]>
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) \ | |
58 | static inline void update_##name(struct btrfs_space_info *sinfo, \ | |
59 | s64 bytes) \ | |
60 | { \ | |
61 | if (bytes < 0 && sinfo->name < -bytes) { \ | |
62 | WARN_ON(1); \ | |
63 | sinfo->name = 0; \ | |
64 | return; \ | |
65 | } \ | |
66 | sinfo->name += bytes; \ | |
67 | } | |
68 | ||
69 | DECLARE_SPACE_INFO_UPDATE(bytes_may_use); | |
e2907c1a | 70 | DECLARE_SPACE_INFO_UPDATE(bytes_pinned); |
9f9b8e8d | 71 | |
5d4f98a2 | 72 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
e72cb923 NB |
73 | struct btrfs_delayed_ref_node *node, u64 parent, |
74 | u64 root_objectid, u64 owner_objectid, | |
75 | u64 owner_offset, int refs_to_drop, | |
76 | struct btrfs_delayed_extent_op *extra_op); | |
5d4f98a2 YZ |
77 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, |
78 | struct extent_buffer *leaf, | |
79 | struct btrfs_extent_item *ei); | |
80 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
81 | u64 parent, u64 root_objectid, |
82 | u64 flags, u64 owner, u64 offset, | |
83 | struct btrfs_key *ins, int ref_mod); | |
84 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, | |
4e6bd4e0 | 85 | struct btrfs_delayed_ref_node *node, |
21ebfbe7 | 86 | struct btrfs_delayed_extent_op *extent_op); |
01458828 | 87 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, u64 flags, |
698d0082 | 88 | int force); |
11833d66 YZ |
89 | static int find_next_key(struct btrfs_path *path, int level, |
90 | struct btrfs_key *key); | |
ab8d0fc4 JM |
91 | static void dump_space_info(struct btrfs_fs_info *fs_info, |
92 | struct btrfs_space_info *info, u64 bytes, | |
9ed74f2d | 93 | int dump_block_groups); |
5d80366e JB |
94 | static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, |
95 | u64 num_bytes); | |
957780eb JB |
96 | static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info, |
97 | struct btrfs_space_info *space_info, | |
98 | u64 num_bytes); | |
99 | static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info, | |
100 | struct btrfs_space_info *space_info, | |
101 | u64 num_bytes); | |
6a63209f | 102 | |
817d52f8 JB |
103 | static noinline int |
104 | block_group_cache_done(struct btrfs_block_group_cache *cache) | |
105 | { | |
106 | smp_mb(); | |
36cce922 JB |
107 | return cache->cached == BTRFS_CACHE_FINISHED || |
108 | cache->cached == BTRFS_CACHE_ERROR; | |
817d52f8 JB |
109 | } |
110 | ||
0f9dd46c JB |
111 | static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits) |
112 | { | |
113 | return (cache->flags & bits) == bits; | |
114 | } | |
115 | ||
758f2dfc | 116 | void btrfs_get_block_group(struct btrfs_block_group_cache *cache) |
11dfe35a JB |
117 | { |
118 | atomic_inc(&cache->count); | |
119 | } | |
120 | ||
121 | void btrfs_put_block_group(struct btrfs_block_group_cache *cache) | |
122 | { | |
f0486c68 YZ |
123 | if (atomic_dec_and_test(&cache->count)) { |
124 | WARN_ON(cache->pinned > 0); | |
125 | WARN_ON(cache->reserved > 0); | |
0966a7b1 QW |
126 | |
127 | /* | |
128 | * If not empty, someone is still holding mutex of | |
129 | * full_stripe_lock, which can only be released by caller. | |
130 | * And it will definitely cause use-after-free when caller | |
131 | * tries to release full stripe lock. | |
132 | * | |
133 | * No better way to resolve, but only to warn. | |
134 | */ | |
135 | WARN_ON(!RB_EMPTY_ROOT(&cache->full_stripe_locks_root.root)); | |
34d52cb6 | 136 | kfree(cache->free_space_ctl); |
11dfe35a | 137 | kfree(cache); |
f0486c68 | 138 | } |
11dfe35a JB |
139 | } |
140 | ||
0f9dd46c JB |
141 | /* |
142 | * this adds the block group to the fs_info rb tree for the block group | |
143 | * cache | |
144 | */ | |
b2950863 | 145 | static int btrfs_add_block_group_cache(struct btrfs_fs_info *info, |
0f9dd46c JB |
146 | struct btrfs_block_group_cache *block_group) |
147 | { | |
148 | struct rb_node **p; | |
149 | struct rb_node *parent = NULL; | |
150 | struct btrfs_block_group_cache *cache; | |
151 | ||
152 | spin_lock(&info->block_group_cache_lock); | |
153 | p = &info->block_group_cache_tree.rb_node; | |
154 | ||
155 | while (*p) { | |
156 | parent = *p; | |
157 | cache = rb_entry(parent, struct btrfs_block_group_cache, | |
158 | cache_node); | |
159 | if (block_group->key.objectid < cache->key.objectid) { | |
160 | p = &(*p)->rb_left; | |
161 | } else if (block_group->key.objectid > cache->key.objectid) { | |
162 | p = &(*p)->rb_right; | |
163 | } else { | |
164 | spin_unlock(&info->block_group_cache_lock); | |
165 | return -EEXIST; | |
166 | } | |
167 | } | |
168 | ||
169 | rb_link_node(&block_group->cache_node, parent, p); | |
170 | rb_insert_color(&block_group->cache_node, | |
171 | &info->block_group_cache_tree); | |
a1897fdd LB |
172 | |
173 | if (info->first_logical_byte > block_group->key.objectid) | |
174 | info->first_logical_byte = block_group->key.objectid; | |
175 | ||
0f9dd46c JB |
176 | spin_unlock(&info->block_group_cache_lock); |
177 | ||
178 | return 0; | |
179 | } | |
180 | ||
181 | /* | |
182 | * This will return the block group at or after bytenr if contains is 0, else | |
183 | * it will return the block group that contains the bytenr | |
184 | */ | |
185 | static struct btrfs_block_group_cache * | |
186 | block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr, | |
187 | int contains) | |
188 | { | |
189 | struct btrfs_block_group_cache *cache, *ret = NULL; | |
190 | struct rb_node *n; | |
191 | u64 end, start; | |
192 | ||
193 | spin_lock(&info->block_group_cache_lock); | |
194 | n = info->block_group_cache_tree.rb_node; | |
195 | ||
196 | while (n) { | |
197 | cache = rb_entry(n, struct btrfs_block_group_cache, | |
198 | cache_node); | |
199 | end = cache->key.objectid + cache->key.offset - 1; | |
200 | start = cache->key.objectid; | |
201 | ||
202 | if (bytenr < start) { | |
203 | if (!contains && (!ret || start < ret->key.objectid)) | |
204 | ret = cache; | |
205 | n = n->rb_left; | |
206 | } else if (bytenr > start) { | |
207 | if (contains && bytenr <= end) { | |
208 | ret = cache; | |
209 | break; | |
210 | } | |
211 | n = n->rb_right; | |
212 | } else { | |
213 | ret = cache; | |
214 | break; | |
215 | } | |
216 | } | |
a1897fdd | 217 | if (ret) { |
11dfe35a | 218 | btrfs_get_block_group(ret); |
a1897fdd LB |
219 | if (bytenr == 0 && info->first_logical_byte > ret->key.objectid) |
220 | info->first_logical_byte = ret->key.objectid; | |
221 | } | |
0f9dd46c JB |
222 | spin_unlock(&info->block_group_cache_lock); |
223 | ||
224 | return ret; | |
225 | } | |
226 | ||
2ff7e61e | 227 | static int add_excluded_extent(struct btrfs_fs_info *fs_info, |
11833d66 | 228 | u64 start, u64 num_bytes) |
817d52f8 | 229 | { |
11833d66 | 230 | u64 end = start + num_bytes - 1; |
0b246afa | 231 | set_extent_bits(&fs_info->freed_extents[0], |
ceeb0ae7 | 232 | start, end, EXTENT_UPTODATE); |
0b246afa | 233 | set_extent_bits(&fs_info->freed_extents[1], |
ceeb0ae7 | 234 | start, end, EXTENT_UPTODATE); |
11833d66 YZ |
235 | return 0; |
236 | } | |
817d52f8 | 237 | |
9e715da8 | 238 | static void free_excluded_extents(struct btrfs_block_group_cache *cache) |
11833d66 | 239 | { |
9e715da8 | 240 | struct btrfs_fs_info *fs_info = cache->fs_info; |
11833d66 | 241 | u64 start, end; |
817d52f8 | 242 | |
11833d66 YZ |
243 | start = cache->key.objectid; |
244 | end = start + cache->key.offset - 1; | |
245 | ||
0b246afa | 246 | clear_extent_bits(&fs_info->freed_extents[0], |
91166212 | 247 | start, end, EXTENT_UPTODATE); |
0b246afa | 248 | clear_extent_bits(&fs_info->freed_extents[1], |
91166212 | 249 | start, end, EXTENT_UPTODATE); |
817d52f8 JB |
250 | } |
251 | ||
3c4da657 | 252 | static int exclude_super_stripes(struct btrfs_block_group_cache *cache) |
817d52f8 | 253 | { |
3c4da657 | 254 | struct btrfs_fs_info *fs_info = cache->fs_info; |
817d52f8 JB |
255 | u64 bytenr; |
256 | u64 *logical; | |
257 | int stripe_len; | |
258 | int i, nr, ret; | |
259 | ||
06b2331f YZ |
260 | if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) { |
261 | stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid; | |
262 | cache->bytes_super += stripe_len; | |
2ff7e61e | 263 | ret = add_excluded_extent(fs_info, cache->key.objectid, |
06b2331f | 264 | stripe_len); |
835d974f JB |
265 | if (ret) |
266 | return ret; | |
06b2331f YZ |
267 | } |
268 | ||
817d52f8 JB |
269 | for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { |
270 | bytenr = btrfs_sb_offset(i); | |
0b246afa | 271 | ret = btrfs_rmap_block(fs_info, cache->key.objectid, |
63a9c7b9 | 272 | bytenr, &logical, &nr, &stripe_len); |
835d974f JB |
273 | if (ret) |
274 | return ret; | |
11833d66 | 275 | |
817d52f8 | 276 | while (nr--) { |
51bf5f0b JB |
277 | u64 start, len; |
278 | ||
279 | if (logical[nr] > cache->key.objectid + | |
280 | cache->key.offset) | |
281 | continue; | |
282 | ||
283 | if (logical[nr] + stripe_len <= cache->key.objectid) | |
284 | continue; | |
285 | ||
286 | start = logical[nr]; | |
287 | if (start < cache->key.objectid) { | |
288 | start = cache->key.objectid; | |
289 | len = (logical[nr] + stripe_len) - start; | |
290 | } else { | |
291 | len = min_t(u64, stripe_len, | |
292 | cache->key.objectid + | |
293 | cache->key.offset - start); | |
294 | } | |
295 | ||
296 | cache->bytes_super += len; | |
2ff7e61e | 297 | ret = add_excluded_extent(fs_info, start, len); |
835d974f JB |
298 | if (ret) { |
299 | kfree(logical); | |
300 | return ret; | |
301 | } | |
817d52f8 | 302 | } |
11833d66 | 303 | |
817d52f8 JB |
304 | kfree(logical); |
305 | } | |
817d52f8 JB |
306 | return 0; |
307 | } | |
308 | ||
11833d66 YZ |
309 | static struct btrfs_caching_control * |
310 | get_caching_control(struct btrfs_block_group_cache *cache) | |
311 | { | |
312 | struct btrfs_caching_control *ctl; | |
313 | ||
314 | spin_lock(&cache->lock); | |
dde5abee JB |
315 | if (!cache->caching_ctl) { |
316 | spin_unlock(&cache->lock); | |
11833d66 YZ |
317 | return NULL; |
318 | } | |
319 | ||
320 | ctl = cache->caching_ctl; | |
1e4f4714 | 321 | refcount_inc(&ctl->count); |
11833d66 YZ |
322 | spin_unlock(&cache->lock); |
323 | return ctl; | |
324 | } | |
325 | ||
326 | static void put_caching_control(struct btrfs_caching_control *ctl) | |
327 | { | |
1e4f4714 | 328 | if (refcount_dec_and_test(&ctl->count)) |
11833d66 YZ |
329 | kfree(ctl); |
330 | } | |
331 | ||
d0bd4560 | 332 | #ifdef CONFIG_BTRFS_DEBUG |
2ff7e61e | 333 | static void fragment_free_space(struct btrfs_block_group_cache *block_group) |
d0bd4560 | 334 | { |
2ff7e61e | 335 | struct btrfs_fs_info *fs_info = block_group->fs_info; |
d0bd4560 JB |
336 | u64 start = block_group->key.objectid; |
337 | u64 len = block_group->key.offset; | |
338 | u64 chunk = block_group->flags & BTRFS_BLOCK_GROUP_METADATA ? | |
0b246afa | 339 | fs_info->nodesize : fs_info->sectorsize; |
d0bd4560 JB |
340 | u64 step = chunk << 1; |
341 | ||
342 | while (len > chunk) { | |
343 | btrfs_remove_free_space(block_group, start, chunk); | |
344 | start += step; | |
345 | if (len < step) | |
346 | len = 0; | |
347 | else | |
348 | len -= step; | |
349 | } | |
350 | } | |
351 | #endif | |
352 | ||
0f9dd46c JB |
353 | /* |
354 | * this is only called by cache_block_group, since we could have freed extents | |
355 | * we need to check the pinned_extents for any extents that can't be used yet | |
356 | * since their free space will be released as soon as the transaction commits. | |
357 | */ | |
a5ed9182 | 358 | u64 add_new_free_space(struct btrfs_block_group_cache *block_group, |
4457c1c7 | 359 | u64 start, u64 end) |
0f9dd46c | 360 | { |
4457c1c7 | 361 | struct btrfs_fs_info *info = block_group->fs_info; |
817d52f8 | 362 | u64 extent_start, extent_end, size, total_added = 0; |
0f9dd46c JB |
363 | int ret; |
364 | ||
365 | while (start < end) { | |
11833d66 | 366 | ret = find_first_extent_bit(info->pinned_extents, start, |
0f9dd46c | 367 | &extent_start, &extent_end, |
e6138876 JB |
368 | EXTENT_DIRTY | EXTENT_UPTODATE, |
369 | NULL); | |
0f9dd46c JB |
370 | if (ret) |
371 | break; | |
372 | ||
06b2331f | 373 | if (extent_start <= start) { |
0f9dd46c JB |
374 | start = extent_end + 1; |
375 | } else if (extent_start > start && extent_start < end) { | |
376 | size = extent_start - start; | |
817d52f8 | 377 | total_added += size; |
ea6a478e JB |
378 | ret = btrfs_add_free_space(block_group, start, |
379 | size); | |
79787eaa | 380 | BUG_ON(ret); /* -ENOMEM or logic error */ |
0f9dd46c JB |
381 | start = extent_end + 1; |
382 | } else { | |
383 | break; | |
384 | } | |
385 | } | |
386 | ||
387 | if (start < end) { | |
388 | size = end - start; | |
817d52f8 | 389 | total_added += size; |
ea6a478e | 390 | ret = btrfs_add_free_space(block_group, start, size); |
79787eaa | 391 | BUG_ON(ret); /* -ENOMEM or logic error */ |
0f9dd46c JB |
392 | } |
393 | ||
817d52f8 | 394 | return total_added; |
0f9dd46c JB |
395 | } |
396 | ||
73fa48b6 | 397 | static int load_extent_tree_free(struct btrfs_caching_control *caching_ctl) |
e37c9e69 | 398 | { |
0b246afa JM |
399 | struct btrfs_block_group_cache *block_group = caching_ctl->block_group; |
400 | struct btrfs_fs_info *fs_info = block_group->fs_info; | |
401 | struct btrfs_root *extent_root = fs_info->extent_root; | |
e37c9e69 | 402 | struct btrfs_path *path; |
5f39d397 | 403 | struct extent_buffer *leaf; |
11833d66 | 404 | struct btrfs_key key; |
817d52f8 | 405 | u64 total_found = 0; |
11833d66 YZ |
406 | u64 last = 0; |
407 | u32 nritems; | |
73fa48b6 | 408 | int ret; |
d0bd4560 | 409 | bool wakeup = true; |
f510cfec | 410 | |
e37c9e69 CM |
411 | path = btrfs_alloc_path(); |
412 | if (!path) | |
73fa48b6 | 413 | return -ENOMEM; |
7d7d6068 | 414 | |
817d52f8 | 415 | last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET); |
11833d66 | 416 | |
d0bd4560 JB |
417 | #ifdef CONFIG_BTRFS_DEBUG |
418 | /* | |
419 | * If we're fragmenting we don't want to make anybody think we can | |
420 | * allocate from this block group until we've had a chance to fragment | |
421 | * the free space. | |
422 | */ | |
2ff7e61e | 423 | if (btrfs_should_fragment_free_space(block_group)) |
d0bd4560 JB |
424 | wakeup = false; |
425 | #endif | |
5cd57b2c | 426 | /* |
817d52f8 JB |
427 | * We don't want to deadlock with somebody trying to allocate a new |
428 | * extent for the extent root while also trying to search the extent | |
429 | * root to add free space. So we skip locking and search the commit | |
430 | * root, since its read-only | |
5cd57b2c CM |
431 | */ |
432 | path->skip_locking = 1; | |
817d52f8 | 433 | path->search_commit_root = 1; |
e4058b54 | 434 | path->reada = READA_FORWARD; |
817d52f8 | 435 | |
e4404d6e | 436 | key.objectid = last; |
e37c9e69 | 437 | key.offset = 0; |
11833d66 | 438 | key.type = BTRFS_EXTENT_ITEM_KEY; |
013f1b12 | 439 | |
52ee28d2 | 440 | next: |
11833d66 | 441 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
e37c9e69 | 442 | if (ret < 0) |
73fa48b6 | 443 | goto out; |
a512bbf8 | 444 | |
11833d66 YZ |
445 | leaf = path->nodes[0]; |
446 | nritems = btrfs_header_nritems(leaf); | |
447 | ||
d397712b | 448 | while (1) { |
7841cb28 | 449 | if (btrfs_fs_closing(fs_info) > 1) { |
f25784b3 | 450 | last = (u64)-1; |
817d52f8 | 451 | break; |
f25784b3 | 452 | } |
817d52f8 | 453 | |
11833d66 YZ |
454 | if (path->slots[0] < nritems) { |
455 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
456 | } else { | |
457 | ret = find_next_key(path, 0, &key); | |
458 | if (ret) | |
e37c9e69 | 459 | break; |
817d52f8 | 460 | |
c9ea7b24 | 461 | if (need_resched() || |
9e351cc8 | 462 | rwsem_is_contended(&fs_info->commit_root_sem)) { |
d0bd4560 JB |
463 | if (wakeup) |
464 | caching_ctl->progress = last; | |
ff5714cc | 465 | btrfs_release_path(path); |
9e351cc8 | 466 | up_read(&fs_info->commit_root_sem); |
589d8ade | 467 | mutex_unlock(&caching_ctl->mutex); |
11833d66 | 468 | cond_resched(); |
73fa48b6 OS |
469 | mutex_lock(&caching_ctl->mutex); |
470 | down_read(&fs_info->commit_root_sem); | |
471 | goto next; | |
589d8ade | 472 | } |
0a3896d0 JB |
473 | |
474 | ret = btrfs_next_leaf(extent_root, path); | |
475 | if (ret < 0) | |
73fa48b6 | 476 | goto out; |
0a3896d0 JB |
477 | if (ret) |
478 | break; | |
589d8ade JB |
479 | leaf = path->nodes[0]; |
480 | nritems = btrfs_header_nritems(leaf); | |
481 | continue; | |
11833d66 | 482 | } |
817d52f8 | 483 | |
52ee28d2 LB |
484 | if (key.objectid < last) { |
485 | key.objectid = last; | |
486 | key.offset = 0; | |
487 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
488 | ||
d0bd4560 JB |
489 | if (wakeup) |
490 | caching_ctl->progress = last; | |
52ee28d2 LB |
491 | btrfs_release_path(path); |
492 | goto next; | |
493 | } | |
494 | ||
11833d66 YZ |
495 | if (key.objectid < block_group->key.objectid) { |
496 | path->slots[0]++; | |
817d52f8 | 497 | continue; |
e37c9e69 | 498 | } |
0f9dd46c | 499 | |
e37c9e69 | 500 | if (key.objectid >= block_group->key.objectid + |
0f9dd46c | 501 | block_group->key.offset) |
e37c9e69 | 502 | break; |
7d7d6068 | 503 | |
3173a18f JB |
504 | if (key.type == BTRFS_EXTENT_ITEM_KEY || |
505 | key.type == BTRFS_METADATA_ITEM_KEY) { | |
4457c1c7 | 506 | total_found += add_new_free_space(block_group, last, |
817d52f8 | 507 | key.objectid); |
3173a18f JB |
508 | if (key.type == BTRFS_METADATA_ITEM_KEY) |
509 | last = key.objectid + | |
da17066c | 510 | fs_info->nodesize; |
3173a18f JB |
511 | else |
512 | last = key.objectid + key.offset; | |
817d52f8 | 513 | |
73fa48b6 | 514 | if (total_found > CACHING_CTL_WAKE_UP) { |
11833d66 | 515 | total_found = 0; |
d0bd4560 JB |
516 | if (wakeup) |
517 | wake_up(&caching_ctl->wait); | |
11833d66 | 518 | } |
817d52f8 | 519 | } |
e37c9e69 CM |
520 | path->slots[0]++; |
521 | } | |
817d52f8 | 522 | ret = 0; |
e37c9e69 | 523 | |
4457c1c7 | 524 | total_found += add_new_free_space(block_group, last, |
817d52f8 JB |
525 | block_group->key.objectid + |
526 | block_group->key.offset); | |
11833d66 | 527 | caching_ctl->progress = (u64)-1; |
817d52f8 | 528 | |
73fa48b6 OS |
529 | out: |
530 | btrfs_free_path(path); | |
531 | return ret; | |
532 | } | |
533 | ||
534 | static noinline void caching_thread(struct btrfs_work *work) | |
535 | { | |
536 | struct btrfs_block_group_cache *block_group; | |
537 | struct btrfs_fs_info *fs_info; | |
538 | struct btrfs_caching_control *caching_ctl; | |
539 | int ret; | |
540 | ||
541 | caching_ctl = container_of(work, struct btrfs_caching_control, work); | |
542 | block_group = caching_ctl->block_group; | |
543 | fs_info = block_group->fs_info; | |
544 | ||
545 | mutex_lock(&caching_ctl->mutex); | |
546 | down_read(&fs_info->commit_root_sem); | |
547 | ||
1e144fb8 OS |
548 | if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) |
549 | ret = load_free_space_tree(caching_ctl); | |
550 | else | |
551 | ret = load_extent_tree_free(caching_ctl); | |
73fa48b6 | 552 | |
817d52f8 | 553 | spin_lock(&block_group->lock); |
11833d66 | 554 | block_group->caching_ctl = NULL; |
73fa48b6 | 555 | block_group->cached = ret ? BTRFS_CACHE_ERROR : BTRFS_CACHE_FINISHED; |
817d52f8 | 556 | spin_unlock(&block_group->lock); |
0f9dd46c | 557 | |
d0bd4560 | 558 | #ifdef CONFIG_BTRFS_DEBUG |
2ff7e61e | 559 | if (btrfs_should_fragment_free_space(block_group)) { |
d0bd4560 JB |
560 | u64 bytes_used; |
561 | ||
562 | spin_lock(&block_group->space_info->lock); | |
563 | spin_lock(&block_group->lock); | |
564 | bytes_used = block_group->key.offset - | |
565 | btrfs_block_group_used(&block_group->item); | |
566 | block_group->space_info->bytes_used += bytes_used >> 1; | |
567 | spin_unlock(&block_group->lock); | |
568 | spin_unlock(&block_group->space_info->lock); | |
2ff7e61e | 569 | fragment_free_space(block_group); |
d0bd4560 JB |
570 | } |
571 | #endif | |
572 | ||
573 | caching_ctl->progress = (u64)-1; | |
11833d66 | 574 | |
9e351cc8 | 575 | up_read(&fs_info->commit_root_sem); |
9e715da8 | 576 | free_excluded_extents(block_group); |
11833d66 | 577 | mutex_unlock(&caching_ctl->mutex); |
73fa48b6 | 578 | |
11833d66 YZ |
579 | wake_up(&caching_ctl->wait); |
580 | ||
581 | put_caching_control(caching_ctl); | |
11dfe35a | 582 | btrfs_put_block_group(block_group); |
817d52f8 JB |
583 | } |
584 | ||
9d66e233 | 585 | static int cache_block_group(struct btrfs_block_group_cache *cache, |
9d66e233 | 586 | int load_cache_only) |
817d52f8 | 587 | { |
291c7d2f | 588 | DEFINE_WAIT(wait); |
11833d66 YZ |
589 | struct btrfs_fs_info *fs_info = cache->fs_info; |
590 | struct btrfs_caching_control *caching_ctl; | |
817d52f8 JB |
591 | int ret = 0; |
592 | ||
291c7d2f | 593 | caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS); |
79787eaa JM |
594 | if (!caching_ctl) |
595 | return -ENOMEM; | |
291c7d2f JB |
596 | |
597 | INIT_LIST_HEAD(&caching_ctl->list); | |
598 | mutex_init(&caching_ctl->mutex); | |
599 | init_waitqueue_head(&caching_ctl->wait); | |
600 | caching_ctl->block_group = cache; | |
601 | caching_ctl->progress = cache->key.objectid; | |
1e4f4714 | 602 | refcount_set(&caching_ctl->count, 1); |
9e0af237 LB |
603 | btrfs_init_work(&caching_ctl->work, btrfs_cache_helper, |
604 | caching_thread, NULL, NULL); | |
291c7d2f JB |
605 | |
606 | spin_lock(&cache->lock); | |
607 | /* | |
608 | * This should be a rare occasion, but this could happen I think in the | |
609 | * case where one thread starts to load the space cache info, and then | |
610 | * some other thread starts a transaction commit which tries to do an | |
611 | * allocation while the other thread is still loading the space cache | |
612 | * info. The previous loop should have kept us from choosing this block | |
613 | * group, but if we've moved to the state where we will wait on caching | |
614 | * block groups we need to first check if we're doing a fast load here, | |
615 | * so we can wait for it to finish, otherwise we could end up allocating | |
616 | * from a block group who's cache gets evicted for one reason or | |
617 | * another. | |
618 | */ | |
619 | while (cache->cached == BTRFS_CACHE_FAST) { | |
620 | struct btrfs_caching_control *ctl; | |
621 | ||
622 | ctl = cache->caching_ctl; | |
1e4f4714 | 623 | refcount_inc(&ctl->count); |
291c7d2f JB |
624 | prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE); |
625 | spin_unlock(&cache->lock); | |
626 | ||
627 | schedule(); | |
628 | ||
629 | finish_wait(&ctl->wait, &wait); | |
630 | put_caching_control(ctl); | |
631 | spin_lock(&cache->lock); | |
632 | } | |
633 | ||
634 | if (cache->cached != BTRFS_CACHE_NO) { | |
635 | spin_unlock(&cache->lock); | |
636 | kfree(caching_ctl); | |
11833d66 | 637 | return 0; |
291c7d2f JB |
638 | } |
639 | WARN_ON(cache->caching_ctl); | |
640 | cache->caching_ctl = caching_ctl; | |
641 | cache->cached = BTRFS_CACHE_FAST; | |
642 | spin_unlock(&cache->lock); | |
11833d66 | 643 | |
d8953d69 | 644 | if (btrfs_test_opt(fs_info, SPACE_CACHE)) { |
cb83b7b8 | 645 | mutex_lock(&caching_ctl->mutex); |
bb6cb1c5 | 646 | ret = load_free_space_cache(cache); |
9d66e233 JB |
647 | |
648 | spin_lock(&cache->lock); | |
649 | if (ret == 1) { | |
291c7d2f | 650 | cache->caching_ctl = NULL; |
9d66e233 JB |
651 | cache->cached = BTRFS_CACHE_FINISHED; |
652 | cache->last_byte_to_unpin = (u64)-1; | |
cb83b7b8 | 653 | caching_ctl->progress = (u64)-1; |
9d66e233 | 654 | } else { |
291c7d2f JB |
655 | if (load_cache_only) { |
656 | cache->caching_ctl = NULL; | |
657 | cache->cached = BTRFS_CACHE_NO; | |
658 | } else { | |
659 | cache->cached = BTRFS_CACHE_STARTED; | |
4f69cb98 | 660 | cache->has_caching_ctl = 1; |
291c7d2f | 661 | } |
9d66e233 JB |
662 | } |
663 | spin_unlock(&cache->lock); | |
d0bd4560 JB |
664 | #ifdef CONFIG_BTRFS_DEBUG |
665 | if (ret == 1 && | |
2ff7e61e | 666 | btrfs_should_fragment_free_space(cache)) { |
d0bd4560 JB |
667 | u64 bytes_used; |
668 | ||
669 | spin_lock(&cache->space_info->lock); | |
670 | spin_lock(&cache->lock); | |
671 | bytes_used = cache->key.offset - | |
672 | btrfs_block_group_used(&cache->item); | |
673 | cache->space_info->bytes_used += bytes_used >> 1; | |
674 | spin_unlock(&cache->lock); | |
675 | spin_unlock(&cache->space_info->lock); | |
2ff7e61e | 676 | fragment_free_space(cache); |
d0bd4560 JB |
677 | } |
678 | #endif | |
cb83b7b8 JB |
679 | mutex_unlock(&caching_ctl->mutex); |
680 | ||
291c7d2f | 681 | wake_up(&caching_ctl->wait); |
3c14874a | 682 | if (ret == 1) { |
291c7d2f | 683 | put_caching_control(caching_ctl); |
9e715da8 | 684 | free_excluded_extents(cache); |
9d66e233 | 685 | return 0; |
3c14874a | 686 | } |
291c7d2f JB |
687 | } else { |
688 | /* | |
1e144fb8 OS |
689 | * We're either using the free space tree or no caching at all. |
690 | * Set cached to the appropriate value and wakeup any waiters. | |
291c7d2f JB |
691 | */ |
692 | spin_lock(&cache->lock); | |
693 | if (load_cache_only) { | |
694 | cache->caching_ctl = NULL; | |
695 | cache->cached = BTRFS_CACHE_NO; | |
696 | } else { | |
697 | cache->cached = BTRFS_CACHE_STARTED; | |
4f69cb98 | 698 | cache->has_caching_ctl = 1; |
291c7d2f JB |
699 | } |
700 | spin_unlock(&cache->lock); | |
701 | wake_up(&caching_ctl->wait); | |
9d66e233 JB |
702 | } |
703 | ||
291c7d2f JB |
704 | if (load_cache_only) { |
705 | put_caching_control(caching_ctl); | |
11833d66 | 706 | return 0; |
817d52f8 | 707 | } |
817d52f8 | 708 | |
9e351cc8 | 709 | down_write(&fs_info->commit_root_sem); |
1e4f4714 | 710 | refcount_inc(&caching_ctl->count); |
11833d66 | 711 | list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups); |
9e351cc8 | 712 | up_write(&fs_info->commit_root_sem); |
11833d66 | 713 | |
11dfe35a | 714 | btrfs_get_block_group(cache); |
11833d66 | 715 | |
e66f0bb1 | 716 | btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work); |
817d52f8 | 717 | |
ef8bbdfe | 718 | return ret; |
e37c9e69 CM |
719 | } |
720 | ||
0f9dd46c JB |
721 | /* |
722 | * return the block group that starts at or after bytenr | |
723 | */ | |
d397712b CM |
724 | static struct btrfs_block_group_cache * |
725 | btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr) | |
0ef3e66b | 726 | { |
e2c89907 | 727 | return block_group_cache_tree_search(info, bytenr, 0); |
0ef3e66b CM |
728 | } |
729 | ||
0f9dd46c | 730 | /* |
9f55684c | 731 | * return the block group that contains the given bytenr |
0f9dd46c | 732 | */ |
d397712b CM |
733 | struct btrfs_block_group_cache *btrfs_lookup_block_group( |
734 | struct btrfs_fs_info *info, | |
735 | u64 bytenr) | |
be744175 | 736 | { |
e2c89907 | 737 | return block_group_cache_tree_search(info, bytenr, 1); |
be744175 | 738 | } |
0b86a832 | 739 | |
0f9dd46c JB |
740 | static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info, |
741 | u64 flags) | |
6324fbf3 | 742 | { |
0f9dd46c | 743 | struct list_head *head = &info->space_info; |
0f9dd46c | 744 | struct btrfs_space_info *found; |
4184ea7f | 745 | |
52ba6929 | 746 | flags &= BTRFS_BLOCK_GROUP_TYPE_MASK; |
b742bb82 | 747 | |
4184ea7f CM |
748 | rcu_read_lock(); |
749 | list_for_each_entry_rcu(found, head, list) { | |
67377734 | 750 | if (found->flags & flags) { |
4184ea7f | 751 | rcu_read_unlock(); |
0f9dd46c | 752 | return found; |
4184ea7f | 753 | } |
0f9dd46c | 754 | } |
4184ea7f | 755 | rcu_read_unlock(); |
0f9dd46c | 756 | return NULL; |
6324fbf3 CM |
757 | } |
758 | ||
ddf30cf0 | 759 | static void add_pinned_bytes(struct btrfs_fs_info *fs_info, |
14ae4ec1 | 760 | struct btrfs_ref *ref, int sign) |
0d9f824d OS |
761 | { |
762 | struct btrfs_space_info *space_info; | |
14ae4ec1 | 763 | s64 num_bytes; |
0d9f824d OS |
764 | u64 flags; |
765 | ||
14ae4ec1 QW |
766 | ASSERT(sign == 1 || sign == -1); |
767 | num_bytes = sign * ref->len; | |
ddf30cf0 QW |
768 | if (ref->type == BTRFS_REF_METADATA) { |
769 | if (ref->tree_ref.root == BTRFS_CHUNK_TREE_OBJECTID) | |
0d9f824d OS |
770 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
771 | else | |
772 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
773 | } else { | |
774 | flags = BTRFS_BLOCK_GROUP_DATA; | |
775 | } | |
776 | ||
777 | space_info = __find_space_info(fs_info, flags); | |
55e8196a | 778 | ASSERT(space_info); |
dec59fa3 EL |
779 | percpu_counter_add_batch(&space_info->total_bytes_pinned, num_bytes, |
780 | BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
0d9f824d OS |
781 | } |
782 | ||
4184ea7f CM |
783 | /* |
784 | * after adding space to the filesystem, we need to clear the full flags | |
785 | * on all the space infos. | |
786 | */ | |
787 | void btrfs_clear_space_info_full(struct btrfs_fs_info *info) | |
788 | { | |
789 | struct list_head *head = &info->space_info; | |
790 | struct btrfs_space_info *found; | |
791 | ||
792 | rcu_read_lock(); | |
793 | list_for_each_entry_rcu(found, head, list) | |
794 | found->full = 0; | |
795 | rcu_read_unlock(); | |
796 | } | |
797 | ||
1a4ed8fd | 798 | /* simple helper to search for an existing data extent at a given offset */ |
2ff7e61e | 799 | int btrfs_lookup_data_extent(struct btrfs_fs_info *fs_info, u64 start, u64 len) |
e02119d5 CM |
800 | { |
801 | int ret; | |
802 | struct btrfs_key key; | |
31840ae1 | 803 | struct btrfs_path *path; |
e02119d5 | 804 | |
31840ae1 | 805 | path = btrfs_alloc_path(); |
d8926bb3 MF |
806 | if (!path) |
807 | return -ENOMEM; | |
808 | ||
e02119d5 CM |
809 | key.objectid = start; |
810 | key.offset = len; | |
3173a18f | 811 | key.type = BTRFS_EXTENT_ITEM_KEY; |
0b246afa | 812 | ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0); |
31840ae1 | 813 | btrfs_free_path(path); |
7bb86316 CM |
814 | return ret; |
815 | } | |
816 | ||
a22285a6 | 817 | /* |
3173a18f | 818 | * helper function to lookup reference count and flags of a tree block. |
a22285a6 YZ |
819 | * |
820 | * the head node for delayed ref is used to store the sum of all the | |
821 | * reference count modifications queued up in the rbtree. the head | |
822 | * node may also store the extent flags to set. This way you can check | |
823 | * to see what the reference count and extent flags would be if all of | |
824 | * the delayed refs are not processed. | |
825 | */ | |
826 | int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans, | |
2ff7e61e | 827 | struct btrfs_fs_info *fs_info, u64 bytenr, |
3173a18f | 828 | u64 offset, int metadata, u64 *refs, u64 *flags) |
a22285a6 YZ |
829 | { |
830 | struct btrfs_delayed_ref_head *head; | |
831 | struct btrfs_delayed_ref_root *delayed_refs; | |
832 | struct btrfs_path *path; | |
833 | struct btrfs_extent_item *ei; | |
834 | struct extent_buffer *leaf; | |
835 | struct btrfs_key key; | |
836 | u32 item_size; | |
837 | u64 num_refs; | |
838 | u64 extent_flags; | |
839 | int ret; | |
840 | ||
3173a18f JB |
841 | /* |
842 | * If we don't have skinny metadata, don't bother doing anything | |
843 | * different | |
844 | */ | |
0b246afa JM |
845 | if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) { |
846 | offset = fs_info->nodesize; | |
3173a18f JB |
847 | metadata = 0; |
848 | } | |
849 | ||
a22285a6 YZ |
850 | path = btrfs_alloc_path(); |
851 | if (!path) | |
852 | return -ENOMEM; | |
853 | ||
a22285a6 YZ |
854 | if (!trans) { |
855 | path->skip_locking = 1; | |
856 | path->search_commit_root = 1; | |
857 | } | |
639eefc8 FDBM |
858 | |
859 | search_again: | |
860 | key.objectid = bytenr; | |
861 | key.offset = offset; | |
862 | if (metadata) | |
863 | key.type = BTRFS_METADATA_ITEM_KEY; | |
864 | else | |
865 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
866 | ||
0b246afa | 867 | ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0); |
a22285a6 YZ |
868 | if (ret < 0) |
869 | goto out_free; | |
870 | ||
3173a18f | 871 | if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) { |
74be9510 FDBM |
872 | if (path->slots[0]) { |
873 | path->slots[0]--; | |
874 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
875 | path->slots[0]); | |
876 | if (key.objectid == bytenr && | |
877 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
0b246afa | 878 | key.offset == fs_info->nodesize) |
74be9510 FDBM |
879 | ret = 0; |
880 | } | |
3173a18f JB |
881 | } |
882 | ||
a22285a6 YZ |
883 | if (ret == 0) { |
884 | leaf = path->nodes[0]; | |
885 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
886 | if (item_size >= sizeof(*ei)) { | |
887 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
888 | struct btrfs_extent_item); | |
889 | num_refs = btrfs_extent_refs(leaf, ei); | |
890 | extent_flags = btrfs_extent_flags(leaf, ei); | |
891 | } else { | |
ba3c2b19 NB |
892 | ret = -EINVAL; |
893 | btrfs_print_v0_err(fs_info); | |
894 | if (trans) | |
895 | btrfs_abort_transaction(trans, ret); | |
896 | else | |
897 | btrfs_handle_fs_error(fs_info, ret, NULL); | |
898 | ||
899 | goto out_free; | |
a22285a6 | 900 | } |
ba3c2b19 | 901 | |
a22285a6 YZ |
902 | BUG_ON(num_refs == 0); |
903 | } else { | |
904 | num_refs = 0; | |
905 | extent_flags = 0; | |
906 | ret = 0; | |
907 | } | |
908 | ||
909 | if (!trans) | |
910 | goto out; | |
911 | ||
912 | delayed_refs = &trans->transaction->delayed_refs; | |
913 | spin_lock(&delayed_refs->lock); | |
f72ad18e | 914 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
a22285a6 YZ |
915 | if (head) { |
916 | if (!mutex_trylock(&head->mutex)) { | |
d278850e | 917 | refcount_inc(&head->refs); |
a22285a6 YZ |
918 | spin_unlock(&delayed_refs->lock); |
919 | ||
b3b4aa74 | 920 | btrfs_release_path(path); |
a22285a6 | 921 | |
8cc33e5c DS |
922 | /* |
923 | * Mutex was contended, block until it's released and try | |
924 | * again | |
925 | */ | |
a22285a6 YZ |
926 | mutex_lock(&head->mutex); |
927 | mutex_unlock(&head->mutex); | |
d278850e | 928 | btrfs_put_delayed_ref_head(head); |
639eefc8 | 929 | goto search_again; |
a22285a6 | 930 | } |
d7df2c79 | 931 | spin_lock(&head->lock); |
a22285a6 YZ |
932 | if (head->extent_op && head->extent_op->update_flags) |
933 | extent_flags |= head->extent_op->flags_to_set; | |
934 | else | |
935 | BUG_ON(num_refs == 0); | |
936 | ||
d278850e | 937 | num_refs += head->ref_mod; |
d7df2c79 | 938 | spin_unlock(&head->lock); |
a22285a6 YZ |
939 | mutex_unlock(&head->mutex); |
940 | } | |
941 | spin_unlock(&delayed_refs->lock); | |
942 | out: | |
943 | WARN_ON(num_refs == 0); | |
944 | if (refs) | |
945 | *refs = num_refs; | |
946 | if (flags) | |
947 | *flags = extent_flags; | |
948 | out_free: | |
949 | btrfs_free_path(path); | |
950 | return ret; | |
951 | } | |
952 | ||
d8d5f3e1 CM |
953 | /* |
954 | * Back reference rules. Back refs have three main goals: | |
955 | * | |
956 | * 1) differentiate between all holders of references to an extent so that | |
957 | * when a reference is dropped we can make sure it was a valid reference | |
958 | * before freeing the extent. | |
959 | * | |
960 | * 2) Provide enough information to quickly find the holders of an extent | |
961 | * if we notice a given block is corrupted or bad. | |
962 | * | |
963 | * 3) Make it easy to migrate blocks for FS shrinking or storage pool | |
964 | * maintenance. This is actually the same as #2, but with a slightly | |
965 | * different use case. | |
966 | * | |
5d4f98a2 YZ |
967 | * There are two kinds of back refs. The implicit back refs is optimized |
968 | * for pointers in non-shared tree blocks. For a given pointer in a block, | |
969 | * back refs of this kind provide information about the block's owner tree | |
970 | * and the pointer's key. These information allow us to find the block by | |
971 | * b-tree searching. The full back refs is for pointers in tree blocks not | |
972 | * referenced by their owner trees. The location of tree block is recorded | |
973 | * in the back refs. Actually the full back refs is generic, and can be | |
974 | * used in all cases the implicit back refs is used. The major shortcoming | |
975 | * of the full back refs is its overhead. Every time a tree block gets | |
976 | * COWed, we have to update back refs entry for all pointers in it. | |
977 | * | |
978 | * For a newly allocated tree block, we use implicit back refs for | |
979 | * pointers in it. This means most tree related operations only involve | |
980 | * implicit back refs. For a tree block created in old transaction, the | |
981 | * only way to drop a reference to it is COW it. So we can detect the | |
982 | * event that tree block loses its owner tree's reference and do the | |
983 | * back refs conversion. | |
984 | * | |
01327610 | 985 | * When a tree block is COWed through a tree, there are four cases: |
5d4f98a2 YZ |
986 | * |
987 | * The reference count of the block is one and the tree is the block's | |
988 | * owner tree. Nothing to do in this case. | |
989 | * | |
990 | * The reference count of the block is one and the tree is not the | |
991 | * block's owner tree. In this case, full back refs is used for pointers | |
992 | * in the block. Remove these full back refs, add implicit back refs for | |
993 | * every pointers in the new block. | |
994 | * | |
995 | * The reference count of the block is greater than one and the tree is | |
996 | * the block's owner tree. In this case, implicit back refs is used for | |
997 | * pointers in the block. Add full back refs for every pointers in the | |
998 | * block, increase lower level extents' reference counts. The original | |
999 | * implicit back refs are entailed to the new block. | |
1000 | * | |
1001 | * The reference count of the block is greater than one and the tree is | |
1002 | * not the block's owner tree. Add implicit back refs for every pointer in | |
1003 | * the new block, increase lower level extents' reference count. | |
1004 | * | |
1005 | * Back Reference Key composing: | |
1006 | * | |
1007 | * The key objectid corresponds to the first byte in the extent, | |
1008 | * The key type is used to differentiate between types of back refs. | |
1009 | * There are different meanings of the key offset for different types | |
1010 | * of back refs. | |
1011 | * | |
d8d5f3e1 CM |
1012 | * File extents can be referenced by: |
1013 | * | |
1014 | * - multiple snapshots, subvolumes, or different generations in one subvol | |
31840ae1 | 1015 | * - different files inside a single subvolume |
d8d5f3e1 CM |
1016 | * - different offsets inside a file (bookend extents in file.c) |
1017 | * | |
5d4f98a2 | 1018 | * The extent ref structure for the implicit back refs has fields for: |
d8d5f3e1 CM |
1019 | * |
1020 | * - Objectid of the subvolume root | |
d8d5f3e1 | 1021 | * - objectid of the file holding the reference |
5d4f98a2 YZ |
1022 | * - original offset in the file |
1023 | * - how many bookend extents | |
d8d5f3e1 | 1024 | * |
5d4f98a2 YZ |
1025 | * The key offset for the implicit back refs is hash of the first |
1026 | * three fields. | |
d8d5f3e1 | 1027 | * |
5d4f98a2 | 1028 | * The extent ref structure for the full back refs has field for: |
d8d5f3e1 | 1029 | * |
5d4f98a2 | 1030 | * - number of pointers in the tree leaf |
d8d5f3e1 | 1031 | * |
5d4f98a2 YZ |
1032 | * The key offset for the implicit back refs is the first byte of |
1033 | * the tree leaf | |
d8d5f3e1 | 1034 | * |
5d4f98a2 YZ |
1035 | * When a file extent is allocated, The implicit back refs is used. |
1036 | * the fields are filled in: | |
d8d5f3e1 | 1037 | * |
5d4f98a2 | 1038 | * (root_key.objectid, inode objectid, offset in file, 1) |
d8d5f3e1 | 1039 | * |
5d4f98a2 YZ |
1040 | * When a file extent is removed file truncation, we find the |
1041 | * corresponding implicit back refs and check the following fields: | |
d8d5f3e1 | 1042 | * |
5d4f98a2 | 1043 | * (btrfs_header_owner(leaf), inode objectid, offset in file) |
d8d5f3e1 | 1044 | * |
5d4f98a2 | 1045 | * Btree extents can be referenced by: |
d8d5f3e1 | 1046 | * |
5d4f98a2 | 1047 | * - Different subvolumes |
d8d5f3e1 | 1048 | * |
5d4f98a2 YZ |
1049 | * Both the implicit back refs and the full back refs for tree blocks |
1050 | * only consist of key. The key offset for the implicit back refs is | |
1051 | * objectid of block's owner tree. The key offset for the full back refs | |
1052 | * is the first byte of parent block. | |
d8d5f3e1 | 1053 | * |
5d4f98a2 YZ |
1054 | * When implicit back refs is used, information about the lowest key and |
1055 | * level of the tree block are required. These information are stored in | |
1056 | * tree block info structure. | |
d8d5f3e1 | 1057 | */ |
31840ae1 | 1058 | |
167ce953 LB |
1059 | /* |
1060 | * is_data == BTRFS_REF_TYPE_BLOCK, tree block type is required, | |
52042d8e | 1061 | * is_data == BTRFS_REF_TYPE_DATA, data type is requiried, |
167ce953 LB |
1062 | * is_data == BTRFS_REF_TYPE_ANY, either type is OK. |
1063 | */ | |
1064 | int btrfs_get_extent_inline_ref_type(const struct extent_buffer *eb, | |
1065 | struct btrfs_extent_inline_ref *iref, | |
1066 | enum btrfs_inline_ref_type is_data) | |
1067 | { | |
1068 | int type = btrfs_extent_inline_ref_type(eb, iref); | |
64ecdb64 | 1069 | u64 offset = btrfs_extent_inline_ref_offset(eb, iref); |
167ce953 LB |
1070 | |
1071 | if (type == BTRFS_TREE_BLOCK_REF_KEY || | |
1072 | type == BTRFS_SHARED_BLOCK_REF_KEY || | |
1073 | type == BTRFS_SHARED_DATA_REF_KEY || | |
1074 | type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1075 | if (is_data == BTRFS_REF_TYPE_BLOCK) { | |
64ecdb64 | 1076 | if (type == BTRFS_TREE_BLOCK_REF_KEY) |
167ce953 | 1077 | return type; |
64ecdb64 LB |
1078 | if (type == BTRFS_SHARED_BLOCK_REF_KEY) { |
1079 | ASSERT(eb->fs_info); | |
1080 | /* | |
1081 | * Every shared one has parent tree | |
1082 | * block, which must be aligned to | |
1083 | * nodesize. | |
1084 | */ | |
1085 | if (offset && | |
1086 | IS_ALIGNED(offset, eb->fs_info->nodesize)) | |
1087 | return type; | |
1088 | } | |
167ce953 | 1089 | } else if (is_data == BTRFS_REF_TYPE_DATA) { |
64ecdb64 | 1090 | if (type == BTRFS_EXTENT_DATA_REF_KEY) |
167ce953 | 1091 | return type; |
64ecdb64 LB |
1092 | if (type == BTRFS_SHARED_DATA_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 LB |
1103 | } else { |
1104 | ASSERT(is_data == BTRFS_REF_TYPE_ANY); | |
1105 | return type; | |
1106 | } | |
1107 | } | |
1108 | ||
1109 | btrfs_print_leaf((struct extent_buffer *)eb); | |
1110 | btrfs_err(eb->fs_info, "eb %llu invalid extent inline ref type %d", | |
1111 | eb->start, type); | |
1112 | WARN_ON(1); | |
1113 | ||
1114 | return BTRFS_REF_TYPE_INVALID; | |
1115 | } | |
1116 | ||
5d4f98a2 YZ |
1117 | static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset) |
1118 | { | |
1119 | u32 high_crc = ~(u32)0; | |
1120 | u32 low_crc = ~(u32)0; | |
1121 | __le64 lenum; | |
1122 | ||
1123 | lenum = cpu_to_le64(root_objectid); | |
9678c543 | 1124 | high_crc = crc32c(high_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 1125 | lenum = cpu_to_le64(owner); |
9678c543 | 1126 | low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 1127 | lenum = cpu_to_le64(offset); |
9678c543 | 1128 | low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 YZ |
1129 | |
1130 | return ((u64)high_crc << 31) ^ (u64)low_crc; | |
1131 | } | |
1132 | ||
1133 | static u64 hash_extent_data_ref_item(struct extent_buffer *leaf, | |
1134 | struct btrfs_extent_data_ref *ref) | |
1135 | { | |
1136 | return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref), | |
1137 | btrfs_extent_data_ref_objectid(leaf, ref), | |
1138 | btrfs_extent_data_ref_offset(leaf, ref)); | |
1139 | } | |
1140 | ||
1141 | static int match_extent_data_ref(struct extent_buffer *leaf, | |
1142 | struct btrfs_extent_data_ref *ref, | |
1143 | u64 root_objectid, u64 owner, u64 offset) | |
1144 | { | |
1145 | if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid || | |
1146 | btrfs_extent_data_ref_objectid(leaf, ref) != owner || | |
1147 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
1148 | return 0; | |
1149 | return 1; | |
1150 | } | |
1151 | ||
1152 | static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
1153 | struct btrfs_path *path, |
1154 | u64 bytenr, u64 parent, | |
1155 | u64 root_objectid, | |
1156 | u64 owner, u64 offset) | |
1157 | { | |
bd1d53ef | 1158 | struct btrfs_root *root = trans->fs_info->extent_root; |
5d4f98a2 YZ |
1159 | struct btrfs_key key; |
1160 | struct btrfs_extent_data_ref *ref; | |
31840ae1 | 1161 | struct extent_buffer *leaf; |
5d4f98a2 | 1162 | u32 nritems; |
74493f7a | 1163 | int ret; |
5d4f98a2 YZ |
1164 | int recow; |
1165 | int err = -ENOENT; | |
74493f7a | 1166 | |
31840ae1 | 1167 | key.objectid = bytenr; |
5d4f98a2 YZ |
1168 | if (parent) { |
1169 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1170 | key.offset = parent; | |
1171 | } else { | |
1172 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1173 | key.offset = hash_extent_data_ref(root_objectid, | |
1174 | owner, offset); | |
1175 | } | |
1176 | again: | |
1177 | recow = 0; | |
1178 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1179 | if (ret < 0) { | |
1180 | err = ret; | |
1181 | goto fail; | |
1182 | } | |
31840ae1 | 1183 | |
5d4f98a2 YZ |
1184 | if (parent) { |
1185 | if (!ret) | |
1186 | return 0; | |
5d4f98a2 | 1187 | goto fail; |
31840ae1 ZY |
1188 | } |
1189 | ||
1190 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1191 | nritems = btrfs_header_nritems(leaf); |
1192 | while (1) { | |
1193 | if (path->slots[0] >= nritems) { | |
1194 | ret = btrfs_next_leaf(root, path); | |
1195 | if (ret < 0) | |
1196 | err = ret; | |
1197 | if (ret) | |
1198 | goto fail; | |
1199 | ||
1200 | leaf = path->nodes[0]; | |
1201 | nritems = btrfs_header_nritems(leaf); | |
1202 | recow = 1; | |
1203 | } | |
1204 | ||
1205 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1206 | if (key.objectid != bytenr || | |
1207 | key.type != BTRFS_EXTENT_DATA_REF_KEY) | |
1208 | goto fail; | |
1209 | ||
1210 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1211 | struct btrfs_extent_data_ref); | |
1212 | ||
1213 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1214 | owner, offset)) { | |
1215 | if (recow) { | |
b3b4aa74 | 1216 | btrfs_release_path(path); |
5d4f98a2 YZ |
1217 | goto again; |
1218 | } | |
1219 | err = 0; | |
1220 | break; | |
1221 | } | |
1222 | path->slots[0]++; | |
31840ae1 | 1223 | } |
5d4f98a2 YZ |
1224 | fail: |
1225 | return err; | |
31840ae1 ZY |
1226 | } |
1227 | ||
5d4f98a2 | 1228 | static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1229 | struct btrfs_path *path, |
1230 | u64 bytenr, u64 parent, | |
1231 | u64 root_objectid, u64 owner, | |
1232 | u64 offset, int refs_to_add) | |
31840ae1 | 1233 | { |
62b895af | 1234 | struct btrfs_root *root = trans->fs_info->extent_root; |
31840ae1 ZY |
1235 | struct btrfs_key key; |
1236 | struct extent_buffer *leaf; | |
5d4f98a2 | 1237 | u32 size; |
31840ae1 ZY |
1238 | u32 num_refs; |
1239 | int ret; | |
74493f7a | 1240 | |
74493f7a | 1241 | key.objectid = bytenr; |
5d4f98a2 YZ |
1242 | if (parent) { |
1243 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1244 | key.offset = parent; | |
1245 | size = sizeof(struct btrfs_shared_data_ref); | |
1246 | } else { | |
1247 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1248 | key.offset = hash_extent_data_ref(root_objectid, | |
1249 | owner, offset); | |
1250 | size = sizeof(struct btrfs_extent_data_ref); | |
1251 | } | |
74493f7a | 1252 | |
5d4f98a2 YZ |
1253 | ret = btrfs_insert_empty_item(trans, root, path, &key, size); |
1254 | if (ret && ret != -EEXIST) | |
1255 | goto fail; | |
1256 | ||
1257 | leaf = path->nodes[0]; | |
1258 | if (parent) { | |
1259 | struct btrfs_shared_data_ref *ref; | |
31840ae1 | 1260 | ref = btrfs_item_ptr(leaf, path->slots[0], |
5d4f98a2 YZ |
1261 | struct btrfs_shared_data_ref); |
1262 | if (ret == 0) { | |
1263 | btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add); | |
1264 | } else { | |
1265 | num_refs = btrfs_shared_data_ref_count(leaf, ref); | |
1266 | num_refs += refs_to_add; | |
1267 | btrfs_set_shared_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1268 | } |
5d4f98a2 YZ |
1269 | } else { |
1270 | struct btrfs_extent_data_ref *ref; | |
1271 | while (ret == -EEXIST) { | |
1272 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1273 | struct btrfs_extent_data_ref); | |
1274 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1275 | owner, offset)) | |
1276 | break; | |
b3b4aa74 | 1277 | btrfs_release_path(path); |
5d4f98a2 YZ |
1278 | key.offset++; |
1279 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1280 | size); | |
1281 | if (ret && ret != -EEXIST) | |
1282 | goto fail; | |
31840ae1 | 1283 | |
5d4f98a2 YZ |
1284 | leaf = path->nodes[0]; |
1285 | } | |
1286 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1287 | struct btrfs_extent_data_ref); | |
1288 | if (ret == 0) { | |
1289 | btrfs_set_extent_data_ref_root(leaf, ref, | |
1290 | root_objectid); | |
1291 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
1292 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
1293 | btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add); | |
1294 | } else { | |
1295 | num_refs = btrfs_extent_data_ref_count(leaf, ref); | |
1296 | num_refs += refs_to_add; | |
1297 | btrfs_set_extent_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1298 | } |
31840ae1 | 1299 | } |
5d4f98a2 YZ |
1300 | btrfs_mark_buffer_dirty(leaf); |
1301 | ret = 0; | |
1302 | fail: | |
b3b4aa74 | 1303 | btrfs_release_path(path); |
7bb86316 | 1304 | return ret; |
74493f7a CM |
1305 | } |
1306 | ||
5d4f98a2 | 1307 | static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 | 1308 | struct btrfs_path *path, |
fcebe456 | 1309 | int refs_to_drop, int *last_ref) |
31840ae1 | 1310 | { |
5d4f98a2 YZ |
1311 | struct btrfs_key key; |
1312 | struct btrfs_extent_data_ref *ref1 = NULL; | |
1313 | struct btrfs_shared_data_ref *ref2 = NULL; | |
31840ae1 | 1314 | struct extent_buffer *leaf; |
5d4f98a2 | 1315 | u32 num_refs = 0; |
31840ae1 ZY |
1316 | int ret = 0; |
1317 | ||
1318 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1319 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
1320 | ||
1321 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1322 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1323 | struct btrfs_extent_data_ref); | |
1324 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1325 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1326 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1327 | struct btrfs_shared_data_ref); | |
1328 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
6d8ff4e4 | 1329 | } else if (unlikely(key.type == BTRFS_EXTENT_REF_V0_KEY)) { |
ba3c2b19 NB |
1330 | btrfs_print_v0_err(trans->fs_info); |
1331 | btrfs_abort_transaction(trans, -EINVAL); | |
1332 | return -EINVAL; | |
5d4f98a2 YZ |
1333 | } else { |
1334 | BUG(); | |
1335 | } | |
1336 | ||
56bec294 CM |
1337 | BUG_ON(num_refs < refs_to_drop); |
1338 | num_refs -= refs_to_drop; | |
5d4f98a2 | 1339 | |
31840ae1 | 1340 | if (num_refs == 0) { |
e9f6290d | 1341 | ret = btrfs_del_item(trans, trans->fs_info->extent_root, path); |
fcebe456 | 1342 | *last_ref = 1; |
31840ae1 | 1343 | } else { |
5d4f98a2 YZ |
1344 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) |
1345 | btrfs_set_extent_data_ref_count(leaf, ref1, num_refs); | |
1346 | else if (key.type == BTRFS_SHARED_DATA_REF_KEY) | |
1347 | btrfs_set_shared_data_ref_count(leaf, ref2, num_refs); | |
31840ae1 ZY |
1348 | btrfs_mark_buffer_dirty(leaf); |
1349 | } | |
31840ae1 ZY |
1350 | return ret; |
1351 | } | |
1352 | ||
9ed0dea0 | 1353 | static noinline u32 extent_data_ref_count(struct btrfs_path *path, |
5d4f98a2 | 1354 | struct btrfs_extent_inline_ref *iref) |
15916de8 | 1355 | { |
5d4f98a2 YZ |
1356 | struct btrfs_key key; |
1357 | struct extent_buffer *leaf; | |
1358 | struct btrfs_extent_data_ref *ref1; | |
1359 | struct btrfs_shared_data_ref *ref2; | |
1360 | u32 num_refs = 0; | |
3de28d57 | 1361 | int type; |
5d4f98a2 YZ |
1362 | |
1363 | leaf = path->nodes[0]; | |
1364 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
ba3c2b19 NB |
1365 | |
1366 | BUG_ON(key.type == BTRFS_EXTENT_REF_V0_KEY); | |
5d4f98a2 | 1367 | if (iref) { |
3de28d57 LB |
1368 | /* |
1369 | * If type is invalid, we should have bailed out earlier than | |
1370 | * this call. | |
1371 | */ | |
1372 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA); | |
1373 | ASSERT(type != BTRFS_REF_TYPE_INVALID); | |
1374 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
5d4f98a2 YZ |
1375 | ref1 = (struct btrfs_extent_data_ref *)(&iref->offset); |
1376 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1377 | } else { | |
1378 | ref2 = (struct btrfs_shared_data_ref *)(iref + 1); | |
1379 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1380 | } | |
1381 | } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1382 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1383 | struct btrfs_extent_data_ref); | |
1384 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1385 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1386 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1387 | struct btrfs_shared_data_ref); | |
1388 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
5d4f98a2 YZ |
1389 | } else { |
1390 | WARN_ON(1); | |
1391 | } | |
1392 | return num_refs; | |
1393 | } | |
15916de8 | 1394 | |
5d4f98a2 | 1395 | static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1396 | struct btrfs_path *path, |
1397 | u64 bytenr, u64 parent, | |
1398 | u64 root_objectid) | |
1f3c79a2 | 1399 | { |
b8582eea | 1400 | struct btrfs_root *root = trans->fs_info->extent_root; |
5d4f98a2 | 1401 | struct btrfs_key key; |
1f3c79a2 | 1402 | int ret; |
1f3c79a2 | 1403 | |
5d4f98a2 YZ |
1404 | key.objectid = bytenr; |
1405 | if (parent) { | |
1406 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1407 | key.offset = parent; | |
1408 | } else { | |
1409 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1410 | key.offset = root_objectid; | |
1f3c79a2 LH |
1411 | } |
1412 | ||
5d4f98a2 YZ |
1413 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1414 | if (ret > 0) | |
1415 | ret = -ENOENT; | |
5d4f98a2 | 1416 | return ret; |
1f3c79a2 LH |
1417 | } |
1418 | ||
5d4f98a2 | 1419 | static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1420 | struct btrfs_path *path, |
1421 | u64 bytenr, u64 parent, | |
1422 | u64 root_objectid) | |
31840ae1 | 1423 | { |
5d4f98a2 | 1424 | struct btrfs_key key; |
31840ae1 | 1425 | int ret; |
31840ae1 | 1426 | |
5d4f98a2 YZ |
1427 | key.objectid = bytenr; |
1428 | if (parent) { | |
1429 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1430 | key.offset = parent; | |
1431 | } else { | |
1432 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1433 | key.offset = root_objectid; | |
1434 | } | |
1435 | ||
10728404 | 1436 | ret = btrfs_insert_empty_item(trans, trans->fs_info->extent_root, |
87bde3cd | 1437 | path, &key, 0); |
b3b4aa74 | 1438 | btrfs_release_path(path); |
31840ae1 ZY |
1439 | return ret; |
1440 | } | |
1441 | ||
5d4f98a2 | 1442 | static inline int extent_ref_type(u64 parent, u64 owner) |
31840ae1 | 1443 | { |
5d4f98a2 YZ |
1444 | int type; |
1445 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1446 | if (parent > 0) | |
1447 | type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1448 | else | |
1449 | type = BTRFS_TREE_BLOCK_REF_KEY; | |
1450 | } else { | |
1451 | if (parent > 0) | |
1452 | type = BTRFS_SHARED_DATA_REF_KEY; | |
1453 | else | |
1454 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
1455 | } | |
1456 | return type; | |
31840ae1 | 1457 | } |
56bec294 | 1458 | |
2c47e605 YZ |
1459 | static int find_next_key(struct btrfs_path *path, int level, |
1460 | struct btrfs_key *key) | |
56bec294 | 1461 | |
02217ed2 | 1462 | { |
2c47e605 | 1463 | for (; level < BTRFS_MAX_LEVEL; level++) { |
5d4f98a2 YZ |
1464 | if (!path->nodes[level]) |
1465 | break; | |
5d4f98a2 YZ |
1466 | if (path->slots[level] + 1 >= |
1467 | btrfs_header_nritems(path->nodes[level])) | |
1468 | continue; | |
1469 | if (level == 0) | |
1470 | btrfs_item_key_to_cpu(path->nodes[level], key, | |
1471 | path->slots[level] + 1); | |
1472 | else | |
1473 | btrfs_node_key_to_cpu(path->nodes[level], key, | |
1474 | path->slots[level] + 1); | |
1475 | return 0; | |
1476 | } | |
1477 | return 1; | |
1478 | } | |
037e6390 | 1479 | |
5d4f98a2 YZ |
1480 | /* |
1481 | * look for inline back ref. if back ref is found, *ref_ret is set | |
1482 | * to the address of inline back ref, and 0 is returned. | |
1483 | * | |
1484 | * if back ref isn't found, *ref_ret is set to the address where it | |
1485 | * should be inserted, and -ENOENT is returned. | |
1486 | * | |
1487 | * if insert is true and there are too many inline back refs, the path | |
1488 | * points to the extent item, and -EAGAIN is returned. | |
1489 | * | |
1490 | * NOTE: inline back refs are ordered in the same way that back ref | |
1491 | * items in the tree are ordered. | |
1492 | */ | |
1493 | static noinline_for_stack | |
1494 | int lookup_inline_extent_backref(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
1495 | struct btrfs_path *path, |
1496 | struct btrfs_extent_inline_ref **ref_ret, | |
1497 | u64 bytenr, u64 num_bytes, | |
1498 | u64 parent, u64 root_objectid, | |
1499 | u64 owner, u64 offset, int insert) | |
1500 | { | |
867cc1fb | 1501 | struct btrfs_fs_info *fs_info = trans->fs_info; |
87bde3cd | 1502 | struct btrfs_root *root = fs_info->extent_root; |
5d4f98a2 YZ |
1503 | struct btrfs_key key; |
1504 | struct extent_buffer *leaf; | |
1505 | struct btrfs_extent_item *ei; | |
1506 | struct btrfs_extent_inline_ref *iref; | |
1507 | u64 flags; | |
1508 | u64 item_size; | |
1509 | unsigned long ptr; | |
1510 | unsigned long end; | |
1511 | int extra_size; | |
1512 | int type; | |
1513 | int want; | |
1514 | int ret; | |
1515 | int err = 0; | |
0b246afa | 1516 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); |
3de28d57 | 1517 | int needed; |
26b8003f | 1518 | |
db94535d | 1519 | key.objectid = bytenr; |
31840ae1 | 1520 | key.type = BTRFS_EXTENT_ITEM_KEY; |
56bec294 | 1521 | key.offset = num_bytes; |
31840ae1 | 1522 | |
5d4f98a2 YZ |
1523 | want = extent_ref_type(parent, owner); |
1524 | if (insert) { | |
1525 | extra_size = btrfs_extent_inline_ref_size(want); | |
85d4198e | 1526 | path->keep_locks = 1; |
5d4f98a2 YZ |
1527 | } else |
1528 | extra_size = -1; | |
3173a18f JB |
1529 | |
1530 | /* | |
16d1c062 NB |
1531 | * Owner is our level, so we can just add one to get the level for the |
1532 | * block we are interested in. | |
3173a18f JB |
1533 | */ |
1534 | if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1535 | key.type = BTRFS_METADATA_ITEM_KEY; | |
1536 | key.offset = owner; | |
1537 | } | |
1538 | ||
1539 | again: | |
5d4f98a2 | 1540 | ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1); |
b9473439 | 1541 | if (ret < 0) { |
5d4f98a2 YZ |
1542 | err = ret; |
1543 | goto out; | |
1544 | } | |
3173a18f JB |
1545 | |
1546 | /* | |
1547 | * We may be a newly converted file system which still has the old fat | |
1548 | * extent entries for metadata, so try and see if we have one of those. | |
1549 | */ | |
1550 | if (ret > 0 && skinny_metadata) { | |
1551 | skinny_metadata = false; | |
1552 | if (path->slots[0]) { | |
1553 | path->slots[0]--; | |
1554 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
1555 | path->slots[0]); | |
1556 | if (key.objectid == bytenr && | |
1557 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
1558 | key.offset == num_bytes) | |
1559 | ret = 0; | |
1560 | } | |
1561 | if (ret) { | |
9ce49a0b | 1562 | key.objectid = bytenr; |
3173a18f JB |
1563 | key.type = BTRFS_EXTENT_ITEM_KEY; |
1564 | key.offset = num_bytes; | |
1565 | btrfs_release_path(path); | |
1566 | goto again; | |
1567 | } | |
1568 | } | |
1569 | ||
79787eaa JM |
1570 | if (ret && !insert) { |
1571 | err = -ENOENT; | |
1572 | goto out; | |
fae7f21c | 1573 | } else if (WARN_ON(ret)) { |
492104c8 | 1574 | err = -EIO; |
492104c8 | 1575 | goto out; |
79787eaa | 1576 | } |
5d4f98a2 YZ |
1577 | |
1578 | leaf = path->nodes[0]; | |
1579 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
6d8ff4e4 | 1580 | if (unlikely(item_size < sizeof(*ei))) { |
ba3c2b19 NB |
1581 | err = -EINVAL; |
1582 | btrfs_print_v0_err(fs_info); | |
1583 | btrfs_abort_transaction(trans, err); | |
1584 | goto out; | |
1585 | } | |
5d4f98a2 | 1586 | |
5d4f98a2 YZ |
1587 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1588 | flags = btrfs_extent_flags(leaf, ei); | |
1589 | ||
1590 | ptr = (unsigned long)(ei + 1); | |
1591 | end = (unsigned long)ei + item_size; | |
1592 | ||
3173a18f | 1593 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) { |
5d4f98a2 YZ |
1594 | ptr += sizeof(struct btrfs_tree_block_info); |
1595 | BUG_ON(ptr > end); | |
5d4f98a2 YZ |
1596 | } |
1597 | ||
3de28d57 LB |
1598 | if (owner >= BTRFS_FIRST_FREE_OBJECTID) |
1599 | needed = BTRFS_REF_TYPE_DATA; | |
1600 | else | |
1601 | needed = BTRFS_REF_TYPE_BLOCK; | |
1602 | ||
5d4f98a2 YZ |
1603 | err = -ENOENT; |
1604 | while (1) { | |
1605 | if (ptr >= end) { | |
1606 | WARN_ON(ptr > end); | |
1607 | break; | |
1608 | } | |
1609 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
3de28d57 LB |
1610 | type = btrfs_get_extent_inline_ref_type(leaf, iref, needed); |
1611 | if (type == BTRFS_REF_TYPE_INVALID) { | |
af431dcb | 1612 | err = -EUCLEAN; |
3de28d57 LB |
1613 | goto out; |
1614 | } | |
1615 | ||
5d4f98a2 YZ |
1616 | if (want < type) |
1617 | break; | |
1618 | if (want > type) { | |
1619 | ptr += btrfs_extent_inline_ref_size(type); | |
1620 | continue; | |
1621 | } | |
1622 | ||
1623 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1624 | struct btrfs_extent_data_ref *dref; | |
1625 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1626 | if (match_extent_data_ref(leaf, dref, root_objectid, | |
1627 | owner, offset)) { | |
1628 | err = 0; | |
1629 | break; | |
1630 | } | |
1631 | if (hash_extent_data_ref_item(leaf, dref) < | |
1632 | hash_extent_data_ref(root_objectid, owner, offset)) | |
1633 | break; | |
1634 | } else { | |
1635 | u64 ref_offset; | |
1636 | ref_offset = btrfs_extent_inline_ref_offset(leaf, iref); | |
1637 | if (parent > 0) { | |
1638 | if (parent == ref_offset) { | |
1639 | err = 0; | |
1640 | break; | |
1641 | } | |
1642 | if (ref_offset < parent) | |
1643 | break; | |
1644 | } else { | |
1645 | if (root_objectid == ref_offset) { | |
1646 | err = 0; | |
1647 | break; | |
1648 | } | |
1649 | if (ref_offset < root_objectid) | |
1650 | break; | |
1651 | } | |
1652 | } | |
1653 | ptr += btrfs_extent_inline_ref_size(type); | |
1654 | } | |
1655 | if (err == -ENOENT && insert) { | |
1656 | if (item_size + extra_size >= | |
1657 | BTRFS_MAX_EXTENT_ITEM_SIZE(root)) { | |
1658 | err = -EAGAIN; | |
1659 | goto out; | |
1660 | } | |
1661 | /* | |
1662 | * To add new inline back ref, we have to make sure | |
1663 | * there is no corresponding back ref item. | |
1664 | * For simplicity, we just do not add new inline back | |
1665 | * ref if there is any kind of item for this block | |
1666 | */ | |
2c47e605 YZ |
1667 | if (find_next_key(path, 0, &key) == 0 && |
1668 | key.objectid == bytenr && | |
85d4198e | 1669 | key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) { |
5d4f98a2 YZ |
1670 | err = -EAGAIN; |
1671 | goto out; | |
1672 | } | |
1673 | } | |
1674 | *ref_ret = (struct btrfs_extent_inline_ref *)ptr; | |
1675 | out: | |
85d4198e | 1676 | if (insert) { |
5d4f98a2 YZ |
1677 | path->keep_locks = 0; |
1678 | btrfs_unlock_up_safe(path, 1); | |
1679 | } | |
1680 | return err; | |
1681 | } | |
1682 | ||
1683 | /* | |
1684 | * helper to add new inline back ref | |
1685 | */ | |
1686 | static noinline_for_stack | |
87bde3cd | 1687 | void setup_inline_extent_backref(struct btrfs_fs_info *fs_info, |
143bede5 JM |
1688 | struct btrfs_path *path, |
1689 | struct btrfs_extent_inline_ref *iref, | |
1690 | u64 parent, u64 root_objectid, | |
1691 | u64 owner, u64 offset, int refs_to_add, | |
1692 | struct btrfs_delayed_extent_op *extent_op) | |
5d4f98a2 YZ |
1693 | { |
1694 | struct extent_buffer *leaf; | |
1695 | struct btrfs_extent_item *ei; | |
1696 | unsigned long ptr; | |
1697 | unsigned long end; | |
1698 | unsigned long item_offset; | |
1699 | u64 refs; | |
1700 | int size; | |
1701 | int type; | |
5d4f98a2 YZ |
1702 | |
1703 | leaf = path->nodes[0]; | |
1704 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1705 | item_offset = (unsigned long)iref - (unsigned long)ei; | |
1706 | ||
1707 | type = extent_ref_type(parent, owner); | |
1708 | size = btrfs_extent_inline_ref_size(type); | |
1709 | ||
c71dd880 | 1710 | btrfs_extend_item(path, size); |
5d4f98a2 YZ |
1711 | |
1712 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1713 | refs = btrfs_extent_refs(leaf, ei); | |
1714 | refs += refs_to_add; | |
1715 | btrfs_set_extent_refs(leaf, ei, refs); | |
1716 | if (extent_op) | |
1717 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1718 | ||
1719 | ptr = (unsigned long)ei + item_offset; | |
1720 | end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]); | |
1721 | if (ptr < end - size) | |
1722 | memmove_extent_buffer(leaf, ptr + size, ptr, | |
1723 | end - size - ptr); | |
1724 | ||
1725 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1726 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
1727 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1728 | struct btrfs_extent_data_ref *dref; | |
1729 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1730 | btrfs_set_extent_data_ref_root(leaf, dref, root_objectid); | |
1731 | btrfs_set_extent_data_ref_objectid(leaf, dref, owner); | |
1732 | btrfs_set_extent_data_ref_offset(leaf, dref, offset); | |
1733 | btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add); | |
1734 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1735 | struct btrfs_shared_data_ref *sref; | |
1736 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1737 | btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add); | |
1738 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1739 | } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) { | |
1740 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1741 | } else { | |
1742 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
1743 | } | |
1744 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1745 | } |
1746 | ||
1747 | static int lookup_extent_backref(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
1748 | struct btrfs_path *path, |
1749 | struct btrfs_extent_inline_ref **ref_ret, | |
1750 | u64 bytenr, u64 num_bytes, u64 parent, | |
1751 | u64 root_objectid, u64 owner, u64 offset) | |
1752 | { | |
1753 | int ret; | |
1754 | ||
867cc1fb NB |
1755 | ret = lookup_inline_extent_backref(trans, path, ref_ret, bytenr, |
1756 | num_bytes, parent, root_objectid, | |
1757 | owner, offset, 0); | |
5d4f98a2 | 1758 | if (ret != -ENOENT) |
54aa1f4d | 1759 | return ret; |
5d4f98a2 | 1760 | |
b3b4aa74 | 1761 | btrfs_release_path(path); |
5d4f98a2 YZ |
1762 | *ref_ret = NULL; |
1763 | ||
1764 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
b8582eea NB |
1765 | ret = lookup_tree_block_ref(trans, path, bytenr, parent, |
1766 | root_objectid); | |
5d4f98a2 | 1767 | } else { |
bd1d53ef NB |
1768 | ret = lookup_extent_data_ref(trans, path, bytenr, parent, |
1769 | root_objectid, owner, offset); | |
b9473439 | 1770 | } |
5d4f98a2 YZ |
1771 | return ret; |
1772 | } | |
31840ae1 | 1773 | |
5d4f98a2 YZ |
1774 | /* |
1775 | * helper to update/remove inline back ref | |
1776 | */ | |
1777 | static noinline_for_stack | |
61a18f1c | 1778 | void update_inline_extent_backref(struct btrfs_path *path, |
143bede5 JM |
1779 | struct btrfs_extent_inline_ref *iref, |
1780 | int refs_to_mod, | |
fcebe456 JB |
1781 | struct btrfs_delayed_extent_op *extent_op, |
1782 | int *last_ref) | |
5d4f98a2 | 1783 | { |
61a18f1c | 1784 | struct extent_buffer *leaf = path->nodes[0]; |
5d4f98a2 YZ |
1785 | struct btrfs_extent_item *ei; |
1786 | struct btrfs_extent_data_ref *dref = NULL; | |
1787 | struct btrfs_shared_data_ref *sref = NULL; | |
1788 | unsigned long ptr; | |
1789 | unsigned long end; | |
1790 | u32 item_size; | |
1791 | int size; | |
1792 | int type; | |
5d4f98a2 YZ |
1793 | u64 refs; |
1794 | ||
5d4f98a2 YZ |
1795 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1796 | refs = btrfs_extent_refs(leaf, ei); | |
1797 | WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0); | |
1798 | refs += refs_to_mod; | |
1799 | btrfs_set_extent_refs(leaf, ei, refs); | |
1800 | if (extent_op) | |
1801 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1802 | ||
3de28d57 LB |
1803 | /* |
1804 | * If type is invalid, we should have bailed out after | |
1805 | * lookup_inline_extent_backref(). | |
1806 | */ | |
1807 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_ANY); | |
1808 | ASSERT(type != BTRFS_REF_TYPE_INVALID); | |
5d4f98a2 YZ |
1809 | |
1810 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1811 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1812 | refs = btrfs_extent_data_ref_count(leaf, dref); | |
1813 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1814 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1815 | refs = btrfs_shared_data_ref_count(leaf, sref); | |
1816 | } else { | |
1817 | refs = 1; | |
1818 | BUG_ON(refs_to_mod != -1); | |
56bec294 | 1819 | } |
31840ae1 | 1820 | |
5d4f98a2 YZ |
1821 | BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod); |
1822 | refs += refs_to_mod; | |
1823 | ||
1824 | if (refs > 0) { | |
1825 | if (type == BTRFS_EXTENT_DATA_REF_KEY) | |
1826 | btrfs_set_extent_data_ref_count(leaf, dref, refs); | |
1827 | else | |
1828 | btrfs_set_shared_data_ref_count(leaf, sref, refs); | |
1829 | } else { | |
fcebe456 | 1830 | *last_ref = 1; |
5d4f98a2 YZ |
1831 | size = btrfs_extent_inline_ref_size(type); |
1832 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1833 | ptr = (unsigned long)iref; | |
1834 | end = (unsigned long)ei + item_size; | |
1835 | if (ptr + size < end) | |
1836 | memmove_extent_buffer(leaf, ptr, ptr + size, | |
1837 | end - ptr - size); | |
1838 | item_size -= size; | |
78ac4f9e | 1839 | btrfs_truncate_item(path, item_size, 1); |
5d4f98a2 YZ |
1840 | } |
1841 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1842 | } |
1843 | ||
1844 | static noinline_for_stack | |
1845 | int insert_inline_extent_backref(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
1846 | struct btrfs_path *path, |
1847 | u64 bytenr, u64 num_bytes, u64 parent, | |
1848 | u64 root_objectid, u64 owner, | |
1849 | u64 offset, int refs_to_add, | |
1850 | struct btrfs_delayed_extent_op *extent_op) | |
1851 | { | |
1852 | struct btrfs_extent_inline_ref *iref; | |
1853 | int ret; | |
1854 | ||
867cc1fb NB |
1855 | ret = lookup_inline_extent_backref(trans, path, &iref, bytenr, |
1856 | num_bytes, parent, root_objectid, | |
1857 | owner, offset, 1); | |
5d4f98a2 YZ |
1858 | if (ret == 0) { |
1859 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID); | |
61a18f1c NB |
1860 | update_inline_extent_backref(path, iref, refs_to_add, |
1861 | extent_op, NULL); | |
5d4f98a2 | 1862 | } else if (ret == -ENOENT) { |
a639cdeb | 1863 | setup_inline_extent_backref(trans->fs_info, path, iref, parent, |
143bede5 JM |
1864 | root_objectid, owner, offset, |
1865 | refs_to_add, extent_op); | |
1866 | ret = 0; | |
771ed689 | 1867 | } |
5d4f98a2 YZ |
1868 | return ret; |
1869 | } | |
31840ae1 | 1870 | |
5d4f98a2 | 1871 | static int insert_extent_backref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1872 | struct btrfs_path *path, |
1873 | u64 bytenr, u64 parent, u64 root_objectid, | |
1874 | u64 owner, u64 offset, int refs_to_add) | |
1875 | { | |
1876 | int ret; | |
1877 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1878 | BUG_ON(refs_to_add != 1); | |
10728404 NB |
1879 | ret = insert_tree_block_ref(trans, path, bytenr, parent, |
1880 | root_objectid); | |
5d4f98a2 | 1881 | } else { |
62b895af NB |
1882 | ret = insert_extent_data_ref(trans, path, bytenr, parent, |
1883 | root_objectid, owner, offset, | |
1884 | refs_to_add); | |
5d4f98a2 YZ |
1885 | } |
1886 | return ret; | |
1887 | } | |
56bec294 | 1888 | |
5d4f98a2 | 1889 | static int remove_extent_backref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1890 | struct btrfs_path *path, |
1891 | struct btrfs_extent_inline_ref *iref, | |
fcebe456 | 1892 | int refs_to_drop, int is_data, int *last_ref) |
5d4f98a2 | 1893 | { |
143bede5 | 1894 | int ret = 0; |
b9473439 | 1895 | |
5d4f98a2 YZ |
1896 | BUG_ON(!is_data && refs_to_drop != 1); |
1897 | if (iref) { | |
61a18f1c NB |
1898 | update_inline_extent_backref(path, iref, -refs_to_drop, NULL, |
1899 | last_ref); | |
5d4f98a2 | 1900 | } else if (is_data) { |
e9f6290d | 1901 | ret = remove_extent_data_ref(trans, path, refs_to_drop, |
fcebe456 | 1902 | last_ref); |
5d4f98a2 | 1903 | } else { |
fcebe456 | 1904 | *last_ref = 1; |
87cc7a8a | 1905 | ret = btrfs_del_item(trans, trans->fs_info->extent_root, path); |
5d4f98a2 YZ |
1906 | } |
1907 | return ret; | |
1908 | } | |
1909 | ||
d04c6b88 JM |
1910 | static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len, |
1911 | u64 *discarded_bytes) | |
5d4f98a2 | 1912 | { |
86557861 JM |
1913 | int j, ret = 0; |
1914 | u64 bytes_left, end; | |
4d89d377 | 1915 | u64 aligned_start = ALIGN(start, 1 << 9); |
d04c6b88 | 1916 | |
4d89d377 JM |
1917 | if (WARN_ON(start != aligned_start)) { |
1918 | len -= aligned_start - start; | |
1919 | len = round_down(len, 1 << 9); | |
1920 | start = aligned_start; | |
1921 | } | |
d04c6b88 | 1922 | |
4d89d377 | 1923 | *discarded_bytes = 0; |
86557861 JM |
1924 | |
1925 | if (!len) | |
1926 | return 0; | |
1927 | ||
1928 | end = start + len; | |
1929 | bytes_left = len; | |
1930 | ||
1931 | /* Skip any superblocks on this device. */ | |
1932 | for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) { | |
1933 | u64 sb_start = btrfs_sb_offset(j); | |
1934 | u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE; | |
1935 | u64 size = sb_start - start; | |
1936 | ||
1937 | if (!in_range(sb_start, start, bytes_left) && | |
1938 | !in_range(sb_end, start, bytes_left) && | |
1939 | !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE)) | |
1940 | continue; | |
1941 | ||
1942 | /* | |
1943 | * Superblock spans beginning of range. Adjust start and | |
1944 | * try again. | |
1945 | */ | |
1946 | if (sb_start <= start) { | |
1947 | start += sb_end - start; | |
1948 | if (start > end) { | |
1949 | bytes_left = 0; | |
1950 | break; | |
1951 | } | |
1952 | bytes_left = end - start; | |
1953 | continue; | |
1954 | } | |
1955 | ||
1956 | if (size) { | |
1957 | ret = blkdev_issue_discard(bdev, start >> 9, size >> 9, | |
1958 | GFP_NOFS, 0); | |
1959 | if (!ret) | |
1960 | *discarded_bytes += size; | |
1961 | else if (ret != -EOPNOTSUPP) | |
1962 | return ret; | |
1963 | } | |
1964 | ||
1965 | start = sb_end; | |
1966 | if (start > end) { | |
1967 | bytes_left = 0; | |
1968 | break; | |
1969 | } | |
1970 | bytes_left = end - start; | |
1971 | } | |
1972 | ||
1973 | if (bytes_left) { | |
1974 | ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9, | |
4d89d377 JM |
1975 | GFP_NOFS, 0); |
1976 | if (!ret) | |
86557861 | 1977 | *discarded_bytes += bytes_left; |
4d89d377 | 1978 | } |
d04c6b88 | 1979 | return ret; |
5d4f98a2 | 1980 | } |
5d4f98a2 | 1981 | |
2ff7e61e | 1982 | int btrfs_discard_extent(struct btrfs_fs_info *fs_info, u64 bytenr, |
1edb647b | 1983 | u64 num_bytes, u64 *actual_bytes) |
5d4f98a2 | 1984 | { |
5d4f98a2 | 1985 | int ret; |
5378e607 | 1986 | u64 discarded_bytes = 0; |
a1d3c478 | 1987 | struct btrfs_bio *bbio = NULL; |
5d4f98a2 | 1988 | |
e244a0ae | 1989 | |
2999241d FM |
1990 | /* |
1991 | * Avoid races with device replace and make sure our bbio has devices | |
1992 | * associated to its stripes that don't go away while we are discarding. | |
1993 | */ | |
0b246afa | 1994 | btrfs_bio_counter_inc_blocked(fs_info); |
5d4f98a2 | 1995 | /* Tell the block device(s) that the sectors can be discarded */ |
0b246afa JM |
1996 | ret = btrfs_map_block(fs_info, BTRFS_MAP_DISCARD, bytenr, &num_bytes, |
1997 | &bbio, 0); | |
79787eaa | 1998 | /* Error condition is -ENOMEM */ |
5d4f98a2 | 1999 | if (!ret) { |
a1d3c478 | 2000 | struct btrfs_bio_stripe *stripe = bbio->stripes; |
5d4f98a2 YZ |
2001 | int i; |
2002 | ||
5d4f98a2 | 2003 | |
a1d3c478 | 2004 | for (i = 0; i < bbio->num_stripes; i++, stripe++) { |
d04c6b88 | 2005 | u64 bytes; |
38b5f68e AJ |
2006 | struct request_queue *req_q; |
2007 | ||
627e0873 FM |
2008 | if (!stripe->dev->bdev) { |
2009 | ASSERT(btrfs_test_opt(fs_info, DEGRADED)); | |
2010 | continue; | |
2011 | } | |
38b5f68e AJ |
2012 | req_q = bdev_get_queue(stripe->dev->bdev); |
2013 | if (!blk_queue_discard(req_q)) | |
d5e2003c JB |
2014 | continue; |
2015 | ||
5378e607 LD |
2016 | ret = btrfs_issue_discard(stripe->dev->bdev, |
2017 | stripe->physical, | |
d04c6b88 JM |
2018 | stripe->length, |
2019 | &bytes); | |
5378e607 | 2020 | if (!ret) |
d04c6b88 | 2021 | discarded_bytes += bytes; |
5378e607 | 2022 | else if (ret != -EOPNOTSUPP) |
79787eaa | 2023 | break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */ |
d5e2003c JB |
2024 | |
2025 | /* | |
2026 | * Just in case we get back EOPNOTSUPP for some reason, | |
2027 | * just ignore the return value so we don't screw up | |
2028 | * people calling discard_extent. | |
2029 | */ | |
2030 | ret = 0; | |
5d4f98a2 | 2031 | } |
6e9606d2 | 2032 | btrfs_put_bbio(bbio); |
5d4f98a2 | 2033 | } |
0b246afa | 2034 | btrfs_bio_counter_dec(fs_info); |
5378e607 LD |
2035 | |
2036 | if (actual_bytes) | |
2037 | *actual_bytes = discarded_bytes; | |
2038 | ||
5d4f98a2 | 2039 | |
53b381b3 DW |
2040 | if (ret == -EOPNOTSUPP) |
2041 | ret = 0; | |
5d4f98a2 | 2042 | return ret; |
5d4f98a2 YZ |
2043 | } |
2044 | ||
79787eaa | 2045 | /* Can return -ENOMEM */ |
5d4f98a2 | 2046 | int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
82fa113f | 2047 | struct btrfs_ref *generic_ref) |
5d4f98a2 | 2048 | { |
82fa113f | 2049 | struct btrfs_fs_info *fs_info = trans->fs_info; |
d7eae340 | 2050 | int old_ref_mod, new_ref_mod; |
5d4f98a2 | 2051 | int ret; |
66d7e7f0 | 2052 | |
82fa113f QW |
2053 | ASSERT(generic_ref->type != BTRFS_REF_NOT_SET && |
2054 | generic_ref->action); | |
2055 | BUG_ON(generic_ref->type == BTRFS_REF_METADATA && | |
2056 | generic_ref->tree_ref.root == BTRFS_TREE_LOG_OBJECTID); | |
5d4f98a2 | 2057 | |
82fa113f QW |
2058 | if (generic_ref->type == BTRFS_REF_METADATA) |
2059 | ret = btrfs_add_delayed_tree_ref(trans, generic_ref, | |
ed4f255b | 2060 | NULL, &old_ref_mod, &new_ref_mod); |
82fa113f QW |
2061 | else |
2062 | ret = btrfs_add_delayed_data_ref(trans, generic_ref, 0, | |
d7eae340 | 2063 | &old_ref_mod, &new_ref_mod); |
d7eae340 | 2064 | |
82fa113f | 2065 | btrfs_ref_tree_mod(fs_info, generic_ref); |
8a5040f7 | 2066 | |
ddf30cf0 | 2067 | if (ret == 0 && old_ref_mod < 0 && new_ref_mod >= 0) |
14ae4ec1 | 2068 | add_pinned_bytes(fs_info, generic_ref, -1); |
d7eae340 | 2069 | |
5d4f98a2 YZ |
2070 | return ret; |
2071 | } | |
2072 | ||
bd3c685e NB |
2073 | /* |
2074 | * __btrfs_inc_extent_ref - insert backreference for a given extent | |
2075 | * | |
2076 | * @trans: Handle of transaction | |
2077 | * | |
2078 | * @node: The delayed ref node used to get the bytenr/length for | |
2079 | * extent whose references are incremented. | |
2080 | * | |
2081 | * @parent: If this is a shared extent (BTRFS_SHARED_DATA_REF_KEY/ | |
2082 | * BTRFS_SHARED_BLOCK_REF_KEY) then it holds the logical | |
2083 | * bytenr of the parent block. Since new extents are always | |
2084 | * created with indirect references, this will only be the case | |
2085 | * when relocating a shared extent. In that case, root_objectid | |
2086 | * will be BTRFS_TREE_RELOC_OBJECTID. Otheriwse, parent must | |
2087 | * be 0 | |
2088 | * | |
2089 | * @root_objectid: The id of the root where this modification has originated, | |
2090 | * this can be either one of the well-known metadata trees or | |
2091 | * the subvolume id which references this extent. | |
2092 | * | |
2093 | * @owner: For data extents it is the inode number of the owning file. | |
2094 | * For metadata extents this parameter holds the level in the | |
2095 | * tree of the extent. | |
2096 | * | |
2097 | * @offset: For metadata extents the offset is ignored and is currently | |
2098 | * always passed as 0. For data extents it is the fileoffset | |
2099 | * this extent belongs to. | |
2100 | * | |
2101 | * @refs_to_add Number of references to add | |
2102 | * | |
2103 | * @extent_op Pointer to a structure, holding information necessary when | |
2104 | * updating a tree block's flags | |
2105 | * | |
2106 | */ | |
5d4f98a2 | 2107 | static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
c682f9b3 | 2108 | struct btrfs_delayed_ref_node *node, |
5d4f98a2 YZ |
2109 | u64 parent, u64 root_objectid, |
2110 | u64 owner, u64 offset, int refs_to_add, | |
2111 | struct btrfs_delayed_extent_op *extent_op) | |
2112 | { | |
2113 | struct btrfs_path *path; | |
2114 | struct extent_buffer *leaf; | |
2115 | struct btrfs_extent_item *item; | |
fcebe456 | 2116 | struct btrfs_key key; |
c682f9b3 QW |
2117 | u64 bytenr = node->bytenr; |
2118 | u64 num_bytes = node->num_bytes; | |
5d4f98a2 YZ |
2119 | u64 refs; |
2120 | int ret; | |
5d4f98a2 YZ |
2121 | |
2122 | path = btrfs_alloc_path(); | |
2123 | if (!path) | |
2124 | return -ENOMEM; | |
2125 | ||
e4058b54 | 2126 | path->reada = READA_FORWARD; |
5d4f98a2 YZ |
2127 | path->leave_spinning = 1; |
2128 | /* this will setup the path even if it fails to insert the back ref */ | |
a639cdeb NB |
2129 | ret = insert_inline_extent_backref(trans, path, bytenr, num_bytes, |
2130 | parent, root_objectid, owner, | |
2131 | offset, refs_to_add, extent_op); | |
0ed4792a | 2132 | if ((ret < 0 && ret != -EAGAIN) || !ret) |
5d4f98a2 | 2133 | goto out; |
fcebe456 JB |
2134 | |
2135 | /* | |
2136 | * Ok we had -EAGAIN which means we didn't have space to insert and | |
2137 | * inline extent ref, so just update the reference count and add a | |
2138 | * normal backref. | |
2139 | */ | |
5d4f98a2 | 2140 | leaf = path->nodes[0]; |
fcebe456 | 2141 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
5d4f98a2 YZ |
2142 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
2143 | refs = btrfs_extent_refs(leaf, item); | |
2144 | btrfs_set_extent_refs(leaf, item, refs + refs_to_add); | |
2145 | if (extent_op) | |
2146 | __run_delayed_extent_op(extent_op, leaf, item); | |
56bec294 | 2147 | |
5d4f98a2 | 2148 | btrfs_mark_buffer_dirty(leaf); |
b3b4aa74 | 2149 | btrfs_release_path(path); |
56bec294 | 2150 | |
e4058b54 | 2151 | path->reada = READA_FORWARD; |
b9473439 | 2152 | path->leave_spinning = 1; |
56bec294 | 2153 | /* now insert the actual backref */ |
37593410 NB |
2154 | ret = insert_extent_backref(trans, path, bytenr, parent, root_objectid, |
2155 | owner, offset, refs_to_add); | |
79787eaa | 2156 | if (ret) |
66642832 | 2157 | btrfs_abort_transaction(trans, ret); |
5d4f98a2 | 2158 | out: |
56bec294 | 2159 | btrfs_free_path(path); |
30d133fc | 2160 | return ret; |
56bec294 CM |
2161 | } |
2162 | ||
5d4f98a2 | 2163 | static int run_delayed_data_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
2164 | struct btrfs_delayed_ref_node *node, |
2165 | struct btrfs_delayed_extent_op *extent_op, | |
2166 | int insert_reserved) | |
56bec294 | 2167 | { |
5d4f98a2 YZ |
2168 | int ret = 0; |
2169 | struct btrfs_delayed_data_ref *ref; | |
2170 | struct btrfs_key ins; | |
2171 | u64 parent = 0; | |
2172 | u64 ref_root = 0; | |
2173 | u64 flags = 0; | |
2174 | ||
2175 | ins.objectid = node->bytenr; | |
2176 | ins.offset = node->num_bytes; | |
2177 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
2178 | ||
2179 | ref = btrfs_delayed_node_to_data_ref(node); | |
2bf98ef3 | 2180 | trace_run_delayed_data_ref(trans->fs_info, node, ref, node->action); |
599c75ec | 2181 | |
5d4f98a2 YZ |
2182 | if (node->type == BTRFS_SHARED_DATA_REF_KEY) |
2183 | parent = ref->parent; | |
fcebe456 | 2184 | ref_root = ref->root; |
5d4f98a2 YZ |
2185 | |
2186 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
3173a18f | 2187 | if (extent_op) |
5d4f98a2 | 2188 | flags |= extent_op->flags_to_set; |
ef89b824 NB |
2189 | ret = alloc_reserved_file_extent(trans, parent, ref_root, |
2190 | flags, ref->objectid, | |
2191 | ref->offset, &ins, | |
2192 | node->ref_mod); | |
5d4f98a2 | 2193 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2590d0f1 NB |
2194 | ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root, |
2195 | ref->objectid, ref->offset, | |
2196 | node->ref_mod, extent_op); | |
5d4f98a2 | 2197 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
e72cb923 | 2198 | ret = __btrfs_free_extent(trans, node, parent, |
5d4f98a2 YZ |
2199 | ref_root, ref->objectid, |
2200 | ref->offset, node->ref_mod, | |
c682f9b3 | 2201 | extent_op); |
5d4f98a2 YZ |
2202 | } else { |
2203 | BUG(); | |
2204 | } | |
2205 | return ret; | |
2206 | } | |
2207 | ||
2208 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
2209 | struct extent_buffer *leaf, | |
2210 | struct btrfs_extent_item *ei) | |
2211 | { | |
2212 | u64 flags = btrfs_extent_flags(leaf, ei); | |
2213 | if (extent_op->update_flags) { | |
2214 | flags |= extent_op->flags_to_set; | |
2215 | btrfs_set_extent_flags(leaf, ei, flags); | |
2216 | } | |
2217 | ||
2218 | if (extent_op->update_key) { | |
2219 | struct btrfs_tree_block_info *bi; | |
2220 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)); | |
2221 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
2222 | btrfs_set_tree_block_key(leaf, bi, &extent_op->key); | |
2223 | } | |
2224 | } | |
2225 | ||
2226 | static int run_delayed_extent_op(struct btrfs_trans_handle *trans, | |
d278850e | 2227 | struct btrfs_delayed_ref_head *head, |
5d4f98a2 YZ |
2228 | struct btrfs_delayed_extent_op *extent_op) |
2229 | { | |
20b9a2d6 | 2230 | struct btrfs_fs_info *fs_info = trans->fs_info; |
5d4f98a2 YZ |
2231 | struct btrfs_key key; |
2232 | struct btrfs_path *path; | |
2233 | struct btrfs_extent_item *ei; | |
2234 | struct extent_buffer *leaf; | |
2235 | u32 item_size; | |
56bec294 | 2236 | int ret; |
5d4f98a2 | 2237 | int err = 0; |
b1c79e09 | 2238 | int metadata = !extent_op->is_data; |
5d4f98a2 | 2239 | |
79787eaa JM |
2240 | if (trans->aborted) |
2241 | return 0; | |
2242 | ||
0b246afa | 2243 | if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) |
3173a18f JB |
2244 | metadata = 0; |
2245 | ||
5d4f98a2 YZ |
2246 | path = btrfs_alloc_path(); |
2247 | if (!path) | |
2248 | return -ENOMEM; | |
2249 | ||
d278850e | 2250 | key.objectid = head->bytenr; |
5d4f98a2 | 2251 | |
3173a18f | 2252 | if (metadata) { |
3173a18f | 2253 | key.type = BTRFS_METADATA_ITEM_KEY; |
b1c79e09 | 2254 | key.offset = extent_op->level; |
3173a18f JB |
2255 | } else { |
2256 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
d278850e | 2257 | key.offset = head->num_bytes; |
3173a18f JB |
2258 | } |
2259 | ||
2260 | again: | |
e4058b54 | 2261 | path->reada = READA_FORWARD; |
5d4f98a2 | 2262 | path->leave_spinning = 1; |
0b246afa | 2263 | ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 1); |
5d4f98a2 YZ |
2264 | if (ret < 0) { |
2265 | err = ret; | |
2266 | goto out; | |
2267 | } | |
2268 | if (ret > 0) { | |
3173a18f | 2269 | if (metadata) { |
55994887 FDBM |
2270 | if (path->slots[0] > 0) { |
2271 | path->slots[0]--; | |
2272 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
2273 | path->slots[0]); | |
d278850e | 2274 | if (key.objectid == head->bytenr && |
55994887 | 2275 | key.type == BTRFS_EXTENT_ITEM_KEY && |
d278850e | 2276 | key.offset == head->num_bytes) |
55994887 FDBM |
2277 | ret = 0; |
2278 | } | |
2279 | if (ret > 0) { | |
2280 | btrfs_release_path(path); | |
2281 | metadata = 0; | |
3173a18f | 2282 | |
d278850e JB |
2283 | key.objectid = head->bytenr; |
2284 | key.offset = head->num_bytes; | |
55994887 FDBM |
2285 | key.type = BTRFS_EXTENT_ITEM_KEY; |
2286 | goto again; | |
2287 | } | |
2288 | } else { | |
2289 | err = -EIO; | |
2290 | goto out; | |
3173a18f | 2291 | } |
5d4f98a2 YZ |
2292 | } |
2293 | ||
2294 | leaf = path->nodes[0]; | |
2295 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
ba3c2b19 | 2296 | |
6d8ff4e4 | 2297 | if (unlikely(item_size < sizeof(*ei))) { |
ba3c2b19 NB |
2298 | err = -EINVAL; |
2299 | btrfs_print_v0_err(fs_info); | |
2300 | btrfs_abort_transaction(trans, err); | |
2301 | goto out; | |
2302 | } | |
2303 | ||
5d4f98a2 YZ |
2304 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
2305 | __run_delayed_extent_op(extent_op, leaf, ei); | |
56bec294 | 2306 | |
5d4f98a2 YZ |
2307 | btrfs_mark_buffer_dirty(leaf); |
2308 | out: | |
2309 | btrfs_free_path(path); | |
2310 | return err; | |
56bec294 CM |
2311 | } |
2312 | ||
5d4f98a2 | 2313 | static int run_delayed_tree_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
2314 | struct btrfs_delayed_ref_node *node, |
2315 | struct btrfs_delayed_extent_op *extent_op, | |
2316 | int insert_reserved) | |
56bec294 CM |
2317 | { |
2318 | int ret = 0; | |
5d4f98a2 | 2319 | struct btrfs_delayed_tree_ref *ref; |
5d4f98a2 YZ |
2320 | u64 parent = 0; |
2321 | u64 ref_root = 0; | |
56bec294 | 2322 | |
5d4f98a2 | 2323 | ref = btrfs_delayed_node_to_tree_ref(node); |
f97806f2 | 2324 | trace_run_delayed_tree_ref(trans->fs_info, node, ref, node->action); |
599c75ec | 2325 | |
5d4f98a2 YZ |
2326 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) |
2327 | parent = ref->parent; | |
fcebe456 | 2328 | ref_root = ref->root; |
5d4f98a2 | 2329 | |
02794222 | 2330 | if (node->ref_mod != 1) { |
f97806f2 | 2331 | btrfs_err(trans->fs_info, |
02794222 LB |
2332 | "btree block(%llu) has %d references rather than 1: action %d ref_root %llu parent %llu", |
2333 | node->bytenr, node->ref_mod, node->action, ref_root, | |
2334 | parent); | |
2335 | return -EIO; | |
2336 | } | |
5d4f98a2 | 2337 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { |
3173a18f | 2338 | BUG_ON(!extent_op || !extent_op->update_flags); |
21ebfbe7 | 2339 | ret = alloc_reserved_tree_block(trans, node, extent_op); |
5d4f98a2 | 2340 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2590d0f1 NB |
2341 | ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root, |
2342 | ref->level, 0, 1, extent_op); | |
5d4f98a2 | 2343 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
e72cb923 | 2344 | ret = __btrfs_free_extent(trans, node, parent, ref_root, |
c682f9b3 | 2345 | ref->level, 0, 1, extent_op); |
5d4f98a2 YZ |
2346 | } else { |
2347 | BUG(); | |
2348 | } | |
56bec294 CM |
2349 | return ret; |
2350 | } | |
2351 | ||
2352 | /* helper function to actually process a single delayed ref entry */ | |
5d4f98a2 | 2353 | static int run_one_delayed_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
2354 | struct btrfs_delayed_ref_node *node, |
2355 | struct btrfs_delayed_extent_op *extent_op, | |
2356 | int insert_reserved) | |
56bec294 | 2357 | { |
79787eaa JM |
2358 | int ret = 0; |
2359 | ||
857cc2fc JB |
2360 | if (trans->aborted) { |
2361 | if (insert_reserved) | |
5fac7f9e | 2362 | btrfs_pin_extent(trans->fs_info, node->bytenr, |
857cc2fc | 2363 | node->num_bytes, 1); |
79787eaa | 2364 | return 0; |
857cc2fc | 2365 | } |
79787eaa | 2366 | |
5d4f98a2 YZ |
2367 | if (node->type == BTRFS_TREE_BLOCK_REF_KEY || |
2368 | node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
f97806f2 | 2369 | ret = run_delayed_tree_ref(trans, node, extent_op, |
5d4f98a2 YZ |
2370 | insert_reserved); |
2371 | else if (node->type == BTRFS_EXTENT_DATA_REF_KEY || | |
2372 | node->type == BTRFS_SHARED_DATA_REF_KEY) | |
2bf98ef3 | 2373 | ret = run_delayed_data_ref(trans, node, extent_op, |
5d4f98a2 YZ |
2374 | insert_reserved); |
2375 | else | |
2376 | BUG(); | |
80ee54bf JB |
2377 | if (ret && insert_reserved) |
2378 | btrfs_pin_extent(trans->fs_info, node->bytenr, | |
2379 | node->num_bytes, 1); | |
5d4f98a2 | 2380 | return ret; |
56bec294 CM |
2381 | } |
2382 | ||
c6fc2454 | 2383 | static inline struct btrfs_delayed_ref_node * |
56bec294 CM |
2384 | select_delayed_ref(struct btrfs_delayed_ref_head *head) |
2385 | { | |
cffc3374 FM |
2386 | struct btrfs_delayed_ref_node *ref; |
2387 | ||
e3d03965 | 2388 | if (RB_EMPTY_ROOT(&head->ref_tree.rb_root)) |
c6fc2454 | 2389 | return NULL; |
d7df2c79 | 2390 | |
cffc3374 FM |
2391 | /* |
2392 | * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first. | |
2393 | * This is to prevent a ref count from going down to zero, which deletes | |
2394 | * the extent item from the extent tree, when there still are references | |
2395 | * to add, which would fail because they would not find the extent item. | |
2396 | */ | |
1d57ee94 WX |
2397 | if (!list_empty(&head->ref_add_list)) |
2398 | return list_first_entry(&head->ref_add_list, | |
2399 | struct btrfs_delayed_ref_node, add_list); | |
2400 | ||
e3d03965 | 2401 | ref = rb_entry(rb_first_cached(&head->ref_tree), |
0e0adbcf | 2402 | struct btrfs_delayed_ref_node, ref_node); |
1d57ee94 WX |
2403 | ASSERT(list_empty(&ref->add_list)); |
2404 | return ref; | |
56bec294 CM |
2405 | } |
2406 | ||
2eadaa22 JB |
2407 | static void unselect_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs, |
2408 | struct btrfs_delayed_ref_head *head) | |
2409 | { | |
2410 | spin_lock(&delayed_refs->lock); | |
2411 | head->processing = 0; | |
2412 | delayed_refs->num_heads_ready++; | |
2413 | spin_unlock(&delayed_refs->lock); | |
2414 | btrfs_delayed_ref_unlock(head); | |
2415 | } | |
2416 | ||
bedc6617 JB |
2417 | static struct btrfs_delayed_extent_op *cleanup_extent_op( |
2418 | struct btrfs_delayed_ref_head *head) | |
b00e6250 JB |
2419 | { |
2420 | struct btrfs_delayed_extent_op *extent_op = head->extent_op; | |
b00e6250 JB |
2421 | |
2422 | if (!extent_op) | |
bedc6617 JB |
2423 | return NULL; |
2424 | ||
b00e6250 | 2425 | if (head->must_insert_reserved) { |
bedc6617 | 2426 | head->extent_op = NULL; |
b00e6250 | 2427 | btrfs_free_delayed_extent_op(extent_op); |
bedc6617 | 2428 | return NULL; |
b00e6250 | 2429 | } |
bedc6617 JB |
2430 | return extent_op; |
2431 | } | |
2432 | ||
2433 | static int run_and_cleanup_extent_op(struct btrfs_trans_handle *trans, | |
2434 | struct btrfs_delayed_ref_head *head) | |
2435 | { | |
2436 | struct btrfs_delayed_extent_op *extent_op; | |
2437 | int ret; | |
2438 | ||
2439 | extent_op = cleanup_extent_op(head); | |
2440 | if (!extent_op) | |
2441 | return 0; | |
2442 | head->extent_op = NULL; | |
b00e6250 | 2443 | spin_unlock(&head->lock); |
20b9a2d6 | 2444 | ret = run_delayed_extent_op(trans, head, extent_op); |
b00e6250 JB |
2445 | btrfs_free_delayed_extent_op(extent_op); |
2446 | return ret ? ret : 1; | |
2447 | } | |
2448 | ||
31890da0 JB |
2449 | void btrfs_cleanup_ref_head_accounting(struct btrfs_fs_info *fs_info, |
2450 | struct btrfs_delayed_ref_root *delayed_refs, | |
2451 | struct btrfs_delayed_ref_head *head) | |
07c47775 | 2452 | { |
ba2c4d4e | 2453 | int nr_items = 1; /* Dropping this ref head update. */ |
07c47775 JB |
2454 | |
2455 | if (head->total_ref_mod < 0) { | |
2456 | struct btrfs_space_info *space_info; | |
2457 | u64 flags; | |
2458 | ||
2459 | if (head->is_data) | |
2460 | flags = BTRFS_BLOCK_GROUP_DATA; | |
2461 | else if (head->is_system) | |
2462 | flags = BTRFS_BLOCK_GROUP_SYSTEM; | |
2463 | else | |
2464 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
2465 | space_info = __find_space_info(fs_info, flags); | |
2466 | ASSERT(space_info); | |
2467 | percpu_counter_add_batch(&space_info->total_bytes_pinned, | |
2468 | -head->num_bytes, | |
2469 | BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
2470 | ||
ba2c4d4e JB |
2471 | /* |
2472 | * We had csum deletions accounted for in our delayed refs rsv, | |
2473 | * we need to drop the csum leaves for this update from our | |
2474 | * delayed_refs_rsv. | |
2475 | */ | |
07c47775 JB |
2476 | if (head->is_data) { |
2477 | spin_lock(&delayed_refs->lock); | |
2478 | delayed_refs->pending_csums -= head->num_bytes; | |
2479 | spin_unlock(&delayed_refs->lock); | |
ba2c4d4e JB |
2480 | nr_items += btrfs_csum_bytes_to_leaves(fs_info, |
2481 | head->num_bytes); | |
07c47775 JB |
2482 | } |
2483 | } | |
2484 | ||
ba2c4d4e | 2485 | btrfs_delayed_refs_rsv_release(fs_info, nr_items); |
07c47775 JB |
2486 | } |
2487 | ||
194ab0bc | 2488 | static int cleanup_ref_head(struct btrfs_trans_handle *trans, |
194ab0bc JB |
2489 | struct btrfs_delayed_ref_head *head) |
2490 | { | |
f9871edd NB |
2491 | |
2492 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
194ab0bc JB |
2493 | struct btrfs_delayed_ref_root *delayed_refs; |
2494 | int ret; | |
2495 | ||
2496 | delayed_refs = &trans->transaction->delayed_refs; | |
2497 | ||
bedc6617 | 2498 | ret = run_and_cleanup_extent_op(trans, head); |
194ab0bc JB |
2499 | if (ret < 0) { |
2500 | unselect_delayed_ref_head(delayed_refs, head); | |
2501 | btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret); | |
2502 | return ret; | |
2503 | } else if (ret) { | |
2504 | return ret; | |
2505 | } | |
2506 | ||
2507 | /* | |
2508 | * Need to drop our head ref lock and re-acquire the delayed ref lock | |
2509 | * and then re-check to make sure nobody got added. | |
2510 | */ | |
2511 | spin_unlock(&head->lock); | |
2512 | spin_lock(&delayed_refs->lock); | |
2513 | spin_lock(&head->lock); | |
e3d03965 | 2514 | if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root) || head->extent_op) { |
194ab0bc JB |
2515 | spin_unlock(&head->lock); |
2516 | spin_unlock(&delayed_refs->lock); | |
2517 | return 1; | |
2518 | } | |
d7baffda | 2519 | btrfs_delete_ref_head(delayed_refs, head); |
c1103f7a | 2520 | spin_unlock(&head->lock); |
1e7a1421 | 2521 | spin_unlock(&delayed_refs->lock); |
c1103f7a | 2522 | |
c1103f7a | 2523 | if (head->must_insert_reserved) { |
d278850e JB |
2524 | btrfs_pin_extent(fs_info, head->bytenr, |
2525 | head->num_bytes, 1); | |
c1103f7a | 2526 | if (head->is_data) { |
d278850e JB |
2527 | ret = btrfs_del_csums(trans, fs_info, head->bytenr, |
2528 | head->num_bytes); | |
c1103f7a JB |
2529 | } |
2530 | } | |
2531 | ||
31890da0 | 2532 | btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head); |
07c47775 JB |
2533 | |
2534 | trace_run_delayed_ref_head(fs_info, head, 0); | |
c1103f7a | 2535 | btrfs_delayed_ref_unlock(head); |
d278850e | 2536 | btrfs_put_delayed_ref_head(head); |
194ab0bc JB |
2537 | return 0; |
2538 | } | |
2539 | ||
b1cdbcb5 NB |
2540 | static struct btrfs_delayed_ref_head *btrfs_obtain_ref_head( |
2541 | struct btrfs_trans_handle *trans) | |
2542 | { | |
2543 | struct btrfs_delayed_ref_root *delayed_refs = | |
2544 | &trans->transaction->delayed_refs; | |
2545 | struct btrfs_delayed_ref_head *head = NULL; | |
2546 | int ret; | |
2547 | ||
2548 | spin_lock(&delayed_refs->lock); | |
5637c74b | 2549 | head = btrfs_select_ref_head(delayed_refs); |
b1cdbcb5 NB |
2550 | if (!head) { |
2551 | spin_unlock(&delayed_refs->lock); | |
2552 | return head; | |
2553 | } | |
2554 | ||
2555 | /* | |
2556 | * Grab the lock that says we are going to process all the refs for | |
2557 | * this head | |
2558 | */ | |
9e920a6f | 2559 | ret = btrfs_delayed_ref_lock(delayed_refs, head); |
b1cdbcb5 NB |
2560 | spin_unlock(&delayed_refs->lock); |
2561 | ||
2562 | /* | |
2563 | * We may have dropped the spin lock to get the head mutex lock, and | |
2564 | * that might have given someone else time to free the head. If that's | |
2565 | * true, it has been removed from our list and we can move on. | |
2566 | */ | |
2567 | if (ret == -EAGAIN) | |
2568 | head = ERR_PTR(-EAGAIN); | |
2569 | ||
2570 | return head; | |
2571 | } | |
2572 | ||
e7261386 NB |
2573 | static int btrfs_run_delayed_refs_for_head(struct btrfs_trans_handle *trans, |
2574 | struct btrfs_delayed_ref_head *locked_ref, | |
2575 | unsigned long *run_refs) | |
2576 | { | |
2577 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
2578 | struct btrfs_delayed_ref_root *delayed_refs; | |
2579 | struct btrfs_delayed_extent_op *extent_op; | |
2580 | struct btrfs_delayed_ref_node *ref; | |
2581 | int must_insert_reserved = 0; | |
2582 | int ret; | |
2583 | ||
2584 | delayed_refs = &trans->transaction->delayed_refs; | |
2585 | ||
0110a4c4 NB |
2586 | lockdep_assert_held(&locked_ref->mutex); |
2587 | lockdep_assert_held(&locked_ref->lock); | |
2588 | ||
e7261386 NB |
2589 | while ((ref = select_delayed_ref(locked_ref))) { |
2590 | if (ref->seq && | |
2591 | btrfs_check_delayed_seq(fs_info, ref->seq)) { | |
2592 | spin_unlock(&locked_ref->lock); | |
2593 | unselect_delayed_ref_head(delayed_refs, locked_ref); | |
2594 | return -EAGAIN; | |
2595 | } | |
2596 | ||
2597 | (*run_refs)++; | |
2598 | ref->in_tree = 0; | |
2599 | rb_erase_cached(&ref->ref_node, &locked_ref->ref_tree); | |
2600 | RB_CLEAR_NODE(&ref->ref_node); | |
2601 | if (!list_empty(&ref->add_list)) | |
2602 | list_del(&ref->add_list); | |
2603 | /* | |
2604 | * When we play the delayed ref, also correct the ref_mod on | |
2605 | * head | |
2606 | */ | |
2607 | switch (ref->action) { | |
2608 | case BTRFS_ADD_DELAYED_REF: | |
2609 | case BTRFS_ADD_DELAYED_EXTENT: | |
2610 | locked_ref->ref_mod -= ref->ref_mod; | |
2611 | break; | |
2612 | case BTRFS_DROP_DELAYED_REF: | |
2613 | locked_ref->ref_mod += ref->ref_mod; | |
2614 | break; | |
2615 | default: | |
2616 | WARN_ON(1); | |
2617 | } | |
2618 | atomic_dec(&delayed_refs->num_entries); | |
2619 | ||
2620 | /* | |
2621 | * Record the must_insert_reserved flag before we drop the | |
2622 | * spin lock. | |
2623 | */ | |
2624 | must_insert_reserved = locked_ref->must_insert_reserved; | |
2625 | locked_ref->must_insert_reserved = 0; | |
2626 | ||
2627 | extent_op = locked_ref->extent_op; | |
2628 | locked_ref->extent_op = NULL; | |
2629 | spin_unlock(&locked_ref->lock); | |
2630 | ||
2631 | ret = run_one_delayed_ref(trans, ref, extent_op, | |
2632 | must_insert_reserved); | |
2633 | ||
2634 | btrfs_free_delayed_extent_op(extent_op); | |
2635 | if (ret) { | |
2636 | unselect_delayed_ref_head(delayed_refs, locked_ref); | |
2637 | btrfs_put_delayed_ref(ref); | |
2638 | btrfs_debug(fs_info, "run_one_delayed_ref returned %d", | |
2639 | ret); | |
2640 | return ret; | |
2641 | } | |
2642 | ||
2643 | btrfs_put_delayed_ref(ref); | |
2644 | cond_resched(); | |
2645 | ||
2646 | spin_lock(&locked_ref->lock); | |
2647 | btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref); | |
2648 | } | |
2649 | ||
2650 | return 0; | |
2651 | } | |
2652 | ||
79787eaa JM |
2653 | /* |
2654 | * Returns 0 on success or if called with an already aborted transaction. | |
2655 | * Returns -ENOMEM or -EIO on failure and will abort the transaction. | |
2656 | */ | |
d7df2c79 | 2657 | static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, |
d7df2c79 | 2658 | unsigned long nr) |
56bec294 | 2659 | { |
0a1e458a | 2660 | struct btrfs_fs_info *fs_info = trans->fs_info; |
56bec294 | 2661 | struct btrfs_delayed_ref_root *delayed_refs; |
56bec294 | 2662 | struct btrfs_delayed_ref_head *locked_ref = NULL; |
0a2b2a84 | 2663 | ktime_t start = ktime_get(); |
56bec294 | 2664 | int ret; |
d7df2c79 | 2665 | unsigned long count = 0; |
0a2b2a84 | 2666 | unsigned long actual_count = 0; |
56bec294 CM |
2667 | |
2668 | delayed_refs = &trans->transaction->delayed_refs; | |
0110a4c4 | 2669 | do { |
56bec294 | 2670 | if (!locked_ref) { |
b1cdbcb5 | 2671 | locked_ref = btrfs_obtain_ref_head(trans); |
0110a4c4 NB |
2672 | if (IS_ERR_OR_NULL(locked_ref)) { |
2673 | if (PTR_ERR(locked_ref) == -EAGAIN) { | |
2674 | continue; | |
2675 | } else { | |
2676 | break; | |
2677 | } | |
56bec294 | 2678 | } |
0110a4c4 | 2679 | count++; |
56bec294 | 2680 | } |
2c3cf7d5 FM |
2681 | /* |
2682 | * We need to try and merge add/drops of the same ref since we | |
2683 | * can run into issues with relocate dropping the implicit ref | |
2684 | * and then it being added back again before the drop can | |
2685 | * finish. If we merged anything we need to re-loop so we can | |
2686 | * get a good ref. | |
2687 | * Or we can get node references of the same type that weren't | |
2688 | * merged when created due to bumps in the tree mod seq, and | |
2689 | * we need to merge them to prevent adding an inline extent | |
2690 | * backref before dropping it (triggering a BUG_ON at | |
2691 | * insert_inline_extent_backref()). | |
2692 | */ | |
d7df2c79 | 2693 | spin_lock(&locked_ref->lock); |
be97f133 | 2694 | btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref); |
ae1e206b | 2695 | |
0110a4c4 NB |
2696 | ret = btrfs_run_delayed_refs_for_head(trans, locked_ref, |
2697 | &actual_count); | |
2698 | if (ret < 0 && ret != -EAGAIN) { | |
2699 | /* | |
2700 | * Error, btrfs_run_delayed_refs_for_head already | |
2701 | * unlocked everything so just bail out | |
2702 | */ | |
2703 | return ret; | |
2704 | } else if (!ret) { | |
2705 | /* | |
2706 | * Success, perform the usual cleanup of a processed | |
2707 | * head | |
2708 | */ | |
f9871edd | 2709 | ret = cleanup_ref_head(trans, locked_ref); |
194ab0bc | 2710 | if (ret > 0 ) { |
b00e6250 JB |
2711 | /* We dropped our lock, we need to loop. */ |
2712 | ret = 0; | |
d7df2c79 | 2713 | continue; |
194ab0bc JB |
2714 | } else if (ret) { |
2715 | return ret; | |
5d4f98a2 | 2716 | } |
22cd2e7d | 2717 | } |
1ce7a5ec | 2718 | |
b00e6250 | 2719 | /* |
0110a4c4 NB |
2720 | * Either success case or btrfs_run_delayed_refs_for_head |
2721 | * returned -EAGAIN, meaning we need to select another head | |
b00e6250 | 2722 | */ |
b00e6250 | 2723 | |
0110a4c4 | 2724 | locked_ref = NULL; |
c3e69d58 | 2725 | cond_resched(); |
0110a4c4 | 2726 | } while ((nr != -1 && count < nr) || locked_ref); |
0a2b2a84 JB |
2727 | |
2728 | /* | |
2729 | * We don't want to include ref heads since we can have empty ref heads | |
2730 | * and those will drastically skew our runtime down since we just do | |
2731 | * accounting, no actual extent tree updates. | |
2732 | */ | |
2733 | if (actual_count > 0) { | |
2734 | u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start)); | |
2735 | u64 avg; | |
2736 | ||
2737 | /* | |
2738 | * We weigh the current average higher than our current runtime | |
2739 | * to avoid large swings in the average. | |
2740 | */ | |
2741 | spin_lock(&delayed_refs->lock); | |
2742 | avg = fs_info->avg_delayed_ref_runtime * 3 + runtime; | |
f8c269d7 | 2743 | fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */ |
0a2b2a84 JB |
2744 | spin_unlock(&delayed_refs->lock); |
2745 | } | |
d7df2c79 | 2746 | return 0; |
c3e69d58 CM |
2747 | } |
2748 | ||
709c0486 AJ |
2749 | #ifdef SCRAMBLE_DELAYED_REFS |
2750 | /* | |
2751 | * Normally delayed refs get processed in ascending bytenr order. This | |
2752 | * correlates in most cases to the order added. To expose dependencies on this | |
2753 | * order, we start to process the tree in the middle instead of the beginning | |
2754 | */ | |
2755 | static u64 find_middle(struct rb_root *root) | |
2756 | { | |
2757 | struct rb_node *n = root->rb_node; | |
2758 | struct btrfs_delayed_ref_node *entry; | |
2759 | int alt = 1; | |
2760 | u64 middle; | |
2761 | u64 first = 0, last = 0; | |
2762 | ||
2763 | n = rb_first(root); | |
2764 | if (n) { | |
2765 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2766 | first = entry->bytenr; | |
2767 | } | |
2768 | n = rb_last(root); | |
2769 | if (n) { | |
2770 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2771 | last = entry->bytenr; | |
2772 | } | |
2773 | n = root->rb_node; | |
2774 | ||
2775 | while (n) { | |
2776 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2777 | WARN_ON(!entry->in_tree); | |
2778 | ||
2779 | middle = entry->bytenr; | |
2780 | ||
2781 | if (alt) | |
2782 | n = n->rb_left; | |
2783 | else | |
2784 | n = n->rb_right; | |
2785 | ||
2786 | alt = 1 - alt; | |
2787 | } | |
2788 | return middle; | |
2789 | } | |
2790 | #endif | |
2791 | ||
2ff7e61e | 2792 | static inline u64 heads_to_leaves(struct btrfs_fs_info *fs_info, u64 heads) |
1be41b78 JB |
2793 | { |
2794 | u64 num_bytes; | |
2795 | ||
2796 | num_bytes = heads * (sizeof(struct btrfs_extent_item) + | |
2797 | sizeof(struct btrfs_extent_inline_ref)); | |
0b246afa | 2798 | if (!btrfs_fs_incompat(fs_info, SKINNY_METADATA)) |
1be41b78 JB |
2799 | num_bytes += heads * sizeof(struct btrfs_tree_block_info); |
2800 | ||
2801 | /* | |
2802 | * We don't ever fill up leaves all the way so multiply by 2 just to be | |
01327610 | 2803 | * closer to what we're really going to want to use. |
1be41b78 | 2804 | */ |
0b246afa | 2805 | return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(fs_info)); |
1be41b78 JB |
2806 | } |
2807 | ||
1262133b JB |
2808 | /* |
2809 | * Takes the number of bytes to be csumm'ed and figures out how many leaves it | |
2810 | * would require to store the csums for that many bytes. | |
2811 | */ | |
2ff7e61e | 2812 | u64 btrfs_csum_bytes_to_leaves(struct btrfs_fs_info *fs_info, u64 csum_bytes) |
1262133b JB |
2813 | { |
2814 | u64 csum_size; | |
2815 | u64 num_csums_per_leaf; | |
2816 | u64 num_csums; | |
2817 | ||
0b246afa | 2818 | csum_size = BTRFS_MAX_ITEM_SIZE(fs_info); |
1262133b | 2819 | num_csums_per_leaf = div64_u64(csum_size, |
0b246afa JM |
2820 | (u64)btrfs_super_csum_size(fs_info->super_copy)); |
2821 | num_csums = div64_u64(csum_bytes, fs_info->sectorsize); | |
1262133b JB |
2822 | num_csums += num_csums_per_leaf - 1; |
2823 | num_csums = div64_u64(num_csums, num_csums_per_leaf); | |
2824 | return num_csums; | |
2825 | } | |
2826 | ||
64403612 | 2827 | bool btrfs_check_space_for_delayed_refs(struct btrfs_fs_info *fs_info) |
1be41b78 | 2828 | { |
64403612 JB |
2829 | struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv; |
2830 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
2831 | bool ret = false; | |
2832 | u64 reserved; | |
1be41b78 | 2833 | |
64403612 JB |
2834 | spin_lock(&global_rsv->lock); |
2835 | reserved = global_rsv->reserved; | |
2836 | spin_unlock(&global_rsv->lock); | |
1be41b78 JB |
2837 | |
2838 | /* | |
64403612 JB |
2839 | * Since the global reserve is just kind of magic we don't really want |
2840 | * to rely on it to save our bacon, so if our size is more than the | |
2841 | * delayed_refs_rsv and the global rsv then it's time to think about | |
2842 | * bailing. | |
1be41b78 | 2843 | */ |
64403612 JB |
2844 | spin_lock(&delayed_refs_rsv->lock); |
2845 | reserved += delayed_refs_rsv->reserved; | |
2846 | if (delayed_refs_rsv->size >= reserved) | |
2847 | ret = true; | |
2848 | spin_unlock(&delayed_refs_rsv->lock); | |
1be41b78 JB |
2849 | return ret; |
2850 | } | |
2851 | ||
7c861627 | 2852 | int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans) |
0a2b2a84 | 2853 | { |
0a2b2a84 JB |
2854 | u64 num_entries = |
2855 | atomic_read(&trans->transaction->delayed_refs.num_entries); | |
2856 | u64 avg_runtime; | |
a79b7d4b | 2857 | u64 val; |
0a2b2a84 JB |
2858 | |
2859 | smp_mb(); | |
7c861627 | 2860 | avg_runtime = trans->fs_info->avg_delayed_ref_runtime; |
a79b7d4b | 2861 | val = num_entries * avg_runtime; |
dc1a90c6 | 2862 | if (val >= NSEC_PER_SEC) |
0a2b2a84 | 2863 | return 1; |
a79b7d4b CM |
2864 | if (val >= NSEC_PER_SEC / 2) |
2865 | return 2; | |
0a2b2a84 | 2866 | |
64403612 | 2867 | return btrfs_check_space_for_delayed_refs(trans->fs_info); |
0a2b2a84 JB |
2868 | } |
2869 | ||
c3e69d58 CM |
2870 | /* |
2871 | * this starts processing the delayed reference count updates and | |
2872 | * extent insertions we have queued up so far. count can be | |
2873 | * 0, which means to process everything in the tree at the start | |
2874 | * of the run (but not newly added entries), or it can be some target | |
2875 | * number you'd like to process. | |
79787eaa JM |
2876 | * |
2877 | * Returns 0 on success or if called with an aborted transaction | |
2878 | * Returns <0 on error and aborts the transaction | |
c3e69d58 CM |
2879 | */ |
2880 | int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, | |
c79a70b1 | 2881 | unsigned long count) |
c3e69d58 | 2882 | { |
c79a70b1 | 2883 | struct btrfs_fs_info *fs_info = trans->fs_info; |
c3e69d58 CM |
2884 | struct rb_node *node; |
2885 | struct btrfs_delayed_ref_root *delayed_refs; | |
c46effa6 | 2886 | struct btrfs_delayed_ref_head *head; |
c3e69d58 CM |
2887 | int ret; |
2888 | int run_all = count == (unsigned long)-1; | |
c3e69d58 | 2889 | |
79787eaa JM |
2890 | /* We'll clean this up in btrfs_cleanup_transaction */ |
2891 | if (trans->aborted) | |
2892 | return 0; | |
2893 | ||
0b246afa | 2894 | if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags)) |
511711af CM |
2895 | return 0; |
2896 | ||
c3e69d58 | 2897 | delayed_refs = &trans->transaction->delayed_refs; |
26455d33 | 2898 | if (count == 0) |
d7df2c79 | 2899 | count = atomic_read(&delayed_refs->num_entries) * 2; |
bb721703 | 2900 | |
c3e69d58 | 2901 | again: |
709c0486 AJ |
2902 | #ifdef SCRAMBLE_DELAYED_REFS |
2903 | delayed_refs->run_delayed_start = find_middle(&delayed_refs->root); | |
2904 | #endif | |
0a1e458a | 2905 | ret = __btrfs_run_delayed_refs(trans, count); |
d7df2c79 | 2906 | if (ret < 0) { |
66642832 | 2907 | btrfs_abort_transaction(trans, ret); |
d7df2c79 | 2908 | return ret; |
eb099670 | 2909 | } |
c3e69d58 | 2910 | |
56bec294 | 2911 | if (run_all) { |
119e80df | 2912 | btrfs_create_pending_block_groups(trans); |
ea658bad | 2913 | |
d7df2c79 | 2914 | spin_lock(&delayed_refs->lock); |
5c9d028b | 2915 | node = rb_first_cached(&delayed_refs->href_root); |
d7df2c79 JB |
2916 | if (!node) { |
2917 | spin_unlock(&delayed_refs->lock); | |
56bec294 | 2918 | goto out; |
d7df2c79 | 2919 | } |
d278850e JB |
2920 | head = rb_entry(node, struct btrfs_delayed_ref_head, |
2921 | href_node); | |
2922 | refcount_inc(&head->refs); | |
2923 | spin_unlock(&delayed_refs->lock); | |
e9d0b13b | 2924 | |
d278850e JB |
2925 | /* Mutex was contended, block until it's released and retry. */ |
2926 | mutex_lock(&head->mutex); | |
2927 | mutex_unlock(&head->mutex); | |
56bec294 | 2928 | |
d278850e | 2929 | btrfs_put_delayed_ref_head(head); |
d7df2c79 | 2930 | cond_resched(); |
56bec294 | 2931 | goto again; |
5f39d397 | 2932 | } |
54aa1f4d | 2933 | out: |
a28ec197 CM |
2934 | return 0; |
2935 | } | |
2936 | ||
5d4f98a2 | 2937 | int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans, |
5d4f98a2 | 2938 | u64 bytenr, u64 num_bytes, u64 flags, |
b1c79e09 | 2939 | int level, int is_data) |
5d4f98a2 YZ |
2940 | { |
2941 | struct btrfs_delayed_extent_op *extent_op; | |
2942 | int ret; | |
2943 | ||
78a6184a | 2944 | extent_op = btrfs_alloc_delayed_extent_op(); |
5d4f98a2 YZ |
2945 | if (!extent_op) |
2946 | return -ENOMEM; | |
2947 | ||
2948 | extent_op->flags_to_set = flags; | |
35b3ad50 DS |
2949 | extent_op->update_flags = true; |
2950 | extent_op->update_key = false; | |
2951 | extent_op->is_data = is_data ? true : false; | |
b1c79e09 | 2952 | extent_op->level = level; |
5d4f98a2 | 2953 | |
c6e340bc | 2954 | ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op); |
5d4f98a2 | 2955 | if (ret) |
78a6184a | 2956 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 YZ |
2957 | return ret; |
2958 | } | |
2959 | ||
e4c3b2dc | 2960 | static noinline int check_delayed_ref(struct btrfs_root *root, |
5d4f98a2 YZ |
2961 | struct btrfs_path *path, |
2962 | u64 objectid, u64 offset, u64 bytenr) | |
2963 | { | |
2964 | struct btrfs_delayed_ref_head *head; | |
2965 | struct btrfs_delayed_ref_node *ref; | |
2966 | struct btrfs_delayed_data_ref *data_ref; | |
2967 | struct btrfs_delayed_ref_root *delayed_refs; | |
e4c3b2dc | 2968 | struct btrfs_transaction *cur_trans; |
0e0adbcf | 2969 | struct rb_node *node; |
5d4f98a2 YZ |
2970 | int ret = 0; |
2971 | ||
998ac6d2 | 2972 | spin_lock(&root->fs_info->trans_lock); |
e4c3b2dc | 2973 | cur_trans = root->fs_info->running_transaction; |
998ac6d2 | 2974 | if (cur_trans) |
2975 | refcount_inc(&cur_trans->use_count); | |
2976 | spin_unlock(&root->fs_info->trans_lock); | |
e4c3b2dc LB |
2977 | if (!cur_trans) |
2978 | return 0; | |
2979 | ||
2980 | delayed_refs = &cur_trans->delayed_refs; | |
5d4f98a2 | 2981 | spin_lock(&delayed_refs->lock); |
f72ad18e | 2982 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
d7df2c79 JB |
2983 | if (!head) { |
2984 | spin_unlock(&delayed_refs->lock); | |
998ac6d2 | 2985 | btrfs_put_transaction(cur_trans); |
d7df2c79 JB |
2986 | return 0; |
2987 | } | |
5d4f98a2 YZ |
2988 | |
2989 | if (!mutex_trylock(&head->mutex)) { | |
d278850e | 2990 | refcount_inc(&head->refs); |
5d4f98a2 YZ |
2991 | spin_unlock(&delayed_refs->lock); |
2992 | ||
b3b4aa74 | 2993 | btrfs_release_path(path); |
5d4f98a2 | 2994 | |
8cc33e5c DS |
2995 | /* |
2996 | * Mutex was contended, block until it's released and let | |
2997 | * caller try again | |
2998 | */ | |
5d4f98a2 YZ |
2999 | mutex_lock(&head->mutex); |
3000 | mutex_unlock(&head->mutex); | |
d278850e | 3001 | btrfs_put_delayed_ref_head(head); |
998ac6d2 | 3002 | btrfs_put_transaction(cur_trans); |
5d4f98a2 YZ |
3003 | return -EAGAIN; |
3004 | } | |
d7df2c79 | 3005 | spin_unlock(&delayed_refs->lock); |
5d4f98a2 | 3006 | |
d7df2c79 | 3007 | spin_lock(&head->lock); |
0e0adbcf JB |
3008 | /* |
3009 | * XXX: We should replace this with a proper search function in the | |
3010 | * future. | |
3011 | */ | |
e3d03965 LB |
3012 | for (node = rb_first_cached(&head->ref_tree); node; |
3013 | node = rb_next(node)) { | |
0e0adbcf | 3014 | ref = rb_entry(node, struct btrfs_delayed_ref_node, ref_node); |
d7df2c79 JB |
3015 | /* If it's a shared ref we know a cross reference exists */ |
3016 | if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) { | |
3017 | ret = 1; | |
3018 | break; | |
3019 | } | |
5d4f98a2 | 3020 | |
d7df2c79 | 3021 | data_ref = btrfs_delayed_node_to_data_ref(ref); |
5d4f98a2 | 3022 | |
d7df2c79 JB |
3023 | /* |
3024 | * If our ref doesn't match the one we're currently looking at | |
3025 | * then we have a cross reference. | |
3026 | */ | |
3027 | if (data_ref->root != root->root_key.objectid || | |
3028 | data_ref->objectid != objectid || | |
3029 | data_ref->offset != offset) { | |
3030 | ret = 1; | |
3031 | break; | |
3032 | } | |
5d4f98a2 | 3033 | } |
d7df2c79 | 3034 | spin_unlock(&head->lock); |
5d4f98a2 | 3035 | mutex_unlock(&head->mutex); |
998ac6d2 | 3036 | btrfs_put_transaction(cur_trans); |
5d4f98a2 YZ |
3037 | return ret; |
3038 | } | |
3039 | ||
e4c3b2dc | 3040 | static noinline int check_committed_ref(struct btrfs_root *root, |
5d4f98a2 YZ |
3041 | struct btrfs_path *path, |
3042 | u64 objectid, u64 offset, u64 bytenr) | |
be20aa9d | 3043 | { |
0b246afa JM |
3044 | struct btrfs_fs_info *fs_info = root->fs_info; |
3045 | struct btrfs_root *extent_root = fs_info->extent_root; | |
f321e491 | 3046 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
3047 | struct btrfs_extent_data_ref *ref; |
3048 | struct btrfs_extent_inline_ref *iref; | |
3049 | struct btrfs_extent_item *ei; | |
f321e491 | 3050 | struct btrfs_key key; |
5d4f98a2 | 3051 | u32 item_size; |
3de28d57 | 3052 | int type; |
be20aa9d | 3053 | int ret; |
925baedd | 3054 | |
be20aa9d | 3055 | key.objectid = bytenr; |
31840ae1 | 3056 | key.offset = (u64)-1; |
f321e491 | 3057 | key.type = BTRFS_EXTENT_ITEM_KEY; |
be20aa9d | 3058 | |
be20aa9d CM |
3059 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
3060 | if (ret < 0) | |
3061 | goto out; | |
79787eaa | 3062 | BUG_ON(ret == 0); /* Corruption */ |
80ff3856 YZ |
3063 | |
3064 | ret = -ENOENT; | |
3065 | if (path->slots[0] == 0) | |
31840ae1 | 3066 | goto out; |
be20aa9d | 3067 | |
31840ae1 | 3068 | path->slots[0]--; |
f321e491 | 3069 | leaf = path->nodes[0]; |
5d4f98a2 | 3070 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
be20aa9d | 3071 | |
5d4f98a2 | 3072 | if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY) |
be20aa9d | 3073 | goto out; |
f321e491 | 3074 | |
5d4f98a2 YZ |
3075 | ret = 1; |
3076 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
5d4f98a2 | 3077 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
bd09835d | 3078 | |
5d4f98a2 YZ |
3079 | if (item_size != sizeof(*ei) + |
3080 | btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY)) | |
3081 | goto out; | |
be20aa9d | 3082 | |
5d4f98a2 YZ |
3083 | if (btrfs_extent_generation(leaf, ei) <= |
3084 | btrfs_root_last_snapshot(&root->root_item)) | |
3085 | goto out; | |
3086 | ||
3087 | iref = (struct btrfs_extent_inline_ref *)(ei + 1); | |
3de28d57 LB |
3088 | |
3089 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA); | |
3090 | if (type != BTRFS_EXTENT_DATA_REF_KEY) | |
5d4f98a2 YZ |
3091 | goto out; |
3092 | ||
3093 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
3094 | if (btrfs_extent_refs(leaf, ei) != | |
3095 | btrfs_extent_data_ref_count(leaf, ref) || | |
3096 | btrfs_extent_data_ref_root(leaf, ref) != | |
3097 | root->root_key.objectid || | |
3098 | btrfs_extent_data_ref_objectid(leaf, ref) != objectid || | |
3099 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
3100 | goto out; | |
3101 | ||
3102 | ret = 0; | |
3103 | out: | |
3104 | return ret; | |
3105 | } | |
3106 | ||
e4c3b2dc LB |
3107 | int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset, |
3108 | u64 bytenr) | |
5d4f98a2 YZ |
3109 | { |
3110 | struct btrfs_path *path; | |
3111 | int ret; | |
5d4f98a2 YZ |
3112 | |
3113 | path = btrfs_alloc_path(); | |
3114 | if (!path) | |
9132c4ff | 3115 | return -ENOMEM; |
5d4f98a2 YZ |
3116 | |
3117 | do { | |
e4c3b2dc | 3118 | ret = check_committed_ref(root, path, objectid, |
5d4f98a2 YZ |
3119 | offset, bytenr); |
3120 | if (ret && ret != -ENOENT) | |
f321e491 | 3121 | goto out; |
80ff3856 | 3122 | |
380fd066 MT |
3123 | ret = check_delayed_ref(root, path, objectid, offset, bytenr); |
3124 | } while (ret == -EAGAIN); | |
5d4f98a2 | 3125 | |
be20aa9d | 3126 | out: |
80ff3856 | 3127 | btrfs_free_path(path); |
f0486c68 YZ |
3128 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) |
3129 | WARN_ON(ret > 0); | |
f321e491 | 3130 | return ret; |
be20aa9d | 3131 | } |
c5739bba | 3132 | |
5d4f98a2 | 3133 | static int __btrfs_mod_ref(struct btrfs_trans_handle *trans, |
b7a9f29f | 3134 | struct btrfs_root *root, |
5d4f98a2 | 3135 | struct extent_buffer *buf, |
e339a6b0 | 3136 | int full_backref, int inc) |
31840ae1 | 3137 | { |
0b246afa | 3138 | struct btrfs_fs_info *fs_info = root->fs_info; |
31840ae1 | 3139 | u64 bytenr; |
5d4f98a2 YZ |
3140 | u64 num_bytes; |
3141 | u64 parent; | |
31840ae1 | 3142 | u64 ref_root; |
31840ae1 | 3143 | u32 nritems; |
31840ae1 ZY |
3144 | struct btrfs_key key; |
3145 | struct btrfs_file_extent_item *fi; | |
82fa113f QW |
3146 | struct btrfs_ref generic_ref = { 0 }; |
3147 | bool for_reloc = btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC); | |
31840ae1 | 3148 | int i; |
82fa113f | 3149 | int action; |
31840ae1 ZY |
3150 | int level; |
3151 | int ret = 0; | |
fccb84c9 | 3152 | |
0b246afa | 3153 | if (btrfs_is_testing(fs_info)) |
faa2dbf0 | 3154 | return 0; |
fccb84c9 | 3155 | |
31840ae1 | 3156 | ref_root = btrfs_header_owner(buf); |
31840ae1 ZY |
3157 | nritems = btrfs_header_nritems(buf); |
3158 | level = btrfs_header_level(buf); | |
3159 | ||
27cdeb70 | 3160 | if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0) |
5d4f98a2 | 3161 | return 0; |
31840ae1 | 3162 | |
5d4f98a2 YZ |
3163 | if (full_backref) |
3164 | parent = buf->start; | |
3165 | else | |
3166 | parent = 0; | |
82fa113f QW |
3167 | if (inc) |
3168 | action = BTRFS_ADD_DELAYED_REF; | |
3169 | else | |
3170 | action = BTRFS_DROP_DELAYED_REF; | |
5d4f98a2 YZ |
3171 | |
3172 | for (i = 0; i < nritems; i++) { | |
31840ae1 | 3173 | if (level == 0) { |
5d4f98a2 | 3174 | btrfs_item_key_to_cpu(buf, &key, i); |
962a298f | 3175 | if (key.type != BTRFS_EXTENT_DATA_KEY) |
31840ae1 | 3176 | continue; |
5d4f98a2 | 3177 | fi = btrfs_item_ptr(buf, i, |
31840ae1 ZY |
3178 | struct btrfs_file_extent_item); |
3179 | if (btrfs_file_extent_type(buf, fi) == | |
3180 | BTRFS_FILE_EXTENT_INLINE) | |
3181 | continue; | |
3182 | bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
3183 | if (bytenr == 0) | |
3184 | continue; | |
5d4f98a2 YZ |
3185 | |
3186 | num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi); | |
3187 | key.offset -= btrfs_file_extent_offset(buf, fi); | |
82fa113f QW |
3188 | btrfs_init_generic_ref(&generic_ref, action, bytenr, |
3189 | num_bytes, parent); | |
3190 | generic_ref.real_root = root->root_key.objectid; | |
3191 | btrfs_init_data_ref(&generic_ref, ref_root, key.objectid, | |
3192 | key.offset); | |
3193 | generic_ref.skip_qgroup = for_reloc; | |
dd28b6a5 | 3194 | if (inc) |
82fa113f | 3195 | ret = btrfs_inc_extent_ref(trans, &generic_ref); |
dd28b6a5 | 3196 | else |
ffd4bb2a | 3197 | ret = btrfs_free_extent(trans, &generic_ref); |
31840ae1 ZY |
3198 | if (ret) |
3199 | goto fail; | |
3200 | } else { | |
5d4f98a2 | 3201 | bytenr = btrfs_node_blockptr(buf, i); |
0b246afa | 3202 | num_bytes = fs_info->nodesize; |
82fa113f QW |
3203 | btrfs_init_generic_ref(&generic_ref, action, bytenr, |
3204 | num_bytes, parent); | |
3205 | generic_ref.real_root = root->root_key.objectid; | |
3206 | btrfs_init_tree_ref(&generic_ref, level - 1, ref_root); | |
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 | } | |
3215 | } | |
3216 | return 0; | |
3217 | fail: | |
5d4f98a2 YZ |
3218 | return ret; |
3219 | } | |
3220 | ||
3221 | int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
e339a6b0 | 3222 | struct extent_buffer *buf, int full_backref) |
5d4f98a2 | 3223 | { |
e339a6b0 | 3224 | return __btrfs_mod_ref(trans, root, buf, full_backref, 1); |
5d4f98a2 YZ |
3225 | } |
3226 | ||
3227 | int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
e339a6b0 | 3228 | struct extent_buffer *buf, int full_backref) |
5d4f98a2 | 3229 | { |
e339a6b0 | 3230 | return __btrfs_mod_ref(trans, root, buf, full_backref, 0); |
31840ae1 ZY |
3231 | } |
3232 | ||
9078a3e1 | 3233 | static int write_one_cache_group(struct btrfs_trans_handle *trans, |
9078a3e1 CM |
3234 | struct btrfs_path *path, |
3235 | struct btrfs_block_group_cache *cache) | |
3236 | { | |
39db232d | 3237 | struct btrfs_fs_info *fs_info = trans->fs_info; |
9078a3e1 | 3238 | int ret; |
0b246afa | 3239 | struct btrfs_root *extent_root = fs_info->extent_root; |
5f39d397 CM |
3240 | unsigned long bi; |
3241 | struct extent_buffer *leaf; | |
9078a3e1 | 3242 | |
9078a3e1 | 3243 | ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1); |
df95e7f0 JB |
3244 | if (ret) { |
3245 | if (ret > 0) | |
3246 | ret = -ENOENT; | |
54aa1f4d | 3247 | goto fail; |
df95e7f0 | 3248 | } |
5f39d397 CM |
3249 | |
3250 | leaf = path->nodes[0]; | |
3251 | bi = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
3252 | write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item)); | |
3253 | btrfs_mark_buffer_dirty(leaf); | |
54aa1f4d | 3254 | fail: |
24b89d08 | 3255 | btrfs_release_path(path); |
df95e7f0 | 3256 | return ret; |
9078a3e1 CM |
3257 | |
3258 | } | |
3259 | ||
f87b7eb8 DS |
3260 | static struct btrfs_block_group_cache *next_block_group( |
3261 | struct btrfs_block_group_cache *cache) | |
4a8c9a62 | 3262 | { |
f87b7eb8 | 3263 | struct btrfs_fs_info *fs_info = cache->fs_info; |
4a8c9a62 | 3264 | struct rb_node *node; |
292cbd51 | 3265 | |
0b246afa | 3266 | spin_lock(&fs_info->block_group_cache_lock); |
292cbd51 FM |
3267 | |
3268 | /* If our block group was removed, we need a full search. */ | |
3269 | if (RB_EMPTY_NODE(&cache->cache_node)) { | |
3270 | const u64 next_bytenr = cache->key.objectid + cache->key.offset; | |
3271 | ||
0b246afa | 3272 | spin_unlock(&fs_info->block_group_cache_lock); |
292cbd51 | 3273 | btrfs_put_block_group(cache); |
0b246afa | 3274 | cache = btrfs_lookup_first_block_group(fs_info, next_bytenr); return cache; |
292cbd51 | 3275 | } |
4a8c9a62 YZ |
3276 | node = rb_next(&cache->cache_node); |
3277 | btrfs_put_block_group(cache); | |
3278 | if (node) { | |
3279 | cache = rb_entry(node, struct btrfs_block_group_cache, | |
3280 | cache_node); | |
11dfe35a | 3281 | btrfs_get_block_group(cache); |
4a8c9a62 YZ |
3282 | } else |
3283 | cache = NULL; | |
0b246afa | 3284 | spin_unlock(&fs_info->block_group_cache_lock); |
4a8c9a62 YZ |
3285 | return cache; |
3286 | } | |
3287 | ||
0af3d00b JB |
3288 | static int cache_save_setup(struct btrfs_block_group_cache *block_group, |
3289 | struct btrfs_trans_handle *trans, | |
3290 | struct btrfs_path *path) | |
3291 | { | |
0b246afa JM |
3292 | struct btrfs_fs_info *fs_info = block_group->fs_info; |
3293 | struct btrfs_root *root = fs_info->tree_root; | |
0af3d00b | 3294 | struct inode *inode = NULL; |
364ecf36 | 3295 | struct extent_changeset *data_reserved = NULL; |
0af3d00b | 3296 | u64 alloc_hint = 0; |
2b20982e | 3297 | int dcs = BTRFS_DC_ERROR; |
f8c269d7 | 3298 | u64 num_pages = 0; |
0af3d00b JB |
3299 | int retries = 0; |
3300 | int ret = 0; | |
3301 | ||
3302 | /* | |
3303 | * If this block group is smaller than 100 megs don't bother caching the | |
3304 | * block group. | |
3305 | */ | |
ee22184b | 3306 | if (block_group->key.offset < (100 * SZ_1M)) { |
0af3d00b JB |
3307 | spin_lock(&block_group->lock); |
3308 | block_group->disk_cache_state = BTRFS_DC_WRITTEN; | |
3309 | spin_unlock(&block_group->lock); | |
3310 | return 0; | |
3311 | } | |
3312 | ||
0c0ef4bc JB |
3313 | if (trans->aborted) |
3314 | return 0; | |
0af3d00b | 3315 | again: |
7949f339 | 3316 | inode = lookup_free_space_inode(block_group, path); |
0af3d00b JB |
3317 | if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) { |
3318 | ret = PTR_ERR(inode); | |
b3b4aa74 | 3319 | btrfs_release_path(path); |
0af3d00b JB |
3320 | goto out; |
3321 | } | |
3322 | ||
3323 | if (IS_ERR(inode)) { | |
3324 | BUG_ON(retries); | |
3325 | retries++; | |
3326 | ||
3327 | if (block_group->ro) | |
3328 | goto out_free; | |
3329 | ||
4ca75f1b | 3330 | ret = create_free_space_inode(trans, block_group, path); |
0af3d00b JB |
3331 | if (ret) |
3332 | goto out_free; | |
3333 | goto again; | |
3334 | } | |
3335 | ||
3336 | /* | |
3337 | * We want to set the generation to 0, that way if anything goes wrong | |
3338 | * from here on out we know not to trust this cache when we load up next | |
3339 | * time. | |
3340 | */ | |
3341 | BTRFS_I(inode)->generation = 0; | |
3342 | ret = btrfs_update_inode(trans, root, inode); | |
0c0ef4bc JB |
3343 | if (ret) { |
3344 | /* | |
3345 | * So theoretically we could recover from this, simply set the | |
3346 | * super cache generation to 0 so we know to invalidate the | |
3347 | * cache, but then we'd have to keep track of the block groups | |
3348 | * that fail this way so we know we _have_ to reset this cache | |
3349 | * before the next commit or risk reading stale cache. So to | |
3350 | * limit our exposure to horrible edge cases lets just abort the | |
3351 | * transaction, this only happens in really bad situations | |
3352 | * anyway. | |
3353 | */ | |
66642832 | 3354 | btrfs_abort_transaction(trans, ret); |
0c0ef4bc JB |
3355 | goto out_put; |
3356 | } | |
0af3d00b JB |
3357 | WARN_ON(ret); |
3358 | ||
8e138e0d JB |
3359 | /* We've already setup this transaction, go ahead and exit */ |
3360 | if (block_group->cache_generation == trans->transid && | |
3361 | i_size_read(inode)) { | |
3362 | dcs = BTRFS_DC_SETUP; | |
3363 | goto out_put; | |
3364 | } | |
3365 | ||
0af3d00b | 3366 | if (i_size_read(inode) > 0) { |
2ff7e61e | 3367 | ret = btrfs_check_trunc_cache_free_space(fs_info, |
0b246afa | 3368 | &fs_info->global_block_rsv); |
7b61cd92 MX |
3369 | if (ret) |
3370 | goto out_put; | |
3371 | ||
77ab86bf | 3372 | ret = btrfs_truncate_free_space_cache(trans, NULL, inode); |
0af3d00b JB |
3373 | if (ret) |
3374 | goto out_put; | |
3375 | } | |
3376 | ||
3377 | spin_lock(&block_group->lock); | |
cf7c1ef6 | 3378 | if (block_group->cached != BTRFS_CACHE_FINISHED || |
0b246afa | 3379 | !btrfs_test_opt(fs_info, SPACE_CACHE)) { |
cf7c1ef6 LB |
3380 | /* |
3381 | * don't bother trying to write stuff out _if_ | |
3382 | * a) we're not cached, | |
1a79c1f2 LB |
3383 | * b) we're with nospace_cache mount option, |
3384 | * c) we're with v2 space_cache (FREE_SPACE_TREE). | |
cf7c1ef6 | 3385 | */ |
2b20982e | 3386 | dcs = BTRFS_DC_WRITTEN; |
0af3d00b JB |
3387 | spin_unlock(&block_group->lock); |
3388 | goto out_put; | |
3389 | } | |
3390 | spin_unlock(&block_group->lock); | |
3391 | ||
2968b1f4 JB |
3392 | /* |
3393 | * We hit an ENOSPC when setting up the cache in this transaction, just | |
3394 | * skip doing the setup, we've already cleared the cache so we're safe. | |
3395 | */ | |
3396 | if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) { | |
3397 | ret = -ENOSPC; | |
3398 | goto out_put; | |
3399 | } | |
3400 | ||
6fc823b1 JB |
3401 | /* |
3402 | * Try to preallocate enough space based on how big the block group is. | |
3403 | * Keep in mind this has to include any pinned space which could end up | |
3404 | * taking up quite a bit since it's not folded into the other space | |
3405 | * cache. | |
3406 | */ | |
ee22184b | 3407 | num_pages = div_u64(block_group->key.offset, SZ_256M); |
0af3d00b JB |
3408 | if (!num_pages) |
3409 | num_pages = 1; | |
3410 | ||
0af3d00b | 3411 | num_pages *= 16; |
09cbfeaf | 3412 | num_pages *= PAGE_SIZE; |
0af3d00b | 3413 | |
364ecf36 | 3414 | ret = btrfs_check_data_free_space(inode, &data_reserved, 0, num_pages); |
0af3d00b JB |
3415 | if (ret) |
3416 | goto out_put; | |
3417 | ||
3418 | ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages, | |
3419 | num_pages, num_pages, | |
3420 | &alloc_hint); | |
2968b1f4 JB |
3421 | /* |
3422 | * Our cache requires contiguous chunks so that we don't modify a bunch | |
3423 | * of metadata or split extents when writing the cache out, which means | |
3424 | * we can enospc if we are heavily fragmented in addition to just normal | |
3425 | * out of space conditions. So if we hit this just skip setting up any | |
3426 | * other block groups for this transaction, maybe we'll unpin enough | |
3427 | * space the next time around. | |
3428 | */ | |
2b20982e JB |
3429 | if (!ret) |
3430 | dcs = BTRFS_DC_SETUP; | |
2968b1f4 JB |
3431 | else if (ret == -ENOSPC) |
3432 | set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags); | |
c09544e0 | 3433 | |
0af3d00b JB |
3434 | out_put: |
3435 | iput(inode); | |
3436 | out_free: | |
b3b4aa74 | 3437 | btrfs_release_path(path); |
0af3d00b JB |
3438 | out: |
3439 | spin_lock(&block_group->lock); | |
e65cbb94 | 3440 | if (!ret && dcs == BTRFS_DC_SETUP) |
5b0e95bf | 3441 | block_group->cache_generation = trans->transid; |
2b20982e | 3442 | block_group->disk_cache_state = dcs; |
0af3d00b JB |
3443 | spin_unlock(&block_group->lock); |
3444 | ||
364ecf36 | 3445 | extent_changeset_free(data_reserved); |
0af3d00b JB |
3446 | return ret; |
3447 | } | |
3448 | ||
bbebb3e0 | 3449 | int btrfs_setup_space_cache(struct btrfs_trans_handle *trans) |
dcdf7f6d | 3450 | { |
bbebb3e0 | 3451 | struct btrfs_fs_info *fs_info = trans->fs_info; |
dcdf7f6d JB |
3452 | struct btrfs_block_group_cache *cache, *tmp; |
3453 | struct btrfs_transaction *cur_trans = trans->transaction; | |
3454 | struct btrfs_path *path; | |
3455 | ||
3456 | if (list_empty(&cur_trans->dirty_bgs) || | |
0b246afa | 3457 | !btrfs_test_opt(fs_info, SPACE_CACHE)) |
dcdf7f6d JB |
3458 | return 0; |
3459 | ||
3460 | path = btrfs_alloc_path(); | |
3461 | if (!path) | |
3462 | return -ENOMEM; | |
3463 | ||
3464 | /* Could add new block groups, use _safe just in case */ | |
3465 | list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs, | |
3466 | dirty_list) { | |
3467 | if (cache->disk_cache_state == BTRFS_DC_CLEAR) | |
3468 | cache_save_setup(cache, trans, path); | |
3469 | } | |
3470 | ||
3471 | btrfs_free_path(path); | |
3472 | return 0; | |
3473 | } | |
3474 | ||
1bbc621e CM |
3475 | /* |
3476 | * transaction commit does final block group cache writeback during a | |
3477 | * critical section where nothing is allowed to change the FS. This is | |
3478 | * required in order for the cache to actually match the block group, | |
3479 | * but can introduce a lot of latency into the commit. | |
3480 | * | |
3481 | * So, btrfs_start_dirty_block_groups is here to kick off block group | |
3482 | * cache IO. There's a chance we'll have to redo some of it if the | |
3483 | * block group changes again during the commit, but it greatly reduces | |
3484 | * the commit latency by getting rid of the easy block groups while | |
3485 | * we're still allowing others to join the commit. | |
3486 | */ | |
21217054 | 3487 | int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans) |
9078a3e1 | 3488 | { |
21217054 | 3489 | struct btrfs_fs_info *fs_info = trans->fs_info; |
4a8c9a62 | 3490 | struct btrfs_block_group_cache *cache; |
ce93ec54 JB |
3491 | struct btrfs_transaction *cur_trans = trans->transaction; |
3492 | int ret = 0; | |
c9dc4c65 | 3493 | int should_put; |
1bbc621e CM |
3494 | struct btrfs_path *path = NULL; |
3495 | LIST_HEAD(dirty); | |
3496 | struct list_head *io = &cur_trans->io_bgs; | |
c9dc4c65 | 3497 | int num_started = 0; |
1bbc621e CM |
3498 | int loops = 0; |
3499 | ||
3500 | spin_lock(&cur_trans->dirty_bgs_lock); | |
b58d1a9e FM |
3501 | if (list_empty(&cur_trans->dirty_bgs)) { |
3502 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3503 | return 0; | |
1bbc621e | 3504 | } |
b58d1a9e | 3505 | list_splice_init(&cur_trans->dirty_bgs, &dirty); |
1bbc621e | 3506 | spin_unlock(&cur_trans->dirty_bgs_lock); |
ce93ec54 | 3507 | |
1bbc621e | 3508 | again: |
1bbc621e CM |
3509 | /* |
3510 | * make sure all the block groups on our dirty list actually | |
3511 | * exist | |
3512 | */ | |
6c686b35 | 3513 | btrfs_create_pending_block_groups(trans); |
1bbc621e CM |
3514 | |
3515 | if (!path) { | |
3516 | path = btrfs_alloc_path(); | |
3517 | if (!path) | |
3518 | return -ENOMEM; | |
3519 | } | |
3520 | ||
b58d1a9e FM |
3521 | /* |
3522 | * cache_write_mutex is here only to save us from balance or automatic | |
3523 | * removal of empty block groups deleting this block group while we are | |
3524 | * writing out the cache | |
3525 | */ | |
3526 | mutex_lock(&trans->transaction->cache_write_mutex); | |
1bbc621e | 3527 | while (!list_empty(&dirty)) { |
ba2c4d4e JB |
3528 | bool drop_reserve = true; |
3529 | ||
1bbc621e CM |
3530 | cache = list_first_entry(&dirty, |
3531 | struct btrfs_block_group_cache, | |
3532 | dirty_list); | |
1bbc621e CM |
3533 | /* |
3534 | * this can happen if something re-dirties a block | |
3535 | * group that is already under IO. Just wait for it to | |
3536 | * finish and then do it all again | |
3537 | */ | |
3538 | if (!list_empty(&cache->io_list)) { | |
3539 | list_del_init(&cache->io_list); | |
afdb5718 | 3540 | btrfs_wait_cache_io(trans, cache, path); |
1bbc621e CM |
3541 | btrfs_put_block_group(cache); |
3542 | } | |
3543 | ||
3544 | ||
3545 | /* | |
3546 | * btrfs_wait_cache_io uses the cache->dirty_list to decide | |
3547 | * if it should update the cache_state. Don't delete | |
3548 | * until after we wait. | |
3549 | * | |
3550 | * Since we're not running in the commit critical section | |
3551 | * we need the dirty_bgs_lock to protect from update_block_group | |
3552 | */ | |
3553 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3554 | list_del_init(&cache->dirty_list); | |
3555 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3556 | ||
3557 | should_put = 1; | |
3558 | ||
3559 | cache_save_setup(cache, trans, path); | |
3560 | ||
3561 | if (cache->disk_cache_state == BTRFS_DC_SETUP) { | |
3562 | cache->io_ctl.inode = NULL; | |
fe041534 | 3563 | ret = btrfs_write_out_cache(trans, cache, path); |
1bbc621e CM |
3564 | if (ret == 0 && cache->io_ctl.inode) { |
3565 | num_started++; | |
3566 | should_put = 0; | |
3567 | ||
3568 | /* | |
45ae2c18 NB |
3569 | * The cache_write_mutex is protecting the |
3570 | * io_list, also refer to the definition of | |
3571 | * btrfs_transaction::io_bgs for more details | |
1bbc621e CM |
3572 | */ |
3573 | list_add_tail(&cache->io_list, io); | |
3574 | } else { | |
3575 | /* | |
3576 | * if we failed to write the cache, the | |
3577 | * generation will be bad and life goes on | |
3578 | */ | |
3579 | ret = 0; | |
3580 | } | |
3581 | } | |
ff1f8250 | 3582 | if (!ret) { |
39db232d | 3583 | ret = write_one_cache_group(trans, path, cache); |
ff1f8250 FM |
3584 | /* |
3585 | * Our block group might still be attached to the list | |
3586 | * of new block groups in the transaction handle of some | |
3587 | * other task (struct btrfs_trans_handle->new_bgs). This | |
3588 | * means its block group item isn't yet in the extent | |
3589 | * tree. If this happens ignore the error, as we will | |
3590 | * try again later in the critical section of the | |
3591 | * transaction commit. | |
3592 | */ | |
3593 | if (ret == -ENOENT) { | |
3594 | ret = 0; | |
3595 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3596 | if (list_empty(&cache->dirty_list)) { | |
3597 | list_add_tail(&cache->dirty_list, | |
3598 | &cur_trans->dirty_bgs); | |
3599 | btrfs_get_block_group(cache); | |
ba2c4d4e | 3600 | drop_reserve = false; |
ff1f8250 FM |
3601 | } |
3602 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3603 | } else if (ret) { | |
66642832 | 3604 | btrfs_abort_transaction(trans, ret); |
ff1f8250 FM |
3605 | } |
3606 | } | |
1bbc621e | 3607 | |
52042d8e | 3608 | /* if it's not on the io list, we need to put the block group */ |
1bbc621e CM |
3609 | if (should_put) |
3610 | btrfs_put_block_group(cache); | |
ba2c4d4e JB |
3611 | if (drop_reserve) |
3612 | btrfs_delayed_refs_rsv_release(fs_info, 1); | |
1bbc621e CM |
3613 | |
3614 | if (ret) | |
3615 | break; | |
b58d1a9e FM |
3616 | |
3617 | /* | |
3618 | * Avoid blocking other tasks for too long. It might even save | |
3619 | * us from writing caches for block groups that are going to be | |
3620 | * removed. | |
3621 | */ | |
3622 | mutex_unlock(&trans->transaction->cache_write_mutex); | |
3623 | mutex_lock(&trans->transaction->cache_write_mutex); | |
1bbc621e | 3624 | } |
b58d1a9e | 3625 | mutex_unlock(&trans->transaction->cache_write_mutex); |
1bbc621e CM |
3626 | |
3627 | /* | |
3628 | * go through delayed refs for all the stuff we've just kicked off | |
3629 | * and then loop back (just once) | |
3630 | */ | |
c79a70b1 | 3631 | ret = btrfs_run_delayed_refs(trans, 0); |
1bbc621e CM |
3632 | if (!ret && loops == 0) { |
3633 | loops++; | |
3634 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3635 | list_splice_init(&cur_trans->dirty_bgs, &dirty); | |
b58d1a9e FM |
3636 | /* |
3637 | * dirty_bgs_lock protects us from concurrent block group | |
3638 | * deletes too (not just cache_write_mutex). | |
3639 | */ | |
3640 | if (!list_empty(&dirty)) { | |
3641 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3642 | goto again; | |
3643 | } | |
1bbc621e | 3644 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c79a1751 | 3645 | } else if (ret < 0) { |
2ff7e61e | 3646 | btrfs_cleanup_dirty_bgs(cur_trans, fs_info); |
1bbc621e CM |
3647 | } |
3648 | ||
3649 | btrfs_free_path(path); | |
3650 | return ret; | |
3651 | } | |
3652 | ||
5742d15f | 3653 | int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans) |
1bbc621e | 3654 | { |
5742d15f | 3655 | struct btrfs_fs_info *fs_info = trans->fs_info; |
1bbc621e CM |
3656 | struct btrfs_block_group_cache *cache; |
3657 | struct btrfs_transaction *cur_trans = trans->transaction; | |
3658 | int ret = 0; | |
3659 | int should_put; | |
3660 | struct btrfs_path *path; | |
3661 | struct list_head *io = &cur_trans->io_bgs; | |
3662 | int num_started = 0; | |
9078a3e1 CM |
3663 | |
3664 | path = btrfs_alloc_path(); | |
3665 | if (!path) | |
3666 | return -ENOMEM; | |
3667 | ||
ce93ec54 | 3668 | /* |
e44081ef FM |
3669 | * Even though we are in the critical section of the transaction commit, |
3670 | * we can still have concurrent tasks adding elements to this | |
3671 | * transaction's list of dirty block groups. These tasks correspond to | |
3672 | * endio free space workers started when writeback finishes for a | |
3673 | * space cache, which run inode.c:btrfs_finish_ordered_io(), and can | |
3674 | * allocate new block groups as a result of COWing nodes of the root | |
3675 | * tree when updating the free space inode. The writeback for the space | |
3676 | * caches is triggered by an earlier call to | |
3677 | * btrfs_start_dirty_block_groups() and iterations of the following | |
3678 | * loop. | |
3679 | * Also we want to do the cache_save_setup first and then run the | |
ce93ec54 JB |
3680 | * delayed refs to make sure we have the best chance at doing this all |
3681 | * in one shot. | |
3682 | */ | |
e44081ef | 3683 | spin_lock(&cur_trans->dirty_bgs_lock); |
ce93ec54 JB |
3684 | while (!list_empty(&cur_trans->dirty_bgs)) { |
3685 | cache = list_first_entry(&cur_trans->dirty_bgs, | |
3686 | struct btrfs_block_group_cache, | |
3687 | dirty_list); | |
c9dc4c65 CM |
3688 | |
3689 | /* | |
3690 | * this can happen if cache_save_setup re-dirties a block | |
3691 | * group that is already under IO. Just wait for it to | |
3692 | * finish and then do it all again | |
3693 | */ | |
3694 | if (!list_empty(&cache->io_list)) { | |
e44081ef | 3695 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 | 3696 | list_del_init(&cache->io_list); |
afdb5718 | 3697 | btrfs_wait_cache_io(trans, cache, path); |
c9dc4c65 | 3698 | btrfs_put_block_group(cache); |
e44081ef | 3699 | spin_lock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 CM |
3700 | } |
3701 | ||
1bbc621e CM |
3702 | /* |
3703 | * don't remove from the dirty list until after we've waited | |
3704 | * on any pending IO | |
3705 | */ | |
ce93ec54 | 3706 | list_del_init(&cache->dirty_list); |
e44081ef | 3707 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 CM |
3708 | should_put = 1; |
3709 | ||
1bbc621e | 3710 | cache_save_setup(cache, trans, path); |
c9dc4c65 | 3711 | |
ce93ec54 | 3712 | if (!ret) |
c79a70b1 | 3713 | ret = btrfs_run_delayed_refs(trans, |
2ff7e61e | 3714 | (unsigned long) -1); |
c9dc4c65 CM |
3715 | |
3716 | if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) { | |
3717 | cache->io_ctl.inode = NULL; | |
fe041534 | 3718 | ret = btrfs_write_out_cache(trans, cache, path); |
c9dc4c65 CM |
3719 | if (ret == 0 && cache->io_ctl.inode) { |
3720 | num_started++; | |
3721 | should_put = 0; | |
1bbc621e | 3722 | list_add_tail(&cache->io_list, io); |
c9dc4c65 CM |
3723 | } else { |
3724 | /* | |
3725 | * if we failed to write the cache, the | |
3726 | * generation will be bad and life goes on | |
3727 | */ | |
3728 | ret = 0; | |
3729 | } | |
3730 | } | |
ff1f8250 | 3731 | if (!ret) { |
39db232d | 3732 | ret = write_one_cache_group(trans, path, cache); |
2bc0bb5f FM |
3733 | /* |
3734 | * One of the free space endio workers might have | |
3735 | * created a new block group while updating a free space | |
3736 | * cache's inode (at inode.c:btrfs_finish_ordered_io()) | |
3737 | * and hasn't released its transaction handle yet, in | |
3738 | * which case the new block group is still attached to | |
3739 | * its transaction handle and its creation has not | |
3740 | * finished yet (no block group item in the extent tree | |
3741 | * yet, etc). If this is the case, wait for all free | |
3742 | * space endio workers to finish and retry. This is a | |
3743 | * a very rare case so no need for a more efficient and | |
3744 | * complex approach. | |
3745 | */ | |
3746 | if (ret == -ENOENT) { | |
3747 | wait_event(cur_trans->writer_wait, | |
3748 | atomic_read(&cur_trans->num_writers) == 1); | |
39db232d | 3749 | ret = write_one_cache_group(trans, path, cache); |
2bc0bb5f | 3750 | } |
ff1f8250 | 3751 | if (ret) |
66642832 | 3752 | btrfs_abort_transaction(trans, ret); |
ff1f8250 | 3753 | } |
c9dc4c65 CM |
3754 | |
3755 | /* if its not on the io list, we need to put the block group */ | |
3756 | if (should_put) | |
3757 | btrfs_put_block_group(cache); | |
ba2c4d4e | 3758 | btrfs_delayed_refs_rsv_release(fs_info, 1); |
e44081ef | 3759 | spin_lock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 | 3760 | } |
e44081ef | 3761 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 | 3762 | |
45ae2c18 NB |
3763 | /* |
3764 | * Refer to the definition of io_bgs member for details why it's safe | |
3765 | * to use it without any locking | |
3766 | */ | |
1bbc621e CM |
3767 | while (!list_empty(io)) { |
3768 | cache = list_first_entry(io, struct btrfs_block_group_cache, | |
c9dc4c65 CM |
3769 | io_list); |
3770 | list_del_init(&cache->io_list); | |
afdb5718 | 3771 | btrfs_wait_cache_io(trans, cache, path); |
0cb59c99 JB |
3772 | btrfs_put_block_group(cache); |
3773 | } | |
3774 | ||
9078a3e1 | 3775 | btrfs_free_path(path); |
ce93ec54 | 3776 | return ret; |
9078a3e1 CM |
3777 | } |
3778 | ||
2ff7e61e | 3779 | int btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr) |
d2fb3437 YZ |
3780 | { |
3781 | struct btrfs_block_group_cache *block_group; | |
3782 | int readonly = 0; | |
3783 | ||
0b246afa | 3784 | block_group = btrfs_lookup_block_group(fs_info, bytenr); |
d2fb3437 YZ |
3785 | if (!block_group || block_group->ro) |
3786 | readonly = 1; | |
3787 | if (block_group) | |
fa9c0d79 | 3788 | btrfs_put_block_group(block_group); |
d2fb3437 YZ |
3789 | return readonly; |
3790 | } | |
3791 | ||
f78c436c FM |
3792 | bool btrfs_inc_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr) |
3793 | { | |
3794 | struct btrfs_block_group_cache *bg; | |
3795 | bool ret = true; | |
3796 | ||
3797 | bg = btrfs_lookup_block_group(fs_info, bytenr); | |
3798 | if (!bg) | |
3799 | return false; | |
3800 | ||
3801 | spin_lock(&bg->lock); | |
3802 | if (bg->ro) | |
3803 | ret = false; | |
3804 | else | |
3805 | atomic_inc(&bg->nocow_writers); | |
3806 | spin_unlock(&bg->lock); | |
3807 | ||
3808 | /* no put on block group, done by btrfs_dec_nocow_writers */ | |
3809 | if (!ret) | |
3810 | btrfs_put_block_group(bg); | |
3811 | ||
3812 | return ret; | |
3813 | ||
3814 | } | |
3815 | ||
3816 | void btrfs_dec_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr) | |
3817 | { | |
3818 | struct btrfs_block_group_cache *bg; | |
3819 | ||
3820 | bg = btrfs_lookup_block_group(fs_info, bytenr); | |
3821 | ASSERT(bg); | |
3822 | if (atomic_dec_and_test(&bg->nocow_writers)) | |
4625956a | 3823 | wake_up_var(&bg->nocow_writers); |
f78c436c FM |
3824 | /* |
3825 | * Once for our lookup and once for the lookup done by a previous call | |
3826 | * to btrfs_inc_nocow_writers() | |
3827 | */ | |
3828 | btrfs_put_block_group(bg); | |
3829 | btrfs_put_block_group(bg); | |
3830 | } | |
3831 | ||
f78c436c FM |
3832 | void btrfs_wait_nocow_writers(struct btrfs_block_group_cache *bg) |
3833 | { | |
4625956a | 3834 | wait_var_event(&bg->nocow_writers, !atomic_read(&bg->nocow_writers)); |
f78c436c FM |
3835 | } |
3836 | ||
6ab0a202 JM |
3837 | static const char *alloc_name(u64 flags) |
3838 | { | |
3839 | switch (flags) { | |
3840 | case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA: | |
3841 | return "mixed"; | |
3842 | case BTRFS_BLOCK_GROUP_METADATA: | |
3843 | return "metadata"; | |
3844 | case BTRFS_BLOCK_GROUP_DATA: | |
3845 | return "data"; | |
3846 | case BTRFS_BLOCK_GROUP_SYSTEM: | |
3847 | return "system"; | |
3848 | default: | |
3849 | WARN_ON(1); | |
3850 | return "invalid-combination"; | |
3851 | }; | |
3852 | } | |
3853 | ||
4ca61683 | 3854 | static int create_space_info(struct btrfs_fs_info *info, u64 flags) |
2be12ef7 NB |
3855 | { |
3856 | ||
3857 | struct btrfs_space_info *space_info; | |
3858 | int i; | |
3859 | int ret; | |
3860 | ||
3861 | space_info = kzalloc(sizeof(*space_info), GFP_NOFS); | |
3862 | if (!space_info) | |
3863 | return -ENOMEM; | |
3864 | ||
3865 | ret = percpu_counter_init(&space_info->total_bytes_pinned, 0, | |
3866 | GFP_KERNEL); | |
3867 | if (ret) { | |
3868 | kfree(space_info); | |
3869 | return ret; | |
3870 | } | |
3871 | ||
3872 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) | |
3873 | INIT_LIST_HEAD(&space_info->block_groups[i]); | |
3874 | init_rwsem(&space_info->groups_sem); | |
3875 | spin_lock_init(&space_info->lock); | |
3876 | space_info->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK; | |
3877 | space_info->force_alloc = CHUNK_ALLOC_NO_FORCE; | |
3878 | init_waitqueue_head(&space_info->wait); | |
3879 | INIT_LIST_HEAD(&space_info->ro_bgs); | |
3880 | INIT_LIST_HEAD(&space_info->tickets); | |
3881 | INIT_LIST_HEAD(&space_info->priority_tickets); | |
3882 | ||
3883 | ret = kobject_init_and_add(&space_info->kobj, &space_info_ktype, | |
3884 | info->space_info_kobj, "%s", | |
3885 | alloc_name(space_info->flags)); | |
3886 | if (ret) { | |
450ff834 | 3887 | kobject_put(&space_info->kobj); |
2be12ef7 NB |
3888 | return ret; |
3889 | } | |
3890 | ||
2be12ef7 NB |
3891 | list_add_rcu(&space_info->list, &info->space_info); |
3892 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
3893 | info->data_sinfo = space_info; | |
3894 | ||
3895 | return ret; | |
3896 | } | |
3897 | ||
d2006e6d | 3898 | static void update_space_info(struct btrfs_fs_info *info, u64 flags, |
593060d7 | 3899 | u64 total_bytes, u64 bytes_used, |
e40edf2d | 3900 | u64 bytes_readonly, |
593060d7 CM |
3901 | struct btrfs_space_info **space_info) |
3902 | { | |
3903 | struct btrfs_space_info *found; | |
b742bb82 YZ |
3904 | int factor; |
3905 | ||
46df06b8 | 3906 | factor = btrfs_bg_type_to_factor(flags); |
593060d7 CM |
3907 | |
3908 | found = __find_space_info(info, flags); | |
d2006e6d NB |
3909 | ASSERT(found); |
3910 | spin_lock(&found->lock); | |
3911 | found->total_bytes += total_bytes; | |
3912 | found->disk_total += total_bytes * factor; | |
3913 | found->bytes_used += bytes_used; | |
3914 | found->disk_used += bytes_used * factor; | |
3915 | found->bytes_readonly += bytes_readonly; | |
3916 | if (total_bytes > 0) | |
3917 | found->full = 0; | |
3918 | space_info_add_new_bytes(info, found, total_bytes - | |
3919 | bytes_used - bytes_readonly); | |
3920 | spin_unlock(&found->lock); | |
3921 | *space_info = found; | |
593060d7 CM |
3922 | } |
3923 | ||
8790d502 CM |
3924 | static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
3925 | { | |
899c81ea ID |
3926 | u64 extra_flags = chunk_to_extended(flags) & |
3927 | BTRFS_EXTENDED_PROFILE_MASK; | |
a46d11a8 | 3928 | |
de98ced9 | 3929 | write_seqlock(&fs_info->profiles_lock); |
a46d11a8 ID |
3930 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
3931 | fs_info->avail_data_alloc_bits |= extra_flags; | |
3932 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
3933 | fs_info->avail_metadata_alloc_bits |= extra_flags; | |
3934 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
3935 | fs_info->avail_system_alloc_bits |= extra_flags; | |
de98ced9 | 3936 | write_sequnlock(&fs_info->profiles_lock); |
8790d502 | 3937 | } |
593060d7 | 3938 | |
fc67c450 ID |
3939 | /* |
3940 | * returns target flags in extended format or 0 if restripe for this | |
3941 | * chunk_type is not in progress | |
c6664b42 | 3942 | * |
dccdb07b | 3943 | * should be called with balance_lock held |
fc67c450 ID |
3944 | */ |
3945 | static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags) | |
3946 | { | |
3947 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
3948 | u64 target = 0; | |
3949 | ||
fc67c450 ID |
3950 | if (!bctl) |
3951 | return 0; | |
3952 | ||
3953 | if (flags & BTRFS_BLOCK_GROUP_DATA && | |
3954 | bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
3955 | target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target; | |
3956 | } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM && | |
3957 | bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
3958 | target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target; | |
3959 | } else if (flags & BTRFS_BLOCK_GROUP_METADATA && | |
3960 | bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
3961 | target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target; | |
3962 | } | |
3963 | ||
3964 | return target; | |
3965 | } | |
3966 | ||
a46d11a8 ID |
3967 | /* |
3968 | * @flags: available profiles in extended format (see ctree.h) | |
3969 | * | |
e4d8ec0f ID |
3970 | * Returns reduced profile in chunk format. If profile changing is in |
3971 | * progress (either running or paused) picks the target profile (if it's | |
3972 | * already available), otherwise falls back to plain reducing. | |
a46d11a8 | 3973 | */ |
2ff7e61e | 3974 | static u64 btrfs_reduce_alloc_profile(struct btrfs_fs_info *fs_info, u64 flags) |
ec44a35c | 3975 | { |
0b246afa | 3976 | u64 num_devices = fs_info->fs_devices->rw_devices; |
fc67c450 | 3977 | u64 target; |
9c170b26 ZL |
3978 | u64 raid_type; |
3979 | u64 allowed = 0; | |
a061fc8d | 3980 | |
fc67c450 ID |
3981 | /* |
3982 | * see if restripe for this chunk_type is in progress, if so | |
3983 | * try to reduce to the target profile | |
3984 | */ | |
0b246afa JM |
3985 | spin_lock(&fs_info->balance_lock); |
3986 | target = get_restripe_target(fs_info, flags); | |
fc67c450 ID |
3987 | if (target) { |
3988 | /* pick target profile only if it's already available */ | |
3989 | if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) { | |
0b246afa | 3990 | spin_unlock(&fs_info->balance_lock); |
fc67c450 | 3991 | return extended_to_chunk(target); |
e4d8ec0f ID |
3992 | } |
3993 | } | |
0b246afa | 3994 | spin_unlock(&fs_info->balance_lock); |
e4d8ec0f | 3995 | |
53b381b3 | 3996 | /* First, mask out the RAID levels which aren't possible */ |
9c170b26 ZL |
3997 | for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) { |
3998 | if (num_devices >= btrfs_raid_array[raid_type].devs_min) | |
41a6e891 | 3999 | allowed |= btrfs_raid_array[raid_type].bg_flag; |
9c170b26 ZL |
4000 | } |
4001 | allowed &= flags; | |
4002 | ||
4003 | if (allowed & BTRFS_BLOCK_GROUP_RAID6) | |
4004 | allowed = BTRFS_BLOCK_GROUP_RAID6; | |
4005 | else if (allowed & BTRFS_BLOCK_GROUP_RAID5) | |
4006 | allowed = BTRFS_BLOCK_GROUP_RAID5; | |
4007 | else if (allowed & BTRFS_BLOCK_GROUP_RAID10) | |
4008 | allowed = BTRFS_BLOCK_GROUP_RAID10; | |
4009 | else if (allowed & BTRFS_BLOCK_GROUP_RAID1) | |
4010 | allowed = BTRFS_BLOCK_GROUP_RAID1; | |
4011 | else if (allowed & BTRFS_BLOCK_GROUP_RAID0) | |
4012 | allowed = BTRFS_BLOCK_GROUP_RAID0; | |
4013 | ||
4014 | flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK; | |
4015 | ||
4016 | return extended_to_chunk(flags | allowed); | |
ec44a35c CM |
4017 | } |
4018 | ||
2ff7e61e | 4019 | static u64 get_alloc_profile(struct btrfs_fs_info *fs_info, u64 orig_flags) |
6a63209f | 4020 | { |
de98ced9 | 4021 | unsigned seq; |
f8213bdc | 4022 | u64 flags; |
de98ced9 MX |
4023 | |
4024 | do { | |
f8213bdc | 4025 | flags = orig_flags; |
0b246afa | 4026 | seq = read_seqbegin(&fs_info->profiles_lock); |
de98ced9 MX |
4027 | |
4028 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
0b246afa | 4029 | flags |= fs_info->avail_data_alloc_bits; |
de98ced9 | 4030 | else if (flags & BTRFS_BLOCK_GROUP_SYSTEM) |
0b246afa | 4031 | flags |= fs_info->avail_system_alloc_bits; |
de98ced9 | 4032 | else if (flags & BTRFS_BLOCK_GROUP_METADATA) |
0b246afa JM |
4033 | flags |= fs_info->avail_metadata_alloc_bits; |
4034 | } while (read_seqretry(&fs_info->profiles_lock, seq)); | |
6fef8df1 | 4035 | |
2ff7e61e | 4036 | return btrfs_reduce_alloc_profile(fs_info, flags); |
6a63209f JB |
4037 | } |
4038 | ||
1b86826d | 4039 | static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data) |
9ed74f2d | 4040 | { |
0b246afa | 4041 | struct btrfs_fs_info *fs_info = root->fs_info; |
b742bb82 | 4042 | u64 flags; |
53b381b3 | 4043 | u64 ret; |
9ed74f2d | 4044 | |
b742bb82 YZ |
4045 | if (data) |
4046 | flags = BTRFS_BLOCK_GROUP_DATA; | |
0b246afa | 4047 | else if (root == fs_info->chunk_root) |
b742bb82 | 4048 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
9ed74f2d | 4049 | else |
b742bb82 | 4050 | flags = BTRFS_BLOCK_GROUP_METADATA; |
9ed74f2d | 4051 | |
2ff7e61e | 4052 | ret = get_alloc_profile(fs_info, flags); |
53b381b3 | 4053 | return ret; |
6a63209f | 4054 | } |
9ed74f2d | 4055 | |
1b86826d JM |
4056 | u64 btrfs_data_alloc_profile(struct btrfs_fs_info *fs_info) |
4057 | { | |
4058 | return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_DATA); | |
4059 | } | |
4060 | ||
4061 | u64 btrfs_metadata_alloc_profile(struct btrfs_fs_info *fs_info) | |
4062 | { | |
4063 | return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_METADATA); | |
4064 | } | |
4065 | ||
4066 | u64 btrfs_system_alloc_profile(struct btrfs_fs_info *fs_info) | |
4067 | { | |
4068 | return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); | |
4069 | } | |
4070 | ||
4136135b LB |
4071 | static u64 btrfs_space_info_used(struct btrfs_space_info *s_info, |
4072 | bool may_use_included) | |
4073 | { | |
4074 | ASSERT(s_info); | |
4075 | return s_info->bytes_used + s_info->bytes_reserved + | |
4076 | s_info->bytes_pinned + s_info->bytes_readonly + | |
4077 | (may_use_included ? s_info->bytes_may_use : 0); | |
4078 | } | |
4079 | ||
04f4f916 | 4080 | int btrfs_alloc_data_chunk_ondemand(struct btrfs_inode *inode, u64 bytes) |
6a63209f | 4081 | { |
04f4f916 | 4082 | struct btrfs_root *root = inode->root; |
b4d7c3c9 | 4083 | struct btrfs_fs_info *fs_info = root->fs_info; |
1174cade | 4084 | struct btrfs_space_info *data_sinfo = fs_info->data_sinfo; |
ab6e2410 | 4085 | u64 used; |
94b947b2 | 4086 | int ret = 0; |
c99f1b0c ZL |
4087 | int need_commit = 2; |
4088 | int have_pinned_space; | |
6a63209f | 4089 | |
6a63209f | 4090 | /* make sure bytes are sectorsize aligned */ |
0b246afa | 4091 | bytes = ALIGN(bytes, fs_info->sectorsize); |
6a63209f | 4092 | |
9dced186 | 4093 | if (btrfs_is_free_space_inode(inode)) { |
c99f1b0c | 4094 | need_commit = 0; |
9dced186 | 4095 | ASSERT(current->journal_info); |
0af3d00b JB |
4096 | } |
4097 | ||
6a63209f JB |
4098 | again: |
4099 | /* make sure we have enough space to handle the data first */ | |
4100 | spin_lock(&data_sinfo->lock); | |
4136135b | 4101 | used = btrfs_space_info_used(data_sinfo, true); |
ab6e2410 JB |
4102 | |
4103 | if (used + bytes > data_sinfo->total_bytes) { | |
4e06bdd6 | 4104 | struct btrfs_trans_handle *trans; |
9ed74f2d | 4105 | |
6a63209f JB |
4106 | /* |
4107 | * if we don't have enough free bytes in this space then we need | |
4108 | * to alloc a new chunk. | |
4109 | */ | |
b9fd47cd | 4110 | if (!data_sinfo->full) { |
6a63209f | 4111 | u64 alloc_target; |
9ed74f2d | 4112 | |
0e4f8f88 | 4113 | data_sinfo->force_alloc = CHUNK_ALLOC_FORCE; |
6a63209f | 4114 | spin_unlock(&data_sinfo->lock); |
1174cade | 4115 | |
1b86826d | 4116 | alloc_target = btrfs_data_alloc_profile(fs_info); |
9dced186 MX |
4117 | /* |
4118 | * It is ugly that we don't call nolock join | |
4119 | * transaction for the free space inode case here. | |
4120 | * But it is safe because we only do the data space | |
4121 | * reservation for the free space cache in the | |
4122 | * transaction context, the common join transaction | |
4123 | * just increase the counter of the current transaction | |
4124 | * handler, doesn't try to acquire the trans_lock of | |
4125 | * the fs. | |
4126 | */ | |
7a7eaa40 | 4127 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
4128 | if (IS_ERR(trans)) |
4129 | return PTR_ERR(trans); | |
9ed74f2d | 4130 | |
01458828 | 4131 | ret = do_chunk_alloc(trans, alloc_target, |
0e4f8f88 | 4132 | CHUNK_ALLOC_NO_FORCE); |
3a45bb20 | 4133 | btrfs_end_transaction(trans); |
d52a5b5f MX |
4134 | if (ret < 0) { |
4135 | if (ret != -ENOSPC) | |
4136 | return ret; | |
c99f1b0c ZL |
4137 | else { |
4138 | have_pinned_space = 1; | |
d52a5b5f | 4139 | goto commit_trans; |
c99f1b0c | 4140 | } |
d52a5b5f | 4141 | } |
9ed74f2d | 4142 | |
6a63209f JB |
4143 | goto again; |
4144 | } | |
f2bb8f5c JB |
4145 | |
4146 | /* | |
b150a4f1 | 4147 | * If we don't have enough pinned space to deal with this |
94b947b2 ZL |
4148 | * allocation, and no removed chunk in current transaction, |
4149 | * don't bother committing the transaction. | |
f2bb8f5c | 4150 | */ |
dec59fa3 | 4151 | have_pinned_space = __percpu_counter_compare( |
c99f1b0c | 4152 | &data_sinfo->total_bytes_pinned, |
dec59fa3 EL |
4153 | used + bytes - data_sinfo->total_bytes, |
4154 | BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
6a63209f | 4155 | spin_unlock(&data_sinfo->lock); |
6a63209f | 4156 | |
4e06bdd6 | 4157 | /* commit the current transaction and try again */ |
d52a5b5f | 4158 | commit_trans: |
92e2f7e3 | 4159 | if (need_commit) { |
c99f1b0c | 4160 | need_commit--; |
b150a4f1 | 4161 | |
e1746e83 | 4162 | if (need_commit > 0) { |
82b3e53b | 4163 | btrfs_start_delalloc_roots(fs_info, -1); |
6374e57a | 4164 | btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, |
0b246afa | 4165 | (u64)-1); |
e1746e83 | 4166 | } |
9a4e7276 | 4167 | |
7a7eaa40 | 4168 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
4169 | if (IS_ERR(trans)) |
4170 | return PTR_ERR(trans); | |
c99f1b0c | 4171 | if (have_pinned_space >= 0 || |
3204d33c JB |
4172 | test_bit(BTRFS_TRANS_HAVE_FREE_BGS, |
4173 | &trans->transaction->flags) || | |
c99f1b0c | 4174 | need_commit > 0) { |
3a45bb20 | 4175 | ret = btrfs_commit_transaction(trans); |
94b947b2 ZL |
4176 | if (ret) |
4177 | return ret; | |
d7c15171 | 4178 | /* |
c2d6cb16 FM |
4179 | * The cleaner kthread might still be doing iput |
4180 | * operations. Wait for it to finish so that | |
034f784d JB |
4181 | * more space is released. We don't need to |
4182 | * explicitly run the delayed iputs here because | |
4183 | * the commit_transaction would have woken up | |
4184 | * the cleaner. | |
d7c15171 | 4185 | */ |
034f784d JB |
4186 | ret = btrfs_wait_on_delayed_iputs(fs_info); |
4187 | if (ret) | |
4188 | return ret; | |
94b947b2 ZL |
4189 | goto again; |
4190 | } else { | |
3a45bb20 | 4191 | btrfs_end_transaction(trans); |
94b947b2 | 4192 | } |
4e06bdd6 | 4193 | } |
9ed74f2d | 4194 | |
0b246afa | 4195 | trace_btrfs_space_reservation(fs_info, |
cab45e22 JM |
4196 | "space_info:enospc", |
4197 | data_sinfo->flags, bytes, 1); | |
6a63209f JB |
4198 | return -ENOSPC; |
4199 | } | |
9f9b8e8d | 4200 | update_bytes_may_use(data_sinfo, bytes); |
0b246afa | 4201 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 4202 | data_sinfo->flags, bytes, 1); |
6a63209f | 4203 | spin_unlock(&data_sinfo->lock); |
6a63209f | 4204 | |
4559b0a7 | 4205 | return 0; |
9ed74f2d | 4206 | } |
6a63209f | 4207 | |
364ecf36 QW |
4208 | int btrfs_check_data_free_space(struct inode *inode, |
4209 | struct extent_changeset **reserved, u64 start, u64 len) | |
4ceff079 | 4210 | { |
0b246afa | 4211 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4ceff079 QW |
4212 | int ret; |
4213 | ||
4214 | /* align the range */ | |
0b246afa JM |
4215 | len = round_up(start + len, fs_info->sectorsize) - |
4216 | round_down(start, fs_info->sectorsize); | |
4217 | start = round_down(start, fs_info->sectorsize); | |
4ceff079 | 4218 | |
04f4f916 | 4219 | ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode), len); |
4ceff079 QW |
4220 | if (ret < 0) |
4221 | return ret; | |
4222 | ||
1e5ec2e7 | 4223 | /* Use new btrfs_qgroup_reserve_data to reserve precious data space. */ |
364ecf36 | 4224 | ret = btrfs_qgroup_reserve_data(inode, reserved, start, len); |
7bc329c1 | 4225 | if (ret < 0) |
1e5ec2e7 | 4226 | btrfs_free_reserved_data_space_noquota(inode, start, len); |
364ecf36 QW |
4227 | else |
4228 | ret = 0; | |
4ceff079 QW |
4229 | return ret; |
4230 | } | |
4231 | ||
4ceff079 QW |
4232 | /* |
4233 | * Called if we need to clear a data reservation for this inode | |
4234 | * Normally in a error case. | |
4235 | * | |
51773bec QW |
4236 | * This one will *NOT* use accurate qgroup reserved space API, just for case |
4237 | * which we can't sleep and is sure it won't affect qgroup reserved space. | |
4238 | * Like clear_bit_hook(). | |
4ceff079 | 4239 | */ |
51773bec QW |
4240 | void btrfs_free_reserved_data_space_noquota(struct inode *inode, u64 start, |
4241 | u64 len) | |
4ceff079 | 4242 | { |
0b246afa | 4243 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4ceff079 QW |
4244 | struct btrfs_space_info *data_sinfo; |
4245 | ||
4246 | /* Make sure the range is aligned to sectorsize */ | |
0b246afa JM |
4247 | len = round_up(start + len, fs_info->sectorsize) - |
4248 | round_down(start, fs_info->sectorsize); | |
4249 | start = round_down(start, fs_info->sectorsize); | |
4ceff079 | 4250 | |
0b246afa | 4251 | data_sinfo = fs_info->data_sinfo; |
4ceff079 | 4252 | spin_lock(&data_sinfo->lock); |
9f9b8e8d | 4253 | update_bytes_may_use(data_sinfo, -len); |
0b246afa | 4254 | trace_btrfs_space_reservation(fs_info, "space_info", |
4ceff079 QW |
4255 | data_sinfo->flags, len, 0); |
4256 | spin_unlock(&data_sinfo->lock); | |
4257 | } | |
4258 | ||
51773bec QW |
4259 | /* |
4260 | * Called if we need to clear a data reservation for this inode | |
4261 | * Normally in a error case. | |
4262 | * | |
01327610 | 4263 | * This one will handle the per-inode data rsv map for accurate reserved |
51773bec QW |
4264 | * space framework. |
4265 | */ | |
bc42bda2 QW |
4266 | void btrfs_free_reserved_data_space(struct inode *inode, |
4267 | struct extent_changeset *reserved, u64 start, u64 len) | |
51773bec | 4268 | { |
0c476a5d JM |
4269 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4270 | ||
4271 | /* Make sure the range is aligned to sectorsize */ | |
da17066c JM |
4272 | len = round_up(start + len, root->fs_info->sectorsize) - |
4273 | round_down(start, root->fs_info->sectorsize); | |
4274 | start = round_down(start, root->fs_info->sectorsize); | |
0c476a5d | 4275 | |
51773bec | 4276 | btrfs_free_reserved_data_space_noquota(inode, start, len); |
bc42bda2 | 4277 | btrfs_qgroup_free_data(inode, reserved, start, len); |
51773bec QW |
4278 | } |
4279 | ||
97e728d4 | 4280 | static void force_metadata_allocation(struct btrfs_fs_info *info) |
e3ccfa98 | 4281 | { |
97e728d4 JB |
4282 | struct list_head *head = &info->space_info; |
4283 | struct btrfs_space_info *found; | |
e3ccfa98 | 4284 | |
97e728d4 JB |
4285 | rcu_read_lock(); |
4286 | list_for_each_entry_rcu(found, head, list) { | |
4287 | if (found->flags & BTRFS_BLOCK_GROUP_METADATA) | |
0e4f8f88 | 4288 | found->force_alloc = CHUNK_ALLOC_FORCE; |
e3ccfa98 | 4289 | } |
97e728d4 | 4290 | rcu_read_unlock(); |
e3ccfa98 JB |
4291 | } |
4292 | ||
3c76cd84 MX |
4293 | static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global) |
4294 | { | |
4295 | return (global->size << 1); | |
4296 | } | |
4297 | ||
2ff7e61e | 4298 | static int should_alloc_chunk(struct btrfs_fs_info *fs_info, |
698d0082 | 4299 | struct btrfs_space_info *sinfo, int force) |
32c00aff | 4300 | { |
8d8aafee | 4301 | u64 bytes_used = btrfs_space_info_used(sinfo, false); |
e5bc2458 | 4302 | u64 thresh; |
e3ccfa98 | 4303 | |
0e4f8f88 CM |
4304 | if (force == CHUNK_ALLOC_FORCE) |
4305 | return 1; | |
4306 | ||
4307 | /* | |
4308 | * in limited mode, we want to have some free space up to | |
4309 | * about 1% of the FS size. | |
4310 | */ | |
4311 | if (force == CHUNK_ALLOC_LIMITED) { | |
0b246afa | 4312 | thresh = btrfs_super_total_bytes(fs_info->super_copy); |
ee22184b | 4313 | thresh = max_t(u64, SZ_64M, div_factor_fine(thresh, 1)); |
0e4f8f88 | 4314 | |
8d8aafee | 4315 | if (sinfo->total_bytes - bytes_used < thresh) |
0e4f8f88 CM |
4316 | return 1; |
4317 | } | |
0e4f8f88 | 4318 | |
8d8aafee | 4319 | if (bytes_used + SZ_2M < div_factor(sinfo->total_bytes, 8)) |
14ed0ca6 | 4320 | return 0; |
424499db | 4321 | return 1; |
32c00aff JB |
4322 | } |
4323 | ||
2ff7e61e | 4324 | static u64 get_profile_num_devs(struct btrfs_fs_info *fs_info, u64 type) |
15d1ff81 LB |
4325 | { |
4326 | u64 num_dev; | |
4327 | ||
53b381b3 DW |
4328 | if (type & (BTRFS_BLOCK_GROUP_RAID10 | |
4329 | BTRFS_BLOCK_GROUP_RAID0 | | |
4330 | BTRFS_BLOCK_GROUP_RAID5 | | |
4331 | BTRFS_BLOCK_GROUP_RAID6)) | |
0b246afa | 4332 | num_dev = fs_info->fs_devices->rw_devices; |
15d1ff81 LB |
4333 | else if (type & BTRFS_BLOCK_GROUP_RAID1) |
4334 | num_dev = 2; | |
4335 | else | |
4336 | num_dev = 1; /* DUP or single */ | |
4337 | ||
39c2d7fa | 4338 | return num_dev; |
15d1ff81 LB |
4339 | } |
4340 | ||
39c2d7fa FM |
4341 | /* |
4342 | * If @is_allocation is true, reserve space in the system space info necessary | |
4343 | * for allocating a chunk, otherwise if it's false, reserve space necessary for | |
4344 | * removing a chunk. | |
4345 | */ | |
451a2c13 | 4346 | void check_system_chunk(struct btrfs_trans_handle *trans, u64 type) |
15d1ff81 | 4347 | { |
451a2c13 | 4348 | struct btrfs_fs_info *fs_info = trans->fs_info; |
15d1ff81 LB |
4349 | struct btrfs_space_info *info; |
4350 | u64 left; | |
4351 | u64 thresh; | |
4fbcdf66 | 4352 | int ret = 0; |
39c2d7fa | 4353 | u64 num_devs; |
4fbcdf66 FM |
4354 | |
4355 | /* | |
4356 | * Needed because we can end up allocating a system chunk and for an | |
4357 | * atomic and race free space reservation in the chunk block reserve. | |
4358 | */ | |
a32bf9a3 | 4359 | lockdep_assert_held(&fs_info->chunk_mutex); |
15d1ff81 | 4360 | |
0b246afa | 4361 | info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); |
15d1ff81 | 4362 | spin_lock(&info->lock); |
4136135b | 4363 | left = info->total_bytes - btrfs_space_info_used(info, true); |
15d1ff81 LB |
4364 | spin_unlock(&info->lock); |
4365 | ||
2ff7e61e | 4366 | num_devs = get_profile_num_devs(fs_info, type); |
39c2d7fa FM |
4367 | |
4368 | /* num_devs device items to update and 1 chunk item to add or remove */ | |
0b246afa JM |
4369 | thresh = btrfs_calc_trunc_metadata_size(fs_info, num_devs) + |
4370 | btrfs_calc_trans_metadata_size(fs_info, 1); | |
39c2d7fa | 4371 | |
0b246afa JM |
4372 | if (left < thresh && btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
4373 | btrfs_info(fs_info, "left=%llu, need=%llu, flags=%llu", | |
4374 | left, thresh, type); | |
4375 | dump_space_info(fs_info, info, 0, 0); | |
15d1ff81 LB |
4376 | } |
4377 | ||
4378 | if (left < thresh) { | |
1b86826d | 4379 | u64 flags = btrfs_system_alloc_profile(fs_info); |
15d1ff81 | 4380 | |
4fbcdf66 FM |
4381 | /* |
4382 | * Ignore failure to create system chunk. We might end up not | |
4383 | * needing it, as we might not need to COW all nodes/leafs from | |
4384 | * the paths we visit in the chunk tree (they were already COWed | |
4385 | * or created in the current transaction for example). | |
4386 | */ | |
c216b203 | 4387 | ret = btrfs_alloc_chunk(trans, flags); |
4fbcdf66 FM |
4388 | } |
4389 | ||
4390 | if (!ret) { | |
0b246afa JM |
4391 | ret = btrfs_block_rsv_add(fs_info->chunk_root, |
4392 | &fs_info->chunk_block_rsv, | |
4fbcdf66 FM |
4393 | thresh, BTRFS_RESERVE_NO_FLUSH); |
4394 | if (!ret) | |
4395 | trans->chunk_bytes_reserved += thresh; | |
15d1ff81 LB |
4396 | } |
4397 | } | |
4398 | ||
28b737f6 LB |
4399 | /* |
4400 | * If force is CHUNK_ALLOC_FORCE: | |
4401 | * - return 1 if it successfully allocates a chunk, | |
4402 | * - return errors including -ENOSPC otherwise. | |
4403 | * If force is NOT CHUNK_ALLOC_FORCE: | |
4404 | * - return 0 if it doesn't need to allocate a new chunk, | |
4405 | * - return 1 if it successfully allocates a chunk, | |
4406 | * - return errors including -ENOSPC otherwise. | |
4407 | */ | |
01458828 NB |
4408 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, u64 flags, |
4409 | int force) | |
9ed74f2d | 4410 | { |
01458828 | 4411 | struct btrfs_fs_info *fs_info = trans->fs_info; |
6324fbf3 | 4412 | struct btrfs_space_info *space_info; |
2556fbb0 NB |
4413 | bool wait_for_alloc = false; |
4414 | bool should_alloc = false; | |
9ed74f2d | 4415 | int ret = 0; |
9ed74f2d | 4416 | |
c6b305a8 JB |
4417 | /* Don't re-enter if we're already allocating a chunk */ |
4418 | if (trans->allocating_chunk) | |
4419 | return -ENOSPC; | |
4420 | ||
0b246afa | 4421 | space_info = __find_space_info(fs_info, flags); |
dc2d3005 | 4422 | ASSERT(space_info); |
9ed74f2d | 4423 | |
2556fbb0 NB |
4424 | do { |
4425 | spin_lock(&space_info->lock); | |
4426 | if (force < space_info->force_alloc) | |
4427 | force = space_info->force_alloc; | |
4428 | should_alloc = should_alloc_chunk(fs_info, space_info, force); | |
4429 | if (space_info->full) { | |
4430 | /* No more free physical space */ | |
4431 | if (should_alloc) | |
4432 | ret = -ENOSPC; | |
4433 | else | |
4434 | ret = 0; | |
4435 | spin_unlock(&space_info->lock); | |
4436 | return ret; | |
4437 | } else if (!should_alloc) { | |
4438 | spin_unlock(&space_info->lock); | |
4439 | return 0; | |
4440 | } else if (space_info->chunk_alloc) { | |
4441 | /* | |
4442 | * Someone is already allocating, so we need to block | |
4443 | * until this someone is finished and then loop to | |
4444 | * recheck if we should continue with our allocation | |
4445 | * attempt. | |
4446 | */ | |
4447 | wait_for_alloc = true; | |
4448 | spin_unlock(&space_info->lock); | |
4449 | mutex_lock(&fs_info->chunk_mutex); | |
4450 | mutex_unlock(&fs_info->chunk_mutex); | |
4451 | } else { | |
4452 | /* Proceed with allocation */ | |
4453 | space_info->chunk_alloc = 1; | |
4454 | wait_for_alloc = false; | |
4455 | spin_unlock(&space_info->lock); | |
4456 | } | |
6d74119f | 4457 | |
1e1c50a9 | 4458 | cond_resched(); |
2556fbb0 | 4459 | } while (wait_for_alloc); |
6d74119f | 4460 | |
2556fbb0 | 4461 | mutex_lock(&fs_info->chunk_mutex); |
c6b305a8 JB |
4462 | trans->allocating_chunk = true; |
4463 | ||
67377734 JB |
4464 | /* |
4465 | * If we have mixed data/metadata chunks we want to make sure we keep | |
4466 | * allocating mixed chunks instead of individual chunks. | |
4467 | */ | |
4468 | if (btrfs_mixed_space_info(space_info)) | |
4469 | flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA); | |
4470 | ||
97e728d4 JB |
4471 | /* |
4472 | * if we're doing a data chunk, go ahead and make sure that | |
4473 | * we keep a reasonable number of metadata chunks allocated in the | |
4474 | * FS as well. | |
4475 | */ | |
9ed74f2d | 4476 | if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) { |
97e728d4 JB |
4477 | fs_info->data_chunk_allocations++; |
4478 | if (!(fs_info->data_chunk_allocations % | |
4479 | fs_info->metadata_ratio)) | |
4480 | force_metadata_allocation(fs_info); | |
9ed74f2d JB |
4481 | } |
4482 | ||
15d1ff81 LB |
4483 | /* |
4484 | * Check if we have enough space in SYSTEM chunk because we may need | |
4485 | * to update devices. | |
4486 | */ | |
451a2c13 | 4487 | check_system_chunk(trans, flags); |
15d1ff81 | 4488 | |
c216b203 | 4489 | ret = btrfs_alloc_chunk(trans, flags); |
c6b305a8 | 4490 | trans->allocating_chunk = false; |
92b8e897 | 4491 | |
9ed74f2d | 4492 | spin_lock(&space_info->lock); |
57f1642e NB |
4493 | if (ret < 0) { |
4494 | if (ret == -ENOSPC) | |
4495 | space_info->full = 1; | |
4496 | else | |
4497 | goto out; | |
4498 | } else { | |
424499db | 4499 | ret = 1; |
21a94f7a | 4500 | space_info->max_extent_size = 0; |
57f1642e | 4501 | } |
6d74119f | 4502 | |
0e4f8f88 | 4503 | space_info->force_alloc = CHUNK_ALLOC_NO_FORCE; |
a81cb9a2 | 4504 | out: |
6d74119f | 4505 | space_info->chunk_alloc = 0; |
9ed74f2d | 4506 | spin_unlock(&space_info->lock); |
a25c75d5 | 4507 | mutex_unlock(&fs_info->chunk_mutex); |
00d80e34 FM |
4508 | /* |
4509 | * When we allocate a new chunk we reserve space in the chunk block | |
4510 | * reserve to make sure we can COW nodes/leafs in the chunk tree or | |
4511 | * add new nodes/leafs to it if we end up needing to do it when | |
4512 | * inserting the chunk item and updating device items as part of the | |
4513 | * second phase of chunk allocation, performed by | |
4514 | * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a | |
4515 | * large number of new block groups to create in our transaction | |
4516 | * handle's new_bgs list to avoid exhausting the chunk block reserve | |
4517 | * in extreme cases - like having a single transaction create many new | |
4518 | * block groups when starting to write out the free space caches of all | |
4519 | * the block groups that were made dirty during the lifetime of the | |
4520 | * transaction. | |
4521 | */ | |
5ce55557 | 4522 | if (trans->chunk_bytes_reserved >= (u64)SZ_2M) |
6c686b35 | 4523 | btrfs_create_pending_block_groups(trans); |
5ce55557 | 4524 | |
0f9dd46c | 4525 | return ret; |
6324fbf3 | 4526 | } |
9ed74f2d | 4527 | |
c1c4919b | 4528 | static int can_overcommit(struct btrfs_fs_info *fs_info, |
a80c8dcf | 4529 | struct btrfs_space_info *space_info, u64 bytes, |
c1c4919b JM |
4530 | enum btrfs_reserve_flush_enum flush, |
4531 | bool system_chunk) | |
a80c8dcf | 4532 | { |
0b246afa | 4533 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; |
957780eb | 4534 | u64 profile; |
3c76cd84 | 4535 | u64 space_size; |
a80c8dcf JB |
4536 | u64 avail; |
4537 | u64 used; | |
46df06b8 | 4538 | int factor; |
a80c8dcf | 4539 | |
957780eb JB |
4540 | /* Don't overcommit when in mixed mode. */ |
4541 | if (space_info->flags & BTRFS_BLOCK_GROUP_DATA) | |
4542 | return 0; | |
4543 | ||
c1c4919b JM |
4544 | if (system_chunk) |
4545 | profile = btrfs_system_alloc_profile(fs_info); | |
4546 | else | |
4547 | profile = btrfs_metadata_alloc_profile(fs_info); | |
4548 | ||
4136135b | 4549 | used = btrfs_space_info_used(space_info, false); |
96f1bb57 | 4550 | |
96f1bb57 JB |
4551 | /* |
4552 | * We only want to allow over committing if we have lots of actual space | |
4553 | * free, but if we don't have enough space to handle the global reserve | |
4554 | * space then we could end up having a real enospc problem when trying | |
4555 | * to allocate a chunk or some other such important allocation. | |
4556 | */ | |
3c76cd84 MX |
4557 | spin_lock(&global_rsv->lock); |
4558 | space_size = calc_global_rsv_need_space(global_rsv); | |
4559 | spin_unlock(&global_rsv->lock); | |
4560 | if (used + space_size >= space_info->total_bytes) | |
96f1bb57 JB |
4561 | return 0; |
4562 | ||
4563 | used += space_info->bytes_may_use; | |
a80c8dcf | 4564 | |
a5ed45f8 | 4565 | avail = atomic64_read(&fs_info->free_chunk_space); |
a80c8dcf JB |
4566 | |
4567 | /* | |
4568 | * If we have dup, raid1 or raid10 then only half of the free | |
52042d8e | 4569 | * space is actually usable. For raid56, the space info used |
53b381b3 DW |
4570 | * doesn't include the parity drive, so we don't have to |
4571 | * change the math | |
a80c8dcf | 4572 | */ |
46df06b8 DS |
4573 | factor = btrfs_bg_type_to_factor(profile); |
4574 | avail = div_u64(avail, factor); | |
a80c8dcf JB |
4575 | |
4576 | /* | |
561c294d MX |
4577 | * If we aren't flushing all things, let us overcommit up to |
4578 | * 1/2th of the space. If we can flush, don't let us overcommit | |
4579 | * too much, let it overcommit up to 1/8 of the space. | |
a80c8dcf | 4580 | */ |
08e007d2 | 4581 | if (flush == BTRFS_RESERVE_FLUSH_ALL) |
14575aef | 4582 | avail >>= 3; |
a80c8dcf | 4583 | else |
14575aef | 4584 | avail >>= 1; |
a80c8dcf | 4585 | |
14575aef | 4586 | if (used + bytes < space_info->total_bytes + avail) |
a80c8dcf JB |
4587 | return 1; |
4588 | return 0; | |
4589 | } | |
4590 | ||
2ff7e61e | 4591 | static void btrfs_writeback_inodes_sb_nr(struct btrfs_fs_info *fs_info, |
6c255e67 | 4592 | unsigned long nr_pages, int nr_items) |
da633a42 | 4593 | { |
0b246afa | 4594 | struct super_block *sb = fs_info->sb; |
da633a42 | 4595 | |
925a6efb JB |
4596 | if (down_read_trylock(&sb->s_umount)) { |
4597 | writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE); | |
4598 | up_read(&sb->s_umount); | |
4599 | } else { | |
da633a42 MX |
4600 | /* |
4601 | * We needn't worry the filesystem going from r/w to r/o though | |
4602 | * we don't acquire ->s_umount mutex, because the filesystem | |
4603 | * should guarantee the delalloc inodes list be empty after | |
4604 | * the filesystem is readonly(all dirty pages are written to | |
4605 | * the disk). | |
4606 | */ | |
82b3e53b | 4607 | btrfs_start_delalloc_roots(fs_info, nr_items); |
98ad69cf | 4608 | if (!current->journal_info) |
0b246afa | 4609 | btrfs_wait_ordered_roots(fs_info, nr_items, 0, (u64)-1); |
da633a42 MX |
4610 | } |
4611 | } | |
4612 | ||
6374e57a | 4613 | static inline u64 calc_reclaim_items_nr(struct btrfs_fs_info *fs_info, |
2ff7e61e | 4614 | u64 to_reclaim) |
18cd8ea6 MX |
4615 | { |
4616 | u64 bytes; | |
6374e57a | 4617 | u64 nr; |
18cd8ea6 | 4618 | |
2ff7e61e | 4619 | bytes = btrfs_calc_trans_metadata_size(fs_info, 1); |
6374e57a | 4620 | nr = div64_u64(to_reclaim, bytes); |
18cd8ea6 MX |
4621 | if (!nr) |
4622 | nr = 1; | |
4623 | return nr; | |
4624 | } | |
4625 | ||
ee22184b | 4626 | #define EXTENT_SIZE_PER_ITEM SZ_256K |
c61a16a7 | 4627 | |
9ed74f2d | 4628 | /* |
5da9d01b | 4629 | * shrink metadata reservation for delalloc |
9ed74f2d | 4630 | */ |
c1c4919b JM |
4631 | static void shrink_delalloc(struct btrfs_fs_info *fs_info, u64 to_reclaim, |
4632 | u64 orig, bool wait_ordered) | |
5da9d01b | 4633 | { |
0019f10d | 4634 | struct btrfs_space_info *space_info; |
663350ac | 4635 | struct btrfs_trans_handle *trans; |
f4c738c2 | 4636 | u64 delalloc_bytes; |
4297ff84 | 4637 | u64 dio_bytes; |
420829d8 | 4638 | u64 async_pages; |
6374e57a | 4639 | u64 items; |
b1953bce | 4640 | long time_left; |
d3ee29e3 MX |
4641 | unsigned long nr_pages; |
4642 | int loops; | |
5da9d01b | 4643 | |
c61a16a7 | 4644 | /* Calc the number of the pages we need flush for space reservation */ |
2ff7e61e | 4645 | items = calc_reclaim_items_nr(fs_info, to_reclaim); |
6374e57a | 4646 | to_reclaim = items * EXTENT_SIZE_PER_ITEM; |
c61a16a7 | 4647 | |
663350ac | 4648 | trans = (struct btrfs_trans_handle *)current->journal_info; |
69fe2d75 | 4649 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
bf9022e0 | 4650 | |
963d678b | 4651 | delalloc_bytes = percpu_counter_sum_positive( |
0b246afa | 4652 | &fs_info->delalloc_bytes); |
4297ff84 JB |
4653 | dio_bytes = percpu_counter_sum_positive(&fs_info->dio_bytes); |
4654 | if (delalloc_bytes == 0 && dio_bytes == 0) { | |
fdb5effd | 4655 | if (trans) |
f4c738c2 | 4656 | return; |
38c135af | 4657 | if (wait_ordered) |
0b246afa | 4658 | btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1); |
f4c738c2 | 4659 | return; |
fdb5effd JB |
4660 | } |
4661 | ||
4297ff84 JB |
4662 | /* |
4663 | * If we are doing more ordered than delalloc we need to just wait on | |
4664 | * ordered extents, otherwise we'll waste time trying to flush delalloc | |
4665 | * that likely won't give us the space back we need. | |
4666 | */ | |
4667 | if (dio_bytes > delalloc_bytes) | |
4668 | wait_ordered = true; | |
4669 | ||
d3ee29e3 | 4670 | loops = 0; |
4297ff84 | 4671 | while ((delalloc_bytes || dio_bytes) && loops < 3) { |
420829d8 NB |
4672 | nr_pages = min(delalloc_bytes, to_reclaim) >> PAGE_SHIFT; |
4673 | ||
4674 | /* | |
4675 | * Triggers inode writeback for up to nr_pages. This will invoke | |
4676 | * ->writepages callback and trigger delalloc filling | |
4677 | * (btrfs_run_delalloc_range()). | |
4678 | */ | |
2ff7e61e | 4679 | btrfs_writeback_inodes_sb_nr(fs_info, nr_pages, items); |
420829d8 | 4680 | |
dea31f52 | 4681 | /* |
420829d8 NB |
4682 | * We need to wait for the compressed pages to start before |
4683 | * we continue. | |
dea31f52 | 4684 | */ |
420829d8 NB |
4685 | async_pages = atomic_read(&fs_info->async_delalloc_pages); |
4686 | if (!async_pages) | |
9f3a074d MX |
4687 | goto skip_async; |
4688 | ||
420829d8 NB |
4689 | /* |
4690 | * Calculate how many compressed pages we want to be written | |
4691 | * before we continue. I.e if there are more async pages than we | |
4692 | * require wait_event will wait until nr_pages are written. | |
4693 | */ | |
4694 | if (async_pages <= nr_pages) | |
4695 | async_pages = 0; | |
9f3a074d | 4696 | else |
420829d8 | 4697 | async_pages -= nr_pages; |
dea31f52 | 4698 | |
0b246afa JM |
4699 | wait_event(fs_info->async_submit_wait, |
4700 | atomic_read(&fs_info->async_delalloc_pages) <= | |
420829d8 | 4701 | (int)async_pages); |
9f3a074d | 4702 | skip_async: |
0019f10d | 4703 | spin_lock(&space_info->lock); |
957780eb JB |
4704 | if (list_empty(&space_info->tickets) && |
4705 | list_empty(&space_info->priority_tickets)) { | |
4706 | spin_unlock(&space_info->lock); | |
4707 | break; | |
4708 | } | |
0019f10d | 4709 | spin_unlock(&space_info->lock); |
5da9d01b | 4710 | |
36e39c40 | 4711 | loops++; |
f104d044 | 4712 | if (wait_ordered && !trans) { |
0b246afa | 4713 | btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1); |
f104d044 | 4714 | } else { |
f4c738c2 | 4715 | time_left = schedule_timeout_killable(1); |
f104d044 JB |
4716 | if (time_left) |
4717 | break; | |
4718 | } | |
963d678b | 4719 | delalloc_bytes = percpu_counter_sum_positive( |
0b246afa | 4720 | &fs_info->delalloc_bytes); |
4297ff84 | 4721 | dio_bytes = percpu_counter_sum_positive(&fs_info->dio_bytes); |
5da9d01b | 4722 | } |
5da9d01b YZ |
4723 | } |
4724 | ||
996478ca | 4725 | struct reserve_ticket { |
f91587e4 | 4726 | u64 orig_bytes; |
996478ca JB |
4727 | u64 bytes; |
4728 | int error; | |
4729 | struct list_head list; | |
4730 | wait_queue_head_t wait; | |
4731 | }; | |
4732 | ||
663350ac JB |
4733 | /** |
4734 | * maybe_commit_transaction - possibly commit the transaction if its ok to | |
4735 | * @root - the root we're allocating for | |
4736 | * @bytes - the number of bytes we want to reserve | |
4737 | * @force - force the commit | |
8bb8ab2e | 4738 | * |
663350ac JB |
4739 | * This will check to make sure that committing the transaction will actually |
4740 | * get us somewhere and then commit the transaction if it does. Otherwise it | |
4741 | * will return -ENOSPC. | |
8bb8ab2e | 4742 | */ |
0c9ab349 | 4743 | static int may_commit_transaction(struct btrfs_fs_info *fs_info, |
996478ca | 4744 | struct btrfs_space_info *space_info) |
663350ac | 4745 | { |
996478ca | 4746 | struct reserve_ticket *ticket = NULL; |
0b246afa | 4747 | struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_block_rsv; |
4c8edbc7 | 4748 | struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv; |
663350ac | 4749 | struct btrfs_trans_handle *trans; |
4c8edbc7 JB |
4750 | u64 bytes_needed; |
4751 | u64 reclaim_bytes = 0; | |
663350ac JB |
4752 | |
4753 | trans = (struct btrfs_trans_handle *)current->journal_info; | |
4754 | if (trans) | |
4755 | return -EAGAIN; | |
4756 | ||
996478ca JB |
4757 | spin_lock(&space_info->lock); |
4758 | if (!list_empty(&space_info->priority_tickets)) | |
4759 | ticket = list_first_entry(&space_info->priority_tickets, | |
4760 | struct reserve_ticket, list); | |
4761 | else if (!list_empty(&space_info->tickets)) | |
4762 | ticket = list_first_entry(&space_info->tickets, | |
4763 | struct reserve_ticket, list); | |
4c8edbc7 | 4764 | bytes_needed = (ticket) ? ticket->bytes : 0; |
996478ca JB |
4765 | spin_unlock(&space_info->lock); |
4766 | ||
4c8edbc7 | 4767 | if (!bytes_needed) |
996478ca | 4768 | return 0; |
663350ac | 4769 | |
d89dbefb JB |
4770 | trans = btrfs_join_transaction(fs_info->extent_root); |
4771 | if (IS_ERR(trans)) | |
4772 | return PTR_ERR(trans); | |
4773 | ||
4774 | /* | |
4775 | * See if there is enough pinned space to make this reservation, or if | |
4776 | * we have block groups that are going to be freed, allowing us to | |
4777 | * possibly do a chunk allocation the next loop through. | |
4778 | */ | |
4779 | if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags) || | |
4780 | __percpu_counter_compare(&space_info->total_bytes_pinned, | |
4781 | bytes_needed, | |
4782 | BTRFS_TOTAL_BYTES_PINNED_BATCH) >= 0) | |
663350ac | 4783 | goto commit; |
663350ac JB |
4784 | |
4785 | /* | |
4786 | * See if there is some space in the delayed insertion reservation for | |
4787 | * this reservation. | |
4788 | */ | |
4789 | if (space_info != delayed_rsv->space_info) | |
d89dbefb | 4790 | goto enospc; |
663350ac JB |
4791 | |
4792 | spin_lock(&delayed_rsv->lock); | |
4c8edbc7 | 4793 | reclaim_bytes += delayed_rsv->reserved; |
057aac3e NB |
4794 | spin_unlock(&delayed_rsv->lock); |
4795 | ||
4c8edbc7 JB |
4796 | spin_lock(&delayed_refs_rsv->lock); |
4797 | reclaim_bytes += delayed_refs_rsv->reserved; | |
4798 | spin_unlock(&delayed_refs_rsv->lock); | |
4799 | if (reclaim_bytes >= bytes_needed) | |
4800 | goto commit; | |
4801 | bytes_needed -= reclaim_bytes; | |
4802 | ||
dec59fa3 | 4803 | if (__percpu_counter_compare(&space_info->total_bytes_pinned, |
4c8edbc7 | 4804 | bytes_needed, |
d89dbefb JB |
4805 | BTRFS_TOTAL_BYTES_PINNED_BATCH) < 0) |
4806 | goto enospc; | |
663350ac JB |
4807 | |
4808 | commit: | |
3a45bb20 | 4809 | return btrfs_commit_transaction(trans); |
d89dbefb JB |
4810 | enospc: |
4811 | btrfs_end_transaction(trans); | |
4812 | return -ENOSPC; | |
663350ac JB |
4813 | } |
4814 | ||
e38ae7a0 NB |
4815 | /* |
4816 | * Try to flush some data based on policy set by @state. This is only advisory | |
4817 | * and may fail for various reasons. The caller is supposed to examine the | |
4818 | * state of @space_info to detect the outcome. | |
4819 | */ | |
4820 | static void flush_space(struct btrfs_fs_info *fs_info, | |
96c3f433 | 4821 | struct btrfs_space_info *space_info, u64 num_bytes, |
7bdd6277 | 4822 | int state) |
96c3f433 | 4823 | { |
a9b3311e | 4824 | struct btrfs_root *root = fs_info->extent_root; |
96c3f433 JB |
4825 | struct btrfs_trans_handle *trans; |
4826 | int nr; | |
f4c738c2 | 4827 | int ret = 0; |
96c3f433 JB |
4828 | |
4829 | switch (state) { | |
96c3f433 JB |
4830 | case FLUSH_DELAYED_ITEMS_NR: |
4831 | case FLUSH_DELAYED_ITEMS: | |
18cd8ea6 | 4832 | if (state == FLUSH_DELAYED_ITEMS_NR) |
2ff7e61e | 4833 | nr = calc_reclaim_items_nr(fs_info, num_bytes) * 2; |
18cd8ea6 | 4834 | else |
96c3f433 | 4835 | nr = -1; |
18cd8ea6 | 4836 | |
96c3f433 JB |
4837 | trans = btrfs_join_transaction(root); |
4838 | if (IS_ERR(trans)) { | |
4839 | ret = PTR_ERR(trans); | |
4840 | break; | |
4841 | } | |
e5c304e6 | 4842 | ret = btrfs_run_delayed_items_nr(trans, nr); |
3a45bb20 | 4843 | btrfs_end_transaction(trans); |
96c3f433 | 4844 | break; |
67b0fd63 JB |
4845 | case FLUSH_DELALLOC: |
4846 | case FLUSH_DELALLOC_WAIT: | |
7bdd6277 | 4847 | shrink_delalloc(fs_info, num_bytes * 2, num_bytes, |
67b0fd63 JB |
4848 | state == FLUSH_DELALLOC_WAIT); |
4849 | break; | |
413df725 JB |
4850 | case FLUSH_DELAYED_REFS_NR: |
4851 | case FLUSH_DELAYED_REFS: | |
4852 | trans = btrfs_join_transaction(root); | |
4853 | if (IS_ERR(trans)) { | |
4854 | ret = PTR_ERR(trans); | |
4855 | break; | |
4856 | } | |
4857 | if (state == FLUSH_DELAYED_REFS_NR) | |
4858 | nr = calc_reclaim_items_nr(fs_info, num_bytes); | |
4859 | else | |
4860 | nr = 0; | |
4861 | btrfs_run_delayed_refs(trans, nr); | |
4862 | btrfs_end_transaction(trans); | |
4863 | break; | |
ea658bad | 4864 | case ALLOC_CHUNK: |
450114fc | 4865 | case ALLOC_CHUNK_FORCE: |
ea658bad JB |
4866 | trans = btrfs_join_transaction(root); |
4867 | if (IS_ERR(trans)) { | |
4868 | ret = PTR_ERR(trans); | |
4869 | break; | |
4870 | } | |
01458828 | 4871 | ret = do_chunk_alloc(trans, |
1b86826d | 4872 | btrfs_metadata_alloc_profile(fs_info), |
450114fc JB |
4873 | (state == ALLOC_CHUNK) ? |
4874 | CHUNK_ALLOC_NO_FORCE : CHUNK_ALLOC_FORCE); | |
3a45bb20 | 4875 | btrfs_end_transaction(trans); |
eecba891 | 4876 | if (ret > 0 || ret == -ENOSPC) |
ea658bad JB |
4877 | ret = 0; |
4878 | break; | |
96c3f433 | 4879 | case COMMIT_TRANS: |
3ec9a4c8 JB |
4880 | /* |
4881 | * If we have pending delayed iputs then we could free up a | |
4882 | * bunch of pinned space, so make sure we run the iputs before | |
4883 | * we do our pinned bytes check below. | |
4884 | */ | |
3ec9a4c8 | 4885 | btrfs_run_delayed_iputs(fs_info); |
034f784d | 4886 | btrfs_wait_on_delayed_iputs(fs_info); |
3ec9a4c8 | 4887 | |
996478ca | 4888 | ret = may_commit_transaction(fs_info, space_info); |
96c3f433 JB |
4889 | break; |
4890 | default: | |
4891 | ret = -ENOSPC; | |
4892 | break; | |
4893 | } | |
4894 | ||
7bdd6277 NB |
4895 | trace_btrfs_flush_space(fs_info, space_info->flags, num_bytes, state, |
4896 | ret); | |
e38ae7a0 | 4897 | return; |
96c3f433 | 4898 | } |
21c7e756 MX |
4899 | |
4900 | static inline u64 | |
c1c4919b JM |
4901 | btrfs_calc_reclaim_metadata_size(struct btrfs_fs_info *fs_info, |
4902 | struct btrfs_space_info *space_info, | |
4903 | bool system_chunk) | |
21c7e756 | 4904 | { |
957780eb | 4905 | struct reserve_ticket *ticket; |
21c7e756 MX |
4906 | u64 used; |
4907 | u64 expected; | |
957780eb | 4908 | u64 to_reclaim = 0; |
21c7e756 | 4909 | |
957780eb JB |
4910 | list_for_each_entry(ticket, &space_info->tickets, list) |
4911 | to_reclaim += ticket->bytes; | |
4912 | list_for_each_entry(ticket, &space_info->priority_tickets, list) | |
4913 | to_reclaim += ticket->bytes; | |
4914 | if (to_reclaim) | |
4915 | return to_reclaim; | |
21c7e756 | 4916 | |
e0af2484 | 4917 | to_reclaim = min_t(u64, num_online_cpus() * SZ_1M, SZ_16M); |
c1c4919b JM |
4918 | if (can_overcommit(fs_info, space_info, to_reclaim, |
4919 | BTRFS_RESERVE_FLUSH_ALL, system_chunk)) | |
e0af2484 WX |
4920 | return 0; |
4921 | ||
0eee8a49 NB |
4922 | used = btrfs_space_info_used(space_info, true); |
4923 | ||
c1c4919b JM |
4924 | if (can_overcommit(fs_info, space_info, SZ_1M, |
4925 | BTRFS_RESERVE_FLUSH_ALL, system_chunk)) | |
21c7e756 MX |
4926 | expected = div_factor_fine(space_info->total_bytes, 95); |
4927 | else | |
4928 | expected = div_factor_fine(space_info->total_bytes, 90); | |
4929 | ||
4930 | if (used > expected) | |
4931 | to_reclaim = used - expected; | |
4932 | else | |
4933 | to_reclaim = 0; | |
4934 | to_reclaim = min(to_reclaim, space_info->bytes_may_use + | |
4935 | space_info->bytes_reserved); | |
21c7e756 MX |
4936 | return to_reclaim; |
4937 | } | |
4938 | ||
c1c4919b JM |
4939 | static inline int need_do_async_reclaim(struct btrfs_fs_info *fs_info, |
4940 | struct btrfs_space_info *space_info, | |
4941 | u64 used, bool system_chunk) | |
21c7e756 | 4942 | { |
365c5313 JB |
4943 | u64 thresh = div_factor_fine(space_info->total_bytes, 98); |
4944 | ||
4945 | /* If we're just plain full then async reclaim just slows us down. */ | |
baee8790 | 4946 | if ((space_info->bytes_used + space_info->bytes_reserved) >= thresh) |
365c5313 JB |
4947 | return 0; |
4948 | ||
c1c4919b JM |
4949 | if (!btrfs_calc_reclaim_metadata_size(fs_info, space_info, |
4950 | system_chunk)) | |
d38b349c JB |
4951 | return 0; |
4952 | ||
0b246afa JM |
4953 | return (used >= thresh && !btrfs_fs_closing(fs_info) && |
4954 | !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state)); | |
21c7e756 MX |
4955 | } |
4956 | ||
f91587e4 | 4957 | static bool wake_all_tickets(struct list_head *head) |
21c7e756 | 4958 | { |
957780eb | 4959 | struct reserve_ticket *ticket; |
25ce459c | 4960 | |
957780eb JB |
4961 | while (!list_empty(head)) { |
4962 | ticket = list_first_entry(head, struct reserve_ticket, list); | |
4963 | list_del_init(&ticket->list); | |
4964 | ticket->error = -ENOSPC; | |
4965 | wake_up(&ticket->wait); | |
f91587e4 JB |
4966 | if (ticket->bytes != ticket->orig_bytes) |
4967 | return true; | |
21c7e756 | 4968 | } |
f91587e4 | 4969 | return false; |
21c7e756 MX |
4970 | } |
4971 | ||
957780eb JB |
4972 | /* |
4973 | * This is for normal flushers, we can wait all goddamned day if we want to. We | |
4974 | * will loop and continuously try to flush as long as we are making progress. | |
4975 | * We count progress as clearing off tickets each time we have to loop. | |
4976 | */ | |
21c7e756 MX |
4977 | static void btrfs_async_reclaim_metadata_space(struct work_struct *work) |
4978 | { | |
4979 | struct btrfs_fs_info *fs_info; | |
4980 | struct btrfs_space_info *space_info; | |
4981 | u64 to_reclaim; | |
4982 | int flush_state; | |
957780eb | 4983 | int commit_cycles = 0; |
ce129655 | 4984 | u64 last_tickets_id; |
21c7e756 MX |
4985 | |
4986 | fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work); | |
4987 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); | |
4988 | ||
957780eb | 4989 | spin_lock(&space_info->lock); |
c1c4919b JM |
4990 | to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info, |
4991 | false); | |
957780eb JB |
4992 | if (!to_reclaim) { |
4993 | space_info->flush = 0; | |
4994 | spin_unlock(&space_info->lock); | |
21c7e756 | 4995 | return; |
957780eb | 4996 | } |
ce129655 | 4997 | last_tickets_id = space_info->tickets_id; |
957780eb | 4998 | spin_unlock(&space_info->lock); |
21c7e756 MX |
4999 | |
5000 | flush_state = FLUSH_DELAYED_ITEMS_NR; | |
957780eb | 5001 | do { |
e38ae7a0 | 5002 | flush_space(fs_info, space_info, to_reclaim, flush_state); |
957780eb JB |
5003 | spin_lock(&space_info->lock); |
5004 | if (list_empty(&space_info->tickets)) { | |
5005 | space_info->flush = 0; | |
5006 | spin_unlock(&space_info->lock); | |
5007 | return; | |
5008 | } | |
c1c4919b JM |
5009 | to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, |
5010 | space_info, | |
5011 | false); | |
ce129655 | 5012 | if (last_tickets_id == space_info->tickets_id) { |
957780eb JB |
5013 | flush_state++; |
5014 | } else { | |
ce129655 | 5015 | last_tickets_id = space_info->tickets_id; |
957780eb JB |
5016 | flush_state = FLUSH_DELAYED_ITEMS_NR; |
5017 | if (commit_cycles) | |
5018 | commit_cycles--; | |
5019 | } | |
5020 | ||
450114fc JB |
5021 | /* |
5022 | * We don't want to force a chunk allocation until we've tried | |
5023 | * pretty hard to reclaim space. Think of the case where we | |
5024 | * freed up a bunch of space and so have a lot of pinned space | |
5025 | * to reclaim. We would rather use that than possibly create a | |
5026 | * underutilized metadata chunk. So if this is our first run | |
5027 | * through the flushing state machine skip ALLOC_CHUNK_FORCE and | |
5028 | * commit the transaction. If nothing has changed the next go | |
5029 | * around then we can force a chunk allocation. | |
5030 | */ | |
5031 | if (flush_state == ALLOC_CHUNK_FORCE && !commit_cycles) | |
5032 | flush_state++; | |
5033 | ||
957780eb JB |
5034 | if (flush_state > COMMIT_TRANS) { |
5035 | commit_cycles++; | |
5036 | if (commit_cycles > 2) { | |
f91587e4 JB |
5037 | if (wake_all_tickets(&space_info->tickets)) { |
5038 | flush_state = FLUSH_DELAYED_ITEMS_NR; | |
5039 | commit_cycles--; | |
5040 | } else { | |
5041 | space_info->flush = 0; | |
5042 | } | |
957780eb JB |
5043 | } else { |
5044 | flush_state = FLUSH_DELAYED_ITEMS_NR; | |
5045 | } | |
5046 | } | |
5047 | spin_unlock(&space_info->lock); | |
5048 | } while (flush_state <= COMMIT_TRANS); | |
5049 | } | |
5050 | ||
5051 | void btrfs_init_async_reclaim_work(struct work_struct *work) | |
5052 | { | |
5053 | INIT_WORK(work, btrfs_async_reclaim_metadata_space); | |
5054 | } | |
5055 | ||
8a1bbe1d JB |
5056 | static const enum btrfs_flush_state priority_flush_states[] = { |
5057 | FLUSH_DELAYED_ITEMS_NR, | |
5058 | FLUSH_DELAYED_ITEMS, | |
5059 | ALLOC_CHUNK, | |
5060 | }; | |
5061 | ||
957780eb JB |
5062 | static void priority_reclaim_metadata_space(struct btrfs_fs_info *fs_info, |
5063 | struct btrfs_space_info *space_info, | |
5064 | struct reserve_ticket *ticket) | |
5065 | { | |
5066 | u64 to_reclaim; | |
8a1bbe1d | 5067 | int flush_state; |
957780eb JB |
5068 | |
5069 | spin_lock(&space_info->lock); | |
c1c4919b JM |
5070 | to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info, |
5071 | false); | |
957780eb JB |
5072 | if (!to_reclaim) { |
5073 | spin_unlock(&space_info->lock); | |
5074 | return; | |
5075 | } | |
5076 | spin_unlock(&space_info->lock); | |
5077 | ||
8a1bbe1d | 5078 | flush_state = 0; |
21c7e756 | 5079 | do { |
8a1bbe1d JB |
5080 | flush_space(fs_info, space_info, to_reclaim, |
5081 | priority_flush_states[flush_state]); | |
21c7e756 | 5082 | flush_state++; |
957780eb JB |
5083 | spin_lock(&space_info->lock); |
5084 | if (ticket->bytes == 0) { | |
5085 | spin_unlock(&space_info->lock); | |
21c7e756 | 5086 | return; |
957780eb JB |
5087 | } |
5088 | spin_unlock(&space_info->lock); | |
8a1bbe1d | 5089 | } while (flush_state < ARRAY_SIZE(priority_flush_states)); |
21c7e756 MX |
5090 | } |
5091 | ||
957780eb JB |
5092 | static int wait_reserve_ticket(struct btrfs_fs_info *fs_info, |
5093 | struct btrfs_space_info *space_info, | |
f91587e4 | 5094 | struct reserve_ticket *ticket) |
957780eb | 5095 | |
21c7e756 | 5096 | { |
957780eb | 5097 | DEFINE_WAIT(wait); |
f91587e4 | 5098 | u64 reclaim_bytes = 0; |
957780eb JB |
5099 | int ret = 0; |
5100 | ||
5101 | spin_lock(&space_info->lock); | |
5102 | while (ticket->bytes > 0 && ticket->error == 0) { | |
5103 | ret = prepare_to_wait_event(&ticket->wait, &wait, TASK_KILLABLE); | |
5104 | if (ret) { | |
5105 | ret = -EINTR; | |
5106 | break; | |
5107 | } | |
5108 | spin_unlock(&space_info->lock); | |
5109 | ||
5110 | schedule(); | |
5111 | ||
5112 | finish_wait(&ticket->wait, &wait); | |
5113 | spin_lock(&space_info->lock); | |
5114 | } | |
5115 | if (!ret) | |
5116 | ret = ticket->error; | |
5117 | if (!list_empty(&ticket->list)) | |
5118 | list_del_init(&ticket->list); | |
f91587e4 JB |
5119 | if (ticket->bytes && ticket->bytes < ticket->orig_bytes) |
5120 | reclaim_bytes = ticket->orig_bytes - ticket->bytes; | |
957780eb JB |
5121 | spin_unlock(&space_info->lock); |
5122 | ||
f91587e4 JB |
5123 | if (reclaim_bytes) |
5124 | space_info_add_old_bytes(fs_info, space_info, reclaim_bytes); | |
957780eb | 5125 | return ret; |
21c7e756 MX |
5126 | } |
5127 | ||
4a92b1b8 JB |
5128 | /** |
5129 | * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space | |
5130 | * @root - the root we're allocating for | |
957780eb | 5131 | * @space_info - the space info we want to allocate from |
4a92b1b8 | 5132 | * @orig_bytes - the number of bytes we want |
48fc7f7e | 5133 | * @flush - whether or not we can flush to make our reservation |
8bb8ab2e | 5134 | * |
01327610 | 5135 | * This will reserve orig_bytes number of bytes from the space info associated |
4a92b1b8 JB |
5136 | * with the block_rsv. If there is not enough space it will make an attempt to |
5137 | * flush out space to make room. It will do this by flushing delalloc if | |
5138 | * possible or committing the transaction. If flush is 0 then no attempts to | |
5139 | * regain reservations will be made and this will fail if there is not enough | |
5140 | * space already. | |
8bb8ab2e | 5141 | */ |
c1c4919b | 5142 | static int __reserve_metadata_bytes(struct btrfs_fs_info *fs_info, |
957780eb JB |
5143 | struct btrfs_space_info *space_info, |
5144 | u64 orig_bytes, | |
c1c4919b JM |
5145 | enum btrfs_reserve_flush_enum flush, |
5146 | bool system_chunk) | |
9ed74f2d | 5147 | { |
957780eb | 5148 | struct reserve_ticket ticket; |
2bf64758 | 5149 | u64 used; |
f91587e4 | 5150 | u64 reclaim_bytes = 0; |
8bb8ab2e | 5151 | int ret = 0; |
9ed74f2d | 5152 | |
957780eb | 5153 | ASSERT(orig_bytes); |
8ca17f0f | 5154 | ASSERT(!current->journal_info || flush != BTRFS_RESERVE_FLUSH_ALL); |
fdb5effd | 5155 | |
8bb8ab2e | 5156 | spin_lock(&space_info->lock); |
fdb5effd | 5157 | ret = -ENOSPC; |
4136135b | 5158 | used = btrfs_space_info_used(space_info, true); |
9ed74f2d | 5159 | |
8bb8ab2e | 5160 | /* |
957780eb JB |
5161 | * If we have enough space then hooray, make our reservation and carry |
5162 | * on. If not see if we can overcommit, and if we can, hooray carry on. | |
5163 | * If not things get more complicated. | |
8bb8ab2e | 5164 | */ |
957780eb | 5165 | if (used + orig_bytes <= space_info->total_bytes) { |
9f9b8e8d | 5166 | update_bytes_may_use(space_info, orig_bytes); |
0b246afa JM |
5167 | trace_btrfs_space_reservation(fs_info, "space_info", |
5168 | space_info->flags, orig_bytes, 1); | |
957780eb | 5169 | ret = 0; |
c1c4919b JM |
5170 | } else if (can_overcommit(fs_info, space_info, orig_bytes, flush, |
5171 | system_chunk)) { | |
9f9b8e8d | 5172 | update_bytes_may_use(space_info, orig_bytes); |
0b246afa JM |
5173 | trace_btrfs_space_reservation(fs_info, "space_info", |
5174 | space_info->flags, orig_bytes, 1); | |
44734ed1 | 5175 | ret = 0; |
2bf64758 JB |
5176 | } |
5177 | ||
8bb8ab2e | 5178 | /* |
957780eb JB |
5179 | * If we couldn't make a reservation then setup our reservation ticket |
5180 | * and kick the async worker if it's not already running. | |
08e007d2 | 5181 | * |
957780eb JB |
5182 | * If we are a priority flusher then we just need to add our ticket to |
5183 | * the list and we will do our own flushing further down. | |
8bb8ab2e | 5184 | */ |
72bcd99d | 5185 | if (ret && flush != BTRFS_RESERVE_NO_FLUSH) { |
f91587e4 | 5186 | ticket.orig_bytes = orig_bytes; |
957780eb JB |
5187 | ticket.bytes = orig_bytes; |
5188 | ticket.error = 0; | |
5189 | init_waitqueue_head(&ticket.wait); | |
5190 | if (flush == BTRFS_RESERVE_FLUSH_ALL) { | |
5191 | list_add_tail(&ticket.list, &space_info->tickets); | |
5192 | if (!space_info->flush) { | |
5193 | space_info->flush = 1; | |
0b246afa | 5194 | trace_btrfs_trigger_flush(fs_info, |
f376df2b JB |
5195 | space_info->flags, |
5196 | orig_bytes, flush, | |
5197 | "enospc"); | |
957780eb | 5198 | queue_work(system_unbound_wq, |
c1c4919b | 5199 | &fs_info->async_reclaim_work); |
957780eb JB |
5200 | } |
5201 | } else { | |
5202 | list_add_tail(&ticket.list, | |
5203 | &space_info->priority_tickets); | |
5204 | } | |
21c7e756 MX |
5205 | } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) { |
5206 | used += orig_bytes; | |
f6acfd50 JB |
5207 | /* |
5208 | * We will do the space reservation dance during log replay, | |
5209 | * which means we won't have fs_info->fs_root set, so don't do | |
5210 | * the async reclaim as we will panic. | |
5211 | */ | |
0b246afa | 5212 | if (!test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags) && |
c1c4919b JM |
5213 | need_do_async_reclaim(fs_info, space_info, |
5214 | used, system_chunk) && | |
0b246afa JM |
5215 | !work_busy(&fs_info->async_reclaim_work)) { |
5216 | trace_btrfs_trigger_flush(fs_info, space_info->flags, | |
5217 | orig_bytes, flush, "preempt"); | |
21c7e756 | 5218 | queue_work(system_unbound_wq, |
0b246afa | 5219 | &fs_info->async_reclaim_work); |
f376df2b | 5220 | } |
8bb8ab2e | 5221 | } |
f0486c68 | 5222 | spin_unlock(&space_info->lock); |
08e007d2 | 5223 | if (!ret || flush == BTRFS_RESERVE_NO_FLUSH) |
957780eb | 5224 | return ret; |
f0486c68 | 5225 | |
957780eb | 5226 | if (flush == BTRFS_RESERVE_FLUSH_ALL) |
f91587e4 | 5227 | return wait_reserve_ticket(fs_info, space_info, &ticket); |
08e007d2 | 5228 | |
957780eb | 5229 | ret = 0; |
0b246afa | 5230 | priority_reclaim_metadata_space(fs_info, space_info, &ticket); |
957780eb JB |
5231 | spin_lock(&space_info->lock); |
5232 | if (ticket.bytes) { | |
f91587e4 JB |
5233 | if (ticket.bytes < orig_bytes) |
5234 | reclaim_bytes = orig_bytes - ticket.bytes; | |
957780eb JB |
5235 | list_del_init(&ticket.list); |
5236 | ret = -ENOSPC; | |
5237 | } | |
5238 | spin_unlock(&space_info->lock); | |
f91587e4 JB |
5239 | |
5240 | if (reclaim_bytes) | |
5241 | space_info_add_old_bytes(fs_info, space_info, reclaim_bytes); | |
957780eb JB |
5242 | ASSERT(list_empty(&ticket.list)); |
5243 | return ret; | |
5244 | } | |
8bb8ab2e | 5245 | |
957780eb JB |
5246 | /** |
5247 | * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space | |
5248 | * @root - the root we're allocating for | |
5249 | * @block_rsv - the block_rsv we're allocating for | |
5250 | * @orig_bytes - the number of bytes we want | |
5251 | * @flush - whether or not we can flush to make our reservation | |
5252 | * | |
52042d8e | 5253 | * This will reserve orig_bytes number of bytes from the space info associated |
957780eb JB |
5254 | * with the block_rsv. If there is not enough space it will make an attempt to |
5255 | * flush out space to make room. It will do this by flushing delalloc if | |
5256 | * possible or committing the transaction. If flush is 0 then no attempts to | |
5257 | * regain reservations will be made and this will fail if there is not enough | |
5258 | * space already. | |
5259 | */ | |
5260 | static int reserve_metadata_bytes(struct btrfs_root *root, | |
5261 | struct btrfs_block_rsv *block_rsv, | |
5262 | u64 orig_bytes, | |
5263 | enum btrfs_reserve_flush_enum flush) | |
5264 | { | |
0b246afa JM |
5265 | struct btrfs_fs_info *fs_info = root->fs_info; |
5266 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
957780eb | 5267 | int ret; |
c1c4919b | 5268 | bool system_chunk = (root == fs_info->chunk_root); |
957780eb | 5269 | |
c1c4919b JM |
5270 | ret = __reserve_metadata_bytes(fs_info, block_rsv->space_info, |
5271 | orig_bytes, flush, system_chunk); | |
5d80366e JB |
5272 | if (ret == -ENOSPC && |
5273 | unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) { | |
5d80366e JB |
5274 | if (block_rsv != global_rsv && |
5275 | !block_rsv_use_bytes(global_rsv, orig_bytes)) | |
5276 | ret = 0; | |
5277 | } | |
9a3daff3 | 5278 | if (ret == -ENOSPC) { |
0b246afa | 5279 | trace_btrfs_space_reservation(fs_info, "space_info:enospc", |
957780eb JB |
5280 | block_rsv->space_info->flags, |
5281 | orig_bytes, 1); | |
9a3daff3 NB |
5282 | |
5283 | if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) | |
5284 | dump_space_info(fs_info, block_rsv->space_info, | |
5285 | orig_bytes, 0); | |
5286 | } | |
f0486c68 YZ |
5287 | return ret; |
5288 | } | |
5289 | ||
79787eaa JM |
5290 | static struct btrfs_block_rsv *get_block_rsv( |
5291 | const struct btrfs_trans_handle *trans, | |
5292 | const struct btrfs_root *root) | |
f0486c68 | 5293 | { |
0b246afa | 5294 | struct btrfs_fs_info *fs_info = root->fs_info; |
4c13d758 JB |
5295 | struct btrfs_block_rsv *block_rsv = NULL; |
5296 | ||
e9cf439f | 5297 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) || |
0b246afa JM |
5298 | (root == fs_info->csum_root && trans->adding_csums) || |
5299 | (root == fs_info->uuid_root)) | |
f7a81ea4 SB |
5300 | block_rsv = trans->block_rsv; |
5301 | ||
4c13d758 | 5302 | if (!block_rsv) |
f0486c68 YZ |
5303 | block_rsv = root->block_rsv; |
5304 | ||
5305 | if (!block_rsv) | |
0b246afa | 5306 | block_rsv = &fs_info->empty_block_rsv; |
f0486c68 YZ |
5307 | |
5308 | return block_rsv; | |
5309 | } | |
5310 | ||
5311 | static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, | |
5312 | u64 num_bytes) | |
5313 | { | |
5314 | int ret = -ENOSPC; | |
5315 | spin_lock(&block_rsv->lock); | |
5316 | if (block_rsv->reserved >= num_bytes) { | |
5317 | block_rsv->reserved -= num_bytes; | |
5318 | if (block_rsv->reserved < block_rsv->size) | |
5319 | block_rsv->full = 0; | |
5320 | ret = 0; | |
5321 | } | |
5322 | spin_unlock(&block_rsv->lock); | |
5323 | return ret; | |
5324 | } | |
5325 | ||
5326 | static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv, | |
3a584174 | 5327 | u64 num_bytes, bool update_size) |
f0486c68 YZ |
5328 | { |
5329 | spin_lock(&block_rsv->lock); | |
5330 | block_rsv->reserved += num_bytes; | |
5331 | if (update_size) | |
5332 | block_rsv->size += num_bytes; | |
5333 | else if (block_rsv->reserved >= block_rsv->size) | |
5334 | block_rsv->full = 1; | |
5335 | spin_unlock(&block_rsv->lock); | |
5336 | } | |
5337 | ||
d52be818 JB |
5338 | int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info, |
5339 | struct btrfs_block_rsv *dest, u64 num_bytes, | |
5340 | int min_factor) | |
5341 | { | |
5342 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
5343 | u64 min_bytes; | |
5344 | ||
5345 | if (global_rsv->space_info != dest->space_info) | |
5346 | return -ENOSPC; | |
5347 | ||
5348 | spin_lock(&global_rsv->lock); | |
5349 | min_bytes = div_factor(global_rsv->size, min_factor); | |
5350 | if (global_rsv->reserved < min_bytes + num_bytes) { | |
5351 | spin_unlock(&global_rsv->lock); | |
5352 | return -ENOSPC; | |
5353 | } | |
5354 | global_rsv->reserved -= num_bytes; | |
5355 | if (global_rsv->reserved < global_rsv->size) | |
5356 | global_rsv->full = 0; | |
5357 | spin_unlock(&global_rsv->lock); | |
5358 | ||
3a584174 | 5359 | block_rsv_add_bytes(dest, num_bytes, true); |
d52be818 JB |
5360 | return 0; |
5361 | } | |
5362 | ||
ba2c4d4e JB |
5363 | /** |
5364 | * btrfs_migrate_to_delayed_refs_rsv - transfer bytes to our delayed refs rsv. | |
5365 | * @fs_info - the fs info for our fs. | |
5366 | * @src - the source block rsv to transfer from. | |
5367 | * @num_bytes - the number of bytes to transfer. | |
5368 | * | |
5369 | * This transfers up to the num_bytes amount from the src rsv to the | |
5370 | * delayed_refs_rsv. Any extra bytes are returned to the space info. | |
5371 | */ | |
5372 | void btrfs_migrate_to_delayed_refs_rsv(struct btrfs_fs_info *fs_info, | |
5373 | struct btrfs_block_rsv *src, | |
5374 | u64 num_bytes) | |
5375 | { | |
5376 | struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv; | |
5377 | u64 to_free = 0; | |
5378 | ||
5379 | spin_lock(&src->lock); | |
5380 | src->reserved -= num_bytes; | |
5381 | src->size -= num_bytes; | |
5382 | spin_unlock(&src->lock); | |
5383 | ||
5384 | spin_lock(&delayed_refs_rsv->lock); | |
5385 | if (delayed_refs_rsv->size > delayed_refs_rsv->reserved) { | |
5386 | u64 delta = delayed_refs_rsv->size - | |
5387 | delayed_refs_rsv->reserved; | |
5388 | if (num_bytes > delta) { | |
5389 | to_free = num_bytes - delta; | |
5390 | num_bytes = delta; | |
5391 | } | |
5392 | } else { | |
5393 | to_free = num_bytes; | |
5394 | num_bytes = 0; | |
5395 | } | |
5396 | ||
5397 | if (num_bytes) | |
5398 | delayed_refs_rsv->reserved += num_bytes; | |
5399 | if (delayed_refs_rsv->reserved >= delayed_refs_rsv->size) | |
5400 | delayed_refs_rsv->full = 1; | |
5401 | spin_unlock(&delayed_refs_rsv->lock); | |
5402 | ||
5403 | if (num_bytes) | |
5404 | trace_btrfs_space_reservation(fs_info, "delayed_refs_rsv", | |
5405 | 0, num_bytes, 1); | |
5406 | if (to_free) | |
5407 | space_info_add_old_bytes(fs_info, delayed_refs_rsv->space_info, | |
5408 | to_free); | |
5409 | } | |
5410 | ||
5411 | /** | |
5412 | * btrfs_delayed_refs_rsv_refill - refill based on our delayed refs usage. | |
5413 | * @fs_info - the fs_info for our fs. | |
5414 | * @flush - control how we can flush for this reservation. | |
5415 | * | |
5416 | * This will refill the delayed block_rsv up to 1 items size worth of space and | |
5417 | * will return -ENOSPC if we can't make the reservation. | |
5418 | */ | |
5419 | int btrfs_delayed_refs_rsv_refill(struct btrfs_fs_info *fs_info, | |
5420 | enum btrfs_reserve_flush_enum flush) | |
5421 | { | |
5422 | struct btrfs_block_rsv *block_rsv = &fs_info->delayed_refs_rsv; | |
5423 | u64 limit = btrfs_calc_trans_metadata_size(fs_info, 1); | |
5424 | u64 num_bytes = 0; | |
5425 | int ret = -ENOSPC; | |
5426 | ||
5427 | spin_lock(&block_rsv->lock); | |
5428 | if (block_rsv->reserved < block_rsv->size) { | |
5429 | num_bytes = block_rsv->size - block_rsv->reserved; | |
5430 | num_bytes = min(num_bytes, limit); | |
5431 | } | |
5432 | spin_unlock(&block_rsv->lock); | |
5433 | ||
5434 | if (!num_bytes) | |
5435 | return 0; | |
5436 | ||
5437 | ret = reserve_metadata_bytes(fs_info->extent_root, block_rsv, | |
5438 | num_bytes, flush); | |
5439 | if (ret) | |
5440 | return ret; | |
5441 | block_rsv_add_bytes(block_rsv, num_bytes, 0); | |
5442 | trace_btrfs_space_reservation(fs_info, "delayed_refs_rsv", | |
5443 | 0, num_bytes, 1); | |
5444 | return 0; | |
5445 | } | |
5446 | ||
957780eb JB |
5447 | /* |
5448 | * This is for space we already have accounted in space_info->bytes_may_use, so | |
5449 | * basically when we're returning space from block_rsv's. | |
5450 | */ | |
5451 | static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info, | |
5452 | struct btrfs_space_info *space_info, | |
5453 | u64 num_bytes) | |
5454 | { | |
5455 | struct reserve_ticket *ticket; | |
5456 | struct list_head *head; | |
5457 | u64 used; | |
5458 | enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_NO_FLUSH; | |
5459 | bool check_overcommit = false; | |
5460 | ||
5461 | spin_lock(&space_info->lock); | |
5462 | head = &space_info->priority_tickets; | |
5463 | ||
5464 | /* | |
5465 | * If we are over our limit then we need to check and see if we can | |
5466 | * overcommit, and if we can't then we just need to free up our space | |
5467 | * and not satisfy any requests. | |
5468 | */ | |
0eee8a49 | 5469 | used = btrfs_space_info_used(space_info, true); |
957780eb JB |
5470 | if (used - num_bytes >= space_info->total_bytes) |
5471 | check_overcommit = true; | |
5472 | again: | |
5473 | while (!list_empty(head) && num_bytes) { | |
5474 | ticket = list_first_entry(head, struct reserve_ticket, | |
5475 | list); | |
5476 | /* | |
5477 | * We use 0 bytes because this space is already reserved, so | |
5478 | * adding the ticket space would be a double count. | |
5479 | */ | |
5480 | if (check_overcommit && | |
c1c4919b | 5481 | !can_overcommit(fs_info, space_info, 0, flush, false)) |
957780eb JB |
5482 | break; |
5483 | if (num_bytes >= ticket->bytes) { | |
5484 | list_del_init(&ticket->list); | |
5485 | num_bytes -= ticket->bytes; | |
5486 | ticket->bytes = 0; | |
ce129655 | 5487 | space_info->tickets_id++; |
957780eb JB |
5488 | wake_up(&ticket->wait); |
5489 | } else { | |
5490 | ticket->bytes -= num_bytes; | |
5491 | num_bytes = 0; | |
5492 | } | |
5493 | } | |
5494 | ||
5495 | if (num_bytes && head == &space_info->priority_tickets) { | |
5496 | head = &space_info->tickets; | |
5497 | flush = BTRFS_RESERVE_FLUSH_ALL; | |
5498 | goto again; | |
5499 | } | |
9f9b8e8d | 5500 | update_bytes_may_use(space_info, -num_bytes); |
957780eb JB |
5501 | trace_btrfs_space_reservation(fs_info, "space_info", |
5502 | space_info->flags, num_bytes, 0); | |
5503 | spin_unlock(&space_info->lock); | |
5504 | } | |
5505 | ||
5506 | /* | |
5507 | * This is for newly allocated space that isn't accounted in | |
5508 | * space_info->bytes_may_use yet. So if we allocate a chunk or unpin an extent | |
5509 | * we use this helper. | |
5510 | */ | |
5511 | static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info, | |
5512 | struct btrfs_space_info *space_info, | |
5513 | u64 num_bytes) | |
5514 | { | |
5515 | struct reserve_ticket *ticket; | |
5516 | struct list_head *head = &space_info->priority_tickets; | |
5517 | ||
5518 | again: | |
5519 | while (!list_empty(head) && num_bytes) { | |
5520 | ticket = list_first_entry(head, struct reserve_ticket, | |
5521 | list); | |
5522 | if (num_bytes >= ticket->bytes) { | |
5523 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5524 | space_info->flags, | |
5525 | ticket->bytes, 1); | |
5526 | list_del_init(&ticket->list); | |
5527 | num_bytes -= ticket->bytes; | |
9f9b8e8d | 5528 | update_bytes_may_use(space_info, ticket->bytes); |
957780eb | 5529 | ticket->bytes = 0; |
ce129655 | 5530 | space_info->tickets_id++; |
957780eb JB |
5531 | wake_up(&ticket->wait); |
5532 | } else { | |
5533 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5534 | space_info->flags, | |
5535 | num_bytes, 1); | |
9f9b8e8d | 5536 | update_bytes_may_use(space_info, num_bytes); |
957780eb JB |
5537 | ticket->bytes -= num_bytes; |
5538 | num_bytes = 0; | |
5539 | } | |
5540 | } | |
5541 | ||
5542 | if (num_bytes && head == &space_info->priority_tickets) { | |
5543 | head = &space_info->tickets; | |
5544 | goto again; | |
5545 | } | |
5546 | } | |
5547 | ||
69fe2d75 | 5548 | static u64 block_rsv_release_bytes(struct btrfs_fs_info *fs_info, |
8c2a3ca2 | 5549 | struct btrfs_block_rsv *block_rsv, |
ff6bc37e QW |
5550 | struct btrfs_block_rsv *dest, u64 num_bytes, |
5551 | u64 *qgroup_to_release_ret) | |
f0486c68 YZ |
5552 | { |
5553 | struct btrfs_space_info *space_info = block_rsv->space_info; | |
ff6bc37e | 5554 | u64 qgroup_to_release = 0; |
69fe2d75 | 5555 | u64 ret; |
f0486c68 YZ |
5556 | |
5557 | spin_lock(&block_rsv->lock); | |
ff6bc37e | 5558 | if (num_bytes == (u64)-1) { |
f0486c68 | 5559 | num_bytes = block_rsv->size; |
ff6bc37e QW |
5560 | qgroup_to_release = block_rsv->qgroup_rsv_size; |
5561 | } | |
f0486c68 YZ |
5562 | block_rsv->size -= num_bytes; |
5563 | if (block_rsv->reserved >= block_rsv->size) { | |
5564 | num_bytes = block_rsv->reserved - block_rsv->size; | |
5565 | block_rsv->reserved = block_rsv->size; | |
5566 | block_rsv->full = 1; | |
5567 | } else { | |
5568 | num_bytes = 0; | |
5569 | } | |
ff6bc37e QW |
5570 | if (block_rsv->qgroup_rsv_reserved >= block_rsv->qgroup_rsv_size) { |
5571 | qgroup_to_release = block_rsv->qgroup_rsv_reserved - | |
5572 | block_rsv->qgroup_rsv_size; | |
5573 | block_rsv->qgroup_rsv_reserved = block_rsv->qgroup_rsv_size; | |
5574 | } else { | |
5575 | qgroup_to_release = 0; | |
5576 | } | |
f0486c68 YZ |
5577 | spin_unlock(&block_rsv->lock); |
5578 | ||
69fe2d75 | 5579 | ret = num_bytes; |
f0486c68 YZ |
5580 | if (num_bytes > 0) { |
5581 | if (dest) { | |
e9e22899 JB |
5582 | spin_lock(&dest->lock); |
5583 | if (!dest->full) { | |
5584 | u64 bytes_to_add; | |
5585 | ||
5586 | bytes_to_add = dest->size - dest->reserved; | |
5587 | bytes_to_add = min(num_bytes, bytes_to_add); | |
5588 | dest->reserved += bytes_to_add; | |
5589 | if (dest->reserved >= dest->size) | |
5590 | dest->full = 1; | |
5591 | num_bytes -= bytes_to_add; | |
5592 | } | |
5593 | spin_unlock(&dest->lock); | |
5594 | } | |
957780eb JB |
5595 | if (num_bytes) |
5596 | space_info_add_old_bytes(fs_info, space_info, | |
5597 | num_bytes); | |
9ed74f2d | 5598 | } |
ff6bc37e QW |
5599 | if (qgroup_to_release_ret) |
5600 | *qgroup_to_release_ret = qgroup_to_release; | |
69fe2d75 | 5601 | return ret; |
f0486c68 | 5602 | } |
4e06bdd6 | 5603 | |
25d609f8 JB |
5604 | int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src, |
5605 | struct btrfs_block_rsv *dst, u64 num_bytes, | |
3a584174 | 5606 | bool update_size) |
f0486c68 YZ |
5607 | { |
5608 | int ret; | |
9ed74f2d | 5609 | |
f0486c68 YZ |
5610 | ret = block_rsv_use_bytes(src, num_bytes); |
5611 | if (ret) | |
5612 | return ret; | |
9ed74f2d | 5613 | |
25d609f8 | 5614 | block_rsv_add_bytes(dst, num_bytes, update_size); |
9ed74f2d JB |
5615 | return 0; |
5616 | } | |
5617 | ||
66d8f3dd | 5618 | void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type) |
9ed74f2d | 5619 | { |
f0486c68 YZ |
5620 | memset(rsv, 0, sizeof(*rsv)); |
5621 | spin_lock_init(&rsv->lock); | |
66d8f3dd | 5622 | rsv->type = type; |
f0486c68 YZ |
5623 | } |
5624 | ||
69fe2d75 JB |
5625 | void btrfs_init_metadata_block_rsv(struct btrfs_fs_info *fs_info, |
5626 | struct btrfs_block_rsv *rsv, | |
5627 | unsigned short type) | |
5628 | { | |
5629 | btrfs_init_block_rsv(rsv, type); | |
5630 | rsv->space_info = __find_space_info(fs_info, | |
5631 | BTRFS_BLOCK_GROUP_METADATA); | |
5632 | } | |
5633 | ||
2ff7e61e | 5634 | struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_fs_info *fs_info, |
66d8f3dd | 5635 | unsigned short type) |
f0486c68 YZ |
5636 | { |
5637 | struct btrfs_block_rsv *block_rsv; | |
9ed74f2d | 5638 | |
f0486c68 YZ |
5639 | block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS); |
5640 | if (!block_rsv) | |
5641 | return NULL; | |
9ed74f2d | 5642 | |
69fe2d75 | 5643 | btrfs_init_metadata_block_rsv(fs_info, block_rsv, type); |
f0486c68 YZ |
5644 | return block_rsv; |
5645 | } | |
9ed74f2d | 5646 | |
2ff7e61e | 5647 | void btrfs_free_block_rsv(struct btrfs_fs_info *fs_info, |
f0486c68 YZ |
5648 | struct btrfs_block_rsv *rsv) |
5649 | { | |
2aaa6655 JB |
5650 | if (!rsv) |
5651 | return; | |
2ff7e61e | 5652 | btrfs_block_rsv_release(fs_info, rsv, (u64)-1); |
dabdb640 | 5653 | kfree(rsv); |
9ed74f2d JB |
5654 | } |
5655 | ||
08e007d2 MX |
5656 | int btrfs_block_rsv_add(struct btrfs_root *root, |
5657 | struct btrfs_block_rsv *block_rsv, u64 num_bytes, | |
5658 | enum btrfs_reserve_flush_enum flush) | |
9ed74f2d | 5659 | { |
f0486c68 | 5660 | int ret; |
9ed74f2d | 5661 | |
f0486c68 YZ |
5662 | if (num_bytes == 0) |
5663 | return 0; | |
8bb8ab2e | 5664 | |
61b520a9 | 5665 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
5a2cb25a | 5666 | if (!ret) |
3a584174 | 5667 | block_rsv_add_bytes(block_rsv, num_bytes, true); |
9ed74f2d | 5668 | |
f0486c68 | 5669 | return ret; |
f0486c68 | 5670 | } |
9ed74f2d | 5671 | |
2ff7e61e | 5672 | int btrfs_block_rsv_check(struct btrfs_block_rsv *block_rsv, int min_factor) |
f0486c68 YZ |
5673 | { |
5674 | u64 num_bytes = 0; | |
f0486c68 | 5675 | int ret = -ENOSPC; |
9ed74f2d | 5676 | |
f0486c68 YZ |
5677 | if (!block_rsv) |
5678 | return 0; | |
9ed74f2d | 5679 | |
f0486c68 | 5680 | spin_lock(&block_rsv->lock); |
36ba022a JB |
5681 | num_bytes = div_factor(block_rsv->size, min_factor); |
5682 | if (block_rsv->reserved >= num_bytes) | |
5683 | ret = 0; | |
5684 | spin_unlock(&block_rsv->lock); | |
9ed74f2d | 5685 | |
36ba022a JB |
5686 | return ret; |
5687 | } | |
5688 | ||
08e007d2 MX |
5689 | int btrfs_block_rsv_refill(struct btrfs_root *root, |
5690 | struct btrfs_block_rsv *block_rsv, u64 min_reserved, | |
5691 | enum btrfs_reserve_flush_enum flush) | |
36ba022a JB |
5692 | { |
5693 | u64 num_bytes = 0; | |
5694 | int ret = -ENOSPC; | |
5695 | ||
5696 | if (!block_rsv) | |
5697 | return 0; | |
5698 | ||
5699 | spin_lock(&block_rsv->lock); | |
5700 | num_bytes = min_reserved; | |
13553e52 | 5701 | if (block_rsv->reserved >= num_bytes) |
f0486c68 | 5702 | ret = 0; |
13553e52 | 5703 | else |
f0486c68 | 5704 | num_bytes -= block_rsv->reserved; |
f0486c68 | 5705 | spin_unlock(&block_rsv->lock); |
13553e52 | 5706 | |
f0486c68 YZ |
5707 | if (!ret) |
5708 | return 0; | |
5709 | ||
aa38a711 | 5710 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
dabdb640 | 5711 | if (!ret) { |
3a584174 | 5712 | block_rsv_add_bytes(block_rsv, num_bytes, false); |
f0486c68 | 5713 | return 0; |
6a63209f | 5714 | } |
9ed74f2d | 5715 | |
13553e52 | 5716 | return ret; |
f0486c68 YZ |
5717 | } |
5718 | ||
ba2c4d4e JB |
5719 | static u64 __btrfs_block_rsv_release(struct btrfs_fs_info *fs_info, |
5720 | struct btrfs_block_rsv *block_rsv, | |
5721 | u64 num_bytes, u64 *qgroup_to_release) | |
5722 | { | |
5723 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
5724 | struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_refs_rsv; | |
5725 | struct btrfs_block_rsv *target = delayed_rsv; | |
5726 | ||
5727 | if (target->full || target == block_rsv) | |
5728 | target = global_rsv; | |
5729 | ||
5730 | if (block_rsv->space_info != target->space_info) | |
5731 | target = NULL; | |
5732 | ||
5733 | return block_rsv_release_bytes(fs_info, block_rsv, target, num_bytes, | |
5734 | qgroup_to_release); | |
5735 | } | |
5736 | ||
5737 | void btrfs_block_rsv_release(struct btrfs_fs_info *fs_info, | |
5738 | struct btrfs_block_rsv *block_rsv, | |
5739 | u64 num_bytes) | |
5740 | { | |
5741 | __btrfs_block_rsv_release(fs_info, block_rsv, num_bytes, NULL); | |
5742 | } | |
5743 | ||
69fe2d75 JB |
5744 | /** |
5745 | * btrfs_inode_rsv_release - release any excessive reservation. | |
5746 | * @inode - the inode we need to release from. | |
43b18595 QW |
5747 | * @qgroup_free - free or convert qgroup meta. |
5748 | * Unlike normal operation, qgroup meta reservation needs to know if we are | |
5749 | * freeing qgroup reservation or just converting it into per-trans. Normally | |
5750 | * @qgroup_free is true for error handling, and false for normal release. | |
69fe2d75 JB |
5751 | * |
5752 | * This is the same as btrfs_block_rsv_release, except that it handles the | |
5753 | * tracepoint for the reservation. | |
5754 | */ | |
43b18595 | 5755 | static void btrfs_inode_rsv_release(struct btrfs_inode *inode, bool qgroup_free) |
69fe2d75 JB |
5756 | { |
5757 | struct btrfs_fs_info *fs_info = inode->root->fs_info; | |
69fe2d75 JB |
5758 | struct btrfs_block_rsv *block_rsv = &inode->block_rsv; |
5759 | u64 released = 0; | |
ff6bc37e | 5760 | u64 qgroup_to_release = 0; |
69fe2d75 JB |
5761 | |
5762 | /* | |
5763 | * Since we statically set the block_rsv->size we just want to say we | |
5764 | * are releasing 0 bytes, and then we'll just get the reservation over | |
5765 | * the size free'd. | |
5766 | */ | |
ba2c4d4e JB |
5767 | released = __btrfs_block_rsv_release(fs_info, block_rsv, 0, |
5768 | &qgroup_to_release); | |
69fe2d75 JB |
5769 | if (released > 0) |
5770 | trace_btrfs_space_reservation(fs_info, "delalloc", | |
5771 | btrfs_ino(inode), released, 0); | |
43b18595 | 5772 | if (qgroup_free) |
ff6bc37e | 5773 | btrfs_qgroup_free_meta_prealloc(inode->root, qgroup_to_release); |
43b18595 | 5774 | else |
ff6bc37e QW |
5775 | btrfs_qgroup_convert_reserved_meta(inode->root, |
5776 | qgroup_to_release); | |
69fe2d75 JB |
5777 | } |
5778 | ||
ba2c4d4e JB |
5779 | /** |
5780 | * btrfs_delayed_refs_rsv_release - release a ref head's reservation. | |
5781 | * @fs_info - the fs_info for our fs. | |
5782 | * @nr - the number of items to drop. | |
5783 | * | |
5784 | * This drops the delayed ref head's count from the delayed refs rsv and frees | |
5785 | * any excess reservation we had. | |
5786 | */ | |
5787 | void btrfs_delayed_refs_rsv_release(struct btrfs_fs_info *fs_info, int nr) | |
f0486c68 | 5788 | { |
ba2c4d4e | 5789 | struct btrfs_block_rsv *block_rsv = &fs_info->delayed_refs_rsv; |
0b246afa | 5790 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; |
ba2c4d4e JB |
5791 | u64 num_bytes = btrfs_calc_trans_metadata_size(fs_info, nr); |
5792 | u64 released = 0; | |
0b246afa | 5793 | |
ba2c4d4e JB |
5794 | released = block_rsv_release_bytes(fs_info, block_rsv, global_rsv, |
5795 | num_bytes, NULL); | |
5796 | if (released) | |
5797 | trace_btrfs_space_reservation(fs_info, "delayed_refs_rsv", | |
5798 | 0, released, 0); | |
6a63209f JB |
5799 | } |
5800 | ||
8929ecfa YZ |
5801 | static void update_global_block_rsv(struct btrfs_fs_info *fs_info) |
5802 | { | |
5803 | struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv; | |
5804 | struct btrfs_space_info *sinfo = block_rsv->space_info; | |
5805 | u64 num_bytes; | |
6a63209f | 5806 | |
ae2e4728 JB |
5807 | /* |
5808 | * The global block rsv is based on the size of the extent tree, the | |
5809 | * checksum tree and the root tree. If the fs is empty we want to set | |
5810 | * it to a minimal amount for safety. | |
5811 | */ | |
5812 | num_bytes = btrfs_root_used(&fs_info->extent_root->root_item) + | |
5813 | btrfs_root_used(&fs_info->csum_root->root_item) + | |
5814 | btrfs_root_used(&fs_info->tree_root->root_item); | |
5815 | num_bytes = max_t(u64, num_bytes, SZ_16M); | |
33b4d47f | 5816 | |
8929ecfa | 5817 | spin_lock(&sinfo->lock); |
1f699d38 | 5818 | spin_lock(&block_rsv->lock); |
4e06bdd6 | 5819 | |
ee22184b | 5820 | block_rsv->size = min_t(u64, num_bytes, SZ_512M); |
4e06bdd6 | 5821 | |
fb4b10e5 | 5822 | if (block_rsv->reserved < block_rsv->size) { |
4136135b | 5823 | num_bytes = btrfs_space_info_used(sinfo, true); |
fb4b10e5 JB |
5824 | if (sinfo->total_bytes > num_bytes) { |
5825 | num_bytes = sinfo->total_bytes - num_bytes; | |
5826 | num_bytes = min(num_bytes, | |
5827 | block_rsv->size - block_rsv->reserved); | |
5828 | block_rsv->reserved += num_bytes; | |
9f9b8e8d | 5829 | update_bytes_may_use(sinfo, num_bytes); |
fb4b10e5 JB |
5830 | trace_btrfs_space_reservation(fs_info, "space_info", |
5831 | sinfo->flags, num_bytes, | |
5832 | 1); | |
5833 | } | |
5834 | } else if (block_rsv->reserved > block_rsv->size) { | |
8929ecfa | 5835 | num_bytes = block_rsv->reserved - block_rsv->size; |
9f9b8e8d | 5836 | update_bytes_may_use(sinfo, -num_bytes); |
8c2a3ca2 | 5837 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 5838 | sinfo->flags, num_bytes, 0); |
8929ecfa | 5839 | block_rsv->reserved = block_rsv->size; |
8929ecfa | 5840 | } |
182608c8 | 5841 | |
fb4b10e5 JB |
5842 | if (block_rsv->reserved == block_rsv->size) |
5843 | block_rsv->full = 1; | |
5844 | else | |
5845 | block_rsv->full = 0; | |
5846 | ||
8929ecfa | 5847 | spin_unlock(&block_rsv->lock); |
1f699d38 | 5848 | spin_unlock(&sinfo->lock); |
6a63209f JB |
5849 | } |
5850 | ||
f0486c68 | 5851 | static void init_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 5852 | { |
f0486c68 | 5853 | struct btrfs_space_info *space_info; |
6a63209f | 5854 | |
f0486c68 YZ |
5855 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); |
5856 | fs_info->chunk_block_rsv.space_info = space_info; | |
6a63209f | 5857 | |
f0486c68 | 5858 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
8929ecfa | 5859 | fs_info->global_block_rsv.space_info = space_info; |
f0486c68 YZ |
5860 | fs_info->trans_block_rsv.space_info = space_info; |
5861 | fs_info->empty_block_rsv.space_info = space_info; | |
6d668dda | 5862 | fs_info->delayed_block_rsv.space_info = space_info; |
ba2c4d4e | 5863 | fs_info->delayed_refs_rsv.space_info = space_info; |
f0486c68 | 5864 | |
ba2c4d4e JB |
5865 | fs_info->extent_root->block_rsv = &fs_info->delayed_refs_rsv; |
5866 | fs_info->csum_root->block_rsv = &fs_info->delayed_refs_rsv; | |
8929ecfa YZ |
5867 | fs_info->dev_root->block_rsv = &fs_info->global_block_rsv; |
5868 | fs_info->tree_root->block_rsv = &fs_info->global_block_rsv; | |
3a6cad90 SB |
5869 | if (fs_info->quota_root) |
5870 | fs_info->quota_root->block_rsv = &fs_info->global_block_rsv; | |
f0486c68 | 5871 | fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv; |
8929ecfa | 5872 | |
8929ecfa | 5873 | update_global_block_rsv(fs_info); |
6a63209f JB |
5874 | } |
5875 | ||
8929ecfa | 5876 | static void release_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 5877 | { |
8c2a3ca2 | 5878 | block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL, |
ff6bc37e | 5879 | (u64)-1, NULL); |
8929ecfa YZ |
5880 | WARN_ON(fs_info->trans_block_rsv.size > 0); |
5881 | WARN_ON(fs_info->trans_block_rsv.reserved > 0); | |
5882 | WARN_ON(fs_info->chunk_block_rsv.size > 0); | |
5883 | WARN_ON(fs_info->chunk_block_rsv.reserved > 0); | |
6d668dda JB |
5884 | WARN_ON(fs_info->delayed_block_rsv.size > 0); |
5885 | WARN_ON(fs_info->delayed_block_rsv.reserved > 0); | |
ba2c4d4e JB |
5886 | WARN_ON(fs_info->delayed_refs_rsv.reserved > 0); |
5887 | WARN_ON(fs_info->delayed_refs_rsv.size > 0); | |
fcb80c2a JB |
5888 | } |
5889 | ||
ba2c4d4e JB |
5890 | /* |
5891 | * btrfs_update_delayed_refs_rsv - adjust the size of the delayed refs rsv | |
5892 | * @trans - the trans that may have generated delayed refs | |
5893 | * | |
5894 | * This is to be called anytime we may have adjusted trans->delayed_ref_updates, | |
5895 | * it'll calculate the additional size and add it to the delayed_refs_rsv. | |
5896 | */ | |
5897 | void btrfs_update_delayed_refs_rsv(struct btrfs_trans_handle *trans) | |
5898 | { | |
5899 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
5900 | struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_refs_rsv; | |
5901 | u64 num_bytes; | |
5902 | ||
5903 | if (!trans->delayed_ref_updates) | |
5904 | return; | |
5905 | ||
5906 | num_bytes = btrfs_calc_trans_metadata_size(fs_info, | |
5907 | trans->delayed_ref_updates); | |
5908 | spin_lock(&delayed_rsv->lock); | |
5909 | delayed_rsv->size += num_bytes; | |
5910 | delayed_rsv->full = 0; | |
5911 | spin_unlock(&delayed_rsv->lock); | |
5912 | trans->delayed_ref_updates = 0; | |
5913 | } | |
6a63209f | 5914 | |
4fbcdf66 FM |
5915 | /* |
5916 | * To be called after all the new block groups attached to the transaction | |
5917 | * handle have been created (btrfs_create_pending_block_groups()). | |
5918 | */ | |
5919 | void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans) | |
5920 | { | |
64b63580 | 5921 | struct btrfs_fs_info *fs_info = trans->fs_info; |
4fbcdf66 FM |
5922 | |
5923 | if (!trans->chunk_bytes_reserved) | |
5924 | return; | |
5925 | ||
5926 | WARN_ON_ONCE(!list_empty(&trans->new_bgs)); | |
5927 | ||
5928 | block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL, | |
ff6bc37e | 5929 | trans->chunk_bytes_reserved, NULL); |
4fbcdf66 FM |
5930 | trans->chunk_bytes_reserved = 0; |
5931 | } | |
5932 | ||
d5c12070 MX |
5933 | /* |
5934 | * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation | |
5935 | * root: the root of the parent directory | |
5936 | * rsv: block reservation | |
5937 | * items: the number of items that we need do reservation | |
a5b7f429 | 5938 | * use_global_rsv: allow fallback to the global block reservation |
d5c12070 MX |
5939 | * |
5940 | * This function is used to reserve the space for snapshot/subvolume | |
5941 | * creation and deletion. Those operations are different with the | |
5942 | * common file/directory operations, they change two fs/file trees | |
5943 | * and root tree, the number of items that the qgroup reserves is | |
5944 | * different with the free space reservation. So we can not use | |
01327610 | 5945 | * the space reservation mechanism in start_transaction(). |
d5c12070 MX |
5946 | */ |
5947 | int btrfs_subvolume_reserve_metadata(struct btrfs_root *root, | |
a5b7f429 | 5948 | struct btrfs_block_rsv *rsv, int items, |
ee3441b4 | 5949 | bool use_global_rsv) |
a22285a6 | 5950 | { |
a5b7f429 | 5951 | u64 qgroup_num_bytes = 0; |
d5c12070 MX |
5952 | u64 num_bytes; |
5953 | int ret; | |
0b246afa JM |
5954 | struct btrfs_fs_info *fs_info = root->fs_info; |
5955 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
d5c12070 | 5956 | |
0b246afa | 5957 | if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) { |
d5c12070 | 5958 | /* One for parent inode, two for dir entries */ |
a5b7f429 LF |
5959 | qgroup_num_bytes = 3 * fs_info->nodesize; |
5960 | ret = btrfs_qgroup_reserve_meta_prealloc(root, | |
5961 | qgroup_num_bytes, true); | |
d5c12070 MX |
5962 | if (ret) |
5963 | return ret; | |
d5c12070 MX |
5964 | } |
5965 | ||
0b246afa JM |
5966 | num_bytes = btrfs_calc_trans_metadata_size(fs_info, items); |
5967 | rsv->space_info = __find_space_info(fs_info, | |
d5c12070 MX |
5968 | BTRFS_BLOCK_GROUP_METADATA); |
5969 | ret = btrfs_block_rsv_add(root, rsv, num_bytes, | |
5970 | BTRFS_RESERVE_FLUSH_ALL); | |
ee3441b4 JM |
5971 | |
5972 | if (ret == -ENOSPC && use_global_rsv) | |
3a584174 | 5973 | ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes, true); |
ee3441b4 | 5974 | |
a5b7f429 LF |
5975 | if (ret && qgroup_num_bytes) |
5976 | btrfs_qgroup_free_meta_prealloc(root, qgroup_num_bytes); | |
d5c12070 MX |
5977 | |
5978 | return ret; | |
5979 | } | |
5980 | ||
2ff7e61e | 5981 | void btrfs_subvolume_release_metadata(struct btrfs_fs_info *fs_info, |
7775c818 | 5982 | struct btrfs_block_rsv *rsv) |
d5c12070 | 5983 | { |
2ff7e61e | 5984 | btrfs_block_rsv_release(fs_info, rsv, (u64)-1); |
97e728d4 JB |
5985 | } |
5986 | ||
69fe2d75 JB |
5987 | static void btrfs_calculate_inode_block_rsv_size(struct btrfs_fs_info *fs_info, |
5988 | struct btrfs_inode *inode) | |
9e0baf60 | 5989 | { |
69fe2d75 JB |
5990 | struct btrfs_block_rsv *block_rsv = &inode->block_rsv; |
5991 | u64 reserve_size = 0; | |
ff6bc37e | 5992 | u64 qgroup_rsv_size = 0; |
69fe2d75 JB |
5993 | u64 csum_leaves; |
5994 | unsigned outstanding_extents; | |
9e0baf60 | 5995 | |
69fe2d75 JB |
5996 | lockdep_assert_held(&inode->lock); |
5997 | outstanding_extents = inode->outstanding_extents; | |
5998 | if (outstanding_extents) | |
5999 | reserve_size = btrfs_calc_trans_metadata_size(fs_info, | |
6000 | outstanding_extents + 1); | |
6001 | csum_leaves = btrfs_csum_bytes_to_leaves(fs_info, | |
6002 | inode->csum_bytes); | |
6003 | reserve_size += btrfs_calc_trans_metadata_size(fs_info, | |
6004 | csum_leaves); | |
ff6bc37e QW |
6005 | /* |
6006 | * For qgroup rsv, the calculation is very simple: | |
6007 | * account one nodesize for each outstanding extent | |
6008 | * | |
6009 | * This is overestimating in most cases. | |
6010 | */ | |
139a5617 | 6011 | qgroup_rsv_size = (u64)outstanding_extents * fs_info->nodesize; |
9e0baf60 | 6012 | |
69fe2d75 JB |
6013 | spin_lock(&block_rsv->lock); |
6014 | block_rsv->size = reserve_size; | |
ff6bc37e | 6015 | block_rsv->qgroup_rsv_size = qgroup_rsv_size; |
69fe2d75 | 6016 | spin_unlock(&block_rsv->lock); |
0ca1f7ce | 6017 | } |
c146afad | 6018 | |
c8eaeac7 JB |
6019 | static void calc_inode_reservations(struct btrfs_fs_info *fs_info, |
6020 | u64 num_bytes, u64 *meta_reserve, | |
6021 | u64 *qgroup_reserve) | |
6022 | { | |
6023 | u64 nr_extents = count_max_extents(num_bytes); | |
6024 | u64 csum_leaves = btrfs_csum_bytes_to_leaves(fs_info, num_bytes); | |
6025 | ||
6026 | /* We add one for the inode update at finish ordered time */ | |
6027 | *meta_reserve = btrfs_calc_trans_metadata_size(fs_info, | |
6028 | nr_extents + csum_leaves + 1); | |
6029 | *qgroup_reserve = nr_extents * fs_info->nodesize; | |
6030 | } | |
6031 | ||
9f3db423 | 6032 | int btrfs_delalloc_reserve_metadata(struct btrfs_inode *inode, u64 num_bytes) |
0ca1f7ce | 6033 | { |
c8eaeac7 JB |
6034 | struct btrfs_root *root = inode->root; |
6035 | struct btrfs_fs_info *fs_info = root->fs_info; | |
6036 | struct btrfs_block_rsv *block_rsv = &inode->block_rsv; | |
6037 | u64 meta_reserve, qgroup_reserve; | |
69fe2d75 | 6038 | unsigned nr_extents; |
08e007d2 | 6039 | enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL; |
eb6b88d9 | 6040 | int ret = 0; |
c64c2bd8 | 6041 | bool delalloc_lock = true; |
6324fbf3 | 6042 | |
c64c2bd8 JB |
6043 | /* If we are a free space inode we need to not flush since we will be in |
6044 | * the middle of a transaction commit. We also don't need the delalloc | |
6045 | * mutex since we won't race with anybody. We need this mostly to make | |
6046 | * lockdep shut its filthy mouth. | |
bac357dc JB |
6047 | * |
6048 | * If we have a transaction open (can happen if we call truncate_block | |
6049 | * from truncate), then we need FLUSH_LIMIT so we don't deadlock. | |
c64c2bd8 JB |
6050 | */ |
6051 | if (btrfs_is_free_space_inode(inode)) { | |
08e007d2 | 6052 | flush = BTRFS_RESERVE_NO_FLUSH; |
c64c2bd8 | 6053 | delalloc_lock = false; |
da07d4ab NB |
6054 | } else { |
6055 | if (current->journal_info) | |
6056 | flush = BTRFS_RESERVE_FLUSH_LIMIT; | |
c09544e0 | 6057 | |
da07d4ab NB |
6058 | if (btrfs_transaction_in_commit(fs_info)) |
6059 | schedule_timeout(1); | |
6060 | } | |
ec44a35c | 6061 | |
c64c2bd8 | 6062 | if (delalloc_lock) |
9f3db423 | 6063 | mutex_lock(&inode->delalloc_mutex); |
c64c2bd8 | 6064 | |
0b246afa | 6065 | num_bytes = ALIGN(num_bytes, fs_info->sectorsize); |
69fe2d75 | 6066 | |
c8eaeac7 JB |
6067 | /* |
6068 | * We always want to do it this way, every other way is wrong and ends | |
6069 | * in tears. Pre-reserving the amount we are going to add will always | |
6070 | * be the right way, because otherwise if we have enough parallelism we | |
6071 | * could end up with thousands of inodes all holding little bits of | |
6072 | * reservations they were able to make previously and the only way to | |
6073 | * reclaim that space is to ENOSPC out the operations and clear | |
6074 | * everything out and try again, which is bad. This way we just | |
6075 | * over-reserve slightly, and clean up the mess when we are done. | |
6076 | */ | |
6077 | calc_inode_reservations(fs_info, num_bytes, &meta_reserve, | |
6078 | &qgroup_reserve); | |
6079 | ret = btrfs_qgroup_reserve_meta_prealloc(root, qgroup_reserve, true); | |
6080 | if (ret) | |
6081 | goto out_fail; | |
6082 | ret = reserve_metadata_bytes(root, block_rsv, meta_reserve, flush); | |
6083 | if (ret) | |
6084 | goto out_qgroup; | |
6085 | ||
6086 | /* | |
6087 | * Now we need to update our outstanding extents and csum bytes _first_ | |
6088 | * and then add the reservation to the block_rsv. This keeps us from | |
6089 | * racing with an ordered completion or some such that would think it | |
6090 | * needs to free the reservation we just made. | |
6091 | */ | |
9f3db423 | 6092 | spin_lock(&inode->lock); |
69fe2d75 | 6093 | nr_extents = count_max_extents(num_bytes); |
8b62f87b | 6094 | btrfs_mod_outstanding_extents(inode, nr_extents); |
69fe2d75 JB |
6095 | inode->csum_bytes += num_bytes; |
6096 | btrfs_calculate_inode_block_rsv_size(fs_info, inode); | |
9f3db423 | 6097 | spin_unlock(&inode->lock); |
57a45ced | 6098 | |
c8eaeac7 JB |
6099 | /* Now we can safely add our space to our block rsv */ |
6100 | block_rsv_add_bytes(block_rsv, meta_reserve, false); | |
6101 | trace_btrfs_space_reservation(root->fs_info, "delalloc", | |
6102 | btrfs_ino(inode), meta_reserve, 1); | |
6103 | ||
6104 | spin_lock(&block_rsv->lock); | |
6105 | block_rsv->qgroup_rsv_reserved += qgroup_reserve; | |
6106 | spin_unlock(&block_rsv->lock); | |
25179201 | 6107 | |
c64c2bd8 | 6108 | if (delalloc_lock) |
9f3db423 | 6109 | mutex_unlock(&inode->delalloc_mutex); |
0ca1f7ce | 6110 | return 0; |
c8eaeac7 JB |
6111 | out_qgroup: |
6112 | btrfs_qgroup_free_meta_prealloc(root, qgroup_reserve); | |
88e081bf | 6113 | out_fail: |
43b18595 | 6114 | btrfs_inode_rsv_release(inode, true); |
88e081bf | 6115 | if (delalloc_lock) |
9f3db423 | 6116 | mutex_unlock(&inode->delalloc_mutex); |
88e081bf | 6117 | return ret; |
0ca1f7ce YZ |
6118 | } |
6119 | ||
7709cde3 JB |
6120 | /** |
6121 | * btrfs_delalloc_release_metadata - release a metadata reservation for an inode | |
8b62f87b JB |
6122 | * @inode: the inode to release the reservation for. |
6123 | * @num_bytes: the number of bytes we are releasing. | |
43b18595 | 6124 | * @qgroup_free: free qgroup reservation or convert it to per-trans reservation |
7709cde3 JB |
6125 | * |
6126 | * This will release the metadata reservation for an inode. This can be called | |
6127 | * once we complete IO for a given set of bytes to release their metadata | |
8b62f87b | 6128 | * reservations, or on error for the same reason. |
7709cde3 | 6129 | */ |
43b18595 QW |
6130 | void btrfs_delalloc_release_metadata(struct btrfs_inode *inode, u64 num_bytes, |
6131 | bool qgroup_free) | |
0ca1f7ce | 6132 | { |
3ffbd68c | 6133 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
0ca1f7ce | 6134 | |
0b246afa | 6135 | num_bytes = ALIGN(num_bytes, fs_info->sectorsize); |
691fa059 | 6136 | spin_lock(&inode->lock); |
69fe2d75 JB |
6137 | inode->csum_bytes -= num_bytes; |
6138 | btrfs_calculate_inode_block_rsv_size(fs_info, inode); | |
691fa059 | 6139 | spin_unlock(&inode->lock); |
0ca1f7ce | 6140 | |
0b246afa | 6141 | if (btrfs_is_testing(fs_info)) |
6a3891c5 JB |
6142 | return; |
6143 | ||
43b18595 | 6144 | btrfs_inode_rsv_release(inode, qgroup_free); |
0ca1f7ce YZ |
6145 | } |
6146 | ||
8b62f87b JB |
6147 | /** |
6148 | * btrfs_delalloc_release_extents - release our outstanding_extents | |
6149 | * @inode: the inode to balance the reservation for. | |
6150 | * @num_bytes: the number of bytes we originally reserved with | |
43b18595 | 6151 | * @qgroup_free: do we need to free qgroup meta reservation or convert them. |
8b62f87b JB |
6152 | * |
6153 | * When we reserve space we increase outstanding_extents for the extents we may | |
6154 | * add. Once we've set the range as delalloc or created our ordered extents we | |
6155 | * have outstanding_extents to track the real usage, so we use this to free our | |
6156 | * temporarily tracked outstanding_extents. This _must_ be used in conjunction | |
6157 | * with btrfs_delalloc_reserve_metadata. | |
6158 | */ | |
43b18595 QW |
6159 | void btrfs_delalloc_release_extents(struct btrfs_inode *inode, u64 num_bytes, |
6160 | bool qgroup_free) | |
8b62f87b | 6161 | { |
3ffbd68c | 6162 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
8b62f87b | 6163 | unsigned num_extents; |
8b62f87b JB |
6164 | |
6165 | spin_lock(&inode->lock); | |
6166 | num_extents = count_max_extents(num_bytes); | |
6167 | btrfs_mod_outstanding_extents(inode, -num_extents); | |
69fe2d75 | 6168 | btrfs_calculate_inode_block_rsv_size(fs_info, inode); |
8b62f87b JB |
6169 | spin_unlock(&inode->lock); |
6170 | ||
8b62f87b JB |
6171 | if (btrfs_is_testing(fs_info)) |
6172 | return; | |
6173 | ||
43b18595 | 6174 | btrfs_inode_rsv_release(inode, qgroup_free); |
8b62f87b JB |
6175 | } |
6176 | ||
1ada3a62 | 6177 | /** |
7cf5b976 | 6178 | * btrfs_delalloc_reserve_space - reserve data and metadata space for |
1ada3a62 QW |
6179 | * delalloc |
6180 | * @inode: inode we're writing to | |
6181 | * @start: start range we are writing to | |
6182 | * @len: how long the range we are writing to | |
364ecf36 QW |
6183 | * @reserved: mandatory parameter, record actually reserved qgroup ranges of |
6184 | * current reservation. | |
1ada3a62 | 6185 | * |
1ada3a62 QW |
6186 | * This will do the following things |
6187 | * | |
6188 | * o reserve space in data space info for num bytes | |
6189 | * and reserve precious corresponding qgroup space | |
6190 | * (Done in check_data_free_space) | |
6191 | * | |
6192 | * o reserve space for metadata space, based on the number of outstanding | |
6193 | * extents and how much csums will be needed | |
6194 | * also reserve metadata space in a per root over-reserve method. | |
6195 | * o add to the inodes->delalloc_bytes | |
6196 | * o add it to the fs_info's delalloc inodes list. | |
6197 | * (Above 3 all done in delalloc_reserve_metadata) | |
6198 | * | |
6199 | * Return 0 for success | |
6200 | * Return <0 for error(-ENOSPC or -EQUOT) | |
6201 | */ | |
364ecf36 QW |
6202 | int btrfs_delalloc_reserve_space(struct inode *inode, |
6203 | struct extent_changeset **reserved, u64 start, u64 len) | |
1ada3a62 QW |
6204 | { |
6205 | int ret; | |
6206 | ||
364ecf36 | 6207 | ret = btrfs_check_data_free_space(inode, reserved, start, len); |
1ada3a62 QW |
6208 | if (ret < 0) |
6209 | return ret; | |
9f3db423 | 6210 | ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len); |
1ada3a62 | 6211 | if (ret < 0) |
bc42bda2 | 6212 | btrfs_free_reserved_data_space(inode, *reserved, start, len); |
1ada3a62 QW |
6213 | return ret; |
6214 | } | |
6215 | ||
7709cde3 | 6216 | /** |
7cf5b976 | 6217 | * btrfs_delalloc_release_space - release data and metadata space for delalloc |
1ada3a62 QW |
6218 | * @inode: inode we're releasing space for |
6219 | * @start: start position of the space already reserved | |
6220 | * @len: the len of the space already reserved | |
8b62f87b | 6221 | * @release_bytes: the len of the space we consumed or didn't use |
1ada3a62 QW |
6222 | * |
6223 | * This function will release the metadata space that was not used and will | |
6224 | * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes | |
6225 | * list if there are no delalloc bytes left. | |
6226 | * Also it will handle the qgroup reserved space. | |
6227 | */ | |
bc42bda2 | 6228 | void btrfs_delalloc_release_space(struct inode *inode, |
8b62f87b | 6229 | struct extent_changeset *reserved, |
43b18595 | 6230 | u64 start, u64 len, bool qgroup_free) |
1ada3a62 | 6231 | { |
43b18595 | 6232 | btrfs_delalloc_release_metadata(BTRFS_I(inode), len, qgroup_free); |
bc42bda2 | 6233 | btrfs_free_reserved_data_space(inode, reserved, start, len); |
6324fbf3 CM |
6234 | } |
6235 | ||
ce93ec54 | 6236 | static int update_block_group(struct btrfs_trans_handle *trans, |
6b279408 | 6237 | u64 bytenr, u64 num_bytes, int alloc) |
9078a3e1 | 6238 | { |
6b279408 | 6239 | struct btrfs_fs_info *info = trans->fs_info; |
0af3d00b | 6240 | struct btrfs_block_group_cache *cache = NULL; |
db94535d | 6241 | u64 total = num_bytes; |
9078a3e1 | 6242 | u64 old_val; |
db94535d | 6243 | u64 byte_in_group; |
0af3d00b | 6244 | int factor; |
ba2c4d4e | 6245 | int ret = 0; |
3e1ad54f | 6246 | |
5d4f98a2 | 6247 | /* block accounting for super block */ |
eb73c1b7 | 6248 | spin_lock(&info->delalloc_root_lock); |
6c41761f | 6249 | old_val = btrfs_super_bytes_used(info->super_copy); |
5d4f98a2 YZ |
6250 | if (alloc) |
6251 | old_val += num_bytes; | |
6252 | else | |
6253 | old_val -= num_bytes; | |
6c41761f | 6254 | btrfs_set_super_bytes_used(info->super_copy, old_val); |
eb73c1b7 | 6255 | spin_unlock(&info->delalloc_root_lock); |
5d4f98a2 | 6256 | |
d397712b | 6257 | while (total) { |
db94535d | 6258 | cache = btrfs_lookup_block_group(info, bytenr); |
ba2c4d4e JB |
6259 | if (!cache) { |
6260 | ret = -ENOENT; | |
6261 | break; | |
6262 | } | |
46df06b8 DS |
6263 | factor = btrfs_bg_type_to_factor(cache->flags); |
6264 | ||
9d66e233 JB |
6265 | /* |
6266 | * If this block group has free space cache written out, we | |
6267 | * need to make sure to load it if we are removing space. This | |
6268 | * is because we need the unpinning stage to actually add the | |
6269 | * space back to the block group, otherwise we will leak space. | |
6270 | */ | |
6271 | if (!alloc && cache->cached == BTRFS_CACHE_NO) | |
f6373bf3 | 6272 | cache_block_group(cache, 1); |
0af3d00b | 6273 | |
db94535d CM |
6274 | byte_in_group = bytenr - cache->key.objectid; |
6275 | WARN_ON(byte_in_group > cache->key.offset); | |
9078a3e1 | 6276 | |
25179201 | 6277 | spin_lock(&cache->space_info->lock); |
c286ac48 | 6278 | spin_lock(&cache->lock); |
0af3d00b | 6279 | |
6202df69 | 6280 | if (btrfs_test_opt(info, SPACE_CACHE) && |
0af3d00b JB |
6281 | cache->disk_cache_state < BTRFS_DC_CLEAR) |
6282 | cache->disk_cache_state = BTRFS_DC_CLEAR; | |
6283 | ||
9078a3e1 | 6284 | old_val = btrfs_block_group_used(&cache->item); |
db94535d | 6285 | num_bytes = min(total, cache->key.offset - byte_in_group); |
cd1bc465 | 6286 | if (alloc) { |
db94535d | 6287 | old_val += num_bytes; |
11833d66 YZ |
6288 | btrfs_set_block_group_used(&cache->item, old_val); |
6289 | cache->reserved -= num_bytes; | |
11833d66 | 6290 | cache->space_info->bytes_reserved -= num_bytes; |
b742bb82 YZ |
6291 | cache->space_info->bytes_used += num_bytes; |
6292 | cache->space_info->disk_used += num_bytes * factor; | |
c286ac48 | 6293 | spin_unlock(&cache->lock); |
25179201 | 6294 | spin_unlock(&cache->space_info->lock); |
cd1bc465 | 6295 | } else { |
db94535d | 6296 | old_val -= num_bytes; |
ae0ab003 FM |
6297 | btrfs_set_block_group_used(&cache->item, old_val); |
6298 | cache->pinned += num_bytes; | |
e2907c1a | 6299 | update_bytes_pinned(cache->space_info, num_bytes); |
ae0ab003 FM |
6300 | cache->space_info->bytes_used -= num_bytes; |
6301 | cache->space_info->disk_used -= num_bytes * factor; | |
6302 | spin_unlock(&cache->lock); | |
6303 | spin_unlock(&cache->space_info->lock); | |
47ab2a6c | 6304 | |
0b246afa | 6305 | trace_btrfs_space_reservation(info, "pinned", |
c51e7bb1 JB |
6306 | cache->space_info->flags, |
6307 | num_bytes, 1); | |
dec59fa3 EL |
6308 | percpu_counter_add_batch(&cache->space_info->total_bytes_pinned, |
6309 | num_bytes, | |
6310 | BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
ae0ab003 FM |
6311 | set_extent_dirty(info->pinned_extents, |
6312 | bytenr, bytenr + num_bytes - 1, | |
6313 | GFP_NOFS | __GFP_NOFAIL); | |
cd1bc465 | 6314 | } |
1bbc621e CM |
6315 | |
6316 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
6317 | if (list_empty(&cache->dirty_list)) { | |
6318 | list_add_tail(&cache->dirty_list, | |
6319 | &trans->transaction->dirty_bgs); | |
ba2c4d4e | 6320 | trans->delayed_ref_updates++; |
1bbc621e CM |
6321 | btrfs_get_block_group(cache); |
6322 | } | |
6323 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
6324 | ||
036a9348 FM |
6325 | /* |
6326 | * No longer have used bytes in this block group, queue it for | |
6327 | * deletion. We do this after adding the block group to the | |
6328 | * dirty list to avoid races between cleaner kthread and space | |
6329 | * cache writeout. | |
6330 | */ | |
031f24da QW |
6331 | if (!alloc && old_val == 0) |
6332 | btrfs_mark_bg_unused(cache); | |
036a9348 | 6333 | |
fa9c0d79 | 6334 | btrfs_put_block_group(cache); |
db94535d CM |
6335 | total -= num_bytes; |
6336 | bytenr += num_bytes; | |
9078a3e1 | 6337 | } |
ba2c4d4e JB |
6338 | |
6339 | /* Modified block groups are accounted for in the delayed_refs_rsv. */ | |
6340 | btrfs_update_delayed_refs_rsv(trans); | |
6341 | return ret; | |
9078a3e1 | 6342 | } |
6324fbf3 | 6343 | |
2ff7e61e | 6344 | static u64 first_logical_byte(struct btrfs_fs_info *fs_info, u64 search_start) |
a061fc8d | 6345 | { |
0f9dd46c | 6346 | struct btrfs_block_group_cache *cache; |
d2fb3437 | 6347 | u64 bytenr; |
0f9dd46c | 6348 | |
0b246afa JM |
6349 | spin_lock(&fs_info->block_group_cache_lock); |
6350 | bytenr = fs_info->first_logical_byte; | |
6351 | spin_unlock(&fs_info->block_group_cache_lock); | |
a1897fdd LB |
6352 | |
6353 | if (bytenr < (u64)-1) | |
6354 | return bytenr; | |
6355 | ||
0b246afa | 6356 | cache = btrfs_lookup_first_block_group(fs_info, search_start); |
0f9dd46c | 6357 | if (!cache) |
a061fc8d | 6358 | return 0; |
0f9dd46c | 6359 | |
d2fb3437 | 6360 | bytenr = cache->key.objectid; |
fa9c0d79 | 6361 | btrfs_put_block_group(cache); |
d2fb3437 YZ |
6362 | |
6363 | return bytenr; | |
a061fc8d CM |
6364 | } |
6365 | ||
fdf08605 | 6366 | static int pin_down_extent(struct btrfs_block_group_cache *cache, |
f0486c68 | 6367 | u64 bytenr, u64 num_bytes, int reserved) |
324ae4df | 6368 | { |
fdf08605 DS |
6369 | struct btrfs_fs_info *fs_info = cache->fs_info; |
6370 | ||
11833d66 YZ |
6371 | spin_lock(&cache->space_info->lock); |
6372 | spin_lock(&cache->lock); | |
6373 | cache->pinned += num_bytes; | |
e2907c1a | 6374 | update_bytes_pinned(cache->space_info, num_bytes); |
11833d66 YZ |
6375 | if (reserved) { |
6376 | cache->reserved -= num_bytes; | |
6377 | cache->space_info->bytes_reserved -= num_bytes; | |
6378 | } | |
6379 | spin_unlock(&cache->lock); | |
6380 | spin_unlock(&cache->space_info->lock); | |
68b38550 | 6381 | |
0b246afa | 6382 | trace_btrfs_space_reservation(fs_info, "pinned", |
c51e7bb1 | 6383 | cache->space_info->flags, num_bytes, 1); |
dec59fa3 EL |
6384 | percpu_counter_add_batch(&cache->space_info->total_bytes_pinned, |
6385 | num_bytes, BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
0b246afa | 6386 | set_extent_dirty(fs_info->pinned_extents, bytenr, |
f0486c68 YZ |
6387 | bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL); |
6388 | return 0; | |
6389 | } | |
68b38550 | 6390 | |
f0486c68 YZ |
6391 | /* |
6392 | * this function must be called within transaction | |
6393 | */ | |
2ff7e61e | 6394 | int btrfs_pin_extent(struct btrfs_fs_info *fs_info, |
f0486c68 YZ |
6395 | u64 bytenr, u64 num_bytes, int reserved) |
6396 | { | |
6397 | struct btrfs_block_group_cache *cache; | |
68b38550 | 6398 | |
0b246afa | 6399 | cache = btrfs_lookup_block_group(fs_info, bytenr); |
79787eaa | 6400 | BUG_ON(!cache); /* Logic error */ |
f0486c68 | 6401 | |
fdf08605 | 6402 | pin_down_extent(cache, bytenr, num_bytes, reserved); |
f0486c68 YZ |
6403 | |
6404 | btrfs_put_block_group(cache); | |
11833d66 YZ |
6405 | return 0; |
6406 | } | |
6407 | ||
f0486c68 | 6408 | /* |
e688b725 CM |
6409 | * this function must be called within transaction |
6410 | */ | |
2ff7e61e | 6411 | int btrfs_pin_extent_for_log_replay(struct btrfs_fs_info *fs_info, |
e688b725 CM |
6412 | u64 bytenr, u64 num_bytes) |
6413 | { | |
6414 | struct btrfs_block_group_cache *cache; | |
b50c6e25 | 6415 | int ret; |
e688b725 | 6416 | |
0b246afa | 6417 | cache = btrfs_lookup_block_group(fs_info, bytenr); |
b50c6e25 JB |
6418 | if (!cache) |
6419 | return -EINVAL; | |
e688b725 CM |
6420 | |
6421 | /* | |
6422 | * pull in the free space cache (if any) so that our pin | |
6423 | * removes the free space from the cache. We have load_only set | |
6424 | * to one because the slow code to read in the free extents does check | |
6425 | * the pinned extents. | |
6426 | */ | |
f6373bf3 | 6427 | cache_block_group(cache, 1); |
e688b725 | 6428 | |
fdf08605 | 6429 | pin_down_extent(cache, bytenr, num_bytes, 0); |
e688b725 CM |
6430 | |
6431 | /* remove us from the free space cache (if we're there at all) */ | |
b50c6e25 | 6432 | ret = btrfs_remove_free_space(cache, bytenr, num_bytes); |
e688b725 | 6433 | btrfs_put_block_group(cache); |
b50c6e25 | 6434 | return ret; |
e688b725 CM |
6435 | } |
6436 | ||
2ff7e61e JM |
6437 | static int __exclude_logged_extent(struct btrfs_fs_info *fs_info, |
6438 | u64 start, u64 num_bytes) | |
8c2a1a30 JB |
6439 | { |
6440 | int ret; | |
6441 | struct btrfs_block_group_cache *block_group; | |
6442 | struct btrfs_caching_control *caching_ctl; | |
6443 | ||
0b246afa | 6444 | block_group = btrfs_lookup_block_group(fs_info, start); |
8c2a1a30 JB |
6445 | if (!block_group) |
6446 | return -EINVAL; | |
6447 | ||
6448 | cache_block_group(block_group, 0); | |
6449 | caching_ctl = get_caching_control(block_group); | |
6450 | ||
6451 | if (!caching_ctl) { | |
6452 | /* Logic error */ | |
6453 | BUG_ON(!block_group_cache_done(block_group)); | |
6454 | ret = btrfs_remove_free_space(block_group, start, num_bytes); | |
6455 | } else { | |
6456 | mutex_lock(&caching_ctl->mutex); | |
6457 | ||
6458 | if (start >= caching_ctl->progress) { | |
2ff7e61e | 6459 | ret = add_excluded_extent(fs_info, start, num_bytes); |
8c2a1a30 JB |
6460 | } else if (start + num_bytes <= caching_ctl->progress) { |
6461 | ret = btrfs_remove_free_space(block_group, | |
6462 | start, num_bytes); | |
6463 | } else { | |
6464 | num_bytes = caching_ctl->progress - start; | |
6465 | ret = btrfs_remove_free_space(block_group, | |
6466 | start, num_bytes); | |
6467 | if (ret) | |
6468 | goto out_lock; | |
6469 | ||
6470 | num_bytes = (start + num_bytes) - | |
6471 | caching_ctl->progress; | |
6472 | start = caching_ctl->progress; | |
2ff7e61e | 6473 | ret = add_excluded_extent(fs_info, start, num_bytes); |
8c2a1a30 JB |
6474 | } |
6475 | out_lock: | |
6476 | mutex_unlock(&caching_ctl->mutex); | |
6477 | put_caching_control(caching_ctl); | |
6478 | } | |
6479 | btrfs_put_block_group(block_group); | |
6480 | return ret; | |
6481 | } | |
6482 | ||
bcdc428c | 6483 | int btrfs_exclude_logged_extents(struct extent_buffer *eb) |
8c2a1a30 | 6484 | { |
bcdc428c | 6485 | struct btrfs_fs_info *fs_info = eb->fs_info; |
8c2a1a30 JB |
6486 | struct btrfs_file_extent_item *item; |
6487 | struct btrfs_key key; | |
6488 | int found_type; | |
6489 | int i; | |
b89311ef | 6490 | int ret = 0; |
8c2a1a30 | 6491 | |
2ff7e61e | 6492 | if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) |
8c2a1a30 JB |
6493 | return 0; |
6494 | ||
6495 | for (i = 0; i < btrfs_header_nritems(eb); i++) { | |
6496 | btrfs_item_key_to_cpu(eb, &key, i); | |
6497 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
6498 | continue; | |
6499 | item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item); | |
6500 | found_type = btrfs_file_extent_type(eb, item); | |
6501 | if (found_type == BTRFS_FILE_EXTENT_INLINE) | |
6502 | continue; | |
6503 | if (btrfs_file_extent_disk_bytenr(eb, item) == 0) | |
6504 | continue; | |
6505 | key.objectid = btrfs_file_extent_disk_bytenr(eb, item); | |
6506 | key.offset = btrfs_file_extent_disk_num_bytes(eb, item); | |
b89311ef GJ |
6507 | ret = __exclude_logged_extent(fs_info, key.objectid, key.offset); |
6508 | if (ret) | |
6509 | break; | |
8c2a1a30 JB |
6510 | } |
6511 | ||
b89311ef | 6512 | return ret; |
8c2a1a30 JB |
6513 | } |
6514 | ||
9cfa3e34 FM |
6515 | static void |
6516 | btrfs_inc_block_group_reservations(struct btrfs_block_group_cache *bg) | |
6517 | { | |
6518 | atomic_inc(&bg->reservations); | |
6519 | } | |
6520 | ||
6521 | void btrfs_dec_block_group_reservations(struct btrfs_fs_info *fs_info, | |
6522 | const u64 start) | |
6523 | { | |
6524 | struct btrfs_block_group_cache *bg; | |
6525 | ||
6526 | bg = btrfs_lookup_block_group(fs_info, start); | |
6527 | ASSERT(bg); | |
6528 | if (atomic_dec_and_test(&bg->reservations)) | |
4625956a | 6529 | wake_up_var(&bg->reservations); |
9cfa3e34 FM |
6530 | btrfs_put_block_group(bg); |
6531 | } | |
6532 | ||
9cfa3e34 FM |
6533 | void btrfs_wait_block_group_reservations(struct btrfs_block_group_cache *bg) |
6534 | { | |
6535 | struct btrfs_space_info *space_info = bg->space_info; | |
6536 | ||
6537 | ASSERT(bg->ro); | |
6538 | ||
6539 | if (!(bg->flags & BTRFS_BLOCK_GROUP_DATA)) | |
6540 | return; | |
6541 | ||
6542 | /* | |
6543 | * Our block group is read only but before we set it to read only, | |
6544 | * some task might have had allocated an extent from it already, but it | |
6545 | * has not yet created a respective ordered extent (and added it to a | |
6546 | * root's list of ordered extents). | |
6547 | * Therefore wait for any task currently allocating extents, since the | |
6548 | * block group's reservations counter is incremented while a read lock | |
6549 | * on the groups' semaphore is held and decremented after releasing | |
6550 | * the read access on that semaphore and creating the ordered extent. | |
6551 | */ | |
6552 | down_write(&space_info->groups_sem); | |
6553 | up_write(&space_info->groups_sem); | |
6554 | ||
4625956a | 6555 | wait_var_event(&bg->reservations, !atomic_read(&bg->reservations)); |
9cfa3e34 FM |
6556 | } |
6557 | ||
fb25e914 | 6558 | /** |
4824f1f4 | 6559 | * btrfs_add_reserved_bytes - update the block_group and space info counters |
fb25e914 | 6560 | * @cache: The cache we are manipulating |
18513091 WX |
6561 | * @ram_bytes: The number of bytes of file content, and will be same to |
6562 | * @num_bytes except for the compress path. | |
fb25e914 | 6563 | * @num_bytes: The number of bytes in question |
e570fd27 | 6564 | * @delalloc: The blocks are allocated for the delalloc write |
fb25e914 | 6565 | * |
745699ef XW |
6566 | * This is called by the allocator when it reserves space. If this is a |
6567 | * reservation and the block group has become read only we cannot make the | |
6568 | * reservation and return -EAGAIN, otherwise this function always succeeds. | |
f0486c68 | 6569 | */ |
4824f1f4 | 6570 | static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache *cache, |
18513091 | 6571 | u64 ram_bytes, u64 num_bytes, int delalloc) |
11833d66 | 6572 | { |
fb25e914 | 6573 | struct btrfs_space_info *space_info = cache->space_info; |
f0486c68 | 6574 | int ret = 0; |
79787eaa | 6575 | |
fb25e914 JB |
6576 | spin_lock(&space_info->lock); |
6577 | spin_lock(&cache->lock); | |
4824f1f4 WX |
6578 | if (cache->ro) { |
6579 | ret = -EAGAIN; | |
fb25e914 | 6580 | } else { |
4824f1f4 WX |
6581 | cache->reserved += num_bytes; |
6582 | space_info->bytes_reserved += num_bytes; | |
9f9b8e8d | 6583 | update_bytes_may_use(space_info, -ram_bytes); |
e570fd27 | 6584 | if (delalloc) |
4824f1f4 | 6585 | cache->delalloc_bytes += num_bytes; |
324ae4df | 6586 | } |
fb25e914 JB |
6587 | spin_unlock(&cache->lock); |
6588 | spin_unlock(&space_info->lock); | |
f0486c68 | 6589 | return ret; |
324ae4df | 6590 | } |
9078a3e1 | 6591 | |
4824f1f4 WX |
6592 | /** |
6593 | * btrfs_free_reserved_bytes - update the block_group and space info counters | |
6594 | * @cache: The cache we are manipulating | |
6595 | * @num_bytes: The number of bytes in question | |
6596 | * @delalloc: The blocks are allocated for the delalloc write | |
6597 | * | |
6598 | * This is called by somebody who is freeing space that was never actually used | |
6599 | * on disk. For example if you reserve some space for a new leaf in transaction | |
6600 | * A and before transaction A commits you free that leaf, you call this with | |
6601 | * reserve set to 0 in order to clear the reservation. | |
6602 | */ | |
6603 | ||
556f3ca8 | 6604 | static void btrfs_free_reserved_bytes(struct btrfs_block_group_cache *cache, |
6605 | u64 num_bytes, int delalloc) | |
4824f1f4 WX |
6606 | { |
6607 | struct btrfs_space_info *space_info = cache->space_info; | |
4824f1f4 WX |
6608 | |
6609 | spin_lock(&space_info->lock); | |
6610 | spin_lock(&cache->lock); | |
6611 | if (cache->ro) | |
6612 | space_info->bytes_readonly += num_bytes; | |
6613 | cache->reserved -= num_bytes; | |
6614 | space_info->bytes_reserved -= num_bytes; | |
21a94f7a | 6615 | space_info->max_extent_size = 0; |
4824f1f4 WX |
6616 | |
6617 | if (delalloc) | |
6618 | cache->delalloc_bytes -= num_bytes; | |
6619 | spin_unlock(&cache->lock); | |
6620 | spin_unlock(&space_info->lock); | |
4824f1f4 | 6621 | } |
8b74c03e | 6622 | void btrfs_prepare_extent_commit(struct btrfs_fs_info *fs_info) |
e8569813 | 6623 | { |
11833d66 YZ |
6624 | struct btrfs_caching_control *next; |
6625 | struct btrfs_caching_control *caching_ctl; | |
6626 | struct btrfs_block_group_cache *cache; | |
e8569813 | 6627 | |
9e351cc8 | 6628 | down_write(&fs_info->commit_root_sem); |
25179201 | 6629 | |
11833d66 YZ |
6630 | list_for_each_entry_safe(caching_ctl, next, |
6631 | &fs_info->caching_block_groups, list) { | |
6632 | cache = caching_ctl->block_group; | |
6633 | if (block_group_cache_done(cache)) { | |
6634 | cache->last_byte_to_unpin = (u64)-1; | |
6635 | list_del_init(&caching_ctl->list); | |
6636 | put_caching_control(caching_ctl); | |
e8569813 | 6637 | } else { |
11833d66 | 6638 | cache->last_byte_to_unpin = caching_ctl->progress; |
e8569813 | 6639 | } |
e8569813 | 6640 | } |
11833d66 YZ |
6641 | |
6642 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) | |
6643 | fs_info->pinned_extents = &fs_info->freed_extents[1]; | |
6644 | else | |
6645 | fs_info->pinned_extents = &fs_info->freed_extents[0]; | |
6646 | ||
9e351cc8 | 6647 | up_write(&fs_info->commit_root_sem); |
8929ecfa YZ |
6648 | |
6649 | update_global_block_rsv(fs_info); | |
e8569813 ZY |
6650 | } |
6651 | ||
c759c4e1 JB |
6652 | /* |
6653 | * Returns the free cluster for the given space info and sets empty_cluster to | |
6654 | * what it should be based on the mount options. | |
6655 | */ | |
6656 | static struct btrfs_free_cluster * | |
2ff7e61e JM |
6657 | fetch_cluster_info(struct btrfs_fs_info *fs_info, |
6658 | struct btrfs_space_info *space_info, u64 *empty_cluster) | |
c759c4e1 JB |
6659 | { |
6660 | struct btrfs_free_cluster *ret = NULL; | |
c759c4e1 JB |
6661 | |
6662 | *empty_cluster = 0; | |
6663 | if (btrfs_mixed_space_info(space_info)) | |
6664 | return ret; | |
6665 | ||
c759c4e1 | 6666 | if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) { |
0b246afa | 6667 | ret = &fs_info->meta_alloc_cluster; |
583b7231 HK |
6668 | if (btrfs_test_opt(fs_info, SSD)) |
6669 | *empty_cluster = SZ_2M; | |
6670 | else | |
ee22184b | 6671 | *empty_cluster = SZ_64K; |
583b7231 HK |
6672 | } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) && |
6673 | btrfs_test_opt(fs_info, SSD_SPREAD)) { | |
6674 | *empty_cluster = SZ_2M; | |
0b246afa | 6675 | ret = &fs_info->data_alloc_cluster; |
c759c4e1 JB |
6676 | } |
6677 | ||
6678 | return ret; | |
6679 | } | |
6680 | ||
2ff7e61e JM |
6681 | static int unpin_extent_range(struct btrfs_fs_info *fs_info, |
6682 | u64 start, u64 end, | |
678886bd | 6683 | const bool return_free_space) |
ccd467d6 | 6684 | { |
11833d66 | 6685 | struct btrfs_block_group_cache *cache = NULL; |
7b398f8e JB |
6686 | struct btrfs_space_info *space_info; |
6687 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
c759c4e1 | 6688 | struct btrfs_free_cluster *cluster = NULL; |
11833d66 | 6689 | u64 len; |
c759c4e1 JB |
6690 | u64 total_unpinned = 0; |
6691 | u64 empty_cluster = 0; | |
7b398f8e | 6692 | bool readonly; |
ccd467d6 | 6693 | |
11833d66 | 6694 | while (start <= end) { |
7b398f8e | 6695 | readonly = false; |
11833d66 YZ |
6696 | if (!cache || |
6697 | start >= cache->key.objectid + cache->key.offset) { | |
6698 | if (cache) | |
6699 | btrfs_put_block_group(cache); | |
c759c4e1 | 6700 | total_unpinned = 0; |
11833d66 | 6701 | cache = btrfs_lookup_block_group(fs_info, start); |
79787eaa | 6702 | BUG_ON(!cache); /* Logic error */ |
c759c4e1 | 6703 | |
2ff7e61e | 6704 | cluster = fetch_cluster_info(fs_info, |
c759c4e1 JB |
6705 | cache->space_info, |
6706 | &empty_cluster); | |
6707 | empty_cluster <<= 1; | |
11833d66 YZ |
6708 | } |
6709 | ||
6710 | len = cache->key.objectid + cache->key.offset - start; | |
6711 | len = min(len, end + 1 - start); | |
6712 | ||
6713 | if (start < cache->last_byte_to_unpin) { | |
6714 | len = min(len, cache->last_byte_to_unpin - start); | |
678886bd FM |
6715 | if (return_free_space) |
6716 | btrfs_add_free_space(cache, start, len); | |
11833d66 YZ |
6717 | } |
6718 | ||
f0486c68 | 6719 | start += len; |
c759c4e1 | 6720 | total_unpinned += len; |
7b398f8e | 6721 | space_info = cache->space_info; |
f0486c68 | 6722 | |
c759c4e1 JB |
6723 | /* |
6724 | * If this space cluster has been marked as fragmented and we've | |
6725 | * unpinned enough in this block group to potentially allow a | |
6726 | * cluster to be created inside of it go ahead and clear the | |
6727 | * fragmented check. | |
6728 | */ | |
6729 | if (cluster && cluster->fragmented && | |
6730 | total_unpinned > empty_cluster) { | |
6731 | spin_lock(&cluster->lock); | |
6732 | cluster->fragmented = 0; | |
6733 | spin_unlock(&cluster->lock); | |
6734 | } | |
6735 | ||
7b398f8e | 6736 | spin_lock(&space_info->lock); |
11833d66 YZ |
6737 | spin_lock(&cache->lock); |
6738 | cache->pinned -= len; | |
e2907c1a | 6739 | update_bytes_pinned(space_info, -len); |
c51e7bb1 JB |
6740 | |
6741 | trace_btrfs_space_reservation(fs_info, "pinned", | |
6742 | space_info->flags, len, 0); | |
4f4db217 | 6743 | space_info->max_extent_size = 0; |
dec59fa3 EL |
6744 | percpu_counter_add_batch(&space_info->total_bytes_pinned, |
6745 | -len, BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
7b398f8e JB |
6746 | if (cache->ro) { |
6747 | space_info->bytes_readonly += len; | |
6748 | readonly = true; | |
6749 | } | |
11833d66 | 6750 | spin_unlock(&cache->lock); |
957780eb JB |
6751 | if (!readonly && return_free_space && |
6752 | global_rsv->space_info == space_info) { | |
6753 | u64 to_add = len; | |
92ac58ec | 6754 | |
7b398f8e JB |
6755 | spin_lock(&global_rsv->lock); |
6756 | if (!global_rsv->full) { | |
957780eb JB |
6757 | to_add = min(len, global_rsv->size - |
6758 | global_rsv->reserved); | |
6759 | global_rsv->reserved += to_add; | |
9f9b8e8d | 6760 | update_bytes_may_use(space_info, to_add); |
7b398f8e JB |
6761 | if (global_rsv->reserved >= global_rsv->size) |
6762 | global_rsv->full = 1; | |
957780eb JB |
6763 | trace_btrfs_space_reservation(fs_info, |
6764 | "space_info", | |
6765 | space_info->flags, | |
6766 | to_add, 1); | |
6767 | len -= to_add; | |
7b398f8e JB |
6768 | } |
6769 | spin_unlock(&global_rsv->lock); | |
957780eb JB |
6770 | /* Add to any tickets we may have */ |
6771 | if (len) | |
6772 | space_info_add_new_bytes(fs_info, space_info, | |
6773 | len); | |
7b398f8e JB |
6774 | } |
6775 | spin_unlock(&space_info->lock); | |
ccd467d6 | 6776 | } |
11833d66 YZ |
6777 | |
6778 | if (cache) | |
6779 | btrfs_put_block_group(cache); | |
ccd467d6 CM |
6780 | return 0; |
6781 | } | |
6782 | ||
5ead2dd0 | 6783 | int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans) |
a28ec197 | 6784 | { |
5ead2dd0 | 6785 | struct btrfs_fs_info *fs_info = trans->fs_info; |
e33e17ee JM |
6786 | struct btrfs_block_group_cache *block_group, *tmp; |
6787 | struct list_head *deleted_bgs; | |
11833d66 | 6788 | struct extent_io_tree *unpin; |
1a5bc167 CM |
6789 | u64 start; |
6790 | u64 end; | |
a28ec197 | 6791 | int ret; |
a28ec197 | 6792 | |
11833d66 YZ |
6793 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) |
6794 | unpin = &fs_info->freed_extents[1]; | |
6795 | else | |
6796 | unpin = &fs_info->freed_extents[0]; | |
6797 | ||
e33e17ee | 6798 | while (!trans->aborted) { |
0e6ec385 FM |
6799 | struct extent_state *cached_state = NULL; |
6800 | ||
d4b450cd | 6801 | mutex_lock(&fs_info->unused_bg_unpin_mutex); |
1a5bc167 | 6802 | ret = find_first_extent_bit(unpin, 0, &start, &end, |
0e6ec385 | 6803 | EXTENT_DIRTY, &cached_state); |
d4b450cd FM |
6804 | if (ret) { |
6805 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); | |
a28ec197 | 6806 | break; |
d4b450cd | 6807 | } |
1f3c79a2 | 6808 | |
0b246afa | 6809 | if (btrfs_test_opt(fs_info, DISCARD)) |
2ff7e61e | 6810 | ret = btrfs_discard_extent(fs_info, start, |
5378e607 | 6811 | end + 1 - start, NULL); |
1f3c79a2 | 6812 | |
0e6ec385 | 6813 | clear_extent_dirty(unpin, start, end, &cached_state); |
2ff7e61e | 6814 | unpin_extent_range(fs_info, start, end, true); |
d4b450cd | 6815 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
0e6ec385 | 6816 | free_extent_state(cached_state); |
b9473439 | 6817 | cond_resched(); |
a28ec197 | 6818 | } |
817d52f8 | 6819 | |
e33e17ee JM |
6820 | /* |
6821 | * Transaction is finished. We don't need the lock anymore. We | |
6822 | * do need to clean up the block groups in case of a transaction | |
6823 | * abort. | |
6824 | */ | |
6825 | deleted_bgs = &trans->transaction->deleted_bgs; | |
6826 | list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) { | |
6827 | u64 trimmed = 0; | |
6828 | ||
6829 | ret = -EROFS; | |
6830 | if (!trans->aborted) | |
2ff7e61e | 6831 | ret = btrfs_discard_extent(fs_info, |
e33e17ee JM |
6832 | block_group->key.objectid, |
6833 | block_group->key.offset, | |
6834 | &trimmed); | |
6835 | ||
6836 | list_del_init(&block_group->bg_list); | |
6837 | btrfs_put_block_group_trimming(block_group); | |
6838 | btrfs_put_block_group(block_group); | |
6839 | ||
6840 | if (ret) { | |
6841 | const char *errstr = btrfs_decode_error(ret); | |
6842 | btrfs_warn(fs_info, | |
913e1535 | 6843 | "discard failed while removing blockgroup: errno=%d %s", |
e33e17ee JM |
6844 | ret, errstr); |
6845 | } | |
6846 | } | |
6847 | ||
e20d96d6 CM |
6848 | return 0; |
6849 | } | |
6850 | ||
5d4f98a2 | 6851 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
e72cb923 NB |
6852 | struct btrfs_delayed_ref_node *node, u64 parent, |
6853 | u64 root_objectid, u64 owner_objectid, | |
6854 | u64 owner_offset, int refs_to_drop, | |
6855 | struct btrfs_delayed_extent_op *extent_op) | |
a28ec197 | 6856 | { |
e72cb923 | 6857 | struct btrfs_fs_info *info = trans->fs_info; |
e2fa7227 | 6858 | struct btrfs_key key; |
5d4f98a2 | 6859 | struct btrfs_path *path; |
1261ec42 | 6860 | struct btrfs_root *extent_root = info->extent_root; |
5f39d397 | 6861 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
6862 | struct btrfs_extent_item *ei; |
6863 | struct btrfs_extent_inline_ref *iref; | |
a28ec197 | 6864 | int ret; |
5d4f98a2 | 6865 | int is_data; |
952fccac CM |
6866 | int extent_slot = 0; |
6867 | int found_extent = 0; | |
6868 | int num_to_del = 1; | |
5d4f98a2 YZ |
6869 | u32 item_size; |
6870 | u64 refs; | |
c682f9b3 QW |
6871 | u64 bytenr = node->bytenr; |
6872 | u64 num_bytes = node->num_bytes; | |
fcebe456 | 6873 | int last_ref = 0; |
0b246afa | 6874 | bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA); |
037e6390 | 6875 | |
5caf2a00 | 6876 | path = btrfs_alloc_path(); |
54aa1f4d CM |
6877 | if (!path) |
6878 | return -ENOMEM; | |
5f26f772 | 6879 | |
e4058b54 | 6880 | path->reada = READA_FORWARD; |
b9473439 | 6881 | path->leave_spinning = 1; |
5d4f98a2 YZ |
6882 | |
6883 | is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID; | |
6884 | BUG_ON(!is_data && refs_to_drop != 1); | |
6885 | ||
3173a18f | 6886 | if (is_data) |
897ca819 | 6887 | skinny_metadata = false; |
3173a18f | 6888 | |
fbe4801b NB |
6889 | ret = lookup_extent_backref(trans, path, &iref, bytenr, num_bytes, |
6890 | parent, root_objectid, owner_objectid, | |
5d4f98a2 | 6891 | owner_offset); |
7bb86316 | 6892 | if (ret == 0) { |
952fccac | 6893 | extent_slot = path->slots[0]; |
5d4f98a2 YZ |
6894 | while (extent_slot >= 0) { |
6895 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
952fccac | 6896 | extent_slot); |
5d4f98a2 | 6897 | if (key.objectid != bytenr) |
952fccac | 6898 | break; |
5d4f98a2 YZ |
6899 | if (key.type == BTRFS_EXTENT_ITEM_KEY && |
6900 | key.offset == num_bytes) { | |
952fccac CM |
6901 | found_extent = 1; |
6902 | break; | |
6903 | } | |
3173a18f JB |
6904 | if (key.type == BTRFS_METADATA_ITEM_KEY && |
6905 | key.offset == owner_objectid) { | |
6906 | found_extent = 1; | |
6907 | break; | |
6908 | } | |
952fccac CM |
6909 | if (path->slots[0] - extent_slot > 5) |
6910 | break; | |
5d4f98a2 | 6911 | extent_slot--; |
952fccac | 6912 | } |
a79865c6 | 6913 | |
31840ae1 | 6914 | if (!found_extent) { |
5d4f98a2 | 6915 | BUG_ON(iref); |
87cc7a8a | 6916 | ret = remove_extent_backref(trans, path, NULL, |
87bde3cd | 6917 | refs_to_drop, |
fcebe456 | 6918 | is_data, &last_ref); |
005d6427 | 6919 | if (ret) { |
66642832 | 6920 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
6921 | goto out; |
6922 | } | |
b3b4aa74 | 6923 | btrfs_release_path(path); |
b9473439 | 6924 | path->leave_spinning = 1; |
5d4f98a2 YZ |
6925 | |
6926 | key.objectid = bytenr; | |
6927 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
6928 | key.offset = num_bytes; | |
6929 | ||
3173a18f JB |
6930 | if (!is_data && skinny_metadata) { |
6931 | key.type = BTRFS_METADATA_ITEM_KEY; | |
6932 | key.offset = owner_objectid; | |
6933 | } | |
6934 | ||
31840ae1 ZY |
6935 | ret = btrfs_search_slot(trans, extent_root, |
6936 | &key, path, -1, 1); | |
3173a18f JB |
6937 | if (ret > 0 && skinny_metadata && path->slots[0]) { |
6938 | /* | |
6939 | * Couldn't find our skinny metadata item, | |
6940 | * see if we have ye olde extent item. | |
6941 | */ | |
6942 | path->slots[0]--; | |
6943 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
6944 | path->slots[0]); | |
6945 | if (key.objectid == bytenr && | |
6946 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
6947 | key.offset == num_bytes) | |
6948 | ret = 0; | |
6949 | } | |
6950 | ||
6951 | if (ret > 0 && skinny_metadata) { | |
6952 | skinny_metadata = false; | |
9ce49a0b | 6953 | key.objectid = bytenr; |
3173a18f JB |
6954 | key.type = BTRFS_EXTENT_ITEM_KEY; |
6955 | key.offset = num_bytes; | |
6956 | btrfs_release_path(path); | |
6957 | ret = btrfs_search_slot(trans, extent_root, | |
6958 | &key, path, -1, 1); | |
6959 | } | |
6960 | ||
f3465ca4 | 6961 | if (ret) { |
5d163e0e JM |
6962 | btrfs_err(info, |
6963 | "umm, got %d back from search, was looking for %llu", | |
6964 | ret, bytenr); | |
b783e62d | 6965 | if (ret > 0) |
a4f78750 | 6966 | btrfs_print_leaf(path->nodes[0]); |
f3465ca4 | 6967 | } |
005d6427 | 6968 | if (ret < 0) { |
66642832 | 6969 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
6970 | goto out; |
6971 | } | |
31840ae1 ZY |
6972 | extent_slot = path->slots[0]; |
6973 | } | |
fae7f21c | 6974 | } else if (WARN_ON(ret == -ENOENT)) { |
a4f78750 | 6975 | btrfs_print_leaf(path->nodes[0]); |
c2cf52eb SK |
6976 | btrfs_err(info, |
6977 | "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu", | |
c1c9ff7c GU |
6978 | bytenr, parent, root_objectid, owner_objectid, |
6979 | owner_offset); | |
66642832 | 6980 | btrfs_abort_transaction(trans, ret); |
c4a050bb | 6981 | goto out; |
79787eaa | 6982 | } else { |
66642832 | 6983 | btrfs_abort_transaction(trans, ret); |
005d6427 | 6984 | goto out; |
7bb86316 | 6985 | } |
5f39d397 CM |
6986 | |
6987 | leaf = path->nodes[0]; | |
5d4f98a2 | 6988 | item_size = btrfs_item_size_nr(leaf, extent_slot); |
6d8ff4e4 | 6989 | if (unlikely(item_size < sizeof(*ei))) { |
ba3c2b19 NB |
6990 | ret = -EINVAL; |
6991 | btrfs_print_v0_err(info); | |
6992 | btrfs_abort_transaction(trans, ret); | |
6993 | goto out; | |
6994 | } | |
952fccac | 6995 | ei = btrfs_item_ptr(leaf, extent_slot, |
123abc88 | 6996 | struct btrfs_extent_item); |
3173a18f JB |
6997 | if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID && |
6998 | key.type == BTRFS_EXTENT_ITEM_KEY) { | |
5d4f98a2 YZ |
6999 | struct btrfs_tree_block_info *bi; |
7000 | BUG_ON(item_size < sizeof(*ei) + sizeof(*bi)); | |
7001 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
7002 | WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi)); | |
7003 | } | |
56bec294 | 7004 | |
5d4f98a2 | 7005 | refs = btrfs_extent_refs(leaf, ei); |
32b02538 | 7006 | if (refs < refs_to_drop) { |
5d163e0e JM |
7007 | btrfs_err(info, |
7008 | "trying to drop %d refs but we only have %Lu for bytenr %Lu", | |
7009 | refs_to_drop, refs, bytenr); | |
32b02538 | 7010 | ret = -EINVAL; |
66642832 | 7011 | btrfs_abort_transaction(trans, ret); |
32b02538 JB |
7012 | goto out; |
7013 | } | |
56bec294 | 7014 | refs -= refs_to_drop; |
5f39d397 | 7015 | |
5d4f98a2 YZ |
7016 | if (refs > 0) { |
7017 | if (extent_op) | |
7018 | __run_delayed_extent_op(extent_op, leaf, ei); | |
7019 | /* | |
7020 | * In the case of inline back ref, reference count will | |
7021 | * be updated by remove_extent_backref | |
952fccac | 7022 | */ |
5d4f98a2 YZ |
7023 | if (iref) { |
7024 | BUG_ON(!found_extent); | |
7025 | } else { | |
7026 | btrfs_set_extent_refs(leaf, ei, refs); | |
7027 | btrfs_mark_buffer_dirty(leaf); | |
7028 | } | |
7029 | if (found_extent) { | |
87cc7a8a NB |
7030 | ret = remove_extent_backref(trans, path, iref, |
7031 | refs_to_drop, is_data, | |
7032 | &last_ref); | |
005d6427 | 7033 | if (ret) { |
66642832 | 7034 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7035 | goto out; |
7036 | } | |
952fccac | 7037 | } |
5d4f98a2 | 7038 | } else { |
5d4f98a2 YZ |
7039 | if (found_extent) { |
7040 | BUG_ON(is_data && refs_to_drop != | |
9ed0dea0 | 7041 | extent_data_ref_count(path, iref)); |
5d4f98a2 YZ |
7042 | if (iref) { |
7043 | BUG_ON(path->slots[0] != extent_slot); | |
7044 | } else { | |
7045 | BUG_ON(path->slots[0] != extent_slot + 1); | |
7046 | path->slots[0] = extent_slot; | |
7047 | num_to_del = 2; | |
7048 | } | |
78fae27e | 7049 | } |
b9473439 | 7050 | |
fcebe456 | 7051 | last_ref = 1; |
952fccac CM |
7052 | ret = btrfs_del_items(trans, extent_root, path, path->slots[0], |
7053 | num_to_del); | |
005d6427 | 7054 | if (ret) { |
66642832 | 7055 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7056 | goto out; |
7057 | } | |
b3b4aa74 | 7058 | btrfs_release_path(path); |
21af804c | 7059 | |
5d4f98a2 | 7060 | if (is_data) { |
5b4aacef | 7061 | ret = btrfs_del_csums(trans, info, bytenr, num_bytes); |
005d6427 | 7062 | if (ret) { |
66642832 | 7063 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7064 | goto out; |
7065 | } | |
459931ec CM |
7066 | } |
7067 | ||
e7355e50 | 7068 | ret = add_to_free_space_tree(trans, bytenr, num_bytes); |
1e144fb8 | 7069 | if (ret) { |
66642832 | 7070 | btrfs_abort_transaction(trans, ret); |
1e144fb8 OS |
7071 | goto out; |
7072 | } | |
7073 | ||
6b279408 | 7074 | ret = update_block_group(trans, bytenr, num_bytes, 0); |
005d6427 | 7075 | if (ret) { |
66642832 | 7076 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7077 | goto out; |
7078 | } | |
a28ec197 | 7079 | } |
fcebe456 JB |
7080 | btrfs_release_path(path); |
7081 | ||
79787eaa | 7082 | out: |
5caf2a00 | 7083 | btrfs_free_path(path); |
a28ec197 CM |
7084 | return ret; |
7085 | } | |
7086 | ||
1887be66 | 7087 | /* |
f0486c68 | 7088 | * when we free an block, it is possible (and likely) that we free the last |
1887be66 CM |
7089 | * delayed ref for that extent as well. This searches the delayed ref tree for |
7090 | * a given extent, and if there are no other delayed refs to be processed, it | |
7091 | * removes it from the tree. | |
7092 | */ | |
7093 | static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans, | |
2ff7e61e | 7094 | u64 bytenr) |
1887be66 CM |
7095 | { |
7096 | struct btrfs_delayed_ref_head *head; | |
7097 | struct btrfs_delayed_ref_root *delayed_refs; | |
f0486c68 | 7098 | int ret = 0; |
1887be66 CM |
7099 | |
7100 | delayed_refs = &trans->transaction->delayed_refs; | |
7101 | spin_lock(&delayed_refs->lock); | |
f72ad18e | 7102 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
1887be66 | 7103 | if (!head) |
cf93da7b | 7104 | goto out_delayed_unlock; |
1887be66 | 7105 | |
d7df2c79 | 7106 | spin_lock(&head->lock); |
e3d03965 | 7107 | if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root)) |
1887be66 CM |
7108 | goto out; |
7109 | ||
bedc6617 JB |
7110 | if (cleanup_extent_op(head) != NULL) |
7111 | goto out; | |
5d4f98a2 | 7112 | |
1887be66 CM |
7113 | /* |
7114 | * waiting for the lock here would deadlock. If someone else has it | |
7115 | * locked they are already in the process of dropping it anyway | |
7116 | */ | |
7117 | if (!mutex_trylock(&head->mutex)) | |
7118 | goto out; | |
7119 | ||
d7baffda | 7120 | btrfs_delete_ref_head(delayed_refs, head); |
d7df2c79 | 7121 | head->processing = 0; |
d7baffda | 7122 | |
d7df2c79 | 7123 | spin_unlock(&head->lock); |
1887be66 CM |
7124 | spin_unlock(&delayed_refs->lock); |
7125 | ||
f0486c68 YZ |
7126 | BUG_ON(head->extent_op); |
7127 | if (head->must_insert_reserved) | |
7128 | ret = 1; | |
7129 | ||
31890da0 | 7130 | btrfs_cleanup_ref_head_accounting(trans->fs_info, delayed_refs, head); |
f0486c68 | 7131 | mutex_unlock(&head->mutex); |
d278850e | 7132 | btrfs_put_delayed_ref_head(head); |
f0486c68 | 7133 | return ret; |
1887be66 | 7134 | out: |
d7df2c79 | 7135 | spin_unlock(&head->lock); |
cf93da7b CM |
7136 | |
7137 | out_delayed_unlock: | |
1887be66 CM |
7138 | spin_unlock(&delayed_refs->lock); |
7139 | return 0; | |
7140 | } | |
7141 | ||
f0486c68 YZ |
7142 | void btrfs_free_tree_block(struct btrfs_trans_handle *trans, |
7143 | struct btrfs_root *root, | |
7144 | struct extent_buffer *buf, | |
5581a51a | 7145 | u64 parent, int last_ref) |
f0486c68 | 7146 | { |
0b246afa | 7147 | struct btrfs_fs_info *fs_info = root->fs_info; |
ed4f255b | 7148 | struct btrfs_ref generic_ref = { 0 }; |
b150a4f1 | 7149 | int pin = 1; |
f0486c68 YZ |
7150 | int ret; |
7151 | ||
ed4f255b QW |
7152 | btrfs_init_generic_ref(&generic_ref, BTRFS_DROP_DELAYED_REF, |
7153 | buf->start, buf->len, parent); | |
7154 | btrfs_init_tree_ref(&generic_ref, btrfs_header_level(buf), | |
7155 | root->root_key.objectid); | |
7156 | ||
f0486c68 | 7157 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { |
d7eae340 OS |
7158 | int old_ref_mod, new_ref_mod; |
7159 | ||
8a5040f7 | 7160 | btrfs_ref_tree_mod(fs_info, &generic_ref); |
ed4f255b | 7161 | ret = btrfs_add_delayed_tree_ref(trans, &generic_ref, NULL, |
d7eae340 | 7162 | &old_ref_mod, &new_ref_mod); |
79787eaa | 7163 | BUG_ON(ret); /* -ENOMEM */ |
d7eae340 | 7164 | pin = old_ref_mod >= 0 && new_ref_mod < 0; |
f0486c68 YZ |
7165 | } |
7166 | ||
0a16c7d7 | 7167 | if (last_ref && btrfs_header_generation(buf) == trans->transid) { |
6219872d FM |
7168 | struct btrfs_block_group_cache *cache; |
7169 | ||
f0486c68 | 7170 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { |
2ff7e61e | 7171 | ret = check_ref_cleanup(trans, buf->start); |
f0486c68 | 7172 | if (!ret) |
37be25bc | 7173 | goto out; |
f0486c68 YZ |
7174 | } |
7175 | ||
4da8b76d | 7176 | pin = 0; |
0b246afa | 7177 | cache = btrfs_lookup_block_group(fs_info, buf->start); |
6219872d | 7178 | |
f0486c68 | 7179 | if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) { |
fdf08605 | 7180 | pin_down_extent(cache, buf->start, buf->len, 1); |
6219872d | 7181 | btrfs_put_block_group(cache); |
37be25bc | 7182 | goto out; |
f0486c68 YZ |
7183 | } |
7184 | ||
7185 | WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)); | |
7186 | ||
7187 | btrfs_add_free_space(cache, buf->start, buf->len); | |
4824f1f4 | 7188 | btrfs_free_reserved_bytes(cache, buf->len, 0); |
6219872d | 7189 | btrfs_put_block_group(cache); |
71ff6437 | 7190 | trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len); |
f0486c68 YZ |
7191 | } |
7192 | out: | |
b150a4f1 | 7193 | if (pin) |
14ae4ec1 | 7194 | add_pinned_bytes(fs_info, &generic_ref, 1); |
b150a4f1 | 7195 | |
0a16c7d7 OS |
7196 | if (last_ref) { |
7197 | /* | |
7198 | * Deleting the buffer, clear the corrupt flag since it doesn't | |
7199 | * matter anymore. | |
7200 | */ | |
7201 | clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags); | |
7202 | } | |
f0486c68 YZ |
7203 | } |
7204 | ||
79787eaa | 7205 | /* Can return -ENOMEM */ |
ffd4bb2a | 7206 | int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_ref *ref) |
925baedd | 7207 | { |
ffd4bb2a | 7208 | struct btrfs_fs_info *fs_info = trans->fs_info; |
d7eae340 | 7209 | int old_ref_mod, new_ref_mod; |
925baedd CM |
7210 | int ret; |
7211 | ||
f5ee5c9a | 7212 | if (btrfs_is_testing(fs_info)) |
faa2dbf0 | 7213 | return 0; |
fccb84c9 | 7214 | |
56bec294 CM |
7215 | /* |
7216 | * tree log blocks never actually go into the extent allocation | |
7217 | * tree, just update pinning info and exit early. | |
56bec294 | 7218 | */ |
ffd4bb2a QW |
7219 | if ((ref->type == BTRFS_REF_METADATA && |
7220 | ref->tree_ref.root == BTRFS_TREE_LOG_OBJECTID) || | |
7221 | (ref->type == BTRFS_REF_DATA && | |
7222 | ref->data_ref.ref_root == BTRFS_TREE_LOG_OBJECTID)) { | |
b9473439 | 7223 | /* unlocks the pinned mutex */ |
ffd4bb2a | 7224 | btrfs_pin_extent(fs_info, ref->bytenr, ref->len, 1); |
d7eae340 | 7225 | old_ref_mod = new_ref_mod = 0; |
56bec294 | 7226 | ret = 0; |
ffd4bb2a QW |
7227 | } else if (ref->type == BTRFS_REF_METADATA) { |
7228 | ret = btrfs_add_delayed_tree_ref(trans, ref, NULL, | |
d7eae340 | 7229 | &old_ref_mod, &new_ref_mod); |
5d4f98a2 | 7230 | } else { |
ffd4bb2a | 7231 | ret = btrfs_add_delayed_data_ref(trans, ref, 0, |
d7eae340 | 7232 | &old_ref_mod, &new_ref_mod); |
56bec294 | 7233 | } |
d7eae340 | 7234 | |
ffd4bb2a QW |
7235 | if (!((ref->type == BTRFS_REF_METADATA && |
7236 | ref->tree_ref.root == BTRFS_TREE_LOG_OBJECTID) || | |
7237 | (ref->type == BTRFS_REF_DATA && | |
7238 | ref->data_ref.ref_root == BTRFS_TREE_LOG_OBJECTID))) | |
7239 | btrfs_ref_tree_mod(fs_info, ref); | |
8a5040f7 | 7240 | |
ddf30cf0 | 7241 | if (ret == 0 && old_ref_mod >= 0 && new_ref_mod < 0) |
14ae4ec1 | 7242 | add_pinned_bytes(fs_info, ref, 1); |
d7eae340 | 7243 | |
925baedd CM |
7244 | return ret; |
7245 | } | |
7246 | ||
817d52f8 JB |
7247 | /* |
7248 | * when we wait for progress in the block group caching, its because | |
7249 | * our allocation attempt failed at least once. So, we must sleep | |
7250 | * and let some progress happen before we try again. | |
7251 | * | |
7252 | * This function will sleep at least once waiting for new free space to | |
7253 | * show up, and then it will check the block group free space numbers | |
7254 | * for our min num_bytes. Another option is to have it go ahead | |
7255 | * and look in the rbtree for a free extent of a given size, but this | |
7256 | * is a good start. | |
36cce922 JB |
7257 | * |
7258 | * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using | |
7259 | * any of the information in this block group. | |
817d52f8 | 7260 | */ |
36cce922 | 7261 | static noinline void |
817d52f8 JB |
7262 | wait_block_group_cache_progress(struct btrfs_block_group_cache *cache, |
7263 | u64 num_bytes) | |
7264 | { | |
11833d66 | 7265 | struct btrfs_caching_control *caching_ctl; |
817d52f8 | 7266 | |
11833d66 YZ |
7267 | caching_ctl = get_caching_control(cache); |
7268 | if (!caching_ctl) | |
36cce922 | 7269 | return; |
817d52f8 | 7270 | |
11833d66 | 7271 | wait_event(caching_ctl->wait, block_group_cache_done(cache) || |
34d52cb6 | 7272 | (cache->free_space_ctl->free_space >= num_bytes)); |
11833d66 YZ |
7273 | |
7274 | put_caching_control(caching_ctl); | |
11833d66 YZ |
7275 | } |
7276 | ||
7277 | static noinline int | |
7278 | wait_block_group_cache_done(struct btrfs_block_group_cache *cache) | |
7279 | { | |
7280 | struct btrfs_caching_control *caching_ctl; | |
36cce922 | 7281 | int ret = 0; |
11833d66 YZ |
7282 | |
7283 | caching_ctl = get_caching_control(cache); | |
7284 | if (!caching_ctl) | |
36cce922 | 7285 | return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0; |
11833d66 YZ |
7286 | |
7287 | wait_event(caching_ctl->wait, block_group_cache_done(cache)); | |
36cce922 JB |
7288 | if (cache->cached == BTRFS_CACHE_ERROR) |
7289 | ret = -EIO; | |
11833d66 | 7290 | put_caching_control(caching_ctl); |
36cce922 | 7291 | return ret; |
817d52f8 JB |
7292 | } |
7293 | ||
7294 | enum btrfs_loop_type { | |
285ff5af JB |
7295 | LOOP_CACHING_NOWAIT = 0, |
7296 | LOOP_CACHING_WAIT = 1, | |
7297 | LOOP_ALLOC_CHUNK = 2, | |
7298 | LOOP_NO_EMPTY_SIZE = 3, | |
817d52f8 JB |
7299 | }; |
7300 | ||
e570fd27 MX |
7301 | static inline void |
7302 | btrfs_lock_block_group(struct btrfs_block_group_cache *cache, | |
7303 | int delalloc) | |
7304 | { | |
7305 | if (delalloc) | |
7306 | down_read(&cache->data_rwsem); | |
7307 | } | |
7308 | ||
7309 | static inline void | |
7310 | btrfs_grab_block_group(struct btrfs_block_group_cache *cache, | |
7311 | int delalloc) | |
7312 | { | |
7313 | btrfs_get_block_group(cache); | |
7314 | if (delalloc) | |
7315 | down_read(&cache->data_rwsem); | |
7316 | } | |
7317 | ||
7318 | static struct btrfs_block_group_cache * | |
7319 | btrfs_lock_cluster(struct btrfs_block_group_cache *block_group, | |
7320 | struct btrfs_free_cluster *cluster, | |
7321 | int delalloc) | |
7322 | { | |
89771cc9 | 7323 | struct btrfs_block_group_cache *used_bg = NULL; |
6719afdc | 7324 | |
e570fd27 | 7325 | spin_lock(&cluster->refill_lock); |
6719afdc GU |
7326 | while (1) { |
7327 | used_bg = cluster->block_group; | |
7328 | if (!used_bg) | |
7329 | return NULL; | |
7330 | ||
7331 | if (used_bg == block_group) | |
e570fd27 MX |
7332 | return used_bg; |
7333 | ||
6719afdc | 7334 | btrfs_get_block_group(used_bg); |
e570fd27 | 7335 | |
6719afdc GU |
7336 | if (!delalloc) |
7337 | return used_bg; | |
e570fd27 | 7338 | |
6719afdc GU |
7339 | if (down_read_trylock(&used_bg->data_rwsem)) |
7340 | return used_bg; | |
e570fd27 | 7341 | |
6719afdc | 7342 | spin_unlock(&cluster->refill_lock); |
e570fd27 | 7343 | |
e321f8a8 LB |
7344 | /* We should only have one-level nested. */ |
7345 | down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING); | |
e570fd27 | 7346 | |
6719afdc GU |
7347 | spin_lock(&cluster->refill_lock); |
7348 | if (used_bg == cluster->block_group) | |
7349 | return used_bg; | |
e570fd27 | 7350 | |
6719afdc GU |
7351 | up_read(&used_bg->data_rwsem); |
7352 | btrfs_put_block_group(used_bg); | |
7353 | } | |
e570fd27 MX |
7354 | } |
7355 | ||
7356 | static inline void | |
7357 | btrfs_release_block_group(struct btrfs_block_group_cache *cache, | |
7358 | int delalloc) | |
7359 | { | |
7360 | if (delalloc) | |
7361 | up_read(&cache->data_rwsem); | |
7362 | btrfs_put_block_group(cache); | |
7363 | } | |
7364 | ||
b4bd745d QW |
7365 | /* |
7366 | * Structure used internally for find_free_extent() function. Wraps needed | |
7367 | * parameters. | |
7368 | */ | |
7369 | struct find_free_extent_ctl { | |
7370 | /* Basic allocation info */ | |
7371 | u64 ram_bytes; | |
7372 | u64 num_bytes; | |
7373 | u64 empty_size; | |
7374 | u64 flags; | |
7375 | int delalloc; | |
7376 | ||
7377 | /* Where to start the search inside the bg */ | |
7378 | u64 search_start; | |
7379 | ||
7380 | /* For clustered allocation */ | |
7381 | u64 empty_cluster; | |
7382 | ||
7383 | bool have_caching_bg; | |
7384 | bool orig_have_caching_bg; | |
7385 | ||
7386 | /* RAID index, converted from flags */ | |
7387 | int index; | |
7388 | ||
e72d79d6 QW |
7389 | /* |
7390 | * Current loop number, check find_free_extent_update_loop() for details | |
7391 | */ | |
b4bd745d QW |
7392 | int loop; |
7393 | ||
7394 | /* | |
7395 | * Whether we're refilling a cluster, if true we need to re-search | |
7396 | * current block group but don't try to refill the cluster again. | |
7397 | */ | |
7398 | bool retry_clustered; | |
7399 | ||
7400 | /* | |
7401 | * Whether we're updating free space cache, if true we need to re-search | |
7402 | * current block group but don't try updating free space cache again. | |
7403 | */ | |
7404 | bool retry_unclustered; | |
7405 | ||
7406 | /* If current block group is cached */ | |
7407 | int cached; | |
7408 | ||
7409 | /* Max contiguous hole found */ | |
7410 | u64 max_extent_size; | |
7411 | ||
7412 | /* Total free space from free space cache, not always contiguous */ | |
7413 | u64 total_free_space; | |
7414 | ||
7415 | /* Found result */ | |
7416 | u64 found_offset; | |
7417 | }; | |
7418 | ||
d06e3bb6 QW |
7419 | |
7420 | /* | |
7421 | * Helper function for find_free_extent(). | |
7422 | * | |
7423 | * Return -ENOENT to inform caller that we need fallback to unclustered mode. | |
7424 | * Return -EAGAIN to inform caller that we need to re-search this block group | |
7425 | * Return >0 to inform caller that we find nothing | |
7426 | * Return 0 means we have found a location and set ffe_ctl->found_offset. | |
7427 | */ | |
7428 | static int find_free_extent_clustered(struct btrfs_block_group_cache *bg, | |
7429 | struct btrfs_free_cluster *last_ptr, | |
7430 | struct find_free_extent_ctl *ffe_ctl, | |
7431 | struct btrfs_block_group_cache **cluster_bg_ret) | |
7432 | { | |
d06e3bb6 QW |
7433 | struct btrfs_block_group_cache *cluster_bg; |
7434 | u64 aligned_cluster; | |
7435 | u64 offset; | |
7436 | int ret; | |
7437 | ||
7438 | cluster_bg = btrfs_lock_cluster(bg, last_ptr, ffe_ctl->delalloc); | |
7439 | if (!cluster_bg) | |
7440 | goto refill_cluster; | |
7441 | if (cluster_bg != bg && (cluster_bg->ro || | |
7442 | !block_group_bits(cluster_bg, ffe_ctl->flags))) | |
7443 | goto release_cluster; | |
7444 | ||
7445 | offset = btrfs_alloc_from_cluster(cluster_bg, last_ptr, | |
7446 | ffe_ctl->num_bytes, cluster_bg->key.objectid, | |
7447 | &ffe_ctl->max_extent_size); | |
7448 | if (offset) { | |
7449 | /* We have a block, we're done */ | |
7450 | spin_unlock(&last_ptr->refill_lock); | |
7451 | trace_btrfs_reserve_extent_cluster(cluster_bg, | |
7452 | ffe_ctl->search_start, ffe_ctl->num_bytes); | |
7453 | *cluster_bg_ret = cluster_bg; | |
7454 | ffe_ctl->found_offset = offset; | |
7455 | return 0; | |
7456 | } | |
7457 | WARN_ON(last_ptr->block_group != cluster_bg); | |
7458 | ||
7459 | release_cluster: | |
7460 | /* | |
7461 | * If we are on LOOP_NO_EMPTY_SIZE, we can't set up a new clusters, so | |
7462 | * lets just skip it and let the allocator find whatever block it can | |
7463 | * find. If we reach this point, we will have tried the cluster | |
7464 | * allocator plenty of times and not have found anything, so we are | |
7465 | * likely way too fragmented for the clustering stuff to find anything. | |
7466 | * | |
7467 | * However, if the cluster is taken from the current block group, | |
7468 | * release the cluster first, so that we stand a better chance of | |
7469 | * succeeding in the unclustered allocation. | |
7470 | */ | |
7471 | if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE && cluster_bg != bg) { | |
7472 | spin_unlock(&last_ptr->refill_lock); | |
7473 | btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc); | |
7474 | return -ENOENT; | |
7475 | } | |
7476 | ||
7477 | /* This cluster didn't work out, free it and start over */ | |
7478 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
7479 | ||
7480 | if (cluster_bg != bg) | |
7481 | btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc); | |
7482 | ||
7483 | refill_cluster: | |
7484 | if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE) { | |
7485 | spin_unlock(&last_ptr->refill_lock); | |
7486 | return -ENOENT; | |
7487 | } | |
7488 | ||
7489 | aligned_cluster = max_t(u64, | |
7490 | ffe_ctl->empty_cluster + ffe_ctl->empty_size, | |
7491 | bg->full_stripe_len); | |
2ceeae2e DS |
7492 | ret = btrfs_find_space_cluster(bg, last_ptr, ffe_ctl->search_start, |
7493 | ffe_ctl->num_bytes, aligned_cluster); | |
d06e3bb6 QW |
7494 | if (ret == 0) { |
7495 | /* Now pull our allocation out of this cluster */ | |
7496 | offset = btrfs_alloc_from_cluster(bg, last_ptr, | |
7497 | ffe_ctl->num_bytes, ffe_ctl->search_start, | |
7498 | &ffe_ctl->max_extent_size); | |
7499 | if (offset) { | |
7500 | /* We found one, proceed */ | |
7501 | spin_unlock(&last_ptr->refill_lock); | |
7502 | trace_btrfs_reserve_extent_cluster(bg, | |
7503 | ffe_ctl->search_start, | |
7504 | ffe_ctl->num_bytes); | |
7505 | ffe_ctl->found_offset = offset; | |
7506 | return 0; | |
7507 | } | |
7508 | } else if (!ffe_ctl->cached && ffe_ctl->loop > LOOP_CACHING_NOWAIT && | |
7509 | !ffe_ctl->retry_clustered) { | |
7510 | spin_unlock(&last_ptr->refill_lock); | |
7511 | ||
7512 | ffe_ctl->retry_clustered = true; | |
7513 | wait_block_group_cache_progress(bg, ffe_ctl->num_bytes + | |
7514 | ffe_ctl->empty_cluster + ffe_ctl->empty_size); | |
7515 | return -EAGAIN; | |
7516 | } | |
7517 | /* | |
7518 | * At this point we either didn't find a cluster or we weren't able to | |
7519 | * allocate a block from our cluster. Free the cluster we've been | |
7520 | * trying to use, and go to the next block group. | |
7521 | */ | |
7522 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
7523 | spin_unlock(&last_ptr->refill_lock); | |
7524 | return 1; | |
7525 | } | |
7526 | ||
e1a41848 QW |
7527 | /* |
7528 | * Return >0 to inform caller that we find nothing | |
7529 | * Return 0 when we found an free extent and set ffe_ctrl->found_offset | |
7530 | * Return -EAGAIN to inform caller that we need to re-search this block group | |
7531 | */ | |
7532 | static int find_free_extent_unclustered(struct btrfs_block_group_cache *bg, | |
7533 | struct btrfs_free_cluster *last_ptr, | |
7534 | struct find_free_extent_ctl *ffe_ctl) | |
7535 | { | |
7536 | u64 offset; | |
7537 | ||
7538 | /* | |
7539 | * We are doing an unclustered allocation, set the fragmented flag so | |
7540 | * we don't bother trying to setup a cluster again until we get more | |
7541 | * space. | |
7542 | */ | |
7543 | if (unlikely(last_ptr)) { | |
7544 | spin_lock(&last_ptr->lock); | |
7545 | last_ptr->fragmented = 1; | |
7546 | spin_unlock(&last_ptr->lock); | |
7547 | } | |
7548 | if (ffe_ctl->cached) { | |
7549 | struct btrfs_free_space_ctl *free_space_ctl; | |
7550 | ||
7551 | free_space_ctl = bg->free_space_ctl; | |
7552 | spin_lock(&free_space_ctl->tree_lock); | |
7553 | if (free_space_ctl->free_space < | |
7554 | ffe_ctl->num_bytes + ffe_ctl->empty_cluster + | |
7555 | ffe_ctl->empty_size) { | |
7556 | ffe_ctl->total_free_space = max_t(u64, | |
7557 | ffe_ctl->total_free_space, | |
7558 | free_space_ctl->free_space); | |
7559 | spin_unlock(&free_space_ctl->tree_lock); | |
7560 | return 1; | |
7561 | } | |
7562 | spin_unlock(&free_space_ctl->tree_lock); | |
7563 | } | |
7564 | ||
7565 | offset = btrfs_find_space_for_alloc(bg, ffe_ctl->search_start, | |
7566 | ffe_ctl->num_bytes, ffe_ctl->empty_size, | |
7567 | &ffe_ctl->max_extent_size); | |
7568 | ||
7569 | /* | |
7570 | * If we didn't find a chunk, and we haven't failed on this block group | |
7571 | * before, and this block group is in the middle of caching and we are | |
7572 | * ok with waiting, then go ahead and wait for progress to be made, and | |
7573 | * set @retry_unclustered to true. | |
7574 | * | |
7575 | * If @retry_unclustered is true then we've already waited on this | |
7576 | * block group once and should move on to the next block group. | |
7577 | */ | |
7578 | if (!offset && !ffe_ctl->retry_unclustered && !ffe_ctl->cached && | |
7579 | ffe_ctl->loop > LOOP_CACHING_NOWAIT) { | |
7580 | wait_block_group_cache_progress(bg, ffe_ctl->num_bytes + | |
7581 | ffe_ctl->empty_size); | |
7582 | ffe_ctl->retry_unclustered = true; | |
7583 | return -EAGAIN; | |
7584 | } else if (!offset) { | |
7585 | return 1; | |
7586 | } | |
7587 | ffe_ctl->found_offset = offset; | |
7588 | return 0; | |
7589 | } | |
7590 | ||
e72d79d6 QW |
7591 | /* |
7592 | * Return >0 means caller needs to re-search for free extent | |
7593 | * Return 0 means we have the needed free extent. | |
7594 | * Return <0 means we failed to locate any free extent. | |
7595 | */ | |
7596 | static int find_free_extent_update_loop(struct btrfs_fs_info *fs_info, | |
7597 | struct btrfs_free_cluster *last_ptr, | |
7598 | struct btrfs_key *ins, | |
7599 | struct find_free_extent_ctl *ffe_ctl, | |
7600 | int full_search, bool use_cluster) | |
7601 | { | |
7602 | struct btrfs_root *root = fs_info->extent_root; | |
7603 | int ret; | |
7604 | ||
7605 | if ((ffe_ctl->loop == LOOP_CACHING_NOWAIT) && | |
7606 | ffe_ctl->have_caching_bg && !ffe_ctl->orig_have_caching_bg) | |
7607 | ffe_ctl->orig_have_caching_bg = true; | |
7608 | ||
7609 | if (!ins->objectid && ffe_ctl->loop >= LOOP_CACHING_WAIT && | |
7610 | ffe_ctl->have_caching_bg) | |
7611 | return 1; | |
7612 | ||
7613 | if (!ins->objectid && ++(ffe_ctl->index) < BTRFS_NR_RAID_TYPES) | |
7614 | return 1; | |
7615 | ||
7616 | if (ins->objectid) { | |
7617 | if (!use_cluster && last_ptr) { | |
7618 | spin_lock(&last_ptr->lock); | |
7619 | last_ptr->window_start = ins->objectid; | |
7620 | spin_unlock(&last_ptr->lock); | |
7621 | } | |
7622 | return 0; | |
7623 | } | |
7624 | ||
7625 | /* | |
7626 | * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking | |
7627 | * caching kthreads as we move along | |
7628 | * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching | |
7629 | * LOOP_ALLOC_CHUNK, force a chunk allocation and try again | |
7630 | * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try | |
7631 | * again | |
7632 | */ | |
7633 | if (ffe_ctl->loop < LOOP_NO_EMPTY_SIZE) { | |
7634 | ffe_ctl->index = 0; | |
7635 | if (ffe_ctl->loop == LOOP_CACHING_NOWAIT) { | |
7636 | /* | |
7637 | * We want to skip the LOOP_CACHING_WAIT step if we | |
7638 | * don't have any uncached bgs and we've already done a | |
7639 | * full search through. | |
7640 | */ | |
7641 | if (ffe_ctl->orig_have_caching_bg || !full_search) | |
7642 | ffe_ctl->loop = LOOP_CACHING_WAIT; | |
7643 | else | |
7644 | ffe_ctl->loop = LOOP_ALLOC_CHUNK; | |
7645 | } else { | |
7646 | ffe_ctl->loop++; | |
7647 | } | |
7648 | ||
7649 | if (ffe_ctl->loop == LOOP_ALLOC_CHUNK) { | |
7650 | struct btrfs_trans_handle *trans; | |
7651 | int exist = 0; | |
7652 | ||
7653 | trans = current->journal_info; | |
7654 | if (trans) | |
7655 | exist = 1; | |
7656 | else | |
7657 | trans = btrfs_join_transaction(root); | |
7658 | ||
7659 | if (IS_ERR(trans)) { | |
7660 | ret = PTR_ERR(trans); | |
7661 | return ret; | |
7662 | } | |
7663 | ||
7664 | ret = do_chunk_alloc(trans, ffe_ctl->flags, | |
7665 | CHUNK_ALLOC_FORCE); | |
7666 | ||
7667 | /* | |
7668 | * If we can't allocate a new chunk we've already looped | |
7669 | * through at least once, move on to the NO_EMPTY_SIZE | |
7670 | * case. | |
7671 | */ | |
7672 | if (ret == -ENOSPC) | |
7673 | ffe_ctl->loop = LOOP_NO_EMPTY_SIZE; | |
7674 | ||
7675 | /* Do not bail out on ENOSPC since we can do more. */ | |
7676 | if (ret < 0 && ret != -ENOSPC) | |
7677 | btrfs_abort_transaction(trans, ret); | |
7678 | else | |
7679 | ret = 0; | |
7680 | if (!exist) | |
7681 | btrfs_end_transaction(trans); | |
7682 | if (ret) | |
7683 | return ret; | |
7684 | } | |
7685 | ||
7686 | if (ffe_ctl->loop == LOOP_NO_EMPTY_SIZE) { | |
7687 | /* | |
7688 | * Don't loop again if we already have no empty_size and | |
7689 | * no empty_cluster. | |
7690 | */ | |
7691 | if (ffe_ctl->empty_size == 0 && | |
7692 | ffe_ctl->empty_cluster == 0) | |
7693 | return -ENOSPC; | |
7694 | ffe_ctl->empty_size = 0; | |
7695 | ffe_ctl->empty_cluster = 0; | |
7696 | } | |
7697 | return 1; | |
7698 | } | |
7699 | return -ENOSPC; | |
7700 | } | |
7701 | ||
fec577fb CM |
7702 | /* |
7703 | * walks the btree of allocated extents and find a hole of a given size. | |
7704 | * The key ins is changed to record the hole: | |
a4820398 | 7705 | * ins->objectid == start position |
62e2749e | 7706 | * ins->flags = BTRFS_EXTENT_ITEM_KEY |
a4820398 | 7707 | * ins->offset == the size of the hole. |
fec577fb | 7708 | * Any available blocks before search_start are skipped. |
a4820398 MX |
7709 | * |
7710 | * If there is no suitable free space, we will record the max size of | |
7711 | * the free space extent currently. | |
e72d79d6 QW |
7712 | * |
7713 | * The overall logic and call chain: | |
7714 | * | |
7715 | * find_free_extent() | |
7716 | * |- Iterate through all block groups | |
7717 | * | |- Get a valid block group | |
7718 | * | |- Try to do clustered allocation in that block group | |
7719 | * | |- Try to do unclustered allocation in that block group | |
7720 | * | |- Check if the result is valid | |
7721 | * | | |- If valid, then exit | |
7722 | * | |- Jump to next block group | |
7723 | * | | |
7724 | * |- Push harder to find free extents | |
7725 | * |- If not found, re-iterate all block groups | |
fec577fb | 7726 | */ |
87bde3cd | 7727 | static noinline int find_free_extent(struct btrfs_fs_info *fs_info, |
18513091 WX |
7728 | u64 ram_bytes, u64 num_bytes, u64 empty_size, |
7729 | u64 hint_byte, struct btrfs_key *ins, | |
7730 | u64 flags, int delalloc) | |
fec577fb | 7731 | { |
80eb234a | 7732 | int ret = 0; |
fa9c0d79 | 7733 | struct btrfs_free_cluster *last_ptr = NULL; |
80eb234a | 7734 | struct btrfs_block_group_cache *block_group = NULL; |
b4bd745d | 7735 | struct find_free_extent_ctl ffe_ctl = {0}; |
80eb234a | 7736 | struct btrfs_space_info *space_info; |
67377734 | 7737 | bool use_cluster = true; |
a5e681d9 | 7738 | bool full_search = false; |
fec577fb | 7739 | |
0b246afa | 7740 | WARN_ON(num_bytes < fs_info->sectorsize); |
b4bd745d QW |
7741 | |
7742 | ffe_ctl.ram_bytes = ram_bytes; | |
7743 | ffe_ctl.num_bytes = num_bytes; | |
7744 | ffe_ctl.empty_size = empty_size; | |
7745 | ffe_ctl.flags = flags; | |
7746 | ffe_ctl.search_start = 0; | |
7747 | ffe_ctl.retry_clustered = false; | |
7748 | ffe_ctl.retry_unclustered = false; | |
7749 | ffe_ctl.delalloc = delalloc; | |
7750 | ffe_ctl.index = btrfs_bg_flags_to_raid_index(flags); | |
7751 | ffe_ctl.have_caching_bg = false; | |
7752 | ffe_ctl.orig_have_caching_bg = false; | |
7753 | ffe_ctl.found_offset = 0; | |
7754 | ||
962a298f | 7755 | ins->type = BTRFS_EXTENT_ITEM_KEY; |
80eb234a JB |
7756 | ins->objectid = 0; |
7757 | ins->offset = 0; | |
b1a4d965 | 7758 | |
71ff6437 | 7759 | trace_find_free_extent(fs_info, num_bytes, empty_size, flags); |
3f7de037 | 7760 | |
0b246afa | 7761 | space_info = __find_space_info(fs_info, flags); |
1b1d1f66 | 7762 | if (!space_info) { |
0b246afa | 7763 | btrfs_err(fs_info, "No space info for %llu", flags); |
1b1d1f66 JB |
7764 | return -ENOSPC; |
7765 | } | |
2552d17e | 7766 | |
67377734 | 7767 | /* |
4f4db217 JB |
7768 | * If our free space is heavily fragmented we may not be able to make |
7769 | * big contiguous allocations, so instead of doing the expensive search | |
7770 | * for free space, simply return ENOSPC with our max_extent_size so we | |
7771 | * can go ahead and search for a more manageable chunk. | |
7772 | * | |
7773 | * If our max_extent_size is large enough for our allocation simply | |
7774 | * disable clustering since we will likely not be able to find enough | |
7775 | * space to create a cluster and induce latency trying. | |
67377734 | 7776 | */ |
4f4db217 JB |
7777 | if (unlikely(space_info->max_extent_size)) { |
7778 | spin_lock(&space_info->lock); | |
7779 | if (space_info->max_extent_size && | |
7780 | num_bytes > space_info->max_extent_size) { | |
7781 | ins->offset = space_info->max_extent_size; | |
7782 | spin_unlock(&space_info->lock); | |
7783 | return -ENOSPC; | |
7784 | } else if (space_info->max_extent_size) { | |
7785 | use_cluster = false; | |
7786 | } | |
7787 | spin_unlock(&space_info->lock); | |
fa9c0d79 | 7788 | } |
0f9dd46c | 7789 | |
b4bd745d QW |
7790 | last_ptr = fetch_cluster_info(fs_info, space_info, |
7791 | &ffe_ctl.empty_cluster); | |
239b14b3 | 7792 | if (last_ptr) { |
fa9c0d79 CM |
7793 | spin_lock(&last_ptr->lock); |
7794 | if (last_ptr->block_group) | |
7795 | hint_byte = last_ptr->window_start; | |
c759c4e1 JB |
7796 | if (last_ptr->fragmented) { |
7797 | /* | |
7798 | * We still set window_start so we can keep track of the | |
7799 | * last place we found an allocation to try and save | |
7800 | * some time. | |
7801 | */ | |
7802 | hint_byte = last_ptr->window_start; | |
7803 | use_cluster = false; | |
7804 | } | |
fa9c0d79 | 7805 | spin_unlock(&last_ptr->lock); |
239b14b3 | 7806 | } |
fa9c0d79 | 7807 | |
b4bd745d QW |
7808 | ffe_ctl.search_start = max(ffe_ctl.search_start, |
7809 | first_logical_byte(fs_info, 0)); | |
7810 | ffe_ctl.search_start = max(ffe_ctl.search_start, hint_byte); | |
7811 | if (ffe_ctl.search_start == hint_byte) { | |
7812 | block_group = btrfs_lookup_block_group(fs_info, | |
7813 | ffe_ctl.search_start); | |
817d52f8 JB |
7814 | /* |
7815 | * we don't want to use the block group if it doesn't match our | |
7816 | * allocation bits, or if its not cached. | |
ccf0e725 JB |
7817 | * |
7818 | * However if we are re-searching with an ideal block group | |
7819 | * picked out then we don't care that the block group is cached. | |
817d52f8 | 7820 | */ |
b6919a58 | 7821 | if (block_group && block_group_bits(block_group, flags) && |
285ff5af | 7822 | block_group->cached != BTRFS_CACHE_NO) { |
2552d17e | 7823 | down_read(&space_info->groups_sem); |
44fb5511 CM |
7824 | if (list_empty(&block_group->list) || |
7825 | block_group->ro) { | |
7826 | /* | |
7827 | * someone is removing this block group, | |
7828 | * we can't jump into the have_block_group | |
7829 | * target because our list pointers are not | |
7830 | * valid | |
7831 | */ | |
7832 | btrfs_put_block_group(block_group); | |
7833 | up_read(&space_info->groups_sem); | |
ccf0e725 | 7834 | } else { |
b4bd745d | 7835 | ffe_ctl.index = btrfs_bg_flags_to_raid_index( |
3e72ee88 | 7836 | block_group->flags); |
e570fd27 | 7837 | btrfs_lock_block_group(block_group, delalloc); |
44fb5511 | 7838 | goto have_block_group; |
ccf0e725 | 7839 | } |
2552d17e | 7840 | } else if (block_group) { |
fa9c0d79 | 7841 | btrfs_put_block_group(block_group); |
2552d17e | 7842 | } |
42e70e7a | 7843 | } |
2552d17e | 7844 | search: |
b4bd745d QW |
7845 | ffe_ctl.have_caching_bg = false; |
7846 | if (ffe_ctl.index == btrfs_bg_flags_to_raid_index(flags) || | |
7847 | ffe_ctl.index == 0) | |
a5e681d9 | 7848 | full_search = true; |
80eb234a | 7849 | down_read(&space_info->groups_sem); |
b4bd745d QW |
7850 | list_for_each_entry(block_group, |
7851 | &space_info->block_groups[ffe_ctl.index], list) { | |
14443937 JM |
7852 | /* If the block group is read-only, we can skip it entirely. */ |
7853 | if (unlikely(block_group->ro)) | |
7854 | continue; | |
7855 | ||
e570fd27 | 7856 | btrfs_grab_block_group(block_group, delalloc); |
b4bd745d | 7857 | ffe_ctl.search_start = block_group->key.objectid; |
42e70e7a | 7858 | |
83a50de9 CM |
7859 | /* |
7860 | * this can happen if we end up cycling through all the | |
7861 | * raid types, but we want to make sure we only allocate | |
7862 | * for the proper type. | |
7863 | */ | |
b6919a58 | 7864 | if (!block_group_bits(block_group, flags)) { |
bece2e82 | 7865 | u64 extra = BTRFS_BLOCK_GROUP_DUP | |
83a50de9 | 7866 | BTRFS_BLOCK_GROUP_RAID1 | |
53b381b3 DW |
7867 | BTRFS_BLOCK_GROUP_RAID5 | |
7868 | BTRFS_BLOCK_GROUP_RAID6 | | |
83a50de9 CM |
7869 | BTRFS_BLOCK_GROUP_RAID10; |
7870 | ||
7871 | /* | |
7872 | * if they asked for extra copies and this block group | |
7873 | * doesn't provide them, bail. This does allow us to | |
7874 | * fill raid0 from raid1. | |
7875 | */ | |
b6919a58 | 7876 | if ((flags & extra) && !(block_group->flags & extra)) |
83a50de9 CM |
7877 | goto loop; |
7878 | } | |
7879 | ||
2552d17e | 7880 | have_block_group: |
b4bd745d QW |
7881 | ffe_ctl.cached = block_group_cache_done(block_group); |
7882 | if (unlikely(!ffe_ctl.cached)) { | |
7883 | ffe_ctl.have_caching_bg = true; | |
f6373bf3 | 7884 | ret = cache_block_group(block_group, 0); |
1d4284bd CM |
7885 | BUG_ON(ret < 0); |
7886 | ret = 0; | |
817d52f8 JB |
7887 | } |
7888 | ||
36cce922 JB |
7889 | if (unlikely(block_group->cached == BTRFS_CACHE_ERROR)) |
7890 | goto loop; | |
0f9dd46c | 7891 | |
0a24325e | 7892 | /* |
062c05c4 AO |
7893 | * Ok we want to try and use the cluster allocator, so |
7894 | * lets look there | |
0a24325e | 7895 | */ |
c759c4e1 | 7896 | if (last_ptr && use_cluster) { |
d06e3bb6 | 7897 | struct btrfs_block_group_cache *cluster_bg = NULL; |
fa9c0d79 | 7898 | |
d06e3bb6 QW |
7899 | ret = find_free_extent_clustered(block_group, last_ptr, |
7900 | &ffe_ctl, &cluster_bg); | |
062c05c4 | 7901 | |
fa9c0d79 | 7902 | if (ret == 0) { |
d06e3bb6 QW |
7903 | if (cluster_bg && cluster_bg != block_group) { |
7904 | btrfs_release_block_group(block_group, | |
7905 | delalloc); | |
7906 | block_group = cluster_bg; | |
fa9c0d79 | 7907 | } |
d06e3bb6 QW |
7908 | goto checks; |
7909 | } else if (ret == -EAGAIN) { | |
817d52f8 | 7910 | goto have_block_group; |
d06e3bb6 QW |
7911 | } else if (ret > 0) { |
7912 | goto loop; | |
fa9c0d79 | 7913 | } |
d06e3bb6 | 7914 | /* ret == -ENOENT case falls through */ |
fa9c0d79 CM |
7915 | } |
7916 | ||
e1a41848 QW |
7917 | ret = find_free_extent_unclustered(block_group, last_ptr, |
7918 | &ffe_ctl); | |
7919 | if (ret == -EAGAIN) | |
817d52f8 | 7920 | goto have_block_group; |
e1a41848 | 7921 | else if (ret > 0) |
1cdda9b8 | 7922 | goto loop; |
e1a41848 | 7923 | /* ret == 0 case falls through */ |
fa9c0d79 | 7924 | checks: |
b4bd745d QW |
7925 | ffe_ctl.search_start = round_up(ffe_ctl.found_offset, |
7926 | fs_info->stripesize); | |
25179201 | 7927 | |
2552d17e | 7928 | /* move on to the next group */ |
b4bd745d | 7929 | if (ffe_ctl.search_start + num_bytes > |
215a63d1 | 7930 | block_group->key.objectid + block_group->key.offset) { |
b4bd745d QW |
7931 | btrfs_add_free_space(block_group, ffe_ctl.found_offset, |
7932 | num_bytes); | |
2552d17e | 7933 | goto loop; |
6226cb0a | 7934 | } |
f5a31e16 | 7935 | |
b4bd745d QW |
7936 | if (ffe_ctl.found_offset < ffe_ctl.search_start) |
7937 | btrfs_add_free_space(block_group, ffe_ctl.found_offset, | |
7938 | ffe_ctl.search_start - ffe_ctl.found_offset); | |
2552d17e | 7939 | |
18513091 WX |
7940 | ret = btrfs_add_reserved_bytes(block_group, ram_bytes, |
7941 | num_bytes, delalloc); | |
f0486c68 | 7942 | if (ret == -EAGAIN) { |
b4bd745d QW |
7943 | btrfs_add_free_space(block_group, ffe_ctl.found_offset, |
7944 | num_bytes); | |
2552d17e | 7945 | goto loop; |
0f9dd46c | 7946 | } |
9cfa3e34 | 7947 | btrfs_inc_block_group_reservations(block_group); |
0b86a832 | 7948 | |
f0486c68 | 7949 | /* we are all good, lets return */ |
b4bd745d | 7950 | ins->objectid = ffe_ctl.search_start; |
2552d17e | 7951 | ins->offset = num_bytes; |
d2fb3437 | 7952 | |
b4bd745d QW |
7953 | trace_btrfs_reserve_extent(block_group, ffe_ctl.search_start, |
7954 | num_bytes); | |
e570fd27 | 7955 | btrfs_release_block_group(block_group, delalloc); |
2552d17e JB |
7956 | break; |
7957 | loop: | |
b4bd745d QW |
7958 | ffe_ctl.retry_clustered = false; |
7959 | ffe_ctl.retry_unclustered = false; | |
3e72ee88 | 7960 | BUG_ON(btrfs_bg_flags_to_raid_index(block_group->flags) != |
b4bd745d | 7961 | ffe_ctl.index); |
e570fd27 | 7962 | btrfs_release_block_group(block_group, delalloc); |
14443937 | 7963 | cond_resched(); |
2552d17e JB |
7964 | } |
7965 | up_read(&space_info->groups_sem); | |
7966 | ||
e72d79d6 QW |
7967 | ret = find_free_extent_update_loop(fs_info, last_ptr, ins, &ffe_ctl, |
7968 | full_search, use_cluster); | |
7969 | if (ret > 0) | |
b742bb82 YZ |
7970 | goto search; |
7971 | ||
4f4db217 | 7972 | if (ret == -ENOSPC) { |
b4bd745d QW |
7973 | /* |
7974 | * Use ffe_ctl->total_free_space as fallback if we can't find | |
7975 | * any contiguous hole. | |
7976 | */ | |
7977 | if (!ffe_ctl.max_extent_size) | |
7978 | ffe_ctl.max_extent_size = ffe_ctl.total_free_space; | |
4f4db217 | 7979 | spin_lock(&space_info->lock); |
b4bd745d | 7980 | space_info->max_extent_size = ffe_ctl.max_extent_size; |
4f4db217 | 7981 | spin_unlock(&space_info->lock); |
b4bd745d | 7982 | ins->offset = ffe_ctl.max_extent_size; |
4f4db217 | 7983 | } |
0f70abe2 | 7984 | return ret; |
fec577fb | 7985 | } |
ec44a35c | 7986 | |
b78e5616 JB |
7987 | #define DUMP_BLOCK_RSV(fs_info, rsv_name) \ |
7988 | do { \ | |
7989 | struct btrfs_block_rsv *__rsv = &(fs_info)->rsv_name; \ | |
7990 | spin_lock(&__rsv->lock); \ | |
7991 | btrfs_info(fs_info, #rsv_name ": size %llu reserved %llu", \ | |
7992 | __rsv->size, __rsv->reserved); \ | |
7993 | spin_unlock(&__rsv->lock); \ | |
7994 | } while (0) | |
7995 | ||
ab8d0fc4 JM |
7996 | static void dump_space_info(struct btrfs_fs_info *fs_info, |
7997 | struct btrfs_space_info *info, u64 bytes, | |
9ed74f2d | 7998 | int dump_block_groups) |
0f9dd46c JB |
7999 | { |
8000 | struct btrfs_block_group_cache *cache; | |
b742bb82 | 8001 | int index = 0; |
0f9dd46c | 8002 | |
9ed74f2d | 8003 | spin_lock(&info->lock); |
ab8d0fc4 JM |
8004 | btrfs_info(fs_info, "space_info %llu has %llu free, is %sfull", |
8005 | info->flags, | |
4136135b LB |
8006 | info->total_bytes - btrfs_space_info_used(info, true), |
8007 | info->full ? "" : "not "); | |
ab8d0fc4 JM |
8008 | btrfs_info(fs_info, |
8009 | "space_info total=%llu, used=%llu, pinned=%llu, reserved=%llu, may_use=%llu, readonly=%llu", | |
8010 | info->total_bytes, info->bytes_used, info->bytes_pinned, | |
8011 | info->bytes_reserved, info->bytes_may_use, | |
8012 | info->bytes_readonly); | |
9ed74f2d JB |
8013 | spin_unlock(&info->lock); |
8014 | ||
b78e5616 JB |
8015 | DUMP_BLOCK_RSV(fs_info, global_block_rsv); |
8016 | DUMP_BLOCK_RSV(fs_info, trans_block_rsv); | |
8017 | DUMP_BLOCK_RSV(fs_info, chunk_block_rsv); | |
8018 | DUMP_BLOCK_RSV(fs_info, delayed_block_rsv); | |
8019 | DUMP_BLOCK_RSV(fs_info, delayed_refs_rsv); | |
8020 | ||
9ed74f2d JB |
8021 | if (!dump_block_groups) |
8022 | return; | |
0f9dd46c | 8023 | |
80eb234a | 8024 | down_read(&info->groups_sem); |
b742bb82 YZ |
8025 | again: |
8026 | list_for_each_entry(cache, &info->block_groups[index], list) { | |
0f9dd46c | 8027 | spin_lock(&cache->lock); |
ab8d0fc4 JM |
8028 | btrfs_info(fs_info, |
8029 | "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s", | |
8030 | cache->key.objectid, cache->key.offset, | |
8031 | btrfs_block_group_used(&cache->item), cache->pinned, | |
8032 | cache->reserved, cache->ro ? "[readonly]" : ""); | |
0f9dd46c JB |
8033 | btrfs_dump_free_space(cache, bytes); |
8034 | spin_unlock(&cache->lock); | |
8035 | } | |
b742bb82 YZ |
8036 | if (++index < BTRFS_NR_RAID_TYPES) |
8037 | goto again; | |
80eb234a | 8038 | up_read(&info->groups_sem); |
0f9dd46c | 8039 | } |
e8569813 | 8040 | |
6f47c706 NB |
8041 | /* |
8042 | * btrfs_reserve_extent - entry point to the extent allocator. Tries to find a | |
8043 | * hole that is at least as big as @num_bytes. | |
8044 | * | |
8045 | * @root - The root that will contain this extent | |
8046 | * | |
8047 | * @ram_bytes - The amount of space in ram that @num_bytes take. This | |
8048 | * is used for accounting purposes. This value differs | |
8049 | * from @num_bytes only in the case of compressed extents. | |
8050 | * | |
8051 | * @num_bytes - Number of bytes to allocate on-disk. | |
8052 | * | |
8053 | * @min_alloc_size - Indicates the minimum amount of space that the | |
8054 | * allocator should try to satisfy. In some cases | |
8055 | * @num_bytes may be larger than what is required and if | |
8056 | * the filesystem is fragmented then allocation fails. | |
8057 | * However, the presence of @min_alloc_size gives a | |
8058 | * chance to try and satisfy the smaller allocation. | |
8059 | * | |
8060 | * @empty_size - A hint that you plan on doing more COW. This is the | |
8061 | * size in bytes the allocator should try to find free | |
8062 | * next to the block it returns. This is just a hint and | |
8063 | * may be ignored by the allocator. | |
8064 | * | |
8065 | * @hint_byte - Hint to the allocator to start searching above the byte | |
8066 | * address passed. It might be ignored. | |
8067 | * | |
8068 | * @ins - This key is modified to record the found hole. It will | |
8069 | * have the following values: | |
8070 | * ins->objectid == start position | |
8071 | * ins->flags = BTRFS_EXTENT_ITEM_KEY | |
8072 | * ins->offset == the size of the hole. | |
8073 | * | |
8074 | * @is_data - Boolean flag indicating whether an extent is | |
8075 | * allocated for data (true) or metadata (false) | |
8076 | * | |
8077 | * @delalloc - Boolean flag indicating whether this allocation is for | |
8078 | * delalloc or not. If 'true' data_rwsem of block groups | |
8079 | * is going to be acquired. | |
8080 | * | |
8081 | * | |
8082 | * Returns 0 when an allocation succeeded or < 0 when an error occurred. In | |
8083 | * case -ENOSPC is returned then @ins->offset will contain the size of the | |
8084 | * largest available hole the allocator managed to find. | |
8085 | */ | |
18513091 | 8086 | int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes, |
11833d66 YZ |
8087 | u64 num_bytes, u64 min_alloc_size, |
8088 | u64 empty_size, u64 hint_byte, | |
e570fd27 | 8089 | struct btrfs_key *ins, int is_data, int delalloc) |
fec577fb | 8090 | { |
ab8d0fc4 | 8091 | struct btrfs_fs_info *fs_info = root->fs_info; |
36af4e07 | 8092 | bool final_tried = num_bytes == min_alloc_size; |
b6919a58 | 8093 | u64 flags; |
fec577fb | 8094 | int ret; |
925baedd | 8095 | |
1b86826d | 8096 | flags = get_alloc_profile_by_root(root, is_data); |
98d20f67 | 8097 | again: |
0b246afa | 8098 | WARN_ON(num_bytes < fs_info->sectorsize); |
87bde3cd | 8099 | ret = find_free_extent(fs_info, ram_bytes, num_bytes, empty_size, |
18513091 | 8100 | hint_byte, ins, flags, delalloc); |
9cfa3e34 | 8101 | if (!ret && !is_data) { |
ab8d0fc4 | 8102 | btrfs_dec_block_group_reservations(fs_info, ins->objectid); |
9cfa3e34 | 8103 | } else if (ret == -ENOSPC) { |
a4820398 MX |
8104 | if (!final_tried && ins->offset) { |
8105 | num_bytes = min(num_bytes >> 1, ins->offset); | |
da17066c | 8106 | num_bytes = round_down(num_bytes, |
0b246afa | 8107 | fs_info->sectorsize); |
9e622d6b | 8108 | num_bytes = max(num_bytes, min_alloc_size); |
18513091 | 8109 | ram_bytes = num_bytes; |
9e622d6b MX |
8110 | if (num_bytes == min_alloc_size) |
8111 | final_tried = true; | |
8112 | goto again; | |
ab8d0fc4 | 8113 | } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
9e622d6b MX |
8114 | struct btrfs_space_info *sinfo; |
8115 | ||
ab8d0fc4 | 8116 | sinfo = __find_space_info(fs_info, flags); |
0b246afa | 8117 | btrfs_err(fs_info, |
5d163e0e JM |
8118 | "allocation failed flags %llu, wanted %llu", |
8119 | flags, num_bytes); | |
53804280 | 8120 | if (sinfo) |
ab8d0fc4 | 8121 | dump_space_info(fs_info, sinfo, num_bytes, 1); |
9e622d6b | 8122 | } |
925baedd | 8123 | } |
0f9dd46c JB |
8124 | |
8125 | return ret; | |
e6dcd2dc CM |
8126 | } |
8127 | ||
2ff7e61e | 8128 | static int __btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info, |
e570fd27 MX |
8129 | u64 start, u64 len, |
8130 | int pin, int delalloc) | |
65b51a00 | 8131 | { |
0f9dd46c | 8132 | struct btrfs_block_group_cache *cache; |
1f3c79a2 | 8133 | int ret = 0; |
0f9dd46c | 8134 | |
0b246afa | 8135 | cache = btrfs_lookup_block_group(fs_info, start); |
0f9dd46c | 8136 | if (!cache) { |
0b246afa JM |
8137 | btrfs_err(fs_info, "Unable to find block group for %llu", |
8138 | start); | |
0f9dd46c JB |
8139 | return -ENOSPC; |
8140 | } | |
1f3c79a2 | 8141 | |
e688b725 | 8142 | if (pin) |
fdf08605 | 8143 | pin_down_extent(cache, start, len, 1); |
e688b725 | 8144 | else { |
0b246afa | 8145 | if (btrfs_test_opt(fs_info, DISCARD)) |
2ff7e61e | 8146 | ret = btrfs_discard_extent(fs_info, start, len, NULL); |
e688b725 | 8147 | btrfs_add_free_space(cache, start, len); |
4824f1f4 | 8148 | btrfs_free_reserved_bytes(cache, len, delalloc); |
71ff6437 | 8149 | trace_btrfs_reserved_extent_free(fs_info, start, len); |
e688b725 | 8150 | } |
31193213 | 8151 | |
fa9c0d79 | 8152 | btrfs_put_block_group(cache); |
e6dcd2dc CM |
8153 | return ret; |
8154 | } | |
8155 | ||
2ff7e61e | 8156 | int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info, |
e570fd27 | 8157 | u64 start, u64 len, int delalloc) |
e688b725 | 8158 | { |
2ff7e61e | 8159 | return __btrfs_free_reserved_extent(fs_info, start, len, 0, delalloc); |
e688b725 CM |
8160 | } |
8161 | ||
2ff7e61e | 8162 | int btrfs_free_and_pin_reserved_extent(struct btrfs_fs_info *fs_info, |
e688b725 CM |
8163 | u64 start, u64 len) |
8164 | { | |
2ff7e61e | 8165 | return __btrfs_free_reserved_extent(fs_info, start, len, 1, 0); |
e688b725 CM |
8166 | } |
8167 | ||
5d4f98a2 | 8168 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
8169 | u64 parent, u64 root_objectid, |
8170 | u64 flags, u64 owner, u64 offset, | |
8171 | struct btrfs_key *ins, int ref_mod) | |
e6dcd2dc | 8172 | { |
ef89b824 | 8173 | struct btrfs_fs_info *fs_info = trans->fs_info; |
e6dcd2dc | 8174 | int ret; |
e6dcd2dc | 8175 | struct btrfs_extent_item *extent_item; |
5d4f98a2 | 8176 | struct btrfs_extent_inline_ref *iref; |
e6dcd2dc | 8177 | struct btrfs_path *path; |
5d4f98a2 YZ |
8178 | struct extent_buffer *leaf; |
8179 | int type; | |
8180 | u32 size; | |
26b8003f | 8181 | |
5d4f98a2 YZ |
8182 | if (parent > 0) |
8183 | type = BTRFS_SHARED_DATA_REF_KEY; | |
8184 | else | |
8185 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
58176a96 | 8186 | |
5d4f98a2 | 8187 | size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type); |
7bb86316 CM |
8188 | |
8189 | path = btrfs_alloc_path(); | |
db5b493a TI |
8190 | if (!path) |
8191 | return -ENOMEM; | |
47e4bb98 | 8192 | |
b9473439 | 8193 | path->leave_spinning = 1; |
5d4f98a2 YZ |
8194 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, |
8195 | ins, size); | |
79787eaa JM |
8196 | if (ret) { |
8197 | btrfs_free_path(path); | |
8198 | return ret; | |
8199 | } | |
0f9dd46c | 8200 | |
5d4f98a2 YZ |
8201 | leaf = path->nodes[0]; |
8202 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
47e4bb98 | 8203 | struct btrfs_extent_item); |
5d4f98a2 YZ |
8204 | btrfs_set_extent_refs(leaf, extent_item, ref_mod); |
8205 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
8206 | btrfs_set_extent_flags(leaf, extent_item, | |
8207 | flags | BTRFS_EXTENT_FLAG_DATA); | |
8208 | ||
8209 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
8210 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
8211 | if (parent > 0) { | |
8212 | struct btrfs_shared_data_ref *ref; | |
8213 | ref = (struct btrfs_shared_data_ref *)(iref + 1); | |
8214 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
8215 | btrfs_set_shared_data_ref_count(leaf, ref, ref_mod); | |
8216 | } else { | |
8217 | struct btrfs_extent_data_ref *ref; | |
8218 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
8219 | btrfs_set_extent_data_ref_root(leaf, ref, root_objectid); | |
8220 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
8221 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
8222 | btrfs_set_extent_data_ref_count(leaf, ref, ref_mod); | |
8223 | } | |
47e4bb98 CM |
8224 | |
8225 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
7bb86316 | 8226 | btrfs_free_path(path); |
f510cfec | 8227 | |
25a356d3 | 8228 | ret = remove_from_free_space_tree(trans, ins->objectid, ins->offset); |
1e144fb8 OS |
8229 | if (ret) |
8230 | return ret; | |
8231 | ||
6b279408 | 8232 | ret = update_block_group(trans, ins->objectid, ins->offset, 1); |
79787eaa | 8233 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb | 8234 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
c1c9ff7c | 8235 | ins->objectid, ins->offset); |
f5947066 CM |
8236 | BUG(); |
8237 | } | |
71ff6437 | 8238 | trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid, ins->offset); |
e6dcd2dc CM |
8239 | return ret; |
8240 | } | |
8241 | ||
5d4f98a2 | 8242 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, |
4e6bd4e0 | 8243 | struct btrfs_delayed_ref_node *node, |
21ebfbe7 | 8244 | struct btrfs_delayed_extent_op *extent_op) |
e6dcd2dc | 8245 | { |
9dcdbe01 | 8246 | struct btrfs_fs_info *fs_info = trans->fs_info; |
e6dcd2dc | 8247 | int ret; |
5d4f98a2 | 8248 | struct btrfs_extent_item *extent_item; |
4e6bd4e0 | 8249 | struct btrfs_key extent_key; |
5d4f98a2 YZ |
8250 | struct btrfs_tree_block_info *block_info; |
8251 | struct btrfs_extent_inline_ref *iref; | |
8252 | struct btrfs_path *path; | |
8253 | struct extent_buffer *leaf; | |
4e6bd4e0 | 8254 | struct btrfs_delayed_tree_ref *ref; |
3173a18f | 8255 | u32 size = sizeof(*extent_item) + sizeof(*iref); |
4e6bd4e0 | 8256 | u64 num_bytes; |
21ebfbe7 | 8257 | u64 flags = extent_op->flags_to_set; |
0b246afa | 8258 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); |
3173a18f | 8259 | |
4e6bd4e0 NB |
8260 | ref = btrfs_delayed_node_to_tree_ref(node); |
8261 | ||
4e6bd4e0 NB |
8262 | extent_key.objectid = node->bytenr; |
8263 | if (skinny_metadata) { | |
8264 | extent_key.offset = ref->level; | |
8265 | extent_key.type = BTRFS_METADATA_ITEM_KEY; | |
8266 | num_bytes = fs_info->nodesize; | |
8267 | } else { | |
8268 | extent_key.offset = node->num_bytes; | |
8269 | extent_key.type = BTRFS_EXTENT_ITEM_KEY; | |
3173a18f | 8270 | size += sizeof(*block_info); |
4e6bd4e0 NB |
8271 | num_bytes = node->num_bytes; |
8272 | } | |
1c2308f8 | 8273 | |
5d4f98a2 | 8274 | path = btrfs_alloc_path(); |
80ee54bf | 8275 | if (!path) |
d8926bb3 | 8276 | return -ENOMEM; |
56bec294 | 8277 | |
5d4f98a2 YZ |
8278 | path->leave_spinning = 1; |
8279 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, | |
4e6bd4e0 | 8280 | &extent_key, size); |
79787eaa | 8281 | if (ret) { |
dd825259 | 8282 | btrfs_free_path(path); |
79787eaa JM |
8283 | return ret; |
8284 | } | |
5d4f98a2 YZ |
8285 | |
8286 | leaf = path->nodes[0]; | |
8287 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
8288 | struct btrfs_extent_item); | |
8289 | btrfs_set_extent_refs(leaf, extent_item, 1); | |
8290 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
8291 | btrfs_set_extent_flags(leaf, extent_item, | |
8292 | flags | BTRFS_EXTENT_FLAG_TREE_BLOCK); | |
5d4f98a2 | 8293 | |
3173a18f JB |
8294 | if (skinny_metadata) { |
8295 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
8296 | } else { | |
8297 | block_info = (struct btrfs_tree_block_info *)(extent_item + 1); | |
21ebfbe7 | 8298 | btrfs_set_tree_block_key(leaf, block_info, &extent_op->key); |
4e6bd4e0 | 8299 | btrfs_set_tree_block_level(leaf, block_info, ref->level); |
3173a18f JB |
8300 | iref = (struct btrfs_extent_inline_ref *)(block_info + 1); |
8301 | } | |
5d4f98a2 | 8302 | |
d4b20733 | 8303 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) { |
5d4f98a2 YZ |
8304 | BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); |
8305 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
8306 | BTRFS_SHARED_BLOCK_REF_KEY); | |
d4b20733 | 8307 | btrfs_set_extent_inline_ref_offset(leaf, iref, ref->parent); |
5d4f98a2 YZ |
8308 | } else { |
8309 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
8310 | BTRFS_TREE_BLOCK_REF_KEY); | |
4e6bd4e0 | 8311 | btrfs_set_extent_inline_ref_offset(leaf, iref, ref->root); |
5d4f98a2 YZ |
8312 | } |
8313 | ||
8314 | btrfs_mark_buffer_dirty(leaf); | |
8315 | btrfs_free_path(path); | |
8316 | ||
4e6bd4e0 NB |
8317 | ret = remove_from_free_space_tree(trans, extent_key.objectid, |
8318 | num_bytes); | |
1e144fb8 OS |
8319 | if (ret) |
8320 | return ret; | |
8321 | ||
6b279408 | 8322 | ret = update_block_group(trans, extent_key.objectid, |
6202df69 | 8323 | fs_info->nodesize, 1); |
79787eaa | 8324 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb | 8325 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
4e6bd4e0 | 8326 | extent_key.objectid, extent_key.offset); |
5d4f98a2 YZ |
8327 | BUG(); |
8328 | } | |
0be5dc67 | 8329 | |
4e6bd4e0 | 8330 | trace_btrfs_reserved_extent_alloc(fs_info, extent_key.objectid, |
0b246afa | 8331 | fs_info->nodesize); |
5d4f98a2 YZ |
8332 | return ret; |
8333 | } | |
8334 | ||
8335 | int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
84f7d8e6 | 8336 | struct btrfs_root *root, u64 owner, |
5846a3c2 QW |
8337 | u64 offset, u64 ram_bytes, |
8338 | struct btrfs_key *ins) | |
5d4f98a2 | 8339 | { |
76675593 | 8340 | struct btrfs_ref generic_ref = { 0 }; |
5d4f98a2 YZ |
8341 | int ret; |
8342 | ||
84f7d8e6 | 8343 | BUG_ON(root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID); |
5d4f98a2 | 8344 | |
76675593 QW |
8345 | btrfs_init_generic_ref(&generic_ref, BTRFS_ADD_DELAYED_EXTENT, |
8346 | ins->objectid, ins->offset, 0); | |
8347 | btrfs_init_data_ref(&generic_ref, root->root_key.objectid, owner, offset); | |
8a5040f7 | 8348 | btrfs_ref_tree_mod(root->fs_info, &generic_ref); |
76675593 QW |
8349 | ret = btrfs_add_delayed_data_ref(trans, &generic_ref, |
8350 | ram_bytes, NULL, NULL); | |
e6dcd2dc CM |
8351 | return ret; |
8352 | } | |
e02119d5 CM |
8353 | |
8354 | /* | |
8355 | * this is used by the tree logging recovery code. It records that | |
8356 | * an extent has been allocated and makes sure to clear the free | |
8357 | * space cache bits as well | |
8358 | */ | |
5d4f98a2 | 8359 | int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
8360 | u64 root_objectid, u64 owner, u64 offset, |
8361 | struct btrfs_key *ins) | |
e02119d5 | 8362 | { |
61da2abf | 8363 | struct btrfs_fs_info *fs_info = trans->fs_info; |
e02119d5 CM |
8364 | int ret; |
8365 | struct btrfs_block_group_cache *block_group; | |
ed7a6948 | 8366 | struct btrfs_space_info *space_info; |
11833d66 | 8367 | |
8c2a1a30 JB |
8368 | /* |
8369 | * Mixed block groups will exclude before processing the log so we only | |
01327610 | 8370 | * need to do the exclude dance if this fs isn't mixed. |
8c2a1a30 | 8371 | */ |
0b246afa | 8372 | if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) { |
2ff7e61e JM |
8373 | ret = __exclude_logged_extent(fs_info, ins->objectid, |
8374 | ins->offset); | |
b50c6e25 | 8375 | if (ret) |
8c2a1a30 | 8376 | return ret; |
11833d66 YZ |
8377 | } |
8378 | ||
0b246afa | 8379 | block_group = btrfs_lookup_block_group(fs_info, ins->objectid); |
8c2a1a30 JB |
8380 | if (!block_group) |
8381 | return -EINVAL; | |
8382 | ||
ed7a6948 WX |
8383 | space_info = block_group->space_info; |
8384 | spin_lock(&space_info->lock); | |
8385 | spin_lock(&block_group->lock); | |
8386 | space_info->bytes_reserved += ins->offset; | |
8387 | block_group->reserved += ins->offset; | |
8388 | spin_unlock(&block_group->lock); | |
8389 | spin_unlock(&space_info->lock); | |
8390 | ||
ef89b824 NB |
8391 | ret = alloc_reserved_file_extent(trans, 0, root_objectid, 0, owner, |
8392 | offset, ins, 1); | |
b50c6e25 | 8393 | btrfs_put_block_group(block_group); |
e02119d5 CM |
8394 | return ret; |
8395 | } | |
8396 | ||
48a3b636 ES |
8397 | static struct extent_buffer * |
8398 | btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
bc877d28 | 8399 | u64 bytenr, int level, u64 owner) |
65b51a00 | 8400 | { |
0b246afa | 8401 | struct btrfs_fs_info *fs_info = root->fs_info; |
65b51a00 CM |
8402 | struct extent_buffer *buf; |
8403 | ||
2ff7e61e | 8404 | buf = btrfs_find_create_tree_block(fs_info, bytenr); |
c871b0f2 LB |
8405 | if (IS_ERR(buf)) |
8406 | return buf; | |
8407 | ||
b72c3aba QW |
8408 | /* |
8409 | * Extra safety check in case the extent tree is corrupted and extent | |
8410 | * allocator chooses to use a tree block which is already used and | |
8411 | * locked. | |
8412 | */ | |
8413 | if (buf->lock_owner == current->pid) { | |
8414 | btrfs_err_rl(fs_info, | |
8415 | "tree block %llu owner %llu already locked by pid=%d, extent tree corruption detected", | |
8416 | buf->start, btrfs_header_owner(buf), current->pid); | |
8417 | free_extent_buffer(buf); | |
8418 | return ERR_PTR(-EUCLEAN); | |
8419 | } | |
8420 | ||
85d4e461 | 8421 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level); |
65b51a00 | 8422 | btrfs_tree_lock(buf); |
6a884d7d | 8423 | btrfs_clean_tree_block(buf); |
3083ee2e | 8424 | clear_bit(EXTENT_BUFFER_STALE, &buf->bflags); |
b4ce94de | 8425 | |
8bead258 | 8426 | btrfs_set_lock_blocking_write(buf); |
4db8c528 | 8427 | set_extent_buffer_uptodate(buf); |
b4ce94de | 8428 | |
bc877d28 NB |
8429 | memzero_extent_buffer(buf, 0, sizeof(struct btrfs_header)); |
8430 | btrfs_set_header_level(buf, level); | |
8431 | btrfs_set_header_bytenr(buf, buf->start); | |
8432 | btrfs_set_header_generation(buf, trans->transid); | |
8433 | btrfs_set_header_backref_rev(buf, BTRFS_MIXED_BACKREF_REV); | |
8434 | btrfs_set_header_owner(buf, owner); | |
de37aa51 | 8435 | write_extent_buffer_fsid(buf, fs_info->fs_devices->metadata_uuid); |
bc877d28 | 8436 | write_extent_buffer_chunk_tree_uuid(buf, fs_info->chunk_tree_uuid); |
d0c803c4 | 8437 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { |
656f30db | 8438 | buf->log_index = root->log_transid % 2; |
8cef4e16 YZ |
8439 | /* |
8440 | * we allow two log transactions at a time, use different | |
52042d8e | 8441 | * EXTENT bit to differentiate dirty pages. |
8cef4e16 | 8442 | */ |
656f30db | 8443 | if (buf->log_index == 0) |
8cef4e16 YZ |
8444 | set_extent_dirty(&root->dirty_log_pages, buf->start, |
8445 | buf->start + buf->len - 1, GFP_NOFS); | |
8446 | else | |
8447 | set_extent_new(&root->dirty_log_pages, buf->start, | |
3744dbeb | 8448 | buf->start + buf->len - 1); |
d0c803c4 | 8449 | } else { |
656f30db | 8450 | buf->log_index = -1; |
d0c803c4 | 8451 | set_extent_dirty(&trans->transaction->dirty_pages, buf->start, |
65b51a00 | 8452 | buf->start + buf->len - 1, GFP_NOFS); |
d0c803c4 | 8453 | } |
64c12921 | 8454 | trans->dirty = true; |
b4ce94de | 8455 | /* this returns a buffer locked for blocking */ |
65b51a00 CM |
8456 | return buf; |
8457 | } | |
8458 | ||
f0486c68 YZ |
8459 | static struct btrfs_block_rsv * |
8460 | use_block_rsv(struct btrfs_trans_handle *trans, | |
8461 | struct btrfs_root *root, u32 blocksize) | |
8462 | { | |
0b246afa | 8463 | struct btrfs_fs_info *fs_info = root->fs_info; |
f0486c68 | 8464 | struct btrfs_block_rsv *block_rsv; |
0b246afa | 8465 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; |
f0486c68 | 8466 | int ret; |
d88033db | 8467 | bool global_updated = false; |
f0486c68 YZ |
8468 | |
8469 | block_rsv = get_block_rsv(trans, root); | |
8470 | ||
b586b323 MX |
8471 | if (unlikely(block_rsv->size == 0)) |
8472 | goto try_reserve; | |
d88033db | 8473 | again: |
f0486c68 YZ |
8474 | ret = block_rsv_use_bytes(block_rsv, blocksize); |
8475 | if (!ret) | |
8476 | return block_rsv; | |
8477 | ||
b586b323 MX |
8478 | if (block_rsv->failfast) |
8479 | return ERR_PTR(ret); | |
8480 | ||
d88033db MX |
8481 | if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) { |
8482 | global_updated = true; | |
0b246afa | 8483 | update_global_block_rsv(fs_info); |
d88033db MX |
8484 | goto again; |
8485 | } | |
8486 | ||
ba2c4d4e JB |
8487 | /* |
8488 | * The global reserve still exists to save us from ourselves, so don't | |
8489 | * warn_on if we are short on our delayed refs reserve. | |
8490 | */ | |
8491 | if (block_rsv->type != BTRFS_BLOCK_RSV_DELREFS && | |
8492 | btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { | |
b586b323 MX |
8493 | static DEFINE_RATELIMIT_STATE(_rs, |
8494 | DEFAULT_RATELIMIT_INTERVAL * 10, | |
8495 | /*DEFAULT_RATELIMIT_BURST*/ 1); | |
8496 | if (__ratelimit(&_rs)) | |
8497 | WARN(1, KERN_DEBUG | |
efe120a0 | 8498 | "BTRFS: block rsv returned %d\n", ret); |
b586b323 MX |
8499 | } |
8500 | try_reserve: | |
8501 | ret = reserve_metadata_bytes(root, block_rsv, blocksize, | |
8502 | BTRFS_RESERVE_NO_FLUSH); | |
8503 | if (!ret) | |
8504 | return block_rsv; | |
8505 | /* | |
8506 | * If we couldn't reserve metadata bytes try and use some from | |
5881cfc9 MX |
8507 | * the global reserve if its space type is the same as the global |
8508 | * reservation. | |
b586b323 | 8509 | */ |
5881cfc9 MX |
8510 | if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL && |
8511 | block_rsv->space_info == global_rsv->space_info) { | |
b586b323 MX |
8512 | ret = block_rsv_use_bytes(global_rsv, blocksize); |
8513 | if (!ret) | |
8514 | return global_rsv; | |
8515 | } | |
8516 | return ERR_PTR(ret); | |
f0486c68 YZ |
8517 | } |
8518 | ||
8c2a3ca2 JB |
8519 | static void unuse_block_rsv(struct btrfs_fs_info *fs_info, |
8520 | struct btrfs_block_rsv *block_rsv, u32 blocksize) | |
f0486c68 | 8521 | { |
3a584174 | 8522 | block_rsv_add_bytes(block_rsv, blocksize, false); |
ff6bc37e | 8523 | block_rsv_release_bytes(fs_info, block_rsv, NULL, 0, NULL); |
f0486c68 YZ |
8524 | } |
8525 | ||
fec577fb | 8526 | /* |
f0486c68 | 8527 | * finds a free extent and does all the dirty work required for allocation |
67b7859e | 8528 | * returns the tree buffer or an ERR_PTR on error. |
fec577fb | 8529 | */ |
4d75f8a9 | 8530 | struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans, |
310712b2 OS |
8531 | struct btrfs_root *root, |
8532 | u64 parent, u64 root_objectid, | |
8533 | const struct btrfs_disk_key *key, | |
8534 | int level, u64 hint, | |
8535 | u64 empty_size) | |
fec577fb | 8536 | { |
0b246afa | 8537 | struct btrfs_fs_info *fs_info = root->fs_info; |
e2fa7227 | 8538 | struct btrfs_key ins; |
f0486c68 | 8539 | struct btrfs_block_rsv *block_rsv; |
5f39d397 | 8540 | struct extent_buffer *buf; |
67b7859e | 8541 | struct btrfs_delayed_extent_op *extent_op; |
ed4f255b | 8542 | struct btrfs_ref generic_ref = { 0 }; |
f0486c68 YZ |
8543 | u64 flags = 0; |
8544 | int ret; | |
0b246afa JM |
8545 | u32 blocksize = fs_info->nodesize; |
8546 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); | |
fec577fb | 8547 | |
05653ef3 | 8548 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
0b246afa | 8549 | if (btrfs_is_testing(fs_info)) { |
faa2dbf0 | 8550 | buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr, |
bc877d28 | 8551 | level, root_objectid); |
faa2dbf0 JB |
8552 | if (!IS_ERR(buf)) |
8553 | root->alloc_bytenr += blocksize; | |
8554 | return buf; | |
8555 | } | |
05653ef3 | 8556 | #endif |
fccb84c9 | 8557 | |
f0486c68 YZ |
8558 | block_rsv = use_block_rsv(trans, root, blocksize); |
8559 | if (IS_ERR(block_rsv)) | |
8560 | return ERR_CAST(block_rsv); | |
8561 | ||
18513091 | 8562 | ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize, |
e570fd27 | 8563 | empty_size, hint, &ins, 0, 0); |
67b7859e OS |
8564 | if (ret) |
8565 | goto out_unuse; | |
55c69072 | 8566 | |
bc877d28 NB |
8567 | buf = btrfs_init_new_buffer(trans, root, ins.objectid, level, |
8568 | root_objectid); | |
67b7859e OS |
8569 | if (IS_ERR(buf)) { |
8570 | ret = PTR_ERR(buf); | |
8571 | goto out_free_reserved; | |
8572 | } | |
f0486c68 YZ |
8573 | |
8574 | if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
8575 | if (parent == 0) | |
8576 | parent = ins.objectid; | |
8577 | flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
8578 | } else | |
8579 | BUG_ON(parent > 0); | |
8580 | ||
8581 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) { | |
78a6184a | 8582 | extent_op = btrfs_alloc_delayed_extent_op(); |
67b7859e OS |
8583 | if (!extent_op) { |
8584 | ret = -ENOMEM; | |
8585 | goto out_free_buf; | |
8586 | } | |
f0486c68 YZ |
8587 | if (key) |
8588 | memcpy(&extent_op->key, key, sizeof(extent_op->key)); | |
8589 | else | |
8590 | memset(&extent_op->key, 0, sizeof(extent_op->key)); | |
8591 | extent_op->flags_to_set = flags; | |
35b3ad50 DS |
8592 | extent_op->update_key = skinny_metadata ? false : true; |
8593 | extent_op->update_flags = true; | |
8594 | extent_op->is_data = false; | |
b1c79e09 | 8595 | extent_op->level = level; |
f0486c68 | 8596 | |
ed4f255b QW |
8597 | btrfs_init_generic_ref(&generic_ref, BTRFS_ADD_DELAYED_EXTENT, |
8598 | ins.objectid, ins.offset, parent); | |
8599 | generic_ref.real_root = root->root_key.objectid; | |
8600 | btrfs_init_tree_ref(&generic_ref, level, root_objectid); | |
8a5040f7 | 8601 | btrfs_ref_tree_mod(fs_info, &generic_ref); |
ed4f255b | 8602 | ret = btrfs_add_delayed_tree_ref(trans, &generic_ref, |
7be07912 | 8603 | extent_op, NULL, NULL); |
67b7859e OS |
8604 | if (ret) |
8605 | goto out_free_delayed; | |
f0486c68 | 8606 | } |
fec577fb | 8607 | return buf; |
67b7859e OS |
8608 | |
8609 | out_free_delayed: | |
8610 | btrfs_free_delayed_extent_op(extent_op); | |
8611 | out_free_buf: | |
8612 | free_extent_buffer(buf); | |
8613 | out_free_reserved: | |
2ff7e61e | 8614 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0); |
67b7859e | 8615 | out_unuse: |
0b246afa | 8616 | unuse_block_rsv(fs_info, block_rsv, blocksize); |
67b7859e | 8617 | return ERR_PTR(ret); |
fec577fb | 8618 | } |
a28ec197 | 8619 | |
2c47e605 YZ |
8620 | struct walk_control { |
8621 | u64 refs[BTRFS_MAX_LEVEL]; | |
8622 | u64 flags[BTRFS_MAX_LEVEL]; | |
8623 | struct btrfs_key update_progress; | |
aea6f028 JB |
8624 | struct btrfs_key drop_progress; |
8625 | int drop_level; | |
2c47e605 YZ |
8626 | int stage; |
8627 | int level; | |
8628 | int shared_level; | |
8629 | int update_ref; | |
8630 | int keep_locks; | |
1c4850e2 YZ |
8631 | int reada_slot; |
8632 | int reada_count; | |
78c52d9e | 8633 | int restarted; |
2c47e605 YZ |
8634 | }; |
8635 | ||
8636 | #define DROP_REFERENCE 1 | |
8637 | #define UPDATE_BACKREF 2 | |
8638 | ||
1c4850e2 YZ |
8639 | static noinline void reada_walk_down(struct btrfs_trans_handle *trans, |
8640 | struct btrfs_root *root, | |
8641 | struct walk_control *wc, | |
8642 | struct btrfs_path *path) | |
6407bf6d | 8643 | { |
0b246afa | 8644 | struct btrfs_fs_info *fs_info = root->fs_info; |
1c4850e2 YZ |
8645 | u64 bytenr; |
8646 | u64 generation; | |
8647 | u64 refs; | |
94fcca9f | 8648 | u64 flags; |
5d4f98a2 | 8649 | u32 nritems; |
1c4850e2 YZ |
8650 | struct btrfs_key key; |
8651 | struct extent_buffer *eb; | |
6407bf6d | 8652 | int ret; |
1c4850e2 YZ |
8653 | int slot; |
8654 | int nread = 0; | |
6407bf6d | 8655 | |
1c4850e2 YZ |
8656 | if (path->slots[wc->level] < wc->reada_slot) { |
8657 | wc->reada_count = wc->reada_count * 2 / 3; | |
8658 | wc->reada_count = max(wc->reada_count, 2); | |
8659 | } else { | |
8660 | wc->reada_count = wc->reada_count * 3 / 2; | |
8661 | wc->reada_count = min_t(int, wc->reada_count, | |
0b246afa | 8662 | BTRFS_NODEPTRS_PER_BLOCK(fs_info)); |
1c4850e2 | 8663 | } |
7bb86316 | 8664 | |
1c4850e2 YZ |
8665 | eb = path->nodes[wc->level]; |
8666 | nritems = btrfs_header_nritems(eb); | |
bd56b302 | 8667 | |
1c4850e2 YZ |
8668 | for (slot = path->slots[wc->level]; slot < nritems; slot++) { |
8669 | if (nread >= wc->reada_count) | |
8670 | break; | |
bd56b302 | 8671 | |
2dd3e67b | 8672 | cond_resched(); |
1c4850e2 YZ |
8673 | bytenr = btrfs_node_blockptr(eb, slot); |
8674 | generation = btrfs_node_ptr_generation(eb, slot); | |
2dd3e67b | 8675 | |
1c4850e2 YZ |
8676 | if (slot == path->slots[wc->level]) |
8677 | goto reada; | |
5d4f98a2 | 8678 | |
1c4850e2 YZ |
8679 | if (wc->stage == UPDATE_BACKREF && |
8680 | generation <= root->root_key.offset) | |
bd56b302 CM |
8681 | continue; |
8682 | ||
94fcca9f | 8683 | /* We don't lock the tree block, it's OK to be racy here */ |
2ff7e61e | 8684 | ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, |
3173a18f JB |
8685 | wc->level - 1, 1, &refs, |
8686 | &flags); | |
79787eaa JM |
8687 | /* We don't care about errors in readahead. */ |
8688 | if (ret < 0) | |
8689 | continue; | |
94fcca9f YZ |
8690 | BUG_ON(refs == 0); |
8691 | ||
1c4850e2 | 8692 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 YZ |
8693 | if (refs == 1) |
8694 | goto reada; | |
bd56b302 | 8695 | |
94fcca9f YZ |
8696 | if (wc->level == 1 && |
8697 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8698 | continue; | |
1c4850e2 YZ |
8699 | if (!wc->update_ref || |
8700 | generation <= root->root_key.offset) | |
8701 | continue; | |
8702 | btrfs_node_key_to_cpu(eb, &key, slot); | |
8703 | ret = btrfs_comp_cpu_keys(&key, | |
8704 | &wc->update_progress); | |
8705 | if (ret < 0) | |
8706 | continue; | |
94fcca9f YZ |
8707 | } else { |
8708 | if (wc->level == 1 && | |
8709 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8710 | continue; | |
6407bf6d | 8711 | } |
1c4850e2 | 8712 | reada: |
2ff7e61e | 8713 | readahead_tree_block(fs_info, bytenr); |
1c4850e2 | 8714 | nread++; |
20524f02 | 8715 | } |
1c4850e2 | 8716 | wc->reada_slot = slot; |
20524f02 | 8717 | } |
2c47e605 | 8718 | |
f82d02d9 | 8719 | /* |
2c016dc2 | 8720 | * helper to process tree block while walking down the tree. |
2c47e605 | 8721 | * |
2c47e605 YZ |
8722 | * when wc->stage == UPDATE_BACKREF, this function updates |
8723 | * back refs for pointers in the block. | |
8724 | * | |
8725 | * NOTE: return value 1 means we should stop walking down. | |
f82d02d9 | 8726 | */ |
2c47e605 | 8727 | static noinline int walk_down_proc(struct btrfs_trans_handle *trans, |
5d4f98a2 | 8728 | struct btrfs_root *root, |
2c47e605 | 8729 | struct btrfs_path *path, |
94fcca9f | 8730 | struct walk_control *wc, int lookup_info) |
f82d02d9 | 8731 | { |
2ff7e61e | 8732 | struct btrfs_fs_info *fs_info = root->fs_info; |
2c47e605 YZ |
8733 | int level = wc->level; |
8734 | struct extent_buffer *eb = path->nodes[level]; | |
2c47e605 | 8735 | u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF; |
f82d02d9 YZ |
8736 | int ret; |
8737 | ||
2c47e605 YZ |
8738 | if (wc->stage == UPDATE_BACKREF && |
8739 | btrfs_header_owner(eb) != root->root_key.objectid) | |
8740 | return 1; | |
f82d02d9 | 8741 | |
2c47e605 YZ |
8742 | /* |
8743 | * when reference count of tree block is 1, it won't increase | |
8744 | * again. once full backref flag is set, we never clear it. | |
8745 | */ | |
94fcca9f YZ |
8746 | if (lookup_info && |
8747 | ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) || | |
8748 | (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) { | |
2c47e605 | 8749 | BUG_ON(!path->locks[level]); |
2ff7e61e | 8750 | ret = btrfs_lookup_extent_info(trans, fs_info, |
3173a18f | 8751 | eb->start, level, 1, |
2c47e605 YZ |
8752 | &wc->refs[level], |
8753 | &wc->flags[level]); | |
79787eaa JM |
8754 | BUG_ON(ret == -ENOMEM); |
8755 | if (ret) | |
8756 | return ret; | |
2c47e605 YZ |
8757 | BUG_ON(wc->refs[level] == 0); |
8758 | } | |
5d4f98a2 | 8759 | |
2c47e605 YZ |
8760 | if (wc->stage == DROP_REFERENCE) { |
8761 | if (wc->refs[level] > 1) | |
8762 | return 1; | |
f82d02d9 | 8763 | |
2c47e605 | 8764 | if (path->locks[level] && !wc->keep_locks) { |
bd681513 | 8765 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
8766 | path->locks[level] = 0; |
8767 | } | |
8768 | return 0; | |
8769 | } | |
f82d02d9 | 8770 | |
2c47e605 YZ |
8771 | /* wc->stage == UPDATE_BACKREF */ |
8772 | if (!(wc->flags[level] & flag)) { | |
8773 | BUG_ON(!path->locks[level]); | |
e339a6b0 | 8774 | ret = btrfs_inc_ref(trans, root, eb, 1); |
79787eaa | 8775 | BUG_ON(ret); /* -ENOMEM */ |
e339a6b0 | 8776 | ret = btrfs_dec_ref(trans, root, eb, 0); |
79787eaa | 8777 | BUG_ON(ret); /* -ENOMEM */ |
f5c8daa5 | 8778 | ret = btrfs_set_disk_extent_flags(trans, eb->start, |
b1c79e09 JB |
8779 | eb->len, flag, |
8780 | btrfs_header_level(eb), 0); | |
79787eaa | 8781 | BUG_ON(ret); /* -ENOMEM */ |
2c47e605 YZ |
8782 | wc->flags[level] |= flag; |
8783 | } | |
8784 | ||
8785 | /* | |
8786 | * the block is shared by multiple trees, so it's not good to | |
8787 | * keep the tree lock | |
8788 | */ | |
8789 | if (path->locks[level] && level > 0) { | |
bd681513 | 8790 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
8791 | path->locks[level] = 0; |
8792 | } | |
8793 | return 0; | |
8794 | } | |
8795 | ||
78c52d9e JB |
8796 | /* |
8797 | * This is used to verify a ref exists for this root to deal with a bug where we | |
8798 | * would have a drop_progress key that hadn't been updated properly. | |
8799 | */ | |
8800 | static int check_ref_exists(struct btrfs_trans_handle *trans, | |
8801 | struct btrfs_root *root, u64 bytenr, u64 parent, | |
8802 | int level) | |
8803 | { | |
8804 | struct btrfs_path *path; | |
8805 | struct btrfs_extent_inline_ref *iref; | |
8806 | int ret; | |
8807 | ||
8808 | path = btrfs_alloc_path(); | |
8809 | if (!path) | |
8810 | return -ENOMEM; | |
8811 | ||
8812 | ret = lookup_extent_backref(trans, path, &iref, bytenr, | |
8813 | root->fs_info->nodesize, parent, | |
8814 | root->root_key.objectid, level, 0); | |
8815 | btrfs_free_path(path); | |
8816 | if (ret == -ENOENT) | |
8817 | return 0; | |
8818 | if (ret < 0) | |
8819 | return ret; | |
8820 | return 1; | |
8821 | } | |
8822 | ||
1c4850e2 | 8823 | /* |
2c016dc2 | 8824 | * helper to process tree block pointer. |
1c4850e2 YZ |
8825 | * |
8826 | * when wc->stage == DROP_REFERENCE, this function checks | |
8827 | * reference count of the block pointed to. if the block | |
8828 | * is shared and we need update back refs for the subtree | |
8829 | * rooted at the block, this function changes wc->stage to | |
8830 | * UPDATE_BACKREF. if the block is shared and there is no | |
8831 | * need to update back, this function drops the reference | |
8832 | * to the block. | |
8833 | * | |
8834 | * NOTE: return value 1 means we should stop walking down. | |
8835 | */ | |
8836 | static noinline int do_walk_down(struct btrfs_trans_handle *trans, | |
8837 | struct btrfs_root *root, | |
8838 | struct btrfs_path *path, | |
94fcca9f | 8839 | struct walk_control *wc, int *lookup_info) |
1c4850e2 | 8840 | { |
0b246afa | 8841 | struct btrfs_fs_info *fs_info = root->fs_info; |
1c4850e2 YZ |
8842 | u64 bytenr; |
8843 | u64 generation; | |
8844 | u64 parent; | |
1c4850e2 | 8845 | struct btrfs_key key; |
581c1760 | 8846 | struct btrfs_key first_key; |
ffd4bb2a | 8847 | struct btrfs_ref ref = { 0 }; |
1c4850e2 YZ |
8848 | struct extent_buffer *next; |
8849 | int level = wc->level; | |
8850 | int reada = 0; | |
8851 | int ret = 0; | |
1152651a | 8852 | bool need_account = false; |
1c4850e2 YZ |
8853 | |
8854 | generation = btrfs_node_ptr_generation(path->nodes[level], | |
8855 | path->slots[level]); | |
8856 | /* | |
8857 | * if the lower level block was created before the snapshot | |
8858 | * was created, we know there is no need to update back refs | |
8859 | * for the subtree | |
8860 | */ | |
8861 | if (wc->stage == UPDATE_BACKREF && | |
94fcca9f YZ |
8862 | generation <= root->root_key.offset) { |
8863 | *lookup_info = 1; | |
1c4850e2 | 8864 | return 1; |
94fcca9f | 8865 | } |
1c4850e2 YZ |
8866 | |
8867 | bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]); | |
581c1760 QW |
8868 | btrfs_node_key_to_cpu(path->nodes[level], &first_key, |
8869 | path->slots[level]); | |
1c4850e2 | 8870 | |
0b246afa | 8871 | next = find_extent_buffer(fs_info, bytenr); |
1c4850e2 | 8872 | if (!next) { |
2ff7e61e | 8873 | next = btrfs_find_create_tree_block(fs_info, bytenr); |
c871b0f2 LB |
8874 | if (IS_ERR(next)) |
8875 | return PTR_ERR(next); | |
8876 | ||
b2aaaa3b JB |
8877 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, next, |
8878 | level - 1); | |
1c4850e2 YZ |
8879 | reada = 1; |
8880 | } | |
8881 | btrfs_tree_lock(next); | |
8bead258 | 8882 | btrfs_set_lock_blocking_write(next); |
1c4850e2 | 8883 | |
2ff7e61e | 8884 | ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1, |
94fcca9f YZ |
8885 | &wc->refs[level - 1], |
8886 | &wc->flags[level - 1]); | |
4867268c JB |
8887 | if (ret < 0) |
8888 | goto out_unlock; | |
79787eaa | 8889 | |
c2cf52eb | 8890 | if (unlikely(wc->refs[level - 1] == 0)) { |
0b246afa | 8891 | btrfs_err(fs_info, "Missing references."); |
4867268c JB |
8892 | ret = -EIO; |
8893 | goto out_unlock; | |
c2cf52eb | 8894 | } |
94fcca9f | 8895 | *lookup_info = 0; |
1c4850e2 | 8896 | |
94fcca9f | 8897 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 | 8898 | if (wc->refs[level - 1] > 1) { |
1152651a | 8899 | need_account = true; |
94fcca9f YZ |
8900 | if (level == 1 && |
8901 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8902 | goto skip; | |
8903 | ||
1c4850e2 YZ |
8904 | if (!wc->update_ref || |
8905 | generation <= root->root_key.offset) | |
8906 | goto skip; | |
8907 | ||
8908 | btrfs_node_key_to_cpu(path->nodes[level], &key, | |
8909 | path->slots[level]); | |
8910 | ret = btrfs_comp_cpu_keys(&key, &wc->update_progress); | |
8911 | if (ret < 0) | |
8912 | goto skip; | |
8913 | ||
8914 | wc->stage = UPDATE_BACKREF; | |
8915 | wc->shared_level = level - 1; | |
8916 | } | |
94fcca9f YZ |
8917 | } else { |
8918 | if (level == 1 && | |
8919 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8920 | goto skip; | |
1c4850e2 YZ |
8921 | } |
8922 | ||
b9fab919 | 8923 | if (!btrfs_buffer_uptodate(next, generation, 0)) { |
1c4850e2 YZ |
8924 | btrfs_tree_unlock(next); |
8925 | free_extent_buffer(next); | |
8926 | next = NULL; | |
94fcca9f | 8927 | *lookup_info = 1; |
1c4850e2 YZ |
8928 | } |
8929 | ||
8930 | if (!next) { | |
8931 | if (reada && level == 1) | |
8932 | reada_walk_down(trans, root, wc, path); | |
581c1760 QW |
8933 | next = read_tree_block(fs_info, bytenr, generation, level - 1, |
8934 | &first_key); | |
64c043de LB |
8935 | if (IS_ERR(next)) { |
8936 | return PTR_ERR(next); | |
8937 | } else if (!extent_buffer_uptodate(next)) { | |
416bc658 | 8938 | free_extent_buffer(next); |
97d9a8a4 | 8939 | return -EIO; |
416bc658 | 8940 | } |
1c4850e2 | 8941 | btrfs_tree_lock(next); |
8bead258 | 8942 | btrfs_set_lock_blocking_write(next); |
1c4850e2 YZ |
8943 | } |
8944 | ||
8945 | level--; | |
4867268c JB |
8946 | ASSERT(level == btrfs_header_level(next)); |
8947 | if (level != btrfs_header_level(next)) { | |
8948 | btrfs_err(root->fs_info, "mismatched level"); | |
8949 | ret = -EIO; | |
8950 | goto out_unlock; | |
8951 | } | |
1c4850e2 YZ |
8952 | path->nodes[level] = next; |
8953 | path->slots[level] = 0; | |
bd681513 | 8954 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
1c4850e2 YZ |
8955 | wc->level = level; |
8956 | if (wc->level == 1) | |
8957 | wc->reada_slot = 0; | |
8958 | return 0; | |
8959 | skip: | |
8960 | wc->refs[level - 1] = 0; | |
8961 | wc->flags[level - 1] = 0; | |
94fcca9f YZ |
8962 | if (wc->stage == DROP_REFERENCE) { |
8963 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
8964 | parent = path->nodes[level]->start; | |
8965 | } else { | |
4867268c | 8966 | ASSERT(root->root_key.objectid == |
94fcca9f | 8967 | btrfs_header_owner(path->nodes[level])); |
4867268c JB |
8968 | if (root->root_key.objectid != |
8969 | btrfs_header_owner(path->nodes[level])) { | |
8970 | btrfs_err(root->fs_info, | |
8971 | "mismatched block owner"); | |
8972 | ret = -EIO; | |
8973 | goto out_unlock; | |
8974 | } | |
94fcca9f YZ |
8975 | parent = 0; |
8976 | } | |
1c4850e2 | 8977 | |
78c52d9e JB |
8978 | /* |
8979 | * If we had a drop_progress we need to verify the refs are set | |
8980 | * as expected. If we find our ref then we know that from here | |
8981 | * on out everything should be correct, and we can clear the | |
8982 | * ->restarted flag. | |
8983 | */ | |
8984 | if (wc->restarted) { | |
8985 | ret = check_ref_exists(trans, root, bytenr, parent, | |
8986 | level - 1); | |
8987 | if (ret < 0) | |
8988 | goto out_unlock; | |
8989 | if (ret == 0) | |
8990 | goto no_delete; | |
8991 | ret = 0; | |
8992 | wc->restarted = 0; | |
8993 | } | |
8994 | ||
2cd86d30 QW |
8995 | /* |
8996 | * Reloc tree doesn't contribute to qgroup numbers, and we have | |
8997 | * already accounted them at merge time (replace_path), | |
8998 | * thus we could skip expensive subtree trace here. | |
8999 | */ | |
9000 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID && | |
9001 | need_account) { | |
deb40627 | 9002 | ret = btrfs_qgroup_trace_subtree(trans, next, |
33d1f05c | 9003 | generation, level - 1); |
1152651a | 9004 | if (ret) { |
0b246afa | 9005 | btrfs_err_rl(fs_info, |
5d163e0e JM |
9006 | "Error %d accounting shared subtree. Quota is out of sync, rescan required.", |
9007 | ret); | |
1152651a MF |
9008 | } |
9009 | } | |
aea6f028 JB |
9010 | |
9011 | /* | |
9012 | * We need to update the next key in our walk control so we can | |
9013 | * update the drop_progress key accordingly. We don't care if | |
9014 | * find_next_key doesn't find a key because that means we're at | |
9015 | * the end and are going to clean up now. | |
9016 | */ | |
9017 | wc->drop_level = level; | |
9018 | find_next_key(path, level, &wc->drop_progress); | |
9019 | ||
ffd4bb2a QW |
9020 | btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr, |
9021 | fs_info->nodesize, parent); | |
9022 | btrfs_init_tree_ref(&ref, level - 1, root->root_key.objectid); | |
9023 | ret = btrfs_free_extent(trans, &ref); | |
4867268c JB |
9024 | if (ret) |
9025 | goto out_unlock; | |
1c4850e2 | 9026 | } |
78c52d9e | 9027 | no_delete: |
4867268c JB |
9028 | *lookup_info = 1; |
9029 | ret = 1; | |
9030 | ||
9031 | out_unlock: | |
1c4850e2 YZ |
9032 | btrfs_tree_unlock(next); |
9033 | free_extent_buffer(next); | |
4867268c JB |
9034 | |
9035 | return ret; | |
1c4850e2 YZ |
9036 | } |
9037 | ||
2c47e605 | 9038 | /* |
2c016dc2 | 9039 | * helper to process tree block while walking up the tree. |
2c47e605 YZ |
9040 | * |
9041 | * when wc->stage == DROP_REFERENCE, this function drops | |
9042 | * reference count on the block. | |
9043 | * | |
9044 | * when wc->stage == UPDATE_BACKREF, this function changes | |
9045 | * wc->stage back to DROP_REFERENCE if we changed wc->stage | |
9046 | * to UPDATE_BACKREF previously while processing the block. | |
9047 | * | |
9048 | * NOTE: return value 1 means we should stop walking up. | |
9049 | */ | |
9050 | static noinline int walk_up_proc(struct btrfs_trans_handle *trans, | |
9051 | struct btrfs_root *root, | |
9052 | struct btrfs_path *path, | |
9053 | struct walk_control *wc) | |
9054 | { | |
0b246afa | 9055 | struct btrfs_fs_info *fs_info = root->fs_info; |
f0486c68 | 9056 | int ret; |
2c47e605 YZ |
9057 | int level = wc->level; |
9058 | struct extent_buffer *eb = path->nodes[level]; | |
9059 | u64 parent = 0; | |
9060 | ||
9061 | if (wc->stage == UPDATE_BACKREF) { | |
9062 | BUG_ON(wc->shared_level < level); | |
9063 | if (level < wc->shared_level) | |
9064 | goto out; | |
9065 | ||
2c47e605 YZ |
9066 | ret = find_next_key(path, level + 1, &wc->update_progress); |
9067 | if (ret > 0) | |
9068 | wc->update_ref = 0; | |
9069 | ||
9070 | wc->stage = DROP_REFERENCE; | |
9071 | wc->shared_level = -1; | |
9072 | path->slots[level] = 0; | |
9073 | ||
9074 | /* | |
9075 | * check reference count again if the block isn't locked. | |
9076 | * we should start walking down the tree again if reference | |
9077 | * count is one. | |
9078 | */ | |
9079 | if (!path->locks[level]) { | |
9080 | BUG_ON(level == 0); | |
9081 | btrfs_tree_lock(eb); | |
8bead258 | 9082 | btrfs_set_lock_blocking_write(eb); |
bd681513 | 9083 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 | 9084 | |
2ff7e61e | 9085 | ret = btrfs_lookup_extent_info(trans, fs_info, |
3173a18f | 9086 | eb->start, level, 1, |
2c47e605 YZ |
9087 | &wc->refs[level], |
9088 | &wc->flags[level]); | |
79787eaa JM |
9089 | if (ret < 0) { |
9090 | btrfs_tree_unlock_rw(eb, path->locks[level]); | |
3268a246 | 9091 | path->locks[level] = 0; |
79787eaa JM |
9092 | return ret; |
9093 | } | |
2c47e605 YZ |
9094 | BUG_ON(wc->refs[level] == 0); |
9095 | if (wc->refs[level] == 1) { | |
bd681513 | 9096 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
3268a246 | 9097 | path->locks[level] = 0; |
2c47e605 YZ |
9098 | return 1; |
9099 | } | |
f82d02d9 | 9100 | } |
2c47e605 | 9101 | } |
f82d02d9 | 9102 | |
2c47e605 YZ |
9103 | /* wc->stage == DROP_REFERENCE */ |
9104 | BUG_ON(wc->refs[level] > 1 && !path->locks[level]); | |
5d4f98a2 | 9105 | |
2c47e605 YZ |
9106 | if (wc->refs[level] == 1) { |
9107 | if (level == 0) { | |
9108 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
e339a6b0 | 9109 | ret = btrfs_dec_ref(trans, root, eb, 1); |
2c47e605 | 9110 | else |
e339a6b0 | 9111 | ret = btrfs_dec_ref(trans, root, eb, 0); |
79787eaa | 9112 | BUG_ON(ret); /* -ENOMEM */ |
c4140cbf QW |
9113 | if (is_fstree(root->root_key.objectid)) { |
9114 | ret = btrfs_qgroup_trace_leaf_items(trans, eb); | |
9115 | if (ret) { | |
9116 | btrfs_err_rl(fs_info, | |
9117 | "error %d accounting leaf items, quota is out of sync, rescan required", | |
5d163e0e | 9118 | ret); |
c4140cbf | 9119 | } |
1152651a | 9120 | } |
2c47e605 | 9121 | } |
6a884d7d | 9122 | /* make block locked assertion in btrfs_clean_tree_block happy */ |
2c47e605 YZ |
9123 | if (!path->locks[level] && |
9124 | btrfs_header_generation(eb) == trans->transid) { | |
9125 | btrfs_tree_lock(eb); | |
8bead258 | 9126 | btrfs_set_lock_blocking_write(eb); |
bd681513 | 9127 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 | 9128 | } |
6a884d7d | 9129 | btrfs_clean_tree_block(eb); |
2c47e605 YZ |
9130 | } |
9131 | ||
9132 | if (eb == root->node) { | |
9133 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
9134 | parent = eb->start; | |
65c6e82b QW |
9135 | else if (root->root_key.objectid != btrfs_header_owner(eb)) |
9136 | goto owner_mismatch; | |
2c47e605 YZ |
9137 | } else { |
9138 | if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
9139 | parent = path->nodes[level + 1]->start; | |
65c6e82b QW |
9140 | else if (root->root_key.objectid != |
9141 | btrfs_header_owner(path->nodes[level + 1])) | |
9142 | goto owner_mismatch; | |
f82d02d9 | 9143 | } |
f82d02d9 | 9144 | |
5581a51a | 9145 | btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1); |
2c47e605 YZ |
9146 | out: |
9147 | wc->refs[level] = 0; | |
9148 | wc->flags[level] = 0; | |
f0486c68 | 9149 | return 0; |
65c6e82b QW |
9150 | |
9151 | owner_mismatch: | |
9152 | btrfs_err_rl(fs_info, "unexpected tree owner, have %llu expect %llu", | |
9153 | btrfs_header_owner(eb), root->root_key.objectid); | |
9154 | return -EUCLEAN; | |
2c47e605 YZ |
9155 | } |
9156 | ||
9157 | static noinline int walk_down_tree(struct btrfs_trans_handle *trans, | |
9158 | struct btrfs_root *root, | |
9159 | struct btrfs_path *path, | |
9160 | struct walk_control *wc) | |
9161 | { | |
2c47e605 | 9162 | int level = wc->level; |
94fcca9f | 9163 | int lookup_info = 1; |
2c47e605 YZ |
9164 | int ret; |
9165 | ||
9166 | while (level >= 0) { | |
94fcca9f | 9167 | ret = walk_down_proc(trans, root, path, wc, lookup_info); |
2c47e605 YZ |
9168 | if (ret > 0) |
9169 | break; | |
9170 | ||
9171 | if (level == 0) | |
9172 | break; | |
9173 | ||
7a7965f8 YZ |
9174 | if (path->slots[level] >= |
9175 | btrfs_header_nritems(path->nodes[level])) | |
9176 | break; | |
9177 | ||
94fcca9f | 9178 | ret = do_walk_down(trans, root, path, wc, &lookup_info); |
1c4850e2 YZ |
9179 | if (ret > 0) { |
9180 | path->slots[level]++; | |
9181 | continue; | |
90d2c51d MX |
9182 | } else if (ret < 0) |
9183 | return ret; | |
1c4850e2 | 9184 | level = wc->level; |
f82d02d9 | 9185 | } |
f82d02d9 YZ |
9186 | return 0; |
9187 | } | |
9188 | ||
d397712b | 9189 | static noinline int walk_up_tree(struct btrfs_trans_handle *trans, |
98ed5174 | 9190 | struct btrfs_root *root, |
f82d02d9 | 9191 | struct btrfs_path *path, |
2c47e605 | 9192 | struct walk_control *wc, int max_level) |
20524f02 | 9193 | { |
2c47e605 | 9194 | int level = wc->level; |
20524f02 | 9195 | int ret; |
9f3a7427 | 9196 | |
2c47e605 YZ |
9197 | path->slots[level] = btrfs_header_nritems(path->nodes[level]); |
9198 | while (level < max_level && path->nodes[level]) { | |
9199 | wc->level = level; | |
9200 | if (path->slots[level] + 1 < | |
9201 | btrfs_header_nritems(path->nodes[level])) { | |
9202 | path->slots[level]++; | |
20524f02 CM |
9203 | return 0; |
9204 | } else { | |
2c47e605 YZ |
9205 | ret = walk_up_proc(trans, root, path, wc); |
9206 | if (ret > 0) | |
9207 | return 0; | |
65c6e82b QW |
9208 | if (ret < 0) |
9209 | return ret; | |
bd56b302 | 9210 | |
2c47e605 | 9211 | if (path->locks[level]) { |
bd681513 CM |
9212 | btrfs_tree_unlock_rw(path->nodes[level], |
9213 | path->locks[level]); | |
2c47e605 | 9214 | path->locks[level] = 0; |
f82d02d9 | 9215 | } |
2c47e605 YZ |
9216 | free_extent_buffer(path->nodes[level]); |
9217 | path->nodes[level] = NULL; | |
9218 | level++; | |
20524f02 CM |
9219 | } |
9220 | } | |
9221 | return 1; | |
9222 | } | |
9223 | ||
9aca1d51 | 9224 | /* |
2c47e605 YZ |
9225 | * drop a subvolume tree. |
9226 | * | |
9227 | * this function traverses the tree freeing any blocks that only | |
9228 | * referenced by the tree. | |
9229 | * | |
9230 | * when a shared tree block is found. this function decreases its | |
9231 | * reference count by one. if update_ref is true, this function | |
9232 | * also make sure backrefs for the shared block and all lower level | |
9233 | * blocks are properly updated. | |
9d1a2a3a DS |
9234 | * |
9235 | * If called with for_reloc == 0, may exit early with -EAGAIN | |
9aca1d51 | 9236 | */ |
2c536799 | 9237 | int btrfs_drop_snapshot(struct btrfs_root *root, |
66d7e7f0 AJ |
9238 | struct btrfs_block_rsv *block_rsv, int update_ref, |
9239 | int for_reloc) | |
20524f02 | 9240 | { |
ab8d0fc4 | 9241 | struct btrfs_fs_info *fs_info = root->fs_info; |
5caf2a00 | 9242 | struct btrfs_path *path; |
2c47e605 | 9243 | struct btrfs_trans_handle *trans; |
ab8d0fc4 | 9244 | struct btrfs_root *tree_root = fs_info->tree_root; |
9f3a7427 | 9245 | struct btrfs_root_item *root_item = &root->root_item; |
2c47e605 YZ |
9246 | struct walk_control *wc; |
9247 | struct btrfs_key key; | |
9248 | int err = 0; | |
9249 | int ret; | |
9250 | int level; | |
d29a9f62 | 9251 | bool root_dropped = false; |
20524f02 | 9252 | |
4fd786e6 | 9253 | btrfs_debug(fs_info, "Drop subvolume %llu", root->root_key.objectid); |
1152651a | 9254 | |
5caf2a00 | 9255 | path = btrfs_alloc_path(); |
cb1b69f4 TI |
9256 | if (!path) { |
9257 | err = -ENOMEM; | |
9258 | goto out; | |
9259 | } | |
20524f02 | 9260 | |
2c47e605 | 9261 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
38a1a919 MF |
9262 | if (!wc) { |
9263 | btrfs_free_path(path); | |
cb1b69f4 TI |
9264 | err = -ENOMEM; |
9265 | goto out; | |
38a1a919 | 9266 | } |
2c47e605 | 9267 | |
a22285a6 | 9268 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
9269 | if (IS_ERR(trans)) { |
9270 | err = PTR_ERR(trans); | |
9271 | goto out_free; | |
9272 | } | |
98d5dc13 | 9273 | |
0568e82d JB |
9274 | err = btrfs_run_delayed_items(trans); |
9275 | if (err) | |
9276 | goto out_end_trans; | |
9277 | ||
3fd0a558 YZ |
9278 | if (block_rsv) |
9279 | trans->block_rsv = block_rsv; | |
2c47e605 | 9280 | |
83354f07 JB |
9281 | /* |
9282 | * This will help us catch people modifying the fs tree while we're | |
9283 | * dropping it. It is unsafe to mess with the fs tree while it's being | |
9284 | * dropped as we unlock the root node and parent nodes as we walk down | |
9285 | * the tree, assuming nothing will change. If something does change | |
9286 | * then we'll have stale information and drop references to blocks we've | |
9287 | * already dropped. | |
9288 | */ | |
9289 | set_bit(BTRFS_ROOT_DELETING, &root->state); | |
9f3a7427 | 9290 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { |
2c47e605 | 9291 | level = btrfs_header_level(root->node); |
5d4f98a2 | 9292 | path->nodes[level] = btrfs_lock_root_node(root); |
8bead258 | 9293 | btrfs_set_lock_blocking_write(path->nodes[level]); |
9f3a7427 | 9294 | path->slots[level] = 0; |
bd681513 | 9295 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
9296 | memset(&wc->update_progress, 0, |
9297 | sizeof(wc->update_progress)); | |
9f3a7427 | 9298 | } else { |
9f3a7427 | 9299 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); |
2c47e605 YZ |
9300 | memcpy(&wc->update_progress, &key, |
9301 | sizeof(wc->update_progress)); | |
9302 | ||
6702ed49 | 9303 | level = root_item->drop_level; |
2c47e605 | 9304 | BUG_ON(level == 0); |
6702ed49 | 9305 | path->lowest_level = level; |
2c47e605 YZ |
9306 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
9307 | path->lowest_level = 0; | |
9308 | if (ret < 0) { | |
9309 | err = ret; | |
79787eaa | 9310 | goto out_end_trans; |
9f3a7427 | 9311 | } |
1c4850e2 | 9312 | WARN_ON(ret > 0); |
2c47e605 | 9313 | |
7d9eb12c CM |
9314 | /* |
9315 | * unlock our path, this is safe because only this | |
9316 | * function is allowed to delete this snapshot | |
9317 | */ | |
5d4f98a2 | 9318 | btrfs_unlock_up_safe(path, 0); |
2c47e605 YZ |
9319 | |
9320 | level = btrfs_header_level(root->node); | |
9321 | while (1) { | |
9322 | btrfs_tree_lock(path->nodes[level]); | |
8bead258 | 9323 | btrfs_set_lock_blocking_write(path->nodes[level]); |
fec386ac | 9324 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 | 9325 | |
2ff7e61e | 9326 | ret = btrfs_lookup_extent_info(trans, fs_info, |
2c47e605 | 9327 | path->nodes[level]->start, |
3173a18f | 9328 | level, 1, &wc->refs[level], |
2c47e605 | 9329 | &wc->flags[level]); |
79787eaa JM |
9330 | if (ret < 0) { |
9331 | err = ret; | |
9332 | goto out_end_trans; | |
9333 | } | |
2c47e605 YZ |
9334 | BUG_ON(wc->refs[level] == 0); |
9335 | ||
9336 | if (level == root_item->drop_level) | |
9337 | break; | |
9338 | ||
9339 | btrfs_tree_unlock(path->nodes[level]); | |
fec386ac | 9340 | path->locks[level] = 0; |
2c47e605 YZ |
9341 | WARN_ON(wc->refs[level] != 1); |
9342 | level--; | |
9343 | } | |
9f3a7427 | 9344 | } |
2c47e605 | 9345 | |
78c52d9e | 9346 | wc->restarted = test_bit(BTRFS_ROOT_DEAD_TREE, &root->state); |
2c47e605 YZ |
9347 | wc->level = level; |
9348 | wc->shared_level = -1; | |
9349 | wc->stage = DROP_REFERENCE; | |
9350 | wc->update_ref = update_ref; | |
9351 | wc->keep_locks = 0; | |
0b246afa | 9352 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info); |
2c47e605 | 9353 | |
d397712b | 9354 | while (1) { |
9d1a2a3a | 9355 | |
2c47e605 YZ |
9356 | ret = walk_down_tree(trans, root, path, wc); |
9357 | if (ret < 0) { | |
9358 | err = ret; | |
20524f02 | 9359 | break; |
2c47e605 | 9360 | } |
9aca1d51 | 9361 | |
2c47e605 YZ |
9362 | ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL); |
9363 | if (ret < 0) { | |
9364 | err = ret; | |
20524f02 | 9365 | break; |
2c47e605 YZ |
9366 | } |
9367 | ||
9368 | if (ret > 0) { | |
9369 | BUG_ON(wc->stage != DROP_REFERENCE); | |
e7a84565 CM |
9370 | break; |
9371 | } | |
2c47e605 YZ |
9372 | |
9373 | if (wc->stage == DROP_REFERENCE) { | |
aea6f028 JB |
9374 | wc->drop_level = wc->level; |
9375 | btrfs_node_key_to_cpu(path->nodes[wc->drop_level], | |
9376 | &wc->drop_progress, | |
9377 | path->slots[wc->drop_level]); | |
9378 | } | |
9379 | btrfs_cpu_key_to_disk(&root_item->drop_progress, | |
9380 | &wc->drop_progress); | |
9381 | root_item->drop_level = wc->drop_level; | |
2c47e605 YZ |
9382 | |
9383 | BUG_ON(wc->level == 0); | |
3a45bb20 | 9384 | if (btrfs_should_end_transaction(trans) || |
2ff7e61e | 9385 | (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) { |
2c47e605 YZ |
9386 | ret = btrfs_update_root(trans, tree_root, |
9387 | &root->root_key, | |
9388 | root_item); | |
79787eaa | 9389 | if (ret) { |
66642832 | 9390 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9391 | err = ret; |
9392 | goto out_end_trans; | |
9393 | } | |
2c47e605 | 9394 | |
3a45bb20 | 9395 | btrfs_end_transaction_throttle(trans); |
2ff7e61e | 9396 | if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) { |
ab8d0fc4 JM |
9397 | btrfs_debug(fs_info, |
9398 | "drop snapshot early exit"); | |
3c8f2422 JB |
9399 | err = -EAGAIN; |
9400 | goto out_free; | |
9401 | } | |
9402 | ||
a22285a6 | 9403 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
9404 | if (IS_ERR(trans)) { |
9405 | err = PTR_ERR(trans); | |
9406 | goto out_free; | |
9407 | } | |
3fd0a558 YZ |
9408 | if (block_rsv) |
9409 | trans->block_rsv = block_rsv; | |
c3e69d58 | 9410 | } |
20524f02 | 9411 | } |
b3b4aa74 | 9412 | btrfs_release_path(path); |
79787eaa JM |
9413 | if (err) |
9414 | goto out_end_trans; | |
2c47e605 | 9415 | |
ab9ce7d4 | 9416 | ret = btrfs_del_root(trans, &root->root_key); |
79787eaa | 9417 | if (ret) { |
66642832 | 9418 | btrfs_abort_transaction(trans, ret); |
e19182c0 | 9419 | err = ret; |
79787eaa JM |
9420 | goto out_end_trans; |
9421 | } | |
2c47e605 | 9422 | |
76dda93c | 9423 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { |
cb517eab MX |
9424 | ret = btrfs_find_root(tree_root, &root->root_key, path, |
9425 | NULL, NULL); | |
79787eaa | 9426 | if (ret < 0) { |
66642832 | 9427 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9428 | err = ret; |
9429 | goto out_end_trans; | |
9430 | } else if (ret > 0) { | |
84cd948c JB |
9431 | /* if we fail to delete the orphan item this time |
9432 | * around, it'll get picked up the next time. | |
9433 | * | |
9434 | * The most common failure here is just -ENOENT. | |
9435 | */ | |
9436 | btrfs_del_orphan_item(trans, tree_root, | |
9437 | root->root_key.objectid); | |
76dda93c YZ |
9438 | } |
9439 | } | |
9440 | ||
27cdeb70 | 9441 | if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) { |
2b9dbef2 | 9442 | btrfs_add_dropped_root(trans, root); |
76dda93c YZ |
9443 | } else { |
9444 | free_extent_buffer(root->node); | |
9445 | free_extent_buffer(root->commit_root); | |
b0feb9d9 | 9446 | btrfs_put_fs_root(root); |
76dda93c | 9447 | } |
d29a9f62 | 9448 | root_dropped = true; |
79787eaa | 9449 | out_end_trans: |
3a45bb20 | 9450 | btrfs_end_transaction_throttle(trans); |
79787eaa | 9451 | out_free: |
2c47e605 | 9452 | kfree(wc); |
5caf2a00 | 9453 | btrfs_free_path(path); |
cb1b69f4 | 9454 | out: |
d29a9f62 JB |
9455 | /* |
9456 | * So if we need to stop dropping the snapshot for whatever reason we | |
9457 | * need to make sure to add it back to the dead root list so that we | |
9458 | * keep trying to do the work later. This also cleans up roots if we | |
9459 | * don't have it in the radix (like when we recover after a power fail | |
9460 | * or unmount) so we don't leak memory. | |
9461 | */ | |
897ca819 | 9462 | if (!for_reloc && !root_dropped) |
d29a9f62 | 9463 | btrfs_add_dead_root(root); |
90515e7f | 9464 | if (err && err != -EAGAIN) |
ab8d0fc4 | 9465 | btrfs_handle_fs_error(fs_info, err, NULL); |
2c536799 | 9466 | return err; |
20524f02 | 9467 | } |
9078a3e1 | 9468 | |
2c47e605 YZ |
9469 | /* |
9470 | * drop subtree rooted at tree block 'node'. | |
9471 | * | |
9472 | * NOTE: this function will unlock and release tree block 'node' | |
66d7e7f0 | 9473 | * only used by relocation code |
2c47e605 | 9474 | */ |
f82d02d9 YZ |
9475 | int btrfs_drop_subtree(struct btrfs_trans_handle *trans, |
9476 | struct btrfs_root *root, | |
9477 | struct extent_buffer *node, | |
9478 | struct extent_buffer *parent) | |
9479 | { | |
0b246afa | 9480 | struct btrfs_fs_info *fs_info = root->fs_info; |
f82d02d9 | 9481 | struct btrfs_path *path; |
2c47e605 | 9482 | struct walk_control *wc; |
f82d02d9 YZ |
9483 | int level; |
9484 | int parent_level; | |
9485 | int ret = 0; | |
9486 | int wret; | |
9487 | ||
2c47e605 YZ |
9488 | BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); |
9489 | ||
f82d02d9 | 9490 | path = btrfs_alloc_path(); |
db5b493a TI |
9491 | if (!path) |
9492 | return -ENOMEM; | |
f82d02d9 | 9493 | |
2c47e605 | 9494 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
db5b493a TI |
9495 | if (!wc) { |
9496 | btrfs_free_path(path); | |
9497 | return -ENOMEM; | |
9498 | } | |
2c47e605 | 9499 | |
b9447ef8 | 9500 | btrfs_assert_tree_locked(parent); |
f82d02d9 YZ |
9501 | parent_level = btrfs_header_level(parent); |
9502 | extent_buffer_get(parent); | |
9503 | path->nodes[parent_level] = parent; | |
9504 | path->slots[parent_level] = btrfs_header_nritems(parent); | |
9505 | ||
b9447ef8 | 9506 | btrfs_assert_tree_locked(node); |
f82d02d9 | 9507 | level = btrfs_header_level(node); |
f82d02d9 YZ |
9508 | path->nodes[level] = node; |
9509 | path->slots[level] = 0; | |
bd681513 | 9510 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
9511 | |
9512 | wc->refs[parent_level] = 1; | |
9513 | wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
9514 | wc->level = level; | |
9515 | wc->shared_level = -1; | |
9516 | wc->stage = DROP_REFERENCE; | |
9517 | wc->update_ref = 0; | |
9518 | wc->keep_locks = 1; | |
0b246afa | 9519 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info); |
f82d02d9 YZ |
9520 | |
9521 | while (1) { | |
2c47e605 YZ |
9522 | wret = walk_down_tree(trans, root, path, wc); |
9523 | if (wret < 0) { | |
f82d02d9 | 9524 | ret = wret; |
f82d02d9 | 9525 | break; |
2c47e605 | 9526 | } |
f82d02d9 | 9527 | |
2c47e605 | 9528 | wret = walk_up_tree(trans, root, path, wc, parent_level); |
f82d02d9 YZ |
9529 | if (wret < 0) |
9530 | ret = wret; | |
9531 | if (wret != 0) | |
9532 | break; | |
9533 | } | |
9534 | ||
2c47e605 | 9535 | kfree(wc); |
f82d02d9 YZ |
9536 | btrfs_free_path(path); |
9537 | return ret; | |
9538 | } | |
9539 | ||
6202df69 | 9540 | static u64 update_block_group_flags(struct btrfs_fs_info *fs_info, u64 flags) |
ec44a35c CM |
9541 | { |
9542 | u64 num_devices; | |
fc67c450 | 9543 | u64 stripped; |
e4d8ec0f | 9544 | |
fc67c450 ID |
9545 | /* |
9546 | * if restripe for this chunk_type is on pick target profile and | |
9547 | * return, otherwise do the usual balance | |
9548 | */ | |
6202df69 | 9549 | stripped = get_restripe_target(fs_info, flags); |
fc67c450 ID |
9550 | if (stripped) |
9551 | return extended_to_chunk(stripped); | |
e4d8ec0f | 9552 | |
6202df69 | 9553 | num_devices = fs_info->fs_devices->rw_devices; |
cd02dca5 | 9554 | |
fc67c450 | 9555 | stripped = BTRFS_BLOCK_GROUP_RAID0 | |
53b381b3 | 9556 | BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 | |
fc67c450 ID |
9557 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10; |
9558 | ||
ec44a35c CM |
9559 | if (num_devices == 1) { |
9560 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
9561 | stripped = flags & ~stripped; | |
9562 | ||
9563 | /* turn raid0 into single device chunks */ | |
9564 | if (flags & BTRFS_BLOCK_GROUP_RAID0) | |
9565 | return stripped; | |
9566 | ||
9567 | /* turn mirroring into duplication */ | |
9568 | if (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
9569 | BTRFS_BLOCK_GROUP_RAID10)) | |
9570 | return stripped | BTRFS_BLOCK_GROUP_DUP; | |
ec44a35c CM |
9571 | } else { |
9572 | /* they already had raid on here, just return */ | |
ec44a35c CM |
9573 | if (flags & stripped) |
9574 | return flags; | |
9575 | ||
9576 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
9577 | stripped = flags & ~stripped; | |
9578 | ||
9579 | /* switch duplicated blocks with raid1 */ | |
9580 | if (flags & BTRFS_BLOCK_GROUP_DUP) | |
9581 | return stripped | BTRFS_BLOCK_GROUP_RAID1; | |
9582 | ||
e3176ca2 | 9583 | /* this is drive concat, leave it alone */ |
ec44a35c | 9584 | } |
e3176ca2 | 9585 | |
ec44a35c CM |
9586 | return flags; |
9587 | } | |
9588 | ||
868f401a | 9589 | static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force) |
0ef3e66b | 9590 | { |
f0486c68 YZ |
9591 | struct btrfs_space_info *sinfo = cache->space_info; |
9592 | u64 num_bytes; | |
3ece54e5 | 9593 | u64 sinfo_used; |
199c36ea | 9594 | u64 min_allocable_bytes; |
f0486c68 | 9595 | int ret = -ENOSPC; |
0ef3e66b | 9596 | |
199c36ea MX |
9597 | /* |
9598 | * We need some metadata space and system metadata space for | |
9599 | * allocating chunks in some corner cases until we force to set | |
9600 | * it to be readonly. | |
9601 | */ | |
9602 | if ((sinfo->flags & | |
9603 | (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) && | |
9604 | !force) | |
ee22184b | 9605 | min_allocable_bytes = SZ_1M; |
199c36ea MX |
9606 | else |
9607 | min_allocable_bytes = 0; | |
9608 | ||
f0486c68 YZ |
9609 | spin_lock(&sinfo->lock); |
9610 | spin_lock(&cache->lock); | |
61cfea9b W |
9611 | |
9612 | if (cache->ro) { | |
868f401a | 9613 | cache->ro++; |
61cfea9b W |
9614 | ret = 0; |
9615 | goto out; | |
9616 | } | |
9617 | ||
f0486c68 YZ |
9618 | num_bytes = cache->key.offset - cache->reserved - cache->pinned - |
9619 | cache->bytes_super - btrfs_block_group_used(&cache->item); | |
3ece54e5 | 9620 | sinfo_used = btrfs_space_info_used(sinfo, true); |
f0486c68 | 9621 | |
3ece54e5 QW |
9622 | if (sinfo_used + num_bytes + min_allocable_bytes <= |
9623 | sinfo->total_bytes) { | |
f0486c68 | 9624 | sinfo->bytes_readonly += num_bytes; |
868f401a | 9625 | cache->ro++; |
633c0aad | 9626 | list_add_tail(&cache->ro_list, &sinfo->ro_bgs); |
f0486c68 YZ |
9627 | ret = 0; |
9628 | } | |
61cfea9b | 9629 | out: |
f0486c68 YZ |
9630 | spin_unlock(&cache->lock); |
9631 | spin_unlock(&sinfo->lock); | |
3ece54e5 QW |
9632 | if (ret == -ENOSPC && btrfs_test_opt(cache->fs_info, ENOSPC_DEBUG)) { |
9633 | btrfs_info(cache->fs_info, | |
9634 | "unable to make block group %llu ro", | |
9635 | cache->key.objectid); | |
9636 | btrfs_info(cache->fs_info, | |
9637 | "sinfo_used=%llu bg_num_bytes=%llu min_allocable=%llu", | |
9638 | sinfo_used, num_bytes, min_allocable_bytes); | |
9639 | dump_space_info(cache->fs_info, cache->space_info, 0, 0); | |
9640 | } | |
f0486c68 YZ |
9641 | return ret; |
9642 | } | |
7d9eb12c | 9643 | |
c83488af | 9644 | int btrfs_inc_block_group_ro(struct btrfs_block_group_cache *cache) |
c286ac48 | 9645 | |
f0486c68 | 9646 | { |
c83488af | 9647 | struct btrfs_fs_info *fs_info = cache->fs_info; |
f0486c68 YZ |
9648 | struct btrfs_trans_handle *trans; |
9649 | u64 alloc_flags; | |
9650 | int ret; | |
7d9eb12c | 9651 | |
1bbc621e | 9652 | again: |
5e00f193 | 9653 | trans = btrfs_join_transaction(fs_info->extent_root); |
79787eaa JM |
9654 | if (IS_ERR(trans)) |
9655 | return PTR_ERR(trans); | |
5d4f98a2 | 9656 | |
1bbc621e CM |
9657 | /* |
9658 | * we're not allowed to set block groups readonly after the dirty | |
9659 | * block groups cache has started writing. If it already started, | |
9660 | * back off and let this transaction commit | |
9661 | */ | |
0b246afa | 9662 | mutex_lock(&fs_info->ro_block_group_mutex); |
3204d33c | 9663 | if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) { |
1bbc621e CM |
9664 | u64 transid = trans->transid; |
9665 | ||
0b246afa | 9666 | mutex_unlock(&fs_info->ro_block_group_mutex); |
3a45bb20 | 9667 | btrfs_end_transaction(trans); |
1bbc621e | 9668 | |
2ff7e61e | 9669 | ret = btrfs_wait_for_commit(fs_info, transid); |
1bbc621e CM |
9670 | if (ret) |
9671 | return ret; | |
9672 | goto again; | |
9673 | } | |
9674 | ||
153c35b6 CM |
9675 | /* |
9676 | * if we are changing raid levels, try to allocate a corresponding | |
9677 | * block group with the new raid level. | |
9678 | */ | |
0b246afa | 9679 | alloc_flags = update_block_group_flags(fs_info, cache->flags); |
153c35b6 | 9680 | if (alloc_flags != cache->flags) { |
01458828 | 9681 | ret = do_chunk_alloc(trans, alloc_flags, |
153c35b6 CM |
9682 | CHUNK_ALLOC_FORCE); |
9683 | /* | |
9684 | * ENOSPC is allowed here, we may have enough space | |
9685 | * already allocated at the new raid level to | |
9686 | * carry on | |
9687 | */ | |
9688 | if (ret == -ENOSPC) | |
9689 | ret = 0; | |
9690 | if (ret < 0) | |
9691 | goto out; | |
9692 | } | |
1bbc621e | 9693 | |
868f401a | 9694 | ret = inc_block_group_ro(cache, 0); |
f0486c68 YZ |
9695 | if (!ret) |
9696 | goto out; | |
2ff7e61e | 9697 | alloc_flags = get_alloc_profile(fs_info, cache->space_info->flags); |
01458828 | 9698 | ret = do_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE); |
f0486c68 YZ |
9699 | if (ret < 0) |
9700 | goto out; | |
868f401a | 9701 | ret = inc_block_group_ro(cache, 0); |
f0486c68 | 9702 | out: |
2f081088 | 9703 | if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) { |
0b246afa | 9704 | alloc_flags = update_block_group_flags(fs_info, cache->flags); |
34441361 | 9705 | mutex_lock(&fs_info->chunk_mutex); |
451a2c13 | 9706 | check_system_chunk(trans, alloc_flags); |
34441361 | 9707 | mutex_unlock(&fs_info->chunk_mutex); |
2f081088 | 9708 | } |
0b246afa | 9709 | mutex_unlock(&fs_info->ro_block_group_mutex); |
2f081088 | 9710 | |
3a45bb20 | 9711 | btrfs_end_transaction(trans); |
f0486c68 YZ |
9712 | return ret; |
9713 | } | |
5d4f98a2 | 9714 | |
43a7e99d | 9715 | int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans, u64 type) |
c87f08ca | 9716 | { |
43a7e99d | 9717 | u64 alloc_flags = get_alloc_profile(trans->fs_info, type); |
2ff7e61e | 9718 | |
01458828 | 9719 | return do_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE); |
c87f08ca CM |
9720 | } |
9721 | ||
6d07bcec MX |
9722 | /* |
9723 | * helper to account the unused space of all the readonly block group in the | |
633c0aad | 9724 | * space_info. takes mirrors into account. |
6d07bcec | 9725 | */ |
633c0aad | 9726 | u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo) |
6d07bcec MX |
9727 | { |
9728 | struct btrfs_block_group_cache *block_group; | |
9729 | u64 free_bytes = 0; | |
9730 | int factor; | |
9731 | ||
01327610 | 9732 | /* It's df, we don't care if it's racy */ |
633c0aad JB |
9733 | if (list_empty(&sinfo->ro_bgs)) |
9734 | return 0; | |
9735 | ||
9736 | spin_lock(&sinfo->lock); | |
9737 | list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) { | |
6d07bcec MX |
9738 | spin_lock(&block_group->lock); |
9739 | ||
9740 | if (!block_group->ro) { | |
9741 | spin_unlock(&block_group->lock); | |
9742 | continue; | |
9743 | } | |
9744 | ||
46df06b8 | 9745 | factor = btrfs_bg_type_to_factor(block_group->flags); |
6d07bcec MX |
9746 | free_bytes += (block_group->key.offset - |
9747 | btrfs_block_group_used(&block_group->item)) * | |
9748 | factor; | |
9749 | ||
9750 | spin_unlock(&block_group->lock); | |
9751 | } | |
6d07bcec MX |
9752 | spin_unlock(&sinfo->lock); |
9753 | ||
9754 | return free_bytes; | |
9755 | } | |
9756 | ||
2ff7e61e | 9757 | void btrfs_dec_block_group_ro(struct btrfs_block_group_cache *cache) |
5d4f98a2 | 9758 | { |
f0486c68 YZ |
9759 | struct btrfs_space_info *sinfo = cache->space_info; |
9760 | u64 num_bytes; | |
9761 | ||
9762 | BUG_ON(!cache->ro); | |
9763 | ||
9764 | spin_lock(&sinfo->lock); | |
9765 | spin_lock(&cache->lock); | |
868f401a Z |
9766 | if (!--cache->ro) { |
9767 | num_bytes = cache->key.offset - cache->reserved - | |
9768 | cache->pinned - cache->bytes_super - | |
9769 | btrfs_block_group_used(&cache->item); | |
9770 | sinfo->bytes_readonly -= num_bytes; | |
9771 | list_del_init(&cache->ro_list); | |
9772 | } | |
f0486c68 YZ |
9773 | spin_unlock(&cache->lock); |
9774 | spin_unlock(&sinfo->lock); | |
5d4f98a2 YZ |
9775 | } |
9776 | ||
ba1bf481 | 9777 | /* |
52042d8e | 9778 | * Checks to see if it's even possible to relocate this block group. |
ba1bf481 JB |
9779 | * |
9780 | * @return - -1 if it's not a good idea to relocate this block group, 0 if its | |
9781 | * ok to go ahead and try. | |
9782 | */ | |
6bccf3ab | 9783 | int btrfs_can_relocate(struct btrfs_fs_info *fs_info, u64 bytenr) |
1a40e23b | 9784 | { |
ba1bf481 JB |
9785 | struct btrfs_block_group_cache *block_group; |
9786 | struct btrfs_space_info *space_info; | |
0b246afa | 9787 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
ba1bf481 | 9788 | struct btrfs_device *device; |
cdcb725c | 9789 | u64 min_free; |
6719db6a JB |
9790 | u64 dev_min = 1; |
9791 | u64 dev_nr = 0; | |
4a5e98f5 | 9792 | u64 target; |
0305bc27 | 9793 | int debug; |
cdcb725c | 9794 | int index; |
ba1bf481 JB |
9795 | int full = 0; |
9796 | int ret = 0; | |
1a40e23b | 9797 | |
0b246afa | 9798 | debug = btrfs_test_opt(fs_info, ENOSPC_DEBUG); |
0305bc27 | 9799 | |
0b246afa | 9800 | block_group = btrfs_lookup_block_group(fs_info, bytenr); |
1a40e23b | 9801 | |
ba1bf481 | 9802 | /* odd, couldn't find the block group, leave it alone */ |
0305bc27 QW |
9803 | if (!block_group) { |
9804 | if (debug) | |
0b246afa | 9805 | btrfs_warn(fs_info, |
0305bc27 QW |
9806 | "can't find block group for bytenr %llu", |
9807 | bytenr); | |
ba1bf481 | 9808 | return -1; |
0305bc27 | 9809 | } |
1a40e23b | 9810 | |
cdcb725c | 9811 | min_free = btrfs_block_group_used(&block_group->item); |
9812 | ||
ba1bf481 | 9813 | /* no bytes used, we're good */ |
cdcb725c | 9814 | if (!min_free) |
1a40e23b ZY |
9815 | goto out; |
9816 | ||
ba1bf481 JB |
9817 | space_info = block_group->space_info; |
9818 | spin_lock(&space_info->lock); | |
17d217fe | 9819 | |
ba1bf481 | 9820 | full = space_info->full; |
17d217fe | 9821 | |
ba1bf481 JB |
9822 | /* |
9823 | * if this is the last block group we have in this space, we can't | |
7ce618db CM |
9824 | * relocate it unless we're able to allocate a new chunk below. |
9825 | * | |
9826 | * Otherwise, we need to make sure we have room in the space to handle | |
9827 | * all of the extents from this block group. If we can, we're good | |
ba1bf481 | 9828 | */ |
7ce618db | 9829 | if ((space_info->total_bytes != block_group->key.offset) && |
4136135b LB |
9830 | (btrfs_space_info_used(space_info, false) + min_free < |
9831 | space_info->total_bytes)) { | |
ba1bf481 JB |
9832 | spin_unlock(&space_info->lock); |
9833 | goto out; | |
17d217fe | 9834 | } |
ba1bf481 | 9835 | spin_unlock(&space_info->lock); |
ea8c2819 | 9836 | |
ba1bf481 JB |
9837 | /* |
9838 | * ok we don't have enough space, but maybe we have free space on our | |
9839 | * devices to allocate new chunks for relocation, so loop through our | |
4a5e98f5 ID |
9840 | * alloc devices and guess if we have enough space. if this block |
9841 | * group is going to be restriped, run checks against the target | |
9842 | * profile instead of the current one. | |
ba1bf481 JB |
9843 | */ |
9844 | ret = -1; | |
ea8c2819 | 9845 | |
cdcb725c | 9846 | /* |
9847 | * index: | |
9848 | * 0: raid10 | |
9849 | * 1: raid1 | |
9850 | * 2: dup | |
9851 | * 3: raid0 | |
9852 | * 4: single | |
9853 | */ | |
0b246afa | 9854 | target = get_restripe_target(fs_info, block_group->flags); |
4a5e98f5 | 9855 | if (target) { |
3e72ee88 | 9856 | index = btrfs_bg_flags_to_raid_index(extended_to_chunk(target)); |
4a5e98f5 ID |
9857 | } else { |
9858 | /* | |
9859 | * this is just a balance, so if we were marked as full | |
9860 | * we know there is no space for a new chunk | |
9861 | */ | |
0305bc27 QW |
9862 | if (full) { |
9863 | if (debug) | |
0b246afa JM |
9864 | btrfs_warn(fs_info, |
9865 | "no space to alloc new chunk for block group %llu", | |
9866 | block_group->key.objectid); | |
4a5e98f5 | 9867 | goto out; |
0305bc27 | 9868 | } |
4a5e98f5 | 9869 | |
3e72ee88 | 9870 | index = btrfs_bg_flags_to_raid_index(block_group->flags); |
4a5e98f5 ID |
9871 | } |
9872 | ||
e6ec716f | 9873 | if (index == BTRFS_RAID_RAID10) { |
cdcb725c | 9874 | dev_min = 4; |
6719db6a JB |
9875 | /* Divide by 2 */ |
9876 | min_free >>= 1; | |
e6ec716f | 9877 | } else if (index == BTRFS_RAID_RAID1) { |
cdcb725c | 9878 | dev_min = 2; |
e6ec716f | 9879 | } else if (index == BTRFS_RAID_DUP) { |
6719db6a JB |
9880 | /* Multiply by 2 */ |
9881 | min_free <<= 1; | |
e6ec716f | 9882 | } else if (index == BTRFS_RAID_RAID0) { |
cdcb725c | 9883 | dev_min = fs_devices->rw_devices; |
47c5713f | 9884 | min_free = div64_u64(min_free, dev_min); |
cdcb725c | 9885 | } |
9886 | ||
0b246afa | 9887 | mutex_lock(&fs_info->chunk_mutex); |
ba1bf481 | 9888 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { |
7bfc837d | 9889 | u64 dev_offset; |
56bec294 | 9890 | |
ba1bf481 JB |
9891 | /* |
9892 | * check to make sure we can actually find a chunk with enough | |
9893 | * space to fit our block group in. | |
9894 | */ | |
63a212ab | 9895 | if (device->total_bytes > device->bytes_used + min_free && |
401e29c1 | 9896 | !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
60dfdf25 | 9897 | ret = find_free_dev_extent(device, min_free, |
7bfc837d | 9898 | &dev_offset, NULL); |
ba1bf481 | 9899 | if (!ret) |
cdcb725c | 9900 | dev_nr++; |
9901 | ||
9902 | if (dev_nr >= dev_min) | |
73e48b27 | 9903 | break; |
cdcb725c | 9904 | |
ba1bf481 | 9905 | ret = -1; |
725c8463 | 9906 | } |
edbd8d4e | 9907 | } |
0305bc27 | 9908 | if (debug && ret == -1) |
0b246afa JM |
9909 | btrfs_warn(fs_info, |
9910 | "no space to allocate a new chunk for block group %llu", | |
9911 | block_group->key.objectid); | |
9912 | mutex_unlock(&fs_info->chunk_mutex); | |
edbd8d4e | 9913 | out: |
ba1bf481 | 9914 | btrfs_put_block_group(block_group); |
edbd8d4e CM |
9915 | return ret; |
9916 | } | |
9917 | ||
6bccf3ab JM |
9918 | static int find_first_block_group(struct btrfs_fs_info *fs_info, |
9919 | struct btrfs_path *path, | |
9920 | struct btrfs_key *key) | |
0b86a832 | 9921 | { |
6bccf3ab | 9922 | struct btrfs_root *root = fs_info->extent_root; |
925baedd | 9923 | int ret = 0; |
0b86a832 CM |
9924 | struct btrfs_key found_key; |
9925 | struct extent_buffer *leaf; | |
514c7dca QW |
9926 | struct btrfs_block_group_item bg; |
9927 | u64 flags; | |
0b86a832 | 9928 | int slot; |
edbd8d4e | 9929 | |
0b86a832 CM |
9930 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); |
9931 | if (ret < 0) | |
925baedd CM |
9932 | goto out; |
9933 | ||
d397712b | 9934 | while (1) { |
0b86a832 | 9935 | slot = path->slots[0]; |
edbd8d4e | 9936 | leaf = path->nodes[0]; |
0b86a832 CM |
9937 | if (slot >= btrfs_header_nritems(leaf)) { |
9938 | ret = btrfs_next_leaf(root, path); | |
9939 | if (ret == 0) | |
9940 | continue; | |
9941 | if (ret < 0) | |
925baedd | 9942 | goto out; |
0b86a832 | 9943 | break; |
edbd8d4e | 9944 | } |
0b86a832 | 9945 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
edbd8d4e | 9946 | |
0b86a832 | 9947 | if (found_key.objectid >= key->objectid && |
925baedd | 9948 | found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) { |
6fb37b75 LB |
9949 | struct extent_map_tree *em_tree; |
9950 | struct extent_map *em; | |
9951 | ||
9952 | em_tree = &root->fs_info->mapping_tree.map_tree; | |
9953 | read_lock(&em_tree->lock); | |
9954 | em = lookup_extent_mapping(em_tree, found_key.objectid, | |
9955 | found_key.offset); | |
9956 | read_unlock(&em_tree->lock); | |
9957 | if (!em) { | |
0b246afa | 9958 | btrfs_err(fs_info, |
6fb37b75 LB |
9959 | "logical %llu len %llu found bg but no related chunk", |
9960 | found_key.objectid, found_key.offset); | |
9961 | ret = -ENOENT; | |
514c7dca QW |
9962 | } else if (em->start != found_key.objectid || |
9963 | em->len != found_key.offset) { | |
9964 | btrfs_err(fs_info, | |
9965 | "block group %llu len %llu mismatch with chunk %llu len %llu", | |
9966 | found_key.objectid, found_key.offset, | |
9967 | em->start, em->len); | |
9968 | ret = -EUCLEAN; | |
6fb37b75 | 9969 | } else { |
514c7dca QW |
9970 | read_extent_buffer(leaf, &bg, |
9971 | btrfs_item_ptr_offset(leaf, slot), | |
9972 | sizeof(bg)); | |
9973 | flags = btrfs_block_group_flags(&bg) & | |
9974 | BTRFS_BLOCK_GROUP_TYPE_MASK; | |
9975 | ||
9976 | if (flags != (em->map_lookup->type & | |
9977 | BTRFS_BLOCK_GROUP_TYPE_MASK)) { | |
9978 | btrfs_err(fs_info, | |
9979 | "block group %llu len %llu type flags 0x%llx mismatch with chunk type flags 0x%llx", | |
9980 | found_key.objectid, | |
9981 | found_key.offset, flags, | |
9982 | (BTRFS_BLOCK_GROUP_TYPE_MASK & | |
9983 | em->map_lookup->type)); | |
9984 | ret = -EUCLEAN; | |
9985 | } else { | |
9986 | ret = 0; | |
9987 | } | |
6fb37b75 | 9988 | } |
187ee58c | 9989 | free_extent_map(em); |
925baedd CM |
9990 | goto out; |
9991 | } | |
0b86a832 | 9992 | path->slots[0]++; |
edbd8d4e | 9993 | } |
925baedd | 9994 | out: |
0b86a832 | 9995 | return ret; |
edbd8d4e CM |
9996 | } |
9997 | ||
0af3d00b JB |
9998 | void btrfs_put_block_group_cache(struct btrfs_fs_info *info) |
9999 | { | |
10000 | struct btrfs_block_group_cache *block_group; | |
10001 | u64 last = 0; | |
10002 | ||
10003 | while (1) { | |
10004 | struct inode *inode; | |
10005 | ||
10006 | block_group = btrfs_lookup_first_block_group(info, last); | |
10007 | while (block_group) { | |
3aa7c7a3 | 10008 | wait_block_group_cache_done(block_group); |
0af3d00b JB |
10009 | spin_lock(&block_group->lock); |
10010 | if (block_group->iref) | |
10011 | break; | |
10012 | spin_unlock(&block_group->lock); | |
f87b7eb8 | 10013 | block_group = next_block_group(block_group); |
0af3d00b JB |
10014 | } |
10015 | if (!block_group) { | |
10016 | if (last == 0) | |
10017 | break; | |
10018 | last = 0; | |
10019 | continue; | |
10020 | } | |
10021 | ||
10022 | inode = block_group->inode; | |
10023 | block_group->iref = 0; | |
10024 | block_group->inode = NULL; | |
10025 | spin_unlock(&block_group->lock); | |
f3bca802 | 10026 | ASSERT(block_group->io_ctl.inode == NULL); |
0af3d00b JB |
10027 | iput(inode); |
10028 | last = block_group->key.objectid + block_group->key.offset; | |
10029 | btrfs_put_block_group(block_group); | |
10030 | } | |
10031 | } | |
10032 | ||
5cdd7db6 FM |
10033 | /* |
10034 | * Must be called only after stopping all workers, since we could have block | |
10035 | * group caching kthreads running, and therefore they could race with us if we | |
10036 | * freed the block groups before stopping them. | |
10037 | */ | |
1a40e23b ZY |
10038 | int btrfs_free_block_groups(struct btrfs_fs_info *info) |
10039 | { | |
10040 | struct btrfs_block_group_cache *block_group; | |
4184ea7f | 10041 | struct btrfs_space_info *space_info; |
11833d66 | 10042 | struct btrfs_caching_control *caching_ctl; |
1a40e23b ZY |
10043 | struct rb_node *n; |
10044 | ||
9e351cc8 | 10045 | down_write(&info->commit_root_sem); |
11833d66 YZ |
10046 | while (!list_empty(&info->caching_block_groups)) { |
10047 | caching_ctl = list_entry(info->caching_block_groups.next, | |
10048 | struct btrfs_caching_control, list); | |
10049 | list_del(&caching_ctl->list); | |
10050 | put_caching_control(caching_ctl); | |
10051 | } | |
9e351cc8 | 10052 | up_write(&info->commit_root_sem); |
11833d66 | 10053 | |
47ab2a6c JB |
10054 | spin_lock(&info->unused_bgs_lock); |
10055 | while (!list_empty(&info->unused_bgs)) { | |
10056 | block_group = list_first_entry(&info->unused_bgs, | |
10057 | struct btrfs_block_group_cache, | |
10058 | bg_list); | |
10059 | list_del_init(&block_group->bg_list); | |
10060 | btrfs_put_block_group(block_group); | |
10061 | } | |
10062 | spin_unlock(&info->unused_bgs_lock); | |
10063 | ||
1a40e23b ZY |
10064 | spin_lock(&info->block_group_cache_lock); |
10065 | while ((n = rb_last(&info->block_group_cache_tree)) != NULL) { | |
10066 | block_group = rb_entry(n, struct btrfs_block_group_cache, | |
10067 | cache_node); | |
1a40e23b ZY |
10068 | rb_erase(&block_group->cache_node, |
10069 | &info->block_group_cache_tree); | |
01eacb27 | 10070 | RB_CLEAR_NODE(&block_group->cache_node); |
d899e052 YZ |
10071 | spin_unlock(&info->block_group_cache_lock); |
10072 | ||
80eb234a | 10073 | down_write(&block_group->space_info->groups_sem); |
1a40e23b | 10074 | list_del(&block_group->list); |
80eb234a | 10075 | up_write(&block_group->space_info->groups_sem); |
d2fb3437 | 10076 | |
3c14874a JB |
10077 | /* |
10078 | * We haven't cached this block group, which means we could | |
10079 | * possibly have excluded extents on this block group. | |
10080 | */ | |
36cce922 JB |
10081 | if (block_group->cached == BTRFS_CACHE_NO || |
10082 | block_group->cached == BTRFS_CACHE_ERROR) | |
9e715da8 | 10083 | free_excluded_extents(block_group); |
3c14874a | 10084 | |
817d52f8 | 10085 | btrfs_remove_free_space_cache(block_group); |
5cdd7db6 | 10086 | ASSERT(block_group->cached != BTRFS_CACHE_STARTED); |
f3bca802 LB |
10087 | ASSERT(list_empty(&block_group->dirty_list)); |
10088 | ASSERT(list_empty(&block_group->io_list)); | |
10089 | ASSERT(list_empty(&block_group->bg_list)); | |
10090 | ASSERT(atomic_read(&block_group->count) == 1); | |
11dfe35a | 10091 | btrfs_put_block_group(block_group); |
d899e052 YZ |
10092 | |
10093 | spin_lock(&info->block_group_cache_lock); | |
1a40e23b ZY |
10094 | } |
10095 | spin_unlock(&info->block_group_cache_lock); | |
4184ea7f CM |
10096 | |
10097 | /* now that all the block groups are freed, go through and | |
10098 | * free all the space_info structs. This is only called during | |
10099 | * the final stages of unmount, and so we know nobody is | |
10100 | * using them. We call synchronize_rcu() once before we start, | |
10101 | * just to be on the safe side. | |
10102 | */ | |
10103 | synchronize_rcu(); | |
10104 | ||
8929ecfa YZ |
10105 | release_global_block_rsv(info); |
10106 | ||
67871254 | 10107 | while (!list_empty(&info->space_info)) { |
6ab0a202 JM |
10108 | int i; |
10109 | ||
4184ea7f CM |
10110 | space_info = list_entry(info->space_info.next, |
10111 | struct btrfs_space_info, | |
10112 | list); | |
d555b6c3 JB |
10113 | |
10114 | /* | |
10115 | * Do not hide this behind enospc_debug, this is actually | |
10116 | * important and indicates a real bug if this happens. | |
10117 | */ | |
10118 | if (WARN_ON(space_info->bytes_pinned > 0 || | |
b069e0c3 | 10119 | space_info->bytes_reserved > 0 || |
d555b6c3 | 10120 | space_info->bytes_may_use > 0)) |
ab8d0fc4 | 10121 | dump_space_info(info, space_info, 0, 0); |
4184ea7f | 10122 | list_del(&space_info->list); |
6ab0a202 JM |
10123 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { |
10124 | struct kobject *kobj; | |
c1895442 JM |
10125 | kobj = space_info->block_group_kobjs[i]; |
10126 | space_info->block_group_kobjs[i] = NULL; | |
10127 | if (kobj) { | |
6ab0a202 JM |
10128 | kobject_del(kobj); |
10129 | kobject_put(kobj); | |
10130 | } | |
10131 | } | |
10132 | kobject_del(&space_info->kobj); | |
10133 | kobject_put(&space_info->kobj); | |
4184ea7f | 10134 | } |
1a40e23b ZY |
10135 | return 0; |
10136 | } | |
10137 | ||
75cb379d JM |
10138 | /* link_block_group will queue up kobjects to add when we're reclaim-safe */ |
10139 | void btrfs_add_raid_kobjects(struct btrfs_fs_info *fs_info) | |
10140 | { | |
10141 | struct btrfs_space_info *space_info; | |
10142 | struct raid_kobject *rkobj; | |
10143 | LIST_HEAD(list); | |
10144 | int index; | |
10145 | int ret = 0; | |
10146 | ||
10147 | spin_lock(&fs_info->pending_raid_kobjs_lock); | |
10148 | list_splice_init(&fs_info->pending_raid_kobjs, &list); | |
10149 | spin_unlock(&fs_info->pending_raid_kobjs_lock); | |
10150 | ||
10151 | list_for_each_entry(rkobj, &list, list) { | |
10152 | space_info = __find_space_info(fs_info, rkobj->flags); | |
10153 | index = btrfs_bg_flags_to_raid_index(rkobj->flags); | |
10154 | ||
10155 | ret = kobject_add(&rkobj->kobj, &space_info->kobj, | |
10156 | "%s", get_raid_name(index)); | |
10157 | if (ret) { | |
10158 | kobject_put(&rkobj->kobj); | |
10159 | break; | |
10160 | } | |
10161 | } | |
10162 | if (ret) | |
10163 | btrfs_warn(fs_info, | |
10164 | "failed to add kobject for block cache, ignoring"); | |
10165 | } | |
10166 | ||
c434d21c | 10167 | static void link_block_group(struct btrfs_block_group_cache *cache) |
b742bb82 | 10168 | { |
c434d21c | 10169 | struct btrfs_space_info *space_info = cache->space_info; |
75cb379d | 10170 | struct btrfs_fs_info *fs_info = cache->fs_info; |
3e72ee88 | 10171 | int index = btrfs_bg_flags_to_raid_index(cache->flags); |
ed55b6ac | 10172 | bool first = false; |
b742bb82 YZ |
10173 | |
10174 | down_write(&space_info->groups_sem); | |
ed55b6ac JM |
10175 | if (list_empty(&space_info->block_groups[index])) |
10176 | first = true; | |
10177 | list_add_tail(&cache->list, &space_info->block_groups[index]); | |
10178 | up_write(&space_info->groups_sem); | |
10179 | ||
10180 | if (first) { | |
75cb379d JM |
10181 | struct raid_kobject *rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS); |
10182 | if (!rkobj) { | |
10183 | btrfs_warn(cache->fs_info, | |
10184 | "couldn't alloc memory for raid level kobject"); | |
10185 | return; | |
6ab0a202 | 10186 | } |
75cb379d JM |
10187 | rkobj->flags = cache->flags; |
10188 | kobject_init(&rkobj->kobj, &btrfs_raid_ktype); | |
10189 | ||
10190 | spin_lock(&fs_info->pending_raid_kobjs_lock); | |
10191 | list_add_tail(&rkobj->list, &fs_info->pending_raid_kobjs); | |
10192 | spin_unlock(&fs_info->pending_raid_kobjs_lock); | |
c1895442 | 10193 | space_info->block_group_kobjs[index] = &rkobj->kobj; |
6ab0a202 | 10194 | } |
b742bb82 YZ |
10195 | } |
10196 | ||
920e4a58 | 10197 | static struct btrfs_block_group_cache * |
2ff7e61e JM |
10198 | btrfs_create_block_group_cache(struct btrfs_fs_info *fs_info, |
10199 | u64 start, u64 size) | |
920e4a58 MX |
10200 | { |
10201 | struct btrfs_block_group_cache *cache; | |
10202 | ||
10203 | cache = kzalloc(sizeof(*cache), GFP_NOFS); | |
10204 | if (!cache) | |
10205 | return NULL; | |
10206 | ||
10207 | cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl), | |
10208 | GFP_NOFS); | |
10209 | if (!cache->free_space_ctl) { | |
10210 | kfree(cache); | |
10211 | return NULL; | |
10212 | } | |
10213 | ||
10214 | cache->key.objectid = start; | |
10215 | cache->key.offset = size; | |
10216 | cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; | |
10217 | ||
0b246afa | 10218 | cache->fs_info = fs_info; |
e4ff5fb5 | 10219 | cache->full_stripe_len = btrfs_full_stripe_len(fs_info, start); |
1e144fb8 OS |
10220 | set_free_space_tree_thresholds(cache); |
10221 | ||
920e4a58 MX |
10222 | atomic_set(&cache->count, 1); |
10223 | spin_lock_init(&cache->lock); | |
e570fd27 | 10224 | init_rwsem(&cache->data_rwsem); |
920e4a58 MX |
10225 | INIT_LIST_HEAD(&cache->list); |
10226 | INIT_LIST_HEAD(&cache->cluster_list); | |
47ab2a6c | 10227 | INIT_LIST_HEAD(&cache->bg_list); |
633c0aad | 10228 | INIT_LIST_HEAD(&cache->ro_list); |
ce93ec54 | 10229 | INIT_LIST_HEAD(&cache->dirty_list); |
c9dc4c65 | 10230 | INIT_LIST_HEAD(&cache->io_list); |
920e4a58 | 10231 | btrfs_init_free_space_ctl(cache); |
04216820 | 10232 | atomic_set(&cache->trimming, 0); |
a5ed9182 | 10233 | mutex_init(&cache->free_space_lock); |
0966a7b1 | 10234 | btrfs_init_full_stripe_locks_tree(&cache->full_stripe_locks_root); |
920e4a58 MX |
10235 | |
10236 | return cache; | |
10237 | } | |
10238 | ||
7ef49515 QW |
10239 | |
10240 | /* | |
10241 | * Iterate all chunks and verify that each of them has the corresponding block | |
10242 | * group | |
10243 | */ | |
10244 | static int check_chunk_block_group_mappings(struct btrfs_fs_info *fs_info) | |
10245 | { | |
10246 | struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree; | |
10247 | struct extent_map *em; | |
10248 | struct btrfs_block_group_cache *bg; | |
10249 | u64 start = 0; | |
10250 | int ret = 0; | |
10251 | ||
10252 | while (1) { | |
10253 | read_lock(&map_tree->map_tree.lock); | |
10254 | /* | |
10255 | * lookup_extent_mapping will return the first extent map | |
10256 | * intersecting the range, so setting @len to 1 is enough to | |
10257 | * get the first chunk. | |
10258 | */ | |
10259 | em = lookup_extent_mapping(&map_tree->map_tree, start, 1); | |
10260 | read_unlock(&map_tree->map_tree.lock); | |
10261 | if (!em) | |
10262 | break; | |
10263 | ||
10264 | bg = btrfs_lookup_block_group(fs_info, em->start); | |
10265 | if (!bg) { | |
10266 | btrfs_err(fs_info, | |
10267 | "chunk start=%llu len=%llu doesn't have corresponding block group", | |
10268 | em->start, em->len); | |
10269 | ret = -EUCLEAN; | |
10270 | free_extent_map(em); | |
10271 | break; | |
10272 | } | |
10273 | if (bg->key.objectid != em->start || | |
10274 | bg->key.offset != em->len || | |
10275 | (bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK) != | |
10276 | (em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK)) { | |
10277 | btrfs_err(fs_info, | |
10278 | "chunk start=%llu len=%llu flags=0x%llx doesn't match block group start=%llu len=%llu flags=0x%llx", | |
10279 | em->start, em->len, | |
10280 | em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK, | |
10281 | bg->key.objectid, bg->key.offset, | |
10282 | bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK); | |
10283 | ret = -EUCLEAN; | |
10284 | free_extent_map(em); | |
10285 | btrfs_put_block_group(bg); | |
10286 | break; | |
10287 | } | |
10288 | start = em->start + em->len; | |
10289 | free_extent_map(em); | |
10290 | btrfs_put_block_group(bg); | |
10291 | } | |
10292 | return ret; | |
10293 | } | |
10294 | ||
5b4aacef | 10295 | int btrfs_read_block_groups(struct btrfs_fs_info *info) |
9078a3e1 CM |
10296 | { |
10297 | struct btrfs_path *path; | |
10298 | int ret; | |
9078a3e1 | 10299 | struct btrfs_block_group_cache *cache; |
6324fbf3 | 10300 | struct btrfs_space_info *space_info; |
9078a3e1 CM |
10301 | struct btrfs_key key; |
10302 | struct btrfs_key found_key; | |
5f39d397 | 10303 | struct extent_buffer *leaf; |
0af3d00b JB |
10304 | int need_clear = 0; |
10305 | u64 cache_gen; | |
49303381 LB |
10306 | u64 feature; |
10307 | int mixed; | |
10308 | ||
10309 | feature = btrfs_super_incompat_flags(info->super_copy); | |
10310 | mixed = !!(feature & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS); | |
96b5179d | 10311 | |
9078a3e1 | 10312 | key.objectid = 0; |
0b86a832 | 10313 | key.offset = 0; |
962a298f | 10314 | key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; |
9078a3e1 CM |
10315 | path = btrfs_alloc_path(); |
10316 | if (!path) | |
10317 | return -ENOMEM; | |
e4058b54 | 10318 | path->reada = READA_FORWARD; |
9078a3e1 | 10319 | |
0b246afa JM |
10320 | cache_gen = btrfs_super_cache_generation(info->super_copy); |
10321 | if (btrfs_test_opt(info, SPACE_CACHE) && | |
10322 | btrfs_super_generation(info->super_copy) != cache_gen) | |
0af3d00b | 10323 | need_clear = 1; |
0b246afa | 10324 | if (btrfs_test_opt(info, CLEAR_CACHE)) |
88c2ba3b | 10325 | need_clear = 1; |
0af3d00b | 10326 | |
d397712b | 10327 | while (1) { |
6bccf3ab | 10328 | ret = find_first_block_group(info, path, &key); |
b742bb82 YZ |
10329 | if (ret > 0) |
10330 | break; | |
0b86a832 CM |
10331 | if (ret != 0) |
10332 | goto error; | |
920e4a58 | 10333 | |
5f39d397 CM |
10334 | leaf = path->nodes[0]; |
10335 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
920e4a58 | 10336 | |
2ff7e61e | 10337 | cache = btrfs_create_block_group_cache(info, found_key.objectid, |
920e4a58 | 10338 | found_key.offset); |
9078a3e1 | 10339 | if (!cache) { |
0b86a832 | 10340 | ret = -ENOMEM; |
f0486c68 | 10341 | goto error; |
9078a3e1 | 10342 | } |
96303081 | 10343 | |
cf7c1ef6 LB |
10344 | if (need_clear) { |
10345 | /* | |
10346 | * When we mount with old space cache, we need to | |
10347 | * set BTRFS_DC_CLEAR and set dirty flag. | |
10348 | * | |
10349 | * a) Setting 'BTRFS_DC_CLEAR' makes sure that we | |
10350 | * truncate the old free space cache inode and | |
10351 | * setup a new one. | |
10352 | * b) Setting 'dirty flag' makes sure that we flush | |
10353 | * the new space cache info onto disk. | |
10354 | */ | |
0b246afa | 10355 | if (btrfs_test_opt(info, SPACE_CACHE)) |
ce93ec54 | 10356 | cache->disk_cache_state = BTRFS_DC_CLEAR; |
cf7c1ef6 | 10357 | } |
0af3d00b | 10358 | |
5f39d397 CM |
10359 | read_extent_buffer(leaf, &cache->item, |
10360 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
10361 | sizeof(cache->item)); | |
920e4a58 | 10362 | cache->flags = btrfs_block_group_flags(&cache->item); |
49303381 LB |
10363 | if (!mixed && |
10364 | ((cache->flags & BTRFS_BLOCK_GROUP_METADATA) && | |
10365 | (cache->flags & BTRFS_BLOCK_GROUP_DATA))) { | |
10366 | btrfs_err(info, | |
10367 | "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups", | |
10368 | cache->key.objectid); | |
10369 | ret = -EINVAL; | |
10370 | goto error; | |
10371 | } | |
0b86a832 | 10372 | |
9078a3e1 | 10373 | key.objectid = found_key.objectid + found_key.offset; |
b3b4aa74 | 10374 | btrfs_release_path(path); |
34d52cb6 | 10375 | |
3c14874a JB |
10376 | /* |
10377 | * We need to exclude the super stripes now so that the space | |
10378 | * info has super bytes accounted for, otherwise we'll think | |
10379 | * we have more space than we actually do. | |
10380 | */ | |
3c4da657 | 10381 | ret = exclude_super_stripes(cache); |
835d974f JB |
10382 | if (ret) { |
10383 | /* | |
10384 | * We may have excluded something, so call this just in | |
10385 | * case. | |
10386 | */ | |
9e715da8 | 10387 | free_excluded_extents(cache); |
920e4a58 | 10388 | btrfs_put_block_group(cache); |
835d974f JB |
10389 | goto error; |
10390 | } | |
3c14874a | 10391 | |
817d52f8 JB |
10392 | /* |
10393 | * check for two cases, either we are full, and therefore | |
10394 | * don't need to bother with the caching work since we won't | |
10395 | * find any space, or we are empty, and we can just add all | |
52042d8e | 10396 | * the space in and be done with it. This saves us _a_lot_ of |
817d52f8 JB |
10397 | * time, particularly in the full case. |
10398 | */ | |
10399 | if (found_key.offset == btrfs_block_group_used(&cache->item)) { | |
11833d66 | 10400 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 10401 | cache->cached = BTRFS_CACHE_FINISHED; |
9e715da8 | 10402 | free_excluded_extents(cache); |
817d52f8 | 10403 | } else if (btrfs_block_group_used(&cache->item) == 0) { |
11833d66 | 10404 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 10405 | cache->cached = BTRFS_CACHE_FINISHED; |
4457c1c7 | 10406 | add_new_free_space(cache, found_key.objectid, |
817d52f8 JB |
10407 | found_key.objectid + |
10408 | found_key.offset); | |
9e715da8 | 10409 | free_excluded_extents(cache); |
817d52f8 | 10410 | } |
96b5179d | 10411 | |
0b246afa | 10412 | ret = btrfs_add_block_group_cache(info, cache); |
8c579fe7 JB |
10413 | if (ret) { |
10414 | btrfs_remove_free_space_cache(cache); | |
10415 | btrfs_put_block_group(cache); | |
10416 | goto error; | |
10417 | } | |
10418 | ||
0b246afa | 10419 | trace_btrfs_add_block_group(info, cache, 0); |
d2006e6d NB |
10420 | update_space_info(info, cache->flags, found_key.offset, |
10421 | btrfs_block_group_used(&cache->item), | |
10422 | cache->bytes_super, &space_info); | |
8c579fe7 | 10423 | |
6324fbf3 | 10424 | cache->space_info = space_info; |
1b2da372 | 10425 | |
c434d21c | 10426 | link_block_group(cache); |
0f9dd46c | 10427 | |
0b246afa | 10428 | set_avail_alloc_bits(info, cache->flags); |
2ff7e61e | 10429 | if (btrfs_chunk_readonly(info, cache->key.objectid)) { |
868f401a | 10430 | inc_block_group_ro(cache, 1); |
47ab2a6c | 10431 | } else if (btrfs_block_group_used(&cache->item) == 0) { |
031f24da QW |
10432 | ASSERT(list_empty(&cache->bg_list)); |
10433 | btrfs_mark_bg_unused(cache); | |
47ab2a6c | 10434 | } |
9078a3e1 | 10435 | } |
b742bb82 | 10436 | |
0b246afa | 10437 | list_for_each_entry_rcu(space_info, &info->space_info, list) { |
2ff7e61e | 10438 | if (!(get_alloc_profile(info, space_info->flags) & |
b742bb82 YZ |
10439 | (BTRFS_BLOCK_GROUP_RAID10 | |
10440 | BTRFS_BLOCK_GROUP_RAID1 | | |
53b381b3 DW |
10441 | BTRFS_BLOCK_GROUP_RAID5 | |
10442 | BTRFS_BLOCK_GROUP_RAID6 | | |
b742bb82 YZ |
10443 | BTRFS_BLOCK_GROUP_DUP))) |
10444 | continue; | |
10445 | /* | |
10446 | * avoid allocating from un-mirrored block group if there are | |
10447 | * mirrored block groups. | |
10448 | */ | |
1095cc0d | 10449 | list_for_each_entry(cache, |
10450 | &space_info->block_groups[BTRFS_RAID_RAID0], | |
10451 | list) | |
868f401a | 10452 | inc_block_group_ro(cache, 1); |
1095cc0d | 10453 | list_for_each_entry(cache, |
10454 | &space_info->block_groups[BTRFS_RAID_SINGLE], | |
10455 | list) | |
868f401a | 10456 | inc_block_group_ro(cache, 1); |
9078a3e1 | 10457 | } |
f0486c68 | 10458 | |
75cb379d | 10459 | btrfs_add_raid_kobjects(info); |
f0486c68 | 10460 | init_global_block_rsv(info); |
7ef49515 | 10461 | ret = check_chunk_block_group_mappings(info); |
0b86a832 | 10462 | error: |
9078a3e1 | 10463 | btrfs_free_path(path); |
0b86a832 | 10464 | return ret; |
9078a3e1 | 10465 | } |
6324fbf3 | 10466 | |
6c686b35 | 10467 | void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans) |
ea658bad | 10468 | { |
6c686b35 | 10469 | struct btrfs_fs_info *fs_info = trans->fs_info; |
545e3366 | 10470 | struct btrfs_block_group_cache *block_group; |
0b246afa | 10471 | struct btrfs_root *extent_root = fs_info->extent_root; |
ea658bad JB |
10472 | struct btrfs_block_group_item item; |
10473 | struct btrfs_key key; | |
10474 | int ret = 0; | |
10475 | ||
5ce55557 FM |
10476 | if (!trans->can_flush_pending_bgs) |
10477 | return; | |
10478 | ||
545e3366 JB |
10479 | while (!list_empty(&trans->new_bgs)) { |
10480 | block_group = list_first_entry(&trans->new_bgs, | |
10481 | struct btrfs_block_group_cache, | |
10482 | bg_list); | |
ea658bad | 10483 | if (ret) |
c92f6be3 | 10484 | goto next; |
ea658bad JB |
10485 | |
10486 | spin_lock(&block_group->lock); | |
10487 | memcpy(&item, &block_group->item, sizeof(item)); | |
10488 | memcpy(&key, &block_group->key, sizeof(key)); | |
10489 | spin_unlock(&block_group->lock); | |
10490 | ||
10491 | ret = btrfs_insert_item(trans, extent_root, &key, &item, | |
10492 | sizeof(item)); | |
10493 | if (ret) | |
66642832 | 10494 | btrfs_abort_transaction(trans, ret); |
97aff912 | 10495 | ret = btrfs_finish_chunk_alloc(trans, key.objectid, key.offset); |
6df9a95e | 10496 | if (ret) |
66642832 | 10497 | btrfs_abort_transaction(trans, ret); |
e4e0711c | 10498 | add_block_group_free_space(trans, block_group); |
1e144fb8 | 10499 | /* already aborted the transaction if it failed. */ |
c92f6be3 | 10500 | next: |
ba2c4d4e | 10501 | btrfs_delayed_refs_rsv_release(fs_info, 1); |
c92f6be3 | 10502 | list_del_init(&block_group->bg_list); |
ea658bad | 10503 | } |
5ce55557 | 10504 | btrfs_trans_release_chunk_metadata(trans); |
ea658bad JB |
10505 | } |
10506 | ||
e7e02096 | 10507 | int btrfs_make_block_group(struct btrfs_trans_handle *trans, u64 bytes_used, |
0174484d | 10508 | u64 type, u64 chunk_offset, u64 size) |
6324fbf3 | 10509 | { |
e7e02096 | 10510 | struct btrfs_fs_info *fs_info = trans->fs_info; |
6324fbf3 | 10511 | struct btrfs_block_group_cache *cache; |
0b246afa | 10512 | int ret; |
6324fbf3 | 10513 | |
90787766 | 10514 | btrfs_set_log_full_commit(trans); |
e02119d5 | 10515 | |
2ff7e61e | 10516 | cache = btrfs_create_block_group_cache(fs_info, chunk_offset, size); |
0f9dd46c JB |
10517 | if (!cache) |
10518 | return -ENOMEM; | |
34d52cb6 | 10519 | |
6324fbf3 | 10520 | btrfs_set_block_group_used(&cache->item, bytes_used); |
0174484d NB |
10521 | btrfs_set_block_group_chunk_objectid(&cache->item, |
10522 | BTRFS_FIRST_CHUNK_TREE_OBJECTID); | |
6324fbf3 CM |
10523 | btrfs_set_block_group_flags(&cache->item, type); |
10524 | ||
920e4a58 | 10525 | cache->flags = type; |
11833d66 | 10526 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 10527 | cache->cached = BTRFS_CACHE_FINISHED; |
1e144fb8 | 10528 | cache->needs_free_space = 1; |
3c4da657 | 10529 | ret = exclude_super_stripes(cache); |
835d974f JB |
10530 | if (ret) { |
10531 | /* | |
10532 | * We may have excluded something, so call this just in | |
10533 | * case. | |
10534 | */ | |
9e715da8 | 10535 | free_excluded_extents(cache); |
920e4a58 | 10536 | btrfs_put_block_group(cache); |
835d974f JB |
10537 | return ret; |
10538 | } | |
96303081 | 10539 | |
4457c1c7 | 10540 | add_new_free_space(cache, chunk_offset, chunk_offset + size); |
817d52f8 | 10541 | |
9e715da8 | 10542 | free_excluded_extents(cache); |
11833d66 | 10543 | |
d0bd4560 | 10544 | #ifdef CONFIG_BTRFS_DEBUG |
2ff7e61e | 10545 | if (btrfs_should_fragment_free_space(cache)) { |
d0bd4560 JB |
10546 | u64 new_bytes_used = size - bytes_used; |
10547 | ||
10548 | bytes_used += new_bytes_used >> 1; | |
2ff7e61e | 10549 | fragment_free_space(cache); |
d0bd4560 JB |
10550 | } |
10551 | #endif | |
2e6e5183 | 10552 | /* |
2be12ef7 NB |
10553 | * Ensure the corresponding space_info object is created and |
10554 | * assigned to our block group. We want our bg to be added to the rbtree | |
10555 | * with its ->space_info set. | |
2e6e5183 | 10556 | */ |
2be12ef7 | 10557 | cache->space_info = __find_space_info(fs_info, cache->flags); |
dc2d3005 | 10558 | ASSERT(cache->space_info); |
2e6e5183 | 10559 | |
0b246afa | 10560 | ret = btrfs_add_block_group_cache(fs_info, cache); |
8c579fe7 JB |
10561 | if (ret) { |
10562 | btrfs_remove_free_space_cache(cache); | |
10563 | btrfs_put_block_group(cache); | |
10564 | return ret; | |
10565 | } | |
10566 | ||
2e6e5183 FM |
10567 | /* |
10568 | * Now that our block group has its ->space_info set and is inserted in | |
10569 | * the rbtree, update the space info's counters. | |
10570 | */ | |
0b246afa | 10571 | trace_btrfs_add_block_group(fs_info, cache, 1); |
d2006e6d | 10572 | update_space_info(fs_info, cache->flags, size, bytes_used, |
e40edf2d | 10573 | cache->bytes_super, &cache->space_info); |
0b246afa | 10574 | update_global_block_rsv(fs_info); |
1b2da372 | 10575 | |
c434d21c | 10576 | link_block_group(cache); |
6324fbf3 | 10577 | |
47ab2a6c | 10578 | list_add_tail(&cache->bg_list, &trans->new_bgs); |
ba2c4d4e JB |
10579 | trans->delayed_ref_updates++; |
10580 | btrfs_update_delayed_refs_rsv(trans); | |
6324fbf3 | 10581 | |
0b246afa | 10582 | set_avail_alloc_bits(fs_info, type); |
6324fbf3 CM |
10583 | return 0; |
10584 | } | |
1a40e23b | 10585 | |
10ea00f5 ID |
10586 | static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
10587 | { | |
899c81ea ID |
10588 | u64 extra_flags = chunk_to_extended(flags) & |
10589 | BTRFS_EXTENDED_PROFILE_MASK; | |
10ea00f5 | 10590 | |
de98ced9 | 10591 | write_seqlock(&fs_info->profiles_lock); |
10ea00f5 ID |
10592 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
10593 | fs_info->avail_data_alloc_bits &= ~extra_flags; | |
10594 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
10595 | fs_info->avail_metadata_alloc_bits &= ~extra_flags; | |
10596 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
10597 | fs_info->avail_system_alloc_bits &= ~extra_flags; | |
de98ced9 | 10598 | write_sequnlock(&fs_info->profiles_lock); |
10ea00f5 ID |
10599 | } |
10600 | ||
1a40e23b | 10601 | int btrfs_remove_block_group(struct btrfs_trans_handle *trans, |
5a98ec01 | 10602 | u64 group_start, struct extent_map *em) |
1a40e23b | 10603 | { |
5a98ec01 | 10604 | struct btrfs_fs_info *fs_info = trans->fs_info; |
6bccf3ab | 10605 | struct btrfs_root *root = fs_info->extent_root; |
1a40e23b ZY |
10606 | struct btrfs_path *path; |
10607 | struct btrfs_block_group_cache *block_group; | |
44fb5511 | 10608 | struct btrfs_free_cluster *cluster; |
0b246afa | 10609 | struct btrfs_root *tree_root = fs_info->tree_root; |
1a40e23b | 10610 | struct btrfs_key key; |
0af3d00b | 10611 | struct inode *inode; |
c1895442 | 10612 | struct kobject *kobj = NULL; |
1a40e23b | 10613 | int ret; |
10ea00f5 | 10614 | int index; |
89a55897 | 10615 | int factor; |
4f69cb98 | 10616 | struct btrfs_caching_control *caching_ctl = NULL; |
04216820 | 10617 | bool remove_em; |
ba2c4d4e | 10618 | bool remove_rsv = false; |
1a40e23b | 10619 | |
6bccf3ab | 10620 | block_group = btrfs_lookup_block_group(fs_info, group_start); |
1a40e23b | 10621 | BUG_ON(!block_group); |
c146afad | 10622 | BUG_ON(!block_group->ro); |
1a40e23b | 10623 | |
4ed0a7a3 | 10624 | trace_btrfs_remove_block_group(block_group); |
9f7c43c9 | 10625 | /* |
10626 | * Free the reserved super bytes from this block group before | |
10627 | * remove it. | |
10628 | */ | |
9e715da8 | 10629 | free_excluded_extents(block_group); |
fd708b81 JB |
10630 | btrfs_free_ref_tree_range(fs_info, block_group->key.objectid, |
10631 | block_group->key.offset); | |
9f7c43c9 | 10632 | |
1a40e23b | 10633 | memcpy(&key, &block_group->key, sizeof(key)); |
3e72ee88 | 10634 | index = btrfs_bg_flags_to_raid_index(block_group->flags); |
46df06b8 | 10635 | factor = btrfs_bg_type_to_factor(block_group->flags); |
1a40e23b | 10636 | |
44fb5511 | 10637 | /* make sure this block group isn't part of an allocation cluster */ |
0b246afa | 10638 | cluster = &fs_info->data_alloc_cluster; |
44fb5511 CM |
10639 | spin_lock(&cluster->refill_lock); |
10640 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
10641 | spin_unlock(&cluster->refill_lock); | |
10642 | ||
10643 | /* | |
10644 | * make sure this block group isn't part of a metadata | |
10645 | * allocation cluster | |
10646 | */ | |
0b246afa | 10647 | cluster = &fs_info->meta_alloc_cluster; |
44fb5511 CM |
10648 | spin_lock(&cluster->refill_lock); |
10649 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
10650 | spin_unlock(&cluster->refill_lock); | |
10651 | ||
1a40e23b | 10652 | path = btrfs_alloc_path(); |
d8926bb3 MF |
10653 | if (!path) { |
10654 | ret = -ENOMEM; | |
10655 | goto out; | |
10656 | } | |
1a40e23b | 10657 | |
1bbc621e CM |
10658 | /* |
10659 | * get the inode first so any iput calls done for the io_list | |
10660 | * aren't the final iput (no unlinks allowed now) | |
10661 | */ | |
7949f339 | 10662 | inode = lookup_free_space_inode(block_group, path); |
1bbc621e CM |
10663 | |
10664 | mutex_lock(&trans->transaction->cache_write_mutex); | |
10665 | /* | |
52042d8e | 10666 | * Make sure our free space cache IO is done before removing the |
1bbc621e CM |
10667 | * free space inode |
10668 | */ | |
10669 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
10670 | if (!list_empty(&block_group->io_list)) { | |
10671 | list_del_init(&block_group->io_list); | |
10672 | ||
10673 | WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode); | |
10674 | ||
10675 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
afdb5718 | 10676 | btrfs_wait_cache_io(trans, block_group, path); |
1bbc621e CM |
10677 | btrfs_put_block_group(block_group); |
10678 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
10679 | } | |
10680 | ||
10681 | if (!list_empty(&block_group->dirty_list)) { | |
10682 | list_del_init(&block_group->dirty_list); | |
ba2c4d4e | 10683 | remove_rsv = true; |
1bbc621e CM |
10684 | btrfs_put_block_group(block_group); |
10685 | } | |
10686 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
10687 | mutex_unlock(&trans->transaction->cache_write_mutex); | |
10688 | ||
0af3d00b | 10689 | if (!IS_ERR(inode)) { |
73f2e545 | 10690 | ret = btrfs_orphan_add(trans, BTRFS_I(inode)); |
79787eaa JM |
10691 | if (ret) { |
10692 | btrfs_add_delayed_iput(inode); | |
10693 | goto out; | |
10694 | } | |
0af3d00b JB |
10695 | clear_nlink(inode); |
10696 | /* One for the block groups ref */ | |
10697 | spin_lock(&block_group->lock); | |
10698 | if (block_group->iref) { | |
10699 | block_group->iref = 0; | |
10700 | block_group->inode = NULL; | |
10701 | spin_unlock(&block_group->lock); | |
10702 | iput(inode); | |
10703 | } else { | |
10704 | spin_unlock(&block_group->lock); | |
10705 | } | |
10706 | /* One for our lookup ref */ | |
455757c3 | 10707 | btrfs_add_delayed_iput(inode); |
0af3d00b JB |
10708 | } |
10709 | ||
10710 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
10711 | key.offset = block_group->key.objectid; | |
10712 | key.type = 0; | |
10713 | ||
10714 | ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1); | |
10715 | if (ret < 0) | |
10716 | goto out; | |
10717 | if (ret > 0) | |
b3b4aa74 | 10718 | btrfs_release_path(path); |
0af3d00b JB |
10719 | if (ret == 0) { |
10720 | ret = btrfs_del_item(trans, tree_root, path); | |
10721 | if (ret) | |
10722 | goto out; | |
b3b4aa74 | 10723 | btrfs_release_path(path); |
0af3d00b JB |
10724 | } |
10725 | ||
0b246afa | 10726 | spin_lock(&fs_info->block_group_cache_lock); |
1a40e23b | 10727 | rb_erase(&block_group->cache_node, |
0b246afa | 10728 | &fs_info->block_group_cache_tree); |
292cbd51 | 10729 | RB_CLEAR_NODE(&block_group->cache_node); |
a1897fdd | 10730 | |
0b246afa JM |
10731 | if (fs_info->first_logical_byte == block_group->key.objectid) |
10732 | fs_info->first_logical_byte = (u64)-1; | |
10733 | spin_unlock(&fs_info->block_group_cache_lock); | |
817d52f8 | 10734 | |
80eb234a | 10735 | down_write(&block_group->space_info->groups_sem); |
44fb5511 CM |
10736 | /* |
10737 | * we must use list_del_init so people can check to see if they | |
10738 | * are still on the list after taking the semaphore | |
10739 | */ | |
10740 | list_del_init(&block_group->list); | |
6ab0a202 | 10741 | if (list_empty(&block_group->space_info->block_groups[index])) { |
c1895442 JM |
10742 | kobj = block_group->space_info->block_group_kobjs[index]; |
10743 | block_group->space_info->block_group_kobjs[index] = NULL; | |
0b246afa | 10744 | clear_avail_alloc_bits(fs_info, block_group->flags); |
6ab0a202 | 10745 | } |
80eb234a | 10746 | up_write(&block_group->space_info->groups_sem); |
c1895442 JM |
10747 | if (kobj) { |
10748 | kobject_del(kobj); | |
10749 | kobject_put(kobj); | |
10750 | } | |
1a40e23b | 10751 | |
4f69cb98 FM |
10752 | if (block_group->has_caching_ctl) |
10753 | caching_ctl = get_caching_control(block_group); | |
817d52f8 | 10754 | if (block_group->cached == BTRFS_CACHE_STARTED) |
11833d66 | 10755 | wait_block_group_cache_done(block_group); |
4f69cb98 | 10756 | if (block_group->has_caching_ctl) { |
0b246afa | 10757 | down_write(&fs_info->commit_root_sem); |
4f69cb98 FM |
10758 | if (!caching_ctl) { |
10759 | struct btrfs_caching_control *ctl; | |
10760 | ||
10761 | list_for_each_entry(ctl, | |
0b246afa | 10762 | &fs_info->caching_block_groups, list) |
4f69cb98 FM |
10763 | if (ctl->block_group == block_group) { |
10764 | caching_ctl = ctl; | |
1e4f4714 | 10765 | refcount_inc(&caching_ctl->count); |
4f69cb98 FM |
10766 | break; |
10767 | } | |
10768 | } | |
10769 | if (caching_ctl) | |
10770 | list_del_init(&caching_ctl->list); | |
0b246afa | 10771 | up_write(&fs_info->commit_root_sem); |
4f69cb98 FM |
10772 | if (caching_ctl) { |
10773 | /* Once for the caching bgs list and once for us. */ | |
10774 | put_caching_control(caching_ctl); | |
10775 | put_caching_control(caching_ctl); | |
10776 | } | |
10777 | } | |
817d52f8 | 10778 | |
ce93ec54 | 10779 | spin_lock(&trans->transaction->dirty_bgs_lock); |
9a0ec83d NB |
10780 | WARN_ON(!list_empty(&block_group->dirty_list)); |
10781 | WARN_ON(!list_empty(&block_group->io_list)); | |
ce93ec54 | 10782 | spin_unlock(&trans->transaction->dirty_bgs_lock); |
9a0ec83d | 10783 | |
817d52f8 JB |
10784 | btrfs_remove_free_space_cache(block_group); |
10785 | ||
c146afad | 10786 | spin_lock(&block_group->space_info->lock); |
75c68e9f | 10787 | list_del_init(&block_group->ro_list); |
18d018ad | 10788 | |
0b246afa | 10789 | if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
18d018ad ZL |
10790 | WARN_ON(block_group->space_info->total_bytes |
10791 | < block_group->key.offset); | |
10792 | WARN_ON(block_group->space_info->bytes_readonly | |
10793 | < block_group->key.offset); | |
10794 | WARN_ON(block_group->space_info->disk_total | |
10795 | < block_group->key.offset * factor); | |
10796 | } | |
c146afad YZ |
10797 | block_group->space_info->total_bytes -= block_group->key.offset; |
10798 | block_group->space_info->bytes_readonly -= block_group->key.offset; | |
89a55897 | 10799 | block_group->space_info->disk_total -= block_group->key.offset * factor; |
18d018ad | 10800 | |
c146afad | 10801 | spin_unlock(&block_group->space_info->lock); |
283bb197 | 10802 | |
0af3d00b JB |
10803 | memcpy(&key, &block_group->key, sizeof(key)); |
10804 | ||
34441361 | 10805 | mutex_lock(&fs_info->chunk_mutex); |
04216820 FM |
10806 | spin_lock(&block_group->lock); |
10807 | block_group->removed = 1; | |
10808 | /* | |
10809 | * At this point trimming can't start on this block group, because we | |
10810 | * removed the block group from the tree fs_info->block_group_cache_tree | |
10811 | * so no one can't find it anymore and even if someone already got this | |
10812 | * block group before we removed it from the rbtree, they have already | |
10813 | * incremented block_group->trimming - if they didn't, they won't find | |
10814 | * any free space entries because we already removed them all when we | |
10815 | * called btrfs_remove_free_space_cache(). | |
10816 | * | |
10817 | * And we must not remove the extent map from the fs_info->mapping_tree | |
10818 | * to prevent the same logical address range and physical device space | |
10819 | * ranges from being reused for a new block group. This is because our | |
10820 | * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is | |
10821 | * completely transactionless, so while it is trimming a range the | |
10822 | * currently running transaction might finish and a new one start, | |
10823 | * allowing for new block groups to be created that can reuse the same | |
10824 | * physical device locations unless we take this special care. | |
e33e17ee JM |
10825 | * |
10826 | * There may also be an implicit trim operation if the file system | |
10827 | * is mounted with -odiscard. The same protections must remain | |
10828 | * in place until the extents have been discarded completely when | |
10829 | * the transaction commit has completed. | |
04216820 FM |
10830 | */ |
10831 | remove_em = (atomic_read(&block_group->trimming) == 0); | |
04216820 | 10832 | spin_unlock(&block_group->lock); |
04216820 FM |
10833 | |
10834 | if (remove_em) { | |
10835 | struct extent_map_tree *em_tree; | |
10836 | ||
0b246afa | 10837 | em_tree = &fs_info->mapping_tree.map_tree; |
04216820 FM |
10838 | write_lock(&em_tree->lock); |
10839 | remove_extent_mapping(em_tree, em); | |
10840 | write_unlock(&em_tree->lock); | |
10841 | /* once for the tree */ | |
10842 | free_extent_map(em); | |
10843 | } | |
10844 | ||
34441361 | 10845 | mutex_unlock(&fs_info->chunk_mutex); |
8dbcd10f | 10846 | |
f3f72779 | 10847 | ret = remove_block_group_free_space(trans, block_group); |
1e144fb8 OS |
10848 | if (ret) |
10849 | goto out; | |
10850 | ||
fa9c0d79 CM |
10851 | btrfs_put_block_group(block_group); |
10852 | btrfs_put_block_group(block_group); | |
1a40e23b ZY |
10853 | |
10854 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
10855 | if (ret > 0) | |
10856 | ret = -EIO; | |
10857 | if (ret < 0) | |
10858 | goto out; | |
10859 | ||
10860 | ret = btrfs_del_item(trans, root, path); | |
10861 | out: | |
ba2c4d4e JB |
10862 | if (remove_rsv) |
10863 | btrfs_delayed_refs_rsv_release(fs_info, 1); | |
1a40e23b ZY |
10864 | btrfs_free_path(path); |
10865 | return ret; | |
10866 | } | |
acce952b | 10867 | |
8eab77ff | 10868 | struct btrfs_trans_handle * |
7fd01182 FM |
10869 | btrfs_start_trans_remove_block_group(struct btrfs_fs_info *fs_info, |
10870 | const u64 chunk_offset) | |
8eab77ff | 10871 | { |
7fd01182 FM |
10872 | struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree; |
10873 | struct extent_map *em; | |
10874 | struct map_lookup *map; | |
10875 | unsigned int num_items; | |
10876 | ||
10877 | read_lock(&em_tree->lock); | |
10878 | em = lookup_extent_mapping(em_tree, chunk_offset, 1); | |
10879 | read_unlock(&em_tree->lock); | |
10880 | ASSERT(em && em->start == chunk_offset); | |
10881 | ||
8eab77ff | 10882 | /* |
7fd01182 FM |
10883 | * We need to reserve 3 + N units from the metadata space info in order |
10884 | * to remove a block group (done at btrfs_remove_chunk() and at | |
10885 | * btrfs_remove_block_group()), which are used for: | |
10886 | * | |
8eab77ff FM |
10887 | * 1 unit for adding the free space inode's orphan (located in the tree |
10888 | * of tree roots). | |
7fd01182 FM |
10889 | * 1 unit for deleting the block group item (located in the extent |
10890 | * tree). | |
10891 | * 1 unit for deleting the free space item (located in tree of tree | |
10892 | * roots). | |
10893 | * N units for deleting N device extent items corresponding to each | |
10894 | * stripe (located in the device tree). | |
10895 | * | |
10896 | * In order to remove a block group we also need to reserve units in the | |
10897 | * system space info in order to update the chunk tree (update one or | |
10898 | * more device items and remove one chunk item), but this is done at | |
10899 | * btrfs_remove_chunk() through a call to check_system_chunk(). | |
8eab77ff | 10900 | */ |
95617d69 | 10901 | map = em->map_lookup; |
7fd01182 FM |
10902 | num_items = 3 + map->num_stripes; |
10903 | free_extent_map(em); | |
10904 | ||
8eab77ff | 10905 | return btrfs_start_transaction_fallback_global_rsv(fs_info->extent_root, |
7fd01182 | 10906 | num_items, 1); |
8eab77ff FM |
10907 | } |
10908 | ||
47ab2a6c JB |
10909 | /* |
10910 | * Process the unused_bgs list and remove any that don't have any allocated | |
10911 | * space inside of them. | |
10912 | */ | |
10913 | void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info) | |
10914 | { | |
10915 | struct btrfs_block_group_cache *block_group; | |
10916 | struct btrfs_space_info *space_info; | |
47ab2a6c JB |
10917 | struct btrfs_trans_handle *trans; |
10918 | int ret = 0; | |
10919 | ||
afcdd129 | 10920 | if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags)) |
47ab2a6c JB |
10921 | return; |
10922 | ||
10923 | spin_lock(&fs_info->unused_bgs_lock); | |
10924 | while (!list_empty(&fs_info->unused_bgs)) { | |
10925 | u64 start, end; | |
e33e17ee | 10926 | int trimming; |
47ab2a6c JB |
10927 | |
10928 | block_group = list_first_entry(&fs_info->unused_bgs, | |
10929 | struct btrfs_block_group_cache, | |
10930 | bg_list); | |
47ab2a6c | 10931 | list_del_init(&block_group->bg_list); |
aefbe9a6 ZL |
10932 | |
10933 | space_info = block_group->space_info; | |
10934 | ||
47ab2a6c JB |
10935 | if (ret || btrfs_mixed_space_info(space_info)) { |
10936 | btrfs_put_block_group(block_group); | |
10937 | continue; | |
10938 | } | |
10939 | spin_unlock(&fs_info->unused_bgs_lock); | |
10940 | ||
d5f2e33b | 10941 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
67c5e7d4 | 10942 | |
47ab2a6c JB |
10943 | /* Don't want to race with allocators so take the groups_sem */ |
10944 | down_write(&space_info->groups_sem); | |
10945 | spin_lock(&block_group->lock); | |
43794446 | 10946 | if (block_group->reserved || block_group->pinned || |
47ab2a6c | 10947 | btrfs_block_group_used(&block_group->item) || |
19c4d2f9 | 10948 | block_group->ro || |
aefbe9a6 | 10949 | list_is_singular(&block_group->list)) { |
47ab2a6c JB |
10950 | /* |
10951 | * We want to bail if we made new allocations or have | |
10952 | * outstanding allocations in this block group. We do | |
10953 | * the ro check in case balance is currently acting on | |
10954 | * this block group. | |
10955 | */ | |
4ed0a7a3 | 10956 | trace_btrfs_skip_unused_block_group(block_group); |
47ab2a6c JB |
10957 | spin_unlock(&block_group->lock); |
10958 | up_write(&space_info->groups_sem); | |
10959 | goto next; | |
10960 | } | |
10961 | spin_unlock(&block_group->lock); | |
10962 | ||
10963 | /* We don't want to force the issue, only flip if it's ok. */ | |
868f401a | 10964 | ret = inc_block_group_ro(block_group, 0); |
47ab2a6c JB |
10965 | up_write(&space_info->groups_sem); |
10966 | if (ret < 0) { | |
10967 | ret = 0; | |
10968 | goto next; | |
10969 | } | |
10970 | ||
10971 | /* | |
10972 | * Want to do this before we do anything else so we can recover | |
10973 | * properly if we fail to join the transaction. | |
10974 | */ | |
7fd01182 FM |
10975 | trans = btrfs_start_trans_remove_block_group(fs_info, |
10976 | block_group->key.objectid); | |
47ab2a6c | 10977 | if (IS_ERR(trans)) { |
2ff7e61e | 10978 | btrfs_dec_block_group_ro(block_group); |
47ab2a6c JB |
10979 | ret = PTR_ERR(trans); |
10980 | goto next; | |
10981 | } | |
10982 | ||
10983 | /* | |
10984 | * We could have pending pinned extents for this block group, | |
10985 | * just delete them, we don't care about them anymore. | |
10986 | */ | |
10987 | start = block_group->key.objectid; | |
10988 | end = start + block_group->key.offset - 1; | |
d4b450cd FM |
10989 | /* |
10990 | * Hold the unused_bg_unpin_mutex lock to avoid racing with | |
10991 | * btrfs_finish_extent_commit(). If we are at transaction N, | |
10992 | * another task might be running finish_extent_commit() for the | |
10993 | * previous transaction N - 1, and have seen a range belonging | |
10994 | * to the block group in freed_extents[] before we were able to | |
10995 | * clear the whole block group range from freed_extents[]. This | |
10996 | * means that task can lookup for the block group after we | |
10997 | * unpinned it from freed_extents[] and removed it, leading to | |
10998 | * a BUG_ON() at btrfs_unpin_extent_range(). | |
10999 | */ | |
11000 | mutex_lock(&fs_info->unused_bg_unpin_mutex); | |
758eb51e | 11001 | ret = clear_extent_bits(&fs_info->freed_extents[0], start, end, |
91166212 | 11002 | EXTENT_DIRTY); |
758eb51e | 11003 | if (ret) { |
d4b450cd | 11004 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
2ff7e61e | 11005 | btrfs_dec_block_group_ro(block_group); |
758eb51e FM |
11006 | goto end_trans; |
11007 | } | |
11008 | ret = clear_extent_bits(&fs_info->freed_extents[1], start, end, | |
91166212 | 11009 | EXTENT_DIRTY); |
758eb51e | 11010 | if (ret) { |
d4b450cd | 11011 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
2ff7e61e | 11012 | btrfs_dec_block_group_ro(block_group); |
758eb51e FM |
11013 | goto end_trans; |
11014 | } | |
d4b450cd | 11015 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
47ab2a6c JB |
11016 | |
11017 | /* Reset pinned so btrfs_put_block_group doesn't complain */ | |
c30666d4 ZL |
11018 | spin_lock(&space_info->lock); |
11019 | spin_lock(&block_group->lock); | |
11020 | ||
e2907c1a | 11021 | update_bytes_pinned(space_info, -block_group->pinned); |
c30666d4 | 11022 | space_info->bytes_readonly += block_group->pinned; |
dec59fa3 EL |
11023 | percpu_counter_add_batch(&space_info->total_bytes_pinned, |
11024 | -block_group->pinned, | |
11025 | BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
47ab2a6c JB |
11026 | block_group->pinned = 0; |
11027 | ||
c30666d4 ZL |
11028 | spin_unlock(&block_group->lock); |
11029 | spin_unlock(&space_info->lock); | |
11030 | ||
e33e17ee | 11031 | /* DISCARD can flip during remount */ |
0b246afa | 11032 | trimming = btrfs_test_opt(fs_info, DISCARD); |
e33e17ee JM |
11033 | |
11034 | /* Implicit trim during transaction commit. */ | |
11035 | if (trimming) | |
11036 | btrfs_get_block_group_trimming(block_group); | |
11037 | ||
47ab2a6c JB |
11038 | /* |
11039 | * Btrfs_remove_chunk will abort the transaction if things go | |
11040 | * horribly wrong. | |
11041 | */ | |
97aff912 | 11042 | ret = btrfs_remove_chunk(trans, block_group->key.objectid); |
e33e17ee JM |
11043 | |
11044 | if (ret) { | |
11045 | if (trimming) | |
11046 | btrfs_put_block_group_trimming(block_group); | |
11047 | goto end_trans; | |
11048 | } | |
11049 | ||
11050 | /* | |
11051 | * If we're not mounted with -odiscard, we can just forget | |
11052 | * about this block group. Otherwise we'll need to wait | |
11053 | * until transaction commit to do the actual discard. | |
11054 | */ | |
11055 | if (trimming) { | |
348a0013 FM |
11056 | spin_lock(&fs_info->unused_bgs_lock); |
11057 | /* | |
11058 | * A concurrent scrub might have added us to the list | |
11059 | * fs_info->unused_bgs, so use a list_move operation | |
11060 | * to add the block group to the deleted_bgs list. | |
11061 | */ | |
e33e17ee JM |
11062 | list_move(&block_group->bg_list, |
11063 | &trans->transaction->deleted_bgs); | |
348a0013 | 11064 | spin_unlock(&fs_info->unused_bgs_lock); |
e33e17ee JM |
11065 | btrfs_get_block_group(block_group); |
11066 | } | |
758eb51e | 11067 | end_trans: |
3a45bb20 | 11068 | btrfs_end_transaction(trans); |
47ab2a6c | 11069 | next: |
d5f2e33b | 11070 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
47ab2a6c JB |
11071 | btrfs_put_block_group(block_group); |
11072 | spin_lock(&fs_info->unused_bgs_lock); | |
11073 | } | |
11074 | spin_unlock(&fs_info->unused_bgs_lock); | |
11075 | } | |
11076 | ||
c59021f8 | 11077 | int btrfs_init_space_info(struct btrfs_fs_info *fs_info) |
11078 | { | |
1aba86d6 | 11079 | struct btrfs_super_block *disk_super; |
11080 | u64 features; | |
11081 | u64 flags; | |
11082 | int mixed = 0; | |
c59021f8 | 11083 | int ret; |
11084 | ||
6c41761f | 11085 | disk_super = fs_info->super_copy; |
1aba86d6 | 11086 | if (!btrfs_super_root(disk_super)) |
0dc924c5 | 11087 | return -EINVAL; |
c59021f8 | 11088 | |
1aba86d6 | 11089 | features = btrfs_super_incompat_flags(disk_super); |
11090 | if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
11091 | mixed = 1; | |
c59021f8 | 11092 | |
1aba86d6 | 11093 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
4ca61683 | 11094 | ret = create_space_info(fs_info, flags); |
c59021f8 | 11095 | if (ret) |
1aba86d6 | 11096 | goto out; |
c59021f8 | 11097 | |
1aba86d6 | 11098 | if (mixed) { |
11099 | flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA; | |
4ca61683 | 11100 | ret = create_space_info(fs_info, flags); |
1aba86d6 | 11101 | } else { |
11102 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
4ca61683 | 11103 | ret = create_space_info(fs_info, flags); |
1aba86d6 | 11104 | if (ret) |
11105 | goto out; | |
11106 | ||
11107 | flags = BTRFS_BLOCK_GROUP_DATA; | |
4ca61683 | 11108 | ret = create_space_info(fs_info, flags); |
1aba86d6 | 11109 | } |
11110 | out: | |
c59021f8 | 11111 | return ret; |
11112 | } | |
11113 | ||
2ff7e61e JM |
11114 | int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info, |
11115 | u64 start, u64 end) | |
acce952b | 11116 | { |
2ff7e61e | 11117 | return unpin_extent_range(fs_info, start, end, false); |
acce952b | 11118 | } |
11119 | ||
499f377f JM |
11120 | /* |
11121 | * It used to be that old block groups would be left around forever. | |
11122 | * Iterating over them would be enough to trim unused space. Since we | |
11123 | * now automatically remove them, we also need to iterate over unallocated | |
11124 | * space. | |
11125 | * | |
11126 | * We don't want a transaction for this since the discard may take a | |
11127 | * substantial amount of time. We don't require that a transaction be | |
11128 | * running, but we do need to take a running transaction into account | |
fee7acc3 JM |
11129 | * to ensure that we're not discarding chunks that were released or |
11130 | * allocated in the current transaction. | |
499f377f JM |
11131 | * |
11132 | * Holding the chunks lock will prevent other threads from allocating | |
11133 | * or releasing chunks, but it won't prevent a running transaction | |
11134 | * from committing and releasing the memory that the pending chunks | |
11135 | * list head uses. For that, we need to take a reference to the | |
fee7acc3 JM |
11136 | * transaction and hold the commit root sem. We only need to hold |
11137 | * it while performing the free space search since we have already | |
11138 | * held back allocations. | |
499f377f | 11139 | */ |
8103d10b | 11140 | static int btrfs_trim_free_extents(struct btrfs_device *device, u64 *trimmed) |
499f377f | 11141 | { |
8103d10b | 11142 | u64 start = SZ_1M, len = 0, end = 0; |
499f377f JM |
11143 | int ret; |
11144 | ||
11145 | *trimmed = 0; | |
11146 | ||
0be88e36 JM |
11147 | /* Discard not supported = nothing to do. */ |
11148 | if (!blk_queue_discard(bdev_get_queue(device->bdev))) | |
11149 | return 0; | |
11150 | ||
52042d8e | 11151 | /* Not writable = nothing to do. */ |
ebbede42 | 11152 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
499f377f JM |
11153 | return 0; |
11154 | ||
11155 | /* No free space = nothing to do. */ | |
11156 | if (device->total_bytes <= device->bytes_used) | |
11157 | return 0; | |
11158 | ||
11159 | ret = 0; | |
11160 | ||
11161 | while (1) { | |
fb456252 | 11162 | struct btrfs_fs_info *fs_info = device->fs_info; |
499f377f JM |
11163 | u64 bytes; |
11164 | ||
11165 | ret = mutex_lock_interruptible(&fs_info->chunk_mutex); | |
11166 | if (ret) | |
fee7acc3 | 11167 | break; |
499f377f | 11168 | |
929be17a NB |
11169 | find_first_clear_extent_bit(&device->alloc_state, start, |
11170 | &start, &end, | |
11171 | CHUNK_TRIMMED | CHUNK_ALLOCATED); | |
11172 | /* | |
11173 | * If find_first_clear_extent_bit find a range that spans the | |
11174 | * end of the device it will set end to -1, in this case it's up | |
11175 | * to the caller to trim the value to the size of the device. | |
11176 | */ | |
11177 | end = min(end, device->total_bytes - 1); | |
11178 | len = end - start + 1; | |
499f377f | 11179 | |
929be17a NB |
11180 | /* We didn't find any extents */ |
11181 | if (!len) { | |
499f377f | 11182 | mutex_unlock(&fs_info->chunk_mutex); |
929be17a | 11183 | ret = 0; |
499f377f JM |
11184 | break; |
11185 | } | |
11186 | ||
929be17a NB |
11187 | ret = btrfs_issue_discard(device->bdev, start, len, |
11188 | &bytes); | |
11189 | if (!ret) | |
11190 | set_extent_bits(&device->alloc_state, start, | |
11191 | start + bytes - 1, | |
11192 | CHUNK_TRIMMED); | |
499f377f JM |
11193 | mutex_unlock(&fs_info->chunk_mutex); |
11194 | ||
11195 | if (ret) | |
11196 | break; | |
11197 | ||
11198 | start += len; | |
11199 | *trimmed += bytes; | |
11200 | ||
11201 | if (fatal_signal_pending(current)) { | |
11202 | ret = -ERESTARTSYS; | |
11203 | break; | |
11204 | } | |
11205 | ||
11206 | cond_resched(); | |
11207 | } | |
11208 | ||
11209 | return ret; | |
11210 | } | |
11211 | ||
93bba24d QW |
11212 | /* |
11213 | * Trim the whole filesystem by: | |
11214 | * 1) trimming the free space in each block group | |
11215 | * 2) trimming the unallocated space on each device | |
11216 | * | |
11217 | * This will also continue trimming even if a block group or device encounters | |
11218 | * an error. The return value will be the last error, or 0 if nothing bad | |
11219 | * happens. | |
11220 | */ | |
2ff7e61e | 11221 | int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range) |
f7039b1d | 11222 | { |
f7039b1d | 11223 | struct btrfs_block_group_cache *cache = NULL; |
499f377f JM |
11224 | struct btrfs_device *device; |
11225 | struct list_head *devices; | |
f7039b1d LD |
11226 | u64 group_trimmed; |
11227 | u64 start; | |
11228 | u64 end; | |
11229 | u64 trimmed = 0; | |
93bba24d QW |
11230 | u64 bg_failed = 0; |
11231 | u64 dev_failed = 0; | |
11232 | int bg_ret = 0; | |
11233 | int dev_ret = 0; | |
f7039b1d LD |
11234 | int ret = 0; |
11235 | ||
6ba9fc8e | 11236 | cache = btrfs_lookup_first_block_group(fs_info, range->start); |
f87b7eb8 | 11237 | for (; cache; cache = next_block_group(cache)) { |
f7039b1d LD |
11238 | if (cache->key.objectid >= (range->start + range->len)) { |
11239 | btrfs_put_block_group(cache); | |
11240 | break; | |
11241 | } | |
11242 | ||
11243 | start = max(range->start, cache->key.objectid); | |
11244 | end = min(range->start + range->len, | |
11245 | cache->key.objectid + cache->key.offset); | |
11246 | ||
11247 | if (end - start >= range->minlen) { | |
11248 | if (!block_group_cache_done(cache)) { | |
f6373bf3 | 11249 | ret = cache_block_group(cache, 0); |
1be41b78 | 11250 | if (ret) { |
93bba24d QW |
11251 | bg_failed++; |
11252 | bg_ret = ret; | |
11253 | continue; | |
1be41b78 JB |
11254 | } |
11255 | ret = wait_block_group_cache_done(cache); | |
11256 | if (ret) { | |
93bba24d QW |
11257 | bg_failed++; |
11258 | bg_ret = ret; | |
11259 | continue; | |
1be41b78 | 11260 | } |
f7039b1d LD |
11261 | } |
11262 | ret = btrfs_trim_block_group(cache, | |
11263 | &group_trimmed, | |
11264 | start, | |
11265 | end, | |
11266 | range->minlen); | |
11267 | ||
11268 | trimmed += group_trimmed; | |
11269 | if (ret) { | |
93bba24d QW |
11270 | bg_failed++; |
11271 | bg_ret = ret; | |
11272 | continue; | |
f7039b1d LD |
11273 | } |
11274 | } | |
f7039b1d LD |
11275 | } |
11276 | ||
93bba24d QW |
11277 | if (bg_failed) |
11278 | btrfs_warn(fs_info, | |
11279 | "failed to trim %llu block group(s), last error %d", | |
11280 | bg_failed, bg_ret); | |
0b246afa | 11281 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
d4e329de JM |
11282 | devices = &fs_info->fs_devices->devices; |
11283 | list_for_each_entry(device, devices, dev_list) { | |
8103d10b | 11284 | ret = btrfs_trim_free_extents(device, &group_trimmed); |
93bba24d QW |
11285 | if (ret) { |
11286 | dev_failed++; | |
11287 | dev_ret = ret; | |
499f377f | 11288 | break; |
93bba24d | 11289 | } |
499f377f JM |
11290 | |
11291 | trimmed += group_trimmed; | |
11292 | } | |
0b246afa | 11293 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
499f377f | 11294 | |
93bba24d QW |
11295 | if (dev_failed) |
11296 | btrfs_warn(fs_info, | |
11297 | "failed to trim %llu device(s), last error %d", | |
11298 | dev_failed, dev_ret); | |
f7039b1d | 11299 | range->len = trimmed; |
93bba24d QW |
11300 | if (bg_ret) |
11301 | return bg_ret; | |
11302 | return dev_ret; | |
f7039b1d | 11303 | } |
8257b2dc MX |
11304 | |
11305 | /* | |
ea14b57f | 11306 | * btrfs_{start,end}_write_no_snapshotting() are similar to |
9ea24bbe FM |
11307 | * mnt_{want,drop}_write(), they are used to prevent some tasks from writing |
11308 | * data into the page cache through nocow before the subvolume is snapshoted, | |
11309 | * but flush the data into disk after the snapshot creation, or to prevent | |
ea14b57f | 11310 | * operations while snapshotting is ongoing and that cause the snapshot to be |
9ea24bbe | 11311 | * inconsistent (writes followed by expanding truncates for example). |
8257b2dc | 11312 | */ |
ea14b57f | 11313 | void btrfs_end_write_no_snapshotting(struct btrfs_root *root) |
8257b2dc MX |
11314 | { |
11315 | percpu_counter_dec(&root->subv_writers->counter); | |
093258e6 | 11316 | cond_wake_up(&root->subv_writers->wait); |
8257b2dc MX |
11317 | } |
11318 | ||
ea14b57f | 11319 | int btrfs_start_write_no_snapshotting(struct btrfs_root *root) |
8257b2dc | 11320 | { |
ea14b57f | 11321 | if (atomic_read(&root->will_be_snapshotted)) |
8257b2dc MX |
11322 | return 0; |
11323 | ||
11324 | percpu_counter_inc(&root->subv_writers->counter); | |
11325 | /* | |
11326 | * Make sure counter is updated before we check for snapshot creation. | |
11327 | */ | |
11328 | smp_mb(); | |
ea14b57f DS |
11329 | if (atomic_read(&root->will_be_snapshotted)) { |
11330 | btrfs_end_write_no_snapshotting(root); | |
8257b2dc MX |
11331 | return 0; |
11332 | } | |
11333 | return 1; | |
11334 | } | |
0bc19f90 | 11335 | |
0bc19f90 ZL |
11336 | void btrfs_wait_for_snapshot_creation(struct btrfs_root *root) |
11337 | { | |
11338 | while (true) { | |
11339 | int ret; | |
11340 | ||
ea14b57f | 11341 | ret = btrfs_start_write_no_snapshotting(root); |
0bc19f90 ZL |
11342 | if (ret) |
11343 | break; | |
4625956a PZ |
11344 | wait_var_event(&root->will_be_snapshotted, |
11345 | !atomic_read(&root->will_be_snapshotted)); | |
0bc19f90 ZL |
11346 | } |
11347 | } | |
031f24da QW |
11348 | |
11349 | void btrfs_mark_bg_unused(struct btrfs_block_group_cache *bg) | |
11350 | { | |
11351 | struct btrfs_fs_info *fs_info = bg->fs_info; | |
11352 | ||
11353 | spin_lock(&fs_info->unused_bgs_lock); | |
11354 | if (list_empty(&bg->bg_list)) { | |
11355 | btrfs_get_block_group(bg); | |
11356 | trace_btrfs_add_unused_block_group(bg); | |
11357 | list_add_tail(&bg->bg_list, &fs_info->unused_bgs); | |
11358 | } | |
11359 | spin_unlock(&fs_info->unused_bgs_lock); | |
11360 | } |