs_info->extent_root; |
925baedd | 9710 | int ret = 0; |
0b86a832 CM |
9711 | struct btrfs_key found_key; |
9712 | struct extent_buffer *leaf; | |
9713 | int slot; | |
edbd8d4e | 9714 | |
0b86a832 CM |
9715 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); |
9716 | if (ret < 0) | |
925baedd CM |
9717 | goto out; |
9718 | ||
d397712b | 9719 | while (1) { |
0b86a832 | 9720 | slot = path->slots[0]; |
edbd8d4e | 9721 | leaf = path->nodes[0]; |
0b86a832 CM |
9722 | if (slot >= btrfs_header_nritems(leaf)) { |
9723 | ret = btrfs_next_leaf(root, path); | |
9724 | if (ret == 0) | |
9725 | continue; | |
9726 | if (ret < 0) | |
925baedd | 9727 | goto out; |
0b86a832 | 9728 | break; |
edbd8d4e | 9729 | } |
0b86a832 | 9730 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
edbd8d4e | 9731 | |
0b86a832 | 9732 | if (found_key.objectid >= key->objectid && |
925baedd | 9733 | found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) { |
6fb37b75 LB |
9734 | struct extent_map_tree *em_tree; |
9735 | struct extent_map *em; | |
9736 | ||
9737 | em_tree = &root->fs_info->mapping_tree.map_tree; | |
9738 | read_lock(&em_tree->lock); | |
9739 | em = lookup_extent_mapping(em_tree, found_key.objectid, | |
9740 | found_key.offset); | |
9741 | read_unlock(&em_tree->lock); | |
9742 | if (!em) { | |
0b246afa | 9743 | btrfs_err(fs_info, |
6fb37b75 LB |
9744 | "logical %llu len %llu found bg but no related chunk", |
9745 | found_key.objectid, found_key.offset); | |
9746 | ret = -ENOENT; | |
9747 | } else { | |
9748 | ret = 0; | |
9749 | } | |
187ee58c | 9750 | free_extent_map(em); |
925baedd CM |
9751 | goto out; |
9752 | } | |
0b86a832 | 9753 | path->slots[0]++; |
edbd8d4e | 9754 | } |
925baedd | 9755 | out: |
0b86a832 | 9756 | return ret; |
edbd8d4e CM |
9757 | } |
9758 | ||
0af3d00b JB |
9759 | void btrfs_put_block_group_cache(struct btrfs_fs_info *info) |
9760 | { | |
9761 | struct btrfs_block_group_cache *block_group; | |
9762 | u64 last = 0; | |
9763 | ||
9764 | while (1) { | |
9765 | struct inode *inode; | |
9766 | ||
9767 | block_group = btrfs_lookup_first_block_group(info, last); | |
9768 | while (block_group) { | |
9769 | spin_lock(&block_group->lock); | |
9770 | if (block_group->iref) | |
9771 | break; | |
9772 | spin_unlock(&block_group->lock); | |
2ff7e61e | 9773 | block_group = next_block_group(info, block_group); |
0af3d00b JB |
9774 | } |
9775 | if (!block_group) { | |
9776 | if (last == 0) | |
9777 | break; | |
9778 | last = 0; | |
9779 | continue; | |
9780 | } | |
9781 | ||
9782 | inode = block_group->inode; | |
9783 | block_group->iref = 0; | |
9784 | block_group->inode = NULL; | |
9785 | spin_unlock(&block_group->lock); | |
f3bca802 | 9786 | ASSERT(block_group->io_ctl.inode == NULL); |
0af3d00b JB |
9787 | iput(inode); |
9788 | last = block_group->key.objectid + block_group->key.offset; | |
9789 | btrfs_put_block_group(block_group); | |
9790 | } | |
9791 | } | |
9792 | ||
5cdd7db6 FM |
9793 | /* |
9794 | * Must be called only after stopping all workers, since we could have block | |
9795 | * group caching kthreads running, and therefore they could race with us if we | |
9796 | * freed the block groups before stopping them. | |
9797 | */ | |
1a40e23b ZY |
9798 | int btrfs_free_block_groups(struct btrfs_fs_info *info) |
9799 | { | |
9800 | struct btrfs_block_group_cache *block_group; | |
4184ea7f | 9801 | struct btrfs_space_info *space_info; |
11833d66 | 9802 | struct btrfs_caching_control *caching_ctl; |
1a40e23b ZY |
9803 | struct rb_node *n; |
9804 | ||
9e351cc8 | 9805 | down_write(&info->commit_root_sem); |
11833d66 YZ |
9806 | while (!list_empty(&info->caching_block_groups)) { |
9807 | caching_ctl = list_entry(info->caching_block_groups.next, | |
9808 | struct btrfs_caching_control, list); | |
9809 | list_del(&caching_ctl->list); | |
9810 | put_caching_control(caching_ctl); | |
9811 | } | |
9e351cc8 | 9812 | up_write(&info->commit_root_sem); |
11833d66 | 9813 | |
47ab2a6c JB |
9814 | spin_lock(&info->unused_bgs_lock); |
9815 | while (!list_empty(&info->unused_bgs)) { | |
9816 | block_group = list_first_entry(&info->unused_bgs, | |
9817 | struct btrfs_block_group_cache, | |
9818 | bg_list); | |
9819 | list_del_init(&block_group->bg_list); | |
9820 | btrfs_put_block_group(block_group); | |
9821 | } | |
9822 | spin_unlock(&info->unused_bgs_lock); | |
9823 | ||
1a40e23b ZY |
9824 | spin_lock(&info->block_group_cache_lock); |
9825 | while ((n = rb_last(&info->block_group_cache_tree)) != NULL) { | |
9826 | block_group = rb_entry(n, struct btrfs_block_group_cache, | |
9827 | cache_node); | |
1a40e23b ZY |
9828 | rb_erase(&block_group->cache_node, |
9829 | &info->block_group_cache_tree); | |
01eacb27 | 9830 | RB_CLEAR_NODE(&block_group->cache_node); |
d899e052 YZ |
9831 | spin_unlock(&info->block_group_cache_lock); |
9832 | ||
80eb234a | 9833 | down_write(&block_group->space_info->groups_sem); |
1a40e23b | 9834 | list_del(&block_group->list); |
80eb234a | 9835 | up_write(&block_group->space_info->groups_sem); |
d2fb3437 | 9836 | |
3c14874a JB |
9837 | /* |
9838 | * We haven't cached this block group, which means we could | |
9839 | * possibly have excluded extents on this block group. | |
9840 | */ | |
36cce922 JB |
9841 | if (block_group->cached == BTRFS_CACHE_NO || |
9842 | block_group->cached == BTRFS_CACHE_ERROR) | |
2ff7e61e | 9843 | free_excluded_extents(info, block_group); |
3c14874a | 9844 | |
817d52f8 | 9845 | btrfs_remove_free_space_cache(block_group); |
5cdd7db6 | 9846 | ASSERT(block_group->cached != BTRFS_CACHE_STARTED); |
f3bca802 LB |
9847 | ASSERT(list_empty(&block_group->dirty_list)); |
9848 | ASSERT(list_empty(&block_group->io_list)); | |
9849 | ASSERT(list_empty(&block_group->bg_list)); | |
9850 | ASSERT(atomic_read(&block_group->count) == 1); | |
11dfe35a | 9851 | btrfs_put_block_group(block_group); |
d899e052 YZ |
9852 | |
9853 | spin_lock(&info->block_group_cache_lock); | |
1a40e23b ZY |
9854 | } |
9855 | spin_unlock(&info->block_group_cache_lock); | |
4184ea7f CM |
9856 | |
9857 | /* now that all the block groups are freed, go through and | |
9858 | * free all the space_info structs. This is only called during | |
9859 | * the final stages of unmount, and so we know nobody is | |
9860 | * using them. We call synchronize_rcu() once before we start, | |
9861 | * just to be on the safe side. | |
9862 | */ | |
9863 | synchronize_rcu(); | |
9864 | ||
8929ecfa YZ |
9865 | release_global_block_rsv(info); |
9866 | ||
67871254 | 9867 | while (!list_empty(&info->space_info)) { |
6ab0a202 JM |
9868 | int i; |
9869 | ||
4184ea7f CM |
9870 | space_info = list_entry(info->space_info.next, |
9871 | struct btrfs_space_info, | |
9872 | list); | |
d555b6c3 JB |
9873 | |
9874 | /* | |
9875 | * Do not hide this behind enospc_debug, this is actually | |
9876 | * important and indicates a real bug if this happens. | |
9877 | */ | |
9878 | if (WARN_ON(space_info->bytes_pinned > 0 || | |
b069e0c3 | 9879 | space_info->bytes_reserved > 0 || |
d555b6c3 | 9880 | space_info->bytes_may_use > 0)) |
ab8d0fc4 | 9881 | dump_space_info(info, space_info, 0, 0); |
4184ea7f | 9882 | list_del(&space_info->list); |
6ab0a202 JM |
9883 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { |
9884 | struct kobject *kobj; | |
c1895442 JM |
9885 | kobj = space_info->block_group_kobjs[i]; |
9886 | space_info->block_group_kobjs[i] = NULL; | |
9887 | if (kobj) { | |
6ab0a202 JM |
9888 | kobject_del(kobj); |
9889 | kobject_put(kobj); | |
9890 | } | |
9891 | } | |
9892 | kobject_del(&space_info->kobj); | |
9893 | kobject_put(&space_info->kobj); | |
4184ea7f | 9894 | } |
1a40e23b ZY |
9895 | return 0; |
9896 | } | |
9897 | ||
b742bb82 YZ |
9898 | static void __link_block_group(struct btrfs_space_info *space_info, |
9899 | struct btrfs_block_group_cache *cache) | |
9900 | { | |
9901 | int index = get_block_group_index(cache); | |
ed55b6ac | 9902 | bool first = false; |
b742bb82 YZ |
9903 | |
9904 | down_write(&space_info->groups_sem); | |
ed55b6ac JM |
9905 | if (list_empty(&space_info->block_groups[index])) |
9906 | first = true; | |
9907 | list_add_tail(&cache->list, &space_info->block_groups[index]); | |
9908 | up_write(&space_info->groups_sem); | |
9909 | ||
9910 | if (first) { | |
c1895442 | 9911 | struct raid_kobject *rkobj; |
6ab0a202 JM |
9912 | int ret; |
9913 | ||
c1895442 JM |
9914 | rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS); |
9915 | if (!rkobj) | |
9916 | goto out_err; | |
9917 | rkobj->raid_type = index; | |
9918 | kobject_init(&rkobj->kobj, &btrfs_raid_ktype); | |
9919 | ret = kobject_add(&rkobj->kobj, &space_info->kobj, | |
9920 | "%s", get_raid_name(index)); | |
6ab0a202 | 9921 | if (ret) { |
c1895442 JM |
9922 | kobject_put(&rkobj->kobj); |
9923 | goto out_err; | |
6ab0a202 | 9924 | } |
c1895442 | 9925 | space_info->block_group_kobjs[index] = &rkobj->kobj; |
6ab0a202 | 9926 | } |
c1895442 JM |
9927 | |
9928 | return; | |
9929 | out_err: | |
ab8d0fc4 JM |
9930 | btrfs_warn(cache->fs_info, |
9931 | "failed to add kobject for block cache, ignoring"); | |
b742bb82 YZ |
9932 | } |
9933 | ||
920e4a58 | 9934 | static struct btrfs_block_group_cache * |
2ff7e61e JM |
9935 | btrfs_create_block_group_cache(struct btrfs_fs_info *fs_info, |
9936 | u64 start, u64 size) | |
920e4a58 MX |
9937 | { |
9938 | struct btrfs_block_group_cache *cache; | |
9939 | ||
9940 | cache = kzalloc(sizeof(*cache), GFP_NOFS); | |
9941 | if (!cache) | |
9942 | return NULL; | |
9943 | ||
9944 | cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl), | |
9945 | GFP_NOFS); | |
9946 | if (!cache->free_space_ctl) { | |
9947 | kfree(cache); | |
9948 | return NULL; | |
9949 | } | |
9950 | ||
9951 | cache->key.objectid = start; | |
9952 | cache->key.offset = size; | |
9953 | cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; | |
9954 | ||
0b246afa JM |
9955 | cache->sectorsize = fs_info->sectorsize; |
9956 | cache->fs_info = fs_info; | |
2ff7e61e JM |
9957 | cache->full_stripe_len = btrfs_full_stripe_len(fs_info, |
9958 | &fs_info->mapping_tree, | |
9959 | start); | |
1e144fb8 OS |
9960 | set_free_space_tree_thresholds(cache); |
9961 | ||
920e4a58 MX |
9962 | atomic_set(&cache->count, 1); |
9963 | spin_lock_init(&cache->lock); | |
e570fd27 | 9964 | init_rwsem(&cache->data_rwsem); |
920e4a58 MX |
9965 | INIT_LIST_HEAD(&cache->list); |
9966 | INIT_LIST_HEAD(&cache->cluster_list); | |
47ab2a6c | 9967 | INIT_LIST_HEAD(&cache->bg_list); |
633c0aad | 9968 | INIT_LIST_HEAD(&cache->ro_list); |
ce93ec54 | 9969 | INIT_LIST_HEAD(&cache->dirty_list); |
c9dc4c65 | 9970 | INIT_LIST_HEAD(&cache->io_list); |
920e4a58 | 9971 | btrfs_init_free_space_ctl(cache); |
04216820 | 9972 | atomic_set(&cache->trimming, 0); |
a5ed9182 | 9973 | mutex_init(&cache->free_space_lock); |
0966a7b1 | 9974 | btrfs_init_full_stripe_locks_tree(&cache->full_stripe_locks_root); |
920e4a58 MX |
9975 | |
9976 | return cache; | |
9977 | } | |
9978 | ||
5b4aacef | 9979 | int btrfs_read_block_groups(struct btrfs_fs_info *info) |
9078a3e1 CM |
9980 | { |
9981 | struct btrfs_path *path; | |
9982 | int ret; | |
9078a3e1 | 9983 | struct btrfs_block_group_cache *cache; |
6324fbf3 | 9984 | struct btrfs_space_info *space_info; |
9078a3e1 CM |
9985 | struct btrfs_key key; |
9986 | struct btrfs_key found_key; | |
5f39d397 | 9987 | struct extent_buffer *leaf; |
0af3d00b JB |
9988 | int need_clear = 0; |
9989 | u64 cache_gen; | |
49303381 LB |
9990 | u64 feature; |
9991 | int mixed; | |
9992 | ||
9993 | feature = btrfs_super_incompat_flags(info->super_copy); | |
9994 | mixed = !!(feature & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS); | |
96b5179d | 9995 | |
9078a3e1 | 9996 | key.objectid = 0; |
0b86a832 | 9997 | key.offset = 0; |
962a298f | 9998 | key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; |
9078a3e1 CM |
9999 | path = btrfs_alloc_path(); |
10000 | if (!path) | |
10001 | return -ENOMEM; | |
e4058b54 | 10002 | path->reada = READA_FORWARD; |
9078a3e1 | 10003 | |
0b246afa JM |
10004 | cache_gen = btrfs_super_cache_generation(info->super_copy); |
10005 | if (btrfs_test_opt(info, SPACE_CACHE) && | |
10006 | btrfs_super_generation(info->super_copy) != cache_gen) | |
0af3d00b | 10007 | need_clear = 1; |
0b246afa | 10008 | if (btrfs_test_opt(info, CLEAR_CACHE)) |
88c2ba3b | 10009 | need_clear = 1; |
0af3d00b | 10010 | |
d397712b | 10011 | while (1) { |
6bccf3ab | 10012 | ret = find_first_block_group(info, path, &key); |
b742bb82 YZ |
10013 | if (ret > 0) |
10014 | break; | |
0b86a832 CM |
10015 | if (ret != 0) |
10016 | goto error; | |
920e4a58 | 10017 | |
5f39d397 CM |
10018 | leaf = path->nodes[0]; |
10019 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
920e4a58 | 10020 | |
2ff7e61e | 10021 | cache = btrfs_create_block_group_cache(info, found_key.objectid, |
920e4a58 | 10022 | found_key.offset); |
9078a3e1 | 10023 | if (!cache) { |
0b86a832 | 10024 | ret = -ENOMEM; |
f0486c68 | 10025 | goto error; |
9078a3e1 | 10026 | } |
96303081 | 10027 | |
cf7c1ef6 LB |
10028 | if (need_clear) { |
10029 | /* | |
10030 | * When we mount with old space cache, we need to | |
10031 | * set BTRFS_DC_CLEAR and set dirty flag. | |
10032 | * | |
10033 | * a) Setting 'BTRFS_DC_CLEAR' makes sure that we | |
10034 | * truncate the old free space cache inode and | |
10035 | * setup a new one. | |
10036 | * b) Setting 'dirty flag' makes sure that we flush | |
10037 | * the new space cache info onto disk. | |
10038 | */ | |
0b246afa | 10039 | if (btrfs_test_opt(info, SPACE_CACHE)) |
ce93ec54 | 10040 | cache->disk_cache_state = BTRFS_DC_CLEAR; |
cf7c1ef6 | 10041 | } |
0af3d00b | 10042 | |
5f39d397 CM |
10043 | read_extent_buffer(leaf, &cache->item, |
10044 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
10045 | sizeof(cache->item)); | |
920e4a58 | 10046 | cache->flags = btrfs_block_group_flags(&cache->item); |
49303381 LB |
10047 | if (!mixed && |
10048 | ((cache->flags & BTRFS_BLOCK_GROUP_METADATA) && | |
10049 | (cache->flags & BTRFS_BLOCK_GROUP_DATA))) { | |
10050 | btrfs_err(info, | |
10051 | "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups", | |
10052 | cache->key.objectid); | |
10053 | ret = -EINVAL; | |
10054 | goto error; | |
10055 | } | |
0b86a832 | 10056 | |
9078a3e1 | 10057 | key.objectid = found_key.objectid + found_key.offset; |
b3b4aa74 | 10058 | btrfs_release_path(path); |
34d52cb6 | 10059 | |
3c14874a JB |
10060 | /* |
10061 | * We need to exclude the super stripes now so that the space | |
10062 | * info has super bytes accounted for, otherwise we'll think | |
10063 | * we have more space than we actually do. | |
10064 | */ | |
2ff7e61e | 10065 | ret = exclude_super_stripes(info, cache); |
835d974f JB |
10066 | if (ret) { |
10067 | /* | |
10068 | * We may have excluded something, so call this just in | |
10069 | * case. | |
10070 | */ | |
2ff7e61e | 10071 | free_excluded_extents(info, cache); |
920e4a58 | 10072 | btrfs_put_block_group(cache); |
835d974f JB |
10073 | goto error; |
10074 | } | |
3c14874a | 10075 | |
817d52f8 JB |
10076 | /* |
10077 | * check for two cases, either we are full, and therefore | |
10078 | * don't need to bother with the caching work since we won't | |
10079 | * find any space, or we are empty, and we can just add all | |
10080 | * the space in and be done with it. This saves us _alot_ of | |
10081 | * time, particularly in the full case. | |
10082 | */ | |
10083 | if (found_key.offset == btrfs_block_group_used(&cache->item)) { | |
11833d66 | 10084 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 10085 | cache->cached = BTRFS_CACHE_FINISHED; |
2ff7e61e | 10086 | free_excluded_extents(info, cache); |
817d52f8 | 10087 | } else if (btrfs_block_group_used(&cache->item) == 0) { |
11833d66 | 10088 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 10089 | cache->cached = BTRFS_CACHE_FINISHED; |
0b246afa | 10090 | add_new_free_space(cache, info, |
817d52f8 JB |
10091 | found_key.objectid, |
10092 | found_key.objectid + | |
10093 | found_key.offset); | |
2ff7e61e | 10094 | free_excluded_extents(info, cache); |
817d52f8 | 10095 | } |
96b5179d | 10096 | |
0b246afa | 10097 | ret = btrfs_add_block_group_cache(info, cache); |
8c579fe7 JB |
10098 | if (ret) { |
10099 | btrfs_remove_free_space_cache(cache); | |
10100 | btrfs_put_block_group(cache); | |
10101 | goto error; | |
10102 | } | |
10103 | ||
0b246afa | 10104 | trace_btrfs_add_block_group(info, cache, 0); |
d2006e6d NB |
10105 | update_space_info(info, cache->flags, found_key.offset, |
10106 | btrfs_block_group_used(&cache->item), | |
10107 | cache->bytes_super, &space_info); | |
8c579fe7 | 10108 | |
6324fbf3 | 10109 | cache->space_info = space_info; |
1b2da372 | 10110 | |
b742bb82 | 10111 | __link_block_group(space_info, cache); |
0f9dd46c | 10112 | |
0b246afa | 10113 | set_avail_alloc_bits(info, cache->flags); |
2ff7e61e | 10114 | if (btrfs_chunk_readonly(info, cache->key.objectid)) { |
868f401a | 10115 | inc_block_group_ro(cache, 1); |
47ab2a6c JB |
10116 | } else if (btrfs_block_group_used(&cache->item) == 0) { |
10117 | spin_lock(&info->unused_bgs_lock); | |
10118 | /* Should always be true but just in case. */ | |
10119 | if (list_empty(&cache->bg_list)) { | |
10120 | btrfs_get_block_group(cache); | |
10121 | list_add_tail(&cache->bg_list, | |
10122 | &info->unused_bgs); | |
10123 | } | |
10124 | spin_unlock(&info->unused_bgs_lock); | |
10125 | } | |
9078a3e1 | 10126 | } |
b742bb82 | 10127 | |
0b246afa | 10128 | list_for_each_entry_rcu(space_info, &info->space_info, list) { |
2ff7e61e | 10129 | if (!(get_alloc_profile(info, space_info->flags) & |
b742bb82 YZ |
10130 | (BTRFS_BLOCK_GROUP_RAID10 | |
10131 | BTRFS_BLOCK_GROUP_RAID1 | | |
53b381b3 DW |
10132 | BTRFS_BLOCK_GROUP_RAID5 | |
10133 | BTRFS_BLOCK_GROUP_RAID6 | | |
b742bb82 YZ |
10134 | BTRFS_BLOCK_GROUP_DUP))) |
10135 | continue; | |
10136 | /* | |
10137 | * avoid allocating from un-mirrored block group if there are | |
10138 | * mirrored block groups. | |
10139 | */ | |
1095cc0d | 10140 | list_for_each_entry(cache, |
10141 | &space_info->block_groups[BTRFS_RAID_RAID0], | |
10142 | list) | |
868f401a | 10143 | inc_block_group_ro(cache, 1); |
1095cc0d | 10144 | list_for_each_entry(cache, |
10145 | &space_info->block_groups[BTRFS_RAID_SINGLE], | |
10146 | list) | |
868f401a | 10147 | inc_block_group_ro(cache, 1); |
9078a3e1 | 10148 | } |
f0486c68 YZ |
10149 | |
10150 | init_global_block_rsv(info); | |
0b86a832 CM |
10151 | ret = 0; |
10152 | error: | |
9078a3e1 | 10153 | btrfs_free_path(path); |
0b86a832 | 10154 | return ret; |
9078a3e1 | 10155 | } |
6324fbf3 | 10156 | |
ea658bad | 10157 | void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans, |
2ff7e61e | 10158 | struct btrfs_fs_info *fs_info) |
ea658bad JB |
10159 | { |
10160 | struct btrfs_block_group_cache *block_group, *tmp; | |
0b246afa | 10161 | struct btrfs_root *extent_root = fs_info->extent_root; |
ea658bad JB |
10162 | struct btrfs_block_group_item item; |
10163 | struct btrfs_key key; | |
10164 | int ret = 0; | |
d9a0540a | 10165 | bool can_flush_pending_bgs = trans->can_flush_pending_bgs; |
ea658bad | 10166 | |
d9a0540a | 10167 | trans->can_flush_pending_bgs = false; |
47ab2a6c | 10168 | list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) { |
ea658bad | 10169 | if (ret) |
c92f6be3 | 10170 | goto next; |
ea658bad JB |
10171 | |
10172 | spin_lock(&block_group->lock); | |
10173 | memcpy(&item, &block_group->item, sizeof(item)); | |
10174 | memcpy(&key, &block_group->key, sizeof(key)); | |
10175 | spin_unlock(&block_group->lock); | |
10176 | ||
10177 | ret = btrfs_insert_item(trans, extent_root, &key, &item, | |
10178 | sizeof(item)); | |
10179 | if (ret) | |
66642832 | 10180 | btrfs_abort_transaction(trans, ret); |
0b246afa JM |
10181 | ret = btrfs_finish_chunk_alloc(trans, fs_info, key.objectid, |
10182 | key.offset); | |
6df9a95e | 10183 | if (ret) |
66642832 | 10184 | btrfs_abort_transaction(trans, ret); |
0b246afa | 10185 | add_block_group_free_space(trans, fs_info, block_group); |
1e144fb8 | 10186 | /* already aborted the transaction if it failed. */ |
c92f6be3 FM |
10187 | next: |
10188 | list_del_init(&block_group->bg_list); | |
ea658bad | 10189 | } |
d9a0540a | 10190 | trans->can_flush_pending_bgs = can_flush_pending_bgs; |
ea658bad JB |
10191 | } |
10192 | ||
6324fbf3 | 10193 | int btrfs_make_block_group(struct btrfs_trans_handle *trans, |
2ff7e61e | 10194 | struct btrfs_fs_info *fs_info, u64 bytes_used, |
e17cade2 | 10195 | u64 type, u64 chunk_objectid, u64 chunk_offset, |
6324fbf3 CM |
10196 | u64 size) |
10197 | { | |
6324fbf3 | 10198 | struct btrfs_block_group_cache *cache; |
0b246afa | 10199 | int ret; |
6324fbf3 | 10200 | |
0b246afa | 10201 | btrfs_set_log_full_commit(fs_info, trans); |
e02119d5 | 10202 | |
2ff7e61e | 10203 | cache = btrfs_create_block_group_cache(fs_info, chunk_offset, size); |
0f9dd46c JB |
10204 | if (!cache) |
10205 | return -ENOMEM; | |
34d52cb6 | 10206 | |
6324fbf3 | 10207 | btrfs_set_block_group_used(&cache->item, bytes_used); |
6324fbf3 | 10208 | btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid); |
6324fbf3 CM |
10209 | btrfs_set_block_group_flags(&cache->item, type); |
10210 | ||
920e4a58 | 10211 | cache->flags = type; |
11833d66 | 10212 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 10213 | cache->cached = BTRFS_CACHE_FINISHED; |
1e144fb8 | 10214 | cache->needs_free_space = 1; |
2ff7e61e | 10215 | ret = exclude_super_stripes(fs_info, cache); |
835d974f JB |
10216 | if (ret) { |
10217 | /* | |
10218 | * We may have excluded something, so call this just in | |
10219 | * case. | |
10220 | */ | |
2ff7e61e | 10221 | free_excluded_extents(fs_info, cache); |
920e4a58 | 10222 | btrfs_put_block_group(cache); |
835d974f JB |
10223 | return ret; |
10224 | } | |
96303081 | 10225 | |
0b246afa | 10226 | add_new_free_space(cache, fs_info, chunk_offset, chunk_offset + size); |
817d52f8 | 10227 | |
2ff7e61e | 10228 | free_excluded_extents(fs_info, cache); |
11833d66 | 10229 | |
d0bd4560 | 10230 | #ifdef CONFIG_BTRFS_DEBUG |
2ff7e61e | 10231 | if (btrfs_should_fragment_free_space(cache)) { |
d0bd4560 JB |
10232 | u64 new_bytes_used = size - bytes_used; |
10233 | ||
10234 | bytes_used += new_bytes_used >> 1; | |
2ff7e61e | 10235 | fragment_free_space(cache); |
d0bd4560 JB |
10236 | } |
10237 | #endif | |
2e6e5183 | 10238 | /* |
2be12ef7 NB |
10239 | * Ensure the corresponding space_info object is created and |
10240 | * assigned to our block group. We want our bg to be added to the rbtree | |
10241 | * with its ->space_info set. | |
2e6e5183 | 10242 | */ |
2be12ef7 NB |
10243 | cache->space_info = __find_space_info(fs_info, cache->flags); |
10244 | if (!cache->space_info) { | |
10245 | ret = create_space_info(fs_info, cache->flags, | |
10246 | &cache->space_info); | |
10247 | if (ret) { | |
10248 | btrfs_remove_free_space_cache(cache); | |
10249 | btrfs_put_block_group(cache); | |
10250 | return ret; | |
10251 | } | |
2e6e5183 FM |
10252 | } |
10253 | ||
0b246afa | 10254 | ret = btrfs_add_block_group_cache(fs_info, cache); |
8c579fe7 JB |
10255 | if (ret) { |
10256 | btrfs_remove_free_space_cache(cache); | |
10257 | btrfs_put_block_group(cache); | |
10258 | return ret; | |
10259 | } | |
10260 | ||
2e6e5183 FM |
10261 | /* |
10262 | * Now that our block group has its ->space_info set and is inserted in | |
10263 | * the rbtree, update the space info's counters. | |
10264 | */ | |
0b246afa | 10265 | trace_btrfs_add_block_group(fs_info, cache, 1); |
d2006e6d | 10266 | update_space_info(fs_info, cache->flags, size, bytes_used, |
e40edf2d | 10267 | cache->bytes_super, &cache->space_info); |
0b246afa | 10268 | update_global_block_rsv(fs_info); |
1b2da372 | 10269 | |
b742bb82 | 10270 | __link_block_group(cache->space_info, cache); |
6324fbf3 | 10271 | |
47ab2a6c | 10272 | list_add_tail(&cache->bg_list, &trans->new_bgs); |
6324fbf3 | 10273 | |
0b246afa | 10274 | set_avail_alloc_bits(fs_info, type); |
6324fbf3 CM |
10275 | return 0; |
10276 | } | |
1a40e23b | 10277 | |
10ea00f5 ID |
10278 | static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
10279 | { | |
899c81ea ID |
10280 | u64 extra_flags = chunk_to_extended(flags) & |
10281 | BTRFS_EXTENDED_PROFILE_MASK; | |
10ea00f5 | 10282 | |
de98ced9 | 10283 | write_seqlock(&fs_info->profiles_lock); |
10ea00f5 ID |
10284 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
10285 | fs_info->avail_data_alloc_bits &= ~extra_flags; | |
10286 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
10287 | fs_info->avail_metadata_alloc_bits &= ~extra_flags; | |
10288 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
10289 | fs_info->avail_system_alloc_bits &= ~extra_flags; | |
de98ced9 | 10290 | write_sequnlock(&fs_info->profiles_lock); |
10ea00f5 ID |
10291 | } |
10292 | ||
1a40e23b | 10293 | int btrfs_remove_block_group(struct btrfs_trans_handle *trans, |
6bccf3ab | 10294 | struct btrfs_fs_info *fs_info, u64 group_start, |
04216820 | 10295 | struct extent_map *em) |
1a40e23b | 10296 | { |
6bccf3ab | 10297 | struct btrfs_root *root = fs_info->extent_root; |
1a40e23b ZY |
10298 | struct btrfs_path *path; |
10299 | struct btrfs_block_group_cache *block_group; | |
44fb5511 | 10300 | struct btrfs_free_cluster *cluster; |
0b246afa | 10301 | struct btrfs_root *tree_root = fs_info->tree_root; |
1a40e23b | 10302 | struct btrfs_key key; |
0af3d00b | 10303 | struct inode *inode; |
c1895442 | 10304 | struct kobject *kobj = NULL; |
1a40e23b | 10305 | int ret; |
10ea00f5 | 10306 | int index; |
89a55897 | 10307 | int factor; |
4f69cb98 | 10308 | struct btrfs_caching_control *caching_ctl = NULL; |
04216820 | 10309 | bool remove_em; |
1a40e23b | 10310 | |
6bccf3ab | 10311 | block_group = btrfs_lookup_block_group(fs_info, group_start); |
1a40e23b | 10312 | BUG_ON(!block_group); |
c146afad | 10313 | BUG_ON(!block_group->ro); |
1a40e23b | 10314 | |
9f7c43c9 | 10315 | /* |
10316 | * Free the reserved super bytes from this block group before | |
10317 | * remove it. | |
10318 | */ | |
2ff7e61e | 10319 | free_excluded_extents(fs_info, block_group); |
9f7c43c9 | 10320 | |
1a40e23b | 10321 | memcpy(&key, &block_group->key, sizeof(key)); |
10ea00f5 | 10322 | index = get_block_group_index(block_group); |
89a55897 JB |
10323 | if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP | |
10324 | BTRFS_BLOCK_GROUP_RAID1 | | |
10325 | BTRFS_BLOCK_GROUP_RAID10)) | |
10326 | factor = 2; | |
10327 | else | |
10328 | factor = 1; | |
1a40e23b | 10329 | |
44fb5511 | 10330 | /* make sure this block group isn't part of an allocation cluster */ |
0b246afa | 10331 | cluster = &fs_info->data_alloc_cluster; |
44fb5511 CM |
10332 | spin_lock(&cluster->refill_lock); |
10333 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
10334 | spin_unlock(&cluster->refill_lock); | |
10335 | ||
10336 | /* | |
10337 | * make sure this block group isn't part of a metadata | |
10338 | * allocation cluster | |
10339 | */ | |
0b246afa | 10340 | cluster = &fs_info->meta_alloc_cluster; |
44fb5511 CM |
10341 | spin_lock(&cluster->refill_lock); |
10342 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
10343 | spin_unlock(&cluster->refill_lock); | |
10344 | ||
1a40e23b | 10345 | path = btrfs_alloc_path(); |
d8926bb3 MF |
10346 | if (!path) { |
10347 | ret = -ENOMEM; | |
10348 | goto out; | |
10349 | } | |
1a40e23b | 10350 | |
1bbc621e CM |
10351 | /* |
10352 | * get the inode first so any iput calls done for the io_list | |
10353 | * aren't the final iput (no unlinks allowed now) | |
10354 | */ | |
77ab86bf | 10355 | inode = lookup_free_space_inode(fs_info, block_group, path); |
1bbc621e CM |
10356 | |
10357 | mutex_lock(&trans->transaction->cache_write_mutex); | |
10358 | /* | |
10359 | * make sure our free spache cache IO is done before remove the | |
10360 | * free space inode | |
10361 | */ | |
10362 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
10363 | if (!list_empty(&block_group->io_list)) { | |
10364 | list_del_init(&block_group->io_list); | |
10365 | ||
10366 | WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode); | |
10367 | ||
10368 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
afdb5718 | 10369 | btrfs_wait_cache_io(trans, block_group, path); |
1bbc621e CM |
10370 | btrfs_put_block_group(block_group); |
10371 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
10372 | } | |
10373 | ||
10374 | if (!list_empty(&block_group->dirty_list)) { | |
10375 | list_del_init(&block_group->dirty_list); | |
10376 | btrfs_put_block_group(block_group); | |
10377 | } | |
10378 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
10379 | mutex_unlock(&trans->transaction->cache_write_mutex); | |
10380 | ||
0af3d00b | 10381 | if (!IS_ERR(inode)) { |
73f2e545 | 10382 | ret = btrfs_orphan_add(trans, BTRFS_I(inode)); |
79787eaa JM |
10383 | if (ret) { |
10384 | btrfs_add_delayed_iput(inode); | |
10385 | goto out; | |
10386 | } | |
0af3d00b JB |
10387 | clear_nlink(inode); |
10388 | /* One for the block groups ref */ | |
10389 | spin_lock(&block_group->lock); | |
10390 | if (block_group->iref) { | |
10391 | block_group->iref = 0; | |
10392 | block_group->inode = NULL; | |
10393 | spin_unlock(&block_group->lock); | |
10394 | iput(inode); | |
10395 | } else { | |
10396 | spin_unlock(&block_group->lock); | |
10397 | } | |
10398 | /* One for our lookup ref */ | |
455757c3 | 10399 | btrfs_add_delayed_iput(inode); |
0af3d00b JB |
10400 | } |
10401 | ||
10402 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
10403 | key.offset = block_group->key.objectid; | |
10404 | key.type = 0; | |
10405 | ||
10406 | ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1); | |
10407 | if (ret < 0) | |
10408 | goto out; | |
10409 | if (ret > 0) | |
b3b4aa74 | 10410 | btrfs_release_path(path); |
0af3d00b JB |
10411 | if (ret == 0) { |
10412 | ret = btrfs_del_item(trans, tree_root, path); | |
10413 | if (ret) | |
10414 | goto out; | |
b3b4aa74 | 10415 | btrfs_release_path(path); |
0af3d00b JB |
10416 | } |
10417 | ||
0b246afa | 10418 | spin_lock(&fs_info->block_group_cache_lock); |
1a40e23b | 10419 | rb_erase(&block_group->cache_node, |
0b246afa | 10420 | &fs_info->block_group_cache_tree); |
292cbd51 | 10421 | RB_CLEAR_NODE(&block_group->cache_node); |
a1897fdd | 10422 | |
0b246afa JM |
10423 | if (fs_info->first_logical_byte == block_group->key.objectid) |
10424 | fs_info->first_logical_byte = (u64)-1; | |
10425 | spin_unlock(&fs_info->block_group_cache_lock); | |
817d52f8 | 10426 | |
80eb234a | 10427 | down_write(&block_group->space_info->groups_sem); |
44fb5511 CM |
10428 | /* |
10429 | * we must use list_del_init so people can check to see if they | |
10430 | * are still on the list after taking the semaphore | |
10431 | */ | |
10432 | list_del_init(&block_group->list); | |
6ab0a202 | 10433 | if (list_empty(&block_group->space_info->block_groups[index])) { |
c1895442 JM |
10434 | kobj = block_group->space_info->block_group_kobjs[index]; |
10435 | block_group->space_info->block_group_kobjs[index] = NULL; | |
0b246afa | 10436 | clear_avail_alloc_bits(fs_info, block_group->flags); |
6ab0a202 | 10437 | } |
80eb234a | 10438 | up_write(&block_group->space_info->groups_sem); |
c1895442 JM |
10439 | if (kobj) { |
10440 | kobject_del(kobj); | |
10441 | kobject_put(kobj); | |
10442 | } | |
1a40e23b | 10443 | |
4f69cb98 FM |
10444 | if (block_group->has_caching_ctl) |
10445 | caching_ctl = get_caching_control(block_group); | |
817d52f8 | 10446 | if (block_group->cached == BTRFS_CACHE_STARTED) |
11833d66 | 10447 | wait_block_group_cache_done(block_group); |
4f69cb98 | 10448 | if (block_group->has_caching_ctl) { |
0b246afa | 10449 | down_write(&fs_info->commit_root_sem); |
4f69cb98 FM |
10450 | if (!caching_ctl) { |
10451 | struct btrfs_caching_control *ctl; | |
10452 | ||
10453 | list_for_each_entry(ctl, | |
0b246afa | 10454 | &fs_info->caching_block_groups, list) |
4f69cb98 FM |
10455 | if (ctl->block_group == block_group) { |
10456 | caching_ctl = ctl; | |
1e4f4714 | 10457 | refcount_inc(&caching_ctl->count); |
4f69cb98 FM |
10458 | break; |
10459 | } | |
10460 | } | |
10461 | if (caching_ctl) | |
10462 | list_del_init(&caching_ctl->list); | |
0b246afa | 10463 | up_write(&fs_info->commit_root_sem); |
4f69cb98 FM |
10464 | if (caching_ctl) { |
10465 | /* Once for the caching bgs list and once for us. */ | |
10466 | put_caching_control(caching_ctl); | |
10467 | put_caching_control(caching_ctl); | |
10468 | } | |
10469 | } | |
817d52f8 | 10470 | |
ce93ec54 JB |
10471 | spin_lock(&trans->transaction->dirty_bgs_lock); |
10472 | if (!list_empty(&block_group->dirty_list)) { | |
1bbc621e CM |
10473 | WARN_ON(1); |
10474 | } | |
10475 | if (!list_empty(&block_group->io_list)) { | |
10476 | WARN_ON(1); | |
ce93ec54 JB |
10477 | } |
10478 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
817d52f8 JB |
10479 | btrfs_remove_free_space_cache(block_group); |
10480 | ||
c146afad | 10481 | spin_lock(&block_group->space_info->lock); |
75c68e9f | 10482 | list_del_init(&block_group->ro_list); |
18d018ad | 10483 | |
0b246afa | 10484 | if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
18d018ad ZL |
10485 | WARN_ON(block_group->space_info->total_bytes |
10486 | < block_group->key.offset); | |
10487 | WARN_ON(block_group->space_info->bytes_readonly | |
10488 | < block_group->key.offset); | |
10489 | WARN_ON(block_group->space_info->disk_total | |
10490 | < block_group->key.offset * factor); | |
10491 | } | |
c146afad YZ |
10492 | block_group->space_info->total_bytes -= block_group->key.offset; |
10493 | block_group->space_info->bytes_readonly -= block_group->key.offset; | |
89a55897 | 10494 | block_group->space_info->disk_total -= block_group->key.offset * factor; |
18d018ad | 10495 | |
c146afad | 10496 | spin_unlock(&block_group->space_info->lock); |
283bb197 | 10497 | |
0af3d00b JB |
10498 | memcpy(&key, &block_group->key, sizeof(key)); |
10499 | ||
34441361 | 10500 | mutex_lock(&fs_info->chunk_mutex); |
495e64f4 FM |
10501 | if (!list_empty(&em->list)) { |
10502 | /* We're in the transaction->pending_chunks list. */ | |
10503 | free_extent_map(em); | |
10504 | } | |
04216820 FM |
10505 | spin_lock(&block_group->lock); |
10506 | block_group->removed = 1; | |
10507 | /* | |
10508 | * At this point trimming can't start on this block group, because we | |
10509 | * removed the block group from the tree fs_info->block_group_cache_tree | |
10510 | * so no one can't find it anymore and even if someone already got this | |
10511 | * block group before we removed it from the rbtree, they have already | |
10512 | * incremented block_group->trimming - if they didn't, they won't find | |
10513 | * any free space entries because we already removed them all when we | |
10514 | * called btrfs_remove_free_space_cache(). | |
10515 | * | |
10516 | * And we must not remove the extent map from the fs_info->mapping_tree | |
10517 | * to prevent the same logical address range and physical device space | |
10518 | * ranges from being reused for a new block group. This is because our | |
10519 | * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is | |
10520 | * completely transactionless, so while it is trimming a range the | |
10521 | * currently running transaction might finish and a new one start, | |
10522 | * allowing for new block groups to be created that can reuse the same | |
10523 | * physical device locations unless we take this special care. | |
e33e17ee JM |
10524 | * |
10525 | * There may also be an implicit trim operation if the file system | |
10526 | * is mounted with -odiscard. The same protections must remain | |
10527 | * in place until the extents have been discarded completely when | |
10528 | * the transaction commit has completed. | |
04216820 FM |
10529 | */ |
10530 | remove_em = (atomic_read(&block_group->trimming) == 0); | |
10531 | /* | |
10532 | * Make sure a trimmer task always sees the em in the pinned_chunks list | |
10533 | * if it sees block_group->removed == 1 (needs to lock block_group->lock | |
10534 | * before checking block_group->removed). | |
10535 | */ | |
10536 | if (!remove_em) { | |
10537 | /* | |
10538 | * Our em might be in trans->transaction->pending_chunks which | |
10539 | * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks), | |
10540 | * and so is the fs_info->pinned_chunks list. | |
10541 | * | |
10542 | * So at this point we must be holding the chunk_mutex to avoid | |
10543 | * any races with chunk allocation (more specifically at | |
10544 | * volumes.c:contains_pending_extent()), to ensure it always | |
10545 | * sees the em, either in the pending_chunks list or in the | |
10546 | * pinned_chunks list. | |
10547 | */ | |
0b246afa | 10548 | list_move_tail(&em->list, &fs_info->pinned_chunks); |
04216820 FM |
10549 | } |
10550 | spin_unlock(&block_group->lock); | |
04216820 FM |
10551 | |
10552 | if (remove_em) { | |
10553 | struct extent_map_tree *em_tree; | |
10554 | ||
0b246afa | 10555 | em_tree = &fs_info->mapping_tree.map_tree; |
04216820 | 10556 | write_lock(&em_tree->lock); |
8dbcd10f FM |
10557 | /* |
10558 | * The em might be in the pending_chunks list, so make sure the | |
10559 | * chunk mutex is locked, since remove_extent_mapping() will | |
10560 | * delete us from that list. | |
10561 | */ | |
04216820 FM |
10562 | remove_extent_mapping(em_tree, em); |
10563 | write_unlock(&em_tree->lock); | |
10564 | /* once for the tree */ | |
10565 | free_extent_map(em); | |
10566 | } | |
10567 | ||
34441361 | 10568 | mutex_unlock(&fs_info->chunk_mutex); |
8dbcd10f | 10569 | |
0b246afa | 10570 | ret = remove_block_group_free_space(trans, fs_info, block_group); |
1e144fb8 OS |
10571 | if (ret) |
10572 | goto out; | |
10573 | ||
fa9c0d79 CM |
10574 | btrfs_put_block_group(block_group); |
10575 | btrfs_put_block_group(block_group); | |
1a40e23b ZY |
10576 | |
10577 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
10578 | if (ret > 0) | |
10579 | ret = -EIO; | |
10580 | if (ret < 0) | |
10581 | goto out; | |
10582 | ||
10583 | ret = btrfs_del_item(trans, root, path); | |
10584 | out: | |
10585 | btrfs_free_path(path); | |
10586 | return ret; | |
10587 | } | |
acce952b | 10588 | |
8eab77ff | 10589 | struct btrfs_trans_handle * |
7fd01182 FM |
10590 | btrfs_start_trans_remove_block_group(struct btrfs_fs_info *fs_info, |
10591 | const u64 chunk_offset) | |
8eab77ff | 10592 | { |
7fd01182 FM |
10593 | struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree; |
10594 | struct extent_map *em; | |
10595 | struct map_lookup *map; | |
10596 | unsigned int num_items; | |
10597 | ||
10598 | read_lock(&em_tree->lock); | |
10599 | em = lookup_extent_mapping(em_tree, chunk_offset, 1); | |
10600 | read_unlock(&em_tree->lock); | |
10601 | ASSERT(em && em->start == chunk_offset); | |
10602 | ||
8eab77ff | 10603 | /* |
7fd01182 FM |
10604 | * We need to reserve 3 + N units from the metadata space info in order |
10605 | * to remove a block group (done at btrfs_remove_chunk() and at | |
10606 | * btrfs_remove_block_group()), which are used for: | |
10607 | * | |
8eab77ff FM |
10608 | * 1 unit for adding the free space inode's orphan (located in the tree |
10609 | * of tree roots). | |
7fd01182 FM |
10610 | * 1 unit for deleting the block group item (located in the extent |
10611 | * tree). | |
10612 | * 1 unit for deleting the free space item (located in tree of tree | |
10613 | * roots). | |
10614 | * N units for deleting N device extent items corresponding to each | |
10615 | * stripe (located in the device tree). | |
10616 | * | |
10617 | * In order to remove a block group we also need to reserve units in the | |
10618 | * system space info in order to update the chunk tree (update one or | |
10619 | * more device items and remove one chunk item), but this is done at | |
10620 | * btrfs_remove_chunk() through a call to check_system_chunk(). | |
8eab77ff | 10621 | */ |
95617d69 | 10622 | map = em->map_lookup; |
7fd01182 FM |
10623 | num_items = 3 + map->num_stripes; |
10624 | free_extent_map(em); | |
10625 | ||
8eab77ff | 10626 | return btrfs_start_transaction_fallback_global_rsv(fs_info->extent_root, |
7fd01182 | 10627 | num_items, 1); |
8eab77ff FM |
10628 | } |
10629 | ||
47ab2a6c JB |
10630 | /* |
10631 | * Process the unused_bgs list and remove any that don't have any allocated | |
10632 | * space inside of them. | |
10633 | */ | |
10634 | void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info) | |
10635 | { | |
10636 | struct btrfs_block_group_cache *block_group; | |
10637 | struct btrfs_space_info *space_info; | |
47ab2a6c JB |
10638 | struct btrfs_trans_handle *trans; |
10639 | int ret = 0; | |
10640 | ||
afcdd129 | 10641 | if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags)) |
47ab2a6c JB |
10642 | return; |
10643 | ||
10644 | spin_lock(&fs_info->unused_bgs_lock); | |
10645 | while (!list_empty(&fs_info->unused_bgs)) { | |
10646 | u64 start, end; | |
e33e17ee | 10647 | int trimming; |
47ab2a6c JB |
10648 | |
10649 | block_group = list_first_entry(&fs_info->unused_bgs, | |
10650 | struct btrfs_block_group_cache, | |
10651 | bg_list); | |
47ab2a6c | 10652 | list_del_init(&block_group->bg_list); |
aefbe9a6 ZL |
10653 | |
10654 | space_info = block_group->space_info; | |
10655 | ||
47ab2a6c JB |
10656 | if (ret || btrfs_mixed_space_info(space_info)) { |
10657 | btrfs_put_block_group(block_group); | |
10658 | continue; | |
10659 | } | |
10660 | spin_unlock(&fs_info->unused_bgs_lock); | |
10661 | ||
d5f2e33b | 10662 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
67c5e7d4 | 10663 | |
47ab2a6c JB |
10664 | /* Don't want to race with allocators so take the groups_sem */ |
10665 | down_write(&space_info->groups_sem); | |
10666 | spin_lock(&block_group->lock); | |
10667 | if (block_group->reserved || | |
10668 | btrfs_block_group_used(&block_group->item) || | |
19c4d2f9 | 10669 | block_group->ro || |
aefbe9a6 | 10670 | list_is_singular(&block_group->list)) { |
47ab2a6c JB |
10671 | /* |
10672 | * We want to bail if we made new allocations or have | |
10673 | * outstanding allocations in this block group. We do | |
10674 | * the ro check in case balance is currently acting on | |
10675 | * this block group. | |
10676 | */ | |
10677 | spin_unlock(&block_group->lock); | |
10678 | up_write(&space_info->groups_sem); | |
10679 | goto next; | |
10680 | } | |
10681 | spin_unlock(&block_group->lock); | |
10682 | ||
10683 | /* We don't want to force the issue, only flip if it's ok. */ | |
868f401a | 10684 | ret = inc_block_group_ro(block_group, 0); |
47ab2a6c JB |
10685 | up_write(&space_info->groups_sem); |
10686 | if (ret < 0) { | |
10687 | ret = 0; | |
10688 | goto next; | |
10689 | } | |
10690 | ||
10691 | /* | |
10692 | * Want to do this before we do anything else so we can recover | |
10693 | * properly if we fail to join the transaction. | |
10694 | */ | |
7fd01182 FM |
10695 | trans = btrfs_start_trans_remove_block_group(fs_info, |
10696 | block_group->key.objectid); | |
47ab2a6c | 10697 | if (IS_ERR(trans)) { |
2ff7e61e | 10698 | btrfs_dec_block_group_ro(block_group); |
47ab2a6c JB |
10699 | ret = PTR_ERR(trans); |
10700 | goto next; | |
10701 | } | |
10702 | ||
10703 | /* | |
10704 | * We could have pending pinned extents for this block group, | |
10705 | * just delete them, we don't care about them anymore. | |
10706 | */ | |
10707 | start = block_group->key.objectid; | |
10708 | end = start + block_group->key.offset - 1; | |
d4b450cd FM |
10709 | /* |
10710 | * Hold the unused_bg_unpin_mutex lock to avoid racing with | |
10711 | * btrfs_finish_extent_commit(). If we are at transaction N, | |
10712 | * another task might be running finish_extent_commit() for the | |
10713 | * previous transaction N - 1, and have seen a range belonging | |
10714 | * to the block group in freed_extents[] before we were able to | |
10715 | * clear the whole block group range from freed_extents[]. This | |
10716 | * means that task can lookup for the block group after we | |
10717 | * unpinned it from freed_extents[] and removed it, leading to | |
10718 | * a BUG_ON() at btrfs_unpin_extent_range(). | |
10719 | */ | |
10720 | mutex_lock(&fs_info->unused_bg_unpin_mutex); | |
758eb51e | 10721 | ret = clear_extent_bits(&fs_info->freed_extents[0], start, end, |
91166212 | 10722 | EXTENT_DIRTY); |
758eb51e | 10723 | if (ret) { |
d4b450cd | 10724 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
2ff7e61e | 10725 | btrfs_dec_block_group_ro(block_group); |
758eb51e FM |
10726 | goto end_trans; |
10727 | } | |
10728 | ret = clear_extent_bits(&fs_info->freed_extents[1], start, end, | |
91166212 | 10729 | EXTENT_DIRTY); |
758eb51e | 10730 | if (ret) { |
d4b450cd | 10731 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
2ff7e61e | 10732 | btrfs_dec_block_group_ro(block_group); |
758eb51e FM |
10733 | goto end_trans; |
10734 | } | |
d4b450cd | 10735 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
47ab2a6c JB |
10736 | |
10737 | /* Reset pinned so btrfs_put_block_group doesn't complain */ | |
c30666d4 ZL |
10738 | spin_lock(&space_info->lock); |
10739 | spin_lock(&block_group->lock); | |
10740 | ||
10741 | space_info->bytes_pinned -= block_group->pinned; | |
10742 | space_info->bytes_readonly += block_group->pinned; | |
10743 | percpu_counter_add(&space_info->total_bytes_pinned, | |
10744 | -block_group->pinned); | |
47ab2a6c JB |
10745 | block_group->pinned = 0; |
10746 | ||
c30666d4 ZL |
10747 | spin_unlock(&block_group->lock); |
10748 | spin_unlock(&space_info->lock); | |
10749 | ||
e33e17ee | 10750 | /* DISCARD can flip during remount */ |
0b246afa | 10751 | trimming = btrfs_test_opt(fs_info, DISCARD); |
e33e17ee JM |
10752 | |
10753 | /* Implicit trim during transaction commit. */ | |
10754 | if (trimming) | |
10755 | btrfs_get_block_group_trimming(block_group); | |
10756 | ||
47ab2a6c JB |
10757 | /* |
10758 | * Btrfs_remove_chunk will abort the transaction if things go | |
10759 | * horribly wrong. | |
10760 | */ | |
5b4aacef | 10761 | ret = btrfs_remove_chunk(trans, fs_info, |
47ab2a6c | 10762 | block_group->key.objectid); |
e33e17ee JM |
10763 | |
10764 | if (ret) { | |
10765 | if (trimming) | |
10766 | btrfs_put_block_group_trimming(block_group); | |
10767 | goto end_trans; | |
10768 | } | |
10769 | ||
10770 | /* | |
10771 | * If we're not mounted with -odiscard, we can just forget | |
10772 | * about this block group. Otherwise we'll need to wait | |
10773 | * until transaction commit to do the actual discard. | |
10774 | */ | |
10775 | if (trimming) { | |
348a0013 FM |
10776 | spin_lock(&fs_info->unused_bgs_lock); |
10777 | /* | |
10778 | * A concurrent scrub might have added us to the list | |
10779 | * fs_info->unused_bgs, so use a list_move operation | |
10780 | * to add the block group to the deleted_bgs list. | |
10781 | */ | |
e33e17ee JM |
10782 | list_move(&block_group->bg_list, |
10783 | &trans->transaction->deleted_bgs); | |
348a0013 | 10784 | spin_unlock(&fs_info->unused_bgs_lock); |
e33e17ee JM |
10785 | btrfs_get_block_group(block_group); |
10786 | } | |
758eb51e | 10787 | end_trans: |
3a45bb20 | 10788 | btrfs_end_transaction(trans); |
47ab2a6c | 10789 | next: |
d5f2e33b | 10790 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
47ab2a6c JB |
10791 | btrfs_put_block_group(block_group); |
10792 | spin_lock(&fs_info->unused_bgs_lock); | |
10793 | } | |
10794 | spin_unlock(&fs_info->unused_bgs_lock); | |
10795 | } | |
10796 | ||
c59021f8 | 10797 | int btrfs_init_space_info(struct btrfs_fs_info *fs_info) |
10798 | { | |
10799 | struct btrfs_space_info *space_info; | |
1aba86d6 | 10800 | struct btrfs_super_block *disk_super; |
10801 | u64 features; | |
10802 | u64 flags; | |
10803 | int mixed = 0; | |
c59021f8 | 10804 | int ret; |
10805 | ||
6c41761f | 10806 | disk_super = fs_info->super_copy; |
1aba86d6 | 10807 | if (!btrfs_super_root(disk_super)) |
0dc924c5 | 10808 | return -EINVAL; |
c59021f8 | 10809 | |
1aba86d6 | 10810 | features = btrfs_super_incompat_flags(disk_super); |
10811 | if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
10812 | mixed = 1; | |
c59021f8 | 10813 | |
1aba86d6 | 10814 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
2be12ef7 | 10815 | ret = create_space_info(fs_info, flags, &space_info); |
c59021f8 | 10816 | if (ret) |
1aba86d6 | 10817 | goto out; |
c59021f8 | 10818 | |
1aba86d6 | 10819 | if (mixed) { |
10820 | flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA; | |
2be12ef7 | 10821 | ret = create_space_info(fs_info, flags, &space_info); |
1aba86d6 | 10822 | } else { |
10823 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
2be12ef7 | 10824 | ret = create_space_info(fs_info, flags, &space_info); |
1aba86d6 | 10825 | if (ret) |
10826 | goto out; | |
10827 | ||
10828 | flags = BTRFS_BLOCK_GROUP_DATA; | |
2be12ef7 | 10829 | ret = create_space_info(fs_info, flags, &space_info); |
1aba86d6 | 10830 | } |
10831 | out: | |
c59021f8 | 10832 | return ret; |
10833 | } | |
10834 | ||
2ff7e61e JM |
10835 | int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info, |
10836 | u64 start, u64 end) | |
acce952b | 10837 | { |
2ff7e61e | 10838 | return unpin_extent_range(fs_info, start, end, false); |
acce952b | 10839 | } |
10840 | ||
499f377f JM |
10841 | /* |
10842 | * It used to be that old block groups would be left around forever. | |
10843 | * Iterating over them would be enough to trim unused space. Since we | |
10844 | * now automatically remove them, we also need to iterate over unallocated | |
10845 | * space. | |
10846 | * | |
10847 | * We don't want a transaction for this since the discard may take a | |
10848 | * substantial amount of time. We don't require that a transaction be | |
10849 | * running, but we do need to take a running transaction into account | |
10850 | * to ensure that we're not discarding chunks that were released in | |
10851 | * the current transaction. | |
10852 | * | |
10853 | * Holding the chunks lock will prevent other threads from allocating | |
10854 | * or releasing chunks, but it won't prevent a running transaction | |
10855 | * from committing and releasing the memory that the pending chunks | |
10856 | * list head uses. For that, we need to take a reference to the | |
10857 | * transaction. | |
10858 | */ | |
10859 | static int btrfs_trim_free_extents(struct btrfs_device *device, | |
10860 | u64 minlen, u64 *trimmed) | |
10861 | { | |
10862 | u64 start = 0, len = 0; | |
10863 | int ret; | |
10864 | ||
10865 | *trimmed = 0; | |
10866 | ||
10867 | /* Not writeable = nothing to do. */ | |
10868 | if (!device->writeable) | |
10869 | return 0; | |
10870 | ||
10871 | /* No free space = nothing to do. */ | |
10872 | if (device->total_bytes <= device->bytes_used) | |
10873 | return 0; | |
10874 | ||
10875 | ret = 0; | |
10876 | ||
10877 | while (1) { | |
fb456252 | 10878 | struct btrfs_fs_info *fs_info = device->fs_info; |
499f377f JM |
10879 | struct btrfs_transaction *trans; |
10880 | u64 bytes; | |
10881 | ||
10882 | ret = mutex_lock_interruptible(&fs_info->chunk_mutex); | |
10883 | if (ret) | |
10884 | return ret; | |
10885 | ||
10886 | down_read(&fs_info->commit_root_sem); | |
10887 | ||
10888 | spin_lock(&fs_info->trans_lock); | |
10889 | trans = fs_info->running_transaction; | |
10890 | if (trans) | |
9b64f57d | 10891 | refcount_inc(&trans->use_count); |
499f377f JM |
10892 | spin_unlock(&fs_info->trans_lock); |
10893 | ||
10894 | ret = find_free_dev_extent_start(trans, device, minlen, start, | |
10895 | &start, &len); | |
10896 | if (trans) | |
10897 | btrfs_put_transaction(trans); | |
10898 | ||
10899 | if (ret) { | |
10900 | up_read(&fs_info->commit_root_sem); | |
10901 | mutex_unlock(&fs_info->chunk_mutex); | |
10902 | if (ret == -ENOSPC) | |
10903 | ret = 0; | |
10904 | break; | |
10905 | } | |
10906 | ||
10907 | ret = btrfs_issue_discard(device->bdev, start, len, &bytes); | |
10908 | up_read(&fs_info->commit_root_sem); | |
10909 | mutex_unlock(&fs_info->chunk_mutex); | |
10910 | ||
10911 | if (ret) | |
10912 | break; | |
10913 | ||
10914 | start += len; | |
10915 | *trimmed += bytes; | |
10916 | ||
10917 | if (fatal_signal_pending(current)) { | |
10918 | ret = -ERESTARTSYS; | |
10919 | break; | |
10920 | } | |
10921 | ||
10922 | cond_resched(); | |
10923 | } | |
10924 | ||
10925 | return ret; | |
10926 | } | |
10927 | ||
2ff7e61e | 10928 | int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range) |
f7039b1d | 10929 | { |
f7039b1d | 10930 | struct btrfs_block_group_cache *cache = NULL; |
499f377f JM |
10931 | struct btrfs_device *device; |
10932 | struct list_head *devices; | |
f7039b1d LD |
10933 | u64 group_trimmed; |
10934 | u64 start; | |
10935 | u64 end; | |
10936 | u64 trimmed = 0; | |
2cac13e4 | 10937 | u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy); |
f7039b1d LD |
10938 | int ret = 0; |
10939 | ||
2cac13e4 LB |
10940 | /* |
10941 | * try to trim all FS space, our block group may start from non-zero. | |
10942 | */ | |
10943 | if (range->len == total_bytes) | |
10944 | cache = btrfs_lookup_first_block_group(fs_info, range->start); | |
10945 | else | |
10946 | cache = btrfs_lookup_block_group(fs_info, range->start); | |
f7039b1d LD |
10947 | |
10948 | while (cache) { | |
10949 | if (cache->key.objectid >= (range->start + range->len)) { | |
10950 | btrfs_put_block_group(cache); | |
10951 | break; | |
10952 | } | |
10953 | ||
10954 | start = max(range->start, cache->key.objectid); | |
10955 | end = min(range->start + range->len, | |
10956 | cache->key.objectid + cache->key.offset); | |
10957 | ||
10958 | if (end - start >= range->minlen) { | |
10959 | if (!block_group_cache_done(cache)) { | |
f6373bf3 | 10960 | ret = cache_block_group(cache, 0); |
1be41b78 JB |
10961 | if (ret) { |
10962 | btrfs_put_block_group(cache); | |
10963 | break; | |
10964 | } | |
10965 | ret = wait_block_group_cache_done(cache); | |
10966 | if (ret) { | |
10967 | btrfs_put_block_group(cache); | |
10968 | break; | |
10969 | } | |
f7039b1d LD |
10970 | } |
10971 | ret = btrfs_trim_block_group(cache, | |
10972 | &group_trimmed, | |
10973 | start, | |
10974 | end, | |
10975 | range->minlen); | |
10976 | ||
10977 | trimmed += group_trimmed; | |
10978 | if (ret) { | |
10979 | btrfs_put_block_group(cache); | |
10980 | break; | |
10981 | } | |
10982 | } | |
10983 | ||
2ff7e61e | 10984 | cache = next_block_group(fs_info, cache); |
f7039b1d LD |
10985 | } |
10986 | ||
0b246afa JM |
10987 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
10988 | devices = &fs_info->fs_devices->alloc_list; | |
499f377f JM |
10989 | list_for_each_entry(device, devices, dev_alloc_list) { |
10990 | ret = btrfs_trim_free_extents(device, range->minlen, | |
10991 | &group_trimmed); | |
10992 | if (ret) | |
10993 | break; | |
10994 | ||
10995 | trimmed += group_trimmed; | |
10996 | } | |
0b246afa | 10997 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
499f377f | 10998 | |
f7039b1d LD |
10999 | range->len = trimmed; |
11000 | return ret; | |
11001 | } | |
8257b2dc MX |
11002 | |
11003 | /* | |
9ea24bbe FM |
11004 | * btrfs_{start,end}_write_no_snapshoting() are similar to |
11005 | * mnt_{want,drop}_write(), they are used to prevent some tasks from writing | |
11006 | * data into the page cache through nocow before the subvolume is snapshoted, | |
11007 | * but flush the data into disk after the snapshot creation, or to prevent | |
11008 | * operations while snapshoting is ongoing and that cause the snapshot to be | |
11009 | * inconsistent (writes followed by expanding truncates for example). | |
8257b2dc | 11010 | */ |
9ea24bbe | 11011 | void btrfs_end_write_no_snapshoting(struct btrfs_root *root) |
8257b2dc MX |
11012 | { |
11013 | percpu_counter_dec(&root->subv_writers->counter); | |
11014 | /* | |
a83342aa | 11015 | * Make sure counter is updated before we wake up waiters. |
8257b2dc MX |
11016 | */ |
11017 | smp_mb(); | |
11018 | if (waitqueue_active(&root->subv_writers->wait)) | |
11019 | wake_up(&root->subv_writers->wait); | |
11020 | } | |
11021 | ||
9ea24bbe | 11022 | int btrfs_start_write_no_snapshoting(struct btrfs_root *root) |
8257b2dc | 11023 | { |
ee39b432 | 11024 | if (atomic_read(&root->will_be_snapshoted)) |
8257b2dc MX |
11025 | return 0; |
11026 | ||
11027 | percpu_counter_inc(&root->subv_writers->counter); | |
11028 | /* | |
11029 | * Make sure counter is updated before we check for snapshot creation. | |
11030 | */ | |
11031 | smp_mb(); | |
ee39b432 | 11032 | if (atomic_read(&root->will_be_snapshoted)) { |
9ea24bbe | 11033 | btrfs_end_write_no_snapshoting(root); |
8257b2dc MX |
11034 | return 0; |
11035 | } | |
11036 | return 1; | |
11037 | } | |
0bc19f90 ZL |
11038 | |
11039 | static int wait_snapshoting_atomic_t(atomic_t *a) | |
11040 | { | |
11041 | schedule(); | |
11042 | return 0; | |
11043 | } | |
11044 | ||
11045 | void btrfs_wait_for_snapshot_creation(struct btrfs_root *root) | |
11046 | { | |
11047 | while (true) { | |
11048 | int ret; | |
11049 | ||
11050 | ret = btrfs_start_write_no_snapshoting(root); | |
11051 | if (ret) | |
11052 | break; | |
11053 | wait_on_atomic_t(&root->will_be_snapshoted, | |
11054 | wait_snapshoting_atomic_t, | |
11055 | TASK_UNINTERRUPTIBLE); | |
11056 | } | |
11057 | } |