Commit | Line | Data |
---|---|---|
c1d7c514 | 1 | // SPDX-License-Identifier: GPL-2.0 |
6cbd5570 CM |
2 | /* |
3 | * Copyright (C) 2007 Oracle. All rights reserved. | |
6cbd5570 | 4 | */ |
c1d7c514 | 5 | |
ec6b910f | 6 | #include <linux/sched.h> |
f361bf4a | 7 | #include <linux/sched/signal.h> |
edbd8d4e | 8 | #include <linux/pagemap.h> |
ec44a35c | 9 | #include <linux/writeback.h> |
21af804c | 10 | #include <linux/blkdev.h> |
b7a9f29f | 11 | #include <linux/sort.h> |
4184ea7f | 12 | #include <linux/rcupdate.h> |
817d52f8 | 13 | #include <linux/kthread.h> |
5a0e3ad6 | 14 | #include <linux/slab.h> |
dff51cd1 | 15 | #include <linux/ratelimit.h> |
b150a4f1 | 16 | #include <linux/percpu_counter.h> |
69fe2d75 | 17 | #include <linux/lockdep.h> |
9678c543 | 18 | #include <linux/crc32c.h> |
995946dd | 19 | #include "tree-log.h" |
fec577fb CM |
20 | #include "disk-io.h" |
21 | #include "print-tree.h" | |
0b86a832 | 22 | #include "volumes.h" |
53b381b3 | 23 | #include "raid56.h" |
925baedd | 24 | #include "locking.h" |
fa9c0d79 | 25 | #include "free-space-cache.h" |
1e144fb8 | 26 | #include "free-space-tree.h" |
3fed40cc | 27 | #include "math.h" |
6ab0a202 | 28 | #include "sysfs.h" |
fcebe456 | 29 | #include "qgroup.h" |
fd708b81 | 30 | #include "ref-verify.h" |
fec577fb | 31 | |
709c0486 AJ |
32 | #undef SCRAMBLE_DELAYED_REFS |
33 | ||
9e622d6b MX |
34 | /* |
35 | * control flags for do_chunk_alloc's force field | |
0e4f8f88 CM |
36 | * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk |
37 | * if we really need one. | |
38 | * | |
0e4f8f88 CM |
39 | * CHUNK_ALLOC_LIMITED means to only try and allocate one |
40 | * if we have very few chunks already allocated. This is | |
41 | * used as part of the clustering code to help make sure | |
42 | * we have a good pool of storage to cluster in, without | |
43 | * filling the FS with empty chunks | |
44 | * | |
9e622d6b MX |
45 | * CHUNK_ALLOC_FORCE means it must try to allocate one |
46 | * | |
0e4f8f88 CM |
47 | */ |
48 | enum { | |
49 | CHUNK_ALLOC_NO_FORCE = 0, | |
9e622d6b MX |
50 | CHUNK_ALLOC_LIMITED = 1, |
51 | CHUNK_ALLOC_FORCE = 2, | |
0e4f8f88 CM |
52 | }; |
53 | ||
5d4f98a2 | 54 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
e72cb923 NB |
55 | struct btrfs_delayed_ref_node *node, u64 parent, |
56 | u64 root_objectid, u64 owner_objectid, | |
57 | u64 owner_offset, int refs_to_drop, | |
58 | struct btrfs_delayed_extent_op *extra_op); | |
5d4f98a2 YZ |
59 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, |
60 | struct extent_buffer *leaf, | |
61 | struct btrfs_extent_item *ei); | |
62 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
63 | u64 parent, u64 root_objectid, |
64 | u64 flags, u64 owner, u64 offset, | |
65 | struct btrfs_key *ins, int ref_mod); | |
66 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, | |
4e6bd4e0 | 67 | struct btrfs_delayed_ref_node *node, |
21ebfbe7 | 68 | struct btrfs_delayed_extent_op *extent_op); |
01458828 | 69 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, u64 flags, |
698d0082 | 70 | int force); |
11833d66 YZ |
71 | static int find_next_key(struct btrfs_path *path, int level, |
72 | struct btrfs_key *key); | |
ab8d0fc4 JM |
73 | static void dump_space_info(struct btrfs_fs_info *fs_info, |
74 | struct btrfs_space_info *info, u64 bytes, | |
9ed74f2d | 75 | int dump_block_groups); |
5d80366e JB |
76 | static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, |
77 | u64 num_bytes); | |
957780eb JB |
78 | static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info, |
79 | struct btrfs_space_info *space_info, | |
80 | u64 num_bytes); | |
81 | static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info, | |
82 | struct btrfs_space_info *space_info, | |
83 | u64 num_bytes); | |
6a63209f | 84 | |
817d52f8 JB |
85 | static noinline int |
86 | block_group_cache_done(struct btrfs_block_group_cache *cache) | |
87 | { | |
88 | smp_mb(); | |
36cce922 JB |
89 | return cache->cached == BTRFS_CACHE_FINISHED || |
90 | cache->cached == BTRFS_CACHE_ERROR; | |
817d52f8 JB |
91 | } |
92 | ||
0f9dd46c JB |
93 | static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits) |
94 | { | |
95 | return (cache->flags & bits) == bits; | |
96 | } | |
97 | ||
758f2dfc | 98 | void btrfs_get_block_group(struct btrfs_block_group_cache *cache) |
11dfe35a JB |
99 | { |
100 | atomic_inc(&cache->count); | |
101 | } | |
102 | ||
103 | void btrfs_put_block_group(struct btrfs_block_group_cache *cache) | |
104 | { | |
f0486c68 YZ |
105 | if (atomic_dec_and_test(&cache->count)) { |
106 | WARN_ON(cache->pinned > 0); | |
107 | WARN_ON(cache->reserved > 0); | |
0966a7b1 QW |
108 | |
109 | /* | |
110 | * If not empty, someone is still holding mutex of | |
111 | * full_stripe_lock, which can only be released by caller. | |
112 | * And it will definitely cause use-after-free when caller | |
113 | * tries to release full stripe lock. | |
114 | * | |
115 | * No better way to resolve, but only to warn. | |
116 | */ | |
117 | WARN_ON(!RB_EMPTY_ROOT(&cache->full_stripe_locks_root.root)); | |
34d52cb6 | 118 | kfree(cache->free_space_ctl); |
11dfe35a | 119 | kfree(cache); |
f0486c68 | 120 | } |
11dfe35a JB |
121 | } |
122 | ||
0f9dd46c JB |
123 | /* |
124 | * this adds the block group to the fs_info rb tree for the block group | |
125 | * cache | |
126 | */ | |
b2950863 | 127 | static int btrfs_add_block_group_cache(struct btrfs_fs_info *info, |
0f9dd46c JB |
128 | struct btrfs_block_group_cache *block_group) |
129 | { | |
130 | struct rb_node **p; | |
131 | struct rb_node *parent = NULL; | |
132 | struct btrfs_block_group_cache *cache; | |
133 | ||
134 | spin_lock(&info->block_group_cache_lock); | |
135 | p = &info->block_group_cache_tree.rb_node; | |
136 | ||
137 | while (*p) { | |
138 | parent = *p; | |
139 | cache = rb_entry(parent, struct btrfs_block_group_cache, | |
140 | cache_node); | |
141 | if (block_group->key.objectid < cache->key.objectid) { | |
142 | p = &(*p)->rb_left; | |
143 | } else if (block_group->key.objectid > cache->key.objectid) { | |
144 | p = &(*p)->rb_right; | |
145 | } else { | |
146 | spin_unlock(&info->block_group_cache_lock); | |
147 | return -EEXIST; | |
148 | } | |
149 | } | |
150 | ||
151 | rb_link_node(&block_group->cache_node, parent, p); | |
152 | rb_insert_color(&block_group->cache_node, | |
153 | &info->block_group_cache_tree); | |
a1897fdd LB |
154 | |
155 | if (info->first_logical_byte > block_group->key.objectid) | |
156 | info->first_logical_byte = block_group->key.objectid; | |
157 | ||
0f9dd46c JB |
158 | spin_unlock(&info->block_group_cache_lock); |
159 | ||
160 | return 0; | |
161 | } | |
162 | ||
163 | /* | |
164 | * This will return the block group at or after bytenr if contains is 0, else | |
165 | * it will return the block group that contains the bytenr | |
166 | */ | |
167 | static struct btrfs_block_group_cache * | |
168 | block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr, | |
169 | int contains) | |
170 | { | |
171 | struct btrfs_block_group_cache *cache, *ret = NULL; | |
172 | struct rb_node *n; | |
173 | u64 end, start; | |
174 | ||
175 | spin_lock(&info->block_group_cache_lock); | |
176 | n = info->block_group_cache_tree.rb_node; | |
177 | ||
178 | while (n) { | |
179 | cache = rb_entry(n, struct btrfs_block_group_cache, | |
180 | cache_node); | |
181 | end = cache->key.objectid + cache->key.offset - 1; | |
182 | start = cache->key.objectid; | |
183 | ||
184 | if (bytenr < start) { | |
185 | if (!contains && (!ret || start < ret->key.objectid)) | |
186 | ret = cache; | |
187 | n = n->rb_left; | |
188 | } else if (bytenr > start) { | |
189 | if (contains && bytenr <= end) { | |
190 | ret = cache; | |
191 | break; | |
192 | } | |
193 | n = n->rb_right; | |
194 | } else { | |
195 | ret = cache; | |
196 | break; | |
197 | } | |
198 | } | |
a1897fdd | 199 | if (ret) { |
11dfe35a | 200 | btrfs_get_block_group(ret); |
a1897fdd LB |
201 | if (bytenr == 0 && info->first_logical_byte > ret->key.objectid) |
202 | info->first_logical_byte = ret->key.objectid; | |
203 | } | |
0f9dd46c JB |
204 | spin_unlock(&info->block_group_cache_lock); |
205 | ||
206 | return ret; | |
207 | } | |
208 | ||
2ff7e61e | 209 | static int add_excluded_extent(struct btrfs_fs_info *fs_info, |
11833d66 | 210 | u64 start, u64 num_bytes) |
817d52f8 | 211 | { |
11833d66 | 212 | u64 end = start + num_bytes - 1; |
0b246afa | 213 | set_extent_bits(&fs_info->freed_extents[0], |
ceeb0ae7 | 214 | start, end, EXTENT_UPTODATE); |
0b246afa | 215 | set_extent_bits(&fs_info->freed_extents[1], |
ceeb0ae7 | 216 | start, end, EXTENT_UPTODATE); |
11833d66 YZ |
217 | return 0; |
218 | } | |
817d52f8 | 219 | |
9e715da8 | 220 | static void free_excluded_extents(struct btrfs_block_group_cache *cache) |
11833d66 | 221 | { |
9e715da8 | 222 | struct btrfs_fs_info *fs_info = cache->fs_info; |
11833d66 | 223 | u64 start, end; |
817d52f8 | 224 | |
11833d66 YZ |
225 | start = cache->key.objectid; |
226 | end = start + cache->key.offset - 1; | |
227 | ||
0b246afa | 228 | clear_extent_bits(&fs_info->freed_extents[0], |
91166212 | 229 | start, end, EXTENT_UPTODATE); |
0b246afa | 230 | clear_extent_bits(&fs_info->freed_extents[1], |
91166212 | 231 | start, end, EXTENT_UPTODATE); |
817d52f8 JB |
232 | } |
233 | ||
3c4da657 | 234 | static int exclude_super_stripes(struct btrfs_block_group_cache *cache) |
817d52f8 | 235 | { |
3c4da657 | 236 | struct btrfs_fs_info *fs_info = cache->fs_info; |
817d52f8 JB |
237 | u64 bytenr; |
238 | u64 *logical; | |
239 | int stripe_len; | |
240 | int i, nr, ret; | |
241 | ||
06b2331f YZ |
242 | if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) { |
243 | stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid; | |
244 | cache->bytes_super += stripe_len; | |
2ff7e61e | 245 | ret = add_excluded_extent(fs_info, cache->key.objectid, |
06b2331f | 246 | stripe_len); |
835d974f JB |
247 | if (ret) |
248 | return ret; | |
06b2331f YZ |
249 | } |
250 | ||
817d52f8 JB |
251 | for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { |
252 | bytenr = btrfs_sb_offset(i); | |
0b246afa | 253 | ret = btrfs_rmap_block(fs_info, cache->key.objectid, |
63a9c7b9 | 254 | bytenr, &logical, &nr, &stripe_len); |
835d974f JB |
255 | if (ret) |
256 | return ret; | |
11833d66 | 257 | |
817d52f8 | 258 | while (nr--) { |
51bf5f0b JB |
259 | u64 start, len; |
260 | ||
261 | if (logical[nr] > cache->key.objectid + | |
262 | cache->key.offset) | |
263 | continue; | |
264 | ||
265 | if (logical[nr] + stripe_len <= cache->key.objectid) | |
266 | continue; | |
267 | ||
268 | start = logical[nr]; | |
269 | if (start < cache->key.objectid) { | |
270 | start = cache->key.objectid; | |
271 | len = (logical[nr] + stripe_len) - start; | |
272 | } else { | |
273 | len = min_t(u64, stripe_len, | |
274 | cache->key.objectid + | |
275 | cache->key.offset - start); | |
276 | } | |
277 | ||
278 | cache->bytes_super += len; | |
2ff7e61e | 279 | ret = add_excluded_extent(fs_info, start, len); |
835d974f JB |
280 | if (ret) { |
281 | kfree(logical); | |
282 | return ret; | |
283 | } | |
817d52f8 | 284 | } |
11833d66 | 285 | |
817d52f8 JB |
286 | kfree(logical); |
287 | } | |
817d52f8 JB |
288 | return 0; |
289 | } | |
290 | ||
11833d66 YZ |
291 | static struct btrfs_caching_control * |
292 | get_caching_control(struct btrfs_block_group_cache *cache) | |
293 | { | |
294 | struct btrfs_caching_control *ctl; | |
295 | ||
296 | spin_lock(&cache->lock); | |
dde5abee JB |
297 | if (!cache->caching_ctl) { |
298 | spin_unlock(&cache->lock); | |
11833d66 YZ |
299 | return NULL; |
300 | } | |
301 | ||
302 | ctl = cache->caching_ctl; | |
1e4f4714 | 303 | refcount_inc(&ctl->count); |
11833d66 YZ |
304 | spin_unlock(&cache->lock); |
305 | return ctl; | |
306 | } | |
307 | ||
308 | static void put_caching_control(struct btrfs_caching_control *ctl) | |
309 | { | |
1e4f4714 | 310 | if (refcount_dec_and_test(&ctl->count)) |
11833d66 YZ |
311 | kfree(ctl); |
312 | } | |
313 | ||
d0bd4560 | 314 | #ifdef CONFIG_BTRFS_DEBUG |
2ff7e61e | 315 | static void fragment_free_space(struct btrfs_block_group_cache *block_group) |
d0bd4560 | 316 | { |
2ff7e61e | 317 | struct btrfs_fs_info *fs_info = block_group->fs_info; |
d0bd4560 JB |
318 | u64 start = block_group->key.objectid; |
319 | u64 len = block_group->key.offset; | |
320 | u64 chunk = block_group->flags & BTRFS_BLOCK_GROUP_METADATA ? | |
0b246afa | 321 | fs_info->nodesize : fs_info->sectorsize; |
d0bd4560 JB |
322 | u64 step = chunk << 1; |
323 | ||
324 | while (len > chunk) { | |
325 | btrfs_remove_free_space(block_group, start, chunk); | |
326 | start += step; | |
327 | if (len < step) | |
328 | len = 0; | |
329 | else | |
330 | len -= step; | |
331 | } | |
332 | } | |
333 | #endif | |
334 | ||
0f9dd46c JB |
335 | /* |
336 | * this is only called by cache_block_group, since we could have freed extents | |
337 | * we need to check the pinned_extents for any extents that can't be used yet | |
338 | * since their free space will be released as soon as the transaction commits. | |
339 | */ | |
a5ed9182 | 340 | u64 add_new_free_space(struct btrfs_block_group_cache *block_group, |
4457c1c7 | 341 | u64 start, u64 end) |
0f9dd46c | 342 | { |
4457c1c7 | 343 | struct btrfs_fs_info *info = block_group->fs_info; |
817d52f8 | 344 | u64 extent_start, extent_end, size, total_added = 0; |
0f9dd46c JB |
345 | int ret; |
346 | ||
347 | while (start < end) { | |
11833d66 | 348 | ret = find_first_extent_bit(info->pinned_extents, start, |
0f9dd46c | 349 | &extent_start, &extent_end, |
e6138876 JB |
350 | EXTENT_DIRTY | EXTENT_UPTODATE, |
351 | NULL); | |
0f9dd46c JB |
352 | if (ret) |
353 | break; | |
354 | ||
06b2331f | 355 | if (extent_start <= start) { |
0f9dd46c JB |
356 | start = extent_end + 1; |
357 | } else if (extent_start > start && extent_start < end) { | |
358 | size = extent_start - start; | |
817d52f8 | 359 | total_added += size; |
ea6a478e JB |
360 | ret = btrfs_add_free_space(block_group, start, |
361 | size); | |
79787eaa | 362 | BUG_ON(ret); /* -ENOMEM or logic error */ |
0f9dd46c JB |
363 | start = extent_end + 1; |
364 | } else { | |
365 | break; | |
366 | } | |
367 | } | |
368 | ||
369 | if (start < end) { | |
370 | size = end - start; | |
817d52f8 | 371 | total_added += size; |
ea6a478e | 372 | ret = btrfs_add_free_space(block_group, start, size); |
79787eaa | 373 | BUG_ON(ret); /* -ENOMEM or logic error */ |
0f9dd46c JB |
374 | } |
375 | ||
817d52f8 | 376 | return total_added; |
0f9dd46c JB |
377 | } |
378 | ||
73fa48b6 | 379 | static int load_extent_tree_free(struct btrfs_caching_control *caching_ctl) |
e37c9e69 | 380 | { |
0b246afa JM |
381 | struct btrfs_block_group_cache *block_group = caching_ctl->block_group; |
382 | struct btrfs_fs_info *fs_info = block_group->fs_info; | |
383 | struct btrfs_root *extent_root = fs_info->extent_root; | |
e37c9e69 | 384 | struct btrfs_path *path; |
5f39d397 | 385 | struct extent_buffer *leaf; |
11833d66 | 386 | struct btrfs_key key; |
817d52f8 | 387 | u64 total_found = 0; |
11833d66 YZ |
388 | u64 last = 0; |
389 | u32 nritems; | |
73fa48b6 | 390 | int ret; |
d0bd4560 | 391 | bool wakeup = true; |
f510cfec | 392 | |
e37c9e69 CM |
393 | path = btrfs_alloc_path(); |
394 | if (!path) | |
73fa48b6 | 395 | return -ENOMEM; |
7d7d6068 | 396 | |
817d52f8 | 397 | last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET); |
11833d66 | 398 | |
d0bd4560 JB |
399 | #ifdef CONFIG_BTRFS_DEBUG |
400 | /* | |
401 | * If we're fragmenting we don't want to make anybody think we can | |
402 | * allocate from this block group until we've had a chance to fragment | |
403 | * the free space. | |
404 | */ | |
2ff7e61e | 405 | if (btrfs_should_fragment_free_space(block_group)) |
d0bd4560 JB |
406 | wakeup = false; |
407 | #endif | |
5cd57b2c | 408 | /* |
817d52f8 JB |
409 | * We don't want to deadlock with somebody trying to allocate a new |
410 | * extent for the extent root while also trying to search the extent | |
411 | * root to add free space. So we skip locking and search the commit | |
412 | * root, since its read-only | |
5cd57b2c CM |
413 | */ |
414 | path->skip_locking = 1; | |
817d52f8 | 415 | path->search_commit_root = 1; |
e4058b54 | 416 | path->reada = READA_FORWARD; |
817d52f8 | 417 | |
e4404d6e | 418 | key.objectid = last; |
e37c9e69 | 419 | key.offset = 0; |
11833d66 | 420 | key.type = BTRFS_EXTENT_ITEM_KEY; |
013f1b12 | 421 | |
52ee28d2 | 422 | next: |
11833d66 | 423 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
e37c9e69 | 424 | if (ret < 0) |
73fa48b6 | 425 | goto out; |
a512bbf8 | 426 | |
11833d66 YZ |
427 | leaf = path->nodes[0]; |
428 | nritems = btrfs_header_nritems(leaf); | |
429 | ||
d397712b | 430 | while (1) { |
7841cb28 | 431 | if (btrfs_fs_closing(fs_info) > 1) { |
f25784b3 | 432 | last = (u64)-1; |
817d52f8 | 433 | break; |
f25784b3 | 434 | } |
817d52f8 | 435 | |
11833d66 YZ |
436 | if (path->slots[0] < nritems) { |
437 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
438 | } else { | |
439 | ret = find_next_key(path, 0, &key); | |
440 | if (ret) | |
e37c9e69 | 441 | break; |
817d52f8 | 442 | |
c9ea7b24 | 443 | if (need_resched() || |
9e351cc8 | 444 | rwsem_is_contended(&fs_info->commit_root_sem)) { |
d0bd4560 JB |
445 | if (wakeup) |
446 | caching_ctl->progress = last; | |
ff5714cc | 447 | btrfs_release_path(path); |
9e351cc8 | 448 | up_read(&fs_info->commit_root_sem); |
589d8ade | 449 | mutex_unlock(&caching_ctl->mutex); |
11833d66 | 450 | cond_resched(); |
73fa48b6 OS |
451 | mutex_lock(&caching_ctl->mutex); |
452 | down_read(&fs_info->commit_root_sem); | |
453 | goto next; | |
589d8ade | 454 | } |
0a3896d0 JB |
455 | |
456 | ret = btrfs_next_leaf(extent_root, path); | |
457 | if (ret < 0) | |
73fa48b6 | 458 | goto out; |
0a3896d0 JB |
459 | if (ret) |
460 | break; | |
589d8ade JB |
461 | leaf = path->nodes[0]; |
462 | nritems = btrfs_header_nritems(leaf); | |
463 | continue; | |
11833d66 | 464 | } |
817d52f8 | 465 | |
52ee28d2 LB |
466 | if (key.objectid < last) { |
467 | key.objectid = last; | |
468 | key.offset = 0; | |
469 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
470 | ||
d0bd4560 JB |
471 | if (wakeup) |
472 | caching_ctl->progress = last; | |
52ee28d2 LB |
473 | btrfs_release_path(path); |
474 | goto next; | |
475 | } | |
476 | ||
11833d66 YZ |
477 | if (key.objectid < block_group->key.objectid) { |
478 | path->slots[0]++; | |
817d52f8 | 479 | continue; |
e37c9e69 | 480 | } |
0f9dd46c | 481 | |
e37c9e69 | 482 | if (key.objectid >= block_group->key.objectid + |
0f9dd46c | 483 | block_group->key.offset) |
e37c9e69 | 484 | break; |
7d7d6068 | 485 | |
3173a18f JB |
486 | if (key.type == BTRFS_EXTENT_ITEM_KEY || |
487 | key.type == BTRFS_METADATA_ITEM_KEY) { | |
4457c1c7 | 488 | total_found += add_new_free_space(block_group, last, |
817d52f8 | 489 | key.objectid); |
3173a18f JB |
490 | if (key.type == BTRFS_METADATA_ITEM_KEY) |
491 | last = key.objectid + | |
da17066c | 492 | fs_info->nodesize; |
3173a18f JB |
493 | else |
494 | last = key.objectid + key.offset; | |
817d52f8 | 495 | |
73fa48b6 | 496 | if (total_found > CACHING_CTL_WAKE_UP) { |
11833d66 | 497 | total_found = 0; |
d0bd4560 JB |
498 | if (wakeup) |
499 | wake_up(&caching_ctl->wait); | |
11833d66 | 500 | } |
817d52f8 | 501 | } |
e37c9e69 CM |
502 | path->slots[0]++; |
503 | } | |
817d52f8 | 504 | ret = 0; |
e37c9e69 | 505 | |
4457c1c7 | 506 | total_found += add_new_free_space(block_group, last, |
817d52f8 JB |
507 | block_group->key.objectid + |
508 | block_group->key.offset); | |
11833d66 | 509 | caching_ctl->progress = (u64)-1; |
817d52f8 | 510 | |
73fa48b6 OS |
511 | out: |
512 | btrfs_free_path(path); | |
513 | return ret; | |
514 | } | |
515 | ||
516 | static noinline void caching_thread(struct btrfs_work *work) | |
517 | { | |
518 | struct btrfs_block_group_cache *block_group; | |
519 | struct btrfs_fs_info *fs_info; | |
520 | struct btrfs_caching_control *caching_ctl; | |
521 | int ret; | |
522 | ||
523 | caching_ctl = container_of(work, struct btrfs_caching_control, work); | |
524 | block_group = caching_ctl->block_group; | |
525 | fs_info = block_group->fs_info; | |
526 | ||
527 | mutex_lock(&caching_ctl->mutex); | |
528 | down_read(&fs_info->commit_root_sem); | |
529 | ||
1e144fb8 OS |
530 | if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) |
531 | ret = load_free_space_tree(caching_ctl); | |
532 | else | |
533 | ret = load_extent_tree_free(caching_ctl); | |
73fa48b6 | 534 | |
817d52f8 | 535 | spin_lock(&block_group->lock); |
11833d66 | 536 | block_group->caching_ctl = NULL; |
73fa48b6 | 537 | block_group->cached = ret ? BTRFS_CACHE_ERROR : BTRFS_CACHE_FINISHED; |
817d52f8 | 538 | spin_unlock(&block_group->lock); |
0f9dd46c | 539 | |
d0bd4560 | 540 | #ifdef CONFIG_BTRFS_DEBUG |
2ff7e61e | 541 | if (btrfs_should_fragment_free_space(block_group)) { |
d0bd4560 JB |
542 | u64 bytes_used; |
543 | ||
544 | spin_lock(&block_group->space_info->lock); | |
545 | spin_lock(&block_group->lock); | |
546 | bytes_used = block_group->key.offset - | |
547 | btrfs_block_group_used(&block_group->item); | |
548 | block_group->space_info->bytes_used += bytes_used >> 1; | |
549 | spin_unlock(&block_group->lock); | |
550 | spin_unlock(&block_group->space_info->lock); | |
2ff7e61e | 551 | fragment_free_space(block_group); |
d0bd4560 JB |
552 | } |
553 | #endif | |
554 | ||
555 | caching_ctl->progress = (u64)-1; | |
11833d66 | 556 | |
9e351cc8 | 557 | up_read(&fs_info->commit_root_sem); |
9e715da8 | 558 | free_excluded_extents(block_group); |
11833d66 | 559 | mutex_unlock(&caching_ctl->mutex); |
73fa48b6 | 560 | |
11833d66 YZ |
561 | wake_up(&caching_ctl->wait); |
562 | ||
563 | put_caching_control(caching_ctl); | |
11dfe35a | 564 | btrfs_put_block_group(block_group); |
817d52f8 JB |
565 | } |
566 | ||
9d66e233 | 567 | static int cache_block_group(struct btrfs_block_group_cache *cache, |
9d66e233 | 568 | int load_cache_only) |
817d52f8 | 569 | { |
291c7d2f | 570 | DEFINE_WAIT(wait); |
11833d66 YZ |
571 | struct btrfs_fs_info *fs_info = cache->fs_info; |
572 | struct btrfs_caching_control *caching_ctl; | |
817d52f8 JB |
573 | int ret = 0; |
574 | ||
291c7d2f | 575 | caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS); |
79787eaa JM |
576 | if (!caching_ctl) |
577 | return -ENOMEM; | |
291c7d2f JB |
578 | |
579 | INIT_LIST_HEAD(&caching_ctl->list); | |
580 | mutex_init(&caching_ctl->mutex); | |
581 | init_waitqueue_head(&caching_ctl->wait); | |
582 | caching_ctl->block_group = cache; | |
583 | caching_ctl->progress = cache->key.objectid; | |
1e4f4714 | 584 | refcount_set(&caching_ctl->count, 1); |
9e0af237 LB |
585 | btrfs_init_work(&caching_ctl->work, btrfs_cache_helper, |
586 | caching_thread, NULL, NULL); | |
291c7d2f JB |
587 | |
588 | spin_lock(&cache->lock); | |
589 | /* | |
590 | * This should be a rare occasion, but this could happen I think in the | |
591 | * case where one thread starts to load the space cache info, and then | |
592 | * some other thread starts a transaction commit which tries to do an | |
593 | * allocation while the other thread is still loading the space cache | |
594 | * info. The previous loop should have kept us from choosing this block | |
595 | * group, but if we've moved to the state where we will wait on caching | |
596 | * block groups we need to first check if we're doing a fast load here, | |
597 | * so we can wait for it to finish, otherwise we could end up allocating | |
598 | * from a block group who's cache gets evicted for one reason or | |
599 | * another. | |
600 | */ | |
601 | while (cache->cached == BTRFS_CACHE_FAST) { | |
602 | struct btrfs_caching_control *ctl; | |
603 | ||
604 | ctl = cache->caching_ctl; | |
1e4f4714 | 605 | refcount_inc(&ctl->count); |
291c7d2f JB |
606 | prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE); |
607 | spin_unlock(&cache->lock); | |
608 | ||
609 | schedule(); | |
610 | ||
611 | finish_wait(&ctl->wait, &wait); | |
612 | put_caching_control(ctl); | |
613 | spin_lock(&cache->lock); | |
614 | } | |
615 | ||
616 | if (cache->cached != BTRFS_CACHE_NO) { | |
617 | spin_unlock(&cache->lock); | |
618 | kfree(caching_ctl); | |
11833d66 | 619 | return 0; |
291c7d2f JB |
620 | } |
621 | WARN_ON(cache->caching_ctl); | |
622 | cache->caching_ctl = caching_ctl; | |
623 | cache->cached = BTRFS_CACHE_FAST; | |
624 | spin_unlock(&cache->lock); | |
11833d66 | 625 | |
d8953d69 | 626 | if (btrfs_test_opt(fs_info, SPACE_CACHE)) { |
cb83b7b8 | 627 | mutex_lock(&caching_ctl->mutex); |
9d66e233 JB |
628 | ret = load_free_space_cache(fs_info, cache); |
629 | ||
630 | spin_lock(&cache->lock); | |
631 | if (ret == 1) { | |
291c7d2f | 632 | cache->caching_ctl = NULL; |
9d66e233 JB |
633 | cache->cached = BTRFS_CACHE_FINISHED; |
634 | cache->last_byte_to_unpin = (u64)-1; | |
cb83b7b8 | 635 | caching_ctl->progress = (u64)-1; |
9d66e233 | 636 | } else { |
291c7d2f JB |
637 | if (load_cache_only) { |
638 | cache->caching_ctl = NULL; | |
639 | cache->cached = BTRFS_CACHE_NO; | |
640 | } else { | |
641 | cache->cached = BTRFS_CACHE_STARTED; | |
4f69cb98 | 642 | cache->has_caching_ctl = 1; |
291c7d2f | 643 | } |
9d66e233 JB |
644 | } |
645 | spin_unlock(&cache->lock); | |
d0bd4560 JB |
646 | #ifdef CONFIG_BTRFS_DEBUG |
647 | if (ret == 1 && | |
2ff7e61e | 648 | btrfs_should_fragment_free_space(cache)) { |
d0bd4560 JB |
649 | u64 bytes_used; |
650 | ||
651 | spin_lock(&cache->space_info->lock); | |
652 | spin_lock(&cache->lock); | |
653 | bytes_used = cache->key.offset - | |
654 | btrfs_block_group_used(&cache->item); | |
655 | cache->space_info->bytes_used += bytes_used >> 1; | |
656 | spin_unlock(&cache->lock); | |
657 | spin_unlock(&cache->space_info->lock); | |
2ff7e61e | 658 | fragment_free_space(cache); |
d0bd4560 JB |
659 | } |
660 | #endif | |
cb83b7b8 JB |
661 | mutex_unlock(&caching_ctl->mutex); |
662 | ||
291c7d2f | 663 | wake_up(&caching_ctl->wait); |
3c14874a | 664 | if (ret == 1) { |
291c7d2f | 665 | put_caching_control(caching_ctl); |
9e715da8 | 666 | free_excluded_extents(cache); |
9d66e233 | 667 | return 0; |
3c14874a | 668 | } |
291c7d2f JB |
669 | } else { |
670 | /* | |
1e144fb8 OS |
671 | * We're either using the free space tree or no caching at all. |
672 | * Set cached to the appropriate value and wakeup any waiters. | |
291c7d2f JB |
673 | */ |
674 | spin_lock(&cache->lock); | |
675 | if (load_cache_only) { | |
676 | cache->caching_ctl = NULL; | |
677 | cache->cached = BTRFS_CACHE_NO; | |
678 | } else { | |
679 | cache->cached = BTRFS_CACHE_STARTED; | |
4f69cb98 | 680 | cache->has_caching_ctl = 1; |
291c7d2f JB |
681 | } |
682 | spin_unlock(&cache->lock); | |
683 | wake_up(&caching_ctl->wait); | |
9d66e233 JB |
684 | } |
685 | ||
291c7d2f JB |
686 | if (load_cache_only) { |
687 | put_caching_control(caching_ctl); | |
11833d66 | 688 | return 0; |
817d52f8 | 689 | } |
817d52f8 | 690 | |
9e351cc8 | 691 | down_write(&fs_info->commit_root_sem); |
1e4f4714 | 692 | refcount_inc(&caching_ctl->count); |
11833d66 | 693 | list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups); |
9e351cc8 | 694 | up_write(&fs_info->commit_root_sem); |
11833d66 | 695 | |
11dfe35a | 696 | btrfs_get_block_group(cache); |
11833d66 | 697 | |
e66f0bb1 | 698 | btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work); |
817d52f8 | 699 | |
ef8bbdfe | 700 | return ret; |
e37c9e69 CM |
701 | } |
702 | ||
0f9dd46c JB |
703 | /* |
704 | * return the block group that starts at or after bytenr | |
705 | */ | |
d397712b CM |
706 | static struct btrfs_block_group_cache * |
707 | btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr) | |
0ef3e66b | 708 | { |
e2c89907 | 709 | return block_group_cache_tree_search(info, bytenr, 0); |
0ef3e66b CM |
710 | } |
711 | ||
0f9dd46c | 712 | /* |
9f55684c | 713 | * return the block group that contains the given bytenr |
0f9dd46c | 714 | */ |
d397712b CM |
715 | struct btrfs_block_group_cache *btrfs_lookup_block_group( |
716 | struct btrfs_fs_info *info, | |
717 | u64 bytenr) | |
be744175 | 718 | { |
e2c89907 | 719 | return block_group_cache_tree_search(info, bytenr, 1); |
be744175 | 720 | } |
0b86a832 | 721 | |
0f9dd46c JB |
722 | static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info, |
723 | u64 flags) | |
6324fbf3 | 724 | { |
0f9dd46c | 725 | struct list_head *head = &info->space_info; |
0f9dd46c | 726 | struct btrfs_space_info *found; |
4184ea7f | 727 | |
52ba6929 | 728 | flags &= BTRFS_BLOCK_GROUP_TYPE_MASK; |
b742bb82 | 729 | |
4184ea7f CM |
730 | rcu_read_lock(); |
731 | list_for_each_entry_rcu(found, head, list) { | |
67377734 | 732 | if (found->flags & flags) { |
4184ea7f | 733 | rcu_read_unlock(); |
0f9dd46c | 734 | return found; |
4184ea7f | 735 | } |
0f9dd46c | 736 | } |
4184ea7f | 737 | rcu_read_unlock(); |
0f9dd46c | 738 | return NULL; |
6324fbf3 CM |
739 | } |
740 | ||
0d9f824d | 741 | static void add_pinned_bytes(struct btrfs_fs_info *fs_info, s64 num_bytes, |
29d2b84c | 742 | bool metadata, u64 root_objectid) |
0d9f824d OS |
743 | { |
744 | struct btrfs_space_info *space_info; | |
745 | u64 flags; | |
746 | ||
29d2b84c | 747 | if (metadata) { |
0d9f824d OS |
748 | if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID) |
749 | flags = BTRFS_BLOCK_GROUP_SYSTEM; | |
750 | else | |
751 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
752 | } else { | |
753 | flags = BTRFS_BLOCK_GROUP_DATA; | |
754 | } | |
755 | ||
756 | space_info = __find_space_info(fs_info, flags); | |
55e8196a | 757 | ASSERT(space_info); |
0d9f824d OS |
758 | percpu_counter_add(&space_info->total_bytes_pinned, num_bytes); |
759 | } | |
760 | ||
4184ea7f CM |
761 | /* |
762 | * after adding space to the filesystem, we need to clear the full flags | |
763 | * on all the space infos. | |
764 | */ | |
765 | void btrfs_clear_space_info_full(struct btrfs_fs_info *info) | |
766 | { | |
767 | struct list_head *head = &info->space_info; | |
768 | struct btrfs_space_info *found; | |
769 | ||
770 | rcu_read_lock(); | |
771 | list_for_each_entry_rcu(found, head, list) | |
772 | found->full = 0; | |
773 | rcu_read_unlock(); | |
774 | } | |
775 | ||
1a4ed8fd | 776 | /* simple helper to search for an existing data extent at a given offset */ |
2ff7e61e | 777 | int btrfs_lookup_data_extent(struct btrfs_fs_info *fs_info, u64 start, u64 len) |
e02119d5 CM |
778 | { |
779 | int ret; | |
780 | struct btrfs_key key; | |
31840ae1 | 781 | struct btrfs_path *path; |
e02119d5 | 782 | |
31840ae1 | 783 | path = btrfs_alloc_path(); |
d8926bb3 MF |
784 | if (!path) |
785 | return -ENOMEM; | |
786 | ||
e02119d5 CM |
787 | key.objectid = start; |
788 | key.offset = len; | |
3173a18f | 789 | key.type = BTRFS_EXTENT_ITEM_KEY; |
0b246afa | 790 | ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0); |
31840ae1 | 791 | btrfs_free_path(path); |
7bb86316 CM |
792 | return ret; |
793 | } | |
794 | ||
a22285a6 | 795 | /* |
3173a18f | 796 | * helper function to lookup reference count and flags of a tree block. |
a22285a6 YZ |
797 | * |
798 | * the head node for delayed ref is used to store the sum of all the | |
799 | * reference count modifications queued up in the rbtree. the head | |
800 | * node may also store the extent flags to set. This way you can check | |
801 | * to see what the reference count and extent flags would be if all of | |
802 | * the delayed refs are not processed. | |
803 | */ | |
804 | int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans, | |
2ff7e61e | 805 | struct btrfs_fs_info *fs_info, u64 bytenr, |
3173a18f | 806 | u64 offset, int metadata, u64 *refs, u64 *flags) |
a22285a6 YZ |
807 | { |
808 | struct btrfs_delayed_ref_head *head; | |
809 | struct btrfs_delayed_ref_root *delayed_refs; | |
810 | struct btrfs_path *path; | |
811 | struct btrfs_extent_item *ei; | |
812 | struct extent_buffer *leaf; | |
813 | struct btrfs_key key; | |
814 | u32 item_size; | |
815 | u64 num_refs; | |
816 | u64 extent_flags; | |
817 | int ret; | |
818 | ||
3173a18f JB |
819 | /* |
820 | * If we don't have skinny metadata, don't bother doing anything | |
821 | * different | |
822 | */ | |
0b246afa JM |
823 | if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) { |
824 | offset = fs_info->nodesize; | |
3173a18f JB |
825 | metadata = 0; |
826 | } | |
827 | ||
a22285a6 YZ |
828 | path = btrfs_alloc_path(); |
829 | if (!path) | |
830 | return -ENOMEM; | |
831 | ||
a22285a6 YZ |
832 | if (!trans) { |
833 | path->skip_locking = 1; | |
834 | path->search_commit_root = 1; | |
835 | } | |
639eefc8 FDBM |
836 | |
837 | search_again: | |
838 | key.objectid = bytenr; | |
839 | key.offset = offset; | |
840 | if (metadata) | |
841 | key.type = BTRFS_METADATA_ITEM_KEY; | |
842 | else | |
843 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
844 | ||
0b246afa | 845 | ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0); |
a22285a6 YZ |
846 | if (ret < 0) |
847 | goto out_free; | |
848 | ||
3173a18f | 849 | if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) { |
74be9510 FDBM |
850 | if (path->slots[0]) { |
851 | path->slots[0]--; | |
852 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
853 | path->slots[0]); | |
854 | if (key.objectid == bytenr && | |
855 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
0b246afa | 856 | key.offset == fs_info->nodesize) |
74be9510 FDBM |
857 | ret = 0; |
858 | } | |
3173a18f JB |
859 | } |
860 | ||
a22285a6 YZ |
861 | if (ret == 0) { |
862 | leaf = path->nodes[0]; | |
863 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
864 | if (item_size >= sizeof(*ei)) { | |
865 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
866 | struct btrfs_extent_item); | |
867 | num_refs = btrfs_extent_refs(leaf, ei); | |
868 | extent_flags = btrfs_extent_flags(leaf, ei); | |
869 | } else { | |
870 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
871 | struct btrfs_extent_item_v0 *ei0; | |
872 | BUG_ON(item_size != sizeof(*ei0)); | |
873 | ei0 = btrfs_item_ptr(leaf, path->slots[0], | |
874 | struct btrfs_extent_item_v0); | |
875 | num_refs = btrfs_extent_refs_v0(leaf, ei0); | |
876 | /* FIXME: this isn't correct for data */ | |
877 | extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
878 | #else | |
879 | BUG(); | |
880 | #endif | |
881 | } | |
882 | BUG_ON(num_refs == 0); | |
883 | } else { | |
884 | num_refs = 0; | |
885 | extent_flags = 0; | |
886 | ret = 0; | |
887 | } | |
888 | ||
889 | if (!trans) | |
890 | goto out; | |
891 | ||
892 | delayed_refs = &trans->transaction->delayed_refs; | |
893 | spin_lock(&delayed_refs->lock); | |
f72ad18e | 894 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
a22285a6 YZ |
895 | if (head) { |
896 | if (!mutex_trylock(&head->mutex)) { | |
d278850e | 897 | refcount_inc(&head->refs); |
a22285a6 YZ |
898 | spin_unlock(&delayed_refs->lock); |
899 | ||
b3b4aa74 | 900 | btrfs_release_path(path); |
a22285a6 | 901 | |
8cc33e5c DS |
902 | /* |
903 | * Mutex was contended, block until it's released and try | |
904 | * again | |
905 | */ | |
a22285a6 YZ |
906 | mutex_lock(&head->mutex); |
907 | mutex_unlock(&head->mutex); | |
d278850e | 908 | btrfs_put_delayed_ref_head(head); |
639eefc8 | 909 | goto search_again; |
a22285a6 | 910 | } |
d7df2c79 | 911 | spin_lock(&head->lock); |
a22285a6 YZ |
912 | if (head->extent_op && head->extent_op->update_flags) |
913 | extent_flags |= head->extent_op->flags_to_set; | |
914 | else | |
915 | BUG_ON(num_refs == 0); | |
916 | ||
d278850e | 917 | num_refs += head->ref_mod; |
d7df2c79 | 918 | spin_unlock(&head->lock); |
a22285a6 YZ |
919 | mutex_unlock(&head->mutex); |
920 | } | |
921 | spin_unlock(&delayed_refs->lock); | |
922 | out: | |
923 | WARN_ON(num_refs == 0); | |
924 | if (refs) | |
925 | *refs = num_refs; | |
926 | if (flags) | |
927 | *flags = extent_flags; | |
928 | out_free: | |
929 | btrfs_free_path(path); | |
930 | return ret; | |
931 | } | |
932 | ||
d8d5f3e1 CM |
933 | /* |
934 | * Back reference rules. Back refs have three main goals: | |
935 | * | |
936 | * 1) differentiate between all holders of references to an extent so that | |
937 | * when a reference is dropped we can make sure it was a valid reference | |
938 | * before freeing the extent. | |
939 | * | |
940 | * 2) Provide enough information to quickly find the holders of an extent | |
941 | * if we notice a given block is corrupted or bad. | |
942 | * | |
943 | * 3) Make it easy to migrate blocks for FS shrinking or storage pool | |
944 | * maintenance. This is actually the same as #2, but with a slightly | |
945 | * different use case. | |
946 | * | |
5d4f98a2 YZ |
947 | * There are two kinds of back refs. The implicit back refs is optimized |
948 | * for pointers in non-shared tree blocks. For a given pointer in a block, | |
949 | * back refs of this kind provide information about the block's owner tree | |
950 | * and the pointer's key. These information allow us to find the block by | |
951 | * b-tree searching. The full back refs is for pointers in tree blocks not | |
952 | * referenced by their owner trees. The location of tree block is recorded | |
953 | * in the back refs. Actually the full back refs is generic, and can be | |
954 | * used in all cases the implicit back refs is used. The major shortcoming | |
955 | * of the full back refs is its overhead. Every time a tree block gets | |
956 | * COWed, we have to update back refs entry for all pointers in it. | |
957 | * | |
958 | * For a newly allocated tree block, we use implicit back refs for | |
959 | * pointers in it. This means most tree related operations only involve | |
960 | * implicit back refs. For a tree block created in old transaction, the | |
961 | * only way to drop a reference to it is COW it. So we can detect the | |
962 | * event that tree block loses its owner tree's reference and do the | |
963 | * back refs conversion. | |
964 | * | |
01327610 | 965 | * When a tree block is COWed through a tree, there are four cases: |
5d4f98a2 YZ |
966 | * |
967 | * The reference count of the block is one and the tree is the block's | |
968 | * owner tree. Nothing to do in this case. | |
969 | * | |
970 | * The reference count of the block is one and the tree is not the | |
971 | * block's owner tree. In this case, full back refs is used for pointers | |
972 | * in the block. Remove these full back refs, add implicit back refs for | |
973 | * every pointers in the new block. | |
974 | * | |
975 | * The reference count of the block is greater than one and the tree is | |
976 | * the block's owner tree. In this case, implicit back refs is used for | |
977 | * pointers in the block. Add full back refs for every pointers in the | |
978 | * block, increase lower level extents' reference counts. The original | |
979 | * implicit back refs are entailed to the new block. | |
980 | * | |
981 | * The reference count of the block is greater than one and the tree is | |
982 | * not the block's owner tree. Add implicit back refs for every pointer in | |
983 | * the new block, increase lower level extents' reference count. | |
984 | * | |
985 | * Back Reference Key composing: | |
986 | * | |
987 | * The key objectid corresponds to the first byte in the extent, | |
988 | * The key type is used to differentiate between types of back refs. | |
989 | * There are different meanings of the key offset for different types | |
990 | * of back refs. | |
991 | * | |
d8d5f3e1 CM |
992 | * File extents can be referenced by: |
993 | * | |
994 | * - multiple snapshots, subvolumes, or different generations in one subvol | |
31840ae1 | 995 | * - different files inside a single subvolume |
d8d5f3e1 CM |
996 | * - different offsets inside a file (bookend extents in file.c) |
997 | * | |
5d4f98a2 | 998 | * The extent ref structure for the implicit back refs has fields for: |
d8d5f3e1 CM |
999 | * |
1000 | * - Objectid of the subvolume root | |
d8d5f3e1 | 1001 | * - objectid of the file holding the reference |
5d4f98a2 YZ |
1002 | * - original offset in the file |
1003 | * - how many bookend extents | |
d8d5f3e1 | 1004 | * |
5d4f98a2 YZ |
1005 | * The key offset for the implicit back refs is hash of the first |
1006 | * three fields. | |
d8d5f3e1 | 1007 | * |
5d4f98a2 | 1008 | * The extent ref structure for the full back refs has field for: |
d8d5f3e1 | 1009 | * |
5d4f98a2 | 1010 | * - number of pointers in the tree leaf |
d8d5f3e1 | 1011 | * |
5d4f98a2 YZ |
1012 | * The key offset for the implicit back refs is the first byte of |
1013 | * the tree leaf | |
d8d5f3e1 | 1014 | * |
5d4f98a2 YZ |
1015 | * When a file extent is allocated, The implicit back refs is used. |
1016 | * the fields are filled in: | |
d8d5f3e1 | 1017 | * |
5d4f98a2 | 1018 | * (root_key.objectid, inode objectid, offset in file, 1) |
d8d5f3e1 | 1019 | * |
5d4f98a2 YZ |
1020 | * When a file extent is removed file truncation, we find the |
1021 | * corresponding implicit back refs and check the following fields: | |
d8d5f3e1 | 1022 | * |
5d4f98a2 | 1023 | * (btrfs_header_owner(leaf), inode objectid, offset in file) |
d8d5f3e1 | 1024 | * |
5d4f98a2 | 1025 | * Btree extents can be referenced by: |
d8d5f3e1 | 1026 | * |
5d4f98a2 | 1027 | * - Different subvolumes |
d8d5f3e1 | 1028 | * |
5d4f98a2 YZ |
1029 | * Both the implicit back refs and the full back refs for tree blocks |
1030 | * only consist of key. The key offset for the implicit back refs is | |
1031 | * objectid of block's owner tree. The key offset for the full back refs | |
1032 | * is the first byte of parent block. | |
d8d5f3e1 | 1033 | * |
5d4f98a2 YZ |
1034 | * When implicit back refs is used, information about the lowest key and |
1035 | * level of the tree block are required. These information are stored in | |
1036 | * tree block info structure. | |
d8d5f3e1 | 1037 | */ |
31840ae1 | 1038 | |
5d4f98a2 YZ |
1039 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
1040 | static int convert_extent_item_v0(struct btrfs_trans_handle *trans, | |
87bde3cd | 1041 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1042 | struct btrfs_path *path, |
1043 | u64 owner, u32 extra_size) | |
7bb86316 | 1044 | { |
87bde3cd | 1045 | struct btrfs_root *root = fs_info->extent_root; |
5d4f98a2 YZ |
1046 | struct btrfs_extent_item *item; |
1047 | struct btrfs_extent_item_v0 *ei0; | |
1048 | struct btrfs_extent_ref_v0 *ref0; | |
1049 | struct btrfs_tree_block_info *bi; | |
1050 | struct extent_buffer *leaf; | |
7bb86316 | 1051 | struct btrfs_key key; |
5d4f98a2 YZ |
1052 | struct btrfs_key found_key; |
1053 | u32 new_size = sizeof(*item); | |
1054 | u64 refs; | |
1055 | int ret; | |
1056 | ||
1057 | leaf = path->nodes[0]; | |
1058 | BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0)); | |
1059 | ||
1060 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1061 | ei0 = btrfs_item_ptr(leaf, path->slots[0], | |
1062 | struct btrfs_extent_item_v0); | |
1063 | refs = btrfs_extent_refs_v0(leaf, ei0); | |
1064 | ||
1065 | if (owner == (u64)-1) { | |
1066 | while (1) { | |
1067 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1068 | ret = btrfs_next_leaf(root, path); | |
1069 | if (ret < 0) | |
1070 | return ret; | |
79787eaa | 1071 | BUG_ON(ret > 0); /* Corruption */ |
5d4f98a2 YZ |
1072 | leaf = path->nodes[0]; |
1073 | } | |
1074 | btrfs_item_key_to_cpu(leaf, &found_key, | |
1075 | path->slots[0]); | |
1076 | BUG_ON(key.objectid != found_key.objectid); | |
1077 | if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) { | |
1078 | path->slots[0]++; | |
1079 | continue; | |
1080 | } | |
1081 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1082 | struct btrfs_extent_ref_v0); | |
1083 | owner = btrfs_ref_objectid_v0(leaf, ref0); | |
1084 | break; | |
1085 | } | |
1086 | } | |
b3b4aa74 | 1087 | btrfs_release_path(path); |
5d4f98a2 YZ |
1088 | |
1089 | if (owner < BTRFS_FIRST_FREE_OBJECTID) | |
1090 | new_size += sizeof(*bi); | |
1091 | ||
1092 | new_size -= sizeof(*ei0); | |
1093 | ret = btrfs_search_slot(trans, root, &key, path, | |
1094 | new_size + extra_size, 1); | |
1095 | if (ret < 0) | |
1096 | return ret; | |
79787eaa | 1097 | BUG_ON(ret); /* Corruption */ |
5d4f98a2 | 1098 | |
87bde3cd | 1099 | btrfs_extend_item(fs_info, path, new_size); |
5d4f98a2 YZ |
1100 | |
1101 | leaf = path->nodes[0]; | |
1102 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1103 | btrfs_set_extent_refs(leaf, item, refs); | |
1104 | /* FIXME: get real generation */ | |
1105 | btrfs_set_extent_generation(leaf, item, 0); | |
1106 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1107 | btrfs_set_extent_flags(leaf, item, | |
1108 | BTRFS_EXTENT_FLAG_TREE_BLOCK | | |
1109 | BTRFS_BLOCK_FLAG_FULL_BACKREF); | |
1110 | bi = (struct btrfs_tree_block_info *)(item + 1); | |
1111 | /* FIXME: get first key of the block */ | |
b159fa28 | 1112 | memzero_extent_buffer(leaf, (unsigned long)bi, sizeof(*bi)); |
5d4f98a2 YZ |
1113 | btrfs_set_tree_block_level(leaf, bi, (int)owner); |
1114 | } else { | |
1115 | btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA); | |
1116 | } | |
1117 | btrfs_mark_buffer_dirty(leaf); | |
1118 | return 0; | |
1119 | } | |
1120 | #endif | |
1121 | ||
167ce953 LB |
1122 | /* |
1123 | * is_data == BTRFS_REF_TYPE_BLOCK, tree block type is required, | |
1124 | * is_data == BTRFS_REF_TYPE_DATA, data type is requried, | |
1125 | * is_data == BTRFS_REF_TYPE_ANY, either type is OK. | |
1126 | */ | |
1127 | int btrfs_get_extent_inline_ref_type(const struct extent_buffer *eb, | |
1128 | struct btrfs_extent_inline_ref *iref, | |
1129 | enum btrfs_inline_ref_type is_data) | |
1130 | { | |
1131 | int type = btrfs_extent_inline_ref_type(eb, iref); | |
64ecdb64 | 1132 | u64 offset = btrfs_extent_inline_ref_offset(eb, iref); |
167ce953 LB |
1133 | |
1134 | if (type == BTRFS_TREE_BLOCK_REF_KEY || | |
1135 | type == BTRFS_SHARED_BLOCK_REF_KEY || | |
1136 | type == BTRFS_SHARED_DATA_REF_KEY || | |
1137 | type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1138 | if (is_data == BTRFS_REF_TYPE_BLOCK) { | |
64ecdb64 | 1139 | if (type == BTRFS_TREE_BLOCK_REF_KEY) |
167ce953 | 1140 | return type; |
64ecdb64 LB |
1141 | if (type == BTRFS_SHARED_BLOCK_REF_KEY) { |
1142 | ASSERT(eb->fs_info); | |
1143 | /* | |
1144 | * Every shared one has parent tree | |
1145 | * block, which must be aligned to | |
1146 | * nodesize. | |
1147 | */ | |
1148 | if (offset && | |
1149 | IS_ALIGNED(offset, eb->fs_info->nodesize)) | |
1150 | return type; | |
1151 | } | |
167ce953 | 1152 | } else if (is_data == BTRFS_REF_TYPE_DATA) { |
64ecdb64 | 1153 | if (type == BTRFS_EXTENT_DATA_REF_KEY) |
167ce953 | 1154 | return type; |
64ecdb64 LB |
1155 | if (type == BTRFS_SHARED_DATA_REF_KEY) { |
1156 | ASSERT(eb->fs_info); | |
1157 | /* | |
1158 | * Every shared one has parent tree | |
1159 | * block, which must be aligned to | |
1160 | * nodesize. | |
1161 | */ | |
1162 | if (offset && | |
1163 | IS_ALIGNED(offset, eb->fs_info->nodesize)) | |
1164 | return type; | |
1165 | } | |
167ce953 LB |
1166 | } else { |
1167 | ASSERT(is_data == BTRFS_REF_TYPE_ANY); | |
1168 | return type; | |
1169 | } | |
1170 | } | |
1171 | ||
1172 | btrfs_print_leaf((struct extent_buffer *)eb); | |
1173 | btrfs_err(eb->fs_info, "eb %llu invalid extent inline ref type %d", | |
1174 | eb->start, type); | |
1175 | WARN_ON(1); | |
1176 | ||
1177 | return BTRFS_REF_TYPE_INVALID; | |
1178 | } | |
1179 | ||
5d4f98a2 YZ |
1180 | static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset) |
1181 | { | |
1182 | u32 high_crc = ~(u32)0; | |
1183 | u32 low_crc = ~(u32)0; | |
1184 | __le64 lenum; | |
1185 | ||
1186 | lenum = cpu_to_le64(root_objectid); | |
9678c543 | 1187 | high_crc = crc32c(high_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 1188 | lenum = cpu_to_le64(owner); |
9678c543 | 1189 | low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 1190 | lenum = cpu_to_le64(offset); |
9678c543 | 1191 | low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 YZ |
1192 | |
1193 | return ((u64)high_crc << 31) ^ (u64)low_crc; | |
1194 | } | |
1195 | ||
1196 | static u64 hash_extent_data_ref_item(struct extent_buffer *leaf, | |
1197 | struct btrfs_extent_data_ref *ref) | |
1198 | { | |
1199 | return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref), | |
1200 | btrfs_extent_data_ref_objectid(leaf, ref), | |
1201 | btrfs_extent_data_ref_offset(leaf, ref)); | |
1202 | } | |
1203 | ||
1204 | static int match_extent_data_ref(struct extent_buffer *leaf, | |
1205 | struct btrfs_extent_data_ref *ref, | |
1206 | u64 root_objectid, u64 owner, u64 offset) | |
1207 | { | |
1208 | if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid || | |
1209 | btrfs_extent_data_ref_objectid(leaf, ref) != owner || | |
1210 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
1211 | return 0; | |
1212 | return 1; | |
1213 | } | |
1214 | ||
1215 | static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
1216 | struct btrfs_path *path, |
1217 | u64 bytenr, u64 parent, | |
1218 | u64 root_objectid, | |
1219 | u64 owner, u64 offset) | |
1220 | { | |
bd1d53ef | 1221 | struct btrfs_root *root = trans->fs_info->extent_root; |
5d4f98a2 YZ |
1222 | struct btrfs_key key; |
1223 | struct btrfs_extent_data_ref *ref; | |
31840ae1 | 1224 | struct extent_buffer *leaf; |
5d4f98a2 | 1225 | u32 nritems; |
74493f7a | 1226 | int ret; |
5d4f98a2 YZ |
1227 | int recow; |
1228 | int err = -ENOENT; | |
74493f7a | 1229 | |
31840ae1 | 1230 | key.objectid = bytenr; |
5d4f98a2 YZ |
1231 | if (parent) { |
1232 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1233 | key.offset = parent; | |
1234 | } else { | |
1235 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1236 | key.offset = hash_extent_data_ref(root_objectid, | |
1237 | owner, offset); | |
1238 | } | |
1239 | again: | |
1240 | recow = 0; | |
1241 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1242 | if (ret < 0) { | |
1243 | err = ret; | |
1244 | goto fail; | |
1245 | } | |
31840ae1 | 1246 | |
5d4f98a2 YZ |
1247 | if (parent) { |
1248 | if (!ret) | |
1249 | return 0; | |
1250 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1251 | key.type = BTRFS_EXTENT_REF_V0_KEY; | |
b3b4aa74 | 1252 | btrfs_release_path(path); |
5d4f98a2 YZ |
1253 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1254 | if (ret < 0) { | |
1255 | err = ret; | |
1256 | goto fail; | |
1257 | } | |
1258 | if (!ret) | |
1259 | return 0; | |
1260 | #endif | |
1261 | goto fail; | |
31840ae1 ZY |
1262 | } |
1263 | ||
1264 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1265 | nritems = btrfs_header_nritems(leaf); |
1266 | while (1) { | |
1267 | if (path->slots[0] >= nritems) { | |
1268 | ret = btrfs_next_leaf(root, path); | |
1269 | if (ret < 0) | |
1270 | err = ret; | |
1271 | if (ret) | |
1272 | goto fail; | |
1273 | ||
1274 | leaf = path->nodes[0]; | |
1275 | nritems = btrfs_header_nritems(leaf); | |
1276 | recow = 1; | |
1277 | } | |
1278 | ||
1279 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1280 | if (key.objectid != bytenr || | |
1281 | key.type != BTRFS_EXTENT_DATA_REF_KEY) | |
1282 | goto fail; | |
1283 | ||
1284 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1285 | struct btrfs_extent_data_ref); | |
1286 | ||
1287 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1288 | owner, offset)) { | |
1289 | if (recow) { | |
b3b4aa74 | 1290 | btrfs_release_path(path); |
5d4f98a2 YZ |
1291 | goto again; |
1292 | } | |
1293 | err = 0; | |
1294 | break; | |
1295 | } | |
1296 | path->slots[0]++; | |
31840ae1 | 1297 | } |
5d4f98a2 YZ |
1298 | fail: |
1299 | return err; | |
31840ae1 ZY |
1300 | } |
1301 | ||
5d4f98a2 | 1302 | static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1303 | struct btrfs_path *path, |
1304 | u64 bytenr, u64 parent, | |
1305 | u64 root_objectid, u64 owner, | |
1306 | u64 offset, int refs_to_add) | |
31840ae1 | 1307 | { |
62b895af | 1308 | struct btrfs_root *root = trans->fs_info->extent_root; |
31840ae1 ZY |
1309 | struct btrfs_key key; |
1310 | struct extent_buffer *leaf; | |
5d4f98a2 | 1311 | u32 size; |
31840ae1 ZY |
1312 | u32 num_refs; |
1313 | int ret; | |
74493f7a | 1314 | |
74493f7a | 1315 | key.objectid = bytenr; |
5d4f98a2 YZ |
1316 | if (parent) { |
1317 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1318 | key.offset = parent; | |
1319 | size = sizeof(struct btrfs_shared_data_ref); | |
1320 | } else { | |
1321 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1322 | key.offset = hash_extent_data_ref(root_objectid, | |
1323 | owner, offset); | |
1324 | size = sizeof(struct btrfs_extent_data_ref); | |
1325 | } | |
74493f7a | 1326 | |
5d4f98a2 YZ |
1327 | ret = btrfs_insert_empty_item(trans, root, path, &key, size); |
1328 | if (ret && ret != -EEXIST) | |
1329 | goto fail; | |
1330 | ||
1331 | leaf = path->nodes[0]; | |
1332 | if (parent) { | |
1333 | struct btrfs_shared_data_ref *ref; | |
31840ae1 | 1334 | ref = btrfs_item_ptr(leaf, path->slots[0], |
5d4f98a2 YZ |
1335 | struct btrfs_shared_data_ref); |
1336 | if (ret == 0) { | |
1337 | btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add); | |
1338 | } else { | |
1339 | num_refs = btrfs_shared_data_ref_count(leaf, ref); | |
1340 | num_refs += refs_to_add; | |
1341 | btrfs_set_shared_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1342 | } |
5d4f98a2 YZ |
1343 | } else { |
1344 | struct btrfs_extent_data_ref *ref; | |
1345 | while (ret == -EEXIST) { | |
1346 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1347 | struct btrfs_extent_data_ref); | |
1348 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1349 | owner, offset)) | |
1350 | break; | |
b3b4aa74 | 1351 | btrfs_release_path(path); |
5d4f98a2 YZ |
1352 | key.offset++; |
1353 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1354 | size); | |
1355 | if (ret && ret != -EEXIST) | |
1356 | goto fail; | |
31840ae1 | 1357 | |
5d4f98a2 YZ |
1358 | leaf = path->nodes[0]; |
1359 | } | |
1360 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1361 | struct btrfs_extent_data_ref); | |
1362 | if (ret == 0) { | |
1363 | btrfs_set_extent_data_ref_root(leaf, ref, | |
1364 | root_objectid); | |
1365 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
1366 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
1367 | btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add); | |
1368 | } else { | |
1369 | num_refs = btrfs_extent_data_ref_count(leaf, ref); | |
1370 | num_refs += refs_to_add; | |
1371 | btrfs_set_extent_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1372 | } |
31840ae1 | 1373 | } |
5d4f98a2 YZ |
1374 | btrfs_mark_buffer_dirty(leaf); |
1375 | ret = 0; | |
1376 | fail: | |
b3b4aa74 | 1377 | btrfs_release_path(path); |
7bb86316 | 1378 | return ret; |
74493f7a CM |
1379 | } |
1380 | ||
5d4f98a2 | 1381 | static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 | 1382 | struct btrfs_path *path, |
fcebe456 | 1383 | int refs_to_drop, int *last_ref) |
31840ae1 | 1384 | { |
5d4f98a2 YZ |
1385 | struct btrfs_key key; |
1386 | struct btrfs_extent_data_ref *ref1 = NULL; | |
1387 | struct btrfs_shared_data_ref *ref2 = NULL; | |
31840ae1 | 1388 | struct extent_buffer *leaf; |
5d4f98a2 | 1389 | u32 num_refs = 0; |
31840ae1 ZY |
1390 | int ret = 0; |
1391 | ||
1392 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1393 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
1394 | ||
1395 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1396 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1397 | struct btrfs_extent_data_ref); | |
1398 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1399 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1400 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1401 | struct btrfs_shared_data_ref); | |
1402 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1403 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1404 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
1405 | struct btrfs_extent_ref_v0 *ref0; | |
1406 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1407 | struct btrfs_extent_ref_v0); | |
1408 | num_refs = btrfs_ref_count_v0(leaf, ref0); | |
1409 | #endif | |
1410 | } else { | |
1411 | BUG(); | |
1412 | } | |
1413 | ||
56bec294 CM |
1414 | BUG_ON(num_refs < refs_to_drop); |
1415 | num_refs -= refs_to_drop; | |
5d4f98a2 | 1416 | |
31840ae1 | 1417 | if (num_refs == 0) { |
e9f6290d | 1418 | ret = btrfs_del_item(trans, trans->fs_info->extent_root, path); |
fcebe456 | 1419 | *last_ref = 1; |
31840ae1 | 1420 | } else { |
5d4f98a2 YZ |
1421 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) |
1422 | btrfs_set_extent_data_ref_count(leaf, ref1, num_refs); | |
1423 | else if (key.type == BTRFS_SHARED_DATA_REF_KEY) | |
1424 | btrfs_set_shared_data_ref_count(leaf, ref2, num_refs); | |
1425 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1426 | else { | |
1427 | struct btrfs_extent_ref_v0 *ref0; | |
1428 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1429 | struct btrfs_extent_ref_v0); | |
1430 | btrfs_set_ref_count_v0(leaf, ref0, num_refs); | |
1431 | } | |
1432 | #endif | |
31840ae1 ZY |
1433 | btrfs_mark_buffer_dirty(leaf); |
1434 | } | |
31840ae1 ZY |
1435 | return ret; |
1436 | } | |
1437 | ||
9ed0dea0 | 1438 | static noinline u32 extent_data_ref_count(struct btrfs_path *path, |
5d4f98a2 | 1439 | struct btrfs_extent_inline_ref *iref) |
15916de8 | 1440 | { |
5d4f98a2 YZ |
1441 | struct btrfs_key key; |
1442 | struct extent_buffer *leaf; | |
1443 | struct btrfs_extent_data_ref *ref1; | |
1444 | struct btrfs_shared_data_ref *ref2; | |
1445 | u32 num_refs = 0; | |
3de28d57 | 1446 | int type; |
5d4f98a2 YZ |
1447 | |
1448 | leaf = path->nodes[0]; | |
1449 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1450 | if (iref) { | |
3de28d57 LB |
1451 | /* |
1452 | * If type is invalid, we should have bailed out earlier than | |
1453 | * this call. | |
1454 | */ | |
1455 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA); | |
1456 | ASSERT(type != BTRFS_REF_TYPE_INVALID); | |
1457 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
5d4f98a2 YZ |
1458 | ref1 = (struct btrfs_extent_data_ref *)(&iref->offset); |
1459 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1460 | } else { | |
1461 | ref2 = (struct btrfs_shared_data_ref *)(iref + 1); | |
1462 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1463 | } | |
1464 | } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1465 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1466 | struct btrfs_extent_data_ref); | |
1467 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1468 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1469 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1470 | struct btrfs_shared_data_ref); | |
1471 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1472 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1473 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
1474 | struct btrfs_extent_ref_v0 *ref0; | |
1475 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1476 | struct btrfs_extent_ref_v0); | |
1477 | num_refs = btrfs_ref_count_v0(leaf, ref0); | |
4b4e25f2 | 1478 | #endif |
5d4f98a2 YZ |
1479 | } else { |
1480 | WARN_ON(1); | |
1481 | } | |
1482 | return num_refs; | |
1483 | } | |
15916de8 | 1484 | |
5d4f98a2 | 1485 | static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1486 | struct btrfs_path *path, |
1487 | u64 bytenr, u64 parent, | |
1488 | u64 root_objectid) | |
1f3c79a2 | 1489 | { |
b8582eea | 1490 | struct btrfs_root *root = trans->fs_info->extent_root; |
5d4f98a2 | 1491 | struct btrfs_key key; |
1f3c79a2 | 1492 | int ret; |
1f3c79a2 | 1493 | |
5d4f98a2 YZ |
1494 | key.objectid = bytenr; |
1495 | if (parent) { | |
1496 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1497 | key.offset = parent; | |
1498 | } else { | |
1499 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1500 | key.offset = root_objectid; | |
1f3c79a2 LH |
1501 | } |
1502 | ||
5d4f98a2 YZ |
1503 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1504 | if (ret > 0) | |
1505 | ret = -ENOENT; | |
1506 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1507 | if (ret == -ENOENT && parent) { | |
b3b4aa74 | 1508 | btrfs_release_path(path); |
5d4f98a2 YZ |
1509 | key.type = BTRFS_EXTENT_REF_V0_KEY; |
1510 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1511 | if (ret > 0) | |
1512 | ret = -ENOENT; | |
1513 | } | |
1f3c79a2 | 1514 | #endif |
5d4f98a2 | 1515 | return ret; |
1f3c79a2 LH |
1516 | } |
1517 | ||
5d4f98a2 | 1518 | static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1519 | struct btrfs_path *path, |
1520 | u64 bytenr, u64 parent, | |
1521 | u64 root_objectid) | |
31840ae1 | 1522 | { |
5d4f98a2 | 1523 | struct btrfs_key key; |
31840ae1 | 1524 | int ret; |
31840ae1 | 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; | |
1533 | } | |
1534 | ||
10728404 | 1535 | ret = btrfs_insert_empty_item(trans, trans->fs_info->extent_root, |
87bde3cd | 1536 | path, &key, 0); |
b3b4aa74 | 1537 | btrfs_release_path(path); |
31840ae1 ZY |
1538 | return ret; |
1539 | } | |
1540 | ||
5d4f98a2 | 1541 | static inline int extent_ref_type(u64 parent, u64 owner) |
31840ae1 | 1542 | { |
5d4f98a2 YZ |
1543 | int type; |
1544 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1545 | if (parent > 0) | |
1546 | type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1547 | else | |
1548 | type = BTRFS_TREE_BLOCK_REF_KEY; | |
1549 | } else { | |
1550 | if (parent > 0) | |
1551 | type = BTRFS_SHARED_DATA_REF_KEY; | |
1552 | else | |
1553 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
1554 | } | |
1555 | return type; | |
31840ae1 | 1556 | } |
56bec294 | 1557 | |
2c47e605 YZ |
1558 | static int find_next_key(struct btrfs_path *path, int level, |
1559 | struct btrfs_key *key) | |
56bec294 | 1560 | |
02217ed2 | 1561 | { |
2c47e605 | 1562 | for (; level < BTRFS_MAX_LEVEL; level++) { |
5d4f98a2 YZ |
1563 | if (!path->nodes[level]) |
1564 | break; | |
5d4f98a2 YZ |
1565 | if (path->slots[level] + 1 >= |
1566 | btrfs_header_nritems(path->nodes[level])) | |
1567 | continue; | |
1568 | if (level == 0) | |
1569 | btrfs_item_key_to_cpu(path->nodes[level], key, | |
1570 | path->slots[level] + 1); | |
1571 | else | |
1572 | btrfs_node_key_to_cpu(path->nodes[level], key, | |
1573 | path->slots[level] + 1); | |
1574 | return 0; | |
1575 | } | |
1576 | return 1; | |
1577 | } | |
037e6390 | 1578 | |
5d4f98a2 YZ |
1579 | /* |
1580 | * look for inline back ref. if back ref is found, *ref_ret is set | |
1581 | * to the address of inline back ref, and 0 is returned. | |
1582 | * | |
1583 | * if back ref isn't found, *ref_ret is set to the address where it | |
1584 | * should be inserted, and -ENOENT is returned. | |
1585 | * | |
1586 | * if insert is true and there are too many inline back refs, the path | |
1587 | * points to the extent item, and -EAGAIN is returned. | |
1588 | * | |
1589 | * NOTE: inline back refs are ordered in the same way that back ref | |
1590 | * items in the tree are ordered. | |
1591 | */ | |
1592 | static noinline_for_stack | |
1593 | int lookup_inline_extent_backref(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
1594 | struct btrfs_path *path, |
1595 | struct btrfs_extent_inline_ref **ref_ret, | |
1596 | u64 bytenr, u64 num_bytes, | |
1597 | u64 parent, u64 root_objectid, | |
1598 | u64 owner, u64 offset, int insert) | |
1599 | { | |
867cc1fb | 1600 | struct btrfs_fs_info *fs_info = trans->fs_info; |
87bde3cd | 1601 | struct btrfs_root *root = fs_info->extent_root; |
5d4f98a2 YZ |
1602 | struct btrfs_key key; |
1603 | struct extent_buffer *leaf; | |
1604 | struct btrfs_extent_item *ei; | |
1605 | struct btrfs_extent_inline_ref *iref; | |
1606 | u64 flags; | |
1607 | u64 item_size; | |
1608 | unsigned long ptr; | |
1609 | unsigned long end; | |
1610 | int extra_size; | |
1611 | int type; | |
1612 | int want; | |
1613 | int ret; | |
1614 | int err = 0; | |
0b246afa | 1615 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); |
3de28d57 | 1616 | int needed; |
26b8003f | 1617 | |
db94535d | 1618 | key.objectid = bytenr; |
31840ae1 | 1619 | key.type = BTRFS_EXTENT_ITEM_KEY; |
56bec294 | 1620 | key.offset = num_bytes; |
31840ae1 | 1621 | |
5d4f98a2 YZ |
1622 | want = extent_ref_type(parent, owner); |
1623 | if (insert) { | |
1624 | extra_size = btrfs_extent_inline_ref_size(want); | |
85d4198e | 1625 | path->keep_locks = 1; |
5d4f98a2 YZ |
1626 | } else |
1627 | extra_size = -1; | |
3173a18f JB |
1628 | |
1629 | /* | |
16d1c062 NB |
1630 | * Owner is our level, so we can just add one to get the level for the |
1631 | * block we are interested in. | |
3173a18f JB |
1632 | */ |
1633 | if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1634 | key.type = BTRFS_METADATA_ITEM_KEY; | |
1635 | key.offset = owner; | |
1636 | } | |
1637 | ||
1638 | again: | |
5d4f98a2 | 1639 | ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1); |
b9473439 | 1640 | if (ret < 0) { |
5d4f98a2 YZ |
1641 | err = ret; |
1642 | goto out; | |
1643 | } | |
3173a18f JB |
1644 | |
1645 | /* | |
1646 | * We may be a newly converted file system which still has the old fat | |
1647 | * extent entries for metadata, so try and see if we have one of those. | |
1648 | */ | |
1649 | if (ret > 0 && skinny_metadata) { | |
1650 | skinny_metadata = false; | |
1651 | if (path->slots[0]) { | |
1652 | path->slots[0]--; | |
1653 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
1654 | path->slots[0]); | |
1655 | if (key.objectid == bytenr && | |
1656 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
1657 | key.offset == num_bytes) | |
1658 | ret = 0; | |
1659 | } | |
1660 | if (ret) { | |
9ce49a0b | 1661 | key.objectid = bytenr; |
3173a18f JB |
1662 | key.type = BTRFS_EXTENT_ITEM_KEY; |
1663 | key.offset = num_bytes; | |
1664 | btrfs_release_path(path); | |
1665 | goto again; | |
1666 | } | |
1667 | } | |
1668 | ||
79787eaa JM |
1669 | if (ret && !insert) { |
1670 | err = -ENOENT; | |
1671 | goto out; | |
fae7f21c | 1672 | } else if (WARN_ON(ret)) { |
492104c8 | 1673 | err = -EIO; |
492104c8 | 1674 | goto out; |
79787eaa | 1675 | } |
5d4f98a2 YZ |
1676 | |
1677 | leaf = path->nodes[0]; | |
1678 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1679 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1680 | if (item_size < sizeof(*ei)) { | |
1681 | if (!insert) { | |
1682 | err = -ENOENT; | |
1683 | goto out; | |
1684 | } | |
87bde3cd | 1685 | ret = convert_extent_item_v0(trans, fs_info, path, owner, |
5d4f98a2 YZ |
1686 | extra_size); |
1687 | if (ret < 0) { | |
1688 | err = ret; | |
1689 | goto out; | |
1690 | } | |
1691 | leaf = path->nodes[0]; | |
1692 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1693 | } | |
1694 | #endif | |
1695 | BUG_ON(item_size < sizeof(*ei)); | |
1696 | ||
5d4f98a2 YZ |
1697 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1698 | flags = btrfs_extent_flags(leaf, ei); | |
1699 | ||
1700 | ptr = (unsigned long)(ei + 1); | |
1701 | end = (unsigned long)ei + item_size; | |
1702 | ||
3173a18f | 1703 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) { |
5d4f98a2 YZ |
1704 | ptr += sizeof(struct btrfs_tree_block_info); |
1705 | BUG_ON(ptr > end); | |
5d4f98a2 YZ |
1706 | } |
1707 | ||
3de28d57 LB |
1708 | if (owner >= BTRFS_FIRST_FREE_OBJECTID) |
1709 | needed = BTRFS_REF_TYPE_DATA; | |
1710 | else | |
1711 | needed = BTRFS_REF_TYPE_BLOCK; | |
1712 | ||
5d4f98a2 YZ |
1713 | err = -ENOENT; |
1714 | while (1) { | |
1715 | if (ptr >= end) { | |
1716 | WARN_ON(ptr > end); | |
1717 | break; | |
1718 | } | |
1719 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
3de28d57 LB |
1720 | type = btrfs_get_extent_inline_ref_type(leaf, iref, needed); |
1721 | if (type == BTRFS_REF_TYPE_INVALID) { | |
1722 | err = -EINVAL; | |
1723 | goto out; | |
1724 | } | |
1725 | ||
5d4f98a2 YZ |
1726 | if (want < type) |
1727 | break; | |
1728 | if (want > type) { | |
1729 | ptr += btrfs_extent_inline_ref_size(type); | |
1730 | continue; | |
1731 | } | |
1732 | ||
1733 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1734 | struct btrfs_extent_data_ref *dref; | |
1735 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1736 | if (match_extent_data_ref(leaf, dref, root_objectid, | |
1737 | owner, offset)) { | |
1738 | err = 0; | |
1739 | break; | |
1740 | } | |
1741 | if (hash_extent_data_ref_item(leaf, dref) < | |
1742 | hash_extent_data_ref(root_objectid, owner, offset)) | |
1743 | break; | |
1744 | } else { | |
1745 | u64 ref_offset; | |
1746 | ref_offset = btrfs_extent_inline_ref_offset(leaf, iref); | |
1747 | if (parent > 0) { | |
1748 | if (parent == ref_offset) { | |
1749 | err = 0; | |
1750 | break; | |
1751 | } | |
1752 | if (ref_offset < parent) | |
1753 | break; | |
1754 | } else { | |
1755 | if (root_objectid == ref_offset) { | |
1756 | err = 0; | |
1757 | break; | |
1758 | } | |
1759 | if (ref_offset < root_objectid) | |
1760 | break; | |
1761 | } | |
1762 | } | |
1763 | ptr += btrfs_extent_inline_ref_size(type); | |
1764 | } | |
1765 | if (err == -ENOENT && insert) { | |
1766 | if (item_size + extra_size >= | |
1767 | BTRFS_MAX_EXTENT_ITEM_SIZE(root)) { | |
1768 | err = -EAGAIN; | |
1769 | goto out; | |
1770 | } | |
1771 | /* | |
1772 | * To add new inline back ref, we have to make sure | |
1773 | * there is no corresponding back ref item. | |
1774 | * For simplicity, we just do not add new inline back | |
1775 | * ref if there is any kind of item for this block | |
1776 | */ | |
2c47e605 YZ |
1777 | if (find_next_key(path, 0, &key) == 0 && |
1778 | key.objectid == bytenr && | |
85d4198e | 1779 | key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) { |
5d4f98a2 YZ |
1780 | err = -EAGAIN; |
1781 | goto out; | |
1782 | } | |
1783 | } | |
1784 | *ref_ret = (struct btrfs_extent_inline_ref *)ptr; | |
1785 | out: | |
85d4198e | 1786 | if (insert) { |
5d4f98a2 YZ |
1787 | path->keep_locks = 0; |
1788 | btrfs_unlock_up_safe(path, 1); | |
1789 | } | |
1790 | return err; | |
1791 | } | |
1792 | ||
1793 | /* | |
1794 | * helper to add new inline back ref | |
1795 | */ | |
1796 | static noinline_for_stack | |
87bde3cd | 1797 | void setup_inline_extent_backref(struct btrfs_fs_info *fs_info, |
143bede5 JM |
1798 | struct btrfs_path *path, |
1799 | struct btrfs_extent_inline_ref *iref, | |
1800 | u64 parent, u64 root_objectid, | |
1801 | u64 owner, u64 offset, int refs_to_add, | |
1802 | struct btrfs_delayed_extent_op *extent_op) | |
5d4f98a2 YZ |
1803 | { |
1804 | struct extent_buffer *leaf; | |
1805 | struct btrfs_extent_item *ei; | |
1806 | unsigned long ptr; | |
1807 | unsigned long end; | |
1808 | unsigned long item_offset; | |
1809 | u64 refs; | |
1810 | int size; | |
1811 | int type; | |
5d4f98a2 YZ |
1812 | |
1813 | leaf = path->nodes[0]; | |
1814 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1815 | item_offset = (unsigned long)iref - (unsigned long)ei; | |
1816 | ||
1817 | type = extent_ref_type(parent, owner); | |
1818 | size = btrfs_extent_inline_ref_size(type); | |
1819 | ||
87bde3cd | 1820 | btrfs_extend_item(fs_info, path, size); |
5d4f98a2 YZ |
1821 | |
1822 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1823 | refs = btrfs_extent_refs(leaf, ei); | |
1824 | refs += refs_to_add; | |
1825 | btrfs_set_extent_refs(leaf, ei, refs); | |
1826 | if (extent_op) | |
1827 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1828 | ||
1829 | ptr = (unsigned long)ei + item_offset; | |
1830 | end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]); | |
1831 | if (ptr < end - size) | |
1832 | memmove_extent_buffer(leaf, ptr + size, ptr, | |
1833 | end - size - ptr); | |
1834 | ||
1835 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1836 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
1837 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1838 | struct btrfs_extent_data_ref *dref; | |
1839 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1840 | btrfs_set_extent_data_ref_root(leaf, dref, root_objectid); | |
1841 | btrfs_set_extent_data_ref_objectid(leaf, dref, owner); | |
1842 | btrfs_set_extent_data_ref_offset(leaf, dref, offset); | |
1843 | btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add); | |
1844 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1845 | struct btrfs_shared_data_ref *sref; | |
1846 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1847 | btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add); | |
1848 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1849 | } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) { | |
1850 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1851 | } else { | |
1852 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
1853 | } | |
1854 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1855 | } |
1856 | ||
1857 | static int lookup_extent_backref(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
1858 | struct btrfs_path *path, |
1859 | struct btrfs_extent_inline_ref **ref_ret, | |
1860 | u64 bytenr, u64 num_bytes, u64 parent, | |
1861 | u64 root_objectid, u64 owner, u64 offset) | |
1862 | { | |
1863 | int ret; | |
1864 | ||
867cc1fb NB |
1865 | ret = lookup_inline_extent_backref(trans, path, ref_ret, bytenr, |
1866 | num_bytes, parent, root_objectid, | |
1867 | owner, offset, 0); | |
5d4f98a2 | 1868 | if (ret != -ENOENT) |
54aa1f4d | 1869 | return ret; |
5d4f98a2 | 1870 | |
b3b4aa74 | 1871 | btrfs_release_path(path); |
5d4f98a2 YZ |
1872 | *ref_ret = NULL; |
1873 | ||
1874 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
b8582eea NB |
1875 | ret = lookup_tree_block_ref(trans, path, bytenr, parent, |
1876 | root_objectid); | |
5d4f98a2 | 1877 | } else { |
bd1d53ef NB |
1878 | ret = lookup_extent_data_ref(trans, path, bytenr, parent, |
1879 | root_objectid, owner, offset); | |
b9473439 | 1880 | } |
5d4f98a2 YZ |
1881 | return ret; |
1882 | } | |
31840ae1 | 1883 | |
5d4f98a2 YZ |
1884 | /* |
1885 | * helper to update/remove inline back ref | |
1886 | */ | |
1887 | static noinline_for_stack | |
61a18f1c | 1888 | void update_inline_extent_backref(struct btrfs_path *path, |
143bede5 JM |
1889 | struct btrfs_extent_inline_ref *iref, |
1890 | int refs_to_mod, | |
fcebe456 JB |
1891 | struct btrfs_delayed_extent_op *extent_op, |
1892 | int *last_ref) | |
5d4f98a2 | 1893 | { |
61a18f1c NB |
1894 | struct extent_buffer *leaf = path->nodes[0]; |
1895 | struct btrfs_fs_info *fs_info = leaf->fs_info; | |
5d4f98a2 YZ |
1896 | struct btrfs_extent_item *ei; |
1897 | struct btrfs_extent_data_ref *dref = NULL; | |
1898 | struct btrfs_shared_data_ref *sref = NULL; | |
1899 | unsigned long ptr; | |
1900 | unsigned long end; | |
1901 | u32 item_size; | |
1902 | int size; | |
1903 | int type; | |
5d4f98a2 YZ |
1904 | u64 refs; |
1905 | ||
5d4f98a2 YZ |
1906 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1907 | refs = btrfs_extent_refs(leaf, ei); | |
1908 | WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0); | |
1909 | refs += refs_to_mod; | |
1910 | btrfs_set_extent_refs(leaf, ei, refs); | |
1911 | if (extent_op) | |
1912 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1913 | ||
3de28d57 LB |
1914 | /* |
1915 | * If type is invalid, we should have bailed out after | |
1916 | * lookup_inline_extent_backref(). | |
1917 | */ | |
1918 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_ANY); | |
1919 | ASSERT(type != BTRFS_REF_TYPE_INVALID); | |
5d4f98a2 YZ |
1920 | |
1921 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1922 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1923 | refs = btrfs_extent_data_ref_count(leaf, dref); | |
1924 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1925 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1926 | refs = btrfs_shared_data_ref_count(leaf, sref); | |
1927 | } else { | |
1928 | refs = 1; | |
1929 | BUG_ON(refs_to_mod != -1); | |
56bec294 | 1930 | } |
31840ae1 | 1931 | |
5d4f98a2 YZ |
1932 | BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod); |
1933 | refs += refs_to_mod; | |
1934 | ||
1935 | if (refs > 0) { | |
1936 | if (type == BTRFS_EXTENT_DATA_REF_KEY) | |
1937 | btrfs_set_extent_data_ref_count(leaf, dref, refs); | |
1938 | else | |
1939 | btrfs_set_shared_data_ref_count(leaf, sref, refs); | |
1940 | } else { | |
fcebe456 | 1941 | *last_ref = 1; |
5d4f98a2 YZ |
1942 | size = btrfs_extent_inline_ref_size(type); |
1943 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1944 | ptr = (unsigned long)iref; | |
1945 | end = (unsigned long)ei + item_size; | |
1946 | if (ptr + size < end) | |
1947 | memmove_extent_buffer(leaf, ptr, ptr + size, | |
1948 | end - ptr - size); | |
1949 | item_size -= size; | |
87bde3cd | 1950 | btrfs_truncate_item(fs_info, path, item_size, 1); |
5d4f98a2 YZ |
1951 | } |
1952 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1953 | } |
1954 | ||
1955 | static noinline_for_stack | |
1956 | int insert_inline_extent_backref(struct btrfs_trans_handle *trans, | |
87bde3cd | 1957 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1958 | struct btrfs_path *path, |
1959 | u64 bytenr, u64 num_bytes, u64 parent, | |
1960 | u64 root_objectid, u64 owner, | |
1961 | u64 offset, int refs_to_add, | |
1962 | struct btrfs_delayed_extent_op *extent_op) | |
1963 | { | |
1964 | struct btrfs_extent_inline_ref *iref; | |
1965 | int ret; | |
1966 | ||
867cc1fb NB |
1967 | ret = lookup_inline_extent_backref(trans, path, &iref, bytenr, |
1968 | num_bytes, parent, root_objectid, | |
1969 | owner, offset, 1); | |
5d4f98a2 YZ |
1970 | if (ret == 0) { |
1971 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID); | |
61a18f1c NB |
1972 | update_inline_extent_backref(path, iref, refs_to_add, |
1973 | extent_op, NULL); | |
5d4f98a2 | 1974 | } else if (ret == -ENOENT) { |
87bde3cd | 1975 | setup_inline_extent_backref(fs_info, path, iref, parent, |
143bede5 JM |
1976 | root_objectid, owner, offset, |
1977 | refs_to_add, extent_op); | |
1978 | ret = 0; | |
771ed689 | 1979 | } |
5d4f98a2 YZ |
1980 | return ret; |
1981 | } | |
31840ae1 | 1982 | |
5d4f98a2 | 1983 | static int insert_extent_backref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1984 | struct btrfs_path *path, |
1985 | u64 bytenr, u64 parent, u64 root_objectid, | |
1986 | u64 owner, u64 offset, int refs_to_add) | |
1987 | { | |
1988 | int ret; | |
1989 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1990 | BUG_ON(refs_to_add != 1); | |
10728404 NB |
1991 | ret = insert_tree_block_ref(trans, path, bytenr, parent, |
1992 | root_objectid); | |
5d4f98a2 | 1993 | } else { |
62b895af NB |
1994 | ret = insert_extent_data_ref(trans, path, bytenr, parent, |
1995 | root_objectid, owner, offset, | |
1996 | refs_to_add); | |
5d4f98a2 YZ |
1997 | } |
1998 | return ret; | |
1999 | } | |
56bec294 | 2000 | |
5d4f98a2 | 2001 | static int remove_extent_backref(struct btrfs_trans_handle *trans, |
87bde3cd | 2002 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
2003 | struct btrfs_path *path, |
2004 | struct btrfs_extent_inline_ref *iref, | |
fcebe456 | 2005 | int refs_to_drop, int is_data, int *last_ref) |
5d4f98a2 | 2006 | { |
143bede5 | 2007 | int ret = 0; |
b9473439 | 2008 | |
5d4f98a2 YZ |
2009 | BUG_ON(!is_data && refs_to_drop != 1); |
2010 | if (iref) { | |
61a18f1c NB |
2011 | update_inline_extent_backref(path, iref, -refs_to_drop, NULL, |
2012 | last_ref); | |
5d4f98a2 | 2013 | } else if (is_data) { |
e9f6290d | 2014 | ret = remove_extent_data_ref(trans, path, refs_to_drop, |
fcebe456 | 2015 | last_ref); |
5d4f98a2 | 2016 | } else { |
fcebe456 | 2017 | *last_ref = 1; |
87bde3cd | 2018 | ret = btrfs_del_item(trans, fs_info->extent_root, path); |
5d4f98a2 YZ |
2019 | } |
2020 | return ret; | |
2021 | } | |
2022 | ||
86557861 | 2023 | #define in_range(b, first, len) ((b) >= (first) && (b) < (first) + (len)) |
d04c6b88 JM |
2024 | static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len, |
2025 | u64 *discarded_bytes) | |
5d4f98a2 | 2026 | { |
86557861 JM |
2027 | int j, ret = 0; |
2028 | u64 bytes_left, end; | |
4d89d377 | 2029 | u64 aligned_start = ALIGN(start, 1 << 9); |
d04c6b88 | 2030 | |
4d89d377 JM |
2031 | if (WARN_ON(start != aligned_start)) { |
2032 | len -= aligned_start - start; | |
2033 | len = round_down(len, 1 << 9); | |
2034 | start = aligned_start; | |
2035 | } | |
d04c6b88 | 2036 | |
4d89d377 | 2037 | *discarded_bytes = 0; |
86557861 JM |
2038 | |
2039 | if (!len) | |
2040 | return 0; | |
2041 | ||
2042 | end = start + len; | |
2043 | bytes_left = len; | |
2044 | ||
2045 | /* Skip any superblocks on this device. */ | |
2046 | for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) { | |
2047 | u64 sb_start = btrfs_sb_offset(j); | |
2048 | u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE; | |
2049 | u64 size = sb_start - start; | |
2050 | ||
2051 | if (!in_range(sb_start, start, bytes_left) && | |
2052 | !in_range(sb_end, start, bytes_left) && | |
2053 | !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE)) | |
2054 | continue; | |
2055 | ||
2056 | /* | |
2057 | * Superblock spans beginning of range. Adjust start and | |
2058 | * try again. | |
2059 | */ | |
2060 | if (sb_start <= start) { | |
2061 | start += sb_end - start; | |
2062 | if (start > end) { | |
2063 | bytes_left = 0; | |
2064 | break; | |
2065 | } | |
2066 | bytes_left = end - start; | |
2067 | continue; | |
2068 | } | |
2069 | ||
2070 | if (size) { | |
2071 | ret = blkdev_issue_discard(bdev, start >> 9, size >> 9, | |
2072 | GFP_NOFS, 0); | |
2073 | if (!ret) | |
2074 | *discarded_bytes += size; | |
2075 | else if (ret != -EOPNOTSUPP) | |
2076 | return ret; | |
2077 | } | |
2078 | ||
2079 | start = sb_end; | |
2080 | if (start > end) { | |
2081 | bytes_left = 0; | |
2082 | break; | |
2083 | } | |
2084 | bytes_left = end - start; | |
2085 | } | |
2086 | ||
2087 | if (bytes_left) { | |
2088 | ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9, | |
4d89d377 JM |
2089 | GFP_NOFS, 0); |
2090 | if (!ret) | |
86557861 | 2091 | *discarded_bytes += bytes_left; |
4d89d377 | 2092 | } |
d04c6b88 | 2093 | return ret; |
5d4f98a2 | 2094 | } |
5d4f98a2 | 2095 | |
2ff7e61e | 2096 | int btrfs_discard_extent(struct btrfs_fs_info *fs_info, u64 bytenr, |
1edb647b | 2097 | u64 num_bytes, u64 *actual_bytes) |
5d4f98a2 | 2098 | { |
5d4f98a2 | 2099 | int ret; |
5378e607 | 2100 | u64 discarded_bytes = 0; |
a1d3c478 | 2101 | struct btrfs_bio *bbio = NULL; |
5d4f98a2 | 2102 | |
e244a0ae | 2103 | |
2999241d FM |
2104 | /* |
2105 | * Avoid races with device replace and make sure our bbio has devices | |
2106 | * associated to its stripes that don't go away while we are discarding. | |
2107 | */ | |
0b246afa | 2108 | btrfs_bio_counter_inc_blocked(fs_info); |
5d4f98a2 | 2109 | /* Tell the block device(s) that the sectors can be discarded */ |
0b246afa JM |
2110 | ret = btrfs_map_block(fs_info, BTRFS_MAP_DISCARD, bytenr, &num_bytes, |
2111 | &bbio, 0); | |
79787eaa | 2112 | /* Error condition is -ENOMEM */ |
5d4f98a2 | 2113 | if (!ret) { |
a1d3c478 | 2114 | struct btrfs_bio_stripe *stripe = bbio->stripes; |
5d4f98a2 YZ |
2115 | int i; |
2116 | ||
5d4f98a2 | 2117 | |
a1d3c478 | 2118 | for (i = 0; i < bbio->num_stripes; i++, stripe++) { |
d04c6b88 | 2119 | u64 bytes; |
38b5f68e AJ |
2120 | struct request_queue *req_q; |
2121 | ||
627e0873 FM |
2122 | if (!stripe->dev->bdev) { |
2123 | ASSERT(btrfs_test_opt(fs_info, DEGRADED)); | |
2124 | continue; | |
2125 | } | |
38b5f68e AJ |
2126 | req_q = bdev_get_queue(stripe->dev->bdev); |
2127 | if (!blk_queue_discard(req_q)) | |
d5e2003c JB |
2128 | continue; |
2129 | ||
5378e607 LD |
2130 | ret = btrfs_issue_discard(stripe->dev->bdev, |
2131 | stripe->physical, | |
d04c6b88 JM |
2132 | stripe->length, |
2133 | &bytes); | |
5378e607 | 2134 | if (!ret) |
d04c6b88 | 2135 | discarded_bytes += bytes; |
5378e607 | 2136 | else if (ret != -EOPNOTSUPP) |
79787eaa | 2137 | break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */ |
d5e2003c JB |
2138 | |
2139 | /* | |
2140 | * Just in case we get back EOPNOTSUPP for some reason, | |
2141 | * just ignore the return value so we don't screw up | |
2142 | * people calling discard_extent. | |
2143 | */ | |
2144 | ret = 0; | |
5d4f98a2 | 2145 | } |
6e9606d2 | 2146 | btrfs_put_bbio(bbio); |
5d4f98a2 | 2147 | } |
0b246afa | 2148 | btrfs_bio_counter_dec(fs_info); |
5378e607 LD |
2149 | |
2150 | if (actual_bytes) | |
2151 | *actual_bytes = discarded_bytes; | |
2152 | ||
5d4f98a2 | 2153 | |
53b381b3 DW |
2154 | if (ret == -EOPNOTSUPP) |
2155 | ret = 0; | |
5d4f98a2 | 2156 | return ret; |
5d4f98a2 YZ |
2157 | } |
2158 | ||
79787eaa | 2159 | /* Can return -ENOMEM */ |
5d4f98a2 | 2160 | int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
84f7d8e6 | 2161 | struct btrfs_root *root, |
5d4f98a2 | 2162 | u64 bytenr, u64 num_bytes, u64 parent, |
b06c4bf5 | 2163 | u64 root_objectid, u64 owner, u64 offset) |
5d4f98a2 | 2164 | { |
84f7d8e6 | 2165 | struct btrfs_fs_info *fs_info = root->fs_info; |
d7eae340 | 2166 | int old_ref_mod, new_ref_mod; |
5d4f98a2 | 2167 | int ret; |
66d7e7f0 | 2168 | |
5d4f98a2 YZ |
2169 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID && |
2170 | root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
2171 | ||
fd708b81 JB |
2172 | btrfs_ref_tree_mod(root, bytenr, num_bytes, parent, root_objectid, |
2173 | owner, offset, BTRFS_ADD_DELAYED_REF); | |
2174 | ||
5d4f98a2 | 2175 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { |
44e1c47d | 2176 | ret = btrfs_add_delayed_tree_ref(trans, bytenr, |
7be07912 OS |
2177 | num_bytes, parent, |
2178 | root_objectid, (int)owner, | |
2179 | BTRFS_ADD_DELAYED_REF, NULL, | |
d7eae340 | 2180 | &old_ref_mod, &new_ref_mod); |
5d4f98a2 | 2181 | } else { |
88a979c6 | 2182 | ret = btrfs_add_delayed_data_ref(trans, bytenr, |
7be07912 OS |
2183 | num_bytes, parent, |
2184 | root_objectid, owner, offset, | |
d7eae340 OS |
2185 | 0, BTRFS_ADD_DELAYED_REF, |
2186 | &old_ref_mod, &new_ref_mod); | |
5d4f98a2 | 2187 | } |
d7eae340 | 2188 | |
29d2b84c NB |
2189 | if (ret == 0 && old_ref_mod < 0 && new_ref_mod >= 0) { |
2190 | bool metadata = owner < BTRFS_FIRST_FREE_OBJECTID; | |
2191 | ||
2192 | add_pinned_bytes(fs_info, -num_bytes, metadata, root_objectid); | |
2193 | } | |
d7eae340 | 2194 | |
5d4f98a2 YZ |
2195 | return ret; |
2196 | } | |
2197 | ||
bd3c685e NB |
2198 | /* |
2199 | * __btrfs_inc_extent_ref - insert backreference for a given extent | |
2200 | * | |
2201 | * @trans: Handle of transaction | |
2202 | * | |
2203 | * @node: The delayed ref node used to get the bytenr/length for | |
2204 | * extent whose references are incremented. | |
2205 | * | |
2206 | * @parent: If this is a shared extent (BTRFS_SHARED_DATA_REF_KEY/ | |
2207 | * BTRFS_SHARED_BLOCK_REF_KEY) then it holds the logical | |
2208 | * bytenr of the parent block. Since new extents are always | |
2209 | * created with indirect references, this will only be the case | |
2210 | * when relocating a shared extent. In that case, root_objectid | |
2211 | * will be BTRFS_TREE_RELOC_OBJECTID. Otheriwse, parent must | |
2212 | * be 0 | |
2213 | * | |
2214 | * @root_objectid: The id of the root where this modification has originated, | |
2215 | * this can be either one of the well-known metadata trees or | |
2216 | * the subvolume id which references this extent. | |
2217 | * | |
2218 | * @owner: For data extents it is the inode number of the owning file. | |
2219 | * For metadata extents this parameter holds the level in the | |
2220 | * tree of the extent. | |
2221 | * | |
2222 | * @offset: For metadata extents the offset is ignored and is currently | |
2223 | * always passed as 0. For data extents it is the fileoffset | |
2224 | * this extent belongs to. | |
2225 | * | |
2226 | * @refs_to_add Number of references to add | |
2227 | * | |
2228 | * @extent_op Pointer to a structure, holding information necessary when | |
2229 | * updating a tree block's flags | |
2230 | * | |
2231 | */ | |
5d4f98a2 | 2232 | static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
c682f9b3 | 2233 | struct btrfs_delayed_ref_node *node, |
5d4f98a2 YZ |
2234 | u64 parent, u64 root_objectid, |
2235 | u64 owner, u64 offset, int refs_to_add, | |
2236 | struct btrfs_delayed_extent_op *extent_op) | |
2237 | { | |
2590d0f1 | 2238 | struct btrfs_fs_info *fs_info = trans->fs_info; |
5d4f98a2 YZ |
2239 | struct btrfs_path *path; |
2240 | struct extent_buffer *leaf; | |
2241 | struct btrfs_extent_item *item; | |
fcebe456 | 2242 | struct btrfs_key key; |
c682f9b3 QW |
2243 | u64 bytenr = node->bytenr; |
2244 | u64 num_bytes = node->num_bytes; | |
5d4f98a2 YZ |
2245 | u64 refs; |
2246 | int ret; | |
5d4f98a2 YZ |
2247 | |
2248 | path = btrfs_alloc_path(); | |
2249 | if (!path) | |
2250 | return -ENOMEM; | |
2251 | ||
e4058b54 | 2252 | path->reada = READA_FORWARD; |
5d4f98a2 YZ |
2253 | path->leave_spinning = 1; |
2254 | /* this will setup the path even if it fails to insert the back ref */ | |
87bde3cd JM |
2255 | ret = insert_inline_extent_backref(trans, fs_info, path, bytenr, |
2256 | num_bytes, parent, root_objectid, | |
2257 | owner, offset, | |
5d4f98a2 | 2258 | refs_to_add, extent_op); |
0ed4792a | 2259 | if ((ret < 0 && ret != -EAGAIN) || !ret) |
5d4f98a2 | 2260 | goto out; |
fcebe456 JB |
2261 | |
2262 | /* | |
2263 | * Ok we had -EAGAIN which means we didn't have space to insert and | |
2264 | * inline extent ref, so just update the reference count and add a | |
2265 | * normal backref. | |
2266 | */ | |
5d4f98a2 | 2267 | leaf = path->nodes[0]; |
fcebe456 | 2268 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
5d4f98a2 YZ |
2269 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
2270 | refs = btrfs_extent_refs(leaf, item); | |
2271 | btrfs_set_extent_refs(leaf, item, refs + refs_to_add); | |
2272 | if (extent_op) | |
2273 | __run_delayed_extent_op(extent_op, leaf, item); | |
56bec294 | 2274 | |
5d4f98a2 | 2275 | btrfs_mark_buffer_dirty(leaf); |
b3b4aa74 | 2276 | btrfs_release_path(path); |
56bec294 | 2277 | |
e4058b54 | 2278 | path->reada = READA_FORWARD; |
b9473439 | 2279 | path->leave_spinning = 1; |
56bec294 | 2280 | /* now insert the actual backref */ |
37593410 NB |
2281 | ret = insert_extent_backref(trans, path, bytenr, parent, root_objectid, |
2282 | owner, offset, refs_to_add); | |
79787eaa | 2283 | if (ret) |
66642832 | 2284 | btrfs_abort_transaction(trans, ret); |
5d4f98a2 | 2285 | out: |
56bec294 | 2286 | btrfs_free_path(path); |
30d133fc | 2287 | return ret; |
56bec294 CM |
2288 | } |
2289 | ||
5d4f98a2 | 2290 | static int run_delayed_data_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
2291 | struct btrfs_delayed_ref_node *node, |
2292 | struct btrfs_delayed_extent_op *extent_op, | |
2293 | int insert_reserved) | |
56bec294 | 2294 | { |
5d4f98a2 YZ |
2295 | int ret = 0; |
2296 | struct btrfs_delayed_data_ref *ref; | |
2297 | struct btrfs_key ins; | |
2298 | u64 parent = 0; | |
2299 | u64 ref_root = 0; | |
2300 | u64 flags = 0; | |
2301 | ||
2302 | ins.objectid = node->bytenr; | |
2303 | ins.offset = node->num_bytes; | |
2304 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
2305 | ||
2306 | ref = btrfs_delayed_node_to_data_ref(node); | |
2bf98ef3 | 2307 | trace_run_delayed_data_ref(trans->fs_info, node, ref, node->action); |
599c75ec | 2308 | |
5d4f98a2 YZ |
2309 | if (node->type == BTRFS_SHARED_DATA_REF_KEY) |
2310 | parent = ref->parent; | |
fcebe456 | 2311 | ref_root = ref->root; |
5d4f98a2 YZ |
2312 | |
2313 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
3173a18f | 2314 | if (extent_op) |
5d4f98a2 | 2315 | flags |= extent_op->flags_to_set; |
ef89b824 NB |
2316 | ret = alloc_reserved_file_extent(trans, parent, ref_root, |
2317 | flags, ref->objectid, | |
2318 | ref->offset, &ins, | |
2319 | node->ref_mod); | |
5d4f98a2 | 2320 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2590d0f1 NB |
2321 | ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root, |
2322 | ref->objectid, ref->offset, | |
2323 | node->ref_mod, extent_op); | |
5d4f98a2 | 2324 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
e72cb923 | 2325 | ret = __btrfs_free_extent(trans, node, parent, |
5d4f98a2 YZ |
2326 | ref_root, ref->objectid, |
2327 | ref->offset, node->ref_mod, | |
c682f9b3 | 2328 | extent_op); |
5d4f98a2 YZ |
2329 | } else { |
2330 | BUG(); | |
2331 | } | |
2332 | return ret; | |
2333 | } | |
2334 | ||
2335 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
2336 | struct extent_buffer *leaf, | |
2337 | struct btrfs_extent_item *ei) | |
2338 | { | |
2339 | u64 flags = btrfs_extent_flags(leaf, ei); | |
2340 | if (extent_op->update_flags) { | |
2341 | flags |= extent_op->flags_to_set; | |
2342 | btrfs_set_extent_flags(leaf, ei, flags); | |
2343 | } | |
2344 | ||
2345 | if (extent_op->update_key) { | |
2346 | struct btrfs_tree_block_info *bi; | |
2347 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)); | |
2348 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
2349 | btrfs_set_tree_block_key(leaf, bi, &extent_op->key); | |
2350 | } | |
2351 | } | |
2352 | ||
2353 | static int run_delayed_extent_op(struct btrfs_trans_handle *trans, | |
d278850e | 2354 | struct btrfs_delayed_ref_head *head, |
5d4f98a2 YZ |
2355 | struct btrfs_delayed_extent_op *extent_op) |
2356 | { | |
20b9a2d6 | 2357 | struct btrfs_fs_info *fs_info = trans->fs_info; |
5d4f98a2 YZ |
2358 | struct btrfs_key key; |
2359 | struct btrfs_path *path; | |
2360 | struct btrfs_extent_item *ei; | |
2361 | struct extent_buffer *leaf; | |
2362 | u32 item_size; | |
56bec294 | 2363 | int ret; |
5d4f98a2 | 2364 | int err = 0; |
b1c79e09 | 2365 | int metadata = !extent_op->is_data; |
5d4f98a2 | 2366 | |
79787eaa JM |
2367 | if (trans->aborted) |
2368 | return 0; | |
2369 | ||
0b246afa | 2370 | if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) |
3173a18f JB |
2371 | metadata = 0; |
2372 | ||
5d4f98a2 YZ |
2373 | path = btrfs_alloc_path(); |
2374 | if (!path) | |
2375 | return -ENOMEM; | |
2376 | ||
d278850e | 2377 | key.objectid = head->bytenr; |
5d4f98a2 | 2378 | |
3173a18f | 2379 | if (metadata) { |
3173a18f | 2380 | key.type = BTRFS_METADATA_ITEM_KEY; |
b1c79e09 | 2381 | key.offset = extent_op->level; |
3173a18f JB |
2382 | } else { |
2383 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
d278850e | 2384 | key.offset = head->num_bytes; |
3173a18f JB |
2385 | } |
2386 | ||
2387 | again: | |
e4058b54 | 2388 | path->reada = READA_FORWARD; |
5d4f98a2 | 2389 | path->leave_spinning = 1; |
0b246afa | 2390 | ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 1); |
5d4f98a2 YZ |
2391 | if (ret < 0) { |
2392 | err = ret; | |
2393 | goto out; | |
2394 | } | |
2395 | if (ret > 0) { | |
3173a18f | 2396 | if (metadata) { |
55994887 FDBM |
2397 | if (path->slots[0] > 0) { |
2398 | path->slots[0]--; | |
2399 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
2400 | path->slots[0]); | |
d278850e | 2401 | if (key.objectid == head->bytenr && |
55994887 | 2402 | key.type == BTRFS_EXTENT_ITEM_KEY && |
d278850e | 2403 | key.offset == head->num_bytes) |
55994887 FDBM |
2404 | ret = 0; |
2405 | } | |
2406 | if (ret > 0) { | |
2407 | btrfs_release_path(path); | |
2408 | metadata = 0; | |
3173a18f | 2409 | |
d278850e JB |
2410 | key.objectid = head->bytenr; |
2411 | key.offset = head->num_bytes; | |
55994887 FDBM |
2412 | key.type = BTRFS_EXTENT_ITEM_KEY; |
2413 | goto again; | |
2414 | } | |
2415 | } else { | |
2416 | err = -EIO; | |
2417 | goto out; | |
3173a18f | 2418 | } |
5d4f98a2 YZ |
2419 | } |
2420 | ||
2421 | leaf = path->nodes[0]; | |
2422 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2423 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
2424 | if (item_size < sizeof(*ei)) { | |
87bde3cd | 2425 | ret = convert_extent_item_v0(trans, fs_info, path, (u64)-1, 0); |
5d4f98a2 YZ |
2426 | if (ret < 0) { |
2427 | err = ret; | |
2428 | goto out; | |
2429 | } | |
2430 | leaf = path->nodes[0]; | |
2431 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2432 | } | |
2433 | #endif | |
2434 | BUG_ON(item_size < sizeof(*ei)); | |
2435 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
2436 | __run_delayed_extent_op(extent_op, leaf, ei); | |
56bec294 | 2437 | |
5d4f98a2 YZ |
2438 | btrfs_mark_buffer_dirty(leaf); |
2439 | out: | |
2440 | btrfs_free_path(path); | |
2441 | return err; | |
56bec294 CM |
2442 | } |
2443 | ||
5d4f98a2 | 2444 | static int run_delayed_tree_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
2445 | struct btrfs_delayed_ref_node *node, |
2446 | struct btrfs_delayed_extent_op *extent_op, | |
2447 | int insert_reserved) | |
56bec294 CM |
2448 | { |
2449 | int ret = 0; | |
5d4f98a2 | 2450 | struct btrfs_delayed_tree_ref *ref; |
5d4f98a2 YZ |
2451 | u64 parent = 0; |
2452 | u64 ref_root = 0; | |
56bec294 | 2453 | |
5d4f98a2 | 2454 | ref = btrfs_delayed_node_to_tree_ref(node); |
f97806f2 | 2455 | trace_run_delayed_tree_ref(trans->fs_info, node, ref, node->action); |
599c75ec | 2456 | |
5d4f98a2 YZ |
2457 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) |
2458 | parent = ref->parent; | |
fcebe456 | 2459 | ref_root = ref->root; |
5d4f98a2 | 2460 | |
02794222 | 2461 | if (node->ref_mod != 1) { |
f97806f2 | 2462 | btrfs_err(trans->fs_info, |
02794222 LB |
2463 | "btree block(%llu) has %d references rather than 1: action %d ref_root %llu parent %llu", |
2464 | node->bytenr, node->ref_mod, node->action, ref_root, | |
2465 | parent); | |
2466 | return -EIO; | |
2467 | } | |
5d4f98a2 | 2468 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { |
3173a18f | 2469 | BUG_ON(!extent_op || !extent_op->update_flags); |
21ebfbe7 | 2470 | ret = alloc_reserved_tree_block(trans, node, extent_op); |
5d4f98a2 | 2471 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2590d0f1 NB |
2472 | ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root, |
2473 | ref->level, 0, 1, extent_op); | |
5d4f98a2 | 2474 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
e72cb923 | 2475 | ret = __btrfs_free_extent(trans, node, parent, ref_root, |
c682f9b3 | 2476 | ref->level, 0, 1, extent_op); |
5d4f98a2 YZ |
2477 | } else { |
2478 | BUG(); | |
2479 | } | |
56bec294 CM |
2480 | return ret; |
2481 | } | |
2482 | ||
2483 | /* helper function to actually process a single delayed ref entry */ | |
5d4f98a2 | 2484 | static int run_one_delayed_ref(struct btrfs_trans_handle *trans, |
2ff7e61e | 2485 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
2486 | struct btrfs_delayed_ref_node *node, |
2487 | struct btrfs_delayed_extent_op *extent_op, | |
2488 | int insert_reserved) | |
56bec294 | 2489 | { |
79787eaa JM |
2490 | int ret = 0; |
2491 | ||
857cc2fc JB |
2492 | if (trans->aborted) { |
2493 | if (insert_reserved) | |
2ff7e61e | 2494 | btrfs_pin_extent(fs_info, node->bytenr, |
857cc2fc | 2495 | node->num_bytes, 1); |
79787eaa | 2496 | return 0; |
857cc2fc | 2497 | } |
79787eaa | 2498 | |
5d4f98a2 YZ |
2499 | if (node->type == BTRFS_TREE_BLOCK_REF_KEY || |
2500 | node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
f97806f2 | 2501 | ret = run_delayed_tree_ref(trans, node, extent_op, |
5d4f98a2 YZ |
2502 | insert_reserved); |
2503 | else if (node->type == BTRFS_EXTENT_DATA_REF_KEY || | |
2504 | node->type == BTRFS_SHARED_DATA_REF_KEY) | |
2bf98ef3 | 2505 | ret = run_delayed_data_ref(trans, node, extent_op, |
5d4f98a2 YZ |
2506 | insert_reserved); |
2507 | else | |
2508 | BUG(); | |
2509 | return ret; | |
56bec294 CM |
2510 | } |
2511 | ||
c6fc2454 | 2512 | static inline struct btrfs_delayed_ref_node * |
56bec294 CM |
2513 | select_delayed_ref(struct btrfs_delayed_ref_head *head) |
2514 | { | |
cffc3374 FM |
2515 | struct btrfs_delayed_ref_node *ref; |
2516 | ||
0e0adbcf | 2517 | if (RB_EMPTY_ROOT(&head->ref_tree)) |
c6fc2454 | 2518 | return NULL; |
d7df2c79 | 2519 | |
cffc3374 FM |
2520 | /* |
2521 | * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first. | |
2522 | * This is to prevent a ref count from going down to zero, which deletes | |
2523 | * the extent item from the extent tree, when there still are references | |
2524 | * to add, which would fail because they would not find the extent item. | |
2525 | */ | |
1d57ee94 WX |
2526 | if (!list_empty(&head->ref_add_list)) |
2527 | return list_first_entry(&head->ref_add_list, | |
2528 | struct btrfs_delayed_ref_node, add_list); | |
2529 | ||
0e0adbcf JB |
2530 | ref = rb_entry(rb_first(&head->ref_tree), |
2531 | struct btrfs_delayed_ref_node, ref_node); | |
1d57ee94 WX |
2532 | ASSERT(list_empty(&ref->add_list)); |
2533 | return ref; | |
56bec294 CM |
2534 | } |
2535 | ||
2eadaa22 JB |
2536 | static void unselect_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs, |
2537 | struct btrfs_delayed_ref_head *head) | |
2538 | { | |
2539 | spin_lock(&delayed_refs->lock); | |
2540 | head->processing = 0; | |
2541 | delayed_refs->num_heads_ready++; | |
2542 | spin_unlock(&delayed_refs->lock); | |
2543 | btrfs_delayed_ref_unlock(head); | |
2544 | } | |
2545 | ||
b00e6250 | 2546 | static int cleanup_extent_op(struct btrfs_trans_handle *trans, |
b00e6250 JB |
2547 | struct btrfs_delayed_ref_head *head) |
2548 | { | |
2549 | struct btrfs_delayed_extent_op *extent_op = head->extent_op; | |
2550 | int ret; | |
2551 | ||
2552 | if (!extent_op) | |
2553 | return 0; | |
2554 | head->extent_op = NULL; | |
2555 | if (head->must_insert_reserved) { | |
2556 | btrfs_free_delayed_extent_op(extent_op); | |
2557 | return 0; | |
2558 | } | |
2559 | spin_unlock(&head->lock); | |
20b9a2d6 | 2560 | ret = run_delayed_extent_op(trans, head, extent_op); |
b00e6250 JB |
2561 | btrfs_free_delayed_extent_op(extent_op); |
2562 | return ret ? ret : 1; | |
2563 | } | |
2564 | ||
194ab0bc | 2565 | static int cleanup_ref_head(struct btrfs_trans_handle *trans, |
194ab0bc JB |
2566 | struct btrfs_delayed_ref_head *head) |
2567 | { | |
f9871edd NB |
2568 | |
2569 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
194ab0bc JB |
2570 | struct btrfs_delayed_ref_root *delayed_refs; |
2571 | int ret; | |
2572 | ||
2573 | delayed_refs = &trans->transaction->delayed_refs; | |
2574 | ||
c4d56d4a | 2575 | ret = cleanup_extent_op(trans, head); |
194ab0bc JB |
2576 | if (ret < 0) { |
2577 | unselect_delayed_ref_head(delayed_refs, head); | |
2578 | btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret); | |
2579 | return ret; | |
2580 | } else if (ret) { | |
2581 | return ret; | |
2582 | } | |
2583 | ||
2584 | /* | |
2585 | * Need to drop our head ref lock and re-acquire the delayed ref lock | |
2586 | * and then re-check to make sure nobody got added. | |
2587 | */ | |
2588 | spin_unlock(&head->lock); | |
2589 | spin_lock(&delayed_refs->lock); | |
2590 | spin_lock(&head->lock); | |
0e0adbcf | 2591 | if (!RB_EMPTY_ROOT(&head->ref_tree) || head->extent_op) { |
194ab0bc JB |
2592 | spin_unlock(&head->lock); |
2593 | spin_unlock(&delayed_refs->lock); | |
2594 | return 1; | |
2595 | } | |
194ab0bc JB |
2596 | delayed_refs->num_heads--; |
2597 | rb_erase(&head->href_node, &delayed_refs->href_root); | |
d278850e | 2598 | RB_CLEAR_NODE(&head->href_node); |
c1103f7a | 2599 | spin_unlock(&head->lock); |
1e7a1421 | 2600 | spin_unlock(&delayed_refs->lock); |
c1103f7a JB |
2601 | atomic_dec(&delayed_refs->num_entries); |
2602 | ||
d278850e | 2603 | trace_run_delayed_ref_head(fs_info, head, 0); |
c1103f7a JB |
2604 | |
2605 | if (head->total_ref_mod < 0) { | |
5e388e95 NB |
2606 | struct btrfs_space_info *space_info; |
2607 | u64 flags; | |
c1103f7a | 2608 | |
5e388e95 NB |
2609 | if (head->is_data) |
2610 | flags = BTRFS_BLOCK_GROUP_DATA; | |
2611 | else if (head->is_system) | |
2612 | flags = BTRFS_BLOCK_GROUP_SYSTEM; | |
2613 | else | |
2614 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
2615 | space_info = __find_space_info(fs_info, flags); | |
2616 | ASSERT(space_info); | |
2617 | percpu_counter_add(&space_info->total_bytes_pinned, | |
d278850e | 2618 | -head->num_bytes); |
c1103f7a JB |
2619 | |
2620 | if (head->is_data) { | |
2621 | spin_lock(&delayed_refs->lock); | |
d278850e | 2622 | delayed_refs->pending_csums -= head->num_bytes; |
c1103f7a JB |
2623 | spin_unlock(&delayed_refs->lock); |
2624 | } | |
2625 | } | |
2626 | ||
2627 | if (head->must_insert_reserved) { | |
d278850e JB |
2628 | btrfs_pin_extent(fs_info, head->bytenr, |
2629 | head->num_bytes, 1); | |
c1103f7a | 2630 | if (head->is_data) { |
d278850e JB |
2631 | ret = btrfs_del_csums(trans, fs_info, head->bytenr, |
2632 | head->num_bytes); | |
c1103f7a JB |
2633 | } |
2634 | } | |
2635 | ||
2636 | /* Also free its reserved qgroup space */ | |
2637 | btrfs_qgroup_free_delayed_ref(fs_info, head->qgroup_ref_root, | |
2638 | head->qgroup_reserved); | |
2639 | btrfs_delayed_ref_unlock(head); | |
d278850e | 2640 | btrfs_put_delayed_ref_head(head); |
194ab0bc JB |
2641 | return 0; |
2642 | } | |
2643 | ||
79787eaa JM |
2644 | /* |
2645 | * Returns 0 on success or if called with an already aborted transaction. | |
2646 | * Returns -ENOMEM or -EIO on failure and will abort the transaction. | |
2647 | */ | |
d7df2c79 | 2648 | static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, |
d7df2c79 | 2649 | unsigned long nr) |
56bec294 | 2650 | { |
0a1e458a | 2651 | struct btrfs_fs_info *fs_info = trans->fs_info; |
56bec294 CM |
2652 | struct btrfs_delayed_ref_root *delayed_refs; |
2653 | struct btrfs_delayed_ref_node *ref; | |
2654 | struct btrfs_delayed_ref_head *locked_ref = NULL; | |
5d4f98a2 | 2655 | struct btrfs_delayed_extent_op *extent_op; |
0a2b2a84 | 2656 | ktime_t start = ktime_get(); |
56bec294 | 2657 | int ret; |
d7df2c79 | 2658 | unsigned long count = 0; |
0a2b2a84 | 2659 | unsigned long actual_count = 0; |
56bec294 | 2660 | int must_insert_reserved = 0; |
56bec294 CM |
2661 | |
2662 | delayed_refs = &trans->transaction->delayed_refs; | |
56bec294 CM |
2663 | while (1) { |
2664 | if (!locked_ref) { | |
d7df2c79 | 2665 | if (count >= nr) |
56bec294 | 2666 | break; |
56bec294 | 2667 | |
d7df2c79 JB |
2668 | spin_lock(&delayed_refs->lock); |
2669 | locked_ref = btrfs_select_ref_head(trans); | |
2670 | if (!locked_ref) { | |
2671 | spin_unlock(&delayed_refs->lock); | |
2672 | break; | |
2673 | } | |
c3e69d58 CM |
2674 | |
2675 | /* grab the lock that says we are going to process | |
2676 | * all the refs for this head */ | |
2677 | ret = btrfs_delayed_ref_lock(trans, locked_ref); | |
d7df2c79 | 2678 | spin_unlock(&delayed_refs->lock); |
c3e69d58 CM |
2679 | /* |
2680 | * we may have dropped the spin lock to get the head | |
2681 | * mutex lock, and that might have given someone else | |
2682 | * time to free the head. If that's true, it has been | |
2683 | * removed from our list and we can move on. | |
2684 | */ | |
2685 | if (ret == -EAGAIN) { | |
2686 | locked_ref = NULL; | |
2687 | count++; | |
2688 | continue; | |
56bec294 CM |
2689 | } |
2690 | } | |
a28ec197 | 2691 | |
2c3cf7d5 FM |
2692 | /* |
2693 | * We need to try and merge add/drops of the same ref since we | |
2694 | * can run into issues with relocate dropping the implicit ref | |
2695 | * and then it being added back again before the drop can | |
2696 | * finish. If we merged anything we need to re-loop so we can | |
2697 | * get a good ref. | |
2698 | * Or we can get node references of the same type that weren't | |
2699 | * merged when created due to bumps in the tree mod seq, and | |
2700 | * we need to merge them to prevent adding an inline extent | |
2701 | * backref before dropping it (triggering a BUG_ON at | |
2702 | * insert_inline_extent_backref()). | |
2703 | */ | |
d7df2c79 | 2704 | spin_lock(&locked_ref->lock); |
be97f133 | 2705 | btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref); |
ae1e206b | 2706 | |
d1270cd9 AJ |
2707 | ref = select_delayed_ref(locked_ref); |
2708 | ||
2709 | if (ref && ref->seq && | |
41d0bd3b | 2710 | btrfs_check_delayed_seq(fs_info, ref->seq)) { |
d7df2c79 | 2711 | spin_unlock(&locked_ref->lock); |
2eadaa22 | 2712 | unselect_delayed_ref_head(delayed_refs, locked_ref); |
d7df2c79 | 2713 | locked_ref = NULL; |
d1270cd9 | 2714 | cond_resched(); |
27a377db | 2715 | count++; |
d1270cd9 AJ |
2716 | continue; |
2717 | } | |
2718 | ||
c1103f7a JB |
2719 | /* |
2720 | * We're done processing refs in this ref_head, clean everything | |
2721 | * up and move on to the next ref_head. | |
2722 | */ | |
56bec294 | 2723 | if (!ref) { |
f9871edd | 2724 | ret = cleanup_ref_head(trans, locked_ref); |
194ab0bc | 2725 | if (ret > 0 ) { |
b00e6250 JB |
2726 | /* We dropped our lock, we need to loop. */ |
2727 | ret = 0; | |
d7df2c79 | 2728 | continue; |
194ab0bc JB |
2729 | } else if (ret) { |
2730 | return ret; | |
5d4f98a2 | 2731 | } |
c1103f7a JB |
2732 | locked_ref = NULL; |
2733 | count++; | |
2734 | continue; | |
2735 | } | |
02217ed2 | 2736 | |
c1103f7a JB |
2737 | actual_count++; |
2738 | ref->in_tree = 0; | |
0e0adbcf JB |
2739 | rb_erase(&ref->ref_node, &locked_ref->ref_tree); |
2740 | RB_CLEAR_NODE(&ref->ref_node); | |
c1103f7a JB |
2741 | if (!list_empty(&ref->add_list)) |
2742 | list_del(&ref->add_list); | |
2743 | /* | |
2744 | * When we play the delayed ref, also correct the ref_mod on | |
2745 | * head | |
2746 | */ | |
2747 | switch (ref->action) { | |
2748 | case BTRFS_ADD_DELAYED_REF: | |
2749 | case BTRFS_ADD_DELAYED_EXTENT: | |
d278850e | 2750 | locked_ref->ref_mod -= ref->ref_mod; |
c1103f7a JB |
2751 | break; |
2752 | case BTRFS_DROP_DELAYED_REF: | |
d278850e | 2753 | locked_ref->ref_mod += ref->ref_mod; |
c1103f7a JB |
2754 | break; |
2755 | default: | |
2756 | WARN_ON(1); | |
22cd2e7d | 2757 | } |
1ce7a5ec JB |
2758 | atomic_dec(&delayed_refs->num_entries); |
2759 | ||
b00e6250 JB |
2760 | /* |
2761 | * Record the must-insert_reserved flag before we drop the spin | |
2762 | * lock. | |
2763 | */ | |
2764 | must_insert_reserved = locked_ref->must_insert_reserved; | |
2765 | locked_ref->must_insert_reserved = 0; | |
2766 | ||
2767 | extent_op = locked_ref->extent_op; | |
2768 | locked_ref->extent_op = NULL; | |
d7df2c79 | 2769 | spin_unlock(&locked_ref->lock); |
925baedd | 2770 | |
2ff7e61e | 2771 | ret = run_one_delayed_ref(trans, fs_info, ref, extent_op, |
56bec294 | 2772 | must_insert_reserved); |
eb099670 | 2773 | |
78a6184a | 2774 | btrfs_free_delayed_extent_op(extent_op); |
79787eaa | 2775 | if (ret) { |
2eadaa22 | 2776 | unselect_delayed_ref_head(delayed_refs, locked_ref); |
093486c4 | 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 | btrfs_put_delayed_ref(ref); |
2784 | count++; | |
c3e69d58 | 2785 | cond_resched(); |
c3e69d58 | 2786 | } |
0a2b2a84 JB |
2787 | |
2788 | /* | |
2789 | * We don't want to include ref heads since we can have empty ref heads | |
2790 | * and those will drastically skew our runtime down since we just do | |
2791 | * accounting, no actual extent tree updates. | |
2792 | */ | |
2793 | if (actual_count > 0) { | |
2794 | u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start)); | |
2795 | u64 avg; | |
2796 | ||
2797 | /* | |
2798 | * We weigh the current average higher than our current runtime | |
2799 | * to avoid large swings in the average. | |
2800 | */ | |
2801 | spin_lock(&delayed_refs->lock); | |
2802 | avg = fs_info->avg_delayed_ref_runtime * 3 + runtime; | |
f8c269d7 | 2803 | fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */ |
0a2b2a84 JB |
2804 | spin_unlock(&delayed_refs->lock); |
2805 | } | |
d7df2c79 | 2806 | return 0; |
c3e69d58 CM |
2807 | } |
2808 | ||
709c0486 AJ |
2809 | #ifdef SCRAMBLE_DELAYED_REFS |
2810 | /* | |
2811 | * Normally delayed refs get processed in ascending bytenr order. This | |
2812 | * correlates in most cases to the order added. To expose dependencies on this | |
2813 | * order, we start to process the tree in the middle instead of the beginning | |
2814 | */ | |
2815 | static u64 find_middle(struct rb_root *root) | |
2816 | { | |
2817 | struct rb_node *n = root->rb_node; | |
2818 | struct btrfs_delayed_ref_node *entry; | |
2819 | int alt = 1; | |
2820 | u64 middle; | |
2821 | u64 first = 0, last = 0; | |
2822 | ||
2823 | n = rb_first(root); | |
2824 | if (n) { | |
2825 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2826 | first = entry->bytenr; | |
2827 | } | |
2828 | n = rb_last(root); | |
2829 | if (n) { | |
2830 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2831 | last = entry->bytenr; | |
2832 | } | |
2833 | n = root->rb_node; | |
2834 | ||
2835 | while (n) { | |
2836 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2837 | WARN_ON(!entry->in_tree); | |
2838 | ||
2839 | middle = entry->bytenr; | |
2840 | ||
2841 | if (alt) | |
2842 | n = n->rb_left; | |
2843 | else | |
2844 | n = n->rb_right; | |
2845 | ||
2846 | alt = 1 - alt; | |
2847 | } | |
2848 | return middle; | |
2849 | } | |
2850 | #endif | |
2851 | ||
2ff7e61e | 2852 | static inline u64 heads_to_leaves(struct btrfs_fs_info *fs_info, u64 heads) |
1be41b78 JB |
2853 | { |
2854 | u64 num_bytes; | |
2855 | ||
2856 | num_bytes = heads * (sizeof(struct btrfs_extent_item) + | |
2857 | sizeof(struct btrfs_extent_inline_ref)); | |
0b246afa | 2858 | if (!btrfs_fs_incompat(fs_info, SKINNY_METADATA)) |
1be41b78 JB |
2859 | num_bytes += heads * sizeof(struct btrfs_tree_block_info); |
2860 | ||
2861 | /* | |
2862 | * We don't ever fill up leaves all the way so multiply by 2 just to be | |
01327610 | 2863 | * closer to what we're really going to want to use. |
1be41b78 | 2864 | */ |
0b246afa | 2865 | return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(fs_info)); |
1be41b78 JB |
2866 | } |
2867 | ||
1262133b JB |
2868 | /* |
2869 | * Takes the number of bytes to be csumm'ed and figures out how many leaves it | |
2870 | * would require to store the csums for that many bytes. | |
2871 | */ | |
2ff7e61e | 2872 | u64 btrfs_csum_bytes_to_leaves(struct btrfs_fs_info *fs_info, u64 csum_bytes) |
1262133b JB |
2873 | { |
2874 | u64 csum_size; | |
2875 | u64 num_csums_per_leaf; | |
2876 | u64 num_csums; | |
2877 | ||
0b246afa | 2878 | csum_size = BTRFS_MAX_ITEM_SIZE(fs_info); |
1262133b | 2879 | num_csums_per_leaf = div64_u64(csum_size, |
0b246afa JM |
2880 | (u64)btrfs_super_csum_size(fs_info->super_copy)); |
2881 | num_csums = div64_u64(csum_bytes, fs_info->sectorsize); | |
1262133b JB |
2882 | num_csums += num_csums_per_leaf - 1; |
2883 | num_csums = div64_u64(num_csums, num_csums_per_leaf); | |
2884 | return num_csums; | |
2885 | } | |
2886 | ||
0a2b2a84 | 2887 | int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans, |
2ff7e61e | 2888 | struct btrfs_fs_info *fs_info) |
1be41b78 JB |
2889 | { |
2890 | struct btrfs_block_rsv *global_rsv; | |
2891 | u64 num_heads = trans->transaction->delayed_refs.num_heads_ready; | |
1262133b | 2892 | u64 csum_bytes = trans->transaction->delayed_refs.pending_csums; |
165c8b02 | 2893 | unsigned int num_dirty_bgs = trans->transaction->num_dirty_bgs; |
cb723e49 | 2894 | u64 num_bytes, num_dirty_bgs_bytes; |
1be41b78 JB |
2895 | int ret = 0; |
2896 | ||
0b246afa | 2897 | num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1); |
2ff7e61e | 2898 | num_heads = heads_to_leaves(fs_info, num_heads); |
1be41b78 | 2899 | if (num_heads > 1) |
0b246afa | 2900 | num_bytes += (num_heads - 1) * fs_info->nodesize; |
1be41b78 | 2901 | num_bytes <<= 1; |
2ff7e61e JM |
2902 | num_bytes += btrfs_csum_bytes_to_leaves(fs_info, csum_bytes) * |
2903 | fs_info->nodesize; | |
0b246afa | 2904 | num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(fs_info, |
cb723e49 | 2905 | num_dirty_bgs); |
0b246afa | 2906 | global_rsv = &fs_info->global_block_rsv; |
1be41b78 JB |
2907 | |
2908 | /* | |
2909 | * If we can't allocate any more chunks lets make sure we have _lots_ of | |
2910 | * wiggle room since running delayed refs can create more delayed refs. | |
2911 | */ | |
cb723e49 JB |
2912 | if (global_rsv->space_info->full) { |
2913 | num_dirty_bgs_bytes <<= 1; | |
1be41b78 | 2914 | num_bytes <<= 1; |
cb723e49 | 2915 | } |
1be41b78 JB |
2916 | |
2917 | spin_lock(&global_rsv->lock); | |
cb723e49 | 2918 | if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes) |
1be41b78 JB |
2919 | ret = 1; |
2920 | spin_unlock(&global_rsv->lock); | |
2921 | return ret; | |
2922 | } | |
2923 | ||
0a2b2a84 | 2924 | int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans, |
2ff7e61e | 2925 | struct btrfs_fs_info *fs_info) |
0a2b2a84 | 2926 | { |
0a2b2a84 JB |
2927 | u64 num_entries = |
2928 | atomic_read(&trans->transaction->delayed_refs.num_entries); | |
2929 | u64 avg_runtime; | |
a79b7d4b | 2930 | u64 val; |
0a2b2a84 JB |
2931 | |
2932 | smp_mb(); | |
2933 | avg_runtime = fs_info->avg_delayed_ref_runtime; | |
a79b7d4b | 2934 | val = num_entries * avg_runtime; |
dc1a90c6 | 2935 | if (val >= NSEC_PER_SEC) |
0a2b2a84 | 2936 | return 1; |
a79b7d4b CM |
2937 | if (val >= NSEC_PER_SEC / 2) |
2938 | return 2; | |
0a2b2a84 | 2939 | |
2ff7e61e | 2940 | return btrfs_check_space_for_delayed_refs(trans, fs_info); |
0a2b2a84 JB |
2941 | } |
2942 | ||
a79b7d4b CM |
2943 | struct async_delayed_refs { |
2944 | struct btrfs_root *root; | |
31b9655f | 2945 | u64 transid; |
a79b7d4b CM |
2946 | int count; |
2947 | int error; | |
2948 | int sync; | |
2949 | struct completion wait; | |
2950 | struct btrfs_work work; | |
2951 | }; | |
2952 | ||
2ff7e61e JM |
2953 | static inline struct async_delayed_refs * |
2954 | to_async_delayed_refs(struct btrfs_work *work) | |
2955 | { | |
2956 | return container_of(work, struct async_delayed_refs, work); | |
2957 | } | |
2958 | ||
a79b7d4b CM |
2959 | static void delayed_ref_async_start(struct btrfs_work *work) |
2960 | { | |
2ff7e61e | 2961 | struct async_delayed_refs *async = to_async_delayed_refs(work); |
a79b7d4b | 2962 | struct btrfs_trans_handle *trans; |
2ff7e61e | 2963 | struct btrfs_fs_info *fs_info = async->root->fs_info; |
a79b7d4b CM |
2964 | int ret; |
2965 | ||
0f873eca | 2966 | /* if the commit is already started, we don't need to wait here */ |
2ff7e61e | 2967 | if (btrfs_transaction_blocked(fs_info)) |
31b9655f | 2968 | goto done; |
31b9655f | 2969 | |
0f873eca CM |
2970 | trans = btrfs_join_transaction(async->root); |
2971 | if (IS_ERR(trans)) { | |
2972 | async->error = PTR_ERR(trans); | |
a79b7d4b CM |
2973 | goto done; |
2974 | } | |
2975 | ||
2976 | /* | |
01327610 | 2977 | * trans->sync means that when we call end_transaction, we won't |
a79b7d4b CM |
2978 | * wait on delayed refs |
2979 | */ | |
2980 | trans->sync = true; | |
0f873eca CM |
2981 | |
2982 | /* Don't bother flushing if we got into a different transaction */ | |
2983 | if (trans->transid > async->transid) | |
2984 | goto end; | |
2985 | ||
c79a70b1 | 2986 | ret = btrfs_run_delayed_refs(trans, async->count); |
a79b7d4b CM |
2987 | if (ret) |
2988 | async->error = ret; | |
0f873eca | 2989 | end: |
3a45bb20 | 2990 | ret = btrfs_end_transaction(trans); |
a79b7d4b CM |
2991 | if (ret && !async->error) |
2992 | async->error = ret; | |
2993 | done: | |
2994 | if (async->sync) | |
2995 | complete(&async->wait); | |
2996 | else | |
2997 | kfree(async); | |
2998 | } | |
2999 | ||
2ff7e61e | 3000 | int btrfs_async_run_delayed_refs(struct btrfs_fs_info *fs_info, |
31b9655f | 3001 | unsigned long count, u64 transid, int wait) |
a79b7d4b CM |
3002 | { |
3003 | struct async_delayed_refs *async; | |
3004 | int ret; | |
3005 | ||
3006 | async = kmalloc(sizeof(*async), GFP_NOFS); | |
3007 | if (!async) | |
3008 | return -ENOMEM; | |
3009 | ||
0b246afa | 3010 | async->root = fs_info->tree_root; |
a79b7d4b CM |
3011 | async->count = count; |
3012 | async->error = 0; | |
31b9655f | 3013 | async->transid = transid; |
a79b7d4b CM |
3014 | if (wait) |
3015 | async->sync = 1; | |
3016 | else | |
3017 | async->sync = 0; | |
3018 | init_completion(&async->wait); | |
3019 | ||
9e0af237 LB |
3020 | btrfs_init_work(&async->work, btrfs_extent_refs_helper, |
3021 | delayed_ref_async_start, NULL, NULL); | |
a79b7d4b | 3022 | |
0b246afa | 3023 | btrfs_queue_work(fs_info->extent_workers, &async->work); |
a79b7d4b CM |
3024 | |
3025 | if (wait) { | |
3026 | wait_for_completion(&async->wait); | |
3027 | ret = async->error; | |
3028 | kfree(async); | |
3029 | return ret; | |
3030 | } | |
3031 | return 0; | |
3032 | } | |
3033 | ||
c3e69d58 CM |
3034 | /* |
3035 | * this starts processing the delayed reference count updates and | |
3036 | * extent insertions we have queued up so far. count can be | |
3037 | * 0, which means to process everything in the tree at the start | |
3038 | * of the run (but not newly added entries), or it can be some target | |
3039 | * number you'd like to process. | |
79787eaa JM |
3040 | * |
3041 | * Returns 0 on success or if called with an aborted transaction | |
3042 | * Returns <0 on error and aborts the transaction | |
c3e69d58 CM |
3043 | */ |
3044 | int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, | |
c79a70b1 | 3045 | unsigned long count) |
c3e69d58 | 3046 | { |
c79a70b1 | 3047 | struct btrfs_fs_info *fs_info = trans->fs_info; |
c3e69d58 CM |
3048 | struct rb_node *node; |
3049 | struct btrfs_delayed_ref_root *delayed_refs; | |
c46effa6 | 3050 | struct btrfs_delayed_ref_head *head; |
c3e69d58 CM |
3051 | int ret; |
3052 | int run_all = count == (unsigned long)-1; | |
d9a0540a | 3053 | bool can_flush_pending_bgs = trans->can_flush_pending_bgs; |
c3e69d58 | 3054 | |
79787eaa JM |
3055 | /* We'll clean this up in btrfs_cleanup_transaction */ |
3056 | if (trans->aborted) | |
3057 | return 0; | |
3058 | ||
0b246afa | 3059 | if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags)) |
511711af CM |
3060 | return 0; |
3061 | ||
c3e69d58 | 3062 | delayed_refs = &trans->transaction->delayed_refs; |
26455d33 | 3063 | if (count == 0) |
d7df2c79 | 3064 | count = atomic_read(&delayed_refs->num_entries) * 2; |
bb721703 | 3065 | |
c3e69d58 | 3066 | again: |
709c0486 AJ |
3067 | #ifdef SCRAMBLE_DELAYED_REFS |
3068 | delayed_refs->run_delayed_start = find_middle(&delayed_refs->root); | |
3069 | #endif | |
d9a0540a | 3070 | trans->can_flush_pending_bgs = false; |
0a1e458a | 3071 | ret = __btrfs_run_delayed_refs(trans, count); |
d7df2c79 | 3072 | if (ret < 0) { |
66642832 | 3073 | btrfs_abort_transaction(trans, ret); |
d7df2c79 | 3074 | return ret; |
eb099670 | 3075 | } |
c3e69d58 | 3076 | |
56bec294 | 3077 | if (run_all) { |
d7df2c79 | 3078 | if (!list_empty(&trans->new_bgs)) |
6c686b35 | 3079 | btrfs_create_pending_block_groups(trans); |
ea658bad | 3080 | |
d7df2c79 | 3081 | spin_lock(&delayed_refs->lock); |
c46effa6 | 3082 | node = rb_first(&delayed_refs->href_root); |
d7df2c79 JB |
3083 | if (!node) { |
3084 | spin_unlock(&delayed_refs->lock); | |
56bec294 | 3085 | goto out; |
d7df2c79 | 3086 | } |
d278850e JB |
3087 | head = rb_entry(node, struct btrfs_delayed_ref_head, |
3088 | href_node); | |
3089 | refcount_inc(&head->refs); | |
3090 | spin_unlock(&delayed_refs->lock); | |
e9d0b13b | 3091 | |
d278850e JB |
3092 | /* Mutex was contended, block until it's released and retry. */ |
3093 | mutex_lock(&head->mutex); | |
3094 | mutex_unlock(&head->mutex); | |
56bec294 | 3095 | |
d278850e | 3096 | btrfs_put_delayed_ref_head(head); |
d7df2c79 | 3097 | cond_resched(); |
56bec294 | 3098 | goto again; |
5f39d397 | 3099 | } |
54aa1f4d | 3100 | out: |
d9a0540a | 3101 | trans->can_flush_pending_bgs = can_flush_pending_bgs; |
a28ec197 CM |
3102 | return 0; |
3103 | } | |
3104 | ||
5d4f98a2 | 3105 | int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans, |
2ff7e61e | 3106 | struct btrfs_fs_info *fs_info, |
5d4f98a2 | 3107 | u64 bytenr, u64 num_bytes, u64 flags, |
b1c79e09 | 3108 | int level, int is_data) |
5d4f98a2 YZ |
3109 | { |
3110 | struct btrfs_delayed_extent_op *extent_op; | |
3111 | int ret; | |
3112 | ||
78a6184a | 3113 | extent_op = btrfs_alloc_delayed_extent_op(); |
5d4f98a2 YZ |
3114 | if (!extent_op) |
3115 | return -ENOMEM; | |
3116 | ||
3117 | extent_op->flags_to_set = flags; | |
35b3ad50 DS |
3118 | extent_op->update_flags = true; |
3119 | extent_op->update_key = false; | |
3120 | extent_op->is_data = is_data ? true : false; | |
b1c79e09 | 3121 | extent_op->level = level; |
5d4f98a2 | 3122 | |
0b246afa | 3123 | ret = btrfs_add_delayed_extent_op(fs_info, trans, bytenr, |
66d7e7f0 | 3124 | num_bytes, extent_op); |
5d4f98a2 | 3125 | if (ret) |
78a6184a | 3126 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 YZ |
3127 | return ret; |
3128 | } | |
3129 | ||
e4c3b2dc | 3130 | static noinline int check_delayed_ref(struct btrfs_root *root, |
5d4f98a2 YZ |
3131 | struct btrfs_path *path, |
3132 | u64 objectid, u64 offset, u64 bytenr) | |
3133 | { | |
3134 | struct btrfs_delayed_ref_head *head; | |
3135 | struct btrfs_delayed_ref_node *ref; | |
3136 | struct btrfs_delayed_data_ref *data_ref; | |
3137 | struct btrfs_delayed_ref_root *delayed_refs; | |
e4c3b2dc | 3138 | struct btrfs_transaction *cur_trans; |
0e0adbcf | 3139 | struct rb_node *node; |
5d4f98a2 YZ |
3140 | int ret = 0; |
3141 | ||
998ac6d2 | 3142 | spin_lock(&root->fs_info->trans_lock); |
e4c3b2dc | 3143 | cur_trans = root->fs_info->running_transaction; |
998ac6d2 | 3144 | if (cur_trans) |
3145 | refcount_inc(&cur_trans->use_count); | |
3146 | spin_unlock(&root->fs_info->trans_lock); | |
e4c3b2dc LB |
3147 | if (!cur_trans) |
3148 | return 0; | |
3149 | ||
3150 | delayed_refs = &cur_trans->delayed_refs; | |
5d4f98a2 | 3151 | spin_lock(&delayed_refs->lock); |
f72ad18e | 3152 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
d7df2c79 JB |
3153 | if (!head) { |
3154 | spin_unlock(&delayed_refs->lock); | |
998ac6d2 | 3155 | btrfs_put_transaction(cur_trans); |
d7df2c79 JB |
3156 | return 0; |
3157 | } | |
5d4f98a2 YZ |
3158 | |
3159 | if (!mutex_trylock(&head->mutex)) { | |
d278850e | 3160 | refcount_inc(&head->refs); |
5d4f98a2 YZ |
3161 | spin_unlock(&delayed_refs->lock); |
3162 | ||
b3b4aa74 | 3163 | btrfs_release_path(path); |
5d4f98a2 | 3164 | |
8cc33e5c DS |
3165 | /* |
3166 | * Mutex was contended, block until it's released and let | |
3167 | * caller try again | |
3168 | */ | |
5d4f98a2 YZ |
3169 | mutex_lock(&head->mutex); |
3170 | mutex_unlock(&head->mutex); | |
d278850e | 3171 | btrfs_put_delayed_ref_head(head); |
998ac6d2 | 3172 | btrfs_put_transaction(cur_trans); |
5d4f98a2 YZ |
3173 | return -EAGAIN; |
3174 | } | |
d7df2c79 | 3175 | spin_unlock(&delayed_refs->lock); |
5d4f98a2 | 3176 | |
d7df2c79 | 3177 | spin_lock(&head->lock); |
0e0adbcf JB |
3178 | /* |
3179 | * XXX: We should replace this with a proper search function in the | |
3180 | * future. | |
3181 | */ | |
3182 | for (node = rb_first(&head->ref_tree); node; node = rb_next(node)) { | |
3183 | ref = rb_entry(node, struct btrfs_delayed_ref_node, ref_node); | |
d7df2c79 JB |
3184 | /* If it's a shared ref we know a cross reference exists */ |
3185 | if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) { | |
3186 | ret = 1; | |
3187 | break; | |
3188 | } | |
5d4f98a2 | 3189 | |
d7df2c79 | 3190 | data_ref = btrfs_delayed_node_to_data_ref(ref); |
5d4f98a2 | 3191 | |
d7df2c79 JB |
3192 | /* |
3193 | * If our ref doesn't match the one we're currently looking at | |
3194 | * then we have a cross reference. | |
3195 | */ | |
3196 | if (data_ref->root != root->root_key.objectid || | |
3197 | data_ref->objectid != objectid || | |
3198 | data_ref->offset != offset) { | |
3199 | ret = 1; | |
3200 | break; | |
3201 | } | |
5d4f98a2 | 3202 | } |
d7df2c79 | 3203 | spin_unlock(&head->lock); |
5d4f98a2 | 3204 | mutex_unlock(&head->mutex); |
998ac6d2 | 3205 | btrfs_put_transaction(cur_trans); |
5d4f98a2 YZ |
3206 | return ret; |
3207 | } | |
3208 | ||
e4c3b2dc | 3209 | static noinline int check_committed_ref(struct btrfs_root *root, |
5d4f98a2 YZ |
3210 | struct btrfs_path *path, |
3211 | u64 objectid, u64 offset, u64 bytenr) | |
be20aa9d | 3212 | { |
0b246afa JM |
3213 | struct btrfs_fs_info *fs_info = root->fs_info; |
3214 | struct btrfs_root *extent_root = fs_info->extent_root; | |
f321e491 | 3215 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
3216 | struct btrfs_extent_data_ref *ref; |
3217 | struct btrfs_extent_inline_ref *iref; | |
3218 | struct btrfs_extent_item *ei; | |
f321e491 | 3219 | struct btrfs_key key; |
5d4f98a2 | 3220 | u32 item_size; |
3de28d57 | 3221 | int type; |
be20aa9d | 3222 | int ret; |
925baedd | 3223 | |
be20aa9d | 3224 | key.objectid = bytenr; |
31840ae1 | 3225 | key.offset = (u64)-1; |
f321e491 | 3226 | key.type = BTRFS_EXTENT_ITEM_KEY; |
be20aa9d | 3227 | |
be20aa9d CM |
3228 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
3229 | if (ret < 0) | |
3230 | goto out; | |
79787eaa | 3231 | BUG_ON(ret == 0); /* Corruption */ |
80ff3856 YZ |
3232 | |
3233 | ret = -ENOENT; | |
3234 | if (path->slots[0] == 0) | |
31840ae1 | 3235 | goto out; |
be20aa9d | 3236 | |
31840ae1 | 3237 | path->slots[0]--; |
f321e491 | 3238 | leaf = path->nodes[0]; |
5d4f98a2 | 3239 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
be20aa9d | 3240 | |
5d4f98a2 | 3241 | if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY) |
be20aa9d | 3242 | goto out; |
f321e491 | 3243 | |
5d4f98a2 YZ |
3244 | ret = 1; |
3245 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
3246 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
3247 | if (item_size < sizeof(*ei)) { | |
3248 | WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0)); | |
3249 | goto out; | |
3250 | } | |
3251 | #endif | |
3252 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
bd09835d | 3253 | |
5d4f98a2 YZ |
3254 | if (item_size != sizeof(*ei) + |
3255 | btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY)) | |
3256 | goto out; | |
be20aa9d | 3257 | |
5d4f98a2 YZ |
3258 | if (btrfs_extent_generation(leaf, ei) <= |
3259 | btrfs_root_last_snapshot(&root->root_item)) | |
3260 | goto out; | |
3261 | ||
3262 | iref = (struct btrfs_extent_inline_ref *)(ei + 1); | |
3de28d57 LB |
3263 | |
3264 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA); | |
3265 | if (type != BTRFS_EXTENT_DATA_REF_KEY) | |
5d4f98a2 YZ |
3266 | goto out; |
3267 | ||
3268 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
3269 | if (btrfs_extent_refs(leaf, ei) != | |
3270 | btrfs_extent_data_ref_count(leaf, ref) || | |
3271 | btrfs_extent_data_ref_root(leaf, ref) != | |
3272 | root->root_key.objectid || | |
3273 | btrfs_extent_data_ref_objectid(leaf, ref) != objectid || | |
3274 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
3275 | goto out; | |
3276 | ||
3277 | ret = 0; | |
3278 | out: | |
3279 | return ret; | |
3280 | } | |
3281 | ||
e4c3b2dc LB |
3282 | int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset, |
3283 | u64 bytenr) | |
5d4f98a2 YZ |
3284 | { |
3285 | struct btrfs_path *path; | |
3286 | int ret; | |
3287 | int ret2; | |
3288 | ||
3289 | path = btrfs_alloc_path(); | |
3290 | if (!path) | |
9132c4ff | 3291 | return -ENOMEM; |
5d4f98a2 YZ |
3292 | |
3293 | do { | |
e4c3b2dc | 3294 | ret = check_committed_ref(root, path, objectid, |
5d4f98a2 YZ |
3295 | offset, bytenr); |
3296 | if (ret && ret != -ENOENT) | |
f321e491 | 3297 | goto out; |
80ff3856 | 3298 | |
e4c3b2dc | 3299 | ret2 = check_delayed_ref(root, path, objectid, |
5d4f98a2 YZ |
3300 | offset, bytenr); |
3301 | } while (ret2 == -EAGAIN); | |
3302 | ||
3303 | if (ret2 && ret2 != -ENOENT) { | |
3304 | ret = ret2; | |
3305 | goto out; | |
f321e491 | 3306 | } |
5d4f98a2 YZ |
3307 | |
3308 | if (ret != -ENOENT || ret2 != -ENOENT) | |
3309 | ret = 0; | |
be20aa9d | 3310 | out: |
80ff3856 | 3311 | btrfs_free_path(path); |
f0486c68 YZ |
3312 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) |
3313 | WARN_ON(ret > 0); | |
f321e491 | 3314 | return ret; |
be20aa9d | 3315 | } |
c5739bba | 3316 | |
5d4f98a2 | 3317 | static int __btrfs_mod_ref(struct btrfs_trans_handle *trans, |
b7a9f29f | 3318 | struct btrfs_root *root, |
5d4f98a2 | 3319 | struct extent_buffer *buf, |
e339a6b0 | 3320 | int full_backref, int inc) |
31840ae1 | 3321 | { |
0b246afa | 3322 | struct btrfs_fs_info *fs_info = root->fs_info; |
31840ae1 | 3323 | u64 bytenr; |
5d4f98a2 YZ |
3324 | u64 num_bytes; |
3325 | u64 parent; | |
31840ae1 | 3326 | u64 ref_root; |
31840ae1 | 3327 | u32 nritems; |
31840ae1 ZY |
3328 | struct btrfs_key key; |
3329 | struct btrfs_file_extent_item *fi; | |
3330 | int i; | |
3331 | int level; | |
3332 | int ret = 0; | |
2ff7e61e | 3333 | int (*process_func)(struct btrfs_trans_handle *, |
84f7d8e6 | 3334 | struct btrfs_root *, |
b06c4bf5 | 3335 | u64, u64, u64, u64, u64, u64); |
31840ae1 | 3336 | |
fccb84c9 | 3337 | |
0b246afa | 3338 | if (btrfs_is_testing(fs_info)) |
faa2dbf0 | 3339 | return 0; |
fccb84c9 | 3340 | |
31840ae1 | 3341 | ref_root = btrfs_header_owner(buf); |
31840ae1 ZY |
3342 | nritems = btrfs_header_nritems(buf); |
3343 | level = btrfs_header_level(buf); | |
3344 | ||
27cdeb70 | 3345 | if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0) |
5d4f98a2 | 3346 | return 0; |
31840ae1 | 3347 | |
5d4f98a2 YZ |
3348 | if (inc) |
3349 | process_func = btrfs_inc_extent_ref; | |
3350 | else | |
3351 | process_func = btrfs_free_extent; | |
31840ae1 | 3352 | |
5d4f98a2 YZ |
3353 | if (full_backref) |
3354 | parent = buf->start; | |
3355 | else | |
3356 | parent = 0; | |
3357 | ||
3358 | for (i = 0; i < nritems; i++) { | |
31840ae1 | 3359 | if (level == 0) { |
5d4f98a2 | 3360 | btrfs_item_key_to_cpu(buf, &key, i); |
962a298f | 3361 | if (key.type != BTRFS_EXTENT_DATA_KEY) |
31840ae1 | 3362 | continue; |
5d4f98a2 | 3363 | fi = btrfs_item_ptr(buf, i, |
31840ae1 ZY |
3364 | struct btrfs_file_extent_item); |
3365 | if (btrfs_file_extent_type(buf, fi) == | |
3366 | BTRFS_FILE_EXTENT_INLINE) | |
3367 | continue; | |
3368 | bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
3369 | if (bytenr == 0) | |
3370 | continue; | |
5d4f98a2 YZ |
3371 | |
3372 | num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi); | |
3373 | key.offset -= btrfs_file_extent_offset(buf, fi); | |
84f7d8e6 | 3374 | ret = process_func(trans, root, bytenr, num_bytes, |
5d4f98a2 | 3375 | parent, ref_root, key.objectid, |
b06c4bf5 | 3376 | key.offset); |
31840ae1 ZY |
3377 | if (ret) |
3378 | goto fail; | |
3379 | } else { | |
5d4f98a2 | 3380 | bytenr = btrfs_node_blockptr(buf, i); |
0b246afa | 3381 | num_bytes = fs_info->nodesize; |
84f7d8e6 | 3382 | ret = process_func(trans, root, bytenr, num_bytes, |
b06c4bf5 | 3383 | parent, ref_root, level - 1, 0); |
31840ae1 ZY |
3384 | if (ret) |
3385 | goto fail; | |
3386 | } | |
3387 | } | |
3388 | return 0; | |
3389 | fail: | |
5d4f98a2 YZ |
3390 | return ret; |
3391 | } | |
3392 | ||
3393 | int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
e339a6b0 | 3394 | struct extent_buffer *buf, int full_backref) |
5d4f98a2 | 3395 | { |
e339a6b0 | 3396 | return __btrfs_mod_ref(trans, root, buf, full_backref, 1); |
5d4f98a2 YZ |
3397 | } |
3398 | ||
3399 | int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
e339a6b0 | 3400 | struct extent_buffer *buf, int full_backref) |
5d4f98a2 | 3401 | { |
e339a6b0 | 3402 | return __btrfs_mod_ref(trans, root, buf, full_backref, 0); |
31840ae1 ZY |
3403 | } |
3404 | ||
9078a3e1 | 3405 | static int write_one_cache_group(struct btrfs_trans_handle *trans, |
2ff7e61e | 3406 | struct btrfs_fs_info *fs_info, |
9078a3e1 CM |
3407 | struct btrfs_path *path, |
3408 | struct btrfs_block_group_cache *cache) | |
3409 | { | |
3410 | int ret; | |
0b246afa | 3411 | struct btrfs_root *extent_root = fs_info->extent_root; |
5f39d397 CM |
3412 | unsigned long bi; |
3413 | struct extent_buffer *leaf; | |
9078a3e1 | 3414 | |
9078a3e1 | 3415 | ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1); |
df95e7f0 JB |
3416 | if (ret) { |
3417 | if (ret > 0) | |
3418 | ret = -ENOENT; | |
54aa1f4d | 3419 | goto fail; |
df95e7f0 | 3420 | } |
5f39d397 CM |
3421 | |
3422 | leaf = path->nodes[0]; | |
3423 | bi = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
3424 | write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item)); | |
3425 | btrfs_mark_buffer_dirty(leaf); | |
54aa1f4d | 3426 | fail: |
24b89d08 | 3427 | btrfs_release_path(path); |
df95e7f0 | 3428 | return ret; |
9078a3e1 CM |
3429 | |
3430 | } | |
3431 | ||
4a8c9a62 | 3432 | static struct btrfs_block_group_cache * |
2ff7e61e | 3433 | next_block_group(struct btrfs_fs_info *fs_info, |
4a8c9a62 YZ |
3434 | struct btrfs_block_group_cache *cache) |
3435 | { | |
3436 | struct rb_node *node; | |
292cbd51 | 3437 | |
0b246afa | 3438 | spin_lock(&fs_info->block_group_cache_lock); |
292cbd51 FM |
3439 | |
3440 | /* If our block group was removed, we need a full search. */ | |
3441 | if (RB_EMPTY_NODE(&cache->cache_node)) { | |
3442 | const u64 next_bytenr = cache->key.objectid + cache->key.offset; | |
3443 | ||
0b246afa | 3444 | spin_unlock(&fs_info->block_group_cache_lock); |
292cbd51 | 3445 | btrfs_put_block_group(cache); |
0b246afa | 3446 | cache = btrfs_lookup_first_block_group(fs_info, next_bytenr); return cache; |
292cbd51 | 3447 | } |
4a8c9a62 YZ |
3448 | node = rb_next(&cache->cache_node); |
3449 | btrfs_put_block_group(cache); | |
3450 | if (node) { | |
3451 | cache = rb_entry(node, struct btrfs_block_group_cache, | |
3452 | cache_node); | |
11dfe35a | 3453 | btrfs_get_block_group(cache); |
4a8c9a62 YZ |
3454 | } else |
3455 | cache = NULL; | |
0b246afa | 3456 | spin_unlock(&fs_info->block_group_cache_lock); |
4a8c9a62 YZ |
3457 | return cache; |
3458 | } | |
3459 | ||
0af3d00b JB |
3460 | static int cache_save_setup(struct btrfs_block_group_cache *block_group, |
3461 | struct btrfs_trans_handle *trans, | |
3462 | struct btrfs_path *path) | |
3463 | { | |
0b246afa JM |
3464 | struct btrfs_fs_info *fs_info = block_group->fs_info; |
3465 | struct btrfs_root *root = fs_info->tree_root; | |
0af3d00b | 3466 | struct inode *inode = NULL; |
364ecf36 | 3467 | struct extent_changeset *data_reserved = NULL; |
0af3d00b | 3468 | u64 alloc_hint = 0; |
2b20982e | 3469 | int dcs = BTRFS_DC_ERROR; |
f8c269d7 | 3470 | u64 num_pages = 0; |
0af3d00b JB |
3471 | int retries = 0; |
3472 | int ret = 0; | |
3473 | ||
3474 | /* | |
3475 | * If this block group is smaller than 100 megs don't bother caching the | |
3476 | * block group. | |
3477 | */ | |
ee22184b | 3478 | if (block_group->key.offset < (100 * SZ_1M)) { |
0af3d00b JB |
3479 | spin_lock(&block_group->lock); |
3480 | block_group->disk_cache_state = BTRFS_DC_WRITTEN; | |
3481 | spin_unlock(&block_group->lock); | |
3482 | return 0; | |
3483 | } | |
3484 | ||
0c0ef4bc JB |
3485 | if (trans->aborted) |
3486 | return 0; | |
0af3d00b | 3487 | again: |
77ab86bf | 3488 | inode = lookup_free_space_inode(fs_info, block_group, path); |
0af3d00b JB |
3489 | if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) { |
3490 | ret = PTR_ERR(inode); | |
b3b4aa74 | 3491 | btrfs_release_path(path); |
0af3d00b JB |
3492 | goto out; |
3493 | } | |
3494 | ||
3495 | if (IS_ERR(inode)) { | |
3496 | BUG_ON(retries); | |
3497 | retries++; | |
3498 | ||
3499 | if (block_group->ro) | |
3500 | goto out_free; | |
3501 | ||
77ab86bf JM |
3502 | ret = create_free_space_inode(fs_info, trans, block_group, |
3503 | path); | |
0af3d00b JB |
3504 | if (ret) |
3505 | goto out_free; | |
3506 | goto again; | |
3507 | } | |
3508 | ||
3509 | /* | |
3510 | * We want to set the generation to 0, that way if anything goes wrong | |
3511 | * from here on out we know not to trust this cache when we load up next | |
3512 | * time. | |
3513 | */ | |
3514 | BTRFS_I(inode)->generation = 0; | |
3515 | ret = btrfs_update_inode(trans, root, inode); | |
0c0ef4bc JB |
3516 | if (ret) { |
3517 | /* | |
3518 | * So theoretically we could recover from this, simply set the | |
3519 | * super cache generation to 0 so we know to invalidate the | |
3520 | * cache, but then we'd have to keep track of the block groups | |
3521 | * that fail this way so we know we _have_ to reset this cache | |
3522 | * before the next commit or risk reading stale cache. So to | |
3523 | * limit our exposure to horrible edge cases lets just abort the | |
3524 | * transaction, this only happens in really bad situations | |
3525 | * anyway. | |
3526 | */ | |
66642832 | 3527 | btrfs_abort_transaction(trans, ret); |
0c0ef4bc JB |
3528 | goto out_put; |
3529 | } | |
0af3d00b JB |
3530 | WARN_ON(ret); |
3531 | ||
8e138e0d JB |
3532 | /* We've already setup this transaction, go ahead and exit */ |
3533 | if (block_group->cache_generation == trans->transid && | |
3534 | i_size_read(inode)) { | |
3535 | dcs = BTRFS_DC_SETUP; | |
3536 | goto out_put; | |
3537 | } | |
3538 | ||
0af3d00b | 3539 | if (i_size_read(inode) > 0) { |
2ff7e61e | 3540 | ret = btrfs_check_trunc_cache_free_space(fs_info, |
0b246afa | 3541 | &fs_info->global_block_rsv); |
7b61cd92 MX |
3542 | if (ret) |
3543 | goto out_put; | |
3544 | ||
77ab86bf | 3545 | ret = btrfs_truncate_free_space_cache(trans, NULL, inode); |
0af3d00b JB |
3546 | if (ret) |
3547 | goto out_put; | |
3548 | } | |
3549 | ||
3550 | spin_lock(&block_group->lock); | |
cf7c1ef6 | 3551 | if (block_group->cached != BTRFS_CACHE_FINISHED || |
0b246afa | 3552 | !btrfs_test_opt(fs_info, SPACE_CACHE)) { |
cf7c1ef6 LB |
3553 | /* |
3554 | * don't bother trying to write stuff out _if_ | |
3555 | * a) we're not cached, | |
1a79c1f2 LB |
3556 | * b) we're with nospace_cache mount option, |
3557 | * c) we're with v2 space_cache (FREE_SPACE_TREE). | |
cf7c1ef6 | 3558 | */ |
2b20982e | 3559 | dcs = BTRFS_DC_WRITTEN; |
0af3d00b JB |
3560 | spin_unlock(&block_group->lock); |
3561 | goto out_put; | |
3562 | } | |
3563 | spin_unlock(&block_group->lock); | |
3564 | ||
2968b1f4 JB |
3565 | /* |
3566 | * We hit an ENOSPC when setting up the cache in this transaction, just | |
3567 | * skip doing the setup, we've already cleared the cache so we're safe. | |
3568 | */ | |
3569 | if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) { | |
3570 | ret = -ENOSPC; | |
3571 | goto out_put; | |
3572 | } | |
3573 | ||
6fc823b1 JB |
3574 | /* |
3575 | * Try to preallocate enough space based on how big the block group is. | |
3576 | * Keep in mind this has to include any pinned space which could end up | |
3577 | * taking up quite a bit since it's not folded into the other space | |
3578 | * cache. | |
3579 | */ | |
ee22184b | 3580 | num_pages = div_u64(block_group->key.offset, SZ_256M); |
0af3d00b JB |
3581 | if (!num_pages) |
3582 | num_pages = 1; | |
3583 | ||
0af3d00b | 3584 | num_pages *= 16; |
09cbfeaf | 3585 | num_pages *= PAGE_SIZE; |
0af3d00b | 3586 | |
364ecf36 | 3587 | ret = btrfs_check_data_free_space(inode, &data_reserved, 0, num_pages); |
0af3d00b JB |
3588 | if (ret) |
3589 | goto out_put; | |
3590 | ||
3591 | ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages, | |
3592 | num_pages, num_pages, | |
3593 | &alloc_hint); | |
2968b1f4 JB |
3594 | /* |
3595 | * Our cache requires contiguous chunks so that we don't modify a bunch | |
3596 | * of metadata or split extents when writing the cache out, which means | |
3597 | * we can enospc if we are heavily fragmented in addition to just normal | |
3598 | * out of space conditions. So if we hit this just skip setting up any | |
3599 | * other block groups for this transaction, maybe we'll unpin enough | |
3600 | * space the next time around. | |
3601 | */ | |
2b20982e JB |
3602 | if (!ret) |
3603 | dcs = BTRFS_DC_SETUP; | |
2968b1f4 JB |
3604 | else if (ret == -ENOSPC) |
3605 | set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags); | |
c09544e0 | 3606 | |
0af3d00b JB |
3607 | out_put: |
3608 | iput(inode); | |
3609 | out_free: | |
b3b4aa74 | 3610 | btrfs_release_path(path); |
0af3d00b JB |
3611 | out: |
3612 | spin_lock(&block_group->lock); | |
e65cbb94 | 3613 | if (!ret && dcs == BTRFS_DC_SETUP) |
5b0e95bf | 3614 | block_group->cache_generation = trans->transid; |
2b20982e | 3615 | block_group->disk_cache_state = dcs; |
0af3d00b JB |
3616 | spin_unlock(&block_group->lock); |
3617 | ||
364ecf36 | 3618 | extent_changeset_free(data_reserved); |
0af3d00b JB |
3619 | return ret; |
3620 | } | |
3621 | ||
dcdf7f6d | 3622 | int btrfs_setup_space_cache(struct btrfs_trans_handle *trans, |
2ff7e61e | 3623 | struct btrfs_fs_info *fs_info) |
dcdf7f6d JB |
3624 | { |
3625 | struct btrfs_block_group_cache *cache, *tmp; | |
3626 | struct btrfs_transaction *cur_trans = trans->transaction; | |
3627 | struct btrfs_path *path; | |
3628 | ||
3629 | if (list_empty(&cur_trans->dirty_bgs) || | |
0b246afa | 3630 | !btrfs_test_opt(fs_info, SPACE_CACHE)) |
dcdf7f6d JB |
3631 | return 0; |
3632 | ||
3633 | path = btrfs_alloc_path(); | |
3634 | if (!path) | |
3635 | return -ENOMEM; | |
3636 | ||
3637 | /* Could add new block groups, use _safe just in case */ | |
3638 | list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs, | |
3639 | dirty_list) { | |
3640 | if (cache->disk_cache_state == BTRFS_DC_CLEAR) | |
3641 | cache_save_setup(cache, trans, path); | |
3642 | } | |
3643 | ||
3644 | btrfs_free_path(path); | |
3645 | return 0; | |
3646 | } | |
3647 | ||
1bbc621e CM |
3648 | /* |
3649 | * transaction commit does final block group cache writeback during a | |
3650 | * critical section where nothing is allowed to change the FS. This is | |
3651 | * required in order for the cache to actually match the block group, | |
3652 | * but can introduce a lot of latency into the commit. | |
3653 | * | |
3654 | * So, btrfs_start_dirty_block_groups is here to kick off block group | |
3655 | * cache IO. There's a chance we'll have to redo some of it if the | |
3656 | * block group changes again during the commit, but it greatly reduces | |
3657 | * the commit latency by getting rid of the easy block groups while | |
3658 | * we're still allowing others to join the commit. | |
3659 | */ | |
21217054 | 3660 | int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans) |
9078a3e1 | 3661 | { |
21217054 | 3662 | struct btrfs_fs_info *fs_info = trans->fs_info; |
4a8c9a62 | 3663 | struct btrfs_block_group_cache *cache; |
ce93ec54 JB |
3664 | struct btrfs_transaction *cur_trans = trans->transaction; |
3665 | int ret = 0; | |
c9dc4c65 | 3666 | int should_put; |
1bbc621e CM |
3667 | struct btrfs_path *path = NULL; |
3668 | LIST_HEAD(dirty); | |
3669 | struct list_head *io = &cur_trans->io_bgs; | |
c9dc4c65 | 3670 | int num_started = 0; |
1bbc621e CM |
3671 | int loops = 0; |
3672 | ||
3673 | spin_lock(&cur_trans->dirty_bgs_lock); | |
b58d1a9e FM |
3674 | if (list_empty(&cur_trans->dirty_bgs)) { |
3675 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3676 | return 0; | |
1bbc621e | 3677 | } |
b58d1a9e | 3678 | list_splice_init(&cur_trans->dirty_bgs, &dirty); |
1bbc621e | 3679 | spin_unlock(&cur_trans->dirty_bgs_lock); |
ce93ec54 | 3680 | |
1bbc621e | 3681 | again: |
1bbc621e CM |
3682 | /* |
3683 | * make sure all the block groups on our dirty list actually | |
3684 | * exist | |
3685 | */ | |
6c686b35 | 3686 | btrfs_create_pending_block_groups(trans); |
1bbc621e CM |
3687 | |
3688 | if (!path) { | |
3689 | path = btrfs_alloc_path(); | |
3690 | if (!path) | |
3691 | return -ENOMEM; | |
3692 | } | |
3693 | ||
b58d1a9e FM |
3694 | /* |
3695 | * cache_write_mutex is here only to save us from balance or automatic | |
3696 | * removal of empty block groups deleting this block group while we are | |
3697 | * writing out the cache | |
3698 | */ | |
3699 | mutex_lock(&trans->transaction->cache_write_mutex); | |
1bbc621e CM |
3700 | while (!list_empty(&dirty)) { |
3701 | cache = list_first_entry(&dirty, | |
3702 | struct btrfs_block_group_cache, | |
3703 | dirty_list); | |
1bbc621e CM |
3704 | /* |
3705 | * this can happen if something re-dirties a block | |
3706 | * group that is already under IO. Just wait for it to | |
3707 | * finish and then do it all again | |
3708 | */ | |
3709 | if (!list_empty(&cache->io_list)) { | |
3710 | list_del_init(&cache->io_list); | |
afdb5718 | 3711 | btrfs_wait_cache_io(trans, cache, path); |
1bbc621e CM |
3712 | btrfs_put_block_group(cache); |
3713 | } | |
3714 | ||
3715 | ||
3716 | /* | |
3717 | * btrfs_wait_cache_io uses the cache->dirty_list to decide | |
3718 | * if it should update the cache_state. Don't delete | |
3719 | * until after we wait. | |
3720 | * | |
3721 | * Since we're not running in the commit critical section | |
3722 | * we need the dirty_bgs_lock to protect from update_block_group | |
3723 | */ | |
3724 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3725 | list_del_init(&cache->dirty_list); | |
3726 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3727 | ||
3728 | should_put = 1; | |
3729 | ||
3730 | cache_save_setup(cache, trans, path); | |
3731 | ||
3732 | if (cache->disk_cache_state == BTRFS_DC_SETUP) { | |
3733 | cache->io_ctl.inode = NULL; | |
0b246afa | 3734 | ret = btrfs_write_out_cache(fs_info, trans, |
5b4aacef | 3735 | cache, path); |
1bbc621e CM |
3736 | if (ret == 0 && cache->io_ctl.inode) { |
3737 | num_started++; | |
3738 | should_put = 0; | |
3739 | ||
3740 | /* | |
45ae2c18 NB |
3741 | * The cache_write_mutex is protecting the |
3742 | * io_list, also refer to the definition of | |
3743 | * btrfs_transaction::io_bgs for more details | |
1bbc621e CM |
3744 | */ |
3745 | list_add_tail(&cache->io_list, io); | |
3746 | } else { | |
3747 | /* | |
3748 | * if we failed to write the cache, the | |
3749 | * generation will be bad and life goes on | |
3750 | */ | |
3751 | ret = 0; | |
3752 | } | |
3753 | } | |
ff1f8250 | 3754 | if (!ret) { |
2ff7e61e JM |
3755 | ret = write_one_cache_group(trans, fs_info, |
3756 | path, cache); | |
ff1f8250 FM |
3757 | /* |
3758 | * Our block group might still be attached to the list | |
3759 | * of new block groups in the transaction handle of some | |
3760 | * other task (struct btrfs_trans_handle->new_bgs). This | |
3761 | * means its block group item isn't yet in the extent | |
3762 | * tree. If this happens ignore the error, as we will | |
3763 | * try again later in the critical section of the | |
3764 | * transaction commit. | |
3765 | */ | |
3766 | if (ret == -ENOENT) { | |
3767 | ret = 0; | |
3768 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3769 | if (list_empty(&cache->dirty_list)) { | |
3770 | list_add_tail(&cache->dirty_list, | |
3771 | &cur_trans->dirty_bgs); | |
3772 | btrfs_get_block_group(cache); | |
3773 | } | |
3774 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3775 | } else if (ret) { | |
66642832 | 3776 | btrfs_abort_transaction(trans, ret); |
ff1f8250 FM |
3777 | } |
3778 | } | |
1bbc621e CM |
3779 | |
3780 | /* if its not on the io list, we need to put the block group */ | |
3781 | if (should_put) | |
3782 | btrfs_put_block_group(cache); | |
3783 | ||
3784 | if (ret) | |
3785 | break; | |
b58d1a9e FM |
3786 | |
3787 | /* | |
3788 | * Avoid blocking other tasks for too long. It might even save | |
3789 | * us from writing caches for block groups that are going to be | |
3790 | * removed. | |
3791 | */ | |
3792 | mutex_unlock(&trans->transaction->cache_write_mutex); | |
3793 | mutex_lock(&trans->transaction->cache_write_mutex); | |
1bbc621e | 3794 | } |
b58d1a9e | 3795 | mutex_unlock(&trans->transaction->cache_write_mutex); |
1bbc621e CM |
3796 | |
3797 | /* | |
3798 | * go through delayed refs for all the stuff we've just kicked off | |
3799 | * and then loop back (just once) | |
3800 | */ | |
c79a70b1 | 3801 | ret = btrfs_run_delayed_refs(trans, 0); |
1bbc621e CM |
3802 | if (!ret && loops == 0) { |
3803 | loops++; | |
3804 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3805 | list_splice_init(&cur_trans->dirty_bgs, &dirty); | |
b58d1a9e FM |
3806 | /* |
3807 | * dirty_bgs_lock protects us from concurrent block group | |
3808 | * deletes too (not just cache_write_mutex). | |
3809 | */ | |
3810 | if (!list_empty(&dirty)) { | |
3811 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3812 | goto again; | |
3813 | } | |
1bbc621e | 3814 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c79a1751 | 3815 | } else if (ret < 0) { |
2ff7e61e | 3816 | btrfs_cleanup_dirty_bgs(cur_trans, fs_info); |
1bbc621e CM |
3817 | } |
3818 | ||
3819 | btrfs_free_path(path); | |
3820 | return ret; | |
3821 | } | |
3822 | ||
3823 | int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans, | |
2ff7e61e | 3824 | struct btrfs_fs_info *fs_info) |
1bbc621e CM |
3825 | { |
3826 | struct btrfs_block_group_cache *cache; | |
3827 | struct btrfs_transaction *cur_trans = trans->transaction; | |
3828 | int ret = 0; | |
3829 | int should_put; | |
3830 | struct btrfs_path *path; | |
3831 | struct list_head *io = &cur_trans->io_bgs; | |
3832 | int num_started = 0; | |
9078a3e1 CM |
3833 | |
3834 | path = btrfs_alloc_path(); | |
3835 | if (!path) | |
3836 | return -ENOMEM; | |
3837 | ||
ce93ec54 | 3838 | /* |
e44081ef FM |
3839 | * Even though we are in the critical section of the transaction commit, |
3840 | * we can still have concurrent tasks adding elements to this | |
3841 | * transaction's list of dirty block groups. These tasks correspond to | |
3842 | * endio free space workers started when writeback finishes for a | |
3843 | * space cache, which run inode.c:btrfs_finish_ordered_io(), and can | |
3844 | * allocate new block groups as a result of COWing nodes of the root | |
3845 | * tree when updating the free space inode. The writeback for the space | |
3846 | * caches is triggered by an earlier call to | |
3847 | * btrfs_start_dirty_block_groups() and iterations of the following | |
3848 | * loop. | |
3849 | * Also we want to do the cache_save_setup first and then run the | |
ce93ec54 JB |
3850 | * delayed refs to make sure we have the best chance at doing this all |
3851 | * in one shot. | |
3852 | */ | |
e44081ef | 3853 | spin_lock(&cur_trans->dirty_bgs_lock); |
ce93ec54 JB |
3854 | while (!list_empty(&cur_trans->dirty_bgs)) { |
3855 | cache = list_first_entry(&cur_trans->dirty_bgs, | |
3856 | struct btrfs_block_group_cache, | |
3857 | dirty_list); | |
c9dc4c65 CM |
3858 | |
3859 | /* | |
3860 | * this can happen if cache_save_setup re-dirties a block | |
3861 | * group that is already under IO. Just wait for it to | |
3862 | * finish and then do it all again | |
3863 | */ | |
3864 | if (!list_empty(&cache->io_list)) { | |
e44081ef | 3865 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 | 3866 | list_del_init(&cache->io_list); |
afdb5718 | 3867 | btrfs_wait_cache_io(trans, cache, path); |
c9dc4c65 | 3868 | btrfs_put_block_group(cache); |
e44081ef | 3869 | spin_lock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 CM |
3870 | } |
3871 | ||
1bbc621e CM |
3872 | /* |
3873 | * don't remove from the dirty list until after we've waited | |
3874 | * on any pending IO | |
3875 | */ | |
ce93ec54 | 3876 | list_del_init(&cache->dirty_list); |
e44081ef | 3877 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 CM |
3878 | should_put = 1; |
3879 | ||
1bbc621e | 3880 | cache_save_setup(cache, trans, path); |
c9dc4c65 | 3881 | |
ce93ec54 | 3882 | if (!ret) |
c79a70b1 | 3883 | ret = btrfs_run_delayed_refs(trans, |
2ff7e61e | 3884 | (unsigned long) -1); |
c9dc4c65 CM |
3885 | |
3886 | if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) { | |
3887 | cache->io_ctl.inode = NULL; | |
0b246afa | 3888 | ret = btrfs_write_out_cache(fs_info, trans, |
5b4aacef | 3889 | cache, path); |
c9dc4c65 CM |
3890 | if (ret == 0 && cache->io_ctl.inode) { |
3891 | num_started++; | |
3892 | should_put = 0; | |
1bbc621e | 3893 | list_add_tail(&cache->io_list, io); |
c9dc4c65 CM |
3894 | } else { |
3895 | /* | |
3896 | * if we failed to write the cache, the | |
3897 | * generation will be bad and life goes on | |
3898 | */ | |
3899 | ret = 0; | |
3900 | } | |
3901 | } | |
ff1f8250 | 3902 | if (!ret) { |
2ff7e61e JM |
3903 | ret = write_one_cache_group(trans, fs_info, |
3904 | path, cache); | |
2bc0bb5f FM |
3905 | /* |
3906 | * One of the free space endio workers might have | |
3907 | * created a new block group while updating a free space | |
3908 | * cache's inode (at inode.c:btrfs_finish_ordered_io()) | |
3909 | * and hasn't released its transaction handle yet, in | |
3910 | * which case the new block group is still attached to | |
3911 | * its transaction handle and its creation has not | |
3912 | * finished yet (no block group item in the extent tree | |
3913 | * yet, etc). If this is the case, wait for all free | |
3914 | * space endio workers to finish and retry. This is a | |
3915 | * a very rare case so no need for a more efficient and | |
3916 | * complex approach. | |
3917 | */ | |
3918 | if (ret == -ENOENT) { | |
3919 | wait_event(cur_trans->writer_wait, | |
3920 | atomic_read(&cur_trans->num_writers) == 1); | |
2ff7e61e JM |
3921 | ret = write_one_cache_group(trans, fs_info, |
3922 | path, cache); | |
2bc0bb5f | 3923 | } |
ff1f8250 | 3924 | if (ret) |
66642832 | 3925 | btrfs_abort_transaction(trans, ret); |
ff1f8250 | 3926 | } |
c9dc4c65 CM |
3927 | |
3928 | /* if its not on the io list, we need to put the block group */ | |
3929 | if (should_put) | |
3930 | btrfs_put_block_group(cache); | |
e44081ef | 3931 | spin_lock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 | 3932 | } |
e44081ef | 3933 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 | 3934 | |
45ae2c18 NB |
3935 | /* |
3936 | * Refer to the definition of io_bgs member for details why it's safe | |
3937 | * to use it without any locking | |
3938 | */ | |
1bbc621e CM |
3939 | while (!list_empty(io)) { |
3940 | cache = list_first_entry(io, struct btrfs_block_group_cache, | |
c9dc4c65 CM |
3941 | io_list); |
3942 | list_del_init(&cache->io_list); | |
afdb5718 | 3943 | btrfs_wait_cache_io(trans, cache, path); |
0cb59c99 JB |
3944 | btrfs_put_block_group(cache); |
3945 | } | |
3946 | ||
9078a3e1 | 3947 | btrfs_free_path(path); |
ce93ec54 | 3948 | return ret; |
9078a3e1 CM |
3949 | } |
3950 | ||
2ff7e61e | 3951 | int btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr) |
d2fb3437 YZ |
3952 | { |
3953 | struct btrfs_block_group_cache *block_group; | |
3954 | int readonly = 0; | |
3955 | ||
0b246afa | 3956 | block_group = btrfs_lookup_block_group(fs_info, bytenr); |
d2fb3437 YZ |
3957 | if (!block_group || block_group->ro) |
3958 | readonly = 1; | |
3959 | if (block_group) | |
fa9c0d79 | 3960 | btrfs_put_block_group(block_group); |
d2fb3437 YZ |
3961 | return readonly; |
3962 | } | |
3963 | ||
f78c436c FM |
3964 | bool btrfs_inc_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr) |
3965 | { | |
3966 | struct btrfs_block_group_cache *bg; | |
3967 | bool ret = true; | |
3968 | ||
3969 | bg = btrfs_lookup_block_group(fs_info, bytenr); | |
3970 | if (!bg) | |
3971 | return false; | |
3972 | ||
3973 | spin_lock(&bg->lock); | |
3974 | if (bg->ro) | |
3975 | ret = false; | |
3976 | else | |
3977 | atomic_inc(&bg->nocow_writers); | |
3978 | spin_unlock(&bg->lock); | |
3979 | ||
3980 | /* no put on block group, done by btrfs_dec_nocow_writers */ | |
3981 | if (!ret) | |
3982 | btrfs_put_block_group(bg); | |
3983 | ||
3984 | return ret; | |
3985 | ||
3986 | } | |
3987 | ||
3988 | void btrfs_dec_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr) | |
3989 | { | |
3990 | struct btrfs_block_group_cache *bg; | |
3991 | ||
3992 | bg = btrfs_lookup_block_group(fs_info, bytenr); | |
3993 | ASSERT(bg); | |
3994 | if (atomic_dec_and_test(&bg->nocow_writers)) | |
4625956a | 3995 | wake_up_var(&bg->nocow_writers); |
f78c436c FM |
3996 | /* |
3997 | * Once for our lookup and once for the lookup done by a previous call | |
3998 | * to btrfs_inc_nocow_writers() | |
3999 | */ | |
4000 | btrfs_put_block_group(bg); | |
4001 | btrfs_put_block_group(bg); | |
4002 | } | |
4003 | ||
f78c436c FM |
4004 | void btrfs_wait_nocow_writers(struct btrfs_block_group_cache *bg) |
4005 | { | |
4625956a | 4006 | wait_var_event(&bg->nocow_writers, !atomic_read(&bg->nocow_writers)); |
f78c436c FM |
4007 | } |
4008 | ||
6ab0a202 JM |
4009 | static const char *alloc_name(u64 flags) |
4010 | { | |
4011 | switch (flags) { | |
4012 | case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA: | |
4013 | return "mixed"; | |
4014 | case BTRFS_BLOCK_GROUP_METADATA: | |
4015 | return "metadata"; | |
4016 | case BTRFS_BLOCK_GROUP_DATA: | |
4017 | return "data"; | |
4018 | case BTRFS_BLOCK_GROUP_SYSTEM: | |
4019 | return "system"; | |
4020 | default: | |
4021 | WARN_ON(1); | |
4022 | return "invalid-combination"; | |
4023 | }; | |
4024 | } | |
4025 | ||
4ca61683 | 4026 | static int create_space_info(struct btrfs_fs_info *info, u64 flags) |
2be12ef7 NB |
4027 | { |
4028 | ||
4029 | struct btrfs_space_info *space_info; | |
4030 | int i; | |
4031 | int ret; | |
4032 | ||
4033 | space_info = kzalloc(sizeof(*space_info), GFP_NOFS); | |
4034 | if (!space_info) | |
4035 | return -ENOMEM; | |
4036 | ||
4037 | ret = percpu_counter_init(&space_info->total_bytes_pinned, 0, | |
4038 | GFP_KERNEL); | |
4039 | if (ret) { | |
4040 | kfree(space_info); | |
4041 | return ret; | |
4042 | } | |
4043 | ||
4044 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) | |
4045 | INIT_LIST_HEAD(&space_info->block_groups[i]); | |
4046 | init_rwsem(&space_info->groups_sem); | |
4047 | spin_lock_init(&space_info->lock); | |
4048 | space_info->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK; | |
4049 | space_info->force_alloc = CHUNK_ALLOC_NO_FORCE; | |
4050 | init_waitqueue_head(&space_info->wait); | |
4051 | INIT_LIST_HEAD(&space_info->ro_bgs); | |
4052 | INIT_LIST_HEAD(&space_info->tickets); | |
4053 | INIT_LIST_HEAD(&space_info->priority_tickets); | |
4054 | ||
4055 | ret = kobject_init_and_add(&space_info->kobj, &space_info_ktype, | |
4056 | info->space_info_kobj, "%s", | |
4057 | alloc_name(space_info->flags)); | |
4058 | if (ret) { | |
4059 | percpu_counter_destroy(&space_info->total_bytes_pinned); | |
4060 | kfree(space_info); | |
4061 | return ret; | |
4062 | } | |
4063 | ||
2be12ef7 NB |
4064 | list_add_rcu(&space_info->list, &info->space_info); |
4065 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
4066 | info->data_sinfo = space_info; | |
4067 | ||
4068 | return ret; | |
4069 | } | |
4070 | ||
d2006e6d | 4071 | static void update_space_info(struct btrfs_fs_info *info, u64 flags, |
593060d7 | 4072 | u64 total_bytes, u64 bytes_used, |
e40edf2d | 4073 | u64 bytes_readonly, |
593060d7 CM |
4074 | struct btrfs_space_info **space_info) |
4075 | { | |
4076 | struct btrfs_space_info *found; | |
b742bb82 YZ |
4077 | int factor; |
4078 | ||
4079 | if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
4080 | BTRFS_BLOCK_GROUP_RAID10)) | |
4081 | factor = 2; | |
4082 | else | |
4083 | factor = 1; | |
593060d7 CM |
4084 | |
4085 | found = __find_space_info(info, flags); | |
d2006e6d NB |
4086 | ASSERT(found); |
4087 | spin_lock(&found->lock); | |
4088 | found->total_bytes += total_bytes; | |
4089 | found->disk_total += total_bytes * factor; | |
4090 | found->bytes_used += bytes_used; | |
4091 | found->disk_used += bytes_used * factor; | |
4092 | found->bytes_readonly += bytes_readonly; | |
4093 | if (total_bytes > 0) | |
4094 | found->full = 0; | |
4095 | space_info_add_new_bytes(info, found, total_bytes - | |
4096 | bytes_used - bytes_readonly); | |
4097 | spin_unlock(&found->lock); | |
4098 | *space_info = found; | |
593060d7 CM |
4099 | } |
4100 | ||
8790d502 CM |
4101 | static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
4102 | { | |
899c81ea ID |
4103 | u64 extra_flags = chunk_to_extended(flags) & |
4104 | BTRFS_EXTENDED_PROFILE_MASK; | |
a46d11a8 | 4105 | |
de98ced9 | 4106 | write_seqlock(&fs_info->profiles_lock); |
a46d11a8 ID |
4107 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
4108 | fs_info->avail_data_alloc_bits |= extra_flags; | |
4109 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
4110 | fs_info->avail_metadata_alloc_bits |= extra_flags; | |
4111 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
4112 | fs_info->avail_system_alloc_bits |= extra_flags; | |
de98ced9 | 4113 | write_sequnlock(&fs_info->profiles_lock); |
8790d502 | 4114 | } |
593060d7 | 4115 | |
fc67c450 ID |
4116 | /* |
4117 | * returns target flags in extended format or 0 if restripe for this | |
4118 | * chunk_type is not in progress | |
c6664b42 | 4119 | * |
dccdb07b | 4120 | * should be called with balance_lock held |
fc67c450 ID |
4121 | */ |
4122 | static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags) | |
4123 | { | |
4124 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
4125 | u64 target = 0; | |
4126 | ||
fc67c450 ID |
4127 | if (!bctl) |
4128 | return 0; | |
4129 | ||
4130 | if (flags & BTRFS_BLOCK_GROUP_DATA && | |
4131 | bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
4132 | target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target; | |
4133 | } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM && | |
4134 | bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
4135 | target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target; | |
4136 | } else if (flags & BTRFS_BLOCK_GROUP_METADATA && | |
4137 | bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
4138 | target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target; | |
4139 | } | |
4140 | ||
4141 | return target; | |
4142 | } | |
4143 | ||
a46d11a8 ID |
4144 | /* |
4145 | * @flags: available profiles in extended format (see ctree.h) | |
4146 | * | |
e4d8ec0f ID |
4147 | * Returns reduced profile in chunk format. If profile changing is in |
4148 | * progress (either running or paused) picks the target profile (if it's | |
4149 | * already available), otherwise falls back to plain reducing. | |
a46d11a8 | 4150 | */ |
2ff7e61e | 4151 | static u64 btrfs_reduce_alloc_profile(struct btrfs_fs_info *fs_info, u64 flags) |
ec44a35c | 4152 | { |
0b246afa | 4153 | u64 num_devices = fs_info->fs_devices->rw_devices; |
fc67c450 | 4154 | u64 target; |
9c170b26 ZL |
4155 | u64 raid_type; |
4156 | u64 allowed = 0; | |
a061fc8d | 4157 | |
fc67c450 ID |
4158 | /* |
4159 | * see if restripe for this chunk_type is in progress, if so | |
4160 | * try to reduce to the target profile | |
4161 | */ | |
0b246afa JM |
4162 | spin_lock(&fs_info->balance_lock); |
4163 | target = get_restripe_target(fs_info, flags); | |
fc67c450 ID |
4164 | if (target) { |
4165 | /* pick target profile only if it's already available */ | |
4166 | if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) { | |
0b246afa | 4167 | spin_unlock(&fs_info->balance_lock); |
fc67c450 | 4168 | return extended_to_chunk(target); |
e4d8ec0f ID |
4169 | } |
4170 | } | |
0b246afa | 4171 | spin_unlock(&fs_info->balance_lock); |
e4d8ec0f | 4172 | |
53b381b3 | 4173 | /* First, mask out the RAID levels which aren't possible */ |
9c170b26 ZL |
4174 | for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) { |
4175 | if (num_devices >= btrfs_raid_array[raid_type].devs_min) | |
41a6e891 | 4176 | allowed |= btrfs_raid_array[raid_type].bg_flag; |
9c170b26 ZL |
4177 | } |
4178 | allowed &= flags; | |
4179 | ||
4180 | if (allowed & BTRFS_BLOCK_GROUP_RAID6) | |
4181 | allowed = BTRFS_BLOCK_GROUP_RAID6; | |
4182 | else if (allowed & BTRFS_BLOCK_GROUP_RAID5) | |
4183 | allowed = BTRFS_BLOCK_GROUP_RAID5; | |
4184 | else if (allowed & BTRFS_BLOCK_GROUP_RAID10) | |
4185 | allowed = BTRFS_BLOCK_GROUP_RAID10; | |
4186 | else if (allowed & BTRFS_BLOCK_GROUP_RAID1) | |
4187 | allowed = BTRFS_BLOCK_GROUP_RAID1; | |
4188 | else if (allowed & BTRFS_BLOCK_GROUP_RAID0) | |
4189 | allowed = BTRFS_BLOCK_GROUP_RAID0; | |
4190 | ||
4191 | flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK; | |
4192 | ||
4193 | return extended_to_chunk(flags | allowed); | |
ec44a35c CM |
4194 | } |
4195 | ||
2ff7e61e | 4196 | static u64 get_alloc_profile(struct btrfs_fs_info *fs_info, u64 orig_flags) |
6a63209f | 4197 | { |
de98ced9 | 4198 | unsigned seq; |
f8213bdc | 4199 | u64 flags; |
de98ced9 MX |
4200 | |
4201 | do { | |
f8213bdc | 4202 | flags = orig_flags; |
0b246afa | 4203 | seq = read_seqbegin(&fs_info->profiles_lock); |
de98ced9 MX |
4204 | |
4205 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
0b246afa | 4206 | flags |= fs_info->avail_data_alloc_bits; |
de98ced9 | 4207 | else if (flags & BTRFS_BLOCK_GROUP_SYSTEM) |
0b246afa | 4208 | flags |= fs_info->avail_system_alloc_bits; |
de98ced9 | 4209 | else if (flags & BTRFS_BLOCK_GROUP_METADATA) |
0b246afa JM |
4210 | flags |= fs_info->avail_metadata_alloc_bits; |
4211 | } while (read_seqretry(&fs_info->profiles_lock, seq)); | |
6fef8df1 | 4212 | |
2ff7e61e | 4213 | return btrfs_reduce_alloc_profile(fs_info, flags); |
6a63209f JB |
4214 | } |
4215 | ||
1b86826d | 4216 | static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data) |
9ed74f2d | 4217 | { |
0b246afa | 4218 | struct btrfs_fs_info *fs_info = root->fs_info; |
b742bb82 | 4219 | u64 flags; |
53b381b3 | 4220 | u64 ret; |
9ed74f2d | 4221 | |
b742bb82 YZ |
4222 | if (data) |
4223 | flags = BTRFS_BLOCK_GROUP_DATA; | |
0b246afa | 4224 | else if (root == fs_info->chunk_root) |
b742bb82 | 4225 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
9ed74f2d | 4226 | else |
b742bb82 | 4227 | flags = BTRFS_BLOCK_GROUP_METADATA; |
9ed74f2d | 4228 | |
2ff7e61e | 4229 | ret = get_alloc_profile(fs_info, flags); |
53b381b3 | 4230 | return ret; |
6a63209f | 4231 | } |
9ed74f2d | 4232 | |
1b86826d JM |
4233 | u64 btrfs_data_alloc_profile(struct btrfs_fs_info *fs_info) |
4234 | { | |
4235 | return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_DATA); | |
4236 | } | |
4237 | ||
4238 | u64 btrfs_metadata_alloc_profile(struct btrfs_fs_info *fs_info) | |
4239 | { | |
4240 | return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_METADATA); | |
4241 | } | |
4242 | ||
4243 | u64 btrfs_system_alloc_profile(struct btrfs_fs_info *fs_info) | |
4244 | { | |
4245 | return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); | |
4246 | } | |
4247 | ||
4136135b LB |
4248 | static u64 btrfs_space_info_used(struct btrfs_space_info *s_info, |
4249 | bool may_use_included) | |
4250 | { | |
4251 | ASSERT(s_info); | |
4252 | return s_info->bytes_used + s_info->bytes_reserved + | |
4253 | s_info->bytes_pinned + s_info->bytes_readonly + | |
4254 | (may_use_included ? s_info->bytes_may_use : 0); | |
4255 | } | |
4256 | ||
04f4f916 | 4257 | int btrfs_alloc_data_chunk_ondemand(struct btrfs_inode *inode, u64 bytes) |
6a63209f | 4258 | { |
04f4f916 | 4259 | struct btrfs_root *root = inode->root; |
b4d7c3c9 | 4260 | struct btrfs_fs_info *fs_info = root->fs_info; |
1174cade | 4261 | struct btrfs_space_info *data_sinfo = fs_info->data_sinfo; |
ab6e2410 | 4262 | u64 used; |
94b947b2 | 4263 | int ret = 0; |
c99f1b0c ZL |
4264 | int need_commit = 2; |
4265 | int have_pinned_space; | |
6a63209f | 4266 | |
6a63209f | 4267 | /* make sure bytes are sectorsize aligned */ |
0b246afa | 4268 | bytes = ALIGN(bytes, fs_info->sectorsize); |
6a63209f | 4269 | |
9dced186 | 4270 | if (btrfs_is_free_space_inode(inode)) { |
c99f1b0c | 4271 | need_commit = 0; |
9dced186 | 4272 | ASSERT(current->journal_info); |
0af3d00b JB |
4273 | } |
4274 | ||
6a63209f JB |
4275 | again: |
4276 | /* make sure we have enough space to handle the data first */ | |
4277 | spin_lock(&data_sinfo->lock); | |
4136135b | 4278 | used = btrfs_space_info_used(data_sinfo, true); |
ab6e2410 JB |
4279 | |
4280 | if (used + bytes > data_sinfo->total_bytes) { | |
4e06bdd6 | 4281 | struct btrfs_trans_handle *trans; |
9ed74f2d | 4282 | |
6a63209f JB |
4283 | /* |
4284 | * if we don't have enough free bytes in this space then we need | |
4285 | * to alloc a new chunk. | |
4286 | */ | |
b9fd47cd | 4287 | if (!data_sinfo->full) { |
6a63209f | 4288 | u64 alloc_target; |
9ed74f2d | 4289 | |
0e4f8f88 | 4290 | data_sinfo->force_alloc = CHUNK_ALLOC_FORCE; |
6a63209f | 4291 | spin_unlock(&data_sinfo->lock); |
1174cade | 4292 | |
1b86826d | 4293 | alloc_target = btrfs_data_alloc_profile(fs_info); |
9dced186 MX |
4294 | /* |
4295 | * It is ugly that we don't call nolock join | |
4296 | * transaction for the free space inode case here. | |
4297 | * But it is safe because we only do the data space | |
4298 | * reservation for the free space cache in the | |
4299 | * transaction context, the common join transaction | |
4300 | * just increase the counter of the current transaction | |
4301 | * handler, doesn't try to acquire the trans_lock of | |
4302 | * the fs. | |
4303 | */ | |
7a7eaa40 | 4304 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
4305 | if (IS_ERR(trans)) |
4306 | return PTR_ERR(trans); | |
9ed74f2d | 4307 | |
01458828 | 4308 | ret = do_chunk_alloc(trans, alloc_target, |
0e4f8f88 | 4309 | CHUNK_ALLOC_NO_FORCE); |
3a45bb20 | 4310 | btrfs_end_transaction(trans); |
d52a5b5f MX |
4311 | if (ret < 0) { |
4312 | if (ret != -ENOSPC) | |
4313 | return ret; | |
c99f1b0c ZL |
4314 | else { |
4315 | have_pinned_space = 1; | |
d52a5b5f | 4316 | goto commit_trans; |
c99f1b0c | 4317 | } |
d52a5b5f | 4318 | } |
9ed74f2d | 4319 | |
6a63209f JB |
4320 | goto again; |
4321 | } | |
f2bb8f5c JB |
4322 | |
4323 | /* | |
b150a4f1 | 4324 | * If we don't have enough pinned space to deal with this |
94b947b2 ZL |
4325 | * allocation, and no removed chunk in current transaction, |
4326 | * don't bother committing the transaction. | |
f2bb8f5c | 4327 | */ |
c99f1b0c ZL |
4328 | have_pinned_space = percpu_counter_compare( |
4329 | &data_sinfo->total_bytes_pinned, | |
4330 | used + bytes - data_sinfo->total_bytes); | |
6a63209f | 4331 | spin_unlock(&data_sinfo->lock); |
6a63209f | 4332 | |
4e06bdd6 | 4333 | /* commit the current transaction and try again */ |
d52a5b5f | 4334 | commit_trans: |
92e2f7e3 | 4335 | if (need_commit) { |
c99f1b0c | 4336 | need_commit--; |
b150a4f1 | 4337 | |
e1746e83 | 4338 | if (need_commit > 0) { |
82b3e53b | 4339 | btrfs_start_delalloc_roots(fs_info, -1); |
6374e57a | 4340 | btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, |
0b246afa | 4341 | (u64)-1); |
e1746e83 | 4342 | } |
9a4e7276 | 4343 | |
7a7eaa40 | 4344 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
4345 | if (IS_ERR(trans)) |
4346 | return PTR_ERR(trans); | |
c99f1b0c | 4347 | if (have_pinned_space >= 0 || |
3204d33c JB |
4348 | test_bit(BTRFS_TRANS_HAVE_FREE_BGS, |
4349 | &trans->transaction->flags) || | |
c99f1b0c | 4350 | need_commit > 0) { |
3a45bb20 | 4351 | ret = btrfs_commit_transaction(trans); |
94b947b2 ZL |
4352 | if (ret) |
4353 | return ret; | |
d7c15171 | 4354 | /* |
c2d6cb16 FM |
4355 | * The cleaner kthread might still be doing iput |
4356 | * operations. Wait for it to finish so that | |
4357 | * more space is released. | |
d7c15171 | 4358 | */ |
0b246afa JM |
4359 | mutex_lock(&fs_info->cleaner_delayed_iput_mutex); |
4360 | mutex_unlock(&fs_info->cleaner_delayed_iput_mutex); | |
94b947b2 ZL |
4361 | goto again; |
4362 | } else { | |
3a45bb20 | 4363 | btrfs_end_transaction(trans); |
94b947b2 | 4364 | } |
4e06bdd6 | 4365 | } |
9ed74f2d | 4366 | |
0b246afa | 4367 | trace_btrfs_space_reservation(fs_info, |
cab45e22 JM |
4368 | "space_info:enospc", |
4369 | data_sinfo->flags, bytes, 1); | |
6a63209f JB |
4370 | return -ENOSPC; |
4371 | } | |
4372 | data_sinfo->bytes_may_use += bytes; | |
0b246afa | 4373 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 4374 | data_sinfo->flags, bytes, 1); |
6a63209f | 4375 | spin_unlock(&data_sinfo->lock); |
6a63209f | 4376 | |
237c0e9f | 4377 | return ret; |
9ed74f2d | 4378 | } |
6a63209f | 4379 | |
364ecf36 QW |
4380 | int btrfs_check_data_free_space(struct inode *inode, |
4381 | struct extent_changeset **reserved, u64 start, u64 len) | |
4ceff079 | 4382 | { |
0b246afa | 4383 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4ceff079 QW |
4384 | int ret; |
4385 | ||
4386 | /* align the range */ | |
0b246afa JM |
4387 | len = round_up(start + len, fs_info->sectorsize) - |
4388 | round_down(start, fs_info->sectorsize); | |
4389 | start = round_down(start, fs_info->sectorsize); | |
4ceff079 | 4390 | |
04f4f916 | 4391 | ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode), len); |
4ceff079 QW |
4392 | if (ret < 0) |
4393 | return ret; | |
4394 | ||
1e5ec2e7 | 4395 | /* Use new btrfs_qgroup_reserve_data to reserve precious data space. */ |
364ecf36 | 4396 | ret = btrfs_qgroup_reserve_data(inode, reserved, start, len); |
7bc329c1 | 4397 | if (ret < 0) |
1e5ec2e7 | 4398 | btrfs_free_reserved_data_space_noquota(inode, start, len); |
364ecf36 QW |
4399 | else |
4400 | ret = 0; | |
4ceff079 QW |
4401 | return ret; |
4402 | } | |
4403 | ||
4ceff079 QW |
4404 | /* |
4405 | * Called if we need to clear a data reservation for this inode | |
4406 | * Normally in a error case. | |
4407 | * | |
51773bec QW |
4408 | * This one will *NOT* use accurate qgroup reserved space API, just for case |
4409 | * which we can't sleep and is sure it won't affect qgroup reserved space. | |
4410 | * Like clear_bit_hook(). | |
4ceff079 | 4411 | */ |
51773bec QW |
4412 | void btrfs_free_reserved_data_space_noquota(struct inode *inode, u64 start, |
4413 | u64 len) | |
4ceff079 | 4414 | { |
0b246afa | 4415 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4ceff079 QW |
4416 | struct btrfs_space_info *data_sinfo; |
4417 | ||
4418 | /* Make sure the range is aligned to sectorsize */ | |
0b246afa JM |
4419 | len = round_up(start + len, fs_info->sectorsize) - |
4420 | round_down(start, fs_info->sectorsize); | |
4421 | start = round_down(start, fs_info->sectorsize); | |
4ceff079 | 4422 | |
0b246afa | 4423 | data_sinfo = fs_info->data_sinfo; |
4ceff079 QW |
4424 | spin_lock(&data_sinfo->lock); |
4425 | if (WARN_ON(data_sinfo->bytes_may_use < len)) | |
4426 | data_sinfo->bytes_may_use = 0; | |
4427 | else | |
4428 | data_sinfo->bytes_may_use -= len; | |
0b246afa | 4429 | trace_btrfs_space_reservation(fs_info, "space_info", |
4ceff079 QW |
4430 | data_sinfo->flags, len, 0); |
4431 | spin_unlock(&data_sinfo->lock); | |
4432 | } | |
4433 | ||
51773bec QW |
4434 | /* |
4435 | * Called if we need to clear a data reservation for this inode | |
4436 | * Normally in a error case. | |
4437 | * | |
01327610 | 4438 | * This one will handle the per-inode data rsv map for accurate reserved |
51773bec QW |
4439 | * space framework. |
4440 | */ | |
bc42bda2 QW |
4441 | void btrfs_free_reserved_data_space(struct inode *inode, |
4442 | struct extent_changeset *reserved, u64 start, u64 len) | |
51773bec | 4443 | { |
0c476a5d JM |
4444 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4445 | ||
4446 | /* Make sure the range is aligned to sectorsize */ | |
da17066c JM |
4447 | len = round_up(start + len, root->fs_info->sectorsize) - |
4448 | round_down(start, root->fs_info->sectorsize); | |
4449 | start = round_down(start, root->fs_info->sectorsize); | |
0c476a5d | 4450 | |
51773bec | 4451 | btrfs_free_reserved_data_space_noquota(inode, start, len); |
bc42bda2 | 4452 | btrfs_qgroup_free_data(inode, reserved, start, len); |
51773bec QW |
4453 | } |
4454 | ||
97e728d4 | 4455 | static void force_metadata_allocation(struct btrfs_fs_info *info) |
e3ccfa98 | 4456 | { |
97e728d4 JB |
4457 | struct list_head *head = &info->space_info; |
4458 | struct btrfs_space_info *found; | |
e3ccfa98 | 4459 | |
97e728d4 JB |
4460 | rcu_read_lock(); |
4461 | list_for_each_entry_rcu(found, head, list) { | |
4462 | if (found->flags & BTRFS_BLOCK_GROUP_METADATA) | |
0e4f8f88 | 4463 | found->force_alloc = CHUNK_ALLOC_FORCE; |
e3ccfa98 | 4464 | } |
97e728d4 | 4465 | rcu_read_unlock(); |
e3ccfa98 JB |
4466 | } |
4467 | ||
3c76cd84 MX |
4468 | static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global) |
4469 | { | |
4470 | return (global->size << 1); | |
4471 | } | |
4472 | ||
2ff7e61e | 4473 | static int should_alloc_chunk(struct btrfs_fs_info *fs_info, |
698d0082 | 4474 | struct btrfs_space_info *sinfo, int force) |
32c00aff | 4475 | { |
0b246afa | 4476 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; |
8d8aafee | 4477 | u64 bytes_used = btrfs_space_info_used(sinfo, false); |
e5bc2458 | 4478 | u64 thresh; |
e3ccfa98 | 4479 | |
0e4f8f88 CM |
4480 | if (force == CHUNK_ALLOC_FORCE) |
4481 | return 1; | |
4482 | ||
fb25e914 JB |
4483 | /* |
4484 | * We need to take into account the global rsv because for all intents | |
4485 | * and purposes it's used space. Don't worry about locking the | |
4486 | * global_rsv, it doesn't change except when the transaction commits. | |
4487 | */ | |
54338b5c | 4488 | if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA) |
8d8aafee | 4489 | bytes_used += calc_global_rsv_need_space(global_rsv); |
fb25e914 | 4490 | |
0e4f8f88 CM |
4491 | /* |
4492 | * in limited mode, we want to have some free space up to | |
4493 | * about 1% of the FS size. | |
4494 | */ | |
4495 | if (force == CHUNK_ALLOC_LIMITED) { | |
0b246afa | 4496 | thresh = btrfs_super_total_bytes(fs_info->super_copy); |
ee22184b | 4497 | thresh = max_t(u64, SZ_64M, div_factor_fine(thresh, 1)); |
0e4f8f88 | 4498 | |
8d8aafee | 4499 | if (sinfo->total_bytes - bytes_used < thresh) |
0e4f8f88 CM |
4500 | return 1; |
4501 | } | |
0e4f8f88 | 4502 | |
8d8aafee | 4503 | if (bytes_used + SZ_2M < div_factor(sinfo->total_bytes, 8)) |
14ed0ca6 | 4504 | return 0; |
424499db | 4505 | return 1; |
32c00aff JB |
4506 | } |
4507 | ||
2ff7e61e | 4508 | static u64 get_profile_num_devs(struct btrfs_fs_info *fs_info, u64 type) |
15d1ff81 LB |
4509 | { |
4510 | u64 num_dev; | |
4511 | ||
53b381b3 DW |
4512 | if (type & (BTRFS_BLOCK_GROUP_RAID10 | |
4513 | BTRFS_BLOCK_GROUP_RAID0 | | |
4514 | BTRFS_BLOCK_GROUP_RAID5 | | |
4515 | BTRFS_BLOCK_GROUP_RAID6)) | |
0b246afa | 4516 | num_dev = fs_info->fs_devices->rw_devices; |
15d1ff81 LB |
4517 | else if (type & BTRFS_BLOCK_GROUP_RAID1) |
4518 | num_dev = 2; | |
4519 | else | |
4520 | num_dev = 1; /* DUP or single */ | |
4521 | ||
39c2d7fa | 4522 | return num_dev; |
15d1ff81 LB |
4523 | } |
4524 | ||
39c2d7fa FM |
4525 | /* |
4526 | * If @is_allocation is true, reserve space in the system space info necessary | |
4527 | * for allocating a chunk, otherwise if it's false, reserve space necessary for | |
4528 | * removing a chunk. | |
4529 | */ | |
451a2c13 | 4530 | void check_system_chunk(struct btrfs_trans_handle *trans, u64 type) |
15d1ff81 | 4531 | { |
451a2c13 | 4532 | struct btrfs_fs_info *fs_info = trans->fs_info; |
15d1ff81 LB |
4533 | struct btrfs_space_info *info; |
4534 | u64 left; | |
4535 | u64 thresh; | |
4fbcdf66 | 4536 | int ret = 0; |
39c2d7fa | 4537 | u64 num_devs; |
4fbcdf66 FM |
4538 | |
4539 | /* | |
4540 | * Needed because we can end up allocating a system chunk and for an | |
4541 | * atomic and race free space reservation in the chunk block reserve. | |
4542 | */ | |
a32bf9a3 | 4543 | lockdep_assert_held(&fs_info->chunk_mutex); |
15d1ff81 | 4544 | |
0b246afa | 4545 | info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); |
15d1ff81 | 4546 | spin_lock(&info->lock); |
4136135b | 4547 | left = info->total_bytes - btrfs_space_info_used(info, true); |
15d1ff81 LB |
4548 | spin_unlock(&info->lock); |
4549 | ||
2ff7e61e | 4550 | num_devs = get_profile_num_devs(fs_info, type); |
39c2d7fa FM |
4551 | |
4552 | /* num_devs device items to update and 1 chunk item to add or remove */ | |
0b246afa JM |
4553 | thresh = btrfs_calc_trunc_metadata_size(fs_info, num_devs) + |
4554 | btrfs_calc_trans_metadata_size(fs_info, 1); | |
39c2d7fa | 4555 | |
0b246afa JM |
4556 | if (left < thresh && btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
4557 | btrfs_info(fs_info, "left=%llu, need=%llu, flags=%llu", | |
4558 | left, thresh, type); | |
4559 | dump_space_info(fs_info, info, 0, 0); | |
15d1ff81 LB |
4560 | } |
4561 | ||
4562 | if (left < thresh) { | |
1b86826d | 4563 | u64 flags = btrfs_system_alloc_profile(fs_info); |
15d1ff81 | 4564 | |
4fbcdf66 FM |
4565 | /* |
4566 | * Ignore failure to create system chunk. We might end up not | |
4567 | * needing it, as we might not need to COW all nodes/leafs from | |
4568 | * the paths we visit in the chunk tree (they were already COWed | |
4569 | * or created in the current transaction for example). | |
4570 | */ | |
c216b203 | 4571 | ret = btrfs_alloc_chunk(trans, flags); |
4fbcdf66 FM |
4572 | } |
4573 | ||
4574 | if (!ret) { | |
0b246afa JM |
4575 | ret = btrfs_block_rsv_add(fs_info->chunk_root, |
4576 | &fs_info->chunk_block_rsv, | |
4fbcdf66 FM |
4577 | thresh, BTRFS_RESERVE_NO_FLUSH); |
4578 | if (!ret) | |
4579 | trans->chunk_bytes_reserved += thresh; | |
15d1ff81 LB |
4580 | } |
4581 | } | |
4582 | ||
28b737f6 LB |
4583 | /* |
4584 | * If force is CHUNK_ALLOC_FORCE: | |
4585 | * - return 1 if it successfully allocates a chunk, | |
4586 | * - return errors including -ENOSPC otherwise. | |
4587 | * If force is NOT CHUNK_ALLOC_FORCE: | |
4588 | * - return 0 if it doesn't need to allocate a new chunk, | |
4589 | * - return 1 if it successfully allocates a chunk, | |
4590 | * - return errors including -ENOSPC otherwise. | |
4591 | */ | |
01458828 NB |
4592 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, u64 flags, |
4593 | int force) | |
9ed74f2d | 4594 | { |
01458828 | 4595 | struct btrfs_fs_info *fs_info = trans->fs_info; |
6324fbf3 | 4596 | struct btrfs_space_info *space_info; |
6d74119f | 4597 | int wait_for_alloc = 0; |
9ed74f2d | 4598 | int ret = 0; |
9ed74f2d | 4599 | |
c6b305a8 JB |
4600 | /* Don't re-enter if we're already allocating a chunk */ |
4601 | if (trans->allocating_chunk) | |
4602 | return -ENOSPC; | |
4603 | ||
0b246afa | 4604 | space_info = __find_space_info(fs_info, flags); |
dc2d3005 | 4605 | ASSERT(space_info); |
9ed74f2d | 4606 | |
6d74119f | 4607 | again: |
25179201 | 4608 | spin_lock(&space_info->lock); |
9e622d6b | 4609 | if (force < space_info->force_alloc) |
0e4f8f88 | 4610 | force = space_info->force_alloc; |
25179201 | 4611 | if (space_info->full) { |
2ff7e61e | 4612 | if (should_alloc_chunk(fs_info, space_info, force)) |
09fb99a6 FDBM |
4613 | ret = -ENOSPC; |
4614 | else | |
4615 | ret = 0; | |
25179201 | 4616 | spin_unlock(&space_info->lock); |
09fb99a6 | 4617 | return ret; |
9ed74f2d JB |
4618 | } |
4619 | ||
2ff7e61e | 4620 | if (!should_alloc_chunk(fs_info, space_info, force)) { |
25179201 | 4621 | spin_unlock(&space_info->lock); |
6d74119f JB |
4622 | return 0; |
4623 | } else if (space_info->chunk_alloc) { | |
4624 | wait_for_alloc = 1; | |
4625 | } else { | |
4626 | space_info->chunk_alloc = 1; | |
9ed74f2d | 4627 | } |
0e4f8f88 | 4628 | |
25179201 | 4629 | spin_unlock(&space_info->lock); |
9ed74f2d | 4630 | |
6d74119f JB |
4631 | mutex_lock(&fs_info->chunk_mutex); |
4632 | ||
4633 | /* | |
4634 | * The chunk_mutex is held throughout the entirety of a chunk | |
4635 | * allocation, so once we've acquired the chunk_mutex we know that the | |
4636 | * other guy is done and we need to recheck and see if we should | |
4637 | * allocate. | |
4638 | */ | |
4639 | if (wait_for_alloc) { | |
4640 | mutex_unlock(&fs_info->chunk_mutex); | |
4641 | wait_for_alloc = 0; | |
1e1c50a9 | 4642 | cond_resched(); |
6d74119f JB |
4643 | goto again; |
4644 | } | |
4645 | ||
c6b305a8 JB |
4646 | trans->allocating_chunk = true; |
4647 | ||
67377734 JB |
4648 | /* |
4649 | * If we have mixed data/metadata chunks we want to make sure we keep | |
4650 | * allocating mixed chunks instead of individual chunks. | |
4651 | */ | |
4652 | if (btrfs_mixed_space_info(space_info)) | |
4653 | flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA); | |
4654 | ||
97e728d4 JB |
4655 | /* |
4656 | * if we're doing a data chunk, go ahead and make sure that | |
4657 | * we keep a reasonable number of metadata chunks allocated in the | |
4658 | * FS as well. | |
4659 | */ | |
9ed74f2d | 4660 | if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) { |
97e728d4 JB |
4661 | fs_info->data_chunk_allocations++; |
4662 | if (!(fs_info->data_chunk_allocations % | |
4663 | fs_info->metadata_ratio)) | |
4664 | force_metadata_allocation(fs_info); | |
9ed74f2d JB |
4665 | } |
4666 | ||
15d1ff81 LB |
4667 | /* |
4668 | * Check if we have enough space in SYSTEM chunk because we may need | |
4669 | * to update devices. | |
4670 | */ | |
451a2c13 | 4671 | check_system_chunk(trans, flags); |
15d1ff81 | 4672 | |
c216b203 | 4673 | ret = btrfs_alloc_chunk(trans, flags); |
c6b305a8 | 4674 | trans->allocating_chunk = false; |
92b8e897 | 4675 | |
9ed74f2d | 4676 | spin_lock(&space_info->lock); |
57f1642e NB |
4677 | if (ret < 0) { |
4678 | if (ret == -ENOSPC) | |
4679 | space_info->full = 1; | |
4680 | else | |
4681 | goto out; | |
4682 | } else { | |
424499db | 4683 | ret = 1; |
57f1642e | 4684 | } |
6d74119f | 4685 | |
0e4f8f88 | 4686 | space_info->force_alloc = CHUNK_ALLOC_NO_FORCE; |
a81cb9a2 | 4687 | out: |
6d74119f | 4688 | space_info->chunk_alloc = 0; |
9ed74f2d | 4689 | spin_unlock(&space_info->lock); |
a25c75d5 | 4690 | mutex_unlock(&fs_info->chunk_mutex); |
00d80e34 FM |
4691 | /* |
4692 | * When we allocate a new chunk we reserve space in the chunk block | |
4693 | * reserve to make sure we can COW nodes/leafs in the chunk tree or | |
4694 | * add new nodes/leafs to it if we end up needing to do it when | |
4695 | * inserting the chunk item and updating device items as part of the | |
4696 | * second phase of chunk allocation, performed by | |
4697 | * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a | |
4698 | * large number of new block groups to create in our transaction | |
4699 | * handle's new_bgs list to avoid exhausting the chunk block reserve | |
4700 | * in extreme cases - like having a single transaction create many new | |
4701 | * block groups when starting to write out the free space caches of all | |
4702 | * the block groups that were made dirty during the lifetime of the | |
4703 | * transaction. | |
4704 | */ | |
d9a0540a | 4705 | if (trans->can_flush_pending_bgs && |
ee22184b | 4706 | trans->chunk_bytes_reserved >= (u64)SZ_2M) { |
6c686b35 | 4707 | btrfs_create_pending_block_groups(trans); |
00d80e34 FM |
4708 | btrfs_trans_release_chunk_metadata(trans); |
4709 | } | |
0f9dd46c | 4710 | return ret; |
6324fbf3 | 4711 | } |
9ed74f2d | 4712 | |
c1c4919b | 4713 | static int can_overcommit(struct btrfs_fs_info *fs_info, |
a80c8dcf | 4714 | struct btrfs_space_info *space_info, u64 bytes, |
c1c4919b JM |
4715 | enum btrfs_reserve_flush_enum flush, |
4716 | bool system_chunk) | |
a80c8dcf | 4717 | { |
0b246afa | 4718 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; |
957780eb | 4719 | u64 profile; |
3c76cd84 | 4720 | u64 space_size; |
a80c8dcf JB |
4721 | u64 avail; |
4722 | u64 used; | |
4723 | ||
957780eb JB |
4724 | /* Don't overcommit when in mixed mode. */ |
4725 | if (space_info->flags & BTRFS_BLOCK_GROUP_DATA) | |
4726 | return 0; | |
4727 | ||
c1c4919b JM |
4728 | if (system_chunk) |
4729 | profile = btrfs_system_alloc_profile(fs_info); | |
4730 | else | |
4731 | profile = btrfs_metadata_alloc_profile(fs_info); | |
4732 | ||
4136135b | 4733 | used = btrfs_space_info_used(space_info, false); |
96f1bb57 | 4734 | |
96f1bb57 JB |
4735 | /* |
4736 | * We only want to allow over committing if we have lots of actual space | |
4737 | * free, but if we don't have enough space to handle the global reserve | |
4738 | * space then we could end up having a real enospc problem when trying | |
4739 | * to allocate a chunk or some other such important allocation. | |
4740 | */ | |
3c76cd84 MX |
4741 | spin_lock(&global_rsv->lock); |
4742 | space_size = calc_global_rsv_need_space(global_rsv); | |
4743 | spin_unlock(&global_rsv->lock); | |
4744 | if (used + space_size >= space_info->total_bytes) | |
96f1bb57 JB |
4745 | return 0; |
4746 | ||
4747 | used += space_info->bytes_may_use; | |
a80c8dcf | 4748 | |
a5ed45f8 | 4749 | avail = atomic64_read(&fs_info->free_chunk_space); |
a80c8dcf JB |
4750 | |
4751 | /* | |
4752 | * If we have dup, raid1 or raid10 then only half of the free | |
53b381b3 DW |
4753 | * space is actually useable. For raid56, the space info used |
4754 | * doesn't include the parity drive, so we don't have to | |
4755 | * change the math | |
a80c8dcf JB |
4756 | */ |
4757 | if (profile & (BTRFS_BLOCK_GROUP_DUP | | |
4758 | BTRFS_BLOCK_GROUP_RAID1 | | |
4759 | BTRFS_BLOCK_GROUP_RAID10)) | |
4760 | avail >>= 1; | |
4761 | ||
4762 | /* | |
561c294d MX |
4763 | * If we aren't flushing all things, let us overcommit up to |
4764 | * 1/2th of the space. If we can flush, don't let us overcommit | |
4765 | * too much, let it overcommit up to 1/8 of the space. | |
a80c8dcf | 4766 | */ |
08e007d2 | 4767 | if (flush == BTRFS_RESERVE_FLUSH_ALL) |
14575aef | 4768 | avail >>= 3; |
a80c8dcf | 4769 | else |
14575aef | 4770 | avail >>= 1; |
a80c8dcf | 4771 | |
14575aef | 4772 | if (used + bytes < space_info->total_bytes + avail) |
a80c8dcf JB |
4773 | return 1; |
4774 | return 0; | |
4775 | } | |
4776 | ||
2ff7e61e | 4777 | static void btrfs_writeback_inodes_sb_nr(struct btrfs_fs_info *fs_info, |
6c255e67 | 4778 | unsigned long nr_pages, int nr_items) |
da633a42 | 4779 | { |
0b246afa | 4780 | struct super_block *sb = fs_info->sb; |
da633a42 | 4781 | |
925a6efb JB |
4782 | if (down_read_trylock(&sb->s_umount)) { |
4783 | writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE); | |
4784 | up_read(&sb->s_umount); | |
4785 | } else { | |
da633a42 MX |
4786 | /* |
4787 | * We needn't worry the filesystem going from r/w to r/o though | |
4788 | * we don't acquire ->s_umount mutex, because the filesystem | |
4789 | * should guarantee the delalloc inodes list be empty after | |
4790 | * the filesystem is readonly(all dirty pages are written to | |
4791 | * the disk). | |
4792 | */ | |
82b3e53b | 4793 | btrfs_start_delalloc_roots(fs_info, nr_items); |
98ad69cf | 4794 | if (!current->journal_info) |
0b246afa | 4795 | btrfs_wait_ordered_roots(fs_info, nr_items, 0, (u64)-1); |
da633a42 MX |
4796 | } |
4797 | } | |
4798 | ||
6374e57a | 4799 | static inline u64 calc_reclaim_items_nr(struct btrfs_fs_info *fs_info, |
2ff7e61e | 4800 | u64 to_reclaim) |
18cd8ea6 MX |
4801 | { |
4802 | u64 bytes; | |
6374e57a | 4803 | u64 nr; |
18cd8ea6 | 4804 | |
2ff7e61e | 4805 | bytes = btrfs_calc_trans_metadata_size(fs_info, 1); |
6374e57a | 4806 | nr = div64_u64(to_reclaim, bytes); |
18cd8ea6 MX |
4807 | if (!nr) |
4808 | nr = 1; | |
4809 | return nr; | |
4810 | } | |
4811 | ||
ee22184b | 4812 | #define EXTENT_SIZE_PER_ITEM SZ_256K |
c61a16a7 | 4813 | |
9ed74f2d | 4814 | /* |
5da9d01b | 4815 | * shrink metadata reservation for delalloc |
9ed74f2d | 4816 | */ |
c1c4919b JM |
4817 | static void shrink_delalloc(struct btrfs_fs_info *fs_info, u64 to_reclaim, |
4818 | u64 orig, bool wait_ordered) | |
5da9d01b | 4819 | { |
0019f10d | 4820 | struct btrfs_space_info *space_info; |
663350ac | 4821 | struct btrfs_trans_handle *trans; |
f4c738c2 | 4822 | u64 delalloc_bytes; |
5da9d01b | 4823 | u64 max_reclaim; |
6374e57a | 4824 | u64 items; |
b1953bce | 4825 | long time_left; |
d3ee29e3 MX |
4826 | unsigned long nr_pages; |
4827 | int loops; | |
5da9d01b | 4828 | |
c61a16a7 | 4829 | /* Calc the number of the pages we need flush for space reservation */ |
2ff7e61e | 4830 | items = calc_reclaim_items_nr(fs_info, to_reclaim); |
6374e57a | 4831 | to_reclaim = items * EXTENT_SIZE_PER_ITEM; |
c61a16a7 | 4832 | |
663350ac | 4833 | trans = (struct btrfs_trans_handle *)current->journal_info; |
69fe2d75 | 4834 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
bf9022e0 | 4835 | |
963d678b | 4836 | delalloc_bytes = percpu_counter_sum_positive( |
0b246afa | 4837 | &fs_info->delalloc_bytes); |
f4c738c2 | 4838 | if (delalloc_bytes == 0) { |
fdb5effd | 4839 | if (trans) |
f4c738c2 | 4840 | return; |
38c135af | 4841 | if (wait_ordered) |
0b246afa | 4842 | btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1); |
f4c738c2 | 4843 | return; |
fdb5effd JB |
4844 | } |
4845 | ||
d3ee29e3 | 4846 | loops = 0; |
f4c738c2 JB |
4847 | while (delalloc_bytes && loops < 3) { |
4848 | max_reclaim = min(delalloc_bytes, to_reclaim); | |
09cbfeaf | 4849 | nr_pages = max_reclaim >> PAGE_SHIFT; |
2ff7e61e | 4850 | btrfs_writeback_inodes_sb_nr(fs_info, nr_pages, items); |
dea31f52 JB |
4851 | /* |
4852 | * We need to wait for the async pages to actually start before | |
4853 | * we do anything. | |
4854 | */ | |
0b246afa | 4855 | max_reclaim = atomic_read(&fs_info->async_delalloc_pages); |
9f3a074d MX |
4856 | if (!max_reclaim) |
4857 | goto skip_async; | |
4858 | ||
4859 | if (max_reclaim <= nr_pages) | |
4860 | max_reclaim = 0; | |
4861 | else | |
4862 | max_reclaim -= nr_pages; | |
dea31f52 | 4863 | |
0b246afa JM |
4864 | wait_event(fs_info->async_submit_wait, |
4865 | atomic_read(&fs_info->async_delalloc_pages) <= | |
9f3a074d MX |
4866 | (int)max_reclaim); |
4867 | skip_async: | |
0019f10d | 4868 | spin_lock(&space_info->lock); |
957780eb JB |
4869 | if (list_empty(&space_info->tickets) && |
4870 | list_empty(&space_info->priority_tickets)) { | |
4871 | spin_unlock(&space_info->lock); | |
4872 | break; | |
4873 | } | |
0019f10d | 4874 | spin_unlock(&space_info->lock); |
5da9d01b | 4875 | |
36e39c40 | 4876 | loops++; |
f104d044 | 4877 | if (wait_ordered && !trans) { |
0b246afa | 4878 | btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1); |
f104d044 | 4879 | } else { |
f4c738c2 | 4880 | time_left = schedule_timeout_killable(1); |
f104d044 JB |
4881 | if (time_left) |
4882 | break; | |
4883 | } | |
963d678b | 4884 | delalloc_bytes = percpu_counter_sum_positive( |
0b246afa | 4885 | &fs_info->delalloc_bytes); |
5da9d01b | 4886 | } |
5da9d01b YZ |
4887 | } |
4888 | ||
996478ca JB |
4889 | struct reserve_ticket { |
4890 | u64 bytes; | |
4891 | int error; | |
4892 | struct list_head list; | |
4893 | wait_queue_head_t wait; | |
4894 | }; | |
4895 | ||
663350ac JB |
4896 | /** |
4897 | * maybe_commit_transaction - possibly commit the transaction if its ok to | |
4898 | * @root - the root we're allocating for | |
4899 | * @bytes - the number of bytes we want to reserve | |
4900 | * @force - force the commit | |
8bb8ab2e | 4901 | * |
663350ac JB |
4902 | * This will check to make sure that committing the transaction will actually |
4903 | * get us somewhere and then commit the transaction if it does. Otherwise it | |
4904 | * will return -ENOSPC. | |
8bb8ab2e | 4905 | */ |
0c9ab349 | 4906 | static int may_commit_transaction(struct btrfs_fs_info *fs_info, |
996478ca | 4907 | struct btrfs_space_info *space_info) |
663350ac | 4908 | { |
996478ca | 4909 | struct reserve_ticket *ticket = NULL; |
0b246afa | 4910 | struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_block_rsv; |
663350ac | 4911 | struct btrfs_trans_handle *trans; |
996478ca | 4912 | u64 bytes; |
663350ac JB |
4913 | |
4914 | trans = (struct btrfs_trans_handle *)current->journal_info; | |
4915 | if (trans) | |
4916 | return -EAGAIN; | |
4917 | ||
996478ca JB |
4918 | spin_lock(&space_info->lock); |
4919 | if (!list_empty(&space_info->priority_tickets)) | |
4920 | ticket = list_first_entry(&space_info->priority_tickets, | |
4921 | struct reserve_ticket, list); | |
4922 | else if (!list_empty(&space_info->tickets)) | |
4923 | ticket = list_first_entry(&space_info->tickets, | |
4924 | struct reserve_ticket, list); | |
4925 | bytes = (ticket) ? ticket->bytes : 0; | |
4926 | spin_unlock(&space_info->lock); | |
4927 | ||
4928 | if (!bytes) | |
4929 | return 0; | |
663350ac JB |
4930 | |
4931 | /* See if there is enough pinned space to make this reservation */ | |
b150a4f1 | 4932 | if (percpu_counter_compare(&space_info->total_bytes_pinned, |
0424c548 | 4933 | bytes) >= 0) |
663350ac | 4934 | goto commit; |
663350ac JB |
4935 | |
4936 | /* | |
4937 | * See if there is some space in the delayed insertion reservation for | |
4938 | * this reservation. | |
4939 | */ | |
4940 | if (space_info != delayed_rsv->space_info) | |
4941 | return -ENOSPC; | |
4942 | ||
4943 | spin_lock(&delayed_rsv->lock); | |
996478ca JB |
4944 | if (delayed_rsv->size > bytes) |
4945 | bytes = 0; | |
4946 | else | |
4947 | bytes -= delayed_rsv->size; | |
057aac3e NB |
4948 | spin_unlock(&delayed_rsv->lock); |
4949 | ||
b150a4f1 | 4950 | if (percpu_counter_compare(&space_info->total_bytes_pinned, |
996478ca | 4951 | bytes) < 0) { |
663350ac JB |
4952 | return -ENOSPC; |
4953 | } | |
663350ac JB |
4954 | |
4955 | commit: | |
a9b3311e | 4956 | trans = btrfs_join_transaction(fs_info->extent_root); |
663350ac JB |
4957 | if (IS_ERR(trans)) |
4958 | return -ENOSPC; | |
4959 | ||
3a45bb20 | 4960 | return btrfs_commit_transaction(trans); |
663350ac JB |
4961 | } |
4962 | ||
e38ae7a0 NB |
4963 | /* |
4964 | * Try to flush some data based on policy set by @state. This is only advisory | |
4965 | * and may fail for various reasons. The caller is supposed to examine the | |
4966 | * state of @space_info to detect the outcome. | |
4967 | */ | |
4968 | static void flush_space(struct btrfs_fs_info *fs_info, | |
96c3f433 | 4969 | struct btrfs_space_info *space_info, u64 num_bytes, |
7bdd6277 | 4970 | int state) |
96c3f433 | 4971 | { |
a9b3311e | 4972 | struct btrfs_root *root = fs_info->extent_root; |
96c3f433 JB |
4973 | struct btrfs_trans_handle *trans; |
4974 | int nr; | |
f4c738c2 | 4975 | int ret = 0; |
96c3f433 JB |
4976 | |
4977 | switch (state) { | |
96c3f433 JB |
4978 | case FLUSH_DELAYED_ITEMS_NR: |
4979 | case FLUSH_DELAYED_ITEMS: | |
18cd8ea6 | 4980 | if (state == FLUSH_DELAYED_ITEMS_NR) |
2ff7e61e | 4981 | nr = calc_reclaim_items_nr(fs_info, num_bytes) * 2; |
18cd8ea6 | 4982 | else |
96c3f433 | 4983 | nr = -1; |
18cd8ea6 | 4984 | |
96c3f433 JB |
4985 | trans = btrfs_join_transaction(root); |
4986 | if (IS_ERR(trans)) { | |
4987 | ret = PTR_ERR(trans); | |
4988 | break; | |
4989 | } | |
e5c304e6 | 4990 | ret = btrfs_run_delayed_items_nr(trans, nr); |
3a45bb20 | 4991 | btrfs_end_transaction(trans); |
96c3f433 | 4992 | break; |
67b0fd63 JB |
4993 | case FLUSH_DELALLOC: |
4994 | case FLUSH_DELALLOC_WAIT: | |
7bdd6277 | 4995 | shrink_delalloc(fs_info, num_bytes * 2, num_bytes, |
67b0fd63 JB |
4996 | state == FLUSH_DELALLOC_WAIT); |
4997 | break; | |
ea658bad JB |
4998 | case ALLOC_CHUNK: |
4999 | trans = btrfs_join_transaction(root); | |
5000 | if (IS_ERR(trans)) { | |
5001 | ret = PTR_ERR(trans); | |
5002 | break; | |
5003 | } | |
01458828 | 5004 | ret = do_chunk_alloc(trans, |
1b86826d | 5005 | btrfs_metadata_alloc_profile(fs_info), |
ea658bad | 5006 | CHUNK_ALLOC_NO_FORCE); |
3a45bb20 | 5007 | btrfs_end_transaction(trans); |
eecba891 | 5008 | if (ret > 0 || ret == -ENOSPC) |
ea658bad JB |
5009 | ret = 0; |
5010 | break; | |
96c3f433 | 5011 | case COMMIT_TRANS: |
996478ca | 5012 | ret = may_commit_transaction(fs_info, space_info); |
96c3f433 JB |
5013 | break; |
5014 | default: | |
5015 | ret = -ENOSPC; | |
5016 | break; | |
5017 | } | |
5018 | ||
7bdd6277 NB |
5019 | trace_btrfs_flush_space(fs_info, space_info->flags, num_bytes, state, |
5020 | ret); | |
e38ae7a0 | 5021 | return; |
96c3f433 | 5022 | } |
21c7e756 MX |
5023 | |
5024 | static inline u64 | |
c1c4919b JM |
5025 | btrfs_calc_reclaim_metadata_size(struct btrfs_fs_info *fs_info, |
5026 | struct btrfs_space_info *space_info, | |
5027 | bool system_chunk) | |
21c7e756 | 5028 | { |
957780eb | 5029 | struct reserve_ticket *ticket; |
21c7e756 MX |
5030 | u64 used; |
5031 | u64 expected; | |
957780eb | 5032 | u64 to_reclaim = 0; |
21c7e756 | 5033 | |
957780eb JB |
5034 | list_for_each_entry(ticket, &space_info->tickets, list) |
5035 | to_reclaim += ticket->bytes; | |
5036 | list_for_each_entry(ticket, &space_info->priority_tickets, list) | |
5037 | to_reclaim += ticket->bytes; | |
5038 | if (to_reclaim) | |
5039 | return to_reclaim; | |
21c7e756 | 5040 | |
e0af2484 | 5041 | to_reclaim = min_t(u64, num_online_cpus() * SZ_1M, SZ_16M); |
c1c4919b JM |
5042 | if (can_overcommit(fs_info, space_info, to_reclaim, |
5043 | BTRFS_RESERVE_FLUSH_ALL, system_chunk)) | |
e0af2484 WX |
5044 | return 0; |
5045 | ||
0eee8a49 NB |
5046 | used = btrfs_space_info_used(space_info, true); |
5047 | ||
c1c4919b JM |
5048 | if (can_overcommit(fs_info, space_info, SZ_1M, |
5049 | BTRFS_RESERVE_FLUSH_ALL, system_chunk)) | |
21c7e756 MX |
5050 | expected = div_factor_fine(space_info->total_bytes, 95); |
5051 | else | |
5052 | expected = div_factor_fine(space_info->total_bytes, 90); | |
5053 | ||
5054 | if (used > expected) | |
5055 | to_reclaim = used - expected; | |
5056 | else | |
5057 | to_reclaim = 0; | |
5058 | to_reclaim = min(to_reclaim, space_info->bytes_may_use + | |
5059 | space_info->bytes_reserved); | |
21c7e756 MX |
5060 | return to_reclaim; |
5061 | } | |
5062 | ||
c1c4919b JM |
5063 | static inline int need_do_async_reclaim(struct btrfs_fs_info *fs_info, |
5064 | struct btrfs_space_info *space_info, | |
5065 | u64 used, bool system_chunk) | |
21c7e756 | 5066 | { |
365c5313 JB |
5067 | u64 thresh = div_factor_fine(space_info->total_bytes, 98); |
5068 | ||
5069 | /* If we're just plain full then async reclaim just slows us down. */ | |
baee8790 | 5070 | if ((space_info->bytes_used + space_info->bytes_reserved) >= thresh) |
365c5313 JB |
5071 | return 0; |
5072 | ||
c1c4919b JM |
5073 | if (!btrfs_calc_reclaim_metadata_size(fs_info, space_info, |
5074 | system_chunk)) | |
d38b349c JB |
5075 | return 0; |
5076 | ||
0b246afa JM |
5077 | return (used >= thresh && !btrfs_fs_closing(fs_info) && |
5078 | !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state)); | |
21c7e756 MX |
5079 | } |
5080 | ||
957780eb | 5081 | static void wake_all_tickets(struct list_head *head) |
21c7e756 | 5082 | { |
957780eb | 5083 | struct reserve_ticket *ticket; |
25ce459c | 5084 | |
957780eb JB |
5085 | while (!list_empty(head)) { |
5086 | ticket = list_first_entry(head, struct reserve_ticket, list); | |
5087 | list_del_init(&ticket->list); | |
5088 | ticket->error = -ENOSPC; | |
5089 | wake_up(&ticket->wait); | |
21c7e756 | 5090 | } |
21c7e756 MX |
5091 | } |
5092 | ||
957780eb JB |
5093 | /* |
5094 | * This is for normal flushers, we can wait all goddamned day if we want to. We | |
5095 | * will loop and continuously try to flush as long as we are making progress. | |
5096 | * We count progress as clearing off tickets each time we have to loop. | |
5097 | */ | |
21c7e756 MX |
5098 | static void btrfs_async_reclaim_metadata_space(struct work_struct *work) |
5099 | { | |
5100 | struct btrfs_fs_info *fs_info; | |
5101 | struct btrfs_space_info *space_info; | |
5102 | u64 to_reclaim; | |
5103 | int flush_state; | |
957780eb | 5104 | int commit_cycles = 0; |
ce129655 | 5105 | u64 last_tickets_id; |
21c7e756 MX |
5106 | |
5107 | fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work); | |
5108 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); | |
5109 | ||
957780eb | 5110 | spin_lock(&space_info->lock); |
c1c4919b JM |
5111 | to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info, |
5112 | false); | |
957780eb JB |
5113 | if (!to_reclaim) { |
5114 | space_info->flush = 0; | |
5115 | spin_unlock(&space_info->lock); | |
21c7e756 | 5116 | return; |
957780eb | 5117 | } |
ce129655 | 5118 | last_tickets_id = space_info->tickets_id; |
957780eb | 5119 | spin_unlock(&space_info->lock); |
21c7e756 MX |
5120 | |
5121 | flush_state = FLUSH_DELAYED_ITEMS_NR; | |
957780eb | 5122 | do { |
e38ae7a0 | 5123 | flush_space(fs_info, space_info, to_reclaim, flush_state); |
957780eb JB |
5124 | spin_lock(&space_info->lock); |
5125 | if (list_empty(&space_info->tickets)) { | |
5126 | space_info->flush = 0; | |
5127 | spin_unlock(&space_info->lock); | |
5128 | return; | |
5129 | } | |
c1c4919b JM |
5130 | to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, |
5131 | space_info, | |
5132 | false); | |
ce129655 | 5133 | if (last_tickets_id == space_info->tickets_id) { |
957780eb JB |
5134 | flush_state++; |
5135 | } else { | |
ce129655 | 5136 | last_tickets_id = space_info->tickets_id; |
957780eb JB |
5137 | flush_state = FLUSH_DELAYED_ITEMS_NR; |
5138 | if (commit_cycles) | |
5139 | commit_cycles--; | |
5140 | } | |
5141 | ||
5142 | if (flush_state > COMMIT_TRANS) { | |
5143 | commit_cycles++; | |
5144 | if (commit_cycles > 2) { | |
5145 | wake_all_tickets(&space_info->tickets); | |
5146 | space_info->flush = 0; | |
5147 | } else { | |
5148 | flush_state = FLUSH_DELAYED_ITEMS_NR; | |
5149 | } | |
5150 | } | |
5151 | spin_unlock(&space_info->lock); | |
5152 | } while (flush_state <= COMMIT_TRANS); | |
5153 | } | |
5154 | ||
5155 | void btrfs_init_async_reclaim_work(struct work_struct *work) | |
5156 | { | |
5157 | INIT_WORK(work, btrfs_async_reclaim_metadata_space); | |
5158 | } | |
5159 | ||
5160 | static void priority_reclaim_metadata_space(struct btrfs_fs_info *fs_info, | |
5161 | struct btrfs_space_info *space_info, | |
5162 | struct reserve_ticket *ticket) | |
5163 | { | |
5164 | u64 to_reclaim; | |
5165 | int flush_state = FLUSH_DELAYED_ITEMS_NR; | |
5166 | ||
5167 | spin_lock(&space_info->lock); | |
c1c4919b JM |
5168 | to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info, |
5169 | false); | |
957780eb JB |
5170 | if (!to_reclaim) { |
5171 | spin_unlock(&space_info->lock); | |
5172 | return; | |
5173 | } | |
5174 | spin_unlock(&space_info->lock); | |
5175 | ||
21c7e756 | 5176 | do { |
7bdd6277 | 5177 | flush_space(fs_info, space_info, to_reclaim, flush_state); |
21c7e756 | 5178 | flush_state++; |
957780eb JB |
5179 | spin_lock(&space_info->lock); |
5180 | if (ticket->bytes == 0) { | |
5181 | spin_unlock(&space_info->lock); | |
21c7e756 | 5182 | return; |
957780eb JB |
5183 | } |
5184 | spin_unlock(&space_info->lock); | |
5185 | ||
5186 | /* | |
5187 | * Priority flushers can't wait on delalloc without | |
5188 | * deadlocking. | |
5189 | */ | |
5190 | if (flush_state == FLUSH_DELALLOC || | |
5191 | flush_state == FLUSH_DELALLOC_WAIT) | |
5192 | flush_state = ALLOC_CHUNK; | |
365c5313 | 5193 | } while (flush_state < COMMIT_TRANS); |
21c7e756 MX |
5194 | } |
5195 | ||
957780eb JB |
5196 | static int wait_reserve_ticket(struct btrfs_fs_info *fs_info, |
5197 | struct btrfs_space_info *space_info, | |
5198 | struct reserve_ticket *ticket, u64 orig_bytes) | |
5199 | ||
21c7e756 | 5200 | { |
957780eb JB |
5201 | DEFINE_WAIT(wait); |
5202 | int ret = 0; | |
5203 | ||
5204 | spin_lock(&space_info->lock); | |
5205 | while (ticket->bytes > 0 && ticket->error == 0) { | |
5206 | ret = prepare_to_wait_event(&ticket->wait, &wait, TASK_KILLABLE); | |
5207 | if (ret) { | |
5208 | ret = -EINTR; | |
5209 | break; | |
5210 | } | |
5211 | spin_unlock(&space_info->lock); | |
5212 | ||
5213 | schedule(); | |
5214 | ||
5215 | finish_wait(&ticket->wait, &wait); | |
5216 | spin_lock(&space_info->lock); | |
5217 | } | |
5218 | if (!ret) | |
5219 | ret = ticket->error; | |
5220 | if (!list_empty(&ticket->list)) | |
5221 | list_del_init(&ticket->list); | |
5222 | if (ticket->bytes && ticket->bytes < orig_bytes) { | |
5223 | u64 num_bytes = orig_bytes - ticket->bytes; | |
5224 | space_info->bytes_may_use -= num_bytes; | |
5225 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5226 | space_info->flags, num_bytes, 0); | |
5227 | } | |
5228 | spin_unlock(&space_info->lock); | |
5229 | ||
5230 | return ret; | |
21c7e756 MX |
5231 | } |
5232 | ||
4a92b1b8 JB |
5233 | /** |
5234 | * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space | |
5235 | * @root - the root we're allocating for | |
957780eb | 5236 | * @space_info - the space info we want to allocate from |
4a92b1b8 | 5237 | * @orig_bytes - the number of bytes we want |
48fc7f7e | 5238 | * @flush - whether or not we can flush to make our reservation |
8bb8ab2e | 5239 | * |
01327610 | 5240 | * This will reserve orig_bytes number of bytes from the space info associated |
4a92b1b8 JB |
5241 | * with the block_rsv. If there is not enough space it will make an attempt to |
5242 | * flush out space to make room. It will do this by flushing delalloc if | |
5243 | * possible or committing the transaction. If flush is 0 then no attempts to | |
5244 | * regain reservations will be made and this will fail if there is not enough | |
5245 | * space already. | |
8bb8ab2e | 5246 | */ |
c1c4919b | 5247 | static int __reserve_metadata_bytes(struct btrfs_fs_info *fs_info, |
957780eb JB |
5248 | struct btrfs_space_info *space_info, |
5249 | u64 orig_bytes, | |
c1c4919b JM |
5250 | enum btrfs_reserve_flush_enum flush, |
5251 | bool system_chunk) | |
9ed74f2d | 5252 | { |
957780eb | 5253 | struct reserve_ticket ticket; |
2bf64758 | 5254 | u64 used; |
8bb8ab2e | 5255 | int ret = 0; |
9ed74f2d | 5256 | |
957780eb | 5257 | ASSERT(orig_bytes); |
8ca17f0f | 5258 | ASSERT(!current->journal_info || flush != BTRFS_RESERVE_FLUSH_ALL); |
fdb5effd | 5259 | |
8bb8ab2e | 5260 | spin_lock(&space_info->lock); |
fdb5effd | 5261 | ret = -ENOSPC; |
4136135b | 5262 | used = btrfs_space_info_used(space_info, true); |
9ed74f2d | 5263 | |
8bb8ab2e | 5264 | /* |
957780eb JB |
5265 | * If we have enough space then hooray, make our reservation and carry |
5266 | * on. If not see if we can overcommit, and if we can, hooray carry on. | |
5267 | * If not things get more complicated. | |
8bb8ab2e | 5268 | */ |
957780eb JB |
5269 | if (used + orig_bytes <= space_info->total_bytes) { |
5270 | space_info->bytes_may_use += orig_bytes; | |
0b246afa JM |
5271 | trace_btrfs_space_reservation(fs_info, "space_info", |
5272 | space_info->flags, orig_bytes, 1); | |
957780eb | 5273 | ret = 0; |
c1c4919b JM |
5274 | } else if (can_overcommit(fs_info, space_info, orig_bytes, flush, |
5275 | system_chunk)) { | |
44734ed1 | 5276 | space_info->bytes_may_use += orig_bytes; |
0b246afa JM |
5277 | trace_btrfs_space_reservation(fs_info, "space_info", |
5278 | space_info->flags, orig_bytes, 1); | |
44734ed1 | 5279 | ret = 0; |
2bf64758 JB |
5280 | } |
5281 | ||
8bb8ab2e | 5282 | /* |
957780eb JB |
5283 | * If we couldn't make a reservation then setup our reservation ticket |
5284 | * and kick the async worker if it's not already running. | |
08e007d2 | 5285 | * |
957780eb JB |
5286 | * If we are a priority flusher then we just need to add our ticket to |
5287 | * the list and we will do our own flushing further down. | |
8bb8ab2e | 5288 | */ |
72bcd99d | 5289 | if (ret && flush != BTRFS_RESERVE_NO_FLUSH) { |
957780eb JB |
5290 | ticket.bytes = orig_bytes; |
5291 | ticket.error = 0; | |
5292 | init_waitqueue_head(&ticket.wait); | |
5293 | if (flush == BTRFS_RESERVE_FLUSH_ALL) { | |
5294 | list_add_tail(&ticket.list, &space_info->tickets); | |
5295 | if (!space_info->flush) { | |
5296 | space_info->flush = 1; | |
0b246afa | 5297 | trace_btrfs_trigger_flush(fs_info, |
f376df2b JB |
5298 | space_info->flags, |
5299 | orig_bytes, flush, | |
5300 | "enospc"); | |
957780eb | 5301 | queue_work(system_unbound_wq, |
c1c4919b | 5302 | &fs_info->async_reclaim_work); |
957780eb JB |
5303 | } |
5304 | } else { | |
5305 | list_add_tail(&ticket.list, | |
5306 | &space_info->priority_tickets); | |
5307 | } | |
21c7e756 MX |
5308 | } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) { |
5309 | used += orig_bytes; | |
f6acfd50 JB |
5310 | /* |
5311 | * We will do the space reservation dance during log replay, | |
5312 | * which means we won't have fs_info->fs_root set, so don't do | |
5313 | * the async reclaim as we will panic. | |
5314 | */ | |
0b246afa | 5315 | if (!test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags) && |
c1c4919b JM |
5316 | need_do_async_reclaim(fs_info, space_info, |
5317 | used, system_chunk) && | |
0b246afa JM |
5318 | !work_busy(&fs_info->async_reclaim_work)) { |
5319 | trace_btrfs_trigger_flush(fs_info, space_info->flags, | |
5320 | orig_bytes, flush, "preempt"); | |
21c7e756 | 5321 | queue_work(system_unbound_wq, |
0b246afa | 5322 | &fs_info->async_reclaim_work); |
f376df2b | 5323 | } |
8bb8ab2e | 5324 | } |
f0486c68 | 5325 | spin_unlock(&space_info->lock); |
08e007d2 | 5326 | if (!ret || flush == BTRFS_RESERVE_NO_FLUSH) |
957780eb | 5327 | return ret; |
f0486c68 | 5328 | |
957780eb | 5329 | if (flush == BTRFS_RESERVE_FLUSH_ALL) |
0b246afa | 5330 | return wait_reserve_ticket(fs_info, space_info, &ticket, |
957780eb | 5331 | orig_bytes); |
08e007d2 | 5332 | |
957780eb | 5333 | ret = 0; |
0b246afa | 5334 | priority_reclaim_metadata_space(fs_info, space_info, &ticket); |
957780eb JB |
5335 | spin_lock(&space_info->lock); |
5336 | if (ticket.bytes) { | |
5337 | if (ticket.bytes < orig_bytes) { | |
5338 | u64 num_bytes = orig_bytes - ticket.bytes; | |
5339 | space_info->bytes_may_use -= num_bytes; | |
0b246afa JM |
5340 | trace_btrfs_space_reservation(fs_info, "space_info", |
5341 | space_info->flags, | |
5342 | num_bytes, 0); | |
08e007d2 | 5343 | |
957780eb JB |
5344 | } |
5345 | list_del_init(&ticket.list); | |
5346 | ret = -ENOSPC; | |
5347 | } | |
5348 | spin_unlock(&space_info->lock); | |
5349 | ASSERT(list_empty(&ticket.list)); | |
5350 | return ret; | |
5351 | } | |
8bb8ab2e | 5352 | |
957780eb JB |
5353 | /** |
5354 | * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space | |
5355 | * @root - the root we're allocating for | |
5356 | * @block_rsv - the block_rsv we're allocating for | |
5357 | * @orig_bytes - the number of bytes we want | |
5358 | * @flush - whether or not we can flush to make our reservation | |
5359 | * | |
5360 | * This will reserve orgi_bytes number of bytes from the space info associated | |
5361 | * with the block_rsv. If there is not enough space it will make an attempt to | |
5362 | * flush out space to make room. It will do this by flushing delalloc if | |
5363 | * possible or committing the transaction. If flush is 0 then no attempts to | |
5364 | * regain reservations will be made and this will fail if there is not enough | |
5365 | * space already. | |
5366 | */ | |
5367 | static int reserve_metadata_bytes(struct btrfs_root *root, | |
5368 | struct btrfs_block_rsv *block_rsv, | |
5369 | u64 orig_bytes, | |
5370 | enum btrfs_reserve_flush_enum flush) | |
5371 | { | |
0b246afa JM |
5372 | struct btrfs_fs_info *fs_info = root->fs_info; |
5373 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
957780eb | 5374 | int ret; |
c1c4919b | 5375 | bool system_chunk = (root == fs_info->chunk_root); |
957780eb | 5376 | |
c1c4919b JM |
5377 | ret = __reserve_metadata_bytes(fs_info, block_rsv->space_info, |
5378 | orig_bytes, flush, system_chunk); | |
5d80366e JB |
5379 | if (ret == -ENOSPC && |
5380 | unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) { | |
5d80366e JB |
5381 | if (block_rsv != global_rsv && |
5382 | !block_rsv_use_bytes(global_rsv, orig_bytes)) | |
5383 | ret = 0; | |
5384 | } | |
9a3daff3 | 5385 | if (ret == -ENOSPC) { |
0b246afa | 5386 | trace_btrfs_space_reservation(fs_info, "space_info:enospc", |
957780eb JB |
5387 | block_rsv->space_info->flags, |
5388 | orig_bytes, 1); | |
9a3daff3 NB |
5389 | |
5390 | if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) | |
5391 | dump_space_info(fs_info, block_rsv->space_info, | |
5392 | orig_bytes, 0); | |
5393 | } | |
f0486c68 YZ |
5394 | return ret; |
5395 | } | |
5396 | ||
79787eaa JM |
5397 | static struct btrfs_block_rsv *get_block_rsv( |
5398 | const struct btrfs_trans_handle *trans, | |
5399 | const struct btrfs_root *root) | |
f0486c68 | 5400 | { |
0b246afa | 5401 | struct btrfs_fs_info *fs_info = root->fs_info; |
4c13d758 JB |
5402 | struct btrfs_block_rsv *block_rsv = NULL; |
5403 | ||
e9cf439f | 5404 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) || |
0b246afa JM |
5405 | (root == fs_info->csum_root && trans->adding_csums) || |
5406 | (root == fs_info->uuid_root)) | |
f7a81ea4 SB |
5407 | block_rsv = trans->block_rsv; |
5408 | ||
4c13d758 | 5409 | if (!block_rsv) |
f0486c68 YZ |
5410 | block_rsv = root->block_rsv; |
5411 | ||
5412 | if (!block_rsv) | |
0b246afa | 5413 | block_rsv = &fs_info->empty_block_rsv; |
f0486c68 YZ |
5414 | |
5415 | return block_rsv; | |
5416 | } | |
5417 | ||
5418 | static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, | |
5419 | u64 num_bytes) | |
5420 | { | |
5421 | int ret = -ENOSPC; | |
5422 | spin_lock(&block_rsv->lock); | |
5423 | if (block_rsv->reserved >= num_bytes) { | |
5424 | block_rsv->reserved -= num_bytes; | |
5425 | if (block_rsv->reserved < block_rsv->size) | |
5426 | block_rsv->full = 0; | |
5427 | ret = 0; | |
5428 | } | |
5429 | spin_unlock(&block_rsv->lock); | |
5430 | return ret; | |
5431 | } | |
5432 | ||
5433 | static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv, | |
5434 | u64 num_bytes, int update_size) | |
5435 | { | |
5436 | spin_lock(&block_rsv->lock); | |
5437 | block_rsv->reserved += num_bytes; | |
5438 | if (update_size) | |
5439 | block_rsv->size += num_bytes; | |
5440 | else if (block_rsv->reserved >= block_rsv->size) | |
5441 | block_rsv->full = 1; | |
5442 | spin_unlock(&block_rsv->lock); | |
5443 | } | |
5444 | ||
d52be818 JB |
5445 | int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info, |
5446 | struct btrfs_block_rsv *dest, u64 num_bytes, | |
5447 | int min_factor) | |
5448 | { | |
5449 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
5450 | u64 min_bytes; | |
5451 | ||
5452 | if (global_rsv->space_info != dest->space_info) | |
5453 | return -ENOSPC; | |
5454 | ||
5455 | spin_lock(&global_rsv->lock); | |
5456 | min_bytes = div_factor(global_rsv->size, min_factor); | |
5457 | if (global_rsv->reserved < min_bytes + num_bytes) { | |
5458 | spin_unlock(&global_rsv->lock); | |
5459 | return -ENOSPC; | |
5460 | } | |
5461 | global_rsv->reserved -= num_bytes; | |
5462 | if (global_rsv->reserved < global_rsv->size) | |
5463 | global_rsv->full = 0; | |
5464 | spin_unlock(&global_rsv->lock); | |
5465 | ||
5466 | block_rsv_add_bytes(dest, num_bytes, 1); | |
5467 | return 0; | |
5468 | } | |
5469 | ||
957780eb JB |
5470 | /* |
5471 | * This is for space we already have accounted in space_info->bytes_may_use, so | |
5472 | * basically when we're returning space from block_rsv's. | |
5473 | */ | |
5474 | static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info, | |
5475 | struct btrfs_space_info *space_info, | |
5476 | u64 num_bytes) | |
5477 | { | |
5478 | struct reserve_ticket *ticket; | |
5479 | struct list_head *head; | |
5480 | u64 used; | |
5481 | enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_NO_FLUSH; | |
5482 | bool check_overcommit = false; | |
5483 | ||
5484 | spin_lock(&space_info->lock); | |
5485 | head = &space_info->priority_tickets; | |
5486 | ||
5487 | /* | |
5488 | * If we are over our limit then we need to check and see if we can | |
5489 | * overcommit, and if we can't then we just need to free up our space | |
5490 | * and not satisfy any requests. | |
5491 | */ | |
0eee8a49 | 5492 | used = btrfs_space_info_used(space_info, true); |
957780eb JB |
5493 | if (used - num_bytes >= space_info->total_bytes) |
5494 | check_overcommit = true; | |
5495 | again: | |
5496 | while (!list_empty(head) && num_bytes) { | |
5497 | ticket = list_first_entry(head, struct reserve_ticket, | |
5498 | list); | |
5499 | /* | |
5500 | * We use 0 bytes because this space is already reserved, so | |
5501 | * adding the ticket space would be a double count. | |
5502 | */ | |
5503 | if (check_overcommit && | |
c1c4919b | 5504 | !can_overcommit(fs_info, space_info, 0, flush, false)) |
957780eb JB |
5505 | break; |
5506 | if (num_bytes >= ticket->bytes) { | |
5507 | list_del_init(&ticket->list); | |
5508 | num_bytes -= ticket->bytes; | |
5509 | ticket->bytes = 0; | |
ce129655 | 5510 | space_info->tickets_id++; |
957780eb JB |
5511 | wake_up(&ticket->wait); |
5512 | } else { | |
5513 | ticket->bytes -= num_bytes; | |
5514 | num_bytes = 0; | |
5515 | } | |
5516 | } | |
5517 | ||
5518 | if (num_bytes && head == &space_info->priority_tickets) { | |
5519 | head = &space_info->tickets; | |
5520 | flush = BTRFS_RESERVE_FLUSH_ALL; | |
5521 | goto again; | |
5522 | } | |
5523 | space_info->bytes_may_use -= num_bytes; | |
5524 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5525 | space_info->flags, num_bytes, 0); | |
5526 | spin_unlock(&space_info->lock); | |
5527 | } | |
5528 | ||
5529 | /* | |
5530 | * This is for newly allocated space that isn't accounted in | |
5531 | * space_info->bytes_may_use yet. So if we allocate a chunk or unpin an extent | |
5532 | * we use this helper. | |
5533 | */ | |
5534 | static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info, | |
5535 | struct btrfs_space_info *space_info, | |
5536 | u64 num_bytes) | |
5537 | { | |
5538 | struct reserve_ticket *ticket; | |
5539 | struct list_head *head = &space_info->priority_tickets; | |
5540 | ||
5541 | again: | |
5542 | while (!list_empty(head) && num_bytes) { | |
5543 | ticket = list_first_entry(head, struct reserve_ticket, | |
5544 | list); | |
5545 | if (num_bytes >= ticket->bytes) { | |
5546 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5547 | space_info->flags, | |
5548 | ticket->bytes, 1); | |
5549 | list_del_init(&ticket->list); | |
5550 | num_bytes -= ticket->bytes; | |
5551 | space_info->bytes_may_use += ticket->bytes; | |
5552 | ticket->bytes = 0; | |
ce129655 | 5553 | space_info->tickets_id++; |
957780eb JB |
5554 | wake_up(&ticket->wait); |
5555 | } else { | |
5556 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5557 | space_info->flags, | |
5558 | num_bytes, 1); | |
5559 | space_info->bytes_may_use += num_bytes; | |
5560 | ticket->bytes -= num_bytes; | |
5561 | num_bytes = 0; | |
5562 | } | |
5563 | } | |
5564 | ||
5565 | if (num_bytes && head == &space_info->priority_tickets) { | |
5566 | head = &space_info->tickets; | |
5567 | goto again; | |
5568 | } | |
5569 | } | |
5570 | ||
69fe2d75 | 5571 | static u64 block_rsv_release_bytes(struct btrfs_fs_info *fs_info, |
8c2a3ca2 | 5572 | struct btrfs_block_rsv *block_rsv, |
ff6bc37e QW |
5573 | struct btrfs_block_rsv *dest, u64 num_bytes, |
5574 | u64 *qgroup_to_release_ret) | |
f0486c68 YZ |
5575 | { |
5576 | struct btrfs_space_info *space_info = block_rsv->space_info; | |
ff6bc37e | 5577 | u64 qgroup_to_release = 0; |
69fe2d75 | 5578 | u64 ret; |
f0486c68 YZ |
5579 | |
5580 | spin_lock(&block_rsv->lock); | |
ff6bc37e | 5581 | if (num_bytes == (u64)-1) { |
f0486c68 | 5582 | num_bytes = block_rsv->size; |
ff6bc37e QW |
5583 | qgroup_to_release = block_rsv->qgroup_rsv_size; |
5584 | } | |
f0486c68 YZ |
5585 | block_rsv->size -= num_bytes; |
5586 | if (block_rsv->reserved >= block_rsv->size) { | |
5587 | num_bytes = block_rsv->reserved - block_rsv->size; | |
5588 | block_rsv->reserved = block_rsv->size; | |
5589 | block_rsv->full = 1; | |
5590 | } else { | |
5591 | num_bytes = 0; | |
5592 | } | |
ff6bc37e QW |
5593 | if (block_rsv->qgroup_rsv_reserved >= block_rsv->qgroup_rsv_size) { |
5594 | qgroup_to_release = block_rsv->qgroup_rsv_reserved - | |
5595 | block_rsv->qgroup_rsv_size; | |
5596 | block_rsv->qgroup_rsv_reserved = block_rsv->qgroup_rsv_size; | |
5597 | } else { | |
5598 | qgroup_to_release = 0; | |
5599 | } | |
f0486c68 YZ |
5600 | spin_unlock(&block_rsv->lock); |
5601 | ||
69fe2d75 | 5602 | ret = num_bytes; |
f0486c68 YZ |
5603 | if (num_bytes > 0) { |
5604 | if (dest) { | |
e9e22899 JB |
5605 | spin_lock(&dest->lock); |
5606 | if (!dest->full) { | |
5607 | u64 bytes_to_add; | |
5608 | ||
5609 | bytes_to_add = dest->size - dest->reserved; | |
5610 | bytes_to_add = min(num_bytes, bytes_to_add); | |
5611 | dest->reserved += bytes_to_add; | |
5612 | if (dest->reserved >= dest->size) | |
5613 | dest->full = 1; | |
5614 | num_bytes -= bytes_to_add; | |
5615 | } | |
5616 | spin_unlock(&dest->lock); | |
5617 | } | |
957780eb JB |
5618 | if (num_bytes) |
5619 | space_info_add_old_bytes(fs_info, space_info, | |
5620 | num_bytes); | |
9ed74f2d | 5621 | } |
ff6bc37e QW |
5622 | if (qgroup_to_release_ret) |
5623 | *qgroup_to_release_ret = qgroup_to_release; | |
69fe2d75 | 5624 | return ret; |
f0486c68 | 5625 | } |
4e06bdd6 | 5626 | |
25d609f8 JB |
5627 | int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src, |
5628 | struct btrfs_block_rsv *dst, u64 num_bytes, | |
5629 | int update_size) | |
f0486c68 YZ |
5630 | { |
5631 | int ret; | |
9ed74f2d | 5632 | |
f0486c68 YZ |
5633 | ret = block_rsv_use_bytes(src, num_bytes); |
5634 | if (ret) | |
5635 | return ret; | |
9ed74f2d | 5636 | |
25d609f8 | 5637 | block_rsv_add_bytes(dst, num_bytes, update_size); |
9ed74f2d JB |
5638 | return 0; |
5639 | } | |
5640 | ||
66d8f3dd | 5641 | void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type) |
9ed74f2d | 5642 | { |
f0486c68 YZ |
5643 | memset(rsv, 0, sizeof(*rsv)); |
5644 | spin_lock_init(&rsv->lock); | |
66d8f3dd | 5645 | rsv->type = type; |
f0486c68 YZ |
5646 | } |
5647 | ||
69fe2d75 JB |
5648 | void btrfs_init_metadata_block_rsv(struct btrfs_fs_info *fs_info, |
5649 | struct btrfs_block_rsv *rsv, | |
5650 | unsigned short type) | |
5651 | { | |
5652 | btrfs_init_block_rsv(rsv, type); | |
5653 | rsv->space_info = __find_space_info(fs_info, | |
5654 | BTRFS_BLOCK_GROUP_METADATA); | |
5655 | } | |
5656 | ||
2ff7e61e | 5657 | struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_fs_info *fs_info, |
66d8f3dd | 5658 | unsigned short type) |
f0486c68 YZ |
5659 | { |
5660 | struct btrfs_block_rsv *block_rsv; | |
9ed74f2d | 5661 | |
f0486c68 YZ |
5662 | block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS); |
5663 | if (!block_rsv) | |
5664 | return NULL; | |
9ed74f2d | 5665 | |
69fe2d75 | 5666 | btrfs_init_metadata_block_rsv(fs_info, block_rsv, type); |
f0486c68 YZ |
5667 | return block_rsv; |
5668 | } | |
9ed74f2d | 5669 | |
2ff7e61e | 5670 | void btrfs_free_block_rsv(struct btrfs_fs_info *fs_info, |
f0486c68 YZ |
5671 | struct btrfs_block_rsv *rsv) |
5672 | { | |
2aaa6655 JB |
5673 | if (!rsv) |
5674 | return; | |
2ff7e61e | 5675 | btrfs_block_rsv_release(fs_info, rsv, (u64)-1); |
dabdb640 | 5676 | kfree(rsv); |
9ed74f2d JB |
5677 | } |
5678 | ||
cdfb080e CM |
5679 | void __btrfs_free_block_rsv(struct btrfs_block_rsv *rsv) |
5680 | { | |
5681 | kfree(rsv); | |
5682 | } | |
5683 | ||
08e007d2 MX |
5684 | int btrfs_block_rsv_add(struct btrfs_root *root, |
5685 | struct btrfs_block_rsv *block_rsv, u64 num_bytes, | |
5686 | enum btrfs_reserve_flush_enum flush) | |
9ed74f2d | 5687 | { |
f0486c68 | 5688 | int ret; |
9ed74f2d | 5689 | |
f0486c68 YZ |
5690 | if (num_bytes == 0) |
5691 | return 0; | |
8bb8ab2e | 5692 | |
61b520a9 | 5693 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
f0486c68 YZ |
5694 | if (!ret) { |
5695 | block_rsv_add_bytes(block_rsv, num_bytes, 1); | |
5696 | return 0; | |
5697 | } | |
9ed74f2d | 5698 | |
f0486c68 | 5699 | return ret; |
f0486c68 | 5700 | } |
9ed74f2d | 5701 | |
2ff7e61e | 5702 | int btrfs_block_rsv_check(struct btrfs_block_rsv *block_rsv, int min_factor) |
f0486c68 YZ |
5703 | { |
5704 | u64 num_bytes = 0; | |
f0486c68 | 5705 | int ret = -ENOSPC; |
9ed74f2d | 5706 | |
f0486c68 YZ |
5707 | if (!block_rsv) |
5708 | return 0; | |
9ed74f2d | 5709 | |
f0486c68 | 5710 | spin_lock(&block_rsv->lock); |
36ba022a JB |
5711 | num_bytes = div_factor(block_rsv->size, min_factor); |
5712 | if (block_rsv->reserved >= num_bytes) | |
5713 | ret = 0; | |
5714 | spin_unlock(&block_rsv->lock); | |
9ed74f2d | 5715 | |
36ba022a JB |
5716 | return ret; |
5717 | } | |
5718 | ||
08e007d2 MX |
5719 | int btrfs_block_rsv_refill(struct btrfs_root *root, |
5720 | struct btrfs_block_rsv *block_rsv, u64 min_reserved, | |
5721 | enum btrfs_reserve_flush_enum flush) | |
36ba022a JB |
5722 | { |
5723 | u64 num_bytes = 0; | |
5724 | int ret = -ENOSPC; | |
5725 | ||
5726 | if (!block_rsv) | |
5727 | return 0; | |
5728 | ||
5729 | spin_lock(&block_rsv->lock); | |
5730 | num_bytes = min_reserved; | |
13553e52 | 5731 | if (block_rsv->reserved >= num_bytes) |
f0486c68 | 5732 | ret = 0; |
13553e52 | 5733 | else |
f0486c68 | 5734 | num_bytes -= block_rsv->reserved; |
f0486c68 | 5735 | spin_unlock(&block_rsv->lock); |
13553e52 | 5736 | |
f0486c68 YZ |
5737 | if (!ret) |
5738 | return 0; | |
5739 | ||
aa38a711 | 5740 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
dabdb640 JB |
5741 | if (!ret) { |
5742 | block_rsv_add_bytes(block_rsv, num_bytes, 0); | |
f0486c68 | 5743 | return 0; |
6a63209f | 5744 | } |
9ed74f2d | 5745 | |
13553e52 | 5746 | return ret; |
f0486c68 YZ |
5747 | } |
5748 | ||
69fe2d75 JB |
5749 | /** |
5750 | * btrfs_inode_rsv_refill - refill the inode block rsv. | |
5751 | * @inode - the inode we are refilling. | |
5752 | * @flush - the flusing restriction. | |
5753 | * | |
5754 | * Essentially the same as btrfs_block_rsv_refill, except it uses the | |
5755 | * block_rsv->size as the minimum size. We'll either refill the missing amount | |
5756 | * or return if we already have enough space. This will also handle the resreve | |
5757 | * tracepoint for the reserved amount. | |
5758 | */ | |
3f2dd7a0 QW |
5759 | static int btrfs_inode_rsv_refill(struct btrfs_inode *inode, |
5760 | enum btrfs_reserve_flush_enum flush) | |
69fe2d75 JB |
5761 | { |
5762 | struct btrfs_root *root = inode->root; | |
5763 | struct btrfs_block_rsv *block_rsv = &inode->block_rsv; | |
5764 | u64 num_bytes = 0; | |
ff6bc37e | 5765 | u64 qgroup_num_bytes = 0; |
69fe2d75 JB |
5766 | int ret = -ENOSPC; |
5767 | ||
5768 | spin_lock(&block_rsv->lock); | |
5769 | if (block_rsv->reserved < block_rsv->size) | |
5770 | num_bytes = block_rsv->size - block_rsv->reserved; | |
ff6bc37e QW |
5771 | if (block_rsv->qgroup_rsv_reserved < block_rsv->qgroup_rsv_size) |
5772 | qgroup_num_bytes = block_rsv->qgroup_rsv_size - | |
5773 | block_rsv->qgroup_rsv_reserved; | |
69fe2d75 JB |
5774 | spin_unlock(&block_rsv->lock); |
5775 | ||
5776 | if (num_bytes == 0) | |
5777 | return 0; | |
5778 | ||
ff6bc37e | 5779 | ret = btrfs_qgroup_reserve_meta_prealloc(root, qgroup_num_bytes, true); |
43b18595 QW |
5780 | if (ret) |
5781 | return ret; | |
69fe2d75 JB |
5782 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
5783 | if (!ret) { | |
5784 | block_rsv_add_bytes(block_rsv, num_bytes, 0); | |
5785 | trace_btrfs_space_reservation(root->fs_info, "delalloc", | |
5786 | btrfs_ino(inode), num_bytes, 1); | |
ff6bc37e QW |
5787 | |
5788 | /* Don't forget to increase qgroup_rsv_reserved */ | |
5789 | spin_lock(&block_rsv->lock); | |
5790 | block_rsv->qgroup_rsv_reserved += qgroup_num_bytes; | |
5791 | spin_unlock(&block_rsv->lock); | |
5792 | } else | |
5793 | btrfs_qgroup_free_meta_prealloc(root, qgroup_num_bytes); | |
69fe2d75 JB |
5794 | return ret; |
5795 | } | |
5796 | ||
5797 | /** | |
5798 | * btrfs_inode_rsv_release - release any excessive reservation. | |
5799 | * @inode - the inode we need to release from. | |
43b18595 QW |
5800 | * @qgroup_free - free or convert qgroup meta. |
5801 | * Unlike normal operation, qgroup meta reservation needs to know if we are | |
5802 | * freeing qgroup reservation or just converting it into per-trans. Normally | |
5803 | * @qgroup_free is true for error handling, and false for normal release. | |
69fe2d75 JB |
5804 | * |
5805 | * This is the same as btrfs_block_rsv_release, except that it handles the | |
5806 | * tracepoint for the reservation. | |
5807 | */ | |
43b18595 | 5808 | static void btrfs_inode_rsv_release(struct btrfs_inode *inode, bool qgroup_free) |
69fe2d75 JB |
5809 | { |
5810 | struct btrfs_fs_info *fs_info = inode->root->fs_info; | |
5811 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
5812 | struct btrfs_block_rsv *block_rsv = &inode->block_rsv; | |
5813 | u64 released = 0; | |
ff6bc37e | 5814 | u64 qgroup_to_release = 0; |
69fe2d75 JB |
5815 | |
5816 | /* | |
5817 | * Since we statically set the block_rsv->size we just want to say we | |
5818 | * are releasing 0 bytes, and then we'll just get the reservation over | |
5819 | * the size free'd. | |
5820 | */ | |
ff6bc37e QW |
5821 | released = block_rsv_release_bytes(fs_info, block_rsv, global_rsv, 0, |
5822 | &qgroup_to_release); | |
69fe2d75 JB |
5823 | if (released > 0) |
5824 | trace_btrfs_space_reservation(fs_info, "delalloc", | |
5825 | btrfs_ino(inode), released, 0); | |
43b18595 | 5826 | if (qgroup_free) |
ff6bc37e | 5827 | btrfs_qgroup_free_meta_prealloc(inode->root, qgroup_to_release); |
43b18595 | 5828 | else |
ff6bc37e QW |
5829 | btrfs_qgroup_convert_reserved_meta(inode->root, |
5830 | qgroup_to_release); | |
69fe2d75 JB |
5831 | } |
5832 | ||
2ff7e61e | 5833 | void btrfs_block_rsv_release(struct btrfs_fs_info *fs_info, |
f0486c68 YZ |
5834 | struct btrfs_block_rsv *block_rsv, |
5835 | u64 num_bytes) | |
5836 | { | |
0b246afa JM |
5837 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; |
5838 | ||
17504584 | 5839 | if (global_rsv == block_rsv || |
f0486c68 YZ |
5840 | block_rsv->space_info != global_rsv->space_info) |
5841 | global_rsv = NULL; | |
ff6bc37e | 5842 | block_rsv_release_bytes(fs_info, block_rsv, global_rsv, num_bytes, NULL); |
6a63209f JB |
5843 | } |
5844 | ||
8929ecfa YZ |
5845 | static void update_global_block_rsv(struct btrfs_fs_info *fs_info) |
5846 | { | |
5847 | struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv; | |
5848 | struct btrfs_space_info *sinfo = block_rsv->space_info; | |
5849 | u64 num_bytes; | |
6a63209f | 5850 | |
ae2e4728 JB |
5851 | /* |
5852 | * The global block rsv is based on the size of the extent tree, the | |
5853 | * checksum tree and the root tree. If the fs is empty we want to set | |
5854 | * it to a minimal amount for safety. | |
5855 | */ | |
5856 | num_bytes = btrfs_root_used(&fs_info->extent_root->root_item) + | |
5857 | btrfs_root_used(&fs_info->csum_root->root_item) + | |
5858 | btrfs_root_used(&fs_info->tree_root->root_item); | |
5859 | num_bytes = max_t(u64, num_bytes, SZ_16M); | |
33b4d47f | 5860 | |
8929ecfa | 5861 | spin_lock(&sinfo->lock); |
1f699d38 | 5862 | spin_lock(&block_rsv->lock); |
4e06bdd6 | 5863 | |
ee22184b | 5864 | block_rsv->size = min_t(u64, num_bytes, SZ_512M); |
4e06bdd6 | 5865 | |
fb4b10e5 | 5866 | if (block_rsv->reserved < block_rsv->size) { |
4136135b | 5867 | num_bytes = btrfs_space_info_used(sinfo, true); |
fb4b10e5 JB |
5868 | if (sinfo->total_bytes > num_bytes) { |
5869 | num_bytes = sinfo->total_bytes - num_bytes; | |
5870 | num_bytes = min(num_bytes, | |
5871 | block_rsv->size - block_rsv->reserved); | |
5872 | block_rsv->reserved += num_bytes; | |
5873 | sinfo->bytes_may_use += num_bytes; | |
5874 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5875 | sinfo->flags, num_bytes, | |
5876 | 1); | |
5877 | } | |
5878 | } else if (block_rsv->reserved > block_rsv->size) { | |
8929ecfa | 5879 | num_bytes = block_rsv->reserved - block_rsv->size; |
fb25e914 | 5880 | sinfo->bytes_may_use -= num_bytes; |
8c2a3ca2 | 5881 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 5882 | sinfo->flags, num_bytes, 0); |
8929ecfa | 5883 | block_rsv->reserved = block_rsv->size; |
8929ecfa | 5884 | } |
182608c8 | 5885 | |
fb4b10e5 JB |
5886 | if (block_rsv->reserved == block_rsv->size) |
5887 | block_rsv->full = 1; | |
5888 | else | |
5889 | block_rsv->full = 0; | |
5890 | ||
8929ecfa | 5891 | spin_unlock(&block_rsv->lock); |
1f699d38 | 5892 | spin_unlock(&sinfo->lock); |
6a63209f JB |
5893 | } |
5894 | ||
f0486c68 | 5895 | static void init_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 5896 | { |
f0486c68 | 5897 | struct btrfs_space_info *space_info; |
6a63209f | 5898 | |
f0486c68 YZ |
5899 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); |
5900 | fs_info->chunk_block_rsv.space_info = space_info; | |
6a63209f | 5901 | |
f0486c68 | 5902 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
8929ecfa | 5903 | fs_info->global_block_rsv.space_info = space_info; |
f0486c68 YZ |
5904 | fs_info->trans_block_rsv.space_info = space_info; |
5905 | fs_info->empty_block_rsv.space_info = space_info; | |
6d668dda | 5906 | fs_info->delayed_block_rsv.space_info = space_info; |
f0486c68 | 5907 | |
8929ecfa YZ |
5908 | fs_info->extent_root->block_rsv = &fs_info->global_block_rsv; |
5909 | fs_info->csum_root->block_rsv = &fs_info->global_block_rsv; | |
5910 | fs_info->dev_root->block_rsv = &fs_info->global_block_rsv; | |
5911 | fs_info->tree_root->block_rsv = &fs_info->global_block_rsv; | |
3a6cad90 SB |
5912 | if (fs_info->quota_root) |
5913 | fs_info->quota_root->block_rsv = &fs_info->global_block_rsv; | |
f0486c68 | 5914 | fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv; |
8929ecfa | 5915 | |
8929ecfa | 5916 | update_global_block_rsv(fs_info); |
6a63209f JB |
5917 | } |
5918 | ||
8929ecfa | 5919 | static void release_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 5920 | { |
8c2a3ca2 | 5921 | block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL, |
ff6bc37e | 5922 | (u64)-1, NULL); |
8929ecfa YZ |
5923 | WARN_ON(fs_info->trans_block_rsv.size > 0); |
5924 | WARN_ON(fs_info->trans_block_rsv.reserved > 0); | |
5925 | WARN_ON(fs_info->chunk_block_rsv.size > 0); | |
5926 | WARN_ON(fs_info->chunk_block_rsv.reserved > 0); | |
6d668dda JB |
5927 | WARN_ON(fs_info->delayed_block_rsv.size > 0); |
5928 | WARN_ON(fs_info->delayed_block_rsv.reserved > 0); | |
fcb80c2a JB |
5929 | } |
5930 | ||
6a63209f | 5931 | |
4fbcdf66 FM |
5932 | /* |
5933 | * To be called after all the new block groups attached to the transaction | |
5934 | * handle have been created (btrfs_create_pending_block_groups()). | |
5935 | */ | |
5936 | void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans) | |
5937 | { | |
64b63580 | 5938 | struct btrfs_fs_info *fs_info = trans->fs_info; |
4fbcdf66 FM |
5939 | |
5940 | if (!trans->chunk_bytes_reserved) | |
5941 | return; | |
5942 | ||
5943 | WARN_ON_ONCE(!list_empty(&trans->new_bgs)); | |
5944 | ||
5945 | block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL, | |
ff6bc37e | 5946 | trans->chunk_bytes_reserved, NULL); |
4fbcdf66 FM |
5947 | trans->chunk_bytes_reserved = 0; |
5948 | } | |
5949 | ||
d5c12070 MX |
5950 | /* |
5951 | * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation | |
5952 | * root: the root of the parent directory | |
5953 | * rsv: block reservation | |
5954 | * items: the number of items that we need do reservation | |
5955 | * qgroup_reserved: used to return the reserved size in qgroup | |
5956 | * | |
5957 | * This function is used to reserve the space for snapshot/subvolume | |
5958 | * creation and deletion. Those operations are different with the | |
5959 | * common file/directory operations, they change two fs/file trees | |
5960 | * and root tree, the number of items that the qgroup reserves is | |
5961 | * different with the free space reservation. So we can not use | |
01327610 | 5962 | * the space reservation mechanism in start_transaction(). |
d5c12070 MX |
5963 | */ |
5964 | int btrfs_subvolume_reserve_metadata(struct btrfs_root *root, | |
5965 | struct btrfs_block_rsv *rsv, | |
5966 | int items, | |
ee3441b4 | 5967 | bool use_global_rsv) |
a22285a6 | 5968 | { |
d5c12070 MX |
5969 | u64 num_bytes; |
5970 | int ret; | |
0b246afa JM |
5971 | struct btrfs_fs_info *fs_info = root->fs_info; |
5972 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
d5c12070 | 5973 | |
0b246afa | 5974 | if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) { |
d5c12070 | 5975 | /* One for parent inode, two for dir entries */ |
0b246afa | 5976 | num_bytes = 3 * fs_info->nodesize; |
733e03a0 | 5977 | ret = btrfs_qgroup_reserve_meta_prealloc(root, num_bytes, true); |
d5c12070 MX |
5978 | if (ret) |
5979 | return ret; | |
5980 | } else { | |
5981 | num_bytes = 0; | |
5982 | } | |
5983 | ||
0b246afa JM |
5984 | num_bytes = btrfs_calc_trans_metadata_size(fs_info, items); |
5985 | rsv->space_info = __find_space_info(fs_info, | |
d5c12070 MX |
5986 | BTRFS_BLOCK_GROUP_METADATA); |
5987 | ret = btrfs_block_rsv_add(root, rsv, num_bytes, | |
5988 | BTRFS_RESERVE_FLUSH_ALL); | |
ee3441b4 JM |
5989 | |
5990 | if (ret == -ENOSPC && use_global_rsv) | |
25d609f8 | 5991 | ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes, 1); |
ee3441b4 | 5992 | |
c4c129db GJ |
5993 | if (ret && num_bytes) |
5994 | btrfs_qgroup_free_meta_prealloc(root, num_bytes); | |
d5c12070 MX |
5995 | |
5996 | return ret; | |
5997 | } | |
5998 | ||
2ff7e61e | 5999 | void btrfs_subvolume_release_metadata(struct btrfs_fs_info *fs_info, |
7775c818 | 6000 | struct btrfs_block_rsv *rsv) |
d5c12070 | 6001 | { |
2ff7e61e | 6002 | btrfs_block_rsv_release(fs_info, rsv, (u64)-1); |
97e728d4 JB |
6003 | } |
6004 | ||
69fe2d75 JB |
6005 | static void btrfs_calculate_inode_block_rsv_size(struct btrfs_fs_info *fs_info, |
6006 | struct btrfs_inode *inode) | |
9e0baf60 | 6007 | { |
69fe2d75 JB |
6008 | struct btrfs_block_rsv *block_rsv = &inode->block_rsv; |
6009 | u64 reserve_size = 0; | |
ff6bc37e | 6010 | u64 qgroup_rsv_size = 0; |
69fe2d75 JB |
6011 | u64 csum_leaves; |
6012 | unsigned outstanding_extents; | |
9e0baf60 | 6013 | |
69fe2d75 JB |
6014 | lockdep_assert_held(&inode->lock); |
6015 | outstanding_extents = inode->outstanding_extents; | |
6016 | if (outstanding_extents) | |
6017 | reserve_size = btrfs_calc_trans_metadata_size(fs_info, | |
6018 | outstanding_extents + 1); | |
6019 | csum_leaves = btrfs_csum_bytes_to_leaves(fs_info, | |
6020 | inode->csum_bytes); | |
6021 | reserve_size += btrfs_calc_trans_metadata_size(fs_info, | |
6022 | csum_leaves); | |
ff6bc37e QW |
6023 | /* |
6024 | * For qgroup rsv, the calculation is very simple: | |
6025 | * account one nodesize for each outstanding extent | |
6026 | * | |
6027 | * This is overestimating in most cases. | |
6028 | */ | |
6029 | qgroup_rsv_size = outstanding_extents * fs_info->nodesize; | |
9e0baf60 | 6030 | |
69fe2d75 JB |
6031 | spin_lock(&block_rsv->lock); |
6032 | block_rsv->size = reserve_size; | |
ff6bc37e | 6033 | block_rsv->qgroup_rsv_size = qgroup_rsv_size; |
69fe2d75 | 6034 | spin_unlock(&block_rsv->lock); |
0ca1f7ce | 6035 | } |
c146afad | 6036 | |
9f3db423 | 6037 | int btrfs_delalloc_reserve_metadata(struct btrfs_inode *inode, u64 num_bytes) |
0ca1f7ce | 6038 | { |
9f3db423 | 6039 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
69fe2d75 | 6040 | unsigned nr_extents; |
08e007d2 | 6041 | enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL; |
eb6b88d9 | 6042 | int ret = 0; |
c64c2bd8 | 6043 | bool delalloc_lock = true; |
6324fbf3 | 6044 | |
c64c2bd8 JB |
6045 | /* If we are a free space inode we need to not flush since we will be in |
6046 | * the middle of a transaction commit. We also don't need the delalloc | |
6047 | * mutex since we won't race with anybody. We need this mostly to make | |
6048 | * lockdep shut its filthy mouth. | |
bac357dc JB |
6049 | * |
6050 | * If we have a transaction open (can happen if we call truncate_block | |
6051 | * from truncate), then we need FLUSH_LIMIT so we don't deadlock. | |
c64c2bd8 JB |
6052 | */ |
6053 | if (btrfs_is_free_space_inode(inode)) { | |
08e007d2 | 6054 | flush = BTRFS_RESERVE_NO_FLUSH; |
c64c2bd8 | 6055 | delalloc_lock = false; |
da07d4ab NB |
6056 | } else { |
6057 | if (current->journal_info) | |
6058 | flush = BTRFS_RESERVE_FLUSH_LIMIT; | |
c09544e0 | 6059 | |
da07d4ab NB |
6060 | if (btrfs_transaction_in_commit(fs_info)) |
6061 | schedule_timeout(1); | |
6062 | } | |
ec44a35c | 6063 | |
c64c2bd8 | 6064 | if (delalloc_lock) |
9f3db423 | 6065 | mutex_lock(&inode->delalloc_mutex); |
c64c2bd8 | 6066 | |
0b246afa | 6067 | num_bytes = ALIGN(num_bytes, fs_info->sectorsize); |
69fe2d75 JB |
6068 | |
6069 | /* Add our new extents and calculate the new rsv size. */ | |
9f3db423 | 6070 | spin_lock(&inode->lock); |
69fe2d75 | 6071 | nr_extents = count_max_extents(num_bytes); |
8b62f87b | 6072 | btrfs_mod_outstanding_extents(inode, nr_extents); |
69fe2d75 JB |
6073 | inode->csum_bytes += num_bytes; |
6074 | btrfs_calculate_inode_block_rsv_size(fs_info, inode); | |
9f3db423 | 6075 | spin_unlock(&inode->lock); |
57a45ced | 6076 | |
69fe2d75 | 6077 | ret = btrfs_inode_rsv_refill(inode, flush); |
43b18595 | 6078 | if (unlikely(ret)) |
88e081bf | 6079 | goto out_fail; |
25179201 | 6080 | |
c64c2bd8 | 6081 | if (delalloc_lock) |
9f3db423 | 6082 | mutex_unlock(&inode->delalloc_mutex); |
0ca1f7ce | 6083 | return 0; |
88e081bf WS |
6084 | |
6085 | out_fail: | |
9f3db423 | 6086 | spin_lock(&inode->lock); |
8b62f87b JB |
6087 | nr_extents = count_max_extents(num_bytes); |
6088 | btrfs_mod_outstanding_extents(inode, -nr_extents); | |
69fe2d75 JB |
6089 | inode->csum_bytes -= num_bytes; |
6090 | btrfs_calculate_inode_block_rsv_size(fs_info, inode); | |
9f3db423 | 6091 | spin_unlock(&inode->lock); |
88e081bf | 6092 | |
43b18595 | 6093 | btrfs_inode_rsv_release(inode, true); |
88e081bf | 6094 | if (delalloc_lock) |
9f3db423 | 6095 | mutex_unlock(&inode->delalloc_mutex); |
88e081bf | 6096 | return ret; |
0ca1f7ce YZ |
6097 | } |
6098 | ||
7709cde3 JB |
6099 | /** |
6100 | * btrfs_delalloc_release_metadata - release a metadata reservation for an inode | |
8b62f87b JB |
6101 | * @inode: the inode to release the reservation for. |
6102 | * @num_bytes: the number of bytes we are releasing. | |
43b18595 | 6103 | * @qgroup_free: free qgroup reservation or convert it to per-trans reservation |
7709cde3 JB |
6104 | * |
6105 | * This will release the metadata reservation for an inode. This can be called | |
6106 | * once we complete IO for a given set of bytes to release their metadata | |
8b62f87b | 6107 | * reservations, or on error for the same reason. |
7709cde3 | 6108 | */ |
43b18595 QW |
6109 | void btrfs_delalloc_release_metadata(struct btrfs_inode *inode, u64 num_bytes, |
6110 | bool qgroup_free) | |
0ca1f7ce | 6111 | { |
691fa059 | 6112 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
0ca1f7ce | 6113 | |
0b246afa | 6114 | num_bytes = ALIGN(num_bytes, fs_info->sectorsize); |
691fa059 | 6115 | spin_lock(&inode->lock); |
69fe2d75 JB |
6116 | inode->csum_bytes -= num_bytes; |
6117 | btrfs_calculate_inode_block_rsv_size(fs_info, inode); | |
691fa059 | 6118 | spin_unlock(&inode->lock); |
0ca1f7ce | 6119 | |
0b246afa | 6120 | if (btrfs_is_testing(fs_info)) |
6a3891c5 JB |
6121 | return; |
6122 | ||
43b18595 | 6123 | btrfs_inode_rsv_release(inode, qgroup_free); |
0ca1f7ce YZ |
6124 | } |
6125 | ||
8b62f87b JB |
6126 | /** |
6127 | * btrfs_delalloc_release_extents - release our outstanding_extents | |
6128 | * @inode: the inode to balance the reservation for. | |
6129 | * @num_bytes: the number of bytes we originally reserved with | |
43b18595 | 6130 | * @qgroup_free: do we need to free qgroup meta reservation or convert them. |
8b62f87b JB |
6131 | * |
6132 | * When we reserve space we increase outstanding_extents for the extents we may | |
6133 | * add. Once we've set the range as delalloc or created our ordered extents we | |
6134 | * have outstanding_extents to track the real usage, so we use this to free our | |
6135 | * temporarily tracked outstanding_extents. This _must_ be used in conjunction | |
6136 | * with btrfs_delalloc_reserve_metadata. | |
6137 | */ | |
43b18595 QW |
6138 | void btrfs_delalloc_release_extents(struct btrfs_inode *inode, u64 num_bytes, |
6139 | bool qgroup_free) | |
8b62f87b JB |
6140 | { |
6141 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); | |
6142 | unsigned num_extents; | |
8b62f87b JB |
6143 | |
6144 | spin_lock(&inode->lock); | |
6145 | num_extents = count_max_extents(num_bytes); | |
6146 | btrfs_mod_outstanding_extents(inode, -num_extents); | |
69fe2d75 | 6147 | btrfs_calculate_inode_block_rsv_size(fs_info, inode); |
8b62f87b JB |
6148 | spin_unlock(&inode->lock); |
6149 | ||
8b62f87b JB |
6150 | if (btrfs_is_testing(fs_info)) |
6151 | return; | |
6152 | ||
43b18595 | 6153 | btrfs_inode_rsv_release(inode, qgroup_free); |
8b62f87b JB |
6154 | } |
6155 | ||
1ada3a62 | 6156 | /** |
7cf5b976 | 6157 | * btrfs_delalloc_reserve_space - reserve data and metadata space for |
1ada3a62 QW |
6158 | * delalloc |
6159 | * @inode: inode we're writing to | |
6160 | * @start: start range we are writing to | |
6161 | * @len: how long the range we are writing to | |
364ecf36 QW |
6162 | * @reserved: mandatory parameter, record actually reserved qgroup ranges of |
6163 | * current reservation. | |
1ada3a62 | 6164 | * |
1ada3a62 QW |
6165 | * This will do the following things |
6166 | * | |
6167 | * o reserve space in data space info for num bytes | |
6168 | * and reserve precious corresponding qgroup space | |
6169 | * (Done in check_data_free_space) | |
6170 | * | |
6171 | * o reserve space for metadata space, based on the number of outstanding | |
6172 | * extents and how much csums will be needed | |
6173 | * also reserve metadata space in a per root over-reserve method. | |
6174 | * o add to the inodes->delalloc_bytes | |
6175 | * o add it to the fs_info's delalloc inodes list. | |
6176 | * (Above 3 all done in delalloc_reserve_metadata) | |
6177 | * | |
6178 | * Return 0 for success | |
6179 | * Return <0 for error(-ENOSPC or -EQUOT) | |
6180 | */ | |
364ecf36 QW |
6181 | int btrfs_delalloc_reserve_space(struct inode *inode, |
6182 | struct extent_changeset **reserved, u64 start, u64 len) | |
1ada3a62 QW |
6183 | { |
6184 | int ret; | |
6185 | ||
364ecf36 | 6186 | ret = btrfs_check_data_free_space(inode, reserved, start, len); |
1ada3a62 QW |
6187 | if (ret < 0) |
6188 | return ret; | |
9f3db423 | 6189 | ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len); |
1ada3a62 | 6190 | if (ret < 0) |
bc42bda2 | 6191 | btrfs_free_reserved_data_space(inode, *reserved, start, len); |
1ada3a62 QW |
6192 | return ret; |
6193 | } | |
6194 | ||
7709cde3 | 6195 | /** |
7cf5b976 | 6196 | * btrfs_delalloc_release_space - release data and metadata space for delalloc |
1ada3a62 QW |
6197 | * @inode: inode we're releasing space for |
6198 | * @start: start position of the space already reserved | |
6199 | * @len: the len of the space already reserved | |
8b62f87b | 6200 | * @release_bytes: the len of the space we consumed or didn't use |
1ada3a62 QW |
6201 | * |
6202 | * This function will release the metadata space that was not used and will | |
6203 | * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes | |
6204 | * list if there are no delalloc bytes left. | |
6205 | * Also it will handle the qgroup reserved space. | |
6206 | */ | |
bc42bda2 | 6207 | void btrfs_delalloc_release_space(struct inode *inode, |
8b62f87b | 6208 | struct extent_changeset *reserved, |
43b18595 | 6209 | u64 start, u64 len, bool qgroup_free) |
1ada3a62 | 6210 | { |
43b18595 | 6211 | btrfs_delalloc_release_metadata(BTRFS_I(inode), len, qgroup_free); |
bc42bda2 | 6212 | btrfs_free_reserved_data_space(inode, reserved, start, len); |
6324fbf3 CM |
6213 | } |
6214 | ||
ce93ec54 | 6215 | static int update_block_group(struct btrfs_trans_handle *trans, |
6202df69 | 6216 | struct btrfs_fs_info *info, u64 bytenr, |
ce93ec54 | 6217 | u64 num_bytes, int alloc) |
9078a3e1 | 6218 | { |
0af3d00b | 6219 | struct btrfs_block_group_cache *cache = NULL; |
db94535d | 6220 | u64 total = num_bytes; |
9078a3e1 | 6221 | u64 old_val; |
db94535d | 6222 | u64 byte_in_group; |
0af3d00b | 6223 | int factor; |
3e1ad54f | 6224 | |
5d4f98a2 | 6225 | /* block accounting for super block */ |
eb73c1b7 | 6226 | spin_lock(&info->delalloc_root_lock); |
6c41761f | 6227 | old_val = btrfs_super_bytes_used(info->super_copy); |
5d4f98a2 YZ |
6228 | if (alloc) |
6229 | old_val += num_bytes; | |
6230 | else | |
6231 | old_val -= num_bytes; | |
6c41761f | 6232 | btrfs_set_super_bytes_used(info->super_copy, old_val); |
eb73c1b7 | 6233 | spin_unlock(&info->delalloc_root_lock); |
5d4f98a2 | 6234 | |
d397712b | 6235 | while (total) { |
db94535d | 6236 | cache = btrfs_lookup_block_group(info, bytenr); |
f3465ca4 | 6237 | if (!cache) |
79787eaa | 6238 | return -ENOENT; |
b742bb82 YZ |
6239 | if (cache->flags & (BTRFS_BLOCK_GROUP_DUP | |
6240 | BTRFS_BLOCK_GROUP_RAID1 | | |
6241 | BTRFS_BLOCK_GROUP_RAID10)) | |
6242 | factor = 2; | |
6243 | else | |
6244 | factor = 1; | |
9d66e233 JB |
6245 | /* |
6246 | * If this block group has free space cache written out, we | |
6247 | * need to make sure to load it if we are removing space. This | |
6248 | * is because we need the unpinning stage to actually add the | |
6249 | * space back to the block group, otherwise we will leak space. | |
6250 | */ | |
6251 | if (!alloc && cache->cached == BTRFS_CACHE_NO) | |
f6373bf3 | 6252 | cache_block_group(cache, 1); |
0af3d00b | 6253 | |
db94535d CM |
6254 | byte_in_group = bytenr - cache->key.objectid; |
6255 | WARN_ON(byte_in_group > cache->key.offset); | |
9078a3e1 | 6256 | |
25179201 | 6257 | spin_lock(&cache->space_info->lock); |
c286ac48 | 6258 | spin_lock(&cache->lock); |
0af3d00b | 6259 | |
6202df69 | 6260 | if (btrfs_test_opt(info, SPACE_CACHE) && |
0af3d00b JB |
6261 | cache->disk_cache_state < BTRFS_DC_CLEAR) |
6262 | cache->disk_cache_state = BTRFS_DC_CLEAR; | |
6263 | ||
9078a3e1 | 6264 | old_val = btrfs_block_group_used(&cache->item); |
db94535d | 6265 | num_bytes = min(total, cache->key.offset - byte_in_group); |
cd1bc465 | 6266 | if (alloc) { |
db94535d | 6267 | old_val += num_bytes; |
11833d66 YZ |
6268 | btrfs_set_block_group_used(&cache->item, old_val); |
6269 | cache->reserved -= num_bytes; | |
11833d66 | 6270 | cache->space_info->bytes_reserved -= num_bytes; |
b742bb82 YZ |
6271 | cache->space_info->bytes_used += num_bytes; |
6272 | cache->space_info->disk_used += num_bytes * factor; | |
c286ac48 | 6273 | spin_unlock(&cache->lock); |
25179201 | 6274 | spin_unlock(&cache->space_info->lock); |
cd1bc465 | 6275 | } else { |
db94535d | 6276 | old_val -= num_bytes; |
ae0ab003 FM |
6277 | btrfs_set_block_group_used(&cache->item, old_val); |
6278 | cache->pinned += num_bytes; | |
6279 | cache->space_info->bytes_pinned += num_bytes; | |
6280 | cache->space_info->bytes_used -= num_bytes; | |
6281 | cache->space_info->disk_used -= num_bytes * factor; | |
6282 | spin_unlock(&cache->lock); | |
6283 | spin_unlock(&cache->space_info->lock); | |
47ab2a6c | 6284 | |
0b246afa | 6285 | trace_btrfs_space_reservation(info, "pinned", |
c51e7bb1 JB |
6286 | cache->space_info->flags, |
6287 | num_bytes, 1); | |
d7eae340 OS |
6288 | percpu_counter_add(&cache->space_info->total_bytes_pinned, |
6289 | num_bytes); | |
ae0ab003 FM |
6290 | set_extent_dirty(info->pinned_extents, |
6291 | bytenr, bytenr + num_bytes - 1, | |
6292 | GFP_NOFS | __GFP_NOFAIL); | |
cd1bc465 | 6293 | } |
1bbc621e CM |
6294 | |
6295 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
6296 | if (list_empty(&cache->dirty_list)) { | |
6297 | list_add_tail(&cache->dirty_list, | |
6298 | &trans->transaction->dirty_bgs); | |
bece2e82 | 6299 | trans->transaction->num_dirty_bgs++; |
1bbc621e CM |
6300 | btrfs_get_block_group(cache); |
6301 | } | |
6302 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
6303 | ||
036a9348 FM |
6304 | /* |
6305 | * No longer have used bytes in this block group, queue it for | |
6306 | * deletion. We do this after adding the block group to the | |
6307 | * dirty list to avoid races between cleaner kthread and space | |
6308 | * cache writeout. | |
6309 | */ | |
6310 | if (!alloc && old_val == 0) { | |
6311 | spin_lock(&info->unused_bgs_lock); | |
6312 | if (list_empty(&cache->bg_list)) { | |
6313 | btrfs_get_block_group(cache); | |
4ed0a7a3 | 6314 | trace_btrfs_add_unused_block_group(cache); |
036a9348 FM |
6315 | list_add_tail(&cache->bg_list, |
6316 | &info->unused_bgs); | |
6317 | } | |
6318 | spin_unlock(&info->unused_bgs_lock); | |
6319 | } | |
6320 | ||
fa9c0d79 | 6321 | btrfs_put_block_group(cache); |
db94535d CM |
6322 | total -= num_bytes; |
6323 | bytenr += num_bytes; | |
9078a3e1 CM |
6324 | } |
6325 | return 0; | |
6326 | } | |
6324fbf3 | 6327 | |
2ff7e61e | 6328 | static u64 first_logical_byte(struct btrfs_fs_info *fs_info, u64 search_start) |
a061fc8d | 6329 | { |
0f9dd46c | 6330 | struct btrfs_block_group_cache *cache; |
d2fb3437 | 6331 | u64 bytenr; |
0f9dd46c | 6332 | |
0b246afa JM |
6333 | spin_lock(&fs_info->block_group_cache_lock); |
6334 | bytenr = fs_info->first_logical_byte; | |
6335 | spin_unlock(&fs_info->block_group_cache_lock); | |
a1897fdd LB |
6336 | |
6337 | if (bytenr < (u64)-1) | |
6338 | return bytenr; | |
6339 | ||
0b246afa | 6340 | cache = btrfs_lookup_first_block_group(fs_info, search_start); |
0f9dd46c | 6341 | if (!cache) |
a061fc8d | 6342 | return 0; |
0f9dd46c | 6343 | |
d2fb3437 | 6344 | bytenr = cache->key.objectid; |
fa9c0d79 | 6345 | btrfs_put_block_group(cache); |
d2fb3437 YZ |
6346 | |
6347 | return bytenr; | |
a061fc8d CM |
6348 | } |
6349 | ||
2ff7e61e | 6350 | static int pin_down_extent(struct btrfs_fs_info *fs_info, |
f0486c68 YZ |
6351 | struct btrfs_block_group_cache *cache, |
6352 | u64 bytenr, u64 num_bytes, int reserved) | |
324ae4df | 6353 | { |
11833d66 YZ |
6354 | spin_lock(&cache->space_info->lock); |
6355 | spin_lock(&cache->lock); | |
6356 | cache->pinned += num_bytes; | |
6357 | cache->space_info->bytes_pinned += num_bytes; | |
6358 | if (reserved) { | |
6359 | cache->reserved -= num_bytes; | |
6360 | cache->space_info->bytes_reserved -= num_bytes; | |
6361 | } | |
6362 | spin_unlock(&cache->lock); | |
6363 | spin_unlock(&cache->space_info->lock); | |
68b38550 | 6364 | |
0b246afa | 6365 | trace_btrfs_space_reservation(fs_info, "pinned", |
c51e7bb1 | 6366 | cache->space_info->flags, num_bytes, 1); |
4da8b76d | 6367 | percpu_counter_add(&cache->space_info->total_bytes_pinned, num_bytes); |
0b246afa | 6368 | set_extent_dirty(fs_info->pinned_extents, bytenr, |
f0486c68 YZ |
6369 | bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL); |
6370 | return 0; | |
6371 | } | |
68b38550 | 6372 | |
f0486c68 YZ |
6373 | /* |
6374 | * this function must be called within transaction | |
6375 | */ | |
2ff7e61e | 6376 | int btrfs_pin_extent(struct btrfs_fs_info *fs_info, |
f0486c68 YZ |
6377 | u64 bytenr, u64 num_bytes, int reserved) |
6378 | { | |
6379 | struct btrfs_block_group_cache *cache; | |
68b38550 | 6380 | |
0b246afa | 6381 | cache = btrfs_lookup_block_group(fs_info, bytenr); |
79787eaa | 6382 | BUG_ON(!cache); /* Logic error */ |
f0486c68 | 6383 | |
2ff7e61e | 6384 | pin_down_extent(fs_info, cache, bytenr, num_bytes, reserved); |
f0486c68 YZ |
6385 | |
6386 | btrfs_put_block_group(cache); | |
11833d66 YZ |
6387 | return 0; |
6388 | } | |
6389 | ||
f0486c68 | 6390 | /* |
e688b725 CM |
6391 | * this function must be called within transaction |
6392 | */ | |
2ff7e61e | 6393 | int btrfs_pin_extent_for_log_replay(struct btrfs_fs_info *fs_info, |
e688b725 CM |
6394 | u64 bytenr, u64 num_bytes) |
6395 | { | |
6396 | struct btrfs_block_group_cache *cache; | |
b50c6e25 | 6397 | int ret; |
e688b725 | 6398 | |
0b246afa | 6399 | cache = btrfs_lookup_block_group(fs_info, bytenr); |
b50c6e25 JB |
6400 | if (!cache) |
6401 | return -EINVAL; | |
e688b725 CM |
6402 | |
6403 | /* | |
6404 | * pull in the free space cache (if any) so that our pin | |
6405 | * removes the free space from the cache. We have load_only set | |
6406 | * to one because the slow code to read in the free extents does check | |
6407 | * the pinned extents. | |
6408 | */ | |
f6373bf3 | 6409 | cache_block_group(cache, 1); |
e688b725 | 6410 | |
2ff7e61e | 6411 | pin_down_extent(fs_info, cache, bytenr, num_bytes, 0); |
e688b725 CM |
6412 | |
6413 | /* remove us from the free space cache (if we're there at all) */ | |
b50c6e25 | 6414 | ret = btrfs_remove_free_space(cache, bytenr, num_bytes); |
e688b725 | 6415 | btrfs_put_block_group(cache); |
b50c6e25 | 6416 | return ret; |
e688b725 CM |
6417 | } |
6418 | ||
2ff7e61e JM |
6419 | static int __exclude_logged_extent(struct btrfs_fs_info *fs_info, |
6420 | u64 start, u64 num_bytes) | |
8c2a1a30 JB |
6421 | { |
6422 | int ret; | |
6423 | struct btrfs_block_group_cache *block_group; | |
6424 | struct btrfs_caching_control *caching_ctl; | |
6425 | ||
0b246afa | 6426 | block_group = btrfs_lookup_block_group(fs_info, start); |
8c2a1a30 JB |
6427 | if (!block_group) |
6428 | return -EINVAL; | |
6429 | ||
6430 | cache_block_group(block_group, 0); | |
6431 | caching_ctl = get_caching_control(block_group); | |
6432 | ||
6433 | if (!caching_ctl) { | |
6434 | /* Logic error */ | |
6435 | BUG_ON(!block_group_cache_done(block_group)); | |
6436 | ret = btrfs_remove_free_space(block_group, start, num_bytes); | |
6437 | } else { | |
6438 | mutex_lock(&caching_ctl->mutex); | |
6439 | ||
6440 | if (start >= caching_ctl->progress) { | |
2ff7e61e | 6441 | ret = add_excluded_extent(fs_info, start, num_bytes); |
8c2a1a30 JB |
6442 | } else if (start + num_bytes <= caching_ctl->progress) { |
6443 | ret = btrfs_remove_free_space(block_group, | |
6444 | start, num_bytes); | |
6445 | } else { | |
6446 | num_bytes = caching_ctl->progress - start; | |
6447 | ret = btrfs_remove_free_space(block_group, | |
6448 | start, num_bytes); | |
6449 | if (ret) | |
6450 | goto out_lock; | |
6451 | ||
6452 | num_bytes = (start + num_bytes) - | |
6453 | caching_ctl->progress; | |
6454 | start = caching_ctl->progress; | |
2ff7e61e | 6455 | ret = add_excluded_extent(fs_info, start, num_bytes); |
8c2a1a30 JB |
6456 | } |
6457 | out_lock: | |
6458 | mutex_unlock(&caching_ctl->mutex); | |
6459 | put_caching_control(caching_ctl); | |
6460 | } | |
6461 | btrfs_put_block_group(block_group); | |
6462 | return ret; | |
6463 | } | |
6464 | ||
2ff7e61e | 6465 | int btrfs_exclude_logged_extents(struct btrfs_fs_info *fs_info, |
8c2a1a30 JB |
6466 | struct extent_buffer *eb) |
6467 | { | |
6468 | struct btrfs_file_extent_item *item; | |
6469 | struct btrfs_key key; | |
6470 | int found_type; | |
6471 | int i; | |
b89311ef | 6472 | int ret = 0; |
8c2a1a30 | 6473 | |
2ff7e61e | 6474 | if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) |
8c2a1a30 JB |
6475 | return 0; |
6476 | ||
6477 | for (i = 0; i < btrfs_header_nritems(eb); i++) { | |
6478 | btrfs_item_key_to_cpu(eb, &key, i); | |
6479 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
6480 | continue; | |
6481 | item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item); | |
6482 | found_type = btrfs_file_extent_type(eb, item); | |
6483 | if (found_type == BTRFS_FILE_EXTENT_INLINE) | |
6484 | continue; | |
6485 | if (btrfs_file_extent_disk_bytenr(eb, item) == 0) | |
6486 | continue; | |
6487 | key.objectid = btrfs_file_extent_disk_bytenr(eb, item); | |
6488 | key.offset = btrfs_file_extent_disk_num_bytes(eb, item); | |
b89311ef GJ |
6489 | ret = __exclude_logged_extent(fs_info, key.objectid, key.offset); |
6490 | if (ret) | |
6491 | break; | |
8c2a1a30 JB |
6492 | } |
6493 | ||
b89311ef | 6494 | return ret; |
8c2a1a30 JB |
6495 | } |
6496 | ||
9cfa3e34 FM |
6497 | static void |
6498 | btrfs_inc_block_group_reservations(struct btrfs_block_group_cache *bg) | |
6499 | { | |
6500 | atomic_inc(&bg->reservations); | |
6501 | } | |
6502 | ||
6503 | void btrfs_dec_block_group_reservations(struct btrfs_fs_info *fs_info, | |
6504 | const u64 start) | |
6505 | { | |
6506 | struct btrfs_block_group_cache *bg; | |
6507 | ||
6508 | bg = btrfs_lookup_block_group(fs_info, start); | |
6509 | ASSERT(bg); | |
6510 | if (atomic_dec_and_test(&bg->reservations)) | |
4625956a | 6511 | wake_up_var(&bg->reservations); |
9cfa3e34 FM |
6512 | btrfs_put_block_group(bg); |
6513 | } | |
6514 | ||
9cfa3e34 FM |
6515 | void btrfs_wait_block_group_reservations(struct btrfs_block_group_cache *bg) |
6516 | { | |
6517 | struct btrfs_space_info *space_info = bg->space_info; | |
6518 | ||
6519 | ASSERT(bg->ro); | |
6520 | ||
6521 | if (!(bg->flags & BTRFS_BLOCK_GROUP_DATA)) | |
6522 | return; | |
6523 | ||
6524 | /* | |
6525 | * Our block group is read only but before we set it to read only, | |
6526 | * some task might have had allocated an extent from it already, but it | |
6527 | * has not yet created a respective ordered extent (and added it to a | |
6528 | * root's list of ordered extents). | |
6529 | * Therefore wait for any task currently allocating extents, since the | |
6530 | * block group's reservations counter is incremented while a read lock | |
6531 | * on the groups' semaphore is held and decremented after releasing | |
6532 | * the read access on that semaphore and creating the ordered extent. | |
6533 | */ | |
6534 | down_write(&space_info->groups_sem); | |
6535 | up_write(&space_info->groups_sem); | |
6536 | ||
4625956a | 6537 | wait_var_event(&bg->reservations, !atomic_read(&bg->reservations)); |
9cfa3e34 FM |
6538 | } |
6539 | ||
fb25e914 | 6540 | /** |
4824f1f4 | 6541 | * btrfs_add_reserved_bytes - update the block_group and space info counters |
fb25e914 | 6542 | * @cache: The cache we are manipulating |
18513091 WX |
6543 | * @ram_bytes: The number of bytes of file content, and will be same to |
6544 | * @num_bytes except for the compress path. | |
fb25e914 | 6545 | * @num_bytes: The number of bytes in question |
e570fd27 | 6546 | * @delalloc: The blocks are allocated for the delalloc write |
fb25e914 | 6547 | * |
745699ef XW |
6548 | * This is called by the allocator when it reserves space. If this is a |
6549 | * reservation and the block group has become read only we cannot make the | |
6550 | * reservation and return -EAGAIN, otherwise this function always succeeds. | |
f0486c68 | 6551 | */ |
4824f1f4 | 6552 | static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache *cache, |
18513091 | 6553 | u64 ram_bytes, u64 num_bytes, int delalloc) |
11833d66 | 6554 | { |
fb25e914 | 6555 | struct btrfs_space_info *space_info = cache->space_info; |
f0486c68 | 6556 | int ret = 0; |
79787eaa | 6557 | |
fb25e914 JB |
6558 | spin_lock(&space_info->lock); |
6559 | spin_lock(&cache->lock); | |
4824f1f4 WX |
6560 | if (cache->ro) { |
6561 | ret = -EAGAIN; | |
fb25e914 | 6562 | } else { |
4824f1f4 WX |
6563 | cache->reserved += num_bytes; |
6564 | space_info->bytes_reserved += num_bytes; | |
e570fd27 | 6565 | |
18513091 WX |
6566 | trace_btrfs_space_reservation(cache->fs_info, |
6567 | "space_info", space_info->flags, | |
6568 | ram_bytes, 0); | |
6569 | space_info->bytes_may_use -= ram_bytes; | |
e570fd27 | 6570 | if (delalloc) |
4824f1f4 | 6571 | cache->delalloc_bytes += num_bytes; |
324ae4df | 6572 | } |
fb25e914 JB |
6573 | spin_unlock(&cache->lock); |
6574 | spin_unlock(&space_info->lock); | |
f0486c68 | 6575 | return ret; |
324ae4df | 6576 | } |
9078a3e1 | 6577 | |
4824f1f4 WX |
6578 | /** |
6579 | * btrfs_free_reserved_bytes - update the block_group and space info counters | |
6580 | * @cache: The cache we are manipulating | |
6581 | * @num_bytes: The number of bytes in question | |
6582 | * @delalloc: The blocks are allocated for the delalloc write | |
6583 | * | |
6584 | * This is called by somebody who is freeing space that was never actually used | |
6585 | * on disk. For example if you reserve some space for a new leaf in transaction | |
6586 | * A and before transaction A commits you free that leaf, you call this with | |
6587 | * reserve set to 0 in order to clear the reservation. | |
6588 | */ | |
6589 | ||
6590 | static int btrfs_free_reserved_bytes(struct btrfs_block_group_cache *cache, | |
6591 | u64 num_bytes, int delalloc) | |
6592 | { | |
6593 | struct btrfs_space_info *space_info = cache->space_info; | |
6594 | int ret = 0; | |
6595 | ||
6596 | spin_lock(&space_info->lock); | |
6597 | spin_lock(&cache->lock); | |
6598 | if (cache->ro) | |
6599 | space_info->bytes_readonly += num_bytes; | |
6600 | cache->reserved -= num_bytes; | |
6601 | space_info->bytes_reserved -= num_bytes; | |
6602 | ||
6603 | if (delalloc) | |
6604 | cache->delalloc_bytes -= num_bytes; | |
6605 | spin_unlock(&cache->lock); | |
6606 | spin_unlock(&space_info->lock); | |
6607 | return ret; | |
6608 | } | |
8b74c03e | 6609 | void btrfs_prepare_extent_commit(struct btrfs_fs_info *fs_info) |
e8569813 | 6610 | { |
11833d66 YZ |
6611 | struct btrfs_caching_control *next; |
6612 | struct btrfs_caching_control *caching_ctl; | |
6613 | struct btrfs_block_group_cache *cache; | |
e8569813 | 6614 | |
9e351cc8 | 6615 | down_write(&fs_info->commit_root_sem); |
25179201 | 6616 | |
11833d66 YZ |
6617 | list_for_each_entry_safe(caching_ctl, next, |
6618 | &fs_info->caching_block_groups, list) { | |
6619 | cache = caching_ctl->block_group; | |
6620 | if (block_group_cache_done(cache)) { | |
6621 | cache->last_byte_to_unpin = (u64)-1; | |
6622 | list_del_init(&caching_ctl->list); | |
6623 | put_caching_control(caching_ctl); | |
e8569813 | 6624 | } else { |
11833d66 | 6625 | cache->last_byte_to_unpin = caching_ctl->progress; |
e8569813 | 6626 | } |
e8569813 | 6627 | } |
11833d66 YZ |
6628 | |
6629 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) | |
6630 | fs_info->pinned_extents = &fs_info->freed_extents[1]; | |
6631 | else | |
6632 | fs_info->pinned_extents = &fs_info->freed_extents[0]; | |
6633 | ||
9e351cc8 | 6634 | up_write(&fs_info->commit_root_sem); |
8929ecfa YZ |
6635 | |
6636 | update_global_block_rsv(fs_info); | |
e8569813 ZY |
6637 | } |
6638 | ||
c759c4e1 JB |
6639 | /* |
6640 | * Returns the free cluster for the given space info and sets empty_cluster to | |
6641 | * what it should be based on the mount options. | |
6642 | */ | |
6643 | static struct btrfs_free_cluster * | |
2ff7e61e JM |
6644 | fetch_cluster_info(struct btrfs_fs_info *fs_info, |
6645 | struct btrfs_space_info *space_info, u64 *empty_cluster) | |
c759c4e1 JB |
6646 | { |
6647 | struct btrfs_free_cluster *ret = NULL; | |
c759c4e1 JB |
6648 | |
6649 | *empty_cluster = 0; | |
6650 | if (btrfs_mixed_space_info(space_info)) | |
6651 | return ret; | |
6652 | ||
c759c4e1 | 6653 | if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) { |
0b246afa | 6654 | ret = &fs_info->meta_alloc_cluster; |
583b7231 HK |
6655 | if (btrfs_test_opt(fs_info, SSD)) |
6656 | *empty_cluster = SZ_2M; | |
6657 | else | |
ee22184b | 6658 | *empty_cluster = SZ_64K; |
583b7231 HK |
6659 | } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) && |
6660 | btrfs_test_opt(fs_info, SSD_SPREAD)) { | |
6661 | *empty_cluster = SZ_2M; | |
0b246afa | 6662 | ret = &fs_info->data_alloc_cluster; |
c759c4e1 JB |
6663 | } |
6664 | ||
6665 | return ret; | |
6666 | } | |
6667 | ||
2ff7e61e JM |
6668 | static int unpin_extent_range(struct btrfs_fs_info *fs_info, |
6669 | u64 start, u64 end, | |
678886bd | 6670 | const bool return_free_space) |
ccd467d6 | 6671 | { |
11833d66 | 6672 | struct btrfs_block_group_cache *cache = NULL; |
7b398f8e JB |
6673 | struct btrfs_space_info *space_info; |
6674 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
c759c4e1 | 6675 | struct btrfs_free_cluster *cluster = NULL; |
11833d66 | 6676 | u64 len; |
c759c4e1 JB |
6677 | u64 total_unpinned = 0; |
6678 | u64 empty_cluster = 0; | |
7b398f8e | 6679 | bool readonly; |
ccd467d6 | 6680 | |
11833d66 | 6681 | while (start <= end) { |
7b398f8e | 6682 | readonly = false; |
11833d66 YZ |
6683 | if (!cache || |
6684 | start >= cache->key.objectid + cache->key.offset) { | |
6685 | if (cache) | |
6686 | btrfs_put_block_group(cache); | |
c759c4e1 | 6687 | total_unpinned = 0; |
11833d66 | 6688 | cache = btrfs_lookup_block_group(fs_info, start); |
79787eaa | 6689 | BUG_ON(!cache); /* Logic error */ |
c759c4e1 | 6690 | |
2ff7e61e | 6691 | cluster = fetch_cluster_info(fs_info, |
c759c4e1 JB |
6692 | cache->space_info, |
6693 | &empty_cluster); | |
6694 | empty_cluster <<= 1; | |
11833d66 YZ |
6695 | } |
6696 | ||
6697 | len = cache->key.objectid + cache->key.offset - start; | |
6698 | len = min(len, end + 1 - start); | |
6699 | ||
6700 | if (start < cache->last_byte_to_unpin) { | |
6701 | len = min(len, cache->last_byte_to_unpin - start); | |
678886bd FM |
6702 | if (return_free_space) |
6703 | btrfs_add_free_space(cache, start, len); | |
11833d66 YZ |
6704 | } |
6705 | ||
f0486c68 | 6706 | start += len; |
c759c4e1 | 6707 | total_unpinned += len; |
7b398f8e | 6708 | space_info = cache->space_info; |
f0486c68 | 6709 | |
c759c4e1 JB |
6710 | /* |
6711 | * If this space cluster has been marked as fragmented and we've | |
6712 | * unpinned enough in this block group to potentially allow a | |
6713 | * cluster to be created inside of it go ahead and clear the | |
6714 | * fragmented check. | |
6715 | */ | |
6716 | if (cluster && cluster->fragmented && | |
6717 | total_unpinned > empty_cluster) { | |
6718 | spin_lock(&cluster->lock); | |
6719 | cluster->fragmented = 0; | |
6720 | spin_unlock(&cluster->lock); | |
6721 | } | |
6722 | ||
7b398f8e | 6723 | spin_lock(&space_info->lock); |
11833d66 YZ |
6724 | spin_lock(&cache->lock); |
6725 | cache->pinned -= len; | |
7b398f8e | 6726 | space_info->bytes_pinned -= len; |
c51e7bb1 JB |
6727 | |
6728 | trace_btrfs_space_reservation(fs_info, "pinned", | |
6729 | space_info->flags, len, 0); | |
4f4db217 | 6730 | space_info->max_extent_size = 0; |
d288db5d | 6731 | percpu_counter_add(&space_info->total_bytes_pinned, -len); |
7b398f8e JB |
6732 | if (cache->ro) { |
6733 | space_info->bytes_readonly += len; | |
6734 | readonly = true; | |
6735 | } | |
11833d66 | 6736 | spin_unlock(&cache->lock); |
957780eb JB |
6737 | if (!readonly && return_free_space && |
6738 | global_rsv->space_info == space_info) { | |
6739 | u64 to_add = len; | |
92ac58ec | 6740 | |
7b398f8e JB |
6741 | spin_lock(&global_rsv->lock); |
6742 | if (!global_rsv->full) { | |
957780eb JB |
6743 | to_add = min(len, global_rsv->size - |
6744 | global_rsv->reserved); | |
6745 | global_rsv->reserved += to_add; | |
6746 | space_info->bytes_may_use += to_add; | |
7b398f8e JB |
6747 | if (global_rsv->reserved >= global_rsv->size) |
6748 | global_rsv->full = 1; | |
957780eb JB |
6749 | trace_btrfs_space_reservation(fs_info, |
6750 | "space_info", | |
6751 | space_info->flags, | |
6752 | to_add, 1); | |
6753 | len -= to_add; | |
7b398f8e JB |
6754 | } |
6755 | spin_unlock(&global_rsv->lock); | |
957780eb JB |
6756 | /* Add to any tickets we may have */ |
6757 | if (len) | |
6758 | space_info_add_new_bytes(fs_info, space_info, | |
6759 | len); | |
7b398f8e JB |
6760 | } |
6761 | spin_unlock(&space_info->lock); | |
ccd467d6 | 6762 | } |
11833d66 YZ |
6763 | |
6764 | if (cache) | |
6765 | btrfs_put_block_group(cache); | |
ccd467d6 CM |
6766 | return 0; |
6767 | } | |
6768 | ||
5ead2dd0 | 6769 | int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans) |
a28ec197 | 6770 | { |
5ead2dd0 | 6771 | struct btrfs_fs_info *fs_info = trans->fs_info; |
e33e17ee JM |
6772 | struct btrfs_block_group_cache *block_group, *tmp; |
6773 | struct list_head *deleted_bgs; | |
11833d66 | 6774 | struct extent_io_tree *unpin; |
1a5bc167 CM |
6775 | u64 start; |
6776 | u64 end; | |
a28ec197 | 6777 | int ret; |
a28ec197 | 6778 | |
11833d66 YZ |
6779 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) |
6780 | unpin = &fs_info->freed_extents[1]; | |
6781 | else | |
6782 | unpin = &fs_info->freed_extents[0]; | |
6783 | ||
e33e17ee | 6784 | while (!trans->aborted) { |
d4b450cd | 6785 | mutex_lock(&fs_info->unused_bg_unpin_mutex); |
1a5bc167 | 6786 | ret = find_first_extent_bit(unpin, 0, &start, &end, |
e6138876 | 6787 | EXTENT_DIRTY, NULL); |
d4b450cd FM |
6788 | if (ret) { |
6789 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); | |
a28ec197 | 6790 | break; |
d4b450cd | 6791 | } |
1f3c79a2 | 6792 | |
0b246afa | 6793 | if (btrfs_test_opt(fs_info, DISCARD)) |
2ff7e61e | 6794 | ret = btrfs_discard_extent(fs_info, start, |
5378e607 | 6795 | end + 1 - start, NULL); |
1f3c79a2 | 6796 | |
af6f8f60 | 6797 | clear_extent_dirty(unpin, start, end); |
2ff7e61e | 6798 | unpin_extent_range(fs_info, start, end, true); |
d4b450cd | 6799 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
b9473439 | 6800 | cond_resched(); |
a28ec197 | 6801 | } |
817d52f8 | 6802 | |
e33e17ee JM |
6803 | /* |
6804 | * Transaction is finished. We don't need the lock anymore. We | |
6805 | * do need to clean up the block groups in case of a transaction | |
6806 | * abort. | |
6807 | */ | |
6808 | deleted_bgs = &trans->transaction->deleted_bgs; | |
6809 | list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) { | |
6810 | u64 trimmed = 0; | |
6811 | ||
6812 | ret = -EROFS; | |
6813 | if (!trans->aborted) | |
2ff7e61e | 6814 | ret = btrfs_discard_extent(fs_info, |
e33e17ee JM |
6815 | block_group->key.objectid, |
6816 | block_group->key.offset, | |
6817 | &trimmed); | |
6818 | ||
6819 | list_del_init(&block_group->bg_list); | |
6820 | btrfs_put_block_group_trimming(block_group); | |
6821 | btrfs_put_block_group(block_group); | |
6822 | ||
6823 | if (ret) { | |
6824 | const char *errstr = btrfs_decode_error(ret); | |
6825 | btrfs_warn(fs_info, | |
913e1535 | 6826 | "discard failed while removing blockgroup: errno=%d %s", |
e33e17ee JM |
6827 | ret, errstr); |
6828 | } | |
6829 | } | |
6830 | ||
e20d96d6 CM |
6831 | return 0; |
6832 | } | |
6833 | ||
5d4f98a2 | 6834 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
e72cb923 NB |
6835 | struct btrfs_delayed_ref_node *node, u64 parent, |
6836 | u64 root_objectid, u64 owner_objectid, | |
6837 | u64 owner_offset, int refs_to_drop, | |
6838 | struct btrfs_delayed_extent_op *extent_op) | |
a28ec197 | 6839 | { |
e72cb923 | 6840 | struct btrfs_fs_info *info = trans->fs_info; |
e2fa7227 | 6841 | struct btrfs_key key; |
5d4f98a2 | 6842 | struct btrfs_path *path; |
1261ec42 | 6843 | struct btrfs_root *extent_root = info->extent_root; |
5f39d397 | 6844 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
6845 | struct btrfs_extent_item *ei; |
6846 | struct btrfs_extent_inline_ref *iref; | |
a28ec197 | 6847 | int ret; |
5d4f98a2 | 6848 | int is_data; |
952fccac CM |
6849 | int extent_slot = 0; |
6850 | int found_extent = 0; | |
6851 | int num_to_del = 1; | |
5d4f98a2 YZ |
6852 | u32 item_size; |
6853 | u64 refs; | |
c682f9b3 QW |
6854 | u64 bytenr = node->bytenr; |
6855 | u64 num_bytes = node->num_bytes; | |
fcebe456 | 6856 | int last_ref = 0; |
0b246afa | 6857 | bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA); |
037e6390 | 6858 | |
5caf2a00 | 6859 | path = btrfs_alloc_path(); |
54aa1f4d CM |
6860 | if (!path) |
6861 | return -ENOMEM; | |
5f26f772 | 6862 | |
e4058b54 | 6863 | path->reada = READA_FORWARD; |
b9473439 | 6864 | path->leave_spinning = 1; |
5d4f98a2 YZ |
6865 | |
6866 | is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID; | |
6867 | BUG_ON(!is_data && refs_to_drop != 1); | |
6868 | ||
3173a18f | 6869 | if (is_data) |
897ca819 | 6870 | skinny_metadata = false; |
3173a18f | 6871 | |
fbe4801b NB |
6872 | ret = lookup_extent_backref(trans, path, &iref, bytenr, num_bytes, |
6873 | parent, root_objectid, owner_objectid, | |
5d4f98a2 | 6874 | owner_offset); |
7bb86316 | 6875 | if (ret == 0) { |
952fccac | 6876 | extent_slot = path->slots[0]; |
5d4f98a2 YZ |
6877 | while (extent_slot >= 0) { |
6878 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
952fccac | 6879 | extent_slot); |
5d4f98a2 | 6880 | if (key.objectid != bytenr) |
952fccac | 6881 | break; |
5d4f98a2 YZ |
6882 | if (key.type == BTRFS_EXTENT_ITEM_KEY && |
6883 | key.offset == num_bytes) { | |
952fccac CM |
6884 | found_extent = 1; |
6885 | break; | |
6886 | } | |
3173a18f JB |
6887 | if (key.type == BTRFS_METADATA_ITEM_KEY && |
6888 | key.offset == owner_objectid) { | |
6889 | found_extent = 1; | |
6890 | break; | |
6891 | } | |
952fccac CM |
6892 | if (path->slots[0] - extent_slot > 5) |
6893 | break; | |
5d4f98a2 | 6894 | extent_slot--; |
952fccac | 6895 | } |
5d4f98a2 YZ |
6896 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
6897 | item_size = btrfs_item_size_nr(path->nodes[0], extent_slot); | |
6898 | if (found_extent && item_size < sizeof(*ei)) | |
6899 | found_extent = 0; | |
6900 | #endif | |
31840ae1 | 6901 | if (!found_extent) { |
5d4f98a2 | 6902 | BUG_ON(iref); |
87bde3cd JM |
6903 | ret = remove_extent_backref(trans, info, path, NULL, |
6904 | refs_to_drop, | |
fcebe456 | 6905 | is_data, &last_ref); |
005d6427 | 6906 | if (ret) { |
66642832 | 6907 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
6908 | goto out; |
6909 | } | |
b3b4aa74 | 6910 | btrfs_release_path(path); |
b9473439 | 6911 | path->leave_spinning = 1; |
5d4f98a2 YZ |
6912 | |
6913 | key.objectid = bytenr; | |
6914 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
6915 | key.offset = num_bytes; | |
6916 | ||
3173a18f JB |
6917 | if (!is_data && skinny_metadata) { |
6918 | key.type = BTRFS_METADATA_ITEM_KEY; | |
6919 | key.offset = owner_objectid; | |
6920 | } | |
6921 | ||
31840ae1 ZY |
6922 | ret = btrfs_search_slot(trans, extent_root, |
6923 | &key, path, -1, 1); | |
3173a18f JB |
6924 | if (ret > 0 && skinny_metadata && path->slots[0]) { |
6925 | /* | |
6926 | * Couldn't find our skinny metadata item, | |
6927 | * see if we have ye olde extent item. | |
6928 | */ | |
6929 | path->slots[0]--; | |
6930 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
6931 | path->slots[0]); | |
6932 | if (key.objectid == bytenr && | |
6933 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
6934 | key.offset == num_bytes) | |
6935 | ret = 0; | |
6936 | } | |
6937 | ||
6938 | if (ret > 0 && skinny_metadata) { | |
6939 | skinny_metadata = false; | |
9ce49a0b | 6940 | key.objectid = bytenr; |
3173a18f JB |
6941 | key.type = BTRFS_EXTENT_ITEM_KEY; |
6942 | key.offset = num_bytes; | |
6943 | btrfs_release_path(path); | |
6944 | ret = btrfs_search_slot(trans, extent_root, | |
6945 | &key, path, -1, 1); | |
6946 | } | |
6947 | ||
f3465ca4 | 6948 | if (ret) { |
5d163e0e JM |
6949 | btrfs_err(info, |
6950 | "umm, got %d back from search, was looking for %llu", | |
6951 | ret, bytenr); | |
b783e62d | 6952 | if (ret > 0) |
a4f78750 | 6953 | btrfs_print_leaf(path->nodes[0]); |
f3465ca4 | 6954 | } |
005d6427 | 6955 | if (ret < 0) { |
66642832 | 6956 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
6957 | goto out; |
6958 | } | |
31840ae1 ZY |
6959 | extent_slot = path->slots[0]; |
6960 | } | |
fae7f21c | 6961 | } else if (WARN_ON(ret == -ENOENT)) { |
a4f78750 | 6962 | btrfs_print_leaf(path->nodes[0]); |
c2cf52eb SK |
6963 | btrfs_err(info, |
6964 | "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu", | |
c1c9ff7c GU |
6965 | bytenr, parent, root_objectid, owner_objectid, |
6966 | owner_offset); | |
66642832 | 6967 | btrfs_abort_transaction(trans, ret); |
c4a050bb | 6968 | goto out; |
79787eaa | 6969 | } else { |
66642832 | 6970 | btrfs_abort_transaction(trans, ret); |
005d6427 | 6971 | goto out; |
7bb86316 | 6972 | } |
5f39d397 CM |
6973 | |
6974 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
6975 | item_size = btrfs_item_size_nr(leaf, extent_slot); |
6976 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
6977 | if (item_size < sizeof(*ei)) { | |
6978 | BUG_ON(found_extent || extent_slot != path->slots[0]); | |
87bde3cd JM |
6979 | ret = convert_extent_item_v0(trans, info, path, owner_objectid, |
6980 | 0); | |
005d6427 | 6981 | if (ret < 0) { |
66642832 | 6982 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
6983 | goto out; |
6984 | } | |
5d4f98a2 | 6985 | |
b3b4aa74 | 6986 | btrfs_release_path(path); |
5d4f98a2 YZ |
6987 | path->leave_spinning = 1; |
6988 | ||
6989 | key.objectid = bytenr; | |
6990 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
6991 | key.offset = num_bytes; | |
6992 | ||
6993 | ret = btrfs_search_slot(trans, extent_root, &key, path, | |
6994 | -1, 1); | |
6995 | if (ret) { | |
5d163e0e JM |
6996 | btrfs_err(info, |
6997 | "umm, got %d back from search, was looking for %llu", | |
c1c9ff7c | 6998 | ret, bytenr); |
a4f78750 | 6999 | btrfs_print_leaf(path->nodes[0]); |
5d4f98a2 | 7000 | } |
005d6427 | 7001 | if (ret < 0) { |
66642832 | 7002 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7003 | goto out; |
7004 | } | |
7005 | ||
5d4f98a2 YZ |
7006 | extent_slot = path->slots[0]; |
7007 | leaf = path->nodes[0]; | |
7008 | item_size = btrfs_item_size_nr(leaf, extent_slot); | |
7009 | } | |
7010 | #endif | |
7011 | BUG_ON(item_size < sizeof(*ei)); | |
952fccac | 7012 | ei = btrfs_item_ptr(leaf, extent_slot, |
123abc88 | 7013 | struct btrfs_extent_item); |
3173a18f JB |
7014 | if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID && |
7015 | key.type == BTRFS_EXTENT_ITEM_KEY) { | |
5d4f98a2 YZ |
7016 | struct btrfs_tree_block_info *bi; |
7017 | BUG_ON(item_size < sizeof(*ei) + sizeof(*bi)); | |
7018 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
7019 | WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi)); | |
7020 | } | |
56bec294 | 7021 | |
5d4f98a2 | 7022 | refs = btrfs_extent_refs(leaf, ei); |
32b02538 | 7023 | if (refs < refs_to_drop) { |
5d163e0e JM |
7024 | btrfs_err(info, |
7025 | "trying to drop %d refs but we only have %Lu for bytenr %Lu", | |
7026 | refs_to_drop, refs, bytenr); | |
32b02538 | 7027 | ret = -EINVAL; |
66642832 | 7028 | btrfs_abort_transaction(trans, ret); |
32b02538 JB |
7029 | goto out; |
7030 | } | |
56bec294 | 7031 | refs -= refs_to_drop; |
5f39d397 | 7032 | |
5d4f98a2 YZ |
7033 | if (refs > 0) { |
7034 | if (extent_op) | |
7035 | __run_delayed_extent_op(extent_op, leaf, ei); | |
7036 | /* | |
7037 | * In the case of inline back ref, reference count will | |
7038 | * be updated by remove_extent_backref | |
952fccac | 7039 | */ |
5d4f98a2 YZ |
7040 | if (iref) { |
7041 | BUG_ON(!found_extent); | |
7042 | } else { | |
7043 | btrfs_set_extent_refs(leaf, ei, refs); | |
7044 | btrfs_mark_buffer_dirty(leaf); | |
7045 | } | |
7046 | if (found_extent) { | |
87bde3cd | 7047 | ret = remove_extent_backref(trans, info, path, |
5d4f98a2 | 7048 | iref, refs_to_drop, |
fcebe456 | 7049 | is_data, &last_ref); |
005d6427 | 7050 | if (ret) { |
66642832 | 7051 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7052 | goto out; |
7053 | } | |
952fccac | 7054 | } |
5d4f98a2 | 7055 | } else { |
5d4f98a2 YZ |
7056 | if (found_extent) { |
7057 | BUG_ON(is_data && refs_to_drop != | |
9ed0dea0 | 7058 | extent_data_ref_count(path, iref)); |
5d4f98a2 YZ |
7059 | if (iref) { |
7060 | BUG_ON(path->slots[0] != extent_slot); | |
7061 | } else { | |
7062 | BUG_ON(path->slots[0] != extent_slot + 1); | |
7063 | path->slots[0] = extent_slot; | |
7064 | num_to_del = 2; | |
7065 | } | |
78fae27e | 7066 | } |
b9473439 | 7067 | |
fcebe456 | 7068 | last_ref = 1; |
952fccac CM |
7069 | ret = btrfs_del_items(trans, extent_root, path, path->slots[0], |
7070 | num_to_del); | |
005d6427 | 7071 | if (ret) { |
66642832 | 7072 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7073 | goto out; |
7074 | } | |
b3b4aa74 | 7075 | btrfs_release_path(path); |
21af804c | 7076 | |
5d4f98a2 | 7077 | if (is_data) { |
5b4aacef | 7078 | ret = btrfs_del_csums(trans, info, bytenr, num_bytes); |
005d6427 | 7079 | if (ret) { |
66642832 | 7080 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7081 | goto out; |
7082 | } | |
459931ec CM |
7083 | } |
7084 | ||
e7355e50 | 7085 | ret = add_to_free_space_tree(trans, bytenr, num_bytes); |
1e144fb8 | 7086 | if (ret) { |
66642832 | 7087 | btrfs_abort_transaction(trans, ret); |
1e144fb8 OS |
7088 | goto out; |
7089 | } | |
7090 | ||
0b246afa | 7091 | ret = update_block_group(trans, info, bytenr, num_bytes, 0); |
005d6427 | 7092 | if (ret) { |
66642832 | 7093 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7094 | goto out; |
7095 | } | |
a28ec197 | 7096 | } |
fcebe456 JB |
7097 | btrfs_release_path(path); |
7098 | ||
79787eaa | 7099 | out: |
5caf2a00 | 7100 | btrfs_free_path(path); |
a28ec197 CM |
7101 | return ret; |
7102 | } | |
7103 | ||
1887be66 | 7104 | /* |
f0486c68 | 7105 | * when we free an block, it is possible (and likely) that we free the last |
1887be66 CM |
7106 | * delayed ref for that extent as well. This searches the delayed ref tree for |
7107 | * a given extent, and if there are no other delayed refs to be processed, it | |
7108 | * removes it from the tree. | |
7109 | */ | |
7110 | static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans, | |
2ff7e61e | 7111 | u64 bytenr) |
1887be66 CM |
7112 | { |
7113 | struct btrfs_delayed_ref_head *head; | |
7114 | struct btrfs_delayed_ref_root *delayed_refs; | |
f0486c68 | 7115 | int ret = 0; |
1887be66 CM |
7116 | |
7117 | delayed_refs = &trans->transaction->delayed_refs; | |
7118 | spin_lock(&delayed_refs->lock); | |
f72ad18e | 7119 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
1887be66 | 7120 | if (!head) |
cf93da7b | 7121 | goto out_delayed_unlock; |
1887be66 | 7122 | |
d7df2c79 | 7123 | spin_lock(&head->lock); |
0e0adbcf | 7124 | if (!RB_EMPTY_ROOT(&head->ref_tree)) |
1887be66 CM |
7125 | goto out; |
7126 | ||
5d4f98a2 YZ |
7127 | if (head->extent_op) { |
7128 | if (!head->must_insert_reserved) | |
7129 | goto out; | |
78a6184a | 7130 | btrfs_free_delayed_extent_op(head->extent_op); |
5d4f98a2 YZ |
7131 | head->extent_op = NULL; |
7132 | } | |
7133 | ||
1887be66 CM |
7134 | /* |
7135 | * waiting for the lock here would deadlock. If someone else has it | |
7136 | * locked they are already in the process of dropping it anyway | |
7137 | */ | |
7138 | if (!mutex_trylock(&head->mutex)) | |
7139 | goto out; | |
7140 | ||
7141 | /* | |
7142 | * at this point we have a head with no other entries. Go | |
7143 | * ahead and process it. | |
7144 | */ | |
c46effa6 | 7145 | rb_erase(&head->href_node, &delayed_refs->href_root); |
d278850e | 7146 | RB_CLEAR_NODE(&head->href_node); |
d7df2c79 | 7147 | atomic_dec(&delayed_refs->num_entries); |
1887be66 CM |
7148 | |
7149 | /* | |
7150 | * we don't take a ref on the node because we're removing it from the | |
7151 | * tree, so we just steal the ref the tree was holding. | |
7152 | */ | |
c3e69d58 | 7153 | delayed_refs->num_heads--; |
d7df2c79 | 7154 | if (head->processing == 0) |
c3e69d58 | 7155 | delayed_refs->num_heads_ready--; |
d7df2c79 JB |
7156 | head->processing = 0; |
7157 | spin_unlock(&head->lock); | |
1887be66 CM |
7158 | spin_unlock(&delayed_refs->lock); |
7159 | ||
f0486c68 YZ |
7160 | BUG_ON(head->extent_op); |
7161 | if (head->must_insert_reserved) | |
7162 | ret = 1; | |
7163 | ||
7164 | mutex_unlock(&head->mutex); | |
d278850e | 7165 | btrfs_put_delayed_ref_head(head); |
f0486c68 | 7166 | return ret; |
1887be66 | 7167 | out: |
d7df2c79 | 7168 | spin_unlock(&head->lock); |
cf93da7b CM |
7169 | |
7170 | out_delayed_unlock: | |
1887be66 CM |
7171 | spin_unlock(&delayed_refs->lock); |
7172 | return 0; | |
7173 | } | |
7174 | ||
f0486c68 YZ |
7175 | void btrfs_free_tree_block(struct btrfs_trans_handle *trans, |
7176 | struct btrfs_root *root, | |
7177 | struct extent_buffer *buf, | |
5581a51a | 7178 | u64 parent, int last_ref) |
f0486c68 | 7179 | { |
0b246afa | 7180 | struct btrfs_fs_info *fs_info = root->fs_info; |
b150a4f1 | 7181 | int pin = 1; |
f0486c68 YZ |
7182 | int ret; |
7183 | ||
7184 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { | |
d7eae340 OS |
7185 | int old_ref_mod, new_ref_mod; |
7186 | ||
fd708b81 JB |
7187 | btrfs_ref_tree_mod(root, buf->start, buf->len, parent, |
7188 | root->root_key.objectid, | |
7189 | btrfs_header_level(buf), 0, | |
7190 | BTRFS_DROP_DELAYED_REF); | |
44e1c47d | 7191 | ret = btrfs_add_delayed_tree_ref(trans, buf->start, |
7be07912 | 7192 | buf->len, parent, |
0b246afa JM |
7193 | root->root_key.objectid, |
7194 | btrfs_header_level(buf), | |
7be07912 | 7195 | BTRFS_DROP_DELAYED_REF, NULL, |
d7eae340 | 7196 | &old_ref_mod, &new_ref_mod); |
79787eaa | 7197 | BUG_ON(ret); /* -ENOMEM */ |
d7eae340 | 7198 | pin = old_ref_mod >= 0 && new_ref_mod < 0; |
f0486c68 YZ |
7199 | } |
7200 | ||
0a16c7d7 | 7201 | if (last_ref && btrfs_header_generation(buf) == trans->transid) { |
6219872d FM |
7202 | struct btrfs_block_group_cache *cache; |
7203 | ||
f0486c68 | 7204 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { |
2ff7e61e | 7205 | ret = check_ref_cleanup(trans, buf->start); |
f0486c68 | 7206 | if (!ret) |
37be25bc | 7207 | goto out; |
f0486c68 YZ |
7208 | } |
7209 | ||
4da8b76d | 7210 | pin = 0; |
0b246afa | 7211 | cache = btrfs_lookup_block_group(fs_info, buf->start); |
6219872d | 7212 | |
f0486c68 | 7213 | if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) { |
2ff7e61e JM |
7214 | pin_down_extent(fs_info, cache, buf->start, |
7215 | buf->len, 1); | |
6219872d | 7216 | btrfs_put_block_group(cache); |
37be25bc | 7217 | goto out; |
f0486c68 YZ |
7218 | } |
7219 | ||
7220 | WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)); | |
7221 | ||
7222 | btrfs_add_free_space(cache, buf->start, buf->len); | |
4824f1f4 | 7223 | btrfs_free_reserved_bytes(cache, buf->len, 0); |
6219872d | 7224 | btrfs_put_block_group(cache); |
71ff6437 | 7225 | trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len); |
f0486c68 YZ |
7226 | } |
7227 | out: | |
b150a4f1 | 7228 | if (pin) |
29d2b84c | 7229 | add_pinned_bytes(fs_info, buf->len, true, |
b150a4f1 JB |
7230 | root->root_key.objectid); |
7231 | ||
0a16c7d7 OS |
7232 | if (last_ref) { |
7233 | /* | |
7234 | * Deleting the buffer, clear the corrupt flag since it doesn't | |
7235 | * matter anymore. | |
7236 | */ | |
7237 | clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags); | |
7238 | } | |
f0486c68 YZ |
7239 | } |
7240 | ||
79787eaa | 7241 | /* Can return -ENOMEM */ |
2ff7e61e | 7242 | int btrfs_free_extent(struct btrfs_trans_handle *trans, |
84f7d8e6 | 7243 | struct btrfs_root *root, |
66d7e7f0 | 7244 | u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid, |
b06c4bf5 | 7245 | u64 owner, u64 offset) |
925baedd | 7246 | { |
84f7d8e6 | 7247 | struct btrfs_fs_info *fs_info = root->fs_info; |
d7eae340 | 7248 | int old_ref_mod, new_ref_mod; |
925baedd CM |
7249 | int ret; |
7250 | ||
f5ee5c9a | 7251 | if (btrfs_is_testing(fs_info)) |
faa2dbf0 | 7252 | return 0; |
fccb84c9 | 7253 | |
fd708b81 JB |
7254 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) |
7255 | btrfs_ref_tree_mod(root, bytenr, num_bytes, parent, | |
7256 | root_objectid, owner, offset, | |
7257 | BTRFS_DROP_DELAYED_REF); | |
7258 | ||
56bec294 CM |
7259 | /* |
7260 | * tree log blocks never actually go into the extent allocation | |
7261 | * tree, just update pinning info and exit early. | |
56bec294 | 7262 | */ |
5d4f98a2 YZ |
7263 | if (root_objectid == BTRFS_TREE_LOG_OBJECTID) { |
7264 | WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID); | |
b9473439 | 7265 | /* unlocks the pinned mutex */ |
2ff7e61e | 7266 | btrfs_pin_extent(fs_info, bytenr, num_bytes, 1); |
d7eae340 | 7267 | old_ref_mod = new_ref_mod = 0; |
56bec294 | 7268 | ret = 0; |
5d4f98a2 | 7269 | } else if (owner < BTRFS_FIRST_FREE_OBJECTID) { |
44e1c47d | 7270 | ret = btrfs_add_delayed_tree_ref(trans, bytenr, |
7be07912 OS |
7271 | num_bytes, parent, |
7272 | root_objectid, (int)owner, | |
7273 | BTRFS_DROP_DELAYED_REF, NULL, | |
d7eae340 | 7274 | &old_ref_mod, &new_ref_mod); |
5d4f98a2 | 7275 | } else { |
88a979c6 | 7276 | ret = btrfs_add_delayed_data_ref(trans, bytenr, |
7be07912 OS |
7277 | num_bytes, parent, |
7278 | root_objectid, owner, offset, | |
7279 | 0, BTRFS_DROP_DELAYED_REF, | |
d7eae340 | 7280 | &old_ref_mod, &new_ref_mod); |
56bec294 | 7281 | } |
d7eae340 | 7282 | |
29d2b84c NB |
7283 | if (ret == 0 && old_ref_mod >= 0 && new_ref_mod < 0) { |
7284 | bool metadata = owner < BTRFS_FIRST_FREE_OBJECTID; | |
7285 | ||
7286 | add_pinned_bytes(fs_info, num_bytes, metadata, root_objectid); | |
7287 | } | |
d7eae340 | 7288 | |
925baedd CM |
7289 | return ret; |
7290 | } | |
7291 | ||
817d52f8 JB |
7292 | /* |
7293 | * when we wait for progress in the block group caching, its because | |
7294 | * our allocation attempt failed at least once. So, we must sleep | |
7295 | * and let some progress happen before we try again. | |
7296 | * | |
7297 | * This function will sleep at least once waiting for new free space to | |
7298 | * show up, and then it will check the block group free space numbers | |
7299 | * for our min num_bytes. Another option is to have it go ahead | |
7300 | * and look in the rbtree for a free extent of a given size, but this | |
7301 | * is a good start. | |
36cce922 JB |
7302 | * |
7303 | * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using | |
7304 | * any of the information in this block group. | |
817d52f8 | 7305 | */ |
36cce922 | 7306 | static noinline void |
817d52f8 JB |
7307 | wait_block_group_cache_progress(struct btrfs_block_group_cache *cache, |
7308 | u64 num_bytes) | |
7309 | { | |
11833d66 | 7310 | struct btrfs_caching_control *caching_ctl; |
817d52f8 | 7311 | |
11833d66 YZ |
7312 | caching_ctl = get_caching_control(cache); |
7313 | if (!caching_ctl) | |
36cce922 | 7314 | return; |
817d52f8 | 7315 | |
11833d66 | 7316 | wait_event(caching_ctl->wait, block_group_cache_done(cache) || |
34d52cb6 | 7317 | (cache->free_space_ctl->free_space >= num_bytes)); |
11833d66 YZ |
7318 | |
7319 | put_caching_control(caching_ctl); | |
11833d66 YZ |
7320 | } |
7321 | ||
7322 | static noinline int | |
7323 | wait_block_group_cache_done(struct btrfs_block_group_cache *cache) | |
7324 | { | |
7325 | struct btrfs_caching_control *caching_ctl; | |
36cce922 | 7326 | int ret = 0; |
11833d66 YZ |
7327 | |
7328 | caching_ctl = get_caching_control(cache); | |
7329 | if (!caching_ctl) | |
36cce922 | 7330 | return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0; |
11833d66 YZ |
7331 | |
7332 | wait_event(caching_ctl->wait, block_group_cache_done(cache)); | |
36cce922 JB |
7333 | if (cache->cached == BTRFS_CACHE_ERROR) |
7334 | ret = -EIO; | |
11833d66 | 7335 | put_caching_control(caching_ctl); |
36cce922 | 7336 | return ret; |
817d52f8 JB |
7337 | } |
7338 | ||
7339 | enum btrfs_loop_type { | |
285ff5af JB |
7340 | LOOP_CACHING_NOWAIT = 0, |
7341 | LOOP_CACHING_WAIT = 1, | |
7342 | LOOP_ALLOC_CHUNK = 2, | |
7343 | LOOP_NO_EMPTY_SIZE = 3, | |
817d52f8 JB |
7344 | }; |
7345 | ||
e570fd27 MX |
7346 | static inline void |
7347 | btrfs_lock_block_group(struct btrfs_block_group_cache *cache, | |
7348 | int delalloc) | |
7349 | { | |
7350 | if (delalloc) | |
7351 | down_read(&cache->data_rwsem); | |
7352 | } | |
7353 | ||
7354 | static inline void | |
7355 | btrfs_grab_block_group(struct btrfs_block_group_cache *cache, | |
7356 | int delalloc) | |
7357 | { | |
7358 | btrfs_get_block_group(cache); | |
7359 | if (delalloc) | |
7360 | down_read(&cache->data_rwsem); | |
7361 | } | |
7362 | ||
7363 | static struct btrfs_block_group_cache * | |
7364 | btrfs_lock_cluster(struct btrfs_block_group_cache *block_group, | |
7365 | struct btrfs_free_cluster *cluster, | |
7366 | int delalloc) | |
7367 | { | |
89771cc9 | 7368 | struct btrfs_block_group_cache *used_bg = NULL; |
6719afdc | 7369 | |
e570fd27 | 7370 | spin_lock(&cluster->refill_lock); |
6719afdc GU |
7371 | while (1) { |
7372 | used_bg = cluster->block_group; | |
7373 | if (!used_bg) | |
7374 | return NULL; | |
7375 | ||
7376 | if (used_bg == block_group) | |
e570fd27 MX |
7377 | return used_bg; |
7378 | ||
6719afdc | 7379 | btrfs_get_block_group(used_bg); |
e570fd27 | 7380 | |
6719afdc GU |
7381 | if (!delalloc) |
7382 | return used_bg; | |
e570fd27 | 7383 | |
6719afdc GU |
7384 | if (down_read_trylock(&used_bg->data_rwsem)) |
7385 | return used_bg; | |
e570fd27 | 7386 | |
6719afdc | 7387 | spin_unlock(&cluster->refill_lock); |
e570fd27 | 7388 | |
e321f8a8 LB |
7389 | /* We should only have one-level nested. */ |
7390 | down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING); | |
e570fd27 | 7391 | |
6719afdc GU |
7392 | spin_lock(&cluster->refill_lock); |
7393 | if (used_bg == cluster->block_group) | |
7394 | return used_bg; | |
e570fd27 | 7395 | |
6719afdc GU |
7396 | up_read(&used_bg->data_rwsem); |
7397 | btrfs_put_block_group(used_bg); | |
7398 | } | |
e570fd27 MX |
7399 | } |
7400 | ||
7401 | static inline void | |
7402 | btrfs_release_block_group(struct btrfs_block_group_cache *cache, | |
7403 | int delalloc) | |
7404 | { | |
7405 | if (delalloc) | |
7406 | up_read(&cache->data_rwsem); | |
7407 | btrfs_put_block_group(cache); | |
7408 | } | |
7409 | ||
fec577fb CM |
7410 | /* |
7411 | * walks the btree of allocated extents and find a hole of a given size. | |
7412 | * The key ins is changed to record the hole: | |
a4820398 | 7413 | * ins->objectid == start position |
62e2749e | 7414 | * ins->flags = BTRFS_EXTENT_ITEM_KEY |
a4820398 | 7415 | * ins->offset == the size of the hole. |
fec577fb | 7416 | * Any available blocks before search_start are skipped. |
a4820398 MX |
7417 | * |
7418 | * If there is no suitable free space, we will record the max size of | |
7419 | * the free space extent currently. | |
fec577fb | 7420 | */ |
87bde3cd | 7421 | static noinline int find_free_extent(struct btrfs_fs_info *fs_info, |
18513091 WX |
7422 | u64 ram_bytes, u64 num_bytes, u64 empty_size, |
7423 | u64 hint_byte, struct btrfs_key *ins, | |
7424 | u64 flags, int delalloc) | |
fec577fb | 7425 | { |
80eb234a | 7426 | int ret = 0; |
0b246afa | 7427 | struct btrfs_root *root = fs_info->extent_root; |
fa9c0d79 | 7428 | struct btrfs_free_cluster *last_ptr = NULL; |
80eb234a | 7429 | struct btrfs_block_group_cache *block_group = NULL; |
81c9ad23 | 7430 | u64 search_start = 0; |
a4820398 | 7431 | u64 max_extent_size = 0; |
c759c4e1 | 7432 | u64 empty_cluster = 0; |
80eb234a | 7433 | struct btrfs_space_info *space_info; |
fa9c0d79 | 7434 | int loop = 0; |
3e72ee88 | 7435 | int index = btrfs_bg_flags_to_raid_index(flags); |
0a24325e | 7436 | bool failed_cluster_refill = false; |
1cdda9b8 | 7437 | bool failed_alloc = false; |
67377734 | 7438 | bool use_cluster = true; |
60d2adbb | 7439 | bool have_caching_bg = false; |
13a0db5a | 7440 | bool orig_have_caching_bg = false; |
a5e681d9 | 7441 | bool full_search = false; |
fec577fb | 7442 | |
0b246afa | 7443 | WARN_ON(num_bytes < fs_info->sectorsize); |
962a298f | 7444 | ins->type = BTRFS_EXTENT_ITEM_KEY; |
80eb234a JB |
7445 | ins->objectid = 0; |
7446 | ins->offset = 0; | |
b1a4d965 | 7447 | |
71ff6437 | 7448 | trace_find_free_extent(fs_info, num_bytes, empty_size, flags); |
3f7de037 | 7449 | |
0b246afa | 7450 | space_info = __find_space_info(fs_info, flags); |
1b1d1f66 | 7451 | if (!space_info) { |
0b246afa | 7452 | btrfs_err(fs_info, "No space info for %llu", flags); |
1b1d1f66 JB |
7453 | return -ENOSPC; |
7454 | } | |
2552d17e | 7455 | |
67377734 | 7456 | /* |
4f4db217 JB |
7457 | * If our free space is heavily fragmented we may not be able to make |
7458 | * big contiguous allocations, so instead of doing the expensive search | |
7459 | * for free space, simply return ENOSPC with our max_extent_size so we | |
7460 | * can go ahead and search for a more manageable chunk. | |
7461 | * | |
7462 | * If our max_extent_size is large enough for our allocation simply | |
7463 | * disable clustering since we will likely not be able to find enough | |
7464 | * space to create a cluster and induce latency trying. | |
67377734 | 7465 | */ |
4f4db217 JB |
7466 | if (unlikely(space_info->max_extent_size)) { |
7467 | spin_lock(&space_info->lock); | |
7468 | if (space_info->max_extent_size && | |
7469 | num_bytes > space_info->max_extent_size) { | |
7470 | ins->offset = space_info->max_extent_size; | |
7471 | spin_unlock(&space_info->lock); | |
7472 | return -ENOSPC; | |
7473 | } else if (space_info->max_extent_size) { | |
7474 | use_cluster = false; | |
7475 | } | |
7476 | spin_unlock(&space_info->lock); | |
fa9c0d79 | 7477 | } |
0f9dd46c | 7478 | |
2ff7e61e | 7479 | last_ptr = fetch_cluster_info(fs_info, space_info, &empty_cluster); |
239b14b3 | 7480 | if (last_ptr) { |
fa9c0d79 CM |
7481 | spin_lock(&last_ptr->lock); |
7482 | if (last_ptr->block_group) | |
7483 | hint_byte = last_ptr->window_start; | |
c759c4e1 JB |
7484 | if (last_ptr->fragmented) { |
7485 | /* | |
7486 | * We still set window_start so we can keep track of the | |
7487 | * last place we found an allocation to try and save | |
7488 | * some time. | |
7489 | */ | |
7490 | hint_byte = last_ptr->window_start; | |
7491 | use_cluster = false; | |
7492 | } | |
fa9c0d79 | 7493 | spin_unlock(&last_ptr->lock); |
239b14b3 | 7494 | } |
fa9c0d79 | 7495 | |
2ff7e61e | 7496 | search_start = max(search_start, first_logical_byte(fs_info, 0)); |
239b14b3 | 7497 | search_start = max(search_start, hint_byte); |
2552d17e | 7498 | if (search_start == hint_byte) { |
0b246afa | 7499 | block_group = btrfs_lookup_block_group(fs_info, search_start); |
817d52f8 JB |
7500 | /* |
7501 | * we don't want to use the block group if it doesn't match our | |
7502 | * allocation bits, or if its not cached. | |
ccf0e725 JB |
7503 | * |
7504 | * However if we are re-searching with an ideal block group | |
7505 | * picked out then we don't care that the block group is cached. | |
817d52f8 | 7506 | */ |
b6919a58 | 7507 | if (block_group && block_group_bits(block_group, flags) && |
285ff5af | 7508 | block_group->cached != BTRFS_CACHE_NO) { |
2552d17e | 7509 | down_read(&space_info->groups_sem); |
44fb5511 CM |
7510 | if (list_empty(&block_group->list) || |
7511 | block_group->ro) { | |
7512 | /* | |
7513 | * someone is removing this block group, | |
7514 | * we can't jump into the have_block_group | |
7515 | * target because our list pointers are not | |
7516 | * valid | |
7517 | */ | |
7518 | btrfs_put_block_group(block_group); | |
7519 | up_read(&space_info->groups_sem); | |
ccf0e725 | 7520 | } else { |
3e72ee88 QW |
7521 | index = btrfs_bg_flags_to_raid_index( |
7522 | block_group->flags); | |
e570fd27 | 7523 | btrfs_lock_block_group(block_group, delalloc); |
44fb5511 | 7524 | goto have_block_group; |
ccf0e725 | 7525 | } |
2552d17e | 7526 | } else if (block_group) { |
fa9c0d79 | 7527 | btrfs_put_block_group(block_group); |
2552d17e | 7528 | } |
42e70e7a | 7529 | } |
2552d17e | 7530 | search: |
60d2adbb | 7531 | have_caching_bg = false; |
3e72ee88 | 7532 | if (index == 0 || index == btrfs_bg_flags_to_raid_index(flags)) |
a5e681d9 | 7533 | full_search = true; |
80eb234a | 7534 | down_read(&space_info->groups_sem); |
b742bb82 YZ |
7535 | list_for_each_entry(block_group, &space_info->block_groups[index], |
7536 | list) { | |
6226cb0a | 7537 | u64 offset; |
817d52f8 | 7538 | int cached; |
8a1413a2 | 7539 | |
14443937 JM |
7540 | /* If the block group is read-only, we can skip it entirely. */ |
7541 | if (unlikely(block_group->ro)) | |
7542 | continue; | |
7543 | ||
e570fd27 | 7544 | btrfs_grab_block_group(block_group, delalloc); |
2552d17e | 7545 | search_start = block_group->key.objectid; |
42e70e7a | 7546 | |
83a50de9 CM |
7547 | /* |
7548 | * this can happen if we end up cycling through all the | |
7549 | * raid types, but we want to make sure we only allocate | |
7550 | * for the proper type. | |
7551 | */ | |
b6919a58 | 7552 | if (!block_group_bits(block_group, flags)) { |
bece2e82 | 7553 | u64 extra = BTRFS_BLOCK_GROUP_DUP | |
83a50de9 | 7554 | BTRFS_BLOCK_GROUP_RAID1 | |
53b381b3 DW |
7555 | BTRFS_BLOCK_GROUP_RAID5 | |
7556 | BTRFS_BLOCK_GROUP_RAID6 | | |
83a50de9 CM |
7557 | BTRFS_BLOCK_GROUP_RAID10; |
7558 | ||
7559 | /* | |
7560 | * if they asked for extra copies and this block group | |
7561 | * doesn't provide them, bail. This does allow us to | |
7562 | * fill raid0 from raid1. | |
7563 | */ | |
b6919a58 | 7564 | if ((flags & extra) && !(block_group->flags & extra)) |
83a50de9 CM |
7565 | goto loop; |
7566 | } | |
7567 | ||
2552d17e | 7568 | have_block_group: |
291c7d2f JB |
7569 | cached = block_group_cache_done(block_group); |
7570 | if (unlikely(!cached)) { | |
a5e681d9 | 7571 | have_caching_bg = true; |
f6373bf3 | 7572 | ret = cache_block_group(block_group, 0); |
1d4284bd CM |
7573 | BUG_ON(ret < 0); |
7574 | ret = 0; | |
817d52f8 JB |
7575 | } |
7576 | ||
36cce922 JB |
7577 | if (unlikely(block_group->cached == BTRFS_CACHE_ERROR)) |
7578 | goto loop; | |
0f9dd46c | 7579 | |
0a24325e | 7580 | /* |
062c05c4 AO |
7581 | * Ok we want to try and use the cluster allocator, so |
7582 | * lets look there | |
0a24325e | 7583 | */ |
c759c4e1 | 7584 | if (last_ptr && use_cluster) { |
215a63d1 | 7585 | struct btrfs_block_group_cache *used_block_group; |
8de972b4 | 7586 | unsigned long aligned_cluster; |
fa9c0d79 CM |
7587 | /* |
7588 | * the refill lock keeps out other | |
7589 | * people trying to start a new cluster | |
7590 | */ | |
e570fd27 MX |
7591 | used_block_group = btrfs_lock_cluster(block_group, |
7592 | last_ptr, | |
7593 | delalloc); | |
7594 | if (!used_block_group) | |
44fb5511 | 7595 | goto refill_cluster; |
274bd4fb | 7596 | |
e570fd27 MX |
7597 | if (used_block_group != block_group && |
7598 | (used_block_group->ro || | |
7599 | !block_group_bits(used_block_group, flags))) | |
7600 | goto release_cluster; | |
44fb5511 | 7601 | |
274bd4fb | 7602 | offset = btrfs_alloc_from_cluster(used_block_group, |
a4820398 MX |
7603 | last_ptr, |
7604 | num_bytes, | |
7605 | used_block_group->key.objectid, | |
7606 | &max_extent_size); | |
fa9c0d79 CM |
7607 | if (offset) { |
7608 | /* we have a block, we're done */ | |
7609 | spin_unlock(&last_ptr->refill_lock); | |
3dca5c94 | 7610 | trace_btrfs_reserve_extent_cluster( |
89d4346a MX |
7611 | used_block_group, |
7612 | search_start, num_bytes); | |
215a63d1 | 7613 | if (used_block_group != block_group) { |
e570fd27 MX |
7614 | btrfs_release_block_group(block_group, |
7615 | delalloc); | |
215a63d1 MX |
7616 | block_group = used_block_group; |
7617 | } | |
fa9c0d79 CM |
7618 | goto checks; |
7619 | } | |
7620 | ||
274bd4fb | 7621 | WARN_ON(last_ptr->block_group != used_block_group); |
e570fd27 | 7622 | release_cluster: |
062c05c4 AO |
7623 | /* If we are on LOOP_NO_EMPTY_SIZE, we can't |
7624 | * set up a new clusters, so lets just skip it | |
7625 | * and let the allocator find whatever block | |
7626 | * it can find. If we reach this point, we | |
7627 | * will have tried the cluster allocator | |
7628 | * plenty of times and not have found | |
7629 | * anything, so we are likely way too | |
7630 | * fragmented for the clustering stuff to find | |
a5f6f719 AO |
7631 | * anything. |
7632 | * | |
7633 | * However, if the cluster is taken from the | |
7634 | * current block group, release the cluster | |
7635 | * first, so that we stand a better chance of | |
7636 | * succeeding in the unclustered | |
7637 | * allocation. */ | |
7638 | if (loop >= LOOP_NO_EMPTY_SIZE && | |
e570fd27 | 7639 | used_block_group != block_group) { |
062c05c4 | 7640 | spin_unlock(&last_ptr->refill_lock); |
e570fd27 MX |
7641 | btrfs_release_block_group(used_block_group, |
7642 | delalloc); | |
062c05c4 AO |
7643 | goto unclustered_alloc; |
7644 | } | |
7645 | ||
fa9c0d79 CM |
7646 | /* |
7647 | * this cluster didn't work out, free it and | |
7648 | * start over | |
7649 | */ | |
7650 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
7651 | ||
e570fd27 MX |
7652 | if (used_block_group != block_group) |
7653 | btrfs_release_block_group(used_block_group, | |
7654 | delalloc); | |
7655 | refill_cluster: | |
a5f6f719 AO |
7656 | if (loop >= LOOP_NO_EMPTY_SIZE) { |
7657 | spin_unlock(&last_ptr->refill_lock); | |
7658 | goto unclustered_alloc; | |
7659 | } | |
7660 | ||
8de972b4 CM |
7661 | aligned_cluster = max_t(unsigned long, |
7662 | empty_cluster + empty_size, | |
7663 | block_group->full_stripe_len); | |
7664 | ||
fa9c0d79 | 7665 | /* allocate a cluster in this block group */ |
2ff7e61e | 7666 | ret = btrfs_find_space_cluster(fs_info, block_group, |
00361589 JB |
7667 | last_ptr, search_start, |
7668 | num_bytes, | |
7669 | aligned_cluster); | |
fa9c0d79 CM |
7670 | if (ret == 0) { |
7671 | /* | |
7672 | * now pull our allocation out of this | |
7673 | * cluster | |
7674 | */ | |
7675 | offset = btrfs_alloc_from_cluster(block_group, | |
a4820398 MX |
7676 | last_ptr, |
7677 | num_bytes, | |
7678 | search_start, | |
7679 | &max_extent_size); | |
fa9c0d79 CM |
7680 | if (offset) { |
7681 | /* we found one, proceed */ | |
7682 | spin_unlock(&last_ptr->refill_lock); | |
3dca5c94 | 7683 | trace_btrfs_reserve_extent_cluster( |
3f7de037 JB |
7684 | block_group, search_start, |
7685 | num_bytes); | |
fa9c0d79 CM |
7686 | goto checks; |
7687 | } | |
0a24325e JB |
7688 | } else if (!cached && loop > LOOP_CACHING_NOWAIT |
7689 | && !failed_cluster_refill) { | |
817d52f8 JB |
7690 | spin_unlock(&last_ptr->refill_lock); |
7691 | ||
0a24325e | 7692 | failed_cluster_refill = true; |
817d52f8 JB |
7693 | wait_block_group_cache_progress(block_group, |
7694 | num_bytes + empty_cluster + empty_size); | |
7695 | goto have_block_group; | |
fa9c0d79 | 7696 | } |
817d52f8 | 7697 | |
fa9c0d79 CM |
7698 | /* |
7699 | * at this point we either didn't find a cluster | |
7700 | * or we weren't able to allocate a block from our | |
7701 | * cluster. Free the cluster we've been trying | |
7702 | * to use, and go to the next block group | |
7703 | */ | |
0a24325e | 7704 | btrfs_return_cluster_to_free_space(NULL, last_ptr); |
fa9c0d79 | 7705 | spin_unlock(&last_ptr->refill_lock); |
0a24325e | 7706 | goto loop; |
fa9c0d79 CM |
7707 | } |
7708 | ||
062c05c4 | 7709 | unclustered_alloc: |
c759c4e1 JB |
7710 | /* |
7711 | * We are doing an unclustered alloc, set the fragmented flag so | |
7712 | * we don't bother trying to setup a cluster again until we get | |
7713 | * more space. | |
7714 | */ | |
7715 | if (unlikely(last_ptr)) { | |
7716 | spin_lock(&last_ptr->lock); | |
7717 | last_ptr->fragmented = 1; | |
7718 | spin_unlock(&last_ptr->lock); | |
7719 | } | |
0c9b36e0 LB |
7720 | if (cached) { |
7721 | struct btrfs_free_space_ctl *ctl = | |
7722 | block_group->free_space_ctl; | |
7723 | ||
7724 | spin_lock(&ctl->tree_lock); | |
7725 | if (ctl->free_space < | |
7726 | num_bytes + empty_cluster + empty_size) { | |
7727 | if (ctl->free_space > max_extent_size) | |
7728 | max_extent_size = ctl->free_space; | |
7729 | spin_unlock(&ctl->tree_lock); | |
7730 | goto loop; | |
7731 | } | |
7732 | spin_unlock(&ctl->tree_lock); | |
a5f6f719 | 7733 | } |
a5f6f719 | 7734 | |
6226cb0a | 7735 | offset = btrfs_find_space_for_alloc(block_group, search_start, |
a4820398 MX |
7736 | num_bytes, empty_size, |
7737 | &max_extent_size); | |
1cdda9b8 JB |
7738 | /* |
7739 | * If we didn't find a chunk, and we haven't failed on this | |
7740 | * block group before, and this block group is in the middle of | |
7741 | * caching and we are ok with waiting, then go ahead and wait | |
7742 | * for progress to be made, and set failed_alloc to true. | |
7743 | * | |
7744 | * If failed_alloc is true then we've already waited on this | |
7745 | * block group once and should move on to the next block group. | |
7746 | */ | |
7747 | if (!offset && !failed_alloc && !cached && | |
7748 | loop > LOOP_CACHING_NOWAIT) { | |
817d52f8 | 7749 | wait_block_group_cache_progress(block_group, |
1cdda9b8 JB |
7750 | num_bytes + empty_size); |
7751 | failed_alloc = true; | |
817d52f8 | 7752 | goto have_block_group; |
1cdda9b8 JB |
7753 | } else if (!offset) { |
7754 | goto loop; | |
817d52f8 | 7755 | } |
fa9c0d79 | 7756 | checks: |
0b246afa | 7757 | search_start = ALIGN(offset, fs_info->stripesize); |
25179201 | 7758 | |
2552d17e JB |
7759 | /* move on to the next group */ |
7760 | if (search_start + num_bytes > | |
215a63d1 MX |
7761 | block_group->key.objectid + block_group->key.offset) { |
7762 | btrfs_add_free_space(block_group, offset, num_bytes); | |
2552d17e | 7763 | goto loop; |
6226cb0a | 7764 | } |
f5a31e16 | 7765 | |
f0486c68 | 7766 | if (offset < search_start) |
215a63d1 | 7767 | btrfs_add_free_space(block_group, offset, |
f0486c68 YZ |
7768 | search_start - offset); |
7769 | BUG_ON(offset > search_start); | |
2552d17e | 7770 | |
18513091 WX |
7771 | ret = btrfs_add_reserved_bytes(block_group, ram_bytes, |
7772 | num_bytes, delalloc); | |
f0486c68 | 7773 | if (ret == -EAGAIN) { |
215a63d1 | 7774 | btrfs_add_free_space(block_group, offset, num_bytes); |
2552d17e | 7775 | goto loop; |
0f9dd46c | 7776 | } |
9cfa3e34 | 7777 | btrfs_inc_block_group_reservations(block_group); |
0b86a832 | 7778 | |
f0486c68 | 7779 | /* we are all good, lets return */ |
2552d17e JB |
7780 | ins->objectid = search_start; |
7781 | ins->offset = num_bytes; | |
d2fb3437 | 7782 | |
3dca5c94 | 7783 | trace_btrfs_reserve_extent(block_group, search_start, num_bytes); |
e570fd27 | 7784 | btrfs_release_block_group(block_group, delalloc); |
2552d17e JB |
7785 | break; |
7786 | loop: | |
0a24325e | 7787 | failed_cluster_refill = false; |
1cdda9b8 | 7788 | failed_alloc = false; |
3e72ee88 QW |
7789 | BUG_ON(btrfs_bg_flags_to_raid_index(block_group->flags) != |
7790 | index); | |
e570fd27 | 7791 | btrfs_release_block_group(block_group, delalloc); |
14443937 | 7792 | cond_resched(); |
2552d17e JB |
7793 | } |
7794 | up_read(&space_info->groups_sem); | |
7795 | ||
13a0db5a | 7796 | if ((loop == LOOP_CACHING_NOWAIT) && have_caching_bg |
7797 | && !orig_have_caching_bg) | |
7798 | orig_have_caching_bg = true; | |
7799 | ||
60d2adbb MX |
7800 | if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg) |
7801 | goto search; | |
7802 | ||
b742bb82 YZ |
7803 | if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES) |
7804 | goto search; | |
7805 | ||
285ff5af | 7806 | /* |
ccf0e725 JB |
7807 | * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking |
7808 | * caching kthreads as we move along | |
817d52f8 JB |
7809 | * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching |
7810 | * LOOP_ALLOC_CHUNK, force a chunk allocation and try again | |
7811 | * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try | |
7812 | * again | |
fa9c0d79 | 7813 | */ |
723bda20 | 7814 | if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) { |
b742bb82 | 7815 | index = 0; |
a5e681d9 JB |
7816 | if (loop == LOOP_CACHING_NOWAIT) { |
7817 | /* | |
7818 | * We want to skip the LOOP_CACHING_WAIT step if we | |
01327610 | 7819 | * don't have any uncached bgs and we've already done a |
a5e681d9 JB |
7820 | * full search through. |
7821 | */ | |
13a0db5a | 7822 | if (orig_have_caching_bg || !full_search) |
a5e681d9 JB |
7823 | loop = LOOP_CACHING_WAIT; |
7824 | else | |
7825 | loop = LOOP_ALLOC_CHUNK; | |
7826 | } else { | |
7827 | loop++; | |
7828 | } | |
7829 | ||
817d52f8 | 7830 | if (loop == LOOP_ALLOC_CHUNK) { |
00361589 | 7831 | struct btrfs_trans_handle *trans; |
f017f15f WS |
7832 | int exist = 0; |
7833 | ||
7834 | trans = current->journal_info; | |
7835 | if (trans) | |
7836 | exist = 1; | |
7837 | else | |
7838 | trans = btrfs_join_transaction(root); | |
00361589 | 7839 | |
00361589 JB |
7840 | if (IS_ERR(trans)) { |
7841 | ret = PTR_ERR(trans); | |
7842 | goto out; | |
7843 | } | |
7844 | ||
01458828 | 7845 | ret = do_chunk_alloc(trans, flags, CHUNK_ALLOC_FORCE); |
a5e681d9 JB |
7846 | |
7847 | /* | |
7848 | * If we can't allocate a new chunk we've already looped | |
7849 | * through at least once, move on to the NO_EMPTY_SIZE | |
7850 | * case. | |
7851 | */ | |
7852 | if (ret == -ENOSPC) | |
7853 | loop = LOOP_NO_EMPTY_SIZE; | |
7854 | ||
ea658bad JB |
7855 | /* |
7856 | * Do not bail out on ENOSPC since we | |
7857 | * can do more things. | |
7858 | */ | |
00361589 | 7859 | if (ret < 0 && ret != -ENOSPC) |
66642832 | 7860 | btrfs_abort_transaction(trans, ret); |
00361589 JB |
7861 | else |
7862 | ret = 0; | |
f017f15f | 7863 | if (!exist) |
3a45bb20 | 7864 | btrfs_end_transaction(trans); |
00361589 | 7865 | if (ret) |
ea658bad | 7866 | goto out; |
2552d17e JB |
7867 | } |
7868 | ||
723bda20 | 7869 | if (loop == LOOP_NO_EMPTY_SIZE) { |
a5e681d9 JB |
7870 | /* |
7871 | * Don't loop again if we already have no empty_size and | |
7872 | * no empty_cluster. | |
7873 | */ | |
7874 | if (empty_size == 0 && | |
7875 | empty_cluster == 0) { | |
7876 | ret = -ENOSPC; | |
7877 | goto out; | |
7878 | } | |
723bda20 JB |
7879 | empty_size = 0; |
7880 | empty_cluster = 0; | |
fa9c0d79 | 7881 | } |
723bda20 JB |
7882 | |
7883 | goto search; | |
2552d17e JB |
7884 | } else if (!ins->objectid) { |
7885 | ret = -ENOSPC; | |
d82a6f1d | 7886 | } else if (ins->objectid) { |
c759c4e1 JB |
7887 | if (!use_cluster && last_ptr) { |
7888 | spin_lock(&last_ptr->lock); | |
7889 | last_ptr->window_start = ins->objectid; | |
7890 | spin_unlock(&last_ptr->lock); | |
7891 | } | |
80eb234a | 7892 | ret = 0; |
be744175 | 7893 | } |
79787eaa | 7894 | out: |
4f4db217 JB |
7895 | if (ret == -ENOSPC) { |
7896 | spin_lock(&space_info->lock); | |
7897 | space_info->max_extent_size = max_extent_size; | |
7898 | spin_unlock(&space_info->lock); | |
a4820398 | 7899 | ins->offset = max_extent_size; |
4f4db217 | 7900 | } |
0f70abe2 | 7901 | return ret; |
fec577fb | 7902 | } |
ec44a35c | 7903 | |
ab8d0fc4 JM |
7904 | static void dump_space_info(struct btrfs_fs_info *fs_info, |
7905 | struct btrfs_space_info *info, u64 bytes, | |
9ed74f2d | 7906 | int dump_block_groups) |
0f9dd46c JB |
7907 | { |
7908 | struct btrfs_block_group_cache *cache; | |
b742bb82 | 7909 | int index = 0; |
0f9dd46c | 7910 | |
9ed74f2d | 7911 | spin_lock(&info->lock); |
ab8d0fc4 JM |
7912 | btrfs_info(fs_info, "space_info %llu has %llu free, is %sfull", |
7913 | info->flags, | |
4136135b LB |
7914 | info->total_bytes - btrfs_space_info_used(info, true), |
7915 | info->full ? "" : "not "); | |
ab8d0fc4 JM |
7916 | btrfs_info(fs_info, |
7917 | "space_info total=%llu, used=%llu, pinned=%llu, reserved=%llu, may_use=%llu, readonly=%llu", | |
7918 | info->total_bytes, info->bytes_used, info->bytes_pinned, | |
7919 | info->bytes_reserved, info->bytes_may_use, | |
7920 | info->bytes_readonly); | |
9ed74f2d JB |
7921 | spin_unlock(&info->lock); |
7922 | ||
7923 | if (!dump_block_groups) | |
7924 | return; | |
0f9dd46c | 7925 | |
80eb234a | 7926 | down_read(&info->groups_sem); |
b742bb82 YZ |
7927 | again: |
7928 | list_for_each_entry(cache, &info->block_groups[index], list) { | |
0f9dd46c | 7929 | spin_lock(&cache->lock); |
ab8d0fc4 JM |
7930 | btrfs_info(fs_info, |
7931 | "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s", | |
7932 | cache->key.objectid, cache->key.offset, | |
7933 | btrfs_block_group_used(&cache->item), cache->pinned, | |
7934 | cache->reserved, cache->ro ? "[readonly]" : ""); | |
0f9dd46c JB |
7935 | btrfs_dump_free_space(cache, bytes); |
7936 | spin_unlock(&cache->lock); | |
7937 | } | |
b742bb82 YZ |
7938 | if (++index < BTRFS_NR_RAID_TYPES) |
7939 | goto again; | |
80eb234a | 7940 | up_read(&info->groups_sem); |
0f9dd46c | 7941 | } |
e8569813 | 7942 | |
6f47c706 NB |
7943 | /* |
7944 | * btrfs_reserve_extent - entry point to the extent allocator. Tries to find a | |
7945 | * hole that is at least as big as @num_bytes. | |
7946 | * | |
7947 | * @root - The root that will contain this extent | |
7948 | * | |
7949 | * @ram_bytes - The amount of space in ram that @num_bytes take. This | |
7950 | * is used for accounting purposes. This value differs | |
7951 | * from @num_bytes only in the case of compressed extents. | |
7952 | * | |
7953 | * @num_bytes - Number of bytes to allocate on-disk. | |
7954 | * | |
7955 | * @min_alloc_size - Indicates the minimum amount of space that the | |
7956 | * allocator should try to satisfy. In some cases | |
7957 | * @num_bytes may be larger than what is required and if | |
7958 | * the filesystem is fragmented then allocation fails. | |
7959 | * However, the presence of @min_alloc_size gives a | |
7960 | * chance to try and satisfy the smaller allocation. | |
7961 | * | |
7962 | * @empty_size - A hint that you plan on doing more COW. This is the | |
7963 | * size in bytes the allocator should try to find free | |
7964 | * next to the block it returns. This is just a hint and | |
7965 | * may be ignored by the allocator. | |
7966 | * | |
7967 | * @hint_byte - Hint to the allocator to start searching above the byte | |
7968 | * address passed. It might be ignored. | |
7969 | * | |
7970 | * @ins - This key is modified to record the found hole. It will | |
7971 | * have the following values: | |
7972 | * ins->objectid == start position | |
7973 | * ins->flags = BTRFS_EXTENT_ITEM_KEY | |
7974 | * ins->offset == the size of the hole. | |
7975 | * | |
7976 | * @is_data - Boolean flag indicating whether an extent is | |
7977 | * allocated for data (true) or metadata (false) | |
7978 | * | |
7979 | * @delalloc - Boolean flag indicating whether this allocation is for | |
7980 | * delalloc or not. If 'true' data_rwsem of block groups | |
7981 | * is going to be acquired. | |
7982 | * | |
7983 | * | |
7984 | * Returns 0 when an allocation succeeded or < 0 when an error occurred. In | |
7985 | * case -ENOSPC is returned then @ins->offset will contain the size of the | |
7986 | * largest available hole the allocator managed to find. | |
7987 | */ | |
18513091 | 7988 | int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes, |
11833d66 YZ |
7989 | u64 num_bytes, u64 min_alloc_size, |
7990 | u64 empty_size, u64 hint_byte, | |
e570fd27 | 7991 | struct btrfs_key *ins, int is_data, int delalloc) |
fec577fb | 7992 | { |
ab8d0fc4 | 7993 | struct btrfs_fs_info *fs_info = root->fs_info; |
36af4e07 | 7994 | bool final_tried = num_bytes == min_alloc_size; |
b6919a58 | 7995 | u64 flags; |
fec577fb | 7996 | int ret; |
925baedd | 7997 | |
1b86826d | 7998 | flags = get_alloc_profile_by_root(root, is_data); |
98d20f67 | 7999 | again: |
0b246afa | 8000 | WARN_ON(num_bytes < fs_info->sectorsize); |
87bde3cd | 8001 | ret = find_free_extent(fs_info, ram_bytes, num_bytes, empty_size, |
18513091 | 8002 | hint_byte, ins, flags, delalloc); |
9cfa3e34 | 8003 | if (!ret && !is_data) { |
ab8d0fc4 | 8004 | btrfs_dec_block_group_reservations(fs_info, ins->objectid); |
9cfa3e34 | 8005 | } else if (ret == -ENOSPC) { |
a4820398 MX |
8006 | if (!final_tried && ins->offset) { |
8007 | num_bytes = min(num_bytes >> 1, ins->offset); | |
da17066c | 8008 | num_bytes = round_down(num_bytes, |
0b246afa | 8009 | fs_info->sectorsize); |
9e622d6b | 8010 | num_bytes = max(num_bytes, min_alloc_size); |
18513091 | 8011 | ram_bytes = num_bytes; |
9e622d6b MX |
8012 | if (num_bytes == min_alloc_size) |
8013 | final_tried = true; | |
8014 | goto again; | |
ab8d0fc4 | 8015 | } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
9e622d6b MX |
8016 | struct btrfs_space_info *sinfo; |
8017 | ||
ab8d0fc4 | 8018 | sinfo = __find_space_info(fs_info, flags); |
0b246afa | 8019 | btrfs_err(fs_info, |
5d163e0e JM |
8020 | "allocation failed flags %llu, wanted %llu", |
8021 | flags, num_bytes); | |
53804280 | 8022 | if (sinfo) |
ab8d0fc4 | 8023 | dump_space_info(fs_info, sinfo, num_bytes, 1); |
9e622d6b | 8024 | } |
925baedd | 8025 | } |
0f9dd46c JB |
8026 | |
8027 | return ret; | |
e6dcd2dc CM |
8028 | } |
8029 | ||
2ff7e61e | 8030 | static int __btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info, |
e570fd27 MX |
8031 | u64 start, u64 len, |
8032 | int pin, int delalloc) | |
65b51a00 | 8033 | { |
0f9dd46c | 8034 | struct btrfs_block_group_cache *cache; |
1f3c79a2 | 8035 | int ret = 0; |
0f9dd46c | 8036 | |
0b246afa | 8037 | cache = btrfs_lookup_block_group(fs_info, start); |
0f9dd46c | 8038 | if (!cache) { |
0b246afa JM |
8039 | btrfs_err(fs_info, "Unable to find block group for %llu", |
8040 | start); | |
0f9dd46c JB |
8041 | return -ENOSPC; |
8042 | } | |
1f3c79a2 | 8043 | |
e688b725 | 8044 | if (pin) |
2ff7e61e | 8045 | pin_down_extent(fs_info, cache, start, len, 1); |
e688b725 | 8046 | else { |
0b246afa | 8047 | if (btrfs_test_opt(fs_info, DISCARD)) |
2ff7e61e | 8048 | ret = btrfs_discard_extent(fs_info, start, len, NULL); |
e688b725 | 8049 | btrfs_add_free_space(cache, start, len); |
4824f1f4 | 8050 | btrfs_free_reserved_bytes(cache, len, delalloc); |
71ff6437 | 8051 | trace_btrfs_reserved_extent_free(fs_info, start, len); |
e688b725 | 8052 | } |
31193213 | 8053 | |
fa9c0d79 | 8054 | btrfs_put_block_group(cache); |
e6dcd2dc CM |
8055 | return ret; |
8056 | } | |
8057 | ||
2ff7e61e | 8058 | int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info, |
e570fd27 | 8059 | u64 start, u64 len, int delalloc) |
e688b725 | 8060 | { |
2ff7e61e | 8061 | return __btrfs_free_reserved_extent(fs_info, start, len, 0, delalloc); |
e688b725 CM |
8062 | } |
8063 | ||
2ff7e61e | 8064 | int btrfs_free_and_pin_reserved_extent(struct btrfs_fs_info *fs_info, |
e688b725 CM |
8065 | u64 start, u64 len) |
8066 | { | |
2ff7e61e | 8067 | return __btrfs_free_reserved_extent(fs_info, start, len, 1, 0); |
e688b725 CM |
8068 | } |
8069 | ||
5d4f98a2 | 8070 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
8071 | u64 parent, u64 root_objectid, |
8072 | u64 flags, u64 owner, u64 offset, | |
8073 | struct btrfs_key *ins, int ref_mod) | |
e6dcd2dc | 8074 | { |
ef89b824 | 8075 | struct btrfs_fs_info *fs_info = trans->fs_info; |
e6dcd2dc | 8076 | int ret; |
e6dcd2dc | 8077 | struct btrfs_extent_item *extent_item; |
5d4f98a2 | 8078 | struct btrfs_extent_inline_ref *iref; |
e6dcd2dc | 8079 | struct btrfs_path *path; |
5d4f98a2 YZ |
8080 | struct extent_buffer *leaf; |
8081 | int type; | |
8082 | u32 size; | |
26b8003f | 8083 | |
5d4f98a2 YZ |
8084 | if (parent > 0) |
8085 | type = BTRFS_SHARED_DATA_REF_KEY; | |
8086 | else | |
8087 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
58176a96 | 8088 | |
5d4f98a2 | 8089 | size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type); |
7bb86316 CM |
8090 | |
8091 | path = btrfs_alloc_path(); | |
db5b493a TI |
8092 | if (!path) |
8093 | return -ENOMEM; | |
47e4bb98 | 8094 | |
b9473439 | 8095 | path->leave_spinning = 1; |
5d4f98a2 YZ |
8096 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, |
8097 | ins, size); | |
79787eaa JM |
8098 | if (ret) { |
8099 | btrfs_free_path(path); | |
8100 | return ret; | |
8101 | } | |
0f9dd46c | 8102 | |
5d4f98a2 YZ |
8103 | leaf = path->nodes[0]; |
8104 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
47e4bb98 | 8105 | struct btrfs_extent_item); |
5d4f98a2 YZ |
8106 | btrfs_set_extent_refs(leaf, extent_item, ref_mod); |
8107 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
8108 | btrfs_set_extent_flags(leaf, extent_item, | |
8109 | flags | BTRFS_EXTENT_FLAG_DATA); | |
8110 | ||
8111 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
8112 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
8113 | if (parent > 0) { | |
8114 | struct btrfs_shared_data_ref *ref; | |
8115 | ref = (struct btrfs_shared_data_ref *)(iref + 1); | |
8116 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
8117 | btrfs_set_shared_data_ref_count(leaf, ref, ref_mod); | |
8118 | } else { | |
8119 | struct btrfs_extent_data_ref *ref; | |
8120 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
8121 | btrfs_set_extent_data_ref_root(leaf, ref, root_objectid); | |
8122 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
8123 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
8124 | btrfs_set_extent_data_ref_count(leaf, ref, ref_mod); | |
8125 | } | |
47e4bb98 CM |
8126 | |
8127 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
7bb86316 | 8128 | btrfs_free_path(path); |
f510cfec | 8129 | |
25a356d3 | 8130 | ret = remove_from_free_space_tree(trans, ins->objectid, ins->offset); |
1e144fb8 OS |
8131 | if (ret) |
8132 | return ret; | |
8133 | ||
6202df69 | 8134 | ret = update_block_group(trans, fs_info, ins->objectid, ins->offset, 1); |
79787eaa | 8135 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb | 8136 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
c1c9ff7c | 8137 | ins->objectid, ins->offset); |
f5947066 CM |
8138 | BUG(); |
8139 | } | |
71ff6437 | 8140 | trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid, ins->offset); |
e6dcd2dc CM |
8141 | return ret; |
8142 | } | |
8143 | ||
5d4f98a2 | 8144 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, |
4e6bd4e0 | 8145 | struct btrfs_delayed_ref_node *node, |
21ebfbe7 | 8146 | struct btrfs_delayed_extent_op *extent_op) |
e6dcd2dc | 8147 | { |
9dcdbe01 | 8148 | struct btrfs_fs_info *fs_info = trans->fs_info; |
e6dcd2dc | 8149 | int ret; |
5d4f98a2 | 8150 | struct btrfs_extent_item *extent_item; |
4e6bd4e0 | 8151 | struct btrfs_key extent_key; |
5d4f98a2 YZ |
8152 | struct btrfs_tree_block_info *block_info; |
8153 | struct btrfs_extent_inline_ref *iref; | |
8154 | struct btrfs_path *path; | |
8155 | struct extent_buffer *leaf; | |
4e6bd4e0 | 8156 | struct btrfs_delayed_tree_ref *ref; |
3173a18f | 8157 | u32 size = sizeof(*extent_item) + sizeof(*iref); |
4e6bd4e0 | 8158 | u64 num_bytes; |
21ebfbe7 | 8159 | u64 flags = extent_op->flags_to_set; |
0b246afa | 8160 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); |
3173a18f | 8161 | |
4e6bd4e0 NB |
8162 | ref = btrfs_delayed_node_to_tree_ref(node); |
8163 | ||
4e6bd4e0 NB |
8164 | extent_key.objectid = node->bytenr; |
8165 | if (skinny_metadata) { | |
8166 | extent_key.offset = ref->level; | |
8167 | extent_key.type = BTRFS_METADATA_ITEM_KEY; | |
8168 | num_bytes = fs_info->nodesize; | |
8169 | } else { | |
8170 | extent_key.offset = node->num_bytes; | |
8171 | extent_key.type = BTRFS_EXTENT_ITEM_KEY; | |
3173a18f | 8172 | size += sizeof(*block_info); |
4e6bd4e0 NB |
8173 | num_bytes = node->num_bytes; |
8174 | } | |
1c2308f8 | 8175 | |
5d4f98a2 | 8176 | path = btrfs_alloc_path(); |
857cc2fc | 8177 | if (!path) { |
4e6bd4e0 NB |
8178 | btrfs_free_and_pin_reserved_extent(fs_info, |
8179 | extent_key.objectid, | |
0b246afa | 8180 | fs_info->nodesize); |
d8926bb3 | 8181 | return -ENOMEM; |
857cc2fc | 8182 | } |
56bec294 | 8183 | |
5d4f98a2 YZ |
8184 | path->leave_spinning = 1; |
8185 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, | |
4e6bd4e0 | 8186 | &extent_key, size); |
79787eaa | 8187 | if (ret) { |
dd825259 | 8188 | btrfs_free_path(path); |
4e6bd4e0 NB |
8189 | btrfs_free_and_pin_reserved_extent(fs_info, |
8190 | extent_key.objectid, | |
0b246afa | 8191 | fs_info->nodesize); |
79787eaa JM |
8192 | return ret; |
8193 | } | |
5d4f98a2 YZ |
8194 | |
8195 | leaf = path->nodes[0]; | |
8196 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
8197 | struct btrfs_extent_item); | |
8198 | btrfs_set_extent_refs(leaf, extent_item, 1); | |
8199 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
8200 | btrfs_set_extent_flags(leaf, extent_item, | |
8201 | flags | BTRFS_EXTENT_FLAG_TREE_BLOCK); | |
5d4f98a2 | 8202 | |
3173a18f JB |
8203 | if (skinny_metadata) { |
8204 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
8205 | } else { | |
8206 | block_info = (struct btrfs_tree_block_info *)(extent_item + 1); | |
21ebfbe7 | 8207 | btrfs_set_tree_block_key(leaf, block_info, &extent_op->key); |
4e6bd4e0 | 8208 | btrfs_set_tree_block_level(leaf, block_info, ref->level); |
3173a18f JB |
8209 | iref = (struct btrfs_extent_inline_ref *)(block_info + 1); |
8210 | } | |
5d4f98a2 | 8211 | |
d4b20733 | 8212 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) { |
5d4f98a2 YZ |
8213 | BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); |
8214 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
8215 | BTRFS_SHARED_BLOCK_REF_KEY); | |
d4b20733 | 8216 | btrfs_set_extent_inline_ref_offset(leaf, iref, ref->parent); |
5d4f98a2 YZ |
8217 | } else { |
8218 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
8219 | BTRFS_TREE_BLOCK_REF_KEY); | |
4e6bd4e0 | 8220 | btrfs_set_extent_inline_ref_offset(leaf, iref, ref->root); |
5d4f98a2 YZ |
8221 | } |
8222 | ||
8223 | btrfs_mark_buffer_dirty(leaf); | |
8224 | btrfs_free_path(path); | |
8225 | ||
4e6bd4e0 NB |
8226 | ret = remove_from_free_space_tree(trans, extent_key.objectid, |
8227 | num_bytes); | |
1e144fb8 OS |
8228 | if (ret) |
8229 | return ret; | |
8230 | ||
4e6bd4e0 | 8231 | ret = update_block_group(trans, fs_info, extent_key.objectid, |
6202df69 | 8232 | fs_info->nodesize, 1); |
79787eaa | 8233 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb | 8234 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
4e6bd4e0 | 8235 | extent_key.objectid, extent_key.offset); |
5d4f98a2 YZ |
8236 | BUG(); |
8237 | } | |
0be5dc67 | 8238 | |
4e6bd4e0 | 8239 | trace_btrfs_reserved_extent_alloc(fs_info, extent_key.objectid, |
0b246afa | 8240 | fs_info->nodesize); |
5d4f98a2 YZ |
8241 | return ret; |
8242 | } | |
8243 | ||
8244 | int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
84f7d8e6 | 8245 | struct btrfs_root *root, u64 owner, |
5846a3c2 QW |
8246 | u64 offset, u64 ram_bytes, |
8247 | struct btrfs_key *ins) | |
5d4f98a2 YZ |
8248 | { |
8249 | int ret; | |
8250 | ||
84f7d8e6 | 8251 | BUG_ON(root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID); |
5d4f98a2 | 8252 | |
fd708b81 JB |
8253 | btrfs_ref_tree_mod(root, ins->objectid, ins->offset, 0, |
8254 | root->root_key.objectid, owner, offset, | |
8255 | BTRFS_ADD_DELAYED_EXTENT); | |
8256 | ||
88a979c6 | 8257 | ret = btrfs_add_delayed_data_ref(trans, ins->objectid, |
84f7d8e6 JB |
8258 | ins->offset, 0, |
8259 | root->root_key.objectid, owner, | |
7be07912 OS |
8260 | offset, ram_bytes, |
8261 | BTRFS_ADD_DELAYED_EXTENT, NULL, NULL); | |
e6dcd2dc CM |
8262 | return ret; |
8263 | } | |
e02119d5 CM |
8264 | |
8265 | /* | |
8266 | * this is used by the tree logging recovery code. It records that | |
8267 | * an extent has been allocated and makes sure to clear the free | |
8268 | * space cache bits as well | |
8269 | */ | |
5d4f98a2 | 8270 | int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans, |
2ff7e61e | 8271 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
8272 | u64 root_objectid, u64 owner, u64 offset, |
8273 | struct btrfs_key *ins) | |
e02119d5 CM |
8274 | { |
8275 | int ret; | |
8276 | struct btrfs_block_group_cache *block_group; | |
ed7a6948 | 8277 | struct btrfs_space_info *space_info; |
11833d66 | 8278 | |
8c2a1a30 JB |
8279 | /* |
8280 | * Mixed block groups will exclude before processing the log so we only | |
01327610 | 8281 | * need to do the exclude dance if this fs isn't mixed. |
8c2a1a30 | 8282 | */ |
0b246afa | 8283 | if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) { |
2ff7e61e JM |
8284 | ret = __exclude_logged_extent(fs_info, ins->objectid, |
8285 | ins->offset); | |
b50c6e25 | 8286 | if (ret) |
8c2a1a30 | 8287 | return ret; |
11833d66 YZ |
8288 | } |
8289 | ||
0b246afa | 8290 | block_group = btrfs_lookup_block_group(fs_info, ins->objectid); |
8c2a1a30 JB |
8291 | if (!block_group) |
8292 | return -EINVAL; | |
8293 | ||
ed7a6948 WX |
8294 | space_info = block_group->space_info; |
8295 | spin_lock(&space_info->lock); | |
8296 | spin_lock(&block_group->lock); | |
8297 | space_info->bytes_reserved += ins->offset; | |
8298 | block_group->reserved += ins->offset; | |
8299 | spin_unlock(&block_group->lock); | |
8300 | spin_unlock(&space_info->lock); | |
8301 | ||
ef89b824 NB |
8302 | ret = alloc_reserved_file_extent(trans, 0, root_objectid, 0, owner, |
8303 | offset, ins, 1); | |
b50c6e25 | 8304 | btrfs_put_block_group(block_group); |
e02119d5 CM |
8305 | return ret; |
8306 | } | |
8307 | ||
48a3b636 ES |
8308 | static struct extent_buffer * |
8309 | btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
fe864576 | 8310 | u64 bytenr, int level) |
65b51a00 | 8311 | { |
0b246afa | 8312 | struct btrfs_fs_info *fs_info = root->fs_info; |
65b51a00 CM |
8313 | struct extent_buffer *buf; |
8314 | ||
2ff7e61e | 8315 | buf = btrfs_find_create_tree_block(fs_info, bytenr); |
c871b0f2 LB |
8316 | if (IS_ERR(buf)) |
8317 | return buf; | |
8318 | ||
65b51a00 | 8319 | btrfs_set_header_generation(buf, trans->transid); |
85d4e461 | 8320 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level); |
65b51a00 | 8321 | btrfs_tree_lock(buf); |
7c302b49 | 8322 | clean_tree_block(fs_info, buf); |
3083ee2e | 8323 | clear_bit(EXTENT_BUFFER_STALE, &buf->bflags); |
b4ce94de CM |
8324 | |
8325 | btrfs_set_lock_blocking(buf); | |
4db8c528 | 8326 | set_extent_buffer_uptodate(buf); |
b4ce94de | 8327 | |
d0c803c4 | 8328 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { |
656f30db | 8329 | buf->log_index = root->log_transid % 2; |
8cef4e16 YZ |
8330 | /* |
8331 | * we allow two log transactions at a time, use different | |
8332 | * EXENT bit to differentiate dirty pages. | |
8333 | */ | |
656f30db | 8334 | if (buf->log_index == 0) |
8cef4e16 YZ |
8335 | set_extent_dirty(&root->dirty_log_pages, buf->start, |
8336 | buf->start + buf->len - 1, GFP_NOFS); | |
8337 | else | |
8338 | set_extent_new(&root->dirty_log_pages, buf->start, | |
3744dbeb | 8339 | buf->start + buf->len - 1); |
d0c803c4 | 8340 | } else { |
656f30db | 8341 | buf->log_index = -1; |
d0c803c4 | 8342 | set_extent_dirty(&trans->transaction->dirty_pages, buf->start, |
65b51a00 | 8343 | buf->start + buf->len - 1, GFP_NOFS); |
d0c803c4 | 8344 | } |
64c12921 | 8345 | trans->dirty = true; |
b4ce94de | 8346 | /* this returns a buffer locked for blocking */ |
65b51a00 CM |
8347 | return buf; |
8348 | } | |
8349 | ||
f0486c68 YZ |
8350 | static struct btrfs_block_rsv * |
8351 | use_block_rsv(struct btrfs_trans_handle *trans, | |
8352 | struct btrfs_root *root, u32 blocksize) | |
8353 | { | |
0b246afa | 8354 | struct btrfs_fs_info *fs_info = root->fs_info; |
f0486c68 | 8355 | struct btrfs_block_rsv *block_rsv; |
0b246afa | 8356 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; |
f0486c68 | 8357 | int ret; |
d88033db | 8358 | bool global_updated = false; |
f0486c68 YZ |
8359 | |
8360 | block_rsv = get_block_rsv(trans, root); | |
8361 | ||
b586b323 MX |
8362 | if (unlikely(block_rsv->size == 0)) |
8363 | goto try_reserve; | |
d88033db | 8364 | again: |
f0486c68 YZ |
8365 | ret = block_rsv_use_bytes(block_rsv, blocksize); |
8366 | if (!ret) | |
8367 | return block_rsv; | |
8368 | ||
b586b323 MX |
8369 | if (block_rsv->failfast) |
8370 | return ERR_PTR(ret); | |
8371 | ||
d88033db MX |
8372 | if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) { |
8373 | global_updated = true; | |
0b246afa | 8374 | update_global_block_rsv(fs_info); |
d88033db MX |
8375 | goto again; |
8376 | } | |
8377 | ||
0b246afa | 8378 | if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
b586b323 MX |
8379 | static DEFINE_RATELIMIT_STATE(_rs, |
8380 | DEFAULT_RATELIMIT_INTERVAL * 10, | |
8381 | /*DEFAULT_RATELIMIT_BURST*/ 1); | |
8382 | if (__ratelimit(&_rs)) | |
8383 | WARN(1, KERN_DEBUG | |
efe120a0 | 8384 | "BTRFS: block rsv returned %d\n", ret); |
b586b323 MX |
8385 | } |
8386 | try_reserve: | |
8387 | ret = reserve_metadata_bytes(root, block_rsv, blocksize, | |
8388 | BTRFS_RESERVE_NO_FLUSH); | |
8389 | if (!ret) | |
8390 | return block_rsv; | |
8391 | /* | |
8392 | * If we couldn't reserve metadata bytes try and use some from | |
5881cfc9 MX |
8393 | * the global reserve if its space type is the same as the global |
8394 | * reservation. | |
b586b323 | 8395 | */ |
5881cfc9 MX |
8396 | if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL && |
8397 | block_rsv->space_info == global_rsv->space_info) { | |
b586b323 MX |
8398 | ret = block_rsv_use_bytes(global_rsv, blocksize); |
8399 | if (!ret) | |
8400 | return global_rsv; | |
8401 | } | |
8402 | return ERR_PTR(ret); | |
f0486c68 YZ |
8403 | } |
8404 | ||
8c2a3ca2 JB |
8405 | static void unuse_block_rsv(struct btrfs_fs_info *fs_info, |
8406 | struct btrfs_block_rsv *block_rsv, u32 blocksize) | |
f0486c68 YZ |
8407 | { |
8408 | block_rsv_add_bytes(block_rsv, blocksize, 0); | |
ff6bc37e | 8409 | block_rsv_release_bytes(fs_info, block_rsv, NULL, 0, NULL); |
f0486c68 YZ |
8410 | } |
8411 | ||
fec577fb | 8412 | /* |
f0486c68 | 8413 | * finds a free extent and does all the dirty work required for allocation |
67b7859e | 8414 | * returns the tree buffer or an ERR_PTR on error. |
fec577fb | 8415 | */ |
4d75f8a9 | 8416 | struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans, |
310712b2 OS |
8417 | struct btrfs_root *root, |
8418 | u64 parent, u64 root_objectid, | |
8419 | const struct btrfs_disk_key *key, | |
8420 | int level, u64 hint, | |
8421 | u64 empty_size) | |
fec577fb | 8422 | { |
0b246afa | 8423 | struct btrfs_fs_info *fs_info = root->fs_info; |
e2fa7227 | 8424 | struct btrfs_key ins; |
f0486c68 | 8425 | struct btrfs_block_rsv *block_rsv; |
5f39d397 | 8426 | struct extent_buffer *buf; |
67b7859e | 8427 | struct btrfs_delayed_extent_op *extent_op; |
f0486c68 YZ |
8428 | u64 flags = 0; |
8429 | int ret; | |
0b246afa JM |
8430 | u32 blocksize = fs_info->nodesize; |
8431 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); | |
fec577fb | 8432 | |
05653ef3 | 8433 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
0b246afa | 8434 | if (btrfs_is_testing(fs_info)) { |
faa2dbf0 | 8435 | buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr, |
fe864576 | 8436 | level); |
faa2dbf0 JB |
8437 | if (!IS_ERR(buf)) |
8438 | root->alloc_bytenr += blocksize; | |
8439 | return buf; | |
8440 | } | |
05653ef3 | 8441 | #endif |
fccb84c9 | 8442 | |
f0486c68 YZ |
8443 | block_rsv = use_block_rsv(trans, root, blocksize); |
8444 | if (IS_ERR(block_rsv)) | |
8445 | return ERR_CAST(block_rsv); | |
8446 | ||
18513091 | 8447 | ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize, |
e570fd27 | 8448 | empty_size, hint, &ins, 0, 0); |
67b7859e OS |
8449 | if (ret) |
8450 | goto out_unuse; | |
55c69072 | 8451 | |
fe864576 | 8452 | buf = btrfs_init_new_buffer(trans, root, ins.objectid, level); |
67b7859e OS |
8453 | if (IS_ERR(buf)) { |
8454 | ret = PTR_ERR(buf); | |
8455 | goto out_free_reserved; | |
8456 | } | |
f0486c68 YZ |
8457 | |
8458 | if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
8459 | if (parent == 0) | |
8460 | parent = ins.objectid; | |
8461 | flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
8462 | } else | |
8463 | BUG_ON(parent > 0); | |
8464 | ||
8465 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) { | |
78a6184a | 8466 | extent_op = btrfs_alloc_delayed_extent_op(); |
67b7859e OS |
8467 | if (!extent_op) { |
8468 | ret = -ENOMEM; | |
8469 | goto out_free_buf; | |
8470 | } | |
f0486c68 YZ |
8471 | if (key) |
8472 | memcpy(&extent_op->key, key, sizeof(extent_op->key)); | |
8473 | else | |
8474 | memset(&extent_op->key, 0, sizeof(extent_op->key)); | |
8475 | extent_op->flags_to_set = flags; | |
35b3ad50 DS |
8476 | extent_op->update_key = skinny_metadata ? false : true; |
8477 | extent_op->update_flags = true; | |
8478 | extent_op->is_data = false; | |
b1c79e09 | 8479 | extent_op->level = level; |
f0486c68 | 8480 | |
fd708b81 JB |
8481 | btrfs_ref_tree_mod(root, ins.objectid, ins.offset, parent, |
8482 | root_objectid, level, 0, | |
8483 | BTRFS_ADD_DELAYED_EXTENT); | |
44e1c47d | 8484 | ret = btrfs_add_delayed_tree_ref(trans, ins.objectid, |
7be07912 OS |
8485 | ins.offset, parent, |
8486 | root_objectid, level, | |
67b7859e | 8487 | BTRFS_ADD_DELAYED_EXTENT, |
7be07912 | 8488 | extent_op, NULL, NULL); |
67b7859e OS |
8489 | if (ret) |
8490 | goto out_free_delayed; | |
f0486c68 | 8491 | } |
fec577fb | 8492 | return buf; |
67b7859e OS |
8493 | |
8494 | out_free_delayed: | |
8495 | btrfs_free_delayed_extent_op(extent_op); | |
8496 | out_free_buf: | |
8497 | free_extent_buffer(buf); | |
8498 | out_free_reserved: | |
2ff7e61e | 8499 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0); |
67b7859e | 8500 | out_unuse: |
0b246afa | 8501 | unuse_block_rsv(fs_info, block_rsv, blocksize); |
67b7859e | 8502 | return ERR_PTR(ret); |
fec577fb | 8503 | } |
a28ec197 | 8504 | |
2c47e605 YZ |
8505 | struct walk_control { |
8506 | u64 refs[BTRFS_MAX_LEVEL]; | |
8507 | u64 flags[BTRFS_MAX_LEVEL]; | |
8508 | struct btrfs_key update_progress; | |
8509 | int stage; | |
8510 | int level; | |
8511 | int shared_level; | |
8512 | int update_ref; | |
8513 | int keep_locks; | |
1c4850e2 YZ |
8514 | int reada_slot; |
8515 | int reada_count; | |
66d7e7f0 | 8516 | int for_reloc; |
2c47e605 YZ |
8517 | }; |
8518 | ||
8519 | #define DROP_REFERENCE 1 | |
8520 | #define UPDATE_BACKREF 2 | |
8521 | ||
1c4850e2 YZ |
8522 | static noinline void reada_walk_down(struct btrfs_trans_handle *trans, |
8523 | struct btrfs_root *root, | |
8524 | struct walk_control *wc, | |
8525 | struct btrfs_path *path) | |
6407bf6d | 8526 | { |
0b246afa | 8527 | struct btrfs_fs_info *fs_info = root->fs_info; |
1c4850e2 YZ |
8528 | u64 bytenr; |
8529 | u64 generation; | |
8530 | u64 refs; | |
94fcca9f | 8531 | u64 flags; |
5d4f98a2 | 8532 | u32 nritems; |
1c4850e2 YZ |
8533 | struct btrfs_key key; |
8534 | struct extent_buffer *eb; | |
6407bf6d | 8535 | int ret; |
1c4850e2 YZ |
8536 | int slot; |
8537 | int nread = 0; | |
6407bf6d | 8538 | |
1c4850e2 YZ |
8539 | if (path->slots[wc->level] < wc->reada_slot) { |
8540 | wc->reada_count = wc->reada_count * 2 / 3; | |
8541 | wc->reada_count = max(wc->reada_count, 2); | |
8542 | } else { | |
8543 | wc->reada_count = wc->reada_count * 3 / 2; | |
8544 | wc->reada_count = min_t(int, wc->reada_count, | |
0b246afa | 8545 | BTRFS_NODEPTRS_PER_BLOCK(fs_info)); |
1c4850e2 | 8546 | } |
7bb86316 | 8547 | |
1c4850e2 YZ |
8548 | eb = path->nodes[wc->level]; |
8549 | nritems = btrfs_header_nritems(eb); | |
bd56b302 | 8550 | |
1c4850e2 YZ |
8551 | for (slot = path->slots[wc->level]; slot < nritems; slot++) { |
8552 | if (nread >= wc->reada_count) | |
8553 | break; | |
bd56b302 | 8554 | |
2dd3e67b | 8555 | cond_resched(); |
1c4850e2 YZ |
8556 | bytenr = btrfs_node_blockptr(eb, slot); |
8557 | generation = btrfs_node_ptr_generation(eb, slot); | |
2dd3e67b | 8558 | |
1c4850e2 YZ |
8559 | if (slot == path->slots[wc->level]) |
8560 | goto reada; | |
5d4f98a2 | 8561 | |
1c4850e2 YZ |
8562 | if (wc->stage == UPDATE_BACKREF && |
8563 | generation <= root->root_key.offset) | |
bd56b302 CM |
8564 | continue; |
8565 | ||
94fcca9f | 8566 | /* We don't lock the tree block, it's OK to be racy here */ |
2ff7e61e | 8567 | ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, |
3173a18f JB |
8568 | wc->level - 1, 1, &refs, |
8569 | &flags); | |
79787eaa JM |
8570 | /* We don't care about errors in readahead. */ |
8571 | if (ret < 0) | |
8572 | continue; | |
94fcca9f YZ |
8573 | BUG_ON(refs == 0); |
8574 | ||
1c4850e2 | 8575 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 YZ |
8576 | if (refs == 1) |
8577 | goto reada; | |
bd56b302 | 8578 | |
94fcca9f YZ |
8579 | if (wc->level == 1 && |
8580 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8581 | continue; | |
1c4850e2 YZ |
8582 | if (!wc->update_ref || |
8583 | generation <= root->root_key.offset) | |
8584 | continue; | |
8585 | btrfs_node_key_to_cpu(eb, &key, slot); | |
8586 | ret = btrfs_comp_cpu_keys(&key, | |
8587 | &wc->update_progress); | |
8588 | if (ret < 0) | |
8589 | continue; | |
94fcca9f YZ |
8590 | } else { |
8591 | if (wc->level == 1 && | |
8592 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8593 | continue; | |
6407bf6d | 8594 | } |
1c4850e2 | 8595 | reada: |
2ff7e61e | 8596 | readahead_tree_block(fs_info, bytenr); |
1c4850e2 | 8597 | nread++; |
20524f02 | 8598 | } |
1c4850e2 | 8599 | wc->reada_slot = slot; |
20524f02 | 8600 | } |
2c47e605 | 8601 | |
f82d02d9 | 8602 | /* |
2c016dc2 | 8603 | * helper to process tree block while walking down the tree. |
2c47e605 | 8604 | * |
2c47e605 YZ |
8605 | * when wc->stage == UPDATE_BACKREF, this function updates |
8606 | * back refs for pointers in the block. | |
8607 | * | |
8608 | * NOTE: return value 1 means we should stop walking down. | |
f82d02d9 | 8609 | */ |
2c47e605 | 8610 | static noinline int walk_down_proc(struct btrfs_trans_handle *trans, |
5d4f98a2 | 8611 | struct btrfs_root *root, |
2c47e605 | 8612 | struct btrfs_path *path, |
94fcca9f | 8613 | struct walk_control *wc, int lookup_info) |
f82d02d9 | 8614 | { |
2ff7e61e | 8615 | struct btrfs_fs_info *fs_info = root->fs_info; |
2c47e605 YZ |
8616 | int level = wc->level; |
8617 | struct extent_buffer *eb = path->nodes[level]; | |
2c47e605 | 8618 | u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF; |
f82d02d9 YZ |
8619 | int ret; |
8620 | ||
2c47e605 YZ |
8621 | if (wc->stage == UPDATE_BACKREF && |
8622 | btrfs_header_owner(eb) != root->root_key.objectid) | |
8623 | return 1; | |
f82d02d9 | 8624 | |
2c47e605 YZ |
8625 | /* |
8626 | * when reference count of tree block is 1, it won't increase | |
8627 | * again. once full backref flag is set, we never clear it. | |
8628 | */ | |
94fcca9f YZ |
8629 | if (lookup_info && |
8630 | ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) || | |
8631 | (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) { | |
2c47e605 | 8632 | BUG_ON(!path->locks[level]); |
2ff7e61e | 8633 | ret = btrfs_lookup_extent_info(trans, fs_info, |
3173a18f | 8634 | eb->start, level, 1, |
2c47e605 YZ |
8635 | &wc->refs[level], |
8636 | &wc->flags[level]); | |
79787eaa JM |
8637 | BUG_ON(ret == -ENOMEM); |
8638 | if (ret) | |
8639 | return ret; | |
2c47e605 YZ |
8640 | BUG_ON(wc->refs[level] == 0); |
8641 | } | |
5d4f98a2 | 8642 | |
2c47e605 YZ |
8643 | if (wc->stage == DROP_REFERENCE) { |
8644 | if (wc->refs[level] > 1) | |
8645 | return 1; | |
f82d02d9 | 8646 | |
2c47e605 | 8647 | if (path->locks[level] && !wc->keep_locks) { |
bd681513 | 8648 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
8649 | path->locks[level] = 0; |
8650 | } | |
8651 | return 0; | |
8652 | } | |
f82d02d9 | 8653 | |
2c47e605 YZ |
8654 | /* wc->stage == UPDATE_BACKREF */ |
8655 | if (!(wc->flags[level] & flag)) { | |
8656 | BUG_ON(!path->locks[level]); | |
e339a6b0 | 8657 | ret = btrfs_inc_ref(trans, root, eb, 1); |
79787eaa | 8658 | BUG_ON(ret); /* -ENOMEM */ |
e339a6b0 | 8659 | ret = btrfs_dec_ref(trans, root, eb, 0); |
79787eaa | 8660 | BUG_ON(ret); /* -ENOMEM */ |
2ff7e61e | 8661 | ret = btrfs_set_disk_extent_flags(trans, fs_info, eb->start, |
b1c79e09 JB |
8662 | eb->len, flag, |
8663 | btrfs_header_level(eb), 0); | |
79787eaa | 8664 | BUG_ON(ret); /* -ENOMEM */ |
2c47e605 YZ |
8665 | wc->flags[level] |= flag; |
8666 | } | |
8667 | ||
8668 | /* | |
8669 | * the block is shared by multiple trees, so it's not good to | |
8670 | * keep the tree lock | |
8671 | */ | |
8672 | if (path->locks[level] && level > 0) { | |
bd681513 | 8673 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
8674 | path->locks[level] = 0; |
8675 | } | |
8676 | return 0; | |
8677 | } | |
8678 | ||
1c4850e2 | 8679 | /* |
2c016dc2 | 8680 | * helper to process tree block pointer. |
1c4850e2 YZ |
8681 | * |
8682 | * when wc->stage == DROP_REFERENCE, this function checks | |
8683 | * reference count of the block pointed to. if the block | |
8684 | * is shared and we need update back refs for the subtree | |
8685 | * rooted at the block, this function changes wc->stage to | |
8686 | * UPDATE_BACKREF. if the block is shared and there is no | |
8687 | * need to update back, this function drops the reference | |
8688 | * to the block. | |
8689 | * | |
8690 | * NOTE: return value 1 means we should stop walking down. | |
8691 | */ | |
8692 | static noinline int do_walk_down(struct btrfs_trans_handle *trans, | |
8693 | struct btrfs_root *root, | |
8694 | struct btrfs_path *path, | |
94fcca9f | 8695 | struct walk_control *wc, int *lookup_info) |
1c4850e2 | 8696 | { |
0b246afa | 8697 | struct btrfs_fs_info *fs_info = root->fs_info; |
1c4850e2 YZ |
8698 | u64 bytenr; |
8699 | u64 generation; | |
8700 | u64 parent; | |
8701 | u32 blocksize; | |
8702 | struct btrfs_key key; | |
581c1760 | 8703 | struct btrfs_key first_key; |
1c4850e2 YZ |
8704 | struct extent_buffer *next; |
8705 | int level = wc->level; | |
8706 | int reada = 0; | |
8707 | int ret = 0; | |
1152651a | 8708 | bool need_account = false; |
1c4850e2 YZ |
8709 | |
8710 | generation = btrfs_node_ptr_generation(path->nodes[level], | |
8711 | path->slots[level]); | |
8712 | /* | |
8713 | * if the lower level block was created before the snapshot | |
8714 | * was created, we know there is no need to update back refs | |
8715 | * for the subtree | |
8716 | */ | |
8717 | if (wc->stage == UPDATE_BACKREF && | |
94fcca9f YZ |
8718 | generation <= root->root_key.offset) { |
8719 | *lookup_info = 1; | |
1c4850e2 | 8720 | return 1; |
94fcca9f | 8721 | } |
1c4850e2 YZ |
8722 | |
8723 | bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]); | |
581c1760 QW |
8724 | btrfs_node_key_to_cpu(path->nodes[level], &first_key, |
8725 | path->slots[level]); | |
0b246afa | 8726 | blocksize = fs_info->nodesize; |
1c4850e2 | 8727 | |
0b246afa | 8728 | next = find_extent_buffer(fs_info, bytenr); |
1c4850e2 | 8729 | if (!next) { |
2ff7e61e | 8730 | next = btrfs_find_create_tree_block(fs_info, bytenr); |
c871b0f2 LB |
8731 | if (IS_ERR(next)) |
8732 | return PTR_ERR(next); | |
8733 | ||
b2aaaa3b JB |
8734 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, next, |
8735 | level - 1); | |
1c4850e2 YZ |
8736 | reada = 1; |
8737 | } | |
8738 | btrfs_tree_lock(next); | |
8739 | btrfs_set_lock_blocking(next); | |
8740 | ||
2ff7e61e | 8741 | ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1, |
94fcca9f YZ |
8742 | &wc->refs[level - 1], |
8743 | &wc->flags[level - 1]); | |
4867268c JB |
8744 | if (ret < 0) |
8745 | goto out_unlock; | |
79787eaa | 8746 | |
c2cf52eb | 8747 | if (unlikely(wc->refs[level - 1] == 0)) { |
0b246afa | 8748 | btrfs_err(fs_info, "Missing references."); |
4867268c JB |
8749 | ret = -EIO; |
8750 | goto out_unlock; | |
c2cf52eb | 8751 | } |
94fcca9f | 8752 | *lookup_info = 0; |
1c4850e2 | 8753 | |
94fcca9f | 8754 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 | 8755 | if (wc->refs[level - 1] > 1) { |
1152651a | 8756 | need_account = true; |
94fcca9f YZ |
8757 | if (level == 1 && |
8758 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8759 | goto skip; | |
8760 | ||
1c4850e2 YZ |
8761 | if (!wc->update_ref || |
8762 | generation <= root->root_key.offset) | |
8763 | goto skip; | |
8764 | ||
8765 | btrfs_node_key_to_cpu(path->nodes[level], &key, | |
8766 | path->slots[level]); | |
8767 | ret = btrfs_comp_cpu_keys(&key, &wc->update_progress); | |
8768 | if (ret < 0) | |
8769 | goto skip; | |
8770 | ||
8771 | wc->stage = UPDATE_BACKREF; | |
8772 | wc->shared_level = level - 1; | |
8773 | } | |
94fcca9f YZ |
8774 | } else { |
8775 | if (level == 1 && | |
8776 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8777 | goto skip; | |
1c4850e2 YZ |
8778 | } |
8779 | ||
b9fab919 | 8780 | if (!btrfs_buffer_uptodate(next, generation, 0)) { |
1c4850e2 YZ |
8781 | btrfs_tree_unlock(next); |
8782 | free_extent_buffer(next); | |
8783 | next = NULL; | |
94fcca9f | 8784 | *lookup_info = 1; |
1c4850e2 YZ |
8785 | } |
8786 | ||
8787 | if (!next) { | |
8788 | if (reada && level == 1) | |
8789 | reada_walk_down(trans, root, wc, path); | |
581c1760 QW |
8790 | next = read_tree_block(fs_info, bytenr, generation, level - 1, |
8791 | &first_key); | |
64c043de LB |
8792 | if (IS_ERR(next)) { |
8793 | return PTR_ERR(next); | |
8794 | } else if (!extent_buffer_uptodate(next)) { | |
416bc658 | 8795 | free_extent_buffer(next); |
97d9a8a4 | 8796 | return -EIO; |
416bc658 | 8797 | } |
1c4850e2 YZ |
8798 | btrfs_tree_lock(next); |
8799 | btrfs_set_lock_blocking(next); | |
8800 | } | |
8801 | ||
8802 | level--; | |
4867268c JB |
8803 | ASSERT(level == btrfs_header_level(next)); |
8804 | if (level != btrfs_header_level(next)) { | |
8805 | btrfs_err(root->fs_info, "mismatched level"); | |
8806 | ret = -EIO; | |
8807 | goto out_unlock; | |
8808 | } | |
1c4850e2 YZ |
8809 | path->nodes[level] = next; |
8810 | path->slots[level] = 0; | |
bd681513 | 8811 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
1c4850e2 YZ |
8812 | wc->level = level; |
8813 | if (wc->level == 1) | |
8814 | wc->reada_slot = 0; | |
8815 | return 0; | |
8816 | skip: | |
8817 | wc->refs[level - 1] = 0; | |
8818 | wc->flags[level - 1] = 0; | |
94fcca9f YZ |
8819 | if (wc->stage == DROP_REFERENCE) { |
8820 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
8821 | parent = path->nodes[level]->start; | |
8822 | } else { | |
4867268c | 8823 | ASSERT(root->root_key.objectid == |
94fcca9f | 8824 | btrfs_header_owner(path->nodes[level])); |
4867268c JB |
8825 | if (root->root_key.objectid != |
8826 | btrfs_header_owner(path->nodes[level])) { | |
8827 | btrfs_err(root->fs_info, | |
8828 | "mismatched block owner"); | |
8829 | ret = -EIO; | |
8830 | goto out_unlock; | |
8831 | } | |
94fcca9f YZ |
8832 | parent = 0; |
8833 | } | |
1c4850e2 | 8834 | |
1152651a | 8835 | if (need_account) { |
33d1f05c QW |
8836 | ret = btrfs_qgroup_trace_subtree(trans, root, next, |
8837 | generation, level - 1); | |
1152651a | 8838 | if (ret) { |
0b246afa | 8839 | btrfs_err_rl(fs_info, |
5d163e0e JM |
8840 | "Error %d accounting shared subtree. Quota is out of sync, rescan required.", |
8841 | ret); | |
1152651a MF |
8842 | } |
8843 | } | |
84f7d8e6 | 8844 | ret = btrfs_free_extent(trans, root, bytenr, blocksize, |
2ff7e61e JM |
8845 | parent, root->root_key.objectid, |
8846 | level - 1, 0); | |
4867268c JB |
8847 | if (ret) |
8848 | goto out_unlock; | |
1c4850e2 | 8849 | } |
4867268c JB |
8850 | |
8851 | *lookup_info = 1; | |
8852 | ret = 1; | |
8853 | ||
8854 | out_unlock: | |
1c4850e2 YZ |
8855 | btrfs_tree_unlock(next); |
8856 | free_extent_buffer(next); | |
4867268c JB |
8857 | |
8858 | return ret; | |
1c4850e2 YZ |
8859 | } |
8860 | ||
2c47e605 | 8861 | /* |
2c016dc2 | 8862 | * helper to process tree block while walking up the tree. |
2c47e605 YZ |
8863 | * |
8864 | * when wc->stage == DROP_REFERENCE, this function drops | |
8865 | * reference count on the block. | |
8866 | * | |
8867 | * when wc->stage == UPDATE_BACKREF, this function changes | |
8868 | * wc->stage back to DROP_REFERENCE if we changed wc->stage | |
8869 | * to UPDATE_BACKREF previously while processing the block. | |
8870 | * | |
8871 | * NOTE: return value 1 means we should stop walking up. | |
8872 | */ | |
8873 | static noinline int walk_up_proc(struct btrfs_trans_handle *trans, | |
8874 | struct btrfs_root *root, | |
8875 | struct btrfs_path *path, | |
8876 | struct walk_control *wc) | |
8877 | { | |
0b246afa | 8878 | struct btrfs_fs_info *fs_info = root->fs_info; |
f0486c68 | 8879 | int ret; |
2c47e605 YZ |
8880 | int level = wc->level; |
8881 | struct extent_buffer *eb = path->nodes[level]; | |
8882 | u64 parent = 0; | |
8883 | ||
8884 | if (wc->stage == UPDATE_BACKREF) { | |
8885 | BUG_ON(wc->shared_level < level); | |
8886 | if (level < wc->shared_level) | |
8887 | goto out; | |
8888 | ||
2c47e605 YZ |
8889 | ret = find_next_key(path, level + 1, &wc->update_progress); |
8890 | if (ret > 0) | |
8891 | wc->update_ref = 0; | |
8892 | ||
8893 | wc->stage = DROP_REFERENCE; | |
8894 | wc->shared_level = -1; | |
8895 | path->slots[level] = 0; | |
8896 | ||
8897 | /* | |
8898 | * check reference count again if the block isn't locked. | |
8899 | * we should start walking down the tree again if reference | |
8900 | * count is one. | |
8901 | */ | |
8902 | if (!path->locks[level]) { | |
8903 | BUG_ON(level == 0); | |
8904 | btrfs_tree_lock(eb); | |
8905 | btrfs_set_lock_blocking(eb); | |
bd681513 | 8906 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 | 8907 | |
2ff7e61e | 8908 | ret = btrfs_lookup_extent_info(trans, fs_info, |
3173a18f | 8909 | eb->start, level, 1, |
2c47e605 YZ |
8910 | &wc->refs[level], |
8911 | &wc->flags[level]); | |
79787eaa JM |
8912 | if (ret < 0) { |
8913 | btrfs_tree_unlock_rw(eb, path->locks[level]); | |
3268a246 | 8914 | path->locks[level] = 0; |
79787eaa JM |
8915 | return ret; |
8916 | } | |
2c47e605 YZ |
8917 | BUG_ON(wc->refs[level] == 0); |
8918 | if (wc->refs[level] == 1) { | |
bd681513 | 8919 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
3268a246 | 8920 | path->locks[level] = 0; |
2c47e605 YZ |
8921 | return 1; |
8922 | } | |
f82d02d9 | 8923 | } |
2c47e605 | 8924 | } |
f82d02d9 | 8925 | |
2c47e605 YZ |
8926 | /* wc->stage == DROP_REFERENCE */ |
8927 | BUG_ON(wc->refs[level] > 1 && !path->locks[level]); | |
5d4f98a2 | 8928 | |
2c47e605 YZ |
8929 | if (wc->refs[level] == 1) { |
8930 | if (level == 0) { | |
8931 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
e339a6b0 | 8932 | ret = btrfs_dec_ref(trans, root, eb, 1); |
2c47e605 | 8933 | else |
e339a6b0 | 8934 | ret = btrfs_dec_ref(trans, root, eb, 0); |
79787eaa | 8935 | BUG_ON(ret); /* -ENOMEM */ |
2ff7e61e | 8936 | ret = btrfs_qgroup_trace_leaf_items(trans, fs_info, eb); |
1152651a | 8937 | if (ret) { |
0b246afa | 8938 | btrfs_err_rl(fs_info, |
5d163e0e JM |
8939 | "error %d accounting leaf items. Quota is out of sync, rescan required.", |
8940 | ret); | |
1152651a | 8941 | } |
2c47e605 YZ |
8942 | } |
8943 | /* make block locked assertion in clean_tree_block happy */ | |
8944 | if (!path->locks[level] && | |
8945 | btrfs_header_generation(eb) == trans->transid) { | |
8946 | btrfs_tree_lock(eb); | |
8947 | btrfs_set_lock_blocking(eb); | |
bd681513 | 8948 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 | 8949 | } |
7c302b49 | 8950 | clean_tree_block(fs_info, eb); |
2c47e605 YZ |
8951 | } |
8952 | ||
8953 | if (eb == root->node) { | |
8954 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
8955 | parent = eb->start; | |
8956 | else | |
8957 | BUG_ON(root->root_key.objectid != | |
8958 | btrfs_header_owner(eb)); | |
8959 | } else { | |
8960 | if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
8961 | parent = path->nodes[level + 1]->start; | |
8962 | else | |
8963 | BUG_ON(root->root_key.objectid != | |
8964 | btrfs_header_owner(path->nodes[level + 1])); | |
f82d02d9 | 8965 | } |
f82d02d9 | 8966 | |
5581a51a | 8967 | btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1); |
2c47e605 YZ |
8968 | out: |
8969 | wc->refs[level] = 0; | |
8970 | wc->flags[level] = 0; | |
f0486c68 | 8971 | return 0; |
2c47e605 YZ |
8972 | } |
8973 | ||
8974 | static noinline int walk_down_tree(struct btrfs_trans_handle *trans, | |
8975 | struct btrfs_root *root, | |
8976 | struct btrfs_path *path, | |
8977 | struct walk_control *wc) | |
8978 | { | |
2c47e605 | 8979 | int level = wc->level; |
94fcca9f | 8980 | int lookup_info = 1; |
2c47e605 YZ |
8981 | int ret; |
8982 | ||
8983 | while (level >= 0) { | |
94fcca9f | 8984 | ret = walk_down_proc(trans, root, path, wc, lookup_info); |
2c47e605 YZ |
8985 | if (ret > 0) |
8986 | break; | |
8987 | ||
8988 | if (level == 0) | |
8989 | break; | |
8990 | ||
7a7965f8 YZ |
8991 | if (path->slots[level] >= |
8992 | btrfs_header_nritems(path->nodes[level])) | |
8993 | break; | |
8994 | ||
94fcca9f | 8995 | ret = do_walk_down(trans, root, path, wc, &lookup_info); |
1c4850e2 YZ |
8996 | if (ret > 0) { |
8997 | path->slots[level]++; | |
8998 | continue; | |
90d2c51d MX |
8999 | } else if (ret < 0) |
9000 | return ret; | |
1c4850e2 | 9001 | level = wc->level; |
f82d02d9 | 9002 | } |
f82d02d9 YZ |
9003 | return 0; |
9004 | } | |
9005 | ||
d397712b | 9006 | static noinline int walk_up_tree(struct btrfs_trans_handle *trans, |
98ed5174 | 9007 | struct btrfs_root *root, |
f82d02d9 | 9008 | struct btrfs_path *path, |
2c47e605 | 9009 | struct walk_control *wc, int max_level) |
20524f02 | 9010 | { |
2c47e605 | 9011 | int level = wc->level; |
20524f02 | 9012 | int ret; |
9f3a7427 | 9013 | |
2c47e605 YZ |
9014 | path->slots[level] = btrfs_header_nritems(path->nodes[level]); |
9015 | while (level < max_level && path->nodes[level]) { | |
9016 | wc->level = level; | |
9017 | if (path->slots[level] + 1 < | |
9018 | btrfs_header_nritems(path->nodes[level])) { | |
9019 | path->slots[level]++; | |
20524f02 CM |
9020 | return 0; |
9021 | } else { | |
2c47e605 YZ |
9022 | ret = walk_up_proc(trans, root, path, wc); |
9023 | if (ret > 0) | |
9024 | return 0; | |
bd56b302 | 9025 | |
2c47e605 | 9026 | if (path->locks[level]) { |
bd681513 CM |
9027 | btrfs_tree_unlock_rw(path->nodes[level], |
9028 | path->locks[level]); | |
2c47e605 | 9029 | path->locks[level] = 0; |
f82d02d9 | 9030 | } |
2c47e605 YZ |
9031 | free_extent_buffer(path->nodes[level]); |
9032 | path->nodes[level] = NULL; | |
9033 | level++; | |
20524f02 CM |
9034 | } |
9035 | } | |
9036 | return 1; | |
9037 | } | |
9038 | ||
9aca1d51 | 9039 | /* |
2c47e605 YZ |
9040 | * drop a subvolume tree. |
9041 | * | |
9042 | * this function traverses the tree freeing any blocks that only | |
9043 | * referenced by the tree. | |
9044 | * | |
9045 | * when a shared tree block is found. this function decreases its | |
9046 | * reference count by one. if update_ref is true, this function | |
9047 | * also make sure backrefs for the shared block and all lower level | |
9048 | * blocks are properly updated. | |
9d1a2a3a DS |
9049 | * |
9050 | * If called with for_reloc == 0, may exit early with -EAGAIN | |
9aca1d51 | 9051 | */ |
2c536799 | 9052 | int btrfs_drop_snapshot(struct btrfs_root *root, |
66d7e7f0 AJ |
9053 | struct btrfs_block_rsv *block_rsv, int update_ref, |
9054 | int for_reloc) | |
20524f02 | 9055 | { |
ab8d0fc4 | 9056 | struct btrfs_fs_info *fs_info = root->fs_info; |
5caf2a00 | 9057 | struct btrfs_path *path; |
2c47e605 | 9058 | struct btrfs_trans_handle *trans; |
ab8d0fc4 | 9059 | struct btrfs_root *tree_root = fs_info->tree_root; |
9f3a7427 | 9060 | struct btrfs_root_item *root_item = &root->root_item; |
2c47e605 YZ |
9061 | struct walk_control *wc; |
9062 | struct btrfs_key key; | |
9063 | int err = 0; | |
9064 | int ret; | |
9065 | int level; | |
d29a9f62 | 9066 | bool root_dropped = false; |
20524f02 | 9067 | |
ab8d0fc4 | 9068 | btrfs_debug(fs_info, "Drop subvolume %llu", root->objectid); |
1152651a | 9069 | |
5caf2a00 | 9070 | path = btrfs_alloc_path(); |
cb1b69f4 TI |
9071 | if (!path) { |
9072 | err = -ENOMEM; | |
9073 | goto out; | |
9074 | } | |
20524f02 | 9075 | |
2c47e605 | 9076 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
38a1a919 MF |
9077 | if (!wc) { |
9078 | btrfs_free_path(path); | |
cb1b69f4 TI |
9079 | err = -ENOMEM; |
9080 | goto out; | |
38a1a919 | 9081 | } |
2c47e605 | 9082 | |
a22285a6 | 9083 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
9084 | if (IS_ERR(trans)) { |
9085 | err = PTR_ERR(trans); | |
9086 | goto out_free; | |
9087 | } | |
98d5dc13 | 9088 | |
3fd0a558 YZ |
9089 | if (block_rsv) |
9090 | trans->block_rsv = block_rsv; | |
2c47e605 | 9091 | |
9f3a7427 | 9092 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { |
2c47e605 | 9093 | level = btrfs_header_level(root->node); |
5d4f98a2 YZ |
9094 | path->nodes[level] = btrfs_lock_root_node(root); |
9095 | btrfs_set_lock_blocking(path->nodes[level]); | |
9f3a7427 | 9096 | path->slots[level] = 0; |
bd681513 | 9097 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
9098 | memset(&wc->update_progress, 0, |
9099 | sizeof(wc->update_progress)); | |
9f3a7427 | 9100 | } else { |
9f3a7427 | 9101 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); |
2c47e605 YZ |
9102 | memcpy(&wc->update_progress, &key, |
9103 | sizeof(wc->update_progress)); | |
9104 | ||
6702ed49 | 9105 | level = root_item->drop_level; |
2c47e605 | 9106 | BUG_ON(level == 0); |
6702ed49 | 9107 | path->lowest_level = level; |
2c47e605 YZ |
9108 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
9109 | path->lowest_level = 0; | |
9110 | if (ret < 0) { | |
9111 | err = ret; | |
79787eaa | 9112 | goto out_end_trans; |
9f3a7427 | 9113 | } |
1c4850e2 | 9114 | WARN_ON(ret > 0); |
2c47e605 | 9115 | |
7d9eb12c CM |
9116 | /* |
9117 | * unlock our path, this is safe because only this | |
9118 | * function is allowed to delete this snapshot | |
9119 | */ | |
5d4f98a2 | 9120 | btrfs_unlock_up_safe(path, 0); |
2c47e605 YZ |
9121 | |
9122 | level = btrfs_header_level(root->node); | |
9123 | while (1) { | |
9124 | btrfs_tree_lock(path->nodes[level]); | |
9125 | btrfs_set_lock_blocking(path->nodes[level]); | |
fec386ac | 9126 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 | 9127 | |
2ff7e61e | 9128 | ret = btrfs_lookup_extent_info(trans, fs_info, |
2c47e605 | 9129 | path->nodes[level]->start, |
3173a18f | 9130 | level, 1, &wc->refs[level], |
2c47e605 | 9131 | &wc->flags[level]); |
79787eaa JM |
9132 | if (ret < 0) { |
9133 | err = ret; | |
9134 | goto out_end_trans; | |
9135 | } | |
2c47e605 YZ |
9136 | BUG_ON(wc->refs[level] == 0); |
9137 | ||
9138 | if (level == root_item->drop_level) | |
9139 | break; | |
9140 | ||
9141 | btrfs_tree_unlock(path->nodes[level]); | |
fec386ac | 9142 | path->locks[level] = 0; |
2c47e605 YZ |
9143 | WARN_ON(wc->refs[level] != 1); |
9144 | level--; | |
9145 | } | |
9f3a7427 | 9146 | } |
2c47e605 YZ |
9147 | |
9148 | wc->level = level; | |
9149 | wc->shared_level = -1; | |
9150 | wc->stage = DROP_REFERENCE; | |
9151 | wc->update_ref = update_ref; | |
9152 | wc->keep_locks = 0; | |
66d7e7f0 | 9153 | wc->for_reloc = for_reloc; |
0b246afa | 9154 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info); |
2c47e605 | 9155 | |
d397712b | 9156 | while (1) { |
9d1a2a3a | 9157 | |
2c47e605 YZ |
9158 | ret = walk_down_tree(trans, root, path, wc); |
9159 | if (ret < 0) { | |
9160 | err = ret; | |
20524f02 | 9161 | break; |
2c47e605 | 9162 | } |
9aca1d51 | 9163 | |
2c47e605 YZ |
9164 | ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL); |
9165 | if (ret < 0) { | |
9166 | err = ret; | |
20524f02 | 9167 | break; |
2c47e605 YZ |
9168 | } |
9169 | ||
9170 | if (ret > 0) { | |
9171 | BUG_ON(wc->stage != DROP_REFERENCE); | |
e7a84565 CM |
9172 | break; |
9173 | } | |
2c47e605 YZ |
9174 | |
9175 | if (wc->stage == DROP_REFERENCE) { | |
9176 | level = wc->level; | |
9177 | btrfs_node_key(path->nodes[level], | |
9178 | &root_item->drop_progress, | |
9179 | path->slots[level]); | |
9180 | root_item->drop_level = level; | |
9181 | } | |
9182 | ||
9183 | BUG_ON(wc->level == 0); | |
3a45bb20 | 9184 | if (btrfs_should_end_transaction(trans) || |
2ff7e61e | 9185 | (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) { |
2c47e605 YZ |
9186 | ret = btrfs_update_root(trans, tree_root, |
9187 | &root->root_key, | |
9188 | root_item); | |
79787eaa | 9189 | if (ret) { |
66642832 | 9190 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9191 | err = ret; |
9192 | goto out_end_trans; | |
9193 | } | |
2c47e605 | 9194 | |
3a45bb20 | 9195 | btrfs_end_transaction_throttle(trans); |
2ff7e61e | 9196 | if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) { |
ab8d0fc4 JM |
9197 | btrfs_debug(fs_info, |
9198 | "drop snapshot early exit"); | |
3c8f2422 JB |
9199 | err = -EAGAIN; |
9200 | goto out_free; | |
9201 | } | |
9202 | ||
a22285a6 | 9203 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
9204 | if (IS_ERR(trans)) { |
9205 | err = PTR_ERR(trans); | |
9206 | goto out_free; | |
9207 | } | |
3fd0a558 YZ |
9208 | if (block_rsv) |
9209 | trans->block_rsv = block_rsv; | |
c3e69d58 | 9210 | } |
20524f02 | 9211 | } |
b3b4aa74 | 9212 | btrfs_release_path(path); |
79787eaa JM |
9213 | if (err) |
9214 | goto out_end_trans; | |
2c47e605 | 9215 | |
1cd5447e | 9216 | ret = btrfs_del_root(trans, fs_info, &root->root_key); |
79787eaa | 9217 | if (ret) { |
66642832 | 9218 | btrfs_abort_transaction(trans, ret); |
e19182c0 | 9219 | err = ret; |
79787eaa JM |
9220 | goto out_end_trans; |
9221 | } | |
2c47e605 | 9222 | |
76dda93c | 9223 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { |
cb517eab MX |
9224 | ret = btrfs_find_root(tree_root, &root->root_key, path, |
9225 | NULL, NULL); | |
79787eaa | 9226 | if (ret < 0) { |
66642832 | 9227 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9228 | err = ret; |
9229 | goto out_end_trans; | |
9230 | } else if (ret > 0) { | |
84cd948c JB |
9231 | /* if we fail to delete the orphan item this time |
9232 | * around, it'll get picked up the next time. | |
9233 | * | |
9234 | * The most common failure here is just -ENOENT. | |
9235 | */ | |
9236 | btrfs_del_orphan_item(trans, tree_root, | |
9237 | root->root_key.objectid); | |
76dda93c YZ |
9238 | } |
9239 | } | |
9240 | ||
27cdeb70 | 9241 | if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) { |
2b9dbef2 | 9242 | btrfs_add_dropped_root(trans, root); |
76dda93c YZ |
9243 | } else { |
9244 | free_extent_buffer(root->node); | |
9245 | free_extent_buffer(root->commit_root); | |
b0feb9d9 | 9246 | btrfs_put_fs_root(root); |
76dda93c | 9247 | } |
d29a9f62 | 9248 | root_dropped = true; |
79787eaa | 9249 | out_end_trans: |
3a45bb20 | 9250 | btrfs_end_transaction_throttle(trans); |
79787eaa | 9251 | out_free: |
2c47e605 | 9252 | kfree(wc); |
5caf2a00 | 9253 | btrfs_free_path(path); |
cb1b69f4 | 9254 | out: |
d29a9f62 JB |
9255 | /* |
9256 | * So if we need to stop dropping the snapshot for whatever reason we | |
9257 | * need to make sure to add it back to the dead root list so that we | |
9258 | * keep trying to do the work later. This also cleans up roots if we | |
9259 | * don't have it in the radix (like when we recover after a power fail | |
9260 | * or unmount) so we don't leak memory. | |
9261 | */ | |
897ca819 | 9262 | if (!for_reloc && !root_dropped) |
d29a9f62 | 9263 | btrfs_add_dead_root(root); |
90515e7f | 9264 | if (err && err != -EAGAIN) |
ab8d0fc4 | 9265 | btrfs_handle_fs_error(fs_info, err, NULL); |
2c536799 | 9266 | return err; |
20524f02 | 9267 | } |
9078a3e1 | 9268 | |
2c47e605 YZ |
9269 | /* |
9270 | * drop subtree rooted at tree block 'node'. | |
9271 | * | |
9272 | * NOTE: this function will unlock and release tree block 'node' | |
66d7e7f0 | 9273 | * only used by relocation code |
2c47e605 | 9274 | */ |
f82d02d9 YZ |
9275 | int btrfs_drop_subtree(struct btrfs_trans_handle *trans, |
9276 | struct btrfs_root *root, | |
9277 | struct extent_buffer *node, | |
9278 | struct extent_buffer *parent) | |
9279 | { | |
0b246afa | 9280 | struct btrfs_fs_info *fs_info = root->fs_info; |
f82d02d9 | 9281 | struct btrfs_path *path; |
2c47e605 | 9282 | struct walk_control *wc; |
f82d02d9 YZ |
9283 | int level; |
9284 | int parent_level; | |
9285 | int ret = 0; | |
9286 | int wret; | |
9287 | ||
2c47e605 YZ |
9288 | BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); |
9289 | ||
f82d02d9 | 9290 | path = btrfs_alloc_path(); |
db5b493a TI |
9291 | if (!path) |
9292 | return -ENOMEM; | |
f82d02d9 | 9293 | |
2c47e605 | 9294 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
db5b493a TI |
9295 | if (!wc) { |
9296 | btrfs_free_path(path); | |
9297 | return -ENOMEM; | |
9298 | } | |
2c47e605 | 9299 | |
b9447ef8 | 9300 | btrfs_assert_tree_locked(parent); |
f82d02d9 YZ |
9301 | parent_level = btrfs_header_level(parent); |
9302 | extent_buffer_get(parent); | |
9303 | path->nodes[parent_level] = parent; | |
9304 | path->slots[parent_level] = btrfs_header_nritems(parent); | |
9305 | ||
b9447ef8 | 9306 | btrfs_assert_tree_locked(node); |
f82d02d9 | 9307 | level = btrfs_header_level(node); |
f82d02d9 YZ |
9308 | path->nodes[level] = node; |
9309 | path->slots[level] = 0; | |
bd681513 | 9310 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
9311 | |
9312 | wc->refs[parent_level] = 1; | |
9313 | wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
9314 | wc->level = level; | |
9315 | wc->shared_level = -1; | |
9316 | wc->stage = DROP_REFERENCE; | |
9317 | wc->update_ref = 0; | |
9318 | wc->keep_locks = 1; | |
66d7e7f0 | 9319 | wc->for_reloc = 1; |
0b246afa | 9320 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info); |
f82d02d9 YZ |
9321 | |
9322 | while (1) { | |
2c47e605 YZ |
9323 | wret = walk_down_tree(trans, root, path, wc); |
9324 | if (wret < 0) { | |
f82d02d9 | 9325 | ret = wret; |
f82d02d9 | 9326 | break; |
2c47e605 | 9327 | } |
f82d02d9 | 9328 | |
2c47e605 | 9329 | wret = walk_up_tree(trans, root, path, wc, parent_level); |
f82d02d9 YZ |
9330 | if (wret < 0) |
9331 | ret = wret; | |
9332 | if (wret != 0) | |
9333 | break; | |
9334 | } | |
9335 | ||
2c47e605 | 9336 | kfree(wc); |
f82d02d9 YZ |
9337 | btrfs_free_path(path); |
9338 | return ret; | |
9339 | } | |
9340 | ||
6202df69 | 9341 | static u64 update_block_group_flags(struct btrfs_fs_info *fs_info, u64 flags) |
ec44a35c CM |
9342 | { |
9343 | u64 num_devices; | |
fc67c450 | 9344 | u64 stripped; |
e4d8ec0f | 9345 | |
fc67c450 ID |
9346 | /* |
9347 | * if restripe for this chunk_type is on pick target profile and | |
9348 | * return, otherwise do the usual balance | |
9349 | */ | |
6202df69 | 9350 | stripped = get_restripe_target(fs_info, flags); |
fc67c450 ID |
9351 | if (stripped) |
9352 | return extended_to_chunk(stripped); | |
e4d8ec0f | 9353 | |
6202df69 | 9354 | num_devices = fs_info->fs_devices->rw_devices; |
cd02dca5 | 9355 | |
fc67c450 | 9356 | stripped = BTRFS_BLOCK_GROUP_RAID0 | |
53b381b3 | 9357 | BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 | |
fc67c450 ID |
9358 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10; |
9359 | ||
ec44a35c CM |
9360 | if (num_devices == 1) { |
9361 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
9362 | stripped = flags & ~stripped; | |
9363 | ||
9364 | /* turn raid0 into single device chunks */ | |
9365 | if (flags & BTRFS_BLOCK_GROUP_RAID0) | |
9366 | return stripped; | |
9367 | ||
9368 | /* turn mirroring into duplication */ | |
9369 | if (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
9370 | BTRFS_BLOCK_GROUP_RAID10)) | |
9371 | return stripped | BTRFS_BLOCK_GROUP_DUP; | |
ec44a35c CM |
9372 | } else { |
9373 | /* they already had raid on here, just return */ | |
ec44a35c CM |
9374 | if (flags & stripped) |
9375 | return flags; | |
9376 | ||
9377 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
9378 | stripped = flags & ~stripped; | |
9379 | ||
9380 | /* switch duplicated blocks with raid1 */ | |
9381 | if (flags & BTRFS_BLOCK_GROUP_DUP) | |
9382 | return stripped | BTRFS_BLOCK_GROUP_RAID1; | |
9383 | ||
e3176ca2 | 9384 | /* this is drive concat, leave it alone */ |
ec44a35c | 9385 | } |
e3176ca2 | 9386 | |
ec44a35c CM |
9387 | return flags; |
9388 | } | |
9389 | ||
868f401a | 9390 | static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force) |
0ef3e66b | 9391 | { |
f0486c68 YZ |
9392 | struct btrfs_space_info *sinfo = cache->space_info; |
9393 | u64 num_bytes; | |
199c36ea | 9394 | u64 min_allocable_bytes; |
f0486c68 | 9395 | int ret = -ENOSPC; |
0ef3e66b | 9396 | |
199c36ea MX |
9397 | /* |
9398 | * We need some metadata space and system metadata space for | |
9399 | * allocating chunks in some corner cases until we force to set | |
9400 | * it to be readonly. | |
9401 | */ | |
9402 | if ((sinfo->flags & | |
9403 | (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) && | |
9404 | !force) | |
ee22184b | 9405 | min_allocable_bytes = SZ_1M; |
199c36ea MX |
9406 | else |
9407 | min_allocable_bytes = 0; | |
9408 | ||
f0486c68 YZ |
9409 | spin_lock(&sinfo->lock); |
9410 | spin_lock(&cache->lock); | |
61cfea9b W |
9411 | |
9412 | if (cache->ro) { | |
868f401a | 9413 | cache->ro++; |
61cfea9b W |
9414 | ret = 0; |
9415 | goto out; | |
9416 | } | |
9417 | ||
f0486c68 YZ |
9418 | num_bytes = cache->key.offset - cache->reserved - cache->pinned - |
9419 | cache->bytes_super - btrfs_block_group_used(&cache->item); | |
9420 | ||
4136135b | 9421 | if (btrfs_space_info_used(sinfo, true) + num_bytes + |
37be25bc | 9422 | min_allocable_bytes <= sinfo->total_bytes) { |
f0486c68 | 9423 | sinfo->bytes_readonly += num_bytes; |
868f401a | 9424 | cache->ro++; |
633c0aad | 9425 | list_add_tail(&cache->ro_list, &sinfo->ro_bgs); |
f0486c68 YZ |
9426 | ret = 0; |
9427 | } | |
61cfea9b | 9428 | out: |
f0486c68 YZ |
9429 | spin_unlock(&cache->lock); |
9430 | spin_unlock(&sinfo->lock); | |
9431 | return ret; | |
9432 | } | |
7d9eb12c | 9433 | |
5e00f193 | 9434 | int btrfs_inc_block_group_ro(struct btrfs_fs_info *fs_info, |
f0486c68 | 9435 | struct btrfs_block_group_cache *cache) |
c286ac48 | 9436 | |
f0486c68 YZ |
9437 | { |
9438 | struct btrfs_trans_handle *trans; | |
9439 | u64 alloc_flags; | |
9440 | int ret; | |
7d9eb12c | 9441 | |
1bbc621e | 9442 | again: |
5e00f193 | 9443 | trans = btrfs_join_transaction(fs_info->extent_root); |
79787eaa JM |
9444 | if (IS_ERR(trans)) |
9445 | return PTR_ERR(trans); | |
5d4f98a2 | 9446 | |
1bbc621e CM |
9447 | /* |
9448 | * we're not allowed to set block groups readonly after the dirty | |
9449 | * block groups cache has started writing. If it already started, | |
9450 | * back off and let this transaction commit | |
9451 | */ | |
0b246afa | 9452 | mutex_lock(&fs_info->ro_block_group_mutex); |
3204d33c | 9453 | if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) { |
1bbc621e CM |
9454 | u64 transid = trans->transid; |
9455 | ||
0b246afa | 9456 | mutex_unlock(&fs_info->ro_block_group_mutex); |
3a45bb20 | 9457 | btrfs_end_transaction(trans); |
1bbc621e | 9458 | |
2ff7e61e | 9459 | ret = btrfs_wait_for_commit(fs_info, transid); |
1bbc621e CM |
9460 | if (ret) |
9461 | return ret; | |
9462 | goto again; | |
9463 | } | |
9464 | ||
153c35b6 CM |
9465 | /* |
9466 | * if we are changing raid levels, try to allocate a corresponding | |
9467 | * block group with the new raid level. | |
9468 | */ | |
0b246afa | 9469 | alloc_flags = update_block_group_flags(fs_info, cache->flags); |
153c35b6 | 9470 | if (alloc_flags != cache->flags) { |
01458828 | 9471 | ret = do_chunk_alloc(trans, alloc_flags, |
153c35b6 CM |
9472 | CHUNK_ALLOC_FORCE); |
9473 | /* | |
9474 | * ENOSPC is allowed here, we may have enough space | |
9475 | * already allocated at the new raid level to | |
9476 | * carry on | |
9477 | */ | |
9478 | if (ret == -ENOSPC) | |
9479 | ret = 0; | |
9480 | if (ret < 0) | |
9481 | goto out; | |
9482 | } | |
1bbc621e | 9483 | |
868f401a | 9484 | ret = inc_block_group_ro(cache, 0); |
f0486c68 YZ |
9485 | if (!ret) |
9486 | goto out; | |
2ff7e61e | 9487 | alloc_flags = get_alloc_profile(fs_info, cache->space_info->flags); |
01458828 | 9488 | ret = do_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE); |
f0486c68 YZ |
9489 | if (ret < 0) |
9490 | goto out; | |
868f401a | 9491 | ret = inc_block_group_ro(cache, 0); |
f0486c68 | 9492 | out: |
2f081088 | 9493 | if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) { |
0b246afa | 9494 | alloc_flags = update_block_group_flags(fs_info, cache->flags); |
34441361 | 9495 | mutex_lock(&fs_info->chunk_mutex); |
451a2c13 | 9496 | check_system_chunk(trans, alloc_flags); |
34441361 | 9497 | mutex_unlock(&fs_info->chunk_mutex); |
2f081088 | 9498 | } |
0b246afa | 9499 | mutex_unlock(&fs_info->ro_block_group_mutex); |
2f081088 | 9500 | |
3a45bb20 | 9501 | btrfs_end_transaction(trans); |
f0486c68 YZ |
9502 | return ret; |
9503 | } | |
5d4f98a2 | 9504 | |
c87f08ca | 9505 | int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans, |
2ff7e61e | 9506 | struct btrfs_fs_info *fs_info, u64 type) |
c87f08ca | 9507 | { |
2ff7e61e JM |
9508 | u64 alloc_flags = get_alloc_profile(fs_info, type); |
9509 | ||
01458828 | 9510 | return do_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE); |
c87f08ca CM |
9511 | } |
9512 | ||
6d07bcec MX |
9513 | /* |
9514 | * helper to account the unused space of all the readonly block group in the | |
633c0aad | 9515 | * space_info. takes mirrors into account. |
6d07bcec | 9516 | */ |
633c0aad | 9517 | u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo) |
6d07bcec MX |
9518 | { |
9519 | struct btrfs_block_group_cache *block_group; | |
9520 | u64 free_bytes = 0; | |
9521 | int factor; | |
9522 | ||
01327610 | 9523 | /* It's df, we don't care if it's racy */ |
633c0aad JB |
9524 | if (list_empty(&sinfo->ro_bgs)) |
9525 | return 0; | |
9526 | ||
9527 | spin_lock(&sinfo->lock); | |
9528 | list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) { | |
6d07bcec MX |
9529 | spin_lock(&block_group->lock); |
9530 | ||
9531 | if (!block_group->ro) { | |
9532 | spin_unlock(&block_group->lock); | |
9533 | continue; | |
9534 | } | |
9535 | ||
9536 | if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
9537 | BTRFS_BLOCK_GROUP_RAID10 | | |
9538 | BTRFS_BLOCK_GROUP_DUP)) | |
9539 | factor = 2; | |
9540 | else | |
9541 | factor = 1; | |
9542 | ||
9543 | free_bytes += (block_group->key.offset - | |
9544 | btrfs_block_group_used(&block_group->item)) * | |
9545 | factor; | |
9546 | ||
9547 | spin_unlock(&block_group->lock); | |
9548 | } | |
6d07bcec MX |
9549 | spin_unlock(&sinfo->lock); |
9550 | ||
9551 | return free_bytes; | |
9552 | } | |
9553 | ||
2ff7e61e | 9554 | void btrfs_dec_block_group_ro(struct btrfs_block_group_cache *cache) |
5d4f98a2 | 9555 | { |
f0486c68 YZ |
9556 | struct btrfs_space_info *sinfo = cache->space_info; |
9557 | u64 num_bytes; | |
9558 | ||
9559 | BUG_ON(!cache->ro); | |
9560 | ||
9561 | spin_lock(&sinfo->lock); | |
9562 | spin_lock(&cache->lock); | |
868f401a Z |
9563 | if (!--cache->ro) { |
9564 | num_bytes = cache->key.offset - cache->reserved - | |
9565 | cache->pinned - cache->bytes_super - | |
9566 | btrfs_block_group_used(&cache->item); | |
9567 | sinfo->bytes_readonly -= num_bytes; | |
9568 | list_del_init(&cache->ro_list); | |
9569 | } | |
f0486c68 YZ |
9570 | spin_unlock(&cache->lock); |
9571 | spin_unlock(&sinfo->lock); | |
5d4f98a2 YZ |
9572 | } |
9573 | ||
ba1bf481 JB |
9574 | /* |
9575 | * checks to see if its even possible to relocate this block group. | |
9576 | * | |
9577 | * @return - -1 if it's not a good idea to relocate this block group, 0 if its | |
9578 | * ok to go ahead and try. | |
9579 | */ | |
6bccf3ab | 9580 | int btrfs_can_relocate(struct btrfs_fs_info *fs_info, u64 bytenr) |
1a40e23b | 9581 | { |
6bccf3ab | 9582 | struct btrfs_root *root = fs_info->extent_root; |
ba1bf481 JB |
9583 | struct btrfs_block_group_cache *block_group; |
9584 | struct btrfs_space_info *space_info; | |
0b246afa | 9585 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
ba1bf481 | 9586 | struct btrfs_device *device; |
6df9a95e | 9587 | struct btrfs_trans_handle *trans; |
cdcb725c | 9588 | u64 min_free; |
6719db6a JB |
9589 | u64 dev_min = 1; |
9590 | u64 dev_nr = 0; | |
4a5e98f5 | 9591 | u64 target; |
0305bc27 | 9592 | int debug; |
cdcb725c | 9593 | int index; |
ba1bf481 JB |
9594 | int full = 0; |
9595 | int ret = 0; | |
1a40e23b | 9596 | |
0b246afa | 9597 | debug = btrfs_test_opt(fs_info, ENOSPC_DEBUG); |
0305bc27 | 9598 | |
0b246afa | 9599 | block_group = btrfs_lookup_block_group(fs_info, bytenr); |
1a40e23b | 9600 | |
ba1bf481 | 9601 | /* odd, couldn't find the block group, leave it alone */ |
0305bc27 QW |
9602 | if (!block_group) { |
9603 | if (debug) | |
0b246afa | 9604 | btrfs_warn(fs_info, |
0305bc27 QW |
9605 | "can't find block group for bytenr %llu", |
9606 | bytenr); | |
ba1bf481 | 9607 | return -1; |
0305bc27 | 9608 | } |
1a40e23b | 9609 | |
cdcb725c | 9610 | min_free = btrfs_block_group_used(&block_group->item); |
9611 | ||
ba1bf481 | 9612 | /* no bytes used, we're good */ |
cdcb725c | 9613 | if (!min_free) |
1a40e23b ZY |
9614 | goto out; |
9615 | ||
ba1bf481 JB |
9616 | space_info = block_group->space_info; |
9617 | spin_lock(&space_info->lock); | |
17d217fe | 9618 | |
ba1bf481 | 9619 | full = space_info->full; |
17d217fe | 9620 | |
ba1bf481 JB |
9621 | /* |
9622 | * if this is the last block group we have in this space, we can't | |
7ce618db CM |
9623 | * relocate it unless we're able to allocate a new chunk below. |
9624 | * | |
9625 | * Otherwise, we need to make sure we have room in the space to handle | |
9626 | * all of the extents from this block group. If we can, we're good | |
ba1bf481 | 9627 | */ |
7ce618db | 9628 | if ((space_info->total_bytes != block_group->key.offset) && |
4136135b LB |
9629 | (btrfs_space_info_used(space_info, false) + min_free < |
9630 | space_info->total_bytes)) { | |
ba1bf481 JB |
9631 | spin_unlock(&space_info->lock); |
9632 | goto out; | |
17d217fe | 9633 | } |
ba1bf481 | 9634 | spin_unlock(&space_info->lock); |
ea8c2819 | 9635 | |
ba1bf481 JB |
9636 | /* |
9637 | * ok we don't have enough space, but maybe we have free space on our | |
9638 | * devices to allocate new chunks for relocation, so loop through our | |
4a5e98f5 ID |
9639 | * alloc devices and guess if we have enough space. if this block |
9640 | * group is going to be restriped, run checks against the target | |
9641 | * profile instead of the current one. | |
ba1bf481 JB |
9642 | */ |
9643 | ret = -1; | |
ea8c2819 | 9644 | |
cdcb725c | 9645 | /* |
9646 | * index: | |
9647 | * 0: raid10 | |
9648 | * 1: raid1 | |
9649 | * 2: dup | |
9650 | * 3: raid0 | |
9651 | * 4: single | |
9652 | */ | |
0b246afa | 9653 | target = get_restripe_target(fs_info, block_group->flags); |
4a5e98f5 | 9654 | if (target) { |
3e72ee88 | 9655 | index = btrfs_bg_flags_to_raid_index(extended_to_chunk(target)); |
4a5e98f5 ID |
9656 | } else { |
9657 | /* | |
9658 | * this is just a balance, so if we were marked as full | |
9659 | * we know there is no space for a new chunk | |
9660 | */ | |
0305bc27 QW |
9661 | if (full) { |
9662 | if (debug) | |
0b246afa JM |
9663 | btrfs_warn(fs_info, |
9664 | "no space to alloc new chunk for block group %llu", | |
9665 | block_group->key.objectid); | |
4a5e98f5 | 9666 | goto out; |
0305bc27 | 9667 | } |
4a5e98f5 | 9668 | |
3e72ee88 | 9669 | index = btrfs_bg_flags_to_raid_index(block_group->flags); |
4a5e98f5 ID |
9670 | } |
9671 | ||
e6ec716f | 9672 | if (index == BTRFS_RAID_RAID10) { |
cdcb725c | 9673 | dev_min = 4; |
6719db6a JB |
9674 | /* Divide by 2 */ |
9675 | min_free >>= 1; | |
e6ec716f | 9676 | } else if (index == BTRFS_RAID_RAID1) { |
cdcb725c | 9677 | dev_min = 2; |
e6ec716f | 9678 | } else if (index == BTRFS_RAID_DUP) { |
6719db6a JB |
9679 | /* Multiply by 2 */ |
9680 | min_free <<= 1; | |
e6ec716f | 9681 | } else if (index == BTRFS_RAID_RAID0) { |
cdcb725c | 9682 | dev_min = fs_devices->rw_devices; |
47c5713f | 9683 | min_free = div64_u64(min_free, dev_min); |
cdcb725c | 9684 | } |
9685 | ||
6df9a95e JB |
9686 | /* We need to do this so that we can look at pending chunks */ |
9687 | trans = btrfs_join_transaction(root); | |
9688 | if (IS_ERR(trans)) { | |
9689 | ret = PTR_ERR(trans); | |
9690 | goto out; | |
9691 | } | |
9692 | ||
0b246afa | 9693 | mutex_lock(&fs_info->chunk_mutex); |
ba1bf481 | 9694 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { |
7bfc837d | 9695 | u64 dev_offset; |
56bec294 | 9696 | |
ba1bf481 JB |
9697 | /* |
9698 | * check to make sure we can actually find a chunk with enough | |
9699 | * space to fit our block group in. | |
9700 | */ | |
63a212ab | 9701 | if (device->total_bytes > device->bytes_used + min_free && |
401e29c1 | 9702 | !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
6df9a95e | 9703 | ret = find_free_dev_extent(trans, device, min_free, |
7bfc837d | 9704 | &dev_offset, NULL); |
ba1bf481 | 9705 | if (!ret) |
cdcb725c | 9706 | dev_nr++; |
9707 | ||
9708 | if (dev_nr >= dev_min) | |
73e48b27 | 9709 | break; |
cdcb725c | 9710 | |
ba1bf481 | 9711 | ret = -1; |
725c8463 | 9712 | } |
edbd8d4e | 9713 | } |
0305bc27 | 9714 | if (debug && ret == -1) |
0b246afa JM |
9715 | btrfs_warn(fs_info, |
9716 | "no space to allocate a new chunk for block group %llu", | |
9717 | block_group->key.objectid); | |
9718 | mutex_unlock(&fs_info->chunk_mutex); | |
3a45bb20 | 9719 | btrfs_end_transaction(trans); |
edbd8d4e | 9720 | out: |
ba1bf481 | 9721 | btrfs_put_block_group(block_group); |
edbd8d4e CM |
9722 | return ret; |
9723 | } | |
9724 | ||
6bccf3ab JM |
9725 | static int find_first_block_group(struct btrfs_fs_info *fs_info, |
9726 | struct btrfs_path *path, | |
9727 | struct btrfs_key *key) | |
0b86a832 | 9728 | { |
6bccf3ab | 9729 | struct btrfs_root *root = fs_info->extent_root; |
925baedd | 9730 | int ret = 0; |
0b86a832 CM |
9731 | struct btrfs_key found_key; |
9732 | struct extent_buffer *leaf; | |
9733 | int slot; | |
edbd8d4e | 9734 | |
0b86a832 CM |
9735 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); |
9736 | if (ret < 0) | |
925baedd CM |
9737 | goto out; |
9738 | ||
d397712b | 9739 | while (1) { |
0b86a832 | 9740 | slot = path->slots[0]; |
edbd8d4e | 9741 | leaf = path->nodes[0]; |
0b86a832 CM |
9742 | if (slot >= btrfs_header_nritems(leaf)) { |
9743 | ret = btrfs_next_leaf(root, path); | |
9744 | if (ret == 0) | |
9745 | continue; | |
9746 | if (ret < 0) | |
925baedd | 9747 | goto out; |
0b86a832 | 9748 | break; |
edbd8d4e | 9749 | } |
0b86a832 | 9750 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
edbd8d4e | 9751 | |
0b86a832 | 9752 | if (found_key.objectid >= key->objectid && |
925baedd | 9753 | found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) { |
6fb37b75 LB |
9754 | struct extent_map_tree *em_tree; |
9755 | struct extent_map *em; | |
9756 | ||
9757 | em_tree = &root->fs_info->mapping_tree.map_tree; | |
9758 | read_lock(&em_tree->lock); | |
9759 | em = lookup_extent_mapping(em_tree, found_key.objectid, | |
9760 | found_key.offset); | |
9761 | read_unlock(&em_tree->lock); | |
9762 | if (!em) { | |
0b246afa | 9763 | btrfs_err(fs_info, |
6fb37b75 LB |
9764 | "logical %llu len %llu found bg but no related chunk", |
9765 | found_key.objectid, found_key.offset); | |
9766 | ret = -ENOENT; | |
9767 | } else { | |
9768 | ret = 0; | |
9769 | } | |
187ee58c | 9770 | free_extent_map(em); |
925baedd CM |
9771 | goto out; |
9772 | } | |
0b86a832 | 9773 | path->slots[0]++; |
edbd8d4e | 9774 | } |
925baedd | 9775 | out: |
0b86a832 | 9776 | return ret; |
edbd8d4e CM |
9777 | } |
9778 | ||
0af3d00b JB |
9779 | void btrfs_put_block_group_cache(struct btrfs_fs_info *info) |
9780 | { | |
9781 | struct btrfs_block_group_cache *block_group; | |
9782 | u64 last = 0; | |
9783 | ||
9784 | while (1) { | |
9785 | struct inode *inode; | |
9786 | ||
9787 | block_group = btrfs_lookup_first_block_group(info, last); | |
9788 | while (block_group) { | |
9789 | spin_lock(&block_group->lock); | |
9790 | if (block_group->iref) | |
9791 | break; | |
9792 | spin_unlock(&block_group->lock); | |
2ff7e61e | 9793 | block_group = next_block_group(info, block_group); |
0af3d00b JB |
9794 | } |
9795 | if (!block_group) { | |
9796 | if (last == 0) | |
9797 | break; | |
9798 | last = 0; | |
9799 | continue; | |
9800 | } | |
9801 | ||
9802 | inode = block_group->inode; | |
9803 | block_group->iref = 0; | |
9804 | block_group->inode = NULL; | |
9805 | spin_unlock(&block_group->lock); | |
f3bca802 | 9806 | ASSERT(block_group->io_ctl.inode == NULL); |
0af3d00b JB |
9807 | iput(inode); |
9808 | last = block_group->key.objectid + block_group->key.offset; | |
9809 | btrfs_put_block_group(block_group); | |
9810 | } | |
9811 | } | |
9812 | ||
5cdd7db6 FM |
9813 | /* |
9814 | * Must be called only after stopping all workers, since we could have block | |
9815 | * group caching kthreads running, and therefore they could race with us if we | |
9816 | * freed the block groups before stopping them. | |
9817 | */ | |
1a40e23b ZY |
9818 | int btrfs_free_block_groups(struct btrfs_fs_info *info) |
9819 | { | |
9820 | struct btrfs_block_group_cache *block_group; | |
4184ea7f | 9821 | struct btrfs_space_info *space_info; |
11833d66 | 9822 | struct btrfs_caching_control *caching_ctl; |
1a40e23b ZY |
9823 | struct rb_node *n; |
9824 | ||
9e351cc8 | 9825 | down_write(&info->commit_root_sem); |
11833d66 YZ |
9826 | while (!list_empty(&info->caching_block_groups)) { |
9827 | caching_ctl = list_entry(info->caching_block_groups.next, | |
9828 | struct btrfs_caching_control, list); | |
9829 | list_del(&caching_ctl->list); | |
9830 | put_caching_control(caching_ctl); | |
9831 | } | |
9e351cc8 | 9832 | up_write(&info->commit_root_sem); |
11833d66 | 9833 | |
47ab2a6c JB |
9834 | spin_lock(&info->unused_bgs_lock); |
9835 | while (!list_empty(&info->unused_bgs)) { | |
9836 | block_group = list_first_entry(&info->unused_bgs, | |
9837 | struct btrfs_block_group_cache, | |
9838 | bg_list); | |
9839 | list_del_init(&block_group->bg_list); | |
9840 | btrfs_put_block_group(block_group); | |
9841 | } | |
9842 | spin_unlock(&info->unused_bgs_lock); | |
9843 | ||
1a40e23b ZY |
9844 | spin_lock(&info->block_group_cache_lock); |
9845 | while ((n = rb_last(&info->block_group_cache_tree)) != NULL) { | |
9846 | block_group = rb_entry(n, struct btrfs_block_group_cache, | |
9847 | cache_node); | |
1a40e23b ZY |
9848 | rb_erase(&block_group->cache_node, |
9849 | &info->block_group_cache_tree); | |
01eacb27 | 9850 | RB_CLEAR_NODE(&block_group->cache_node); |
d899e052 YZ |
9851 | spin_unlock(&info->block_group_cache_lock); |
9852 | ||
80eb234a | 9853 | down_write(&block_group->space_info->groups_sem); |
1a40e23b | 9854 | list_del(&block_group->list); |
80eb234a | 9855 | up_write(&block_group->space_info->groups_sem); |
d2fb3437 | 9856 | |
3c14874a JB |
9857 | /* |
9858 | * We haven't cached this block group, which means we could | |
9859 | * possibly have excluded extents on this block group. | |
9860 | */ | |
36cce922 JB |
9861 | if (block_group->cached == BTRFS_CACHE_NO || |
9862 | block_group->cached == BTRFS_CACHE_ERROR) | |
9e715da8 | 9863 | free_excluded_extents(block_group); |
3c14874a | 9864 | |
817d52f8 | 9865 | btrfs_remove_free_space_cache(block_group); |
5cdd7db6 | 9866 | ASSERT(block_group->cached != BTRFS_CACHE_STARTED); |
f3bca802 LB |
9867 | ASSERT(list_empty(&block_group->dirty_list)); |
9868 | ASSERT(list_empty(&block_group->io_list)); | |
9869 | ASSERT(list_empty(&block_group->bg_list)); | |
9870 | ASSERT(atomic_read(&block_group->count) == 1); | |
11dfe35a | 9871 | btrfs_put_block_group(block_group); |
d899e052 YZ |
9872 | |
9873 | spin_lock(&info->block_group_cache_lock); | |
1a40e23b ZY |
9874 | } |
9875 | spin_unlock(&info->block_group_cache_lock); | |
4184ea7f CM |
9876 | |
9877 | /* now that all the block groups are freed, go through and | |
9878 | * free all the space_info structs. This is only called during | |
9879 | * the final stages of unmount, and so we know nobody is | |
9880 | * using them. We call synchronize_rcu() once before we start, | |
9881 | * just to be on the safe side. | |
9882 | */ | |
9883 | synchronize_rcu(); | |
9884 | ||
8929ecfa YZ |
9885 | release_global_block_rsv(info); |
9886 | ||
67871254 | 9887 | while (!list_empty(&info->space_info)) { |
6ab0a202 JM |
9888 | int i; |
9889 | ||
4184ea7f CM |
9890 | space_info = list_entry(info->space_info.next, |
9891 | struct btrfs_space_info, | |
9892 | list); | |
d555b6c3 JB |
9893 | |
9894 | /* | |
9895 | * Do not hide this behind enospc_debug, this is actually | |
9896 | * important and indicates a real bug if this happens. | |
9897 | */ | |
9898 | if (WARN_ON(space_info->bytes_pinned > 0 || | |
b069e0c3 | 9899 | space_info->bytes_reserved > 0 || |
d555b6c3 | 9900 | space_info->bytes_may_use > 0)) |
ab8d0fc4 | 9901 | dump_space_info(info, space_info, 0, 0); |
4184ea7f | 9902 | list_del(&space_info->list); |
6ab0a202 JM |
9903 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { |
9904 | struct kobject *kobj; | |
c1895442 JM |
9905 | kobj = space_info->block_group_kobjs[i]; |
9906 | space_info->block_group_kobjs[i] = NULL; | |
9907 | if (kobj) { | |
6ab0a202 JM |
9908 | kobject_del(kobj); |
9909 | kobject_put(kobj); | |
9910 | } | |
9911 | } | |
9912 | kobject_del(&space_info->kobj); | |
9913 | kobject_put(&space_info->kobj); | |
4184ea7f | 9914 | } |
1a40e23b ZY |
9915 | return 0; |
9916 | } | |
9917 | ||
75cb379d JM |
9918 | /* link_block_group will queue up kobjects to add when we're reclaim-safe */ |
9919 | void btrfs_add_raid_kobjects(struct btrfs_fs_info *fs_info) | |
9920 | { | |
9921 | struct btrfs_space_info *space_info; | |
9922 | struct raid_kobject *rkobj; | |
9923 | LIST_HEAD(list); | |
9924 | int index; | |
9925 | int ret = 0; | |
9926 | ||
9927 | spin_lock(&fs_info->pending_raid_kobjs_lock); | |
9928 | list_splice_init(&fs_info->pending_raid_kobjs, &list); | |
9929 | spin_unlock(&fs_info->pending_raid_kobjs_lock); | |
9930 | ||
9931 | list_for_each_entry(rkobj, &list, list) { | |
9932 | space_info = __find_space_info(fs_info, rkobj->flags); | |
9933 | index = btrfs_bg_flags_to_raid_index(rkobj->flags); | |
9934 | ||
9935 | ret = kobject_add(&rkobj->kobj, &space_info->kobj, | |
9936 | "%s", get_raid_name(index)); | |
9937 | if (ret) { | |
9938 | kobject_put(&rkobj->kobj); | |
9939 | break; | |
9940 | } | |
9941 | } | |
9942 | if (ret) | |
9943 | btrfs_warn(fs_info, | |
9944 | "failed to add kobject for block cache, ignoring"); | |
9945 | } | |
9946 | ||
c434d21c | 9947 | static void link_block_group(struct btrfs_block_group_cache *cache) |
b742bb82 | 9948 | { |
c434d21c | 9949 | struct btrfs_space_info *space_info = cache->space_info; |
75cb379d | 9950 | struct btrfs_fs_info *fs_info = cache->fs_info; |
3e72ee88 | 9951 | int index = btrfs_bg_flags_to_raid_index(cache->flags); |
ed55b6ac | 9952 | bool first = false; |
b742bb82 YZ |
9953 | |
9954 | down_write(&space_info->groups_sem); | |
ed55b6ac JM |
9955 | if (list_empty(&space_info->block_groups[index])) |
9956 | first = true; | |
9957 | list_add_tail(&cache->list, &space_info->block_groups[index]); | |
9958 | up_write(&space_info->groups_sem); | |
9959 | ||
9960 | if (first) { | |
75cb379d JM |
9961 | struct raid_kobject *rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS); |
9962 | if (!rkobj) { | |
9963 | btrfs_warn(cache->fs_info, | |
9964 | "couldn't alloc memory for raid level kobject"); | |
9965 | return; | |
6ab0a202 | 9966 | } |
75cb379d JM |
9967 | rkobj->flags = cache->flags; |
9968 | kobject_init(&rkobj->kobj, &btrfs_raid_ktype); | |
9969 | ||
9970 | spin_lock(&fs_info->pending_raid_kobjs_lock); | |
9971 | list_add_tail(&rkobj->list, &fs_info->pending_raid_kobjs); | |
9972 | spin_unlock(&fs_info->pending_raid_kobjs_lock); | |
c1895442 | 9973 | space_info->block_group_kobjs[index] = &rkobj->kobj; |
6ab0a202 | 9974 | } |
b742bb82 YZ |
9975 | } |
9976 | ||
920e4a58 | 9977 | static struct btrfs_block_group_cache * |
2ff7e61e JM |
9978 | btrfs_create_block_group_cache(struct btrfs_fs_info *fs_info, |
9979 | u64 start, u64 size) | |
920e4a58 MX |
9980 | { |
9981 | struct btrfs_block_group_cache *cache; | |
9982 | ||
9983 | cache = kzalloc(sizeof(*cache), GFP_NOFS); | |
9984 | if (!cache) | |
9985 | return NULL; | |
9986 | ||
9987 | cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl), | |
9988 | GFP_NOFS); | |
9989 | if (!cache->free_space_ctl) { | |
9990 | kfree(cache); | |
9991 | return NULL; | |
9992 | } | |
9993 | ||
9994 | cache->key.objectid = start; | |
9995 | cache->key.offset = size; | |
9996 | cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; | |
9997 | ||
0b246afa | 9998 | cache->fs_info = fs_info; |
e4ff5fb5 | 9999 | cache->full_stripe_len = btrfs_full_stripe_len(fs_info, start); |
1e144fb8 OS |
10000 | set_free_space_tree_thresholds(cache); |
10001 | ||
920e4a58 MX |
10002 | atomic_set(&cache->count, 1); |
10003 | spin_lock_init(&cache->lock); | |
e570fd27 | 10004 | init_rwsem(&cache->data_rwsem); |
920e4a58 MX |
10005 | INIT_LIST_HEAD(&cache->list); |
10006 | INIT_LIST_HEAD(&cache->cluster_list); | |
47ab2a6c | 10007 | INIT_LIST_HEAD(&cache->bg_list); |
633c0aad | 10008 | INIT_LIST_HEAD(&cache->ro_list); |
ce93ec54 | 10009 | INIT_LIST_HEAD(&cache->dirty_list); |
c9dc4c65 | 10010 | INIT_LIST_HEAD(&cache->io_list); |
920e4a58 | 10011 | btrfs_init_free_space_ctl(cache); |
04216820 | 10012 | atomic_set(&cache->trimming, 0); |
a5ed9182 | 10013 | mutex_init(&cache->free_space_lock); |
0966a7b1 | 10014 | btrfs_init_full_stripe_locks_tree(&cache->full_stripe_locks_root); |
920e4a58 MX |
10015 | |
10016 | return cache; | |
10017 | } | |
10018 | ||
5b4aacef | 10019 | int btrfs_read_block_groups(struct btrfs_fs_info *info) |
9078a3e1 CM |
10020 | { |
10021 | struct btrfs_path *path; | |
10022 | int ret; | |
9078a3e1 | 10023 | struct btrfs_block_group_cache *cache; |
6324fbf3 | 10024 | struct btrfs_space_info *space_info; |
9078a3e1 CM |
10025 | struct btrfs_key key; |
10026 | struct btrfs_key found_key; | |
5f39d397 | 10027 | struct extent_buffer *leaf; |
0af3d00b JB |
10028 | int need_clear = 0; |
10029 | u64 cache_gen; | |
49303381 LB |
10030 | u64 feature; |
10031 | int mixed; | |
10032 | ||
10033 | feature = btrfs_super_incompat_flags(info->super_copy); | |
10034 | mixed = !!(feature & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS); | |
96b5179d | 10035 | |
9078a3e1 | 10036 | key.objectid = 0; |
0b86a832 | 10037 | key.offset = 0; |
962a298f | 10038 | key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; |
9078a3e1 CM |
10039 | path = btrfs_alloc_path(); |
10040 | if (!path) | |
10041 | return -ENOMEM; | |
e4058b54 | 10042 | path->reada = READA_FORWARD; |
9078a3e1 | 10043 | |
0b246afa JM |
10044 | cache_gen = btrfs_super_cache_generation(info->super_copy); |
10045 | if (btrfs_test_opt(info, SPACE_CACHE) && | |
10046 | btrfs_super_generation(info->super_copy) != cache_gen) | |
0af3d00b | 10047 | need_clear = 1; |
0b246afa | 10048 | if (btrfs_test_opt(info, CLEAR_CACHE)) |
88c2ba3b | 10049 | need_clear = 1; |
0af3d00b | 10050 | |
d397712b | 10051 | while (1) { |
6bccf3ab | 10052 | ret = find_first_block_group(info, path, &key); |
b742bb82 YZ |
10053 | if (ret > 0) |
10054 | break; | |
0b86a832 CM |
10055 | if (ret != 0) |
10056 | goto error; | |
920e4a58 | 10057 | |
5f39d397 CM |
10058 | leaf = path->nodes[0]; |
10059 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
920e4a58 | 10060 | |
2ff7e61e | 10061 | cache = btrfs_create_block_group_cache(info, found_key.objectid, |
920e4a58 | 10062 | found_key.offset); |
9078a3e1 | 10063 | if (!cache) { |
0b86a832 | 10064 | ret = -ENOMEM; |
f0486c68 | 10065 | goto error; |
9078a3e1 | 10066 | } |
96303081 | 10067 | |
cf7c1ef6 LB |
10068 | if (need_clear) { |
10069 | /* | |
10070 | * When we mount with old space cache, we need to | |
10071 | * set BTRFS_DC_CLEAR and set dirty flag. | |
10072 | * | |
10073 | * a) Setting 'BTRFS_DC_CLEAR' makes sure that we | |
10074 | * truncate the old free space cache inode and | |
10075 | * setup a new one. | |
10076 | * b) Setting 'dirty flag' makes sure that we flush | |
10077 | * the new space cache info onto disk. | |
10078 | */ | |
0b246afa | 10079 | if (btrfs_test_opt(info, SPACE_CACHE)) |
ce93ec54 | 10080 | cache->disk_cache_state = BTRFS_DC_CLEAR; |
cf7c1ef6 | 10081 | } |
0af3d00b | 10082 | |
5f39d397 CM |
10083 | read_extent_buffer(leaf, &cache->item, |
10084 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
10085 | sizeof(cache->item)); | |
920e4a58 | 10086 | cache->flags = btrfs_block_group_flags(&cache->item); |
49303381 LB |
10087 | if (!mixed && |
10088 | ((cache->flags & BTRFS_BLOCK_GROUP_METADATA) && | |
10089 | (cache->flags & BTRFS_BLOCK_GROUP_DATA))) { | |
10090 | btrfs_err(info, | |
10091 | "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups", | |
10092 | cache->key.objectid); | |
10093 | ret = -EINVAL; | |
10094 | goto error; | |
10095 | } | |
0b86a832 | 10096 | |
9078a3e1 | 10097 | key.objectid = found_key.objectid + found_key.offset; |
b3b4aa74 | 10098 | btrfs_release_path(path); |
34d52cb6 | 10099 | |
3c14874a JB |
10100 | /* |
10101 | * We need to exclude the super stripes now so that the space | |
10102 | * info has super bytes accounted for, otherwise we'll think | |
10103 | * we have more space than we actually do. | |
10104 | */ | |
3c4da657 | 10105 | ret = exclude_super_stripes(cache); |
835d974f JB |
10106 | if (ret) { |
10107 | /* | |
10108 | * We may have excluded something, so call this just in | |
10109 | * case. | |
10110 | */ | |
9e715da8 | 10111 | free_excluded_extents(cache); |
920e4a58 | 10112 | btrfs_put_block_group(cache); |
835d974f JB |
10113 | goto error; |
10114 | } | |
3c14874a | 10115 | |
817d52f8 JB |
10116 | /* |
10117 | * check for two cases, either we are full, and therefore | |
10118 | * don't need to bother with the caching work since we won't | |
10119 | * find any space, or we are empty, and we can just add all | |
10120 | * the space in and be done with it. This saves us _alot_ of | |
10121 | * time, particularly in the full case. | |
10122 | */ | |
10123 | if (found_key.offset == btrfs_block_group_used(&cache->item)) { | |
11833d66 | 10124 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 10125 | cache->cached = BTRFS_CACHE_FINISHED; |
9e715da8 | 10126 | free_excluded_extents(cache); |
817d52f8 | 10127 | } else if (btrfs_block_group_used(&cache->item) == 0) { |
11833d66 | 10128 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 10129 | cache->cached = BTRFS_CACHE_FINISHED; |
4457c1c7 | 10130 | add_new_free_space(cache, found_key.objectid, |
817d52f8 JB |
10131 | found_key.objectid + |
10132 | found_key.offset); | |
9e715da8 | 10133 | free_excluded_extents(cache); |
817d52f8 | 10134 | } |
96b5179d | 10135 | |
0b246afa | 10136 | ret = btrfs_add_block_group_cache(info, cache); |
8c579fe7 JB |
10137 | if (ret) { |
10138 | btrfs_remove_free_space_cache(cache); | |
10139 | btrfs_put_block_group(cache); | |
10140 | goto error; | |
10141 | } | |
10142 | ||
0b246afa | 10143 | trace_btrfs_add_block_group(info, cache, 0); |
d2006e6d NB |
10144 | update_space_info(info, cache->flags, found_key.offset, |
10145 | btrfs_block_group_used(&cache->item), | |
10146 | cache->bytes_super, &space_info); | |
8c579fe7 | 10147 | |
6324fbf3 | 10148 | cache->space_info = space_info; |
1b2da372 | 10149 | |
c434d21c | 10150 | link_block_group(cache); |
0f9dd46c | 10151 | |
0b246afa | 10152 | set_avail_alloc_bits(info, cache->flags); |
2ff7e61e | 10153 | if (btrfs_chunk_readonly(info, cache->key.objectid)) { |
868f401a | 10154 | inc_block_group_ro(cache, 1); |
47ab2a6c JB |
10155 | } else if (btrfs_block_group_used(&cache->item) == 0) { |
10156 | spin_lock(&info->unused_bgs_lock); | |
10157 | /* Should always be true but just in case. */ | |
10158 | if (list_empty(&cache->bg_list)) { | |
10159 | btrfs_get_block_group(cache); | |
4ed0a7a3 | 10160 | trace_btrfs_add_unused_block_group(cache); |
47ab2a6c JB |
10161 | list_add_tail(&cache->bg_list, |
10162 | &info->unused_bgs); | |
10163 | } | |
10164 | spin_unlock(&info->unused_bgs_lock); | |
10165 | } | |
9078a3e1 | 10166 | } |
b742bb82 | 10167 | |
0b246afa | 10168 | list_for_each_entry_rcu(space_info, &info->space_info, list) { |
2ff7e61e | 10169 | if (!(get_alloc_profile(info, space_info->flags) & |
b742bb82 YZ |
10170 | (BTRFS_BLOCK_GROUP_RAID10 | |
10171 | BTRFS_BLOCK_GROUP_RAID1 | | |
53b381b3 DW |
10172 | BTRFS_BLOCK_GROUP_RAID5 | |
10173 | BTRFS_BLOCK_GROUP_RAID6 | | |
b742bb82 YZ |
10174 | BTRFS_BLOCK_GROUP_DUP))) |
10175 | continue; | |
10176 | /* | |
10177 | * avoid allocating from un-mirrored block group if there are | |
10178 | * mirrored block groups. | |
10179 | */ | |
1095cc0d | 10180 | list_for_each_entry(cache, |
10181 | &space_info->block_groups[BTRFS_RAID_RAID0], | |
10182 | list) | |
868f401a | 10183 | inc_block_group_ro(cache, 1); |
1095cc0d | 10184 | list_for_each_entry(cache, |
10185 | &space_info->block_groups[BTRFS_RAID_SINGLE], | |
10186 | list) | |
868f401a | 10187 | inc_block_group_ro(cache, 1); |
9078a3e1 | 10188 | } |
f0486c68 | 10189 | |
75cb379d | 10190 | btrfs_add_raid_kobjects(info); |
f0486c68 | 10191 | init_global_block_rsv(info); |
0b86a832 CM |
10192 | ret = 0; |
10193 | error: | |
9078a3e1 | 10194 | btrfs_free_path(path); |
0b86a832 | 10195 | return ret; |
9078a3e1 | 10196 | } |
6324fbf3 | 10197 | |
6c686b35 | 10198 | void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans) |
ea658bad | 10199 | { |
6c686b35 | 10200 | struct btrfs_fs_info *fs_info = trans->fs_info; |
ea658bad | 10201 | struct btrfs_block_group_cache *block_group, *tmp; |
0b246afa | 10202 | struct btrfs_root *extent_root = fs_info->extent_root; |
ea658bad JB |
10203 | struct btrfs_block_group_item item; |
10204 | struct btrfs_key key; | |
10205 | int ret = 0; | |
d9a0540a | 10206 | bool can_flush_pending_bgs = trans->can_flush_pending_bgs; |
ea658bad | 10207 | |
d9a0540a | 10208 | trans->can_flush_pending_bgs = false; |
47ab2a6c | 10209 | list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) { |
ea658bad | 10210 | if (ret) |
c92f6be3 | 10211 | goto next; |
ea658bad JB |
10212 | |
10213 | spin_lock(&block_group->lock); | |
10214 | memcpy(&item, &block_group->item, sizeof(item)); | |
10215 | memcpy(&key, &block_group->key, sizeof(key)); | |
10216 | spin_unlock(&block_group->lock); | |
10217 | ||
10218 | ret = btrfs_insert_item(trans, extent_root, &key, &item, | |
10219 | sizeof(item)); | |
10220 | if (ret) | |
66642832 | 10221 | btrfs_abort_transaction(trans, ret); |
0b246afa JM |
10222 | ret = btrfs_finish_chunk_alloc(trans, fs_info, key.objectid, |
10223 | key.offset); | |
6df9a95e | 10224 | if (ret) |
66642832 | 10225 | btrfs_abort_transaction(trans, ret); |
e4e0711c | 10226 | add_block_group_free_space(trans, block_group); |
1e144fb8 | 10227 | /* already aborted the transaction if it failed. */ |
c92f6be3 FM |
10228 | next: |
10229 | list_del_init(&block_group->bg_list); | |
ea658bad | 10230 | } |
d9a0540a | 10231 | trans->can_flush_pending_bgs = can_flush_pending_bgs; |
ea658bad JB |
10232 | } |
10233 | ||
e7e02096 | 10234 | int btrfs_make_block_group(struct btrfs_trans_handle *trans, u64 bytes_used, |
0174484d | 10235 | u64 type, u64 chunk_offset, u64 size) |
6324fbf3 | 10236 | { |
e7e02096 | 10237 | struct btrfs_fs_info *fs_info = trans->fs_info; |
6324fbf3 | 10238 | struct btrfs_block_group_cache *cache; |
0b246afa | 10239 | int ret; |
6324fbf3 | 10240 | |
0b246afa | 10241 | btrfs_set_log_full_commit(fs_info, trans); |
e02119d5 | 10242 | |
2ff7e61e | 10243 | cache = btrfs_create_block_group_cache(fs_info, chunk_offset, size); |
0f9dd46c JB |
10244 | if (!cache) |
10245 | return -ENOMEM; | |
34d52cb6 | 10246 | |
6324fbf3 | 10247 | btrfs_set_block_group_used(&cache->item, bytes_used); |
0174484d NB |
10248 | btrfs_set_block_group_chunk_objectid(&cache->item, |
10249 | BTRFS_FIRST_CHUNK_TREE_OBJECTID); | |
6324fbf3 CM |
10250 | btrfs_set_block_group_flags(&cache->item, type); |
10251 | ||
920e4a58 | 10252 | cache->flags = type; |
11833d66 | 10253 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 10254 | cache->cached = BTRFS_CACHE_FINISHED; |
1e144fb8 | 10255 | cache->needs_free_space = 1; |
3c4da657 | 10256 | ret = exclude_super_stripes(cache); |
835d974f JB |
10257 | if (ret) { |
10258 | /* | |
10259 | * We may have excluded something, so call this just in | |
10260 | * case. | |
10261 | */ | |
9e715da8 | 10262 | free_excluded_extents(cache); |
920e4a58 | 10263 | btrfs_put_block_group(cache); |
835d974f JB |
10264 | return ret; |
10265 | } | |
96303081 | 10266 | |
4457c1c7 | 10267 | add_new_free_space(cache, chunk_offset, chunk_offset + size); |
817d52f8 | 10268 | |
9e715da8 | 10269 | free_excluded_extents(cache); |
11833d66 | 10270 | |
d0bd4560 | 10271 | #ifdef CONFIG_BTRFS_DEBUG |
2ff7e61e | 10272 | if (btrfs_should_fragment_free_space(cache)) { |
d0bd4560 JB |
10273 | u64 new_bytes_used = size - bytes_used; |
10274 | ||
10275 | bytes_used += new_bytes_used >> 1; | |
2ff7e61e | 10276 | fragment_free_space(cache); |
d0bd4560 JB |
10277 | } |
10278 | #endif | |
2e6e5183 | 10279 | /* |
2be12ef7 NB |
10280 | * Ensure the corresponding space_info object is created and |
10281 | * assigned to our block group. We want our bg to be added to the rbtree | |
10282 | * with its ->space_info set. | |
2e6e5183 | 10283 | */ |
2be12ef7 | 10284 | cache->space_info = __find_space_info(fs_info, cache->flags); |
dc2d3005 | 10285 | ASSERT(cache->space_info); |
2e6e5183 | 10286 | |
0b246afa | 10287 | ret = btrfs_add_block_group_cache(fs_info, cache); |
8c579fe7 JB |
10288 | if (ret) { |
10289 | btrfs_remove_free_space_cache(cache); | |
10290 | btrfs_put_block_group(cache); | |
10291 | return ret; | |
10292 | } | |
10293 | ||
2e6e5183 FM |
10294 | /* |
10295 | * Now that our block group has its ->space_info set and is inserted in | |
10296 | * the rbtree, update the space info's counters. | |
10297 | */ | |
0b246afa | 10298 | trace_btrfs_add_block_group(fs_info, cache, 1); |
d2006e6d | 10299 | update_space_info(fs_info, cache->flags, size, bytes_used, |
e40edf2d | 10300 | cache->bytes_super, &cache->space_info); |
0b246afa | 10301 | update_global_block_rsv(fs_info); |
1b2da372 | 10302 | |
c434d21c | 10303 | link_block_group(cache); |
6324fbf3 | 10304 | |
47ab2a6c | 10305 | list_add_tail(&cache->bg_list, &trans->new_bgs); |
6324fbf3 | 10306 | |
0b246afa | 10307 | set_avail_alloc_bits(fs_info, type); |
6324fbf3 CM |
10308 | return 0; |
10309 | } | |
1a40e23b | 10310 | |
10ea00f5 ID |
10311 | static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
10312 | { | |
899c81ea ID |
10313 | u64 extra_flags = chunk_to_extended(flags) & |
10314 | BTRFS_EXTENDED_PROFILE_MASK; | |
10ea00f5 | 10315 | |
de98ced9 | 10316 | write_seqlock(&fs_info->profiles_lock); |
10ea00f5 ID |
10317 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
10318 | fs_info->avail_data_alloc_bits &= ~extra_flags; | |
10319 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
10320 | fs_info->avail_metadata_alloc_bits &= ~extra_flags; | |
10321 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
10322 | fs_info->avail_system_alloc_bits &= ~extra_flags; | |
de98ced9 | 10323 | write_sequnlock(&fs_info->profiles_lock); |
10ea00f5 ID |
10324 | } |
10325 | ||
1a40e23b | 10326 | int btrfs_remove_block_group(struct btrfs_trans_handle *trans, |
5a98ec01 | 10327 | u64 group_start, struct extent_map *em) |
1a40e23b | 10328 | { |
5a98ec01 | 10329 | struct btrfs_fs_info *fs_info = trans->fs_info; |
6bccf3ab | 10330 | struct btrfs_root *root = fs_info->extent_root; |
1a40e23b ZY |
10331 | struct btrfs_path *path; |
10332 | struct btrfs_block_group_cache *block_group; | |
44fb5511 | 10333 | struct btrfs_free_cluster *cluster; |
0b246afa | 10334 | struct btrfs_root *tree_root = fs_info->tree_root; |
1a40e23b | 10335 | struct btrfs_key key; |
0af3d00b | 10336 | struct inode *inode; |
c1895442 | 10337 | struct kobject *kobj = NULL; |
1a40e23b | 10338 | int ret; |
10ea00f5 | 10339 | int index; |
89a55897 | 10340 | int factor; |
4f69cb98 | 10341 | struct btrfs_caching_control *caching_ctl = NULL; |
04216820 | 10342 | bool remove_em; |
1a40e23b | 10343 | |
6bccf3ab | 10344 | block_group = btrfs_lookup_block_group(fs_info, group_start); |
1a40e23b | 10345 | BUG_ON(!block_group); |
c146afad | 10346 | BUG_ON(!block_group->ro); |
1a40e23b | 10347 | |
4ed0a7a3 | 10348 | trace_btrfs_remove_block_group(block_group); |
9f7c43c9 | 10349 | /* |
10350 | * Free the reserved super bytes from this block group before | |
10351 | * remove it. | |
10352 | */ | |
9e715da8 | 10353 | free_excluded_extents(block_group); |
fd708b81 JB |
10354 | btrfs_free_ref_tree_range(fs_info, block_group->key.objectid, |
10355 | block_group->key.offset); | |
9f7c43c9 | 10356 | |
1a40e23b | 10357 | memcpy(&key, &block_group->key, sizeof(key)); |
3e72ee88 | 10358 | index = btrfs_bg_flags_to_raid_index(block_group->flags); |
89a55897 JB |
10359 | if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP | |
10360 | BTRFS_BLOCK_GROUP_RAID1 | | |
10361 | BTRFS_BLOCK_GROUP_RAID10)) | |
10362 | factor = 2; | |
10363 | else | |
10364 | factor = 1; | |
1a40e23b | 10365 | |
44fb5511 | 10366 | /* make sure this block group isn't part of an allocation cluster */ |
0b246afa | 10367 | cluster = &fs_info->data_alloc_cluster; |
44fb5511 CM |
10368 | spin_lock(&cluster->refill_lock); |
10369 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
10370 | spin_unlock(&cluster->refill_lock); | |
10371 | ||
10372 | /* | |
10373 | * make sure this block group isn't part of a metadata | |
10374 | * allocation cluster | |
10375 | */ | |
0b246afa | 10376 | cluster = &fs_info->meta_alloc_cluster; |
44fb5511 CM |
10377 | spin_lock(&cluster->refill_lock); |
10378 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
10379 | spin_unlock(&cluster->refill_lock); | |
10380 | ||
1a40e23b | 10381 | path = btrfs_alloc_path(); |
d8926bb3 MF |
10382 | if (!path) { |
10383 | ret = -ENOMEM; | |
10384 | goto out; | |
10385 | } | |
1a40e23b | 10386 | |
1bbc621e CM |
10387 | /* |
10388 | * get the inode first so any iput calls done for the io_list | |
10389 | * aren't the final iput (no unlinks allowed now) | |
10390 | */ | |
77ab86bf | 10391 | inode = lookup_free_space_inode(fs_info, block_group, path); |
1bbc621e CM |
10392 | |
10393 | mutex_lock(&trans->transaction->cache_write_mutex); | |
10394 | /* | |
10395 | * make sure our free spache cache IO is done before remove the | |
10396 | * free space inode | |
10397 | */ | |
10398 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
10399 | if (!list_empty(&block_group->io_list)) { | |
10400 | list_del_init(&block_group->io_list); | |
10401 | ||
10402 | WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode); | |
10403 | ||
10404 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
afdb5718 | 10405 | btrfs_wait_cache_io(trans, block_group, path); |
1bbc621e CM |
10406 | btrfs_put_block_group(block_group); |
10407 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
10408 | } | |
10409 | ||
10410 | if (!list_empty(&block_group->dirty_list)) { | |
10411 | list_del_init(&block_group->dirty_list); | |
10412 | btrfs_put_block_group(block_group); | |
10413 | } | |
10414 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
10415 | mutex_unlock(&trans->transaction->cache_write_mutex); | |
10416 | ||
0af3d00b | 10417 | if (!IS_ERR(inode)) { |
73f2e545 | 10418 | ret = btrfs_orphan_add(trans, BTRFS_I(inode)); |
79787eaa JM |
10419 | if (ret) { |
10420 | btrfs_add_delayed_iput(inode); | |
10421 | goto out; | |
10422 | } | |
0af3d00b JB |
10423 | clear_nlink(inode); |
10424 | /* One for the block groups ref */ | |
10425 | spin_lock(&block_group->lock); | |
10426 | if (block_group->iref) { | |
10427 | block_group->iref = 0; | |
10428 | block_group->inode = NULL; | |
10429 | spin_unlock(&block_group->lock); | |
10430 | iput(inode); | |
10431 | } else { | |
10432 | spin_unlock(&block_group->lock); | |
10433 | } | |
10434 | /* One for our lookup ref */ | |
455757c3 | 10435 | btrfs_add_delayed_iput(inode); |
0af3d00b JB |
10436 | } |
10437 | ||
10438 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
10439 | key.offset = block_group->key.objectid; | |
10440 | key.type = 0; | |
10441 | ||
10442 | ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1); | |
10443 | if (ret < 0) | |
10444 | goto out; | |
10445 | if (ret > 0) | |
b3b4aa74 | 10446 | btrfs_release_path(path); |
0af3d00b JB |
10447 | if (ret == 0) { |
10448 | ret = btrfs_del_item(trans, tree_root, path); | |
10449 | if (ret) | |
10450 | goto out; | |
b3b4aa74 | 10451 | btrfs_release_path(path); |
0af3d00b JB |
10452 | } |
10453 | ||
0b246afa | 10454 | spin_lock(&fs_info->block_group_cache_lock); |
1a40e23b | 10455 | rb_erase(&block_group->cache_node, |
0b246afa | 10456 | &fs_info->block_group_cache_tree); |
292cbd51 | 10457 | RB_CLEAR_NODE(&block_group->cache_node); |
a1897fdd | 10458 | |
0b246afa JM |
10459 | if (fs_info->first_logical_byte == block_group->key.objectid) |
10460 | fs_info->first_logical_byte = (u64)-1; | |
10461 | spin_unlock(&fs_info->block_group_cache_lock); | |
817d52f8 | 10462 | |
80eb234a | 10463 | down_write(&block_group->space_info->groups_sem); |
44fb5511 CM |
10464 | /* |
10465 | * we must use list_del_init so people can check to see if they | |
10466 | * are still on the list after taking the semaphore | |
10467 | */ | |
10468 | list_del_init(&block_group->list); | |
6ab0a202 | 10469 | if (list_empty(&block_group->space_info->block_groups[index])) { |
c1895442 JM |
10470 | kobj = block_group->space_info->block_group_kobjs[index]; |
10471 | block_group->space_info->block_group_kobjs[index] = NULL; | |
0b246afa | 10472 | clear_avail_alloc_bits(fs_info, block_group->flags); |
6ab0a202 | 10473 | } |
80eb234a | 10474 | up_write(&block_group->space_info->groups_sem); |
c1895442 JM |
10475 | if (kobj) { |
10476 | kobject_del(kobj); | |
10477 | kobject_put(kobj); | |
10478 | } | |
1a40e23b | 10479 | |
4f69cb98 FM |
10480 | if (block_group->has_caching_ctl) |
10481 | caching_ctl = get_caching_control(block_group); | |
817d52f8 | 10482 | if (block_group->cached == BTRFS_CACHE_STARTED) |
11833d66 | 10483 | wait_block_group_cache_done(block_group); |
4f69cb98 | 10484 | if (block_group->has_caching_ctl) { |
0b246afa | 10485 | down_write(&fs_info->commit_root_sem); |
4f69cb98 FM |
10486 | if (!caching_ctl) { |
10487 | struct btrfs_caching_control *ctl; | |
10488 | ||
10489 | list_for_each_entry(ctl, | |
0b246afa | 10490 | &fs_info->caching_block_groups, list) |
4f69cb98 FM |
10491 | if (ctl->block_group == block_group) { |
10492 | caching_ctl = ctl; | |
1e4f4714 | 10493 | refcount_inc(&caching_ctl->count); |
4f69cb98 FM |
10494 | break; |
10495 | } | |
10496 | } | |
10497 | if (caching_ctl) | |
10498 | list_del_init(&caching_ctl->list); | |
0b246afa | 10499 | up_write(&fs_info->commit_root_sem); |
4f69cb98 FM |
10500 | if (caching_ctl) { |
10501 | /* Once for the caching bgs list and once for us. */ | |
10502 | put_caching_control(caching_ctl); | |
10503 | put_caching_control(caching_ctl); | |
10504 | } | |
10505 | } | |
817d52f8 | 10506 | |
ce93ec54 JB |
10507 | spin_lock(&trans->transaction->dirty_bgs_lock); |
10508 | if (!list_empty(&block_group->dirty_list)) { | |
1bbc621e CM |
10509 | WARN_ON(1); |
10510 | } | |
10511 | if (!list_empty(&block_group->io_list)) { | |
10512 | WARN_ON(1); | |
ce93ec54 JB |
10513 | } |
10514 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
817d52f8 JB |
10515 | btrfs_remove_free_space_cache(block_group); |
10516 | ||
c146afad | 10517 | spin_lock(&block_group->space_info->lock); |
75c68e9f | 10518 | list_del_init(&block_group->ro_list); |
18d018ad | 10519 | |
0b246afa | 10520 | if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
18d018ad ZL |
10521 | WARN_ON(block_group->space_info->total_bytes |
10522 | < block_group->key.offset); | |
10523 | WARN_ON(block_group->space_info->bytes_readonly | |
10524 | < block_group->key.offset); | |
10525 | WARN_ON(block_group->space_info->disk_total | |
10526 | < block_group->key.offset * factor); | |
10527 | } | |
c146afad YZ |
10528 | block_group->space_info->total_bytes -= block_group->key.offset; |
10529 | block_group->space_info->bytes_readonly -= block_group->key.offset; | |
89a55897 | 10530 | block_group->space_info->disk_total -= block_group->key.offset * factor; |
18d018ad | 10531 | |
c146afad | 10532 | spin_unlock(&block_group->space_info->lock); |
283bb197 | 10533 | |
0af3d00b JB |
10534 | memcpy(&key, &block_group->key, sizeof(key)); |
10535 | ||
34441361 | 10536 | mutex_lock(&fs_info->chunk_mutex); |
495e64f4 FM |
10537 | if (!list_empty(&em->list)) { |
10538 | /* We're in the transaction->pending_chunks list. */ | |
10539 | free_extent_map(em); | |
10540 | } | |
04216820 FM |
10541 | spin_lock(&block_group->lock); |
10542 | block_group->removed = 1; | |
10543 | /* | |
10544 | * At this point trimming can't start on this block group, because we | |
10545 | * removed the block group from the tree fs_info->block_group_cache_tree | |
10546 | * so no one can't find it anymore and even if someone already got this | |
10547 | * block group before we removed it from the rbtree, they have already | |
10548 | * incremented block_group->trimming - if they didn't, they won't find | |
10549 | * any free space entries because we already removed them all when we | |
10550 | * called btrfs_remove_free_space_cache(). | |
10551 | * | |
10552 | * And we must not remove the extent map from the fs_info->mapping_tree | |
10553 | * to prevent the same logical address range and physical device space | |
10554 | * ranges from being reused for a new block group. This is because our | |
10555 | * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is | |
10556 | * completely transactionless, so while it is trimming a range the | |
10557 | * currently running transaction might finish and a new one start, | |
10558 | * allowing for new block groups to be created that can reuse the same | |
10559 | * physical device locations unless we take this special care. | |
e33e17ee JM |
10560 | * |
10561 | * There may also be an implicit trim operation if the file system | |
10562 | * is mounted with -odiscard. The same protections must remain | |
10563 | * in place until the extents have been discarded completely when | |
10564 | * the transaction commit has completed. | |
04216820 FM |
10565 | */ |
10566 | remove_em = (atomic_read(&block_group->trimming) == 0); | |
10567 | /* | |
10568 | * Make sure a trimmer task always sees the em in the pinned_chunks list | |
10569 | * if it sees block_group->removed == 1 (needs to lock block_group->lock | |
10570 | * before checking block_group->removed). | |
10571 | */ | |
10572 | if (!remove_em) { | |
10573 | /* | |
10574 | * Our em might be in trans->transaction->pending_chunks which | |
10575 | * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks), | |
10576 | * and so is the fs_info->pinned_chunks list. | |
10577 | * | |
10578 | * So at this point we must be holding the chunk_mutex to avoid | |
10579 | * any races with chunk allocation (more specifically at | |
10580 | * volumes.c:contains_pending_extent()), to ensure it always | |
10581 | * sees the em, either in the pending_chunks list or in the | |
10582 | * pinned_chunks list. | |
10583 | */ | |
0b246afa | 10584 | list_move_tail(&em->list, &fs_info->pinned_chunks); |
04216820 FM |
10585 | } |
10586 | spin_unlock(&block_group->lock); | |
04216820 FM |
10587 | |
10588 | if (remove_em) { | |
10589 | struct extent_map_tree *em_tree; | |
10590 | ||
0b246afa | 10591 | em_tree = &fs_info->mapping_tree.map_tree; |
04216820 | 10592 | write_lock(&em_tree->lock); |
8dbcd10f FM |
10593 | /* |
10594 | * The em might be in the pending_chunks list, so make sure the | |
10595 | * chunk mutex is locked, since remove_extent_mapping() will | |
10596 | * delete us from that list. | |
10597 | */ | |
04216820 FM |
10598 | remove_extent_mapping(em_tree, em); |
10599 | write_unlock(&em_tree->lock); | |
10600 | /* once for the tree */ | |
10601 | free_extent_map(em); | |
10602 | } | |
10603 | ||
34441361 | 10604 | mutex_unlock(&fs_info->chunk_mutex); |
8dbcd10f | 10605 | |
f3f72779 | 10606 | ret = remove_block_group_free_space(trans, block_group); |
1e144fb8 OS |
10607 | if (ret) |
10608 | goto out; | |
10609 | ||
fa9c0d79 CM |
10610 | btrfs_put_block_group(block_group); |
10611 | btrfs_put_block_group(block_group); | |
1a40e23b ZY |
10612 | |
10613 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
10614 | if (ret > 0) | |
10615 | ret = -EIO; | |
10616 | if (ret < 0) | |
10617 | goto out; | |
10618 | ||
10619 | ret = btrfs_del_item(trans, root, path); | |
10620 | out: | |
10621 | btrfs_free_path(path); | |
10622 | return ret; | |
10623 | } | |
acce952b | 10624 | |
8eab77ff | 10625 | struct btrfs_trans_handle * |
7fd01182 FM |
10626 | btrfs_start_trans_remove_block_group(struct btrfs_fs_info *fs_info, |
10627 | const u64 chunk_offset) | |
8eab77ff | 10628 | { |
7fd01182 FM |
10629 | struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree; |
10630 | struct extent_map *em; | |
10631 | struct map_lookup *map; | |
10632 | unsigned int num_items; | |
10633 | ||
10634 | read_lock(&em_tree->lock); | |
10635 | em = lookup_extent_mapping(em_tree, chunk_offset, 1); | |
10636 | read_unlock(&em_tree->lock); | |
10637 | ASSERT(em && em->start == chunk_offset); | |
10638 | ||
8eab77ff | 10639 | /* |
7fd01182 FM |
10640 | * We need to reserve 3 + N units from the metadata space info in order |
10641 | * to remove a block group (done at btrfs_remove_chunk() and at | |
10642 | * btrfs_remove_block_group()), which are used for: | |
10643 | * | |
8eab77ff FM |
10644 | * 1 unit for adding the free space inode's orphan (located in the tree |
10645 | * of tree roots). | |
7fd01182 FM |
10646 | * 1 unit for deleting the block group item (located in the extent |
10647 | * tree). | |
10648 | * 1 unit for deleting the free space item (located in tree of tree | |
10649 | * roots). | |
10650 | * N units for deleting N device extent items corresponding to each | |
10651 | * stripe (located in the device tree). | |
10652 | * | |
10653 | * In order to remove a block group we also need to reserve units in the | |
10654 | * system space info in order to update the chunk tree (update one or | |
10655 | * more device items and remove one chunk item), but this is done at | |
10656 | * btrfs_remove_chunk() through a call to check_system_chunk(). | |
8eab77ff | 10657 | */ |
95617d69 | 10658 | map = em->map_lookup; |
7fd01182 FM |
10659 | num_items = 3 + map->num_stripes; |
10660 | free_extent_map(em); | |
10661 | ||
8eab77ff | 10662 | return btrfs_start_transaction_fallback_global_rsv(fs_info->extent_root, |
7fd01182 | 10663 | num_items, 1); |
8eab77ff FM |
10664 | } |
10665 | ||
47ab2a6c JB |
10666 | /* |
10667 | * Process the unused_bgs list and remove any that don't have any allocated | |
10668 | * space inside of them. | |
10669 | */ | |
10670 | void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info) | |
10671 | { | |
10672 | struct btrfs_block_group_cache *block_group; | |
10673 | struct btrfs_space_info *space_info; | |
47ab2a6c JB |
10674 | struct btrfs_trans_handle *trans; |
10675 | int ret = 0; | |
10676 | ||
afcdd129 | 10677 | if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags)) |
47ab2a6c JB |
10678 | return; |
10679 | ||
10680 | spin_lock(&fs_info->unused_bgs_lock); | |
10681 | while (!list_empty(&fs_info->unused_bgs)) { | |
10682 | u64 start, end; | |
e33e17ee | 10683 | int trimming; |
47ab2a6c JB |
10684 | |
10685 | block_group = list_first_entry(&fs_info->unused_bgs, | |
10686 | struct btrfs_block_group_cache, | |
10687 | bg_list); | |
47ab2a6c | 10688 | list_del_init(&block_group->bg_list); |
aefbe9a6 ZL |
10689 | |
10690 | space_info = block_group->space_info; | |
10691 | ||
47ab2a6c JB |
10692 | if (ret || btrfs_mixed_space_info(space_info)) { |
10693 | btrfs_put_block_group(block_group); | |
10694 | continue; | |
10695 | } | |
10696 | spin_unlock(&fs_info->unused_bgs_lock); | |
10697 | ||
d5f2e33b | 10698 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
67c5e7d4 | 10699 | |
47ab2a6c JB |
10700 | /* Don't want to race with allocators so take the groups_sem */ |
10701 | down_write(&space_info->groups_sem); | |
10702 | spin_lock(&block_group->lock); | |
10703 | if (block_group->reserved || | |
10704 | btrfs_block_group_used(&block_group->item) || | |
19c4d2f9 | 10705 | block_group->ro || |
aefbe9a6 | 10706 | list_is_singular(&block_group->list)) { |
47ab2a6c JB |
10707 | /* |
10708 | * We want to bail if we made new allocations or have | |
10709 | * outstanding allocations in this block group. We do | |
10710 | * the ro check in case balance is currently acting on | |
10711 | * this block group. | |
10712 | */ | |
4ed0a7a3 | 10713 | trace_btrfs_skip_unused_block_group(block_group); |
47ab2a6c JB |
10714 | spin_unlock(&block_group->lock); |
10715 | up_write(&space_info->groups_sem); | |
10716 | goto next; | |
10717 | } | |
10718 | spin_unlock(&block_group->lock); | |
10719 | ||
10720 | /* We don't want to force the issue, only flip if it's ok. */ | |
868f401a | 10721 | ret = inc_block_group_ro(block_group, 0); |
47ab2a6c JB |
10722 | up_write(&space_info->groups_sem); |
10723 | if (ret < 0) { | |
10724 | ret = 0; | |
10725 | goto next; | |
10726 | } | |
10727 | ||
10728 | /* | |
10729 | * Want to do this before we do anything else so we can recover | |
10730 | * properly if we fail to join the transaction. | |
10731 | */ | |
7fd01182 FM |
10732 | trans = btrfs_start_trans_remove_block_group(fs_info, |
10733 | block_group->key.objectid); | |
47ab2a6c | 10734 | if (IS_ERR(trans)) { |
2ff7e61e | 10735 | btrfs_dec_block_group_ro(block_group); |
47ab2a6c JB |
10736 | ret = PTR_ERR(trans); |
10737 | goto next; | |
10738 | } | |
10739 | ||
10740 | /* | |
10741 | * We could have pending pinned extents for this block group, | |
10742 | * just delete them, we don't care about them anymore. | |
10743 | */ | |
10744 | start = block_group->key.objectid; | |
10745 | end = start + block_group->key.offset - 1; | |
d4b450cd FM |
10746 | /* |
10747 | * Hold the unused_bg_unpin_mutex lock to avoid racing with | |
10748 | * btrfs_finish_extent_commit(). If we are at transaction N, | |
10749 | * another task might be running finish_extent_commit() for the | |
10750 | * previous transaction N - 1, and have seen a range belonging | |
10751 | * to the block group in freed_extents[] before we were able to | |
10752 | * clear the whole block group range from freed_extents[]. This | |
10753 | * means that task can lookup for the block group after we | |
10754 | * unpinned it from freed_extents[] and removed it, leading to | |
10755 | * a BUG_ON() at btrfs_unpin_extent_range(). | |
10756 | */ | |
10757 | mutex_lock(&fs_info->unused_bg_unpin_mutex); | |
758eb51e | 10758 | ret = clear_extent_bits(&fs_info->freed_extents[0], start, end, |
91166212 | 10759 | EXTENT_DIRTY); |
758eb51e | 10760 | if (ret) { |
d4b450cd | 10761 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
2ff7e61e | 10762 | btrfs_dec_block_group_ro(block_group); |
758eb51e FM |
10763 | goto end_trans; |
10764 | } | |
10765 | ret = clear_extent_bits(&fs_info->freed_extents[1], start, end, | |
91166212 | 10766 | EXTENT_DIRTY); |
758eb51e | 10767 | if (ret) { |
d4b450cd | 10768 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
2ff7e61e | 10769 | btrfs_dec_block_group_ro(block_group); |
758eb51e FM |
10770 | goto end_trans; |
10771 | } | |
d4b450cd | 10772 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
47ab2a6c JB |
10773 | |
10774 | /* Reset pinned so btrfs_put_block_group doesn't complain */ | |
c30666d4 ZL |
10775 | spin_lock(&space_info->lock); |
10776 | spin_lock(&block_group->lock); | |
10777 | ||
10778 | space_info->bytes_pinned -= block_group->pinned; | |
10779 | space_info->bytes_readonly += block_group->pinned; | |
10780 | percpu_counter_add(&space_info->total_bytes_pinned, | |
10781 | -block_group->pinned); | |
47ab2a6c JB |
10782 | block_group->pinned = 0; |
10783 | ||
c30666d4 ZL |
10784 | spin_unlock(&block_group->lock); |
10785 | spin_unlock(&space_info->lock); | |
10786 | ||
e33e17ee | 10787 | /* DISCARD can flip during remount */ |
0b246afa | 10788 | trimming = btrfs_test_opt(fs_info, DISCARD); |
e33e17ee JM |
10789 | |
10790 | /* Implicit trim during transaction commit. */ | |
10791 | if (trimming) | |
10792 | btrfs_get_block_group_trimming(block_group); | |
10793 | ||
47ab2a6c JB |
10794 | /* |
10795 | * Btrfs_remove_chunk will abort the transaction if things go | |
10796 | * horribly wrong. | |
10797 | */ | |
5b4aacef | 10798 | ret = btrfs_remove_chunk(trans, fs_info, |
47ab2a6c | 10799 | block_group->key.objectid); |
e33e17ee JM |
10800 | |
10801 | if (ret) { | |
10802 | if (trimming) | |
10803 | btrfs_put_block_group_trimming(block_group); | |
10804 | goto end_trans; | |
10805 | } | |
10806 | ||
10807 | /* | |
10808 | * If we're not mounted with -odiscard, we can just forget | |
10809 | * about this block group. Otherwise we'll need to wait | |
10810 | * until transaction commit to do the actual discard. | |
10811 | */ | |
10812 | if (trimming) { | |
348a0013 FM |
10813 | spin_lock(&fs_info->unused_bgs_lock); |
10814 | /* | |
10815 | * A concurrent scrub might have added us to the list | |
10816 | * fs_info->unused_bgs, so use a list_move operation | |
10817 | * to add the block group to the deleted_bgs list. | |
10818 | */ | |
e33e17ee JM |
10819 | list_move(&block_group->bg_list, |
10820 | &trans->transaction->deleted_bgs); | |
348a0013 | 10821 | spin_unlock(&fs_info->unused_bgs_lock); |
e33e17ee JM |
10822 | btrfs_get_block_group(block_group); |
10823 | } | |
758eb51e | 10824 | end_trans: |
3a45bb20 | 10825 | btrfs_end_transaction(trans); |
47ab2a6c | 10826 | next: |
d5f2e33b | 10827 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
47ab2a6c JB |
10828 | btrfs_put_block_group(block_group); |
10829 | spin_lock(&fs_info->unused_bgs_lock); | |
10830 | } | |
10831 | spin_unlock(&fs_info->unused_bgs_lock); | |
10832 | } | |
10833 | ||
c59021f8 | 10834 | int btrfs_init_space_info(struct btrfs_fs_info *fs_info) |
10835 | { | |
1aba86d6 | 10836 | struct btrfs_super_block *disk_super; |
10837 | u64 features; | |
10838 | u64 flags; | |
10839 | int mixed = 0; | |
c59021f8 | 10840 | int ret; |
10841 | ||
6c41761f | 10842 | disk_super = fs_info->super_copy; |
1aba86d6 | 10843 | if (!btrfs_super_root(disk_super)) |
0dc924c5 | 10844 | return -EINVAL; |
c59021f8 | 10845 | |
1aba86d6 | 10846 | features = btrfs_super_incompat_flags(disk_super); |
10847 | if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
10848 | mixed = 1; | |
c59021f8 | 10849 | |
1aba86d6 | 10850 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
4ca61683 | 10851 | ret = create_space_info(fs_info, flags); |
c59021f8 | 10852 | if (ret) |
1aba86d6 | 10853 | goto out; |
c59021f8 | 10854 | |
1aba86d6 | 10855 | if (mixed) { |
10856 | flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA; | |
4ca61683 | 10857 | ret = create_space_info(fs_info, flags); |
1aba86d6 | 10858 | } else { |
10859 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
4ca61683 | 10860 | ret = create_space_info(fs_info, flags); |
1aba86d6 | 10861 | if (ret) |
10862 | goto out; | |
10863 | ||
10864 | flags = BTRFS_BLOCK_GROUP_DATA; | |
4ca61683 | 10865 | ret = create_space_info(fs_info, flags); |
1aba86d6 | 10866 | } |
10867 | out: | |
c59021f8 | 10868 | return ret; |
10869 | } | |
10870 | ||
2ff7e61e JM |
10871 | int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info, |
10872 | u64 start, u64 end) | |
acce952b | 10873 | { |
2ff7e61e | 10874 | return unpin_extent_range(fs_info, start, end, false); |
acce952b | 10875 | } |
10876 | ||
499f377f JM |
10877 | /* |
10878 | * It used to be that old block groups would be left around forever. | |
10879 | * Iterating over them would be enough to trim unused space. Since we | |
10880 | * now automatically remove them, we also need to iterate over unallocated | |
10881 | * space. | |
10882 | * | |
10883 | * We don't want a transaction for this since the discard may take a | |
10884 | * substantial amount of time. We don't require that a transaction be | |
10885 | * running, but we do need to take a running transaction into account | |
10886 | * to ensure that we're not discarding chunks that were released in | |
10887 | * the current transaction. | |
10888 | * | |
10889 | * Holding the chunks lock will prevent other threads from allocating | |
10890 | * or releasing chunks, but it won't prevent a running transaction | |
10891 | * from committing and releasing the memory that the pending chunks | |
10892 | * list head uses. For that, we need to take a reference to the | |
10893 | * transaction. | |
10894 | */ | |
10895 | static int btrfs_trim_free_extents(struct btrfs_device *device, | |
10896 | u64 minlen, u64 *trimmed) | |
10897 | { | |
10898 | u64 start = 0, len = 0; | |
10899 | int ret; | |
10900 | ||
10901 | *trimmed = 0; | |
10902 | ||
10903 | /* Not writeable = nothing to do. */ | |
ebbede42 | 10904 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
499f377f JM |
10905 | return 0; |
10906 | ||
10907 | /* No free space = nothing to do. */ | |
10908 | if (device->total_bytes <= device->bytes_used) | |
10909 | return 0; | |
10910 | ||
10911 | ret = 0; | |
10912 | ||
10913 | while (1) { | |
fb456252 | 10914 | struct btrfs_fs_info *fs_info = device->fs_info; |
499f377f JM |
10915 | struct btrfs_transaction *trans; |
10916 | u64 bytes; | |
10917 | ||
10918 | ret = mutex_lock_interruptible(&fs_info->chunk_mutex); | |
10919 | if (ret) | |
10920 | return ret; | |
10921 | ||
10922 | down_read(&fs_info->commit_root_sem); | |
10923 | ||
10924 | spin_lock(&fs_info->trans_lock); | |
10925 | trans = fs_info->running_transaction; | |
10926 | if (trans) | |
9b64f57d | 10927 | refcount_inc(&trans->use_count); |
499f377f JM |
10928 | spin_unlock(&fs_info->trans_lock); |
10929 | ||
10930 | ret = find_free_dev_extent_start(trans, device, minlen, start, | |
10931 | &start, &len); | |
10932 | if (trans) | |
10933 | btrfs_put_transaction(trans); | |
10934 | ||
10935 | if (ret) { | |
10936 | up_read(&fs_info->commit_root_sem); | |
10937 | mutex_unlock(&fs_info->chunk_mutex); | |
10938 | if (ret == -ENOSPC) | |
10939 | ret = 0; | |
10940 | break; | |
10941 | } | |
10942 | ||
10943 | ret = btrfs_issue_discard(device->bdev, start, len, &bytes); | |
10944 | up_read(&fs_info->commit_root_sem); | |
10945 | mutex_unlock(&fs_info->chunk_mutex); | |
10946 | ||
10947 | if (ret) | |
10948 | break; | |
10949 | ||
10950 | start += len; | |
10951 | *trimmed += bytes; | |
10952 | ||
10953 | if (fatal_signal_pending(current)) { | |
10954 | ret = -ERESTARTSYS; | |
10955 | break; | |
10956 | } | |
10957 | ||
10958 | cond_resched(); | |
10959 | } | |
10960 | ||
10961 | return ret; | |
10962 | } | |
10963 | ||
2ff7e61e | 10964 | int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range) |
f7039b1d | 10965 | { |
f7039b1d | 10966 | struct btrfs_block_group_cache *cache = NULL; |
499f377f JM |
10967 | struct btrfs_device *device; |
10968 | struct list_head *devices; | |
f7039b1d LD |
10969 | u64 group_trimmed; |
10970 | u64 start; | |
10971 | u64 end; | |
10972 | u64 trimmed = 0; | |
2cac13e4 | 10973 | u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy); |
f7039b1d LD |
10974 | int ret = 0; |
10975 | ||
2cac13e4 LB |
10976 | /* |
10977 | * try to trim all FS space, our block group may start from non-zero. | |
10978 | */ | |
10979 | if (range->len == total_bytes) | |
10980 | cache = btrfs_lookup_first_block_group(fs_info, range->start); | |
10981 | else | |
10982 | cache = btrfs_lookup_block_group(fs_info, range->start); | |
f7039b1d LD |
10983 | |
10984 | while (cache) { | |
10985 | if (cache->key.objectid >= (range->start + range->len)) { | |
10986 | btrfs_put_block_group(cache); | |
10987 | break; | |
10988 | } | |
10989 | ||
10990 | start = max(range->start, cache->key.objectid); | |
10991 | end = min(range->start + range->len, | |
10992 | cache->key.objectid + cache->key.offset); | |
10993 | ||
10994 | if (end - start >= range->minlen) { | |
10995 | if (!block_group_cache_done(cache)) { | |
f6373bf3 | 10996 | ret = cache_block_group(cache, 0); |
1be41b78 JB |
10997 | if (ret) { |
10998 | btrfs_put_block_group(cache); | |
10999 | break; | |
11000 | } | |
11001 | ret = wait_block_group_cache_done(cache); | |
11002 | if (ret) { | |
11003 | btrfs_put_block_group(cache); | |
11004 | break; | |
11005 | } | |
f7039b1d LD |
11006 | } |
11007 | ret = btrfs_trim_block_group(cache, | |
11008 | &group_trimmed, | |
11009 | start, | |
11010 | end, | |
11011 | range->minlen); | |
11012 | ||
11013 | trimmed += group_trimmed; | |
11014 | if (ret) { | |
11015 | btrfs_put_block_group(cache); | |
11016 | break; | |
11017 | } | |
11018 | } | |
11019 | ||
2ff7e61e | 11020 | cache = next_block_group(fs_info, cache); |
f7039b1d LD |
11021 | } |
11022 | ||
0b246afa JM |
11023 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
11024 | devices = &fs_info->fs_devices->alloc_list; | |
499f377f JM |
11025 | list_for_each_entry(device, devices, dev_alloc_list) { |
11026 | ret = btrfs_trim_free_extents(device, range->minlen, | |
11027 | &group_trimmed); | |
11028 | if (ret) | |
11029 | break; | |
11030 | ||
11031 | trimmed += group_trimmed; | |
11032 | } | |
0b246afa | 11033 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
499f377f | 11034 | |
f7039b1d LD |
11035 | range->len = trimmed; |
11036 | return ret; | |
11037 | } | |
8257b2dc MX |
11038 | |
11039 | /* | |
ea14b57f | 11040 | * btrfs_{start,end}_write_no_snapshotting() are similar to |
9ea24bbe FM |
11041 | * mnt_{want,drop}_write(), they are used to prevent some tasks from writing |
11042 | * data into the page cache through nocow before the subvolume is snapshoted, | |
11043 | * but flush the data into disk after the snapshot creation, or to prevent | |
ea14b57f | 11044 | * operations while snapshotting is ongoing and that cause the snapshot to be |
9ea24bbe | 11045 | * inconsistent (writes followed by expanding truncates for example). |
8257b2dc | 11046 | */ |
ea14b57f | 11047 | void btrfs_end_write_no_snapshotting(struct btrfs_root *root) |
8257b2dc MX |
11048 | { |
11049 | percpu_counter_dec(&root->subv_writers->counter); | |
093258e6 | 11050 | cond_wake_up(&root->subv_writers->wait); |
8257b2dc MX |
11051 | } |
11052 | ||
ea14b57f | 11053 | int btrfs_start_write_no_snapshotting(struct btrfs_root *root) |
8257b2dc | 11054 | { |
ea14b57f | 11055 | if (atomic_read(&root->will_be_snapshotted)) |
8257b2dc MX |
11056 | return 0; |
11057 | ||
11058 | percpu_counter_inc(&root->subv_writers->counter); | |
11059 | /* | |
11060 | * Make sure counter is updated before we check for snapshot creation. | |
11061 | */ | |
11062 | smp_mb(); | |
ea14b57f DS |
11063 | if (atomic_read(&root->will_be_snapshotted)) { |
11064 | btrfs_end_write_no_snapshotting(root); | |
8257b2dc MX |
11065 | return 0; |
11066 | } | |
11067 | return 1; | |
11068 | } | |
0bc19f90 | 11069 | |
0bc19f90 ZL |
11070 | void btrfs_wait_for_snapshot_creation(struct btrfs_root *root) |
11071 | { | |
11072 | while (true) { | |
11073 | int ret; | |
11074 | ||
ea14b57f | 11075 | ret = btrfs_start_write_no_snapshotting(root); |
0bc19f90 ZL |
11076 | if (ret) |
11077 | break; | |
4625956a PZ |
11078 | wait_var_event(&root->will_be_snapshotted, |
11079 | !atomic_read(&root->will_be_snapshotted)); | |
0bc19f90 ZL |
11080 | } |
11081 | } |