]>
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 | ||
167ce953 LB |
1151 | /* |
1152 | * is_data == BTRFS_REF_TYPE_BLOCK, tree block type is required, | |
1153 | * is_data == BTRFS_REF_TYPE_DATA, data type is requried, | |
1154 | * is_data == BTRFS_REF_TYPE_ANY, either type is OK. | |
1155 | */ | |
1156 | int btrfs_get_extent_inline_ref_type(const struct extent_buffer *eb, | |
1157 | struct btrfs_extent_inline_ref *iref, | |
1158 | enum btrfs_inline_ref_type is_data) | |
1159 | { | |
1160 | int type = btrfs_extent_inline_ref_type(eb, iref); | |
64ecdb64 | 1161 | u64 offset = btrfs_extent_inline_ref_offset(eb, iref); |
167ce953 LB |
1162 | |
1163 | if (type == BTRFS_TREE_BLOCK_REF_KEY || | |
1164 | type == BTRFS_SHARED_BLOCK_REF_KEY || | |
1165 | type == BTRFS_SHARED_DATA_REF_KEY || | |
1166 | type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1167 | if (is_data == BTRFS_REF_TYPE_BLOCK) { | |
64ecdb64 | 1168 | if (type == BTRFS_TREE_BLOCK_REF_KEY) |
167ce953 | 1169 | return type; |
64ecdb64 LB |
1170 | if (type == BTRFS_SHARED_BLOCK_REF_KEY) { |
1171 | ASSERT(eb->fs_info); | |
1172 | /* | |
e24ef300 QW |
1173 | * Every shared one has parent tree block, |
1174 | * which must be aligned to sector size. | |
64ecdb64 LB |
1175 | */ |
1176 | if (offset && | |
e24ef300 | 1177 | IS_ALIGNED(offset, eb->fs_info->sectorsize)) |
64ecdb64 LB |
1178 | return type; |
1179 | } | |
167ce953 | 1180 | } else if (is_data == BTRFS_REF_TYPE_DATA) { |
64ecdb64 | 1181 | if (type == BTRFS_EXTENT_DATA_REF_KEY) |
167ce953 | 1182 | return type; |
64ecdb64 LB |
1183 | if (type == BTRFS_SHARED_DATA_REF_KEY) { |
1184 | ASSERT(eb->fs_info); | |
1185 | /* | |
e24ef300 QW |
1186 | * Every shared one has parent tree block, |
1187 | * which must be aligned to sector size. | |
64ecdb64 LB |
1188 | */ |
1189 | if (offset && | |
e24ef300 | 1190 | IS_ALIGNED(offset, eb->fs_info->sectorsize)) |
64ecdb64 LB |
1191 | return type; |
1192 | } | |
167ce953 LB |
1193 | } else { |
1194 | ASSERT(is_data == BTRFS_REF_TYPE_ANY); | |
1195 | return type; | |
1196 | } | |
1197 | } | |
1198 | ||
1199 | btrfs_print_leaf((struct extent_buffer *)eb); | |
e24ef300 QW |
1200 | btrfs_err(eb->fs_info, |
1201 | "eb %llu iref 0x%lx invalid extent inline ref type %d", | |
1202 | eb->start, (unsigned long)iref, type); | |
167ce953 LB |
1203 | WARN_ON(1); |
1204 | ||
1205 | return BTRFS_REF_TYPE_INVALID; | |
1206 | } | |
1207 | ||
5d4f98a2 YZ |
1208 | static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset) |
1209 | { | |
1210 | u32 high_crc = ~(u32)0; | |
1211 | u32 low_crc = ~(u32)0; | |
1212 | __le64 lenum; | |
1213 | ||
1214 | lenum = cpu_to_le64(root_objectid); | |
14a958e6 | 1215 | high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 1216 | lenum = cpu_to_le64(owner); |
14a958e6 | 1217 | low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 1218 | lenum = cpu_to_le64(offset); |
14a958e6 | 1219 | low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 YZ |
1220 | |
1221 | return ((u64)high_crc << 31) ^ (u64)low_crc; | |
1222 | } | |
1223 | ||
1224 | static u64 hash_extent_data_ref_item(struct extent_buffer *leaf, | |
1225 | struct btrfs_extent_data_ref *ref) | |
1226 | { | |
1227 | return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref), | |
1228 | btrfs_extent_data_ref_objectid(leaf, ref), | |
1229 | btrfs_extent_data_ref_offset(leaf, ref)); | |
1230 | } | |
1231 | ||
1232 | static int match_extent_data_ref(struct extent_buffer *leaf, | |
1233 | struct btrfs_extent_data_ref *ref, | |
1234 | u64 root_objectid, u64 owner, u64 offset) | |
1235 | { | |
1236 | if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid || | |
1237 | btrfs_extent_data_ref_objectid(leaf, ref) != owner || | |
1238 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
1239 | return 0; | |
1240 | return 1; | |
1241 | } | |
1242 | ||
1243 | static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans, | |
87bde3cd | 1244 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1245 | struct btrfs_path *path, |
1246 | u64 bytenr, u64 parent, | |
1247 | u64 root_objectid, | |
1248 | u64 owner, u64 offset) | |
1249 | { | |
87bde3cd | 1250 | struct btrfs_root *root = fs_info->extent_root; |
5d4f98a2 YZ |
1251 | struct btrfs_key key; |
1252 | struct btrfs_extent_data_ref *ref; | |
31840ae1 | 1253 | struct extent_buffer *leaf; |
5d4f98a2 | 1254 | u32 nritems; |
74493f7a | 1255 | int ret; |
5d4f98a2 YZ |
1256 | int recow; |
1257 | int err = -ENOENT; | |
74493f7a | 1258 | |
31840ae1 | 1259 | key.objectid = bytenr; |
5d4f98a2 YZ |
1260 | if (parent) { |
1261 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1262 | key.offset = parent; | |
1263 | } else { | |
1264 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1265 | key.offset = hash_extent_data_ref(root_objectid, | |
1266 | owner, offset); | |
1267 | } | |
1268 | again: | |
1269 | recow = 0; | |
1270 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1271 | if (ret < 0) { | |
1272 | err = ret; | |
1273 | goto fail; | |
1274 | } | |
31840ae1 | 1275 | |
5d4f98a2 YZ |
1276 | if (parent) { |
1277 | if (!ret) | |
1278 | return 0; | |
1279 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1280 | key.type = BTRFS_EXTENT_REF_V0_KEY; | |
b3b4aa74 | 1281 | btrfs_release_path(path); |
5d4f98a2 YZ |
1282 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1283 | if (ret < 0) { | |
1284 | err = ret; | |
1285 | goto fail; | |
1286 | } | |
1287 | if (!ret) | |
1288 | return 0; | |
1289 | #endif | |
1290 | goto fail; | |
31840ae1 ZY |
1291 | } |
1292 | ||
1293 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1294 | nritems = btrfs_header_nritems(leaf); |
1295 | while (1) { | |
1296 | if (path->slots[0] >= nritems) { | |
1297 | ret = btrfs_next_leaf(root, path); | |
1298 | if (ret < 0) | |
1299 | err = ret; | |
1300 | if (ret) | |
1301 | goto fail; | |
1302 | ||
1303 | leaf = path->nodes[0]; | |
1304 | nritems = btrfs_header_nritems(leaf); | |
1305 | recow = 1; | |
1306 | } | |
1307 | ||
1308 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1309 | if (key.objectid != bytenr || | |
1310 | key.type != BTRFS_EXTENT_DATA_REF_KEY) | |
1311 | goto fail; | |
1312 | ||
1313 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1314 | struct btrfs_extent_data_ref); | |
1315 | ||
1316 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1317 | owner, offset)) { | |
1318 | if (recow) { | |
b3b4aa74 | 1319 | btrfs_release_path(path); |
5d4f98a2 YZ |
1320 | goto again; |
1321 | } | |
1322 | err = 0; | |
1323 | break; | |
1324 | } | |
1325 | path->slots[0]++; | |
31840ae1 | 1326 | } |
5d4f98a2 YZ |
1327 | fail: |
1328 | return err; | |
31840ae1 ZY |
1329 | } |
1330 | ||
5d4f98a2 | 1331 | static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans, |
87bde3cd | 1332 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1333 | struct btrfs_path *path, |
1334 | u64 bytenr, u64 parent, | |
1335 | u64 root_objectid, u64 owner, | |
1336 | u64 offset, int refs_to_add) | |
31840ae1 | 1337 | { |
87bde3cd | 1338 | struct btrfs_root *root = fs_info->extent_root; |
31840ae1 ZY |
1339 | struct btrfs_key key; |
1340 | struct extent_buffer *leaf; | |
5d4f98a2 | 1341 | u32 size; |
31840ae1 ZY |
1342 | u32 num_refs; |
1343 | int ret; | |
74493f7a | 1344 | |
74493f7a | 1345 | key.objectid = bytenr; |
5d4f98a2 YZ |
1346 | if (parent) { |
1347 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1348 | key.offset = parent; | |
1349 | size = sizeof(struct btrfs_shared_data_ref); | |
1350 | } else { | |
1351 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1352 | key.offset = hash_extent_data_ref(root_objectid, | |
1353 | owner, offset); | |
1354 | size = sizeof(struct btrfs_extent_data_ref); | |
1355 | } | |
74493f7a | 1356 | |
5d4f98a2 YZ |
1357 | ret = btrfs_insert_empty_item(trans, root, path, &key, size); |
1358 | if (ret && ret != -EEXIST) | |
1359 | goto fail; | |
1360 | ||
1361 | leaf = path->nodes[0]; | |
1362 | if (parent) { | |
1363 | struct btrfs_shared_data_ref *ref; | |
31840ae1 | 1364 | ref = btrfs_item_ptr(leaf, path->slots[0], |
5d4f98a2 YZ |
1365 | struct btrfs_shared_data_ref); |
1366 | if (ret == 0) { | |
1367 | btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add); | |
1368 | } else { | |
1369 | num_refs = btrfs_shared_data_ref_count(leaf, ref); | |
1370 | num_refs += refs_to_add; | |
1371 | btrfs_set_shared_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1372 | } |
5d4f98a2 YZ |
1373 | } else { |
1374 | struct btrfs_extent_data_ref *ref; | |
1375 | while (ret == -EEXIST) { | |
1376 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1377 | struct btrfs_extent_data_ref); | |
1378 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1379 | owner, offset)) | |
1380 | break; | |
b3b4aa74 | 1381 | btrfs_release_path(path); |
5d4f98a2 YZ |
1382 | key.offset++; |
1383 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1384 | size); | |
1385 | if (ret && ret != -EEXIST) | |
1386 | goto fail; | |
31840ae1 | 1387 | |
5d4f98a2 YZ |
1388 | leaf = path->nodes[0]; |
1389 | } | |
1390 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1391 | struct btrfs_extent_data_ref); | |
1392 | if (ret == 0) { | |
1393 | btrfs_set_extent_data_ref_root(leaf, ref, | |
1394 | root_objectid); | |
1395 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
1396 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
1397 | btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add); | |
1398 | } else { | |
1399 | num_refs = btrfs_extent_data_ref_count(leaf, ref); | |
1400 | num_refs += refs_to_add; | |
1401 | btrfs_set_extent_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1402 | } |
31840ae1 | 1403 | } |
5d4f98a2 YZ |
1404 | btrfs_mark_buffer_dirty(leaf); |
1405 | ret = 0; | |
1406 | fail: | |
b3b4aa74 | 1407 | btrfs_release_path(path); |
7bb86316 | 1408 | return ret; |
74493f7a CM |
1409 | } |
1410 | ||
5d4f98a2 | 1411 | static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans, |
87bde3cd | 1412 | struct btrfs_fs_info *fs_info, |
5d4f98a2 | 1413 | struct btrfs_path *path, |
fcebe456 | 1414 | int refs_to_drop, int *last_ref) |
31840ae1 | 1415 | { |
5d4f98a2 YZ |
1416 | struct btrfs_key key; |
1417 | struct btrfs_extent_data_ref *ref1 = NULL; | |
1418 | struct btrfs_shared_data_ref *ref2 = NULL; | |
31840ae1 | 1419 | struct extent_buffer *leaf; |
5d4f98a2 | 1420 | u32 num_refs = 0; |
31840ae1 ZY |
1421 | int ret = 0; |
1422 | ||
1423 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1424 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
1425 | ||
1426 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1427 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1428 | struct btrfs_extent_data_ref); | |
1429 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1430 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1431 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1432 | struct btrfs_shared_data_ref); | |
1433 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1434 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1435 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
1436 | struct btrfs_extent_ref_v0 *ref0; | |
1437 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1438 | struct btrfs_extent_ref_v0); | |
1439 | num_refs = btrfs_ref_count_v0(leaf, ref0); | |
1440 | #endif | |
1441 | } else { | |
1442 | BUG(); | |
1443 | } | |
1444 | ||
56bec294 CM |
1445 | BUG_ON(num_refs < refs_to_drop); |
1446 | num_refs -= refs_to_drop; | |
5d4f98a2 | 1447 | |
31840ae1 | 1448 | if (num_refs == 0) { |
87bde3cd | 1449 | ret = btrfs_del_item(trans, fs_info->extent_root, path); |
fcebe456 | 1450 | *last_ref = 1; |
31840ae1 | 1451 | } else { |
5d4f98a2 YZ |
1452 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) |
1453 | btrfs_set_extent_data_ref_count(leaf, ref1, num_refs); | |
1454 | else if (key.type == BTRFS_SHARED_DATA_REF_KEY) | |
1455 | btrfs_set_shared_data_ref_count(leaf, ref2, num_refs); | |
1456 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1457 | else { | |
1458 | struct btrfs_extent_ref_v0 *ref0; | |
1459 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1460 | struct btrfs_extent_ref_v0); | |
1461 | btrfs_set_ref_count_v0(leaf, ref0, num_refs); | |
1462 | } | |
1463 | #endif | |
31840ae1 ZY |
1464 | btrfs_mark_buffer_dirty(leaf); |
1465 | } | |
31840ae1 ZY |
1466 | return ret; |
1467 | } | |
1468 | ||
9ed0dea0 | 1469 | static noinline u32 extent_data_ref_count(struct btrfs_path *path, |
5d4f98a2 | 1470 | struct btrfs_extent_inline_ref *iref) |
15916de8 | 1471 | { |
5d4f98a2 YZ |
1472 | struct btrfs_key key; |
1473 | struct extent_buffer *leaf; | |
1474 | struct btrfs_extent_data_ref *ref1; | |
1475 | struct btrfs_shared_data_ref *ref2; | |
1476 | u32 num_refs = 0; | |
3de28d57 | 1477 | int type; |
5d4f98a2 YZ |
1478 | |
1479 | leaf = path->nodes[0]; | |
1480 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1481 | if (iref) { | |
3de28d57 LB |
1482 | /* |
1483 | * If type is invalid, we should have bailed out earlier than | |
1484 | * this call. | |
1485 | */ | |
1486 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA); | |
1487 | ASSERT(type != BTRFS_REF_TYPE_INVALID); | |
1488 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
5d4f98a2 YZ |
1489 | ref1 = (struct btrfs_extent_data_ref *)(&iref->offset); |
1490 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1491 | } else { | |
1492 | ref2 = (struct btrfs_shared_data_ref *)(iref + 1); | |
1493 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1494 | } | |
1495 | } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1496 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1497 | struct btrfs_extent_data_ref); | |
1498 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1499 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1500 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1501 | struct btrfs_shared_data_ref); | |
1502 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1503 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1504 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
1505 | struct btrfs_extent_ref_v0 *ref0; | |
1506 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1507 | struct btrfs_extent_ref_v0); | |
1508 | num_refs = btrfs_ref_count_v0(leaf, ref0); | |
4b4e25f2 | 1509 | #endif |
5d4f98a2 YZ |
1510 | } else { |
1511 | WARN_ON(1); | |
1512 | } | |
1513 | return num_refs; | |
1514 | } | |
15916de8 | 1515 | |
5d4f98a2 | 1516 | static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans, |
87bde3cd | 1517 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1518 | struct btrfs_path *path, |
1519 | u64 bytenr, u64 parent, | |
1520 | u64 root_objectid) | |
1f3c79a2 | 1521 | { |
87bde3cd | 1522 | struct btrfs_root *root = fs_info->extent_root; |
5d4f98a2 | 1523 | struct btrfs_key key; |
1f3c79a2 | 1524 | int ret; |
1f3c79a2 | 1525 | |
5d4f98a2 YZ |
1526 | key.objectid = bytenr; |
1527 | if (parent) { | |
1528 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1529 | key.offset = parent; | |
1530 | } else { | |
1531 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1532 | key.offset = root_objectid; | |
1f3c79a2 LH |
1533 | } |
1534 | ||
5d4f98a2 YZ |
1535 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1536 | if (ret > 0) | |
1537 | ret = -ENOENT; | |
1538 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1539 | if (ret == -ENOENT && parent) { | |
b3b4aa74 | 1540 | btrfs_release_path(path); |
5d4f98a2 YZ |
1541 | key.type = BTRFS_EXTENT_REF_V0_KEY; |
1542 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1543 | if (ret > 0) | |
1544 | ret = -ENOENT; | |
1545 | } | |
1f3c79a2 | 1546 | #endif |
5d4f98a2 | 1547 | return ret; |
1f3c79a2 LH |
1548 | } |
1549 | ||
5d4f98a2 | 1550 | static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans, |
87bde3cd | 1551 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1552 | struct btrfs_path *path, |
1553 | u64 bytenr, u64 parent, | |
1554 | u64 root_objectid) | |
31840ae1 | 1555 | { |
5d4f98a2 | 1556 | struct btrfs_key key; |
31840ae1 | 1557 | int ret; |
31840ae1 | 1558 | |
5d4f98a2 YZ |
1559 | key.objectid = bytenr; |
1560 | if (parent) { | |
1561 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1562 | key.offset = parent; | |
1563 | } else { | |
1564 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1565 | key.offset = root_objectid; | |
1566 | } | |
1567 | ||
87bde3cd JM |
1568 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, |
1569 | path, &key, 0); | |
b3b4aa74 | 1570 | btrfs_release_path(path); |
31840ae1 ZY |
1571 | return ret; |
1572 | } | |
1573 | ||
5d4f98a2 | 1574 | static inline int extent_ref_type(u64 parent, u64 owner) |
31840ae1 | 1575 | { |
5d4f98a2 YZ |
1576 | int type; |
1577 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1578 | if (parent > 0) | |
1579 | type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1580 | else | |
1581 | type = BTRFS_TREE_BLOCK_REF_KEY; | |
1582 | } else { | |
1583 | if (parent > 0) | |
1584 | type = BTRFS_SHARED_DATA_REF_KEY; | |
1585 | else | |
1586 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
1587 | } | |
1588 | return type; | |
31840ae1 | 1589 | } |
56bec294 | 1590 | |
2c47e605 YZ |
1591 | static int find_next_key(struct btrfs_path *path, int level, |
1592 | struct btrfs_key *key) | |
56bec294 | 1593 | |
02217ed2 | 1594 | { |
2c47e605 | 1595 | for (; level < BTRFS_MAX_LEVEL; level++) { |
5d4f98a2 YZ |
1596 | if (!path->nodes[level]) |
1597 | break; | |
5d4f98a2 YZ |
1598 | if (path->slots[level] + 1 >= |
1599 | btrfs_header_nritems(path->nodes[level])) | |
1600 | continue; | |
1601 | if (level == 0) | |
1602 | btrfs_item_key_to_cpu(path->nodes[level], key, | |
1603 | path->slots[level] + 1); | |
1604 | else | |
1605 | btrfs_node_key_to_cpu(path->nodes[level], key, | |
1606 | path->slots[level] + 1); | |
1607 | return 0; | |
1608 | } | |
1609 | return 1; | |
1610 | } | |
037e6390 | 1611 | |
5d4f98a2 YZ |
1612 | /* |
1613 | * look for inline back ref. if back ref is found, *ref_ret is set | |
1614 | * to the address of inline back ref, and 0 is returned. | |
1615 | * | |
1616 | * if back ref isn't found, *ref_ret is set to the address where it | |
1617 | * should be inserted, and -ENOENT is returned. | |
1618 | * | |
1619 | * if insert is true and there are too many inline back refs, the path | |
1620 | * points to the extent item, and -EAGAIN is returned. | |
1621 | * | |
1622 | * NOTE: inline back refs are ordered in the same way that back ref | |
1623 | * items in the tree are ordered. | |
1624 | */ | |
1625 | static noinline_for_stack | |
1626 | int lookup_inline_extent_backref(struct btrfs_trans_handle *trans, | |
87bde3cd | 1627 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1628 | struct btrfs_path *path, |
1629 | struct btrfs_extent_inline_ref **ref_ret, | |
1630 | u64 bytenr, u64 num_bytes, | |
1631 | u64 parent, u64 root_objectid, | |
1632 | u64 owner, u64 offset, int insert) | |
1633 | { | |
87bde3cd | 1634 | struct btrfs_root *root = fs_info->extent_root; |
5d4f98a2 YZ |
1635 | struct btrfs_key key; |
1636 | struct extent_buffer *leaf; | |
1637 | struct btrfs_extent_item *ei; | |
1638 | struct btrfs_extent_inline_ref *iref; | |
1639 | u64 flags; | |
1640 | u64 item_size; | |
1641 | unsigned long ptr; | |
1642 | unsigned long end; | |
1643 | int extra_size; | |
1644 | int type; | |
1645 | int want; | |
1646 | int ret; | |
1647 | int err = 0; | |
0b246afa | 1648 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); |
3de28d57 | 1649 | int needed; |
26b8003f | 1650 | |
db94535d | 1651 | key.objectid = bytenr; |
31840ae1 | 1652 | key.type = BTRFS_EXTENT_ITEM_KEY; |
56bec294 | 1653 | key.offset = num_bytes; |
31840ae1 | 1654 | |
5d4f98a2 YZ |
1655 | want = extent_ref_type(parent, owner); |
1656 | if (insert) { | |
1657 | extra_size = btrfs_extent_inline_ref_size(want); | |
85d4198e | 1658 | path->keep_locks = 1; |
5d4f98a2 YZ |
1659 | } else |
1660 | extra_size = -1; | |
3173a18f JB |
1661 | |
1662 | /* | |
1663 | * Owner is our parent level, so we can just add one to get the level | |
1664 | * for the block we are interested in. | |
1665 | */ | |
1666 | if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1667 | key.type = BTRFS_METADATA_ITEM_KEY; | |
1668 | key.offset = owner; | |
1669 | } | |
1670 | ||
1671 | again: | |
5d4f98a2 | 1672 | ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1); |
b9473439 | 1673 | if (ret < 0) { |
5d4f98a2 YZ |
1674 | err = ret; |
1675 | goto out; | |
1676 | } | |
3173a18f JB |
1677 | |
1678 | /* | |
1679 | * We may be a newly converted file system which still has the old fat | |
1680 | * extent entries for metadata, so try and see if we have one of those. | |
1681 | */ | |
1682 | if (ret > 0 && skinny_metadata) { | |
1683 | skinny_metadata = false; | |
1684 | if (path->slots[0]) { | |
1685 | path->slots[0]--; | |
1686 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
1687 | path->slots[0]); | |
1688 | if (key.objectid == bytenr && | |
1689 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
1690 | key.offset == num_bytes) | |
1691 | ret = 0; | |
1692 | } | |
1693 | if (ret) { | |
9ce49a0b | 1694 | key.objectid = bytenr; |
3173a18f JB |
1695 | key.type = BTRFS_EXTENT_ITEM_KEY; |
1696 | key.offset = num_bytes; | |
1697 | btrfs_release_path(path); | |
1698 | goto again; | |
1699 | } | |
1700 | } | |
1701 | ||
79787eaa JM |
1702 | if (ret && !insert) { |
1703 | err = -ENOENT; | |
1704 | goto out; | |
fae7f21c | 1705 | } else if (WARN_ON(ret)) { |
492104c8 | 1706 | err = -EIO; |
492104c8 | 1707 | goto out; |
79787eaa | 1708 | } |
5d4f98a2 YZ |
1709 | |
1710 | leaf = path->nodes[0]; | |
1711 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1712 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1713 | if (item_size < sizeof(*ei)) { | |
1714 | if (!insert) { | |
1715 | err = -ENOENT; | |
1716 | goto out; | |
1717 | } | |
87bde3cd | 1718 | ret = convert_extent_item_v0(trans, fs_info, path, owner, |
5d4f98a2 YZ |
1719 | extra_size); |
1720 | if (ret < 0) { | |
1721 | err = ret; | |
1722 | goto out; | |
1723 | } | |
1724 | leaf = path->nodes[0]; | |
1725 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1726 | } | |
1727 | #endif | |
1728 | BUG_ON(item_size < sizeof(*ei)); | |
1729 | ||
5d4f98a2 YZ |
1730 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1731 | flags = btrfs_extent_flags(leaf, ei); | |
1732 | ||
1733 | ptr = (unsigned long)(ei + 1); | |
1734 | end = (unsigned long)ei + item_size; | |
1735 | ||
3173a18f | 1736 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) { |
5d4f98a2 YZ |
1737 | ptr += sizeof(struct btrfs_tree_block_info); |
1738 | BUG_ON(ptr > end); | |
5d4f98a2 YZ |
1739 | } |
1740 | ||
3de28d57 LB |
1741 | if (owner >= BTRFS_FIRST_FREE_OBJECTID) |
1742 | needed = BTRFS_REF_TYPE_DATA; | |
1743 | else | |
1744 | needed = BTRFS_REF_TYPE_BLOCK; | |
1745 | ||
5d4f98a2 YZ |
1746 | err = -ENOENT; |
1747 | while (1) { | |
1748 | if (ptr >= end) { | |
1749 | WARN_ON(ptr > end); | |
1750 | break; | |
1751 | } | |
1752 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
3de28d57 LB |
1753 | type = btrfs_get_extent_inline_ref_type(leaf, iref, needed); |
1754 | if (type == BTRFS_REF_TYPE_INVALID) { | |
1755 | err = -EINVAL; | |
1756 | goto out; | |
1757 | } | |
1758 | ||
5d4f98a2 YZ |
1759 | if (want < type) |
1760 | break; | |
1761 | if (want > type) { | |
1762 | ptr += btrfs_extent_inline_ref_size(type); | |
1763 | continue; | |
1764 | } | |
1765 | ||
1766 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1767 | struct btrfs_extent_data_ref *dref; | |
1768 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1769 | if (match_extent_data_ref(leaf, dref, root_objectid, | |
1770 | owner, offset)) { | |
1771 | err = 0; | |
1772 | break; | |
1773 | } | |
1774 | if (hash_extent_data_ref_item(leaf, dref) < | |
1775 | hash_extent_data_ref(root_objectid, owner, offset)) | |
1776 | break; | |
1777 | } else { | |
1778 | u64 ref_offset; | |
1779 | ref_offset = btrfs_extent_inline_ref_offset(leaf, iref); | |
1780 | if (parent > 0) { | |
1781 | if (parent == ref_offset) { | |
1782 | err = 0; | |
1783 | break; | |
1784 | } | |
1785 | if (ref_offset < parent) | |
1786 | break; | |
1787 | } else { | |
1788 | if (root_objectid == ref_offset) { | |
1789 | err = 0; | |
1790 | break; | |
1791 | } | |
1792 | if (ref_offset < root_objectid) | |
1793 | break; | |
1794 | } | |
1795 | } | |
1796 | ptr += btrfs_extent_inline_ref_size(type); | |
1797 | } | |
1798 | if (err == -ENOENT && insert) { | |
1799 | if (item_size + extra_size >= | |
1800 | BTRFS_MAX_EXTENT_ITEM_SIZE(root)) { | |
1801 | err = -EAGAIN; | |
1802 | goto out; | |
1803 | } | |
1804 | /* | |
1805 | * To add new inline back ref, we have to make sure | |
1806 | * there is no corresponding back ref item. | |
1807 | * For simplicity, we just do not add new inline back | |
1808 | * ref if there is any kind of item for this block | |
1809 | */ | |
2c47e605 YZ |
1810 | if (find_next_key(path, 0, &key) == 0 && |
1811 | key.objectid == bytenr && | |
85d4198e | 1812 | key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) { |
5d4f98a2 YZ |
1813 | err = -EAGAIN; |
1814 | goto out; | |
1815 | } | |
1816 | } | |
1817 | *ref_ret = (struct btrfs_extent_inline_ref *)ptr; | |
1818 | out: | |
85d4198e | 1819 | if (insert) { |
5d4f98a2 YZ |
1820 | path->keep_locks = 0; |
1821 | btrfs_unlock_up_safe(path, 1); | |
1822 | } | |
1823 | return err; | |
1824 | } | |
1825 | ||
1826 | /* | |
1827 | * helper to add new inline back ref | |
1828 | */ | |
1829 | static noinline_for_stack | |
87bde3cd | 1830 | void setup_inline_extent_backref(struct btrfs_fs_info *fs_info, |
143bede5 JM |
1831 | struct btrfs_path *path, |
1832 | struct btrfs_extent_inline_ref *iref, | |
1833 | u64 parent, u64 root_objectid, | |
1834 | u64 owner, u64 offset, int refs_to_add, | |
1835 | struct btrfs_delayed_extent_op *extent_op) | |
5d4f98a2 YZ |
1836 | { |
1837 | struct extent_buffer *leaf; | |
1838 | struct btrfs_extent_item *ei; | |
1839 | unsigned long ptr; | |
1840 | unsigned long end; | |
1841 | unsigned long item_offset; | |
1842 | u64 refs; | |
1843 | int size; | |
1844 | int type; | |
5d4f98a2 YZ |
1845 | |
1846 | leaf = path->nodes[0]; | |
1847 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1848 | item_offset = (unsigned long)iref - (unsigned long)ei; | |
1849 | ||
1850 | type = extent_ref_type(parent, owner); | |
1851 | size = btrfs_extent_inline_ref_size(type); | |
1852 | ||
87bde3cd | 1853 | btrfs_extend_item(fs_info, path, size); |
5d4f98a2 YZ |
1854 | |
1855 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1856 | refs = btrfs_extent_refs(leaf, ei); | |
1857 | refs += refs_to_add; | |
1858 | btrfs_set_extent_refs(leaf, ei, refs); | |
1859 | if (extent_op) | |
1860 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1861 | ||
1862 | ptr = (unsigned long)ei + item_offset; | |
1863 | end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]); | |
1864 | if (ptr < end - size) | |
1865 | memmove_extent_buffer(leaf, ptr + size, ptr, | |
1866 | end - size - ptr); | |
1867 | ||
1868 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1869 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
1870 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1871 | struct btrfs_extent_data_ref *dref; | |
1872 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1873 | btrfs_set_extent_data_ref_root(leaf, dref, root_objectid); | |
1874 | btrfs_set_extent_data_ref_objectid(leaf, dref, owner); | |
1875 | btrfs_set_extent_data_ref_offset(leaf, dref, offset); | |
1876 | btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add); | |
1877 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1878 | struct btrfs_shared_data_ref *sref; | |
1879 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1880 | btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add); | |
1881 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1882 | } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) { | |
1883 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1884 | } else { | |
1885 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
1886 | } | |
1887 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1888 | } |
1889 | ||
1890 | static int lookup_extent_backref(struct btrfs_trans_handle *trans, | |
87bde3cd | 1891 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1892 | struct btrfs_path *path, |
1893 | struct btrfs_extent_inline_ref **ref_ret, | |
1894 | u64 bytenr, u64 num_bytes, u64 parent, | |
1895 | u64 root_objectid, u64 owner, u64 offset) | |
1896 | { | |
1897 | int ret; | |
1898 | ||
87bde3cd | 1899 | ret = lookup_inline_extent_backref(trans, fs_info, path, ref_ret, |
5d4f98a2 YZ |
1900 | bytenr, num_bytes, parent, |
1901 | root_objectid, owner, offset, 0); | |
1902 | if (ret != -ENOENT) | |
54aa1f4d | 1903 | return ret; |
5d4f98a2 | 1904 | |
b3b4aa74 | 1905 | btrfs_release_path(path); |
5d4f98a2 YZ |
1906 | *ref_ret = NULL; |
1907 | ||
1908 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
87bde3cd JM |
1909 | ret = lookup_tree_block_ref(trans, fs_info, path, bytenr, |
1910 | parent, root_objectid); | |
5d4f98a2 | 1911 | } else { |
87bde3cd JM |
1912 | ret = lookup_extent_data_ref(trans, fs_info, path, bytenr, |
1913 | parent, root_objectid, owner, | |
1914 | offset); | |
b9473439 | 1915 | } |
5d4f98a2 YZ |
1916 | return ret; |
1917 | } | |
31840ae1 | 1918 | |
5d4f98a2 YZ |
1919 | /* |
1920 | * helper to update/remove inline back ref | |
1921 | */ | |
1922 | static noinline_for_stack | |
87bde3cd | 1923 | void update_inline_extent_backref(struct btrfs_fs_info *fs_info, |
143bede5 JM |
1924 | struct btrfs_path *path, |
1925 | struct btrfs_extent_inline_ref *iref, | |
1926 | int refs_to_mod, | |
fcebe456 JB |
1927 | struct btrfs_delayed_extent_op *extent_op, |
1928 | int *last_ref) | |
5d4f98a2 YZ |
1929 | { |
1930 | struct extent_buffer *leaf; | |
1931 | struct btrfs_extent_item *ei; | |
1932 | struct btrfs_extent_data_ref *dref = NULL; | |
1933 | struct btrfs_shared_data_ref *sref = NULL; | |
1934 | unsigned long ptr; | |
1935 | unsigned long end; | |
1936 | u32 item_size; | |
1937 | int size; | |
1938 | int type; | |
5d4f98a2 YZ |
1939 | u64 refs; |
1940 | ||
1941 | leaf = path->nodes[0]; | |
1942 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1943 | refs = btrfs_extent_refs(leaf, ei); | |
1944 | WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0); | |
1945 | refs += refs_to_mod; | |
1946 | btrfs_set_extent_refs(leaf, ei, refs); | |
1947 | if (extent_op) | |
1948 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1949 | ||
3de28d57 LB |
1950 | /* |
1951 | * If type is invalid, we should have bailed out after | |
1952 | * lookup_inline_extent_backref(). | |
1953 | */ | |
1954 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_ANY); | |
1955 | ASSERT(type != BTRFS_REF_TYPE_INVALID); | |
5d4f98a2 YZ |
1956 | |
1957 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1958 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1959 | refs = btrfs_extent_data_ref_count(leaf, dref); | |
1960 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1961 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1962 | refs = btrfs_shared_data_ref_count(leaf, sref); | |
1963 | } else { | |
1964 | refs = 1; | |
1965 | BUG_ON(refs_to_mod != -1); | |
56bec294 | 1966 | } |
31840ae1 | 1967 | |
5d4f98a2 YZ |
1968 | BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod); |
1969 | refs += refs_to_mod; | |
1970 | ||
1971 | if (refs > 0) { | |
1972 | if (type == BTRFS_EXTENT_DATA_REF_KEY) | |
1973 | btrfs_set_extent_data_ref_count(leaf, dref, refs); | |
1974 | else | |
1975 | btrfs_set_shared_data_ref_count(leaf, sref, refs); | |
1976 | } else { | |
fcebe456 | 1977 | *last_ref = 1; |
5d4f98a2 YZ |
1978 | size = btrfs_extent_inline_ref_size(type); |
1979 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1980 | ptr = (unsigned long)iref; | |
1981 | end = (unsigned long)ei + item_size; | |
1982 | if (ptr + size < end) | |
1983 | memmove_extent_buffer(leaf, ptr, ptr + size, | |
1984 | end - ptr - size); | |
1985 | item_size -= size; | |
87bde3cd | 1986 | btrfs_truncate_item(fs_info, path, item_size, 1); |
5d4f98a2 YZ |
1987 | } |
1988 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1989 | } |
1990 | ||
1991 | static noinline_for_stack | |
1992 | int insert_inline_extent_backref(struct btrfs_trans_handle *trans, | |
87bde3cd | 1993 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1994 | struct btrfs_path *path, |
1995 | u64 bytenr, u64 num_bytes, u64 parent, | |
1996 | u64 root_objectid, u64 owner, | |
1997 | u64 offset, int refs_to_add, | |
1998 | struct btrfs_delayed_extent_op *extent_op) | |
1999 | { | |
2000 | struct btrfs_extent_inline_ref *iref; | |
2001 | int ret; | |
2002 | ||
87bde3cd | 2003 | ret = lookup_inline_extent_backref(trans, fs_info, path, &iref, |
5d4f98a2 YZ |
2004 | bytenr, num_bytes, parent, |
2005 | root_objectid, owner, offset, 1); | |
2006 | if (ret == 0) { | |
2007 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID); | |
87bde3cd | 2008 | update_inline_extent_backref(fs_info, path, iref, |
fcebe456 | 2009 | refs_to_add, extent_op, NULL); |
5d4f98a2 | 2010 | } else if (ret == -ENOENT) { |
87bde3cd | 2011 | setup_inline_extent_backref(fs_info, path, iref, parent, |
143bede5 JM |
2012 | root_objectid, owner, offset, |
2013 | refs_to_add, extent_op); | |
2014 | ret = 0; | |
771ed689 | 2015 | } |
5d4f98a2 YZ |
2016 | return ret; |
2017 | } | |
31840ae1 | 2018 | |
5d4f98a2 | 2019 | static int insert_extent_backref(struct btrfs_trans_handle *trans, |
87bde3cd | 2020 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
2021 | struct btrfs_path *path, |
2022 | u64 bytenr, u64 parent, u64 root_objectid, | |
2023 | u64 owner, u64 offset, int refs_to_add) | |
2024 | { | |
2025 | int ret; | |
2026 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
2027 | BUG_ON(refs_to_add != 1); | |
87bde3cd | 2028 | ret = insert_tree_block_ref(trans, fs_info, path, bytenr, |
5d4f98a2 YZ |
2029 | parent, root_objectid); |
2030 | } else { | |
87bde3cd | 2031 | ret = insert_extent_data_ref(trans, fs_info, path, bytenr, |
5d4f98a2 YZ |
2032 | parent, root_objectid, |
2033 | owner, offset, refs_to_add); | |
2034 | } | |
2035 | return ret; | |
2036 | } | |
56bec294 | 2037 | |
5d4f98a2 | 2038 | static int remove_extent_backref(struct btrfs_trans_handle *trans, |
87bde3cd | 2039 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
2040 | struct btrfs_path *path, |
2041 | struct btrfs_extent_inline_ref *iref, | |
fcebe456 | 2042 | int refs_to_drop, int is_data, int *last_ref) |
5d4f98a2 | 2043 | { |
143bede5 | 2044 | int ret = 0; |
b9473439 | 2045 | |
5d4f98a2 YZ |
2046 | BUG_ON(!is_data && refs_to_drop != 1); |
2047 | if (iref) { | |
87bde3cd | 2048 | update_inline_extent_backref(fs_info, path, iref, |
fcebe456 | 2049 | -refs_to_drop, NULL, last_ref); |
5d4f98a2 | 2050 | } else if (is_data) { |
87bde3cd | 2051 | ret = remove_extent_data_ref(trans, fs_info, path, refs_to_drop, |
fcebe456 | 2052 | last_ref); |
5d4f98a2 | 2053 | } else { |
fcebe456 | 2054 | *last_ref = 1; |
87bde3cd | 2055 | ret = btrfs_del_item(trans, fs_info->extent_root, path); |
5d4f98a2 YZ |
2056 | } |
2057 | return ret; | |
2058 | } | |
2059 | ||
86557861 | 2060 | #define in_range(b, first, len) ((b) >= (first) && (b) < (first) + (len)) |
d04c6b88 JM |
2061 | static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len, |
2062 | u64 *discarded_bytes) | |
5d4f98a2 | 2063 | { |
86557861 JM |
2064 | int j, ret = 0; |
2065 | u64 bytes_left, end; | |
4d89d377 | 2066 | u64 aligned_start = ALIGN(start, 1 << 9); |
d04c6b88 | 2067 | |
4d89d377 JM |
2068 | if (WARN_ON(start != aligned_start)) { |
2069 | len -= aligned_start - start; | |
2070 | len = round_down(len, 1 << 9); | |
2071 | start = aligned_start; | |
2072 | } | |
d04c6b88 | 2073 | |
4d89d377 | 2074 | *discarded_bytes = 0; |
86557861 JM |
2075 | |
2076 | if (!len) | |
2077 | return 0; | |
2078 | ||
2079 | end = start + len; | |
2080 | bytes_left = len; | |
2081 | ||
2082 | /* Skip any superblocks on this device. */ | |
2083 | for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) { | |
2084 | u64 sb_start = btrfs_sb_offset(j); | |
2085 | u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE; | |
2086 | u64 size = sb_start - start; | |
2087 | ||
2088 | if (!in_range(sb_start, start, bytes_left) && | |
2089 | !in_range(sb_end, start, bytes_left) && | |
2090 | !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE)) | |
2091 | continue; | |
2092 | ||
2093 | /* | |
2094 | * Superblock spans beginning of range. Adjust start and | |
2095 | * try again. | |
2096 | */ | |
2097 | if (sb_start <= start) { | |
2098 | start += sb_end - start; | |
2099 | if (start > end) { | |
2100 | bytes_left = 0; | |
2101 | break; | |
2102 | } | |
2103 | bytes_left = end - start; | |
2104 | continue; | |
2105 | } | |
2106 | ||
2107 | if (size) { | |
2108 | ret = blkdev_issue_discard(bdev, start >> 9, size >> 9, | |
2109 | GFP_NOFS, 0); | |
2110 | if (!ret) | |
2111 | *discarded_bytes += size; | |
2112 | else if (ret != -EOPNOTSUPP) | |
2113 | return ret; | |
2114 | } | |
2115 | ||
2116 | start = sb_end; | |
2117 | if (start > end) { | |
2118 | bytes_left = 0; | |
2119 | break; | |
2120 | } | |
2121 | bytes_left = end - start; | |
2122 | } | |
2123 | ||
2124 | if (bytes_left) { | |
2125 | ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9, | |
4d89d377 JM |
2126 | GFP_NOFS, 0); |
2127 | if (!ret) | |
86557861 | 2128 | *discarded_bytes += bytes_left; |
4d89d377 | 2129 | } |
d04c6b88 | 2130 | return ret; |
5d4f98a2 | 2131 | } |
5d4f98a2 | 2132 | |
2ff7e61e | 2133 | int btrfs_discard_extent(struct btrfs_fs_info *fs_info, u64 bytenr, |
1edb647b | 2134 | u64 num_bytes, u64 *actual_bytes) |
5d4f98a2 | 2135 | { |
5d4f98a2 | 2136 | int ret; |
5378e607 | 2137 | u64 discarded_bytes = 0; |
a1d3c478 | 2138 | struct btrfs_bio *bbio = NULL; |
5d4f98a2 | 2139 | |
e244a0ae | 2140 | |
2999241d FM |
2141 | /* |
2142 | * Avoid races with device replace and make sure our bbio has devices | |
2143 | * associated to its stripes that don't go away while we are discarding. | |
2144 | */ | |
0b246afa | 2145 | btrfs_bio_counter_inc_blocked(fs_info); |
5d4f98a2 | 2146 | /* Tell the block device(s) that the sectors can be discarded */ |
0b246afa JM |
2147 | ret = btrfs_map_block(fs_info, BTRFS_MAP_DISCARD, bytenr, &num_bytes, |
2148 | &bbio, 0); | |
79787eaa | 2149 | /* Error condition is -ENOMEM */ |
5d4f98a2 | 2150 | if (!ret) { |
a1d3c478 | 2151 | struct btrfs_bio_stripe *stripe = bbio->stripes; |
5d4f98a2 YZ |
2152 | int i; |
2153 | ||
5d4f98a2 | 2154 | |
a1d3c478 | 2155 | for (i = 0; i < bbio->num_stripes; i++, stripe++) { |
d04c6b88 | 2156 | u64 bytes; |
d5e2003c JB |
2157 | if (!stripe->dev->can_discard) |
2158 | continue; | |
2159 | ||
5378e607 LD |
2160 | ret = btrfs_issue_discard(stripe->dev->bdev, |
2161 | stripe->physical, | |
d04c6b88 JM |
2162 | stripe->length, |
2163 | &bytes); | |
5378e607 | 2164 | if (!ret) |
d04c6b88 | 2165 | discarded_bytes += bytes; |
5378e607 | 2166 | else if (ret != -EOPNOTSUPP) |
79787eaa | 2167 | break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */ |
d5e2003c JB |
2168 | |
2169 | /* | |
2170 | * Just in case we get back EOPNOTSUPP for some reason, | |
2171 | * just ignore the return value so we don't screw up | |
2172 | * people calling discard_extent. | |
2173 | */ | |
2174 | ret = 0; | |
5d4f98a2 | 2175 | } |
6e9606d2 | 2176 | btrfs_put_bbio(bbio); |
5d4f98a2 | 2177 | } |
0b246afa | 2178 | btrfs_bio_counter_dec(fs_info); |
5378e607 LD |
2179 | |
2180 | if (actual_bytes) | |
2181 | *actual_bytes = discarded_bytes; | |
2182 | ||
5d4f98a2 | 2183 | |
53b381b3 DW |
2184 | if (ret == -EOPNOTSUPP) |
2185 | ret = 0; | |
5d4f98a2 | 2186 | return ret; |
5d4f98a2 YZ |
2187 | } |
2188 | ||
79787eaa | 2189 | /* Can return -ENOMEM */ |
5d4f98a2 | 2190 | int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
2ff7e61e | 2191 | struct btrfs_fs_info *fs_info, |
5d4f98a2 | 2192 | u64 bytenr, u64 num_bytes, u64 parent, |
b06c4bf5 | 2193 | u64 root_objectid, u64 owner, u64 offset) |
5d4f98a2 | 2194 | { |
d7eae340 | 2195 | int old_ref_mod, new_ref_mod; |
5d4f98a2 | 2196 | int ret; |
66d7e7f0 | 2197 | |
5d4f98a2 YZ |
2198 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID && |
2199 | root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
2200 | ||
2201 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
66d7e7f0 | 2202 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr, |
7be07912 OS |
2203 | num_bytes, parent, |
2204 | root_objectid, (int)owner, | |
2205 | BTRFS_ADD_DELAYED_REF, NULL, | |
d7eae340 | 2206 | &old_ref_mod, &new_ref_mod); |
5d4f98a2 | 2207 | } else { |
66d7e7f0 | 2208 | ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr, |
7be07912 OS |
2209 | num_bytes, parent, |
2210 | root_objectid, owner, offset, | |
d7eae340 OS |
2211 | 0, BTRFS_ADD_DELAYED_REF, |
2212 | &old_ref_mod, &new_ref_mod); | |
5d4f98a2 | 2213 | } |
d7eae340 OS |
2214 | |
2215 | if (ret == 0 && old_ref_mod < 0 && new_ref_mod >= 0) | |
2216 | add_pinned_bytes(fs_info, -num_bytes, owner, root_objectid); | |
2217 | ||
5d4f98a2 YZ |
2218 | return ret; |
2219 | } | |
2220 | ||
2221 | static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, | |
2ff7e61e | 2222 | struct btrfs_fs_info *fs_info, |
c682f9b3 | 2223 | struct btrfs_delayed_ref_node *node, |
5d4f98a2 YZ |
2224 | u64 parent, u64 root_objectid, |
2225 | u64 owner, u64 offset, int refs_to_add, | |
2226 | struct btrfs_delayed_extent_op *extent_op) | |
2227 | { | |
2228 | struct btrfs_path *path; | |
2229 | struct extent_buffer *leaf; | |
2230 | struct btrfs_extent_item *item; | |
fcebe456 | 2231 | struct btrfs_key key; |
c682f9b3 QW |
2232 | u64 bytenr = node->bytenr; |
2233 | u64 num_bytes = node->num_bytes; | |
5d4f98a2 YZ |
2234 | u64 refs; |
2235 | int ret; | |
5d4f98a2 YZ |
2236 | |
2237 | path = btrfs_alloc_path(); | |
2238 | if (!path) | |
2239 | return -ENOMEM; | |
2240 | ||
e4058b54 | 2241 | path->reada = READA_FORWARD; |
5d4f98a2 YZ |
2242 | path->leave_spinning = 1; |
2243 | /* this will setup the path even if it fails to insert the back ref */ | |
87bde3cd JM |
2244 | ret = insert_inline_extent_backref(trans, fs_info, path, bytenr, |
2245 | num_bytes, parent, root_objectid, | |
2246 | owner, offset, | |
5d4f98a2 | 2247 | refs_to_add, extent_op); |
0ed4792a | 2248 | if ((ret < 0 && ret != -EAGAIN) || !ret) |
5d4f98a2 | 2249 | goto out; |
fcebe456 JB |
2250 | |
2251 | /* | |
2252 | * Ok we had -EAGAIN which means we didn't have space to insert and | |
2253 | * inline extent ref, so just update the reference count and add a | |
2254 | * normal backref. | |
2255 | */ | |
5d4f98a2 | 2256 | leaf = path->nodes[0]; |
fcebe456 | 2257 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
5d4f98a2 YZ |
2258 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
2259 | refs = btrfs_extent_refs(leaf, item); | |
2260 | btrfs_set_extent_refs(leaf, item, refs + refs_to_add); | |
2261 | if (extent_op) | |
2262 | __run_delayed_extent_op(extent_op, leaf, item); | |
56bec294 | 2263 | |
5d4f98a2 | 2264 | btrfs_mark_buffer_dirty(leaf); |
b3b4aa74 | 2265 | btrfs_release_path(path); |
56bec294 | 2266 | |
e4058b54 | 2267 | path->reada = READA_FORWARD; |
b9473439 | 2268 | path->leave_spinning = 1; |
56bec294 | 2269 | /* now insert the actual backref */ |
87bde3cd JM |
2270 | ret = insert_extent_backref(trans, fs_info, path, bytenr, parent, |
2271 | root_objectid, owner, offset, refs_to_add); | |
79787eaa | 2272 | if (ret) |
66642832 | 2273 | btrfs_abort_transaction(trans, ret); |
5d4f98a2 | 2274 | out: |
56bec294 | 2275 | btrfs_free_path(path); |
30d133fc | 2276 | return ret; |
56bec294 CM |
2277 | } |
2278 | ||
5d4f98a2 | 2279 | static int run_delayed_data_ref(struct btrfs_trans_handle *trans, |
2ff7e61e | 2280 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
2281 | struct btrfs_delayed_ref_node *node, |
2282 | struct btrfs_delayed_extent_op *extent_op, | |
2283 | int insert_reserved) | |
56bec294 | 2284 | { |
5d4f98a2 YZ |
2285 | int ret = 0; |
2286 | struct btrfs_delayed_data_ref *ref; | |
2287 | struct btrfs_key ins; | |
2288 | u64 parent = 0; | |
2289 | u64 ref_root = 0; | |
2290 | u64 flags = 0; | |
2291 | ||
2292 | ins.objectid = node->bytenr; | |
2293 | ins.offset = node->num_bytes; | |
2294 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
2295 | ||
2296 | ref = btrfs_delayed_node_to_data_ref(node); | |
0b246afa | 2297 | trace_run_delayed_data_ref(fs_info, node, ref, node->action); |
599c75ec | 2298 | |
5d4f98a2 YZ |
2299 | if (node->type == BTRFS_SHARED_DATA_REF_KEY) |
2300 | parent = ref->parent; | |
fcebe456 | 2301 | ref_root = ref->root; |
5d4f98a2 YZ |
2302 | |
2303 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
3173a18f | 2304 | if (extent_op) |
5d4f98a2 | 2305 | flags |= extent_op->flags_to_set; |
2ff7e61e | 2306 | ret = alloc_reserved_file_extent(trans, fs_info, |
5d4f98a2 YZ |
2307 | parent, ref_root, flags, |
2308 | ref->objectid, ref->offset, | |
2309 | &ins, node->ref_mod); | |
5d4f98a2 | 2310 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2ff7e61e | 2311 | ret = __btrfs_inc_extent_ref(trans, fs_info, node, parent, |
5d4f98a2 YZ |
2312 | ref_root, ref->objectid, |
2313 | ref->offset, node->ref_mod, | |
c682f9b3 | 2314 | extent_op); |
5d4f98a2 | 2315 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
2ff7e61e | 2316 | ret = __btrfs_free_extent(trans, fs_info, node, parent, |
5d4f98a2 YZ |
2317 | ref_root, ref->objectid, |
2318 | ref->offset, node->ref_mod, | |
c682f9b3 | 2319 | extent_op); |
5d4f98a2 YZ |
2320 | } else { |
2321 | BUG(); | |
2322 | } | |
2323 | return ret; | |
2324 | } | |
2325 | ||
2326 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
2327 | struct extent_buffer *leaf, | |
2328 | struct btrfs_extent_item *ei) | |
2329 | { | |
2330 | u64 flags = btrfs_extent_flags(leaf, ei); | |
2331 | if (extent_op->update_flags) { | |
2332 | flags |= extent_op->flags_to_set; | |
2333 | btrfs_set_extent_flags(leaf, ei, flags); | |
2334 | } | |
2335 | ||
2336 | if (extent_op->update_key) { | |
2337 | struct btrfs_tree_block_info *bi; | |
2338 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)); | |
2339 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
2340 | btrfs_set_tree_block_key(leaf, bi, &extent_op->key); | |
2341 | } | |
2342 | } | |
2343 | ||
2344 | static int run_delayed_extent_op(struct btrfs_trans_handle *trans, | |
2ff7e61e | 2345 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
2346 | struct btrfs_delayed_ref_node *node, |
2347 | struct btrfs_delayed_extent_op *extent_op) | |
2348 | { | |
2349 | struct btrfs_key key; | |
2350 | struct btrfs_path *path; | |
2351 | struct btrfs_extent_item *ei; | |
2352 | struct extent_buffer *leaf; | |
2353 | u32 item_size; | |
56bec294 | 2354 | int ret; |
5d4f98a2 | 2355 | int err = 0; |
b1c79e09 | 2356 | int metadata = !extent_op->is_data; |
5d4f98a2 | 2357 | |
79787eaa JM |
2358 | if (trans->aborted) |
2359 | return 0; | |
2360 | ||
0b246afa | 2361 | if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) |
3173a18f JB |
2362 | metadata = 0; |
2363 | ||
5d4f98a2 YZ |
2364 | path = btrfs_alloc_path(); |
2365 | if (!path) | |
2366 | return -ENOMEM; | |
2367 | ||
2368 | key.objectid = node->bytenr; | |
5d4f98a2 | 2369 | |
3173a18f | 2370 | if (metadata) { |
3173a18f | 2371 | key.type = BTRFS_METADATA_ITEM_KEY; |
b1c79e09 | 2372 | key.offset = extent_op->level; |
3173a18f JB |
2373 | } else { |
2374 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
2375 | key.offset = node->num_bytes; | |
2376 | } | |
2377 | ||
2378 | again: | |
e4058b54 | 2379 | path->reada = READA_FORWARD; |
5d4f98a2 | 2380 | path->leave_spinning = 1; |
0b246afa | 2381 | ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 1); |
5d4f98a2 YZ |
2382 | if (ret < 0) { |
2383 | err = ret; | |
2384 | goto out; | |
2385 | } | |
2386 | if (ret > 0) { | |
3173a18f | 2387 | if (metadata) { |
55994887 FDBM |
2388 | if (path->slots[0] > 0) { |
2389 | path->slots[0]--; | |
2390 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
2391 | path->slots[0]); | |
2392 | if (key.objectid == node->bytenr && | |
2393 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
2394 | key.offset == node->num_bytes) | |
2395 | ret = 0; | |
2396 | } | |
2397 | if (ret > 0) { | |
2398 | btrfs_release_path(path); | |
2399 | metadata = 0; | |
3173a18f | 2400 | |
55994887 FDBM |
2401 | key.objectid = node->bytenr; |
2402 | key.offset = node->num_bytes; | |
2403 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
2404 | goto again; | |
2405 | } | |
2406 | } else { | |
2407 | err = -EIO; | |
2408 | goto out; | |
3173a18f | 2409 | } |
5d4f98a2 YZ |
2410 | } |
2411 | ||
2412 | leaf = path->nodes[0]; | |
2413 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2414 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
2415 | if (item_size < sizeof(*ei)) { | |
87bde3cd | 2416 | ret = convert_extent_item_v0(trans, fs_info, path, (u64)-1, 0); |
5d4f98a2 YZ |
2417 | if (ret < 0) { |
2418 | err = ret; | |
2419 | goto out; | |
2420 | } | |
2421 | leaf = path->nodes[0]; | |
2422 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2423 | } | |
2424 | #endif | |
2425 | BUG_ON(item_size < sizeof(*ei)); | |
2426 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
2427 | __run_delayed_extent_op(extent_op, leaf, ei); | |
56bec294 | 2428 | |
5d4f98a2 YZ |
2429 | btrfs_mark_buffer_dirty(leaf); |
2430 | out: | |
2431 | btrfs_free_path(path); | |
2432 | return err; | |
56bec294 CM |
2433 | } |
2434 | ||
5d4f98a2 | 2435 | static int run_delayed_tree_ref(struct btrfs_trans_handle *trans, |
2ff7e61e | 2436 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
2437 | struct btrfs_delayed_ref_node *node, |
2438 | struct btrfs_delayed_extent_op *extent_op, | |
2439 | int insert_reserved) | |
56bec294 CM |
2440 | { |
2441 | int ret = 0; | |
5d4f98a2 YZ |
2442 | struct btrfs_delayed_tree_ref *ref; |
2443 | struct btrfs_key ins; | |
2444 | u64 parent = 0; | |
2445 | u64 ref_root = 0; | |
0b246afa | 2446 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); |
56bec294 | 2447 | |
5d4f98a2 | 2448 | ref = btrfs_delayed_node_to_tree_ref(node); |
0b246afa | 2449 | trace_run_delayed_tree_ref(fs_info, node, ref, node->action); |
599c75ec | 2450 | |
5d4f98a2 YZ |
2451 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) |
2452 | parent = ref->parent; | |
fcebe456 | 2453 | ref_root = ref->root; |
5d4f98a2 | 2454 | |
3173a18f JB |
2455 | ins.objectid = node->bytenr; |
2456 | if (skinny_metadata) { | |
2457 | ins.offset = ref->level; | |
2458 | ins.type = BTRFS_METADATA_ITEM_KEY; | |
2459 | } else { | |
2460 | ins.offset = node->num_bytes; | |
2461 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
2462 | } | |
2463 | ||
02794222 | 2464 | if (node->ref_mod != 1) { |
2ff7e61e | 2465 | btrfs_err(fs_info, |
02794222 LB |
2466 | "btree block(%llu) has %d references rather than 1: action %d ref_root %llu parent %llu", |
2467 | node->bytenr, node->ref_mod, node->action, ref_root, | |
2468 | parent); | |
2469 | return -EIO; | |
2470 | } | |
5d4f98a2 | 2471 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { |
3173a18f | 2472 | BUG_ON(!extent_op || !extent_op->update_flags); |
2ff7e61e | 2473 | ret = alloc_reserved_tree_block(trans, fs_info, |
5d4f98a2 YZ |
2474 | parent, ref_root, |
2475 | extent_op->flags_to_set, | |
2476 | &extent_op->key, | |
b06c4bf5 | 2477 | ref->level, &ins); |
5d4f98a2 | 2478 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2ff7e61e | 2479 | ret = __btrfs_inc_extent_ref(trans, fs_info, node, |
c682f9b3 QW |
2480 | parent, ref_root, |
2481 | ref->level, 0, 1, | |
fcebe456 | 2482 | extent_op); |
5d4f98a2 | 2483 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
2ff7e61e | 2484 | ret = __btrfs_free_extent(trans, fs_info, node, |
c682f9b3 QW |
2485 | parent, ref_root, |
2486 | ref->level, 0, 1, extent_op); | |
5d4f98a2 YZ |
2487 | } else { |
2488 | BUG(); | |
2489 | } | |
56bec294 CM |
2490 | return ret; |
2491 | } | |
2492 | ||
2493 | /* helper function to actually process a single delayed ref entry */ | |
5d4f98a2 | 2494 | static int run_one_delayed_ref(struct btrfs_trans_handle *trans, |
2ff7e61e | 2495 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
2496 | struct btrfs_delayed_ref_node *node, |
2497 | struct btrfs_delayed_extent_op *extent_op, | |
2498 | int insert_reserved) | |
56bec294 | 2499 | { |
79787eaa JM |
2500 | int ret = 0; |
2501 | ||
857cc2fc JB |
2502 | if (trans->aborted) { |
2503 | if (insert_reserved) | |
2ff7e61e | 2504 | btrfs_pin_extent(fs_info, node->bytenr, |
857cc2fc | 2505 | node->num_bytes, 1); |
79787eaa | 2506 | return 0; |
857cc2fc | 2507 | } |
79787eaa | 2508 | |
5d4f98a2 | 2509 | if (btrfs_delayed_ref_is_head(node)) { |
56bec294 CM |
2510 | struct btrfs_delayed_ref_head *head; |
2511 | /* | |
2512 | * we've hit the end of the chain and we were supposed | |
2513 | * to insert this extent into the tree. But, it got | |
2514 | * deleted before we ever needed to insert it, so all | |
2515 | * we have to do is clean up the accounting | |
2516 | */ | |
5d4f98a2 YZ |
2517 | BUG_ON(extent_op); |
2518 | head = btrfs_delayed_node_to_head(node); | |
0b246afa | 2519 | trace_run_delayed_ref_head(fs_info, node, head, node->action); |
599c75ec | 2520 | |
d7eae340 OS |
2521 | if (head->total_ref_mod < 0) { |
2522 | struct btrfs_block_group_cache *cache; | |
2523 | ||
2524 | cache = btrfs_lookup_block_group(fs_info, node->bytenr); | |
2525 | ASSERT(cache); | |
2526 | percpu_counter_add(&cache->space_info->total_bytes_pinned, | |
2527 | -node->num_bytes); | |
2528 | btrfs_put_block_group(cache); | |
2529 | } | |
2530 | ||
56bec294 | 2531 | if (insert_reserved) { |
2ff7e61e | 2532 | btrfs_pin_extent(fs_info, node->bytenr, |
f0486c68 | 2533 | node->num_bytes, 1); |
5d4f98a2 | 2534 | if (head->is_data) { |
0b246afa | 2535 | ret = btrfs_del_csums(trans, fs_info, |
5d4f98a2 YZ |
2536 | node->bytenr, |
2537 | node->num_bytes); | |
5d4f98a2 | 2538 | } |
56bec294 | 2539 | } |
297d750b QW |
2540 | |
2541 | /* Also free its reserved qgroup space */ | |
0b246afa | 2542 | btrfs_qgroup_free_delayed_ref(fs_info, head->qgroup_ref_root, |
297d750b | 2543 | head->qgroup_reserved); |
79787eaa | 2544 | return ret; |
56bec294 CM |
2545 | } |
2546 | ||
5d4f98a2 YZ |
2547 | if (node->type == BTRFS_TREE_BLOCK_REF_KEY || |
2548 | node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
2ff7e61e | 2549 | ret = run_delayed_tree_ref(trans, fs_info, node, extent_op, |
5d4f98a2 YZ |
2550 | insert_reserved); |
2551 | else if (node->type == BTRFS_EXTENT_DATA_REF_KEY || | |
2552 | node->type == BTRFS_SHARED_DATA_REF_KEY) | |
2ff7e61e | 2553 | ret = run_delayed_data_ref(trans, fs_info, node, extent_op, |
5d4f98a2 YZ |
2554 | insert_reserved); |
2555 | else | |
2556 | BUG(); | |
2557 | return ret; | |
56bec294 CM |
2558 | } |
2559 | ||
c6fc2454 | 2560 | static inline struct btrfs_delayed_ref_node * |
56bec294 CM |
2561 | select_delayed_ref(struct btrfs_delayed_ref_head *head) |
2562 | { | |
cffc3374 FM |
2563 | struct btrfs_delayed_ref_node *ref; |
2564 | ||
c6fc2454 QW |
2565 | if (list_empty(&head->ref_list)) |
2566 | return NULL; | |
d7df2c79 | 2567 | |
cffc3374 FM |
2568 | /* |
2569 | * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first. | |
2570 | * This is to prevent a ref count from going down to zero, which deletes | |
2571 | * the extent item from the extent tree, when there still are references | |
2572 | * to add, which would fail because they would not find the extent item. | |
2573 | */ | |
1d57ee94 WX |
2574 | if (!list_empty(&head->ref_add_list)) |
2575 | return list_first_entry(&head->ref_add_list, | |
2576 | struct btrfs_delayed_ref_node, add_list); | |
2577 | ||
2578 | ref = list_first_entry(&head->ref_list, struct btrfs_delayed_ref_node, | |
2579 | list); | |
2580 | ASSERT(list_empty(&ref->add_list)); | |
2581 | return ref; | |
56bec294 CM |
2582 | } |
2583 | ||
79787eaa JM |
2584 | /* |
2585 | * Returns 0 on success or if called with an already aborted transaction. | |
2586 | * Returns -ENOMEM or -EIO on failure and will abort the transaction. | |
2587 | */ | |
d7df2c79 | 2588 | static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, |
2ff7e61e | 2589 | struct btrfs_fs_info *fs_info, |
d7df2c79 | 2590 | unsigned long nr) |
56bec294 | 2591 | { |
56bec294 CM |
2592 | struct btrfs_delayed_ref_root *delayed_refs; |
2593 | struct btrfs_delayed_ref_node *ref; | |
2594 | struct btrfs_delayed_ref_head *locked_ref = NULL; | |
5d4f98a2 | 2595 | struct btrfs_delayed_extent_op *extent_op; |
0a2b2a84 | 2596 | ktime_t start = ktime_get(); |
56bec294 | 2597 | int ret; |
d7df2c79 | 2598 | unsigned long count = 0; |
0a2b2a84 | 2599 | unsigned long actual_count = 0; |
56bec294 | 2600 | int must_insert_reserved = 0; |
56bec294 CM |
2601 | |
2602 | delayed_refs = &trans->transaction->delayed_refs; | |
56bec294 CM |
2603 | while (1) { |
2604 | if (!locked_ref) { | |
d7df2c79 | 2605 | if (count >= nr) |
56bec294 | 2606 | break; |
56bec294 | 2607 | |
d7df2c79 JB |
2608 | spin_lock(&delayed_refs->lock); |
2609 | locked_ref = btrfs_select_ref_head(trans); | |
2610 | if (!locked_ref) { | |
2611 | spin_unlock(&delayed_refs->lock); | |
2612 | break; | |
2613 | } | |
c3e69d58 CM |
2614 | |
2615 | /* grab the lock that says we are going to process | |
2616 | * all the refs for this head */ | |
2617 | ret = btrfs_delayed_ref_lock(trans, locked_ref); | |
d7df2c79 | 2618 | spin_unlock(&delayed_refs->lock); |
c3e69d58 CM |
2619 | /* |
2620 | * we may have dropped the spin lock to get the head | |
2621 | * mutex lock, and that might have given someone else | |
2622 | * time to free the head. If that's true, it has been | |
2623 | * removed from our list and we can move on. | |
2624 | */ | |
2625 | if (ret == -EAGAIN) { | |
2626 | locked_ref = NULL; | |
2627 | count++; | |
2628 | continue; | |
56bec294 CM |
2629 | } |
2630 | } | |
a28ec197 | 2631 | |
2c3cf7d5 FM |
2632 | /* |
2633 | * We need to try and merge add/drops of the same ref since we | |
2634 | * can run into issues with relocate dropping the implicit ref | |
2635 | * and then it being added back again before the drop can | |
2636 | * finish. If we merged anything we need to re-loop so we can | |
2637 | * get a good ref. | |
2638 | * Or we can get node references of the same type that weren't | |
2639 | * merged when created due to bumps in the tree mod seq, and | |
2640 | * we need to merge them to prevent adding an inline extent | |
2641 | * backref before dropping it (triggering a BUG_ON at | |
2642 | * insert_inline_extent_backref()). | |
2643 | */ | |
d7df2c79 | 2644 | spin_lock(&locked_ref->lock); |
2c3cf7d5 FM |
2645 | btrfs_merge_delayed_refs(trans, fs_info, delayed_refs, |
2646 | locked_ref); | |
ae1e206b | 2647 | |
d1270cd9 AJ |
2648 | /* |
2649 | * locked_ref is the head node, so we have to go one | |
2650 | * node back for any delayed ref updates | |
2651 | */ | |
2652 | ref = select_delayed_ref(locked_ref); | |
2653 | ||
2654 | if (ref && ref->seq && | |
097b8a7c | 2655 | btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) { |
d7df2c79 | 2656 | spin_unlock(&locked_ref->lock); |
d7df2c79 JB |
2657 | spin_lock(&delayed_refs->lock); |
2658 | locked_ref->processing = 0; | |
d1270cd9 AJ |
2659 | delayed_refs->num_heads_ready++; |
2660 | spin_unlock(&delayed_refs->lock); | |
d0280996 | 2661 | btrfs_delayed_ref_unlock(locked_ref); |
d7df2c79 | 2662 | locked_ref = NULL; |
d1270cd9 | 2663 | cond_resched(); |
27a377db | 2664 | count++; |
d1270cd9 AJ |
2665 | continue; |
2666 | } | |
2667 | ||
56bec294 CM |
2668 | /* |
2669 | * record the must insert reserved flag before we | |
2670 | * drop the spin lock. | |
2671 | */ | |
2672 | must_insert_reserved = locked_ref->must_insert_reserved; | |
2673 | locked_ref->must_insert_reserved = 0; | |
7bb86316 | 2674 | |
5d4f98a2 YZ |
2675 | extent_op = locked_ref->extent_op; |
2676 | locked_ref->extent_op = NULL; | |
2677 | ||
56bec294 | 2678 | if (!ref) { |
d7df2c79 JB |
2679 | |
2680 | ||
56bec294 CM |
2681 | /* All delayed refs have been processed, Go ahead |
2682 | * and send the head node to run_one_delayed_ref, | |
2683 | * so that any accounting fixes can happen | |
2684 | */ | |
2685 | ref = &locked_ref->node; | |
5d4f98a2 YZ |
2686 | |
2687 | if (extent_op && must_insert_reserved) { | |
78a6184a | 2688 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 YZ |
2689 | extent_op = NULL; |
2690 | } | |
2691 | ||
2692 | if (extent_op) { | |
d7df2c79 | 2693 | spin_unlock(&locked_ref->lock); |
2ff7e61e | 2694 | ret = run_delayed_extent_op(trans, fs_info, |
5d4f98a2 | 2695 | ref, extent_op); |
78a6184a | 2696 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 | 2697 | |
79787eaa | 2698 | if (ret) { |
857cc2fc JB |
2699 | /* |
2700 | * Need to reset must_insert_reserved if | |
2701 | * there was an error so the abort stuff | |
2702 | * can cleanup the reserved space | |
2703 | * properly. | |
2704 | */ | |
2705 | if (must_insert_reserved) | |
2706 | locked_ref->must_insert_reserved = 1; | |
aa7c8da3 | 2707 | spin_lock(&delayed_refs->lock); |
d7df2c79 | 2708 | locked_ref->processing = 0; |
aa7c8da3 JM |
2709 | delayed_refs->num_heads_ready++; |
2710 | spin_unlock(&delayed_refs->lock); | |
5d163e0e JM |
2711 | btrfs_debug(fs_info, |
2712 | "run_delayed_extent_op returned %d", | |
2713 | ret); | |
093486c4 | 2714 | btrfs_delayed_ref_unlock(locked_ref); |
79787eaa JM |
2715 | return ret; |
2716 | } | |
d7df2c79 | 2717 | continue; |
5d4f98a2 | 2718 | } |
02217ed2 | 2719 | |
d7df2c79 | 2720 | /* |
01327610 | 2721 | * Need to drop our head ref lock and re-acquire the |
d7df2c79 JB |
2722 | * delayed ref lock and then re-check to make sure |
2723 | * nobody got added. | |
2724 | */ | |
2725 | spin_unlock(&locked_ref->lock); | |
2726 | spin_lock(&delayed_refs->lock); | |
2727 | spin_lock(&locked_ref->lock); | |
c6fc2454 | 2728 | if (!list_empty(&locked_ref->ref_list) || |
573a0755 | 2729 | locked_ref->extent_op) { |
d7df2c79 JB |
2730 | spin_unlock(&locked_ref->lock); |
2731 | spin_unlock(&delayed_refs->lock); | |
2732 | continue; | |
2733 | } | |
2734 | ref->in_tree = 0; | |
2735 | delayed_refs->num_heads--; | |
c46effa6 LB |
2736 | rb_erase(&locked_ref->href_node, |
2737 | &delayed_refs->href_root); | |
d7df2c79 JB |
2738 | spin_unlock(&delayed_refs->lock); |
2739 | } else { | |
0a2b2a84 | 2740 | actual_count++; |
d7df2c79 | 2741 | ref->in_tree = 0; |
c6fc2454 | 2742 | list_del(&ref->list); |
1d57ee94 WX |
2743 | if (!list_empty(&ref->add_list)) |
2744 | list_del(&ref->add_list); | |
c46effa6 | 2745 | } |
d7df2c79 JB |
2746 | atomic_dec(&delayed_refs->num_entries); |
2747 | ||
093486c4 | 2748 | if (!btrfs_delayed_ref_is_head(ref)) { |
22cd2e7d AJ |
2749 | /* |
2750 | * when we play the delayed ref, also correct the | |
2751 | * ref_mod on head | |
2752 | */ | |
2753 | switch (ref->action) { | |
2754 | case BTRFS_ADD_DELAYED_REF: | |
2755 | case BTRFS_ADD_DELAYED_EXTENT: | |
2756 | locked_ref->node.ref_mod -= ref->ref_mod; | |
2757 | break; | |
2758 | case BTRFS_DROP_DELAYED_REF: | |
2759 | locked_ref->node.ref_mod += ref->ref_mod; | |
2760 | break; | |
2761 | default: | |
2762 | WARN_ON(1); | |
2763 | } | |
2764 | } | |
d7df2c79 | 2765 | spin_unlock(&locked_ref->lock); |
925baedd | 2766 | |
2ff7e61e | 2767 | ret = run_one_delayed_ref(trans, fs_info, ref, extent_op, |
56bec294 | 2768 | must_insert_reserved); |
eb099670 | 2769 | |
78a6184a | 2770 | btrfs_free_delayed_extent_op(extent_op); |
79787eaa | 2771 | if (ret) { |
9d1032cc | 2772 | spin_lock(&delayed_refs->lock); |
d7df2c79 | 2773 | locked_ref->processing = 0; |
9d1032cc WX |
2774 | delayed_refs->num_heads_ready++; |
2775 | spin_unlock(&delayed_refs->lock); | |
093486c4 MX |
2776 | btrfs_delayed_ref_unlock(locked_ref); |
2777 | btrfs_put_delayed_ref(ref); | |
5d163e0e JM |
2778 | btrfs_debug(fs_info, "run_one_delayed_ref returned %d", |
2779 | ret); | |
79787eaa JM |
2780 | return ret; |
2781 | } | |
2782 | ||
093486c4 MX |
2783 | /* |
2784 | * If this node is a head, that means all the refs in this head | |
2785 | * have been dealt with, and we will pick the next head to deal | |
2786 | * with, so we must unlock the head and drop it from the cluster | |
2787 | * list before we release it. | |
2788 | */ | |
2789 | if (btrfs_delayed_ref_is_head(ref)) { | |
1262133b JB |
2790 | if (locked_ref->is_data && |
2791 | locked_ref->total_ref_mod < 0) { | |
2792 | spin_lock(&delayed_refs->lock); | |
2793 | delayed_refs->pending_csums -= ref->num_bytes; | |
2794 | spin_unlock(&delayed_refs->lock); | |
2795 | } | |
093486c4 MX |
2796 | btrfs_delayed_ref_unlock(locked_ref); |
2797 | locked_ref = NULL; | |
2798 | } | |
2799 | btrfs_put_delayed_ref(ref); | |
2800 | count++; | |
c3e69d58 | 2801 | cond_resched(); |
c3e69d58 | 2802 | } |
0a2b2a84 JB |
2803 | |
2804 | /* | |
2805 | * We don't want to include ref heads since we can have empty ref heads | |
2806 | * and those will drastically skew our runtime down since we just do | |
2807 | * accounting, no actual extent tree updates. | |
2808 | */ | |
2809 | if (actual_count > 0) { | |
2810 | u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start)); | |
2811 | u64 avg; | |
2812 | ||
2813 | /* | |
2814 | * We weigh the current average higher than our current runtime | |
2815 | * to avoid large swings in the average. | |
2816 | */ | |
2817 | spin_lock(&delayed_refs->lock); | |
2818 | avg = fs_info->avg_delayed_ref_runtime * 3 + runtime; | |
f8c269d7 | 2819 | fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */ |
0a2b2a84 JB |
2820 | spin_unlock(&delayed_refs->lock); |
2821 | } | |
d7df2c79 | 2822 | return 0; |
c3e69d58 CM |
2823 | } |
2824 | ||
709c0486 AJ |
2825 | #ifdef SCRAMBLE_DELAYED_REFS |
2826 | /* | |
2827 | * Normally delayed refs get processed in ascending bytenr order. This | |
2828 | * correlates in most cases to the order added. To expose dependencies on this | |
2829 | * order, we start to process the tree in the middle instead of the beginning | |
2830 | */ | |
2831 | static u64 find_middle(struct rb_root *root) | |
2832 | { | |
2833 | struct rb_node *n = root->rb_node; | |
2834 | struct btrfs_delayed_ref_node *entry; | |
2835 | int alt = 1; | |
2836 | u64 middle; | |
2837 | u64 first = 0, last = 0; | |
2838 | ||
2839 | n = rb_first(root); | |
2840 | if (n) { | |
2841 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2842 | first = entry->bytenr; | |
2843 | } | |
2844 | n = rb_last(root); | |
2845 | if (n) { | |
2846 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2847 | last = entry->bytenr; | |
2848 | } | |
2849 | n = root->rb_node; | |
2850 | ||
2851 | while (n) { | |
2852 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2853 | WARN_ON(!entry->in_tree); | |
2854 | ||
2855 | middle = entry->bytenr; | |
2856 | ||
2857 | if (alt) | |
2858 | n = n->rb_left; | |
2859 | else | |
2860 | n = n->rb_right; | |
2861 | ||
2862 | alt = 1 - alt; | |
2863 | } | |
2864 | return middle; | |
2865 | } | |
2866 | #endif | |
2867 | ||
2ff7e61e | 2868 | static inline u64 heads_to_leaves(struct btrfs_fs_info *fs_info, u64 heads) |
1be41b78 JB |
2869 | { |
2870 | u64 num_bytes; | |
2871 | ||
2872 | num_bytes = heads * (sizeof(struct btrfs_extent_item) + | |
2873 | sizeof(struct btrfs_extent_inline_ref)); | |
0b246afa | 2874 | if (!btrfs_fs_incompat(fs_info, SKINNY_METADATA)) |
1be41b78 JB |
2875 | num_bytes += heads * sizeof(struct btrfs_tree_block_info); |
2876 | ||
2877 | /* | |
2878 | * We don't ever fill up leaves all the way so multiply by 2 just to be | |
01327610 | 2879 | * closer to what we're really going to want to use. |
1be41b78 | 2880 | */ |
0b246afa | 2881 | return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(fs_info)); |
1be41b78 JB |
2882 | } |
2883 | ||
1262133b JB |
2884 | /* |
2885 | * Takes the number of bytes to be csumm'ed and figures out how many leaves it | |
2886 | * would require to store the csums for that many bytes. | |
2887 | */ | |
2ff7e61e | 2888 | u64 btrfs_csum_bytes_to_leaves(struct btrfs_fs_info *fs_info, u64 csum_bytes) |
1262133b JB |
2889 | { |
2890 | u64 csum_size; | |
2891 | u64 num_csums_per_leaf; | |
2892 | u64 num_csums; | |
2893 | ||
0b246afa | 2894 | csum_size = BTRFS_MAX_ITEM_SIZE(fs_info); |
1262133b | 2895 | num_csums_per_leaf = div64_u64(csum_size, |
0b246afa JM |
2896 | (u64)btrfs_super_csum_size(fs_info->super_copy)); |
2897 | num_csums = div64_u64(csum_bytes, fs_info->sectorsize); | |
1262133b JB |
2898 | num_csums += num_csums_per_leaf - 1; |
2899 | num_csums = div64_u64(num_csums, num_csums_per_leaf); | |
2900 | return num_csums; | |
2901 | } | |
2902 | ||
0a2b2a84 | 2903 | int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans, |
2ff7e61e | 2904 | struct btrfs_fs_info *fs_info) |
1be41b78 JB |
2905 | { |
2906 | struct btrfs_block_rsv *global_rsv; | |
2907 | u64 num_heads = trans->transaction->delayed_refs.num_heads_ready; | |
1262133b | 2908 | u64 csum_bytes = trans->transaction->delayed_refs.pending_csums; |
cb723e49 JB |
2909 | u64 num_dirty_bgs = trans->transaction->num_dirty_bgs; |
2910 | u64 num_bytes, num_dirty_bgs_bytes; | |
1be41b78 JB |
2911 | int ret = 0; |
2912 | ||
0b246afa | 2913 | num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1); |
2ff7e61e | 2914 | num_heads = heads_to_leaves(fs_info, num_heads); |
1be41b78 | 2915 | if (num_heads > 1) |
0b246afa | 2916 | num_bytes += (num_heads - 1) * fs_info->nodesize; |
1be41b78 | 2917 | num_bytes <<= 1; |
2ff7e61e JM |
2918 | num_bytes += btrfs_csum_bytes_to_leaves(fs_info, csum_bytes) * |
2919 | fs_info->nodesize; | |
0b246afa | 2920 | num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(fs_info, |
cb723e49 | 2921 | num_dirty_bgs); |
0b246afa | 2922 | global_rsv = &fs_info->global_block_rsv; |
1be41b78 JB |
2923 | |
2924 | /* | |
2925 | * If we can't allocate any more chunks lets make sure we have _lots_ of | |
2926 | * wiggle room since running delayed refs can create more delayed refs. | |
2927 | */ | |
cb723e49 JB |
2928 | if (global_rsv->space_info->full) { |
2929 | num_dirty_bgs_bytes <<= 1; | |
1be41b78 | 2930 | num_bytes <<= 1; |
cb723e49 | 2931 | } |
1be41b78 JB |
2932 | |
2933 | spin_lock(&global_rsv->lock); | |
cb723e49 | 2934 | if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes) |
1be41b78 JB |
2935 | ret = 1; |
2936 | spin_unlock(&global_rsv->lock); | |
2937 | return ret; | |
2938 | } | |
2939 | ||
0a2b2a84 | 2940 | int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans, |
2ff7e61e | 2941 | struct btrfs_fs_info *fs_info) |
0a2b2a84 | 2942 | { |
0a2b2a84 JB |
2943 | u64 num_entries = |
2944 | atomic_read(&trans->transaction->delayed_refs.num_entries); | |
2945 | u64 avg_runtime; | |
a79b7d4b | 2946 | u64 val; |
0a2b2a84 JB |
2947 | |
2948 | smp_mb(); | |
2949 | avg_runtime = fs_info->avg_delayed_ref_runtime; | |
a79b7d4b | 2950 | val = num_entries * avg_runtime; |
dc1a90c6 | 2951 | if (val >= NSEC_PER_SEC) |
0a2b2a84 | 2952 | return 1; |
a79b7d4b CM |
2953 | if (val >= NSEC_PER_SEC / 2) |
2954 | return 2; | |
0a2b2a84 | 2955 | |
2ff7e61e | 2956 | return btrfs_check_space_for_delayed_refs(trans, fs_info); |
0a2b2a84 JB |
2957 | } |
2958 | ||
a79b7d4b CM |
2959 | struct async_delayed_refs { |
2960 | struct btrfs_root *root; | |
31b9655f | 2961 | u64 transid; |
a79b7d4b CM |
2962 | int count; |
2963 | int error; | |
2964 | int sync; | |
2965 | struct completion wait; | |
2966 | struct btrfs_work work; | |
2967 | }; | |
2968 | ||
2ff7e61e JM |
2969 | static inline struct async_delayed_refs * |
2970 | to_async_delayed_refs(struct btrfs_work *work) | |
2971 | { | |
2972 | return container_of(work, struct async_delayed_refs, work); | |
2973 | } | |
2974 | ||
a79b7d4b CM |
2975 | static void delayed_ref_async_start(struct btrfs_work *work) |
2976 | { | |
2ff7e61e | 2977 | struct async_delayed_refs *async = to_async_delayed_refs(work); |
a79b7d4b | 2978 | struct btrfs_trans_handle *trans; |
2ff7e61e | 2979 | struct btrfs_fs_info *fs_info = async->root->fs_info; |
a79b7d4b CM |
2980 | int ret; |
2981 | ||
0f873eca | 2982 | /* if the commit is already started, we don't need to wait here */ |
2ff7e61e | 2983 | if (btrfs_transaction_blocked(fs_info)) |
31b9655f | 2984 | goto done; |
31b9655f | 2985 | |
0f873eca CM |
2986 | trans = btrfs_join_transaction(async->root); |
2987 | if (IS_ERR(trans)) { | |
2988 | async->error = PTR_ERR(trans); | |
a79b7d4b CM |
2989 | goto done; |
2990 | } | |
2991 | ||
2992 | /* | |
01327610 | 2993 | * trans->sync means that when we call end_transaction, we won't |
a79b7d4b CM |
2994 | * wait on delayed refs |
2995 | */ | |
2996 | trans->sync = true; | |
0f873eca CM |
2997 | |
2998 | /* Don't bother flushing if we got into a different transaction */ | |
2999 | if (trans->transid > async->transid) | |
3000 | goto end; | |
3001 | ||
2ff7e61e | 3002 | ret = btrfs_run_delayed_refs(trans, fs_info, async->count); |
a79b7d4b CM |
3003 | if (ret) |
3004 | async->error = ret; | |
0f873eca | 3005 | end: |
3a45bb20 | 3006 | ret = btrfs_end_transaction(trans); |
a79b7d4b CM |
3007 | if (ret && !async->error) |
3008 | async->error = ret; | |
3009 | done: | |
3010 | if (async->sync) | |
3011 | complete(&async->wait); | |
3012 | else | |
3013 | kfree(async); | |
3014 | } | |
3015 | ||
2ff7e61e | 3016 | int btrfs_async_run_delayed_refs(struct btrfs_fs_info *fs_info, |
31b9655f | 3017 | unsigned long count, u64 transid, int wait) |
a79b7d4b CM |
3018 | { |
3019 | struct async_delayed_refs *async; | |
3020 | int ret; | |
3021 | ||
3022 | async = kmalloc(sizeof(*async), GFP_NOFS); | |
3023 | if (!async) | |
3024 | return -ENOMEM; | |
3025 | ||
0b246afa | 3026 | async->root = fs_info->tree_root; |
a79b7d4b CM |
3027 | async->count = count; |
3028 | async->error = 0; | |
31b9655f | 3029 | async->transid = transid; |
a79b7d4b CM |
3030 | if (wait) |
3031 | async->sync = 1; | |
3032 | else | |
3033 | async->sync = 0; | |
3034 | init_completion(&async->wait); | |
3035 | ||
9e0af237 LB |
3036 | btrfs_init_work(&async->work, btrfs_extent_refs_helper, |
3037 | delayed_ref_async_start, NULL, NULL); | |
a79b7d4b | 3038 | |
0b246afa | 3039 | btrfs_queue_work(fs_info->extent_workers, &async->work); |
a79b7d4b CM |
3040 | |
3041 | if (wait) { | |
3042 | wait_for_completion(&async->wait); | |
3043 | ret = async->error; | |
3044 | kfree(async); | |
3045 | return ret; | |
3046 | } | |
3047 | return 0; | |
3048 | } | |
3049 | ||
c3e69d58 CM |
3050 | /* |
3051 | * this starts processing the delayed reference count updates and | |
3052 | * extent insertions we have queued up so far. count can be | |
3053 | * 0, which means to process everything in the tree at the start | |
3054 | * of the run (but not newly added entries), or it can be some target | |
3055 | * number you'd like to process. | |
79787eaa JM |
3056 | * |
3057 | * Returns 0 on success or if called with an aborted transaction | |
3058 | * Returns <0 on error and aborts the transaction | |
c3e69d58 CM |
3059 | */ |
3060 | int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, | |
2ff7e61e | 3061 | struct btrfs_fs_info *fs_info, unsigned long count) |
c3e69d58 CM |
3062 | { |
3063 | struct rb_node *node; | |
3064 | struct btrfs_delayed_ref_root *delayed_refs; | |
c46effa6 | 3065 | struct btrfs_delayed_ref_head *head; |
c3e69d58 CM |
3066 | int ret; |
3067 | int run_all = count == (unsigned long)-1; | |
d9a0540a | 3068 | bool can_flush_pending_bgs = trans->can_flush_pending_bgs; |
c3e69d58 | 3069 | |
79787eaa JM |
3070 | /* We'll clean this up in btrfs_cleanup_transaction */ |
3071 | if (trans->aborted) | |
3072 | return 0; | |
3073 | ||
0b246afa | 3074 | if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags)) |
511711af CM |
3075 | return 0; |
3076 | ||
c3e69d58 | 3077 | delayed_refs = &trans->transaction->delayed_refs; |
26455d33 | 3078 | if (count == 0) |
d7df2c79 | 3079 | count = atomic_read(&delayed_refs->num_entries) * 2; |
bb721703 | 3080 | |
c3e69d58 | 3081 | again: |
709c0486 AJ |
3082 | #ifdef SCRAMBLE_DELAYED_REFS |
3083 | delayed_refs->run_delayed_start = find_middle(&delayed_refs->root); | |
3084 | #endif | |
d9a0540a | 3085 | trans->can_flush_pending_bgs = false; |
2ff7e61e | 3086 | ret = __btrfs_run_delayed_refs(trans, fs_info, count); |
d7df2c79 | 3087 | if (ret < 0) { |
66642832 | 3088 | btrfs_abort_transaction(trans, ret); |
d7df2c79 | 3089 | return ret; |
eb099670 | 3090 | } |
c3e69d58 | 3091 | |
56bec294 | 3092 | if (run_all) { |
d7df2c79 | 3093 | if (!list_empty(&trans->new_bgs)) |
2ff7e61e | 3094 | btrfs_create_pending_block_groups(trans, fs_info); |
ea658bad | 3095 | |
d7df2c79 | 3096 | spin_lock(&delayed_refs->lock); |
c46effa6 | 3097 | node = rb_first(&delayed_refs->href_root); |
d7df2c79 JB |
3098 | if (!node) { |
3099 | spin_unlock(&delayed_refs->lock); | |
56bec294 | 3100 | goto out; |
d7df2c79 | 3101 | } |
e9d0b13b | 3102 | |
56bec294 | 3103 | while (node) { |
c46effa6 LB |
3104 | head = rb_entry(node, struct btrfs_delayed_ref_head, |
3105 | href_node); | |
3106 | if (btrfs_delayed_ref_is_head(&head->node)) { | |
3107 | struct btrfs_delayed_ref_node *ref; | |
5caf2a00 | 3108 | |
c46effa6 | 3109 | ref = &head->node; |
6df8cdf5 | 3110 | refcount_inc(&ref->refs); |
56bec294 CM |
3111 | |
3112 | spin_unlock(&delayed_refs->lock); | |
8cc33e5c DS |
3113 | /* |
3114 | * Mutex was contended, block until it's | |
3115 | * released and try again | |
3116 | */ | |
56bec294 CM |
3117 | mutex_lock(&head->mutex); |
3118 | mutex_unlock(&head->mutex); | |
3119 | ||
3120 | btrfs_put_delayed_ref(ref); | |
1887be66 | 3121 | cond_resched(); |
56bec294 | 3122 | goto again; |
c46effa6 LB |
3123 | } else { |
3124 | WARN_ON(1); | |
56bec294 CM |
3125 | } |
3126 | node = rb_next(node); | |
3127 | } | |
3128 | spin_unlock(&delayed_refs->lock); | |
d7df2c79 | 3129 | cond_resched(); |
56bec294 | 3130 | goto again; |
5f39d397 | 3131 | } |
54aa1f4d | 3132 | out: |
d9a0540a | 3133 | trans->can_flush_pending_bgs = can_flush_pending_bgs; |
a28ec197 CM |
3134 | return 0; |
3135 | } | |
3136 | ||
5d4f98a2 | 3137 | int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans, |
2ff7e61e | 3138 | struct btrfs_fs_info *fs_info, |
5d4f98a2 | 3139 | u64 bytenr, u64 num_bytes, u64 flags, |
b1c79e09 | 3140 | int level, int is_data) |
5d4f98a2 YZ |
3141 | { |
3142 | struct btrfs_delayed_extent_op *extent_op; | |
3143 | int ret; | |
3144 | ||
78a6184a | 3145 | extent_op = btrfs_alloc_delayed_extent_op(); |
5d4f98a2 YZ |
3146 | if (!extent_op) |
3147 | return -ENOMEM; | |
3148 | ||
3149 | extent_op->flags_to_set = flags; | |
35b3ad50 DS |
3150 | extent_op->update_flags = true; |
3151 | extent_op->update_key = false; | |
3152 | extent_op->is_data = is_data ? true : false; | |
b1c79e09 | 3153 | extent_op->level = level; |
5d4f98a2 | 3154 | |
0b246afa | 3155 | ret = btrfs_add_delayed_extent_op(fs_info, trans, bytenr, |
66d7e7f0 | 3156 | num_bytes, extent_op); |
5d4f98a2 | 3157 | if (ret) |
78a6184a | 3158 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 YZ |
3159 | return ret; |
3160 | } | |
3161 | ||
e4c3b2dc | 3162 | static noinline int check_delayed_ref(struct btrfs_root *root, |
5d4f98a2 YZ |
3163 | struct btrfs_path *path, |
3164 | u64 objectid, u64 offset, u64 bytenr) | |
3165 | { | |
3166 | struct btrfs_delayed_ref_head *head; | |
3167 | struct btrfs_delayed_ref_node *ref; | |
3168 | struct btrfs_delayed_data_ref *data_ref; | |
3169 | struct btrfs_delayed_ref_root *delayed_refs; | |
e4c3b2dc | 3170 | struct btrfs_transaction *cur_trans; |
5d4f98a2 YZ |
3171 | int ret = 0; |
3172 | ||
b0e5b437 | 3173 | spin_lock(&root->fs_info->trans_lock); |
e4c3b2dc | 3174 | cur_trans = root->fs_info->running_transaction; |
b0e5b437 | 3175 | if (cur_trans) |
3176 | refcount_inc(&cur_trans->use_count); | |
3177 | spin_unlock(&root->fs_info->trans_lock); | |
e4c3b2dc LB |
3178 | if (!cur_trans) |
3179 | return 0; | |
3180 | ||
3181 | delayed_refs = &cur_trans->delayed_refs; | |
5d4f98a2 | 3182 | spin_lock(&delayed_refs->lock); |
f72ad18e | 3183 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
d7df2c79 JB |
3184 | if (!head) { |
3185 | spin_unlock(&delayed_refs->lock); | |
b0e5b437 | 3186 | btrfs_put_transaction(cur_trans); |
d7df2c79 JB |
3187 | return 0; |
3188 | } | |
5d4f98a2 YZ |
3189 | |
3190 | if (!mutex_trylock(&head->mutex)) { | |
6df8cdf5 | 3191 | refcount_inc(&head->node.refs); |
5d4f98a2 YZ |
3192 | spin_unlock(&delayed_refs->lock); |
3193 | ||
b3b4aa74 | 3194 | btrfs_release_path(path); |
5d4f98a2 | 3195 | |
8cc33e5c DS |
3196 | /* |
3197 | * Mutex was contended, block until it's released and let | |
3198 | * caller try again | |
3199 | */ | |
5d4f98a2 YZ |
3200 | mutex_lock(&head->mutex); |
3201 | mutex_unlock(&head->mutex); | |
3202 | btrfs_put_delayed_ref(&head->node); | |
b0e5b437 | 3203 | btrfs_put_transaction(cur_trans); |
5d4f98a2 YZ |
3204 | return -EAGAIN; |
3205 | } | |
d7df2c79 | 3206 | spin_unlock(&delayed_refs->lock); |
5d4f98a2 | 3207 | |
d7df2c79 | 3208 | spin_lock(&head->lock); |
c6fc2454 | 3209 | list_for_each_entry(ref, &head->ref_list, list) { |
d7df2c79 JB |
3210 | /* If it's a shared ref we know a cross reference exists */ |
3211 | if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) { | |
3212 | ret = 1; | |
3213 | break; | |
3214 | } | |
5d4f98a2 | 3215 | |
d7df2c79 | 3216 | data_ref = btrfs_delayed_node_to_data_ref(ref); |
5d4f98a2 | 3217 | |
d7df2c79 JB |
3218 | /* |
3219 | * If our ref doesn't match the one we're currently looking at | |
3220 | * then we have a cross reference. | |
3221 | */ | |
3222 | if (data_ref->root != root->root_key.objectid || | |
3223 | data_ref->objectid != objectid || | |
3224 | data_ref->offset != offset) { | |
3225 | ret = 1; | |
3226 | break; | |
3227 | } | |
5d4f98a2 | 3228 | } |
d7df2c79 | 3229 | spin_unlock(&head->lock); |
5d4f98a2 | 3230 | mutex_unlock(&head->mutex); |
b0e5b437 | 3231 | btrfs_put_transaction(cur_trans); |
5d4f98a2 YZ |
3232 | return ret; |
3233 | } | |
3234 | ||
e4c3b2dc | 3235 | static noinline int check_committed_ref(struct btrfs_root *root, |
5d4f98a2 YZ |
3236 | struct btrfs_path *path, |
3237 | u64 objectid, u64 offset, u64 bytenr) | |
be20aa9d | 3238 | { |
0b246afa JM |
3239 | struct btrfs_fs_info *fs_info = root->fs_info; |
3240 | struct btrfs_root *extent_root = fs_info->extent_root; | |
f321e491 | 3241 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
3242 | struct btrfs_extent_data_ref *ref; |
3243 | struct btrfs_extent_inline_ref *iref; | |
3244 | struct btrfs_extent_item *ei; | |
f321e491 | 3245 | struct btrfs_key key; |
5d4f98a2 | 3246 | u32 item_size; |
3de28d57 | 3247 | int type; |
be20aa9d | 3248 | int ret; |
925baedd | 3249 | |
be20aa9d | 3250 | key.objectid = bytenr; |
31840ae1 | 3251 | key.offset = (u64)-1; |
f321e491 | 3252 | key.type = BTRFS_EXTENT_ITEM_KEY; |
be20aa9d | 3253 | |
be20aa9d CM |
3254 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
3255 | if (ret < 0) | |
3256 | goto out; | |
79787eaa | 3257 | BUG_ON(ret == 0); /* Corruption */ |
80ff3856 YZ |
3258 | |
3259 | ret = -ENOENT; | |
3260 | if (path->slots[0] == 0) | |
31840ae1 | 3261 | goto out; |
be20aa9d | 3262 | |
31840ae1 | 3263 | path->slots[0]--; |
f321e491 | 3264 | leaf = path->nodes[0]; |
5d4f98a2 | 3265 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
be20aa9d | 3266 | |
5d4f98a2 | 3267 | if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY) |
be20aa9d | 3268 | goto out; |
f321e491 | 3269 | |
5d4f98a2 YZ |
3270 | ret = 1; |
3271 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
3272 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
3273 | if (item_size < sizeof(*ei)) { | |
3274 | WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0)); | |
3275 | goto out; | |
3276 | } | |
3277 | #endif | |
3278 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
bd09835d | 3279 | |
5d4f98a2 YZ |
3280 | if (item_size != sizeof(*ei) + |
3281 | btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY)) | |
3282 | goto out; | |
be20aa9d | 3283 | |
5d4f98a2 YZ |
3284 | if (btrfs_extent_generation(leaf, ei) <= |
3285 | btrfs_root_last_snapshot(&root->root_item)) | |
3286 | goto out; | |
3287 | ||
3288 | iref = (struct btrfs_extent_inline_ref *)(ei + 1); | |
3de28d57 LB |
3289 | |
3290 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA); | |
3291 | if (type != BTRFS_EXTENT_DATA_REF_KEY) | |
5d4f98a2 YZ |
3292 | goto out; |
3293 | ||
3294 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
3295 | if (btrfs_extent_refs(leaf, ei) != | |
3296 | btrfs_extent_data_ref_count(leaf, ref) || | |
3297 | btrfs_extent_data_ref_root(leaf, ref) != | |
3298 | root->root_key.objectid || | |
3299 | btrfs_extent_data_ref_objectid(leaf, ref) != objectid || | |
3300 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
3301 | goto out; | |
3302 | ||
3303 | ret = 0; | |
3304 | out: | |
3305 | return ret; | |
3306 | } | |
3307 | ||
e4c3b2dc LB |
3308 | int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset, |
3309 | u64 bytenr) | |
5d4f98a2 YZ |
3310 | { |
3311 | struct btrfs_path *path; | |
3312 | int ret; | |
3313 | int ret2; | |
3314 | ||
3315 | path = btrfs_alloc_path(); | |
3316 | if (!path) | |
3317 | return -ENOENT; | |
3318 | ||
3319 | do { | |
e4c3b2dc | 3320 | ret = check_committed_ref(root, path, objectid, |
5d4f98a2 YZ |
3321 | offset, bytenr); |
3322 | if (ret && ret != -ENOENT) | |
f321e491 | 3323 | goto out; |
80ff3856 | 3324 | |
e4c3b2dc | 3325 | ret2 = check_delayed_ref(root, path, objectid, |
5d4f98a2 YZ |
3326 | offset, bytenr); |
3327 | } while (ret2 == -EAGAIN); | |
3328 | ||
3329 | if (ret2 && ret2 != -ENOENT) { | |
3330 | ret = ret2; | |
3331 | goto out; | |
f321e491 | 3332 | } |
5d4f98a2 YZ |
3333 | |
3334 | if (ret != -ENOENT || ret2 != -ENOENT) | |
3335 | ret = 0; | |
be20aa9d | 3336 | out: |
80ff3856 | 3337 | btrfs_free_path(path); |
f0486c68 YZ |
3338 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) |
3339 | WARN_ON(ret > 0); | |
f321e491 | 3340 | return ret; |
be20aa9d | 3341 | } |
c5739bba | 3342 | |
5d4f98a2 | 3343 | static int __btrfs_mod_ref(struct btrfs_trans_handle *trans, |
b7a9f29f | 3344 | struct btrfs_root *root, |
5d4f98a2 | 3345 | struct extent_buffer *buf, |
e339a6b0 | 3346 | int full_backref, int inc) |
31840ae1 | 3347 | { |
0b246afa | 3348 | struct btrfs_fs_info *fs_info = root->fs_info; |
31840ae1 | 3349 | u64 bytenr; |
5d4f98a2 YZ |
3350 | u64 num_bytes; |
3351 | u64 parent; | |
31840ae1 | 3352 | u64 ref_root; |
31840ae1 | 3353 | u32 nritems; |
31840ae1 ZY |
3354 | struct btrfs_key key; |
3355 | struct btrfs_file_extent_item *fi; | |
3356 | int i; | |
3357 | int level; | |
3358 | int ret = 0; | |
2ff7e61e JM |
3359 | int (*process_func)(struct btrfs_trans_handle *, |
3360 | struct btrfs_fs_info *, | |
b06c4bf5 | 3361 | u64, u64, u64, u64, u64, u64); |
31840ae1 | 3362 | |
fccb84c9 | 3363 | |
0b246afa | 3364 | if (btrfs_is_testing(fs_info)) |
faa2dbf0 | 3365 | return 0; |
fccb84c9 | 3366 | |
31840ae1 | 3367 | ref_root = btrfs_header_owner(buf); |
31840ae1 ZY |
3368 | nritems = btrfs_header_nritems(buf); |
3369 | level = btrfs_header_level(buf); | |
3370 | ||
27cdeb70 | 3371 | if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0) |
5d4f98a2 | 3372 | return 0; |
31840ae1 | 3373 | |
5d4f98a2 YZ |
3374 | if (inc) |
3375 | process_func = btrfs_inc_extent_ref; | |
3376 | else | |
3377 | process_func = btrfs_free_extent; | |
31840ae1 | 3378 | |
5d4f98a2 YZ |
3379 | if (full_backref) |
3380 | parent = buf->start; | |
3381 | else | |
3382 | parent = 0; | |
3383 | ||
3384 | for (i = 0; i < nritems; i++) { | |
31840ae1 | 3385 | if (level == 0) { |
5d4f98a2 | 3386 | btrfs_item_key_to_cpu(buf, &key, i); |
962a298f | 3387 | if (key.type != BTRFS_EXTENT_DATA_KEY) |
31840ae1 | 3388 | continue; |
5d4f98a2 | 3389 | fi = btrfs_item_ptr(buf, i, |
31840ae1 ZY |
3390 | struct btrfs_file_extent_item); |
3391 | if (btrfs_file_extent_type(buf, fi) == | |
3392 | BTRFS_FILE_EXTENT_INLINE) | |
3393 | continue; | |
3394 | bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
3395 | if (bytenr == 0) | |
3396 | continue; | |
5d4f98a2 YZ |
3397 | |
3398 | num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi); | |
3399 | key.offset -= btrfs_file_extent_offset(buf, fi); | |
2ff7e61e | 3400 | ret = process_func(trans, fs_info, bytenr, num_bytes, |
5d4f98a2 | 3401 | parent, ref_root, key.objectid, |
b06c4bf5 | 3402 | key.offset); |
31840ae1 ZY |
3403 | if (ret) |
3404 | goto fail; | |
3405 | } else { | |
5d4f98a2 | 3406 | bytenr = btrfs_node_blockptr(buf, i); |
0b246afa | 3407 | num_bytes = fs_info->nodesize; |
2ff7e61e | 3408 | ret = process_func(trans, fs_info, bytenr, num_bytes, |
b06c4bf5 | 3409 | parent, ref_root, level - 1, 0); |
31840ae1 ZY |
3410 | if (ret) |
3411 | goto fail; | |
3412 | } | |
3413 | } | |
3414 | return 0; | |
3415 | fail: | |
5d4f98a2 YZ |
3416 | return ret; |
3417 | } | |
3418 | ||
3419 | int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
e339a6b0 | 3420 | struct extent_buffer *buf, int full_backref) |
5d4f98a2 | 3421 | { |
e339a6b0 | 3422 | return __btrfs_mod_ref(trans, root, buf, full_backref, 1); |
5d4f98a2 YZ |
3423 | } |
3424 | ||
3425 | int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
e339a6b0 | 3426 | struct extent_buffer *buf, int full_backref) |
5d4f98a2 | 3427 | { |
e339a6b0 | 3428 | return __btrfs_mod_ref(trans, root, buf, full_backref, 0); |
31840ae1 ZY |
3429 | } |
3430 | ||
9078a3e1 | 3431 | static int write_one_cache_group(struct btrfs_trans_handle *trans, |
2ff7e61e | 3432 | struct btrfs_fs_info *fs_info, |
9078a3e1 CM |
3433 | struct btrfs_path *path, |
3434 | struct btrfs_block_group_cache *cache) | |
3435 | { | |
3436 | int ret; | |
0b246afa | 3437 | struct btrfs_root *extent_root = fs_info->extent_root; |
5f39d397 CM |
3438 | unsigned long bi; |
3439 | struct extent_buffer *leaf; | |
9078a3e1 | 3440 | |
9078a3e1 | 3441 | ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1); |
df95e7f0 JB |
3442 | if (ret) { |
3443 | if (ret > 0) | |
3444 | ret = -ENOENT; | |
54aa1f4d | 3445 | goto fail; |
df95e7f0 | 3446 | } |
5f39d397 CM |
3447 | |
3448 | leaf = path->nodes[0]; | |
3449 | bi = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
3450 | write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item)); | |
3451 | btrfs_mark_buffer_dirty(leaf); | |
54aa1f4d | 3452 | fail: |
24b89d08 | 3453 | btrfs_release_path(path); |
df95e7f0 | 3454 | return ret; |
9078a3e1 CM |
3455 | |
3456 | } | |
3457 | ||
4a8c9a62 | 3458 | static struct btrfs_block_group_cache * |
2ff7e61e | 3459 | next_block_group(struct btrfs_fs_info *fs_info, |
4a8c9a62 YZ |
3460 | struct btrfs_block_group_cache *cache) |
3461 | { | |
3462 | struct rb_node *node; | |
292cbd51 | 3463 | |
0b246afa | 3464 | spin_lock(&fs_info->block_group_cache_lock); |
292cbd51 FM |
3465 | |
3466 | /* If our block group was removed, we need a full search. */ | |
3467 | if (RB_EMPTY_NODE(&cache->cache_node)) { | |
3468 | const u64 next_bytenr = cache->key.objectid + cache->key.offset; | |
3469 | ||
0b246afa | 3470 | spin_unlock(&fs_info->block_group_cache_lock); |
292cbd51 | 3471 | btrfs_put_block_group(cache); |
0b246afa | 3472 | cache = btrfs_lookup_first_block_group(fs_info, next_bytenr); return cache; |
292cbd51 | 3473 | } |
4a8c9a62 YZ |
3474 | node = rb_next(&cache->cache_node); |
3475 | btrfs_put_block_group(cache); | |
3476 | if (node) { | |
3477 | cache = rb_entry(node, struct btrfs_block_group_cache, | |
3478 | cache_node); | |
11dfe35a | 3479 | btrfs_get_block_group(cache); |
4a8c9a62 YZ |
3480 | } else |
3481 | cache = NULL; | |
0b246afa | 3482 | spin_unlock(&fs_info->block_group_cache_lock); |
4a8c9a62 YZ |
3483 | return cache; |
3484 | } | |
3485 | ||
0af3d00b JB |
3486 | static int cache_save_setup(struct btrfs_block_group_cache *block_group, |
3487 | struct btrfs_trans_handle *trans, | |
3488 | struct btrfs_path *path) | |
3489 | { | |
0b246afa JM |
3490 | struct btrfs_fs_info *fs_info = block_group->fs_info; |
3491 | struct btrfs_root *root = fs_info->tree_root; | |
0af3d00b | 3492 | struct inode *inode = NULL; |
364ecf36 | 3493 | struct extent_changeset *data_reserved = NULL; |
0af3d00b | 3494 | u64 alloc_hint = 0; |
2b20982e | 3495 | int dcs = BTRFS_DC_ERROR; |
f8c269d7 | 3496 | u64 num_pages = 0; |
0af3d00b JB |
3497 | int retries = 0; |
3498 | int ret = 0; | |
3499 | ||
3500 | /* | |
3501 | * If this block group is smaller than 100 megs don't bother caching the | |
3502 | * block group. | |
3503 | */ | |
ee22184b | 3504 | if (block_group->key.offset < (100 * SZ_1M)) { |
0af3d00b JB |
3505 | spin_lock(&block_group->lock); |
3506 | block_group->disk_cache_state = BTRFS_DC_WRITTEN; | |
3507 | spin_unlock(&block_group->lock); | |
3508 | return 0; | |
3509 | } | |
3510 | ||
0c0ef4bc JB |
3511 | if (trans->aborted) |
3512 | return 0; | |
0af3d00b | 3513 | again: |
77ab86bf | 3514 | inode = lookup_free_space_inode(fs_info, block_group, path); |
0af3d00b JB |
3515 | if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) { |
3516 | ret = PTR_ERR(inode); | |
b3b4aa74 | 3517 | btrfs_release_path(path); |
0af3d00b JB |
3518 | goto out; |
3519 | } | |
3520 | ||
3521 | if (IS_ERR(inode)) { | |
3522 | BUG_ON(retries); | |
3523 | retries++; | |
3524 | ||
3525 | if (block_group->ro) | |
3526 | goto out_free; | |
3527 | ||
77ab86bf JM |
3528 | ret = create_free_space_inode(fs_info, trans, block_group, |
3529 | path); | |
0af3d00b JB |
3530 | if (ret) |
3531 | goto out_free; | |
3532 | goto again; | |
3533 | } | |
3534 | ||
3535 | /* | |
3536 | * We want to set the generation to 0, that way if anything goes wrong | |
3537 | * from here on out we know not to trust this cache when we load up next | |
3538 | * time. | |
3539 | */ | |
3540 | BTRFS_I(inode)->generation = 0; | |
3541 | ret = btrfs_update_inode(trans, root, inode); | |
0c0ef4bc JB |
3542 | if (ret) { |
3543 | /* | |
3544 | * So theoretically we could recover from this, simply set the | |
3545 | * super cache generation to 0 so we know to invalidate the | |
3546 | * cache, but then we'd have to keep track of the block groups | |
3547 | * that fail this way so we know we _have_ to reset this cache | |
3548 | * before the next commit or risk reading stale cache. So to | |
3549 | * limit our exposure to horrible edge cases lets just abort the | |
3550 | * transaction, this only happens in really bad situations | |
3551 | * anyway. | |
3552 | */ | |
66642832 | 3553 | btrfs_abort_transaction(trans, ret); |
0c0ef4bc JB |
3554 | goto out_put; |
3555 | } | |
0af3d00b JB |
3556 | WARN_ON(ret); |
3557 | ||
28580e75 JB |
3558 | /* We've already setup this transaction, go ahead and exit */ |
3559 | if (block_group->cache_generation == trans->transid && | |
3560 | i_size_read(inode)) { | |
3561 | dcs = BTRFS_DC_SETUP; | |
3562 | goto out_put; | |
3563 | } | |
3564 | ||
0af3d00b | 3565 | if (i_size_read(inode) > 0) { |
2ff7e61e | 3566 | ret = btrfs_check_trunc_cache_free_space(fs_info, |
0b246afa | 3567 | &fs_info->global_block_rsv); |
7b61cd92 MX |
3568 | if (ret) |
3569 | goto out_put; | |
3570 | ||
77ab86bf | 3571 | ret = btrfs_truncate_free_space_cache(trans, NULL, inode); |
0af3d00b JB |
3572 | if (ret) |
3573 | goto out_put; | |
3574 | } | |
3575 | ||
3576 | spin_lock(&block_group->lock); | |
cf7c1ef6 | 3577 | if (block_group->cached != BTRFS_CACHE_FINISHED || |
0b246afa | 3578 | !btrfs_test_opt(fs_info, SPACE_CACHE)) { |
cf7c1ef6 LB |
3579 | /* |
3580 | * don't bother trying to write stuff out _if_ | |
3581 | * a) we're not cached, | |
1a79c1f2 LB |
3582 | * b) we're with nospace_cache mount option, |
3583 | * c) we're with v2 space_cache (FREE_SPACE_TREE). | |
cf7c1ef6 | 3584 | */ |
2b20982e | 3585 | dcs = BTRFS_DC_WRITTEN; |
0af3d00b JB |
3586 | spin_unlock(&block_group->lock); |
3587 | goto out_put; | |
3588 | } | |
3589 | spin_unlock(&block_group->lock); | |
3590 | ||
2968b1f4 JB |
3591 | /* |
3592 | * We hit an ENOSPC when setting up the cache in this transaction, just | |
3593 | * skip doing the setup, we've already cleared the cache so we're safe. | |
3594 | */ | |
3595 | if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) { | |
3596 | ret = -ENOSPC; | |
3597 | goto out_put; | |
3598 | } | |
3599 | ||
6fc823b1 JB |
3600 | /* |
3601 | * Try to preallocate enough space based on how big the block group is. | |
3602 | * Keep in mind this has to include any pinned space which could end up | |
3603 | * taking up quite a bit since it's not folded into the other space | |
3604 | * cache. | |
3605 | */ | |
ee22184b | 3606 | num_pages = div_u64(block_group->key.offset, SZ_256M); |
0af3d00b JB |
3607 | if (!num_pages) |
3608 | num_pages = 1; | |
3609 | ||
0af3d00b | 3610 | num_pages *= 16; |
09cbfeaf | 3611 | num_pages *= PAGE_SIZE; |
0af3d00b | 3612 | |
364ecf36 | 3613 | ret = btrfs_check_data_free_space(inode, &data_reserved, 0, num_pages); |
0af3d00b JB |
3614 | if (ret) |
3615 | goto out_put; | |
3616 | ||
3617 | ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages, | |
3618 | num_pages, num_pages, | |
3619 | &alloc_hint); | |
2968b1f4 JB |
3620 | /* |
3621 | * Our cache requires contiguous chunks so that we don't modify a bunch | |
3622 | * of metadata or split extents when writing the cache out, which means | |
3623 | * we can enospc if we are heavily fragmented in addition to just normal | |
3624 | * out of space conditions. So if we hit this just skip setting up any | |
3625 | * other block groups for this transaction, maybe we'll unpin enough | |
3626 | * space the next time around. | |
3627 | */ | |
2b20982e JB |
3628 | if (!ret) |
3629 | dcs = BTRFS_DC_SETUP; | |
2968b1f4 JB |
3630 | else if (ret == -ENOSPC) |
3631 | set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags); | |
c09544e0 | 3632 | |
0af3d00b JB |
3633 | out_put: |
3634 | iput(inode); | |
3635 | out_free: | |
b3b4aa74 | 3636 | btrfs_release_path(path); |
0af3d00b JB |
3637 | out: |
3638 | spin_lock(&block_group->lock); | |
e65cbb94 | 3639 | if (!ret && dcs == BTRFS_DC_SETUP) |
5b0e95bf | 3640 | block_group->cache_generation = trans->transid; |
2b20982e | 3641 | block_group->disk_cache_state = dcs; |
0af3d00b JB |
3642 | spin_unlock(&block_group->lock); |
3643 | ||
364ecf36 | 3644 | extent_changeset_free(data_reserved); |
0af3d00b JB |
3645 | return ret; |
3646 | } | |
3647 | ||
dcdf7f6d | 3648 | int btrfs_setup_space_cache(struct btrfs_trans_handle *trans, |
2ff7e61e | 3649 | struct btrfs_fs_info *fs_info) |
dcdf7f6d JB |
3650 | { |
3651 | struct btrfs_block_group_cache *cache, *tmp; | |
3652 | struct btrfs_transaction *cur_trans = trans->transaction; | |
3653 | struct btrfs_path *path; | |
3654 | ||
3655 | if (list_empty(&cur_trans->dirty_bgs) || | |
0b246afa | 3656 | !btrfs_test_opt(fs_info, SPACE_CACHE)) |
dcdf7f6d JB |
3657 | return 0; |
3658 | ||
3659 | path = btrfs_alloc_path(); | |
3660 | if (!path) | |
3661 | return -ENOMEM; | |
3662 | ||
3663 | /* Could add new block groups, use _safe just in case */ | |
3664 | list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs, | |
3665 | dirty_list) { | |
3666 | if (cache->disk_cache_state == BTRFS_DC_CLEAR) | |
3667 | cache_save_setup(cache, trans, path); | |
3668 | } | |
3669 | ||
3670 | btrfs_free_path(path); | |
3671 | return 0; | |
3672 | } | |
3673 | ||
1bbc621e CM |
3674 | /* |
3675 | * transaction commit does final block group cache writeback during a | |
3676 | * critical section where nothing is allowed to change the FS. This is | |
3677 | * required in order for the cache to actually match the block group, | |
3678 | * but can introduce a lot of latency into the commit. | |
3679 | * | |
3680 | * So, btrfs_start_dirty_block_groups is here to kick off block group | |
3681 | * cache IO. There's a chance we'll have to redo some of it if the | |
3682 | * block group changes again during the commit, but it greatly reduces | |
3683 | * the commit latency by getting rid of the easy block groups while | |
3684 | * we're still allowing others to join the commit. | |
3685 | */ | |
3686 | int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans, | |
2ff7e61e | 3687 | struct btrfs_fs_info *fs_info) |
9078a3e1 | 3688 | { |
4a8c9a62 | 3689 | struct btrfs_block_group_cache *cache; |
ce93ec54 JB |
3690 | struct btrfs_transaction *cur_trans = trans->transaction; |
3691 | int ret = 0; | |
c9dc4c65 | 3692 | int should_put; |
1bbc621e CM |
3693 | struct btrfs_path *path = NULL; |
3694 | LIST_HEAD(dirty); | |
3695 | struct list_head *io = &cur_trans->io_bgs; | |
c9dc4c65 | 3696 | int num_started = 0; |
1bbc621e CM |
3697 | int loops = 0; |
3698 | ||
3699 | spin_lock(&cur_trans->dirty_bgs_lock); | |
b58d1a9e FM |
3700 | if (list_empty(&cur_trans->dirty_bgs)) { |
3701 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3702 | return 0; | |
1bbc621e | 3703 | } |
b58d1a9e | 3704 | list_splice_init(&cur_trans->dirty_bgs, &dirty); |
1bbc621e | 3705 | spin_unlock(&cur_trans->dirty_bgs_lock); |
ce93ec54 | 3706 | |
1bbc621e | 3707 | again: |
1bbc621e CM |
3708 | /* |
3709 | * make sure all the block groups on our dirty list actually | |
3710 | * exist | |
3711 | */ | |
2ff7e61e | 3712 | btrfs_create_pending_block_groups(trans, fs_info); |
1bbc621e CM |
3713 | |
3714 | if (!path) { | |
3715 | path = btrfs_alloc_path(); | |
3716 | if (!path) | |
3717 | return -ENOMEM; | |
3718 | } | |
3719 | ||
b58d1a9e FM |
3720 | /* |
3721 | * cache_write_mutex is here only to save us from balance or automatic | |
3722 | * removal of empty block groups deleting this block group while we are | |
3723 | * writing out the cache | |
3724 | */ | |
3725 | mutex_lock(&trans->transaction->cache_write_mutex); | |
1bbc621e CM |
3726 | while (!list_empty(&dirty)) { |
3727 | cache = list_first_entry(&dirty, | |
3728 | struct btrfs_block_group_cache, | |
3729 | dirty_list); | |
1bbc621e CM |
3730 | /* |
3731 | * this can happen if something re-dirties a block | |
3732 | * group that is already under IO. Just wait for it to | |
3733 | * finish and then do it all again | |
3734 | */ | |
3735 | if (!list_empty(&cache->io_list)) { | |
3736 | list_del_init(&cache->io_list); | |
afdb5718 | 3737 | btrfs_wait_cache_io(trans, cache, path); |
1bbc621e CM |
3738 | btrfs_put_block_group(cache); |
3739 | } | |
3740 | ||
3741 | ||
3742 | /* | |
3743 | * btrfs_wait_cache_io uses the cache->dirty_list to decide | |
3744 | * if it should update the cache_state. Don't delete | |
3745 | * until after we wait. | |
3746 | * | |
3747 | * Since we're not running in the commit critical section | |
3748 | * we need the dirty_bgs_lock to protect from update_block_group | |
3749 | */ | |
3750 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3751 | list_del_init(&cache->dirty_list); | |
3752 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3753 | ||
3754 | should_put = 1; | |
3755 | ||
3756 | cache_save_setup(cache, trans, path); | |
3757 | ||
3758 | if (cache->disk_cache_state == BTRFS_DC_SETUP) { | |
3759 | cache->io_ctl.inode = NULL; | |
0b246afa | 3760 | ret = btrfs_write_out_cache(fs_info, trans, |
5b4aacef | 3761 | cache, path); |
1bbc621e CM |
3762 | if (ret == 0 && cache->io_ctl.inode) { |
3763 | num_started++; | |
3764 | should_put = 0; | |
3765 | ||
3766 | /* | |
3767 | * the cache_write_mutex is protecting | |
3768 | * the io_list | |
3769 | */ | |
3770 | list_add_tail(&cache->io_list, io); | |
3771 | } else { | |
3772 | /* | |
3773 | * if we failed to write the cache, the | |
3774 | * generation will be bad and life goes on | |
3775 | */ | |
3776 | ret = 0; | |
3777 | } | |
3778 | } | |
ff1f8250 | 3779 | if (!ret) { |
2ff7e61e JM |
3780 | ret = write_one_cache_group(trans, fs_info, |
3781 | path, cache); | |
ff1f8250 FM |
3782 | /* |
3783 | * Our block group might still be attached to the list | |
3784 | * of new block groups in the transaction handle of some | |
3785 | * other task (struct btrfs_trans_handle->new_bgs). This | |
3786 | * means its block group item isn't yet in the extent | |
3787 | * tree. If this happens ignore the error, as we will | |
3788 | * try again later in the critical section of the | |
3789 | * transaction commit. | |
3790 | */ | |
3791 | if (ret == -ENOENT) { | |
3792 | ret = 0; | |
3793 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3794 | if (list_empty(&cache->dirty_list)) { | |
3795 | list_add_tail(&cache->dirty_list, | |
3796 | &cur_trans->dirty_bgs); | |
3797 | btrfs_get_block_group(cache); | |
3798 | } | |
3799 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3800 | } else if (ret) { | |
66642832 | 3801 | btrfs_abort_transaction(trans, ret); |
ff1f8250 FM |
3802 | } |
3803 | } | |
1bbc621e CM |
3804 | |
3805 | /* if its not on the io list, we need to put the block group */ | |
3806 | if (should_put) | |
3807 | btrfs_put_block_group(cache); | |
3808 | ||
3809 | if (ret) | |
3810 | break; | |
b58d1a9e FM |
3811 | |
3812 | /* | |
3813 | * Avoid blocking other tasks for too long. It might even save | |
3814 | * us from writing caches for block groups that are going to be | |
3815 | * removed. | |
3816 | */ | |
3817 | mutex_unlock(&trans->transaction->cache_write_mutex); | |
3818 | mutex_lock(&trans->transaction->cache_write_mutex); | |
1bbc621e | 3819 | } |
b58d1a9e | 3820 | mutex_unlock(&trans->transaction->cache_write_mutex); |
1bbc621e CM |
3821 | |
3822 | /* | |
3823 | * go through delayed refs for all the stuff we've just kicked off | |
3824 | * and then loop back (just once) | |
3825 | */ | |
2ff7e61e | 3826 | ret = btrfs_run_delayed_refs(trans, fs_info, 0); |
1bbc621e CM |
3827 | if (!ret && loops == 0) { |
3828 | loops++; | |
3829 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3830 | list_splice_init(&cur_trans->dirty_bgs, &dirty); | |
b58d1a9e FM |
3831 | /* |
3832 | * dirty_bgs_lock protects us from concurrent block group | |
3833 | * deletes too (not just cache_write_mutex). | |
3834 | */ | |
3835 | if (!list_empty(&dirty)) { | |
3836 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3837 | goto again; | |
3838 | } | |
1bbc621e | 3839 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c79a1751 | 3840 | } else if (ret < 0) { |
2ff7e61e | 3841 | btrfs_cleanup_dirty_bgs(cur_trans, fs_info); |
1bbc621e CM |
3842 | } |
3843 | ||
3844 | btrfs_free_path(path); | |
3845 | return ret; | |
3846 | } | |
3847 | ||
3848 | int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans, | |
2ff7e61e | 3849 | struct btrfs_fs_info *fs_info) |
1bbc621e CM |
3850 | { |
3851 | struct btrfs_block_group_cache *cache; | |
3852 | struct btrfs_transaction *cur_trans = trans->transaction; | |
3853 | int ret = 0; | |
3854 | int should_put; | |
3855 | struct btrfs_path *path; | |
3856 | struct list_head *io = &cur_trans->io_bgs; | |
3857 | int num_started = 0; | |
9078a3e1 CM |
3858 | |
3859 | path = btrfs_alloc_path(); | |
3860 | if (!path) | |
3861 | return -ENOMEM; | |
3862 | ||
ce93ec54 | 3863 | /* |
e44081ef FM |
3864 | * Even though we are in the critical section of the transaction commit, |
3865 | * we can still have concurrent tasks adding elements to this | |
3866 | * transaction's list of dirty block groups. These tasks correspond to | |
3867 | * endio free space workers started when writeback finishes for a | |
3868 | * space cache, which run inode.c:btrfs_finish_ordered_io(), and can | |
3869 | * allocate new block groups as a result of COWing nodes of the root | |
3870 | * tree when updating the free space inode. The writeback for the space | |
3871 | * caches is triggered by an earlier call to | |
3872 | * btrfs_start_dirty_block_groups() and iterations of the following | |
3873 | * loop. | |
3874 | * Also we want to do the cache_save_setup first and then run the | |
ce93ec54 JB |
3875 | * delayed refs to make sure we have the best chance at doing this all |
3876 | * in one shot. | |
3877 | */ | |
e44081ef | 3878 | spin_lock(&cur_trans->dirty_bgs_lock); |
ce93ec54 JB |
3879 | while (!list_empty(&cur_trans->dirty_bgs)) { |
3880 | cache = list_first_entry(&cur_trans->dirty_bgs, | |
3881 | struct btrfs_block_group_cache, | |
3882 | dirty_list); | |
c9dc4c65 CM |
3883 | |
3884 | /* | |
3885 | * this can happen if cache_save_setup re-dirties a block | |
3886 | * group that is already under IO. Just wait for it to | |
3887 | * finish and then do it all again | |
3888 | */ | |
3889 | if (!list_empty(&cache->io_list)) { | |
e44081ef | 3890 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 | 3891 | list_del_init(&cache->io_list); |
afdb5718 | 3892 | btrfs_wait_cache_io(trans, cache, path); |
c9dc4c65 | 3893 | btrfs_put_block_group(cache); |
e44081ef | 3894 | spin_lock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 CM |
3895 | } |
3896 | ||
1bbc621e CM |
3897 | /* |
3898 | * don't remove from the dirty list until after we've waited | |
3899 | * on any pending IO | |
3900 | */ | |
ce93ec54 | 3901 | list_del_init(&cache->dirty_list); |
e44081ef | 3902 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 CM |
3903 | should_put = 1; |
3904 | ||
1bbc621e | 3905 | cache_save_setup(cache, trans, path); |
c9dc4c65 | 3906 | |
ce93ec54 | 3907 | if (!ret) |
2ff7e61e JM |
3908 | ret = btrfs_run_delayed_refs(trans, fs_info, |
3909 | (unsigned long) -1); | |
c9dc4c65 CM |
3910 | |
3911 | if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) { | |
3912 | cache->io_ctl.inode = NULL; | |
0b246afa | 3913 | ret = btrfs_write_out_cache(fs_info, trans, |
5b4aacef | 3914 | cache, path); |
c9dc4c65 CM |
3915 | if (ret == 0 && cache->io_ctl.inode) { |
3916 | num_started++; | |
3917 | should_put = 0; | |
1bbc621e | 3918 | list_add_tail(&cache->io_list, io); |
c9dc4c65 CM |
3919 | } else { |
3920 | /* | |
3921 | * if we failed to write the cache, the | |
3922 | * generation will be bad and life goes on | |
3923 | */ | |
3924 | ret = 0; | |
3925 | } | |
3926 | } | |
ff1f8250 | 3927 | if (!ret) { |
2ff7e61e JM |
3928 | ret = write_one_cache_group(trans, fs_info, |
3929 | path, cache); | |
2bc0bb5f FM |
3930 | /* |
3931 | * One of the free space endio workers might have | |
3932 | * created a new block group while updating a free space | |
3933 | * cache's inode (at inode.c:btrfs_finish_ordered_io()) | |
3934 | * and hasn't released its transaction handle yet, in | |
3935 | * which case the new block group is still attached to | |
3936 | * its transaction handle and its creation has not | |
3937 | * finished yet (no block group item in the extent tree | |
3938 | * yet, etc). If this is the case, wait for all free | |
3939 | * space endio workers to finish and retry. This is a | |
3940 | * a very rare case so no need for a more efficient and | |
3941 | * complex approach. | |
3942 | */ | |
3943 | if (ret == -ENOENT) { | |
3944 | wait_event(cur_trans->writer_wait, | |
3945 | atomic_read(&cur_trans->num_writers) == 1); | |
2ff7e61e JM |
3946 | ret = write_one_cache_group(trans, fs_info, |
3947 | path, cache); | |
2bc0bb5f | 3948 | } |
ff1f8250 | 3949 | if (ret) |
66642832 | 3950 | btrfs_abort_transaction(trans, ret); |
ff1f8250 | 3951 | } |
c9dc4c65 CM |
3952 | |
3953 | /* if its not on the io list, we need to put the block group */ | |
3954 | if (should_put) | |
3955 | btrfs_put_block_group(cache); | |
e44081ef | 3956 | spin_lock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 | 3957 | } |
e44081ef | 3958 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 | 3959 | |
1bbc621e CM |
3960 | while (!list_empty(io)) { |
3961 | cache = list_first_entry(io, struct btrfs_block_group_cache, | |
c9dc4c65 CM |
3962 | io_list); |
3963 | list_del_init(&cache->io_list); | |
afdb5718 | 3964 | btrfs_wait_cache_io(trans, cache, path); |
0cb59c99 JB |
3965 | btrfs_put_block_group(cache); |
3966 | } | |
3967 | ||
9078a3e1 | 3968 | btrfs_free_path(path); |
ce93ec54 | 3969 | return ret; |
9078a3e1 CM |
3970 | } |
3971 | ||
2ff7e61e | 3972 | int btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr) |
d2fb3437 YZ |
3973 | { |
3974 | struct btrfs_block_group_cache *block_group; | |
3975 | int readonly = 0; | |
3976 | ||
0b246afa | 3977 | block_group = btrfs_lookup_block_group(fs_info, bytenr); |
d2fb3437 YZ |
3978 | if (!block_group || block_group->ro) |
3979 | readonly = 1; | |
3980 | if (block_group) | |
fa9c0d79 | 3981 | btrfs_put_block_group(block_group); |
d2fb3437 YZ |
3982 | return readonly; |
3983 | } | |
3984 | ||
f78c436c FM |
3985 | bool btrfs_inc_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr) |
3986 | { | |
3987 | struct btrfs_block_group_cache *bg; | |
3988 | bool ret = true; | |
3989 | ||
3990 | bg = btrfs_lookup_block_group(fs_info, bytenr); | |
3991 | if (!bg) | |
3992 | return false; | |
3993 | ||
3994 | spin_lock(&bg->lock); | |
3995 | if (bg->ro) | |
3996 | ret = false; | |
3997 | else | |
3998 | atomic_inc(&bg->nocow_writers); | |
3999 | spin_unlock(&bg->lock); | |
4000 | ||
4001 | /* no put on block group, done by btrfs_dec_nocow_writers */ | |
4002 | if (!ret) | |
4003 | btrfs_put_block_group(bg); | |
4004 | ||
4005 | return ret; | |
4006 | ||
4007 | } | |
4008 | ||
4009 | void btrfs_dec_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr) | |
4010 | { | |
4011 | struct btrfs_block_group_cache *bg; | |
4012 | ||
4013 | bg = btrfs_lookup_block_group(fs_info, bytenr); | |
4014 | ASSERT(bg); | |
4015 | if (atomic_dec_and_test(&bg->nocow_writers)) | |
4016 | wake_up_atomic_t(&bg->nocow_writers); | |
4017 | /* | |
4018 | * Once for our lookup and once for the lookup done by a previous call | |
4019 | * to btrfs_inc_nocow_writers() | |
4020 | */ | |
4021 | btrfs_put_block_group(bg); | |
4022 | btrfs_put_block_group(bg); | |
4023 | } | |
4024 | ||
4025 | static int btrfs_wait_nocow_writers_atomic_t(atomic_t *a) | |
4026 | { | |
4027 | schedule(); | |
4028 | return 0; | |
4029 | } | |
4030 | ||
4031 | void btrfs_wait_nocow_writers(struct btrfs_block_group_cache *bg) | |
4032 | { | |
4033 | wait_on_atomic_t(&bg->nocow_writers, | |
4034 | btrfs_wait_nocow_writers_atomic_t, | |
4035 | TASK_UNINTERRUPTIBLE); | |
4036 | } | |
4037 | ||
6ab0a202 JM |
4038 | static const char *alloc_name(u64 flags) |
4039 | { | |
4040 | switch (flags) { | |
4041 | case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA: | |
4042 | return "mixed"; | |
4043 | case BTRFS_BLOCK_GROUP_METADATA: | |
4044 | return "metadata"; | |
4045 | case BTRFS_BLOCK_GROUP_DATA: | |
4046 | return "data"; | |
4047 | case BTRFS_BLOCK_GROUP_SYSTEM: | |
4048 | return "system"; | |
4049 | default: | |
4050 | WARN_ON(1); | |
4051 | return "invalid-combination"; | |
4052 | }; | |
4053 | } | |
4054 | ||
2be12ef7 NB |
4055 | static int create_space_info(struct btrfs_fs_info *info, u64 flags, |
4056 | struct btrfs_space_info **new) | |
4057 | { | |
4058 | ||
4059 | struct btrfs_space_info *space_info; | |
4060 | int i; | |
4061 | int ret; | |
4062 | ||
4063 | space_info = kzalloc(sizeof(*space_info), GFP_NOFS); | |
4064 | if (!space_info) | |
4065 | return -ENOMEM; | |
4066 | ||
4067 | ret = percpu_counter_init(&space_info->total_bytes_pinned, 0, | |
4068 | GFP_KERNEL); | |
4069 | if (ret) { | |
4070 | kfree(space_info); | |
4071 | return ret; | |
4072 | } | |
4073 | ||
4074 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) | |
4075 | INIT_LIST_HEAD(&space_info->block_groups[i]); | |
4076 | init_rwsem(&space_info->groups_sem); | |
4077 | spin_lock_init(&space_info->lock); | |
4078 | space_info->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK; | |
4079 | space_info->force_alloc = CHUNK_ALLOC_NO_FORCE; | |
4080 | init_waitqueue_head(&space_info->wait); | |
4081 | INIT_LIST_HEAD(&space_info->ro_bgs); | |
4082 | INIT_LIST_HEAD(&space_info->tickets); | |
4083 | INIT_LIST_HEAD(&space_info->priority_tickets); | |
4084 | ||
4085 | ret = kobject_init_and_add(&space_info->kobj, &space_info_ktype, | |
4086 | info->space_info_kobj, "%s", | |
4087 | alloc_name(space_info->flags)); | |
4088 | if (ret) { | |
535dedd9 | 4089 | kobject_put(&space_info->kobj); |
2be12ef7 NB |
4090 | return ret; |
4091 | } | |
4092 | ||
4093 | *new = space_info; | |
4094 | list_add_rcu(&space_info->list, &info->space_info); | |
4095 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
4096 | info->data_sinfo = space_info; | |
4097 | ||
4098 | return ret; | |
4099 | } | |
4100 | ||
d2006e6d | 4101 | static void update_space_info(struct btrfs_fs_info *info, u64 flags, |
593060d7 | 4102 | u64 total_bytes, u64 bytes_used, |
e40edf2d | 4103 | u64 bytes_readonly, |
593060d7 CM |
4104 | struct btrfs_space_info **space_info) |
4105 | { | |
4106 | struct btrfs_space_info *found; | |
b742bb82 YZ |
4107 | int factor; |
4108 | ||
4109 | if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
4110 | BTRFS_BLOCK_GROUP_RAID10)) | |
4111 | factor = 2; | |
4112 | else | |
4113 | factor = 1; | |
593060d7 CM |
4114 | |
4115 | found = __find_space_info(info, flags); | |
d2006e6d NB |
4116 | ASSERT(found); |
4117 | spin_lock(&found->lock); | |
4118 | found->total_bytes += total_bytes; | |
4119 | found->disk_total += total_bytes * factor; | |
4120 | found->bytes_used += bytes_used; | |
4121 | found->disk_used += bytes_used * factor; | |
4122 | found->bytes_readonly += bytes_readonly; | |
4123 | if (total_bytes > 0) | |
4124 | found->full = 0; | |
4125 | space_info_add_new_bytes(info, found, total_bytes - | |
4126 | bytes_used - bytes_readonly); | |
4127 | spin_unlock(&found->lock); | |
4128 | *space_info = found; | |
593060d7 CM |
4129 | } |
4130 | ||
8790d502 CM |
4131 | static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
4132 | { | |
899c81ea ID |
4133 | u64 extra_flags = chunk_to_extended(flags) & |
4134 | BTRFS_EXTENDED_PROFILE_MASK; | |
a46d11a8 | 4135 | |
de98ced9 | 4136 | write_seqlock(&fs_info->profiles_lock); |
a46d11a8 ID |
4137 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
4138 | fs_info->avail_data_alloc_bits |= extra_flags; | |
4139 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
4140 | fs_info->avail_metadata_alloc_bits |= extra_flags; | |
4141 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
4142 | fs_info->avail_system_alloc_bits |= extra_flags; | |
de98ced9 | 4143 | write_sequnlock(&fs_info->profiles_lock); |
8790d502 | 4144 | } |
593060d7 | 4145 | |
fc67c450 ID |
4146 | /* |
4147 | * returns target flags in extended format or 0 if restripe for this | |
4148 | * chunk_type is not in progress | |
c6664b42 ID |
4149 | * |
4150 | * should be called with either volume_mutex or balance_lock held | |
fc67c450 ID |
4151 | */ |
4152 | static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags) | |
4153 | { | |
4154 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
4155 | u64 target = 0; | |
4156 | ||
fc67c450 ID |
4157 | if (!bctl) |
4158 | return 0; | |
4159 | ||
4160 | if (flags & BTRFS_BLOCK_GROUP_DATA && | |
4161 | bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
4162 | target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target; | |
4163 | } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM && | |
4164 | bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
4165 | target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target; | |
4166 | } else if (flags & BTRFS_BLOCK_GROUP_METADATA && | |
4167 | bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
4168 | target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target; | |
4169 | } | |
4170 | ||
4171 | return target; | |
4172 | } | |
4173 | ||
a46d11a8 ID |
4174 | /* |
4175 | * @flags: available profiles in extended format (see ctree.h) | |
4176 | * | |
e4d8ec0f ID |
4177 | * Returns reduced profile in chunk format. If profile changing is in |
4178 | * progress (either running or paused) picks the target profile (if it's | |
4179 | * already available), otherwise falls back to plain reducing. | |
a46d11a8 | 4180 | */ |
2ff7e61e | 4181 | static u64 btrfs_reduce_alloc_profile(struct btrfs_fs_info *fs_info, u64 flags) |
ec44a35c | 4182 | { |
0b246afa | 4183 | u64 num_devices = fs_info->fs_devices->rw_devices; |
fc67c450 | 4184 | u64 target; |
9c170b26 ZL |
4185 | u64 raid_type; |
4186 | u64 allowed = 0; | |
a061fc8d | 4187 | |
fc67c450 ID |
4188 | /* |
4189 | * see if restripe for this chunk_type is in progress, if so | |
4190 | * try to reduce to the target profile | |
4191 | */ | |
0b246afa JM |
4192 | spin_lock(&fs_info->balance_lock); |
4193 | target = get_restripe_target(fs_info, flags); | |
fc67c450 ID |
4194 | if (target) { |
4195 | /* pick target profile only if it's already available */ | |
4196 | if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) { | |
0b246afa | 4197 | spin_unlock(&fs_info->balance_lock); |
fc67c450 | 4198 | return extended_to_chunk(target); |
e4d8ec0f ID |
4199 | } |
4200 | } | |
0b246afa | 4201 | spin_unlock(&fs_info->balance_lock); |
e4d8ec0f | 4202 | |
53b381b3 | 4203 | /* First, mask out the RAID levels which aren't possible */ |
9c170b26 ZL |
4204 | for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) { |
4205 | if (num_devices >= btrfs_raid_array[raid_type].devs_min) | |
4206 | allowed |= btrfs_raid_group[raid_type]; | |
4207 | } | |
4208 | allowed &= flags; | |
4209 | ||
4210 | if (allowed & BTRFS_BLOCK_GROUP_RAID6) | |
4211 | allowed = BTRFS_BLOCK_GROUP_RAID6; | |
4212 | else if (allowed & BTRFS_BLOCK_GROUP_RAID5) | |
4213 | allowed = BTRFS_BLOCK_GROUP_RAID5; | |
4214 | else if (allowed & BTRFS_BLOCK_GROUP_RAID10) | |
4215 | allowed = BTRFS_BLOCK_GROUP_RAID10; | |
4216 | else if (allowed & BTRFS_BLOCK_GROUP_RAID1) | |
4217 | allowed = BTRFS_BLOCK_GROUP_RAID1; | |
4218 | else if (allowed & BTRFS_BLOCK_GROUP_RAID0) | |
4219 | allowed = BTRFS_BLOCK_GROUP_RAID0; | |
4220 | ||
4221 | flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK; | |
4222 | ||
4223 | return extended_to_chunk(flags | allowed); | |
ec44a35c CM |
4224 | } |
4225 | ||
2ff7e61e | 4226 | static u64 get_alloc_profile(struct btrfs_fs_info *fs_info, u64 orig_flags) |
6a63209f | 4227 | { |
de98ced9 | 4228 | unsigned seq; |
f8213bdc | 4229 | u64 flags; |
de98ced9 MX |
4230 | |
4231 | do { | |
f8213bdc | 4232 | flags = orig_flags; |
0b246afa | 4233 | seq = read_seqbegin(&fs_info->profiles_lock); |
de98ced9 MX |
4234 | |
4235 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
0b246afa | 4236 | flags |= fs_info->avail_data_alloc_bits; |
de98ced9 | 4237 | else if (flags & BTRFS_BLOCK_GROUP_SYSTEM) |
0b246afa | 4238 | flags |= fs_info->avail_system_alloc_bits; |
de98ced9 | 4239 | else if (flags & BTRFS_BLOCK_GROUP_METADATA) |
0b246afa JM |
4240 | flags |= fs_info->avail_metadata_alloc_bits; |
4241 | } while (read_seqretry(&fs_info->profiles_lock, seq)); | |
6fef8df1 | 4242 | |
2ff7e61e | 4243 | return btrfs_reduce_alloc_profile(fs_info, flags); |
6a63209f JB |
4244 | } |
4245 | ||
1b86826d | 4246 | static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data) |
9ed74f2d | 4247 | { |
0b246afa | 4248 | struct btrfs_fs_info *fs_info = root->fs_info; |
b742bb82 | 4249 | u64 flags; |
53b381b3 | 4250 | u64 ret; |
9ed74f2d | 4251 | |
b742bb82 YZ |
4252 | if (data) |
4253 | flags = BTRFS_BLOCK_GROUP_DATA; | |
0b246afa | 4254 | else if (root == fs_info->chunk_root) |
b742bb82 | 4255 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
9ed74f2d | 4256 | else |
b742bb82 | 4257 | flags = BTRFS_BLOCK_GROUP_METADATA; |
9ed74f2d | 4258 | |
2ff7e61e | 4259 | ret = get_alloc_profile(fs_info, flags); |
53b381b3 | 4260 | return ret; |
6a63209f | 4261 | } |
9ed74f2d | 4262 | |
1b86826d JM |
4263 | u64 btrfs_data_alloc_profile(struct btrfs_fs_info *fs_info) |
4264 | { | |
4265 | return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_DATA); | |
4266 | } | |
4267 | ||
4268 | u64 btrfs_metadata_alloc_profile(struct btrfs_fs_info *fs_info) | |
4269 | { | |
4270 | return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_METADATA); | |
4271 | } | |
4272 | ||
4273 | u64 btrfs_system_alloc_profile(struct btrfs_fs_info *fs_info) | |
4274 | { | |
4275 | return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); | |
4276 | } | |
4277 | ||
4136135b LB |
4278 | static u64 btrfs_space_info_used(struct btrfs_space_info *s_info, |
4279 | bool may_use_included) | |
4280 | { | |
4281 | ASSERT(s_info); | |
4282 | return s_info->bytes_used + s_info->bytes_reserved + | |
4283 | s_info->bytes_pinned + s_info->bytes_readonly + | |
4284 | (may_use_included ? s_info->bytes_may_use : 0); | |
4285 | } | |
4286 | ||
04f4f916 | 4287 | int btrfs_alloc_data_chunk_ondemand(struct btrfs_inode *inode, u64 bytes) |
6a63209f | 4288 | { |
04f4f916 | 4289 | struct btrfs_root *root = inode->root; |
b4d7c3c9 | 4290 | struct btrfs_fs_info *fs_info = root->fs_info; |
1174cade | 4291 | struct btrfs_space_info *data_sinfo = fs_info->data_sinfo; |
ab6e2410 | 4292 | u64 used; |
94b947b2 | 4293 | int ret = 0; |
c99f1b0c ZL |
4294 | int need_commit = 2; |
4295 | int have_pinned_space; | |
6a63209f | 4296 | |
6a63209f | 4297 | /* make sure bytes are sectorsize aligned */ |
0b246afa | 4298 | bytes = ALIGN(bytes, fs_info->sectorsize); |
6a63209f | 4299 | |
9dced186 | 4300 | if (btrfs_is_free_space_inode(inode)) { |
c99f1b0c | 4301 | need_commit = 0; |
9dced186 | 4302 | ASSERT(current->journal_info); |
0af3d00b JB |
4303 | } |
4304 | ||
6a63209f JB |
4305 | again: |
4306 | /* make sure we have enough space to handle the data first */ | |
4307 | spin_lock(&data_sinfo->lock); | |
4136135b | 4308 | used = btrfs_space_info_used(data_sinfo, true); |
ab6e2410 JB |
4309 | |
4310 | if (used + bytes > data_sinfo->total_bytes) { | |
4e06bdd6 | 4311 | struct btrfs_trans_handle *trans; |
9ed74f2d | 4312 | |
6a63209f JB |
4313 | /* |
4314 | * if we don't have enough free bytes in this space then we need | |
4315 | * to alloc a new chunk. | |
4316 | */ | |
b9fd47cd | 4317 | if (!data_sinfo->full) { |
6a63209f | 4318 | u64 alloc_target; |
9ed74f2d | 4319 | |
0e4f8f88 | 4320 | data_sinfo->force_alloc = CHUNK_ALLOC_FORCE; |
6a63209f | 4321 | spin_unlock(&data_sinfo->lock); |
1174cade | 4322 | |
1b86826d | 4323 | alloc_target = btrfs_data_alloc_profile(fs_info); |
9dced186 MX |
4324 | /* |
4325 | * It is ugly that we don't call nolock join | |
4326 | * transaction for the free space inode case here. | |
4327 | * But it is safe because we only do the data space | |
4328 | * reservation for the free space cache in the | |
4329 | * transaction context, the common join transaction | |
4330 | * just increase the counter of the current transaction | |
4331 | * handler, doesn't try to acquire the trans_lock of | |
4332 | * the fs. | |
4333 | */ | |
7a7eaa40 | 4334 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
4335 | if (IS_ERR(trans)) |
4336 | return PTR_ERR(trans); | |
9ed74f2d | 4337 | |
2ff7e61e | 4338 | ret = do_chunk_alloc(trans, fs_info, alloc_target, |
0e4f8f88 | 4339 | CHUNK_ALLOC_NO_FORCE); |
3a45bb20 | 4340 | btrfs_end_transaction(trans); |
d52a5b5f MX |
4341 | if (ret < 0) { |
4342 | if (ret != -ENOSPC) | |
4343 | return ret; | |
c99f1b0c ZL |
4344 | else { |
4345 | have_pinned_space = 1; | |
d52a5b5f | 4346 | goto commit_trans; |
c99f1b0c | 4347 | } |
d52a5b5f | 4348 | } |
9ed74f2d | 4349 | |
6a63209f JB |
4350 | goto again; |
4351 | } | |
f2bb8f5c JB |
4352 | |
4353 | /* | |
b150a4f1 | 4354 | * If we don't have enough pinned space to deal with this |
94b947b2 ZL |
4355 | * allocation, and no removed chunk in current transaction, |
4356 | * don't bother committing the transaction. | |
f2bb8f5c | 4357 | */ |
c99f1b0c ZL |
4358 | have_pinned_space = percpu_counter_compare( |
4359 | &data_sinfo->total_bytes_pinned, | |
4360 | used + bytes - data_sinfo->total_bytes); | |
6a63209f | 4361 | spin_unlock(&data_sinfo->lock); |
6a63209f | 4362 | |
4e06bdd6 | 4363 | /* commit the current transaction and try again */ |
d52a5b5f | 4364 | commit_trans: |
c99f1b0c | 4365 | if (need_commit && |
0b246afa | 4366 | !atomic_read(&fs_info->open_ioctl_trans)) { |
c99f1b0c | 4367 | need_commit--; |
b150a4f1 | 4368 | |
e1746e83 ZL |
4369 | if (need_commit > 0) { |
4370 | btrfs_start_delalloc_roots(fs_info, 0, -1); | |
6374e57a | 4371 | btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, |
0b246afa | 4372 | (u64)-1); |
e1746e83 | 4373 | } |
9a4e7276 | 4374 | |
7a7eaa40 | 4375 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
4376 | if (IS_ERR(trans)) |
4377 | return PTR_ERR(trans); | |
c99f1b0c | 4378 | if (have_pinned_space >= 0 || |
3204d33c JB |
4379 | test_bit(BTRFS_TRANS_HAVE_FREE_BGS, |
4380 | &trans->transaction->flags) || | |
c99f1b0c | 4381 | need_commit > 0) { |
3a45bb20 | 4382 | ret = btrfs_commit_transaction(trans); |
94b947b2 ZL |
4383 | if (ret) |
4384 | return ret; | |
d7c15171 | 4385 | /* |
c2d6cb16 FM |
4386 | * The cleaner kthread might still be doing iput |
4387 | * operations. Wait for it to finish so that | |
4388 | * more space is released. | |
d7c15171 | 4389 | */ |
0b246afa JM |
4390 | mutex_lock(&fs_info->cleaner_delayed_iput_mutex); |
4391 | mutex_unlock(&fs_info->cleaner_delayed_iput_mutex); | |
94b947b2 ZL |
4392 | goto again; |
4393 | } else { | |
3a45bb20 | 4394 | btrfs_end_transaction(trans); |
94b947b2 | 4395 | } |
4e06bdd6 | 4396 | } |
9ed74f2d | 4397 | |
0b246afa | 4398 | trace_btrfs_space_reservation(fs_info, |
cab45e22 JM |
4399 | "space_info:enospc", |
4400 | data_sinfo->flags, bytes, 1); | |
6a63209f JB |
4401 | return -ENOSPC; |
4402 | } | |
4403 | data_sinfo->bytes_may_use += bytes; | |
0b246afa | 4404 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 4405 | data_sinfo->flags, bytes, 1); |
6a63209f | 4406 | spin_unlock(&data_sinfo->lock); |
6a63209f | 4407 | |
e7457f97 | 4408 | return 0; |
9ed74f2d | 4409 | } |
6a63209f | 4410 | |
364ecf36 QW |
4411 | int btrfs_check_data_free_space(struct inode *inode, |
4412 | struct extent_changeset **reserved, u64 start, u64 len) | |
4ceff079 | 4413 | { |
0b246afa | 4414 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4ceff079 QW |
4415 | int ret; |
4416 | ||
4417 | /* align the range */ | |
0b246afa JM |
4418 | len = round_up(start + len, fs_info->sectorsize) - |
4419 | round_down(start, fs_info->sectorsize); | |
4420 | start = round_down(start, fs_info->sectorsize); | |
4ceff079 | 4421 | |
04f4f916 | 4422 | ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode), len); |
4ceff079 QW |
4423 | if (ret < 0) |
4424 | return ret; | |
4425 | ||
1e5ec2e7 | 4426 | /* Use new btrfs_qgroup_reserve_data to reserve precious data space. */ |
364ecf36 | 4427 | ret = btrfs_qgroup_reserve_data(inode, reserved, start, len); |
7bc329c1 | 4428 | if (ret < 0) |
1e5ec2e7 | 4429 | btrfs_free_reserved_data_space_noquota(inode, start, len); |
364ecf36 QW |
4430 | else |
4431 | ret = 0; | |
4ceff079 QW |
4432 | return ret; |
4433 | } | |
4434 | ||
4ceff079 QW |
4435 | /* |
4436 | * Called if we need to clear a data reservation for this inode | |
4437 | * Normally in a error case. | |
4438 | * | |
51773bec QW |
4439 | * This one will *NOT* use accurate qgroup reserved space API, just for case |
4440 | * which we can't sleep and is sure it won't affect qgroup reserved space. | |
4441 | * Like clear_bit_hook(). | |
4ceff079 | 4442 | */ |
51773bec QW |
4443 | void btrfs_free_reserved_data_space_noquota(struct inode *inode, u64 start, |
4444 | u64 len) | |
4ceff079 | 4445 | { |
0b246afa | 4446 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4ceff079 QW |
4447 | struct btrfs_space_info *data_sinfo; |
4448 | ||
4449 | /* Make sure the range is aligned to sectorsize */ | |
0b246afa JM |
4450 | len = round_up(start + len, fs_info->sectorsize) - |
4451 | round_down(start, fs_info->sectorsize); | |
4452 | start = round_down(start, fs_info->sectorsize); | |
4ceff079 | 4453 | |
0b246afa | 4454 | data_sinfo = fs_info->data_sinfo; |
4ceff079 QW |
4455 | spin_lock(&data_sinfo->lock); |
4456 | if (WARN_ON(data_sinfo->bytes_may_use < len)) | |
4457 | data_sinfo->bytes_may_use = 0; | |
4458 | else | |
4459 | data_sinfo->bytes_may_use -= len; | |
0b246afa | 4460 | trace_btrfs_space_reservation(fs_info, "space_info", |
4ceff079 QW |
4461 | data_sinfo->flags, len, 0); |
4462 | spin_unlock(&data_sinfo->lock); | |
4463 | } | |
4464 | ||
51773bec QW |
4465 | /* |
4466 | * Called if we need to clear a data reservation for this inode | |
4467 | * Normally in a error case. | |
4468 | * | |
01327610 | 4469 | * This one will handle the per-inode data rsv map for accurate reserved |
51773bec QW |
4470 | * space framework. |
4471 | */ | |
bc42bda2 QW |
4472 | void btrfs_free_reserved_data_space(struct inode *inode, |
4473 | struct extent_changeset *reserved, u64 start, u64 len) | |
51773bec | 4474 | { |
0c476a5d JM |
4475 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4476 | ||
4477 | /* Make sure the range is aligned to sectorsize */ | |
da17066c JM |
4478 | len = round_up(start + len, root->fs_info->sectorsize) - |
4479 | round_down(start, root->fs_info->sectorsize); | |
4480 | start = round_down(start, root->fs_info->sectorsize); | |
0c476a5d | 4481 | |
51773bec | 4482 | btrfs_free_reserved_data_space_noquota(inode, start, len); |
bc42bda2 | 4483 | btrfs_qgroup_free_data(inode, reserved, start, len); |
51773bec QW |
4484 | } |
4485 | ||
97e728d4 | 4486 | static void force_metadata_allocation(struct btrfs_fs_info *info) |
e3ccfa98 | 4487 | { |
97e728d4 JB |
4488 | struct list_head *head = &info->space_info; |
4489 | struct btrfs_space_info *found; | |
e3ccfa98 | 4490 | |
97e728d4 JB |
4491 | rcu_read_lock(); |
4492 | list_for_each_entry_rcu(found, head, list) { | |
4493 | if (found->flags & BTRFS_BLOCK_GROUP_METADATA) | |
0e4f8f88 | 4494 | found->force_alloc = CHUNK_ALLOC_FORCE; |
e3ccfa98 | 4495 | } |
97e728d4 | 4496 | rcu_read_unlock(); |
e3ccfa98 JB |
4497 | } |
4498 | ||
3c76cd84 MX |
4499 | static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global) |
4500 | { | |
4501 | return (global->size << 1); | |
4502 | } | |
4503 | ||
2ff7e61e | 4504 | static int should_alloc_chunk(struct btrfs_fs_info *fs_info, |
698d0082 | 4505 | struct btrfs_space_info *sinfo, int force) |
32c00aff | 4506 | { |
0b246afa | 4507 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; |
8d8aafee | 4508 | u64 bytes_used = btrfs_space_info_used(sinfo, false); |
e5bc2458 | 4509 | u64 thresh; |
e3ccfa98 | 4510 | |
0e4f8f88 CM |
4511 | if (force == CHUNK_ALLOC_FORCE) |
4512 | return 1; | |
4513 | ||
fb25e914 JB |
4514 | /* |
4515 | * We need to take into account the global rsv because for all intents | |
4516 | * and purposes it's used space. Don't worry about locking the | |
4517 | * global_rsv, it doesn't change except when the transaction commits. | |
4518 | */ | |
54338b5c | 4519 | if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA) |
8d8aafee | 4520 | bytes_used += calc_global_rsv_need_space(global_rsv); |
fb25e914 | 4521 | |
0e4f8f88 CM |
4522 | /* |
4523 | * in limited mode, we want to have some free space up to | |
4524 | * about 1% of the FS size. | |
4525 | */ | |
4526 | if (force == CHUNK_ALLOC_LIMITED) { | |
0b246afa | 4527 | thresh = btrfs_super_total_bytes(fs_info->super_copy); |
ee22184b | 4528 | thresh = max_t(u64, SZ_64M, div_factor_fine(thresh, 1)); |
0e4f8f88 | 4529 | |
8d8aafee | 4530 | if (sinfo->total_bytes - bytes_used < thresh) |
0e4f8f88 CM |
4531 | return 1; |
4532 | } | |
0e4f8f88 | 4533 | |
8d8aafee | 4534 | if (bytes_used + SZ_2M < div_factor(sinfo->total_bytes, 8)) |
14ed0ca6 | 4535 | return 0; |
424499db | 4536 | return 1; |
32c00aff JB |
4537 | } |
4538 | ||
2ff7e61e | 4539 | static u64 get_profile_num_devs(struct btrfs_fs_info *fs_info, u64 type) |
15d1ff81 LB |
4540 | { |
4541 | u64 num_dev; | |
4542 | ||
53b381b3 DW |
4543 | if (type & (BTRFS_BLOCK_GROUP_RAID10 | |
4544 | BTRFS_BLOCK_GROUP_RAID0 | | |
4545 | BTRFS_BLOCK_GROUP_RAID5 | | |
4546 | BTRFS_BLOCK_GROUP_RAID6)) | |
0b246afa | 4547 | num_dev = fs_info->fs_devices->rw_devices; |
15d1ff81 LB |
4548 | else if (type & BTRFS_BLOCK_GROUP_RAID1) |
4549 | num_dev = 2; | |
4550 | else | |
4551 | num_dev = 1; /* DUP or single */ | |
4552 | ||
39c2d7fa | 4553 | return num_dev; |
15d1ff81 LB |
4554 | } |
4555 | ||
39c2d7fa FM |
4556 | /* |
4557 | * If @is_allocation is true, reserve space in the system space info necessary | |
4558 | * for allocating a chunk, otherwise if it's false, reserve space necessary for | |
4559 | * removing a chunk. | |
4560 | */ | |
4561 | void check_system_chunk(struct btrfs_trans_handle *trans, | |
2ff7e61e | 4562 | struct btrfs_fs_info *fs_info, u64 type) |
15d1ff81 LB |
4563 | { |
4564 | struct btrfs_space_info *info; | |
4565 | u64 left; | |
4566 | u64 thresh; | |
4fbcdf66 | 4567 | int ret = 0; |
39c2d7fa | 4568 | u64 num_devs; |
4fbcdf66 FM |
4569 | |
4570 | /* | |
4571 | * Needed because we can end up allocating a system chunk and for an | |
4572 | * atomic and race free space reservation in the chunk block reserve. | |
4573 | */ | |
0b246afa | 4574 | ASSERT(mutex_is_locked(&fs_info->chunk_mutex)); |
15d1ff81 | 4575 | |
0b246afa | 4576 | info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); |
15d1ff81 | 4577 | spin_lock(&info->lock); |
4136135b | 4578 | left = info->total_bytes - btrfs_space_info_used(info, true); |
15d1ff81 LB |
4579 | spin_unlock(&info->lock); |
4580 | ||
2ff7e61e | 4581 | num_devs = get_profile_num_devs(fs_info, type); |
39c2d7fa FM |
4582 | |
4583 | /* num_devs device items to update and 1 chunk item to add or remove */ | |
0b246afa JM |
4584 | thresh = btrfs_calc_trunc_metadata_size(fs_info, num_devs) + |
4585 | btrfs_calc_trans_metadata_size(fs_info, 1); | |
39c2d7fa | 4586 | |
0b246afa JM |
4587 | if (left < thresh && btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
4588 | btrfs_info(fs_info, "left=%llu, need=%llu, flags=%llu", | |
4589 | left, thresh, type); | |
4590 | dump_space_info(fs_info, info, 0, 0); | |
15d1ff81 LB |
4591 | } |
4592 | ||
4593 | if (left < thresh) { | |
1b86826d | 4594 | u64 flags = btrfs_system_alloc_profile(fs_info); |
15d1ff81 | 4595 | |
4fbcdf66 FM |
4596 | /* |
4597 | * Ignore failure to create system chunk. We might end up not | |
4598 | * needing it, as we might not need to COW all nodes/leafs from | |
4599 | * the paths we visit in the chunk tree (they were already COWed | |
4600 | * or created in the current transaction for example). | |
4601 | */ | |
2ff7e61e | 4602 | ret = btrfs_alloc_chunk(trans, fs_info, flags); |
4fbcdf66 FM |
4603 | } |
4604 | ||
4605 | if (!ret) { | |
0b246afa JM |
4606 | ret = btrfs_block_rsv_add(fs_info->chunk_root, |
4607 | &fs_info->chunk_block_rsv, | |
4fbcdf66 FM |
4608 | thresh, BTRFS_RESERVE_NO_FLUSH); |
4609 | if (!ret) | |
4610 | trans->chunk_bytes_reserved += thresh; | |
15d1ff81 LB |
4611 | } |
4612 | } | |
4613 | ||
28b737f6 LB |
4614 | /* |
4615 | * If force is CHUNK_ALLOC_FORCE: | |
4616 | * - return 1 if it successfully allocates a chunk, | |
4617 | * - return errors including -ENOSPC otherwise. | |
4618 | * If force is NOT CHUNK_ALLOC_FORCE: | |
4619 | * - return 0 if it doesn't need to allocate a new chunk, | |
4620 | * - return 1 if it successfully allocates a chunk, | |
4621 | * - return errors including -ENOSPC otherwise. | |
4622 | */ | |
6324fbf3 | 4623 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
2ff7e61e | 4624 | struct btrfs_fs_info *fs_info, u64 flags, int force) |
9ed74f2d | 4625 | { |
6324fbf3 | 4626 | struct btrfs_space_info *space_info; |
6d74119f | 4627 | int wait_for_alloc = 0; |
9ed74f2d | 4628 | int ret = 0; |
9ed74f2d | 4629 | |
c6b305a8 JB |
4630 | /* Don't re-enter if we're already allocating a chunk */ |
4631 | if (trans->allocating_chunk) | |
4632 | return -ENOSPC; | |
4633 | ||
0b246afa | 4634 | space_info = __find_space_info(fs_info, flags); |
593060d7 | 4635 | if (!space_info) { |
2be12ef7 NB |
4636 | ret = create_space_info(fs_info, flags, &space_info); |
4637 | if (ret) | |
4638 | return ret; | |
9ed74f2d JB |
4639 | } |
4640 | ||
6d74119f | 4641 | again: |
25179201 | 4642 | spin_lock(&space_info->lock); |
9e622d6b | 4643 | if (force < space_info->force_alloc) |
0e4f8f88 | 4644 | force = space_info->force_alloc; |
25179201 | 4645 | if (space_info->full) { |
2ff7e61e | 4646 | if (should_alloc_chunk(fs_info, space_info, force)) |
09fb99a6 FDBM |
4647 | ret = -ENOSPC; |
4648 | else | |
4649 | ret = 0; | |
25179201 | 4650 | spin_unlock(&space_info->lock); |
09fb99a6 | 4651 | return ret; |
9ed74f2d JB |
4652 | } |
4653 | ||
2ff7e61e | 4654 | if (!should_alloc_chunk(fs_info, space_info, force)) { |
25179201 | 4655 | spin_unlock(&space_info->lock); |
6d74119f JB |
4656 | return 0; |
4657 | } else if (space_info->chunk_alloc) { | |
4658 | wait_for_alloc = 1; | |
4659 | } else { | |
4660 | space_info->chunk_alloc = 1; | |
9ed74f2d | 4661 | } |
0e4f8f88 | 4662 | |
25179201 | 4663 | spin_unlock(&space_info->lock); |
9ed74f2d | 4664 | |
6d74119f JB |
4665 | mutex_lock(&fs_info->chunk_mutex); |
4666 | ||
4667 | /* | |
4668 | * The chunk_mutex is held throughout the entirety of a chunk | |
4669 | * allocation, so once we've acquired the chunk_mutex we know that the | |
4670 | * other guy is done and we need to recheck and see if we should | |
4671 | * allocate. | |
4672 | */ | |
4673 | if (wait_for_alloc) { | |
4674 | mutex_unlock(&fs_info->chunk_mutex); | |
4675 | wait_for_alloc = 0; | |
370b3353 | 4676 | cond_resched(); |
6d74119f JB |
4677 | goto again; |
4678 | } | |
4679 | ||
c6b305a8 JB |
4680 | trans->allocating_chunk = true; |
4681 | ||
67377734 JB |
4682 | /* |
4683 | * If we have mixed data/metadata chunks we want to make sure we keep | |
4684 | * allocating mixed chunks instead of individual chunks. | |
4685 | */ | |
4686 | if (btrfs_mixed_space_info(space_info)) | |
4687 | flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA); | |
4688 | ||
97e728d4 JB |
4689 | /* |
4690 | * if we're doing a data chunk, go ahead and make sure that | |
4691 | * we keep a reasonable number of metadata chunks allocated in the | |
4692 | * FS as well. | |
4693 | */ | |
9ed74f2d | 4694 | if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) { |
97e728d4 JB |
4695 | fs_info->data_chunk_allocations++; |
4696 | if (!(fs_info->data_chunk_allocations % | |
4697 | fs_info->metadata_ratio)) | |
4698 | force_metadata_allocation(fs_info); | |
9ed74f2d JB |
4699 | } |
4700 | ||
15d1ff81 LB |
4701 | /* |
4702 | * Check if we have enough space in SYSTEM chunk because we may need | |
4703 | * to update devices. | |
4704 | */ | |
2ff7e61e | 4705 | check_system_chunk(trans, fs_info, flags); |
15d1ff81 | 4706 | |
2ff7e61e | 4707 | ret = btrfs_alloc_chunk(trans, fs_info, flags); |
c6b305a8 | 4708 | trans->allocating_chunk = false; |
92b8e897 | 4709 | |
9ed74f2d | 4710 | spin_lock(&space_info->lock); |
a81cb9a2 AO |
4711 | if (ret < 0 && ret != -ENOSPC) |
4712 | goto out; | |
9ed74f2d | 4713 | if (ret) |
6324fbf3 | 4714 | space_info->full = 1; |
424499db YZ |
4715 | else |
4716 | ret = 1; | |
6d74119f | 4717 | |
0e4f8f88 | 4718 | space_info->force_alloc = CHUNK_ALLOC_NO_FORCE; |
a81cb9a2 | 4719 | out: |
6d74119f | 4720 | space_info->chunk_alloc = 0; |
9ed74f2d | 4721 | spin_unlock(&space_info->lock); |
a25c75d5 | 4722 | mutex_unlock(&fs_info->chunk_mutex); |
00d80e34 FM |
4723 | /* |
4724 | * When we allocate a new chunk we reserve space in the chunk block | |
4725 | * reserve to make sure we can COW nodes/leafs in the chunk tree or | |
4726 | * add new nodes/leafs to it if we end up needing to do it when | |
4727 | * inserting the chunk item and updating device items as part of the | |
4728 | * second phase of chunk allocation, performed by | |
4729 | * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a | |
4730 | * large number of new block groups to create in our transaction | |
4731 | * handle's new_bgs list to avoid exhausting the chunk block reserve | |
4732 | * in extreme cases - like having a single transaction create many new | |
4733 | * block groups when starting to write out the free space caches of all | |
4734 | * the block groups that were made dirty during the lifetime of the | |
4735 | * transaction. | |
4736 | */ | |
d9a0540a | 4737 | if (trans->can_flush_pending_bgs && |
ee22184b | 4738 | trans->chunk_bytes_reserved >= (u64)SZ_2M) { |
2ff7e61e | 4739 | btrfs_create_pending_block_groups(trans, fs_info); |
00d80e34 FM |
4740 | btrfs_trans_release_chunk_metadata(trans); |
4741 | } | |
0f9dd46c | 4742 | return ret; |
6324fbf3 | 4743 | } |
9ed74f2d | 4744 | |
c1c4919b | 4745 | static int can_overcommit(struct btrfs_fs_info *fs_info, |
a80c8dcf | 4746 | struct btrfs_space_info *space_info, u64 bytes, |
c1c4919b JM |
4747 | enum btrfs_reserve_flush_enum flush, |
4748 | bool system_chunk) | |
a80c8dcf | 4749 | { |
0b246afa | 4750 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; |
957780eb | 4751 | u64 profile; |
3c76cd84 | 4752 | u64 space_size; |
a80c8dcf JB |
4753 | u64 avail; |
4754 | u64 used; | |
4755 | ||
957780eb JB |
4756 | /* Don't overcommit when in mixed mode. */ |
4757 | if (space_info->flags & BTRFS_BLOCK_GROUP_DATA) | |
4758 | return 0; | |
4759 | ||
c1c4919b JM |
4760 | if (system_chunk) |
4761 | profile = btrfs_system_alloc_profile(fs_info); | |
4762 | else | |
4763 | profile = btrfs_metadata_alloc_profile(fs_info); | |
4764 | ||
4136135b | 4765 | used = btrfs_space_info_used(space_info, false); |
96f1bb57 | 4766 | |
96f1bb57 JB |
4767 | /* |
4768 | * We only want to allow over committing if we have lots of actual space | |
4769 | * free, but if we don't have enough space to handle the global reserve | |
4770 | * space then we could end up having a real enospc problem when trying | |
4771 | * to allocate a chunk or some other such important allocation. | |
4772 | */ | |
3c76cd84 MX |
4773 | spin_lock(&global_rsv->lock); |
4774 | space_size = calc_global_rsv_need_space(global_rsv); | |
4775 | spin_unlock(&global_rsv->lock); | |
4776 | if (used + space_size >= space_info->total_bytes) | |
96f1bb57 JB |
4777 | return 0; |
4778 | ||
4779 | used += space_info->bytes_may_use; | |
a80c8dcf | 4780 | |
a5ed45f8 | 4781 | avail = atomic64_read(&fs_info->free_chunk_space); |
a80c8dcf JB |
4782 | |
4783 | /* | |
4784 | * If we have dup, raid1 or raid10 then only half of the free | |
53b381b3 DW |
4785 | * space is actually useable. For raid56, the space info used |
4786 | * doesn't include the parity drive, so we don't have to | |
4787 | * change the math | |
a80c8dcf JB |
4788 | */ |
4789 | if (profile & (BTRFS_BLOCK_GROUP_DUP | | |
4790 | BTRFS_BLOCK_GROUP_RAID1 | | |
4791 | BTRFS_BLOCK_GROUP_RAID10)) | |
4792 | avail >>= 1; | |
4793 | ||
4794 | /* | |
561c294d MX |
4795 | * If we aren't flushing all things, let us overcommit up to |
4796 | * 1/2th of the space. If we can flush, don't let us overcommit | |
4797 | * too much, let it overcommit up to 1/8 of the space. | |
a80c8dcf | 4798 | */ |
08e007d2 | 4799 | if (flush == BTRFS_RESERVE_FLUSH_ALL) |
14575aef | 4800 | avail >>= 3; |
a80c8dcf | 4801 | else |
14575aef | 4802 | avail >>= 1; |
a80c8dcf | 4803 | |
14575aef | 4804 | if (used + bytes < space_info->total_bytes + avail) |
a80c8dcf JB |
4805 | return 1; |
4806 | return 0; | |
4807 | } | |
4808 | ||
2ff7e61e | 4809 | static void btrfs_writeback_inodes_sb_nr(struct btrfs_fs_info *fs_info, |
6c255e67 | 4810 | unsigned long nr_pages, int nr_items) |
da633a42 | 4811 | { |
0b246afa | 4812 | struct super_block *sb = fs_info->sb; |
da633a42 | 4813 | |
925a6efb JB |
4814 | if (down_read_trylock(&sb->s_umount)) { |
4815 | writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE); | |
4816 | up_read(&sb->s_umount); | |
4817 | } else { | |
da633a42 MX |
4818 | /* |
4819 | * We needn't worry the filesystem going from r/w to r/o though | |
4820 | * we don't acquire ->s_umount mutex, because the filesystem | |
4821 | * should guarantee the delalloc inodes list be empty after | |
4822 | * the filesystem is readonly(all dirty pages are written to | |
4823 | * the disk). | |
4824 | */ | |
0b246afa | 4825 | btrfs_start_delalloc_roots(fs_info, 0, nr_items); |
98ad69cf | 4826 | if (!current->journal_info) |
0b246afa | 4827 | btrfs_wait_ordered_roots(fs_info, nr_items, 0, (u64)-1); |
da633a42 MX |
4828 | } |
4829 | } | |
4830 | ||
6374e57a | 4831 | static inline u64 calc_reclaim_items_nr(struct btrfs_fs_info *fs_info, |
2ff7e61e | 4832 | u64 to_reclaim) |
18cd8ea6 MX |
4833 | { |
4834 | u64 bytes; | |
6374e57a | 4835 | u64 nr; |
18cd8ea6 | 4836 | |
2ff7e61e | 4837 | bytes = btrfs_calc_trans_metadata_size(fs_info, 1); |
6374e57a | 4838 | nr = div64_u64(to_reclaim, bytes); |
18cd8ea6 MX |
4839 | if (!nr) |
4840 | nr = 1; | |
4841 | return nr; | |
4842 | } | |
4843 | ||
ee22184b | 4844 | #define EXTENT_SIZE_PER_ITEM SZ_256K |
c61a16a7 | 4845 | |
9ed74f2d | 4846 | /* |
5da9d01b | 4847 | * shrink metadata reservation for delalloc |
9ed74f2d | 4848 | */ |
c1c4919b JM |
4849 | static void shrink_delalloc(struct btrfs_fs_info *fs_info, u64 to_reclaim, |
4850 | u64 orig, bool wait_ordered) | |
5da9d01b | 4851 | { |
0ca1f7ce | 4852 | struct btrfs_block_rsv *block_rsv; |
0019f10d | 4853 | struct btrfs_space_info *space_info; |
663350ac | 4854 | struct btrfs_trans_handle *trans; |
f4c738c2 | 4855 | u64 delalloc_bytes; |
5da9d01b | 4856 | u64 max_reclaim; |
6374e57a | 4857 | u64 items; |
b1953bce | 4858 | long time_left; |
d3ee29e3 MX |
4859 | unsigned long nr_pages; |
4860 | int loops; | |
08e007d2 | 4861 | enum btrfs_reserve_flush_enum flush; |
5da9d01b | 4862 | |
c61a16a7 | 4863 | /* Calc the number of the pages we need flush for space reservation */ |
2ff7e61e | 4864 | items = calc_reclaim_items_nr(fs_info, to_reclaim); |
6374e57a | 4865 | to_reclaim = items * EXTENT_SIZE_PER_ITEM; |
c61a16a7 | 4866 | |
663350ac | 4867 | trans = (struct btrfs_trans_handle *)current->journal_info; |
0b246afa | 4868 | block_rsv = &fs_info->delalloc_block_rsv; |
0019f10d | 4869 | space_info = block_rsv->space_info; |
bf9022e0 | 4870 | |
963d678b | 4871 | delalloc_bytes = percpu_counter_sum_positive( |
0b246afa | 4872 | &fs_info->delalloc_bytes); |
f4c738c2 | 4873 | if (delalloc_bytes == 0) { |
fdb5effd | 4874 | if (trans) |
f4c738c2 | 4875 | return; |
38c135af | 4876 | if (wait_ordered) |
0b246afa | 4877 | btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1); |
f4c738c2 | 4878 | return; |
fdb5effd JB |
4879 | } |
4880 | ||
d3ee29e3 | 4881 | loops = 0; |
f4c738c2 JB |
4882 | while (delalloc_bytes && loops < 3) { |
4883 | max_reclaim = min(delalloc_bytes, to_reclaim); | |
09cbfeaf | 4884 | nr_pages = max_reclaim >> PAGE_SHIFT; |
2ff7e61e | 4885 | btrfs_writeback_inodes_sb_nr(fs_info, nr_pages, items); |
dea31f52 JB |
4886 | /* |
4887 | * We need to wait for the async pages to actually start before | |
4888 | * we do anything. | |
4889 | */ | |
0b246afa | 4890 | max_reclaim = atomic_read(&fs_info->async_delalloc_pages); |
9f3a074d MX |
4891 | if (!max_reclaim) |
4892 | goto skip_async; | |
4893 | ||
4894 | if (max_reclaim <= nr_pages) | |
4895 | max_reclaim = 0; | |
4896 | else | |
4897 | max_reclaim -= nr_pages; | |
dea31f52 | 4898 | |
0b246afa JM |
4899 | wait_event(fs_info->async_submit_wait, |
4900 | atomic_read(&fs_info->async_delalloc_pages) <= | |
9f3a074d MX |
4901 | (int)max_reclaim); |
4902 | skip_async: | |
08e007d2 MX |
4903 | if (!trans) |
4904 | flush = BTRFS_RESERVE_FLUSH_ALL; | |
4905 | else | |
4906 | flush = BTRFS_RESERVE_NO_FLUSH; | |
0019f10d | 4907 | spin_lock(&space_info->lock); |
957780eb JB |
4908 | if (list_empty(&space_info->tickets) && |
4909 | list_empty(&space_info->priority_tickets)) { | |
4910 | spin_unlock(&space_info->lock); | |
4911 | break; | |
4912 | } | |
0019f10d | 4913 | spin_unlock(&space_info->lock); |
5da9d01b | 4914 | |
36e39c40 | 4915 | loops++; |
f104d044 | 4916 | if (wait_ordered && !trans) { |
0b246afa | 4917 | btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1); |
f104d044 | 4918 | } else { |
f4c738c2 | 4919 | time_left = schedule_timeout_killable(1); |
f104d044 JB |
4920 | if (time_left) |
4921 | break; | |
4922 | } | |
963d678b | 4923 | delalloc_bytes = percpu_counter_sum_positive( |
0b246afa | 4924 | &fs_info->delalloc_bytes); |
5da9d01b | 4925 | } |
5da9d01b YZ |
4926 | } |
4927 | ||
f1117628 JB |
4928 | struct reserve_ticket { |
4929 | u64 bytes; | |
4930 | int error; | |
4931 | struct list_head list; | |
4932 | wait_queue_head_t wait; | |
4933 | }; | |
4934 | ||
663350ac JB |
4935 | /** |
4936 | * maybe_commit_transaction - possibly commit the transaction if its ok to | |
4937 | * @root - the root we're allocating for | |
4938 | * @bytes - the number of bytes we want to reserve | |
4939 | * @force - force the commit | |
8bb8ab2e | 4940 | * |
663350ac JB |
4941 | * This will check to make sure that committing the transaction will actually |
4942 | * get us somewhere and then commit the transaction if it does. Otherwise it | |
4943 | * will return -ENOSPC. | |
8bb8ab2e | 4944 | */ |
0c9ab349 | 4945 | static int may_commit_transaction(struct btrfs_fs_info *fs_info, |
f1117628 | 4946 | struct btrfs_space_info *space_info) |
663350ac | 4947 | { |
f1117628 | 4948 | struct reserve_ticket *ticket = NULL; |
0b246afa | 4949 | struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_block_rsv; |
663350ac | 4950 | struct btrfs_trans_handle *trans; |
f1117628 | 4951 | u64 bytes; |
663350ac JB |
4952 | |
4953 | trans = (struct btrfs_trans_handle *)current->journal_info; | |
4954 | if (trans) | |
4955 | return -EAGAIN; | |
4956 | ||
f1117628 JB |
4957 | spin_lock(&space_info->lock); |
4958 | if (!list_empty(&space_info->priority_tickets)) | |
4959 | ticket = list_first_entry(&space_info->priority_tickets, | |
4960 | struct reserve_ticket, list); | |
4961 | else if (!list_empty(&space_info->tickets)) | |
4962 | ticket = list_first_entry(&space_info->tickets, | |
4963 | struct reserve_ticket, list); | |
4964 | bytes = (ticket) ? ticket->bytes : 0; | |
4965 | spin_unlock(&space_info->lock); | |
4966 | ||
4967 | if (!bytes) | |
4968 | return 0; | |
663350ac JB |
4969 | |
4970 | /* See if there is enough pinned space to make this reservation */ | |
b150a4f1 | 4971 | if (percpu_counter_compare(&space_info->total_bytes_pinned, |
0424c548 | 4972 | bytes) >= 0) |
663350ac | 4973 | goto commit; |
663350ac JB |
4974 | |
4975 | /* | |
4976 | * See if there is some space in the delayed insertion reservation for | |
4977 | * this reservation. | |
4978 | */ | |
4979 | if (space_info != delayed_rsv->space_info) | |
4980 | return -ENOSPC; | |
4981 | ||
4982 | spin_lock(&delayed_rsv->lock); | |
f1117628 JB |
4983 | if (delayed_rsv->size > bytes) |
4984 | bytes = 0; | |
4985 | else | |
4986 | bytes -= delayed_rsv->size; | |
b150a4f1 | 4987 | if (percpu_counter_compare(&space_info->total_bytes_pinned, |
f1117628 | 4988 | bytes) < 0) { |
663350ac JB |
4989 | spin_unlock(&delayed_rsv->lock); |
4990 | return -ENOSPC; | |
4991 | } | |
4992 | spin_unlock(&delayed_rsv->lock); | |
4993 | ||
4994 | commit: | |
a9b3311e | 4995 | trans = btrfs_join_transaction(fs_info->extent_root); |
663350ac JB |
4996 | if (IS_ERR(trans)) |
4997 | return -ENOSPC; | |
4998 | ||
3a45bb20 | 4999 | return btrfs_commit_transaction(trans); |
663350ac JB |
5000 | } |
5001 | ||
e38ae7a0 NB |
5002 | /* |
5003 | * Try to flush some data based on policy set by @state. This is only advisory | |
5004 | * and may fail for various reasons. The caller is supposed to examine the | |
5005 | * state of @space_info to detect the outcome. | |
5006 | */ | |
5007 | static void flush_space(struct btrfs_fs_info *fs_info, | |
96c3f433 | 5008 | struct btrfs_space_info *space_info, u64 num_bytes, |
7bdd6277 | 5009 | int state) |
96c3f433 | 5010 | { |
a9b3311e | 5011 | struct btrfs_root *root = fs_info->extent_root; |
96c3f433 JB |
5012 | struct btrfs_trans_handle *trans; |
5013 | int nr; | |
f4c738c2 | 5014 | int ret = 0; |
96c3f433 JB |
5015 | |
5016 | switch (state) { | |
96c3f433 JB |
5017 | case FLUSH_DELAYED_ITEMS_NR: |
5018 | case FLUSH_DELAYED_ITEMS: | |
18cd8ea6 | 5019 | if (state == FLUSH_DELAYED_ITEMS_NR) |
2ff7e61e | 5020 | nr = calc_reclaim_items_nr(fs_info, num_bytes) * 2; |
18cd8ea6 | 5021 | else |
96c3f433 | 5022 | nr = -1; |
18cd8ea6 | 5023 | |
96c3f433 JB |
5024 | trans = btrfs_join_transaction(root); |
5025 | if (IS_ERR(trans)) { | |
5026 | ret = PTR_ERR(trans); | |
5027 | break; | |
5028 | } | |
2ff7e61e | 5029 | ret = btrfs_run_delayed_items_nr(trans, fs_info, nr); |
3a45bb20 | 5030 | btrfs_end_transaction(trans); |
96c3f433 | 5031 | break; |
67b0fd63 JB |
5032 | case FLUSH_DELALLOC: |
5033 | case FLUSH_DELALLOC_WAIT: | |
7bdd6277 | 5034 | shrink_delalloc(fs_info, num_bytes * 2, num_bytes, |
67b0fd63 JB |
5035 | state == FLUSH_DELALLOC_WAIT); |
5036 | break; | |
ea658bad JB |
5037 | case ALLOC_CHUNK: |
5038 | trans = btrfs_join_transaction(root); | |
5039 | if (IS_ERR(trans)) { | |
5040 | ret = PTR_ERR(trans); | |
5041 | break; | |
5042 | } | |
2ff7e61e | 5043 | ret = do_chunk_alloc(trans, fs_info, |
1b86826d | 5044 | btrfs_metadata_alloc_profile(fs_info), |
ea658bad | 5045 | CHUNK_ALLOC_NO_FORCE); |
3a45bb20 | 5046 | btrfs_end_transaction(trans); |
eecba891 | 5047 | if (ret > 0 || ret == -ENOSPC) |
ea658bad JB |
5048 | ret = 0; |
5049 | break; | |
96c3f433 | 5050 | case COMMIT_TRANS: |
f1117628 | 5051 | ret = may_commit_transaction(fs_info, space_info); |
96c3f433 JB |
5052 | break; |
5053 | default: | |
5054 | ret = -ENOSPC; | |
5055 | break; | |
5056 | } | |
5057 | ||
7bdd6277 NB |
5058 | trace_btrfs_flush_space(fs_info, space_info->flags, num_bytes, state, |
5059 | ret); | |
e38ae7a0 | 5060 | return; |
96c3f433 | 5061 | } |
21c7e756 MX |
5062 | |
5063 | static inline u64 | |
c1c4919b JM |
5064 | btrfs_calc_reclaim_metadata_size(struct btrfs_fs_info *fs_info, |
5065 | struct btrfs_space_info *space_info, | |
5066 | bool system_chunk) | |
21c7e756 | 5067 | { |
957780eb | 5068 | struct reserve_ticket *ticket; |
21c7e756 MX |
5069 | u64 used; |
5070 | u64 expected; | |
957780eb | 5071 | u64 to_reclaim = 0; |
21c7e756 | 5072 | |
957780eb JB |
5073 | list_for_each_entry(ticket, &space_info->tickets, list) |
5074 | to_reclaim += ticket->bytes; | |
5075 | list_for_each_entry(ticket, &space_info->priority_tickets, list) | |
5076 | to_reclaim += ticket->bytes; | |
5077 | if (to_reclaim) | |
5078 | return to_reclaim; | |
21c7e756 | 5079 | |
e0af2484 | 5080 | to_reclaim = min_t(u64, num_online_cpus() * SZ_1M, SZ_16M); |
c1c4919b JM |
5081 | if (can_overcommit(fs_info, space_info, to_reclaim, |
5082 | BTRFS_RESERVE_FLUSH_ALL, system_chunk)) | |
e0af2484 WX |
5083 | return 0; |
5084 | ||
0eee8a49 NB |
5085 | used = btrfs_space_info_used(space_info, true); |
5086 | ||
c1c4919b JM |
5087 | if (can_overcommit(fs_info, space_info, SZ_1M, |
5088 | BTRFS_RESERVE_FLUSH_ALL, system_chunk)) | |
21c7e756 MX |
5089 | expected = div_factor_fine(space_info->total_bytes, 95); |
5090 | else | |
5091 | expected = div_factor_fine(space_info->total_bytes, 90); | |
5092 | ||
5093 | if (used > expected) | |
5094 | to_reclaim = used - expected; | |
5095 | else | |
5096 | to_reclaim = 0; | |
5097 | to_reclaim = min(to_reclaim, space_info->bytes_may_use + | |
5098 | space_info->bytes_reserved); | |
21c7e756 MX |
5099 | return to_reclaim; |
5100 | } | |
5101 | ||
c1c4919b JM |
5102 | static inline int need_do_async_reclaim(struct btrfs_fs_info *fs_info, |
5103 | struct btrfs_space_info *space_info, | |
5104 | u64 used, bool system_chunk) | |
21c7e756 | 5105 | { |
365c5313 JB |
5106 | u64 thresh = div_factor_fine(space_info->total_bytes, 98); |
5107 | ||
5108 | /* If we're just plain full then async reclaim just slows us down. */ | |
baee8790 | 5109 | if ((space_info->bytes_used + space_info->bytes_reserved) >= thresh) |
365c5313 JB |
5110 | return 0; |
5111 | ||
c1c4919b JM |
5112 | if (!btrfs_calc_reclaim_metadata_size(fs_info, space_info, |
5113 | system_chunk)) | |
d38b349c JB |
5114 | return 0; |
5115 | ||
0b246afa JM |
5116 | return (used >= thresh && !btrfs_fs_closing(fs_info) && |
5117 | !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state)); | |
21c7e756 MX |
5118 | } |
5119 | ||
957780eb | 5120 | static void wake_all_tickets(struct list_head *head) |
21c7e756 | 5121 | { |
957780eb | 5122 | struct reserve_ticket *ticket; |
25ce459c | 5123 | |
957780eb JB |
5124 | while (!list_empty(head)) { |
5125 | ticket = list_first_entry(head, struct reserve_ticket, list); | |
5126 | list_del_init(&ticket->list); | |
5127 | ticket->error = -ENOSPC; | |
5128 | wake_up(&ticket->wait); | |
21c7e756 | 5129 | } |
21c7e756 MX |
5130 | } |
5131 | ||
957780eb JB |
5132 | /* |
5133 | * This is for normal flushers, we can wait all goddamned day if we want to. We | |
5134 | * will loop and continuously try to flush as long as we are making progress. | |
5135 | * We count progress as clearing off tickets each time we have to loop. | |
5136 | */ | |
21c7e756 MX |
5137 | static void btrfs_async_reclaim_metadata_space(struct work_struct *work) |
5138 | { | |
5139 | struct btrfs_fs_info *fs_info; | |
5140 | struct btrfs_space_info *space_info; | |
5141 | u64 to_reclaim; | |
5142 | int flush_state; | |
957780eb | 5143 | int commit_cycles = 0; |
ce129655 | 5144 | u64 last_tickets_id; |
21c7e756 MX |
5145 | |
5146 | fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work); | |
5147 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); | |
5148 | ||
957780eb | 5149 | spin_lock(&space_info->lock); |
c1c4919b JM |
5150 | to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info, |
5151 | false); | |
957780eb JB |
5152 | if (!to_reclaim) { |
5153 | space_info->flush = 0; | |
5154 | spin_unlock(&space_info->lock); | |
21c7e756 | 5155 | return; |
957780eb | 5156 | } |
ce129655 | 5157 | last_tickets_id = space_info->tickets_id; |
957780eb | 5158 | spin_unlock(&space_info->lock); |
21c7e756 MX |
5159 | |
5160 | flush_state = FLUSH_DELAYED_ITEMS_NR; | |
957780eb | 5161 | do { |
e38ae7a0 | 5162 | flush_space(fs_info, space_info, to_reclaim, flush_state); |
957780eb JB |
5163 | spin_lock(&space_info->lock); |
5164 | if (list_empty(&space_info->tickets)) { | |
5165 | space_info->flush = 0; | |
5166 | spin_unlock(&space_info->lock); | |
5167 | return; | |
5168 | } | |
c1c4919b JM |
5169 | to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, |
5170 | space_info, | |
5171 | false); | |
ce129655 | 5172 | if (last_tickets_id == space_info->tickets_id) { |
957780eb JB |
5173 | flush_state++; |
5174 | } else { | |
ce129655 | 5175 | last_tickets_id = space_info->tickets_id; |
957780eb JB |
5176 | flush_state = FLUSH_DELAYED_ITEMS_NR; |
5177 | if (commit_cycles) | |
5178 | commit_cycles--; | |
5179 | } | |
5180 | ||
5181 | if (flush_state > COMMIT_TRANS) { | |
5182 | commit_cycles++; | |
5183 | if (commit_cycles > 2) { | |
5184 | wake_all_tickets(&space_info->tickets); | |
5185 | space_info->flush = 0; | |
5186 | } else { | |
5187 | flush_state = FLUSH_DELAYED_ITEMS_NR; | |
5188 | } | |
5189 | } | |
5190 | spin_unlock(&space_info->lock); | |
5191 | } while (flush_state <= COMMIT_TRANS); | |
5192 | } | |
5193 | ||
5194 | void btrfs_init_async_reclaim_work(struct work_struct *work) | |
5195 | { | |
5196 | INIT_WORK(work, btrfs_async_reclaim_metadata_space); | |
5197 | } | |
5198 | ||
5199 | static void priority_reclaim_metadata_space(struct btrfs_fs_info *fs_info, | |
5200 | struct btrfs_space_info *space_info, | |
5201 | struct reserve_ticket *ticket) | |
5202 | { | |
5203 | u64 to_reclaim; | |
5204 | int flush_state = FLUSH_DELAYED_ITEMS_NR; | |
5205 | ||
5206 | spin_lock(&space_info->lock); | |
c1c4919b JM |
5207 | to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info, |
5208 | false); | |
957780eb JB |
5209 | if (!to_reclaim) { |
5210 | spin_unlock(&space_info->lock); | |
5211 | return; | |
5212 | } | |
5213 | spin_unlock(&space_info->lock); | |
5214 | ||
21c7e756 | 5215 | do { |
7bdd6277 | 5216 | flush_space(fs_info, space_info, to_reclaim, flush_state); |
21c7e756 | 5217 | flush_state++; |
957780eb JB |
5218 | spin_lock(&space_info->lock); |
5219 | if (ticket->bytes == 0) { | |
5220 | spin_unlock(&space_info->lock); | |
21c7e756 | 5221 | return; |
957780eb JB |
5222 | } |
5223 | spin_unlock(&space_info->lock); | |
5224 | ||
5225 | /* | |
5226 | * Priority flushers can't wait on delalloc without | |
5227 | * deadlocking. | |
5228 | */ | |
5229 | if (flush_state == FLUSH_DELALLOC || | |
5230 | flush_state == FLUSH_DELALLOC_WAIT) | |
5231 | flush_state = ALLOC_CHUNK; | |
365c5313 | 5232 | } while (flush_state < COMMIT_TRANS); |
21c7e756 MX |
5233 | } |
5234 | ||
957780eb JB |
5235 | static int wait_reserve_ticket(struct btrfs_fs_info *fs_info, |
5236 | struct btrfs_space_info *space_info, | |
5237 | struct reserve_ticket *ticket, u64 orig_bytes) | |
5238 | ||
21c7e756 | 5239 | { |
957780eb JB |
5240 | DEFINE_WAIT(wait); |
5241 | int ret = 0; | |
5242 | ||
5243 | spin_lock(&space_info->lock); | |
5244 | while (ticket->bytes > 0 && ticket->error == 0) { | |
5245 | ret = prepare_to_wait_event(&ticket->wait, &wait, TASK_KILLABLE); | |
5246 | if (ret) { | |
5247 | ret = -EINTR; | |
5248 | break; | |
5249 | } | |
5250 | spin_unlock(&space_info->lock); | |
5251 | ||
5252 | schedule(); | |
5253 | ||
5254 | finish_wait(&ticket->wait, &wait); | |
5255 | spin_lock(&space_info->lock); | |
5256 | } | |
5257 | if (!ret) | |
5258 | ret = ticket->error; | |
5259 | if (!list_empty(&ticket->list)) | |
5260 | list_del_init(&ticket->list); | |
5261 | if (ticket->bytes && ticket->bytes < orig_bytes) { | |
5262 | u64 num_bytes = orig_bytes - ticket->bytes; | |
5263 | space_info->bytes_may_use -= num_bytes; | |
5264 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5265 | space_info->flags, num_bytes, 0); | |
5266 | } | |
5267 | spin_unlock(&space_info->lock); | |
5268 | ||
5269 | return ret; | |
21c7e756 MX |
5270 | } |
5271 | ||
4a92b1b8 JB |
5272 | /** |
5273 | * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space | |
5274 | * @root - the root we're allocating for | |
957780eb | 5275 | * @space_info - the space info we want to allocate from |
4a92b1b8 | 5276 | * @orig_bytes - the number of bytes we want |
48fc7f7e | 5277 | * @flush - whether or not we can flush to make our reservation |
8bb8ab2e | 5278 | * |
01327610 | 5279 | * This will reserve orig_bytes number of bytes from the space info associated |
4a92b1b8 JB |
5280 | * with the block_rsv. If there is not enough space it will make an attempt to |
5281 | * flush out space to make room. It will do this by flushing delalloc if | |
5282 | * possible or committing the transaction. If flush is 0 then no attempts to | |
5283 | * regain reservations will be made and this will fail if there is not enough | |
5284 | * space already. | |
8bb8ab2e | 5285 | */ |
c1c4919b | 5286 | static int __reserve_metadata_bytes(struct btrfs_fs_info *fs_info, |
957780eb JB |
5287 | struct btrfs_space_info *space_info, |
5288 | u64 orig_bytes, | |
c1c4919b JM |
5289 | enum btrfs_reserve_flush_enum flush, |
5290 | bool system_chunk) | |
9ed74f2d | 5291 | { |
957780eb | 5292 | struct reserve_ticket ticket; |
2bf64758 | 5293 | u64 used; |
8bb8ab2e | 5294 | int ret = 0; |
9ed74f2d | 5295 | |
957780eb | 5296 | ASSERT(orig_bytes); |
8ca17f0f | 5297 | ASSERT(!current->journal_info || flush != BTRFS_RESERVE_FLUSH_ALL); |
fdb5effd | 5298 | |
8bb8ab2e | 5299 | spin_lock(&space_info->lock); |
fdb5effd | 5300 | ret = -ENOSPC; |
4136135b | 5301 | used = btrfs_space_info_used(space_info, true); |
9ed74f2d | 5302 | |
8bb8ab2e | 5303 | /* |
957780eb JB |
5304 | * If we have enough space then hooray, make our reservation and carry |
5305 | * on. If not see if we can overcommit, and if we can, hooray carry on. | |
5306 | * If not things get more complicated. | |
8bb8ab2e | 5307 | */ |
957780eb JB |
5308 | if (used + orig_bytes <= space_info->total_bytes) { |
5309 | space_info->bytes_may_use += orig_bytes; | |
0b246afa JM |
5310 | trace_btrfs_space_reservation(fs_info, "space_info", |
5311 | space_info->flags, orig_bytes, 1); | |
957780eb | 5312 | ret = 0; |
c1c4919b JM |
5313 | } else if (can_overcommit(fs_info, space_info, orig_bytes, flush, |
5314 | system_chunk)) { | |
44734ed1 | 5315 | space_info->bytes_may_use += orig_bytes; |
0b246afa JM |
5316 | trace_btrfs_space_reservation(fs_info, "space_info", |
5317 | space_info->flags, orig_bytes, 1); | |
44734ed1 | 5318 | ret = 0; |
2bf64758 JB |
5319 | } |
5320 | ||
8bb8ab2e | 5321 | /* |
957780eb JB |
5322 | * If we couldn't make a reservation then setup our reservation ticket |
5323 | * and kick the async worker if it's not already running. | |
08e007d2 | 5324 | * |
957780eb JB |
5325 | * If we are a priority flusher then we just need to add our ticket to |
5326 | * the list and we will do our own flushing further down. | |
8bb8ab2e | 5327 | */ |
72bcd99d | 5328 | if (ret && flush != BTRFS_RESERVE_NO_FLUSH) { |
957780eb JB |
5329 | ticket.bytes = orig_bytes; |
5330 | ticket.error = 0; | |
5331 | init_waitqueue_head(&ticket.wait); | |
5332 | if (flush == BTRFS_RESERVE_FLUSH_ALL) { | |
5333 | list_add_tail(&ticket.list, &space_info->tickets); | |
5334 | if (!space_info->flush) { | |
5335 | space_info->flush = 1; | |
0b246afa | 5336 | trace_btrfs_trigger_flush(fs_info, |
f376df2b JB |
5337 | space_info->flags, |
5338 | orig_bytes, flush, | |
5339 | "enospc"); | |
957780eb | 5340 | queue_work(system_unbound_wq, |
c1c4919b | 5341 | &fs_info->async_reclaim_work); |
957780eb JB |
5342 | } |
5343 | } else { | |
5344 | list_add_tail(&ticket.list, | |
5345 | &space_info->priority_tickets); | |
5346 | } | |
21c7e756 MX |
5347 | } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) { |
5348 | used += orig_bytes; | |
f6acfd50 JB |
5349 | /* |
5350 | * We will do the space reservation dance during log replay, | |
5351 | * which means we won't have fs_info->fs_root set, so don't do | |
5352 | * the async reclaim as we will panic. | |
5353 | */ | |
0b246afa | 5354 | if (!test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags) && |
c1c4919b JM |
5355 | need_do_async_reclaim(fs_info, space_info, |
5356 | used, system_chunk) && | |
0b246afa JM |
5357 | !work_busy(&fs_info->async_reclaim_work)) { |
5358 | trace_btrfs_trigger_flush(fs_info, space_info->flags, | |
5359 | orig_bytes, flush, "preempt"); | |
21c7e756 | 5360 | queue_work(system_unbound_wq, |
0b246afa | 5361 | &fs_info->async_reclaim_work); |
f376df2b | 5362 | } |
8bb8ab2e | 5363 | } |
f0486c68 | 5364 | spin_unlock(&space_info->lock); |
08e007d2 | 5365 | if (!ret || flush == BTRFS_RESERVE_NO_FLUSH) |
957780eb | 5366 | return ret; |
f0486c68 | 5367 | |
957780eb | 5368 | if (flush == BTRFS_RESERVE_FLUSH_ALL) |
0b246afa | 5369 | return wait_reserve_ticket(fs_info, space_info, &ticket, |
957780eb | 5370 | orig_bytes); |
08e007d2 | 5371 | |
957780eb | 5372 | ret = 0; |
0b246afa | 5373 | priority_reclaim_metadata_space(fs_info, space_info, &ticket); |
957780eb JB |
5374 | spin_lock(&space_info->lock); |
5375 | if (ticket.bytes) { | |
5376 | if (ticket.bytes < orig_bytes) { | |
5377 | u64 num_bytes = orig_bytes - ticket.bytes; | |
5378 | space_info->bytes_may_use -= num_bytes; | |
0b246afa JM |
5379 | trace_btrfs_space_reservation(fs_info, "space_info", |
5380 | space_info->flags, | |
5381 | num_bytes, 0); | |
08e007d2 | 5382 | |
957780eb JB |
5383 | } |
5384 | list_del_init(&ticket.list); | |
5385 | ret = -ENOSPC; | |
5386 | } | |
5387 | spin_unlock(&space_info->lock); | |
5388 | ASSERT(list_empty(&ticket.list)); | |
5389 | return ret; | |
5390 | } | |
8bb8ab2e | 5391 | |
957780eb JB |
5392 | /** |
5393 | * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space | |
5394 | * @root - the root we're allocating for | |
5395 | * @block_rsv - the block_rsv we're allocating for | |
5396 | * @orig_bytes - the number of bytes we want | |
5397 | * @flush - whether or not we can flush to make our reservation | |
5398 | * | |
5399 | * This will reserve orgi_bytes number of bytes from the space info associated | |
5400 | * with the block_rsv. If there is not enough space it will make an attempt to | |
5401 | * flush out space to make room. It will do this by flushing delalloc if | |
5402 | * possible or committing the transaction. If flush is 0 then no attempts to | |
5403 | * regain reservations will be made and this will fail if there is not enough | |
5404 | * space already. | |
5405 | */ | |
5406 | static int reserve_metadata_bytes(struct btrfs_root *root, | |
5407 | struct btrfs_block_rsv *block_rsv, | |
5408 | u64 orig_bytes, | |
5409 | enum btrfs_reserve_flush_enum flush) | |
5410 | { | |
0b246afa JM |
5411 | struct btrfs_fs_info *fs_info = root->fs_info; |
5412 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
957780eb | 5413 | int ret; |
c1c4919b | 5414 | bool system_chunk = (root == fs_info->chunk_root); |
957780eb | 5415 | |
c1c4919b JM |
5416 | ret = __reserve_metadata_bytes(fs_info, block_rsv->space_info, |
5417 | orig_bytes, flush, system_chunk); | |
5d80366e JB |
5418 | if (ret == -ENOSPC && |
5419 | unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) { | |
5d80366e JB |
5420 | if (block_rsv != global_rsv && |
5421 | !block_rsv_use_bytes(global_rsv, orig_bytes)) | |
5422 | ret = 0; | |
5423 | } | |
cab45e22 | 5424 | if (ret == -ENOSPC) |
0b246afa | 5425 | trace_btrfs_space_reservation(fs_info, "space_info:enospc", |
957780eb JB |
5426 | block_rsv->space_info->flags, |
5427 | orig_bytes, 1); | |
f0486c68 YZ |
5428 | return ret; |
5429 | } | |
5430 | ||
79787eaa JM |
5431 | static struct btrfs_block_rsv *get_block_rsv( |
5432 | const struct btrfs_trans_handle *trans, | |
5433 | const struct btrfs_root *root) | |
f0486c68 | 5434 | { |
0b246afa | 5435 | struct btrfs_fs_info *fs_info = root->fs_info; |
4c13d758 JB |
5436 | struct btrfs_block_rsv *block_rsv = NULL; |
5437 | ||
e9cf439f | 5438 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) || |
0b246afa JM |
5439 | (root == fs_info->csum_root && trans->adding_csums) || |
5440 | (root == fs_info->uuid_root)) | |
f7a81ea4 SB |
5441 | block_rsv = trans->block_rsv; |
5442 | ||
4c13d758 | 5443 | if (!block_rsv) |
f0486c68 YZ |
5444 | block_rsv = root->block_rsv; |
5445 | ||
5446 | if (!block_rsv) | |
0b246afa | 5447 | block_rsv = &fs_info->empty_block_rsv; |
f0486c68 YZ |
5448 | |
5449 | return block_rsv; | |
5450 | } | |
5451 | ||
5452 | static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, | |
5453 | u64 num_bytes) | |
5454 | { | |
5455 | int ret = -ENOSPC; | |
5456 | spin_lock(&block_rsv->lock); | |
5457 | if (block_rsv->reserved >= num_bytes) { | |
5458 | block_rsv->reserved -= num_bytes; | |
5459 | if (block_rsv->reserved < block_rsv->size) | |
5460 | block_rsv->full = 0; | |
5461 | ret = 0; | |
5462 | } | |
5463 | spin_unlock(&block_rsv->lock); | |
5464 | return ret; | |
5465 | } | |
5466 | ||
5467 | static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv, | |
5468 | u64 num_bytes, int update_size) | |
5469 | { | |
5470 | spin_lock(&block_rsv->lock); | |
5471 | block_rsv->reserved += num_bytes; | |
5472 | if (update_size) | |
5473 | block_rsv->size += num_bytes; | |
5474 | else if (block_rsv->reserved >= block_rsv->size) | |
5475 | block_rsv->full = 1; | |
5476 | spin_unlock(&block_rsv->lock); | |
5477 | } | |
5478 | ||
d52be818 JB |
5479 | int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info, |
5480 | struct btrfs_block_rsv *dest, u64 num_bytes, | |
5481 | int min_factor) | |
5482 | { | |
5483 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
5484 | u64 min_bytes; | |
5485 | ||
5486 | if (global_rsv->space_info != dest->space_info) | |
5487 | return -ENOSPC; | |
5488 | ||
5489 | spin_lock(&global_rsv->lock); | |
5490 | min_bytes = div_factor(global_rsv->size, min_factor); | |
5491 | if (global_rsv->reserved < min_bytes + num_bytes) { | |
5492 | spin_unlock(&global_rsv->lock); | |
5493 | return -ENOSPC; | |
5494 | } | |
5495 | global_rsv->reserved -= num_bytes; | |
5496 | if (global_rsv->reserved < global_rsv->size) | |
5497 | global_rsv->full = 0; | |
5498 | spin_unlock(&global_rsv->lock); | |
5499 | ||
5500 | block_rsv_add_bytes(dest, num_bytes, 1); | |
5501 | return 0; | |
5502 | } | |
5503 | ||
957780eb JB |
5504 | /* |
5505 | * This is for space we already have accounted in space_info->bytes_may_use, so | |
5506 | * basically when we're returning space from block_rsv's. | |
5507 | */ | |
5508 | static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info, | |
5509 | struct btrfs_space_info *space_info, | |
5510 | u64 num_bytes) | |
5511 | { | |
5512 | struct reserve_ticket *ticket; | |
5513 | struct list_head *head; | |
5514 | u64 used; | |
5515 | enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_NO_FLUSH; | |
5516 | bool check_overcommit = false; | |
5517 | ||
5518 | spin_lock(&space_info->lock); | |
5519 | head = &space_info->priority_tickets; | |
5520 | ||
5521 | /* | |
5522 | * If we are over our limit then we need to check and see if we can | |
5523 | * overcommit, and if we can't then we just need to free up our space | |
5524 | * and not satisfy any requests. | |
5525 | */ | |
0eee8a49 | 5526 | used = btrfs_space_info_used(space_info, true); |
957780eb JB |
5527 | if (used - num_bytes >= space_info->total_bytes) |
5528 | check_overcommit = true; | |
5529 | again: | |
5530 | while (!list_empty(head) && num_bytes) { | |
5531 | ticket = list_first_entry(head, struct reserve_ticket, | |
5532 | list); | |
5533 | /* | |
5534 | * We use 0 bytes because this space is already reserved, so | |
5535 | * adding the ticket space would be a double count. | |
5536 | */ | |
5537 | if (check_overcommit && | |
c1c4919b | 5538 | !can_overcommit(fs_info, space_info, 0, flush, false)) |
957780eb JB |
5539 | break; |
5540 | if (num_bytes >= ticket->bytes) { | |
5541 | list_del_init(&ticket->list); | |
5542 | num_bytes -= ticket->bytes; | |
5543 | ticket->bytes = 0; | |
ce129655 | 5544 | space_info->tickets_id++; |
957780eb JB |
5545 | wake_up(&ticket->wait); |
5546 | } else { | |
5547 | ticket->bytes -= num_bytes; | |
5548 | num_bytes = 0; | |
5549 | } | |
5550 | } | |
5551 | ||
5552 | if (num_bytes && head == &space_info->priority_tickets) { | |
5553 | head = &space_info->tickets; | |
5554 | flush = BTRFS_RESERVE_FLUSH_ALL; | |
5555 | goto again; | |
5556 | } | |
5557 | space_info->bytes_may_use -= num_bytes; | |
5558 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5559 | space_info->flags, num_bytes, 0); | |
5560 | spin_unlock(&space_info->lock); | |
5561 | } | |
5562 | ||
5563 | /* | |
5564 | * This is for newly allocated space that isn't accounted in | |
5565 | * space_info->bytes_may_use yet. So if we allocate a chunk or unpin an extent | |
5566 | * we use this helper. | |
5567 | */ | |
5568 | static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info, | |
5569 | struct btrfs_space_info *space_info, | |
5570 | u64 num_bytes) | |
5571 | { | |
5572 | struct reserve_ticket *ticket; | |
5573 | struct list_head *head = &space_info->priority_tickets; | |
5574 | ||
5575 | again: | |
5576 | while (!list_empty(head) && num_bytes) { | |
5577 | ticket = list_first_entry(head, struct reserve_ticket, | |
5578 | list); | |
5579 | if (num_bytes >= ticket->bytes) { | |
5580 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5581 | space_info->flags, | |
5582 | ticket->bytes, 1); | |
5583 | list_del_init(&ticket->list); | |
5584 | num_bytes -= ticket->bytes; | |
5585 | space_info->bytes_may_use += ticket->bytes; | |
5586 | ticket->bytes = 0; | |
ce129655 | 5587 | space_info->tickets_id++; |
957780eb JB |
5588 | wake_up(&ticket->wait); |
5589 | } else { | |
5590 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5591 | space_info->flags, | |
5592 | num_bytes, 1); | |
5593 | space_info->bytes_may_use += num_bytes; | |
5594 | ticket->bytes -= num_bytes; | |
5595 | num_bytes = 0; | |
5596 | } | |
5597 | } | |
5598 | ||
5599 | if (num_bytes && head == &space_info->priority_tickets) { | |
5600 | head = &space_info->tickets; | |
5601 | goto again; | |
5602 | } | |
5603 | } | |
5604 | ||
8c2a3ca2 JB |
5605 | static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info, |
5606 | struct btrfs_block_rsv *block_rsv, | |
62a45b60 | 5607 | struct btrfs_block_rsv *dest, u64 num_bytes) |
f0486c68 YZ |
5608 | { |
5609 | struct btrfs_space_info *space_info = block_rsv->space_info; | |
5610 | ||
5611 | spin_lock(&block_rsv->lock); | |
5612 | if (num_bytes == (u64)-1) | |
5613 | num_bytes = block_rsv->size; | |
5614 | block_rsv->size -= num_bytes; | |
5615 | if (block_rsv->reserved >= block_rsv->size) { | |
5616 | num_bytes = block_rsv->reserved - block_rsv->size; | |
5617 | block_rsv->reserved = block_rsv->size; | |
5618 | block_rsv->full = 1; | |
5619 | } else { | |
5620 | num_bytes = 0; | |
5621 | } | |
5622 | spin_unlock(&block_rsv->lock); | |
5623 | ||
5624 | if (num_bytes > 0) { | |
5625 | if (dest) { | |
e9e22899 JB |
5626 | spin_lock(&dest->lock); |
5627 | if (!dest->full) { | |
5628 | u64 bytes_to_add; | |
5629 | ||
5630 | bytes_to_add = dest->size - dest->reserved; | |
5631 | bytes_to_add = min(num_bytes, bytes_to_add); | |
5632 | dest->reserved += bytes_to_add; | |
5633 | if (dest->reserved >= dest->size) | |
5634 | dest->full = 1; | |
5635 | num_bytes -= bytes_to_add; | |
5636 | } | |
5637 | spin_unlock(&dest->lock); | |
5638 | } | |
957780eb JB |
5639 | if (num_bytes) |
5640 | space_info_add_old_bytes(fs_info, space_info, | |
5641 | num_bytes); | |
9ed74f2d | 5642 | } |
f0486c68 | 5643 | } |
4e06bdd6 | 5644 | |
25d609f8 JB |
5645 | int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src, |
5646 | struct btrfs_block_rsv *dst, u64 num_bytes, | |
5647 | int update_size) | |
f0486c68 YZ |
5648 | { |
5649 | int ret; | |
9ed74f2d | 5650 | |
f0486c68 YZ |
5651 | ret = block_rsv_use_bytes(src, num_bytes); |
5652 | if (ret) | |
5653 | return ret; | |
9ed74f2d | 5654 | |
25d609f8 | 5655 | block_rsv_add_bytes(dst, num_bytes, update_size); |
9ed74f2d JB |
5656 | return 0; |
5657 | } | |
5658 | ||
66d8f3dd | 5659 | void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type) |
9ed74f2d | 5660 | { |
f0486c68 YZ |
5661 | memset(rsv, 0, sizeof(*rsv)); |
5662 | spin_lock_init(&rsv->lock); | |
66d8f3dd | 5663 | rsv->type = type; |
f0486c68 YZ |
5664 | } |
5665 | ||
2ff7e61e | 5666 | struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_fs_info *fs_info, |
66d8f3dd | 5667 | unsigned short type) |
f0486c68 YZ |
5668 | { |
5669 | struct btrfs_block_rsv *block_rsv; | |
9ed74f2d | 5670 | |
f0486c68 YZ |
5671 | block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS); |
5672 | if (!block_rsv) | |
5673 | return NULL; | |
9ed74f2d | 5674 | |
66d8f3dd | 5675 | btrfs_init_block_rsv(block_rsv, type); |
f0486c68 YZ |
5676 | block_rsv->space_info = __find_space_info(fs_info, |
5677 | BTRFS_BLOCK_GROUP_METADATA); | |
f0486c68 YZ |
5678 | return block_rsv; |
5679 | } | |
9ed74f2d | 5680 | |
2ff7e61e | 5681 | void btrfs_free_block_rsv(struct btrfs_fs_info *fs_info, |
f0486c68 YZ |
5682 | struct btrfs_block_rsv *rsv) |
5683 | { | |
2aaa6655 JB |
5684 | if (!rsv) |
5685 | return; | |
2ff7e61e | 5686 | btrfs_block_rsv_release(fs_info, rsv, (u64)-1); |
dabdb640 | 5687 | kfree(rsv); |
9ed74f2d JB |
5688 | } |
5689 | ||
cdfb080e CM |
5690 | void __btrfs_free_block_rsv(struct btrfs_block_rsv *rsv) |
5691 | { | |
5692 | kfree(rsv); | |
5693 | } | |
5694 | ||
08e007d2 MX |
5695 | int btrfs_block_rsv_add(struct btrfs_root *root, |
5696 | struct btrfs_block_rsv *block_rsv, u64 num_bytes, | |
5697 | enum btrfs_reserve_flush_enum flush) | |
9ed74f2d | 5698 | { |
f0486c68 | 5699 | int ret; |
9ed74f2d | 5700 | |
f0486c68 YZ |
5701 | if (num_bytes == 0) |
5702 | return 0; | |
8bb8ab2e | 5703 | |
61b520a9 | 5704 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
f0486c68 YZ |
5705 | if (!ret) { |
5706 | block_rsv_add_bytes(block_rsv, num_bytes, 1); | |
5707 | return 0; | |
5708 | } | |
9ed74f2d | 5709 | |
f0486c68 | 5710 | return ret; |
f0486c68 | 5711 | } |
9ed74f2d | 5712 | |
2ff7e61e | 5713 | int btrfs_block_rsv_check(struct btrfs_block_rsv *block_rsv, int min_factor) |
f0486c68 YZ |
5714 | { |
5715 | u64 num_bytes = 0; | |
f0486c68 | 5716 | int ret = -ENOSPC; |
9ed74f2d | 5717 | |
f0486c68 YZ |
5718 | if (!block_rsv) |
5719 | return 0; | |
9ed74f2d | 5720 | |
f0486c68 | 5721 | spin_lock(&block_rsv->lock); |
36ba022a JB |
5722 | num_bytes = div_factor(block_rsv->size, min_factor); |
5723 | if (block_rsv->reserved >= num_bytes) | |
5724 | ret = 0; | |
5725 | spin_unlock(&block_rsv->lock); | |
9ed74f2d | 5726 | |
36ba022a JB |
5727 | return ret; |
5728 | } | |
5729 | ||
08e007d2 MX |
5730 | int btrfs_block_rsv_refill(struct btrfs_root *root, |
5731 | struct btrfs_block_rsv *block_rsv, u64 min_reserved, | |
5732 | enum btrfs_reserve_flush_enum flush) | |
36ba022a JB |
5733 | { |
5734 | u64 num_bytes = 0; | |
5735 | int ret = -ENOSPC; | |
5736 | ||
5737 | if (!block_rsv) | |
5738 | return 0; | |
5739 | ||
5740 | spin_lock(&block_rsv->lock); | |
5741 | num_bytes = min_reserved; | |
13553e52 | 5742 | if (block_rsv->reserved >= num_bytes) |
f0486c68 | 5743 | ret = 0; |
13553e52 | 5744 | else |
f0486c68 | 5745 | num_bytes -= block_rsv->reserved; |
f0486c68 | 5746 | spin_unlock(&block_rsv->lock); |
13553e52 | 5747 | |
f0486c68 YZ |
5748 | if (!ret) |
5749 | return 0; | |
5750 | ||
aa38a711 | 5751 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
dabdb640 JB |
5752 | if (!ret) { |
5753 | block_rsv_add_bytes(block_rsv, num_bytes, 0); | |
f0486c68 | 5754 | return 0; |
6a63209f | 5755 | } |
9ed74f2d | 5756 | |
13553e52 | 5757 | return ret; |
f0486c68 YZ |
5758 | } |
5759 | ||
2ff7e61e | 5760 | void btrfs_block_rsv_release(struct btrfs_fs_info *fs_info, |
f0486c68 YZ |
5761 | struct btrfs_block_rsv *block_rsv, |
5762 | u64 num_bytes) | |
5763 | { | |
0b246afa JM |
5764 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; |
5765 | ||
17504584 | 5766 | if (global_rsv == block_rsv || |
f0486c68 YZ |
5767 | block_rsv->space_info != global_rsv->space_info) |
5768 | global_rsv = NULL; | |
0b246afa | 5769 | block_rsv_release_bytes(fs_info, block_rsv, global_rsv, num_bytes); |
6a63209f JB |
5770 | } |
5771 | ||
8929ecfa YZ |
5772 | static void update_global_block_rsv(struct btrfs_fs_info *fs_info) |
5773 | { | |
5774 | struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv; | |
5775 | struct btrfs_space_info *sinfo = block_rsv->space_info; | |
5776 | u64 num_bytes; | |
6a63209f | 5777 | |
ae2e4728 JB |
5778 | /* |
5779 | * The global block rsv is based on the size of the extent tree, the | |
5780 | * checksum tree and the root tree. If the fs is empty we want to set | |
5781 | * it to a minimal amount for safety. | |
5782 | */ | |
5783 | num_bytes = btrfs_root_used(&fs_info->extent_root->root_item) + | |
5784 | btrfs_root_used(&fs_info->csum_root->root_item) + | |
5785 | btrfs_root_used(&fs_info->tree_root->root_item); | |
5786 | num_bytes = max_t(u64, num_bytes, SZ_16M); | |
33b4d47f | 5787 | |
8929ecfa | 5788 | spin_lock(&sinfo->lock); |
1f699d38 | 5789 | spin_lock(&block_rsv->lock); |
4e06bdd6 | 5790 | |
ee22184b | 5791 | block_rsv->size = min_t(u64, num_bytes, SZ_512M); |
4e06bdd6 | 5792 | |
fb4b10e5 | 5793 | if (block_rsv->reserved < block_rsv->size) { |
4136135b | 5794 | num_bytes = btrfs_space_info_used(sinfo, true); |
fb4b10e5 JB |
5795 | if (sinfo->total_bytes > num_bytes) { |
5796 | num_bytes = sinfo->total_bytes - num_bytes; | |
5797 | num_bytes = min(num_bytes, | |
5798 | block_rsv->size - block_rsv->reserved); | |
5799 | block_rsv->reserved += num_bytes; | |
5800 | sinfo->bytes_may_use += num_bytes; | |
5801 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5802 | sinfo->flags, num_bytes, | |
5803 | 1); | |
5804 | } | |
5805 | } else if (block_rsv->reserved > block_rsv->size) { | |
8929ecfa | 5806 | num_bytes = block_rsv->reserved - block_rsv->size; |
fb25e914 | 5807 | sinfo->bytes_may_use -= num_bytes; |
8c2a3ca2 | 5808 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 5809 | sinfo->flags, num_bytes, 0); |
8929ecfa | 5810 | block_rsv->reserved = block_rsv->size; |
8929ecfa | 5811 | } |
182608c8 | 5812 | |
fb4b10e5 JB |
5813 | if (block_rsv->reserved == block_rsv->size) |
5814 | block_rsv->full = 1; | |
5815 | else | |
5816 | block_rsv->full = 0; | |
5817 | ||
8929ecfa | 5818 | spin_unlock(&block_rsv->lock); |
1f699d38 | 5819 | spin_unlock(&sinfo->lock); |
6a63209f JB |
5820 | } |
5821 | ||
f0486c68 | 5822 | static void init_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 5823 | { |
f0486c68 | 5824 | struct btrfs_space_info *space_info; |
6a63209f | 5825 | |
f0486c68 YZ |
5826 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); |
5827 | fs_info->chunk_block_rsv.space_info = space_info; | |
6a63209f | 5828 | |
f0486c68 | 5829 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
8929ecfa | 5830 | fs_info->global_block_rsv.space_info = space_info; |
8929ecfa | 5831 | fs_info->delalloc_block_rsv.space_info = space_info; |
f0486c68 YZ |
5832 | fs_info->trans_block_rsv.space_info = space_info; |
5833 | fs_info->empty_block_rsv.space_info = space_info; | |
6d668dda | 5834 | fs_info->delayed_block_rsv.space_info = space_info; |
f0486c68 | 5835 | |
8929ecfa YZ |
5836 | fs_info->extent_root->block_rsv = &fs_info->global_block_rsv; |
5837 | fs_info->csum_root->block_rsv = &fs_info->global_block_rsv; | |
5838 | fs_info->dev_root->block_rsv = &fs_info->global_block_rsv; | |
5839 | fs_info->tree_root->block_rsv = &fs_info->global_block_rsv; | |
3a6cad90 SB |
5840 | if (fs_info->quota_root) |
5841 | fs_info->quota_root->block_rsv = &fs_info->global_block_rsv; | |
f0486c68 | 5842 | fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv; |
8929ecfa | 5843 | |
8929ecfa | 5844 | update_global_block_rsv(fs_info); |
6a63209f JB |
5845 | } |
5846 | ||
8929ecfa | 5847 | static void release_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 5848 | { |
8c2a3ca2 JB |
5849 | block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL, |
5850 | (u64)-1); | |
8929ecfa YZ |
5851 | WARN_ON(fs_info->delalloc_block_rsv.size > 0); |
5852 | WARN_ON(fs_info->delalloc_block_rsv.reserved > 0); | |
5853 | WARN_ON(fs_info->trans_block_rsv.size > 0); | |
5854 | WARN_ON(fs_info->trans_block_rsv.reserved > 0); | |
5855 | WARN_ON(fs_info->chunk_block_rsv.size > 0); | |
5856 | WARN_ON(fs_info->chunk_block_rsv.reserved > 0); | |
6d668dda JB |
5857 | WARN_ON(fs_info->delayed_block_rsv.size > 0); |
5858 | WARN_ON(fs_info->delayed_block_rsv.reserved > 0); | |
fcb80c2a JB |
5859 | } |
5860 | ||
a22285a6 | 5861 | void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans, |
2ff7e61e | 5862 | struct btrfs_fs_info *fs_info) |
6a63209f | 5863 | { |
0e721106 JB |
5864 | if (!trans->block_rsv) |
5865 | return; | |
5866 | ||
a22285a6 YZ |
5867 | if (!trans->bytes_reserved) |
5868 | return; | |
6a63209f | 5869 | |
0b246afa | 5870 | trace_btrfs_space_reservation(fs_info, "transaction", |
2bcc0328 | 5871 | trans->transid, trans->bytes_reserved, 0); |
2ff7e61e JM |
5872 | btrfs_block_rsv_release(fs_info, trans->block_rsv, |
5873 | trans->bytes_reserved); | |
a22285a6 YZ |
5874 | trans->bytes_reserved = 0; |
5875 | } | |
6a63209f | 5876 | |
4fbcdf66 FM |
5877 | /* |
5878 | * To be called after all the new block groups attached to the transaction | |
5879 | * handle have been created (btrfs_create_pending_block_groups()). | |
5880 | */ | |
5881 | void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans) | |
5882 | { | |
64b63580 | 5883 | struct btrfs_fs_info *fs_info = trans->fs_info; |
4fbcdf66 FM |
5884 | |
5885 | if (!trans->chunk_bytes_reserved) | |
5886 | return; | |
5887 | ||
5888 | WARN_ON_ONCE(!list_empty(&trans->new_bgs)); | |
5889 | ||
5890 | block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL, | |
5891 | trans->chunk_bytes_reserved); | |
5892 | trans->chunk_bytes_reserved = 0; | |
5893 | } | |
5894 | ||
79787eaa | 5895 | /* Can only return 0 or -ENOSPC */ |
d68fc57b | 5896 | int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans, |
8ed7a2a0 | 5897 | struct btrfs_inode *inode) |
d68fc57b | 5898 | { |
8ed7a2a0 NB |
5899 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
5900 | struct btrfs_root *root = inode->root; | |
40acc3ee JB |
5901 | /* |
5902 | * We always use trans->block_rsv here as we will have reserved space | |
5903 | * for our orphan when starting the transaction, using get_block_rsv() | |
5904 | * here will sometimes make us choose the wrong block rsv as we could be | |
5905 | * doing a reloc inode for a non refcounted root. | |
5906 | */ | |
5907 | struct btrfs_block_rsv *src_rsv = trans->block_rsv; | |
d68fc57b YZ |
5908 | struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv; |
5909 | ||
5910 | /* | |
fcb80c2a JB |
5911 | * We need to hold space in order to delete our orphan item once we've |
5912 | * added it, so this takes the reservation so we can release it later | |
5913 | * when we are truly done with the orphan item. | |
d68fc57b | 5914 | */ |
0b246afa JM |
5915 | u64 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1); |
5916 | ||
8ed7a2a0 NB |
5917 | trace_btrfs_space_reservation(fs_info, "orphan", btrfs_ino(inode), |
5918 | num_bytes, 1); | |
25d609f8 | 5919 | return btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, 1); |
6a63209f JB |
5920 | } |
5921 | ||
703b391a | 5922 | void btrfs_orphan_release_metadata(struct btrfs_inode *inode) |
97e728d4 | 5923 | { |
703b391a NB |
5924 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
5925 | struct btrfs_root *root = inode->root; | |
0b246afa JM |
5926 | u64 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1); |
5927 | ||
703b391a NB |
5928 | trace_btrfs_space_reservation(fs_info, "orphan", btrfs_ino(inode), |
5929 | num_bytes, 0); | |
2ff7e61e | 5930 | btrfs_block_rsv_release(fs_info, root->orphan_block_rsv, num_bytes); |
d68fc57b | 5931 | } |
97e728d4 | 5932 | |
d5c12070 MX |
5933 | /* |
5934 | * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation | |
5935 | * root: the root of the parent directory | |
5936 | * rsv: block reservation | |
5937 | * items: the number of items that we need do reservation | |
5938 | * qgroup_reserved: used to return the reserved size in qgroup | |
5939 | * | |
5940 | * This function is used to reserve the space for snapshot/subvolume | |
5941 | * creation and deletion. Those operations are different with the | |
5942 | * common file/directory operations, they change two fs/file trees | |
5943 | * and root tree, the number of items that the qgroup reserves is | |
5944 | * different with the free space reservation. So we can not use | |
01327610 | 5945 | * the space reservation mechanism in start_transaction(). |
d5c12070 MX |
5946 | */ |
5947 | int btrfs_subvolume_reserve_metadata(struct btrfs_root *root, | |
5948 | struct btrfs_block_rsv *rsv, | |
5949 | int items, | |
ee3441b4 JM |
5950 | u64 *qgroup_reserved, |
5951 | bool use_global_rsv) | |
a22285a6 | 5952 | { |
d5c12070 MX |
5953 | u64 num_bytes; |
5954 | int ret; | |
0b246afa JM |
5955 | struct btrfs_fs_info *fs_info = root->fs_info; |
5956 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
d5c12070 | 5957 | |
0b246afa | 5958 | if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) { |
d5c12070 | 5959 | /* One for parent inode, two for dir entries */ |
0b246afa | 5960 | num_bytes = 3 * fs_info->nodesize; |
003d7c59 | 5961 | ret = btrfs_qgroup_reserve_meta(root, num_bytes, true); |
d5c12070 MX |
5962 | if (ret) |
5963 | return ret; | |
5964 | } else { | |
5965 | num_bytes = 0; | |
5966 | } | |
5967 | ||
5968 | *qgroup_reserved = num_bytes; | |
5969 | ||
0b246afa JM |
5970 | num_bytes = btrfs_calc_trans_metadata_size(fs_info, items); |
5971 | rsv->space_info = __find_space_info(fs_info, | |
d5c12070 MX |
5972 | BTRFS_BLOCK_GROUP_METADATA); |
5973 | ret = btrfs_block_rsv_add(root, rsv, num_bytes, | |
5974 | BTRFS_RESERVE_FLUSH_ALL); | |
ee3441b4 JM |
5975 | |
5976 | if (ret == -ENOSPC && use_global_rsv) | |
25d609f8 | 5977 | ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes, 1); |
ee3441b4 | 5978 | |
7174109c QW |
5979 | if (ret && *qgroup_reserved) |
5980 | btrfs_qgroup_free_meta(root, *qgroup_reserved); | |
d5c12070 MX |
5981 | |
5982 | return ret; | |
5983 | } | |
5984 | ||
2ff7e61e | 5985 | void btrfs_subvolume_release_metadata(struct btrfs_fs_info *fs_info, |
7775c818 | 5986 | struct btrfs_block_rsv *rsv) |
d5c12070 | 5987 | { |
2ff7e61e | 5988 | btrfs_block_rsv_release(fs_info, rsv, (u64)-1); |
97e728d4 JB |
5989 | } |
5990 | ||
7709cde3 JB |
5991 | /** |
5992 | * drop_outstanding_extent - drop an outstanding extent | |
5993 | * @inode: the inode we're dropping the extent for | |
01327610 | 5994 | * @num_bytes: the number of bytes we're releasing. |
7709cde3 JB |
5995 | * |
5996 | * This is called when we are freeing up an outstanding extent, either called | |
5997 | * after an error or after an extent is written. This will return the number of | |
5998 | * reserved extents that need to be freed. This must be called with | |
5999 | * BTRFS_I(inode)->lock held. | |
6000 | */ | |
baa3ba39 NB |
6001 | static unsigned drop_outstanding_extent(struct btrfs_inode *inode, |
6002 | u64 num_bytes) | |
9e0baf60 | 6003 | { |
7fd2ae21 | 6004 | unsigned drop_inode_space = 0; |
9e0baf60 | 6005 | unsigned dropped_extents = 0; |
823bb20a | 6006 | unsigned num_extents; |
9e0baf60 | 6007 | |
823bb20a | 6008 | num_extents = count_max_extents(num_bytes); |
dcab6a3b | 6009 | ASSERT(num_extents); |
baa3ba39 NB |
6010 | ASSERT(inode->outstanding_extents >= num_extents); |
6011 | inode->outstanding_extents -= num_extents; | |
9e0baf60 | 6012 | |
baa3ba39 | 6013 | if (inode->outstanding_extents == 0 && |
72ac3c0d | 6014 | test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED, |
baa3ba39 | 6015 | &inode->runtime_flags)) |
7fd2ae21 | 6016 | drop_inode_space = 1; |
7fd2ae21 | 6017 | |
9e0baf60 | 6018 | /* |
01327610 | 6019 | * If we have more or the same amount of outstanding extents than we have |
9e0baf60 JB |
6020 | * reserved then we need to leave the reserved extents count alone. |
6021 | */ | |
baa3ba39 | 6022 | if (inode->outstanding_extents >= inode->reserved_extents) |
7fd2ae21 | 6023 | return drop_inode_space; |
9e0baf60 | 6024 | |
baa3ba39 NB |
6025 | dropped_extents = inode->reserved_extents - inode->outstanding_extents; |
6026 | inode->reserved_extents -= dropped_extents; | |
7fd2ae21 | 6027 | return dropped_extents + drop_inode_space; |
9e0baf60 JB |
6028 | } |
6029 | ||
7709cde3 | 6030 | /** |
01327610 NS |
6031 | * calc_csum_metadata_size - return the amount of metadata space that must be |
6032 | * reserved/freed for the given bytes. | |
7709cde3 JB |
6033 | * @inode: the inode we're manipulating |
6034 | * @num_bytes: the number of bytes in question | |
6035 | * @reserve: 1 if we are reserving space, 0 if we are freeing space | |
6036 | * | |
6037 | * This adjusts the number of csum_bytes in the inode and then returns the | |
6038 | * correct amount of metadata that must either be reserved or freed. We | |
6039 | * calculate how many checksums we can fit into one leaf and then divide the | |
6040 | * number of bytes that will need to be checksumed by this value to figure out | |
6041 | * how many checksums will be required. If we are adding bytes then the number | |
6042 | * may go up and we will return the number of additional bytes that must be | |
6043 | * reserved. If it is going down we will return the number of bytes that must | |
6044 | * be freed. | |
6045 | * | |
6046 | * This must be called with BTRFS_I(inode)->lock held. | |
6047 | */ | |
0e6bf9b1 | 6048 | static u64 calc_csum_metadata_size(struct btrfs_inode *inode, u64 num_bytes, |
7709cde3 | 6049 | int reserve) |
6324fbf3 | 6050 | { |
0e6bf9b1 | 6051 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
1262133b | 6052 | u64 old_csums, num_csums; |
7709cde3 | 6053 | |
0e6bf9b1 | 6054 | if (inode->flags & BTRFS_INODE_NODATASUM && inode->csum_bytes == 0) |
7709cde3 JB |
6055 | return 0; |
6056 | ||
0e6bf9b1 | 6057 | old_csums = btrfs_csum_bytes_to_leaves(fs_info, inode->csum_bytes); |
7709cde3 | 6058 | if (reserve) |
0e6bf9b1 | 6059 | inode->csum_bytes += num_bytes; |
7709cde3 | 6060 | else |
0e6bf9b1 NB |
6061 | inode->csum_bytes -= num_bytes; |
6062 | num_csums = btrfs_csum_bytes_to_leaves(fs_info, inode->csum_bytes); | |
7709cde3 JB |
6063 | |
6064 | /* No change, no need to reserve more */ | |
6065 | if (old_csums == num_csums) | |
6066 | return 0; | |
6067 | ||
6068 | if (reserve) | |
0b246afa | 6069 | return btrfs_calc_trans_metadata_size(fs_info, |
7709cde3 JB |
6070 | num_csums - old_csums); |
6071 | ||
0b246afa | 6072 | return btrfs_calc_trans_metadata_size(fs_info, old_csums - num_csums); |
0ca1f7ce | 6073 | } |
c146afad | 6074 | |
9f3db423 | 6075 | int btrfs_delalloc_reserve_metadata(struct btrfs_inode *inode, u64 num_bytes) |
0ca1f7ce | 6076 | { |
9f3db423 NB |
6077 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
6078 | struct btrfs_root *root = inode->root; | |
0b246afa | 6079 | struct btrfs_block_rsv *block_rsv = &fs_info->delalloc_block_rsv; |
9e0baf60 | 6080 | u64 to_reserve = 0; |
660d3f6c | 6081 | u64 csum_bytes; |
823bb20a | 6082 | unsigned nr_extents; |
08e007d2 | 6083 | enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL; |
eb6b88d9 | 6084 | int ret = 0; |
c64c2bd8 | 6085 | bool delalloc_lock = true; |
88e081bf WS |
6086 | u64 to_free = 0; |
6087 | unsigned dropped; | |
48c3d480 | 6088 | bool release_extra = false; |
6324fbf3 | 6089 | |
c64c2bd8 JB |
6090 | /* If we are a free space inode we need to not flush since we will be in |
6091 | * the middle of a transaction commit. We also don't need the delalloc | |
6092 | * mutex since we won't race with anybody. We need this mostly to make | |
6093 | * lockdep shut its filthy mouth. | |
bac357dc JB |
6094 | * |
6095 | * If we have a transaction open (can happen if we call truncate_block | |
6096 | * from truncate), then we need FLUSH_LIMIT so we don't deadlock. | |
c64c2bd8 JB |
6097 | */ |
6098 | if (btrfs_is_free_space_inode(inode)) { | |
08e007d2 | 6099 | flush = BTRFS_RESERVE_NO_FLUSH; |
c64c2bd8 | 6100 | delalloc_lock = false; |
bac357dc JB |
6101 | } else if (current->journal_info) { |
6102 | flush = BTRFS_RESERVE_FLUSH_LIMIT; | |
c64c2bd8 | 6103 | } |
c09544e0 | 6104 | |
08e007d2 | 6105 | if (flush != BTRFS_RESERVE_NO_FLUSH && |
0b246afa | 6106 | btrfs_transaction_in_commit(fs_info)) |
0ca1f7ce | 6107 | schedule_timeout(1); |
ec44a35c | 6108 | |
c64c2bd8 | 6109 | if (delalloc_lock) |
9f3db423 | 6110 | mutex_lock(&inode->delalloc_mutex); |
c64c2bd8 | 6111 | |
0b246afa | 6112 | num_bytes = ALIGN(num_bytes, fs_info->sectorsize); |
8bb8ab2e | 6113 | |
9f3db423 | 6114 | spin_lock(&inode->lock); |
823bb20a | 6115 | nr_extents = count_max_extents(num_bytes); |
9f3db423 | 6116 | inode->outstanding_extents += nr_extents; |
9e0baf60 | 6117 | |
48c3d480 | 6118 | nr_extents = 0; |
9f3db423 NB |
6119 | if (inode->outstanding_extents > inode->reserved_extents) |
6120 | nr_extents += inode->outstanding_extents - | |
6121 | inode->reserved_extents; | |
57a45ced | 6122 | |
48c3d480 | 6123 | /* We always want to reserve a slot for updating the inode. */ |
0b246afa | 6124 | to_reserve = btrfs_calc_trans_metadata_size(fs_info, nr_extents + 1); |
7709cde3 | 6125 | to_reserve += calc_csum_metadata_size(inode, num_bytes, 1); |
9f3db423 NB |
6126 | csum_bytes = inode->csum_bytes; |
6127 | spin_unlock(&inode->lock); | |
57a45ced | 6128 | |
0b246afa | 6129 | if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) { |
7174109c | 6130 | ret = btrfs_qgroup_reserve_meta(root, |
003d7c59 | 6131 | nr_extents * fs_info->nodesize, true); |
88e081bf WS |
6132 | if (ret) |
6133 | goto out_fail; | |
6134 | } | |
c5567237 | 6135 | |
48c3d480 | 6136 | ret = btrfs_block_rsv_add(root, block_rsv, to_reserve, flush); |
88e081bf | 6137 | if (unlikely(ret)) { |
da17066c | 6138 | btrfs_qgroup_free_meta(root, |
0b246afa | 6139 | nr_extents * fs_info->nodesize); |
88e081bf | 6140 | goto out_fail; |
9e0baf60 | 6141 | } |
25179201 | 6142 | |
9f3db423 | 6143 | spin_lock(&inode->lock); |
48c3d480 | 6144 | if (test_and_set_bit(BTRFS_INODE_DELALLOC_META_RESERVED, |
9f3db423 | 6145 | &inode->runtime_flags)) { |
0b246afa | 6146 | to_reserve -= btrfs_calc_trans_metadata_size(fs_info, 1); |
48c3d480 | 6147 | release_extra = true; |
660d3f6c | 6148 | } |
9f3db423 NB |
6149 | inode->reserved_extents += nr_extents; |
6150 | spin_unlock(&inode->lock); | |
c64c2bd8 JB |
6151 | |
6152 | if (delalloc_lock) | |
9f3db423 | 6153 | mutex_unlock(&inode->delalloc_mutex); |
660d3f6c | 6154 | |
8c2a3ca2 | 6155 | if (to_reserve) |
0b246afa | 6156 | trace_btrfs_space_reservation(fs_info, "delalloc", |
9f3db423 | 6157 | btrfs_ino(inode), to_reserve, 1); |
48c3d480 | 6158 | if (release_extra) |
2ff7e61e | 6159 | btrfs_block_rsv_release(fs_info, block_rsv, |
0b246afa | 6160 | btrfs_calc_trans_metadata_size(fs_info, 1)); |
0ca1f7ce | 6161 | return 0; |
88e081bf WS |
6162 | |
6163 | out_fail: | |
9f3db423 | 6164 | spin_lock(&inode->lock); |
dcab6a3b | 6165 | dropped = drop_outstanding_extent(inode, num_bytes); |
88e081bf WS |
6166 | /* |
6167 | * If the inodes csum_bytes is the same as the original | |
6168 | * csum_bytes then we know we haven't raced with any free()ers | |
6169 | * so we can just reduce our inodes csum bytes and carry on. | |
88e081bf | 6170 | */ |
9f3db423 | 6171 | if (inode->csum_bytes == csum_bytes) { |
88e081bf | 6172 | calc_csum_metadata_size(inode, num_bytes, 0); |
f4881bc7 | 6173 | } else { |
9f3db423 | 6174 | u64 orig_csum_bytes = inode->csum_bytes; |
f4881bc7 JB |
6175 | u64 bytes; |
6176 | ||
6177 | /* | |
6178 | * This is tricky, but first we need to figure out how much we | |
01327610 | 6179 | * freed from any free-ers that occurred during this |
f4881bc7 JB |
6180 | * reservation, so we reset ->csum_bytes to the csum_bytes |
6181 | * before we dropped our lock, and then call the free for the | |
6182 | * number of bytes that were freed while we were trying our | |
6183 | * reservation. | |
6184 | */ | |
9f3db423 NB |
6185 | bytes = csum_bytes - inode->csum_bytes; |
6186 | inode->csum_bytes = csum_bytes; | |
f4881bc7 JB |
6187 | to_free = calc_csum_metadata_size(inode, bytes, 0); |
6188 | ||
6189 | ||
6190 | /* | |
6191 | * Now we need to see how much we would have freed had we not | |
6192 | * been making this reservation and our ->csum_bytes were not | |
6193 | * artificially inflated. | |
6194 | */ | |
9f3db423 | 6195 | inode->csum_bytes = csum_bytes - num_bytes; |
f4881bc7 JB |
6196 | bytes = csum_bytes - orig_csum_bytes; |
6197 | bytes = calc_csum_metadata_size(inode, bytes, 0); | |
6198 | ||
6199 | /* | |
6200 | * Now reset ->csum_bytes to what it should be. If bytes is | |
01327610 | 6201 | * more than to_free then we would have freed more space had we |
f4881bc7 JB |
6202 | * not had an artificially high ->csum_bytes, so we need to free |
6203 | * the remainder. If bytes is the same or less then we don't | |
6204 | * need to do anything, the other free-ers did the correct | |
6205 | * thing. | |
6206 | */ | |
9f3db423 | 6207 | inode->csum_bytes = orig_csum_bytes - num_bytes; |
f4881bc7 JB |
6208 | if (bytes > to_free) |
6209 | to_free = bytes - to_free; | |
6210 | else | |
6211 | to_free = 0; | |
6212 | } | |
9f3db423 | 6213 | spin_unlock(&inode->lock); |
e2d1f923 | 6214 | if (dropped) |
0b246afa | 6215 | to_free += btrfs_calc_trans_metadata_size(fs_info, dropped); |
88e081bf WS |
6216 | |
6217 | if (to_free) { | |
2ff7e61e | 6218 | btrfs_block_rsv_release(fs_info, block_rsv, to_free); |
0b246afa | 6219 | trace_btrfs_space_reservation(fs_info, "delalloc", |
9f3db423 | 6220 | btrfs_ino(inode), to_free, 0); |
88e081bf WS |
6221 | } |
6222 | if (delalloc_lock) | |
9f3db423 | 6223 | mutex_unlock(&inode->delalloc_mutex); |
88e081bf | 6224 | return ret; |
0ca1f7ce YZ |
6225 | } |
6226 | ||
7709cde3 JB |
6227 | /** |
6228 | * btrfs_delalloc_release_metadata - release a metadata reservation for an inode | |
6229 | * @inode: the inode to release the reservation for | |
6230 | * @num_bytes: the number of bytes we're releasing | |
6231 | * | |
6232 | * This will release the metadata reservation for an inode. This can be called | |
6233 | * once we complete IO for a given set of bytes to release their metadata | |
6234 | * reservations. | |
6235 | */ | |
691fa059 | 6236 | void btrfs_delalloc_release_metadata(struct btrfs_inode *inode, u64 num_bytes) |
0ca1f7ce | 6237 | { |
691fa059 | 6238 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
9e0baf60 JB |
6239 | u64 to_free = 0; |
6240 | unsigned dropped; | |
0ca1f7ce | 6241 | |
0b246afa | 6242 | num_bytes = ALIGN(num_bytes, fs_info->sectorsize); |
691fa059 | 6243 | spin_lock(&inode->lock); |
dcab6a3b | 6244 | dropped = drop_outstanding_extent(inode, num_bytes); |
97e728d4 | 6245 | |
0934856d MX |
6246 | if (num_bytes) |
6247 | to_free = calc_csum_metadata_size(inode, num_bytes, 0); | |
691fa059 | 6248 | spin_unlock(&inode->lock); |
9e0baf60 | 6249 | if (dropped > 0) |
0b246afa | 6250 | to_free += btrfs_calc_trans_metadata_size(fs_info, dropped); |
0ca1f7ce | 6251 | |
0b246afa | 6252 | if (btrfs_is_testing(fs_info)) |
6a3891c5 JB |
6253 | return; |
6254 | ||
691fa059 NB |
6255 | trace_btrfs_space_reservation(fs_info, "delalloc", btrfs_ino(inode), |
6256 | to_free, 0); | |
c5567237 | 6257 | |
2ff7e61e | 6258 | btrfs_block_rsv_release(fs_info, &fs_info->delalloc_block_rsv, to_free); |
0ca1f7ce YZ |
6259 | } |
6260 | ||
1ada3a62 | 6261 | /** |
7cf5b976 | 6262 | * btrfs_delalloc_reserve_space - reserve data and metadata space for |
1ada3a62 QW |
6263 | * delalloc |
6264 | * @inode: inode we're writing to | |
6265 | * @start: start range we are writing to | |
6266 | * @len: how long the range we are writing to | |
364ecf36 QW |
6267 | * @reserved: mandatory parameter, record actually reserved qgroup ranges of |
6268 | * current reservation. | |
1ada3a62 | 6269 | * |
1ada3a62 QW |
6270 | * This will do the following things |
6271 | * | |
6272 | * o reserve space in data space info for num bytes | |
6273 | * and reserve precious corresponding qgroup space | |
6274 | * (Done in check_data_free_space) | |
6275 | * | |
6276 | * o reserve space for metadata space, based on the number of outstanding | |
6277 | * extents and how much csums will be needed | |
6278 | * also reserve metadata space in a per root over-reserve method. | |
6279 | * o add to the inodes->delalloc_bytes | |
6280 | * o add it to the fs_info's delalloc inodes list. | |
6281 | * (Above 3 all done in delalloc_reserve_metadata) | |
6282 | * | |
6283 | * Return 0 for success | |
6284 | * Return <0 for error(-ENOSPC or -EQUOT) | |
6285 | */ | |
364ecf36 QW |
6286 | int btrfs_delalloc_reserve_space(struct inode *inode, |
6287 | struct extent_changeset **reserved, u64 start, u64 len) | |
1ada3a62 QW |
6288 | { |
6289 | int ret; | |
6290 | ||
364ecf36 | 6291 | ret = btrfs_check_data_free_space(inode, reserved, start, len); |
1ada3a62 QW |
6292 | if (ret < 0) |
6293 | return ret; | |
9f3db423 | 6294 | ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len); |
1ada3a62 | 6295 | if (ret < 0) |
bc42bda2 | 6296 | btrfs_free_reserved_data_space(inode, *reserved, start, len); |
1ada3a62 QW |
6297 | return ret; |
6298 | } | |
6299 | ||
7709cde3 | 6300 | /** |
7cf5b976 | 6301 | * btrfs_delalloc_release_space - release data and metadata space for delalloc |
1ada3a62 QW |
6302 | * @inode: inode we're releasing space for |
6303 | * @start: start position of the space already reserved | |
6304 | * @len: the len of the space already reserved | |
6305 | * | |
6306 | * This must be matched with a call to btrfs_delalloc_reserve_space. This is | |
6307 | * called in the case that we don't need the metadata AND data reservations | |
6308 | * anymore. So if there is an error or we insert an inline extent. | |
6309 | * | |
6310 | * This function will release the metadata space that was not used and will | |
6311 | * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes | |
6312 | * list if there are no delalloc bytes left. | |
6313 | * Also it will handle the qgroup reserved space. | |
6314 | */ | |
bc42bda2 QW |
6315 | void btrfs_delalloc_release_space(struct inode *inode, |
6316 | struct extent_changeset *reserved, u64 start, u64 len) | |
1ada3a62 | 6317 | { |
691fa059 | 6318 | btrfs_delalloc_release_metadata(BTRFS_I(inode), len); |
bc42bda2 | 6319 | btrfs_free_reserved_data_space(inode, reserved, start, len); |
6324fbf3 CM |
6320 | } |
6321 | ||
ce93ec54 | 6322 | static int update_block_group(struct btrfs_trans_handle *trans, |
6202df69 | 6323 | struct btrfs_fs_info *info, u64 bytenr, |
ce93ec54 | 6324 | u64 num_bytes, int alloc) |
9078a3e1 | 6325 | { |
0af3d00b | 6326 | struct btrfs_block_group_cache *cache = NULL; |
db94535d | 6327 | u64 total = num_bytes; |
9078a3e1 | 6328 | u64 old_val; |
db94535d | 6329 | u64 byte_in_group; |
0af3d00b | 6330 | int factor; |
3e1ad54f | 6331 | |
5d4f98a2 | 6332 | /* block accounting for super block */ |
eb73c1b7 | 6333 | spin_lock(&info->delalloc_root_lock); |
6c41761f | 6334 | old_val = btrfs_super_bytes_used(info->super_copy); |
5d4f98a2 YZ |
6335 | if (alloc) |
6336 | old_val += num_bytes; | |
6337 | else | |
6338 | old_val -= num_bytes; | |
6c41761f | 6339 | btrfs_set_super_bytes_used(info->super_copy, old_val); |
eb73c1b7 | 6340 | spin_unlock(&info->delalloc_root_lock); |
5d4f98a2 | 6341 | |
d397712b | 6342 | while (total) { |
db94535d | 6343 | cache = btrfs_lookup_block_group(info, bytenr); |
f3465ca4 | 6344 | if (!cache) |
79787eaa | 6345 | return -ENOENT; |
b742bb82 YZ |
6346 | if (cache->flags & (BTRFS_BLOCK_GROUP_DUP | |
6347 | BTRFS_BLOCK_GROUP_RAID1 | | |
6348 | BTRFS_BLOCK_GROUP_RAID10)) | |
6349 | factor = 2; | |
6350 | else | |
6351 | factor = 1; | |
9d66e233 JB |
6352 | /* |
6353 | * If this block group has free space cache written out, we | |
6354 | * need to make sure to load it if we are removing space. This | |
6355 | * is because we need the unpinning stage to actually add the | |
6356 | * space back to the block group, otherwise we will leak space. | |
6357 | */ | |
6358 | if (!alloc && cache->cached == BTRFS_CACHE_NO) | |
f6373bf3 | 6359 | cache_block_group(cache, 1); |
0af3d00b | 6360 | |
db94535d CM |
6361 | byte_in_group = bytenr - cache->key.objectid; |
6362 | WARN_ON(byte_in_group > cache->key.offset); | |
9078a3e1 | 6363 | |
25179201 | 6364 | spin_lock(&cache->space_info->lock); |
c286ac48 | 6365 | spin_lock(&cache->lock); |
0af3d00b | 6366 | |
6202df69 | 6367 | if (btrfs_test_opt(info, SPACE_CACHE) && |
0af3d00b JB |
6368 | cache->disk_cache_state < BTRFS_DC_CLEAR) |
6369 | cache->disk_cache_state = BTRFS_DC_CLEAR; | |
6370 | ||
9078a3e1 | 6371 | old_val = btrfs_block_group_used(&cache->item); |
db94535d | 6372 | num_bytes = min(total, cache->key.offset - byte_in_group); |
cd1bc465 | 6373 | if (alloc) { |
db94535d | 6374 | old_val += num_bytes; |
11833d66 YZ |
6375 | btrfs_set_block_group_used(&cache->item, old_val); |
6376 | cache->reserved -= num_bytes; | |
11833d66 | 6377 | cache->space_info->bytes_reserved -= num_bytes; |
b742bb82 YZ |
6378 | cache->space_info->bytes_used += num_bytes; |
6379 | cache->space_info->disk_used += num_bytes * factor; | |
c286ac48 | 6380 | spin_unlock(&cache->lock); |
25179201 | 6381 | spin_unlock(&cache->space_info->lock); |
cd1bc465 | 6382 | } else { |
db94535d | 6383 | old_val -= num_bytes; |
ae0ab003 FM |
6384 | btrfs_set_block_group_used(&cache->item, old_val); |
6385 | cache->pinned += num_bytes; | |
6386 | cache->space_info->bytes_pinned += num_bytes; | |
6387 | cache->space_info->bytes_used -= num_bytes; | |
6388 | cache->space_info->disk_used -= num_bytes * factor; | |
6389 | spin_unlock(&cache->lock); | |
6390 | spin_unlock(&cache->space_info->lock); | |
47ab2a6c | 6391 | |
0b246afa | 6392 | trace_btrfs_space_reservation(info, "pinned", |
c51e7bb1 JB |
6393 | cache->space_info->flags, |
6394 | num_bytes, 1); | |
d7eae340 OS |
6395 | percpu_counter_add(&cache->space_info->total_bytes_pinned, |
6396 | num_bytes); | |
ae0ab003 FM |
6397 | set_extent_dirty(info->pinned_extents, |
6398 | bytenr, bytenr + num_bytes - 1, | |
6399 | GFP_NOFS | __GFP_NOFAIL); | |
cd1bc465 | 6400 | } |
1bbc621e CM |
6401 | |
6402 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
6403 | if (list_empty(&cache->dirty_list)) { | |
6404 | list_add_tail(&cache->dirty_list, | |
6405 | &trans->transaction->dirty_bgs); | |
6406 | trans->transaction->num_dirty_bgs++; | |
6407 | btrfs_get_block_group(cache); | |
6408 | } | |
6409 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
6410 | ||
036a9348 FM |
6411 | /* |
6412 | * No longer have used bytes in this block group, queue it for | |
6413 | * deletion. We do this after adding the block group to the | |
6414 | * dirty list to avoid races between cleaner kthread and space | |
6415 | * cache writeout. | |
6416 | */ | |
6417 | if (!alloc && old_val == 0) { | |
6418 | spin_lock(&info->unused_bgs_lock); | |
6419 | if (list_empty(&cache->bg_list)) { | |
6420 | btrfs_get_block_group(cache); | |
6421 | list_add_tail(&cache->bg_list, | |
6422 | &info->unused_bgs); | |
6423 | } | |
6424 | spin_unlock(&info->unused_bgs_lock); | |
6425 | } | |
6426 | ||
fa9c0d79 | 6427 | btrfs_put_block_group(cache); |
db94535d CM |
6428 | total -= num_bytes; |
6429 | bytenr += num_bytes; | |
9078a3e1 CM |
6430 | } |
6431 | return 0; | |
6432 | } | |
6324fbf3 | 6433 | |
2ff7e61e | 6434 | static u64 first_logical_byte(struct btrfs_fs_info *fs_info, u64 search_start) |
a061fc8d | 6435 | { |
0f9dd46c | 6436 | struct btrfs_block_group_cache *cache; |
d2fb3437 | 6437 | u64 bytenr; |
0f9dd46c | 6438 | |
0b246afa JM |
6439 | spin_lock(&fs_info->block_group_cache_lock); |
6440 | bytenr = fs_info->first_logical_byte; | |
6441 | spin_unlock(&fs_info->block_group_cache_lock); | |
a1897fdd LB |
6442 | |
6443 | if (bytenr < (u64)-1) | |
6444 | return bytenr; | |
6445 | ||
0b246afa | 6446 | cache = btrfs_lookup_first_block_group(fs_info, search_start); |
0f9dd46c | 6447 | if (!cache) |
a061fc8d | 6448 | return 0; |
0f9dd46c | 6449 | |
d2fb3437 | 6450 | bytenr = cache->key.objectid; |
fa9c0d79 | 6451 | btrfs_put_block_group(cache); |
d2fb3437 YZ |
6452 | |
6453 | return bytenr; | |
a061fc8d CM |
6454 | } |
6455 | ||
2ff7e61e | 6456 | static int pin_down_extent(struct btrfs_fs_info *fs_info, |
f0486c68 YZ |
6457 | struct btrfs_block_group_cache *cache, |
6458 | u64 bytenr, u64 num_bytes, int reserved) | |
324ae4df | 6459 | { |
11833d66 YZ |
6460 | spin_lock(&cache->space_info->lock); |
6461 | spin_lock(&cache->lock); | |
6462 | cache->pinned += num_bytes; | |
6463 | cache->space_info->bytes_pinned += num_bytes; | |
6464 | if (reserved) { | |
6465 | cache->reserved -= num_bytes; | |
6466 | cache->space_info->bytes_reserved -= num_bytes; | |
6467 | } | |
6468 | spin_unlock(&cache->lock); | |
6469 | spin_unlock(&cache->space_info->lock); | |
68b38550 | 6470 | |
0b246afa | 6471 | trace_btrfs_space_reservation(fs_info, "pinned", |
c51e7bb1 | 6472 | cache->space_info->flags, num_bytes, 1); |
4da8b76d | 6473 | percpu_counter_add(&cache->space_info->total_bytes_pinned, num_bytes); |
0b246afa | 6474 | set_extent_dirty(fs_info->pinned_extents, bytenr, |
f0486c68 YZ |
6475 | bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL); |
6476 | return 0; | |
6477 | } | |
68b38550 | 6478 | |
f0486c68 YZ |
6479 | /* |
6480 | * this function must be called within transaction | |
6481 | */ | |
2ff7e61e | 6482 | int btrfs_pin_extent(struct btrfs_fs_info *fs_info, |
f0486c68 YZ |
6483 | u64 bytenr, u64 num_bytes, int reserved) |
6484 | { | |
6485 | struct btrfs_block_group_cache *cache; | |
68b38550 | 6486 | |
0b246afa | 6487 | cache = btrfs_lookup_block_group(fs_info, bytenr); |
79787eaa | 6488 | BUG_ON(!cache); /* Logic error */ |
f0486c68 | 6489 | |
2ff7e61e | 6490 | pin_down_extent(fs_info, cache, bytenr, num_bytes, reserved); |
f0486c68 YZ |
6491 | |
6492 | btrfs_put_block_group(cache); | |
11833d66 YZ |
6493 | return 0; |
6494 | } | |
6495 | ||
f0486c68 | 6496 | /* |
e688b725 CM |
6497 | * this function must be called within transaction |
6498 | */ | |
2ff7e61e | 6499 | int btrfs_pin_extent_for_log_replay(struct btrfs_fs_info *fs_info, |
e688b725 CM |
6500 | u64 bytenr, u64 num_bytes) |
6501 | { | |
6502 | struct btrfs_block_group_cache *cache; | |
b50c6e25 | 6503 | int ret; |
e688b725 | 6504 | |
0b246afa | 6505 | cache = btrfs_lookup_block_group(fs_info, bytenr); |
b50c6e25 JB |
6506 | if (!cache) |
6507 | return -EINVAL; | |
e688b725 CM |
6508 | |
6509 | /* | |
6510 | * pull in the free space cache (if any) so that our pin | |
6511 | * removes the free space from the cache. We have load_only set | |
6512 | * to one because the slow code to read in the free extents does check | |
6513 | * the pinned extents. | |
6514 | */ | |
f6373bf3 | 6515 | cache_block_group(cache, 1); |
e688b725 | 6516 | |
2ff7e61e | 6517 | pin_down_extent(fs_info, cache, bytenr, num_bytes, 0); |
e688b725 CM |
6518 | |
6519 | /* remove us from the free space cache (if we're there at all) */ | |
b50c6e25 | 6520 | ret = btrfs_remove_free_space(cache, bytenr, num_bytes); |
e688b725 | 6521 | btrfs_put_block_group(cache); |
b50c6e25 | 6522 | return ret; |
e688b725 CM |
6523 | } |
6524 | ||
2ff7e61e JM |
6525 | static int __exclude_logged_extent(struct btrfs_fs_info *fs_info, |
6526 | u64 start, u64 num_bytes) | |
8c2a1a30 JB |
6527 | { |
6528 | int ret; | |
6529 | struct btrfs_block_group_cache *block_group; | |
6530 | struct btrfs_caching_control *caching_ctl; | |
6531 | ||
0b246afa | 6532 | block_group = btrfs_lookup_block_group(fs_info, start); |
8c2a1a30 JB |
6533 | if (!block_group) |
6534 | return -EINVAL; | |
6535 | ||
6536 | cache_block_group(block_group, 0); | |
6537 | caching_ctl = get_caching_control(block_group); | |
6538 | ||
6539 | if (!caching_ctl) { | |
6540 | /* Logic error */ | |
6541 | BUG_ON(!block_group_cache_done(block_group)); | |
6542 | ret = btrfs_remove_free_space(block_group, start, num_bytes); | |
6543 | } else { | |
6544 | mutex_lock(&caching_ctl->mutex); | |
6545 | ||
6546 | if (start >= caching_ctl->progress) { | |
2ff7e61e | 6547 | ret = add_excluded_extent(fs_info, start, num_bytes); |
8c2a1a30 JB |
6548 | } else if (start + num_bytes <= caching_ctl->progress) { |
6549 | ret = btrfs_remove_free_space(block_group, | |
6550 | start, num_bytes); | |
6551 | } else { | |
6552 | num_bytes = caching_ctl->progress - start; | |
6553 | ret = btrfs_remove_free_space(block_group, | |
6554 | start, num_bytes); | |
6555 | if (ret) | |
6556 | goto out_lock; | |
6557 | ||
6558 | num_bytes = (start + num_bytes) - | |
6559 | caching_ctl->progress; | |
6560 | start = caching_ctl->progress; | |
2ff7e61e | 6561 | ret = add_excluded_extent(fs_info, start, num_bytes); |
8c2a1a30 JB |
6562 | } |
6563 | out_lock: | |
6564 | mutex_unlock(&caching_ctl->mutex); | |
6565 | put_caching_control(caching_ctl); | |
6566 | } | |
6567 | btrfs_put_block_group(block_group); | |
6568 | return ret; | |
6569 | } | |
6570 | ||
2ff7e61e | 6571 | int btrfs_exclude_logged_extents(struct btrfs_fs_info *fs_info, |
8c2a1a30 JB |
6572 | struct extent_buffer *eb) |
6573 | { | |
6574 | struct btrfs_file_extent_item *item; | |
6575 | struct btrfs_key key; | |
6576 | int found_type; | |
6577 | int i; | |
6578 | ||
2ff7e61e | 6579 | if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) |
8c2a1a30 JB |
6580 | return 0; |
6581 | ||
6582 | for (i = 0; i < btrfs_header_nritems(eb); i++) { | |
6583 | btrfs_item_key_to_cpu(eb, &key, i); | |
6584 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
6585 | continue; | |
6586 | item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item); | |
6587 | found_type = btrfs_file_extent_type(eb, item); | |
6588 | if (found_type == BTRFS_FILE_EXTENT_INLINE) | |
6589 | continue; | |
6590 | if (btrfs_file_extent_disk_bytenr(eb, item) == 0) | |
6591 | continue; | |
6592 | key.objectid = btrfs_file_extent_disk_bytenr(eb, item); | |
6593 | key.offset = btrfs_file_extent_disk_num_bytes(eb, item); | |
2ff7e61e | 6594 | __exclude_logged_extent(fs_info, key.objectid, key.offset); |
8c2a1a30 JB |
6595 | } |
6596 | ||
6597 | return 0; | |
6598 | } | |
6599 | ||
9cfa3e34 FM |
6600 | static void |
6601 | btrfs_inc_block_group_reservations(struct btrfs_block_group_cache *bg) | |
6602 | { | |
6603 | atomic_inc(&bg->reservations); | |
6604 | } | |
6605 | ||
6606 | void btrfs_dec_block_group_reservations(struct btrfs_fs_info *fs_info, | |
6607 | const u64 start) | |
6608 | { | |
6609 | struct btrfs_block_group_cache *bg; | |
6610 | ||
6611 | bg = btrfs_lookup_block_group(fs_info, start); | |
6612 | ASSERT(bg); | |
6613 | if (atomic_dec_and_test(&bg->reservations)) | |
6614 | wake_up_atomic_t(&bg->reservations); | |
6615 | btrfs_put_block_group(bg); | |
6616 | } | |
6617 | ||
6618 | static int btrfs_wait_bg_reservations_atomic_t(atomic_t *a) | |
6619 | { | |
6620 | schedule(); | |
6621 | return 0; | |
6622 | } | |
6623 | ||
6624 | void btrfs_wait_block_group_reservations(struct btrfs_block_group_cache *bg) | |
6625 | { | |
6626 | struct btrfs_space_info *space_info = bg->space_info; | |
6627 | ||
6628 | ASSERT(bg->ro); | |
6629 | ||
6630 | if (!(bg->flags & BTRFS_BLOCK_GROUP_DATA)) | |
6631 | return; | |
6632 | ||
6633 | /* | |
6634 | * Our block group is read only but before we set it to read only, | |
6635 | * some task might have had allocated an extent from it already, but it | |
6636 | * has not yet created a respective ordered extent (and added it to a | |
6637 | * root's list of ordered extents). | |
6638 | * Therefore wait for any task currently allocating extents, since the | |
6639 | * block group's reservations counter is incremented while a read lock | |
6640 | * on the groups' semaphore is held and decremented after releasing | |
6641 | * the read access on that semaphore and creating the ordered extent. | |
6642 | */ | |
6643 | down_write(&space_info->groups_sem); | |
6644 | up_write(&space_info->groups_sem); | |
6645 | ||
6646 | wait_on_atomic_t(&bg->reservations, | |
6647 | btrfs_wait_bg_reservations_atomic_t, | |
6648 | TASK_UNINTERRUPTIBLE); | |
6649 | } | |
6650 | ||
fb25e914 | 6651 | /** |
4824f1f4 | 6652 | * btrfs_add_reserved_bytes - update the block_group and space info counters |
fb25e914 | 6653 | * @cache: The cache we are manipulating |
18513091 WX |
6654 | * @ram_bytes: The number of bytes of file content, and will be same to |
6655 | * @num_bytes except for the compress path. | |
fb25e914 | 6656 | * @num_bytes: The number of bytes in question |
e570fd27 | 6657 | * @delalloc: The blocks are allocated for the delalloc write |
fb25e914 | 6658 | * |
745699ef XW |
6659 | * This is called by the allocator when it reserves space. If this is a |
6660 | * reservation and the block group has become read only we cannot make the | |
6661 | * reservation and return -EAGAIN, otherwise this function always succeeds. | |
f0486c68 | 6662 | */ |
4824f1f4 | 6663 | static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache *cache, |
18513091 | 6664 | u64 ram_bytes, u64 num_bytes, int delalloc) |
11833d66 | 6665 | { |
fb25e914 | 6666 | struct btrfs_space_info *space_info = cache->space_info; |
f0486c68 | 6667 | int ret = 0; |
79787eaa | 6668 | |
fb25e914 JB |
6669 | spin_lock(&space_info->lock); |
6670 | spin_lock(&cache->lock); | |
4824f1f4 WX |
6671 | if (cache->ro) { |
6672 | ret = -EAGAIN; | |
fb25e914 | 6673 | } else { |
4824f1f4 WX |
6674 | cache->reserved += num_bytes; |
6675 | space_info->bytes_reserved += num_bytes; | |
e570fd27 | 6676 | |
18513091 WX |
6677 | trace_btrfs_space_reservation(cache->fs_info, |
6678 | "space_info", space_info->flags, | |
6679 | ram_bytes, 0); | |
6680 | space_info->bytes_may_use -= ram_bytes; | |
e570fd27 | 6681 | if (delalloc) |
4824f1f4 | 6682 | cache->delalloc_bytes += num_bytes; |
324ae4df | 6683 | } |
fb25e914 JB |
6684 | spin_unlock(&cache->lock); |
6685 | spin_unlock(&space_info->lock); | |
f0486c68 | 6686 | return ret; |
324ae4df | 6687 | } |
9078a3e1 | 6688 | |
4824f1f4 WX |
6689 | /** |
6690 | * btrfs_free_reserved_bytes - update the block_group and space info counters | |
6691 | * @cache: The cache we are manipulating | |
6692 | * @num_bytes: The number of bytes in question | |
6693 | * @delalloc: The blocks are allocated for the delalloc write | |
6694 | * | |
6695 | * This is called by somebody who is freeing space that was never actually used | |
6696 | * on disk. For example if you reserve some space for a new leaf in transaction | |
6697 | * A and before transaction A commits you free that leaf, you call this with | |
6698 | * reserve set to 0 in order to clear the reservation. | |
6699 | */ | |
6700 | ||
6701 | static int btrfs_free_reserved_bytes(struct btrfs_block_group_cache *cache, | |
6702 | u64 num_bytes, int delalloc) | |
6703 | { | |
6704 | struct btrfs_space_info *space_info = cache->space_info; | |
6705 | int ret = 0; | |
6706 | ||
6707 | spin_lock(&space_info->lock); | |
6708 | spin_lock(&cache->lock); | |
6709 | if (cache->ro) | |
6710 | space_info->bytes_readonly += num_bytes; | |
6711 | cache->reserved -= num_bytes; | |
6712 | space_info->bytes_reserved -= num_bytes; | |
6713 | ||
6714 | if (delalloc) | |
6715 | cache->delalloc_bytes -= num_bytes; | |
6716 | spin_unlock(&cache->lock); | |
6717 | spin_unlock(&space_info->lock); | |
6718 | return ret; | |
6719 | } | |
8b74c03e | 6720 | void btrfs_prepare_extent_commit(struct btrfs_fs_info *fs_info) |
e8569813 | 6721 | { |
11833d66 YZ |
6722 | struct btrfs_caching_control *next; |
6723 | struct btrfs_caching_control *caching_ctl; | |
6724 | struct btrfs_block_group_cache *cache; | |
e8569813 | 6725 | |
9e351cc8 | 6726 | down_write(&fs_info->commit_root_sem); |
25179201 | 6727 | |
11833d66 YZ |
6728 | list_for_each_entry_safe(caching_ctl, next, |
6729 | &fs_info->caching_block_groups, list) { | |
6730 | cache = caching_ctl->block_group; | |
6731 | if (block_group_cache_done(cache)) { | |
6732 | cache->last_byte_to_unpin = (u64)-1; | |
6733 | list_del_init(&caching_ctl->list); | |
6734 | put_caching_control(caching_ctl); | |
e8569813 | 6735 | } else { |
11833d66 | 6736 | cache->last_byte_to_unpin = caching_ctl->progress; |
e8569813 | 6737 | } |
e8569813 | 6738 | } |
11833d66 YZ |
6739 | |
6740 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) | |
6741 | fs_info->pinned_extents = &fs_info->freed_extents[1]; | |
6742 | else | |
6743 | fs_info->pinned_extents = &fs_info->freed_extents[0]; | |
6744 | ||
9e351cc8 | 6745 | up_write(&fs_info->commit_root_sem); |
8929ecfa YZ |
6746 | |
6747 | update_global_block_rsv(fs_info); | |
e8569813 ZY |
6748 | } |
6749 | ||
c759c4e1 JB |
6750 | /* |
6751 | * Returns the free cluster for the given space info and sets empty_cluster to | |
6752 | * what it should be based on the mount options. | |
6753 | */ | |
6754 | static struct btrfs_free_cluster * | |
2ff7e61e JM |
6755 | fetch_cluster_info(struct btrfs_fs_info *fs_info, |
6756 | struct btrfs_space_info *space_info, u64 *empty_cluster) | |
c759c4e1 JB |
6757 | { |
6758 | struct btrfs_free_cluster *ret = NULL; | |
c759c4e1 JB |
6759 | |
6760 | *empty_cluster = 0; | |
6761 | if (btrfs_mixed_space_info(space_info)) | |
6762 | return ret; | |
6763 | ||
c759c4e1 | 6764 | if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) { |
0b246afa | 6765 | ret = &fs_info->meta_alloc_cluster; |
583b7231 HK |
6766 | if (btrfs_test_opt(fs_info, SSD)) |
6767 | *empty_cluster = SZ_2M; | |
6768 | else | |
ee22184b | 6769 | *empty_cluster = SZ_64K; |
583b7231 HK |
6770 | } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) && |
6771 | btrfs_test_opt(fs_info, SSD_SPREAD)) { | |
6772 | *empty_cluster = SZ_2M; | |
0b246afa | 6773 | ret = &fs_info->data_alloc_cluster; |
c759c4e1 JB |
6774 | } |
6775 | ||
6776 | return ret; | |
6777 | } | |
6778 | ||
2ff7e61e JM |
6779 | static int unpin_extent_range(struct btrfs_fs_info *fs_info, |
6780 | u64 start, u64 end, | |
678886bd | 6781 | const bool return_free_space) |
ccd467d6 | 6782 | { |
11833d66 | 6783 | struct btrfs_block_group_cache *cache = NULL; |
7b398f8e JB |
6784 | struct btrfs_space_info *space_info; |
6785 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
c759c4e1 | 6786 | struct btrfs_free_cluster *cluster = NULL; |
11833d66 | 6787 | u64 len; |
c759c4e1 JB |
6788 | u64 total_unpinned = 0; |
6789 | u64 empty_cluster = 0; | |
7b398f8e | 6790 | bool readonly; |
ccd467d6 | 6791 | |
11833d66 | 6792 | while (start <= end) { |
7b398f8e | 6793 | readonly = false; |
11833d66 YZ |
6794 | if (!cache || |
6795 | start >= cache->key.objectid + cache->key.offset) { | |
6796 | if (cache) | |
6797 | btrfs_put_block_group(cache); | |
c759c4e1 | 6798 | total_unpinned = 0; |
11833d66 | 6799 | cache = btrfs_lookup_block_group(fs_info, start); |
79787eaa | 6800 | BUG_ON(!cache); /* Logic error */ |
c759c4e1 | 6801 | |
2ff7e61e | 6802 | cluster = fetch_cluster_info(fs_info, |
c759c4e1 JB |
6803 | cache->space_info, |
6804 | &empty_cluster); | |
6805 | empty_cluster <<= 1; | |
11833d66 YZ |
6806 | } |
6807 | ||
6808 | len = cache->key.objectid + cache->key.offset - start; | |
6809 | len = min(len, end + 1 - start); | |
6810 | ||
6811 | if (start < cache->last_byte_to_unpin) { | |
6812 | len = min(len, cache->last_byte_to_unpin - start); | |
678886bd FM |
6813 | if (return_free_space) |
6814 | btrfs_add_free_space(cache, start, len); | |
11833d66 YZ |
6815 | } |
6816 | ||
f0486c68 | 6817 | start += len; |
c759c4e1 | 6818 | total_unpinned += len; |
7b398f8e | 6819 | space_info = cache->space_info; |
f0486c68 | 6820 | |
c759c4e1 JB |
6821 | /* |
6822 | * If this space cluster has been marked as fragmented and we've | |
6823 | * unpinned enough in this block group to potentially allow a | |
6824 | * cluster to be created inside of it go ahead and clear the | |
6825 | * fragmented check. | |
6826 | */ | |
6827 | if (cluster && cluster->fragmented && | |
6828 | total_unpinned > empty_cluster) { | |
6829 | spin_lock(&cluster->lock); | |
6830 | cluster->fragmented = 0; | |
6831 | spin_unlock(&cluster->lock); | |
6832 | } | |
6833 | ||
7b398f8e | 6834 | spin_lock(&space_info->lock); |
11833d66 YZ |
6835 | spin_lock(&cache->lock); |
6836 | cache->pinned -= len; | |
7b398f8e | 6837 | space_info->bytes_pinned -= len; |
c51e7bb1 JB |
6838 | |
6839 | trace_btrfs_space_reservation(fs_info, "pinned", | |
6840 | space_info->flags, len, 0); | |
4f4db217 | 6841 | space_info->max_extent_size = 0; |
d288db5d | 6842 | percpu_counter_add(&space_info->total_bytes_pinned, -len); |
7b398f8e JB |
6843 | if (cache->ro) { |
6844 | space_info->bytes_readonly += len; | |
6845 | readonly = true; | |
6846 | } | |
11833d66 | 6847 | spin_unlock(&cache->lock); |
957780eb JB |
6848 | if (!readonly && return_free_space && |
6849 | global_rsv->space_info == space_info) { | |
6850 | u64 to_add = len; | |
92ac58ec | 6851 | |
7b398f8e JB |
6852 | spin_lock(&global_rsv->lock); |
6853 | if (!global_rsv->full) { | |
957780eb JB |
6854 | to_add = min(len, global_rsv->size - |
6855 | global_rsv->reserved); | |
6856 | global_rsv->reserved += to_add; | |
6857 | space_info->bytes_may_use += to_add; | |
7b398f8e JB |
6858 | if (global_rsv->reserved >= global_rsv->size) |
6859 | global_rsv->full = 1; | |
957780eb JB |
6860 | trace_btrfs_space_reservation(fs_info, |
6861 | "space_info", | |
6862 | space_info->flags, | |
6863 | to_add, 1); | |
6864 | len -= to_add; | |
7b398f8e JB |
6865 | } |
6866 | spin_unlock(&global_rsv->lock); | |
957780eb JB |
6867 | /* Add to any tickets we may have */ |
6868 | if (len) | |
6869 | space_info_add_new_bytes(fs_info, space_info, | |
6870 | len); | |
7b398f8e JB |
6871 | } |
6872 | spin_unlock(&space_info->lock); | |
ccd467d6 | 6873 | } |
11833d66 YZ |
6874 | |
6875 | if (cache) | |
6876 | btrfs_put_block_group(cache); | |
ccd467d6 CM |
6877 | return 0; |
6878 | } | |
6879 | ||
6880 | int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans, | |
2ff7e61e | 6881 | struct btrfs_fs_info *fs_info) |
a28ec197 | 6882 | { |
e33e17ee JM |
6883 | struct btrfs_block_group_cache *block_group, *tmp; |
6884 | struct list_head *deleted_bgs; | |
11833d66 | 6885 | struct extent_io_tree *unpin; |
1a5bc167 CM |
6886 | u64 start; |
6887 | u64 end; | |
a28ec197 | 6888 | int ret; |
a28ec197 | 6889 | |
11833d66 YZ |
6890 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) |
6891 | unpin = &fs_info->freed_extents[1]; | |
6892 | else | |
6893 | unpin = &fs_info->freed_extents[0]; | |
6894 | ||
e33e17ee | 6895 | while (!trans->aborted) { |
d4b450cd | 6896 | mutex_lock(&fs_info->unused_bg_unpin_mutex); |
1a5bc167 | 6897 | ret = find_first_extent_bit(unpin, 0, &start, &end, |
e6138876 | 6898 | EXTENT_DIRTY, NULL); |
d4b450cd FM |
6899 | if (ret) { |
6900 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); | |
a28ec197 | 6901 | break; |
d4b450cd | 6902 | } |
1f3c79a2 | 6903 | |
0b246afa | 6904 | if (btrfs_test_opt(fs_info, DISCARD)) |
2ff7e61e | 6905 | ret = btrfs_discard_extent(fs_info, start, |
5378e607 | 6906 | end + 1 - start, NULL); |
1f3c79a2 | 6907 | |
af6f8f60 | 6908 | clear_extent_dirty(unpin, start, end); |
2ff7e61e | 6909 | unpin_extent_range(fs_info, start, end, true); |
d4b450cd | 6910 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
b9473439 | 6911 | cond_resched(); |
a28ec197 | 6912 | } |
817d52f8 | 6913 | |
e33e17ee JM |
6914 | /* |
6915 | * Transaction is finished. We don't need the lock anymore. We | |
6916 | * do need to clean up the block groups in case of a transaction | |
6917 | * abort. | |
6918 | */ | |
6919 | deleted_bgs = &trans->transaction->deleted_bgs; | |
6920 | list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) { | |
6921 | u64 trimmed = 0; | |
6922 | ||
6923 | ret = -EROFS; | |
6924 | if (!trans->aborted) | |
2ff7e61e | 6925 | ret = btrfs_discard_extent(fs_info, |
e33e17ee JM |
6926 | block_group->key.objectid, |
6927 | block_group->key.offset, | |
6928 | &trimmed); | |
6929 | ||
6930 | list_del_init(&block_group->bg_list); | |
6931 | btrfs_put_block_group_trimming(block_group); | |
6932 | btrfs_put_block_group(block_group); | |
6933 | ||
6934 | if (ret) { | |
6935 | const char *errstr = btrfs_decode_error(ret); | |
6936 | btrfs_warn(fs_info, | |
913e1535 | 6937 | "discard failed while removing blockgroup: errno=%d %s", |
e33e17ee JM |
6938 | ret, errstr); |
6939 | } | |
6940 | } | |
6941 | ||
e20d96d6 CM |
6942 | return 0; |
6943 | } | |
6944 | ||
5d4f98a2 | 6945 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
2ff7e61e | 6946 | struct btrfs_fs_info *info, |
c682f9b3 | 6947 | struct btrfs_delayed_ref_node *node, u64 parent, |
5d4f98a2 YZ |
6948 | u64 root_objectid, u64 owner_objectid, |
6949 | u64 owner_offset, int refs_to_drop, | |
c682f9b3 | 6950 | struct btrfs_delayed_extent_op *extent_op) |
a28ec197 | 6951 | { |
e2fa7227 | 6952 | struct btrfs_key key; |
5d4f98a2 | 6953 | struct btrfs_path *path; |
1261ec42 | 6954 | struct btrfs_root *extent_root = info->extent_root; |
5f39d397 | 6955 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
6956 | struct btrfs_extent_item *ei; |
6957 | struct btrfs_extent_inline_ref *iref; | |
a28ec197 | 6958 | int ret; |
5d4f98a2 | 6959 | int is_data; |
952fccac CM |
6960 | int extent_slot = 0; |
6961 | int found_extent = 0; | |
6962 | int num_to_del = 1; | |
5d4f98a2 YZ |
6963 | u32 item_size; |
6964 | u64 refs; | |
c682f9b3 QW |
6965 | u64 bytenr = node->bytenr; |
6966 | u64 num_bytes = node->num_bytes; | |
fcebe456 | 6967 | int last_ref = 0; |
0b246afa | 6968 | bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA); |
037e6390 | 6969 | |
5caf2a00 | 6970 | path = btrfs_alloc_path(); |
54aa1f4d CM |
6971 | if (!path) |
6972 | return -ENOMEM; | |
5f26f772 | 6973 | |
e4058b54 | 6974 | path->reada = READA_FORWARD; |
b9473439 | 6975 | path->leave_spinning = 1; |
5d4f98a2 YZ |
6976 | |
6977 | is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID; | |
6978 | BUG_ON(!is_data && refs_to_drop != 1); | |
6979 | ||
3173a18f JB |
6980 | if (is_data) |
6981 | skinny_metadata = 0; | |
6982 | ||
87bde3cd | 6983 | ret = lookup_extent_backref(trans, info, path, &iref, |
5d4f98a2 YZ |
6984 | bytenr, num_bytes, parent, |
6985 | root_objectid, owner_objectid, | |
6986 | owner_offset); | |
7bb86316 | 6987 | if (ret == 0) { |
952fccac | 6988 | extent_slot = path->slots[0]; |
5d4f98a2 YZ |
6989 | while (extent_slot >= 0) { |
6990 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
952fccac | 6991 | extent_slot); |
5d4f98a2 | 6992 | if (key.objectid != bytenr) |
952fccac | 6993 | break; |
5d4f98a2 YZ |
6994 | if (key.type == BTRFS_EXTENT_ITEM_KEY && |
6995 | key.offset == num_bytes) { | |
952fccac CM |
6996 | found_extent = 1; |
6997 | break; | |
6998 | } | |
3173a18f JB |
6999 | if (key.type == BTRFS_METADATA_ITEM_KEY && |
7000 | key.offset == owner_objectid) { | |
7001 | found_extent = 1; | |
7002 | break; | |
7003 | } | |
952fccac CM |
7004 | if (path->slots[0] - extent_slot > 5) |
7005 | break; | |
5d4f98a2 | 7006 | extent_slot--; |
952fccac | 7007 | } |
5d4f98a2 YZ |
7008 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
7009 | item_size = btrfs_item_size_nr(path->nodes[0], extent_slot); | |
7010 | if (found_extent && item_size < sizeof(*ei)) | |
7011 | found_extent = 0; | |
7012 | #endif | |
31840ae1 | 7013 | if (!found_extent) { |
5d4f98a2 | 7014 | BUG_ON(iref); |
87bde3cd JM |
7015 | ret = remove_extent_backref(trans, info, path, NULL, |
7016 | refs_to_drop, | |
fcebe456 | 7017 | is_data, &last_ref); |
005d6427 | 7018 | if (ret) { |
66642832 | 7019 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7020 | goto out; |
7021 | } | |
b3b4aa74 | 7022 | btrfs_release_path(path); |
b9473439 | 7023 | path->leave_spinning = 1; |
5d4f98a2 YZ |
7024 | |
7025 | key.objectid = bytenr; | |
7026 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
7027 | key.offset = num_bytes; | |
7028 | ||
3173a18f JB |
7029 | if (!is_data && skinny_metadata) { |
7030 | key.type = BTRFS_METADATA_ITEM_KEY; | |
7031 | key.offset = owner_objectid; | |
7032 | } | |
7033 | ||
31840ae1 ZY |
7034 | ret = btrfs_search_slot(trans, extent_root, |
7035 | &key, path, -1, 1); | |
3173a18f JB |
7036 | if (ret > 0 && skinny_metadata && path->slots[0]) { |
7037 | /* | |
7038 | * Couldn't find our skinny metadata item, | |
7039 | * see if we have ye olde extent item. | |
7040 | */ | |
7041 | path->slots[0]--; | |
7042 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
7043 | path->slots[0]); | |
7044 | if (key.objectid == bytenr && | |
7045 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
7046 | key.offset == num_bytes) | |
7047 | ret = 0; | |
7048 | } | |
7049 | ||
7050 | if (ret > 0 && skinny_metadata) { | |
7051 | skinny_metadata = false; | |
9ce49a0b | 7052 | key.objectid = bytenr; |
3173a18f JB |
7053 | key.type = BTRFS_EXTENT_ITEM_KEY; |
7054 | key.offset = num_bytes; | |
7055 | btrfs_release_path(path); | |
7056 | ret = btrfs_search_slot(trans, extent_root, | |
7057 | &key, path, -1, 1); | |
7058 | } | |
7059 | ||
f3465ca4 | 7060 | if (ret) { |
5d163e0e JM |
7061 | btrfs_err(info, |
7062 | "umm, got %d back from search, was looking for %llu", | |
7063 | ret, bytenr); | |
b783e62d | 7064 | if (ret > 0) |
a4f78750 | 7065 | btrfs_print_leaf(path->nodes[0]); |
f3465ca4 | 7066 | } |
005d6427 | 7067 | if (ret < 0) { |
66642832 | 7068 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7069 | goto out; |
7070 | } | |
31840ae1 ZY |
7071 | extent_slot = path->slots[0]; |
7072 | } | |
fae7f21c | 7073 | } else if (WARN_ON(ret == -ENOENT)) { |
a4f78750 | 7074 | btrfs_print_leaf(path->nodes[0]); |
c2cf52eb SK |
7075 | btrfs_err(info, |
7076 | "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu", | |
c1c9ff7c GU |
7077 | bytenr, parent, root_objectid, owner_objectid, |
7078 | owner_offset); | |
66642832 | 7079 | btrfs_abort_transaction(trans, ret); |
c4a050bb | 7080 | goto out; |
79787eaa | 7081 | } else { |
66642832 | 7082 | btrfs_abort_transaction(trans, ret); |
005d6427 | 7083 | goto out; |
7bb86316 | 7084 | } |
5f39d397 CM |
7085 | |
7086 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
7087 | item_size = btrfs_item_size_nr(leaf, extent_slot); |
7088 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
7089 | if (item_size < sizeof(*ei)) { | |
7090 | BUG_ON(found_extent || extent_slot != path->slots[0]); | |
87bde3cd JM |
7091 | ret = convert_extent_item_v0(trans, info, path, owner_objectid, |
7092 | 0); | |
005d6427 | 7093 | if (ret < 0) { |
66642832 | 7094 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7095 | goto out; |
7096 | } | |
5d4f98a2 | 7097 | |
b3b4aa74 | 7098 | btrfs_release_path(path); |
5d4f98a2 YZ |
7099 | path->leave_spinning = 1; |
7100 | ||
7101 | key.objectid = bytenr; | |
7102 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
7103 | key.offset = num_bytes; | |
7104 | ||
7105 | ret = btrfs_search_slot(trans, extent_root, &key, path, | |
7106 | -1, 1); | |
7107 | if (ret) { | |
5d163e0e JM |
7108 | btrfs_err(info, |
7109 | "umm, got %d back from search, was looking for %llu", | |
c1c9ff7c | 7110 | ret, bytenr); |
a4f78750 | 7111 | btrfs_print_leaf(path->nodes[0]); |
5d4f98a2 | 7112 | } |
005d6427 | 7113 | if (ret < 0) { |
66642832 | 7114 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7115 | goto out; |
7116 | } | |
7117 | ||
5d4f98a2 YZ |
7118 | extent_slot = path->slots[0]; |
7119 | leaf = path->nodes[0]; | |
7120 | item_size = btrfs_item_size_nr(leaf, extent_slot); | |
7121 | } | |
7122 | #endif | |
7123 | BUG_ON(item_size < sizeof(*ei)); | |
952fccac | 7124 | ei = btrfs_item_ptr(leaf, extent_slot, |
123abc88 | 7125 | struct btrfs_extent_item); |
3173a18f JB |
7126 | if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID && |
7127 | key.type == BTRFS_EXTENT_ITEM_KEY) { | |
5d4f98a2 YZ |
7128 | struct btrfs_tree_block_info *bi; |
7129 | BUG_ON(item_size < sizeof(*ei) + sizeof(*bi)); | |
7130 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
7131 | WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi)); | |
7132 | } | |
56bec294 | 7133 | |
5d4f98a2 | 7134 | refs = btrfs_extent_refs(leaf, ei); |
32b02538 | 7135 | if (refs < refs_to_drop) { |
5d163e0e JM |
7136 | btrfs_err(info, |
7137 | "trying to drop %d refs but we only have %Lu for bytenr %Lu", | |
7138 | refs_to_drop, refs, bytenr); | |
32b02538 | 7139 | ret = -EINVAL; |
66642832 | 7140 | btrfs_abort_transaction(trans, ret); |
32b02538 JB |
7141 | goto out; |
7142 | } | |
56bec294 | 7143 | refs -= refs_to_drop; |
5f39d397 | 7144 | |
5d4f98a2 YZ |
7145 | if (refs > 0) { |
7146 | if (extent_op) | |
7147 | __run_delayed_extent_op(extent_op, leaf, ei); | |
7148 | /* | |
7149 | * In the case of inline back ref, reference count will | |
7150 | * be updated by remove_extent_backref | |
952fccac | 7151 | */ |
5d4f98a2 YZ |
7152 | if (iref) { |
7153 | BUG_ON(!found_extent); | |
7154 | } else { | |
7155 | btrfs_set_extent_refs(leaf, ei, refs); | |
7156 | btrfs_mark_buffer_dirty(leaf); | |
7157 | } | |
7158 | if (found_extent) { | |
87bde3cd | 7159 | ret = remove_extent_backref(trans, info, path, |
5d4f98a2 | 7160 | iref, refs_to_drop, |
fcebe456 | 7161 | is_data, &last_ref); |
005d6427 | 7162 | if (ret) { |
66642832 | 7163 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7164 | goto out; |
7165 | } | |
952fccac | 7166 | } |
5d4f98a2 | 7167 | } else { |
5d4f98a2 YZ |
7168 | if (found_extent) { |
7169 | BUG_ON(is_data && refs_to_drop != | |
9ed0dea0 | 7170 | extent_data_ref_count(path, iref)); |
5d4f98a2 YZ |
7171 | if (iref) { |
7172 | BUG_ON(path->slots[0] != extent_slot); | |
7173 | } else { | |
7174 | BUG_ON(path->slots[0] != extent_slot + 1); | |
7175 | path->slots[0] = extent_slot; | |
7176 | num_to_del = 2; | |
7177 | } | |
78fae27e | 7178 | } |
b9473439 | 7179 | |
fcebe456 | 7180 | last_ref = 1; |
952fccac CM |
7181 | ret = btrfs_del_items(trans, extent_root, path, path->slots[0], |
7182 | num_to_del); | |
005d6427 | 7183 | if (ret) { |
66642832 | 7184 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7185 | goto out; |
7186 | } | |
b3b4aa74 | 7187 | btrfs_release_path(path); |
21af804c | 7188 | |
5d4f98a2 | 7189 | if (is_data) { |
5b4aacef | 7190 | ret = btrfs_del_csums(trans, info, bytenr, num_bytes); |
005d6427 | 7191 | if (ret) { |
66642832 | 7192 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7193 | goto out; |
7194 | } | |
459931ec CM |
7195 | } |
7196 | ||
0b246afa | 7197 | ret = add_to_free_space_tree(trans, info, bytenr, num_bytes); |
1e144fb8 | 7198 | if (ret) { |
66642832 | 7199 | btrfs_abort_transaction(trans, ret); |
1e144fb8 OS |
7200 | goto out; |
7201 | } | |
7202 | ||
0b246afa | 7203 | ret = update_block_group(trans, info, bytenr, num_bytes, 0); |
005d6427 | 7204 | if (ret) { |
66642832 | 7205 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7206 | goto out; |
7207 | } | |
a28ec197 | 7208 | } |
fcebe456 JB |
7209 | btrfs_release_path(path); |
7210 | ||
79787eaa | 7211 | out: |
5caf2a00 | 7212 | btrfs_free_path(path); |
a28ec197 CM |
7213 | return ret; |
7214 | } | |
7215 | ||
1887be66 | 7216 | /* |
f0486c68 | 7217 | * when we free an block, it is possible (and likely) that we free the last |
1887be66 CM |
7218 | * delayed ref for that extent as well. This searches the delayed ref tree for |
7219 | * a given extent, and if there are no other delayed refs to be processed, it | |
7220 | * removes it from the tree. | |
7221 | */ | |
7222 | static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans, | |
2ff7e61e | 7223 | u64 bytenr) |
1887be66 CM |
7224 | { |
7225 | struct btrfs_delayed_ref_head *head; | |
7226 | struct btrfs_delayed_ref_root *delayed_refs; | |
f0486c68 | 7227 | int ret = 0; |
1887be66 CM |
7228 | |
7229 | delayed_refs = &trans->transaction->delayed_refs; | |
7230 | spin_lock(&delayed_refs->lock); | |
f72ad18e | 7231 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
1887be66 | 7232 | if (!head) |
cf93da7b | 7233 | goto out_delayed_unlock; |
1887be66 | 7234 | |
d7df2c79 | 7235 | spin_lock(&head->lock); |
c6fc2454 | 7236 | if (!list_empty(&head->ref_list)) |
1887be66 CM |
7237 | goto out; |
7238 | ||
5d4f98a2 YZ |
7239 | if (head->extent_op) { |
7240 | if (!head->must_insert_reserved) | |
7241 | goto out; | |
78a6184a | 7242 | btrfs_free_delayed_extent_op(head->extent_op); |
5d4f98a2 YZ |
7243 | head->extent_op = NULL; |
7244 | } | |
7245 | ||
1887be66 CM |
7246 | /* |
7247 | * waiting for the lock here would deadlock. If someone else has it | |
7248 | * locked they are already in the process of dropping it anyway | |
7249 | */ | |
7250 | if (!mutex_trylock(&head->mutex)) | |
7251 | goto out; | |
7252 | ||
7253 | /* | |
7254 | * at this point we have a head with no other entries. Go | |
7255 | * ahead and process it. | |
7256 | */ | |
7257 | head->node.in_tree = 0; | |
c46effa6 | 7258 | rb_erase(&head->href_node, &delayed_refs->href_root); |
c3e69d58 | 7259 | |
d7df2c79 | 7260 | atomic_dec(&delayed_refs->num_entries); |
1887be66 CM |
7261 | |
7262 | /* | |
7263 | * we don't take a ref on the node because we're removing it from the | |
7264 | * tree, so we just steal the ref the tree was holding. | |
7265 | */ | |
c3e69d58 | 7266 | delayed_refs->num_heads--; |
d7df2c79 | 7267 | if (head->processing == 0) |
c3e69d58 | 7268 | delayed_refs->num_heads_ready--; |
d7df2c79 JB |
7269 | head->processing = 0; |
7270 | spin_unlock(&head->lock); | |
1887be66 CM |
7271 | spin_unlock(&delayed_refs->lock); |
7272 | ||
f0486c68 YZ |
7273 | BUG_ON(head->extent_op); |
7274 | if (head->must_insert_reserved) | |
7275 | ret = 1; | |
7276 | ||
7277 | mutex_unlock(&head->mutex); | |
1887be66 | 7278 | btrfs_put_delayed_ref(&head->node); |
f0486c68 | 7279 | return ret; |
1887be66 | 7280 | out: |
d7df2c79 | 7281 | spin_unlock(&head->lock); |
cf93da7b CM |
7282 | |
7283 | out_delayed_unlock: | |
1887be66 CM |
7284 | spin_unlock(&delayed_refs->lock); |
7285 | return 0; | |
7286 | } | |
7287 | ||
f0486c68 YZ |
7288 | void btrfs_free_tree_block(struct btrfs_trans_handle *trans, |
7289 | struct btrfs_root *root, | |
7290 | struct extent_buffer *buf, | |
5581a51a | 7291 | u64 parent, int last_ref) |
f0486c68 | 7292 | { |
0b246afa | 7293 | struct btrfs_fs_info *fs_info = root->fs_info; |
b150a4f1 | 7294 | int pin = 1; |
f0486c68 YZ |
7295 | int ret; |
7296 | ||
7297 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { | |
d7eae340 OS |
7298 | int old_ref_mod, new_ref_mod; |
7299 | ||
7be07912 OS |
7300 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, buf->start, |
7301 | buf->len, parent, | |
0b246afa JM |
7302 | root->root_key.objectid, |
7303 | btrfs_header_level(buf), | |
7be07912 | 7304 | BTRFS_DROP_DELAYED_REF, NULL, |
d7eae340 | 7305 | &old_ref_mod, &new_ref_mod); |
79787eaa | 7306 | BUG_ON(ret); /* -ENOMEM */ |
d7eae340 | 7307 | pin = old_ref_mod >= 0 && new_ref_mod < 0; |
f0486c68 YZ |
7308 | } |
7309 | ||
0a16c7d7 | 7310 | if (last_ref && btrfs_header_generation(buf) == trans->transid) { |
6219872d FM |
7311 | struct btrfs_block_group_cache *cache; |
7312 | ||
f0486c68 | 7313 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { |
2ff7e61e | 7314 | ret = check_ref_cleanup(trans, buf->start); |
f0486c68 | 7315 | if (!ret) |
37be25bc | 7316 | goto out; |
f0486c68 YZ |
7317 | } |
7318 | ||
4da8b76d | 7319 | pin = 0; |
0b246afa | 7320 | cache = btrfs_lookup_block_group(fs_info, buf->start); |
6219872d | 7321 | |
f0486c68 | 7322 | if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) { |
2ff7e61e JM |
7323 | pin_down_extent(fs_info, cache, buf->start, |
7324 | buf->len, 1); | |
6219872d | 7325 | btrfs_put_block_group(cache); |
37be25bc | 7326 | goto out; |
f0486c68 YZ |
7327 | } |
7328 | ||
7329 | WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)); | |
7330 | ||
7331 | btrfs_add_free_space(cache, buf->start, buf->len); | |
4824f1f4 | 7332 | btrfs_free_reserved_bytes(cache, buf->len, 0); |
6219872d | 7333 | btrfs_put_block_group(cache); |
71ff6437 | 7334 | trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len); |
f0486c68 YZ |
7335 | } |
7336 | out: | |
b150a4f1 | 7337 | if (pin) |
0b246afa | 7338 | add_pinned_bytes(fs_info, buf->len, btrfs_header_level(buf), |
b150a4f1 JB |
7339 | root->root_key.objectid); |
7340 | ||
0a16c7d7 OS |
7341 | if (last_ref) { |
7342 | /* | |
7343 | * Deleting the buffer, clear the corrupt flag since it doesn't | |
7344 | * matter anymore. | |
7345 | */ | |
7346 | clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags); | |
7347 | } | |
f0486c68 YZ |
7348 | } |
7349 | ||
79787eaa | 7350 | /* Can return -ENOMEM */ |
2ff7e61e JM |
7351 | int btrfs_free_extent(struct btrfs_trans_handle *trans, |
7352 | struct btrfs_fs_info *fs_info, | |
66d7e7f0 | 7353 | u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid, |
b06c4bf5 | 7354 | u64 owner, u64 offset) |
925baedd | 7355 | { |
d7eae340 | 7356 | int old_ref_mod, new_ref_mod; |
925baedd CM |
7357 | int ret; |
7358 | ||
f5ee5c9a | 7359 | if (btrfs_is_testing(fs_info)) |
faa2dbf0 | 7360 | return 0; |
fccb84c9 | 7361 | |
b150a4f1 | 7362 | |
56bec294 CM |
7363 | /* |
7364 | * tree log blocks never actually go into the extent allocation | |
7365 | * tree, just update pinning info and exit early. | |
56bec294 | 7366 | */ |
5d4f98a2 YZ |
7367 | if (root_objectid == BTRFS_TREE_LOG_OBJECTID) { |
7368 | WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID); | |
b9473439 | 7369 | /* unlocks the pinned mutex */ |
2ff7e61e | 7370 | btrfs_pin_extent(fs_info, bytenr, num_bytes, 1); |
d7eae340 | 7371 | old_ref_mod = new_ref_mod = 0; |
56bec294 | 7372 | ret = 0; |
5d4f98a2 | 7373 | } else if (owner < BTRFS_FIRST_FREE_OBJECTID) { |
66d7e7f0 | 7374 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr, |
7be07912 OS |
7375 | num_bytes, parent, |
7376 | root_objectid, (int)owner, | |
7377 | BTRFS_DROP_DELAYED_REF, NULL, | |
d7eae340 | 7378 | &old_ref_mod, &new_ref_mod); |
5d4f98a2 | 7379 | } else { |
66d7e7f0 | 7380 | ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr, |
7be07912 OS |
7381 | num_bytes, parent, |
7382 | root_objectid, owner, offset, | |
7383 | 0, BTRFS_DROP_DELAYED_REF, | |
d7eae340 | 7384 | &old_ref_mod, &new_ref_mod); |
56bec294 | 7385 | } |
d7eae340 OS |
7386 | |
7387 | if (ret == 0 && old_ref_mod >= 0 && new_ref_mod < 0) | |
7388 | add_pinned_bytes(fs_info, num_bytes, owner, root_objectid); | |
7389 | ||
925baedd CM |
7390 | return ret; |
7391 | } | |
7392 | ||
817d52f8 JB |
7393 | /* |
7394 | * when we wait for progress in the block group caching, its because | |
7395 | * our allocation attempt failed at least once. So, we must sleep | |
7396 | * and let some progress happen before we try again. | |
7397 | * | |
7398 | * This function will sleep at least once waiting for new free space to | |
7399 | * show up, and then it will check the block group free space numbers | |
7400 | * for our min num_bytes. Another option is to have it go ahead | |
7401 | * and look in the rbtree for a free extent of a given size, but this | |
7402 | * is a good start. | |
36cce922 JB |
7403 | * |
7404 | * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using | |
7405 | * any of the information in this block group. | |
817d52f8 | 7406 | */ |
36cce922 | 7407 | static noinline void |
817d52f8 JB |
7408 | wait_block_group_cache_progress(struct btrfs_block_group_cache *cache, |
7409 | u64 num_bytes) | |
7410 | { | |
11833d66 | 7411 | struct btrfs_caching_control *caching_ctl; |
817d52f8 | 7412 | |
11833d66 YZ |
7413 | caching_ctl = get_caching_control(cache); |
7414 | if (!caching_ctl) | |
36cce922 | 7415 | return; |
817d52f8 | 7416 | |
11833d66 | 7417 | wait_event(caching_ctl->wait, block_group_cache_done(cache) || |
34d52cb6 | 7418 | (cache->free_space_ctl->free_space >= num_bytes)); |
11833d66 YZ |
7419 | |
7420 | put_caching_control(caching_ctl); | |
11833d66 YZ |
7421 | } |
7422 | ||
7423 | static noinline int | |
7424 | wait_block_group_cache_done(struct btrfs_block_group_cache *cache) | |
7425 | { | |
7426 | struct btrfs_caching_control *caching_ctl; | |
36cce922 | 7427 | int ret = 0; |
11833d66 YZ |
7428 | |
7429 | caching_ctl = get_caching_control(cache); | |
7430 | if (!caching_ctl) | |
36cce922 | 7431 | return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0; |
11833d66 YZ |
7432 | |
7433 | wait_event(caching_ctl->wait, block_group_cache_done(cache)); | |
36cce922 JB |
7434 | if (cache->cached == BTRFS_CACHE_ERROR) |
7435 | ret = -EIO; | |
11833d66 | 7436 | put_caching_control(caching_ctl); |
36cce922 | 7437 | return ret; |
817d52f8 JB |
7438 | } |
7439 | ||
31e50229 | 7440 | int __get_raid_index(u64 flags) |
b742bb82 | 7441 | { |
7738a53a | 7442 | if (flags & BTRFS_BLOCK_GROUP_RAID10) |
e6ec716f | 7443 | return BTRFS_RAID_RAID10; |
7738a53a | 7444 | else if (flags & BTRFS_BLOCK_GROUP_RAID1) |
e6ec716f | 7445 | return BTRFS_RAID_RAID1; |
7738a53a | 7446 | else if (flags & BTRFS_BLOCK_GROUP_DUP) |
e6ec716f | 7447 | return BTRFS_RAID_DUP; |
7738a53a | 7448 | else if (flags & BTRFS_BLOCK_GROUP_RAID0) |
e6ec716f | 7449 | return BTRFS_RAID_RAID0; |
53b381b3 | 7450 | else if (flags & BTRFS_BLOCK_GROUP_RAID5) |
e942f883 | 7451 | return BTRFS_RAID_RAID5; |
53b381b3 | 7452 | else if (flags & BTRFS_BLOCK_GROUP_RAID6) |
e942f883 | 7453 | return BTRFS_RAID_RAID6; |
7738a53a | 7454 | |
e942f883 | 7455 | return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */ |
b742bb82 YZ |
7456 | } |
7457 | ||
6ab0a202 | 7458 | int get_block_group_index(struct btrfs_block_group_cache *cache) |
7738a53a | 7459 | { |
31e50229 | 7460 | return __get_raid_index(cache->flags); |
7738a53a ID |
7461 | } |
7462 | ||
6ab0a202 JM |
7463 | static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = { |
7464 | [BTRFS_RAID_RAID10] = "raid10", | |
7465 | [BTRFS_RAID_RAID1] = "raid1", | |
7466 | [BTRFS_RAID_DUP] = "dup", | |
7467 | [BTRFS_RAID_RAID0] = "raid0", | |
7468 | [BTRFS_RAID_SINGLE] = "single", | |
7469 | [BTRFS_RAID_RAID5] = "raid5", | |
7470 | [BTRFS_RAID_RAID6] = "raid6", | |
7471 | }; | |
7472 | ||
1b8e5df6 | 7473 | static const char *get_raid_name(enum btrfs_raid_types type) |
6ab0a202 JM |
7474 | { |
7475 | if (type >= BTRFS_NR_RAID_TYPES) | |
7476 | return NULL; | |
7477 | ||
7478 | return btrfs_raid_type_names[type]; | |
7479 | } | |
7480 | ||
817d52f8 | 7481 | enum btrfs_loop_type { |
285ff5af JB |
7482 | LOOP_CACHING_NOWAIT = 0, |
7483 | LOOP_CACHING_WAIT = 1, | |
7484 | LOOP_ALLOC_CHUNK = 2, | |
7485 | LOOP_NO_EMPTY_SIZE = 3, | |
817d52f8 JB |
7486 | }; |
7487 | ||
e570fd27 MX |
7488 | static inline void |
7489 | btrfs_lock_block_group(struct btrfs_block_group_cache *cache, | |
7490 | int delalloc) | |
7491 | { | |
7492 | if (delalloc) | |
7493 | down_read(&cache->data_rwsem); | |
7494 | } | |
7495 | ||
7496 | static inline void | |
7497 | btrfs_grab_block_group(struct btrfs_block_group_cache *cache, | |
7498 | int delalloc) | |
7499 | { | |
7500 | btrfs_get_block_group(cache); | |
7501 | if (delalloc) | |
7502 | down_read(&cache->data_rwsem); | |
7503 | } | |
7504 | ||
7505 | static struct btrfs_block_group_cache * | |
7506 | btrfs_lock_cluster(struct btrfs_block_group_cache *block_group, | |
7507 | struct btrfs_free_cluster *cluster, | |
7508 | int delalloc) | |
7509 | { | |
89771cc9 | 7510 | struct btrfs_block_group_cache *used_bg = NULL; |
6719afdc | 7511 | |
e570fd27 | 7512 | spin_lock(&cluster->refill_lock); |
6719afdc GU |
7513 | while (1) { |
7514 | used_bg = cluster->block_group; | |
7515 | if (!used_bg) | |
7516 | return NULL; | |
7517 | ||
7518 | if (used_bg == block_group) | |
e570fd27 MX |
7519 | return used_bg; |
7520 | ||
6719afdc | 7521 | btrfs_get_block_group(used_bg); |
e570fd27 | 7522 | |
6719afdc GU |
7523 | if (!delalloc) |
7524 | return used_bg; | |
e570fd27 | 7525 | |
6719afdc GU |
7526 | if (down_read_trylock(&used_bg->data_rwsem)) |
7527 | return used_bg; | |
e570fd27 | 7528 | |
6719afdc | 7529 | spin_unlock(&cluster->refill_lock); |
e570fd27 | 7530 | |
e321f8a8 LB |
7531 | /* We should only have one-level nested. */ |
7532 | down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING); | |
e570fd27 | 7533 | |
6719afdc GU |
7534 | spin_lock(&cluster->refill_lock); |
7535 | if (used_bg == cluster->block_group) | |
7536 | return used_bg; | |
e570fd27 | 7537 | |
6719afdc GU |
7538 | up_read(&used_bg->data_rwsem); |
7539 | btrfs_put_block_group(used_bg); | |
7540 | } | |
e570fd27 MX |
7541 | } |
7542 | ||
7543 | static inline void | |
7544 | btrfs_release_block_group(struct btrfs_block_group_cache *cache, | |
7545 | int delalloc) | |
7546 | { | |
7547 | if (delalloc) | |
7548 | up_read(&cache->data_rwsem); | |
7549 | btrfs_put_block_group(cache); | |
7550 | } | |
7551 | ||
fec577fb CM |
7552 | /* |
7553 | * walks the btree of allocated extents and find a hole of a given size. | |
7554 | * The key ins is changed to record the hole: | |
a4820398 | 7555 | * ins->objectid == start position |
62e2749e | 7556 | * ins->flags = BTRFS_EXTENT_ITEM_KEY |
a4820398 | 7557 | * ins->offset == the size of the hole. |
fec577fb | 7558 | * Any available blocks before search_start are skipped. |
a4820398 MX |
7559 | * |
7560 | * If there is no suitable free space, we will record the max size of | |
7561 | * the free space extent currently. | |
fec577fb | 7562 | */ |
87bde3cd | 7563 | static noinline int find_free_extent(struct btrfs_fs_info *fs_info, |
18513091 WX |
7564 | u64 ram_bytes, u64 num_bytes, u64 empty_size, |
7565 | u64 hint_byte, struct btrfs_key *ins, | |
7566 | u64 flags, int delalloc) | |
fec577fb | 7567 | { |
80eb234a | 7568 | int ret = 0; |
0b246afa | 7569 | struct btrfs_root *root = fs_info->extent_root; |
fa9c0d79 | 7570 | struct btrfs_free_cluster *last_ptr = NULL; |
80eb234a | 7571 | struct btrfs_block_group_cache *block_group = NULL; |
81c9ad23 | 7572 | u64 search_start = 0; |
a4820398 | 7573 | u64 max_extent_size = 0; |
7509d4f9 | 7574 | u64 max_free_space = 0; |
c759c4e1 | 7575 | u64 empty_cluster = 0; |
80eb234a | 7576 | struct btrfs_space_info *space_info; |
fa9c0d79 | 7577 | int loop = 0; |
b6919a58 | 7578 | int index = __get_raid_index(flags); |
0a24325e | 7579 | bool failed_cluster_refill = false; |
1cdda9b8 | 7580 | bool failed_alloc = false; |
67377734 | 7581 | bool use_cluster = true; |
60d2adbb | 7582 | bool have_caching_bg = false; |
13a0db5a | 7583 | bool orig_have_caching_bg = false; |
a5e681d9 | 7584 | bool full_search = false; |
fec577fb | 7585 | |
0b246afa | 7586 | WARN_ON(num_bytes < fs_info->sectorsize); |
962a298f | 7587 | ins->type = BTRFS_EXTENT_ITEM_KEY; |
80eb234a JB |
7588 | ins->objectid = 0; |
7589 | ins->offset = 0; | |
b1a4d965 | 7590 | |
71ff6437 | 7591 | trace_find_free_extent(fs_info, num_bytes, empty_size, flags); |
3f7de037 | 7592 | |
0b246afa | 7593 | space_info = __find_space_info(fs_info, flags); |
1b1d1f66 | 7594 | if (!space_info) { |
0b246afa | 7595 | btrfs_err(fs_info, "No space info for %llu", flags); |
1b1d1f66 JB |
7596 | return -ENOSPC; |
7597 | } | |
2552d17e | 7598 | |
67377734 | 7599 | /* |
4f4db217 JB |
7600 | * If our free space is heavily fragmented we may not be able to make |
7601 | * big contiguous allocations, so instead of doing the expensive search | |
7602 | * for free space, simply return ENOSPC with our max_extent_size so we | |
7603 | * can go ahead and search for a more manageable chunk. | |
7604 | * | |
7605 | * If our max_extent_size is large enough for our allocation simply | |
7606 | * disable clustering since we will likely not be able to find enough | |
7607 | * space to create a cluster and induce latency trying. | |
67377734 | 7608 | */ |
4f4db217 JB |
7609 | if (unlikely(space_info->max_extent_size)) { |
7610 | spin_lock(&space_info->lock); | |
7611 | if (space_info->max_extent_size && | |
7612 | num_bytes > space_info->max_extent_size) { | |
7613 | ins->offset = space_info->max_extent_size; | |
7614 | spin_unlock(&space_info->lock); | |
7615 | return -ENOSPC; | |
7616 | } else if (space_info->max_extent_size) { | |
7617 | use_cluster = false; | |
7618 | } | |
7619 | spin_unlock(&space_info->lock); | |
fa9c0d79 | 7620 | } |
0f9dd46c | 7621 | |
2ff7e61e | 7622 | last_ptr = fetch_cluster_info(fs_info, space_info, &empty_cluster); |
239b14b3 | 7623 | if (last_ptr) { |
fa9c0d79 CM |
7624 | spin_lock(&last_ptr->lock); |
7625 | if (last_ptr->block_group) | |
7626 | hint_byte = last_ptr->window_start; | |
c759c4e1 JB |
7627 | if (last_ptr->fragmented) { |
7628 | /* | |
7629 | * We still set window_start so we can keep track of the | |
7630 | * last place we found an allocation to try and save | |
7631 | * some time. | |
7632 | */ | |
7633 | hint_byte = last_ptr->window_start; | |
7634 | use_cluster = false; | |
7635 | } | |
fa9c0d79 | 7636 | spin_unlock(&last_ptr->lock); |
239b14b3 | 7637 | } |
fa9c0d79 | 7638 | |
2ff7e61e | 7639 | search_start = max(search_start, first_logical_byte(fs_info, 0)); |
239b14b3 | 7640 | search_start = max(search_start, hint_byte); |
2552d17e | 7641 | if (search_start == hint_byte) { |
0b246afa | 7642 | block_group = btrfs_lookup_block_group(fs_info, search_start); |
817d52f8 JB |
7643 | /* |
7644 | * we don't want to use the block group if it doesn't match our | |
7645 | * allocation bits, or if its not cached. | |
ccf0e725 JB |
7646 | * |
7647 | * However if we are re-searching with an ideal block group | |
7648 | * picked out then we don't care that the block group is cached. | |
817d52f8 | 7649 | */ |
b6919a58 | 7650 | if (block_group && block_group_bits(block_group, flags) && |
285ff5af | 7651 | block_group->cached != BTRFS_CACHE_NO) { |
2552d17e | 7652 | down_read(&space_info->groups_sem); |
44fb5511 CM |
7653 | if (list_empty(&block_group->list) || |
7654 | block_group->ro) { | |
7655 | /* | |
7656 | * someone is removing this block group, | |
7657 | * we can't jump into the have_block_group | |
7658 | * target because our list pointers are not | |
7659 | * valid | |
7660 | */ | |
7661 | btrfs_put_block_group(block_group); | |
7662 | up_read(&space_info->groups_sem); | |
ccf0e725 | 7663 | } else { |
b742bb82 | 7664 | index = get_block_group_index(block_group); |
e570fd27 | 7665 | btrfs_lock_block_group(block_group, delalloc); |
44fb5511 | 7666 | goto have_block_group; |
ccf0e725 | 7667 | } |
2552d17e | 7668 | } else if (block_group) { |
fa9c0d79 | 7669 | btrfs_put_block_group(block_group); |
2552d17e | 7670 | } |
42e70e7a | 7671 | } |
2552d17e | 7672 | search: |
60d2adbb | 7673 | have_caching_bg = false; |
a5e681d9 JB |
7674 | if (index == 0 || index == __get_raid_index(flags)) |
7675 | full_search = true; | |
80eb234a | 7676 | down_read(&space_info->groups_sem); |
b742bb82 YZ |
7677 | list_for_each_entry(block_group, &space_info->block_groups[index], |
7678 | list) { | |
6226cb0a | 7679 | u64 offset; |
817d52f8 | 7680 | int cached; |
8a1413a2 | 7681 | |
14443937 JM |
7682 | /* If the block group is read-only, we can skip it entirely. */ |
7683 | if (unlikely(block_group->ro)) | |
7684 | continue; | |
7685 | ||
e570fd27 | 7686 | btrfs_grab_block_group(block_group, delalloc); |
2552d17e | 7687 | search_start = block_group->key.objectid; |
42e70e7a | 7688 | |
83a50de9 CM |
7689 | /* |
7690 | * this can happen if we end up cycling through all the | |
7691 | * raid types, but we want to make sure we only allocate | |
7692 | * for the proper type. | |
7693 | */ | |
b6919a58 | 7694 | if (!block_group_bits(block_group, flags)) { |
83a50de9 CM |
7695 | u64 extra = BTRFS_BLOCK_GROUP_DUP | |
7696 | BTRFS_BLOCK_GROUP_RAID1 | | |
53b381b3 DW |
7697 | BTRFS_BLOCK_GROUP_RAID5 | |
7698 | BTRFS_BLOCK_GROUP_RAID6 | | |
83a50de9 CM |
7699 | BTRFS_BLOCK_GROUP_RAID10; |
7700 | ||
7701 | /* | |
7702 | * if they asked for extra copies and this block group | |
7703 | * doesn't provide them, bail. This does allow us to | |
7704 | * fill raid0 from raid1. | |
7705 | */ | |
b6919a58 | 7706 | if ((flags & extra) && !(block_group->flags & extra)) |
83a50de9 | 7707 | goto loop; |
cd107319 QW |
7708 | |
7709 | /* | |
7710 | * This block group has different flags than we want. | |
7711 | * It's possible that we have MIXED_GROUP flag but no | |
7712 | * block group is mixed. Just skip such block group. | |
7713 | */ | |
7714 | btrfs_release_block_group(block_group, delalloc); | |
7715 | continue; | |
83a50de9 CM |
7716 | } |
7717 | ||
2552d17e | 7718 | have_block_group: |
291c7d2f JB |
7719 | cached = block_group_cache_done(block_group); |
7720 | if (unlikely(!cached)) { | |
a5e681d9 | 7721 | have_caching_bg = true; |
f6373bf3 | 7722 | ret = cache_block_group(block_group, 0); |
1d4284bd CM |
7723 | BUG_ON(ret < 0); |
7724 | ret = 0; | |
817d52f8 JB |
7725 | } |
7726 | ||
36cce922 JB |
7727 | if (unlikely(block_group->cached == BTRFS_CACHE_ERROR)) |
7728 | goto loop; | |
0f9dd46c | 7729 | |
0a24325e | 7730 | /* |
062c05c4 AO |
7731 | * Ok we want to try and use the cluster allocator, so |
7732 | * lets look there | |
0a24325e | 7733 | */ |
c759c4e1 | 7734 | if (last_ptr && use_cluster) { |
215a63d1 | 7735 | struct btrfs_block_group_cache *used_block_group; |
8de972b4 | 7736 | unsigned long aligned_cluster; |
fa9c0d79 CM |
7737 | /* |
7738 | * the refill lock keeps out other | |
7739 | * people trying to start a new cluster | |
7740 | */ | |
e570fd27 MX |
7741 | used_block_group = btrfs_lock_cluster(block_group, |
7742 | last_ptr, | |
7743 | delalloc); | |
7744 | if (!used_block_group) | |
44fb5511 | 7745 | goto refill_cluster; |
274bd4fb | 7746 | |
e570fd27 MX |
7747 | if (used_block_group != block_group && |
7748 | (used_block_group->ro || | |
7749 | !block_group_bits(used_block_group, flags))) | |
7750 | goto release_cluster; | |
44fb5511 | 7751 | |
274bd4fb | 7752 | offset = btrfs_alloc_from_cluster(used_block_group, |
a4820398 MX |
7753 | last_ptr, |
7754 | num_bytes, | |
7755 | used_block_group->key.objectid, | |
7756 | &max_extent_size); | |
fa9c0d79 CM |
7757 | if (offset) { |
7758 | /* we have a block, we're done */ | |
7759 | spin_unlock(&last_ptr->refill_lock); | |
71ff6437 | 7760 | trace_btrfs_reserve_extent_cluster(fs_info, |
89d4346a MX |
7761 | used_block_group, |
7762 | search_start, num_bytes); | |
215a63d1 | 7763 | if (used_block_group != block_group) { |
e570fd27 MX |
7764 | btrfs_release_block_group(block_group, |
7765 | delalloc); | |
215a63d1 MX |
7766 | block_group = used_block_group; |
7767 | } | |
fa9c0d79 CM |
7768 | goto checks; |
7769 | } | |
7770 | ||
274bd4fb | 7771 | WARN_ON(last_ptr->block_group != used_block_group); |
e570fd27 | 7772 | release_cluster: |
062c05c4 AO |
7773 | /* If we are on LOOP_NO_EMPTY_SIZE, we can't |
7774 | * set up a new clusters, so lets just skip it | |
7775 | * and let the allocator find whatever block | |
7776 | * it can find. If we reach this point, we | |
7777 | * will have tried the cluster allocator | |
7778 | * plenty of times and not have found | |
7779 | * anything, so we are likely way too | |
7780 | * fragmented for the clustering stuff to find | |
a5f6f719 AO |
7781 | * anything. |
7782 | * | |
7783 | * However, if the cluster is taken from the | |
7784 | * current block group, release the cluster | |
7785 | * first, so that we stand a better chance of | |
7786 | * succeeding in the unclustered | |
7787 | * allocation. */ | |
7788 | if (loop >= LOOP_NO_EMPTY_SIZE && | |
e570fd27 | 7789 | used_block_group != block_group) { |
062c05c4 | 7790 | spin_unlock(&last_ptr->refill_lock); |
e570fd27 MX |
7791 | btrfs_release_block_group(used_block_group, |
7792 | delalloc); | |
062c05c4 AO |
7793 | goto unclustered_alloc; |
7794 | } | |
7795 | ||
fa9c0d79 CM |
7796 | /* |
7797 | * this cluster didn't work out, free it and | |
7798 | * start over | |
7799 | */ | |
7800 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
7801 | ||
e570fd27 MX |
7802 | if (used_block_group != block_group) |
7803 | btrfs_release_block_group(used_block_group, | |
7804 | delalloc); | |
7805 | refill_cluster: | |
a5f6f719 AO |
7806 | if (loop >= LOOP_NO_EMPTY_SIZE) { |
7807 | spin_unlock(&last_ptr->refill_lock); | |
7808 | goto unclustered_alloc; | |
7809 | } | |
7810 | ||
8de972b4 CM |
7811 | aligned_cluster = max_t(unsigned long, |
7812 | empty_cluster + empty_size, | |
7813 | block_group->full_stripe_len); | |
7814 | ||
fa9c0d79 | 7815 | /* allocate a cluster in this block group */ |
2ff7e61e | 7816 | ret = btrfs_find_space_cluster(fs_info, block_group, |
00361589 JB |
7817 | last_ptr, search_start, |
7818 | num_bytes, | |
7819 | aligned_cluster); | |
fa9c0d79 CM |
7820 | if (ret == 0) { |
7821 | /* | |
7822 | * now pull our allocation out of this | |
7823 | * cluster | |
7824 | */ | |
7825 | offset = btrfs_alloc_from_cluster(block_group, | |
a4820398 MX |
7826 | last_ptr, |
7827 | num_bytes, | |
7828 | search_start, | |
7829 | &max_extent_size); | |
fa9c0d79 CM |
7830 | if (offset) { |
7831 | /* we found one, proceed */ | |
7832 | spin_unlock(&last_ptr->refill_lock); | |
71ff6437 | 7833 | trace_btrfs_reserve_extent_cluster(fs_info, |
3f7de037 JB |
7834 | block_group, search_start, |
7835 | num_bytes); | |
fa9c0d79 CM |
7836 | goto checks; |
7837 | } | |
0a24325e JB |
7838 | } else if (!cached && loop > LOOP_CACHING_NOWAIT |
7839 | && !failed_cluster_refill) { | |
817d52f8 JB |
7840 | spin_unlock(&last_ptr->refill_lock); |
7841 | ||
0a24325e | 7842 | failed_cluster_refill = true; |
817d52f8 JB |
7843 | wait_block_group_cache_progress(block_group, |
7844 | num_bytes + empty_cluster + empty_size); | |
7845 | goto have_block_group; | |
fa9c0d79 | 7846 | } |
817d52f8 | 7847 | |
fa9c0d79 CM |
7848 | /* |
7849 | * at this point we either didn't find a cluster | |
7850 | * or we weren't able to allocate a block from our | |
7851 | * cluster. Free the cluster we've been trying | |
7852 | * to use, and go to the next block group | |
7853 | */ | |
0a24325e | 7854 | btrfs_return_cluster_to_free_space(NULL, last_ptr); |
fa9c0d79 | 7855 | spin_unlock(&last_ptr->refill_lock); |
0a24325e | 7856 | goto loop; |
fa9c0d79 CM |
7857 | } |
7858 | ||
062c05c4 | 7859 | unclustered_alloc: |
c759c4e1 JB |
7860 | /* |
7861 | * We are doing an unclustered alloc, set the fragmented flag so | |
7862 | * we don't bother trying to setup a cluster again until we get | |
7863 | * more space. | |
7864 | */ | |
7865 | if (unlikely(last_ptr)) { | |
7866 | spin_lock(&last_ptr->lock); | |
7867 | last_ptr->fragmented = 1; | |
7868 | spin_unlock(&last_ptr->lock); | |
7869 | } | |
0c9b36e0 LB |
7870 | if (cached) { |
7871 | struct btrfs_free_space_ctl *ctl = | |
7872 | block_group->free_space_ctl; | |
7873 | ||
7874 | spin_lock(&ctl->tree_lock); | |
7875 | if (ctl->free_space < | |
7876 | num_bytes + empty_cluster + empty_size) { | |
7509d4f9 JB |
7877 | max_free_space = max(max_free_space, |
7878 | ctl->free_space); | |
0c9b36e0 LB |
7879 | spin_unlock(&ctl->tree_lock); |
7880 | goto loop; | |
7881 | } | |
7882 | spin_unlock(&ctl->tree_lock); | |
a5f6f719 | 7883 | } |
a5f6f719 | 7884 | |
6226cb0a | 7885 | offset = btrfs_find_space_for_alloc(block_group, search_start, |
a4820398 MX |
7886 | num_bytes, empty_size, |
7887 | &max_extent_size); | |
1cdda9b8 JB |
7888 | /* |
7889 | * If we didn't find a chunk, and we haven't failed on this | |
7890 | * block group before, and this block group is in the middle of | |
7891 | * caching and we are ok with waiting, then go ahead and wait | |
7892 | * for progress to be made, and set failed_alloc to true. | |
7893 | * | |
7894 | * If failed_alloc is true then we've already waited on this | |
7895 | * block group once and should move on to the next block group. | |
7896 | */ | |
7897 | if (!offset && !failed_alloc && !cached && | |
7898 | loop > LOOP_CACHING_NOWAIT) { | |
817d52f8 | 7899 | wait_block_group_cache_progress(block_group, |
1cdda9b8 JB |
7900 | num_bytes + empty_size); |
7901 | failed_alloc = true; | |
817d52f8 | 7902 | goto have_block_group; |
1cdda9b8 JB |
7903 | } else if (!offset) { |
7904 | goto loop; | |
817d52f8 | 7905 | } |
fa9c0d79 | 7906 | checks: |
0b246afa | 7907 | search_start = ALIGN(offset, fs_info->stripesize); |
25179201 | 7908 | |
2552d17e JB |
7909 | /* move on to the next group */ |
7910 | if (search_start + num_bytes > | |
215a63d1 MX |
7911 | block_group->key.objectid + block_group->key.offset) { |
7912 | btrfs_add_free_space(block_group, offset, num_bytes); | |
2552d17e | 7913 | goto loop; |
6226cb0a | 7914 | } |
f5a31e16 | 7915 | |
f0486c68 | 7916 | if (offset < search_start) |
215a63d1 | 7917 | btrfs_add_free_space(block_group, offset, |
f0486c68 YZ |
7918 | search_start - offset); |
7919 | BUG_ON(offset > search_start); | |
2552d17e | 7920 | |
18513091 WX |
7921 | ret = btrfs_add_reserved_bytes(block_group, ram_bytes, |
7922 | num_bytes, delalloc); | |
f0486c68 | 7923 | if (ret == -EAGAIN) { |
215a63d1 | 7924 | btrfs_add_free_space(block_group, offset, num_bytes); |
2552d17e | 7925 | goto loop; |
0f9dd46c | 7926 | } |
9cfa3e34 | 7927 | btrfs_inc_block_group_reservations(block_group); |
0b86a832 | 7928 | |
f0486c68 | 7929 | /* we are all good, lets return */ |
2552d17e JB |
7930 | ins->objectid = search_start; |
7931 | ins->offset = num_bytes; | |
d2fb3437 | 7932 | |
71ff6437 | 7933 | trace_btrfs_reserve_extent(fs_info, block_group, |
3f7de037 | 7934 | search_start, num_bytes); |
e570fd27 | 7935 | btrfs_release_block_group(block_group, delalloc); |
2552d17e JB |
7936 | break; |
7937 | loop: | |
0a24325e | 7938 | failed_cluster_refill = false; |
1cdda9b8 | 7939 | failed_alloc = false; |
b742bb82 | 7940 | BUG_ON(index != get_block_group_index(block_group)); |
e570fd27 | 7941 | btrfs_release_block_group(block_group, delalloc); |
14443937 | 7942 | cond_resched(); |
2552d17e JB |
7943 | } |
7944 | up_read(&space_info->groups_sem); | |
7945 | ||
13a0db5a | 7946 | if ((loop == LOOP_CACHING_NOWAIT) && have_caching_bg |
7947 | && !orig_have_caching_bg) | |
7948 | orig_have_caching_bg = true; | |
7949 | ||
60d2adbb MX |
7950 | if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg) |
7951 | goto search; | |
7952 | ||
b742bb82 YZ |
7953 | if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES) |
7954 | goto search; | |
7955 | ||
285ff5af | 7956 | /* |
ccf0e725 JB |
7957 | * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking |
7958 | * caching kthreads as we move along | |
817d52f8 JB |
7959 | * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching |
7960 | * LOOP_ALLOC_CHUNK, force a chunk allocation and try again | |
7961 | * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try | |
7962 | * again | |
fa9c0d79 | 7963 | */ |
723bda20 | 7964 | if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) { |
b742bb82 | 7965 | index = 0; |
a5e681d9 JB |
7966 | if (loop == LOOP_CACHING_NOWAIT) { |
7967 | /* | |
7968 | * We want to skip the LOOP_CACHING_WAIT step if we | |
01327610 | 7969 | * don't have any uncached bgs and we've already done a |
a5e681d9 JB |
7970 | * full search through. |
7971 | */ | |
13a0db5a | 7972 | if (orig_have_caching_bg || !full_search) |
a5e681d9 JB |
7973 | loop = LOOP_CACHING_WAIT; |
7974 | else | |
7975 | loop = LOOP_ALLOC_CHUNK; | |
7976 | } else { | |
7977 | loop++; | |
7978 | } | |
7979 | ||
817d52f8 | 7980 | if (loop == LOOP_ALLOC_CHUNK) { |
00361589 | 7981 | struct btrfs_trans_handle *trans; |
f017f15f WS |
7982 | int exist = 0; |
7983 | ||
7984 | trans = current->journal_info; | |
7985 | if (trans) | |
7986 | exist = 1; | |
7987 | else | |
7988 | trans = btrfs_join_transaction(root); | |
00361589 | 7989 | |
00361589 JB |
7990 | if (IS_ERR(trans)) { |
7991 | ret = PTR_ERR(trans); | |
7992 | goto out; | |
7993 | } | |
7994 | ||
2ff7e61e | 7995 | ret = do_chunk_alloc(trans, fs_info, flags, |
ea658bad | 7996 | CHUNK_ALLOC_FORCE); |
a5e681d9 JB |
7997 | |
7998 | /* | |
7999 | * If we can't allocate a new chunk we've already looped | |
8000 | * through at least once, move on to the NO_EMPTY_SIZE | |
8001 | * case. | |
8002 | */ | |
8003 | if (ret == -ENOSPC) | |
8004 | loop = LOOP_NO_EMPTY_SIZE; | |
8005 | ||
ea658bad JB |
8006 | /* |
8007 | * Do not bail out on ENOSPC since we | |
8008 | * can do more things. | |
8009 | */ | |
00361589 | 8010 | if (ret < 0 && ret != -ENOSPC) |
66642832 | 8011 | btrfs_abort_transaction(trans, ret); |
00361589 JB |
8012 | else |
8013 | ret = 0; | |
f017f15f | 8014 | if (!exist) |
3a45bb20 | 8015 | btrfs_end_transaction(trans); |
00361589 | 8016 | if (ret) |
ea658bad | 8017 | goto out; |
2552d17e JB |
8018 | } |
8019 | ||
723bda20 | 8020 | if (loop == LOOP_NO_EMPTY_SIZE) { |
a5e681d9 JB |
8021 | /* |
8022 | * Don't loop again if we already have no empty_size and | |
8023 | * no empty_cluster. | |
8024 | */ | |
8025 | if (empty_size == 0 && | |
8026 | empty_cluster == 0) { | |
8027 | ret = -ENOSPC; | |
8028 | goto out; | |
8029 | } | |
723bda20 JB |
8030 | empty_size = 0; |
8031 | empty_cluster = 0; | |
fa9c0d79 | 8032 | } |
723bda20 JB |
8033 | |
8034 | goto search; | |
2552d17e JB |
8035 | } else if (!ins->objectid) { |
8036 | ret = -ENOSPC; | |
d82a6f1d | 8037 | } else if (ins->objectid) { |
c759c4e1 JB |
8038 | if (!use_cluster && last_ptr) { |
8039 | spin_lock(&last_ptr->lock); | |
8040 | last_ptr->window_start = ins->objectid; | |
8041 | spin_unlock(&last_ptr->lock); | |
8042 | } | |
80eb234a | 8043 | ret = 0; |
be744175 | 8044 | } |
79787eaa | 8045 | out: |
4f4db217 | 8046 | if (ret == -ENOSPC) { |
7509d4f9 JB |
8047 | if (!max_extent_size) |
8048 | max_extent_size = max_free_space; | |
4f4db217 JB |
8049 | spin_lock(&space_info->lock); |
8050 | space_info->max_extent_size = max_extent_size; | |
8051 | spin_unlock(&space_info->lock); | |
a4820398 | 8052 | ins->offset = max_extent_size; |
4f4db217 | 8053 | } |
0f70abe2 | 8054 | return ret; |
fec577fb | 8055 | } |
ec44a35c | 8056 | |
ab8d0fc4 JM |
8057 | static void dump_space_info(struct btrfs_fs_info *fs_info, |
8058 | struct btrfs_space_info *info, u64 bytes, | |
9ed74f2d | 8059 | int dump_block_groups) |
0f9dd46c JB |
8060 | { |
8061 | struct btrfs_block_group_cache *cache; | |
b742bb82 | 8062 | int index = 0; |
0f9dd46c | 8063 | |
9ed74f2d | 8064 | spin_lock(&info->lock); |
ab8d0fc4 JM |
8065 | btrfs_info(fs_info, "space_info %llu has %llu free, is %sfull", |
8066 | info->flags, | |
4136135b LB |
8067 | info->total_bytes - btrfs_space_info_used(info, true), |
8068 | info->full ? "" : "not "); | |
ab8d0fc4 JM |
8069 | btrfs_info(fs_info, |
8070 | "space_info total=%llu, used=%llu, pinned=%llu, reserved=%llu, may_use=%llu, readonly=%llu", | |
8071 | info->total_bytes, info->bytes_used, info->bytes_pinned, | |
8072 | info->bytes_reserved, info->bytes_may_use, | |
8073 | info->bytes_readonly); | |
9ed74f2d JB |
8074 | spin_unlock(&info->lock); |
8075 | ||
8076 | if (!dump_block_groups) | |
8077 | return; | |
0f9dd46c | 8078 | |
80eb234a | 8079 | down_read(&info->groups_sem); |
b742bb82 YZ |
8080 | again: |
8081 | list_for_each_entry(cache, &info->block_groups[index], list) { | |
0f9dd46c | 8082 | spin_lock(&cache->lock); |
ab8d0fc4 JM |
8083 | btrfs_info(fs_info, |
8084 | "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s", | |
8085 | cache->key.objectid, cache->key.offset, | |
8086 | btrfs_block_group_used(&cache->item), cache->pinned, | |
8087 | cache->reserved, cache->ro ? "[readonly]" : ""); | |
0f9dd46c JB |
8088 | btrfs_dump_free_space(cache, bytes); |
8089 | spin_unlock(&cache->lock); | |
8090 | } | |
b742bb82 YZ |
8091 | if (++index < BTRFS_NR_RAID_TYPES) |
8092 | goto again; | |
80eb234a | 8093 | up_read(&info->groups_sem); |
0f9dd46c | 8094 | } |
e8569813 | 8095 | |
18513091 | 8096 | int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes, |
11833d66 YZ |
8097 | u64 num_bytes, u64 min_alloc_size, |
8098 | u64 empty_size, u64 hint_byte, | |
e570fd27 | 8099 | struct btrfs_key *ins, int is_data, int delalloc) |
fec577fb | 8100 | { |
ab8d0fc4 | 8101 | struct btrfs_fs_info *fs_info = root->fs_info; |
36af4e07 | 8102 | bool final_tried = num_bytes == min_alloc_size; |
b6919a58 | 8103 | u64 flags; |
fec577fb | 8104 | int ret; |
925baedd | 8105 | |
1b86826d | 8106 | flags = get_alloc_profile_by_root(root, is_data); |
98d20f67 | 8107 | again: |
0b246afa | 8108 | WARN_ON(num_bytes < fs_info->sectorsize); |
87bde3cd | 8109 | ret = find_free_extent(fs_info, ram_bytes, num_bytes, empty_size, |
18513091 | 8110 | hint_byte, ins, flags, delalloc); |
9cfa3e34 | 8111 | if (!ret && !is_data) { |
ab8d0fc4 | 8112 | btrfs_dec_block_group_reservations(fs_info, ins->objectid); |
9cfa3e34 | 8113 | } else if (ret == -ENOSPC) { |
a4820398 MX |
8114 | if (!final_tried && ins->offset) { |
8115 | num_bytes = min(num_bytes >> 1, ins->offset); | |
da17066c | 8116 | num_bytes = round_down(num_bytes, |
0b246afa | 8117 | fs_info->sectorsize); |
9e622d6b | 8118 | num_bytes = max(num_bytes, min_alloc_size); |
18513091 | 8119 | ram_bytes = num_bytes; |
9e622d6b MX |
8120 | if (num_bytes == min_alloc_size) |
8121 | final_tried = true; | |
8122 | goto again; | |
ab8d0fc4 | 8123 | } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
9e622d6b MX |
8124 | struct btrfs_space_info *sinfo; |
8125 | ||
ab8d0fc4 | 8126 | sinfo = __find_space_info(fs_info, flags); |
0b246afa | 8127 | btrfs_err(fs_info, |
5d163e0e JM |
8128 | "allocation failed flags %llu, wanted %llu", |
8129 | flags, num_bytes); | |
53804280 | 8130 | if (sinfo) |
ab8d0fc4 | 8131 | dump_space_info(fs_info, sinfo, num_bytes, 1); |
9e622d6b | 8132 | } |
925baedd | 8133 | } |
0f9dd46c JB |
8134 | |
8135 | return ret; | |
e6dcd2dc CM |
8136 | } |
8137 | ||
2ff7e61e | 8138 | static int __btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info, |
e570fd27 MX |
8139 | u64 start, u64 len, |
8140 | int pin, int delalloc) | |
65b51a00 | 8141 | { |
0f9dd46c | 8142 | struct btrfs_block_group_cache *cache; |
1f3c79a2 | 8143 | int ret = 0; |
0f9dd46c | 8144 | |
0b246afa | 8145 | cache = btrfs_lookup_block_group(fs_info, start); |
0f9dd46c | 8146 | if (!cache) { |
0b246afa JM |
8147 | btrfs_err(fs_info, "Unable to find block group for %llu", |
8148 | start); | |
0f9dd46c JB |
8149 | return -ENOSPC; |
8150 | } | |
1f3c79a2 | 8151 | |
e688b725 | 8152 | if (pin) |
2ff7e61e | 8153 | pin_down_extent(fs_info, cache, start, len, 1); |
e688b725 | 8154 | else { |
0b246afa | 8155 | if (btrfs_test_opt(fs_info, DISCARD)) |
2ff7e61e | 8156 | ret = btrfs_discard_extent(fs_info, start, len, NULL); |
e688b725 | 8157 | btrfs_add_free_space(cache, start, len); |
4824f1f4 | 8158 | btrfs_free_reserved_bytes(cache, len, delalloc); |
71ff6437 | 8159 | trace_btrfs_reserved_extent_free(fs_info, start, len); |
e688b725 | 8160 | } |
31193213 | 8161 | |
fa9c0d79 | 8162 | btrfs_put_block_group(cache); |
e6dcd2dc CM |
8163 | return ret; |
8164 | } | |
8165 | ||
2ff7e61e | 8166 | int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info, |
e570fd27 | 8167 | u64 start, u64 len, int delalloc) |
e688b725 | 8168 | { |
2ff7e61e | 8169 | return __btrfs_free_reserved_extent(fs_info, start, len, 0, delalloc); |
e688b725 CM |
8170 | } |
8171 | ||
2ff7e61e | 8172 | int btrfs_free_and_pin_reserved_extent(struct btrfs_fs_info *fs_info, |
e688b725 CM |
8173 | u64 start, u64 len) |
8174 | { | |
2ff7e61e | 8175 | return __btrfs_free_reserved_extent(fs_info, start, len, 1, 0); |
e688b725 CM |
8176 | } |
8177 | ||
5d4f98a2 | 8178 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, |
2ff7e61e | 8179 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
8180 | u64 parent, u64 root_objectid, |
8181 | u64 flags, u64 owner, u64 offset, | |
8182 | struct btrfs_key *ins, int ref_mod) | |
e6dcd2dc CM |
8183 | { |
8184 | int ret; | |
e6dcd2dc | 8185 | struct btrfs_extent_item *extent_item; |
5d4f98a2 | 8186 | struct btrfs_extent_inline_ref *iref; |
e6dcd2dc | 8187 | struct btrfs_path *path; |
5d4f98a2 YZ |
8188 | struct extent_buffer *leaf; |
8189 | int type; | |
8190 | u32 size; | |
26b8003f | 8191 | |
5d4f98a2 YZ |
8192 | if (parent > 0) |
8193 | type = BTRFS_SHARED_DATA_REF_KEY; | |
8194 | else | |
8195 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
58176a96 | 8196 | |
5d4f98a2 | 8197 | size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type); |
7bb86316 CM |
8198 | |
8199 | path = btrfs_alloc_path(); | |
db5b493a TI |
8200 | if (!path) |
8201 | return -ENOMEM; | |
47e4bb98 | 8202 | |
b9473439 | 8203 | path->leave_spinning = 1; |
5d4f98a2 YZ |
8204 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, |
8205 | ins, size); | |
79787eaa JM |
8206 | if (ret) { |
8207 | btrfs_free_path(path); | |
8208 | return ret; | |
8209 | } | |
0f9dd46c | 8210 | |
5d4f98a2 YZ |
8211 | leaf = path->nodes[0]; |
8212 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
47e4bb98 | 8213 | struct btrfs_extent_item); |
5d4f98a2 YZ |
8214 | btrfs_set_extent_refs(leaf, extent_item, ref_mod); |
8215 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
8216 | btrfs_set_extent_flags(leaf, extent_item, | |
8217 | flags | BTRFS_EXTENT_FLAG_DATA); | |
8218 | ||
8219 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
8220 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
8221 | if (parent > 0) { | |
8222 | struct btrfs_shared_data_ref *ref; | |
8223 | ref = (struct btrfs_shared_data_ref *)(iref + 1); | |
8224 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
8225 | btrfs_set_shared_data_ref_count(leaf, ref, ref_mod); | |
8226 | } else { | |
8227 | struct btrfs_extent_data_ref *ref; | |
8228 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
8229 | btrfs_set_extent_data_ref_root(leaf, ref, root_objectid); | |
8230 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
8231 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
8232 | btrfs_set_extent_data_ref_count(leaf, ref, ref_mod); | |
8233 | } | |
47e4bb98 CM |
8234 | |
8235 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
7bb86316 | 8236 | btrfs_free_path(path); |
f510cfec | 8237 | |
1e144fb8 OS |
8238 | ret = remove_from_free_space_tree(trans, fs_info, ins->objectid, |
8239 | ins->offset); | |
8240 | if (ret) | |
8241 | return ret; | |
8242 | ||
6202df69 | 8243 | ret = update_block_group(trans, fs_info, ins->objectid, ins->offset, 1); |
79787eaa | 8244 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb | 8245 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
c1c9ff7c | 8246 | ins->objectid, ins->offset); |
f5947066 CM |
8247 | BUG(); |
8248 | } | |
71ff6437 | 8249 | trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid, ins->offset); |
e6dcd2dc CM |
8250 | return ret; |
8251 | } | |
8252 | ||
5d4f98a2 | 8253 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, |
2ff7e61e | 8254 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
8255 | u64 parent, u64 root_objectid, |
8256 | u64 flags, struct btrfs_disk_key *key, | |
b06c4bf5 | 8257 | int level, struct btrfs_key *ins) |
e6dcd2dc CM |
8258 | { |
8259 | int ret; | |
5d4f98a2 YZ |
8260 | struct btrfs_extent_item *extent_item; |
8261 | struct btrfs_tree_block_info *block_info; | |
8262 | struct btrfs_extent_inline_ref *iref; | |
8263 | struct btrfs_path *path; | |
8264 | struct extent_buffer *leaf; | |
3173a18f | 8265 | u32 size = sizeof(*extent_item) + sizeof(*iref); |
fcebe456 | 8266 | u64 num_bytes = ins->offset; |
0b246afa | 8267 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); |
3173a18f JB |
8268 | |
8269 | if (!skinny_metadata) | |
8270 | size += sizeof(*block_info); | |
1c2308f8 | 8271 | |
5d4f98a2 | 8272 | path = btrfs_alloc_path(); |
857cc2fc | 8273 | if (!path) { |
2ff7e61e | 8274 | btrfs_free_and_pin_reserved_extent(fs_info, ins->objectid, |
0b246afa | 8275 | fs_info->nodesize); |
d8926bb3 | 8276 | return -ENOMEM; |
857cc2fc | 8277 | } |
56bec294 | 8278 | |
5d4f98a2 YZ |
8279 | path->leave_spinning = 1; |
8280 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, | |
8281 | ins, size); | |
79787eaa | 8282 | if (ret) { |
dd825259 | 8283 | btrfs_free_path(path); |
2ff7e61e | 8284 | btrfs_free_and_pin_reserved_extent(fs_info, ins->objectid, |
0b246afa | 8285 | fs_info->nodesize); |
79787eaa JM |
8286 | return ret; |
8287 | } | |
5d4f98a2 YZ |
8288 | |
8289 | leaf = path->nodes[0]; | |
8290 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
8291 | struct btrfs_extent_item); | |
8292 | btrfs_set_extent_refs(leaf, extent_item, 1); | |
8293 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
8294 | btrfs_set_extent_flags(leaf, extent_item, | |
8295 | flags | BTRFS_EXTENT_FLAG_TREE_BLOCK); | |
5d4f98a2 | 8296 | |
3173a18f JB |
8297 | if (skinny_metadata) { |
8298 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
0b246afa | 8299 | num_bytes = fs_info->nodesize; |
3173a18f JB |
8300 | } else { |
8301 | block_info = (struct btrfs_tree_block_info *)(extent_item + 1); | |
8302 | btrfs_set_tree_block_key(leaf, block_info, key); | |
8303 | btrfs_set_tree_block_level(leaf, block_info, level); | |
8304 | iref = (struct btrfs_extent_inline_ref *)(block_info + 1); | |
8305 | } | |
5d4f98a2 | 8306 | |
5d4f98a2 YZ |
8307 | if (parent > 0) { |
8308 | BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); | |
8309 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
8310 | BTRFS_SHARED_BLOCK_REF_KEY); | |
8311 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
8312 | } else { | |
8313 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
8314 | BTRFS_TREE_BLOCK_REF_KEY); | |
8315 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
8316 | } | |
8317 | ||
8318 | btrfs_mark_buffer_dirty(leaf); | |
8319 | btrfs_free_path(path); | |
8320 | ||
1e144fb8 OS |
8321 | ret = remove_from_free_space_tree(trans, fs_info, ins->objectid, |
8322 | num_bytes); | |
8323 | if (ret) | |
8324 | return ret; | |
8325 | ||
6202df69 JM |
8326 | ret = update_block_group(trans, fs_info, ins->objectid, |
8327 | fs_info->nodesize, 1); | |
79787eaa | 8328 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb | 8329 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
c1c9ff7c | 8330 | ins->objectid, ins->offset); |
5d4f98a2 YZ |
8331 | BUG(); |
8332 | } | |
0be5dc67 | 8333 | |
71ff6437 | 8334 | trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid, |
0b246afa | 8335 | fs_info->nodesize); |
5d4f98a2 YZ |
8336 | return ret; |
8337 | } | |
8338 | ||
8339 | int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
5d4f98a2 | 8340 | u64 root_objectid, u64 owner, |
5846a3c2 QW |
8341 | u64 offset, u64 ram_bytes, |
8342 | struct btrfs_key *ins) | |
5d4f98a2 | 8343 | { |
2ff7e61e | 8344 | struct btrfs_fs_info *fs_info = trans->fs_info; |
5d4f98a2 YZ |
8345 | int ret; |
8346 | ||
8347 | BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
8348 | ||
0b246afa | 8349 | ret = btrfs_add_delayed_data_ref(fs_info, trans, ins->objectid, |
7be07912 OS |
8350 | ins->offset, 0, root_objectid, owner, |
8351 | offset, ram_bytes, | |
8352 | BTRFS_ADD_DELAYED_EXTENT, NULL, NULL); | |
e6dcd2dc CM |
8353 | return ret; |
8354 | } | |
e02119d5 CM |
8355 | |
8356 | /* | |
8357 | * this is used by the tree logging recovery code. It records that | |
8358 | * an extent has been allocated and makes sure to clear the free | |
8359 | * space cache bits as well | |
8360 | */ | |
5d4f98a2 | 8361 | int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans, |
2ff7e61e | 8362 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
8363 | u64 root_objectid, u64 owner, u64 offset, |
8364 | struct btrfs_key *ins) | |
e02119d5 CM |
8365 | { |
8366 | int ret; | |
8367 | struct btrfs_block_group_cache *block_group; | |
ed7a6948 | 8368 | struct btrfs_space_info *space_info; |
11833d66 | 8369 | |
8c2a1a30 JB |
8370 | /* |
8371 | * Mixed block groups will exclude before processing the log so we only | |
01327610 | 8372 | * need to do the exclude dance if this fs isn't mixed. |
8c2a1a30 | 8373 | */ |
0b246afa | 8374 | if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) { |
2ff7e61e JM |
8375 | ret = __exclude_logged_extent(fs_info, ins->objectid, |
8376 | ins->offset); | |
b50c6e25 | 8377 | if (ret) |
8c2a1a30 | 8378 | return ret; |
11833d66 YZ |
8379 | } |
8380 | ||
0b246afa | 8381 | block_group = btrfs_lookup_block_group(fs_info, ins->objectid); |
8c2a1a30 JB |
8382 | if (!block_group) |
8383 | return -EINVAL; | |
8384 | ||
ed7a6948 WX |
8385 | space_info = block_group->space_info; |
8386 | spin_lock(&space_info->lock); | |
8387 | spin_lock(&block_group->lock); | |
8388 | space_info->bytes_reserved += ins->offset; | |
8389 | block_group->reserved += ins->offset; | |
8390 | spin_unlock(&block_group->lock); | |
8391 | spin_unlock(&space_info->lock); | |
8392 | ||
2ff7e61e | 8393 | ret = alloc_reserved_file_extent(trans, fs_info, 0, root_objectid, |
5d4f98a2 | 8394 | 0, owner, offset, ins, 1); |
b50c6e25 | 8395 | btrfs_put_block_group(block_group); |
e02119d5 CM |
8396 | return ret; |
8397 | } | |
8398 | ||
48a3b636 ES |
8399 | static struct extent_buffer * |
8400 | btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
fe864576 | 8401 | u64 bytenr, int level) |
65b51a00 | 8402 | { |
0b246afa | 8403 | struct btrfs_fs_info *fs_info = root->fs_info; |
65b51a00 CM |
8404 | struct extent_buffer *buf; |
8405 | ||
2ff7e61e | 8406 | buf = btrfs_find_create_tree_block(fs_info, bytenr); |
c871b0f2 LB |
8407 | if (IS_ERR(buf)) |
8408 | return buf; | |
8409 | ||
d36ac4a3 QW |
8410 | /* |
8411 | * Extra safety check in case the extent tree is corrupted and extent | |
8412 | * allocator chooses to use a tree block which is already used and | |
8413 | * locked. | |
8414 | */ | |
8415 | if (buf->lock_owner == current->pid) { | |
8416 | btrfs_err_rl(fs_info, | |
8417 | "tree block %llu owner %llu already locked by pid=%d, extent tree corruption detected", | |
8418 | buf->start, btrfs_header_owner(buf), current->pid); | |
8419 | free_extent_buffer(buf); | |
8420 | return ERR_PTR(-EUCLEAN); | |
8421 | } | |
8422 | ||
65b51a00 | 8423 | btrfs_set_header_generation(buf, trans->transid); |
85d4e461 | 8424 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level); |
65b51a00 | 8425 | btrfs_tree_lock(buf); |
7c302b49 | 8426 | clean_tree_block(fs_info, buf); |
3083ee2e | 8427 | clear_bit(EXTENT_BUFFER_STALE, &buf->bflags); |
b4ce94de CM |
8428 | |
8429 | btrfs_set_lock_blocking(buf); | |
4db8c528 | 8430 | set_extent_buffer_uptodate(buf); |
b4ce94de | 8431 | |
d0c803c4 | 8432 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { |
656f30db | 8433 | buf->log_index = root->log_transid % 2; |
8cef4e16 YZ |
8434 | /* |
8435 | * we allow two log transactions at a time, use different | |
8436 | * EXENT bit to differentiate dirty pages. | |
8437 | */ | |
656f30db | 8438 | if (buf->log_index == 0) |
8cef4e16 YZ |
8439 | set_extent_dirty(&root->dirty_log_pages, buf->start, |
8440 | buf->start + buf->len - 1, GFP_NOFS); | |
8441 | else | |
8442 | set_extent_new(&root->dirty_log_pages, buf->start, | |
3744dbeb | 8443 | buf->start + buf->len - 1); |
d0c803c4 | 8444 | } else { |
656f30db | 8445 | buf->log_index = -1; |
d0c803c4 | 8446 | set_extent_dirty(&trans->transaction->dirty_pages, buf->start, |
65b51a00 | 8447 | buf->start + buf->len - 1, GFP_NOFS); |
d0c803c4 | 8448 | } |
64c12921 | 8449 | trans->dirty = true; |
b4ce94de | 8450 | /* this returns a buffer locked for blocking */ |
65b51a00 CM |
8451 | return buf; |
8452 | } | |
8453 | ||
f0486c68 YZ |
8454 | static struct btrfs_block_rsv * |
8455 | use_block_rsv(struct btrfs_trans_handle *trans, | |
8456 | struct btrfs_root *root, u32 blocksize) | |
8457 | { | |
0b246afa | 8458 | struct btrfs_fs_info *fs_info = root->fs_info; |
f0486c68 | 8459 | struct btrfs_block_rsv *block_rsv; |
0b246afa | 8460 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; |
f0486c68 | 8461 | int ret; |
d88033db | 8462 | bool global_updated = false; |
f0486c68 YZ |
8463 | |
8464 | block_rsv = get_block_rsv(trans, root); | |
8465 | ||
b586b323 MX |
8466 | if (unlikely(block_rsv->size == 0)) |
8467 | goto try_reserve; | |
d88033db | 8468 | again: |
f0486c68 YZ |
8469 | ret = block_rsv_use_bytes(block_rsv, blocksize); |
8470 | if (!ret) | |
8471 | return block_rsv; | |
8472 | ||
b586b323 MX |
8473 | if (block_rsv->failfast) |
8474 | return ERR_PTR(ret); | |
8475 | ||
d88033db MX |
8476 | if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) { |
8477 | global_updated = true; | |
0b246afa | 8478 | update_global_block_rsv(fs_info); |
d88033db MX |
8479 | goto again; |
8480 | } | |
8481 | ||
0b246afa | 8482 | if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
b586b323 MX |
8483 | static DEFINE_RATELIMIT_STATE(_rs, |
8484 | DEFAULT_RATELIMIT_INTERVAL * 10, | |
8485 | /*DEFAULT_RATELIMIT_BURST*/ 1); | |
8486 | if (__ratelimit(&_rs)) | |
8487 | WARN(1, KERN_DEBUG | |
efe120a0 | 8488 | "BTRFS: block rsv returned %d\n", ret); |
b586b323 MX |
8489 | } |
8490 | try_reserve: | |
8491 | ret = reserve_metadata_bytes(root, block_rsv, blocksize, | |
8492 | BTRFS_RESERVE_NO_FLUSH); | |
8493 | if (!ret) | |
8494 | return block_rsv; | |
8495 | /* | |
8496 | * If we couldn't reserve metadata bytes try and use some from | |
5881cfc9 MX |
8497 | * the global reserve if its space type is the same as the global |
8498 | * reservation. | |
b586b323 | 8499 | */ |
5881cfc9 MX |
8500 | if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL && |
8501 | block_rsv->space_info == global_rsv->space_info) { | |
b586b323 MX |
8502 | ret = block_rsv_use_bytes(global_rsv, blocksize); |
8503 | if (!ret) | |
8504 | return global_rsv; | |
8505 | } | |
8506 | return ERR_PTR(ret); | |
f0486c68 YZ |
8507 | } |
8508 | ||
8c2a3ca2 JB |
8509 | static void unuse_block_rsv(struct btrfs_fs_info *fs_info, |
8510 | struct btrfs_block_rsv *block_rsv, u32 blocksize) | |
f0486c68 YZ |
8511 | { |
8512 | block_rsv_add_bytes(block_rsv, blocksize, 0); | |
8c2a3ca2 | 8513 | block_rsv_release_bytes(fs_info, block_rsv, NULL, 0); |
f0486c68 YZ |
8514 | } |
8515 | ||
fec577fb | 8516 | /* |
f0486c68 | 8517 | * finds a free extent and does all the dirty work required for allocation |
67b7859e | 8518 | * returns the tree buffer or an ERR_PTR on error. |
fec577fb | 8519 | */ |
4d75f8a9 | 8520 | struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans, |
310712b2 OS |
8521 | struct btrfs_root *root, |
8522 | u64 parent, u64 root_objectid, | |
8523 | const struct btrfs_disk_key *key, | |
8524 | int level, u64 hint, | |
8525 | u64 empty_size) | |
fec577fb | 8526 | { |
0b246afa | 8527 | struct btrfs_fs_info *fs_info = root->fs_info; |
e2fa7227 | 8528 | struct btrfs_key ins; |
f0486c68 | 8529 | struct btrfs_block_rsv *block_rsv; |
5f39d397 | 8530 | struct extent_buffer *buf; |
67b7859e | 8531 | struct btrfs_delayed_extent_op *extent_op; |
f0486c68 YZ |
8532 | u64 flags = 0; |
8533 | int ret; | |
0b246afa JM |
8534 | u32 blocksize = fs_info->nodesize; |
8535 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); | |
fec577fb | 8536 | |
05653ef3 | 8537 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
0b246afa | 8538 | if (btrfs_is_testing(fs_info)) { |
faa2dbf0 | 8539 | buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr, |
fe864576 | 8540 | level); |
faa2dbf0 JB |
8541 | if (!IS_ERR(buf)) |
8542 | root->alloc_bytenr += blocksize; | |
8543 | return buf; | |
8544 | } | |
05653ef3 | 8545 | #endif |
fccb84c9 | 8546 | |
f0486c68 YZ |
8547 | block_rsv = use_block_rsv(trans, root, blocksize); |
8548 | if (IS_ERR(block_rsv)) | |
8549 | return ERR_CAST(block_rsv); | |
8550 | ||
18513091 | 8551 | ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize, |
e570fd27 | 8552 | empty_size, hint, &ins, 0, 0); |
67b7859e OS |
8553 | if (ret) |
8554 | goto out_unuse; | |
55c69072 | 8555 | |
fe864576 | 8556 | buf = btrfs_init_new_buffer(trans, root, ins.objectid, level); |
67b7859e OS |
8557 | if (IS_ERR(buf)) { |
8558 | ret = PTR_ERR(buf); | |
8559 | goto out_free_reserved; | |
8560 | } | |
f0486c68 YZ |
8561 | |
8562 | if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
8563 | if (parent == 0) | |
8564 | parent = ins.objectid; | |
8565 | flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
8566 | } else | |
8567 | BUG_ON(parent > 0); | |
8568 | ||
8569 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) { | |
78a6184a | 8570 | extent_op = btrfs_alloc_delayed_extent_op(); |
67b7859e OS |
8571 | if (!extent_op) { |
8572 | ret = -ENOMEM; | |
8573 | goto out_free_buf; | |
8574 | } | |
f0486c68 YZ |
8575 | if (key) |
8576 | memcpy(&extent_op->key, key, sizeof(extent_op->key)); | |
8577 | else | |
8578 | memset(&extent_op->key, 0, sizeof(extent_op->key)); | |
8579 | extent_op->flags_to_set = flags; | |
35b3ad50 DS |
8580 | extent_op->update_key = skinny_metadata ? false : true; |
8581 | extent_op->update_flags = true; | |
8582 | extent_op->is_data = false; | |
b1c79e09 | 8583 | extent_op->level = level; |
f0486c68 | 8584 | |
7be07912 OS |
8585 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, ins.objectid, |
8586 | ins.offset, parent, | |
8587 | root_objectid, level, | |
67b7859e | 8588 | BTRFS_ADD_DELAYED_EXTENT, |
7be07912 | 8589 | extent_op, NULL, NULL); |
67b7859e OS |
8590 | if (ret) |
8591 | goto out_free_delayed; | |
f0486c68 | 8592 | } |
fec577fb | 8593 | return buf; |
67b7859e OS |
8594 | |
8595 | out_free_delayed: | |
8596 | btrfs_free_delayed_extent_op(extent_op); | |
8597 | out_free_buf: | |
8598 | free_extent_buffer(buf); | |
8599 | out_free_reserved: | |
2ff7e61e | 8600 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0); |
67b7859e | 8601 | out_unuse: |
0b246afa | 8602 | unuse_block_rsv(fs_info, block_rsv, blocksize); |
67b7859e | 8603 | return ERR_PTR(ret); |
fec577fb | 8604 | } |
a28ec197 | 8605 | |
2c47e605 YZ |
8606 | struct walk_control { |
8607 | u64 refs[BTRFS_MAX_LEVEL]; | |
8608 | u64 flags[BTRFS_MAX_LEVEL]; | |
8609 | struct btrfs_key update_progress; | |
8610 | int stage; | |
8611 | int level; | |
8612 | int shared_level; | |
8613 | int update_ref; | |
8614 | int keep_locks; | |
1c4850e2 YZ |
8615 | int reada_slot; |
8616 | int reada_count; | |
66d7e7f0 | 8617 | int for_reloc; |
2c47e605 YZ |
8618 | }; |
8619 | ||
8620 | #define DROP_REFERENCE 1 | |
8621 | #define UPDATE_BACKREF 2 | |
8622 | ||
1c4850e2 YZ |
8623 | static noinline void reada_walk_down(struct btrfs_trans_handle *trans, |
8624 | struct btrfs_root *root, | |
8625 | struct walk_control *wc, | |
8626 | struct btrfs_path *path) | |
6407bf6d | 8627 | { |
0b246afa | 8628 | struct btrfs_fs_info *fs_info = root->fs_info; |
1c4850e2 YZ |
8629 | u64 bytenr; |
8630 | u64 generation; | |
8631 | u64 refs; | |
94fcca9f | 8632 | u64 flags; |
5d4f98a2 | 8633 | u32 nritems; |
1c4850e2 YZ |
8634 | struct btrfs_key key; |
8635 | struct extent_buffer *eb; | |
6407bf6d | 8636 | int ret; |
1c4850e2 YZ |
8637 | int slot; |
8638 | int nread = 0; | |
6407bf6d | 8639 | |
1c4850e2 YZ |
8640 | if (path->slots[wc->level] < wc->reada_slot) { |
8641 | wc->reada_count = wc->reada_count * 2 / 3; | |
8642 | wc->reada_count = max(wc->reada_count, 2); | |
8643 | } else { | |
8644 | wc->reada_count = wc->reada_count * 3 / 2; | |
8645 | wc->reada_count = min_t(int, wc->reada_count, | |
0b246afa | 8646 | BTRFS_NODEPTRS_PER_BLOCK(fs_info)); |
1c4850e2 | 8647 | } |
7bb86316 | 8648 | |
1c4850e2 YZ |
8649 | eb = path->nodes[wc->level]; |
8650 | nritems = btrfs_header_nritems(eb); | |
bd56b302 | 8651 | |
1c4850e2 YZ |
8652 | for (slot = path->slots[wc->level]; slot < nritems; slot++) { |
8653 | if (nread >= wc->reada_count) | |
8654 | break; | |
bd56b302 | 8655 | |
2dd3e67b | 8656 | cond_resched(); |
1c4850e2 YZ |
8657 | bytenr = btrfs_node_blockptr(eb, slot); |
8658 | generation = btrfs_node_ptr_generation(eb, slot); | |
2dd3e67b | 8659 | |
1c4850e2 YZ |
8660 | if (slot == path->slots[wc->level]) |
8661 | goto reada; | |
5d4f98a2 | 8662 | |
1c4850e2 YZ |
8663 | if (wc->stage == UPDATE_BACKREF && |
8664 | generation <= root->root_key.offset) | |
bd56b302 CM |
8665 | continue; |
8666 | ||
94fcca9f | 8667 | /* We don't lock the tree block, it's OK to be racy here */ |
2ff7e61e | 8668 | ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, |
3173a18f JB |
8669 | wc->level - 1, 1, &refs, |
8670 | &flags); | |
79787eaa JM |
8671 | /* We don't care about errors in readahead. */ |
8672 | if (ret < 0) | |
8673 | continue; | |
94fcca9f YZ |
8674 | BUG_ON(refs == 0); |
8675 | ||
1c4850e2 | 8676 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 YZ |
8677 | if (refs == 1) |
8678 | goto reada; | |
bd56b302 | 8679 | |
94fcca9f YZ |
8680 | if (wc->level == 1 && |
8681 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8682 | continue; | |
1c4850e2 YZ |
8683 | if (!wc->update_ref || |
8684 | generation <= root->root_key.offset) | |
8685 | continue; | |
8686 | btrfs_node_key_to_cpu(eb, &key, slot); | |
8687 | ret = btrfs_comp_cpu_keys(&key, | |
8688 | &wc->update_progress); | |
8689 | if (ret < 0) | |
8690 | continue; | |
94fcca9f YZ |
8691 | } else { |
8692 | if (wc->level == 1 && | |
8693 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8694 | continue; | |
6407bf6d | 8695 | } |
1c4850e2 | 8696 | reada: |
2ff7e61e | 8697 | readahead_tree_block(fs_info, bytenr); |
1c4850e2 | 8698 | nread++; |
20524f02 | 8699 | } |
1c4850e2 | 8700 | wc->reada_slot = slot; |
20524f02 | 8701 | } |
2c47e605 | 8702 | |
f82d02d9 | 8703 | /* |
2c016dc2 | 8704 | * helper to process tree block while walking down the tree. |
2c47e605 | 8705 | * |
2c47e605 YZ |
8706 | * when wc->stage == UPDATE_BACKREF, this function updates |
8707 | * back refs for pointers in the block. | |
8708 | * | |
8709 | * NOTE: return value 1 means we should stop walking down. | |
f82d02d9 | 8710 | */ |
2c47e605 | 8711 | static noinline int walk_down_proc(struct btrfs_trans_handle *trans, |
5d4f98a2 | 8712 | struct btrfs_root *root, |
2c47e605 | 8713 | struct btrfs_path *path, |
94fcca9f | 8714 | struct walk_control *wc, int lookup_info) |
f82d02d9 | 8715 | { |
2ff7e61e | 8716 | struct btrfs_fs_info *fs_info = root->fs_info; |
2c47e605 YZ |
8717 | int level = wc->level; |
8718 | struct extent_buffer *eb = path->nodes[level]; | |
2c47e605 | 8719 | u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF; |
f82d02d9 YZ |
8720 | int ret; |
8721 | ||
2c47e605 YZ |
8722 | if (wc->stage == UPDATE_BACKREF && |
8723 | btrfs_header_owner(eb) != root->root_key.objectid) | |
8724 | return 1; | |
f82d02d9 | 8725 | |
2c47e605 YZ |
8726 | /* |
8727 | * when reference count of tree block is 1, it won't increase | |
8728 | * again. once full backref flag is set, we never clear it. | |
8729 | */ | |
94fcca9f YZ |
8730 | if (lookup_info && |
8731 | ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) || | |
8732 | (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) { | |
2c47e605 | 8733 | BUG_ON(!path->locks[level]); |
2ff7e61e | 8734 | ret = btrfs_lookup_extent_info(trans, fs_info, |
3173a18f | 8735 | eb->start, level, 1, |
2c47e605 YZ |
8736 | &wc->refs[level], |
8737 | &wc->flags[level]); | |
79787eaa JM |
8738 | BUG_ON(ret == -ENOMEM); |
8739 | if (ret) | |
8740 | return ret; | |
2c47e605 YZ |
8741 | BUG_ON(wc->refs[level] == 0); |
8742 | } | |
5d4f98a2 | 8743 | |
2c47e605 YZ |
8744 | if (wc->stage == DROP_REFERENCE) { |
8745 | if (wc->refs[level] > 1) | |
8746 | return 1; | |
f82d02d9 | 8747 | |
2c47e605 | 8748 | if (path->locks[level] && !wc->keep_locks) { |
bd681513 | 8749 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
8750 | path->locks[level] = 0; |
8751 | } | |
8752 | return 0; | |
8753 | } | |
f82d02d9 | 8754 | |
2c47e605 YZ |
8755 | /* wc->stage == UPDATE_BACKREF */ |
8756 | if (!(wc->flags[level] & flag)) { | |
8757 | BUG_ON(!path->locks[level]); | |
e339a6b0 | 8758 | ret = btrfs_inc_ref(trans, root, eb, 1); |
79787eaa | 8759 | BUG_ON(ret); /* -ENOMEM */ |
e339a6b0 | 8760 | ret = btrfs_dec_ref(trans, root, eb, 0); |
79787eaa | 8761 | BUG_ON(ret); /* -ENOMEM */ |
2ff7e61e | 8762 | ret = btrfs_set_disk_extent_flags(trans, fs_info, eb->start, |
b1c79e09 JB |
8763 | eb->len, flag, |
8764 | btrfs_header_level(eb), 0); | |
79787eaa | 8765 | BUG_ON(ret); /* -ENOMEM */ |
2c47e605 YZ |
8766 | wc->flags[level] |= flag; |
8767 | } | |
8768 | ||
8769 | /* | |
8770 | * the block is shared by multiple trees, so it's not good to | |
8771 | * keep the tree lock | |
8772 | */ | |
8773 | if (path->locks[level] && level > 0) { | |
bd681513 | 8774 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
8775 | path->locks[level] = 0; |
8776 | } | |
8777 | return 0; | |
8778 | } | |
8779 | ||
1c4850e2 | 8780 | /* |
2c016dc2 | 8781 | * helper to process tree block pointer. |
1c4850e2 YZ |
8782 | * |
8783 | * when wc->stage == DROP_REFERENCE, this function checks | |
8784 | * reference count of the block pointed to. if the block | |
8785 | * is shared and we need update back refs for the subtree | |
8786 | * rooted at the block, this function changes wc->stage to | |
8787 | * UPDATE_BACKREF. if the block is shared and there is no | |
8788 | * need to update back, this function drops the reference | |
8789 | * to the block. | |
8790 | * | |
8791 | * NOTE: return value 1 means we should stop walking down. | |
8792 | */ | |
8793 | static noinline int do_walk_down(struct btrfs_trans_handle *trans, | |
8794 | struct btrfs_root *root, | |
8795 | struct btrfs_path *path, | |
94fcca9f | 8796 | struct walk_control *wc, int *lookup_info) |
1c4850e2 | 8797 | { |
0b246afa | 8798 | struct btrfs_fs_info *fs_info = root->fs_info; |
1c4850e2 YZ |
8799 | u64 bytenr; |
8800 | u64 generation; | |
8801 | u64 parent; | |
8802 | u32 blocksize; | |
8803 | struct btrfs_key key; | |
8804 | struct extent_buffer *next; | |
8805 | int level = wc->level; | |
8806 | int reada = 0; | |
8807 | int ret = 0; | |
1152651a | 8808 | bool need_account = false; |
1c4850e2 YZ |
8809 | |
8810 | generation = btrfs_node_ptr_generation(path->nodes[level], | |
8811 | path->slots[level]); | |
8812 | /* | |
8813 | * if the lower level block was created before the snapshot | |
8814 | * was created, we know there is no need to update back refs | |
8815 | * for the subtree | |
8816 | */ | |
8817 | if (wc->stage == UPDATE_BACKREF && | |
94fcca9f YZ |
8818 | generation <= root->root_key.offset) { |
8819 | *lookup_info = 1; | |
1c4850e2 | 8820 | return 1; |
94fcca9f | 8821 | } |
1c4850e2 YZ |
8822 | |
8823 | bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]); | |
0b246afa | 8824 | blocksize = fs_info->nodesize; |
1c4850e2 | 8825 | |
0b246afa | 8826 | next = find_extent_buffer(fs_info, bytenr); |
1c4850e2 | 8827 | if (!next) { |
2ff7e61e | 8828 | next = btrfs_find_create_tree_block(fs_info, bytenr); |
c871b0f2 LB |
8829 | if (IS_ERR(next)) |
8830 | return PTR_ERR(next); | |
8831 | ||
b2aaaa3b JB |
8832 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, next, |
8833 | level - 1); | |
1c4850e2 YZ |
8834 | reada = 1; |
8835 | } | |
8836 | btrfs_tree_lock(next); | |
8837 | btrfs_set_lock_blocking(next); | |
8838 | ||
2ff7e61e | 8839 | ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1, |
94fcca9f YZ |
8840 | &wc->refs[level - 1], |
8841 | &wc->flags[level - 1]); | |
4867268c JB |
8842 | if (ret < 0) |
8843 | goto out_unlock; | |
79787eaa | 8844 | |
c2cf52eb | 8845 | if (unlikely(wc->refs[level - 1] == 0)) { |
0b246afa | 8846 | btrfs_err(fs_info, "Missing references."); |
4867268c JB |
8847 | ret = -EIO; |
8848 | goto out_unlock; | |
c2cf52eb | 8849 | } |
94fcca9f | 8850 | *lookup_info = 0; |
1c4850e2 | 8851 | |
94fcca9f | 8852 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 | 8853 | if (wc->refs[level - 1] > 1) { |
1152651a | 8854 | need_account = true; |
94fcca9f YZ |
8855 | if (level == 1 && |
8856 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8857 | goto skip; | |
8858 | ||
1c4850e2 YZ |
8859 | if (!wc->update_ref || |
8860 | generation <= root->root_key.offset) | |
8861 | goto skip; | |
8862 | ||
8863 | btrfs_node_key_to_cpu(path->nodes[level], &key, | |
8864 | path->slots[level]); | |
8865 | ret = btrfs_comp_cpu_keys(&key, &wc->update_progress); | |
8866 | if (ret < 0) | |
8867 | goto skip; | |
8868 | ||
8869 | wc->stage = UPDATE_BACKREF; | |
8870 | wc->shared_level = level - 1; | |
8871 | } | |
94fcca9f YZ |
8872 | } else { |
8873 | if (level == 1 && | |
8874 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8875 | goto skip; | |
1c4850e2 YZ |
8876 | } |
8877 | ||
b9fab919 | 8878 | if (!btrfs_buffer_uptodate(next, generation, 0)) { |
1c4850e2 YZ |
8879 | btrfs_tree_unlock(next); |
8880 | free_extent_buffer(next); | |
8881 | next = NULL; | |
94fcca9f | 8882 | *lookup_info = 1; |
1c4850e2 YZ |
8883 | } |
8884 | ||
8885 | if (!next) { | |
8886 | if (reada && level == 1) | |
8887 | reada_walk_down(trans, root, wc, path); | |
2ff7e61e | 8888 | next = read_tree_block(fs_info, bytenr, generation); |
64c043de LB |
8889 | if (IS_ERR(next)) { |
8890 | return PTR_ERR(next); | |
8891 | } else if (!extent_buffer_uptodate(next)) { | |
416bc658 | 8892 | free_extent_buffer(next); |
97d9a8a4 | 8893 | return -EIO; |
416bc658 | 8894 | } |
1c4850e2 YZ |
8895 | btrfs_tree_lock(next); |
8896 | btrfs_set_lock_blocking(next); | |
8897 | } | |
8898 | ||
8899 | level--; | |
4867268c JB |
8900 | ASSERT(level == btrfs_header_level(next)); |
8901 | if (level != btrfs_header_level(next)) { | |
8902 | btrfs_err(root->fs_info, "mismatched level"); | |
8903 | ret = -EIO; | |
8904 | goto out_unlock; | |
8905 | } | |
1c4850e2 YZ |
8906 | path->nodes[level] = next; |
8907 | path->slots[level] = 0; | |
bd681513 | 8908 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
1c4850e2 YZ |
8909 | wc->level = level; |
8910 | if (wc->level == 1) | |
8911 | wc->reada_slot = 0; | |
8912 | return 0; | |
8913 | skip: | |
8914 | wc->refs[level - 1] = 0; | |
8915 | wc->flags[level - 1] = 0; | |
94fcca9f YZ |
8916 | if (wc->stage == DROP_REFERENCE) { |
8917 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
8918 | parent = path->nodes[level]->start; | |
8919 | } else { | |
4867268c | 8920 | ASSERT(root->root_key.objectid == |
94fcca9f | 8921 | btrfs_header_owner(path->nodes[level])); |
4867268c JB |
8922 | if (root->root_key.objectid != |
8923 | btrfs_header_owner(path->nodes[level])) { | |
8924 | btrfs_err(root->fs_info, | |
8925 | "mismatched block owner"); | |
8926 | ret = -EIO; | |
8927 | goto out_unlock; | |
8928 | } | |
94fcca9f YZ |
8929 | parent = 0; |
8930 | } | |
1c4850e2 | 8931 | |
1152651a | 8932 | if (need_account) { |
33d1f05c QW |
8933 | ret = btrfs_qgroup_trace_subtree(trans, root, next, |
8934 | generation, level - 1); | |
1152651a | 8935 | if (ret) { |
0b246afa | 8936 | btrfs_err_rl(fs_info, |
5d163e0e JM |
8937 | "Error %d accounting shared subtree. Quota is out of sync, rescan required.", |
8938 | ret); | |
1152651a MF |
8939 | } |
8940 | } | |
2ff7e61e JM |
8941 | ret = btrfs_free_extent(trans, fs_info, bytenr, blocksize, |
8942 | parent, root->root_key.objectid, | |
8943 | level - 1, 0); | |
4867268c JB |
8944 | if (ret) |
8945 | goto out_unlock; | |
1c4850e2 | 8946 | } |
4867268c JB |
8947 | |
8948 | *lookup_info = 1; | |
8949 | ret = 1; | |
8950 | ||
8951 | out_unlock: | |
1c4850e2 YZ |
8952 | btrfs_tree_unlock(next); |
8953 | free_extent_buffer(next); | |
4867268c JB |
8954 | |
8955 | return ret; | |
1c4850e2 YZ |
8956 | } |
8957 | ||
2c47e605 | 8958 | /* |
2c016dc2 | 8959 | * helper to process tree block while walking up the tree. |
2c47e605 YZ |
8960 | * |
8961 | * when wc->stage == DROP_REFERENCE, this function drops | |
8962 | * reference count on the block. | |
8963 | * | |
8964 | * when wc->stage == UPDATE_BACKREF, this function changes | |
8965 | * wc->stage back to DROP_REFERENCE if we changed wc->stage | |
8966 | * to UPDATE_BACKREF previously while processing the block. | |
8967 | * | |
8968 | * NOTE: return value 1 means we should stop walking up. | |
8969 | */ | |
8970 | static noinline int walk_up_proc(struct btrfs_trans_handle *trans, | |
8971 | struct btrfs_root *root, | |
8972 | struct btrfs_path *path, | |
8973 | struct walk_control *wc) | |
8974 | { | |
0b246afa | 8975 | struct btrfs_fs_info *fs_info = root->fs_info; |
f0486c68 | 8976 | int ret; |
2c47e605 YZ |
8977 | int level = wc->level; |
8978 | struct extent_buffer *eb = path->nodes[level]; | |
8979 | u64 parent = 0; | |
8980 | ||
8981 | if (wc->stage == UPDATE_BACKREF) { | |
8982 | BUG_ON(wc->shared_level < level); | |
8983 | if (level < wc->shared_level) | |
8984 | goto out; | |
8985 | ||
2c47e605 YZ |
8986 | ret = find_next_key(path, level + 1, &wc->update_progress); |
8987 | if (ret > 0) | |
8988 | wc->update_ref = 0; | |
8989 | ||
8990 | wc->stage = DROP_REFERENCE; | |
8991 | wc->shared_level = -1; | |
8992 | path->slots[level] = 0; | |
8993 | ||
8994 | /* | |
8995 | * check reference count again if the block isn't locked. | |
8996 | * we should start walking down the tree again if reference | |
8997 | * count is one. | |
8998 | */ | |
8999 | if (!path->locks[level]) { | |
9000 | BUG_ON(level == 0); | |
9001 | btrfs_tree_lock(eb); | |
9002 | btrfs_set_lock_blocking(eb); | |
bd681513 | 9003 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 | 9004 | |
2ff7e61e | 9005 | ret = btrfs_lookup_extent_info(trans, fs_info, |
3173a18f | 9006 | eb->start, level, 1, |
2c47e605 YZ |
9007 | &wc->refs[level], |
9008 | &wc->flags[level]); | |
79787eaa JM |
9009 | if (ret < 0) { |
9010 | btrfs_tree_unlock_rw(eb, path->locks[level]); | |
3268a246 | 9011 | path->locks[level] = 0; |
79787eaa JM |
9012 | return ret; |
9013 | } | |
2c47e605 YZ |
9014 | BUG_ON(wc->refs[level] == 0); |
9015 | if (wc->refs[level] == 1) { | |
bd681513 | 9016 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
3268a246 | 9017 | path->locks[level] = 0; |
2c47e605 YZ |
9018 | return 1; |
9019 | } | |
f82d02d9 | 9020 | } |
2c47e605 | 9021 | } |
f82d02d9 | 9022 | |
2c47e605 YZ |
9023 | /* wc->stage == DROP_REFERENCE */ |
9024 | BUG_ON(wc->refs[level] > 1 && !path->locks[level]); | |
5d4f98a2 | 9025 | |
2c47e605 YZ |
9026 | if (wc->refs[level] == 1) { |
9027 | if (level == 0) { | |
9028 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
e339a6b0 | 9029 | ret = btrfs_dec_ref(trans, root, eb, 1); |
2c47e605 | 9030 | else |
e339a6b0 | 9031 | ret = btrfs_dec_ref(trans, root, eb, 0); |
79787eaa | 9032 | BUG_ON(ret); /* -ENOMEM */ |
2ff7e61e | 9033 | ret = btrfs_qgroup_trace_leaf_items(trans, fs_info, eb); |
1152651a | 9034 | if (ret) { |
0b246afa | 9035 | btrfs_err_rl(fs_info, |
5d163e0e JM |
9036 | "error %d accounting leaf items. Quota is out of sync, rescan required.", |
9037 | ret); | |
1152651a | 9038 | } |
2c47e605 YZ |
9039 | } |
9040 | /* make block locked assertion in clean_tree_block happy */ | |
9041 | if (!path->locks[level] && | |
9042 | btrfs_header_generation(eb) == trans->transid) { | |
9043 | btrfs_tree_lock(eb); | |
9044 | btrfs_set_lock_blocking(eb); | |
bd681513 | 9045 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 | 9046 | } |
7c302b49 | 9047 | clean_tree_block(fs_info, eb); |
2c47e605 YZ |
9048 | } |
9049 | ||
9050 | if (eb == root->node) { | |
9051 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
9052 | parent = eb->start; | |
61e3643d QW |
9053 | else if (root->root_key.objectid != btrfs_header_owner(eb)) |
9054 | goto owner_mismatch; | |
2c47e605 YZ |
9055 | } else { |
9056 | if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
9057 | parent = path->nodes[level + 1]->start; | |
61e3643d QW |
9058 | else if (root->root_key.objectid != |
9059 | btrfs_header_owner(path->nodes[level + 1])) | |
9060 | goto owner_mismatch; | |
f82d02d9 | 9061 | } |
f82d02d9 | 9062 | |
5581a51a | 9063 | btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1); |
2c47e605 YZ |
9064 | out: |
9065 | wc->refs[level] = 0; | |
9066 | wc->flags[level] = 0; | |
f0486c68 | 9067 | return 0; |
61e3643d QW |
9068 | |
9069 | owner_mismatch: | |
9070 | btrfs_err_rl(fs_info, "unexpected tree owner, have %llu expect %llu", | |
9071 | btrfs_header_owner(eb), root->root_key.objectid); | |
9072 | return -EUCLEAN; | |
2c47e605 YZ |
9073 | } |
9074 | ||
9075 | static noinline int walk_down_tree(struct btrfs_trans_handle *trans, | |
9076 | struct btrfs_root *root, | |
9077 | struct btrfs_path *path, | |
9078 | struct walk_control *wc) | |
9079 | { | |
2c47e605 | 9080 | int level = wc->level; |
94fcca9f | 9081 | int lookup_info = 1; |
2c47e605 YZ |
9082 | int ret; |
9083 | ||
9084 | while (level >= 0) { | |
94fcca9f | 9085 | ret = walk_down_proc(trans, root, path, wc, lookup_info); |
2c47e605 YZ |
9086 | if (ret > 0) |
9087 | break; | |
9088 | ||
9089 | if (level == 0) | |
9090 | break; | |
9091 | ||
7a7965f8 YZ |
9092 | if (path->slots[level] >= |
9093 | btrfs_header_nritems(path->nodes[level])) | |
9094 | break; | |
9095 | ||
94fcca9f | 9096 | ret = do_walk_down(trans, root, path, wc, &lookup_info); |
1c4850e2 YZ |
9097 | if (ret > 0) { |
9098 | path->slots[level]++; | |
9099 | continue; | |
90d2c51d MX |
9100 | } else if (ret < 0) |
9101 | return ret; | |
1c4850e2 | 9102 | level = wc->level; |
f82d02d9 | 9103 | } |
f82d02d9 YZ |
9104 | return 0; |
9105 | } | |
9106 | ||
d397712b | 9107 | static noinline int walk_up_tree(struct btrfs_trans_handle *trans, |
98ed5174 | 9108 | struct btrfs_root *root, |
f82d02d9 | 9109 | struct btrfs_path *path, |
2c47e605 | 9110 | struct walk_control *wc, int max_level) |
20524f02 | 9111 | { |
2c47e605 | 9112 | int level = wc->level; |
20524f02 | 9113 | int ret; |
9f3a7427 | 9114 | |
2c47e605 YZ |
9115 | path->slots[level] = btrfs_header_nritems(path->nodes[level]); |
9116 | while (level < max_level && path->nodes[level]) { | |
9117 | wc->level = level; | |
9118 | if (path->slots[level] + 1 < | |
9119 | btrfs_header_nritems(path->nodes[level])) { | |
9120 | path->slots[level]++; | |
20524f02 CM |
9121 | return 0; |
9122 | } else { | |
2c47e605 YZ |
9123 | ret = walk_up_proc(trans, root, path, wc); |
9124 | if (ret > 0) | |
9125 | return 0; | |
61e3643d QW |
9126 | if (ret < 0) |
9127 | return ret; | |
bd56b302 | 9128 | |
2c47e605 | 9129 | if (path->locks[level]) { |
bd681513 CM |
9130 | btrfs_tree_unlock_rw(path->nodes[level], |
9131 | path->locks[level]); | |
2c47e605 | 9132 | path->locks[level] = 0; |
f82d02d9 | 9133 | } |
2c47e605 YZ |
9134 | free_extent_buffer(path->nodes[level]); |
9135 | path->nodes[level] = NULL; | |
9136 | level++; | |
20524f02 CM |
9137 | } |
9138 | } | |
9139 | return 1; | |
9140 | } | |
9141 | ||
9aca1d51 | 9142 | /* |
2c47e605 YZ |
9143 | * drop a subvolume tree. |
9144 | * | |
9145 | * this function traverses the tree freeing any blocks that only | |
9146 | * referenced by the tree. | |
9147 | * | |
9148 | * when a shared tree block is found. this function decreases its | |
9149 | * reference count by one. if update_ref is true, this function | |
9150 | * also make sure backrefs for the shared block and all lower level | |
9151 | * blocks are properly updated. | |
9d1a2a3a DS |
9152 | * |
9153 | * If called with for_reloc == 0, may exit early with -EAGAIN | |
9aca1d51 | 9154 | */ |
2c536799 | 9155 | int btrfs_drop_snapshot(struct btrfs_root *root, |
66d7e7f0 AJ |
9156 | struct btrfs_block_rsv *block_rsv, int update_ref, |
9157 | int for_reloc) | |
20524f02 | 9158 | { |
ab8d0fc4 | 9159 | struct btrfs_fs_info *fs_info = root->fs_info; |
5caf2a00 | 9160 | struct btrfs_path *path; |
2c47e605 | 9161 | struct btrfs_trans_handle *trans; |
ab8d0fc4 | 9162 | struct btrfs_root *tree_root = fs_info->tree_root; |
9f3a7427 | 9163 | struct btrfs_root_item *root_item = &root->root_item; |
2c47e605 YZ |
9164 | struct walk_control *wc; |
9165 | struct btrfs_key key; | |
9166 | int err = 0; | |
9167 | int ret; | |
9168 | int level; | |
d29a9f62 | 9169 | bool root_dropped = false; |
20524f02 | 9170 | |
ab8d0fc4 | 9171 | btrfs_debug(fs_info, "Drop subvolume %llu", root->objectid); |
1152651a | 9172 | |
5caf2a00 | 9173 | path = btrfs_alloc_path(); |
cb1b69f4 TI |
9174 | if (!path) { |
9175 | err = -ENOMEM; | |
9176 | goto out; | |
9177 | } | |
20524f02 | 9178 | |
2c47e605 | 9179 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
38a1a919 MF |
9180 | if (!wc) { |
9181 | btrfs_free_path(path); | |
cb1b69f4 TI |
9182 | err = -ENOMEM; |
9183 | goto out; | |
38a1a919 | 9184 | } |
2c47e605 | 9185 | |
a22285a6 | 9186 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
9187 | if (IS_ERR(trans)) { |
9188 | err = PTR_ERR(trans); | |
9189 | goto out_free; | |
9190 | } | |
98d5dc13 | 9191 | |
3fd0a558 YZ |
9192 | if (block_rsv) |
9193 | trans->block_rsv = block_rsv; | |
2c47e605 | 9194 | |
9f3a7427 | 9195 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { |
2c47e605 | 9196 | level = btrfs_header_level(root->node); |
5d4f98a2 YZ |
9197 | path->nodes[level] = btrfs_lock_root_node(root); |
9198 | btrfs_set_lock_blocking(path->nodes[level]); | |
9f3a7427 | 9199 | path->slots[level] = 0; |
bd681513 | 9200 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
9201 | memset(&wc->update_progress, 0, |
9202 | sizeof(wc->update_progress)); | |
9f3a7427 | 9203 | } else { |
9f3a7427 | 9204 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); |
2c47e605 YZ |
9205 | memcpy(&wc->update_progress, &key, |
9206 | sizeof(wc->update_progress)); | |
9207 | ||
6702ed49 | 9208 | level = root_item->drop_level; |
2c47e605 | 9209 | BUG_ON(level == 0); |
6702ed49 | 9210 | path->lowest_level = level; |
2c47e605 YZ |
9211 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
9212 | path->lowest_level = 0; | |
9213 | if (ret < 0) { | |
9214 | err = ret; | |
79787eaa | 9215 | goto out_end_trans; |
9f3a7427 | 9216 | } |
1c4850e2 | 9217 | WARN_ON(ret > 0); |
2c47e605 | 9218 | |
7d9eb12c CM |
9219 | /* |
9220 | * unlock our path, this is safe because only this | |
9221 | * function is allowed to delete this snapshot | |
9222 | */ | |
5d4f98a2 | 9223 | btrfs_unlock_up_safe(path, 0); |
2c47e605 YZ |
9224 | |
9225 | level = btrfs_header_level(root->node); | |
9226 | while (1) { | |
9227 | btrfs_tree_lock(path->nodes[level]); | |
9228 | btrfs_set_lock_blocking(path->nodes[level]); | |
fec386ac | 9229 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 | 9230 | |
2ff7e61e | 9231 | ret = btrfs_lookup_extent_info(trans, fs_info, |
2c47e605 | 9232 | path->nodes[level]->start, |
3173a18f | 9233 | level, 1, &wc->refs[level], |
2c47e605 | 9234 | &wc->flags[level]); |
79787eaa JM |
9235 | if (ret < 0) { |
9236 | err = ret; | |
9237 | goto out_end_trans; | |
9238 | } | |
2c47e605 YZ |
9239 | BUG_ON(wc->refs[level] == 0); |
9240 | ||
9241 | if (level == root_item->drop_level) | |
9242 | break; | |
9243 | ||
9244 | btrfs_tree_unlock(path->nodes[level]); | |
fec386ac | 9245 | path->locks[level] = 0; |
2c47e605 YZ |
9246 | WARN_ON(wc->refs[level] != 1); |
9247 | level--; | |
9248 | } | |
9f3a7427 | 9249 | } |
2c47e605 YZ |
9250 | |
9251 | wc->level = level; | |
9252 | wc->shared_level = -1; | |
9253 | wc->stage = DROP_REFERENCE; | |
9254 | wc->update_ref = update_ref; | |
9255 | wc->keep_locks = 0; | |
66d7e7f0 | 9256 | wc->for_reloc = for_reloc; |
0b246afa | 9257 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info); |
2c47e605 | 9258 | |
d397712b | 9259 | while (1) { |
9d1a2a3a | 9260 | |
2c47e605 YZ |
9261 | ret = walk_down_tree(trans, root, path, wc); |
9262 | if (ret < 0) { | |
9263 | err = ret; | |
20524f02 | 9264 | break; |
2c47e605 | 9265 | } |
9aca1d51 | 9266 | |
2c47e605 YZ |
9267 | ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL); |
9268 | if (ret < 0) { | |
9269 | err = ret; | |
20524f02 | 9270 | break; |
2c47e605 YZ |
9271 | } |
9272 | ||
9273 | if (ret > 0) { | |
9274 | BUG_ON(wc->stage != DROP_REFERENCE); | |
e7a84565 CM |
9275 | break; |
9276 | } | |
2c47e605 YZ |
9277 | |
9278 | if (wc->stage == DROP_REFERENCE) { | |
9279 | level = wc->level; | |
9280 | btrfs_node_key(path->nodes[level], | |
9281 | &root_item->drop_progress, | |
9282 | path->slots[level]); | |
9283 | root_item->drop_level = level; | |
9284 | } | |
9285 | ||
9286 | BUG_ON(wc->level == 0); | |
3a45bb20 | 9287 | if (btrfs_should_end_transaction(trans) || |
2ff7e61e | 9288 | (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) { |
2c47e605 YZ |
9289 | ret = btrfs_update_root(trans, tree_root, |
9290 | &root->root_key, | |
9291 | root_item); | |
79787eaa | 9292 | if (ret) { |
66642832 | 9293 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9294 | err = ret; |
9295 | goto out_end_trans; | |
9296 | } | |
2c47e605 | 9297 | |
3a45bb20 | 9298 | btrfs_end_transaction_throttle(trans); |
2ff7e61e | 9299 | if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) { |
ab8d0fc4 JM |
9300 | btrfs_debug(fs_info, |
9301 | "drop snapshot early exit"); | |
3c8f2422 JB |
9302 | err = -EAGAIN; |
9303 | goto out_free; | |
9304 | } | |
9305 | ||
a22285a6 | 9306 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
9307 | if (IS_ERR(trans)) { |
9308 | err = PTR_ERR(trans); | |
9309 | goto out_free; | |
9310 | } | |
3fd0a558 YZ |
9311 | if (block_rsv) |
9312 | trans->block_rsv = block_rsv; | |
c3e69d58 | 9313 | } |
20524f02 | 9314 | } |
b3b4aa74 | 9315 | btrfs_release_path(path); |
79787eaa JM |
9316 | if (err) |
9317 | goto out_end_trans; | |
2c47e605 | 9318 | |
1cd5447e | 9319 | ret = btrfs_del_root(trans, fs_info, &root->root_key); |
79787eaa | 9320 | if (ret) { |
66642832 | 9321 | btrfs_abort_transaction(trans, ret); |
85ab61fd | 9322 | err = ret; |
79787eaa JM |
9323 | goto out_end_trans; |
9324 | } | |
2c47e605 | 9325 | |
76dda93c | 9326 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { |
cb517eab MX |
9327 | ret = btrfs_find_root(tree_root, &root->root_key, path, |
9328 | NULL, NULL); | |
79787eaa | 9329 | if (ret < 0) { |
66642832 | 9330 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9331 | err = ret; |
9332 | goto out_end_trans; | |
9333 | } else if (ret > 0) { | |
84cd948c JB |
9334 | /* if we fail to delete the orphan item this time |
9335 | * around, it'll get picked up the next time. | |
9336 | * | |
9337 | * The most common failure here is just -ENOENT. | |
9338 | */ | |
9339 | btrfs_del_orphan_item(trans, tree_root, | |
9340 | root->root_key.objectid); | |
76dda93c YZ |
9341 | } |
9342 | } | |
9343 | ||
27cdeb70 | 9344 | if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) { |
2b9dbef2 | 9345 | btrfs_add_dropped_root(trans, root); |
76dda93c YZ |
9346 | } else { |
9347 | free_extent_buffer(root->node); | |
9348 | free_extent_buffer(root->commit_root); | |
b0feb9d9 | 9349 | btrfs_put_fs_root(root); |
76dda93c | 9350 | } |
d29a9f62 | 9351 | root_dropped = true; |
79787eaa | 9352 | out_end_trans: |
3a45bb20 | 9353 | btrfs_end_transaction_throttle(trans); |
79787eaa | 9354 | out_free: |
2c47e605 | 9355 | kfree(wc); |
5caf2a00 | 9356 | btrfs_free_path(path); |
cb1b69f4 | 9357 | out: |
d29a9f62 JB |
9358 | /* |
9359 | * So if we need to stop dropping the snapshot for whatever reason we | |
9360 | * need to make sure to add it back to the dead root list so that we | |
9361 | * keep trying to do the work later. This also cleans up roots if we | |
9362 | * don't have it in the radix (like when we recover after a power fail | |
9363 | * or unmount) so we don't leak memory. | |
9364 | */ | |
b37b39cd | 9365 | if (!for_reloc && root_dropped == false) |
d29a9f62 | 9366 | btrfs_add_dead_root(root); |
90515e7f | 9367 | if (err && err != -EAGAIN) |
ab8d0fc4 | 9368 | btrfs_handle_fs_error(fs_info, err, NULL); |
2c536799 | 9369 | return err; |
20524f02 | 9370 | } |
9078a3e1 | 9371 | |
2c47e605 YZ |
9372 | /* |
9373 | * drop subtree rooted at tree block 'node'. | |
9374 | * | |
9375 | * NOTE: this function will unlock and release tree block 'node' | |
66d7e7f0 | 9376 | * only used by relocation code |
2c47e605 | 9377 | */ |
f82d02d9 YZ |
9378 | int btrfs_drop_subtree(struct btrfs_trans_handle *trans, |
9379 | struct btrfs_root *root, | |
9380 | struct extent_buffer *node, | |
9381 | struct extent_buffer *parent) | |
9382 | { | |
0b246afa | 9383 | struct btrfs_fs_info *fs_info = root->fs_info; |
f82d02d9 | 9384 | struct btrfs_path *path; |
2c47e605 | 9385 | struct walk_control *wc; |
f82d02d9 YZ |
9386 | int level; |
9387 | int parent_level; | |
9388 | int ret = 0; | |
9389 | int wret; | |
9390 | ||
2c47e605 YZ |
9391 | BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); |
9392 | ||
f82d02d9 | 9393 | path = btrfs_alloc_path(); |
db5b493a TI |
9394 | if (!path) |
9395 | return -ENOMEM; | |
f82d02d9 | 9396 | |
2c47e605 | 9397 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
db5b493a TI |
9398 | if (!wc) { |
9399 | btrfs_free_path(path); | |
9400 | return -ENOMEM; | |
9401 | } | |
2c47e605 | 9402 | |
b9447ef8 | 9403 | btrfs_assert_tree_locked(parent); |
f82d02d9 YZ |
9404 | parent_level = btrfs_header_level(parent); |
9405 | extent_buffer_get(parent); | |
9406 | path->nodes[parent_level] = parent; | |
9407 | path->slots[parent_level] = btrfs_header_nritems(parent); | |
9408 | ||
b9447ef8 | 9409 | btrfs_assert_tree_locked(node); |
f82d02d9 | 9410 | level = btrfs_header_level(node); |
f82d02d9 YZ |
9411 | path->nodes[level] = node; |
9412 | path->slots[level] = 0; | |
bd681513 | 9413 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
9414 | |
9415 | wc->refs[parent_level] = 1; | |
9416 | wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
9417 | wc->level = level; | |
9418 | wc->shared_level = -1; | |
9419 | wc->stage = DROP_REFERENCE; | |
9420 | wc->update_ref = 0; | |
9421 | wc->keep_locks = 1; | |
66d7e7f0 | 9422 | wc->for_reloc = 1; |
0b246afa | 9423 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info); |
f82d02d9 YZ |
9424 | |
9425 | while (1) { | |
2c47e605 YZ |
9426 | wret = walk_down_tree(trans, root, path, wc); |
9427 | if (wret < 0) { | |
f82d02d9 | 9428 | ret = wret; |
f82d02d9 | 9429 | break; |
2c47e605 | 9430 | } |
f82d02d9 | 9431 | |
2c47e605 | 9432 | wret = walk_up_tree(trans, root, path, wc, parent_level); |
f82d02d9 YZ |
9433 | if (wret < 0) |
9434 | ret = wret; | |
9435 | if (wret != 0) | |
9436 | break; | |
9437 | } | |
9438 | ||
2c47e605 | 9439 | kfree(wc); |
f82d02d9 YZ |
9440 | btrfs_free_path(path); |
9441 | return ret; | |
9442 | } | |
9443 | ||
6202df69 | 9444 | static u64 update_block_group_flags(struct btrfs_fs_info *fs_info, u64 flags) |
ec44a35c CM |
9445 | { |
9446 | u64 num_devices; | |
fc67c450 | 9447 | u64 stripped; |
e4d8ec0f | 9448 | |
fc67c450 ID |
9449 | /* |
9450 | * if restripe for this chunk_type is on pick target profile and | |
9451 | * return, otherwise do the usual balance | |
9452 | */ | |
6202df69 | 9453 | stripped = get_restripe_target(fs_info, flags); |
fc67c450 ID |
9454 | if (stripped) |
9455 | return extended_to_chunk(stripped); | |
e4d8ec0f | 9456 | |
6202df69 | 9457 | num_devices = fs_info->fs_devices->rw_devices; |
cd02dca5 | 9458 | |
fc67c450 | 9459 | stripped = BTRFS_BLOCK_GROUP_RAID0 | |
53b381b3 | 9460 | BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 | |
fc67c450 ID |
9461 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10; |
9462 | ||
ec44a35c CM |
9463 | if (num_devices == 1) { |
9464 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
9465 | stripped = flags & ~stripped; | |
9466 | ||
9467 | /* turn raid0 into single device chunks */ | |
9468 | if (flags & BTRFS_BLOCK_GROUP_RAID0) | |
9469 | return stripped; | |
9470 | ||
9471 | /* turn mirroring into duplication */ | |
9472 | if (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
9473 | BTRFS_BLOCK_GROUP_RAID10)) | |
9474 | return stripped | BTRFS_BLOCK_GROUP_DUP; | |
ec44a35c CM |
9475 | } else { |
9476 | /* they already had raid on here, just return */ | |
ec44a35c CM |
9477 | if (flags & stripped) |
9478 | return flags; | |
9479 | ||
9480 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
9481 | stripped = flags & ~stripped; | |
9482 | ||
9483 | /* switch duplicated blocks with raid1 */ | |
9484 | if (flags & BTRFS_BLOCK_GROUP_DUP) | |
9485 | return stripped | BTRFS_BLOCK_GROUP_RAID1; | |
9486 | ||
e3176ca2 | 9487 | /* this is drive concat, leave it alone */ |
ec44a35c | 9488 | } |
e3176ca2 | 9489 | |
ec44a35c CM |
9490 | return flags; |
9491 | } | |
9492 | ||
868f401a | 9493 | static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force) |
0ef3e66b | 9494 | { |
f0486c68 YZ |
9495 | struct btrfs_space_info *sinfo = cache->space_info; |
9496 | u64 num_bytes; | |
199c36ea | 9497 | u64 min_allocable_bytes; |
f0486c68 | 9498 | int ret = -ENOSPC; |
0ef3e66b | 9499 | |
199c36ea MX |
9500 | /* |
9501 | * We need some metadata space and system metadata space for | |
9502 | * allocating chunks in some corner cases until we force to set | |
9503 | * it to be readonly. | |
9504 | */ | |
9505 | if ((sinfo->flags & | |
9506 | (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) && | |
9507 | !force) | |
ee22184b | 9508 | min_allocable_bytes = SZ_1M; |
199c36ea MX |
9509 | else |
9510 | min_allocable_bytes = 0; | |
9511 | ||
f0486c68 YZ |
9512 | spin_lock(&sinfo->lock); |
9513 | spin_lock(&cache->lock); | |
61cfea9b W |
9514 | |
9515 | if (cache->ro) { | |
868f401a | 9516 | cache->ro++; |
61cfea9b W |
9517 | ret = 0; |
9518 | goto out; | |
9519 | } | |
9520 | ||
f0486c68 YZ |
9521 | num_bytes = cache->key.offset - cache->reserved - cache->pinned - |
9522 | cache->bytes_super - btrfs_block_group_used(&cache->item); | |
9523 | ||
4136135b | 9524 | if (btrfs_space_info_used(sinfo, true) + num_bytes + |
37be25bc | 9525 | min_allocable_bytes <= sinfo->total_bytes) { |
f0486c68 | 9526 | sinfo->bytes_readonly += num_bytes; |
868f401a | 9527 | cache->ro++; |
633c0aad | 9528 | list_add_tail(&cache->ro_list, &sinfo->ro_bgs); |
f0486c68 YZ |
9529 | ret = 0; |
9530 | } | |
61cfea9b | 9531 | out: |
f0486c68 YZ |
9532 | spin_unlock(&cache->lock); |
9533 | spin_unlock(&sinfo->lock); | |
9534 | return ret; | |
9535 | } | |
7d9eb12c | 9536 | |
5e00f193 | 9537 | int btrfs_inc_block_group_ro(struct btrfs_fs_info *fs_info, |
f0486c68 | 9538 | struct btrfs_block_group_cache *cache) |
c286ac48 | 9539 | |
f0486c68 YZ |
9540 | { |
9541 | struct btrfs_trans_handle *trans; | |
9542 | u64 alloc_flags; | |
9543 | int ret; | |
7d9eb12c | 9544 | |
1bbc621e | 9545 | again: |
5e00f193 | 9546 | trans = btrfs_join_transaction(fs_info->extent_root); |
79787eaa JM |
9547 | if (IS_ERR(trans)) |
9548 | return PTR_ERR(trans); | |
5d4f98a2 | 9549 | |
1bbc621e CM |
9550 | /* |
9551 | * we're not allowed to set block groups readonly after the dirty | |
9552 | * block groups cache has started writing. If it already started, | |
9553 | * back off and let this transaction commit | |
9554 | */ | |
0b246afa | 9555 | mutex_lock(&fs_info->ro_block_group_mutex); |
3204d33c | 9556 | if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) { |
1bbc621e CM |
9557 | u64 transid = trans->transid; |
9558 | ||
0b246afa | 9559 | mutex_unlock(&fs_info->ro_block_group_mutex); |
3a45bb20 | 9560 | btrfs_end_transaction(trans); |
1bbc621e | 9561 | |
2ff7e61e | 9562 | ret = btrfs_wait_for_commit(fs_info, transid); |
1bbc621e CM |
9563 | if (ret) |
9564 | return ret; | |
9565 | goto again; | |
9566 | } | |
9567 | ||
153c35b6 CM |
9568 | /* |
9569 | * if we are changing raid levels, try to allocate a corresponding | |
9570 | * block group with the new raid level. | |
9571 | */ | |
0b246afa | 9572 | alloc_flags = update_block_group_flags(fs_info, cache->flags); |
153c35b6 | 9573 | if (alloc_flags != cache->flags) { |
2ff7e61e | 9574 | ret = do_chunk_alloc(trans, fs_info, alloc_flags, |
153c35b6 CM |
9575 | CHUNK_ALLOC_FORCE); |
9576 | /* | |
9577 | * ENOSPC is allowed here, we may have enough space | |
9578 | * already allocated at the new raid level to | |
9579 | * carry on | |
9580 | */ | |
9581 | if (ret == -ENOSPC) | |
9582 | ret = 0; | |
9583 | if (ret < 0) | |
9584 | goto out; | |
9585 | } | |
1bbc621e | 9586 | |
868f401a | 9587 | ret = inc_block_group_ro(cache, 0); |
f0486c68 YZ |
9588 | if (!ret) |
9589 | goto out; | |
2ff7e61e JM |
9590 | alloc_flags = get_alloc_profile(fs_info, cache->space_info->flags); |
9591 | ret = do_chunk_alloc(trans, fs_info, alloc_flags, | |
0e4f8f88 | 9592 | CHUNK_ALLOC_FORCE); |
f0486c68 YZ |
9593 | if (ret < 0) |
9594 | goto out; | |
868f401a | 9595 | ret = inc_block_group_ro(cache, 0); |
f0486c68 | 9596 | out: |
2f081088 | 9597 | if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) { |
0b246afa | 9598 | alloc_flags = update_block_group_flags(fs_info, cache->flags); |
34441361 | 9599 | mutex_lock(&fs_info->chunk_mutex); |
2ff7e61e | 9600 | check_system_chunk(trans, fs_info, alloc_flags); |
34441361 | 9601 | mutex_unlock(&fs_info->chunk_mutex); |
2f081088 | 9602 | } |
0b246afa | 9603 | mutex_unlock(&fs_info->ro_block_group_mutex); |
2f081088 | 9604 | |
3a45bb20 | 9605 | btrfs_end_transaction(trans); |
f0486c68 YZ |
9606 | return ret; |
9607 | } | |
5d4f98a2 | 9608 | |
c87f08ca | 9609 | int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans, |
2ff7e61e | 9610 | struct btrfs_fs_info *fs_info, u64 type) |
c87f08ca | 9611 | { |
2ff7e61e JM |
9612 | u64 alloc_flags = get_alloc_profile(fs_info, type); |
9613 | ||
9614 | return do_chunk_alloc(trans, fs_info, alloc_flags, CHUNK_ALLOC_FORCE); | |
c87f08ca CM |
9615 | } |
9616 | ||
6d07bcec MX |
9617 | /* |
9618 | * helper to account the unused space of all the readonly block group in the | |
633c0aad | 9619 | * space_info. takes mirrors into account. |
6d07bcec | 9620 | */ |
633c0aad | 9621 | u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo) |
6d07bcec MX |
9622 | { |
9623 | struct btrfs_block_group_cache *block_group; | |
9624 | u64 free_bytes = 0; | |
9625 | int factor; | |
9626 | ||
01327610 | 9627 | /* It's df, we don't care if it's racy */ |
633c0aad JB |
9628 | if (list_empty(&sinfo->ro_bgs)) |
9629 | return 0; | |
9630 | ||
9631 | spin_lock(&sinfo->lock); | |
9632 | list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) { | |
6d07bcec MX |
9633 | spin_lock(&block_group->lock); |
9634 | ||
9635 | if (!block_group->ro) { | |
9636 | spin_unlock(&block_group->lock); | |
9637 | continue; | |
9638 | } | |
9639 | ||
9640 | if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
9641 | BTRFS_BLOCK_GROUP_RAID10 | | |
9642 | BTRFS_BLOCK_GROUP_DUP)) | |
9643 | factor = 2; | |
9644 | else | |
9645 | factor = 1; | |
9646 | ||
9647 | free_bytes += (block_group->key.offset - | |
9648 | btrfs_block_group_used(&block_group->item)) * | |
9649 | factor; | |
9650 | ||
9651 | spin_unlock(&block_group->lock); | |
9652 | } | |
6d07bcec MX |
9653 | spin_unlock(&sinfo->lock); |
9654 | ||
9655 | return free_bytes; | |
9656 | } | |
9657 | ||
2ff7e61e | 9658 | void btrfs_dec_block_group_ro(struct btrfs_block_group_cache *cache) |
5d4f98a2 | 9659 | { |
f0486c68 YZ |
9660 | struct btrfs_space_info *sinfo = cache->space_info; |
9661 | u64 num_bytes; | |
9662 | ||
9663 | BUG_ON(!cache->ro); | |
9664 | ||
9665 | spin_lock(&sinfo->lock); | |
9666 | spin_lock(&cache->lock); | |
868f401a Z |
9667 | if (!--cache->ro) { |
9668 | num_bytes = cache->key.offset - cache->reserved - | |
9669 | cache->pinned - cache->bytes_super - | |
9670 | btrfs_block_group_used(&cache->item); | |
9671 | sinfo->bytes_readonly -= num_bytes; | |
9672 | list_del_init(&cache->ro_list); | |
9673 | } | |
f0486c68 YZ |
9674 | spin_unlock(&cache->lock); |
9675 | spin_unlock(&sinfo->lock); | |
5d4f98a2 YZ |
9676 | } |
9677 | ||
ba1bf481 JB |
9678 | /* |
9679 | * checks to see if its even possible to relocate this block group. | |
9680 | * | |
9681 | * @return - -1 if it's not a good idea to relocate this block group, 0 if its | |
9682 | * ok to go ahead and try. | |
9683 | */ | |
6bccf3ab | 9684 | int btrfs_can_relocate(struct btrfs_fs_info *fs_info, u64 bytenr) |
1a40e23b | 9685 | { |
6bccf3ab | 9686 | struct btrfs_root *root = fs_info->extent_root; |
ba1bf481 JB |
9687 | struct btrfs_block_group_cache *block_group; |
9688 | struct btrfs_space_info *space_info; | |
0b246afa | 9689 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
ba1bf481 | 9690 | struct btrfs_device *device; |
6df9a95e | 9691 | struct btrfs_trans_handle *trans; |
cdcb725c | 9692 | u64 min_free; |
6719db6a JB |
9693 | u64 dev_min = 1; |
9694 | u64 dev_nr = 0; | |
4a5e98f5 | 9695 | u64 target; |
0305bc27 | 9696 | int debug; |
cdcb725c | 9697 | int index; |
ba1bf481 JB |
9698 | int full = 0; |
9699 | int ret = 0; | |
1a40e23b | 9700 | |
0b246afa | 9701 | debug = btrfs_test_opt(fs_info, ENOSPC_DEBUG); |
0305bc27 | 9702 | |
0b246afa | 9703 | block_group = btrfs_lookup_block_group(fs_info, bytenr); |
1a40e23b | 9704 | |
ba1bf481 | 9705 | /* odd, couldn't find the block group, leave it alone */ |
0305bc27 QW |
9706 | if (!block_group) { |
9707 | if (debug) | |
0b246afa | 9708 | btrfs_warn(fs_info, |
0305bc27 QW |
9709 | "can't find block group for bytenr %llu", |
9710 | bytenr); | |
ba1bf481 | 9711 | return -1; |
0305bc27 | 9712 | } |
1a40e23b | 9713 | |
cdcb725c | 9714 | min_free = btrfs_block_group_used(&block_group->item); |
9715 | ||
ba1bf481 | 9716 | /* no bytes used, we're good */ |
cdcb725c | 9717 | if (!min_free) |
1a40e23b ZY |
9718 | goto out; |
9719 | ||
ba1bf481 JB |
9720 | space_info = block_group->space_info; |
9721 | spin_lock(&space_info->lock); | |
17d217fe | 9722 | |
ba1bf481 | 9723 | full = space_info->full; |
17d217fe | 9724 | |
ba1bf481 JB |
9725 | /* |
9726 | * if this is the last block group we have in this space, we can't | |
7ce618db CM |
9727 | * relocate it unless we're able to allocate a new chunk below. |
9728 | * | |
9729 | * Otherwise, we need to make sure we have room in the space to handle | |
9730 | * all of the extents from this block group. If we can, we're good | |
ba1bf481 | 9731 | */ |
7ce618db | 9732 | if ((space_info->total_bytes != block_group->key.offset) && |
4136135b LB |
9733 | (btrfs_space_info_used(space_info, false) + min_free < |
9734 | space_info->total_bytes)) { | |
ba1bf481 JB |
9735 | spin_unlock(&space_info->lock); |
9736 | goto out; | |
17d217fe | 9737 | } |
ba1bf481 | 9738 | spin_unlock(&space_info->lock); |
ea8c2819 | 9739 | |
ba1bf481 JB |
9740 | /* |
9741 | * ok we don't have enough space, but maybe we have free space on our | |
9742 | * devices to allocate new chunks for relocation, so loop through our | |
4a5e98f5 ID |
9743 | * alloc devices and guess if we have enough space. if this block |
9744 | * group is going to be restriped, run checks against the target | |
9745 | * profile instead of the current one. | |
ba1bf481 JB |
9746 | */ |
9747 | ret = -1; | |
ea8c2819 | 9748 | |
cdcb725c | 9749 | /* |
9750 | * index: | |
9751 | * 0: raid10 | |
9752 | * 1: raid1 | |
9753 | * 2: dup | |
9754 | * 3: raid0 | |
9755 | * 4: single | |
9756 | */ | |
0b246afa | 9757 | target = get_restripe_target(fs_info, block_group->flags); |
4a5e98f5 | 9758 | if (target) { |
31e50229 | 9759 | index = __get_raid_index(extended_to_chunk(target)); |
4a5e98f5 ID |
9760 | } else { |
9761 | /* | |
9762 | * this is just a balance, so if we were marked as full | |
9763 | * we know there is no space for a new chunk | |
9764 | */ | |
0305bc27 QW |
9765 | if (full) { |
9766 | if (debug) | |
0b246afa JM |
9767 | btrfs_warn(fs_info, |
9768 | "no space to alloc new chunk for block group %llu", | |
9769 | block_group->key.objectid); | |
4a5e98f5 | 9770 | goto out; |
0305bc27 | 9771 | } |
4a5e98f5 ID |
9772 | |
9773 | index = get_block_group_index(block_group); | |
9774 | } | |
9775 | ||
e6ec716f | 9776 | if (index == BTRFS_RAID_RAID10) { |
cdcb725c | 9777 | dev_min = 4; |
6719db6a JB |
9778 | /* Divide by 2 */ |
9779 | min_free >>= 1; | |
e6ec716f | 9780 | } else if (index == BTRFS_RAID_RAID1) { |
cdcb725c | 9781 | dev_min = 2; |
e6ec716f | 9782 | } else if (index == BTRFS_RAID_DUP) { |
6719db6a JB |
9783 | /* Multiply by 2 */ |
9784 | min_free <<= 1; | |
e6ec716f | 9785 | } else if (index == BTRFS_RAID_RAID0) { |
cdcb725c | 9786 | dev_min = fs_devices->rw_devices; |
47c5713f | 9787 | min_free = div64_u64(min_free, dev_min); |
cdcb725c | 9788 | } |
9789 | ||
6df9a95e JB |
9790 | /* We need to do this so that we can look at pending chunks */ |
9791 | trans = btrfs_join_transaction(root); | |
9792 | if (IS_ERR(trans)) { | |
9793 | ret = PTR_ERR(trans); | |
9794 | goto out; | |
9795 | } | |
9796 | ||
0b246afa | 9797 | mutex_lock(&fs_info->chunk_mutex); |
ba1bf481 | 9798 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { |
7bfc837d | 9799 | u64 dev_offset; |
56bec294 | 9800 | |
ba1bf481 JB |
9801 | /* |
9802 | * check to make sure we can actually find a chunk with enough | |
9803 | * space to fit our block group in. | |
9804 | */ | |
63a212ab SB |
9805 | if (device->total_bytes > device->bytes_used + min_free && |
9806 | !device->is_tgtdev_for_dev_replace) { | |
6df9a95e | 9807 | ret = find_free_dev_extent(trans, device, min_free, |
7bfc837d | 9808 | &dev_offset, NULL); |
ba1bf481 | 9809 | if (!ret) |
cdcb725c | 9810 | dev_nr++; |
9811 | ||
9812 | if (dev_nr >= dev_min) | |
73e48b27 | 9813 | break; |
cdcb725c | 9814 | |
ba1bf481 | 9815 | ret = -1; |
725c8463 | 9816 | } |
edbd8d4e | 9817 | } |
0305bc27 | 9818 | if (debug && ret == -1) |
0b246afa JM |
9819 | btrfs_warn(fs_info, |
9820 | "no space to allocate a new chunk for block group %llu", | |
9821 | block_group->key.objectid); | |
9822 | mutex_unlock(&fs_info->chunk_mutex); | |
3a45bb20 | 9823 | btrfs_end_transaction(trans); |
edbd8d4e | 9824 | out: |
ba1bf481 | 9825 | btrfs_put_block_group(block_group); |
edbd8d4e CM |
9826 | return ret; |
9827 | } | |
9828 | ||
6bccf3ab JM |
9829 | static int find_first_block_group(struct btrfs_fs_info *fs_info, |
9830 | struct btrfs_path *path, | |
9831 | struct btrfs_key *key) | |
0b86a832 | 9832 | { |
6bccf3ab | 9833 | struct btrfs_root *root = fs_info->extent_root; |
925baedd | 9834 | int ret = 0; |
0b86a832 CM |
9835 | struct btrfs_key found_key; |
9836 | struct extent_buffer *leaf; | |
34407a17 QW |
9837 | struct btrfs_block_group_item bg; |
9838 | u64 flags; | |
0b86a832 | 9839 | int slot; |
edbd8d4e | 9840 | |
0b86a832 CM |
9841 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); |
9842 | if (ret < 0) | |
925baedd CM |
9843 | goto out; |
9844 | ||
d397712b | 9845 | while (1) { |
0b86a832 | 9846 | slot = path->slots[0]; |
edbd8d4e | 9847 | leaf = path->nodes[0]; |
0b86a832 CM |
9848 | if (slot >= btrfs_header_nritems(leaf)) { |
9849 | ret = btrfs_next_leaf(root, path); | |
9850 | if (ret == 0) | |
9851 | continue; | |
9852 | if (ret < 0) | |
925baedd | 9853 | goto out; |
0b86a832 | 9854 | break; |
edbd8d4e | 9855 | } |
0b86a832 | 9856 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
edbd8d4e | 9857 | |
0b86a832 | 9858 | if (found_key.objectid >= key->objectid && |
925baedd | 9859 | found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) { |
6fb37b75 LB |
9860 | struct extent_map_tree *em_tree; |
9861 | struct extent_map *em; | |
9862 | ||
9863 | em_tree = &root->fs_info->mapping_tree.map_tree; | |
9864 | read_lock(&em_tree->lock); | |
9865 | em = lookup_extent_mapping(em_tree, found_key.objectid, | |
9866 | found_key.offset); | |
9867 | read_unlock(&em_tree->lock); | |
9868 | if (!em) { | |
0b246afa | 9869 | btrfs_err(fs_info, |
6fb37b75 LB |
9870 | "logical %llu len %llu found bg but no related chunk", |
9871 | found_key.objectid, found_key.offset); | |
9872 | ret = -ENOENT; | |
34407a17 QW |
9873 | } else if (em->start != found_key.objectid || |
9874 | em->len != found_key.offset) { | |
9875 | btrfs_err(fs_info, | |
9876 | "block group %llu len %llu mismatch with chunk %llu len %llu", | |
9877 | found_key.objectid, found_key.offset, | |
9878 | em->start, em->len); | |
9879 | ret = -EUCLEAN; | |
6fb37b75 | 9880 | } else { |
34407a17 QW |
9881 | read_extent_buffer(leaf, &bg, |
9882 | btrfs_item_ptr_offset(leaf, slot), | |
9883 | sizeof(bg)); | |
9884 | flags = btrfs_block_group_flags(&bg) & | |
9885 | BTRFS_BLOCK_GROUP_TYPE_MASK; | |
9886 | ||
9887 | if (flags != (em->map_lookup->type & | |
9888 | BTRFS_BLOCK_GROUP_TYPE_MASK)) { | |
9889 | btrfs_err(fs_info, | |
9890 | "block group %llu len %llu type flags 0x%llx mismatch with chunk type flags 0x%llx", | |
9891 | found_key.objectid, | |
9892 | found_key.offset, flags, | |
9893 | (BTRFS_BLOCK_GROUP_TYPE_MASK & | |
9894 | em->map_lookup->type)); | |
9895 | ret = -EUCLEAN; | |
9896 | } else { | |
9897 | ret = 0; | |
9898 | } | |
6fb37b75 | 9899 | } |
187ee58c | 9900 | free_extent_map(em); |
925baedd CM |
9901 | goto out; |
9902 | } | |
0b86a832 | 9903 | path->slots[0]++; |
edbd8d4e | 9904 | } |
925baedd | 9905 | out: |
0b86a832 | 9906 | return ret; |
edbd8d4e CM |
9907 | } |
9908 | ||
0af3d00b JB |
9909 | void btrfs_put_block_group_cache(struct btrfs_fs_info *info) |
9910 | { | |
9911 | struct btrfs_block_group_cache *block_group; | |
9912 | u64 last = 0; | |
9913 | ||
9914 | while (1) { | |
9915 | struct inode *inode; | |
9916 | ||
9917 | block_group = btrfs_lookup_first_block_group(info, last); | |
9918 | while (block_group) { | |
9ce0f8b6 | 9919 | wait_block_group_cache_done(block_group); |
0af3d00b JB |
9920 | spin_lock(&block_group->lock); |
9921 | if (block_group->iref) | |
9922 | break; | |
9923 | spin_unlock(&block_group->lock); | |
2ff7e61e | 9924 | block_group = next_block_group(info, block_group); |
0af3d00b JB |
9925 | } |
9926 | if (!block_group) { | |
9927 | if (last == 0) | |
9928 | break; | |
9929 | last = 0; | |
9930 | continue; | |
9931 | } | |
9932 | ||
9933 | inode = block_group->inode; | |
9934 | block_group->iref = 0; | |
9935 | block_group->inode = NULL; | |
9936 | spin_unlock(&block_group->lock); | |
f3bca802 | 9937 | ASSERT(block_group->io_ctl.inode == NULL); |
0af3d00b JB |
9938 | iput(inode); |
9939 | last = block_group->key.objectid + block_group->key.offset; | |
9940 | btrfs_put_block_group(block_group); | |
9941 | } | |
9942 | } | |
9943 | ||
5cdd7db6 FM |
9944 | /* |
9945 | * Must be called only after stopping all workers, since we could have block | |
9946 | * group caching kthreads running, and therefore they could race with us if we | |
9947 | * freed the block groups before stopping them. | |
9948 | */ | |
1a40e23b ZY |
9949 | int btrfs_free_block_groups(struct btrfs_fs_info *info) |
9950 | { | |
9951 | struct btrfs_block_group_cache *block_group; | |
4184ea7f | 9952 | struct btrfs_space_info *space_info; |
11833d66 | 9953 | struct btrfs_caching_control *caching_ctl; |
1a40e23b ZY |
9954 | struct rb_node *n; |
9955 | ||
9e351cc8 | 9956 | down_write(&info->commit_root_sem); |
11833d66 YZ |
9957 | while (!list_empty(&info->caching_block_groups)) { |
9958 | caching_ctl = list_entry(info->caching_block_groups.next, | |
9959 | struct btrfs_caching_control, list); | |
9960 | list_del(&caching_ctl->list); | |
9961 | put_caching_control(caching_ctl); | |
9962 | } | |
9e351cc8 | 9963 | up_write(&info->commit_root_sem); |
11833d66 | 9964 | |
47ab2a6c JB |
9965 | spin_lock(&info->unused_bgs_lock); |
9966 | while (!list_empty(&info->unused_bgs)) { | |
9967 | block_group = list_first_entry(&info->unused_bgs, | |
9968 | struct btrfs_block_group_cache, | |
9969 | bg_list); | |
9970 | list_del_init(&block_group->bg_list); | |
9971 | btrfs_put_block_group(block_group); | |
9972 | } | |
9973 | spin_unlock(&info->unused_bgs_lock); | |
9974 | ||
1a40e23b ZY |
9975 | spin_lock(&info->block_group_cache_lock); |
9976 | while ((n = rb_last(&info->block_group_cache_tree)) != NULL) { | |
9977 | block_group = rb_entry(n, struct btrfs_block_group_cache, | |
9978 | cache_node); | |
1a40e23b ZY |
9979 | rb_erase(&block_group->cache_node, |
9980 | &info->block_group_cache_tree); | |
01eacb27 | 9981 | RB_CLEAR_NODE(&block_group->cache_node); |
d899e052 YZ |
9982 | spin_unlock(&info->block_group_cache_lock); |
9983 | ||
80eb234a | 9984 | down_write(&block_group->space_info->groups_sem); |
1a40e23b | 9985 | list_del(&block_group->list); |
80eb234a | 9986 | up_write(&block_group->space_info->groups_sem); |
d2fb3437 | 9987 | |
3c14874a JB |
9988 | /* |
9989 | * We haven't cached this block group, which means we could | |
9990 | * possibly have excluded extents on this block group. | |
9991 | */ | |
36cce922 JB |
9992 | if (block_group->cached == BTRFS_CACHE_NO || |
9993 | block_group->cached == BTRFS_CACHE_ERROR) | |
2ff7e61e | 9994 | free_excluded_extents(info, block_group); |
3c14874a | 9995 | |
817d52f8 | 9996 | btrfs_remove_free_space_cache(block_group); |
5cdd7db6 | 9997 | ASSERT(block_group->cached != BTRFS_CACHE_STARTED); |
f3bca802 LB |
9998 | ASSERT(list_empty(&block_group->dirty_list)); |
9999 | ASSERT(list_empty(&block_group->io_list)); | |
10000 | ASSERT(list_empty(&block_group->bg_list)); | |
10001 | ASSERT(atomic_read(&block_group->count) == 1); | |
11dfe35a | 10002 | btrfs_put_block_group(block_group); |
d899e052 YZ |
10003 | |
10004 | spin_lock(&info->block_group_cache_lock); | |
1a40e23b ZY |
10005 | } |
10006 | spin_unlock(&info->block_group_cache_lock); | |
4184ea7f CM |
10007 | |
10008 | /* now that all the block groups are freed, go through and | |
10009 | * free all the space_info structs. This is only called during | |
10010 | * the final stages of unmount, and so we know nobody is | |
10011 | * using them. We call synchronize_rcu() once before we start, | |
10012 | * just to be on the safe side. | |
10013 | */ | |
10014 | synchronize_rcu(); | |
10015 | ||
8929ecfa YZ |
10016 | release_global_block_rsv(info); |
10017 | ||
67871254 | 10018 | while (!list_empty(&info->space_info)) { |
6ab0a202 JM |
10019 | int i; |
10020 | ||
4184ea7f CM |
10021 | space_info = list_entry(info->space_info.next, |
10022 | struct btrfs_space_info, | |
10023 | list); | |
d555b6c3 JB |
10024 | |
10025 | /* | |
10026 | * Do not hide this behind enospc_debug, this is actually | |
10027 | * important and indicates a real bug if this happens. | |
10028 | */ | |
10029 | if (WARN_ON(space_info->bytes_pinned > 0 || | |
b069e0c3 | 10030 | space_info->bytes_reserved > 0 || |
d555b6c3 | 10031 | space_info->bytes_may_use > 0)) |
ab8d0fc4 | 10032 | dump_space_info(info, space_info, 0, 0); |
4184ea7f | 10033 | list_del(&space_info->list); |
6ab0a202 JM |
10034 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { |
10035 | struct kobject *kobj; | |
c1895442 JM |
10036 | kobj = space_info->block_group_kobjs[i]; |
10037 | space_info->block_group_kobjs[i] = NULL; | |
10038 | if (kobj) { | |
6ab0a202 JM |
10039 | kobject_del(kobj); |
10040 | kobject_put(kobj); | |
10041 | } | |
10042 | } | |
10043 | kobject_del(&space_info->kobj); | |
10044 | kobject_put(&space_info->kobj); | |
4184ea7f | 10045 | } |
1a40e23b ZY |
10046 | return 0; |
10047 | } | |
10048 | ||
b742bb82 YZ |
10049 | static void __link_block_group(struct btrfs_space_info *space_info, |
10050 | struct btrfs_block_group_cache *cache) | |
10051 | { | |
10052 | int index = get_block_group_index(cache); | |
ed55b6ac | 10053 | bool first = false; |
b742bb82 YZ |
10054 | |
10055 | down_write(&space_info->groups_sem); | |
ed55b6ac JM |
10056 | if (list_empty(&space_info->block_groups[index])) |
10057 | first = true; | |
10058 | list_add_tail(&cache->list, &space_info->block_groups[index]); | |
10059 | up_write(&space_info->groups_sem); | |
10060 | ||
10061 | if (first) { | |
c1895442 | 10062 | struct raid_kobject *rkobj; |
6ab0a202 JM |
10063 | int ret; |
10064 | ||
c1895442 JM |
10065 | rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS); |
10066 | if (!rkobj) | |
10067 | goto out_err; | |
10068 | rkobj->raid_type = index; | |
10069 | kobject_init(&rkobj->kobj, &btrfs_raid_ktype); | |
10070 | ret = kobject_add(&rkobj->kobj, &space_info->kobj, | |
10071 | "%s", get_raid_name(index)); | |
6ab0a202 | 10072 | if (ret) { |
c1895442 JM |
10073 | kobject_put(&rkobj->kobj); |
10074 | goto out_err; | |
6ab0a202 | 10075 | } |
c1895442 | 10076 | space_info->block_group_kobjs[index] = &rkobj->kobj; |
6ab0a202 | 10077 | } |
c1895442 JM |
10078 | |
10079 | return; | |
10080 | out_err: | |
ab8d0fc4 JM |
10081 | btrfs_warn(cache->fs_info, |
10082 | "failed to add kobject for block cache, ignoring"); | |
b742bb82 YZ |
10083 | } |
10084 | ||
920e4a58 | 10085 | static struct btrfs_block_group_cache * |
2ff7e61e JM |
10086 | btrfs_create_block_group_cache(struct btrfs_fs_info *fs_info, |
10087 | u64 start, u64 size) | |
920e4a58 MX |
10088 | { |
10089 | struct btrfs_block_group_cache *cache; | |
10090 | ||
10091 | cache = kzalloc(sizeof(*cache), GFP_NOFS); | |
10092 | if (!cache) | |
10093 | return NULL; | |
10094 | ||
10095 | cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl), | |
10096 | GFP_NOFS); | |
10097 | if (!cache->free_space_ctl) { | |
10098 | kfree(cache); | |
10099 | return NULL; | |
10100 | } | |
10101 | ||
10102 | cache->key.objectid = start; | |
10103 | cache->key.offset = size; | |
10104 | cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; | |
10105 | ||
0b246afa | 10106 | cache->fs_info = fs_info; |
e4ff5fb5 | 10107 | cache->full_stripe_len = btrfs_full_stripe_len(fs_info, start); |
1e144fb8 OS |
10108 | set_free_space_tree_thresholds(cache); |
10109 | ||
920e4a58 MX |
10110 | atomic_set(&cache->count, 1); |
10111 | spin_lock_init(&cache->lock); | |
e570fd27 | 10112 | init_rwsem(&cache->data_rwsem); |
920e4a58 MX |
10113 | INIT_LIST_HEAD(&cache->list); |
10114 | INIT_LIST_HEAD(&cache->cluster_list); | |
47ab2a6c | 10115 | INIT_LIST_HEAD(&cache->bg_list); |
633c0aad | 10116 | INIT_LIST_HEAD(&cache->ro_list); |
ce93ec54 | 10117 | INIT_LIST_HEAD(&cache->dirty_list); |
c9dc4c65 | 10118 | INIT_LIST_HEAD(&cache->io_list); |
920e4a58 | 10119 | btrfs_init_free_space_ctl(cache); |
04216820 | 10120 | atomic_set(&cache->trimming, 0); |
a5ed9182 | 10121 | mutex_init(&cache->free_space_lock); |
0966a7b1 | 10122 | btrfs_init_full_stripe_locks_tree(&cache->full_stripe_locks_root); |
920e4a58 MX |
10123 | |
10124 | return cache; | |
10125 | } | |
10126 | ||
895586ec QW |
10127 | |
10128 | /* | |
10129 | * Iterate all chunks and verify that each of them has the corresponding block | |
10130 | * group | |
10131 | */ | |
10132 | static int check_chunk_block_group_mappings(struct btrfs_fs_info *fs_info) | |
10133 | { | |
10134 | struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree; | |
10135 | struct extent_map *em; | |
10136 | struct btrfs_block_group_cache *bg; | |
10137 | u64 start = 0; | |
10138 | int ret = 0; | |
10139 | ||
10140 | while (1) { | |
10141 | read_lock(&map_tree->map_tree.lock); | |
10142 | /* | |
10143 | * lookup_extent_mapping will return the first extent map | |
10144 | * intersecting the range, so setting @len to 1 is enough to | |
10145 | * get the first chunk. | |
10146 | */ | |
10147 | em = lookup_extent_mapping(&map_tree->map_tree, start, 1); | |
10148 | read_unlock(&map_tree->map_tree.lock); | |
10149 | if (!em) | |
10150 | break; | |
10151 | ||
10152 | bg = btrfs_lookup_block_group(fs_info, em->start); | |
10153 | if (!bg) { | |
10154 | btrfs_err(fs_info, | |
10155 | "chunk start=%llu len=%llu doesn't have corresponding block group", | |
10156 | em->start, em->len); | |
10157 | ret = -EUCLEAN; | |
10158 | free_extent_map(em); | |
10159 | break; | |
10160 | } | |
10161 | if (bg->key.objectid != em->start || | |
10162 | bg->key.offset != em->len || | |
10163 | (bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK) != | |
10164 | (em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK)) { | |
10165 | btrfs_err(fs_info, | |
10166 | "chunk start=%llu len=%llu flags=0x%llx doesn't match block group start=%llu len=%llu flags=0x%llx", | |
10167 | em->start, em->len, | |
10168 | em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK, | |
10169 | bg->key.objectid, bg->key.offset, | |
10170 | bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK); | |
10171 | ret = -EUCLEAN; | |
10172 | free_extent_map(em); | |
10173 | btrfs_put_block_group(bg); | |
10174 | break; | |
10175 | } | |
10176 | start = em->start + em->len; | |
10177 | free_extent_map(em); | |
10178 | btrfs_put_block_group(bg); | |
10179 | } | |
10180 | return ret; | |
10181 | } | |
10182 | ||
5b4aacef | 10183 | int btrfs_read_block_groups(struct btrfs_fs_info *info) |
9078a3e1 CM |
10184 | { |
10185 | struct btrfs_path *path; | |
10186 | int ret; | |
9078a3e1 | 10187 | struct btrfs_block_group_cache *cache; |
6324fbf3 | 10188 | struct btrfs_space_info *space_info; |
9078a3e1 CM |
10189 | struct btrfs_key key; |
10190 | struct btrfs_key found_key; | |
5f39d397 | 10191 | struct extent_buffer *leaf; |
0af3d00b JB |
10192 | int need_clear = 0; |
10193 | u64 cache_gen; | |
49303381 LB |
10194 | u64 feature; |
10195 | int mixed; | |
10196 | ||
10197 | feature = btrfs_super_incompat_flags(info->super_copy); | |
10198 | mixed = !!(feature & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS); | |
96b5179d | 10199 | |
9078a3e1 | 10200 | key.objectid = 0; |
0b86a832 | 10201 | key.offset = 0; |
962a298f | 10202 | key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; |
9078a3e1 CM |
10203 | path = btrfs_alloc_path(); |
10204 | if (!path) | |
10205 | return -ENOMEM; | |
e4058b54 | 10206 | path->reada = READA_FORWARD; |
9078a3e1 | 10207 | |
0b246afa JM |
10208 | cache_gen = btrfs_super_cache_generation(info->super_copy); |
10209 | if (btrfs_test_opt(info, SPACE_CACHE) && | |
10210 | btrfs_super_generation(info->super_copy) != cache_gen) | |
0af3d00b | 10211 | need_clear = 1; |
0b246afa | 10212 | if (btrfs_test_opt(info, CLEAR_CACHE)) |
88c2ba3b | 10213 | need_clear = 1; |
0af3d00b | 10214 | |
d397712b | 10215 | while (1) { |
6bccf3ab | 10216 | ret = find_first_block_group(info, path, &key); |
b742bb82 YZ |
10217 | if (ret > 0) |
10218 | break; | |
0b86a832 CM |
10219 | if (ret != 0) |
10220 | goto error; | |
920e4a58 | 10221 | |
5f39d397 CM |
10222 | leaf = path->nodes[0]; |
10223 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
920e4a58 | 10224 | |
2ff7e61e | 10225 | cache = btrfs_create_block_group_cache(info, found_key.objectid, |
920e4a58 | 10226 | found_key.offset); |
9078a3e1 | 10227 | if (!cache) { |
0b86a832 | 10228 | ret = -ENOMEM; |
f0486c68 | 10229 | goto error; |
9078a3e1 | 10230 | } |
96303081 | 10231 | |
cf7c1ef6 LB |
10232 | if (need_clear) { |
10233 | /* | |
10234 | * When we mount with old space cache, we need to | |
10235 | * set BTRFS_DC_CLEAR and set dirty flag. | |
10236 | * | |
10237 | * a) Setting 'BTRFS_DC_CLEAR' makes sure that we | |
10238 | * truncate the old free space cache inode and | |
10239 | * setup a new one. | |
10240 | * b) Setting 'dirty flag' makes sure that we flush | |
10241 | * the new space cache info onto disk. | |
10242 | */ | |
0b246afa | 10243 | if (btrfs_test_opt(info, SPACE_CACHE)) |
ce93ec54 | 10244 | cache->disk_cache_state = BTRFS_DC_CLEAR; |
cf7c1ef6 | 10245 | } |
0af3d00b | 10246 | |
5f39d397 CM |
10247 | read_extent_buffer(leaf, &cache->item, |
10248 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
10249 | sizeof(cache->item)); | |
920e4a58 | 10250 | cache->flags = btrfs_block_group_flags(&cache->item); |
49303381 LB |
10251 | if (!mixed && |
10252 | ((cache->flags & BTRFS_BLOCK_GROUP_METADATA) && | |
10253 | (cache->flags & BTRFS_BLOCK_GROUP_DATA))) { | |
10254 | btrfs_err(info, | |
10255 | "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups", | |
10256 | cache->key.objectid); | |
e06e89fe | 10257 | btrfs_put_block_group(cache); |
49303381 LB |
10258 | ret = -EINVAL; |
10259 | goto error; | |
10260 | } | |
0b86a832 | 10261 | |
9078a3e1 | 10262 | key.objectid = found_key.objectid + found_key.offset; |
b3b4aa74 | 10263 | btrfs_release_path(path); |
34d52cb6 | 10264 | |
3c14874a JB |
10265 | /* |
10266 | * We need to exclude the super stripes now so that the space | |
10267 | * info has super bytes accounted for, otherwise we'll think | |
10268 | * we have more space than we actually do. | |
10269 | */ | |
2ff7e61e | 10270 | ret = exclude_super_stripes(info, cache); |
835d974f JB |
10271 | if (ret) { |
10272 | /* | |
10273 | * We may have excluded something, so call this just in | |
10274 | * case. | |
10275 | */ | |
2ff7e61e | 10276 | free_excluded_extents(info, cache); |
920e4a58 | 10277 | btrfs_put_block_group(cache); |
835d974f JB |
10278 | goto error; |
10279 | } | |
3c14874a | 10280 | |
817d52f8 JB |
10281 | /* |
10282 | * check for two cases, either we are full, and therefore | |
10283 | * don't need to bother with the caching work since we won't | |
10284 | * find any space, or we are empty, and we can just add all | |
10285 | * the space in and be done with it. This saves us _alot_ of | |
10286 | * time, particularly in the full case. | |
10287 | */ | |
10288 | if (found_key.offset == btrfs_block_group_used(&cache->item)) { | |
11833d66 | 10289 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 10290 | cache->cached = BTRFS_CACHE_FINISHED; |
2ff7e61e | 10291 | free_excluded_extents(info, cache); |
817d52f8 | 10292 | } else if (btrfs_block_group_used(&cache->item) == 0) { |
11833d66 | 10293 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 10294 | cache->cached = BTRFS_CACHE_FINISHED; |
0b246afa | 10295 | add_new_free_space(cache, info, |
817d52f8 JB |
10296 | found_key.objectid, |
10297 | found_key.objectid + | |
10298 | found_key.offset); | |
2ff7e61e | 10299 | free_excluded_extents(info, cache); |
817d52f8 | 10300 | } |
96b5179d | 10301 | |
0b246afa | 10302 | ret = btrfs_add_block_group_cache(info, cache); |
8c579fe7 JB |
10303 | if (ret) { |
10304 | btrfs_remove_free_space_cache(cache); | |
10305 | btrfs_put_block_group(cache); | |
10306 | goto error; | |
10307 | } | |
10308 | ||
0b246afa | 10309 | trace_btrfs_add_block_group(info, cache, 0); |
d2006e6d NB |
10310 | update_space_info(info, cache->flags, found_key.offset, |
10311 | btrfs_block_group_used(&cache->item), | |
10312 | cache->bytes_super, &space_info); | |
8c579fe7 | 10313 | |
6324fbf3 | 10314 | cache->space_info = space_info; |
1b2da372 | 10315 | |
b742bb82 | 10316 | __link_block_group(space_info, cache); |
0f9dd46c | 10317 | |
0b246afa | 10318 | set_avail_alloc_bits(info, cache->flags); |
2ff7e61e | 10319 | if (btrfs_chunk_readonly(info, cache->key.objectid)) { |
868f401a | 10320 | inc_block_group_ro(cache, 1); |
47ab2a6c JB |
10321 | } else if (btrfs_block_group_used(&cache->item) == 0) { |
10322 | spin_lock(&info->unused_bgs_lock); | |
10323 | /* Should always be true but just in case. */ | |
10324 | if (list_empty(&cache->bg_list)) { | |
10325 | btrfs_get_block_group(cache); | |
10326 | list_add_tail(&cache->bg_list, | |
10327 | &info->unused_bgs); | |
10328 | } | |
10329 | spin_unlock(&info->unused_bgs_lock); | |
10330 | } | |
9078a3e1 | 10331 | } |
b742bb82 | 10332 | |
0b246afa | 10333 | list_for_each_entry_rcu(space_info, &info->space_info, list) { |
2ff7e61e | 10334 | if (!(get_alloc_profile(info, space_info->flags) & |
b742bb82 YZ |
10335 | (BTRFS_BLOCK_GROUP_RAID10 | |
10336 | BTRFS_BLOCK_GROUP_RAID1 | | |
53b381b3 DW |
10337 | BTRFS_BLOCK_GROUP_RAID5 | |
10338 | BTRFS_BLOCK_GROUP_RAID6 | | |
b742bb82 YZ |
10339 | BTRFS_BLOCK_GROUP_DUP))) |
10340 | continue; | |
10341 | /* | |
10342 | * avoid allocating from un-mirrored block group if there are | |
10343 | * mirrored block groups. | |
10344 | */ | |
1095cc0d | 10345 | list_for_each_entry(cache, |
10346 | &space_info->block_groups[BTRFS_RAID_RAID0], | |
10347 | list) | |
868f401a | 10348 | inc_block_group_ro(cache, 1); |
1095cc0d | 10349 | list_for_each_entry(cache, |
10350 | &space_info->block_groups[BTRFS_RAID_SINGLE], | |
10351 | list) | |
868f401a | 10352 | inc_block_group_ro(cache, 1); |
9078a3e1 | 10353 | } |
f0486c68 YZ |
10354 | |
10355 | init_global_block_rsv(info); | |
895586ec | 10356 | ret = check_chunk_block_group_mappings(info); |
0b86a832 | 10357 | error: |
9078a3e1 | 10358 | btrfs_free_path(path); |
0b86a832 | 10359 | return ret; |
9078a3e1 | 10360 | } |
6324fbf3 | 10361 | |
ea658bad | 10362 | void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans, |
2ff7e61e | 10363 | struct btrfs_fs_info *fs_info) |
ea658bad | 10364 | { |
0ae66cf1 | 10365 | struct btrfs_block_group_cache *block_group; |
0b246afa | 10366 | struct btrfs_root *extent_root = fs_info->extent_root; |
ea658bad JB |
10367 | struct btrfs_block_group_item item; |
10368 | struct btrfs_key key; | |
10369 | int ret = 0; | |
d9a0540a | 10370 | bool can_flush_pending_bgs = trans->can_flush_pending_bgs; |
ea658bad | 10371 | |
d9a0540a | 10372 | trans->can_flush_pending_bgs = false; |
0ae66cf1 JB |
10373 | while (!list_empty(&trans->new_bgs)) { |
10374 | block_group = list_first_entry(&trans->new_bgs, | |
10375 | struct btrfs_block_group_cache, | |
10376 | bg_list); | |
ea658bad | 10377 | if (ret) |
c92f6be3 | 10378 | goto next; |
ea658bad JB |
10379 | |
10380 | spin_lock(&block_group->lock); | |
10381 | memcpy(&item, &block_group->item, sizeof(item)); | |
10382 | memcpy(&key, &block_group->key, sizeof(key)); | |
10383 | spin_unlock(&block_group->lock); | |
10384 | ||
10385 | ret = btrfs_insert_item(trans, extent_root, &key, &item, | |
10386 | sizeof(item)); | |
10387 | if (ret) | |
66642832 | 10388 | btrfs_abort_transaction(trans, ret); |
0b246afa JM |
10389 | ret = btrfs_finish_chunk_alloc(trans, fs_info, key.objectid, |
10390 | key.offset); | |
6df9a95e | 10391 | if (ret) |
66642832 | 10392 | btrfs_abort_transaction(trans, ret); |
0b246afa | 10393 | add_block_group_free_space(trans, fs_info, block_group); |
1e144fb8 | 10394 | /* already aborted the transaction if it failed. */ |
c92f6be3 FM |
10395 | next: |
10396 | list_del_init(&block_group->bg_list); | |
ea658bad | 10397 | } |
d9a0540a | 10398 | trans->can_flush_pending_bgs = can_flush_pending_bgs; |
ea658bad JB |
10399 | } |
10400 | ||
6324fbf3 | 10401 | int btrfs_make_block_group(struct btrfs_trans_handle *trans, |
2ff7e61e | 10402 | struct btrfs_fs_info *fs_info, u64 bytes_used, |
0174484d | 10403 | u64 type, u64 chunk_offset, u64 size) |
6324fbf3 | 10404 | { |
6324fbf3 | 10405 | struct btrfs_block_group_cache *cache; |
0b246afa | 10406 | int ret; |
6324fbf3 | 10407 | |
0b246afa | 10408 | btrfs_set_log_full_commit(fs_info, trans); |
e02119d5 | 10409 | |
2ff7e61e | 10410 | cache = btrfs_create_block_group_cache(fs_info, chunk_offset, size); |
0f9dd46c JB |
10411 | if (!cache) |
10412 | return -ENOMEM; | |
34d52cb6 | 10413 | |
6324fbf3 | 10414 | btrfs_set_block_group_used(&cache->item, bytes_used); |
0174484d NB |
10415 | btrfs_set_block_group_chunk_objectid(&cache->item, |
10416 | BTRFS_FIRST_CHUNK_TREE_OBJECTID); | |
6324fbf3 CM |
10417 | btrfs_set_block_group_flags(&cache->item, type); |
10418 | ||
920e4a58 | 10419 | cache->flags = type; |
11833d66 | 10420 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 10421 | cache->cached = BTRFS_CACHE_FINISHED; |
1e144fb8 | 10422 | cache->needs_free_space = 1; |
2ff7e61e | 10423 | ret = exclude_super_stripes(fs_info, cache); |
835d974f JB |
10424 | if (ret) { |
10425 | /* | |
10426 | * We may have excluded something, so call this just in | |
10427 | * case. | |
10428 | */ | |
2ff7e61e | 10429 | free_excluded_extents(fs_info, cache); |
920e4a58 | 10430 | btrfs_put_block_group(cache); |
835d974f JB |
10431 | return ret; |
10432 | } | |
96303081 | 10433 | |
0b246afa | 10434 | add_new_free_space(cache, fs_info, chunk_offset, chunk_offset + size); |
817d52f8 | 10435 | |
2ff7e61e | 10436 | free_excluded_extents(fs_info, cache); |
11833d66 | 10437 | |
d0bd4560 | 10438 | #ifdef CONFIG_BTRFS_DEBUG |
2ff7e61e | 10439 | if (btrfs_should_fragment_free_space(cache)) { |
d0bd4560 JB |
10440 | u64 new_bytes_used = size - bytes_used; |
10441 | ||
10442 | bytes_used += new_bytes_used >> 1; | |
2ff7e61e | 10443 | fragment_free_space(cache); |
d0bd4560 JB |
10444 | } |
10445 | #endif | |
2e6e5183 | 10446 | /* |
2be12ef7 NB |
10447 | * Ensure the corresponding space_info object is created and |
10448 | * assigned to our block group. We want our bg to be added to the rbtree | |
10449 | * with its ->space_info set. | |
2e6e5183 | 10450 | */ |
2be12ef7 NB |
10451 | cache->space_info = __find_space_info(fs_info, cache->flags); |
10452 | if (!cache->space_info) { | |
10453 | ret = create_space_info(fs_info, cache->flags, | |
10454 | &cache->space_info); | |
10455 | if (ret) { | |
10456 | btrfs_remove_free_space_cache(cache); | |
10457 | btrfs_put_block_group(cache); | |
10458 | return ret; | |
10459 | } | |
2e6e5183 FM |
10460 | } |
10461 | ||
0b246afa | 10462 | ret = btrfs_add_block_group_cache(fs_info, cache); |
8c579fe7 JB |
10463 | if (ret) { |
10464 | btrfs_remove_free_space_cache(cache); | |
10465 | btrfs_put_block_group(cache); | |
10466 | return ret; | |
10467 | } | |
10468 | ||
2e6e5183 FM |
10469 | /* |
10470 | * Now that our block group has its ->space_info set and is inserted in | |
10471 | * the rbtree, update the space info's counters. | |
10472 | */ | |
0b246afa | 10473 | trace_btrfs_add_block_group(fs_info, cache, 1); |
d2006e6d | 10474 | update_space_info(fs_info, cache->flags, size, bytes_used, |
e40edf2d | 10475 | cache->bytes_super, &cache->space_info); |
0b246afa | 10476 | update_global_block_rsv(fs_info); |
1b2da372 | 10477 | |
b742bb82 | 10478 | __link_block_group(cache->space_info, cache); |
6324fbf3 | 10479 | |
47ab2a6c | 10480 | list_add_tail(&cache->bg_list, &trans->new_bgs); |
6324fbf3 | 10481 | |
0b246afa | 10482 | set_avail_alloc_bits(fs_info, type); |
6324fbf3 CM |
10483 | return 0; |
10484 | } | |
1a40e23b | 10485 | |
10ea00f5 ID |
10486 | static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
10487 | { | |
899c81ea ID |
10488 | u64 extra_flags = chunk_to_extended(flags) & |
10489 | BTRFS_EXTENDED_PROFILE_MASK; | |
10ea00f5 | 10490 | |
de98ced9 | 10491 | write_seqlock(&fs_info->profiles_lock); |
10ea00f5 ID |
10492 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
10493 | fs_info->avail_data_alloc_bits &= ~extra_flags; | |
10494 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
10495 | fs_info->avail_metadata_alloc_bits &= ~extra_flags; | |
10496 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
10497 | fs_info->avail_system_alloc_bits &= ~extra_flags; | |
de98ced9 | 10498 | write_sequnlock(&fs_info->profiles_lock); |
10ea00f5 ID |
10499 | } |
10500 | ||
1a40e23b | 10501 | int btrfs_remove_block_group(struct btrfs_trans_handle *trans, |
6bccf3ab | 10502 | struct btrfs_fs_info *fs_info, u64 group_start, |
04216820 | 10503 | struct extent_map *em) |
1a40e23b | 10504 | { |
6bccf3ab | 10505 | struct btrfs_root *root = fs_info->extent_root; |
1a40e23b ZY |
10506 | struct btrfs_path *path; |
10507 | struct btrfs_block_group_cache *block_group; | |
44fb5511 | 10508 | struct btrfs_free_cluster *cluster; |
0b246afa | 10509 | struct btrfs_root *tree_root = fs_info->tree_root; |
1a40e23b | 10510 | struct btrfs_key key; |
0af3d00b | 10511 | struct inode *inode; |
c1895442 | 10512 | struct kobject *kobj = NULL; |
1a40e23b | 10513 | int ret; |
10ea00f5 | 10514 | int index; |
89a55897 | 10515 | int factor; |
4f69cb98 | 10516 | struct btrfs_caching_control *caching_ctl = NULL; |
04216820 | 10517 | bool remove_em; |
1a40e23b | 10518 | |
6bccf3ab | 10519 | block_group = btrfs_lookup_block_group(fs_info, group_start); |
1a40e23b | 10520 | BUG_ON(!block_group); |
c146afad | 10521 | BUG_ON(!block_group->ro); |
1a40e23b | 10522 | |
9f7c43c9 | 10523 | /* |
10524 | * Free the reserved super bytes from this block group before | |
10525 | * remove it. | |
10526 | */ | |
2ff7e61e | 10527 | free_excluded_extents(fs_info, block_group); |
9f7c43c9 | 10528 | |
1a40e23b | 10529 | memcpy(&key, &block_group->key, sizeof(key)); |
10ea00f5 | 10530 | index = get_block_group_index(block_group); |
89a55897 JB |
10531 | if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP | |
10532 | BTRFS_BLOCK_GROUP_RAID1 | | |
10533 | BTRFS_BLOCK_GROUP_RAID10)) | |
10534 | factor = 2; | |
10535 | else | |
10536 | factor = 1; | |
1a40e23b | 10537 | |
44fb5511 | 10538 | /* make sure this block group isn't part of an allocation cluster */ |
0b246afa | 10539 | cluster = &fs_info->data_alloc_cluster; |
44fb5511 CM |
10540 | spin_lock(&cluster->refill_lock); |
10541 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
10542 | spin_unlock(&cluster->refill_lock); | |
10543 | ||
10544 | /* | |
10545 | * make sure this block group isn't part of a metadata | |
10546 | * allocation cluster | |
10547 | */ | |
0b246afa | 10548 | cluster = &fs_info->meta_alloc_cluster; |
44fb5511 CM |
10549 | spin_lock(&cluster->refill_lock); |
10550 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
10551 | spin_unlock(&cluster->refill_lock); | |
10552 | ||
1a40e23b | 10553 | path = btrfs_alloc_path(); |
d8926bb3 MF |
10554 | if (!path) { |
10555 | ret = -ENOMEM; | |
b74639ca | 10556 | goto out; |
d8926bb3 | 10557 | } |
1a40e23b | 10558 | |
1bbc621e CM |
10559 | /* |
10560 | * get the inode first so any iput calls done for the io_list | |
10561 | * aren't the final iput (no unlinks allowed now) | |
10562 | */ | |
77ab86bf | 10563 | inode = lookup_free_space_inode(fs_info, block_group, path); |
1bbc621e CM |
10564 | |
10565 | mutex_lock(&trans->transaction->cache_write_mutex); | |
10566 | /* | |
10567 | * make sure our free spache cache IO is done before remove the | |
10568 | * free space inode | |
10569 | */ | |
10570 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
10571 | if (!list_empty(&block_group->io_list)) { | |
10572 | list_del_init(&block_group->io_list); | |
10573 | ||
10574 | WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode); | |
10575 | ||
10576 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
afdb5718 | 10577 | btrfs_wait_cache_io(trans, block_group, path); |
1bbc621e CM |
10578 | btrfs_put_block_group(block_group); |
10579 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
10580 | } | |
10581 | ||
10582 | if (!list_empty(&block_group->dirty_list)) { | |
10583 | list_del_init(&block_group->dirty_list); | |
10584 | btrfs_put_block_group(block_group); | |
10585 | } | |
10586 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
10587 | mutex_unlock(&trans->transaction->cache_write_mutex); | |
10588 | ||
0af3d00b | 10589 | if (!IS_ERR(inode)) { |
73f2e545 | 10590 | ret = btrfs_orphan_add(trans, BTRFS_I(inode)); |
79787eaa JM |
10591 | if (ret) { |
10592 | btrfs_add_delayed_iput(inode); | |
b74639ca | 10593 | goto out; |
79787eaa | 10594 | } |
0af3d00b JB |
10595 | clear_nlink(inode); |
10596 | /* One for the block groups ref */ | |
10597 | spin_lock(&block_group->lock); | |
10598 | if (block_group->iref) { | |
10599 | block_group->iref = 0; | |
10600 | block_group->inode = NULL; | |
10601 | spin_unlock(&block_group->lock); | |
10602 | iput(inode); | |
10603 | } else { | |
10604 | spin_unlock(&block_group->lock); | |
10605 | } | |
10606 | /* One for our lookup ref */ | |
455757c3 | 10607 | btrfs_add_delayed_iput(inode); |
0af3d00b JB |
10608 | } |
10609 | ||
10610 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
10611 | key.offset = block_group->key.objectid; | |
10612 | key.type = 0; | |
10613 | ||
10614 | ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1); | |
10615 | if (ret < 0) | |
b74639ca | 10616 | goto out; |
0af3d00b | 10617 | if (ret > 0) |
b3b4aa74 | 10618 | btrfs_release_path(path); |
0af3d00b JB |
10619 | if (ret == 0) { |
10620 | ret = btrfs_del_item(trans, tree_root, path); | |
10621 | if (ret) | |
b74639ca | 10622 | goto out; |
b3b4aa74 | 10623 | btrfs_release_path(path); |
0af3d00b JB |
10624 | } |
10625 | ||
0b246afa | 10626 | spin_lock(&fs_info->block_group_cache_lock); |
1a40e23b | 10627 | rb_erase(&block_group->cache_node, |
0b246afa | 10628 | &fs_info->block_group_cache_tree); |
292cbd51 | 10629 | RB_CLEAR_NODE(&block_group->cache_node); |
a1897fdd | 10630 | |
b74639ca FM |
10631 | /* Once for the block groups rbtree */ |
10632 | btrfs_put_block_group(block_group); | |
10633 | ||
0b246afa JM |
10634 | if (fs_info->first_logical_byte == block_group->key.objectid) |
10635 | fs_info->first_logical_byte = (u64)-1; | |
10636 | spin_unlock(&fs_info->block_group_cache_lock); | |
817d52f8 | 10637 | |
80eb234a | 10638 | down_write(&block_group->space_info->groups_sem); |
44fb5511 CM |
10639 | /* |
10640 | * we must use list_del_init so people can check to see if they | |
10641 | * are still on the list after taking the semaphore | |
10642 | */ | |
10643 | list_del_init(&block_group->list); | |
6ab0a202 | 10644 | if (list_empty(&block_group->space_info->block_groups[index])) { |
c1895442 JM |
10645 | kobj = block_group->space_info->block_group_kobjs[index]; |
10646 | block_group->space_info->block_group_kobjs[index] = NULL; | |
0b246afa | 10647 | clear_avail_alloc_bits(fs_info, block_group->flags); |
6ab0a202 | 10648 | } |
80eb234a | 10649 | up_write(&block_group->space_info->groups_sem); |
c1895442 JM |
10650 | if (kobj) { |
10651 | kobject_del(kobj); | |
10652 | kobject_put(kobj); | |
10653 | } | |
1a40e23b | 10654 | |
4f69cb98 FM |
10655 | if (block_group->has_caching_ctl) |
10656 | caching_ctl = get_caching_control(block_group); | |
817d52f8 | 10657 | if (block_group->cached == BTRFS_CACHE_STARTED) |
11833d66 | 10658 | wait_block_group_cache_done(block_group); |
4f69cb98 | 10659 | if (block_group->has_caching_ctl) { |
0b246afa | 10660 | down_write(&fs_info->commit_root_sem); |
4f69cb98 FM |
10661 | if (!caching_ctl) { |
10662 | struct btrfs_caching_control *ctl; | |
10663 | ||
10664 | list_for_each_entry(ctl, | |
0b246afa | 10665 | &fs_info->caching_block_groups, list) |
4f69cb98 FM |
10666 | if (ctl->block_group == block_group) { |
10667 | caching_ctl = ctl; | |
1e4f4714 | 10668 | refcount_inc(&caching_ctl->count); |
4f69cb98 FM |
10669 | break; |
10670 | } | |
10671 | } | |
10672 | if (caching_ctl) | |
10673 | list_del_init(&caching_ctl->list); | |
0b246afa | 10674 | up_write(&fs_info->commit_root_sem); |
4f69cb98 FM |
10675 | if (caching_ctl) { |
10676 | /* Once for the caching bgs list and once for us. */ | |
10677 | put_caching_control(caching_ctl); | |
10678 | put_caching_control(caching_ctl); | |
10679 | } | |
10680 | } | |
817d52f8 | 10681 | |
ce93ec54 JB |
10682 | spin_lock(&trans->transaction->dirty_bgs_lock); |
10683 | if (!list_empty(&block_group->dirty_list)) { | |
1bbc621e CM |
10684 | WARN_ON(1); |
10685 | } | |
10686 | if (!list_empty(&block_group->io_list)) { | |
10687 | WARN_ON(1); | |
ce93ec54 JB |
10688 | } |
10689 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
817d52f8 JB |
10690 | btrfs_remove_free_space_cache(block_group); |
10691 | ||
c146afad | 10692 | spin_lock(&block_group->space_info->lock); |
75c68e9f | 10693 | list_del_init(&block_group->ro_list); |
18d018ad | 10694 | |
0b246afa | 10695 | if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
18d018ad ZL |
10696 | WARN_ON(block_group->space_info->total_bytes |
10697 | < block_group->key.offset); | |
10698 | WARN_ON(block_group->space_info->bytes_readonly | |
10699 | < block_group->key.offset); | |
10700 | WARN_ON(block_group->space_info->disk_total | |
10701 | < block_group->key.offset * factor); | |
10702 | } | |
c146afad YZ |
10703 | block_group->space_info->total_bytes -= block_group->key.offset; |
10704 | block_group->space_info->bytes_readonly -= block_group->key.offset; | |
89a55897 | 10705 | block_group->space_info->disk_total -= block_group->key.offset * factor; |
18d018ad | 10706 | |
c146afad | 10707 | spin_unlock(&block_group->space_info->lock); |
283bb197 | 10708 | |
0af3d00b JB |
10709 | memcpy(&key, &block_group->key, sizeof(key)); |
10710 | ||
34441361 | 10711 | mutex_lock(&fs_info->chunk_mutex); |
495e64f4 FM |
10712 | if (!list_empty(&em->list)) { |
10713 | /* We're in the transaction->pending_chunks list. */ | |
10714 | free_extent_map(em); | |
10715 | } | |
04216820 FM |
10716 | spin_lock(&block_group->lock); |
10717 | block_group->removed = 1; | |
10718 | /* | |
10719 | * At this point trimming can't start on this block group, because we | |
10720 | * removed the block group from the tree fs_info->block_group_cache_tree | |
10721 | * so no one can't find it anymore and even if someone already got this | |
10722 | * block group before we removed it from the rbtree, they have already | |
10723 | * incremented block_group->trimming - if they didn't, they won't find | |
10724 | * any free space entries because we already removed them all when we | |
10725 | * called btrfs_remove_free_space_cache(). | |
10726 | * | |
10727 | * And we must not remove the extent map from the fs_info->mapping_tree | |
10728 | * to prevent the same logical address range and physical device space | |
10729 | * ranges from being reused for a new block group. This is because our | |
10730 | * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is | |
10731 | * completely transactionless, so while it is trimming a range the | |
10732 | * currently running transaction might finish and a new one start, | |
10733 | * allowing for new block groups to be created that can reuse the same | |
10734 | * physical device locations unless we take this special care. | |
e33e17ee JM |
10735 | * |
10736 | * There may also be an implicit trim operation if the file system | |
10737 | * is mounted with -odiscard. The same protections must remain | |
10738 | * in place until the extents have been discarded completely when | |
10739 | * the transaction commit has completed. | |
04216820 FM |
10740 | */ |
10741 | remove_em = (atomic_read(&block_group->trimming) == 0); | |
10742 | /* | |
10743 | * Make sure a trimmer task always sees the em in the pinned_chunks list | |
10744 | * if it sees block_group->removed == 1 (needs to lock block_group->lock | |
10745 | * before checking block_group->removed). | |
10746 | */ | |
10747 | if (!remove_em) { | |
10748 | /* | |
10749 | * Our em might be in trans->transaction->pending_chunks which | |
10750 | * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks), | |
10751 | * and so is the fs_info->pinned_chunks list. | |
10752 | * | |
10753 | * So at this point we must be holding the chunk_mutex to avoid | |
10754 | * any races with chunk allocation (more specifically at | |
10755 | * volumes.c:contains_pending_extent()), to ensure it always | |
10756 | * sees the em, either in the pending_chunks list or in the | |
10757 | * pinned_chunks list. | |
10758 | */ | |
0b246afa | 10759 | list_move_tail(&em->list, &fs_info->pinned_chunks); |
04216820 FM |
10760 | } |
10761 | spin_unlock(&block_group->lock); | |
04216820 FM |
10762 | |
10763 | if (remove_em) { | |
10764 | struct extent_map_tree *em_tree; | |
10765 | ||
0b246afa | 10766 | em_tree = &fs_info->mapping_tree.map_tree; |
04216820 | 10767 | write_lock(&em_tree->lock); |
8dbcd10f FM |
10768 | /* |
10769 | * The em might be in the pending_chunks list, so make sure the | |
10770 | * chunk mutex is locked, since remove_extent_mapping() will | |
10771 | * delete us from that list. | |
10772 | */ | |
04216820 FM |
10773 | remove_extent_mapping(em_tree, em); |
10774 | write_unlock(&em_tree->lock); | |
10775 | /* once for the tree */ | |
10776 | free_extent_map(em); | |
10777 | } | |
10778 | ||
34441361 | 10779 | mutex_unlock(&fs_info->chunk_mutex); |
8dbcd10f | 10780 | |
0b246afa | 10781 | ret = remove_block_group_free_space(trans, fs_info, block_group); |
1e144fb8 | 10782 | if (ret) |
b74639ca | 10783 | goto out; |
1a40e23b ZY |
10784 | |
10785 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
10786 | if (ret > 0) | |
10787 | ret = -EIO; | |
10788 | if (ret < 0) | |
10789 | goto out; | |
10790 | ||
10791 | ret = btrfs_del_item(trans, root, path); | |
4f7ade41 | 10792 | |
b74639ca | 10793 | out: |
4f7ade41 XY |
10794 | /* Once for the lookup reference */ |
10795 | btrfs_put_block_group(block_group); | |
1a40e23b ZY |
10796 | btrfs_free_path(path); |
10797 | return ret; | |
10798 | } | |
acce952b | 10799 | |
8eab77ff | 10800 | struct btrfs_trans_handle * |
7fd01182 FM |
10801 | btrfs_start_trans_remove_block_group(struct btrfs_fs_info *fs_info, |
10802 | const u64 chunk_offset) | |
8eab77ff | 10803 | { |
7fd01182 FM |
10804 | struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree; |
10805 | struct extent_map *em; | |
10806 | struct map_lookup *map; | |
10807 | unsigned int num_items; | |
10808 | ||
10809 | read_lock(&em_tree->lock); | |
10810 | em = lookup_extent_mapping(em_tree, chunk_offset, 1); | |
10811 | read_unlock(&em_tree->lock); | |
10812 | ASSERT(em && em->start == chunk_offset); | |
10813 | ||
8eab77ff | 10814 | /* |
7fd01182 FM |
10815 | * We need to reserve 3 + N units from the metadata space info in order |
10816 | * to remove a block group (done at btrfs_remove_chunk() and at | |
10817 | * btrfs_remove_block_group()), which are used for: | |
10818 | * | |
8eab77ff FM |
10819 | * 1 unit for adding the free space inode's orphan (located in the tree |
10820 | * of tree roots). | |
7fd01182 FM |
10821 | * 1 unit for deleting the block group item (located in the extent |
10822 | * tree). | |
10823 | * 1 unit for deleting the free space item (located in tree of tree | |
10824 | * roots). | |
10825 | * N units for deleting N device extent items corresponding to each | |
10826 | * stripe (located in the device tree). | |
10827 | * | |
10828 | * In order to remove a block group we also need to reserve units in the | |
10829 | * system space info in order to update the chunk tree (update one or | |
10830 | * more device items and remove one chunk item), but this is done at | |
10831 | * btrfs_remove_chunk() through a call to check_system_chunk(). | |
8eab77ff | 10832 | */ |
95617d69 | 10833 | map = em->map_lookup; |
7fd01182 FM |
10834 | num_items = 3 + map->num_stripes; |
10835 | free_extent_map(em); | |
10836 | ||
8eab77ff | 10837 | return btrfs_start_transaction_fallback_global_rsv(fs_info->extent_root, |
7fd01182 | 10838 | num_items, 1); |
8eab77ff FM |
10839 | } |
10840 | ||
47ab2a6c JB |
10841 | /* |
10842 | * Process the unused_bgs list and remove any that don't have any allocated | |
10843 | * space inside of them. | |
10844 | */ | |
10845 | void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info) | |
10846 | { | |
10847 | struct btrfs_block_group_cache *block_group; | |
10848 | struct btrfs_space_info *space_info; | |
47ab2a6c JB |
10849 | struct btrfs_trans_handle *trans; |
10850 | int ret = 0; | |
10851 | ||
afcdd129 | 10852 | if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags)) |
47ab2a6c JB |
10853 | return; |
10854 | ||
10855 | spin_lock(&fs_info->unused_bgs_lock); | |
10856 | while (!list_empty(&fs_info->unused_bgs)) { | |
10857 | u64 start, end; | |
e33e17ee | 10858 | int trimming; |
47ab2a6c JB |
10859 | |
10860 | block_group = list_first_entry(&fs_info->unused_bgs, | |
10861 | struct btrfs_block_group_cache, | |
10862 | bg_list); | |
47ab2a6c | 10863 | list_del_init(&block_group->bg_list); |
aefbe9a6 ZL |
10864 | |
10865 | space_info = block_group->space_info; | |
10866 | ||
47ab2a6c JB |
10867 | if (ret || btrfs_mixed_space_info(space_info)) { |
10868 | btrfs_put_block_group(block_group); | |
10869 | continue; | |
10870 | } | |
10871 | spin_unlock(&fs_info->unused_bgs_lock); | |
10872 | ||
d5f2e33b | 10873 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
67c5e7d4 | 10874 | |
47ab2a6c JB |
10875 | /* Don't want to race with allocators so take the groups_sem */ |
10876 | down_write(&space_info->groups_sem); | |
10877 | spin_lock(&block_group->lock); | |
33f4c5c6 | 10878 | if (block_group->reserved || block_group->pinned || |
47ab2a6c | 10879 | btrfs_block_group_used(&block_group->item) || |
19c4d2f9 | 10880 | block_group->ro || |
aefbe9a6 | 10881 | list_is_singular(&block_group->list)) { |
47ab2a6c JB |
10882 | /* |
10883 | * We want to bail if we made new allocations or have | |
10884 | * outstanding allocations in this block group. We do | |
10885 | * the ro check in case balance is currently acting on | |
10886 | * this block group. | |
10887 | */ | |
10888 | spin_unlock(&block_group->lock); | |
10889 | up_write(&space_info->groups_sem); | |
10890 | goto next; | |
10891 | } | |
10892 | spin_unlock(&block_group->lock); | |
10893 | ||
10894 | /* We don't want to force the issue, only flip if it's ok. */ | |
868f401a | 10895 | ret = inc_block_group_ro(block_group, 0); |
47ab2a6c JB |
10896 | up_write(&space_info->groups_sem); |
10897 | if (ret < 0) { | |
10898 | ret = 0; | |
10899 | goto next; | |
10900 | } | |
10901 | ||
10902 | /* | |
10903 | * Want to do this before we do anything else so we can recover | |
10904 | * properly if we fail to join the transaction. | |
10905 | */ | |
7fd01182 FM |
10906 | trans = btrfs_start_trans_remove_block_group(fs_info, |
10907 | block_group->key.objectid); | |
47ab2a6c | 10908 | if (IS_ERR(trans)) { |
2ff7e61e | 10909 | btrfs_dec_block_group_ro(block_group); |
47ab2a6c JB |
10910 | ret = PTR_ERR(trans); |
10911 | goto next; | |
10912 | } | |
10913 | ||
10914 | /* | |
10915 | * We could have pending pinned extents for this block group, | |
10916 | * just delete them, we don't care about them anymore. | |
10917 | */ | |
10918 | start = block_group->key.objectid; | |
10919 | end = start + block_group->key.offset - 1; | |
d4b450cd FM |
10920 | /* |
10921 | * Hold the unused_bg_unpin_mutex lock to avoid racing with | |
10922 | * btrfs_finish_extent_commit(). If we are at transaction N, | |
10923 | * another task might be running finish_extent_commit() for the | |
10924 | * previous transaction N - 1, and have seen a range belonging | |
10925 | * to the block group in freed_extents[] before we were able to | |
10926 | * clear the whole block group range from freed_extents[]. This | |
10927 | * means that task can lookup for the block group after we | |
10928 | * unpinned it from freed_extents[] and removed it, leading to | |
10929 | * a BUG_ON() at btrfs_unpin_extent_range(). | |
10930 | */ | |
10931 | mutex_lock(&fs_info->unused_bg_unpin_mutex); | |
758eb51e | 10932 | ret = clear_extent_bits(&fs_info->freed_extents[0], start, end, |
91166212 | 10933 | EXTENT_DIRTY); |
758eb51e | 10934 | if (ret) { |
d4b450cd | 10935 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
2ff7e61e | 10936 | btrfs_dec_block_group_ro(block_group); |
758eb51e FM |
10937 | goto end_trans; |
10938 | } | |
10939 | ret = clear_extent_bits(&fs_info->freed_extents[1], start, end, | |
91166212 | 10940 | EXTENT_DIRTY); |
758eb51e | 10941 | if (ret) { |
d4b450cd | 10942 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
2ff7e61e | 10943 | btrfs_dec_block_group_ro(block_group); |
758eb51e FM |
10944 | goto end_trans; |
10945 | } | |
d4b450cd | 10946 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
47ab2a6c JB |
10947 | |
10948 | /* Reset pinned so btrfs_put_block_group doesn't complain */ | |
c30666d4 ZL |
10949 | spin_lock(&space_info->lock); |
10950 | spin_lock(&block_group->lock); | |
10951 | ||
10952 | space_info->bytes_pinned -= block_group->pinned; | |
10953 | space_info->bytes_readonly += block_group->pinned; | |
10954 | percpu_counter_add(&space_info->total_bytes_pinned, | |
10955 | -block_group->pinned); | |
47ab2a6c JB |
10956 | block_group->pinned = 0; |
10957 | ||
c30666d4 ZL |
10958 | spin_unlock(&block_group->lock); |
10959 | spin_unlock(&space_info->lock); | |
10960 | ||
e33e17ee | 10961 | /* DISCARD can flip during remount */ |
0b246afa | 10962 | trimming = btrfs_test_opt(fs_info, DISCARD); |
e33e17ee JM |
10963 | |
10964 | /* Implicit trim during transaction commit. */ | |
10965 | if (trimming) | |
10966 | btrfs_get_block_group_trimming(block_group); | |
10967 | ||
47ab2a6c JB |
10968 | /* |
10969 | * Btrfs_remove_chunk will abort the transaction if things go | |
10970 | * horribly wrong. | |
10971 | */ | |
5b4aacef | 10972 | ret = btrfs_remove_chunk(trans, fs_info, |
47ab2a6c | 10973 | block_group->key.objectid); |
e33e17ee JM |
10974 | |
10975 | if (ret) { | |
10976 | if (trimming) | |
10977 | btrfs_put_block_group_trimming(block_group); | |
10978 | goto end_trans; | |
10979 | } | |
10980 | ||
10981 | /* | |
10982 | * If we're not mounted with -odiscard, we can just forget | |
10983 | * about this block group. Otherwise we'll need to wait | |
10984 | * until transaction commit to do the actual discard. | |
10985 | */ | |
10986 | if (trimming) { | |
348a0013 FM |
10987 | spin_lock(&fs_info->unused_bgs_lock); |
10988 | /* | |
10989 | * A concurrent scrub might have added us to the list | |
10990 | * fs_info->unused_bgs, so use a list_move operation | |
10991 | * to add the block group to the deleted_bgs list. | |
10992 | */ | |
e33e17ee JM |
10993 | list_move(&block_group->bg_list, |
10994 | &trans->transaction->deleted_bgs); | |
348a0013 | 10995 | spin_unlock(&fs_info->unused_bgs_lock); |
e33e17ee JM |
10996 | btrfs_get_block_group(block_group); |
10997 | } | |
758eb51e | 10998 | end_trans: |
3a45bb20 | 10999 | btrfs_end_transaction(trans); |
47ab2a6c | 11000 | next: |
d5f2e33b | 11001 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
47ab2a6c JB |
11002 | btrfs_put_block_group(block_group); |
11003 | spin_lock(&fs_info->unused_bgs_lock); | |
11004 | } | |
11005 | spin_unlock(&fs_info->unused_bgs_lock); | |
11006 | } | |
11007 | ||
c59021f8 | 11008 | int btrfs_init_space_info(struct btrfs_fs_info *fs_info) |
11009 | { | |
11010 | struct btrfs_space_info *space_info; | |
1aba86d6 | 11011 | struct btrfs_super_block *disk_super; |
11012 | u64 features; | |
11013 | u64 flags; | |
11014 | int mixed = 0; | |
c59021f8 | 11015 | int ret; |
11016 | ||
6c41761f | 11017 | disk_super = fs_info->super_copy; |
1aba86d6 | 11018 | if (!btrfs_super_root(disk_super)) |
0dc924c5 | 11019 | return -EINVAL; |
c59021f8 | 11020 | |
1aba86d6 | 11021 | features = btrfs_super_incompat_flags(disk_super); |
11022 | if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
11023 | mixed = 1; | |
c59021f8 | 11024 | |
1aba86d6 | 11025 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
2be12ef7 | 11026 | ret = create_space_info(fs_info, flags, &space_info); |
c59021f8 | 11027 | if (ret) |
1aba86d6 | 11028 | goto out; |
c59021f8 | 11029 | |
1aba86d6 | 11030 | if (mixed) { |
11031 | flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA; | |
2be12ef7 | 11032 | ret = create_space_info(fs_info, flags, &space_info); |
1aba86d6 | 11033 | } else { |
11034 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
2be12ef7 | 11035 | ret = create_space_info(fs_info, flags, &space_info); |
1aba86d6 | 11036 | if (ret) |
11037 | goto out; | |
11038 | ||
11039 | flags = BTRFS_BLOCK_GROUP_DATA; | |
2be12ef7 | 11040 | ret = create_space_info(fs_info, flags, &space_info); |
1aba86d6 | 11041 | } |
11042 | out: | |
c59021f8 | 11043 | return ret; |
11044 | } | |
11045 | ||
2ff7e61e JM |
11046 | int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info, |
11047 | u64 start, u64 end) | |
acce952b | 11048 | { |
2ff7e61e | 11049 | return unpin_extent_range(fs_info, start, end, false); |
acce952b | 11050 | } |
11051 | ||
499f377f JM |
11052 | /* |
11053 | * It used to be that old block groups would be left around forever. | |
11054 | * Iterating over them would be enough to trim unused space. Since we | |
11055 | * now automatically remove them, we also need to iterate over unallocated | |
11056 | * space. | |
11057 | * | |
11058 | * We don't want a transaction for this since the discard may take a | |
11059 | * substantial amount of time. We don't require that a transaction be | |
11060 | * running, but we do need to take a running transaction into account | |
11061 | * to ensure that we're not discarding chunks that were released in | |
11062 | * the current transaction. | |
11063 | * | |
11064 | * Holding the chunks lock will prevent other threads from allocating | |
11065 | * or releasing chunks, but it won't prevent a running transaction | |
11066 | * from committing and releasing the memory that the pending chunks | |
11067 | * list head uses. For that, we need to take a reference to the | |
11068 | * transaction. | |
11069 | */ | |
11070 | static int btrfs_trim_free_extents(struct btrfs_device *device, | |
6250c25a | 11071 | u64 minlen, u64 *trimmed) |
499f377f | 11072 | { |
6250c25a | 11073 | u64 start = 0, len = 0; |
499f377f JM |
11074 | int ret; |
11075 | ||
11076 | *trimmed = 0; | |
11077 | ||
da05e937 JM |
11078 | /* Discard not supported = nothing to do. */ |
11079 | if (!blk_queue_discard(bdev_get_queue(device->bdev))) | |
11080 | return 0; | |
11081 | ||
499f377f JM |
11082 | /* Not writeable = nothing to do. */ |
11083 | if (!device->writeable) | |
11084 | return 0; | |
11085 | ||
11086 | /* No free space = nothing to do. */ | |
11087 | if (device->total_bytes <= device->bytes_used) | |
11088 | return 0; | |
11089 | ||
11090 | ret = 0; | |
11091 | ||
11092 | while (1) { | |
fb456252 | 11093 | struct btrfs_fs_info *fs_info = device->fs_info; |
499f377f JM |
11094 | struct btrfs_transaction *trans; |
11095 | u64 bytes; | |
11096 | ||
11097 | ret = mutex_lock_interruptible(&fs_info->chunk_mutex); | |
11098 | if (ret) | |
11099 | return ret; | |
11100 | ||
11101 | down_read(&fs_info->commit_root_sem); | |
11102 | ||
11103 | spin_lock(&fs_info->trans_lock); | |
11104 | trans = fs_info->running_transaction; | |
11105 | if (trans) | |
9b64f57d | 11106 | refcount_inc(&trans->use_count); |
499f377f JM |
11107 | spin_unlock(&fs_info->trans_lock); |
11108 | ||
6250c25a DS |
11109 | ret = find_free_dev_extent_start(trans, device, minlen, start, |
11110 | &start, &len); | |
499f377f JM |
11111 | if (trans) |
11112 | btrfs_put_transaction(trans); | |
11113 | ||
11114 | if (ret) { | |
11115 | up_read(&fs_info->commit_root_sem); | |
11116 | mutex_unlock(&fs_info->chunk_mutex); | |
11117 | if (ret == -ENOSPC) | |
11118 | ret = 0; | |
11119 | break; | |
11120 | } | |
11121 | ||
11122 | ret = btrfs_issue_discard(device->bdev, start, len, &bytes); | |
11123 | up_read(&fs_info->commit_root_sem); | |
11124 | mutex_unlock(&fs_info->chunk_mutex); | |
11125 | ||
11126 | if (ret) | |
11127 | break; | |
11128 | ||
11129 | start += len; | |
11130 | *trimmed += bytes; | |
11131 | ||
11132 | if (fatal_signal_pending(current)) { | |
11133 | ret = -ERESTARTSYS; | |
11134 | break; | |
11135 | } | |
11136 | ||
11137 | cond_resched(); | |
11138 | } | |
11139 | ||
11140 | return ret; | |
11141 | } | |
11142 | ||
7b23644c QW |
11143 | /* |
11144 | * Trim the whole filesystem by: | |
11145 | * 1) trimming the free space in each block group | |
11146 | * 2) trimming the unallocated space on each device | |
11147 | * | |
11148 | * This will also continue trimming even if a block group or device encounters | |
11149 | * an error. The return value will be the last error, or 0 if nothing bad | |
11150 | * happens. | |
11151 | */ | |
2ff7e61e | 11152 | int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range) |
f7039b1d | 11153 | { |
f7039b1d | 11154 | struct btrfs_block_group_cache *cache = NULL; |
499f377f JM |
11155 | struct btrfs_device *device; |
11156 | struct list_head *devices; | |
f7039b1d LD |
11157 | u64 group_trimmed; |
11158 | u64 start; | |
11159 | u64 end; | |
11160 | u64 trimmed = 0; | |
7b23644c QW |
11161 | u64 bg_failed = 0; |
11162 | u64 dev_failed = 0; | |
11163 | int bg_ret = 0; | |
11164 | int dev_ret = 0; | |
f7039b1d LD |
11165 | int ret = 0; |
11166 | ||
2fdb337b | 11167 | cache = btrfs_lookup_first_block_group(fs_info, range->start); |
7b23644c | 11168 | for (; cache; cache = next_block_group(fs_info, cache)) { |
f7039b1d LD |
11169 | if (cache->key.objectid >= (range->start + range->len)) { |
11170 | btrfs_put_block_group(cache); | |
11171 | break; | |
11172 | } | |
11173 | ||
11174 | start = max(range->start, cache->key.objectid); | |
11175 | end = min(range->start + range->len, | |
11176 | cache->key.objectid + cache->key.offset); | |
11177 | ||
11178 | if (end - start >= range->minlen) { | |
11179 | if (!block_group_cache_done(cache)) { | |
f6373bf3 | 11180 | ret = cache_block_group(cache, 0); |
1be41b78 | 11181 | if (ret) { |
7b23644c QW |
11182 | bg_failed++; |
11183 | bg_ret = ret; | |
11184 | continue; | |
1be41b78 JB |
11185 | } |
11186 | ret = wait_block_group_cache_done(cache); | |
11187 | if (ret) { | |
7b23644c QW |
11188 | bg_failed++; |
11189 | bg_ret = ret; | |
11190 | continue; | |
1be41b78 | 11191 | } |
f7039b1d LD |
11192 | } |
11193 | ret = btrfs_trim_block_group(cache, | |
11194 | &group_trimmed, | |
11195 | start, | |
11196 | end, | |
11197 | range->minlen); | |
11198 | ||
11199 | trimmed += group_trimmed; | |
11200 | if (ret) { | |
7b23644c QW |
11201 | bg_failed++; |
11202 | bg_ret = ret; | |
11203 | continue; | |
f7039b1d LD |
11204 | } |
11205 | } | |
f7039b1d LD |
11206 | } |
11207 | ||
7b23644c QW |
11208 | if (bg_failed) |
11209 | btrfs_warn(fs_info, | |
11210 | "failed to trim %llu block group(s), last error %d", | |
11211 | bg_failed, bg_ret); | |
0b246afa | 11212 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
a7d24c89 JM |
11213 | devices = &fs_info->fs_devices->devices; |
11214 | list_for_each_entry(device, devices, dev_list) { | |
6250c25a DS |
11215 | ret = btrfs_trim_free_extents(device, range->minlen, |
11216 | &group_trimmed); | |
7b23644c QW |
11217 | if (ret) { |
11218 | dev_failed++; | |
11219 | dev_ret = ret; | |
499f377f | 11220 | break; |
7b23644c | 11221 | } |
499f377f JM |
11222 | |
11223 | trimmed += group_trimmed; | |
11224 | } | |
0b246afa | 11225 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
499f377f | 11226 | |
7b23644c QW |
11227 | if (dev_failed) |
11228 | btrfs_warn(fs_info, | |
11229 | "failed to trim %llu device(s), last error %d", | |
11230 | dev_failed, dev_ret); | |
f7039b1d | 11231 | range->len = trimmed; |
7b23644c QW |
11232 | if (bg_ret) |
11233 | return bg_ret; | |
11234 | return dev_ret; | |
f7039b1d | 11235 | } |
8257b2dc MX |
11236 | |
11237 | /* | |
ea14b57f | 11238 | * btrfs_{start,end}_write_no_snapshotting() are similar to |
9ea24bbe FM |
11239 | * mnt_{want,drop}_write(), they are used to prevent some tasks from writing |
11240 | * data into the page cache through nocow before the subvolume is snapshoted, | |
11241 | * but flush the data into disk after the snapshot creation, or to prevent | |
ea14b57f | 11242 | * operations while snapshotting is ongoing and that cause the snapshot to be |
9ea24bbe | 11243 | * inconsistent (writes followed by expanding truncates for example). |
8257b2dc | 11244 | */ |
ea14b57f | 11245 | void btrfs_end_write_no_snapshotting(struct btrfs_root *root) |
8257b2dc MX |
11246 | { |
11247 | percpu_counter_dec(&root->subv_writers->counter); | |
11248 | /* | |
a83342aa | 11249 | * Make sure counter is updated before we wake up waiters. |
8257b2dc MX |
11250 | */ |
11251 | smp_mb(); | |
11252 | if (waitqueue_active(&root->subv_writers->wait)) | |
11253 | wake_up(&root->subv_writers->wait); | |
11254 | } | |
11255 | ||
ea14b57f | 11256 | int btrfs_start_write_no_snapshotting(struct btrfs_root *root) |
8257b2dc | 11257 | { |
ea14b57f | 11258 | if (atomic_read(&root->will_be_snapshotted)) |
8257b2dc MX |
11259 | return 0; |
11260 | ||
11261 | percpu_counter_inc(&root->subv_writers->counter); | |
11262 | /* | |
11263 | * Make sure counter is updated before we check for snapshot creation. | |
11264 | */ | |
11265 | smp_mb(); | |
ea14b57f DS |
11266 | if (atomic_read(&root->will_be_snapshotted)) { |
11267 | btrfs_end_write_no_snapshotting(root); | |
8257b2dc MX |
11268 | return 0; |
11269 | } | |
11270 | return 1; | |
11271 | } | |
0bc19f90 | 11272 | |
ea14b57f | 11273 | static int wait_snapshotting_atomic_t(atomic_t *a) |
0bc19f90 ZL |
11274 | { |
11275 | schedule(); | |
11276 | return 0; | |
11277 | } | |
11278 | ||
11279 | void btrfs_wait_for_snapshot_creation(struct btrfs_root *root) | |
11280 | { | |
11281 | while (true) { | |
11282 | int ret; | |
11283 | ||
ea14b57f | 11284 | ret = btrfs_start_write_no_snapshotting(root); |
0bc19f90 ZL |
11285 | if (ret) |
11286 | break; | |
ea14b57f DS |
11287 | wait_on_atomic_t(&root->will_be_snapshotted, |
11288 | wait_snapshotting_atomic_t, | |
0bc19f90 ZL |
11289 | TASK_UNINTERRUPTIBLE); |
11290 | } | |
11291 | } |