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 | ||
9f9b8e8d QW |
54 | /* |
55 | * Declare a helper function to detect underflow of various space info members | |
56 | */ | |
57 | #define DECLARE_SPACE_INFO_UPDATE(name) \ | |
480b9b4d QW |
58 | static inline void update_##name(struct btrfs_fs_info *fs_info, \ |
59 | struct btrfs_space_info *sinfo, \ | |
9f9b8e8d QW |
60 | s64 bytes) \ |
61 | { \ | |
0185f364 | 62 | lockdep_assert_held(&sinfo->lock); \ |
480b9b4d | 63 | trace_update_##name(fs_info, sinfo, sinfo->name, bytes); \ |
9f9b8e8d QW |
64 | if (bytes < 0 && sinfo->name < -bytes) { \ |
65 | WARN_ON(1); \ | |
66 | sinfo->name = 0; \ | |
67 | return; \ | |
68 | } \ | |
69 | sinfo->name += bytes; \ | |
70 | } | |
71 | ||
72 | DECLARE_SPACE_INFO_UPDATE(bytes_may_use); | |
e2907c1a | 73 | DECLARE_SPACE_INFO_UPDATE(bytes_pinned); |
9f9b8e8d | 74 | |
5d4f98a2 | 75 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
e72cb923 NB |
76 | struct btrfs_delayed_ref_node *node, u64 parent, |
77 | u64 root_objectid, u64 owner_objectid, | |
78 | u64 owner_offset, int refs_to_drop, | |
79 | struct btrfs_delayed_extent_op *extra_op); | |
5d4f98a2 YZ |
80 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, |
81 | struct extent_buffer *leaf, | |
82 | struct btrfs_extent_item *ei); | |
83 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
84 | u64 parent, u64 root_objectid, |
85 | u64 flags, u64 owner, u64 offset, | |
86 | struct btrfs_key *ins, int ref_mod); | |
87 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, | |
4e6bd4e0 | 88 | struct btrfs_delayed_ref_node *node, |
21ebfbe7 | 89 | struct btrfs_delayed_extent_op *extent_op); |
01458828 | 90 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, u64 flags, |
698d0082 | 91 | int force); |
11833d66 YZ |
92 | static int find_next_key(struct btrfs_path *path, int level, |
93 | struct btrfs_key *key); | |
ab8d0fc4 JM |
94 | static void dump_space_info(struct btrfs_fs_info *fs_info, |
95 | struct btrfs_space_info *info, u64 bytes, | |
9ed74f2d | 96 | int dump_block_groups); |
5d80366e JB |
97 | static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, |
98 | u64 num_bytes); | |
957780eb JB |
99 | static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info, |
100 | struct btrfs_space_info *space_info, | |
101 | u64 num_bytes); | |
102 | static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info, | |
103 | struct btrfs_space_info *space_info, | |
104 | u64 num_bytes); | |
6a63209f | 105 | |
817d52f8 JB |
106 | static noinline int |
107 | block_group_cache_done(struct btrfs_block_group_cache *cache) | |
108 | { | |
109 | smp_mb(); | |
36cce922 JB |
110 | return cache->cached == BTRFS_CACHE_FINISHED || |
111 | cache->cached == BTRFS_CACHE_ERROR; | |
817d52f8 JB |
112 | } |
113 | ||
0f9dd46c JB |
114 | static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits) |
115 | { | |
116 | return (cache->flags & bits) == bits; | |
117 | } | |
118 | ||
758f2dfc | 119 | void btrfs_get_block_group(struct btrfs_block_group_cache *cache) |
11dfe35a JB |
120 | { |
121 | atomic_inc(&cache->count); | |
122 | } | |
123 | ||
124 | void btrfs_put_block_group(struct btrfs_block_group_cache *cache) | |
125 | { | |
f0486c68 YZ |
126 | if (atomic_dec_and_test(&cache->count)) { |
127 | WARN_ON(cache->pinned > 0); | |
128 | WARN_ON(cache->reserved > 0); | |
0966a7b1 QW |
129 | |
130 | /* | |
131 | * If not empty, someone is still holding mutex of | |
132 | * full_stripe_lock, which can only be released by caller. | |
133 | * And it will definitely cause use-after-free when caller | |
134 | * tries to release full stripe lock. | |
135 | * | |
136 | * No better way to resolve, but only to warn. | |
137 | */ | |
138 | WARN_ON(!RB_EMPTY_ROOT(&cache->full_stripe_locks_root.root)); | |
34d52cb6 | 139 | kfree(cache->free_space_ctl); |
11dfe35a | 140 | kfree(cache); |
f0486c68 | 141 | } |
11dfe35a JB |
142 | } |
143 | ||
0f9dd46c JB |
144 | /* |
145 | * this adds the block group to the fs_info rb tree for the block group | |
146 | * cache | |
147 | */ | |
b2950863 | 148 | static int btrfs_add_block_group_cache(struct btrfs_fs_info *info, |
0f9dd46c JB |
149 | struct btrfs_block_group_cache *block_group) |
150 | { | |
151 | struct rb_node **p; | |
152 | struct rb_node *parent = NULL; | |
153 | struct btrfs_block_group_cache *cache; | |
154 | ||
155 | spin_lock(&info->block_group_cache_lock); | |
156 | p = &info->block_group_cache_tree.rb_node; | |
157 | ||
158 | while (*p) { | |
159 | parent = *p; | |
160 | cache = rb_entry(parent, struct btrfs_block_group_cache, | |
161 | cache_node); | |
162 | if (block_group->key.objectid < cache->key.objectid) { | |
163 | p = &(*p)->rb_left; | |
164 | } else if (block_group->key.objectid > cache->key.objectid) { | |
165 | p = &(*p)->rb_right; | |
166 | } else { | |
167 | spin_unlock(&info->block_group_cache_lock); | |
168 | return -EEXIST; | |
169 | } | |
170 | } | |
171 | ||
172 | rb_link_node(&block_group->cache_node, parent, p); | |
173 | rb_insert_color(&block_group->cache_node, | |
174 | &info->block_group_cache_tree); | |
a1897fdd LB |
175 | |
176 | if (info->first_logical_byte > block_group->key.objectid) | |
177 | info->first_logical_byte = block_group->key.objectid; | |
178 | ||
0f9dd46c JB |
179 | spin_unlock(&info->block_group_cache_lock); |
180 | ||
181 | return 0; | |
182 | } | |
183 | ||
184 | /* | |
185 | * This will return the block group at or after bytenr if contains is 0, else | |
186 | * it will return the block group that contains the bytenr | |
187 | */ | |
188 | static struct btrfs_block_group_cache * | |
189 | block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr, | |
190 | int contains) | |
191 | { | |
192 | struct btrfs_block_group_cache *cache, *ret = NULL; | |
193 | struct rb_node *n; | |
194 | u64 end, start; | |
195 | ||
196 | spin_lock(&info->block_group_cache_lock); | |
197 | n = info->block_group_cache_tree.rb_node; | |
198 | ||
199 | while (n) { | |
200 | cache = rb_entry(n, struct btrfs_block_group_cache, | |
201 | cache_node); | |
202 | end = cache->key.objectid + cache->key.offset - 1; | |
203 | start = cache->key.objectid; | |
204 | ||
205 | if (bytenr < start) { | |
206 | if (!contains && (!ret || start < ret->key.objectid)) | |
207 | ret = cache; | |
208 | n = n->rb_left; | |
209 | } else if (bytenr > start) { | |
210 | if (contains && bytenr <= end) { | |
211 | ret = cache; | |
212 | break; | |
213 | } | |
214 | n = n->rb_right; | |
215 | } else { | |
216 | ret = cache; | |
217 | break; | |
218 | } | |
219 | } | |
a1897fdd | 220 | if (ret) { |
11dfe35a | 221 | btrfs_get_block_group(ret); |
a1897fdd LB |
222 | if (bytenr == 0 && info->first_logical_byte > ret->key.objectid) |
223 | info->first_logical_byte = ret->key.objectid; | |
224 | } | |
0f9dd46c JB |
225 | spin_unlock(&info->block_group_cache_lock); |
226 | ||
227 | return ret; | |
228 | } | |
229 | ||
2ff7e61e | 230 | static int add_excluded_extent(struct btrfs_fs_info *fs_info, |
11833d66 | 231 | u64 start, u64 num_bytes) |
817d52f8 | 232 | { |
11833d66 | 233 | u64 end = start + num_bytes - 1; |
0b246afa | 234 | set_extent_bits(&fs_info->freed_extents[0], |
ceeb0ae7 | 235 | start, end, EXTENT_UPTODATE); |
0b246afa | 236 | set_extent_bits(&fs_info->freed_extents[1], |
ceeb0ae7 | 237 | start, end, EXTENT_UPTODATE); |
11833d66 YZ |
238 | return 0; |
239 | } | |
817d52f8 | 240 | |
9e715da8 | 241 | static void free_excluded_extents(struct btrfs_block_group_cache *cache) |
11833d66 | 242 | { |
9e715da8 | 243 | struct btrfs_fs_info *fs_info = cache->fs_info; |
11833d66 | 244 | u64 start, end; |
817d52f8 | 245 | |
11833d66 YZ |
246 | start = cache->key.objectid; |
247 | end = start + cache->key.offset - 1; | |
248 | ||
0b246afa | 249 | clear_extent_bits(&fs_info->freed_extents[0], |
91166212 | 250 | start, end, EXTENT_UPTODATE); |
0b246afa | 251 | clear_extent_bits(&fs_info->freed_extents[1], |
91166212 | 252 | start, end, EXTENT_UPTODATE); |
817d52f8 JB |
253 | } |
254 | ||
3c4da657 | 255 | static int exclude_super_stripes(struct btrfs_block_group_cache *cache) |
817d52f8 | 256 | { |
3c4da657 | 257 | struct btrfs_fs_info *fs_info = cache->fs_info; |
817d52f8 JB |
258 | u64 bytenr; |
259 | u64 *logical; | |
260 | int stripe_len; | |
261 | int i, nr, ret; | |
262 | ||
06b2331f YZ |
263 | if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) { |
264 | stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid; | |
265 | cache->bytes_super += stripe_len; | |
2ff7e61e | 266 | ret = add_excluded_extent(fs_info, cache->key.objectid, |
06b2331f | 267 | stripe_len); |
835d974f JB |
268 | if (ret) |
269 | return ret; | |
06b2331f YZ |
270 | } |
271 | ||
817d52f8 JB |
272 | for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { |
273 | bytenr = btrfs_sb_offset(i); | |
0b246afa | 274 | ret = btrfs_rmap_block(fs_info, cache->key.objectid, |
63a9c7b9 | 275 | bytenr, &logical, &nr, &stripe_len); |
835d974f JB |
276 | if (ret) |
277 | return ret; | |
11833d66 | 278 | |
817d52f8 | 279 | while (nr--) { |
51bf5f0b JB |
280 | u64 start, len; |
281 | ||
282 | if (logical[nr] > cache->key.objectid + | |
283 | cache->key.offset) | |
284 | continue; | |
285 | ||
286 | if (logical[nr] + stripe_len <= cache->key.objectid) | |
287 | continue; | |
288 | ||
289 | start = logical[nr]; | |
290 | if (start < cache->key.objectid) { | |
291 | start = cache->key.objectid; | |
292 | len = (logical[nr] + stripe_len) - start; | |
293 | } else { | |
294 | len = min_t(u64, stripe_len, | |
295 | cache->key.objectid + | |
296 | cache->key.offset - start); | |
297 | } | |
298 | ||
299 | cache->bytes_super += len; | |
2ff7e61e | 300 | ret = add_excluded_extent(fs_info, start, len); |
835d974f JB |
301 | if (ret) { |
302 | kfree(logical); | |
303 | return ret; | |
304 | } | |
817d52f8 | 305 | } |
11833d66 | 306 | |
817d52f8 JB |
307 | kfree(logical); |
308 | } | |
817d52f8 JB |
309 | return 0; |
310 | } | |
311 | ||
11833d66 YZ |
312 | static struct btrfs_caching_control * |
313 | get_caching_control(struct btrfs_block_group_cache *cache) | |
314 | { | |
315 | struct btrfs_caching_control *ctl; | |
316 | ||
317 | spin_lock(&cache->lock); | |
dde5abee JB |
318 | if (!cache->caching_ctl) { |
319 | spin_unlock(&cache->lock); | |
11833d66 YZ |
320 | return NULL; |
321 | } | |
322 | ||
323 | ctl = cache->caching_ctl; | |
1e4f4714 | 324 | refcount_inc(&ctl->count); |
11833d66 YZ |
325 | spin_unlock(&cache->lock); |
326 | return ctl; | |
327 | } | |
328 | ||
329 | static void put_caching_control(struct btrfs_caching_control *ctl) | |
330 | { | |
1e4f4714 | 331 | if (refcount_dec_and_test(&ctl->count)) |
11833d66 YZ |
332 | kfree(ctl); |
333 | } | |
334 | ||
d0bd4560 | 335 | #ifdef CONFIG_BTRFS_DEBUG |
2ff7e61e | 336 | static void fragment_free_space(struct btrfs_block_group_cache *block_group) |
d0bd4560 | 337 | { |
2ff7e61e | 338 | struct btrfs_fs_info *fs_info = block_group->fs_info; |
d0bd4560 JB |
339 | u64 start = block_group->key.objectid; |
340 | u64 len = block_group->key.offset; | |
341 | u64 chunk = block_group->flags & BTRFS_BLOCK_GROUP_METADATA ? | |
0b246afa | 342 | fs_info->nodesize : fs_info->sectorsize; |
d0bd4560 JB |
343 | u64 step = chunk << 1; |
344 | ||
345 | while (len > chunk) { | |
346 | btrfs_remove_free_space(block_group, start, chunk); | |
347 | start += step; | |
348 | if (len < step) | |
349 | len = 0; | |
350 | else | |
351 | len -= step; | |
352 | } | |
353 | } | |
354 | #endif | |
355 | ||
0f9dd46c JB |
356 | /* |
357 | * this is only called by cache_block_group, since we could have freed extents | |
358 | * we need to check the pinned_extents for any extents that can't be used yet | |
359 | * since their free space will be released as soon as the transaction commits. | |
360 | */ | |
a5ed9182 | 361 | u64 add_new_free_space(struct btrfs_block_group_cache *block_group, |
4457c1c7 | 362 | u64 start, u64 end) |
0f9dd46c | 363 | { |
4457c1c7 | 364 | struct btrfs_fs_info *info = block_group->fs_info; |
817d52f8 | 365 | u64 extent_start, extent_end, size, total_added = 0; |
0f9dd46c JB |
366 | int ret; |
367 | ||
368 | while (start < end) { | |
11833d66 | 369 | ret = find_first_extent_bit(info->pinned_extents, start, |
0f9dd46c | 370 | &extent_start, &extent_end, |
e6138876 JB |
371 | EXTENT_DIRTY | EXTENT_UPTODATE, |
372 | NULL); | |
0f9dd46c JB |
373 | if (ret) |
374 | break; | |
375 | ||
06b2331f | 376 | if (extent_start <= start) { |
0f9dd46c JB |
377 | start = extent_end + 1; |
378 | } else if (extent_start > start && extent_start < end) { | |
379 | size = extent_start - start; | |
817d52f8 | 380 | total_added += size; |
ea6a478e JB |
381 | ret = btrfs_add_free_space(block_group, start, |
382 | size); | |
79787eaa | 383 | BUG_ON(ret); /* -ENOMEM or logic error */ |
0f9dd46c JB |
384 | start = extent_end + 1; |
385 | } else { | |
386 | break; | |
387 | } | |
388 | } | |
389 | ||
390 | if (start < end) { | |
391 | size = end - start; | |
817d52f8 | 392 | total_added += size; |
ea6a478e | 393 | ret = btrfs_add_free_space(block_group, start, size); |
79787eaa | 394 | BUG_ON(ret); /* -ENOMEM or logic error */ |
0f9dd46c JB |
395 | } |
396 | ||
817d52f8 | 397 | return total_added; |
0f9dd46c JB |
398 | } |
399 | ||
73fa48b6 | 400 | static int load_extent_tree_free(struct btrfs_caching_control *caching_ctl) |
e37c9e69 | 401 | { |
0b246afa JM |
402 | struct btrfs_block_group_cache *block_group = caching_ctl->block_group; |
403 | struct btrfs_fs_info *fs_info = block_group->fs_info; | |
404 | struct btrfs_root *extent_root = fs_info->extent_root; | |
e37c9e69 | 405 | struct btrfs_path *path; |
5f39d397 | 406 | struct extent_buffer *leaf; |
11833d66 | 407 | struct btrfs_key key; |
817d52f8 | 408 | u64 total_found = 0; |
11833d66 YZ |
409 | u64 last = 0; |
410 | u32 nritems; | |
73fa48b6 | 411 | int ret; |
d0bd4560 | 412 | bool wakeup = true; |
f510cfec | 413 | |
e37c9e69 CM |
414 | path = btrfs_alloc_path(); |
415 | if (!path) | |
73fa48b6 | 416 | return -ENOMEM; |
7d7d6068 | 417 | |
817d52f8 | 418 | last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET); |
11833d66 | 419 | |
d0bd4560 JB |
420 | #ifdef CONFIG_BTRFS_DEBUG |
421 | /* | |
422 | * If we're fragmenting we don't want to make anybody think we can | |
423 | * allocate from this block group until we've had a chance to fragment | |
424 | * the free space. | |
425 | */ | |
2ff7e61e | 426 | if (btrfs_should_fragment_free_space(block_group)) |
d0bd4560 JB |
427 | wakeup = false; |
428 | #endif | |
5cd57b2c | 429 | /* |
817d52f8 JB |
430 | * We don't want to deadlock with somebody trying to allocate a new |
431 | * extent for the extent root while also trying to search the extent | |
432 | * root to add free space. So we skip locking and search the commit | |
433 | * root, since its read-only | |
5cd57b2c CM |
434 | */ |
435 | path->skip_locking = 1; | |
817d52f8 | 436 | path->search_commit_root = 1; |
e4058b54 | 437 | path->reada = READA_FORWARD; |
817d52f8 | 438 | |
e4404d6e | 439 | key.objectid = last; |
e37c9e69 | 440 | key.offset = 0; |
11833d66 | 441 | key.type = BTRFS_EXTENT_ITEM_KEY; |
013f1b12 | 442 | |
52ee28d2 | 443 | next: |
11833d66 | 444 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
e37c9e69 | 445 | if (ret < 0) |
73fa48b6 | 446 | goto out; |
a512bbf8 | 447 | |
11833d66 YZ |
448 | leaf = path->nodes[0]; |
449 | nritems = btrfs_header_nritems(leaf); | |
450 | ||
d397712b | 451 | while (1) { |
7841cb28 | 452 | if (btrfs_fs_closing(fs_info) > 1) { |
f25784b3 | 453 | last = (u64)-1; |
817d52f8 | 454 | break; |
f25784b3 | 455 | } |
817d52f8 | 456 | |
11833d66 YZ |
457 | if (path->slots[0] < nritems) { |
458 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
459 | } else { | |
460 | ret = find_next_key(path, 0, &key); | |
461 | if (ret) | |
e37c9e69 | 462 | break; |
817d52f8 | 463 | |
c9ea7b24 | 464 | if (need_resched() || |
9e351cc8 | 465 | rwsem_is_contended(&fs_info->commit_root_sem)) { |
d0bd4560 JB |
466 | if (wakeup) |
467 | caching_ctl->progress = last; | |
ff5714cc | 468 | btrfs_release_path(path); |
9e351cc8 | 469 | up_read(&fs_info->commit_root_sem); |
589d8ade | 470 | mutex_unlock(&caching_ctl->mutex); |
11833d66 | 471 | cond_resched(); |
73fa48b6 OS |
472 | mutex_lock(&caching_ctl->mutex); |
473 | down_read(&fs_info->commit_root_sem); | |
474 | goto next; | |
589d8ade | 475 | } |
0a3896d0 JB |
476 | |
477 | ret = btrfs_next_leaf(extent_root, path); | |
478 | if (ret < 0) | |
73fa48b6 | 479 | goto out; |
0a3896d0 JB |
480 | if (ret) |
481 | break; | |
589d8ade JB |
482 | leaf = path->nodes[0]; |
483 | nritems = btrfs_header_nritems(leaf); | |
484 | continue; | |
11833d66 | 485 | } |
817d52f8 | 486 | |
52ee28d2 LB |
487 | if (key.objectid < last) { |
488 | key.objectid = last; | |
489 | key.offset = 0; | |
490 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
491 | ||
d0bd4560 JB |
492 | if (wakeup) |
493 | caching_ctl->progress = last; | |
52ee28d2 LB |
494 | btrfs_release_path(path); |
495 | goto next; | |
496 | } | |
497 | ||
11833d66 YZ |
498 | if (key.objectid < block_group->key.objectid) { |
499 | path->slots[0]++; | |
817d52f8 | 500 | continue; |
e37c9e69 | 501 | } |
0f9dd46c | 502 | |
e37c9e69 | 503 | if (key.objectid >= block_group->key.objectid + |
0f9dd46c | 504 | block_group->key.offset) |
e37c9e69 | 505 | break; |
7d7d6068 | 506 | |
3173a18f JB |
507 | if (key.type == BTRFS_EXTENT_ITEM_KEY || |
508 | key.type == BTRFS_METADATA_ITEM_KEY) { | |
4457c1c7 | 509 | total_found += add_new_free_space(block_group, last, |
817d52f8 | 510 | key.objectid); |
3173a18f JB |
511 | if (key.type == BTRFS_METADATA_ITEM_KEY) |
512 | last = key.objectid + | |
da17066c | 513 | fs_info->nodesize; |
3173a18f JB |
514 | else |
515 | last = key.objectid + key.offset; | |
817d52f8 | 516 | |
73fa48b6 | 517 | if (total_found > CACHING_CTL_WAKE_UP) { |
11833d66 | 518 | total_found = 0; |
d0bd4560 JB |
519 | if (wakeup) |
520 | wake_up(&caching_ctl->wait); | |
11833d66 | 521 | } |
817d52f8 | 522 | } |
e37c9e69 CM |
523 | path->slots[0]++; |
524 | } | |
817d52f8 | 525 | ret = 0; |
e37c9e69 | 526 | |
4457c1c7 | 527 | total_found += add_new_free_space(block_group, last, |
817d52f8 JB |
528 | block_group->key.objectid + |
529 | block_group->key.offset); | |
11833d66 | 530 | caching_ctl->progress = (u64)-1; |
817d52f8 | 531 | |
73fa48b6 OS |
532 | out: |
533 | btrfs_free_path(path); | |
534 | return ret; | |
535 | } | |
536 | ||
537 | static noinline void caching_thread(struct btrfs_work *work) | |
538 | { | |
539 | struct btrfs_block_group_cache *block_group; | |
540 | struct btrfs_fs_info *fs_info; | |
541 | struct btrfs_caching_control *caching_ctl; | |
542 | int ret; | |
543 | ||
544 | caching_ctl = container_of(work, struct btrfs_caching_control, work); | |
545 | block_group = caching_ctl->block_group; | |
546 | fs_info = block_group->fs_info; | |
547 | ||
548 | mutex_lock(&caching_ctl->mutex); | |
549 | down_read(&fs_info->commit_root_sem); | |
550 | ||
1e144fb8 OS |
551 | if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) |
552 | ret = load_free_space_tree(caching_ctl); | |
553 | else | |
554 | ret = load_extent_tree_free(caching_ctl); | |
73fa48b6 | 555 | |
817d52f8 | 556 | spin_lock(&block_group->lock); |
11833d66 | 557 | block_group->caching_ctl = NULL; |
73fa48b6 | 558 | block_group->cached = ret ? BTRFS_CACHE_ERROR : BTRFS_CACHE_FINISHED; |
817d52f8 | 559 | spin_unlock(&block_group->lock); |
0f9dd46c | 560 | |
d0bd4560 | 561 | #ifdef CONFIG_BTRFS_DEBUG |
2ff7e61e | 562 | if (btrfs_should_fragment_free_space(block_group)) { |
d0bd4560 JB |
563 | u64 bytes_used; |
564 | ||
565 | spin_lock(&block_group->space_info->lock); | |
566 | spin_lock(&block_group->lock); | |
567 | bytes_used = block_group->key.offset - | |
568 | btrfs_block_group_used(&block_group->item); | |
569 | block_group->space_info->bytes_used += bytes_used >> 1; | |
570 | spin_unlock(&block_group->lock); | |
571 | spin_unlock(&block_group->space_info->lock); | |
2ff7e61e | 572 | fragment_free_space(block_group); |
d0bd4560 JB |
573 | } |
574 | #endif | |
575 | ||
576 | caching_ctl->progress = (u64)-1; | |
11833d66 | 577 | |
9e351cc8 | 578 | up_read(&fs_info->commit_root_sem); |
9e715da8 | 579 | free_excluded_extents(block_group); |
11833d66 | 580 | mutex_unlock(&caching_ctl->mutex); |
73fa48b6 | 581 | |
11833d66 YZ |
582 | wake_up(&caching_ctl->wait); |
583 | ||
584 | put_caching_control(caching_ctl); | |
11dfe35a | 585 | btrfs_put_block_group(block_group); |
817d52f8 JB |
586 | } |
587 | ||
9d66e233 | 588 | static int cache_block_group(struct btrfs_block_group_cache *cache, |
9d66e233 | 589 | int load_cache_only) |
817d52f8 | 590 | { |
291c7d2f | 591 | DEFINE_WAIT(wait); |
11833d66 YZ |
592 | struct btrfs_fs_info *fs_info = cache->fs_info; |
593 | struct btrfs_caching_control *caching_ctl; | |
817d52f8 JB |
594 | int ret = 0; |
595 | ||
291c7d2f | 596 | caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS); |
79787eaa JM |
597 | if (!caching_ctl) |
598 | return -ENOMEM; | |
291c7d2f JB |
599 | |
600 | INIT_LIST_HEAD(&caching_ctl->list); | |
601 | mutex_init(&caching_ctl->mutex); | |
602 | init_waitqueue_head(&caching_ctl->wait); | |
603 | caching_ctl->block_group = cache; | |
604 | caching_ctl->progress = cache->key.objectid; | |
1e4f4714 | 605 | refcount_set(&caching_ctl->count, 1); |
9e0af237 LB |
606 | btrfs_init_work(&caching_ctl->work, btrfs_cache_helper, |
607 | caching_thread, NULL, NULL); | |
291c7d2f JB |
608 | |
609 | spin_lock(&cache->lock); | |
610 | /* | |
611 | * This should be a rare occasion, but this could happen I think in the | |
612 | * case where one thread starts to load the space cache info, and then | |
613 | * some other thread starts a transaction commit which tries to do an | |
614 | * allocation while the other thread is still loading the space cache | |
615 | * info. The previous loop should have kept us from choosing this block | |
616 | * group, but if we've moved to the state where we will wait on caching | |
617 | * block groups we need to first check if we're doing a fast load here, | |
618 | * so we can wait for it to finish, otherwise we could end up allocating | |
619 | * from a block group who's cache gets evicted for one reason or | |
620 | * another. | |
621 | */ | |
622 | while (cache->cached == BTRFS_CACHE_FAST) { | |
623 | struct btrfs_caching_control *ctl; | |
624 | ||
625 | ctl = cache->caching_ctl; | |
1e4f4714 | 626 | refcount_inc(&ctl->count); |
291c7d2f JB |
627 | prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE); |
628 | spin_unlock(&cache->lock); | |
629 | ||
630 | schedule(); | |
631 | ||
632 | finish_wait(&ctl->wait, &wait); | |
633 | put_caching_control(ctl); | |
634 | spin_lock(&cache->lock); | |
635 | } | |
636 | ||
637 | if (cache->cached != BTRFS_CACHE_NO) { | |
638 | spin_unlock(&cache->lock); | |
639 | kfree(caching_ctl); | |
11833d66 | 640 | return 0; |
291c7d2f JB |
641 | } |
642 | WARN_ON(cache->caching_ctl); | |
643 | cache->caching_ctl = caching_ctl; | |
644 | cache->cached = BTRFS_CACHE_FAST; | |
645 | spin_unlock(&cache->lock); | |
11833d66 | 646 | |
d8953d69 | 647 | if (btrfs_test_opt(fs_info, SPACE_CACHE)) { |
cb83b7b8 | 648 | mutex_lock(&caching_ctl->mutex); |
bb6cb1c5 | 649 | ret = load_free_space_cache(cache); |
9d66e233 JB |
650 | |
651 | spin_lock(&cache->lock); | |
652 | if (ret == 1) { | |
291c7d2f | 653 | cache->caching_ctl = NULL; |
9d66e233 JB |
654 | cache->cached = BTRFS_CACHE_FINISHED; |
655 | cache->last_byte_to_unpin = (u64)-1; | |
cb83b7b8 | 656 | caching_ctl->progress = (u64)-1; |
9d66e233 | 657 | } else { |
291c7d2f JB |
658 | if (load_cache_only) { |
659 | cache->caching_ctl = NULL; | |
660 | cache->cached = BTRFS_CACHE_NO; | |
661 | } else { | |
662 | cache->cached = BTRFS_CACHE_STARTED; | |
4f69cb98 | 663 | cache->has_caching_ctl = 1; |
291c7d2f | 664 | } |
9d66e233 JB |
665 | } |
666 | spin_unlock(&cache->lock); | |
d0bd4560 JB |
667 | #ifdef CONFIG_BTRFS_DEBUG |
668 | if (ret == 1 && | |
2ff7e61e | 669 | btrfs_should_fragment_free_space(cache)) { |
d0bd4560 JB |
670 | u64 bytes_used; |
671 | ||
672 | spin_lock(&cache->space_info->lock); | |
673 | spin_lock(&cache->lock); | |
674 | bytes_used = cache->key.offset - | |
675 | btrfs_block_group_used(&cache->item); | |
676 | cache->space_info->bytes_used += bytes_used >> 1; | |
677 | spin_unlock(&cache->lock); | |
678 | spin_unlock(&cache->space_info->lock); | |
2ff7e61e | 679 | fragment_free_space(cache); |
d0bd4560 JB |
680 | } |
681 | #endif | |
cb83b7b8 JB |
682 | mutex_unlock(&caching_ctl->mutex); |
683 | ||
291c7d2f | 684 | wake_up(&caching_ctl->wait); |
3c14874a | 685 | if (ret == 1) { |
291c7d2f | 686 | put_caching_control(caching_ctl); |
9e715da8 | 687 | free_excluded_extents(cache); |
9d66e233 | 688 | return 0; |
3c14874a | 689 | } |
291c7d2f JB |
690 | } else { |
691 | /* | |
1e144fb8 OS |
692 | * We're either using the free space tree or no caching at all. |
693 | * Set cached to the appropriate value and wakeup any waiters. | |
291c7d2f JB |
694 | */ |
695 | spin_lock(&cache->lock); | |
696 | if (load_cache_only) { | |
697 | cache->caching_ctl = NULL; | |
698 | cache->cached = BTRFS_CACHE_NO; | |
699 | } else { | |
700 | cache->cached = BTRFS_CACHE_STARTED; | |
4f69cb98 | 701 | cache->has_caching_ctl = 1; |
291c7d2f JB |
702 | } |
703 | spin_unlock(&cache->lock); | |
704 | wake_up(&caching_ctl->wait); | |
9d66e233 JB |
705 | } |
706 | ||
291c7d2f JB |
707 | if (load_cache_only) { |
708 | put_caching_control(caching_ctl); | |
11833d66 | 709 | return 0; |
817d52f8 | 710 | } |
817d52f8 | 711 | |
9e351cc8 | 712 | down_write(&fs_info->commit_root_sem); |
1e4f4714 | 713 | refcount_inc(&caching_ctl->count); |
11833d66 | 714 | list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups); |
9e351cc8 | 715 | up_write(&fs_info->commit_root_sem); |
11833d66 | 716 | |
11dfe35a | 717 | btrfs_get_block_group(cache); |
11833d66 | 718 | |
e66f0bb1 | 719 | btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work); |
817d52f8 | 720 | |
ef8bbdfe | 721 | return ret; |
e37c9e69 CM |
722 | } |
723 | ||
0f9dd46c JB |
724 | /* |
725 | * return the block group that starts at or after bytenr | |
726 | */ | |
d397712b CM |
727 | static struct btrfs_block_group_cache * |
728 | btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr) | |
0ef3e66b | 729 | { |
e2c89907 | 730 | return block_group_cache_tree_search(info, bytenr, 0); |
0ef3e66b CM |
731 | } |
732 | ||
0f9dd46c | 733 | /* |
9f55684c | 734 | * return the block group that contains the given bytenr |
0f9dd46c | 735 | */ |
d397712b CM |
736 | struct btrfs_block_group_cache *btrfs_lookup_block_group( |
737 | struct btrfs_fs_info *info, | |
738 | u64 bytenr) | |
be744175 | 739 | { |
e2c89907 | 740 | return block_group_cache_tree_search(info, bytenr, 1); |
be744175 | 741 | } |
0b86a832 | 742 | |
0f9dd46c JB |
743 | static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info, |
744 | u64 flags) | |
6324fbf3 | 745 | { |
0f9dd46c | 746 | struct list_head *head = &info->space_info; |
0f9dd46c | 747 | struct btrfs_space_info *found; |
4184ea7f | 748 | |
52ba6929 | 749 | flags &= BTRFS_BLOCK_GROUP_TYPE_MASK; |
b742bb82 | 750 | |
4184ea7f CM |
751 | rcu_read_lock(); |
752 | list_for_each_entry_rcu(found, head, list) { | |
67377734 | 753 | if (found->flags & flags) { |
4184ea7f | 754 | rcu_read_unlock(); |
0f9dd46c | 755 | return found; |
4184ea7f | 756 | } |
0f9dd46c | 757 | } |
4184ea7f | 758 | rcu_read_unlock(); |
0f9dd46c | 759 | return NULL; |
6324fbf3 CM |
760 | } |
761 | ||
78192442 | 762 | static u64 generic_ref_to_space_flags(struct btrfs_ref *ref) |
0d9f824d | 763 | { |
ddf30cf0 QW |
764 | if (ref->type == BTRFS_REF_METADATA) { |
765 | if (ref->tree_ref.root == BTRFS_CHUNK_TREE_OBJECTID) | |
78192442 | 766 | return BTRFS_BLOCK_GROUP_SYSTEM; |
0d9f824d | 767 | else |
78192442 | 768 | return BTRFS_BLOCK_GROUP_METADATA; |
0d9f824d | 769 | } |
78192442 QW |
770 | return BTRFS_BLOCK_GROUP_DATA; |
771 | } | |
772 | ||
773 | static void add_pinned_bytes(struct btrfs_fs_info *fs_info, | |
774 | struct btrfs_ref *ref) | |
775 | { | |
776 | struct btrfs_space_info *space_info; | |
777 | u64 flags = generic_ref_to_space_flags(ref); | |
778 | ||
779 | space_info = __find_space_info(fs_info, flags); | |
780 | ASSERT(space_info); | |
781 | percpu_counter_add_batch(&space_info->total_bytes_pinned, ref->len, | |
782 | BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
783 | } | |
784 | ||
785 | static void sub_pinned_bytes(struct btrfs_fs_info *fs_info, | |
786 | struct btrfs_ref *ref) | |
787 | { | |
788 | struct btrfs_space_info *space_info; | |
789 | u64 flags = generic_ref_to_space_flags(ref); | |
0d9f824d OS |
790 | |
791 | space_info = __find_space_info(fs_info, flags); | |
55e8196a | 792 | ASSERT(space_info); |
78192442 | 793 | percpu_counter_add_batch(&space_info->total_bytes_pinned, -ref->len, |
dec59fa3 | 794 | BTRFS_TOTAL_BYTES_PINNED_BATCH); |
0d9f824d OS |
795 | } |
796 | ||
4184ea7f CM |
797 | /* |
798 | * after adding space to the filesystem, we need to clear the full flags | |
799 | * on all the space infos. | |
800 | */ | |
801 | void btrfs_clear_space_info_full(struct btrfs_fs_info *info) | |
802 | { | |
803 | struct list_head *head = &info->space_info; | |
804 | struct btrfs_space_info *found; | |
805 | ||
806 | rcu_read_lock(); | |
807 | list_for_each_entry_rcu(found, head, list) | |
808 | found->full = 0; | |
809 | rcu_read_unlock(); | |
810 | } | |
811 | ||
1a4ed8fd | 812 | /* simple helper to search for an existing data extent at a given offset */ |
2ff7e61e | 813 | int btrfs_lookup_data_extent(struct btrfs_fs_info *fs_info, u64 start, u64 len) |
e02119d5 CM |
814 | { |
815 | int ret; | |
816 | struct btrfs_key key; | |
31840ae1 | 817 | struct btrfs_path *path; |
e02119d5 | 818 | |
31840ae1 | 819 | path = btrfs_alloc_path(); |
d8926bb3 MF |
820 | if (!path) |
821 | return -ENOMEM; | |
822 | ||
e02119d5 CM |
823 | key.objectid = start; |
824 | key.offset = len; | |
3173a18f | 825 | key.type = BTRFS_EXTENT_ITEM_KEY; |
0b246afa | 826 | ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0); |
31840ae1 | 827 | btrfs_free_path(path); |
7bb86316 CM |
828 | return ret; |
829 | } | |
830 | ||
a22285a6 | 831 | /* |
3173a18f | 832 | * helper function to lookup reference count and flags of a tree block. |
a22285a6 YZ |
833 | * |
834 | * the head node for delayed ref is used to store the sum of all the | |
835 | * reference count modifications queued up in the rbtree. the head | |
836 | * node may also store the extent flags to set. This way you can check | |
837 | * to see what the reference count and extent flags would be if all of | |
838 | * the delayed refs are not processed. | |
839 | */ | |
840 | int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans, | |
2ff7e61e | 841 | struct btrfs_fs_info *fs_info, u64 bytenr, |
3173a18f | 842 | u64 offset, int metadata, u64 *refs, u64 *flags) |
a22285a6 YZ |
843 | { |
844 | struct btrfs_delayed_ref_head *head; | |
845 | struct btrfs_delayed_ref_root *delayed_refs; | |
846 | struct btrfs_path *path; | |
847 | struct btrfs_extent_item *ei; | |
848 | struct extent_buffer *leaf; | |
849 | struct btrfs_key key; | |
850 | u32 item_size; | |
851 | u64 num_refs; | |
852 | u64 extent_flags; | |
853 | int ret; | |
854 | ||
3173a18f JB |
855 | /* |
856 | * If we don't have skinny metadata, don't bother doing anything | |
857 | * different | |
858 | */ | |
0b246afa JM |
859 | if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) { |
860 | offset = fs_info->nodesize; | |
3173a18f JB |
861 | metadata = 0; |
862 | } | |
863 | ||
a22285a6 YZ |
864 | path = btrfs_alloc_path(); |
865 | if (!path) | |
866 | return -ENOMEM; | |
867 | ||
a22285a6 YZ |
868 | if (!trans) { |
869 | path->skip_locking = 1; | |
870 | path->search_commit_root = 1; | |
871 | } | |
639eefc8 FDBM |
872 | |
873 | search_again: | |
874 | key.objectid = bytenr; | |
875 | key.offset = offset; | |
876 | if (metadata) | |
877 | key.type = BTRFS_METADATA_ITEM_KEY; | |
878 | else | |
879 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
880 | ||
0b246afa | 881 | ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0); |
a22285a6 YZ |
882 | if (ret < 0) |
883 | goto out_free; | |
884 | ||
3173a18f | 885 | if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) { |
74be9510 FDBM |
886 | if (path->slots[0]) { |
887 | path->slots[0]--; | |
888 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
889 | path->slots[0]); | |
890 | if (key.objectid == bytenr && | |
891 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
0b246afa | 892 | key.offset == fs_info->nodesize) |
74be9510 FDBM |
893 | ret = 0; |
894 | } | |
3173a18f JB |
895 | } |
896 | ||
a22285a6 YZ |
897 | if (ret == 0) { |
898 | leaf = path->nodes[0]; | |
899 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
900 | if (item_size >= sizeof(*ei)) { | |
901 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
902 | struct btrfs_extent_item); | |
903 | num_refs = btrfs_extent_refs(leaf, ei); | |
904 | extent_flags = btrfs_extent_flags(leaf, ei); | |
905 | } else { | |
ba3c2b19 NB |
906 | ret = -EINVAL; |
907 | btrfs_print_v0_err(fs_info); | |
908 | if (trans) | |
909 | btrfs_abort_transaction(trans, ret); | |
910 | else | |
911 | btrfs_handle_fs_error(fs_info, ret, NULL); | |
912 | ||
913 | goto out_free; | |
a22285a6 | 914 | } |
ba3c2b19 | 915 | |
a22285a6 YZ |
916 | BUG_ON(num_refs == 0); |
917 | } else { | |
918 | num_refs = 0; | |
919 | extent_flags = 0; | |
920 | ret = 0; | |
921 | } | |
922 | ||
923 | if (!trans) | |
924 | goto out; | |
925 | ||
926 | delayed_refs = &trans->transaction->delayed_refs; | |
927 | spin_lock(&delayed_refs->lock); | |
f72ad18e | 928 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
a22285a6 YZ |
929 | if (head) { |
930 | if (!mutex_trylock(&head->mutex)) { | |
d278850e | 931 | refcount_inc(&head->refs); |
a22285a6 YZ |
932 | spin_unlock(&delayed_refs->lock); |
933 | ||
b3b4aa74 | 934 | btrfs_release_path(path); |
a22285a6 | 935 | |
8cc33e5c DS |
936 | /* |
937 | * Mutex was contended, block until it's released and try | |
938 | * again | |
939 | */ | |
a22285a6 YZ |
940 | mutex_lock(&head->mutex); |
941 | mutex_unlock(&head->mutex); | |
d278850e | 942 | btrfs_put_delayed_ref_head(head); |
639eefc8 | 943 | goto search_again; |
a22285a6 | 944 | } |
d7df2c79 | 945 | spin_lock(&head->lock); |
a22285a6 YZ |
946 | if (head->extent_op && head->extent_op->update_flags) |
947 | extent_flags |= head->extent_op->flags_to_set; | |
948 | else | |
949 | BUG_ON(num_refs == 0); | |
950 | ||
d278850e | 951 | num_refs += head->ref_mod; |
d7df2c79 | 952 | spin_unlock(&head->lock); |
a22285a6 YZ |
953 | mutex_unlock(&head->mutex); |
954 | } | |
955 | spin_unlock(&delayed_refs->lock); | |
956 | out: | |
957 | WARN_ON(num_refs == 0); | |
958 | if (refs) | |
959 | *refs = num_refs; | |
960 | if (flags) | |
961 | *flags = extent_flags; | |
962 | out_free: | |
963 | btrfs_free_path(path); | |
964 | return ret; | |
965 | } | |
966 | ||
d8d5f3e1 CM |
967 | /* |
968 | * Back reference rules. Back refs have three main goals: | |
969 | * | |
970 | * 1) differentiate between all holders of references to an extent so that | |
971 | * when a reference is dropped we can make sure it was a valid reference | |
972 | * before freeing the extent. | |
973 | * | |
974 | * 2) Provide enough information to quickly find the holders of an extent | |
975 | * if we notice a given block is corrupted or bad. | |
976 | * | |
977 | * 3) Make it easy to migrate blocks for FS shrinking or storage pool | |
978 | * maintenance. This is actually the same as #2, but with a slightly | |
979 | * different use case. | |
980 | * | |
5d4f98a2 YZ |
981 | * There are two kinds of back refs. The implicit back refs is optimized |
982 | * for pointers in non-shared tree blocks. For a given pointer in a block, | |
983 | * back refs of this kind provide information about the block's owner tree | |
984 | * and the pointer's key. These information allow us to find the block by | |
985 | * b-tree searching. The full back refs is for pointers in tree blocks not | |
986 | * referenced by their owner trees. The location of tree block is recorded | |
987 | * in the back refs. Actually the full back refs is generic, and can be | |
988 | * used in all cases the implicit back refs is used. The major shortcoming | |
989 | * of the full back refs is its overhead. Every time a tree block gets | |
990 | * COWed, we have to update back refs entry for all pointers in it. | |
991 | * | |
992 | * For a newly allocated tree block, we use implicit back refs for | |
993 | * pointers in it. This means most tree related operations only involve | |
994 | * implicit back refs. For a tree block created in old transaction, the | |
995 | * only way to drop a reference to it is COW it. So we can detect the | |
996 | * event that tree block loses its owner tree's reference and do the | |
997 | * back refs conversion. | |
998 | * | |
01327610 | 999 | * When a tree block is COWed through a tree, there are four cases: |
5d4f98a2 YZ |
1000 | * |
1001 | * The reference count of the block is one and the tree is the block's | |
1002 | * owner tree. Nothing to do in this case. | |
1003 | * | |
1004 | * The reference count of the block is one and the tree is not the | |
1005 | * block's owner tree. In this case, full back refs is used for pointers | |
1006 | * in the block. Remove these full back refs, add implicit back refs for | |
1007 | * every pointers in the new block. | |
1008 | * | |
1009 | * The reference count of the block is greater than one and the tree is | |
1010 | * the block's owner tree. In this case, implicit back refs is used for | |
1011 | * pointers in the block. Add full back refs for every pointers in the | |
1012 | * block, increase lower level extents' reference counts. The original | |
1013 | * implicit back refs are entailed to the new block. | |
1014 | * | |
1015 | * The reference count of the block is greater than one and the tree is | |
1016 | * not the block's owner tree. Add implicit back refs for every pointer in | |
1017 | * the new block, increase lower level extents' reference count. | |
1018 | * | |
1019 | * Back Reference Key composing: | |
1020 | * | |
1021 | * The key objectid corresponds to the first byte in the extent, | |
1022 | * The key type is used to differentiate between types of back refs. | |
1023 | * There are different meanings of the key offset for different types | |
1024 | * of back refs. | |
1025 | * | |
d8d5f3e1 CM |
1026 | * File extents can be referenced by: |
1027 | * | |
1028 | * - multiple snapshots, subvolumes, or different generations in one subvol | |
31840ae1 | 1029 | * - different files inside a single subvolume |
d8d5f3e1 CM |
1030 | * - different offsets inside a file (bookend extents in file.c) |
1031 | * | |
5d4f98a2 | 1032 | * The extent ref structure for the implicit back refs has fields for: |
d8d5f3e1 CM |
1033 | * |
1034 | * - Objectid of the subvolume root | |
d8d5f3e1 | 1035 | * - objectid of the file holding the reference |
5d4f98a2 YZ |
1036 | * - original offset in the file |
1037 | * - how many bookend extents | |
d8d5f3e1 | 1038 | * |
5d4f98a2 YZ |
1039 | * The key offset for the implicit back refs is hash of the first |
1040 | * three fields. | |
d8d5f3e1 | 1041 | * |
5d4f98a2 | 1042 | * The extent ref structure for the full back refs has field for: |
d8d5f3e1 | 1043 | * |
5d4f98a2 | 1044 | * - number of pointers in the tree leaf |
d8d5f3e1 | 1045 | * |
5d4f98a2 YZ |
1046 | * The key offset for the implicit back refs is the first byte of |
1047 | * the tree leaf | |
d8d5f3e1 | 1048 | * |
5d4f98a2 YZ |
1049 | * When a file extent is allocated, The implicit back refs is used. |
1050 | * the fields are filled in: | |
d8d5f3e1 | 1051 | * |
5d4f98a2 | 1052 | * (root_key.objectid, inode objectid, offset in file, 1) |
d8d5f3e1 | 1053 | * |
5d4f98a2 YZ |
1054 | * When a file extent is removed file truncation, we find the |
1055 | * corresponding implicit back refs and check the following fields: | |
d8d5f3e1 | 1056 | * |
5d4f98a2 | 1057 | * (btrfs_header_owner(leaf), inode objectid, offset in file) |
d8d5f3e1 | 1058 | * |
5d4f98a2 | 1059 | * Btree extents can be referenced by: |
d8d5f3e1 | 1060 | * |
5d4f98a2 | 1061 | * - Different subvolumes |
d8d5f3e1 | 1062 | * |
5d4f98a2 YZ |
1063 | * Both the implicit back refs and the full back refs for tree blocks |
1064 | * only consist of key. The key offset for the implicit back refs is | |
1065 | * objectid of block's owner tree. The key offset for the full back refs | |
1066 | * is the first byte of parent block. | |
d8d5f3e1 | 1067 | * |
5d4f98a2 YZ |
1068 | * When implicit back refs is used, information about the lowest key and |
1069 | * level of the tree block are required. These information are stored in | |
1070 | * tree block info structure. | |
d8d5f3e1 | 1071 | */ |
31840ae1 | 1072 | |
167ce953 LB |
1073 | /* |
1074 | * is_data == BTRFS_REF_TYPE_BLOCK, tree block type is required, | |
52042d8e | 1075 | * is_data == BTRFS_REF_TYPE_DATA, data type is requiried, |
167ce953 LB |
1076 | * is_data == BTRFS_REF_TYPE_ANY, either type is OK. |
1077 | */ | |
1078 | int btrfs_get_extent_inline_ref_type(const struct extent_buffer *eb, | |
1079 | struct btrfs_extent_inline_ref *iref, | |
1080 | enum btrfs_inline_ref_type is_data) | |
1081 | { | |
1082 | int type = btrfs_extent_inline_ref_type(eb, iref); | |
64ecdb64 | 1083 | u64 offset = btrfs_extent_inline_ref_offset(eb, iref); |
167ce953 LB |
1084 | |
1085 | if (type == BTRFS_TREE_BLOCK_REF_KEY || | |
1086 | type == BTRFS_SHARED_BLOCK_REF_KEY || | |
1087 | type == BTRFS_SHARED_DATA_REF_KEY || | |
1088 | type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1089 | if (is_data == BTRFS_REF_TYPE_BLOCK) { | |
64ecdb64 | 1090 | if (type == BTRFS_TREE_BLOCK_REF_KEY) |
167ce953 | 1091 | return type; |
64ecdb64 LB |
1092 | if (type == BTRFS_SHARED_BLOCK_REF_KEY) { |
1093 | ASSERT(eb->fs_info); | |
1094 | /* | |
1095 | * Every shared one has parent tree | |
1096 | * block, which must be aligned to | |
1097 | * nodesize. | |
1098 | */ | |
1099 | if (offset && | |
1100 | IS_ALIGNED(offset, eb->fs_info->nodesize)) | |
1101 | return type; | |
1102 | } | |
167ce953 | 1103 | } else if (is_data == BTRFS_REF_TYPE_DATA) { |
64ecdb64 | 1104 | if (type == BTRFS_EXTENT_DATA_REF_KEY) |
167ce953 | 1105 | return type; |
64ecdb64 LB |
1106 | if (type == BTRFS_SHARED_DATA_REF_KEY) { |
1107 | ASSERT(eb->fs_info); | |
1108 | /* | |
1109 | * Every shared one has parent tree | |
1110 | * block, which must be aligned to | |
1111 | * nodesize. | |
1112 | */ | |
1113 | if (offset && | |
1114 | IS_ALIGNED(offset, eb->fs_info->nodesize)) | |
1115 | return type; | |
1116 | } | |
167ce953 LB |
1117 | } else { |
1118 | ASSERT(is_data == BTRFS_REF_TYPE_ANY); | |
1119 | return type; | |
1120 | } | |
1121 | } | |
1122 | ||
1123 | btrfs_print_leaf((struct extent_buffer *)eb); | |
1124 | btrfs_err(eb->fs_info, "eb %llu invalid extent inline ref type %d", | |
1125 | eb->start, type); | |
1126 | WARN_ON(1); | |
1127 | ||
1128 | return BTRFS_REF_TYPE_INVALID; | |
1129 | } | |
1130 | ||
5d4f98a2 YZ |
1131 | static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset) |
1132 | { | |
1133 | u32 high_crc = ~(u32)0; | |
1134 | u32 low_crc = ~(u32)0; | |
1135 | __le64 lenum; | |
1136 | ||
1137 | lenum = cpu_to_le64(root_objectid); | |
9678c543 | 1138 | high_crc = crc32c(high_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 1139 | lenum = cpu_to_le64(owner); |
9678c543 | 1140 | low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 1141 | lenum = cpu_to_le64(offset); |
9678c543 | 1142 | low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 YZ |
1143 | |
1144 | return ((u64)high_crc << 31) ^ (u64)low_crc; | |
1145 | } | |
1146 | ||
1147 | static u64 hash_extent_data_ref_item(struct extent_buffer *leaf, | |
1148 | struct btrfs_extent_data_ref *ref) | |
1149 | { | |
1150 | return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref), | |
1151 | btrfs_extent_data_ref_objectid(leaf, ref), | |
1152 | btrfs_extent_data_ref_offset(leaf, ref)); | |
1153 | } | |
1154 | ||
1155 | static int match_extent_data_ref(struct extent_buffer *leaf, | |
1156 | struct btrfs_extent_data_ref *ref, | |
1157 | u64 root_objectid, u64 owner, u64 offset) | |
1158 | { | |
1159 | if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid || | |
1160 | btrfs_extent_data_ref_objectid(leaf, ref) != owner || | |
1161 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
1162 | return 0; | |
1163 | return 1; | |
1164 | } | |
1165 | ||
1166 | static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
1167 | struct btrfs_path *path, |
1168 | u64 bytenr, u64 parent, | |
1169 | u64 root_objectid, | |
1170 | u64 owner, u64 offset) | |
1171 | { | |
bd1d53ef | 1172 | struct btrfs_root *root = trans->fs_info->extent_root; |
5d4f98a2 YZ |
1173 | struct btrfs_key key; |
1174 | struct btrfs_extent_data_ref *ref; | |
31840ae1 | 1175 | struct extent_buffer *leaf; |
5d4f98a2 | 1176 | u32 nritems; |
74493f7a | 1177 | int ret; |
5d4f98a2 YZ |
1178 | int recow; |
1179 | int err = -ENOENT; | |
74493f7a | 1180 | |
31840ae1 | 1181 | key.objectid = bytenr; |
5d4f98a2 YZ |
1182 | if (parent) { |
1183 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1184 | key.offset = parent; | |
1185 | } else { | |
1186 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1187 | key.offset = hash_extent_data_ref(root_objectid, | |
1188 | owner, offset); | |
1189 | } | |
1190 | again: | |
1191 | recow = 0; | |
1192 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1193 | if (ret < 0) { | |
1194 | err = ret; | |
1195 | goto fail; | |
1196 | } | |
31840ae1 | 1197 | |
5d4f98a2 YZ |
1198 | if (parent) { |
1199 | if (!ret) | |
1200 | return 0; | |
5d4f98a2 | 1201 | goto fail; |
31840ae1 ZY |
1202 | } |
1203 | ||
1204 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1205 | nritems = btrfs_header_nritems(leaf); |
1206 | while (1) { | |
1207 | if (path->slots[0] >= nritems) { | |
1208 | ret = btrfs_next_leaf(root, path); | |
1209 | if (ret < 0) | |
1210 | err = ret; | |
1211 | if (ret) | |
1212 | goto fail; | |
1213 | ||
1214 | leaf = path->nodes[0]; | |
1215 | nritems = btrfs_header_nritems(leaf); | |
1216 | recow = 1; | |
1217 | } | |
1218 | ||
1219 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1220 | if (key.objectid != bytenr || | |
1221 | key.type != BTRFS_EXTENT_DATA_REF_KEY) | |
1222 | goto fail; | |
1223 | ||
1224 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1225 | struct btrfs_extent_data_ref); | |
1226 | ||
1227 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1228 | owner, offset)) { | |
1229 | if (recow) { | |
b3b4aa74 | 1230 | btrfs_release_path(path); |
5d4f98a2 YZ |
1231 | goto again; |
1232 | } | |
1233 | err = 0; | |
1234 | break; | |
1235 | } | |
1236 | path->slots[0]++; | |
31840ae1 | 1237 | } |
5d4f98a2 YZ |
1238 | fail: |
1239 | return err; | |
31840ae1 ZY |
1240 | } |
1241 | ||
5d4f98a2 | 1242 | static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1243 | struct btrfs_path *path, |
1244 | u64 bytenr, u64 parent, | |
1245 | u64 root_objectid, u64 owner, | |
1246 | u64 offset, int refs_to_add) | |
31840ae1 | 1247 | { |
62b895af | 1248 | struct btrfs_root *root = trans->fs_info->extent_root; |
31840ae1 ZY |
1249 | struct btrfs_key key; |
1250 | struct extent_buffer *leaf; | |
5d4f98a2 | 1251 | u32 size; |
31840ae1 ZY |
1252 | u32 num_refs; |
1253 | int ret; | |
74493f7a | 1254 | |
74493f7a | 1255 | key.objectid = bytenr; |
5d4f98a2 YZ |
1256 | if (parent) { |
1257 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1258 | key.offset = parent; | |
1259 | size = sizeof(struct btrfs_shared_data_ref); | |
1260 | } else { | |
1261 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1262 | key.offset = hash_extent_data_ref(root_objectid, | |
1263 | owner, offset); | |
1264 | size = sizeof(struct btrfs_extent_data_ref); | |
1265 | } | |
74493f7a | 1266 | |
5d4f98a2 YZ |
1267 | ret = btrfs_insert_empty_item(trans, root, path, &key, size); |
1268 | if (ret && ret != -EEXIST) | |
1269 | goto fail; | |
1270 | ||
1271 | leaf = path->nodes[0]; | |
1272 | if (parent) { | |
1273 | struct btrfs_shared_data_ref *ref; | |
31840ae1 | 1274 | ref = btrfs_item_ptr(leaf, path->slots[0], |
5d4f98a2 YZ |
1275 | struct btrfs_shared_data_ref); |
1276 | if (ret == 0) { | |
1277 | btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add); | |
1278 | } else { | |
1279 | num_refs = btrfs_shared_data_ref_count(leaf, ref); | |
1280 | num_refs += refs_to_add; | |
1281 | btrfs_set_shared_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1282 | } |
5d4f98a2 YZ |
1283 | } else { |
1284 | struct btrfs_extent_data_ref *ref; | |
1285 | while (ret == -EEXIST) { | |
1286 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1287 | struct btrfs_extent_data_ref); | |
1288 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1289 | owner, offset)) | |
1290 | break; | |
b3b4aa74 | 1291 | btrfs_release_path(path); |
5d4f98a2 YZ |
1292 | key.offset++; |
1293 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1294 | size); | |
1295 | if (ret && ret != -EEXIST) | |
1296 | goto fail; | |
31840ae1 | 1297 | |
5d4f98a2 YZ |
1298 | leaf = path->nodes[0]; |
1299 | } | |
1300 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1301 | struct btrfs_extent_data_ref); | |
1302 | if (ret == 0) { | |
1303 | btrfs_set_extent_data_ref_root(leaf, ref, | |
1304 | root_objectid); | |
1305 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
1306 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
1307 | btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add); | |
1308 | } else { | |
1309 | num_refs = btrfs_extent_data_ref_count(leaf, ref); | |
1310 | num_refs += refs_to_add; | |
1311 | btrfs_set_extent_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1312 | } |
31840ae1 | 1313 | } |
5d4f98a2 YZ |
1314 | btrfs_mark_buffer_dirty(leaf); |
1315 | ret = 0; | |
1316 | fail: | |
b3b4aa74 | 1317 | btrfs_release_path(path); |
7bb86316 | 1318 | return ret; |
74493f7a CM |
1319 | } |
1320 | ||
5d4f98a2 | 1321 | static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 | 1322 | struct btrfs_path *path, |
fcebe456 | 1323 | int refs_to_drop, int *last_ref) |
31840ae1 | 1324 | { |
5d4f98a2 YZ |
1325 | struct btrfs_key key; |
1326 | struct btrfs_extent_data_ref *ref1 = NULL; | |
1327 | struct btrfs_shared_data_ref *ref2 = NULL; | |
31840ae1 | 1328 | struct extent_buffer *leaf; |
5d4f98a2 | 1329 | u32 num_refs = 0; |
31840ae1 ZY |
1330 | int ret = 0; |
1331 | ||
1332 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1333 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
1334 | ||
1335 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1336 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1337 | struct btrfs_extent_data_ref); | |
1338 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1339 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1340 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1341 | struct btrfs_shared_data_ref); | |
1342 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
6d8ff4e4 | 1343 | } else if (unlikely(key.type == BTRFS_EXTENT_REF_V0_KEY)) { |
ba3c2b19 NB |
1344 | btrfs_print_v0_err(trans->fs_info); |
1345 | btrfs_abort_transaction(trans, -EINVAL); | |
1346 | return -EINVAL; | |
5d4f98a2 YZ |
1347 | } else { |
1348 | BUG(); | |
1349 | } | |
1350 | ||
56bec294 CM |
1351 | BUG_ON(num_refs < refs_to_drop); |
1352 | num_refs -= refs_to_drop; | |
5d4f98a2 | 1353 | |
31840ae1 | 1354 | if (num_refs == 0) { |
e9f6290d | 1355 | ret = btrfs_del_item(trans, trans->fs_info->extent_root, path); |
fcebe456 | 1356 | *last_ref = 1; |
31840ae1 | 1357 | } else { |
5d4f98a2 YZ |
1358 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) |
1359 | btrfs_set_extent_data_ref_count(leaf, ref1, num_refs); | |
1360 | else if (key.type == BTRFS_SHARED_DATA_REF_KEY) | |
1361 | btrfs_set_shared_data_ref_count(leaf, ref2, num_refs); | |
31840ae1 ZY |
1362 | btrfs_mark_buffer_dirty(leaf); |
1363 | } | |
31840ae1 ZY |
1364 | return ret; |
1365 | } | |
1366 | ||
9ed0dea0 | 1367 | static noinline u32 extent_data_ref_count(struct btrfs_path *path, |
5d4f98a2 | 1368 | struct btrfs_extent_inline_ref *iref) |
15916de8 | 1369 | { |
5d4f98a2 YZ |
1370 | struct btrfs_key key; |
1371 | struct extent_buffer *leaf; | |
1372 | struct btrfs_extent_data_ref *ref1; | |
1373 | struct btrfs_shared_data_ref *ref2; | |
1374 | u32 num_refs = 0; | |
3de28d57 | 1375 | int type; |
5d4f98a2 YZ |
1376 | |
1377 | leaf = path->nodes[0]; | |
1378 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
ba3c2b19 NB |
1379 | |
1380 | BUG_ON(key.type == BTRFS_EXTENT_REF_V0_KEY); | |
5d4f98a2 | 1381 | if (iref) { |
3de28d57 LB |
1382 | /* |
1383 | * If type is invalid, we should have bailed out earlier than | |
1384 | * this call. | |
1385 | */ | |
1386 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA); | |
1387 | ASSERT(type != BTRFS_REF_TYPE_INVALID); | |
1388 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
5d4f98a2 YZ |
1389 | ref1 = (struct btrfs_extent_data_ref *)(&iref->offset); |
1390 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1391 | } else { | |
1392 | ref2 = (struct btrfs_shared_data_ref *)(iref + 1); | |
1393 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1394 | } | |
1395 | } else 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); | |
5d4f98a2 YZ |
1403 | } else { |
1404 | WARN_ON(1); | |
1405 | } | |
1406 | return num_refs; | |
1407 | } | |
15916de8 | 1408 | |
5d4f98a2 | 1409 | static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1410 | struct btrfs_path *path, |
1411 | u64 bytenr, u64 parent, | |
1412 | u64 root_objectid) | |
1f3c79a2 | 1413 | { |
b8582eea | 1414 | struct btrfs_root *root = trans->fs_info->extent_root; |
5d4f98a2 | 1415 | struct btrfs_key key; |
1f3c79a2 | 1416 | int ret; |
1f3c79a2 | 1417 | |
5d4f98a2 YZ |
1418 | key.objectid = bytenr; |
1419 | if (parent) { | |
1420 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1421 | key.offset = parent; | |
1422 | } else { | |
1423 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1424 | key.offset = root_objectid; | |
1f3c79a2 LH |
1425 | } |
1426 | ||
5d4f98a2 YZ |
1427 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1428 | if (ret > 0) | |
1429 | ret = -ENOENT; | |
5d4f98a2 | 1430 | return ret; |
1f3c79a2 LH |
1431 | } |
1432 | ||
5d4f98a2 | 1433 | static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1434 | struct btrfs_path *path, |
1435 | u64 bytenr, u64 parent, | |
1436 | u64 root_objectid) | |
31840ae1 | 1437 | { |
5d4f98a2 | 1438 | struct btrfs_key key; |
31840ae1 | 1439 | int ret; |
31840ae1 | 1440 | |
5d4f98a2 YZ |
1441 | key.objectid = bytenr; |
1442 | if (parent) { | |
1443 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1444 | key.offset = parent; | |
1445 | } else { | |
1446 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1447 | key.offset = root_objectid; | |
1448 | } | |
1449 | ||
10728404 | 1450 | ret = btrfs_insert_empty_item(trans, trans->fs_info->extent_root, |
87bde3cd | 1451 | path, &key, 0); |
b3b4aa74 | 1452 | btrfs_release_path(path); |
31840ae1 ZY |
1453 | return ret; |
1454 | } | |
1455 | ||
5d4f98a2 | 1456 | static inline int extent_ref_type(u64 parent, u64 owner) |
31840ae1 | 1457 | { |
5d4f98a2 YZ |
1458 | int type; |
1459 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1460 | if (parent > 0) | |
1461 | type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1462 | else | |
1463 | type = BTRFS_TREE_BLOCK_REF_KEY; | |
1464 | } else { | |
1465 | if (parent > 0) | |
1466 | type = BTRFS_SHARED_DATA_REF_KEY; | |
1467 | else | |
1468 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
1469 | } | |
1470 | return type; | |
31840ae1 | 1471 | } |
56bec294 | 1472 | |
2c47e605 YZ |
1473 | static int find_next_key(struct btrfs_path *path, int level, |
1474 | struct btrfs_key *key) | |
56bec294 | 1475 | |
02217ed2 | 1476 | { |
2c47e605 | 1477 | for (; level < BTRFS_MAX_LEVEL; level++) { |
5d4f98a2 YZ |
1478 | if (!path->nodes[level]) |
1479 | break; | |
5d4f98a2 YZ |
1480 | if (path->slots[level] + 1 >= |
1481 | btrfs_header_nritems(path->nodes[level])) | |
1482 | continue; | |
1483 | if (level == 0) | |
1484 | btrfs_item_key_to_cpu(path->nodes[level], key, | |
1485 | path->slots[level] + 1); | |
1486 | else | |
1487 | btrfs_node_key_to_cpu(path->nodes[level], key, | |
1488 | path->slots[level] + 1); | |
1489 | return 0; | |
1490 | } | |
1491 | return 1; | |
1492 | } | |
037e6390 | 1493 | |
5d4f98a2 YZ |
1494 | /* |
1495 | * look for inline back ref. if back ref is found, *ref_ret is set | |
1496 | * to the address of inline back ref, and 0 is returned. | |
1497 | * | |
1498 | * if back ref isn't found, *ref_ret is set to the address where it | |
1499 | * should be inserted, and -ENOENT is returned. | |
1500 | * | |
1501 | * if insert is true and there are too many inline back refs, the path | |
1502 | * points to the extent item, and -EAGAIN is returned. | |
1503 | * | |
1504 | * NOTE: inline back refs are ordered in the same way that back ref | |
1505 | * items in the tree are ordered. | |
1506 | */ | |
1507 | static noinline_for_stack | |
1508 | int lookup_inline_extent_backref(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
1509 | struct btrfs_path *path, |
1510 | struct btrfs_extent_inline_ref **ref_ret, | |
1511 | u64 bytenr, u64 num_bytes, | |
1512 | u64 parent, u64 root_objectid, | |
1513 | u64 owner, u64 offset, int insert) | |
1514 | { | |
867cc1fb | 1515 | struct btrfs_fs_info *fs_info = trans->fs_info; |
87bde3cd | 1516 | struct btrfs_root *root = fs_info->extent_root; |
5d4f98a2 YZ |
1517 | struct btrfs_key key; |
1518 | struct extent_buffer *leaf; | |
1519 | struct btrfs_extent_item *ei; | |
1520 | struct btrfs_extent_inline_ref *iref; | |
1521 | u64 flags; | |
1522 | u64 item_size; | |
1523 | unsigned long ptr; | |
1524 | unsigned long end; | |
1525 | int extra_size; | |
1526 | int type; | |
1527 | int want; | |
1528 | int ret; | |
1529 | int err = 0; | |
0b246afa | 1530 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); |
3de28d57 | 1531 | int needed; |
26b8003f | 1532 | |
db94535d | 1533 | key.objectid = bytenr; |
31840ae1 | 1534 | key.type = BTRFS_EXTENT_ITEM_KEY; |
56bec294 | 1535 | key.offset = num_bytes; |
31840ae1 | 1536 | |
5d4f98a2 YZ |
1537 | want = extent_ref_type(parent, owner); |
1538 | if (insert) { | |
1539 | extra_size = btrfs_extent_inline_ref_size(want); | |
85d4198e | 1540 | path->keep_locks = 1; |
5d4f98a2 YZ |
1541 | } else |
1542 | extra_size = -1; | |
3173a18f JB |
1543 | |
1544 | /* | |
16d1c062 NB |
1545 | * Owner is our level, so we can just add one to get the level for the |
1546 | * block we are interested in. | |
3173a18f JB |
1547 | */ |
1548 | if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1549 | key.type = BTRFS_METADATA_ITEM_KEY; | |
1550 | key.offset = owner; | |
1551 | } | |
1552 | ||
1553 | again: | |
5d4f98a2 | 1554 | ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1); |
b9473439 | 1555 | if (ret < 0) { |
5d4f98a2 YZ |
1556 | err = ret; |
1557 | goto out; | |
1558 | } | |
3173a18f JB |
1559 | |
1560 | /* | |
1561 | * We may be a newly converted file system which still has the old fat | |
1562 | * extent entries for metadata, so try and see if we have one of those. | |
1563 | */ | |
1564 | if (ret > 0 && skinny_metadata) { | |
1565 | skinny_metadata = false; | |
1566 | if (path->slots[0]) { | |
1567 | path->slots[0]--; | |
1568 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
1569 | path->slots[0]); | |
1570 | if (key.objectid == bytenr && | |
1571 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
1572 | key.offset == num_bytes) | |
1573 | ret = 0; | |
1574 | } | |
1575 | if (ret) { | |
9ce49a0b | 1576 | key.objectid = bytenr; |
3173a18f JB |
1577 | key.type = BTRFS_EXTENT_ITEM_KEY; |
1578 | key.offset = num_bytes; | |
1579 | btrfs_release_path(path); | |
1580 | goto again; | |
1581 | } | |
1582 | } | |
1583 | ||
79787eaa JM |
1584 | if (ret && !insert) { |
1585 | err = -ENOENT; | |
1586 | goto out; | |
fae7f21c | 1587 | } else if (WARN_ON(ret)) { |
492104c8 | 1588 | err = -EIO; |
492104c8 | 1589 | goto out; |
79787eaa | 1590 | } |
5d4f98a2 YZ |
1591 | |
1592 | leaf = path->nodes[0]; | |
1593 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
6d8ff4e4 | 1594 | if (unlikely(item_size < sizeof(*ei))) { |
ba3c2b19 NB |
1595 | err = -EINVAL; |
1596 | btrfs_print_v0_err(fs_info); | |
1597 | btrfs_abort_transaction(trans, err); | |
1598 | goto out; | |
1599 | } | |
5d4f98a2 | 1600 | |
5d4f98a2 YZ |
1601 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1602 | flags = btrfs_extent_flags(leaf, ei); | |
1603 | ||
1604 | ptr = (unsigned long)(ei + 1); | |
1605 | end = (unsigned long)ei + item_size; | |
1606 | ||
3173a18f | 1607 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) { |
5d4f98a2 YZ |
1608 | ptr += sizeof(struct btrfs_tree_block_info); |
1609 | BUG_ON(ptr > end); | |
5d4f98a2 YZ |
1610 | } |
1611 | ||
3de28d57 LB |
1612 | if (owner >= BTRFS_FIRST_FREE_OBJECTID) |
1613 | needed = BTRFS_REF_TYPE_DATA; | |
1614 | else | |
1615 | needed = BTRFS_REF_TYPE_BLOCK; | |
1616 | ||
5d4f98a2 YZ |
1617 | err = -ENOENT; |
1618 | while (1) { | |
1619 | if (ptr >= end) { | |
1620 | WARN_ON(ptr > end); | |
1621 | break; | |
1622 | } | |
1623 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
3de28d57 LB |
1624 | type = btrfs_get_extent_inline_ref_type(leaf, iref, needed); |
1625 | if (type == BTRFS_REF_TYPE_INVALID) { | |
af431dcb | 1626 | err = -EUCLEAN; |
3de28d57 LB |
1627 | goto out; |
1628 | } | |
1629 | ||
5d4f98a2 YZ |
1630 | if (want < type) |
1631 | break; | |
1632 | if (want > type) { | |
1633 | ptr += btrfs_extent_inline_ref_size(type); | |
1634 | continue; | |
1635 | } | |
1636 | ||
1637 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1638 | struct btrfs_extent_data_ref *dref; | |
1639 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1640 | if (match_extent_data_ref(leaf, dref, root_objectid, | |
1641 | owner, offset)) { | |
1642 | err = 0; | |
1643 | break; | |
1644 | } | |
1645 | if (hash_extent_data_ref_item(leaf, dref) < | |
1646 | hash_extent_data_ref(root_objectid, owner, offset)) | |
1647 | break; | |
1648 | } else { | |
1649 | u64 ref_offset; | |
1650 | ref_offset = btrfs_extent_inline_ref_offset(leaf, iref); | |
1651 | if (parent > 0) { | |
1652 | if (parent == ref_offset) { | |
1653 | err = 0; | |
1654 | break; | |
1655 | } | |
1656 | if (ref_offset < parent) | |
1657 | break; | |
1658 | } else { | |
1659 | if (root_objectid == ref_offset) { | |
1660 | err = 0; | |
1661 | break; | |
1662 | } | |
1663 | if (ref_offset < root_objectid) | |
1664 | break; | |
1665 | } | |
1666 | } | |
1667 | ptr += btrfs_extent_inline_ref_size(type); | |
1668 | } | |
1669 | if (err == -ENOENT && insert) { | |
1670 | if (item_size + extra_size >= | |
1671 | BTRFS_MAX_EXTENT_ITEM_SIZE(root)) { | |
1672 | err = -EAGAIN; | |
1673 | goto out; | |
1674 | } | |
1675 | /* | |
1676 | * To add new inline back ref, we have to make sure | |
1677 | * there is no corresponding back ref item. | |
1678 | * For simplicity, we just do not add new inline back | |
1679 | * ref if there is any kind of item for this block | |
1680 | */ | |
2c47e605 YZ |
1681 | if (find_next_key(path, 0, &key) == 0 && |
1682 | key.objectid == bytenr && | |
85d4198e | 1683 | key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) { |
5d4f98a2 YZ |
1684 | err = -EAGAIN; |
1685 | goto out; | |
1686 | } | |
1687 | } | |
1688 | *ref_ret = (struct btrfs_extent_inline_ref *)ptr; | |
1689 | out: | |
85d4198e | 1690 | if (insert) { |
5d4f98a2 YZ |
1691 | path->keep_locks = 0; |
1692 | btrfs_unlock_up_safe(path, 1); | |
1693 | } | |
1694 | return err; | |
1695 | } | |
1696 | ||
1697 | /* | |
1698 | * helper to add new inline back ref | |
1699 | */ | |
1700 | static noinline_for_stack | |
87bde3cd | 1701 | void setup_inline_extent_backref(struct btrfs_fs_info *fs_info, |
143bede5 JM |
1702 | struct btrfs_path *path, |
1703 | struct btrfs_extent_inline_ref *iref, | |
1704 | u64 parent, u64 root_objectid, | |
1705 | u64 owner, u64 offset, int refs_to_add, | |
1706 | struct btrfs_delayed_extent_op *extent_op) | |
5d4f98a2 YZ |
1707 | { |
1708 | struct extent_buffer *leaf; | |
1709 | struct btrfs_extent_item *ei; | |
1710 | unsigned long ptr; | |
1711 | unsigned long end; | |
1712 | unsigned long item_offset; | |
1713 | u64 refs; | |
1714 | int size; | |
1715 | int type; | |
5d4f98a2 YZ |
1716 | |
1717 | leaf = path->nodes[0]; | |
1718 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1719 | item_offset = (unsigned long)iref - (unsigned long)ei; | |
1720 | ||
1721 | type = extent_ref_type(parent, owner); | |
1722 | size = btrfs_extent_inline_ref_size(type); | |
1723 | ||
c71dd880 | 1724 | btrfs_extend_item(path, size); |
5d4f98a2 YZ |
1725 | |
1726 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1727 | refs = btrfs_extent_refs(leaf, ei); | |
1728 | refs += refs_to_add; | |
1729 | btrfs_set_extent_refs(leaf, ei, refs); | |
1730 | if (extent_op) | |
1731 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1732 | ||
1733 | ptr = (unsigned long)ei + item_offset; | |
1734 | end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]); | |
1735 | if (ptr < end - size) | |
1736 | memmove_extent_buffer(leaf, ptr + size, ptr, | |
1737 | end - size - ptr); | |
1738 | ||
1739 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1740 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
1741 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1742 | struct btrfs_extent_data_ref *dref; | |
1743 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1744 | btrfs_set_extent_data_ref_root(leaf, dref, root_objectid); | |
1745 | btrfs_set_extent_data_ref_objectid(leaf, dref, owner); | |
1746 | btrfs_set_extent_data_ref_offset(leaf, dref, offset); | |
1747 | btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add); | |
1748 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1749 | struct btrfs_shared_data_ref *sref; | |
1750 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1751 | btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add); | |
1752 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1753 | } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) { | |
1754 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1755 | } else { | |
1756 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
1757 | } | |
1758 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1759 | } |
1760 | ||
1761 | static int lookup_extent_backref(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
1762 | struct btrfs_path *path, |
1763 | struct btrfs_extent_inline_ref **ref_ret, | |
1764 | u64 bytenr, u64 num_bytes, u64 parent, | |
1765 | u64 root_objectid, u64 owner, u64 offset) | |
1766 | { | |
1767 | int ret; | |
1768 | ||
867cc1fb NB |
1769 | ret = lookup_inline_extent_backref(trans, path, ref_ret, bytenr, |
1770 | num_bytes, parent, root_objectid, | |
1771 | owner, offset, 0); | |
5d4f98a2 | 1772 | if (ret != -ENOENT) |
54aa1f4d | 1773 | return ret; |
5d4f98a2 | 1774 | |
b3b4aa74 | 1775 | btrfs_release_path(path); |
5d4f98a2 YZ |
1776 | *ref_ret = NULL; |
1777 | ||
1778 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
b8582eea NB |
1779 | ret = lookup_tree_block_ref(trans, path, bytenr, parent, |
1780 | root_objectid); | |
5d4f98a2 | 1781 | } else { |
bd1d53ef NB |
1782 | ret = lookup_extent_data_ref(trans, path, bytenr, parent, |
1783 | root_objectid, owner, offset); | |
b9473439 | 1784 | } |
5d4f98a2 YZ |
1785 | return ret; |
1786 | } | |
31840ae1 | 1787 | |
5d4f98a2 YZ |
1788 | /* |
1789 | * helper to update/remove inline back ref | |
1790 | */ | |
1791 | static noinline_for_stack | |
61a18f1c | 1792 | void update_inline_extent_backref(struct btrfs_path *path, |
143bede5 JM |
1793 | struct btrfs_extent_inline_ref *iref, |
1794 | int refs_to_mod, | |
fcebe456 JB |
1795 | struct btrfs_delayed_extent_op *extent_op, |
1796 | int *last_ref) | |
5d4f98a2 | 1797 | { |
61a18f1c | 1798 | struct extent_buffer *leaf = path->nodes[0]; |
5d4f98a2 YZ |
1799 | struct btrfs_extent_item *ei; |
1800 | struct btrfs_extent_data_ref *dref = NULL; | |
1801 | struct btrfs_shared_data_ref *sref = NULL; | |
1802 | unsigned long ptr; | |
1803 | unsigned long end; | |
1804 | u32 item_size; | |
1805 | int size; | |
1806 | int type; | |
5d4f98a2 YZ |
1807 | u64 refs; |
1808 | ||
5d4f98a2 YZ |
1809 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1810 | refs = btrfs_extent_refs(leaf, ei); | |
1811 | WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0); | |
1812 | refs += refs_to_mod; | |
1813 | btrfs_set_extent_refs(leaf, ei, refs); | |
1814 | if (extent_op) | |
1815 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1816 | ||
3de28d57 LB |
1817 | /* |
1818 | * If type is invalid, we should have bailed out after | |
1819 | * lookup_inline_extent_backref(). | |
1820 | */ | |
1821 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_ANY); | |
1822 | ASSERT(type != BTRFS_REF_TYPE_INVALID); | |
5d4f98a2 YZ |
1823 | |
1824 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1825 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1826 | refs = btrfs_extent_data_ref_count(leaf, dref); | |
1827 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1828 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1829 | refs = btrfs_shared_data_ref_count(leaf, sref); | |
1830 | } else { | |
1831 | refs = 1; | |
1832 | BUG_ON(refs_to_mod != -1); | |
56bec294 | 1833 | } |
31840ae1 | 1834 | |
5d4f98a2 YZ |
1835 | BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod); |
1836 | refs += refs_to_mod; | |
1837 | ||
1838 | if (refs > 0) { | |
1839 | if (type == BTRFS_EXTENT_DATA_REF_KEY) | |
1840 | btrfs_set_extent_data_ref_count(leaf, dref, refs); | |
1841 | else | |
1842 | btrfs_set_shared_data_ref_count(leaf, sref, refs); | |
1843 | } else { | |
fcebe456 | 1844 | *last_ref = 1; |
5d4f98a2 YZ |
1845 | size = btrfs_extent_inline_ref_size(type); |
1846 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1847 | ptr = (unsigned long)iref; | |
1848 | end = (unsigned long)ei + item_size; | |
1849 | if (ptr + size < end) | |
1850 | memmove_extent_buffer(leaf, ptr, ptr + size, | |
1851 | end - ptr - size); | |
1852 | item_size -= size; | |
78ac4f9e | 1853 | btrfs_truncate_item(path, item_size, 1); |
5d4f98a2 YZ |
1854 | } |
1855 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1856 | } |
1857 | ||
1858 | static noinline_for_stack | |
1859 | int insert_inline_extent_backref(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
1860 | struct btrfs_path *path, |
1861 | u64 bytenr, u64 num_bytes, u64 parent, | |
1862 | u64 root_objectid, u64 owner, | |
1863 | u64 offset, int refs_to_add, | |
1864 | struct btrfs_delayed_extent_op *extent_op) | |
1865 | { | |
1866 | struct btrfs_extent_inline_ref *iref; | |
1867 | int ret; | |
1868 | ||
867cc1fb NB |
1869 | ret = lookup_inline_extent_backref(trans, path, &iref, bytenr, |
1870 | num_bytes, parent, root_objectid, | |
1871 | owner, offset, 1); | |
5d4f98a2 YZ |
1872 | if (ret == 0) { |
1873 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID); | |
61a18f1c NB |
1874 | update_inline_extent_backref(path, iref, refs_to_add, |
1875 | extent_op, NULL); | |
5d4f98a2 | 1876 | } else if (ret == -ENOENT) { |
a639cdeb | 1877 | setup_inline_extent_backref(trans->fs_info, path, iref, parent, |
143bede5 JM |
1878 | root_objectid, owner, offset, |
1879 | refs_to_add, extent_op); | |
1880 | ret = 0; | |
771ed689 | 1881 | } |
5d4f98a2 YZ |
1882 | return ret; |
1883 | } | |
31840ae1 | 1884 | |
5d4f98a2 | 1885 | static int insert_extent_backref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1886 | struct btrfs_path *path, |
1887 | u64 bytenr, u64 parent, u64 root_objectid, | |
1888 | u64 owner, u64 offset, int refs_to_add) | |
1889 | { | |
1890 | int ret; | |
1891 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1892 | BUG_ON(refs_to_add != 1); | |
10728404 NB |
1893 | ret = insert_tree_block_ref(trans, path, bytenr, parent, |
1894 | root_objectid); | |
5d4f98a2 | 1895 | } else { |
62b895af NB |
1896 | ret = insert_extent_data_ref(trans, path, bytenr, parent, |
1897 | root_objectid, owner, offset, | |
1898 | refs_to_add); | |
5d4f98a2 YZ |
1899 | } |
1900 | return ret; | |
1901 | } | |
56bec294 | 1902 | |
5d4f98a2 | 1903 | static int remove_extent_backref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1904 | struct btrfs_path *path, |
1905 | struct btrfs_extent_inline_ref *iref, | |
fcebe456 | 1906 | int refs_to_drop, int is_data, int *last_ref) |
5d4f98a2 | 1907 | { |
143bede5 | 1908 | int ret = 0; |
b9473439 | 1909 | |
5d4f98a2 YZ |
1910 | BUG_ON(!is_data && refs_to_drop != 1); |
1911 | if (iref) { | |
61a18f1c NB |
1912 | update_inline_extent_backref(path, iref, -refs_to_drop, NULL, |
1913 | last_ref); | |
5d4f98a2 | 1914 | } else if (is_data) { |
e9f6290d | 1915 | ret = remove_extent_data_ref(trans, path, refs_to_drop, |
fcebe456 | 1916 | last_ref); |
5d4f98a2 | 1917 | } else { |
fcebe456 | 1918 | *last_ref = 1; |
87cc7a8a | 1919 | ret = btrfs_del_item(trans, trans->fs_info->extent_root, path); |
5d4f98a2 YZ |
1920 | } |
1921 | return ret; | |
1922 | } | |
1923 | ||
d04c6b88 JM |
1924 | static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len, |
1925 | u64 *discarded_bytes) | |
5d4f98a2 | 1926 | { |
86557861 JM |
1927 | int j, ret = 0; |
1928 | u64 bytes_left, end; | |
4d89d377 | 1929 | u64 aligned_start = ALIGN(start, 1 << 9); |
d04c6b88 | 1930 | |
4d89d377 JM |
1931 | if (WARN_ON(start != aligned_start)) { |
1932 | len -= aligned_start - start; | |
1933 | len = round_down(len, 1 << 9); | |
1934 | start = aligned_start; | |
1935 | } | |
d04c6b88 | 1936 | |
4d89d377 | 1937 | *discarded_bytes = 0; |
86557861 JM |
1938 | |
1939 | if (!len) | |
1940 | return 0; | |
1941 | ||
1942 | end = start + len; | |
1943 | bytes_left = len; | |
1944 | ||
1945 | /* Skip any superblocks on this device. */ | |
1946 | for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) { | |
1947 | u64 sb_start = btrfs_sb_offset(j); | |
1948 | u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE; | |
1949 | u64 size = sb_start - start; | |
1950 | ||
1951 | if (!in_range(sb_start, start, bytes_left) && | |
1952 | !in_range(sb_end, start, bytes_left) && | |
1953 | !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE)) | |
1954 | continue; | |
1955 | ||
1956 | /* | |
1957 | * Superblock spans beginning of range. Adjust start and | |
1958 | * try again. | |
1959 | */ | |
1960 | if (sb_start <= start) { | |
1961 | start += sb_end - start; | |
1962 | if (start > end) { | |
1963 | bytes_left = 0; | |
1964 | break; | |
1965 | } | |
1966 | bytes_left = end - start; | |
1967 | continue; | |
1968 | } | |
1969 | ||
1970 | if (size) { | |
1971 | ret = blkdev_issue_discard(bdev, start >> 9, size >> 9, | |
1972 | GFP_NOFS, 0); | |
1973 | if (!ret) | |
1974 | *discarded_bytes += size; | |
1975 | else if (ret != -EOPNOTSUPP) | |
1976 | return ret; | |
1977 | } | |
1978 | ||
1979 | start = sb_end; | |
1980 | if (start > end) { | |
1981 | bytes_left = 0; | |
1982 | break; | |
1983 | } | |
1984 | bytes_left = end - start; | |
1985 | } | |
1986 | ||
1987 | if (bytes_left) { | |
1988 | ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9, | |
4d89d377 JM |
1989 | GFP_NOFS, 0); |
1990 | if (!ret) | |
86557861 | 1991 | *discarded_bytes += bytes_left; |
4d89d377 | 1992 | } |
d04c6b88 | 1993 | return ret; |
5d4f98a2 | 1994 | } |
5d4f98a2 | 1995 | |
2ff7e61e | 1996 | int btrfs_discard_extent(struct btrfs_fs_info *fs_info, u64 bytenr, |
1edb647b | 1997 | u64 num_bytes, u64 *actual_bytes) |
5d4f98a2 | 1998 | { |
5d4f98a2 | 1999 | int ret; |
5378e607 | 2000 | u64 discarded_bytes = 0; |
a1d3c478 | 2001 | struct btrfs_bio *bbio = NULL; |
5d4f98a2 | 2002 | |
e244a0ae | 2003 | |
2999241d FM |
2004 | /* |
2005 | * Avoid races with device replace and make sure our bbio has devices | |
2006 | * associated to its stripes that don't go away while we are discarding. | |
2007 | */ | |
0b246afa | 2008 | btrfs_bio_counter_inc_blocked(fs_info); |
5d4f98a2 | 2009 | /* Tell the block device(s) that the sectors can be discarded */ |
0b246afa JM |
2010 | ret = btrfs_map_block(fs_info, BTRFS_MAP_DISCARD, bytenr, &num_bytes, |
2011 | &bbio, 0); | |
79787eaa | 2012 | /* Error condition is -ENOMEM */ |
5d4f98a2 | 2013 | if (!ret) { |
a1d3c478 | 2014 | struct btrfs_bio_stripe *stripe = bbio->stripes; |
5d4f98a2 YZ |
2015 | int i; |
2016 | ||
5d4f98a2 | 2017 | |
a1d3c478 | 2018 | for (i = 0; i < bbio->num_stripes; i++, stripe++) { |
d04c6b88 | 2019 | u64 bytes; |
38b5f68e AJ |
2020 | struct request_queue *req_q; |
2021 | ||
627e0873 FM |
2022 | if (!stripe->dev->bdev) { |
2023 | ASSERT(btrfs_test_opt(fs_info, DEGRADED)); | |
2024 | continue; | |
2025 | } | |
38b5f68e AJ |
2026 | req_q = bdev_get_queue(stripe->dev->bdev); |
2027 | if (!blk_queue_discard(req_q)) | |
d5e2003c JB |
2028 | continue; |
2029 | ||
5378e607 LD |
2030 | ret = btrfs_issue_discard(stripe->dev->bdev, |
2031 | stripe->physical, | |
d04c6b88 JM |
2032 | stripe->length, |
2033 | &bytes); | |
5378e607 | 2034 | if (!ret) |
d04c6b88 | 2035 | discarded_bytes += bytes; |
5378e607 | 2036 | else if (ret != -EOPNOTSUPP) |
79787eaa | 2037 | break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */ |
d5e2003c JB |
2038 | |
2039 | /* | |
2040 | * Just in case we get back EOPNOTSUPP for some reason, | |
2041 | * just ignore the return value so we don't screw up | |
2042 | * people calling discard_extent. | |
2043 | */ | |
2044 | ret = 0; | |
5d4f98a2 | 2045 | } |
6e9606d2 | 2046 | btrfs_put_bbio(bbio); |
5d4f98a2 | 2047 | } |
0b246afa | 2048 | btrfs_bio_counter_dec(fs_info); |
5378e607 LD |
2049 | |
2050 | if (actual_bytes) | |
2051 | *actual_bytes = discarded_bytes; | |
2052 | ||
5d4f98a2 | 2053 | |
53b381b3 DW |
2054 | if (ret == -EOPNOTSUPP) |
2055 | ret = 0; | |
5d4f98a2 | 2056 | return ret; |
5d4f98a2 YZ |
2057 | } |
2058 | ||
79787eaa | 2059 | /* Can return -ENOMEM */ |
5d4f98a2 | 2060 | int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
82fa113f | 2061 | struct btrfs_ref *generic_ref) |
5d4f98a2 | 2062 | { |
82fa113f | 2063 | struct btrfs_fs_info *fs_info = trans->fs_info; |
d7eae340 | 2064 | int old_ref_mod, new_ref_mod; |
5d4f98a2 | 2065 | int ret; |
66d7e7f0 | 2066 | |
82fa113f QW |
2067 | ASSERT(generic_ref->type != BTRFS_REF_NOT_SET && |
2068 | generic_ref->action); | |
2069 | BUG_ON(generic_ref->type == BTRFS_REF_METADATA && | |
2070 | generic_ref->tree_ref.root == BTRFS_TREE_LOG_OBJECTID); | |
5d4f98a2 | 2071 | |
82fa113f QW |
2072 | if (generic_ref->type == BTRFS_REF_METADATA) |
2073 | ret = btrfs_add_delayed_tree_ref(trans, generic_ref, | |
ed4f255b | 2074 | NULL, &old_ref_mod, &new_ref_mod); |
82fa113f QW |
2075 | else |
2076 | ret = btrfs_add_delayed_data_ref(trans, generic_ref, 0, | |
d7eae340 | 2077 | &old_ref_mod, &new_ref_mod); |
d7eae340 | 2078 | |
82fa113f | 2079 | btrfs_ref_tree_mod(fs_info, generic_ref); |
8a5040f7 | 2080 | |
ddf30cf0 | 2081 | if (ret == 0 && old_ref_mod < 0 && new_ref_mod >= 0) |
78192442 | 2082 | sub_pinned_bytes(fs_info, generic_ref); |
d7eae340 | 2083 | |
5d4f98a2 YZ |
2084 | return ret; |
2085 | } | |
2086 | ||
bd3c685e NB |
2087 | /* |
2088 | * __btrfs_inc_extent_ref - insert backreference for a given extent | |
2089 | * | |
2090 | * @trans: Handle of transaction | |
2091 | * | |
2092 | * @node: The delayed ref node used to get the bytenr/length for | |
2093 | * extent whose references are incremented. | |
2094 | * | |
2095 | * @parent: If this is a shared extent (BTRFS_SHARED_DATA_REF_KEY/ | |
2096 | * BTRFS_SHARED_BLOCK_REF_KEY) then it holds the logical | |
2097 | * bytenr of the parent block. Since new extents are always | |
2098 | * created with indirect references, this will only be the case | |
2099 | * when relocating a shared extent. In that case, root_objectid | |
2100 | * will be BTRFS_TREE_RELOC_OBJECTID. Otheriwse, parent must | |
2101 | * be 0 | |
2102 | * | |
2103 | * @root_objectid: The id of the root where this modification has originated, | |
2104 | * this can be either one of the well-known metadata trees or | |
2105 | * the subvolume id which references this extent. | |
2106 | * | |
2107 | * @owner: For data extents it is the inode number of the owning file. | |
2108 | * For metadata extents this parameter holds the level in the | |
2109 | * tree of the extent. | |
2110 | * | |
2111 | * @offset: For metadata extents the offset is ignored and is currently | |
2112 | * always passed as 0. For data extents it is the fileoffset | |
2113 | * this extent belongs to. | |
2114 | * | |
2115 | * @refs_to_add Number of references to add | |
2116 | * | |
2117 | * @extent_op Pointer to a structure, holding information necessary when | |
2118 | * updating a tree block's flags | |
2119 | * | |
2120 | */ | |
5d4f98a2 | 2121 | static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
c682f9b3 | 2122 | struct btrfs_delayed_ref_node *node, |
5d4f98a2 YZ |
2123 | u64 parent, u64 root_objectid, |
2124 | u64 owner, u64 offset, int refs_to_add, | |
2125 | struct btrfs_delayed_extent_op *extent_op) | |
2126 | { | |
2127 | struct btrfs_path *path; | |
2128 | struct extent_buffer *leaf; | |
2129 | struct btrfs_extent_item *item; | |
fcebe456 | 2130 | struct btrfs_key key; |
c682f9b3 QW |
2131 | u64 bytenr = node->bytenr; |
2132 | u64 num_bytes = node->num_bytes; | |
5d4f98a2 YZ |
2133 | u64 refs; |
2134 | int ret; | |
5d4f98a2 YZ |
2135 | |
2136 | path = btrfs_alloc_path(); | |
2137 | if (!path) | |
2138 | return -ENOMEM; | |
2139 | ||
e4058b54 | 2140 | path->reada = READA_FORWARD; |
5d4f98a2 YZ |
2141 | path->leave_spinning = 1; |
2142 | /* this will setup the path even if it fails to insert the back ref */ | |
a639cdeb NB |
2143 | ret = insert_inline_extent_backref(trans, path, bytenr, num_bytes, |
2144 | parent, root_objectid, owner, | |
2145 | offset, refs_to_add, extent_op); | |
0ed4792a | 2146 | if ((ret < 0 && ret != -EAGAIN) || !ret) |
5d4f98a2 | 2147 | goto out; |
fcebe456 JB |
2148 | |
2149 | /* | |
2150 | * Ok we had -EAGAIN which means we didn't have space to insert and | |
2151 | * inline extent ref, so just update the reference count and add a | |
2152 | * normal backref. | |
2153 | */ | |
5d4f98a2 | 2154 | leaf = path->nodes[0]; |
fcebe456 | 2155 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
5d4f98a2 YZ |
2156 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
2157 | refs = btrfs_extent_refs(leaf, item); | |
2158 | btrfs_set_extent_refs(leaf, item, refs + refs_to_add); | |
2159 | if (extent_op) | |
2160 | __run_delayed_extent_op(extent_op, leaf, item); | |
56bec294 | 2161 | |
5d4f98a2 | 2162 | btrfs_mark_buffer_dirty(leaf); |
b3b4aa74 | 2163 | btrfs_release_path(path); |
56bec294 | 2164 | |
e4058b54 | 2165 | path->reada = READA_FORWARD; |
b9473439 | 2166 | path->leave_spinning = 1; |
56bec294 | 2167 | /* now insert the actual backref */ |
37593410 NB |
2168 | ret = insert_extent_backref(trans, path, bytenr, parent, root_objectid, |
2169 | owner, offset, refs_to_add); | |
79787eaa | 2170 | if (ret) |
66642832 | 2171 | btrfs_abort_transaction(trans, ret); |
5d4f98a2 | 2172 | out: |
56bec294 | 2173 | btrfs_free_path(path); |
30d133fc | 2174 | return ret; |
56bec294 CM |
2175 | } |
2176 | ||
5d4f98a2 | 2177 | static int run_delayed_data_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
2178 | struct btrfs_delayed_ref_node *node, |
2179 | struct btrfs_delayed_extent_op *extent_op, | |
2180 | int insert_reserved) | |
56bec294 | 2181 | { |
5d4f98a2 YZ |
2182 | int ret = 0; |
2183 | struct btrfs_delayed_data_ref *ref; | |
2184 | struct btrfs_key ins; | |
2185 | u64 parent = 0; | |
2186 | u64 ref_root = 0; | |
2187 | u64 flags = 0; | |
2188 | ||
2189 | ins.objectid = node->bytenr; | |
2190 | ins.offset = node->num_bytes; | |
2191 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
2192 | ||
2193 | ref = btrfs_delayed_node_to_data_ref(node); | |
2bf98ef3 | 2194 | trace_run_delayed_data_ref(trans->fs_info, node, ref, node->action); |
599c75ec | 2195 | |
5d4f98a2 YZ |
2196 | if (node->type == BTRFS_SHARED_DATA_REF_KEY) |
2197 | parent = ref->parent; | |
fcebe456 | 2198 | ref_root = ref->root; |
5d4f98a2 YZ |
2199 | |
2200 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
3173a18f | 2201 | if (extent_op) |
5d4f98a2 | 2202 | flags |= extent_op->flags_to_set; |
ef89b824 NB |
2203 | ret = alloc_reserved_file_extent(trans, parent, ref_root, |
2204 | flags, ref->objectid, | |
2205 | ref->offset, &ins, | |
2206 | node->ref_mod); | |
5d4f98a2 | 2207 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2590d0f1 NB |
2208 | ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root, |
2209 | ref->objectid, ref->offset, | |
2210 | node->ref_mod, extent_op); | |
5d4f98a2 | 2211 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
e72cb923 | 2212 | ret = __btrfs_free_extent(trans, node, parent, |
5d4f98a2 YZ |
2213 | ref_root, ref->objectid, |
2214 | ref->offset, node->ref_mod, | |
c682f9b3 | 2215 | extent_op); |
5d4f98a2 YZ |
2216 | } else { |
2217 | BUG(); | |
2218 | } | |
2219 | return ret; | |
2220 | } | |
2221 | ||
2222 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
2223 | struct extent_buffer *leaf, | |
2224 | struct btrfs_extent_item *ei) | |
2225 | { | |
2226 | u64 flags = btrfs_extent_flags(leaf, ei); | |
2227 | if (extent_op->update_flags) { | |
2228 | flags |= extent_op->flags_to_set; | |
2229 | btrfs_set_extent_flags(leaf, ei, flags); | |
2230 | } | |
2231 | ||
2232 | if (extent_op->update_key) { | |
2233 | struct btrfs_tree_block_info *bi; | |
2234 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)); | |
2235 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
2236 | btrfs_set_tree_block_key(leaf, bi, &extent_op->key); | |
2237 | } | |
2238 | } | |
2239 | ||
2240 | static int run_delayed_extent_op(struct btrfs_trans_handle *trans, | |
d278850e | 2241 | struct btrfs_delayed_ref_head *head, |
5d4f98a2 YZ |
2242 | struct btrfs_delayed_extent_op *extent_op) |
2243 | { | |
20b9a2d6 | 2244 | struct btrfs_fs_info *fs_info = trans->fs_info; |
5d4f98a2 YZ |
2245 | struct btrfs_key key; |
2246 | struct btrfs_path *path; | |
2247 | struct btrfs_extent_item *ei; | |
2248 | struct extent_buffer *leaf; | |
2249 | u32 item_size; | |
56bec294 | 2250 | int ret; |
5d4f98a2 | 2251 | int err = 0; |
b1c79e09 | 2252 | int metadata = !extent_op->is_data; |
5d4f98a2 | 2253 | |
79787eaa JM |
2254 | if (trans->aborted) |
2255 | return 0; | |
2256 | ||
0b246afa | 2257 | if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) |
3173a18f JB |
2258 | metadata = 0; |
2259 | ||
5d4f98a2 YZ |
2260 | path = btrfs_alloc_path(); |
2261 | if (!path) | |
2262 | return -ENOMEM; | |
2263 | ||
d278850e | 2264 | key.objectid = head->bytenr; |
5d4f98a2 | 2265 | |
3173a18f | 2266 | if (metadata) { |
3173a18f | 2267 | key.type = BTRFS_METADATA_ITEM_KEY; |
b1c79e09 | 2268 | key.offset = extent_op->level; |
3173a18f JB |
2269 | } else { |
2270 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
d278850e | 2271 | key.offset = head->num_bytes; |
3173a18f JB |
2272 | } |
2273 | ||
2274 | again: | |
e4058b54 | 2275 | path->reada = READA_FORWARD; |
5d4f98a2 | 2276 | path->leave_spinning = 1; |
0b246afa | 2277 | ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 1); |
5d4f98a2 YZ |
2278 | if (ret < 0) { |
2279 | err = ret; | |
2280 | goto out; | |
2281 | } | |
2282 | if (ret > 0) { | |
3173a18f | 2283 | if (metadata) { |
55994887 FDBM |
2284 | if (path->slots[0] > 0) { |
2285 | path->slots[0]--; | |
2286 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
2287 | path->slots[0]); | |
d278850e | 2288 | if (key.objectid == head->bytenr && |
55994887 | 2289 | key.type == BTRFS_EXTENT_ITEM_KEY && |
d278850e | 2290 | key.offset == head->num_bytes) |
55994887 FDBM |
2291 | ret = 0; |
2292 | } | |
2293 | if (ret > 0) { | |
2294 | btrfs_release_path(path); | |
2295 | metadata = 0; | |
3173a18f | 2296 | |
d278850e JB |
2297 | key.objectid = head->bytenr; |
2298 | key.offset = head->num_bytes; | |
55994887 FDBM |
2299 | key.type = BTRFS_EXTENT_ITEM_KEY; |
2300 | goto again; | |
2301 | } | |
2302 | } else { | |
2303 | err = -EIO; | |
2304 | goto out; | |
3173a18f | 2305 | } |
5d4f98a2 YZ |
2306 | } |
2307 | ||
2308 | leaf = path->nodes[0]; | |
2309 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
ba3c2b19 | 2310 | |
6d8ff4e4 | 2311 | if (unlikely(item_size < sizeof(*ei))) { |
ba3c2b19 NB |
2312 | err = -EINVAL; |
2313 | btrfs_print_v0_err(fs_info); | |
2314 | btrfs_abort_transaction(trans, err); | |
2315 | goto out; | |
2316 | } | |
2317 | ||
5d4f98a2 YZ |
2318 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
2319 | __run_delayed_extent_op(extent_op, leaf, ei); | |
56bec294 | 2320 | |
5d4f98a2 YZ |
2321 | btrfs_mark_buffer_dirty(leaf); |
2322 | out: | |
2323 | btrfs_free_path(path); | |
2324 | return err; | |
56bec294 CM |
2325 | } |
2326 | ||
5d4f98a2 | 2327 | static int run_delayed_tree_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
2328 | struct btrfs_delayed_ref_node *node, |
2329 | struct btrfs_delayed_extent_op *extent_op, | |
2330 | int insert_reserved) | |
56bec294 CM |
2331 | { |
2332 | int ret = 0; | |
5d4f98a2 | 2333 | struct btrfs_delayed_tree_ref *ref; |
5d4f98a2 YZ |
2334 | u64 parent = 0; |
2335 | u64 ref_root = 0; | |
56bec294 | 2336 | |
5d4f98a2 | 2337 | ref = btrfs_delayed_node_to_tree_ref(node); |
f97806f2 | 2338 | trace_run_delayed_tree_ref(trans->fs_info, node, ref, node->action); |
599c75ec | 2339 | |
5d4f98a2 YZ |
2340 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) |
2341 | parent = ref->parent; | |
fcebe456 | 2342 | ref_root = ref->root; |
5d4f98a2 | 2343 | |
02794222 | 2344 | if (node->ref_mod != 1) { |
f97806f2 | 2345 | btrfs_err(trans->fs_info, |
02794222 LB |
2346 | "btree block(%llu) has %d references rather than 1: action %d ref_root %llu parent %llu", |
2347 | node->bytenr, node->ref_mod, node->action, ref_root, | |
2348 | parent); | |
2349 | return -EIO; | |
2350 | } | |
5d4f98a2 | 2351 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { |
3173a18f | 2352 | BUG_ON(!extent_op || !extent_op->update_flags); |
21ebfbe7 | 2353 | ret = alloc_reserved_tree_block(trans, node, extent_op); |
5d4f98a2 | 2354 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2590d0f1 NB |
2355 | ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root, |
2356 | ref->level, 0, 1, extent_op); | |
5d4f98a2 | 2357 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
e72cb923 | 2358 | ret = __btrfs_free_extent(trans, node, parent, ref_root, |
c682f9b3 | 2359 | ref->level, 0, 1, extent_op); |
5d4f98a2 YZ |
2360 | } else { |
2361 | BUG(); | |
2362 | } | |
56bec294 CM |
2363 | return ret; |
2364 | } | |
2365 | ||
2366 | /* helper function to actually process a single delayed ref entry */ | |
5d4f98a2 | 2367 | static int run_one_delayed_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
2368 | struct btrfs_delayed_ref_node *node, |
2369 | struct btrfs_delayed_extent_op *extent_op, | |
2370 | int insert_reserved) | |
56bec294 | 2371 | { |
79787eaa JM |
2372 | int ret = 0; |
2373 | ||
857cc2fc JB |
2374 | if (trans->aborted) { |
2375 | if (insert_reserved) | |
5fac7f9e | 2376 | btrfs_pin_extent(trans->fs_info, node->bytenr, |
857cc2fc | 2377 | node->num_bytes, 1); |
79787eaa | 2378 | return 0; |
857cc2fc | 2379 | } |
79787eaa | 2380 | |
5d4f98a2 YZ |
2381 | if (node->type == BTRFS_TREE_BLOCK_REF_KEY || |
2382 | node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
f97806f2 | 2383 | ret = run_delayed_tree_ref(trans, node, extent_op, |
5d4f98a2 YZ |
2384 | insert_reserved); |
2385 | else if (node->type == BTRFS_EXTENT_DATA_REF_KEY || | |
2386 | node->type == BTRFS_SHARED_DATA_REF_KEY) | |
2bf98ef3 | 2387 | ret = run_delayed_data_ref(trans, node, extent_op, |
5d4f98a2 YZ |
2388 | insert_reserved); |
2389 | else | |
2390 | BUG(); | |
80ee54bf JB |
2391 | if (ret && insert_reserved) |
2392 | btrfs_pin_extent(trans->fs_info, node->bytenr, | |
2393 | node->num_bytes, 1); | |
5d4f98a2 | 2394 | return ret; |
56bec294 CM |
2395 | } |
2396 | ||
c6fc2454 | 2397 | static inline struct btrfs_delayed_ref_node * |
56bec294 CM |
2398 | select_delayed_ref(struct btrfs_delayed_ref_head *head) |
2399 | { | |
cffc3374 FM |
2400 | struct btrfs_delayed_ref_node *ref; |
2401 | ||
e3d03965 | 2402 | if (RB_EMPTY_ROOT(&head->ref_tree.rb_root)) |
c6fc2454 | 2403 | return NULL; |
d7df2c79 | 2404 | |
cffc3374 FM |
2405 | /* |
2406 | * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first. | |
2407 | * This is to prevent a ref count from going down to zero, which deletes | |
2408 | * the extent item from the extent tree, when there still are references | |
2409 | * to add, which would fail because they would not find the extent item. | |
2410 | */ | |
1d57ee94 WX |
2411 | if (!list_empty(&head->ref_add_list)) |
2412 | return list_first_entry(&head->ref_add_list, | |
2413 | struct btrfs_delayed_ref_node, add_list); | |
2414 | ||
e3d03965 | 2415 | ref = rb_entry(rb_first_cached(&head->ref_tree), |
0e0adbcf | 2416 | struct btrfs_delayed_ref_node, ref_node); |
1d57ee94 WX |
2417 | ASSERT(list_empty(&ref->add_list)); |
2418 | return ref; | |
56bec294 CM |
2419 | } |
2420 | ||
2eadaa22 JB |
2421 | static void unselect_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs, |
2422 | struct btrfs_delayed_ref_head *head) | |
2423 | { | |
2424 | spin_lock(&delayed_refs->lock); | |
2425 | head->processing = 0; | |
2426 | delayed_refs->num_heads_ready++; | |
2427 | spin_unlock(&delayed_refs->lock); | |
2428 | btrfs_delayed_ref_unlock(head); | |
2429 | } | |
2430 | ||
bedc6617 JB |
2431 | static struct btrfs_delayed_extent_op *cleanup_extent_op( |
2432 | struct btrfs_delayed_ref_head *head) | |
b00e6250 JB |
2433 | { |
2434 | struct btrfs_delayed_extent_op *extent_op = head->extent_op; | |
b00e6250 JB |
2435 | |
2436 | if (!extent_op) | |
bedc6617 JB |
2437 | return NULL; |
2438 | ||
b00e6250 | 2439 | if (head->must_insert_reserved) { |
bedc6617 | 2440 | head->extent_op = NULL; |
b00e6250 | 2441 | btrfs_free_delayed_extent_op(extent_op); |
bedc6617 | 2442 | return NULL; |
b00e6250 | 2443 | } |
bedc6617 JB |
2444 | return extent_op; |
2445 | } | |
2446 | ||
2447 | static int run_and_cleanup_extent_op(struct btrfs_trans_handle *trans, | |
2448 | struct btrfs_delayed_ref_head *head) | |
2449 | { | |
2450 | struct btrfs_delayed_extent_op *extent_op; | |
2451 | int ret; | |
2452 | ||
2453 | extent_op = cleanup_extent_op(head); | |
2454 | if (!extent_op) | |
2455 | return 0; | |
2456 | head->extent_op = NULL; | |
b00e6250 | 2457 | spin_unlock(&head->lock); |
20b9a2d6 | 2458 | ret = run_delayed_extent_op(trans, head, extent_op); |
b00e6250 JB |
2459 | btrfs_free_delayed_extent_op(extent_op); |
2460 | return ret ? ret : 1; | |
2461 | } | |
2462 | ||
31890da0 JB |
2463 | void btrfs_cleanup_ref_head_accounting(struct btrfs_fs_info *fs_info, |
2464 | struct btrfs_delayed_ref_root *delayed_refs, | |
2465 | struct btrfs_delayed_ref_head *head) | |
07c47775 | 2466 | { |
ba2c4d4e | 2467 | int nr_items = 1; /* Dropping this ref head update. */ |
07c47775 JB |
2468 | |
2469 | if (head->total_ref_mod < 0) { | |
2470 | struct btrfs_space_info *space_info; | |
2471 | u64 flags; | |
2472 | ||
2473 | if (head->is_data) | |
2474 | flags = BTRFS_BLOCK_GROUP_DATA; | |
2475 | else if (head->is_system) | |
2476 | flags = BTRFS_BLOCK_GROUP_SYSTEM; | |
2477 | else | |
2478 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
2479 | space_info = __find_space_info(fs_info, flags); | |
2480 | ASSERT(space_info); | |
2481 | percpu_counter_add_batch(&space_info->total_bytes_pinned, | |
2482 | -head->num_bytes, | |
2483 | BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
2484 | ||
ba2c4d4e JB |
2485 | /* |
2486 | * We had csum deletions accounted for in our delayed refs rsv, | |
2487 | * we need to drop the csum leaves for this update from our | |
2488 | * delayed_refs_rsv. | |
2489 | */ | |
07c47775 JB |
2490 | if (head->is_data) { |
2491 | spin_lock(&delayed_refs->lock); | |
2492 | delayed_refs->pending_csums -= head->num_bytes; | |
2493 | spin_unlock(&delayed_refs->lock); | |
ba2c4d4e JB |
2494 | nr_items += btrfs_csum_bytes_to_leaves(fs_info, |
2495 | head->num_bytes); | |
07c47775 JB |
2496 | } |
2497 | } | |
2498 | ||
ba2c4d4e | 2499 | btrfs_delayed_refs_rsv_release(fs_info, nr_items); |
07c47775 JB |
2500 | } |
2501 | ||
194ab0bc | 2502 | static int cleanup_ref_head(struct btrfs_trans_handle *trans, |
194ab0bc JB |
2503 | struct btrfs_delayed_ref_head *head) |
2504 | { | |
f9871edd NB |
2505 | |
2506 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
194ab0bc JB |
2507 | struct btrfs_delayed_ref_root *delayed_refs; |
2508 | int ret; | |
2509 | ||
2510 | delayed_refs = &trans->transaction->delayed_refs; | |
2511 | ||
bedc6617 | 2512 | ret = run_and_cleanup_extent_op(trans, head); |
194ab0bc JB |
2513 | if (ret < 0) { |
2514 | unselect_delayed_ref_head(delayed_refs, head); | |
2515 | btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret); | |
2516 | return ret; | |
2517 | } else if (ret) { | |
2518 | return ret; | |
2519 | } | |
2520 | ||
2521 | /* | |
2522 | * Need to drop our head ref lock and re-acquire the delayed ref lock | |
2523 | * and then re-check to make sure nobody got added. | |
2524 | */ | |
2525 | spin_unlock(&head->lock); | |
2526 | spin_lock(&delayed_refs->lock); | |
2527 | spin_lock(&head->lock); | |
e3d03965 | 2528 | if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root) || head->extent_op) { |
194ab0bc JB |
2529 | spin_unlock(&head->lock); |
2530 | spin_unlock(&delayed_refs->lock); | |
2531 | return 1; | |
2532 | } | |
d7baffda | 2533 | btrfs_delete_ref_head(delayed_refs, head); |
c1103f7a | 2534 | spin_unlock(&head->lock); |
1e7a1421 | 2535 | spin_unlock(&delayed_refs->lock); |
c1103f7a | 2536 | |
c1103f7a | 2537 | if (head->must_insert_reserved) { |
d278850e JB |
2538 | btrfs_pin_extent(fs_info, head->bytenr, |
2539 | head->num_bytes, 1); | |
c1103f7a | 2540 | if (head->is_data) { |
d278850e JB |
2541 | ret = btrfs_del_csums(trans, fs_info, head->bytenr, |
2542 | head->num_bytes); | |
c1103f7a JB |
2543 | } |
2544 | } | |
2545 | ||
31890da0 | 2546 | btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head); |
07c47775 JB |
2547 | |
2548 | trace_run_delayed_ref_head(fs_info, head, 0); | |
c1103f7a | 2549 | btrfs_delayed_ref_unlock(head); |
d278850e | 2550 | btrfs_put_delayed_ref_head(head); |
194ab0bc JB |
2551 | return 0; |
2552 | } | |
2553 | ||
b1cdbcb5 NB |
2554 | static struct btrfs_delayed_ref_head *btrfs_obtain_ref_head( |
2555 | struct btrfs_trans_handle *trans) | |
2556 | { | |
2557 | struct btrfs_delayed_ref_root *delayed_refs = | |
2558 | &trans->transaction->delayed_refs; | |
2559 | struct btrfs_delayed_ref_head *head = NULL; | |
2560 | int ret; | |
2561 | ||
2562 | spin_lock(&delayed_refs->lock); | |
5637c74b | 2563 | head = btrfs_select_ref_head(delayed_refs); |
b1cdbcb5 NB |
2564 | if (!head) { |
2565 | spin_unlock(&delayed_refs->lock); | |
2566 | return head; | |
2567 | } | |
2568 | ||
2569 | /* | |
2570 | * Grab the lock that says we are going to process all the refs for | |
2571 | * this head | |
2572 | */ | |
9e920a6f | 2573 | ret = btrfs_delayed_ref_lock(delayed_refs, head); |
b1cdbcb5 NB |
2574 | spin_unlock(&delayed_refs->lock); |
2575 | ||
2576 | /* | |
2577 | * We may have dropped the spin lock to get the head mutex lock, and | |
2578 | * that might have given someone else time to free the head. If that's | |
2579 | * true, it has been removed from our list and we can move on. | |
2580 | */ | |
2581 | if (ret == -EAGAIN) | |
2582 | head = ERR_PTR(-EAGAIN); | |
2583 | ||
2584 | return head; | |
2585 | } | |
2586 | ||
e7261386 NB |
2587 | static int btrfs_run_delayed_refs_for_head(struct btrfs_trans_handle *trans, |
2588 | struct btrfs_delayed_ref_head *locked_ref, | |
2589 | unsigned long *run_refs) | |
2590 | { | |
2591 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
2592 | struct btrfs_delayed_ref_root *delayed_refs; | |
2593 | struct btrfs_delayed_extent_op *extent_op; | |
2594 | struct btrfs_delayed_ref_node *ref; | |
2595 | int must_insert_reserved = 0; | |
2596 | int ret; | |
2597 | ||
2598 | delayed_refs = &trans->transaction->delayed_refs; | |
2599 | ||
0110a4c4 NB |
2600 | lockdep_assert_held(&locked_ref->mutex); |
2601 | lockdep_assert_held(&locked_ref->lock); | |
2602 | ||
e7261386 NB |
2603 | while ((ref = select_delayed_ref(locked_ref))) { |
2604 | if (ref->seq && | |
2605 | btrfs_check_delayed_seq(fs_info, ref->seq)) { | |
2606 | spin_unlock(&locked_ref->lock); | |
2607 | unselect_delayed_ref_head(delayed_refs, locked_ref); | |
2608 | return -EAGAIN; | |
2609 | } | |
2610 | ||
2611 | (*run_refs)++; | |
2612 | ref->in_tree = 0; | |
2613 | rb_erase_cached(&ref->ref_node, &locked_ref->ref_tree); | |
2614 | RB_CLEAR_NODE(&ref->ref_node); | |
2615 | if (!list_empty(&ref->add_list)) | |
2616 | list_del(&ref->add_list); | |
2617 | /* | |
2618 | * When we play the delayed ref, also correct the ref_mod on | |
2619 | * head | |
2620 | */ | |
2621 | switch (ref->action) { | |
2622 | case BTRFS_ADD_DELAYED_REF: | |
2623 | case BTRFS_ADD_DELAYED_EXTENT: | |
2624 | locked_ref->ref_mod -= ref->ref_mod; | |
2625 | break; | |
2626 | case BTRFS_DROP_DELAYED_REF: | |
2627 | locked_ref->ref_mod += ref->ref_mod; | |
2628 | break; | |
2629 | default: | |
2630 | WARN_ON(1); | |
2631 | } | |
2632 | atomic_dec(&delayed_refs->num_entries); | |
2633 | ||
2634 | /* | |
2635 | * Record the must_insert_reserved flag before we drop the | |
2636 | * spin lock. | |
2637 | */ | |
2638 | must_insert_reserved = locked_ref->must_insert_reserved; | |
2639 | locked_ref->must_insert_reserved = 0; | |
2640 | ||
2641 | extent_op = locked_ref->extent_op; | |
2642 | locked_ref->extent_op = NULL; | |
2643 | spin_unlock(&locked_ref->lock); | |
2644 | ||
2645 | ret = run_one_delayed_ref(trans, ref, extent_op, | |
2646 | must_insert_reserved); | |
2647 | ||
2648 | btrfs_free_delayed_extent_op(extent_op); | |
2649 | if (ret) { | |
2650 | unselect_delayed_ref_head(delayed_refs, locked_ref); | |
2651 | btrfs_put_delayed_ref(ref); | |
2652 | btrfs_debug(fs_info, "run_one_delayed_ref returned %d", | |
2653 | ret); | |
2654 | return ret; | |
2655 | } | |
2656 | ||
2657 | btrfs_put_delayed_ref(ref); | |
2658 | cond_resched(); | |
2659 | ||
2660 | spin_lock(&locked_ref->lock); | |
2661 | btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref); | |
2662 | } | |
2663 | ||
2664 | return 0; | |
2665 | } | |
2666 | ||
79787eaa JM |
2667 | /* |
2668 | * Returns 0 on success or if called with an already aborted transaction. | |
2669 | * Returns -ENOMEM or -EIO on failure and will abort the transaction. | |
2670 | */ | |
d7df2c79 | 2671 | static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, |
d7df2c79 | 2672 | unsigned long nr) |
56bec294 | 2673 | { |
0a1e458a | 2674 | struct btrfs_fs_info *fs_info = trans->fs_info; |
56bec294 | 2675 | struct btrfs_delayed_ref_root *delayed_refs; |
56bec294 | 2676 | struct btrfs_delayed_ref_head *locked_ref = NULL; |
0a2b2a84 | 2677 | ktime_t start = ktime_get(); |
56bec294 | 2678 | int ret; |
d7df2c79 | 2679 | unsigned long count = 0; |
0a2b2a84 | 2680 | unsigned long actual_count = 0; |
56bec294 CM |
2681 | |
2682 | delayed_refs = &trans->transaction->delayed_refs; | |
0110a4c4 | 2683 | do { |
56bec294 | 2684 | if (!locked_ref) { |
b1cdbcb5 | 2685 | locked_ref = btrfs_obtain_ref_head(trans); |
0110a4c4 NB |
2686 | if (IS_ERR_OR_NULL(locked_ref)) { |
2687 | if (PTR_ERR(locked_ref) == -EAGAIN) { | |
2688 | continue; | |
2689 | } else { | |
2690 | break; | |
2691 | } | |
56bec294 | 2692 | } |
0110a4c4 | 2693 | count++; |
56bec294 | 2694 | } |
2c3cf7d5 FM |
2695 | /* |
2696 | * We need to try and merge add/drops of the same ref since we | |
2697 | * can run into issues with relocate dropping the implicit ref | |
2698 | * and then it being added back again before the drop can | |
2699 | * finish. If we merged anything we need to re-loop so we can | |
2700 | * get a good ref. | |
2701 | * Or we can get node references of the same type that weren't | |
2702 | * merged when created due to bumps in the tree mod seq, and | |
2703 | * we need to merge them to prevent adding an inline extent | |
2704 | * backref before dropping it (triggering a BUG_ON at | |
2705 | * insert_inline_extent_backref()). | |
2706 | */ | |
d7df2c79 | 2707 | spin_lock(&locked_ref->lock); |
be97f133 | 2708 | btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref); |
ae1e206b | 2709 | |
0110a4c4 NB |
2710 | ret = btrfs_run_delayed_refs_for_head(trans, locked_ref, |
2711 | &actual_count); | |
2712 | if (ret < 0 && ret != -EAGAIN) { | |
2713 | /* | |
2714 | * Error, btrfs_run_delayed_refs_for_head already | |
2715 | * unlocked everything so just bail out | |
2716 | */ | |
2717 | return ret; | |
2718 | } else if (!ret) { | |
2719 | /* | |
2720 | * Success, perform the usual cleanup of a processed | |
2721 | * head | |
2722 | */ | |
f9871edd | 2723 | ret = cleanup_ref_head(trans, locked_ref); |
194ab0bc | 2724 | if (ret > 0 ) { |
b00e6250 JB |
2725 | /* We dropped our lock, we need to loop. */ |
2726 | ret = 0; | |
d7df2c79 | 2727 | continue; |
194ab0bc JB |
2728 | } else if (ret) { |
2729 | return ret; | |
5d4f98a2 | 2730 | } |
22cd2e7d | 2731 | } |
1ce7a5ec | 2732 | |
b00e6250 | 2733 | /* |
0110a4c4 NB |
2734 | * Either success case or btrfs_run_delayed_refs_for_head |
2735 | * returned -EAGAIN, meaning we need to select another head | |
b00e6250 | 2736 | */ |
b00e6250 | 2737 | |
0110a4c4 | 2738 | locked_ref = NULL; |
c3e69d58 | 2739 | cond_resched(); |
0110a4c4 | 2740 | } while ((nr != -1 && count < nr) || locked_ref); |
0a2b2a84 JB |
2741 | |
2742 | /* | |
2743 | * We don't want to include ref heads since we can have empty ref heads | |
2744 | * and those will drastically skew our runtime down since we just do | |
2745 | * accounting, no actual extent tree updates. | |
2746 | */ | |
2747 | if (actual_count > 0) { | |
2748 | u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start)); | |
2749 | u64 avg; | |
2750 | ||
2751 | /* | |
2752 | * We weigh the current average higher than our current runtime | |
2753 | * to avoid large swings in the average. | |
2754 | */ | |
2755 | spin_lock(&delayed_refs->lock); | |
2756 | avg = fs_info->avg_delayed_ref_runtime * 3 + runtime; | |
f8c269d7 | 2757 | fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */ |
0a2b2a84 JB |
2758 | spin_unlock(&delayed_refs->lock); |
2759 | } | |
d7df2c79 | 2760 | return 0; |
c3e69d58 CM |
2761 | } |
2762 | ||
709c0486 AJ |
2763 | #ifdef SCRAMBLE_DELAYED_REFS |
2764 | /* | |
2765 | * Normally delayed refs get processed in ascending bytenr order. This | |
2766 | * correlates in most cases to the order added. To expose dependencies on this | |
2767 | * order, we start to process the tree in the middle instead of the beginning | |
2768 | */ | |
2769 | static u64 find_middle(struct rb_root *root) | |
2770 | { | |
2771 | struct rb_node *n = root->rb_node; | |
2772 | struct btrfs_delayed_ref_node *entry; | |
2773 | int alt = 1; | |
2774 | u64 middle; | |
2775 | u64 first = 0, last = 0; | |
2776 | ||
2777 | n = rb_first(root); | |
2778 | if (n) { | |
2779 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2780 | first = entry->bytenr; | |
2781 | } | |
2782 | n = rb_last(root); | |
2783 | if (n) { | |
2784 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2785 | last = entry->bytenr; | |
2786 | } | |
2787 | n = root->rb_node; | |
2788 | ||
2789 | while (n) { | |
2790 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2791 | WARN_ON(!entry->in_tree); | |
2792 | ||
2793 | middle = entry->bytenr; | |
2794 | ||
2795 | if (alt) | |
2796 | n = n->rb_left; | |
2797 | else | |
2798 | n = n->rb_right; | |
2799 | ||
2800 | alt = 1 - alt; | |
2801 | } | |
2802 | return middle; | |
2803 | } | |
2804 | #endif | |
2805 | ||
2ff7e61e | 2806 | static inline u64 heads_to_leaves(struct btrfs_fs_info *fs_info, u64 heads) |
1be41b78 JB |
2807 | { |
2808 | u64 num_bytes; | |
2809 | ||
2810 | num_bytes = heads * (sizeof(struct btrfs_extent_item) + | |
2811 | sizeof(struct btrfs_extent_inline_ref)); | |
0b246afa | 2812 | if (!btrfs_fs_incompat(fs_info, SKINNY_METADATA)) |
1be41b78 JB |
2813 | num_bytes += heads * sizeof(struct btrfs_tree_block_info); |
2814 | ||
2815 | /* | |
2816 | * We don't ever fill up leaves all the way so multiply by 2 just to be | |
01327610 | 2817 | * closer to what we're really going to want to use. |
1be41b78 | 2818 | */ |
0b246afa | 2819 | return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(fs_info)); |
1be41b78 JB |
2820 | } |
2821 | ||
1262133b JB |
2822 | /* |
2823 | * Takes the number of bytes to be csumm'ed and figures out how many leaves it | |
2824 | * would require to store the csums for that many bytes. | |
2825 | */ | |
2ff7e61e | 2826 | u64 btrfs_csum_bytes_to_leaves(struct btrfs_fs_info *fs_info, u64 csum_bytes) |
1262133b JB |
2827 | { |
2828 | u64 csum_size; | |
2829 | u64 num_csums_per_leaf; | |
2830 | u64 num_csums; | |
2831 | ||
0b246afa | 2832 | csum_size = BTRFS_MAX_ITEM_SIZE(fs_info); |
1262133b | 2833 | num_csums_per_leaf = div64_u64(csum_size, |
0b246afa JM |
2834 | (u64)btrfs_super_csum_size(fs_info->super_copy)); |
2835 | num_csums = div64_u64(csum_bytes, fs_info->sectorsize); | |
1262133b JB |
2836 | num_csums += num_csums_per_leaf - 1; |
2837 | num_csums = div64_u64(num_csums, num_csums_per_leaf); | |
2838 | return num_csums; | |
2839 | } | |
2840 | ||
64403612 | 2841 | bool btrfs_check_space_for_delayed_refs(struct btrfs_fs_info *fs_info) |
1be41b78 | 2842 | { |
64403612 JB |
2843 | struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv; |
2844 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
2845 | bool ret = false; | |
2846 | u64 reserved; | |
1be41b78 | 2847 | |
64403612 JB |
2848 | spin_lock(&global_rsv->lock); |
2849 | reserved = global_rsv->reserved; | |
2850 | spin_unlock(&global_rsv->lock); | |
1be41b78 JB |
2851 | |
2852 | /* | |
64403612 JB |
2853 | * Since the global reserve is just kind of magic we don't really want |
2854 | * to rely on it to save our bacon, so if our size is more than the | |
2855 | * delayed_refs_rsv and the global rsv then it's time to think about | |
2856 | * bailing. | |
1be41b78 | 2857 | */ |
64403612 JB |
2858 | spin_lock(&delayed_refs_rsv->lock); |
2859 | reserved += delayed_refs_rsv->reserved; | |
2860 | if (delayed_refs_rsv->size >= reserved) | |
2861 | ret = true; | |
2862 | spin_unlock(&delayed_refs_rsv->lock); | |
1be41b78 JB |
2863 | return ret; |
2864 | } | |
2865 | ||
7c861627 | 2866 | int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans) |
0a2b2a84 | 2867 | { |
0a2b2a84 JB |
2868 | u64 num_entries = |
2869 | atomic_read(&trans->transaction->delayed_refs.num_entries); | |
2870 | u64 avg_runtime; | |
a79b7d4b | 2871 | u64 val; |
0a2b2a84 JB |
2872 | |
2873 | smp_mb(); | |
7c861627 | 2874 | avg_runtime = trans->fs_info->avg_delayed_ref_runtime; |
a79b7d4b | 2875 | val = num_entries * avg_runtime; |
dc1a90c6 | 2876 | if (val >= NSEC_PER_SEC) |
0a2b2a84 | 2877 | return 1; |
a79b7d4b CM |
2878 | if (val >= NSEC_PER_SEC / 2) |
2879 | return 2; | |
0a2b2a84 | 2880 | |
64403612 | 2881 | return btrfs_check_space_for_delayed_refs(trans->fs_info); |
0a2b2a84 JB |
2882 | } |
2883 | ||
c3e69d58 CM |
2884 | /* |
2885 | * this starts processing the delayed reference count updates and | |
2886 | * extent insertions we have queued up so far. count can be | |
2887 | * 0, which means to process everything in the tree at the start | |
2888 | * of the run (but not newly added entries), or it can be some target | |
2889 | * number you'd like to process. | |
79787eaa JM |
2890 | * |
2891 | * Returns 0 on success or if called with an aborted transaction | |
2892 | * Returns <0 on error and aborts the transaction | |
c3e69d58 CM |
2893 | */ |
2894 | int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, | |
c79a70b1 | 2895 | unsigned long count) |
c3e69d58 | 2896 | { |
c79a70b1 | 2897 | struct btrfs_fs_info *fs_info = trans->fs_info; |
c3e69d58 CM |
2898 | struct rb_node *node; |
2899 | struct btrfs_delayed_ref_root *delayed_refs; | |
c46effa6 | 2900 | struct btrfs_delayed_ref_head *head; |
c3e69d58 CM |
2901 | int ret; |
2902 | int run_all = count == (unsigned long)-1; | |
c3e69d58 | 2903 | |
79787eaa JM |
2904 | /* We'll clean this up in btrfs_cleanup_transaction */ |
2905 | if (trans->aborted) | |
2906 | return 0; | |
2907 | ||
0b246afa | 2908 | if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags)) |
511711af CM |
2909 | return 0; |
2910 | ||
c3e69d58 | 2911 | delayed_refs = &trans->transaction->delayed_refs; |
26455d33 | 2912 | if (count == 0) |
d7df2c79 | 2913 | count = atomic_read(&delayed_refs->num_entries) * 2; |
bb721703 | 2914 | |
c3e69d58 | 2915 | again: |
709c0486 AJ |
2916 | #ifdef SCRAMBLE_DELAYED_REFS |
2917 | delayed_refs->run_delayed_start = find_middle(&delayed_refs->root); | |
2918 | #endif | |
0a1e458a | 2919 | ret = __btrfs_run_delayed_refs(trans, count); |
d7df2c79 | 2920 | if (ret < 0) { |
66642832 | 2921 | btrfs_abort_transaction(trans, ret); |
d7df2c79 | 2922 | return ret; |
eb099670 | 2923 | } |
c3e69d58 | 2924 | |
56bec294 | 2925 | if (run_all) { |
119e80df | 2926 | btrfs_create_pending_block_groups(trans); |
ea658bad | 2927 | |
d7df2c79 | 2928 | spin_lock(&delayed_refs->lock); |
5c9d028b | 2929 | node = rb_first_cached(&delayed_refs->href_root); |
d7df2c79 JB |
2930 | if (!node) { |
2931 | spin_unlock(&delayed_refs->lock); | |
56bec294 | 2932 | goto out; |
d7df2c79 | 2933 | } |
d278850e JB |
2934 | head = rb_entry(node, struct btrfs_delayed_ref_head, |
2935 | href_node); | |
2936 | refcount_inc(&head->refs); | |
2937 | spin_unlock(&delayed_refs->lock); | |
e9d0b13b | 2938 | |
d278850e JB |
2939 | /* Mutex was contended, block until it's released and retry. */ |
2940 | mutex_lock(&head->mutex); | |
2941 | mutex_unlock(&head->mutex); | |
56bec294 | 2942 | |
d278850e | 2943 | btrfs_put_delayed_ref_head(head); |
d7df2c79 | 2944 | cond_resched(); |
56bec294 | 2945 | goto again; |
5f39d397 | 2946 | } |
54aa1f4d | 2947 | out: |
a28ec197 CM |
2948 | return 0; |
2949 | } | |
2950 | ||
5d4f98a2 | 2951 | int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans, |
5d4f98a2 | 2952 | u64 bytenr, u64 num_bytes, u64 flags, |
b1c79e09 | 2953 | int level, int is_data) |
5d4f98a2 YZ |
2954 | { |
2955 | struct btrfs_delayed_extent_op *extent_op; | |
2956 | int ret; | |
2957 | ||
78a6184a | 2958 | extent_op = btrfs_alloc_delayed_extent_op(); |
5d4f98a2 YZ |
2959 | if (!extent_op) |
2960 | return -ENOMEM; | |
2961 | ||
2962 | extent_op->flags_to_set = flags; | |
35b3ad50 DS |
2963 | extent_op->update_flags = true; |
2964 | extent_op->update_key = false; | |
2965 | extent_op->is_data = is_data ? true : false; | |
b1c79e09 | 2966 | extent_op->level = level; |
5d4f98a2 | 2967 | |
c6e340bc | 2968 | ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op); |
5d4f98a2 | 2969 | if (ret) |
78a6184a | 2970 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 YZ |
2971 | return ret; |
2972 | } | |
2973 | ||
e4c3b2dc | 2974 | static noinline int check_delayed_ref(struct btrfs_root *root, |
5d4f98a2 YZ |
2975 | struct btrfs_path *path, |
2976 | u64 objectid, u64 offset, u64 bytenr) | |
2977 | { | |
2978 | struct btrfs_delayed_ref_head *head; | |
2979 | struct btrfs_delayed_ref_node *ref; | |
2980 | struct btrfs_delayed_data_ref *data_ref; | |
2981 | struct btrfs_delayed_ref_root *delayed_refs; | |
e4c3b2dc | 2982 | struct btrfs_transaction *cur_trans; |
0e0adbcf | 2983 | struct rb_node *node; |
5d4f98a2 YZ |
2984 | int ret = 0; |
2985 | ||
998ac6d2 | 2986 | spin_lock(&root->fs_info->trans_lock); |
e4c3b2dc | 2987 | cur_trans = root->fs_info->running_transaction; |
998ac6d2 | 2988 | if (cur_trans) |
2989 | refcount_inc(&cur_trans->use_count); | |
2990 | spin_unlock(&root->fs_info->trans_lock); | |
e4c3b2dc LB |
2991 | if (!cur_trans) |
2992 | return 0; | |
2993 | ||
2994 | delayed_refs = &cur_trans->delayed_refs; | |
5d4f98a2 | 2995 | spin_lock(&delayed_refs->lock); |
f72ad18e | 2996 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
d7df2c79 JB |
2997 | if (!head) { |
2998 | spin_unlock(&delayed_refs->lock); | |
998ac6d2 | 2999 | btrfs_put_transaction(cur_trans); |
d7df2c79 JB |
3000 | return 0; |
3001 | } | |
5d4f98a2 YZ |
3002 | |
3003 | if (!mutex_trylock(&head->mutex)) { | |
d278850e | 3004 | refcount_inc(&head->refs); |
5d4f98a2 YZ |
3005 | spin_unlock(&delayed_refs->lock); |
3006 | ||
b3b4aa74 | 3007 | btrfs_release_path(path); |
5d4f98a2 | 3008 | |
8cc33e5c DS |
3009 | /* |
3010 | * Mutex was contended, block until it's released and let | |
3011 | * caller try again | |
3012 | */ | |
5d4f98a2 YZ |
3013 | mutex_lock(&head->mutex); |
3014 | mutex_unlock(&head->mutex); | |
d278850e | 3015 | btrfs_put_delayed_ref_head(head); |
998ac6d2 | 3016 | btrfs_put_transaction(cur_trans); |
5d4f98a2 YZ |
3017 | return -EAGAIN; |
3018 | } | |
d7df2c79 | 3019 | spin_unlock(&delayed_refs->lock); |
5d4f98a2 | 3020 | |
d7df2c79 | 3021 | spin_lock(&head->lock); |
0e0adbcf JB |
3022 | /* |
3023 | * XXX: We should replace this with a proper search function in the | |
3024 | * future. | |
3025 | */ | |
e3d03965 LB |
3026 | for (node = rb_first_cached(&head->ref_tree); node; |
3027 | node = rb_next(node)) { | |
0e0adbcf | 3028 | ref = rb_entry(node, struct btrfs_delayed_ref_node, ref_node); |
d7df2c79 JB |
3029 | /* If it's a shared ref we know a cross reference exists */ |
3030 | if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) { | |
3031 | ret = 1; | |
3032 | break; | |
3033 | } | |
5d4f98a2 | 3034 | |
d7df2c79 | 3035 | data_ref = btrfs_delayed_node_to_data_ref(ref); |
5d4f98a2 | 3036 | |
d7df2c79 JB |
3037 | /* |
3038 | * If our ref doesn't match the one we're currently looking at | |
3039 | * then we have a cross reference. | |
3040 | */ | |
3041 | if (data_ref->root != root->root_key.objectid || | |
3042 | data_ref->objectid != objectid || | |
3043 | data_ref->offset != offset) { | |
3044 | ret = 1; | |
3045 | break; | |
3046 | } | |
5d4f98a2 | 3047 | } |
d7df2c79 | 3048 | spin_unlock(&head->lock); |
5d4f98a2 | 3049 | mutex_unlock(&head->mutex); |
998ac6d2 | 3050 | btrfs_put_transaction(cur_trans); |
5d4f98a2 YZ |
3051 | return ret; |
3052 | } | |
3053 | ||
e4c3b2dc | 3054 | static noinline int check_committed_ref(struct btrfs_root *root, |
5d4f98a2 YZ |
3055 | struct btrfs_path *path, |
3056 | u64 objectid, u64 offset, u64 bytenr) | |
be20aa9d | 3057 | { |
0b246afa JM |
3058 | struct btrfs_fs_info *fs_info = root->fs_info; |
3059 | struct btrfs_root *extent_root = fs_info->extent_root; | |
f321e491 | 3060 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
3061 | struct btrfs_extent_data_ref *ref; |
3062 | struct btrfs_extent_inline_ref *iref; | |
3063 | struct btrfs_extent_item *ei; | |
f321e491 | 3064 | struct btrfs_key key; |
5d4f98a2 | 3065 | u32 item_size; |
3de28d57 | 3066 | int type; |
be20aa9d | 3067 | int ret; |
925baedd | 3068 | |
be20aa9d | 3069 | key.objectid = bytenr; |
31840ae1 | 3070 | key.offset = (u64)-1; |
f321e491 | 3071 | key.type = BTRFS_EXTENT_ITEM_KEY; |
be20aa9d | 3072 | |
be20aa9d CM |
3073 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
3074 | if (ret < 0) | |
3075 | goto out; | |
79787eaa | 3076 | BUG_ON(ret == 0); /* Corruption */ |
80ff3856 YZ |
3077 | |
3078 | ret = -ENOENT; | |
3079 | if (path->slots[0] == 0) | |
31840ae1 | 3080 | goto out; |
be20aa9d | 3081 | |
31840ae1 | 3082 | path->slots[0]--; |
f321e491 | 3083 | leaf = path->nodes[0]; |
5d4f98a2 | 3084 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
be20aa9d | 3085 | |
5d4f98a2 | 3086 | if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY) |
be20aa9d | 3087 | goto out; |
f321e491 | 3088 | |
5d4f98a2 YZ |
3089 | ret = 1; |
3090 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
5d4f98a2 | 3091 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
bd09835d | 3092 | |
5d4f98a2 YZ |
3093 | if (item_size != sizeof(*ei) + |
3094 | btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY)) | |
3095 | goto out; | |
be20aa9d | 3096 | |
5d4f98a2 YZ |
3097 | if (btrfs_extent_generation(leaf, ei) <= |
3098 | btrfs_root_last_snapshot(&root->root_item)) | |
3099 | goto out; | |
3100 | ||
3101 | iref = (struct btrfs_extent_inline_ref *)(ei + 1); | |
3de28d57 LB |
3102 | |
3103 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA); | |
3104 | if (type != BTRFS_EXTENT_DATA_REF_KEY) | |
5d4f98a2 YZ |
3105 | goto out; |
3106 | ||
3107 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
3108 | if (btrfs_extent_refs(leaf, ei) != | |
3109 | btrfs_extent_data_ref_count(leaf, ref) || | |
3110 | btrfs_extent_data_ref_root(leaf, ref) != | |
3111 | root->root_key.objectid || | |
3112 | btrfs_extent_data_ref_objectid(leaf, ref) != objectid || | |
3113 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
3114 | goto out; | |
3115 | ||
3116 | ret = 0; | |
3117 | out: | |
3118 | return ret; | |
3119 | } | |
3120 | ||
e4c3b2dc LB |
3121 | int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset, |
3122 | u64 bytenr) | |
5d4f98a2 YZ |
3123 | { |
3124 | struct btrfs_path *path; | |
3125 | int ret; | |
5d4f98a2 YZ |
3126 | |
3127 | path = btrfs_alloc_path(); | |
3128 | if (!path) | |
9132c4ff | 3129 | return -ENOMEM; |
5d4f98a2 YZ |
3130 | |
3131 | do { | |
e4c3b2dc | 3132 | ret = check_committed_ref(root, path, objectid, |
5d4f98a2 YZ |
3133 | offset, bytenr); |
3134 | if (ret && ret != -ENOENT) | |
f321e491 | 3135 | goto out; |
80ff3856 | 3136 | |
380fd066 MT |
3137 | ret = check_delayed_ref(root, path, objectid, offset, bytenr); |
3138 | } while (ret == -EAGAIN); | |
5d4f98a2 | 3139 | |
be20aa9d | 3140 | out: |
80ff3856 | 3141 | btrfs_free_path(path); |
f0486c68 YZ |
3142 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) |
3143 | WARN_ON(ret > 0); | |
f321e491 | 3144 | return ret; |
be20aa9d | 3145 | } |
c5739bba | 3146 | |
5d4f98a2 | 3147 | static int __btrfs_mod_ref(struct btrfs_trans_handle *trans, |
b7a9f29f | 3148 | struct btrfs_root *root, |
5d4f98a2 | 3149 | struct extent_buffer *buf, |
e339a6b0 | 3150 | int full_backref, int inc) |
31840ae1 | 3151 | { |
0b246afa | 3152 | struct btrfs_fs_info *fs_info = root->fs_info; |
31840ae1 | 3153 | u64 bytenr; |
5d4f98a2 YZ |
3154 | u64 num_bytes; |
3155 | u64 parent; | |
31840ae1 | 3156 | u64 ref_root; |
31840ae1 | 3157 | u32 nritems; |
31840ae1 ZY |
3158 | struct btrfs_key key; |
3159 | struct btrfs_file_extent_item *fi; | |
82fa113f QW |
3160 | struct btrfs_ref generic_ref = { 0 }; |
3161 | bool for_reloc = btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC); | |
31840ae1 | 3162 | int i; |
82fa113f | 3163 | int action; |
31840ae1 ZY |
3164 | int level; |
3165 | int ret = 0; | |
fccb84c9 | 3166 | |
0b246afa | 3167 | if (btrfs_is_testing(fs_info)) |
faa2dbf0 | 3168 | return 0; |
fccb84c9 | 3169 | |
31840ae1 | 3170 | ref_root = btrfs_header_owner(buf); |
31840ae1 ZY |
3171 | nritems = btrfs_header_nritems(buf); |
3172 | level = btrfs_header_level(buf); | |
3173 | ||
27cdeb70 | 3174 | if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0) |
5d4f98a2 | 3175 | return 0; |
31840ae1 | 3176 | |
5d4f98a2 YZ |
3177 | if (full_backref) |
3178 | parent = buf->start; | |
3179 | else | |
3180 | parent = 0; | |
82fa113f QW |
3181 | if (inc) |
3182 | action = BTRFS_ADD_DELAYED_REF; | |
3183 | else | |
3184 | action = BTRFS_DROP_DELAYED_REF; | |
5d4f98a2 YZ |
3185 | |
3186 | for (i = 0; i < nritems; i++) { | |
31840ae1 | 3187 | if (level == 0) { |
5d4f98a2 | 3188 | btrfs_item_key_to_cpu(buf, &key, i); |
962a298f | 3189 | if (key.type != BTRFS_EXTENT_DATA_KEY) |
31840ae1 | 3190 | continue; |
5d4f98a2 | 3191 | fi = btrfs_item_ptr(buf, i, |
31840ae1 ZY |
3192 | struct btrfs_file_extent_item); |
3193 | if (btrfs_file_extent_type(buf, fi) == | |
3194 | BTRFS_FILE_EXTENT_INLINE) | |
3195 | continue; | |
3196 | bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
3197 | if (bytenr == 0) | |
3198 | continue; | |
5d4f98a2 YZ |
3199 | |
3200 | num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi); | |
3201 | key.offset -= btrfs_file_extent_offset(buf, fi); | |
82fa113f QW |
3202 | btrfs_init_generic_ref(&generic_ref, action, bytenr, |
3203 | num_bytes, parent); | |
3204 | generic_ref.real_root = root->root_key.objectid; | |
3205 | btrfs_init_data_ref(&generic_ref, ref_root, key.objectid, | |
3206 | key.offset); | |
3207 | generic_ref.skip_qgroup = for_reloc; | |
dd28b6a5 | 3208 | if (inc) |
82fa113f | 3209 | ret = btrfs_inc_extent_ref(trans, &generic_ref); |
dd28b6a5 | 3210 | else |
ffd4bb2a | 3211 | ret = btrfs_free_extent(trans, &generic_ref); |
31840ae1 ZY |
3212 | if (ret) |
3213 | goto fail; | |
3214 | } else { | |
5d4f98a2 | 3215 | bytenr = btrfs_node_blockptr(buf, i); |
0b246afa | 3216 | num_bytes = fs_info->nodesize; |
82fa113f QW |
3217 | btrfs_init_generic_ref(&generic_ref, action, bytenr, |
3218 | num_bytes, parent); | |
3219 | generic_ref.real_root = root->root_key.objectid; | |
3220 | btrfs_init_tree_ref(&generic_ref, level - 1, ref_root); | |
3221 | generic_ref.skip_qgroup = for_reloc; | |
dd28b6a5 | 3222 | if (inc) |
82fa113f | 3223 | ret = btrfs_inc_extent_ref(trans, &generic_ref); |
dd28b6a5 | 3224 | else |
ffd4bb2a | 3225 | ret = btrfs_free_extent(trans, &generic_ref); |
31840ae1 ZY |
3226 | if (ret) |
3227 | goto fail; | |
3228 | } | |
3229 | } | |
3230 | return 0; | |
3231 | fail: | |
5d4f98a2 YZ |
3232 | return ret; |
3233 | } | |
3234 | ||
3235 | int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
e339a6b0 | 3236 | struct extent_buffer *buf, int full_backref) |
5d4f98a2 | 3237 | { |
e339a6b0 | 3238 | return __btrfs_mod_ref(trans, root, buf, full_backref, 1); |
5d4f98a2 YZ |
3239 | } |
3240 | ||
3241 | int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
e339a6b0 | 3242 | struct extent_buffer *buf, int full_backref) |
5d4f98a2 | 3243 | { |
e339a6b0 | 3244 | return __btrfs_mod_ref(trans, root, buf, full_backref, 0); |
31840ae1 ZY |
3245 | } |
3246 | ||
9078a3e1 | 3247 | static int write_one_cache_group(struct btrfs_trans_handle *trans, |
9078a3e1 CM |
3248 | struct btrfs_path *path, |
3249 | struct btrfs_block_group_cache *cache) | |
3250 | { | |
39db232d | 3251 | struct btrfs_fs_info *fs_info = trans->fs_info; |
9078a3e1 | 3252 | int ret; |
0b246afa | 3253 | struct btrfs_root *extent_root = fs_info->extent_root; |
5f39d397 CM |
3254 | unsigned long bi; |
3255 | struct extent_buffer *leaf; | |
9078a3e1 | 3256 | |
9078a3e1 | 3257 | ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1); |
df95e7f0 JB |
3258 | if (ret) { |
3259 | if (ret > 0) | |
3260 | ret = -ENOENT; | |
54aa1f4d | 3261 | goto fail; |
df95e7f0 | 3262 | } |
5f39d397 CM |
3263 | |
3264 | leaf = path->nodes[0]; | |
3265 | bi = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
3266 | write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item)); | |
3267 | btrfs_mark_buffer_dirty(leaf); | |
54aa1f4d | 3268 | fail: |
24b89d08 | 3269 | btrfs_release_path(path); |
df95e7f0 | 3270 | return ret; |
9078a3e1 CM |
3271 | |
3272 | } | |
3273 | ||
f87b7eb8 DS |
3274 | static struct btrfs_block_group_cache *next_block_group( |
3275 | struct btrfs_block_group_cache *cache) | |
4a8c9a62 | 3276 | { |
f87b7eb8 | 3277 | struct btrfs_fs_info *fs_info = cache->fs_info; |
4a8c9a62 | 3278 | struct rb_node *node; |
292cbd51 | 3279 | |
0b246afa | 3280 | spin_lock(&fs_info->block_group_cache_lock); |
292cbd51 FM |
3281 | |
3282 | /* If our block group was removed, we need a full search. */ | |
3283 | if (RB_EMPTY_NODE(&cache->cache_node)) { | |
3284 | const u64 next_bytenr = cache->key.objectid + cache->key.offset; | |
3285 | ||
0b246afa | 3286 | spin_unlock(&fs_info->block_group_cache_lock); |
292cbd51 | 3287 | btrfs_put_block_group(cache); |
0b246afa | 3288 | cache = btrfs_lookup_first_block_group(fs_info, next_bytenr); return cache; |
292cbd51 | 3289 | } |
4a8c9a62 YZ |
3290 | node = rb_next(&cache->cache_node); |
3291 | btrfs_put_block_group(cache); | |
3292 | if (node) { | |
3293 | cache = rb_entry(node, struct btrfs_block_group_cache, | |
3294 | cache_node); | |
11dfe35a | 3295 | btrfs_get_block_group(cache); |
4a8c9a62 YZ |
3296 | } else |
3297 | cache = NULL; | |
0b246afa | 3298 | spin_unlock(&fs_info->block_group_cache_lock); |
4a8c9a62 YZ |
3299 | return cache; |
3300 | } | |
3301 | ||
0af3d00b JB |
3302 | static int cache_save_setup(struct btrfs_block_group_cache *block_group, |
3303 | struct btrfs_trans_handle *trans, | |
3304 | struct btrfs_path *path) | |
3305 | { | |
0b246afa JM |
3306 | struct btrfs_fs_info *fs_info = block_group->fs_info; |
3307 | struct btrfs_root *root = fs_info->tree_root; | |
0af3d00b | 3308 | struct inode *inode = NULL; |
364ecf36 | 3309 | struct extent_changeset *data_reserved = NULL; |
0af3d00b | 3310 | u64 alloc_hint = 0; |
2b20982e | 3311 | int dcs = BTRFS_DC_ERROR; |
f8c269d7 | 3312 | u64 num_pages = 0; |
0af3d00b JB |
3313 | int retries = 0; |
3314 | int ret = 0; | |
3315 | ||
3316 | /* | |
3317 | * If this block group is smaller than 100 megs don't bother caching the | |
3318 | * block group. | |
3319 | */ | |
ee22184b | 3320 | if (block_group->key.offset < (100 * SZ_1M)) { |
0af3d00b JB |
3321 | spin_lock(&block_group->lock); |
3322 | block_group->disk_cache_state = BTRFS_DC_WRITTEN; | |
3323 | spin_unlock(&block_group->lock); | |
3324 | return 0; | |
3325 | } | |
3326 | ||
0c0ef4bc JB |
3327 | if (trans->aborted) |
3328 | return 0; | |
0af3d00b | 3329 | again: |
7949f339 | 3330 | inode = lookup_free_space_inode(block_group, path); |
0af3d00b JB |
3331 | if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) { |
3332 | ret = PTR_ERR(inode); | |
b3b4aa74 | 3333 | btrfs_release_path(path); |
0af3d00b JB |
3334 | goto out; |
3335 | } | |
3336 | ||
3337 | if (IS_ERR(inode)) { | |
3338 | BUG_ON(retries); | |
3339 | retries++; | |
3340 | ||
3341 | if (block_group->ro) | |
3342 | goto out_free; | |
3343 | ||
4ca75f1b | 3344 | ret = create_free_space_inode(trans, block_group, path); |
0af3d00b JB |
3345 | if (ret) |
3346 | goto out_free; | |
3347 | goto again; | |
3348 | } | |
3349 | ||
3350 | /* | |
3351 | * We want to set the generation to 0, that way if anything goes wrong | |
3352 | * from here on out we know not to trust this cache when we load up next | |
3353 | * time. | |
3354 | */ | |
3355 | BTRFS_I(inode)->generation = 0; | |
3356 | ret = btrfs_update_inode(trans, root, inode); | |
0c0ef4bc JB |
3357 | if (ret) { |
3358 | /* | |
3359 | * So theoretically we could recover from this, simply set the | |
3360 | * super cache generation to 0 so we know to invalidate the | |
3361 | * cache, but then we'd have to keep track of the block groups | |
3362 | * that fail this way so we know we _have_ to reset this cache | |
3363 | * before the next commit or risk reading stale cache. So to | |
3364 | * limit our exposure to horrible edge cases lets just abort the | |
3365 | * transaction, this only happens in really bad situations | |
3366 | * anyway. | |
3367 | */ | |
66642832 | 3368 | btrfs_abort_transaction(trans, ret); |
0c0ef4bc JB |
3369 | goto out_put; |
3370 | } | |
0af3d00b JB |
3371 | WARN_ON(ret); |
3372 | ||
8e138e0d JB |
3373 | /* We've already setup this transaction, go ahead and exit */ |
3374 | if (block_group->cache_generation == trans->transid && | |
3375 | i_size_read(inode)) { | |
3376 | dcs = BTRFS_DC_SETUP; | |
3377 | goto out_put; | |
3378 | } | |
3379 | ||
0af3d00b | 3380 | if (i_size_read(inode) > 0) { |
2ff7e61e | 3381 | ret = btrfs_check_trunc_cache_free_space(fs_info, |
0b246afa | 3382 | &fs_info->global_block_rsv); |
7b61cd92 MX |
3383 | if (ret) |
3384 | goto out_put; | |
3385 | ||
77ab86bf | 3386 | ret = btrfs_truncate_free_space_cache(trans, NULL, inode); |
0af3d00b JB |
3387 | if (ret) |
3388 | goto out_put; | |
3389 | } | |
3390 | ||
3391 | spin_lock(&block_group->lock); | |
cf7c1ef6 | 3392 | if (block_group->cached != BTRFS_CACHE_FINISHED || |
0b246afa | 3393 | !btrfs_test_opt(fs_info, SPACE_CACHE)) { |
cf7c1ef6 LB |
3394 | /* |
3395 | * don't bother trying to write stuff out _if_ | |
3396 | * a) we're not cached, | |
1a79c1f2 LB |
3397 | * b) we're with nospace_cache mount option, |
3398 | * c) we're with v2 space_cache (FREE_SPACE_TREE). | |
cf7c1ef6 | 3399 | */ |
2b20982e | 3400 | dcs = BTRFS_DC_WRITTEN; |
0af3d00b JB |
3401 | spin_unlock(&block_group->lock); |
3402 | goto out_put; | |
3403 | } | |
3404 | spin_unlock(&block_group->lock); | |
3405 | ||
2968b1f4 JB |
3406 | /* |
3407 | * We hit an ENOSPC when setting up the cache in this transaction, just | |
3408 | * skip doing the setup, we've already cleared the cache so we're safe. | |
3409 | */ | |
3410 | if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) { | |
3411 | ret = -ENOSPC; | |
3412 | goto out_put; | |
3413 | } | |
3414 | ||
6fc823b1 JB |
3415 | /* |
3416 | * Try to preallocate enough space based on how big the block group is. | |
3417 | * Keep in mind this has to include any pinned space which could end up | |
3418 | * taking up quite a bit since it's not folded into the other space | |
3419 | * cache. | |
3420 | */ | |
ee22184b | 3421 | num_pages = div_u64(block_group->key.offset, SZ_256M); |
0af3d00b JB |
3422 | if (!num_pages) |
3423 | num_pages = 1; | |
3424 | ||
0af3d00b | 3425 | num_pages *= 16; |
09cbfeaf | 3426 | num_pages *= PAGE_SIZE; |
0af3d00b | 3427 | |
364ecf36 | 3428 | ret = btrfs_check_data_free_space(inode, &data_reserved, 0, num_pages); |
0af3d00b JB |
3429 | if (ret) |
3430 | goto out_put; | |
3431 | ||
3432 | ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages, | |
3433 | num_pages, num_pages, | |
3434 | &alloc_hint); | |
2968b1f4 JB |
3435 | /* |
3436 | * Our cache requires contiguous chunks so that we don't modify a bunch | |
3437 | * of metadata or split extents when writing the cache out, which means | |
3438 | * we can enospc if we are heavily fragmented in addition to just normal | |
3439 | * out of space conditions. So if we hit this just skip setting up any | |
3440 | * other block groups for this transaction, maybe we'll unpin enough | |
3441 | * space the next time around. | |
3442 | */ | |
2b20982e JB |
3443 | if (!ret) |
3444 | dcs = BTRFS_DC_SETUP; | |
2968b1f4 JB |
3445 | else if (ret == -ENOSPC) |
3446 | set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags); | |
c09544e0 | 3447 | |
0af3d00b JB |
3448 | out_put: |
3449 | iput(inode); | |
3450 | out_free: | |
b3b4aa74 | 3451 | btrfs_release_path(path); |
0af3d00b JB |
3452 | out: |
3453 | spin_lock(&block_group->lock); | |
e65cbb94 | 3454 | if (!ret && dcs == BTRFS_DC_SETUP) |
5b0e95bf | 3455 | block_group->cache_generation = trans->transid; |
2b20982e | 3456 | block_group->disk_cache_state = dcs; |
0af3d00b JB |
3457 | spin_unlock(&block_group->lock); |
3458 | ||
364ecf36 | 3459 | extent_changeset_free(data_reserved); |
0af3d00b JB |
3460 | return ret; |
3461 | } | |
3462 | ||
bbebb3e0 | 3463 | int btrfs_setup_space_cache(struct btrfs_trans_handle *trans) |
dcdf7f6d | 3464 | { |
bbebb3e0 | 3465 | struct btrfs_fs_info *fs_info = trans->fs_info; |
dcdf7f6d JB |
3466 | struct btrfs_block_group_cache *cache, *tmp; |
3467 | struct btrfs_transaction *cur_trans = trans->transaction; | |
3468 | struct btrfs_path *path; | |
3469 | ||
3470 | if (list_empty(&cur_trans->dirty_bgs) || | |
0b246afa | 3471 | !btrfs_test_opt(fs_info, SPACE_CACHE)) |
dcdf7f6d JB |
3472 | return 0; |
3473 | ||
3474 | path = btrfs_alloc_path(); | |
3475 | if (!path) | |
3476 | return -ENOMEM; | |
3477 | ||
3478 | /* Could add new block groups, use _safe just in case */ | |
3479 | list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs, | |
3480 | dirty_list) { | |
3481 | if (cache->disk_cache_state == BTRFS_DC_CLEAR) | |
3482 | cache_save_setup(cache, trans, path); | |
3483 | } | |
3484 | ||
3485 | btrfs_free_path(path); | |
3486 | return 0; | |
3487 | } | |
3488 | ||
1bbc621e CM |
3489 | /* |
3490 | * transaction commit does final block group cache writeback during a | |
3491 | * critical section where nothing is allowed to change the FS. This is | |
3492 | * required in order for the cache to actually match the block group, | |
3493 | * but can introduce a lot of latency into the commit. | |
3494 | * | |
3495 | * So, btrfs_start_dirty_block_groups is here to kick off block group | |
3496 | * cache IO. There's a chance we'll have to redo some of it if the | |
3497 | * block group changes again during the commit, but it greatly reduces | |
3498 | * the commit latency by getting rid of the easy block groups while | |
3499 | * we're still allowing others to join the commit. | |
3500 | */ | |
21217054 | 3501 | int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans) |
9078a3e1 | 3502 | { |
21217054 | 3503 | struct btrfs_fs_info *fs_info = trans->fs_info; |
4a8c9a62 | 3504 | struct btrfs_block_group_cache *cache; |
ce93ec54 JB |
3505 | struct btrfs_transaction *cur_trans = trans->transaction; |
3506 | int ret = 0; | |
c9dc4c65 | 3507 | int should_put; |
1bbc621e CM |
3508 | struct btrfs_path *path = NULL; |
3509 | LIST_HEAD(dirty); | |
3510 | struct list_head *io = &cur_trans->io_bgs; | |
c9dc4c65 | 3511 | int num_started = 0; |
1bbc621e CM |
3512 | int loops = 0; |
3513 | ||
3514 | spin_lock(&cur_trans->dirty_bgs_lock); | |
b58d1a9e FM |
3515 | if (list_empty(&cur_trans->dirty_bgs)) { |
3516 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3517 | return 0; | |
1bbc621e | 3518 | } |
b58d1a9e | 3519 | list_splice_init(&cur_trans->dirty_bgs, &dirty); |
1bbc621e | 3520 | spin_unlock(&cur_trans->dirty_bgs_lock); |
ce93ec54 | 3521 | |
1bbc621e | 3522 | again: |
1bbc621e CM |
3523 | /* |
3524 | * make sure all the block groups on our dirty list actually | |
3525 | * exist | |
3526 | */ | |
6c686b35 | 3527 | btrfs_create_pending_block_groups(trans); |
1bbc621e CM |
3528 | |
3529 | if (!path) { | |
3530 | path = btrfs_alloc_path(); | |
3531 | if (!path) | |
3532 | return -ENOMEM; | |
3533 | } | |
3534 | ||
b58d1a9e FM |
3535 | /* |
3536 | * cache_write_mutex is here only to save us from balance or automatic | |
3537 | * removal of empty block groups deleting this block group while we are | |
3538 | * writing out the cache | |
3539 | */ | |
3540 | mutex_lock(&trans->transaction->cache_write_mutex); | |
1bbc621e | 3541 | while (!list_empty(&dirty)) { |
ba2c4d4e JB |
3542 | bool drop_reserve = true; |
3543 | ||
1bbc621e CM |
3544 | cache = list_first_entry(&dirty, |
3545 | struct btrfs_block_group_cache, | |
3546 | dirty_list); | |
1bbc621e CM |
3547 | /* |
3548 | * this can happen if something re-dirties a block | |
3549 | * group that is already under IO. Just wait for it to | |
3550 | * finish and then do it all again | |
3551 | */ | |
3552 | if (!list_empty(&cache->io_list)) { | |
3553 | list_del_init(&cache->io_list); | |
afdb5718 | 3554 | btrfs_wait_cache_io(trans, cache, path); |
1bbc621e CM |
3555 | btrfs_put_block_group(cache); |
3556 | } | |
3557 | ||
3558 | ||
3559 | /* | |
3560 | * btrfs_wait_cache_io uses the cache->dirty_list to decide | |
3561 | * if it should update the cache_state. Don't delete | |
3562 | * until after we wait. | |
3563 | * | |
3564 | * Since we're not running in the commit critical section | |
3565 | * we need the dirty_bgs_lock to protect from update_block_group | |
3566 | */ | |
3567 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3568 | list_del_init(&cache->dirty_list); | |
3569 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3570 | ||
3571 | should_put = 1; | |
3572 | ||
3573 | cache_save_setup(cache, trans, path); | |
3574 | ||
3575 | if (cache->disk_cache_state == BTRFS_DC_SETUP) { | |
3576 | cache->io_ctl.inode = NULL; | |
fe041534 | 3577 | ret = btrfs_write_out_cache(trans, cache, path); |
1bbc621e CM |
3578 | if (ret == 0 && cache->io_ctl.inode) { |
3579 | num_started++; | |
3580 | should_put = 0; | |
3581 | ||
3582 | /* | |
45ae2c18 NB |
3583 | * The cache_write_mutex is protecting the |
3584 | * io_list, also refer to the definition of | |
3585 | * btrfs_transaction::io_bgs for more details | |
1bbc621e CM |
3586 | */ |
3587 | list_add_tail(&cache->io_list, io); | |
3588 | } else { | |
3589 | /* | |
3590 | * if we failed to write the cache, the | |
3591 | * generation will be bad and life goes on | |
3592 | */ | |
3593 | ret = 0; | |
3594 | } | |
3595 | } | |
ff1f8250 | 3596 | if (!ret) { |
39db232d | 3597 | ret = write_one_cache_group(trans, path, cache); |
ff1f8250 FM |
3598 | /* |
3599 | * Our block group might still be attached to the list | |
3600 | * of new block groups in the transaction handle of some | |
3601 | * other task (struct btrfs_trans_handle->new_bgs). This | |
3602 | * means its block group item isn't yet in the extent | |
3603 | * tree. If this happens ignore the error, as we will | |
3604 | * try again later in the critical section of the | |
3605 | * transaction commit. | |
3606 | */ | |
3607 | if (ret == -ENOENT) { | |
3608 | ret = 0; | |
3609 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3610 | if (list_empty(&cache->dirty_list)) { | |
3611 | list_add_tail(&cache->dirty_list, | |
3612 | &cur_trans->dirty_bgs); | |
3613 | btrfs_get_block_group(cache); | |
ba2c4d4e | 3614 | drop_reserve = false; |
ff1f8250 FM |
3615 | } |
3616 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3617 | } else if (ret) { | |
66642832 | 3618 | btrfs_abort_transaction(trans, ret); |
ff1f8250 FM |
3619 | } |
3620 | } | |
1bbc621e | 3621 | |
52042d8e | 3622 | /* if it's not on the io list, we need to put the block group */ |
1bbc621e CM |
3623 | if (should_put) |
3624 | btrfs_put_block_group(cache); | |
ba2c4d4e JB |
3625 | if (drop_reserve) |
3626 | btrfs_delayed_refs_rsv_release(fs_info, 1); | |
1bbc621e CM |
3627 | |
3628 | if (ret) | |
3629 | break; | |
b58d1a9e FM |
3630 | |
3631 | /* | |
3632 | * Avoid blocking other tasks for too long. It might even save | |
3633 | * us from writing caches for block groups that are going to be | |
3634 | * removed. | |
3635 | */ | |
3636 | mutex_unlock(&trans->transaction->cache_write_mutex); | |
3637 | mutex_lock(&trans->transaction->cache_write_mutex); | |
1bbc621e | 3638 | } |
b58d1a9e | 3639 | mutex_unlock(&trans->transaction->cache_write_mutex); |
1bbc621e CM |
3640 | |
3641 | /* | |
3642 | * go through delayed refs for all the stuff we've just kicked off | |
3643 | * and then loop back (just once) | |
3644 | */ | |
c79a70b1 | 3645 | ret = btrfs_run_delayed_refs(trans, 0); |
1bbc621e CM |
3646 | if (!ret && loops == 0) { |
3647 | loops++; | |
3648 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3649 | list_splice_init(&cur_trans->dirty_bgs, &dirty); | |
b58d1a9e FM |
3650 | /* |
3651 | * dirty_bgs_lock protects us from concurrent block group | |
3652 | * deletes too (not just cache_write_mutex). | |
3653 | */ | |
3654 | if (!list_empty(&dirty)) { | |
3655 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3656 | goto again; | |
3657 | } | |
1bbc621e | 3658 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c79a1751 | 3659 | } else if (ret < 0) { |
2ff7e61e | 3660 | btrfs_cleanup_dirty_bgs(cur_trans, fs_info); |
1bbc621e CM |
3661 | } |
3662 | ||
3663 | btrfs_free_path(path); | |
3664 | return ret; | |
3665 | } | |
3666 | ||
5742d15f | 3667 | int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans) |
1bbc621e | 3668 | { |
5742d15f | 3669 | struct btrfs_fs_info *fs_info = trans->fs_info; |
1bbc621e CM |
3670 | struct btrfs_block_group_cache *cache; |
3671 | struct btrfs_transaction *cur_trans = trans->transaction; | |
3672 | int ret = 0; | |
3673 | int should_put; | |
3674 | struct btrfs_path *path; | |
3675 | struct list_head *io = &cur_trans->io_bgs; | |
3676 | int num_started = 0; | |
9078a3e1 CM |
3677 | |
3678 | path = btrfs_alloc_path(); | |
3679 | if (!path) | |
3680 | return -ENOMEM; | |
3681 | ||
ce93ec54 | 3682 | /* |
e44081ef FM |
3683 | * Even though we are in the critical section of the transaction commit, |
3684 | * we can still have concurrent tasks adding elements to this | |
3685 | * transaction's list of dirty block groups. These tasks correspond to | |
3686 | * endio free space workers started when writeback finishes for a | |
3687 | * space cache, which run inode.c:btrfs_finish_ordered_io(), and can | |
3688 | * allocate new block groups as a result of COWing nodes of the root | |
3689 | * tree when updating the free space inode. The writeback for the space | |
3690 | * caches is triggered by an earlier call to | |
3691 | * btrfs_start_dirty_block_groups() and iterations of the following | |
3692 | * loop. | |
3693 | * Also we want to do the cache_save_setup first and then run the | |
ce93ec54 JB |
3694 | * delayed refs to make sure we have the best chance at doing this all |
3695 | * in one shot. | |
3696 | */ | |
e44081ef | 3697 | spin_lock(&cur_trans->dirty_bgs_lock); |
ce93ec54 JB |
3698 | while (!list_empty(&cur_trans->dirty_bgs)) { |
3699 | cache = list_first_entry(&cur_trans->dirty_bgs, | |
3700 | struct btrfs_block_group_cache, | |
3701 | dirty_list); | |
c9dc4c65 CM |
3702 | |
3703 | /* | |
3704 | * this can happen if cache_save_setup re-dirties a block | |
3705 | * group that is already under IO. Just wait for it to | |
3706 | * finish and then do it all again | |
3707 | */ | |
3708 | if (!list_empty(&cache->io_list)) { | |
e44081ef | 3709 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 | 3710 | list_del_init(&cache->io_list); |
afdb5718 | 3711 | btrfs_wait_cache_io(trans, cache, path); |
c9dc4c65 | 3712 | btrfs_put_block_group(cache); |
e44081ef | 3713 | spin_lock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 CM |
3714 | } |
3715 | ||
1bbc621e CM |
3716 | /* |
3717 | * don't remove from the dirty list until after we've waited | |
3718 | * on any pending IO | |
3719 | */ | |
ce93ec54 | 3720 | list_del_init(&cache->dirty_list); |
e44081ef | 3721 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 CM |
3722 | should_put = 1; |
3723 | ||
1bbc621e | 3724 | cache_save_setup(cache, trans, path); |
c9dc4c65 | 3725 | |
ce93ec54 | 3726 | if (!ret) |
c79a70b1 | 3727 | ret = btrfs_run_delayed_refs(trans, |
2ff7e61e | 3728 | (unsigned long) -1); |
c9dc4c65 CM |
3729 | |
3730 | if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) { | |
3731 | cache->io_ctl.inode = NULL; | |
fe041534 | 3732 | ret = btrfs_write_out_cache(trans, cache, path); |
c9dc4c65 CM |
3733 | if (ret == 0 && cache->io_ctl.inode) { |
3734 | num_started++; | |
3735 | should_put = 0; | |
1bbc621e | 3736 | list_add_tail(&cache->io_list, io); |
c9dc4c65 CM |
3737 | } else { |
3738 | /* | |
3739 | * if we failed to write the cache, the | |
3740 | * generation will be bad and life goes on | |
3741 | */ | |
3742 | ret = 0; | |
3743 | } | |
3744 | } | |
ff1f8250 | 3745 | if (!ret) { |
39db232d | 3746 | ret = write_one_cache_group(trans, path, cache); |
2bc0bb5f FM |
3747 | /* |
3748 | * One of the free space endio workers might have | |
3749 | * created a new block group while updating a free space | |
3750 | * cache's inode (at inode.c:btrfs_finish_ordered_io()) | |
3751 | * and hasn't released its transaction handle yet, in | |
3752 | * which case the new block group is still attached to | |
3753 | * its transaction handle and its creation has not | |
3754 | * finished yet (no block group item in the extent tree | |
3755 | * yet, etc). If this is the case, wait for all free | |
3756 | * space endio workers to finish and retry. This is a | |
3757 | * a very rare case so no need for a more efficient and | |
3758 | * complex approach. | |
3759 | */ | |
3760 | if (ret == -ENOENT) { | |
3761 | wait_event(cur_trans->writer_wait, | |
3762 | atomic_read(&cur_trans->num_writers) == 1); | |
39db232d | 3763 | ret = write_one_cache_group(trans, path, cache); |
2bc0bb5f | 3764 | } |
ff1f8250 | 3765 | if (ret) |
66642832 | 3766 | btrfs_abort_transaction(trans, ret); |
ff1f8250 | 3767 | } |
c9dc4c65 CM |
3768 | |
3769 | /* if its not on the io list, we need to put the block group */ | |
3770 | if (should_put) | |
3771 | btrfs_put_block_group(cache); | |
ba2c4d4e | 3772 | btrfs_delayed_refs_rsv_release(fs_info, 1); |
e44081ef | 3773 | spin_lock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 | 3774 | } |
e44081ef | 3775 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 | 3776 | |
45ae2c18 NB |
3777 | /* |
3778 | * Refer to the definition of io_bgs member for details why it's safe | |
3779 | * to use it without any locking | |
3780 | */ | |
1bbc621e CM |
3781 | while (!list_empty(io)) { |
3782 | cache = list_first_entry(io, struct btrfs_block_group_cache, | |
c9dc4c65 CM |
3783 | io_list); |
3784 | list_del_init(&cache->io_list); | |
afdb5718 | 3785 | btrfs_wait_cache_io(trans, cache, path); |
0cb59c99 JB |
3786 | btrfs_put_block_group(cache); |
3787 | } | |
3788 | ||
9078a3e1 | 3789 | btrfs_free_path(path); |
ce93ec54 | 3790 | return ret; |
9078a3e1 CM |
3791 | } |
3792 | ||
2ff7e61e | 3793 | int btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr) |
d2fb3437 YZ |
3794 | { |
3795 | struct btrfs_block_group_cache *block_group; | |
3796 | int readonly = 0; | |
3797 | ||
0b246afa | 3798 | block_group = btrfs_lookup_block_group(fs_info, bytenr); |
d2fb3437 YZ |
3799 | if (!block_group || block_group->ro) |
3800 | readonly = 1; | |
3801 | if (block_group) | |
fa9c0d79 | 3802 | btrfs_put_block_group(block_group); |
d2fb3437 YZ |
3803 | return readonly; |
3804 | } | |
3805 | ||
f78c436c FM |
3806 | bool btrfs_inc_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr) |
3807 | { | |
3808 | struct btrfs_block_group_cache *bg; | |
3809 | bool ret = true; | |
3810 | ||
3811 | bg = btrfs_lookup_block_group(fs_info, bytenr); | |
3812 | if (!bg) | |
3813 | return false; | |
3814 | ||
3815 | spin_lock(&bg->lock); | |
3816 | if (bg->ro) | |
3817 | ret = false; | |
3818 | else | |
3819 | atomic_inc(&bg->nocow_writers); | |
3820 | spin_unlock(&bg->lock); | |
3821 | ||
3822 | /* no put on block group, done by btrfs_dec_nocow_writers */ | |
3823 | if (!ret) | |
3824 | btrfs_put_block_group(bg); | |
3825 | ||
3826 | return ret; | |
3827 | ||
3828 | } | |
3829 | ||
3830 | void btrfs_dec_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr) | |
3831 | { | |
3832 | struct btrfs_block_group_cache *bg; | |
3833 | ||
3834 | bg = btrfs_lookup_block_group(fs_info, bytenr); | |
3835 | ASSERT(bg); | |
3836 | if (atomic_dec_and_test(&bg->nocow_writers)) | |
4625956a | 3837 | wake_up_var(&bg->nocow_writers); |
f78c436c FM |
3838 | /* |
3839 | * Once for our lookup and once for the lookup done by a previous call | |
3840 | * to btrfs_inc_nocow_writers() | |
3841 | */ | |
3842 | btrfs_put_block_group(bg); | |
3843 | btrfs_put_block_group(bg); | |
3844 | } | |
3845 | ||
f78c436c FM |
3846 | void btrfs_wait_nocow_writers(struct btrfs_block_group_cache *bg) |
3847 | { | |
4625956a | 3848 | wait_var_event(&bg->nocow_writers, !atomic_read(&bg->nocow_writers)); |
f78c436c FM |
3849 | } |
3850 | ||
6ab0a202 JM |
3851 | static const char *alloc_name(u64 flags) |
3852 | { | |
3853 | switch (flags) { | |
3854 | case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA: | |
3855 | return "mixed"; | |
3856 | case BTRFS_BLOCK_GROUP_METADATA: | |
3857 | return "metadata"; | |
3858 | case BTRFS_BLOCK_GROUP_DATA: | |
3859 | return "data"; | |
3860 | case BTRFS_BLOCK_GROUP_SYSTEM: | |
3861 | return "system"; | |
3862 | default: | |
3863 | WARN_ON(1); | |
3864 | return "invalid-combination"; | |
3865 | }; | |
3866 | } | |
3867 | ||
4ca61683 | 3868 | static int create_space_info(struct btrfs_fs_info *info, u64 flags) |
2be12ef7 NB |
3869 | { |
3870 | ||
3871 | struct btrfs_space_info *space_info; | |
3872 | int i; | |
3873 | int ret; | |
3874 | ||
3875 | space_info = kzalloc(sizeof(*space_info), GFP_NOFS); | |
3876 | if (!space_info) | |
3877 | return -ENOMEM; | |
3878 | ||
3879 | ret = percpu_counter_init(&space_info->total_bytes_pinned, 0, | |
3880 | GFP_KERNEL); | |
3881 | if (ret) { | |
3882 | kfree(space_info); | |
3883 | return ret; | |
3884 | } | |
3885 | ||
3886 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) | |
3887 | INIT_LIST_HEAD(&space_info->block_groups[i]); | |
3888 | init_rwsem(&space_info->groups_sem); | |
3889 | spin_lock_init(&space_info->lock); | |
3890 | space_info->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK; | |
3891 | space_info->force_alloc = CHUNK_ALLOC_NO_FORCE; | |
3892 | init_waitqueue_head(&space_info->wait); | |
3893 | INIT_LIST_HEAD(&space_info->ro_bgs); | |
3894 | INIT_LIST_HEAD(&space_info->tickets); | |
3895 | INIT_LIST_HEAD(&space_info->priority_tickets); | |
3896 | ||
3897 | ret = kobject_init_and_add(&space_info->kobj, &space_info_ktype, | |
3898 | info->space_info_kobj, "%s", | |
3899 | alloc_name(space_info->flags)); | |
3900 | if (ret) { | |
450ff834 | 3901 | kobject_put(&space_info->kobj); |
2be12ef7 NB |
3902 | return ret; |
3903 | } | |
3904 | ||
2be12ef7 NB |
3905 | list_add_rcu(&space_info->list, &info->space_info); |
3906 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
3907 | info->data_sinfo = space_info; | |
3908 | ||
3909 | return ret; | |
3910 | } | |
3911 | ||
d2006e6d | 3912 | static void update_space_info(struct btrfs_fs_info *info, u64 flags, |
593060d7 | 3913 | u64 total_bytes, u64 bytes_used, |
e40edf2d | 3914 | u64 bytes_readonly, |
593060d7 CM |
3915 | struct btrfs_space_info **space_info) |
3916 | { | |
3917 | struct btrfs_space_info *found; | |
b742bb82 YZ |
3918 | int factor; |
3919 | ||
46df06b8 | 3920 | factor = btrfs_bg_type_to_factor(flags); |
593060d7 CM |
3921 | |
3922 | found = __find_space_info(info, flags); | |
d2006e6d NB |
3923 | ASSERT(found); |
3924 | spin_lock(&found->lock); | |
3925 | found->total_bytes += total_bytes; | |
3926 | found->disk_total += total_bytes * factor; | |
3927 | found->bytes_used += bytes_used; | |
3928 | found->disk_used += bytes_used * factor; | |
3929 | found->bytes_readonly += bytes_readonly; | |
3930 | if (total_bytes > 0) | |
3931 | found->full = 0; | |
3932 | space_info_add_new_bytes(info, found, total_bytes - | |
3933 | bytes_used - bytes_readonly); | |
3934 | spin_unlock(&found->lock); | |
3935 | *space_info = found; | |
593060d7 CM |
3936 | } |
3937 | ||
8790d502 CM |
3938 | static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
3939 | { | |
899c81ea ID |
3940 | u64 extra_flags = chunk_to_extended(flags) & |
3941 | BTRFS_EXTENDED_PROFILE_MASK; | |
a46d11a8 | 3942 | |
de98ced9 | 3943 | write_seqlock(&fs_info->profiles_lock); |
a46d11a8 ID |
3944 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
3945 | fs_info->avail_data_alloc_bits |= extra_flags; | |
3946 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
3947 | fs_info->avail_metadata_alloc_bits |= extra_flags; | |
3948 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
3949 | fs_info->avail_system_alloc_bits |= extra_flags; | |
de98ced9 | 3950 | write_sequnlock(&fs_info->profiles_lock); |
8790d502 | 3951 | } |
593060d7 | 3952 | |
fc67c450 ID |
3953 | /* |
3954 | * returns target flags in extended format or 0 if restripe for this | |
3955 | * chunk_type is not in progress | |
c6664b42 | 3956 | * |
dccdb07b | 3957 | * should be called with balance_lock held |
fc67c450 ID |
3958 | */ |
3959 | static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags) | |
3960 | { | |
3961 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
3962 | u64 target = 0; | |
3963 | ||
fc67c450 ID |
3964 | if (!bctl) |
3965 | return 0; | |
3966 | ||
3967 | if (flags & BTRFS_BLOCK_GROUP_DATA && | |
3968 | bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
3969 | target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target; | |
3970 | } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM && | |
3971 | bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
3972 | target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target; | |
3973 | } else if (flags & BTRFS_BLOCK_GROUP_METADATA && | |
3974 | bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
3975 | target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target; | |
3976 | } | |
3977 | ||
3978 | return target; | |
3979 | } | |
3980 | ||
a46d11a8 ID |
3981 | /* |
3982 | * @flags: available profiles in extended format (see ctree.h) | |
3983 | * | |
e4d8ec0f ID |
3984 | * Returns reduced profile in chunk format. If profile changing is in |
3985 | * progress (either running or paused) picks the target profile (if it's | |
3986 | * already available), otherwise falls back to plain reducing. | |
a46d11a8 | 3987 | */ |
2ff7e61e | 3988 | static u64 btrfs_reduce_alloc_profile(struct btrfs_fs_info *fs_info, u64 flags) |
ec44a35c | 3989 | { |
0b246afa | 3990 | u64 num_devices = fs_info->fs_devices->rw_devices; |
fc67c450 | 3991 | u64 target; |
9c170b26 ZL |
3992 | u64 raid_type; |
3993 | u64 allowed = 0; | |
a061fc8d | 3994 | |
fc67c450 ID |
3995 | /* |
3996 | * see if restripe for this chunk_type is in progress, if so | |
3997 | * try to reduce to the target profile | |
3998 | */ | |
0b246afa JM |
3999 | spin_lock(&fs_info->balance_lock); |
4000 | target = get_restripe_target(fs_info, flags); | |
fc67c450 ID |
4001 | if (target) { |
4002 | /* pick target profile only if it's already available */ | |
4003 | if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) { | |
0b246afa | 4004 | spin_unlock(&fs_info->balance_lock); |
fc67c450 | 4005 | return extended_to_chunk(target); |
e4d8ec0f ID |
4006 | } |
4007 | } | |
0b246afa | 4008 | spin_unlock(&fs_info->balance_lock); |
e4d8ec0f | 4009 | |
53b381b3 | 4010 | /* First, mask out the RAID levels which aren't possible */ |
9c170b26 ZL |
4011 | for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) { |
4012 | if (num_devices >= btrfs_raid_array[raid_type].devs_min) | |
41a6e891 | 4013 | allowed |= btrfs_raid_array[raid_type].bg_flag; |
9c170b26 ZL |
4014 | } |
4015 | allowed &= flags; | |
4016 | ||
4017 | if (allowed & BTRFS_BLOCK_GROUP_RAID6) | |
4018 | allowed = BTRFS_BLOCK_GROUP_RAID6; | |
4019 | else if (allowed & BTRFS_BLOCK_GROUP_RAID5) | |
4020 | allowed = BTRFS_BLOCK_GROUP_RAID5; | |
4021 | else if (allowed & BTRFS_BLOCK_GROUP_RAID10) | |
4022 | allowed = BTRFS_BLOCK_GROUP_RAID10; | |
4023 | else if (allowed & BTRFS_BLOCK_GROUP_RAID1) | |
4024 | allowed = BTRFS_BLOCK_GROUP_RAID1; | |
4025 | else if (allowed & BTRFS_BLOCK_GROUP_RAID0) | |
4026 | allowed = BTRFS_BLOCK_GROUP_RAID0; | |
4027 | ||
4028 | flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK; | |
4029 | ||
4030 | return extended_to_chunk(flags | allowed); | |
ec44a35c CM |
4031 | } |
4032 | ||
2ff7e61e | 4033 | static u64 get_alloc_profile(struct btrfs_fs_info *fs_info, u64 orig_flags) |
6a63209f | 4034 | { |
de98ced9 | 4035 | unsigned seq; |
f8213bdc | 4036 | u64 flags; |
de98ced9 MX |
4037 | |
4038 | do { | |
f8213bdc | 4039 | flags = orig_flags; |
0b246afa | 4040 | seq = read_seqbegin(&fs_info->profiles_lock); |
de98ced9 MX |
4041 | |
4042 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
0b246afa | 4043 | flags |= fs_info->avail_data_alloc_bits; |
de98ced9 | 4044 | else if (flags & BTRFS_BLOCK_GROUP_SYSTEM) |
0b246afa | 4045 | flags |= fs_info->avail_system_alloc_bits; |
de98ced9 | 4046 | else if (flags & BTRFS_BLOCK_GROUP_METADATA) |
0b246afa JM |
4047 | flags |= fs_info->avail_metadata_alloc_bits; |
4048 | } while (read_seqretry(&fs_info->profiles_lock, seq)); | |
6fef8df1 | 4049 | |
2ff7e61e | 4050 | return btrfs_reduce_alloc_profile(fs_info, flags); |
6a63209f JB |
4051 | } |
4052 | ||
1b86826d | 4053 | static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data) |
9ed74f2d | 4054 | { |
0b246afa | 4055 | struct btrfs_fs_info *fs_info = root->fs_info; |
b742bb82 | 4056 | u64 flags; |
53b381b3 | 4057 | u64 ret; |
9ed74f2d | 4058 | |
b742bb82 YZ |
4059 | if (data) |
4060 | flags = BTRFS_BLOCK_GROUP_DATA; | |
0b246afa | 4061 | else if (root == fs_info->chunk_root) |
b742bb82 | 4062 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
9ed74f2d | 4063 | else |
b742bb82 | 4064 | flags = BTRFS_BLOCK_GROUP_METADATA; |
9ed74f2d | 4065 | |
2ff7e61e | 4066 | ret = get_alloc_profile(fs_info, flags); |
53b381b3 | 4067 | return ret; |
6a63209f | 4068 | } |
9ed74f2d | 4069 | |
1b86826d JM |
4070 | u64 btrfs_data_alloc_profile(struct btrfs_fs_info *fs_info) |
4071 | { | |
4072 | return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_DATA); | |
4073 | } | |
4074 | ||
4075 | u64 btrfs_metadata_alloc_profile(struct btrfs_fs_info *fs_info) | |
4076 | { | |
4077 | return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_METADATA); | |
4078 | } | |
4079 | ||
4080 | u64 btrfs_system_alloc_profile(struct btrfs_fs_info *fs_info) | |
4081 | { | |
4082 | return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); | |
4083 | } | |
4084 | ||
4136135b LB |
4085 | static u64 btrfs_space_info_used(struct btrfs_space_info *s_info, |
4086 | bool may_use_included) | |
4087 | { | |
4088 | ASSERT(s_info); | |
4089 | return s_info->bytes_used + s_info->bytes_reserved + | |
4090 | s_info->bytes_pinned + s_info->bytes_readonly + | |
4091 | (may_use_included ? s_info->bytes_may_use : 0); | |
4092 | } | |
4093 | ||
04f4f916 | 4094 | int btrfs_alloc_data_chunk_ondemand(struct btrfs_inode *inode, u64 bytes) |
6a63209f | 4095 | { |
04f4f916 | 4096 | struct btrfs_root *root = inode->root; |
b4d7c3c9 | 4097 | struct btrfs_fs_info *fs_info = root->fs_info; |
1174cade | 4098 | struct btrfs_space_info *data_sinfo = fs_info->data_sinfo; |
ab6e2410 | 4099 | u64 used; |
94b947b2 | 4100 | int ret = 0; |
c99f1b0c ZL |
4101 | int need_commit = 2; |
4102 | int have_pinned_space; | |
6a63209f | 4103 | |
6a63209f | 4104 | /* make sure bytes are sectorsize aligned */ |
0b246afa | 4105 | bytes = ALIGN(bytes, fs_info->sectorsize); |
6a63209f | 4106 | |
9dced186 | 4107 | if (btrfs_is_free_space_inode(inode)) { |
c99f1b0c | 4108 | need_commit = 0; |
9dced186 | 4109 | ASSERT(current->journal_info); |
0af3d00b JB |
4110 | } |
4111 | ||
6a63209f JB |
4112 | again: |
4113 | /* make sure we have enough space to handle the data first */ | |
4114 | spin_lock(&data_sinfo->lock); | |
4136135b | 4115 | used = btrfs_space_info_used(data_sinfo, true); |
ab6e2410 JB |
4116 | |
4117 | if (used + bytes > data_sinfo->total_bytes) { | |
4e06bdd6 | 4118 | struct btrfs_trans_handle *trans; |
9ed74f2d | 4119 | |
6a63209f JB |
4120 | /* |
4121 | * if we don't have enough free bytes in this space then we need | |
4122 | * to alloc a new chunk. | |
4123 | */ | |
b9fd47cd | 4124 | if (!data_sinfo->full) { |
6a63209f | 4125 | u64 alloc_target; |
9ed74f2d | 4126 | |
0e4f8f88 | 4127 | data_sinfo->force_alloc = CHUNK_ALLOC_FORCE; |
6a63209f | 4128 | spin_unlock(&data_sinfo->lock); |
1174cade | 4129 | |
1b86826d | 4130 | alloc_target = btrfs_data_alloc_profile(fs_info); |
9dced186 MX |
4131 | /* |
4132 | * It is ugly that we don't call nolock join | |
4133 | * transaction for the free space inode case here. | |
4134 | * But it is safe because we only do the data space | |
4135 | * reservation for the free space cache in the | |
4136 | * transaction context, the common join transaction | |
4137 | * just increase the counter of the current transaction | |
4138 | * handler, doesn't try to acquire the trans_lock of | |
4139 | * the fs. | |
4140 | */ | |
7a7eaa40 | 4141 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
4142 | if (IS_ERR(trans)) |
4143 | return PTR_ERR(trans); | |
9ed74f2d | 4144 | |
01458828 | 4145 | ret = do_chunk_alloc(trans, alloc_target, |
0e4f8f88 | 4146 | CHUNK_ALLOC_NO_FORCE); |
3a45bb20 | 4147 | btrfs_end_transaction(trans); |
d52a5b5f MX |
4148 | if (ret < 0) { |
4149 | if (ret != -ENOSPC) | |
4150 | return ret; | |
c99f1b0c ZL |
4151 | else { |
4152 | have_pinned_space = 1; | |
d52a5b5f | 4153 | goto commit_trans; |
c99f1b0c | 4154 | } |
d52a5b5f | 4155 | } |
9ed74f2d | 4156 | |
6a63209f JB |
4157 | goto again; |
4158 | } | |
f2bb8f5c JB |
4159 | |
4160 | /* | |
b150a4f1 | 4161 | * If we don't have enough pinned space to deal with this |
94b947b2 ZL |
4162 | * allocation, and no removed chunk in current transaction, |
4163 | * don't bother committing the transaction. | |
f2bb8f5c | 4164 | */ |
dec59fa3 | 4165 | have_pinned_space = __percpu_counter_compare( |
c99f1b0c | 4166 | &data_sinfo->total_bytes_pinned, |
dec59fa3 EL |
4167 | used + bytes - data_sinfo->total_bytes, |
4168 | BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
6a63209f | 4169 | spin_unlock(&data_sinfo->lock); |
6a63209f | 4170 | |
4e06bdd6 | 4171 | /* commit the current transaction and try again */ |
d52a5b5f | 4172 | commit_trans: |
92e2f7e3 | 4173 | if (need_commit) { |
c99f1b0c | 4174 | need_commit--; |
b150a4f1 | 4175 | |
e1746e83 | 4176 | if (need_commit > 0) { |
82b3e53b | 4177 | btrfs_start_delalloc_roots(fs_info, -1); |
6374e57a | 4178 | btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, |
0b246afa | 4179 | (u64)-1); |
e1746e83 | 4180 | } |
9a4e7276 | 4181 | |
7a7eaa40 | 4182 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
4183 | if (IS_ERR(trans)) |
4184 | return PTR_ERR(trans); | |
c99f1b0c | 4185 | if (have_pinned_space >= 0 || |
3204d33c JB |
4186 | test_bit(BTRFS_TRANS_HAVE_FREE_BGS, |
4187 | &trans->transaction->flags) || | |
c99f1b0c | 4188 | need_commit > 0) { |
3a45bb20 | 4189 | ret = btrfs_commit_transaction(trans); |
94b947b2 ZL |
4190 | if (ret) |
4191 | return ret; | |
d7c15171 | 4192 | /* |
c2d6cb16 FM |
4193 | * The cleaner kthread might still be doing iput |
4194 | * operations. Wait for it to finish so that | |
034f784d JB |
4195 | * more space is released. We don't need to |
4196 | * explicitly run the delayed iputs here because | |
4197 | * the commit_transaction would have woken up | |
4198 | * the cleaner. | |
d7c15171 | 4199 | */ |
034f784d JB |
4200 | ret = btrfs_wait_on_delayed_iputs(fs_info); |
4201 | if (ret) | |
4202 | return ret; | |
94b947b2 ZL |
4203 | goto again; |
4204 | } else { | |
3a45bb20 | 4205 | btrfs_end_transaction(trans); |
94b947b2 | 4206 | } |
4e06bdd6 | 4207 | } |
9ed74f2d | 4208 | |
0b246afa | 4209 | trace_btrfs_space_reservation(fs_info, |
cab45e22 JM |
4210 | "space_info:enospc", |
4211 | data_sinfo->flags, bytes, 1); | |
6a63209f JB |
4212 | return -ENOSPC; |
4213 | } | |
480b9b4d | 4214 | update_bytes_may_use(fs_info, data_sinfo, bytes); |
0b246afa | 4215 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 4216 | data_sinfo->flags, bytes, 1); |
6a63209f | 4217 | spin_unlock(&data_sinfo->lock); |
6a63209f | 4218 | |
4559b0a7 | 4219 | return 0; |
9ed74f2d | 4220 | } |
6a63209f | 4221 | |
364ecf36 QW |
4222 | int btrfs_check_data_free_space(struct inode *inode, |
4223 | struct extent_changeset **reserved, u64 start, u64 len) | |
4ceff079 | 4224 | { |
0b246afa | 4225 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4ceff079 QW |
4226 | int ret; |
4227 | ||
4228 | /* align the range */ | |
0b246afa JM |
4229 | len = round_up(start + len, fs_info->sectorsize) - |
4230 | round_down(start, fs_info->sectorsize); | |
4231 | start = round_down(start, fs_info->sectorsize); | |
4ceff079 | 4232 | |
04f4f916 | 4233 | ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode), len); |
4ceff079 QW |
4234 | if (ret < 0) |
4235 | return ret; | |
4236 | ||
1e5ec2e7 | 4237 | /* Use new btrfs_qgroup_reserve_data to reserve precious data space. */ |
364ecf36 | 4238 | ret = btrfs_qgroup_reserve_data(inode, reserved, start, len); |
7bc329c1 | 4239 | if (ret < 0) |
1e5ec2e7 | 4240 | btrfs_free_reserved_data_space_noquota(inode, start, len); |
364ecf36 QW |
4241 | else |
4242 | ret = 0; | |
4ceff079 QW |
4243 | return ret; |
4244 | } | |
4245 | ||
4ceff079 QW |
4246 | /* |
4247 | * Called if we need to clear a data reservation for this inode | |
4248 | * Normally in a error case. | |
4249 | * | |
51773bec QW |
4250 | * This one will *NOT* use accurate qgroup reserved space API, just for case |
4251 | * which we can't sleep and is sure it won't affect qgroup reserved space. | |
4252 | * Like clear_bit_hook(). | |
4ceff079 | 4253 | */ |
51773bec QW |
4254 | void btrfs_free_reserved_data_space_noquota(struct inode *inode, u64 start, |
4255 | u64 len) | |
4ceff079 | 4256 | { |
0b246afa | 4257 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4ceff079 QW |
4258 | struct btrfs_space_info *data_sinfo; |
4259 | ||
4260 | /* Make sure the range is aligned to sectorsize */ | |
0b246afa JM |
4261 | len = round_up(start + len, fs_info->sectorsize) - |
4262 | round_down(start, fs_info->sectorsize); | |
4263 | start = round_down(start, fs_info->sectorsize); | |
4ceff079 | 4264 | |
0b246afa | 4265 | data_sinfo = fs_info->data_sinfo; |
4ceff079 | 4266 | spin_lock(&data_sinfo->lock); |
480b9b4d | 4267 | update_bytes_may_use(fs_info, data_sinfo, -len); |
0b246afa | 4268 | trace_btrfs_space_reservation(fs_info, "space_info", |
4ceff079 QW |
4269 | data_sinfo->flags, len, 0); |
4270 | spin_unlock(&data_sinfo->lock); | |
4271 | } | |
4272 | ||
51773bec QW |
4273 | /* |
4274 | * Called if we need to clear a data reservation for this inode | |
4275 | * Normally in a error case. | |
4276 | * | |
01327610 | 4277 | * This one will handle the per-inode data rsv map for accurate reserved |
51773bec QW |
4278 | * space framework. |
4279 | */ | |
bc42bda2 QW |
4280 | void btrfs_free_reserved_data_space(struct inode *inode, |
4281 | struct extent_changeset *reserved, u64 start, u64 len) | |
51773bec | 4282 | { |
0c476a5d JM |
4283 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4284 | ||
4285 | /* Make sure the range is aligned to sectorsize */ | |
da17066c JM |
4286 | len = round_up(start + len, root->fs_info->sectorsize) - |
4287 | round_down(start, root->fs_info->sectorsize); | |
4288 | start = round_down(start, root->fs_info->sectorsize); | |
0c476a5d | 4289 | |
51773bec | 4290 | btrfs_free_reserved_data_space_noquota(inode, start, len); |
bc42bda2 | 4291 | btrfs_qgroup_free_data(inode, reserved, start, len); |
51773bec QW |
4292 | } |
4293 | ||
97e728d4 | 4294 | static void force_metadata_allocation(struct btrfs_fs_info *info) |
e3ccfa98 | 4295 | { |
97e728d4 JB |
4296 | struct list_head *head = &info->space_info; |
4297 | struct btrfs_space_info *found; | |
e3ccfa98 | 4298 | |
97e728d4 JB |
4299 | rcu_read_lock(); |
4300 | list_for_each_entry_rcu(found, head, list) { | |
4301 | if (found->flags & BTRFS_BLOCK_GROUP_METADATA) | |
0e4f8f88 | 4302 | found->force_alloc = CHUNK_ALLOC_FORCE; |
e3ccfa98 | 4303 | } |
97e728d4 | 4304 | rcu_read_unlock(); |
e3ccfa98 JB |
4305 | } |
4306 | ||
3c76cd84 MX |
4307 | static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global) |
4308 | { | |
4309 | return (global->size << 1); | |
4310 | } | |
4311 | ||
2ff7e61e | 4312 | static int should_alloc_chunk(struct btrfs_fs_info *fs_info, |
698d0082 | 4313 | struct btrfs_space_info *sinfo, int force) |
32c00aff | 4314 | { |
8d8aafee | 4315 | u64 bytes_used = btrfs_space_info_used(sinfo, false); |
e5bc2458 | 4316 | u64 thresh; |
e3ccfa98 | 4317 | |
0e4f8f88 CM |
4318 | if (force == CHUNK_ALLOC_FORCE) |
4319 | return 1; | |
4320 | ||
4321 | /* | |
4322 | * in limited mode, we want to have some free space up to | |
4323 | * about 1% of the FS size. | |
4324 | */ | |
4325 | if (force == CHUNK_ALLOC_LIMITED) { | |
0b246afa | 4326 | thresh = btrfs_super_total_bytes(fs_info->super_copy); |
ee22184b | 4327 | thresh = max_t(u64, SZ_64M, div_factor_fine(thresh, 1)); |
0e4f8f88 | 4328 | |
8d8aafee | 4329 | if (sinfo->total_bytes - bytes_used < thresh) |
0e4f8f88 CM |
4330 | return 1; |
4331 | } | |
0e4f8f88 | 4332 | |
8d8aafee | 4333 | if (bytes_used + SZ_2M < div_factor(sinfo->total_bytes, 8)) |
14ed0ca6 | 4334 | return 0; |
424499db | 4335 | return 1; |
32c00aff JB |
4336 | } |
4337 | ||
2ff7e61e | 4338 | static u64 get_profile_num_devs(struct btrfs_fs_info *fs_info, u64 type) |
15d1ff81 LB |
4339 | { |
4340 | u64 num_dev; | |
4341 | ||
9fa02ac7 DS |
4342 | num_dev = btrfs_raid_array[btrfs_bg_flags_to_raid_index(type)].devs_max; |
4343 | if (!num_dev) | |
0b246afa | 4344 | num_dev = fs_info->fs_devices->rw_devices; |
15d1ff81 | 4345 | |
39c2d7fa | 4346 | return num_dev; |
15d1ff81 LB |
4347 | } |
4348 | ||
39c2d7fa FM |
4349 | /* |
4350 | * If @is_allocation is true, reserve space in the system space info necessary | |
4351 | * for allocating a chunk, otherwise if it's false, reserve space necessary for | |
4352 | * removing a chunk. | |
4353 | */ | |
451a2c13 | 4354 | void check_system_chunk(struct btrfs_trans_handle *trans, u64 type) |
15d1ff81 | 4355 | { |
451a2c13 | 4356 | struct btrfs_fs_info *fs_info = trans->fs_info; |
15d1ff81 LB |
4357 | struct btrfs_space_info *info; |
4358 | u64 left; | |
4359 | u64 thresh; | |
4fbcdf66 | 4360 | int ret = 0; |
39c2d7fa | 4361 | u64 num_devs; |
4fbcdf66 FM |
4362 | |
4363 | /* | |
4364 | * Needed because we can end up allocating a system chunk and for an | |
4365 | * atomic and race free space reservation in the chunk block reserve. | |
4366 | */ | |
a32bf9a3 | 4367 | lockdep_assert_held(&fs_info->chunk_mutex); |
15d1ff81 | 4368 | |
0b246afa | 4369 | info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); |
15d1ff81 | 4370 | spin_lock(&info->lock); |
4136135b | 4371 | left = info->total_bytes - btrfs_space_info_used(info, true); |
15d1ff81 LB |
4372 | spin_unlock(&info->lock); |
4373 | ||
2ff7e61e | 4374 | num_devs = get_profile_num_devs(fs_info, type); |
39c2d7fa FM |
4375 | |
4376 | /* num_devs device items to update and 1 chunk item to add or remove */ | |
0b246afa JM |
4377 | thresh = btrfs_calc_trunc_metadata_size(fs_info, num_devs) + |
4378 | btrfs_calc_trans_metadata_size(fs_info, 1); | |
39c2d7fa | 4379 | |
0b246afa JM |
4380 | if (left < thresh && btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
4381 | btrfs_info(fs_info, "left=%llu, need=%llu, flags=%llu", | |
4382 | left, thresh, type); | |
4383 | dump_space_info(fs_info, info, 0, 0); | |
15d1ff81 LB |
4384 | } |
4385 | ||
4386 | if (left < thresh) { | |
1b86826d | 4387 | u64 flags = btrfs_system_alloc_profile(fs_info); |
15d1ff81 | 4388 | |
4fbcdf66 FM |
4389 | /* |
4390 | * Ignore failure to create system chunk. We might end up not | |
4391 | * needing it, as we might not need to COW all nodes/leafs from | |
4392 | * the paths we visit in the chunk tree (they were already COWed | |
4393 | * or created in the current transaction for example). | |
4394 | */ | |
c216b203 | 4395 | ret = btrfs_alloc_chunk(trans, flags); |
4fbcdf66 FM |
4396 | } |
4397 | ||
4398 | if (!ret) { | |
0b246afa JM |
4399 | ret = btrfs_block_rsv_add(fs_info->chunk_root, |
4400 | &fs_info->chunk_block_rsv, | |
4fbcdf66 FM |
4401 | thresh, BTRFS_RESERVE_NO_FLUSH); |
4402 | if (!ret) | |
4403 | trans->chunk_bytes_reserved += thresh; | |
15d1ff81 LB |
4404 | } |
4405 | } | |
4406 | ||
28b737f6 LB |
4407 | /* |
4408 | * If force is CHUNK_ALLOC_FORCE: | |
4409 | * - return 1 if it successfully allocates a chunk, | |
4410 | * - return errors including -ENOSPC otherwise. | |
4411 | * If force is NOT CHUNK_ALLOC_FORCE: | |
4412 | * - return 0 if it doesn't need to allocate a new chunk, | |
4413 | * - return 1 if it successfully allocates a chunk, | |
4414 | * - return errors including -ENOSPC otherwise. | |
4415 | */ | |
01458828 NB |
4416 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, u64 flags, |
4417 | int force) | |
9ed74f2d | 4418 | { |
01458828 | 4419 | struct btrfs_fs_info *fs_info = trans->fs_info; |
6324fbf3 | 4420 | struct btrfs_space_info *space_info; |
2556fbb0 NB |
4421 | bool wait_for_alloc = false; |
4422 | bool should_alloc = false; | |
9ed74f2d | 4423 | int ret = 0; |
9ed74f2d | 4424 | |
c6b305a8 JB |
4425 | /* Don't re-enter if we're already allocating a chunk */ |
4426 | if (trans->allocating_chunk) | |
4427 | return -ENOSPC; | |
4428 | ||
0b246afa | 4429 | space_info = __find_space_info(fs_info, flags); |
dc2d3005 | 4430 | ASSERT(space_info); |
9ed74f2d | 4431 | |
2556fbb0 NB |
4432 | do { |
4433 | spin_lock(&space_info->lock); | |
4434 | if (force < space_info->force_alloc) | |
4435 | force = space_info->force_alloc; | |
4436 | should_alloc = should_alloc_chunk(fs_info, space_info, force); | |
4437 | if (space_info->full) { | |
4438 | /* No more free physical space */ | |
4439 | if (should_alloc) | |
4440 | ret = -ENOSPC; | |
4441 | else | |
4442 | ret = 0; | |
4443 | spin_unlock(&space_info->lock); | |
4444 | return ret; | |
4445 | } else if (!should_alloc) { | |
4446 | spin_unlock(&space_info->lock); | |
4447 | return 0; | |
4448 | } else if (space_info->chunk_alloc) { | |
4449 | /* | |
4450 | * Someone is already allocating, so we need to block | |
4451 | * until this someone is finished and then loop to | |
4452 | * recheck if we should continue with our allocation | |
4453 | * attempt. | |
4454 | */ | |
4455 | wait_for_alloc = true; | |
4456 | spin_unlock(&space_info->lock); | |
4457 | mutex_lock(&fs_info->chunk_mutex); | |
4458 | mutex_unlock(&fs_info->chunk_mutex); | |
4459 | } else { | |
4460 | /* Proceed with allocation */ | |
4461 | space_info->chunk_alloc = 1; | |
4462 | wait_for_alloc = false; | |
4463 | spin_unlock(&space_info->lock); | |
4464 | } | |
6d74119f | 4465 | |
1e1c50a9 | 4466 | cond_resched(); |
2556fbb0 | 4467 | } while (wait_for_alloc); |
6d74119f | 4468 | |
2556fbb0 | 4469 | mutex_lock(&fs_info->chunk_mutex); |
c6b305a8 JB |
4470 | trans->allocating_chunk = true; |
4471 | ||
67377734 JB |
4472 | /* |
4473 | * If we have mixed data/metadata chunks we want to make sure we keep | |
4474 | * allocating mixed chunks instead of individual chunks. | |
4475 | */ | |
4476 | if (btrfs_mixed_space_info(space_info)) | |
4477 | flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA); | |
4478 | ||
97e728d4 JB |
4479 | /* |
4480 | * if we're doing a data chunk, go ahead and make sure that | |
4481 | * we keep a reasonable number of metadata chunks allocated in the | |
4482 | * FS as well. | |
4483 | */ | |
9ed74f2d | 4484 | if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) { |
97e728d4 JB |
4485 | fs_info->data_chunk_allocations++; |
4486 | if (!(fs_info->data_chunk_allocations % | |
4487 | fs_info->metadata_ratio)) | |
4488 | force_metadata_allocation(fs_info); | |
9ed74f2d JB |
4489 | } |
4490 | ||
15d1ff81 LB |
4491 | /* |
4492 | * Check if we have enough space in SYSTEM chunk because we may need | |
4493 | * to update devices. | |
4494 | */ | |
451a2c13 | 4495 | check_system_chunk(trans, flags); |
15d1ff81 | 4496 | |
c216b203 | 4497 | ret = btrfs_alloc_chunk(trans, flags); |
c6b305a8 | 4498 | trans->allocating_chunk = false; |
92b8e897 | 4499 | |
9ed74f2d | 4500 | spin_lock(&space_info->lock); |
57f1642e NB |
4501 | if (ret < 0) { |
4502 | if (ret == -ENOSPC) | |
4503 | space_info->full = 1; | |
4504 | else | |
4505 | goto out; | |
4506 | } else { | |
424499db | 4507 | ret = 1; |
21a94f7a | 4508 | space_info->max_extent_size = 0; |
57f1642e | 4509 | } |
6d74119f | 4510 | |
0e4f8f88 | 4511 | space_info->force_alloc = CHUNK_ALLOC_NO_FORCE; |
a81cb9a2 | 4512 | out: |
6d74119f | 4513 | space_info->chunk_alloc = 0; |
9ed74f2d | 4514 | spin_unlock(&space_info->lock); |
a25c75d5 | 4515 | mutex_unlock(&fs_info->chunk_mutex); |
00d80e34 FM |
4516 | /* |
4517 | * When we allocate a new chunk we reserve space in the chunk block | |
4518 | * reserve to make sure we can COW nodes/leafs in the chunk tree or | |
4519 | * add new nodes/leafs to it if we end up needing to do it when | |
4520 | * inserting the chunk item and updating device items as part of the | |
4521 | * second phase of chunk allocation, performed by | |
4522 | * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a | |
4523 | * large number of new block groups to create in our transaction | |
4524 | * handle's new_bgs list to avoid exhausting the chunk block reserve | |
4525 | * in extreme cases - like having a single transaction create many new | |
4526 | * block groups when starting to write out the free space caches of all | |
4527 | * the block groups that were made dirty during the lifetime of the | |
4528 | * transaction. | |
4529 | */ | |
5ce55557 | 4530 | if (trans->chunk_bytes_reserved >= (u64)SZ_2M) |
6c686b35 | 4531 | btrfs_create_pending_block_groups(trans); |
5ce55557 | 4532 | |
0f9dd46c | 4533 | return ret; |
6324fbf3 | 4534 | } |
9ed74f2d | 4535 | |
c1c4919b | 4536 | static int can_overcommit(struct btrfs_fs_info *fs_info, |
a80c8dcf | 4537 | struct btrfs_space_info *space_info, u64 bytes, |
c1c4919b JM |
4538 | enum btrfs_reserve_flush_enum flush, |
4539 | bool system_chunk) | |
a80c8dcf | 4540 | { |
0b246afa | 4541 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; |
957780eb | 4542 | u64 profile; |
3c76cd84 | 4543 | u64 space_size; |
a80c8dcf JB |
4544 | u64 avail; |
4545 | u64 used; | |
46df06b8 | 4546 | int factor; |
a80c8dcf | 4547 | |
957780eb JB |
4548 | /* Don't overcommit when in mixed mode. */ |
4549 | if (space_info->flags & BTRFS_BLOCK_GROUP_DATA) | |
4550 | return 0; | |
4551 | ||
c1c4919b JM |
4552 | if (system_chunk) |
4553 | profile = btrfs_system_alloc_profile(fs_info); | |
4554 | else | |
4555 | profile = btrfs_metadata_alloc_profile(fs_info); | |
4556 | ||
4136135b | 4557 | used = btrfs_space_info_used(space_info, false); |
96f1bb57 | 4558 | |
96f1bb57 JB |
4559 | /* |
4560 | * We only want to allow over committing if we have lots of actual space | |
4561 | * free, but if we don't have enough space to handle the global reserve | |
4562 | * space then we could end up having a real enospc problem when trying | |
4563 | * to allocate a chunk or some other such important allocation. | |
4564 | */ | |
3c76cd84 MX |
4565 | spin_lock(&global_rsv->lock); |
4566 | space_size = calc_global_rsv_need_space(global_rsv); | |
4567 | spin_unlock(&global_rsv->lock); | |
4568 | if (used + space_size >= space_info->total_bytes) | |
96f1bb57 JB |
4569 | return 0; |
4570 | ||
4571 | used += space_info->bytes_may_use; | |
a80c8dcf | 4572 | |
a5ed45f8 | 4573 | avail = atomic64_read(&fs_info->free_chunk_space); |
a80c8dcf JB |
4574 | |
4575 | /* | |
4576 | * If we have dup, raid1 or raid10 then only half of the free | |
52042d8e | 4577 | * space is actually usable. For raid56, the space info used |
53b381b3 DW |
4578 | * doesn't include the parity drive, so we don't have to |
4579 | * change the math | |
a80c8dcf | 4580 | */ |
46df06b8 DS |
4581 | factor = btrfs_bg_type_to_factor(profile); |
4582 | avail = div_u64(avail, factor); | |
a80c8dcf JB |
4583 | |
4584 | /* | |
561c294d MX |
4585 | * If we aren't flushing all things, let us overcommit up to |
4586 | * 1/2th of the space. If we can flush, don't let us overcommit | |
4587 | * too much, let it overcommit up to 1/8 of the space. | |
a80c8dcf | 4588 | */ |
08e007d2 | 4589 | if (flush == BTRFS_RESERVE_FLUSH_ALL) |
14575aef | 4590 | avail >>= 3; |
a80c8dcf | 4591 | else |
14575aef | 4592 | avail >>= 1; |
a80c8dcf | 4593 | |
14575aef | 4594 | if (used + bytes < space_info->total_bytes + avail) |
a80c8dcf JB |
4595 | return 1; |
4596 | return 0; | |
4597 | } | |
4598 | ||
2ff7e61e | 4599 | static void btrfs_writeback_inodes_sb_nr(struct btrfs_fs_info *fs_info, |
6c255e67 | 4600 | unsigned long nr_pages, int nr_items) |
da633a42 | 4601 | { |
0b246afa | 4602 | struct super_block *sb = fs_info->sb; |
da633a42 | 4603 | |
925a6efb JB |
4604 | if (down_read_trylock(&sb->s_umount)) { |
4605 | writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE); | |
4606 | up_read(&sb->s_umount); | |
4607 | } else { | |
da633a42 MX |
4608 | /* |
4609 | * We needn't worry the filesystem going from r/w to r/o though | |
4610 | * we don't acquire ->s_umount mutex, because the filesystem | |
4611 | * should guarantee the delalloc inodes list be empty after | |
4612 | * the filesystem is readonly(all dirty pages are written to | |
4613 | * the disk). | |
4614 | */ | |
82b3e53b | 4615 | btrfs_start_delalloc_roots(fs_info, nr_items); |
98ad69cf | 4616 | if (!current->journal_info) |
0b246afa | 4617 | btrfs_wait_ordered_roots(fs_info, nr_items, 0, (u64)-1); |
da633a42 MX |
4618 | } |
4619 | } | |
4620 | ||
6374e57a | 4621 | static inline u64 calc_reclaim_items_nr(struct btrfs_fs_info *fs_info, |
2ff7e61e | 4622 | u64 to_reclaim) |
18cd8ea6 MX |
4623 | { |
4624 | u64 bytes; | |
6374e57a | 4625 | u64 nr; |
18cd8ea6 | 4626 | |
2ff7e61e | 4627 | bytes = btrfs_calc_trans_metadata_size(fs_info, 1); |
6374e57a | 4628 | nr = div64_u64(to_reclaim, bytes); |
18cd8ea6 MX |
4629 | if (!nr) |
4630 | nr = 1; | |
4631 | return nr; | |
4632 | } | |
4633 | ||
ee22184b | 4634 | #define EXTENT_SIZE_PER_ITEM SZ_256K |
c61a16a7 | 4635 | |
9ed74f2d | 4636 | /* |
5da9d01b | 4637 | * shrink metadata reservation for delalloc |
9ed74f2d | 4638 | */ |
c1c4919b JM |
4639 | static void shrink_delalloc(struct btrfs_fs_info *fs_info, u64 to_reclaim, |
4640 | u64 orig, bool wait_ordered) | |
5da9d01b | 4641 | { |
0019f10d | 4642 | struct btrfs_space_info *space_info; |
663350ac | 4643 | struct btrfs_trans_handle *trans; |
f4c738c2 | 4644 | u64 delalloc_bytes; |
4297ff84 | 4645 | u64 dio_bytes; |
420829d8 | 4646 | u64 async_pages; |
6374e57a | 4647 | u64 items; |
b1953bce | 4648 | long time_left; |
d3ee29e3 MX |
4649 | unsigned long nr_pages; |
4650 | int loops; | |
5da9d01b | 4651 | |
c61a16a7 | 4652 | /* Calc the number of the pages we need flush for space reservation */ |
2ff7e61e | 4653 | items = calc_reclaim_items_nr(fs_info, to_reclaim); |
6374e57a | 4654 | to_reclaim = items * EXTENT_SIZE_PER_ITEM; |
c61a16a7 | 4655 | |
663350ac | 4656 | trans = (struct btrfs_trans_handle *)current->journal_info; |
69fe2d75 | 4657 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
bf9022e0 | 4658 | |
963d678b | 4659 | delalloc_bytes = percpu_counter_sum_positive( |
0b246afa | 4660 | &fs_info->delalloc_bytes); |
4297ff84 JB |
4661 | dio_bytes = percpu_counter_sum_positive(&fs_info->dio_bytes); |
4662 | if (delalloc_bytes == 0 && dio_bytes == 0) { | |
fdb5effd | 4663 | if (trans) |
f4c738c2 | 4664 | return; |
38c135af | 4665 | if (wait_ordered) |
0b246afa | 4666 | btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1); |
f4c738c2 | 4667 | return; |
fdb5effd JB |
4668 | } |
4669 | ||
4297ff84 JB |
4670 | /* |
4671 | * If we are doing more ordered than delalloc we need to just wait on | |
4672 | * ordered extents, otherwise we'll waste time trying to flush delalloc | |
4673 | * that likely won't give us the space back we need. | |
4674 | */ | |
4675 | if (dio_bytes > delalloc_bytes) | |
4676 | wait_ordered = true; | |
4677 | ||
d3ee29e3 | 4678 | loops = 0; |
4297ff84 | 4679 | while ((delalloc_bytes || dio_bytes) && loops < 3) { |
420829d8 NB |
4680 | nr_pages = min(delalloc_bytes, to_reclaim) >> PAGE_SHIFT; |
4681 | ||
4682 | /* | |
4683 | * Triggers inode writeback for up to nr_pages. This will invoke | |
4684 | * ->writepages callback and trigger delalloc filling | |
4685 | * (btrfs_run_delalloc_range()). | |
4686 | */ | |
2ff7e61e | 4687 | btrfs_writeback_inodes_sb_nr(fs_info, nr_pages, items); |
420829d8 | 4688 | |
dea31f52 | 4689 | /* |
420829d8 NB |
4690 | * We need to wait for the compressed pages to start before |
4691 | * we continue. | |
dea31f52 | 4692 | */ |
420829d8 NB |
4693 | async_pages = atomic_read(&fs_info->async_delalloc_pages); |
4694 | if (!async_pages) | |
9f3a074d MX |
4695 | goto skip_async; |
4696 | ||
420829d8 NB |
4697 | /* |
4698 | * Calculate how many compressed pages we want to be written | |
4699 | * before we continue. I.e if there are more async pages than we | |
4700 | * require wait_event will wait until nr_pages are written. | |
4701 | */ | |
4702 | if (async_pages <= nr_pages) | |
4703 | async_pages = 0; | |
9f3a074d | 4704 | else |
420829d8 | 4705 | async_pages -= nr_pages; |
dea31f52 | 4706 | |
0b246afa JM |
4707 | wait_event(fs_info->async_submit_wait, |
4708 | atomic_read(&fs_info->async_delalloc_pages) <= | |
420829d8 | 4709 | (int)async_pages); |
9f3a074d | 4710 | skip_async: |
0019f10d | 4711 | spin_lock(&space_info->lock); |
957780eb JB |
4712 | if (list_empty(&space_info->tickets) && |
4713 | list_empty(&space_info->priority_tickets)) { | |
4714 | spin_unlock(&space_info->lock); | |
4715 | break; | |
4716 | } | |
0019f10d | 4717 | spin_unlock(&space_info->lock); |
5da9d01b | 4718 | |
36e39c40 | 4719 | loops++; |
f104d044 | 4720 | if (wait_ordered && !trans) { |
0b246afa | 4721 | btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1); |
f104d044 | 4722 | } else { |
f4c738c2 | 4723 | time_left = schedule_timeout_killable(1); |
f104d044 JB |
4724 | if (time_left) |
4725 | break; | |
4726 | } | |
963d678b | 4727 | delalloc_bytes = percpu_counter_sum_positive( |
0b246afa | 4728 | &fs_info->delalloc_bytes); |
4297ff84 | 4729 | dio_bytes = percpu_counter_sum_positive(&fs_info->dio_bytes); |
5da9d01b | 4730 | } |
5da9d01b YZ |
4731 | } |
4732 | ||
996478ca | 4733 | struct reserve_ticket { |
f91587e4 | 4734 | u64 orig_bytes; |
996478ca JB |
4735 | u64 bytes; |
4736 | int error; | |
4737 | struct list_head list; | |
4738 | wait_queue_head_t wait; | |
4739 | }; | |
4740 | ||
663350ac JB |
4741 | /** |
4742 | * maybe_commit_transaction - possibly commit the transaction if its ok to | |
4743 | * @root - the root we're allocating for | |
4744 | * @bytes - the number of bytes we want to reserve | |
4745 | * @force - force the commit | |
8bb8ab2e | 4746 | * |
663350ac JB |
4747 | * This will check to make sure that committing the transaction will actually |
4748 | * get us somewhere and then commit the transaction if it does. Otherwise it | |
4749 | * will return -ENOSPC. | |
8bb8ab2e | 4750 | */ |
0c9ab349 | 4751 | static int may_commit_transaction(struct btrfs_fs_info *fs_info, |
996478ca | 4752 | struct btrfs_space_info *space_info) |
663350ac | 4753 | { |
996478ca | 4754 | struct reserve_ticket *ticket = NULL; |
0b246afa | 4755 | struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_block_rsv; |
4c8edbc7 | 4756 | struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv; |
663350ac | 4757 | struct btrfs_trans_handle *trans; |
4c8edbc7 JB |
4758 | u64 bytes_needed; |
4759 | u64 reclaim_bytes = 0; | |
663350ac JB |
4760 | |
4761 | trans = (struct btrfs_trans_handle *)current->journal_info; | |
4762 | if (trans) | |
4763 | return -EAGAIN; | |
4764 | ||
996478ca JB |
4765 | spin_lock(&space_info->lock); |
4766 | if (!list_empty(&space_info->priority_tickets)) | |
4767 | ticket = list_first_entry(&space_info->priority_tickets, | |
4768 | struct reserve_ticket, list); | |
4769 | else if (!list_empty(&space_info->tickets)) | |
4770 | ticket = list_first_entry(&space_info->tickets, | |
4771 | struct reserve_ticket, list); | |
4c8edbc7 | 4772 | bytes_needed = (ticket) ? ticket->bytes : 0; |
996478ca JB |
4773 | spin_unlock(&space_info->lock); |
4774 | ||
4c8edbc7 | 4775 | if (!bytes_needed) |
996478ca | 4776 | return 0; |
663350ac | 4777 | |
d89dbefb JB |
4778 | trans = btrfs_join_transaction(fs_info->extent_root); |
4779 | if (IS_ERR(trans)) | |
4780 | return PTR_ERR(trans); | |
4781 | ||
4782 | /* | |
4783 | * See if there is enough pinned space to make this reservation, or if | |
4784 | * we have block groups that are going to be freed, allowing us to | |
4785 | * possibly do a chunk allocation the next loop through. | |
4786 | */ | |
4787 | if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags) || | |
4788 | __percpu_counter_compare(&space_info->total_bytes_pinned, | |
4789 | bytes_needed, | |
4790 | BTRFS_TOTAL_BYTES_PINNED_BATCH) >= 0) | |
663350ac | 4791 | goto commit; |
663350ac JB |
4792 | |
4793 | /* | |
4794 | * See if there is some space in the delayed insertion reservation for | |
4795 | * this reservation. | |
4796 | */ | |
4797 | if (space_info != delayed_rsv->space_info) | |
d89dbefb | 4798 | goto enospc; |
663350ac JB |
4799 | |
4800 | spin_lock(&delayed_rsv->lock); | |
4c8edbc7 | 4801 | reclaim_bytes += delayed_rsv->reserved; |
057aac3e NB |
4802 | spin_unlock(&delayed_rsv->lock); |
4803 | ||
4c8edbc7 JB |
4804 | spin_lock(&delayed_refs_rsv->lock); |
4805 | reclaim_bytes += delayed_refs_rsv->reserved; | |
4806 | spin_unlock(&delayed_refs_rsv->lock); | |
4807 | if (reclaim_bytes >= bytes_needed) | |
4808 | goto commit; | |
4809 | bytes_needed -= reclaim_bytes; | |
4810 | ||
dec59fa3 | 4811 | if (__percpu_counter_compare(&space_info->total_bytes_pinned, |
4c8edbc7 | 4812 | bytes_needed, |
d89dbefb JB |
4813 | BTRFS_TOTAL_BYTES_PINNED_BATCH) < 0) |
4814 | goto enospc; | |
663350ac JB |
4815 | |
4816 | commit: | |
3a45bb20 | 4817 | return btrfs_commit_transaction(trans); |
d89dbefb JB |
4818 | enospc: |
4819 | btrfs_end_transaction(trans); | |
4820 | return -ENOSPC; | |
663350ac JB |
4821 | } |
4822 | ||
e38ae7a0 NB |
4823 | /* |
4824 | * Try to flush some data based on policy set by @state. This is only advisory | |
4825 | * and may fail for various reasons. The caller is supposed to examine the | |
4826 | * state of @space_info to detect the outcome. | |
4827 | */ | |
4828 | static void flush_space(struct btrfs_fs_info *fs_info, | |
96c3f433 | 4829 | struct btrfs_space_info *space_info, u64 num_bytes, |
7bdd6277 | 4830 | int state) |
96c3f433 | 4831 | { |
a9b3311e | 4832 | struct btrfs_root *root = fs_info->extent_root; |
96c3f433 JB |
4833 | struct btrfs_trans_handle *trans; |
4834 | int nr; | |
f4c738c2 | 4835 | int ret = 0; |
96c3f433 JB |
4836 | |
4837 | switch (state) { | |
96c3f433 JB |
4838 | case FLUSH_DELAYED_ITEMS_NR: |
4839 | case FLUSH_DELAYED_ITEMS: | |
18cd8ea6 | 4840 | if (state == FLUSH_DELAYED_ITEMS_NR) |
2ff7e61e | 4841 | nr = calc_reclaim_items_nr(fs_info, num_bytes) * 2; |
18cd8ea6 | 4842 | else |
96c3f433 | 4843 | nr = -1; |
18cd8ea6 | 4844 | |
96c3f433 JB |
4845 | trans = btrfs_join_transaction(root); |
4846 | if (IS_ERR(trans)) { | |
4847 | ret = PTR_ERR(trans); | |
4848 | break; | |
4849 | } | |
e5c304e6 | 4850 | ret = btrfs_run_delayed_items_nr(trans, nr); |
3a45bb20 | 4851 | btrfs_end_transaction(trans); |
96c3f433 | 4852 | break; |
67b0fd63 JB |
4853 | case FLUSH_DELALLOC: |
4854 | case FLUSH_DELALLOC_WAIT: | |
7bdd6277 | 4855 | shrink_delalloc(fs_info, num_bytes * 2, num_bytes, |
67b0fd63 JB |
4856 | state == FLUSH_DELALLOC_WAIT); |
4857 | break; | |
413df725 JB |
4858 | case FLUSH_DELAYED_REFS_NR: |
4859 | case FLUSH_DELAYED_REFS: | |
4860 | trans = btrfs_join_transaction(root); | |
4861 | if (IS_ERR(trans)) { | |
4862 | ret = PTR_ERR(trans); | |
4863 | break; | |
4864 | } | |
4865 | if (state == FLUSH_DELAYED_REFS_NR) | |
4866 | nr = calc_reclaim_items_nr(fs_info, num_bytes); | |
4867 | else | |
4868 | nr = 0; | |
4869 | btrfs_run_delayed_refs(trans, nr); | |
4870 | btrfs_end_transaction(trans); | |
4871 | break; | |
ea658bad | 4872 | case ALLOC_CHUNK: |
450114fc | 4873 | case ALLOC_CHUNK_FORCE: |
ea658bad JB |
4874 | trans = btrfs_join_transaction(root); |
4875 | if (IS_ERR(trans)) { | |
4876 | ret = PTR_ERR(trans); | |
4877 | break; | |
4878 | } | |
01458828 | 4879 | ret = do_chunk_alloc(trans, |
1b86826d | 4880 | btrfs_metadata_alloc_profile(fs_info), |
450114fc JB |
4881 | (state == ALLOC_CHUNK) ? |
4882 | CHUNK_ALLOC_NO_FORCE : CHUNK_ALLOC_FORCE); | |
3a45bb20 | 4883 | btrfs_end_transaction(trans); |
eecba891 | 4884 | if (ret > 0 || ret == -ENOSPC) |
ea658bad JB |
4885 | ret = 0; |
4886 | break; | |
96c3f433 | 4887 | case COMMIT_TRANS: |
3ec9a4c8 JB |
4888 | /* |
4889 | * If we have pending delayed iputs then we could free up a | |
4890 | * bunch of pinned space, so make sure we run the iputs before | |
4891 | * we do our pinned bytes check below. | |
4892 | */ | |
3ec9a4c8 | 4893 | btrfs_run_delayed_iputs(fs_info); |
034f784d | 4894 | btrfs_wait_on_delayed_iputs(fs_info); |
3ec9a4c8 | 4895 | |
996478ca | 4896 | ret = may_commit_transaction(fs_info, space_info); |
96c3f433 JB |
4897 | break; |
4898 | default: | |
4899 | ret = -ENOSPC; | |
4900 | break; | |
4901 | } | |
4902 | ||
7bdd6277 NB |
4903 | trace_btrfs_flush_space(fs_info, space_info->flags, num_bytes, state, |
4904 | ret); | |
e38ae7a0 | 4905 | return; |
96c3f433 | 4906 | } |
21c7e756 MX |
4907 | |
4908 | static inline u64 | |
c1c4919b JM |
4909 | btrfs_calc_reclaim_metadata_size(struct btrfs_fs_info *fs_info, |
4910 | struct btrfs_space_info *space_info, | |
4911 | bool system_chunk) | |
21c7e756 | 4912 | { |
957780eb | 4913 | struct reserve_ticket *ticket; |
21c7e756 MX |
4914 | u64 used; |
4915 | u64 expected; | |
957780eb | 4916 | u64 to_reclaim = 0; |
21c7e756 | 4917 | |
957780eb JB |
4918 | list_for_each_entry(ticket, &space_info->tickets, list) |
4919 | to_reclaim += ticket->bytes; | |
4920 | list_for_each_entry(ticket, &space_info->priority_tickets, list) | |
4921 | to_reclaim += ticket->bytes; | |
4922 | if (to_reclaim) | |
4923 | return to_reclaim; | |
21c7e756 | 4924 | |
e0af2484 | 4925 | to_reclaim = min_t(u64, num_online_cpus() * SZ_1M, SZ_16M); |
c1c4919b JM |
4926 | if (can_overcommit(fs_info, space_info, to_reclaim, |
4927 | BTRFS_RESERVE_FLUSH_ALL, system_chunk)) | |
e0af2484 WX |
4928 | return 0; |
4929 | ||
0eee8a49 NB |
4930 | used = btrfs_space_info_used(space_info, true); |
4931 | ||
c1c4919b JM |
4932 | if (can_overcommit(fs_info, space_info, SZ_1M, |
4933 | BTRFS_RESERVE_FLUSH_ALL, system_chunk)) | |
21c7e756 MX |
4934 | expected = div_factor_fine(space_info->total_bytes, 95); |
4935 | else | |
4936 | expected = div_factor_fine(space_info->total_bytes, 90); | |
4937 | ||
4938 | if (used > expected) | |
4939 | to_reclaim = used - expected; | |
4940 | else | |
4941 | to_reclaim = 0; | |
4942 | to_reclaim = min(to_reclaim, space_info->bytes_may_use + | |
4943 | space_info->bytes_reserved); | |
21c7e756 MX |
4944 | return to_reclaim; |
4945 | } | |
4946 | ||
c1c4919b JM |
4947 | static inline int need_do_async_reclaim(struct btrfs_fs_info *fs_info, |
4948 | struct btrfs_space_info *space_info, | |
4949 | u64 used, bool system_chunk) | |
21c7e756 | 4950 | { |
365c5313 JB |
4951 | u64 thresh = div_factor_fine(space_info->total_bytes, 98); |
4952 | ||
4953 | /* If we're just plain full then async reclaim just slows us down. */ | |
baee8790 | 4954 | if ((space_info->bytes_used + space_info->bytes_reserved) >= thresh) |
365c5313 JB |
4955 | return 0; |
4956 | ||
c1c4919b JM |
4957 | if (!btrfs_calc_reclaim_metadata_size(fs_info, space_info, |
4958 | system_chunk)) | |
d38b349c JB |
4959 | return 0; |
4960 | ||
0b246afa JM |
4961 | return (used >= thresh && !btrfs_fs_closing(fs_info) && |
4962 | !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state)); | |
21c7e756 MX |
4963 | } |
4964 | ||
f91587e4 | 4965 | static bool wake_all_tickets(struct list_head *head) |
21c7e756 | 4966 | { |
957780eb | 4967 | struct reserve_ticket *ticket; |
25ce459c | 4968 | |
957780eb JB |
4969 | while (!list_empty(head)) { |
4970 | ticket = list_first_entry(head, struct reserve_ticket, list); | |
4971 | list_del_init(&ticket->list); | |
4972 | ticket->error = -ENOSPC; | |
4973 | wake_up(&ticket->wait); | |
f91587e4 JB |
4974 | if (ticket->bytes != ticket->orig_bytes) |
4975 | return true; | |
21c7e756 | 4976 | } |
f91587e4 | 4977 | return false; |
21c7e756 MX |
4978 | } |
4979 | ||
957780eb JB |
4980 | /* |
4981 | * This is for normal flushers, we can wait all goddamned day if we want to. We | |
4982 | * will loop and continuously try to flush as long as we are making progress. | |
4983 | * We count progress as clearing off tickets each time we have to loop. | |
4984 | */ | |
21c7e756 MX |
4985 | static void btrfs_async_reclaim_metadata_space(struct work_struct *work) |
4986 | { | |
4987 | struct btrfs_fs_info *fs_info; | |
4988 | struct btrfs_space_info *space_info; | |
4989 | u64 to_reclaim; | |
4990 | int flush_state; | |
957780eb | 4991 | int commit_cycles = 0; |
ce129655 | 4992 | u64 last_tickets_id; |
21c7e756 MX |
4993 | |
4994 | fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work); | |
4995 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); | |
4996 | ||
957780eb | 4997 | spin_lock(&space_info->lock); |
c1c4919b JM |
4998 | to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info, |
4999 | false); | |
957780eb JB |
5000 | if (!to_reclaim) { |
5001 | space_info->flush = 0; | |
5002 | spin_unlock(&space_info->lock); | |
21c7e756 | 5003 | return; |
957780eb | 5004 | } |
ce129655 | 5005 | last_tickets_id = space_info->tickets_id; |
957780eb | 5006 | spin_unlock(&space_info->lock); |
21c7e756 MX |
5007 | |
5008 | flush_state = FLUSH_DELAYED_ITEMS_NR; | |
957780eb | 5009 | do { |
e38ae7a0 | 5010 | flush_space(fs_info, space_info, to_reclaim, flush_state); |
957780eb JB |
5011 | spin_lock(&space_info->lock); |
5012 | if (list_empty(&space_info->tickets)) { | |
5013 | space_info->flush = 0; | |
5014 | spin_unlock(&space_info->lock); | |
5015 | return; | |
5016 | } | |
c1c4919b JM |
5017 | to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, |
5018 | space_info, | |
5019 | false); | |
ce129655 | 5020 | if (last_tickets_id == space_info->tickets_id) { |
957780eb JB |
5021 | flush_state++; |
5022 | } else { | |
ce129655 | 5023 | last_tickets_id = space_info->tickets_id; |
957780eb JB |
5024 | flush_state = FLUSH_DELAYED_ITEMS_NR; |
5025 | if (commit_cycles) | |
5026 | commit_cycles--; | |
5027 | } | |
5028 | ||
450114fc JB |
5029 | /* |
5030 | * We don't want to force a chunk allocation until we've tried | |
5031 | * pretty hard to reclaim space. Think of the case where we | |
5032 | * freed up a bunch of space and so have a lot of pinned space | |
5033 | * to reclaim. We would rather use that than possibly create a | |
5034 | * underutilized metadata chunk. So if this is our first run | |
5035 | * through the flushing state machine skip ALLOC_CHUNK_FORCE and | |
5036 | * commit the transaction. If nothing has changed the next go | |
5037 | * around then we can force a chunk allocation. | |
5038 | */ | |
5039 | if (flush_state == ALLOC_CHUNK_FORCE && !commit_cycles) | |
5040 | flush_state++; | |
5041 | ||
957780eb JB |
5042 | if (flush_state > COMMIT_TRANS) { |
5043 | commit_cycles++; | |
5044 | if (commit_cycles > 2) { | |
f91587e4 JB |
5045 | if (wake_all_tickets(&space_info->tickets)) { |
5046 | flush_state = FLUSH_DELAYED_ITEMS_NR; | |
5047 | commit_cycles--; | |
5048 | } else { | |
5049 | space_info->flush = 0; | |
5050 | } | |
957780eb JB |
5051 | } else { |
5052 | flush_state = FLUSH_DELAYED_ITEMS_NR; | |
5053 | } | |
5054 | } | |
5055 | spin_unlock(&space_info->lock); | |
5056 | } while (flush_state <= COMMIT_TRANS); | |
5057 | } | |
5058 | ||
5059 | void btrfs_init_async_reclaim_work(struct work_struct *work) | |
5060 | { | |
5061 | INIT_WORK(work, btrfs_async_reclaim_metadata_space); | |
5062 | } | |
5063 | ||
8a1bbe1d JB |
5064 | static const enum btrfs_flush_state priority_flush_states[] = { |
5065 | FLUSH_DELAYED_ITEMS_NR, | |
5066 | FLUSH_DELAYED_ITEMS, | |
5067 | ALLOC_CHUNK, | |
5068 | }; | |
5069 | ||
957780eb JB |
5070 | static void priority_reclaim_metadata_space(struct btrfs_fs_info *fs_info, |
5071 | struct btrfs_space_info *space_info, | |
5072 | struct reserve_ticket *ticket) | |
5073 | { | |
5074 | u64 to_reclaim; | |
8a1bbe1d | 5075 | int flush_state; |
957780eb JB |
5076 | |
5077 | spin_lock(&space_info->lock); | |
c1c4919b JM |
5078 | to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info, |
5079 | false); | |
957780eb JB |
5080 | if (!to_reclaim) { |
5081 | spin_unlock(&space_info->lock); | |
5082 | return; | |
5083 | } | |
5084 | spin_unlock(&space_info->lock); | |
5085 | ||
8a1bbe1d | 5086 | flush_state = 0; |
21c7e756 | 5087 | do { |
8a1bbe1d JB |
5088 | flush_space(fs_info, space_info, to_reclaim, |
5089 | priority_flush_states[flush_state]); | |
21c7e756 | 5090 | flush_state++; |
957780eb JB |
5091 | spin_lock(&space_info->lock); |
5092 | if (ticket->bytes == 0) { | |
5093 | spin_unlock(&space_info->lock); | |
21c7e756 | 5094 | return; |
957780eb JB |
5095 | } |
5096 | spin_unlock(&space_info->lock); | |
8a1bbe1d | 5097 | } while (flush_state < ARRAY_SIZE(priority_flush_states)); |
21c7e756 MX |
5098 | } |
5099 | ||
957780eb JB |
5100 | static int wait_reserve_ticket(struct btrfs_fs_info *fs_info, |
5101 | struct btrfs_space_info *space_info, | |
f91587e4 | 5102 | struct reserve_ticket *ticket) |
957780eb | 5103 | |
21c7e756 | 5104 | { |
957780eb | 5105 | DEFINE_WAIT(wait); |
f91587e4 | 5106 | u64 reclaim_bytes = 0; |
957780eb JB |
5107 | int ret = 0; |
5108 | ||
5109 | spin_lock(&space_info->lock); | |
5110 | while (ticket->bytes > 0 && ticket->error == 0) { | |
5111 | ret = prepare_to_wait_event(&ticket->wait, &wait, TASK_KILLABLE); | |
5112 | if (ret) { | |
5113 | ret = -EINTR; | |
5114 | break; | |
5115 | } | |
5116 | spin_unlock(&space_info->lock); | |
5117 | ||
5118 | schedule(); | |
5119 | ||
5120 | finish_wait(&ticket->wait, &wait); | |
5121 | spin_lock(&space_info->lock); | |
5122 | } | |
5123 | if (!ret) | |
5124 | ret = ticket->error; | |
5125 | if (!list_empty(&ticket->list)) | |
5126 | list_del_init(&ticket->list); | |
f91587e4 JB |
5127 | if (ticket->bytes && ticket->bytes < ticket->orig_bytes) |
5128 | reclaim_bytes = ticket->orig_bytes - ticket->bytes; | |
957780eb JB |
5129 | spin_unlock(&space_info->lock); |
5130 | ||
f91587e4 JB |
5131 | if (reclaim_bytes) |
5132 | space_info_add_old_bytes(fs_info, space_info, reclaim_bytes); | |
957780eb | 5133 | return ret; |
21c7e756 MX |
5134 | } |
5135 | ||
4a92b1b8 JB |
5136 | /** |
5137 | * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space | |
5138 | * @root - the root we're allocating for | |
957780eb | 5139 | * @space_info - the space info we want to allocate from |
4a92b1b8 | 5140 | * @orig_bytes - the number of bytes we want |
48fc7f7e | 5141 | * @flush - whether or not we can flush to make our reservation |
8bb8ab2e | 5142 | * |
01327610 | 5143 | * This will reserve orig_bytes number of bytes from the space info associated |
4a92b1b8 JB |
5144 | * with the block_rsv. If there is not enough space it will make an attempt to |
5145 | * flush out space to make room. It will do this by flushing delalloc if | |
5146 | * possible or committing the transaction. If flush is 0 then no attempts to | |
5147 | * regain reservations will be made and this will fail if there is not enough | |
5148 | * space already. | |
8bb8ab2e | 5149 | */ |
c1c4919b | 5150 | static int __reserve_metadata_bytes(struct btrfs_fs_info *fs_info, |
957780eb JB |
5151 | struct btrfs_space_info *space_info, |
5152 | u64 orig_bytes, | |
c1c4919b JM |
5153 | enum btrfs_reserve_flush_enum flush, |
5154 | bool system_chunk) | |
9ed74f2d | 5155 | { |
957780eb | 5156 | struct reserve_ticket ticket; |
2bf64758 | 5157 | u64 used; |
f91587e4 | 5158 | u64 reclaim_bytes = 0; |
8bb8ab2e | 5159 | int ret = 0; |
9ed74f2d | 5160 | |
957780eb | 5161 | ASSERT(orig_bytes); |
8ca17f0f | 5162 | ASSERT(!current->journal_info || flush != BTRFS_RESERVE_FLUSH_ALL); |
fdb5effd | 5163 | |
8bb8ab2e | 5164 | spin_lock(&space_info->lock); |
fdb5effd | 5165 | ret = -ENOSPC; |
4136135b | 5166 | used = btrfs_space_info_used(space_info, true); |
9ed74f2d | 5167 | |
8bb8ab2e | 5168 | /* |
957780eb JB |
5169 | * If we have enough space then hooray, make our reservation and carry |
5170 | * on. If not see if we can overcommit, and if we can, hooray carry on. | |
5171 | * If not things get more complicated. | |
8bb8ab2e | 5172 | */ |
957780eb | 5173 | if (used + orig_bytes <= space_info->total_bytes) { |
480b9b4d | 5174 | update_bytes_may_use(fs_info, space_info, orig_bytes); |
0b246afa JM |
5175 | trace_btrfs_space_reservation(fs_info, "space_info", |
5176 | space_info->flags, orig_bytes, 1); | |
957780eb | 5177 | ret = 0; |
c1c4919b JM |
5178 | } else if (can_overcommit(fs_info, space_info, orig_bytes, flush, |
5179 | system_chunk)) { | |
480b9b4d | 5180 | update_bytes_may_use(fs_info, space_info, orig_bytes); |
0b246afa JM |
5181 | trace_btrfs_space_reservation(fs_info, "space_info", |
5182 | space_info->flags, orig_bytes, 1); | |
44734ed1 | 5183 | ret = 0; |
2bf64758 JB |
5184 | } |
5185 | ||
8bb8ab2e | 5186 | /* |
957780eb JB |
5187 | * If we couldn't make a reservation then setup our reservation ticket |
5188 | * and kick the async worker if it's not already running. | |
08e007d2 | 5189 | * |
957780eb JB |
5190 | * If we are a priority flusher then we just need to add our ticket to |
5191 | * the list and we will do our own flushing further down. | |
8bb8ab2e | 5192 | */ |
72bcd99d | 5193 | if (ret && flush != BTRFS_RESERVE_NO_FLUSH) { |
f91587e4 | 5194 | ticket.orig_bytes = orig_bytes; |
957780eb JB |
5195 | ticket.bytes = orig_bytes; |
5196 | ticket.error = 0; | |
5197 | init_waitqueue_head(&ticket.wait); | |
5198 | if (flush == BTRFS_RESERVE_FLUSH_ALL) { | |
5199 | list_add_tail(&ticket.list, &space_info->tickets); | |
5200 | if (!space_info->flush) { | |
5201 | space_info->flush = 1; | |
0b246afa | 5202 | trace_btrfs_trigger_flush(fs_info, |
f376df2b JB |
5203 | space_info->flags, |
5204 | orig_bytes, flush, | |
5205 | "enospc"); | |
957780eb | 5206 | queue_work(system_unbound_wq, |
c1c4919b | 5207 | &fs_info->async_reclaim_work); |
957780eb JB |
5208 | } |
5209 | } else { | |
5210 | list_add_tail(&ticket.list, | |
5211 | &space_info->priority_tickets); | |
5212 | } | |
21c7e756 MX |
5213 | } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) { |
5214 | used += orig_bytes; | |
f6acfd50 JB |
5215 | /* |
5216 | * We will do the space reservation dance during log replay, | |
5217 | * which means we won't have fs_info->fs_root set, so don't do | |
5218 | * the async reclaim as we will panic. | |
5219 | */ | |
0b246afa | 5220 | if (!test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags) && |
c1c4919b JM |
5221 | need_do_async_reclaim(fs_info, space_info, |
5222 | used, system_chunk) && | |
0b246afa JM |
5223 | !work_busy(&fs_info->async_reclaim_work)) { |
5224 | trace_btrfs_trigger_flush(fs_info, space_info->flags, | |
5225 | orig_bytes, flush, "preempt"); | |
21c7e756 | 5226 | queue_work(system_unbound_wq, |
0b246afa | 5227 | &fs_info->async_reclaim_work); |
f376df2b | 5228 | } |
8bb8ab2e | 5229 | } |
f0486c68 | 5230 | spin_unlock(&space_info->lock); |
08e007d2 | 5231 | if (!ret || flush == BTRFS_RESERVE_NO_FLUSH) |
957780eb | 5232 | return ret; |
f0486c68 | 5233 | |
957780eb | 5234 | if (flush == BTRFS_RESERVE_FLUSH_ALL) |
f91587e4 | 5235 | return wait_reserve_ticket(fs_info, space_info, &ticket); |
08e007d2 | 5236 | |
957780eb | 5237 | ret = 0; |
0b246afa | 5238 | priority_reclaim_metadata_space(fs_info, space_info, &ticket); |
957780eb JB |
5239 | spin_lock(&space_info->lock); |
5240 | if (ticket.bytes) { | |
f91587e4 JB |
5241 | if (ticket.bytes < orig_bytes) |
5242 | reclaim_bytes = orig_bytes - ticket.bytes; | |
957780eb JB |
5243 | list_del_init(&ticket.list); |
5244 | ret = -ENOSPC; | |
5245 | } | |
5246 | spin_unlock(&space_info->lock); | |
f91587e4 JB |
5247 | |
5248 | if (reclaim_bytes) | |
5249 | space_info_add_old_bytes(fs_info, space_info, reclaim_bytes); | |
957780eb JB |
5250 | ASSERT(list_empty(&ticket.list)); |
5251 | return ret; | |
5252 | } | |
8bb8ab2e | 5253 | |
957780eb JB |
5254 | /** |
5255 | * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space | |
5256 | * @root - the root we're allocating for | |
5257 | * @block_rsv - the block_rsv we're allocating for | |
5258 | * @orig_bytes - the number of bytes we want | |
5259 | * @flush - whether or not we can flush to make our reservation | |
5260 | * | |
52042d8e | 5261 | * This will reserve orig_bytes number of bytes from the space info associated |
957780eb JB |
5262 | * with the block_rsv. If there is not enough space it will make an attempt to |
5263 | * flush out space to make room. It will do this by flushing delalloc if | |
5264 | * possible or committing the transaction. If flush is 0 then no attempts to | |
5265 | * regain reservations will be made and this will fail if there is not enough | |
5266 | * space already. | |
5267 | */ | |
5268 | static int reserve_metadata_bytes(struct btrfs_root *root, | |
5269 | struct btrfs_block_rsv *block_rsv, | |
5270 | u64 orig_bytes, | |
5271 | enum btrfs_reserve_flush_enum flush) | |
5272 | { | |
0b246afa JM |
5273 | struct btrfs_fs_info *fs_info = root->fs_info; |
5274 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
957780eb | 5275 | int ret; |
c1c4919b | 5276 | bool system_chunk = (root == fs_info->chunk_root); |
957780eb | 5277 | |
c1c4919b JM |
5278 | ret = __reserve_metadata_bytes(fs_info, block_rsv->space_info, |
5279 | orig_bytes, flush, system_chunk); | |
5d80366e JB |
5280 | if (ret == -ENOSPC && |
5281 | unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) { | |
5d80366e JB |
5282 | if (block_rsv != global_rsv && |
5283 | !block_rsv_use_bytes(global_rsv, orig_bytes)) | |
5284 | ret = 0; | |
5285 | } | |
9a3daff3 | 5286 | if (ret == -ENOSPC) { |
0b246afa | 5287 | trace_btrfs_space_reservation(fs_info, "space_info:enospc", |
957780eb JB |
5288 | block_rsv->space_info->flags, |
5289 | orig_bytes, 1); | |
9a3daff3 NB |
5290 | |
5291 | if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) | |
5292 | dump_space_info(fs_info, block_rsv->space_info, | |
5293 | orig_bytes, 0); | |
5294 | } | |
f0486c68 YZ |
5295 | return ret; |
5296 | } | |
5297 | ||
79787eaa JM |
5298 | static struct btrfs_block_rsv *get_block_rsv( |
5299 | const struct btrfs_trans_handle *trans, | |
5300 | const struct btrfs_root *root) | |
f0486c68 | 5301 | { |
0b246afa | 5302 | struct btrfs_fs_info *fs_info = root->fs_info; |
4c13d758 JB |
5303 | struct btrfs_block_rsv *block_rsv = NULL; |
5304 | ||
e9cf439f | 5305 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) || |
0b246afa JM |
5306 | (root == fs_info->csum_root && trans->adding_csums) || |
5307 | (root == fs_info->uuid_root)) | |
f7a81ea4 SB |
5308 | block_rsv = trans->block_rsv; |
5309 | ||
4c13d758 | 5310 | if (!block_rsv) |
f0486c68 YZ |
5311 | block_rsv = root->block_rsv; |
5312 | ||
5313 | if (!block_rsv) | |
0b246afa | 5314 | block_rsv = &fs_info->empty_block_rsv; |
f0486c68 YZ |
5315 | |
5316 | return block_rsv; | |
5317 | } | |
5318 | ||
5319 | static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, | |
5320 | u64 num_bytes) | |
5321 | { | |
5322 | int ret = -ENOSPC; | |
5323 | spin_lock(&block_rsv->lock); | |
5324 | if (block_rsv->reserved >= num_bytes) { | |
5325 | block_rsv->reserved -= num_bytes; | |
5326 | if (block_rsv->reserved < block_rsv->size) | |
5327 | block_rsv->full = 0; | |
5328 | ret = 0; | |
5329 | } | |
5330 | spin_unlock(&block_rsv->lock); | |
5331 | return ret; | |
5332 | } | |
5333 | ||
5334 | static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv, | |
3a584174 | 5335 | u64 num_bytes, bool update_size) |
f0486c68 YZ |
5336 | { |
5337 | spin_lock(&block_rsv->lock); | |
5338 | block_rsv->reserved += num_bytes; | |
5339 | if (update_size) | |
5340 | block_rsv->size += num_bytes; | |
5341 | else if (block_rsv->reserved >= block_rsv->size) | |
5342 | block_rsv->full = 1; | |
5343 | spin_unlock(&block_rsv->lock); | |
5344 | } | |
5345 | ||
d52be818 JB |
5346 | int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info, |
5347 | struct btrfs_block_rsv *dest, u64 num_bytes, | |
5348 | int min_factor) | |
5349 | { | |
5350 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
5351 | u64 min_bytes; | |
5352 | ||
5353 | if (global_rsv->space_info != dest->space_info) | |
5354 | return -ENOSPC; | |
5355 | ||
5356 | spin_lock(&global_rsv->lock); | |
5357 | min_bytes = div_factor(global_rsv->size, min_factor); | |
5358 | if (global_rsv->reserved < min_bytes + num_bytes) { | |
5359 | spin_unlock(&global_rsv->lock); | |
5360 | return -ENOSPC; | |
5361 | } | |
5362 | global_rsv->reserved -= num_bytes; | |
5363 | if (global_rsv->reserved < global_rsv->size) | |
5364 | global_rsv->full = 0; | |
5365 | spin_unlock(&global_rsv->lock); | |
5366 | ||
3a584174 | 5367 | block_rsv_add_bytes(dest, num_bytes, true); |
d52be818 JB |
5368 | return 0; |
5369 | } | |
5370 | ||
ba2c4d4e JB |
5371 | /** |
5372 | * btrfs_migrate_to_delayed_refs_rsv - transfer bytes to our delayed refs rsv. | |
5373 | * @fs_info - the fs info for our fs. | |
5374 | * @src - the source block rsv to transfer from. | |
5375 | * @num_bytes - the number of bytes to transfer. | |
5376 | * | |
5377 | * This transfers up to the num_bytes amount from the src rsv to the | |
5378 | * delayed_refs_rsv. Any extra bytes are returned to the space info. | |
5379 | */ | |
5380 | void btrfs_migrate_to_delayed_refs_rsv(struct btrfs_fs_info *fs_info, | |
5381 | struct btrfs_block_rsv *src, | |
5382 | u64 num_bytes) | |
5383 | { | |
5384 | struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv; | |
5385 | u64 to_free = 0; | |
5386 | ||
5387 | spin_lock(&src->lock); | |
5388 | src->reserved -= num_bytes; | |
5389 | src->size -= num_bytes; | |
5390 | spin_unlock(&src->lock); | |
5391 | ||
5392 | spin_lock(&delayed_refs_rsv->lock); | |
5393 | if (delayed_refs_rsv->size > delayed_refs_rsv->reserved) { | |
5394 | u64 delta = delayed_refs_rsv->size - | |
5395 | delayed_refs_rsv->reserved; | |
5396 | if (num_bytes > delta) { | |
5397 | to_free = num_bytes - delta; | |
5398 | num_bytes = delta; | |
5399 | } | |
5400 | } else { | |
5401 | to_free = num_bytes; | |
5402 | num_bytes = 0; | |
5403 | } | |
5404 | ||
5405 | if (num_bytes) | |
5406 | delayed_refs_rsv->reserved += num_bytes; | |
5407 | if (delayed_refs_rsv->reserved >= delayed_refs_rsv->size) | |
5408 | delayed_refs_rsv->full = 1; | |
5409 | spin_unlock(&delayed_refs_rsv->lock); | |
5410 | ||
5411 | if (num_bytes) | |
5412 | trace_btrfs_space_reservation(fs_info, "delayed_refs_rsv", | |
5413 | 0, num_bytes, 1); | |
5414 | if (to_free) | |
5415 | space_info_add_old_bytes(fs_info, delayed_refs_rsv->space_info, | |
5416 | to_free); | |
5417 | } | |
5418 | ||
5419 | /** | |
5420 | * btrfs_delayed_refs_rsv_refill - refill based on our delayed refs usage. | |
5421 | * @fs_info - the fs_info for our fs. | |
5422 | * @flush - control how we can flush for this reservation. | |
5423 | * | |
5424 | * This will refill the delayed block_rsv up to 1 items size worth of space and | |
5425 | * will return -ENOSPC if we can't make the reservation. | |
5426 | */ | |
5427 | int btrfs_delayed_refs_rsv_refill(struct btrfs_fs_info *fs_info, | |
5428 | enum btrfs_reserve_flush_enum flush) | |
5429 | { | |
5430 | struct btrfs_block_rsv *block_rsv = &fs_info->delayed_refs_rsv; | |
5431 | u64 limit = btrfs_calc_trans_metadata_size(fs_info, 1); | |
5432 | u64 num_bytes = 0; | |
5433 | int ret = -ENOSPC; | |
5434 | ||
5435 | spin_lock(&block_rsv->lock); | |
5436 | if (block_rsv->reserved < block_rsv->size) { | |
5437 | num_bytes = block_rsv->size - block_rsv->reserved; | |
5438 | num_bytes = min(num_bytes, limit); | |
5439 | } | |
5440 | spin_unlock(&block_rsv->lock); | |
5441 | ||
5442 | if (!num_bytes) | |
5443 | return 0; | |
5444 | ||
5445 | ret = reserve_metadata_bytes(fs_info->extent_root, block_rsv, | |
5446 | num_bytes, flush); | |
5447 | if (ret) | |
5448 | return ret; | |
5449 | block_rsv_add_bytes(block_rsv, num_bytes, 0); | |
5450 | trace_btrfs_space_reservation(fs_info, "delayed_refs_rsv", | |
5451 | 0, num_bytes, 1); | |
5452 | return 0; | |
5453 | } | |
5454 | ||
957780eb JB |
5455 | /* |
5456 | * This is for space we already have accounted in space_info->bytes_may_use, so | |
5457 | * basically when we're returning space from block_rsv's. | |
5458 | */ | |
5459 | static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info, | |
5460 | struct btrfs_space_info *space_info, | |
5461 | u64 num_bytes) | |
5462 | { | |
5463 | struct reserve_ticket *ticket; | |
5464 | struct list_head *head; | |
5465 | u64 used; | |
5466 | enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_NO_FLUSH; | |
5467 | bool check_overcommit = false; | |
5468 | ||
5469 | spin_lock(&space_info->lock); | |
5470 | head = &space_info->priority_tickets; | |
5471 | ||
5472 | /* | |
5473 | * If we are over our limit then we need to check and see if we can | |
5474 | * overcommit, and if we can't then we just need to free up our space | |
5475 | * and not satisfy any requests. | |
5476 | */ | |
0eee8a49 | 5477 | used = btrfs_space_info_used(space_info, true); |
957780eb JB |
5478 | if (used - num_bytes >= space_info->total_bytes) |
5479 | check_overcommit = true; | |
5480 | again: | |
5481 | while (!list_empty(head) && num_bytes) { | |
5482 | ticket = list_first_entry(head, struct reserve_ticket, | |
5483 | list); | |
5484 | /* | |
5485 | * We use 0 bytes because this space is already reserved, so | |
5486 | * adding the ticket space would be a double count. | |
5487 | */ | |
5488 | if (check_overcommit && | |
c1c4919b | 5489 | !can_overcommit(fs_info, space_info, 0, flush, false)) |
957780eb JB |
5490 | break; |
5491 | if (num_bytes >= ticket->bytes) { | |
5492 | list_del_init(&ticket->list); | |
5493 | num_bytes -= ticket->bytes; | |
5494 | ticket->bytes = 0; | |
ce129655 | 5495 | space_info->tickets_id++; |
957780eb JB |
5496 | wake_up(&ticket->wait); |
5497 | } else { | |
5498 | ticket->bytes -= num_bytes; | |
5499 | num_bytes = 0; | |
5500 | } | |
5501 | } | |
5502 | ||
5503 | if (num_bytes && head == &space_info->priority_tickets) { | |
5504 | head = &space_info->tickets; | |
5505 | flush = BTRFS_RESERVE_FLUSH_ALL; | |
5506 | goto again; | |
5507 | } | |
480b9b4d | 5508 | update_bytes_may_use(fs_info, space_info, -num_bytes); |
957780eb JB |
5509 | trace_btrfs_space_reservation(fs_info, "space_info", |
5510 | space_info->flags, num_bytes, 0); | |
5511 | spin_unlock(&space_info->lock); | |
5512 | } | |
5513 | ||
5514 | /* | |
5515 | * This is for newly allocated space that isn't accounted in | |
5516 | * space_info->bytes_may_use yet. So if we allocate a chunk or unpin an extent | |
5517 | * we use this helper. | |
5518 | */ | |
5519 | static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info, | |
5520 | struct btrfs_space_info *space_info, | |
5521 | u64 num_bytes) | |
5522 | { | |
5523 | struct reserve_ticket *ticket; | |
5524 | struct list_head *head = &space_info->priority_tickets; | |
5525 | ||
5526 | again: | |
5527 | while (!list_empty(head) && num_bytes) { | |
5528 | ticket = list_first_entry(head, struct reserve_ticket, | |
5529 | list); | |
5530 | if (num_bytes >= ticket->bytes) { | |
5531 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5532 | space_info->flags, | |
5533 | ticket->bytes, 1); | |
5534 | list_del_init(&ticket->list); | |
5535 | num_bytes -= ticket->bytes; | |
480b9b4d QW |
5536 | update_bytes_may_use(fs_info, space_info, |
5537 | ticket->bytes); | |
957780eb | 5538 | ticket->bytes = 0; |
ce129655 | 5539 | space_info->tickets_id++; |
957780eb JB |
5540 | wake_up(&ticket->wait); |
5541 | } else { | |
5542 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5543 | space_info->flags, | |
5544 | num_bytes, 1); | |
480b9b4d | 5545 | update_bytes_may_use(fs_info, space_info, num_bytes); |
957780eb JB |
5546 | ticket->bytes -= num_bytes; |
5547 | num_bytes = 0; | |
5548 | } | |
5549 | } | |
5550 | ||
5551 | if (num_bytes && head == &space_info->priority_tickets) { | |
5552 | head = &space_info->tickets; | |
5553 | goto again; | |
5554 | } | |
5555 | } | |
5556 | ||
69fe2d75 | 5557 | static u64 block_rsv_release_bytes(struct btrfs_fs_info *fs_info, |
8c2a3ca2 | 5558 | struct btrfs_block_rsv *block_rsv, |
ff6bc37e QW |
5559 | struct btrfs_block_rsv *dest, u64 num_bytes, |
5560 | u64 *qgroup_to_release_ret) | |
f0486c68 YZ |
5561 | { |
5562 | struct btrfs_space_info *space_info = block_rsv->space_info; | |
ff6bc37e | 5563 | u64 qgroup_to_release = 0; |
69fe2d75 | 5564 | u64 ret; |
f0486c68 YZ |
5565 | |
5566 | spin_lock(&block_rsv->lock); | |
ff6bc37e | 5567 | if (num_bytes == (u64)-1) { |
f0486c68 | 5568 | num_bytes = block_rsv->size; |
ff6bc37e QW |
5569 | qgroup_to_release = block_rsv->qgroup_rsv_size; |
5570 | } | |
f0486c68 YZ |
5571 | block_rsv->size -= num_bytes; |
5572 | if (block_rsv->reserved >= block_rsv->size) { | |
5573 | num_bytes = block_rsv->reserved - block_rsv->size; | |
5574 | block_rsv->reserved = block_rsv->size; | |
5575 | block_rsv->full = 1; | |
5576 | } else { | |
5577 | num_bytes = 0; | |
5578 | } | |
ff6bc37e QW |
5579 | if (block_rsv->qgroup_rsv_reserved >= block_rsv->qgroup_rsv_size) { |
5580 | qgroup_to_release = block_rsv->qgroup_rsv_reserved - | |
5581 | block_rsv->qgroup_rsv_size; | |
5582 | block_rsv->qgroup_rsv_reserved = block_rsv->qgroup_rsv_size; | |
5583 | } else { | |
5584 | qgroup_to_release = 0; | |
5585 | } | |
f0486c68 YZ |
5586 | spin_unlock(&block_rsv->lock); |
5587 | ||
69fe2d75 | 5588 | ret = num_bytes; |
f0486c68 YZ |
5589 | if (num_bytes > 0) { |
5590 | if (dest) { | |
e9e22899 JB |
5591 | spin_lock(&dest->lock); |
5592 | if (!dest->full) { | |
5593 | u64 bytes_to_add; | |
5594 | ||
5595 | bytes_to_add = dest->size - dest->reserved; | |
5596 | bytes_to_add = min(num_bytes, bytes_to_add); | |
5597 | dest->reserved += bytes_to_add; | |
5598 | if (dest->reserved >= dest->size) | |
5599 | dest->full = 1; | |
5600 | num_bytes -= bytes_to_add; | |
5601 | } | |
5602 | spin_unlock(&dest->lock); | |
5603 | } | |
957780eb JB |
5604 | if (num_bytes) |
5605 | space_info_add_old_bytes(fs_info, space_info, | |
5606 | num_bytes); | |
9ed74f2d | 5607 | } |
ff6bc37e QW |
5608 | if (qgroup_to_release_ret) |
5609 | *qgroup_to_release_ret = qgroup_to_release; | |
69fe2d75 | 5610 | return ret; |
f0486c68 | 5611 | } |
4e06bdd6 | 5612 | |
25d609f8 JB |
5613 | int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src, |
5614 | struct btrfs_block_rsv *dst, u64 num_bytes, | |
3a584174 | 5615 | bool update_size) |
f0486c68 YZ |
5616 | { |
5617 | int ret; | |
9ed74f2d | 5618 | |
f0486c68 YZ |
5619 | ret = block_rsv_use_bytes(src, num_bytes); |
5620 | if (ret) | |
5621 | return ret; | |
9ed74f2d | 5622 | |
25d609f8 | 5623 | block_rsv_add_bytes(dst, num_bytes, update_size); |
9ed74f2d JB |
5624 | return 0; |
5625 | } | |
5626 | ||
66d8f3dd | 5627 | void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type) |
9ed74f2d | 5628 | { |
f0486c68 YZ |
5629 | memset(rsv, 0, sizeof(*rsv)); |
5630 | spin_lock_init(&rsv->lock); | |
66d8f3dd | 5631 | rsv->type = type; |
f0486c68 YZ |
5632 | } |
5633 | ||
69fe2d75 JB |
5634 | void btrfs_init_metadata_block_rsv(struct btrfs_fs_info *fs_info, |
5635 | struct btrfs_block_rsv *rsv, | |
5636 | unsigned short type) | |
5637 | { | |
5638 | btrfs_init_block_rsv(rsv, type); | |
5639 | rsv->space_info = __find_space_info(fs_info, | |
5640 | BTRFS_BLOCK_GROUP_METADATA); | |
5641 | } | |
5642 | ||
2ff7e61e | 5643 | struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_fs_info *fs_info, |
66d8f3dd | 5644 | unsigned short type) |
f0486c68 YZ |
5645 | { |
5646 | struct btrfs_block_rsv *block_rsv; | |
9ed74f2d | 5647 | |
f0486c68 YZ |
5648 | block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS); |
5649 | if (!block_rsv) | |
5650 | return NULL; | |
9ed74f2d | 5651 | |
69fe2d75 | 5652 | btrfs_init_metadata_block_rsv(fs_info, block_rsv, type); |
f0486c68 YZ |
5653 | return block_rsv; |
5654 | } | |
9ed74f2d | 5655 | |
2ff7e61e | 5656 | void btrfs_free_block_rsv(struct btrfs_fs_info *fs_info, |
f0486c68 YZ |
5657 | struct btrfs_block_rsv *rsv) |
5658 | { | |
2aaa6655 JB |
5659 | if (!rsv) |
5660 | return; | |
2ff7e61e | 5661 | btrfs_block_rsv_release(fs_info, rsv, (u64)-1); |
dabdb640 | 5662 | kfree(rsv); |
9ed74f2d JB |
5663 | } |
5664 | ||
08e007d2 MX |
5665 | int btrfs_block_rsv_add(struct btrfs_root *root, |
5666 | struct btrfs_block_rsv *block_rsv, u64 num_bytes, | |
5667 | enum btrfs_reserve_flush_enum flush) | |
9ed74f2d | 5668 | { |
f0486c68 | 5669 | int ret; |
9ed74f2d | 5670 | |
f0486c68 YZ |
5671 | if (num_bytes == 0) |
5672 | return 0; | |
8bb8ab2e | 5673 | |
61b520a9 | 5674 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
5a2cb25a | 5675 | if (!ret) |
3a584174 | 5676 | block_rsv_add_bytes(block_rsv, num_bytes, true); |
9ed74f2d | 5677 | |
f0486c68 | 5678 | return ret; |
f0486c68 | 5679 | } |
9ed74f2d | 5680 | |
2ff7e61e | 5681 | int btrfs_block_rsv_check(struct btrfs_block_rsv *block_rsv, int min_factor) |
f0486c68 YZ |
5682 | { |
5683 | u64 num_bytes = 0; | |
f0486c68 | 5684 | int ret = -ENOSPC; |
9ed74f2d | 5685 | |
f0486c68 YZ |
5686 | if (!block_rsv) |
5687 | return 0; | |
9ed74f2d | 5688 | |
f0486c68 | 5689 | spin_lock(&block_rsv->lock); |
36ba022a JB |
5690 | num_bytes = div_factor(block_rsv->size, min_factor); |
5691 | if (block_rsv->reserved >= num_bytes) | |
5692 | ret = 0; | |
5693 | spin_unlock(&block_rsv->lock); | |
9ed74f2d | 5694 | |
36ba022a JB |
5695 | return ret; |
5696 | } | |
5697 | ||
08e007d2 MX |
5698 | int btrfs_block_rsv_refill(struct btrfs_root *root, |
5699 | struct btrfs_block_rsv *block_rsv, u64 min_reserved, | |
5700 | enum btrfs_reserve_flush_enum flush) | |
36ba022a JB |
5701 | { |
5702 | u64 num_bytes = 0; | |
5703 | int ret = -ENOSPC; | |
5704 | ||
5705 | if (!block_rsv) | |
5706 | return 0; | |
5707 | ||
5708 | spin_lock(&block_rsv->lock); | |
5709 | num_bytes = min_reserved; | |
13553e52 | 5710 | if (block_rsv->reserved >= num_bytes) |
f0486c68 | 5711 | ret = 0; |
13553e52 | 5712 | else |
f0486c68 | 5713 | num_bytes -= block_rsv->reserved; |
f0486c68 | 5714 | spin_unlock(&block_rsv->lock); |
13553e52 | 5715 | |
f0486c68 YZ |
5716 | if (!ret) |
5717 | return 0; | |
5718 | ||
aa38a711 | 5719 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
dabdb640 | 5720 | if (!ret) { |
3a584174 | 5721 | block_rsv_add_bytes(block_rsv, num_bytes, false); |
f0486c68 | 5722 | return 0; |
6a63209f | 5723 | } |
9ed74f2d | 5724 | |
13553e52 | 5725 | return ret; |
f0486c68 YZ |
5726 | } |
5727 | ||
ba2c4d4e JB |
5728 | static u64 __btrfs_block_rsv_release(struct btrfs_fs_info *fs_info, |
5729 | struct btrfs_block_rsv *block_rsv, | |
5730 | u64 num_bytes, u64 *qgroup_to_release) | |
5731 | { | |
5732 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
5733 | struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_refs_rsv; | |
5734 | struct btrfs_block_rsv *target = delayed_rsv; | |
5735 | ||
5736 | if (target->full || target == block_rsv) | |
5737 | target = global_rsv; | |
5738 | ||
5739 | if (block_rsv->space_info != target->space_info) | |
5740 | target = NULL; | |
5741 | ||
5742 | return block_rsv_release_bytes(fs_info, block_rsv, target, num_bytes, | |
5743 | qgroup_to_release); | |
5744 | } | |
5745 | ||
5746 | void btrfs_block_rsv_release(struct btrfs_fs_info *fs_info, | |
5747 | struct btrfs_block_rsv *block_rsv, | |
5748 | u64 num_bytes) | |
5749 | { | |
5750 | __btrfs_block_rsv_release(fs_info, block_rsv, num_bytes, NULL); | |
5751 | } | |
5752 | ||
69fe2d75 JB |
5753 | /** |
5754 | * btrfs_inode_rsv_release - release any excessive reservation. | |
5755 | * @inode - the inode we need to release from. | |
43b18595 QW |
5756 | * @qgroup_free - free or convert qgroup meta. |
5757 | * Unlike normal operation, qgroup meta reservation needs to know if we are | |
5758 | * freeing qgroup reservation or just converting it into per-trans. Normally | |
5759 | * @qgroup_free is true for error handling, and false for normal release. | |
69fe2d75 JB |
5760 | * |
5761 | * This is the same as btrfs_block_rsv_release, except that it handles the | |
5762 | * tracepoint for the reservation. | |
5763 | */ | |
43b18595 | 5764 | static void btrfs_inode_rsv_release(struct btrfs_inode *inode, bool qgroup_free) |
69fe2d75 JB |
5765 | { |
5766 | struct btrfs_fs_info *fs_info = inode->root->fs_info; | |
69fe2d75 JB |
5767 | struct btrfs_block_rsv *block_rsv = &inode->block_rsv; |
5768 | u64 released = 0; | |
ff6bc37e | 5769 | u64 qgroup_to_release = 0; |
69fe2d75 JB |
5770 | |
5771 | /* | |
5772 | * Since we statically set the block_rsv->size we just want to say we | |
5773 | * are releasing 0 bytes, and then we'll just get the reservation over | |
5774 | * the size free'd. | |
5775 | */ | |
ba2c4d4e JB |
5776 | released = __btrfs_block_rsv_release(fs_info, block_rsv, 0, |
5777 | &qgroup_to_release); | |
69fe2d75 JB |
5778 | if (released > 0) |
5779 | trace_btrfs_space_reservation(fs_info, "delalloc", | |
5780 | btrfs_ino(inode), released, 0); | |
43b18595 | 5781 | if (qgroup_free) |
ff6bc37e | 5782 | btrfs_qgroup_free_meta_prealloc(inode->root, qgroup_to_release); |
43b18595 | 5783 | else |
ff6bc37e QW |
5784 | btrfs_qgroup_convert_reserved_meta(inode->root, |
5785 | qgroup_to_release); | |
69fe2d75 JB |
5786 | } |
5787 | ||
ba2c4d4e JB |
5788 | /** |
5789 | * btrfs_delayed_refs_rsv_release - release a ref head's reservation. | |
5790 | * @fs_info - the fs_info for our fs. | |
5791 | * @nr - the number of items to drop. | |
5792 | * | |
5793 | * This drops the delayed ref head's count from the delayed refs rsv and frees | |
5794 | * any excess reservation we had. | |
5795 | */ | |
5796 | void btrfs_delayed_refs_rsv_release(struct btrfs_fs_info *fs_info, int nr) | |
f0486c68 | 5797 | { |
ba2c4d4e | 5798 | struct btrfs_block_rsv *block_rsv = &fs_info->delayed_refs_rsv; |
0b246afa | 5799 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; |
ba2c4d4e JB |
5800 | u64 num_bytes = btrfs_calc_trans_metadata_size(fs_info, nr); |
5801 | u64 released = 0; | |
0b246afa | 5802 | |
ba2c4d4e JB |
5803 | released = block_rsv_release_bytes(fs_info, block_rsv, global_rsv, |
5804 | num_bytes, NULL); | |
5805 | if (released) | |
5806 | trace_btrfs_space_reservation(fs_info, "delayed_refs_rsv", | |
5807 | 0, released, 0); | |
6a63209f JB |
5808 | } |
5809 | ||
8929ecfa YZ |
5810 | static void update_global_block_rsv(struct btrfs_fs_info *fs_info) |
5811 | { | |
5812 | struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv; | |
5813 | struct btrfs_space_info *sinfo = block_rsv->space_info; | |
5814 | u64 num_bytes; | |
6a63209f | 5815 | |
ae2e4728 JB |
5816 | /* |
5817 | * The global block rsv is based on the size of the extent tree, the | |
5818 | * checksum tree and the root tree. If the fs is empty we want to set | |
5819 | * it to a minimal amount for safety. | |
5820 | */ | |
5821 | num_bytes = btrfs_root_used(&fs_info->extent_root->root_item) + | |
5822 | btrfs_root_used(&fs_info->csum_root->root_item) + | |
5823 | btrfs_root_used(&fs_info->tree_root->root_item); | |
5824 | num_bytes = max_t(u64, num_bytes, SZ_16M); | |
33b4d47f | 5825 | |
8929ecfa | 5826 | spin_lock(&sinfo->lock); |
1f699d38 | 5827 | spin_lock(&block_rsv->lock); |
4e06bdd6 | 5828 | |
ee22184b | 5829 | block_rsv->size = min_t(u64, num_bytes, SZ_512M); |
4e06bdd6 | 5830 | |
fb4b10e5 | 5831 | if (block_rsv->reserved < block_rsv->size) { |
4136135b | 5832 | num_bytes = btrfs_space_info_used(sinfo, true); |
fb4b10e5 JB |
5833 | if (sinfo->total_bytes > num_bytes) { |
5834 | num_bytes = sinfo->total_bytes - num_bytes; | |
5835 | num_bytes = min(num_bytes, | |
5836 | block_rsv->size - block_rsv->reserved); | |
5837 | block_rsv->reserved += num_bytes; | |
480b9b4d | 5838 | update_bytes_may_use(fs_info, sinfo, num_bytes); |
fb4b10e5 JB |
5839 | trace_btrfs_space_reservation(fs_info, "space_info", |
5840 | sinfo->flags, num_bytes, | |
5841 | 1); | |
5842 | } | |
5843 | } else if (block_rsv->reserved > block_rsv->size) { | |
8929ecfa | 5844 | num_bytes = block_rsv->reserved - block_rsv->size; |
480b9b4d | 5845 | update_bytes_may_use(fs_info, sinfo, -num_bytes); |
8c2a3ca2 | 5846 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 5847 | sinfo->flags, num_bytes, 0); |
8929ecfa | 5848 | block_rsv->reserved = block_rsv->size; |
8929ecfa | 5849 | } |
182608c8 | 5850 | |
fb4b10e5 JB |
5851 | if (block_rsv->reserved == block_rsv->size) |
5852 | block_rsv->full = 1; | |
5853 | else | |
5854 | block_rsv->full = 0; | |
5855 | ||
8929ecfa | 5856 | spin_unlock(&block_rsv->lock); |
1f699d38 | 5857 | spin_unlock(&sinfo->lock); |
6a63209f JB |
5858 | } |
5859 | ||
f0486c68 | 5860 | static void init_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 5861 | { |
f0486c68 | 5862 | struct btrfs_space_info *space_info; |
6a63209f | 5863 | |
f0486c68 YZ |
5864 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); |
5865 | fs_info->chunk_block_rsv.space_info = space_info; | |
6a63209f | 5866 | |
f0486c68 | 5867 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
8929ecfa | 5868 | fs_info->global_block_rsv.space_info = space_info; |
f0486c68 YZ |
5869 | fs_info->trans_block_rsv.space_info = space_info; |
5870 | fs_info->empty_block_rsv.space_info = space_info; | |
6d668dda | 5871 | fs_info->delayed_block_rsv.space_info = space_info; |
ba2c4d4e | 5872 | fs_info->delayed_refs_rsv.space_info = space_info; |
f0486c68 | 5873 | |
ba2c4d4e JB |
5874 | fs_info->extent_root->block_rsv = &fs_info->delayed_refs_rsv; |
5875 | fs_info->csum_root->block_rsv = &fs_info->delayed_refs_rsv; | |
8929ecfa YZ |
5876 | fs_info->dev_root->block_rsv = &fs_info->global_block_rsv; |
5877 | fs_info->tree_root->block_rsv = &fs_info->global_block_rsv; | |
3a6cad90 SB |
5878 | if (fs_info->quota_root) |
5879 | fs_info->quota_root->block_rsv = &fs_info->global_block_rsv; | |
f0486c68 | 5880 | fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv; |
8929ecfa | 5881 | |
8929ecfa | 5882 | update_global_block_rsv(fs_info); |
6a63209f JB |
5883 | } |
5884 | ||
8929ecfa | 5885 | static void release_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 5886 | { |
8c2a3ca2 | 5887 | block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL, |
ff6bc37e | 5888 | (u64)-1, NULL); |
8929ecfa YZ |
5889 | WARN_ON(fs_info->trans_block_rsv.size > 0); |
5890 | WARN_ON(fs_info->trans_block_rsv.reserved > 0); | |
5891 | WARN_ON(fs_info->chunk_block_rsv.size > 0); | |
5892 | WARN_ON(fs_info->chunk_block_rsv.reserved > 0); | |
6d668dda JB |
5893 | WARN_ON(fs_info->delayed_block_rsv.size > 0); |
5894 | WARN_ON(fs_info->delayed_block_rsv.reserved > 0); | |
ba2c4d4e JB |
5895 | WARN_ON(fs_info->delayed_refs_rsv.reserved > 0); |
5896 | WARN_ON(fs_info->delayed_refs_rsv.size > 0); | |
fcb80c2a JB |
5897 | } |
5898 | ||
ba2c4d4e JB |
5899 | /* |
5900 | * btrfs_update_delayed_refs_rsv - adjust the size of the delayed refs rsv | |
5901 | * @trans - the trans that may have generated delayed refs | |
5902 | * | |
5903 | * This is to be called anytime we may have adjusted trans->delayed_ref_updates, | |
5904 | * it'll calculate the additional size and add it to the delayed_refs_rsv. | |
5905 | */ | |
5906 | void btrfs_update_delayed_refs_rsv(struct btrfs_trans_handle *trans) | |
5907 | { | |
5908 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
5909 | struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_refs_rsv; | |
5910 | u64 num_bytes; | |
5911 | ||
5912 | if (!trans->delayed_ref_updates) | |
5913 | return; | |
5914 | ||
5915 | num_bytes = btrfs_calc_trans_metadata_size(fs_info, | |
5916 | trans->delayed_ref_updates); | |
5917 | spin_lock(&delayed_rsv->lock); | |
5918 | delayed_rsv->size += num_bytes; | |
5919 | delayed_rsv->full = 0; | |
5920 | spin_unlock(&delayed_rsv->lock); | |
5921 | trans->delayed_ref_updates = 0; | |
5922 | } | |
6a63209f | 5923 | |
4fbcdf66 FM |
5924 | /* |
5925 | * To be called after all the new block groups attached to the transaction | |
5926 | * handle have been created (btrfs_create_pending_block_groups()). | |
5927 | */ | |
5928 | void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans) | |
5929 | { | |
64b63580 | 5930 | struct btrfs_fs_info *fs_info = trans->fs_info; |
4fbcdf66 FM |
5931 | |
5932 | if (!trans->chunk_bytes_reserved) | |
5933 | return; | |
5934 | ||
5935 | WARN_ON_ONCE(!list_empty(&trans->new_bgs)); | |
5936 | ||
5937 | block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL, | |
ff6bc37e | 5938 | trans->chunk_bytes_reserved, NULL); |
4fbcdf66 FM |
5939 | trans->chunk_bytes_reserved = 0; |
5940 | } | |
5941 | ||
d5c12070 MX |
5942 | /* |
5943 | * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation | |
5944 | * root: the root of the parent directory | |
5945 | * rsv: block reservation | |
5946 | * items: the number of items that we need do reservation | |
a5b7f429 | 5947 | * use_global_rsv: allow fallback to the global block reservation |
d5c12070 MX |
5948 | * |
5949 | * This function is used to reserve the space for snapshot/subvolume | |
5950 | * creation and deletion. Those operations are different with the | |
5951 | * common file/directory operations, they change two fs/file trees | |
5952 | * and root tree, the number of items that the qgroup reserves is | |
5953 | * different with the free space reservation. So we can not use | |
01327610 | 5954 | * the space reservation mechanism in start_transaction(). |
d5c12070 MX |
5955 | */ |
5956 | int btrfs_subvolume_reserve_metadata(struct btrfs_root *root, | |
a5b7f429 | 5957 | struct btrfs_block_rsv *rsv, int items, |
ee3441b4 | 5958 | bool use_global_rsv) |
a22285a6 | 5959 | { |
a5b7f429 | 5960 | u64 qgroup_num_bytes = 0; |
d5c12070 MX |
5961 | u64 num_bytes; |
5962 | int ret; | |
0b246afa JM |
5963 | struct btrfs_fs_info *fs_info = root->fs_info; |
5964 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
d5c12070 | 5965 | |
0b246afa | 5966 | if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) { |
d5c12070 | 5967 | /* One for parent inode, two for dir entries */ |
a5b7f429 LF |
5968 | qgroup_num_bytes = 3 * fs_info->nodesize; |
5969 | ret = btrfs_qgroup_reserve_meta_prealloc(root, | |
5970 | qgroup_num_bytes, true); | |
d5c12070 MX |
5971 | if (ret) |
5972 | return ret; | |
d5c12070 MX |
5973 | } |
5974 | ||
0b246afa JM |
5975 | num_bytes = btrfs_calc_trans_metadata_size(fs_info, items); |
5976 | rsv->space_info = __find_space_info(fs_info, | |
d5c12070 MX |
5977 | BTRFS_BLOCK_GROUP_METADATA); |
5978 | ret = btrfs_block_rsv_add(root, rsv, num_bytes, | |
5979 | BTRFS_RESERVE_FLUSH_ALL); | |
ee3441b4 JM |
5980 | |
5981 | if (ret == -ENOSPC && use_global_rsv) | |
3a584174 | 5982 | ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes, true); |
ee3441b4 | 5983 | |
a5b7f429 LF |
5984 | if (ret && qgroup_num_bytes) |
5985 | btrfs_qgroup_free_meta_prealloc(root, qgroup_num_bytes); | |
d5c12070 MX |
5986 | |
5987 | return ret; | |
5988 | } | |
5989 | ||
2ff7e61e | 5990 | void btrfs_subvolume_release_metadata(struct btrfs_fs_info *fs_info, |
7775c818 | 5991 | struct btrfs_block_rsv *rsv) |
d5c12070 | 5992 | { |
2ff7e61e | 5993 | btrfs_block_rsv_release(fs_info, rsv, (u64)-1); |
97e728d4 JB |
5994 | } |
5995 | ||
69fe2d75 JB |
5996 | static void btrfs_calculate_inode_block_rsv_size(struct btrfs_fs_info *fs_info, |
5997 | struct btrfs_inode *inode) | |
9e0baf60 | 5998 | { |
69fe2d75 JB |
5999 | struct btrfs_block_rsv *block_rsv = &inode->block_rsv; |
6000 | u64 reserve_size = 0; | |
ff6bc37e | 6001 | u64 qgroup_rsv_size = 0; |
69fe2d75 JB |
6002 | u64 csum_leaves; |
6003 | unsigned outstanding_extents; | |
9e0baf60 | 6004 | |
69fe2d75 JB |
6005 | lockdep_assert_held(&inode->lock); |
6006 | outstanding_extents = inode->outstanding_extents; | |
6007 | if (outstanding_extents) | |
6008 | reserve_size = btrfs_calc_trans_metadata_size(fs_info, | |
6009 | outstanding_extents + 1); | |
6010 | csum_leaves = btrfs_csum_bytes_to_leaves(fs_info, | |
6011 | inode->csum_bytes); | |
6012 | reserve_size += btrfs_calc_trans_metadata_size(fs_info, | |
6013 | csum_leaves); | |
ff6bc37e QW |
6014 | /* |
6015 | * For qgroup rsv, the calculation is very simple: | |
6016 | * account one nodesize for each outstanding extent | |
6017 | * | |
6018 | * This is overestimating in most cases. | |
6019 | */ | |
139a5617 | 6020 | qgroup_rsv_size = (u64)outstanding_extents * fs_info->nodesize; |
9e0baf60 | 6021 | |
69fe2d75 JB |
6022 | spin_lock(&block_rsv->lock); |
6023 | block_rsv->size = reserve_size; | |
ff6bc37e | 6024 | block_rsv->qgroup_rsv_size = qgroup_rsv_size; |
69fe2d75 | 6025 | spin_unlock(&block_rsv->lock); |
0ca1f7ce | 6026 | } |
c146afad | 6027 | |
c8eaeac7 JB |
6028 | static void calc_inode_reservations(struct btrfs_fs_info *fs_info, |
6029 | u64 num_bytes, u64 *meta_reserve, | |
6030 | u64 *qgroup_reserve) | |
6031 | { | |
6032 | u64 nr_extents = count_max_extents(num_bytes); | |
6033 | u64 csum_leaves = btrfs_csum_bytes_to_leaves(fs_info, num_bytes); | |
6034 | ||
6035 | /* We add one for the inode update at finish ordered time */ | |
6036 | *meta_reserve = btrfs_calc_trans_metadata_size(fs_info, | |
6037 | nr_extents + csum_leaves + 1); | |
6038 | *qgroup_reserve = nr_extents * fs_info->nodesize; | |
6039 | } | |
6040 | ||
9f3db423 | 6041 | int btrfs_delalloc_reserve_metadata(struct btrfs_inode *inode, u64 num_bytes) |
0ca1f7ce | 6042 | { |
c8eaeac7 JB |
6043 | struct btrfs_root *root = inode->root; |
6044 | struct btrfs_fs_info *fs_info = root->fs_info; | |
6045 | struct btrfs_block_rsv *block_rsv = &inode->block_rsv; | |
6046 | u64 meta_reserve, qgroup_reserve; | |
69fe2d75 | 6047 | unsigned nr_extents; |
08e007d2 | 6048 | enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL; |
eb6b88d9 | 6049 | int ret = 0; |
c64c2bd8 | 6050 | bool delalloc_lock = true; |
6324fbf3 | 6051 | |
c64c2bd8 JB |
6052 | /* If we are a free space inode we need to not flush since we will be in |
6053 | * the middle of a transaction commit. We also don't need the delalloc | |
6054 | * mutex since we won't race with anybody. We need this mostly to make | |
6055 | * lockdep shut its filthy mouth. | |
bac357dc JB |
6056 | * |
6057 | * If we have a transaction open (can happen if we call truncate_block | |
6058 | * from truncate), then we need FLUSH_LIMIT so we don't deadlock. | |
c64c2bd8 JB |
6059 | */ |
6060 | if (btrfs_is_free_space_inode(inode)) { | |
08e007d2 | 6061 | flush = BTRFS_RESERVE_NO_FLUSH; |
c64c2bd8 | 6062 | delalloc_lock = false; |
da07d4ab NB |
6063 | } else { |
6064 | if (current->journal_info) | |
6065 | flush = BTRFS_RESERVE_FLUSH_LIMIT; | |
c09544e0 | 6066 | |
da07d4ab NB |
6067 | if (btrfs_transaction_in_commit(fs_info)) |
6068 | schedule_timeout(1); | |
6069 | } | |
ec44a35c | 6070 | |
c64c2bd8 | 6071 | if (delalloc_lock) |
9f3db423 | 6072 | mutex_lock(&inode->delalloc_mutex); |
c64c2bd8 | 6073 | |
0b246afa | 6074 | num_bytes = ALIGN(num_bytes, fs_info->sectorsize); |
69fe2d75 | 6075 | |
c8eaeac7 JB |
6076 | /* |
6077 | * We always want to do it this way, every other way is wrong and ends | |
6078 | * in tears. Pre-reserving the amount we are going to add will always | |
6079 | * be the right way, because otherwise if we have enough parallelism we | |
6080 | * could end up with thousands of inodes all holding little bits of | |
6081 | * reservations they were able to make previously and the only way to | |
6082 | * reclaim that space is to ENOSPC out the operations and clear | |
6083 | * everything out and try again, which is bad. This way we just | |
6084 | * over-reserve slightly, and clean up the mess when we are done. | |
6085 | */ | |
6086 | calc_inode_reservations(fs_info, num_bytes, &meta_reserve, | |
6087 | &qgroup_reserve); | |
6088 | ret = btrfs_qgroup_reserve_meta_prealloc(root, qgroup_reserve, true); | |
6089 | if (ret) | |
6090 | goto out_fail; | |
6091 | ret = reserve_metadata_bytes(root, block_rsv, meta_reserve, flush); | |
6092 | if (ret) | |
6093 | goto out_qgroup; | |
6094 | ||
6095 | /* | |
6096 | * Now we need to update our outstanding extents and csum bytes _first_ | |
6097 | * and then add the reservation to the block_rsv. This keeps us from | |
6098 | * racing with an ordered completion or some such that would think it | |
6099 | * needs to free the reservation we just made. | |
6100 | */ | |
9f3db423 | 6101 | spin_lock(&inode->lock); |
69fe2d75 | 6102 | nr_extents = count_max_extents(num_bytes); |
8b62f87b | 6103 | btrfs_mod_outstanding_extents(inode, nr_extents); |
69fe2d75 JB |
6104 | inode->csum_bytes += num_bytes; |
6105 | btrfs_calculate_inode_block_rsv_size(fs_info, inode); | |
9f3db423 | 6106 | spin_unlock(&inode->lock); |
57a45ced | 6107 | |
c8eaeac7 JB |
6108 | /* Now we can safely add our space to our block rsv */ |
6109 | block_rsv_add_bytes(block_rsv, meta_reserve, false); | |
6110 | trace_btrfs_space_reservation(root->fs_info, "delalloc", | |
6111 | btrfs_ino(inode), meta_reserve, 1); | |
6112 | ||
6113 | spin_lock(&block_rsv->lock); | |
6114 | block_rsv->qgroup_rsv_reserved += qgroup_reserve; | |
6115 | spin_unlock(&block_rsv->lock); | |
25179201 | 6116 | |
c64c2bd8 | 6117 | if (delalloc_lock) |
9f3db423 | 6118 | mutex_unlock(&inode->delalloc_mutex); |
0ca1f7ce | 6119 | return 0; |
c8eaeac7 JB |
6120 | out_qgroup: |
6121 | btrfs_qgroup_free_meta_prealloc(root, qgroup_reserve); | |
88e081bf | 6122 | out_fail: |
43b18595 | 6123 | btrfs_inode_rsv_release(inode, true); |
88e081bf | 6124 | if (delalloc_lock) |
9f3db423 | 6125 | mutex_unlock(&inode->delalloc_mutex); |
88e081bf | 6126 | return ret; |
0ca1f7ce YZ |
6127 | } |
6128 | ||
7709cde3 JB |
6129 | /** |
6130 | * btrfs_delalloc_release_metadata - release a metadata reservation for an inode | |
8b62f87b JB |
6131 | * @inode: the inode to release the reservation for. |
6132 | * @num_bytes: the number of bytes we are releasing. | |
43b18595 | 6133 | * @qgroup_free: free qgroup reservation or convert it to per-trans reservation |
7709cde3 JB |
6134 | * |
6135 | * This will release the metadata reservation for an inode. This can be called | |
6136 | * once we complete IO for a given set of bytes to release their metadata | |
8b62f87b | 6137 | * reservations, or on error for the same reason. |
7709cde3 | 6138 | */ |
43b18595 QW |
6139 | void btrfs_delalloc_release_metadata(struct btrfs_inode *inode, u64 num_bytes, |
6140 | bool qgroup_free) | |
0ca1f7ce | 6141 | { |
3ffbd68c | 6142 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
0ca1f7ce | 6143 | |
0b246afa | 6144 | num_bytes = ALIGN(num_bytes, fs_info->sectorsize); |
691fa059 | 6145 | spin_lock(&inode->lock); |
69fe2d75 JB |
6146 | inode->csum_bytes -= num_bytes; |
6147 | btrfs_calculate_inode_block_rsv_size(fs_info, inode); | |
691fa059 | 6148 | spin_unlock(&inode->lock); |
0ca1f7ce | 6149 | |
0b246afa | 6150 | if (btrfs_is_testing(fs_info)) |
6a3891c5 JB |
6151 | return; |
6152 | ||
43b18595 | 6153 | btrfs_inode_rsv_release(inode, qgroup_free); |
0ca1f7ce YZ |
6154 | } |
6155 | ||
8b62f87b JB |
6156 | /** |
6157 | * btrfs_delalloc_release_extents - release our outstanding_extents | |
6158 | * @inode: the inode to balance the reservation for. | |
6159 | * @num_bytes: the number of bytes we originally reserved with | |
43b18595 | 6160 | * @qgroup_free: do we need to free qgroup meta reservation or convert them. |
8b62f87b JB |
6161 | * |
6162 | * When we reserve space we increase outstanding_extents for the extents we may | |
6163 | * add. Once we've set the range as delalloc or created our ordered extents we | |
6164 | * have outstanding_extents to track the real usage, so we use this to free our | |
6165 | * temporarily tracked outstanding_extents. This _must_ be used in conjunction | |
6166 | * with btrfs_delalloc_reserve_metadata. | |
6167 | */ | |
43b18595 QW |
6168 | void btrfs_delalloc_release_extents(struct btrfs_inode *inode, u64 num_bytes, |
6169 | bool qgroup_free) | |
8b62f87b | 6170 | { |
3ffbd68c | 6171 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
8b62f87b | 6172 | unsigned num_extents; |
8b62f87b JB |
6173 | |
6174 | spin_lock(&inode->lock); | |
6175 | num_extents = count_max_extents(num_bytes); | |
6176 | btrfs_mod_outstanding_extents(inode, -num_extents); | |
69fe2d75 | 6177 | btrfs_calculate_inode_block_rsv_size(fs_info, inode); |
8b62f87b JB |
6178 | spin_unlock(&inode->lock); |
6179 | ||
8b62f87b JB |
6180 | if (btrfs_is_testing(fs_info)) |
6181 | return; | |
6182 | ||
43b18595 | 6183 | btrfs_inode_rsv_release(inode, qgroup_free); |
8b62f87b JB |
6184 | } |
6185 | ||
1ada3a62 | 6186 | /** |
7cf5b976 | 6187 | * btrfs_delalloc_reserve_space - reserve data and metadata space for |
1ada3a62 QW |
6188 | * delalloc |
6189 | * @inode: inode we're writing to | |
6190 | * @start: start range we are writing to | |
6191 | * @len: how long the range we are writing to | |
364ecf36 QW |
6192 | * @reserved: mandatory parameter, record actually reserved qgroup ranges of |
6193 | * current reservation. | |
1ada3a62 | 6194 | * |
1ada3a62 QW |
6195 | * This will do the following things |
6196 | * | |
6197 | * o reserve space in data space info for num bytes | |
6198 | * and reserve precious corresponding qgroup space | |
6199 | * (Done in check_data_free_space) | |
6200 | * | |
6201 | * o reserve space for metadata space, based on the number of outstanding | |
6202 | * extents and how much csums will be needed | |
6203 | * also reserve metadata space in a per root over-reserve method. | |
6204 | * o add to the inodes->delalloc_bytes | |
6205 | * o add it to the fs_info's delalloc inodes list. | |
6206 | * (Above 3 all done in delalloc_reserve_metadata) | |
6207 | * | |
6208 | * Return 0 for success | |
6209 | * Return <0 for error(-ENOSPC or -EQUOT) | |
6210 | */ | |
364ecf36 QW |
6211 | int btrfs_delalloc_reserve_space(struct inode *inode, |
6212 | struct extent_changeset **reserved, u64 start, u64 len) | |
1ada3a62 QW |
6213 | { |
6214 | int ret; | |
6215 | ||
364ecf36 | 6216 | ret = btrfs_check_data_free_space(inode, reserved, start, len); |
1ada3a62 QW |
6217 | if (ret < 0) |
6218 | return ret; | |
9f3db423 | 6219 | ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len); |
1ada3a62 | 6220 | if (ret < 0) |
bc42bda2 | 6221 | btrfs_free_reserved_data_space(inode, *reserved, start, len); |
1ada3a62 QW |
6222 | return ret; |
6223 | } | |
6224 | ||
7709cde3 | 6225 | /** |
7cf5b976 | 6226 | * btrfs_delalloc_release_space - release data and metadata space for delalloc |
1ada3a62 QW |
6227 | * @inode: inode we're releasing space for |
6228 | * @start: start position of the space already reserved | |
6229 | * @len: the len of the space already reserved | |
8b62f87b | 6230 | * @release_bytes: the len of the space we consumed or didn't use |
1ada3a62 QW |
6231 | * |
6232 | * This function will release the metadata space that was not used and will | |
6233 | * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes | |
6234 | * list if there are no delalloc bytes left. | |
6235 | * Also it will handle the qgroup reserved space. | |
6236 | */ | |
bc42bda2 | 6237 | void btrfs_delalloc_release_space(struct inode *inode, |
8b62f87b | 6238 | struct extent_changeset *reserved, |
43b18595 | 6239 | u64 start, u64 len, bool qgroup_free) |
1ada3a62 | 6240 | { |
43b18595 | 6241 | btrfs_delalloc_release_metadata(BTRFS_I(inode), len, qgroup_free); |
bc42bda2 | 6242 | btrfs_free_reserved_data_space(inode, reserved, start, len); |
6324fbf3 CM |
6243 | } |
6244 | ||
ce93ec54 | 6245 | static int update_block_group(struct btrfs_trans_handle *trans, |
6b279408 | 6246 | u64 bytenr, u64 num_bytes, int alloc) |
9078a3e1 | 6247 | { |
6b279408 | 6248 | struct btrfs_fs_info *info = trans->fs_info; |
0af3d00b | 6249 | struct btrfs_block_group_cache *cache = NULL; |
db94535d | 6250 | u64 total = num_bytes; |
9078a3e1 | 6251 | u64 old_val; |
db94535d | 6252 | u64 byte_in_group; |
0af3d00b | 6253 | int factor; |
ba2c4d4e | 6254 | int ret = 0; |
3e1ad54f | 6255 | |
5d4f98a2 | 6256 | /* block accounting for super block */ |
eb73c1b7 | 6257 | spin_lock(&info->delalloc_root_lock); |
6c41761f | 6258 | old_val = btrfs_super_bytes_used(info->super_copy); |
5d4f98a2 YZ |
6259 | if (alloc) |
6260 | old_val += num_bytes; | |
6261 | else | |
6262 | old_val -= num_bytes; | |
6c41761f | 6263 | btrfs_set_super_bytes_used(info->super_copy, old_val); |
eb73c1b7 | 6264 | spin_unlock(&info->delalloc_root_lock); |
5d4f98a2 | 6265 | |
d397712b | 6266 | while (total) { |
db94535d | 6267 | cache = btrfs_lookup_block_group(info, bytenr); |
ba2c4d4e JB |
6268 | if (!cache) { |
6269 | ret = -ENOENT; | |
6270 | break; | |
6271 | } | |
46df06b8 DS |
6272 | factor = btrfs_bg_type_to_factor(cache->flags); |
6273 | ||
9d66e233 JB |
6274 | /* |
6275 | * If this block group has free space cache written out, we | |
6276 | * need to make sure to load it if we are removing space. This | |
6277 | * is because we need the unpinning stage to actually add the | |
6278 | * space back to the block group, otherwise we will leak space. | |
6279 | */ | |
6280 | if (!alloc && cache->cached == BTRFS_CACHE_NO) | |
f6373bf3 | 6281 | cache_block_group(cache, 1); |
0af3d00b | 6282 | |
db94535d CM |
6283 | byte_in_group = bytenr - cache->key.objectid; |
6284 | WARN_ON(byte_in_group > cache->key.offset); | |
9078a3e1 | 6285 | |
25179201 | 6286 | spin_lock(&cache->space_info->lock); |
c286ac48 | 6287 | spin_lock(&cache->lock); |
0af3d00b | 6288 | |
6202df69 | 6289 | if (btrfs_test_opt(info, SPACE_CACHE) && |
0af3d00b JB |
6290 | cache->disk_cache_state < BTRFS_DC_CLEAR) |
6291 | cache->disk_cache_state = BTRFS_DC_CLEAR; | |
6292 | ||
9078a3e1 | 6293 | old_val = btrfs_block_group_used(&cache->item); |
db94535d | 6294 | num_bytes = min(total, cache->key.offset - byte_in_group); |
cd1bc465 | 6295 | if (alloc) { |
db94535d | 6296 | old_val += num_bytes; |
11833d66 YZ |
6297 | btrfs_set_block_group_used(&cache->item, old_val); |
6298 | cache->reserved -= num_bytes; | |
11833d66 | 6299 | cache->space_info->bytes_reserved -= num_bytes; |
b742bb82 YZ |
6300 | cache->space_info->bytes_used += num_bytes; |
6301 | cache->space_info->disk_used += num_bytes * factor; | |
c286ac48 | 6302 | spin_unlock(&cache->lock); |
25179201 | 6303 | spin_unlock(&cache->space_info->lock); |
cd1bc465 | 6304 | } else { |
db94535d | 6305 | old_val -= num_bytes; |
ae0ab003 FM |
6306 | btrfs_set_block_group_used(&cache->item, old_val); |
6307 | cache->pinned += num_bytes; | |
480b9b4d | 6308 | update_bytes_pinned(info, cache->space_info, num_bytes); |
ae0ab003 FM |
6309 | cache->space_info->bytes_used -= num_bytes; |
6310 | cache->space_info->disk_used -= num_bytes * factor; | |
6311 | spin_unlock(&cache->lock); | |
6312 | spin_unlock(&cache->space_info->lock); | |
47ab2a6c | 6313 | |
0b246afa | 6314 | trace_btrfs_space_reservation(info, "pinned", |
c51e7bb1 JB |
6315 | cache->space_info->flags, |
6316 | num_bytes, 1); | |
dec59fa3 EL |
6317 | percpu_counter_add_batch(&cache->space_info->total_bytes_pinned, |
6318 | num_bytes, | |
6319 | BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
ae0ab003 FM |
6320 | set_extent_dirty(info->pinned_extents, |
6321 | bytenr, bytenr + num_bytes - 1, | |
6322 | GFP_NOFS | __GFP_NOFAIL); | |
cd1bc465 | 6323 | } |
1bbc621e CM |
6324 | |
6325 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
6326 | if (list_empty(&cache->dirty_list)) { | |
6327 | list_add_tail(&cache->dirty_list, | |
6328 | &trans->transaction->dirty_bgs); | |
ba2c4d4e | 6329 | trans->delayed_ref_updates++; |
1bbc621e CM |
6330 | btrfs_get_block_group(cache); |
6331 | } | |
6332 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
6333 | ||
036a9348 FM |
6334 | /* |
6335 | * No longer have used bytes in this block group, queue it for | |
6336 | * deletion. We do this after adding the block group to the | |
6337 | * dirty list to avoid races between cleaner kthread and space | |
6338 | * cache writeout. | |
6339 | */ | |
031f24da QW |
6340 | if (!alloc && old_val == 0) |
6341 | btrfs_mark_bg_unused(cache); | |
036a9348 | 6342 | |
fa9c0d79 | 6343 | btrfs_put_block_group(cache); |
db94535d CM |
6344 | total -= num_bytes; |
6345 | bytenr += num_bytes; | |
9078a3e1 | 6346 | } |
ba2c4d4e JB |
6347 | |
6348 | /* Modified block groups are accounted for in the delayed_refs_rsv. */ | |
6349 | btrfs_update_delayed_refs_rsv(trans); | |
6350 | return ret; | |
9078a3e1 | 6351 | } |
6324fbf3 | 6352 | |
2ff7e61e | 6353 | static u64 first_logical_byte(struct btrfs_fs_info *fs_info, u64 search_start) |
a061fc8d | 6354 | { |
0f9dd46c | 6355 | struct btrfs_block_group_cache *cache; |
d2fb3437 | 6356 | u64 bytenr; |
0f9dd46c | 6357 | |
0b246afa JM |
6358 | spin_lock(&fs_info->block_group_cache_lock); |
6359 | bytenr = fs_info->first_logical_byte; | |
6360 | spin_unlock(&fs_info->block_group_cache_lock); | |
a1897fdd LB |
6361 | |
6362 | if (bytenr < (u64)-1) | |
6363 | return bytenr; | |
6364 | ||
0b246afa | 6365 | cache = btrfs_lookup_first_block_group(fs_info, search_start); |
0f9dd46c | 6366 | if (!cache) |
a061fc8d | 6367 | return 0; |
0f9dd46c | 6368 | |
d2fb3437 | 6369 | bytenr = cache->key.objectid; |
fa9c0d79 | 6370 | btrfs_put_block_group(cache); |
d2fb3437 YZ |
6371 | |
6372 | return bytenr; | |
a061fc8d CM |
6373 | } |
6374 | ||
fdf08605 | 6375 | static int pin_down_extent(struct btrfs_block_group_cache *cache, |
f0486c68 | 6376 | u64 bytenr, u64 num_bytes, int reserved) |
324ae4df | 6377 | { |
fdf08605 DS |
6378 | struct btrfs_fs_info *fs_info = cache->fs_info; |
6379 | ||
11833d66 YZ |
6380 | spin_lock(&cache->space_info->lock); |
6381 | spin_lock(&cache->lock); | |
6382 | cache->pinned += num_bytes; | |
480b9b4d | 6383 | update_bytes_pinned(fs_info, cache->space_info, num_bytes); |
11833d66 YZ |
6384 | if (reserved) { |
6385 | cache->reserved -= num_bytes; | |
6386 | cache->space_info->bytes_reserved -= num_bytes; | |
6387 | } | |
6388 | spin_unlock(&cache->lock); | |
6389 | spin_unlock(&cache->space_info->lock); | |
68b38550 | 6390 | |
0b246afa | 6391 | trace_btrfs_space_reservation(fs_info, "pinned", |
c51e7bb1 | 6392 | cache->space_info->flags, num_bytes, 1); |
dec59fa3 EL |
6393 | percpu_counter_add_batch(&cache->space_info->total_bytes_pinned, |
6394 | num_bytes, BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
0b246afa | 6395 | set_extent_dirty(fs_info->pinned_extents, bytenr, |
f0486c68 YZ |
6396 | bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL); |
6397 | return 0; | |
6398 | } | |
68b38550 | 6399 | |
f0486c68 YZ |
6400 | /* |
6401 | * this function must be called within transaction | |
6402 | */ | |
2ff7e61e | 6403 | int btrfs_pin_extent(struct btrfs_fs_info *fs_info, |
f0486c68 YZ |
6404 | u64 bytenr, u64 num_bytes, int reserved) |
6405 | { | |
6406 | struct btrfs_block_group_cache *cache; | |
68b38550 | 6407 | |
0b246afa | 6408 | cache = btrfs_lookup_block_group(fs_info, bytenr); |
79787eaa | 6409 | BUG_ON(!cache); /* Logic error */ |
f0486c68 | 6410 | |
fdf08605 | 6411 | pin_down_extent(cache, bytenr, num_bytes, reserved); |
f0486c68 YZ |
6412 | |
6413 | btrfs_put_block_group(cache); | |
11833d66 YZ |
6414 | return 0; |
6415 | } | |
6416 | ||
f0486c68 | 6417 | /* |
e688b725 CM |
6418 | * this function must be called within transaction |
6419 | */ | |
2ff7e61e | 6420 | int btrfs_pin_extent_for_log_replay(struct btrfs_fs_info *fs_info, |
e688b725 CM |
6421 | u64 bytenr, u64 num_bytes) |
6422 | { | |
6423 | struct btrfs_block_group_cache *cache; | |
b50c6e25 | 6424 | int ret; |
e688b725 | 6425 | |
0b246afa | 6426 | cache = btrfs_lookup_block_group(fs_info, bytenr); |
b50c6e25 JB |
6427 | if (!cache) |
6428 | return -EINVAL; | |
e688b725 CM |
6429 | |
6430 | /* | |
6431 | * pull in the free space cache (if any) so that our pin | |
6432 | * removes the free space from the cache. We have load_only set | |
6433 | * to one because the slow code to read in the free extents does check | |
6434 | * the pinned extents. | |
6435 | */ | |
f6373bf3 | 6436 | cache_block_group(cache, 1); |
e688b725 | 6437 | |
fdf08605 | 6438 | pin_down_extent(cache, bytenr, num_bytes, 0); |
e688b725 CM |
6439 | |
6440 | /* remove us from the free space cache (if we're there at all) */ | |
b50c6e25 | 6441 | ret = btrfs_remove_free_space(cache, bytenr, num_bytes); |
e688b725 | 6442 | btrfs_put_block_group(cache); |
b50c6e25 | 6443 | return ret; |
e688b725 CM |
6444 | } |
6445 | ||
2ff7e61e JM |
6446 | static int __exclude_logged_extent(struct btrfs_fs_info *fs_info, |
6447 | u64 start, u64 num_bytes) | |
8c2a1a30 JB |
6448 | { |
6449 | int ret; | |
6450 | struct btrfs_block_group_cache *block_group; | |
6451 | struct btrfs_caching_control *caching_ctl; | |
6452 | ||
0b246afa | 6453 | block_group = btrfs_lookup_block_group(fs_info, start); |
8c2a1a30 JB |
6454 | if (!block_group) |
6455 | return -EINVAL; | |
6456 | ||
6457 | cache_block_group(block_group, 0); | |
6458 | caching_ctl = get_caching_control(block_group); | |
6459 | ||
6460 | if (!caching_ctl) { | |
6461 | /* Logic error */ | |
6462 | BUG_ON(!block_group_cache_done(block_group)); | |
6463 | ret = btrfs_remove_free_space(block_group, start, num_bytes); | |
6464 | } else { | |
6465 | mutex_lock(&caching_ctl->mutex); | |
6466 | ||
6467 | if (start >= caching_ctl->progress) { | |
2ff7e61e | 6468 | ret = add_excluded_extent(fs_info, start, num_bytes); |
8c2a1a30 JB |
6469 | } else if (start + num_bytes <= caching_ctl->progress) { |
6470 | ret = btrfs_remove_free_space(block_group, | |
6471 | start, num_bytes); | |
6472 | } else { | |
6473 | num_bytes = caching_ctl->progress - start; | |
6474 | ret = btrfs_remove_free_space(block_group, | |
6475 | start, num_bytes); | |
6476 | if (ret) | |
6477 | goto out_lock; | |
6478 | ||
6479 | num_bytes = (start + num_bytes) - | |
6480 | caching_ctl->progress; | |
6481 | start = caching_ctl->progress; | |
2ff7e61e | 6482 | ret = add_excluded_extent(fs_info, start, num_bytes); |
8c2a1a30 JB |
6483 | } |
6484 | out_lock: | |
6485 | mutex_unlock(&caching_ctl->mutex); | |
6486 | put_caching_control(caching_ctl); | |
6487 | } | |
6488 | btrfs_put_block_group(block_group); | |
6489 | return ret; | |
6490 | } | |
6491 | ||
bcdc428c | 6492 | int btrfs_exclude_logged_extents(struct extent_buffer *eb) |
8c2a1a30 | 6493 | { |
bcdc428c | 6494 | struct btrfs_fs_info *fs_info = eb->fs_info; |
8c2a1a30 JB |
6495 | struct btrfs_file_extent_item *item; |
6496 | struct btrfs_key key; | |
6497 | int found_type; | |
6498 | int i; | |
b89311ef | 6499 | int ret = 0; |
8c2a1a30 | 6500 | |
2ff7e61e | 6501 | if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) |
8c2a1a30 JB |
6502 | return 0; |
6503 | ||
6504 | for (i = 0; i < btrfs_header_nritems(eb); i++) { | |
6505 | btrfs_item_key_to_cpu(eb, &key, i); | |
6506 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
6507 | continue; | |
6508 | item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item); | |
6509 | found_type = btrfs_file_extent_type(eb, item); | |
6510 | if (found_type == BTRFS_FILE_EXTENT_INLINE) | |
6511 | continue; | |
6512 | if (btrfs_file_extent_disk_bytenr(eb, item) == 0) | |
6513 | continue; | |
6514 | key.objectid = btrfs_file_extent_disk_bytenr(eb, item); | |
6515 | key.offset = btrfs_file_extent_disk_num_bytes(eb, item); | |
b89311ef GJ |
6516 | ret = __exclude_logged_extent(fs_info, key.objectid, key.offset); |
6517 | if (ret) | |
6518 | break; | |
8c2a1a30 JB |
6519 | } |
6520 | ||
b89311ef | 6521 | return ret; |
8c2a1a30 JB |
6522 | } |
6523 | ||
9cfa3e34 FM |
6524 | static void |
6525 | btrfs_inc_block_group_reservations(struct btrfs_block_group_cache *bg) | |
6526 | { | |
6527 | atomic_inc(&bg->reservations); | |
6528 | } | |
6529 | ||
6530 | void btrfs_dec_block_group_reservations(struct btrfs_fs_info *fs_info, | |
6531 | const u64 start) | |
6532 | { | |
6533 | struct btrfs_block_group_cache *bg; | |
6534 | ||
6535 | bg = btrfs_lookup_block_group(fs_info, start); | |
6536 | ASSERT(bg); | |
6537 | if (atomic_dec_and_test(&bg->reservations)) | |
4625956a | 6538 | wake_up_var(&bg->reservations); |
9cfa3e34 FM |
6539 | btrfs_put_block_group(bg); |
6540 | } | |
6541 | ||
9cfa3e34 FM |
6542 | void btrfs_wait_block_group_reservations(struct btrfs_block_group_cache *bg) |
6543 | { | |
6544 | struct btrfs_space_info *space_info = bg->space_info; | |
6545 | ||
6546 | ASSERT(bg->ro); | |
6547 | ||
6548 | if (!(bg->flags & BTRFS_BLOCK_GROUP_DATA)) | |
6549 | return; | |
6550 | ||
6551 | /* | |
6552 | * Our block group is read only but before we set it to read only, | |
6553 | * some task might have had allocated an extent from it already, but it | |
6554 | * has not yet created a respective ordered extent (and added it to a | |
6555 | * root's list of ordered extents). | |
6556 | * Therefore wait for any task currently allocating extents, since the | |
6557 | * block group's reservations counter is incremented while a read lock | |
6558 | * on the groups' semaphore is held and decremented after releasing | |
6559 | * the read access on that semaphore and creating the ordered extent. | |
6560 | */ | |
6561 | down_write(&space_info->groups_sem); | |
6562 | up_write(&space_info->groups_sem); | |
6563 | ||
4625956a | 6564 | wait_var_event(&bg->reservations, !atomic_read(&bg->reservations)); |
9cfa3e34 FM |
6565 | } |
6566 | ||
fb25e914 | 6567 | /** |
4824f1f4 | 6568 | * btrfs_add_reserved_bytes - update the block_group and space info counters |
fb25e914 | 6569 | * @cache: The cache we are manipulating |
18513091 WX |
6570 | * @ram_bytes: The number of bytes of file content, and will be same to |
6571 | * @num_bytes except for the compress path. | |
fb25e914 | 6572 | * @num_bytes: The number of bytes in question |
e570fd27 | 6573 | * @delalloc: The blocks are allocated for the delalloc write |
fb25e914 | 6574 | * |
745699ef XW |
6575 | * This is called by the allocator when it reserves space. If this is a |
6576 | * reservation and the block group has become read only we cannot make the | |
6577 | * reservation and return -EAGAIN, otherwise this function always succeeds. | |
f0486c68 | 6578 | */ |
4824f1f4 | 6579 | static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache *cache, |
18513091 | 6580 | u64 ram_bytes, u64 num_bytes, int delalloc) |
11833d66 | 6581 | { |
fb25e914 | 6582 | struct btrfs_space_info *space_info = cache->space_info; |
f0486c68 | 6583 | int ret = 0; |
79787eaa | 6584 | |
fb25e914 JB |
6585 | spin_lock(&space_info->lock); |
6586 | spin_lock(&cache->lock); | |
4824f1f4 WX |
6587 | if (cache->ro) { |
6588 | ret = -EAGAIN; | |
fb25e914 | 6589 | } else { |
4824f1f4 WX |
6590 | cache->reserved += num_bytes; |
6591 | space_info->bytes_reserved += num_bytes; | |
480b9b4d | 6592 | update_bytes_may_use(cache->fs_info, space_info, -ram_bytes); |
e570fd27 | 6593 | if (delalloc) |
4824f1f4 | 6594 | cache->delalloc_bytes += num_bytes; |
324ae4df | 6595 | } |
fb25e914 JB |
6596 | spin_unlock(&cache->lock); |
6597 | spin_unlock(&space_info->lock); | |
f0486c68 | 6598 | return ret; |
324ae4df | 6599 | } |
9078a3e1 | 6600 | |
4824f1f4 WX |
6601 | /** |
6602 | * btrfs_free_reserved_bytes - update the block_group and space info counters | |
6603 | * @cache: The cache we are manipulating | |
6604 | * @num_bytes: The number of bytes in question | |
6605 | * @delalloc: The blocks are allocated for the delalloc write | |
6606 | * | |
6607 | * This is called by somebody who is freeing space that was never actually used | |
6608 | * on disk. For example if you reserve some space for a new leaf in transaction | |
6609 | * A and before transaction A commits you free that leaf, you call this with | |
6610 | * reserve set to 0 in order to clear the reservation. | |
6611 | */ | |
6612 | ||
556f3ca8 | 6613 | static void btrfs_free_reserved_bytes(struct btrfs_block_group_cache *cache, |
6614 | u64 num_bytes, int delalloc) | |
4824f1f4 WX |
6615 | { |
6616 | struct btrfs_space_info *space_info = cache->space_info; | |
4824f1f4 WX |
6617 | |
6618 | spin_lock(&space_info->lock); | |
6619 | spin_lock(&cache->lock); | |
6620 | if (cache->ro) | |
6621 | space_info->bytes_readonly += num_bytes; | |
6622 | cache->reserved -= num_bytes; | |
6623 | space_info->bytes_reserved -= num_bytes; | |
21a94f7a | 6624 | space_info->max_extent_size = 0; |
4824f1f4 WX |
6625 | |
6626 | if (delalloc) | |
6627 | cache->delalloc_bytes -= num_bytes; | |
6628 | spin_unlock(&cache->lock); | |
6629 | spin_unlock(&space_info->lock); | |
4824f1f4 | 6630 | } |
8b74c03e | 6631 | void btrfs_prepare_extent_commit(struct btrfs_fs_info *fs_info) |
e8569813 | 6632 | { |
11833d66 YZ |
6633 | struct btrfs_caching_control *next; |
6634 | struct btrfs_caching_control *caching_ctl; | |
6635 | struct btrfs_block_group_cache *cache; | |
e8569813 | 6636 | |
9e351cc8 | 6637 | down_write(&fs_info->commit_root_sem); |
25179201 | 6638 | |
11833d66 YZ |
6639 | list_for_each_entry_safe(caching_ctl, next, |
6640 | &fs_info->caching_block_groups, list) { | |
6641 | cache = caching_ctl->block_group; | |
6642 | if (block_group_cache_done(cache)) { | |
6643 | cache->last_byte_to_unpin = (u64)-1; | |
6644 | list_del_init(&caching_ctl->list); | |
6645 | put_caching_control(caching_ctl); | |
e8569813 | 6646 | } else { |
11833d66 | 6647 | cache->last_byte_to_unpin = caching_ctl->progress; |
e8569813 | 6648 | } |
e8569813 | 6649 | } |
11833d66 YZ |
6650 | |
6651 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) | |
6652 | fs_info->pinned_extents = &fs_info->freed_extents[1]; | |
6653 | else | |
6654 | fs_info->pinned_extents = &fs_info->freed_extents[0]; | |
6655 | ||
9e351cc8 | 6656 | up_write(&fs_info->commit_root_sem); |
8929ecfa YZ |
6657 | |
6658 | update_global_block_rsv(fs_info); | |
e8569813 ZY |
6659 | } |
6660 | ||
c759c4e1 JB |
6661 | /* |
6662 | * Returns the free cluster for the given space info and sets empty_cluster to | |
6663 | * what it should be based on the mount options. | |
6664 | */ | |
6665 | static struct btrfs_free_cluster * | |
2ff7e61e JM |
6666 | fetch_cluster_info(struct btrfs_fs_info *fs_info, |
6667 | struct btrfs_space_info *space_info, u64 *empty_cluster) | |
c759c4e1 JB |
6668 | { |
6669 | struct btrfs_free_cluster *ret = NULL; | |
c759c4e1 JB |
6670 | |
6671 | *empty_cluster = 0; | |
6672 | if (btrfs_mixed_space_info(space_info)) | |
6673 | return ret; | |
6674 | ||
c759c4e1 | 6675 | if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) { |
0b246afa | 6676 | ret = &fs_info->meta_alloc_cluster; |
583b7231 HK |
6677 | if (btrfs_test_opt(fs_info, SSD)) |
6678 | *empty_cluster = SZ_2M; | |
6679 | else | |
ee22184b | 6680 | *empty_cluster = SZ_64K; |
583b7231 HK |
6681 | } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) && |
6682 | btrfs_test_opt(fs_info, SSD_SPREAD)) { | |
6683 | *empty_cluster = SZ_2M; | |
0b246afa | 6684 | ret = &fs_info->data_alloc_cluster; |
c759c4e1 JB |
6685 | } |
6686 | ||
6687 | return ret; | |
6688 | } | |
6689 | ||
2ff7e61e JM |
6690 | static int unpin_extent_range(struct btrfs_fs_info *fs_info, |
6691 | u64 start, u64 end, | |
678886bd | 6692 | const bool return_free_space) |
ccd467d6 | 6693 | { |
11833d66 | 6694 | struct btrfs_block_group_cache *cache = NULL; |
7b398f8e JB |
6695 | struct btrfs_space_info *space_info; |
6696 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
c759c4e1 | 6697 | struct btrfs_free_cluster *cluster = NULL; |
11833d66 | 6698 | u64 len; |
c759c4e1 JB |
6699 | u64 total_unpinned = 0; |
6700 | u64 empty_cluster = 0; | |
7b398f8e | 6701 | bool readonly; |
ccd467d6 | 6702 | |
11833d66 | 6703 | while (start <= end) { |
7b398f8e | 6704 | readonly = false; |
11833d66 YZ |
6705 | if (!cache || |
6706 | start >= cache->key.objectid + cache->key.offset) { | |
6707 | if (cache) | |
6708 | btrfs_put_block_group(cache); | |
c759c4e1 | 6709 | total_unpinned = 0; |
11833d66 | 6710 | cache = btrfs_lookup_block_group(fs_info, start); |
79787eaa | 6711 | BUG_ON(!cache); /* Logic error */ |
c759c4e1 | 6712 | |
2ff7e61e | 6713 | cluster = fetch_cluster_info(fs_info, |
c759c4e1 JB |
6714 | cache->space_info, |
6715 | &empty_cluster); | |
6716 | empty_cluster <<= 1; | |
11833d66 YZ |
6717 | } |
6718 | ||
6719 | len = cache->key.objectid + cache->key.offset - start; | |
6720 | len = min(len, end + 1 - start); | |
6721 | ||
6722 | if (start < cache->last_byte_to_unpin) { | |
6723 | len = min(len, cache->last_byte_to_unpin - start); | |
678886bd FM |
6724 | if (return_free_space) |
6725 | btrfs_add_free_space(cache, start, len); | |
11833d66 YZ |
6726 | } |
6727 | ||
f0486c68 | 6728 | start += len; |
c759c4e1 | 6729 | total_unpinned += len; |
7b398f8e | 6730 | space_info = cache->space_info; |
f0486c68 | 6731 | |
c759c4e1 JB |
6732 | /* |
6733 | * If this space cluster has been marked as fragmented and we've | |
6734 | * unpinned enough in this block group to potentially allow a | |
6735 | * cluster to be created inside of it go ahead and clear the | |
6736 | * fragmented check. | |
6737 | */ | |
6738 | if (cluster && cluster->fragmented && | |
6739 | total_unpinned > empty_cluster) { | |
6740 | spin_lock(&cluster->lock); | |
6741 | cluster->fragmented = 0; | |
6742 | spin_unlock(&cluster->lock); | |
6743 | } | |
6744 | ||
7b398f8e | 6745 | spin_lock(&space_info->lock); |
11833d66 YZ |
6746 | spin_lock(&cache->lock); |
6747 | cache->pinned -= len; | |
480b9b4d | 6748 | update_bytes_pinned(fs_info, space_info, -len); |
c51e7bb1 JB |
6749 | |
6750 | trace_btrfs_space_reservation(fs_info, "pinned", | |
6751 | space_info->flags, len, 0); | |
4f4db217 | 6752 | space_info->max_extent_size = 0; |
dec59fa3 EL |
6753 | percpu_counter_add_batch(&space_info->total_bytes_pinned, |
6754 | -len, BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
7b398f8e JB |
6755 | if (cache->ro) { |
6756 | space_info->bytes_readonly += len; | |
6757 | readonly = true; | |
6758 | } | |
11833d66 | 6759 | spin_unlock(&cache->lock); |
957780eb JB |
6760 | if (!readonly && return_free_space && |
6761 | global_rsv->space_info == space_info) { | |
6762 | u64 to_add = len; | |
92ac58ec | 6763 | |
7b398f8e JB |
6764 | spin_lock(&global_rsv->lock); |
6765 | if (!global_rsv->full) { | |
957780eb JB |
6766 | to_add = min(len, global_rsv->size - |
6767 | global_rsv->reserved); | |
6768 | global_rsv->reserved += to_add; | |
480b9b4d QW |
6769 | update_bytes_may_use(fs_info, space_info, |
6770 | to_add); | |
7b398f8e JB |
6771 | if (global_rsv->reserved >= global_rsv->size) |
6772 | global_rsv->full = 1; | |
957780eb JB |
6773 | trace_btrfs_space_reservation(fs_info, |
6774 | "space_info", | |
6775 | space_info->flags, | |
6776 | to_add, 1); | |
6777 | len -= to_add; | |
7b398f8e JB |
6778 | } |
6779 | spin_unlock(&global_rsv->lock); | |
957780eb JB |
6780 | /* Add to any tickets we may have */ |
6781 | if (len) | |
6782 | space_info_add_new_bytes(fs_info, space_info, | |
6783 | len); | |
7b398f8e JB |
6784 | } |
6785 | spin_unlock(&space_info->lock); | |
ccd467d6 | 6786 | } |
11833d66 YZ |
6787 | |
6788 | if (cache) | |
6789 | btrfs_put_block_group(cache); | |
ccd467d6 CM |
6790 | return 0; |
6791 | } | |
6792 | ||
5ead2dd0 | 6793 | int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans) |
a28ec197 | 6794 | { |
5ead2dd0 | 6795 | struct btrfs_fs_info *fs_info = trans->fs_info; |
e33e17ee JM |
6796 | struct btrfs_block_group_cache *block_group, *tmp; |
6797 | struct list_head *deleted_bgs; | |
11833d66 | 6798 | struct extent_io_tree *unpin; |
1a5bc167 CM |
6799 | u64 start; |
6800 | u64 end; | |
a28ec197 | 6801 | int ret; |
a28ec197 | 6802 | |
11833d66 YZ |
6803 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) |
6804 | unpin = &fs_info->freed_extents[1]; | |
6805 | else | |
6806 | unpin = &fs_info->freed_extents[0]; | |
6807 | ||
e33e17ee | 6808 | while (!trans->aborted) { |
0e6ec385 FM |
6809 | struct extent_state *cached_state = NULL; |
6810 | ||
d4b450cd | 6811 | mutex_lock(&fs_info->unused_bg_unpin_mutex); |
1a5bc167 | 6812 | ret = find_first_extent_bit(unpin, 0, &start, &end, |
0e6ec385 | 6813 | EXTENT_DIRTY, &cached_state); |
d4b450cd FM |
6814 | if (ret) { |
6815 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); | |
a28ec197 | 6816 | break; |
d4b450cd | 6817 | } |
1f3c79a2 | 6818 | |
0b246afa | 6819 | if (btrfs_test_opt(fs_info, DISCARD)) |
2ff7e61e | 6820 | ret = btrfs_discard_extent(fs_info, start, |
5378e607 | 6821 | end + 1 - start, NULL); |
1f3c79a2 | 6822 | |
0e6ec385 | 6823 | clear_extent_dirty(unpin, start, end, &cached_state); |
2ff7e61e | 6824 | unpin_extent_range(fs_info, start, end, true); |
d4b450cd | 6825 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
0e6ec385 | 6826 | free_extent_state(cached_state); |
b9473439 | 6827 | cond_resched(); |
a28ec197 | 6828 | } |
817d52f8 | 6829 | |
e33e17ee JM |
6830 | /* |
6831 | * Transaction is finished. We don't need the lock anymore. We | |
6832 | * do need to clean up the block groups in case of a transaction | |
6833 | * abort. | |
6834 | */ | |
6835 | deleted_bgs = &trans->transaction->deleted_bgs; | |
6836 | list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) { | |
6837 | u64 trimmed = 0; | |
6838 | ||
6839 | ret = -EROFS; | |
6840 | if (!trans->aborted) | |
2ff7e61e | 6841 | ret = btrfs_discard_extent(fs_info, |
e33e17ee JM |
6842 | block_group->key.objectid, |
6843 | block_group->key.offset, | |
6844 | &trimmed); | |
6845 | ||
6846 | list_del_init(&block_group->bg_list); | |
6847 | btrfs_put_block_group_trimming(block_group); | |
6848 | btrfs_put_block_group(block_group); | |
6849 | ||
6850 | if (ret) { | |
6851 | const char *errstr = btrfs_decode_error(ret); | |
6852 | btrfs_warn(fs_info, | |
913e1535 | 6853 | "discard failed while removing blockgroup: errno=%d %s", |
e33e17ee JM |
6854 | ret, errstr); |
6855 | } | |
6856 | } | |
6857 | ||
e20d96d6 CM |
6858 | return 0; |
6859 | } | |
6860 | ||
5d4f98a2 | 6861 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
e72cb923 NB |
6862 | struct btrfs_delayed_ref_node *node, u64 parent, |
6863 | u64 root_objectid, u64 owner_objectid, | |
6864 | u64 owner_offset, int refs_to_drop, | |
6865 | struct btrfs_delayed_extent_op *extent_op) | |
a28ec197 | 6866 | { |
e72cb923 | 6867 | struct btrfs_fs_info *info = trans->fs_info; |
e2fa7227 | 6868 | struct btrfs_key key; |
5d4f98a2 | 6869 | struct btrfs_path *path; |
1261ec42 | 6870 | struct btrfs_root *extent_root = info->extent_root; |
5f39d397 | 6871 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
6872 | struct btrfs_extent_item *ei; |
6873 | struct btrfs_extent_inline_ref *iref; | |
a28ec197 | 6874 | int ret; |
5d4f98a2 | 6875 | int is_data; |
952fccac CM |
6876 | int extent_slot = 0; |
6877 | int found_extent = 0; | |
6878 | int num_to_del = 1; | |
5d4f98a2 YZ |
6879 | u32 item_size; |
6880 | u64 refs; | |
c682f9b3 QW |
6881 | u64 bytenr = node->bytenr; |
6882 | u64 num_bytes = node->num_bytes; | |
fcebe456 | 6883 | int last_ref = 0; |
0b246afa | 6884 | bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA); |
037e6390 | 6885 | |
5caf2a00 | 6886 | path = btrfs_alloc_path(); |
54aa1f4d CM |
6887 | if (!path) |
6888 | return -ENOMEM; | |
5f26f772 | 6889 | |
e4058b54 | 6890 | path->reada = READA_FORWARD; |
b9473439 | 6891 | path->leave_spinning = 1; |
5d4f98a2 YZ |
6892 | |
6893 | is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID; | |
6894 | BUG_ON(!is_data && refs_to_drop != 1); | |
6895 | ||
3173a18f | 6896 | if (is_data) |
897ca819 | 6897 | skinny_metadata = false; |
3173a18f | 6898 | |
fbe4801b NB |
6899 | ret = lookup_extent_backref(trans, path, &iref, bytenr, num_bytes, |
6900 | parent, root_objectid, owner_objectid, | |
5d4f98a2 | 6901 | owner_offset); |
7bb86316 | 6902 | if (ret == 0) { |
952fccac | 6903 | extent_slot = path->slots[0]; |
5d4f98a2 YZ |
6904 | while (extent_slot >= 0) { |
6905 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
952fccac | 6906 | extent_slot); |
5d4f98a2 | 6907 | if (key.objectid != bytenr) |
952fccac | 6908 | break; |
5d4f98a2 YZ |
6909 | if (key.type == BTRFS_EXTENT_ITEM_KEY && |
6910 | key.offset == num_bytes) { | |
952fccac CM |
6911 | found_extent = 1; |
6912 | break; | |
6913 | } | |
3173a18f JB |
6914 | if (key.type == BTRFS_METADATA_ITEM_KEY && |
6915 | key.offset == owner_objectid) { | |
6916 | found_extent = 1; | |
6917 | break; | |
6918 | } | |
952fccac CM |
6919 | if (path->slots[0] - extent_slot > 5) |
6920 | break; | |
5d4f98a2 | 6921 | extent_slot--; |
952fccac | 6922 | } |
a79865c6 | 6923 | |
31840ae1 | 6924 | if (!found_extent) { |
5d4f98a2 | 6925 | BUG_ON(iref); |
87cc7a8a | 6926 | ret = remove_extent_backref(trans, path, NULL, |
87bde3cd | 6927 | refs_to_drop, |
fcebe456 | 6928 | is_data, &last_ref); |
005d6427 | 6929 | if (ret) { |
66642832 | 6930 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
6931 | goto out; |
6932 | } | |
b3b4aa74 | 6933 | btrfs_release_path(path); |
b9473439 | 6934 | path->leave_spinning = 1; |
5d4f98a2 YZ |
6935 | |
6936 | key.objectid = bytenr; | |
6937 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
6938 | key.offset = num_bytes; | |
6939 | ||
3173a18f JB |
6940 | if (!is_data && skinny_metadata) { |
6941 | key.type = BTRFS_METADATA_ITEM_KEY; | |
6942 | key.offset = owner_objectid; | |
6943 | } | |
6944 | ||
31840ae1 ZY |
6945 | ret = btrfs_search_slot(trans, extent_root, |
6946 | &key, path, -1, 1); | |
3173a18f JB |
6947 | if (ret > 0 && skinny_metadata && path->slots[0]) { |
6948 | /* | |
6949 | * Couldn't find our skinny metadata item, | |
6950 | * see if we have ye olde extent item. | |
6951 | */ | |
6952 | path->slots[0]--; | |
6953 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
6954 | path->slots[0]); | |
6955 | if (key.objectid == bytenr && | |
6956 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
6957 | key.offset == num_bytes) | |
6958 | ret = 0; | |
6959 | } | |
6960 | ||
6961 | if (ret > 0 && skinny_metadata) { | |
6962 | skinny_metadata = false; | |
9ce49a0b | 6963 | key.objectid = bytenr; |
3173a18f JB |
6964 | key.type = BTRFS_EXTENT_ITEM_KEY; |
6965 | key.offset = num_bytes; | |
6966 | btrfs_release_path(path); | |
6967 | ret = btrfs_search_slot(trans, extent_root, | |
6968 | &key, path, -1, 1); | |
6969 | } | |
6970 | ||
f3465ca4 | 6971 | if (ret) { |
5d163e0e JM |
6972 | btrfs_err(info, |
6973 | "umm, got %d back from search, was looking for %llu", | |
6974 | ret, bytenr); | |
b783e62d | 6975 | if (ret > 0) |
a4f78750 | 6976 | btrfs_print_leaf(path->nodes[0]); |
f3465ca4 | 6977 | } |
005d6427 | 6978 | if (ret < 0) { |
66642832 | 6979 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
6980 | goto out; |
6981 | } | |
31840ae1 ZY |
6982 | extent_slot = path->slots[0]; |
6983 | } | |
fae7f21c | 6984 | } else if (WARN_ON(ret == -ENOENT)) { |
a4f78750 | 6985 | btrfs_print_leaf(path->nodes[0]); |
c2cf52eb SK |
6986 | btrfs_err(info, |
6987 | "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu", | |
c1c9ff7c GU |
6988 | bytenr, parent, root_objectid, owner_objectid, |
6989 | owner_offset); | |
66642832 | 6990 | btrfs_abort_transaction(trans, ret); |
c4a050bb | 6991 | goto out; |
79787eaa | 6992 | } else { |
66642832 | 6993 | btrfs_abort_transaction(trans, ret); |
005d6427 | 6994 | goto out; |
7bb86316 | 6995 | } |
5f39d397 CM |
6996 | |
6997 | leaf = path->nodes[0]; | |
5d4f98a2 | 6998 | item_size = btrfs_item_size_nr(leaf, extent_slot); |
6d8ff4e4 | 6999 | if (unlikely(item_size < sizeof(*ei))) { |
ba3c2b19 NB |
7000 | ret = -EINVAL; |
7001 | btrfs_print_v0_err(info); | |
7002 | btrfs_abort_transaction(trans, ret); | |
7003 | goto out; | |
7004 | } | |
952fccac | 7005 | ei = btrfs_item_ptr(leaf, extent_slot, |
123abc88 | 7006 | struct btrfs_extent_item); |
3173a18f JB |
7007 | if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID && |
7008 | key.type == BTRFS_EXTENT_ITEM_KEY) { | |
5d4f98a2 YZ |
7009 | struct btrfs_tree_block_info *bi; |
7010 | BUG_ON(item_size < sizeof(*ei) + sizeof(*bi)); | |
7011 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
7012 | WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi)); | |
7013 | } | |
56bec294 | 7014 | |
5d4f98a2 | 7015 | refs = btrfs_extent_refs(leaf, ei); |
32b02538 | 7016 | if (refs < refs_to_drop) { |
5d163e0e JM |
7017 | btrfs_err(info, |
7018 | "trying to drop %d refs but we only have %Lu for bytenr %Lu", | |
7019 | refs_to_drop, refs, bytenr); | |
32b02538 | 7020 | ret = -EINVAL; |
66642832 | 7021 | btrfs_abort_transaction(trans, ret); |
32b02538 JB |
7022 | goto out; |
7023 | } | |
56bec294 | 7024 | refs -= refs_to_drop; |
5f39d397 | 7025 | |
5d4f98a2 YZ |
7026 | if (refs > 0) { |
7027 | if (extent_op) | |
7028 | __run_delayed_extent_op(extent_op, leaf, ei); | |
7029 | /* | |
7030 | * In the case of inline back ref, reference count will | |
7031 | * be updated by remove_extent_backref | |
952fccac | 7032 | */ |
5d4f98a2 YZ |
7033 | if (iref) { |
7034 | BUG_ON(!found_extent); | |
7035 | } else { | |
7036 | btrfs_set_extent_refs(leaf, ei, refs); | |
7037 | btrfs_mark_buffer_dirty(leaf); | |
7038 | } | |
7039 | if (found_extent) { | |
87cc7a8a NB |
7040 | ret = remove_extent_backref(trans, path, iref, |
7041 | refs_to_drop, is_data, | |
7042 | &last_ref); | |
005d6427 | 7043 | if (ret) { |
66642832 | 7044 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7045 | goto out; |
7046 | } | |
952fccac | 7047 | } |
5d4f98a2 | 7048 | } else { |
5d4f98a2 YZ |
7049 | if (found_extent) { |
7050 | BUG_ON(is_data && refs_to_drop != | |
9ed0dea0 | 7051 | extent_data_ref_count(path, iref)); |
5d4f98a2 YZ |
7052 | if (iref) { |
7053 | BUG_ON(path->slots[0] != extent_slot); | |
7054 | } else { | |
7055 | BUG_ON(path->slots[0] != extent_slot + 1); | |
7056 | path->slots[0] = extent_slot; | |
7057 | num_to_del = 2; | |
7058 | } | |
78fae27e | 7059 | } |
b9473439 | 7060 | |
fcebe456 | 7061 | last_ref = 1; |
952fccac CM |
7062 | ret = btrfs_del_items(trans, extent_root, path, path->slots[0], |
7063 | num_to_del); | |
005d6427 | 7064 | if (ret) { |
66642832 | 7065 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7066 | goto out; |
7067 | } | |
b3b4aa74 | 7068 | btrfs_release_path(path); |
21af804c | 7069 | |
5d4f98a2 | 7070 | if (is_data) { |
5b4aacef | 7071 | ret = btrfs_del_csums(trans, info, bytenr, num_bytes); |
005d6427 | 7072 | if (ret) { |
66642832 | 7073 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7074 | goto out; |
7075 | } | |
459931ec CM |
7076 | } |
7077 | ||
e7355e50 | 7078 | ret = add_to_free_space_tree(trans, bytenr, num_bytes); |
1e144fb8 | 7079 | if (ret) { |
66642832 | 7080 | btrfs_abort_transaction(trans, ret); |
1e144fb8 OS |
7081 | goto out; |
7082 | } | |
7083 | ||
6b279408 | 7084 | ret = update_block_group(trans, bytenr, num_bytes, 0); |
005d6427 | 7085 | if (ret) { |
66642832 | 7086 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7087 | goto out; |
7088 | } | |
a28ec197 | 7089 | } |
fcebe456 JB |
7090 | btrfs_release_path(path); |
7091 | ||
79787eaa | 7092 | out: |
5caf2a00 | 7093 | btrfs_free_path(path); |
a28ec197 CM |
7094 | return ret; |
7095 | } | |
7096 | ||
1887be66 | 7097 | /* |
f0486c68 | 7098 | * when we free an block, it is possible (and likely) that we free the last |
1887be66 CM |
7099 | * delayed ref for that extent as well. This searches the delayed ref tree for |
7100 | * a given extent, and if there are no other delayed refs to be processed, it | |
7101 | * removes it from the tree. | |
7102 | */ | |
7103 | static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans, | |
2ff7e61e | 7104 | u64 bytenr) |
1887be66 CM |
7105 | { |
7106 | struct btrfs_delayed_ref_head *head; | |
7107 | struct btrfs_delayed_ref_root *delayed_refs; | |
f0486c68 | 7108 | int ret = 0; |
1887be66 CM |
7109 | |
7110 | delayed_refs = &trans->transaction->delayed_refs; | |
7111 | spin_lock(&delayed_refs->lock); | |
f72ad18e | 7112 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
1887be66 | 7113 | if (!head) |
cf93da7b | 7114 | goto out_delayed_unlock; |
1887be66 | 7115 | |
d7df2c79 | 7116 | spin_lock(&head->lock); |
e3d03965 | 7117 | if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root)) |
1887be66 CM |
7118 | goto out; |
7119 | ||
bedc6617 JB |
7120 | if (cleanup_extent_op(head) != NULL) |
7121 | goto out; | |
5d4f98a2 | 7122 | |
1887be66 CM |
7123 | /* |
7124 | * waiting for the lock here would deadlock. If someone else has it | |
7125 | * locked they are already in the process of dropping it anyway | |
7126 | */ | |
7127 | if (!mutex_trylock(&head->mutex)) | |
7128 | goto out; | |
7129 | ||
d7baffda | 7130 | btrfs_delete_ref_head(delayed_refs, head); |
d7df2c79 | 7131 | head->processing = 0; |
d7baffda | 7132 | |
d7df2c79 | 7133 | spin_unlock(&head->lock); |
1887be66 CM |
7134 | spin_unlock(&delayed_refs->lock); |
7135 | ||
f0486c68 YZ |
7136 | BUG_ON(head->extent_op); |
7137 | if (head->must_insert_reserved) | |
7138 | ret = 1; | |
7139 | ||
31890da0 | 7140 | btrfs_cleanup_ref_head_accounting(trans->fs_info, delayed_refs, head); |
f0486c68 | 7141 | mutex_unlock(&head->mutex); |
d278850e | 7142 | btrfs_put_delayed_ref_head(head); |
f0486c68 | 7143 | return ret; |
1887be66 | 7144 | out: |
d7df2c79 | 7145 | spin_unlock(&head->lock); |
cf93da7b CM |
7146 | |
7147 | out_delayed_unlock: | |
1887be66 CM |
7148 | spin_unlock(&delayed_refs->lock); |
7149 | return 0; | |
7150 | } | |
7151 | ||
f0486c68 YZ |
7152 | void btrfs_free_tree_block(struct btrfs_trans_handle *trans, |
7153 | struct btrfs_root *root, | |
7154 | struct extent_buffer *buf, | |
5581a51a | 7155 | u64 parent, int last_ref) |
f0486c68 | 7156 | { |
0b246afa | 7157 | struct btrfs_fs_info *fs_info = root->fs_info; |
ed4f255b | 7158 | struct btrfs_ref generic_ref = { 0 }; |
b150a4f1 | 7159 | int pin = 1; |
f0486c68 YZ |
7160 | int ret; |
7161 | ||
ed4f255b QW |
7162 | btrfs_init_generic_ref(&generic_ref, BTRFS_DROP_DELAYED_REF, |
7163 | buf->start, buf->len, parent); | |
7164 | btrfs_init_tree_ref(&generic_ref, btrfs_header_level(buf), | |
7165 | root->root_key.objectid); | |
7166 | ||
f0486c68 | 7167 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { |
d7eae340 OS |
7168 | int old_ref_mod, new_ref_mod; |
7169 | ||
8a5040f7 | 7170 | btrfs_ref_tree_mod(fs_info, &generic_ref); |
ed4f255b | 7171 | ret = btrfs_add_delayed_tree_ref(trans, &generic_ref, NULL, |
d7eae340 | 7172 | &old_ref_mod, &new_ref_mod); |
79787eaa | 7173 | BUG_ON(ret); /* -ENOMEM */ |
d7eae340 | 7174 | pin = old_ref_mod >= 0 && new_ref_mod < 0; |
f0486c68 YZ |
7175 | } |
7176 | ||
0a16c7d7 | 7177 | if (last_ref && btrfs_header_generation(buf) == trans->transid) { |
6219872d FM |
7178 | struct btrfs_block_group_cache *cache; |
7179 | ||
f0486c68 | 7180 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { |
2ff7e61e | 7181 | ret = check_ref_cleanup(trans, buf->start); |
f0486c68 | 7182 | if (!ret) |
37be25bc | 7183 | goto out; |
f0486c68 YZ |
7184 | } |
7185 | ||
4da8b76d | 7186 | pin = 0; |
0b246afa | 7187 | cache = btrfs_lookup_block_group(fs_info, buf->start); |
6219872d | 7188 | |
f0486c68 | 7189 | if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) { |
fdf08605 | 7190 | pin_down_extent(cache, buf->start, buf->len, 1); |
6219872d | 7191 | btrfs_put_block_group(cache); |
37be25bc | 7192 | goto out; |
f0486c68 YZ |
7193 | } |
7194 | ||
7195 | WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)); | |
7196 | ||
7197 | btrfs_add_free_space(cache, buf->start, buf->len); | |
4824f1f4 | 7198 | btrfs_free_reserved_bytes(cache, buf->len, 0); |
6219872d | 7199 | btrfs_put_block_group(cache); |
71ff6437 | 7200 | trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len); |
f0486c68 YZ |
7201 | } |
7202 | out: | |
b150a4f1 | 7203 | if (pin) |
78192442 | 7204 | add_pinned_bytes(fs_info, &generic_ref); |
b150a4f1 | 7205 | |
0a16c7d7 OS |
7206 | if (last_ref) { |
7207 | /* | |
7208 | * Deleting the buffer, clear the corrupt flag since it doesn't | |
7209 | * matter anymore. | |
7210 | */ | |
7211 | clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags); | |
7212 | } | |
f0486c68 YZ |
7213 | } |
7214 | ||
79787eaa | 7215 | /* Can return -ENOMEM */ |
ffd4bb2a | 7216 | int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_ref *ref) |
925baedd | 7217 | { |
ffd4bb2a | 7218 | struct btrfs_fs_info *fs_info = trans->fs_info; |
d7eae340 | 7219 | int old_ref_mod, new_ref_mod; |
925baedd CM |
7220 | int ret; |
7221 | ||
f5ee5c9a | 7222 | if (btrfs_is_testing(fs_info)) |
faa2dbf0 | 7223 | return 0; |
fccb84c9 | 7224 | |
56bec294 CM |
7225 | /* |
7226 | * tree log blocks never actually go into the extent allocation | |
7227 | * tree, just update pinning info and exit early. | |
56bec294 | 7228 | */ |
ffd4bb2a QW |
7229 | if ((ref->type == BTRFS_REF_METADATA && |
7230 | ref->tree_ref.root == BTRFS_TREE_LOG_OBJECTID) || | |
7231 | (ref->type == BTRFS_REF_DATA && | |
7232 | ref->data_ref.ref_root == BTRFS_TREE_LOG_OBJECTID)) { | |
b9473439 | 7233 | /* unlocks the pinned mutex */ |
ffd4bb2a | 7234 | btrfs_pin_extent(fs_info, ref->bytenr, ref->len, 1); |
d7eae340 | 7235 | old_ref_mod = new_ref_mod = 0; |
56bec294 | 7236 | ret = 0; |
ffd4bb2a QW |
7237 | } else if (ref->type == BTRFS_REF_METADATA) { |
7238 | ret = btrfs_add_delayed_tree_ref(trans, ref, NULL, | |
d7eae340 | 7239 | &old_ref_mod, &new_ref_mod); |
5d4f98a2 | 7240 | } else { |
ffd4bb2a | 7241 | ret = btrfs_add_delayed_data_ref(trans, ref, 0, |
d7eae340 | 7242 | &old_ref_mod, &new_ref_mod); |
56bec294 | 7243 | } |
d7eae340 | 7244 | |
ffd4bb2a QW |
7245 | if (!((ref->type == BTRFS_REF_METADATA && |
7246 | ref->tree_ref.root == BTRFS_TREE_LOG_OBJECTID) || | |
7247 | (ref->type == BTRFS_REF_DATA && | |
7248 | ref->data_ref.ref_root == BTRFS_TREE_LOG_OBJECTID))) | |
7249 | btrfs_ref_tree_mod(fs_info, ref); | |
8a5040f7 | 7250 | |
ddf30cf0 | 7251 | if (ret == 0 && old_ref_mod >= 0 && new_ref_mod < 0) |
78192442 | 7252 | add_pinned_bytes(fs_info, ref); |
d7eae340 | 7253 | |
925baedd CM |
7254 | return ret; |
7255 | } | |
7256 | ||
817d52f8 JB |
7257 | /* |
7258 | * when we wait for progress in the block group caching, its because | |
7259 | * our allocation attempt failed at least once. So, we must sleep | |
7260 | * and let some progress happen before we try again. | |
7261 | * | |
7262 | * This function will sleep at least once waiting for new free space to | |
7263 | * show up, and then it will check the block group free space numbers | |
7264 | * for our min num_bytes. Another option is to have it go ahead | |
7265 | * and look in the rbtree for a free extent of a given size, but this | |
7266 | * is a good start. | |
36cce922 JB |
7267 | * |
7268 | * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using | |
7269 | * any of the information in this block group. | |
817d52f8 | 7270 | */ |
36cce922 | 7271 | static noinline void |
817d52f8 JB |
7272 | wait_block_group_cache_progress(struct btrfs_block_group_cache *cache, |
7273 | u64 num_bytes) | |
7274 | { | |
11833d66 | 7275 | struct btrfs_caching_control *caching_ctl; |
817d52f8 | 7276 | |
11833d66 YZ |
7277 | caching_ctl = get_caching_control(cache); |
7278 | if (!caching_ctl) | |
36cce922 | 7279 | return; |
817d52f8 | 7280 | |
11833d66 | 7281 | wait_event(caching_ctl->wait, block_group_cache_done(cache) || |
34d52cb6 | 7282 | (cache->free_space_ctl->free_space >= num_bytes)); |
11833d66 YZ |
7283 | |
7284 | put_caching_control(caching_ctl); | |
11833d66 YZ |
7285 | } |
7286 | ||
7287 | static noinline int | |
7288 | wait_block_group_cache_done(struct btrfs_block_group_cache *cache) | |
7289 | { | |
7290 | struct btrfs_caching_control *caching_ctl; | |
36cce922 | 7291 | int ret = 0; |
11833d66 YZ |
7292 | |
7293 | caching_ctl = get_caching_control(cache); | |
7294 | if (!caching_ctl) | |
36cce922 | 7295 | return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0; |
11833d66 YZ |
7296 | |
7297 | wait_event(caching_ctl->wait, block_group_cache_done(cache)); | |
36cce922 JB |
7298 | if (cache->cached == BTRFS_CACHE_ERROR) |
7299 | ret = -EIO; | |
11833d66 | 7300 | put_caching_control(caching_ctl); |
36cce922 | 7301 | return ret; |
817d52f8 JB |
7302 | } |
7303 | ||
7304 | enum btrfs_loop_type { | |
285ff5af JB |
7305 | LOOP_CACHING_NOWAIT = 0, |
7306 | LOOP_CACHING_WAIT = 1, | |
7307 | LOOP_ALLOC_CHUNK = 2, | |
7308 | LOOP_NO_EMPTY_SIZE = 3, | |
817d52f8 JB |
7309 | }; |
7310 | ||
e570fd27 MX |
7311 | static inline void |
7312 | btrfs_lock_block_group(struct btrfs_block_group_cache *cache, | |
7313 | int delalloc) | |
7314 | { | |
7315 | if (delalloc) | |
7316 | down_read(&cache->data_rwsem); | |
7317 | } | |
7318 | ||
7319 | static inline void | |
7320 | btrfs_grab_block_group(struct btrfs_block_group_cache *cache, | |
7321 | int delalloc) | |
7322 | { | |
7323 | btrfs_get_block_group(cache); | |
7324 | if (delalloc) | |
7325 | down_read(&cache->data_rwsem); | |
7326 | } | |
7327 | ||
7328 | static struct btrfs_block_group_cache * | |
7329 | btrfs_lock_cluster(struct btrfs_block_group_cache *block_group, | |
7330 | struct btrfs_free_cluster *cluster, | |
7331 | int delalloc) | |
7332 | { | |
89771cc9 | 7333 | struct btrfs_block_group_cache *used_bg = NULL; |
6719afdc | 7334 | |
e570fd27 | 7335 | spin_lock(&cluster->refill_lock); |
6719afdc GU |
7336 | while (1) { |
7337 | used_bg = cluster->block_group; | |
7338 | if (!used_bg) | |
7339 | return NULL; | |
7340 | ||
7341 | if (used_bg == block_group) | |
e570fd27 MX |
7342 | return used_bg; |
7343 | ||
6719afdc | 7344 | btrfs_get_block_group(used_bg); |
e570fd27 | 7345 | |
6719afdc GU |
7346 | if (!delalloc) |
7347 | return used_bg; | |
e570fd27 | 7348 | |
6719afdc GU |
7349 | if (down_read_trylock(&used_bg->data_rwsem)) |
7350 | return used_bg; | |
e570fd27 | 7351 | |
6719afdc | 7352 | spin_unlock(&cluster->refill_lock); |
e570fd27 | 7353 | |
e321f8a8 LB |
7354 | /* We should only have one-level nested. */ |
7355 | down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING); | |
e570fd27 | 7356 | |
6719afdc GU |
7357 | spin_lock(&cluster->refill_lock); |
7358 | if (used_bg == cluster->block_group) | |
7359 | return used_bg; | |
e570fd27 | 7360 | |
6719afdc GU |
7361 | up_read(&used_bg->data_rwsem); |
7362 | btrfs_put_block_group(used_bg); | |
7363 | } | |
e570fd27 MX |
7364 | } |
7365 | ||
7366 | static inline void | |
7367 | btrfs_release_block_group(struct btrfs_block_group_cache *cache, | |
7368 | int delalloc) | |
7369 | { | |
7370 | if (delalloc) | |
7371 | up_read(&cache->data_rwsem); | |
7372 | btrfs_put_block_group(cache); | |
7373 | } | |
7374 | ||
b4bd745d QW |
7375 | /* |
7376 | * Structure used internally for find_free_extent() function. Wraps needed | |
7377 | * parameters. | |
7378 | */ | |
7379 | struct find_free_extent_ctl { | |
7380 | /* Basic allocation info */ | |
7381 | u64 ram_bytes; | |
7382 | u64 num_bytes; | |
7383 | u64 empty_size; | |
7384 | u64 flags; | |
7385 | int delalloc; | |
7386 | ||
7387 | /* Where to start the search inside the bg */ | |
7388 | u64 search_start; | |
7389 | ||
7390 | /* For clustered allocation */ | |
7391 | u64 empty_cluster; | |
7392 | ||
7393 | bool have_caching_bg; | |
7394 | bool orig_have_caching_bg; | |
7395 | ||
7396 | /* RAID index, converted from flags */ | |
7397 | int index; | |
7398 | ||
e72d79d6 QW |
7399 | /* |
7400 | * Current loop number, check find_free_extent_update_loop() for details | |
7401 | */ | |
b4bd745d QW |
7402 | int loop; |
7403 | ||
7404 | /* | |
7405 | * Whether we're refilling a cluster, if true we need to re-search | |
7406 | * current block group but don't try to refill the cluster again. | |
7407 | */ | |
7408 | bool retry_clustered; | |
7409 | ||
7410 | /* | |
7411 | * Whether we're updating free space cache, if true we need to re-search | |
7412 | * current block group but don't try updating free space cache again. | |
7413 | */ | |
7414 | bool retry_unclustered; | |
7415 | ||
7416 | /* If current block group is cached */ | |
7417 | int cached; | |
7418 | ||
7419 | /* Max contiguous hole found */ | |
7420 | u64 max_extent_size; | |
7421 | ||
7422 | /* Total free space from free space cache, not always contiguous */ | |
7423 | u64 total_free_space; | |
7424 | ||
7425 | /* Found result */ | |
7426 | u64 found_offset; | |
7427 | }; | |
7428 | ||
d06e3bb6 QW |
7429 | |
7430 | /* | |
7431 | * Helper function for find_free_extent(). | |
7432 | * | |
7433 | * Return -ENOENT to inform caller that we need fallback to unclustered mode. | |
7434 | * Return -EAGAIN to inform caller that we need to re-search this block group | |
7435 | * Return >0 to inform caller that we find nothing | |
7436 | * Return 0 means we have found a location and set ffe_ctl->found_offset. | |
7437 | */ | |
7438 | static int find_free_extent_clustered(struct btrfs_block_group_cache *bg, | |
7439 | struct btrfs_free_cluster *last_ptr, | |
7440 | struct find_free_extent_ctl *ffe_ctl, | |
7441 | struct btrfs_block_group_cache **cluster_bg_ret) | |
7442 | { | |
d06e3bb6 QW |
7443 | struct btrfs_block_group_cache *cluster_bg; |
7444 | u64 aligned_cluster; | |
7445 | u64 offset; | |
7446 | int ret; | |
7447 | ||
7448 | cluster_bg = btrfs_lock_cluster(bg, last_ptr, ffe_ctl->delalloc); | |
7449 | if (!cluster_bg) | |
7450 | goto refill_cluster; | |
7451 | if (cluster_bg != bg && (cluster_bg->ro || | |
7452 | !block_group_bits(cluster_bg, ffe_ctl->flags))) | |
7453 | goto release_cluster; | |
7454 | ||
7455 | offset = btrfs_alloc_from_cluster(cluster_bg, last_ptr, | |
7456 | ffe_ctl->num_bytes, cluster_bg->key.objectid, | |
7457 | &ffe_ctl->max_extent_size); | |
7458 | if (offset) { | |
7459 | /* We have a block, we're done */ | |
7460 | spin_unlock(&last_ptr->refill_lock); | |
7461 | trace_btrfs_reserve_extent_cluster(cluster_bg, | |
7462 | ffe_ctl->search_start, ffe_ctl->num_bytes); | |
7463 | *cluster_bg_ret = cluster_bg; | |
7464 | ffe_ctl->found_offset = offset; | |
7465 | return 0; | |
7466 | } | |
7467 | WARN_ON(last_ptr->block_group != cluster_bg); | |
7468 | ||
7469 | release_cluster: | |
7470 | /* | |
7471 | * If we are on LOOP_NO_EMPTY_SIZE, we can't set up a new clusters, so | |
7472 | * lets just skip it and let the allocator find whatever block it can | |
7473 | * find. If we reach this point, we will have tried the cluster | |
7474 | * allocator plenty of times and not have found anything, so we are | |
7475 | * likely way too fragmented for the clustering stuff to find anything. | |
7476 | * | |
7477 | * However, if the cluster is taken from the current block group, | |
7478 | * release the cluster first, so that we stand a better chance of | |
7479 | * succeeding in the unclustered allocation. | |
7480 | */ | |
7481 | if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE && cluster_bg != bg) { | |
7482 | spin_unlock(&last_ptr->refill_lock); | |
7483 | btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc); | |
7484 | return -ENOENT; | |
7485 | } | |
7486 | ||
7487 | /* This cluster didn't work out, free it and start over */ | |
7488 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
7489 | ||
7490 | if (cluster_bg != bg) | |
7491 | btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc); | |
7492 | ||
7493 | refill_cluster: | |
7494 | if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE) { | |
7495 | spin_unlock(&last_ptr->refill_lock); | |
7496 | return -ENOENT; | |
7497 | } | |
7498 | ||
7499 | aligned_cluster = max_t(u64, | |
7500 | ffe_ctl->empty_cluster + ffe_ctl->empty_size, | |
7501 | bg->full_stripe_len); | |
2ceeae2e DS |
7502 | ret = btrfs_find_space_cluster(bg, last_ptr, ffe_ctl->search_start, |
7503 | ffe_ctl->num_bytes, aligned_cluster); | |
d06e3bb6 QW |
7504 | if (ret == 0) { |
7505 | /* Now pull our allocation out of this cluster */ | |
7506 | offset = btrfs_alloc_from_cluster(bg, last_ptr, | |
7507 | ffe_ctl->num_bytes, ffe_ctl->search_start, | |
7508 | &ffe_ctl->max_extent_size); | |
7509 | if (offset) { | |
7510 | /* We found one, proceed */ | |
7511 | spin_unlock(&last_ptr->refill_lock); | |
7512 | trace_btrfs_reserve_extent_cluster(bg, | |
7513 | ffe_ctl->search_start, | |
7514 | ffe_ctl->num_bytes); | |
7515 | ffe_ctl->found_offset = offset; | |
7516 | return 0; | |
7517 | } | |
7518 | } else if (!ffe_ctl->cached && ffe_ctl->loop > LOOP_CACHING_NOWAIT && | |
7519 | !ffe_ctl->retry_clustered) { | |
7520 | spin_unlock(&last_ptr->refill_lock); | |
7521 | ||
7522 | ffe_ctl->retry_clustered = true; | |
7523 | wait_block_group_cache_progress(bg, ffe_ctl->num_bytes + | |
7524 | ffe_ctl->empty_cluster + ffe_ctl->empty_size); | |
7525 | return -EAGAIN; | |
7526 | } | |
7527 | /* | |
7528 | * At this point we either didn't find a cluster or we weren't able to | |
7529 | * allocate a block from our cluster. Free the cluster we've been | |
7530 | * trying to use, and go to the next block group. | |
7531 | */ | |
7532 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
7533 | spin_unlock(&last_ptr->refill_lock); | |
7534 | return 1; | |
7535 | } | |
7536 | ||
e1a41848 QW |
7537 | /* |
7538 | * Return >0 to inform caller that we find nothing | |
7539 | * Return 0 when we found an free extent and set ffe_ctrl->found_offset | |
7540 | * Return -EAGAIN to inform caller that we need to re-search this block group | |
7541 | */ | |
7542 | static int find_free_extent_unclustered(struct btrfs_block_group_cache *bg, | |
7543 | struct btrfs_free_cluster *last_ptr, | |
7544 | struct find_free_extent_ctl *ffe_ctl) | |
7545 | { | |
7546 | u64 offset; | |
7547 | ||
7548 | /* | |
7549 | * We are doing an unclustered allocation, set the fragmented flag so | |
7550 | * we don't bother trying to setup a cluster again until we get more | |
7551 | * space. | |
7552 | */ | |
7553 | if (unlikely(last_ptr)) { | |
7554 | spin_lock(&last_ptr->lock); | |
7555 | last_ptr->fragmented = 1; | |
7556 | spin_unlock(&last_ptr->lock); | |
7557 | } | |
7558 | if (ffe_ctl->cached) { | |
7559 | struct btrfs_free_space_ctl *free_space_ctl; | |
7560 | ||
7561 | free_space_ctl = bg->free_space_ctl; | |
7562 | spin_lock(&free_space_ctl->tree_lock); | |
7563 | if (free_space_ctl->free_space < | |
7564 | ffe_ctl->num_bytes + ffe_ctl->empty_cluster + | |
7565 | ffe_ctl->empty_size) { | |
7566 | ffe_ctl->total_free_space = max_t(u64, | |
7567 | ffe_ctl->total_free_space, | |
7568 | free_space_ctl->free_space); | |
7569 | spin_unlock(&free_space_ctl->tree_lock); | |
7570 | return 1; | |
7571 | } | |
7572 | spin_unlock(&free_space_ctl->tree_lock); | |
7573 | } | |
7574 | ||
7575 | offset = btrfs_find_space_for_alloc(bg, ffe_ctl->search_start, | |
7576 | ffe_ctl->num_bytes, ffe_ctl->empty_size, | |
7577 | &ffe_ctl->max_extent_size); | |
7578 | ||
7579 | /* | |
7580 | * If we didn't find a chunk, and we haven't failed on this block group | |
7581 | * before, and this block group is in the middle of caching and we are | |
7582 | * ok with waiting, then go ahead and wait for progress to be made, and | |
7583 | * set @retry_unclustered to true. | |
7584 | * | |
7585 | * If @retry_unclustered is true then we've already waited on this | |
7586 | * block group once and should move on to the next block group. | |
7587 | */ | |
7588 | if (!offset && !ffe_ctl->retry_unclustered && !ffe_ctl->cached && | |
7589 | ffe_ctl->loop > LOOP_CACHING_NOWAIT) { | |
7590 | wait_block_group_cache_progress(bg, ffe_ctl->num_bytes + | |
7591 | ffe_ctl->empty_size); | |
7592 | ffe_ctl->retry_unclustered = true; | |
7593 | return -EAGAIN; | |
7594 | } else if (!offset) { | |
7595 | return 1; | |
7596 | } | |
7597 | ffe_ctl->found_offset = offset; | |
7598 | return 0; | |
7599 | } | |
7600 | ||
e72d79d6 QW |
7601 | /* |
7602 | * Return >0 means caller needs to re-search for free extent | |
7603 | * Return 0 means we have the needed free extent. | |
7604 | * Return <0 means we failed to locate any free extent. | |
7605 | */ | |
7606 | static int find_free_extent_update_loop(struct btrfs_fs_info *fs_info, | |
7607 | struct btrfs_free_cluster *last_ptr, | |
7608 | struct btrfs_key *ins, | |
7609 | struct find_free_extent_ctl *ffe_ctl, | |
7610 | int full_search, bool use_cluster) | |
7611 | { | |
7612 | struct btrfs_root *root = fs_info->extent_root; | |
7613 | int ret; | |
7614 | ||
7615 | if ((ffe_ctl->loop == LOOP_CACHING_NOWAIT) && | |
7616 | ffe_ctl->have_caching_bg && !ffe_ctl->orig_have_caching_bg) | |
7617 | ffe_ctl->orig_have_caching_bg = true; | |
7618 | ||
7619 | if (!ins->objectid && ffe_ctl->loop >= LOOP_CACHING_WAIT && | |
7620 | ffe_ctl->have_caching_bg) | |
7621 | return 1; | |
7622 | ||
7623 | if (!ins->objectid && ++(ffe_ctl->index) < BTRFS_NR_RAID_TYPES) | |
7624 | return 1; | |
7625 | ||
7626 | if (ins->objectid) { | |
7627 | if (!use_cluster && last_ptr) { | |
7628 | spin_lock(&last_ptr->lock); | |
7629 | last_ptr->window_start = ins->objectid; | |
7630 | spin_unlock(&last_ptr->lock); | |
7631 | } | |
7632 | return 0; | |
7633 | } | |
7634 | ||
7635 | /* | |
7636 | * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking | |
7637 | * caching kthreads as we move along | |
7638 | * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching | |
7639 | * LOOP_ALLOC_CHUNK, force a chunk allocation and try again | |
7640 | * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try | |
7641 | * again | |
7642 | */ | |
7643 | if (ffe_ctl->loop < LOOP_NO_EMPTY_SIZE) { | |
7644 | ffe_ctl->index = 0; | |
7645 | if (ffe_ctl->loop == LOOP_CACHING_NOWAIT) { | |
7646 | /* | |
7647 | * We want to skip the LOOP_CACHING_WAIT step if we | |
7648 | * don't have any uncached bgs and we've already done a | |
7649 | * full search through. | |
7650 | */ | |
7651 | if (ffe_ctl->orig_have_caching_bg || !full_search) | |
7652 | ffe_ctl->loop = LOOP_CACHING_WAIT; | |
7653 | else | |
7654 | ffe_ctl->loop = LOOP_ALLOC_CHUNK; | |
7655 | } else { | |
7656 | ffe_ctl->loop++; | |
7657 | } | |
7658 | ||
7659 | if (ffe_ctl->loop == LOOP_ALLOC_CHUNK) { | |
7660 | struct btrfs_trans_handle *trans; | |
7661 | int exist = 0; | |
7662 | ||
7663 | trans = current->journal_info; | |
7664 | if (trans) | |
7665 | exist = 1; | |
7666 | else | |
7667 | trans = btrfs_join_transaction(root); | |
7668 | ||
7669 | if (IS_ERR(trans)) { | |
7670 | ret = PTR_ERR(trans); | |
7671 | return ret; | |
7672 | } | |
7673 | ||
7674 | ret = do_chunk_alloc(trans, ffe_ctl->flags, | |
7675 | CHUNK_ALLOC_FORCE); | |
7676 | ||
7677 | /* | |
7678 | * If we can't allocate a new chunk we've already looped | |
7679 | * through at least once, move on to the NO_EMPTY_SIZE | |
7680 | * case. | |
7681 | */ | |
7682 | if (ret == -ENOSPC) | |
7683 | ffe_ctl->loop = LOOP_NO_EMPTY_SIZE; | |
7684 | ||
7685 | /* Do not bail out on ENOSPC since we can do more. */ | |
7686 | if (ret < 0 && ret != -ENOSPC) | |
7687 | btrfs_abort_transaction(trans, ret); | |
7688 | else | |
7689 | ret = 0; | |
7690 | if (!exist) | |
7691 | btrfs_end_transaction(trans); | |
7692 | if (ret) | |
7693 | return ret; | |
7694 | } | |
7695 | ||
7696 | if (ffe_ctl->loop == LOOP_NO_EMPTY_SIZE) { | |
7697 | /* | |
7698 | * Don't loop again if we already have no empty_size and | |
7699 | * no empty_cluster. | |
7700 | */ | |
7701 | if (ffe_ctl->empty_size == 0 && | |
7702 | ffe_ctl->empty_cluster == 0) | |
7703 | return -ENOSPC; | |
7704 | ffe_ctl->empty_size = 0; | |
7705 | ffe_ctl->empty_cluster = 0; | |
7706 | } | |
7707 | return 1; | |
7708 | } | |
7709 | return -ENOSPC; | |
7710 | } | |
7711 | ||
fec577fb CM |
7712 | /* |
7713 | * walks the btree of allocated extents and find a hole of a given size. | |
7714 | * The key ins is changed to record the hole: | |
a4820398 | 7715 | * ins->objectid == start position |
62e2749e | 7716 | * ins->flags = BTRFS_EXTENT_ITEM_KEY |
a4820398 | 7717 | * ins->offset == the size of the hole. |
fec577fb | 7718 | * Any available blocks before search_start are skipped. |
a4820398 MX |
7719 | * |
7720 | * If there is no suitable free space, we will record the max size of | |
7721 | * the free space extent currently. | |
e72d79d6 QW |
7722 | * |
7723 | * The overall logic and call chain: | |
7724 | * | |
7725 | * find_free_extent() | |
7726 | * |- Iterate through all block groups | |
7727 | * | |- Get a valid block group | |
7728 | * | |- Try to do clustered allocation in that block group | |
7729 | * | |- Try to do unclustered allocation in that block group | |
7730 | * | |- Check if the result is valid | |
7731 | * | | |- If valid, then exit | |
7732 | * | |- Jump to next block group | |
7733 | * | | |
7734 | * |- Push harder to find free extents | |
7735 | * |- If not found, re-iterate all block groups | |
fec577fb | 7736 | */ |
87bde3cd | 7737 | static noinline int find_free_extent(struct btrfs_fs_info *fs_info, |
18513091 WX |
7738 | u64 ram_bytes, u64 num_bytes, u64 empty_size, |
7739 | u64 hint_byte, struct btrfs_key *ins, | |
7740 | u64 flags, int delalloc) | |
fec577fb | 7741 | { |
80eb234a | 7742 | int ret = 0; |
fa9c0d79 | 7743 | struct btrfs_free_cluster *last_ptr = NULL; |
80eb234a | 7744 | struct btrfs_block_group_cache *block_group = NULL; |
b4bd745d | 7745 | struct find_free_extent_ctl ffe_ctl = {0}; |
80eb234a | 7746 | struct btrfs_space_info *space_info; |
67377734 | 7747 | bool use_cluster = true; |
a5e681d9 | 7748 | bool full_search = false; |
fec577fb | 7749 | |
0b246afa | 7750 | WARN_ON(num_bytes < fs_info->sectorsize); |
b4bd745d QW |
7751 | |
7752 | ffe_ctl.ram_bytes = ram_bytes; | |
7753 | ffe_ctl.num_bytes = num_bytes; | |
7754 | ffe_ctl.empty_size = empty_size; | |
7755 | ffe_ctl.flags = flags; | |
7756 | ffe_ctl.search_start = 0; | |
7757 | ffe_ctl.retry_clustered = false; | |
7758 | ffe_ctl.retry_unclustered = false; | |
7759 | ffe_ctl.delalloc = delalloc; | |
7760 | ffe_ctl.index = btrfs_bg_flags_to_raid_index(flags); | |
7761 | ffe_ctl.have_caching_bg = false; | |
7762 | ffe_ctl.orig_have_caching_bg = false; | |
7763 | ffe_ctl.found_offset = 0; | |
7764 | ||
962a298f | 7765 | ins->type = BTRFS_EXTENT_ITEM_KEY; |
80eb234a JB |
7766 | ins->objectid = 0; |
7767 | ins->offset = 0; | |
b1a4d965 | 7768 | |
71ff6437 | 7769 | trace_find_free_extent(fs_info, num_bytes, empty_size, flags); |
3f7de037 | 7770 | |
0b246afa | 7771 | space_info = __find_space_info(fs_info, flags); |
1b1d1f66 | 7772 | if (!space_info) { |
0b246afa | 7773 | btrfs_err(fs_info, "No space info for %llu", flags); |
1b1d1f66 JB |
7774 | return -ENOSPC; |
7775 | } | |
2552d17e | 7776 | |
67377734 | 7777 | /* |
4f4db217 JB |
7778 | * If our free space is heavily fragmented we may not be able to make |
7779 | * big contiguous allocations, so instead of doing the expensive search | |
7780 | * for free space, simply return ENOSPC with our max_extent_size so we | |
7781 | * can go ahead and search for a more manageable chunk. | |
7782 | * | |
7783 | * If our max_extent_size is large enough for our allocation simply | |
7784 | * disable clustering since we will likely not be able to find enough | |
7785 | * space to create a cluster and induce latency trying. | |
67377734 | 7786 | */ |
4f4db217 JB |
7787 | if (unlikely(space_info->max_extent_size)) { |
7788 | spin_lock(&space_info->lock); | |
7789 | if (space_info->max_extent_size && | |
7790 | num_bytes > space_info->max_extent_size) { | |
7791 | ins->offset = space_info->max_extent_size; | |
7792 | spin_unlock(&space_info->lock); | |
7793 | return -ENOSPC; | |
7794 | } else if (space_info->max_extent_size) { | |
7795 | use_cluster = false; | |
7796 | } | |
7797 | spin_unlock(&space_info->lock); | |
fa9c0d79 | 7798 | } |
0f9dd46c | 7799 | |
b4bd745d QW |
7800 | last_ptr = fetch_cluster_info(fs_info, space_info, |
7801 | &ffe_ctl.empty_cluster); | |
239b14b3 | 7802 | if (last_ptr) { |
fa9c0d79 CM |
7803 | spin_lock(&last_ptr->lock); |
7804 | if (last_ptr->block_group) | |
7805 | hint_byte = last_ptr->window_start; | |
c759c4e1 JB |
7806 | if (last_ptr->fragmented) { |
7807 | /* | |
7808 | * We still set window_start so we can keep track of the | |
7809 | * last place we found an allocation to try and save | |
7810 | * some time. | |
7811 | */ | |
7812 | hint_byte = last_ptr->window_start; | |
7813 | use_cluster = false; | |
7814 | } | |
fa9c0d79 | 7815 | spin_unlock(&last_ptr->lock); |
239b14b3 | 7816 | } |
fa9c0d79 | 7817 | |
b4bd745d QW |
7818 | ffe_ctl.search_start = max(ffe_ctl.search_start, |
7819 | first_logical_byte(fs_info, 0)); | |
7820 | ffe_ctl.search_start = max(ffe_ctl.search_start, hint_byte); | |
7821 | if (ffe_ctl.search_start == hint_byte) { | |
7822 | block_group = btrfs_lookup_block_group(fs_info, | |
7823 | ffe_ctl.search_start); | |
817d52f8 JB |
7824 | /* |
7825 | * we don't want to use the block group if it doesn't match our | |
7826 | * allocation bits, or if its not cached. | |
ccf0e725 JB |
7827 | * |
7828 | * However if we are re-searching with an ideal block group | |
7829 | * picked out then we don't care that the block group is cached. | |
817d52f8 | 7830 | */ |
b6919a58 | 7831 | if (block_group && block_group_bits(block_group, flags) && |
285ff5af | 7832 | block_group->cached != BTRFS_CACHE_NO) { |
2552d17e | 7833 | down_read(&space_info->groups_sem); |
44fb5511 CM |
7834 | if (list_empty(&block_group->list) || |
7835 | block_group->ro) { | |
7836 | /* | |
7837 | * someone is removing this block group, | |
7838 | * we can't jump into the have_block_group | |
7839 | * target because our list pointers are not | |
7840 | * valid | |
7841 | */ | |
7842 | btrfs_put_block_group(block_group); | |
7843 | up_read(&space_info->groups_sem); | |
ccf0e725 | 7844 | } else { |
b4bd745d | 7845 | ffe_ctl.index = btrfs_bg_flags_to_raid_index( |
3e72ee88 | 7846 | block_group->flags); |
e570fd27 | 7847 | btrfs_lock_block_group(block_group, delalloc); |
44fb5511 | 7848 | goto have_block_group; |
ccf0e725 | 7849 | } |
2552d17e | 7850 | } else if (block_group) { |
fa9c0d79 | 7851 | btrfs_put_block_group(block_group); |
2552d17e | 7852 | } |
42e70e7a | 7853 | } |
2552d17e | 7854 | search: |
b4bd745d QW |
7855 | ffe_ctl.have_caching_bg = false; |
7856 | if (ffe_ctl.index == btrfs_bg_flags_to_raid_index(flags) || | |
7857 | ffe_ctl.index == 0) | |
a5e681d9 | 7858 | full_search = true; |
80eb234a | 7859 | down_read(&space_info->groups_sem); |
b4bd745d QW |
7860 | list_for_each_entry(block_group, |
7861 | &space_info->block_groups[ffe_ctl.index], list) { | |
14443937 JM |
7862 | /* If the block group is read-only, we can skip it entirely. */ |
7863 | if (unlikely(block_group->ro)) | |
7864 | continue; | |
7865 | ||
e570fd27 | 7866 | btrfs_grab_block_group(block_group, delalloc); |
b4bd745d | 7867 | ffe_ctl.search_start = block_group->key.objectid; |
42e70e7a | 7868 | |
83a50de9 CM |
7869 | /* |
7870 | * this can happen if we end up cycling through all the | |
7871 | * raid types, but we want to make sure we only allocate | |
7872 | * for the proper type. | |
7873 | */ | |
b6919a58 | 7874 | if (!block_group_bits(block_group, flags)) { |
bece2e82 | 7875 | u64 extra = BTRFS_BLOCK_GROUP_DUP | |
83a50de9 | 7876 | BTRFS_BLOCK_GROUP_RAID1 | |
53b381b3 DW |
7877 | BTRFS_BLOCK_GROUP_RAID5 | |
7878 | BTRFS_BLOCK_GROUP_RAID6 | | |
83a50de9 CM |
7879 | BTRFS_BLOCK_GROUP_RAID10; |
7880 | ||
7881 | /* | |
7882 | * if they asked for extra copies and this block group | |
7883 | * doesn't provide them, bail. This does allow us to | |
7884 | * fill raid0 from raid1. | |
7885 | */ | |
b6919a58 | 7886 | if ((flags & extra) && !(block_group->flags & extra)) |
83a50de9 CM |
7887 | goto loop; |
7888 | } | |
7889 | ||
2552d17e | 7890 | have_block_group: |
b4bd745d QW |
7891 | ffe_ctl.cached = block_group_cache_done(block_group); |
7892 | if (unlikely(!ffe_ctl.cached)) { | |
7893 | ffe_ctl.have_caching_bg = true; | |
f6373bf3 | 7894 | ret = cache_block_group(block_group, 0); |
1d4284bd CM |
7895 | BUG_ON(ret < 0); |
7896 | ret = 0; | |
817d52f8 JB |
7897 | } |
7898 | ||
36cce922 JB |
7899 | if (unlikely(block_group->cached == BTRFS_CACHE_ERROR)) |
7900 | goto loop; | |
0f9dd46c | 7901 | |
0a24325e | 7902 | /* |
062c05c4 AO |
7903 | * Ok we want to try and use the cluster allocator, so |
7904 | * lets look there | |
0a24325e | 7905 | */ |
c759c4e1 | 7906 | if (last_ptr && use_cluster) { |
d06e3bb6 | 7907 | struct btrfs_block_group_cache *cluster_bg = NULL; |
fa9c0d79 | 7908 | |
d06e3bb6 QW |
7909 | ret = find_free_extent_clustered(block_group, last_ptr, |
7910 | &ffe_ctl, &cluster_bg); | |
062c05c4 | 7911 | |
fa9c0d79 | 7912 | if (ret == 0) { |
d06e3bb6 QW |
7913 | if (cluster_bg && cluster_bg != block_group) { |
7914 | btrfs_release_block_group(block_group, | |
7915 | delalloc); | |
7916 | block_group = cluster_bg; | |
fa9c0d79 | 7917 | } |
d06e3bb6 QW |
7918 | goto checks; |
7919 | } else if (ret == -EAGAIN) { | |
817d52f8 | 7920 | goto have_block_group; |
d06e3bb6 QW |
7921 | } else if (ret > 0) { |
7922 | goto loop; | |
fa9c0d79 | 7923 | } |
d06e3bb6 | 7924 | /* ret == -ENOENT case falls through */ |
fa9c0d79 CM |
7925 | } |
7926 | ||
e1a41848 QW |
7927 | ret = find_free_extent_unclustered(block_group, last_ptr, |
7928 | &ffe_ctl); | |
7929 | if (ret == -EAGAIN) | |
817d52f8 | 7930 | goto have_block_group; |
e1a41848 | 7931 | else if (ret > 0) |
1cdda9b8 | 7932 | goto loop; |
e1a41848 | 7933 | /* ret == 0 case falls through */ |
fa9c0d79 | 7934 | checks: |
b4bd745d QW |
7935 | ffe_ctl.search_start = round_up(ffe_ctl.found_offset, |
7936 | fs_info->stripesize); | |
25179201 | 7937 | |
2552d17e | 7938 | /* move on to the next group */ |
b4bd745d | 7939 | if (ffe_ctl.search_start + num_bytes > |
215a63d1 | 7940 | block_group->key.objectid + block_group->key.offset) { |
b4bd745d QW |
7941 | btrfs_add_free_space(block_group, ffe_ctl.found_offset, |
7942 | num_bytes); | |
2552d17e | 7943 | goto loop; |
6226cb0a | 7944 | } |
f5a31e16 | 7945 | |
b4bd745d QW |
7946 | if (ffe_ctl.found_offset < ffe_ctl.search_start) |
7947 | btrfs_add_free_space(block_group, ffe_ctl.found_offset, | |
7948 | ffe_ctl.search_start - ffe_ctl.found_offset); | |
2552d17e | 7949 | |
18513091 WX |
7950 | ret = btrfs_add_reserved_bytes(block_group, ram_bytes, |
7951 | num_bytes, delalloc); | |
f0486c68 | 7952 | if (ret == -EAGAIN) { |
b4bd745d QW |
7953 | btrfs_add_free_space(block_group, ffe_ctl.found_offset, |
7954 | num_bytes); | |
2552d17e | 7955 | goto loop; |
0f9dd46c | 7956 | } |
9cfa3e34 | 7957 | btrfs_inc_block_group_reservations(block_group); |
0b86a832 | 7958 | |
f0486c68 | 7959 | /* we are all good, lets return */ |
b4bd745d | 7960 | ins->objectid = ffe_ctl.search_start; |
2552d17e | 7961 | ins->offset = num_bytes; |
d2fb3437 | 7962 | |
b4bd745d QW |
7963 | trace_btrfs_reserve_extent(block_group, ffe_ctl.search_start, |
7964 | num_bytes); | |
e570fd27 | 7965 | btrfs_release_block_group(block_group, delalloc); |
2552d17e JB |
7966 | break; |
7967 | loop: | |
b4bd745d QW |
7968 | ffe_ctl.retry_clustered = false; |
7969 | ffe_ctl.retry_unclustered = false; | |
3e72ee88 | 7970 | BUG_ON(btrfs_bg_flags_to_raid_index(block_group->flags) != |
b4bd745d | 7971 | ffe_ctl.index); |
e570fd27 | 7972 | btrfs_release_block_group(block_group, delalloc); |
14443937 | 7973 | cond_resched(); |
2552d17e JB |
7974 | } |
7975 | up_read(&space_info->groups_sem); | |
7976 | ||
e72d79d6 QW |
7977 | ret = find_free_extent_update_loop(fs_info, last_ptr, ins, &ffe_ctl, |
7978 | full_search, use_cluster); | |
7979 | if (ret > 0) | |
b742bb82 YZ |
7980 | goto search; |
7981 | ||
4f4db217 | 7982 | if (ret == -ENOSPC) { |
b4bd745d QW |
7983 | /* |
7984 | * Use ffe_ctl->total_free_space as fallback if we can't find | |
7985 | * any contiguous hole. | |
7986 | */ | |
7987 | if (!ffe_ctl.max_extent_size) | |
7988 | ffe_ctl.max_extent_size = ffe_ctl.total_free_space; | |
4f4db217 | 7989 | spin_lock(&space_info->lock); |
b4bd745d | 7990 | space_info->max_extent_size = ffe_ctl.max_extent_size; |
4f4db217 | 7991 | spin_unlock(&space_info->lock); |
b4bd745d | 7992 | ins->offset = ffe_ctl.max_extent_size; |
4f4db217 | 7993 | } |
0f70abe2 | 7994 | return ret; |
fec577fb | 7995 | } |
ec44a35c | 7996 | |
b78e5616 JB |
7997 | #define DUMP_BLOCK_RSV(fs_info, rsv_name) \ |
7998 | do { \ | |
7999 | struct btrfs_block_rsv *__rsv = &(fs_info)->rsv_name; \ | |
8000 | spin_lock(&__rsv->lock); \ | |
8001 | btrfs_info(fs_info, #rsv_name ": size %llu reserved %llu", \ | |
8002 | __rsv->size, __rsv->reserved); \ | |
8003 | spin_unlock(&__rsv->lock); \ | |
8004 | } while (0) | |
8005 | ||
ab8d0fc4 JM |
8006 | static void dump_space_info(struct btrfs_fs_info *fs_info, |
8007 | struct btrfs_space_info *info, u64 bytes, | |
9ed74f2d | 8008 | int dump_block_groups) |
0f9dd46c JB |
8009 | { |
8010 | struct btrfs_block_group_cache *cache; | |
b742bb82 | 8011 | int index = 0; |
0f9dd46c | 8012 | |
9ed74f2d | 8013 | spin_lock(&info->lock); |
ab8d0fc4 JM |
8014 | btrfs_info(fs_info, "space_info %llu has %llu free, is %sfull", |
8015 | info->flags, | |
4136135b LB |
8016 | info->total_bytes - btrfs_space_info_used(info, true), |
8017 | info->full ? "" : "not "); | |
ab8d0fc4 JM |
8018 | btrfs_info(fs_info, |
8019 | "space_info total=%llu, used=%llu, pinned=%llu, reserved=%llu, may_use=%llu, readonly=%llu", | |
8020 | info->total_bytes, info->bytes_used, info->bytes_pinned, | |
8021 | info->bytes_reserved, info->bytes_may_use, | |
8022 | info->bytes_readonly); | |
9ed74f2d JB |
8023 | spin_unlock(&info->lock); |
8024 | ||
b78e5616 JB |
8025 | DUMP_BLOCK_RSV(fs_info, global_block_rsv); |
8026 | DUMP_BLOCK_RSV(fs_info, trans_block_rsv); | |
8027 | DUMP_BLOCK_RSV(fs_info, chunk_block_rsv); | |
8028 | DUMP_BLOCK_RSV(fs_info, delayed_block_rsv); | |
8029 | DUMP_BLOCK_RSV(fs_info, delayed_refs_rsv); | |
8030 | ||
9ed74f2d JB |
8031 | if (!dump_block_groups) |
8032 | return; | |
0f9dd46c | 8033 | |
80eb234a | 8034 | down_read(&info->groups_sem); |
b742bb82 YZ |
8035 | again: |
8036 | list_for_each_entry(cache, &info->block_groups[index], list) { | |
0f9dd46c | 8037 | spin_lock(&cache->lock); |
ab8d0fc4 JM |
8038 | btrfs_info(fs_info, |
8039 | "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s", | |
8040 | cache->key.objectid, cache->key.offset, | |
8041 | btrfs_block_group_used(&cache->item), cache->pinned, | |
8042 | cache->reserved, cache->ro ? "[readonly]" : ""); | |
0f9dd46c JB |
8043 | btrfs_dump_free_space(cache, bytes); |
8044 | spin_unlock(&cache->lock); | |
8045 | } | |
b742bb82 YZ |
8046 | if (++index < BTRFS_NR_RAID_TYPES) |
8047 | goto again; | |
80eb234a | 8048 | up_read(&info->groups_sem); |
0f9dd46c | 8049 | } |
e8569813 | 8050 | |
6f47c706 NB |
8051 | /* |
8052 | * btrfs_reserve_extent - entry point to the extent allocator. Tries to find a | |
8053 | * hole that is at least as big as @num_bytes. | |
8054 | * | |
8055 | * @root - The root that will contain this extent | |
8056 | * | |
8057 | * @ram_bytes - The amount of space in ram that @num_bytes take. This | |
8058 | * is used for accounting purposes. This value differs | |
8059 | * from @num_bytes only in the case of compressed extents. | |
8060 | * | |
8061 | * @num_bytes - Number of bytes to allocate on-disk. | |
8062 | * | |
8063 | * @min_alloc_size - Indicates the minimum amount of space that the | |
8064 | * allocator should try to satisfy. In some cases | |
8065 | * @num_bytes may be larger than what is required and if | |
8066 | * the filesystem is fragmented then allocation fails. | |
8067 | * However, the presence of @min_alloc_size gives a | |
8068 | * chance to try and satisfy the smaller allocation. | |
8069 | * | |
8070 | * @empty_size - A hint that you plan on doing more COW. This is the | |
8071 | * size in bytes the allocator should try to find free | |
8072 | * next to the block it returns. This is just a hint and | |
8073 | * may be ignored by the allocator. | |
8074 | * | |
8075 | * @hint_byte - Hint to the allocator to start searching above the byte | |
8076 | * address passed. It might be ignored. | |
8077 | * | |
8078 | * @ins - This key is modified to record the found hole. It will | |
8079 | * have the following values: | |
8080 | * ins->objectid == start position | |
8081 | * ins->flags = BTRFS_EXTENT_ITEM_KEY | |
8082 | * ins->offset == the size of the hole. | |
8083 | * | |
8084 | * @is_data - Boolean flag indicating whether an extent is | |
8085 | * allocated for data (true) or metadata (false) | |
8086 | * | |
8087 | * @delalloc - Boolean flag indicating whether this allocation is for | |
8088 | * delalloc or not. If 'true' data_rwsem of block groups | |
8089 | * is going to be acquired. | |
8090 | * | |
8091 | * | |
8092 | * Returns 0 when an allocation succeeded or < 0 when an error occurred. In | |
8093 | * case -ENOSPC is returned then @ins->offset will contain the size of the | |
8094 | * largest available hole the allocator managed to find. | |
8095 | */ | |
18513091 | 8096 | int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes, |
11833d66 YZ |
8097 | u64 num_bytes, u64 min_alloc_size, |
8098 | u64 empty_size, u64 hint_byte, | |
e570fd27 | 8099 | struct btrfs_key *ins, int is_data, int delalloc) |
fec577fb | 8100 | { |
ab8d0fc4 | 8101 | struct btrfs_fs_info *fs_info = root->fs_info; |
36af4e07 | 8102 | bool final_tried = num_bytes == min_alloc_size; |
b6919a58 | 8103 | u64 flags; |
fec577fb | 8104 | int ret; |
925baedd | 8105 | |
1b86826d | 8106 | flags = get_alloc_profile_by_root(root, is_data); |
98d20f67 | 8107 | again: |
0b246afa | 8108 | WARN_ON(num_bytes < fs_info->sectorsize); |
87bde3cd | 8109 | ret = find_free_extent(fs_info, ram_bytes, num_bytes, empty_size, |
18513091 | 8110 | hint_byte, ins, flags, delalloc); |
9cfa3e34 | 8111 | if (!ret && !is_data) { |
ab8d0fc4 | 8112 | btrfs_dec_block_group_reservations(fs_info, ins->objectid); |
9cfa3e34 | 8113 | } else if (ret == -ENOSPC) { |
a4820398 MX |
8114 | if (!final_tried && ins->offset) { |
8115 | num_bytes = min(num_bytes >> 1, ins->offset); | |
da17066c | 8116 | num_bytes = round_down(num_bytes, |
0b246afa | 8117 | fs_info->sectorsize); |
9e622d6b | 8118 | num_bytes = max(num_bytes, min_alloc_size); |
18513091 | 8119 | ram_bytes = num_bytes; |
9e622d6b MX |
8120 | if (num_bytes == min_alloc_size) |
8121 | final_tried = true; | |
8122 | goto again; | |
ab8d0fc4 | 8123 | } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
9e622d6b MX |
8124 | struct btrfs_space_info *sinfo; |
8125 | ||
ab8d0fc4 | 8126 | sinfo = __find_space_info(fs_info, flags); |
0b246afa | 8127 | btrfs_err(fs_info, |
5d163e0e JM |
8128 | "allocation failed flags %llu, wanted %llu", |
8129 | flags, num_bytes); | |
53804280 | 8130 | if (sinfo) |
ab8d0fc4 | 8131 | dump_space_info(fs_info, sinfo, num_bytes, 1); |
9e622d6b | 8132 | } |
925baedd | 8133 | } |
0f9dd46c JB |
8134 | |
8135 | return ret; | |
e6dcd2dc CM |
8136 | } |
8137 | ||
2ff7e61e | 8138 | static int __btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info, |
e570fd27 MX |
8139 | u64 start, u64 len, |
8140 | int pin, int delalloc) | |
65b51a00 | 8141 | { |
0f9dd46c | 8142 | struct btrfs_block_group_cache *cache; |
1f3c79a2 | 8143 | int ret = 0; |
0f9dd46c | 8144 | |
0b246afa | 8145 | cache = btrfs_lookup_block_group(fs_info, start); |
0f9dd46c | 8146 | if (!cache) { |
0b246afa JM |
8147 | btrfs_err(fs_info, "Unable to find block group for %llu", |
8148 | start); | |
0f9dd46c JB |
8149 | return -ENOSPC; |
8150 | } | |
1f3c79a2 | 8151 | |
e688b725 | 8152 | if (pin) |
fdf08605 | 8153 | pin_down_extent(cache, start, len, 1); |
e688b725 | 8154 | else { |
0b246afa | 8155 | if (btrfs_test_opt(fs_info, DISCARD)) |
2ff7e61e | 8156 | ret = btrfs_discard_extent(fs_info, start, len, NULL); |
e688b725 | 8157 | btrfs_add_free_space(cache, start, len); |
4824f1f4 | 8158 | btrfs_free_reserved_bytes(cache, len, delalloc); |
71ff6437 | 8159 | trace_btrfs_reserved_extent_free(fs_info, start, len); |
e688b725 | 8160 | } |
31193213 | 8161 | |
fa9c0d79 | 8162 | btrfs_put_block_group(cache); |
e6dcd2dc CM |
8163 | return ret; |
8164 | } | |
8165 | ||
2ff7e61e | 8166 | int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info, |
e570fd27 | 8167 | u64 start, u64 len, int delalloc) |
e688b725 | 8168 | { |
2ff7e61e | 8169 | return __btrfs_free_reserved_extent(fs_info, start, len, 0, delalloc); |
e688b725 CM |
8170 | } |
8171 | ||
2ff7e61e | 8172 | int btrfs_free_and_pin_reserved_extent(struct btrfs_fs_info *fs_info, |
e688b725 CM |
8173 | u64 start, u64 len) |
8174 | { | |
2ff7e61e | 8175 | return __btrfs_free_reserved_extent(fs_info, start, len, 1, 0); |
e688b725 CM |
8176 | } |
8177 | ||
5d4f98a2 | 8178 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
8179 | u64 parent, u64 root_objectid, |
8180 | u64 flags, u64 owner, u64 offset, | |
8181 | struct btrfs_key *ins, int ref_mod) | |
e6dcd2dc | 8182 | { |
ef89b824 | 8183 | struct btrfs_fs_info *fs_info = trans->fs_info; |
e6dcd2dc | 8184 | int ret; |
e6dcd2dc | 8185 | struct btrfs_extent_item *extent_item; |
5d4f98a2 | 8186 | struct btrfs_extent_inline_ref *iref; |
e6dcd2dc | 8187 | struct btrfs_path *path; |
5d4f98a2 YZ |
8188 | struct extent_buffer *leaf; |
8189 | int type; | |
8190 | u32 size; | |
26b8003f | 8191 | |
5d4f98a2 YZ |
8192 | if (parent > 0) |
8193 | type = BTRFS_SHARED_DATA_REF_KEY; | |
8194 | else | |
8195 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
58176a96 | 8196 | |
5d4f98a2 | 8197 | size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type); |
7bb86316 CM |
8198 | |
8199 | path = btrfs_alloc_path(); | |
db5b493a TI |
8200 | if (!path) |
8201 | return -ENOMEM; | |
47e4bb98 | 8202 | |
b9473439 | 8203 | path->leave_spinning = 1; |
5d4f98a2 YZ |
8204 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, |
8205 | ins, size); | |
79787eaa JM |
8206 | if (ret) { |
8207 | btrfs_free_path(path); | |
8208 | return ret; | |
8209 | } | |
0f9dd46c | 8210 | |
5d4f98a2 YZ |
8211 | leaf = path->nodes[0]; |
8212 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
47e4bb98 | 8213 | struct btrfs_extent_item); |
5d4f98a2 YZ |
8214 | btrfs_set_extent_refs(leaf, extent_item, ref_mod); |
8215 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
8216 | btrfs_set_extent_flags(leaf, extent_item, | |
8217 | flags | BTRFS_EXTENT_FLAG_DATA); | |
8218 | ||
8219 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
8220 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
8221 | if (parent > 0) { | |
8222 | struct btrfs_shared_data_ref *ref; | |
8223 | ref = (struct btrfs_shared_data_ref *)(iref + 1); | |
8224 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
8225 | btrfs_set_shared_data_ref_count(leaf, ref, ref_mod); | |
8226 | } else { | |
8227 | struct btrfs_extent_data_ref *ref; | |
8228 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
8229 | btrfs_set_extent_data_ref_root(leaf, ref, root_objectid); | |
8230 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
8231 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
8232 | btrfs_set_extent_data_ref_count(leaf, ref, ref_mod); | |
8233 | } | |
47e4bb98 CM |
8234 | |
8235 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
7bb86316 | 8236 | btrfs_free_path(path); |
f510cfec | 8237 | |
25a356d3 | 8238 | ret = remove_from_free_space_tree(trans, ins->objectid, ins->offset); |
1e144fb8 OS |
8239 | if (ret) |
8240 | return ret; | |
8241 | ||
6b279408 | 8242 | ret = update_block_group(trans, ins->objectid, ins->offset, 1); |
79787eaa | 8243 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb | 8244 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
c1c9ff7c | 8245 | ins->objectid, ins->offset); |
f5947066 CM |
8246 | BUG(); |
8247 | } | |
71ff6437 | 8248 | trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid, ins->offset); |
e6dcd2dc CM |
8249 | return ret; |
8250 | } | |
8251 | ||
5d4f98a2 | 8252 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, |
4e6bd4e0 | 8253 | struct btrfs_delayed_ref_node *node, |
21ebfbe7 | 8254 | struct btrfs_delayed_extent_op *extent_op) |
e6dcd2dc | 8255 | { |
9dcdbe01 | 8256 | struct btrfs_fs_info *fs_info = trans->fs_info; |
e6dcd2dc | 8257 | int ret; |
5d4f98a2 | 8258 | struct btrfs_extent_item *extent_item; |
4e6bd4e0 | 8259 | struct btrfs_key extent_key; |
5d4f98a2 YZ |
8260 | struct btrfs_tree_block_info *block_info; |
8261 | struct btrfs_extent_inline_ref *iref; | |
8262 | struct btrfs_path *path; | |
8263 | struct extent_buffer *leaf; | |
4e6bd4e0 | 8264 | struct btrfs_delayed_tree_ref *ref; |
3173a18f | 8265 | u32 size = sizeof(*extent_item) + sizeof(*iref); |
4e6bd4e0 | 8266 | u64 num_bytes; |
21ebfbe7 | 8267 | u64 flags = extent_op->flags_to_set; |
0b246afa | 8268 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); |
3173a18f | 8269 | |
4e6bd4e0 NB |
8270 | ref = btrfs_delayed_node_to_tree_ref(node); |
8271 | ||
4e6bd4e0 NB |
8272 | extent_key.objectid = node->bytenr; |
8273 | if (skinny_metadata) { | |
8274 | extent_key.offset = ref->level; | |
8275 | extent_key.type = BTRFS_METADATA_ITEM_KEY; | |
8276 | num_bytes = fs_info->nodesize; | |
8277 | } else { | |
8278 | extent_key.offset = node->num_bytes; | |
8279 | extent_key.type = BTRFS_EXTENT_ITEM_KEY; | |
3173a18f | 8280 | size += sizeof(*block_info); |
4e6bd4e0 NB |
8281 | num_bytes = node->num_bytes; |
8282 | } | |
1c2308f8 | 8283 | |
5d4f98a2 | 8284 | path = btrfs_alloc_path(); |
80ee54bf | 8285 | if (!path) |
d8926bb3 | 8286 | return -ENOMEM; |
56bec294 | 8287 | |
5d4f98a2 YZ |
8288 | path->leave_spinning = 1; |
8289 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, | |
4e6bd4e0 | 8290 | &extent_key, size); |
79787eaa | 8291 | if (ret) { |
dd825259 | 8292 | btrfs_free_path(path); |
79787eaa JM |
8293 | return ret; |
8294 | } | |
5d4f98a2 YZ |
8295 | |
8296 | leaf = path->nodes[0]; | |
8297 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
8298 | struct btrfs_extent_item); | |
8299 | btrfs_set_extent_refs(leaf, extent_item, 1); | |
8300 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
8301 | btrfs_set_extent_flags(leaf, extent_item, | |
8302 | flags | BTRFS_EXTENT_FLAG_TREE_BLOCK); | |
5d4f98a2 | 8303 | |
3173a18f JB |
8304 | if (skinny_metadata) { |
8305 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
8306 | } else { | |
8307 | block_info = (struct btrfs_tree_block_info *)(extent_item + 1); | |
21ebfbe7 | 8308 | btrfs_set_tree_block_key(leaf, block_info, &extent_op->key); |
4e6bd4e0 | 8309 | btrfs_set_tree_block_level(leaf, block_info, ref->level); |
3173a18f JB |
8310 | iref = (struct btrfs_extent_inline_ref *)(block_info + 1); |
8311 | } | |
5d4f98a2 | 8312 | |
d4b20733 | 8313 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) { |
5d4f98a2 YZ |
8314 | BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); |
8315 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
8316 | BTRFS_SHARED_BLOCK_REF_KEY); | |
d4b20733 | 8317 | btrfs_set_extent_inline_ref_offset(leaf, iref, ref->parent); |
5d4f98a2 YZ |
8318 | } else { |
8319 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
8320 | BTRFS_TREE_BLOCK_REF_KEY); | |
4e6bd4e0 | 8321 | btrfs_set_extent_inline_ref_offset(leaf, iref, ref->root); |
5d4f98a2 YZ |
8322 | } |
8323 | ||
8324 | btrfs_mark_buffer_dirty(leaf); | |
8325 | btrfs_free_path(path); | |
8326 | ||
4e6bd4e0 NB |
8327 | ret = remove_from_free_space_tree(trans, extent_key.objectid, |
8328 | num_bytes); | |
1e144fb8 OS |
8329 | if (ret) |
8330 | return ret; | |
8331 | ||
6b279408 | 8332 | ret = update_block_group(trans, extent_key.objectid, |
6202df69 | 8333 | fs_info->nodesize, 1); |
79787eaa | 8334 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb | 8335 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
4e6bd4e0 | 8336 | extent_key.objectid, extent_key.offset); |
5d4f98a2 YZ |
8337 | BUG(); |
8338 | } | |
0be5dc67 | 8339 | |
4e6bd4e0 | 8340 | trace_btrfs_reserved_extent_alloc(fs_info, extent_key.objectid, |
0b246afa | 8341 | fs_info->nodesize); |
5d4f98a2 YZ |
8342 | return ret; |
8343 | } | |
8344 | ||
8345 | int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
84f7d8e6 | 8346 | struct btrfs_root *root, u64 owner, |
5846a3c2 QW |
8347 | u64 offset, u64 ram_bytes, |
8348 | struct btrfs_key *ins) | |
5d4f98a2 | 8349 | { |
76675593 | 8350 | struct btrfs_ref generic_ref = { 0 }; |
5d4f98a2 YZ |
8351 | int ret; |
8352 | ||
84f7d8e6 | 8353 | BUG_ON(root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID); |
5d4f98a2 | 8354 | |
76675593 QW |
8355 | btrfs_init_generic_ref(&generic_ref, BTRFS_ADD_DELAYED_EXTENT, |
8356 | ins->objectid, ins->offset, 0); | |
8357 | btrfs_init_data_ref(&generic_ref, root->root_key.objectid, owner, offset); | |
8a5040f7 | 8358 | btrfs_ref_tree_mod(root->fs_info, &generic_ref); |
76675593 QW |
8359 | ret = btrfs_add_delayed_data_ref(trans, &generic_ref, |
8360 | ram_bytes, NULL, NULL); | |
e6dcd2dc CM |
8361 | return ret; |
8362 | } | |
e02119d5 CM |
8363 | |
8364 | /* | |
8365 | * this is used by the tree logging recovery code. It records that | |
8366 | * an extent has been allocated and makes sure to clear the free | |
8367 | * space cache bits as well | |
8368 | */ | |
5d4f98a2 | 8369 | int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
8370 | u64 root_objectid, u64 owner, u64 offset, |
8371 | struct btrfs_key *ins) | |
e02119d5 | 8372 | { |
61da2abf | 8373 | struct btrfs_fs_info *fs_info = trans->fs_info; |
e02119d5 CM |
8374 | int ret; |
8375 | struct btrfs_block_group_cache *block_group; | |
ed7a6948 | 8376 | struct btrfs_space_info *space_info; |
11833d66 | 8377 | |
8c2a1a30 JB |
8378 | /* |
8379 | * Mixed block groups will exclude before processing the log so we only | |
01327610 | 8380 | * need to do the exclude dance if this fs isn't mixed. |
8c2a1a30 | 8381 | */ |
0b246afa | 8382 | if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) { |
2ff7e61e JM |
8383 | ret = __exclude_logged_extent(fs_info, ins->objectid, |
8384 | ins->offset); | |
b50c6e25 | 8385 | if (ret) |
8c2a1a30 | 8386 | return ret; |
11833d66 YZ |
8387 | } |
8388 | ||
0b246afa | 8389 | block_group = btrfs_lookup_block_group(fs_info, ins->objectid); |
8c2a1a30 JB |
8390 | if (!block_group) |
8391 | return -EINVAL; | |
8392 | ||
ed7a6948 WX |
8393 | space_info = block_group->space_info; |
8394 | spin_lock(&space_info->lock); | |
8395 | spin_lock(&block_group->lock); | |
8396 | space_info->bytes_reserved += ins->offset; | |
8397 | block_group->reserved += ins->offset; | |
8398 | spin_unlock(&block_group->lock); | |
8399 | spin_unlock(&space_info->lock); | |
8400 | ||
ef89b824 NB |
8401 | ret = alloc_reserved_file_extent(trans, 0, root_objectid, 0, owner, |
8402 | offset, ins, 1); | |
b50c6e25 | 8403 | btrfs_put_block_group(block_group); |
e02119d5 CM |
8404 | return ret; |
8405 | } | |
8406 | ||
48a3b636 ES |
8407 | static struct extent_buffer * |
8408 | btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
bc877d28 | 8409 | u64 bytenr, int level, u64 owner) |
65b51a00 | 8410 | { |
0b246afa | 8411 | struct btrfs_fs_info *fs_info = root->fs_info; |
65b51a00 CM |
8412 | struct extent_buffer *buf; |
8413 | ||
2ff7e61e | 8414 | buf = btrfs_find_create_tree_block(fs_info, bytenr); |
c871b0f2 LB |
8415 | if (IS_ERR(buf)) |
8416 | return buf; | |
8417 | ||
b72c3aba QW |
8418 | /* |
8419 | * Extra safety check in case the extent tree is corrupted and extent | |
8420 | * allocator chooses to use a tree block which is already used and | |
8421 | * locked. | |
8422 | */ | |
8423 | if (buf->lock_owner == current->pid) { | |
8424 | btrfs_err_rl(fs_info, | |
8425 | "tree block %llu owner %llu already locked by pid=%d, extent tree corruption detected", | |
8426 | buf->start, btrfs_header_owner(buf), current->pid); | |
8427 | free_extent_buffer(buf); | |
8428 | return ERR_PTR(-EUCLEAN); | |
8429 | } | |
8430 | ||
85d4e461 | 8431 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level); |
65b51a00 | 8432 | btrfs_tree_lock(buf); |
6a884d7d | 8433 | btrfs_clean_tree_block(buf); |
3083ee2e | 8434 | clear_bit(EXTENT_BUFFER_STALE, &buf->bflags); |
b4ce94de | 8435 | |
8bead258 | 8436 | btrfs_set_lock_blocking_write(buf); |
4db8c528 | 8437 | set_extent_buffer_uptodate(buf); |
b4ce94de | 8438 | |
bc877d28 NB |
8439 | memzero_extent_buffer(buf, 0, sizeof(struct btrfs_header)); |
8440 | btrfs_set_header_level(buf, level); | |
8441 | btrfs_set_header_bytenr(buf, buf->start); | |
8442 | btrfs_set_header_generation(buf, trans->transid); | |
8443 | btrfs_set_header_backref_rev(buf, BTRFS_MIXED_BACKREF_REV); | |
8444 | btrfs_set_header_owner(buf, owner); | |
de37aa51 | 8445 | write_extent_buffer_fsid(buf, fs_info->fs_devices->metadata_uuid); |
bc877d28 | 8446 | write_extent_buffer_chunk_tree_uuid(buf, fs_info->chunk_tree_uuid); |
d0c803c4 | 8447 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { |
656f30db | 8448 | buf->log_index = root->log_transid % 2; |
8cef4e16 YZ |
8449 | /* |
8450 | * we allow two log transactions at a time, use different | |
52042d8e | 8451 | * EXTENT bit to differentiate dirty pages. |
8cef4e16 | 8452 | */ |
656f30db | 8453 | if (buf->log_index == 0) |
8cef4e16 YZ |
8454 | set_extent_dirty(&root->dirty_log_pages, buf->start, |
8455 | buf->start + buf->len - 1, GFP_NOFS); | |
8456 | else | |
8457 | set_extent_new(&root->dirty_log_pages, buf->start, | |
3744dbeb | 8458 | buf->start + buf->len - 1); |
d0c803c4 | 8459 | } else { |
656f30db | 8460 | buf->log_index = -1; |
d0c803c4 | 8461 | set_extent_dirty(&trans->transaction->dirty_pages, buf->start, |
65b51a00 | 8462 | buf->start + buf->len - 1, GFP_NOFS); |
d0c803c4 | 8463 | } |
64c12921 | 8464 | trans->dirty = true; |
b4ce94de | 8465 | /* this returns a buffer locked for blocking */ |
65b51a00 CM |
8466 | return buf; |
8467 | } | |
8468 | ||
f0486c68 YZ |
8469 | static struct btrfs_block_rsv * |
8470 | use_block_rsv(struct btrfs_trans_handle *trans, | |
8471 | struct btrfs_root *root, u32 blocksize) | |
8472 | { | |
0b246afa | 8473 | struct btrfs_fs_info *fs_info = root->fs_info; |
f0486c68 | 8474 | struct btrfs_block_rsv *block_rsv; |
0b246afa | 8475 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; |
f0486c68 | 8476 | int ret; |
d88033db | 8477 | bool global_updated = false; |
f0486c68 YZ |
8478 | |
8479 | block_rsv = get_block_rsv(trans, root); | |
8480 | ||
b586b323 MX |
8481 | if (unlikely(block_rsv->size == 0)) |
8482 | goto try_reserve; | |
d88033db | 8483 | again: |
f0486c68 YZ |
8484 | ret = block_rsv_use_bytes(block_rsv, blocksize); |
8485 | if (!ret) | |
8486 | return block_rsv; | |
8487 | ||
b586b323 MX |
8488 | if (block_rsv->failfast) |
8489 | return ERR_PTR(ret); | |
8490 | ||
d88033db MX |
8491 | if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) { |
8492 | global_updated = true; | |
0b246afa | 8493 | update_global_block_rsv(fs_info); |
d88033db MX |
8494 | goto again; |
8495 | } | |
8496 | ||
ba2c4d4e JB |
8497 | /* |
8498 | * The global reserve still exists to save us from ourselves, so don't | |
8499 | * warn_on if we are short on our delayed refs reserve. | |
8500 | */ | |
8501 | if (block_rsv->type != BTRFS_BLOCK_RSV_DELREFS && | |
8502 | btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { | |
b586b323 MX |
8503 | static DEFINE_RATELIMIT_STATE(_rs, |
8504 | DEFAULT_RATELIMIT_INTERVAL * 10, | |
8505 | /*DEFAULT_RATELIMIT_BURST*/ 1); | |
8506 | if (__ratelimit(&_rs)) | |
8507 | WARN(1, KERN_DEBUG | |
efe120a0 | 8508 | "BTRFS: block rsv returned %d\n", ret); |
b586b323 MX |
8509 | } |
8510 | try_reserve: | |
8511 | ret = reserve_metadata_bytes(root, block_rsv, blocksize, | |
8512 | BTRFS_RESERVE_NO_FLUSH); | |
8513 | if (!ret) | |
8514 | return block_rsv; | |
8515 | /* | |
8516 | * If we couldn't reserve metadata bytes try and use some from | |
5881cfc9 MX |
8517 | * the global reserve if its space type is the same as the global |
8518 | * reservation. | |
b586b323 | 8519 | */ |
5881cfc9 MX |
8520 | if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL && |
8521 | block_rsv->space_info == global_rsv->space_info) { | |
b586b323 MX |
8522 | ret = block_rsv_use_bytes(global_rsv, blocksize); |
8523 | if (!ret) | |
8524 | return global_rsv; | |
8525 | } | |
8526 | return ERR_PTR(ret); | |
f0486c68 YZ |
8527 | } |
8528 | ||
8c2a3ca2 JB |
8529 | static void unuse_block_rsv(struct btrfs_fs_info *fs_info, |
8530 | struct btrfs_block_rsv *block_rsv, u32 blocksize) | |
f0486c68 | 8531 | { |
3a584174 | 8532 | block_rsv_add_bytes(block_rsv, blocksize, false); |
ff6bc37e | 8533 | block_rsv_release_bytes(fs_info, block_rsv, NULL, 0, NULL); |
f0486c68 YZ |
8534 | } |
8535 | ||
fec577fb | 8536 | /* |
f0486c68 | 8537 | * finds a free extent and does all the dirty work required for allocation |
67b7859e | 8538 | * returns the tree buffer or an ERR_PTR on error. |
fec577fb | 8539 | */ |
4d75f8a9 | 8540 | struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans, |
310712b2 OS |
8541 | struct btrfs_root *root, |
8542 | u64 parent, u64 root_objectid, | |
8543 | const struct btrfs_disk_key *key, | |
8544 | int level, u64 hint, | |
8545 | u64 empty_size) | |
fec577fb | 8546 | { |
0b246afa | 8547 | struct btrfs_fs_info *fs_info = root->fs_info; |
e2fa7227 | 8548 | struct btrfs_key ins; |
f0486c68 | 8549 | struct btrfs_block_rsv *block_rsv; |
5f39d397 | 8550 | struct extent_buffer *buf; |
67b7859e | 8551 | struct btrfs_delayed_extent_op *extent_op; |
ed4f255b | 8552 | struct btrfs_ref generic_ref = { 0 }; |
f0486c68 YZ |
8553 | u64 flags = 0; |
8554 | int ret; | |
0b246afa JM |
8555 | u32 blocksize = fs_info->nodesize; |
8556 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); | |
fec577fb | 8557 | |
05653ef3 | 8558 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
0b246afa | 8559 | if (btrfs_is_testing(fs_info)) { |
faa2dbf0 | 8560 | buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr, |
bc877d28 | 8561 | level, root_objectid); |
faa2dbf0 JB |
8562 | if (!IS_ERR(buf)) |
8563 | root->alloc_bytenr += blocksize; | |
8564 | return buf; | |
8565 | } | |
05653ef3 | 8566 | #endif |
fccb84c9 | 8567 | |
f0486c68 YZ |
8568 | block_rsv = use_block_rsv(trans, root, blocksize); |
8569 | if (IS_ERR(block_rsv)) | |
8570 | return ERR_CAST(block_rsv); | |
8571 | ||
18513091 | 8572 | ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize, |
e570fd27 | 8573 | empty_size, hint, &ins, 0, 0); |
67b7859e OS |
8574 | if (ret) |
8575 | goto out_unuse; | |
55c69072 | 8576 | |
bc877d28 NB |
8577 | buf = btrfs_init_new_buffer(trans, root, ins.objectid, level, |
8578 | root_objectid); | |
67b7859e OS |
8579 | if (IS_ERR(buf)) { |
8580 | ret = PTR_ERR(buf); | |
8581 | goto out_free_reserved; | |
8582 | } | |
f0486c68 YZ |
8583 | |
8584 | if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
8585 | if (parent == 0) | |
8586 | parent = ins.objectid; | |
8587 | flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
8588 | } else | |
8589 | BUG_ON(parent > 0); | |
8590 | ||
8591 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) { | |
78a6184a | 8592 | extent_op = btrfs_alloc_delayed_extent_op(); |
67b7859e OS |
8593 | if (!extent_op) { |
8594 | ret = -ENOMEM; | |
8595 | goto out_free_buf; | |
8596 | } | |
f0486c68 YZ |
8597 | if (key) |
8598 | memcpy(&extent_op->key, key, sizeof(extent_op->key)); | |
8599 | else | |
8600 | memset(&extent_op->key, 0, sizeof(extent_op->key)); | |
8601 | extent_op->flags_to_set = flags; | |
35b3ad50 DS |
8602 | extent_op->update_key = skinny_metadata ? false : true; |
8603 | extent_op->update_flags = true; | |
8604 | extent_op->is_data = false; | |
b1c79e09 | 8605 | extent_op->level = level; |
f0486c68 | 8606 | |
ed4f255b QW |
8607 | btrfs_init_generic_ref(&generic_ref, BTRFS_ADD_DELAYED_EXTENT, |
8608 | ins.objectid, ins.offset, parent); | |
8609 | generic_ref.real_root = root->root_key.objectid; | |
8610 | btrfs_init_tree_ref(&generic_ref, level, root_objectid); | |
8a5040f7 | 8611 | btrfs_ref_tree_mod(fs_info, &generic_ref); |
ed4f255b | 8612 | ret = btrfs_add_delayed_tree_ref(trans, &generic_ref, |
7be07912 | 8613 | extent_op, NULL, NULL); |
67b7859e OS |
8614 | if (ret) |
8615 | goto out_free_delayed; | |
f0486c68 | 8616 | } |
fec577fb | 8617 | return buf; |
67b7859e OS |
8618 | |
8619 | out_free_delayed: | |
8620 | btrfs_free_delayed_extent_op(extent_op); | |
8621 | out_free_buf: | |
8622 | free_extent_buffer(buf); | |
8623 | out_free_reserved: | |
2ff7e61e | 8624 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0); |
67b7859e | 8625 | out_unuse: |
0b246afa | 8626 | unuse_block_rsv(fs_info, block_rsv, blocksize); |
67b7859e | 8627 | return ERR_PTR(ret); |
fec577fb | 8628 | } |
a28ec197 | 8629 | |
2c47e605 YZ |
8630 | struct walk_control { |
8631 | u64 refs[BTRFS_MAX_LEVEL]; | |
8632 | u64 flags[BTRFS_MAX_LEVEL]; | |
8633 | struct btrfs_key update_progress; | |
aea6f028 JB |
8634 | struct btrfs_key drop_progress; |
8635 | int drop_level; | |
2c47e605 YZ |
8636 | int stage; |
8637 | int level; | |
8638 | int shared_level; | |
8639 | int update_ref; | |
8640 | int keep_locks; | |
1c4850e2 YZ |
8641 | int reada_slot; |
8642 | int reada_count; | |
78c52d9e | 8643 | int restarted; |
2c47e605 YZ |
8644 | }; |
8645 | ||
8646 | #define DROP_REFERENCE 1 | |
8647 | #define UPDATE_BACKREF 2 | |
8648 | ||
1c4850e2 YZ |
8649 | static noinline void reada_walk_down(struct btrfs_trans_handle *trans, |
8650 | struct btrfs_root *root, | |
8651 | struct walk_control *wc, | |
8652 | struct btrfs_path *path) | |
6407bf6d | 8653 | { |
0b246afa | 8654 | struct btrfs_fs_info *fs_info = root->fs_info; |
1c4850e2 YZ |
8655 | u64 bytenr; |
8656 | u64 generation; | |
8657 | u64 refs; | |
94fcca9f | 8658 | u64 flags; |
5d4f98a2 | 8659 | u32 nritems; |
1c4850e2 YZ |
8660 | struct btrfs_key key; |
8661 | struct extent_buffer *eb; | |
6407bf6d | 8662 | int ret; |
1c4850e2 YZ |
8663 | int slot; |
8664 | int nread = 0; | |
6407bf6d | 8665 | |
1c4850e2 YZ |
8666 | if (path->slots[wc->level] < wc->reada_slot) { |
8667 | wc->reada_count = wc->reada_count * 2 / 3; | |
8668 | wc->reada_count = max(wc->reada_count, 2); | |
8669 | } else { | |
8670 | wc->reada_count = wc->reada_count * 3 / 2; | |
8671 | wc->reada_count = min_t(int, wc->reada_count, | |
0b246afa | 8672 | BTRFS_NODEPTRS_PER_BLOCK(fs_info)); |
1c4850e2 | 8673 | } |
7bb86316 | 8674 | |
1c4850e2 YZ |
8675 | eb = path->nodes[wc->level]; |
8676 | nritems = btrfs_header_nritems(eb); | |
bd56b302 | 8677 | |
1c4850e2 YZ |
8678 | for (slot = path->slots[wc->level]; slot < nritems; slot++) { |
8679 | if (nread >= wc->reada_count) | |
8680 | break; | |
bd56b302 | 8681 | |
2dd3e67b | 8682 | cond_resched(); |
1c4850e2 YZ |
8683 | bytenr = btrfs_node_blockptr(eb, slot); |
8684 | generation = btrfs_node_ptr_generation(eb, slot); | |
2dd3e67b | 8685 | |
1c4850e2 YZ |
8686 | if (slot == path->slots[wc->level]) |
8687 | goto reada; | |
5d4f98a2 | 8688 | |
1c4850e2 YZ |
8689 | if (wc->stage == UPDATE_BACKREF && |
8690 | generation <= root->root_key.offset) | |
bd56b302 CM |
8691 | continue; |
8692 | ||
94fcca9f | 8693 | /* We don't lock the tree block, it's OK to be racy here */ |
2ff7e61e | 8694 | ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, |
3173a18f JB |
8695 | wc->level - 1, 1, &refs, |
8696 | &flags); | |
79787eaa JM |
8697 | /* We don't care about errors in readahead. */ |
8698 | if (ret < 0) | |
8699 | continue; | |
94fcca9f YZ |
8700 | BUG_ON(refs == 0); |
8701 | ||
1c4850e2 | 8702 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 YZ |
8703 | if (refs == 1) |
8704 | goto reada; | |
bd56b302 | 8705 | |
94fcca9f YZ |
8706 | if (wc->level == 1 && |
8707 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8708 | continue; | |
1c4850e2 YZ |
8709 | if (!wc->update_ref || |
8710 | generation <= root->root_key.offset) | |
8711 | continue; | |
8712 | btrfs_node_key_to_cpu(eb, &key, slot); | |
8713 | ret = btrfs_comp_cpu_keys(&key, | |
8714 | &wc->update_progress); | |
8715 | if (ret < 0) | |
8716 | continue; | |
94fcca9f YZ |
8717 | } else { |
8718 | if (wc->level == 1 && | |
8719 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8720 | continue; | |
6407bf6d | 8721 | } |
1c4850e2 | 8722 | reada: |
2ff7e61e | 8723 | readahead_tree_block(fs_info, bytenr); |
1c4850e2 | 8724 | nread++; |
20524f02 | 8725 | } |
1c4850e2 | 8726 | wc->reada_slot = slot; |
20524f02 | 8727 | } |
2c47e605 | 8728 | |
f82d02d9 | 8729 | /* |
2c016dc2 | 8730 | * helper to process tree block while walking down the tree. |
2c47e605 | 8731 | * |
2c47e605 YZ |
8732 | * when wc->stage == UPDATE_BACKREF, this function updates |
8733 | * back refs for pointers in the block. | |
8734 | * | |
8735 | * NOTE: return value 1 means we should stop walking down. | |
f82d02d9 | 8736 | */ |
2c47e605 | 8737 | static noinline int walk_down_proc(struct btrfs_trans_handle *trans, |
5d4f98a2 | 8738 | struct btrfs_root *root, |
2c47e605 | 8739 | struct btrfs_path *path, |
94fcca9f | 8740 | struct walk_control *wc, int lookup_info) |
f82d02d9 | 8741 | { |
2ff7e61e | 8742 | struct btrfs_fs_info *fs_info = root->fs_info; |
2c47e605 YZ |
8743 | int level = wc->level; |
8744 | struct extent_buffer *eb = path->nodes[level]; | |
2c47e605 | 8745 | u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF; |
f82d02d9 YZ |
8746 | int ret; |
8747 | ||
2c47e605 YZ |
8748 | if (wc->stage == UPDATE_BACKREF && |
8749 | btrfs_header_owner(eb) != root->root_key.objectid) | |
8750 | return 1; | |
f82d02d9 | 8751 | |
2c47e605 YZ |
8752 | /* |
8753 | * when reference count of tree block is 1, it won't increase | |
8754 | * again. once full backref flag is set, we never clear it. | |
8755 | */ | |
94fcca9f YZ |
8756 | if (lookup_info && |
8757 | ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) || | |
8758 | (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) { | |
2c47e605 | 8759 | BUG_ON(!path->locks[level]); |
2ff7e61e | 8760 | ret = btrfs_lookup_extent_info(trans, fs_info, |
3173a18f | 8761 | eb->start, level, 1, |
2c47e605 YZ |
8762 | &wc->refs[level], |
8763 | &wc->flags[level]); | |
79787eaa JM |
8764 | BUG_ON(ret == -ENOMEM); |
8765 | if (ret) | |
8766 | return ret; | |
2c47e605 YZ |
8767 | BUG_ON(wc->refs[level] == 0); |
8768 | } | |
5d4f98a2 | 8769 | |
2c47e605 YZ |
8770 | if (wc->stage == DROP_REFERENCE) { |
8771 | if (wc->refs[level] > 1) | |
8772 | return 1; | |
f82d02d9 | 8773 | |
2c47e605 | 8774 | if (path->locks[level] && !wc->keep_locks) { |
bd681513 | 8775 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
8776 | path->locks[level] = 0; |
8777 | } | |
8778 | return 0; | |
8779 | } | |
f82d02d9 | 8780 | |
2c47e605 YZ |
8781 | /* wc->stage == UPDATE_BACKREF */ |
8782 | if (!(wc->flags[level] & flag)) { | |
8783 | BUG_ON(!path->locks[level]); | |
e339a6b0 | 8784 | ret = btrfs_inc_ref(trans, root, eb, 1); |
79787eaa | 8785 | BUG_ON(ret); /* -ENOMEM */ |
e339a6b0 | 8786 | ret = btrfs_dec_ref(trans, root, eb, 0); |
79787eaa | 8787 | BUG_ON(ret); /* -ENOMEM */ |
f5c8daa5 | 8788 | ret = btrfs_set_disk_extent_flags(trans, eb->start, |
b1c79e09 JB |
8789 | eb->len, flag, |
8790 | btrfs_header_level(eb), 0); | |
79787eaa | 8791 | BUG_ON(ret); /* -ENOMEM */ |
2c47e605 YZ |
8792 | wc->flags[level] |= flag; |
8793 | } | |
8794 | ||
8795 | /* | |
8796 | * the block is shared by multiple trees, so it's not good to | |
8797 | * keep the tree lock | |
8798 | */ | |
8799 | if (path->locks[level] && level > 0) { | |
bd681513 | 8800 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
8801 | path->locks[level] = 0; |
8802 | } | |
8803 | return 0; | |
8804 | } | |
8805 | ||
78c52d9e JB |
8806 | /* |
8807 | * This is used to verify a ref exists for this root to deal with a bug where we | |
8808 | * would have a drop_progress key that hadn't been updated properly. | |
8809 | */ | |
8810 | static int check_ref_exists(struct btrfs_trans_handle *trans, | |
8811 | struct btrfs_root *root, u64 bytenr, u64 parent, | |
8812 | int level) | |
8813 | { | |
8814 | struct btrfs_path *path; | |
8815 | struct btrfs_extent_inline_ref *iref; | |
8816 | int ret; | |
8817 | ||
8818 | path = btrfs_alloc_path(); | |
8819 | if (!path) | |
8820 | return -ENOMEM; | |
8821 | ||
8822 | ret = lookup_extent_backref(trans, path, &iref, bytenr, | |
8823 | root->fs_info->nodesize, parent, | |
8824 | root->root_key.objectid, level, 0); | |
8825 | btrfs_free_path(path); | |
8826 | if (ret == -ENOENT) | |
8827 | return 0; | |
8828 | if (ret < 0) | |
8829 | return ret; | |
8830 | return 1; | |
8831 | } | |
8832 | ||
1c4850e2 | 8833 | /* |
2c016dc2 | 8834 | * helper to process tree block pointer. |
1c4850e2 YZ |
8835 | * |
8836 | * when wc->stage == DROP_REFERENCE, this function checks | |
8837 | * reference count of the block pointed to. if the block | |
8838 | * is shared and we need update back refs for the subtree | |
8839 | * rooted at the block, this function changes wc->stage to | |
8840 | * UPDATE_BACKREF. if the block is shared and there is no | |
8841 | * need to update back, this function drops the reference | |
8842 | * to the block. | |
8843 | * | |
8844 | * NOTE: return value 1 means we should stop walking down. | |
8845 | */ | |
8846 | static noinline int do_walk_down(struct btrfs_trans_handle *trans, | |
8847 | struct btrfs_root *root, | |
8848 | struct btrfs_path *path, | |
94fcca9f | 8849 | struct walk_control *wc, int *lookup_info) |
1c4850e2 | 8850 | { |
0b246afa | 8851 | struct btrfs_fs_info *fs_info = root->fs_info; |
1c4850e2 YZ |
8852 | u64 bytenr; |
8853 | u64 generation; | |
8854 | u64 parent; | |
1c4850e2 | 8855 | struct btrfs_key key; |
581c1760 | 8856 | struct btrfs_key first_key; |
ffd4bb2a | 8857 | struct btrfs_ref ref = { 0 }; |
1c4850e2 YZ |
8858 | struct extent_buffer *next; |
8859 | int level = wc->level; | |
8860 | int reada = 0; | |
8861 | int ret = 0; | |
1152651a | 8862 | bool need_account = false; |
1c4850e2 YZ |
8863 | |
8864 | generation = btrfs_node_ptr_generation(path->nodes[level], | |
8865 | path->slots[level]); | |
8866 | /* | |
8867 | * if the lower level block was created before the snapshot | |
8868 | * was created, we know there is no need to update back refs | |
8869 | * for the subtree | |
8870 | */ | |
8871 | if (wc->stage == UPDATE_BACKREF && | |
94fcca9f YZ |
8872 | generation <= root->root_key.offset) { |
8873 | *lookup_info = 1; | |
1c4850e2 | 8874 | return 1; |
94fcca9f | 8875 | } |
1c4850e2 YZ |
8876 | |
8877 | bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]); | |
581c1760 QW |
8878 | btrfs_node_key_to_cpu(path->nodes[level], &first_key, |
8879 | path->slots[level]); | |
1c4850e2 | 8880 | |
0b246afa | 8881 | next = find_extent_buffer(fs_info, bytenr); |
1c4850e2 | 8882 | if (!next) { |
2ff7e61e | 8883 | next = btrfs_find_create_tree_block(fs_info, bytenr); |
c871b0f2 LB |
8884 | if (IS_ERR(next)) |
8885 | return PTR_ERR(next); | |
8886 | ||
b2aaaa3b JB |
8887 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, next, |
8888 | level - 1); | |
1c4850e2 YZ |
8889 | reada = 1; |
8890 | } | |
8891 | btrfs_tree_lock(next); | |
8bead258 | 8892 | btrfs_set_lock_blocking_write(next); |
1c4850e2 | 8893 | |
2ff7e61e | 8894 | ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1, |
94fcca9f YZ |
8895 | &wc->refs[level - 1], |
8896 | &wc->flags[level - 1]); | |
4867268c JB |
8897 | if (ret < 0) |
8898 | goto out_unlock; | |
79787eaa | 8899 | |
c2cf52eb | 8900 | if (unlikely(wc->refs[level - 1] == 0)) { |
0b246afa | 8901 | btrfs_err(fs_info, "Missing references."); |
4867268c JB |
8902 | ret = -EIO; |
8903 | goto out_unlock; | |
c2cf52eb | 8904 | } |
94fcca9f | 8905 | *lookup_info = 0; |
1c4850e2 | 8906 | |
94fcca9f | 8907 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 | 8908 | if (wc->refs[level - 1] > 1) { |
1152651a | 8909 | need_account = true; |
94fcca9f YZ |
8910 | if (level == 1 && |
8911 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8912 | goto skip; | |
8913 | ||
1c4850e2 YZ |
8914 | if (!wc->update_ref || |
8915 | generation <= root->root_key.offset) | |
8916 | goto skip; | |
8917 | ||
8918 | btrfs_node_key_to_cpu(path->nodes[level], &key, | |
8919 | path->slots[level]); | |
8920 | ret = btrfs_comp_cpu_keys(&key, &wc->update_progress); | |
8921 | if (ret < 0) | |
8922 | goto skip; | |
8923 | ||
8924 | wc->stage = UPDATE_BACKREF; | |
8925 | wc->shared_level = level - 1; | |
8926 | } | |
94fcca9f YZ |
8927 | } else { |
8928 | if (level == 1 && | |
8929 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8930 | goto skip; | |
1c4850e2 YZ |
8931 | } |
8932 | ||
b9fab919 | 8933 | if (!btrfs_buffer_uptodate(next, generation, 0)) { |
1c4850e2 YZ |
8934 | btrfs_tree_unlock(next); |
8935 | free_extent_buffer(next); | |
8936 | next = NULL; | |
94fcca9f | 8937 | *lookup_info = 1; |
1c4850e2 YZ |
8938 | } |
8939 | ||
8940 | if (!next) { | |
8941 | if (reada && level == 1) | |
8942 | reada_walk_down(trans, root, wc, path); | |
581c1760 QW |
8943 | next = read_tree_block(fs_info, bytenr, generation, level - 1, |
8944 | &first_key); | |
64c043de LB |
8945 | if (IS_ERR(next)) { |
8946 | return PTR_ERR(next); | |
8947 | } else if (!extent_buffer_uptodate(next)) { | |
416bc658 | 8948 | free_extent_buffer(next); |
97d9a8a4 | 8949 | return -EIO; |
416bc658 | 8950 | } |
1c4850e2 | 8951 | btrfs_tree_lock(next); |
8bead258 | 8952 | btrfs_set_lock_blocking_write(next); |
1c4850e2 YZ |
8953 | } |
8954 | ||
8955 | level--; | |
4867268c JB |
8956 | ASSERT(level == btrfs_header_level(next)); |
8957 | if (level != btrfs_header_level(next)) { | |
8958 | btrfs_err(root->fs_info, "mismatched level"); | |
8959 | ret = -EIO; | |
8960 | goto out_unlock; | |
8961 | } | |
1c4850e2 YZ |
8962 | path->nodes[level] = next; |
8963 | path->slots[level] = 0; | |
bd681513 | 8964 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
1c4850e2 YZ |
8965 | wc->level = level; |
8966 | if (wc->level == 1) | |
8967 | wc->reada_slot = 0; | |
8968 | return 0; | |
8969 | skip: | |
8970 | wc->refs[level - 1] = 0; | |
8971 | wc->flags[level - 1] = 0; | |
94fcca9f YZ |
8972 | if (wc->stage == DROP_REFERENCE) { |
8973 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
8974 | parent = path->nodes[level]->start; | |
8975 | } else { | |
4867268c | 8976 | ASSERT(root->root_key.objectid == |
94fcca9f | 8977 | btrfs_header_owner(path->nodes[level])); |
4867268c JB |
8978 | if (root->root_key.objectid != |
8979 | btrfs_header_owner(path->nodes[level])) { | |
8980 | btrfs_err(root->fs_info, | |
8981 | "mismatched block owner"); | |
8982 | ret = -EIO; | |
8983 | goto out_unlock; | |
8984 | } | |
94fcca9f YZ |
8985 | parent = 0; |
8986 | } | |
1c4850e2 | 8987 | |
78c52d9e JB |
8988 | /* |
8989 | * If we had a drop_progress we need to verify the refs are set | |
8990 | * as expected. If we find our ref then we know that from here | |
8991 | * on out everything should be correct, and we can clear the | |
8992 | * ->restarted flag. | |
8993 | */ | |
8994 | if (wc->restarted) { | |
8995 | ret = check_ref_exists(trans, root, bytenr, parent, | |
8996 | level - 1); | |
8997 | if (ret < 0) | |
8998 | goto out_unlock; | |
8999 | if (ret == 0) | |
9000 | goto no_delete; | |
9001 | ret = 0; | |
9002 | wc->restarted = 0; | |
9003 | } | |
9004 | ||
2cd86d30 QW |
9005 | /* |
9006 | * Reloc tree doesn't contribute to qgroup numbers, and we have | |
9007 | * already accounted them at merge time (replace_path), | |
9008 | * thus we could skip expensive subtree trace here. | |
9009 | */ | |
9010 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID && | |
9011 | need_account) { | |
deb40627 | 9012 | ret = btrfs_qgroup_trace_subtree(trans, next, |
33d1f05c | 9013 | generation, level - 1); |
1152651a | 9014 | if (ret) { |
0b246afa | 9015 | btrfs_err_rl(fs_info, |
5d163e0e JM |
9016 | "Error %d accounting shared subtree. Quota is out of sync, rescan required.", |
9017 | ret); | |
1152651a MF |
9018 | } |
9019 | } | |
aea6f028 JB |
9020 | |
9021 | /* | |
9022 | * We need to update the next key in our walk control so we can | |
9023 | * update the drop_progress key accordingly. We don't care if | |
9024 | * find_next_key doesn't find a key because that means we're at | |
9025 | * the end and are going to clean up now. | |
9026 | */ | |
9027 | wc->drop_level = level; | |
9028 | find_next_key(path, level, &wc->drop_progress); | |
9029 | ||
ffd4bb2a QW |
9030 | btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr, |
9031 | fs_info->nodesize, parent); | |
9032 | btrfs_init_tree_ref(&ref, level - 1, root->root_key.objectid); | |
9033 | ret = btrfs_free_extent(trans, &ref); | |
4867268c JB |
9034 | if (ret) |
9035 | goto out_unlock; | |
1c4850e2 | 9036 | } |
78c52d9e | 9037 | no_delete: |
4867268c JB |
9038 | *lookup_info = 1; |
9039 | ret = 1; | |
9040 | ||
9041 | out_unlock: | |
1c4850e2 YZ |
9042 | btrfs_tree_unlock(next); |
9043 | free_extent_buffer(next); | |
4867268c JB |
9044 | |
9045 | return ret; | |
1c4850e2 YZ |
9046 | } |
9047 | ||
2c47e605 | 9048 | /* |
2c016dc2 | 9049 | * helper to process tree block while walking up the tree. |
2c47e605 YZ |
9050 | * |
9051 | * when wc->stage == DROP_REFERENCE, this function drops | |
9052 | * reference count on the block. | |
9053 | * | |
9054 | * when wc->stage == UPDATE_BACKREF, this function changes | |
9055 | * wc->stage back to DROP_REFERENCE if we changed wc->stage | |
9056 | * to UPDATE_BACKREF previously while processing the block. | |
9057 | * | |
9058 | * NOTE: return value 1 means we should stop walking up. | |
9059 | */ | |
9060 | static noinline int walk_up_proc(struct btrfs_trans_handle *trans, | |
9061 | struct btrfs_root *root, | |
9062 | struct btrfs_path *path, | |
9063 | struct walk_control *wc) | |
9064 | { | |
0b246afa | 9065 | struct btrfs_fs_info *fs_info = root->fs_info; |
f0486c68 | 9066 | int ret; |
2c47e605 YZ |
9067 | int level = wc->level; |
9068 | struct extent_buffer *eb = path->nodes[level]; | |
9069 | u64 parent = 0; | |
9070 | ||
9071 | if (wc->stage == UPDATE_BACKREF) { | |
9072 | BUG_ON(wc->shared_level < level); | |
9073 | if (level < wc->shared_level) | |
9074 | goto out; | |
9075 | ||
2c47e605 YZ |
9076 | ret = find_next_key(path, level + 1, &wc->update_progress); |
9077 | if (ret > 0) | |
9078 | wc->update_ref = 0; | |
9079 | ||
9080 | wc->stage = DROP_REFERENCE; | |
9081 | wc->shared_level = -1; | |
9082 | path->slots[level] = 0; | |
9083 | ||
9084 | /* | |
9085 | * check reference count again if the block isn't locked. | |
9086 | * we should start walking down the tree again if reference | |
9087 | * count is one. | |
9088 | */ | |
9089 | if (!path->locks[level]) { | |
9090 | BUG_ON(level == 0); | |
9091 | btrfs_tree_lock(eb); | |
8bead258 | 9092 | btrfs_set_lock_blocking_write(eb); |
bd681513 | 9093 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 | 9094 | |
2ff7e61e | 9095 | ret = btrfs_lookup_extent_info(trans, fs_info, |
3173a18f | 9096 | eb->start, level, 1, |
2c47e605 YZ |
9097 | &wc->refs[level], |
9098 | &wc->flags[level]); | |
79787eaa JM |
9099 | if (ret < 0) { |
9100 | btrfs_tree_unlock_rw(eb, path->locks[level]); | |
3268a246 | 9101 | path->locks[level] = 0; |
79787eaa JM |
9102 | return ret; |
9103 | } | |
2c47e605 YZ |
9104 | BUG_ON(wc->refs[level] == 0); |
9105 | if (wc->refs[level] == 1) { | |
bd681513 | 9106 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
3268a246 | 9107 | path->locks[level] = 0; |
2c47e605 YZ |
9108 | return 1; |
9109 | } | |
f82d02d9 | 9110 | } |
2c47e605 | 9111 | } |
f82d02d9 | 9112 | |
2c47e605 YZ |
9113 | /* wc->stage == DROP_REFERENCE */ |
9114 | BUG_ON(wc->refs[level] > 1 && !path->locks[level]); | |
5d4f98a2 | 9115 | |
2c47e605 YZ |
9116 | if (wc->refs[level] == 1) { |
9117 | if (level == 0) { | |
9118 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
e339a6b0 | 9119 | ret = btrfs_dec_ref(trans, root, eb, 1); |
2c47e605 | 9120 | else |
e339a6b0 | 9121 | ret = btrfs_dec_ref(trans, root, eb, 0); |
79787eaa | 9122 | BUG_ON(ret); /* -ENOMEM */ |
c4140cbf QW |
9123 | if (is_fstree(root->root_key.objectid)) { |
9124 | ret = btrfs_qgroup_trace_leaf_items(trans, eb); | |
9125 | if (ret) { | |
9126 | btrfs_err_rl(fs_info, | |
9127 | "error %d accounting leaf items, quota is out of sync, rescan required", | |
5d163e0e | 9128 | ret); |
c4140cbf | 9129 | } |
1152651a | 9130 | } |
2c47e605 | 9131 | } |
6a884d7d | 9132 | /* make block locked assertion in btrfs_clean_tree_block happy */ |
2c47e605 YZ |
9133 | if (!path->locks[level] && |
9134 | btrfs_header_generation(eb) == trans->transid) { | |
9135 | btrfs_tree_lock(eb); | |
8bead258 | 9136 | btrfs_set_lock_blocking_write(eb); |
bd681513 | 9137 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 | 9138 | } |
6a884d7d | 9139 | btrfs_clean_tree_block(eb); |
2c47e605 YZ |
9140 | } |
9141 | ||
9142 | if (eb == root->node) { | |
9143 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
9144 | parent = eb->start; | |
65c6e82b QW |
9145 | else if (root->root_key.objectid != btrfs_header_owner(eb)) |
9146 | goto owner_mismatch; | |
2c47e605 YZ |
9147 | } else { |
9148 | if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
9149 | parent = path->nodes[level + 1]->start; | |
65c6e82b QW |
9150 | else if (root->root_key.objectid != |
9151 | btrfs_header_owner(path->nodes[level + 1])) | |
9152 | goto owner_mismatch; | |
f82d02d9 | 9153 | } |
f82d02d9 | 9154 | |
5581a51a | 9155 | btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1); |
2c47e605 YZ |
9156 | out: |
9157 | wc->refs[level] = 0; | |
9158 | wc->flags[level] = 0; | |
f0486c68 | 9159 | return 0; |
65c6e82b QW |
9160 | |
9161 | owner_mismatch: | |
9162 | btrfs_err_rl(fs_info, "unexpected tree owner, have %llu expect %llu", | |
9163 | btrfs_header_owner(eb), root->root_key.objectid); | |
9164 | return -EUCLEAN; | |
2c47e605 YZ |
9165 | } |
9166 | ||
9167 | static noinline int walk_down_tree(struct btrfs_trans_handle *trans, | |
9168 | struct btrfs_root *root, | |
9169 | struct btrfs_path *path, | |
9170 | struct walk_control *wc) | |
9171 | { | |
2c47e605 | 9172 | int level = wc->level; |
94fcca9f | 9173 | int lookup_info = 1; |
2c47e605 YZ |
9174 | int ret; |
9175 | ||
9176 | while (level >= 0) { | |
94fcca9f | 9177 | ret = walk_down_proc(trans, root, path, wc, lookup_info); |
2c47e605 YZ |
9178 | if (ret > 0) |
9179 | break; | |
9180 | ||
9181 | if (level == 0) | |
9182 | break; | |
9183 | ||
7a7965f8 YZ |
9184 | if (path->slots[level] >= |
9185 | btrfs_header_nritems(path->nodes[level])) | |
9186 | break; | |
9187 | ||
94fcca9f | 9188 | ret = do_walk_down(trans, root, path, wc, &lookup_info); |
1c4850e2 YZ |
9189 | if (ret > 0) { |
9190 | path->slots[level]++; | |
9191 | continue; | |
90d2c51d MX |
9192 | } else if (ret < 0) |
9193 | return ret; | |
1c4850e2 | 9194 | level = wc->level; |
f82d02d9 | 9195 | } |
f82d02d9 YZ |
9196 | return 0; |
9197 | } | |
9198 | ||
d397712b | 9199 | static noinline int walk_up_tree(struct btrfs_trans_handle *trans, |
98ed5174 | 9200 | struct btrfs_root *root, |
f82d02d9 | 9201 | struct btrfs_path *path, |
2c47e605 | 9202 | struct walk_control *wc, int max_level) |
20524f02 | 9203 | { |
2c47e605 | 9204 | int level = wc->level; |
20524f02 | 9205 | int ret; |
9f3a7427 | 9206 | |
2c47e605 YZ |
9207 | path->slots[level] = btrfs_header_nritems(path->nodes[level]); |
9208 | while (level < max_level && path->nodes[level]) { | |
9209 | wc->level = level; | |
9210 | if (path->slots[level] + 1 < | |
9211 | btrfs_header_nritems(path->nodes[level])) { | |
9212 | path->slots[level]++; | |
20524f02 CM |
9213 | return 0; |
9214 | } else { | |
2c47e605 YZ |
9215 | ret = walk_up_proc(trans, root, path, wc); |
9216 | if (ret > 0) | |
9217 | return 0; | |
65c6e82b QW |
9218 | if (ret < 0) |
9219 | return ret; | |
bd56b302 | 9220 | |
2c47e605 | 9221 | if (path->locks[level]) { |
bd681513 CM |
9222 | btrfs_tree_unlock_rw(path->nodes[level], |
9223 | path->locks[level]); | |
2c47e605 | 9224 | path->locks[level] = 0; |
f82d02d9 | 9225 | } |
2c47e605 YZ |
9226 | free_extent_buffer(path->nodes[level]); |
9227 | path->nodes[level] = NULL; | |
9228 | level++; | |
20524f02 CM |
9229 | } |
9230 | } | |
9231 | return 1; | |
9232 | } | |
9233 | ||
9aca1d51 | 9234 | /* |
2c47e605 YZ |
9235 | * drop a subvolume tree. |
9236 | * | |
9237 | * this function traverses the tree freeing any blocks that only | |
9238 | * referenced by the tree. | |
9239 | * | |
9240 | * when a shared tree block is found. this function decreases its | |
9241 | * reference count by one. if update_ref is true, this function | |
9242 | * also make sure backrefs for the shared block and all lower level | |
9243 | * blocks are properly updated. | |
9d1a2a3a DS |
9244 | * |
9245 | * If called with for_reloc == 0, may exit early with -EAGAIN | |
9aca1d51 | 9246 | */ |
2c536799 | 9247 | int btrfs_drop_snapshot(struct btrfs_root *root, |
66d7e7f0 AJ |
9248 | struct btrfs_block_rsv *block_rsv, int update_ref, |
9249 | int for_reloc) | |
20524f02 | 9250 | { |
ab8d0fc4 | 9251 | struct btrfs_fs_info *fs_info = root->fs_info; |
5caf2a00 | 9252 | struct btrfs_path *path; |
2c47e605 | 9253 | struct btrfs_trans_handle *trans; |
ab8d0fc4 | 9254 | struct btrfs_root *tree_root = fs_info->tree_root; |
9f3a7427 | 9255 | struct btrfs_root_item *root_item = &root->root_item; |
2c47e605 YZ |
9256 | struct walk_control *wc; |
9257 | struct btrfs_key key; | |
9258 | int err = 0; | |
9259 | int ret; | |
9260 | int level; | |
d29a9f62 | 9261 | bool root_dropped = false; |
20524f02 | 9262 | |
4fd786e6 | 9263 | btrfs_debug(fs_info, "Drop subvolume %llu", root->root_key.objectid); |
1152651a | 9264 | |
5caf2a00 | 9265 | path = btrfs_alloc_path(); |
cb1b69f4 TI |
9266 | if (!path) { |
9267 | err = -ENOMEM; | |
9268 | goto out; | |
9269 | } | |
20524f02 | 9270 | |
2c47e605 | 9271 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
38a1a919 MF |
9272 | if (!wc) { |
9273 | btrfs_free_path(path); | |
cb1b69f4 TI |
9274 | err = -ENOMEM; |
9275 | goto out; | |
38a1a919 | 9276 | } |
2c47e605 | 9277 | |
a22285a6 | 9278 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
9279 | if (IS_ERR(trans)) { |
9280 | err = PTR_ERR(trans); | |
9281 | goto out_free; | |
9282 | } | |
98d5dc13 | 9283 | |
0568e82d JB |
9284 | err = btrfs_run_delayed_items(trans); |
9285 | if (err) | |
9286 | goto out_end_trans; | |
9287 | ||
3fd0a558 YZ |
9288 | if (block_rsv) |
9289 | trans->block_rsv = block_rsv; | |
2c47e605 | 9290 | |
83354f07 JB |
9291 | /* |
9292 | * This will help us catch people modifying the fs tree while we're | |
9293 | * dropping it. It is unsafe to mess with the fs tree while it's being | |
9294 | * dropped as we unlock the root node and parent nodes as we walk down | |
9295 | * the tree, assuming nothing will change. If something does change | |
9296 | * then we'll have stale information and drop references to blocks we've | |
9297 | * already dropped. | |
9298 | */ | |
9299 | set_bit(BTRFS_ROOT_DELETING, &root->state); | |
9f3a7427 | 9300 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { |
2c47e605 | 9301 | level = btrfs_header_level(root->node); |
5d4f98a2 | 9302 | path->nodes[level] = btrfs_lock_root_node(root); |
8bead258 | 9303 | btrfs_set_lock_blocking_write(path->nodes[level]); |
9f3a7427 | 9304 | path->slots[level] = 0; |
bd681513 | 9305 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
9306 | memset(&wc->update_progress, 0, |
9307 | sizeof(wc->update_progress)); | |
9f3a7427 | 9308 | } else { |
9f3a7427 | 9309 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); |
2c47e605 YZ |
9310 | memcpy(&wc->update_progress, &key, |
9311 | sizeof(wc->update_progress)); | |
9312 | ||
6702ed49 | 9313 | level = root_item->drop_level; |
2c47e605 | 9314 | BUG_ON(level == 0); |
6702ed49 | 9315 | path->lowest_level = level; |
2c47e605 YZ |
9316 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
9317 | path->lowest_level = 0; | |
9318 | if (ret < 0) { | |
9319 | err = ret; | |
79787eaa | 9320 | goto out_end_trans; |
9f3a7427 | 9321 | } |
1c4850e2 | 9322 | WARN_ON(ret > 0); |
2c47e605 | 9323 | |
7d9eb12c CM |
9324 | /* |
9325 | * unlock our path, this is safe because only this | |
9326 | * function is allowed to delete this snapshot | |
9327 | */ | |
5d4f98a2 | 9328 | btrfs_unlock_up_safe(path, 0); |
2c47e605 YZ |
9329 | |
9330 | level = btrfs_header_level(root->node); | |
9331 | while (1) { | |
9332 | btrfs_tree_lock(path->nodes[level]); | |
8bead258 | 9333 | btrfs_set_lock_blocking_write(path->nodes[level]); |
fec386ac | 9334 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 | 9335 | |
2ff7e61e | 9336 | ret = btrfs_lookup_extent_info(trans, fs_info, |
2c47e605 | 9337 | path->nodes[level]->start, |
3173a18f | 9338 | level, 1, &wc->refs[level], |
2c47e605 | 9339 | &wc->flags[level]); |
79787eaa JM |
9340 | if (ret < 0) { |
9341 | err = ret; | |
9342 | goto out_end_trans; | |
9343 | } | |
2c47e605 YZ |
9344 | BUG_ON(wc->refs[level] == 0); |
9345 | ||
9346 | if (level == root_item->drop_level) | |
9347 | break; | |
9348 | ||
9349 | btrfs_tree_unlock(path->nodes[level]); | |
fec386ac | 9350 | path->locks[level] = 0; |
2c47e605 YZ |
9351 | WARN_ON(wc->refs[level] != 1); |
9352 | level--; | |
9353 | } | |
9f3a7427 | 9354 | } |
2c47e605 | 9355 | |
78c52d9e | 9356 | wc->restarted = test_bit(BTRFS_ROOT_DEAD_TREE, &root->state); |
2c47e605 YZ |
9357 | wc->level = level; |
9358 | wc->shared_level = -1; | |
9359 | wc->stage = DROP_REFERENCE; | |
9360 | wc->update_ref = update_ref; | |
9361 | wc->keep_locks = 0; | |
0b246afa | 9362 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info); |
2c47e605 | 9363 | |
d397712b | 9364 | while (1) { |
9d1a2a3a | 9365 | |
2c47e605 YZ |
9366 | ret = walk_down_tree(trans, root, path, wc); |
9367 | if (ret < 0) { | |
9368 | err = ret; | |
20524f02 | 9369 | break; |
2c47e605 | 9370 | } |
9aca1d51 | 9371 | |
2c47e605 YZ |
9372 | ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL); |
9373 | if (ret < 0) { | |
9374 | err = ret; | |
20524f02 | 9375 | break; |
2c47e605 YZ |
9376 | } |
9377 | ||
9378 | if (ret > 0) { | |
9379 | BUG_ON(wc->stage != DROP_REFERENCE); | |
e7a84565 CM |
9380 | break; |
9381 | } | |
2c47e605 YZ |
9382 | |
9383 | if (wc->stage == DROP_REFERENCE) { | |
aea6f028 JB |
9384 | wc->drop_level = wc->level; |
9385 | btrfs_node_key_to_cpu(path->nodes[wc->drop_level], | |
9386 | &wc->drop_progress, | |
9387 | path->slots[wc->drop_level]); | |
9388 | } | |
9389 | btrfs_cpu_key_to_disk(&root_item->drop_progress, | |
9390 | &wc->drop_progress); | |
9391 | root_item->drop_level = wc->drop_level; | |
2c47e605 YZ |
9392 | |
9393 | BUG_ON(wc->level == 0); | |
3a45bb20 | 9394 | if (btrfs_should_end_transaction(trans) || |
2ff7e61e | 9395 | (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) { |
2c47e605 YZ |
9396 | ret = btrfs_update_root(trans, tree_root, |
9397 | &root->root_key, | |
9398 | root_item); | |
79787eaa | 9399 | if (ret) { |
66642832 | 9400 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9401 | err = ret; |
9402 | goto out_end_trans; | |
9403 | } | |
2c47e605 | 9404 | |
3a45bb20 | 9405 | btrfs_end_transaction_throttle(trans); |
2ff7e61e | 9406 | if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) { |
ab8d0fc4 JM |
9407 | btrfs_debug(fs_info, |
9408 | "drop snapshot early exit"); | |
3c8f2422 JB |
9409 | err = -EAGAIN; |
9410 | goto out_free; | |
9411 | } | |
9412 | ||
a22285a6 | 9413 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
9414 | if (IS_ERR(trans)) { |
9415 | err = PTR_ERR(trans); | |
9416 | goto out_free; | |
9417 | } | |
3fd0a558 YZ |
9418 | if (block_rsv) |
9419 | trans->block_rsv = block_rsv; | |
c3e69d58 | 9420 | } |
20524f02 | 9421 | } |
b3b4aa74 | 9422 | btrfs_release_path(path); |
79787eaa JM |
9423 | if (err) |
9424 | goto out_end_trans; | |
2c47e605 | 9425 | |
ab9ce7d4 | 9426 | ret = btrfs_del_root(trans, &root->root_key); |
79787eaa | 9427 | if (ret) { |
66642832 | 9428 | btrfs_abort_transaction(trans, ret); |
e19182c0 | 9429 | err = ret; |
79787eaa JM |
9430 | goto out_end_trans; |
9431 | } | |
2c47e605 | 9432 | |
76dda93c | 9433 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { |
cb517eab MX |
9434 | ret = btrfs_find_root(tree_root, &root->root_key, path, |
9435 | NULL, NULL); | |
79787eaa | 9436 | if (ret < 0) { |
66642832 | 9437 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9438 | err = ret; |
9439 | goto out_end_trans; | |
9440 | } else if (ret > 0) { | |
84cd948c JB |
9441 | /* if we fail to delete the orphan item this time |
9442 | * around, it'll get picked up the next time. | |
9443 | * | |
9444 | * The most common failure here is just -ENOENT. | |
9445 | */ | |
9446 | btrfs_del_orphan_item(trans, tree_root, | |
9447 | root->root_key.objectid); | |
76dda93c YZ |
9448 | } |
9449 | } | |
9450 | ||
27cdeb70 | 9451 | if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) { |
2b9dbef2 | 9452 | btrfs_add_dropped_root(trans, root); |
76dda93c YZ |
9453 | } else { |
9454 | free_extent_buffer(root->node); | |
9455 | free_extent_buffer(root->commit_root); | |
b0feb9d9 | 9456 | btrfs_put_fs_root(root); |
76dda93c | 9457 | } |
d29a9f62 | 9458 | root_dropped = true; |
79787eaa | 9459 | out_end_trans: |
3a45bb20 | 9460 | btrfs_end_transaction_throttle(trans); |
79787eaa | 9461 | out_free: |
2c47e605 | 9462 | kfree(wc); |
5caf2a00 | 9463 | btrfs_free_path(path); |
cb1b69f4 | 9464 | out: |
d29a9f62 JB |
9465 | /* |
9466 | * So if we need to stop dropping the snapshot for whatever reason we | |
9467 | * need to make sure to add it back to the dead root list so that we | |
9468 | * keep trying to do the work later. This also cleans up roots if we | |
9469 | * don't have it in the radix (like when we recover after a power fail | |
9470 | * or unmount) so we don't leak memory. | |
9471 | */ | |
897ca819 | 9472 | if (!for_reloc && !root_dropped) |
d29a9f62 | 9473 | btrfs_add_dead_root(root); |
90515e7f | 9474 | if (err && err != -EAGAIN) |
ab8d0fc4 | 9475 | btrfs_handle_fs_error(fs_info, err, NULL); |
2c536799 | 9476 | return err; |
20524f02 | 9477 | } |
9078a3e1 | 9478 | |
2c47e605 YZ |
9479 | /* |
9480 | * drop subtree rooted at tree block 'node'. | |
9481 | * | |
9482 | * NOTE: this function will unlock and release tree block 'node' | |
66d7e7f0 | 9483 | * only used by relocation code |
2c47e605 | 9484 | */ |
f82d02d9 YZ |
9485 | int btrfs_drop_subtree(struct btrfs_trans_handle *trans, |
9486 | struct btrfs_root *root, | |
9487 | struct extent_buffer *node, | |
9488 | struct extent_buffer *parent) | |
9489 | { | |
0b246afa | 9490 | struct btrfs_fs_info *fs_info = root->fs_info; |
f82d02d9 | 9491 | struct btrfs_path *path; |
2c47e605 | 9492 | struct walk_control *wc; |
f82d02d9 YZ |
9493 | int level; |
9494 | int parent_level; | |
9495 | int ret = 0; | |
9496 | int wret; | |
9497 | ||
2c47e605 YZ |
9498 | BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); |
9499 | ||
f82d02d9 | 9500 | path = btrfs_alloc_path(); |
db5b493a TI |
9501 | if (!path) |
9502 | return -ENOMEM; | |
f82d02d9 | 9503 | |
2c47e605 | 9504 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
db5b493a TI |
9505 | if (!wc) { |
9506 | btrfs_free_path(path); | |
9507 | return -ENOMEM; | |
9508 | } | |
2c47e605 | 9509 | |
b9447ef8 | 9510 | btrfs_assert_tree_locked(parent); |
f82d02d9 YZ |
9511 | parent_level = btrfs_header_level(parent); |
9512 | extent_buffer_get(parent); | |
9513 | path->nodes[parent_level] = parent; | |
9514 | path->slots[parent_level] = btrfs_header_nritems(parent); | |
9515 | ||
b9447ef8 | 9516 | btrfs_assert_tree_locked(node); |
f82d02d9 | 9517 | level = btrfs_header_level(node); |
f82d02d9 YZ |
9518 | path->nodes[level] = node; |
9519 | path->slots[level] = 0; | |
bd681513 | 9520 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
9521 | |
9522 | wc->refs[parent_level] = 1; | |
9523 | wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
9524 | wc->level = level; | |
9525 | wc->shared_level = -1; | |
9526 | wc->stage = DROP_REFERENCE; | |
9527 | wc->update_ref = 0; | |
9528 | wc->keep_locks = 1; | |
0b246afa | 9529 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info); |
f82d02d9 YZ |
9530 | |
9531 | while (1) { | |
2c47e605 YZ |
9532 | wret = walk_down_tree(trans, root, path, wc); |
9533 | if (wret < 0) { | |
f82d02d9 | 9534 | ret = wret; |
f82d02d9 | 9535 | break; |
2c47e605 | 9536 | } |
f82d02d9 | 9537 | |
2c47e605 | 9538 | wret = walk_up_tree(trans, root, path, wc, parent_level); |
f82d02d9 YZ |
9539 | if (wret < 0) |
9540 | ret = wret; | |
9541 | if (wret != 0) | |
9542 | break; | |
9543 | } | |
9544 | ||
2c47e605 | 9545 | kfree(wc); |
f82d02d9 YZ |
9546 | btrfs_free_path(path); |
9547 | return ret; | |
9548 | } | |
9549 | ||
6202df69 | 9550 | static u64 update_block_group_flags(struct btrfs_fs_info *fs_info, u64 flags) |
ec44a35c CM |
9551 | { |
9552 | u64 num_devices; | |
fc67c450 | 9553 | u64 stripped; |
e4d8ec0f | 9554 | |
fc67c450 ID |
9555 | /* |
9556 | * if restripe for this chunk_type is on pick target profile and | |
9557 | * return, otherwise do the usual balance | |
9558 | */ | |
6202df69 | 9559 | stripped = get_restripe_target(fs_info, flags); |
fc67c450 ID |
9560 | if (stripped) |
9561 | return extended_to_chunk(stripped); | |
e4d8ec0f | 9562 | |
6202df69 | 9563 | num_devices = fs_info->fs_devices->rw_devices; |
cd02dca5 | 9564 | |
fc67c450 | 9565 | stripped = BTRFS_BLOCK_GROUP_RAID0 | |
53b381b3 | 9566 | BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 | |
fc67c450 ID |
9567 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10; |
9568 | ||
ec44a35c CM |
9569 | if (num_devices == 1) { |
9570 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
9571 | stripped = flags & ~stripped; | |
9572 | ||
9573 | /* turn raid0 into single device chunks */ | |
9574 | if (flags & BTRFS_BLOCK_GROUP_RAID0) | |
9575 | return stripped; | |
9576 | ||
9577 | /* turn mirroring into duplication */ | |
9578 | if (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
9579 | BTRFS_BLOCK_GROUP_RAID10)) | |
9580 | return stripped | BTRFS_BLOCK_GROUP_DUP; | |
ec44a35c CM |
9581 | } else { |
9582 | /* they already had raid on here, just return */ | |
ec44a35c CM |
9583 | if (flags & stripped) |
9584 | return flags; | |
9585 | ||
9586 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
9587 | stripped = flags & ~stripped; | |
9588 | ||
9589 | /* switch duplicated blocks with raid1 */ | |
9590 | if (flags & BTRFS_BLOCK_GROUP_DUP) | |
9591 | return stripped | BTRFS_BLOCK_GROUP_RAID1; | |
9592 | ||
e3176ca2 | 9593 | /* this is drive concat, leave it alone */ |
ec44a35c | 9594 | } |
e3176ca2 | 9595 | |
ec44a35c CM |
9596 | return flags; |
9597 | } | |
9598 | ||
868f401a | 9599 | static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force) |
0ef3e66b | 9600 | { |
f0486c68 YZ |
9601 | struct btrfs_space_info *sinfo = cache->space_info; |
9602 | u64 num_bytes; | |
3ece54e5 | 9603 | u64 sinfo_used; |
199c36ea | 9604 | u64 min_allocable_bytes; |
f0486c68 | 9605 | int ret = -ENOSPC; |
0ef3e66b | 9606 | |
199c36ea MX |
9607 | /* |
9608 | * We need some metadata space and system metadata space for | |
9609 | * allocating chunks in some corner cases until we force to set | |
9610 | * it to be readonly. | |
9611 | */ | |
9612 | if ((sinfo->flags & | |
9613 | (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) && | |
9614 | !force) | |
ee22184b | 9615 | min_allocable_bytes = SZ_1M; |
199c36ea MX |
9616 | else |
9617 | min_allocable_bytes = 0; | |
9618 | ||
f0486c68 YZ |
9619 | spin_lock(&sinfo->lock); |
9620 | spin_lock(&cache->lock); | |
61cfea9b W |
9621 | |
9622 | if (cache->ro) { | |
868f401a | 9623 | cache->ro++; |
61cfea9b W |
9624 | ret = 0; |
9625 | goto out; | |
9626 | } | |
9627 | ||
f0486c68 YZ |
9628 | num_bytes = cache->key.offset - cache->reserved - cache->pinned - |
9629 | cache->bytes_super - btrfs_block_group_used(&cache->item); | |
3ece54e5 | 9630 | sinfo_used = btrfs_space_info_used(sinfo, true); |
f0486c68 | 9631 | |
3ece54e5 QW |
9632 | if (sinfo_used + num_bytes + min_allocable_bytes <= |
9633 | sinfo->total_bytes) { | |
f0486c68 | 9634 | sinfo->bytes_readonly += num_bytes; |
868f401a | 9635 | cache->ro++; |
633c0aad | 9636 | list_add_tail(&cache->ro_list, &sinfo->ro_bgs); |
f0486c68 YZ |
9637 | ret = 0; |
9638 | } | |
61cfea9b | 9639 | out: |
f0486c68 YZ |
9640 | spin_unlock(&cache->lock); |
9641 | spin_unlock(&sinfo->lock); | |
3ece54e5 QW |
9642 | if (ret == -ENOSPC && btrfs_test_opt(cache->fs_info, ENOSPC_DEBUG)) { |
9643 | btrfs_info(cache->fs_info, | |
9644 | "unable to make block group %llu ro", | |
9645 | cache->key.objectid); | |
9646 | btrfs_info(cache->fs_info, | |
9647 | "sinfo_used=%llu bg_num_bytes=%llu min_allocable=%llu", | |
9648 | sinfo_used, num_bytes, min_allocable_bytes); | |
9649 | dump_space_info(cache->fs_info, cache->space_info, 0, 0); | |
9650 | } | |
f0486c68 YZ |
9651 | return ret; |
9652 | } | |
7d9eb12c | 9653 | |
c83488af | 9654 | int btrfs_inc_block_group_ro(struct btrfs_block_group_cache *cache) |
c286ac48 | 9655 | |
f0486c68 | 9656 | { |
c83488af | 9657 | struct btrfs_fs_info *fs_info = cache->fs_info; |
f0486c68 YZ |
9658 | struct btrfs_trans_handle *trans; |
9659 | u64 alloc_flags; | |
9660 | int ret; | |
7d9eb12c | 9661 | |
1bbc621e | 9662 | again: |
5e00f193 | 9663 | trans = btrfs_join_transaction(fs_info->extent_root); |
79787eaa JM |
9664 | if (IS_ERR(trans)) |
9665 | return PTR_ERR(trans); | |
5d4f98a2 | 9666 | |
1bbc621e CM |
9667 | /* |
9668 | * we're not allowed to set block groups readonly after the dirty | |
9669 | * block groups cache has started writing. If it already started, | |
9670 | * back off and let this transaction commit | |
9671 | */ | |
0b246afa | 9672 | mutex_lock(&fs_info->ro_block_group_mutex); |
3204d33c | 9673 | if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) { |
1bbc621e CM |
9674 | u64 transid = trans->transid; |
9675 | ||
0b246afa | 9676 | mutex_unlock(&fs_info->ro_block_group_mutex); |
3a45bb20 | 9677 | btrfs_end_transaction(trans); |
1bbc621e | 9678 | |
2ff7e61e | 9679 | ret = btrfs_wait_for_commit(fs_info, transid); |
1bbc621e CM |
9680 | if (ret) |
9681 | return ret; | |
9682 | goto again; | |
9683 | } | |
9684 | ||
153c35b6 CM |
9685 | /* |
9686 | * if we are changing raid levels, try to allocate a corresponding | |
9687 | * block group with the new raid level. | |
9688 | */ | |
0b246afa | 9689 | alloc_flags = update_block_group_flags(fs_info, cache->flags); |
153c35b6 | 9690 | if (alloc_flags != cache->flags) { |
01458828 | 9691 | ret = do_chunk_alloc(trans, alloc_flags, |
153c35b6 CM |
9692 | CHUNK_ALLOC_FORCE); |
9693 | /* | |
9694 | * ENOSPC is allowed here, we may have enough space | |
9695 | * already allocated at the new raid level to | |
9696 | * carry on | |
9697 | */ | |
9698 | if (ret == -ENOSPC) | |
9699 | ret = 0; | |
9700 | if (ret < 0) | |
9701 | goto out; | |
9702 | } | |
1bbc621e | 9703 | |
868f401a | 9704 | ret = inc_block_group_ro(cache, 0); |
f0486c68 YZ |
9705 | if (!ret) |
9706 | goto out; | |
2ff7e61e | 9707 | alloc_flags = get_alloc_profile(fs_info, cache->space_info->flags); |
01458828 | 9708 | ret = do_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE); |
f0486c68 YZ |
9709 | if (ret < 0) |
9710 | goto out; | |
868f401a | 9711 | ret = inc_block_group_ro(cache, 0); |
f0486c68 | 9712 | out: |
2f081088 | 9713 | if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) { |
0b246afa | 9714 | alloc_flags = update_block_group_flags(fs_info, cache->flags); |
34441361 | 9715 | mutex_lock(&fs_info->chunk_mutex); |
451a2c13 | 9716 | check_system_chunk(trans, alloc_flags); |
34441361 | 9717 | mutex_unlock(&fs_info->chunk_mutex); |
2f081088 | 9718 | } |
0b246afa | 9719 | mutex_unlock(&fs_info->ro_block_group_mutex); |
2f081088 | 9720 | |
3a45bb20 | 9721 | btrfs_end_transaction(trans); |
f0486c68 YZ |
9722 | return ret; |
9723 | } | |
5d4f98a2 | 9724 | |
43a7e99d | 9725 | int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans, u64 type) |
c87f08ca | 9726 | { |
43a7e99d | 9727 | u64 alloc_flags = get_alloc_profile(trans->fs_info, type); |
2ff7e61e | 9728 | |
01458828 | 9729 | return do_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE); |
c87f08ca CM |
9730 | } |
9731 | ||
6d07bcec MX |
9732 | /* |
9733 | * helper to account the unused space of all the readonly block group in the | |
633c0aad | 9734 | * space_info. takes mirrors into account. |
6d07bcec | 9735 | */ |
633c0aad | 9736 | u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo) |
6d07bcec MX |
9737 | { |
9738 | struct btrfs_block_group_cache *block_group; | |
9739 | u64 free_bytes = 0; | |
9740 | int factor; | |
9741 | ||
01327610 | 9742 | /* It's df, we don't care if it's racy */ |
633c0aad JB |
9743 | if (list_empty(&sinfo->ro_bgs)) |
9744 | return 0; | |
9745 | ||
9746 | spin_lock(&sinfo->lock); | |
9747 | list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) { | |
6d07bcec MX |
9748 | spin_lock(&block_group->lock); |
9749 | ||
9750 | if (!block_group->ro) { | |
9751 | spin_unlock(&block_group->lock); | |
9752 | continue; | |
9753 | } | |
9754 | ||
46df06b8 | 9755 | factor = btrfs_bg_type_to_factor(block_group->flags); |
6d07bcec MX |
9756 | free_bytes += (block_group->key.offset - |
9757 | btrfs_block_group_used(&block_group->item)) * | |
9758 | factor; | |
9759 | ||
9760 | spin_unlock(&block_group->lock); | |
9761 | } | |
6d07bcec MX |
9762 | spin_unlock(&sinfo->lock); |
9763 | ||
9764 | return free_bytes; | |
9765 | } | |
9766 | ||
2ff7e61e | 9767 | void btrfs_dec_block_group_ro(struct btrfs_block_group_cache *cache) |
5d4f98a2 | 9768 | { |
f0486c68 YZ |
9769 | struct btrfs_space_info *sinfo = cache->space_info; |
9770 | u64 num_bytes; | |
9771 | ||
9772 | BUG_ON(!cache->ro); | |
9773 | ||
9774 | spin_lock(&sinfo->lock); | |
9775 | spin_lock(&cache->lock); | |
868f401a Z |
9776 | if (!--cache->ro) { |
9777 | num_bytes = cache->key.offset - cache->reserved - | |
9778 | cache->pinned - cache->bytes_super - | |
9779 | btrfs_block_group_used(&cache->item); | |
9780 | sinfo->bytes_readonly -= num_bytes; | |
9781 | list_del_init(&cache->ro_list); | |
9782 | } | |
f0486c68 YZ |
9783 | spin_unlock(&cache->lock); |
9784 | spin_unlock(&sinfo->lock); | |
5d4f98a2 YZ |
9785 | } |
9786 | ||
ba1bf481 | 9787 | /* |
52042d8e | 9788 | * Checks to see if it's even possible to relocate this block group. |
ba1bf481 JB |
9789 | * |
9790 | * @return - -1 if it's not a good idea to relocate this block group, 0 if its | |
9791 | * ok to go ahead and try. | |
9792 | */ | |
6bccf3ab | 9793 | int btrfs_can_relocate(struct btrfs_fs_info *fs_info, u64 bytenr) |
1a40e23b | 9794 | { |
ba1bf481 JB |
9795 | struct btrfs_block_group_cache *block_group; |
9796 | struct btrfs_space_info *space_info; | |
0b246afa | 9797 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
ba1bf481 | 9798 | struct btrfs_device *device; |
cdcb725c | 9799 | u64 min_free; |
6719db6a JB |
9800 | u64 dev_min = 1; |
9801 | u64 dev_nr = 0; | |
4a5e98f5 | 9802 | u64 target; |
0305bc27 | 9803 | int debug; |
cdcb725c | 9804 | int index; |
ba1bf481 JB |
9805 | int full = 0; |
9806 | int ret = 0; | |
1a40e23b | 9807 | |
0b246afa | 9808 | debug = btrfs_test_opt(fs_info, ENOSPC_DEBUG); |
0305bc27 | 9809 | |
0b246afa | 9810 | block_group = btrfs_lookup_block_group(fs_info, bytenr); |
1a40e23b | 9811 | |
ba1bf481 | 9812 | /* odd, couldn't find the block group, leave it alone */ |
0305bc27 QW |
9813 | if (!block_group) { |
9814 | if (debug) | |
0b246afa | 9815 | btrfs_warn(fs_info, |
0305bc27 QW |
9816 | "can't find block group for bytenr %llu", |
9817 | bytenr); | |
ba1bf481 | 9818 | return -1; |
0305bc27 | 9819 | } |
1a40e23b | 9820 | |
cdcb725c | 9821 | min_free = btrfs_block_group_used(&block_group->item); |
9822 | ||
ba1bf481 | 9823 | /* no bytes used, we're good */ |
cdcb725c | 9824 | if (!min_free) |
1a40e23b ZY |
9825 | goto out; |
9826 | ||
ba1bf481 JB |
9827 | space_info = block_group->space_info; |
9828 | spin_lock(&space_info->lock); | |
17d217fe | 9829 | |
ba1bf481 | 9830 | full = space_info->full; |
17d217fe | 9831 | |
ba1bf481 JB |
9832 | /* |
9833 | * if this is the last block group we have in this space, we can't | |
7ce618db CM |
9834 | * relocate it unless we're able to allocate a new chunk below. |
9835 | * | |
9836 | * Otherwise, we need to make sure we have room in the space to handle | |
9837 | * all of the extents from this block group. If we can, we're good | |
ba1bf481 | 9838 | */ |
7ce618db | 9839 | if ((space_info->total_bytes != block_group->key.offset) && |
4136135b LB |
9840 | (btrfs_space_info_used(space_info, false) + min_free < |
9841 | space_info->total_bytes)) { | |
ba1bf481 JB |
9842 | spin_unlock(&space_info->lock); |
9843 | goto out; | |
17d217fe | 9844 | } |
ba1bf481 | 9845 | spin_unlock(&space_info->lock); |
ea8c2819 | 9846 | |
ba1bf481 JB |
9847 | /* |
9848 | * ok we don't have enough space, but maybe we have free space on our | |
9849 | * devices to allocate new chunks for relocation, so loop through our | |
4a5e98f5 ID |
9850 | * alloc devices and guess if we have enough space. if this block |
9851 | * group is going to be restriped, run checks against the target | |
9852 | * profile instead of the current one. | |
ba1bf481 JB |
9853 | */ |
9854 | ret = -1; | |
ea8c2819 | 9855 | |
cdcb725c | 9856 | /* |
9857 | * index: | |
9858 | * 0: raid10 | |
9859 | * 1: raid1 | |
9860 | * 2: dup | |
9861 | * 3: raid0 | |
9862 | * 4: single | |
9863 | */ | |
0b246afa | 9864 | target = get_restripe_target(fs_info, block_group->flags); |
4a5e98f5 | 9865 | if (target) { |
3e72ee88 | 9866 | index = btrfs_bg_flags_to_raid_index(extended_to_chunk(target)); |
4a5e98f5 ID |
9867 | } else { |
9868 | /* | |
9869 | * this is just a balance, so if we were marked as full | |
9870 | * we know there is no space for a new chunk | |
9871 | */ | |
0305bc27 QW |
9872 | if (full) { |
9873 | if (debug) | |
0b246afa JM |
9874 | btrfs_warn(fs_info, |
9875 | "no space to alloc new chunk for block group %llu", | |
9876 | block_group->key.objectid); | |
4a5e98f5 | 9877 | goto out; |
0305bc27 | 9878 | } |
4a5e98f5 | 9879 | |
3e72ee88 | 9880 | index = btrfs_bg_flags_to_raid_index(block_group->flags); |
4a5e98f5 ID |
9881 | } |
9882 | ||
e6ec716f | 9883 | if (index == BTRFS_RAID_RAID10) { |
cdcb725c | 9884 | dev_min = 4; |
6719db6a JB |
9885 | /* Divide by 2 */ |
9886 | min_free >>= 1; | |
e6ec716f | 9887 | } else if (index == BTRFS_RAID_RAID1) { |
cdcb725c | 9888 | dev_min = 2; |
e6ec716f | 9889 | } else if (index == BTRFS_RAID_DUP) { |
6719db6a JB |
9890 | /* Multiply by 2 */ |
9891 | min_free <<= 1; | |
e6ec716f | 9892 | } else if (index == BTRFS_RAID_RAID0) { |
cdcb725c | 9893 | dev_min = fs_devices->rw_devices; |
47c5713f | 9894 | min_free = div64_u64(min_free, dev_min); |
cdcb725c | 9895 | } |
9896 | ||
0b246afa | 9897 | mutex_lock(&fs_info->chunk_mutex); |
ba1bf481 | 9898 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { |
7bfc837d | 9899 | u64 dev_offset; |
56bec294 | 9900 | |
ba1bf481 JB |
9901 | /* |
9902 | * check to make sure we can actually find a chunk with enough | |
9903 | * space to fit our block group in. | |
9904 | */ | |
63a212ab | 9905 | if (device->total_bytes > device->bytes_used + min_free && |
401e29c1 | 9906 | !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
60dfdf25 | 9907 | ret = find_free_dev_extent(device, min_free, |
7bfc837d | 9908 | &dev_offset, NULL); |
ba1bf481 | 9909 | if (!ret) |
cdcb725c | 9910 | dev_nr++; |
9911 | ||
9912 | if (dev_nr >= dev_min) | |
73e48b27 | 9913 | break; |
cdcb725c | 9914 | |
ba1bf481 | 9915 | ret = -1; |
725c8463 | 9916 | } |
edbd8d4e | 9917 | } |
0305bc27 | 9918 | if (debug && ret == -1) |
0b246afa JM |
9919 | btrfs_warn(fs_info, |
9920 | "no space to allocate a new chunk for block group %llu", | |
9921 | block_group->key.objectid); | |
9922 | mutex_unlock(&fs_info->chunk_mutex); | |
edbd8d4e | 9923 | out: |
ba1bf481 | 9924 | btrfs_put_block_group(block_group); |
edbd8d4e CM |
9925 | return ret; |
9926 | } | |
9927 | ||
6bccf3ab JM |
9928 | static int find_first_block_group(struct btrfs_fs_info *fs_info, |
9929 | struct btrfs_path *path, | |
9930 | struct btrfs_key *key) | |
0b86a832 | 9931 | { |
6bccf3ab | 9932 | struct btrfs_root *root = fs_info->extent_root; |
925baedd | 9933 | int ret = 0; |
0b86a832 CM |
9934 | struct btrfs_key found_key; |
9935 | struct extent_buffer *leaf; | |
514c7dca QW |
9936 | struct btrfs_block_group_item bg; |
9937 | u64 flags; | |
0b86a832 | 9938 | int slot; |
edbd8d4e | 9939 | |
0b86a832 CM |
9940 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); |
9941 | if (ret < 0) | |
925baedd CM |
9942 | goto out; |
9943 | ||
d397712b | 9944 | while (1) { |
0b86a832 | 9945 | slot = path->slots[0]; |
edbd8d4e | 9946 | leaf = path->nodes[0]; |
0b86a832 CM |
9947 | if (slot >= btrfs_header_nritems(leaf)) { |
9948 | ret = btrfs_next_leaf(root, path); | |
9949 | if (ret == 0) | |
9950 | continue; | |
9951 | if (ret < 0) | |
925baedd | 9952 | goto out; |
0b86a832 | 9953 | break; |
edbd8d4e | 9954 | } |
0b86a832 | 9955 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
edbd8d4e | 9956 | |
0b86a832 | 9957 | if (found_key.objectid >= key->objectid && |
925baedd | 9958 | found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) { |
6fb37b75 LB |
9959 | struct extent_map_tree *em_tree; |
9960 | struct extent_map *em; | |
9961 | ||
c8bf1b67 | 9962 | em_tree = &root->fs_info->mapping_tree; |
6fb37b75 LB |
9963 | read_lock(&em_tree->lock); |
9964 | em = lookup_extent_mapping(em_tree, found_key.objectid, | |
9965 | found_key.offset); | |
9966 | read_unlock(&em_tree->lock); | |
9967 | if (!em) { | |
0b246afa | 9968 | btrfs_err(fs_info, |
6fb37b75 LB |
9969 | "logical %llu len %llu found bg but no related chunk", |
9970 | found_key.objectid, found_key.offset); | |
9971 | ret = -ENOENT; | |
514c7dca QW |
9972 | } else if (em->start != found_key.objectid || |
9973 | em->len != found_key.offset) { | |
9974 | btrfs_err(fs_info, | |
9975 | "block group %llu len %llu mismatch with chunk %llu len %llu", | |
9976 | found_key.objectid, found_key.offset, | |
9977 | em->start, em->len); | |
9978 | ret = -EUCLEAN; | |
6fb37b75 | 9979 | } else { |
514c7dca QW |
9980 | read_extent_buffer(leaf, &bg, |
9981 | btrfs_item_ptr_offset(leaf, slot), | |
9982 | sizeof(bg)); | |
9983 | flags = btrfs_block_group_flags(&bg) & | |
9984 | BTRFS_BLOCK_GROUP_TYPE_MASK; | |
9985 | ||
9986 | if (flags != (em->map_lookup->type & | |
9987 | BTRFS_BLOCK_GROUP_TYPE_MASK)) { | |
9988 | btrfs_err(fs_info, | |
9989 | "block group %llu len %llu type flags 0x%llx mismatch with chunk type flags 0x%llx", | |
9990 | found_key.objectid, | |
9991 | found_key.offset, flags, | |
9992 | (BTRFS_BLOCK_GROUP_TYPE_MASK & | |
9993 | em->map_lookup->type)); | |
9994 | ret = -EUCLEAN; | |
9995 | } else { | |
9996 | ret = 0; | |
9997 | } | |
6fb37b75 | 9998 | } |
187ee58c | 9999 | free_extent_map(em); |
925baedd CM |
10000 | goto out; |
10001 | } | |
0b86a832 | 10002 | path->slots[0]++; |
edbd8d4e | 10003 | } |
925baedd | 10004 | out: |
0b86a832 | 10005 | return ret; |
edbd8d4e CM |
10006 | } |
10007 | ||
0af3d00b JB |
10008 | void btrfs_put_block_group_cache(struct btrfs_fs_info *info) |
10009 | { | |
10010 | struct btrfs_block_group_cache *block_group; | |
10011 | u64 last = 0; | |
10012 | ||
10013 | while (1) { | |
10014 | struct inode *inode; | |
10015 | ||
10016 | block_group = btrfs_lookup_first_block_group(info, last); | |
10017 | while (block_group) { | |
3aa7c7a3 | 10018 | wait_block_group_cache_done(block_group); |
0af3d00b JB |
10019 | spin_lock(&block_group->lock); |
10020 | if (block_group->iref) | |
10021 | break; | |
10022 | spin_unlock(&block_group->lock); | |
f87b7eb8 | 10023 | block_group = next_block_group(block_group); |
0af3d00b JB |
10024 | } |
10025 | if (!block_group) { | |
10026 | if (last == 0) | |
10027 | break; | |
10028 | last = 0; | |
10029 | continue; | |
10030 | } | |
10031 | ||
10032 | inode = block_group->inode; | |
10033 | block_group->iref = 0; | |
10034 | block_group->inode = NULL; | |
10035 | spin_unlock(&block_group->lock); | |
f3bca802 | 10036 | ASSERT(block_group->io_ctl.inode == NULL); |
0af3d00b JB |
10037 | iput(inode); |
10038 | last = block_group->key.objectid + block_group->key.offset; | |
10039 | btrfs_put_block_group(block_group); | |
10040 | } | |
10041 | } | |
10042 | ||
5cdd7db6 FM |
10043 | /* |
10044 | * Must be called only after stopping all workers, since we could have block | |
10045 | * group caching kthreads running, and therefore they could race with us if we | |
10046 | * freed the block groups before stopping them. | |
10047 | */ | |
1a40e23b ZY |
10048 | int btrfs_free_block_groups(struct btrfs_fs_info *info) |
10049 | { | |
10050 | struct btrfs_block_group_cache *block_group; | |
4184ea7f | 10051 | struct btrfs_space_info *space_info; |
11833d66 | 10052 | struct btrfs_caching_control *caching_ctl; |
1a40e23b ZY |
10053 | struct rb_node *n; |
10054 | ||
9e351cc8 | 10055 | down_write(&info->commit_root_sem); |
11833d66 YZ |
10056 | while (!list_empty(&info->caching_block_groups)) { |
10057 | caching_ctl = list_entry(info->caching_block_groups.next, | |
10058 | struct btrfs_caching_control, list); | |
10059 | list_del(&caching_ctl->list); | |
10060 | put_caching_control(caching_ctl); | |
10061 | } | |
9e351cc8 | 10062 | up_write(&info->commit_root_sem); |
11833d66 | 10063 | |
47ab2a6c JB |
10064 | spin_lock(&info->unused_bgs_lock); |
10065 | while (!list_empty(&info->unused_bgs)) { | |
10066 | block_group = list_first_entry(&info->unused_bgs, | |
10067 | struct btrfs_block_group_cache, | |
10068 | bg_list); | |
10069 | list_del_init(&block_group->bg_list); | |
10070 | btrfs_put_block_group(block_group); | |
10071 | } | |
10072 | spin_unlock(&info->unused_bgs_lock); | |
10073 | ||
1a40e23b ZY |
10074 | spin_lock(&info->block_group_cache_lock); |
10075 | while ((n = rb_last(&info->block_group_cache_tree)) != NULL) { | |
10076 | block_group = rb_entry(n, struct btrfs_block_group_cache, | |
10077 | cache_node); | |
1a40e23b ZY |
10078 | rb_erase(&block_group->cache_node, |
10079 | &info->block_group_cache_tree); | |
01eacb27 | 10080 | RB_CLEAR_NODE(&block_group->cache_node); |
d899e052 YZ |
10081 | spin_unlock(&info->block_group_cache_lock); |
10082 | ||
80eb234a | 10083 | down_write(&block_group->space_info->groups_sem); |
1a40e23b | 10084 | list_del(&block_group->list); |
80eb234a | 10085 | up_write(&block_group->space_info->groups_sem); |
d2fb3437 | 10086 | |
3c14874a JB |
10087 | /* |
10088 | * We haven't cached this block group, which means we could | |
10089 | * possibly have excluded extents on this block group. | |
10090 | */ | |
36cce922 JB |
10091 | if (block_group->cached == BTRFS_CACHE_NO || |
10092 | block_group->cached == BTRFS_CACHE_ERROR) | |
9e715da8 | 10093 | free_excluded_extents(block_group); |
3c14874a | 10094 | |
817d52f8 | 10095 | btrfs_remove_free_space_cache(block_group); |
5cdd7db6 | 10096 | ASSERT(block_group->cached != BTRFS_CACHE_STARTED); |
f3bca802 LB |
10097 | ASSERT(list_empty(&block_group->dirty_list)); |
10098 | ASSERT(list_empty(&block_group->io_list)); | |
10099 | ASSERT(list_empty(&block_group->bg_list)); | |
10100 | ASSERT(atomic_read(&block_group->count) == 1); | |
11dfe35a | 10101 | btrfs_put_block_group(block_group); |
d899e052 YZ |
10102 | |
10103 | spin_lock(&info->block_group_cache_lock); | |
1a40e23b ZY |
10104 | } |
10105 | spin_unlock(&info->block_group_cache_lock); | |
4184ea7f CM |
10106 | |
10107 | /* now that all the block groups are freed, go through and | |
10108 | * free all the space_info structs. This is only called during | |
10109 | * the final stages of unmount, and so we know nobody is | |
10110 | * using them. We call synchronize_rcu() once before we start, | |
10111 | * just to be on the safe side. | |
10112 | */ | |
10113 | synchronize_rcu(); | |
10114 | ||
8929ecfa YZ |
10115 | release_global_block_rsv(info); |
10116 | ||
67871254 | 10117 | while (!list_empty(&info->space_info)) { |
6ab0a202 JM |
10118 | int i; |
10119 | ||
4184ea7f CM |
10120 | space_info = list_entry(info->space_info.next, |
10121 | struct btrfs_space_info, | |
10122 | list); | |
d555b6c3 JB |
10123 | |
10124 | /* | |
10125 | * Do not hide this behind enospc_debug, this is actually | |
10126 | * important and indicates a real bug if this happens. | |
10127 | */ | |
10128 | if (WARN_ON(space_info->bytes_pinned > 0 || | |
b069e0c3 | 10129 | space_info->bytes_reserved > 0 || |
d555b6c3 | 10130 | space_info->bytes_may_use > 0)) |
ab8d0fc4 | 10131 | dump_space_info(info, space_info, 0, 0); |
4184ea7f | 10132 | list_del(&space_info->list); |
6ab0a202 JM |
10133 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { |
10134 | struct kobject *kobj; | |
c1895442 JM |
10135 | kobj = space_info->block_group_kobjs[i]; |
10136 | space_info->block_group_kobjs[i] = NULL; | |
10137 | if (kobj) { | |
6ab0a202 JM |
10138 | kobject_del(kobj); |
10139 | kobject_put(kobj); | |
10140 | } | |
10141 | } | |
10142 | kobject_del(&space_info->kobj); | |
10143 | kobject_put(&space_info->kobj); | |
4184ea7f | 10144 | } |
1a40e23b ZY |
10145 | return 0; |
10146 | } | |
10147 | ||
75cb379d JM |
10148 | /* link_block_group will queue up kobjects to add when we're reclaim-safe */ |
10149 | void btrfs_add_raid_kobjects(struct btrfs_fs_info *fs_info) | |
10150 | { | |
10151 | struct btrfs_space_info *space_info; | |
10152 | struct raid_kobject *rkobj; | |
10153 | LIST_HEAD(list); | |
75cb379d JM |
10154 | int ret = 0; |
10155 | ||
10156 | spin_lock(&fs_info->pending_raid_kobjs_lock); | |
10157 | list_splice_init(&fs_info->pending_raid_kobjs, &list); | |
10158 | spin_unlock(&fs_info->pending_raid_kobjs_lock); | |
10159 | ||
10160 | list_for_each_entry(rkobj, &list, list) { | |
10161 | space_info = __find_space_info(fs_info, rkobj->flags); | |
75cb379d JM |
10162 | |
10163 | ret = kobject_add(&rkobj->kobj, &space_info->kobj, | |
158da513 | 10164 | "%s", btrfs_bg_type_to_raid_name(rkobj->flags)); |
75cb379d JM |
10165 | if (ret) { |
10166 | kobject_put(&rkobj->kobj); | |
10167 | break; | |
10168 | } | |
10169 | } | |
10170 | if (ret) | |
10171 | btrfs_warn(fs_info, | |
10172 | "failed to add kobject for block cache, ignoring"); | |
10173 | } | |
10174 | ||
c434d21c | 10175 | static void link_block_group(struct btrfs_block_group_cache *cache) |
b742bb82 | 10176 | { |
c434d21c | 10177 | struct btrfs_space_info *space_info = cache->space_info; |
75cb379d | 10178 | struct btrfs_fs_info *fs_info = cache->fs_info; |
3e72ee88 | 10179 | int index = btrfs_bg_flags_to_raid_index(cache->flags); |
ed55b6ac | 10180 | bool first = false; |
b742bb82 YZ |
10181 | |
10182 | down_write(&space_info->groups_sem); | |
ed55b6ac JM |
10183 | if (list_empty(&space_info->block_groups[index])) |
10184 | first = true; | |
10185 | list_add_tail(&cache->list, &space_info->block_groups[index]); | |
10186 | up_write(&space_info->groups_sem); | |
10187 | ||
10188 | if (first) { | |
75cb379d JM |
10189 | struct raid_kobject *rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS); |
10190 | if (!rkobj) { | |
10191 | btrfs_warn(cache->fs_info, | |
10192 | "couldn't alloc memory for raid level kobject"); | |
10193 | return; | |
6ab0a202 | 10194 | } |
75cb379d JM |
10195 | rkobj->flags = cache->flags; |
10196 | kobject_init(&rkobj->kobj, &btrfs_raid_ktype); | |
10197 | ||
10198 | spin_lock(&fs_info->pending_raid_kobjs_lock); | |
10199 | list_add_tail(&rkobj->list, &fs_info->pending_raid_kobjs); | |
10200 | spin_unlock(&fs_info->pending_raid_kobjs_lock); | |
c1895442 | 10201 | space_info->block_group_kobjs[index] = &rkobj->kobj; |
6ab0a202 | 10202 | } |
b742bb82 YZ |
10203 | } |
10204 | ||
920e4a58 | 10205 | static struct btrfs_block_group_cache * |
2ff7e61e JM |
10206 | btrfs_create_block_group_cache(struct btrfs_fs_info *fs_info, |
10207 | u64 start, u64 size) | |
920e4a58 MX |
10208 | { |
10209 | struct btrfs_block_group_cache *cache; | |
10210 | ||
10211 | cache = kzalloc(sizeof(*cache), GFP_NOFS); | |
10212 | if (!cache) | |
10213 | return NULL; | |
10214 | ||
10215 | cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl), | |
10216 | GFP_NOFS); | |
10217 | if (!cache->free_space_ctl) { | |
10218 | kfree(cache); | |
10219 | return NULL; | |
10220 | } | |
10221 | ||
10222 | cache->key.objectid = start; | |
10223 | cache->key.offset = size; | |
10224 | cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; | |
10225 | ||
0b246afa | 10226 | cache->fs_info = fs_info; |
e4ff5fb5 | 10227 | cache->full_stripe_len = btrfs_full_stripe_len(fs_info, start); |
1e144fb8 OS |
10228 | set_free_space_tree_thresholds(cache); |
10229 | ||
920e4a58 MX |
10230 | atomic_set(&cache->count, 1); |
10231 | spin_lock_init(&cache->lock); | |
e570fd27 | 10232 | init_rwsem(&cache->data_rwsem); |
920e4a58 MX |
10233 | INIT_LIST_HEAD(&cache->list); |
10234 | INIT_LIST_HEAD(&cache->cluster_list); | |
47ab2a6c | 10235 | INIT_LIST_HEAD(&cache->bg_list); |
633c0aad | 10236 | INIT_LIST_HEAD(&cache->ro_list); |
ce93ec54 | 10237 | INIT_LIST_HEAD(&cache->dirty_list); |
c9dc4c65 | 10238 | INIT_LIST_HEAD(&cache->io_list); |
920e4a58 | 10239 | btrfs_init_free_space_ctl(cache); |
04216820 | 10240 | atomic_set(&cache->trimming, 0); |
a5ed9182 | 10241 | mutex_init(&cache->free_space_lock); |
0966a7b1 | 10242 | btrfs_init_full_stripe_locks_tree(&cache->full_stripe_locks_root); |
920e4a58 MX |
10243 | |
10244 | return cache; | |
10245 | } | |
10246 | ||
7ef49515 QW |
10247 | |
10248 | /* | |
10249 | * Iterate all chunks and verify that each of them has the corresponding block | |
10250 | * group | |
10251 | */ | |
10252 | static int check_chunk_block_group_mappings(struct btrfs_fs_info *fs_info) | |
10253 | { | |
c8bf1b67 | 10254 | struct extent_map_tree *map_tree = &fs_info->mapping_tree; |
7ef49515 QW |
10255 | struct extent_map *em; |
10256 | struct btrfs_block_group_cache *bg; | |
10257 | u64 start = 0; | |
10258 | int ret = 0; | |
10259 | ||
10260 | while (1) { | |
c8bf1b67 | 10261 | read_lock(&map_tree->lock); |
7ef49515 QW |
10262 | /* |
10263 | * lookup_extent_mapping will return the first extent map | |
10264 | * intersecting the range, so setting @len to 1 is enough to | |
10265 | * get the first chunk. | |
10266 | */ | |
c8bf1b67 DS |
10267 | em = lookup_extent_mapping(map_tree, start, 1); |
10268 | read_unlock(&map_tree->lock); | |
7ef49515 QW |
10269 | if (!em) |
10270 | break; | |
10271 | ||
10272 | bg = btrfs_lookup_block_group(fs_info, em->start); | |
10273 | if (!bg) { | |
10274 | btrfs_err(fs_info, | |
10275 | "chunk start=%llu len=%llu doesn't have corresponding block group", | |
10276 | em->start, em->len); | |
10277 | ret = -EUCLEAN; | |
10278 | free_extent_map(em); | |
10279 | break; | |
10280 | } | |
10281 | if (bg->key.objectid != em->start || | |
10282 | bg->key.offset != em->len || | |
10283 | (bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK) != | |
10284 | (em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK)) { | |
10285 | btrfs_err(fs_info, | |
10286 | "chunk start=%llu len=%llu flags=0x%llx doesn't match block group start=%llu len=%llu flags=0x%llx", | |
10287 | em->start, em->len, | |
10288 | em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK, | |
10289 | bg->key.objectid, bg->key.offset, | |
10290 | bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK); | |
10291 | ret = -EUCLEAN; | |
10292 | free_extent_map(em); | |
10293 | btrfs_put_block_group(bg); | |
10294 | break; | |
10295 | } | |
10296 | start = em->start + em->len; | |
10297 | free_extent_map(em); | |
10298 | btrfs_put_block_group(bg); | |
10299 | } | |
10300 | return ret; | |
10301 | } | |
10302 | ||
5b4aacef | 10303 | int btrfs_read_block_groups(struct btrfs_fs_info *info) |
9078a3e1 CM |
10304 | { |
10305 | struct btrfs_path *path; | |
10306 | int ret; | |
9078a3e1 | 10307 | struct btrfs_block_group_cache *cache; |
6324fbf3 | 10308 | struct btrfs_space_info *space_info; |
9078a3e1 CM |
10309 | struct btrfs_key key; |
10310 | struct btrfs_key found_key; | |
5f39d397 | 10311 | struct extent_buffer *leaf; |
0af3d00b JB |
10312 | int need_clear = 0; |
10313 | u64 cache_gen; | |
49303381 LB |
10314 | u64 feature; |
10315 | int mixed; | |
10316 | ||
10317 | feature = btrfs_super_incompat_flags(info->super_copy); | |
10318 | mixed = !!(feature & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS); | |
96b5179d | 10319 | |
9078a3e1 | 10320 | key.objectid = 0; |
0b86a832 | 10321 | key.offset = 0; |
962a298f | 10322 | key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; |
9078a3e1 CM |
10323 | path = btrfs_alloc_path(); |
10324 | if (!path) | |
10325 | return -ENOMEM; | |
e4058b54 | 10326 | path->reada = READA_FORWARD; |
9078a3e1 | 10327 | |
0b246afa JM |
10328 | cache_gen = btrfs_super_cache_generation(info->super_copy); |
10329 | if (btrfs_test_opt(info, SPACE_CACHE) && | |
10330 | btrfs_super_generation(info->super_copy) != cache_gen) | |
0af3d00b | 10331 | need_clear = 1; |
0b246afa | 10332 | if (btrfs_test_opt(info, CLEAR_CACHE)) |
88c2ba3b | 10333 | need_clear = 1; |
0af3d00b | 10334 | |
d397712b | 10335 | while (1) { |
6bccf3ab | 10336 | ret = find_first_block_group(info, path, &key); |
b742bb82 YZ |
10337 | if (ret > 0) |
10338 | break; | |
0b86a832 CM |
10339 | if (ret != 0) |
10340 | goto error; | |
920e4a58 | 10341 | |
5f39d397 CM |
10342 | leaf = path->nodes[0]; |
10343 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
920e4a58 | 10344 | |
2ff7e61e | 10345 | cache = btrfs_create_block_group_cache(info, found_key.objectid, |
920e4a58 | 10346 | found_key.offset); |
9078a3e1 | 10347 | if (!cache) { |
0b86a832 | 10348 | ret = -ENOMEM; |
f0486c68 | 10349 | goto error; |
9078a3e1 | 10350 | } |
96303081 | 10351 | |
cf7c1ef6 LB |
10352 | if (need_clear) { |
10353 | /* | |
10354 | * When we mount with old space cache, we need to | |
10355 | * set BTRFS_DC_CLEAR and set dirty flag. | |
10356 | * | |
10357 | * a) Setting 'BTRFS_DC_CLEAR' makes sure that we | |
10358 | * truncate the old free space cache inode and | |
10359 | * setup a new one. | |
10360 | * b) Setting 'dirty flag' makes sure that we flush | |
10361 | * the new space cache info onto disk. | |
10362 | */ | |
0b246afa | 10363 | if (btrfs_test_opt(info, SPACE_CACHE)) |
ce93ec54 | 10364 | cache->disk_cache_state = BTRFS_DC_CLEAR; |
cf7c1ef6 | 10365 | } |
0af3d00b | 10366 | |
5f39d397 CM |
10367 | read_extent_buffer(leaf, &cache->item, |
10368 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
10369 | sizeof(cache->item)); | |
920e4a58 | 10370 | cache->flags = btrfs_block_group_flags(&cache->item); |
49303381 LB |
10371 | if (!mixed && |
10372 | ((cache->flags & BTRFS_BLOCK_GROUP_METADATA) && | |
10373 | (cache->flags & BTRFS_BLOCK_GROUP_DATA))) { | |
10374 | btrfs_err(info, | |
10375 | "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups", | |
10376 | cache->key.objectid); | |
10377 | ret = -EINVAL; | |
10378 | goto error; | |
10379 | } | |
0b86a832 | 10380 | |
9078a3e1 | 10381 | key.objectid = found_key.objectid + found_key.offset; |
b3b4aa74 | 10382 | btrfs_release_path(path); |
34d52cb6 | 10383 | |
3c14874a JB |
10384 | /* |
10385 | * We need to exclude the super stripes now so that the space | |
10386 | * info has super bytes accounted for, otherwise we'll think | |
10387 | * we have more space than we actually do. | |
10388 | */ | |
3c4da657 | 10389 | ret = exclude_super_stripes(cache); |
835d974f JB |
10390 | if (ret) { |
10391 | /* | |
10392 | * We may have excluded something, so call this just in | |
10393 | * case. | |
10394 | */ | |
9e715da8 | 10395 | free_excluded_extents(cache); |
920e4a58 | 10396 | btrfs_put_block_group(cache); |
835d974f JB |
10397 | goto error; |
10398 | } | |
3c14874a | 10399 | |
817d52f8 JB |
10400 | /* |
10401 | * check for two cases, either we are full, and therefore | |
10402 | * don't need to bother with the caching work since we won't | |
10403 | * find any space, or we are empty, and we can just add all | |
52042d8e | 10404 | * the space in and be done with it. This saves us _a_lot_ of |
817d52f8 JB |
10405 | * time, particularly in the full case. |
10406 | */ | |
10407 | if (found_key.offset == btrfs_block_group_used(&cache->item)) { | |
11833d66 | 10408 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 10409 | cache->cached = BTRFS_CACHE_FINISHED; |
9e715da8 | 10410 | free_excluded_extents(cache); |
817d52f8 | 10411 | } else if (btrfs_block_group_used(&cache->item) == 0) { |
11833d66 | 10412 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 10413 | cache->cached = BTRFS_CACHE_FINISHED; |
4457c1c7 | 10414 | add_new_free_space(cache, found_key.objectid, |
817d52f8 JB |
10415 | found_key.objectid + |
10416 | found_key.offset); | |
9e715da8 | 10417 | free_excluded_extents(cache); |
817d52f8 | 10418 | } |
96b5179d | 10419 | |
0b246afa | 10420 | ret = btrfs_add_block_group_cache(info, cache); |
8c579fe7 JB |
10421 | if (ret) { |
10422 | btrfs_remove_free_space_cache(cache); | |
10423 | btrfs_put_block_group(cache); | |
10424 | goto error; | |
10425 | } | |
10426 | ||
0b246afa | 10427 | trace_btrfs_add_block_group(info, cache, 0); |
d2006e6d NB |
10428 | update_space_info(info, cache->flags, found_key.offset, |
10429 | btrfs_block_group_used(&cache->item), | |
10430 | cache->bytes_super, &space_info); | |
8c579fe7 | 10431 | |
6324fbf3 | 10432 | cache->space_info = space_info; |
1b2da372 | 10433 | |
c434d21c | 10434 | link_block_group(cache); |
0f9dd46c | 10435 | |
0b246afa | 10436 | set_avail_alloc_bits(info, cache->flags); |
2ff7e61e | 10437 | if (btrfs_chunk_readonly(info, cache->key.objectid)) { |
868f401a | 10438 | inc_block_group_ro(cache, 1); |
47ab2a6c | 10439 | } else if (btrfs_block_group_used(&cache->item) == 0) { |
031f24da QW |
10440 | ASSERT(list_empty(&cache->bg_list)); |
10441 | btrfs_mark_bg_unused(cache); | |
47ab2a6c | 10442 | } |
9078a3e1 | 10443 | } |
b742bb82 | 10444 | |
0b246afa | 10445 | list_for_each_entry_rcu(space_info, &info->space_info, list) { |
2ff7e61e | 10446 | if (!(get_alloc_profile(info, space_info->flags) & |
b742bb82 YZ |
10447 | (BTRFS_BLOCK_GROUP_RAID10 | |
10448 | BTRFS_BLOCK_GROUP_RAID1 | | |
53b381b3 DW |
10449 | BTRFS_BLOCK_GROUP_RAID5 | |
10450 | BTRFS_BLOCK_GROUP_RAID6 | | |
b742bb82 YZ |
10451 | BTRFS_BLOCK_GROUP_DUP))) |
10452 | continue; | |
10453 | /* | |
10454 | * avoid allocating from un-mirrored block group if there are | |
10455 | * mirrored block groups. | |
10456 | */ | |
1095cc0d | 10457 | list_for_each_entry(cache, |
10458 | &space_info->block_groups[BTRFS_RAID_RAID0], | |
10459 | list) | |
868f401a | 10460 | inc_block_group_ro(cache, 1); |
1095cc0d | 10461 | list_for_each_entry(cache, |
10462 | &space_info->block_groups[BTRFS_RAID_SINGLE], | |
10463 | list) | |
868f401a | 10464 | inc_block_group_ro(cache, 1); |
9078a3e1 | 10465 | } |
f0486c68 | 10466 | |
75cb379d | 10467 | btrfs_add_raid_kobjects(info); |
f0486c68 | 10468 | init_global_block_rsv(info); |
7ef49515 | 10469 | ret = check_chunk_block_group_mappings(info); |
0b86a832 | 10470 | error: |
9078a3e1 | 10471 | btrfs_free_path(path); |
0b86a832 | 10472 | return ret; |
9078a3e1 | 10473 | } |
6324fbf3 | 10474 | |
6c686b35 | 10475 | void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans) |
ea658bad | 10476 | { |
6c686b35 | 10477 | struct btrfs_fs_info *fs_info = trans->fs_info; |
545e3366 | 10478 | struct btrfs_block_group_cache *block_group; |
0b246afa | 10479 | struct btrfs_root *extent_root = fs_info->extent_root; |
ea658bad JB |
10480 | struct btrfs_block_group_item item; |
10481 | struct btrfs_key key; | |
10482 | int ret = 0; | |
10483 | ||
5ce55557 FM |
10484 | if (!trans->can_flush_pending_bgs) |
10485 | return; | |
10486 | ||
545e3366 JB |
10487 | while (!list_empty(&trans->new_bgs)) { |
10488 | block_group = list_first_entry(&trans->new_bgs, | |
10489 | struct btrfs_block_group_cache, | |
10490 | bg_list); | |
ea658bad | 10491 | if (ret) |
c92f6be3 | 10492 | goto next; |
ea658bad JB |
10493 | |
10494 | spin_lock(&block_group->lock); | |
10495 | memcpy(&item, &block_group->item, sizeof(item)); | |
10496 | memcpy(&key, &block_group->key, sizeof(key)); | |
10497 | spin_unlock(&block_group->lock); | |
10498 | ||
10499 | ret = btrfs_insert_item(trans, extent_root, &key, &item, | |
10500 | sizeof(item)); | |
10501 | if (ret) | |
66642832 | 10502 | btrfs_abort_transaction(trans, ret); |
97aff912 | 10503 | ret = btrfs_finish_chunk_alloc(trans, key.objectid, key.offset); |
6df9a95e | 10504 | if (ret) |
66642832 | 10505 | btrfs_abort_transaction(trans, ret); |
e4e0711c | 10506 | add_block_group_free_space(trans, block_group); |
1e144fb8 | 10507 | /* already aborted the transaction if it failed. */ |
c92f6be3 | 10508 | next: |
ba2c4d4e | 10509 | btrfs_delayed_refs_rsv_release(fs_info, 1); |
c92f6be3 | 10510 | list_del_init(&block_group->bg_list); |
ea658bad | 10511 | } |
5ce55557 | 10512 | btrfs_trans_release_chunk_metadata(trans); |
ea658bad JB |
10513 | } |
10514 | ||
e7e02096 | 10515 | int btrfs_make_block_group(struct btrfs_trans_handle *trans, u64 bytes_used, |
0174484d | 10516 | u64 type, u64 chunk_offset, u64 size) |
6324fbf3 | 10517 | { |
e7e02096 | 10518 | struct btrfs_fs_info *fs_info = trans->fs_info; |
6324fbf3 | 10519 | struct btrfs_block_group_cache *cache; |
0b246afa | 10520 | int ret; |
6324fbf3 | 10521 | |
90787766 | 10522 | btrfs_set_log_full_commit(trans); |
e02119d5 | 10523 | |
2ff7e61e | 10524 | cache = btrfs_create_block_group_cache(fs_info, chunk_offset, size); |
0f9dd46c JB |
10525 | if (!cache) |
10526 | return -ENOMEM; | |
34d52cb6 | 10527 | |
6324fbf3 | 10528 | btrfs_set_block_group_used(&cache->item, bytes_used); |
0174484d NB |
10529 | btrfs_set_block_group_chunk_objectid(&cache->item, |
10530 | BTRFS_FIRST_CHUNK_TREE_OBJECTID); | |
6324fbf3 CM |
10531 | btrfs_set_block_group_flags(&cache->item, type); |
10532 | ||
920e4a58 | 10533 | cache->flags = type; |
11833d66 | 10534 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 10535 | cache->cached = BTRFS_CACHE_FINISHED; |
1e144fb8 | 10536 | cache->needs_free_space = 1; |
3c4da657 | 10537 | ret = exclude_super_stripes(cache); |
835d974f JB |
10538 | if (ret) { |
10539 | /* | |
10540 | * We may have excluded something, so call this just in | |
10541 | * case. | |
10542 | */ | |
9e715da8 | 10543 | free_excluded_extents(cache); |
920e4a58 | 10544 | btrfs_put_block_group(cache); |
835d974f JB |
10545 | return ret; |
10546 | } | |
96303081 | 10547 | |
4457c1c7 | 10548 | add_new_free_space(cache, chunk_offset, chunk_offset + size); |
817d52f8 | 10549 | |
9e715da8 | 10550 | free_excluded_extents(cache); |
11833d66 | 10551 | |
d0bd4560 | 10552 | #ifdef CONFIG_BTRFS_DEBUG |
2ff7e61e | 10553 | if (btrfs_should_fragment_free_space(cache)) { |
d0bd4560 JB |
10554 | u64 new_bytes_used = size - bytes_used; |
10555 | ||
10556 | bytes_used += new_bytes_used >> 1; | |
2ff7e61e | 10557 | fragment_free_space(cache); |
d0bd4560 JB |
10558 | } |
10559 | #endif | |
2e6e5183 | 10560 | /* |
2be12ef7 NB |
10561 | * Ensure the corresponding space_info object is created and |
10562 | * assigned to our block group. We want our bg to be added to the rbtree | |
10563 | * with its ->space_info set. | |
2e6e5183 | 10564 | */ |
2be12ef7 | 10565 | cache->space_info = __find_space_info(fs_info, cache->flags); |
dc2d3005 | 10566 | ASSERT(cache->space_info); |
2e6e5183 | 10567 | |
0b246afa | 10568 | ret = btrfs_add_block_group_cache(fs_info, cache); |
8c579fe7 JB |
10569 | if (ret) { |
10570 | btrfs_remove_free_space_cache(cache); | |
10571 | btrfs_put_block_group(cache); | |
10572 | return ret; | |
10573 | } | |
10574 | ||
2e6e5183 FM |
10575 | /* |
10576 | * Now that our block group has its ->space_info set and is inserted in | |
10577 | * the rbtree, update the space info's counters. | |
10578 | */ | |
0b246afa | 10579 | trace_btrfs_add_block_group(fs_info, cache, 1); |
d2006e6d | 10580 | update_space_info(fs_info, cache->flags, size, bytes_used, |
e40edf2d | 10581 | cache->bytes_super, &cache->space_info); |
0b246afa | 10582 | update_global_block_rsv(fs_info); |
1b2da372 | 10583 | |
c434d21c | 10584 | link_block_group(cache); |
6324fbf3 | 10585 | |
47ab2a6c | 10586 | list_add_tail(&cache->bg_list, &trans->new_bgs); |
ba2c4d4e JB |
10587 | trans->delayed_ref_updates++; |
10588 | btrfs_update_delayed_refs_rsv(trans); | |
6324fbf3 | 10589 | |
0b246afa | 10590 | set_avail_alloc_bits(fs_info, type); |
6324fbf3 CM |
10591 | return 0; |
10592 | } | |
1a40e23b | 10593 | |
10ea00f5 ID |
10594 | static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
10595 | { | |
899c81ea ID |
10596 | u64 extra_flags = chunk_to_extended(flags) & |
10597 | BTRFS_EXTENDED_PROFILE_MASK; | |
10ea00f5 | 10598 | |
de98ced9 | 10599 | write_seqlock(&fs_info->profiles_lock); |
10ea00f5 ID |
10600 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
10601 | fs_info->avail_data_alloc_bits &= ~extra_flags; | |
10602 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
10603 | fs_info->avail_metadata_alloc_bits &= ~extra_flags; | |
10604 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
10605 | fs_info->avail_system_alloc_bits &= ~extra_flags; | |
de98ced9 | 10606 | write_sequnlock(&fs_info->profiles_lock); |
10ea00f5 ID |
10607 | } |
10608 | ||
1a40e23b | 10609 | int btrfs_remove_block_group(struct btrfs_trans_handle *trans, |
5a98ec01 | 10610 | u64 group_start, struct extent_map *em) |
1a40e23b | 10611 | { |
5a98ec01 | 10612 | struct btrfs_fs_info *fs_info = trans->fs_info; |
6bccf3ab | 10613 | struct btrfs_root *root = fs_info->extent_root; |
1a40e23b ZY |
10614 | struct btrfs_path *path; |
10615 | struct btrfs_block_group_cache *block_group; | |
44fb5511 | 10616 | struct btrfs_free_cluster *cluster; |
0b246afa | 10617 | struct btrfs_root *tree_root = fs_info->tree_root; |
1a40e23b | 10618 | struct btrfs_key key; |
0af3d00b | 10619 | struct inode *inode; |
c1895442 | 10620 | struct kobject *kobj = NULL; |
1a40e23b | 10621 | int ret; |
10ea00f5 | 10622 | int index; |
89a55897 | 10623 | int factor; |
4f69cb98 | 10624 | struct btrfs_caching_control *caching_ctl = NULL; |
04216820 | 10625 | bool remove_em; |
ba2c4d4e | 10626 | bool remove_rsv = false; |
1a40e23b | 10627 | |
6bccf3ab | 10628 | block_group = btrfs_lookup_block_group(fs_info, group_start); |
1a40e23b | 10629 | BUG_ON(!block_group); |
c146afad | 10630 | BUG_ON(!block_group->ro); |
1a40e23b | 10631 | |
4ed0a7a3 | 10632 | trace_btrfs_remove_block_group(block_group); |
9f7c43c9 | 10633 | /* |
10634 | * Free the reserved super bytes from this block group before | |
10635 | * remove it. | |
10636 | */ | |
9e715da8 | 10637 | free_excluded_extents(block_group); |
fd708b81 JB |
10638 | btrfs_free_ref_tree_range(fs_info, block_group->key.objectid, |
10639 | block_group->key.offset); | |
9f7c43c9 | 10640 | |
1a40e23b | 10641 | memcpy(&key, &block_group->key, sizeof(key)); |
3e72ee88 | 10642 | index = btrfs_bg_flags_to_raid_index(block_group->flags); |
46df06b8 | 10643 | factor = btrfs_bg_type_to_factor(block_group->flags); |
1a40e23b | 10644 | |
44fb5511 | 10645 | /* make sure this block group isn't part of an allocation cluster */ |
0b246afa | 10646 | cluster = &fs_info->data_alloc_cluster; |
44fb5511 CM |
10647 | spin_lock(&cluster->refill_lock); |
10648 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
10649 | spin_unlock(&cluster->refill_lock); | |
10650 | ||
10651 | /* | |
10652 | * make sure this block group isn't part of a metadata | |
10653 | * allocation cluster | |
10654 | */ | |
0b246afa | 10655 | cluster = &fs_info->meta_alloc_cluster; |
44fb5511 CM |
10656 | spin_lock(&cluster->refill_lock); |
10657 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
10658 | spin_unlock(&cluster->refill_lock); | |
10659 | ||
1a40e23b | 10660 | path = btrfs_alloc_path(); |
d8926bb3 MF |
10661 | if (!path) { |
10662 | ret = -ENOMEM; | |
10663 | goto out; | |
10664 | } | |
1a40e23b | 10665 | |
1bbc621e CM |
10666 | /* |
10667 | * get the inode first so any iput calls done for the io_list | |
10668 | * aren't the final iput (no unlinks allowed now) | |
10669 | */ | |
7949f339 | 10670 | inode = lookup_free_space_inode(block_group, path); |
1bbc621e CM |
10671 | |
10672 | mutex_lock(&trans->transaction->cache_write_mutex); | |
10673 | /* | |
52042d8e | 10674 | * Make sure our free space cache IO is done before removing the |
1bbc621e CM |
10675 | * free space inode |
10676 | */ | |
10677 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
10678 | if (!list_empty(&block_group->io_list)) { | |
10679 | list_del_init(&block_group->io_list); | |
10680 | ||
10681 | WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode); | |
10682 | ||
10683 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
afdb5718 | 10684 | btrfs_wait_cache_io(trans, block_group, path); |
1bbc621e CM |
10685 | btrfs_put_block_group(block_group); |
10686 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
10687 | } | |
10688 | ||
10689 | if (!list_empty(&block_group->dirty_list)) { | |
10690 | list_del_init(&block_group->dirty_list); | |
ba2c4d4e | 10691 | remove_rsv = true; |
1bbc621e CM |
10692 | btrfs_put_block_group(block_group); |
10693 | } | |
10694 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
10695 | mutex_unlock(&trans->transaction->cache_write_mutex); | |
10696 | ||
0af3d00b | 10697 | if (!IS_ERR(inode)) { |
73f2e545 | 10698 | ret = btrfs_orphan_add(trans, BTRFS_I(inode)); |
79787eaa JM |
10699 | if (ret) { |
10700 | btrfs_add_delayed_iput(inode); | |
10701 | goto out; | |
10702 | } | |
0af3d00b JB |
10703 | clear_nlink(inode); |
10704 | /* One for the block groups ref */ | |
10705 | spin_lock(&block_group->lock); | |
10706 | if (block_group->iref) { | |
10707 | block_group->iref = 0; | |
10708 | block_group->inode = NULL; | |
10709 | spin_unlock(&block_group->lock); | |
10710 | iput(inode); | |
10711 | } else { | |
10712 | spin_unlock(&block_group->lock); | |
10713 | } | |
10714 | /* One for our lookup ref */ | |
455757c3 | 10715 | btrfs_add_delayed_iput(inode); |
0af3d00b JB |
10716 | } |
10717 | ||
10718 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
10719 | key.offset = block_group->key.objectid; | |
10720 | key.type = 0; | |
10721 | ||
10722 | ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1); | |
10723 | if (ret < 0) | |
10724 | goto out; | |
10725 | if (ret > 0) | |
b3b4aa74 | 10726 | btrfs_release_path(path); |
0af3d00b JB |
10727 | if (ret == 0) { |
10728 | ret = btrfs_del_item(trans, tree_root, path); | |
10729 | if (ret) | |
10730 | goto out; | |
b3b4aa74 | 10731 | btrfs_release_path(path); |
0af3d00b JB |
10732 | } |
10733 | ||
0b246afa | 10734 | spin_lock(&fs_info->block_group_cache_lock); |
1a40e23b | 10735 | rb_erase(&block_group->cache_node, |
0b246afa | 10736 | &fs_info->block_group_cache_tree); |
292cbd51 | 10737 | RB_CLEAR_NODE(&block_group->cache_node); |
a1897fdd | 10738 | |
0b246afa JM |
10739 | if (fs_info->first_logical_byte == block_group->key.objectid) |
10740 | fs_info->first_logical_byte = (u64)-1; | |
10741 | spin_unlock(&fs_info->block_group_cache_lock); | |
817d52f8 | 10742 | |
80eb234a | 10743 | down_write(&block_group->space_info->groups_sem); |
44fb5511 CM |
10744 | /* |
10745 | * we must use list_del_init so people can check to see if they | |
10746 | * are still on the list after taking the semaphore | |
10747 | */ | |
10748 | list_del_init(&block_group->list); | |
6ab0a202 | 10749 | if (list_empty(&block_group->space_info->block_groups[index])) { |
c1895442 JM |
10750 | kobj = block_group->space_info->block_group_kobjs[index]; |
10751 | block_group->space_info->block_group_kobjs[index] = NULL; | |
0b246afa | 10752 | clear_avail_alloc_bits(fs_info, block_group->flags); |
6ab0a202 | 10753 | } |
80eb234a | 10754 | up_write(&block_group->space_info->groups_sem); |
c1895442 JM |
10755 | if (kobj) { |
10756 | kobject_del(kobj); | |
10757 | kobject_put(kobj); | |
10758 | } | |
1a40e23b | 10759 | |
4f69cb98 FM |
10760 | if (block_group->has_caching_ctl) |
10761 | caching_ctl = get_caching_control(block_group); | |
817d52f8 | 10762 | if (block_group->cached == BTRFS_CACHE_STARTED) |
11833d66 | 10763 | wait_block_group_cache_done(block_group); |
4f69cb98 | 10764 | if (block_group->has_caching_ctl) { |
0b246afa | 10765 | down_write(&fs_info->commit_root_sem); |
4f69cb98 FM |
10766 | if (!caching_ctl) { |
10767 | struct btrfs_caching_control *ctl; | |
10768 | ||
10769 | list_for_each_entry(ctl, | |
0b246afa | 10770 | &fs_info->caching_block_groups, list) |
4f69cb98 FM |
10771 | if (ctl->block_group == block_group) { |
10772 | caching_ctl = ctl; | |
1e4f4714 | 10773 | refcount_inc(&caching_ctl->count); |
4f69cb98 FM |
10774 | break; |
10775 | } | |
10776 | } | |
10777 | if (caching_ctl) | |
10778 | list_del_init(&caching_ctl->list); | |
0b246afa | 10779 | up_write(&fs_info->commit_root_sem); |
4f69cb98 FM |
10780 | if (caching_ctl) { |
10781 | /* Once for the caching bgs list and once for us. */ | |
10782 | put_caching_control(caching_ctl); | |
10783 | put_caching_control(caching_ctl); | |
10784 | } | |
10785 | } | |
817d52f8 | 10786 | |
ce93ec54 | 10787 | spin_lock(&trans->transaction->dirty_bgs_lock); |
9a0ec83d NB |
10788 | WARN_ON(!list_empty(&block_group->dirty_list)); |
10789 | WARN_ON(!list_empty(&block_group->io_list)); | |
ce93ec54 | 10790 | spin_unlock(&trans->transaction->dirty_bgs_lock); |
9a0ec83d | 10791 | |
817d52f8 JB |
10792 | btrfs_remove_free_space_cache(block_group); |
10793 | ||
c146afad | 10794 | spin_lock(&block_group->space_info->lock); |
75c68e9f | 10795 | list_del_init(&block_group->ro_list); |
18d018ad | 10796 | |
0b246afa | 10797 | if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
18d018ad ZL |
10798 | WARN_ON(block_group->space_info->total_bytes |
10799 | < block_group->key.offset); | |
10800 | WARN_ON(block_group->space_info->bytes_readonly | |
10801 | < block_group->key.offset); | |
10802 | WARN_ON(block_group->space_info->disk_total | |
10803 | < block_group->key.offset * factor); | |
10804 | } | |
c146afad YZ |
10805 | block_group->space_info->total_bytes -= block_group->key.offset; |
10806 | block_group->space_info->bytes_readonly -= block_group->key.offset; | |
89a55897 | 10807 | block_group->space_info->disk_total -= block_group->key.offset * factor; |
18d018ad | 10808 | |
c146afad | 10809 | spin_unlock(&block_group->space_info->lock); |
283bb197 | 10810 | |
0af3d00b JB |
10811 | memcpy(&key, &block_group->key, sizeof(key)); |
10812 | ||
34441361 | 10813 | mutex_lock(&fs_info->chunk_mutex); |
04216820 FM |
10814 | spin_lock(&block_group->lock); |
10815 | block_group->removed = 1; | |
10816 | /* | |
10817 | * At this point trimming can't start on this block group, because we | |
10818 | * removed the block group from the tree fs_info->block_group_cache_tree | |
10819 | * so no one can't find it anymore and even if someone already got this | |
10820 | * block group before we removed it from the rbtree, they have already | |
10821 | * incremented block_group->trimming - if they didn't, they won't find | |
10822 | * any free space entries because we already removed them all when we | |
10823 | * called btrfs_remove_free_space_cache(). | |
10824 | * | |
10825 | * And we must not remove the extent map from the fs_info->mapping_tree | |
10826 | * to prevent the same logical address range and physical device space | |
10827 | * ranges from being reused for a new block group. This is because our | |
10828 | * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is | |
10829 | * completely transactionless, so while it is trimming a range the | |
10830 | * currently running transaction might finish and a new one start, | |
10831 | * allowing for new block groups to be created that can reuse the same | |
10832 | * physical device locations unless we take this special care. | |
e33e17ee JM |
10833 | * |
10834 | * There may also be an implicit trim operation if the file system | |
10835 | * is mounted with -odiscard. The same protections must remain | |
10836 | * in place until the extents have been discarded completely when | |
10837 | * the transaction commit has completed. | |
04216820 FM |
10838 | */ |
10839 | remove_em = (atomic_read(&block_group->trimming) == 0); | |
04216820 | 10840 | spin_unlock(&block_group->lock); |
04216820 | 10841 | |
34441361 | 10842 | mutex_unlock(&fs_info->chunk_mutex); |
8dbcd10f | 10843 | |
f3f72779 | 10844 | ret = remove_block_group_free_space(trans, block_group); |
1e144fb8 OS |
10845 | if (ret) |
10846 | goto out; | |
10847 | ||
fa9c0d79 CM |
10848 | btrfs_put_block_group(block_group); |
10849 | btrfs_put_block_group(block_group); | |
1a40e23b ZY |
10850 | |
10851 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
10852 | if (ret > 0) | |
10853 | ret = -EIO; | |
10854 | if (ret < 0) | |
10855 | goto out; | |
10856 | ||
10857 | ret = btrfs_del_item(trans, root, path); | |
8eaf40c0 FM |
10858 | if (ret) |
10859 | goto out; | |
10860 | ||
10861 | if (remove_em) { | |
10862 | struct extent_map_tree *em_tree; | |
10863 | ||
c8bf1b67 | 10864 | em_tree = &fs_info->mapping_tree; |
8eaf40c0 FM |
10865 | write_lock(&em_tree->lock); |
10866 | remove_extent_mapping(em_tree, em); | |
10867 | write_unlock(&em_tree->lock); | |
10868 | /* once for the tree */ | |
10869 | free_extent_map(em); | |
10870 | } | |
1a40e23b | 10871 | out: |
ba2c4d4e JB |
10872 | if (remove_rsv) |
10873 | btrfs_delayed_refs_rsv_release(fs_info, 1); | |
1a40e23b ZY |
10874 | btrfs_free_path(path); |
10875 | return ret; | |
10876 | } | |
acce952b | 10877 | |
8eab77ff | 10878 | struct btrfs_trans_handle * |
7fd01182 FM |
10879 | btrfs_start_trans_remove_block_group(struct btrfs_fs_info *fs_info, |
10880 | const u64 chunk_offset) | |
8eab77ff | 10881 | { |
c8bf1b67 | 10882 | struct extent_map_tree *em_tree = &fs_info->mapping_tree; |
7fd01182 FM |
10883 | struct extent_map *em; |
10884 | struct map_lookup *map; | |
10885 | unsigned int num_items; | |
10886 | ||
10887 | read_lock(&em_tree->lock); | |
10888 | em = lookup_extent_mapping(em_tree, chunk_offset, 1); | |
10889 | read_unlock(&em_tree->lock); | |
10890 | ASSERT(em && em->start == chunk_offset); | |
10891 | ||
8eab77ff | 10892 | /* |
7fd01182 FM |
10893 | * We need to reserve 3 + N units from the metadata space info in order |
10894 | * to remove a block group (done at btrfs_remove_chunk() and at | |
10895 | * btrfs_remove_block_group()), which are used for: | |
10896 | * | |
8eab77ff FM |
10897 | * 1 unit for adding the free space inode's orphan (located in the tree |
10898 | * of tree roots). | |
7fd01182 FM |
10899 | * 1 unit for deleting the block group item (located in the extent |
10900 | * tree). | |
10901 | * 1 unit for deleting the free space item (located in tree of tree | |
10902 | * roots). | |
10903 | * N units for deleting N device extent items corresponding to each | |
10904 | * stripe (located in the device tree). | |
10905 | * | |
10906 | * In order to remove a block group we also need to reserve units in the | |
10907 | * system space info in order to update the chunk tree (update one or | |
10908 | * more device items and remove one chunk item), but this is done at | |
10909 | * btrfs_remove_chunk() through a call to check_system_chunk(). | |
8eab77ff | 10910 | */ |
95617d69 | 10911 | map = em->map_lookup; |
7fd01182 FM |
10912 | num_items = 3 + map->num_stripes; |
10913 | free_extent_map(em); | |
10914 | ||
8eab77ff | 10915 | return btrfs_start_transaction_fallback_global_rsv(fs_info->extent_root, |
7fd01182 | 10916 | num_items, 1); |
8eab77ff FM |
10917 | } |
10918 | ||
47ab2a6c JB |
10919 | /* |
10920 | * Process the unused_bgs list and remove any that don't have any allocated | |
10921 | * space inside of them. | |
10922 | */ | |
10923 | void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info) | |
10924 | { | |
10925 | struct btrfs_block_group_cache *block_group; | |
10926 | struct btrfs_space_info *space_info; | |
47ab2a6c JB |
10927 | struct btrfs_trans_handle *trans; |
10928 | int ret = 0; | |
10929 | ||
afcdd129 | 10930 | if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags)) |
47ab2a6c JB |
10931 | return; |
10932 | ||
10933 | spin_lock(&fs_info->unused_bgs_lock); | |
10934 | while (!list_empty(&fs_info->unused_bgs)) { | |
10935 | u64 start, end; | |
e33e17ee | 10936 | int trimming; |
47ab2a6c JB |
10937 | |
10938 | block_group = list_first_entry(&fs_info->unused_bgs, | |
10939 | struct btrfs_block_group_cache, | |
10940 | bg_list); | |
47ab2a6c | 10941 | list_del_init(&block_group->bg_list); |
aefbe9a6 ZL |
10942 | |
10943 | space_info = block_group->space_info; | |
10944 | ||
47ab2a6c JB |
10945 | if (ret || btrfs_mixed_space_info(space_info)) { |
10946 | btrfs_put_block_group(block_group); | |
10947 | continue; | |
10948 | } | |
10949 | spin_unlock(&fs_info->unused_bgs_lock); | |
10950 | ||
d5f2e33b | 10951 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
67c5e7d4 | 10952 | |
47ab2a6c JB |
10953 | /* Don't want to race with allocators so take the groups_sem */ |
10954 | down_write(&space_info->groups_sem); | |
10955 | spin_lock(&block_group->lock); | |
43794446 | 10956 | if (block_group->reserved || block_group->pinned || |
47ab2a6c | 10957 | btrfs_block_group_used(&block_group->item) || |
19c4d2f9 | 10958 | block_group->ro || |
aefbe9a6 | 10959 | list_is_singular(&block_group->list)) { |
47ab2a6c JB |
10960 | /* |
10961 | * We want to bail if we made new allocations or have | |
10962 | * outstanding allocations in this block group. We do | |
10963 | * the ro check in case balance is currently acting on | |
10964 | * this block group. | |
10965 | */ | |
4ed0a7a3 | 10966 | trace_btrfs_skip_unused_block_group(block_group); |
47ab2a6c JB |
10967 | spin_unlock(&block_group->lock); |
10968 | up_write(&space_info->groups_sem); | |
10969 | goto next; | |
10970 | } | |
10971 | spin_unlock(&block_group->lock); | |
10972 | ||
10973 | /* We don't want to force the issue, only flip if it's ok. */ | |
868f401a | 10974 | ret = inc_block_group_ro(block_group, 0); |
47ab2a6c JB |
10975 | up_write(&space_info->groups_sem); |
10976 | if (ret < 0) { | |
10977 | ret = 0; | |
10978 | goto next; | |
10979 | } | |
10980 | ||
10981 | /* | |
10982 | * Want to do this before we do anything else so we can recover | |
10983 | * properly if we fail to join the transaction. | |
10984 | */ | |
7fd01182 FM |
10985 | trans = btrfs_start_trans_remove_block_group(fs_info, |
10986 | block_group->key.objectid); | |
47ab2a6c | 10987 | if (IS_ERR(trans)) { |
2ff7e61e | 10988 | btrfs_dec_block_group_ro(block_group); |
47ab2a6c JB |
10989 | ret = PTR_ERR(trans); |
10990 | goto next; | |
10991 | } | |
10992 | ||
10993 | /* | |
10994 | * We could have pending pinned extents for this block group, | |
10995 | * just delete them, we don't care about them anymore. | |
10996 | */ | |
10997 | start = block_group->key.objectid; | |
10998 | end = start + block_group->key.offset - 1; | |
d4b450cd FM |
10999 | /* |
11000 | * Hold the unused_bg_unpin_mutex lock to avoid racing with | |
11001 | * btrfs_finish_extent_commit(). If we are at transaction N, | |
11002 | * another task might be running finish_extent_commit() for the | |
11003 | * previous transaction N - 1, and have seen a range belonging | |
11004 | * to the block group in freed_extents[] before we were able to | |
11005 | * clear the whole block group range from freed_extents[]. This | |
11006 | * means that task can lookup for the block group after we | |
11007 | * unpinned it from freed_extents[] and removed it, leading to | |
11008 | * a BUG_ON() at btrfs_unpin_extent_range(). | |
11009 | */ | |
11010 | mutex_lock(&fs_info->unused_bg_unpin_mutex); | |
758eb51e | 11011 | ret = clear_extent_bits(&fs_info->freed_extents[0], start, end, |
91166212 | 11012 | EXTENT_DIRTY); |
758eb51e | 11013 | if (ret) { |
d4b450cd | 11014 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
2ff7e61e | 11015 | btrfs_dec_block_group_ro(block_group); |
758eb51e FM |
11016 | goto end_trans; |
11017 | } | |
11018 | ret = clear_extent_bits(&fs_info->freed_extents[1], start, end, | |
91166212 | 11019 | EXTENT_DIRTY); |
758eb51e | 11020 | if (ret) { |
d4b450cd | 11021 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
2ff7e61e | 11022 | btrfs_dec_block_group_ro(block_group); |
758eb51e FM |
11023 | goto end_trans; |
11024 | } | |
d4b450cd | 11025 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
47ab2a6c JB |
11026 | |
11027 | /* Reset pinned so btrfs_put_block_group doesn't complain */ | |
c30666d4 ZL |
11028 | spin_lock(&space_info->lock); |
11029 | spin_lock(&block_group->lock); | |
11030 | ||
480b9b4d | 11031 | update_bytes_pinned(fs_info, space_info, -block_group->pinned); |
c30666d4 | 11032 | space_info->bytes_readonly += block_group->pinned; |
dec59fa3 EL |
11033 | percpu_counter_add_batch(&space_info->total_bytes_pinned, |
11034 | -block_group->pinned, | |
11035 | BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
47ab2a6c JB |
11036 | block_group->pinned = 0; |
11037 | ||
c30666d4 ZL |
11038 | spin_unlock(&block_group->lock); |
11039 | spin_unlock(&space_info->lock); | |
11040 | ||
e33e17ee | 11041 | /* DISCARD can flip during remount */ |
0b246afa | 11042 | trimming = btrfs_test_opt(fs_info, DISCARD); |
e33e17ee JM |
11043 | |
11044 | /* Implicit trim during transaction commit. */ | |
11045 | if (trimming) | |
11046 | btrfs_get_block_group_trimming(block_group); | |
11047 | ||
47ab2a6c JB |
11048 | /* |
11049 | * Btrfs_remove_chunk will abort the transaction if things go | |
11050 | * horribly wrong. | |
11051 | */ | |
97aff912 | 11052 | ret = btrfs_remove_chunk(trans, block_group->key.objectid); |
e33e17ee JM |
11053 | |
11054 | if (ret) { | |
11055 | if (trimming) | |
11056 | btrfs_put_block_group_trimming(block_group); | |
11057 | goto end_trans; | |
11058 | } | |
11059 | ||
11060 | /* | |
11061 | * If we're not mounted with -odiscard, we can just forget | |
11062 | * about this block group. Otherwise we'll need to wait | |
11063 | * until transaction commit to do the actual discard. | |
11064 | */ | |
11065 | if (trimming) { | |
348a0013 FM |
11066 | spin_lock(&fs_info->unused_bgs_lock); |
11067 | /* | |
11068 | * A concurrent scrub might have added us to the list | |
11069 | * fs_info->unused_bgs, so use a list_move operation | |
11070 | * to add the block group to the deleted_bgs list. | |
11071 | */ | |
e33e17ee JM |
11072 | list_move(&block_group->bg_list, |
11073 | &trans->transaction->deleted_bgs); | |
348a0013 | 11074 | spin_unlock(&fs_info->unused_bgs_lock); |
e33e17ee JM |
11075 | btrfs_get_block_group(block_group); |
11076 | } | |
758eb51e | 11077 | end_trans: |
3a45bb20 | 11078 | btrfs_end_transaction(trans); |
47ab2a6c | 11079 | next: |
d5f2e33b | 11080 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
47ab2a6c JB |
11081 | btrfs_put_block_group(block_group); |
11082 | spin_lock(&fs_info->unused_bgs_lock); | |
11083 | } | |
11084 | spin_unlock(&fs_info->unused_bgs_lock); | |
11085 | } | |
11086 | ||
c59021f8 | 11087 | int btrfs_init_space_info(struct btrfs_fs_info *fs_info) |
11088 | { | |
1aba86d6 | 11089 | struct btrfs_super_block *disk_super; |
11090 | u64 features; | |
11091 | u64 flags; | |
11092 | int mixed = 0; | |
c59021f8 | 11093 | int ret; |
11094 | ||
6c41761f | 11095 | disk_super = fs_info->super_copy; |
1aba86d6 | 11096 | if (!btrfs_super_root(disk_super)) |
0dc924c5 | 11097 | return -EINVAL; |
c59021f8 | 11098 | |
1aba86d6 | 11099 | features = btrfs_super_incompat_flags(disk_super); |
11100 | if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
11101 | mixed = 1; | |
c59021f8 | 11102 | |
1aba86d6 | 11103 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
4ca61683 | 11104 | ret = create_space_info(fs_info, flags); |
c59021f8 | 11105 | if (ret) |
1aba86d6 | 11106 | goto out; |
c59021f8 | 11107 | |
1aba86d6 | 11108 | if (mixed) { |
11109 | flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA; | |
4ca61683 | 11110 | ret = create_space_info(fs_info, flags); |
1aba86d6 | 11111 | } else { |
11112 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
4ca61683 | 11113 | ret = create_space_info(fs_info, flags); |
1aba86d6 | 11114 | if (ret) |
11115 | goto out; | |
11116 | ||
11117 | flags = BTRFS_BLOCK_GROUP_DATA; | |
4ca61683 | 11118 | ret = create_space_info(fs_info, flags); |
1aba86d6 | 11119 | } |
11120 | out: | |
c59021f8 | 11121 | return ret; |
11122 | } | |
11123 | ||
2ff7e61e JM |
11124 | int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info, |
11125 | u64 start, u64 end) | |
acce952b | 11126 | { |
2ff7e61e | 11127 | return unpin_extent_range(fs_info, start, end, false); |
acce952b | 11128 | } |
11129 | ||
499f377f JM |
11130 | /* |
11131 | * It used to be that old block groups would be left around forever. | |
11132 | * Iterating over them would be enough to trim unused space. Since we | |
11133 | * now automatically remove them, we also need to iterate over unallocated | |
11134 | * space. | |
11135 | * | |
11136 | * We don't want a transaction for this since the discard may take a | |
11137 | * substantial amount of time. We don't require that a transaction be | |
11138 | * running, but we do need to take a running transaction into account | |
fee7acc3 JM |
11139 | * to ensure that we're not discarding chunks that were released or |
11140 | * allocated in the current transaction. | |
499f377f JM |
11141 | * |
11142 | * Holding the chunks lock will prevent other threads from allocating | |
11143 | * or releasing chunks, but it won't prevent a running transaction | |
11144 | * from committing and releasing the memory that the pending chunks | |
11145 | * list head uses. For that, we need to take a reference to the | |
fee7acc3 JM |
11146 | * transaction and hold the commit root sem. We only need to hold |
11147 | * it while performing the free space search since we have already | |
11148 | * held back allocations. | |
499f377f | 11149 | */ |
8103d10b | 11150 | static int btrfs_trim_free_extents(struct btrfs_device *device, u64 *trimmed) |
499f377f | 11151 | { |
8103d10b | 11152 | u64 start = SZ_1M, len = 0, end = 0; |
499f377f JM |
11153 | int ret; |
11154 | ||
11155 | *trimmed = 0; | |
11156 | ||
0be88e36 JM |
11157 | /* Discard not supported = nothing to do. */ |
11158 | if (!blk_queue_discard(bdev_get_queue(device->bdev))) | |
11159 | return 0; | |
11160 | ||
52042d8e | 11161 | /* Not writable = nothing to do. */ |
ebbede42 | 11162 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
499f377f JM |
11163 | return 0; |
11164 | ||
11165 | /* No free space = nothing to do. */ | |
11166 | if (device->total_bytes <= device->bytes_used) | |
11167 | return 0; | |
11168 | ||
11169 | ret = 0; | |
11170 | ||
11171 | while (1) { | |
fb456252 | 11172 | struct btrfs_fs_info *fs_info = device->fs_info; |
499f377f JM |
11173 | u64 bytes; |
11174 | ||
11175 | ret = mutex_lock_interruptible(&fs_info->chunk_mutex); | |
11176 | if (ret) | |
fee7acc3 | 11177 | break; |
499f377f | 11178 | |
929be17a NB |
11179 | find_first_clear_extent_bit(&device->alloc_state, start, |
11180 | &start, &end, | |
11181 | CHUNK_TRIMMED | CHUNK_ALLOCATED); | |
11182 | /* | |
11183 | * If find_first_clear_extent_bit find a range that spans the | |
11184 | * end of the device it will set end to -1, in this case it's up | |
11185 | * to the caller to trim the value to the size of the device. | |
11186 | */ | |
11187 | end = min(end, device->total_bytes - 1); | |
11188 | len = end - start + 1; | |
499f377f | 11189 | |
929be17a NB |
11190 | /* We didn't find any extents */ |
11191 | if (!len) { | |
499f377f | 11192 | mutex_unlock(&fs_info->chunk_mutex); |
929be17a | 11193 | ret = 0; |
499f377f JM |
11194 | break; |
11195 | } | |
11196 | ||
929be17a NB |
11197 | ret = btrfs_issue_discard(device->bdev, start, len, |
11198 | &bytes); | |
11199 | if (!ret) | |
11200 | set_extent_bits(&device->alloc_state, start, | |
11201 | start + bytes - 1, | |
11202 | CHUNK_TRIMMED); | |
499f377f JM |
11203 | mutex_unlock(&fs_info->chunk_mutex); |
11204 | ||
11205 | if (ret) | |
11206 | break; | |
11207 | ||
11208 | start += len; | |
11209 | *trimmed += bytes; | |
11210 | ||
11211 | if (fatal_signal_pending(current)) { | |
11212 | ret = -ERESTARTSYS; | |
11213 | break; | |
11214 | } | |
11215 | ||
11216 | cond_resched(); | |
11217 | } | |
11218 | ||
11219 | return ret; | |
11220 | } | |
11221 | ||
93bba24d QW |
11222 | /* |
11223 | * Trim the whole filesystem by: | |
11224 | * 1) trimming the free space in each block group | |
11225 | * 2) trimming the unallocated space on each device | |
11226 | * | |
11227 | * This will also continue trimming even if a block group or device encounters | |
11228 | * an error. The return value will be the last error, or 0 if nothing bad | |
11229 | * happens. | |
11230 | */ | |
2ff7e61e | 11231 | int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range) |
f7039b1d | 11232 | { |
f7039b1d | 11233 | struct btrfs_block_group_cache *cache = NULL; |
499f377f JM |
11234 | struct btrfs_device *device; |
11235 | struct list_head *devices; | |
f7039b1d LD |
11236 | u64 group_trimmed; |
11237 | u64 start; | |
11238 | u64 end; | |
11239 | u64 trimmed = 0; | |
93bba24d QW |
11240 | u64 bg_failed = 0; |
11241 | u64 dev_failed = 0; | |
11242 | int bg_ret = 0; | |
11243 | int dev_ret = 0; | |
f7039b1d LD |
11244 | int ret = 0; |
11245 | ||
6ba9fc8e | 11246 | cache = btrfs_lookup_first_block_group(fs_info, range->start); |
f87b7eb8 | 11247 | for (; cache; cache = next_block_group(cache)) { |
f7039b1d LD |
11248 | if (cache->key.objectid >= (range->start + range->len)) { |
11249 | btrfs_put_block_group(cache); | |
11250 | break; | |
11251 | } | |
11252 | ||
11253 | start = max(range->start, cache->key.objectid); | |
11254 | end = min(range->start + range->len, | |
11255 | cache->key.objectid + cache->key.offset); | |
11256 | ||
11257 | if (end - start >= range->minlen) { | |
11258 | if (!block_group_cache_done(cache)) { | |
f6373bf3 | 11259 | ret = cache_block_group(cache, 0); |
1be41b78 | 11260 | if (ret) { |
93bba24d QW |
11261 | bg_failed++; |
11262 | bg_ret = ret; | |
11263 | continue; | |
1be41b78 JB |
11264 | } |
11265 | ret = wait_block_group_cache_done(cache); | |
11266 | if (ret) { | |
93bba24d QW |
11267 | bg_failed++; |
11268 | bg_ret = ret; | |
11269 | continue; | |
1be41b78 | 11270 | } |
f7039b1d LD |
11271 | } |
11272 | ret = btrfs_trim_block_group(cache, | |
11273 | &group_trimmed, | |
11274 | start, | |
11275 | end, | |
11276 | range->minlen); | |
11277 | ||
11278 | trimmed += group_trimmed; | |
11279 | if (ret) { | |
93bba24d QW |
11280 | bg_failed++; |
11281 | bg_ret = ret; | |
11282 | continue; | |
f7039b1d LD |
11283 | } |
11284 | } | |
f7039b1d LD |
11285 | } |
11286 | ||
93bba24d QW |
11287 | if (bg_failed) |
11288 | btrfs_warn(fs_info, | |
11289 | "failed to trim %llu block group(s), last error %d", | |
11290 | bg_failed, bg_ret); | |
0b246afa | 11291 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
d4e329de JM |
11292 | devices = &fs_info->fs_devices->devices; |
11293 | list_for_each_entry(device, devices, dev_list) { | |
8103d10b | 11294 | ret = btrfs_trim_free_extents(device, &group_trimmed); |
93bba24d QW |
11295 | if (ret) { |
11296 | dev_failed++; | |
11297 | dev_ret = ret; | |
499f377f | 11298 | break; |
93bba24d | 11299 | } |
499f377f JM |
11300 | |
11301 | trimmed += group_trimmed; | |
11302 | } | |
0b246afa | 11303 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
499f377f | 11304 | |
93bba24d QW |
11305 | if (dev_failed) |
11306 | btrfs_warn(fs_info, | |
11307 | "failed to trim %llu device(s), last error %d", | |
11308 | dev_failed, dev_ret); | |
f7039b1d | 11309 | range->len = trimmed; |
93bba24d QW |
11310 | if (bg_ret) |
11311 | return bg_ret; | |
11312 | return dev_ret; | |
f7039b1d | 11313 | } |
8257b2dc MX |
11314 | |
11315 | /* | |
ea14b57f | 11316 | * btrfs_{start,end}_write_no_snapshotting() are similar to |
9ea24bbe FM |
11317 | * mnt_{want,drop}_write(), they are used to prevent some tasks from writing |
11318 | * data into the page cache through nocow before the subvolume is snapshoted, | |
11319 | * but flush the data into disk after the snapshot creation, or to prevent | |
ea14b57f | 11320 | * operations while snapshotting is ongoing and that cause the snapshot to be |
9ea24bbe | 11321 | * inconsistent (writes followed by expanding truncates for example). |
8257b2dc | 11322 | */ |
ea14b57f | 11323 | void btrfs_end_write_no_snapshotting(struct btrfs_root *root) |
8257b2dc MX |
11324 | { |
11325 | percpu_counter_dec(&root->subv_writers->counter); | |
093258e6 | 11326 | cond_wake_up(&root->subv_writers->wait); |
8257b2dc MX |
11327 | } |
11328 | ||
ea14b57f | 11329 | int btrfs_start_write_no_snapshotting(struct btrfs_root *root) |
8257b2dc | 11330 | { |
ea14b57f | 11331 | if (atomic_read(&root->will_be_snapshotted)) |
8257b2dc MX |
11332 | return 0; |
11333 | ||
11334 | percpu_counter_inc(&root->subv_writers->counter); | |
11335 | /* | |
11336 | * Make sure counter is updated before we check for snapshot creation. | |
11337 | */ | |
11338 | smp_mb(); | |
ea14b57f DS |
11339 | if (atomic_read(&root->will_be_snapshotted)) { |
11340 | btrfs_end_write_no_snapshotting(root); | |
8257b2dc MX |
11341 | return 0; |
11342 | } | |
11343 | return 1; | |
11344 | } | |
0bc19f90 | 11345 | |
0bc19f90 ZL |
11346 | void btrfs_wait_for_snapshot_creation(struct btrfs_root *root) |
11347 | { | |
11348 | while (true) { | |
11349 | int ret; | |
11350 | ||
ea14b57f | 11351 | ret = btrfs_start_write_no_snapshotting(root); |
0bc19f90 ZL |
11352 | if (ret) |
11353 | break; | |
4625956a PZ |
11354 | wait_var_event(&root->will_be_snapshotted, |
11355 | !atomic_read(&root->will_be_snapshotted)); | |
0bc19f90 ZL |
11356 | } |
11357 | } | |
031f24da QW |
11358 | |
11359 | void btrfs_mark_bg_unused(struct btrfs_block_group_cache *bg) | |
11360 | { | |
11361 | struct btrfs_fs_info *fs_info = bg->fs_info; | |
11362 | ||
11363 | spin_lock(&fs_info->unused_bgs_lock); | |
11364 | if (list_empty(&bg->bg_list)) { | |
11365 | btrfs_get_block_group(bg); | |
11366 | trace_btrfs_add_unused_block_group(bg); | |
11367 | list_add_tail(&bg->bg_list, &fs_info->unused_bgs); | |
11368 | } | |
11369 | spin_unlock(&fs_info->unused_bgs_lock); | |
11370 | } |