Commit | Line | Data |
---|---|---|
6cbd5570 CM |
1 | /* |
2 | * Copyright (C) 2007 Oracle. All rights reserved. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public | |
6 | * License v2 as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, | |
9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public | |
14 | * License along with this program; if not, write to the | |
15 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
16 | * Boston, MA 021110-1307, USA. | |
17 | */ | |
ec6b910f | 18 | #include <linux/sched.h> |
edbd8d4e | 19 | #include <linux/pagemap.h> |
ec44a35c | 20 | #include <linux/writeback.h> |
21af804c | 21 | #include <linux/blkdev.h> |
b7a9f29f | 22 | #include <linux/sort.h> |
4184ea7f | 23 | #include <linux/rcupdate.h> |
817d52f8 | 24 | #include <linux/kthread.h> |
5a0e3ad6 | 25 | #include <linux/slab.h> |
dff51cd1 | 26 | #include <linux/ratelimit.h> |
b150a4f1 | 27 | #include <linux/percpu_counter.h> |
74493f7a | 28 | #include "hash.h" |
995946dd | 29 | #include "tree-log.h" |
fec577fb CM |
30 | #include "disk-io.h" |
31 | #include "print-tree.h" | |
0b86a832 | 32 | #include "volumes.h" |
53b381b3 | 33 | #include "raid56.h" |
925baedd | 34 | #include "locking.h" |
fa9c0d79 | 35 | #include "free-space-cache.h" |
1e144fb8 | 36 | #include "free-space-tree.h" |
3fed40cc | 37 | #include "math.h" |
6ab0a202 | 38 | #include "sysfs.h" |
fcebe456 | 39 | #include "qgroup.h" |
fec577fb | 40 | |
709c0486 AJ |
41 | #undef SCRAMBLE_DELAYED_REFS |
42 | ||
9e622d6b MX |
43 | /* |
44 | * control flags for do_chunk_alloc's force field | |
0e4f8f88 CM |
45 | * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk |
46 | * if we really need one. | |
47 | * | |
0e4f8f88 CM |
48 | * CHUNK_ALLOC_LIMITED means to only try and allocate one |
49 | * if we have very few chunks already allocated. This is | |
50 | * used as part of the clustering code to help make sure | |
51 | * we have a good pool of storage to cluster in, without | |
52 | * filling the FS with empty chunks | |
53 | * | |
9e622d6b MX |
54 | * CHUNK_ALLOC_FORCE means it must try to allocate one |
55 | * | |
0e4f8f88 CM |
56 | */ |
57 | enum { | |
58 | CHUNK_ALLOC_NO_FORCE = 0, | |
9e622d6b MX |
59 | CHUNK_ALLOC_LIMITED = 1, |
60 | CHUNK_ALLOC_FORCE = 2, | |
0e4f8f88 CM |
61 | }; |
62 | ||
fb25e914 JB |
63 | /* |
64 | * Control how reservations are dealt with. | |
65 | * | |
66 | * RESERVE_FREE - freeing a reservation. | |
67 | * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for | |
68 | * ENOSPC accounting | |
69 | * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update | |
70 | * bytes_may_use as the ENOSPC accounting is done elsewhere | |
71 | */ | |
72 | enum { | |
73 | RESERVE_FREE = 0, | |
74 | RESERVE_ALLOC = 1, | |
75 | RESERVE_ALLOC_NO_ACCOUNT = 2, | |
76 | }; | |
77 | ||
ce93ec54 JB |
78 | static int update_block_group(struct btrfs_trans_handle *trans, |
79 | struct btrfs_root *root, u64 bytenr, | |
80 | u64 num_bytes, int alloc); | |
5d4f98a2 YZ |
81 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
82 | struct btrfs_root *root, | |
c682f9b3 | 83 | struct btrfs_delayed_ref_node *node, u64 parent, |
5d4f98a2 YZ |
84 | u64 root_objectid, u64 owner_objectid, |
85 | u64 owner_offset, int refs_to_drop, | |
c682f9b3 | 86 | struct btrfs_delayed_extent_op *extra_op); |
5d4f98a2 YZ |
87 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, |
88 | struct extent_buffer *leaf, | |
89 | struct btrfs_extent_item *ei); | |
90 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
91 | struct btrfs_root *root, | |
92 | u64 parent, u64 root_objectid, | |
93 | u64 flags, u64 owner, u64 offset, | |
94 | struct btrfs_key *ins, int ref_mod); | |
95 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, | |
96 | struct btrfs_root *root, | |
97 | u64 parent, u64 root_objectid, | |
98 | u64 flags, struct btrfs_disk_key *key, | |
b06c4bf5 | 99 | int level, struct btrfs_key *ins); |
6a63209f | 100 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
698d0082 JB |
101 | struct btrfs_root *extent_root, u64 flags, |
102 | int force); | |
11833d66 YZ |
103 | static int find_next_key(struct btrfs_path *path, int level, |
104 | struct btrfs_key *key); | |
9ed74f2d JB |
105 | static void dump_space_info(struct btrfs_space_info *info, u64 bytes, |
106 | int dump_block_groups); | |
fb25e914 | 107 | static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache, |
e570fd27 MX |
108 | u64 num_bytes, int reserve, |
109 | int delalloc); | |
5d80366e JB |
110 | static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, |
111 | u64 num_bytes); | |
48a3b636 ES |
112 | int btrfs_pin_extent(struct btrfs_root *root, |
113 | u64 bytenr, u64 num_bytes, int reserved); | |
6a63209f | 114 | |
817d52f8 JB |
115 | static noinline int |
116 | block_group_cache_done(struct btrfs_block_group_cache *cache) | |
117 | { | |
118 | smp_mb(); | |
36cce922 JB |
119 | return cache->cached == BTRFS_CACHE_FINISHED || |
120 | cache->cached == BTRFS_CACHE_ERROR; | |
817d52f8 JB |
121 | } |
122 | ||
0f9dd46c JB |
123 | static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits) |
124 | { | |
125 | return (cache->flags & bits) == bits; | |
126 | } | |
127 | ||
758f2dfc | 128 | void btrfs_get_block_group(struct btrfs_block_group_cache *cache) |
11dfe35a JB |
129 | { |
130 | atomic_inc(&cache->count); | |
131 | } | |
132 | ||
133 | void btrfs_put_block_group(struct btrfs_block_group_cache *cache) | |
134 | { | |
f0486c68 YZ |
135 | if (atomic_dec_and_test(&cache->count)) { |
136 | WARN_ON(cache->pinned > 0); | |
137 | WARN_ON(cache->reserved > 0); | |
34d52cb6 | 138 | kfree(cache->free_space_ctl); |
11dfe35a | 139 | kfree(cache); |
f0486c68 | 140 | } |
11dfe35a JB |
141 | } |
142 | ||
0f9dd46c JB |
143 | /* |
144 | * this adds the block group to the fs_info rb tree for the block group | |
145 | * cache | |
146 | */ | |
b2950863 | 147 | static int btrfs_add_block_group_cache(struct btrfs_fs_info *info, |
0f9dd46c JB |
148 | struct btrfs_block_group_cache *block_group) |
149 | { | |
150 | struct rb_node **p; | |
151 | struct rb_node *parent = NULL; | |
152 | struct btrfs_block_group_cache *cache; | |
153 | ||
154 | spin_lock(&info->block_group_cache_lock); | |
155 | p = &info->block_group_cache_tree.rb_node; | |
156 | ||
157 | while (*p) { | |
158 | parent = *p; | |
159 | cache = rb_entry(parent, struct btrfs_block_group_cache, | |
160 | cache_node); | |
161 | if (block_group->key.objectid < cache->key.objectid) { | |
162 | p = &(*p)->rb_left; | |
163 | } else if (block_group->key.objectid > cache->key.objectid) { | |
164 | p = &(*p)->rb_right; | |
165 | } else { | |
166 | spin_unlock(&info->block_group_cache_lock); | |
167 | return -EEXIST; | |
168 | } | |
169 | } | |
170 | ||
171 | rb_link_node(&block_group->cache_node, parent, p); | |
172 | rb_insert_color(&block_group->cache_node, | |
173 | &info->block_group_cache_tree); | |
a1897fdd LB |
174 | |
175 | if (info->first_logical_byte > block_group->key.objectid) | |
176 | info->first_logical_byte = block_group->key.objectid; | |
177 | ||
0f9dd46c JB |
178 | spin_unlock(&info->block_group_cache_lock); |
179 | ||
180 | return 0; | |
181 | } | |
182 | ||
183 | /* | |
184 | * This will return the block group at or after bytenr if contains is 0, else | |
185 | * it will return the block group that contains the bytenr | |
186 | */ | |
187 | static struct btrfs_block_group_cache * | |
188 | block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr, | |
189 | int contains) | |
190 | { | |
191 | struct btrfs_block_group_cache *cache, *ret = NULL; | |
192 | struct rb_node *n; | |
193 | u64 end, start; | |
194 | ||
195 | spin_lock(&info->block_group_cache_lock); | |
196 | n = info->block_group_cache_tree.rb_node; | |
197 | ||
198 | while (n) { | |
199 | cache = rb_entry(n, struct btrfs_block_group_cache, | |
200 | cache_node); | |
201 | end = cache->key.objectid + cache->key.offset - 1; | |
202 | start = cache->key.objectid; | |
203 | ||
204 | if (bytenr < start) { | |
205 | if (!contains && (!ret || start < ret->key.objectid)) | |
206 | ret = cache; | |
207 | n = n->rb_left; | |
208 | } else if (bytenr > start) { | |
209 | if (contains && bytenr <= end) { | |
210 | ret = cache; | |
211 | break; | |
212 | } | |
213 | n = n->rb_right; | |
214 | } else { | |
215 | ret = cache; | |
216 | break; | |
217 | } | |
218 | } | |
a1897fdd | 219 | if (ret) { |
11dfe35a | 220 | btrfs_get_block_group(ret); |
a1897fdd LB |
221 | if (bytenr == 0 && info->first_logical_byte > ret->key.objectid) |
222 | info->first_logical_byte = ret->key.objectid; | |
223 | } | |
0f9dd46c JB |
224 | spin_unlock(&info->block_group_cache_lock); |
225 | ||
226 | return ret; | |
227 | } | |
228 | ||
11833d66 YZ |
229 | static int add_excluded_extent(struct btrfs_root *root, |
230 | u64 start, u64 num_bytes) | |
817d52f8 | 231 | { |
11833d66 YZ |
232 | u64 end = start + num_bytes - 1; |
233 | set_extent_bits(&root->fs_info->freed_extents[0], | |
234 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
235 | set_extent_bits(&root->fs_info->freed_extents[1], | |
236 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
237 | return 0; | |
238 | } | |
817d52f8 | 239 | |
11833d66 YZ |
240 | static void free_excluded_extents(struct btrfs_root *root, |
241 | struct btrfs_block_group_cache *cache) | |
242 | { | |
243 | u64 start, end; | |
817d52f8 | 244 | |
11833d66 YZ |
245 | start = cache->key.objectid; |
246 | end = start + cache->key.offset - 1; | |
247 | ||
248 | clear_extent_bits(&root->fs_info->freed_extents[0], | |
249 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
250 | clear_extent_bits(&root->fs_info->freed_extents[1], | |
251 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
817d52f8 JB |
252 | } |
253 | ||
11833d66 YZ |
254 | static int exclude_super_stripes(struct btrfs_root *root, |
255 | struct btrfs_block_group_cache *cache) | |
817d52f8 | 256 | { |
817d52f8 JB |
257 | u64 bytenr; |
258 | u64 *logical; | |
259 | int stripe_len; | |
260 | int i, nr, ret; | |
261 | ||
06b2331f YZ |
262 | if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) { |
263 | stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid; | |
264 | cache->bytes_super += stripe_len; | |
265 | ret = add_excluded_extent(root, cache->key.objectid, | |
266 | stripe_len); | |
835d974f JB |
267 | if (ret) |
268 | return ret; | |
06b2331f YZ |
269 | } |
270 | ||
817d52f8 JB |
271 | for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { |
272 | bytenr = btrfs_sb_offset(i); | |
273 | ret = btrfs_rmap_block(&root->fs_info->mapping_tree, | |
274 | cache->key.objectid, bytenr, | |
275 | 0, &logical, &nr, &stripe_len); | |
835d974f JB |
276 | if (ret) |
277 | return ret; | |
11833d66 | 278 | |
817d52f8 | 279 | while (nr--) { |
51bf5f0b JB |
280 | u64 start, len; |
281 | ||
282 | if (logical[nr] > cache->key.objectid + | |
283 | cache->key.offset) | |
284 | continue; | |
285 | ||
286 | if (logical[nr] + stripe_len <= cache->key.objectid) | |
287 | continue; | |
288 | ||
289 | start = logical[nr]; | |
290 | if (start < cache->key.objectid) { | |
291 | start = cache->key.objectid; | |
292 | len = (logical[nr] + stripe_len) - start; | |
293 | } else { | |
294 | len = min_t(u64, stripe_len, | |
295 | cache->key.objectid + | |
296 | cache->key.offset - start); | |
297 | } | |
298 | ||
299 | cache->bytes_super += len; | |
300 | ret = add_excluded_extent(root, start, len); | |
835d974f JB |
301 | if (ret) { |
302 | kfree(logical); | |
303 | return ret; | |
304 | } | |
817d52f8 | 305 | } |
11833d66 | 306 | |
817d52f8 JB |
307 | kfree(logical); |
308 | } | |
817d52f8 JB |
309 | return 0; |
310 | } | |
311 | ||
11833d66 YZ |
312 | static struct btrfs_caching_control * |
313 | get_caching_control(struct btrfs_block_group_cache *cache) | |
314 | { | |
315 | struct btrfs_caching_control *ctl; | |
316 | ||
317 | spin_lock(&cache->lock); | |
dde5abee JB |
318 | if (!cache->caching_ctl) { |
319 | spin_unlock(&cache->lock); | |
11833d66 YZ |
320 | return NULL; |
321 | } | |
322 | ||
323 | ctl = cache->caching_ctl; | |
324 | atomic_inc(&ctl->count); | |
325 | spin_unlock(&cache->lock); | |
326 | return ctl; | |
327 | } | |
328 | ||
329 | static void put_caching_control(struct btrfs_caching_control *ctl) | |
330 | { | |
331 | if (atomic_dec_and_test(&ctl->count)) | |
332 | kfree(ctl); | |
333 | } | |
334 | ||
d0bd4560 JB |
335 | #ifdef CONFIG_BTRFS_DEBUG |
336 | static void fragment_free_space(struct btrfs_root *root, | |
337 | struct btrfs_block_group_cache *block_group) | |
338 | { | |
339 | u64 start = block_group->key.objectid; | |
340 | u64 len = block_group->key.offset; | |
341 | u64 chunk = block_group->flags & BTRFS_BLOCK_GROUP_METADATA ? | |
342 | root->nodesize : root->sectorsize; | |
343 | u64 step = chunk << 1; | |
344 | ||
345 | while (len > chunk) { | |
346 | btrfs_remove_free_space(block_group, start, chunk); | |
347 | start += step; | |
348 | if (len < step) | |
349 | len = 0; | |
350 | else | |
351 | len -= step; | |
352 | } | |
353 | } | |
354 | #endif | |
355 | ||
0f9dd46c JB |
356 | /* |
357 | * this is only called by cache_block_group, since we could have freed extents | |
358 | * we need to check the pinned_extents for any extents that can't be used yet | |
359 | * since their free space will be released as soon as the transaction commits. | |
360 | */ | |
a5ed9182 OS |
361 | u64 add_new_free_space(struct btrfs_block_group_cache *block_group, |
362 | struct btrfs_fs_info *info, u64 start, u64 end) | |
0f9dd46c | 363 | { |
817d52f8 | 364 | u64 extent_start, extent_end, size, total_added = 0; |
0f9dd46c JB |
365 | int ret; |
366 | ||
367 | while (start < end) { | |
11833d66 | 368 | ret = find_first_extent_bit(info->pinned_extents, start, |
0f9dd46c | 369 | &extent_start, &extent_end, |
e6138876 JB |
370 | EXTENT_DIRTY | EXTENT_UPTODATE, |
371 | NULL); | |
0f9dd46c JB |
372 | if (ret) |
373 | break; | |
374 | ||
06b2331f | 375 | if (extent_start <= start) { |
0f9dd46c JB |
376 | start = extent_end + 1; |
377 | } else if (extent_start > start && extent_start < end) { | |
378 | size = extent_start - start; | |
817d52f8 | 379 | total_added += size; |
ea6a478e JB |
380 | ret = btrfs_add_free_space(block_group, start, |
381 | size); | |
79787eaa | 382 | BUG_ON(ret); /* -ENOMEM or logic error */ |
0f9dd46c JB |
383 | start = extent_end + 1; |
384 | } else { | |
385 | break; | |
386 | } | |
387 | } | |
388 | ||
389 | if (start < end) { | |
390 | size = end - start; | |
817d52f8 | 391 | total_added += size; |
ea6a478e | 392 | ret = btrfs_add_free_space(block_group, start, size); |
79787eaa | 393 | BUG_ON(ret); /* -ENOMEM or logic error */ |
0f9dd46c JB |
394 | } |
395 | ||
817d52f8 | 396 | return total_added; |
0f9dd46c JB |
397 | } |
398 | ||
73fa48b6 | 399 | static int load_extent_tree_free(struct btrfs_caching_control *caching_ctl) |
e37c9e69 | 400 | { |
bab39bf9 JB |
401 | struct btrfs_block_group_cache *block_group; |
402 | struct btrfs_fs_info *fs_info; | |
bab39bf9 | 403 | struct btrfs_root *extent_root; |
e37c9e69 | 404 | struct btrfs_path *path; |
5f39d397 | 405 | struct extent_buffer *leaf; |
11833d66 | 406 | struct btrfs_key key; |
817d52f8 | 407 | u64 total_found = 0; |
11833d66 YZ |
408 | u64 last = 0; |
409 | u32 nritems; | |
73fa48b6 | 410 | int ret; |
d0bd4560 | 411 | bool wakeup = true; |
f510cfec | 412 | |
bab39bf9 JB |
413 | block_group = caching_ctl->block_group; |
414 | fs_info = block_group->fs_info; | |
415 | extent_root = fs_info->extent_root; | |
416 | ||
e37c9e69 CM |
417 | path = btrfs_alloc_path(); |
418 | if (!path) | |
73fa48b6 | 419 | return -ENOMEM; |
7d7d6068 | 420 | |
817d52f8 | 421 | last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET); |
11833d66 | 422 | |
d0bd4560 JB |
423 | #ifdef CONFIG_BTRFS_DEBUG |
424 | /* | |
425 | * If we're fragmenting we don't want to make anybody think we can | |
426 | * allocate from this block group until we've had a chance to fragment | |
427 | * the free space. | |
428 | */ | |
429 | if (btrfs_should_fragment_free_space(extent_root, block_group)) | |
430 | wakeup = false; | |
431 | #endif | |
5cd57b2c | 432 | /* |
817d52f8 JB |
433 | * We don't want to deadlock with somebody trying to allocate a new |
434 | * extent for the extent root while also trying to search the extent | |
435 | * root to add free space. So we skip locking and search the commit | |
436 | * root, since its read-only | |
5cd57b2c CM |
437 | */ |
438 | path->skip_locking = 1; | |
817d52f8 | 439 | path->search_commit_root = 1; |
026fd317 | 440 | path->reada = 1; |
817d52f8 | 441 | |
e4404d6e | 442 | key.objectid = last; |
e37c9e69 | 443 | key.offset = 0; |
11833d66 | 444 | key.type = BTRFS_EXTENT_ITEM_KEY; |
013f1b12 | 445 | |
52ee28d2 | 446 | next: |
11833d66 | 447 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
e37c9e69 | 448 | if (ret < 0) |
73fa48b6 | 449 | goto out; |
a512bbf8 | 450 | |
11833d66 YZ |
451 | leaf = path->nodes[0]; |
452 | nritems = btrfs_header_nritems(leaf); | |
453 | ||
d397712b | 454 | while (1) { |
7841cb28 | 455 | if (btrfs_fs_closing(fs_info) > 1) { |
f25784b3 | 456 | last = (u64)-1; |
817d52f8 | 457 | break; |
f25784b3 | 458 | } |
817d52f8 | 459 | |
11833d66 YZ |
460 | if (path->slots[0] < nritems) { |
461 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
462 | } else { | |
463 | ret = find_next_key(path, 0, &key); | |
464 | if (ret) | |
e37c9e69 | 465 | break; |
817d52f8 | 466 | |
c9ea7b24 | 467 | if (need_resched() || |
9e351cc8 | 468 | rwsem_is_contended(&fs_info->commit_root_sem)) { |
d0bd4560 JB |
469 | if (wakeup) |
470 | caching_ctl->progress = last; | |
ff5714cc | 471 | btrfs_release_path(path); |
9e351cc8 | 472 | up_read(&fs_info->commit_root_sem); |
589d8ade | 473 | mutex_unlock(&caching_ctl->mutex); |
11833d66 | 474 | cond_resched(); |
73fa48b6 OS |
475 | mutex_lock(&caching_ctl->mutex); |
476 | down_read(&fs_info->commit_root_sem); | |
477 | goto next; | |
589d8ade | 478 | } |
0a3896d0 JB |
479 | |
480 | ret = btrfs_next_leaf(extent_root, path); | |
481 | if (ret < 0) | |
73fa48b6 | 482 | goto out; |
0a3896d0 JB |
483 | if (ret) |
484 | break; | |
589d8ade JB |
485 | leaf = path->nodes[0]; |
486 | nritems = btrfs_header_nritems(leaf); | |
487 | continue; | |
11833d66 | 488 | } |
817d52f8 | 489 | |
52ee28d2 LB |
490 | if (key.objectid < last) { |
491 | key.objectid = last; | |
492 | key.offset = 0; | |
493 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
494 | ||
d0bd4560 JB |
495 | if (wakeup) |
496 | caching_ctl->progress = last; | |
52ee28d2 LB |
497 | btrfs_release_path(path); |
498 | goto next; | |
499 | } | |
500 | ||
11833d66 YZ |
501 | if (key.objectid < block_group->key.objectid) { |
502 | path->slots[0]++; | |
817d52f8 | 503 | continue; |
e37c9e69 | 504 | } |
0f9dd46c | 505 | |
e37c9e69 | 506 | if (key.objectid >= block_group->key.objectid + |
0f9dd46c | 507 | block_group->key.offset) |
e37c9e69 | 508 | break; |
7d7d6068 | 509 | |
3173a18f JB |
510 | if (key.type == BTRFS_EXTENT_ITEM_KEY || |
511 | key.type == BTRFS_METADATA_ITEM_KEY) { | |
817d52f8 JB |
512 | total_found += add_new_free_space(block_group, |
513 | fs_info, last, | |
514 | key.objectid); | |
3173a18f JB |
515 | if (key.type == BTRFS_METADATA_ITEM_KEY) |
516 | last = key.objectid + | |
707e8a07 | 517 | fs_info->tree_root->nodesize; |
3173a18f JB |
518 | else |
519 | last = key.objectid + key.offset; | |
817d52f8 | 520 | |
73fa48b6 | 521 | if (total_found > CACHING_CTL_WAKE_UP) { |
11833d66 | 522 | total_found = 0; |
d0bd4560 JB |
523 | if (wakeup) |
524 | wake_up(&caching_ctl->wait); | |
11833d66 | 525 | } |
817d52f8 | 526 | } |
e37c9e69 CM |
527 | path->slots[0]++; |
528 | } | |
817d52f8 | 529 | ret = 0; |
e37c9e69 | 530 | |
817d52f8 JB |
531 | total_found += add_new_free_space(block_group, fs_info, last, |
532 | block_group->key.objectid + | |
533 | block_group->key.offset); | |
11833d66 | 534 | caching_ctl->progress = (u64)-1; |
817d52f8 | 535 | |
73fa48b6 OS |
536 | out: |
537 | btrfs_free_path(path); | |
538 | return ret; | |
539 | } | |
540 | ||
541 | static noinline void caching_thread(struct btrfs_work *work) | |
542 | { | |
543 | struct btrfs_block_group_cache *block_group; | |
544 | struct btrfs_fs_info *fs_info; | |
545 | struct btrfs_caching_control *caching_ctl; | |
546 | int ret; | |
547 | ||
548 | caching_ctl = container_of(work, struct btrfs_caching_control, work); | |
549 | block_group = caching_ctl->block_group; | |
550 | fs_info = block_group->fs_info; | |
551 | ||
552 | mutex_lock(&caching_ctl->mutex); | |
553 | down_read(&fs_info->commit_root_sem); | |
554 | ||
1e144fb8 OS |
555 | if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) |
556 | ret = load_free_space_tree(caching_ctl); | |
557 | else | |
558 | ret = load_extent_tree_free(caching_ctl); | |
73fa48b6 | 559 | |
817d52f8 | 560 | spin_lock(&block_group->lock); |
11833d66 | 561 | block_group->caching_ctl = NULL; |
73fa48b6 | 562 | block_group->cached = ret ? BTRFS_CACHE_ERROR : BTRFS_CACHE_FINISHED; |
817d52f8 | 563 | spin_unlock(&block_group->lock); |
0f9dd46c | 564 | |
d0bd4560 JB |
565 | #ifdef CONFIG_BTRFS_DEBUG |
566 | if (btrfs_should_fragment_free_space(extent_root, block_group)) { | |
567 | u64 bytes_used; | |
568 | ||
569 | spin_lock(&block_group->space_info->lock); | |
570 | spin_lock(&block_group->lock); | |
571 | bytes_used = block_group->key.offset - | |
572 | btrfs_block_group_used(&block_group->item); | |
573 | block_group->space_info->bytes_used += bytes_used >> 1; | |
574 | spin_unlock(&block_group->lock); | |
575 | spin_unlock(&block_group->space_info->lock); | |
576 | fragment_free_space(extent_root, block_group); | |
577 | } | |
578 | #endif | |
579 | ||
580 | caching_ctl->progress = (u64)-1; | |
11833d66 | 581 | |
9e351cc8 | 582 | up_read(&fs_info->commit_root_sem); |
73fa48b6 | 583 | free_excluded_extents(fs_info->extent_root, block_group); |
11833d66 | 584 | mutex_unlock(&caching_ctl->mutex); |
73fa48b6 | 585 | |
11833d66 YZ |
586 | wake_up(&caching_ctl->wait); |
587 | ||
588 | put_caching_control(caching_ctl); | |
11dfe35a | 589 | btrfs_put_block_group(block_group); |
817d52f8 JB |
590 | } |
591 | ||
9d66e233 | 592 | static int cache_block_group(struct btrfs_block_group_cache *cache, |
9d66e233 | 593 | int load_cache_only) |
817d52f8 | 594 | { |
291c7d2f | 595 | DEFINE_WAIT(wait); |
11833d66 YZ |
596 | struct btrfs_fs_info *fs_info = cache->fs_info; |
597 | struct btrfs_caching_control *caching_ctl; | |
817d52f8 JB |
598 | int ret = 0; |
599 | ||
291c7d2f | 600 | caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS); |
79787eaa JM |
601 | if (!caching_ctl) |
602 | return -ENOMEM; | |
291c7d2f JB |
603 | |
604 | INIT_LIST_HEAD(&caching_ctl->list); | |
605 | mutex_init(&caching_ctl->mutex); | |
606 | init_waitqueue_head(&caching_ctl->wait); | |
607 | caching_ctl->block_group = cache; | |
608 | caching_ctl->progress = cache->key.objectid; | |
609 | atomic_set(&caching_ctl->count, 1); | |
9e0af237 LB |
610 | btrfs_init_work(&caching_ctl->work, btrfs_cache_helper, |
611 | caching_thread, NULL, NULL); | |
291c7d2f JB |
612 | |
613 | spin_lock(&cache->lock); | |
614 | /* | |
615 | * This should be a rare occasion, but this could happen I think in the | |
616 | * case where one thread starts to load the space cache info, and then | |
617 | * some other thread starts a transaction commit which tries to do an | |
618 | * allocation while the other thread is still loading the space cache | |
619 | * info. The previous loop should have kept us from choosing this block | |
620 | * group, but if we've moved to the state where we will wait on caching | |
621 | * block groups we need to first check if we're doing a fast load here, | |
622 | * so we can wait for it to finish, otherwise we could end up allocating | |
623 | * from a block group who's cache gets evicted for one reason or | |
624 | * another. | |
625 | */ | |
626 | while (cache->cached == BTRFS_CACHE_FAST) { | |
627 | struct btrfs_caching_control *ctl; | |
628 | ||
629 | ctl = cache->caching_ctl; | |
630 | atomic_inc(&ctl->count); | |
631 | prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE); | |
632 | spin_unlock(&cache->lock); | |
633 | ||
634 | schedule(); | |
635 | ||
636 | finish_wait(&ctl->wait, &wait); | |
637 | put_caching_control(ctl); | |
638 | spin_lock(&cache->lock); | |
639 | } | |
640 | ||
641 | if (cache->cached != BTRFS_CACHE_NO) { | |
642 | spin_unlock(&cache->lock); | |
643 | kfree(caching_ctl); | |
11833d66 | 644 | return 0; |
291c7d2f JB |
645 | } |
646 | WARN_ON(cache->caching_ctl); | |
647 | cache->caching_ctl = caching_ctl; | |
648 | cache->cached = BTRFS_CACHE_FAST; | |
649 | spin_unlock(&cache->lock); | |
11833d66 | 650 | |
d53ba474 | 651 | if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) { |
cb83b7b8 | 652 | mutex_lock(&caching_ctl->mutex); |
9d66e233 JB |
653 | ret = load_free_space_cache(fs_info, cache); |
654 | ||
655 | spin_lock(&cache->lock); | |
656 | if (ret == 1) { | |
291c7d2f | 657 | cache->caching_ctl = NULL; |
9d66e233 JB |
658 | cache->cached = BTRFS_CACHE_FINISHED; |
659 | cache->last_byte_to_unpin = (u64)-1; | |
cb83b7b8 | 660 | caching_ctl->progress = (u64)-1; |
9d66e233 | 661 | } else { |
291c7d2f JB |
662 | if (load_cache_only) { |
663 | cache->caching_ctl = NULL; | |
664 | cache->cached = BTRFS_CACHE_NO; | |
665 | } else { | |
666 | cache->cached = BTRFS_CACHE_STARTED; | |
4f69cb98 | 667 | cache->has_caching_ctl = 1; |
291c7d2f | 668 | } |
9d66e233 JB |
669 | } |
670 | spin_unlock(&cache->lock); | |
d0bd4560 JB |
671 | #ifdef CONFIG_BTRFS_DEBUG |
672 | if (ret == 1 && | |
673 | btrfs_should_fragment_free_space(fs_info->extent_root, | |
674 | cache)) { | |
675 | u64 bytes_used; | |
676 | ||
677 | spin_lock(&cache->space_info->lock); | |
678 | spin_lock(&cache->lock); | |
679 | bytes_used = cache->key.offset - | |
680 | btrfs_block_group_used(&cache->item); | |
681 | cache->space_info->bytes_used += bytes_used >> 1; | |
682 | spin_unlock(&cache->lock); | |
683 | spin_unlock(&cache->space_info->lock); | |
684 | fragment_free_space(fs_info->extent_root, cache); | |
685 | } | |
686 | #endif | |
cb83b7b8 JB |
687 | mutex_unlock(&caching_ctl->mutex); |
688 | ||
291c7d2f | 689 | wake_up(&caching_ctl->wait); |
3c14874a | 690 | if (ret == 1) { |
291c7d2f | 691 | put_caching_control(caching_ctl); |
3c14874a | 692 | free_excluded_extents(fs_info->extent_root, cache); |
9d66e233 | 693 | return 0; |
3c14874a | 694 | } |
291c7d2f JB |
695 | } else { |
696 | /* | |
1e144fb8 OS |
697 | * We're either using the free space tree or no caching at all. |
698 | * Set cached to the appropriate value and wakeup any waiters. | |
291c7d2f JB |
699 | */ |
700 | spin_lock(&cache->lock); | |
701 | if (load_cache_only) { | |
702 | cache->caching_ctl = NULL; | |
703 | cache->cached = BTRFS_CACHE_NO; | |
704 | } else { | |
705 | cache->cached = BTRFS_CACHE_STARTED; | |
4f69cb98 | 706 | cache->has_caching_ctl = 1; |
291c7d2f JB |
707 | } |
708 | spin_unlock(&cache->lock); | |
709 | wake_up(&caching_ctl->wait); | |
9d66e233 JB |
710 | } |
711 | ||
291c7d2f JB |
712 | if (load_cache_only) { |
713 | put_caching_control(caching_ctl); | |
11833d66 | 714 | return 0; |
817d52f8 | 715 | } |
817d52f8 | 716 | |
9e351cc8 | 717 | down_write(&fs_info->commit_root_sem); |
291c7d2f | 718 | atomic_inc(&caching_ctl->count); |
11833d66 | 719 | list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups); |
9e351cc8 | 720 | up_write(&fs_info->commit_root_sem); |
11833d66 | 721 | |
11dfe35a | 722 | btrfs_get_block_group(cache); |
11833d66 | 723 | |
e66f0bb1 | 724 | btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work); |
817d52f8 | 725 | |
ef8bbdfe | 726 | return ret; |
e37c9e69 CM |
727 | } |
728 | ||
0f9dd46c JB |
729 | /* |
730 | * return the block group that starts at or after bytenr | |
731 | */ | |
d397712b CM |
732 | static struct btrfs_block_group_cache * |
733 | btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr) | |
0ef3e66b | 734 | { |
0f9dd46c | 735 | struct btrfs_block_group_cache *cache; |
0ef3e66b | 736 | |
0f9dd46c | 737 | cache = block_group_cache_tree_search(info, bytenr, 0); |
0ef3e66b | 738 | |
0f9dd46c | 739 | return cache; |
0ef3e66b CM |
740 | } |
741 | ||
0f9dd46c | 742 | /* |
9f55684c | 743 | * return the block group that contains the given bytenr |
0f9dd46c | 744 | */ |
d397712b CM |
745 | struct btrfs_block_group_cache *btrfs_lookup_block_group( |
746 | struct btrfs_fs_info *info, | |
747 | u64 bytenr) | |
be744175 | 748 | { |
0f9dd46c | 749 | struct btrfs_block_group_cache *cache; |
be744175 | 750 | |
0f9dd46c | 751 | cache = block_group_cache_tree_search(info, bytenr, 1); |
96b5179d | 752 | |
0f9dd46c | 753 | return cache; |
be744175 | 754 | } |
0b86a832 | 755 | |
0f9dd46c JB |
756 | static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info, |
757 | u64 flags) | |
6324fbf3 | 758 | { |
0f9dd46c | 759 | struct list_head *head = &info->space_info; |
0f9dd46c | 760 | struct btrfs_space_info *found; |
4184ea7f | 761 | |
52ba6929 | 762 | flags &= BTRFS_BLOCK_GROUP_TYPE_MASK; |
b742bb82 | 763 | |
4184ea7f CM |
764 | rcu_read_lock(); |
765 | list_for_each_entry_rcu(found, head, list) { | |
67377734 | 766 | if (found->flags & flags) { |
4184ea7f | 767 | rcu_read_unlock(); |
0f9dd46c | 768 | return found; |
4184ea7f | 769 | } |
0f9dd46c | 770 | } |
4184ea7f | 771 | rcu_read_unlock(); |
0f9dd46c | 772 | return NULL; |
6324fbf3 CM |
773 | } |
774 | ||
4184ea7f CM |
775 | /* |
776 | * after adding space to the filesystem, we need to clear the full flags | |
777 | * on all the space infos. | |
778 | */ | |
779 | void btrfs_clear_space_info_full(struct btrfs_fs_info *info) | |
780 | { | |
781 | struct list_head *head = &info->space_info; | |
782 | struct btrfs_space_info *found; | |
783 | ||
784 | rcu_read_lock(); | |
785 | list_for_each_entry_rcu(found, head, list) | |
786 | found->full = 0; | |
787 | rcu_read_unlock(); | |
788 | } | |
789 | ||
1a4ed8fd FM |
790 | /* simple helper to search for an existing data extent at a given offset */ |
791 | int btrfs_lookup_data_extent(struct btrfs_root *root, u64 start, u64 len) | |
e02119d5 CM |
792 | { |
793 | int ret; | |
794 | struct btrfs_key key; | |
31840ae1 | 795 | struct btrfs_path *path; |
e02119d5 | 796 | |
31840ae1 | 797 | path = btrfs_alloc_path(); |
d8926bb3 MF |
798 | if (!path) |
799 | return -ENOMEM; | |
800 | ||
e02119d5 CM |
801 | key.objectid = start; |
802 | key.offset = len; | |
3173a18f | 803 | key.type = BTRFS_EXTENT_ITEM_KEY; |
e02119d5 CM |
804 | ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path, |
805 | 0, 0); | |
31840ae1 | 806 | btrfs_free_path(path); |
7bb86316 CM |
807 | return ret; |
808 | } | |
809 | ||
a22285a6 | 810 | /* |
3173a18f | 811 | * helper function to lookup reference count and flags of a tree block. |
a22285a6 YZ |
812 | * |
813 | * the head node for delayed ref is used to store the sum of all the | |
814 | * reference count modifications queued up in the rbtree. the head | |
815 | * node may also store the extent flags to set. This way you can check | |
816 | * to see what the reference count and extent flags would be if all of | |
817 | * the delayed refs are not processed. | |
818 | */ | |
819 | int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans, | |
820 | struct btrfs_root *root, u64 bytenr, | |
3173a18f | 821 | u64 offset, int metadata, u64 *refs, u64 *flags) |
a22285a6 YZ |
822 | { |
823 | struct btrfs_delayed_ref_head *head; | |
824 | struct btrfs_delayed_ref_root *delayed_refs; | |
825 | struct btrfs_path *path; | |
826 | struct btrfs_extent_item *ei; | |
827 | struct extent_buffer *leaf; | |
828 | struct btrfs_key key; | |
829 | u32 item_size; | |
830 | u64 num_refs; | |
831 | u64 extent_flags; | |
832 | int ret; | |
833 | ||
3173a18f JB |
834 | /* |
835 | * If we don't have skinny metadata, don't bother doing anything | |
836 | * different | |
837 | */ | |
838 | if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) { | |
707e8a07 | 839 | offset = root->nodesize; |
3173a18f JB |
840 | metadata = 0; |
841 | } | |
842 | ||
a22285a6 YZ |
843 | path = btrfs_alloc_path(); |
844 | if (!path) | |
845 | return -ENOMEM; | |
846 | ||
a22285a6 YZ |
847 | if (!trans) { |
848 | path->skip_locking = 1; | |
849 | path->search_commit_root = 1; | |
850 | } | |
639eefc8 FDBM |
851 | |
852 | search_again: | |
853 | key.objectid = bytenr; | |
854 | key.offset = offset; | |
855 | if (metadata) | |
856 | key.type = BTRFS_METADATA_ITEM_KEY; | |
857 | else | |
858 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
859 | ||
a22285a6 YZ |
860 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, |
861 | &key, path, 0, 0); | |
862 | if (ret < 0) | |
863 | goto out_free; | |
864 | ||
3173a18f | 865 | if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) { |
74be9510 FDBM |
866 | if (path->slots[0]) { |
867 | path->slots[0]--; | |
868 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
869 | path->slots[0]); | |
870 | if (key.objectid == bytenr && | |
871 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
707e8a07 | 872 | key.offset == root->nodesize) |
74be9510 FDBM |
873 | ret = 0; |
874 | } | |
3173a18f JB |
875 | } |
876 | ||
a22285a6 YZ |
877 | if (ret == 0) { |
878 | leaf = path->nodes[0]; | |
879 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
880 | if (item_size >= sizeof(*ei)) { | |
881 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
882 | struct btrfs_extent_item); | |
883 | num_refs = btrfs_extent_refs(leaf, ei); | |
884 | extent_flags = btrfs_extent_flags(leaf, ei); | |
885 | } else { | |
886 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
887 | struct btrfs_extent_item_v0 *ei0; | |
888 | BUG_ON(item_size != sizeof(*ei0)); | |
889 | ei0 = btrfs_item_ptr(leaf, path->slots[0], | |
890 | struct btrfs_extent_item_v0); | |
891 | num_refs = btrfs_extent_refs_v0(leaf, ei0); | |
892 | /* FIXME: this isn't correct for data */ | |
893 | extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
894 | #else | |
895 | BUG(); | |
896 | #endif | |
897 | } | |
898 | BUG_ON(num_refs == 0); | |
899 | } else { | |
900 | num_refs = 0; | |
901 | extent_flags = 0; | |
902 | ret = 0; | |
903 | } | |
904 | ||
905 | if (!trans) | |
906 | goto out; | |
907 | ||
908 | delayed_refs = &trans->transaction->delayed_refs; | |
909 | spin_lock(&delayed_refs->lock); | |
910 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
911 | if (head) { | |
912 | if (!mutex_trylock(&head->mutex)) { | |
913 | atomic_inc(&head->node.refs); | |
914 | spin_unlock(&delayed_refs->lock); | |
915 | ||
b3b4aa74 | 916 | btrfs_release_path(path); |
a22285a6 | 917 | |
8cc33e5c DS |
918 | /* |
919 | * Mutex was contended, block until it's released and try | |
920 | * again | |
921 | */ | |
a22285a6 YZ |
922 | mutex_lock(&head->mutex); |
923 | mutex_unlock(&head->mutex); | |
924 | btrfs_put_delayed_ref(&head->node); | |
639eefc8 | 925 | goto search_again; |
a22285a6 | 926 | } |
d7df2c79 | 927 | spin_lock(&head->lock); |
a22285a6 YZ |
928 | if (head->extent_op && head->extent_op->update_flags) |
929 | extent_flags |= head->extent_op->flags_to_set; | |
930 | else | |
931 | BUG_ON(num_refs == 0); | |
932 | ||
933 | num_refs += head->node.ref_mod; | |
d7df2c79 | 934 | spin_unlock(&head->lock); |
a22285a6 YZ |
935 | mutex_unlock(&head->mutex); |
936 | } | |
937 | spin_unlock(&delayed_refs->lock); | |
938 | out: | |
939 | WARN_ON(num_refs == 0); | |
940 | if (refs) | |
941 | *refs = num_refs; | |
942 | if (flags) | |
943 | *flags = extent_flags; | |
944 | out_free: | |
945 | btrfs_free_path(path); | |
946 | return ret; | |
947 | } | |
948 | ||
d8d5f3e1 CM |
949 | /* |
950 | * Back reference rules. Back refs have three main goals: | |
951 | * | |
952 | * 1) differentiate between all holders of references to an extent so that | |
953 | * when a reference is dropped we can make sure it was a valid reference | |
954 | * before freeing the extent. | |
955 | * | |
956 | * 2) Provide enough information to quickly find the holders of an extent | |
957 | * if we notice a given block is corrupted or bad. | |
958 | * | |
959 | * 3) Make it easy to migrate blocks for FS shrinking or storage pool | |
960 | * maintenance. This is actually the same as #2, but with a slightly | |
961 | * different use case. | |
962 | * | |
5d4f98a2 YZ |
963 | * There are two kinds of back refs. The implicit back refs is optimized |
964 | * for pointers in non-shared tree blocks. For a given pointer in a block, | |
965 | * back refs of this kind provide information about the block's owner tree | |
966 | * and the pointer's key. These information allow us to find the block by | |
967 | * b-tree searching. The full back refs is for pointers in tree blocks not | |
968 | * referenced by their owner trees. The location of tree block is recorded | |
969 | * in the back refs. Actually the full back refs is generic, and can be | |
970 | * used in all cases the implicit back refs is used. The major shortcoming | |
971 | * of the full back refs is its overhead. Every time a tree block gets | |
972 | * COWed, we have to update back refs entry for all pointers in it. | |
973 | * | |
974 | * For a newly allocated tree block, we use implicit back refs for | |
975 | * pointers in it. This means most tree related operations only involve | |
976 | * implicit back refs. For a tree block created in old transaction, the | |
977 | * only way to drop a reference to it is COW it. So we can detect the | |
978 | * event that tree block loses its owner tree's reference and do the | |
979 | * back refs conversion. | |
980 | * | |
981 | * When a tree block is COW'd through a tree, there are four cases: | |
982 | * | |
983 | * The reference count of the block is one and the tree is the block's | |
984 | * owner tree. Nothing to do in this case. | |
985 | * | |
986 | * The reference count of the block is one and the tree is not the | |
987 | * block's owner tree. In this case, full back refs is used for pointers | |
988 | * in the block. Remove these full back refs, add implicit back refs for | |
989 | * every pointers in the new block. | |
990 | * | |
991 | * The reference count of the block is greater than one and the tree is | |
992 | * the block's owner tree. In this case, implicit back refs is used for | |
993 | * pointers in the block. Add full back refs for every pointers in the | |
994 | * block, increase lower level extents' reference counts. The original | |
995 | * implicit back refs are entailed to the new block. | |
996 | * | |
997 | * The reference count of the block is greater than one and the tree is | |
998 | * not the block's owner tree. Add implicit back refs for every pointer in | |
999 | * the new block, increase lower level extents' reference count. | |
1000 | * | |
1001 | * Back Reference Key composing: | |
1002 | * | |
1003 | * The key objectid corresponds to the first byte in the extent, | |
1004 | * The key type is used to differentiate between types of back refs. | |
1005 | * There are different meanings of the key offset for different types | |
1006 | * of back refs. | |
1007 | * | |
d8d5f3e1 CM |
1008 | * File extents can be referenced by: |
1009 | * | |
1010 | * - multiple snapshots, subvolumes, or different generations in one subvol | |
31840ae1 | 1011 | * - different files inside a single subvolume |
d8d5f3e1 CM |
1012 | * - different offsets inside a file (bookend extents in file.c) |
1013 | * | |
5d4f98a2 | 1014 | * The extent ref structure for the implicit back refs has fields for: |
d8d5f3e1 CM |
1015 | * |
1016 | * - Objectid of the subvolume root | |
d8d5f3e1 | 1017 | * - objectid of the file holding the reference |
5d4f98a2 YZ |
1018 | * - original offset in the file |
1019 | * - how many bookend extents | |
d8d5f3e1 | 1020 | * |
5d4f98a2 YZ |
1021 | * The key offset for the implicit back refs is hash of the first |
1022 | * three fields. | |
d8d5f3e1 | 1023 | * |
5d4f98a2 | 1024 | * The extent ref structure for the full back refs has field for: |
d8d5f3e1 | 1025 | * |
5d4f98a2 | 1026 | * - number of pointers in the tree leaf |
d8d5f3e1 | 1027 | * |
5d4f98a2 YZ |
1028 | * The key offset for the implicit back refs is the first byte of |
1029 | * the tree leaf | |
d8d5f3e1 | 1030 | * |
5d4f98a2 YZ |
1031 | * When a file extent is allocated, The implicit back refs is used. |
1032 | * the fields are filled in: | |
d8d5f3e1 | 1033 | * |
5d4f98a2 | 1034 | * (root_key.objectid, inode objectid, offset in file, 1) |
d8d5f3e1 | 1035 | * |
5d4f98a2 YZ |
1036 | * When a file extent is removed file truncation, we find the |
1037 | * corresponding implicit back refs and check the following fields: | |
d8d5f3e1 | 1038 | * |
5d4f98a2 | 1039 | * (btrfs_header_owner(leaf), inode objectid, offset in file) |
d8d5f3e1 | 1040 | * |
5d4f98a2 | 1041 | * Btree extents can be referenced by: |
d8d5f3e1 | 1042 | * |
5d4f98a2 | 1043 | * - Different subvolumes |
d8d5f3e1 | 1044 | * |
5d4f98a2 YZ |
1045 | * Both the implicit back refs and the full back refs for tree blocks |
1046 | * only consist of key. The key offset for the implicit back refs is | |
1047 | * objectid of block's owner tree. The key offset for the full back refs | |
1048 | * is the first byte of parent block. | |
d8d5f3e1 | 1049 | * |
5d4f98a2 YZ |
1050 | * When implicit back refs is used, information about the lowest key and |
1051 | * level of the tree block are required. These information are stored in | |
1052 | * tree block info structure. | |
d8d5f3e1 | 1053 | */ |
31840ae1 | 1054 | |
5d4f98a2 YZ |
1055 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
1056 | static int convert_extent_item_v0(struct btrfs_trans_handle *trans, | |
1057 | struct btrfs_root *root, | |
1058 | struct btrfs_path *path, | |
1059 | u64 owner, u32 extra_size) | |
7bb86316 | 1060 | { |
5d4f98a2 YZ |
1061 | struct btrfs_extent_item *item; |
1062 | struct btrfs_extent_item_v0 *ei0; | |
1063 | struct btrfs_extent_ref_v0 *ref0; | |
1064 | struct btrfs_tree_block_info *bi; | |
1065 | struct extent_buffer *leaf; | |
7bb86316 | 1066 | struct btrfs_key key; |
5d4f98a2 YZ |
1067 | struct btrfs_key found_key; |
1068 | u32 new_size = sizeof(*item); | |
1069 | u64 refs; | |
1070 | int ret; | |
1071 | ||
1072 | leaf = path->nodes[0]; | |
1073 | BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0)); | |
1074 | ||
1075 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1076 | ei0 = btrfs_item_ptr(leaf, path->slots[0], | |
1077 | struct btrfs_extent_item_v0); | |
1078 | refs = btrfs_extent_refs_v0(leaf, ei0); | |
1079 | ||
1080 | if (owner == (u64)-1) { | |
1081 | while (1) { | |
1082 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1083 | ret = btrfs_next_leaf(root, path); | |
1084 | if (ret < 0) | |
1085 | return ret; | |
79787eaa | 1086 | BUG_ON(ret > 0); /* Corruption */ |
5d4f98a2 YZ |
1087 | leaf = path->nodes[0]; |
1088 | } | |
1089 | btrfs_item_key_to_cpu(leaf, &found_key, | |
1090 | path->slots[0]); | |
1091 | BUG_ON(key.objectid != found_key.objectid); | |
1092 | if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) { | |
1093 | path->slots[0]++; | |
1094 | continue; | |
1095 | } | |
1096 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1097 | struct btrfs_extent_ref_v0); | |
1098 | owner = btrfs_ref_objectid_v0(leaf, ref0); | |
1099 | break; | |
1100 | } | |
1101 | } | |
b3b4aa74 | 1102 | btrfs_release_path(path); |
5d4f98a2 YZ |
1103 | |
1104 | if (owner < BTRFS_FIRST_FREE_OBJECTID) | |
1105 | new_size += sizeof(*bi); | |
1106 | ||
1107 | new_size -= sizeof(*ei0); | |
1108 | ret = btrfs_search_slot(trans, root, &key, path, | |
1109 | new_size + extra_size, 1); | |
1110 | if (ret < 0) | |
1111 | return ret; | |
79787eaa | 1112 | BUG_ON(ret); /* Corruption */ |
5d4f98a2 | 1113 | |
4b90c680 | 1114 | btrfs_extend_item(root, path, new_size); |
5d4f98a2 YZ |
1115 | |
1116 | leaf = path->nodes[0]; | |
1117 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1118 | btrfs_set_extent_refs(leaf, item, refs); | |
1119 | /* FIXME: get real generation */ | |
1120 | btrfs_set_extent_generation(leaf, item, 0); | |
1121 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1122 | btrfs_set_extent_flags(leaf, item, | |
1123 | BTRFS_EXTENT_FLAG_TREE_BLOCK | | |
1124 | BTRFS_BLOCK_FLAG_FULL_BACKREF); | |
1125 | bi = (struct btrfs_tree_block_info *)(item + 1); | |
1126 | /* FIXME: get first key of the block */ | |
1127 | memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi)); | |
1128 | btrfs_set_tree_block_level(leaf, bi, (int)owner); | |
1129 | } else { | |
1130 | btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA); | |
1131 | } | |
1132 | btrfs_mark_buffer_dirty(leaf); | |
1133 | return 0; | |
1134 | } | |
1135 | #endif | |
1136 | ||
1137 | static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset) | |
1138 | { | |
1139 | u32 high_crc = ~(u32)0; | |
1140 | u32 low_crc = ~(u32)0; | |
1141 | __le64 lenum; | |
1142 | ||
1143 | lenum = cpu_to_le64(root_objectid); | |
14a958e6 | 1144 | high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 1145 | lenum = cpu_to_le64(owner); |
14a958e6 | 1146 | low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 1147 | lenum = cpu_to_le64(offset); |
14a958e6 | 1148 | low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 YZ |
1149 | |
1150 | return ((u64)high_crc << 31) ^ (u64)low_crc; | |
1151 | } | |
1152 | ||
1153 | static u64 hash_extent_data_ref_item(struct extent_buffer *leaf, | |
1154 | struct btrfs_extent_data_ref *ref) | |
1155 | { | |
1156 | return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref), | |
1157 | btrfs_extent_data_ref_objectid(leaf, ref), | |
1158 | btrfs_extent_data_ref_offset(leaf, ref)); | |
1159 | } | |
1160 | ||
1161 | static int match_extent_data_ref(struct extent_buffer *leaf, | |
1162 | struct btrfs_extent_data_ref *ref, | |
1163 | u64 root_objectid, u64 owner, u64 offset) | |
1164 | { | |
1165 | if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid || | |
1166 | btrfs_extent_data_ref_objectid(leaf, ref) != owner || | |
1167 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
1168 | return 0; | |
1169 | return 1; | |
1170 | } | |
1171 | ||
1172 | static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans, | |
1173 | struct btrfs_root *root, | |
1174 | struct btrfs_path *path, | |
1175 | u64 bytenr, u64 parent, | |
1176 | u64 root_objectid, | |
1177 | u64 owner, u64 offset) | |
1178 | { | |
1179 | struct btrfs_key key; | |
1180 | struct btrfs_extent_data_ref *ref; | |
31840ae1 | 1181 | struct extent_buffer *leaf; |
5d4f98a2 | 1182 | u32 nritems; |
74493f7a | 1183 | int ret; |
5d4f98a2 YZ |
1184 | int recow; |
1185 | int err = -ENOENT; | |
74493f7a | 1186 | |
31840ae1 | 1187 | key.objectid = bytenr; |
5d4f98a2 YZ |
1188 | if (parent) { |
1189 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1190 | key.offset = parent; | |
1191 | } else { | |
1192 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1193 | key.offset = hash_extent_data_ref(root_objectid, | |
1194 | owner, offset); | |
1195 | } | |
1196 | again: | |
1197 | recow = 0; | |
1198 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1199 | if (ret < 0) { | |
1200 | err = ret; | |
1201 | goto fail; | |
1202 | } | |
31840ae1 | 1203 | |
5d4f98a2 YZ |
1204 | if (parent) { |
1205 | if (!ret) | |
1206 | return 0; | |
1207 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1208 | key.type = BTRFS_EXTENT_REF_V0_KEY; | |
b3b4aa74 | 1209 | btrfs_release_path(path); |
5d4f98a2 YZ |
1210 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1211 | if (ret < 0) { | |
1212 | err = ret; | |
1213 | goto fail; | |
1214 | } | |
1215 | if (!ret) | |
1216 | return 0; | |
1217 | #endif | |
1218 | goto fail; | |
31840ae1 ZY |
1219 | } |
1220 | ||
1221 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1222 | nritems = btrfs_header_nritems(leaf); |
1223 | while (1) { | |
1224 | if (path->slots[0] >= nritems) { | |
1225 | ret = btrfs_next_leaf(root, path); | |
1226 | if (ret < 0) | |
1227 | err = ret; | |
1228 | if (ret) | |
1229 | goto fail; | |
1230 | ||
1231 | leaf = path->nodes[0]; | |
1232 | nritems = btrfs_header_nritems(leaf); | |
1233 | recow = 1; | |
1234 | } | |
1235 | ||
1236 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1237 | if (key.objectid != bytenr || | |
1238 | key.type != BTRFS_EXTENT_DATA_REF_KEY) | |
1239 | goto fail; | |
1240 | ||
1241 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1242 | struct btrfs_extent_data_ref); | |
1243 | ||
1244 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1245 | owner, offset)) { | |
1246 | if (recow) { | |
b3b4aa74 | 1247 | btrfs_release_path(path); |
5d4f98a2 YZ |
1248 | goto again; |
1249 | } | |
1250 | err = 0; | |
1251 | break; | |
1252 | } | |
1253 | path->slots[0]++; | |
31840ae1 | 1254 | } |
5d4f98a2 YZ |
1255 | fail: |
1256 | return err; | |
31840ae1 ZY |
1257 | } |
1258 | ||
5d4f98a2 YZ |
1259 | static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans, |
1260 | struct btrfs_root *root, | |
1261 | struct btrfs_path *path, | |
1262 | u64 bytenr, u64 parent, | |
1263 | u64 root_objectid, u64 owner, | |
1264 | u64 offset, int refs_to_add) | |
31840ae1 ZY |
1265 | { |
1266 | struct btrfs_key key; | |
1267 | struct extent_buffer *leaf; | |
5d4f98a2 | 1268 | u32 size; |
31840ae1 ZY |
1269 | u32 num_refs; |
1270 | int ret; | |
74493f7a | 1271 | |
74493f7a | 1272 | key.objectid = bytenr; |
5d4f98a2 YZ |
1273 | if (parent) { |
1274 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1275 | key.offset = parent; | |
1276 | size = sizeof(struct btrfs_shared_data_ref); | |
1277 | } else { | |
1278 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1279 | key.offset = hash_extent_data_ref(root_objectid, | |
1280 | owner, offset); | |
1281 | size = sizeof(struct btrfs_extent_data_ref); | |
1282 | } | |
74493f7a | 1283 | |
5d4f98a2 YZ |
1284 | ret = btrfs_insert_empty_item(trans, root, path, &key, size); |
1285 | if (ret && ret != -EEXIST) | |
1286 | goto fail; | |
1287 | ||
1288 | leaf = path->nodes[0]; | |
1289 | if (parent) { | |
1290 | struct btrfs_shared_data_ref *ref; | |
31840ae1 | 1291 | ref = btrfs_item_ptr(leaf, path->slots[0], |
5d4f98a2 YZ |
1292 | struct btrfs_shared_data_ref); |
1293 | if (ret == 0) { | |
1294 | btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add); | |
1295 | } else { | |
1296 | num_refs = btrfs_shared_data_ref_count(leaf, ref); | |
1297 | num_refs += refs_to_add; | |
1298 | btrfs_set_shared_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1299 | } |
5d4f98a2 YZ |
1300 | } else { |
1301 | struct btrfs_extent_data_ref *ref; | |
1302 | while (ret == -EEXIST) { | |
1303 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1304 | struct btrfs_extent_data_ref); | |
1305 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1306 | owner, offset)) | |
1307 | break; | |
b3b4aa74 | 1308 | btrfs_release_path(path); |
5d4f98a2 YZ |
1309 | key.offset++; |
1310 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1311 | size); | |
1312 | if (ret && ret != -EEXIST) | |
1313 | goto fail; | |
31840ae1 | 1314 | |
5d4f98a2 YZ |
1315 | leaf = path->nodes[0]; |
1316 | } | |
1317 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1318 | struct btrfs_extent_data_ref); | |
1319 | if (ret == 0) { | |
1320 | btrfs_set_extent_data_ref_root(leaf, ref, | |
1321 | root_objectid); | |
1322 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
1323 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
1324 | btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add); | |
1325 | } else { | |
1326 | num_refs = btrfs_extent_data_ref_count(leaf, ref); | |
1327 | num_refs += refs_to_add; | |
1328 | btrfs_set_extent_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1329 | } |
31840ae1 | 1330 | } |
5d4f98a2 YZ |
1331 | btrfs_mark_buffer_dirty(leaf); |
1332 | ret = 0; | |
1333 | fail: | |
b3b4aa74 | 1334 | btrfs_release_path(path); |
7bb86316 | 1335 | return ret; |
74493f7a CM |
1336 | } |
1337 | ||
5d4f98a2 YZ |
1338 | static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans, |
1339 | struct btrfs_root *root, | |
1340 | struct btrfs_path *path, | |
fcebe456 | 1341 | int refs_to_drop, int *last_ref) |
31840ae1 | 1342 | { |
5d4f98a2 YZ |
1343 | struct btrfs_key key; |
1344 | struct btrfs_extent_data_ref *ref1 = NULL; | |
1345 | struct btrfs_shared_data_ref *ref2 = NULL; | |
31840ae1 | 1346 | struct extent_buffer *leaf; |
5d4f98a2 | 1347 | u32 num_refs = 0; |
31840ae1 ZY |
1348 | int ret = 0; |
1349 | ||
1350 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1351 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
1352 | ||
1353 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1354 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1355 | struct btrfs_extent_data_ref); | |
1356 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1357 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1358 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1359 | struct btrfs_shared_data_ref); | |
1360 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1361 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1362 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
1363 | struct btrfs_extent_ref_v0 *ref0; | |
1364 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1365 | struct btrfs_extent_ref_v0); | |
1366 | num_refs = btrfs_ref_count_v0(leaf, ref0); | |
1367 | #endif | |
1368 | } else { | |
1369 | BUG(); | |
1370 | } | |
1371 | ||
56bec294 CM |
1372 | BUG_ON(num_refs < refs_to_drop); |
1373 | num_refs -= refs_to_drop; | |
5d4f98a2 | 1374 | |
31840ae1 ZY |
1375 | if (num_refs == 0) { |
1376 | ret = btrfs_del_item(trans, root, path); | |
fcebe456 | 1377 | *last_ref = 1; |
31840ae1 | 1378 | } else { |
5d4f98a2 YZ |
1379 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) |
1380 | btrfs_set_extent_data_ref_count(leaf, ref1, num_refs); | |
1381 | else if (key.type == BTRFS_SHARED_DATA_REF_KEY) | |
1382 | btrfs_set_shared_data_ref_count(leaf, ref2, num_refs); | |
1383 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1384 | else { | |
1385 | struct btrfs_extent_ref_v0 *ref0; | |
1386 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1387 | struct btrfs_extent_ref_v0); | |
1388 | btrfs_set_ref_count_v0(leaf, ref0, num_refs); | |
1389 | } | |
1390 | #endif | |
31840ae1 ZY |
1391 | btrfs_mark_buffer_dirty(leaf); |
1392 | } | |
31840ae1 ZY |
1393 | return ret; |
1394 | } | |
1395 | ||
9ed0dea0 | 1396 | static noinline u32 extent_data_ref_count(struct btrfs_path *path, |
5d4f98a2 | 1397 | struct btrfs_extent_inline_ref *iref) |
15916de8 | 1398 | { |
5d4f98a2 YZ |
1399 | struct btrfs_key key; |
1400 | struct extent_buffer *leaf; | |
1401 | struct btrfs_extent_data_ref *ref1; | |
1402 | struct btrfs_shared_data_ref *ref2; | |
1403 | u32 num_refs = 0; | |
1404 | ||
1405 | leaf = path->nodes[0]; | |
1406 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1407 | if (iref) { | |
1408 | if (btrfs_extent_inline_ref_type(leaf, iref) == | |
1409 | BTRFS_EXTENT_DATA_REF_KEY) { | |
1410 | ref1 = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1411 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1412 | } else { | |
1413 | ref2 = (struct btrfs_shared_data_ref *)(iref + 1); | |
1414 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1415 | } | |
1416 | } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1417 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1418 | struct btrfs_extent_data_ref); | |
1419 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1420 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1421 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1422 | struct btrfs_shared_data_ref); | |
1423 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1424 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1425 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
1426 | struct btrfs_extent_ref_v0 *ref0; | |
1427 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1428 | struct btrfs_extent_ref_v0); | |
1429 | num_refs = btrfs_ref_count_v0(leaf, ref0); | |
4b4e25f2 | 1430 | #endif |
5d4f98a2 YZ |
1431 | } else { |
1432 | WARN_ON(1); | |
1433 | } | |
1434 | return num_refs; | |
1435 | } | |
15916de8 | 1436 | |
5d4f98a2 YZ |
1437 | static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans, |
1438 | struct btrfs_root *root, | |
1439 | struct btrfs_path *path, | |
1440 | u64 bytenr, u64 parent, | |
1441 | u64 root_objectid) | |
1f3c79a2 | 1442 | { |
5d4f98a2 | 1443 | struct btrfs_key key; |
1f3c79a2 | 1444 | int ret; |
1f3c79a2 | 1445 | |
5d4f98a2 YZ |
1446 | key.objectid = bytenr; |
1447 | if (parent) { | |
1448 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1449 | key.offset = parent; | |
1450 | } else { | |
1451 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1452 | key.offset = root_objectid; | |
1f3c79a2 LH |
1453 | } |
1454 | ||
5d4f98a2 YZ |
1455 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1456 | if (ret > 0) | |
1457 | ret = -ENOENT; | |
1458 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1459 | if (ret == -ENOENT && parent) { | |
b3b4aa74 | 1460 | btrfs_release_path(path); |
5d4f98a2 YZ |
1461 | key.type = BTRFS_EXTENT_REF_V0_KEY; |
1462 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1463 | if (ret > 0) | |
1464 | ret = -ENOENT; | |
1465 | } | |
1f3c79a2 | 1466 | #endif |
5d4f98a2 | 1467 | return ret; |
1f3c79a2 LH |
1468 | } |
1469 | ||
5d4f98a2 YZ |
1470 | static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans, |
1471 | struct btrfs_root *root, | |
1472 | struct btrfs_path *path, | |
1473 | u64 bytenr, u64 parent, | |
1474 | u64 root_objectid) | |
31840ae1 | 1475 | { |
5d4f98a2 | 1476 | struct btrfs_key key; |
31840ae1 | 1477 | int ret; |
31840ae1 | 1478 | |
5d4f98a2 YZ |
1479 | key.objectid = bytenr; |
1480 | if (parent) { | |
1481 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1482 | key.offset = parent; | |
1483 | } else { | |
1484 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1485 | key.offset = root_objectid; | |
1486 | } | |
1487 | ||
1488 | ret = btrfs_insert_empty_item(trans, root, path, &key, 0); | |
b3b4aa74 | 1489 | btrfs_release_path(path); |
31840ae1 ZY |
1490 | return ret; |
1491 | } | |
1492 | ||
5d4f98a2 | 1493 | static inline int extent_ref_type(u64 parent, u64 owner) |
31840ae1 | 1494 | { |
5d4f98a2 YZ |
1495 | int type; |
1496 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1497 | if (parent > 0) | |
1498 | type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1499 | else | |
1500 | type = BTRFS_TREE_BLOCK_REF_KEY; | |
1501 | } else { | |
1502 | if (parent > 0) | |
1503 | type = BTRFS_SHARED_DATA_REF_KEY; | |
1504 | else | |
1505 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
1506 | } | |
1507 | return type; | |
31840ae1 | 1508 | } |
56bec294 | 1509 | |
2c47e605 YZ |
1510 | static int find_next_key(struct btrfs_path *path, int level, |
1511 | struct btrfs_key *key) | |
56bec294 | 1512 | |
02217ed2 | 1513 | { |
2c47e605 | 1514 | for (; level < BTRFS_MAX_LEVEL; level++) { |
5d4f98a2 YZ |
1515 | if (!path->nodes[level]) |
1516 | break; | |
5d4f98a2 YZ |
1517 | if (path->slots[level] + 1 >= |
1518 | btrfs_header_nritems(path->nodes[level])) | |
1519 | continue; | |
1520 | if (level == 0) | |
1521 | btrfs_item_key_to_cpu(path->nodes[level], key, | |
1522 | path->slots[level] + 1); | |
1523 | else | |
1524 | btrfs_node_key_to_cpu(path->nodes[level], key, | |
1525 | path->slots[level] + 1); | |
1526 | return 0; | |
1527 | } | |
1528 | return 1; | |
1529 | } | |
037e6390 | 1530 | |
5d4f98a2 YZ |
1531 | /* |
1532 | * look for inline back ref. if back ref is found, *ref_ret is set | |
1533 | * to the address of inline back ref, and 0 is returned. | |
1534 | * | |
1535 | * if back ref isn't found, *ref_ret is set to the address where it | |
1536 | * should be inserted, and -ENOENT is returned. | |
1537 | * | |
1538 | * if insert is true and there are too many inline back refs, the path | |
1539 | * points to the extent item, and -EAGAIN is returned. | |
1540 | * | |
1541 | * NOTE: inline back refs are ordered in the same way that back ref | |
1542 | * items in the tree are ordered. | |
1543 | */ | |
1544 | static noinline_for_stack | |
1545 | int lookup_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1546 | struct btrfs_root *root, | |
1547 | struct btrfs_path *path, | |
1548 | struct btrfs_extent_inline_ref **ref_ret, | |
1549 | u64 bytenr, u64 num_bytes, | |
1550 | u64 parent, u64 root_objectid, | |
1551 | u64 owner, u64 offset, int insert) | |
1552 | { | |
1553 | struct btrfs_key key; | |
1554 | struct extent_buffer *leaf; | |
1555 | struct btrfs_extent_item *ei; | |
1556 | struct btrfs_extent_inline_ref *iref; | |
1557 | u64 flags; | |
1558 | u64 item_size; | |
1559 | unsigned long ptr; | |
1560 | unsigned long end; | |
1561 | int extra_size; | |
1562 | int type; | |
1563 | int want; | |
1564 | int ret; | |
1565 | int err = 0; | |
3173a18f JB |
1566 | bool skinny_metadata = btrfs_fs_incompat(root->fs_info, |
1567 | SKINNY_METADATA); | |
26b8003f | 1568 | |
db94535d | 1569 | key.objectid = bytenr; |
31840ae1 | 1570 | key.type = BTRFS_EXTENT_ITEM_KEY; |
56bec294 | 1571 | key.offset = num_bytes; |
31840ae1 | 1572 | |
5d4f98a2 YZ |
1573 | want = extent_ref_type(parent, owner); |
1574 | if (insert) { | |
1575 | extra_size = btrfs_extent_inline_ref_size(want); | |
85d4198e | 1576 | path->keep_locks = 1; |
5d4f98a2 YZ |
1577 | } else |
1578 | extra_size = -1; | |
3173a18f JB |
1579 | |
1580 | /* | |
1581 | * Owner is our parent level, so we can just add one to get the level | |
1582 | * for the block we are interested in. | |
1583 | */ | |
1584 | if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1585 | key.type = BTRFS_METADATA_ITEM_KEY; | |
1586 | key.offset = owner; | |
1587 | } | |
1588 | ||
1589 | again: | |
5d4f98a2 | 1590 | ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1); |
b9473439 | 1591 | if (ret < 0) { |
5d4f98a2 YZ |
1592 | err = ret; |
1593 | goto out; | |
1594 | } | |
3173a18f JB |
1595 | |
1596 | /* | |
1597 | * We may be a newly converted file system which still has the old fat | |
1598 | * extent entries for metadata, so try and see if we have one of those. | |
1599 | */ | |
1600 | if (ret > 0 && skinny_metadata) { | |
1601 | skinny_metadata = false; | |
1602 | if (path->slots[0]) { | |
1603 | path->slots[0]--; | |
1604 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
1605 | path->slots[0]); | |
1606 | if (key.objectid == bytenr && | |
1607 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
1608 | key.offset == num_bytes) | |
1609 | ret = 0; | |
1610 | } | |
1611 | if (ret) { | |
9ce49a0b | 1612 | key.objectid = bytenr; |
3173a18f JB |
1613 | key.type = BTRFS_EXTENT_ITEM_KEY; |
1614 | key.offset = num_bytes; | |
1615 | btrfs_release_path(path); | |
1616 | goto again; | |
1617 | } | |
1618 | } | |
1619 | ||
79787eaa JM |
1620 | if (ret && !insert) { |
1621 | err = -ENOENT; | |
1622 | goto out; | |
fae7f21c | 1623 | } else if (WARN_ON(ret)) { |
492104c8 | 1624 | err = -EIO; |
492104c8 | 1625 | goto out; |
79787eaa | 1626 | } |
5d4f98a2 YZ |
1627 | |
1628 | leaf = path->nodes[0]; | |
1629 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1630 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1631 | if (item_size < sizeof(*ei)) { | |
1632 | if (!insert) { | |
1633 | err = -ENOENT; | |
1634 | goto out; | |
1635 | } | |
1636 | ret = convert_extent_item_v0(trans, root, path, owner, | |
1637 | extra_size); | |
1638 | if (ret < 0) { | |
1639 | err = ret; | |
1640 | goto out; | |
1641 | } | |
1642 | leaf = path->nodes[0]; | |
1643 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1644 | } | |
1645 | #endif | |
1646 | BUG_ON(item_size < sizeof(*ei)); | |
1647 | ||
5d4f98a2 YZ |
1648 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1649 | flags = btrfs_extent_flags(leaf, ei); | |
1650 | ||
1651 | ptr = (unsigned long)(ei + 1); | |
1652 | end = (unsigned long)ei + item_size; | |
1653 | ||
3173a18f | 1654 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) { |
5d4f98a2 YZ |
1655 | ptr += sizeof(struct btrfs_tree_block_info); |
1656 | BUG_ON(ptr > end); | |
5d4f98a2 YZ |
1657 | } |
1658 | ||
1659 | err = -ENOENT; | |
1660 | while (1) { | |
1661 | if (ptr >= end) { | |
1662 | WARN_ON(ptr > end); | |
1663 | break; | |
1664 | } | |
1665 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1666 | type = btrfs_extent_inline_ref_type(leaf, iref); | |
1667 | if (want < type) | |
1668 | break; | |
1669 | if (want > type) { | |
1670 | ptr += btrfs_extent_inline_ref_size(type); | |
1671 | continue; | |
1672 | } | |
1673 | ||
1674 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1675 | struct btrfs_extent_data_ref *dref; | |
1676 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1677 | if (match_extent_data_ref(leaf, dref, root_objectid, | |
1678 | owner, offset)) { | |
1679 | err = 0; | |
1680 | break; | |
1681 | } | |
1682 | if (hash_extent_data_ref_item(leaf, dref) < | |
1683 | hash_extent_data_ref(root_objectid, owner, offset)) | |
1684 | break; | |
1685 | } else { | |
1686 | u64 ref_offset; | |
1687 | ref_offset = btrfs_extent_inline_ref_offset(leaf, iref); | |
1688 | if (parent > 0) { | |
1689 | if (parent == ref_offset) { | |
1690 | err = 0; | |
1691 | break; | |
1692 | } | |
1693 | if (ref_offset < parent) | |
1694 | break; | |
1695 | } else { | |
1696 | if (root_objectid == ref_offset) { | |
1697 | err = 0; | |
1698 | break; | |
1699 | } | |
1700 | if (ref_offset < root_objectid) | |
1701 | break; | |
1702 | } | |
1703 | } | |
1704 | ptr += btrfs_extent_inline_ref_size(type); | |
1705 | } | |
1706 | if (err == -ENOENT && insert) { | |
1707 | if (item_size + extra_size >= | |
1708 | BTRFS_MAX_EXTENT_ITEM_SIZE(root)) { | |
1709 | err = -EAGAIN; | |
1710 | goto out; | |
1711 | } | |
1712 | /* | |
1713 | * To add new inline back ref, we have to make sure | |
1714 | * there is no corresponding back ref item. | |
1715 | * For simplicity, we just do not add new inline back | |
1716 | * ref if there is any kind of item for this block | |
1717 | */ | |
2c47e605 YZ |
1718 | if (find_next_key(path, 0, &key) == 0 && |
1719 | key.objectid == bytenr && | |
85d4198e | 1720 | key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) { |
5d4f98a2 YZ |
1721 | err = -EAGAIN; |
1722 | goto out; | |
1723 | } | |
1724 | } | |
1725 | *ref_ret = (struct btrfs_extent_inline_ref *)ptr; | |
1726 | out: | |
85d4198e | 1727 | if (insert) { |
5d4f98a2 YZ |
1728 | path->keep_locks = 0; |
1729 | btrfs_unlock_up_safe(path, 1); | |
1730 | } | |
1731 | return err; | |
1732 | } | |
1733 | ||
1734 | /* | |
1735 | * helper to add new inline back ref | |
1736 | */ | |
1737 | static noinline_for_stack | |
fd279fae | 1738 | void setup_inline_extent_backref(struct btrfs_root *root, |
143bede5 JM |
1739 | struct btrfs_path *path, |
1740 | struct btrfs_extent_inline_ref *iref, | |
1741 | u64 parent, u64 root_objectid, | |
1742 | u64 owner, u64 offset, int refs_to_add, | |
1743 | struct btrfs_delayed_extent_op *extent_op) | |
5d4f98a2 YZ |
1744 | { |
1745 | struct extent_buffer *leaf; | |
1746 | struct btrfs_extent_item *ei; | |
1747 | unsigned long ptr; | |
1748 | unsigned long end; | |
1749 | unsigned long item_offset; | |
1750 | u64 refs; | |
1751 | int size; | |
1752 | int type; | |
5d4f98a2 YZ |
1753 | |
1754 | leaf = path->nodes[0]; | |
1755 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1756 | item_offset = (unsigned long)iref - (unsigned long)ei; | |
1757 | ||
1758 | type = extent_ref_type(parent, owner); | |
1759 | size = btrfs_extent_inline_ref_size(type); | |
1760 | ||
4b90c680 | 1761 | btrfs_extend_item(root, path, size); |
5d4f98a2 YZ |
1762 | |
1763 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1764 | refs = btrfs_extent_refs(leaf, ei); | |
1765 | refs += refs_to_add; | |
1766 | btrfs_set_extent_refs(leaf, ei, refs); | |
1767 | if (extent_op) | |
1768 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1769 | ||
1770 | ptr = (unsigned long)ei + item_offset; | |
1771 | end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]); | |
1772 | if (ptr < end - size) | |
1773 | memmove_extent_buffer(leaf, ptr + size, ptr, | |
1774 | end - size - ptr); | |
1775 | ||
1776 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1777 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
1778 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1779 | struct btrfs_extent_data_ref *dref; | |
1780 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1781 | btrfs_set_extent_data_ref_root(leaf, dref, root_objectid); | |
1782 | btrfs_set_extent_data_ref_objectid(leaf, dref, owner); | |
1783 | btrfs_set_extent_data_ref_offset(leaf, dref, offset); | |
1784 | btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add); | |
1785 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1786 | struct btrfs_shared_data_ref *sref; | |
1787 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1788 | btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add); | |
1789 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1790 | } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) { | |
1791 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1792 | } else { | |
1793 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
1794 | } | |
1795 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1796 | } |
1797 | ||
1798 | static int lookup_extent_backref(struct btrfs_trans_handle *trans, | |
1799 | struct btrfs_root *root, | |
1800 | struct btrfs_path *path, | |
1801 | struct btrfs_extent_inline_ref **ref_ret, | |
1802 | u64 bytenr, u64 num_bytes, u64 parent, | |
1803 | u64 root_objectid, u64 owner, u64 offset) | |
1804 | { | |
1805 | int ret; | |
1806 | ||
1807 | ret = lookup_inline_extent_backref(trans, root, path, ref_ret, | |
1808 | bytenr, num_bytes, parent, | |
1809 | root_objectid, owner, offset, 0); | |
1810 | if (ret != -ENOENT) | |
54aa1f4d | 1811 | return ret; |
5d4f98a2 | 1812 | |
b3b4aa74 | 1813 | btrfs_release_path(path); |
5d4f98a2 YZ |
1814 | *ref_ret = NULL; |
1815 | ||
1816 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1817 | ret = lookup_tree_block_ref(trans, root, path, bytenr, parent, | |
1818 | root_objectid); | |
1819 | } else { | |
1820 | ret = lookup_extent_data_ref(trans, root, path, bytenr, parent, | |
1821 | root_objectid, owner, offset); | |
b9473439 | 1822 | } |
5d4f98a2 YZ |
1823 | return ret; |
1824 | } | |
31840ae1 | 1825 | |
5d4f98a2 YZ |
1826 | /* |
1827 | * helper to update/remove inline back ref | |
1828 | */ | |
1829 | static noinline_for_stack | |
afe5fea7 | 1830 | void update_inline_extent_backref(struct btrfs_root *root, |
143bede5 JM |
1831 | struct btrfs_path *path, |
1832 | struct btrfs_extent_inline_ref *iref, | |
1833 | int refs_to_mod, | |
fcebe456 JB |
1834 | struct btrfs_delayed_extent_op *extent_op, |
1835 | int *last_ref) | |
5d4f98a2 YZ |
1836 | { |
1837 | struct extent_buffer *leaf; | |
1838 | struct btrfs_extent_item *ei; | |
1839 | struct btrfs_extent_data_ref *dref = NULL; | |
1840 | struct btrfs_shared_data_ref *sref = NULL; | |
1841 | unsigned long ptr; | |
1842 | unsigned long end; | |
1843 | u32 item_size; | |
1844 | int size; | |
1845 | int type; | |
5d4f98a2 YZ |
1846 | u64 refs; |
1847 | ||
1848 | leaf = path->nodes[0]; | |
1849 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1850 | refs = btrfs_extent_refs(leaf, ei); | |
1851 | WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0); | |
1852 | refs += refs_to_mod; | |
1853 | btrfs_set_extent_refs(leaf, ei, refs); | |
1854 | if (extent_op) | |
1855 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1856 | ||
1857 | type = btrfs_extent_inline_ref_type(leaf, iref); | |
1858 | ||
1859 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1860 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1861 | refs = btrfs_extent_data_ref_count(leaf, dref); | |
1862 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1863 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1864 | refs = btrfs_shared_data_ref_count(leaf, sref); | |
1865 | } else { | |
1866 | refs = 1; | |
1867 | BUG_ON(refs_to_mod != -1); | |
56bec294 | 1868 | } |
31840ae1 | 1869 | |
5d4f98a2 YZ |
1870 | BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod); |
1871 | refs += refs_to_mod; | |
1872 | ||
1873 | if (refs > 0) { | |
1874 | if (type == BTRFS_EXTENT_DATA_REF_KEY) | |
1875 | btrfs_set_extent_data_ref_count(leaf, dref, refs); | |
1876 | else | |
1877 | btrfs_set_shared_data_ref_count(leaf, sref, refs); | |
1878 | } else { | |
fcebe456 | 1879 | *last_ref = 1; |
5d4f98a2 YZ |
1880 | size = btrfs_extent_inline_ref_size(type); |
1881 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1882 | ptr = (unsigned long)iref; | |
1883 | end = (unsigned long)ei + item_size; | |
1884 | if (ptr + size < end) | |
1885 | memmove_extent_buffer(leaf, ptr, ptr + size, | |
1886 | end - ptr - size); | |
1887 | item_size -= size; | |
afe5fea7 | 1888 | btrfs_truncate_item(root, path, item_size, 1); |
5d4f98a2 YZ |
1889 | } |
1890 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1891 | } |
1892 | ||
1893 | static noinline_for_stack | |
1894 | int insert_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1895 | struct btrfs_root *root, | |
1896 | struct btrfs_path *path, | |
1897 | u64 bytenr, u64 num_bytes, u64 parent, | |
1898 | u64 root_objectid, u64 owner, | |
1899 | u64 offset, int refs_to_add, | |
1900 | struct btrfs_delayed_extent_op *extent_op) | |
1901 | { | |
1902 | struct btrfs_extent_inline_ref *iref; | |
1903 | int ret; | |
1904 | ||
1905 | ret = lookup_inline_extent_backref(trans, root, path, &iref, | |
1906 | bytenr, num_bytes, parent, | |
1907 | root_objectid, owner, offset, 1); | |
1908 | if (ret == 0) { | |
1909 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID); | |
afe5fea7 | 1910 | update_inline_extent_backref(root, path, iref, |
fcebe456 | 1911 | refs_to_add, extent_op, NULL); |
5d4f98a2 | 1912 | } else if (ret == -ENOENT) { |
fd279fae | 1913 | setup_inline_extent_backref(root, path, iref, parent, |
143bede5 JM |
1914 | root_objectid, owner, offset, |
1915 | refs_to_add, extent_op); | |
1916 | ret = 0; | |
771ed689 | 1917 | } |
5d4f98a2 YZ |
1918 | return ret; |
1919 | } | |
31840ae1 | 1920 | |
5d4f98a2 YZ |
1921 | static int insert_extent_backref(struct btrfs_trans_handle *trans, |
1922 | struct btrfs_root *root, | |
1923 | struct btrfs_path *path, | |
1924 | u64 bytenr, u64 parent, u64 root_objectid, | |
1925 | u64 owner, u64 offset, int refs_to_add) | |
1926 | { | |
1927 | int ret; | |
1928 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1929 | BUG_ON(refs_to_add != 1); | |
1930 | ret = insert_tree_block_ref(trans, root, path, bytenr, | |
1931 | parent, root_objectid); | |
1932 | } else { | |
1933 | ret = insert_extent_data_ref(trans, root, path, bytenr, | |
1934 | parent, root_objectid, | |
1935 | owner, offset, refs_to_add); | |
1936 | } | |
1937 | return ret; | |
1938 | } | |
56bec294 | 1939 | |
5d4f98a2 YZ |
1940 | static int remove_extent_backref(struct btrfs_trans_handle *trans, |
1941 | struct btrfs_root *root, | |
1942 | struct btrfs_path *path, | |
1943 | struct btrfs_extent_inline_ref *iref, | |
fcebe456 | 1944 | int refs_to_drop, int is_data, int *last_ref) |
5d4f98a2 | 1945 | { |
143bede5 | 1946 | int ret = 0; |
b9473439 | 1947 | |
5d4f98a2 YZ |
1948 | BUG_ON(!is_data && refs_to_drop != 1); |
1949 | if (iref) { | |
afe5fea7 | 1950 | update_inline_extent_backref(root, path, iref, |
fcebe456 | 1951 | -refs_to_drop, NULL, last_ref); |
5d4f98a2 | 1952 | } else if (is_data) { |
fcebe456 JB |
1953 | ret = remove_extent_data_ref(trans, root, path, refs_to_drop, |
1954 | last_ref); | |
5d4f98a2 | 1955 | } else { |
fcebe456 | 1956 | *last_ref = 1; |
5d4f98a2 YZ |
1957 | ret = btrfs_del_item(trans, root, path); |
1958 | } | |
1959 | return ret; | |
1960 | } | |
1961 | ||
86557861 | 1962 | #define in_range(b, first, len) ((b) >= (first) && (b) < (first) + (len)) |
d04c6b88 JM |
1963 | static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len, |
1964 | u64 *discarded_bytes) | |
5d4f98a2 | 1965 | { |
86557861 JM |
1966 | int j, ret = 0; |
1967 | u64 bytes_left, end; | |
4d89d377 | 1968 | u64 aligned_start = ALIGN(start, 1 << 9); |
d04c6b88 | 1969 | |
4d89d377 JM |
1970 | if (WARN_ON(start != aligned_start)) { |
1971 | len -= aligned_start - start; | |
1972 | len = round_down(len, 1 << 9); | |
1973 | start = aligned_start; | |
1974 | } | |
d04c6b88 | 1975 | |
4d89d377 | 1976 | *discarded_bytes = 0; |
86557861 JM |
1977 | |
1978 | if (!len) | |
1979 | return 0; | |
1980 | ||
1981 | end = start + len; | |
1982 | bytes_left = len; | |
1983 | ||
1984 | /* Skip any superblocks on this device. */ | |
1985 | for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) { | |
1986 | u64 sb_start = btrfs_sb_offset(j); | |
1987 | u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE; | |
1988 | u64 size = sb_start - start; | |
1989 | ||
1990 | if (!in_range(sb_start, start, bytes_left) && | |
1991 | !in_range(sb_end, start, bytes_left) && | |
1992 | !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE)) | |
1993 | continue; | |
1994 | ||
1995 | /* | |
1996 | * Superblock spans beginning of range. Adjust start and | |
1997 | * try again. | |
1998 | */ | |
1999 | if (sb_start <= start) { | |
2000 | start += sb_end - start; | |
2001 | if (start > end) { | |
2002 | bytes_left = 0; | |
2003 | break; | |
2004 | } | |
2005 | bytes_left = end - start; | |
2006 | continue; | |
2007 | } | |
2008 | ||
2009 | if (size) { | |
2010 | ret = blkdev_issue_discard(bdev, start >> 9, size >> 9, | |
2011 | GFP_NOFS, 0); | |
2012 | if (!ret) | |
2013 | *discarded_bytes += size; | |
2014 | else if (ret != -EOPNOTSUPP) | |
2015 | return ret; | |
2016 | } | |
2017 | ||
2018 | start = sb_end; | |
2019 | if (start > end) { | |
2020 | bytes_left = 0; | |
2021 | break; | |
2022 | } | |
2023 | bytes_left = end - start; | |
2024 | } | |
2025 | ||
2026 | if (bytes_left) { | |
2027 | ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9, | |
4d89d377 JM |
2028 | GFP_NOFS, 0); |
2029 | if (!ret) | |
86557861 | 2030 | *discarded_bytes += bytes_left; |
4d89d377 | 2031 | } |
d04c6b88 | 2032 | return ret; |
5d4f98a2 | 2033 | } |
5d4f98a2 | 2034 | |
1edb647b FM |
2035 | int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr, |
2036 | u64 num_bytes, u64 *actual_bytes) | |
5d4f98a2 | 2037 | { |
5d4f98a2 | 2038 | int ret; |
5378e607 | 2039 | u64 discarded_bytes = 0; |
a1d3c478 | 2040 | struct btrfs_bio *bbio = NULL; |
5d4f98a2 | 2041 | |
e244a0ae | 2042 | |
5d4f98a2 | 2043 | /* Tell the block device(s) that the sectors can be discarded */ |
3ec706c8 | 2044 | ret = btrfs_map_block(root->fs_info, REQ_DISCARD, |
a1d3c478 | 2045 | bytenr, &num_bytes, &bbio, 0); |
79787eaa | 2046 | /* Error condition is -ENOMEM */ |
5d4f98a2 | 2047 | if (!ret) { |
a1d3c478 | 2048 | struct btrfs_bio_stripe *stripe = bbio->stripes; |
5d4f98a2 YZ |
2049 | int i; |
2050 | ||
5d4f98a2 | 2051 | |
a1d3c478 | 2052 | for (i = 0; i < bbio->num_stripes; i++, stripe++) { |
d04c6b88 | 2053 | u64 bytes; |
d5e2003c JB |
2054 | if (!stripe->dev->can_discard) |
2055 | continue; | |
2056 | ||
5378e607 LD |
2057 | ret = btrfs_issue_discard(stripe->dev->bdev, |
2058 | stripe->physical, | |
d04c6b88 JM |
2059 | stripe->length, |
2060 | &bytes); | |
5378e607 | 2061 | if (!ret) |
d04c6b88 | 2062 | discarded_bytes += bytes; |
5378e607 | 2063 | else if (ret != -EOPNOTSUPP) |
79787eaa | 2064 | break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */ |
d5e2003c JB |
2065 | |
2066 | /* | |
2067 | * Just in case we get back EOPNOTSUPP for some reason, | |
2068 | * just ignore the return value so we don't screw up | |
2069 | * people calling discard_extent. | |
2070 | */ | |
2071 | ret = 0; | |
5d4f98a2 | 2072 | } |
6e9606d2 | 2073 | btrfs_put_bbio(bbio); |
5d4f98a2 | 2074 | } |
5378e607 LD |
2075 | |
2076 | if (actual_bytes) | |
2077 | *actual_bytes = discarded_bytes; | |
2078 | ||
5d4f98a2 | 2079 | |
53b381b3 DW |
2080 | if (ret == -EOPNOTSUPP) |
2081 | ret = 0; | |
5d4f98a2 | 2082 | return ret; |
5d4f98a2 YZ |
2083 | } |
2084 | ||
79787eaa | 2085 | /* Can return -ENOMEM */ |
5d4f98a2 YZ |
2086 | int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
2087 | struct btrfs_root *root, | |
2088 | u64 bytenr, u64 num_bytes, u64 parent, | |
b06c4bf5 | 2089 | u64 root_objectid, u64 owner, u64 offset) |
5d4f98a2 YZ |
2090 | { |
2091 | int ret; | |
66d7e7f0 AJ |
2092 | struct btrfs_fs_info *fs_info = root->fs_info; |
2093 | ||
5d4f98a2 YZ |
2094 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID && |
2095 | root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
2096 | ||
2097 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
66d7e7f0 AJ |
2098 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr, |
2099 | num_bytes, | |
5d4f98a2 | 2100 | parent, root_objectid, (int)owner, |
b06c4bf5 | 2101 | BTRFS_ADD_DELAYED_REF, NULL); |
5d4f98a2 | 2102 | } else { |
66d7e7f0 | 2103 | ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr, |
5846a3c2 QW |
2104 | num_bytes, parent, root_objectid, |
2105 | owner, offset, 0, | |
b06c4bf5 | 2106 | BTRFS_ADD_DELAYED_REF, NULL); |
5d4f98a2 YZ |
2107 | } |
2108 | return ret; | |
2109 | } | |
2110 | ||
2111 | static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, | |
2112 | struct btrfs_root *root, | |
c682f9b3 | 2113 | struct btrfs_delayed_ref_node *node, |
5d4f98a2 YZ |
2114 | u64 parent, u64 root_objectid, |
2115 | u64 owner, u64 offset, int refs_to_add, | |
2116 | struct btrfs_delayed_extent_op *extent_op) | |
2117 | { | |
fcebe456 | 2118 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 YZ |
2119 | struct btrfs_path *path; |
2120 | struct extent_buffer *leaf; | |
2121 | struct btrfs_extent_item *item; | |
fcebe456 | 2122 | struct btrfs_key key; |
c682f9b3 QW |
2123 | u64 bytenr = node->bytenr; |
2124 | u64 num_bytes = node->num_bytes; | |
5d4f98a2 YZ |
2125 | u64 refs; |
2126 | int ret; | |
5d4f98a2 YZ |
2127 | |
2128 | path = btrfs_alloc_path(); | |
2129 | if (!path) | |
2130 | return -ENOMEM; | |
2131 | ||
2132 | path->reada = 1; | |
2133 | path->leave_spinning = 1; | |
2134 | /* this will setup the path even if it fails to insert the back ref */ | |
fcebe456 JB |
2135 | ret = insert_inline_extent_backref(trans, fs_info->extent_root, path, |
2136 | bytenr, num_bytes, parent, | |
5d4f98a2 YZ |
2137 | root_objectid, owner, offset, |
2138 | refs_to_add, extent_op); | |
0ed4792a | 2139 | if ((ret < 0 && ret != -EAGAIN) || !ret) |
5d4f98a2 | 2140 | goto out; |
fcebe456 JB |
2141 | |
2142 | /* | |
2143 | * Ok we had -EAGAIN which means we didn't have space to insert and | |
2144 | * inline extent ref, so just update the reference count and add a | |
2145 | * normal backref. | |
2146 | */ | |
5d4f98a2 | 2147 | leaf = path->nodes[0]; |
fcebe456 | 2148 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
5d4f98a2 YZ |
2149 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
2150 | refs = btrfs_extent_refs(leaf, item); | |
2151 | btrfs_set_extent_refs(leaf, item, refs + refs_to_add); | |
2152 | if (extent_op) | |
2153 | __run_delayed_extent_op(extent_op, leaf, item); | |
56bec294 | 2154 | |
5d4f98a2 | 2155 | btrfs_mark_buffer_dirty(leaf); |
b3b4aa74 | 2156 | btrfs_release_path(path); |
56bec294 CM |
2157 | |
2158 | path->reada = 1; | |
b9473439 | 2159 | path->leave_spinning = 1; |
56bec294 CM |
2160 | /* now insert the actual backref */ |
2161 | ret = insert_extent_backref(trans, root->fs_info->extent_root, | |
5d4f98a2 YZ |
2162 | path, bytenr, parent, root_objectid, |
2163 | owner, offset, refs_to_add); | |
79787eaa JM |
2164 | if (ret) |
2165 | btrfs_abort_transaction(trans, root, ret); | |
5d4f98a2 | 2166 | out: |
56bec294 | 2167 | btrfs_free_path(path); |
30d133fc | 2168 | return ret; |
56bec294 CM |
2169 | } |
2170 | ||
5d4f98a2 YZ |
2171 | static int run_delayed_data_ref(struct btrfs_trans_handle *trans, |
2172 | struct btrfs_root *root, | |
2173 | struct btrfs_delayed_ref_node *node, | |
2174 | struct btrfs_delayed_extent_op *extent_op, | |
2175 | int insert_reserved) | |
56bec294 | 2176 | { |
5d4f98a2 YZ |
2177 | int ret = 0; |
2178 | struct btrfs_delayed_data_ref *ref; | |
2179 | struct btrfs_key ins; | |
2180 | u64 parent = 0; | |
2181 | u64 ref_root = 0; | |
2182 | u64 flags = 0; | |
2183 | ||
2184 | ins.objectid = node->bytenr; | |
2185 | ins.offset = node->num_bytes; | |
2186 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
2187 | ||
2188 | ref = btrfs_delayed_node_to_data_ref(node); | |
599c75ec LB |
2189 | trace_run_delayed_data_ref(node, ref, node->action); |
2190 | ||
5d4f98a2 YZ |
2191 | if (node->type == BTRFS_SHARED_DATA_REF_KEY) |
2192 | parent = ref->parent; | |
fcebe456 | 2193 | ref_root = ref->root; |
5d4f98a2 YZ |
2194 | |
2195 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
3173a18f | 2196 | if (extent_op) |
5d4f98a2 | 2197 | flags |= extent_op->flags_to_set; |
5d4f98a2 YZ |
2198 | ret = alloc_reserved_file_extent(trans, root, |
2199 | parent, ref_root, flags, | |
2200 | ref->objectid, ref->offset, | |
2201 | &ins, node->ref_mod); | |
5d4f98a2 | 2202 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
c682f9b3 | 2203 | ret = __btrfs_inc_extent_ref(trans, root, node, parent, |
5d4f98a2 YZ |
2204 | ref_root, ref->objectid, |
2205 | ref->offset, node->ref_mod, | |
c682f9b3 | 2206 | extent_op); |
5d4f98a2 | 2207 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
c682f9b3 | 2208 | ret = __btrfs_free_extent(trans, root, node, parent, |
5d4f98a2 YZ |
2209 | ref_root, ref->objectid, |
2210 | ref->offset, node->ref_mod, | |
c682f9b3 | 2211 | extent_op); |
5d4f98a2 YZ |
2212 | } else { |
2213 | BUG(); | |
2214 | } | |
2215 | return ret; | |
2216 | } | |
2217 | ||
2218 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
2219 | struct extent_buffer *leaf, | |
2220 | struct btrfs_extent_item *ei) | |
2221 | { | |
2222 | u64 flags = btrfs_extent_flags(leaf, ei); | |
2223 | if (extent_op->update_flags) { | |
2224 | flags |= extent_op->flags_to_set; | |
2225 | btrfs_set_extent_flags(leaf, ei, flags); | |
2226 | } | |
2227 | ||
2228 | if (extent_op->update_key) { | |
2229 | struct btrfs_tree_block_info *bi; | |
2230 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)); | |
2231 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
2232 | btrfs_set_tree_block_key(leaf, bi, &extent_op->key); | |
2233 | } | |
2234 | } | |
2235 | ||
2236 | static int run_delayed_extent_op(struct btrfs_trans_handle *trans, | |
2237 | struct btrfs_root *root, | |
2238 | struct btrfs_delayed_ref_node *node, | |
2239 | struct btrfs_delayed_extent_op *extent_op) | |
2240 | { | |
2241 | struct btrfs_key key; | |
2242 | struct btrfs_path *path; | |
2243 | struct btrfs_extent_item *ei; | |
2244 | struct extent_buffer *leaf; | |
2245 | u32 item_size; | |
56bec294 | 2246 | int ret; |
5d4f98a2 | 2247 | int err = 0; |
b1c79e09 | 2248 | int metadata = !extent_op->is_data; |
5d4f98a2 | 2249 | |
79787eaa JM |
2250 | if (trans->aborted) |
2251 | return 0; | |
2252 | ||
3173a18f JB |
2253 | if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) |
2254 | metadata = 0; | |
2255 | ||
5d4f98a2 YZ |
2256 | path = btrfs_alloc_path(); |
2257 | if (!path) | |
2258 | return -ENOMEM; | |
2259 | ||
2260 | key.objectid = node->bytenr; | |
5d4f98a2 | 2261 | |
3173a18f | 2262 | if (metadata) { |
3173a18f | 2263 | key.type = BTRFS_METADATA_ITEM_KEY; |
b1c79e09 | 2264 | key.offset = extent_op->level; |
3173a18f JB |
2265 | } else { |
2266 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
2267 | key.offset = node->num_bytes; | |
2268 | } | |
2269 | ||
2270 | again: | |
5d4f98a2 YZ |
2271 | path->reada = 1; |
2272 | path->leave_spinning = 1; | |
2273 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, | |
2274 | path, 0, 1); | |
2275 | if (ret < 0) { | |
2276 | err = ret; | |
2277 | goto out; | |
2278 | } | |
2279 | if (ret > 0) { | |
3173a18f | 2280 | if (metadata) { |
55994887 FDBM |
2281 | if (path->slots[0] > 0) { |
2282 | path->slots[0]--; | |
2283 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
2284 | path->slots[0]); | |
2285 | if (key.objectid == node->bytenr && | |
2286 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
2287 | key.offset == node->num_bytes) | |
2288 | ret = 0; | |
2289 | } | |
2290 | if (ret > 0) { | |
2291 | btrfs_release_path(path); | |
2292 | metadata = 0; | |
3173a18f | 2293 | |
55994887 FDBM |
2294 | key.objectid = node->bytenr; |
2295 | key.offset = node->num_bytes; | |
2296 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
2297 | goto again; | |
2298 | } | |
2299 | } else { | |
2300 | err = -EIO; | |
2301 | goto out; | |
3173a18f | 2302 | } |
5d4f98a2 YZ |
2303 | } |
2304 | ||
2305 | leaf = path->nodes[0]; | |
2306 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2307 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
2308 | if (item_size < sizeof(*ei)) { | |
2309 | ret = convert_extent_item_v0(trans, root->fs_info->extent_root, | |
2310 | path, (u64)-1, 0); | |
2311 | if (ret < 0) { | |
2312 | err = ret; | |
2313 | goto out; | |
2314 | } | |
2315 | leaf = path->nodes[0]; | |
2316 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2317 | } | |
2318 | #endif | |
2319 | BUG_ON(item_size < sizeof(*ei)); | |
2320 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
2321 | __run_delayed_extent_op(extent_op, leaf, ei); | |
56bec294 | 2322 | |
5d4f98a2 YZ |
2323 | btrfs_mark_buffer_dirty(leaf); |
2324 | out: | |
2325 | btrfs_free_path(path); | |
2326 | return err; | |
56bec294 CM |
2327 | } |
2328 | ||
5d4f98a2 YZ |
2329 | static int run_delayed_tree_ref(struct btrfs_trans_handle *trans, |
2330 | struct btrfs_root *root, | |
2331 | struct btrfs_delayed_ref_node *node, | |
2332 | struct btrfs_delayed_extent_op *extent_op, | |
2333 | int insert_reserved) | |
56bec294 CM |
2334 | { |
2335 | int ret = 0; | |
5d4f98a2 YZ |
2336 | struct btrfs_delayed_tree_ref *ref; |
2337 | struct btrfs_key ins; | |
2338 | u64 parent = 0; | |
2339 | u64 ref_root = 0; | |
3173a18f JB |
2340 | bool skinny_metadata = btrfs_fs_incompat(root->fs_info, |
2341 | SKINNY_METADATA); | |
56bec294 | 2342 | |
5d4f98a2 | 2343 | ref = btrfs_delayed_node_to_tree_ref(node); |
599c75ec LB |
2344 | trace_run_delayed_tree_ref(node, ref, node->action); |
2345 | ||
5d4f98a2 YZ |
2346 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) |
2347 | parent = ref->parent; | |
fcebe456 | 2348 | ref_root = ref->root; |
5d4f98a2 | 2349 | |
3173a18f JB |
2350 | ins.objectid = node->bytenr; |
2351 | if (skinny_metadata) { | |
2352 | ins.offset = ref->level; | |
2353 | ins.type = BTRFS_METADATA_ITEM_KEY; | |
2354 | } else { | |
2355 | ins.offset = node->num_bytes; | |
2356 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
2357 | } | |
2358 | ||
5d4f98a2 YZ |
2359 | BUG_ON(node->ref_mod != 1); |
2360 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
3173a18f | 2361 | BUG_ON(!extent_op || !extent_op->update_flags); |
5d4f98a2 YZ |
2362 | ret = alloc_reserved_tree_block(trans, root, |
2363 | parent, ref_root, | |
2364 | extent_op->flags_to_set, | |
2365 | &extent_op->key, | |
b06c4bf5 | 2366 | ref->level, &ins); |
5d4f98a2 | 2367 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
c682f9b3 QW |
2368 | ret = __btrfs_inc_extent_ref(trans, root, node, |
2369 | parent, ref_root, | |
2370 | ref->level, 0, 1, | |
fcebe456 | 2371 | extent_op); |
5d4f98a2 | 2372 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
c682f9b3 QW |
2373 | ret = __btrfs_free_extent(trans, root, node, |
2374 | parent, ref_root, | |
2375 | ref->level, 0, 1, extent_op); | |
5d4f98a2 YZ |
2376 | } else { |
2377 | BUG(); | |
2378 | } | |
56bec294 CM |
2379 | return ret; |
2380 | } | |
2381 | ||
2382 | /* helper function to actually process a single delayed ref entry */ | |
5d4f98a2 YZ |
2383 | static int run_one_delayed_ref(struct btrfs_trans_handle *trans, |
2384 | struct btrfs_root *root, | |
2385 | struct btrfs_delayed_ref_node *node, | |
2386 | struct btrfs_delayed_extent_op *extent_op, | |
2387 | int insert_reserved) | |
56bec294 | 2388 | { |
79787eaa JM |
2389 | int ret = 0; |
2390 | ||
857cc2fc JB |
2391 | if (trans->aborted) { |
2392 | if (insert_reserved) | |
2393 | btrfs_pin_extent(root, node->bytenr, | |
2394 | node->num_bytes, 1); | |
79787eaa | 2395 | return 0; |
857cc2fc | 2396 | } |
79787eaa | 2397 | |
5d4f98a2 | 2398 | if (btrfs_delayed_ref_is_head(node)) { |
56bec294 CM |
2399 | struct btrfs_delayed_ref_head *head; |
2400 | /* | |
2401 | * we've hit the end of the chain and we were supposed | |
2402 | * to insert this extent into the tree. But, it got | |
2403 | * deleted before we ever needed to insert it, so all | |
2404 | * we have to do is clean up the accounting | |
2405 | */ | |
5d4f98a2 YZ |
2406 | BUG_ON(extent_op); |
2407 | head = btrfs_delayed_node_to_head(node); | |
599c75ec LB |
2408 | trace_run_delayed_ref_head(node, head, node->action); |
2409 | ||
56bec294 | 2410 | if (insert_reserved) { |
f0486c68 YZ |
2411 | btrfs_pin_extent(root, node->bytenr, |
2412 | node->num_bytes, 1); | |
5d4f98a2 YZ |
2413 | if (head->is_data) { |
2414 | ret = btrfs_del_csums(trans, root, | |
2415 | node->bytenr, | |
2416 | node->num_bytes); | |
5d4f98a2 | 2417 | } |
56bec294 | 2418 | } |
297d750b QW |
2419 | |
2420 | /* Also free its reserved qgroup space */ | |
2421 | btrfs_qgroup_free_delayed_ref(root->fs_info, | |
2422 | head->qgroup_ref_root, | |
2423 | head->qgroup_reserved); | |
79787eaa | 2424 | return ret; |
56bec294 CM |
2425 | } |
2426 | ||
5d4f98a2 YZ |
2427 | if (node->type == BTRFS_TREE_BLOCK_REF_KEY || |
2428 | node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
2429 | ret = run_delayed_tree_ref(trans, root, node, extent_op, | |
2430 | insert_reserved); | |
2431 | else if (node->type == BTRFS_EXTENT_DATA_REF_KEY || | |
2432 | node->type == BTRFS_SHARED_DATA_REF_KEY) | |
2433 | ret = run_delayed_data_ref(trans, root, node, extent_op, | |
2434 | insert_reserved); | |
2435 | else | |
2436 | BUG(); | |
2437 | return ret; | |
56bec294 CM |
2438 | } |
2439 | ||
c6fc2454 | 2440 | static inline struct btrfs_delayed_ref_node * |
56bec294 CM |
2441 | select_delayed_ref(struct btrfs_delayed_ref_head *head) |
2442 | { | |
cffc3374 FM |
2443 | struct btrfs_delayed_ref_node *ref; |
2444 | ||
c6fc2454 QW |
2445 | if (list_empty(&head->ref_list)) |
2446 | return NULL; | |
d7df2c79 | 2447 | |
cffc3374 FM |
2448 | /* |
2449 | * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first. | |
2450 | * This is to prevent a ref count from going down to zero, which deletes | |
2451 | * the extent item from the extent tree, when there still are references | |
2452 | * to add, which would fail because they would not find the extent item. | |
2453 | */ | |
2454 | list_for_each_entry(ref, &head->ref_list, list) { | |
2455 | if (ref->action == BTRFS_ADD_DELAYED_REF) | |
2456 | return ref; | |
2457 | } | |
2458 | ||
c6fc2454 QW |
2459 | return list_entry(head->ref_list.next, struct btrfs_delayed_ref_node, |
2460 | list); | |
56bec294 CM |
2461 | } |
2462 | ||
79787eaa JM |
2463 | /* |
2464 | * Returns 0 on success or if called with an already aborted transaction. | |
2465 | * Returns -ENOMEM or -EIO on failure and will abort the transaction. | |
2466 | */ | |
d7df2c79 JB |
2467 | static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, |
2468 | struct btrfs_root *root, | |
2469 | unsigned long nr) | |
56bec294 | 2470 | { |
56bec294 CM |
2471 | struct btrfs_delayed_ref_root *delayed_refs; |
2472 | struct btrfs_delayed_ref_node *ref; | |
2473 | struct btrfs_delayed_ref_head *locked_ref = NULL; | |
5d4f98a2 | 2474 | struct btrfs_delayed_extent_op *extent_op; |
097b8a7c | 2475 | struct btrfs_fs_info *fs_info = root->fs_info; |
0a2b2a84 | 2476 | ktime_t start = ktime_get(); |
56bec294 | 2477 | int ret; |
d7df2c79 | 2478 | unsigned long count = 0; |
0a2b2a84 | 2479 | unsigned long actual_count = 0; |
56bec294 | 2480 | int must_insert_reserved = 0; |
56bec294 CM |
2481 | |
2482 | delayed_refs = &trans->transaction->delayed_refs; | |
56bec294 CM |
2483 | while (1) { |
2484 | if (!locked_ref) { | |
d7df2c79 | 2485 | if (count >= nr) |
56bec294 | 2486 | break; |
56bec294 | 2487 | |
d7df2c79 JB |
2488 | spin_lock(&delayed_refs->lock); |
2489 | locked_ref = btrfs_select_ref_head(trans); | |
2490 | if (!locked_ref) { | |
2491 | spin_unlock(&delayed_refs->lock); | |
2492 | break; | |
2493 | } | |
c3e69d58 CM |
2494 | |
2495 | /* grab the lock that says we are going to process | |
2496 | * all the refs for this head */ | |
2497 | ret = btrfs_delayed_ref_lock(trans, locked_ref); | |
d7df2c79 | 2498 | spin_unlock(&delayed_refs->lock); |
c3e69d58 CM |
2499 | /* |
2500 | * we may have dropped the spin lock to get the head | |
2501 | * mutex lock, and that might have given someone else | |
2502 | * time to free the head. If that's true, it has been | |
2503 | * removed from our list and we can move on. | |
2504 | */ | |
2505 | if (ret == -EAGAIN) { | |
2506 | locked_ref = NULL; | |
2507 | count++; | |
2508 | continue; | |
56bec294 CM |
2509 | } |
2510 | } | |
a28ec197 | 2511 | |
2c3cf7d5 FM |
2512 | /* |
2513 | * We need to try and merge add/drops of the same ref since we | |
2514 | * can run into issues with relocate dropping the implicit ref | |
2515 | * and then it being added back again before the drop can | |
2516 | * finish. If we merged anything we need to re-loop so we can | |
2517 | * get a good ref. | |
2518 | * Or we can get node references of the same type that weren't | |
2519 | * merged when created due to bumps in the tree mod seq, and | |
2520 | * we need to merge them to prevent adding an inline extent | |
2521 | * backref before dropping it (triggering a BUG_ON at | |
2522 | * insert_inline_extent_backref()). | |
2523 | */ | |
d7df2c79 | 2524 | spin_lock(&locked_ref->lock); |
2c3cf7d5 FM |
2525 | btrfs_merge_delayed_refs(trans, fs_info, delayed_refs, |
2526 | locked_ref); | |
ae1e206b | 2527 | |
d1270cd9 AJ |
2528 | /* |
2529 | * locked_ref is the head node, so we have to go one | |
2530 | * node back for any delayed ref updates | |
2531 | */ | |
2532 | ref = select_delayed_ref(locked_ref); | |
2533 | ||
2534 | if (ref && ref->seq && | |
097b8a7c | 2535 | btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) { |
d7df2c79 | 2536 | spin_unlock(&locked_ref->lock); |
093486c4 | 2537 | btrfs_delayed_ref_unlock(locked_ref); |
d7df2c79 JB |
2538 | spin_lock(&delayed_refs->lock); |
2539 | locked_ref->processing = 0; | |
d1270cd9 AJ |
2540 | delayed_refs->num_heads_ready++; |
2541 | spin_unlock(&delayed_refs->lock); | |
d7df2c79 | 2542 | locked_ref = NULL; |
d1270cd9 | 2543 | cond_resched(); |
27a377db | 2544 | count++; |
d1270cd9 AJ |
2545 | continue; |
2546 | } | |
2547 | ||
56bec294 CM |
2548 | /* |
2549 | * record the must insert reserved flag before we | |
2550 | * drop the spin lock. | |
2551 | */ | |
2552 | must_insert_reserved = locked_ref->must_insert_reserved; | |
2553 | locked_ref->must_insert_reserved = 0; | |
7bb86316 | 2554 | |
5d4f98a2 YZ |
2555 | extent_op = locked_ref->extent_op; |
2556 | locked_ref->extent_op = NULL; | |
2557 | ||
56bec294 | 2558 | if (!ref) { |
d7df2c79 JB |
2559 | |
2560 | ||
56bec294 CM |
2561 | /* All delayed refs have been processed, Go ahead |
2562 | * and send the head node to run_one_delayed_ref, | |
2563 | * so that any accounting fixes can happen | |
2564 | */ | |
2565 | ref = &locked_ref->node; | |
5d4f98a2 YZ |
2566 | |
2567 | if (extent_op && must_insert_reserved) { | |
78a6184a | 2568 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 YZ |
2569 | extent_op = NULL; |
2570 | } | |
2571 | ||
2572 | if (extent_op) { | |
d7df2c79 | 2573 | spin_unlock(&locked_ref->lock); |
5d4f98a2 YZ |
2574 | ret = run_delayed_extent_op(trans, root, |
2575 | ref, extent_op); | |
78a6184a | 2576 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 | 2577 | |
79787eaa | 2578 | if (ret) { |
857cc2fc JB |
2579 | /* |
2580 | * Need to reset must_insert_reserved if | |
2581 | * there was an error so the abort stuff | |
2582 | * can cleanup the reserved space | |
2583 | * properly. | |
2584 | */ | |
2585 | if (must_insert_reserved) | |
2586 | locked_ref->must_insert_reserved = 1; | |
d7df2c79 | 2587 | locked_ref->processing = 0; |
c2cf52eb | 2588 | btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret); |
093486c4 | 2589 | btrfs_delayed_ref_unlock(locked_ref); |
79787eaa JM |
2590 | return ret; |
2591 | } | |
d7df2c79 | 2592 | continue; |
5d4f98a2 | 2593 | } |
02217ed2 | 2594 | |
d7df2c79 JB |
2595 | /* |
2596 | * Need to drop our head ref lock and re-aqcuire the | |
2597 | * delayed ref lock and then re-check to make sure | |
2598 | * nobody got added. | |
2599 | */ | |
2600 | spin_unlock(&locked_ref->lock); | |
2601 | spin_lock(&delayed_refs->lock); | |
2602 | spin_lock(&locked_ref->lock); | |
c6fc2454 | 2603 | if (!list_empty(&locked_ref->ref_list) || |
573a0755 | 2604 | locked_ref->extent_op) { |
d7df2c79 JB |
2605 | spin_unlock(&locked_ref->lock); |
2606 | spin_unlock(&delayed_refs->lock); | |
2607 | continue; | |
2608 | } | |
2609 | ref->in_tree = 0; | |
2610 | delayed_refs->num_heads--; | |
c46effa6 LB |
2611 | rb_erase(&locked_ref->href_node, |
2612 | &delayed_refs->href_root); | |
d7df2c79 JB |
2613 | spin_unlock(&delayed_refs->lock); |
2614 | } else { | |
0a2b2a84 | 2615 | actual_count++; |
d7df2c79 | 2616 | ref->in_tree = 0; |
c6fc2454 | 2617 | list_del(&ref->list); |
c46effa6 | 2618 | } |
d7df2c79 JB |
2619 | atomic_dec(&delayed_refs->num_entries); |
2620 | ||
093486c4 | 2621 | if (!btrfs_delayed_ref_is_head(ref)) { |
22cd2e7d AJ |
2622 | /* |
2623 | * when we play the delayed ref, also correct the | |
2624 | * ref_mod on head | |
2625 | */ | |
2626 | switch (ref->action) { | |
2627 | case BTRFS_ADD_DELAYED_REF: | |
2628 | case BTRFS_ADD_DELAYED_EXTENT: | |
2629 | locked_ref->node.ref_mod -= ref->ref_mod; | |
2630 | break; | |
2631 | case BTRFS_DROP_DELAYED_REF: | |
2632 | locked_ref->node.ref_mod += ref->ref_mod; | |
2633 | break; | |
2634 | default: | |
2635 | WARN_ON(1); | |
2636 | } | |
2637 | } | |
d7df2c79 | 2638 | spin_unlock(&locked_ref->lock); |
925baedd | 2639 | |
5d4f98a2 | 2640 | ret = run_one_delayed_ref(trans, root, ref, extent_op, |
56bec294 | 2641 | must_insert_reserved); |
eb099670 | 2642 | |
78a6184a | 2643 | btrfs_free_delayed_extent_op(extent_op); |
79787eaa | 2644 | if (ret) { |
d7df2c79 | 2645 | locked_ref->processing = 0; |
093486c4 MX |
2646 | btrfs_delayed_ref_unlock(locked_ref); |
2647 | btrfs_put_delayed_ref(ref); | |
c2cf52eb | 2648 | btrfs_debug(fs_info, "run_one_delayed_ref returned %d", ret); |
79787eaa JM |
2649 | return ret; |
2650 | } | |
2651 | ||
093486c4 MX |
2652 | /* |
2653 | * If this node is a head, that means all the refs in this head | |
2654 | * have been dealt with, and we will pick the next head to deal | |
2655 | * with, so we must unlock the head and drop it from the cluster | |
2656 | * list before we release it. | |
2657 | */ | |
2658 | if (btrfs_delayed_ref_is_head(ref)) { | |
1262133b JB |
2659 | if (locked_ref->is_data && |
2660 | locked_ref->total_ref_mod < 0) { | |
2661 | spin_lock(&delayed_refs->lock); | |
2662 | delayed_refs->pending_csums -= ref->num_bytes; | |
2663 | spin_unlock(&delayed_refs->lock); | |
2664 | } | |
093486c4 MX |
2665 | btrfs_delayed_ref_unlock(locked_ref); |
2666 | locked_ref = NULL; | |
2667 | } | |
2668 | btrfs_put_delayed_ref(ref); | |
2669 | count++; | |
c3e69d58 | 2670 | cond_resched(); |
c3e69d58 | 2671 | } |
0a2b2a84 JB |
2672 | |
2673 | /* | |
2674 | * We don't want to include ref heads since we can have empty ref heads | |
2675 | * and those will drastically skew our runtime down since we just do | |
2676 | * accounting, no actual extent tree updates. | |
2677 | */ | |
2678 | if (actual_count > 0) { | |
2679 | u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start)); | |
2680 | u64 avg; | |
2681 | ||
2682 | /* | |
2683 | * We weigh the current average higher than our current runtime | |
2684 | * to avoid large swings in the average. | |
2685 | */ | |
2686 | spin_lock(&delayed_refs->lock); | |
2687 | avg = fs_info->avg_delayed_ref_runtime * 3 + runtime; | |
f8c269d7 | 2688 | fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */ |
0a2b2a84 JB |
2689 | spin_unlock(&delayed_refs->lock); |
2690 | } | |
d7df2c79 | 2691 | return 0; |
c3e69d58 CM |
2692 | } |
2693 | ||
709c0486 AJ |
2694 | #ifdef SCRAMBLE_DELAYED_REFS |
2695 | /* | |
2696 | * Normally delayed refs get processed in ascending bytenr order. This | |
2697 | * correlates in most cases to the order added. To expose dependencies on this | |
2698 | * order, we start to process the tree in the middle instead of the beginning | |
2699 | */ | |
2700 | static u64 find_middle(struct rb_root *root) | |
2701 | { | |
2702 | struct rb_node *n = root->rb_node; | |
2703 | struct btrfs_delayed_ref_node *entry; | |
2704 | int alt = 1; | |
2705 | u64 middle; | |
2706 | u64 first = 0, last = 0; | |
2707 | ||
2708 | n = rb_first(root); | |
2709 | if (n) { | |
2710 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2711 | first = entry->bytenr; | |
2712 | } | |
2713 | n = rb_last(root); | |
2714 | if (n) { | |
2715 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2716 | last = entry->bytenr; | |
2717 | } | |
2718 | n = root->rb_node; | |
2719 | ||
2720 | while (n) { | |
2721 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2722 | WARN_ON(!entry->in_tree); | |
2723 | ||
2724 | middle = entry->bytenr; | |
2725 | ||
2726 | if (alt) | |
2727 | n = n->rb_left; | |
2728 | else | |
2729 | n = n->rb_right; | |
2730 | ||
2731 | alt = 1 - alt; | |
2732 | } | |
2733 | return middle; | |
2734 | } | |
2735 | #endif | |
2736 | ||
1be41b78 JB |
2737 | static inline u64 heads_to_leaves(struct btrfs_root *root, u64 heads) |
2738 | { | |
2739 | u64 num_bytes; | |
2740 | ||
2741 | num_bytes = heads * (sizeof(struct btrfs_extent_item) + | |
2742 | sizeof(struct btrfs_extent_inline_ref)); | |
2743 | if (!btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) | |
2744 | num_bytes += heads * sizeof(struct btrfs_tree_block_info); | |
2745 | ||
2746 | /* | |
2747 | * We don't ever fill up leaves all the way so multiply by 2 just to be | |
2748 | * closer to what we're really going to want to ouse. | |
2749 | */ | |
f8c269d7 | 2750 | return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(root)); |
1be41b78 JB |
2751 | } |
2752 | ||
1262133b JB |
2753 | /* |
2754 | * Takes the number of bytes to be csumm'ed and figures out how many leaves it | |
2755 | * would require to store the csums for that many bytes. | |
2756 | */ | |
28f75a0e | 2757 | u64 btrfs_csum_bytes_to_leaves(struct btrfs_root *root, u64 csum_bytes) |
1262133b JB |
2758 | { |
2759 | u64 csum_size; | |
2760 | u64 num_csums_per_leaf; | |
2761 | u64 num_csums; | |
2762 | ||
2763 | csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item); | |
2764 | num_csums_per_leaf = div64_u64(csum_size, | |
2765 | (u64)btrfs_super_csum_size(root->fs_info->super_copy)); | |
2766 | num_csums = div64_u64(csum_bytes, root->sectorsize); | |
2767 | num_csums += num_csums_per_leaf - 1; | |
2768 | num_csums = div64_u64(num_csums, num_csums_per_leaf); | |
2769 | return num_csums; | |
2770 | } | |
2771 | ||
0a2b2a84 | 2772 | int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans, |
1be41b78 JB |
2773 | struct btrfs_root *root) |
2774 | { | |
2775 | struct btrfs_block_rsv *global_rsv; | |
2776 | u64 num_heads = trans->transaction->delayed_refs.num_heads_ready; | |
1262133b | 2777 | u64 csum_bytes = trans->transaction->delayed_refs.pending_csums; |
cb723e49 JB |
2778 | u64 num_dirty_bgs = trans->transaction->num_dirty_bgs; |
2779 | u64 num_bytes, num_dirty_bgs_bytes; | |
1be41b78 JB |
2780 | int ret = 0; |
2781 | ||
2782 | num_bytes = btrfs_calc_trans_metadata_size(root, 1); | |
2783 | num_heads = heads_to_leaves(root, num_heads); | |
2784 | if (num_heads > 1) | |
707e8a07 | 2785 | num_bytes += (num_heads - 1) * root->nodesize; |
1be41b78 | 2786 | num_bytes <<= 1; |
28f75a0e | 2787 | num_bytes += btrfs_csum_bytes_to_leaves(root, csum_bytes) * root->nodesize; |
cb723e49 JB |
2788 | num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(root, |
2789 | num_dirty_bgs); | |
1be41b78 JB |
2790 | global_rsv = &root->fs_info->global_block_rsv; |
2791 | ||
2792 | /* | |
2793 | * If we can't allocate any more chunks lets make sure we have _lots_ of | |
2794 | * wiggle room since running delayed refs can create more delayed refs. | |
2795 | */ | |
cb723e49 JB |
2796 | if (global_rsv->space_info->full) { |
2797 | num_dirty_bgs_bytes <<= 1; | |
1be41b78 | 2798 | num_bytes <<= 1; |
cb723e49 | 2799 | } |
1be41b78 JB |
2800 | |
2801 | spin_lock(&global_rsv->lock); | |
cb723e49 | 2802 | if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes) |
1be41b78 JB |
2803 | ret = 1; |
2804 | spin_unlock(&global_rsv->lock); | |
2805 | return ret; | |
2806 | } | |
2807 | ||
0a2b2a84 JB |
2808 | int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans, |
2809 | struct btrfs_root *root) | |
2810 | { | |
2811 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2812 | u64 num_entries = | |
2813 | atomic_read(&trans->transaction->delayed_refs.num_entries); | |
2814 | u64 avg_runtime; | |
a79b7d4b | 2815 | u64 val; |
0a2b2a84 JB |
2816 | |
2817 | smp_mb(); | |
2818 | avg_runtime = fs_info->avg_delayed_ref_runtime; | |
a79b7d4b | 2819 | val = num_entries * avg_runtime; |
0a2b2a84 JB |
2820 | if (num_entries * avg_runtime >= NSEC_PER_SEC) |
2821 | return 1; | |
a79b7d4b CM |
2822 | if (val >= NSEC_PER_SEC / 2) |
2823 | return 2; | |
0a2b2a84 JB |
2824 | |
2825 | return btrfs_check_space_for_delayed_refs(trans, root); | |
2826 | } | |
2827 | ||
a79b7d4b CM |
2828 | struct async_delayed_refs { |
2829 | struct btrfs_root *root; | |
2830 | int count; | |
2831 | int error; | |
2832 | int sync; | |
2833 | struct completion wait; | |
2834 | struct btrfs_work work; | |
2835 | }; | |
2836 | ||
2837 | static void delayed_ref_async_start(struct btrfs_work *work) | |
2838 | { | |
2839 | struct async_delayed_refs *async; | |
2840 | struct btrfs_trans_handle *trans; | |
2841 | int ret; | |
2842 | ||
2843 | async = container_of(work, struct async_delayed_refs, work); | |
2844 | ||
2845 | trans = btrfs_join_transaction(async->root); | |
2846 | if (IS_ERR(trans)) { | |
2847 | async->error = PTR_ERR(trans); | |
2848 | goto done; | |
2849 | } | |
2850 | ||
2851 | /* | |
2852 | * trans->sync means that when we call end_transaciton, we won't | |
2853 | * wait on delayed refs | |
2854 | */ | |
2855 | trans->sync = true; | |
2856 | ret = btrfs_run_delayed_refs(trans, async->root, async->count); | |
2857 | if (ret) | |
2858 | async->error = ret; | |
2859 | ||
2860 | ret = btrfs_end_transaction(trans, async->root); | |
2861 | if (ret && !async->error) | |
2862 | async->error = ret; | |
2863 | done: | |
2864 | if (async->sync) | |
2865 | complete(&async->wait); | |
2866 | else | |
2867 | kfree(async); | |
2868 | } | |
2869 | ||
2870 | int btrfs_async_run_delayed_refs(struct btrfs_root *root, | |
2871 | unsigned long count, int wait) | |
2872 | { | |
2873 | struct async_delayed_refs *async; | |
2874 | int ret; | |
2875 | ||
2876 | async = kmalloc(sizeof(*async), GFP_NOFS); | |
2877 | if (!async) | |
2878 | return -ENOMEM; | |
2879 | ||
2880 | async->root = root->fs_info->tree_root; | |
2881 | async->count = count; | |
2882 | async->error = 0; | |
2883 | if (wait) | |
2884 | async->sync = 1; | |
2885 | else | |
2886 | async->sync = 0; | |
2887 | init_completion(&async->wait); | |
2888 | ||
9e0af237 LB |
2889 | btrfs_init_work(&async->work, btrfs_extent_refs_helper, |
2890 | delayed_ref_async_start, NULL, NULL); | |
a79b7d4b CM |
2891 | |
2892 | btrfs_queue_work(root->fs_info->extent_workers, &async->work); | |
2893 | ||
2894 | if (wait) { | |
2895 | wait_for_completion(&async->wait); | |
2896 | ret = async->error; | |
2897 | kfree(async); | |
2898 | return ret; | |
2899 | } | |
2900 | return 0; | |
2901 | } | |
2902 | ||
c3e69d58 CM |
2903 | /* |
2904 | * this starts processing the delayed reference count updates and | |
2905 | * extent insertions we have queued up so far. count can be | |
2906 | * 0, which means to process everything in the tree at the start | |
2907 | * of the run (but not newly added entries), or it can be some target | |
2908 | * number you'd like to process. | |
79787eaa JM |
2909 | * |
2910 | * Returns 0 on success or if called with an aborted transaction | |
2911 | * Returns <0 on error and aborts the transaction | |
c3e69d58 CM |
2912 | */ |
2913 | int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, | |
2914 | struct btrfs_root *root, unsigned long count) | |
2915 | { | |
2916 | struct rb_node *node; | |
2917 | struct btrfs_delayed_ref_root *delayed_refs; | |
c46effa6 | 2918 | struct btrfs_delayed_ref_head *head; |
c3e69d58 CM |
2919 | int ret; |
2920 | int run_all = count == (unsigned long)-1; | |
d9a0540a | 2921 | bool can_flush_pending_bgs = trans->can_flush_pending_bgs; |
c3e69d58 | 2922 | |
79787eaa JM |
2923 | /* We'll clean this up in btrfs_cleanup_transaction */ |
2924 | if (trans->aborted) | |
2925 | return 0; | |
2926 | ||
c3e69d58 CM |
2927 | if (root == root->fs_info->extent_root) |
2928 | root = root->fs_info->tree_root; | |
2929 | ||
2930 | delayed_refs = &trans->transaction->delayed_refs; | |
26455d33 | 2931 | if (count == 0) |
d7df2c79 | 2932 | count = atomic_read(&delayed_refs->num_entries) * 2; |
bb721703 | 2933 | |
c3e69d58 | 2934 | again: |
709c0486 AJ |
2935 | #ifdef SCRAMBLE_DELAYED_REFS |
2936 | delayed_refs->run_delayed_start = find_middle(&delayed_refs->root); | |
2937 | #endif | |
d9a0540a | 2938 | trans->can_flush_pending_bgs = false; |
d7df2c79 JB |
2939 | ret = __btrfs_run_delayed_refs(trans, root, count); |
2940 | if (ret < 0) { | |
2941 | btrfs_abort_transaction(trans, root, ret); | |
2942 | return ret; | |
eb099670 | 2943 | } |
c3e69d58 | 2944 | |
56bec294 | 2945 | if (run_all) { |
d7df2c79 | 2946 | if (!list_empty(&trans->new_bgs)) |
ea658bad | 2947 | btrfs_create_pending_block_groups(trans, root); |
ea658bad | 2948 | |
d7df2c79 | 2949 | spin_lock(&delayed_refs->lock); |
c46effa6 | 2950 | node = rb_first(&delayed_refs->href_root); |
d7df2c79 JB |
2951 | if (!node) { |
2952 | spin_unlock(&delayed_refs->lock); | |
56bec294 | 2953 | goto out; |
d7df2c79 | 2954 | } |
c3e69d58 | 2955 | count = (unsigned long)-1; |
e9d0b13b | 2956 | |
56bec294 | 2957 | while (node) { |
c46effa6 LB |
2958 | head = rb_entry(node, struct btrfs_delayed_ref_head, |
2959 | href_node); | |
2960 | if (btrfs_delayed_ref_is_head(&head->node)) { | |
2961 | struct btrfs_delayed_ref_node *ref; | |
5caf2a00 | 2962 | |
c46effa6 | 2963 | ref = &head->node; |
56bec294 CM |
2964 | atomic_inc(&ref->refs); |
2965 | ||
2966 | spin_unlock(&delayed_refs->lock); | |
8cc33e5c DS |
2967 | /* |
2968 | * Mutex was contended, block until it's | |
2969 | * released and try again | |
2970 | */ | |
56bec294 CM |
2971 | mutex_lock(&head->mutex); |
2972 | mutex_unlock(&head->mutex); | |
2973 | ||
2974 | btrfs_put_delayed_ref(ref); | |
1887be66 | 2975 | cond_resched(); |
56bec294 | 2976 | goto again; |
c46effa6 LB |
2977 | } else { |
2978 | WARN_ON(1); | |
56bec294 CM |
2979 | } |
2980 | node = rb_next(node); | |
2981 | } | |
2982 | spin_unlock(&delayed_refs->lock); | |
d7df2c79 | 2983 | cond_resched(); |
56bec294 | 2984 | goto again; |
5f39d397 | 2985 | } |
54aa1f4d | 2986 | out: |
edf39272 | 2987 | assert_qgroups_uptodate(trans); |
d9a0540a | 2988 | trans->can_flush_pending_bgs = can_flush_pending_bgs; |
a28ec197 CM |
2989 | return 0; |
2990 | } | |
2991 | ||
5d4f98a2 YZ |
2992 | int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans, |
2993 | struct btrfs_root *root, | |
2994 | u64 bytenr, u64 num_bytes, u64 flags, | |
b1c79e09 | 2995 | int level, int is_data) |
5d4f98a2 YZ |
2996 | { |
2997 | struct btrfs_delayed_extent_op *extent_op; | |
2998 | int ret; | |
2999 | ||
78a6184a | 3000 | extent_op = btrfs_alloc_delayed_extent_op(); |
5d4f98a2 YZ |
3001 | if (!extent_op) |
3002 | return -ENOMEM; | |
3003 | ||
3004 | extent_op->flags_to_set = flags; | |
3005 | extent_op->update_flags = 1; | |
3006 | extent_op->update_key = 0; | |
3007 | extent_op->is_data = is_data ? 1 : 0; | |
b1c79e09 | 3008 | extent_op->level = level; |
5d4f98a2 | 3009 | |
66d7e7f0 AJ |
3010 | ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr, |
3011 | num_bytes, extent_op); | |
5d4f98a2 | 3012 | if (ret) |
78a6184a | 3013 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 YZ |
3014 | return ret; |
3015 | } | |
3016 | ||
3017 | static noinline int check_delayed_ref(struct btrfs_trans_handle *trans, | |
3018 | struct btrfs_root *root, | |
3019 | struct btrfs_path *path, | |
3020 | u64 objectid, u64 offset, u64 bytenr) | |
3021 | { | |
3022 | struct btrfs_delayed_ref_head *head; | |
3023 | struct btrfs_delayed_ref_node *ref; | |
3024 | struct btrfs_delayed_data_ref *data_ref; | |
3025 | struct btrfs_delayed_ref_root *delayed_refs; | |
5d4f98a2 YZ |
3026 | int ret = 0; |
3027 | ||
5d4f98a2 YZ |
3028 | delayed_refs = &trans->transaction->delayed_refs; |
3029 | spin_lock(&delayed_refs->lock); | |
3030 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
d7df2c79 JB |
3031 | if (!head) { |
3032 | spin_unlock(&delayed_refs->lock); | |
3033 | return 0; | |
3034 | } | |
5d4f98a2 YZ |
3035 | |
3036 | if (!mutex_trylock(&head->mutex)) { | |
3037 | atomic_inc(&head->node.refs); | |
3038 | spin_unlock(&delayed_refs->lock); | |
3039 | ||
b3b4aa74 | 3040 | btrfs_release_path(path); |
5d4f98a2 | 3041 | |
8cc33e5c DS |
3042 | /* |
3043 | * Mutex was contended, block until it's released and let | |
3044 | * caller try again | |
3045 | */ | |
5d4f98a2 YZ |
3046 | mutex_lock(&head->mutex); |
3047 | mutex_unlock(&head->mutex); | |
3048 | btrfs_put_delayed_ref(&head->node); | |
3049 | return -EAGAIN; | |
3050 | } | |
d7df2c79 | 3051 | spin_unlock(&delayed_refs->lock); |
5d4f98a2 | 3052 | |
d7df2c79 | 3053 | spin_lock(&head->lock); |
c6fc2454 | 3054 | list_for_each_entry(ref, &head->ref_list, list) { |
d7df2c79 JB |
3055 | /* If it's a shared ref we know a cross reference exists */ |
3056 | if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) { | |
3057 | ret = 1; | |
3058 | break; | |
3059 | } | |
5d4f98a2 | 3060 | |
d7df2c79 | 3061 | data_ref = btrfs_delayed_node_to_data_ref(ref); |
5d4f98a2 | 3062 | |
d7df2c79 JB |
3063 | /* |
3064 | * If our ref doesn't match the one we're currently looking at | |
3065 | * then we have a cross reference. | |
3066 | */ | |
3067 | if (data_ref->root != root->root_key.objectid || | |
3068 | data_ref->objectid != objectid || | |
3069 | data_ref->offset != offset) { | |
3070 | ret = 1; | |
3071 | break; | |
3072 | } | |
5d4f98a2 | 3073 | } |
d7df2c79 | 3074 | spin_unlock(&head->lock); |
5d4f98a2 | 3075 | mutex_unlock(&head->mutex); |
5d4f98a2 YZ |
3076 | return ret; |
3077 | } | |
3078 | ||
3079 | static noinline int check_committed_ref(struct btrfs_trans_handle *trans, | |
3080 | struct btrfs_root *root, | |
3081 | struct btrfs_path *path, | |
3082 | u64 objectid, u64 offset, u64 bytenr) | |
be20aa9d CM |
3083 | { |
3084 | struct btrfs_root *extent_root = root->fs_info->extent_root; | |
f321e491 | 3085 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
3086 | struct btrfs_extent_data_ref *ref; |
3087 | struct btrfs_extent_inline_ref *iref; | |
3088 | struct btrfs_extent_item *ei; | |
f321e491 | 3089 | struct btrfs_key key; |
5d4f98a2 | 3090 | u32 item_size; |
be20aa9d | 3091 | int ret; |
925baedd | 3092 | |
be20aa9d | 3093 | key.objectid = bytenr; |
31840ae1 | 3094 | key.offset = (u64)-1; |
f321e491 | 3095 | key.type = BTRFS_EXTENT_ITEM_KEY; |
be20aa9d | 3096 | |
be20aa9d CM |
3097 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
3098 | if (ret < 0) | |
3099 | goto out; | |
79787eaa | 3100 | BUG_ON(ret == 0); /* Corruption */ |
80ff3856 YZ |
3101 | |
3102 | ret = -ENOENT; | |
3103 | if (path->slots[0] == 0) | |
31840ae1 | 3104 | goto out; |
be20aa9d | 3105 | |
31840ae1 | 3106 | path->slots[0]--; |
f321e491 | 3107 | leaf = path->nodes[0]; |
5d4f98a2 | 3108 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
be20aa9d | 3109 | |
5d4f98a2 | 3110 | if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY) |
be20aa9d | 3111 | goto out; |
f321e491 | 3112 | |
5d4f98a2 YZ |
3113 | ret = 1; |
3114 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
3115 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
3116 | if (item_size < sizeof(*ei)) { | |
3117 | WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0)); | |
3118 | goto out; | |
3119 | } | |
3120 | #endif | |
3121 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
bd09835d | 3122 | |
5d4f98a2 YZ |
3123 | if (item_size != sizeof(*ei) + |
3124 | btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY)) | |
3125 | goto out; | |
be20aa9d | 3126 | |
5d4f98a2 YZ |
3127 | if (btrfs_extent_generation(leaf, ei) <= |
3128 | btrfs_root_last_snapshot(&root->root_item)) | |
3129 | goto out; | |
3130 | ||
3131 | iref = (struct btrfs_extent_inline_ref *)(ei + 1); | |
3132 | if (btrfs_extent_inline_ref_type(leaf, iref) != | |
3133 | BTRFS_EXTENT_DATA_REF_KEY) | |
3134 | goto out; | |
3135 | ||
3136 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
3137 | if (btrfs_extent_refs(leaf, ei) != | |
3138 | btrfs_extent_data_ref_count(leaf, ref) || | |
3139 | btrfs_extent_data_ref_root(leaf, ref) != | |
3140 | root->root_key.objectid || | |
3141 | btrfs_extent_data_ref_objectid(leaf, ref) != objectid || | |
3142 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
3143 | goto out; | |
3144 | ||
3145 | ret = 0; | |
3146 | out: | |
3147 | return ret; | |
3148 | } | |
3149 | ||
3150 | int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans, | |
3151 | struct btrfs_root *root, | |
3152 | u64 objectid, u64 offset, u64 bytenr) | |
3153 | { | |
3154 | struct btrfs_path *path; | |
3155 | int ret; | |
3156 | int ret2; | |
3157 | ||
3158 | path = btrfs_alloc_path(); | |
3159 | if (!path) | |
3160 | return -ENOENT; | |
3161 | ||
3162 | do { | |
3163 | ret = check_committed_ref(trans, root, path, objectid, | |
3164 | offset, bytenr); | |
3165 | if (ret && ret != -ENOENT) | |
f321e491 | 3166 | goto out; |
80ff3856 | 3167 | |
5d4f98a2 YZ |
3168 | ret2 = check_delayed_ref(trans, root, path, objectid, |
3169 | offset, bytenr); | |
3170 | } while (ret2 == -EAGAIN); | |
3171 | ||
3172 | if (ret2 && ret2 != -ENOENT) { | |
3173 | ret = ret2; | |
3174 | goto out; | |
f321e491 | 3175 | } |
5d4f98a2 YZ |
3176 | |
3177 | if (ret != -ENOENT || ret2 != -ENOENT) | |
3178 | ret = 0; | |
be20aa9d | 3179 | out: |
80ff3856 | 3180 | btrfs_free_path(path); |
f0486c68 YZ |
3181 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) |
3182 | WARN_ON(ret > 0); | |
f321e491 | 3183 | return ret; |
be20aa9d | 3184 | } |
c5739bba | 3185 | |
5d4f98a2 | 3186 | static int __btrfs_mod_ref(struct btrfs_trans_handle *trans, |
b7a9f29f | 3187 | struct btrfs_root *root, |
5d4f98a2 | 3188 | struct extent_buffer *buf, |
e339a6b0 | 3189 | int full_backref, int inc) |
31840ae1 ZY |
3190 | { |
3191 | u64 bytenr; | |
5d4f98a2 YZ |
3192 | u64 num_bytes; |
3193 | u64 parent; | |
31840ae1 | 3194 | u64 ref_root; |
31840ae1 | 3195 | u32 nritems; |
31840ae1 ZY |
3196 | struct btrfs_key key; |
3197 | struct btrfs_file_extent_item *fi; | |
3198 | int i; | |
3199 | int level; | |
3200 | int ret = 0; | |
31840ae1 | 3201 | int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *, |
b06c4bf5 | 3202 | u64, u64, u64, u64, u64, u64); |
31840ae1 | 3203 | |
fccb84c9 DS |
3204 | |
3205 | if (btrfs_test_is_dummy_root(root)) | |
faa2dbf0 | 3206 | return 0; |
fccb84c9 | 3207 | |
31840ae1 | 3208 | ref_root = btrfs_header_owner(buf); |
31840ae1 ZY |
3209 | nritems = btrfs_header_nritems(buf); |
3210 | level = btrfs_header_level(buf); | |
3211 | ||
27cdeb70 | 3212 | if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0) |
5d4f98a2 | 3213 | return 0; |
31840ae1 | 3214 | |
5d4f98a2 YZ |
3215 | if (inc) |
3216 | process_func = btrfs_inc_extent_ref; | |
3217 | else | |
3218 | process_func = btrfs_free_extent; | |
31840ae1 | 3219 | |
5d4f98a2 YZ |
3220 | if (full_backref) |
3221 | parent = buf->start; | |
3222 | else | |
3223 | parent = 0; | |
3224 | ||
3225 | for (i = 0; i < nritems; i++) { | |
31840ae1 | 3226 | if (level == 0) { |
5d4f98a2 | 3227 | btrfs_item_key_to_cpu(buf, &key, i); |
962a298f | 3228 | if (key.type != BTRFS_EXTENT_DATA_KEY) |
31840ae1 | 3229 | continue; |
5d4f98a2 | 3230 | fi = btrfs_item_ptr(buf, i, |
31840ae1 ZY |
3231 | struct btrfs_file_extent_item); |
3232 | if (btrfs_file_extent_type(buf, fi) == | |
3233 | BTRFS_FILE_EXTENT_INLINE) | |
3234 | continue; | |
3235 | bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
3236 | if (bytenr == 0) | |
3237 | continue; | |
5d4f98a2 YZ |
3238 | |
3239 | num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi); | |
3240 | key.offset -= btrfs_file_extent_offset(buf, fi); | |
3241 | ret = process_func(trans, root, bytenr, num_bytes, | |
3242 | parent, ref_root, key.objectid, | |
b06c4bf5 | 3243 | key.offset); |
31840ae1 ZY |
3244 | if (ret) |
3245 | goto fail; | |
3246 | } else { | |
5d4f98a2 | 3247 | bytenr = btrfs_node_blockptr(buf, i); |
707e8a07 | 3248 | num_bytes = root->nodesize; |
5d4f98a2 | 3249 | ret = process_func(trans, root, bytenr, num_bytes, |
b06c4bf5 | 3250 | parent, ref_root, level - 1, 0); |
31840ae1 ZY |
3251 | if (ret) |
3252 | goto fail; | |
3253 | } | |
3254 | } | |
3255 | return 0; | |
3256 | fail: | |
5d4f98a2 YZ |
3257 | return ret; |
3258 | } | |
3259 | ||
3260 | int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
e339a6b0 | 3261 | struct extent_buffer *buf, int full_backref) |
5d4f98a2 | 3262 | { |
e339a6b0 | 3263 | return __btrfs_mod_ref(trans, root, buf, full_backref, 1); |
5d4f98a2 YZ |
3264 | } |
3265 | ||
3266 | int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
e339a6b0 | 3267 | struct extent_buffer *buf, int full_backref) |
5d4f98a2 | 3268 | { |
e339a6b0 | 3269 | return __btrfs_mod_ref(trans, root, buf, full_backref, 0); |
31840ae1 ZY |
3270 | } |
3271 | ||
9078a3e1 CM |
3272 | static int write_one_cache_group(struct btrfs_trans_handle *trans, |
3273 | struct btrfs_root *root, | |
3274 | struct btrfs_path *path, | |
3275 | struct btrfs_block_group_cache *cache) | |
3276 | { | |
3277 | int ret; | |
9078a3e1 | 3278 | struct btrfs_root *extent_root = root->fs_info->extent_root; |
5f39d397 CM |
3279 | unsigned long bi; |
3280 | struct extent_buffer *leaf; | |
9078a3e1 | 3281 | |
9078a3e1 | 3282 | ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1); |
df95e7f0 JB |
3283 | if (ret) { |
3284 | if (ret > 0) | |
3285 | ret = -ENOENT; | |
54aa1f4d | 3286 | goto fail; |
df95e7f0 | 3287 | } |
5f39d397 CM |
3288 | |
3289 | leaf = path->nodes[0]; | |
3290 | bi = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
3291 | write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item)); | |
3292 | btrfs_mark_buffer_dirty(leaf); | |
54aa1f4d | 3293 | fail: |
24b89d08 | 3294 | btrfs_release_path(path); |
df95e7f0 | 3295 | return ret; |
9078a3e1 CM |
3296 | |
3297 | } | |
3298 | ||
4a8c9a62 YZ |
3299 | static struct btrfs_block_group_cache * |
3300 | next_block_group(struct btrfs_root *root, | |
3301 | struct btrfs_block_group_cache *cache) | |
3302 | { | |
3303 | struct rb_node *node; | |
292cbd51 | 3304 | |
4a8c9a62 | 3305 | spin_lock(&root->fs_info->block_group_cache_lock); |
292cbd51 FM |
3306 | |
3307 | /* If our block group was removed, we need a full search. */ | |
3308 | if (RB_EMPTY_NODE(&cache->cache_node)) { | |
3309 | const u64 next_bytenr = cache->key.objectid + cache->key.offset; | |
3310 | ||
3311 | spin_unlock(&root->fs_info->block_group_cache_lock); | |
3312 | btrfs_put_block_group(cache); | |
3313 | cache = btrfs_lookup_first_block_group(root->fs_info, | |
3314 | next_bytenr); | |
3315 | return cache; | |
3316 | } | |
4a8c9a62 YZ |
3317 | node = rb_next(&cache->cache_node); |
3318 | btrfs_put_block_group(cache); | |
3319 | if (node) { | |
3320 | cache = rb_entry(node, struct btrfs_block_group_cache, | |
3321 | cache_node); | |
11dfe35a | 3322 | btrfs_get_block_group(cache); |
4a8c9a62 YZ |
3323 | } else |
3324 | cache = NULL; | |
3325 | spin_unlock(&root->fs_info->block_group_cache_lock); | |
3326 | return cache; | |
3327 | } | |
3328 | ||
0af3d00b JB |
3329 | static int cache_save_setup(struct btrfs_block_group_cache *block_group, |
3330 | struct btrfs_trans_handle *trans, | |
3331 | struct btrfs_path *path) | |
3332 | { | |
3333 | struct btrfs_root *root = block_group->fs_info->tree_root; | |
3334 | struct inode *inode = NULL; | |
3335 | u64 alloc_hint = 0; | |
2b20982e | 3336 | int dcs = BTRFS_DC_ERROR; |
f8c269d7 | 3337 | u64 num_pages = 0; |
0af3d00b JB |
3338 | int retries = 0; |
3339 | int ret = 0; | |
3340 | ||
3341 | /* | |
3342 | * If this block group is smaller than 100 megs don't bother caching the | |
3343 | * block group. | |
3344 | */ | |
3345 | if (block_group->key.offset < (100 * 1024 * 1024)) { | |
3346 | spin_lock(&block_group->lock); | |
3347 | block_group->disk_cache_state = BTRFS_DC_WRITTEN; | |
3348 | spin_unlock(&block_group->lock); | |
3349 | return 0; | |
3350 | } | |
3351 | ||
0c0ef4bc JB |
3352 | if (trans->aborted) |
3353 | return 0; | |
0af3d00b JB |
3354 | again: |
3355 | inode = lookup_free_space_inode(root, block_group, path); | |
3356 | if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) { | |
3357 | ret = PTR_ERR(inode); | |
b3b4aa74 | 3358 | btrfs_release_path(path); |
0af3d00b JB |
3359 | goto out; |
3360 | } | |
3361 | ||
3362 | if (IS_ERR(inode)) { | |
3363 | BUG_ON(retries); | |
3364 | retries++; | |
3365 | ||
3366 | if (block_group->ro) | |
3367 | goto out_free; | |
3368 | ||
3369 | ret = create_free_space_inode(root, trans, block_group, path); | |
3370 | if (ret) | |
3371 | goto out_free; | |
3372 | goto again; | |
3373 | } | |
3374 | ||
5b0e95bf JB |
3375 | /* We've already setup this transaction, go ahead and exit */ |
3376 | if (block_group->cache_generation == trans->transid && | |
3377 | i_size_read(inode)) { | |
3378 | dcs = BTRFS_DC_SETUP; | |
3379 | goto out_put; | |
3380 | } | |
3381 | ||
0af3d00b JB |
3382 | /* |
3383 | * We want to set the generation to 0, that way if anything goes wrong | |
3384 | * from here on out we know not to trust this cache when we load up next | |
3385 | * time. | |
3386 | */ | |
3387 | BTRFS_I(inode)->generation = 0; | |
3388 | ret = btrfs_update_inode(trans, root, inode); | |
0c0ef4bc JB |
3389 | if (ret) { |
3390 | /* | |
3391 | * So theoretically we could recover from this, simply set the | |
3392 | * super cache generation to 0 so we know to invalidate the | |
3393 | * cache, but then we'd have to keep track of the block groups | |
3394 | * that fail this way so we know we _have_ to reset this cache | |
3395 | * before the next commit or risk reading stale cache. So to | |
3396 | * limit our exposure to horrible edge cases lets just abort the | |
3397 | * transaction, this only happens in really bad situations | |
3398 | * anyway. | |
3399 | */ | |
3400 | btrfs_abort_transaction(trans, root, ret); | |
3401 | goto out_put; | |
3402 | } | |
0af3d00b JB |
3403 | WARN_ON(ret); |
3404 | ||
3405 | if (i_size_read(inode) > 0) { | |
7b61cd92 MX |
3406 | ret = btrfs_check_trunc_cache_free_space(root, |
3407 | &root->fs_info->global_block_rsv); | |
3408 | if (ret) | |
3409 | goto out_put; | |
3410 | ||
1bbc621e | 3411 | ret = btrfs_truncate_free_space_cache(root, trans, NULL, inode); |
0af3d00b JB |
3412 | if (ret) |
3413 | goto out_put; | |
3414 | } | |
3415 | ||
3416 | spin_lock(&block_group->lock); | |
cf7c1ef6 | 3417 | if (block_group->cached != BTRFS_CACHE_FINISHED || |
e4c88f00 | 3418 | !btrfs_test_opt(root, SPACE_CACHE)) { |
cf7c1ef6 LB |
3419 | /* |
3420 | * don't bother trying to write stuff out _if_ | |
3421 | * a) we're not cached, | |
3422 | * b) we're with nospace_cache mount option. | |
3423 | */ | |
2b20982e | 3424 | dcs = BTRFS_DC_WRITTEN; |
0af3d00b JB |
3425 | spin_unlock(&block_group->lock); |
3426 | goto out_put; | |
3427 | } | |
3428 | spin_unlock(&block_group->lock); | |
3429 | ||
2968b1f4 JB |
3430 | /* |
3431 | * We hit an ENOSPC when setting up the cache in this transaction, just | |
3432 | * skip doing the setup, we've already cleared the cache so we're safe. | |
3433 | */ | |
3434 | if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) { | |
3435 | ret = -ENOSPC; | |
3436 | goto out_put; | |
3437 | } | |
3438 | ||
6fc823b1 JB |
3439 | /* |
3440 | * Try to preallocate enough space based on how big the block group is. | |
3441 | * Keep in mind this has to include any pinned space which could end up | |
3442 | * taking up quite a bit since it's not folded into the other space | |
3443 | * cache. | |
3444 | */ | |
f8c269d7 | 3445 | num_pages = div_u64(block_group->key.offset, 256 * 1024 * 1024); |
0af3d00b JB |
3446 | if (!num_pages) |
3447 | num_pages = 1; | |
3448 | ||
0af3d00b JB |
3449 | num_pages *= 16; |
3450 | num_pages *= PAGE_CACHE_SIZE; | |
3451 | ||
7cf5b976 | 3452 | ret = btrfs_check_data_free_space(inode, 0, num_pages); |
0af3d00b JB |
3453 | if (ret) |
3454 | goto out_put; | |
3455 | ||
3456 | ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages, | |
3457 | num_pages, num_pages, | |
3458 | &alloc_hint); | |
2968b1f4 JB |
3459 | /* |
3460 | * Our cache requires contiguous chunks so that we don't modify a bunch | |
3461 | * of metadata or split extents when writing the cache out, which means | |
3462 | * we can enospc if we are heavily fragmented in addition to just normal | |
3463 | * out of space conditions. So if we hit this just skip setting up any | |
3464 | * other block groups for this transaction, maybe we'll unpin enough | |
3465 | * space the next time around. | |
3466 | */ | |
2b20982e JB |
3467 | if (!ret) |
3468 | dcs = BTRFS_DC_SETUP; | |
2968b1f4 JB |
3469 | else if (ret == -ENOSPC) |
3470 | set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags); | |
7cf5b976 | 3471 | btrfs_free_reserved_data_space(inode, 0, num_pages); |
c09544e0 | 3472 | |
0af3d00b JB |
3473 | out_put: |
3474 | iput(inode); | |
3475 | out_free: | |
b3b4aa74 | 3476 | btrfs_release_path(path); |
0af3d00b JB |
3477 | out: |
3478 | spin_lock(&block_group->lock); | |
e65cbb94 | 3479 | if (!ret && dcs == BTRFS_DC_SETUP) |
5b0e95bf | 3480 | block_group->cache_generation = trans->transid; |
2b20982e | 3481 | block_group->disk_cache_state = dcs; |
0af3d00b JB |
3482 | spin_unlock(&block_group->lock); |
3483 | ||
3484 | return ret; | |
3485 | } | |
3486 | ||
dcdf7f6d JB |
3487 | int btrfs_setup_space_cache(struct btrfs_trans_handle *trans, |
3488 | struct btrfs_root *root) | |
3489 | { | |
3490 | struct btrfs_block_group_cache *cache, *tmp; | |
3491 | struct btrfs_transaction *cur_trans = trans->transaction; | |
3492 | struct btrfs_path *path; | |
3493 | ||
3494 | if (list_empty(&cur_trans->dirty_bgs) || | |
3495 | !btrfs_test_opt(root, SPACE_CACHE)) | |
3496 | return 0; | |
3497 | ||
3498 | path = btrfs_alloc_path(); | |
3499 | if (!path) | |
3500 | return -ENOMEM; | |
3501 | ||
3502 | /* Could add new block groups, use _safe just in case */ | |
3503 | list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs, | |
3504 | dirty_list) { | |
3505 | if (cache->disk_cache_state == BTRFS_DC_CLEAR) | |
3506 | cache_save_setup(cache, trans, path); | |
3507 | } | |
3508 | ||
3509 | btrfs_free_path(path); | |
3510 | return 0; | |
3511 | } | |
3512 | ||
1bbc621e CM |
3513 | /* |
3514 | * transaction commit does final block group cache writeback during a | |
3515 | * critical section where nothing is allowed to change the FS. This is | |
3516 | * required in order for the cache to actually match the block group, | |
3517 | * but can introduce a lot of latency into the commit. | |
3518 | * | |
3519 | * So, btrfs_start_dirty_block_groups is here to kick off block group | |
3520 | * cache IO. There's a chance we'll have to redo some of it if the | |
3521 | * block group changes again during the commit, but it greatly reduces | |
3522 | * the commit latency by getting rid of the easy block groups while | |
3523 | * we're still allowing others to join the commit. | |
3524 | */ | |
3525 | int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans, | |
96b5179d | 3526 | struct btrfs_root *root) |
9078a3e1 | 3527 | { |
4a8c9a62 | 3528 | struct btrfs_block_group_cache *cache; |
ce93ec54 JB |
3529 | struct btrfs_transaction *cur_trans = trans->transaction; |
3530 | int ret = 0; | |
c9dc4c65 | 3531 | int should_put; |
1bbc621e CM |
3532 | struct btrfs_path *path = NULL; |
3533 | LIST_HEAD(dirty); | |
3534 | struct list_head *io = &cur_trans->io_bgs; | |
c9dc4c65 | 3535 | int num_started = 0; |
1bbc621e CM |
3536 | int loops = 0; |
3537 | ||
3538 | spin_lock(&cur_trans->dirty_bgs_lock); | |
b58d1a9e FM |
3539 | if (list_empty(&cur_trans->dirty_bgs)) { |
3540 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3541 | return 0; | |
1bbc621e | 3542 | } |
b58d1a9e | 3543 | list_splice_init(&cur_trans->dirty_bgs, &dirty); |
1bbc621e | 3544 | spin_unlock(&cur_trans->dirty_bgs_lock); |
ce93ec54 | 3545 | |
1bbc621e | 3546 | again: |
1bbc621e CM |
3547 | /* |
3548 | * make sure all the block groups on our dirty list actually | |
3549 | * exist | |
3550 | */ | |
3551 | btrfs_create_pending_block_groups(trans, root); | |
3552 | ||
3553 | if (!path) { | |
3554 | path = btrfs_alloc_path(); | |
3555 | if (!path) | |
3556 | return -ENOMEM; | |
3557 | } | |
3558 | ||
b58d1a9e FM |
3559 | /* |
3560 | * cache_write_mutex is here only to save us from balance or automatic | |
3561 | * removal of empty block groups deleting this block group while we are | |
3562 | * writing out the cache | |
3563 | */ | |
3564 | mutex_lock(&trans->transaction->cache_write_mutex); | |
1bbc621e CM |
3565 | while (!list_empty(&dirty)) { |
3566 | cache = list_first_entry(&dirty, | |
3567 | struct btrfs_block_group_cache, | |
3568 | dirty_list); | |
1bbc621e CM |
3569 | /* |
3570 | * this can happen if something re-dirties a block | |
3571 | * group that is already under IO. Just wait for it to | |
3572 | * finish and then do it all again | |
3573 | */ | |
3574 | if (!list_empty(&cache->io_list)) { | |
3575 | list_del_init(&cache->io_list); | |
3576 | btrfs_wait_cache_io(root, trans, cache, | |
3577 | &cache->io_ctl, path, | |
3578 | cache->key.objectid); | |
3579 | btrfs_put_block_group(cache); | |
3580 | } | |
3581 | ||
3582 | ||
3583 | /* | |
3584 | * btrfs_wait_cache_io uses the cache->dirty_list to decide | |
3585 | * if it should update the cache_state. Don't delete | |
3586 | * until after we wait. | |
3587 | * | |
3588 | * Since we're not running in the commit critical section | |
3589 | * we need the dirty_bgs_lock to protect from update_block_group | |
3590 | */ | |
3591 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3592 | list_del_init(&cache->dirty_list); | |
3593 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3594 | ||
3595 | should_put = 1; | |
3596 | ||
3597 | cache_save_setup(cache, trans, path); | |
3598 | ||
3599 | if (cache->disk_cache_state == BTRFS_DC_SETUP) { | |
3600 | cache->io_ctl.inode = NULL; | |
3601 | ret = btrfs_write_out_cache(root, trans, cache, path); | |
3602 | if (ret == 0 && cache->io_ctl.inode) { | |
3603 | num_started++; | |
3604 | should_put = 0; | |
3605 | ||
3606 | /* | |
3607 | * the cache_write_mutex is protecting | |
3608 | * the io_list | |
3609 | */ | |
3610 | list_add_tail(&cache->io_list, io); | |
3611 | } else { | |
3612 | /* | |
3613 | * if we failed to write the cache, the | |
3614 | * generation will be bad and life goes on | |
3615 | */ | |
3616 | ret = 0; | |
3617 | } | |
3618 | } | |
ff1f8250 | 3619 | if (!ret) { |
1bbc621e | 3620 | ret = write_one_cache_group(trans, root, path, cache); |
ff1f8250 FM |
3621 | /* |
3622 | * Our block group might still be attached to the list | |
3623 | * of new block groups in the transaction handle of some | |
3624 | * other task (struct btrfs_trans_handle->new_bgs). This | |
3625 | * means its block group item isn't yet in the extent | |
3626 | * tree. If this happens ignore the error, as we will | |
3627 | * try again later in the critical section of the | |
3628 | * transaction commit. | |
3629 | */ | |
3630 | if (ret == -ENOENT) { | |
3631 | ret = 0; | |
3632 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3633 | if (list_empty(&cache->dirty_list)) { | |
3634 | list_add_tail(&cache->dirty_list, | |
3635 | &cur_trans->dirty_bgs); | |
3636 | btrfs_get_block_group(cache); | |
3637 | } | |
3638 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3639 | } else if (ret) { | |
3640 | btrfs_abort_transaction(trans, root, ret); | |
3641 | } | |
3642 | } | |
1bbc621e CM |
3643 | |
3644 | /* if its not on the io list, we need to put the block group */ | |
3645 | if (should_put) | |
3646 | btrfs_put_block_group(cache); | |
3647 | ||
3648 | if (ret) | |
3649 | break; | |
b58d1a9e FM |
3650 | |
3651 | /* | |
3652 | * Avoid blocking other tasks for too long. It might even save | |
3653 | * us from writing caches for block groups that are going to be | |
3654 | * removed. | |
3655 | */ | |
3656 | mutex_unlock(&trans->transaction->cache_write_mutex); | |
3657 | mutex_lock(&trans->transaction->cache_write_mutex); | |
1bbc621e | 3658 | } |
b58d1a9e | 3659 | mutex_unlock(&trans->transaction->cache_write_mutex); |
1bbc621e CM |
3660 | |
3661 | /* | |
3662 | * go through delayed refs for all the stuff we've just kicked off | |
3663 | * and then loop back (just once) | |
3664 | */ | |
3665 | ret = btrfs_run_delayed_refs(trans, root, 0); | |
3666 | if (!ret && loops == 0) { | |
3667 | loops++; | |
3668 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3669 | list_splice_init(&cur_trans->dirty_bgs, &dirty); | |
b58d1a9e FM |
3670 | /* |
3671 | * dirty_bgs_lock protects us from concurrent block group | |
3672 | * deletes too (not just cache_write_mutex). | |
3673 | */ | |
3674 | if (!list_empty(&dirty)) { | |
3675 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3676 | goto again; | |
3677 | } | |
1bbc621e | 3678 | spin_unlock(&cur_trans->dirty_bgs_lock); |
1bbc621e CM |
3679 | } |
3680 | ||
3681 | btrfs_free_path(path); | |
3682 | return ret; | |
3683 | } | |
3684 | ||
3685 | int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans, | |
3686 | struct btrfs_root *root) | |
3687 | { | |
3688 | struct btrfs_block_group_cache *cache; | |
3689 | struct btrfs_transaction *cur_trans = trans->transaction; | |
3690 | int ret = 0; | |
3691 | int should_put; | |
3692 | struct btrfs_path *path; | |
3693 | struct list_head *io = &cur_trans->io_bgs; | |
3694 | int num_started = 0; | |
9078a3e1 CM |
3695 | |
3696 | path = btrfs_alloc_path(); | |
3697 | if (!path) | |
3698 | return -ENOMEM; | |
3699 | ||
ce93ec54 | 3700 | /* |
e44081ef FM |
3701 | * Even though we are in the critical section of the transaction commit, |
3702 | * we can still have concurrent tasks adding elements to this | |
3703 | * transaction's list of dirty block groups. These tasks correspond to | |
3704 | * endio free space workers started when writeback finishes for a | |
3705 | * space cache, which run inode.c:btrfs_finish_ordered_io(), and can | |
3706 | * allocate new block groups as a result of COWing nodes of the root | |
3707 | * tree when updating the free space inode. The writeback for the space | |
3708 | * caches is triggered by an earlier call to | |
3709 | * btrfs_start_dirty_block_groups() and iterations of the following | |
3710 | * loop. | |
3711 | * Also we want to do the cache_save_setup first and then run the | |
ce93ec54 JB |
3712 | * delayed refs to make sure we have the best chance at doing this all |
3713 | * in one shot. | |
3714 | */ | |
e44081ef | 3715 | spin_lock(&cur_trans->dirty_bgs_lock); |
ce93ec54 JB |
3716 | while (!list_empty(&cur_trans->dirty_bgs)) { |
3717 | cache = list_first_entry(&cur_trans->dirty_bgs, | |
3718 | struct btrfs_block_group_cache, | |
3719 | dirty_list); | |
c9dc4c65 CM |
3720 | |
3721 | /* | |
3722 | * this can happen if cache_save_setup re-dirties a block | |
3723 | * group that is already under IO. Just wait for it to | |
3724 | * finish and then do it all again | |
3725 | */ | |
3726 | if (!list_empty(&cache->io_list)) { | |
e44081ef | 3727 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 CM |
3728 | list_del_init(&cache->io_list); |
3729 | btrfs_wait_cache_io(root, trans, cache, | |
3730 | &cache->io_ctl, path, | |
3731 | cache->key.objectid); | |
3732 | btrfs_put_block_group(cache); | |
e44081ef | 3733 | spin_lock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 CM |
3734 | } |
3735 | ||
1bbc621e CM |
3736 | /* |
3737 | * don't remove from the dirty list until after we've waited | |
3738 | * on any pending IO | |
3739 | */ | |
ce93ec54 | 3740 | list_del_init(&cache->dirty_list); |
e44081ef | 3741 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 CM |
3742 | should_put = 1; |
3743 | ||
1bbc621e | 3744 | cache_save_setup(cache, trans, path); |
c9dc4c65 | 3745 | |
ce93ec54 | 3746 | if (!ret) |
c9dc4c65 CM |
3747 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long) -1); |
3748 | ||
3749 | if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) { | |
3750 | cache->io_ctl.inode = NULL; | |
3751 | ret = btrfs_write_out_cache(root, trans, cache, path); | |
3752 | if (ret == 0 && cache->io_ctl.inode) { | |
3753 | num_started++; | |
3754 | should_put = 0; | |
1bbc621e | 3755 | list_add_tail(&cache->io_list, io); |
c9dc4c65 CM |
3756 | } else { |
3757 | /* | |
3758 | * if we failed to write the cache, the | |
3759 | * generation will be bad and life goes on | |
3760 | */ | |
3761 | ret = 0; | |
3762 | } | |
3763 | } | |
ff1f8250 | 3764 | if (!ret) { |
ce93ec54 | 3765 | ret = write_one_cache_group(trans, root, path, cache); |
ff1f8250 FM |
3766 | if (ret) |
3767 | btrfs_abort_transaction(trans, root, ret); | |
3768 | } | |
c9dc4c65 CM |
3769 | |
3770 | /* if its not on the io list, we need to put the block group */ | |
3771 | if (should_put) | |
3772 | btrfs_put_block_group(cache); | |
e44081ef | 3773 | spin_lock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 | 3774 | } |
e44081ef | 3775 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 | 3776 | |
1bbc621e CM |
3777 | while (!list_empty(io)) { |
3778 | cache = list_first_entry(io, struct btrfs_block_group_cache, | |
c9dc4c65 CM |
3779 | io_list); |
3780 | list_del_init(&cache->io_list); | |
c9dc4c65 CM |
3781 | btrfs_wait_cache_io(root, trans, cache, |
3782 | &cache->io_ctl, path, cache->key.objectid); | |
0cb59c99 JB |
3783 | btrfs_put_block_group(cache); |
3784 | } | |
3785 | ||
9078a3e1 | 3786 | btrfs_free_path(path); |
ce93ec54 | 3787 | return ret; |
9078a3e1 CM |
3788 | } |
3789 | ||
d2fb3437 YZ |
3790 | int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr) |
3791 | { | |
3792 | struct btrfs_block_group_cache *block_group; | |
3793 | int readonly = 0; | |
3794 | ||
3795 | block_group = btrfs_lookup_block_group(root->fs_info, bytenr); | |
3796 | if (!block_group || block_group->ro) | |
3797 | readonly = 1; | |
3798 | if (block_group) | |
fa9c0d79 | 3799 | btrfs_put_block_group(block_group); |
d2fb3437 YZ |
3800 | return readonly; |
3801 | } | |
3802 | ||
6ab0a202 JM |
3803 | static const char *alloc_name(u64 flags) |
3804 | { | |
3805 | switch (flags) { | |
3806 | case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA: | |
3807 | return "mixed"; | |
3808 | case BTRFS_BLOCK_GROUP_METADATA: | |
3809 | return "metadata"; | |
3810 | case BTRFS_BLOCK_GROUP_DATA: | |
3811 | return "data"; | |
3812 | case BTRFS_BLOCK_GROUP_SYSTEM: | |
3813 | return "system"; | |
3814 | default: | |
3815 | WARN_ON(1); | |
3816 | return "invalid-combination"; | |
3817 | }; | |
3818 | } | |
3819 | ||
593060d7 CM |
3820 | static int update_space_info(struct btrfs_fs_info *info, u64 flags, |
3821 | u64 total_bytes, u64 bytes_used, | |
3822 | struct btrfs_space_info **space_info) | |
3823 | { | |
3824 | struct btrfs_space_info *found; | |
b742bb82 YZ |
3825 | int i; |
3826 | int factor; | |
b150a4f1 | 3827 | int ret; |
b742bb82 YZ |
3828 | |
3829 | if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
3830 | BTRFS_BLOCK_GROUP_RAID10)) | |
3831 | factor = 2; | |
3832 | else | |
3833 | factor = 1; | |
593060d7 CM |
3834 | |
3835 | found = __find_space_info(info, flags); | |
3836 | if (found) { | |
25179201 | 3837 | spin_lock(&found->lock); |
593060d7 | 3838 | found->total_bytes += total_bytes; |
89a55897 | 3839 | found->disk_total += total_bytes * factor; |
593060d7 | 3840 | found->bytes_used += bytes_used; |
b742bb82 | 3841 | found->disk_used += bytes_used * factor; |
2e6e5183 FM |
3842 | if (total_bytes > 0) |
3843 | found->full = 0; | |
25179201 | 3844 | spin_unlock(&found->lock); |
593060d7 CM |
3845 | *space_info = found; |
3846 | return 0; | |
3847 | } | |
c146afad | 3848 | found = kzalloc(sizeof(*found), GFP_NOFS); |
593060d7 CM |
3849 | if (!found) |
3850 | return -ENOMEM; | |
3851 | ||
908c7f19 | 3852 | ret = percpu_counter_init(&found->total_bytes_pinned, 0, GFP_KERNEL); |
b150a4f1 JB |
3853 | if (ret) { |
3854 | kfree(found); | |
3855 | return ret; | |
3856 | } | |
3857 | ||
c1895442 | 3858 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) |
b742bb82 | 3859 | INIT_LIST_HEAD(&found->block_groups[i]); |
80eb234a | 3860 | init_rwsem(&found->groups_sem); |
0f9dd46c | 3861 | spin_lock_init(&found->lock); |
52ba6929 | 3862 | found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK; |
593060d7 | 3863 | found->total_bytes = total_bytes; |
89a55897 | 3864 | found->disk_total = total_bytes * factor; |
593060d7 | 3865 | found->bytes_used = bytes_used; |
b742bb82 | 3866 | found->disk_used = bytes_used * factor; |
593060d7 | 3867 | found->bytes_pinned = 0; |
e8569813 | 3868 | found->bytes_reserved = 0; |
c146afad | 3869 | found->bytes_readonly = 0; |
f0486c68 | 3870 | found->bytes_may_use = 0; |
6af3e3ad | 3871 | found->full = 0; |
4f4db217 | 3872 | found->max_extent_size = 0; |
0e4f8f88 | 3873 | found->force_alloc = CHUNK_ALLOC_NO_FORCE; |
6d74119f | 3874 | found->chunk_alloc = 0; |
fdb5effd JB |
3875 | found->flush = 0; |
3876 | init_waitqueue_head(&found->wait); | |
633c0aad | 3877 | INIT_LIST_HEAD(&found->ro_bgs); |
6ab0a202 JM |
3878 | |
3879 | ret = kobject_init_and_add(&found->kobj, &space_info_ktype, | |
3880 | info->space_info_kobj, "%s", | |
3881 | alloc_name(found->flags)); | |
3882 | if (ret) { | |
3883 | kfree(found); | |
3884 | return ret; | |
3885 | } | |
3886 | ||
593060d7 | 3887 | *space_info = found; |
4184ea7f | 3888 | list_add_rcu(&found->list, &info->space_info); |
b4d7c3c9 LZ |
3889 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
3890 | info->data_sinfo = found; | |
6ab0a202 JM |
3891 | |
3892 | return ret; | |
593060d7 CM |
3893 | } |
3894 | ||
8790d502 CM |
3895 | static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
3896 | { | |
899c81ea ID |
3897 | u64 extra_flags = chunk_to_extended(flags) & |
3898 | BTRFS_EXTENDED_PROFILE_MASK; | |
a46d11a8 | 3899 | |
de98ced9 | 3900 | write_seqlock(&fs_info->profiles_lock); |
a46d11a8 ID |
3901 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
3902 | fs_info->avail_data_alloc_bits |= extra_flags; | |
3903 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
3904 | fs_info->avail_metadata_alloc_bits |= extra_flags; | |
3905 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
3906 | fs_info->avail_system_alloc_bits |= extra_flags; | |
de98ced9 | 3907 | write_sequnlock(&fs_info->profiles_lock); |
8790d502 | 3908 | } |
593060d7 | 3909 | |
fc67c450 ID |
3910 | /* |
3911 | * returns target flags in extended format or 0 if restripe for this | |
3912 | * chunk_type is not in progress | |
c6664b42 ID |
3913 | * |
3914 | * should be called with either volume_mutex or balance_lock held | |
fc67c450 ID |
3915 | */ |
3916 | static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags) | |
3917 | { | |
3918 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
3919 | u64 target = 0; | |
3920 | ||
fc67c450 ID |
3921 | if (!bctl) |
3922 | return 0; | |
3923 | ||
3924 | if (flags & BTRFS_BLOCK_GROUP_DATA && | |
3925 | bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
3926 | target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target; | |
3927 | } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM && | |
3928 | bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
3929 | target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target; | |
3930 | } else if (flags & BTRFS_BLOCK_GROUP_METADATA && | |
3931 | bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
3932 | target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target; | |
3933 | } | |
3934 | ||
3935 | return target; | |
3936 | } | |
3937 | ||
a46d11a8 ID |
3938 | /* |
3939 | * @flags: available profiles in extended format (see ctree.h) | |
3940 | * | |
e4d8ec0f ID |
3941 | * Returns reduced profile in chunk format. If profile changing is in |
3942 | * progress (either running or paused) picks the target profile (if it's | |
3943 | * already available), otherwise falls back to plain reducing. | |
a46d11a8 | 3944 | */ |
48a3b636 | 3945 | static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags) |
ec44a35c | 3946 | { |
95669976 | 3947 | u64 num_devices = root->fs_info->fs_devices->rw_devices; |
fc67c450 | 3948 | u64 target; |
9c170b26 ZL |
3949 | u64 raid_type; |
3950 | u64 allowed = 0; | |
a061fc8d | 3951 | |
fc67c450 ID |
3952 | /* |
3953 | * see if restripe for this chunk_type is in progress, if so | |
3954 | * try to reduce to the target profile | |
3955 | */ | |
e4d8ec0f | 3956 | spin_lock(&root->fs_info->balance_lock); |
fc67c450 ID |
3957 | target = get_restripe_target(root->fs_info, flags); |
3958 | if (target) { | |
3959 | /* pick target profile only if it's already available */ | |
3960 | if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) { | |
e4d8ec0f | 3961 | spin_unlock(&root->fs_info->balance_lock); |
fc67c450 | 3962 | return extended_to_chunk(target); |
e4d8ec0f ID |
3963 | } |
3964 | } | |
3965 | spin_unlock(&root->fs_info->balance_lock); | |
3966 | ||
53b381b3 | 3967 | /* First, mask out the RAID levels which aren't possible */ |
9c170b26 ZL |
3968 | for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) { |
3969 | if (num_devices >= btrfs_raid_array[raid_type].devs_min) | |
3970 | allowed |= btrfs_raid_group[raid_type]; | |
3971 | } | |
3972 | allowed &= flags; | |
3973 | ||
3974 | if (allowed & BTRFS_BLOCK_GROUP_RAID6) | |
3975 | allowed = BTRFS_BLOCK_GROUP_RAID6; | |
3976 | else if (allowed & BTRFS_BLOCK_GROUP_RAID5) | |
3977 | allowed = BTRFS_BLOCK_GROUP_RAID5; | |
3978 | else if (allowed & BTRFS_BLOCK_GROUP_RAID10) | |
3979 | allowed = BTRFS_BLOCK_GROUP_RAID10; | |
3980 | else if (allowed & BTRFS_BLOCK_GROUP_RAID1) | |
3981 | allowed = BTRFS_BLOCK_GROUP_RAID1; | |
3982 | else if (allowed & BTRFS_BLOCK_GROUP_RAID0) | |
3983 | allowed = BTRFS_BLOCK_GROUP_RAID0; | |
3984 | ||
3985 | flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK; | |
3986 | ||
3987 | return extended_to_chunk(flags | allowed); | |
ec44a35c CM |
3988 | } |
3989 | ||
f8213bdc | 3990 | static u64 get_alloc_profile(struct btrfs_root *root, u64 orig_flags) |
6a63209f | 3991 | { |
de98ced9 | 3992 | unsigned seq; |
f8213bdc | 3993 | u64 flags; |
de98ced9 MX |
3994 | |
3995 | do { | |
f8213bdc | 3996 | flags = orig_flags; |
de98ced9 MX |
3997 | seq = read_seqbegin(&root->fs_info->profiles_lock); |
3998 | ||
3999 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
4000 | flags |= root->fs_info->avail_data_alloc_bits; | |
4001 | else if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
4002 | flags |= root->fs_info->avail_system_alloc_bits; | |
4003 | else if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
4004 | flags |= root->fs_info->avail_metadata_alloc_bits; | |
4005 | } while (read_seqretry(&root->fs_info->profiles_lock, seq)); | |
6fef8df1 | 4006 | |
b742bb82 | 4007 | return btrfs_reduce_alloc_profile(root, flags); |
6a63209f JB |
4008 | } |
4009 | ||
6d07bcec | 4010 | u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data) |
9ed74f2d | 4011 | { |
b742bb82 | 4012 | u64 flags; |
53b381b3 | 4013 | u64 ret; |
9ed74f2d | 4014 | |
b742bb82 YZ |
4015 | if (data) |
4016 | flags = BTRFS_BLOCK_GROUP_DATA; | |
4017 | else if (root == root->fs_info->chunk_root) | |
4018 | flags = BTRFS_BLOCK_GROUP_SYSTEM; | |
9ed74f2d | 4019 | else |
b742bb82 | 4020 | flags = BTRFS_BLOCK_GROUP_METADATA; |
9ed74f2d | 4021 | |
53b381b3 DW |
4022 | ret = get_alloc_profile(root, flags); |
4023 | return ret; | |
6a63209f | 4024 | } |
9ed74f2d | 4025 | |
4ceff079 | 4026 | int btrfs_alloc_data_chunk_ondemand(struct inode *inode, u64 bytes) |
6a63209f | 4027 | { |
6a63209f | 4028 | struct btrfs_space_info *data_sinfo; |
0ca1f7ce | 4029 | struct btrfs_root *root = BTRFS_I(inode)->root; |
b4d7c3c9 | 4030 | struct btrfs_fs_info *fs_info = root->fs_info; |
ab6e2410 | 4031 | u64 used; |
94b947b2 | 4032 | int ret = 0; |
c99f1b0c ZL |
4033 | int need_commit = 2; |
4034 | int have_pinned_space; | |
6a63209f | 4035 | |
6a63209f | 4036 | /* make sure bytes are sectorsize aligned */ |
fda2832f | 4037 | bytes = ALIGN(bytes, root->sectorsize); |
6a63209f | 4038 | |
9dced186 | 4039 | if (btrfs_is_free_space_inode(inode)) { |
c99f1b0c | 4040 | need_commit = 0; |
9dced186 | 4041 | ASSERT(current->journal_info); |
0af3d00b JB |
4042 | } |
4043 | ||
b4d7c3c9 | 4044 | data_sinfo = fs_info->data_sinfo; |
33b4d47f CM |
4045 | if (!data_sinfo) |
4046 | goto alloc; | |
9ed74f2d | 4047 | |
6a63209f JB |
4048 | again: |
4049 | /* make sure we have enough space to handle the data first */ | |
4050 | spin_lock(&data_sinfo->lock); | |
8929ecfa YZ |
4051 | used = data_sinfo->bytes_used + data_sinfo->bytes_reserved + |
4052 | data_sinfo->bytes_pinned + data_sinfo->bytes_readonly + | |
4053 | data_sinfo->bytes_may_use; | |
ab6e2410 JB |
4054 | |
4055 | if (used + bytes > data_sinfo->total_bytes) { | |
4e06bdd6 | 4056 | struct btrfs_trans_handle *trans; |
9ed74f2d | 4057 | |
6a63209f JB |
4058 | /* |
4059 | * if we don't have enough free bytes in this space then we need | |
4060 | * to alloc a new chunk. | |
4061 | */ | |
b9fd47cd | 4062 | if (!data_sinfo->full) { |
6a63209f | 4063 | u64 alloc_target; |
9ed74f2d | 4064 | |
0e4f8f88 | 4065 | data_sinfo->force_alloc = CHUNK_ALLOC_FORCE; |
6a63209f | 4066 | spin_unlock(&data_sinfo->lock); |
33b4d47f | 4067 | alloc: |
6a63209f | 4068 | alloc_target = btrfs_get_alloc_profile(root, 1); |
9dced186 MX |
4069 | /* |
4070 | * It is ugly that we don't call nolock join | |
4071 | * transaction for the free space inode case here. | |
4072 | * But it is safe because we only do the data space | |
4073 | * reservation for the free space cache in the | |
4074 | * transaction context, the common join transaction | |
4075 | * just increase the counter of the current transaction | |
4076 | * handler, doesn't try to acquire the trans_lock of | |
4077 | * the fs. | |
4078 | */ | |
7a7eaa40 | 4079 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
4080 | if (IS_ERR(trans)) |
4081 | return PTR_ERR(trans); | |
9ed74f2d | 4082 | |
6a63209f | 4083 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, |
0e4f8f88 CM |
4084 | alloc_target, |
4085 | CHUNK_ALLOC_NO_FORCE); | |
6a63209f | 4086 | btrfs_end_transaction(trans, root); |
d52a5b5f MX |
4087 | if (ret < 0) { |
4088 | if (ret != -ENOSPC) | |
4089 | return ret; | |
c99f1b0c ZL |
4090 | else { |
4091 | have_pinned_space = 1; | |
d52a5b5f | 4092 | goto commit_trans; |
c99f1b0c | 4093 | } |
d52a5b5f | 4094 | } |
9ed74f2d | 4095 | |
b4d7c3c9 LZ |
4096 | if (!data_sinfo) |
4097 | data_sinfo = fs_info->data_sinfo; | |
4098 | ||
6a63209f JB |
4099 | goto again; |
4100 | } | |
f2bb8f5c JB |
4101 | |
4102 | /* | |
b150a4f1 | 4103 | * If we don't have enough pinned space to deal with this |
94b947b2 ZL |
4104 | * allocation, and no removed chunk in current transaction, |
4105 | * don't bother committing the transaction. | |
f2bb8f5c | 4106 | */ |
c99f1b0c ZL |
4107 | have_pinned_space = percpu_counter_compare( |
4108 | &data_sinfo->total_bytes_pinned, | |
4109 | used + bytes - data_sinfo->total_bytes); | |
6a63209f | 4110 | spin_unlock(&data_sinfo->lock); |
6a63209f | 4111 | |
4e06bdd6 | 4112 | /* commit the current transaction and try again */ |
d52a5b5f | 4113 | commit_trans: |
c99f1b0c | 4114 | if (need_commit && |
a4abeea4 | 4115 | !atomic_read(&root->fs_info->open_ioctl_trans)) { |
c99f1b0c | 4116 | need_commit--; |
b150a4f1 | 4117 | |
9a4e7276 ZL |
4118 | if (need_commit > 0) |
4119 | btrfs_wait_ordered_roots(fs_info, -1); | |
4120 | ||
7a7eaa40 | 4121 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
4122 | if (IS_ERR(trans)) |
4123 | return PTR_ERR(trans); | |
c99f1b0c | 4124 | if (have_pinned_space >= 0 || |
3204d33c JB |
4125 | test_bit(BTRFS_TRANS_HAVE_FREE_BGS, |
4126 | &trans->transaction->flags) || | |
c99f1b0c | 4127 | need_commit > 0) { |
94b947b2 ZL |
4128 | ret = btrfs_commit_transaction(trans, root); |
4129 | if (ret) | |
4130 | return ret; | |
d7c15171 ZL |
4131 | /* |
4132 | * make sure that all running delayed iput are | |
4133 | * done | |
4134 | */ | |
4135 | down_write(&root->fs_info->delayed_iput_sem); | |
4136 | up_write(&root->fs_info->delayed_iput_sem); | |
94b947b2 ZL |
4137 | goto again; |
4138 | } else { | |
4139 | btrfs_end_transaction(trans, root); | |
4140 | } | |
4e06bdd6 | 4141 | } |
9ed74f2d | 4142 | |
cab45e22 JM |
4143 | trace_btrfs_space_reservation(root->fs_info, |
4144 | "space_info:enospc", | |
4145 | data_sinfo->flags, bytes, 1); | |
6a63209f JB |
4146 | return -ENOSPC; |
4147 | } | |
4148 | data_sinfo->bytes_may_use += bytes; | |
8c2a3ca2 | 4149 | trace_btrfs_space_reservation(root->fs_info, "space_info", |
2bcc0328 | 4150 | data_sinfo->flags, bytes, 1); |
6a63209f | 4151 | spin_unlock(&data_sinfo->lock); |
6a63209f | 4152 | |
237c0e9f | 4153 | return ret; |
9ed74f2d | 4154 | } |
6a63209f | 4155 | |
4ceff079 QW |
4156 | /* |
4157 | * New check_data_free_space() with ability for precious data reservation | |
4158 | * Will replace old btrfs_check_data_free_space(), but for patch split, | |
4159 | * add a new function first and then replace it. | |
4160 | */ | |
7cf5b976 | 4161 | int btrfs_check_data_free_space(struct inode *inode, u64 start, u64 len) |
4ceff079 QW |
4162 | { |
4163 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4164 | int ret; | |
4165 | ||
4166 | /* align the range */ | |
4167 | len = round_up(start + len, root->sectorsize) - | |
4168 | round_down(start, root->sectorsize); | |
4169 | start = round_down(start, root->sectorsize); | |
4170 | ||
4171 | ret = btrfs_alloc_data_chunk_ondemand(inode, len); | |
4172 | if (ret < 0) | |
4173 | return ret; | |
4174 | ||
94ed938a QW |
4175 | /* |
4176 | * Use new btrfs_qgroup_reserve_data to reserve precious data space | |
4177 | * | |
4178 | * TODO: Find a good method to avoid reserve data space for NOCOW | |
4179 | * range, but don't impact performance on quota disable case. | |
4180 | */ | |
4ceff079 QW |
4181 | ret = btrfs_qgroup_reserve_data(inode, start, len); |
4182 | return ret; | |
4183 | } | |
4184 | ||
4ceff079 QW |
4185 | /* |
4186 | * Called if we need to clear a data reservation for this inode | |
4187 | * Normally in a error case. | |
4188 | * | |
51773bec QW |
4189 | * This one will *NOT* use accurate qgroup reserved space API, just for case |
4190 | * which we can't sleep and is sure it won't affect qgroup reserved space. | |
4191 | * Like clear_bit_hook(). | |
4ceff079 | 4192 | */ |
51773bec QW |
4193 | void btrfs_free_reserved_data_space_noquota(struct inode *inode, u64 start, |
4194 | u64 len) | |
4ceff079 QW |
4195 | { |
4196 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4197 | struct btrfs_space_info *data_sinfo; | |
4198 | ||
4199 | /* Make sure the range is aligned to sectorsize */ | |
4200 | len = round_up(start + len, root->sectorsize) - | |
4201 | round_down(start, root->sectorsize); | |
4202 | start = round_down(start, root->sectorsize); | |
4203 | ||
4ceff079 QW |
4204 | data_sinfo = root->fs_info->data_sinfo; |
4205 | spin_lock(&data_sinfo->lock); | |
4206 | if (WARN_ON(data_sinfo->bytes_may_use < len)) | |
4207 | data_sinfo->bytes_may_use = 0; | |
4208 | else | |
4209 | data_sinfo->bytes_may_use -= len; | |
4210 | trace_btrfs_space_reservation(root->fs_info, "space_info", | |
4211 | data_sinfo->flags, len, 0); | |
4212 | spin_unlock(&data_sinfo->lock); | |
4213 | } | |
4214 | ||
51773bec QW |
4215 | /* |
4216 | * Called if we need to clear a data reservation for this inode | |
4217 | * Normally in a error case. | |
4218 | * | |
4219 | * This one will handle the per-indoe data rsv map for accurate reserved | |
4220 | * space framework. | |
4221 | */ | |
4222 | void btrfs_free_reserved_data_space(struct inode *inode, u64 start, u64 len) | |
4223 | { | |
4224 | btrfs_free_reserved_data_space_noquota(inode, start, len); | |
4225 | btrfs_qgroup_free_data(inode, start, len); | |
4226 | } | |
4227 | ||
97e728d4 | 4228 | static void force_metadata_allocation(struct btrfs_fs_info *info) |
e3ccfa98 | 4229 | { |
97e728d4 JB |
4230 | struct list_head *head = &info->space_info; |
4231 | struct btrfs_space_info *found; | |
e3ccfa98 | 4232 | |
97e728d4 JB |
4233 | rcu_read_lock(); |
4234 | list_for_each_entry_rcu(found, head, list) { | |
4235 | if (found->flags & BTRFS_BLOCK_GROUP_METADATA) | |
0e4f8f88 | 4236 | found->force_alloc = CHUNK_ALLOC_FORCE; |
e3ccfa98 | 4237 | } |
97e728d4 | 4238 | rcu_read_unlock(); |
e3ccfa98 JB |
4239 | } |
4240 | ||
3c76cd84 MX |
4241 | static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global) |
4242 | { | |
4243 | return (global->size << 1); | |
4244 | } | |
4245 | ||
e5bc2458 | 4246 | static int should_alloc_chunk(struct btrfs_root *root, |
698d0082 | 4247 | struct btrfs_space_info *sinfo, int force) |
32c00aff | 4248 | { |
fb25e914 | 4249 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; |
424499db | 4250 | u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly; |
0e4f8f88 | 4251 | u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved; |
e5bc2458 | 4252 | u64 thresh; |
e3ccfa98 | 4253 | |
0e4f8f88 CM |
4254 | if (force == CHUNK_ALLOC_FORCE) |
4255 | return 1; | |
4256 | ||
fb25e914 JB |
4257 | /* |
4258 | * We need to take into account the global rsv because for all intents | |
4259 | * and purposes it's used space. Don't worry about locking the | |
4260 | * global_rsv, it doesn't change except when the transaction commits. | |
4261 | */ | |
54338b5c | 4262 | if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA) |
3c76cd84 | 4263 | num_allocated += calc_global_rsv_need_space(global_rsv); |
fb25e914 | 4264 | |
0e4f8f88 CM |
4265 | /* |
4266 | * in limited mode, we want to have some free space up to | |
4267 | * about 1% of the FS size. | |
4268 | */ | |
4269 | if (force == CHUNK_ALLOC_LIMITED) { | |
6c41761f | 4270 | thresh = btrfs_super_total_bytes(root->fs_info->super_copy); |
0e4f8f88 CM |
4271 | thresh = max_t(u64, 64 * 1024 * 1024, |
4272 | div_factor_fine(thresh, 1)); | |
4273 | ||
4274 | if (num_bytes - num_allocated < thresh) | |
4275 | return 1; | |
4276 | } | |
0e4f8f88 | 4277 | |
698d0082 | 4278 | if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8)) |
14ed0ca6 | 4279 | return 0; |
424499db | 4280 | return 1; |
32c00aff JB |
4281 | } |
4282 | ||
39c2d7fa | 4283 | static u64 get_profile_num_devs(struct btrfs_root *root, u64 type) |
15d1ff81 LB |
4284 | { |
4285 | u64 num_dev; | |
4286 | ||
53b381b3 DW |
4287 | if (type & (BTRFS_BLOCK_GROUP_RAID10 | |
4288 | BTRFS_BLOCK_GROUP_RAID0 | | |
4289 | BTRFS_BLOCK_GROUP_RAID5 | | |
4290 | BTRFS_BLOCK_GROUP_RAID6)) | |
15d1ff81 LB |
4291 | num_dev = root->fs_info->fs_devices->rw_devices; |
4292 | else if (type & BTRFS_BLOCK_GROUP_RAID1) | |
4293 | num_dev = 2; | |
4294 | else | |
4295 | num_dev = 1; /* DUP or single */ | |
4296 | ||
39c2d7fa | 4297 | return num_dev; |
15d1ff81 LB |
4298 | } |
4299 | ||
39c2d7fa FM |
4300 | /* |
4301 | * If @is_allocation is true, reserve space in the system space info necessary | |
4302 | * for allocating a chunk, otherwise if it's false, reserve space necessary for | |
4303 | * removing a chunk. | |
4304 | */ | |
4305 | void check_system_chunk(struct btrfs_trans_handle *trans, | |
4306 | struct btrfs_root *root, | |
4617ea3a | 4307 | u64 type) |
15d1ff81 LB |
4308 | { |
4309 | struct btrfs_space_info *info; | |
4310 | u64 left; | |
4311 | u64 thresh; | |
4fbcdf66 | 4312 | int ret = 0; |
39c2d7fa | 4313 | u64 num_devs; |
4fbcdf66 FM |
4314 | |
4315 | /* | |
4316 | * Needed because we can end up allocating a system chunk and for an | |
4317 | * atomic and race free space reservation in the chunk block reserve. | |
4318 | */ | |
4319 | ASSERT(mutex_is_locked(&root->fs_info->chunk_mutex)); | |
15d1ff81 LB |
4320 | |
4321 | info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM); | |
4322 | spin_lock(&info->lock); | |
4323 | left = info->total_bytes - info->bytes_used - info->bytes_pinned - | |
4fbcdf66 FM |
4324 | info->bytes_reserved - info->bytes_readonly - |
4325 | info->bytes_may_use; | |
15d1ff81 LB |
4326 | spin_unlock(&info->lock); |
4327 | ||
39c2d7fa FM |
4328 | num_devs = get_profile_num_devs(root, type); |
4329 | ||
4330 | /* num_devs device items to update and 1 chunk item to add or remove */ | |
4617ea3a FM |
4331 | thresh = btrfs_calc_trunc_metadata_size(root, num_devs) + |
4332 | btrfs_calc_trans_metadata_size(root, 1); | |
39c2d7fa | 4333 | |
15d1ff81 | 4334 | if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) { |
c2cf52eb SK |
4335 | btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu", |
4336 | left, thresh, type); | |
15d1ff81 LB |
4337 | dump_space_info(info, 0, 0); |
4338 | } | |
4339 | ||
4340 | if (left < thresh) { | |
4341 | u64 flags; | |
4342 | ||
4343 | flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0); | |
4fbcdf66 FM |
4344 | /* |
4345 | * Ignore failure to create system chunk. We might end up not | |
4346 | * needing it, as we might not need to COW all nodes/leafs from | |
4347 | * the paths we visit in the chunk tree (they were already COWed | |
4348 | * or created in the current transaction for example). | |
4349 | */ | |
4350 | ret = btrfs_alloc_chunk(trans, root, flags); | |
4351 | } | |
4352 | ||
4353 | if (!ret) { | |
4354 | ret = btrfs_block_rsv_add(root->fs_info->chunk_root, | |
4355 | &root->fs_info->chunk_block_rsv, | |
4356 | thresh, BTRFS_RESERVE_NO_FLUSH); | |
4357 | if (!ret) | |
4358 | trans->chunk_bytes_reserved += thresh; | |
15d1ff81 LB |
4359 | } |
4360 | } | |
4361 | ||
6324fbf3 | 4362 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
698d0082 | 4363 | struct btrfs_root *extent_root, u64 flags, int force) |
9ed74f2d | 4364 | { |
6324fbf3 | 4365 | struct btrfs_space_info *space_info; |
97e728d4 | 4366 | struct btrfs_fs_info *fs_info = extent_root->fs_info; |
6d74119f | 4367 | int wait_for_alloc = 0; |
9ed74f2d | 4368 | int ret = 0; |
9ed74f2d | 4369 | |
c6b305a8 JB |
4370 | /* Don't re-enter if we're already allocating a chunk */ |
4371 | if (trans->allocating_chunk) | |
4372 | return -ENOSPC; | |
4373 | ||
6324fbf3 | 4374 | space_info = __find_space_info(extent_root->fs_info, flags); |
593060d7 CM |
4375 | if (!space_info) { |
4376 | ret = update_space_info(extent_root->fs_info, flags, | |
4377 | 0, 0, &space_info); | |
79787eaa | 4378 | BUG_ON(ret); /* -ENOMEM */ |
9ed74f2d | 4379 | } |
79787eaa | 4380 | BUG_ON(!space_info); /* Logic error */ |
9ed74f2d | 4381 | |
6d74119f | 4382 | again: |
25179201 | 4383 | spin_lock(&space_info->lock); |
9e622d6b | 4384 | if (force < space_info->force_alloc) |
0e4f8f88 | 4385 | force = space_info->force_alloc; |
25179201 | 4386 | if (space_info->full) { |
09fb99a6 FDBM |
4387 | if (should_alloc_chunk(extent_root, space_info, force)) |
4388 | ret = -ENOSPC; | |
4389 | else | |
4390 | ret = 0; | |
25179201 | 4391 | spin_unlock(&space_info->lock); |
09fb99a6 | 4392 | return ret; |
9ed74f2d JB |
4393 | } |
4394 | ||
698d0082 | 4395 | if (!should_alloc_chunk(extent_root, space_info, force)) { |
25179201 | 4396 | spin_unlock(&space_info->lock); |
6d74119f JB |
4397 | return 0; |
4398 | } else if (space_info->chunk_alloc) { | |
4399 | wait_for_alloc = 1; | |
4400 | } else { | |
4401 | space_info->chunk_alloc = 1; | |
9ed74f2d | 4402 | } |
0e4f8f88 | 4403 | |
25179201 | 4404 | spin_unlock(&space_info->lock); |
9ed74f2d | 4405 | |
6d74119f JB |
4406 | mutex_lock(&fs_info->chunk_mutex); |
4407 | ||
4408 | /* | |
4409 | * The chunk_mutex is held throughout the entirety of a chunk | |
4410 | * allocation, so once we've acquired the chunk_mutex we know that the | |
4411 | * other guy is done and we need to recheck and see if we should | |
4412 | * allocate. | |
4413 | */ | |
4414 | if (wait_for_alloc) { | |
4415 | mutex_unlock(&fs_info->chunk_mutex); | |
4416 | wait_for_alloc = 0; | |
4417 | goto again; | |
4418 | } | |
4419 | ||
c6b305a8 JB |
4420 | trans->allocating_chunk = true; |
4421 | ||
67377734 JB |
4422 | /* |
4423 | * If we have mixed data/metadata chunks we want to make sure we keep | |
4424 | * allocating mixed chunks instead of individual chunks. | |
4425 | */ | |
4426 | if (btrfs_mixed_space_info(space_info)) | |
4427 | flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA); | |
4428 | ||
97e728d4 JB |
4429 | /* |
4430 | * if we're doing a data chunk, go ahead and make sure that | |
4431 | * we keep a reasonable number of metadata chunks allocated in the | |
4432 | * FS as well. | |
4433 | */ | |
9ed74f2d | 4434 | if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) { |
97e728d4 JB |
4435 | fs_info->data_chunk_allocations++; |
4436 | if (!(fs_info->data_chunk_allocations % | |
4437 | fs_info->metadata_ratio)) | |
4438 | force_metadata_allocation(fs_info); | |
9ed74f2d JB |
4439 | } |
4440 | ||
15d1ff81 LB |
4441 | /* |
4442 | * Check if we have enough space in SYSTEM chunk because we may need | |
4443 | * to update devices. | |
4444 | */ | |
4617ea3a | 4445 | check_system_chunk(trans, extent_root, flags); |
15d1ff81 | 4446 | |
2b82032c | 4447 | ret = btrfs_alloc_chunk(trans, extent_root, flags); |
c6b305a8 | 4448 | trans->allocating_chunk = false; |
92b8e897 | 4449 | |
9ed74f2d | 4450 | spin_lock(&space_info->lock); |
a81cb9a2 AO |
4451 | if (ret < 0 && ret != -ENOSPC) |
4452 | goto out; | |
9ed74f2d | 4453 | if (ret) |
6324fbf3 | 4454 | space_info->full = 1; |
424499db YZ |
4455 | else |
4456 | ret = 1; | |
6d74119f | 4457 | |
0e4f8f88 | 4458 | space_info->force_alloc = CHUNK_ALLOC_NO_FORCE; |
a81cb9a2 | 4459 | out: |
6d74119f | 4460 | space_info->chunk_alloc = 0; |
9ed74f2d | 4461 | spin_unlock(&space_info->lock); |
a25c75d5 | 4462 | mutex_unlock(&fs_info->chunk_mutex); |
00d80e34 FM |
4463 | /* |
4464 | * When we allocate a new chunk we reserve space in the chunk block | |
4465 | * reserve to make sure we can COW nodes/leafs in the chunk tree or | |
4466 | * add new nodes/leafs to it if we end up needing to do it when | |
4467 | * inserting the chunk item and updating device items as part of the | |
4468 | * second phase of chunk allocation, performed by | |
4469 | * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a | |
4470 | * large number of new block groups to create in our transaction | |
4471 | * handle's new_bgs list to avoid exhausting the chunk block reserve | |
4472 | * in extreme cases - like having a single transaction create many new | |
4473 | * block groups when starting to write out the free space caches of all | |
4474 | * the block groups that were made dirty during the lifetime of the | |
4475 | * transaction. | |
4476 | */ | |
d9a0540a FM |
4477 | if (trans->can_flush_pending_bgs && |
4478 | trans->chunk_bytes_reserved >= (2 * 1024 * 1024ull)) { | |
00d80e34 FM |
4479 | btrfs_create_pending_block_groups(trans, trans->root); |
4480 | btrfs_trans_release_chunk_metadata(trans); | |
4481 | } | |
0f9dd46c | 4482 | return ret; |
6324fbf3 | 4483 | } |
9ed74f2d | 4484 | |
a80c8dcf JB |
4485 | static int can_overcommit(struct btrfs_root *root, |
4486 | struct btrfs_space_info *space_info, u64 bytes, | |
08e007d2 | 4487 | enum btrfs_reserve_flush_enum flush) |
a80c8dcf | 4488 | { |
96f1bb57 | 4489 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; |
a80c8dcf | 4490 | u64 profile = btrfs_get_alloc_profile(root, 0); |
3c76cd84 | 4491 | u64 space_size; |
a80c8dcf JB |
4492 | u64 avail; |
4493 | u64 used; | |
4494 | ||
4495 | used = space_info->bytes_used + space_info->bytes_reserved + | |
96f1bb57 JB |
4496 | space_info->bytes_pinned + space_info->bytes_readonly; |
4497 | ||
96f1bb57 JB |
4498 | /* |
4499 | * We only want to allow over committing if we have lots of actual space | |
4500 | * free, but if we don't have enough space to handle the global reserve | |
4501 | * space then we could end up having a real enospc problem when trying | |
4502 | * to allocate a chunk or some other such important allocation. | |
4503 | */ | |
3c76cd84 MX |
4504 | spin_lock(&global_rsv->lock); |
4505 | space_size = calc_global_rsv_need_space(global_rsv); | |
4506 | spin_unlock(&global_rsv->lock); | |
4507 | if (used + space_size >= space_info->total_bytes) | |
96f1bb57 JB |
4508 | return 0; |
4509 | ||
4510 | used += space_info->bytes_may_use; | |
a80c8dcf JB |
4511 | |
4512 | spin_lock(&root->fs_info->free_chunk_lock); | |
4513 | avail = root->fs_info->free_chunk_space; | |
4514 | spin_unlock(&root->fs_info->free_chunk_lock); | |
4515 | ||
4516 | /* | |
4517 | * If we have dup, raid1 or raid10 then only half of the free | |
53b381b3 DW |
4518 | * space is actually useable. For raid56, the space info used |
4519 | * doesn't include the parity drive, so we don't have to | |
4520 | * change the math | |
a80c8dcf JB |
4521 | */ |
4522 | if (profile & (BTRFS_BLOCK_GROUP_DUP | | |
4523 | BTRFS_BLOCK_GROUP_RAID1 | | |
4524 | BTRFS_BLOCK_GROUP_RAID10)) | |
4525 | avail >>= 1; | |
4526 | ||
4527 | /* | |
561c294d MX |
4528 | * If we aren't flushing all things, let us overcommit up to |
4529 | * 1/2th of the space. If we can flush, don't let us overcommit | |
4530 | * too much, let it overcommit up to 1/8 of the space. | |
a80c8dcf | 4531 | */ |
08e007d2 | 4532 | if (flush == BTRFS_RESERVE_FLUSH_ALL) |
14575aef | 4533 | avail >>= 3; |
a80c8dcf | 4534 | else |
14575aef | 4535 | avail >>= 1; |
a80c8dcf | 4536 | |
14575aef | 4537 | if (used + bytes < space_info->total_bytes + avail) |
a80c8dcf JB |
4538 | return 1; |
4539 | return 0; | |
4540 | } | |
4541 | ||
48a3b636 | 4542 | static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root, |
6c255e67 | 4543 | unsigned long nr_pages, int nr_items) |
da633a42 MX |
4544 | { |
4545 | struct super_block *sb = root->fs_info->sb; | |
da633a42 | 4546 | |
925a6efb JB |
4547 | if (down_read_trylock(&sb->s_umount)) { |
4548 | writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE); | |
4549 | up_read(&sb->s_umount); | |
4550 | } else { | |
da633a42 MX |
4551 | /* |
4552 | * We needn't worry the filesystem going from r/w to r/o though | |
4553 | * we don't acquire ->s_umount mutex, because the filesystem | |
4554 | * should guarantee the delalloc inodes list be empty after | |
4555 | * the filesystem is readonly(all dirty pages are written to | |
4556 | * the disk). | |
4557 | */ | |
6c255e67 | 4558 | btrfs_start_delalloc_roots(root->fs_info, 0, nr_items); |
98ad69cf | 4559 | if (!current->journal_info) |
6c255e67 | 4560 | btrfs_wait_ordered_roots(root->fs_info, nr_items); |
da633a42 MX |
4561 | } |
4562 | } | |
4563 | ||
18cd8ea6 MX |
4564 | static inline int calc_reclaim_items_nr(struct btrfs_root *root, u64 to_reclaim) |
4565 | { | |
4566 | u64 bytes; | |
4567 | int nr; | |
4568 | ||
4569 | bytes = btrfs_calc_trans_metadata_size(root, 1); | |
4570 | nr = (int)div64_u64(to_reclaim, bytes); | |
4571 | if (!nr) | |
4572 | nr = 1; | |
4573 | return nr; | |
4574 | } | |
4575 | ||
c61a16a7 MX |
4576 | #define EXTENT_SIZE_PER_ITEM (256 * 1024) |
4577 | ||
9ed74f2d | 4578 | /* |
5da9d01b | 4579 | * shrink metadata reservation for delalloc |
9ed74f2d | 4580 | */ |
f4c738c2 JB |
4581 | static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig, |
4582 | bool wait_ordered) | |
5da9d01b | 4583 | { |
0ca1f7ce | 4584 | struct btrfs_block_rsv *block_rsv; |
0019f10d | 4585 | struct btrfs_space_info *space_info; |
663350ac | 4586 | struct btrfs_trans_handle *trans; |
f4c738c2 | 4587 | u64 delalloc_bytes; |
5da9d01b | 4588 | u64 max_reclaim; |
b1953bce | 4589 | long time_left; |
d3ee29e3 MX |
4590 | unsigned long nr_pages; |
4591 | int loops; | |
b0244199 | 4592 | int items; |
08e007d2 | 4593 | enum btrfs_reserve_flush_enum flush; |
5da9d01b | 4594 | |
c61a16a7 | 4595 | /* Calc the number of the pages we need flush for space reservation */ |
b0244199 MX |
4596 | items = calc_reclaim_items_nr(root, to_reclaim); |
4597 | to_reclaim = items * EXTENT_SIZE_PER_ITEM; | |
c61a16a7 | 4598 | |
663350ac | 4599 | trans = (struct btrfs_trans_handle *)current->journal_info; |
0ca1f7ce | 4600 | block_rsv = &root->fs_info->delalloc_block_rsv; |
0019f10d | 4601 | space_info = block_rsv->space_info; |
bf9022e0 | 4602 | |
963d678b MX |
4603 | delalloc_bytes = percpu_counter_sum_positive( |
4604 | &root->fs_info->delalloc_bytes); | |
f4c738c2 | 4605 | if (delalloc_bytes == 0) { |
fdb5effd | 4606 | if (trans) |
f4c738c2 | 4607 | return; |
38c135af | 4608 | if (wait_ordered) |
b0244199 | 4609 | btrfs_wait_ordered_roots(root->fs_info, items); |
f4c738c2 | 4610 | return; |
fdb5effd JB |
4611 | } |
4612 | ||
d3ee29e3 | 4613 | loops = 0; |
f4c738c2 JB |
4614 | while (delalloc_bytes && loops < 3) { |
4615 | max_reclaim = min(delalloc_bytes, to_reclaim); | |
4616 | nr_pages = max_reclaim >> PAGE_CACHE_SHIFT; | |
6c255e67 | 4617 | btrfs_writeback_inodes_sb_nr(root, nr_pages, items); |
dea31f52 JB |
4618 | /* |
4619 | * We need to wait for the async pages to actually start before | |
4620 | * we do anything. | |
4621 | */ | |
9f3a074d MX |
4622 | max_reclaim = atomic_read(&root->fs_info->async_delalloc_pages); |
4623 | if (!max_reclaim) | |
4624 | goto skip_async; | |
4625 | ||
4626 | if (max_reclaim <= nr_pages) | |
4627 | max_reclaim = 0; | |
4628 | else | |
4629 | max_reclaim -= nr_pages; | |
dea31f52 | 4630 | |
9f3a074d MX |
4631 | wait_event(root->fs_info->async_submit_wait, |
4632 | atomic_read(&root->fs_info->async_delalloc_pages) <= | |
4633 | (int)max_reclaim); | |
4634 | skip_async: | |
08e007d2 MX |
4635 | if (!trans) |
4636 | flush = BTRFS_RESERVE_FLUSH_ALL; | |
4637 | else | |
4638 | flush = BTRFS_RESERVE_NO_FLUSH; | |
0019f10d | 4639 | spin_lock(&space_info->lock); |
08e007d2 | 4640 | if (can_overcommit(root, space_info, orig, flush)) { |
f4c738c2 JB |
4641 | spin_unlock(&space_info->lock); |
4642 | break; | |
4643 | } | |
0019f10d | 4644 | spin_unlock(&space_info->lock); |
5da9d01b | 4645 | |
36e39c40 | 4646 | loops++; |
f104d044 | 4647 | if (wait_ordered && !trans) { |
b0244199 | 4648 | btrfs_wait_ordered_roots(root->fs_info, items); |
f104d044 | 4649 | } else { |
f4c738c2 | 4650 | time_left = schedule_timeout_killable(1); |
f104d044 JB |
4651 | if (time_left) |
4652 | break; | |
4653 | } | |
963d678b MX |
4654 | delalloc_bytes = percpu_counter_sum_positive( |
4655 | &root->fs_info->delalloc_bytes); | |
5da9d01b | 4656 | } |
5da9d01b YZ |
4657 | } |
4658 | ||
663350ac JB |
4659 | /** |
4660 | * maybe_commit_transaction - possibly commit the transaction if its ok to | |
4661 | * @root - the root we're allocating for | |
4662 | * @bytes - the number of bytes we want to reserve | |
4663 | * @force - force the commit | |
8bb8ab2e | 4664 | * |
663350ac JB |
4665 | * This will check to make sure that committing the transaction will actually |
4666 | * get us somewhere and then commit the transaction if it does. Otherwise it | |
4667 | * will return -ENOSPC. | |
8bb8ab2e | 4668 | */ |
663350ac JB |
4669 | static int may_commit_transaction(struct btrfs_root *root, |
4670 | struct btrfs_space_info *space_info, | |
4671 | u64 bytes, int force) | |
4672 | { | |
4673 | struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv; | |
4674 | struct btrfs_trans_handle *trans; | |
4675 | ||
4676 | trans = (struct btrfs_trans_handle *)current->journal_info; | |
4677 | if (trans) | |
4678 | return -EAGAIN; | |
4679 | ||
4680 | if (force) | |
4681 | goto commit; | |
4682 | ||
4683 | /* See if there is enough pinned space to make this reservation */ | |
b150a4f1 | 4684 | if (percpu_counter_compare(&space_info->total_bytes_pinned, |
0424c548 | 4685 | bytes) >= 0) |
663350ac | 4686 | goto commit; |
663350ac JB |
4687 | |
4688 | /* | |
4689 | * See if there is some space in the delayed insertion reservation for | |
4690 | * this reservation. | |
4691 | */ | |
4692 | if (space_info != delayed_rsv->space_info) | |
4693 | return -ENOSPC; | |
4694 | ||
4695 | spin_lock(&delayed_rsv->lock); | |
b150a4f1 JB |
4696 | if (percpu_counter_compare(&space_info->total_bytes_pinned, |
4697 | bytes - delayed_rsv->size) >= 0) { | |
663350ac JB |
4698 | spin_unlock(&delayed_rsv->lock); |
4699 | return -ENOSPC; | |
4700 | } | |
4701 | spin_unlock(&delayed_rsv->lock); | |
4702 | ||
4703 | commit: | |
4704 | trans = btrfs_join_transaction(root); | |
4705 | if (IS_ERR(trans)) | |
4706 | return -ENOSPC; | |
4707 | ||
4708 | return btrfs_commit_transaction(trans, root); | |
4709 | } | |
4710 | ||
96c3f433 | 4711 | enum flush_state { |
67b0fd63 JB |
4712 | FLUSH_DELAYED_ITEMS_NR = 1, |
4713 | FLUSH_DELAYED_ITEMS = 2, | |
4714 | FLUSH_DELALLOC = 3, | |
4715 | FLUSH_DELALLOC_WAIT = 4, | |
ea658bad JB |
4716 | ALLOC_CHUNK = 5, |
4717 | COMMIT_TRANS = 6, | |
96c3f433 JB |
4718 | }; |
4719 | ||
4720 | static int flush_space(struct btrfs_root *root, | |
4721 | struct btrfs_space_info *space_info, u64 num_bytes, | |
4722 | u64 orig_bytes, int state) | |
4723 | { | |
4724 | struct btrfs_trans_handle *trans; | |
4725 | int nr; | |
f4c738c2 | 4726 | int ret = 0; |
96c3f433 JB |
4727 | |
4728 | switch (state) { | |
96c3f433 JB |
4729 | case FLUSH_DELAYED_ITEMS_NR: |
4730 | case FLUSH_DELAYED_ITEMS: | |
18cd8ea6 MX |
4731 | if (state == FLUSH_DELAYED_ITEMS_NR) |
4732 | nr = calc_reclaim_items_nr(root, num_bytes) * 2; | |
4733 | else | |
96c3f433 | 4734 | nr = -1; |
18cd8ea6 | 4735 | |
96c3f433 JB |
4736 | trans = btrfs_join_transaction(root); |
4737 | if (IS_ERR(trans)) { | |
4738 | ret = PTR_ERR(trans); | |
4739 | break; | |
4740 | } | |
4741 | ret = btrfs_run_delayed_items_nr(trans, root, nr); | |
4742 | btrfs_end_transaction(trans, root); | |
4743 | break; | |
67b0fd63 JB |
4744 | case FLUSH_DELALLOC: |
4745 | case FLUSH_DELALLOC_WAIT: | |
24af7dd1 | 4746 | shrink_delalloc(root, num_bytes * 2, orig_bytes, |
67b0fd63 JB |
4747 | state == FLUSH_DELALLOC_WAIT); |
4748 | break; | |
ea658bad JB |
4749 | case ALLOC_CHUNK: |
4750 | trans = btrfs_join_transaction(root); | |
4751 | if (IS_ERR(trans)) { | |
4752 | ret = PTR_ERR(trans); | |
4753 | break; | |
4754 | } | |
4755 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, | |
ea658bad JB |
4756 | btrfs_get_alloc_profile(root, 0), |
4757 | CHUNK_ALLOC_NO_FORCE); | |
4758 | btrfs_end_transaction(trans, root); | |
4759 | if (ret == -ENOSPC) | |
4760 | ret = 0; | |
4761 | break; | |
96c3f433 JB |
4762 | case COMMIT_TRANS: |
4763 | ret = may_commit_transaction(root, space_info, orig_bytes, 0); | |
4764 | break; | |
4765 | default: | |
4766 | ret = -ENOSPC; | |
4767 | break; | |
4768 | } | |
4769 | ||
4770 | return ret; | |
4771 | } | |
21c7e756 MX |
4772 | |
4773 | static inline u64 | |
4774 | btrfs_calc_reclaim_metadata_size(struct btrfs_root *root, | |
4775 | struct btrfs_space_info *space_info) | |
4776 | { | |
4777 | u64 used; | |
4778 | u64 expected; | |
4779 | u64 to_reclaim; | |
4780 | ||
4781 | to_reclaim = min_t(u64, num_online_cpus() * 1024 * 1024, | |
4782 | 16 * 1024 * 1024); | |
4783 | spin_lock(&space_info->lock); | |
4784 | if (can_overcommit(root, space_info, to_reclaim, | |
4785 | BTRFS_RESERVE_FLUSH_ALL)) { | |
4786 | to_reclaim = 0; | |
4787 | goto out; | |
4788 | } | |
4789 | ||
4790 | used = space_info->bytes_used + space_info->bytes_reserved + | |
4791 | space_info->bytes_pinned + space_info->bytes_readonly + | |
4792 | space_info->bytes_may_use; | |
4793 | if (can_overcommit(root, space_info, 1024 * 1024, | |
4794 | BTRFS_RESERVE_FLUSH_ALL)) | |
4795 | expected = div_factor_fine(space_info->total_bytes, 95); | |
4796 | else | |
4797 | expected = div_factor_fine(space_info->total_bytes, 90); | |
4798 | ||
4799 | if (used > expected) | |
4800 | to_reclaim = used - expected; | |
4801 | else | |
4802 | to_reclaim = 0; | |
4803 | to_reclaim = min(to_reclaim, space_info->bytes_may_use + | |
4804 | space_info->bytes_reserved); | |
4805 | out: | |
4806 | spin_unlock(&space_info->lock); | |
4807 | ||
4808 | return to_reclaim; | |
4809 | } | |
4810 | ||
4811 | static inline int need_do_async_reclaim(struct btrfs_space_info *space_info, | |
4812 | struct btrfs_fs_info *fs_info, u64 used) | |
4813 | { | |
365c5313 JB |
4814 | u64 thresh = div_factor_fine(space_info->total_bytes, 98); |
4815 | ||
4816 | /* If we're just plain full then async reclaim just slows us down. */ | |
4817 | if (space_info->bytes_used >= thresh) | |
4818 | return 0; | |
4819 | ||
4820 | return (used >= thresh && !btrfs_fs_closing(fs_info) && | |
21c7e756 MX |
4821 | !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state)); |
4822 | } | |
4823 | ||
4824 | static int btrfs_need_do_async_reclaim(struct btrfs_space_info *space_info, | |
25ce459c LB |
4825 | struct btrfs_fs_info *fs_info, |
4826 | int flush_state) | |
21c7e756 MX |
4827 | { |
4828 | u64 used; | |
4829 | ||
4830 | spin_lock(&space_info->lock); | |
25ce459c LB |
4831 | /* |
4832 | * We run out of space and have not got any free space via flush_space, | |
4833 | * so don't bother doing async reclaim. | |
4834 | */ | |
4835 | if (flush_state > COMMIT_TRANS && space_info->full) { | |
4836 | spin_unlock(&space_info->lock); | |
4837 | return 0; | |
4838 | } | |
4839 | ||
21c7e756 MX |
4840 | used = space_info->bytes_used + space_info->bytes_reserved + |
4841 | space_info->bytes_pinned + space_info->bytes_readonly + | |
4842 | space_info->bytes_may_use; | |
4843 | if (need_do_async_reclaim(space_info, fs_info, used)) { | |
4844 | spin_unlock(&space_info->lock); | |
4845 | return 1; | |
4846 | } | |
4847 | spin_unlock(&space_info->lock); | |
4848 | ||
4849 | return 0; | |
4850 | } | |
4851 | ||
4852 | static void btrfs_async_reclaim_metadata_space(struct work_struct *work) | |
4853 | { | |
4854 | struct btrfs_fs_info *fs_info; | |
4855 | struct btrfs_space_info *space_info; | |
4856 | u64 to_reclaim; | |
4857 | int flush_state; | |
4858 | ||
4859 | fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work); | |
4860 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); | |
4861 | ||
4862 | to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root, | |
4863 | space_info); | |
4864 | if (!to_reclaim) | |
4865 | return; | |
4866 | ||
4867 | flush_state = FLUSH_DELAYED_ITEMS_NR; | |
4868 | do { | |
4869 | flush_space(fs_info->fs_root, space_info, to_reclaim, | |
4870 | to_reclaim, flush_state); | |
4871 | flush_state++; | |
25ce459c LB |
4872 | if (!btrfs_need_do_async_reclaim(space_info, fs_info, |
4873 | flush_state)) | |
21c7e756 | 4874 | return; |
365c5313 | 4875 | } while (flush_state < COMMIT_TRANS); |
21c7e756 MX |
4876 | } |
4877 | ||
4878 | void btrfs_init_async_reclaim_work(struct work_struct *work) | |
4879 | { | |
4880 | INIT_WORK(work, btrfs_async_reclaim_metadata_space); | |
4881 | } | |
4882 | ||
4a92b1b8 JB |
4883 | /** |
4884 | * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space | |
4885 | * @root - the root we're allocating for | |
4886 | * @block_rsv - the block_rsv we're allocating for | |
4887 | * @orig_bytes - the number of bytes we want | |
48fc7f7e | 4888 | * @flush - whether or not we can flush to make our reservation |
8bb8ab2e | 4889 | * |
4a92b1b8 JB |
4890 | * This will reserve orgi_bytes number of bytes from the space info associated |
4891 | * with the block_rsv. If there is not enough space it will make an attempt to | |
4892 | * flush out space to make room. It will do this by flushing delalloc if | |
4893 | * possible or committing the transaction. If flush is 0 then no attempts to | |
4894 | * regain reservations will be made and this will fail if there is not enough | |
4895 | * space already. | |
8bb8ab2e | 4896 | */ |
4a92b1b8 | 4897 | static int reserve_metadata_bytes(struct btrfs_root *root, |
8bb8ab2e | 4898 | struct btrfs_block_rsv *block_rsv, |
08e007d2 MX |
4899 | u64 orig_bytes, |
4900 | enum btrfs_reserve_flush_enum flush) | |
9ed74f2d | 4901 | { |
f0486c68 | 4902 | struct btrfs_space_info *space_info = block_rsv->space_info; |
2bf64758 | 4903 | u64 used; |
8bb8ab2e | 4904 | u64 num_bytes = orig_bytes; |
67b0fd63 | 4905 | int flush_state = FLUSH_DELAYED_ITEMS_NR; |
8bb8ab2e | 4906 | int ret = 0; |
fdb5effd | 4907 | bool flushing = false; |
9ed74f2d | 4908 | |
8bb8ab2e | 4909 | again: |
fdb5effd | 4910 | ret = 0; |
8bb8ab2e | 4911 | spin_lock(&space_info->lock); |
fdb5effd | 4912 | /* |
08e007d2 MX |
4913 | * We only want to wait if somebody other than us is flushing and we |
4914 | * are actually allowed to flush all things. | |
fdb5effd | 4915 | */ |
08e007d2 MX |
4916 | while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing && |
4917 | space_info->flush) { | |
fdb5effd JB |
4918 | spin_unlock(&space_info->lock); |
4919 | /* | |
4920 | * If we have a trans handle we can't wait because the flusher | |
4921 | * may have to commit the transaction, which would mean we would | |
4922 | * deadlock since we are waiting for the flusher to finish, but | |
4923 | * hold the current transaction open. | |
4924 | */ | |
663350ac | 4925 | if (current->journal_info) |
fdb5effd | 4926 | return -EAGAIN; |
b9688bb8 AJ |
4927 | ret = wait_event_killable(space_info->wait, !space_info->flush); |
4928 | /* Must have been killed, return */ | |
4929 | if (ret) | |
fdb5effd JB |
4930 | return -EINTR; |
4931 | ||
4932 | spin_lock(&space_info->lock); | |
4933 | } | |
4934 | ||
4935 | ret = -ENOSPC; | |
2bf64758 JB |
4936 | used = space_info->bytes_used + space_info->bytes_reserved + |
4937 | space_info->bytes_pinned + space_info->bytes_readonly + | |
4938 | space_info->bytes_may_use; | |
9ed74f2d | 4939 | |
8bb8ab2e JB |
4940 | /* |
4941 | * The idea here is that we've not already over-reserved the block group | |
4942 | * then we can go ahead and save our reservation first and then start | |
4943 | * flushing if we need to. Otherwise if we've already overcommitted | |
4944 | * lets start flushing stuff first and then come back and try to make | |
4945 | * our reservation. | |
4946 | */ | |
2bf64758 JB |
4947 | if (used <= space_info->total_bytes) { |
4948 | if (used + orig_bytes <= space_info->total_bytes) { | |
fb25e914 | 4949 | space_info->bytes_may_use += orig_bytes; |
8c2a3ca2 | 4950 | trace_btrfs_space_reservation(root->fs_info, |
2bcc0328 | 4951 | "space_info", space_info->flags, orig_bytes, 1); |
8bb8ab2e JB |
4952 | ret = 0; |
4953 | } else { | |
4954 | /* | |
4955 | * Ok set num_bytes to orig_bytes since we aren't | |
4956 | * overocmmitted, this way we only try and reclaim what | |
4957 | * we need. | |
4958 | */ | |
4959 | num_bytes = orig_bytes; | |
4960 | } | |
4961 | } else { | |
4962 | /* | |
4963 | * Ok we're over committed, set num_bytes to the overcommitted | |
4964 | * amount plus the amount of bytes that we need for this | |
4965 | * reservation. | |
4966 | */ | |
2bf64758 | 4967 | num_bytes = used - space_info->total_bytes + |
96c3f433 | 4968 | (orig_bytes * 2); |
8bb8ab2e | 4969 | } |
9ed74f2d | 4970 | |
44734ed1 JB |
4971 | if (ret && can_overcommit(root, space_info, orig_bytes, flush)) { |
4972 | space_info->bytes_may_use += orig_bytes; | |
4973 | trace_btrfs_space_reservation(root->fs_info, "space_info", | |
4974 | space_info->flags, orig_bytes, | |
4975 | 1); | |
4976 | ret = 0; | |
2bf64758 JB |
4977 | } |
4978 | ||
8bb8ab2e JB |
4979 | /* |
4980 | * Couldn't make our reservation, save our place so while we're trying | |
4981 | * to reclaim space we can actually use it instead of somebody else | |
4982 | * stealing it from us. | |
08e007d2 MX |
4983 | * |
4984 | * We make the other tasks wait for the flush only when we can flush | |
4985 | * all things. | |
8bb8ab2e | 4986 | */ |
72bcd99d | 4987 | if (ret && flush != BTRFS_RESERVE_NO_FLUSH) { |
fdb5effd JB |
4988 | flushing = true; |
4989 | space_info->flush = 1; | |
21c7e756 MX |
4990 | } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) { |
4991 | used += orig_bytes; | |
f6acfd50 JB |
4992 | /* |
4993 | * We will do the space reservation dance during log replay, | |
4994 | * which means we won't have fs_info->fs_root set, so don't do | |
4995 | * the async reclaim as we will panic. | |
4996 | */ | |
4997 | if (!root->fs_info->log_root_recovering && | |
4998 | need_do_async_reclaim(space_info, root->fs_info, used) && | |
21c7e756 MX |
4999 | !work_busy(&root->fs_info->async_reclaim_work)) |
5000 | queue_work(system_unbound_wq, | |
5001 | &root->fs_info->async_reclaim_work); | |
8bb8ab2e | 5002 | } |
f0486c68 | 5003 | spin_unlock(&space_info->lock); |
9ed74f2d | 5004 | |
08e007d2 | 5005 | if (!ret || flush == BTRFS_RESERVE_NO_FLUSH) |
8bb8ab2e | 5006 | goto out; |
f0486c68 | 5007 | |
96c3f433 JB |
5008 | ret = flush_space(root, space_info, num_bytes, orig_bytes, |
5009 | flush_state); | |
5010 | flush_state++; | |
08e007d2 MX |
5011 | |
5012 | /* | |
5013 | * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock | |
5014 | * would happen. So skip delalloc flush. | |
5015 | */ | |
5016 | if (flush == BTRFS_RESERVE_FLUSH_LIMIT && | |
5017 | (flush_state == FLUSH_DELALLOC || | |
5018 | flush_state == FLUSH_DELALLOC_WAIT)) | |
5019 | flush_state = ALLOC_CHUNK; | |
5020 | ||
96c3f433 | 5021 | if (!ret) |
8bb8ab2e | 5022 | goto again; |
08e007d2 MX |
5023 | else if (flush == BTRFS_RESERVE_FLUSH_LIMIT && |
5024 | flush_state < COMMIT_TRANS) | |
5025 | goto again; | |
5026 | else if (flush == BTRFS_RESERVE_FLUSH_ALL && | |
5027 | flush_state <= COMMIT_TRANS) | |
8bb8ab2e JB |
5028 | goto again; |
5029 | ||
5030 | out: | |
5d80366e JB |
5031 | if (ret == -ENOSPC && |
5032 | unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) { | |
5033 | struct btrfs_block_rsv *global_rsv = | |
5034 | &root->fs_info->global_block_rsv; | |
5035 | ||
5036 | if (block_rsv != global_rsv && | |
5037 | !block_rsv_use_bytes(global_rsv, orig_bytes)) | |
5038 | ret = 0; | |
5039 | } | |
cab45e22 JM |
5040 | if (ret == -ENOSPC) |
5041 | trace_btrfs_space_reservation(root->fs_info, | |
5042 | "space_info:enospc", | |
5043 | space_info->flags, orig_bytes, 1); | |
fdb5effd | 5044 | if (flushing) { |
8bb8ab2e | 5045 | spin_lock(&space_info->lock); |
fdb5effd JB |
5046 | space_info->flush = 0; |
5047 | wake_up_all(&space_info->wait); | |
8bb8ab2e | 5048 | spin_unlock(&space_info->lock); |
f0486c68 | 5049 | } |
f0486c68 YZ |
5050 | return ret; |
5051 | } | |
5052 | ||
79787eaa JM |
5053 | static struct btrfs_block_rsv *get_block_rsv( |
5054 | const struct btrfs_trans_handle *trans, | |
5055 | const struct btrfs_root *root) | |
f0486c68 | 5056 | { |
4c13d758 JB |
5057 | struct btrfs_block_rsv *block_rsv = NULL; |
5058 | ||
e9cf439f AM |
5059 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) || |
5060 | (root == root->fs_info->csum_root && trans->adding_csums) || | |
5061 | (root == root->fs_info->uuid_root)) | |
f7a81ea4 SB |
5062 | block_rsv = trans->block_rsv; |
5063 | ||
4c13d758 | 5064 | if (!block_rsv) |
f0486c68 YZ |
5065 | block_rsv = root->block_rsv; |
5066 | ||
5067 | if (!block_rsv) | |
5068 | block_rsv = &root->fs_info->empty_block_rsv; | |
5069 | ||
5070 | return block_rsv; | |
5071 | } | |
5072 | ||
5073 | static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, | |
5074 | u64 num_bytes) | |
5075 | { | |
5076 | int ret = -ENOSPC; | |
5077 | spin_lock(&block_rsv->lock); | |
5078 | if (block_rsv->reserved >= num_bytes) { | |
5079 | block_rsv->reserved -= num_bytes; | |
5080 | if (block_rsv->reserved < block_rsv->size) | |
5081 | block_rsv->full = 0; | |
5082 | ret = 0; | |
5083 | } | |
5084 | spin_unlock(&block_rsv->lock); | |
5085 | return ret; | |
5086 | } | |
5087 | ||
5088 | static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv, | |
5089 | u64 num_bytes, int update_size) | |
5090 | { | |
5091 | spin_lock(&block_rsv->lock); | |
5092 | block_rsv->reserved += num_bytes; | |
5093 | if (update_size) | |
5094 | block_rsv->size += num_bytes; | |
5095 | else if (block_rsv->reserved >= block_rsv->size) | |
5096 | block_rsv->full = 1; | |
5097 | spin_unlock(&block_rsv->lock); | |
5098 | } | |
5099 | ||
d52be818 JB |
5100 | int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info, |
5101 | struct btrfs_block_rsv *dest, u64 num_bytes, | |
5102 | int min_factor) | |
5103 | { | |
5104 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
5105 | u64 min_bytes; | |
5106 | ||
5107 | if (global_rsv->space_info != dest->space_info) | |
5108 | return -ENOSPC; | |
5109 | ||
5110 | spin_lock(&global_rsv->lock); | |
5111 | min_bytes = div_factor(global_rsv->size, min_factor); | |
5112 | if (global_rsv->reserved < min_bytes + num_bytes) { | |
5113 | spin_unlock(&global_rsv->lock); | |
5114 | return -ENOSPC; | |
5115 | } | |
5116 | global_rsv->reserved -= num_bytes; | |
5117 | if (global_rsv->reserved < global_rsv->size) | |
5118 | global_rsv->full = 0; | |
5119 | spin_unlock(&global_rsv->lock); | |
5120 | ||
5121 | block_rsv_add_bytes(dest, num_bytes, 1); | |
5122 | return 0; | |
5123 | } | |
5124 | ||
8c2a3ca2 JB |
5125 | static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info, |
5126 | struct btrfs_block_rsv *block_rsv, | |
62a45b60 | 5127 | struct btrfs_block_rsv *dest, u64 num_bytes) |
f0486c68 YZ |
5128 | { |
5129 | struct btrfs_space_info *space_info = block_rsv->space_info; | |
5130 | ||
5131 | spin_lock(&block_rsv->lock); | |
5132 | if (num_bytes == (u64)-1) | |
5133 | num_bytes = block_rsv->size; | |
5134 | block_rsv->size -= num_bytes; | |
5135 | if (block_rsv->reserved >= block_rsv->size) { | |
5136 | num_bytes = block_rsv->reserved - block_rsv->size; | |
5137 | block_rsv->reserved = block_rsv->size; | |
5138 | block_rsv->full = 1; | |
5139 | } else { | |
5140 | num_bytes = 0; | |
5141 | } | |
5142 | spin_unlock(&block_rsv->lock); | |
5143 | ||
5144 | if (num_bytes > 0) { | |
5145 | if (dest) { | |
e9e22899 JB |
5146 | spin_lock(&dest->lock); |
5147 | if (!dest->full) { | |
5148 | u64 bytes_to_add; | |
5149 | ||
5150 | bytes_to_add = dest->size - dest->reserved; | |
5151 | bytes_to_add = min(num_bytes, bytes_to_add); | |
5152 | dest->reserved += bytes_to_add; | |
5153 | if (dest->reserved >= dest->size) | |
5154 | dest->full = 1; | |
5155 | num_bytes -= bytes_to_add; | |
5156 | } | |
5157 | spin_unlock(&dest->lock); | |
5158 | } | |
5159 | if (num_bytes) { | |
f0486c68 | 5160 | spin_lock(&space_info->lock); |
fb25e914 | 5161 | space_info->bytes_may_use -= num_bytes; |
8c2a3ca2 | 5162 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 5163 | space_info->flags, num_bytes, 0); |
f0486c68 | 5164 | spin_unlock(&space_info->lock); |
4e06bdd6 | 5165 | } |
9ed74f2d | 5166 | } |
f0486c68 | 5167 | } |
4e06bdd6 | 5168 | |
f0486c68 YZ |
5169 | static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src, |
5170 | struct btrfs_block_rsv *dst, u64 num_bytes) | |
5171 | { | |
5172 | int ret; | |
9ed74f2d | 5173 | |
f0486c68 YZ |
5174 | ret = block_rsv_use_bytes(src, num_bytes); |
5175 | if (ret) | |
5176 | return ret; | |
9ed74f2d | 5177 | |
f0486c68 | 5178 | block_rsv_add_bytes(dst, num_bytes, 1); |
9ed74f2d JB |
5179 | return 0; |
5180 | } | |
5181 | ||
66d8f3dd | 5182 | void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type) |
9ed74f2d | 5183 | { |
f0486c68 YZ |
5184 | memset(rsv, 0, sizeof(*rsv)); |
5185 | spin_lock_init(&rsv->lock); | |
66d8f3dd | 5186 | rsv->type = type; |
f0486c68 YZ |
5187 | } |
5188 | ||
66d8f3dd MX |
5189 | struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root, |
5190 | unsigned short type) | |
f0486c68 YZ |
5191 | { |
5192 | struct btrfs_block_rsv *block_rsv; | |
5193 | struct btrfs_fs_info *fs_info = root->fs_info; | |
9ed74f2d | 5194 | |
f0486c68 YZ |
5195 | block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS); |
5196 | if (!block_rsv) | |
5197 | return NULL; | |
9ed74f2d | 5198 | |
66d8f3dd | 5199 | btrfs_init_block_rsv(block_rsv, type); |
f0486c68 YZ |
5200 | block_rsv->space_info = __find_space_info(fs_info, |
5201 | BTRFS_BLOCK_GROUP_METADATA); | |
f0486c68 YZ |
5202 | return block_rsv; |
5203 | } | |
9ed74f2d | 5204 | |
f0486c68 YZ |
5205 | void btrfs_free_block_rsv(struct btrfs_root *root, |
5206 | struct btrfs_block_rsv *rsv) | |
5207 | { | |
2aaa6655 JB |
5208 | if (!rsv) |
5209 | return; | |
dabdb640 JB |
5210 | btrfs_block_rsv_release(root, rsv, (u64)-1); |
5211 | kfree(rsv); | |
9ed74f2d JB |
5212 | } |
5213 | ||
cdfb080e CM |
5214 | void __btrfs_free_block_rsv(struct btrfs_block_rsv *rsv) |
5215 | { | |
5216 | kfree(rsv); | |
5217 | } | |
5218 | ||
08e007d2 MX |
5219 | int btrfs_block_rsv_add(struct btrfs_root *root, |
5220 | struct btrfs_block_rsv *block_rsv, u64 num_bytes, | |
5221 | enum btrfs_reserve_flush_enum flush) | |
9ed74f2d | 5222 | { |
f0486c68 | 5223 | int ret; |
9ed74f2d | 5224 | |
f0486c68 YZ |
5225 | if (num_bytes == 0) |
5226 | return 0; | |
8bb8ab2e | 5227 | |
61b520a9 | 5228 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
f0486c68 YZ |
5229 | if (!ret) { |
5230 | block_rsv_add_bytes(block_rsv, num_bytes, 1); | |
5231 | return 0; | |
5232 | } | |
9ed74f2d | 5233 | |
f0486c68 | 5234 | return ret; |
f0486c68 | 5235 | } |
9ed74f2d | 5236 | |
4a92b1b8 | 5237 | int btrfs_block_rsv_check(struct btrfs_root *root, |
36ba022a | 5238 | struct btrfs_block_rsv *block_rsv, int min_factor) |
f0486c68 YZ |
5239 | { |
5240 | u64 num_bytes = 0; | |
f0486c68 | 5241 | int ret = -ENOSPC; |
9ed74f2d | 5242 | |
f0486c68 YZ |
5243 | if (!block_rsv) |
5244 | return 0; | |
9ed74f2d | 5245 | |
f0486c68 | 5246 | spin_lock(&block_rsv->lock); |
36ba022a JB |
5247 | num_bytes = div_factor(block_rsv->size, min_factor); |
5248 | if (block_rsv->reserved >= num_bytes) | |
5249 | ret = 0; | |
5250 | spin_unlock(&block_rsv->lock); | |
9ed74f2d | 5251 | |
36ba022a JB |
5252 | return ret; |
5253 | } | |
5254 | ||
08e007d2 MX |
5255 | int btrfs_block_rsv_refill(struct btrfs_root *root, |
5256 | struct btrfs_block_rsv *block_rsv, u64 min_reserved, | |
5257 | enum btrfs_reserve_flush_enum flush) | |
36ba022a JB |
5258 | { |
5259 | u64 num_bytes = 0; | |
5260 | int ret = -ENOSPC; | |
5261 | ||
5262 | if (!block_rsv) | |
5263 | return 0; | |
5264 | ||
5265 | spin_lock(&block_rsv->lock); | |
5266 | num_bytes = min_reserved; | |
13553e52 | 5267 | if (block_rsv->reserved >= num_bytes) |
f0486c68 | 5268 | ret = 0; |
13553e52 | 5269 | else |
f0486c68 | 5270 | num_bytes -= block_rsv->reserved; |
f0486c68 | 5271 | spin_unlock(&block_rsv->lock); |
13553e52 | 5272 | |
f0486c68 YZ |
5273 | if (!ret) |
5274 | return 0; | |
5275 | ||
aa38a711 | 5276 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
dabdb640 JB |
5277 | if (!ret) { |
5278 | block_rsv_add_bytes(block_rsv, num_bytes, 0); | |
f0486c68 | 5279 | return 0; |
6a63209f | 5280 | } |
9ed74f2d | 5281 | |
13553e52 | 5282 | return ret; |
f0486c68 YZ |
5283 | } |
5284 | ||
5285 | int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv, | |
5286 | struct btrfs_block_rsv *dst_rsv, | |
5287 | u64 num_bytes) | |
5288 | { | |
5289 | return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes); | |
5290 | } | |
5291 | ||
5292 | void btrfs_block_rsv_release(struct btrfs_root *root, | |
5293 | struct btrfs_block_rsv *block_rsv, | |
5294 | u64 num_bytes) | |
5295 | { | |
5296 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; | |
17504584 | 5297 | if (global_rsv == block_rsv || |
f0486c68 YZ |
5298 | block_rsv->space_info != global_rsv->space_info) |
5299 | global_rsv = NULL; | |
8c2a3ca2 JB |
5300 | block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv, |
5301 | num_bytes); | |
6a63209f JB |
5302 | } |
5303 | ||
5304 | /* | |
8929ecfa YZ |
5305 | * helper to calculate size of global block reservation. |
5306 | * the desired value is sum of space used by extent tree, | |
5307 | * checksum tree and root tree | |
6a63209f | 5308 | */ |
8929ecfa | 5309 | static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info) |
6a63209f | 5310 | { |
8929ecfa YZ |
5311 | struct btrfs_space_info *sinfo; |
5312 | u64 num_bytes; | |
5313 | u64 meta_used; | |
5314 | u64 data_used; | |
6c41761f | 5315 | int csum_size = btrfs_super_csum_size(fs_info->super_copy); |
6a63209f | 5316 | |
8929ecfa YZ |
5317 | sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA); |
5318 | spin_lock(&sinfo->lock); | |
5319 | data_used = sinfo->bytes_used; | |
5320 | spin_unlock(&sinfo->lock); | |
33b4d47f | 5321 | |
8929ecfa YZ |
5322 | sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
5323 | spin_lock(&sinfo->lock); | |
6d48755d JB |
5324 | if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA) |
5325 | data_used = 0; | |
8929ecfa YZ |
5326 | meta_used = sinfo->bytes_used; |
5327 | spin_unlock(&sinfo->lock); | |
ab6e2410 | 5328 | |
8929ecfa YZ |
5329 | num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) * |
5330 | csum_size * 2; | |
f8c269d7 | 5331 | num_bytes += div_u64(data_used + meta_used, 50); |
4e06bdd6 | 5332 | |
8929ecfa | 5333 | if (num_bytes * 3 > meta_used) |
f8c269d7 | 5334 | num_bytes = div_u64(meta_used, 3); |
ab6e2410 | 5335 | |
707e8a07 | 5336 | return ALIGN(num_bytes, fs_info->extent_root->nodesize << 10); |
8929ecfa | 5337 | } |
6a63209f | 5338 | |
8929ecfa YZ |
5339 | static void update_global_block_rsv(struct btrfs_fs_info *fs_info) |
5340 | { | |
5341 | struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv; | |
5342 | struct btrfs_space_info *sinfo = block_rsv->space_info; | |
5343 | u64 num_bytes; | |
6a63209f | 5344 | |
8929ecfa | 5345 | num_bytes = calc_global_metadata_size(fs_info); |
33b4d47f | 5346 | |
8929ecfa | 5347 | spin_lock(&sinfo->lock); |
1f699d38 | 5348 | spin_lock(&block_rsv->lock); |
4e06bdd6 | 5349 | |
fdf30d1c | 5350 | block_rsv->size = min_t(u64, num_bytes, 512 * 1024 * 1024); |
4e06bdd6 | 5351 | |
8929ecfa | 5352 | num_bytes = sinfo->bytes_used + sinfo->bytes_pinned + |
6d48755d JB |
5353 | sinfo->bytes_reserved + sinfo->bytes_readonly + |
5354 | sinfo->bytes_may_use; | |
8929ecfa YZ |
5355 | |
5356 | if (sinfo->total_bytes > num_bytes) { | |
5357 | num_bytes = sinfo->total_bytes - num_bytes; | |
5358 | block_rsv->reserved += num_bytes; | |
fb25e914 | 5359 | sinfo->bytes_may_use += num_bytes; |
8c2a3ca2 | 5360 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 5361 | sinfo->flags, num_bytes, 1); |
6a63209f | 5362 | } |
6a63209f | 5363 | |
8929ecfa YZ |
5364 | if (block_rsv->reserved >= block_rsv->size) { |
5365 | num_bytes = block_rsv->reserved - block_rsv->size; | |
fb25e914 | 5366 | sinfo->bytes_may_use -= num_bytes; |
8c2a3ca2 | 5367 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 5368 | sinfo->flags, num_bytes, 0); |
8929ecfa YZ |
5369 | block_rsv->reserved = block_rsv->size; |
5370 | block_rsv->full = 1; | |
5371 | } | |
182608c8 | 5372 | |
8929ecfa | 5373 | spin_unlock(&block_rsv->lock); |
1f699d38 | 5374 | spin_unlock(&sinfo->lock); |
6a63209f JB |
5375 | } |
5376 | ||
f0486c68 | 5377 | static void init_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 5378 | { |
f0486c68 | 5379 | struct btrfs_space_info *space_info; |
6a63209f | 5380 | |
f0486c68 YZ |
5381 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); |
5382 | fs_info->chunk_block_rsv.space_info = space_info; | |
6a63209f | 5383 | |
f0486c68 | 5384 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
8929ecfa | 5385 | fs_info->global_block_rsv.space_info = space_info; |
8929ecfa | 5386 | fs_info->delalloc_block_rsv.space_info = space_info; |
f0486c68 YZ |
5387 | fs_info->trans_block_rsv.space_info = space_info; |
5388 | fs_info->empty_block_rsv.space_info = space_info; | |
6d668dda | 5389 | fs_info->delayed_block_rsv.space_info = space_info; |
f0486c68 | 5390 | |
8929ecfa YZ |
5391 | fs_info->extent_root->block_rsv = &fs_info->global_block_rsv; |
5392 | fs_info->csum_root->block_rsv = &fs_info->global_block_rsv; | |
5393 | fs_info->dev_root->block_rsv = &fs_info->global_block_rsv; | |
5394 | fs_info->tree_root->block_rsv = &fs_info->global_block_rsv; | |
3a6cad90 SB |
5395 | if (fs_info->quota_root) |
5396 | fs_info->quota_root->block_rsv = &fs_info->global_block_rsv; | |
f0486c68 | 5397 | fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv; |
8929ecfa | 5398 | |
8929ecfa | 5399 | update_global_block_rsv(fs_info); |
6a63209f JB |
5400 | } |
5401 | ||
8929ecfa | 5402 | static void release_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 5403 | { |
8c2a3ca2 JB |
5404 | block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL, |
5405 | (u64)-1); | |
8929ecfa YZ |
5406 | WARN_ON(fs_info->delalloc_block_rsv.size > 0); |
5407 | WARN_ON(fs_info->delalloc_block_rsv.reserved > 0); | |
5408 | WARN_ON(fs_info->trans_block_rsv.size > 0); | |
5409 | WARN_ON(fs_info->trans_block_rsv.reserved > 0); | |
5410 | WARN_ON(fs_info->chunk_block_rsv.size > 0); | |
5411 | WARN_ON(fs_info->chunk_block_rsv.reserved > 0); | |
6d668dda JB |
5412 | WARN_ON(fs_info->delayed_block_rsv.size > 0); |
5413 | WARN_ON(fs_info->delayed_block_rsv.reserved > 0); | |
fcb80c2a JB |
5414 | } |
5415 | ||
a22285a6 YZ |
5416 | void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans, |
5417 | struct btrfs_root *root) | |
6a63209f | 5418 | { |
0e721106 JB |
5419 | if (!trans->block_rsv) |
5420 | return; | |
5421 | ||
a22285a6 YZ |
5422 | if (!trans->bytes_reserved) |
5423 | return; | |
6a63209f | 5424 | |
e77266e4 | 5425 | trace_btrfs_space_reservation(root->fs_info, "transaction", |
2bcc0328 | 5426 | trans->transid, trans->bytes_reserved, 0); |
b24e03db | 5427 | btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved); |
a22285a6 YZ |
5428 | trans->bytes_reserved = 0; |
5429 | } | |
6a63209f | 5430 | |
4fbcdf66 FM |
5431 | /* |
5432 | * To be called after all the new block groups attached to the transaction | |
5433 | * handle have been created (btrfs_create_pending_block_groups()). | |
5434 | */ | |
5435 | void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans) | |
5436 | { | |
5437 | struct btrfs_fs_info *fs_info = trans->root->fs_info; | |
5438 | ||
5439 | if (!trans->chunk_bytes_reserved) | |
5440 | return; | |
5441 | ||
5442 | WARN_ON_ONCE(!list_empty(&trans->new_bgs)); | |
5443 | ||
5444 | block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL, | |
5445 | trans->chunk_bytes_reserved); | |
5446 | trans->chunk_bytes_reserved = 0; | |
5447 | } | |
5448 | ||
79787eaa | 5449 | /* Can only return 0 or -ENOSPC */ |
d68fc57b YZ |
5450 | int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans, |
5451 | struct inode *inode) | |
5452 | { | |
5453 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5454 | struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root); | |
5455 | struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv; | |
5456 | ||
5457 | /* | |
fcb80c2a JB |
5458 | * We need to hold space in order to delete our orphan item once we've |
5459 | * added it, so this takes the reservation so we can release it later | |
5460 | * when we are truly done with the orphan item. | |
d68fc57b | 5461 | */ |
ff5714cc | 5462 | u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1); |
8c2a3ca2 JB |
5463 | trace_btrfs_space_reservation(root->fs_info, "orphan", |
5464 | btrfs_ino(inode), num_bytes, 1); | |
d68fc57b | 5465 | return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes); |
6a63209f JB |
5466 | } |
5467 | ||
d68fc57b | 5468 | void btrfs_orphan_release_metadata(struct inode *inode) |
97e728d4 | 5469 | { |
d68fc57b | 5470 | struct btrfs_root *root = BTRFS_I(inode)->root; |
ff5714cc | 5471 | u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1); |
8c2a3ca2 JB |
5472 | trace_btrfs_space_reservation(root->fs_info, "orphan", |
5473 | btrfs_ino(inode), num_bytes, 0); | |
d68fc57b YZ |
5474 | btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes); |
5475 | } | |
97e728d4 | 5476 | |
d5c12070 MX |
5477 | /* |
5478 | * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation | |
5479 | * root: the root of the parent directory | |
5480 | * rsv: block reservation | |
5481 | * items: the number of items that we need do reservation | |
5482 | * qgroup_reserved: used to return the reserved size in qgroup | |
5483 | * | |
5484 | * This function is used to reserve the space for snapshot/subvolume | |
5485 | * creation and deletion. Those operations are different with the | |
5486 | * common file/directory operations, they change two fs/file trees | |
5487 | * and root tree, the number of items that the qgroup reserves is | |
5488 | * different with the free space reservation. So we can not use | |
5489 | * the space reseravtion mechanism in start_transaction(). | |
5490 | */ | |
5491 | int btrfs_subvolume_reserve_metadata(struct btrfs_root *root, | |
5492 | struct btrfs_block_rsv *rsv, | |
5493 | int items, | |
ee3441b4 JM |
5494 | u64 *qgroup_reserved, |
5495 | bool use_global_rsv) | |
a22285a6 | 5496 | { |
d5c12070 MX |
5497 | u64 num_bytes; |
5498 | int ret; | |
ee3441b4 | 5499 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; |
d5c12070 MX |
5500 | |
5501 | if (root->fs_info->quota_enabled) { | |
5502 | /* One for parent inode, two for dir entries */ | |
707e8a07 | 5503 | num_bytes = 3 * root->nodesize; |
7174109c | 5504 | ret = btrfs_qgroup_reserve_meta(root, num_bytes); |
d5c12070 MX |
5505 | if (ret) |
5506 | return ret; | |
5507 | } else { | |
5508 | num_bytes = 0; | |
5509 | } | |
5510 | ||
5511 | *qgroup_reserved = num_bytes; | |
5512 | ||
5513 | num_bytes = btrfs_calc_trans_metadata_size(root, items); | |
5514 | rsv->space_info = __find_space_info(root->fs_info, | |
5515 | BTRFS_BLOCK_GROUP_METADATA); | |
5516 | ret = btrfs_block_rsv_add(root, rsv, num_bytes, | |
5517 | BTRFS_RESERVE_FLUSH_ALL); | |
ee3441b4 JM |
5518 | |
5519 | if (ret == -ENOSPC && use_global_rsv) | |
5520 | ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes); | |
5521 | ||
7174109c QW |
5522 | if (ret && *qgroup_reserved) |
5523 | btrfs_qgroup_free_meta(root, *qgroup_reserved); | |
d5c12070 MX |
5524 | |
5525 | return ret; | |
5526 | } | |
5527 | ||
5528 | void btrfs_subvolume_release_metadata(struct btrfs_root *root, | |
5529 | struct btrfs_block_rsv *rsv, | |
5530 | u64 qgroup_reserved) | |
5531 | { | |
5532 | btrfs_block_rsv_release(root, rsv, (u64)-1); | |
97e728d4 JB |
5533 | } |
5534 | ||
7709cde3 JB |
5535 | /** |
5536 | * drop_outstanding_extent - drop an outstanding extent | |
5537 | * @inode: the inode we're dropping the extent for | |
dcab6a3b | 5538 | * @num_bytes: the number of bytes we're relaseing. |
7709cde3 JB |
5539 | * |
5540 | * This is called when we are freeing up an outstanding extent, either called | |
5541 | * after an error or after an extent is written. This will return the number of | |
5542 | * reserved extents that need to be freed. This must be called with | |
5543 | * BTRFS_I(inode)->lock held. | |
5544 | */ | |
dcab6a3b | 5545 | static unsigned drop_outstanding_extent(struct inode *inode, u64 num_bytes) |
9e0baf60 | 5546 | { |
7fd2ae21 | 5547 | unsigned drop_inode_space = 0; |
9e0baf60 | 5548 | unsigned dropped_extents = 0; |
dcab6a3b | 5549 | unsigned num_extents = 0; |
9e0baf60 | 5550 | |
dcab6a3b JB |
5551 | num_extents = (unsigned)div64_u64(num_bytes + |
5552 | BTRFS_MAX_EXTENT_SIZE - 1, | |
5553 | BTRFS_MAX_EXTENT_SIZE); | |
5554 | ASSERT(num_extents); | |
5555 | ASSERT(BTRFS_I(inode)->outstanding_extents >= num_extents); | |
5556 | BTRFS_I(inode)->outstanding_extents -= num_extents; | |
9e0baf60 | 5557 | |
7fd2ae21 | 5558 | if (BTRFS_I(inode)->outstanding_extents == 0 && |
72ac3c0d JB |
5559 | test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED, |
5560 | &BTRFS_I(inode)->runtime_flags)) | |
7fd2ae21 | 5561 | drop_inode_space = 1; |
7fd2ae21 | 5562 | |
9e0baf60 JB |
5563 | /* |
5564 | * If we have more or the same amount of outsanding extents than we have | |
5565 | * reserved then we need to leave the reserved extents count alone. | |
5566 | */ | |
5567 | if (BTRFS_I(inode)->outstanding_extents >= | |
5568 | BTRFS_I(inode)->reserved_extents) | |
7fd2ae21 | 5569 | return drop_inode_space; |
9e0baf60 JB |
5570 | |
5571 | dropped_extents = BTRFS_I(inode)->reserved_extents - | |
5572 | BTRFS_I(inode)->outstanding_extents; | |
5573 | BTRFS_I(inode)->reserved_extents -= dropped_extents; | |
7fd2ae21 | 5574 | return dropped_extents + drop_inode_space; |
9e0baf60 JB |
5575 | } |
5576 | ||
7709cde3 JB |
5577 | /** |
5578 | * calc_csum_metadata_size - return the amount of metada space that must be | |
5579 | * reserved/free'd for the given bytes. | |
5580 | * @inode: the inode we're manipulating | |
5581 | * @num_bytes: the number of bytes in question | |
5582 | * @reserve: 1 if we are reserving space, 0 if we are freeing space | |
5583 | * | |
5584 | * This adjusts the number of csum_bytes in the inode and then returns the | |
5585 | * correct amount of metadata that must either be reserved or freed. We | |
5586 | * calculate how many checksums we can fit into one leaf and then divide the | |
5587 | * number of bytes that will need to be checksumed by this value to figure out | |
5588 | * how many checksums will be required. If we are adding bytes then the number | |
5589 | * may go up and we will return the number of additional bytes that must be | |
5590 | * reserved. If it is going down we will return the number of bytes that must | |
5591 | * be freed. | |
5592 | * | |
5593 | * This must be called with BTRFS_I(inode)->lock held. | |
5594 | */ | |
5595 | static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes, | |
5596 | int reserve) | |
6324fbf3 | 5597 | { |
7709cde3 | 5598 | struct btrfs_root *root = BTRFS_I(inode)->root; |
1262133b | 5599 | u64 old_csums, num_csums; |
7709cde3 JB |
5600 | |
5601 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM && | |
5602 | BTRFS_I(inode)->csum_bytes == 0) | |
5603 | return 0; | |
5604 | ||
28f75a0e | 5605 | old_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes); |
7709cde3 JB |
5606 | if (reserve) |
5607 | BTRFS_I(inode)->csum_bytes += num_bytes; | |
5608 | else | |
5609 | BTRFS_I(inode)->csum_bytes -= num_bytes; | |
28f75a0e | 5610 | num_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes); |
7709cde3 JB |
5611 | |
5612 | /* No change, no need to reserve more */ | |
5613 | if (old_csums == num_csums) | |
5614 | return 0; | |
5615 | ||
5616 | if (reserve) | |
5617 | return btrfs_calc_trans_metadata_size(root, | |
5618 | num_csums - old_csums); | |
5619 | ||
5620 | return btrfs_calc_trans_metadata_size(root, old_csums - num_csums); | |
0ca1f7ce | 5621 | } |
c146afad | 5622 | |
0ca1f7ce YZ |
5623 | int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes) |
5624 | { | |
5625 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5626 | struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv; | |
9e0baf60 | 5627 | u64 to_reserve = 0; |
660d3f6c | 5628 | u64 csum_bytes; |
9e0baf60 | 5629 | unsigned nr_extents = 0; |
660d3f6c | 5630 | int extra_reserve = 0; |
08e007d2 | 5631 | enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL; |
eb6b88d9 | 5632 | int ret = 0; |
c64c2bd8 | 5633 | bool delalloc_lock = true; |
88e081bf WS |
5634 | u64 to_free = 0; |
5635 | unsigned dropped; | |
6324fbf3 | 5636 | |
c64c2bd8 JB |
5637 | /* If we are a free space inode we need to not flush since we will be in |
5638 | * the middle of a transaction commit. We also don't need the delalloc | |
5639 | * mutex since we won't race with anybody. We need this mostly to make | |
5640 | * lockdep shut its filthy mouth. | |
5641 | */ | |
5642 | if (btrfs_is_free_space_inode(inode)) { | |
08e007d2 | 5643 | flush = BTRFS_RESERVE_NO_FLUSH; |
c64c2bd8 JB |
5644 | delalloc_lock = false; |
5645 | } | |
c09544e0 | 5646 | |
08e007d2 MX |
5647 | if (flush != BTRFS_RESERVE_NO_FLUSH && |
5648 | btrfs_transaction_in_commit(root->fs_info)) | |
0ca1f7ce | 5649 | schedule_timeout(1); |
ec44a35c | 5650 | |
c64c2bd8 JB |
5651 | if (delalloc_lock) |
5652 | mutex_lock(&BTRFS_I(inode)->delalloc_mutex); | |
5653 | ||
0ca1f7ce | 5654 | num_bytes = ALIGN(num_bytes, root->sectorsize); |
8bb8ab2e | 5655 | |
9e0baf60 | 5656 | spin_lock(&BTRFS_I(inode)->lock); |
6a41dd09 JB |
5657 | nr_extents = (unsigned)div64_u64(num_bytes + |
5658 | BTRFS_MAX_EXTENT_SIZE - 1, | |
5659 | BTRFS_MAX_EXTENT_SIZE); | |
5660 | BTRFS_I(inode)->outstanding_extents += nr_extents; | |
5661 | nr_extents = 0; | |
9e0baf60 JB |
5662 | |
5663 | if (BTRFS_I(inode)->outstanding_extents > | |
660d3f6c | 5664 | BTRFS_I(inode)->reserved_extents) |
9e0baf60 JB |
5665 | nr_extents = BTRFS_I(inode)->outstanding_extents - |
5666 | BTRFS_I(inode)->reserved_extents; | |
57a45ced | 5667 | |
7fd2ae21 JB |
5668 | /* |
5669 | * Add an item to reserve for updating the inode when we complete the | |
5670 | * delalloc io. | |
5671 | */ | |
72ac3c0d JB |
5672 | if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED, |
5673 | &BTRFS_I(inode)->runtime_flags)) { | |
7fd2ae21 | 5674 | nr_extents++; |
660d3f6c | 5675 | extra_reserve = 1; |
593060d7 | 5676 | } |
7fd2ae21 JB |
5677 | |
5678 | to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents); | |
7709cde3 | 5679 | to_reserve += calc_csum_metadata_size(inode, num_bytes, 1); |
660d3f6c | 5680 | csum_bytes = BTRFS_I(inode)->csum_bytes; |
9e0baf60 | 5681 | spin_unlock(&BTRFS_I(inode)->lock); |
57a45ced | 5682 | |
88e081bf | 5683 | if (root->fs_info->quota_enabled) { |
7174109c QW |
5684 | ret = btrfs_qgroup_reserve_meta(root, |
5685 | nr_extents * root->nodesize); | |
88e081bf WS |
5686 | if (ret) |
5687 | goto out_fail; | |
5688 | } | |
c5567237 | 5689 | |
88e081bf WS |
5690 | ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush); |
5691 | if (unlikely(ret)) { | |
7174109c | 5692 | btrfs_qgroup_free_meta(root, nr_extents * root->nodesize); |
88e081bf | 5693 | goto out_fail; |
9e0baf60 | 5694 | } |
25179201 | 5695 | |
660d3f6c JB |
5696 | spin_lock(&BTRFS_I(inode)->lock); |
5697 | if (extra_reserve) { | |
72ac3c0d JB |
5698 | set_bit(BTRFS_INODE_DELALLOC_META_RESERVED, |
5699 | &BTRFS_I(inode)->runtime_flags); | |
660d3f6c JB |
5700 | nr_extents--; |
5701 | } | |
5702 | BTRFS_I(inode)->reserved_extents += nr_extents; | |
5703 | spin_unlock(&BTRFS_I(inode)->lock); | |
c64c2bd8 JB |
5704 | |
5705 | if (delalloc_lock) | |
5706 | mutex_unlock(&BTRFS_I(inode)->delalloc_mutex); | |
660d3f6c | 5707 | |
8c2a3ca2 | 5708 | if (to_reserve) |
67871254 | 5709 | trace_btrfs_space_reservation(root->fs_info, "delalloc", |
8c2a3ca2 | 5710 | btrfs_ino(inode), to_reserve, 1); |
0ca1f7ce YZ |
5711 | block_rsv_add_bytes(block_rsv, to_reserve, 1); |
5712 | ||
0ca1f7ce | 5713 | return 0; |
88e081bf WS |
5714 | |
5715 | out_fail: | |
5716 | spin_lock(&BTRFS_I(inode)->lock); | |
dcab6a3b | 5717 | dropped = drop_outstanding_extent(inode, num_bytes); |
88e081bf WS |
5718 | /* |
5719 | * If the inodes csum_bytes is the same as the original | |
5720 | * csum_bytes then we know we haven't raced with any free()ers | |
5721 | * so we can just reduce our inodes csum bytes and carry on. | |
88e081bf | 5722 | */ |
f4881bc7 | 5723 | if (BTRFS_I(inode)->csum_bytes == csum_bytes) { |
88e081bf | 5724 | calc_csum_metadata_size(inode, num_bytes, 0); |
f4881bc7 JB |
5725 | } else { |
5726 | u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes; | |
5727 | u64 bytes; | |
5728 | ||
5729 | /* | |
5730 | * This is tricky, but first we need to figure out how much we | |
5731 | * free'd from any free-ers that occured during this | |
5732 | * reservation, so we reset ->csum_bytes to the csum_bytes | |
5733 | * before we dropped our lock, and then call the free for the | |
5734 | * number of bytes that were freed while we were trying our | |
5735 | * reservation. | |
5736 | */ | |
5737 | bytes = csum_bytes - BTRFS_I(inode)->csum_bytes; | |
5738 | BTRFS_I(inode)->csum_bytes = csum_bytes; | |
5739 | to_free = calc_csum_metadata_size(inode, bytes, 0); | |
5740 | ||
5741 | ||
5742 | /* | |
5743 | * Now we need to see how much we would have freed had we not | |
5744 | * been making this reservation and our ->csum_bytes were not | |
5745 | * artificially inflated. | |
5746 | */ | |
5747 | BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes; | |
5748 | bytes = csum_bytes - orig_csum_bytes; | |
5749 | bytes = calc_csum_metadata_size(inode, bytes, 0); | |
5750 | ||
5751 | /* | |
5752 | * Now reset ->csum_bytes to what it should be. If bytes is | |
5753 | * more than to_free then we would have free'd more space had we | |
5754 | * not had an artificially high ->csum_bytes, so we need to free | |
5755 | * the remainder. If bytes is the same or less then we don't | |
5756 | * need to do anything, the other free-ers did the correct | |
5757 | * thing. | |
5758 | */ | |
5759 | BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes; | |
5760 | if (bytes > to_free) | |
5761 | to_free = bytes - to_free; | |
5762 | else | |
5763 | to_free = 0; | |
5764 | } | |
88e081bf | 5765 | spin_unlock(&BTRFS_I(inode)->lock); |
e2d1f923 | 5766 | if (dropped) |
88e081bf WS |
5767 | to_free += btrfs_calc_trans_metadata_size(root, dropped); |
5768 | ||
5769 | if (to_free) { | |
5770 | btrfs_block_rsv_release(root, block_rsv, to_free); | |
5771 | trace_btrfs_space_reservation(root->fs_info, "delalloc", | |
5772 | btrfs_ino(inode), to_free, 0); | |
5773 | } | |
5774 | if (delalloc_lock) | |
5775 | mutex_unlock(&BTRFS_I(inode)->delalloc_mutex); | |
5776 | return ret; | |
0ca1f7ce YZ |
5777 | } |
5778 | ||
7709cde3 JB |
5779 | /** |
5780 | * btrfs_delalloc_release_metadata - release a metadata reservation for an inode | |
5781 | * @inode: the inode to release the reservation for | |
5782 | * @num_bytes: the number of bytes we're releasing | |
5783 | * | |
5784 | * This will release the metadata reservation for an inode. This can be called | |
5785 | * once we complete IO for a given set of bytes to release their metadata | |
5786 | * reservations. | |
5787 | */ | |
0ca1f7ce YZ |
5788 | void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes) |
5789 | { | |
5790 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
9e0baf60 JB |
5791 | u64 to_free = 0; |
5792 | unsigned dropped; | |
0ca1f7ce YZ |
5793 | |
5794 | num_bytes = ALIGN(num_bytes, root->sectorsize); | |
7709cde3 | 5795 | spin_lock(&BTRFS_I(inode)->lock); |
dcab6a3b | 5796 | dropped = drop_outstanding_extent(inode, num_bytes); |
97e728d4 | 5797 | |
0934856d MX |
5798 | if (num_bytes) |
5799 | to_free = calc_csum_metadata_size(inode, num_bytes, 0); | |
7709cde3 | 5800 | spin_unlock(&BTRFS_I(inode)->lock); |
9e0baf60 JB |
5801 | if (dropped > 0) |
5802 | to_free += btrfs_calc_trans_metadata_size(root, dropped); | |
0ca1f7ce | 5803 | |
6a3891c5 JB |
5804 | if (btrfs_test_is_dummy_root(root)) |
5805 | return; | |
5806 | ||
8c2a3ca2 JB |
5807 | trace_btrfs_space_reservation(root->fs_info, "delalloc", |
5808 | btrfs_ino(inode), to_free, 0); | |
c5567237 | 5809 | |
0ca1f7ce YZ |
5810 | btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv, |
5811 | to_free); | |
5812 | } | |
5813 | ||
1ada3a62 | 5814 | /** |
7cf5b976 | 5815 | * btrfs_delalloc_reserve_space - reserve data and metadata space for |
1ada3a62 QW |
5816 | * delalloc |
5817 | * @inode: inode we're writing to | |
5818 | * @start: start range we are writing to | |
5819 | * @len: how long the range we are writing to | |
5820 | * | |
5821 | * TODO: This function will finally replace old btrfs_delalloc_reserve_space() | |
5822 | * | |
5823 | * This will do the following things | |
5824 | * | |
5825 | * o reserve space in data space info for num bytes | |
5826 | * and reserve precious corresponding qgroup space | |
5827 | * (Done in check_data_free_space) | |
5828 | * | |
5829 | * o reserve space for metadata space, based on the number of outstanding | |
5830 | * extents and how much csums will be needed | |
5831 | * also reserve metadata space in a per root over-reserve method. | |
5832 | * o add to the inodes->delalloc_bytes | |
5833 | * o add it to the fs_info's delalloc inodes list. | |
5834 | * (Above 3 all done in delalloc_reserve_metadata) | |
5835 | * | |
5836 | * Return 0 for success | |
5837 | * Return <0 for error(-ENOSPC or -EQUOT) | |
5838 | */ | |
7cf5b976 | 5839 | int btrfs_delalloc_reserve_space(struct inode *inode, u64 start, u64 len) |
1ada3a62 QW |
5840 | { |
5841 | int ret; | |
5842 | ||
7cf5b976 | 5843 | ret = btrfs_check_data_free_space(inode, start, len); |
1ada3a62 QW |
5844 | if (ret < 0) |
5845 | return ret; | |
5846 | ret = btrfs_delalloc_reserve_metadata(inode, len); | |
5847 | if (ret < 0) | |
7cf5b976 | 5848 | btrfs_free_reserved_data_space(inode, start, len); |
1ada3a62 QW |
5849 | return ret; |
5850 | } | |
5851 | ||
7709cde3 | 5852 | /** |
7cf5b976 | 5853 | * btrfs_delalloc_release_space - release data and metadata space for delalloc |
1ada3a62 QW |
5854 | * @inode: inode we're releasing space for |
5855 | * @start: start position of the space already reserved | |
5856 | * @len: the len of the space already reserved | |
5857 | * | |
5858 | * This must be matched with a call to btrfs_delalloc_reserve_space. This is | |
5859 | * called in the case that we don't need the metadata AND data reservations | |
5860 | * anymore. So if there is an error or we insert an inline extent. | |
5861 | * | |
5862 | * This function will release the metadata space that was not used and will | |
5863 | * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes | |
5864 | * list if there are no delalloc bytes left. | |
5865 | * Also it will handle the qgroup reserved space. | |
5866 | */ | |
7cf5b976 | 5867 | void btrfs_delalloc_release_space(struct inode *inode, u64 start, u64 len) |
1ada3a62 QW |
5868 | { |
5869 | btrfs_delalloc_release_metadata(inode, len); | |
7cf5b976 | 5870 | btrfs_free_reserved_data_space(inode, start, len); |
6324fbf3 CM |
5871 | } |
5872 | ||
ce93ec54 JB |
5873 | static int update_block_group(struct btrfs_trans_handle *trans, |
5874 | struct btrfs_root *root, u64 bytenr, | |
5875 | u64 num_bytes, int alloc) | |
9078a3e1 | 5876 | { |
0af3d00b | 5877 | struct btrfs_block_group_cache *cache = NULL; |
9078a3e1 | 5878 | struct btrfs_fs_info *info = root->fs_info; |
db94535d | 5879 | u64 total = num_bytes; |
9078a3e1 | 5880 | u64 old_val; |
db94535d | 5881 | u64 byte_in_group; |
0af3d00b | 5882 | int factor; |
3e1ad54f | 5883 | |
5d4f98a2 | 5884 | /* block accounting for super block */ |
eb73c1b7 | 5885 | spin_lock(&info->delalloc_root_lock); |
6c41761f | 5886 | old_val = btrfs_super_bytes_used(info->super_copy); |
5d4f98a2 YZ |
5887 | if (alloc) |
5888 | old_val += num_bytes; | |
5889 | else | |
5890 | old_val -= num_bytes; | |
6c41761f | 5891 | btrfs_set_super_bytes_used(info->super_copy, old_val); |
eb73c1b7 | 5892 | spin_unlock(&info->delalloc_root_lock); |
5d4f98a2 | 5893 | |
d397712b | 5894 | while (total) { |
db94535d | 5895 | cache = btrfs_lookup_block_group(info, bytenr); |
f3465ca4 | 5896 | if (!cache) |
79787eaa | 5897 | return -ENOENT; |
b742bb82 YZ |
5898 | if (cache->flags & (BTRFS_BLOCK_GROUP_DUP | |
5899 | BTRFS_BLOCK_GROUP_RAID1 | | |
5900 | BTRFS_BLOCK_GROUP_RAID10)) | |
5901 | factor = 2; | |
5902 | else | |
5903 | factor = 1; | |
9d66e233 JB |
5904 | /* |
5905 | * If this block group has free space cache written out, we | |
5906 | * need to make sure to load it if we are removing space. This | |
5907 | * is because we need the unpinning stage to actually add the | |
5908 | * space back to the block group, otherwise we will leak space. | |
5909 | */ | |
5910 | if (!alloc && cache->cached == BTRFS_CACHE_NO) | |
f6373bf3 | 5911 | cache_block_group(cache, 1); |
0af3d00b | 5912 | |
db94535d CM |
5913 | byte_in_group = bytenr - cache->key.objectid; |
5914 | WARN_ON(byte_in_group > cache->key.offset); | |
9078a3e1 | 5915 | |
25179201 | 5916 | spin_lock(&cache->space_info->lock); |
c286ac48 | 5917 | spin_lock(&cache->lock); |
0af3d00b | 5918 | |
73bc1876 | 5919 | if (btrfs_test_opt(root, SPACE_CACHE) && |
0af3d00b JB |
5920 | cache->disk_cache_state < BTRFS_DC_CLEAR) |
5921 | cache->disk_cache_state = BTRFS_DC_CLEAR; | |
5922 | ||
9078a3e1 | 5923 | old_val = btrfs_block_group_used(&cache->item); |
db94535d | 5924 | num_bytes = min(total, cache->key.offset - byte_in_group); |
cd1bc465 | 5925 | if (alloc) { |
db94535d | 5926 | old_val += num_bytes; |
11833d66 YZ |
5927 | btrfs_set_block_group_used(&cache->item, old_val); |
5928 | cache->reserved -= num_bytes; | |
11833d66 | 5929 | cache->space_info->bytes_reserved -= num_bytes; |
b742bb82 YZ |
5930 | cache->space_info->bytes_used += num_bytes; |
5931 | cache->space_info->disk_used += num_bytes * factor; | |
c286ac48 | 5932 | spin_unlock(&cache->lock); |
25179201 | 5933 | spin_unlock(&cache->space_info->lock); |
cd1bc465 | 5934 | } else { |
db94535d | 5935 | old_val -= num_bytes; |
ae0ab003 FM |
5936 | btrfs_set_block_group_used(&cache->item, old_val); |
5937 | cache->pinned += num_bytes; | |
5938 | cache->space_info->bytes_pinned += num_bytes; | |
5939 | cache->space_info->bytes_used -= num_bytes; | |
5940 | cache->space_info->disk_used -= num_bytes * factor; | |
5941 | spin_unlock(&cache->lock); | |
5942 | spin_unlock(&cache->space_info->lock); | |
47ab2a6c | 5943 | |
ae0ab003 FM |
5944 | set_extent_dirty(info->pinned_extents, |
5945 | bytenr, bytenr + num_bytes - 1, | |
5946 | GFP_NOFS | __GFP_NOFAIL); | |
cd1bc465 | 5947 | } |
1bbc621e CM |
5948 | |
5949 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
5950 | if (list_empty(&cache->dirty_list)) { | |
5951 | list_add_tail(&cache->dirty_list, | |
5952 | &trans->transaction->dirty_bgs); | |
5953 | trans->transaction->num_dirty_bgs++; | |
5954 | btrfs_get_block_group(cache); | |
5955 | } | |
5956 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
5957 | ||
036a9348 FM |
5958 | /* |
5959 | * No longer have used bytes in this block group, queue it for | |
5960 | * deletion. We do this after adding the block group to the | |
5961 | * dirty list to avoid races between cleaner kthread and space | |
5962 | * cache writeout. | |
5963 | */ | |
5964 | if (!alloc && old_val == 0) { | |
5965 | spin_lock(&info->unused_bgs_lock); | |
5966 | if (list_empty(&cache->bg_list)) { | |
5967 | btrfs_get_block_group(cache); | |
5968 | list_add_tail(&cache->bg_list, | |
5969 | &info->unused_bgs); | |
5970 | } | |
5971 | spin_unlock(&info->unused_bgs_lock); | |
5972 | } | |
5973 | ||
fa9c0d79 | 5974 | btrfs_put_block_group(cache); |
db94535d CM |
5975 | total -= num_bytes; |
5976 | bytenr += num_bytes; | |
9078a3e1 CM |
5977 | } |
5978 | return 0; | |
5979 | } | |
6324fbf3 | 5980 | |
a061fc8d CM |
5981 | static u64 first_logical_byte(struct btrfs_root *root, u64 search_start) |
5982 | { | |
0f9dd46c | 5983 | struct btrfs_block_group_cache *cache; |
d2fb3437 | 5984 | u64 bytenr; |
0f9dd46c | 5985 | |
a1897fdd LB |
5986 | spin_lock(&root->fs_info->block_group_cache_lock); |
5987 | bytenr = root->fs_info->first_logical_byte; | |
5988 | spin_unlock(&root->fs_info->block_group_cache_lock); | |
5989 | ||
5990 | if (bytenr < (u64)-1) | |
5991 | return bytenr; | |
5992 | ||
0f9dd46c JB |
5993 | cache = btrfs_lookup_first_block_group(root->fs_info, search_start); |
5994 | if (!cache) | |
a061fc8d | 5995 | return 0; |
0f9dd46c | 5996 | |
d2fb3437 | 5997 | bytenr = cache->key.objectid; |
fa9c0d79 | 5998 | btrfs_put_block_group(cache); |
d2fb3437 YZ |
5999 | |
6000 | return bytenr; | |
a061fc8d CM |
6001 | } |
6002 | ||
f0486c68 YZ |
6003 | static int pin_down_extent(struct btrfs_root *root, |
6004 | struct btrfs_block_group_cache *cache, | |
6005 | u64 bytenr, u64 num_bytes, int reserved) | |
324ae4df | 6006 | { |
11833d66 YZ |
6007 | spin_lock(&cache->space_info->lock); |
6008 | spin_lock(&cache->lock); | |
6009 | cache->pinned += num_bytes; | |
6010 | cache->space_info->bytes_pinned += num_bytes; | |
6011 | if (reserved) { | |
6012 | cache->reserved -= num_bytes; | |
6013 | cache->space_info->bytes_reserved -= num_bytes; | |
6014 | } | |
6015 | spin_unlock(&cache->lock); | |
6016 | spin_unlock(&cache->space_info->lock); | |
68b38550 | 6017 | |
f0486c68 YZ |
6018 | set_extent_dirty(root->fs_info->pinned_extents, bytenr, |
6019 | bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL); | |
e2d1f923 | 6020 | if (reserved) |
0be5dc67 | 6021 | trace_btrfs_reserved_extent_free(root, bytenr, num_bytes); |
f0486c68 YZ |
6022 | return 0; |
6023 | } | |
68b38550 | 6024 | |
f0486c68 YZ |
6025 | /* |
6026 | * this function must be called within transaction | |
6027 | */ | |
6028 | int btrfs_pin_extent(struct btrfs_root *root, | |
6029 | u64 bytenr, u64 num_bytes, int reserved) | |
6030 | { | |
6031 | struct btrfs_block_group_cache *cache; | |
68b38550 | 6032 | |
f0486c68 | 6033 | cache = btrfs_lookup_block_group(root->fs_info, bytenr); |
79787eaa | 6034 | BUG_ON(!cache); /* Logic error */ |
f0486c68 YZ |
6035 | |
6036 | pin_down_extent(root, cache, bytenr, num_bytes, reserved); | |
6037 | ||
6038 | btrfs_put_block_group(cache); | |
11833d66 YZ |
6039 | return 0; |
6040 | } | |
6041 | ||
f0486c68 | 6042 | /* |
e688b725 CM |
6043 | * this function must be called within transaction |
6044 | */ | |
dcfac415 | 6045 | int btrfs_pin_extent_for_log_replay(struct btrfs_root *root, |
e688b725 CM |
6046 | u64 bytenr, u64 num_bytes) |
6047 | { | |
6048 | struct btrfs_block_group_cache *cache; | |
b50c6e25 | 6049 | int ret; |
e688b725 CM |
6050 | |
6051 | cache = btrfs_lookup_block_group(root->fs_info, bytenr); | |
b50c6e25 JB |
6052 | if (!cache) |
6053 | return -EINVAL; | |
e688b725 CM |
6054 | |
6055 | /* | |
6056 | * pull in the free space cache (if any) so that our pin | |
6057 | * removes the free space from the cache. We have load_only set | |
6058 | * to one because the slow code to read in the free extents does check | |
6059 | * the pinned extents. | |
6060 | */ | |
f6373bf3 | 6061 | cache_block_group(cache, 1); |
e688b725 CM |
6062 | |
6063 | pin_down_extent(root, cache, bytenr, num_bytes, 0); | |
6064 | ||
6065 | /* remove us from the free space cache (if we're there at all) */ | |
b50c6e25 | 6066 | ret = btrfs_remove_free_space(cache, bytenr, num_bytes); |
e688b725 | 6067 | btrfs_put_block_group(cache); |
b50c6e25 | 6068 | return ret; |
e688b725 CM |
6069 | } |
6070 | ||
8c2a1a30 JB |
6071 | static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes) |
6072 | { | |
6073 | int ret; | |
6074 | struct btrfs_block_group_cache *block_group; | |
6075 | struct btrfs_caching_control *caching_ctl; | |
6076 | ||
6077 | block_group = btrfs_lookup_block_group(root->fs_info, start); | |
6078 | if (!block_group) | |
6079 | return -EINVAL; | |
6080 | ||
6081 | cache_block_group(block_group, 0); | |
6082 | caching_ctl = get_caching_control(block_group); | |
6083 | ||
6084 | if (!caching_ctl) { | |
6085 | /* Logic error */ | |
6086 | BUG_ON(!block_group_cache_done(block_group)); | |
6087 | ret = btrfs_remove_free_space(block_group, start, num_bytes); | |
6088 | } else { | |
6089 | mutex_lock(&caching_ctl->mutex); | |
6090 | ||
6091 | if (start >= caching_ctl->progress) { | |
6092 | ret = add_excluded_extent(root, start, num_bytes); | |
6093 | } else if (start + num_bytes <= caching_ctl->progress) { | |
6094 | ret = btrfs_remove_free_space(block_group, | |
6095 | start, num_bytes); | |
6096 | } else { | |
6097 | num_bytes = caching_ctl->progress - start; | |
6098 | ret = btrfs_remove_free_space(block_group, | |
6099 | start, num_bytes); | |
6100 | if (ret) | |
6101 | goto out_lock; | |
6102 | ||
6103 | num_bytes = (start + num_bytes) - | |
6104 | caching_ctl->progress; | |
6105 | start = caching_ctl->progress; | |
6106 | ret = add_excluded_extent(root, start, num_bytes); | |
6107 | } | |
6108 | out_lock: | |
6109 | mutex_unlock(&caching_ctl->mutex); | |
6110 | put_caching_control(caching_ctl); | |
6111 | } | |
6112 | btrfs_put_block_group(block_group); | |
6113 | return ret; | |
6114 | } | |
6115 | ||
6116 | int btrfs_exclude_logged_extents(struct btrfs_root *log, | |
6117 | struct extent_buffer *eb) | |
6118 | { | |
6119 | struct btrfs_file_extent_item *item; | |
6120 | struct btrfs_key key; | |
6121 | int found_type; | |
6122 | int i; | |
6123 | ||
6124 | if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS)) | |
6125 | return 0; | |
6126 | ||
6127 | for (i = 0; i < btrfs_header_nritems(eb); i++) { | |
6128 | btrfs_item_key_to_cpu(eb, &key, i); | |
6129 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
6130 | continue; | |
6131 | item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item); | |
6132 | found_type = btrfs_file_extent_type(eb, item); | |
6133 | if (found_type == BTRFS_FILE_EXTENT_INLINE) | |
6134 | continue; | |
6135 | if (btrfs_file_extent_disk_bytenr(eb, item) == 0) | |
6136 | continue; | |
6137 | key.objectid = btrfs_file_extent_disk_bytenr(eb, item); | |
6138 | key.offset = btrfs_file_extent_disk_num_bytes(eb, item); | |
6139 | __exclude_logged_extent(log, key.objectid, key.offset); | |
6140 | } | |
6141 | ||
6142 | return 0; | |
6143 | } | |
6144 | ||
fb25e914 JB |
6145 | /** |
6146 | * btrfs_update_reserved_bytes - update the block_group and space info counters | |
6147 | * @cache: The cache we are manipulating | |
6148 | * @num_bytes: The number of bytes in question | |
6149 | * @reserve: One of the reservation enums | |
e570fd27 | 6150 | * @delalloc: The blocks are allocated for the delalloc write |
fb25e914 JB |
6151 | * |
6152 | * This is called by the allocator when it reserves space, or by somebody who is | |
6153 | * freeing space that was never actually used on disk. For example if you | |
6154 | * reserve some space for a new leaf in transaction A and before transaction A | |
6155 | * commits you free that leaf, you call this with reserve set to 0 in order to | |
6156 | * clear the reservation. | |
6157 | * | |
6158 | * Metadata reservations should be called with RESERVE_ALLOC so we do the proper | |
6159 | * ENOSPC accounting. For data we handle the reservation through clearing the | |
6160 | * delalloc bits in the io_tree. We have to do this since we could end up | |
6161 | * allocating less disk space for the amount of data we have reserved in the | |
6162 | * case of compression. | |
6163 | * | |
6164 | * If this is a reservation and the block group has become read only we cannot | |
6165 | * make the reservation and return -EAGAIN, otherwise this function always | |
6166 | * succeeds. | |
f0486c68 | 6167 | */ |
fb25e914 | 6168 | static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache, |
e570fd27 | 6169 | u64 num_bytes, int reserve, int delalloc) |
11833d66 | 6170 | { |
fb25e914 | 6171 | struct btrfs_space_info *space_info = cache->space_info; |
f0486c68 | 6172 | int ret = 0; |
79787eaa | 6173 | |
fb25e914 JB |
6174 | spin_lock(&space_info->lock); |
6175 | spin_lock(&cache->lock); | |
6176 | if (reserve != RESERVE_FREE) { | |
f0486c68 YZ |
6177 | if (cache->ro) { |
6178 | ret = -EAGAIN; | |
6179 | } else { | |
fb25e914 JB |
6180 | cache->reserved += num_bytes; |
6181 | space_info->bytes_reserved += num_bytes; | |
6182 | if (reserve == RESERVE_ALLOC) { | |
8c2a3ca2 | 6183 | trace_btrfs_space_reservation(cache->fs_info, |
2bcc0328 LB |
6184 | "space_info", space_info->flags, |
6185 | num_bytes, 0); | |
fb25e914 JB |
6186 | space_info->bytes_may_use -= num_bytes; |
6187 | } | |
e570fd27 MX |
6188 | |
6189 | if (delalloc) | |
6190 | cache->delalloc_bytes += num_bytes; | |
f0486c68 | 6191 | } |
fb25e914 JB |
6192 | } else { |
6193 | if (cache->ro) | |
6194 | space_info->bytes_readonly += num_bytes; | |
6195 | cache->reserved -= num_bytes; | |
6196 | space_info->bytes_reserved -= num_bytes; | |
e570fd27 MX |
6197 | |
6198 | if (delalloc) | |
6199 | cache->delalloc_bytes -= num_bytes; | |
324ae4df | 6200 | } |
fb25e914 JB |
6201 | spin_unlock(&cache->lock); |
6202 | spin_unlock(&space_info->lock); | |
f0486c68 | 6203 | return ret; |
324ae4df | 6204 | } |
9078a3e1 | 6205 | |
143bede5 | 6206 | void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans, |
11833d66 | 6207 | struct btrfs_root *root) |
e8569813 | 6208 | { |
e8569813 | 6209 | struct btrfs_fs_info *fs_info = root->fs_info; |
11833d66 YZ |
6210 | struct btrfs_caching_control *next; |
6211 | struct btrfs_caching_control *caching_ctl; | |
6212 | struct btrfs_block_group_cache *cache; | |
e8569813 | 6213 | |
9e351cc8 | 6214 | down_write(&fs_info->commit_root_sem); |
25179201 | 6215 | |
11833d66 YZ |
6216 | list_for_each_entry_safe(caching_ctl, next, |
6217 | &fs_info->caching_block_groups, list) { | |
6218 | cache = caching_ctl->block_group; | |
6219 | if (block_group_cache_done(cache)) { | |
6220 | cache->last_byte_to_unpin = (u64)-1; | |
6221 | list_del_init(&caching_ctl->list); | |
6222 | put_caching_control(caching_ctl); | |
e8569813 | 6223 | } else { |
11833d66 | 6224 | cache->last_byte_to_unpin = caching_ctl->progress; |
e8569813 | 6225 | } |
e8569813 | 6226 | } |
11833d66 YZ |
6227 | |
6228 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) | |
6229 | fs_info->pinned_extents = &fs_info->freed_extents[1]; | |
6230 | else | |
6231 | fs_info->pinned_extents = &fs_info->freed_extents[0]; | |
6232 | ||
9e351cc8 | 6233 | up_write(&fs_info->commit_root_sem); |
8929ecfa YZ |
6234 | |
6235 | update_global_block_rsv(fs_info); | |
e8569813 ZY |
6236 | } |
6237 | ||
c759c4e1 JB |
6238 | /* |
6239 | * Returns the free cluster for the given space info and sets empty_cluster to | |
6240 | * what it should be based on the mount options. | |
6241 | */ | |
6242 | static struct btrfs_free_cluster * | |
6243 | fetch_cluster_info(struct btrfs_root *root, struct btrfs_space_info *space_info, | |
6244 | u64 *empty_cluster) | |
6245 | { | |
6246 | struct btrfs_free_cluster *ret = NULL; | |
6247 | bool ssd = btrfs_test_opt(root, SSD); | |
6248 | ||
6249 | *empty_cluster = 0; | |
6250 | if (btrfs_mixed_space_info(space_info)) | |
6251 | return ret; | |
6252 | ||
6253 | if (ssd) | |
6254 | *empty_cluster = 2 * 1024 * 1024; | |
6255 | if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) { | |
6256 | ret = &root->fs_info->meta_alloc_cluster; | |
6257 | if (!ssd) | |
6258 | *empty_cluster = 64 * 1024; | |
6259 | } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) && ssd) { | |
6260 | ret = &root->fs_info->data_alloc_cluster; | |
6261 | } | |
6262 | ||
6263 | return ret; | |
6264 | } | |
6265 | ||
678886bd FM |
6266 | static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end, |
6267 | const bool return_free_space) | |
ccd467d6 | 6268 | { |
11833d66 YZ |
6269 | struct btrfs_fs_info *fs_info = root->fs_info; |
6270 | struct btrfs_block_group_cache *cache = NULL; | |
7b398f8e JB |
6271 | struct btrfs_space_info *space_info; |
6272 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
c759c4e1 | 6273 | struct btrfs_free_cluster *cluster = NULL; |
11833d66 | 6274 | u64 len; |
c759c4e1 JB |
6275 | u64 total_unpinned = 0; |
6276 | u64 empty_cluster = 0; | |
7b398f8e | 6277 | bool readonly; |
ccd467d6 | 6278 | |
11833d66 | 6279 | while (start <= end) { |
7b398f8e | 6280 | readonly = false; |
11833d66 YZ |
6281 | if (!cache || |
6282 | start >= cache->key.objectid + cache->key.offset) { | |
6283 | if (cache) | |
6284 | btrfs_put_block_group(cache); | |
c759c4e1 | 6285 | total_unpinned = 0; |
11833d66 | 6286 | cache = btrfs_lookup_block_group(fs_info, start); |
79787eaa | 6287 | BUG_ON(!cache); /* Logic error */ |
c759c4e1 JB |
6288 | |
6289 | cluster = fetch_cluster_info(root, | |
6290 | cache->space_info, | |
6291 | &empty_cluster); | |
6292 | empty_cluster <<= 1; | |
11833d66 YZ |
6293 | } |
6294 | ||
6295 | len = cache->key.objectid + cache->key.offset - start; | |
6296 | len = min(len, end + 1 - start); | |
6297 | ||
6298 | if (start < cache->last_byte_to_unpin) { | |
6299 | len = min(len, cache->last_byte_to_unpin - start); | |
678886bd FM |
6300 | if (return_free_space) |
6301 | btrfs_add_free_space(cache, start, len); | |
11833d66 YZ |
6302 | } |
6303 | ||
f0486c68 | 6304 | start += len; |
c759c4e1 | 6305 | total_unpinned += len; |
7b398f8e | 6306 | space_info = cache->space_info; |
f0486c68 | 6307 | |
c759c4e1 JB |
6308 | /* |
6309 | * If this space cluster has been marked as fragmented and we've | |
6310 | * unpinned enough in this block group to potentially allow a | |
6311 | * cluster to be created inside of it go ahead and clear the | |
6312 | * fragmented check. | |
6313 | */ | |
6314 | if (cluster && cluster->fragmented && | |
6315 | total_unpinned > empty_cluster) { | |
6316 | spin_lock(&cluster->lock); | |
6317 | cluster->fragmented = 0; | |
6318 | spin_unlock(&cluster->lock); | |
6319 | } | |
6320 | ||
7b398f8e | 6321 | spin_lock(&space_info->lock); |
11833d66 YZ |
6322 | spin_lock(&cache->lock); |
6323 | cache->pinned -= len; | |
7b398f8e | 6324 | space_info->bytes_pinned -= len; |
4f4db217 | 6325 | space_info->max_extent_size = 0; |
d288db5d | 6326 | percpu_counter_add(&space_info->total_bytes_pinned, -len); |
7b398f8e JB |
6327 | if (cache->ro) { |
6328 | space_info->bytes_readonly += len; | |
6329 | readonly = true; | |
6330 | } | |
11833d66 | 6331 | spin_unlock(&cache->lock); |
7b398f8e JB |
6332 | if (!readonly && global_rsv->space_info == space_info) { |
6333 | spin_lock(&global_rsv->lock); | |
6334 | if (!global_rsv->full) { | |
6335 | len = min(len, global_rsv->size - | |
6336 | global_rsv->reserved); | |
6337 | global_rsv->reserved += len; | |
6338 | space_info->bytes_may_use += len; | |
6339 | if (global_rsv->reserved >= global_rsv->size) | |
6340 | global_rsv->full = 1; | |
6341 | } | |
6342 | spin_unlock(&global_rsv->lock); | |
6343 | } | |
6344 | spin_unlock(&space_info->lock); | |
ccd467d6 | 6345 | } |
11833d66 YZ |
6346 | |
6347 | if (cache) | |
6348 | btrfs_put_block_group(cache); | |
ccd467d6 CM |
6349 | return 0; |
6350 | } | |
6351 | ||
6352 | int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans, | |
11833d66 | 6353 | struct btrfs_root *root) |
a28ec197 | 6354 | { |
11833d66 | 6355 | struct btrfs_fs_info *fs_info = root->fs_info; |
e33e17ee JM |
6356 | struct btrfs_block_group_cache *block_group, *tmp; |
6357 | struct list_head *deleted_bgs; | |
11833d66 | 6358 | struct extent_io_tree *unpin; |
1a5bc167 CM |
6359 | u64 start; |
6360 | u64 end; | |
a28ec197 | 6361 | int ret; |
a28ec197 | 6362 | |
11833d66 YZ |
6363 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) |
6364 | unpin = &fs_info->freed_extents[1]; | |
6365 | else | |
6366 | unpin = &fs_info->freed_extents[0]; | |
6367 | ||
e33e17ee | 6368 | while (!trans->aborted) { |
d4b450cd | 6369 | mutex_lock(&fs_info->unused_bg_unpin_mutex); |
1a5bc167 | 6370 | ret = find_first_extent_bit(unpin, 0, &start, &end, |
e6138876 | 6371 | EXTENT_DIRTY, NULL); |
d4b450cd FM |
6372 | if (ret) { |
6373 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); | |
a28ec197 | 6374 | break; |
d4b450cd | 6375 | } |
1f3c79a2 | 6376 | |
5378e607 LD |
6377 | if (btrfs_test_opt(root, DISCARD)) |
6378 | ret = btrfs_discard_extent(root, start, | |
6379 | end + 1 - start, NULL); | |
1f3c79a2 | 6380 | |
1a5bc167 | 6381 | clear_extent_dirty(unpin, start, end, GFP_NOFS); |
678886bd | 6382 | unpin_extent_range(root, start, end, true); |
d4b450cd | 6383 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
b9473439 | 6384 | cond_resched(); |
a28ec197 | 6385 | } |
817d52f8 | 6386 | |
e33e17ee JM |
6387 | /* |
6388 | * Transaction is finished. We don't need the lock anymore. We | |
6389 | * do need to clean up the block groups in case of a transaction | |
6390 | * abort. | |
6391 | */ | |
6392 | deleted_bgs = &trans->transaction->deleted_bgs; | |
6393 | list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) { | |
6394 | u64 trimmed = 0; | |
6395 | ||
6396 | ret = -EROFS; | |
6397 | if (!trans->aborted) | |
6398 | ret = btrfs_discard_extent(root, | |
6399 | block_group->key.objectid, | |
6400 | block_group->key.offset, | |
6401 | &trimmed); | |
6402 | ||
6403 | list_del_init(&block_group->bg_list); | |
6404 | btrfs_put_block_group_trimming(block_group); | |
6405 | btrfs_put_block_group(block_group); | |
6406 | ||
6407 | if (ret) { | |
6408 | const char *errstr = btrfs_decode_error(ret); | |
6409 | btrfs_warn(fs_info, | |
6410 | "Discard failed while removing blockgroup: errno=%d %s\n", | |
6411 | ret, errstr); | |
6412 | } | |
6413 | } | |
6414 | ||
e20d96d6 CM |
6415 | return 0; |
6416 | } | |
6417 | ||
b150a4f1 JB |
6418 | static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes, |
6419 | u64 owner, u64 root_objectid) | |
6420 | { | |
6421 | struct btrfs_space_info *space_info; | |
6422 | u64 flags; | |
6423 | ||
6424 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
6425 | if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID) | |
6426 | flags = BTRFS_BLOCK_GROUP_SYSTEM; | |
6427 | else | |
6428 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
6429 | } else { | |
6430 | flags = BTRFS_BLOCK_GROUP_DATA; | |
6431 | } | |
6432 | ||
6433 | space_info = __find_space_info(fs_info, flags); | |
6434 | BUG_ON(!space_info); /* Logic bug */ | |
6435 | percpu_counter_add(&space_info->total_bytes_pinned, num_bytes); | |
6436 | } | |
6437 | ||
6438 | ||
5d4f98a2 YZ |
6439 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
6440 | struct btrfs_root *root, | |
c682f9b3 | 6441 | struct btrfs_delayed_ref_node *node, u64 parent, |
5d4f98a2 YZ |
6442 | u64 root_objectid, u64 owner_objectid, |
6443 | u64 owner_offset, int refs_to_drop, | |
c682f9b3 | 6444 | struct btrfs_delayed_extent_op *extent_op) |
a28ec197 | 6445 | { |
e2fa7227 | 6446 | struct btrfs_key key; |
5d4f98a2 | 6447 | struct btrfs_path *path; |
1261ec42 CM |
6448 | struct btrfs_fs_info *info = root->fs_info; |
6449 | struct btrfs_root *extent_root = info->extent_root; | |
5f39d397 | 6450 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
6451 | struct btrfs_extent_item *ei; |
6452 | struct btrfs_extent_inline_ref *iref; | |
a28ec197 | 6453 | int ret; |
5d4f98a2 | 6454 | int is_data; |
952fccac CM |
6455 | int extent_slot = 0; |
6456 | int found_extent = 0; | |
6457 | int num_to_del = 1; | |
5d4f98a2 YZ |
6458 | u32 item_size; |
6459 | u64 refs; | |
c682f9b3 QW |
6460 | u64 bytenr = node->bytenr; |
6461 | u64 num_bytes = node->num_bytes; | |
fcebe456 | 6462 | int last_ref = 0; |
3173a18f JB |
6463 | bool skinny_metadata = btrfs_fs_incompat(root->fs_info, |
6464 | SKINNY_METADATA); | |
037e6390 | 6465 | |
5caf2a00 | 6466 | path = btrfs_alloc_path(); |
54aa1f4d CM |
6467 | if (!path) |
6468 | return -ENOMEM; | |
5f26f772 | 6469 | |
3c12ac72 | 6470 | path->reada = 1; |
b9473439 | 6471 | path->leave_spinning = 1; |
5d4f98a2 YZ |
6472 | |
6473 | is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID; | |
6474 | BUG_ON(!is_data && refs_to_drop != 1); | |
6475 | ||
3173a18f JB |
6476 | if (is_data) |
6477 | skinny_metadata = 0; | |
6478 | ||
5d4f98a2 YZ |
6479 | ret = lookup_extent_backref(trans, extent_root, path, &iref, |
6480 | bytenr, num_bytes, parent, | |
6481 | root_objectid, owner_objectid, | |
6482 | owner_offset); | |
7bb86316 | 6483 | if (ret == 0) { |
952fccac | 6484 | extent_slot = path->slots[0]; |
5d4f98a2 YZ |
6485 | while (extent_slot >= 0) { |
6486 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
952fccac | 6487 | extent_slot); |
5d4f98a2 | 6488 | if (key.objectid != bytenr) |
952fccac | 6489 | break; |
5d4f98a2 YZ |
6490 | if (key.type == BTRFS_EXTENT_ITEM_KEY && |
6491 | key.offset == num_bytes) { | |
952fccac CM |
6492 | found_extent = 1; |
6493 | break; | |
6494 | } | |
3173a18f JB |
6495 | if (key.type == BTRFS_METADATA_ITEM_KEY && |
6496 | key.offset == owner_objectid) { | |
6497 | found_extent = 1; | |
6498 | break; | |
6499 | } | |
952fccac CM |
6500 | if (path->slots[0] - extent_slot > 5) |
6501 | break; | |
5d4f98a2 | 6502 | extent_slot--; |
952fccac | 6503 | } |
5d4f98a2 YZ |
6504 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
6505 | item_size = btrfs_item_size_nr(path->nodes[0], extent_slot); | |
6506 | if (found_extent && item_size < sizeof(*ei)) | |
6507 | found_extent = 0; | |
6508 | #endif | |
31840ae1 | 6509 | if (!found_extent) { |
5d4f98a2 | 6510 | BUG_ON(iref); |
56bec294 | 6511 | ret = remove_extent_backref(trans, extent_root, path, |
5d4f98a2 | 6512 | NULL, refs_to_drop, |
fcebe456 | 6513 | is_data, &last_ref); |
005d6427 DS |
6514 | if (ret) { |
6515 | btrfs_abort_transaction(trans, extent_root, ret); | |
6516 | goto out; | |
6517 | } | |
b3b4aa74 | 6518 | btrfs_release_path(path); |
b9473439 | 6519 | path->leave_spinning = 1; |
5d4f98a2 YZ |
6520 | |
6521 | key.objectid = bytenr; | |
6522 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
6523 | key.offset = num_bytes; | |
6524 | ||
3173a18f JB |
6525 | if (!is_data && skinny_metadata) { |
6526 | key.type = BTRFS_METADATA_ITEM_KEY; | |
6527 | key.offset = owner_objectid; | |
6528 | } | |
6529 | ||
31840ae1 ZY |
6530 | ret = btrfs_search_slot(trans, extent_root, |
6531 | &key, path, -1, 1); | |
3173a18f JB |
6532 | if (ret > 0 && skinny_metadata && path->slots[0]) { |
6533 | /* | |
6534 | * Couldn't find our skinny metadata item, | |
6535 | * see if we have ye olde extent item. | |
6536 | */ | |
6537 | path->slots[0]--; | |
6538 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
6539 | path->slots[0]); | |
6540 | if (key.objectid == bytenr && | |
6541 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
6542 | key.offset == num_bytes) | |
6543 | ret = 0; | |
6544 | } | |
6545 | ||
6546 | if (ret > 0 && skinny_metadata) { | |
6547 | skinny_metadata = false; | |
9ce49a0b | 6548 | key.objectid = bytenr; |
3173a18f JB |
6549 | key.type = BTRFS_EXTENT_ITEM_KEY; |
6550 | key.offset = num_bytes; | |
6551 | btrfs_release_path(path); | |
6552 | ret = btrfs_search_slot(trans, extent_root, | |
6553 | &key, path, -1, 1); | |
6554 | } | |
6555 | ||
f3465ca4 | 6556 | if (ret) { |
c2cf52eb | 6557 | btrfs_err(info, "umm, got %d back from search, was looking for %llu", |
c1c9ff7c | 6558 | ret, bytenr); |
b783e62d JB |
6559 | if (ret > 0) |
6560 | btrfs_print_leaf(extent_root, | |
6561 | path->nodes[0]); | |
f3465ca4 | 6562 | } |
005d6427 DS |
6563 | if (ret < 0) { |
6564 | btrfs_abort_transaction(trans, extent_root, ret); | |
6565 | goto out; | |
6566 | } | |
31840ae1 ZY |
6567 | extent_slot = path->slots[0]; |
6568 | } | |
fae7f21c | 6569 | } else if (WARN_ON(ret == -ENOENT)) { |
7bb86316 | 6570 | btrfs_print_leaf(extent_root, path->nodes[0]); |
c2cf52eb SK |
6571 | btrfs_err(info, |
6572 | "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu", | |
c1c9ff7c GU |
6573 | bytenr, parent, root_objectid, owner_objectid, |
6574 | owner_offset); | |
c4a050bb JB |
6575 | btrfs_abort_transaction(trans, extent_root, ret); |
6576 | goto out; | |
79787eaa | 6577 | } else { |
005d6427 DS |
6578 | btrfs_abort_transaction(trans, extent_root, ret); |
6579 | goto out; | |
7bb86316 | 6580 | } |
5f39d397 CM |
6581 | |
6582 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
6583 | item_size = btrfs_item_size_nr(leaf, extent_slot); |
6584 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
6585 | if (item_size < sizeof(*ei)) { | |
6586 | BUG_ON(found_extent || extent_slot != path->slots[0]); | |
6587 | ret = convert_extent_item_v0(trans, extent_root, path, | |
6588 | owner_objectid, 0); | |
005d6427 DS |
6589 | if (ret < 0) { |
6590 | btrfs_abort_transaction(trans, extent_root, ret); | |
6591 | goto out; | |
6592 | } | |
5d4f98a2 | 6593 | |
b3b4aa74 | 6594 | btrfs_release_path(path); |
5d4f98a2 YZ |
6595 | path->leave_spinning = 1; |
6596 | ||
6597 | key.objectid = bytenr; | |
6598 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
6599 | key.offset = num_bytes; | |
6600 | ||
6601 | ret = btrfs_search_slot(trans, extent_root, &key, path, | |
6602 | -1, 1); | |
6603 | if (ret) { | |
c2cf52eb | 6604 | btrfs_err(info, "umm, got %d back from search, was looking for %llu", |
c1c9ff7c | 6605 | ret, bytenr); |
5d4f98a2 YZ |
6606 | btrfs_print_leaf(extent_root, path->nodes[0]); |
6607 | } | |
005d6427 DS |
6608 | if (ret < 0) { |
6609 | btrfs_abort_transaction(trans, extent_root, ret); | |
6610 | goto out; | |
6611 | } | |
6612 | ||
5d4f98a2 YZ |
6613 | extent_slot = path->slots[0]; |
6614 | leaf = path->nodes[0]; | |
6615 | item_size = btrfs_item_size_nr(leaf, extent_slot); | |
6616 | } | |
6617 | #endif | |
6618 | BUG_ON(item_size < sizeof(*ei)); | |
952fccac | 6619 | ei = btrfs_item_ptr(leaf, extent_slot, |
123abc88 | 6620 | struct btrfs_extent_item); |
3173a18f JB |
6621 | if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID && |
6622 | key.type == BTRFS_EXTENT_ITEM_KEY) { | |
5d4f98a2 YZ |
6623 | struct btrfs_tree_block_info *bi; |
6624 | BUG_ON(item_size < sizeof(*ei) + sizeof(*bi)); | |
6625 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
6626 | WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi)); | |
6627 | } | |
56bec294 | 6628 | |
5d4f98a2 | 6629 | refs = btrfs_extent_refs(leaf, ei); |
32b02538 JB |
6630 | if (refs < refs_to_drop) { |
6631 | btrfs_err(info, "trying to drop %d refs but we only have %Lu " | |
351fd353 | 6632 | "for bytenr %Lu", refs_to_drop, refs, bytenr); |
32b02538 JB |
6633 | ret = -EINVAL; |
6634 | btrfs_abort_transaction(trans, extent_root, ret); | |
6635 | goto out; | |
6636 | } | |
56bec294 | 6637 | refs -= refs_to_drop; |
5f39d397 | 6638 | |
5d4f98a2 YZ |
6639 | if (refs > 0) { |
6640 | if (extent_op) | |
6641 | __run_delayed_extent_op(extent_op, leaf, ei); | |
6642 | /* | |
6643 | * In the case of inline back ref, reference count will | |
6644 | * be updated by remove_extent_backref | |
952fccac | 6645 | */ |
5d4f98a2 YZ |
6646 | if (iref) { |
6647 | BUG_ON(!found_extent); | |
6648 | } else { | |
6649 | btrfs_set_extent_refs(leaf, ei, refs); | |
6650 | btrfs_mark_buffer_dirty(leaf); | |
6651 | } | |
6652 | if (found_extent) { | |
6653 | ret = remove_extent_backref(trans, extent_root, path, | |
6654 | iref, refs_to_drop, | |
fcebe456 | 6655 | is_data, &last_ref); |
005d6427 DS |
6656 | if (ret) { |
6657 | btrfs_abort_transaction(trans, extent_root, ret); | |
6658 | goto out; | |
6659 | } | |
952fccac | 6660 | } |
b150a4f1 JB |
6661 | add_pinned_bytes(root->fs_info, -num_bytes, owner_objectid, |
6662 | root_objectid); | |
5d4f98a2 | 6663 | } else { |
5d4f98a2 YZ |
6664 | if (found_extent) { |
6665 | BUG_ON(is_data && refs_to_drop != | |
9ed0dea0 | 6666 | extent_data_ref_count(path, iref)); |
5d4f98a2 YZ |
6667 | if (iref) { |
6668 | BUG_ON(path->slots[0] != extent_slot); | |
6669 | } else { | |
6670 | BUG_ON(path->slots[0] != extent_slot + 1); | |
6671 | path->slots[0] = extent_slot; | |
6672 | num_to_del = 2; | |
6673 | } | |
78fae27e | 6674 | } |
b9473439 | 6675 | |
fcebe456 | 6676 | last_ref = 1; |
952fccac CM |
6677 | ret = btrfs_del_items(trans, extent_root, path, path->slots[0], |
6678 | num_to_del); | |
005d6427 DS |
6679 | if (ret) { |
6680 | btrfs_abort_transaction(trans, extent_root, ret); | |
6681 | goto out; | |
6682 | } | |
b3b4aa74 | 6683 | btrfs_release_path(path); |
21af804c | 6684 | |
5d4f98a2 | 6685 | if (is_data) { |
459931ec | 6686 | ret = btrfs_del_csums(trans, root, bytenr, num_bytes); |
005d6427 DS |
6687 | if (ret) { |
6688 | btrfs_abort_transaction(trans, extent_root, ret); | |
6689 | goto out; | |
6690 | } | |
459931ec CM |
6691 | } |
6692 | ||
1e144fb8 OS |
6693 | ret = add_to_free_space_tree(trans, root->fs_info, bytenr, |
6694 | num_bytes); | |
6695 | if (ret) { | |
6696 | btrfs_abort_transaction(trans, extent_root, ret); | |
6697 | goto out; | |
6698 | } | |
6699 | ||
ce93ec54 | 6700 | ret = update_block_group(trans, root, bytenr, num_bytes, 0); |
005d6427 DS |
6701 | if (ret) { |
6702 | btrfs_abort_transaction(trans, extent_root, ret); | |
6703 | goto out; | |
6704 | } | |
a28ec197 | 6705 | } |
fcebe456 JB |
6706 | btrfs_release_path(path); |
6707 | ||
79787eaa | 6708 | out: |
5caf2a00 | 6709 | btrfs_free_path(path); |
a28ec197 CM |
6710 | return ret; |
6711 | } | |
6712 | ||
1887be66 | 6713 | /* |
f0486c68 | 6714 | * when we free an block, it is possible (and likely) that we free the last |
1887be66 CM |
6715 | * delayed ref for that extent as well. This searches the delayed ref tree for |
6716 | * a given extent, and if there are no other delayed refs to be processed, it | |
6717 | * removes it from the tree. | |
6718 | */ | |
6719 | static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans, | |
6720 | struct btrfs_root *root, u64 bytenr) | |
6721 | { | |
6722 | struct btrfs_delayed_ref_head *head; | |
6723 | struct btrfs_delayed_ref_root *delayed_refs; | |
f0486c68 | 6724 | int ret = 0; |
1887be66 CM |
6725 | |
6726 | delayed_refs = &trans->transaction->delayed_refs; | |
6727 | spin_lock(&delayed_refs->lock); | |
6728 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
6729 | if (!head) | |
cf93da7b | 6730 | goto out_delayed_unlock; |
1887be66 | 6731 | |
d7df2c79 | 6732 | spin_lock(&head->lock); |
c6fc2454 | 6733 | if (!list_empty(&head->ref_list)) |
1887be66 CM |
6734 | goto out; |
6735 | ||
5d4f98a2 YZ |
6736 | if (head->extent_op) { |
6737 | if (!head->must_insert_reserved) | |
6738 | goto out; | |
78a6184a | 6739 | btrfs_free_delayed_extent_op(head->extent_op); |
5d4f98a2 YZ |
6740 | head->extent_op = NULL; |
6741 | } | |
6742 | ||
1887be66 CM |
6743 | /* |
6744 | * waiting for the lock here would deadlock. If someone else has it | |
6745 | * locked they are already in the process of dropping it anyway | |
6746 | */ | |
6747 | if (!mutex_trylock(&head->mutex)) | |
6748 | goto out; | |
6749 | ||
6750 | /* | |
6751 | * at this point we have a head with no other entries. Go | |
6752 | * ahead and process it. | |
6753 | */ | |
6754 | head->node.in_tree = 0; | |
c46effa6 | 6755 | rb_erase(&head->href_node, &delayed_refs->href_root); |
c3e69d58 | 6756 | |
d7df2c79 | 6757 | atomic_dec(&delayed_refs->num_entries); |
1887be66 CM |
6758 | |
6759 | /* | |
6760 | * we don't take a ref on the node because we're removing it from the | |
6761 | * tree, so we just steal the ref the tree was holding. | |
6762 | */ | |
c3e69d58 | 6763 | delayed_refs->num_heads--; |
d7df2c79 | 6764 | if (head->processing == 0) |
c3e69d58 | 6765 | delayed_refs->num_heads_ready--; |
d7df2c79 JB |
6766 | head->processing = 0; |
6767 | spin_unlock(&head->lock); | |
1887be66 CM |
6768 | spin_unlock(&delayed_refs->lock); |
6769 | ||
f0486c68 YZ |
6770 | BUG_ON(head->extent_op); |
6771 | if (head->must_insert_reserved) | |
6772 | ret = 1; | |
6773 | ||
6774 | mutex_unlock(&head->mutex); | |
1887be66 | 6775 | btrfs_put_delayed_ref(&head->node); |
f0486c68 | 6776 | return ret; |
1887be66 | 6777 | out: |
d7df2c79 | 6778 | spin_unlock(&head->lock); |
cf93da7b CM |
6779 | |
6780 | out_delayed_unlock: | |
1887be66 CM |
6781 | spin_unlock(&delayed_refs->lock); |
6782 | return 0; | |
6783 | } | |
6784 | ||
f0486c68 YZ |
6785 | void btrfs_free_tree_block(struct btrfs_trans_handle *trans, |
6786 | struct btrfs_root *root, | |
6787 | struct extent_buffer *buf, | |
5581a51a | 6788 | u64 parent, int last_ref) |
f0486c68 | 6789 | { |
b150a4f1 | 6790 | int pin = 1; |
f0486c68 YZ |
6791 | int ret; |
6792 | ||
6793 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { | |
66d7e7f0 AJ |
6794 | ret = btrfs_add_delayed_tree_ref(root->fs_info, trans, |
6795 | buf->start, buf->len, | |
6796 | parent, root->root_key.objectid, | |
6797 | btrfs_header_level(buf), | |
b06c4bf5 | 6798 | BTRFS_DROP_DELAYED_REF, NULL); |
79787eaa | 6799 | BUG_ON(ret); /* -ENOMEM */ |
f0486c68 YZ |
6800 | } |
6801 | ||
6802 | if (!last_ref) | |
6803 | return; | |
6804 | ||
f0486c68 | 6805 | if (btrfs_header_generation(buf) == trans->transid) { |
6219872d FM |
6806 | struct btrfs_block_group_cache *cache; |
6807 | ||
f0486c68 YZ |
6808 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { |
6809 | ret = check_ref_cleanup(trans, root, buf->start); | |
6810 | if (!ret) | |
37be25bc | 6811 | goto out; |
f0486c68 YZ |
6812 | } |
6813 | ||
6219872d FM |
6814 | cache = btrfs_lookup_block_group(root->fs_info, buf->start); |
6815 | ||
f0486c68 YZ |
6816 | if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) { |
6817 | pin_down_extent(root, cache, buf->start, buf->len, 1); | |
6219872d | 6818 | btrfs_put_block_group(cache); |
37be25bc | 6819 | goto out; |
f0486c68 YZ |
6820 | } |
6821 | ||
6822 | WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)); | |
6823 | ||
6824 | btrfs_add_free_space(cache, buf->start, buf->len); | |
e570fd27 | 6825 | btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE, 0); |
6219872d | 6826 | btrfs_put_block_group(cache); |
0be5dc67 | 6827 | trace_btrfs_reserved_extent_free(root, buf->start, buf->len); |
b150a4f1 | 6828 | pin = 0; |
f0486c68 YZ |
6829 | } |
6830 | out: | |
b150a4f1 JB |
6831 | if (pin) |
6832 | add_pinned_bytes(root->fs_info, buf->len, | |
6833 | btrfs_header_level(buf), | |
6834 | root->root_key.objectid); | |
6835 | ||
a826d6dc JB |
6836 | /* |
6837 | * Deleting the buffer, clear the corrupt flag since it doesn't matter | |
6838 | * anymore. | |
6839 | */ | |
6840 | clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags); | |
f0486c68 YZ |
6841 | } |
6842 | ||
79787eaa | 6843 | /* Can return -ENOMEM */ |
66d7e7f0 AJ |
6844 | int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
6845 | u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid, | |
b06c4bf5 | 6846 | u64 owner, u64 offset) |
925baedd CM |
6847 | { |
6848 | int ret; | |
66d7e7f0 | 6849 | struct btrfs_fs_info *fs_info = root->fs_info; |
925baedd | 6850 | |
fccb84c9 | 6851 | if (btrfs_test_is_dummy_root(root)) |
faa2dbf0 | 6852 | return 0; |
fccb84c9 | 6853 | |
b150a4f1 JB |
6854 | add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid); |
6855 | ||
56bec294 CM |
6856 | /* |
6857 | * tree log blocks never actually go into the extent allocation | |
6858 | * tree, just update pinning info and exit early. | |
56bec294 | 6859 | */ |
5d4f98a2 YZ |
6860 | if (root_objectid == BTRFS_TREE_LOG_OBJECTID) { |
6861 | WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID); | |
b9473439 | 6862 | /* unlocks the pinned mutex */ |
11833d66 | 6863 | btrfs_pin_extent(root, bytenr, num_bytes, 1); |
56bec294 | 6864 | ret = 0; |
5d4f98a2 | 6865 | } else if (owner < BTRFS_FIRST_FREE_OBJECTID) { |
66d7e7f0 AJ |
6866 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr, |
6867 | num_bytes, | |
5d4f98a2 | 6868 | parent, root_objectid, (int)owner, |
b06c4bf5 | 6869 | BTRFS_DROP_DELAYED_REF, NULL); |
5d4f98a2 | 6870 | } else { |
66d7e7f0 AJ |
6871 | ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr, |
6872 | num_bytes, | |
6873 | parent, root_objectid, owner, | |
5846a3c2 QW |
6874 | offset, 0, |
6875 | BTRFS_DROP_DELAYED_REF, NULL); | |
56bec294 | 6876 | } |
925baedd CM |
6877 | return ret; |
6878 | } | |
6879 | ||
817d52f8 JB |
6880 | /* |
6881 | * when we wait for progress in the block group caching, its because | |
6882 | * our allocation attempt failed at least once. So, we must sleep | |
6883 | * and let some progress happen before we try again. | |
6884 | * | |
6885 | * This function will sleep at least once waiting for new free space to | |
6886 | * show up, and then it will check the block group free space numbers | |
6887 | * for our min num_bytes. Another option is to have it go ahead | |
6888 | * and look in the rbtree for a free extent of a given size, but this | |
6889 | * is a good start. | |
36cce922 JB |
6890 | * |
6891 | * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using | |
6892 | * any of the information in this block group. | |
817d52f8 | 6893 | */ |
36cce922 | 6894 | static noinline void |
817d52f8 JB |
6895 | wait_block_group_cache_progress(struct btrfs_block_group_cache *cache, |
6896 | u64 num_bytes) | |
6897 | { | |
11833d66 | 6898 | struct btrfs_caching_control *caching_ctl; |
817d52f8 | 6899 | |
11833d66 YZ |
6900 | caching_ctl = get_caching_control(cache); |
6901 | if (!caching_ctl) | |
36cce922 | 6902 | return; |
817d52f8 | 6903 | |
11833d66 | 6904 | wait_event(caching_ctl->wait, block_group_cache_done(cache) || |
34d52cb6 | 6905 | (cache->free_space_ctl->free_space >= num_bytes)); |
11833d66 YZ |
6906 | |
6907 | put_caching_control(caching_ctl); | |
11833d66 YZ |
6908 | } |
6909 | ||
6910 | static noinline int | |
6911 | wait_block_group_cache_done(struct btrfs_block_group_cache *cache) | |
6912 | { | |
6913 | struct btrfs_caching_control *caching_ctl; | |
36cce922 | 6914 | int ret = 0; |
11833d66 YZ |
6915 | |
6916 | caching_ctl = get_caching_control(cache); | |
6917 | if (!caching_ctl) | |
36cce922 | 6918 | return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0; |
11833d66 YZ |
6919 | |
6920 | wait_event(caching_ctl->wait, block_group_cache_done(cache)); | |
36cce922 JB |
6921 | if (cache->cached == BTRFS_CACHE_ERROR) |
6922 | ret = -EIO; | |
11833d66 | 6923 | put_caching_control(caching_ctl); |
36cce922 | 6924 | return ret; |
817d52f8 JB |
6925 | } |
6926 | ||
31e50229 | 6927 | int __get_raid_index(u64 flags) |
b742bb82 | 6928 | { |
7738a53a | 6929 | if (flags & BTRFS_BLOCK_GROUP_RAID10) |
e6ec716f | 6930 | return BTRFS_RAID_RAID10; |
7738a53a | 6931 | else if (flags & BTRFS_BLOCK_GROUP_RAID1) |
e6ec716f | 6932 | return BTRFS_RAID_RAID1; |
7738a53a | 6933 | else if (flags & BTRFS_BLOCK_GROUP_DUP) |
e6ec716f | 6934 | return BTRFS_RAID_DUP; |
7738a53a | 6935 | else if (flags & BTRFS_BLOCK_GROUP_RAID0) |
e6ec716f | 6936 | return BTRFS_RAID_RAID0; |
53b381b3 | 6937 | else if (flags & BTRFS_BLOCK_GROUP_RAID5) |
e942f883 | 6938 | return BTRFS_RAID_RAID5; |
53b381b3 | 6939 | else if (flags & BTRFS_BLOCK_GROUP_RAID6) |
e942f883 | 6940 | return BTRFS_RAID_RAID6; |
7738a53a | 6941 | |
e942f883 | 6942 | return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */ |
b742bb82 YZ |
6943 | } |
6944 | ||
6ab0a202 | 6945 | int get_block_group_index(struct btrfs_block_group_cache *cache) |
7738a53a | 6946 | { |
31e50229 | 6947 | return __get_raid_index(cache->flags); |
7738a53a ID |
6948 | } |
6949 | ||
6ab0a202 JM |
6950 | static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = { |
6951 | [BTRFS_RAID_RAID10] = "raid10", | |
6952 | [BTRFS_RAID_RAID1] = "raid1", | |
6953 | [BTRFS_RAID_DUP] = "dup", | |
6954 | [BTRFS_RAID_RAID0] = "raid0", | |
6955 | [BTRFS_RAID_SINGLE] = "single", | |
6956 | [BTRFS_RAID_RAID5] = "raid5", | |
6957 | [BTRFS_RAID_RAID6] = "raid6", | |
6958 | }; | |
6959 | ||
1b8e5df6 | 6960 | static const char *get_raid_name(enum btrfs_raid_types type) |
6ab0a202 JM |
6961 | { |
6962 | if (type >= BTRFS_NR_RAID_TYPES) | |
6963 | return NULL; | |
6964 | ||
6965 | return btrfs_raid_type_names[type]; | |
6966 | } | |
6967 | ||
817d52f8 | 6968 | enum btrfs_loop_type { |
285ff5af JB |
6969 | LOOP_CACHING_NOWAIT = 0, |
6970 | LOOP_CACHING_WAIT = 1, | |
6971 | LOOP_ALLOC_CHUNK = 2, | |
6972 | LOOP_NO_EMPTY_SIZE = 3, | |
817d52f8 JB |
6973 | }; |
6974 | ||
e570fd27 MX |
6975 | static inline void |
6976 | btrfs_lock_block_group(struct btrfs_block_group_cache *cache, | |
6977 | int delalloc) | |
6978 | { | |
6979 | if (delalloc) | |
6980 | down_read(&cache->data_rwsem); | |
6981 | } | |
6982 | ||
6983 | static inline void | |
6984 | btrfs_grab_block_group(struct btrfs_block_group_cache *cache, | |
6985 | int delalloc) | |
6986 | { | |
6987 | btrfs_get_block_group(cache); | |
6988 | if (delalloc) | |
6989 | down_read(&cache->data_rwsem); | |
6990 | } | |
6991 | ||
6992 | static struct btrfs_block_group_cache * | |
6993 | btrfs_lock_cluster(struct btrfs_block_group_cache *block_group, | |
6994 | struct btrfs_free_cluster *cluster, | |
6995 | int delalloc) | |
6996 | { | |
6997 | struct btrfs_block_group_cache *used_bg; | |
6998 | bool locked = false; | |
6999 | again: | |
7000 | spin_lock(&cluster->refill_lock); | |
7001 | if (locked) { | |
7002 | if (used_bg == cluster->block_group) | |
7003 | return used_bg; | |
7004 | ||
7005 | up_read(&used_bg->data_rwsem); | |
7006 | btrfs_put_block_group(used_bg); | |
7007 | } | |
7008 | ||
7009 | used_bg = cluster->block_group; | |
7010 | if (!used_bg) | |
7011 | return NULL; | |
7012 | ||
7013 | if (used_bg == block_group) | |
7014 | return used_bg; | |
7015 | ||
7016 | btrfs_get_block_group(used_bg); | |
7017 | ||
7018 | if (!delalloc) | |
7019 | return used_bg; | |
7020 | ||
7021 | if (down_read_trylock(&used_bg->data_rwsem)) | |
7022 | return used_bg; | |
7023 | ||
7024 | spin_unlock(&cluster->refill_lock); | |
7025 | down_read(&used_bg->data_rwsem); | |
7026 | locked = true; | |
7027 | goto again; | |
7028 | } | |
7029 | ||
7030 | static inline void | |
7031 | btrfs_release_block_group(struct btrfs_block_group_cache *cache, | |
7032 | int delalloc) | |
7033 | { | |
7034 | if (delalloc) | |
7035 | up_read(&cache->data_rwsem); | |
7036 | btrfs_put_block_group(cache); | |
7037 | } | |
7038 | ||
fec577fb CM |
7039 | /* |
7040 | * walks the btree of allocated extents and find a hole of a given size. | |
7041 | * The key ins is changed to record the hole: | |
a4820398 | 7042 | * ins->objectid == start position |
62e2749e | 7043 | * ins->flags = BTRFS_EXTENT_ITEM_KEY |
a4820398 | 7044 | * ins->offset == the size of the hole. |
fec577fb | 7045 | * Any available blocks before search_start are skipped. |
a4820398 MX |
7046 | * |
7047 | * If there is no suitable free space, we will record the max size of | |
7048 | * the free space extent currently. | |
fec577fb | 7049 | */ |
00361589 | 7050 | static noinline int find_free_extent(struct btrfs_root *orig_root, |
98ed5174 | 7051 | u64 num_bytes, u64 empty_size, |
98ed5174 | 7052 | u64 hint_byte, struct btrfs_key *ins, |
e570fd27 | 7053 | u64 flags, int delalloc) |
fec577fb | 7054 | { |
80eb234a | 7055 | int ret = 0; |
d397712b | 7056 | struct btrfs_root *root = orig_root->fs_info->extent_root; |
fa9c0d79 | 7057 | struct btrfs_free_cluster *last_ptr = NULL; |
80eb234a | 7058 | struct btrfs_block_group_cache *block_group = NULL; |
81c9ad23 | 7059 | u64 search_start = 0; |
a4820398 | 7060 | u64 max_extent_size = 0; |
c759c4e1 | 7061 | u64 empty_cluster = 0; |
80eb234a | 7062 | struct btrfs_space_info *space_info; |
fa9c0d79 | 7063 | int loop = 0; |
b6919a58 DS |
7064 | int index = __get_raid_index(flags); |
7065 | int alloc_type = (flags & BTRFS_BLOCK_GROUP_DATA) ? | |
fb25e914 | 7066 | RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC; |
0a24325e | 7067 | bool failed_cluster_refill = false; |
1cdda9b8 | 7068 | bool failed_alloc = false; |
67377734 | 7069 | bool use_cluster = true; |
60d2adbb | 7070 | bool have_caching_bg = false; |
13a0db5a | 7071 | bool orig_have_caching_bg = false; |
a5e681d9 | 7072 | bool full_search = false; |
fec577fb | 7073 | |
db94535d | 7074 | WARN_ON(num_bytes < root->sectorsize); |
962a298f | 7075 | ins->type = BTRFS_EXTENT_ITEM_KEY; |
80eb234a JB |
7076 | ins->objectid = 0; |
7077 | ins->offset = 0; | |
b1a4d965 | 7078 | |
b6919a58 | 7079 | trace_find_free_extent(orig_root, num_bytes, empty_size, flags); |
3f7de037 | 7080 | |
b6919a58 | 7081 | space_info = __find_space_info(root->fs_info, flags); |
1b1d1f66 | 7082 | if (!space_info) { |
b6919a58 | 7083 | btrfs_err(root->fs_info, "No space info for %llu", flags); |
1b1d1f66 JB |
7084 | return -ENOSPC; |
7085 | } | |
2552d17e | 7086 | |
67377734 | 7087 | /* |
4f4db217 JB |
7088 | * If our free space is heavily fragmented we may not be able to make |
7089 | * big contiguous allocations, so instead of doing the expensive search | |
7090 | * for free space, simply return ENOSPC with our max_extent_size so we | |
7091 | * can go ahead and search for a more manageable chunk. | |
7092 | * | |
7093 | * If our max_extent_size is large enough for our allocation simply | |
7094 | * disable clustering since we will likely not be able to find enough | |
7095 | * space to create a cluster and induce latency trying. | |
67377734 | 7096 | */ |
4f4db217 JB |
7097 | if (unlikely(space_info->max_extent_size)) { |
7098 | spin_lock(&space_info->lock); | |
7099 | if (space_info->max_extent_size && | |
7100 | num_bytes > space_info->max_extent_size) { | |
7101 | ins->offset = space_info->max_extent_size; | |
7102 | spin_unlock(&space_info->lock); | |
7103 | return -ENOSPC; | |
7104 | } else if (space_info->max_extent_size) { | |
7105 | use_cluster = false; | |
7106 | } | |
7107 | spin_unlock(&space_info->lock); | |
fa9c0d79 | 7108 | } |
0f9dd46c | 7109 | |
c759c4e1 | 7110 | last_ptr = fetch_cluster_info(orig_root, space_info, &empty_cluster); |
239b14b3 | 7111 | if (last_ptr) { |
fa9c0d79 CM |
7112 | spin_lock(&last_ptr->lock); |
7113 | if (last_ptr->block_group) | |
7114 | hint_byte = last_ptr->window_start; | |
c759c4e1 JB |
7115 | if (last_ptr->fragmented) { |
7116 | /* | |
7117 | * We still set window_start so we can keep track of the | |
7118 | * last place we found an allocation to try and save | |
7119 | * some time. | |
7120 | */ | |
7121 | hint_byte = last_ptr->window_start; | |
7122 | use_cluster = false; | |
7123 | } | |
fa9c0d79 | 7124 | spin_unlock(&last_ptr->lock); |
239b14b3 | 7125 | } |
fa9c0d79 | 7126 | |
a061fc8d | 7127 | search_start = max(search_start, first_logical_byte(root, 0)); |
239b14b3 | 7128 | search_start = max(search_start, hint_byte); |
2552d17e | 7129 | if (search_start == hint_byte) { |
2552d17e JB |
7130 | block_group = btrfs_lookup_block_group(root->fs_info, |
7131 | search_start); | |
817d52f8 JB |
7132 | /* |
7133 | * we don't want to use the block group if it doesn't match our | |
7134 | * allocation bits, or if its not cached. | |
ccf0e725 JB |
7135 | * |
7136 | * However if we are re-searching with an ideal block group | |
7137 | * picked out then we don't care that the block group is cached. | |
817d52f8 | 7138 | */ |
b6919a58 | 7139 | if (block_group && block_group_bits(block_group, flags) && |
285ff5af | 7140 | block_group->cached != BTRFS_CACHE_NO) { |
2552d17e | 7141 | down_read(&space_info->groups_sem); |
44fb5511 CM |
7142 | if (list_empty(&block_group->list) || |
7143 | block_group->ro) { | |
7144 | /* | |
7145 | * someone is removing this block group, | |
7146 | * we can't jump into the have_block_group | |
7147 | * target because our list pointers are not | |
7148 | * valid | |
7149 | */ | |
7150 | btrfs_put_block_group(block_group); | |
7151 | up_read(&space_info->groups_sem); | |
ccf0e725 | 7152 | } else { |
b742bb82 | 7153 | index = get_block_group_index(block_group); |
e570fd27 | 7154 | btrfs_lock_block_group(block_group, delalloc); |
44fb5511 | 7155 | goto have_block_group; |
ccf0e725 | 7156 | } |
2552d17e | 7157 | } else if (block_group) { |
fa9c0d79 | 7158 | btrfs_put_block_group(block_group); |
2552d17e | 7159 | } |
42e70e7a | 7160 | } |
2552d17e | 7161 | search: |
60d2adbb | 7162 | have_caching_bg = false; |
a5e681d9 JB |
7163 | if (index == 0 || index == __get_raid_index(flags)) |
7164 | full_search = true; | |
80eb234a | 7165 | down_read(&space_info->groups_sem); |
b742bb82 YZ |
7166 | list_for_each_entry(block_group, &space_info->block_groups[index], |
7167 | list) { | |
6226cb0a | 7168 | u64 offset; |
817d52f8 | 7169 | int cached; |
8a1413a2 | 7170 | |
e570fd27 | 7171 | btrfs_grab_block_group(block_group, delalloc); |
2552d17e | 7172 | search_start = block_group->key.objectid; |
42e70e7a | 7173 | |
83a50de9 CM |
7174 | /* |
7175 | * this can happen if we end up cycling through all the | |
7176 | * raid types, but we want to make sure we only allocate | |
7177 | * for the proper type. | |
7178 | */ | |
b6919a58 | 7179 | if (!block_group_bits(block_group, flags)) { |
83a50de9 CM |
7180 | u64 extra = BTRFS_BLOCK_GROUP_DUP | |
7181 | BTRFS_BLOCK_GROUP_RAID1 | | |
53b381b3 DW |
7182 | BTRFS_BLOCK_GROUP_RAID5 | |
7183 | BTRFS_BLOCK_GROUP_RAID6 | | |
83a50de9 CM |
7184 | BTRFS_BLOCK_GROUP_RAID10; |
7185 | ||
7186 | /* | |
7187 | * if they asked for extra copies and this block group | |
7188 | * doesn't provide them, bail. This does allow us to | |
7189 | * fill raid0 from raid1. | |
7190 | */ | |
b6919a58 | 7191 | if ((flags & extra) && !(block_group->flags & extra)) |
83a50de9 CM |
7192 | goto loop; |
7193 | } | |
7194 | ||
2552d17e | 7195 | have_block_group: |
291c7d2f JB |
7196 | cached = block_group_cache_done(block_group); |
7197 | if (unlikely(!cached)) { | |
a5e681d9 | 7198 | have_caching_bg = true; |
f6373bf3 | 7199 | ret = cache_block_group(block_group, 0); |
1d4284bd CM |
7200 | BUG_ON(ret < 0); |
7201 | ret = 0; | |
817d52f8 JB |
7202 | } |
7203 | ||
36cce922 JB |
7204 | if (unlikely(block_group->cached == BTRFS_CACHE_ERROR)) |
7205 | goto loop; | |
ea6a478e | 7206 | if (unlikely(block_group->ro)) |
2552d17e | 7207 | goto loop; |
0f9dd46c | 7208 | |
0a24325e | 7209 | /* |
062c05c4 AO |
7210 | * Ok we want to try and use the cluster allocator, so |
7211 | * lets look there | |
0a24325e | 7212 | */ |
c759c4e1 | 7213 | if (last_ptr && use_cluster) { |
215a63d1 | 7214 | struct btrfs_block_group_cache *used_block_group; |
8de972b4 | 7215 | unsigned long aligned_cluster; |
fa9c0d79 CM |
7216 | /* |
7217 | * the refill lock keeps out other | |
7218 | * people trying to start a new cluster | |
7219 | */ | |
e570fd27 MX |
7220 | used_block_group = btrfs_lock_cluster(block_group, |
7221 | last_ptr, | |
7222 | delalloc); | |
7223 | if (!used_block_group) | |
44fb5511 | 7224 | goto refill_cluster; |
274bd4fb | 7225 | |
e570fd27 MX |
7226 | if (used_block_group != block_group && |
7227 | (used_block_group->ro || | |
7228 | !block_group_bits(used_block_group, flags))) | |
7229 | goto release_cluster; | |
44fb5511 | 7230 | |
274bd4fb | 7231 | offset = btrfs_alloc_from_cluster(used_block_group, |
a4820398 MX |
7232 | last_ptr, |
7233 | num_bytes, | |
7234 | used_block_group->key.objectid, | |
7235 | &max_extent_size); | |
fa9c0d79 CM |
7236 | if (offset) { |
7237 | /* we have a block, we're done */ | |
7238 | spin_unlock(&last_ptr->refill_lock); | |
3f7de037 | 7239 | trace_btrfs_reserve_extent_cluster(root, |
89d4346a MX |
7240 | used_block_group, |
7241 | search_start, num_bytes); | |
215a63d1 | 7242 | if (used_block_group != block_group) { |
e570fd27 MX |
7243 | btrfs_release_block_group(block_group, |
7244 | delalloc); | |
215a63d1 MX |
7245 | block_group = used_block_group; |
7246 | } | |
fa9c0d79 CM |
7247 | goto checks; |
7248 | } | |
7249 | ||
274bd4fb | 7250 | WARN_ON(last_ptr->block_group != used_block_group); |
e570fd27 | 7251 | release_cluster: |
062c05c4 AO |
7252 | /* If we are on LOOP_NO_EMPTY_SIZE, we can't |
7253 | * set up a new clusters, so lets just skip it | |
7254 | * and let the allocator find whatever block | |
7255 | * it can find. If we reach this point, we | |
7256 | * will have tried the cluster allocator | |
7257 | * plenty of times and not have found | |
7258 | * anything, so we are likely way too | |
7259 | * fragmented for the clustering stuff to find | |
a5f6f719 AO |
7260 | * anything. |
7261 | * | |
7262 | * However, if the cluster is taken from the | |
7263 | * current block group, release the cluster | |
7264 | * first, so that we stand a better chance of | |
7265 | * succeeding in the unclustered | |
7266 | * allocation. */ | |
7267 | if (loop >= LOOP_NO_EMPTY_SIZE && | |
e570fd27 | 7268 | used_block_group != block_group) { |
062c05c4 | 7269 | spin_unlock(&last_ptr->refill_lock); |
e570fd27 MX |
7270 | btrfs_release_block_group(used_block_group, |
7271 | delalloc); | |
062c05c4 AO |
7272 | goto unclustered_alloc; |
7273 | } | |
7274 | ||
fa9c0d79 CM |
7275 | /* |
7276 | * this cluster didn't work out, free it and | |
7277 | * start over | |
7278 | */ | |
7279 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
7280 | ||
e570fd27 MX |
7281 | if (used_block_group != block_group) |
7282 | btrfs_release_block_group(used_block_group, | |
7283 | delalloc); | |
7284 | refill_cluster: | |
a5f6f719 AO |
7285 | if (loop >= LOOP_NO_EMPTY_SIZE) { |
7286 | spin_unlock(&last_ptr->refill_lock); | |
7287 | goto unclustered_alloc; | |
7288 | } | |
7289 | ||
8de972b4 CM |
7290 | aligned_cluster = max_t(unsigned long, |
7291 | empty_cluster + empty_size, | |
7292 | block_group->full_stripe_len); | |
7293 | ||
fa9c0d79 | 7294 | /* allocate a cluster in this block group */ |
00361589 JB |
7295 | ret = btrfs_find_space_cluster(root, block_group, |
7296 | last_ptr, search_start, | |
7297 | num_bytes, | |
7298 | aligned_cluster); | |
fa9c0d79 CM |
7299 | if (ret == 0) { |
7300 | /* | |
7301 | * now pull our allocation out of this | |
7302 | * cluster | |
7303 | */ | |
7304 | offset = btrfs_alloc_from_cluster(block_group, | |
a4820398 MX |
7305 | last_ptr, |
7306 | num_bytes, | |
7307 | search_start, | |
7308 | &max_extent_size); | |
fa9c0d79 CM |
7309 | if (offset) { |
7310 | /* we found one, proceed */ | |
7311 | spin_unlock(&last_ptr->refill_lock); | |
3f7de037 JB |
7312 | trace_btrfs_reserve_extent_cluster(root, |
7313 | block_group, search_start, | |
7314 | num_bytes); | |
fa9c0d79 CM |
7315 | goto checks; |
7316 | } | |
0a24325e JB |
7317 | } else if (!cached && loop > LOOP_CACHING_NOWAIT |
7318 | && !failed_cluster_refill) { | |
817d52f8 JB |
7319 | spin_unlock(&last_ptr->refill_lock); |
7320 | ||
0a24325e | 7321 | failed_cluster_refill = true; |
817d52f8 JB |
7322 | wait_block_group_cache_progress(block_group, |
7323 | num_bytes + empty_cluster + empty_size); | |
7324 | goto have_block_group; | |
fa9c0d79 | 7325 | } |
817d52f8 | 7326 | |
fa9c0d79 CM |
7327 | /* |
7328 | * at this point we either didn't find a cluster | |
7329 | * or we weren't able to allocate a block from our | |
7330 | * cluster. Free the cluster we've been trying | |
7331 | * to use, and go to the next block group | |
7332 | */ | |
0a24325e | 7333 | btrfs_return_cluster_to_free_space(NULL, last_ptr); |
fa9c0d79 | 7334 | spin_unlock(&last_ptr->refill_lock); |
0a24325e | 7335 | goto loop; |
fa9c0d79 CM |
7336 | } |
7337 | ||
062c05c4 | 7338 | unclustered_alloc: |
c759c4e1 JB |
7339 | /* |
7340 | * We are doing an unclustered alloc, set the fragmented flag so | |
7341 | * we don't bother trying to setup a cluster again until we get | |
7342 | * more space. | |
7343 | */ | |
7344 | if (unlikely(last_ptr)) { | |
7345 | spin_lock(&last_ptr->lock); | |
7346 | last_ptr->fragmented = 1; | |
7347 | spin_unlock(&last_ptr->lock); | |
7348 | } | |
a5f6f719 AO |
7349 | spin_lock(&block_group->free_space_ctl->tree_lock); |
7350 | if (cached && | |
7351 | block_group->free_space_ctl->free_space < | |
7352 | num_bytes + empty_cluster + empty_size) { | |
a4820398 MX |
7353 | if (block_group->free_space_ctl->free_space > |
7354 | max_extent_size) | |
7355 | max_extent_size = | |
7356 | block_group->free_space_ctl->free_space; | |
a5f6f719 AO |
7357 | spin_unlock(&block_group->free_space_ctl->tree_lock); |
7358 | goto loop; | |
7359 | } | |
7360 | spin_unlock(&block_group->free_space_ctl->tree_lock); | |
7361 | ||
6226cb0a | 7362 | offset = btrfs_find_space_for_alloc(block_group, search_start, |
a4820398 MX |
7363 | num_bytes, empty_size, |
7364 | &max_extent_size); | |
1cdda9b8 JB |
7365 | /* |
7366 | * If we didn't find a chunk, and we haven't failed on this | |
7367 | * block group before, and this block group is in the middle of | |
7368 | * caching and we are ok with waiting, then go ahead and wait | |
7369 | * for progress to be made, and set failed_alloc to true. | |
7370 | * | |
7371 | * If failed_alloc is true then we've already waited on this | |
7372 | * block group once and should move on to the next block group. | |
7373 | */ | |
7374 | if (!offset && !failed_alloc && !cached && | |
7375 | loop > LOOP_CACHING_NOWAIT) { | |
817d52f8 | 7376 | wait_block_group_cache_progress(block_group, |
1cdda9b8 JB |
7377 | num_bytes + empty_size); |
7378 | failed_alloc = true; | |
817d52f8 | 7379 | goto have_block_group; |
1cdda9b8 JB |
7380 | } else if (!offset) { |
7381 | goto loop; | |
817d52f8 | 7382 | } |
fa9c0d79 | 7383 | checks: |
4e54b17a | 7384 | search_start = ALIGN(offset, root->stripesize); |
25179201 | 7385 | |
2552d17e JB |
7386 | /* move on to the next group */ |
7387 | if (search_start + num_bytes > | |
215a63d1 MX |
7388 | block_group->key.objectid + block_group->key.offset) { |
7389 | btrfs_add_free_space(block_group, offset, num_bytes); | |
2552d17e | 7390 | goto loop; |
6226cb0a | 7391 | } |
f5a31e16 | 7392 | |
f0486c68 | 7393 | if (offset < search_start) |
215a63d1 | 7394 | btrfs_add_free_space(block_group, offset, |
f0486c68 YZ |
7395 | search_start - offset); |
7396 | BUG_ON(offset > search_start); | |
2552d17e | 7397 | |
215a63d1 | 7398 | ret = btrfs_update_reserved_bytes(block_group, num_bytes, |
e570fd27 | 7399 | alloc_type, delalloc); |
f0486c68 | 7400 | if (ret == -EAGAIN) { |
215a63d1 | 7401 | btrfs_add_free_space(block_group, offset, num_bytes); |
2552d17e | 7402 | goto loop; |
0f9dd46c | 7403 | } |
0b86a832 | 7404 | |
f0486c68 | 7405 | /* we are all good, lets return */ |
2552d17e JB |
7406 | ins->objectid = search_start; |
7407 | ins->offset = num_bytes; | |
d2fb3437 | 7408 | |
3f7de037 JB |
7409 | trace_btrfs_reserve_extent(orig_root, block_group, |
7410 | search_start, num_bytes); | |
e570fd27 | 7411 | btrfs_release_block_group(block_group, delalloc); |
2552d17e JB |
7412 | break; |
7413 | loop: | |
0a24325e | 7414 | failed_cluster_refill = false; |
1cdda9b8 | 7415 | failed_alloc = false; |
b742bb82 | 7416 | BUG_ON(index != get_block_group_index(block_group)); |
e570fd27 | 7417 | btrfs_release_block_group(block_group, delalloc); |
2552d17e JB |
7418 | } |
7419 | up_read(&space_info->groups_sem); | |
7420 | ||
13a0db5a | 7421 | if ((loop == LOOP_CACHING_NOWAIT) && have_caching_bg |
7422 | && !orig_have_caching_bg) | |
7423 | orig_have_caching_bg = true; | |
7424 | ||
60d2adbb MX |
7425 | if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg) |
7426 | goto search; | |
7427 | ||
b742bb82 YZ |
7428 | if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES) |
7429 | goto search; | |
7430 | ||
285ff5af | 7431 | /* |
ccf0e725 JB |
7432 | * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking |
7433 | * caching kthreads as we move along | |
817d52f8 JB |
7434 | * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching |
7435 | * LOOP_ALLOC_CHUNK, force a chunk allocation and try again | |
7436 | * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try | |
7437 | * again | |
fa9c0d79 | 7438 | */ |
723bda20 | 7439 | if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) { |
b742bb82 | 7440 | index = 0; |
a5e681d9 JB |
7441 | if (loop == LOOP_CACHING_NOWAIT) { |
7442 | /* | |
7443 | * We want to skip the LOOP_CACHING_WAIT step if we | |
7444 | * don't have any unached bgs and we've alrelady done a | |
7445 | * full search through. | |
7446 | */ | |
13a0db5a | 7447 | if (orig_have_caching_bg || !full_search) |
a5e681d9 JB |
7448 | loop = LOOP_CACHING_WAIT; |
7449 | else | |
7450 | loop = LOOP_ALLOC_CHUNK; | |
7451 | } else { | |
7452 | loop++; | |
7453 | } | |
7454 | ||
817d52f8 | 7455 | if (loop == LOOP_ALLOC_CHUNK) { |
00361589 | 7456 | struct btrfs_trans_handle *trans; |
f017f15f WS |
7457 | int exist = 0; |
7458 | ||
7459 | trans = current->journal_info; | |
7460 | if (trans) | |
7461 | exist = 1; | |
7462 | else | |
7463 | trans = btrfs_join_transaction(root); | |
00361589 | 7464 | |
00361589 JB |
7465 | if (IS_ERR(trans)) { |
7466 | ret = PTR_ERR(trans); | |
7467 | goto out; | |
7468 | } | |
7469 | ||
b6919a58 | 7470 | ret = do_chunk_alloc(trans, root, flags, |
ea658bad | 7471 | CHUNK_ALLOC_FORCE); |
a5e681d9 JB |
7472 | |
7473 | /* | |
7474 | * If we can't allocate a new chunk we've already looped | |
7475 | * through at least once, move on to the NO_EMPTY_SIZE | |
7476 | * case. | |
7477 | */ | |
7478 | if (ret == -ENOSPC) | |
7479 | loop = LOOP_NO_EMPTY_SIZE; | |
7480 | ||
ea658bad JB |
7481 | /* |
7482 | * Do not bail out on ENOSPC since we | |
7483 | * can do more things. | |
7484 | */ | |
00361589 | 7485 | if (ret < 0 && ret != -ENOSPC) |
ea658bad JB |
7486 | btrfs_abort_transaction(trans, |
7487 | root, ret); | |
00361589 JB |
7488 | else |
7489 | ret = 0; | |
f017f15f WS |
7490 | if (!exist) |
7491 | btrfs_end_transaction(trans, root); | |
00361589 | 7492 | if (ret) |
ea658bad | 7493 | goto out; |
2552d17e JB |
7494 | } |
7495 | ||
723bda20 | 7496 | if (loop == LOOP_NO_EMPTY_SIZE) { |
a5e681d9 JB |
7497 | /* |
7498 | * Don't loop again if we already have no empty_size and | |
7499 | * no empty_cluster. | |
7500 | */ | |
7501 | if (empty_size == 0 && | |
7502 | empty_cluster == 0) { | |
7503 | ret = -ENOSPC; | |
7504 | goto out; | |
7505 | } | |
723bda20 JB |
7506 | empty_size = 0; |
7507 | empty_cluster = 0; | |
fa9c0d79 | 7508 | } |
723bda20 JB |
7509 | |
7510 | goto search; | |
2552d17e JB |
7511 | } else if (!ins->objectid) { |
7512 | ret = -ENOSPC; | |
d82a6f1d | 7513 | } else if (ins->objectid) { |
c759c4e1 JB |
7514 | if (!use_cluster && last_ptr) { |
7515 | spin_lock(&last_ptr->lock); | |
7516 | last_ptr->window_start = ins->objectid; | |
7517 | spin_unlock(&last_ptr->lock); | |
7518 | } | |
80eb234a | 7519 | ret = 0; |
be744175 | 7520 | } |
79787eaa | 7521 | out: |
4f4db217 JB |
7522 | if (ret == -ENOSPC) { |
7523 | spin_lock(&space_info->lock); | |
7524 | space_info->max_extent_size = max_extent_size; | |
7525 | spin_unlock(&space_info->lock); | |
a4820398 | 7526 | ins->offset = max_extent_size; |
4f4db217 | 7527 | } |
0f70abe2 | 7528 | return ret; |
fec577fb | 7529 | } |
ec44a35c | 7530 | |
9ed74f2d JB |
7531 | static void dump_space_info(struct btrfs_space_info *info, u64 bytes, |
7532 | int dump_block_groups) | |
0f9dd46c JB |
7533 | { |
7534 | struct btrfs_block_group_cache *cache; | |
b742bb82 | 7535 | int index = 0; |
0f9dd46c | 7536 | |
9ed74f2d | 7537 | spin_lock(&info->lock); |
efe120a0 | 7538 | printk(KERN_INFO "BTRFS: space_info %llu has %llu free, is %sfull\n", |
c1c9ff7c GU |
7539 | info->flags, |
7540 | info->total_bytes - info->bytes_used - info->bytes_pinned - | |
7541 | info->bytes_reserved - info->bytes_readonly, | |
d397712b | 7542 | (info->full) ? "" : "not "); |
efe120a0 | 7543 | printk(KERN_INFO "BTRFS: space_info total=%llu, used=%llu, pinned=%llu, " |
8929ecfa | 7544 | "reserved=%llu, may_use=%llu, readonly=%llu\n", |
c1c9ff7c GU |
7545 | info->total_bytes, info->bytes_used, info->bytes_pinned, |
7546 | info->bytes_reserved, info->bytes_may_use, | |
7547 | info->bytes_readonly); | |
9ed74f2d JB |
7548 | spin_unlock(&info->lock); |
7549 | ||
7550 | if (!dump_block_groups) | |
7551 | return; | |
0f9dd46c | 7552 | |
80eb234a | 7553 | down_read(&info->groups_sem); |
b742bb82 YZ |
7554 | again: |
7555 | list_for_each_entry(cache, &info->block_groups[index], list) { | |
0f9dd46c | 7556 | spin_lock(&cache->lock); |
efe120a0 FH |
7557 | printk(KERN_INFO "BTRFS: " |
7558 | "block group %llu has %llu bytes, " | |
7559 | "%llu used %llu pinned %llu reserved %s\n", | |
c1c9ff7c GU |
7560 | cache->key.objectid, cache->key.offset, |
7561 | btrfs_block_group_used(&cache->item), cache->pinned, | |
7562 | cache->reserved, cache->ro ? "[readonly]" : ""); | |
0f9dd46c JB |
7563 | btrfs_dump_free_space(cache, bytes); |
7564 | spin_unlock(&cache->lock); | |
7565 | } | |
b742bb82 YZ |
7566 | if (++index < BTRFS_NR_RAID_TYPES) |
7567 | goto again; | |
80eb234a | 7568 | up_read(&info->groups_sem); |
0f9dd46c | 7569 | } |
e8569813 | 7570 | |
00361589 | 7571 | int btrfs_reserve_extent(struct btrfs_root *root, |
11833d66 YZ |
7572 | u64 num_bytes, u64 min_alloc_size, |
7573 | u64 empty_size, u64 hint_byte, | |
e570fd27 | 7574 | struct btrfs_key *ins, int is_data, int delalloc) |
fec577fb | 7575 | { |
36af4e07 | 7576 | bool final_tried = num_bytes == min_alloc_size; |
b6919a58 | 7577 | u64 flags; |
fec577fb | 7578 | int ret; |
925baedd | 7579 | |
b6919a58 | 7580 | flags = btrfs_get_alloc_profile(root, is_data); |
98d20f67 | 7581 | again: |
db94535d | 7582 | WARN_ON(num_bytes < root->sectorsize); |
00361589 | 7583 | ret = find_free_extent(root, num_bytes, empty_size, hint_byte, ins, |
e570fd27 | 7584 | flags, delalloc); |
3b951516 | 7585 | |
9e622d6b | 7586 | if (ret == -ENOSPC) { |
a4820398 MX |
7587 | if (!final_tried && ins->offset) { |
7588 | num_bytes = min(num_bytes >> 1, ins->offset); | |
24542bf7 | 7589 | num_bytes = round_down(num_bytes, root->sectorsize); |
9e622d6b | 7590 | num_bytes = max(num_bytes, min_alloc_size); |
9e622d6b MX |
7591 | if (num_bytes == min_alloc_size) |
7592 | final_tried = true; | |
7593 | goto again; | |
7594 | } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) { | |
7595 | struct btrfs_space_info *sinfo; | |
7596 | ||
b6919a58 | 7597 | sinfo = __find_space_info(root->fs_info, flags); |
c2cf52eb | 7598 | btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu", |
c1c9ff7c | 7599 | flags, num_bytes); |
53804280 JM |
7600 | if (sinfo) |
7601 | dump_space_info(sinfo, num_bytes, 1); | |
9e622d6b | 7602 | } |
925baedd | 7603 | } |
0f9dd46c JB |
7604 | |
7605 | return ret; | |
e6dcd2dc CM |
7606 | } |
7607 | ||
e688b725 | 7608 | static int __btrfs_free_reserved_extent(struct btrfs_root *root, |
e570fd27 MX |
7609 | u64 start, u64 len, |
7610 | int pin, int delalloc) | |
65b51a00 | 7611 | { |
0f9dd46c | 7612 | struct btrfs_block_group_cache *cache; |
1f3c79a2 | 7613 | int ret = 0; |
0f9dd46c | 7614 | |
0f9dd46c JB |
7615 | cache = btrfs_lookup_block_group(root->fs_info, start); |
7616 | if (!cache) { | |
c2cf52eb | 7617 | btrfs_err(root->fs_info, "Unable to find block group for %llu", |
c1c9ff7c | 7618 | start); |
0f9dd46c JB |
7619 | return -ENOSPC; |
7620 | } | |
1f3c79a2 | 7621 | |
e688b725 CM |
7622 | if (pin) |
7623 | pin_down_extent(root, cache, start, len, 1); | |
7624 | else { | |
dcc82f47 FM |
7625 | if (btrfs_test_opt(root, DISCARD)) |
7626 | ret = btrfs_discard_extent(root, start, len, NULL); | |
e688b725 | 7627 | btrfs_add_free_space(cache, start, len); |
e570fd27 | 7628 | btrfs_update_reserved_bytes(cache, len, RESERVE_FREE, delalloc); |
e688b725 | 7629 | } |
31193213 | 7630 | |
fa9c0d79 | 7631 | btrfs_put_block_group(cache); |
817d52f8 | 7632 | |
1abe9b8a | 7633 | trace_btrfs_reserved_extent_free(root, start, len); |
7634 | ||
e6dcd2dc CM |
7635 | return ret; |
7636 | } | |
7637 | ||
e688b725 | 7638 | int btrfs_free_reserved_extent(struct btrfs_root *root, |
e570fd27 | 7639 | u64 start, u64 len, int delalloc) |
e688b725 | 7640 | { |
e570fd27 | 7641 | return __btrfs_free_reserved_extent(root, start, len, 0, delalloc); |
e688b725 CM |
7642 | } |
7643 | ||
7644 | int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root, | |
7645 | u64 start, u64 len) | |
7646 | { | |
e570fd27 | 7647 | return __btrfs_free_reserved_extent(root, start, len, 1, 0); |
e688b725 CM |
7648 | } |
7649 | ||
5d4f98a2 YZ |
7650 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, |
7651 | struct btrfs_root *root, | |
7652 | u64 parent, u64 root_objectid, | |
7653 | u64 flags, u64 owner, u64 offset, | |
7654 | struct btrfs_key *ins, int ref_mod) | |
e6dcd2dc CM |
7655 | { |
7656 | int ret; | |
5d4f98a2 | 7657 | struct btrfs_fs_info *fs_info = root->fs_info; |
e6dcd2dc | 7658 | struct btrfs_extent_item *extent_item; |
5d4f98a2 | 7659 | struct btrfs_extent_inline_ref *iref; |
e6dcd2dc | 7660 | struct btrfs_path *path; |
5d4f98a2 YZ |
7661 | struct extent_buffer *leaf; |
7662 | int type; | |
7663 | u32 size; | |
26b8003f | 7664 | |
5d4f98a2 YZ |
7665 | if (parent > 0) |
7666 | type = BTRFS_SHARED_DATA_REF_KEY; | |
7667 | else | |
7668 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
58176a96 | 7669 | |
5d4f98a2 | 7670 | size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type); |
7bb86316 CM |
7671 | |
7672 | path = btrfs_alloc_path(); | |
db5b493a TI |
7673 | if (!path) |
7674 | return -ENOMEM; | |
47e4bb98 | 7675 | |
b9473439 | 7676 | path->leave_spinning = 1; |
5d4f98a2 YZ |
7677 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, |
7678 | ins, size); | |
79787eaa JM |
7679 | if (ret) { |
7680 | btrfs_free_path(path); | |
7681 | return ret; | |
7682 | } | |
0f9dd46c | 7683 | |
5d4f98a2 YZ |
7684 | leaf = path->nodes[0]; |
7685 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
47e4bb98 | 7686 | struct btrfs_extent_item); |
5d4f98a2 YZ |
7687 | btrfs_set_extent_refs(leaf, extent_item, ref_mod); |
7688 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
7689 | btrfs_set_extent_flags(leaf, extent_item, | |
7690 | flags | BTRFS_EXTENT_FLAG_DATA); | |
7691 | ||
7692 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
7693 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
7694 | if (parent > 0) { | |
7695 | struct btrfs_shared_data_ref *ref; | |
7696 | ref = (struct btrfs_shared_data_ref *)(iref + 1); | |
7697 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
7698 | btrfs_set_shared_data_ref_count(leaf, ref, ref_mod); | |
7699 | } else { | |
7700 | struct btrfs_extent_data_ref *ref; | |
7701 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
7702 | btrfs_set_extent_data_ref_root(leaf, ref, root_objectid); | |
7703 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
7704 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
7705 | btrfs_set_extent_data_ref_count(leaf, ref, ref_mod); | |
7706 | } | |
47e4bb98 CM |
7707 | |
7708 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
7bb86316 | 7709 | btrfs_free_path(path); |
f510cfec | 7710 | |
1e144fb8 OS |
7711 | ret = remove_from_free_space_tree(trans, fs_info, ins->objectid, |
7712 | ins->offset); | |
7713 | if (ret) | |
7714 | return ret; | |
7715 | ||
ce93ec54 | 7716 | ret = update_block_group(trans, root, ins->objectid, ins->offset, 1); |
79787eaa | 7717 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb | 7718 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
c1c9ff7c | 7719 | ins->objectid, ins->offset); |
f5947066 CM |
7720 | BUG(); |
7721 | } | |
0be5dc67 | 7722 | trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset); |
e6dcd2dc CM |
7723 | return ret; |
7724 | } | |
7725 | ||
5d4f98a2 YZ |
7726 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, |
7727 | struct btrfs_root *root, | |
7728 | u64 parent, u64 root_objectid, | |
7729 | u64 flags, struct btrfs_disk_key *key, | |
b06c4bf5 | 7730 | int level, struct btrfs_key *ins) |
e6dcd2dc CM |
7731 | { |
7732 | int ret; | |
5d4f98a2 YZ |
7733 | struct btrfs_fs_info *fs_info = root->fs_info; |
7734 | struct btrfs_extent_item *extent_item; | |
7735 | struct btrfs_tree_block_info *block_info; | |
7736 | struct btrfs_extent_inline_ref *iref; | |
7737 | struct btrfs_path *path; | |
7738 | struct extent_buffer *leaf; | |
3173a18f | 7739 | u32 size = sizeof(*extent_item) + sizeof(*iref); |
fcebe456 | 7740 | u64 num_bytes = ins->offset; |
3173a18f JB |
7741 | bool skinny_metadata = btrfs_fs_incompat(root->fs_info, |
7742 | SKINNY_METADATA); | |
7743 | ||
7744 | if (!skinny_metadata) | |
7745 | size += sizeof(*block_info); | |
1c2308f8 | 7746 | |
5d4f98a2 | 7747 | path = btrfs_alloc_path(); |
857cc2fc JB |
7748 | if (!path) { |
7749 | btrfs_free_and_pin_reserved_extent(root, ins->objectid, | |
707e8a07 | 7750 | root->nodesize); |
d8926bb3 | 7751 | return -ENOMEM; |
857cc2fc | 7752 | } |
56bec294 | 7753 | |
5d4f98a2 YZ |
7754 | path->leave_spinning = 1; |
7755 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, | |
7756 | ins, size); | |
79787eaa | 7757 | if (ret) { |
dd825259 | 7758 | btrfs_free_path(path); |
857cc2fc | 7759 | btrfs_free_and_pin_reserved_extent(root, ins->objectid, |
707e8a07 | 7760 | root->nodesize); |
79787eaa JM |
7761 | return ret; |
7762 | } | |
5d4f98a2 YZ |
7763 | |
7764 | leaf = path->nodes[0]; | |
7765 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
7766 | struct btrfs_extent_item); | |
7767 | btrfs_set_extent_refs(leaf, extent_item, 1); | |
7768 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
7769 | btrfs_set_extent_flags(leaf, extent_item, | |
7770 | flags | BTRFS_EXTENT_FLAG_TREE_BLOCK); | |
5d4f98a2 | 7771 | |
3173a18f JB |
7772 | if (skinny_metadata) { |
7773 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
707e8a07 | 7774 | num_bytes = root->nodesize; |
3173a18f JB |
7775 | } else { |
7776 | block_info = (struct btrfs_tree_block_info *)(extent_item + 1); | |
7777 | btrfs_set_tree_block_key(leaf, block_info, key); | |
7778 | btrfs_set_tree_block_level(leaf, block_info, level); | |
7779 | iref = (struct btrfs_extent_inline_ref *)(block_info + 1); | |
7780 | } | |
5d4f98a2 | 7781 | |
5d4f98a2 YZ |
7782 | if (parent > 0) { |
7783 | BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); | |
7784 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
7785 | BTRFS_SHARED_BLOCK_REF_KEY); | |
7786 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
7787 | } else { | |
7788 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
7789 | BTRFS_TREE_BLOCK_REF_KEY); | |
7790 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
7791 | } | |
7792 | ||
7793 | btrfs_mark_buffer_dirty(leaf); | |
7794 | btrfs_free_path(path); | |
7795 | ||
1e144fb8 OS |
7796 | ret = remove_from_free_space_tree(trans, fs_info, ins->objectid, |
7797 | num_bytes); | |
7798 | if (ret) | |
7799 | return ret; | |
7800 | ||
ce93ec54 JB |
7801 | ret = update_block_group(trans, root, ins->objectid, root->nodesize, |
7802 | 1); | |
79787eaa | 7803 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb | 7804 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
c1c9ff7c | 7805 | ins->objectid, ins->offset); |
5d4f98a2 YZ |
7806 | BUG(); |
7807 | } | |
0be5dc67 | 7808 | |
707e8a07 | 7809 | trace_btrfs_reserved_extent_alloc(root, ins->objectid, root->nodesize); |
5d4f98a2 YZ |
7810 | return ret; |
7811 | } | |
7812 | ||
7813 | int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
7814 | struct btrfs_root *root, | |
7815 | u64 root_objectid, u64 owner, | |
5846a3c2 QW |
7816 | u64 offset, u64 ram_bytes, |
7817 | struct btrfs_key *ins) | |
5d4f98a2 YZ |
7818 | { |
7819 | int ret; | |
7820 | ||
7821 | BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
7822 | ||
66d7e7f0 AJ |
7823 | ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid, |
7824 | ins->offset, 0, | |
7825 | root_objectid, owner, offset, | |
5846a3c2 QW |
7826 | ram_bytes, BTRFS_ADD_DELAYED_EXTENT, |
7827 | NULL); | |
e6dcd2dc CM |
7828 | return ret; |
7829 | } | |
e02119d5 CM |
7830 | |
7831 | /* | |
7832 | * this is used by the tree logging recovery code. It records that | |
7833 | * an extent has been allocated and makes sure to clear the free | |
7834 | * space cache bits as well | |
7835 | */ | |
5d4f98a2 YZ |
7836 | int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans, |
7837 | struct btrfs_root *root, | |
7838 | u64 root_objectid, u64 owner, u64 offset, | |
7839 | struct btrfs_key *ins) | |
e02119d5 CM |
7840 | { |
7841 | int ret; | |
7842 | struct btrfs_block_group_cache *block_group; | |
11833d66 | 7843 | |
8c2a1a30 JB |
7844 | /* |
7845 | * Mixed block groups will exclude before processing the log so we only | |
7846 | * need to do the exlude dance if this fs isn't mixed. | |
7847 | */ | |
7848 | if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) { | |
7849 | ret = __exclude_logged_extent(root, ins->objectid, ins->offset); | |
b50c6e25 | 7850 | if (ret) |
8c2a1a30 | 7851 | return ret; |
11833d66 YZ |
7852 | } |
7853 | ||
8c2a1a30 JB |
7854 | block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid); |
7855 | if (!block_group) | |
7856 | return -EINVAL; | |
7857 | ||
fb25e914 | 7858 | ret = btrfs_update_reserved_bytes(block_group, ins->offset, |
e570fd27 | 7859 | RESERVE_ALLOC_NO_ACCOUNT, 0); |
79787eaa | 7860 | BUG_ON(ret); /* logic error */ |
5d4f98a2 YZ |
7861 | ret = alloc_reserved_file_extent(trans, root, 0, root_objectid, |
7862 | 0, owner, offset, ins, 1); | |
b50c6e25 | 7863 | btrfs_put_block_group(block_group); |
e02119d5 CM |
7864 | return ret; |
7865 | } | |
7866 | ||
48a3b636 ES |
7867 | static struct extent_buffer * |
7868 | btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
fe864576 | 7869 | u64 bytenr, int level) |
65b51a00 CM |
7870 | { |
7871 | struct extent_buffer *buf; | |
7872 | ||
a83fffb7 | 7873 | buf = btrfs_find_create_tree_block(root, bytenr); |
65b51a00 CM |
7874 | if (!buf) |
7875 | return ERR_PTR(-ENOMEM); | |
7876 | btrfs_set_header_generation(buf, trans->transid); | |
85d4e461 | 7877 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level); |
65b51a00 | 7878 | btrfs_tree_lock(buf); |
01d58472 | 7879 | clean_tree_block(trans, root->fs_info, buf); |
3083ee2e | 7880 | clear_bit(EXTENT_BUFFER_STALE, &buf->bflags); |
b4ce94de CM |
7881 | |
7882 | btrfs_set_lock_blocking(buf); | |
4db8c528 | 7883 | set_extent_buffer_uptodate(buf); |
b4ce94de | 7884 | |
d0c803c4 | 7885 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { |
656f30db | 7886 | buf->log_index = root->log_transid % 2; |
8cef4e16 YZ |
7887 | /* |
7888 | * we allow two log transactions at a time, use different | |
7889 | * EXENT bit to differentiate dirty pages. | |
7890 | */ | |
656f30db | 7891 | if (buf->log_index == 0) |
8cef4e16 YZ |
7892 | set_extent_dirty(&root->dirty_log_pages, buf->start, |
7893 | buf->start + buf->len - 1, GFP_NOFS); | |
7894 | else | |
7895 | set_extent_new(&root->dirty_log_pages, buf->start, | |
7896 | buf->start + buf->len - 1, GFP_NOFS); | |
d0c803c4 | 7897 | } else { |
656f30db | 7898 | buf->log_index = -1; |
d0c803c4 | 7899 | set_extent_dirty(&trans->transaction->dirty_pages, buf->start, |
65b51a00 | 7900 | buf->start + buf->len - 1, GFP_NOFS); |
d0c803c4 | 7901 | } |
65b51a00 | 7902 | trans->blocks_used++; |
b4ce94de | 7903 | /* this returns a buffer locked for blocking */ |
65b51a00 CM |
7904 | return buf; |
7905 | } | |
7906 | ||
f0486c68 YZ |
7907 | static struct btrfs_block_rsv * |
7908 | use_block_rsv(struct btrfs_trans_handle *trans, | |
7909 | struct btrfs_root *root, u32 blocksize) | |
7910 | { | |
7911 | struct btrfs_block_rsv *block_rsv; | |
68a82277 | 7912 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; |
f0486c68 | 7913 | int ret; |
d88033db | 7914 | bool global_updated = false; |
f0486c68 YZ |
7915 | |
7916 | block_rsv = get_block_rsv(trans, root); | |
7917 | ||
b586b323 MX |
7918 | if (unlikely(block_rsv->size == 0)) |
7919 | goto try_reserve; | |
d88033db | 7920 | again: |
f0486c68 YZ |
7921 | ret = block_rsv_use_bytes(block_rsv, blocksize); |
7922 | if (!ret) | |
7923 | return block_rsv; | |
7924 | ||
b586b323 MX |
7925 | if (block_rsv->failfast) |
7926 | return ERR_PTR(ret); | |
7927 | ||
d88033db MX |
7928 | if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) { |
7929 | global_updated = true; | |
7930 | update_global_block_rsv(root->fs_info); | |
7931 | goto again; | |
7932 | } | |
7933 | ||
b586b323 MX |
7934 | if (btrfs_test_opt(root, ENOSPC_DEBUG)) { |
7935 | static DEFINE_RATELIMIT_STATE(_rs, | |
7936 | DEFAULT_RATELIMIT_INTERVAL * 10, | |
7937 | /*DEFAULT_RATELIMIT_BURST*/ 1); | |
7938 | if (__ratelimit(&_rs)) | |
7939 | WARN(1, KERN_DEBUG | |
efe120a0 | 7940 | "BTRFS: block rsv returned %d\n", ret); |
b586b323 MX |
7941 | } |
7942 | try_reserve: | |
7943 | ret = reserve_metadata_bytes(root, block_rsv, blocksize, | |
7944 | BTRFS_RESERVE_NO_FLUSH); | |
7945 | if (!ret) | |
7946 | return block_rsv; | |
7947 | /* | |
7948 | * If we couldn't reserve metadata bytes try and use some from | |
5881cfc9 MX |
7949 | * the global reserve if its space type is the same as the global |
7950 | * reservation. | |
b586b323 | 7951 | */ |
5881cfc9 MX |
7952 | if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL && |
7953 | block_rsv->space_info == global_rsv->space_info) { | |
b586b323 MX |
7954 | ret = block_rsv_use_bytes(global_rsv, blocksize); |
7955 | if (!ret) | |
7956 | return global_rsv; | |
7957 | } | |
7958 | return ERR_PTR(ret); | |
f0486c68 YZ |
7959 | } |
7960 | ||
8c2a3ca2 JB |
7961 | static void unuse_block_rsv(struct btrfs_fs_info *fs_info, |
7962 | struct btrfs_block_rsv *block_rsv, u32 blocksize) | |
f0486c68 YZ |
7963 | { |
7964 | block_rsv_add_bytes(block_rsv, blocksize, 0); | |
8c2a3ca2 | 7965 | block_rsv_release_bytes(fs_info, block_rsv, NULL, 0); |
f0486c68 YZ |
7966 | } |
7967 | ||
fec577fb | 7968 | /* |
f0486c68 | 7969 | * finds a free extent and does all the dirty work required for allocation |
67b7859e | 7970 | * returns the tree buffer or an ERR_PTR on error. |
fec577fb | 7971 | */ |
4d75f8a9 DS |
7972 | struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans, |
7973 | struct btrfs_root *root, | |
5d4f98a2 YZ |
7974 | u64 parent, u64 root_objectid, |
7975 | struct btrfs_disk_key *key, int level, | |
5581a51a | 7976 | u64 hint, u64 empty_size) |
fec577fb | 7977 | { |
e2fa7227 | 7978 | struct btrfs_key ins; |
f0486c68 | 7979 | struct btrfs_block_rsv *block_rsv; |
5f39d397 | 7980 | struct extent_buffer *buf; |
67b7859e | 7981 | struct btrfs_delayed_extent_op *extent_op; |
f0486c68 YZ |
7982 | u64 flags = 0; |
7983 | int ret; | |
4d75f8a9 | 7984 | u32 blocksize = root->nodesize; |
3173a18f JB |
7985 | bool skinny_metadata = btrfs_fs_incompat(root->fs_info, |
7986 | SKINNY_METADATA); | |
fec577fb | 7987 | |
fccb84c9 | 7988 | if (btrfs_test_is_dummy_root(root)) { |
faa2dbf0 | 7989 | buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr, |
fe864576 | 7990 | level); |
faa2dbf0 JB |
7991 | if (!IS_ERR(buf)) |
7992 | root->alloc_bytenr += blocksize; | |
7993 | return buf; | |
7994 | } | |
fccb84c9 | 7995 | |
f0486c68 YZ |
7996 | block_rsv = use_block_rsv(trans, root, blocksize); |
7997 | if (IS_ERR(block_rsv)) | |
7998 | return ERR_CAST(block_rsv); | |
7999 | ||
00361589 | 8000 | ret = btrfs_reserve_extent(root, blocksize, blocksize, |
e570fd27 | 8001 | empty_size, hint, &ins, 0, 0); |
67b7859e OS |
8002 | if (ret) |
8003 | goto out_unuse; | |
55c69072 | 8004 | |
fe864576 | 8005 | buf = btrfs_init_new_buffer(trans, root, ins.objectid, level); |
67b7859e OS |
8006 | if (IS_ERR(buf)) { |
8007 | ret = PTR_ERR(buf); | |
8008 | goto out_free_reserved; | |
8009 | } | |
f0486c68 YZ |
8010 | |
8011 | if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
8012 | if (parent == 0) | |
8013 | parent = ins.objectid; | |
8014 | flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
8015 | } else | |
8016 | BUG_ON(parent > 0); | |
8017 | ||
8018 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) { | |
78a6184a | 8019 | extent_op = btrfs_alloc_delayed_extent_op(); |
67b7859e OS |
8020 | if (!extent_op) { |
8021 | ret = -ENOMEM; | |
8022 | goto out_free_buf; | |
8023 | } | |
f0486c68 YZ |
8024 | if (key) |
8025 | memcpy(&extent_op->key, key, sizeof(extent_op->key)); | |
8026 | else | |
8027 | memset(&extent_op->key, 0, sizeof(extent_op->key)); | |
8028 | extent_op->flags_to_set = flags; | |
3173a18f JB |
8029 | if (skinny_metadata) |
8030 | extent_op->update_key = 0; | |
8031 | else | |
8032 | extent_op->update_key = 1; | |
f0486c68 YZ |
8033 | extent_op->update_flags = 1; |
8034 | extent_op->is_data = 0; | |
b1c79e09 | 8035 | extent_op->level = level; |
f0486c68 | 8036 | |
66d7e7f0 | 8037 | ret = btrfs_add_delayed_tree_ref(root->fs_info, trans, |
67b7859e OS |
8038 | ins.objectid, ins.offset, |
8039 | parent, root_objectid, level, | |
8040 | BTRFS_ADD_DELAYED_EXTENT, | |
b06c4bf5 | 8041 | extent_op); |
67b7859e OS |
8042 | if (ret) |
8043 | goto out_free_delayed; | |
f0486c68 | 8044 | } |
fec577fb | 8045 | return buf; |
67b7859e OS |
8046 | |
8047 | out_free_delayed: | |
8048 | btrfs_free_delayed_extent_op(extent_op); | |
8049 | out_free_buf: | |
8050 | free_extent_buffer(buf); | |
8051 | out_free_reserved: | |
8052 | btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 0); | |
8053 | out_unuse: | |
8054 | unuse_block_rsv(root->fs_info, block_rsv, blocksize); | |
8055 | return ERR_PTR(ret); | |
fec577fb | 8056 | } |
a28ec197 | 8057 | |
2c47e605 YZ |
8058 | struct walk_control { |
8059 | u64 refs[BTRFS_MAX_LEVEL]; | |
8060 | u64 flags[BTRFS_MAX_LEVEL]; | |
8061 | struct btrfs_key update_progress; | |
8062 | int stage; | |
8063 | int level; | |
8064 | int shared_level; | |
8065 | int update_ref; | |
8066 | int keep_locks; | |
1c4850e2 YZ |
8067 | int reada_slot; |
8068 | int reada_count; | |
66d7e7f0 | 8069 | int for_reloc; |
2c47e605 YZ |
8070 | }; |
8071 | ||
8072 | #define DROP_REFERENCE 1 | |
8073 | #define UPDATE_BACKREF 2 | |
8074 | ||
1c4850e2 YZ |
8075 | static noinline void reada_walk_down(struct btrfs_trans_handle *trans, |
8076 | struct btrfs_root *root, | |
8077 | struct walk_control *wc, | |
8078 | struct btrfs_path *path) | |
6407bf6d | 8079 | { |
1c4850e2 YZ |
8080 | u64 bytenr; |
8081 | u64 generation; | |
8082 | u64 refs; | |
94fcca9f | 8083 | u64 flags; |
5d4f98a2 | 8084 | u32 nritems; |
1c4850e2 YZ |
8085 | u32 blocksize; |
8086 | struct btrfs_key key; | |
8087 | struct extent_buffer *eb; | |
6407bf6d | 8088 | int ret; |
1c4850e2 YZ |
8089 | int slot; |
8090 | int nread = 0; | |
6407bf6d | 8091 | |
1c4850e2 YZ |
8092 | if (path->slots[wc->level] < wc->reada_slot) { |
8093 | wc->reada_count = wc->reada_count * 2 / 3; | |
8094 | wc->reada_count = max(wc->reada_count, 2); | |
8095 | } else { | |
8096 | wc->reada_count = wc->reada_count * 3 / 2; | |
8097 | wc->reada_count = min_t(int, wc->reada_count, | |
8098 | BTRFS_NODEPTRS_PER_BLOCK(root)); | |
8099 | } | |
7bb86316 | 8100 | |
1c4850e2 YZ |
8101 | eb = path->nodes[wc->level]; |
8102 | nritems = btrfs_header_nritems(eb); | |
707e8a07 | 8103 | blocksize = root->nodesize; |
bd56b302 | 8104 | |
1c4850e2 YZ |
8105 | for (slot = path->slots[wc->level]; slot < nritems; slot++) { |
8106 | if (nread >= wc->reada_count) | |
8107 | break; | |
bd56b302 | 8108 | |
2dd3e67b | 8109 | cond_resched(); |
1c4850e2 YZ |
8110 | bytenr = btrfs_node_blockptr(eb, slot); |
8111 | generation = btrfs_node_ptr_generation(eb, slot); | |
2dd3e67b | 8112 | |
1c4850e2 YZ |
8113 | if (slot == path->slots[wc->level]) |
8114 | goto reada; | |
5d4f98a2 | 8115 | |
1c4850e2 YZ |
8116 | if (wc->stage == UPDATE_BACKREF && |
8117 | generation <= root->root_key.offset) | |
bd56b302 CM |
8118 | continue; |
8119 | ||
94fcca9f | 8120 | /* We don't lock the tree block, it's OK to be racy here */ |
3173a18f JB |
8121 | ret = btrfs_lookup_extent_info(trans, root, bytenr, |
8122 | wc->level - 1, 1, &refs, | |
8123 | &flags); | |
79787eaa JM |
8124 | /* We don't care about errors in readahead. */ |
8125 | if (ret < 0) | |
8126 | continue; | |
94fcca9f YZ |
8127 | BUG_ON(refs == 0); |
8128 | ||
1c4850e2 | 8129 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 YZ |
8130 | if (refs == 1) |
8131 | goto reada; | |
bd56b302 | 8132 | |
94fcca9f YZ |
8133 | if (wc->level == 1 && |
8134 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8135 | continue; | |
1c4850e2 YZ |
8136 | if (!wc->update_ref || |
8137 | generation <= root->root_key.offset) | |
8138 | continue; | |
8139 | btrfs_node_key_to_cpu(eb, &key, slot); | |
8140 | ret = btrfs_comp_cpu_keys(&key, | |
8141 | &wc->update_progress); | |
8142 | if (ret < 0) | |
8143 | continue; | |
94fcca9f YZ |
8144 | } else { |
8145 | if (wc->level == 1 && | |
8146 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8147 | continue; | |
6407bf6d | 8148 | } |
1c4850e2 | 8149 | reada: |
d3e46fea | 8150 | readahead_tree_block(root, bytenr); |
1c4850e2 | 8151 | nread++; |
20524f02 | 8152 | } |
1c4850e2 | 8153 | wc->reada_slot = slot; |
20524f02 | 8154 | } |
2c47e605 | 8155 | |
0ed4792a | 8156 | /* |
82bd101b MF |
8157 | * These may not be seen by the usual inc/dec ref code so we have to |
8158 | * add them here. | |
0ed4792a | 8159 | */ |
82bd101b MF |
8160 | static int record_one_subtree_extent(struct btrfs_trans_handle *trans, |
8161 | struct btrfs_root *root, u64 bytenr, | |
8162 | u64 num_bytes) | |
8163 | { | |
8164 | struct btrfs_qgroup_extent_record *qrecord; | |
8165 | struct btrfs_delayed_ref_root *delayed_refs; | |
8166 | ||
8167 | qrecord = kmalloc(sizeof(*qrecord), GFP_NOFS); | |
8168 | if (!qrecord) | |
8169 | return -ENOMEM; | |
8170 | ||
8171 | qrecord->bytenr = bytenr; | |
8172 | qrecord->num_bytes = num_bytes; | |
8173 | qrecord->old_roots = NULL; | |
8174 | ||
8175 | delayed_refs = &trans->transaction->delayed_refs; | |
8176 | spin_lock(&delayed_refs->lock); | |
8177 | if (btrfs_qgroup_insert_dirty_extent(delayed_refs, qrecord)) | |
8178 | kfree(qrecord); | |
8179 | spin_unlock(&delayed_refs->lock); | |
8180 | ||
8181 | return 0; | |
8182 | } | |
8183 | ||
1152651a MF |
8184 | static int account_leaf_items(struct btrfs_trans_handle *trans, |
8185 | struct btrfs_root *root, | |
8186 | struct extent_buffer *eb) | |
8187 | { | |
8188 | int nr = btrfs_header_nritems(eb); | |
82bd101b | 8189 | int i, extent_type, ret; |
1152651a MF |
8190 | struct btrfs_key key; |
8191 | struct btrfs_file_extent_item *fi; | |
8192 | u64 bytenr, num_bytes; | |
8193 | ||
82bd101b MF |
8194 | /* We can be called directly from walk_up_proc() */ |
8195 | if (!root->fs_info->quota_enabled) | |
8196 | return 0; | |
8197 | ||
1152651a MF |
8198 | for (i = 0; i < nr; i++) { |
8199 | btrfs_item_key_to_cpu(eb, &key, i); | |
8200 | ||
8201 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
8202 | continue; | |
8203 | ||
8204 | fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item); | |
8205 | /* filter out non qgroup-accountable extents */ | |
8206 | extent_type = btrfs_file_extent_type(eb, fi); | |
8207 | ||
8208 | if (extent_type == BTRFS_FILE_EXTENT_INLINE) | |
8209 | continue; | |
8210 | ||
8211 | bytenr = btrfs_file_extent_disk_bytenr(eb, fi); | |
8212 | if (!bytenr) | |
8213 | continue; | |
8214 | ||
8215 | num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi); | |
82bd101b MF |
8216 | |
8217 | ret = record_one_subtree_extent(trans, root, bytenr, num_bytes); | |
8218 | if (ret) | |
8219 | return ret; | |
1152651a MF |
8220 | } |
8221 | return 0; | |
8222 | } | |
8223 | ||
8224 | /* | |
8225 | * Walk up the tree from the bottom, freeing leaves and any interior | |
8226 | * nodes which have had all slots visited. If a node (leaf or | |
8227 | * interior) is freed, the node above it will have it's slot | |
8228 | * incremented. The root node will never be freed. | |
8229 | * | |
8230 | * At the end of this function, we should have a path which has all | |
8231 | * slots incremented to the next position for a search. If we need to | |
8232 | * read a new node it will be NULL and the node above it will have the | |
8233 | * correct slot selected for a later read. | |
8234 | * | |
8235 | * If we increment the root nodes slot counter past the number of | |
8236 | * elements, 1 is returned to signal completion of the search. | |
8237 | */ | |
8238 | static int adjust_slots_upwards(struct btrfs_root *root, | |
8239 | struct btrfs_path *path, int root_level) | |
8240 | { | |
8241 | int level = 0; | |
8242 | int nr, slot; | |
8243 | struct extent_buffer *eb; | |
8244 | ||
8245 | if (root_level == 0) | |
8246 | return 1; | |
8247 | ||
8248 | while (level <= root_level) { | |
8249 | eb = path->nodes[level]; | |
8250 | nr = btrfs_header_nritems(eb); | |
8251 | path->slots[level]++; | |
8252 | slot = path->slots[level]; | |
8253 | if (slot >= nr || level == 0) { | |
8254 | /* | |
8255 | * Don't free the root - we will detect this | |
8256 | * condition after our loop and return a | |
8257 | * positive value for caller to stop walking the tree. | |
8258 | */ | |
8259 | if (level != root_level) { | |
8260 | btrfs_tree_unlock_rw(eb, path->locks[level]); | |
8261 | path->locks[level] = 0; | |
8262 | ||
8263 | free_extent_buffer(eb); | |
8264 | path->nodes[level] = NULL; | |
8265 | path->slots[level] = 0; | |
8266 | } | |
8267 | } else { | |
8268 | /* | |
8269 | * We have a valid slot to walk back down | |
8270 | * from. Stop here so caller can process these | |
8271 | * new nodes. | |
8272 | */ | |
8273 | break; | |
8274 | } | |
8275 | ||
8276 | level++; | |
8277 | } | |
8278 | ||
8279 | eb = path->nodes[root_level]; | |
8280 | if (path->slots[root_level] >= btrfs_header_nritems(eb)) | |
8281 | return 1; | |
8282 | ||
8283 | return 0; | |
8284 | } | |
8285 | ||
8286 | /* | |
8287 | * root_eb is the subtree root and is locked before this function is called. | |
8288 | */ | |
8289 | static int account_shared_subtree(struct btrfs_trans_handle *trans, | |
8290 | struct btrfs_root *root, | |
8291 | struct extent_buffer *root_eb, | |
8292 | u64 root_gen, | |
8293 | int root_level) | |
8294 | { | |
8295 | int ret = 0; | |
8296 | int level; | |
8297 | struct extent_buffer *eb = root_eb; | |
8298 | struct btrfs_path *path = NULL; | |
8299 | ||
8300 | BUG_ON(root_level < 0 || root_level > BTRFS_MAX_LEVEL); | |
8301 | BUG_ON(root_eb == NULL); | |
8302 | ||
8303 | if (!root->fs_info->quota_enabled) | |
8304 | return 0; | |
8305 | ||
8306 | if (!extent_buffer_uptodate(root_eb)) { | |
8307 | ret = btrfs_read_buffer(root_eb, root_gen); | |
8308 | if (ret) | |
8309 | goto out; | |
8310 | } | |
8311 | ||
8312 | if (root_level == 0) { | |
8313 | ret = account_leaf_items(trans, root, root_eb); | |
8314 | goto out; | |
8315 | } | |
8316 | ||
8317 | path = btrfs_alloc_path(); | |
8318 | if (!path) | |
8319 | return -ENOMEM; | |
8320 | ||
8321 | /* | |
8322 | * Walk down the tree. Missing extent blocks are filled in as | |
8323 | * we go. Metadata is accounted every time we read a new | |
8324 | * extent block. | |
8325 | * | |
8326 | * When we reach a leaf, we account for file extent items in it, | |
8327 | * walk back up the tree (adjusting slot pointers as we go) | |
8328 | * and restart the search process. | |
8329 | */ | |
8330 | extent_buffer_get(root_eb); /* For path */ | |
8331 | path->nodes[root_level] = root_eb; | |
8332 | path->slots[root_level] = 0; | |
8333 | path->locks[root_level] = 0; /* so release_path doesn't try to unlock */ | |
8334 | walk_down: | |
8335 | level = root_level; | |
8336 | while (level >= 0) { | |
8337 | if (path->nodes[level] == NULL) { | |
1152651a MF |
8338 | int parent_slot; |
8339 | u64 child_gen; | |
8340 | u64 child_bytenr; | |
8341 | ||
8342 | /* We need to get child blockptr/gen from | |
8343 | * parent before we can read it. */ | |
8344 | eb = path->nodes[level + 1]; | |
8345 | parent_slot = path->slots[level + 1]; | |
8346 | child_bytenr = btrfs_node_blockptr(eb, parent_slot); | |
8347 | child_gen = btrfs_node_ptr_generation(eb, parent_slot); | |
8348 | ||
ce86cd59 | 8349 | eb = read_tree_block(root, child_bytenr, child_gen); |
64c043de LB |
8350 | if (IS_ERR(eb)) { |
8351 | ret = PTR_ERR(eb); | |
8352 | goto out; | |
8353 | } else if (!extent_buffer_uptodate(eb)) { | |
8635eda9 | 8354 | free_extent_buffer(eb); |
64c043de | 8355 | ret = -EIO; |
1152651a MF |
8356 | goto out; |
8357 | } | |
8358 | ||
8359 | path->nodes[level] = eb; | |
8360 | path->slots[level] = 0; | |
8361 | ||
8362 | btrfs_tree_read_lock(eb); | |
8363 | btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK); | |
8364 | path->locks[level] = BTRFS_READ_LOCK_BLOCKING; | |
82bd101b MF |
8365 | |
8366 | ret = record_one_subtree_extent(trans, root, child_bytenr, | |
8367 | root->nodesize); | |
8368 | if (ret) | |
8369 | goto out; | |
1152651a MF |
8370 | } |
8371 | ||
8372 | if (level == 0) { | |
8373 | ret = account_leaf_items(trans, root, path->nodes[level]); | |
8374 | if (ret) | |
8375 | goto out; | |
8376 | ||
8377 | /* Nonzero return here means we completed our search */ | |
8378 | ret = adjust_slots_upwards(root, path, root_level); | |
8379 | if (ret) | |
8380 | break; | |
8381 | ||
8382 | /* Restart search with new slots */ | |
8383 | goto walk_down; | |
8384 | } | |
8385 | ||
8386 | level--; | |
8387 | } | |
8388 | ||
8389 | ret = 0; | |
8390 | out: | |
8391 | btrfs_free_path(path); | |
8392 | ||
8393 | return ret; | |
8394 | } | |
8395 | ||
f82d02d9 | 8396 | /* |
2c016dc2 | 8397 | * helper to process tree block while walking down the tree. |
2c47e605 | 8398 | * |
2c47e605 YZ |
8399 | * when wc->stage == UPDATE_BACKREF, this function updates |
8400 | * back refs for pointers in the block. | |
8401 | * | |
8402 | * NOTE: return value 1 means we should stop walking down. | |
f82d02d9 | 8403 | */ |
2c47e605 | 8404 | static noinline int walk_down_proc(struct btrfs_trans_handle *trans, |
5d4f98a2 | 8405 | struct btrfs_root *root, |
2c47e605 | 8406 | struct btrfs_path *path, |
94fcca9f | 8407 | struct walk_control *wc, int lookup_info) |
f82d02d9 | 8408 | { |
2c47e605 YZ |
8409 | int level = wc->level; |
8410 | struct extent_buffer *eb = path->nodes[level]; | |
2c47e605 | 8411 | u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF; |
f82d02d9 YZ |
8412 | int ret; |
8413 | ||
2c47e605 YZ |
8414 | if (wc->stage == UPDATE_BACKREF && |
8415 | btrfs_header_owner(eb) != root->root_key.objectid) | |
8416 | return 1; | |
f82d02d9 | 8417 | |
2c47e605 YZ |
8418 | /* |
8419 | * when reference count of tree block is 1, it won't increase | |
8420 | * again. once full backref flag is set, we never clear it. | |
8421 | */ | |
94fcca9f YZ |
8422 | if (lookup_info && |
8423 | ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) || | |
8424 | (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) { | |
2c47e605 YZ |
8425 | BUG_ON(!path->locks[level]); |
8426 | ret = btrfs_lookup_extent_info(trans, root, | |
3173a18f | 8427 | eb->start, level, 1, |
2c47e605 YZ |
8428 | &wc->refs[level], |
8429 | &wc->flags[level]); | |
79787eaa JM |
8430 | BUG_ON(ret == -ENOMEM); |
8431 | if (ret) | |
8432 | return ret; | |
2c47e605 YZ |
8433 | BUG_ON(wc->refs[level] == 0); |
8434 | } | |
5d4f98a2 | 8435 | |
2c47e605 YZ |
8436 | if (wc->stage == DROP_REFERENCE) { |
8437 | if (wc->refs[level] > 1) | |
8438 | return 1; | |
f82d02d9 | 8439 | |
2c47e605 | 8440 | if (path->locks[level] && !wc->keep_locks) { |
bd681513 | 8441 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
8442 | path->locks[level] = 0; |
8443 | } | |
8444 | return 0; | |
8445 | } | |
f82d02d9 | 8446 | |
2c47e605 YZ |
8447 | /* wc->stage == UPDATE_BACKREF */ |
8448 | if (!(wc->flags[level] & flag)) { | |
8449 | BUG_ON(!path->locks[level]); | |
e339a6b0 | 8450 | ret = btrfs_inc_ref(trans, root, eb, 1); |
79787eaa | 8451 | BUG_ON(ret); /* -ENOMEM */ |
e339a6b0 | 8452 | ret = btrfs_dec_ref(trans, root, eb, 0); |
79787eaa | 8453 | BUG_ON(ret); /* -ENOMEM */ |
2c47e605 | 8454 | ret = btrfs_set_disk_extent_flags(trans, root, eb->start, |
b1c79e09 JB |
8455 | eb->len, flag, |
8456 | btrfs_header_level(eb), 0); | |
79787eaa | 8457 | BUG_ON(ret); /* -ENOMEM */ |
2c47e605 YZ |
8458 | wc->flags[level] |= flag; |
8459 | } | |
8460 | ||
8461 | /* | |
8462 | * the block is shared by multiple trees, so it's not good to | |
8463 | * keep the tree lock | |
8464 | */ | |
8465 | if (path->locks[level] && level > 0) { | |
bd681513 | 8466 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
8467 | path->locks[level] = 0; |
8468 | } | |
8469 | return 0; | |
8470 | } | |
8471 | ||
1c4850e2 | 8472 | /* |
2c016dc2 | 8473 | * helper to process tree block pointer. |
1c4850e2 YZ |
8474 | * |
8475 | * when wc->stage == DROP_REFERENCE, this function checks | |
8476 | * reference count of the block pointed to. if the block | |
8477 | * is shared and we need update back refs for the subtree | |
8478 | * rooted at the block, this function changes wc->stage to | |
8479 | * UPDATE_BACKREF. if the block is shared and there is no | |
8480 | * need to update back, this function drops the reference | |
8481 | * to the block. | |
8482 | * | |
8483 | * NOTE: return value 1 means we should stop walking down. | |
8484 | */ | |
8485 | static noinline int do_walk_down(struct btrfs_trans_handle *trans, | |
8486 | struct btrfs_root *root, | |
8487 | struct btrfs_path *path, | |
94fcca9f | 8488 | struct walk_control *wc, int *lookup_info) |
1c4850e2 YZ |
8489 | { |
8490 | u64 bytenr; | |
8491 | u64 generation; | |
8492 | u64 parent; | |
8493 | u32 blocksize; | |
8494 | struct btrfs_key key; | |
8495 | struct extent_buffer *next; | |
8496 | int level = wc->level; | |
8497 | int reada = 0; | |
8498 | int ret = 0; | |
1152651a | 8499 | bool need_account = false; |
1c4850e2 YZ |
8500 | |
8501 | generation = btrfs_node_ptr_generation(path->nodes[level], | |
8502 | path->slots[level]); | |
8503 | /* | |
8504 | * if the lower level block was created before the snapshot | |
8505 | * was created, we know there is no need to update back refs | |
8506 | * for the subtree | |
8507 | */ | |
8508 | if (wc->stage == UPDATE_BACKREF && | |
94fcca9f YZ |
8509 | generation <= root->root_key.offset) { |
8510 | *lookup_info = 1; | |
1c4850e2 | 8511 | return 1; |
94fcca9f | 8512 | } |
1c4850e2 YZ |
8513 | |
8514 | bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]); | |
707e8a07 | 8515 | blocksize = root->nodesize; |
1c4850e2 | 8516 | |
01d58472 | 8517 | next = btrfs_find_tree_block(root->fs_info, bytenr); |
1c4850e2 | 8518 | if (!next) { |
a83fffb7 | 8519 | next = btrfs_find_create_tree_block(root, bytenr); |
90d2c51d MX |
8520 | if (!next) |
8521 | return -ENOMEM; | |
b2aaaa3b JB |
8522 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, next, |
8523 | level - 1); | |
1c4850e2 YZ |
8524 | reada = 1; |
8525 | } | |
8526 | btrfs_tree_lock(next); | |
8527 | btrfs_set_lock_blocking(next); | |
8528 | ||
3173a18f | 8529 | ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1, |
94fcca9f YZ |
8530 | &wc->refs[level - 1], |
8531 | &wc->flags[level - 1]); | |
79787eaa JM |
8532 | if (ret < 0) { |
8533 | btrfs_tree_unlock(next); | |
8534 | return ret; | |
8535 | } | |
8536 | ||
c2cf52eb SK |
8537 | if (unlikely(wc->refs[level - 1] == 0)) { |
8538 | btrfs_err(root->fs_info, "Missing references."); | |
8539 | BUG(); | |
8540 | } | |
94fcca9f | 8541 | *lookup_info = 0; |
1c4850e2 | 8542 | |
94fcca9f | 8543 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 | 8544 | if (wc->refs[level - 1] > 1) { |
1152651a | 8545 | need_account = true; |
94fcca9f YZ |
8546 | if (level == 1 && |
8547 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8548 | goto skip; | |
8549 | ||
1c4850e2 YZ |
8550 | if (!wc->update_ref || |
8551 | generation <= root->root_key.offset) | |
8552 | goto skip; | |
8553 | ||
8554 | btrfs_node_key_to_cpu(path->nodes[level], &key, | |
8555 | path->slots[level]); | |
8556 | ret = btrfs_comp_cpu_keys(&key, &wc->update_progress); | |
8557 | if (ret < 0) | |
8558 | goto skip; | |
8559 | ||
8560 | wc->stage = UPDATE_BACKREF; | |
8561 | wc->shared_level = level - 1; | |
8562 | } | |
94fcca9f YZ |
8563 | } else { |
8564 | if (level == 1 && | |
8565 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8566 | goto skip; | |
1c4850e2 YZ |
8567 | } |
8568 | ||
b9fab919 | 8569 | if (!btrfs_buffer_uptodate(next, generation, 0)) { |
1c4850e2 YZ |
8570 | btrfs_tree_unlock(next); |
8571 | free_extent_buffer(next); | |
8572 | next = NULL; | |
94fcca9f | 8573 | *lookup_info = 1; |
1c4850e2 YZ |
8574 | } |
8575 | ||
8576 | if (!next) { | |
8577 | if (reada && level == 1) | |
8578 | reada_walk_down(trans, root, wc, path); | |
ce86cd59 | 8579 | next = read_tree_block(root, bytenr, generation); |
64c043de LB |
8580 | if (IS_ERR(next)) { |
8581 | return PTR_ERR(next); | |
8582 | } else if (!extent_buffer_uptodate(next)) { | |
416bc658 | 8583 | free_extent_buffer(next); |
97d9a8a4 | 8584 | return -EIO; |
416bc658 | 8585 | } |
1c4850e2 YZ |
8586 | btrfs_tree_lock(next); |
8587 | btrfs_set_lock_blocking(next); | |
8588 | } | |
8589 | ||
8590 | level--; | |
8591 | BUG_ON(level != btrfs_header_level(next)); | |
8592 | path->nodes[level] = next; | |
8593 | path->slots[level] = 0; | |
bd681513 | 8594 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
1c4850e2 YZ |
8595 | wc->level = level; |
8596 | if (wc->level == 1) | |
8597 | wc->reada_slot = 0; | |
8598 | return 0; | |
8599 | skip: | |
8600 | wc->refs[level - 1] = 0; | |
8601 | wc->flags[level - 1] = 0; | |
94fcca9f YZ |
8602 | if (wc->stage == DROP_REFERENCE) { |
8603 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
8604 | parent = path->nodes[level]->start; | |
8605 | } else { | |
8606 | BUG_ON(root->root_key.objectid != | |
8607 | btrfs_header_owner(path->nodes[level])); | |
8608 | parent = 0; | |
8609 | } | |
1c4850e2 | 8610 | |
1152651a MF |
8611 | if (need_account) { |
8612 | ret = account_shared_subtree(trans, root, next, | |
8613 | generation, level - 1); | |
8614 | if (ret) { | |
94647322 DS |
8615 | btrfs_err_rl(root->fs_info, |
8616 | "Error " | |
1152651a | 8617 | "%d accounting shared subtree. Quota " |
94647322 DS |
8618 | "is out of sync, rescan required.", |
8619 | ret); | |
1152651a MF |
8620 | } |
8621 | } | |
94fcca9f | 8622 | ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent, |
b06c4bf5 | 8623 | root->root_key.objectid, level - 1, 0); |
79787eaa | 8624 | BUG_ON(ret); /* -ENOMEM */ |
1c4850e2 | 8625 | } |
1c4850e2 YZ |
8626 | btrfs_tree_unlock(next); |
8627 | free_extent_buffer(next); | |
94fcca9f | 8628 | *lookup_info = 1; |
1c4850e2 YZ |
8629 | return 1; |
8630 | } | |
8631 | ||
2c47e605 | 8632 | /* |
2c016dc2 | 8633 | * helper to process tree block while walking up the tree. |
2c47e605 YZ |
8634 | * |
8635 | * when wc->stage == DROP_REFERENCE, this function drops | |
8636 | * reference count on the block. | |
8637 | * | |
8638 | * when wc->stage == UPDATE_BACKREF, this function changes | |
8639 | * wc->stage back to DROP_REFERENCE if we changed wc->stage | |
8640 | * to UPDATE_BACKREF previously while processing the block. | |
8641 | * | |
8642 | * NOTE: return value 1 means we should stop walking up. | |
8643 | */ | |
8644 | static noinline int walk_up_proc(struct btrfs_trans_handle *trans, | |
8645 | struct btrfs_root *root, | |
8646 | struct btrfs_path *path, | |
8647 | struct walk_control *wc) | |
8648 | { | |
f0486c68 | 8649 | int ret; |
2c47e605 YZ |
8650 | int level = wc->level; |
8651 | struct extent_buffer *eb = path->nodes[level]; | |
8652 | u64 parent = 0; | |
8653 | ||
8654 | if (wc->stage == UPDATE_BACKREF) { | |
8655 | BUG_ON(wc->shared_level < level); | |
8656 | if (level < wc->shared_level) | |
8657 | goto out; | |
8658 | ||
2c47e605 YZ |
8659 | ret = find_next_key(path, level + 1, &wc->update_progress); |
8660 | if (ret > 0) | |
8661 | wc->update_ref = 0; | |
8662 | ||
8663 | wc->stage = DROP_REFERENCE; | |
8664 | wc->shared_level = -1; | |
8665 | path->slots[level] = 0; | |
8666 | ||
8667 | /* | |
8668 | * check reference count again if the block isn't locked. | |
8669 | * we should start walking down the tree again if reference | |
8670 | * count is one. | |
8671 | */ | |
8672 | if (!path->locks[level]) { | |
8673 | BUG_ON(level == 0); | |
8674 | btrfs_tree_lock(eb); | |
8675 | btrfs_set_lock_blocking(eb); | |
bd681513 | 8676 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
8677 | |
8678 | ret = btrfs_lookup_extent_info(trans, root, | |
3173a18f | 8679 | eb->start, level, 1, |
2c47e605 YZ |
8680 | &wc->refs[level], |
8681 | &wc->flags[level]); | |
79787eaa JM |
8682 | if (ret < 0) { |
8683 | btrfs_tree_unlock_rw(eb, path->locks[level]); | |
3268a246 | 8684 | path->locks[level] = 0; |
79787eaa JM |
8685 | return ret; |
8686 | } | |
2c47e605 YZ |
8687 | BUG_ON(wc->refs[level] == 0); |
8688 | if (wc->refs[level] == 1) { | |
bd681513 | 8689 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
3268a246 | 8690 | path->locks[level] = 0; |
2c47e605 YZ |
8691 | return 1; |
8692 | } | |
f82d02d9 | 8693 | } |
2c47e605 | 8694 | } |
f82d02d9 | 8695 | |
2c47e605 YZ |
8696 | /* wc->stage == DROP_REFERENCE */ |
8697 | BUG_ON(wc->refs[level] > 1 && !path->locks[level]); | |
5d4f98a2 | 8698 | |
2c47e605 YZ |
8699 | if (wc->refs[level] == 1) { |
8700 | if (level == 0) { | |
8701 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
e339a6b0 | 8702 | ret = btrfs_dec_ref(trans, root, eb, 1); |
2c47e605 | 8703 | else |
e339a6b0 | 8704 | ret = btrfs_dec_ref(trans, root, eb, 0); |
79787eaa | 8705 | BUG_ON(ret); /* -ENOMEM */ |
1152651a MF |
8706 | ret = account_leaf_items(trans, root, eb); |
8707 | if (ret) { | |
94647322 DS |
8708 | btrfs_err_rl(root->fs_info, |
8709 | "error " | |
1152651a | 8710 | "%d accounting leaf items. Quota " |
94647322 DS |
8711 | "is out of sync, rescan required.", |
8712 | ret); | |
1152651a | 8713 | } |
2c47e605 YZ |
8714 | } |
8715 | /* make block locked assertion in clean_tree_block happy */ | |
8716 | if (!path->locks[level] && | |
8717 | btrfs_header_generation(eb) == trans->transid) { | |
8718 | btrfs_tree_lock(eb); | |
8719 | btrfs_set_lock_blocking(eb); | |
bd681513 | 8720 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 | 8721 | } |
01d58472 | 8722 | clean_tree_block(trans, root->fs_info, eb); |
2c47e605 YZ |
8723 | } |
8724 | ||
8725 | if (eb == root->node) { | |
8726 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
8727 | parent = eb->start; | |
8728 | else | |
8729 | BUG_ON(root->root_key.objectid != | |
8730 | btrfs_header_owner(eb)); | |
8731 | } else { | |
8732 | if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
8733 | parent = path->nodes[level + 1]->start; | |
8734 | else | |
8735 | BUG_ON(root->root_key.objectid != | |
8736 | btrfs_header_owner(path->nodes[level + 1])); | |
f82d02d9 | 8737 | } |
f82d02d9 | 8738 | |
5581a51a | 8739 | btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1); |
2c47e605 YZ |
8740 | out: |
8741 | wc->refs[level] = 0; | |
8742 | wc->flags[level] = 0; | |
f0486c68 | 8743 | return 0; |
2c47e605 YZ |
8744 | } |
8745 | ||
8746 | static noinline int walk_down_tree(struct btrfs_trans_handle *trans, | |
8747 | struct btrfs_root *root, | |
8748 | struct btrfs_path *path, | |
8749 | struct walk_control *wc) | |
8750 | { | |
2c47e605 | 8751 | int level = wc->level; |
94fcca9f | 8752 | int lookup_info = 1; |
2c47e605 YZ |
8753 | int ret; |
8754 | ||
8755 | while (level >= 0) { | |
94fcca9f | 8756 | ret = walk_down_proc(trans, root, path, wc, lookup_info); |
2c47e605 YZ |
8757 | if (ret > 0) |
8758 | break; | |
8759 | ||
8760 | if (level == 0) | |
8761 | break; | |
8762 | ||
7a7965f8 YZ |
8763 | if (path->slots[level] >= |
8764 | btrfs_header_nritems(path->nodes[level])) | |
8765 | break; | |
8766 | ||
94fcca9f | 8767 | ret = do_walk_down(trans, root, path, wc, &lookup_info); |
1c4850e2 YZ |
8768 | if (ret > 0) { |
8769 | path->slots[level]++; | |
8770 | continue; | |
90d2c51d MX |
8771 | } else if (ret < 0) |
8772 | return ret; | |
1c4850e2 | 8773 | level = wc->level; |
f82d02d9 | 8774 | } |
f82d02d9 YZ |
8775 | return 0; |
8776 | } | |
8777 | ||
d397712b | 8778 | static noinline int walk_up_tree(struct btrfs_trans_handle *trans, |
98ed5174 | 8779 | struct btrfs_root *root, |
f82d02d9 | 8780 | struct btrfs_path *path, |
2c47e605 | 8781 | struct walk_control *wc, int max_level) |
20524f02 | 8782 | { |
2c47e605 | 8783 | int level = wc->level; |
20524f02 | 8784 | int ret; |
9f3a7427 | 8785 | |
2c47e605 YZ |
8786 | path->slots[level] = btrfs_header_nritems(path->nodes[level]); |
8787 | while (level < max_level && path->nodes[level]) { | |
8788 | wc->level = level; | |
8789 | if (path->slots[level] + 1 < | |
8790 | btrfs_header_nritems(path->nodes[level])) { | |
8791 | path->slots[level]++; | |
20524f02 CM |
8792 | return 0; |
8793 | } else { | |
2c47e605 YZ |
8794 | ret = walk_up_proc(trans, root, path, wc); |
8795 | if (ret > 0) | |
8796 | return 0; | |
bd56b302 | 8797 | |
2c47e605 | 8798 | if (path->locks[level]) { |
bd681513 CM |
8799 | btrfs_tree_unlock_rw(path->nodes[level], |
8800 | path->locks[level]); | |
2c47e605 | 8801 | path->locks[level] = 0; |
f82d02d9 | 8802 | } |
2c47e605 YZ |
8803 | free_extent_buffer(path->nodes[level]); |
8804 | path->nodes[level] = NULL; | |
8805 | level++; | |
20524f02 CM |
8806 | } |
8807 | } | |
8808 | return 1; | |
8809 | } | |
8810 | ||
9aca1d51 | 8811 | /* |
2c47e605 YZ |
8812 | * drop a subvolume tree. |
8813 | * | |
8814 | * this function traverses the tree freeing any blocks that only | |
8815 | * referenced by the tree. | |
8816 | * | |
8817 | * when a shared tree block is found. this function decreases its | |
8818 | * reference count by one. if update_ref is true, this function | |
8819 | * also make sure backrefs for the shared block and all lower level | |
8820 | * blocks are properly updated. | |
9d1a2a3a DS |
8821 | * |
8822 | * If called with for_reloc == 0, may exit early with -EAGAIN | |
9aca1d51 | 8823 | */ |
2c536799 | 8824 | int btrfs_drop_snapshot(struct btrfs_root *root, |
66d7e7f0 AJ |
8825 | struct btrfs_block_rsv *block_rsv, int update_ref, |
8826 | int for_reloc) | |
20524f02 | 8827 | { |
5caf2a00 | 8828 | struct btrfs_path *path; |
2c47e605 YZ |
8829 | struct btrfs_trans_handle *trans; |
8830 | struct btrfs_root *tree_root = root->fs_info->tree_root; | |
9f3a7427 | 8831 | struct btrfs_root_item *root_item = &root->root_item; |
2c47e605 YZ |
8832 | struct walk_control *wc; |
8833 | struct btrfs_key key; | |
8834 | int err = 0; | |
8835 | int ret; | |
8836 | int level; | |
d29a9f62 | 8837 | bool root_dropped = false; |
20524f02 | 8838 | |
1152651a MF |
8839 | btrfs_debug(root->fs_info, "Drop subvolume %llu", root->objectid); |
8840 | ||
5caf2a00 | 8841 | path = btrfs_alloc_path(); |
cb1b69f4 TI |
8842 | if (!path) { |
8843 | err = -ENOMEM; | |
8844 | goto out; | |
8845 | } | |
20524f02 | 8846 | |
2c47e605 | 8847 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
38a1a919 MF |
8848 | if (!wc) { |
8849 | btrfs_free_path(path); | |
cb1b69f4 TI |
8850 | err = -ENOMEM; |
8851 | goto out; | |
38a1a919 | 8852 | } |
2c47e605 | 8853 | |
a22285a6 | 8854 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
8855 | if (IS_ERR(trans)) { |
8856 | err = PTR_ERR(trans); | |
8857 | goto out_free; | |
8858 | } | |
98d5dc13 | 8859 | |
3fd0a558 YZ |
8860 | if (block_rsv) |
8861 | trans->block_rsv = block_rsv; | |
2c47e605 | 8862 | |
9f3a7427 | 8863 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { |
2c47e605 | 8864 | level = btrfs_header_level(root->node); |
5d4f98a2 YZ |
8865 | path->nodes[level] = btrfs_lock_root_node(root); |
8866 | btrfs_set_lock_blocking(path->nodes[level]); | |
9f3a7427 | 8867 | path->slots[level] = 0; |
bd681513 | 8868 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
8869 | memset(&wc->update_progress, 0, |
8870 | sizeof(wc->update_progress)); | |
9f3a7427 | 8871 | } else { |
9f3a7427 | 8872 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); |
2c47e605 YZ |
8873 | memcpy(&wc->update_progress, &key, |
8874 | sizeof(wc->update_progress)); | |
8875 | ||
6702ed49 | 8876 | level = root_item->drop_level; |
2c47e605 | 8877 | BUG_ON(level == 0); |
6702ed49 | 8878 | path->lowest_level = level; |
2c47e605 YZ |
8879 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
8880 | path->lowest_level = 0; | |
8881 | if (ret < 0) { | |
8882 | err = ret; | |
79787eaa | 8883 | goto out_end_trans; |
9f3a7427 | 8884 | } |
1c4850e2 | 8885 | WARN_ON(ret > 0); |
2c47e605 | 8886 | |
7d9eb12c CM |
8887 | /* |
8888 | * unlock our path, this is safe because only this | |
8889 | * function is allowed to delete this snapshot | |
8890 | */ | |
5d4f98a2 | 8891 | btrfs_unlock_up_safe(path, 0); |
2c47e605 YZ |
8892 | |
8893 | level = btrfs_header_level(root->node); | |
8894 | while (1) { | |
8895 | btrfs_tree_lock(path->nodes[level]); | |
8896 | btrfs_set_lock_blocking(path->nodes[level]); | |
fec386ac | 8897 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
8898 | |
8899 | ret = btrfs_lookup_extent_info(trans, root, | |
8900 | path->nodes[level]->start, | |
3173a18f | 8901 | level, 1, &wc->refs[level], |
2c47e605 | 8902 | &wc->flags[level]); |
79787eaa JM |
8903 | if (ret < 0) { |
8904 | err = ret; | |
8905 | goto out_end_trans; | |
8906 | } | |
2c47e605 YZ |
8907 | BUG_ON(wc->refs[level] == 0); |
8908 | ||
8909 | if (level == root_item->drop_level) | |
8910 | break; | |
8911 | ||
8912 | btrfs_tree_unlock(path->nodes[level]); | |
fec386ac | 8913 | path->locks[level] = 0; |
2c47e605 YZ |
8914 | WARN_ON(wc->refs[level] != 1); |
8915 | level--; | |
8916 | } | |
9f3a7427 | 8917 | } |
2c47e605 YZ |
8918 | |
8919 | wc->level = level; | |
8920 | wc->shared_level = -1; | |
8921 | wc->stage = DROP_REFERENCE; | |
8922 | wc->update_ref = update_ref; | |
8923 | wc->keep_locks = 0; | |
66d7e7f0 | 8924 | wc->for_reloc = for_reloc; |
1c4850e2 | 8925 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root); |
2c47e605 | 8926 | |
d397712b | 8927 | while (1) { |
9d1a2a3a | 8928 | |
2c47e605 YZ |
8929 | ret = walk_down_tree(trans, root, path, wc); |
8930 | if (ret < 0) { | |
8931 | err = ret; | |
20524f02 | 8932 | break; |
2c47e605 | 8933 | } |
9aca1d51 | 8934 | |
2c47e605 YZ |
8935 | ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL); |
8936 | if (ret < 0) { | |
8937 | err = ret; | |
20524f02 | 8938 | break; |
2c47e605 YZ |
8939 | } |
8940 | ||
8941 | if (ret > 0) { | |
8942 | BUG_ON(wc->stage != DROP_REFERENCE); | |
e7a84565 CM |
8943 | break; |
8944 | } | |
2c47e605 YZ |
8945 | |
8946 | if (wc->stage == DROP_REFERENCE) { | |
8947 | level = wc->level; | |
8948 | btrfs_node_key(path->nodes[level], | |
8949 | &root_item->drop_progress, | |
8950 | path->slots[level]); | |
8951 | root_item->drop_level = level; | |
8952 | } | |
8953 | ||
8954 | BUG_ON(wc->level == 0); | |
3c8f2422 JB |
8955 | if (btrfs_should_end_transaction(trans, tree_root) || |
8956 | (!for_reloc && btrfs_need_cleaner_sleep(root))) { | |
2c47e605 YZ |
8957 | ret = btrfs_update_root(trans, tree_root, |
8958 | &root->root_key, | |
8959 | root_item); | |
79787eaa JM |
8960 | if (ret) { |
8961 | btrfs_abort_transaction(trans, tree_root, ret); | |
8962 | err = ret; | |
8963 | goto out_end_trans; | |
8964 | } | |
2c47e605 | 8965 | |
3fd0a558 | 8966 | btrfs_end_transaction_throttle(trans, tree_root); |
3c8f2422 | 8967 | if (!for_reloc && btrfs_need_cleaner_sleep(root)) { |
efe120a0 | 8968 | pr_debug("BTRFS: drop snapshot early exit\n"); |
3c8f2422 JB |
8969 | err = -EAGAIN; |
8970 | goto out_free; | |
8971 | } | |
8972 | ||
a22285a6 | 8973 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
8974 | if (IS_ERR(trans)) { |
8975 | err = PTR_ERR(trans); | |
8976 | goto out_free; | |
8977 | } | |
3fd0a558 YZ |
8978 | if (block_rsv) |
8979 | trans->block_rsv = block_rsv; | |
c3e69d58 | 8980 | } |
20524f02 | 8981 | } |
b3b4aa74 | 8982 | btrfs_release_path(path); |
79787eaa JM |
8983 | if (err) |
8984 | goto out_end_trans; | |
2c47e605 YZ |
8985 | |
8986 | ret = btrfs_del_root(trans, tree_root, &root->root_key); | |
79787eaa JM |
8987 | if (ret) { |
8988 | btrfs_abort_transaction(trans, tree_root, ret); | |
8989 | goto out_end_trans; | |
8990 | } | |
2c47e605 | 8991 | |
76dda93c | 8992 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { |
cb517eab MX |
8993 | ret = btrfs_find_root(tree_root, &root->root_key, path, |
8994 | NULL, NULL); | |
79787eaa JM |
8995 | if (ret < 0) { |
8996 | btrfs_abort_transaction(trans, tree_root, ret); | |
8997 | err = ret; | |
8998 | goto out_end_trans; | |
8999 | } else if (ret > 0) { | |
84cd948c JB |
9000 | /* if we fail to delete the orphan item this time |
9001 | * around, it'll get picked up the next time. | |
9002 | * | |
9003 | * The most common failure here is just -ENOENT. | |
9004 | */ | |
9005 | btrfs_del_orphan_item(trans, tree_root, | |
9006 | root->root_key.objectid); | |
76dda93c YZ |
9007 | } |
9008 | } | |
9009 | ||
27cdeb70 | 9010 | if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) { |
2b9dbef2 | 9011 | btrfs_add_dropped_root(trans, root); |
76dda93c YZ |
9012 | } else { |
9013 | free_extent_buffer(root->node); | |
9014 | free_extent_buffer(root->commit_root); | |
b0feb9d9 | 9015 | btrfs_put_fs_root(root); |
76dda93c | 9016 | } |
d29a9f62 | 9017 | root_dropped = true; |
79787eaa | 9018 | out_end_trans: |
3fd0a558 | 9019 | btrfs_end_transaction_throttle(trans, tree_root); |
79787eaa | 9020 | out_free: |
2c47e605 | 9021 | kfree(wc); |
5caf2a00 | 9022 | btrfs_free_path(path); |
cb1b69f4 | 9023 | out: |
d29a9f62 JB |
9024 | /* |
9025 | * So if we need to stop dropping the snapshot for whatever reason we | |
9026 | * need to make sure to add it back to the dead root list so that we | |
9027 | * keep trying to do the work later. This also cleans up roots if we | |
9028 | * don't have it in the radix (like when we recover after a power fail | |
9029 | * or unmount) so we don't leak memory. | |
9030 | */ | |
b37b39cd | 9031 | if (!for_reloc && root_dropped == false) |
d29a9f62 | 9032 | btrfs_add_dead_root(root); |
90515e7f | 9033 | if (err && err != -EAGAIN) |
a4553fef | 9034 | btrfs_std_error(root->fs_info, err, NULL); |
2c536799 | 9035 | return err; |
20524f02 | 9036 | } |
9078a3e1 | 9037 | |
2c47e605 YZ |
9038 | /* |
9039 | * drop subtree rooted at tree block 'node'. | |
9040 | * | |
9041 | * NOTE: this function will unlock and release tree block 'node' | |
66d7e7f0 | 9042 | * only used by relocation code |
2c47e605 | 9043 | */ |
f82d02d9 YZ |
9044 | int btrfs_drop_subtree(struct btrfs_trans_handle *trans, |
9045 | struct btrfs_root *root, | |
9046 | struct extent_buffer *node, | |
9047 | struct extent_buffer *parent) | |
9048 | { | |
9049 | struct btrfs_path *path; | |
2c47e605 | 9050 | struct walk_control *wc; |
f82d02d9 YZ |
9051 | int level; |
9052 | int parent_level; | |
9053 | int ret = 0; | |
9054 | int wret; | |
9055 | ||
2c47e605 YZ |
9056 | BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); |
9057 | ||
f82d02d9 | 9058 | path = btrfs_alloc_path(); |
db5b493a TI |
9059 | if (!path) |
9060 | return -ENOMEM; | |
f82d02d9 | 9061 | |
2c47e605 | 9062 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
db5b493a TI |
9063 | if (!wc) { |
9064 | btrfs_free_path(path); | |
9065 | return -ENOMEM; | |
9066 | } | |
2c47e605 | 9067 | |
b9447ef8 | 9068 | btrfs_assert_tree_locked(parent); |
f82d02d9 YZ |
9069 | parent_level = btrfs_header_level(parent); |
9070 | extent_buffer_get(parent); | |
9071 | path->nodes[parent_level] = parent; | |
9072 | path->slots[parent_level] = btrfs_header_nritems(parent); | |
9073 | ||
b9447ef8 | 9074 | btrfs_assert_tree_locked(node); |
f82d02d9 | 9075 | level = btrfs_header_level(node); |
f82d02d9 YZ |
9076 | path->nodes[level] = node; |
9077 | path->slots[level] = 0; | |
bd681513 | 9078 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
9079 | |
9080 | wc->refs[parent_level] = 1; | |
9081 | wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
9082 | wc->level = level; | |
9083 | wc->shared_level = -1; | |
9084 | wc->stage = DROP_REFERENCE; | |
9085 | wc->update_ref = 0; | |
9086 | wc->keep_locks = 1; | |
66d7e7f0 | 9087 | wc->for_reloc = 1; |
1c4850e2 | 9088 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root); |
f82d02d9 YZ |
9089 | |
9090 | while (1) { | |
2c47e605 YZ |
9091 | wret = walk_down_tree(trans, root, path, wc); |
9092 | if (wret < 0) { | |
f82d02d9 | 9093 | ret = wret; |
f82d02d9 | 9094 | break; |
2c47e605 | 9095 | } |
f82d02d9 | 9096 | |
2c47e605 | 9097 | wret = walk_up_tree(trans, root, path, wc, parent_level); |
f82d02d9 YZ |
9098 | if (wret < 0) |
9099 | ret = wret; | |
9100 | if (wret != 0) | |
9101 | break; | |
9102 | } | |
9103 | ||
2c47e605 | 9104 | kfree(wc); |
f82d02d9 YZ |
9105 | btrfs_free_path(path); |
9106 | return ret; | |
9107 | } | |
9108 | ||
ec44a35c CM |
9109 | static u64 update_block_group_flags(struct btrfs_root *root, u64 flags) |
9110 | { | |
9111 | u64 num_devices; | |
fc67c450 | 9112 | u64 stripped; |
e4d8ec0f | 9113 | |
fc67c450 ID |
9114 | /* |
9115 | * if restripe for this chunk_type is on pick target profile and | |
9116 | * return, otherwise do the usual balance | |
9117 | */ | |
9118 | stripped = get_restripe_target(root->fs_info, flags); | |
9119 | if (stripped) | |
9120 | return extended_to_chunk(stripped); | |
e4d8ec0f | 9121 | |
95669976 | 9122 | num_devices = root->fs_info->fs_devices->rw_devices; |
cd02dca5 | 9123 | |
fc67c450 | 9124 | stripped = BTRFS_BLOCK_GROUP_RAID0 | |
53b381b3 | 9125 | BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 | |
fc67c450 ID |
9126 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10; |
9127 | ||
ec44a35c CM |
9128 | if (num_devices == 1) { |
9129 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
9130 | stripped = flags & ~stripped; | |
9131 | ||
9132 | /* turn raid0 into single device chunks */ | |
9133 | if (flags & BTRFS_BLOCK_GROUP_RAID0) | |
9134 | return stripped; | |
9135 | ||
9136 | /* turn mirroring into duplication */ | |
9137 | if (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
9138 | BTRFS_BLOCK_GROUP_RAID10)) | |
9139 | return stripped | BTRFS_BLOCK_GROUP_DUP; | |
ec44a35c CM |
9140 | } else { |
9141 | /* they already had raid on here, just return */ | |
ec44a35c CM |
9142 | if (flags & stripped) |
9143 | return flags; | |
9144 | ||
9145 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
9146 | stripped = flags & ~stripped; | |
9147 | ||
9148 | /* switch duplicated blocks with raid1 */ | |
9149 | if (flags & BTRFS_BLOCK_GROUP_DUP) | |
9150 | return stripped | BTRFS_BLOCK_GROUP_RAID1; | |
9151 | ||
e3176ca2 | 9152 | /* this is drive concat, leave it alone */ |
ec44a35c | 9153 | } |
e3176ca2 | 9154 | |
ec44a35c CM |
9155 | return flags; |
9156 | } | |
9157 | ||
868f401a | 9158 | static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force) |
0ef3e66b | 9159 | { |
f0486c68 YZ |
9160 | struct btrfs_space_info *sinfo = cache->space_info; |
9161 | u64 num_bytes; | |
199c36ea | 9162 | u64 min_allocable_bytes; |
f0486c68 | 9163 | int ret = -ENOSPC; |
0ef3e66b | 9164 | |
199c36ea MX |
9165 | /* |
9166 | * We need some metadata space and system metadata space for | |
9167 | * allocating chunks in some corner cases until we force to set | |
9168 | * it to be readonly. | |
9169 | */ | |
9170 | if ((sinfo->flags & | |
9171 | (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) && | |
9172 | !force) | |
9173 | min_allocable_bytes = 1 * 1024 * 1024; | |
9174 | else | |
9175 | min_allocable_bytes = 0; | |
9176 | ||
f0486c68 YZ |
9177 | spin_lock(&sinfo->lock); |
9178 | spin_lock(&cache->lock); | |
61cfea9b W |
9179 | |
9180 | if (cache->ro) { | |
868f401a | 9181 | cache->ro++; |
61cfea9b W |
9182 | ret = 0; |
9183 | goto out; | |
9184 | } | |
9185 | ||
f0486c68 YZ |
9186 | num_bytes = cache->key.offset - cache->reserved - cache->pinned - |
9187 | cache->bytes_super - btrfs_block_group_used(&cache->item); | |
9188 | ||
9189 | if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned + | |
37be25bc JB |
9190 | sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes + |
9191 | min_allocable_bytes <= sinfo->total_bytes) { | |
f0486c68 | 9192 | sinfo->bytes_readonly += num_bytes; |
868f401a | 9193 | cache->ro++; |
633c0aad | 9194 | list_add_tail(&cache->ro_list, &sinfo->ro_bgs); |
f0486c68 YZ |
9195 | ret = 0; |
9196 | } | |
61cfea9b | 9197 | out: |
f0486c68 YZ |
9198 | spin_unlock(&cache->lock); |
9199 | spin_unlock(&sinfo->lock); | |
9200 | return ret; | |
9201 | } | |
7d9eb12c | 9202 | |
868f401a | 9203 | int btrfs_inc_block_group_ro(struct btrfs_root *root, |
f0486c68 | 9204 | struct btrfs_block_group_cache *cache) |
c286ac48 | 9205 | |
f0486c68 YZ |
9206 | { |
9207 | struct btrfs_trans_handle *trans; | |
9208 | u64 alloc_flags; | |
9209 | int ret; | |
7d9eb12c | 9210 | |
1bbc621e | 9211 | again: |
ff5714cc | 9212 | trans = btrfs_join_transaction(root); |
79787eaa JM |
9213 | if (IS_ERR(trans)) |
9214 | return PTR_ERR(trans); | |
5d4f98a2 | 9215 | |
1bbc621e CM |
9216 | /* |
9217 | * we're not allowed to set block groups readonly after the dirty | |
9218 | * block groups cache has started writing. If it already started, | |
9219 | * back off and let this transaction commit | |
9220 | */ | |
9221 | mutex_lock(&root->fs_info->ro_block_group_mutex); | |
3204d33c | 9222 | if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) { |
1bbc621e CM |
9223 | u64 transid = trans->transid; |
9224 | ||
9225 | mutex_unlock(&root->fs_info->ro_block_group_mutex); | |
9226 | btrfs_end_transaction(trans, root); | |
9227 | ||
9228 | ret = btrfs_wait_for_commit(root, transid); | |
9229 | if (ret) | |
9230 | return ret; | |
9231 | goto again; | |
9232 | } | |
9233 | ||
153c35b6 CM |
9234 | /* |
9235 | * if we are changing raid levels, try to allocate a corresponding | |
9236 | * block group with the new raid level. | |
9237 | */ | |
9238 | alloc_flags = update_block_group_flags(root, cache->flags); | |
9239 | if (alloc_flags != cache->flags) { | |
9240 | ret = do_chunk_alloc(trans, root, alloc_flags, | |
9241 | CHUNK_ALLOC_FORCE); | |
9242 | /* | |
9243 | * ENOSPC is allowed here, we may have enough space | |
9244 | * already allocated at the new raid level to | |
9245 | * carry on | |
9246 | */ | |
9247 | if (ret == -ENOSPC) | |
9248 | ret = 0; | |
9249 | if (ret < 0) | |
9250 | goto out; | |
9251 | } | |
1bbc621e | 9252 | |
868f401a | 9253 | ret = inc_block_group_ro(cache, 0); |
f0486c68 YZ |
9254 | if (!ret) |
9255 | goto out; | |
9256 | alloc_flags = get_alloc_profile(root, cache->space_info->flags); | |
698d0082 | 9257 | ret = do_chunk_alloc(trans, root, alloc_flags, |
0e4f8f88 | 9258 | CHUNK_ALLOC_FORCE); |
f0486c68 YZ |
9259 | if (ret < 0) |
9260 | goto out; | |
868f401a | 9261 | ret = inc_block_group_ro(cache, 0); |
f0486c68 | 9262 | out: |
2f081088 SL |
9263 | if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) { |
9264 | alloc_flags = update_block_group_flags(root, cache->flags); | |
a9629596 | 9265 | lock_chunks(root->fs_info->chunk_root); |
4617ea3a | 9266 | check_system_chunk(trans, root, alloc_flags); |
a9629596 | 9267 | unlock_chunks(root->fs_info->chunk_root); |
2f081088 | 9268 | } |
1bbc621e | 9269 | mutex_unlock(&root->fs_info->ro_block_group_mutex); |
2f081088 | 9270 | |
f0486c68 YZ |
9271 | btrfs_end_transaction(trans, root); |
9272 | return ret; | |
9273 | } | |
5d4f98a2 | 9274 | |
c87f08ca CM |
9275 | int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans, |
9276 | struct btrfs_root *root, u64 type) | |
9277 | { | |
9278 | u64 alloc_flags = get_alloc_profile(root, type); | |
698d0082 | 9279 | return do_chunk_alloc(trans, root, alloc_flags, |
0e4f8f88 | 9280 | CHUNK_ALLOC_FORCE); |
c87f08ca CM |
9281 | } |
9282 | ||
6d07bcec MX |
9283 | /* |
9284 | * helper to account the unused space of all the readonly block group in the | |
633c0aad | 9285 | * space_info. takes mirrors into account. |
6d07bcec | 9286 | */ |
633c0aad | 9287 | u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo) |
6d07bcec MX |
9288 | { |
9289 | struct btrfs_block_group_cache *block_group; | |
9290 | u64 free_bytes = 0; | |
9291 | int factor; | |
9292 | ||
633c0aad JB |
9293 | /* It's df, we don't care if it's racey */ |
9294 | if (list_empty(&sinfo->ro_bgs)) | |
9295 | return 0; | |
9296 | ||
9297 | spin_lock(&sinfo->lock); | |
9298 | list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) { | |
6d07bcec MX |
9299 | spin_lock(&block_group->lock); |
9300 | ||
9301 | if (!block_group->ro) { | |
9302 | spin_unlock(&block_group->lock); | |
9303 | continue; | |
9304 | } | |
9305 | ||
9306 | if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
9307 | BTRFS_BLOCK_GROUP_RAID10 | | |
9308 | BTRFS_BLOCK_GROUP_DUP)) | |
9309 | factor = 2; | |
9310 | else | |
9311 | factor = 1; | |
9312 | ||
9313 | free_bytes += (block_group->key.offset - | |
9314 | btrfs_block_group_used(&block_group->item)) * | |
9315 | factor; | |
9316 | ||
9317 | spin_unlock(&block_group->lock); | |
9318 | } | |
6d07bcec MX |
9319 | spin_unlock(&sinfo->lock); |
9320 | ||
9321 | return free_bytes; | |
9322 | } | |
9323 | ||
868f401a | 9324 | void btrfs_dec_block_group_ro(struct btrfs_root *root, |
f0486c68 | 9325 | struct btrfs_block_group_cache *cache) |
5d4f98a2 | 9326 | { |
f0486c68 YZ |
9327 | struct btrfs_space_info *sinfo = cache->space_info; |
9328 | u64 num_bytes; | |
9329 | ||
9330 | BUG_ON(!cache->ro); | |
9331 | ||
9332 | spin_lock(&sinfo->lock); | |
9333 | spin_lock(&cache->lock); | |
868f401a Z |
9334 | if (!--cache->ro) { |
9335 | num_bytes = cache->key.offset - cache->reserved - | |
9336 | cache->pinned - cache->bytes_super - | |
9337 | btrfs_block_group_used(&cache->item); | |
9338 | sinfo->bytes_readonly -= num_bytes; | |
9339 | list_del_init(&cache->ro_list); | |
9340 | } | |
f0486c68 YZ |
9341 | spin_unlock(&cache->lock); |
9342 | spin_unlock(&sinfo->lock); | |
5d4f98a2 YZ |
9343 | } |
9344 | ||
ba1bf481 JB |
9345 | /* |
9346 | * checks to see if its even possible to relocate this block group. | |
9347 | * | |
9348 | * @return - -1 if it's not a good idea to relocate this block group, 0 if its | |
9349 | * ok to go ahead and try. | |
9350 | */ | |
9351 | int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr) | |
1a40e23b | 9352 | { |
ba1bf481 JB |
9353 | struct btrfs_block_group_cache *block_group; |
9354 | struct btrfs_space_info *space_info; | |
9355 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; | |
9356 | struct btrfs_device *device; | |
6df9a95e | 9357 | struct btrfs_trans_handle *trans; |
cdcb725c | 9358 | u64 min_free; |
6719db6a JB |
9359 | u64 dev_min = 1; |
9360 | u64 dev_nr = 0; | |
4a5e98f5 | 9361 | u64 target; |
cdcb725c | 9362 | int index; |
ba1bf481 JB |
9363 | int full = 0; |
9364 | int ret = 0; | |
1a40e23b | 9365 | |
ba1bf481 | 9366 | block_group = btrfs_lookup_block_group(root->fs_info, bytenr); |
1a40e23b | 9367 | |
ba1bf481 JB |
9368 | /* odd, couldn't find the block group, leave it alone */ |
9369 | if (!block_group) | |
9370 | return -1; | |
1a40e23b | 9371 | |
cdcb725c | 9372 | min_free = btrfs_block_group_used(&block_group->item); |
9373 | ||
ba1bf481 | 9374 | /* no bytes used, we're good */ |
cdcb725c | 9375 | if (!min_free) |
1a40e23b ZY |
9376 | goto out; |
9377 | ||
ba1bf481 JB |
9378 | space_info = block_group->space_info; |
9379 | spin_lock(&space_info->lock); | |
17d217fe | 9380 | |
ba1bf481 | 9381 | full = space_info->full; |
17d217fe | 9382 | |
ba1bf481 JB |
9383 | /* |
9384 | * if this is the last block group we have in this space, we can't | |
7ce618db CM |
9385 | * relocate it unless we're able to allocate a new chunk below. |
9386 | * | |
9387 | * Otherwise, we need to make sure we have room in the space to handle | |
9388 | * all of the extents from this block group. If we can, we're good | |
ba1bf481 | 9389 | */ |
7ce618db | 9390 | if ((space_info->total_bytes != block_group->key.offset) && |
cdcb725c | 9391 | (space_info->bytes_used + space_info->bytes_reserved + |
9392 | space_info->bytes_pinned + space_info->bytes_readonly + | |
9393 | min_free < space_info->total_bytes)) { | |
ba1bf481 JB |
9394 | spin_unlock(&space_info->lock); |
9395 | goto out; | |
17d217fe | 9396 | } |
ba1bf481 | 9397 | spin_unlock(&space_info->lock); |
ea8c2819 | 9398 | |
ba1bf481 JB |
9399 | /* |
9400 | * ok we don't have enough space, but maybe we have free space on our | |
9401 | * devices to allocate new chunks for relocation, so loop through our | |
4a5e98f5 ID |
9402 | * alloc devices and guess if we have enough space. if this block |
9403 | * group is going to be restriped, run checks against the target | |
9404 | * profile instead of the current one. | |
ba1bf481 JB |
9405 | */ |
9406 | ret = -1; | |
ea8c2819 | 9407 | |
cdcb725c | 9408 | /* |
9409 | * index: | |
9410 | * 0: raid10 | |
9411 | * 1: raid1 | |
9412 | * 2: dup | |
9413 | * 3: raid0 | |
9414 | * 4: single | |
9415 | */ | |
4a5e98f5 ID |
9416 | target = get_restripe_target(root->fs_info, block_group->flags); |
9417 | if (target) { | |
31e50229 | 9418 | index = __get_raid_index(extended_to_chunk(target)); |
4a5e98f5 ID |
9419 | } else { |
9420 | /* | |
9421 | * this is just a balance, so if we were marked as full | |
9422 | * we know there is no space for a new chunk | |
9423 | */ | |
9424 | if (full) | |
9425 | goto out; | |
9426 | ||
9427 | index = get_block_group_index(block_group); | |
9428 | } | |
9429 | ||
e6ec716f | 9430 | if (index == BTRFS_RAID_RAID10) { |
cdcb725c | 9431 | dev_min = 4; |
6719db6a JB |
9432 | /* Divide by 2 */ |
9433 | min_free >>= 1; | |
e6ec716f | 9434 | } else if (index == BTRFS_RAID_RAID1) { |
cdcb725c | 9435 | dev_min = 2; |
e6ec716f | 9436 | } else if (index == BTRFS_RAID_DUP) { |
6719db6a JB |
9437 | /* Multiply by 2 */ |
9438 | min_free <<= 1; | |
e6ec716f | 9439 | } else if (index == BTRFS_RAID_RAID0) { |
cdcb725c | 9440 | dev_min = fs_devices->rw_devices; |
47c5713f | 9441 | min_free = div64_u64(min_free, dev_min); |
cdcb725c | 9442 | } |
9443 | ||
6df9a95e JB |
9444 | /* We need to do this so that we can look at pending chunks */ |
9445 | trans = btrfs_join_transaction(root); | |
9446 | if (IS_ERR(trans)) { | |
9447 | ret = PTR_ERR(trans); | |
9448 | goto out; | |
9449 | } | |
9450 | ||
ba1bf481 JB |
9451 | mutex_lock(&root->fs_info->chunk_mutex); |
9452 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { | |
7bfc837d | 9453 | u64 dev_offset; |
56bec294 | 9454 | |
ba1bf481 JB |
9455 | /* |
9456 | * check to make sure we can actually find a chunk with enough | |
9457 | * space to fit our block group in. | |
9458 | */ | |
63a212ab SB |
9459 | if (device->total_bytes > device->bytes_used + min_free && |
9460 | !device->is_tgtdev_for_dev_replace) { | |
6df9a95e | 9461 | ret = find_free_dev_extent(trans, device, min_free, |
7bfc837d | 9462 | &dev_offset, NULL); |
ba1bf481 | 9463 | if (!ret) |
cdcb725c | 9464 | dev_nr++; |
9465 | ||
9466 | if (dev_nr >= dev_min) | |
73e48b27 | 9467 | break; |
cdcb725c | 9468 | |
ba1bf481 | 9469 | ret = -1; |
725c8463 | 9470 | } |
edbd8d4e | 9471 | } |
ba1bf481 | 9472 | mutex_unlock(&root->fs_info->chunk_mutex); |
6df9a95e | 9473 | btrfs_end_transaction(trans, root); |
edbd8d4e | 9474 | out: |
ba1bf481 | 9475 | btrfs_put_block_group(block_group); |
edbd8d4e CM |
9476 | return ret; |
9477 | } | |
9478 | ||
b2950863 CH |
9479 | static int find_first_block_group(struct btrfs_root *root, |
9480 | struct btrfs_path *path, struct btrfs_key *key) | |
0b86a832 | 9481 | { |
925baedd | 9482 | int ret = 0; |
0b86a832 CM |
9483 | struct btrfs_key found_key; |
9484 | struct extent_buffer *leaf; | |
9485 | int slot; | |
edbd8d4e | 9486 | |
0b86a832 CM |
9487 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); |
9488 | if (ret < 0) | |
925baedd CM |
9489 | goto out; |
9490 | ||
d397712b | 9491 | while (1) { |
0b86a832 | 9492 | slot = path->slots[0]; |
edbd8d4e | 9493 | leaf = path->nodes[0]; |
0b86a832 CM |
9494 | if (slot >= btrfs_header_nritems(leaf)) { |
9495 | ret = btrfs_next_leaf(root, path); | |
9496 | if (ret == 0) | |
9497 | continue; | |
9498 | if (ret < 0) | |
925baedd | 9499 | goto out; |
0b86a832 | 9500 | break; |
edbd8d4e | 9501 | } |
0b86a832 | 9502 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
edbd8d4e | 9503 | |
0b86a832 | 9504 | if (found_key.objectid >= key->objectid && |
925baedd CM |
9505 | found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) { |
9506 | ret = 0; | |
9507 | goto out; | |
9508 | } | |
0b86a832 | 9509 | path->slots[0]++; |
edbd8d4e | 9510 | } |
925baedd | 9511 | out: |
0b86a832 | 9512 | return ret; |
edbd8d4e CM |
9513 | } |
9514 | ||
0af3d00b JB |
9515 | void btrfs_put_block_group_cache(struct btrfs_fs_info *info) |
9516 | { | |
9517 | struct btrfs_block_group_cache *block_group; | |
9518 | u64 last = 0; | |
9519 | ||
9520 | while (1) { | |
9521 | struct inode *inode; | |
9522 | ||
9523 | block_group = btrfs_lookup_first_block_group(info, last); | |
9524 | while (block_group) { | |
9525 | spin_lock(&block_group->lock); | |
9526 | if (block_group->iref) | |
9527 | break; | |
9528 | spin_unlock(&block_group->lock); | |
9529 | block_group = next_block_group(info->tree_root, | |
9530 | block_group); | |
9531 | } | |
9532 | if (!block_group) { | |
9533 | if (last == 0) | |
9534 | break; | |
9535 | last = 0; | |
9536 | continue; | |
9537 | } | |
9538 | ||
9539 | inode = block_group->inode; | |
9540 | block_group->iref = 0; | |
9541 | block_group->inode = NULL; | |
9542 | spin_unlock(&block_group->lock); | |
9543 | iput(inode); | |
9544 | last = block_group->key.objectid + block_group->key.offset; | |
9545 | btrfs_put_block_group(block_group); | |
9546 | } | |
9547 | } | |
9548 | ||
1a40e23b ZY |
9549 | int btrfs_free_block_groups(struct btrfs_fs_info *info) |
9550 | { | |
9551 | struct btrfs_block_group_cache *block_group; | |
4184ea7f | 9552 | struct btrfs_space_info *space_info; |
11833d66 | 9553 | struct btrfs_caching_control *caching_ctl; |
1a40e23b ZY |
9554 | struct rb_node *n; |
9555 | ||
9e351cc8 | 9556 | down_write(&info->commit_root_sem); |
11833d66 YZ |
9557 | while (!list_empty(&info->caching_block_groups)) { |
9558 | caching_ctl = list_entry(info->caching_block_groups.next, | |
9559 | struct btrfs_caching_control, list); | |
9560 | list_del(&caching_ctl->list); | |
9561 | put_caching_control(caching_ctl); | |
9562 | } | |
9e351cc8 | 9563 | up_write(&info->commit_root_sem); |
11833d66 | 9564 | |
47ab2a6c JB |
9565 | spin_lock(&info->unused_bgs_lock); |
9566 | while (!list_empty(&info->unused_bgs)) { | |
9567 | block_group = list_first_entry(&info->unused_bgs, | |
9568 | struct btrfs_block_group_cache, | |
9569 | bg_list); | |
9570 | list_del_init(&block_group->bg_list); | |
9571 | btrfs_put_block_group(block_group); | |
9572 | } | |
9573 | spin_unlock(&info->unused_bgs_lock); | |
9574 | ||
1a40e23b ZY |
9575 | spin_lock(&info->block_group_cache_lock); |
9576 | while ((n = rb_last(&info->block_group_cache_tree)) != NULL) { | |
9577 | block_group = rb_entry(n, struct btrfs_block_group_cache, | |
9578 | cache_node); | |
1a40e23b ZY |
9579 | rb_erase(&block_group->cache_node, |
9580 | &info->block_group_cache_tree); | |
01eacb27 | 9581 | RB_CLEAR_NODE(&block_group->cache_node); |
d899e052 YZ |
9582 | spin_unlock(&info->block_group_cache_lock); |
9583 | ||
80eb234a | 9584 | down_write(&block_group->space_info->groups_sem); |
1a40e23b | 9585 | list_del(&block_group->list); |
80eb234a | 9586 | up_write(&block_group->space_info->groups_sem); |
d2fb3437 | 9587 | |
817d52f8 | 9588 | if (block_group->cached == BTRFS_CACHE_STARTED) |
11833d66 | 9589 | wait_block_group_cache_done(block_group); |
817d52f8 | 9590 | |
3c14874a JB |
9591 | /* |
9592 | * We haven't cached this block group, which means we could | |
9593 | * possibly have excluded extents on this block group. | |
9594 | */ | |
36cce922 JB |
9595 | if (block_group->cached == BTRFS_CACHE_NO || |
9596 | block_group->cached == BTRFS_CACHE_ERROR) | |
3c14874a JB |
9597 | free_excluded_extents(info->extent_root, block_group); |
9598 | ||
817d52f8 | 9599 | btrfs_remove_free_space_cache(block_group); |
11dfe35a | 9600 | btrfs_put_block_group(block_group); |
d899e052 YZ |
9601 | |
9602 | spin_lock(&info->block_group_cache_lock); | |
1a40e23b ZY |
9603 | } |
9604 | spin_unlock(&info->block_group_cache_lock); | |
4184ea7f CM |
9605 | |
9606 | /* now that all the block groups are freed, go through and | |
9607 | * free all the space_info structs. This is only called during | |
9608 | * the final stages of unmount, and so we know nobody is | |
9609 | * using them. We call synchronize_rcu() once before we start, | |
9610 | * just to be on the safe side. | |
9611 | */ | |
9612 | synchronize_rcu(); | |
9613 | ||
8929ecfa YZ |
9614 | release_global_block_rsv(info); |
9615 | ||
67871254 | 9616 | while (!list_empty(&info->space_info)) { |
6ab0a202 JM |
9617 | int i; |
9618 | ||
4184ea7f CM |
9619 | space_info = list_entry(info->space_info.next, |
9620 | struct btrfs_space_info, | |
9621 | list); | |
b069e0c3 | 9622 | if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) { |
fae7f21c | 9623 | if (WARN_ON(space_info->bytes_pinned > 0 || |
b069e0c3 | 9624 | space_info->bytes_reserved > 0 || |
fae7f21c | 9625 | space_info->bytes_may_use > 0)) { |
b069e0c3 DS |
9626 | dump_space_info(space_info, 0, 0); |
9627 | } | |
f0486c68 | 9628 | } |
4184ea7f | 9629 | list_del(&space_info->list); |
6ab0a202 JM |
9630 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { |
9631 | struct kobject *kobj; | |
c1895442 JM |
9632 | kobj = space_info->block_group_kobjs[i]; |
9633 | space_info->block_group_kobjs[i] = NULL; | |
9634 | if (kobj) { | |
6ab0a202 JM |
9635 | kobject_del(kobj); |
9636 | kobject_put(kobj); | |
9637 | } | |
9638 | } | |
9639 | kobject_del(&space_info->kobj); | |
9640 | kobject_put(&space_info->kobj); | |
4184ea7f | 9641 | } |
1a40e23b ZY |
9642 | return 0; |
9643 | } | |
9644 | ||
b742bb82 YZ |
9645 | static void __link_block_group(struct btrfs_space_info *space_info, |
9646 | struct btrfs_block_group_cache *cache) | |
9647 | { | |
9648 | int index = get_block_group_index(cache); | |
ed55b6ac | 9649 | bool first = false; |
b742bb82 YZ |
9650 | |
9651 | down_write(&space_info->groups_sem); | |
ed55b6ac JM |
9652 | if (list_empty(&space_info->block_groups[index])) |
9653 | first = true; | |
9654 | list_add_tail(&cache->list, &space_info->block_groups[index]); | |
9655 | up_write(&space_info->groups_sem); | |
9656 | ||
9657 | if (first) { | |
c1895442 | 9658 | struct raid_kobject *rkobj; |
6ab0a202 JM |
9659 | int ret; |
9660 | ||
c1895442 JM |
9661 | rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS); |
9662 | if (!rkobj) | |
9663 | goto out_err; | |
9664 | rkobj->raid_type = index; | |
9665 | kobject_init(&rkobj->kobj, &btrfs_raid_ktype); | |
9666 | ret = kobject_add(&rkobj->kobj, &space_info->kobj, | |
9667 | "%s", get_raid_name(index)); | |
6ab0a202 | 9668 | if (ret) { |
c1895442 JM |
9669 | kobject_put(&rkobj->kobj); |
9670 | goto out_err; | |
6ab0a202 | 9671 | } |
c1895442 | 9672 | space_info->block_group_kobjs[index] = &rkobj->kobj; |
6ab0a202 | 9673 | } |
c1895442 JM |
9674 | |
9675 | return; | |
9676 | out_err: | |
9677 | pr_warn("BTRFS: failed to add kobject for block cache. ignoring.\n"); | |
b742bb82 YZ |
9678 | } |
9679 | ||
920e4a58 MX |
9680 | static struct btrfs_block_group_cache * |
9681 | btrfs_create_block_group_cache(struct btrfs_root *root, u64 start, u64 size) | |
9682 | { | |
9683 | struct btrfs_block_group_cache *cache; | |
9684 | ||
9685 | cache = kzalloc(sizeof(*cache), GFP_NOFS); | |
9686 | if (!cache) | |
9687 | return NULL; | |
9688 | ||
9689 | cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl), | |
9690 | GFP_NOFS); | |
9691 | if (!cache->free_space_ctl) { | |
9692 | kfree(cache); | |
9693 | return NULL; | |
9694 | } | |
9695 | ||
9696 | cache->key.objectid = start; | |
9697 | cache->key.offset = size; | |
9698 | cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; | |
9699 | ||
9700 | cache->sectorsize = root->sectorsize; | |
9701 | cache->fs_info = root->fs_info; | |
9702 | cache->full_stripe_len = btrfs_full_stripe_len(root, | |
9703 | &root->fs_info->mapping_tree, | |
9704 | start); | |
1e144fb8 OS |
9705 | set_free_space_tree_thresholds(cache); |
9706 | ||
920e4a58 MX |
9707 | atomic_set(&cache->count, 1); |
9708 | spin_lock_init(&cache->lock); | |
e570fd27 | 9709 | init_rwsem(&cache->data_rwsem); |
920e4a58 MX |
9710 | INIT_LIST_HEAD(&cache->list); |
9711 | INIT_LIST_HEAD(&cache->cluster_list); | |
47ab2a6c | 9712 | INIT_LIST_HEAD(&cache->bg_list); |
633c0aad | 9713 | INIT_LIST_HEAD(&cache->ro_list); |
ce93ec54 | 9714 | INIT_LIST_HEAD(&cache->dirty_list); |
c9dc4c65 | 9715 | INIT_LIST_HEAD(&cache->io_list); |
920e4a58 | 9716 | btrfs_init_free_space_ctl(cache); |
04216820 | 9717 | atomic_set(&cache->trimming, 0); |
a5ed9182 | 9718 | mutex_init(&cache->free_space_lock); |
920e4a58 MX |
9719 | |
9720 | return cache; | |
9721 | } | |
9722 | ||
9078a3e1 CM |
9723 | int btrfs_read_block_groups(struct btrfs_root *root) |
9724 | { | |
9725 | struct btrfs_path *path; | |
9726 | int ret; | |
9078a3e1 | 9727 | struct btrfs_block_group_cache *cache; |
be744175 | 9728 | struct btrfs_fs_info *info = root->fs_info; |
6324fbf3 | 9729 | struct btrfs_space_info *space_info; |
9078a3e1 CM |
9730 | struct btrfs_key key; |
9731 | struct btrfs_key found_key; | |
5f39d397 | 9732 | struct extent_buffer *leaf; |
0af3d00b JB |
9733 | int need_clear = 0; |
9734 | u64 cache_gen; | |
96b5179d | 9735 | |
be744175 | 9736 | root = info->extent_root; |
9078a3e1 | 9737 | key.objectid = 0; |
0b86a832 | 9738 | key.offset = 0; |
962a298f | 9739 | key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; |
9078a3e1 CM |
9740 | path = btrfs_alloc_path(); |
9741 | if (!path) | |
9742 | return -ENOMEM; | |
026fd317 | 9743 | path->reada = 1; |
9078a3e1 | 9744 | |
6c41761f | 9745 | cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy); |
73bc1876 | 9746 | if (btrfs_test_opt(root, SPACE_CACHE) && |
6c41761f | 9747 | btrfs_super_generation(root->fs_info->super_copy) != cache_gen) |
0af3d00b | 9748 | need_clear = 1; |
88c2ba3b JB |
9749 | if (btrfs_test_opt(root, CLEAR_CACHE)) |
9750 | need_clear = 1; | |
0af3d00b | 9751 | |
d397712b | 9752 | while (1) { |
0b86a832 | 9753 | ret = find_first_block_group(root, path, &key); |
b742bb82 YZ |
9754 | if (ret > 0) |
9755 | break; | |
0b86a832 CM |
9756 | if (ret != 0) |
9757 | goto error; | |
920e4a58 | 9758 | |
5f39d397 CM |
9759 | leaf = path->nodes[0]; |
9760 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
920e4a58 MX |
9761 | |
9762 | cache = btrfs_create_block_group_cache(root, found_key.objectid, | |
9763 | found_key.offset); | |
9078a3e1 | 9764 | if (!cache) { |
0b86a832 | 9765 | ret = -ENOMEM; |
f0486c68 | 9766 | goto error; |
9078a3e1 | 9767 | } |
96303081 | 9768 | |
cf7c1ef6 LB |
9769 | if (need_clear) { |
9770 | /* | |
9771 | * When we mount with old space cache, we need to | |
9772 | * set BTRFS_DC_CLEAR and set dirty flag. | |
9773 | * | |
9774 | * a) Setting 'BTRFS_DC_CLEAR' makes sure that we | |
9775 | * truncate the old free space cache inode and | |
9776 | * setup a new one. | |
9777 | * b) Setting 'dirty flag' makes sure that we flush | |
9778 | * the new space cache info onto disk. | |
9779 | */ | |
cf7c1ef6 | 9780 | if (btrfs_test_opt(root, SPACE_CACHE)) |
ce93ec54 | 9781 | cache->disk_cache_state = BTRFS_DC_CLEAR; |
cf7c1ef6 | 9782 | } |
0af3d00b | 9783 | |
5f39d397 CM |
9784 | read_extent_buffer(leaf, &cache->item, |
9785 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
9786 | sizeof(cache->item)); | |
920e4a58 | 9787 | cache->flags = btrfs_block_group_flags(&cache->item); |
0b86a832 | 9788 | |
9078a3e1 | 9789 | key.objectid = found_key.objectid + found_key.offset; |
b3b4aa74 | 9790 | btrfs_release_path(path); |
34d52cb6 | 9791 | |
3c14874a JB |
9792 | /* |
9793 | * We need to exclude the super stripes now so that the space | |
9794 | * info has super bytes accounted for, otherwise we'll think | |
9795 | * we have more space than we actually do. | |
9796 | */ | |
835d974f JB |
9797 | ret = exclude_super_stripes(root, cache); |
9798 | if (ret) { | |
9799 | /* | |
9800 | * We may have excluded something, so call this just in | |
9801 | * case. | |
9802 | */ | |
9803 | free_excluded_extents(root, cache); | |
920e4a58 | 9804 | btrfs_put_block_group(cache); |
835d974f JB |
9805 | goto error; |
9806 | } | |
3c14874a | 9807 | |
817d52f8 JB |
9808 | /* |
9809 | * check for two cases, either we are full, and therefore | |
9810 | * don't need to bother with the caching work since we won't | |
9811 | * find any space, or we are empty, and we can just add all | |
9812 | * the space in and be done with it. This saves us _alot_ of | |
9813 | * time, particularly in the full case. | |
9814 | */ | |
9815 | if (found_key.offset == btrfs_block_group_used(&cache->item)) { | |
11833d66 | 9816 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 9817 | cache->cached = BTRFS_CACHE_FINISHED; |
1b2da372 | 9818 | free_excluded_extents(root, cache); |
817d52f8 | 9819 | } else if (btrfs_block_group_used(&cache->item) == 0) { |
11833d66 | 9820 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 JB |
9821 | cache->cached = BTRFS_CACHE_FINISHED; |
9822 | add_new_free_space(cache, root->fs_info, | |
9823 | found_key.objectid, | |
9824 | found_key.objectid + | |
9825 | found_key.offset); | |
11833d66 | 9826 | free_excluded_extents(root, cache); |
817d52f8 | 9827 | } |
96b5179d | 9828 | |
8c579fe7 JB |
9829 | ret = btrfs_add_block_group_cache(root->fs_info, cache); |
9830 | if (ret) { | |
9831 | btrfs_remove_free_space_cache(cache); | |
9832 | btrfs_put_block_group(cache); | |
9833 | goto error; | |
9834 | } | |
9835 | ||
6324fbf3 CM |
9836 | ret = update_space_info(info, cache->flags, found_key.offset, |
9837 | btrfs_block_group_used(&cache->item), | |
9838 | &space_info); | |
8c579fe7 JB |
9839 | if (ret) { |
9840 | btrfs_remove_free_space_cache(cache); | |
9841 | spin_lock(&info->block_group_cache_lock); | |
9842 | rb_erase(&cache->cache_node, | |
9843 | &info->block_group_cache_tree); | |
01eacb27 | 9844 | RB_CLEAR_NODE(&cache->cache_node); |
8c579fe7 JB |
9845 | spin_unlock(&info->block_group_cache_lock); |
9846 | btrfs_put_block_group(cache); | |
9847 | goto error; | |
9848 | } | |
9849 | ||
6324fbf3 | 9850 | cache->space_info = space_info; |
1b2da372 | 9851 | spin_lock(&cache->space_info->lock); |
f0486c68 | 9852 | cache->space_info->bytes_readonly += cache->bytes_super; |
1b2da372 JB |
9853 | spin_unlock(&cache->space_info->lock); |
9854 | ||
b742bb82 | 9855 | __link_block_group(space_info, cache); |
0f9dd46c | 9856 | |
75ccf47d | 9857 | set_avail_alloc_bits(root->fs_info, cache->flags); |
47ab2a6c | 9858 | if (btrfs_chunk_readonly(root, cache->key.objectid)) { |
868f401a | 9859 | inc_block_group_ro(cache, 1); |
47ab2a6c JB |
9860 | } else if (btrfs_block_group_used(&cache->item) == 0) { |
9861 | spin_lock(&info->unused_bgs_lock); | |
9862 | /* Should always be true but just in case. */ | |
9863 | if (list_empty(&cache->bg_list)) { | |
9864 | btrfs_get_block_group(cache); | |
9865 | list_add_tail(&cache->bg_list, | |
9866 | &info->unused_bgs); | |
9867 | } | |
9868 | spin_unlock(&info->unused_bgs_lock); | |
9869 | } | |
9078a3e1 | 9870 | } |
b742bb82 YZ |
9871 | |
9872 | list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) { | |
9873 | if (!(get_alloc_profile(root, space_info->flags) & | |
9874 | (BTRFS_BLOCK_GROUP_RAID10 | | |
9875 | BTRFS_BLOCK_GROUP_RAID1 | | |
53b381b3 DW |
9876 | BTRFS_BLOCK_GROUP_RAID5 | |
9877 | BTRFS_BLOCK_GROUP_RAID6 | | |
b742bb82 YZ |
9878 | BTRFS_BLOCK_GROUP_DUP))) |
9879 | continue; | |
9880 | /* | |
9881 | * avoid allocating from un-mirrored block group if there are | |
9882 | * mirrored block groups. | |
9883 | */ | |
1095cc0d | 9884 | list_for_each_entry(cache, |
9885 | &space_info->block_groups[BTRFS_RAID_RAID0], | |
9886 | list) | |
868f401a | 9887 | inc_block_group_ro(cache, 1); |
1095cc0d | 9888 | list_for_each_entry(cache, |
9889 | &space_info->block_groups[BTRFS_RAID_SINGLE], | |
9890 | list) | |
868f401a | 9891 | inc_block_group_ro(cache, 1); |
9078a3e1 | 9892 | } |
f0486c68 YZ |
9893 | |
9894 | init_global_block_rsv(info); | |
0b86a832 CM |
9895 | ret = 0; |
9896 | error: | |
9078a3e1 | 9897 | btrfs_free_path(path); |
0b86a832 | 9898 | return ret; |
9078a3e1 | 9899 | } |
6324fbf3 | 9900 | |
ea658bad JB |
9901 | void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans, |
9902 | struct btrfs_root *root) | |
9903 | { | |
9904 | struct btrfs_block_group_cache *block_group, *tmp; | |
9905 | struct btrfs_root *extent_root = root->fs_info->extent_root; | |
9906 | struct btrfs_block_group_item item; | |
9907 | struct btrfs_key key; | |
9908 | int ret = 0; | |
d9a0540a | 9909 | bool can_flush_pending_bgs = trans->can_flush_pending_bgs; |
ea658bad | 9910 | |
d9a0540a | 9911 | trans->can_flush_pending_bgs = false; |
47ab2a6c | 9912 | list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) { |
ea658bad | 9913 | if (ret) |
c92f6be3 | 9914 | goto next; |
ea658bad JB |
9915 | |
9916 | spin_lock(&block_group->lock); | |
9917 | memcpy(&item, &block_group->item, sizeof(item)); | |
9918 | memcpy(&key, &block_group->key, sizeof(key)); | |
9919 | spin_unlock(&block_group->lock); | |
9920 | ||
9921 | ret = btrfs_insert_item(trans, extent_root, &key, &item, | |
9922 | sizeof(item)); | |
9923 | if (ret) | |
9924 | btrfs_abort_transaction(trans, extent_root, ret); | |
6df9a95e JB |
9925 | ret = btrfs_finish_chunk_alloc(trans, extent_root, |
9926 | key.objectid, key.offset); | |
9927 | if (ret) | |
9928 | btrfs_abort_transaction(trans, extent_root, ret); | |
1e144fb8 OS |
9929 | add_block_group_free_space(trans, root->fs_info, block_group); |
9930 | /* already aborted the transaction if it failed. */ | |
c92f6be3 FM |
9931 | next: |
9932 | list_del_init(&block_group->bg_list); | |
ea658bad | 9933 | } |
d9a0540a | 9934 | trans->can_flush_pending_bgs = can_flush_pending_bgs; |
ea658bad JB |
9935 | } |
9936 | ||
6324fbf3 CM |
9937 | int btrfs_make_block_group(struct btrfs_trans_handle *trans, |
9938 | struct btrfs_root *root, u64 bytes_used, | |
e17cade2 | 9939 | u64 type, u64 chunk_objectid, u64 chunk_offset, |
6324fbf3 CM |
9940 | u64 size) |
9941 | { | |
9942 | int ret; | |
6324fbf3 CM |
9943 | struct btrfs_root *extent_root; |
9944 | struct btrfs_block_group_cache *cache; | |
6324fbf3 CM |
9945 | |
9946 | extent_root = root->fs_info->extent_root; | |
6324fbf3 | 9947 | |
995946dd | 9948 | btrfs_set_log_full_commit(root->fs_info, trans); |
e02119d5 | 9949 | |
920e4a58 | 9950 | cache = btrfs_create_block_group_cache(root, chunk_offset, size); |
0f9dd46c JB |
9951 | if (!cache) |
9952 | return -ENOMEM; | |
34d52cb6 | 9953 | |
6324fbf3 | 9954 | btrfs_set_block_group_used(&cache->item, bytes_used); |
6324fbf3 | 9955 | btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid); |
6324fbf3 CM |
9956 | btrfs_set_block_group_flags(&cache->item, type); |
9957 | ||
920e4a58 | 9958 | cache->flags = type; |
11833d66 | 9959 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 9960 | cache->cached = BTRFS_CACHE_FINISHED; |
1e144fb8 | 9961 | cache->needs_free_space = 1; |
835d974f JB |
9962 | ret = exclude_super_stripes(root, cache); |
9963 | if (ret) { | |
9964 | /* | |
9965 | * We may have excluded something, so call this just in | |
9966 | * case. | |
9967 | */ | |
9968 | free_excluded_extents(root, cache); | |
920e4a58 | 9969 | btrfs_put_block_group(cache); |
835d974f JB |
9970 | return ret; |
9971 | } | |
96303081 | 9972 | |
817d52f8 JB |
9973 | add_new_free_space(cache, root->fs_info, chunk_offset, |
9974 | chunk_offset + size); | |
9975 | ||
11833d66 YZ |
9976 | free_excluded_extents(root, cache); |
9977 | ||
d0bd4560 JB |
9978 | #ifdef CONFIG_BTRFS_DEBUG |
9979 | if (btrfs_should_fragment_free_space(root, cache)) { | |
9980 | u64 new_bytes_used = size - bytes_used; | |
9981 | ||
9982 | bytes_used += new_bytes_used >> 1; | |
9983 | fragment_free_space(root, cache); | |
9984 | } | |
9985 | #endif | |
2e6e5183 FM |
9986 | /* |
9987 | * Call to ensure the corresponding space_info object is created and | |
9988 | * assigned to our block group, but don't update its counters just yet. | |
9989 | * We want our bg to be added to the rbtree with its ->space_info set. | |
9990 | */ | |
9991 | ret = update_space_info(root->fs_info, cache->flags, 0, 0, | |
9992 | &cache->space_info); | |
9993 | if (ret) { | |
9994 | btrfs_remove_free_space_cache(cache); | |
9995 | btrfs_put_block_group(cache); | |
9996 | return ret; | |
9997 | } | |
9998 | ||
8c579fe7 JB |
9999 | ret = btrfs_add_block_group_cache(root->fs_info, cache); |
10000 | if (ret) { | |
10001 | btrfs_remove_free_space_cache(cache); | |
10002 | btrfs_put_block_group(cache); | |
10003 | return ret; | |
10004 | } | |
10005 | ||
2e6e5183 FM |
10006 | /* |
10007 | * Now that our block group has its ->space_info set and is inserted in | |
10008 | * the rbtree, update the space info's counters. | |
10009 | */ | |
6324fbf3 CM |
10010 | ret = update_space_info(root->fs_info, cache->flags, size, bytes_used, |
10011 | &cache->space_info); | |
8c579fe7 JB |
10012 | if (ret) { |
10013 | btrfs_remove_free_space_cache(cache); | |
10014 | spin_lock(&root->fs_info->block_group_cache_lock); | |
10015 | rb_erase(&cache->cache_node, | |
10016 | &root->fs_info->block_group_cache_tree); | |
01eacb27 | 10017 | RB_CLEAR_NODE(&cache->cache_node); |
8c579fe7 JB |
10018 | spin_unlock(&root->fs_info->block_group_cache_lock); |
10019 | btrfs_put_block_group(cache); | |
10020 | return ret; | |
10021 | } | |
c7c144db | 10022 | update_global_block_rsv(root->fs_info); |
1b2da372 JB |
10023 | |
10024 | spin_lock(&cache->space_info->lock); | |
f0486c68 | 10025 | cache->space_info->bytes_readonly += cache->bytes_super; |
1b2da372 JB |
10026 | spin_unlock(&cache->space_info->lock); |
10027 | ||
b742bb82 | 10028 | __link_block_group(cache->space_info, cache); |
6324fbf3 | 10029 | |
47ab2a6c | 10030 | list_add_tail(&cache->bg_list, &trans->new_bgs); |
6324fbf3 | 10031 | |
d18a2c44 | 10032 | set_avail_alloc_bits(extent_root->fs_info, type); |
925baedd | 10033 | |
6324fbf3 CM |
10034 | return 0; |
10035 | } | |
1a40e23b | 10036 | |
10ea00f5 ID |
10037 | static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
10038 | { | |
899c81ea ID |
10039 | u64 extra_flags = chunk_to_extended(flags) & |
10040 | BTRFS_EXTENDED_PROFILE_MASK; | |
10ea00f5 | 10041 | |
de98ced9 | 10042 | write_seqlock(&fs_info->profiles_lock); |
10ea00f5 ID |
10043 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
10044 | fs_info->avail_data_alloc_bits &= ~extra_flags; | |
10045 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
10046 | fs_info->avail_metadata_alloc_bits &= ~extra_flags; | |
10047 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
10048 | fs_info->avail_system_alloc_bits &= ~extra_flags; | |
de98ced9 | 10049 | write_sequnlock(&fs_info->profiles_lock); |
10ea00f5 ID |
10050 | } |
10051 | ||
1a40e23b | 10052 | int btrfs_remove_block_group(struct btrfs_trans_handle *trans, |
04216820 FM |
10053 | struct btrfs_root *root, u64 group_start, |
10054 | struct extent_map *em) | |
1a40e23b ZY |
10055 | { |
10056 | struct btrfs_path *path; | |
10057 | struct btrfs_block_group_cache *block_group; | |
44fb5511 | 10058 | struct btrfs_free_cluster *cluster; |
0af3d00b | 10059 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
1a40e23b | 10060 | struct btrfs_key key; |
0af3d00b | 10061 | struct inode *inode; |
c1895442 | 10062 | struct kobject *kobj = NULL; |
1a40e23b | 10063 | int ret; |
10ea00f5 | 10064 | int index; |
89a55897 | 10065 | int factor; |
4f69cb98 | 10066 | struct btrfs_caching_control *caching_ctl = NULL; |
04216820 | 10067 | bool remove_em; |
1a40e23b | 10068 | |
1a40e23b ZY |
10069 | root = root->fs_info->extent_root; |
10070 | ||
10071 | block_group = btrfs_lookup_block_group(root->fs_info, group_start); | |
10072 | BUG_ON(!block_group); | |
c146afad | 10073 | BUG_ON(!block_group->ro); |
1a40e23b | 10074 | |
9f7c43c9 | 10075 | /* |
10076 | * Free the reserved super bytes from this block group before | |
10077 | * remove it. | |
10078 | */ | |
10079 | free_excluded_extents(root, block_group); | |
10080 | ||
1a40e23b | 10081 | memcpy(&key, &block_group->key, sizeof(key)); |
10ea00f5 | 10082 | index = get_block_group_index(block_group); |
89a55897 JB |
10083 | if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP | |
10084 | BTRFS_BLOCK_GROUP_RAID1 | | |
10085 | BTRFS_BLOCK_GROUP_RAID10)) | |
10086 | factor = 2; | |
10087 | else | |
10088 | factor = 1; | |
1a40e23b | 10089 | |
44fb5511 CM |
10090 | /* make sure this block group isn't part of an allocation cluster */ |
10091 | cluster = &root->fs_info->data_alloc_cluster; | |
10092 | spin_lock(&cluster->refill_lock); | |
10093 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
10094 | spin_unlock(&cluster->refill_lock); | |
10095 | ||
10096 | /* | |
10097 | * make sure this block group isn't part of a metadata | |
10098 | * allocation cluster | |
10099 | */ | |
10100 | cluster = &root->fs_info->meta_alloc_cluster; | |
10101 | spin_lock(&cluster->refill_lock); | |
10102 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
10103 | spin_unlock(&cluster->refill_lock); | |
10104 | ||
1a40e23b | 10105 | path = btrfs_alloc_path(); |
d8926bb3 MF |
10106 | if (!path) { |
10107 | ret = -ENOMEM; | |
10108 | goto out; | |
10109 | } | |
1a40e23b | 10110 | |
1bbc621e CM |
10111 | /* |
10112 | * get the inode first so any iput calls done for the io_list | |
10113 | * aren't the final iput (no unlinks allowed now) | |
10114 | */ | |
10b2f34d | 10115 | inode = lookup_free_space_inode(tree_root, block_group, path); |
1bbc621e CM |
10116 | |
10117 | mutex_lock(&trans->transaction->cache_write_mutex); | |
10118 | /* | |
10119 | * make sure our free spache cache IO is done before remove the | |
10120 | * free space inode | |
10121 | */ | |
10122 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
10123 | if (!list_empty(&block_group->io_list)) { | |
10124 | list_del_init(&block_group->io_list); | |
10125 | ||
10126 | WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode); | |
10127 | ||
10128 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
10129 | btrfs_wait_cache_io(root, trans, block_group, | |
10130 | &block_group->io_ctl, path, | |
10131 | block_group->key.objectid); | |
10132 | btrfs_put_block_group(block_group); | |
10133 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
10134 | } | |
10135 | ||
10136 | if (!list_empty(&block_group->dirty_list)) { | |
10137 | list_del_init(&block_group->dirty_list); | |
10138 | btrfs_put_block_group(block_group); | |
10139 | } | |
10140 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
10141 | mutex_unlock(&trans->transaction->cache_write_mutex); | |
10142 | ||
0af3d00b | 10143 | if (!IS_ERR(inode)) { |
b532402e | 10144 | ret = btrfs_orphan_add(trans, inode); |
79787eaa JM |
10145 | if (ret) { |
10146 | btrfs_add_delayed_iput(inode); | |
10147 | goto out; | |
10148 | } | |
0af3d00b JB |
10149 | clear_nlink(inode); |
10150 | /* One for the block groups ref */ | |
10151 | spin_lock(&block_group->lock); | |
10152 | if (block_group->iref) { | |
10153 | block_group->iref = 0; | |
10154 | block_group->inode = NULL; | |
10155 | spin_unlock(&block_group->lock); | |
10156 | iput(inode); | |
10157 | } else { | |
10158 | spin_unlock(&block_group->lock); | |
10159 | } | |
10160 | /* One for our lookup ref */ | |
455757c3 | 10161 | btrfs_add_delayed_iput(inode); |
0af3d00b JB |
10162 | } |
10163 | ||
10164 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
10165 | key.offset = block_group->key.objectid; | |
10166 | key.type = 0; | |
10167 | ||
10168 | ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1); | |
10169 | if (ret < 0) | |
10170 | goto out; | |
10171 | if (ret > 0) | |
b3b4aa74 | 10172 | btrfs_release_path(path); |
0af3d00b JB |
10173 | if (ret == 0) { |
10174 | ret = btrfs_del_item(trans, tree_root, path); | |
10175 | if (ret) | |
10176 | goto out; | |
b3b4aa74 | 10177 | btrfs_release_path(path); |
0af3d00b JB |
10178 | } |
10179 | ||
3dfdb934 | 10180 | spin_lock(&root->fs_info->block_group_cache_lock); |
1a40e23b ZY |
10181 | rb_erase(&block_group->cache_node, |
10182 | &root->fs_info->block_group_cache_tree); | |
292cbd51 | 10183 | RB_CLEAR_NODE(&block_group->cache_node); |
a1897fdd LB |
10184 | |
10185 | if (root->fs_info->first_logical_byte == block_group->key.objectid) | |
10186 | root->fs_info->first_logical_byte = (u64)-1; | |
3dfdb934 | 10187 | spin_unlock(&root->fs_info->block_group_cache_lock); |
817d52f8 | 10188 | |
80eb234a | 10189 | down_write(&block_group->space_info->groups_sem); |
44fb5511 CM |
10190 | /* |
10191 | * we must use list_del_init so people can check to see if they | |
10192 | * are still on the list after taking the semaphore | |
10193 | */ | |
10194 | list_del_init(&block_group->list); | |
6ab0a202 | 10195 | if (list_empty(&block_group->space_info->block_groups[index])) { |
c1895442 JM |
10196 | kobj = block_group->space_info->block_group_kobjs[index]; |
10197 | block_group->space_info->block_group_kobjs[index] = NULL; | |
10ea00f5 | 10198 | clear_avail_alloc_bits(root->fs_info, block_group->flags); |
6ab0a202 | 10199 | } |
80eb234a | 10200 | up_write(&block_group->space_info->groups_sem); |
c1895442 JM |
10201 | if (kobj) { |
10202 | kobject_del(kobj); | |
10203 | kobject_put(kobj); | |
10204 | } | |
1a40e23b | 10205 | |
4f69cb98 FM |
10206 | if (block_group->has_caching_ctl) |
10207 | caching_ctl = get_caching_control(block_group); | |
817d52f8 | 10208 | if (block_group->cached == BTRFS_CACHE_STARTED) |
11833d66 | 10209 | wait_block_group_cache_done(block_group); |
4f69cb98 FM |
10210 | if (block_group->has_caching_ctl) { |
10211 | down_write(&root->fs_info->commit_root_sem); | |
10212 | if (!caching_ctl) { | |
10213 | struct btrfs_caching_control *ctl; | |
10214 | ||
10215 | list_for_each_entry(ctl, | |
10216 | &root->fs_info->caching_block_groups, list) | |
10217 | if (ctl->block_group == block_group) { | |
10218 | caching_ctl = ctl; | |
10219 | atomic_inc(&caching_ctl->count); | |
10220 | break; | |
10221 | } | |
10222 | } | |
10223 | if (caching_ctl) | |
10224 | list_del_init(&caching_ctl->list); | |
10225 | up_write(&root->fs_info->commit_root_sem); | |
10226 | if (caching_ctl) { | |
10227 | /* Once for the caching bgs list and once for us. */ | |
10228 | put_caching_control(caching_ctl); | |
10229 | put_caching_control(caching_ctl); | |
10230 | } | |
10231 | } | |
817d52f8 | 10232 | |
ce93ec54 JB |
10233 | spin_lock(&trans->transaction->dirty_bgs_lock); |
10234 | if (!list_empty(&block_group->dirty_list)) { | |
1bbc621e CM |
10235 | WARN_ON(1); |
10236 | } | |
10237 | if (!list_empty(&block_group->io_list)) { | |
10238 | WARN_ON(1); | |
ce93ec54 JB |
10239 | } |
10240 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
817d52f8 JB |
10241 | btrfs_remove_free_space_cache(block_group); |
10242 | ||
c146afad | 10243 | spin_lock(&block_group->space_info->lock); |
75c68e9f | 10244 | list_del_init(&block_group->ro_list); |
18d018ad ZL |
10245 | |
10246 | if (btrfs_test_opt(root, ENOSPC_DEBUG)) { | |
10247 | WARN_ON(block_group->space_info->total_bytes | |
10248 | < block_group->key.offset); | |
10249 | WARN_ON(block_group->space_info->bytes_readonly | |
10250 | < block_group->key.offset); | |
10251 | WARN_ON(block_group->space_info->disk_total | |
10252 | < block_group->key.offset * factor); | |
10253 | } | |
c146afad YZ |
10254 | block_group->space_info->total_bytes -= block_group->key.offset; |
10255 | block_group->space_info->bytes_readonly -= block_group->key.offset; | |
89a55897 | 10256 | block_group->space_info->disk_total -= block_group->key.offset * factor; |
18d018ad | 10257 | |
c146afad | 10258 | spin_unlock(&block_group->space_info->lock); |
283bb197 | 10259 | |
0af3d00b JB |
10260 | memcpy(&key, &block_group->key, sizeof(key)); |
10261 | ||
04216820 | 10262 | lock_chunks(root); |
495e64f4 FM |
10263 | if (!list_empty(&em->list)) { |
10264 | /* We're in the transaction->pending_chunks list. */ | |
10265 | free_extent_map(em); | |
10266 | } | |
04216820 FM |
10267 | spin_lock(&block_group->lock); |
10268 | block_group->removed = 1; | |
10269 | /* | |
10270 | * At this point trimming can't start on this block group, because we | |
10271 | * removed the block group from the tree fs_info->block_group_cache_tree | |
10272 | * so no one can't find it anymore and even if someone already got this | |
10273 | * block group before we removed it from the rbtree, they have already | |
10274 | * incremented block_group->trimming - if they didn't, they won't find | |
10275 | * any free space entries because we already removed them all when we | |
10276 | * called btrfs_remove_free_space_cache(). | |
10277 | * | |
10278 | * And we must not remove the extent map from the fs_info->mapping_tree | |
10279 | * to prevent the same logical address range and physical device space | |
10280 | * ranges from being reused for a new block group. This is because our | |
10281 | * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is | |
10282 | * completely transactionless, so while it is trimming a range the | |
10283 | * currently running transaction might finish and a new one start, | |
10284 | * allowing for new block groups to be created that can reuse the same | |
10285 | * physical device locations unless we take this special care. | |
e33e17ee JM |
10286 | * |
10287 | * There may also be an implicit trim operation if the file system | |
10288 | * is mounted with -odiscard. The same protections must remain | |
10289 | * in place until the extents have been discarded completely when | |
10290 | * the transaction commit has completed. | |
04216820 FM |
10291 | */ |
10292 | remove_em = (atomic_read(&block_group->trimming) == 0); | |
10293 | /* | |
10294 | * Make sure a trimmer task always sees the em in the pinned_chunks list | |
10295 | * if it sees block_group->removed == 1 (needs to lock block_group->lock | |
10296 | * before checking block_group->removed). | |
10297 | */ | |
10298 | if (!remove_em) { | |
10299 | /* | |
10300 | * Our em might be in trans->transaction->pending_chunks which | |
10301 | * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks), | |
10302 | * and so is the fs_info->pinned_chunks list. | |
10303 | * | |
10304 | * So at this point we must be holding the chunk_mutex to avoid | |
10305 | * any races with chunk allocation (more specifically at | |
10306 | * volumes.c:contains_pending_extent()), to ensure it always | |
10307 | * sees the em, either in the pending_chunks list or in the | |
10308 | * pinned_chunks list. | |
10309 | */ | |
10310 | list_move_tail(&em->list, &root->fs_info->pinned_chunks); | |
10311 | } | |
10312 | spin_unlock(&block_group->lock); | |
04216820 FM |
10313 | |
10314 | if (remove_em) { | |
10315 | struct extent_map_tree *em_tree; | |
10316 | ||
10317 | em_tree = &root->fs_info->mapping_tree.map_tree; | |
10318 | write_lock(&em_tree->lock); | |
8dbcd10f FM |
10319 | /* |
10320 | * The em might be in the pending_chunks list, so make sure the | |
10321 | * chunk mutex is locked, since remove_extent_mapping() will | |
10322 | * delete us from that list. | |
10323 | */ | |
04216820 FM |
10324 | remove_extent_mapping(em_tree, em); |
10325 | write_unlock(&em_tree->lock); | |
10326 | /* once for the tree */ | |
10327 | free_extent_map(em); | |
10328 | } | |
10329 | ||
8dbcd10f FM |
10330 | unlock_chunks(root); |
10331 | ||
1e144fb8 OS |
10332 | ret = remove_block_group_free_space(trans, root->fs_info, block_group); |
10333 | if (ret) | |
10334 | goto out; | |
10335 | ||
fa9c0d79 CM |
10336 | btrfs_put_block_group(block_group); |
10337 | btrfs_put_block_group(block_group); | |
1a40e23b ZY |
10338 | |
10339 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
10340 | if (ret > 0) | |
10341 | ret = -EIO; | |
10342 | if (ret < 0) | |
10343 | goto out; | |
10344 | ||
10345 | ret = btrfs_del_item(trans, root, path); | |
10346 | out: | |
10347 | btrfs_free_path(path); | |
10348 | return ret; | |
10349 | } | |
acce952b | 10350 | |
8eab77ff | 10351 | struct btrfs_trans_handle * |
7fd01182 FM |
10352 | btrfs_start_trans_remove_block_group(struct btrfs_fs_info *fs_info, |
10353 | const u64 chunk_offset) | |
8eab77ff | 10354 | { |
7fd01182 FM |
10355 | struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree; |
10356 | struct extent_map *em; | |
10357 | struct map_lookup *map; | |
10358 | unsigned int num_items; | |
10359 | ||
10360 | read_lock(&em_tree->lock); | |
10361 | em = lookup_extent_mapping(em_tree, chunk_offset, 1); | |
10362 | read_unlock(&em_tree->lock); | |
10363 | ASSERT(em && em->start == chunk_offset); | |
10364 | ||
8eab77ff | 10365 | /* |
7fd01182 FM |
10366 | * We need to reserve 3 + N units from the metadata space info in order |
10367 | * to remove a block group (done at btrfs_remove_chunk() and at | |
10368 | * btrfs_remove_block_group()), which are used for: | |
10369 | * | |
8eab77ff FM |
10370 | * 1 unit for adding the free space inode's orphan (located in the tree |
10371 | * of tree roots). | |
7fd01182 FM |
10372 | * 1 unit for deleting the block group item (located in the extent |
10373 | * tree). | |
10374 | * 1 unit for deleting the free space item (located in tree of tree | |
10375 | * roots). | |
10376 | * N units for deleting N device extent items corresponding to each | |
10377 | * stripe (located in the device tree). | |
10378 | * | |
10379 | * In order to remove a block group we also need to reserve units in the | |
10380 | * system space info in order to update the chunk tree (update one or | |
10381 | * more device items and remove one chunk item), but this is done at | |
10382 | * btrfs_remove_chunk() through a call to check_system_chunk(). | |
8eab77ff | 10383 | */ |
7fd01182 FM |
10384 | map = (struct map_lookup *)em->bdev; |
10385 | num_items = 3 + map->num_stripes; | |
10386 | free_extent_map(em); | |
10387 | ||
8eab77ff | 10388 | return btrfs_start_transaction_fallback_global_rsv(fs_info->extent_root, |
7fd01182 | 10389 | num_items, 1); |
8eab77ff FM |
10390 | } |
10391 | ||
47ab2a6c JB |
10392 | /* |
10393 | * Process the unused_bgs list and remove any that don't have any allocated | |
10394 | * space inside of them. | |
10395 | */ | |
10396 | void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info) | |
10397 | { | |
10398 | struct btrfs_block_group_cache *block_group; | |
10399 | struct btrfs_space_info *space_info; | |
10400 | struct btrfs_root *root = fs_info->extent_root; | |
10401 | struct btrfs_trans_handle *trans; | |
10402 | int ret = 0; | |
10403 | ||
10404 | if (!fs_info->open) | |
10405 | return; | |
10406 | ||
10407 | spin_lock(&fs_info->unused_bgs_lock); | |
10408 | while (!list_empty(&fs_info->unused_bgs)) { | |
10409 | u64 start, end; | |
e33e17ee | 10410 | int trimming; |
47ab2a6c JB |
10411 | |
10412 | block_group = list_first_entry(&fs_info->unused_bgs, | |
10413 | struct btrfs_block_group_cache, | |
10414 | bg_list); | |
47ab2a6c | 10415 | list_del_init(&block_group->bg_list); |
aefbe9a6 ZL |
10416 | |
10417 | space_info = block_group->space_info; | |
10418 | ||
47ab2a6c JB |
10419 | if (ret || btrfs_mixed_space_info(space_info)) { |
10420 | btrfs_put_block_group(block_group); | |
10421 | continue; | |
10422 | } | |
10423 | spin_unlock(&fs_info->unused_bgs_lock); | |
10424 | ||
d5f2e33b | 10425 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
67c5e7d4 | 10426 | |
47ab2a6c JB |
10427 | /* Don't want to race with allocators so take the groups_sem */ |
10428 | down_write(&space_info->groups_sem); | |
10429 | spin_lock(&block_group->lock); | |
10430 | if (block_group->reserved || | |
10431 | btrfs_block_group_used(&block_group->item) || | |
aefbe9a6 ZL |
10432 | block_group->ro || |
10433 | list_is_singular(&block_group->list)) { | |
47ab2a6c JB |
10434 | /* |
10435 | * We want to bail if we made new allocations or have | |
10436 | * outstanding allocations in this block group. We do | |
10437 | * the ro check in case balance is currently acting on | |
10438 | * this block group. | |
10439 | */ | |
10440 | spin_unlock(&block_group->lock); | |
10441 | up_write(&space_info->groups_sem); | |
10442 | goto next; | |
10443 | } | |
10444 | spin_unlock(&block_group->lock); | |
10445 | ||
10446 | /* We don't want to force the issue, only flip if it's ok. */ | |
868f401a | 10447 | ret = inc_block_group_ro(block_group, 0); |
47ab2a6c JB |
10448 | up_write(&space_info->groups_sem); |
10449 | if (ret < 0) { | |
10450 | ret = 0; | |
10451 | goto next; | |
10452 | } | |
10453 | ||
10454 | /* | |
10455 | * Want to do this before we do anything else so we can recover | |
10456 | * properly if we fail to join the transaction. | |
10457 | */ | |
7fd01182 FM |
10458 | trans = btrfs_start_trans_remove_block_group(fs_info, |
10459 | block_group->key.objectid); | |
47ab2a6c | 10460 | if (IS_ERR(trans)) { |
868f401a | 10461 | btrfs_dec_block_group_ro(root, block_group); |
47ab2a6c JB |
10462 | ret = PTR_ERR(trans); |
10463 | goto next; | |
10464 | } | |
10465 | ||
10466 | /* | |
10467 | * We could have pending pinned extents for this block group, | |
10468 | * just delete them, we don't care about them anymore. | |
10469 | */ | |
10470 | start = block_group->key.objectid; | |
10471 | end = start + block_group->key.offset - 1; | |
d4b450cd FM |
10472 | /* |
10473 | * Hold the unused_bg_unpin_mutex lock to avoid racing with | |
10474 | * btrfs_finish_extent_commit(). If we are at transaction N, | |
10475 | * another task might be running finish_extent_commit() for the | |
10476 | * previous transaction N - 1, and have seen a range belonging | |
10477 | * to the block group in freed_extents[] before we were able to | |
10478 | * clear the whole block group range from freed_extents[]. This | |
10479 | * means that task can lookup for the block group after we | |
10480 | * unpinned it from freed_extents[] and removed it, leading to | |
10481 | * a BUG_ON() at btrfs_unpin_extent_range(). | |
10482 | */ | |
10483 | mutex_lock(&fs_info->unused_bg_unpin_mutex); | |
758eb51e | 10484 | ret = clear_extent_bits(&fs_info->freed_extents[0], start, end, |
47ab2a6c | 10485 | EXTENT_DIRTY, GFP_NOFS); |
758eb51e | 10486 | if (ret) { |
d4b450cd | 10487 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
868f401a | 10488 | btrfs_dec_block_group_ro(root, block_group); |
758eb51e FM |
10489 | goto end_trans; |
10490 | } | |
10491 | ret = clear_extent_bits(&fs_info->freed_extents[1], start, end, | |
47ab2a6c | 10492 | EXTENT_DIRTY, GFP_NOFS); |
758eb51e | 10493 | if (ret) { |
d4b450cd | 10494 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
868f401a | 10495 | btrfs_dec_block_group_ro(root, block_group); |
758eb51e FM |
10496 | goto end_trans; |
10497 | } | |
d4b450cd | 10498 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
47ab2a6c JB |
10499 | |
10500 | /* Reset pinned so btrfs_put_block_group doesn't complain */ | |
c30666d4 ZL |
10501 | spin_lock(&space_info->lock); |
10502 | spin_lock(&block_group->lock); | |
10503 | ||
10504 | space_info->bytes_pinned -= block_group->pinned; | |
10505 | space_info->bytes_readonly += block_group->pinned; | |
10506 | percpu_counter_add(&space_info->total_bytes_pinned, | |
10507 | -block_group->pinned); | |
47ab2a6c JB |
10508 | block_group->pinned = 0; |
10509 | ||
c30666d4 ZL |
10510 | spin_unlock(&block_group->lock); |
10511 | spin_unlock(&space_info->lock); | |
10512 | ||
e33e17ee JM |
10513 | /* DISCARD can flip during remount */ |
10514 | trimming = btrfs_test_opt(root, DISCARD); | |
10515 | ||
10516 | /* Implicit trim during transaction commit. */ | |
10517 | if (trimming) | |
10518 | btrfs_get_block_group_trimming(block_group); | |
10519 | ||
47ab2a6c JB |
10520 | /* |
10521 | * Btrfs_remove_chunk will abort the transaction if things go | |
10522 | * horribly wrong. | |
10523 | */ | |
10524 | ret = btrfs_remove_chunk(trans, root, | |
10525 | block_group->key.objectid); | |
e33e17ee JM |
10526 | |
10527 | if (ret) { | |
10528 | if (trimming) | |
10529 | btrfs_put_block_group_trimming(block_group); | |
10530 | goto end_trans; | |
10531 | } | |
10532 | ||
10533 | /* | |
10534 | * If we're not mounted with -odiscard, we can just forget | |
10535 | * about this block group. Otherwise we'll need to wait | |
10536 | * until transaction commit to do the actual discard. | |
10537 | */ | |
10538 | if (trimming) { | |
348a0013 FM |
10539 | spin_lock(&fs_info->unused_bgs_lock); |
10540 | /* | |
10541 | * A concurrent scrub might have added us to the list | |
10542 | * fs_info->unused_bgs, so use a list_move operation | |
10543 | * to add the block group to the deleted_bgs list. | |
10544 | */ | |
e33e17ee JM |
10545 | list_move(&block_group->bg_list, |
10546 | &trans->transaction->deleted_bgs); | |
348a0013 | 10547 | spin_unlock(&fs_info->unused_bgs_lock); |
e33e17ee JM |
10548 | btrfs_get_block_group(block_group); |
10549 | } | |
758eb51e | 10550 | end_trans: |
47ab2a6c JB |
10551 | btrfs_end_transaction(trans, root); |
10552 | next: | |
d5f2e33b | 10553 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
47ab2a6c JB |
10554 | btrfs_put_block_group(block_group); |
10555 | spin_lock(&fs_info->unused_bgs_lock); | |
10556 | } | |
10557 | spin_unlock(&fs_info->unused_bgs_lock); | |
10558 | } | |
10559 | ||
c59021f8 | 10560 | int btrfs_init_space_info(struct btrfs_fs_info *fs_info) |
10561 | { | |
10562 | struct btrfs_space_info *space_info; | |
1aba86d6 | 10563 | struct btrfs_super_block *disk_super; |
10564 | u64 features; | |
10565 | u64 flags; | |
10566 | int mixed = 0; | |
c59021f8 | 10567 | int ret; |
10568 | ||
6c41761f | 10569 | disk_super = fs_info->super_copy; |
1aba86d6 | 10570 | if (!btrfs_super_root(disk_super)) |
10571 | return 1; | |
c59021f8 | 10572 | |
1aba86d6 | 10573 | features = btrfs_super_incompat_flags(disk_super); |
10574 | if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
10575 | mixed = 1; | |
c59021f8 | 10576 | |
1aba86d6 | 10577 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
10578 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
c59021f8 | 10579 | if (ret) |
1aba86d6 | 10580 | goto out; |
c59021f8 | 10581 | |
1aba86d6 | 10582 | if (mixed) { |
10583 | flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA; | |
10584 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
10585 | } else { | |
10586 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
10587 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
10588 | if (ret) | |
10589 | goto out; | |
10590 | ||
10591 | flags = BTRFS_BLOCK_GROUP_DATA; | |
10592 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
10593 | } | |
10594 | out: | |
c59021f8 | 10595 | return ret; |
10596 | } | |
10597 | ||
acce952b | 10598 | int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end) |
10599 | { | |
678886bd | 10600 | return unpin_extent_range(root, start, end, false); |
acce952b | 10601 | } |
10602 | ||
499f377f JM |
10603 | /* |
10604 | * It used to be that old block groups would be left around forever. | |
10605 | * Iterating over them would be enough to trim unused space. Since we | |
10606 | * now automatically remove them, we also need to iterate over unallocated | |
10607 | * space. | |
10608 | * | |
10609 | * We don't want a transaction for this since the discard may take a | |
10610 | * substantial amount of time. We don't require that a transaction be | |
10611 | * running, but we do need to take a running transaction into account | |
10612 | * to ensure that we're not discarding chunks that were released in | |
10613 | * the current transaction. | |
10614 | * | |
10615 | * Holding the chunks lock will prevent other threads from allocating | |
10616 | * or releasing chunks, but it won't prevent a running transaction | |
10617 | * from committing and releasing the memory that the pending chunks | |
10618 | * list head uses. For that, we need to take a reference to the | |
10619 | * transaction. | |
10620 | */ | |
10621 | static int btrfs_trim_free_extents(struct btrfs_device *device, | |
10622 | u64 minlen, u64 *trimmed) | |
10623 | { | |
10624 | u64 start = 0, len = 0; | |
10625 | int ret; | |
10626 | ||
10627 | *trimmed = 0; | |
10628 | ||
10629 | /* Not writeable = nothing to do. */ | |
10630 | if (!device->writeable) | |
10631 | return 0; | |
10632 | ||
10633 | /* No free space = nothing to do. */ | |
10634 | if (device->total_bytes <= device->bytes_used) | |
10635 | return 0; | |
10636 | ||
10637 | ret = 0; | |
10638 | ||
10639 | while (1) { | |
10640 | struct btrfs_fs_info *fs_info = device->dev_root->fs_info; | |
10641 | struct btrfs_transaction *trans; | |
10642 | u64 bytes; | |
10643 | ||
10644 | ret = mutex_lock_interruptible(&fs_info->chunk_mutex); | |
10645 | if (ret) | |
10646 | return ret; | |
10647 | ||
10648 | down_read(&fs_info->commit_root_sem); | |
10649 | ||
10650 | spin_lock(&fs_info->trans_lock); | |
10651 | trans = fs_info->running_transaction; | |
10652 | if (trans) | |
10653 | atomic_inc(&trans->use_count); | |
10654 | spin_unlock(&fs_info->trans_lock); | |
10655 | ||
10656 | ret = find_free_dev_extent_start(trans, device, minlen, start, | |
10657 | &start, &len); | |
10658 | if (trans) | |
10659 | btrfs_put_transaction(trans); | |
10660 | ||
10661 | if (ret) { | |
10662 | up_read(&fs_info->commit_root_sem); | |
10663 | mutex_unlock(&fs_info->chunk_mutex); | |
10664 | if (ret == -ENOSPC) | |
10665 | ret = 0; | |
10666 | break; | |
10667 | } | |
10668 | ||
10669 | ret = btrfs_issue_discard(device->bdev, start, len, &bytes); | |
10670 | up_read(&fs_info->commit_root_sem); | |
10671 | mutex_unlock(&fs_info->chunk_mutex); | |
10672 | ||
10673 | if (ret) | |
10674 | break; | |
10675 | ||
10676 | start += len; | |
10677 | *trimmed += bytes; | |
10678 | ||
10679 | if (fatal_signal_pending(current)) { | |
10680 | ret = -ERESTARTSYS; | |
10681 | break; | |
10682 | } | |
10683 | ||
10684 | cond_resched(); | |
10685 | } | |
10686 | ||
10687 | return ret; | |
10688 | } | |
10689 | ||
f7039b1d LD |
10690 | int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range) |
10691 | { | |
10692 | struct btrfs_fs_info *fs_info = root->fs_info; | |
10693 | struct btrfs_block_group_cache *cache = NULL; | |
499f377f JM |
10694 | struct btrfs_device *device; |
10695 | struct list_head *devices; | |
f7039b1d LD |
10696 | u64 group_trimmed; |
10697 | u64 start; | |
10698 | u64 end; | |
10699 | u64 trimmed = 0; | |
2cac13e4 | 10700 | u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy); |
f7039b1d LD |
10701 | int ret = 0; |
10702 | ||
2cac13e4 LB |
10703 | /* |
10704 | * try to trim all FS space, our block group may start from non-zero. | |
10705 | */ | |
10706 | if (range->len == total_bytes) | |
10707 | cache = btrfs_lookup_first_block_group(fs_info, range->start); | |
10708 | else | |
10709 | cache = btrfs_lookup_block_group(fs_info, range->start); | |
f7039b1d LD |
10710 | |
10711 | while (cache) { | |
10712 | if (cache->key.objectid >= (range->start + range->len)) { | |
10713 | btrfs_put_block_group(cache); | |
10714 | break; | |
10715 | } | |
10716 | ||
10717 | start = max(range->start, cache->key.objectid); | |
10718 | end = min(range->start + range->len, | |
10719 | cache->key.objectid + cache->key.offset); | |
10720 | ||
10721 | if (end - start >= range->minlen) { | |
10722 | if (!block_group_cache_done(cache)) { | |
f6373bf3 | 10723 | ret = cache_block_group(cache, 0); |
1be41b78 JB |
10724 | if (ret) { |
10725 | btrfs_put_block_group(cache); | |
10726 | break; | |
10727 | } | |
10728 | ret = wait_block_group_cache_done(cache); | |
10729 | if (ret) { | |
10730 | btrfs_put_block_group(cache); | |
10731 | break; | |
10732 | } | |
f7039b1d LD |
10733 | } |
10734 | ret = btrfs_trim_block_group(cache, | |
10735 | &group_trimmed, | |
10736 | start, | |
10737 | end, | |
10738 | range->minlen); | |
10739 | ||
10740 | trimmed += group_trimmed; | |
10741 | if (ret) { | |
10742 | btrfs_put_block_group(cache); | |
10743 | break; | |
10744 | } | |
10745 | } | |
10746 | ||
10747 | cache = next_block_group(fs_info->tree_root, cache); | |
10748 | } | |
10749 | ||
499f377f JM |
10750 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); |
10751 | devices = &root->fs_info->fs_devices->alloc_list; | |
10752 | list_for_each_entry(device, devices, dev_alloc_list) { | |
10753 | ret = btrfs_trim_free_extents(device, range->minlen, | |
10754 | &group_trimmed); | |
10755 | if (ret) | |
10756 | break; | |
10757 | ||
10758 | trimmed += group_trimmed; | |
10759 | } | |
10760 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); | |
10761 | ||
f7039b1d LD |
10762 | range->len = trimmed; |
10763 | return ret; | |
10764 | } | |
8257b2dc MX |
10765 | |
10766 | /* | |
9ea24bbe FM |
10767 | * btrfs_{start,end}_write_no_snapshoting() are similar to |
10768 | * mnt_{want,drop}_write(), they are used to prevent some tasks from writing | |
10769 | * data into the page cache through nocow before the subvolume is snapshoted, | |
10770 | * but flush the data into disk after the snapshot creation, or to prevent | |
10771 | * operations while snapshoting is ongoing and that cause the snapshot to be | |
10772 | * inconsistent (writes followed by expanding truncates for example). | |
8257b2dc | 10773 | */ |
9ea24bbe | 10774 | void btrfs_end_write_no_snapshoting(struct btrfs_root *root) |
8257b2dc MX |
10775 | { |
10776 | percpu_counter_dec(&root->subv_writers->counter); | |
10777 | /* | |
a83342aa | 10778 | * Make sure counter is updated before we wake up waiters. |
8257b2dc MX |
10779 | */ |
10780 | smp_mb(); | |
10781 | if (waitqueue_active(&root->subv_writers->wait)) | |
10782 | wake_up(&root->subv_writers->wait); | |
10783 | } | |
10784 | ||
9ea24bbe | 10785 | int btrfs_start_write_no_snapshoting(struct btrfs_root *root) |
8257b2dc | 10786 | { |
ee39b432 | 10787 | if (atomic_read(&root->will_be_snapshoted)) |
8257b2dc MX |
10788 | return 0; |
10789 | ||
10790 | percpu_counter_inc(&root->subv_writers->counter); | |
10791 | /* | |
10792 | * Make sure counter is updated before we check for snapshot creation. | |
10793 | */ | |
10794 | smp_mb(); | |
ee39b432 | 10795 | if (atomic_read(&root->will_be_snapshoted)) { |
9ea24bbe | 10796 | btrfs_end_write_no_snapshoting(root); |
8257b2dc MX |
10797 | return 0; |
10798 | } | |
10799 | return 1; | |
10800 | } |