]>
Commit | Line | Data |
---|---|---|
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 | */ | |
18 | #include <linux/sched.h> | |
19 | #include <linux/pagemap.h> | |
20 | #include <linux/writeback.h> | |
21 | #include <linux/blkdev.h> | |
22 | #include <linux/sort.h> | |
23 | #include <linux/rcupdate.h> | |
24 | #include <linux/kthread.h> | |
25 | #include <linux/slab.h> | |
26 | #include <linux/ratelimit.h> | |
27 | #include <linux/percpu_counter.h> | |
28 | #include "hash.h" | |
29 | #include "tree-log.h" | |
30 | #include "disk-io.h" | |
31 | #include "print-tree.h" | |
32 | #include "volumes.h" | |
33 | #include "raid56.h" | |
34 | #include "locking.h" | |
35 | #include "free-space-cache.h" | |
36 | #include "free-space-tree.h" | |
37 | #include "math.h" | |
38 | #include "sysfs.h" | |
39 | #include "qgroup.h" | |
40 | ||
41 | #undef SCRAMBLE_DELAYED_REFS | |
42 | ||
43 | /* | |
44 | * control flags for do_chunk_alloc's force field | |
45 | * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk | |
46 | * if we really need one. | |
47 | * | |
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 | * | |
54 | * CHUNK_ALLOC_FORCE means it must try to allocate one | |
55 | * | |
56 | */ | |
57 | enum { | |
58 | CHUNK_ALLOC_NO_FORCE = 0, | |
59 | CHUNK_ALLOC_LIMITED = 1, | |
60 | CHUNK_ALLOC_FORCE = 2, | |
61 | }; | |
62 | ||
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 | ||
78 | static int update_block_group(struct btrfs_trans_handle *trans, | |
79 | struct btrfs_root *root, u64 bytenr, | |
80 | u64 num_bytes, int alloc); | |
81 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, | |
82 | struct btrfs_root *root, | |
83 | struct btrfs_delayed_ref_node *node, u64 parent, | |
84 | u64 root_objectid, u64 owner_objectid, | |
85 | u64 owner_offset, int refs_to_drop, | |
86 | struct btrfs_delayed_extent_op *extra_op); | |
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, | |
99 | int level, struct btrfs_key *ins); | |
100 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, | |
101 | struct btrfs_root *extent_root, u64 flags, | |
102 | int force); | |
103 | static int find_next_key(struct btrfs_path *path, int level, | |
104 | struct btrfs_key *key); | |
105 | static void dump_space_info(struct btrfs_space_info *info, u64 bytes, | |
106 | int dump_block_groups); | |
107 | static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache, | |
108 | u64 num_bytes, int reserve, | |
109 | int delalloc); | |
110 | static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, | |
111 | u64 num_bytes); | |
112 | int btrfs_pin_extent(struct btrfs_root *root, | |
113 | u64 bytenr, u64 num_bytes, int reserved); | |
114 | ||
115 | static noinline int | |
116 | block_group_cache_done(struct btrfs_block_group_cache *cache) | |
117 | { | |
118 | smp_mb(); | |
119 | return cache->cached == BTRFS_CACHE_FINISHED || | |
120 | cache->cached == BTRFS_CACHE_ERROR; | |
121 | } | |
122 | ||
123 | static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits) | |
124 | { | |
125 | return (cache->flags & bits) == bits; | |
126 | } | |
127 | ||
128 | void btrfs_get_block_group(struct btrfs_block_group_cache *cache) | |
129 | { | |
130 | atomic_inc(&cache->count); | |
131 | } | |
132 | ||
133 | void btrfs_put_block_group(struct btrfs_block_group_cache *cache) | |
134 | { | |
135 | if (atomic_dec_and_test(&cache->count)) { | |
136 | WARN_ON(cache->pinned > 0); | |
137 | WARN_ON(cache->reserved > 0); | |
138 | kfree(cache->free_space_ctl); | |
139 | kfree(cache); | |
140 | } | |
141 | } | |
142 | ||
143 | /* | |
144 | * this adds the block group to the fs_info rb tree for the block group | |
145 | * cache | |
146 | */ | |
147 | static int btrfs_add_block_group_cache(struct btrfs_fs_info *info, | |
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); | |
174 | ||
175 | if (info->first_logical_byte > block_group->key.objectid) | |
176 | info->first_logical_byte = block_group->key.objectid; | |
177 | ||
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 | } | |
219 | if (ret) { | |
220 | btrfs_get_block_group(ret); | |
221 | if (bytenr == 0 && info->first_logical_byte > ret->key.objectid) | |
222 | info->first_logical_byte = ret->key.objectid; | |
223 | } | |
224 | spin_unlock(&info->block_group_cache_lock); | |
225 | ||
226 | return ret; | |
227 | } | |
228 | ||
229 | static int add_excluded_extent(struct btrfs_root *root, | |
230 | u64 start, u64 num_bytes) | |
231 | { | |
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 | } | |
239 | ||
240 | static void free_excluded_extents(struct btrfs_root *root, | |
241 | struct btrfs_block_group_cache *cache) | |
242 | { | |
243 | u64 start, end; | |
244 | ||
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); | |
252 | } | |
253 | ||
254 | static int exclude_super_stripes(struct btrfs_root *root, | |
255 | struct btrfs_block_group_cache *cache) | |
256 | { | |
257 | u64 bytenr; | |
258 | u64 *logical; | |
259 | int stripe_len; | |
260 | int i, nr, ret; | |
261 | ||
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); | |
267 | if (ret) | |
268 | return ret; | |
269 | } | |
270 | ||
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); | |
276 | if (ret) | |
277 | return ret; | |
278 | ||
279 | while (nr--) { | |
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); | |
301 | if (ret) { | |
302 | kfree(logical); | |
303 | return ret; | |
304 | } | |
305 | } | |
306 | ||
307 | kfree(logical); | |
308 | } | |
309 | return 0; | |
310 | } | |
311 | ||
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); | |
318 | if (!cache->caching_ctl) { | |
319 | spin_unlock(&cache->lock); | |
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 | ||
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 | ||
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 | */ | |
361 | u64 add_new_free_space(struct btrfs_block_group_cache *block_group, | |
362 | struct btrfs_fs_info *info, u64 start, u64 end) | |
363 | { | |
364 | u64 extent_start, extent_end, size, total_added = 0; | |
365 | int ret; | |
366 | ||
367 | while (start < end) { | |
368 | ret = find_first_extent_bit(info->pinned_extents, start, | |
369 | &extent_start, &extent_end, | |
370 | EXTENT_DIRTY | EXTENT_UPTODATE, | |
371 | NULL); | |
372 | if (ret) | |
373 | break; | |
374 | ||
375 | if (extent_start <= start) { | |
376 | start = extent_end + 1; | |
377 | } else if (extent_start > start && extent_start < end) { | |
378 | size = extent_start - start; | |
379 | total_added += size; | |
380 | ret = btrfs_add_free_space(block_group, start, | |
381 | size); | |
382 | BUG_ON(ret); /* -ENOMEM or logic error */ | |
383 | start = extent_end + 1; | |
384 | } else { | |
385 | break; | |
386 | } | |
387 | } | |
388 | ||
389 | if (start < end) { | |
390 | size = end - start; | |
391 | total_added += size; | |
392 | ret = btrfs_add_free_space(block_group, start, size); | |
393 | BUG_ON(ret); /* -ENOMEM or logic error */ | |
394 | } | |
395 | ||
396 | return total_added; | |
397 | } | |
398 | ||
399 | static int load_extent_tree_free(struct btrfs_caching_control *caching_ctl) | |
400 | { | |
401 | struct btrfs_block_group_cache *block_group; | |
402 | struct btrfs_fs_info *fs_info; | |
403 | struct btrfs_root *extent_root; | |
404 | struct btrfs_path *path; | |
405 | struct extent_buffer *leaf; | |
406 | struct btrfs_key key; | |
407 | u64 total_found = 0; | |
408 | u64 last = 0; | |
409 | u32 nritems; | |
410 | int ret; | |
411 | bool wakeup = true; | |
412 | ||
413 | block_group = caching_ctl->block_group; | |
414 | fs_info = block_group->fs_info; | |
415 | extent_root = fs_info->extent_root; | |
416 | ||
417 | path = btrfs_alloc_path(); | |
418 | if (!path) | |
419 | return -ENOMEM; | |
420 | ||
421 | last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET); | |
422 | ||
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 | |
432 | /* | |
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 | |
437 | */ | |
438 | path->skip_locking = 1; | |
439 | path->search_commit_root = 1; | |
440 | path->reada = READA_FORWARD; | |
441 | ||
442 | key.objectid = last; | |
443 | key.offset = 0; | |
444 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
445 | ||
446 | next: | |
447 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); | |
448 | if (ret < 0) | |
449 | goto out; | |
450 | ||
451 | leaf = path->nodes[0]; | |
452 | nritems = btrfs_header_nritems(leaf); | |
453 | ||
454 | while (1) { | |
455 | if (btrfs_fs_closing(fs_info) > 1) { | |
456 | last = (u64)-1; | |
457 | break; | |
458 | } | |
459 | ||
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) | |
465 | break; | |
466 | ||
467 | if (need_resched() || | |
468 | rwsem_is_contended(&fs_info->commit_root_sem)) { | |
469 | if (wakeup) | |
470 | caching_ctl->progress = last; | |
471 | btrfs_release_path(path); | |
472 | up_read(&fs_info->commit_root_sem); | |
473 | mutex_unlock(&caching_ctl->mutex); | |
474 | cond_resched(); | |
475 | mutex_lock(&caching_ctl->mutex); | |
476 | down_read(&fs_info->commit_root_sem); | |
477 | goto next; | |
478 | } | |
479 | ||
480 | ret = btrfs_next_leaf(extent_root, path); | |
481 | if (ret < 0) | |
482 | goto out; | |
483 | if (ret) | |
484 | break; | |
485 | leaf = path->nodes[0]; | |
486 | nritems = btrfs_header_nritems(leaf); | |
487 | continue; | |
488 | } | |
489 | ||
490 | if (key.objectid < last) { | |
491 | key.objectid = last; | |
492 | key.offset = 0; | |
493 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
494 | ||
495 | if (wakeup) | |
496 | caching_ctl->progress = last; | |
497 | btrfs_release_path(path); | |
498 | goto next; | |
499 | } | |
500 | ||
501 | if (key.objectid < block_group->key.objectid) { | |
502 | path->slots[0]++; | |
503 | continue; | |
504 | } | |
505 | ||
506 | if (key.objectid >= block_group->key.objectid + | |
507 | block_group->key.offset) | |
508 | break; | |
509 | ||
510 | if (key.type == BTRFS_EXTENT_ITEM_KEY || | |
511 | key.type == BTRFS_METADATA_ITEM_KEY) { | |
512 | total_found += add_new_free_space(block_group, | |
513 | fs_info, last, | |
514 | key.objectid); | |
515 | if (key.type == BTRFS_METADATA_ITEM_KEY) | |
516 | last = key.objectid + | |
517 | fs_info->tree_root->nodesize; | |
518 | else | |
519 | last = key.objectid + key.offset; | |
520 | ||
521 | if (total_found > CACHING_CTL_WAKE_UP) { | |
522 | total_found = 0; | |
523 | if (wakeup) | |
524 | wake_up(&caching_ctl->wait); | |
525 | } | |
526 | } | |
527 | path->slots[0]++; | |
528 | } | |
529 | ret = 0; | |
530 | ||
531 | total_found += add_new_free_space(block_group, fs_info, last, | |
532 | block_group->key.objectid + | |
533 | block_group->key.offset); | |
534 | caching_ctl->progress = (u64)-1; | |
535 | ||
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 | struct btrfs_root *extent_root; | |
547 | int ret; | |
548 | ||
549 | caching_ctl = container_of(work, struct btrfs_caching_control, work); | |
550 | block_group = caching_ctl->block_group; | |
551 | fs_info = block_group->fs_info; | |
552 | extent_root = fs_info->extent_root; | |
553 | ||
554 | mutex_lock(&caching_ctl->mutex); | |
555 | down_read(&fs_info->commit_root_sem); | |
556 | ||
557 | if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) | |
558 | ret = load_free_space_tree(caching_ctl); | |
559 | else | |
560 | ret = load_extent_tree_free(caching_ctl); | |
561 | ||
562 | spin_lock(&block_group->lock); | |
563 | block_group->caching_ctl = NULL; | |
564 | block_group->cached = ret ? BTRFS_CACHE_ERROR : BTRFS_CACHE_FINISHED; | |
565 | spin_unlock(&block_group->lock); | |
566 | ||
567 | #ifdef CONFIG_BTRFS_DEBUG | |
568 | if (btrfs_should_fragment_free_space(extent_root, block_group)) { | |
569 | u64 bytes_used; | |
570 | ||
571 | spin_lock(&block_group->space_info->lock); | |
572 | spin_lock(&block_group->lock); | |
573 | bytes_used = block_group->key.offset - | |
574 | btrfs_block_group_used(&block_group->item); | |
575 | block_group->space_info->bytes_used += bytes_used >> 1; | |
576 | spin_unlock(&block_group->lock); | |
577 | spin_unlock(&block_group->space_info->lock); | |
578 | fragment_free_space(extent_root, block_group); | |
579 | } | |
580 | #endif | |
581 | ||
582 | caching_ctl->progress = (u64)-1; | |
583 | ||
584 | up_read(&fs_info->commit_root_sem); | |
585 | free_excluded_extents(fs_info->extent_root, block_group); | |
586 | mutex_unlock(&caching_ctl->mutex); | |
587 | ||
588 | wake_up(&caching_ctl->wait); | |
589 | ||
590 | put_caching_control(caching_ctl); | |
591 | btrfs_put_block_group(block_group); | |
592 | } | |
593 | ||
594 | static int cache_block_group(struct btrfs_block_group_cache *cache, | |
595 | int load_cache_only) | |
596 | { | |
597 | DEFINE_WAIT(wait); | |
598 | struct btrfs_fs_info *fs_info = cache->fs_info; | |
599 | struct btrfs_caching_control *caching_ctl; | |
600 | int ret = 0; | |
601 | ||
602 | caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS); | |
603 | if (!caching_ctl) | |
604 | return -ENOMEM; | |
605 | ||
606 | INIT_LIST_HEAD(&caching_ctl->list); | |
607 | mutex_init(&caching_ctl->mutex); | |
608 | init_waitqueue_head(&caching_ctl->wait); | |
609 | caching_ctl->block_group = cache; | |
610 | caching_ctl->progress = cache->key.objectid; | |
611 | atomic_set(&caching_ctl->count, 1); | |
612 | btrfs_init_work(&caching_ctl->work, btrfs_cache_helper, | |
613 | caching_thread, NULL, NULL); | |
614 | ||
615 | spin_lock(&cache->lock); | |
616 | /* | |
617 | * This should be a rare occasion, but this could happen I think in the | |
618 | * case where one thread starts to load the space cache info, and then | |
619 | * some other thread starts a transaction commit which tries to do an | |
620 | * allocation while the other thread is still loading the space cache | |
621 | * info. The previous loop should have kept us from choosing this block | |
622 | * group, but if we've moved to the state where we will wait on caching | |
623 | * block groups we need to first check if we're doing a fast load here, | |
624 | * so we can wait for it to finish, otherwise we could end up allocating | |
625 | * from a block group who's cache gets evicted for one reason or | |
626 | * another. | |
627 | */ | |
628 | while (cache->cached == BTRFS_CACHE_FAST) { | |
629 | struct btrfs_caching_control *ctl; | |
630 | ||
631 | ctl = cache->caching_ctl; | |
632 | atomic_inc(&ctl->count); | |
633 | prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE); | |
634 | spin_unlock(&cache->lock); | |
635 | ||
636 | schedule(); | |
637 | ||
638 | finish_wait(&ctl->wait, &wait); | |
639 | put_caching_control(ctl); | |
640 | spin_lock(&cache->lock); | |
641 | } | |
642 | ||
643 | if (cache->cached != BTRFS_CACHE_NO) { | |
644 | spin_unlock(&cache->lock); | |
645 | kfree(caching_ctl); | |
646 | return 0; | |
647 | } | |
648 | WARN_ON(cache->caching_ctl); | |
649 | cache->caching_ctl = caching_ctl; | |
650 | cache->cached = BTRFS_CACHE_FAST; | |
651 | spin_unlock(&cache->lock); | |
652 | ||
653 | if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) { | |
654 | mutex_lock(&caching_ctl->mutex); | |
655 | ret = load_free_space_cache(fs_info, cache); | |
656 | ||
657 | spin_lock(&cache->lock); | |
658 | if (ret == 1) { | |
659 | cache->caching_ctl = NULL; | |
660 | cache->cached = BTRFS_CACHE_FINISHED; | |
661 | cache->last_byte_to_unpin = (u64)-1; | |
662 | caching_ctl->progress = (u64)-1; | |
663 | } else { | |
664 | if (load_cache_only) { | |
665 | cache->caching_ctl = NULL; | |
666 | cache->cached = BTRFS_CACHE_NO; | |
667 | } else { | |
668 | cache->cached = BTRFS_CACHE_STARTED; | |
669 | cache->has_caching_ctl = 1; | |
670 | } | |
671 | } | |
672 | spin_unlock(&cache->lock); | |
673 | #ifdef CONFIG_BTRFS_DEBUG | |
674 | if (ret == 1 && | |
675 | btrfs_should_fragment_free_space(fs_info->extent_root, | |
676 | cache)) { | |
677 | u64 bytes_used; | |
678 | ||
679 | spin_lock(&cache->space_info->lock); | |
680 | spin_lock(&cache->lock); | |
681 | bytes_used = cache->key.offset - | |
682 | btrfs_block_group_used(&cache->item); | |
683 | cache->space_info->bytes_used += bytes_used >> 1; | |
684 | spin_unlock(&cache->lock); | |
685 | spin_unlock(&cache->space_info->lock); | |
686 | fragment_free_space(fs_info->extent_root, cache); | |
687 | } | |
688 | #endif | |
689 | mutex_unlock(&caching_ctl->mutex); | |
690 | ||
691 | wake_up(&caching_ctl->wait); | |
692 | if (ret == 1) { | |
693 | put_caching_control(caching_ctl); | |
694 | free_excluded_extents(fs_info->extent_root, cache); | |
695 | return 0; | |
696 | } | |
697 | } else { | |
698 | /* | |
699 | * We're either using the free space tree or no caching at all. | |
700 | * Set cached to the appropriate value and wakeup any waiters. | |
701 | */ | |
702 | spin_lock(&cache->lock); | |
703 | if (load_cache_only) { | |
704 | cache->caching_ctl = NULL; | |
705 | cache->cached = BTRFS_CACHE_NO; | |
706 | } else { | |
707 | cache->cached = BTRFS_CACHE_STARTED; | |
708 | cache->has_caching_ctl = 1; | |
709 | } | |
710 | spin_unlock(&cache->lock); | |
711 | wake_up(&caching_ctl->wait); | |
712 | } | |
713 | ||
714 | if (load_cache_only) { | |
715 | put_caching_control(caching_ctl); | |
716 | return 0; | |
717 | } | |
718 | ||
719 | down_write(&fs_info->commit_root_sem); | |
720 | atomic_inc(&caching_ctl->count); | |
721 | list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups); | |
722 | up_write(&fs_info->commit_root_sem); | |
723 | ||
724 | btrfs_get_block_group(cache); | |
725 | ||
726 | btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work); | |
727 | ||
728 | return ret; | |
729 | } | |
730 | ||
731 | /* | |
732 | * return the block group that starts at or after bytenr | |
733 | */ | |
734 | static struct btrfs_block_group_cache * | |
735 | btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr) | |
736 | { | |
737 | struct btrfs_block_group_cache *cache; | |
738 | ||
739 | cache = block_group_cache_tree_search(info, bytenr, 0); | |
740 | ||
741 | return cache; | |
742 | } | |
743 | ||
744 | /* | |
745 | * return the block group that contains the given bytenr | |
746 | */ | |
747 | struct btrfs_block_group_cache *btrfs_lookup_block_group( | |
748 | struct btrfs_fs_info *info, | |
749 | u64 bytenr) | |
750 | { | |
751 | struct btrfs_block_group_cache *cache; | |
752 | ||
753 | cache = block_group_cache_tree_search(info, bytenr, 1); | |
754 | ||
755 | return cache; | |
756 | } | |
757 | ||
758 | static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info, | |
759 | u64 flags) | |
760 | { | |
761 | struct list_head *head = &info->space_info; | |
762 | struct btrfs_space_info *found; | |
763 | ||
764 | flags &= BTRFS_BLOCK_GROUP_TYPE_MASK; | |
765 | ||
766 | rcu_read_lock(); | |
767 | list_for_each_entry_rcu(found, head, list) { | |
768 | if (found->flags & flags) { | |
769 | rcu_read_unlock(); | |
770 | return found; | |
771 | } | |
772 | } | |
773 | rcu_read_unlock(); | |
774 | return NULL; | |
775 | } | |
776 | ||
777 | /* | |
778 | * after adding space to the filesystem, we need to clear the full flags | |
779 | * on all the space infos. | |
780 | */ | |
781 | void btrfs_clear_space_info_full(struct btrfs_fs_info *info) | |
782 | { | |
783 | struct list_head *head = &info->space_info; | |
784 | struct btrfs_space_info *found; | |
785 | ||
786 | rcu_read_lock(); | |
787 | list_for_each_entry_rcu(found, head, list) | |
788 | found->full = 0; | |
789 | rcu_read_unlock(); | |
790 | } | |
791 | ||
792 | /* simple helper to search for an existing data extent at a given offset */ | |
793 | int btrfs_lookup_data_extent(struct btrfs_root *root, u64 start, u64 len) | |
794 | { | |
795 | int ret; | |
796 | struct btrfs_key key; | |
797 | struct btrfs_path *path; | |
798 | ||
799 | path = btrfs_alloc_path(); | |
800 | if (!path) | |
801 | return -ENOMEM; | |
802 | ||
803 | key.objectid = start; | |
804 | key.offset = len; | |
805 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
806 | ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path, | |
807 | 0, 0); | |
808 | btrfs_free_path(path); | |
809 | return ret; | |
810 | } | |
811 | ||
812 | /* | |
813 | * helper function to lookup reference count and flags of a tree block. | |
814 | * | |
815 | * the head node for delayed ref is used to store the sum of all the | |
816 | * reference count modifications queued up in the rbtree. the head | |
817 | * node may also store the extent flags to set. This way you can check | |
818 | * to see what the reference count and extent flags would be if all of | |
819 | * the delayed refs are not processed. | |
820 | */ | |
821 | int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans, | |
822 | struct btrfs_root *root, u64 bytenr, | |
823 | u64 offset, int metadata, u64 *refs, u64 *flags) | |
824 | { | |
825 | struct btrfs_delayed_ref_head *head; | |
826 | struct btrfs_delayed_ref_root *delayed_refs; | |
827 | struct btrfs_path *path; | |
828 | struct btrfs_extent_item *ei; | |
829 | struct extent_buffer *leaf; | |
830 | struct btrfs_key key; | |
831 | u32 item_size; | |
832 | u64 num_refs; | |
833 | u64 extent_flags; | |
834 | int ret; | |
835 | ||
836 | /* | |
837 | * If we don't have skinny metadata, don't bother doing anything | |
838 | * different | |
839 | */ | |
840 | if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) { | |
841 | offset = root->nodesize; | |
842 | metadata = 0; | |
843 | } | |
844 | ||
845 | path = btrfs_alloc_path(); | |
846 | if (!path) | |
847 | return -ENOMEM; | |
848 | ||
849 | if (!trans) { | |
850 | path->skip_locking = 1; | |
851 | path->search_commit_root = 1; | |
852 | } | |
853 | ||
854 | search_again: | |
855 | key.objectid = bytenr; | |
856 | key.offset = offset; | |
857 | if (metadata) | |
858 | key.type = BTRFS_METADATA_ITEM_KEY; | |
859 | else | |
860 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
861 | ||
862 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, | |
863 | &key, path, 0, 0); | |
864 | if (ret < 0) | |
865 | goto out_free; | |
866 | ||
867 | if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) { | |
868 | if (path->slots[0]) { | |
869 | path->slots[0]--; | |
870 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
871 | path->slots[0]); | |
872 | if (key.objectid == bytenr && | |
873 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
874 | key.offset == root->nodesize) | |
875 | ret = 0; | |
876 | } | |
877 | } | |
878 | ||
879 | if (ret == 0) { | |
880 | leaf = path->nodes[0]; | |
881 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
882 | if (item_size >= sizeof(*ei)) { | |
883 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
884 | struct btrfs_extent_item); | |
885 | num_refs = btrfs_extent_refs(leaf, ei); | |
886 | extent_flags = btrfs_extent_flags(leaf, ei); | |
887 | } else { | |
888 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
889 | struct btrfs_extent_item_v0 *ei0; | |
890 | BUG_ON(item_size != sizeof(*ei0)); | |
891 | ei0 = btrfs_item_ptr(leaf, path->slots[0], | |
892 | struct btrfs_extent_item_v0); | |
893 | num_refs = btrfs_extent_refs_v0(leaf, ei0); | |
894 | /* FIXME: this isn't correct for data */ | |
895 | extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
896 | #else | |
897 | BUG(); | |
898 | #endif | |
899 | } | |
900 | BUG_ON(num_refs == 0); | |
901 | } else { | |
902 | num_refs = 0; | |
903 | extent_flags = 0; | |
904 | ret = 0; | |
905 | } | |
906 | ||
907 | if (!trans) | |
908 | goto out; | |
909 | ||
910 | delayed_refs = &trans->transaction->delayed_refs; | |
911 | spin_lock(&delayed_refs->lock); | |
912 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
913 | if (head) { | |
914 | if (!mutex_trylock(&head->mutex)) { | |
915 | atomic_inc(&head->node.refs); | |
916 | spin_unlock(&delayed_refs->lock); | |
917 | ||
918 | btrfs_release_path(path); | |
919 | ||
920 | /* | |
921 | * Mutex was contended, block until it's released and try | |
922 | * again | |
923 | */ | |
924 | mutex_lock(&head->mutex); | |
925 | mutex_unlock(&head->mutex); | |
926 | btrfs_put_delayed_ref(&head->node); | |
927 | goto search_again; | |
928 | } | |
929 | spin_lock(&head->lock); | |
930 | if (head->extent_op && head->extent_op->update_flags) | |
931 | extent_flags |= head->extent_op->flags_to_set; | |
932 | else | |
933 | BUG_ON(num_refs == 0); | |
934 | ||
935 | num_refs += head->node.ref_mod; | |
936 | spin_unlock(&head->lock); | |
937 | mutex_unlock(&head->mutex); | |
938 | } | |
939 | spin_unlock(&delayed_refs->lock); | |
940 | out: | |
941 | WARN_ON(num_refs == 0); | |
942 | if (refs) | |
943 | *refs = num_refs; | |
944 | if (flags) | |
945 | *flags = extent_flags; | |
946 | out_free: | |
947 | btrfs_free_path(path); | |
948 | return ret; | |
949 | } | |
950 | ||
951 | /* | |
952 | * Back reference rules. Back refs have three main goals: | |
953 | * | |
954 | * 1) differentiate between all holders of references to an extent so that | |
955 | * when a reference is dropped we can make sure it was a valid reference | |
956 | * before freeing the extent. | |
957 | * | |
958 | * 2) Provide enough information to quickly find the holders of an extent | |
959 | * if we notice a given block is corrupted or bad. | |
960 | * | |
961 | * 3) Make it easy to migrate blocks for FS shrinking or storage pool | |
962 | * maintenance. This is actually the same as #2, but with a slightly | |
963 | * different use case. | |
964 | * | |
965 | * There are two kinds of back refs. The implicit back refs is optimized | |
966 | * for pointers in non-shared tree blocks. For a given pointer in a block, | |
967 | * back refs of this kind provide information about the block's owner tree | |
968 | * and the pointer's key. These information allow us to find the block by | |
969 | * b-tree searching. The full back refs is for pointers in tree blocks not | |
970 | * referenced by their owner trees. The location of tree block is recorded | |
971 | * in the back refs. Actually the full back refs is generic, and can be | |
972 | * used in all cases the implicit back refs is used. The major shortcoming | |
973 | * of the full back refs is its overhead. Every time a tree block gets | |
974 | * COWed, we have to update back refs entry for all pointers in it. | |
975 | * | |
976 | * For a newly allocated tree block, we use implicit back refs for | |
977 | * pointers in it. This means most tree related operations only involve | |
978 | * implicit back refs. For a tree block created in old transaction, the | |
979 | * only way to drop a reference to it is COW it. So we can detect the | |
980 | * event that tree block loses its owner tree's reference and do the | |
981 | * back refs conversion. | |
982 | * | |
983 | * When a tree block is COW'd through a tree, there are four cases: | |
984 | * | |
985 | * The reference count of the block is one and the tree is the block's | |
986 | * owner tree. Nothing to do in this case. | |
987 | * | |
988 | * The reference count of the block is one and the tree is not the | |
989 | * block's owner tree. In this case, full back refs is used for pointers | |
990 | * in the block. Remove these full back refs, add implicit back refs for | |
991 | * every pointers in the new block. | |
992 | * | |
993 | * The reference count of the block is greater than one and the tree is | |
994 | * the block's owner tree. In this case, implicit back refs is used for | |
995 | * pointers in the block. Add full back refs for every pointers in the | |
996 | * block, increase lower level extents' reference counts. The original | |
997 | * implicit back refs are entailed to the new block. | |
998 | * | |
999 | * The reference count of the block is greater than one and the tree is | |
1000 | * not the block's owner tree. Add implicit back refs for every pointer in | |
1001 | * the new block, increase lower level extents' reference count. | |
1002 | * | |
1003 | * Back Reference Key composing: | |
1004 | * | |
1005 | * The key objectid corresponds to the first byte in the extent, | |
1006 | * The key type is used to differentiate between types of back refs. | |
1007 | * There are different meanings of the key offset for different types | |
1008 | * of back refs. | |
1009 | * | |
1010 | * File extents can be referenced by: | |
1011 | * | |
1012 | * - multiple snapshots, subvolumes, or different generations in one subvol | |
1013 | * - different files inside a single subvolume | |
1014 | * - different offsets inside a file (bookend extents in file.c) | |
1015 | * | |
1016 | * The extent ref structure for the implicit back refs has fields for: | |
1017 | * | |
1018 | * - Objectid of the subvolume root | |
1019 | * - objectid of the file holding the reference | |
1020 | * - original offset in the file | |
1021 | * - how many bookend extents | |
1022 | * | |
1023 | * The key offset for the implicit back refs is hash of the first | |
1024 | * three fields. | |
1025 | * | |
1026 | * The extent ref structure for the full back refs has field for: | |
1027 | * | |
1028 | * - number of pointers in the tree leaf | |
1029 | * | |
1030 | * The key offset for the implicit back refs is the first byte of | |
1031 | * the tree leaf | |
1032 | * | |
1033 | * When a file extent is allocated, The implicit back refs is used. | |
1034 | * the fields are filled in: | |
1035 | * | |
1036 | * (root_key.objectid, inode objectid, offset in file, 1) | |
1037 | * | |
1038 | * When a file extent is removed file truncation, we find the | |
1039 | * corresponding implicit back refs and check the following fields: | |
1040 | * | |
1041 | * (btrfs_header_owner(leaf), inode objectid, offset in file) | |
1042 | * | |
1043 | * Btree extents can be referenced by: | |
1044 | * | |
1045 | * - Different subvolumes | |
1046 | * | |
1047 | * Both the implicit back refs and the full back refs for tree blocks | |
1048 | * only consist of key. The key offset for the implicit back refs is | |
1049 | * objectid of block's owner tree. The key offset for the full back refs | |
1050 | * is the first byte of parent block. | |
1051 | * | |
1052 | * When implicit back refs is used, information about the lowest key and | |
1053 | * level of the tree block are required. These information are stored in | |
1054 | * tree block info structure. | |
1055 | */ | |
1056 | ||
1057 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1058 | static int convert_extent_item_v0(struct btrfs_trans_handle *trans, | |
1059 | struct btrfs_root *root, | |
1060 | struct btrfs_path *path, | |
1061 | u64 owner, u32 extra_size) | |
1062 | { | |
1063 | struct btrfs_extent_item *item; | |
1064 | struct btrfs_extent_item_v0 *ei0; | |
1065 | struct btrfs_extent_ref_v0 *ref0; | |
1066 | struct btrfs_tree_block_info *bi; | |
1067 | struct extent_buffer *leaf; | |
1068 | struct btrfs_key key; | |
1069 | struct btrfs_key found_key; | |
1070 | u32 new_size = sizeof(*item); | |
1071 | u64 refs; | |
1072 | int ret; | |
1073 | ||
1074 | leaf = path->nodes[0]; | |
1075 | BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0)); | |
1076 | ||
1077 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1078 | ei0 = btrfs_item_ptr(leaf, path->slots[0], | |
1079 | struct btrfs_extent_item_v0); | |
1080 | refs = btrfs_extent_refs_v0(leaf, ei0); | |
1081 | ||
1082 | if (owner == (u64)-1) { | |
1083 | while (1) { | |
1084 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1085 | ret = btrfs_next_leaf(root, path); | |
1086 | if (ret < 0) | |
1087 | return ret; | |
1088 | BUG_ON(ret > 0); /* Corruption */ | |
1089 | leaf = path->nodes[0]; | |
1090 | } | |
1091 | btrfs_item_key_to_cpu(leaf, &found_key, | |
1092 | path->slots[0]); | |
1093 | BUG_ON(key.objectid != found_key.objectid); | |
1094 | if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) { | |
1095 | path->slots[0]++; | |
1096 | continue; | |
1097 | } | |
1098 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1099 | struct btrfs_extent_ref_v0); | |
1100 | owner = btrfs_ref_objectid_v0(leaf, ref0); | |
1101 | break; | |
1102 | } | |
1103 | } | |
1104 | btrfs_release_path(path); | |
1105 | ||
1106 | if (owner < BTRFS_FIRST_FREE_OBJECTID) | |
1107 | new_size += sizeof(*bi); | |
1108 | ||
1109 | new_size -= sizeof(*ei0); | |
1110 | ret = btrfs_search_slot(trans, root, &key, path, | |
1111 | new_size + extra_size, 1); | |
1112 | if (ret < 0) | |
1113 | return ret; | |
1114 | BUG_ON(ret); /* Corruption */ | |
1115 | ||
1116 | btrfs_extend_item(root, path, new_size); | |
1117 | ||
1118 | leaf = path->nodes[0]; | |
1119 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1120 | btrfs_set_extent_refs(leaf, item, refs); | |
1121 | /* FIXME: get real generation */ | |
1122 | btrfs_set_extent_generation(leaf, item, 0); | |
1123 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1124 | btrfs_set_extent_flags(leaf, item, | |
1125 | BTRFS_EXTENT_FLAG_TREE_BLOCK | | |
1126 | BTRFS_BLOCK_FLAG_FULL_BACKREF); | |
1127 | bi = (struct btrfs_tree_block_info *)(item + 1); | |
1128 | /* FIXME: get first key of the block */ | |
1129 | memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi)); | |
1130 | btrfs_set_tree_block_level(leaf, bi, (int)owner); | |
1131 | } else { | |
1132 | btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA); | |
1133 | } | |
1134 | btrfs_mark_buffer_dirty(leaf); | |
1135 | return 0; | |
1136 | } | |
1137 | #endif | |
1138 | ||
1139 | static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset) | |
1140 | { | |
1141 | u32 high_crc = ~(u32)0; | |
1142 | u32 low_crc = ~(u32)0; | |
1143 | __le64 lenum; | |
1144 | ||
1145 | lenum = cpu_to_le64(root_objectid); | |
1146 | high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum)); | |
1147 | lenum = cpu_to_le64(owner); | |
1148 | low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum)); | |
1149 | lenum = cpu_to_le64(offset); | |
1150 | low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum)); | |
1151 | ||
1152 | return ((u64)high_crc << 31) ^ (u64)low_crc; | |
1153 | } | |
1154 | ||
1155 | static u64 hash_extent_data_ref_item(struct extent_buffer *leaf, | |
1156 | struct btrfs_extent_data_ref *ref) | |
1157 | { | |
1158 | return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref), | |
1159 | btrfs_extent_data_ref_objectid(leaf, ref), | |
1160 | btrfs_extent_data_ref_offset(leaf, ref)); | |
1161 | } | |
1162 | ||
1163 | static int match_extent_data_ref(struct extent_buffer *leaf, | |
1164 | struct btrfs_extent_data_ref *ref, | |
1165 | u64 root_objectid, u64 owner, u64 offset) | |
1166 | { | |
1167 | if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid || | |
1168 | btrfs_extent_data_ref_objectid(leaf, ref) != owner || | |
1169 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
1170 | return 0; | |
1171 | return 1; | |
1172 | } | |
1173 | ||
1174 | static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans, | |
1175 | struct btrfs_root *root, | |
1176 | struct btrfs_path *path, | |
1177 | u64 bytenr, u64 parent, | |
1178 | u64 root_objectid, | |
1179 | u64 owner, u64 offset) | |
1180 | { | |
1181 | struct btrfs_key key; | |
1182 | struct btrfs_extent_data_ref *ref; | |
1183 | struct extent_buffer *leaf; | |
1184 | u32 nritems; | |
1185 | int ret; | |
1186 | int recow; | |
1187 | int err = -ENOENT; | |
1188 | ||
1189 | key.objectid = bytenr; | |
1190 | if (parent) { | |
1191 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1192 | key.offset = parent; | |
1193 | } else { | |
1194 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1195 | key.offset = hash_extent_data_ref(root_objectid, | |
1196 | owner, offset); | |
1197 | } | |
1198 | again: | |
1199 | recow = 0; | |
1200 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1201 | if (ret < 0) { | |
1202 | err = ret; | |
1203 | goto fail; | |
1204 | } | |
1205 | ||
1206 | if (parent) { | |
1207 | if (!ret) | |
1208 | return 0; | |
1209 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1210 | key.type = BTRFS_EXTENT_REF_V0_KEY; | |
1211 | btrfs_release_path(path); | |
1212 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1213 | if (ret < 0) { | |
1214 | err = ret; | |
1215 | goto fail; | |
1216 | } | |
1217 | if (!ret) | |
1218 | return 0; | |
1219 | #endif | |
1220 | goto fail; | |
1221 | } | |
1222 | ||
1223 | leaf = path->nodes[0]; | |
1224 | nritems = btrfs_header_nritems(leaf); | |
1225 | while (1) { | |
1226 | if (path->slots[0] >= nritems) { | |
1227 | ret = btrfs_next_leaf(root, path); | |
1228 | if (ret < 0) | |
1229 | err = ret; | |
1230 | if (ret) | |
1231 | goto fail; | |
1232 | ||
1233 | leaf = path->nodes[0]; | |
1234 | nritems = btrfs_header_nritems(leaf); | |
1235 | recow = 1; | |
1236 | } | |
1237 | ||
1238 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1239 | if (key.objectid != bytenr || | |
1240 | key.type != BTRFS_EXTENT_DATA_REF_KEY) | |
1241 | goto fail; | |
1242 | ||
1243 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1244 | struct btrfs_extent_data_ref); | |
1245 | ||
1246 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1247 | owner, offset)) { | |
1248 | if (recow) { | |
1249 | btrfs_release_path(path); | |
1250 | goto again; | |
1251 | } | |
1252 | err = 0; | |
1253 | break; | |
1254 | } | |
1255 | path->slots[0]++; | |
1256 | } | |
1257 | fail: | |
1258 | return err; | |
1259 | } | |
1260 | ||
1261 | static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans, | |
1262 | struct btrfs_root *root, | |
1263 | struct btrfs_path *path, | |
1264 | u64 bytenr, u64 parent, | |
1265 | u64 root_objectid, u64 owner, | |
1266 | u64 offset, int refs_to_add) | |
1267 | { | |
1268 | struct btrfs_key key; | |
1269 | struct extent_buffer *leaf; | |
1270 | u32 size; | |
1271 | u32 num_refs; | |
1272 | int ret; | |
1273 | ||
1274 | key.objectid = bytenr; | |
1275 | if (parent) { | |
1276 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1277 | key.offset = parent; | |
1278 | size = sizeof(struct btrfs_shared_data_ref); | |
1279 | } else { | |
1280 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1281 | key.offset = hash_extent_data_ref(root_objectid, | |
1282 | owner, offset); | |
1283 | size = sizeof(struct btrfs_extent_data_ref); | |
1284 | } | |
1285 | ||
1286 | ret = btrfs_insert_empty_item(trans, root, path, &key, size); | |
1287 | if (ret && ret != -EEXIST) | |
1288 | goto fail; | |
1289 | ||
1290 | leaf = path->nodes[0]; | |
1291 | if (parent) { | |
1292 | struct btrfs_shared_data_ref *ref; | |
1293 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1294 | struct btrfs_shared_data_ref); | |
1295 | if (ret == 0) { | |
1296 | btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add); | |
1297 | } else { | |
1298 | num_refs = btrfs_shared_data_ref_count(leaf, ref); | |
1299 | num_refs += refs_to_add; | |
1300 | btrfs_set_shared_data_ref_count(leaf, ref, num_refs); | |
1301 | } | |
1302 | } else { | |
1303 | struct btrfs_extent_data_ref *ref; | |
1304 | while (ret == -EEXIST) { | |
1305 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1306 | struct btrfs_extent_data_ref); | |
1307 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1308 | owner, offset)) | |
1309 | break; | |
1310 | btrfs_release_path(path); | |
1311 | key.offset++; | |
1312 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1313 | size); | |
1314 | if (ret && ret != -EEXIST) | |
1315 | goto fail; | |
1316 | ||
1317 | leaf = path->nodes[0]; | |
1318 | } | |
1319 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1320 | struct btrfs_extent_data_ref); | |
1321 | if (ret == 0) { | |
1322 | btrfs_set_extent_data_ref_root(leaf, ref, | |
1323 | root_objectid); | |
1324 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
1325 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
1326 | btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add); | |
1327 | } else { | |
1328 | num_refs = btrfs_extent_data_ref_count(leaf, ref); | |
1329 | num_refs += refs_to_add; | |
1330 | btrfs_set_extent_data_ref_count(leaf, ref, num_refs); | |
1331 | } | |
1332 | } | |
1333 | btrfs_mark_buffer_dirty(leaf); | |
1334 | ret = 0; | |
1335 | fail: | |
1336 | btrfs_release_path(path); | |
1337 | return ret; | |
1338 | } | |
1339 | ||
1340 | static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans, | |
1341 | struct btrfs_root *root, | |
1342 | struct btrfs_path *path, | |
1343 | int refs_to_drop, int *last_ref) | |
1344 | { | |
1345 | struct btrfs_key key; | |
1346 | struct btrfs_extent_data_ref *ref1 = NULL; | |
1347 | struct btrfs_shared_data_ref *ref2 = NULL; | |
1348 | struct extent_buffer *leaf; | |
1349 | u32 num_refs = 0; | |
1350 | int ret = 0; | |
1351 | ||
1352 | leaf = path->nodes[0]; | |
1353 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1354 | ||
1355 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1356 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1357 | struct btrfs_extent_data_ref); | |
1358 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1359 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1360 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1361 | struct btrfs_shared_data_ref); | |
1362 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1363 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1364 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
1365 | struct btrfs_extent_ref_v0 *ref0; | |
1366 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1367 | struct btrfs_extent_ref_v0); | |
1368 | num_refs = btrfs_ref_count_v0(leaf, ref0); | |
1369 | #endif | |
1370 | } else { | |
1371 | BUG(); | |
1372 | } | |
1373 | ||
1374 | BUG_ON(num_refs < refs_to_drop); | |
1375 | num_refs -= refs_to_drop; | |
1376 | ||
1377 | if (num_refs == 0) { | |
1378 | ret = btrfs_del_item(trans, root, path); | |
1379 | *last_ref = 1; | |
1380 | } else { | |
1381 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) | |
1382 | btrfs_set_extent_data_ref_count(leaf, ref1, num_refs); | |
1383 | else if (key.type == BTRFS_SHARED_DATA_REF_KEY) | |
1384 | btrfs_set_shared_data_ref_count(leaf, ref2, num_refs); | |
1385 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1386 | else { | |
1387 | struct btrfs_extent_ref_v0 *ref0; | |
1388 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1389 | struct btrfs_extent_ref_v0); | |
1390 | btrfs_set_ref_count_v0(leaf, ref0, num_refs); | |
1391 | } | |
1392 | #endif | |
1393 | btrfs_mark_buffer_dirty(leaf); | |
1394 | } | |
1395 | return ret; | |
1396 | } | |
1397 | ||
1398 | static noinline u32 extent_data_ref_count(struct btrfs_path *path, | |
1399 | struct btrfs_extent_inline_ref *iref) | |
1400 | { | |
1401 | struct btrfs_key key; | |
1402 | struct extent_buffer *leaf; | |
1403 | struct btrfs_extent_data_ref *ref1; | |
1404 | struct btrfs_shared_data_ref *ref2; | |
1405 | u32 num_refs = 0; | |
1406 | ||
1407 | leaf = path->nodes[0]; | |
1408 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1409 | if (iref) { | |
1410 | if (btrfs_extent_inline_ref_type(leaf, iref) == | |
1411 | BTRFS_EXTENT_DATA_REF_KEY) { | |
1412 | ref1 = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1413 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1414 | } else { | |
1415 | ref2 = (struct btrfs_shared_data_ref *)(iref + 1); | |
1416 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1417 | } | |
1418 | } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1419 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1420 | struct btrfs_extent_data_ref); | |
1421 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1422 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1423 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1424 | struct btrfs_shared_data_ref); | |
1425 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1426 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1427 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
1428 | struct btrfs_extent_ref_v0 *ref0; | |
1429 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1430 | struct btrfs_extent_ref_v0); | |
1431 | num_refs = btrfs_ref_count_v0(leaf, ref0); | |
1432 | #endif | |
1433 | } else { | |
1434 | WARN_ON(1); | |
1435 | } | |
1436 | return num_refs; | |
1437 | } | |
1438 | ||
1439 | static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans, | |
1440 | struct btrfs_root *root, | |
1441 | struct btrfs_path *path, | |
1442 | u64 bytenr, u64 parent, | |
1443 | u64 root_objectid) | |
1444 | { | |
1445 | struct btrfs_key key; | |
1446 | int ret; | |
1447 | ||
1448 | key.objectid = bytenr; | |
1449 | if (parent) { | |
1450 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1451 | key.offset = parent; | |
1452 | } else { | |
1453 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1454 | key.offset = root_objectid; | |
1455 | } | |
1456 | ||
1457 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1458 | if (ret > 0) | |
1459 | ret = -ENOENT; | |
1460 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1461 | if (ret == -ENOENT && parent) { | |
1462 | btrfs_release_path(path); | |
1463 | key.type = BTRFS_EXTENT_REF_V0_KEY; | |
1464 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1465 | if (ret > 0) | |
1466 | ret = -ENOENT; | |
1467 | } | |
1468 | #endif | |
1469 | return ret; | |
1470 | } | |
1471 | ||
1472 | static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans, | |
1473 | struct btrfs_root *root, | |
1474 | struct btrfs_path *path, | |
1475 | u64 bytenr, u64 parent, | |
1476 | u64 root_objectid) | |
1477 | { | |
1478 | struct btrfs_key key; | |
1479 | int ret; | |
1480 | ||
1481 | key.objectid = bytenr; | |
1482 | if (parent) { | |
1483 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1484 | key.offset = parent; | |
1485 | } else { | |
1486 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1487 | key.offset = root_objectid; | |
1488 | } | |
1489 | ||
1490 | ret = btrfs_insert_empty_item(trans, root, path, &key, 0); | |
1491 | btrfs_release_path(path); | |
1492 | return ret; | |
1493 | } | |
1494 | ||
1495 | static inline int extent_ref_type(u64 parent, u64 owner) | |
1496 | { | |
1497 | int type; | |
1498 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1499 | if (parent > 0) | |
1500 | type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1501 | else | |
1502 | type = BTRFS_TREE_BLOCK_REF_KEY; | |
1503 | } else { | |
1504 | if (parent > 0) | |
1505 | type = BTRFS_SHARED_DATA_REF_KEY; | |
1506 | else | |
1507 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
1508 | } | |
1509 | return type; | |
1510 | } | |
1511 | ||
1512 | static int find_next_key(struct btrfs_path *path, int level, | |
1513 | struct btrfs_key *key) | |
1514 | ||
1515 | { | |
1516 | for (; level < BTRFS_MAX_LEVEL; level++) { | |
1517 | if (!path->nodes[level]) | |
1518 | break; | |
1519 | if (path->slots[level] + 1 >= | |
1520 | btrfs_header_nritems(path->nodes[level])) | |
1521 | continue; | |
1522 | if (level == 0) | |
1523 | btrfs_item_key_to_cpu(path->nodes[level], key, | |
1524 | path->slots[level] + 1); | |
1525 | else | |
1526 | btrfs_node_key_to_cpu(path->nodes[level], key, | |
1527 | path->slots[level] + 1); | |
1528 | return 0; | |
1529 | } | |
1530 | return 1; | |
1531 | } | |
1532 | ||
1533 | /* | |
1534 | * look for inline back ref. if back ref is found, *ref_ret is set | |
1535 | * to the address of inline back ref, and 0 is returned. | |
1536 | * | |
1537 | * if back ref isn't found, *ref_ret is set to the address where it | |
1538 | * should be inserted, and -ENOENT is returned. | |
1539 | * | |
1540 | * if insert is true and there are too many inline back refs, the path | |
1541 | * points to the extent item, and -EAGAIN is returned. | |
1542 | * | |
1543 | * NOTE: inline back refs are ordered in the same way that back ref | |
1544 | * items in the tree are ordered. | |
1545 | */ | |
1546 | static noinline_for_stack | |
1547 | int lookup_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1548 | struct btrfs_root *root, | |
1549 | struct btrfs_path *path, | |
1550 | struct btrfs_extent_inline_ref **ref_ret, | |
1551 | u64 bytenr, u64 num_bytes, | |
1552 | u64 parent, u64 root_objectid, | |
1553 | u64 owner, u64 offset, int insert) | |
1554 | { | |
1555 | struct btrfs_key key; | |
1556 | struct extent_buffer *leaf; | |
1557 | struct btrfs_extent_item *ei; | |
1558 | struct btrfs_extent_inline_ref *iref; | |
1559 | u64 flags; | |
1560 | u64 item_size; | |
1561 | unsigned long ptr; | |
1562 | unsigned long end; | |
1563 | int extra_size; | |
1564 | int type; | |
1565 | int want; | |
1566 | int ret; | |
1567 | int err = 0; | |
1568 | bool skinny_metadata = btrfs_fs_incompat(root->fs_info, | |
1569 | SKINNY_METADATA); | |
1570 | ||
1571 | key.objectid = bytenr; | |
1572 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
1573 | key.offset = num_bytes; | |
1574 | ||
1575 | want = extent_ref_type(parent, owner); | |
1576 | if (insert) { | |
1577 | extra_size = btrfs_extent_inline_ref_size(want); | |
1578 | path->keep_locks = 1; | |
1579 | } else | |
1580 | extra_size = -1; | |
1581 | ||
1582 | /* | |
1583 | * Owner is our parent level, so we can just add one to get the level | |
1584 | * for the block we are interested in. | |
1585 | */ | |
1586 | if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1587 | key.type = BTRFS_METADATA_ITEM_KEY; | |
1588 | key.offset = owner; | |
1589 | } | |
1590 | ||
1591 | again: | |
1592 | ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1); | |
1593 | if (ret < 0) { | |
1594 | err = ret; | |
1595 | goto out; | |
1596 | } | |
1597 | ||
1598 | /* | |
1599 | * We may be a newly converted file system which still has the old fat | |
1600 | * extent entries for metadata, so try and see if we have one of those. | |
1601 | */ | |
1602 | if (ret > 0 && skinny_metadata) { | |
1603 | skinny_metadata = false; | |
1604 | if (path->slots[0]) { | |
1605 | path->slots[0]--; | |
1606 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
1607 | path->slots[0]); | |
1608 | if (key.objectid == bytenr && | |
1609 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
1610 | key.offset == num_bytes) | |
1611 | ret = 0; | |
1612 | } | |
1613 | if (ret) { | |
1614 | key.objectid = bytenr; | |
1615 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
1616 | key.offset = num_bytes; | |
1617 | btrfs_release_path(path); | |
1618 | goto again; | |
1619 | } | |
1620 | } | |
1621 | ||
1622 | if (ret && !insert) { | |
1623 | err = -ENOENT; | |
1624 | goto out; | |
1625 | } else if (WARN_ON(ret)) { | |
1626 | err = -EIO; | |
1627 | goto out; | |
1628 | } | |
1629 | ||
1630 | leaf = path->nodes[0]; | |
1631 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1632 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1633 | if (item_size < sizeof(*ei)) { | |
1634 | if (!insert) { | |
1635 | err = -ENOENT; | |
1636 | goto out; | |
1637 | } | |
1638 | ret = convert_extent_item_v0(trans, root, path, owner, | |
1639 | extra_size); | |
1640 | if (ret < 0) { | |
1641 | err = ret; | |
1642 | goto out; | |
1643 | } | |
1644 | leaf = path->nodes[0]; | |
1645 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1646 | } | |
1647 | #endif | |
1648 | BUG_ON(item_size < sizeof(*ei)); | |
1649 | ||
1650 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1651 | flags = btrfs_extent_flags(leaf, ei); | |
1652 | ||
1653 | ptr = (unsigned long)(ei + 1); | |
1654 | end = (unsigned long)ei + item_size; | |
1655 | ||
1656 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) { | |
1657 | ptr += sizeof(struct btrfs_tree_block_info); | |
1658 | BUG_ON(ptr > end); | |
1659 | } | |
1660 | ||
1661 | err = -ENOENT; | |
1662 | while (1) { | |
1663 | if (ptr >= end) { | |
1664 | WARN_ON(ptr > end); | |
1665 | break; | |
1666 | } | |
1667 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1668 | type = btrfs_extent_inline_ref_type(leaf, iref); | |
1669 | if (want < type) | |
1670 | break; | |
1671 | if (want > type) { | |
1672 | ptr += btrfs_extent_inline_ref_size(type); | |
1673 | continue; | |
1674 | } | |
1675 | ||
1676 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1677 | struct btrfs_extent_data_ref *dref; | |
1678 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1679 | if (match_extent_data_ref(leaf, dref, root_objectid, | |
1680 | owner, offset)) { | |
1681 | err = 0; | |
1682 | break; | |
1683 | } | |
1684 | if (hash_extent_data_ref_item(leaf, dref) < | |
1685 | hash_extent_data_ref(root_objectid, owner, offset)) | |
1686 | break; | |
1687 | } else { | |
1688 | u64 ref_offset; | |
1689 | ref_offset = btrfs_extent_inline_ref_offset(leaf, iref); | |
1690 | if (parent > 0) { | |
1691 | if (parent == ref_offset) { | |
1692 | err = 0; | |
1693 | break; | |
1694 | } | |
1695 | if (ref_offset < parent) | |
1696 | break; | |
1697 | } else { | |
1698 | if (root_objectid == ref_offset) { | |
1699 | err = 0; | |
1700 | break; | |
1701 | } | |
1702 | if (ref_offset < root_objectid) | |
1703 | break; | |
1704 | } | |
1705 | } | |
1706 | ptr += btrfs_extent_inline_ref_size(type); | |
1707 | } | |
1708 | if (err == -ENOENT && insert) { | |
1709 | if (item_size + extra_size >= | |
1710 | BTRFS_MAX_EXTENT_ITEM_SIZE(root)) { | |
1711 | err = -EAGAIN; | |
1712 | goto out; | |
1713 | } | |
1714 | /* | |
1715 | * To add new inline back ref, we have to make sure | |
1716 | * there is no corresponding back ref item. | |
1717 | * For simplicity, we just do not add new inline back | |
1718 | * ref if there is any kind of item for this block | |
1719 | */ | |
1720 | if (find_next_key(path, 0, &key) == 0 && | |
1721 | key.objectid == bytenr && | |
1722 | key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) { | |
1723 | err = -EAGAIN; | |
1724 | goto out; | |
1725 | } | |
1726 | } | |
1727 | *ref_ret = (struct btrfs_extent_inline_ref *)ptr; | |
1728 | out: | |
1729 | if (insert) { | |
1730 | path->keep_locks = 0; | |
1731 | btrfs_unlock_up_safe(path, 1); | |
1732 | } | |
1733 | return err; | |
1734 | } | |
1735 | ||
1736 | /* | |
1737 | * helper to add new inline back ref | |
1738 | */ | |
1739 | static noinline_for_stack | |
1740 | void setup_inline_extent_backref(struct btrfs_root *root, | |
1741 | struct btrfs_path *path, | |
1742 | struct btrfs_extent_inline_ref *iref, | |
1743 | u64 parent, u64 root_objectid, | |
1744 | u64 owner, u64 offset, int refs_to_add, | |
1745 | struct btrfs_delayed_extent_op *extent_op) | |
1746 | { | |
1747 | struct extent_buffer *leaf; | |
1748 | struct btrfs_extent_item *ei; | |
1749 | unsigned long ptr; | |
1750 | unsigned long end; | |
1751 | unsigned long item_offset; | |
1752 | u64 refs; | |
1753 | int size; | |
1754 | int type; | |
1755 | ||
1756 | leaf = path->nodes[0]; | |
1757 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1758 | item_offset = (unsigned long)iref - (unsigned long)ei; | |
1759 | ||
1760 | type = extent_ref_type(parent, owner); | |
1761 | size = btrfs_extent_inline_ref_size(type); | |
1762 | ||
1763 | btrfs_extend_item(root, path, size); | |
1764 | ||
1765 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1766 | refs = btrfs_extent_refs(leaf, ei); | |
1767 | refs += refs_to_add; | |
1768 | btrfs_set_extent_refs(leaf, ei, refs); | |
1769 | if (extent_op) | |
1770 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1771 | ||
1772 | ptr = (unsigned long)ei + item_offset; | |
1773 | end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]); | |
1774 | if (ptr < end - size) | |
1775 | memmove_extent_buffer(leaf, ptr + size, ptr, | |
1776 | end - size - ptr); | |
1777 | ||
1778 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1779 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
1780 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1781 | struct btrfs_extent_data_ref *dref; | |
1782 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1783 | btrfs_set_extent_data_ref_root(leaf, dref, root_objectid); | |
1784 | btrfs_set_extent_data_ref_objectid(leaf, dref, owner); | |
1785 | btrfs_set_extent_data_ref_offset(leaf, dref, offset); | |
1786 | btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add); | |
1787 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1788 | struct btrfs_shared_data_ref *sref; | |
1789 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1790 | btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add); | |
1791 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1792 | } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) { | |
1793 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1794 | } else { | |
1795 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
1796 | } | |
1797 | btrfs_mark_buffer_dirty(leaf); | |
1798 | } | |
1799 | ||
1800 | static int lookup_extent_backref(struct btrfs_trans_handle *trans, | |
1801 | struct btrfs_root *root, | |
1802 | struct btrfs_path *path, | |
1803 | struct btrfs_extent_inline_ref **ref_ret, | |
1804 | u64 bytenr, u64 num_bytes, u64 parent, | |
1805 | u64 root_objectid, u64 owner, u64 offset) | |
1806 | { | |
1807 | int ret; | |
1808 | ||
1809 | ret = lookup_inline_extent_backref(trans, root, path, ref_ret, | |
1810 | bytenr, num_bytes, parent, | |
1811 | root_objectid, owner, offset, 0); | |
1812 | if (ret != -ENOENT) | |
1813 | return ret; | |
1814 | ||
1815 | btrfs_release_path(path); | |
1816 | *ref_ret = NULL; | |
1817 | ||
1818 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1819 | ret = lookup_tree_block_ref(trans, root, path, bytenr, parent, | |
1820 | root_objectid); | |
1821 | } else { | |
1822 | ret = lookup_extent_data_ref(trans, root, path, bytenr, parent, | |
1823 | root_objectid, owner, offset); | |
1824 | } | |
1825 | return ret; | |
1826 | } | |
1827 | ||
1828 | /* | |
1829 | * helper to update/remove inline back ref | |
1830 | */ | |
1831 | static noinline_for_stack | |
1832 | void update_inline_extent_backref(struct btrfs_root *root, | |
1833 | struct btrfs_path *path, | |
1834 | struct btrfs_extent_inline_ref *iref, | |
1835 | int refs_to_mod, | |
1836 | struct btrfs_delayed_extent_op *extent_op, | |
1837 | int *last_ref) | |
1838 | { | |
1839 | struct extent_buffer *leaf; | |
1840 | struct btrfs_extent_item *ei; | |
1841 | struct btrfs_extent_data_ref *dref = NULL; | |
1842 | struct btrfs_shared_data_ref *sref = NULL; | |
1843 | unsigned long ptr; | |
1844 | unsigned long end; | |
1845 | u32 item_size; | |
1846 | int size; | |
1847 | int type; | |
1848 | u64 refs; | |
1849 | ||
1850 | leaf = path->nodes[0]; | |
1851 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1852 | refs = btrfs_extent_refs(leaf, ei); | |
1853 | WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0); | |
1854 | refs += refs_to_mod; | |
1855 | btrfs_set_extent_refs(leaf, ei, refs); | |
1856 | if (extent_op) | |
1857 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1858 | ||
1859 | type = btrfs_extent_inline_ref_type(leaf, iref); | |
1860 | ||
1861 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1862 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1863 | refs = btrfs_extent_data_ref_count(leaf, dref); | |
1864 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1865 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1866 | refs = btrfs_shared_data_ref_count(leaf, sref); | |
1867 | } else { | |
1868 | refs = 1; | |
1869 | BUG_ON(refs_to_mod != -1); | |
1870 | } | |
1871 | ||
1872 | BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod); | |
1873 | refs += refs_to_mod; | |
1874 | ||
1875 | if (refs > 0) { | |
1876 | if (type == BTRFS_EXTENT_DATA_REF_KEY) | |
1877 | btrfs_set_extent_data_ref_count(leaf, dref, refs); | |
1878 | else | |
1879 | btrfs_set_shared_data_ref_count(leaf, sref, refs); | |
1880 | } else { | |
1881 | *last_ref = 1; | |
1882 | size = btrfs_extent_inline_ref_size(type); | |
1883 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1884 | ptr = (unsigned long)iref; | |
1885 | end = (unsigned long)ei + item_size; | |
1886 | if (ptr + size < end) | |
1887 | memmove_extent_buffer(leaf, ptr, ptr + size, | |
1888 | end - ptr - size); | |
1889 | item_size -= size; | |
1890 | btrfs_truncate_item(root, path, item_size, 1); | |
1891 | } | |
1892 | btrfs_mark_buffer_dirty(leaf); | |
1893 | } | |
1894 | ||
1895 | static noinline_for_stack | |
1896 | int insert_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1897 | struct btrfs_root *root, | |
1898 | struct btrfs_path *path, | |
1899 | u64 bytenr, u64 num_bytes, u64 parent, | |
1900 | u64 root_objectid, u64 owner, | |
1901 | u64 offset, int refs_to_add, | |
1902 | struct btrfs_delayed_extent_op *extent_op) | |
1903 | { | |
1904 | struct btrfs_extent_inline_ref *iref; | |
1905 | int ret; | |
1906 | ||
1907 | ret = lookup_inline_extent_backref(trans, root, path, &iref, | |
1908 | bytenr, num_bytes, parent, | |
1909 | root_objectid, owner, offset, 1); | |
1910 | if (ret == 0) { | |
1911 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID); | |
1912 | update_inline_extent_backref(root, path, iref, | |
1913 | refs_to_add, extent_op, NULL); | |
1914 | } else if (ret == -ENOENT) { | |
1915 | setup_inline_extent_backref(root, path, iref, parent, | |
1916 | root_objectid, owner, offset, | |
1917 | refs_to_add, extent_op); | |
1918 | ret = 0; | |
1919 | } | |
1920 | return ret; | |
1921 | } | |
1922 | ||
1923 | static int insert_extent_backref(struct btrfs_trans_handle *trans, | |
1924 | struct btrfs_root *root, | |
1925 | struct btrfs_path *path, | |
1926 | u64 bytenr, u64 parent, u64 root_objectid, | |
1927 | u64 owner, u64 offset, int refs_to_add) | |
1928 | { | |
1929 | int ret; | |
1930 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1931 | BUG_ON(refs_to_add != 1); | |
1932 | ret = insert_tree_block_ref(trans, root, path, bytenr, | |
1933 | parent, root_objectid); | |
1934 | } else { | |
1935 | ret = insert_extent_data_ref(trans, root, path, bytenr, | |
1936 | parent, root_objectid, | |
1937 | owner, offset, refs_to_add); | |
1938 | } | |
1939 | return ret; | |
1940 | } | |
1941 | ||
1942 | static int remove_extent_backref(struct btrfs_trans_handle *trans, | |
1943 | struct btrfs_root *root, | |
1944 | struct btrfs_path *path, | |
1945 | struct btrfs_extent_inline_ref *iref, | |
1946 | int refs_to_drop, int is_data, int *last_ref) | |
1947 | { | |
1948 | int ret = 0; | |
1949 | ||
1950 | BUG_ON(!is_data && refs_to_drop != 1); | |
1951 | if (iref) { | |
1952 | update_inline_extent_backref(root, path, iref, | |
1953 | -refs_to_drop, NULL, last_ref); | |
1954 | } else if (is_data) { | |
1955 | ret = remove_extent_data_ref(trans, root, path, refs_to_drop, | |
1956 | last_ref); | |
1957 | } else { | |
1958 | *last_ref = 1; | |
1959 | ret = btrfs_del_item(trans, root, path); | |
1960 | } | |
1961 | return ret; | |
1962 | } | |
1963 | ||
1964 | #define in_range(b, first, len) ((b) >= (first) && (b) < (first) + (len)) | |
1965 | static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len, | |
1966 | u64 *discarded_bytes) | |
1967 | { | |
1968 | int j, ret = 0; | |
1969 | u64 bytes_left, end; | |
1970 | u64 aligned_start = ALIGN(start, 1 << 9); | |
1971 | ||
1972 | if (WARN_ON(start != aligned_start)) { | |
1973 | len -= aligned_start - start; | |
1974 | len = round_down(len, 1 << 9); | |
1975 | start = aligned_start; | |
1976 | } | |
1977 | ||
1978 | *discarded_bytes = 0; | |
1979 | ||
1980 | if (!len) | |
1981 | return 0; | |
1982 | ||
1983 | end = start + len; | |
1984 | bytes_left = len; | |
1985 | ||
1986 | /* Skip any superblocks on this device. */ | |
1987 | for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) { | |
1988 | u64 sb_start = btrfs_sb_offset(j); | |
1989 | u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE; | |
1990 | u64 size = sb_start - start; | |
1991 | ||
1992 | if (!in_range(sb_start, start, bytes_left) && | |
1993 | !in_range(sb_end, start, bytes_left) && | |
1994 | !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE)) | |
1995 | continue; | |
1996 | ||
1997 | /* | |
1998 | * Superblock spans beginning of range. Adjust start and | |
1999 | * try again. | |
2000 | */ | |
2001 | if (sb_start <= start) { | |
2002 | start += sb_end - start; | |
2003 | if (start > end) { | |
2004 | bytes_left = 0; | |
2005 | break; | |
2006 | } | |
2007 | bytes_left = end - start; | |
2008 | continue; | |
2009 | } | |
2010 | ||
2011 | if (size) { | |
2012 | ret = blkdev_issue_discard(bdev, start >> 9, size >> 9, | |
2013 | GFP_NOFS, 0); | |
2014 | if (!ret) | |
2015 | *discarded_bytes += size; | |
2016 | else if (ret != -EOPNOTSUPP) | |
2017 | return ret; | |
2018 | } | |
2019 | ||
2020 | start = sb_end; | |
2021 | if (start > end) { | |
2022 | bytes_left = 0; | |
2023 | break; | |
2024 | } | |
2025 | bytes_left = end - start; | |
2026 | } | |
2027 | ||
2028 | if (bytes_left) { | |
2029 | ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9, | |
2030 | GFP_NOFS, 0); | |
2031 | if (!ret) | |
2032 | *discarded_bytes += bytes_left; | |
2033 | } | |
2034 | return ret; | |
2035 | } | |
2036 | ||
2037 | int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr, | |
2038 | u64 num_bytes, u64 *actual_bytes) | |
2039 | { | |
2040 | int ret; | |
2041 | u64 discarded_bytes = 0; | |
2042 | struct btrfs_bio *bbio = NULL; | |
2043 | ||
2044 | ||
2045 | /* Tell the block device(s) that the sectors can be discarded */ | |
2046 | ret = btrfs_map_block(root->fs_info, REQ_DISCARD, | |
2047 | bytenr, &num_bytes, &bbio, 0); | |
2048 | /* Error condition is -ENOMEM */ | |
2049 | if (!ret) { | |
2050 | struct btrfs_bio_stripe *stripe = bbio->stripes; | |
2051 | int i; | |
2052 | ||
2053 | ||
2054 | for (i = 0; i < bbio->num_stripes; i++, stripe++) { | |
2055 | u64 bytes; | |
2056 | if (!stripe->dev->can_discard) | |
2057 | continue; | |
2058 | ||
2059 | ret = btrfs_issue_discard(stripe->dev->bdev, | |
2060 | stripe->physical, | |
2061 | stripe->length, | |
2062 | &bytes); | |
2063 | if (!ret) | |
2064 | discarded_bytes += bytes; | |
2065 | else if (ret != -EOPNOTSUPP) | |
2066 | break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */ | |
2067 | ||
2068 | /* | |
2069 | * Just in case we get back EOPNOTSUPP for some reason, | |
2070 | * just ignore the return value so we don't screw up | |
2071 | * people calling discard_extent. | |
2072 | */ | |
2073 | ret = 0; | |
2074 | } | |
2075 | btrfs_put_bbio(bbio); | |
2076 | } | |
2077 | ||
2078 | if (actual_bytes) | |
2079 | *actual_bytes = discarded_bytes; | |
2080 | ||
2081 | ||
2082 | if (ret == -EOPNOTSUPP) | |
2083 | ret = 0; | |
2084 | return ret; | |
2085 | } | |
2086 | ||
2087 | /* Can return -ENOMEM */ | |
2088 | int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, | |
2089 | struct btrfs_root *root, | |
2090 | u64 bytenr, u64 num_bytes, u64 parent, | |
2091 | u64 root_objectid, u64 owner, u64 offset) | |
2092 | { | |
2093 | int ret; | |
2094 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2095 | ||
2096 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID && | |
2097 | root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
2098 | ||
2099 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
2100 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr, | |
2101 | num_bytes, | |
2102 | parent, root_objectid, (int)owner, | |
2103 | BTRFS_ADD_DELAYED_REF, NULL); | |
2104 | } else { | |
2105 | ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr, | |
2106 | num_bytes, parent, root_objectid, | |
2107 | owner, offset, 0, | |
2108 | BTRFS_ADD_DELAYED_REF, NULL); | |
2109 | } | |
2110 | return ret; | |
2111 | } | |
2112 | ||
2113 | static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, | |
2114 | struct btrfs_root *root, | |
2115 | struct btrfs_delayed_ref_node *node, | |
2116 | u64 parent, u64 root_objectid, | |
2117 | u64 owner, u64 offset, int refs_to_add, | |
2118 | struct btrfs_delayed_extent_op *extent_op) | |
2119 | { | |
2120 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2121 | struct btrfs_path *path; | |
2122 | struct extent_buffer *leaf; | |
2123 | struct btrfs_extent_item *item; | |
2124 | struct btrfs_key key; | |
2125 | u64 bytenr = node->bytenr; | |
2126 | u64 num_bytes = node->num_bytes; | |
2127 | u64 refs; | |
2128 | int ret; | |
2129 | ||
2130 | path = btrfs_alloc_path(); | |
2131 | if (!path) | |
2132 | return -ENOMEM; | |
2133 | ||
2134 | path->reada = READA_FORWARD; | |
2135 | path->leave_spinning = 1; | |
2136 | /* this will setup the path even if it fails to insert the back ref */ | |
2137 | ret = insert_inline_extent_backref(trans, fs_info->extent_root, path, | |
2138 | bytenr, num_bytes, parent, | |
2139 | root_objectid, owner, offset, | |
2140 | refs_to_add, extent_op); | |
2141 | if ((ret < 0 && ret != -EAGAIN) || !ret) | |
2142 | goto out; | |
2143 | ||
2144 | /* | |
2145 | * Ok we had -EAGAIN which means we didn't have space to insert and | |
2146 | * inline extent ref, so just update the reference count and add a | |
2147 | * normal backref. | |
2148 | */ | |
2149 | leaf = path->nodes[0]; | |
2150 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
2151 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
2152 | refs = btrfs_extent_refs(leaf, item); | |
2153 | btrfs_set_extent_refs(leaf, item, refs + refs_to_add); | |
2154 | if (extent_op) | |
2155 | __run_delayed_extent_op(extent_op, leaf, item); | |
2156 | ||
2157 | btrfs_mark_buffer_dirty(leaf); | |
2158 | btrfs_release_path(path); | |
2159 | ||
2160 | path->reada = READA_FORWARD; | |
2161 | path->leave_spinning = 1; | |
2162 | /* now insert the actual backref */ | |
2163 | ret = insert_extent_backref(trans, root->fs_info->extent_root, | |
2164 | path, bytenr, parent, root_objectid, | |
2165 | owner, offset, refs_to_add); | |
2166 | if (ret) | |
2167 | btrfs_abort_transaction(trans, root, ret); | |
2168 | out: | |
2169 | btrfs_free_path(path); | |
2170 | return ret; | |
2171 | } | |
2172 | ||
2173 | static int run_delayed_data_ref(struct btrfs_trans_handle *trans, | |
2174 | struct btrfs_root *root, | |
2175 | struct btrfs_delayed_ref_node *node, | |
2176 | struct btrfs_delayed_extent_op *extent_op, | |
2177 | int insert_reserved) | |
2178 | { | |
2179 | int ret = 0; | |
2180 | struct btrfs_delayed_data_ref *ref; | |
2181 | struct btrfs_key ins; | |
2182 | u64 parent = 0; | |
2183 | u64 ref_root = 0; | |
2184 | u64 flags = 0; | |
2185 | ||
2186 | ins.objectid = node->bytenr; | |
2187 | ins.offset = node->num_bytes; | |
2188 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
2189 | ||
2190 | ref = btrfs_delayed_node_to_data_ref(node); | |
2191 | trace_run_delayed_data_ref(node, ref, node->action); | |
2192 | ||
2193 | if (node->type == BTRFS_SHARED_DATA_REF_KEY) | |
2194 | parent = ref->parent; | |
2195 | ref_root = ref->root; | |
2196 | ||
2197 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
2198 | if (extent_op) | |
2199 | flags |= extent_op->flags_to_set; | |
2200 | ret = alloc_reserved_file_extent(trans, root, | |
2201 | parent, ref_root, flags, | |
2202 | ref->objectid, ref->offset, | |
2203 | &ins, node->ref_mod); | |
2204 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { | |
2205 | ret = __btrfs_inc_extent_ref(trans, root, node, parent, | |
2206 | ref_root, ref->objectid, | |
2207 | ref->offset, node->ref_mod, | |
2208 | extent_op); | |
2209 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { | |
2210 | ret = __btrfs_free_extent(trans, root, node, parent, | |
2211 | ref_root, ref->objectid, | |
2212 | ref->offset, node->ref_mod, | |
2213 | extent_op); | |
2214 | } else { | |
2215 | BUG(); | |
2216 | } | |
2217 | return ret; | |
2218 | } | |
2219 | ||
2220 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
2221 | struct extent_buffer *leaf, | |
2222 | struct btrfs_extent_item *ei) | |
2223 | { | |
2224 | u64 flags = btrfs_extent_flags(leaf, ei); | |
2225 | if (extent_op->update_flags) { | |
2226 | flags |= extent_op->flags_to_set; | |
2227 | btrfs_set_extent_flags(leaf, ei, flags); | |
2228 | } | |
2229 | ||
2230 | if (extent_op->update_key) { | |
2231 | struct btrfs_tree_block_info *bi; | |
2232 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)); | |
2233 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
2234 | btrfs_set_tree_block_key(leaf, bi, &extent_op->key); | |
2235 | } | |
2236 | } | |
2237 | ||
2238 | static int run_delayed_extent_op(struct btrfs_trans_handle *trans, | |
2239 | struct btrfs_root *root, | |
2240 | struct btrfs_delayed_ref_node *node, | |
2241 | struct btrfs_delayed_extent_op *extent_op) | |
2242 | { | |
2243 | struct btrfs_key key; | |
2244 | struct btrfs_path *path; | |
2245 | struct btrfs_extent_item *ei; | |
2246 | struct extent_buffer *leaf; | |
2247 | u32 item_size; | |
2248 | int ret; | |
2249 | int err = 0; | |
2250 | int metadata = !extent_op->is_data; | |
2251 | ||
2252 | if (trans->aborted) | |
2253 | return 0; | |
2254 | ||
2255 | if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) | |
2256 | metadata = 0; | |
2257 | ||
2258 | path = btrfs_alloc_path(); | |
2259 | if (!path) | |
2260 | return -ENOMEM; | |
2261 | ||
2262 | key.objectid = node->bytenr; | |
2263 | ||
2264 | if (metadata) { | |
2265 | key.type = BTRFS_METADATA_ITEM_KEY; | |
2266 | key.offset = extent_op->level; | |
2267 | } else { | |
2268 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
2269 | key.offset = node->num_bytes; | |
2270 | } | |
2271 | ||
2272 | again: | |
2273 | path->reada = READA_FORWARD; | |
2274 | path->leave_spinning = 1; | |
2275 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, | |
2276 | path, 0, 1); | |
2277 | if (ret < 0) { | |
2278 | err = ret; | |
2279 | goto out; | |
2280 | } | |
2281 | if (ret > 0) { | |
2282 | if (metadata) { | |
2283 | if (path->slots[0] > 0) { | |
2284 | path->slots[0]--; | |
2285 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
2286 | path->slots[0]); | |
2287 | if (key.objectid == node->bytenr && | |
2288 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
2289 | key.offset == node->num_bytes) | |
2290 | ret = 0; | |
2291 | } | |
2292 | if (ret > 0) { | |
2293 | btrfs_release_path(path); | |
2294 | metadata = 0; | |
2295 | ||
2296 | key.objectid = node->bytenr; | |
2297 | key.offset = node->num_bytes; | |
2298 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
2299 | goto again; | |
2300 | } | |
2301 | } else { | |
2302 | err = -EIO; | |
2303 | goto out; | |
2304 | } | |
2305 | } | |
2306 | ||
2307 | leaf = path->nodes[0]; | |
2308 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2309 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
2310 | if (item_size < sizeof(*ei)) { | |
2311 | ret = convert_extent_item_v0(trans, root->fs_info->extent_root, | |
2312 | path, (u64)-1, 0); | |
2313 | if (ret < 0) { | |
2314 | err = ret; | |
2315 | goto out; | |
2316 | } | |
2317 | leaf = path->nodes[0]; | |
2318 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2319 | } | |
2320 | #endif | |
2321 | BUG_ON(item_size < sizeof(*ei)); | |
2322 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
2323 | __run_delayed_extent_op(extent_op, leaf, ei); | |
2324 | ||
2325 | btrfs_mark_buffer_dirty(leaf); | |
2326 | out: | |
2327 | btrfs_free_path(path); | |
2328 | return err; | |
2329 | } | |
2330 | ||
2331 | static int run_delayed_tree_ref(struct btrfs_trans_handle *trans, | |
2332 | struct btrfs_root *root, | |
2333 | struct btrfs_delayed_ref_node *node, | |
2334 | struct btrfs_delayed_extent_op *extent_op, | |
2335 | int insert_reserved) | |
2336 | { | |
2337 | int ret = 0; | |
2338 | struct btrfs_delayed_tree_ref *ref; | |
2339 | struct btrfs_key ins; | |
2340 | u64 parent = 0; | |
2341 | u64 ref_root = 0; | |
2342 | bool skinny_metadata = btrfs_fs_incompat(root->fs_info, | |
2343 | SKINNY_METADATA); | |
2344 | ||
2345 | ref = btrfs_delayed_node_to_tree_ref(node); | |
2346 | trace_run_delayed_tree_ref(node, ref, node->action); | |
2347 | ||
2348 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
2349 | parent = ref->parent; | |
2350 | ref_root = ref->root; | |
2351 | ||
2352 | ins.objectid = node->bytenr; | |
2353 | if (skinny_metadata) { | |
2354 | ins.offset = ref->level; | |
2355 | ins.type = BTRFS_METADATA_ITEM_KEY; | |
2356 | } else { | |
2357 | ins.offset = node->num_bytes; | |
2358 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
2359 | } | |
2360 | ||
2361 | BUG_ON(node->ref_mod != 1); | |
2362 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
2363 | BUG_ON(!extent_op || !extent_op->update_flags); | |
2364 | ret = alloc_reserved_tree_block(trans, root, | |
2365 | parent, ref_root, | |
2366 | extent_op->flags_to_set, | |
2367 | &extent_op->key, | |
2368 | ref->level, &ins); | |
2369 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { | |
2370 | ret = __btrfs_inc_extent_ref(trans, root, node, | |
2371 | parent, ref_root, | |
2372 | ref->level, 0, 1, | |
2373 | extent_op); | |
2374 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { | |
2375 | ret = __btrfs_free_extent(trans, root, node, | |
2376 | parent, ref_root, | |
2377 | ref->level, 0, 1, extent_op); | |
2378 | } else { | |
2379 | BUG(); | |
2380 | } | |
2381 | return ret; | |
2382 | } | |
2383 | ||
2384 | /* helper function to actually process a single delayed ref entry */ | |
2385 | static int run_one_delayed_ref(struct btrfs_trans_handle *trans, | |
2386 | struct btrfs_root *root, | |
2387 | struct btrfs_delayed_ref_node *node, | |
2388 | struct btrfs_delayed_extent_op *extent_op, | |
2389 | int insert_reserved) | |
2390 | { | |
2391 | int ret = 0; | |
2392 | ||
2393 | if (trans->aborted) { | |
2394 | if (insert_reserved) | |
2395 | btrfs_pin_extent(root, node->bytenr, | |
2396 | node->num_bytes, 1); | |
2397 | return 0; | |
2398 | } | |
2399 | ||
2400 | if (btrfs_delayed_ref_is_head(node)) { | |
2401 | struct btrfs_delayed_ref_head *head; | |
2402 | /* | |
2403 | * we've hit the end of the chain and we were supposed | |
2404 | * to insert this extent into the tree. But, it got | |
2405 | * deleted before we ever needed to insert it, so all | |
2406 | * we have to do is clean up the accounting | |
2407 | */ | |
2408 | BUG_ON(extent_op); | |
2409 | head = btrfs_delayed_node_to_head(node); | |
2410 | trace_run_delayed_ref_head(node, head, node->action); | |
2411 | ||
2412 | if (insert_reserved) { | |
2413 | btrfs_pin_extent(root, node->bytenr, | |
2414 | node->num_bytes, 1); | |
2415 | if (head->is_data) { | |
2416 | ret = btrfs_del_csums(trans, root, | |
2417 | node->bytenr, | |
2418 | node->num_bytes); | |
2419 | } | |
2420 | } | |
2421 | ||
2422 | /* Also free its reserved qgroup space */ | |
2423 | btrfs_qgroup_free_delayed_ref(root->fs_info, | |
2424 | head->qgroup_ref_root, | |
2425 | head->qgroup_reserved); | |
2426 | return ret; | |
2427 | } | |
2428 | ||
2429 | if (node->type == BTRFS_TREE_BLOCK_REF_KEY || | |
2430 | node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
2431 | ret = run_delayed_tree_ref(trans, root, node, extent_op, | |
2432 | insert_reserved); | |
2433 | else if (node->type == BTRFS_EXTENT_DATA_REF_KEY || | |
2434 | node->type == BTRFS_SHARED_DATA_REF_KEY) | |
2435 | ret = run_delayed_data_ref(trans, root, node, extent_op, | |
2436 | insert_reserved); | |
2437 | else | |
2438 | BUG(); | |
2439 | return ret; | |
2440 | } | |
2441 | ||
2442 | static inline struct btrfs_delayed_ref_node * | |
2443 | select_delayed_ref(struct btrfs_delayed_ref_head *head) | |
2444 | { | |
2445 | struct btrfs_delayed_ref_node *ref; | |
2446 | ||
2447 | if (list_empty(&head->ref_list)) | |
2448 | return NULL; | |
2449 | ||
2450 | /* | |
2451 | * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first. | |
2452 | * This is to prevent a ref count from going down to zero, which deletes | |
2453 | * the extent item from the extent tree, when there still are references | |
2454 | * to add, which would fail because they would not find the extent item. | |
2455 | */ | |
2456 | list_for_each_entry(ref, &head->ref_list, list) { | |
2457 | if (ref->action == BTRFS_ADD_DELAYED_REF) | |
2458 | return ref; | |
2459 | } | |
2460 | ||
2461 | return list_entry(head->ref_list.next, struct btrfs_delayed_ref_node, | |
2462 | list); | |
2463 | } | |
2464 | ||
2465 | /* | |
2466 | * Returns 0 on success or if called with an already aborted transaction. | |
2467 | * Returns -ENOMEM or -EIO on failure and will abort the transaction. | |
2468 | */ | |
2469 | static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, | |
2470 | struct btrfs_root *root, | |
2471 | unsigned long nr) | |
2472 | { | |
2473 | struct btrfs_delayed_ref_root *delayed_refs; | |
2474 | struct btrfs_delayed_ref_node *ref; | |
2475 | struct btrfs_delayed_ref_head *locked_ref = NULL; | |
2476 | struct btrfs_delayed_extent_op *extent_op; | |
2477 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2478 | ktime_t start = ktime_get(); | |
2479 | int ret; | |
2480 | unsigned long count = 0; | |
2481 | unsigned long actual_count = 0; | |
2482 | int must_insert_reserved = 0; | |
2483 | ||
2484 | delayed_refs = &trans->transaction->delayed_refs; | |
2485 | while (1) { | |
2486 | if (!locked_ref) { | |
2487 | if (count >= nr) | |
2488 | break; | |
2489 | ||
2490 | spin_lock(&delayed_refs->lock); | |
2491 | locked_ref = btrfs_select_ref_head(trans); | |
2492 | if (!locked_ref) { | |
2493 | spin_unlock(&delayed_refs->lock); | |
2494 | break; | |
2495 | } | |
2496 | ||
2497 | /* grab the lock that says we are going to process | |
2498 | * all the refs for this head */ | |
2499 | ret = btrfs_delayed_ref_lock(trans, locked_ref); | |
2500 | spin_unlock(&delayed_refs->lock); | |
2501 | /* | |
2502 | * we may have dropped the spin lock to get the head | |
2503 | * mutex lock, and that might have given someone else | |
2504 | * time to free the head. If that's true, it has been | |
2505 | * removed from our list and we can move on. | |
2506 | */ | |
2507 | if (ret == -EAGAIN) { | |
2508 | locked_ref = NULL; | |
2509 | count++; | |
2510 | continue; | |
2511 | } | |
2512 | } | |
2513 | ||
2514 | /* | |
2515 | * We need to try and merge add/drops of the same ref since we | |
2516 | * can run into issues with relocate dropping the implicit ref | |
2517 | * and then it being added back again before the drop can | |
2518 | * finish. If we merged anything we need to re-loop so we can | |
2519 | * get a good ref. | |
2520 | * Or we can get node references of the same type that weren't | |
2521 | * merged when created due to bumps in the tree mod seq, and | |
2522 | * we need to merge them to prevent adding an inline extent | |
2523 | * backref before dropping it (triggering a BUG_ON at | |
2524 | * insert_inline_extent_backref()). | |
2525 | */ | |
2526 | spin_lock(&locked_ref->lock); | |
2527 | btrfs_merge_delayed_refs(trans, fs_info, delayed_refs, | |
2528 | locked_ref); | |
2529 | ||
2530 | /* | |
2531 | * locked_ref is the head node, so we have to go one | |
2532 | * node back for any delayed ref updates | |
2533 | */ | |
2534 | ref = select_delayed_ref(locked_ref); | |
2535 | ||
2536 | if (ref && ref->seq && | |
2537 | btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) { | |
2538 | spin_unlock(&locked_ref->lock); | |
2539 | btrfs_delayed_ref_unlock(locked_ref); | |
2540 | spin_lock(&delayed_refs->lock); | |
2541 | locked_ref->processing = 0; | |
2542 | delayed_refs->num_heads_ready++; | |
2543 | spin_unlock(&delayed_refs->lock); | |
2544 | locked_ref = NULL; | |
2545 | cond_resched(); | |
2546 | count++; | |
2547 | continue; | |
2548 | } | |
2549 | ||
2550 | /* | |
2551 | * record the must insert reserved flag before we | |
2552 | * drop the spin lock. | |
2553 | */ | |
2554 | must_insert_reserved = locked_ref->must_insert_reserved; | |
2555 | locked_ref->must_insert_reserved = 0; | |
2556 | ||
2557 | extent_op = locked_ref->extent_op; | |
2558 | locked_ref->extent_op = NULL; | |
2559 | ||
2560 | if (!ref) { | |
2561 | ||
2562 | ||
2563 | /* All delayed refs have been processed, Go ahead | |
2564 | * and send the head node to run_one_delayed_ref, | |
2565 | * so that any accounting fixes can happen | |
2566 | */ | |
2567 | ref = &locked_ref->node; | |
2568 | ||
2569 | if (extent_op && must_insert_reserved) { | |
2570 | btrfs_free_delayed_extent_op(extent_op); | |
2571 | extent_op = NULL; | |
2572 | } | |
2573 | ||
2574 | if (extent_op) { | |
2575 | spin_unlock(&locked_ref->lock); | |
2576 | ret = run_delayed_extent_op(trans, root, | |
2577 | ref, extent_op); | |
2578 | btrfs_free_delayed_extent_op(extent_op); | |
2579 | ||
2580 | if (ret) { | |
2581 | /* | |
2582 | * Need to reset must_insert_reserved if | |
2583 | * there was an error so the abort stuff | |
2584 | * can cleanup the reserved space | |
2585 | * properly. | |
2586 | */ | |
2587 | if (must_insert_reserved) | |
2588 | locked_ref->must_insert_reserved = 1; | |
2589 | locked_ref->processing = 0; | |
2590 | btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret); | |
2591 | btrfs_delayed_ref_unlock(locked_ref); | |
2592 | return ret; | |
2593 | } | |
2594 | continue; | |
2595 | } | |
2596 | ||
2597 | /* | |
2598 | * Need to drop our head ref lock and re-aqcuire the | |
2599 | * delayed ref lock and then re-check to make sure | |
2600 | * nobody got added. | |
2601 | */ | |
2602 | spin_unlock(&locked_ref->lock); | |
2603 | spin_lock(&delayed_refs->lock); | |
2604 | spin_lock(&locked_ref->lock); | |
2605 | if (!list_empty(&locked_ref->ref_list) || | |
2606 | locked_ref->extent_op) { | |
2607 | spin_unlock(&locked_ref->lock); | |
2608 | spin_unlock(&delayed_refs->lock); | |
2609 | continue; | |
2610 | } | |
2611 | ref->in_tree = 0; | |
2612 | delayed_refs->num_heads--; | |
2613 | rb_erase(&locked_ref->href_node, | |
2614 | &delayed_refs->href_root); | |
2615 | spin_unlock(&delayed_refs->lock); | |
2616 | } else { | |
2617 | actual_count++; | |
2618 | ref->in_tree = 0; | |
2619 | list_del(&ref->list); | |
2620 | } | |
2621 | atomic_dec(&delayed_refs->num_entries); | |
2622 | ||
2623 | if (!btrfs_delayed_ref_is_head(ref)) { | |
2624 | /* | |
2625 | * when we play the delayed ref, also correct the | |
2626 | * ref_mod on head | |
2627 | */ | |
2628 | switch (ref->action) { | |
2629 | case BTRFS_ADD_DELAYED_REF: | |
2630 | case BTRFS_ADD_DELAYED_EXTENT: | |
2631 | locked_ref->node.ref_mod -= ref->ref_mod; | |
2632 | break; | |
2633 | case BTRFS_DROP_DELAYED_REF: | |
2634 | locked_ref->node.ref_mod += ref->ref_mod; | |
2635 | break; | |
2636 | default: | |
2637 | WARN_ON(1); | |
2638 | } | |
2639 | } | |
2640 | spin_unlock(&locked_ref->lock); | |
2641 | ||
2642 | ret = run_one_delayed_ref(trans, root, ref, extent_op, | |
2643 | must_insert_reserved); | |
2644 | ||
2645 | btrfs_free_delayed_extent_op(extent_op); | |
2646 | if (ret) { | |
2647 | locked_ref->processing = 0; | |
2648 | btrfs_delayed_ref_unlock(locked_ref); | |
2649 | btrfs_put_delayed_ref(ref); | |
2650 | btrfs_debug(fs_info, "run_one_delayed_ref returned %d", ret); | |
2651 | return ret; | |
2652 | } | |
2653 | ||
2654 | /* | |
2655 | * If this node is a head, that means all the refs in this head | |
2656 | * have been dealt with, and we will pick the next head to deal | |
2657 | * with, so we must unlock the head and drop it from the cluster | |
2658 | * list before we release it. | |
2659 | */ | |
2660 | if (btrfs_delayed_ref_is_head(ref)) { | |
2661 | if (locked_ref->is_data && | |
2662 | locked_ref->total_ref_mod < 0) { | |
2663 | spin_lock(&delayed_refs->lock); | |
2664 | delayed_refs->pending_csums -= ref->num_bytes; | |
2665 | spin_unlock(&delayed_refs->lock); | |
2666 | } | |
2667 | btrfs_delayed_ref_unlock(locked_ref); | |
2668 | locked_ref = NULL; | |
2669 | } | |
2670 | btrfs_put_delayed_ref(ref); | |
2671 | count++; | |
2672 | cond_resched(); | |
2673 | } | |
2674 | ||
2675 | /* | |
2676 | * We don't want to include ref heads since we can have empty ref heads | |
2677 | * and those will drastically skew our runtime down since we just do | |
2678 | * accounting, no actual extent tree updates. | |
2679 | */ | |
2680 | if (actual_count > 0) { | |
2681 | u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start)); | |
2682 | u64 avg; | |
2683 | ||
2684 | /* | |
2685 | * We weigh the current average higher than our current runtime | |
2686 | * to avoid large swings in the average. | |
2687 | */ | |
2688 | spin_lock(&delayed_refs->lock); | |
2689 | avg = fs_info->avg_delayed_ref_runtime * 3 + runtime; | |
2690 | fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */ | |
2691 | spin_unlock(&delayed_refs->lock); | |
2692 | } | |
2693 | return 0; | |
2694 | } | |
2695 | ||
2696 | #ifdef SCRAMBLE_DELAYED_REFS | |
2697 | /* | |
2698 | * Normally delayed refs get processed in ascending bytenr order. This | |
2699 | * correlates in most cases to the order added. To expose dependencies on this | |
2700 | * order, we start to process the tree in the middle instead of the beginning | |
2701 | */ | |
2702 | static u64 find_middle(struct rb_root *root) | |
2703 | { | |
2704 | struct rb_node *n = root->rb_node; | |
2705 | struct btrfs_delayed_ref_node *entry; | |
2706 | int alt = 1; | |
2707 | u64 middle; | |
2708 | u64 first = 0, last = 0; | |
2709 | ||
2710 | n = rb_first(root); | |
2711 | if (n) { | |
2712 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2713 | first = entry->bytenr; | |
2714 | } | |
2715 | n = rb_last(root); | |
2716 | if (n) { | |
2717 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2718 | last = entry->bytenr; | |
2719 | } | |
2720 | n = root->rb_node; | |
2721 | ||
2722 | while (n) { | |
2723 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2724 | WARN_ON(!entry->in_tree); | |
2725 | ||
2726 | middle = entry->bytenr; | |
2727 | ||
2728 | if (alt) | |
2729 | n = n->rb_left; | |
2730 | else | |
2731 | n = n->rb_right; | |
2732 | ||
2733 | alt = 1 - alt; | |
2734 | } | |
2735 | return middle; | |
2736 | } | |
2737 | #endif | |
2738 | ||
2739 | static inline u64 heads_to_leaves(struct btrfs_root *root, u64 heads) | |
2740 | { | |
2741 | u64 num_bytes; | |
2742 | ||
2743 | num_bytes = heads * (sizeof(struct btrfs_extent_item) + | |
2744 | sizeof(struct btrfs_extent_inline_ref)); | |
2745 | if (!btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) | |
2746 | num_bytes += heads * sizeof(struct btrfs_tree_block_info); | |
2747 | ||
2748 | /* | |
2749 | * We don't ever fill up leaves all the way so multiply by 2 just to be | |
2750 | * closer to what we're really going to want to ouse. | |
2751 | */ | |
2752 | return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(root)); | |
2753 | } | |
2754 | ||
2755 | /* | |
2756 | * Takes the number of bytes to be csumm'ed and figures out how many leaves it | |
2757 | * would require to store the csums for that many bytes. | |
2758 | */ | |
2759 | u64 btrfs_csum_bytes_to_leaves(struct btrfs_root *root, u64 csum_bytes) | |
2760 | { | |
2761 | u64 csum_size; | |
2762 | u64 num_csums_per_leaf; | |
2763 | u64 num_csums; | |
2764 | ||
2765 | csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item); | |
2766 | num_csums_per_leaf = div64_u64(csum_size, | |
2767 | (u64)btrfs_super_csum_size(root->fs_info->super_copy)); | |
2768 | num_csums = div64_u64(csum_bytes, root->sectorsize); | |
2769 | num_csums += num_csums_per_leaf - 1; | |
2770 | num_csums = div64_u64(num_csums, num_csums_per_leaf); | |
2771 | return num_csums; | |
2772 | } | |
2773 | ||
2774 | int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans, | |
2775 | struct btrfs_root *root) | |
2776 | { | |
2777 | struct btrfs_block_rsv *global_rsv; | |
2778 | u64 num_heads = trans->transaction->delayed_refs.num_heads_ready; | |
2779 | u64 csum_bytes = trans->transaction->delayed_refs.pending_csums; | |
2780 | u64 num_dirty_bgs = trans->transaction->num_dirty_bgs; | |
2781 | u64 num_bytes, num_dirty_bgs_bytes; | |
2782 | int ret = 0; | |
2783 | ||
2784 | num_bytes = btrfs_calc_trans_metadata_size(root, 1); | |
2785 | num_heads = heads_to_leaves(root, num_heads); | |
2786 | if (num_heads > 1) | |
2787 | num_bytes += (num_heads - 1) * root->nodesize; | |
2788 | num_bytes <<= 1; | |
2789 | num_bytes += btrfs_csum_bytes_to_leaves(root, csum_bytes) * root->nodesize; | |
2790 | num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(root, | |
2791 | num_dirty_bgs); | |
2792 | global_rsv = &root->fs_info->global_block_rsv; | |
2793 | ||
2794 | /* | |
2795 | * If we can't allocate any more chunks lets make sure we have _lots_ of | |
2796 | * wiggle room since running delayed refs can create more delayed refs. | |
2797 | */ | |
2798 | if (global_rsv->space_info->full) { | |
2799 | num_dirty_bgs_bytes <<= 1; | |
2800 | num_bytes <<= 1; | |
2801 | } | |
2802 | ||
2803 | spin_lock(&global_rsv->lock); | |
2804 | if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes) | |
2805 | ret = 1; | |
2806 | spin_unlock(&global_rsv->lock); | |
2807 | return ret; | |
2808 | } | |
2809 | ||
2810 | int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans, | |
2811 | struct btrfs_root *root) | |
2812 | { | |
2813 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2814 | u64 num_entries = | |
2815 | atomic_read(&trans->transaction->delayed_refs.num_entries); | |
2816 | u64 avg_runtime; | |
2817 | u64 val; | |
2818 | ||
2819 | smp_mb(); | |
2820 | avg_runtime = fs_info->avg_delayed_ref_runtime; | |
2821 | val = num_entries * avg_runtime; | |
2822 | if (num_entries * avg_runtime >= NSEC_PER_SEC) | |
2823 | return 1; | |
2824 | if (val >= NSEC_PER_SEC / 2) | |
2825 | return 2; | |
2826 | ||
2827 | return btrfs_check_space_for_delayed_refs(trans, root); | |
2828 | } | |
2829 | ||
2830 | struct async_delayed_refs { | |
2831 | struct btrfs_root *root; | |
2832 | int count; | |
2833 | int error; | |
2834 | int sync; | |
2835 | struct completion wait; | |
2836 | struct btrfs_work work; | |
2837 | }; | |
2838 | ||
2839 | static void delayed_ref_async_start(struct btrfs_work *work) | |
2840 | { | |
2841 | struct async_delayed_refs *async; | |
2842 | struct btrfs_trans_handle *trans; | |
2843 | int ret; | |
2844 | ||
2845 | async = container_of(work, struct async_delayed_refs, work); | |
2846 | ||
2847 | trans = btrfs_join_transaction(async->root); | |
2848 | if (IS_ERR(trans)) { | |
2849 | async->error = PTR_ERR(trans); | |
2850 | goto done; | |
2851 | } | |
2852 | ||
2853 | /* | |
2854 | * trans->sync means that when we call end_transaciton, we won't | |
2855 | * wait on delayed refs | |
2856 | */ | |
2857 | trans->sync = true; | |
2858 | ret = btrfs_run_delayed_refs(trans, async->root, async->count); | |
2859 | if (ret) | |
2860 | async->error = ret; | |
2861 | ||
2862 | ret = btrfs_end_transaction(trans, async->root); | |
2863 | if (ret && !async->error) | |
2864 | async->error = ret; | |
2865 | done: | |
2866 | if (async->sync) | |
2867 | complete(&async->wait); | |
2868 | else | |
2869 | kfree(async); | |
2870 | } | |
2871 | ||
2872 | int btrfs_async_run_delayed_refs(struct btrfs_root *root, | |
2873 | unsigned long count, int wait) | |
2874 | { | |
2875 | struct async_delayed_refs *async; | |
2876 | int ret; | |
2877 | ||
2878 | async = kmalloc(sizeof(*async), GFP_NOFS); | |
2879 | if (!async) | |
2880 | return -ENOMEM; | |
2881 | ||
2882 | async->root = root->fs_info->tree_root; | |
2883 | async->count = count; | |
2884 | async->error = 0; | |
2885 | if (wait) | |
2886 | async->sync = 1; | |
2887 | else | |
2888 | async->sync = 0; | |
2889 | init_completion(&async->wait); | |
2890 | ||
2891 | btrfs_init_work(&async->work, btrfs_extent_refs_helper, | |
2892 | delayed_ref_async_start, NULL, NULL); | |
2893 | ||
2894 | btrfs_queue_work(root->fs_info->extent_workers, &async->work); | |
2895 | ||
2896 | if (wait) { | |
2897 | wait_for_completion(&async->wait); | |
2898 | ret = async->error; | |
2899 | kfree(async); | |
2900 | return ret; | |
2901 | } | |
2902 | return 0; | |
2903 | } | |
2904 | ||
2905 | /* | |
2906 | * this starts processing the delayed reference count updates and | |
2907 | * extent insertions we have queued up so far. count can be | |
2908 | * 0, which means to process everything in the tree at the start | |
2909 | * of the run (but not newly added entries), or it can be some target | |
2910 | * number you'd like to process. | |
2911 | * | |
2912 | * Returns 0 on success or if called with an aborted transaction | |
2913 | * Returns <0 on error and aborts the transaction | |
2914 | */ | |
2915 | int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, | |
2916 | struct btrfs_root *root, unsigned long count) | |
2917 | { | |
2918 | struct rb_node *node; | |
2919 | struct btrfs_delayed_ref_root *delayed_refs; | |
2920 | struct btrfs_delayed_ref_head *head; | |
2921 | int ret; | |
2922 | int run_all = count == (unsigned long)-1; | |
2923 | bool can_flush_pending_bgs = trans->can_flush_pending_bgs; | |
2924 | ||
2925 | /* We'll clean this up in btrfs_cleanup_transaction */ | |
2926 | if (trans->aborted) | |
2927 | return 0; | |
2928 | ||
2929 | if (root->fs_info->creating_free_space_tree) | |
2930 | return 0; | |
2931 | ||
2932 | if (root == root->fs_info->extent_root) | |
2933 | root = root->fs_info->tree_root; | |
2934 | ||
2935 | delayed_refs = &trans->transaction->delayed_refs; | |
2936 | if (count == 0) | |
2937 | count = atomic_read(&delayed_refs->num_entries) * 2; | |
2938 | ||
2939 | again: | |
2940 | #ifdef SCRAMBLE_DELAYED_REFS | |
2941 | delayed_refs->run_delayed_start = find_middle(&delayed_refs->root); | |
2942 | #endif | |
2943 | trans->can_flush_pending_bgs = false; | |
2944 | ret = __btrfs_run_delayed_refs(trans, root, count); | |
2945 | if (ret < 0) { | |
2946 | btrfs_abort_transaction(trans, root, ret); | |
2947 | return ret; | |
2948 | } | |
2949 | ||
2950 | if (run_all) { | |
2951 | if (!list_empty(&trans->new_bgs)) | |
2952 | btrfs_create_pending_block_groups(trans, root); | |
2953 | ||
2954 | spin_lock(&delayed_refs->lock); | |
2955 | node = rb_first(&delayed_refs->href_root); | |
2956 | if (!node) { | |
2957 | spin_unlock(&delayed_refs->lock); | |
2958 | goto out; | |
2959 | } | |
2960 | count = (unsigned long)-1; | |
2961 | ||
2962 | while (node) { | |
2963 | head = rb_entry(node, struct btrfs_delayed_ref_head, | |
2964 | href_node); | |
2965 | if (btrfs_delayed_ref_is_head(&head->node)) { | |
2966 | struct btrfs_delayed_ref_node *ref; | |
2967 | ||
2968 | ref = &head->node; | |
2969 | atomic_inc(&ref->refs); | |
2970 | ||
2971 | spin_unlock(&delayed_refs->lock); | |
2972 | /* | |
2973 | * Mutex was contended, block until it's | |
2974 | * released and try again | |
2975 | */ | |
2976 | mutex_lock(&head->mutex); | |
2977 | mutex_unlock(&head->mutex); | |
2978 | ||
2979 | btrfs_put_delayed_ref(ref); | |
2980 | cond_resched(); | |
2981 | goto again; | |
2982 | } else { | |
2983 | WARN_ON(1); | |
2984 | } | |
2985 | node = rb_next(node); | |
2986 | } | |
2987 | spin_unlock(&delayed_refs->lock); | |
2988 | cond_resched(); | |
2989 | goto again; | |
2990 | } | |
2991 | out: | |
2992 | assert_qgroups_uptodate(trans); | |
2993 | trans->can_flush_pending_bgs = can_flush_pending_bgs; | |
2994 | return 0; | |
2995 | } | |
2996 | ||
2997 | int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans, | |
2998 | struct btrfs_root *root, | |
2999 | u64 bytenr, u64 num_bytes, u64 flags, | |
3000 | int level, int is_data) | |
3001 | { | |
3002 | struct btrfs_delayed_extent_op *extent_op; | |
3003 | int ret; | |
3004 | ||
3005 | extent_op = btrfs_alloc_delayed_extent_op(); | |
3006 | if (!extent_op) | |
3007 | return -ENOMEM; | |
3008 | ||
3009 | extent_op->flags_to_set = flags; | |
3010 | extent_op->update_flags = true; | |
3011 | extent_op->update_key = false; | |
3012 | extent_op->is_data = is_data ? true : false; | |
3013 | extent_op->level = level; | |
3014 | ||
3015 | ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr, | |
3016 | num_bytes, extent_op); | |
3017 | if (ret) | |
3018 | btrfs_free_delayed_extent_op(extent_op); | |
3019 | return ret; | |
3020 | } | |
3021 | ||
3022 | static noinline int check_delayed_ref(struct btrfs_trans_handle *trans, | |
3023 | struct btrfs_root *root, | |
3024 | struct btrfs_path *path, | |
3025 | u64 objectid, u64 offset, u64 bytenr) | |
3026 | { | |
3027 | struct btrfs_delayed_ref_head *head; | |
3028 | struct btrfs_delayed_ref_node *ref; | |
3029 | struct btrfs_delayed_data_ref *data_ref; | |
3030 | struct btrfs_delayed_ref_root *delayed_refs; | |
3031 | int ret = 0; | |
3032 | ||
3033 | delayed_refs = &trans->transaction->delayed_refs; | |
3034 | spin_lock(&delayed_refs->lock); | |
3035 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
3036 | if (!head) { | |
3037 | spin_unlock(&delayed_refs->lock); | |
3038 | return 0; | |
3039 | } | |
3040 | ||
3041 | if (!mutex_trylock(&head->mutex)) { | |
3042 | atomic_inc(&head->node.refs); | |
3043 | spin_unlock(&delayed_refs->lock); | |
3044 | ||
3045 | btrfs_release_path(path); | |
3046 | ||
3047 | /* | |
3048 | * Mutex was contended, block until it's released and let | |
3049 | * caller try again | |
3050 | */ | |
3051 | mutex_lock(&head->mutex); | |
3052 | mutex_unlock(&head->mutex); | |
3053 | btrfs_put_delayed_ref(&head->node); | |
3054 | return -EAGAIN; | |
3055 | } | |
3056 | spin_unlock(&delayed_refs->lock); | |
3057 | ||
3058 | spin_lock(&head->lock); | |
3059 | list_for_each_entry(ref, &head->ref_list, list) { | |
3060 | /* If it's a shared ref we know a cross reference exists */ | |
3061 | if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) { | |
3062 | ret = 1; | |
3063 | break; | |
3064 | } | |
3065 | ||
3066 | data_ref = btrfs_delayed_node_to_data_ref(ref); | |
3067 | ||
3068 | /* | |
3069 | * If our ref doesn't match the one we're currently looking at | |
3070 | * then we have a cross reference. | |
3071 | */ | |
3072 | if (data_ref->root != root->root_key.objectid || | |
3073 | data_ref->objectid != objectid || | |
3074 | data_ref->offset != offset) { | |
3075 | ret = 1; | |
3076 | break; | |
3077 | } | |
3078 | } | |
3079 | spin_unlock(&head->lock); | |
3080 | mutex_unlock(&head->mutex); | |
3081 | return ret; | |
3082 | } | |
3083 | ||
3084 | static noinline int check_committed_ref(struct btrfs_trans_handle *trans, | |
3085 | struct btrfs_root *root, | |
3086 | struct btrfs_path *path, | |
3087 | u64 objectid, u64 offset, u64 bytenr) | |
3088 | { | |
3089 | struct btrfs_root *extent_root = root->fs_info->extent_root; | |
3090 | struct extent_buffer *leaf; | |
3091 | struct btrfs_extent_data_ref *ref; | |
3092 | struct btrfs_extent_inline_ref *iref; | |
3093 | struct btrfs_extent_item *ei; | |
3094 | struct btrfs_key key; | |
3095 | u32 item_size; | |
3096 | int ret; | |
3097 | ||
3098 | key.objectid = bytenr; | |
3099 | key.offset = (u64)-1; | |
3100 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
3101 | ||
3102 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); | |
3103 | if (ret < 0) | |
3104 | goto out; | |
3105 | BUG_ON(ret == 0); /* Corruption */ | |
3106 | ||
3107 | ret = -ENOENT; | |
3108 | if (path->slots[0] == 0) | |
3109 | goto out; | |
3110 | ||
3111 | path->slots[0]--; | |
3112 | leaf = path->nodes[0]; | |
3113 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
3114 | ||
3115 | if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY) | |
3116 | goto out; | |
3117 | ||
3118 | ret = 1; | |
3119 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
3120 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
3121 | if (item_size < sizeof(*ei)) { | |
3122 | WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0)); | |
3123 | goto out; | |
3124 | } | |
3125 | #endif | |
3126 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
3127 | ||
3128 | if (item_size != sizeof(*ei) + | |
3129 | btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY)) | |
3130 | goto out; | |
3131 | ||
3132 | if (btrfs_extent_generation(leaf, ei) <= | |
3133 | btrfs_root_last_snapshot(&root->root_item)) | |
3134 | goto out; | |
3135 | ||
3136 | iref = (struct btrfs_extent_inline_ref *)(ei + 1); | |
3137 | if (btrfs_extent_inline_ref_type(leaf, iref) != | |
3138 | BTRFS_EXTENT_DATA_REF_KEY) | |
3139 | goto out; | |
3140 | ||
3141 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
3142 | if (btrfs_extent_refs(leaf, ei) != | |
3143 | btrfs_extent_data_ref_count(leaf, ref) || | |
3144 | btrfs_extent_data_ref_root(leaf, ref) != | |
3145 | root->root_key.objectid || | |
3146 | btrfs_extent_data_ref_objectid(leaf, ref) != objectid || | |
3147 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
3148 | goto out; | |
3149 | ||
3150 | ret = 0; | |
3151 | out: | |
3152 | return ret; | |
3153 | } | |
3154 | ||
3155 | int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans, | |
3156 | struct btrfs_root *root, | |
3157 | u64 objectid, u64 offset, u64 bytenr) | |
3158 | { | |
3159 | struct btrfs_path *path; | |
3160 | int ret; | |
3161 | int ret2; | |
3162 | ||
3163 | path = btrfs_alloc_path(); | |
3164 | if (!path) | |
3165 | return -ENOENT; | |
3166 | ||
3167 | do { | |
3168 | ret = check_committed_ref(trans, root, path, objectid, | |
3169 | offset, bytenr); | |
3170 | if (ret && ret != -ENOENT) | |
3171 | goto out; | |
3172 | ||
3173 | ret2 = check_delayed_ref(trans, root, path, objectid, | |
3174 | offset, bytenr); | |
3175 | } while (ret2 == -EAGAIN); | |
3176 | ||
3177 | if (ret2 && ret2 != -ENOENT) { | |
3178 | ret = ret2; | |
3179 | goto out; | |
3180 | } | |
3181 | ||
3182 | if (ret != -ENOENT || ret2 != -ENOENT) | |
3183 | ret = 0; | |
3184 | out: | |
3185 | btrfs_free_path(path); | |
3186 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) | |
3187 | WARN_ON(ret > 0); | |
3188 | return ret; | |
3189 | } | |
3190 | ||
3191 | static int __btrfs_mod_ref(struct btrfs_trans_handle *trans, | |
3192 | struct btrfs_root *root, | |
3193 | struct extent_buffer *buf, | |
3194 | int full_backref, int inc) | |
3195 | { | |
3196 | u64 bytenr; | |
3197 | u64 num_bytes; | |
3198 | u64 parent; | |
3199 | u64 ref_root; | |
3200 | u32 nritems; | |
3201 | struct btrfs_key key; | |
3202 | struct btrfs_file_extent_item *fi; | |
3203 | int i; | |
3204 | int level; | |
3205 | int ret = 0; | |
3206 | int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *, | |
3207 | u64, u64, u64, u64, u64, u64); | |
3208 | ||
3209 | ||
3210 | if (btrfs_test_is_dummy_root(root)) | |
3211 | return 0; | |
3212 | ||
3213 | ref_root = btrfs_header_owner(buf); | |
3214 | nritems = btrfs_header_nritems(buf); | |
3215 | level = btrfs_header_level(buf); | |
3216 | ||
3217 | if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0) | |
3218 | return 0; | |
3219 | ||
3220 | if (inc) | |
3221 | process_func = btrfs_inc_extent_ref; | |
3222 | else | |
3223 | process_func = btrfs_free_extent; | |
3224 | ||
3225 | if (full_backref) | |
3226 | parent = buf->start; | |
3227 | else | |
3228 | parent = 0; | |
3229 | ||
3230 | for (i = 0; i < nritems; i++) { | |
3231 | if (level == 0) { | |
3232 | btrfs_item_key_to_cpu(buf, &key, i); | |
3233 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
3234 | continue; | |
3235 | fi = btrfs_item_ptr(buf, i, | |
3236 | struct btrfs_file_extent_item); | |
3237 | if (btrfs_file_extent_type(buf, fi) == | |
3238 | BTRFS_FILE_EXTENT_INLINE) | |
3239 | continue; | |
3240 | bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
3241 | if (bytenr == 0) | |
3242 | continue; | |
3243 | ||
3244 | num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi); | |
3245 | key.offset -= btrfs_file_extent_offset(buf, fi); | |
3246 | ret = process_func(trans, root, bytenr, num_bytes, | |
3247 | parent, ref_root, key.objectid, | |
3248 | key.offset); | |
3249 | if (ret) | |
3250 | goto fail; | |
3251 | } else { | |
3252 | bytenr = btrfs_node_blockptr(buf, i); | |
3253 | num_bytes = root->nodesize; | |
3254 | ret = process_func(trans, root, bytenr, num_bytes, | |
3255 | parent, ref_root, level - 1, 0); | |
3256 | if (ret) | |
3257 | goto fail; | |
3258 | } | |
3259 | } | |
3260 | return 0; | |
3261 | fail: | |
3262 | return ret; | |
3263 | } | |
3264 | ||
3265 | int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
3266 | struct extent_buffer *buf, int full_backref) | |
3267 | { | |
3268 | return __btrfs_mod_ref(trans, root, buf, full_backref, 1); | |
3269 | } | |
3270 | ||
3271 | int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
3272 | struct extent_buffer *buf, int full_backref) | |
3273 | { | |
3274 | return __btrfs_mod_ref(trans, root, buf, full_backref, 0); | |
3275 | } | |
3276 | ||
3277 | static int write_one_cache_group(struct btrfs_trans_handle *trans, | |
3278 | struct btrfs_root *root, | |
3279 | struct btrfs_path *path, | |
3280 | struct btrfs_block_group_cache *cache) | |
3281 | { | |
3282 | int ret; | |
3283 | struct btrfs_root *extent_root = root->fs_info->extent_root; | |
3284 | unsigned long bi; | |
3285 | struct extent_buffer *leaf; | |
3286 | ||
3287 | ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1); | |
3288 | if (ret) { | |
3289 | if (ret > 0) | |
3290 | ret = -ENOENT; | |
3291 | goto fail; | |
3292 | } | |
3293 | ||
3294 | leaf = path->nodes[0]; | |
3295 | bi = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
3296 | write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item)); | |
3297 | btrfs_mark_buffer_dirty(leaf); | |
3298 | fail: | |
3299 | btrfs_release_path(path); | |
3300 | return ret; | |
3301 | ||
3302 | } | |
3303 | ||
3304 | static struct btrfs_block_group_cache * | |
3305 | next_block_group(struct btrfs_root *root, | |
3306 | struct btrfs_block_group_cache *cache) | |
3307 | { | |
3308 | struct rb_node *node; | |
3309 | ||
3310 | spin_lock(&root->fs_info->block_group_cache_lock); | |
3311 | ||
3312 | /* If our block group was removed, we need a full search. */ | |
3313 | if (RB_EMPTY_NODE(&cache->cache_node)) { | |
3314 | const u64 next_bytenr = cache->key.objectid + cache->key.offset; | |
3315 | ||
3316 | spin_unlock(&root->fs_info->block_group_cache_lock); | |
3317 | btrfs_put_block_group(cache); | |
3318 | cache = btrfs_lookup_first_block_group(root->fs_info, | |
3319 | next_bytenr); | |
3320 | return cache; | |
3321 | } | |
3322 | node = rb_next(&cache->cache_node); | |
3323 | btrfs_put_block_group(cache); | |
3324 | if (node) { | |
3325 | cache = rb_entry(node, struct btrfs_block_group_cache, | |
3326 | cache_node); | |
3327 | btrfs_get_block_group(cache); | |
3328 | } else | |
3329 | cache = NULL; | |
3330 | spin_unlock(&root->fs_info->block_group_cache_lock); | |
3331 | return cache; | |
3332 | } | |
3333 | ||
3334 | static int cache_save_setup(struct btrfs_block_group_cache *block_group, | |
3335 | struct btrfs_trans_handle *trans, | |
3336 | struct btrfs_path *path) | |
3337 | { | |
3338 | struct btrfs_root *root = block_group->fs_info->tree_root; | |
3339 | struct inode *inode = NULL; | |
3340 | u64 alloc_hint = 0; | |
3341 | int dcs = BTRFS_DC_ERROR; | |
3342 | u64 num_pages = 0; | |
3343 | int retries = 0; | |
3344 | int ret = 0; | |
3345 | ||
3346 | /* | |
3347 | * If this block group is smaller than 100 megs don't bother caching the | |
3348 | * block group. | |
3349 | */ | |
3350 | if (block_group->key.offset < (100 * SZ_1M)) { | |
3351 | spin_lock(&block_group->lock); | |
3352 | block_group->disk_cache_state = BTRFS_DC_WRITTEN; | |
3353 | spin_unlock(&block_group->lock); | |
3354 | return 0; | |
3355 | } | |
3356 | ||
3357 | if (trans->aborted) | |
3358 | return 0; | |
3359 | again: | |
3360 | inode = lookup_free_space_inode(root, block_group, path); | |
3361 | if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) { | |
3362 | ret = PTR_ERR(inode); | |
3363 | btrfs_release_path(path); | |
3364 | goto out; | |
3365 | } | |
3366 | ||
3367 | if (IS_ERR(inode)) { | |
3368 | BUG_ON(retries); | |
3369 | retries++; | |
3370 | ||
3371 | if (block_group->ro) | |
3372 | goto out_free; | |
3373 | ||
3374 | ret = create_free_space_inode(root, trans, block_group, path); | |
3375 | if (ret) | |
3376 | goto out_free; | |
3377 | goto again; | |
3378 | } | |
3379 | ||
3380 | /* We've already setup this transaction, go ahead and exit */ | |
3381 | if (block_group->cache_generation == trans->transid && | |
3382 | i_size_read(inode)) { | |
3383 | dcs = BTRFS_DC_SETUP; | |
3384 | goto out_put; | |
3385 | } | |
3386 | ||
3387 | /* | |
3388 | * We want to set the generation to 0, that way if anything goes wrong | |
3389 | * from here on out we know not to trust this cache when we load up next | |
3390 | * time. | |
3391 | */ | |
3392 | BTRFS_I(inode)->generation = 0; | |
3393 | ret = btrfs_update_inode(trans, root, inode); | |
3394 | if (ret) { | |
3395 | /* | |
3396 | * So theoretically we could recover from this, simply set the | |
3397 | * super cache generation to 0 so we know to invalidate the | |
3398 | * cache, but then we'd have to keep track of the block groups | |
3399 | * that fail this way so we know we _have_ to reset this cache | |
3400 | * before the next commit or risk reading stale cache. So to | |
3401 | * limit our exposure to horrible edge cases lets just abort the | |
3402 | * transaction, this only happens in really bad situations | |
3403 | * anyway. | |
3404 | */ | |
3405 | btrfs_abort_transaction(trans, root, ret); | |
3406 | goto out_put; | |
3407 | } | |
3408 | WARN_ON(ret); | |
3409 | ||
3410 | if (i_size_read(inode) > 0) { | |
3411 | ret = btrfs_check_trunc_cache_free_space(root, | |
3412 | &root->fs_info->global_block_rsv); | |
3413 | if (ret) | |
3414 | goto out_put; | |
3415 | ||
3416 | ret = btrfs_truncate_free_space_cache(root, trans, NULL, inode); | |
3417 | if (ret) | |
3418 | goto out_put; | |
3419 | } | |
3420 | ||
3421 | spin_lock(&block_group->lock); | |
3422 | if (block_group->cached != BTRFS_CACHE_FINISHED || | |
3423 | !btrfs_test_opt(root, SPACE_CACHE)) { | |
3424 | /* | |
3425 | * don't bother trying to write stuff out _if_ | |
3426 | * a) we're not cached, | |
3427 | * b) we're with nospace_cache mount option. | |
3428 | */ | |
3429 | dcs = BTRFS_DC_WRITTEN; | |
3430 | spin_unlock(&block_group->lock); | |
3431 | goto out_put; | |
3432 | } | |
3433 | spin_unlock(&block_group->lock); | |
3434 | ||
3435 | /* | |
3436 | * We hit an ENOSPC when setting up the cache in this transaction, just | |
3437 | * skip doing the setup, we've already cleared the cache so we're safe. | |
3438 | */ | |
3439 | if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) { | |
3440 | ret = -ENOSPC; | |
3441 | goto out_put; | |
3442 | } | |
3443 | ||
3444 | /* | |
3445 | * Try to preallocate enough space based on how big the block group is. | |
3446 | * Keep in mind this has to include any pinned space which could end up | |
3447 | * taking up quite a bit since it's not folded into the other space | |
3448 | * cache. | |
3449 | */ | |
3450 | num_pages = div_u64(block_group->key.offset, SZ_256M); | |
3451 | if (!num_pages) | |
3452 | num_pages = 1; | |
3453 | ||
3454 | num_pages *= 16; | |
3455 | num_pages *= PAGE_SIZE; | |
3456 | ||
3457 | ret = btrfs_check_data_free_space(inode, 0, num_pages); | |
3458 | if (ret) | |
3459 | goto out_put; | |
3460 | ||
3461 | ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages, | |
3462 | num_pages, num_pages, | |
3463 | &alloc_hint); | |
3464 | /* | |
3465 | * Our cache requires contiguous chunks so that we don't modify a bunch | |
3466 | * of metadata or split extents when writing the cache out, which means | |
3467 | * we can enospc if we are heavily fragmented in addition to just normal | |
3468 | * out of space conditions. So if we hit this just skip setting up any | |
3469 | * other block groups for this transaction, maybe we'll unpin enough | |
3470 | * space the next time around. | |
3471 | */ | |
3472 | if (!ret) | |
3473 | dcs = BTRFS_DC_SETUP; | |
3474 | else if (ret == -ENOSPC) | |
3475 | set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags); | |
3476 | btrfs_free_reserved_data_space(inode, 0, num_pages); | |
3477 | ||
3478 | out_put: | |
3479 | iput(inode); | |
3480 | out_free: | |
3481 | btrfs_release_path(path); | |
3482 | out: | |
3483 | spin_lock(&block_group->lock); | |
3484 | if (!ret && dcs == BTRFS_DC_SETUP) | |
3485 | block_group->cache_generation = trans->transid; | |
3486 | block_group->disk_cache_state = dcs; | |
3487 | spin_unlock(&block_group->lock); | |
3488 | ||
3489 | return ret; | |
3490 | } | |
3491 | ||
3492 | int btrfs_setup_space_cache(struct btrfs_trans_handle *trans, | |
3493 | struct btrfs_root *root) | |
3494 | { | |
3495 | struct btrfs_block_group_cache *cache, *tmp; | |
3496 | struct btrfs_transaction *cur_trans = trans->transaction; | |
3497 | struct btrfs_path *path; | |
3498 | ||
3499 | if (list_empty(&cur_trans->dirty_bgs) || | |
3500 | !btrfs_test_opt(root, SPACE_CACHE)) | |
3501 | return 0; | |
3502 | ||
3503 | path = btrfs_alloc_path(); | |
3504 | if (!path) | |
3505 | return -ENOMEM; | |
3506 | ||
3507 | /* Could add new block groups, use _safe just in case */ | |
3508 | list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs, | |
3509 | dirty_list) { | |
3510 | if (cache->disk_cache_state == BTRFS_DC_CLEAR) | |
3511 | cache_save_setup(cache, trans, path); | |
3512 | } | |
3513 | ||
3514 | btrfs_free_path(path); | |
3515 | return 0; | |
3516 | } | |
3517 | ||
3518 | /* | |
3519 | * transaction commit does final block group cache writeback during a | |
3520 | * critical section where nothing is allowed to change the FS. This is | |
3521 | * required in order for the cache to actually match the block group, | |
3522 | * but can introduce a lot of latency into the commit. | |
3523 | * | |
3524 | * So, btrfs_start_dirty_block_groups is here to kick off block group | |
3525 | * cache IO. There's a chance we'll have to redo some of it if the | |
3526 | * block group changes again during the commit, but it greatly reduces | |
3527 | * the commit latency by getting rid of the easy block groups while | |
3528 | * we're still allowing others to join the commit. | |
3529 | */ | |
3530 | int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans, | |
3531 | struct btrfs_root *root) | |
3532 | { | |
3533 | struct btrfs_block_group_cache *cache; | |
3534 | struct btrfs_transaction *cur_trans = trans->transaction; | |
3535 | int ret = 0; | |
3536 | int should_put; | |
3537 | struct btrfs_path *path = NULL; | |
3538 | LIST_HEAD(dirty); | |
3539 | struct list_head *io = &cur_trans->io_bgs; | |
3540 | int num_started = 0; | |
3541 | int loops = 0; | |
3542 | ||
3543 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3544 | if (list_empty(&cur_trans->dirty_bgs)) { | |
3545 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3546 | return 0; | |
3547 | } | |
3548 | list_splice_init(&cur_trans->dirty_bgs, &dirty); | |
3549 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3550 | ||
3551 | again: | |
3552 | /* | |
3553 | * make sure all the block groups on our dirty list actually | |
3554 | * exist | |
3555 | */ | |
3556 | btrfs_create_pending_block_groups(trans, root); | |
3557 | ||
3558 | if (!path) { | |
3559 | path = btrfs_alloc_path(); | |
3560 | if (!path) | |
3561 | return -ENOMEM; | |
3562 | } | |
3563 | ||
3564 | /* | |
3565 | * cache_write_mutex is here only to save us from balance or automatic | |
3566 | * removal of empty block groups deleting this block group while we are | |
3567 | * writing out the cache | |
3568 | */ | |
3569 | mutex_lock(&trans->transaction->cache_write_mutex); | |
3570 | while (!list_empty(&dirty)) { | |
3571 | cache = list_first_entry(&dirty, | |
3572 | struct btrfs_block_group_cache, | |
3573 | dirty_list); | |
3574 | /* | |
3575 | * this can happen if something re-dirties a block | |
3576 | * group that is already under IO. Just wait for it to | |
3577 | * finish and then do it all again | |
3578 | */ | |
3579 | if (!list_empty(&cache->io_list)) { | |
3580 | list_del_init(&cache->io_list); | |
3581 | btrfs_wait_cache_io(root, trans, cache, | |
3582 | &cache->io_ctl, path, | |
3583 | cache->key.objectid); | |
3584 | btrfs_put_block_group(cache); | |
3585 | } | |
3586 | ||
3587 | ||
3588 | /* | |
3589 | * btrfs_wait_cache_io uses the cache->dirty_list to decide | |
3590 | * if it should update the cache_state. Don't delete | |
3591 | * until after we wait. | |
3592 | * | |
3593 | * Since we're not running in the commit critical section | |
3594 | * we need the dirty_bgs_lock to protect from update_block_group | |
3595 | */ | |
3596 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3597 | list_del_init(&cache->dirty_list); | |
3598 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3599 | ||
3600 | should_put = 1; | |
3601 | ||
3602 | cache_save_setup(cache, trans, path); | |
3603 | ||
3604 | if (cache->disk_cache_state == BTRFS_DC_SETUP) { | |
3605 | cache->io_ctl.inode = NULL; | |
3606 | ret = btrfs_write_out_cache(root, trans, cache, path); | |
3607 | if (ret == 0 && cache->io_ctl.inode) { | |
3608 | num_started++; | |
3609 | should_put = 0; | |
3610 | ||
3611 | /* | |
3612 | * the cache_write_mutex is protecting | |
3613 | * the io_list | |
3614 | */ | |
3615 | list_add_tail(&cache->io_list, io); | |
3616 | } else { | |
3617 | /* | |
3618 | * if we failed to write the cache, the | |
3619 | * generation will be bad and life goes on | |
3620 | */ | |
3621 | ret = 0; | |
3622 | } | |
3623 | } | |
3624 | if (!ret) { | |
3625 | ret = write_one_cache_group(trans, root, path, cache); | |
3626 | /* | |
3627 | * Our block group might still be attached to the list | |
3628 | * of new block groups in the transaction handle of some | |
3629 | * other task (struct btrfs_trans_handle->new_bgs). This | |
3630 | * means its block group item isn't yet in the extent | |
3631 | * tree. If this happens ignore the error, as we will | |
3632 | * try again later in the critical section of the | |
3633 | * transaction commit. | |
3634 | */ | |
3635 | if (ret == -ENOENT) { | |
3636 | ret = 0; | |
3637 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3638 | if (list_empty(&cache->dirty_list)) { | |
3639 | list_add_tail(&cache->dirty_list, | |
3640 | &cur_trans->dirty_bgs); | |
3641 | btrfs_get_block_group(cache); | |
3642 | } | |
3643 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3644 | } else if (ret) { | |
3645 | btrfs_abort_transaction(trans, root, ret); | |
3646 | } | |
3647 | } | |
3648 | ||
3649 | /* if its not on the io list, we need to put the block group */ | |
3650 | if (should_put) | |
3651 | btrfs_put_block_group(cache); | |
3652 | ||
3653 | if (ret) | |
3654 | break; | |
3655 | ||
3656 | /* | |
3657 | * Avoid blocking other tasks for too long. It might even save | |
3658 | * us from writing caches for block groups that are going to be | |
3659 | * removed. | |
3660 | */ | |
3661 | mutex_unlock(&trans->transaction->cache_write_mutex); | |
3662 | mutex_lock(&trans->transaction->cache_write_mutex); | |
3663 | } | |
3664 | mutex_unlock(&trans->transaction->cache_write_mutex); | |
3665 | ||
3666 | /* | |
3667 | * go through delayed refs for all the stuff we've just kicked off | |
3668 | * and then loop back (just once) | |
3669 | */ | |
3670 | ret = btrfs_run_delayed_refs(trans, root, 0); | |
3671 | if (!ret && loops == 0) { | |
3672 | loops++; | |
3673 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3674 | list_splice_init(&cur_trans->dirty_bgs, &dirty); | |
3675 | /* | |
3676 | * dirty_bgs_lock protects us from concurrent block group | |
3677 | * deletes too (not just cache_write_mutex). | |
3678 | */ | |
3679 | if (!list_empty(&dirty)) { | |
3680 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3681 | goto again; | |
3682 | } | |
3683 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3684 | } | |
3685 | ||
3686 | btrfs_free_path(path); | |
3687 | return ret; | |
3688 | } | |
3689 | ||
3690 | int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans, | |
3691 | struct btrfs_root *root) | |
3692 | { | |
3693 | struct btrfs_block_group_cache *cache; | |
3694 | struct btrfs_transaction *cur_trans = trans->transaction; | |
3695 | int ret = 0; | |
3696 | int should_put; | |
3697 | struct btrfs_path *path; | |
3698 | struct list_head *io = &cur_trans->io_bgs; | |
3699 | int num_started = 0; | |
3700 | ||
3701 | path = btrfs_alloc_path(); | |
3702 | if (!path) | |
3703 | return -ENOMEM; | |
3704 | ||
3705 | /* | |
3706 | * Even though we are in the critical section of the transaction commit, | |
3707 | * we can still have concurrent tasks adding elements to this | |
3708 | * transaction's list of dirty block groups. These tasks correspond to | |
3709 | * endio free space workers started when writeback finishes for a | |
3710 | * space cache, which run inode.c:btrfs_finish_ordered_io(), and can | |
3711 | * allocate new block groups as a result of COWing nodes of the root | |
3712 | * tree when updating the free space inode. The writeback for the space | |
3713 | * caches is triggered by an earlier call to | |
3714 | * btrfs_start_dirty_block_groups() and iterations of the following | |
3715 | * loop. | |
3716 | * Also we want to do the cache_save_setup first and then run the | |
3717 | * delayed refs to make sure we have the best chance at doing this all | |
3718 | * in one shot. | |
3719 | */ | |
3720 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3721 | while (!list_empty(&cur_trans->dirty_bgs)) { | |
3722 | cache = list_first_entry(&cur_trans->dirty_bgs, | |
3723 | struct btrfs_block_group_cache, | |
3724 | dirty_list); | |
3725 | ||
3726 | /* | |
3727 | * this can happen if cache_save_setup re-dirties a block | |
3728 | * group that is already under IO. Just wait for it to | |
3729 | * finish and then do it all again | |
3730 | */ | |
3731 | if (!list_empty(&cache->io_list)) { | |
3732 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3733 | list_del_init(&cache->io_list); | |
3734 | btrfs_wait_cache_io(root, trans, cache, | |
3735 | &cache->io_ctl, path, | |
3736 | cache->key.objectid); | |
3737 | btrfs_put_block_group(cache); | |
3738 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3739 | } | |
3740 | ||
3741 | /* | |
3742 | * don't remove from the dirty list until after we've waited | |
3743 | * on any pending IO | |
3744 | */ | |
3745 | list_del_init(&cache->dirty_list); | |
3746 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3747 | should_put = 1; | |
3748 | ||
3749 | cache_save_setup(cache, trans, path); | |
3750 | ||
3751 | if (!ret) | |
3752 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long) -1); | |
3753 | ||
3754 | if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) { | |
3755 | cache->io_ctl.inode = NULL; | |
3756 | ret = btrfs_write_out_cache(root, trans, cache, path); | |
3757 | if (ret == 0 && cache->io_ctl.inode) { | |
3758 | num_started++; | |
3759 | should_put = 0; | |
3760 | list_add_tail(&cache->io_list, io); | |
3761 | } else { | |
3762 | /* | |
3763 | * if we failed to write the cache, the | |
3764 | * generation will be bad and life goes on | |
3765 | */ | |
3766 | ret = 0; | |
3767 | } | |
3768 | } | |
3769 | if (!ret) { | |
3770 | ret = write_one_cache_group(trans, root, path, cache); | |
3771 | /* | |
3772 | * One of the free space endio workers might have | |
3773 | * created a new block group while updating a free space | |
3774 | * cache's inode (at inode.c:btrfs_finish_ordered_io()) | |
3775 | * and hasn't released its transaction handle yet, in | |
3776 | * which case the new block group is still attached to | |
3777 | * its transaction handle and its creation has not | |
3778 | * finished yet (no block group item in the extent tree | |
3779 | * yet, etc). If this is the case, wait for all free | |
3780 | * space endio workers to finish and retry. This is a | |
3781 | * a very rare case so no need for a more efficient and | |
3782 | * complex approach. | |
3783 | */ | |
3784 | if (ret == -ENOENT) { | |
3785 | wait_event(cur_trans->writer_wait, | |
3786 | atomic_read(&cur_trans->num_writers) == 1); | |
3787 | ret = write_one_cache_group(trans, root, path, | |
3788 | cache); | |
3789 | } | |
3790 | if (ret) | |
3791 | btrfs_abort_transaction(trans, root, ret); | |
3792 | } | |
3793 | ||
3794 | /* if its not on the io list, we need to put the block group */ | |
3795 | if (should_put) | |
3796 | btrfs_put_block_group(cache); | |
3797 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3798 | } | |
3799 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3800 | ||
3801 | while (!list_empty(io)) { | |
3802 | cache = list_first_entry(io, struct btrfs_block_group_cache, | |
3803 | io_list); | |
3804 | list_del_init(&cache->io_list); | |
3805 | btrfs_wait_cache_io(root, trans, cache, | |
3806 | &cache->io_ctl, path, cache->key.objectid); | |
3807 | btrfs_put_block_group(cache); | |
3808 | } | |
3809 | ||
3810 | btrfs_free_path(path); | |
3811 | return ret; | |
3812 | } | |
3813 | ||
3814 | int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr) | |
3815 | { | |
3816 | struct btrfs_block_group_cache *block_group; | |
3817 | int readonly = 0; | |
3818 | ||
3819 | block_group = btrfs_lookup_block_group(root->fs_info, bytenr); | |
3820 | if (!block_group || block_group->ro) | |
3821 | readonly = 1; | |
3822 | if (block_group) | |
3823 | btrfs_put_block_group(block_group); | |
3824 | return readonly; | |
3825 | } | |
3826 | ||
3827 | static const char *alloc_name(u64 flags) | |
3828 | { | |
3829 | switch (flags) { | |
3830 | case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA: | |
3831 | return "mixed"; | |
3832 | case BTRFS_BLOCK_GROUP_METADATA: | |
3833 | return "metadata"; | |
3834 | case BTRFS_BLOCK_GROUP_DATA: | |
3835 | return "data"; | |
3836 | case BTRFS_BLOCK_GROUP_SYSTEM: | |
3837 | return "system"; | |
3838 | default: | |
3839 | WARN_ON(1); | |
3840 | return "invalid-combination"; | |
3841 | }; | |
3842 | } | |
3843 | ||
3844 | static int update_space_info(struct btrfs_fs_info *info, u64 flags, | |
3845 | u64 total_bytes, u64 bytes_used, | |
3846 | struct btrfs_space_info **space_info) | |
3847 | { | |
3848 | struct btrfs_space_info *found; | |
3849 | int i; | |
3850 | int factor; | |
3851 | int ret; | |
3852 | ||
3853 | if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
3854 | BTRFS_BLOCK_GROUP_RAID10)) | |
3855 | factor = 2; | |
3856 | else | |
3857 | factor = 1; | |
3858 | ||
3859 | found = __find_space_info(info, flags); | |
3860 | if (found) { | |
3861 | spin_lock(&found->lock); | |
3862 | found->total_bytes += total_bytes; | |
3863 | found->disk_total += total_bytes * factor; | |
3864 | found->bytes_used += bytes_used; | |
3865 | found->disk_used += bytes_used * factor; | |
3866 | if (total_bytes > 0) | |
3867 | found->full = 0; | |
3868 | spin_unlock(&found->lock); | |
3869 | *space_info = found; | |
3870 | return 0; | |
3871 | } | |
3872 | found = kzalloc(sizeof(*found), GFP_NOFS); | |
3873 | if (!found) | |
3874 | return -ENOMEM; | |
3875 | ||
3876 | ret = percpu_counter_init(&found->total_bytes_pinned, 0, GFP_KERNEL); | |
3877 | if (ret) { | |
3878 | kfree(found); | |
3879 | return ret; | |
3880 | } | |
3881 | ||
3882 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) | |
3883 | INIT_LIST_HEAD(&found->block_groups[i]); | |
3884 | init_rwsem(&found->groups_sem); | |
3885 | spin_lock_init(&found->lock); | |
3886 | found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK; | |
3887 | found->total_bytes = total_bytes; | |
3888 | found->disk_total = total_bytes * factor; | |
3889 | found->bytes_used = bytes_used; | |
3890 | found->disk_used = bytes_used * factor; | |
3891 | found->bytes_pinned = 0; | |
3892 | found->bytes_reserved = 0; | |
3893 | found->bytes_readonly = 0; | |
3894 | found->bytes_may_use = 0; | |
3895 | found->full = 0; | |
3896 | found->max_extent_size = 0; | |
3897 | found->force_alloc = CHUNK_ALLOC_NO_FORCE; | |
3898 | found->chunk_alloc = 0; | |
3899 | found->flush = 0; | |
3900 | init_waitqueue_head(&found->wait); | |
3901 | INIT_LIST_HEAD(&found->ro_bgs); | |
3902 | ||
3903 | ret = kobject_init_and_add(&found->kobj, &space_info_ktype, | |
3904 | info->space_info_kobj, "%s", | |
3905 | alloc_name(found->flags)); | |
3906 | if (ret) { | |
3907 | kfree(found); | |
3908 | return ret; | |
3909 | } | |
3910 | ||
3911 | *space_info = found; | |
3912 | list_add_rcu(&found->list, &info->space_info); | |
3913 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
3914 | info->data_sinfo = found; | |
3915 | ||
3916 | return ret; | |
3917 | } | |
3918 | ||
3919 | static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) | |
3920 | { | |
3921 | u64 extra_flags = chunk_to_extended(flags) & | |
3922 | BTRFS_EXTENDED_PROFILE_MASK; | |
3923 | ||
3924 | write_seqlock(&fs_info->profiles_lock); | |
3925 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
3926 | fs_info->avail_data_alloc_bits |= extra_flags; | |
3927 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
3928 | fs_info->avail_metadata_alloc_bits |= extra_flags; | |
3929 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
3930 | fs_info->avail_system_alloc_bits |= extra_flags; | |
3931 | write_sequnlock(&fs_info->profiles_lock); | |
3932 | } | |
3933 | ||
3934 | /* | |
3935 | * returns target flags in extended format or 0 if restripe for this | |
3936 | * chunk_type is not in progress | |
3937 | * | |
3938 | * should be called with either volume_mutex or balance_lock held | |
3939 | */ | |
3940 | static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags) | |
3941 | { | |
3942 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
3943 | u64 target = 0; | |
3944 | ||
3945 | if (!bctl) | |
3946 | return 0; | |
3947 | ||
3948 | if (flags & BTRFS_BLOCK_GROUP_DATA && | |
3949 | bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
3950 | target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target; | |
3951 | } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM && | |
3952 | bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
3953 | target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target; | |
3954 | } else if (flags & BTRFS_BLOCK_GROUP_METADATA && | |
3955 | bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
3956 | target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target; | |
3957 | } | |
3958 | ||
3959 | return target; | |
3960 | } | |
3961 | ||
3962 | /* | |
3963 | * @flags: available profiles in extended format (see ctree.h) | |
3964 | * | |
3965 | * Returns reduced profile in chunk format. If profile changing is in | |
3966 | * progress (either running or paused) picks the target profile (if it's | |
3967 | * already available), otherwise falls back to plain reducing. | |
3968 | */ | |
3969 | static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags) | |
3970 | { | |
3971 | u64 num_devices = root->fs_info->fs_devices->rw_devices; | |
3972 | u64 target; | |
3973 | u64 raid_type; | |
3974 | u64 allowed = 0; | |
3975 | ||
3976 | /* | |
3977 | * see if restripe for this chunk_type is in progress, if so | |
3978 | * try to reduce to the target profile | |
3979 | */ | |
3980 | spin_lock(&root->fs_info->balance_lock); | |
3981 | target = get_restripe_target(root->fs_info, flags); | |
3982 | if (target) { | |
3983 | /* pick target profile only if it's already available */ | |
3984 | if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) { | |
3985 | spin_unlock(&root->fs_info->balance_lock); | |
3986 | return extended_to_chunk(target); | |
3987 | } | |
3988 | } | |
3989 | spin_unlock(&root->fs_info->balance_lock); | |
3990 | ||
3991 | /* First, mask out the RAID levels which aren't possible */ | |
3992 | for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) { | |
3993 | if (num_devices >= btrfs_raid_array[raid_type].devs_min) | |
3994 | allowed |= btrfs_raid_group[raid_type]; | |
3995 | } | |
3996 | allowed &= flags; | |
3997 | ||
3998 | if (allowed & BTRFS_BLOCK_GROUP_RAID6) | |
3999 | allowed = BTRFS_BLOCK_GROUP_RAID6; | |
4000 | else if (allowed & BTRFS_BLOCK_GROUP_RAID5) | |
4001 | allowed = BTRFS_BLOCK_GROUP_RAID5; | |
4002 | else if (allowed & BTRFS_BLOCK_GROUP_RAID10) | |
4003 | allowed = BTRFS_BLOCK_GROUP_RAID10; | |
4004 | else if (allowed & BTRFS_BLOCK_GROUP_RAID1) | |
4005 | allowed = BTRFS_BLOCK_GROUP_RAID1; | |
4006 | else if (allowed & BTRFS_BLOCK_GROUP_RAID0) | |
4007 | allowed = BTRFS_BLOCK_GROUP_RAID0; | |
4008 | ||
4009 | flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK; | |
4010 | ||
4011 | return extended_to_chunk(flags | allowed); | |
4012 | } | |
4013 | ||
4014 | static u64 get_alloc_profile(struct btrfs_root *root, u64 orig_flags) | |
4015 | { | |
4016 | unsigned seq; | |
4017 | u64 flags; | |
4018 | ||
4019 | do { | |
4020 | flags = orig_flags; | |
4021 | seq = read_seqbegin(&root->fs_info->profiles_lock); | |
4022 | ||
4023 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
4024 | flags |= root->fs_info->avail_data_alloc_bits; | |
4025 | else if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
4026 | flags |= root->fs_info->avail_system_alloc_bits; | |
4027 | else if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
4028 | flags |= root->fs_info->avail_metadata_alloc_bits; | |
4029 | } while (read_seqretry(&root->fs_info->profiles_lock, seq)); | |
4030 | ||
4031 | return btrfs_reduce_alloc_profile(root, flags); | |
4032 | } | |
4033 | ||
4034 | u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data) | |
4035 | { | |
4036 | u64 flags; | |
4037 | u64 ret; | |
4038 | ||
4039 | if (data) | |
4040 | flags = BTRFS_BLOCK_GROUP_DATA; | |
4041 | else if (root == root->fs_info->chunk_root) | |
4042 | flags = BTRFS_BLOCK_GROUP_SYSTEM; | |
4043 | else | |
4044 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
4045 | ||
4046 | ret = get_alloc_profile(root, flags); | |
4047 | return ret; | |
4048 | } | |
4049 | ||
4050 | int btrfs_alloc_data_chunk_ondemand(struct inode *inode, u64 bytes) | |
4051 | { | |
4052 | struct btrfs_space_info *data_sinfo; | |
4053 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4054 | struct btrfs_fs_info *fs_info = root->fs_info; | |
4055 | u64 used; | |
4056 | int ret = 0; | |
4057 | int need_commit = 2; | |
4058 | int have_pinned_space; | |
4059 | ||
4060 | /* make sure bytes are sectorsize aligned */ | |
4061 | bytes = ALIGN(bytes, root->sectorsize); | |
4062 | ||
4063 | if (btrfs_is_free_space_inode(inode)) { | |
4064 | need_commit = 0; | |
4065 | ASSERT(current->journal_info); | |
4066 | } | |
4067 | ||
4068 | data_sinfo = fs_info->data_sinfo; | |
4069 | if (!data_sinfo) | |
4070 | goto alloc; | |
4071 | ||
4072 | again: | |
4073 | /* make sure we have enough space to handle the data first */ | |
4074 | spin_lock(&data_sinfo->lock); | |
4075 | used = data_sinfo->bytes_used + data_sinfo->bytes_reserved + | |
4076 | data_sinfo->bytes_pinned + data_sinfo->bytes_readonly + | |
4077 | data_sinfo->bytes_may_use; | |
4078 | ||
4079 | if (used + bytes > data_sinfo->total_bytes) { | |
4080 | struct btrfs_trans_handle *trans; | |
4081 | ||
4082 | /* | |
4083 | * if we don't have enough free bytes in this space then we need | |
4084 | * to alloc a new chunk. | |
4085 | */ | |
4086 | if (!data_sinfo->full) { | |
4087 | u64 alloc_target; | |
4088 | ||
4089 | data_sinfo->force_alloc = CHUNK_ALLOC_FORCE; | |
4090 | spin_unlock(&data_sinfo->lock); | |
4091 | alloc: | |
4092 | alloc_target = btrfs_get_alloc_profile(root, 1); | |
4093 | /* | |
4094 | * It is ugly that we don't call nolock join | |
4095 | * transaction for the free space inode case here. | |
4096 | * But it is safe because we only do the data space | |
4097 | * reservation for the free space cache in the | |
4098 | * transaction context, the common join transaction | |
4099 | * just increase the counter of the current transaction | |
4100 | * handler, doesn't try to acquire the trans_lock of | |
4101 | * the fs. | |
4102 | */ | |
4103 | trans = btrfs_join_transaction(root); | |
4104 | if (IS_ERR(trans)) | |
4105 | return PTR_ERR(trans); | |
4106 | ||
4107 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, | |
4108 | alloc_target, | |
4109 | CHUNK_ALLOC_NO_FORCE); | |
4110 | btrfs_end_transaction(trans, root); | |
4111 | if (ret < 0) { | |
4112 | if (ret != -ENOSPC) | |
4113 | return ret; | |
4114 | else { | |
4115 | have_pinned_space = 1; | |
4116 | goto commit_trans; | |
4117 | } | |
4118 | } | |
4119 | ||
4120 | if (!data_sinfo) | |
4121 | data_sinfo = fs_info->data_sinfo; | |
4122 | ||
4123 | goto again; | |
4124 | } | |
4125 | ||
4126 | /* | |
4127 | * If we don't have enough pinned space to deal with this | |
4128 | * allocation, and no removed chunk in current transaction, | |
4129 | * don't bother committing the transaction. | |
4130 | */ | |
4131 | have_pinned_space = percpu_counter_compare( | |
4132 | &data_sinfo->total_bytes_pinned, | |
4133 | used + bytes - data_sinfo->total_bytes); | |
4134 | spin_unlock(&data_sinfo->lock); | |
4135 | ||
4136 | /* commit the current transaction and try again */ | |
4137 | commit_trans: | |
4138 | if (need_commit && | |
4139 | !atomic_read(&root->fs_info->open_ioctl_trans)) { | |
4140 | need_commit--; | |
4141 | ||
4142 | if (need_commit > 0) { | |
4143 | btrfs_start_delalloc_roots(fs_info, 0, -1); | |
4144 | btrfs_wait_ordered_roots(fs_info, -1, 0, (u64)-1); | |
4145 | } | |
4146 | ||
4147 | trans = btrfs_join_transaction(root); | |
4148 | if (IS_ERR(trans)) | |
4149 | return PTR_ERR(trans); | |
4150 | if (have_pinned_space >= 0 || | |
4151 | test_bit(BTRFS_TRANS_HAVE_FREE_BGS, | |
4152 | &trans->transaction->flags) || | |
4153 | need_commit > 0) { | |
4154 | ret = btrfs_commit_transaction(trans, root); | |
4155 | if (ret) | |
4156 | return ret; | |
4157 | /* | |
4158 | * The cleaner kthread might still be doing iput | |
4159 | * operations. Wait for it to finish so that | |
4160 | * more space is released. | |
4161 | */ | |
4162 | mutex_lock(&root->fs_info->cleaner_delayed_iput_mutex); | |
4163 | mutex_unlock(&root->fs_info->cleaner_delayed_iput_mutex); | |
4164 | goto again; | |
4165 | } else { | |
4166 | btrfs_end_transaction(trans, root); | |
4167 | } | |
4168 | } | |
4169 | ||
4170 | trace_btrfs_space_reservation(root->fs_info, | |
4171 | "space_info:enospc", | |
4172 | data_sinfo->flags, bytes, 1); | |
4173 | return -ENOSPC; | |
4174 | } | |
4175 | data_sinfo->bytes_may_use += bytes; | |
4176 | trace_btrfs_space_reservation(root->fs_info, "space_info", | |
4177 | data_sinfo->flags, bytes, 1); | |
4178 | spin_unlock(&data_sinfo->lock); | |
4179 | ||
4180 | return ret; | |
4181 | } | |
4182 | ||
4183 | /* | |
4184 | * New check_data_free_space() with ability for precious data reservation | |
4185 | * Will replace old btrfs_check_data_free_space(), but for patch split, | |
4186 | * add a new function first and then replace it. | |
4187 | */ | |
4188 | int btrfs_check_data_free_space(struct inode *inode, u64 start, u64 len) | |
4189 | { | |
4190 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4191 | int ret; | |
4192 | ||
4193 | /* align the range */ | |
4194 | len = round_up(start + len, root->sectorsize) - | |
4195 | round_down(start, root->sectorsize); | |
4196 | start = round_down(start, root->sectorsize); | |
4197 | ||
4198 | ret = btrfs_alloc_data_chunk_ondemand(inode, len); | |
4199 | if (ret < 0) | |
4200 | return ret; | |
4201 | ||
4202 | /* | |
4203 | * Use new btrfs_qgroup_reserve_data to reserve precious data space | |
4204 | * | |
4205 | * TODO: Find a good method to avoid reserve data space for NOCOW | |
4206 | * range, but don't impact performance on quota disable case. | |
4207 | */ | |
4208 | ret = btrfs_qgroup_reserve_data(inode, start, len); | |
4209 | return ret; | |
4210 | } | |
4211 | ||
4212 | /* | |
4213 | * Called if we need to clear a data reservation for this inode | |
4214 | * Normally in a error case. | |
4215 | * | |
4216 | * This one will *NOT* use accurate qgroup reserved space API, just for case | |
4217 | * which we can't sleep and is sure it won't affect qgroup reserved space. | |
4218 | * Like clear_bit_hook(). | |
4219 | */ | |
4220 | void btrfs_free_reserved_data_space_noquota(struct inode *inode, u64 start, | |
4221 | u64 len) | |
4222 | { | |
4223 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4224 | struct btrfs_space_info *data_sinfo; | |
4225 | ||
4226 | /* Make sure the range is aligned to sectorsize */ | |
4227 | len = round_up(start + len, root->sectorsize) - | |
4228 | round_down(start, root->sectorsize); | |
4229 | start = round_down(start, root->sectorsize); | |
4230 | ||
4231 | data_sinfo = root->fs_info->data_sinfo; | |
4232 | spin_lock(&data_sinfo->lock); | |
4233 | if (WARN_ON(data_sinfo->bytes_may_use < len)) | |
4234 | data_sinfo->bytes_may_use = 0; | |
4235 | else | |
4236 | data_sinfo->bytes_may_use -= len; | |
4237 | trace_btrfs_space_reservation(root->fs_info, "space_info", | |
4238 | data_sinfo->flags, len, 0); | |
4239 | spin_unlock(&data_sinfo->lock); | |
4240 | } | |
4241 | ||
4242 | /* | |
4243 | * Called if we need to clear a data reservation for this inode | |
4244 | * Normally in a error case. | |
4245 | * | |
4246 | * This one will handle the per-indoe data rsv map for accurate reserved | |
4247 | * space framework. | |
4248 | */ | |
4249 | void btrfs_free_reserved_data_space(struct inode *inode, u64 start, u64 len) | |
4250 | { | |
4251 | btrfs_free_reserved_data_space_noquota(inode, start, len); | |
4252 | btrfs_qgroup_free_data(inode, start, len); | |
4253 | } | |
4254 | ||
4255 | static void force_metadata_allocation(struct btrfs_fs_info *info) | |
4256 | { | |
4257 | struct list_head *head = &info->space_info; | |
4258 | struct btrfs_space_info *found; | |
4259 | ||
4260 | rcu_read_lock(); | |
4261 | list_for_each_entry_rcu(found, head, list) { | |
4262 | if (found->flags & BTRFS_BLOCK_GROUP_METADATA) | |
4263 | found->force_alloc = CHUNK_ALLOC_FORCE; | |
4264 | } | |
4265 | rcu_read_unlock(); | |
4266 | } | |
4267 | ||
4268 | static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global) | |
4269 | { | |
4270 | return (global->size << 1); | |
4271 | } | |
4272 | ||
4273 | static int should_alloc_chunk(struct btrfs_root *root, | |
4274 | struct btrfs_space_info *sinfo, int force) | |
4275 | { | |
4276 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; | |
4277 | u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly; | |
4278 | u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved; | |
4279 | u64 thresh; | |
4280 | ||
4281 | if (force == CHUNK_ALLOC_FORCE) | |
4282 | return 1; | |
4283 | ||
4284 | /* | |
4285 | * We need to take into account the global rsv because for all intents | |
4286 | * and purposes it's used space. Don't worry about locking the | |
4287 | * global_rsv, it doesn't change except when the transaction commits. | |
4288 | */ | |
4289 | if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA) | |
4290 | num_allocated += calc_global_rsv_need_space(global_rsv); | |
4291 | ||
4292 | /* | |
4293 | * in limited mode, we want to have some free space up to | |
4294 | * about 1% of the FS size. | |
4295 | */ | |
4296 | if (force == CHUNK_ALLOC_LIMITED) { | |
4297 | thresh = btrfs_super_total_bytes(root->fs_info->super_copy); | |
4298 | thresh = max_t(u64, SZ_64M, div_factor_fine(thresh, 1)); | |
4299 | ||
4300 | if (num_bytes - num_allocated < thresh) | |
4301 | return 1; | |
4302 | } | |
4303 | ||
4304 | if (num_allocated + SZ_2M < div_factor(num_bytes, 8)) | |
4305 | return 0; | |
4306 | return 1; | |
4307 | } | |
4308 | ||
4309 | static u64 get_profile_num_devs(struct btrfs_root *root, u64 type) | |
4310 | { | |
4311 | u64 num_dev; | |
4312 | ||
4313 | if (type & (BTRFS_BLOCK_GROUP_RAID10 | | |
4314 | BTRFS_BLOCK_GROUP_RAID0 | | |
4315 | BTRFS_BLOCK_GROUP_RAID5 | | |
4316 | BTRFS_BLOCK_GROUP_RAID6)) | |
4317 | num_dev = root->fs_info->fs_devices->rw_devices; | |
4318 | else if (type & BTRFS_BLOCK_GROUP_RAID1) | |
4319 | num_dev = 2; | |
4320 | else | |
4321 | num_dev = 1; /* DUP or single */ | |
4322 | ||
4323 | return num_dev; | |
4324 | } | |
4325 | ||
4326 | /* | |
4327 | * If @is_allocation is true, reserve space in the system space info necessary | |
4328 | * for allocating a chunk, otherwise if it's false, reserve space necessary for | |
4329 | * removing a chunk. | |
4330 | */ | |
4331 | void check_system_chunk(struct btrfs_trans_handle *trans, | |
4332 | struct btrfs_root *root, | |
4333 | u64 type) | |
4334 | { | |
4335 | struct btrfs_space_info *info; | |
4336 | u64 left; | |
4337 | u64 thresh; | |
4338 | int ret = 0; | |
4339 | u64 num_devs; | |
4340 | ||
4341 | /* | |
4342 | * Needed because we can end up allocating a system chunk and for an | |
4343 | * atomic and race free space reservation in the chunk block reserve. | |
4344 | */ | |
4345 | ASSERT(mutex_is_locked(&root->fs_info->chunk_mutex)); | |
4346 | ||
4347 | info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM); | |
4348 | spin_lock(&info->lock); | |
4349 | left = info->total_bytes - info->bytes_used - info->bytes_pinned - | |
4350 | info->bytes_reserved - info->bytes_readonly - | |
4351 | info->bytes_may_use; | |
4352 | spin_unlock(&info->lock); | |
4353 | ||
4354 | num_devs = get_profile_num_devs(root, type); | |
4355 | ||
4356 | /* num_devs device items to update and 1 chunk item to add or remove */ | |
4357 | thresh = btrfs_calc_trunc_metadata_size(root, num_devs) + | |
4358 | btrfs_calc_trans_metadata_size(root, 1); | |
4359 | ||
4360 | if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) { | |
4361 | btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu", | |
4362 | left, thresh, type); | |
4363 | dump_space_info(info, 0, 0); | |
4364 | } | |
4365 | ||
4366 | if (left < thresh) { | |
4367 | u64 flags; | |
4368 | ||
4369 | flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0); | |
4370 | /* | |
4371 | * Ignore failure to create system chunk. We might end up not | |
4372 | * needing it, as we might not need to COW all nodes/leafs from | |
4373 | * the paths we visit in the chunk tree (they were already COWed | |
4374 | * or created in the current transaction for example). | |
4375 | */ | |
4376 | ret = btrfs_alloc_chunk(trans, root, flags); | |
4377 | } | |
4378 | ||
4379 | if (!ret) { | |
4380 | ret = btrfs_block_rsv_add(root->fs_info->chunk_root, | |
4381 | &root->fs_info->chunk_block_rsv, | |
4382 | thresh, BTRFS_RESERVE_NO_FLUSH); | |
4383 | if (!ret) | |
4384 | trans->chunk_bytes_reserved += thresh; | |
4385 | } | |
4386 | } | |
4387 | ||
4388 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, | |
4389 | struct btrfs_root *extent_root, u64 flags, int force) | |
4390 | { | |
4391 | struct btrfs_space_info *space_info; | |
4392 | struct btrfs_fs_info *fs_info = extent_root->fs_info; | |
4393 | int wait_for_alloc = 0; | |
4394 | int ret = 0; | |
4395 | ||
4396 | /* Don't re-enter if we're already allocating a chunk */ | |
4397 | if (trans->allocating_chunk) | |
4398 | return -ENOSPC; | |
4399 | ||
4400 | space_info = __find_space_info(extent_root->fs_info, flags); | |
4401 | if (!space_info) { | |
4402 | ret = update_space_info(extent_root->fs_info, flags, | |
4403 | 0, 0, &space_info); | |
4404 | BUG_ON(ret); /* -ENOMEM */ | |
4405 | } | |
4406 | BUG_ON(!space_info); /* Logic error */ | |
4407 | ||
4408 | again: | |
4409 | spin_lock(&space_info->lock); | |
4410 | if (force < space_info->force_alloc) | |
4411 | force = space_info->force_alloc; | |
4412 | if (space_info->full) { | |
4413 | if (should_alloc_chunk(extent_root, space_info, force)) | |
4414 | ret = -ENOSPC; | |
4415 | else | |
4416 | ret = 0; | |
4417 | spin_unlock(&space_info->lock); | |
4418 | return ret; | |
4419 | } | |
4420 | ||
4421 | if (!should_alloc_chunk(extent_root, space_info, force)) { | |
4422 | spin_unlock(&space_info->lock); | |
4423 | return 0; | |
4424 | } else if (space_info->chunk_alloc) { | |
4425 | wait_for_alloc = 1; | |
4426 | } else { | |
4427 | space_info->chunk_alloc = 1; | |
4428 | } | |
4429 | ||
4430 | spin_unlock(&space_info->lock); | |
4431 | ||
4432 | mutex_lock(&fs_info->chunk_mutex); | |
4433 | ||
4434 | /* | |
4435 | * The chunk_mutex is held throughout the entirety of a chunk | |
4436 | * allocation, so once we've acquired the chunk_mutex we know that the | |
4437 | * other guy is done and we need to recheck and see if we should | |
4438 | * allocate. | |
4439 | */ | |
4440 | if (wait_for_alloc) { | |
4441 | mutex_unlock(&fs_info->chunk_mutex); | |
4442 | wait_for_alloc = 0; | |
4443 | goto again; | |
4444 | } | |
4445 | ||
4446 | trans->allocating_chunk = true; | |
4447 | ||
4448 | /* | |
4449 | * If we have mixed data/metadata chunks we want to make sure we keep | |
4450 | * allocating mixed chunks instead of individual chunks. | |
4451 | */ | |
4452 | if (btrfs_mixed_space_info(space_info)) | |
4453 | flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA); | |
4454 | ||
4455 | /* | |
4456 | * if we're doing a data chunk, go ahead and make sure that | |
4457 | * we keep a reasonable number of metadata chunks allocated in the | |
4458 | * FS as well. | |
4459 | */ | |
4460 | if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) { | |
4461 | fs_info->data_chunk_allocations++; | |
4462 | if (!(fs_info->data_chunk_allocations % | |
4463 | fs_info->metadata_ratio)) | |
4464 | force_metadata_allocation(fs_info); | |
4465 | } | |
4466 | ||
4467 | /* | |
4468 | * Check if we have enough space in SYSTEM chunk because we may need | |
4469 | * to update devices. | |
4470 | */ | |
4471 | check_system_chunk(trans, extent_root, flags); | |
4472 | ||
4473 | ret = btrfs_alloc_chunk(trans, extent_root, flags); | |
4474 | trans->allocating_chunk = false; | |
4475 | ||
4476 | spin_lock(&space_info->lock); | |
4477 | if (ret < 0 && ret != -ENOSPC) | |
4478 | goto out; | |
4479 | if (ret) | |
4480 | space_info->full = 1; | |
4481 | else | |
4482 | ret = 1; | |
4483 | ||
4484 | space_info->force_alloc = CHUNK_ALLOC_NO_FORCE; | |
4485 | out: | |
4486 | space_info->chunk_alloc = 0; | |
4487 | spin_unlock(&space_info->lock); | |
4488 | mutex_unlock(&fs_info->chunk_mutex); | |
4489 | /* | |
4490 | * When we allocate a new chunk we reserve space in the chunk block | |
4491 | * reserve to make sure we can COW nodes/leafs in the chunk tree or | |
4492 | * add new nodes/leafs to it if we end up needing to do it when | |
4493 | * inserting the chunk item and updating device items as part of the | |
4494 | * second phase of chunk allocation, performed by | |
4495 | * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a | |
4496 | * large number of new block groups to create in our transaction | |
4497 | * handle's new_bgs list to avoid exhausting the chunk block reserve | |
4498 | * in extreme cases - like having a single transaction create many new | |
4499 | * block groups when starting to write out the free space caches of all | |
4500 | * the block groups that were made dirty during the lifetime of the | |
4501 | * transaction. | |
4502 | */ | |
4503 | if (trans->can_flush_pending_bgs && | |
4504 | trans->chunk_bytes_reserved >= (u64)SZ_2M) { | |
4505 | btrfs_create_pending_block_groups(trans, trans->root); | |
4506 | btrfs_trans_release_chunk_metadata(trans); | |
4507 | } | |
4508 | return ret; | |
4509 | } | |
4510 | ||
4511 | static int can_overcommit(struct btrfs_root *root, | |
4512 | struct btrfs_space_info *space_info, u64 bytes, | |
4513 | enum btrfs_reserve_flush_enum flush) | |
4514 | { | |
4515 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; | |
4516 | u64 profile = btrfs_get_alloc_profile(root, 0); | |
4517 | u64 space_size; | |
4518 | u64 avail; | |
4519 | u64 used; | |
4520 | ||
4521 | used = space_info->bytes_used + space_info->bytes_reserved + | |
4522 | space_info->bytes_pinned + space_info->bytes_readonly; | |
4523 | ||
4524 | /* | |
4525 | * We only want to allow over committing if we have lots of actual space | |
4526 | * free, but if we don't have enough space to handle the global reserve | |
4527 | * space then we could end up having a real enospc problem when trying | |
4528 | * to allocate a chunk or some other such important allocation. | |
4529 | */ | |
4530 | spin_lock(&global_rsv->lock); | |
4531 | space_size = calc_global_rsv_need_space(global_rsv); | |
4532 | spin_unlock(&global_rsv->lock); | |
4533 | if (used + space_size >= space_info->total_bytes) | |
4534 | return 0; | |
4535 | ||
4536 | used += space_info->bytes_may_use; | |
4537 | ||
4538 | spin_lock(&root->fs_info->free_chunk_lock); | |
4539 | avail = root->fs_info->free_chunk_space; | |
4540 | spin_unlock(&root->fs_info->free_chunk_lock); | |
4541 | ||
4542 | /* | |
4543 | * If we have dup, raid1 or raid10 then only half of the free | |
4544 | * space is actually useable. For raid56, the space info used | |
4545 | * doesn't include the parity drive, so we don't have to | |
4546 | * change the math | |
4547 | */ | |
4548 | if (profile & (BTRFS_BLOCK_GROUP_DUP | | |
4549 | BTRFS_BLOCK_GROUP_RAID1 | | |
4550 | BTRFS_BLOCK_GROUP_RAID10)) | |
4551 | avail >>= 1; | |
4552 | ||
4553 | /* | |
4554 | * If we aren't flushing all things, let us overcommit up to | |
4555 | * 1/2th of the space. If we can flush, don't let us overcommit | |
4556 | * too much, let it overcommit up to 1/8 of the space. | |
4557 | */ | |
4558 | if (flush == BTRFS_RESERVE_FLUSH_ALL) | |
4559 | avail >>= 3; | |
4560 | else | |
4561 | avail >>= 1; | |
4562 | ||
4563 | if (used + bytes < space_info->total_bytes + avail) | |
4564 | return 1; | |
4565 | return 0; | |
4566 | } | |
4567 | ||
4568 | static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root, | |
4569 | unsigned long nr_pages, int nr_items) | |
4570 | { | |
4571 | struct super_block *sb = root->fs_info->sb; | |
4572 | ||
4573 | if (down_read_trylock(&sb->s_umount)) { | |
4574 | writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE); | |
4575 | up_read(&sb->s_umount); | |
4576 | } else { | |
4577 | /* | |
4578 | * We needn't worry the filesystem going from r/w to r/o though | |
4579 | * we don't acquire ->s_umount mutex, because the filesystem | |
4580 | * should guarantee the delalloc inodes list be empty after | |
4581 | * the filesystem is readonly(all dirty pages are written to | |
4582 | * the disk). | |
4583 | */ | |
4584 | btrfs_start_delalloc_roots(root->fs_info, 0, nr_items); | |
4585 | if (!current->journal_info) | |
4586 | btrfs_wait_ordered_roots(root->fs_info, nr_items, | |
4587 | 0, (u64)-1); | |
4588 | } | |
4589 | } | |
4590 | ||
4591 | static inline int calc_reclaim_items_nr(struct btrfs_root *root, u64 to_reclaim) | |
4592 | { | |
4593 | u64 bytes; | |
4594 | int nr; | |
4595 | ||
4596 | bytes = btrfs_calc_trans_metadata_size(root, 1); | |
4597 | nr = (int)div64_u64(to_reclaim, bytes); | |
4598 | if (!nr) | |
4599 | nr = 1; | |
4600 | return nr; | |
4601 | } | |
4602 | ||
4603 | #define EXTENT_SIZE_PER_ITEM SZ_256K | |
4604 | ||
4605 | /* | |
4606 | * shrink metadata reservation for delalloc | |
4607 | */ | |
4608 | static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig, | |
4609 | bool wait_ordered) | |
4610 | { | |
4611 | struct btrfs_block_rsv *block_rsv; | |
4612 | struct btrfs_space_info *space_info; | |
4613 | struct btrfs_trans_handle *trans; | |
4614 | u64 delalloc_bytes; | |
4615 | u64 max_reclaim; | |
4616 | long time_left; | |
4617 | unsigned long nr_pages; | |
4618 | int loops; | |
4619 | int items; | |
4620 | enum btrfs_reserve_flush_enum flush; | |
4621 | ||
4622 | /* Calc the number of the pages we need flush for space reservation */ | |
4623 | items = calc_reclaim_items_nr(root, to_reclaim); | |
4624 | to_reclaim = items * EXTENT_SIZE_PER_ITEM; | |
4625 | ||
4626 | trans = (struct btrfs_trans_handle *)current->journal_info; | |
4627 | block_rsv = &root->fs_info->delalloc_block_rsv; | |
4628 | space_info = block_rsv->space_info; | |
4629 | ||
4630 | delalloc_bytes = percpu_counter_sum_positive( | |
4631 | &root->fs_info->delalloc_bytes); | |
4632 | if (delalloc_bytes == 0) { | |
4633 | if (trans) | |
4634 | return; | |
4635 | if (wait_ordered) | |
4636 | btrfs_wait_ordered_roots(root->fs_info, items, | |
4637 | 0, (u64)-1); | |
4638 | return; | |
4639 | } | |
4640 | ||
4641 | loops = 0; | |
4642 | while (delalloc_bytes && loops < 3) { | |
4643 | max_reclaim = min(delalloc_bytes, to_reclaim); | |
4644 | nr_pages = max_reclaim >> PAGE_SHIFT; | |
4645 | btrfs_writeback_inodes_sb_nr(root, nr_pages, items); | |
4646 | /* | |
4647 | * We need to wait for the async pages to actually start before | |
4648 | * we do anything. | |
4649 | */ | |
4650 | max_reclaim = atomic_read(&root->fs_info->async_delalloc_pages); | |
4651 | if (!max_reclaim) | |
4652 | goto skip_async; | |
4653 | ||
4654 | if (max_reclaim <= nr_pages) | |
4655 | max_reclaim = 0; | |
4656 | else | |
4657 | max_reclaim -= nr_pages; | |
4658 | ||
4659 | wait_event(root->fs_info->async_submit_wait, | |
4660 | atomic_read(&root->fs_info->async_delalloc_pages) <= | |
4661 | (int)max_reclaim); | |
4662 | skip_async: | |
4663 | if (!trans) | |
4664 | flush = BTRFS_RESERVE_FLUSH_ALL; | |
4665 | else | |
4666 | flush = BTRFS_RESERVE_NO_FLUSH; | |
4667 | spin_lock(&space_info->lock); | |
4668 | if (can_overcommit(root, space_info, orig, flush)) { | |
4669 | spin_unlock(&space_info->lock); | |
4670 | break; | |
4671 | } | |
4672 | spin_unlock(&space_info->lock); | |
4673 | ||
4674 | loops++; | |
4675 | if (wait_ordered && !trans) { | |
4676 | btrfs_wait_ordered_roots(root->fs_info, items, | |
4677 | 0, (u64)-1); | |
4678 | } else { | |
4679 | time_left = schedule_timeout_killable(1); | |
4680 | if (time_left) | |
4681 | break; | |
4682 | } | |
4683 | delalloc_bytes = percpu_counter_sum_positive( | |
4684 | &root->fs_info->delalloc_bytes); | |
4685 | } | |
4686 | } | |
4687 | ||
4688 | /** | |
4689 | * maybe_commit_transaction - possibly commit the transaction if its ok to | |
4690 | * @root - the root we're allocating for | |
4691 | * @bytes - the number of bytes we want to reserve | |
4692 | * @force - force the commit | |
4693 | * | |
4694 | * This will check to make sure that committing the transaction will actually | |
4695 | * get us somewhere and then commit the transaction if it does. Otherwise it | |
4696 | * will return -ENOSPC. | |
4697 | */ | |
4698 | static int may_commit_transaction(struct btrfs_root *root, | |
4699 | struct btrfs_space_info *space_info, | |
4700 | u64 bytes, int force) | |
4701 | { | |
4702 | struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv; | |
4703 | struct btrfs_trans_handle *trans; | |
4704 | ||
4705 | trans = (struct btrfs_trans_handle *)current->journal_info; | |
4706 | if (trans) | |
4707 | return -EAGAIN; | |
4708 | ||
4709 | if (force) | |
4710 | goto commit; | |
4711 | ||
4712 | /* See if there is enough pinned space to make this reservation */ | |
4713 | if (percpu_counter_compare(&space_info->total_bytes_pinned, | |
4714 | bytes) >= 0) | |
4715 | goto commit; | |
4716 | ||
4717 | /* | |
4718 | * See if there is some space in the delayed insertion reservation for | |
4719 | * this reservation. | |
4720 | */ | |
4721 | if (space_info != delayed_rsv->space_info) | |
4722 | return -ENOSPC; | |
4723 | ||
4724 | spin_lock(&delayed_rsv->lock); | |
4725 | if (percpu_counter_compare(&space_info->total_bytes_pinned, | |
4726 | bytes - delayed_rsv->size) >= 0) { | |
4727 | spin_unlock(&delayed_rsv->lock); | |
4728 | return -ENOSPC; | |
4729 | } | |
4730 | spin_unlock(&delayed_rsv->lock); | |
4731 | ||
4732 | commit: | |
4733 | trans = btrfs_join_transaction(root); | |
4734 | if (IS_ERR(trans)) | |
4735 | return -ENOSPC; | |
4736 | ||
4737 | return btrfs_commit_transaction(trans, root); | |
4738 | } | |
4739 | ||
4740 | enum flush_state { | |
4741 | FLUSH_DELAYED_ITEMS_NR = 1, | |
4742 | FLUSH_DELAYED_ITEMS = 2, | |
4743 | FLUSH_DELALLOC = 3, | |
4744 | FLUSH_DELALLOC_WAIT = 4, | |
4745 | ALLOC_CHUNK = 5, | |
4746 | COMMIT_TRANS = 6, | |
4747 | }; | |
4748 | ||
4749 | static int flush_space(struct btrfs_root *root, | |
4750 | struct btrfs_space_info *space_info, u64 num_bytes, | |
4751 | u64 orig_bytes, int state) | |
4752 | { | |
4753 | struct btrfs_trans_handle *trans; | |
4754 | int nr; | |
4755 | int ret = 0; | |
4756 | ||
4757 | switch (state) { | |
4758 | case FLUSH_DELAYED_ITEMS_NR: | |
4759 | case FLUSH_DELAYED_ITEMS: | |
4760 | if (state == FLUSH_DELAYED_ITEMS_NR) | |
4761 | nr = calc_reclaim_items_nr(root, num_bytes) * 2; | |
4762 | else | |
4763 | nr = -1; | |
4764 | ||
4765 | trans = btrfs_join_transaction(root); | |
4766 | if (IS_ERR(trans)) { | |
4767 | ret = PTR_ERR(trans); | |
4768 | break; | |
4769 | } | |
4770 | ret = btrfs_run_delayed_items_nr(trans, root, nr); | |
4771 | btrfs_end_transaction(trans, root); | |
4772 | break; | |
4773 | case FLUSH_DELALLOC: | |
4774 | case FLUSH_DELALLOC_WAIT: | |
4775 | shrink_delalloc(root, num_bytes * 2, orig_bytes, | |
4776 | state == FLUSH_DELALLOC_WAIT); | |
4777 | break; | |
4778 | case ALLOC_CHUNK: | |
4779 | trans = btrfs_join_transaction(root); | |
4780 | if (IS_ERR(trans)) { | |
4781 | ret = PTR_ERR(trans); | |
4782 | break; | |
4783 | } | |
4784 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, | |
4785 | btrfs_get_alloc_profile(root, 0), | |
4786 | CHUNK_ALLOC_NO_FORCE); | |
4787 | btrfs_end_transaction(trans, root); | |
4788 | if (ret == -ENOSPC) | |
4789 | ret = 0; | |
4790 | break; | |
4791 | case COMMIT_TRANS: | |
4792 | ret = may_commit_transaction(root, space_info, orig_bytes, 0); | |
4793 | break; | |
4794 | default: | |
4795 | ret = -ENOSPC; | |
4796 | break; | |
4797 | } | |
4798 | ||
4799 | return ret; | |
4800 | } | |
4801 | ||
4802 | static inline u64 | |
4803 | btrfs_calc_reclaim_metadata_size(struct btrfs_root *root, | |
4804 | struct btrfs_space_info *space_info) | |
4805 | { | |
4806 | u64 used; | |
4807 | u64 expected; | |
4808 | u64 to_reclaim; | |
4809 | ||
4810 | to_reclaim = min_t(u64, num_online_cpus() * SZ_1M, SZ_16M); | |
4811 | spin_lock(&space_info->lock); | |
4812 | if (can_overcommit(root, space_info, to_reclaim, | |
4813 | BTRFS_RESERVE_FLUSH_ALL)) { | |
4814 | to_reclaim = 0; | |
4815 | goto out; | |
4816 | } | |
4817 | ||
4818 | used = space_info->bytes_used + space_info->bytes_reserved + | |
4819 | space_info->bytes_pinned + space_info->bytes_readonly + | |
4820 | space_info->bytes_may_use; | |
4821 | if (can_overcommit(root, space_info, SZ_1M, BTRFS_RESERVE_FLUSH_ALL)) | |
4822 | expected = div_factor_fine(space_info->total_bytes, 95); | |
4823 | else | |
4824 | expected = div_factor_fine(space_info->total_bytes, 90); | |
4825 | ||
4826 | if (used > expected) | |
4827 | to_reclaim = used - expected; | |
4828 | else | |
4829 | to_reclaim = 0; | |
4830 | to_reclaim = min(to_reclaim, space_info->bytes_may_use + | |
4831 | space_info->bytes_reserved); | |
4832 | out: | |
4833 | spin_unlock(&space_info->lock); | |
4834 | ||
4835 | return to_reclaim; | |
4836 | } | |
4837 | ||
4838 | static inline int need_do_async_reclaim(struct btrfs_space_info *space_info, | |
4839 | struct btrfs_fs_info *fs_info, u64 used) | |
4840 | { | |
4841 | u64 thresh = div_factor_fine(space_info->total_bytes, 98); | |
4842 | ||
4843 | /* If we're just plain full then async reclaim just slows us down. */ | |
4844 | if ((space_info->bytes_used + space_info->bytes_reserved) >= thresh) | |
4845 | return 0; | |
4846 | ||
4847 | return (used >= thresh && !btrfs_fs_closing(fs_info) && | |
4848 | !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state)); | |
4849 | } | |
4850 | ||
4851 | static int btrfs_need_do_async_reclaim(struct btrfs_space_info *space_info, | |
4852 | struct btrfs_fs_info *fs_info, | |
4853 | int flush_state) | |
4854 | { | |
4855 | u64 used; | |
4856 | ||
4857 | spin_lock(&space_info->lock); | |
4858 | /* | |
4859 | * We run out of space and have not got any free space via flush_space, | |
4860 | * so don't bother doing async reclaim. | |
4861 | */ | |
4862 | if (flush_state > COMMIT_TRANS && space_info->full) { | |
4863 | spin_unlock(&space_info->lock); | |
4864 | return 0; | |
4865 | } | |
4866 | ||
4867 | used = space_info->bytes_used + space_info->bytes_reserved + | |
4868 | space_info->bytes_pinned + space_info->bytes_readonly + | |
4869 | space_info->bytes_may_use; | |
4870 | if (need_do_async_reclaim(space_info, fs_info, used)) { | |
4871 | spin_unlock(&space_info->lock); | |
4872 | return 1; | |
4873 | } | |
4874 | spin_unlock(&space_info->lock); | |
4875 | ||
4876 | return 0; | |
4877 | } | |
4878 | ||
4879 | static void btrfs_async_reclaim_metadata_space(struct work_struct *work) | |
4880 | { | |
4881 | struct btrfs_fs_info *fs_info; | |
4882 | struct btrfs_space_info *space_info; | |
4883 | u64 to_reclaim; | |
4884 | int flush_state; | |
4885 | ||
4886 | fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work); | |
4887 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); | |
4888 | ||
4889 | to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root, | |
4890 | space_info); | |
4891 | if (!to_reclaim) | |
4892 | return; | |
4893 | ||
4894 | flush_state = FLUSH_DELAYED_ITEMS_NR; | |
4895 | do { | |
4896 | flush_space(fs_info->fs_root, space_info, to_reclaim, | |
4897 | to_reclaim, flush_state); | |
4898 | flush_state++; | |
4899 | if (!btrfs_need_do_async_reclaim(space_info, fs_info, | |
4900 | flush_state)) | |
4901 | return; | |
4902 | } while (flush_state < COMMIT_TRANS); | |
4903 | } | |
4904 | ||
4905 | void btrfs_init_async_reclaim_work(struct work_struct *work) | |
4906 | { | |
4907 | INIT_WORK(work, btrfs_async_reclaim_metadata_space); | |
4908 | } | |
4909 | ||
4910 | /** | |
4911 | * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space | |
4912 | * @root - the root we're allocating for | |
4913 | * @block_rsv - the block_rsv we're allocating for | |
4914 | * @orig_bytes - the number of bytes we want | |
4915 | * @flush - whether or not we can flush to make our reservation | |
4916 | * | |
4917 | * This will reserve orgi_bytes number of bytes from the space info associated | |
4918 | * with the block_rsv. If there is not enough space it will make an attempt to | |
4919 | * flush out space to make room. It will do this by flushing delalloc if | |
4920 | * possible or committing the transaction. If flush is 0 then no attempts to | |
4921 | * regain reservations will be made and this will fail if there is not enough | |
4922 | * space already. | |
4923 | */ | |
4924 | static int reserve_metadata_bytes(struct btrfs_root *root, | |
4925 | struct btrfs_block_rsv *block_rsv, | |
4926 | u64 orig_bytes, | |
4927 | enum btrfs_reserve_flush_enum flush) | |
4928 | { | |
4929 | struct btrfs_space_info *space_info = block_rsv->space_info; | |
4930 | u64 used; | |
4931 | u64 num_bytes = orig_bytes; | |
4932 | int flush_state = FLUSH_DELAYED_ITEMS_NR; | |
4933 | int ret = 0; | |
4934 | bool flushing = false; | |
4935 | ||
4936 | again: | |
4937 | ret = 0; | |
4938 | spin_lock(&space_info->lock); | |
4939 | /* | |
4940 | * We only want to wait if somebody other than us is flushing and we | |
4941 | * are actually allowed to flush all things. | |
4942 | */ | |
4943 | while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing && | |
4944 | space_info->flush) { | |
4945 | spin_unlock(&space_info->lock); | |
4946 | /* | |
4947 | * If we have a trans handle we can't wait because the flusher | |
4948 | * may have to commit the transaction, which would mean we would | |
4949 | * deadlock since we are waiting for the flusher to finish, but | |
4950 | * hold the current transaction open. | |
4951 | */ | |
4952 | if (current->journal_info) | |
4953 | return -EAGAIN; | |
4954 | ret = wait_event_killable(space_info->wait, !space_info->flush); | |
4955 | /* Must have been killed, return */ | |
4956 | if (ret) | |
4957 | return -EINTR; | |
4958 | ||
4959 | spin_lock(&space_info->lock); | |
4960 | } | |
4961 | ||
4962 | ret = -ENOSPC; | |
4963 | used = space_info->bytes_used + space_info->bytes_reserved + | |
4964 | space_info->bytes_pinned + space_info->bytes_readonly + | |
4965 | space_info->bytes_may_use; | |
4966 | ||
4967 | /* | |
4968 | * The idea here is that we've not already over-reserved the block group | |
4969 | * then we can go ahead and save our reservation first and then start | |
4970 | * flushing if we need to. Otherwise if we've already overcommitted | |
4971 | * lets start flushing stuff first and then come back and try to make | |
4972 | * our reservation. | |
4973 | */ | |
4974 | if (used <= space_info->total_bytes) { | |
4975 | if (used + orig_bytes <= space_info->total_bytes) { | |
4976 | space_info->bytes_may_use += orig_bytes; | |
4977 | trace_btrfs_space_reservation(root->fs_info, | |
4978 | "space_info", space_info->flags, orig_bytes, 1); | |
4979 | ret = 0; | |
4980 | } else { | |
4981 | /* | |
4982 | * Ok set num_bytes to orig_bytes since we aren't | |
4983 | * overocmmitted, this way we only try and reclaim what | |
4984 | * we need. | |
4985 | */ | |
4986 | num_bytes = orig_bytes; | |
4987 | } | |
4988 | } else { | |
4989 | /* | |
4990 | * Ok we're over committed, set num_bytes to the overcommitted | |
4991 | * amount plus the amount of bytes that we need for this | |
4992 | * reservation. | |
4993 | */ | |
4994 | num_bytes = used - space_info->total_bytes + | |
4995 | (orig_bytes * 2); | |
4996 | } | |
4997 | ||
4998 | if (ret && can_overcommit(root, space_info, orig_bytes, flush)) { | |
4999 | space_info->bytes_may_use += orig_bytes; | |
5000 | trace_btrfs_space_reservation(root->fs_info, "space_info", | |
5001 | space_info->flags, orig_bytes, | |
5002 | 1); | |
5003 | ret = 0; | |
5004 | } | |
5005 | ||
5006 | /* | |
5007 | * Couldn't make our reservation, save our place so while we're trying | |
5008 | * to reclaim space we can actually use it instead of somebody else | |
5009 | * stealing it from us. | |
5010 | * | |
5011 | * We make the other tasks wait for the flush only when we can flush | |
5012 | * all things. | |
5013 | */ | |
5014 | if (ret && flush != BTRFS_RESERVE_NO_FLUSH) { | |
5015 | flushing = true; | |
5016 | space_info->flush = 1; | |
5017 | } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) { | |
5018 | used += orig_bytes; | |
5019 | /* | |
5020 | * We will do the space reservation dance during log replay, | |
5021 | * which means we won't have fs_info->fs_root set, so don't do | |
5022 | * the async reclaim as we will panic. | |
5023 | */ | |
5024 | if (!root->fs_info->log_root_recovering && | |
5025 | need_do_async_reclaim(space_info, root->fs_info, used) && | |
5026 | !work_busy(&root->fs_info->async_reclaim_work)) | |
5027 | queue_work(system_unbound_wq, | |
5028 | &root->fs_info->async_reclaim_work); | |
5029 | } | |
5030 | spin_unlock(&space_info->lock); | |
5031 | ||
5032 | if (!ret || flush == BTRFS_RESERVE_NO_FLUSH) | |
5033 | goto out; | |
5034 | ||
5035 | ret = flush_space(root, space_info, num_bytes, orig_bytes, | |
5036 | flush_state); | |
5037 | flush_state++; | |
5038 | ||
5039 | /* | |
5040 | * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock | |
5041 | * would happen. So skip delalloc flush. | |
5042 | */ | |
5043 | if (flush == BTRFS_RESERVE_FLUSH_LIMIT && | |
5044 | (flush_state == FLUSH_DELALLOC || | |
5045 | flush_state == FLUSH_DELALLOC_WAIT)) | |
5046 | flush_state = ALLOC_CHUNK; | |
5047 | ||
5048 | if (!ret) | |
5049 | goto again; | |
5050 | else if (flush == BTRFS_RESERVE_FLUSH_LIMIT && | |
5051 | flush_state < COMMIT_TRANS) | |
5052 | goto again; | |
5053 | else if (flush == BTRFS_RESERVE_FLUSH_ALL && | |
5054 | flush_state <= COMMIT_TRANS) | |
5055 | goto again; | |
5056 | ||
5057 | out: | |
5058 | if (ret == -ENOSPC && | |
5059 | unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) { | |
5060 | struct btrfs_block_rsv *global_rsv = | |
5061 | &root->fs_info->global_block_rsv; | |
5062 | ||
5063 | if (block_rsv != global_rsv && | |
5064 | !block_rsv_use_bytes(global_rsv, orig_bytes)) | |
5065 | ret = 0; | |
5066 | } | |
5067 | if (ret == -ENOSPC) | |
5068 | trace_btrfs_space_reservation(root->fs_info, | |
5069 | "space_info:enospc", | |
5070 | space_info->flags, orig_bytes, 1); | |
5071 | if (flushing) { | |
5072 | spin_lock(&space_info->lock); | |
5073 | space_info->flush = 0; | |
5074 | wake_up_all(&space_info->wait); | |
5075 | spin_unlock(&space_info->lock); | |
5076 | } | |
5077 | return ret; | |
5078 | } | |
5079 | ||
5080 | static struct btrfs_block_rsv *get_block_rsv( | |
5081 | const struct btrfs_trans_handle *trans, | |
5082 | const struct btrfs_root *root) | |
5083 | { | |
5084 | struct btrfs_block_rsv *block_rsv = NULL; | |
5085 | ||
5086 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) || | |
5087 | (root == root->fs_info->csum_root && trans->adding_csums) || | |
5088 | (root == root->fs_info->uuid_root)) | |
5089 | block_rsv = trans->block_rsv; | |
5090 | ||
5091 | if (!block_rsv) | |
5092 | block_rsv = root->block_rsv; | |
5093 | ||
5094 | if (!block_rsv) | |
5095 | block_rsv = &root->fs_info->empty_block_rsv; | |
5096 | ||
5097 | return block_rsv; | |
5098 | } | |
5099 | ||
5100 | static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, | |
5101 | u64 num_bytes) | |
5102 | { | |
5103 | int ret = -ENOSPC; | |
5104 | spin_lock(&block_rsv->lock); | |
5105 | if (block_rsv->reserved >= num_bytes) { | |
5106 | block_rsv->reserved -= num_bytes; | |
5107 | if (block_rsv->reserved < block_rsv->size) | |
5108 | block_rsv->full = 0; | |
5109 | ret = 0; | |
5110 | } | |
5111 | spin_unlock(&block_rsv->lock); | |
5112 | return ret; | |
5113 | } | |
5114 | ||
5115 | static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv, | |
5116 | u64 num_bytes, int update_size) | |
5117 | { | |
5118 | spin_lock(&block_rsv->lock); | |
5119 | block_rsv->reserved += num_bytes; | |
5120 | if (update_size) | |
5121 | block_rsv->size += num_bytes; | |
5122 | else if (block_rsv->reserved >= block_rsv->size) | |
5123 | block_rsv->full = 1; | |
5124 | spin_unlock(&block_rsv->lock); | |
5125 | } | |
5126 | ||
5127 | int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info, | |
5128 | struct btrfs_block_rsv *dest, u64 num_bytes, | |
5129 | int min_factor) | |
5130 | { | |
5131 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
5132 | u64 min_bytes; | |
5133 | ||
5134 | if (global_rsv->space_info != dest->space_info) | |
5135 | return -ENOSPC; | |
5136 | ||
5137 | spin_lock(&global_rsv->lock); | |
5138 | min_bytes = div_factor(global_rsv->size, min_factor); | |
5139 | if (global_rsv->reserved < min_bytes + num_bytes) { | |
5140 | spin_unlock(&global_rsv->lock); | |
5141 | return -ENOSPC; | |
5142 | } | |
5143 | global_rsv->reserved -= num_bytes; | |
5144 | if (global_rsv->reserved < global_rsv->size) | |
5145 | global_rsv->full = 0; | |
5146 | spin_unlock(&global_rsv->lock); | |
5147 | ||
5148 | block_rsv_add_bytes(dest, num_bytes, 1); | |
5149 | return 0; | |
5150 | } | |
5151 | ||
5152 | static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info, | |
5153 | struct btrfs_block_rsv *block_rsv, | |
5154 | struct btrfs_block_rsv *dest, u64 num_bytes) | |
5155 | { | |
5156 | struct btrfs_space_info *space_info = block_rsv->space_info; | |
5157 | ||
5158 | spin_lock(&block_rsv->lock); | |
5159 | if (num_bytes == (u64)-1) | |
5160 | num_bytes = block_rsv->size; | |
5161 | block_rsv->size -= num_bytes; | |
5162 | if (block_rsv->reserved >= block_rsv->size) { | |
5163 | num_bytes = block_rsv->reserved - block_rsv->size; | |
5164 | block_rsv->reserved = block_rsv->size; | |
5165 | block_rsv->full = 1; | |
5166 | } else { | |
5167 | num_bytes = 0; | |
5168 | } | |
5169 | spin_unlock(&block_rsv->lock); | |
5170 | ||
5171 | if (num_bytes > 0) { | |
5172 | if (dest) { | |
5173 | spin_lock(&dest->lock); | |
5174 | if (!dest->full) { | |
5175 | u64 bytes_to_add; | |
5176 | ||
5177 | bytes_to_add = dest->size - dest->reserved; | |
5178 | bytes_to_add = min(num_bytes, bytes_to_add); | |
5179 | dest->reserved += bytes_to_add; | |
5180 | if (dest->reserved >= dest->size) | |
5181 | dest->full = 1; | |
5182 | num_bytes -= bytes_to_add; | |
5183 | } | |
5184 | spin_unlock(&dest->lock); | |
5185 | } | |
5186 | if (num_bytes) { | |
5187 | spin_lock(&space_info->lock); | |
5188 | space_info->bytes_may_use -= num_bytes; | |
5189 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5190 | space_info->flags, num_bytes, 0); | |
5191 | spin_unlock(&space_info->lock); | |
5192 | } | |
5193 | } | |
5194 | } | |
5195 | ||
5196 | static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src, | |
5197 | struct btrfs_block_rsv *dst, u64 num_bytes) | |
5198 | { | |
5199 | int ret; | |
5200 | ||
5201 | ret = block_rsv_use_bytes(src, num_bytes); | |
5202 | if (ret) | |
5203 | return ret; | |
5204 | ||
5205 | block_rsv_add_bytes(dst, num_bytes, 1); | |
5206 | return 0; | |
5207 | } | |
5208 | ||
5209 | void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type) | |
5210 | { | |
5211 | memset(rsv, 0, sizeof(*rsv)); | |
5212 | spin_lock_init(&rsv->lock); | |
5213 | rsv->type = type; | |
5214 | } | |
5215 | ||
5216 | struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root, | |
5217 | unsigned short type) | |
5218 | { | |
5219 | struct btrfs_block_rsv *block_rsv; | |
5220 | struct btrfs_fs_info *fs_info = root->fs_info; | |
5221 | ||
5222 | block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS); | |
5223 | if (!block_rsv) | |
5224 | return NULL; | |
5225 | ||
5226 | btrfs_init_block_rsv(block_rsv, type); | |
5227 | block_rsv->space_info = __find_space_info(fs_info, | |
5228 | BTRFS_BLOCK_GROUP_METADATA); | |
5229 | return block_rsv; | |
5230 | } | |
5231 | ||
5232 | void btrfs_free_block_rsv(struct btrfs_root *root, | |
5233 | struct btrfs_block_rsv *rsv) | |
5234 | { | |
5235 | if (!rsv) | |
5236 | return; | |
5237 | btrfs_block_rsv_release(root, rsv, (u64)-1); | |
5238 | kfree(rsv); | |
5239 | } | |
5240 | ||
5241 | void __btrfs_free_block_rsv(struct btrfs_block_rsv *rsv) | |
5242 | { | |
5243 | kfree(rsv); | |
5244 | } | |
5245 | ||
5246 | int btrfs_block_rsv_add(struct btrfs_root *root, | |
5247 | struct btrfs_block_rsv *block_rsv, u64 num_bytes, | |
5248 | enum btrfs_reserve_flush_enum flush) | |
5249 | { | |
5250 | int ret; | |
5251 | ||
5252 | if (num_bytes == 0) | |
5253 | return 0; | |
5254 | ||
5255 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); | |
5256 | if (!ret) { | |
5257 | block_rsv_add_bytes(block_rsv, num_bytes, 1); | |
5258 | return 0; | |
5259 | } | |
5260 | ||
5261 | return ret; | |
5262 | } | |
5263 | ||
5264 | int btrfs_block_rsv_check(struct btrfs_root *root, | |
5265 | struct btrfs_block_rsv *block_rsv, int min_factor) | |
5266 | { | |
5267 | u64 num_bytes = 0; | |
5268 | int ret = -ENOSPC; | |
5269 | ||
5270 | if (!block_rsv) | |
5271 | return 0; | |
5272 | ||
5273 | spin_lock(&block_rsv->lock); | |
5274 | num_bytes = div_factor(block_rsv->size, min_factor); | |
5275 | if (block_rsv->reserved >= num_bytes) | |
5276 | ret = 0; | |
5277 | spin_unlock(&block_rsv->lock); | |
5278 | ||
5279 | return ret; | |
5280 | } | |
5281 | ||
5282 | int btrfs_block_rsv_refill(struct btrfs_root *root, | |
5283 | struct btrfs_block_rsv *block_rsv, u64 min_reserved, | |
5284 | enum btrfs_reserve_flush_enum flush) | |
5285 | { | |
5286 | u64 num_bytes = 0; | |
5287 | int ret = -ENOSPC; | |
5288 | ||
5289 | if (!block_rsv) | |
5290 | return 0; | |
5291 | ||
5292 | spin_lock(&block_rsv->lock); | |
5293 | num_bytes = min_reserved; | |
5294 | if (block_rsv->reserved >= num_bytes) | |
5295 | ret = 0; | |
5296 | else | |
5297 | num_bytes -= block_rsv->reserved; | |
5298 | spin_unlock(&block_rsv->lock); | |
5299 | ||
5300 | if (!ret) | |
5301 | return 0; | |
5302 | ||
5303 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); | |
5304 | if (!ret) { | |
5305 | block_rsv_add_bytes(block_rsv, num_bytes, 0); | |
5306 | return 0; | |
5307 | } | |
5308 | ||
5309 | return ret; | |
5310 | } | |
5311 | ||
5312 | int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv, | |
5313 | struct btrfs_block_rsv *dst_rsv, | |
5314 | u64 num_bytes) | |
5315 | { | |
5316 | return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes); | |
5317 | } | |
5318 | ||
5319 | void btrfs_block_rsv_release(struct btrfs_root *root, | |
5320 | struct btrfs_block_rsv *block_rsv, | |
5321 | u64 num_bytes) | |
5322 | { | |
5323 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; | |
5324 | if (global_rsv == block_rsv || | |
5325 | block_rsv->space_info != global_rsv->space_info) | |
5326 | global_rsv = NULL; | |
5327 | block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv, | |
5328 | num_bytes); | |
5329 | } | |
5330 | ||
5331 | /* | |
5332 | * helper to calculate size of global block reservation. | |
5333 | * the desired value is sum of space used by extent tree, | |
5334 | * checksum tree and root tree | |
5335 | */ | |
5336 | static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info) | |
5337 | { | |
5338 | struct btrfs_space_info *sinfo; | |
5339 | u64 num_bytes; | |
5340 | u64 meta_used; | |
5341 | u64 data_used; | |
5342 | int csum_size = btrfs_super_csum_size(fs_info->super_copy); | |
5343 | ||
5344 | sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA); | |
5345 | spin_lock(&sinfo->lock); | |
5346 | data_used = sinfo->bytes_used; | |
5347 | spin_unlock(&sinfo->lock); | |
5348 | ||
5349 | sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); | |
5350 | spin_lock(&sinfo->lock); | |
5351 | if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA) | |
5352 | data_used = 0; | |
5353 | meta_used = sinfo->bytes_used; | |
5354 | spin_unlock(&sinfo->lock); | |
5355 | ||
5356 | num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) * | |
5357 | csum_size * 2; | |
5358 | num_bytes += div_u64(data_used + meta_used, 50); | |
5359 | ||
5360 | if (num_bytes * 3 > meta_used) | |
5361 | num_bytes = div_u64(meta_used, 3); | |
5362 | ||
5363 | return ALIGN(num_bytes, fs_info->extent_root->nodesize << 10); | |
5364 | } | |
5365 | ||
5366 | static void update_global_block_rsv(struct btrfs_fs_info *fs_info) | |
5367 | { | |
5368 | struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv; | |
5369 | struct btrfs_space_info *sinfo = block_rsv->space_info; | |
5370 | u64 num_bytes; | |
5371 | ||
5372 | num_bytes = calc_global_metadata_size(fs_info); | |
5373 | ||
5374 | spin_lock(&sinfo->lock); | |
5375 | spin_lock(&block_rsv->lock); | |
5376 | ||
5377 | block_rsv->size = min_t(u64, num_bytes, SZ_512M); | |
5378 | ||
5379 | if (block_rsv->reserved < block_rsv->size) { | |
5380 | num_bytes = sinfo->bytes_used + sinfo->bytes_pinned + | |
5381 | sinfo->bytes_reserved + sinfo->bytes_readonly + | |
5382 | sinfo->bytes_may_use; | |
5383 | if (sinfo->total_bytes > num_bytes) { | |
5384 | num_bytes = sinfo->total_bytes - num_bytes; | |
5385 | num_bytes = min(num_bytes, | |
5386 | block_rsv->size - block_rsv->reserved); | |
5387 | block_rsv->reserved += num_bytes; | |
5388 | sinfo->bytes_may_use += num_bytes; | |
5389 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5390 | sinfo->flags, num_bytes, | |
5391 | 1); | |
5392 | } | |
5393 | } else if (block_rsv->reserved > block_rsv->size) { | |
5394 | num_bytes = block_rsv->reserved - block_rsv->size; | |
5395 | sinfo->bytes_may_use -= num_bytes; | |
5396 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5397 | sinfo->flags, num_bytes, 0); | |
5398 | block_rsv->reserved = block_rsv->size; | |
5399 | } | |
5400 | ||
5401 | if (block_rsv->reserved == block_rsv->size) | |
5402 | block_rsv->full = 1; | |
5403 | else | |
5404 | block_rsv->full = 0; | |
5405 | ||
5406 | spin_unlock(&block_rsv->lock); | |
5407 | spin_unlock(&sinfo->lock); | |
5408 | } | |
5409 | ||
5410 | static void init_global_block_rsv(struct btrfs_fs_info *fs_info) | |
5411 | { | |
5412 | struct btrfs_space_info *space_info; | |
5413 | ||
5414 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); | |
5415 | fs_info->chunk_block_rsv.space_info = space_info; | |
5416 | ||
5417 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); | |
5418 | fs_info->global_block_rsv.space_info = space_info; | |
5419 | fs_info->delalloc_block_rsv.space_info = space_info; | |
5420 | fs_info->trans_block_rsv.space_info = space_info; | |
5421 | fs_info->empty_block_rsv.space_info = space_info; | |
5422 | fs_info->delayed_block_rsv.space_info = space_info; | |
5423 | ||
5424 | fs_info->extent_root->block_rsv = &fs_info->global_block_rsv; | |
5425 | fs_info->csum_root->block_rsv = &fs_info->global_block_rsv; | |
5426 | fs_info->dev_root->block_rsv = &fs_info->global_block_rsv; | |
5427 | fs_info->tree_root->block_rsv = &fs_info->global_block_rsv; | |
5428 | if (fs_info->quota_root) | |
5429 | fs_info->quota_root->block_rsv = &fs_info->global_block_rsv; | |
5430 | fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv; | |
5431 | ||
5432 | update_global_block_rsv(fs_info); | |
5433 | } | |
5434 | ||
5435 | static void release_global_block_rsv(struct btrfs_fs_info *fs_info) | |
5436 | { | |
5437 | block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL, | |
5438 | (u64)-1); | |
5439 | WARN_ON(fs_info->delalloc_block_rsv.size > 0); | |
5440 | WARN_ON(fs_info->delalloc_block_rsv.reserved > 0); | |
5441 | WARN_ON(fs_info->trans_block_rsv.size > 0); | |
5442 | WARN_ON(fs_info->trans_block_rsv.reserved > 0); | |
5443 | WARN_ON(fs_info->chunk_block_rsv.size > 0); | |
5444 | WARN_ON(fs_info->chunk_block_rsv.reserved > 0); | |
5445 | WARN_ON(fs_info->delayed_block_rsv.size > 0); | |
5446 | WARN_ON(fs_info->delayed_block_rsv.reserved > 0); | |
5447 | } | |
5448 | ||
5449 | void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans, | |
5450 | struct btrfs_root *root) | |
5451 | { | |
5452 | if (!trans->block_rsv) | |
5453 | return; | |
5454 | ||
5455 | if (!trans->bytes_reserved) | |
5456 | return; | |
5457 | ||
5458 | trace_btrfs_space_reservation(root->fs_info, "transaction", | |
5459 | trans->transid, trans->bytes_reserved, 0); | |
5460 | btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved); | |
5461 | trans->bytes_reserved = 0; | |
5462 | } | |
5463 | ||
5464 | /* | |
5465 | * To be called after all the new block groups attached to the transaction | |
5466 | * handle have been created (btrfs_create_pending_block_groups()). | |
5467 | */ | |
5468 | void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans) | |
5469 | { | |
5470 | struct btrfs_fs_info *fs_info = trans->root->fs_info; | |
5471 | ||
5472 | if (!trans->chunk_bytes_reserved) | |
5473 | return; | |
5474 | ||
5475 | WARN_ON_ONCE(!list_empty(&trans->new_bgs)); | |
5476 | ||
5477 | block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL, | |
5478 | trans->chunk_bytes_reserved); | |
5479 | trans->chunk_bytes_reserved = 0; | |
5480 | } | |
5481 | ||
5482 | /* Can only return 0 or -ENOSPC */ | |
5483 | int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans, | |
5484 | struct inode *inode) | |
5485 | { | |
5486 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5487 | struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root); | |
5488 | struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv; | |
5489 | ||
5490 | /* | |
5491 | * We need to hold space in order to delete our orphan item once we've | |
5492 | * added it, so this takes the reservation so we can release it later | |
5493 | * when we are truly done with the orphan item. | |
5494 | */ | |
5495 | u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1); | |
5496 | trace_btrfs_space_reservation(root->fs_info, "orphan", | |
5497 | btrfs_ino(inode), num_bytes, 1); | |
5498 | return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes); | |
5499 | } | |
5500 | ||
5501 | void btrfs_orphan_release_metadata(struct inode *inode) | |
5502 | { | |
5503 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5504 | u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1); | |
5505 | trace_btrfs_space_reservation(root->fs_info, "orphan", | |
5506 | btrfs_ino(inode), num_bytes, 0); | |
5507 | btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes); | |
5508 | } | |
5509 | ||
5510 | /* | |
5511 | * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation | |
5512 | * root: the root of the parent directory | |
5513 | * rsv: block reservation | |
5514 | * items: the number of items that we need do reservation | |
5515 | * qgroup_reserved: used to return the reserved size in qgroup | |
5516 | * | |
5517 | * This function is used to reserve the space for snapshot/subvolume | |
5518 | * creation and deletion. Those operations are different with the | |
5519 | * common file/directory operations, they change two fs/file trees | |
5520 | * and root tree, the number of items that the qgroup reserves is | |
5521 | * different with the free space reservation. So we can not use | |
5522 | * the space reseravtion mechanism in start_transaction(). | |
5523 | */ | |
5524 | int btrfs_subvolume_reserve_metadata(struct btrfs_root *root, | |
5525 | struct btrfs_block_rsv *rsv, | |
5526 | int items, | |
5527 | u64 *qgroup_reserved, | |
5528 | bool use_global_rsv) | |
5529 | { | |
5530 | u64 num_bytes; | |
5531 | int ret; | |
5532 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; | |
5533 | ||
5534 | if (root->fs_info->quota_enabled) { | |
5535 | /* One for parent inode, two for dir entries */ | |
5536 | num_bytes = 3 * root->nodesize; | |
5537 | ret = btrfs_qgroup_reserve_meta(root, num_bytes); | |
5538 | if (ret) | |
5539 | return ret; | |
5540 | } else { | |
5541 | num_bytes = 0; | |
5542 | } | |
5543 | ||
5544 | *qgroup_reserved = num_bytes; | |
5545 | ||
5546 | num_bytes = btrfs_calc_trans_metadata_size(root, items); | |
5547 | rsv->space_info = __find_space_info(root->fs_info, | |
5548 | BTRFS_BLOCK_GROUP_METADATA); | |
5549 | ret = btrfs_block_rsv_add(root, rsv, num_bytes, | |
5550 | BTRFS_RESERVE_FLUSH_ALL); | |
5551 | ||
5552 | if (ret == -ENOSPC && use_global_rsv) | |
5553 | ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes); | |
5554 | ||
5555 | if (ret && *qgroup_reserved) | |
5556 | btrfs_qgroup_free_meta(root, *qgroup_reserved); | |
5557 | ||
5558 | return ret; | |
5559 | } | |
5560 | ||
5561 | void btrfs_subvolume_release_metadata(struct btrfs_root *root, | |
5562 | struct btrfs_block_rsv *rsv, | |
5563 | u64 qgroup_reserved) | |
5564 | { | |
5565 | btrfs_block_rsv_release(root, rsv, (u64)-1); | |
5566 | } | |
5567 | ||
5568 | /** | |
5569 | * drop_outstanding_extent - drop an outstanding extent | |
5570 | * @inode: the inode we're dropping the extent for | |
5571 | * @num_bytes: the number of bytes we're relaseing. | |
5572 | * | |
5573 | * This is called when we are freeing up an outstanding extent, either called | |
5574 | * after an error or after an extent is written. This will return the number of | |
5575 | * reserved extents that need to be freed. This must be called with | |
5576 | * BTRFS_I(inode)->lock held. | |
5577 | */ | |
5578 | static unsigned drop_outstanding_extent(struct inode *inode, u64 num_bytes) | |
5579 | { | |
5580 | unsigned drop_inode_space = 0; | |
5581 | unsigned dropped_extents = 0; | |
5582 | unsigned num_extents = 0; | |
5583 | ||
5584 | num_extents = (unsigned)div64_u64(num_bytes + | |
5585 | BTRFS_MAX_EXTENT_SIZE - 1, | |
5586 | BTRFS_MAX_EXTENT_SIZE); | |
5587 | ASSERT(num_extents); | |
5588 | ASSERT(BTRFS_I(inode)->outstanding_extents >= num_extents); | |
5589 | BTRFS_I(inode)->outstanding_extents -= num_extents; | |
5590 | ||
5591 | if (BTRFS_I(inode)->outstanding_extents == 0 && | |
5592 | test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED, | |
5593 | &BTRFS_I(inode)->runtime_flags)) | |
5594 | drop_inode_space = 1; | |
5595 | ||
5596 | /* | |
5597 | * If we have more or the same amount of outsanding extents than we have | |
5598 | * reserved then we need to leave the reserved extents count alone. | |
5599 | */ | |
5600 | if (BTRFS_I(inode)->outstanding_extents >= | |
5601 | BTRFS_I(inode)->reserved_extents) | |
5602 | return drop_inode_space; | |
5603 | ||
5604 | dropped_extents = BTRFS_I(inode)->reserved_extents - | |
5605 | BTRFS_I(inode)->outstanding_extents; | |
5606 | BTRFS_I(inode)->reserved_extents -= dropped_extents; | |
5607 | return dropped_extents + drop_inode_space; | |
5608 | } | |
5609 | ||
5610 | /** | |
5611 | * calc_csum_metadata_size - return the amount of metada space that must be | |
5612 | * reserved/free'd for the given bytes. | |
5613 | * @inode: the inode we're manipulating | |
5614 | * @num_bytes: the number of bytes in question | |
5615 | * @reserve: 1 if we are reserving space, 0 if we are freeing space | |
5616 | * | |
5617 | * This adjusts the number of csum_bytes in the inode and then returns the | |
5618 | * correct amount of metadata that must either be reserved or freed. We | |
5619 | * calculate how many checksums we can fit into one leaf and then divide the | |
5620 | * number of bytes that will need to be checksumed by this value to figure out | |
5621 | * how many checksums will be required. If we are adding bytes then the number | |
5622 | * may go up and we will return the number of additional bytes that must be | |
5623 | * reserved. If it is going down we will return the number of bytes that must | |
5624 | * be freed. | |
5625 | * | |
5626 | * This must be called with BTRFS_I(inode)->lock held. | |
5627 | */ | |
5628 | static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes, | |
5629 | int reserve) | |
5630 | { | |
5631 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5632 | u64 old_csums, num_csums; | |
5633 | ||
5634 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM && | |
5635 | BTRFS_I(inode)->csum_bytes == 0) | |
5636 | return 0; | |
5637 | ||
5638 | old_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes); | |
5639 | if (reserve) | |
5640 | BTRFS_I(inode)->csum_bytes += num_bytes; | |
5641 | else | |
5642 | BTRFS_I(inode)->csum_bytes -= num_bytes; | |
5643 | num_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes); | |
5644 | ||
5645 | /* No change, no need to reserve more */ | |
5646 | if (old_csums == num_csums) | |
5647 | return 0; | |
5648 | ||
5649 | if (reserve) | |
5650 | return btrfs_calc_trans_metadata_size(root, | |
5651 | num_csums - old_csums); | |
5652 | ||
5653 | return btrfs_calc_trans_metadata_size(root, old_csums - num_csums); | |
5654 | } | |
5655 | ||
5656 | int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes) | |
5657 | { | |
5658 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5659 | struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv; | |
5660 | u64 to_reserve = 0; | |
5661 | u64 csum_bytes; | |
5662 | unsigned nr_extents = 0; | |
5663 | int extra_reserve = 0; | |
5664 | enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL; | |
5665 | int ret = 0; | |
5666 | bool delalloc_lock = true; | |
5667 | u64 to_free = 0; | |
5668 | unsigned dropped; | |
5669 | ||
5670 | /* If we are a free space inode we need to not flush since we will be in | |
5671 | * the middle of a transaction commit. We also don't need the delalloc | |
5672 | * mutex since we won't race with anybody. We need this mostly to make | |
5673 | * lockdep shut its filthy mouth. | |
5674 | */ | |
5675 | if (btrfs_is_free_space_inode(inode)) { | |
5676 | flush = BTRFS_RESERVE_NO_FLUSH; | |
5677 | delalloc_lock = false; | |
5678 | } | |
5679 | ||
5680 | if (flush != BTRFS_RESERVE_NO_FLUSH && | |
5681 | btrfs_transaction_in_commit(root->fs_info)) | |
5682 | schedule_timeout(1); | |
5683 | ||
5684 | if (delalloc_lock) | |
5685 | mutex_lock(&BTRFS_I(inode)->delalloc_mutex); | |
5686 | ||
5687 | num_bytes = ALIGN(num_bytes, root->sectorsize); | |
5688 | ||
5689 | spin_lock(&BTRFS_I(inode)->lock); | |
5690 | nr_extents = (unsigned)div64_u64(num_bytes + | |
5691 | BTRFS_MAX_EXTENT_SIZE - 1, | |
5692 | BTRFS_MAX_EXTENT_SIZE); | |
5693 | BTRFS_I(inode)->outstanding_extents += nr_extents; | |
5694 | nr_extents = 0; | |
5695 | ||
5696 | if (BTRFS_I(inode)->outstanding_extents > | |
5697 | BTRFS_I(inode)->reserved_extents) | |
5698 | nr_extents = BTRFS_I(inode)->outstanding_extents - | |
5699 | BTRFS_I(inode)->reserved_extents; | |
5700 | ||
5701 | /* | |
5702 | * Add an item to reserve for updating the inode when we complete the | |
5703 | * delalloc io. | |
5704 | */ | |
5705 | if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED, | |
5706 | &BTRFS_I(inode)->runtime_flags)) { | |
5707 | nr_extents++; | |
5708 | extra_reserve = 1; | |
5709 | } | |
5710 | ||
5711 | to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents); | |
5712 | to_reserve += calc_csum_metadata_size(inode, num_bytes, 1); | |
5713 | csum_bytes = BTRFS_I(inode)->csum_bytes; | |
5714 | spin_unlock(&BTRFS_I(inode)->lock); | |
5715 | ||
5716 | if (root->fs_info->quota_enabled) { | |
5717 | ret = btrfs_qgroup_reserve_meta(root, | |
5718 | nr_extents * root->nodesize); | |
5719 | if (ret) | |
5720 | goto out_fail; | |
5721 | } | |
5722 | ||
5723 | ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush); | |
5724 | if (unlikely(ret)) { | |
5725 | btrfs_qgroup_free_meta(root, nr_extents * root->nodesize); | |
5726 | goto out_fail; | |
5727 | } | |
5728 | ||
5729 | spin_lock(&BTRFS_I(inode)->lock); | |
5730 | if (extra_reserve) { | |
5731 | set_bit(BTRFS_INODE_DELALLOC_META_RESERVED, | |
5732 | &BTRFS_I(inode)->runtime_flags); | |
5733 | nr_extents--; | |
5734 | } | |
5735 | BTRFS_I(inode)->reserved_extents += nr_extents; | |
5736 | spin_unlock(&BTRFS_I(inode)->lock); | |
5737 | ||
5738 | if (delalloc_lock) | |
5739 | mutex_unlock(&BTRFS_I(inode)->delalloc_mutex); | |
5740 | ||
5741 | if (to_reserve) | |
5742 | trace_btrfs_space_reservation(root->fs_info, "delalloc", | |
5743 | btrfs_ino(inode), to_reserve, 1); | |
5744 | block_rsv_add_bytes(block_rsv, to_reserve, 1); | |
5745 | ||
5746 | return 0; | |
5747 | ||
5748 | out_fail: | |
5749 | spin_lock(&BTRFS_I(inode)->lock); | |
5750 | dropped = drop_outstanding_extent(inode, num_bytes); | |
5751 | /* | |
5752 | * If the inodes csum_bytes is the same as the original | |
5753 | * csum_bytes then we know we haven't raced with any free()ers | |
5754 | * so we can just reduce our inodes csum bytes and carry on. | |
5755 | */ | |
5756 | if (BTRFS_I(inode)->csum_bytes == csum_bytes) { | |
5757 | calc_csum_metadata_size(inode, num_bytes, 0); | |
5758 | } else { | |
5759 | u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes; | |
5760 | u64 bytes; | |
5761 | ||
5762 | /* | |
5763 | * This is tricky, but first we need to figure out how much we | |
5764 | * free'd from any free-ers that occurred during this | |
5765 | * reservation, so we reset ->csum_bytes to the csum_bytes | |
5766 | * before we dropped our lock, and then call the free for the | |
5767 | * number of bytes that were freed while we were trying our | |
5768 | * reservation. | |
5769 | */ | |
5770 | bytes = csum_bytes - BTRFS_I(inode)->csum_bytes; | |
5771 | BTRFS_I(inode)->csum_bytes = csum_bytes; | |
5772 | to_free = calc_csum_metadata_size(inode, bytes, 0); | |
5773 | ||
5774 | ||
5775 | /* | |
5776 | * Now we need to see how much we would have freed had we not | |
5777 | * been making this reservation and our ->csum_bytes were not | |
5778 | * artificially inflated. | |
5779 | */ | |
5780 | BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes; | |
5781 | bytes = csum_bytes - orig_csum_bytes; | |
5782 | bytes = calc_csum_metadata_size(inode, bytes, 0); | |
5783 | ||
5784 | /* | |
5785 | * Now reset ->csum_bytes to what it should be. If bytes is | |
5786 | * more than to_free then we would have free'd more space had we | |
5787 | * not had an artificially high ->csum_bytes, so we need to free | |
5788 | * the remainder. If bytes is the same or less then we don't | |
5789 | * need to do anything, the other free-ers did the correct | |
5790 | * thing. | |
5791 | */ | |
5792 | BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes; | |
5793 | if (bytes > to_free) | |
5794 | to_free = bytes - to_free; | |
5795 | else | |
5796 | to_free = 0; | |
5797 | } | |
5798 | spin_unlock(&BTRFS_I(inode)->lock); | |
5799 | if (dropped) | |
5800 | to_free += btrfs_calc_trans_metadata_size(root, dropped); | |
5801 | ||
5802 | if (to_free) { | |
5803 | btrfs_block_rsv_release(root, block_rsv, to_free); | |
5804 | trace_btrfs_space_reservation(root->fs_info, "delalloc", | |
5805 | btrfs_ino(inode), to_free, 0); | |
5806 | } | |
5807 | if (delalloc_lock) | |
5808 | mutex_unlock(&BTRFS_I(inode)->delalloc_mutex); | |
5809 | return ret; | |
5810 | } | |
5811 | ||
5812 | /** | |
5813 | * btrfs_delalloc_release_metadata - release a metadata reservation for an inode | |
5814 | * @inode: the inode to release the reservation for | |
5815 | * @num_bytes: the number of bytes we're releasing | |
5816 | * | |
5817 | * This will release the metadata reservation for an inode. This can be called | |
5818 | * once we complete IO for a given set of bytes to release their metadata | |
5819 | * reservations. | |
5820 | */ | |
5821 | void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes) | |
5822 | { | |
5823 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5824 | u64 to_free = 0; | |
5825 | unsigned dropped; | |
5826 | ||
5827 | num_bytes = ALIGN(num_bytes, root->sectorsize); | |
5828 | spin_lock(&BTRFS_I(inode)->lock); | |
5829 | dropped = drop_outstanding_extent(inode, num_bytes); | |
5830 | ||
5831 | if (num_bytes) | |
5832 | to_free = calc_csum_metadata_size(inode, num_bytes, 0); | |
5833 | spin_unlock(&BTRFS_I(inode)->lock); | |
5834 | if (dropped > 0) | |
5835 | to_free += btrfs_calc_trans_metadata_size(root, dropped); | |
5836 | ||
5837 | if (btrfs_test_is_dummy_root(root)) | |
5838 | return; | |
5839 | ||
5840 | trace_btrfs_space_reservation(root->fs_info, "delalloc", | |
5841 | btrfs_ino(inode), to_free, 0); | |
5842 | ||
5843 | btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv, | |
5844 | to_free); | |
5845 | } | |
5846 | ||
5847 | /** | |
5848 | * btrfs_delalloc_reserve_space - reserve data and metadata space for | |
5849 | * delalloc | |
5850 | * @inode: inode we're writing to | |
5851 | * @start: start range we are writing to | |
5852 | * @len: how long the range we are writing to | |
5853 | * | |
5854 | * TODO: This function will finally replace old btrfs_delalloc_reserve_space() | |
5855 | * | |
5856 | * This will do the following things | |
5857 | * | |
5858 | * o reserve space in data space info for num bytes | |
5859 | * and reserve precious corresponding qgroup space | |
5860 | * (Done in check_data_free_space) | |
5861 | * | |
5862 | * o reserve space for metadata space, based on the number of outstanding | |
5863 | * extents and how much csums will be needed | |
5864 | * also reserve metadata space in a per root over-reserve method. | |
5865 | * o add to the inodes->delalloc_bytes | |
5866 | * o add it to the fs_info's delalloc inodes list. | |
5867 | * (Above 3 all done in delalloc_reserve_metadata) | |
5868 | * | |
5869 | * Return 0 for success | |
5870 | * Return <0 for error(-ENOSPC or -EQUOT) | |
5871 | */ | |
5872 | int btrfs_delalloc_reserve_space(struct inode *inode, u64 start, u64 len) | |
5873 | { | |
5874 | int ret; | |
5875 | ||
5876 | ret = btrfs_check_data_free_space(inode, start, len); | |
5877 | if (ret < 0) | |
5878 | return ret; | |
5879 | ret = btrfs_delalloc_reserve_metadata(inode, len); | |
5880 | if (ret < 0) | |
5881 | btrfs_free_reserved_data_space(inode, start, len); | |
5882 | return ret; | |
5883 | } | |
5884 | ||
5885 | /** | |
5886 | * btrfs_delalloc_release_space - release data and metadata space for delalloc | |
5887 | * @inode: inode we're releasing space for | |
5888 | * @start: start position of the space already reserved | |
5889 | * @len: the len of the space already reserved | |
5890 | * | |
5891 | * This must be matched with a call to btrfs_delalloc_reserve_space. This is | |
5892 | * called in the case that we don't need the metadata AND data reservations | |
5893 | * anymore. So if there is an error or we insert an inline extent. | |
5894 | * | |
5895 | * This function will release the metadata space that was not used and will | |
5896 | * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes | |
5897 | * list if there are no delalloc bytes left. | |
5898 | * Also it will handle the qgroup reserved space. | |
5899 | */ | |
5900 | void btrfs_delalloc_release_space(struct inode *inode, u64 start, u64 len) | |
5901 | { | |
5902 | btrfs_delalloc_release_metadata(inode, len); | |
5903 | btrfs_free_reserved_data_space(inode, start, len); | |
5904 | } | |
5905 | ||
5906 | static int update_block_group(struct btrfs_trans_handle *trans, | |
5907 | struct btrfs_root *root, u64 bytenr, | |
5908 | u64 num_bytes, int alloc) | |
5909 | { | |
5910 | struct btrfs_block_group_cache *cache = NULL; | |
5911 | struct btrfs_fs_info *info = root->fs_info; | |
5912 | u64 total = num_bytes; | |
5913 | u64 old_val; | |
5914 | u64 byte_in_group; | |
5915 | int factor; | |
5916 | ||
5917 | /* block accounting for super block */ | |
5918 | spin_lock(&info->delalloc_root_lock); | |
5919 | old_val = btrfs_super_bytes_used(info->super_copy); | |
5920 | if (alloc) | |
5921 | old_val += num_bytes; | |
5922 | else | |
5923 | old_val -= num_bytes; | |
5924 | btrfs_set_super_bytes_used(info->super_copy, old_val); | |
5925 | spin_unlock(&info->delalloc_root_lock); | |
5926 | ||
5927 | while (total) { | |
5928 | cache = btrfs_lookup_block_group(info, bytenr); | |
5929 | if (!cache) | |
5930 | return -ENOENT; | |
5931 | if (cache->flags & (BTRFS_BLOCK_GROUP_DUP | | |
5932 | BTRFS_BLOCK_GROUP_RAID1 | | |
5933 | BTRFS_BLOCK_GROUP_RAID10)) | |
5934 | factor = 2; | |
5935 | else | |
5936 | factor = 1; | |
5937 | /* | |
5938 | * If this block group has free space cache written out, we | |
5939 | * need to make sure to load it if we are removing space. This | |
5940 | * is because we need the unpinning stage to actually add the | |
5941 | * space back to the block group, otherwise we will leak space. | |
5942 | */ | |
5943 | if (!alloc && cache->cached == BTRFS_CACHE_NO) | |
5944 | cache_block_group(cache, 1); | |
5945 | ||
5946 | byte_in_group = bytenr - cache->key.objectid; | |
5947 | WARN_ON(byte_in_group > cache->key.offset); | |
5948 | ||
5949 | spin_lock(&cache->space_info->lock); | |
5950 | spin_lock(&cache->lock); | |
5951 | ||
5952 | if (btrfs_test_opt(root, SPACE_CACHE) && | |
5953 | cache->disk_cache_state < BTRFS_DC_CLEAR) | |
5954 | cache->disk_cache_state = BTRFS_DC_CLEAR; | |
5955 | ||
5956 | old_val = btrfs_block_group_used(&cache->item); | |
5957 | num_bytes = min(total, cache->key.offset - byte_in_group); | |
5958 | if (alloc) { | |
5959 | old_val += num_bytes; | |
5960 | btrfs_set_block_group_used(&cache->item, old_val); | |
5961 | cache->reserved -= num_bytes; | |
5962 | cache->space_info->bytes_reserved -= num_bytes; | |
5963 | cache->space_info->bytes_used += num_bytes; | |
5964 | cache->space_info->disk_used += num_bytes * factor; | |
5965 | spin_unlock(&cache->lock); | |
5966 | spin_unlock(&cache->space_info->lock); | |
5967 | } else { | |
5968 | old_val -= num_bytes; | |
5969 | btrfs_set_block_group_used(&cache->item, old_val); | |
5970 | cache->pinned += num_bytes; | |
5971 | cache->space_info->bytes_pinned += num_bytes; | |
5972 | cache->space_info->bytes_used -= num_bytes; | |
5973 | cache->space_info->disk_used -= num_bytes * factor; | |
5974 | spin_unlock(&cache->lock); | |
5975 | spin_unlock(&cache->space_info->lock); | |
5976 | ||
5977 | set_extent_dirty(info->pinned_extents, | |
5978 | bytenr, bytenr + num_bytes - 1, | |
5979 | GFP_NOFS | __GFP_NOFAIL); | |
5980 | } | |
5981 | ||
5982 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
5983 | if (list_empty(&cache->dirty_list)) { | |
5984 | list_add_tail(&cache->dirty_list, | |
5985 | &trans->transaction->dirty_bgs); | |
5986 | trans->transaction->num_dirty_bgs++; | |
5987 | btrfs_get_block_group(cache); | |
5988 | } | |
5989 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
5990 | ||
5991 | /* | |
5992 | * No longer have used bytes in this block group, queue it for | |
5993 | * deletion. We do this after adding the block group to the | |
5994 | * dirty list to avoid races between cleaner kthread and space | |
5995 | * cache writeout. | |
5996 | */ | |
5997 | if (!alloc && old_val == 0) { | |
5998 | spin_lock(&info->unused_bgs_lock); | |
5999 | if (list_empty(&cache->bg_list)) { | |
6000 | btrfs_get_block_group(cache); | |
6001 | list_add_tail(&cache->bg_list, | |
6002 | &info->unused_bgs); | |
6003 | } | |
6004 | spin_unlock(&info->unused_bgs_lock); | |
6005 | } | |
6006 | ||
6007 | btrfs_put_block_group(cache); | |
6008 | total -= num_bytes; | |
6009 | bytenr += num_bytes; | |
6010 | } | |
6011 | return 0; | |
6012 | } | |
6013 | ||
6014 | static u64 first_logical_byte(struct btrfs_root *root, u64 search_start) | |
6015 | { | |
6016 | struct btrfs_block_group_cache *cache; | |
6017 | u64 bytenr; | |
6018 | ||
6019 | spin_lock(&root->fs_info->block_group_cache_lock); | |
6020 | bytenr = root->fs_info->first_logical_byte; | |
6021 | spin_unlock(&root->fs_info->block_group_cache_lock); | |
6022 | ||
6023 | if (bytenr < (u64)-1) | |
6024 | return bytenr; | |
6025 | ||
6026 | cache = btrfs_lookup_first_block_group(root->fs_info, search_start); | |
6027 | if (!cache) | |
6028 | return 0; | |
6029 | ||
6030 | bytenr = cache->key.objectid; | |
6031 | btrfs_put_block_group(cache); | |
6032 | ||
6033 | return bytenr; | |
6034 | } | |
6035 | ||
6036 | static int pin_down_extent(struct btrfs_root *root, | |
6037 | struct btrfs_block_group_cache *cache, | |
6038 | u64 bytenr, u64 num_bytes, int reserved) | |
6039 | { | |
6040 | spin_lock(&cache->space_info->lock); | |
6041 | spin_lock(&cache->lock); | |
6042 | cache->pinned += num_bytes; | |
6043 | cache->space_info->bytes_pinned += num_bytes; | |
6044 | if (reserved) { | |
6045 | cache->reserved -= num_bytes; | |
6046 | cache->space_info->bytes_reserved -= num_bytes; | |
6047 | } | |
6048 | spin_unlock(&cache->lock); | |
6049 | spin_unlock(&cache->space_info->lock); | |
6050 | ||
6051 | set_extent_dirty(root->fs_info->pinned_extents, bytenr, | |
6052 | bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL); | |
6053 | if (reserved) | |
6054 | trace_btrfs_reserved_extent_free(root, bytenr, num_bytes); | |
6055 | return 0; | |
6056 | } | |
6057 | ||
6058 | /* | |
6059 | * this function must be called within transaction | |
6060 | */ | |
6061 | int btrfs_pin_extent(struct btrfs_root *root, | |
6062 | u64 bytenr, u64 num_bytes, int reserved) | |
6063 | { | |
6064 | struct btrfs_block_group_cache *cache; | |
6065 | ||
6066 | cache = btrfs_lookup_block_group(root->fs_info, bytenr); | |
6067 | BUG_ON(!cache); /* Logic error */ | |
6068 | ||
6069 | pin_down_extent(root, cache, bytenr, num_bytes, reserved); | |
6070 | ||
6071 | btrfs_put_block_group(cache); | |
6072 | return 0; | |
6073 | } | |
6074 | ||
6075 | /* | |
6076 | * this function must be called within transaction | |
6077 | */ | |
6078 | int btrfs_pin_extent_for_log_replay(struct btrfs_root *root, | |
6079 | u64 bytenr, u64 num_bytes) | |
6080 | { | |
6081 | struct btrfs_block_group_cache *cache; | |
6082 | int ret; | |
6083 | ||
6084 | cache = btrfs_lookup_block_group(root->fs_info, bytenr); | |
6085 | if (!cache) | |
6086 | return -EINVAL; | |
6087 | ||
6088 | /* | |
6089 | * pull in the free space cache (if any) so that our pin | |
6090 | * removes the free space from the cache. We have load_only set | |
6091 | * to one because the slow code to read in the free extents does check | |
6092 | * the pinned extents. | |
6093 | */ | |
6094 | cache_block_group(cache, 1); | |
6095 | ||
6096 | pin_down_extent(root, cache, bytenr, num_bytes, 0); | |
6097 | ||
6098 | /* remove us from the free space cache (if we're there at all) */ | |
6099 | ret = btrfs_remove_free_space(cache, bytenr, num_bytes); | |
6100 | btrfs_put_block_group(cache); | |
6101 | return ret; | |
6102 | } | |
6103 | ||
6104 | static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes) | |
6105 | { | |
6106 | int ret; | |
6107 | struct btrfs_block_group_cache *block_group; | |
6108 | struct btrfs_caching_control *caching_ctl; | |
6109 | ||
6110 | block_group = btrfs_lookup_block_group(root->fs_info, start); | |
6111 | if (!block_group) | |
6112 | return -EINVAL; | |
6113 | ||
6114 | cache_block_group(block_group, 0); | |
6115 | caching_ctl = get_caching_control(block_group); | |
6116 | ||
6117 | if (!caching_ctl) { | |
6118 | /* Logic error */ | |
6119 | BUG_ON(!block_group_cache_done(block_group)); | |
6120 | ret = btrfs_remove_free_space(block_group, start, num_bytes); | |
6121 | } else { | |
6122 | mutex_lock(&caching_ctl->mutex); | |
6123 | ||
6124 | if (start >= caching_ctl->progress) { | |
6125 | ret = add_excluded_extent(root, start, num_bytes); | |
6126 | } else if (start + num_bytes <= caching_ctl->progress) { | |
6127 | ret = btrfs_remove_free_space(block_group, | |
6128 | start, num_bytes); | |
6129 | } else { | |
6130 | num_bytes = caching_ctl->progress - start; | |
6131 | ret = btrfs_remove_free_space(block_group, | |
6132 | start, num_bytes); | |
6133 | if (ret) | |
6134 | goto out_lock; | |
6135 | ||
6136 | num_bytes = (start + num_bytes) - | |
6137 | caching_ctl->progress; | |
6138 | start = caching_ctl->progress; | |
6139 | ret = add_excluded_extent(root, start, num_bytes); | |
6140 | } | |
6141 | out_lock: | |
6142 | mutex_unlock(&caching_ctl->mutex); | |
6143 | put_caching_control(caching_ctl); | |
6144 | } | |
6145 | btrfs_put_block_group(block_group); | |
6146 | return ret; | |
6147 | } | |
6148 | ||
6149 | int btrfs_exclude_logged_extents(struct btrfs_root *log, | |
6150 | struct extent_buffer *eb) | |
6151 | { | |
6152 | struct btrfs_file_extent_item *item; | |
6153 | struct btrfs_key key; | |
6154 | int found_type; | |
6155 | int i; | |
6156 | ||
6157 | if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS)) | |
6158 | return 0; | |
6159 | ||
6160 | for (i = 0; i < btrfs_header_nritems(eb); i++) { | |
6161 | btrfs_item_key_to_cpu(eb, &key, i); | |
6162 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
6163 | continue; | |
6164 | item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item); | |
6165 | found_type = btrfs_file_extent_type(eb, item); | |
6166 | if (found_type == BTRFS_FILE_EXTENT_INLINE) | |
6167 | continue; | |
6168 | if (btrfs_file_extent_disk_bytenr(eb, item) == 0) | |
6169 | continue; | |
6170 | key.objectid = btrfs_file_extent_disk_bytenr(eb, item); | |
6171 | key.offset = btrfs_file_extent_disk_num_bytes(eb, item); | |
6172 | __exclude_logged_extent(log, key.objectid, key.offset); | |
6173 | } | |
6174 | ||
6175 | return 0; | |
6176 | } | |
6177 | ||
6178 | /** | |
6179 | * btrfs_update_reserved_bytes - update the block_group and space info counters | |
6180 | * @cache: The cache we are manipulating | |
6181 | * @num_bytes: The number of bytes in question | |
6182 | * @reserve: One of the reservation enums | |
6183 | * @delalloc: The blocks are allocated for the delalloc write | |
6184 | * | |
6185 | * This is called by the allocator when it reserves space, or by somebody who is | |
6186 | * freeing space that was never actually used on disk. For example if you | |
6187 | * reserve some space for a new leaf in transaction A and before transaction A | |
6188 | * commits you free that leaf, you call this with reserve set to 0 in order to | |
6189 | * clear the reservation. | |
6190 | * | |
6191 | * Metadata reservations should be called with RESERVE_ALLOC so we do the proper | |
6192 | * ENOSPC accounting. For data we handle the reservation through clearing the | |
6193 | * delalloc bits in the io_tree. We have to do this since we could end up | |
6194 | * allocating less disk space for the amount of data we have reserved in the | |
6195 | * case of compression. | |
6196 | * | |
6197 | * If this is a reservation and the block group has become read only we cannot | |
6198 | * make the reservation and return -EAGAIN, otherwise this function always | |
6199 | * succeeds. | |
6200 | */ | |
6201 | static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache, | |
6202 | u64 num_bytes, int reserve, int delalloc) | |
6203 | { | |
6204 | struct btrfs_space_info *space_info = cache->space_info; | |
6205 | int ret = 0; | |
6206 | ||
6207 | spin_lock(&space_info->lock); | |
6208 | spin_lock(&cache->lock); | |
6209 | if (reserve != RESERVE_FREE) { | |
6210 | if (cache->ro) { | |
6211 | ret = -EAGAIN; | |
6212 | } else { | |
6213 | cache->reserved += num_bytes; | |
6214 | space_info->bytes_reserved += num_bytes; | |
6215 | if (reserve == RESERVE_ALLOC) { | |
6216 | trace_btrfs_space_reservation(cache->fs_info, | |
6217 | "space_info", space_info->flags, | |
6218 | num_bytes, 0); | |
6219 | space_info->bytes_may_use -= num_bytes; | |
6220 | } | |
6221 | ||
6222 | if (delalloc) | |
6223 | cache->delalloc_bytes += num_bytes; | |
6224 | } | |
6225 | } else { | |
6226 | if (cache->ro) | |
6227 | space_info->bytes_readonly += num_bytes; | |
6228 | cache->reserved -= num_bytes; | |
6229 | space_info->bytes_reserved -= num_bytes; | |
6230 | ||
6231 | if (delalloc) | |
6232 | cache->delalloc_bytes -= num_bytes; | |
6233 | } | |
6234 | spin_unlock(&cache->lock); | |
6235 | spin_unlock(&space_info->lock); | |
6236 | return ret; | |
6237 | } | |
6238 | ||
6239 | void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans, | |
6240 | struct btrfs_root *root) | |
6241 | { | |
6242 | struct btrfs_fs_info *fs_info = root->fs_info; | |
6243 | struct btrfs_caching_control *next; | |
6244 | struct btrfs_caching_control *caching_ctl; | |
6245 | struct btrfs_block_group_cache *cache; | |
6246 | ||
6247 | down_write(&fs_info->commit_root_sem); | |
6248 | ||
6249 | list_for_each_entry_safe(caching_ctl, next, | |
6250 | &fs_info->caching_block_groups, list) { | |
6251 | cache = caching_ctl->block_group; | |
6252 | if (block_group_cache_done(cache)) { | |
6253 | cache->last_byte_to_unpin = (u64)-1; | |
6254 | list_del_init(&caching_ctl->list); | |
6255 | put_caching_control(caching_ctl); | |
6256 | } else { | |
6257 | cache->last_byte_to_unpin = caching_ctl->progress; | |
6258 | } | |
6259 | } | |
6260 | ||
6261 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) | |
6262 | fs_info->pinned_extents = &fs_info->freed_extents[1]; | |
6263 | else | |
6264 | fs_info->pinned_extents = &fs_info->freed_extents[0]; | |
6265 | ||
6266 | up_write(&fs_info->commit_root_sem); | |
6267 | ||
6268 | update_global_block_rsv(fs_info); | |
6269 | } | |
6270 | ||
6271 | /* | |
6272 | * Returns the free cluster for the given space info and sets empty_cluster to | |
6273 | * what it should be based on the mount options. | |
6274 | */ | |
6275 | static struct btrfs_free_cluster * | |
6276 | fetch_cluster_info(struct btrfs_root *root, struct btrfs_space_info *space_info, | |
6277 | u64 *empty_cluster) | |
6278 | { | |
6279 | struct btrfs_free_cluster *ret = NULL; | |
6280 | bool ssd = btrfs_test_opt(root, SSD); | |
6281 | ||
6282 | *empty_cluster = 0; | |
6283 | if (btrfs_mixed_space_info(space_info)) | |
6284 | return ret; | |
6285 | ||
6286 | if (ssd) | |
6287 | *empty_cluster = SZ_2M; | |
6288 | if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) { | |
6289 | ret = &root->fs_info->meta_alloc_cluster; | |
6290 | if (!ssd) | |
6291 | *empty_cluster = SZ_64K; | |
6292 | } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) && ssd) { | |
6293 | ret = &root->fs_info->data_alloc_cluster; | |
6294 | } | |
6295 | ||
6296 | return ret; | |
6297 | } | |
6298 | ||
6299 | static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end, | |
6300 | const bool return_free_space) | |
6301 | { | |
6302 | struct btrfs_fs_info *fs_info = root->fs_info; | |
6303 | struct btrfs_block_group_cache *cache = NULL; | |
6304 | struct btrfs_space_info *space_info; | |
6305 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
6306 | struct btrfs_free_cluster *cluster = NULL; | |
6307 | u64 len; | |
6308 | u64 total_unpinned = 0; | |
6309 | u64 empty_cluster = 0; | |
6310 | bool readonly; | |
6311 | ||
6312 | while (start <= end) { | |
6313 | readonly = false; | |
6314 | if (!cache || | |
6315 | start >= cache->key.objectid + cache->key.offset) { | |
6316 | if (cache) | |
6317 | btrfs_put_block_group(cache); | |
6318 | total_unpinned = 0; | |
6319 | cache = btrfs_lookup_block_group(fs_info, start); | |
6320 | BUG_ON(!cache); /* Logic error */ | |
6321 | ||
6322 | cluster = fetch_cluster_info(root, | |
6323 | cache->space_info, | |
6324 | &empty_cluster); | |
6325 | empty_cluster <<= 1; | |
6326 | } | |
6327 | ||
6328 | len = cache->key.objectid + cache->key.offset - start; | |
6329 | len = min(len, end + 1 - start); | |
6330 | ||
6331 | if (start < cache->last_byte_to_unpin) { | |
6332 | len = min(len, cache->last_byte_to_unpin - start); | |
6333 | if (return_free_space) | |
6334 | btrfs_add_free_space(cache, start, len); | |
6335 | } | |
6336 | ||
6337 | start += len; | |
6338 | total_unpinned += len; | |
6339 | space_info = cache->space_info; | |
6340 | ||
6341 | /* | |
6342 | * If this space cluster has been marked as fragmented and we've | |
6343 | * unpinned enough in this block group to potentially allow a | |
6344 | * cluster to be created inside of it go ahead and clear the | |
6345 | * fragmented check. | |
6346 | */ | |
6347 | if (cluster && cluster->fragmented && | |
6348 | total_unpinned > empty_cluster) { | |
6349 | spin_lock(&cluster->lock); | |
6350 | cluster->fragmented = 0; | |
6351 | spin_unlock(&cluster->lock); | |
6352 | } | |
6353 | ||
6354 | spin_lock(&space_info->lock); | |
6355 | spin_lock(&cache->lock); | |
6356 | cache->pinned -= len; | |
6357 | space_info->bytes_pinned -= len; | |
6358 | space_info->max_extent_size = 0; | |
6359 | percpu_counter_add(&space_info->total_bytes_pinned, -len); | |
6360 | if (cache->ro) { | |
6361 | space_info->bytes_readonly += len; | |
6362 | readonly = true; | |
6363 | } | |
6364 | spin_unlock(&cache->lock); | |
6365 | if (!readonly && global_rsv->space_info == space_info) { | |
6366 | spin_lock(&global_rsv->lock); | |
6367 | if (!global_rsv->full) { | |
6368 | len = min(len, global_rsv->size - | |
6369 | global_rsv->reserved); | |
6370 | global_rsv->reserved += len; | |
6371 | space_info->bytes_may_use += len; | |
6372 | if (global_rsv->reserved >= global_rsv->size) | |
6373 | global_rsv->full = 1; | |
6374 | } | |
6375 | spin_unlock(&global_rsv->lock); | |
6376 | } | |
6377 | spin_unlock(&space_info->lock); | |
6378 | } | |
6379 | ||
6380 | if (cache) | |
6381 | btrfs_put_block_group(cache); | |
6382 | return 0; | |
6383 | } | |
6384 | ||
6385 | int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans, | |
6386 | struct btrfs_root *root) | |
6387 | { | |
6388 | struct btrfs_fs_info *fs_info = root->fs_info; | |
6389 | struct btrfs_block_group_cache *block_group, *tmp; | |
6390 | struct list_head *deleted_bgs; | |
6391 | struct extent_io_tree *unpin; | |
6392 | u64 start; | |
6393 | u64 end; | |
6394 | int ret; | |
6395 | ||
6396 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) | |
6397 | unpin = &fs_info->freed_extents[1]; | |
6398 | else | |
6399 | unpin = &fs_info->freed_extents[0]; | |
6400 | ||
6401 | while (!trans->aborted) { | |
6402 | mutex_lock(&fs_info->unused_bg_unpin_mutex); | |
6403 | ret = find_first_extent_bit(unpin, 0, &start, &end, | |
6404 | EXTENT_DIRTY, NULL); | |
6405 | if (ret) { | |
6406 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); | |
6407 | break; | |
6408 | } | |
6409 | ||
6410 | if (btrfs_test_opt(root, DISCARD)) | |
6411 | ret = btrfs_discard_extent(root, start, | |
6412 | end + 1 - start, NULL); | |
6413 | ||
6414 | clear_extent_dirty(unpin, start, end, GFP_NOFS); | |
6415 | unpin_extent_range(root, start, end, true); | |
6416 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); | |
6417 | cond_resched(); | |
6418 | } | |
6419 | ||
6420 | /* | |
6421 | * Transaction is finished. We don't need the lock anymore. We | |
6422 | * do need to clean up the block groups in case of a transaction | |
6423 | * abort. | |
6424 | */ | |
6425 | deleted_bgs = &trans->transaction->deleted_bgs; | |
6426 | list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) { | |
6427 | u64 trimmed = 0; | |
6428 | ||
6429 | ret = -EROFS; | |
6430 | if (!trans->aborted) | |
6431 | ret = btrfs_discard_extent(root, | |
6432 | block_group->key.objectid, | |
6433 | block_group->key.offset, | |
6434 | &trimmed); | |
6435 | ||
6436 | list_del_init(&block_group->bg_list); | |
6437 | btrfs_put_block_group_trimming(block_group); | |
6438 | btrfs_put_block_group(block_group); | |
6439 | ||
6440 | if (ret) { | |
6441 | const char *errstr = btrfs_decode_error(ret); | |
6442 | btrfs_warn(fs_info, | |
6443 | "Discard failed while removing blockgroup: errno=%d %s\n", | |
6444 | ret, errstr); | |
6445 | } | |
6446 | } | |
6447 | ||
6448 | return 0; | |
6449 | } | |
6450 | ||
6451 | static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes, | |
6452 | u64 owner, u64 root_objectid) | |
6453 | { | |
6454 | struct btrfs_space_info *space_info; | |
6455 | u64 flags; | |
6456 | ||
6457 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
6458 | if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID) | |
6459 | flags = BTRFS_BLOCK_GROUP_SYSTEM; | |
6460 | else | |
6461 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
6462 | } else { | |
6463 | flags = BTRFS_BLOCK_GROUP_DATA; | |
6464 | } | |
6465 | ||
6466 | space_info = __find_space_info(fs_info, flags); | |
6467 | BUG_ON(!space_info); /* Logic bug */ | |
6468 | percpu_counter_add(&space_info->total_bytes_pinned, num_bytes); | |
6469 | } | |
6470 | ||
6471 | ||
6472 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, | |
6473 | struct btrfs_root *root, | |
6474 | struct btrfs_delayed_ref_node *node, u64 parent, | |
6475 | u64 root_objectid, u64 owner_objectid, | |
6476 | u64 owner_offset, int refs_to_drop, | |
6477 | struct btrfs_delayed_extent_op *extent_op) | |
6478 | { | |
6479 | struct btrfs_key key; | |
6480 | struct btrfs_path *path; | |
6481 | struct btrfs_fs_info *info = root->fs_info; | |
6482 | struct btrfs_root *extent_root = info->extent_root; | |
6483 | struct extent_buffer *leaf; | |
6484 | struct btrfs_extent_item *ei; | |
6485 | struct btrfs_extent_inline_ref *iref; | |
6486 | int ret; | |
6487 | int is_data; | |
6488 | int extent_slot = 0; | |
6489 | int found_extent = 0; | |
6490 | int num_to_del = 1; | |
6491 | u32 item_size; | |
6492 | u64 refs; | |
6493 | u64 bytenr = node->bytenr; | |
6494 | u64 num_bytes = node->num_bytes; | |
6495 | int last_ref = 0; | |
6496 | bool skinny_metadata = btrfs_fs_incompat(root->fs_info, | |
6497 | SKINNY_METADATA); | |
6498 | ||
6499 | path = btrfs_alloc_path(); | |
6500 | if (!path) | |
6501 | return -ENOMEM; | |
6502 | ||
6503 | path->reada = READA_FORWARD; | |
6504 | path->leave_spinning = 1; | |
6505 | ||
6506 | is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID; | |
6507 | BUG_ON(!is_data && refs_to_drop != 1); | |
6508 | ||
6509 | if (is_data) | |
6510 | skinny_metadata = 0; | |
6511 | ||
6512 | ret = lookup_extent_backref(trans, extent_root, path, &iref, | |
6513 | bytenr, num_bytes, parent, | |
6514 | root_objectid, owner_objectid, | |
6515 | owner_offset); | |
6516 | if (ret == 0) { | |
6517 | extent_slot = path->slots[0]; | |
6518 | while (extent_slot >= 0) { | |
6519 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
6520 | extent_slot); | |
6521 | if (key.objectid != bytenr) | |
6522 | break; | |
6523 | if (key.type == BTRFS_EXTENT_ITEM_KEY && | |
6524 | key.offset == num_bytes) { | |
6525 | found_extent = 1; | |
6526 | break; | |
6527 | } | |
6528 | if (key.type == BTRFS_METADATA_ITEM_KEY && | |
6529 | key.offset == owner_objectid) { | |
6530 | found_extent = 1; | |
6531 | break; | |
6532 | } | |
6533 | if (path->slots[0] - extent_slot > 5) | |
6534 | break; | |
6535 | extent_slot--; | |
6536 | } | |
6537 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
6538 | item_size = btrfs_item_size_nr(path->nodes[0], extent_slot); | |
6539 | if (found_extent && item_size < sizeof(*ei)) | |
6540 | found_extent = 0; | |
6541 | #endif | |
6542 | if (!found_extent) { | |
6543 | BUG_ON(iref); | |
6544 | ret = remove_extent_backref(trans, extent_root, path, | |
6545 | NULL, refs_to_drop, | |
6546 | is_data, &last_ref); | |
6547 | if (ret) { | |
6548 | btrfs_abort_transaction(trans, extent_root, ret); | |
6549 | goto out; | |
6550 | } | |
6551 | btrfs_release_path(path); | |
6552 | path->leave_spinning = 1; | |
6553 | ||
6554 | key.objectid = bytenr; | |
6555 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
6556 | key.offset = num_bytes; | |
6557 | ||
6558 | if (!is_data && skinny_metadata) { | |
6559 | key.type = BTRFS_METADATA_ITEM_KEY; | |
6560 | key.offset = owner_objectid; | |
6561 | } | |
6562 | ||
6563 | ret = btrfs_search_slot(trans, extent_root, | |
6564 | &key, path, -1, 1); | |
6565 | if (ret > 0 && skinny_metadata && path->slots[0]) { | |
6566 | /* | |
6567 | * Couldn't find our skinny metadata item, | |
6568 | * see if we have ye olde extent item. | |
6569 | */ | |
6570 | path->slots[0]--; | |
6571 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
6572 | path->slots[0]); | |
6573 | if (key.objectid == bytenr && | |
6574 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
6575 | key.offset == num_bytes) | |
6576 | ret = 0; | |
6577 | } | |
6578 | ||
6579 | if (ret > 0 && skinny_metadata) { | |
6580 | skinny_metadata = false; | |
6581 | key.objectid = bytenr; | |
6582 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
6583 | key.offset = num_bytes; | |
6584 | btrfs_release_path(path); | |
6585 | ret = btrfs_search_slot(trans, extent_root, | |
6586 | &key, path, -1, 1); | |
6587 | } | |
6588 | ||
6589 | if (ret) { | |
6590 | btrfs_err(info, "umm, got %d back from search, was looking for %llu", | |
6591 | ret, bytenr); | |
6592 | if (ret > 0) | |
6593 | btrfs_print_leaf(extent_root, | |
6594 | path->nodes[0]); | |
6595 | } | |
6596 | if (ret < 0) { | |
6597 | btrfs_abort_transaction(trans, extent_root, ret); | |
6598 | goto out; | |
6599 | } | |
6600 | extent_slot = path->slots[0]; | |
6601 | } | |
6602 | } else if (WARN_ON(ret == -ENOENT)) { | |
6603 | btrfs_print_leaf(extent_root, path->nodes[0]); | |
6604 | btrfs_err(info, | |
6605 | "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu", | |
6606 | bytenr, parent, root_objectid, owner_objectid, | |
6607 | owner_offset); | |
6608 | btrfs_abort_transaction(trans, extent_root, ret); | |
6609 | goto out; | |
6610 | } else { | |
6611 | btrfs_abort_transaction(trans, extent_root, ret); | |
6612 | goto out; | |
6613 | } | |
6614 | ||
6615 | leaf = path->nodes[0]; | |
6616 | item_size = btrfs_item_size_nr(leaf, extent_slot); | |
6617 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
6618 | if (item_size < sizeof(*ei)) { | |
6619 | BUG_ON(found_extent || extent_slot != path->slots[0]); | |
6620 | ret = convert_extent_item_v0(trans, extent_root, path, | |
6621 | owner_objectid, 0); | |
6622 | if (ret < 0) { | |
6623 | btrfs_abort_transaction(trans, extent_root, ret); | |
6624 | goto out; | |
6625 | } | |
6626 | ||
6627 | btrfs_release_path(path); | |
6628 | path->leave_spinning = 1; | |
6629 | ||
6630 | key.objectid = bytenr; | |
6631 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
6632 | key.offset = num_bytes; | |
6633 | ||
6634 | ret = btrfs_search_slot(trans, extent_root, &key, path, | |
6635 | -1, 1); | |
6636 | if (ret) { | |
6637 | btrfs_err(info, "umm, got %d back from search, was looking for %llu", | |
6638 | ret, bytenr); | |
6639 | btrfs_print_leaf(extent_root, path->nodes[0]); | |
6640 | } | |
6641 | if (ret < 0) { | |
6642 | btrfs_abort_transaction(trans, extent_root, ret); | |
6643 | goto out; | |
6644 | } | |
6645 | ||
6646 | extent_slot = path->slots[0]; | |
6647 | leaf = path->nodes[0]; | |
6648 | item_size = btrfs_item_size_nr(leaf, extent_slot); | |
6649 | } | |
6650 | #endif | |
6651 | BUG_ON(item_size < sizeof(*ei)); | |
6652 | ei = btrfs_item_ptr(leaf, extent_slot, | |
6653 | struct btrfs_extent_item); | |
6654 | if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID && | |
6655 | key.type == BTRFS_EXTENT_ITEM_KEY) { | |
6656 | struct btrfs_tree_block_info *bi; | |
6657 | BUG_ON(item_size < sizeof(*ei) + sizeof(*bi)); | |
6658 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
6659 | WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi)); | |
6660 | } | |
6661 | ||
6662 | refs = btrfs_extent_refs(leaf, ei); | |
6663 | if (refs < refs_to_drop) { | |
6664 | btrfs_err(info, "trying to drop %d refs but we only have %Lu " | |
6665 | "for bytenr %Lu", refs_to_drop, refs, bytenr); | |
6666 | ret = -EINVAL; | |
6667 | btrfs_abort_transaction(trans, extent_root, ret); | |
6668 | goto out; | |
6669 | } | |
6670 | refs -= refs_to_drop; | |
6671 | ||
6672 | if (refs > 0) { | |
6673 | if (extent_op) | |
6674 | __run_delayed_extent_op(extent_op, leaf, ei); | |
6675 | /* | |
6676 | * In the case of inline back ref, reference count will | |
6677 | * be updated by remove_extent_backref | |
6678 | */ | |
6679 | if (iref) { | |
6680 | BUG_ON(!found_extent); | |
6681 | } else { | |
6682 | btrfs_set_extent_refs(leaf, ei, refs); | |
6683 | btrfs_mark_buffer_dirty(leaf); | |
6684 | } | |
6685 | if (found_extent) { | |
6686 | ret = remove_extent_backref(trans, extent_root, path, | |
6687 | iref, refs_to_drop, | |
6688 | is_data, &last_ref); | |
6689 | if (ret) { | |
6690 | btrfs_abort_transaction(trans, extent_root, ret); | |
6691 | goto out; | |
6692 | } | |
6693 | } | |
6694 | add_pinned_bytes(root->fs_info, -num_bytes, owner_objectid, | |
6695 | root_objectid); | |
6696 | } else { | |
6697 | if (found_extent) { | |
6698 | BUG_ON(is_data && refs_to_drop != | |
6699 | extent_data_ref_count(path, iref)); | |
6700 | if (iref) { | |
6701 | BUG_ON(path->slots[0] != extent_slot); | |
6702 | } else { | |
6703 | BUG_ON(path->slots[0] != extent_slot + 1); | |
6704 | path->slots[0] = extent_slot; | |
6705 | num_to_del = 2; | |
6706 | } | |
6707 | } | |
6708 | ||
6709 | last_ref = 1; | |
6710 | ret = btrfs_del_items(trans, extent_root, path, path->slots[0], | |
6711 | num_to_del); | |
6712 | if (ret) { | |
6713 | btrfs_abort_transaction(trans, extent_root, ret); | |
6714 | goto out; | |
6715 | } | |
6716 | btrfs_release_path(path); | |
6717 | ||
6718 | if (is_data) { | |
6719 | ret = btrfs_del_csums(trans, root, bytenr, num_bytes); | |
6720 | if (ret) { | |
6721 | btrfs_abort_transaction(trans, extent_root, ret); | |
6722 | goto out; | |
6723 | } | |
6724 | } | |
6725 | ||
6726 | ret = add_to_free_space_tree(trans, root->fs_info, bytenr, | |
6727 | num_bytes); | |
6728 | if (ret) { | |
6729 | btrfs_abort_transaction(trans, extent_root, ret); | |
6730 | goto out; | |
6731 | } | |
6732 | ||
6733 | ret = update_block_group(trans, root, bytenr, num_bytes, 0); | |
6734 | if (ret) { | |
6735 | btrfs_abort_transaction(trans, extent_root, ret); | |
6736 | goto out; | |
6737 | } | |
6738 | } | |
6739 | btrfs_release_path(path); | |
6740 | ||
6741 | out: | |
6742 | btrfs_free_path(path); | |
6743 | return ret; | |
6744 | } | |
6745 | ||
6746 | /* | |
6747 | * when we free an block, it is possible (and likely) that we free the last | |
6748 | * delayed ref for that extent as well. This searches the delayed ref tree for | |
6749 | * a given extent, and if there are no other delayed refs to be processed, it | |
6750 | * removes it from the tree. | |
6751 | */ | |
6752 | static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans, | |
6753 | struct btrfs_root *root, u64 bytenr) | |
6754 | { | |
6755 | struct btrfs_delayed_ref_head *head; | |
6756 | struct btrfs_delayed_ref_root *delayed_refs; | |
6757 | int ret = 0; | |
6758 | ||
6759 | delayed_refs = &trans->transaction->delayed_refs; | |
6760 | spin_lock(&delayed_refs->lock); | |
6761 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
6762 | if (!head) | |
6763 | goto out_delayed_unlock; | |
6764 | ||
6765 | spin_lock(&head->lock); | |
6766 | if (!list_empty(&head->ref_list)) | |
6767 | goto out; | |
6768 | ||
6769 | if (head->extent_op) { | |
6770 | if (!head->must_insert_reserved) | |
6771 | goto out; | |
6772 | btrfs_free_delayed_extent_op(head->extent_op); | |
6773 | head->extent_op = NULL; | |
6774 | } | |
6775 | ||
6776 | /* | |
6777 | * waiting for the lock here would deadlock. If someone else has it | |
6778 | * locked they are already in the process of dropping it anyway | |
6779 | */ | |
6780 | if (!mutex_trylock(&head->mutex)) | |
6781 | goto out; | |
6782 | ||
6783 | /* | |
6784 | * at this point we have a head with no other entries. Go | |
6785 | * ahead and process it. | |
6786 | */ | |
6787 | head->node.in_tree = 0; | |
6788 | rb_erase(&head->href_node, &delayed_refs->href_root); | |
6789 | ||
6790 | atomic_dec(&delayed_refs->num_entries); | |
6791 | ||
6792 | /* | |
6793 | * we don't take a ref on the node because we're removing it from the | |
6794 | * tree, so we just steal the ref the tree was holding. | |
6795 | */ | |
6796 | delayed_refs->num_heads--; | |
6797 | if (head->processing == 0) | |
6798 | delayed_refs->num_heads_ready--; | |
6799 | head->processing = 0; | |
6800 | spin_unlock(&head->lock); | |
6801 | spin_unlock(&delayed_refs->lock); | |
6802 | ||
6803 | BUG_ON(head->extent_op); | |
6804 | if (head->must_insert_reserved) | |
6805 | ret = 1; | |
6806 | ||
6807 | mutex_unlock(&head->mutex); | |
6808 | btrfs_put_delayed_ref(&head->node); | |
6809 | return ret; | |
6810 | out: | |
6811 | spin_unlock(&head->lock); | |
6812 | ||
6813 | out_delayed_unlock: | |
6814 | spin_unlock(&delayed_refs->lock); | |
6815 | return 0; | |
6816 | } | |
6817 | ||
6818 | void btrfs_free_tree_block(struct btrfs_trans_handle *trans, | |
6819 | struct btrfs_root *root, | |
6820 | struct extent_buffer *buf, | |
6821 | u64 parent, int last_ref) | |
6822 | { | |
6823 | int pin = 1; | |
6824 | int ret; | |
6825 | ||
6826 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { | |
6827 | ret = btrfs_add_delayed_tree_ref(root->fs_info, trans, | |
6828 | buf->start, buf->len, | |
6829 | parent, root->root_key.objectid, | |
6830 | btrfs_header_level(buf), | |
6831 | BTRFS_DROP_DELAYED_REF, NULL); | |
6832 | BUG_ON(ret); /* -ENOMEM */ | |
6833 | } | |
6834 | ||
6835 | if (!last_ref) | |
6836 | return; | |
6837 | ||
6838 | if (btrfs_header_generation(buf) == trans->transid) { | |
6839 | struct btrfs_block_group_cache *cache; | |
6840 | ||
6841 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { | |
6842 | ret = check_ref_cleanup(trans, root, buf->start); | |
6843 | if (!ret) | |
6844 | goto out; | |
6845 | } | |
6846 | ||
6847 | cache = btrfs_lookup_block_group(root->fs_info, buf->start); | |
6848 | ||
6849 | if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) { | |
6850 | pin_down_extent(root, cache, buf->start, buf->len, 1); | |
6851 | btrfs_put_block_group(cache); | |
6852 | goto out; | |
6853 | } | |
6854 | ||
6855 | WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)); | |
6856 | ||
6857 | btrfs_add_free_space(cache, buf->start, buf->len); | |
6858 | btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE, 0); | |
6859 | btrfs_put_block_group(cache); | |
6860 | trace_btrfs_reserved_extent_free(root, buf->start, buf->len); | |
6861 | pin = 0; | |
6862 | } | |
6863 | out: | |
6864 | if (pin) | |
6865 | add_pinned_bytes(root->fs_info, buf->len, | |
6866 | btrfs_header_level(buf), | |
6867 | root->root_key.objectid); | |
6868 | ||
6869 | /* | |
6870 | * Deleting the buffer, clear the corrupt flag since it doesn't matter | |
6871 | * anymore. | |
6872 | */ | |
6873 | clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags); | |
6874 | } | |
6875 | ||
6876 | /* Can return -ENOMEM */ | |
6877 | int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
6878 | u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid, | |
6879 | u64 owner, u64 offset) | |
6880 | { | |
6881 | int ret; | |
6882 | struct btrfs_fs_info *fs_info = root->fs_info; | |
6883 | ||
6884 | if (btrfs_test_is_dummy_root(root)) | |
6885 | return 0; | |
6886 | ||
6887 | add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid); | |
6888 | ||
6889 | /* | |
6890 | * tree log blocks never actually go into the extent allocation | |
6891 | * tree, just update pinning info and exit early. | |
6892 | */ | |
6893 | if (root_objectid == BTRFS_TREE_LOG_OBJECTID) { | |
6894 | WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID); | |
6895 | /* unlocks the pinned mutex */ | |
6896 | btrfs_pin_extent(root, bytenr, num_bytes, 1); | |
6897 | ret = 0; | |
6898 | } else if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
6899 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr, | |
6900 | num_bytes, | |
6901 | parent, root_objectid, (int)owner, | |
6902 | BTRFS_DROP_DELAYED_REF, NULL); | |
6903 | } else { | |
6904 | ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr, | |
6905 | num_bytes, | |
6906 | parent, root_objectid, owner, | |
6907 | offset, 0, | |
6908 | BTRFS_DROP_DELAYED_REF, NULL); | |
6909 | } | |
6910 | return ret; | |
6911 | } | |
6912 | ||
6913 | /* | |
6914 | * when we wait for progress in the block group caching, its because | |
6915 | * our allocation attempt failed at least once. So, we must sleep | |
6916 | * and let some progress happen before we try again. | |
6917 | * | |
6918 | * This function will sleep at least once waiting for new free space to | |
6919 | * show up, and then it will check the block group free space numbers | |
6920 | * for our min num_bytes. Another option is to have it go ahead | |
6921 | * and look in the rbtree for a free extent of a given size, but this | |
6922 | * is a good start. | |
6923 | * | |
6924 | * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using | |
6925 | * any of the information in this block group. | |
6926 | */ | |
6927 | static noinline void | |
6928 | wait_block_group_cache_progress(struct btrfs_block_group_cache *cache, | |
6929 | u64 num_bytes) | |
6930 | { | |
6931 | struct btrfs_caching_control *caching_ctl; | |
6932 | ||
6933 | caching_ctl = get_caching_control(cache); | |
6934 | if (!caching_ctl) | |
6935 | return; | |
6936 | ||
6937 | wait_event(caching_ctl->wait, block_group_cache_done(cache) || | |
6938 | (cache->free_space_ctl->free_space >= num_bytes)); | |
6939 | ||
6940 | put_caching_control(caching_ctl); | |
6941 | } | |
6942 | ||
6943 | static noinline int | |
6944 | wait_block_group_cache_done(struct btrfs_block_group_cache *cache) | |
6945 | { | |
6946 | struct btrfs_caching_control *caching_ctl; | |
6947 | int ret = 0; | |
6948 | ||
6949 | caching_ctl = get_caching_control(cache); | |
6950 | if (!caching_ctl) | |
6951 | return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0; | |
6952 | ||
6953 | wait_event(caching_ctl->wait, block_group_cache_done(cache)); | |
6954 | if (cache->cached == BTRFS_CACHE_ERROR) | |
6955 | ret = -EIO; | |
6956 | put_caching_control(caching_ctl); | |
6957 | return ret; | |
6958 | } | |
6959 | ||
6960 | int __get_raid_index(u64 flags) | |
6961 | { | |
6962 | if (flags & BTRFS_BLOCK_GROUP_RAID10) | |
6963 | return BTRFS_RAID_RAID10; | |
6964 | else if (flags & BTRFS_BLOCK_GROUP_RAID1) | |
6965 | return BTRFS_RAID_RAID1; | |
6966 | else if (flags & BTRFS_BLOCK_GROUP_DUP) | |
6967 | return BTRFS_RAID_DUP; | |
6968 | else if (flags & BTRFS_BLOCK_GROUP_RAID0) | |
6969 | return BTRFS_RAID_RAID0; | |
6970 | else if (flags & BTRFS_BLOCK_GROUP_RAID5) | |
6971 | return BTRFS_RAID_RAID5; | |
6972 | else if (flags & BTRFS_BLOCK_GROUP_RAID6) | |
6973 | return BTRFS_RAID_RAID6; | |
6974 | ||
6975 | return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */ | |
6976 | } | |
6977 | ||
6978 | int get_block_group_index(struct btrfs_block_group_cache *cache) | |
6979 | { | |
6980 | return __get_raid_index(cache->flags); | |
6981 | } | |
6982 | ||
6983 | static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = { | |
6984 | [BTRFS_RAID_RAID10] = "raid10", | |
6985 | [BTRFS_RAID_RAID1] = "raid1", | |
6986 | [BTRFS_RAID_DUP] = "dup", | |
6987 | [BTRFS_RAID_RAID0] = "raid0", | |
6988 | [BTRFS_RAID_SINGLE] = "single", | |
6989 | [BTRFS_RAID_RAID5] = "raid5", | |
6990 | [BTRFS_RAID_RAID6] = "raid6", | |
6991 | }; | |
6992 | ||
6993 | static const char *get_raid_name(enum btrfs_raid_types type) | |
6994 | { | |
6995 | if (type >= BTRFS_NR_RAID_TYPES) | |
6996 | return NULL; | |
6997 | ||
6998 | return btrfs_raid_type_names[type]; | |
6999 | } | |
7000 | ||
7001 | enum btrfs_loop_type { | |
7002 | LOOP_CACHING_NOWAIT = 0, | |
7003 | LOOP_CACHING_WAIT = 1, | |
7004 | LOOP_ALLOC_CHUNK = 2, | |
7005 | LOOP_NO_EMPTY_SIZE = 3, | |
7006 | }; | |
7007 | ||
7008 | static inline void | |
7009 | btrfs_lock_block_group(struct btrfs_block_group_cache *cache, | |
7010 | int delalloc) | |
7011 | { | |
7012 | if (delalloc) | |
7013 | down_read(&cache->data_rwsem); | |
7014 | } | |
7015 | ||
7016 | static inline void | |
7017 | btrfs_grab_block_group(struct btrfs_block_group_cache *cache, | |
7018 | int delalloc) | |
7019 | { | |
7020 | btrfs_get_block_group(cache); | |
7021 | if (delalloc) | |
7022 | down_read(&cache->data_rwsem); | |
7023 | } | |
7024 | ||
7025 | static struct btrfs_block_group_cache * | |
7026 | btrfs_lock_cluster(struct btrfs_block_group_cache *block_group, | |
7027 | struct btrfs_free_cluster *cluster, | |
7028 | int delalloc) | |
7029 | { | |
7030 | struct btrfs_block_group_cache *used_bg = NULL; | |
7031 | bool locked = false; | |
7032 | again: | |
7033 | spin_lock(&cluster->refill_lock); | |
7034 | if (locked) { | |
7035 | if (used_bg == cluster->block_group) | |
7036 | return used_bg; | |
7037 | ||
7038 | up_read(&used_bg->data_rwsem); | |
7039 | btrfs_put_block_group(used_bg); | |
7040 | } | |
7041 | ||
7042 | used_bg = cluster->block_group; | |
7043 | if (!used_bg) | |
7044 | return NULL; | |
7045 | ||
7046 | if (used_bg == block_group) | |
7047 | return used_bg; | |
7048 | ||
7049 | btrfs_get_block_group(used_bg); | |
7050 | ||
7051 | if (!delalloc) | |
7052 | return used_bg; | |
7053 | ||
7054 | if (down_read_trylock(&used_bg->data_rwsem)) | |
7055 | return used_bg; | |
7056 | ||
7057 | spin_unlock(&cluster->refill_lock); | |
7058 | down_read(&used_bg->data_rwsem); | |
7059 | locked = true; | |
7060 | goto again; | |
7061 | } | |
7062 | ||
7063 | static inline void | |
7064 | btrfs_release_block_group(struct btrfs_block_group_cache *cache, | |
7065 | int delalloc) | |
7066 | { | |
7067 | if (delalloc) | |
7068 | up_read(&cache->data_rwsem); | |
7069 | btrfs_put_block_group(cache); | |
7070 | } | |
7071 | ||
7072 | /* | |
7073 | * walks the btree of allocated extents and find a hole of a given size. | |
7074 | * The key ins is changed to record the hole: | |
7075 | * ins->objectid == start position | |
7076 | * ins->flags = BTRFS_EXTENT_ITEM_KEY | |
7077 | * ins->offset == the size of the hole. | |
7078 | * Any available blocks before search_start are skipped. | |
7079 | * | |
7080 | * If there is no suitable free space, we will record the max size of | |
7081 | * the free space extent currently. | |
7082 | */ | |
7083 | static noinline int find_free_extent(struct btrfs_root *orig_root, | |
7084 | u64 num_bytes, u64 empty_size, | |
7085 | u64 hint_byte, struct btrfs_key *ins, | |
7086 | u64 flags, int delalloc) | |
7087 | { | |
7088 | int ret = 0; | |
7089 | struct btrfs_root *root = orig_root->fs_info->extent_root; | |
7090 | struct btrfs_free_cluster *last_ptr = NULL; | |
7091 | struct btrfs_block_group_cache *block_group = NULL; | |
7092 | u64 search_start = 0; | |
7093 | u64 max_extent_size = 0; | |
7094 | u64 empty_cluster = 0; | |
7095 | struct btrfs_space_info *space_info; | |
7096 | int loop = 0; | |
7097 | int index = __get_raid_index(flags); | |
7098 | int alloc_type = (flags & BTRFS_BLOCK_GROUP_DATA) ? | |
7099 | RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC; | |
7100 | bool failed_cluster_refill = false; | |
7101 | bool failed_alloc = false; | |
7102 | bool use_cluster = true; | |
7103 | bool have_caching_bg = false; | |
7104 | bool orig_have_caching_bg = false; | |
7105 | bool full_search = false; | |
7106 | ||
7107 | WARN_ON(num_bytes < root->sectorsize); | |
7108 | ins->type = BTRFS_EXTENT_ITEM_KEY; | |
7109 | ins->objectid = 0; | |
7110 | ins->offset = 0; | |
7111 | ||
7112 | trace_find_free_extent(orig_root, num_bytes, empty_size, flags); | |
7113 | ||
7114 | space_info = __find_space_info(root->fs_info, flags); | |
7115 | if (!space_info) { | |
7116 | btrfs_err(root->fs_info, "No space info for %llu", flags); | |
7117 | return -ENOSPC; | |
7118 | } | |
7119 | ||
7120 | /* | |
7121 | * If our free space is heavily fragmented we may not be able to make | |
7122 | * big contiguous allocations, so instead of doing the expensive search | |
7123 | * for free space, simply return ENOSPC with our max_extent_size so we | |
7124 | * can go ahead and search for a more manageable chunk. | |
7125 | * | |
7126 | * If our max_extent_size is large enough for our allocation simply | |
7127 | * disable clustering since we will likely not be able to find enough | |
7128 | * space to create a cluster and induce latency trying. | |
7129 | */ | |
7130 | if (unlikely(space_info->max_extent_size)) { | |
7131 | spin_lock(&space_info->lock); | |
7132 | if (space_info->max_extent_size && | |
7133 | num_bytes > space_info->max_extent_size) { | |
7134 | ins->offset = space_info->max_extent_size; | |
7135 | spin_unlock(&space_info->lock); | |
7136 | return -ENOSPC; | |
7137 | } else if (space_info->max_extent_size) { | |
7138 | use_cluster = false; | |
7139 | } | |
7140 | spin_unlock(&space_info->lock); | |
7141 | } | |
7142 | ||
7143 | last_ptr = fetch_cluster_info(orig_root, space_info, &empty_cluster); | |
7144 | if (last_ptr) { | |
7145 | spin_lock(&last_ptr->lock); | |
7146 | if (last_ptr->block_group) | |
7147 | hint_byte = last_ptr->window_start; | |
7148 | if (last_ptr->fragmented) { | |
7149 | /* | |
7150 | * We still set window_start so we can keep track of the | |
7151 | * last place we found an allocation to try and save | |
7152 | * some time. | |
7153 | */ | |
7154 | hint_byte = last_ptr->window_start; | |
7155 | use_cluster = false; | |
7156 | } | |
7157 | spin_unlock(&last_ptr->lock); | |
7158 | } | |
7159 | ||
7160 | search_start = max(search_start, first_logical_byte(root, 0)); | |
7161 | search_start = max(search_start, hint_byte); | |
7162 | if (search_start == hint_byte) { | |
7163 | block_group = btrfs_lookup_block_group(root->fs_info, | |
7164 | search_start); | |
7165 | /* | |
7166 | * we don't want to use the block group if it doesn't match our | |
7167 | * allocation bits, or if its not cached. | |
7168 | * | |
7169 | * However if we are re-searching with an ideal block group | |
7170 | * picked out then we don't care that the block group is cached. | |
7171 | */ | |
7172 | if (block_group && block_group_bits(block_group, flags) && | |
7173 | block_group->cached != BTRFS_CACHE_NO) { | |
7174 | down_read(&space_info->groups_sem); | |
7175 | if (list_empty(&block_group->list) || | |
7176 | block_group->ro) { | |
7177 | /* | |
7178 | * someone is removing this block group, | |
7179 | * we can't jump into the have_block_group | |
7180 | * target because our list pointers are not | |
7181 | * valid | |
7182 | */ | |
7183 | btrfs_put_block_group(block_group); | |
7184 | up_read(&space_info->groups_sem); | |
7185 | } else { | |
7186 | index = get_block_group_index(block_group); | |
7187 | btrfs_lock_block_group(block_group, delalloc); | |
7188 | goto have_block_group; | |
7189 | } | |
7190 | } else if (block_group) { | |
7191 | btrfs_put_block_group(block_group); | |
7192 | } | |
7193 | } | |
7194 | search: | |
7195 | have_caching_bg = false; | |
7196 | if (index == 0 || index == __get_raid_index(flags)) | |
7197 | full_search = true; | |
7198 | down_read(&space_info->groups_sem); | |
7199 | list_for_each_entry(block_group, &space_info->block_groups[index], | |
7200 | list) { | |
7201 | u64 offset; | |
7202 | int cached; | |
7203 | ||
7204 | btrfs_grab_block_group(block_group, delalloc); | |
7205 | search_start = block_group->key.objectid; | |
7206 | ||
7207 | /* | |
7208 | * this can happen if we end up cycling through all the | |
7209 | * raid types, but we want to make sure we only allocate | |
7210 | * for the proper type. | |
7211 | */ | |
7212 | if (!block_group_bits(block_group, flags)) { | |
7213 | u64 extra = BTRFS_BLOCK_GROUP_DUP | | |
7214 | BTRFS_BLOCK_GROUP_RAID1 | | |
7215 | BTRFS_BLOCK_GROUP_RAID5 | | |
7216 | BTRFS_BLOCK_GROUP_RAID6 | | |
7217 | BTRFS_BLOCK_GROUP_RAID10; | |
7218 | ||
7219 | /* | |
7220 | * if they asked for extra copies and this block group | |
7221 | * doesn't provide them, bail. This does allow us to | |
7222 | * fill raid0 from raid1. | |
7223 | */ | |
7224 | if ((flags & extra) && !(block_group->flags & extra)) | |
7225 | goto loop; | |
7226 | } | |
7227 | ||
7228 | have_block_group: | |
7229 | cached = block_group_cache_done(block_group); | |
7230 | if (unlikely(!cached)) { | |
7231 | have_caching_bg = true; | |
7232 | ret = cache_block_group(block_group, 0); | |
7233 | BUG_ON(ret < 0); | |
7234 | ret = 0; | |
7235 | } | |
7236 | ||
7237 | if (unlikely(block_group->cached == BTRFS_CACHE_ERROR)) | |
7238 | goto loop; | |
7239 | if (unlikely(block_group->ro)) | |
7240 | goto loop; | |
7241 | ||
7242 | /* | |
7243 | * Ok we want to try and use the cluster allocator, so | |
7244 | * lets look there | |
7245 | */ | |
7246 | if (last_ptr && use_cluster) { | |
7247 | struct btrfs_block_group_cache *used_block_group; | |
7248 | unsigned long aligned_cluster; | |
7249 | /* | |
7250 | * the refill lock keeps out other | |
7251 | * people trying to start a new cluster | |
7252 | */ | |
7253 | used_block_group = btrfs_lock_cluster(block_group, | |
7254 | last_ptr, | |
7255 | delalloc); | |
7256 | if (!used_block_group) | |
7257 | goto refill_cluster; | |
7258 | ||
7259 | if (used_block_group != block_group && | |
7260 | (used_block_group->ro || | |
7261 | !block_group_bits(used_block_group, flags))) | |
7262 | goto release_cluster; | |
7263 | ||
7264 | offset = btrfs_alloc_from_cluster(used_block_group, | |
7265 | last_ptr, | |
7266 | num_bytes, | |
7267 | used_block_group->key.objectid, | |
7268 | &max_extent_size); | |
7269 | if (offset) { | |
7270 | /* we have a block, we're done */ | |
7271 | spin_unlock(&last_ptr->refill_lock); | |
7272 | trace_btrfs_reserve_extent_cluster(root, | |
7273 | used_block_group, | |
7274 | search_start, num_bytes); | |
7275 | if (used_block_group != block_group) { | |
7276 | btrfs_release_block_group(block_group, | |
7277 | delalloc); | |
7278 | block_group = used_block_group; | |
7279 | } | |
7280 | goto checks; | |
7281 | } | |
7282 | ||
7283 | WARN_ON(last_ptr->block_group != used_block_group); | |
7284 | release_cluster: | |
7285 | /* If we are on LOOP_NO_EMPTY_SIZE, we can't | |
7286 | * set up a new clusters, so lets just skip it | |
7287 | * and let the allocator find whatever block | |
7288 | * it can find. If we reach this point, we | |
7289 | * will have tried the cluster allocator | |
7290 | * plenty of times and not have found | |
7291 | * anything, so we are likely way too | |
7292 | * fragmented for the clustering stuff to find | |
7293 | * anything. | |
7294 | * | |
7295 | * However, if the cluster is taken from the | |
7296 | * current block group, release the cluster | |
7297 | * first, so that we stand a better chance of | |
7298 | * succeeding in the unclustered | |
7299 | * allocation. */ | |
7300 | if (loop >= LOOP_NO_EMPTY_SIZE && | |
7301 | used_block_group != block_group) { | |
7302 | spin_unlock(&last_ptr->refill_lock); | |
7303 | btrfs_release_block_group(used_block_group, | |
7304 | delalloc); | |
7305 | goto unclustered_alloc; | |
7306 | } | |
7307 | ||
7308 | /* | |
7309 | * this cluster didn't work out, free it and | |
7310 | * start over | |
7311 | */ | |
7312 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
7313 | ||
7314 | if (used_block_group != block_group) | |
7315 | btrfs_release_block_group(used_block_group, | |
7316 | delalloc); | |
7317 | refill_cluster: | |
7318 | if (loop >= LOOP_NO_EMPTY_SIZE) { | |
7319 | spin_unlock(&last_ptr->refill_lock); | |
7320 | goto unclustered_alloc; | |
7321 | } | |
7322 | ||
7323 | aligned_cluster = max_t(unsigned long, | |
7324 | empty_cluster + empty_size, | |
7325 | block_group->full_stripe_len); | |
7326 | ||
7327 | /* allocate a cluster in this block group */ | |
7328 | ret = btrfs_find_space_cluster(root, block_group, | |
7329 | last_ptr, search_start, | |
7330 | num_bytes, | |
7331 | aligned_cluster); | |
7332 | if (ret == 0) { | |
7333 | /* | |
7334 | * now pull our allocation out of this | |
7335 | * cluster | |
7336 | */ | |
7337 | offset = btrfs_alloc_from_cluster(block_group, | |
7338 | last_ptr, | |
7339 | num_bytes, | |
7340 | search_start, | |
7341 | &max_extent_size); | |
7342 | if (offset) { | |
7343 | /* we found one, proceed */ | |
7344 | spin_unlock(&last_ptr->refill_lock); | |
7345 | trace_btrfs_reserve_extent_cluster(root, | |
7346 | block_group, search_start, | |
7347 | num_bytes); | |
7348 | goto checks; | |
7349 | } | |
7350 | } else if (!cached && loop > LOOP_CACHING_NOWAIT | |
7351 | && !failed_cluster_refill) { | |
7352 | spin_unlock(&last_ptr->refill_lock); | |
7353 | ||
7354 | failed_cluster_refill = true; | |
7355 | wait_block_group_cache_progress(block_group, | |
7356 | num_bytes + empty_cluster + empty_size); | |
7357 | goto have_block_group; | |
7358 | } | |
7359 | ||
7360 | /* | |
7361 | * at this point we either didn't find a cluster | |
7362 | * or we weren't able to allocate a block from our | |
7363 | * cluster. Free the cluster we've been trying | |
7364 | * to use, and go to the next block group | |
7365 | */ | |
7366 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
7367 | spin_unlock(&last_ptr->refill_lock); | |
7368 | goto loop; | |
7369 | } | |
7370 | ||
7371 | unclustered_alloc: | |
7372 | /* | |
7373 | * We are doing an unclustered alloc, set the fragmented flag so | |
7374 | * we don't bother trying to setup a cluster again until we get | |
7375 | * more space. | |
7376 | */ | |
7377 | if (unlikely(last_ptr)) { | |
7378 | spin_lock(&last_ptr->lock); | |
7379 | last_ptr->fragmented = 1; | |
7380 | spin_unlock(&last_ptr->lock); | |
7381 | } | |
7382 | spin_lock(&block_group->free_space_ctl->tree_lock); | |
7383 | if (cached && | |
7384 | block_group->free_space_ctl->free_space < | |
7385 | num_bytes + empty_cluster + empty_size) { | |
7386 | if (block_group->free_space_ctl->free_space > | |
7387 | max_extent_size) | |
7388 | max_extent_size = | |
7389 | block_group->free_space_ctl->free_space; | |
7390 | spin_unlock(&block_group->free_space_ctl->tree_lock); | |
7391 | goto loop; | |
7392 | } | |
7393 | spin_unlock(&block_group->free_space_ctl->tree_lock); | |
7394 | ||
7395 | offset = btrfs_find_space_for_alloc(block_group, search_start, | |
7396 | num_bytes, empty_size, | |
7397 | &max_extent_size); | |
7398 | /* | |
7399 | * If we didn't find a chunk, and we haven't failed on this | |
7400 | * block group before, and this block group is in the middle of | |
7401 | * caching and we are ok with waiting, then go ahead and wait | |
7402 | * for progress to be made, and set failed_alloc to true. | |
7403 | * | |
7404 | * If failed_alloc is true then we've already waited on this | |
7405 | * block group once and should move on to the next block group. | |
7406 | */ | |
7407 | if (!offset && !failed_alloc && !cached && | |
7408 | loop > LOOP_CACHING_NOWAIT) { | |
7409 | wait_block_group_cache_progress(block_group, | |
7410 | num_bytes + empty_size); | |
7411 | failed_alloc = true; | |
7412 | goto have_block_group; | |
7413 | } else if (!offset) { | |
7414 | goto loop; | |
7415 | } | |
7416 | checks: | |
7417 | search_start = ALIGN(offset, root->stripesize); | |
7418 | ||
7419 | /* move on to the next group */ | |
7420 | if (search_start + num_bytes > | |
7421 | block_group->key.objectid + block_group->key.offset) { | |
7422 | btrfs_add_free_space(block_group, offset, num_bytes); | |
7423 | goto loop; | |
7424 | } | |
7425 | ||
7426 | if (offset < search_start) | |
7427 | btrfs_add_free_space(block_group, offset, | |
7428 | search_start - offset); | |
7429 | BUG_ON(offset > search_start); | |
7430 | ||
7431 | ret = btrfs_update_reserved_bytes(block_group, num_bytes, | |
7432 | alloc_type, delalloc); | |
7433 | if (ret == -EAGAIN) { | |
7434 | btrfs_add_free_space(block_group, offset, num_bytes); | |
7435 | goto loop; | |
7436 | } | |
7437 | ||
7438 | /* we are all good, lets return */ | |
7439 | ins->objectid = search_start; | |
7440 | ins->offset = num_bytes; | |
7441 | ||
7442 | trace_btrfs_reserve_extent(orig_root, block_group, | |
7443 | search_start, num_bytes); | |
7444 | btrfs_release_block_group(block_group, delalloc); | |
7445 | break; | |
7446 | loop: | |
7447 | failed_cluster_refill = false; | |
7448 | failed_alloc = false; | |
7449 | BUG_ON(index != get_block_group_index(block_group)); | |
7450 | btrfs_release_block_group(block_group, delalloc); | |
7451 | } | |
7452 | up_read(&space_info->groups_sem); | |
7453 | ||
7454 | if ((loop == LOOP_CACHING_NOWAIT) && have_caching_bg | |
7455 | && !orig_have_caching_bg) | |
7456 | orig_have_caching_bg = true; | |
7457 | ||
7458 | if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg) | |
7459 | goto search; | |
7460 | ||
7461 | if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES) | |
7462 | goto search; | |
7463 | ||
7464 | /* | |
7465 | * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking | |
7466 | * caching kthreads as we move along | |
7467 | * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching | |
7468 | * LOOP_ALLOC_CHUNK, force a chunk allocation and try again | |
7469 | * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try | |
7470 | * again | |
7471 | */ | |
7472 | if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) { | |
7473 | index = 0; | |
7474 | if (loop == LOOP_CACHING_NOWAIT) { | |
7475 | /* | |
7476 | * We want to skip the LOOP_CACHING_WAIT step if we | |
7477 | * don't have any unached bgs and we've alrelady done a | |
7478 | * full search through. | |
7479 | */ | |
7480 | if (orig_have_caching_bg || !full_search) | |
7481 | loop = LOOP_CACHING_WAIT; | |
7482 | else | |
7483 | loop = LOOP_ALLOC_CHUNK; | |
7484 | } else { | |
7485 | loop++; | |
7486 | } | |
7487 | ||
7488 | if (loop == LOOP_ALLOC_CHUNK) { | |
7489 | struct btrfs_trans_handle *trans; | |
7490 | int exist = 0; | |
7491 | ||
7492 | trans = current->journal_info; | |
7493 | if (trans) | |
7494 | exist = 1; | |
7495 | else | |
7496 | trans = btrfs_join_transaction(root); | |
7497 | ||
7498 | if (IS_ERR(trans)) { | |
7499 | ret = PTR_ERR(trans); | |
7500 | goto out; | |
7501 | } | |
7502 | ||
7503 | ret = do_chunk_alloc(trans, root, flags, | |
7504 | CHUNK_ALLOC_FORCE); | |
7505 | ||
7506 | /* | |
7507 | * If we can't allocate a new chunk we've already looped | |
7508 | * through at least once, move on to the NO_EMPTY_SIZE | |
7509 | * case. | |
7510 | */ | |
7511 | if (ret == -ENOSPC) | |
7512 | loop = LOOP_NO_EMPTY_SIZE; | |
7513 | ||
7514 | /* | |
7515 | * Do not bail out on ENOSPC since we | |
7516 | * can do more things. | |
7517 | */ | |
7518 | if (ret < 0 && ret != -ENOSPC) | |
7519 | btrfs_abort_transaction(trans, | |
7520 | root, ret); | |
7521 | else | |
7522 | ret = 0; | |
7523 | if (!exist) | |
7524 | btrfs_end_transaction(trans, root); | |
7525 | if (ret) | |
7526 | goto out; | |
7527 | } | |
7528 | ||
7529 | if (loop == LOOP_NO_EMPTY_SIZE) { | |
7530 | /* | |
7531 | * Don't loop again if we already have no empty_size and | |
7532 | * no empty_cluster. | |
7533 | */ | |
7534 | if (empty_size == 0 && | |
7535 | empty_cluster == 0) { | |
7536 | ret = -ENOSPC; | |
7537 | goto out; | |
7538 | } | |
7539 | empty_size = 0; | |
7540 | empty_cluster = 0; | |
7541 | } | |
7542 | ||
7543 | goto search; | |
7544 | } else if (!ins->objectid) { | |
7545 | ret = -ENOSPC; | |
7546 | } else if (ins->objectid) { | |
7547 | if (!use_cluster && last_ptr) { | |
7548 | spin_lock(&last_ptr->lock); | |
7549 | last_ptr->window_start = ins->objectid; | |
7550 | spin_unlock(&last_ptr->lock); | |
7551 | } | |
7552 | ret = 0; | |
7553 | } | |
7554 | out: | |
7555 | if (ret == -ENOSPC) { | |
7556 | spin_lock(&space_info->lock); | |
7557 | space_info->max_extent_size = max_extent_size; | |
7558 | spin_unlock(&space_info->lock); | |
7559 | ins->offset = max_extent_size; | |
7560 | } | |
7561 | return ret; | |
7562 | } | |
7563 | ||
7564 | static void dump_space_info(struct btrfs_space_info *info, u64 bytes, | |
7565 | int dump_block_groups) | |
7566 | { | |
7567 | struct btrfs_block_group_cache *cache; | |
7568 | int index = 0; | |
7569 | ||
7570 | spin_lock(&info->lock); | |
7571 | printk(KERN_INFO "BTRFS: space_info %llu has %llu free, is %sfull\n", | |
7572 | info->flags, | |
7573 | info->total_bytes - info->bytes_used - info->bytes_pinned - | |
7574 | info->bytes_reserved - info->bytes_readonly, | |
7575 | (info->full) ? "" : "not "); | |
7576 | printk(KERN_INFO "BTRFS: space_info total=%llu, used=%llu, pinned=%llu, " | |
7577 | "reserved=%llu, may_use=%llu, readonly=%llu\n", | |
7578 | info->total_bytes, info->bytes_used, info->bytes_pinned, | |
7579 | info->bytes_reserved, info->bytes_may_use, | |
7580 | info->bytes_readonly); | |
7581 | spin_unlock(&info->lock); | |
7582 | ||
7583 | if (!dump_block_groups) | |
7584 | return; | |
7585 | ||
7586 | down_read(&info->groups_sem); | |
7587 | again: | |
7588 | list_for_each_entry(cache, &info->block_groups[index], list) { | |
7589 | spin_lock(&cache->lock); | |
7590 | printk(KERN_INFO "BTRFS: " | |
7591 | "block group %llu has %llu bytes, " | |
7592 | "%llu used %llu pinned %llu reserved %s\n", | |
7593 | cache->key.objectid, cache->key.offset, | |
7594 | btrfs_block_group_used(&cache->item), cache->pinned, | |
7595 | cache->reserved, cache->ro ? "[readonly]" : ""); | |
7596 | btrfs_dump_free_space(cache, bytes); | |
7597 | spin_unlock(&cache->lock); | |
7598 | } | |
7599 | if (++index < BTRFS_NR_RAID_TYPES) | |
7600 | goto again; | |
7601 | up_read(&info->groups_sem); | |
7602 | } | |
7603 | ||
7604 | int btrfs_reserve_extent(struct btrfs_root *root, | |
7605 | u64 num_bytes, u64 min_alloc_size, | |
7606 | u64 empty_size, u64 hint_byte, | |
7607 | struct btrfs_key *ins, int is_data, int delalloc) | |
7608 | { | |
7609 | bool final_tried = num_bytes == min_alloc_size; | |
7610 | u64 flags; | |
7611 | int ret; | |
7612 | ||
7613 | flags = btrfs_get_alloc_profile(root, is_data); | |
7614 | again: | |
7615 | WARN_ON(num_bytes < root->sectorsize); | |
7616 | ret = find_free_extent(root, num_bytes, empty_size, hint_byte, ins, | |
7617 | flags, delalloc); | |
7618 | ||
7619 | if (ret == -ENOSPC) { | |
7620 | if (!final_tried && ins->offset) { | |
7621 | num_bytes = min(num_bytes >> 1, ins->offset); | |
7622 | num_bytes = round_down(num_bytes, root->sectorsize); | |
7623 | num_bytes = max(num_bytes, min_alloc_size); | |
7624 | if (num_bytes == min_alloc_size) | |
7625 | final_tried = true; | |
7626 | goto again; | |
7627 | } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) { | |
7628 | struct btrfs_space_info *sinfo; | |
7629 | ||
7630 | sinfo = __find_space_info(root->fs_info, flags); | |
7631 | btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu", | |
7632 | flags, num_bytes); | |
7633 | if (sinfo) | |
7634 | dump_space_info(sinfo, num_bytes, 1); | |
7635 | } | |
7636 | } | |
7637 | ||
7638 | return ret; | |
7639 | } | |
7640 | ||
7641 | static int __btrfs_free_reserved_extent(struct btrfs_root *root, | |
7642 | u64 start, u64 len, | |
7643 | int pin, int delalloc) | |
7644 | { | |
7645 | struct btrfs_block_group_cache *cache; | |
7646 | int ret = 0; | |
7647 | ||
7648 | cache = btrfs_lookup_block_group(root->fs_info, start); | |
7649 | if (!cache) { | |
7650 | btrfs_err(root->fs_info, "Unable to find block group for %llu", | |
7651 | start); | |
7652 | return -ENOSPC; | |
7653 | } | |
7654 | ||
7655 | if (pin) | |
7656 | pin_down_extent(root, cache, start, len, 1); | |
7657 | else { | |
7658 | if (btrfs_test_opt(root, DISCARD)) | |
7659 | ret = btrfs_discard_extent(root, start, len, NULL); | |
7660 | btrfs_add_free_space(cache, start, len); | |
7661 | btrfs_update_reserved_bytes(cache, len, RESERVE_FREE, delalloc); | |
7662 | } | |
7663 | ||
7664 | btrfs_put_block_group(cache); | |
7665 | ||
7666 | trace_btrfs_reserved_extent_free(root, start, len); | |
7667 | ||
7668 | return ret; | |
7669 | } | |
7670 | ||
7671 | int btrfs_free_reserved_extent(struct btrfs_root *root, | |
7672 | u64 start, u64 len, int delalloc) | |
7673 | { | |
7674 | return __btrfs_free_reserved_extent(root, start, len, 0, delalloc); | |
7675 | } | |
7676 | ||
7677 | int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root, | |
7678 | u64 start, u64 len) | |
7679 | { | |
7680 | return __btrfs_free_reserved_extent(root, start, len, 1, 0); | |
7681 | } | |
7682 | ||
7683 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
7684 | struct btrfs_root *root, | |
7685 | u64 parent, u64 root_objectid, | |
7686 | u64 flags, u64 owner, u64 offset, | |
7687 | struct btrfs_key *ins, int ref_mod) | |
7688 | { | |
7689 | int ret; | |
7690 | struct btrfs_fs_info *fs_info = root->fs_info; | |
7691 | struct btrfs_extent_item *extent_item; | |
7692 | struct btrfs_extent_inline_ref *iref; | |
7693 | struct btrfs_path *path; | |
7694 | struct extent_buffer *leaf; | |
7695 | int type; | |
7696 | u32 size; | |
7697 | ||
7698 | if (parent > 0) | |
7699 | type = BTRFS_SHARED_DATA_REF_KEY; | |
7700 | else | |
7701 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
7702 | ||
7703 | size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type); | |
7704 | ||
7705 | path = btrfs_alloc_path(); | |
7706 | if (!path) | |
7707 | return -ENOMEM; | |
7708 | ||
7709 | path->leave_spinning = 1; | |
7710 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, | |
7711 | ins, size); | |
7712 | if (ret) { | |
7713 | btrfs_free_path(path); | |
7714 | return ret; | |
7715 | } | |
7716 | ||
7717 | leaf = path->nodes[0]; | |
7718 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
7719 | struct btrfs_extent_item); | |
7720 | btrfs_set_extent_refs(leaf, extent_item, ref_mod); | |
7721 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
7722 | btrfs_set_extent_flags(leaf, extent_item, | |
7723 | flags | BTRFS_EXTENT_FLAG_DATA); | |
7724 | ||
7725 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
7726 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
7727 | if (parent > 0) { | |
7728 | struct btrfs_shared_data_ref *ref; | |
7729 | ref = (struct btrfs_shared_data_ref *)(iref + 1); | |
7730 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
7731 | btrfs_set_shared_data_ref_count(leaf, ref, ref_mod); | |
7732 | } else { | |
7733 | struct btrfs_extent_data_ref *ref; | |
7734 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
7735 | btrfs_set_extent_data_ref_root(leaf, ref, root_objectid); | |
7736 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
7737 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
7738 | btrfs_set_extent_data_ref_count(leaf, ref, ref_mod); | |
7739 | } | |
7740 | ||
7741 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
7742 | btrfs_free_path(path); | |
7743 | ||
7744 | ret = remove_from_free_space_tree(trans, fs_info, ins->objectid, | |
7745 | ins->offset); | |
7746 | if (ret) | |
7747 | return ret; | |
7748 | ||
7749 | ret = update_block_group(trans, root, ins->objectid, ins->offset, 1); | |
7750 | if (ret) { /* -ENOENT, logic error */ | |
7751 | btrfs_err(fs_info, "update block group failed for %llu %llu", | |
7752 | ins->objectid, ins->offset); | |
7753 | BUG(); | |
7754 | } | |
7755 | trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset); | |
7756 | return ret; | |
7757 | } | |
7758 | ||
7759 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, | |
7760 | struct btrfs_root *root, | |
7761 | u64 parent, u64 root_objectid, | |
7762 | u64 flags, struct btrfs_disk_key *key, | |
7763 | int level, struct btrfs_key *ins) | |
7764 | { | |
7765 | int ret; | |
7766 | struct btrfs_fs_info *fs_info = root->fs_info; | |
7767 | struct btrfs_extent_item *extent_item; | |
7768 | struct btrfs_tree_block_info *block_info; | |
7769 | struct btrfs_extent_inline_ref *iref; | |
7770 | struct btrfs_path *path; | |
7771 | struct extent_buffer *leaf; | |
7772 | u32 size = sizeof(*extent_item) + sizeof(*iref); | |
7773 | u64 num_bytes = ins->offset; | |
7774 | bool skinny_metadata = btrfs_fs_incompat(root->fs_info, | |
7775 | SKINNY_METADATA); | |
7776 | ||
7777 | if (!skinny_metadata) | |
7778 | size += sizeof(*block_info); | |
7779 | ||
7780 | path = btrfs_alloc_path(); | |
7781 | if (!path) { | |
7782 | btrfs_free_and_pin_reserved_extent(root, ins->objectid, | |
7783 | root->nodesize); | |
7784 | return -ENOMEM; | |
7785 | } | |
7786 | ||
7787 | path->leave_spinning = 1; | |
7788 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, | |
7789 | ins, size); | |
7790 | if (ret) { | |
7791 | btrfs_free_path(path); | |
7792 | btrfs_free_and_pin_reserved_extent(root, ins->objectid, | |
7793 | root->nodesize); | |
7794 | return ret; | |
7795 | } | |
7796 | ||
7797 | leaf = path->nodes[0]; | |
7798 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
7799 | struct btrfs_extent_item); | |
7800 | btrfs_set_extent_refs(leaf, extent_item, 1); | |
7801 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
7802 | btrfs_set_extent_flags(leaf, extent_item, | |
7803 | flags | BTRFS_EXTENT_FLAG_TREE_BLOCK); | |
7804 | ||
7805 | if (skinny_metadata) { | |
7806 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
7807 | num_bytes = root->nodesize; | |
7808 | } else { | |
7809 | block_info = (struct btrfs_tree_block_info *)(extent_item + 1); | |
7810 | btrfs_set_tree_block_key(leaf, block_info, key); | |
7811 | btrfs_set_tree_block_level(leaf, block_info, level); | |
7812 | iref = (struct btrfs_extent_inline_ref *)(block_info + 1); | |
7813 | } | |
7814 | ||
7815 | if (parent > 0) { | |
7816 | BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); | |
7817 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
7818 | BTRFS_SHARED_BLOCK_REF_KEY); | |
7819 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
7820 | } else { | |
7821 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
7822 | BTRFS_TREE_BLOCK_REF_KEY); | |
7823 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
7824 | } | |
7825 | ||
7826 | btrfs_mark_buffer_dirty(leaf); | |
7827 | btrfs_free_path(path); | |
7828 | ||
7829 | ret = remove_from_free_space_tree(trans, fs_info, ins->objectid, | |
7830 | num_bytes); | |
7831 | if (ret) | |
7832 | return ret; | |
7833 | ||
7834 | ret = update_block_group(trans, root, ins->objectid, root->nodesize, | |
7835 | 1); | |
7836 | if (ret) { /* -ENOENT, logic error */ | |
7837 | btrfs_err(fs_info, "update block group failed for %llu %llu", | |
7838 | ins->objectid, ins->offset); | |
7839 | BUG(); | |
7840 | } | |
7841 | ||
7842 | trace_btrfs_reserved_extent_alloc(root, ins->objectid, root->nodesize); | |
7843 | return ret; | |
7844 | } | |
7845 | ||
7846 | int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
7847 | struct btrfs_root *root, | |
7848 | u64 root_objectid, u64 owner, | |
7849 | u64 offset, u64 ram_bytes, | |
7850 | struct btrfs_key *ins) | |
7851 | { | |
7852 | int ret; | |
7853 | ||
7854 | BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
7855 | ||
7856 | ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid, | |
7857 | ins->offset, 0, | |
7858 | root_objectid, owner, offset, | |
7859 | ram_bytes, BTRFS_ADD_DELAYED_EXTENT, | |
7860 | NULL); | |
7861 | return ret; | |
7862 | } | |
7863 | ||
7864 | /* | |
7865 | * this is used by the tree logging recovery code. It records that | |
7866 | * an extent has been allocated and makes sure to clear the free | |
7867 | * space cache bits as well | |
7868 | */ | |
7869 | int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans, | |
7870 | struct btrfs_root *root, | |
7871 | u64 root_objectid, u64 owner, u64 offset, | |
7872 | struct btrfs_key *ins) | |
7873 | { | |
7874 | int ret; | |
7875 | struct btrfs_block_group_cache *block_group; | |
7876 | ||
7877 | /* | |
7878 | * Mixed block groups will exclude before processing the log so we only | |
7879 | * need to do the exlude dance if this fs isn't mixed. | |
7880 | */ | |
7881 | if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) { | |
7882 | ret = __exclude_logged_extent(root, ins->objectid, ins->offset); | |
7883 | if (ret) | |
7884 | return ret; | |
7885 | } | |
7886 | ||
7887 | block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid); | |
7888 | if (!block_group) | |
7889 | return -EINVAL; | |
7890 | ||
7891 | ret = btrfs_update_reserved_bytes(block_group, ins->offset, | |
7892 | RESERVE_ALLOC_NO_ACCOUNT, 0); | |
7893 | BUG_ON(ret); /* logic error */ | |
7894 | ret = alloc_reserved_file_extent(trans, root, 0, root_objectid, | |
7895 | 0, owner, offset, ins, 1); | |
7896 | btrfs_put_block_group(block_group); | |
7897 | return ret; | |
7898 | } | |
7899 | ||
7900 | static struct extent_buffer * | |
7901 | btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
7902 | u64 bytenr, int level) | |
7903 | { | |
7904 | struct extent_buffer *buf; | |
7905 | ||
7906 | buf = btrfs_find_create_tree_block(root, bytenr); | |
7907 | if (!buf) | |
7908 | return ERR_PTR(-ENOMEM); | |
7909 | btrfs_set_header_generation(buf, trans->transid); | |
7910 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level); | |
7911 | btrfs_tree_lock(buf); | |
7912 | clean_tree_block(trans, root->fs_info, buf); | |
7913 | clear_bit(EXTENT_BUFFER_STALE, &buf->bflags); | |
7914 | ||
7915 | btrfs_set_lock_blocking(buf); | |
7916 | set_extent_buffer_uptodate(buf); | |
7917 | ||
7918 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { | |
7919 | buf->log_index = root->log_transid % 2; | |
7920 | /* | |
7921 | * we allow two log transactions at a time, use different | |
7922 | * EXENT bit to differentiate dirty pages. | |
7923 | */ | |
7924 | if (buf->log_index == 0) | |
7925 | set_extent_dirty(&root->dirty_log_pages, buf->start, | |
7926 | buf->start + buf->len - 1, GFP_NOFS); | |
7927 | else | |
7928 | set_extent_new(&root->dirty_log_pages, buf->start, | |
7929 | buf->start + buf->len - 1, GFP_NOFS); | |
7930 | } else { | |
7931 | buf->log_index = -1; | |
7932 | set_extent_dirty(&trans->transaction->dirty_pages, buf->start, | |
7933 | buf->start + buf->len - 1, GFP_NOFS); | |
7934 | } | |
7935 | trans->blocks_used++; | |
7936 | /* this returns a buffer locked for blocking */ | |
7937 | return buf; | |
7938 | } | |
7939 | ||
7940 | static struct btrfs_block_rsv * | |
7941 | use_block_rsv(struct btrfs_trans_handle *trans, | |
7942 | struct btrfs_root *root, u32 blocksize) | |
7943 | { | |
7944 | struct btrfs_block_rsv *block_rsv; | |
7945 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; | |
7946 | int ret; | |
7947 | bool global_updated = false; | |
7948 | ||
7949 | block_rsv = get_block_rsv(trans, root); | |
7950 | ||
7951 | if (unlikely(block_rsv->size == 0)) | |
7952 | goto try_reserve; | |
7953 | again: | |
7954 | ret = block_rsv_use_bytes(block_rsv, blocksize); | |
7955 | if (!ret) | |
7956 | return block_rsv; | |
7957 | ||
7958 | if (block_rsv->failfast) | |
7959 | return ERR_PTR(ret); | |
7960 | ||
7961 | if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) { | |
7962 | global_updated = true; | |
7963 | update_global_block_rsv(root->fs_info); | |
7964 | goto again; | |
7965 | } | |
7966 | ||
7967 | if (btrfs_test_opt(root, ENOSPC_DEBUG)) { | |
7968 | static DEFINE_RATELIMIT_STATE(_rs, | |
7969 | DEFAULT_RATELIMIT_INTERVAL * 10, | |
7970 | /*DEFAULT_RATELIMIT_BURST*/ 1); | |
7971 | if (__ratelimit(&_rs)) | |
7972 | WARN(1, KERN_DEBUG | |
7973 | "BTRFS: block rsv returned %d\n", ret); | |
7974 | } | |
7975 | try_reserve: | |
7976 | ret = reserve_metadata_bytes(root, block_rsv, blocksize, | |
7977 | BTRFS_RESERVE_NO_FLUSH); | |
7978 | if (!ret) | |
7979 | return block_rsv; | |
7980 | /* | |
7981 | * If we couldn't reserve metadata bytes try and use some from | |
7982 | * the global reserve if its space type is the same as the global | |
7983 | * reservation. | |
7984 | */ | |
7985 | if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL && | |
7986 | block_rsv->space_info == global_rsv->space_info) { | |
7987 | ret = block_rsv_use_bytes(global_rsv, blocksize); | |
7988 | if (!ret) | |
7989 | return global_rsv; | |
7990 | } | |
7991 | return ERR_PTR(ret); | |
7992 | } | |
7993 | ||
7994 | static void unuse_block_rsv(struct btrfs_fs_info *fs_info, | |
7995 | struct btrfs_block_rsv *block_rsv, u32 blocksize) | |
7996 | { | |
7997 | block_rsv_add_bytes(block_rsv, blocksize, 0); | |
7998 | block_rsv_release_bytes(fs_info, block_rsv, NULL, 0); | |
7999 | } | |
8000 | ||
8001 | /* | |
8002 | * finds a free extent and does all the dirty work required for allocation | |
8003 | * returns the tree buffer or an ERR_PTR on error. | |
8004 | */ | |
8005 | struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans, | |
8006 | struct btrfs_root *root, | |
8007 | u64 parent, u64 root_objectid, | |
8008 | struct btrfs_disk_key *key, int level, | |
8009 | u64 hint, u64 empty_size) | |
8010 | { | |
8011 | struct btrfs_key ins; | |
8012 | struct btrfs_block_rsv *block_rsv; | |
8013 | struct extent_buffer *buf; | |
8014 | struct btrfs_delayed_extent_op *extent_op; | |
8015 | u64 flags = 0; | |
8016 | int ret; | |
8017 | u32 blocksize = root->nodesize; | |
8018 | bool skinny_metadata = btrfs_fs_incompat(root->fs_info, | |
8019 | SKINNY_METADATA); | |
8020 | ||
8021 | if (btrfs_test_is_dummy_root(root)) { | |
8022 | buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr, | |
8023 | level); | |
8024 | if (!IS_ERR(buf)) | |
8025 | root->alloc_bytenr += blocksize; | |
8026 | return buf; | |
8027 | } | |
8028 | ||
8029 | block_rsv = use_block_rsv(trans, root, blocksize); | |
8030 | if (IS_ERR(block_rsv)) | |
8031 | return ERR_CAST(block_rsv); | |
8032 | ||
8033 | ret = btrfs_reserve_extent(root, blocksize, blocksize, | |
8034 | empty_size, hint, &ins, 0, 0); | |
8035 | if (ret) | |
8036 | goto out_unuse; | |
8037 | ||
8038 | buf = btrfs_init_new_buffer(trans, root, ins.objectid, level); | |
8039 | if (IS_ERR(buf)) { | |
8040 | ret = PTR_ERR(buf); | |
8041 | goto out_free_reserved; | |
8042 | } | |
8043 | ||
8044 | if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
8045 | if (parent == 0) | |
8046 | parent = ins.objectid; | |
8047 | flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
8048 | } else | |
8049 | BUG_ON(parent > 0); | |
8050 | ||
8051 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) { | |
8052 | extent_op = btrfs_alloc_delayed_extent_op(); | |
8053 | if (!extent_op) { | |
8054 | ret = -ENOMEM; | |
8055 | goto out_free_buf; | |
8056 | } | |
8057 | if (key) | |
8058 | memcpy(&extent_op->key, key, sizeof(extent_op->key)); | |
8059 | else | |
8060 | memset(&extent_op->key, 0, sizeof(extent_op->key)); | |
8061 | extent_op->flags_to_set = flags; | |
8062 | extent_op->update_key = skinny_metadata ? false : true; | |
8063 | extent_op->update_flags = true; | |
8064 | extent_op->is_data = false; | |
8065 | extent_op->level = level; | |
8066 | ||
8067 | ret = btrfs_add_delayed_tree_ref(root->fs_info, trans, | |
8068 | ins.objectid, ins.offset, | |
8069 | parent, root_objectid, level, | |
8070 | BTRFS_ADD_DELAYED_EXTENT, | |
8071 | extent_op); | |
8072 | if (ret) | |
8073 | goto out_free_delayed; | |
8074 | } | |
8075 | return buf; | |
8076 | ||
8077 | out_free_delayed: | |
8078 | btrfs_free_delayed_extent_op(extent_op); | |
8079 | out_free_buf: | |
8080 | free_extent_buffer(buf); | |
8081 | out_free_reserved: | |
8082 | btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 0); | |
8083 | out_unuse: | |
8084 | unuse_block_rsv(root->fs_info, block_rsv, blocksize); | |
8085 | return ERR_PTR(ret); | |
8086 | } | |
8087 | ||
8088 | struct walk_control { | |
8089 | u64 refs[BTRFS_MAX_LEVEL]; | |
8090 | u64 flags[BTRFS_MAX_LEVEL]; | |
8091 | struct btrfs_key update_progress; | |
8092 | int stage; | |
8093 | int level; | |
8094 | int shared_level; | |
8095 | int update_ref; | |
8096 | int keep_locks; | |
8097 | int reada_slot; | |
8098 | int reada_count; | |
8099 | int for_reloc; | |
8100 | }; | |
8101 | ||
8102 | #define DROP_REFERENCE 1 | |
8103 | #define UPDATE_BACKREF 2 | |
8104 | ||
8105 | static noinline void reada_walk_down(struct btrfs_trans_handle *trans, | |
8106 | struct btrfs_root *root, | |
8107 | struct walk_control *wc, | |
8108 | struct btrfs_path *path) | |
8109 | { | |
8110 | u64 bytenr; | |
8111 | u64 generation; | |
8112 | u64 refs; | |
8113 | u64 flags; | |
8114 | u32 nritems; | |
8115 | u32 blocksize; | |
8116 | struct btrfs_key key; | |
8117 | struct extent_buffer *eb; | |
8118 | int ret; | |
8119 | int slot; | |
8120 | int nread = 0; | |
8121 | ||
8122 | if (path->slots[wc->level] < wc->reada_slot) { | |
8123 | wc->reada_count = wc->reada_count * 2 / 3; | |
8124 | wc->reada_count = max(wc->reada_count, 2); | |
8125 | } else { | |
8126 | wc->reada_count = wc->reada_count * 3 / 2; | |
8127 | wc->reada_count = min_t(int, wc->reada_count, | |
8128 | BTRFS_NODEPTRS_PER_BLOCK(root)); | |
8129 | } | |
8130 | ||
8131 | eb = path->nodes[wc->level]; | |
8132 | nritems = btrfs_header_nritems(eb); | |
8133 | blocksize = root->nodesize; | |
8134 | ||
8135 | for (slot = path->slots[wc->level]; slot < nritems; slot++) { | |
8136 | if (nread >= wc->reada_count) | |
8137 | break; | |
8138 | ||
8139 | cond_resched(); | |
8140 | bytenr = btrfs_node_blockptr(eb, slot); | |
8141 | generation = btrfs_node_ptr_generation(eb, slot); | |
8142 | ||
8143 | if (slot == path->slots[wc->level]) | |
8144 | goto reada; | |
8145 | ||
8146 | if (wc->stage == UPDATE_BACKREF && | |
8147 | generation <= root->root_key.offset) | |
8148 | continue; | |
8149 | ||
8150 | /* We don't lock the tree block, it's OK to be racy here */ | |
8151 | ret = btrfs_lookup_extent_info(trans, root, bytenr, | |
8152 | wc->level - 1, 1, &refs, | |
8153 | &flags); | |
8154 | /* We don't care about errors in readahead. */ | |
8155 | if (ret < 0) | |
8156 | continue; | |
8157 | BUG_ON(refs == 0); | |
8158 | ||
8159 | if (wc->stage == DROP_REFERENCE) { | |
8160 | if (refs == 1) | |
8161 | goto reada; | |
8162 | ||
8163 | if (wc->level == 1 && | |
8164 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8165 | continue; | |
8166 | if (!wc->update_ref || | |
8167 | generation <= root->root_key.offset) | |
8168 | continue; | |
8169 | btrfs_node_key_to_cpu(eb, &key, slot); | |
8170 | ret = btrfs_comp_cpu_keys(&key, | |
8171 | &wc->update_progress); | |
8172 | if (ret < 0) | |
8173 | continue; | |
8174 | } else { | |
8175 | if (wc->level == 1 && | |
8176 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8177 | continue; | |
8178 | } | |
8179 | reada: | |
8180 | readahead_tree_block(root, bytenr); | |
8181 | nread++; | |
8182 | } | |
8183 | wc->reada_slot = slot; | |
8184 | } | |
8185 | ||
8186 | /* | |
8187 | * These may not be seen by the usual inc/dec ref code so we have to | |
8188 | * add them here. | |
8189 | */ | |
8190 | static int record_one_subtree_extent(struct btrfs_trans_handle *trans, | |
8191 | struct btrfs_root *root, u64 bytenr, | |
8192 | u64 num_bytes) | |
8193 | { | |
8194 | struct btrfs_qgroup_extent_record *qrecord; | |
8195 | struct btrfs_delayed_ref_root *delayed_refs; | |
8196 | ||
8197 | qrecord = kmalloc(sizeof(*qrecord), GFP_NOFS); | |
8198 | if (!qrecord) | |
8199 | return -ENOMEM; | |
8200 | ||
8201 | qrecord->bytenr = bytenr; | |
8202 | qrecord->num_bytes = num_bytes; | |
8203 | qrecord->old_roots = NULL; | |
8204 | ||
8205 | delayed_refs = &trans->transaction->delayed_refs; | |
8206 | spin_lock(&delayed_refs->lock); | |
8207 | if (btrfs_qgroup_insert_dirty_extent(delayed_refs, qrecord)) | |
8208 | kfree(qrecord); | |
8209 | spin_unlock(&delayed_refs->lock); | |
8210 | ||
8211 | return 0; | |
8212 | } | |
8213 | ||
8214 | static int account_leaf_items(struct btrfs_trans_handle *trans, | |
8215 | struct btrfs_root *root, | |
8216 | struct extent_buffer *eb) | |
8217 | { | |
8218 | int nr = btrfs_header_nritems(eb); | |
8219 | int i, extent_type, ret; | |
8220 | struct btrfs_key key; | |
8221 | struct btrfs_file_extent_item *fi; | |
8222 | u64 bytenr, num_bytes; | |
8223 | ||
8224 | /* We can be called directly from walk_up_proc() */ | |
8225 | if (!root->fs_info->quota_enabled) | |
8226 | return 0; | |
8227 | ||
8228 | for (i = 0; i < nr; i++) { | |
8229 | btrfs_item_key_to_cpu(eb, &key, i); | |
8230 | ||
8231 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
8232 | continue; | |
8233 | ||
8234 | fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item); | |
8235 | /* filter out non qgroup-accountable extents */ | |
8236 | extent_type = btrfs_file_extent_type(eb, fi); | |
8237 | ||
8238 | if (extent_type == BTRFS_FILE_EXTENT_INLINE) | |
8239 | continue; | |
8240 | ||
8241 | bytenr = btrfs_file_extent_disk_bytenr(eb, fi); | |
8242 | if (!bytenr) | |
8243 | continue; | |
8244 | ||
8245 | num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi); | |
8246 | ||
8247 | ret = record_one_subtree_extent(trans, root, bytenr, num_bytes); | |
8248 | if (ret) | |
8249 | return ret; | |
8250 | } | |
8251 | return 0; | |
8252 | } | |
8253 | ||
8254 | /* | |
8255 | * Walk up the tree from the bottom, freeing leaves and any interior | |
8256 | * nodes which have had all slots visited. If a node (leaf or | |
8257 | * interior) is freed, the node above it will have it's slot | |
8258 | * incremented. The root node will never be freed. | |
8259 | * | |
8260 | * At the end of this function, we should have a path which has all | |
8261 | * slots incremented to the next position for a search. If we need to | |
8262 | * read a new node it will be NULL and the node above it will have the | |
8263 | * correct slot selected for a later read. | |
8264 | * | |
8265 | * If we increment the root nodes slot counter past the number of | |
8266 | * elements, 1 is returned to signal completion of the search. | |
8267 | */ | |
8268 | static int adjust_slots_upwards(struct btrfs_root *root, | |
8269 | struct btrfs_path *path, int root_level) | |
8270 | { | |
8271 | int level = 0; | |
8272 | int nr, slot; | |
8273 | struct extent_buffer *eb; | |
8274 | ||
8275 | if (root_level == 0) | |
8276 | return 1; | |
8277 | ||
8278 | while (level <= root_level) { | |
8279 | eb = path->nodes[level]; | |
8280 | nr = btrfs_header_nritems(eb); | |
8281 | path->slots[level]++; | |
8282 | slot = path->slots[level]; | |
8283 | if (slot >= nr || level == 0) { | |
8284 | /* | |
8285 | * Don't free the root - we will detect this | |
8286 | * condition after our loop and return a | |
8287 | * positive value for caller to stop walking the tree. | |
8288 | */ | |
8289 | if (level != root_level) { | |
8290 | btrfs_tree_unlock_rw(eb, path->locks[level]); | |
8291 | path->locks[level] = 0; | |
8292 | ||
8293 | free_extent_buffer(eb); | |
8294 | path->nodes[level] = NULL; | |
8295 | path->slots[level] = 0; | |
8296 | } | |
8297 | } else { | |
8298 | /* | |
8299 | * We have a valid slot to walk back down | |
8300 | * from. Stop here so caller can process these | |
8301 | * new nodes. | |
8302 | */ | |
8303 | break; | |
8304 | } | |
8305 | ||
8306 | level++; | |
8307 | } | |
8308 | ||
8309 | eb = path->nodes[root_level]; | |
8310 | if (path->slots[root_level] >= btrfs_header_nritems(eb)) | |
8311 | return 1; | |
8312 | ||
8313 | return 0; | |
8314 | } | |
8315 | ||
8316 | /* | |
8317 | * root_eb is the subtree root and is locked before this function is called. | |
8318 | */ | |
8319 | static int account_shared_subtree(struct btrfs_trans_handle *trans, | |
8320 | struct btrfs_root *root, | |
8321 | struct extent_buffer *root_eb, | |
8322 | u64 root_gen, | |
8323 | int root_level) | |
8324 | { | |
8325 | int ret = 0; | |
8326 | int level; | |
8327 | struct extent_buffer *eb = root_eb; | |
8328 | struct btrfs_path *path = NULL; | |
8329 | ||
8330 | BUG_ON(root_level < 0 || root_level > BTRFS_MAX_LEVEL); | |
8331 | BUG_ON(root_eb == NULL); | |
8332 | ||
8333 | if (!root->fs_info->quota_enabled) | |
8334 | return 0; | |
8335 | ||
8336 | if (!extent_buffer_uptodate(root_eb)) { | |
8337 | ret = btrfs_read_buffer(root_eb, root_gen); | |
8338 | if (ret) | |
8339 | goto out; | |
8340 | } | |
8341 | ||
8342 | if (root_level == 0) { | |
8343 | ret = account_leaf_items(trans, root, root_eb); | |
8344 | goto out; | |
8345 | } | |
8346 | ||
8347 | path = btrfs_alloc_path(); | |
8348 | if (!path) | |
8349 | return -ENOMEM; | |
8350 | ||
8351 | /* | |
8352 | * Walk down the tree. Missing extent blocks are filled in as | |
8353 | * we go. Metadata is accounted every time we read a new | |
8354 | * extent block. | |
8355 | * | |
8356 | * When we reach a leaf, we account for file extent items in it, | |
8357 | * walk back up the tree (adjusting slot pointers as we go) | |
8358 | * and restart the search process. | |
8359 | */ | |
8360 | extent_buffer_get(root_eb); /* For path */ | |
8361 | path->nodes[root_level] = root_eb; | |
8362 | path->slots[root_level] = 0; | |
8363 | path->locks[root_level] = 0; /* so release_path doesn't try to unlock */ | |
8364 | walk_down: | |
8365 | level = root_level; | |
8366 | while (level >= 0) { | |
8367 | if (path->nodes[level] == NULL) { | |
8368 | int parent_slot; | |
8369 | u64 child_gen; | |
8370 | u64 child_bytenr; | |
8371 | ||
8372 | /* We need to get child blockptr/gen from | |
8373 | * parent before we can read it. */ | |
8374 | eb = path->nodes[level + 1]; | |
8375 | parent_slot = path->slots[level + 1]; | |
8376 | child_bytenr = btrfs_node_blockptr(eb, parent_slot); | |
8377 | child_gen = btrfs_node_ptr_generation(eb, parent_slot); | |
8378 | ||
8379 | eb = read_tree_block(root, child_bytenr, child_gen); | |
8380 | if (IS_ERR(eb)) { | |
8381 | ret = PTR_ERR(eb); | |
8382 | goto out; | |
8383 | } else if (!extent_buffer_uptodate(eb)) { | |
8384 | free_extent_buffer(eb); | |
8385 | ret = -EIO; | |
8386 | goto out; | |
8387 | } | |
8388 | ||
8389 | path->nodes[level] = eb; | |
8390 | path->slots[level] = 0; | |
8391 | ||
8392 | btrfs_tree_read_lock(eb); | |
8393 | btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK); | |
8394 | path->locks[level] = BTRFS_READ_LOCK_BLOCKING; | |
8395 | ||
8396 | ret = record_one_subtree_extent(trans, root, child_bytenr, | |
8397 | root->nodesize); | |
8398 | if (ret) | |
8399 | goto out; | |
8400 | } | |
8401 | ||
8402 | if (level == 0) { | |
8403 | ret = account_leaf_items(trans, root, path->nodes[level]); | |
8404 | if (ret) | |
8405 | goto out; | |
8406 | ||
8407 | /* Nonzero return here means we completed our search */ | |
8408 | ret = adjust_slots_upwards(root, path, root_level); | |
8409 | if (ret) | |
8410 | break; | |
8411 | ||
8412 | /* Restart search with new slots */ | |
8413 | goto walk_down; | |
8414 | } | |
8415 | ||
8416 | level--; | |
8417 | } | |
8418 | ||
8419 | ret = 0; | |
8420 | out: | |
8421 | btrfs_free_path(path); | |
8422 | ||
8423 | return ret; | |
8424 | } | |
8425 | ||
8426 | /* | |
8427 | * helper to process tree block while walking down the tree. | |
8428 | * | |
8429 | * when wc->stage == UPDATE_BACKREF, this function updates | |
8430 | * back refs for pointers in the block. | |
8431 | * | |
8432 | * NOTE: return value 1 means we should stop walking down. | |
8433 | */ | |
8434 | static noinline int walk_down_proc(struct btrfs_trans_handle *trans, | |
8435 | struct btrfs_root *root, | |
8436 | struct btrfs_path *path, | |
8437 | struct walk_control *wc, int lookup_info) | |
8438 | { | |
8439 | int level = wc->level; | |
8440 | struct extent_buffer *eb = path->nodes[level]; | |
8441 | u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
8442 | int ret; | |
8443 | ||
8444 | if (wc->stage == UPDATE_BACKREF && | |
8445 | btrfs_header_owner(eb) != root->root_key.objectid) | |
8446 | return 1; | |
8447 | ||
8448 | /* | |
8449 | * when reference count of tree block is 1, it won't increase | |
8450 | * again. once full backref flag is set, we never clear it. | |
8451 | */ | |
8452 | if (lookup_info && | |
8453 | ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) || | |
8454 | (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) { | |
8455 | BUG_ON(!path->locks[level]); | |
8456 | ret = btrfs_lookup_extent_info(trans, root, | |
8457 | eb->start, level, 1, | |
8458 | &wc->refs[level], | |
8459 | &wc->flags[level]); | |
8460 | BUG_ON(ret == -ENOMEM); | |
8461 | if (ret) | |
8462 | return ret; | |
8463 | BUG_ON(wc->refs[level] == 0); | |
8464 | } | |
8465 | ||
8466 | if (wc->stage == DROP_REFERENCE) { | |
8467 | if (wc->refs[level] > 1) | |
8468 | return 1; | |
8469 | ||
8470 | if (path->locks[level] && !wc->keep_locks) { | |
8471 | btrfs_tree_unlock_rw(eb, path->locks[level]); | |
8472 | path->locks[level] = 0; | |
8473 | } | |
8474 | return 0; | |
8475 | } | |
8476 | ||
8477 | /* wc->stage == UPDATE_BACKREF */ | |
8478 | if (!(wc->flags[level] & flag)) { | |
8479 | BUG_ON(!path->locks[level]); | |
8480 | ret = btrfs_inc_ref(trans, root, eb, 1); | |
8481 | BUG_ON(ret); /* -ENOMEM */ | |
8482 | ret = btrfs_dec_ref(trans, root, eb, 0); | |
8483 | BUG_ON(ret); /* -ENOMEM */ | |
8484 | ret = btrfs_set_disk_extent_flags(trans, root, eb->start, | |
8485 | eb->len, flag, | |
8486 | btrfs_header_level(eb), 0); | |
8487 | BUG_ON(ret); /* -ENOMEM */ | |
8488 | wc->flags[level] |= flag; | |
8489 | } | |
8490 | ||
8491 | /* | |
8492 | * the block is shared by multiple trees, so it's not good to | |
8493 | * keep the tree lock | |
8494 | */ | |
8495 | if (path->locks[level] && level > 0) { | |
8496 | btrfs_tree_unlock_rw(eb, path->locks[level]); | |
8497 | path->locks[level] = 0; | |
8498 | } | |
8499 | return 0; | |
8500 | } | |
8501 | ||
8502 | /* | |
8503 | * helper to process tree block pointer. | |
8504 | * | |
8505 | * when wc->stage == DROP_REFERENCE, this function checks | |
8506 | * reference count of the block pointed to. if the block | |
8507 | * is shared and we need update back refs for the subtree | |
8508 | * rooted at the block, this function changes wc->stage to | |
8509 | * UPDATE_BACKREF. if the block is shared and there is no | |
8510 | * need to update back, this function drops the reference | |
8511 | * to the block. | |
8512 | * | |
8513 | * NOTE: return value 1 means we should stop walking down. | |
8514 | */ | |
8515 | static noinline int do_walk_down(struct btrfs_trans_handle *trans, | |
8516 | struct btrfs_root *root, | |
8517 | struct btrfs_path *path, | |
8518 | struct walk_control *wc, int *lookup_info) | |
8519 | { | |
8520 | u64 bytenr; | |
8521 | u64 generation; | |
8522 | u64 parent; | |
8523 | u32 blocksize; | |
8524 | struct btrfs_key key; | |
8525 | struct extent_buffer *next; | |
8526 | int level = wc->level; | |
8527 | int reada = 0; | |
8528 | int ret = 0; | |
8529 | bool need_account = false; | |
8530 | ||
8531 | generation = btrfs_node_ptr_generation(path->nodes[level], | |
8532 | path->slots[level]); | |
8533 | /* | |
8534 | * if the lower level block was created before the snapshot | |
8535 | * was created, we know there is no need to update back refs | |
8536 | * for the subtree | |
8537 | */ | |
8538 | if (wc->stage == UPDATE_BACKREF && | |
8539 | generation <= root->root_key.offset) { | |
8540 | *lookup_info = 1; | |
8541 | return 1; | |
8542 | } | |
8543 | ||
8544 | bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]); | |
8545 | blocksize = root->nodesize; | |
8546 | ||
8547 | next = btrfs_find_tree_block(root->fs_info, bytenr); | |
8548 | if (!next) { | |
8549 | next = btrfs_find_create_tree_block(root, bytenr); | |
8550 | if (!next) | |
8551 | return -ENOMEM; | |
8552 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, next, | |
8553 | level - 1); | |
8554 | reada = 1; | |
8555 | } | |
8556 | btrfs_tree_lock(next); | |
8557 | btrfs_set_lock_blocking(next); | |
8558 | ||
8559 | ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1, | |
8560 | &wc->refs[level - 1], | |
8561 | &wc->flags[level - 1]); | |
8562 | if (ret < 0) { | |
8563 | btrfs_tree_unlock(next); | |
8564 | return ret; | |
8565 | } | |
8566 | ||
8567 | if (unlikely(wc->refs[level - 1] == 0)) { | |
8568 | btrfs_err(root->fs_info, "Missing references."); | |
8569 | BUG(); | |
8570 | } | |
8571 | *lookup_info = 0; | |
8572 | ||
8573 | if (wc->stage == DROP_REFERENCE) { | |
8574 | if (wc->refs[level - 1] > 1) { | |
8575 | need_account = true; | |
8576 | if (level == 1 && | |
8577 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8578 | goto skip; | |
8579 | ||
8580 | if (!wc->update_ref || | |
8581 | generation <= root->root_key.offset) | |
8582 | goto skip; | |
8583 | ||
8584 | btrfs_node_key_to_cpu(path->nodes[level], &key, | |
8585 | path->slots[level]); | |
8586 | ret = btrfs_comp_cpu_keys(&key, &wc->update_progress); | |
8587 | if (ret < 0) | |
8588 | goto skip; | |
8589 | ||
8590 | wc->stage = UPDATE_BACKREF; | |
8591 | wc->shared_level = level - 1; | |
8592 | } | |
8593 | } else { | |
8594 | if (level == 1 && | |
8595 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8596 | goto skip; | |
8597 | } | |
8598 | ||
8599 | if (!btrfs_buffer_uptodate(next, generation, 0)) { | |
8600 | btrfs_tree_unlock(next); | |
8601 | free_extent_buffer(next); | |
8602 | next = NULL; | |
8603 | *lookup_info = 1; | |
8604 | } | |
8605 | ||
8606 | if (!next) { | |
8607 | if (reada && level == 1) | |
8608 | reada_walk_down(trans, root, wc, path); | |
8609 | next = read_tree_block(root, bytenr, generation); | |
8610 | if (IS_ERR(next)) { | |
8611 | return PTR_ERR(next); | |
8612 | } else if (!extent_buffer_uptodate(next)) { | |
8613 | free_extent_buffer(next); | |
8614 | return -EIO; | |
8615 | } | |
8616 | btrfs_tree_lock(next); | |
8617 | btrfs_set_lock_blocking(next); | |
8618 | } | |
8619 | ||
8620 | level--; | |
8621 | BUG_ON(level != btrfs_header_level(next)); | |
8622 | path->nodes[level] = next; | |
8623 | path->slots[level] = 0; | |
8624 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; | |
8625 | wc->level = level; | |
8626 | if (wc->level == 1) | |
8627 | wc->reada_slot = 0; | |
8628 | return 0; | |
8629 | skip: | |
8630 | wc->refs[level - 1] = 0; | |
8631 | wc->flags[level - 1] = 0; | |
8632 | if (wc->stage == DROP_REFERENCE) { | |
8633 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
8634 | parent = path->nodes[level]->start; | |
8635 | } else { | |
8636 | BUG_ON(root->root_key.objectid != | |
8637 | btrfs_header_owner(path->nodes[level])); | |
8638 | parent = 0; | |
8639 | } | |
8640 | ||
8641 | if (need_account) { | |
8642 | ret = account_shared_subtree(trans, root, next, | |
8643 | generation, level - 1); | |
8644 | if (ret) { | |
8645 | btrfs_err_rl(root->fs_info, | |
8646 | "Error " | |
8647 | "%d accounting shared subtree. Quota " | |
8648 | "is out of sync, rescan required.", | |
8649 | ret); | |
8650 | } | |
8651 | } | |
8652 | ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent, | |
8653 | root->root_key.objectid, level - 1, 0); | |
8654 | BUG_ON(ret); /* -ENOMEM */ | |
8655 | } | |
8656 | btrfs_tree_unlock(next); | |
8657 | free_extent_buffer(next); | |
8658 | *lookup_info = 1; | |
8659 | return 1; | |
8660 | } | |
8661 | ||
8662 | /* | |
8663 | * helper to process tree block while walking up the tree. | |
8664 | * | |
8665 | * when wc->stage == DROP_REFERENCE, this function drops | |
8666 | * reference count on the block. | |
8667 | * | |
8668 | * when wc->stage == UPDATE_BACKREF, this function changes | |
8669 | * wc->stage back to DROP_REFERENCE if we changed wc->stage | |
8670 | * to UPDATE_BACKREF previously while processing the block. | |
8671 | * | |
8672 | * NOTE: return value 1 means we should stop walking up. | |
8673 | */ | |
8674 | static noinline int walk_up_proc(struct btrfs_trans_handle *trans, | |
8675 | struct btrfs_root *root, | |
8676 | struct btrfs_path *path, | |
8677 | struct walk_control *wc) | |
8678 | { | |
8679 | int ret; | |
8680 | int level = wc->level; | |
8681 | struct extent_buffer *eb = path->nodes[level]; | |
8682 | u64 parent = 0; | |
8683 | ||
8684 | if (wc->stage == UPDATE_BACKREF) { | |
8685 | BUG_ON(wc->shared_level < level); | |
8686 | if (level < wc->shared_level) | |
8687 | goto out; | |
8688 | ||
8689 | ret = find_next_key(path, level + 1, &wc->update_progress); | |
8690 | if (ret > 0) | |
8691 | wc->update_ref = 0; | |
8692 | ||
8693 | wc->stage = DROP_REFERENCE; | |
8694 | wc->shared_level = -1; | |
8695 | path->slots[level] = 0; | |
8696 | ||
8697 | /* | |
8698 | * check reference count again if the block isn't locked. | |
8699 | * we should start walking down the tree again if reference | |
8700 | * count is one. | |
8701 | */ | |
8702 | if (!path->locks[level]) { | |
8703 | BUG_ON(level == 0); | |
8704 | btrfs_tree_lock(eb); | |
8705 | btrfs_set_lock_blocking(eb); | |
8706 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; | |
8707 | ||
8708 | ret = btrfs_lookup_extent_info(trans, root, | |
8709 | eb->start, level, 1, | |
8710 | &wc->refs[level], | |
8711 | &wc->flags[level]); | |
8712 | if (ret < 0) { | |
8713 | btrfs_tree_unlock_rw(eb, path->locks[level]); | |
8714 | path->locks[level] = 0; | |
8715 | return ret; | |
8716 | } | |
8717 | BUG_ON(wc->refs[level] == 0); | |
8718 | if (wc->refs[level] == 1) { | |
8719 | btrfs_tree_unlock_rw(eb, path->locks[level]); | |
8720 | path->locks[level] = 0; | |
8721 | return 1; | |
8722 | } | |
8723 | } | |
8724 | } | |
8725 | ||
8726 | /* wc->stage == DROP_REFERENCE */ | |
8727 | BUG_ON(wc->refs[level] > 1 && !path->locks[level]); | |
8728 | ||
8729 | if (wc->refs[level] == 1) { | |
8730 | if (level == 0) { | |
8731 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
8732 | ret = btrfs_dec_ref(trans, root, eb, 1); | |
8733 | else | |
8734 | ret = btrfs_dec_ref(trans, root, eb, 0); | |
8735 | BUG_ON(ret); /* -ENOMEM */ | |
8736 | ret = account_leaf_items(trans, root, eb); | |
8737 | if (ret) { | |
8738 | btrfs_err_rl(root->fs_info, | |
8739 | "error " | |
8740 | "%d accounting leaf items. Quota " | |
8741 | "is out of sync, rescan required.", | |
8742 | ret); | |
8743 | } | |
8744 | } | |
8745 | /* make block locked assertion in clean_tree_block happy */ | |
8746 | if (!path->locks[level] && | |
8747 | btrfs_header_generation(eb) == trans->transid) { | |
8748 | btrfs_tree_lock(eb); | |
8749 | btrfs_set_lock_blocking(eb); | |
8750 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; | |
8751 | } | |
8752 | clean_tree_block(trans, root->fs_info, eb); | |
8753 | } | |
8754 | ||
8755 | if (eb == root->node) { | |
8756 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
8757 | parent = eb->start; | |
8758 | else | |
8759 | BUG_ON(root->root_key.objectid != | |
8760 | btrfs_header_owner(eb)); | |
8761 | } else { | |
8762 | if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
8763 | parent = path->nodes[level + 1]->start; | |
8764 | else | |
8765 | BUG_ON(root->root_key.objectid != | |
8766 | btrfs_header_owner(path->nodes[level + 1])); | |
8767 | } | |
8768 | ||
8769 | btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1); | |
8770 | out: | |
8771 | wc->refs[level] = 0; | |
8772 | wc->flags[level] = 0; | |
8773 | return 0; | |
8774 | } | |
8775 | ||
8776 | static noinline int walk_down_tree(struct btrfs_trans_handle *trans, | |
8777 | struct btrfs_root *root, | |
8778 | struct btrfs_path *path, | |
8779 | struct walk_control *wc) | |
8780 | { | |
8781 | int level = wc->level; | |
8782 | int lookup_info = 1; | |
8783 | int ret; | |
8784 | ||
8785 | while (level >= 0) { | |
8786 | ret = walk_down_proc(trans, root, path, wc, lookup_info); | |
8787 | if (ret > 0) | |
8788 | break; | |
8789 | ||
8790 | if (level == 0) | |
8791 | break; | |
8792 | ||
8793 | if (path->slots[level] >= | |
8794 | btrfs_header_nritems(path->nodes[level])) | |
8795 | break; | |
8796 | ||
8797 | ret = do_walk_down(trans, root, path, wc, &lookup_info); | |
8798 | if (ret > 0) { | |
8799 | path->slots[level]++; | |
8800 | continue; | |
8801 | } else if (ret < 0) | |
8802 | return ret; | |
8803 | level = wc->level; | |
8804 | } | |
8805 | return 0; | |
8806 | } | |
8807 | ||
8808 | static noinline int walk_up_tree(struct btrfs_trans_handle *trans, | |
8809 | struct btrfs_root *root, | |
8810 | struct btrfs_path *path, | |
8811 | struct walk_control *wc, int max_level) | |
8812 | { | |
8813 | int level = wc->level; | |
8814 | int ret; | |
8815 | ||
8816 | path->slots[level] = btrfs_header_nritems(path->nodes[level]); | |
8817 | while (level < max_level && path->nodes[level]) { | |
8818 | wc->level = level; | |
8819 | if (path->slots[level] + 1 < | |
8820 | btrfs_header_nritems(path->nodes[level])) { | |
8821 | path->slots[level]++; | |
8822 | return 0; | |
8823 | } else { | |
8824 | ret = walk_up_proc(trans, root, path, wc); | |
8825 | if (ret > 0) | |
8826 | return 0; | |
8827 | ||
8828 | if (path->locks[level]) { | |
8829 | btrfs_tree_unlock_rw(path->nodes[level], | |
8830 | path->locks[level]); | |
8831 | path->locks[level] = 0; | |
8832 | } | |
8833 | free_extent_buffer(path->nodes[level]); | |
8834 | path->nodes[level] = NULL; | |
8835 | level++; | |
8836 | } | |
8837 | } | |
8838 | return 1; | |
8839 | } | |
8840 | ||
8841 | /* | |
8842 | * drop a subvolume tree. | |
8843 | * | |
8844 | * this function traverses the tree freeing any blocks that only | |
8845 | * referenced by the tree. | |
8846 | * | |
8847 | * when a shared tree block is found. this function decreases its | |
8848 | * reference count by one. if update_ref is true, this function | |
8849 | * also make sure backrefs for the shared block and all lower level | |
8850 | * blocks are properly updated. | |
8851 | * | |
8852 | * If called with for_reloc == 0, may exit early with -EAGAIN | |
8853 | */ | |
8854 | int btrfs_drop_snapshot(struct btrfs_root *root, | |
8855 | struct btrfs_block_rsv *block_rsv, int update_ref, | |
8856 | int for_reloc) | |
8857 | { | |
8858 | struct btrfs_path *path; | |
8859 | struct btrfs_trans_handle *trans; | |
8860 | struct btrfs_root *tree_root = root->fs_info->tree_root; | |
8861 | struct btrfs_root_item *root_item = &root->root_item; | |
8862 | struct walk_control *wc; | |
8863 | struct btrfs_key key; | |
8864 | int err = 0; | |
8865 | int ret; | |
8866 | int level; | |
8867 | bool root_dropped = false; | |
8868 | ||
8869 | btrfs_debug(root->fs_info, "Drop subvolume %llu", root->objectid); | |
8870 | ||
8871 | path = btrfs_alloc_path(); | |
8872 | if (!path) { | |
8873 | err = -ENOMEM; | |
8874 | goto out; | |
8875 | } | |
8876 | ||
8877 | wc = kzalloc(sizeof(*wc), GFP_NOFS); | |
8878 | if (!wc) { | |
8879 | btrfs_free_path(path); | |
8880 | err = -ENOMEM; | |
8881 | goto out; | |
8882 | } | |
8883 | ||
8884 | trans = btrfs_start_transaction(tree_root, 0); | |
8885 | if (IS_ERR(trans)) { | |
8886 | err = PTR_ERR(trans); | |
8887 | goto out_free; | |
8888 | } | |
8889 | ||
8890 | if (block_rsv) | |
8891 | trans->block_rsv = block_rsv; | |
8892 | ||
8893 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { | |
8894 | level = btrfs_header_level(root->node); | |
8895 | path->nodes[level] = btrfs_lock_root_node(root); | |
8896 | btrfs_set_lock_blocking(path->nodes[level]); | |
8897 | path->slots[level] = 0; | |
8898 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; | |
8899 | memset(&wc->update_progress, 0, | |
8900 | sizeof(wc->update_progress)); | |
8901 | } else { | |
8902 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); | |
8903 | memcpy(&wc->update_progress, &key, | |
8904 | sizeof(wc->update_progress)); | |
8905 | ||
8906 | level = root_item->drop_level; | |
8907 | BUG_ON(level == 0); | |
8908 | path->lowest_level = level; | |
8909 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
8910 | path->lowest_level = 0; | |
8911 | if (ret < 0) { | |
8912 | err = ret; | |
8913 | goto out_end_trans; | |
8914 | } | |
8915 | WARN_ON(ret > 0); | |
8916 | ||
8917 | /* | |
8918 | * unlock our path, this is safe because only this | |
8919 | * function is allowed to delete this snapshot | |
8920 | */ | |
8921 | btrfs_unlock_up_safe(path, 0); | |
8922 | ||
8923 | level = btrfs_header_level(root->node); | |
8924 | while (1) { | |
8925 | btrfs_tree_lock(path->nodes[level]); | |
8926 | btrfs_set_lock_blocking(path->nodes[level]); | |
8927 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; | |
8928 | ||
8929 | ret = btrfs_lookup_extent_info(trans, root, | |
8930 | path->nodes[level]->start, | |
8931 | level, 1, &wc->refs[level], | |
8932 | &wc->flags[level]); | |
8933 | if (ret < 0) { | |
8934 | err = ret; | |
8935 | goto out_end_trans; | |
8936 | } | |
8937 | BUG_ON(wc->refs[level] == 0); | |
8938 | ||
8939 | if (level == root_item->drop_level) | |
8940 | break; | |
8941 | ||
8942 | btrfs_tree_unlock(path->nodes[level]); | |
8943 | path->locks[level] = 0; | |
8944 | WARN_ON(wc->refs[level] != 1); | |
8945 | level--; | |
8946 | } | |
8947 | } | |
8948 | ||
8949 | wc->level = level; | |
8950 | wc->shared_level = -1; | |
8951 | wc->stage = DROP_REFERENCE; | |
8952 | wc->update_ref = update_ref; | |
8953 | wc->keep_locks = 0; | |
8954 | wc->for_reloc = for_reloc; | |
8955 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root); | |
8956 | ||
8957 | while (1) { | |
8958 | ||
8959 | ret = walk_down_tree(trans, root, path, wc); | |
8960 | if (ret < 0) { | |
8961 | err = ret; | |
8962 | break; | |
8963 | } | |
8964 | ||
8965 | ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL); | |
8966 | if (ret < 0) { | |
8967 | err = ret; | |
8968 | break; | |
8969 | } | |
8970 | ||
8971 | if (ret > 0) { | |
8972 | BUG_ON(wc->stage != DROP_REFERENCE); | |
8973 | break; | |
8974 | } | |
8975 | ||
8976 | if (wc->stage == DROP_REFERENCE) { | |
8977 | level = wc->level; | |
8978 | btrfs_node_key(path->nodes[level], | |
8979 | &root_item->drop_progress, | |
8980 | path->slots[level]); | |
8981 | root_item->drop_level = level; | |
8982 | } | |
8983 | ||
8984 | BUG_ON(wc->level == 0); | |
8985 | if (btrfs_should_end_transaction(trans, tree_root) || | |
8986 | (!for_reloc && btrfs_need_cleaner_sleep(root))) { | |
8987 | ret = btrfs_update_root(trans, tree_root, | |
8988 | &root->root_key, | |
8989 | root_item); | |
8990 | if (ret) { | |
8991 | btrfs_abort_transaction(trans, tree_root, ret); | |
8992 | err = ret; | |
8993 | goto out_end_trans; | |
8994 | } | |
8995 | ||
8996 | btrfs_end_transaction_throttle(trans, tree_root); | |
8997 | if (!for_reloc && btrfs_need_cleaner_sleep(root)) { | |
8998 | pr_debug("BTRFS: drop snapshot early exit\n"); | |
8999 | err = -EAGAIN; | |
9000 | goto out_free; | |
9001 | } | |
9002 | ||
9003 | trans = btrfs_start_transaction(tree_root, 0); | |
9004 | if (IS_ERR(trans)) { | |
9005 | err = PTR_ERR(trans); | |
9006 | goto out_free; | |
9007 | } | |
9008 | if (block_rsv) | |
9009 | trans->block_rsv = block_rsv; | |
9010 | } | |
9011 | } | |
9012 | btrfs_release_path(path); | |
9013 | if (err) | |
9014 | goto out_end_trans; | |
9015 | ||
9016 | ret = btrfs_del_root(trans, tree_root, &root->root_key); | |
9017 | if (ret) { | |
9018 | btrfs_abort_transaction(trans, tree_root, ret); | |
9019 | goto out_end_trans; | |
9020 | } | |
9021 | ||
9022 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { | |
9023 | ret = btrfs_find_root(tree_root, &root->root_key, path, | |
9024 | NULL, NULL); | |
9025 | if (ret < 0) { | |
9026 | btrfs_abort_transaction(trans, tree_root, ret); | |
9027 | err = ret; | |
9028 | goto out_end_trans; | |
9029 | } else if (ret > 0) { | |
9030 | /* if we fail to delete the orphan item this time | |
9031 | * around, it'll get picked up the next time. | |
9032 | * | |
9033 | * The most common failure here is just -ENOENT. | |
9034 | */ | |
9035 | btrfs_del_orphan_item(trans, tree_root, | |
9036 | root->root_key.objectid); | |
9037 | } | |
9038 | } | |
9039 | ||
9040 | if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) { | |
9041 | btrfs_add_dropped_root(trans, root); | |
9042 | } else { | |
9043 | free_extent_buffer(root->node); | |
9044 | free_extent_buffer(root->commit_root); | |
9045 | btrfs_put_fs_root(root); | |
9046 | } | |
9047 | root_dropped = true; | |
9048 | out_end_trans: | |
9049 | btrfs_end_transaction_throttle(trans, tree_root); | |
9050 | out_free: | |
9051 | kfree(wc); | |
9052 | btrfs_free_path(path); | |
9053 | out: | |
9054 | /* | |
9055 | * So if we need to stop dropping the snapshot for whatever reason we | |
9056 | * need to make sure to add it back to the dead root list so that we | |
9057 | * keep trying to do the work later. This also cleans up roots if we | |
9058 | * don't have it in the radix (like when we recover after a power fail | |
9059 | * or unmount) so we don't leak memory. | |
9060 | */ | |
9061 | if (!for_reloc && root_dropped == false) | |
9062 | btrfs_add_dead_root(root); | |
9063 | if (err && err != -EAGAIN) | |
9064 | btrfs_std_error(root->fs_info, err, NULL); | |
9065 | return err; | |
9066 | } | |
9067 | ||
9068 | /* | |
9069 | * drop subtree rooted at tree block 'node'. | |
9070 | * | |
9071 | * NOTE: this function will unlock and release tree block 'node' | |
9072 | * only used by relocation code | |
9073 | */ | |
9074 | int btrfs_drop_subtree(struct btrfs_trans_handle *trans, | |
9075 | struct btrfs_root *root, | |
9076 | struct extent_buffer *node, | |
9077 | struct extent_buffer *parent) | |
9078 | { | |
9079 | struct btrfs_path *path; | |
9080 | struct walk_control *wc; | |
9081 | int level; | |
9082 | int parent_level; | |
9083 | int ret = 0; | |
9084 | int wret; | |
9085 | ||
9086 | BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); | |
9087 | ||
9088 | path = btrfs_alloc_path(); | |
9089 | if (!path) | |
9090 | return -ENOMEM; | |
9091 | ||
9092 | wc = kzalloc(sizeof(*wc), GFP_NOFS); | |
9093 | if (!wc) { | |
9094 | btrfs_free_path(path); | |
9095 | return -ENOMEM; | |
9096 | } | |
9097 | ||
9098 | btrfs_assert_tree_locked(parent); | |
9099 | parent_level = btrfs_header_level(parent); | |
9100 | extent_buffer_get(parent); | |
9101 | path->nodes[parent_level] = parent; | |
9102 | path->slots[parent_level] = btrfs_header_nritems(parent); | |
9103 | ||
9104 | btrfs_assert_tree_locked(node); | |
9105 | level = btrfs_header_level(node); | |
9106 | path->nodes[level] = node; | |
9107 | path->slots[level] = 0; | |
9108 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; | |
9109 | ||
9110 | wc->refs[parent_level] = 1; | |
9111 | wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
9112 | wc->level = level; | |
9113 | wc->shared_level = -1; | |
9114 | wc->stage = DROP_REFERENCE; | |
9115 | wc->update_ref = 0; | |
9116 | wc->keep_locks = 1; | |
9117 | wc->for_reloc = 1; | |
9118 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root); | |
9119 | ||
9120 | while (1) { | |
9121 | wret = walk_down_tree(trans, root, path, wc); | |
9122 | if (wret < 0) { | |
9123 | ret = wret; | |
9124 | break; | |
9125 | } | |
9126 | ||
9127 | wret = walk_up_tree(trans, root, path, wc, parent_level); | |
9128 | if (wret < 0) | |
9129 | ret = wret; | |
9130 | if (wret != 0) | |
9131 | break; | |
9132 | } | |
9133 | ||
9134 | kfree(wc); | |
9135 | btrfs_free_path(path); | |
9136 | return ret; | |
9137 | } | |
9138 | ||
9139 | static u64 update_block_group_flags(struct btrfs_root *root, u64 flags) | |
9140 | { | |
9141 | u64 num_devices; | |
9142 | u64 stripped; | |
9143 | ||
9144 | /* | |
9145 | * if restripe for this chunk_type is on pick target profile and | |
9146 | * return, otherwise do the usual balance | |
9147 | */ | |
9148 | stripped = get_restripe_target(root->fs_info, flags); | |
9149 | if (stripped) | |
9150 | return extended_to_chunk(stripped); | |
9151 | ||
9152 | num_devices = root->fs_info->fs_devices->rw_devices; | |
9153 | ||
9154 | stripped = BTRFS_BLOCK_GROUP_RAID0 | | |
9155 | BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 | | |
9156 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10; | |
9157 | ||
9158 | if (num_devices == 1) { | |
9159 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
9160 | stripped = flags & ~stripped; | |
9161 | ||
9162 | /* turn raid0 into single device chunks */ | |
9163 | if (flags & BTRFS_BLOCK_GROUP_RAID0) | |
9164 | return stripped; | |
9165 | ||
9166 | /* turn mirroring into duplication */ | |
9167 | if (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
9168 | BTRFS_BLOCK_GROUP_RAID10)) | |
9169 | return stripped | BTRFS_BLOCK_GROUP_DUP; | |
9170 | } else { | |
9171 | /* they already had raid on here, just return */ | |
9172 | if (flags & stripped) | |
9173 | return flags; | |
9174 | ||
9175 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
9176 | stripped = flags & ~stripped; | |
9177 | ||
9178 | /* switch duplicated blocks with raid1 */ | |
9179 | if (flags & BTRFS_BLOCK_GROUP_DUP) | |
9180 | return stripped | BTRFS_BLOCK_GROUP_RAID1; | |
9181 | ||
9182 | /* this is drive concat, leave it alone */ | |
9183 | } | |
9184 | ||
9185 | return flags; | |
9186 | } | |
9187 | ||
9188 | static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force) | |
9189 | { | |
9190 | struct btrfs_space_info *sinfo = cache->space_info; | |
9191 | u64 num_bytes; | |
9192 | u64 min_allocable_bytes; | |
9193 | int ret = -ENOSPC; | |
9194 | ||
9195 | /* | |
9196 | * We need some metadata space and system metadata space for | |
9197 | * allocating chunks in some corner cases until we force to set | |
9198 | * it to be readonly. | |
9199 | */ | |
9200 | if ((sinfo->flags & | |
9201 | (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) && | |
9202 | !force) | |
9203 | min_allocable_bytes = SZ_1M; | |
9204 | else | |
9205 | min_allocable_bytes = 0; | |
9206 | ||
9207 | spin_lock(&sinfo->lock); | |
9208 | spin_lock(&cache->lock); | |
9209 | ||
9210 | if (cache->ro) { | |
9211 | cache->ro++; | |
9212 | ret = 0; | |
9213 | goto out; | |
9214 | } | |
9215 | ||
9216 | num_bytes = cache->key.offset - cache->reserved - cache->pinned - | |
9217 | cache->bytes_super - btrfs_block_group_used(&cache->item); | |
9218 | ||
9219 | if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned + | |
9220 | sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes + | |
9221 | min_allocable_bytes <= sinfo->total_bytes) { | |
9222 | sinfo->bytes_readonly += num_bytes; | |
9223 | cache->ro++; | |
9224 | list_add_tail(&cache->ro_list, &sinfo->ro_bgs); | |
9225 | ret = 0; | |
9226 | } | |
9227 | out: | |
9228 | spin_unlock(&cache->lock); | |
9229 | spin_unlock(&sinfo->lock); | |
9230 | return ret; | |
9231 | } | |
9232 | ||
9233 | int btrfs_inc_block_group_ro(struct btrfs_root *root, | |
9234 | struct btrfs_block_group_cache *cache) | |
9235 | ||
9236 | { | |
9237 | struct btrfs_trans_handle *trans; | |
9238 | u64 alloc_flags; | |
9239 | int ret; | |
9240 | ||
9241 | again: | |
9242 | trans = btrfs_join_transaction(root); | |
9243 | if (IS_ERR(trans)) | |
9244 | return PTR_ERR(trans); | |
9245 | ||
9246 | /* | |
9247 | * we're not allowed to set block groups readonly after the dirty | |
9248 | * block groups cache has started writing. If it already started, | |
9249 | * back off and let this transaction commit | |
9250 | */ | |
9251 | mutex_lock(&root->fs_info->ro_block_group_mutex); | |
9252 | if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) { | |
9253 | u64 transid = trans->transid; | |
9254 | ||
9255 | mutex_unlock(&root->fs_info->ro_block_group_mutex); | |
9256 | btrfs_end_transaction(trans, root); | |
9257 | ||
9258 | ret = btrfs_wait_for_commit(root, transid); | |
9259 | if (ret) | |
9260 | return ret; | |
9261 | goto again; | |
9262 | } | |
9263 | ||
9264 | /* | |
9265 | * if we are changing raid levels, try to allocate a corresponding | |
9266 | * block group with the new raid level. | |
9267 | */ | |
9268 | alloc_flags = update_block_group_flags(root, cache->flags); | |
9269 | if (alloc_flags != cache->flags) { | |
9270 | ret = do_chunk_alloc(trans, root, alloc_flags, | |
9271 | CHUNK_ALLOC_FORCE); | |
9272 | /* | |
9273 | * ENOSPC is allowed here, we may have enough space | |
9274 | * already allocated at the new raid level to | |
9275 | * carry on | |
9276 | */ | |
9277 | if (ret == -ENOSPC) | |
9278 | ret = 0; | |
9279 | if (ret < 0) | |
9280 | goto out; | |
9281 | } | |
9282 | ||
9283 | ret = inc_block_group_ro(cache, 0); | |
9284 | if (!ret) | |
9285 | goto out; | |
9286 | alloc_flags = get_alloc_profile(root, cache->space_info->flags); | |
9287 | ret = do_chunk_alloc(trans, root, alloc_flags, | |
9288 | CHUNK_ALLOC_FORCE); | |
9289 | if (ret < 0) | |
9290 | goto out; | |
9291 | ret = inc_block_group_ro(cache, 0); | |
9292 | out: | |
9293 | if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) { | |
9294 | alloc_flags = update_block_group_flags(root, cache->flags); | |
9295 | lock_chunks(root->fs_info->chunk_root); | |
9296 | check_system_chunk(trans, root, alloc_flags); | |
9297 | unlock_chunks(root->fs_info->chunk_root); | |
9298 | } | |
9299 | mutex_unlock(&root->fs_info->ro_block_group_mutex); | |
9300 | ||
9301 | btrfs_end_transaction(trans, root); | |
9302 | return ret; | |
9303 | } | |
9304 | ||
9305 | int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans, | |
9306 | struct btrfs_root *root, u64 type) | |
9307 | { | |
9308 | u64 alloc_flags = get_alloc_profile(root, type); | |
9309 | return do_chunk_alloc(trans, root, alloc_flags, | |
9310 | CHUNK_ALLOC_FORCE); | |
9311 | } | |
9312 | ||
9313 | /* | |
9314 | * helper to account the unused space of all the readonly block group in the | |
9315 | * space_info. takes mirrors into account. | |
9316 | */ | |
9317 | u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo) | |
9318 | { | |
9319 | struct btrfs_block_group_cache *block_group; | |
9320 | u64 free_bytes = 0; | |
9321 | int factor; | |
9322 | ||
9323 | /* It's df, we don't care if it's racey */ | |
9324 | if (list_empty(&sinfo->ro_bgs)) | |
9325 | return 0; | |
9326 | ||
9327 | spin_lock(&sinfo->lock); | |
9328 | list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) { | |
9329 | spin_lock(&block_group->lock); | |
9330 | ||
9331 | if (!block_group->ro) { | |
9332 | spin_unlock(&block_group->lock); | |
9333 | continue; | |
9334 | } | |
9335 | ||
9336 | if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
9337 | BTRFS_BLOCK_GROUP_RAID10 | | |
9338 | BTRFS_BLOCK_GROUP_DUP)) | |
9339 | factor = 2; | |
9340 | else | |
9341 | factor = 1; | |
9342 | ||
9343 | free_bytes += (block_group->key.offset - | |
9344 | btrfs_block_group_used(&block_group->item)) * | |
9345 | factor; | |
9346 | ||
9347 | spin_unlock(&block_group->lock); | |
9348 | } | |
9349 | spin_unlock(&sinfo->lock); | |
9350 | ||
9351 | return free_bytes; | |
9352 | } | |
9353 | ||
9354 | void btrfs_dec_block_group_ro(struct btrfs_root *root, | |
9355 | struct btrfs_block_group_cache *cache) | |
9356 | { | |
9357 | struct btrfs_space_info *sinfo = cache->space_info; | |
9358 | u64 num_bytes; | |
9359 | ||
9360 | BUG_ON(!cache->ro); | |
9361 | ||
9362 | spin_lock(&sinfo->lock); | |
9363 | spin_lock(&cache->lock); | |
9364 | if (!--cache->ro) { | |
9365 | num_bytes = cache->key.offset - cache->reserved - | |
9366 | cache->pinned - cache->bytes_super - | |
9367 | btrfs_block_group_used(&cache->item); | |
9368 | sinfo->bytes_readonly -= num_bytes; | |
9369 | list_del_init(&cache->ro_list); | |
9370 | } | |
9371 | spin_unlock(&cache->lock); | |
9372 | spin_unlock(&sinfo->lock); | |
9373 | } | |
9374 | ||
9375 | /* | |
9376 | * checks to see if its even possible to relocate this block group. | |
9377 | * | |
9378 | * @return - -1 if it's not a good idea to relocate this block group, 0 if its | |
9379 | * ok to go ahead and try. | |
9380 | */ | |
9381 | int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr) | |
9382 | { | |
9383 | struct btrfs_block_group_cache *block_group; | |
9384 | struct btrfs_space_info *space_info; | |
9385 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; | |
9386 | struct btrfs_device *device; | |
9387 | struct btrfs_trans_handle *trans; | |
9388 | u64 min_free; | |
9389 | u64 dev_min = 1; | |
9390 | u64 dev_nr = 0; | |
9391 | u64 target; | |
9392 | int debug; | |
9393 | int index; | |
9394 | int full = 0; | |
9395 | int ret = 0; | |
9396 | ||
9397 | debug = btrfs_test_opt(root, ENOSPC_DEBUG); | |
9398 | ||
9399 | block_group = btrfs_lookup_block_group(root->fs_info, bytenr); | |
9400 | ||
9401 | /* odd, couldn't find the block group, leave it alone */ | |
9402 | if (!block_group) { | |
9403 | if (debug) | |
9404 | btrfs_warn(root->fs_info, | |
9405 | "can't find block group for bytenr %llu", | |
9406 | bytenr); | |
9407 | return -1; | |
9408 | } | |
9409 | ||
9410 | min_free = btrfs_block_group_used(&block_group->item); | |
9411 | ||
9412 | /* no bytes used, we're good */ | |
9413 | if (!min_free) | |
9414 | goto out; | |
9415 | ||
9416 | space_info = block_group->space_info; | |
9417 | spin_lock(&space_info->lock); | |
9418 | ||
9419 | full = space_info->full; | |
9420 | ||
9421 | /* | |
9422 | * if this is the last block group we have in this space, we can't | |
9423 | * relocate it unless we're able to allocate a new chunk below. | |
9424 | * | |
9425 | * Otherwise, we need to make sure we have room in the space to handle | |
9426 | * all of the extents from this block group. If we can, we're good | |
9427 | */ | |
9428 | if ((space_info->total_bytes != block_group->key.offset) && | |
9429 | (space_info->bytes_used + space_info->bytes_reserved + | |
9430 | space_info->bytes_pinned + space_info->bytes_readonly + | |
9431 | min_free < space_info->total_bytes)) { | |
9432 | spin_unlock(&space_info->lock); | |
9433 | goto out; | |
9434 | } | |
9435 | spin_unlock(&space_info->lock); | |
9436 | ||
9437 | /* | |
9438 | * ok we don't have enough space, but maybe we have free space on our | |
9439 | * devices to allocate new chunks for relocation, so loop through our | |
9440 | * alloc devices and guess if we have enough space. if this block | |
9441 | * group is going to be restriped, run checks against the target | |
9442 | * profile instead of the current one. | |
9443 | */ | |
9444 | ret = -1; | |
9445 | ||
9446 | /* | |
9447 | * index: | |
9448 | * 0: raid10 | |
9449 | * 1: raid1 | |
9450 | * 2: dup | |
9451 | * 3: raid0 | |
9452 | * 4: single | |
9453 | */ | |
9454 | target = get_restripe_target(root->fs_info, block_group->flags); | |
9455 | if (target) { | |
9456 | index = __get_raid_index(extended_to_chunk(target)); | |
9457 | } else { | |
9458 | /* | |
9459 | * this is just a balance, so if we were marked as full | |
9460 | * we know there is no space for a new chunk | |
9461 | */ | |
9462 | if (full) { | |
9463 | if (debug) | |
9464 | btrfs_warn(root->fs_info, | |
9465 | "no space to alloc new chunk for block group %llu", | |
9466 | block_group->key.objectid); | |
9467 | goto out; | |
9468 | } | |
9469 | ||
9470 | index = get_block_group_index(block_group); | |
9471 | } | |
9472 | ||
9473 | if (index == BTRFS_RAID_RAID10) { | |
9474 | dev_min = 4; | |
9475 | /* Divide by 2 */ | |
9476 | min_free >>= 1; | |
9477 | } else if (index == BTRFS_RAID_RAID1) { | |
9478 | dev_min = 2; | |
9479 | } else if (index == BTRFS_RAID_DUP) { | |
9480 | /* Multiply by 2 */ | |
9481 | min_free <<= 1; | |
9482 | } else if (index == BTRFS_RAID_RAID0) { | |
9483 | dev_min = fs_devices->rw_devices; | |
9484 | min_free = div64_u64(min_free, dev_min); | |
9485 | } | |
9486 | ||
9487 | /* We need to do this so that we can look at pending chunks */ | |
9488 | trans = btrfs_join_transaction(root); | |
9489 | if (IS_ERR(trans)) { | |
9490 | ret = PTR_ERR(trans); | |
9491 | goto out; | |
9492 | } | |
9493 | ||
9494 | mutex_lock(&root->fs_info->chunk_mutex); | |
9495 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { | |
9496 | u64 dev_offset; | |
9497 | ||
9498 | /* | |
9499 | * check to make sure we can actually find a chunk with enough | |
9500 | * space to fit our block group in. | |
9501 | */ | |
9502 | if (device->total_bytes > device->bytes_used + min_free && | |
9503 | !device->is_tgtdev_for_dev_replace) { | |
9504 | ret = find_free_dev_extent(trans, device, min_free, | |
9505 | &dev_offset, NULL); | |
9506 | if (!ret) | |
9507 | dev_nr++; | |
9508 | ||
9509 | if (dev_nr >= dev_min) | |
9510 | break; | |
9511 | ||
9512 | ret = -1; | |
9513 | } | |
9514 | } | |
9515 | if (debug && ret == -1) | |
9516 | btrfs_warn(root->fs_info, | |
9517 | "no space to allocate a new chunk for block group %llu", | |
9518 | block_group->key.objectid); | |
9519 | mutex_unlock(&root->fs_info->chunk_mutex); | |
9520 | btrfs_end_transaction(trans, root); | |
9521 | out: | |
9522 | btrfs_put_block_group(block_group); | |
9523 | return ret; | |
9524 | } | |
9525 | ||
9526 | static int find_first_block_group(struct btrfs_root *root, | |
9527 | struct btrfs_path *path, struct btrfs_key *key) | |
9528 | { | |
9529 | int ret = 0; | |
9530 | struct btrfs_key found_key; | |
9531 | struct extent_buffer *leaf; | |
9532 | int slot; | |
9533 | ||
9534 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); | |
9535 | if (ret < 0) | |
9536 | goto out; | |
9537 | ||
9538 | while (1) { | |
9539 | slot = path->slots[0]; | |
9540 | leaf = path->nodes[0]; | |
9541 | if (slot >= btrfs_header_nritems(leaf)) { | |
9542 | ret = btrfs_next_leaf(root, path); | |
9543 | if (ret == 0) | |
9544 | continue; | |
9545 | if (ret < 0) | |
9546 | goto out; | |
9547 | break; | |
9548 | } | |
9549 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
9550 | ||
9551 | if (found_key.objectid >= key->objectid && | |
9552 | found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) { | |
9553 | ret = 0; | |
9554 | goto out; | |
9555 | } | |
9556 | path->slots[0]++; | |
9557 | } | |
9558 | out: | |
9559 | return ret; | |
9560 | } | |
9561 | ||
9562 | void btrfs_put_block_group_cache(struct btrfs_fs_info *info) | |
9563 | { | |
9564 | struct btrfs_block_group_cache *block_group; | |
9565 | u64 last = 0; | |
9566 | ||
9567 | while (1) { | |
9568 | struct inode *inode; | |
9569 | ||
9570 | block_group = btrfs_lookup_first_block_group(info, last); | |
9571 | while (block_group) { | |
9572 | spin_lock(&block_group->lock); | |
9573 | if (block_group->iref) | |
9574 | break; | |
9575 | spin_unlock(&block_group->lock); | |
9576 | block_group = next_block_group(info->tree_root, | |
9577 | block_group); | |
9578 | } | |
9579 | if (!block_group) { | |
9580 | if (last == 0) | |
9581 | break; | |
9582 | last = 0; | |
9583 | continue; | |
9584 | } | |
9585 | ||
9586 | inode = block_group->inode; | |
9587 | block_group->iref = 0; | |
9588 | block_group->inode = NULL; | |
9589 | spin_unlock(&block_group->lock); | |
9590 | iput(inode); | |
9591 | last = block_group->key.objectid + block_group->key.offset; | |
9592 | btrfs_put_block_group(block_group); | |
9593 | } | |
9594 | } | |
9595 | ||
9596 | int btrfs_free_block_groups(struct btrfs_fs_info *info) | |
9597 | { | |
9598 | struct btrfs_block_group_cache *block_group; | |
9599 | struct btrfs_space_info *space_info; | |
9600 | struct btrfs_caching_control *caching_ctl; | |
9601 | struct rb_node *n; | |
9602 | ||
9603 | down_write(&info->commit_root_sem); | |
9604 | while (!list_empty(&info->caching_block_groups)) { | |
9605 | caching_ctl = list_entry(info->caching_block_groups.next, | |
9606 | struct btrfs_caching_control, list); | |
9607 | list_del(&caching_ctl->list); | |
9608 | put_caching_control(caching_ctl); | |
9609 | } | |
9610 | up_write(&info->commit_root_sem); | |
9611 | ||
9612 | spin_lock(&info->unused_bgs_lock); | |
9613 | while (!list_empty(&info->unused_bgs)) { | |
9614 | block_group = list_first_entry(&info->unused_bgs, | |
9615 | struct btrfs_block_group_cache, | |
9616 | bg_list); | |
9617 | list_del_init(&block_group->bg_list); | |
9618 | btrfs_put_block_group(block_group); | |
9619 | } | |
9620 | spin_unlock(&info->unused_bgs_lock); | |
9621 | ||
9622 | spin_lock(&info->block_group_cache_lock); | |
9623 | while ((n = rb_last(&info->block_group_cache_tree)) != NULL) { | |
9624 | block_group = rb_entry(n, struct btrfs_block_group_cache, | |
9625 | cache_node); | |
9626 | rb_erase(&block_group->cache_node, | |
9627 | &info->block_group_cache_tree); | |
9628 | RB_CLEAR_NODE(&block_group->cache_node); | |
9629 | spin_unlock(&info->block_group_cache_lock); | |
9630 | ||
9631 | down_write(&block_group->space_info->groups_sem); | |
9632 | list_del(&block_group->list); | |
9633 | up_write(&block_group->space_info->groups_sem); | |
9634 | ||
9635 | if (block_group->cached == BTRFS_CACHE_STARTED) | |
9636 | wait_block_group_cache_done(block_group); | |
9637 | ||
9638 | /* | |
9639 | * We haven't cached this block group, which means we could | |
9640 | * possibly have excluded extents on this block group. | |
9641 | */ | |
9642 | if (block_group->cached == BTRFS_CACHE_NO || | |
9643 | block_group->cached == BTRFS_CACHE_ERROR) | |
9644 | free_excluded_extents(info->extent_root, block_group); | |
9645 | ||
9646 | btrfs_remove_free_space_cache(block_group); | |
9647 | btrfs_put_block_group(block_group); | |
9648 | ||
9649 | spin_lock(&info->block_group_cache_lock); | |
9650 | } | |
9651 | spin_unlock(&info->block_group_cache_lock); | |
9652 | ||
9653 | /* now that all the block groups are freed, go through and | |
9654 | * free all the space_info structs. This is only called during | |
9655 | * the final stages of unmount, and so we know nobody is | |
9656 | * using them. We call synchronize_rcu() once before we start, | |
9657 | * just to be on the safe side. | |
9658 | */ | |
9659 | synchronize_rcu(); | |
9660 | ||
9661 | release_global_block_rsv(info); | |
9662 | ||
9663 | while (!list_empty(&info->space_info)) { | |
9664 | int i; | |
9665 | ||
9666 | space_info = list_entry(info->space_info.next, | |
9667 | struct btrfs_space_info, | |
9668 | list); | |
9669 | if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) { | |
9670 | if (WARN_ON(space_info->bytes_pinned > 0 || | |
9671 | space_info->bytes_reserved > 0 || | |
9672 | space_info->bytes_may_use > 0)) { | |
9673 | dump_space_info(space_info, 0, 0); | |
9674 | } | |
9675 | } | |
9676 | list_del(&space_info->list); | |
9677 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { | |
9678 | struct kobject *kobj; | |
9679 | kobj = space_info->block_group_kobjs[i]; | |
9680 | space_info->block_group_kobjs[i] = NULL; | |
9681 | if (kobj) { | |
9682 | kobject_del(kobj); | |
9683 | kobject_put(kobj); | |
9684 | } | |
9685 | } | |
9686 | kobject_del(&space_info->kobj); | |
9687 | kobject_put(&space_info->kobj); | |
9688 | } | |
9689 | return 0; | |
9690 | } | |
9691 | ||
9692 | static void __link_block_group(struct btrfs_space_info *space_info, | |
9693 | struct btrfs_block_group_cache *cache) | |
9694 | { | |
9695 | int index = get_block_group_index(cache); | |
9696 | bool first = false; | |
9697 | ||
9698 | down_write(&space_info->groups_sem); | |
9699 | if (list_empty(&space_info->block_groups[index])) | |
9700 | first = true; | |
9701 | list_add_tail(&cache->list, &space_info->block_groups[index]); | |
9702 | up_write(&space_info->groups_sem); | |
9703 | ||
9704 | if (first) { | |
9705 | struct raid_kobject *rkobj; | |
9706 | int ret; | |
9707 | ||
9708 | rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS); | |
9709 | if (!rkobj) | |
9710 | goto out_err; | |
9711 | rkobj->raid_type = index; | |
9712 | kobject_init(&rkobj->kobj, &btrfs_raid_ktype); | |
9713 | ret = kobject_add(&rkobj->kobj, &space_info->kobj, | |
9714 | "%s", get_raid_name(index)); | |
9715 | if (ret) { | |
9716 | kobject_put(&rkobj->kobj); | |
9717 | goto out_err; | |
9718 | } | |
9719 | space_info->block_group_kobjs[index] = &rkobj->kobj; | |
9720 | } | |
9721 | ||
9722 | return; | |
9723 | out_err: | |
9724 | pr_warn("BTRFS: failed to add kobject for block cache. ignoring.\n"); | |
9725 | } | |
9726 | ||
9727 | static struct btrfs_block_group_cache * | |
9728 | btrfs_create_block_group_cache(struct btrfs_root *root, u64 start, u64 size) | |
9729 | { | |
9730 | struct btrfs_block_group_cache *cache; | |
9731 | ||
9732 | cache = kzalloc(sizeof(*cache), GFP_NOFS); | |
9733 | if (!cache) | |
9734 | return NULL; | |
9735 | ||
9736 | cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl), | |
9737 | GFP_NOFS); | |
9738 | if (!cache->free_space_ctl) { | |
9739 | kfree(cache); | |
9740 | return NULL; | |
9741 | } | |
9742 | ||
9743 | cache->key.objectid = start; | |
9744 | cache->key.offset = size; | |
9745 | cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; | |
9746 | ||
9747 | cache->sectorsize = root->sectorsize; | |
9748 | cache->fs_info = root->fs_info; | |
9749 | cache->full_stripe_len = btrfs_full_stripe_len(root, | |
9750 | &root->fs_info->mapping_tree, | |
9751 | start); | |
9752 | set_free_space_tree_thresholds(cache); | |
9753 | ||
9754 | atomic_set(&cache->count, 1); | |
9755 | spin_lock_init(&cache->lock); | |
9756 | init_rwsem(&cache->data_rwsem); | |
9757 | INIT_LIST_HEAD(&cache->list); | |
9758 | INIT_LIST_HEAD(&cache->cluster_list); | |
9759 | INIT_LIST_HEAD(&cache->bg_list); | |
9760 | INIT_LIST_HEAD(&cache->ro_list); | |
9761 | INIT_LIST_HEAD(&cache->dirty_list); | |
9762 | INIT_LIST_HEAD(&cache->io_list); | |
9763 | btrfs_init_free_space_ctl(cache); | |
9764 | atomic_set(&cache->trimming, 0); | |
9765 | mutex_init(&cache->free_space_lock); | |
9766 | ||
9767 | return cache; | |
9768 | } | |
9769 | ||
9770 | int btrfs_read_block_groups(struct btrfs_root *root) | |
9771 | { | |
9772 | struct btrfs_path *path; | |
9773 | int ret; | |
9774 | struct btrfs_block_group_cache *cache; | |
9775 | struct btrfs_fs_info *info = root->fs_info; | |
9776 | struct btrfs_space_info *space_info; | |
9777 | struct btrfs_key key; | |
9778 | struct btrfs_key found_key; | |
9779 | struct extent_buffer *leaf; | |
9780 | int need_clear = 0; | |
9781 | u64 cache_gen; | |
9782 | ||
9783 | root = info->extent_root; | |
9784 | key.objectid = 0; | |
9785 | key.offset = 0; | |
9786 | key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; | |
9787 | path = btrfs_alloc_path(); | |
9788 | if (!path) | |
9789 | return -ENOMEM; | |
9790 | path->reada = READA_FORWARD; | |
9791 | ||
9792 | cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy); | |
9793 | if (btrfs_test_opt(root, SPACE_CACHE) && | |
9794 | btrfs_super_generation(root->fs_info->super_copy) != cache_gen) | |
9795 | need_clear = 1; | |
9796 | if (btrfs_test_opt(root, CLEAR_CACHE)) | |
9797 | need_clear = 1; | |
9798 | ||
9799 | while (1) { | |
9800 | ret = find_first_block_group(root, path, &key); | |
9801 | if (ret > 0) | |
9802 | break; | |
9803 | if (ret != 0) | |
9804 | goto error; | |
9805 | ||
9806 | leaf = path->nodes[0]; | |
9807 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
9808 | ||
9809 | cache = btrfs_create_block_group_cache(root, found_key.objectid, | |
9810 | found_key.offset); | |
9811 | if (!cache) { | |
9812 | ret = -ENOMEM; | |
9813 | goto error; | |
9814 | } | |
9815 | ||
9816 | if (need_clear) { | |
9817 | /* | |
9818 | * When we mount with old space cache, we need to | |
9819 | * set BTRFS_DC_CLEAR and set dirty flag. | |
9820 | * | |
9821 | * a) Setting 'BTRFS_DC_CLEAR' makes sure that we | |
9822 | * truncate the old free space cache inode and | |
9823 | * setup a new one. | |
9824 | * b) Setting 'dirty flag' makes sure that we flush | |
9825 | * the new space cache info onto disk. | |
9826 | */ | |
9827 | if (btrfs_test_opt(root, SPACE_CACHE)) | |
9828 | cache->disk_cache_state = BTRFS_DC_CLEAR; | |
9829 | } | |
9830 | ||
9831 | read_extent_buffer(leaf, &cache->item, | |
9832 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
9833 | sizeof(cache->item)); | |
9834 | cache->flags = btrfs_block_group_flags(&cache->item); | |
9835 | ||
9836 | key.objectid = found_key.objectid + found_key.offset; | |
9837 | btrfs_release_path(path); | |
9838 | ||
9839 | /* | |
9840 | * We need to exclude the super stripes now so that the space | |
9841 | * info has super bytes accounted for, otherwise we'll think | |
9842 | * we have more space than we actually do. | |
9843 | */ | |
9844 | ret = exclude_super_stripes(root, cache); | |
9845 | if (ret) { | |
9846 | /* | |
9847 | * We may have excluded something, so call this just in | |
9848 | * case. | |
9849 | */ | |
9850 | free_excluded_extents(root, cache); | |
9851 | btrfs_put_block_group(cache); | |
9852 | goto error; | |
9853 | } | |
9854 | ||
9855 | /* | |
9856 | * check for two cases, either we are full, and therefore | |
9857 | * don't need to bother with the caching work since we won't | |
9858 | * find any space, or we are empty, and we can just add all | |
9859 | * the space in and be done with it. This saves us _alot_ of | |
9860 | * time, particularly in the full case. | |
9861 | */ | |
9862 | if (found_key.offset == btrfs_block_group_used(&cache->item)) { | |
9863 | cache->last_byte_to_unpin = (u64)-1; | |
9864 | cache->cached = BTRFS_CACHE_FINISHED; | |
9865 | free_excluded_extents(root, cache); | |
9866 | } else if (btrfs_block_group_used(&cache->item) == 0) { | |
9867 | cache->last_byte_to_unpin = (u64)-1; | |
9868 | cache->cached = BTRFS_CACHE_FINISHED; | |
9869 | add_new_free_space(cache, root->fs_info, | |
9870 | found_key.objectid, | |
9871 | found_key.objectid + | |
9872 | found_key.offset); | |
9873 | free_excluded_extents(root, cache); | |
9874 | } | |
9875 | ||
9876 | ret = btrfs_add_block_group_cache(root->fs_info, cache); | |
9877 | if (ret) { | |
9878 | btrfs_remove_free_space_cache(cache); | |
9879 | btrfs_put_block_group(cache); | |
9880 | goto error; | |
9881 | } | |
9882 | ||
9883 | ret = update_space_info(info, cache->flags, found_key.offset, | |
9884 | btrfs_block_group_used(&cache->item), | |
9885 | &space_info); | |
9886 | if (ret) { | |
9887 | btrfs_remove_free_space_cache(cache); | |
9888 | spin_lock(&info->block_group_cache_lock); | |
9889 | rb_erase(&cache->cache_node, | |
9890 | &info->block_group_cache_tree); | |
9891 | RB_CLEAR_NODE(&cache->cache_node); | |
9892 | spin_unlock(&info->block_group_cache_lock); | |
9893 | btrfs_put_block_group(cache); | |
9894 | goto error; | |
9895 | } | |
9896 | ||
9897 | cache->space_info = space_info; | |
9898 | spin_lock(&cache->space_info->lock); | |
9899 | cache->space_info->bytes_readonly += cache->bytes_super; | |
9900 | spin_unlock(&cache->space_info->lock); | |
9901 | ||
9902 | __link_block_group(space_info, cache); | |
9903 | ||
9904 | set_avail_alloc_bits(root->fs_info, cache->flags); | |
9905 | if (btrfs_chunk_readonly(root, cache->key.objectid)) { | |
9906 | inc_block_group_ro(cache, 1); | |
9907 | } else if (btrfs_block_group_used(&cache->item) == 0) { | |
9908 | spin_lock(&info->unused_bgs_lock); | |
9909 | /* Should always be true but just in case. */ | |
9910 | if (list_empty(&cache->bg_list)) { | |
9911 | btrfs_get_block_group(cache); | |
9912 | list_add_tail(&cache->bg_list, | |
9913 | &info->unused_bgs); | |
9914 | } | |
9915 | spin_unlock(&info->unused_bgs_lock); | |
9916 | } | |
9917 | } | |
9918 | ||
9919 | list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) { | |
9920 | if (!(get_alloc_profile(root, space_info->flags) & | |
9921 | (BTRFS_BLOCK_GROUP_RAID10 | | |
9922 | BTRFS_BLOCK_GROUP_RAID1 | | |
9923 | BTRFS_BLOCK_GROUP_RAID5 | | |
9924 | BTRFS_BLOCK_GROUP_RAID6 | | |
9925 | BTRFS_BLOCK_GROUP_DUP))) | |
9926 | continue; | |
9927 | /* | |
9928 | * avoid allocating from un-mirrored block group if there are | |
9929 | * mirrored block groups. | |
9930 | */ | |
9931 | list_for_each_entry(cache, | |
9932 | &space_info->block_groups[BTRFS_RAID_RAID0], | |
9933 | list) | |
9934 | inc_block_group_ro(cache, 1); | |
9935 | list_for_each_entry(cache, | |
9936 | &space_info->block_groups[BTRFS_RAID_SINGLE], | |
9937 | list) | |
9938 | inc_block_group_ro(cache, 1); | |
9939 | } | |
9940 | ||
9941 | init_global_block_rsv(info); | |
9942 | ret = 0; | |
9943 | error: | |
9944 | btrfs_free_path(path); | |
9945 | return ret; | |
9946 | } | |
9947 | ||
9948 | void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans, | |
9949 | struct btrfs_root *root) | |
9950 | { | |
9951 | struct btrfs_block_group_cache *block_group, *tmp; | |
9952 | struct btrfs_root *extent_root = root->fs_info->extent_root; | |
9953 | struct btrfs_block_group_item item; | |
9954 | struct btrfs_key key; | |
9955 | int ret = 0; | |
9956 | bool can_flush_pending_bgs = trans->can_flush_pending_bgs; | |
9957 | ||
9958 | trans->can_flush_pending_bgs = false; | |
9959 | list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) { | |
9960 | if (ret) | |
9961 | goto next; | |
9962 | ||
9963 | spin_lock(&block_group->lock); | |
9964 | memcpy(&item, &block_group->item, sizeof(item)); | |
9965 | memcpy(&key, &block_group->key, sizeof(key)); | |
9966 | spin_unlock(&block_group->lock); | |
9967 | ||
9968 | ret = btrfs_insert_item(trans, extent_root, &key, &item, | |
9969 | sizeof(item)); | |
9970 | if (ret) | |
9971 | btrfs_abort_transaction(trans, extent_root, ret); | |
9972 | ret = btrfs_finish_chunk_alloc(trans, extent_root, | |
9973 | key.objectid, key.offset); | |
9974 | if (ret) | |
9975 | btrfs_abort_transaction(trans, extent_root, ret); | |
9976 | add_block_group_free_space(trans, root->fs_info, block_group); | |
9977 | /* already aborted the transaction if it failed. */ | |
9978 | next: | |
9979 | list_del_init(&block_group->bg_list); | |
9980 | } | |
9981 | trans->can_flush_pending_bgs = can_flush_pending_bgs; | |
9982 | } | |
9983 | ||
9984 | int btrfs_make_block_group(struct btrfs_trans_handle *trans, | |
9985 | struct btrfs_root *root, u64 bytes_used, | |
9986 | u64 type, u64 chunk_objectid, u64 chunk_offset, | |
9987 | u64 size) | |
9988 | { | |
9989 | int ret; | |
9990 | struct btrfs_root *extent_root; | |
9991 | struct btrfs_block_group_cache *cache; | |
9992 | ||
9993 | extent_root = root->fs_info->extent_root; | |
9994 | ||
9995 | btrfs_set_log_full_commit(root->fs_info, trans); | |
9996 | ||
9997 | cache = btrfs_create_block_group_cache(root, chunk_offset, size); | |
9998 | if (!cache) | |
9999 | return -ENOMEM; | |
10000 | ||
10001 | btrfs_set_block_group_used(&cache->item, bytes_used); | |
10002 | btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid); | |
10003 | btrfs_set_block_group_flags(&cache->item, type); | |
10004 | ||
10005 | cache->flags = type; | |
10006 | cache->last_byte_to_unpin = (u64)-1; | |
10007 | cache->cached = BTRFS_CACHE_FINISHED; | |
10008 | cache->needs_free_space = 1; | |
10009 | ret = exclude_super_stripes(root, cache); | |
10010 | if (ret) { | |
10011 | /* | |
10012 | * We may have excluded something, so call this just in | |
10013 | * case. | |
10014 | */ | |
10015 | free_excluded_extents(root, cache); | |
10016 | btrfs_put_block_group(cache); | |
10017 | return ret; | |
10018 | } | |
10019 | ||
10020 | add_new_free_space(cache, root->fs_info, chunk_offset, | |
10021 | chunk_offset + size); | |
10022 | ||
10023 | free_excluded_extents(root, cache); | |
10024 | ||
10025 | #ifdef CONFIG_BTRFS_DEBUG | |
10026 | if (btrfs_should_fragment_free_space(root, cache)) { | |
10027 | u64 new_bytes_used = size - bytes_used; | |
10028 | ||
10029 | bytes_used += new_bytes_used >> 1; | |
10030 | fragment_free_space(root, cache); | |
10031 | } | |
10032 | #endif | |
10033 | /* | |
10034 | * Call to ensure the corresponding space_info object is created and | |
10035 | * assigned to our block group, but don't update its counters just yet. | |
10036 | * We want our bg to be added to the rbtree with its ->space_info set. | |
10037 | */ | |
10038 | ret = update_space_info(root->fs_info, cache->flags, 0, 0, | |
10039 | &cache->space_info); | |
10040 | if (ret) { | |
10041 | btrfs_remove_free_space_cache(cache); | |
10042 | btrfs_put_block_group(cache); | |
10043 | return ret; | |
10044 | } | |
10045 | ||
10046 | ret = btrfs_add_block_group_cache(root->fs_info, cache); | |
10047 | if (ret) { | |
10048 | btrfs_remove_free_space_cache(cache); | |
10049 | btrfs_put_block_group(cache); | |
10050 | return ret; | |
10051 | } | |
10052 | ||
10053 | /* | |
10054 | * Now that our block group has its ->space_info set and is inserted in | |
10055 | * the rbtree, update the space info's counters. | |
10056 | */ | |
10057 | ret = update_space_info(root->fs_info, cache->flags, size, bytes_used, | |
10058 | &cache->space_info); | |
10059 | if (ret) { | |
10060 | btrfs_remove_free_space_cache(cache); | |
10061 | spin_lock(&root->fs_info->block_group_cache_lock); | |
10062 | rb_erase(&cache->cache_node, | |
10063 | &root->fs_info->block_group_cache_tree); | |
10064 | RB_CLEAR_NODE(&cache->cache_node); | |
10065 | spin_unlock(&root->fs_info->block_group_cache_lock); | |
10066 | btrfs_put_block_group(cache); | |
10067 | return ret; | |
10068 | } | |
10069 | update_global_block_rsv(root->fs_info); | |
10070 | ||
10071 | spin_lock(&cache->space_info->lock); | |
10072 | cache->space_info->bytes_readonly += cache->bytes_super; | |
10073 | spin_unlock(&cache->space_info->lock); | |
10074 | ||
10075 | __link_block_group(cache->space_info, cache); | |
10076 | ||
10077 | list_add_tail(&cache->bg_list, &trans->new_bgs); | |
10078 | ||
10079 | set_avail_alloc_bits(extent_root->fs_info, type); | |
10080 | ||
10081 | return 0; | |
10082 | } | |
10083 | ||
10084 | static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) | |
10085 | { | |
10086 | u64 extra_flags = chunk_to_extended(flags) & | |
10087 | BTRFS_EXTENDED_PROFILE_MASK; | |
10088 | ||
10089 | write_seqlock(&fs_info->profiles_lock); | |
10090 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
10091 | fs_info->avail_data_alloc_bits &= ~extra_flags; | |
10092 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
10093 | fs_info->avail_metadata_alloc_bits &= ~extra_flags; | |
10094 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
10095 | fs_info->avail_system_alloc_bits &= ~extra_flags; | |
10096 | write_sequnlock(&fs_info->profiles_lock); | |
10097 | } | |
10098 | ||
10099 | int btrfs_remove_block_group(struct btrfs_trans_handle *trans, | |
10100 | struct btrfs_root *root, u64 group_start, | |
10101 | struct extent_map *em) | |
10102 | { | |
10103 | struct btrfs_path *path; | |
10104 | struct btrfs_block_group_cache *block_group; | |
10105 | struct btrfs_free_cluster *cluster; | |
10106 | struct btrfs_root *tree_root = root->fs_info->tree_root; | |
10107 | struct btrfs_key key; | |
10108 | struct inode *inode; | |
10109 | struct kobject *kobj = NULL; | |
10110 | int ret; | |
10111 | int index; | |
10112 | int factor; | |
10113 | struct btrfs_caching_control *caching_ctl = NULL; | |
10114 | bool remove_em; | |
10115 | ||
10116 | root = root->fs_info->extent_root; | |
10117 | ||
10118 | block_group = btrfs_lookup_block_group(root->fs_info, group_start); | |
10119 | BUG_ON(!block_group); | |
10120 | BUG_ON(!block_group->ro); | |
10121 | ||
10122 | /* | |
10123 | * Free the reserved super bytes from this block group before | |
10124 | * remove it. | |
10125 | */ | |
10126 | free_excluded_extents(root, block_group); | |
10127 | ||
10128 | memcpy(&key, &block_group->key, sizeof(key)); | |
10129 | index = get_block_group_index(block_group); | |
10130 | if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP | | |
10131 | BTRFS_BLOCK_GROUP_RAID1 | | |
10132 | BTRFS_BLOCK_GROUP_RAID10)) | |
10133 | factor = 2; | |
10134 | else | |
10135 | factor = 1; | |
10136 | ||
10137 | /* make sure this block group isn't part of an allocation cluster */ | |
10138 | cluster = &root->fs_info->data_alloc_cluster; | |
10139 | spin_lock(&cluster->refill_lock); | |
10140 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
10141 | spin_unlock(&cluster->refill_lock); | |
10142 | ||
10143 | /* | |
10144 | * make sure this block group isn't part of a metadata | |
10145 | * allocation cluster | |
10146 | */ | |
10147 | cluster = &root->fs_info->meta_alloc_cluster; | |
10148 | spin_lock(&cluster->refill_lock); | |
10149 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
10150 | spin_unlock(&cluster->refill_lock); | |
10151 | ||
10152 | path = btrfs_alloc_path(); | |
10153 | if (!path) { | |
10154 | ret = -ENOMEM; | |
10155 | goto out; | |
10156 | } | |
10157 | ||
10158 | /* | |
10159 | * get the inode first so any iput calls done for the io_list | |
10160 | * aren't the final iput (no unlinks allowed now) | |
10161 | */ | |
10162 | inode = lookup_free_space_inode(tree_root, block_group, path); | |
10163 | ||
10164 | mutex_lock(&trans->transaction->cache_write_mutex); | |
10165 | /* | |
10166 | * make sure our free spache cache IO is done before remove the | |
10167 | * free space inode | |
10168 | */ | |
10169 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
10170 | if (!list_empty(&block_group->io_list)) { | |
10171 | list_del_init(&block_group->io_list); | |
10172 | ||
10173 | WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode); | |
10174 | ||
10175 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
10176 | btrfs_wait_cache_io(root, trans, block_group, | |
10177 | &block_group->io_ctl, path, | |
10178 | block_group->key.objectid); | |
10179 | btrfs_put_block_group(block_group); | |
10180 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
10181 | } | |
10182 | ||
10183 | if (!list_empty(&block_group->dirty_list)) { | |
10184 | list_del_init(&block_group->dirty_list); | |
10185 | btrfs_put_block_group(block_group); | |
10186 | } | |
10187 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
10188 | mutex_unlock(&trans->transaction->cache_write_mutex); | |
10189 | ||
10190 | if (!IS_ERR(inode)) { | |
10191 | ret = btrfs_orphan_add(trans, inode); | |
10192 | if (ret) { | |
10193 | btrfs_add_delayed_iput(inode); | |
10194 | goto out; | |
10195 | } | |
10196 | clear_nlink(inode); | |
10197 | /* One for the block groups ref */ | |
10198 | spin_lock(&block_group->lock); | |
10199 | if (block_group->iref) { | |
10200 | block_group->iref = 0; | |
10201 | block_group->inode = NULL; | |
10202 | spin_unlock(&block_group->lock); | |
10203 | iput(inode); | |
10204 | } else { | |
10205 | spin_unlock(&block_group->lock); | |
10206 | } | |
10207 | /* One for our lookup ref */ | |
10208 | btrfs_add_delayed_iput(inode); | |
10209 | } | |
10210 | ||
10211 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
10212 | key.offset = block_group->key.objectid; | |
10213 | key.type = 0; | |
10214 | ||
10215 | ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1); | |
10216 | if (ret < 0) | |
10217 | goto out; | |
10218 | if (ret > 0) | |
10219 | btrfs_release_path(path); | |
10220 | if (ret == 0) { | |
10221 | ret = btrfs_del_item(trans, tree_root, path); | |
10222 | if (ret) | |
10223 | goto out; | |
10224 | btrfs_release_path(path); | |
10225 | } | |
10226 | ||
10227 | spin_lock(&root->fs_info->block_group_cache_lock); | |
10228 | rb_erase(&block_group->cache_node, | |
10229 | &root->fs_info->block_group_cache_tree); | |
10230 | RB_CLEAR_NODE(&block_group->cache_node); | |
10231 | ||
10232 | if (root->fs_info->first_logical_byte == block_group->key.objectid) | |
10233 | root->fs_info->first_logical_byte = (u64)-1; | |
10234 | spin_unlock(&root->fs_info->block_group_cache_lock); | |
10235 | ||
10236 | down_write(&block_group->space_info->groups_sem); | |
10237 | /* | |
10238 | * we must use list_del_init so people can check to see if they | |
10239 | * are still on the list after taking the semaphore | |
10240 | */ | |
10241 | list_del_init(&block_group->list); | |
10242 | if (list_empty(&block_group->space_info->block_groups[index])) { | |
10243 | kobj = block_group->space_info->block_group_kobjs[index]; | |
10244 | block_group->space_info->block_group_kobjs[index] = NULL; | |
10245 | clear_avail_alloc_bits(root->fs_info, block_group->flags); | |
10246 | } | |
10247 | up_write(&block_group->space_info->groups_sem); | |
10248 | if (kobj) { | |
10249 | kobject_del(kobj); | |
10250 | kobject_put(kobj); | |
10251 | } | |
10252 | ||
10253 | if (block_group->has_caching_ctl) | |
10254 | caching_ctl = get_caching_control(block_group); | |
10255 | if (block_group->cached == BTRFS_CACHE_STARTED) | |
10256 | wait_block_group_cache_done(block_group); | |
10257 | if (block_group->has_caching_ctl) { | |
10258 | down_write(&root->fs_info->commit_root_sem); | |
10259 | if (!caching_ctl) { | |
10260 | struct btrfs_caching_control *ctl; | |
10261 | ||
10262 | list_for_each_entry(ctl, | |
10263 | &root->fs_info->caching_block_groups, list) | |
10264 | if (ctl->block_group == block_group) { | |
10265 | caching_ctl = ctl; | |
10266 | atomic_inc(&caching_ctl->count); | |
10267 | break; | |
10268 | } | |
10269 | } | |
10270 | if (caching_ctl) | |
10271 | list_del_init(&caching_ctl->list); | |
10272 | up_write(&root->fs_info->commit_root_sem); | |
10273 | if (caching_ctl) { | |
10274 | /* Once for the caching bgs list and once for us. */ | |
10275 | put_caching_control(caching_ctl); | |
10276 | put_caching_control(caching_ctl); | |
10277 | } | |
10278 | } | |
10279 | ||
10280 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
10281 | if (!list_empty(&block_group->dirty_list)) { | |
10282 | WARN_ON(1); | |
10283 | } | |
10284 | if (!list_empty(&block_group->io_list)) { | |
10285 | WARN_ON(1); | |
10286 | } | |
10287 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
10288 | btrfs_remove_free_space_cache(block_group); | |
10289 | ||
10290 | spin_lock(&block_group->space_info->lock); | |
10291 | list_del_init(&block_group->ro_list); | |
10292 | ||
10293 | if (btrfs_test_opt(root, ENOSPC_DEBUG)) { | |
10294 | WARN_ON(block_group->space_info->total_bytes | |
10295 | < block_group->key.offset); | |
10296 | WARN_ON(block_group->space_info->bytes_readonly | |
10297 | < block_group->key.offset); | |
10298 | WARN_ON(block_group->space_info->disk_total | |
10299 | < block_group->key.offset * factor); | |
10300 | } | |
10301 | block_group->space_info->total_bytes -= block_group->key.offset; | |
10302 | block_group->space_info->bytes_readonly -= block_group->key.offset; | |
10303 | block_group->space_info->disk_total -= block_group->key.offset * factor; | |
10304 | ||
10305 | spin_unlock(&block_group->space_info->lock); | |
10306 | ||
10307 | memcpy(&key, &block_group->key, sizeof(key)); | |
10308 | ||
10309 | lock_chunks(root); | |
10310 | if (!list_empty(&em->list)) { | |
10311 | /* We're in the transaction->pending_chunks list. */ | |
10312 | free_extent_map(em); | |
10313 | } | |
10314 | spin_lock(&block_group->lock); | |
10315 | block_group->removed = 1; | |
10316 | /* | |
10317 | * At this point trimming can't start on this block group, because we | |
10318 | * removed the block group from the tree fs_info->block_group_cache_tree | |
10319 | * so no one can't find it anymore and even if someone already got this | |
10320 | * block group before we removed it from the rbtree, they have already | |
10321 | * incremented block_group->trimming - if they didn't, they won't find | |
10322 | * any free space entries because we already removed them all when we | |
10323 | * called btrfs_remove_free_space_cache(). | |
10324 | * | |
10325 | * And we must not remove the extent map from the fs_info->mapping_tree | |
10326 | * to prevent the same logical address range and physical device space | |
10327 | * ranges from being reused for a new block group. This is because our | |
10328 | * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is | |
10329 | * completely transactionless, so while it is trimming a range the | |
10330 | * currently running transaction might finish and a new one start, | |
10331 | * allowing for new block groups to be created that can reuse the same | |
10332 | * physical device locations unless we take this special care. | |
10333 | * | |
10334 | * There may also be an implicit trim operation if the file system | |
10335 | * is mounted with -odiscard. The same protections must remain | |
10336 | * in place until the extents have been discarded completely when | |
10337 | * the transaction commit has completed. | |
10338 | */ | |
10339 | remove_em = (atomic_read(&block_group->trimming) == 0); | |
10340 | /* | |
10341 | * Make sure a trimmer task always sees the em in the pinned_chunks list | |
10342 | * if it sees block_group->removed == 1 (needs to lock block_group->lock | |
10343 | * before checking block_group->removed). | |
10344 | */ | |
10345 | if (!remove_em) { | |
10346 | /* | |
10347 | * Our em might be in trans->transaction->pending_chunks which | |
10348 | * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks), | |
10349 | * and so is the fs_info->pinned_chunks list. | |
10350 | * | |
10351 | * So at this point we must be holding the chunk_mutex to avoid | |
10352 | * any races with chunk allocation (more specifically at | |
10353 | * volumes.c:contains_pending_extent()), to ensure it always | |
10354 | * sees the em, either in the pending_chunks list or in the | |
10355 | * pinned_chunks list. | |
10356 | */ | |
10357 | list_move_tail(&em->list, &root->fs_info->pinned_chunks); | |
10358 | } | |
10359 | spin_unlock(&block_group->lock); | |
10360 | ||
10361 | if (remove_em) { | |
10362 | struct extent_map_tree *em_tree; | |
10363 | ||
10364 | em_tree = &root->fs_info->mapping_tree.map_tree; | |
10365 | write_lock(&em_tree->lock); | |
10366 | /* | |
10367 | * The em might be in the pending_chunks list, so make sure the | |
10368 | * chunk mutex is locked, since remove_extent_mapping() will | |
10369 | * delete us from that list. | |
10370 | */ | |
10371 | remove_extent_mapping(em_tree, em); | |
10372 | write_unlock(&em_tree->lock); | |
10373 | /* once for the tree */ | |
10374 | free_extent_map(em); | |
10375 | } | |
10376 | ||
10377 | unlock_chunks(root); | |
10378 | ||
10379 | ret = remove_block_group_free_space(trans, root->fs_info, block_group); | |
10380 | if (ret) | |
10381 | goto out; | |
10382 | ||
10383 | btrfs_put_block_group(block_group); | |
10384 | btrfs_put_block_group(block_group); | |
10385 | ||
10386 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
10387 | if (ret > 0) | |
10388 | ret = -EIO; | |
10389 | if (ret < 0) | |
10390 | goto out; | |
10391 | ||
10392 | ret = btrfs_del_item(trans, root, path); | |
10393 | out: | |
10394 | btrfs_free_path(path); | |
10395 | return ret; | |
10396 | } | |
10397 | ||
10398 | struct btrfs_trans_handle * | |
10399 | btrfs_start_trans_remove_block_group(struct btrfs_fs_info *fs_info, | |
10400 | const u64 chunk_offset) | |
10401 | { | |
10402 | struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree; | |
10403 | struct extent_map *em; | |
10404 | struct map_lookup *map; | |
10405 | unsigned int num_items; | |
10406 | ||
10407 | read_lock(&em_tree->lock); | |
10408 | em = lookup_extent_mapping(em_tree, chunk_offset, 1); | |
10409 | read_unlock(&em_tree->lock); | |
10410 | ASSERT(em && em->start == chunk_offset); | |
10411 | ||
10412 | /* | |
10413 | * We need to reserve 3 + N units from the metadata space info in order | |
10414 | * to remove a block group (done at btrfs_remove_chunk() and at | |
10415 | * btrfs_remove_block_group()), which are used for: | |
10416 | * | |
10417 | * 1 unit for adding the free space inode's orphan (located in the tree | |
10418 | * of tree roots). | |
10419 | * 1 unit for deleting the block group item (located in the extent | |
10420 | * tree). | |
10421 | * 1 unit for deleting the free space item (located in tree of tree | |
10422 | * roots). | |
10423 | * N units for deleting N device extent items corresponding to each | |
10424 | * stripe (located in the device tree). | |
10425 | * | |
10426 | * In order to remove a block group we also need to reserve units in the | |
10427 | * system space info in order to update the chunk tree (update one or | |
10428 | * more device items and remove one chunk item), but this is done at | |
10429 | * btrfs_remove_chunk() through a call to check_system_chunk(). | |
10430 | */ | |
10431 | map = em->map_lookup; | |
10432 | num_items = 3 + map->num_stripes; | |
10433 | free_extent_map(em); | |
10434 | ||
10435 | return btrfs_start_transaction_fallback_global_rsv(fs_info->extent_root, | |
10436 | num_items, 1); | |
10437 | } | |
10438 | ||
10439 | /* | |
10440 | * Process the unused_bgs list and remove any that don't have any allocated | |
10441 | * space inside of them. | |
10442 | */ | |
10443 | void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info) | |
10444 | { | |
10445 | struct btrfs_block_group_cache *block_group; | |
10446 | struct btrfs_space_info *space_info; | |
10447 | struct btrfs_root *root = fs_info->extent_root; | |
10448 | struct btrfs_trans_handle *trans; | |
10449 | int ret = 0; | |
10450 | ||
10451 | if (!fs_info->open) | |
10452 | return; | |
10453 | ||
10454 | spin_lock(&fs_info->unused_bgs_lock); | |
10455 | while (!list_empty(&fs_info->unused_bgs)) { | |
10456 | u64 start, end; | |
10457 | int trimming; | |
10458 | ||
10459 | block_group = list_first_entry(&fs_info->unused_bgs, | |
10460 | struct btrfs_block_group_cache, | |
10461 | bg_list); | |
10462 | list_del_init(&block_group->bg_list); | |
10463 | ||
10464 | space_info = block_group->space_info; | |
10465 | ||
10466 | if (ret || btrfs_mixed_space_info(space_info)) { | |
10467 | btrfs_put_block_group(block_group); | |
10468 | continue; | |
10469 | } | |
10470 | spin_unlock(&fs_info->unused_bgs_lock); | |
10471 | ||
10472 | mutex_lock(&fs_info->delete_unused_bgs_mutex); | |
10473 | ||
10474 | /* Don't want to race with allocators so take the groups_sem */ | |
10475 | down_write(&space_info->groups_sem); | |
10476 | spin_lock(&block_group->lock); | |
10477 | if (block_group->reserved || | |
10478 | btrfs_block_group_used(&block_group->item) || | |
10479 | block_group->ro || | |
10480 | list_is_singular(&block_group->list)) { | |
10481 | /* | |
10482 | * We want to bail if we made new allocations or have | |
10483 | * outstanding allocations in this block group. We do | |
10484 | * the ro check in case balance is currently acting on | |
10485 | * this block group. | |
10486 | */ | |
10487 | spin_unlock(&block_group->lock); | |
10488 | up_write(&space_info->groups_sem); | |
10489 | goto next; | |
10490 | } | |
10491 | spin_unlock(&block_group->lock); | |
10492 | ||
10493 | /* We don't want to force the issue, only flip if it's ok. */ | |
10494 | ret = inc_block_group_ro(block_group, 0); | |
10495 | up_write(&space_info->groups_sem); | |
10496 | if (ret < 0) { | |
10497 | ret = 0; | |
10498 | goto next; | |
10499 | } | |
10500 | ||
10501 | /* | |
10502 | * Want to do this before we do anything else so we can recover | |
10503 | * properly if we fail to join the transaction. | |
10504 | */ | |
10505 | trans = btrfs_start_trans_remove_block_group(fs_info, | |
10506 | block_group->key.objectid); | |
10507 | if (IS_ERR(trans)) { | |
10508 | btrfs_dec_block_group_ro(root, block_group); | |
10509 | ret = PTR_ERR(trans); | |
10510 | goto next; | |
10511 | } | |
10512 | ||
10513 | /* | |
10514 | * We could have pending pinned extents for this block group, | |
10515 | * just delete them, we don't care about them anymore. | |
10516 | */ | |
10517 | start = block_group->key.objectid; | |
10518 | end = start + block_group->key.offset - 1; | |
10519 | /* | |
10520 | * Hold the unused_bg_unpin_mutex lock to avoid racing with | |
10521 | * btrfs_finish_extent_commit(). If we are at transaction N, | |
10522 | * another task might be running finish_extent_commit() for the | |
10523 | * previous transaction N - 1, and have seen a range belonging | |
10524 | * to the block group in freed_extents[] before we were able to | |
10525 | * clear the whole block group range from freed_extents[]. This | |
10526 | * means that task can lookup for the block group after we | |
10527 | * unpinned it from freed_extents[] and removed it, leading to | |
10528 | * a BUG_ON() at btrfs_unpin_extent_range(). | |
10529 | */ | |
10530 | mutex_lock(&fs_info->unused_bg_unpin_mutex); | |
10531 | ret = clear_extent_bits(&fs_info->freed_extents[0], start, end, | |
10532 | EXTENT_DIRTY, GFP_NOFS); | |
10533 | if (ret) { | |
10534 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); | |
10535 | btrfs_dec_block_group_ro(root, block_group); | |
10536 | goto end_trans; | |
10537 | } | |
10538 | ret = clear_extent_bits(&fs_info->freed_extents[1], start, end, | |
10539 | EXTENT_DIRTY, GFP_NOFS); | |
10540 | if (ret) { | |
10541 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); | |
10542 | btrfs_dec_block_group_ro(root, block_group); | |
10543 | goto end_trans; | |
10544 | } | |
10545 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); | |
10546 | ||
10547 | /* Reset pinned so btrfs_put_block_group doesn't complain */ | |
10548 | spin_lock(&space_info->lock); | |
10549 | spin_lock(&block_group->lock); | |
10550 | ||
10551 | space_info->bytes_pinned -= block_group->pinned; | |
10552 | space_info->bytes_readonly += block_group->pinned; | |
10553 | percpu_counter_add(&space_info->total_bytes_pinned, | |
10554 | -block_group->pinned); | |
10555 | block_group->pinned = 0; | |
10556 | ||
10557 | spin_unlock(&block_group->lock); | |
10558 | spin_unlock(&space_info->lock); | |
10559 | ||
10560 | /* DISCARD can flip during remount */ | |
10561 | trimming = btrfs_test_opt(root, DISCARD); | |
10562 | ||
10563 | /* Implicit trim during transaction commit. */ | |
10564 | if (trimming) | |
10565 | btrfs_get_block_group_trimming(block_group); | |
10566 | ||
10567 | /* | |
10568 | * Btrfs_remove_chunk will abort the transaction if things go | |
10569 | * horribly wrong. | |
10570 | */ | |
10571 | ret = btrfs_remove_chunk(trans, root, | |
10572 | block_group->key.objectid); | |
10573 | ||
10574 | if (ret) { | |
10575 | if (trimming) | |
10576 | btrfs_put_block_group_trimming(block_group); | |
10577 | goto end_trans; | |
10578 | } | |
10579 | ||
10580 | /* | |
10581 | * If we're not mounted with -odiscard, we can just forget | |
10582 | * about this block group. Otherwise we'll need to wait | |
10583 | * until transaction commit to do the actual discard. | |
10584 | */ | |
10585 | if (trimming) { | |
10586 | spin_lock(&fs_info->unused_bgs_lock); | |
10587 | /* | |
10588 | * A concurrent scrub might have added us to the list | |
10589 | * fs_info->unused_bgs, so use a list_move operation | |
10590 | * to add the block group to the deleted_bgs list. | |
10591 | */ | |
10592 | list_move(&block_group->bg_list, | |
10593 | &trans->transaction->deleted_bgs); | |
10594 | spin_unlock(&fs_info->unused_bgs_lock); | |
10595 | btrfs_get_block_group(block_group); | |
10596 | } | |
10597 | end_trans: | |
10598 | btrfs_end_transaction(trans, root); | |
10599 | next: | |
10600 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
10601 | btrfs_put_block_group(block_group); | |
10602 | spin_lock(&fs_info->unused_bgs_lock); | |
10603 | } | |
10604 | spin_unlock(&fs_info->unused_bgs_lock); | |
10605 | } | |
10606 | ||
10607 | int btrfs_init_space_info(struct btrfs_fs_info *fs_info) | |
10608 | { | |
10609 | struct btrfs_space_info *space_info; | |
10610 | struct btrfs_super_block *disk_super; | |
10611 | u64 features; | |
10612 | u64 flags; | |
10613 | int mixed = 0; | |
10614 | int ret; | |
10615 | ||
10616 | disk_super = fs_info->super_copy; | |
10617 | if (!btrfs_super_root(disk_super)) | |
10618 | return -EINVAL; | |
10619 | ||
10620 | features = btrfs_super_incompat_flags(disk_super); | |
10621 | if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
10622 | mixed = 1; | |
10623 | ||
10624 | flags = BTRFS_BLOCK_GROUP_SYSTEM; | |
10625 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
10626 | if (ret) | |
10627 | goto out; | |
10628 | ||
10629 | if (mixed) { | |
10630 | flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA; | |
10631 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
10632 | } else { | |
10633 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
10634 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
10635 | if (ret) | |
10636 | goto out; | |
10637 | ||
10638 | flags = BTRFS_BLOCK_GROUP_DATA; | |
10639 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
10640 | } | |
10641 | out: | |
10642 | return ret; | |
10643 | } | |
10644 | ||
10645 | int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end) | |
10646 | { | |
10647 | return unpin_extent_range(root, start, end, false); | |
10648 | } | |
10649 | ||
10650 | /* | |
10651 | * It used to be that old block groups would be left around forever. | |
10652 | * Iterating over them would be enough to trim unused space. Since we | |
10653 | * now automatically remove them, we also need to iterate over unallocated | |
10654 | * space. | |
10655 | * | |
10656 | * We don't want a transaction for this since the discard may take a | |
10657 | * substantial amount of time. We don't require that a transaction be | |
10658 | * running, but we do need to take a running transaction into account | |
10659 | * to ensure that we're not discarding chunks that were released in | |
10660 | * the current transaction. | |
10661 | * | |
10662 | * Holding the chunks lock will prevent other threads from allocating | |
10663 | * or releasing chunks, but it won't prevent a running transaction | |
10664 | * from committing and releasing the memory that the pending chunks | |
10665 | * list head uses. For that, we need to take a reference to the | |
10666 | * transaction. | |
10667 | */ | |
10668 | static int btrfs_trim_free_extents(struct btrfs_device *device, | |
10669 | u64 minlen, u64 *trimmed) | |
10670 | { | |
10671 | u64 start = 0, len = 0; | |
10672 | int ret; | |
10673 | ||
10674 | *trimmed = 0; | |
10675 | ||
10676 | /* Not writeable = nothing to do. */ | |
10677 | if (!device->writeable) | |
10678 | return 0; | |
10679 | ||
10680 | /* No free space = nothing to do. */ | |
10681 | if (device->total_bytes <= device->bytes_used) | |
10682 | return 0; | |
10683 | ||
10684 | ret = 0; | |
10685 | ||
10686 | while (1) { | |
10687 | struct btrfs_fs_info *fs_info = device->dev_root->fs_info; | |
10688 | struct btrfs_transaction *trans; | |
10689 | u64 bytes; | |
10690 | ||
10691 | ret = mutex_lock_interruptible(&fs_info->chunk_mutex); | |
10692 | if (ret) | |
10693 | return ret; | |
10694 | ||
10695 | down_read(&fs_info->commit_root_sem); | |
10696 | ||
10697 | spin_lock(&fs_info->trans_lock); | |
10698 | trans = fs_info->running_transaction; | |
10699 | if (trans) | |
10700 | atomic_inc(&trans->use_count); | |
10701 | spin_unlock(&fs_info->trans_lock); | |
10702 | ||
10703 | ret = find_free_dev_extent_start(trans, device, minlen, start, | |
10704 | &start, &len); | |
10705 | if (trans) | |
10706 | btrfs_put_transaction(trans); | |
10707 | ||
10708 | if (ret) { | |
10709 | up_read(&fs_info->commit_root_sem); | |
10710 | mutex_unlock(&fs_info->chunk_mutex); | |
10711 | if (ret == -ENOSPC) | |
10712 | ret = 0; | |
10713 | break; | |
10714 | } | |
10715 | ||
10716 | ret = btrfs_issue_discard(device->bdev, start, len, &bytes); | |
10717 | up_read(&fs_info->commit_root_sem); | |
10718 | mutex_unlock(&fs_info->chunk_mutex); | |
10719 | ||
10720 | if (ret) | |
10721 | break; | |
10722 | ||
10723 | start += len; | |
10724 | *trimmed += bytes; | |
10725 | ||
10726 | if (fatal_signal_pending(current)) { | |
10727 | ret = -ERESTARTSYS; | |
10728 | break; | |
10729 | } | |
10730 | ||
10731 | cond_resched(); | |
10732 | } | |
10733 | ||
10734 | return ret; | |
10735 | } | |
10736 | ||
10737 | int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range) | |
10738 | { | |
10739 | struct btrfs_fs_info *fs_info = root->fs_info; | |
10740 | struct btrfs_block_group_cache *cache = NULL; | |
10741 | struct btrfs_device *device; | |
10742 | struct list_head *devices; | |
10743 | u64 group_trimmed; | |
10744 | u64 start; | |
10745 | u64 end; | |
10746 | u64 trimmed = 0; | |
10747 | u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy); | |
10748 | int ret = 0; | |
10749 | ||
10750 | /* | |
10751 | * try to trim all FS space, our block group may start from non-zero. | |
10752 | */ | |
10753 | if (range->len == total_bytes) | |
10754 | cache = btrfs_lookup_first_block_group(fs_info, range->start); | |
10755 | else | |
10756 | cache = btrfs_lookup_block_group(fs_info, range->start); | |
10757 | ||
10758 | while (cache) { | |
10759 | if (cache->key.objectid >= (range->start + range->len)) { | |
10760 | btrfs_put_block_group(cache); | |
10761 | break; | |
10762 | } | |
10763 | ||
10764 | start = max(range->start, cache->key.objectid); | |
10765 | end = min(range->start + range->len, | |
10766 | cache->key.objectid + cache->key.offset); | |
10767 | ||
10768 | if (end - start >= range->minlen) { | |
10769 | if (!block_group_cache_done(cache)) { | |
10770 | ret = cache_block_group(cache, 0); | |
10771 | if (ret) { | |
10772 | btrfs_put_block_group(cache); | |
10773 | break; | |
10774 | } | |
10775 | ret = wait_block_group_cache_done(cache); | |
10776 | if (ret) { | |
10777 | btrfs_put_block_group(cache); | |
10778 | break; | |
10779 | } | |
10780 | } | |
10781 | ret = btrfs_trim_block_group(cache, | |
10782 | &group_trimmed, | |
10783 | start, | |
10784 | end, | |
10785 | range->minlen); | |
10786 | ||
10787 | trimmed += group_trimmed; | |
10788 | if (ret) { | |
10789 | btrfs_put_block_group(cache); | |
10790 | break; | |
10791 | } | |
10792 | } | |
10793 | ||
10794 | cache = next_block_group(fs_info->tree_root, cache); | |
10795 | } | |
10796 | ||
10797 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); | |
10798 | devices = &root->fs_info->fs_devices->alloc_list; | |
10799 | list_for_each_entry(device, devices, dev_alloc_list) { | |
10800 | ret = btrfs_trim_free_extents(device, range->minlen, | |
10801 | &group_trimmed); | |
10802 | if (ret) | |
10803 | break; | |
10804 | ||
10805 | trimmed += group_trimmed; | |
10806 | } | |
10807 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); | |
10808 | ||
10809 | range->len = trimmed; | |
10810 | return ret; | |
10811 | } | |
10812 | ||
10813 | /* | |
10814 | * btrfs_{start,end}_write_no_snapshoting() are similar to | |
10815 | * mnt_{want,drop}_write(), they are used to prevent some tasks from writing | |
10816 | * data into the page cache through nocow before the subvolume is snapshoted, | |
10817 | * but flush the data into disk after the snapshot creation, or to prevent | |
10818 | * operations while snapshoting is ongoing and that cause the snapshot to be | |
10819 | * inconsistent (writes followed by expanding truncates for example). | |
10820 | */ | |
10821 | void btrfs_end_write_no_snapshoting(struct btrfs_root *root) | |
10822 | { | |
10823 | percpu_counter_dec(&root->subv_writers->counter); | |
10824 | /* | |
10825 | * Make sure counter is updated before we wake up waiters. | |
10826 | */ | |
10827 | smp_mb(); | |
10828 | if (waitqueue_active(&root->subv_writers->wait)) | |
10829 | wake_up(&root->subv_writers->wait); | |
10830 | } | |
10831 | ||
10832 | int btrfs_start_write_no_snapshoting(struct btrfs_root *root) | |
10833 | { | |
10834 | if (atomic_read(&root->will_be_snapshoted)) | |
10835 | return 0; | |
10836 | ||
10837 | percpu_counter_inc(&root->subv_writers->counter); | |
10838 | /* | |
10839 | * Make sure counter is updated before we check for snapshot creation. | |
10840 | */ | |
10841 | smp_mb(); | |
10842 | if (atomic_read(&root->will_be_snapshoted)) { | |
10843 | btrfs_end_write_no_snapshoting(root); | |
10844 | return 0; | |
10845 | } | |
10846 | return 1; | |
10847 | } | |
10848 | ||
10849 | static int wait_snapshoting_atomic_t(atomic_t *a) | |
10850 | { | |
10851 | schedule(); | |
10852 | return 0; | |
10853 | } | |
10854 | ||
10855 | void btrfs_wait_for_snapshot_creation(struct btrfs_root *root) | |
10856 | { | |
10857 | while (true) { | |
10858 | int ret; | |
10859 | ||
10860 | ret = btrfs_start_write_no_snapshoting(root); | |
10861 | if (ret) | |
10862 | break; | |
10863 | wait_on_atomic_t(&root->will_be_snapshoted, | |
10864 | wait_snapshoting_atomic_t, | |
10865 | TASK_UNINTERRUPTIBLE); | |
10866 | } | |
10867 | } |