]>
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 "hash.h" | |
22 | #include "crc32c.h" | |
23 | #include "ctree.h" | |
24 | #include "disk-io.h" | |
25 | #include "print-tree.h" | |
26 | #include "transaction.h" | |
27 | #include "volumes.h" | |
28 | #include "locking.h" | |
29 | #include "ref-cache.h" | |
30 | ||
31 | #define BLOCK_GROUP_DATA EXTENT_WRITEBACK | |
32 | #define BLOCK_GROUP_METADATA EXTENT_UPTODATE | |
33 | #define BLOCK_GROUP_SYSTEM EXTENT_NEW | |
34 | ||
35 | #define BLOCK_GROUP_DIRTY EXTENT_DIRTY | |
36 | ||
37 | static int finish_current_insert(struct btrfs_trans_handle *trans, struct | |
38 | btrfs_root *extent_root); | |
39 | static int del_pending_extents(struct btrfs_trans_handle *trans, struct | |
40 | btrfs_root *extent_root); | |
41 | static struct btrfs_block_group_cache * | |
42 | __btrfs_find_block_group(struct btrfs_root *root, | |
43 | struct btrfs_block_group_cache *hint, | |
44 | u64 search_start, int data, int owner); | |
45 | ||
46 | void maybe_lock_mutex(struct btrfs_root *root) | |
47 | { | |
48 | if (root != root->fs_info->extent_root && | |
49 | root != root->fs_info->chunk_root && | |
50 | root != root->fs_info->dev_root) { | |
51 | mutex_lock(&root->fs_info->alloc_mutex); | |
52 | } | |
53 | } | |
54 | ||
55 | void maybe_unlock_mutex(struct btrfs_root *root) | |
56 | { | |
57 | if (root != root->fs_info->extent_root && | |
58 | root != root->fs_info->chunk_root && | |
59 | root != root->fs_info->dev_root) { | |
60 | mutex_unlock(&root->fs_info->alloc_mutex); | |
61 | } | |
62 | } | |
63 | ||
64 | static int cache_block_group(struct btrfs_root *root, | |
65 | struct btrfs_block_group_cache *block_group) | |
66 | { | |
67 | struct btrfs_path *path; | |
68 | int ret; | |
69 | struct btrfs_key key; | |
70 | struct extent_buffer *leaf; | |
71 | struct extent_io_tree *free_space_cache; | |
72 | int slot; | |
73 | u64 last = 0; | |
74 | u64 hole_size; | |
75 | u64 first_free; | |
76 | int found = 0; | |
77 | ||
78 | if (!block_group) | |
79 | return 0; | |
80 | ||
81 | root = root->fs_info->extent_root; | |
82 | free_space_cache = &root->fs_info->free_space_cache; | |
83 | ||
84 | if (block_group->cached) | |
85 | return 0; | |
86 | ||
87 | path = btrfs_alloc_path(); | |
88 | if (!path) | |
89 | return -ENOMEM; | |
90 | ||
91 | path->reada = 2; | |
92 | /* | |
93 | * we get into deadlocks with paths held by callers of this function. | |
94 | * since the alloc_mutex is protecting things right now, just | |
95 | * skip the locking here | |
96 | */ | |
97 | path->skip_locking = 1; | |
98 | first_free = block_group->key.objectid; | |
99 | key.objectid = block_group->key.objectid; | |
100 | key.offset = 0; | |
101 | btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); | |
102 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
103 | if (ret < 0) | |
104 | return ret; | |
105 | ret = btrfs_previous_item(root, path, 0, BTRFS_EXTENT_ITEM_KEY); | |
106 | if (ret < 0) | |
107 | return ret; | |
108 | if (ret == 0) { | |
109 | leaf = path->nodes[0]; | |
110 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
111 | if (key.objectid + key.offset > first_free) | |
112 | first_free = key.objectid + key.offset; | |
113 | } | |
114 | while(1) { | |
115 | leaf = path->nodes[0]; | |
116 | slot = path->slots[0]; | |
117 | if (slot >= btrfs_header_nritems(leaf)) { | |
118 | ret = btrfs_next_leaf(root, path); | |
119 | if (ret < 0) | |
120 | goto err; | |
121 | if (ret == 0) { | |
122 | continue; | |
123 | } else { | |
124 | break; | |
125 | } | |
126 | } | |
127 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
128 | if (key.objectid < block_group->key.objectid) { | |
129 | goto next; | |
130 | } | |
131 | if (key.objectid >= block_group->key.objectid + | |
132 | block_group->key.offset) { | |
133 | break; | |
134 | } | |
135 | ||
136 | if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) { | |
137 | if (!found) { | |
138 | last = first_free; | |
139 | found = 1; | |
140 | } | |
141 | if (key.objectid > last) { | |
142 | hole_size = key.objectid - last; | |
143 | set_extent_dirty(free_space_cache, last, | |
144 | last + hole_size - 1, | |
145 | GFP_NOFS); | |
146 | } | |
147 | last = key.objectid + key.offset; | |
148 | } | |
149 | next: | |
150 | path->slots[0]++; | |
151 | } | |
152 | ||
153 | if (!found) | |
154 | last = first_free; | |
155 | if (block_group->key.objectid + | |
156 | block_group->key.offset > last) { | |
157 | hole_size = block_group->key.objectid + | |
158 | block_group->key.offset - last; | |
159 | set_extent_dirty(free_space_cache, last, | |
160 | last + hole_size - 1, GFP_NOFS); | |
161 | } | |
162 | block_group->cached = 1; | |
163 | err: | |
164 | btrfs_free_path(path); | |
165 | return 0; | |
166 | } | |
167 | ||
168 | struct btrfs_block_group_cache *btrfs_lookup_first_block_group(struct | |
169 | btrfs_fs_info *info, | |
170 | u64 bytenr) | |
171 | { | |
172 | struct extent_io_tree *block_group_cache; | |
173 | struct btrfs_block_group_cache *block_group = NULL; | |
174 | u64 ptr; | |
175 | u64 start; | |
176 | u64 end; | |
177 | int ret; | |
178 | ||
179 | bytenr = max_t(u64, bytenr, | |
180 | BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE); | |
181 | block_group_cache = &info->block_group_cache; | |
182 | ret = find_first_extent_bit(block_group_cache, | |
183 | bytenr, &start, &end, | |
184 | BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA | | |
185 | BLOCK_GROUP_SYSTEM); | |
186 | if (ret) { | |
187 | return NULL; | |
188 | } | |
189 | ret = get_state_private(block_group_cache, start, &ptr); | |
190 | if (ret) | |
191 | return NULL; | |
192 | ||
193 | block_group = (struct btrfs_block_group_cache *)(unsigned long)ptr; | |
194 | return block_group; | |
195 | } | |
196 | ||
197 | struct btrfs_block_group_cache *btrfs_lookup_block_group(struct | |
198 | btrfs_fs_info *info, | |
199 | u64 bytenr) | |
200 | { | |
201 | struct extent_io_tree *block_group_cache; | |
202 | struct btrfs_block_group_cache *block_group = NULL; | |
203 | u64 ptr; | |
204 | u64 start; | |
205 | u64 end; | |
206 | int ret; | |
207 | ||
208 | bytenr = max_t(u64, bytenr, | |
209 | BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE); | |
210 | block_group_cache = &info->block_group_cache; | |
211 | ret = find_first_extent_bit(block_group_cache, | |
212 | bytenr, &start, &end, | |
213 | BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA | | |
214 | BLOCK_GROUP_SYSTEM); | |
215 | if (ret) { | |
216 | return NULL; | |
217 | } | |
218 | ret = get_state_private(block_group_cache, start, &ptr); | |
219 | if (ret) | |
220 | return NULL; | |
221 | ||
222 | block_group = (struct btrfs_block_group_cache *)(unsigned long)ptr; | |
223 | if (block_group->key.objectid <= bytenr && bytenr < | |
224 | block_group->key.objectid + block_group->key.offset) | |
225 | return block_group; | |
226 | return NULL; | |
227 | } | |
228 | ||
229 | static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits) | |
230 | { | |
231 | return (cache->flags & bits) == bits; | |
232 | } | |
233 | ||
234 | static int noinline find_search_start(struct btrfs_root *root, | |
235 | struct btrfs_block_group_cache **cache_ret, | |
236 | u64 *start_ret, u64 num, int data) | |
237 | { | |
238 | int ret; | |
239 | struct btrfs_block_group_cache *cache = *cache_ret; | |
240 | struct extent_io_tree *free_space_cache; | |
241 | struct extent_state *state; | |
242 | u64 last; | |
243 | u64 start = 0; | |
244 | u64 cache_miss = 0; | |
245 | u64 total_fs_bytes; | |
246 | u64 search_start = *start_ret; | |
247 | int wrapped = 0; | |
248 | ||
249 | WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex)); | |
250 | total_fs_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy); | |
251 | free_space_cache = &root->fs_info->free_space_cache; | |
252 | ||
253 | if (!cache) | |
254 | goto out; | |
255 | ||
256 | again: | |
257 | ret = cache_block_group(root, cache); | |
258 | if (ret) { | |
259 | goto out; | |
260 | } | |
261 | ||
262 | last = max(search_start, cache->key.objectid); | |
263 | if (!block_group_bits(cache, data) || cache->ro) | |
264 | goto new_group; | |
265 | ||
266 | spin_lock_irq(&free_space_cache->lock); | |
267 | state = find_first_extent_bit_state(free_space_cache, last, EXTENT_DIRTY); | |
268 | while(1) { | |
269 | if (!state) { | |
270 | if (!cache_miss) | |
271 | cache_miss = last; | |
272 | spin_unlock_irq(&free_space_cache->lock); | |
273 | goto new_group; | |
274 | } | |
275 | ||
276 | start = max(last, state->start); | |
277 | last = state->end + 1; | |
278 | if (last - start < num) { | |
279 | do { | |
280 | state = extent_state_next(state); | |
281 | } while(state && !(state->state & EXTENT_DIRTY)); | |
282 | continue; | |
283 | } | |
284 | spin_unlock_irq(&free_space_cache->lock); | |
285 | if (cache->ro) { | |
286 | goto new_group; | |
287 | } | |
288 | if (start + num > cache->key.objectid + cache->key.offset) | |
289 | goto new_group; | |
290 | if (!block_group_bits(cache, data)) { | |
291 | printk("block group bits don't match %Lu %d\n", cache->flags, data); | |
292 | } | |
293 | *start_ret = start; | |
294 | return 0; | |
295 | } | |
296 | out: | |
297 | cache = btrfs_lookup_block_group(root->fs_info, search_start); | |
298 | if (!cache) { | |
299 | printk("Unable to find block group for %Lu\n", search_start); | |
300 | WARN_ON(1); | |
301 | } | |
302 | return -ENOSPC; | |
303 | ||
304 | new_group: | |
305 | last = cache->key.objectid + cache->key.offset; | |
306 | wrapped: | |
307 | cache = btrfs_lookup_first_block_group(root->fs_info, last); | |
308 | if (!cache || cache->key.objectid >= total_fs_bytes) { | |
309 | no_cache: | |
310 | if (!wrapped) { | |
311 | wrapped = 1; | |
312 | last = search_start; | |
313 | goto wrapped; | |
314 | } | |
315 | goto out; | |
316 | } | |
317 | if (cache_miss && !cache->cached) { | |
318 | cache_block_group(root, cache); | |
319 | last = cache_miss; | |
320 | cache = btrfs_lookup_first_block_group(root->fs_info, last); | |
321 | } | |
322 | cache_miss = 0; | |
323 | cache = btrfs_find_block_group(root, cache, last, data, 0); | |
324 | if (!cache) | |
325 | goto no_cache; | |
326 | *cache_ret = cache; | |
327 | goto again; | |
328 | } | |
329 | ||
330 | static u64 div_factor(u64 num, int factor) | |
331 | { | |
332 | if (factor == 10) | |
333 | return num; | |
334 | num *= factor; | |
335 | do_div(num, 10); | |
336 | return num; | |
337 | } | |
338 | ||
339 | static int block_group_state_bits(u64 flags) | |
340 | { | |
341 | int bits = 0; | |
342 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
343 | bits |= BLOCK_GROUP_DATA; | |
344 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
345 | bits |= BLOCK_GROUP_METADATA; | |
346 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
347 | bits |= BLOCK_GROUP_SYSTEM; | |
348 | return bits; | |
349 | } | |
350 | ||
351 | static struct btrfs_block_group_cache * | |
352 | __btrfs_find_block_group(struct btrfs_root *root, | |
353 | struct btrfs_block_group_cache *hint, | |
354 | u64 search_start, int data, int owner) | |
355 | { | |
356 | struct btrfs_block_group_cache *cache; | |
357 | struct extent_io_tree *block_group_cache; | |
358 | struct btrfs_block_group_cache *found_group = NULL; | |
359 | struct btrfs_fs_info *info = root->fs_info; | |
360 | u64 used; | |
361 | u64 last = 0; | |
362 | u64 start; | |
363 | u64 end; | |
364 | u64 free_check; | |
365 | u64 ptr; | |
366 | int bit; | |
367 | int ret; | |
368 | int full_search = 0; | |
369 | int factor = 10; | |
370 | int wrapped = 0; | |
371 | ||
372 | block_group_cache = &info->block_group_cache; | |
373 | ||
374 | if (data & BTRFS_BLOCK_GROUP_METADATA) | |
375 | factor = 9; | |
376 | ||
377 | bit = block_group_state_bits(data); | |
378 | ||
379 | if (search_start) { | |
380 | struct btrfs_block_group_cache *shint; | |
381 | shint = btrfs_lookup_first_block_group(info, search_start); | |
382 | if (shint && block_group_bits(shint, data) && !shint->ro) { | |
383 | spin_lock(&shint->lock); | |
384 | used = btrfs_block_group_used(&shint->item); | |
385 | if (used + shint->pinned < | |
386 | div_factor(shint->key.offset, factor)) { | |
387 | spin_unlock(&shint->lock); | |
388 | return shint; | |
389 | } | |
390 | spin_unlock(&shint->lock); | |
391 | } | |
392 | } | |
393 | if (hint && !hint->ro && block_group_bits(hint, data)) { | |
394 | spin_lock(&hint->lock); | |
395 | used = btrfs_block_group_used(&hint->item); | |
396 | if (used + hint->pinned < | |
397 | div_factor(hint->key.offset, factor)) { | |
398 | spin_unlock(&hint->lock); | |
399 | return hint; | |
400 | } | |
401 | spin_unlock(&hint->lock); | |
402 | last = hint->key.objectid + hint->key.offset; | |
403 | } else { | |
404 | if (hint) | |
405 | last = max(hint->key.objectid, search_start); | |
406 | else | |
407 | last = search_start; | |
408 | } | |
409 | again: | |
410 | while(1) { | |
411 | ret = find_first_extent_bit(block_group_cache, last, | |
412 | &start, &end, bit); | |
413 | if (ret) | |
414 | break; | |
415 | ||
416 | ret = get_state_private(block_group_cache, start, &ptr); | |
417 | if (ret) { | |
418 | last = end + 1; | |
419 | continue; | |
420 | } | |
421 | ||
422 | cache = (struct btrfs_block_group_cache *)(unsigned long)ptr; | |
423 | spin_lock(&cache->lock); | |
424 | last = cache->key.objectid + cache->key.offset; | |
425 | used = btrfs_block_group_used(&cache->item); | |
426 | ||
427 | if (!cache->ro && block_group_bits(cache, data)) { | |
428 | free_check = div_factor(cache->key.offset, factor); | |
429 | if (used + cache->pinned < free_check) { | |
430 | found_group = cache; | |
431 | spin_unlock(&cache->lock); | |
432 | goto found; | |
433 | } | |
434 | } | |
435 | spin_unlock(&cache->lock); | |
436 | cond_resched(); | |
437 | } | |
438 | if (!wrapped) { | |
439 | last = search_start; | |
440 | wrapped = 1; | |
441 | goto again; | |
442 | } | |
443 | if (!full_search && factor < 10) { | |
444 | last = search_start; | |
445 | full_search = 1; | |
446 | factor = 10; | |
447 | goto again; | |
448 | } | |
449 | found: | |
450 | return found_group; | |
451 | } | |
452 | ||
453 | struct btrfs_block_group_cache *btrfs_find_block_group(struct btrfs_root *root, | |
454 | struct btrfs_block_group_cache | |
455 | *hint, u64 search_start, | |
456 | int data, int owner) | |
457 | { | |
458 | ||
459 | struct btrfs_block_group_cache *ret; | |
460 | ret = __btrfs_find_block_group(root, hint, search_start, data, owner); | |
461 | return ret; | |
462 | } | |
463 | static u64 hash_extent_ref(u64 root_objectid, u64 ref_generation, | |
464 | u64 owner, u64 owner_offset) | |
465 | { | |
466 | u32 high_crc = ~(u32)0; | |
467 | u32 low_crc = ~(u32)0; | |
468 | __le64 lenum; | |
469 | lenum = cpu_to_le64(root_objectid); | |
470 | high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum)); | |
471 | lenum = cpu_to_le64(ref_generation); | |
472 | low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum)); | |
473 | if (owner >= BTRFS_FIRST_FREE_OBJECTID) { | |
474 | lenum = cpu_to_le64(owner); | |
475 | low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum)); | |
476 | lenum = cpu_to_le64(owner_offset); | |
477 | low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum)); | |
478 | } | |
479 | return ((u64)high_crc << 32) | (u64)low_crc; | |
480 | } | |
481 | ||
482 | static int match_extent_ref(struct extent_buffer *leaf, | |
483 | struct btrfs_extent_ref *disk_ref, | |
484 | struct btrfs_extent_ref *cpu_ref) | |
485 | { | |
486 | int ret; | |
487 | int len; | |
488 | ||
489 | if (cpu_ref->objectid) | |
490 | len = sizeof(*cpu_ref); | |
491 | else | |
492 | len = 2 * sizeof(u64); | |
493 | ret = memcmp_extent_buffer(leaf, cpu_ref, (unsigned long)disk_ref, | |
494 | len); | |
495 | return ret == 0; | |
496 | } | |
497 | ||
498 | static int noinline lookup_extent_backref(struct btrfs_trans_handle *trans, | |
499 | struct btrfs_root *root, | |
500 | struct btrfs_path *path, u64 bytenr, | |
501 | u64 root_objectid, | |
502 | u64 ref_generation, u64 owner, | |
503 | u64 owner_offset, int del) | |
504 | { | |
505 | u64 hash; | |
506 | struct btrfs_key key; | |
507 | struct btrfs_key found_key; | |
508 | struct btrfs_extent_ref ref; | |
509 | struct extent_buffer *leaf; | |
510 | struct btrfs_extent_ref *disk_ref; | |
511 | int ret; | |
512 | int ret2; | |
513 | ||
514 | btrfs_set_stack_ref_root(&ref, root_objectid); | |
515 | btrfs_set_stack_ref_generation(&ref, ref_generation); | |
516 | btrfs_set_stack_ref_objectid(&ref, owner); | |
517 | btrfs_set_stack_ref_offset(&ref, owner_offset); | |
518 | ||
519 | hash = hash_extent_ref(root_objectid, ref_generation, owner, | |
520 | owner_offset); | |
521 | key.offset = hash; | |
522 | key.objectid = bytenr; | |
523 | key.type = BTRFS_EXTENT_REF_KEY; | |
524 | ||
525 | while (1) { | |
526 | ret = btrfs_search_slot(trans, root, &key, path, | |
527 | del ? -1 : 0, del); | |
528 | if (ret < 0) | |
529 | goto out; | |
530 | leaf = path->nodes[0]; | |
531 | if (ret != 0) { | |
532 | u32 nritems = btrfs_header_nritems(leaf); | |
533 | if (path->slots[0] >= nritems) { | |
534 | ret2 = btrfs_next_leaf(root, path); | |
535 | if (ret2) | |
536 | goto out; | |
537 | leaf = path->nodes[0]; | |
538 | } | |
539 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
540 | if (found_key.objectid != bytenr || | |
541 | found_key.type != BTRFS_EXTENT_REF_KEY) | |
542 | goto out; | |
543 | key.offset = found_key.offset; | |
544 | if (del) { | |
545 | btrfs_release_path(root, path); | |
546 | continue; | |
547 | } | |
548 | } | |
549 | disk_ref = btrfs_item_ptr(path->nodes[0], | |
550 | path->slots[0], | |
551 | struct btrfs_extent_ref); | |
552 | if (match_extent_ref(path->nodes[0], disk_ref, &ref)) { | |
553 | ret = 0; | |
554 | goto out; | |
555 | } | |
556 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
557 | key.offset = found_key.offset + 1; | |
558 | btrfs_release_path(root, path); | |
559 | } | |
560 | out: | |
561 | return ret; | |
562 | } | |
563 | ||
564 | /* | |
565 | * Back reference rules. Back refs have three main goals: | |
566 | * | |
567 | * 1) differentiate between all holders of references to an extent so that | |
568 | * when a reference is dropped we can make sure it was a valid reference | |
569 | * before freeing the extent. | |
570 | * | |
571 | * 2) Provide enough information to quickly find the holders of an extent | |
572 | * if we notice a given block is corrupted or bad. | |
573 | * | |
574 | * 3) Make it easy to migrate blocks for FS shrinking or storage pool | |
575 | * maintenance. This is actually the same as #2, but with a slightly | |
576 | * different use case. | |
577 | * | |
578 | * File extents can be referenced by: | |
579 | * | |
580 | * - multiple snapshots, subvolumes, or different generations in one subvol | |
581 | * - different files inside a single subvolume (in theory, not implemented yet) | |
582 | * - different offsets inside a file (bookend extents in file.c) | |
583 | * | |
584 | * The extent ref structure has fields for: | |
585 | * | |
586 | * - Objectid of the subvolume root | |
587 | * - Generation number of the tree holding the reference | |
588 | * - objectid of the file holding the reference | |
589 | * - offset in the file corresponding to the key holding the reference | |
590 | * | |
591 | * When a file extent is allocated the fields are filled in: | |
592 | * (root_key.objectid, trans->transid, inode objectid, offset in file) | |
593 | * | |
594 | * When a leaf is cow'd new references are added for every file extent found | |
595 | * in the leaf. It looks the same as the create case, but trans->transid | |
596 | * will be different when the block is cow'd. | |
597 | * | |
598 | * (root_key.objectid, trans->transid, inode objectid, offset in file) | |
599 | * | |
600 | * When a file extent is removed either during snapshot deletion or file | |
601 | * truncation, the corresponding back reference is found | |
602 | * by searching for: | |
603 | * | |
604 | * (btrfs_header_owner(leaf), btrfs_header_generation(leaf), | |
605 | * inode objectid, offset in file) | |
606 | * | |
607 | * Btree extents can be referenced by: | |
608 | * | |
609 | * - Different subvolumes | |
610 | * - Different generations of the same subvolume | |
611 | * | |
612 | * Storing sufficient information for a full reverse mapping of a btree | |
613 | * block would require storing the lowest key of the block in the backref, | |
614 | * and it would require updating that lowest key either before write out or | |
615 | * every time it changed. Instead, the objectid of the lowest key is stored | |
616 | * along with the level of the tree block. This provides a hint | |
617 | * about where in the btree the block can be found. Searches through the | |
618 | * btree only need to look for a pointer to that block, so they stop one | |
619 | * level higher than the level recorded in the backref. | |
620 | * | |
621 | * Some btrees do not do reference counting on their extents. These | |
622 | * include the extent tree and the tree of tree roots. Backrefs for these | |
623 | * trees always have a generation of zero. | |
624 | * | |
625 | * When a tree block is created, back references are inserted: | |
626 | * | |
627 | * (root->root_key.objectid, trans->transid or zero, level, lowest_key_objectid) | |
628 | * | |
629 | * When a tree block is cow'd in a reference counted root, | |
630 | * new back references are added for all the blocks it points to. | |
631 | * These are of the form (trans->transid will have increased since creation): | |
632 | * | |
633 | * (root->root_key.objectid, trans->transid, level, lowest_key_objectid) | |
634 | * | |
635 | * Because the lowest_key_objectid and the level are just hints | |
636 | * they are not used when backrefs are deleted. When a backref is deleted: | |
637 | * | |
638 | * if backref was for a tree root: | |
639 | * root_objectid = root->root_key.objectid | |
640 | * else | |
641 | * root_objectid = btrfs_header_owner(parent) | |
642 | * | |
643 | * (root_objectid, btrfs_header_generation(parent) or zero, 0, 0) | |
644 | * | |
645 | * Back Reference Key hashing: | |
646 | * | |
647 | * Back references have four fields, each 64 bits long. Unfortunately, | |
648 | * This is hashed into a single 64 bit number and placed into the key offset. | |
649 | * The key objectid corresponds to the first byte in the extent, and the | |
650 | * key type is set to BTRFS_EXTENT_REF_KEY | |
651 | */ | |
652 | int btrfs_insert_extent_backref(struct btrfs_trans_handle *trans, | |
653 | struct btrfs_root *root, | |
654 | struct btrfs_path *path, u64 bytenr, | |
655 | u64 root_objectid, u64 ref_generation, | |
656 | u64 owner, u64 owner_offset) | |
657 | { | |
658 | u64 hash; | |
659 | struct btrfs_key key; | |
660 | struct btrfs_extent_ref ref; | |
661 | struct btrfs_extent_ref *disk_ref; | |
662 | int ret; | |
663 | ||
664 | btrfs_set_stack_ref_root(&ref, root_objectid); | |
665 | btrfs_set_stack_ref_generation(&ref, ref_generation); | |
666 | btrfs_set_stack_ref_objectid(&ref, owner); | |
667 | btrfs_set_stack_ref_offset(&ref, owner_offset); | |
668 | ||
669 | hash = hash_extent_ref(root_objectid, ref_generation, owner, | |
670 | owner_offset); | |
671 | key.offset = hash; | |
672 | key.objectid = bytenr; | |
673 | key.type = BTRFS_EXTENT_REF_KEY; | |
674 | ||
675 | ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(ref)); | |
676 | while (ret == -EEXIST) { | |
677 | disk_ref = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
678 | struct btrfs_extent_ref); | |
679 | if (match_extent_ref(path->nodes[0], disk_ref, &ref)) | |
680 | goto out; | |
681 | key.offset++; | |
682 | btrfs_release_path(root, path); | |
683 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
684 | sizeof(ref)); | |
685 | } | |
686 | if (ret) | |
687 | goto out; | |
688 | disk_ref = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
689 | struct btrfs_extent_ref); | |
690 | write_extent_buffer(path->nodes[0], &ref, (unsigned long)disk_ref, | |
691 | sizeof(ref)); | |
692 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
693 | out: | |
694 | btrfs_release_path(root, path); | |
695 | return ret; | |
696 | } | |
697 | ||
698 | static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, | |
699 | struct btrfs_root *root, | |
700 | u64 bytenr, u64 num_bytes, | |
701 | u64 root_objectid, u64 ref_generation, | |
702 | u64 owner, u64 owner_offset) | |
703 | { | |
704 | struct btrfs_path *path; | |
705 | int ret; | |
706 | struct btrfs_key key; | |
707 | struct extent_buffer *l; | |
708 | struct btrfs_extent_item *item; | |
709 | u32 refs; | |
710 | ||
711 | WARN_ON(num_bytes < root->sectorsize); | |
712 | path = btrfs_alloc_path(); | |
713 | if (!path) | |
714 | return -ENOMEM; | |
715 | ||
716 | path->reada = 1; | |
717 | key.objectid = bytenr; | |
718 | btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); | |
719 | key.offset = num_bytes; | |
720 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path, | |
721 | 0, 1); | |
722 | if (ret < 0) | |
723 | return ret; | |
724 | if (ret != 0) { | |
725 | BUG(); | |
726 | } | |
727 | BUG_ON(ret != 0); | |
728 | l = path->nodes[0]; | |
729 | item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item); | |
730 | refs = btrfs_extent_refs(l, item); | |
731 | btrfs_set_extent_refs(l, item, refs + 1); | |
732 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
733 | ||
734 | btrfs_release_path(root->fs_info->extent_root, path); | |
735 | ||
736 | path->reada = 1; | |
737 | ret = btrfs_insert_extent_backref(trans, root->fs_info->extent_root, | |
738 | path, bytenr, root_objectid, | |
739 | ref_generation, owner, owner_offset); | |
740 | BUG_ON(ret); | |
741 | finish_current_insert(trans, root->fs_info->extent_root); | |
742 | del_pending_extents(trans, root->fs_info->extent_root); | |
743 | ||
744 | btrfs_free_path(path); | |
745 | return 0; | |
746 | } | |
747 | ||
748 | int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, | |
749 | struct btrfs_root *root, | |
750 | u64 bytenr, u64 num_bytes, | |
751 | u64 root_objectid, u64 ref_generation, | |
752 | u64 owner, u64 owner_offset) | |
753 | { | |
754 | int ret; | |
755 | ||
756 | mutex_lock(&root->fs_info->alloc_mutex); | |
757 | ret = __btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, | |
758 | root_objectid, ref_generation, | |
759 | owner, owner_offset); | |
760 | mutex_unlock(&root->fs_info->alloc_mutex); | |
761 | return ret; | |
762 | } | |
763 | ||
764 | int btrfs_extent_post_op(struct btrfs_trans_handle *trans, | |
765 | struct btrfs_root *root) | |
766 | { | |
767 | finish_current_insert(trans, root->fs_info->extent_root); | |
768 | del_pending_extents(trans, root->fs_info->extent_root); | |
769 | return 0; | |
770 | } | |
771 | ||
772 | static int lookup_extent_ref(struct btrfs_trans_handle *trans, | |
773 | struct btrfs_root *root, u64 bytenr, | |
774 | u64 num_bytes, u32 *refs) | |
775 | { | |
776 | struct btrfs_path *path; | |
777 | int ret; | |
778 | struct btrfs_key key; | |
779 | struct extent_buffer *l; | |
780 | struct btrfs_extent_item *item; | |
781 | ||
782 | WARN_ON(num_bytes < root->sectorsize); | |
783 | path = btrfs_alloc_path(); | |
784 | path->reada = 1; | |
785 | key.objectid = bytenr; | |
786 | key.offset = num_bytes; | |
787 | btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); | |
788 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path, | |
789 | 0, 0); | |
790 | if (ret < 0) | |
791 | goto out; | |
792 | if (ret != 0) { | |
793 | btrfs_print_leaf(root, path->nodes[0]); | |
794 | printk("failed to find block number %Lu\n", bytenr); | |
795 | BUG(); | |
796 | } | |
797 | l = path->nodes[0]; | |
798 | item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item); | |
799 | *refs = btrfs_extent_refs(l, item); | |
800 | out: | |
801 | btrfs_free_path(path); | |
802 | return 0; | |
803 | } | |
804 | ||
805 | ||
806 | static int get_reference_status(struct btrfs_root *root, u64 bytenr, | |
807 | u64 parent_gen, u64 ref_objectid, | |
808 | u64 *min_generation, u32 *ref_count) | |
809 | { | |
810 | struct btrfs_root *extent_root = root->fs_info->extent_root; | |
811 | struct btrfs_path *path; | |
812 | struct extent_buffer *leaf; | |
813 | struct btrfs_extent_ref *ref_item; | |
814 | struct btrfs_key key; | |
815 | struct btrfs_key found_key; | |
816 | u64 root_objectid = root->root_key.objectid; | |
817 | u64 ref_generation; | |
818 | u32 nritems; | |
819 | int ret; | |
820 | ||
821 | key.objectid = bytenr; | |
822 | key.offset = 0; | |
823 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
824 | ||
825 | path = btrfs_alloc_path(); | |
826 | mutex_lock(&root->fs_info->alloc_mutex); | |
827 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); | |
828 | if (ret < 0) | |
829 | goto out; | |
830 | BUG_ON(ret == 0); | |
831 | ||
832 | leaf = path->nodes[0]; | |
833 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
834 | ||
835 | if (found_key.objectid != bytenr || | |
836 | found_key.type != BTRFS_EXTENT_ITEM_KEY) { | |
837 | ret = 1; | |
838 | goto out; | |
839 | } | |
840 | ||
841 | *ref_count = 0; | |
842 | *min_generation = (u64)-1; | |
843 | ||
844 | while (1) { | |
845 | leaf = path->nodes[0]; | |
846 | nritems = btrfs_header_nritems(leaf); | |
847 | if (path->slots[0] >= nritems) { | |
848 | ret = btrfs_next_leaf(extent_root, path); | |
849 | if (ret < 0) | |
850 | goto out; | |
851 | if (ret == 0) | |
852 | continue; | |
853 | break; | |
854 | } | |
855 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
856 | if (found_key.objectid != bytenr) | |
857 | break; | |
858 | ||
859 | if (found_key.type != BTRFS_EXTENT_REF_KEY) { | |
860 | path->slots[0]++; | |
861 | continue; | |
862 | } | |
863 | ||
864 | ref_item = btrfs_item_ptr(leaf, path->slots[0], | |
865 | struct btrfs_extent_ref); | |
866 | ref_generation = btrfs_ref_generation(leaf, ref_item); | |
867 | /* | |
868 | * For (parent_gen > 0 && parent_gen > ref_gen): | |
869 | * | |
870 | * we reach here through the oldest root, therefore | |
871 | * all other reference from same snapshot should have | |
872 | * a larger generation. | |
873 | */ | |
874 | if ((root_objectid != btrfs_ref_root(leaf, ref_item)) || | |
875 | (parent_gen > 0 && parent_gen > ref_generation) || | |
876 | (ref_objectid >= BTRFS_FIRST_FREE_OBJECTID && | |
877 | ref_objectid != btrfs_ref_objectid(leaf, ref_item))) { | |
878 | if (ref_count) | |
879 | *ref_count = 2; | |
880 | break; | |
881 | } | |
882 | ||
883 | *ref_count = 1; | |
884 | if (*min_generation > ref_generation) | |
885 | *min_generation = ref_generation; | |
886 | ||
887 | path->slots[0]++; | |
888 | } | |
889 | ret = 0; | |
890 | out: | |
891 | mutex_unlock(&root->fs_info->alloc_mutex); | |
892 | btrfs_free_path(path); | |
893 | return ret; | |
894 | } | |
895 | ||
896 | int btrfs_cross_ref_exists(struct btrfs_root *root, | |
897 | struct btrfs_key *key, u64 bytenr) | |
898 | { | |
899 | struct btrfs_trans_handle *trans; | |
900 | struct btrfs_root *old_root; | |
901 | struct btrfs_path *path = NULL; | |
902 | struct extent_buffer *eb; | |
903 | struct btrfs_file_extent_item *item; | |
904 | u64 ref_generation; | |
905 | u64 min_generation; | |
906 | u64 extent_start; | |
907 | u32 ref_count; | |
908 | int level; | |
909 | int ret; | |
910 | ||
911 | BUG_ON(key->type != BTRFS_EXTENT_DATA_KEY); | |
912 | ret = get_reference_status(root, bytenr, 0, key->objectid, | |
913 | &min_generation, &ref_count); | |
914 | if (ret) | |
915 | return ret; | |
916 | ||
917 | if (ref_count != 1) | |
918 | return 1; | |
919 | ||
920 | trans = btrfs_start_transaction(root, 0); | |
921 | old_root = root->dirty_root->root; | |
922 | ref_generation = old_root->root_key.offset; | |
923 | ||
924 | /* all references are created in running transaction */ | |
925 | if (min_generation > ref_generation) { | |
926 | ret = 0; | |
927 | goto out; | |
928 | } | |
929 | ||
930 | path = btrfs_alloc_path(); | |
931 | if (!path) { | |
932 | ret = -ENOMEM; | |
933 | goto out; | |
934 | } | |
935 | ||
936 | path->skip_locking = 1; | |
937 | /* if no item found, the extent is referenced by other snapshot */ | |
938 | ret = btrfs_search_slot(NULL, old_root, key, path, 0, 0); | |
939 | if (ret) | |
940 | goto out; | |
941 | ||
942 | eb = path->nodes[0]; | |
943 | item = btrfs_item_ptr(eb, path->slots[0], | |
944 | struct btrfs_file_extent_item); | |
945 | if (btrfs_file_extent_type(eb, item) != BTRFS_FILE_EXTENT_REG || | |
946 | btrfs_file_extent_disk_bytenr(eb, item) != bytenr) { | |
947 | ret = 1; | |
948 | goto out; | |
949 | } | |
950 | ||
951 | for (level = BTRFS_MAX_LEVEL - 1; level >= -1; level--) { | |
952 | if (level >= 0) { | |
953 | eb = path->nodes[level]; | |
954 | if (!eb) | |
955 | continue; | |
956 | extent_start = eb->start; | |
957 | } else | |
958 | extent_start = bytenr; | |
959 | ||
960 | ret = get_reference_status(root, extent_start, ref_generation, | |
961 | 0, &min_generation, &ref_count); | |
962 | if (ret) | |
963 | goto out; | |
964 | ||
965 | if (ref_count != 1) { | |
966 | ret = 1; | |
967 | goto out; | |
968 | } | |
969 | if (level >= 0) | |
970 | ref_generation = btrfs_header_generation(eb); | |
971 | } | |
972 | ret = 0; | |
973 | out: | |
974 | if (path) | |
975 | btrfs_free_path(path); | |
976 | btrfs_end_transaction(trans, root); | |
977 | return ret; | |
978 | } | |
979 | ||
980 | int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
981 | struct extent_buffer *buf, int cache_ref) | |
982 | { | |
983 | u64 bytenr; | |
984 | u32 nritems; | |
985 | struct btrfs_key key; | |
986 | struct btrfs_file_extent_item *fi; | |
987 | int i; | |
988 | int level; | |
989 | int ret; | |
990 | int faili; | |
991 | int nr_file_extents = 0; | |
992 | ||
993 | if (!root->ref_cows) | |
994 | return 0; | |
995 | ||
996 | level = btrfs_header_level(buf); | |
997 | nritems = btrfs_header_nritems(buf); | |
998 | for (i = 0; i < nritems; i++) { | |
999 | cond_resched(); | |
1000 | if (level == 0) { | |
1001 | u64 disk_bytenr; | |
1002 | btrfs_item_key_to_cpu(buf, &key, i); | |
1003 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) | |
1004 | continue; | |
1005 | fi = btrfs_item_ptr(buf, i, | |
1006 | struct btrfs_file_extent_item); | |
1007 | if (btrfs_file_extent_type(buf, fi) == | |
1008 | BTRFS_FILE_EXTENT_INLINE) | |
1009 | continue; | |
1010 | disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
1011 | if (disk_bytenr == 0) | |
1012 | continue; | |
1013 | ||
1014 | if (buf != root->commit_root) | |
1015 | nr_file_extents++; | |
1016 | ||
1017 | mutex_lock(&root->fs_info->alloc_mutex); | |
1018 | ret = __btrfs_inc_extent_ref(trans, root, disk_bytenr, | |
1019 | btrfs_file_extent_disk_num_bytes(buf, fi), | |
1020 | root->root_key.objectid, trans->transid, | |
1021 | key.objectid, key.offset); | |
1022 | mutex_unlock(&root->fs_info->alloc_mutex); | |
1023 | if (ret) { | |
1024 | faili = i; | |
1025 | WARN_ON(1); | |
1026 | goto fail; | |
1027 | } | |
1028 | } else { | |
1029 | bytenr = btrfs_node_blockptr(buf, i); | |
1030 | btrfs_node_key_to_cpu(buf, &key, i); | |
1031 | ||
1032 | mutex_lock(&root->fs_info->alloc_mutex); | |
1033 | ret = __btrfs_inc_extent_ref(trans, root, bytenr, | |
1034 | btrfs_level_size(root, level - 1), | |
1035 | root->root_key.objectid, | |
1036 | trans->transid, | |
1037 | level - 1, key.objectid); | |
1038 | mutex_unlock(&root->fs_info->alloc_mutex); | |
1039 | if (ret) { | |
1040 | faili = i; | |
1041 | WARN_ON(1); | |
1042 | goto fail; | |
1043 | } | |
1044 | } | |
1045 | } | |
1046 | /* cache orignal leaf block's references */ | |
1047 | if (level == 0 && cache_ref && buf != root->commit_root) { | |
1048 | struct btrfs_leaf_ref *ref; | |
1049 | struct btrfs_extent_info *info; | |
1050 | ||
1051 | ref = btrfs_alloc_leaf_ref(root, nr_file_extents); | |
1052 | if (!ref) { | |
1053 | WARN_ON(1); | |
1054 | goto out; | |
1055 | } | |
1056 | ||
1057 | ref->root_gen = root->root_key.offset; | |
1058 | ref->bytenr = buf->start; | |
1059 | ref->owner = btrfs_header_owner(buf); | |
1060 | ref->generation = btrfs_header_generation(buf); | |
1061 | ref->nritems = nr_file_extents; | |
1062 | info = ref->extents; | |
1063 | ||
1064 | for (i = 0; nr_file_extents > 0 && i < nritems; i++) { | |
1065 | u64 disk_bytenr; | |
1066 | btrfs_item_key_to_cpu(buf, &key, i); | |
1067 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) | |
1068 | continue; | |
1069 | fi = btrfs_item_ptr(buf, i, | |
1070 | struct btrfs_file_extent_item); | |
1071 | if (btrfs_file_extent_type(buf, fi) == | |
1072 | BTRFS_FILE_EXTENT_INLINE) | |
1073 | continue; | |
1074 | disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
1075 | if (disk_bytenr == 0) | |
1076 | continue; | |
1077 | ||
1078 | info->bytenr = disk_bytenr; | |
1079 | info->num_bytes = | |
1080 | btrfs_file_extent_disk_num_bytes(buf, fi); | |
1081 | info->objectid = key.objectid; | |
1082 | info->offset = key.offset; | |
1083 | info++; | |
1084 | } | |
1085 | ||
1086 | BUG_ON(!root->ref_tree); | |
1087 | ret = btrfs_add_leaf_ref(root, ref); | |
1088 | WARN_ON(ret); | |
1089 | btrfs_free_leaf_ref(root, ref); | |
1090 | } | |
1091 | out: | |
1092 | return 0; | |
1093 | fail: | |
1094 | WARN_ON(1); | |
1095 | #if 0 | |
1096 | for (i =0; i < faili; i++) { | |
1097 | if (level == 0) { | |
1098 | u64 disk_bytenr; | |
1099 | btrfs_item_key_to_cpu(buf, &key, i); | |
1100 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) | |
1101 | continue; | |
1102 | fi = btrfs_item_ptr(buf, i, | |
1103 | struct btrfs_file_extent_item); | |
1104 | if (btrfs_file_extent_type(buf, fi) == | |
1105 | BTRFS_FILE_EXTENT_INLINE) | |
1106 | continue; | |
1107 | disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
1108 | if (disk_bytenr == 0) | |
1109 | continue; | |
1110 | err = btrfs_free_extent(trans, root, disk_bytenr, | |
1111 | btrfs_file_extent_disk_num_bytes(buf, | |
1112 | fi), 0); | |
1113 | BUG_ON(err); | |
1114 | } else { | |
1115 | bytenr = btrfs_node_blockptr(buf, i); | |
1116 | err = btrfs_free_extent(trans, root, bytenr, | |
1117 | btrfs_level_size(root, level - 1), 0); | |
1118 | BUG_ON(err); | |
1119 | } | |
1120 | } | |
1121 | #endif | |
1122 | return ret; | |
1123 | } | |
1124 | ||
1125 | static int write_one_cache_group(struct btrfs_trans_handle *trans, | |
1126 | struct btrfs_root *root, | |
1127 | struct btrfs_path *path, | |
1128 | struct btrfs_block_group_cache *cache) | |
1129 | { | |
1130 | int ret; | |
1131 | int pending_ret; | |
1132 | struct btrfs_root *extent_root = root->fs_info->extent_root; | |
1133 | unsigned long bi; | |
1134 | struct extent_buffer *leaf; | |
1135 | ||
1136 | ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1); | |
1137 | if (ret < 0) | |
1138 | goto fail; | |
1139 | BUG_ON(ret); | |
1140 | ||
1141 | leaf = path->nodes[0]; | |
1142 | bi = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
1143 | write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item)); | |
1144 | btrfs_mark_buffer_dirty(leaf); | |
1145 | btrfs_release_path(extent_root, path); | |
1146 | fail: | |
1147 | finish_current_insert(trans, extent_root); | |
1148 | pending_ret = del_pending_extents(trans, extent_root); | |
1149 | if (ret) | |
1150 | return ret; | |
1151 | if (pending_ret) | |
1152 | return pending_ret; | |
1153 | return 0; | |
1154 | ||
1155 | } | |
1156 | ||
1157 | int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans, | |
1158 | struct btrfs_root *root) | |
1159 | { | |
1160 | struct extent_io_tree *block_group_cache; | |
1161 | struct btrfs_block_group_cache *cache; | |
1162 | int ret; | |
1163 | int err = 0; | |
1164 | int werr = 0; | |
1165 | struct btrfs_path *path; | |
1166 | u64 last = 0; | |
1167 | u64 start; | |
1168 | u64 end; | |
1169 | u64 ptr; | |
1170 | ||
1171 | block_group_cache = &root->fs_info->block_group_cache; | |
1172 | path = btrfs_alloc_path(); | |
1173 | if (!path) | |
1174 | return -ENOMEM; | |
1175 | ||
1176 | mutex_lock(&root->fs_info->alloc_mutex); | |
1177 | while(1) { | |
1178 | ret = find_first_extent_bit(block_group_cache, last, | |
1179 | &start, &end, BLOCK_GROUP_DIRTY); | |
1180 | if (ret) | |
1181 | break; | |
1182 | ||
1183 | last = end + 1; | |
1184 | ret = get_state_private(block_group_cache, start, &ptr); | |
1185 | if (ret) | |
1186 | break; | |
1187 | cache = (struct btrfs_block_group_cache *)(unsigned long)ptr; | |
1188 | err = write_one_cache_group(trans, root, | |
1189 | path, cache); | |
1190 | /* | |
1191 | * if we fail to write the cache group, we want | |
1192 | * to keep it marked dirty in hopes that a later | |
1193 | * write will work | |
1194 | */ | |
1195 | if (err) { | |
1196 | werr = err; | |
1197 | continue; | |
1198 | } | |
1199 | clear_extent_bits(block_group_cache, start, end, | |
1200 | BLOCK_GROUP_DIRTY, GFP_NOFS); | |
1201 | } | |
1202 | btrfs_free_path(path); | |
1203 | mutex_unlock(&root->fs_info->alloc_mutex); | |
1204 | return werr; | |
1205 | } | |
1206 | ||
1207 | static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info, | |
1208 | u64 flags) | |
1209 | { | |
1210 | struct list_head *head = &info->space_info; | |
1211 | struct list_head *cur; | |
1212 | struct btrfs_space_info *found; | |
1213 | list_for_each(cur, head) { | |
1214 | found = list_entry(cur, struct btrfs_space_info, list); | |
1215 | if (found->flags == flags) | |
1216 | return found; | |
1217 | } | |
1218 | return NULL; | |
1219 | ||
1220 | } | |
1221 | ||
1222 | static int update_space_info(struct btrfs_fs_info *info, u64 flags, | |
1223 | u64 total_bytes, u64 bytes_used, | |
1224 | struct btrfs_space_info **space_info) | |
1225 | { | |
1226 | struct btrfs_space_info *found; | |
1227 | ||
1228 | found = __find_space_info(info, flags); | |
1229 | if (found) { | |
1230 | found->total_bytes += total_bytes; | |
1231 | found->bytes_used += bytes_used; | |
1232 | found->full = 0; | |
1233 | WARN_ON(found->total_bytes < found->bytes_used); | |
1234 | *space_info = found; | |
1235 | return 0; | |
1236 | } | |
1237 | found = kmalloc(sizeof(*found), GFP_NOFS); | |
1238 | if (!found) | |
1239 | return -ENOMEM; | |
1240 | ||
1241 | list_add(&found->list, &info->space_info); | |
1242 | found->flags = flags; | |
1243 | found->total_bytes = total_bytes; | |
1244 | found->bytes_used = bytes_used; | |
1245 | found->bytes_pinned = 0; | |
1246 | found->full = 0; | |
1247 | found->force_alloc = 0; | |
1248 | *space_info = found; | |
1249 | return 0; | |
1250 | } | |
1251 | ||
1252 | static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) | |
1253 | { | |
1254 | u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 | | |
1255 | BTRFS_BLOCK_GROUP_RAID1 | | |
1256 | BTRFS_BLOCK_GROUP_RAID10 | | |
1257 | BTRFS_BLOCK_GROUP_DUP); | |
1258 | if (extra_flags) { | |
1259 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
1260 | fs_info->avail_data_alloc_bits |= extra_flags; | |
1261 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
1262 | fs_info->avail_metadata_alloc_bits |= extra_flags; | |
1263 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
1264 | fs_info->avail_system_alloc_bits |= extra_flags; | |
1265 | } | |
1266 | } | |
1267 | ||
1268 | static u64 reduce_alloc_profile(struct btrfs_root *root, u64 flags) | |
1269 | { | |
1270 | u64 num_devices = root->fs_info->fs_devices->num_devices; | |
1271 | ||
1272 | if (num_devices == 1) | |
1273 | flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0); | |
1274 | if (num_devices < 4) | |
1275 | flags &= ~BTRFS_BLOCK_GROUP_RAID10; | |
1276 | ||
1277 | if ((flags & BTRFS_BLOCK_GROUP_DUP) && | |
1278 | (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
1279 | BTRFS_BLOCK_GROUP_RAID10))) { | |
1280 | flags &= ~BTRFS_BLOCK_GROUP_DUP; | |
1281 | } | |
1282 | ||
1283 | if ((flags & BTRFS_BLOCK_GROUP_RAID1) && | |
1284 | (flags & BTRFS_BLOCK_GROUP_RAID10)) { | |
1285 | flags &= ~BTRFS_BLOCK_GROUP_RAID1; | |
1286 | } | |
1287 | ||
1288 | if ((flags & BTRFS_BLOCK_GROUP_RAID0) && | |
1289 | ((flags & BTRFS_BLOCK_GROUP_RAID1) | | |
1290 | (flags & BTRFS_BLOCK_GROUP_RAID10) | | |
1291 | (flags & BTRFS_BLOCK_GROUP_DUP))) | |
1292 | flags &= ~BTRFS_BLOCK_GROUP_RAID0; | |
1293 | return flags; | |
1294 | } | |
1295 | ||
1296 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, | |
1297 | struct btrfs_root *extent_root, u64 alloc_bytes, | |
1298 | u64 flags, int force) | |
1299 | { | |
1300 | struct btrfs_space_info *space_info; | |
1301 | u64 thresh; | |
1302 | u64 start; | |
1303 | u64 num_bytes; | |
1304 | int ret; | |
1305 | ||
1306 | flags = reduce_alloc_profile(extent_root, flags); | |
1307 | ||
1308 | space_info = __find_space_info(extent_root->fs_info, flags); | |
1309 | if (!space_info) { | |
1310 | ret = update_space_info(extent_root->fs_info, flags, | |
1311 | 0, 0, &space_info); | |
1312 | BUG_ON(ret); | |
1313 | } | |
1314 | BUG_ON(!space_info); | |
1315 | ||
1316 | if (space_info->force_alloc) { | |
1317 | force = 1; | |
1318 | space_info->force_alloc = 0; | |
1319 | } | |
1320 | if (space_info->full) | |
1321 | goto out; | |
1322 | ||
1323 | thresh = div_factor(space_info->total_bytes, 6); | |
1324 | if (!force && | |
1325 | (space_info->bytes_used + space_info->bytes_pinned + alloc_bytes) < | |
1326 | thresh) | |
1327 | goto out; | |
1328 | ||
1329 | mutex_lock(&extent_root->fs_info->chunk_mutex); | |
1330 | ret = btrfs_alloc_chunk(trans, extent_root, &start, &num_bytes, flags); | |
1331 | if (ret == -ENOSPC) { | |
1332 | printk("space info full %Lu\n", flags); | |
1333 | space_info->full = 1; | |
1334 | goto out_unlock; | |
1335 | } | |
1336 | BUG_ON(ret); | |
1337 | ||
1338 | ret = btrfs_make_block_group(trans, extent_root, 0, flags, | |
1339 | BTRFS_FIRST_CHUNK_TREE_OBJECTID, start, num_bytes); | |
1340 | BUG_ON(ret); | |
1341 | out_unlock: | |
1342 | mutex_unlock(&extent_root->fs_info->chunk_mutex); | |
1343 | out: | |
1344 | return 0; | |
1345 | } | |
1346 | ||
1347 | static int update_block_group(struct btrfs_trans_handle *trans, | |
1348 | struct btrfs_root *root, | |
1349 | u64 bytenr, u64 num_bytes, int alloc, | |
1350 | int mark_free) | |
1351 | { | |
1352 | struct btrfs_block_group_cache *cache; | |
1353 | struct btrfs_fs_info *info = root->fs_info; | |
1354 | u64 total = num_bytes; | |
1355 | u64 old_val; | |
1356 | u64 byte_in_group; | |
1357 | u64 start; | |
1358 | u64 end; | |
1359 | ||
1360 | WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex)); | |
1361 | while(total) { | |
1362 | cache = btrfs_lookup_block_group(info, bytenr); | |
1363 | if (!cache) { | |
1364 | return -1; | |
1365 | } | |
1366 | byte_in_group = bytenr - cache->key.objectid; | |
1367 | WARN_ON(byte_in_group > cache->key.offset); | |
1368 | start = cache->key.objectid; | |
1369 | end = start + cache->key.offset - 1; | |
1370 | set_extent_bits(&info->block_group_cache, start, end, | |
1371 | BLOCK_GROUP_DIRTY, GFP_NOFS); | |
1372 | ||
1373 | spin_lock(&cache->lock); | |
1374 | old_val = btrfs_block_group_used(&cache->item); | |
1375 | num_bytes = min(total, cache->key.offset - byte_in_group); | |
1376 | if (alloc) { | |
1377 | old_val += num_bytes; | |
1378 | cache->space_info->bytes_used += num_bytes; | |
1379 | btrfs_set_block_group_used(&cache->item, old_val); | |
1380 | spin_unlock(&cache->lock); | |
1381 | } else { | |
1382 | old_val -= num_bytes; | |
1383 | cache->space_info->bytes_used -= num_bytes; | |
1384 | btrfs_set_block_group_used(&cache->item, old_val); | |
1385 | spin_unlock(&cache->lock); | |
1386 | if (mark_free) { | |
1387 | set_extent_dirty(&info->free_space_cache, | |
1388 | bytenr, bytenr + num_bytes - 1, | |
1389 | GFP_NOFS); | |
1390 | } | |
1391 | } | |
1392 | total -= num_bytes; | |
1393 | bytenr += num_bytes; | |
1394 | } | |
1395 | return 0; | |
1396 | } | |
1397 | ||
1398 | static u64 first_logical_byte(struct btrfs_root *root, u64 search_start) | |
1399 | { | |
1400 | u64 start; | |
1401 | u64 end; | |
1402 | int ret; | |
1403 | ret = find_first_extent_bit(&root->fs_info->block_group_cache, | |
1404 | search_start, &start, &end, | |
1405 | BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA | | |
1406 | BLOCK_GROUP_SYSTEM); | |
1407 | if (ret) | |
1408 | return 0; | |
1409 | return start; | |
1410 | } | |
1411 | ||
1412 | ||
1413 | static int update_pinned_extents(struct btrfs_root *root, | |
1414 | u64 bytenr, u64 num, int pin) | |
1415 | { | |
1416 | u64 len; | |
1417 | struct btrfs_block_group_cache *cache; | |
1418 | struct btrfs_fs_info *fs_info = root->fs_info; | |
1419 | ||
1420 | WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex)); | |
1421 | if (pin) { | |
1422 | set_extent_dirty(&fs_info->pinned_extents, | |
1423 | bytenr, bytenr + num - 1, GFP_NOFS); | |
1424 | } else { | |
1425 | clear_extent_dirty(&fs_info->pinned_extents, | |
1426 | bytenr, bytenr + num - 1, GFP_NOFS); | |
1427 | } | |
1428 | while (num > 0) { | |
1429 | cache = btrfs_lookup_block_group(fs_info, bytenr); | |
1430 | if (!cache) { | |
1431 | u64 first = first_logical_byte(root, bytenr); | |
1432 | WARN_ON(first < bytenr); | |
1433 | len = min(first - bytenr, num); | |
1434 | } else { | |
1435 | len = min(num, cache->key.offset - | |
1436 | (bytenr - cache->key.objectid)); | |
1437 | } | |
1438 | if (pin) { | |
1439 | if (cache) { | |
1440 | spin_lock(&cache->lock); | |
1441 | cache->pinned += len; | |
1442 | cache->space_info->bytes_pinned += len; | |
1443 | spin_unlock(&cache->lock); | |
1444 | } | |
1445 | fs_info->total_pinned += len; | |
1446 | } else { | |
1447 | if (cache) { | |
1448 | spin_lock(&cache->lock); | |
1449 | cache->pinned -= len; | |
1450 | cache->space_info->bytes_pinned -= len; | |
1451 | spin_unlock(&cache->lock); | |
1452 | } | |
1453 | fs_info->total_pinned -= len; | |
1454 | } | |
1455 | bytenr += len; | |
1456 | num -= len; | |
1457 | } | |
1458 | return 0; | |
1459 | } | |
1460 | ||
1461 | int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy) | |
1462 | { | |
1463 | u64 last = 0; | |
1464 | u64 start; | |
1465 | u64 end; | |
1466 | struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents; | |
1467 | int ret; | |
1468 | ||
1469 | while(1) { | |
1470 | ret = find_first_extent_bit(pinned_extents, last, | |
1471 | &start, &end, EXTENT_DIRTY); | |
1472 | if (ret) | |
1473 | break; | |
1474 | set_extent_dirty(copy, start, end, GFP_NOFS); | |
1475 | last = end + 1; | |
1476 | } | |
1477 | return 0; | |
1478 | } | |
1479 | ||
1480 | int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans, | |
1481 | struct btrfs_root *root, | |
1482 | struct extent_io_tree *unpin) | |
1483 | { | |
1484 | u64 start; | |
1485 | u64 end; | |
1486 | int ret; | |
1487 | struct extent_io_tree *free_space_cache; | |
1488 | free_space_cache = &root->fs_info->free_space_cache; | |
1489 | ||
1490 | mutex_lock(&root->fs_info->alloc_mutex); | |
1491 | while(1) { | |
1492 | ret = find_first_extent_bit(unpin, 0, &start, &end, | |
1493 | EXTENT_DIRTY); | |
1494 | if (ret) | |
1495 | break; | |
1496 | update_pinned_extents(root, start, end + 1 - start, 0); | |
1497 | clear_extent_dirty(unpin, start, end, GFP_NOFS); | |
1498 | set_extent_dirty(free_space_cache, start, end, GFP_NOFS); | |
1499 | if (need_resched()) { | |
1500 | mutex_unlock(&root->fs_info->alloc_mutex); | |
1501 | cond_resched(); | |
1502 | mutex_lock(&root->fs_info->alloc_mutex); | |
1503 | } | |
1504 | } | |
1505 | mutex_unlock(&root->fs_info->alloc_mutex); | |
1506 | return 0; | |
1507 | } | |
1508 | ||
1509 | static int finish_current_insert(struct btrfs_trans_handle *trans, | |
1510 | struct btrfs_root *extent_root) | |
1511 | { | |
1512 | u64 start; | |
1513 | u64 end; | |
1514 | struct btrfs_fs_info *info = extent_root->fs_info; | |
1515 | struct extent_buffer *eb; | |
1516 | struct btrfs_path *path; | |
1517 | struct btrfs_key ins; | |
1518 | struct btrfs_disk_key first; | |
1519 | struct btrfs_extent_item extent_item; | |
1520 | int ret; | |
1521 | int level; | |
1522 | int err = 0; | |
1523 | ||
1524 | WARN_ON(!mutex_is_locked(&extent_root->fs_info->alloc_mutex)); | |
1525 | btrfs_set_stack_extent_refs(&extent_item, 1); | |
1526 | btrfs_set_key_type(&ins, BTRFS_EXTENT_ITEM_KEY); | |
1527 | path = btrfs_alloc_path(); | |
1528 | ||
1529 | while(1) { | |
1530 | ret = find_first_extent_bit(&info->extent_ins, 0, &start, | |
1531 | &end, EXTENT_LOCKED); | |
1532 | if (ret) | |
1533 | break; | |
1534 | ||
1535 | ins.objectid = start; | |
1536 | ins.offset = end + 1 - start; | |
1537 | err = btrfs_insert_item(trans, extent_root, &ins, | |
1538 | &extent_item, sizeof(extent_item)); | |
1539 | clear_extent_bits(&info->extent_ins, start, end, EXTENT_LOCKED, | |
1540 | GFP_NOFS); | |
1541 | ||
1542 | eb = btrfs_find_tree_block(extent_root, ins.objectid, | |
1543 | ins.offset); | |
1544 | ||
1545 | if (!btrfs_buffer_uptodate(eb, trans->transid)) { | |
1546 | mutex_unlock(&extent_root->fs_info->alloc_mutex); | |
1547 | btrfs_read_buffer(eb, trans->transid); | |
1548 | mutex_lock(&extent_root->fs_info->alloc_mutex); | |
1549 | } | |
1550 | ||
1551 | btrfs_tree_lock(eb); | |
1552 | level = btrfs_header_level(eb); | |
1553 | if (level == 0) { | |
1554 | btrfs_item_key(eb, &first, 0); | |
1555 | } else { | |
1556 | btrfs_node_key(eb, &first, 0); | |
1557 | } | |
1558 | btrfs_tree_unlock(eb); | |
1559 | free_extent_buffer(eb); | |
1560 | /* | |
1561 | * the first key is just a hint, so the race we've created | |
1562 | * against reading it is fine | |
1563 | */ | |
1564 | err = btrfs_insert_extent_backref(trans, extent_root, path, | |
1565 | start, extent_root->root_key.objectid, | |
1566 | 0, level, | |
1567 | btrfs_disk_key_objectid(&first)); | |
1568 | BUG_ON(err); | |
1569 | if (need_resched()) { | |
1570 | mutex_unlock(&extent_root->fs_info->alloc_mutex); | |
1571 | cond_resched(); | |
1572 | mutex_lock(&extent_root->fs_info->alloc_mutex); | |
1573 | } | |
1574 | } | |
1575 | btrfs_free_path(path); | |
1576 | return 0; | |
1577 | } | |
1578 | ||
1579 | static int pin_down_bytes(struct btrfs_root *root, u64 bytenr, u32 num_bytes, | |
1580 | int pending) | |
1581 | { | |
1582 | int err = 0; | |
1583 | ||
1584 | WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex)); | |
1585 | if (!pending) { | |
1586 | struct extent_buffer *buf; | |
1587 | buf = btrfs_find_tree_block(root, bytenr, num_bytes); | |
1588 | if (buf) { | |
1589 | if (btrfs_buffer_uptodate(buf, 0) && | |
1590 | btrfs_try_tree_lock(buf)) { | |
1591 | u64 transid = | |
1592 | root->fs_info->running_transaction->transid; | |
1593 | u64 header_transid = | |
1594 | btrfs_header_generation(buf); | |
1595 | if (header_transid == transid && | |
1596 | !btrfs_header_flag(buf, | |
1597 | BTRFS_HEADER_FLAG_WRITTEN)) { | |
1598 | clean_tree_block(NULL, root, buf); | |
1599 | btrfs_tree_unlock(buf); | |
1600 | free_extent_buffer(buf); | |
1601 | return 1; | |
1602 | } | |
1603 | btrfs_tree_unlock(buf); | |
1604 | } | |
1605 | free_extent_buffer(buf); | |
1606 | } | |
1607 | update_pinned_extents(root, bytenr, num_bytes, 1); | |
1608 | } else { | |
1609 | set_extent_bits(&root->fs_info->pending_del, | |
1610 | bytenr, bytenr + num_bytes - 1, | |
1611 | EXTENT_LOCKED, GFP_NOFS); | |
1612 | } | |
1613 | BUG_ON(err < 0); | |
1614 | return 0; | |
1615 | } | |
1616 | ||
1617 | /* | |
1618 | * remove an extent from the root, returns 0 on success | |
1619 | */ | |
1620 | static int __free_extent(struct btrfs_trans_handle *trans, struct btrfs_root | |
1621 | *root, u64 bytenr, u64 num_bytes, | |
1622 | u64 root_objectid, u64 ref_generation, | |
1623 | u64 owner_objectid, u64 owner_offset, int pin, | |
1624 | int mark_free) | |
1625 | { | |
1626 | struct btrfs_path *path; | |
1627 | struct btrfs_key key; | |
1628 | struct btrfs_fs_info *info = root->fs_info; | |
1629 | struct btrfs_root *extent_root = info->extent_root; | |
1630 | struct extent_buffer *leaf; | |
1631 | int ret; | |
1632 | int extent_slot = 0; | |
1633 | int found_extent = 0; | |
1634 | int num_to_del = 1; | |
1635 | struct btrfs_extent_item *ei; | |
1636 | u32 refs; | |
1637 | ||
1638 | WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex)); | |
1639 | key.objectid = bytenr; | |
1640 | btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); | |
1641 | key.offset = num_bytes; | |
1642 | path = btrfs_alloc_path(); | |
1643 | if (!path) | |
1644 | return -ENOMEM; | |
1645 | ||
1646 | path->reada = 1; | |
1647 | ret = lookup_extent_backref(trans, extent_root, path, | |
1648 | bytenr, root_objectid, | |
1649 | ref_generation, | |
1650 | owner_objectid, owner_offset, 1); | |
1651 | if (ret == 0) { | |
1652 | struct btrfs_key found_key; | |
1653 | extent_slot = path->slots[0]; | |
1654 | while(extent_slot > 0) { | |
1655 | extent_slot--; | |
1656 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
1657 | extent_slot); | |
1658 | if (found_key.objectid != bytenr) | |
1659 | break; | |
1660 | if (found_key.type == BTRFS_EXTENT_ITEM_KEY && | |
1661 | found_key.offset == num_bytes) { | |
1662 | found_extent = 1; | |
1663 | break; | |
1664 | } | |
1665 | if (path->slots[0] - extent_slot > 5) | |
1666 | break; | |
1667 | } | |
1668 | if (!found_extent) | |
1669 | ret = btrfs_del_item(trans, extent_root, path); | |
1670 | } else { | |
1671 | btrfs_print_leaf(extent_root, path->nodes[0]); | |
1672 | WARN_ON(1); | |
1673 | printk("Unable to find ref byte nr %Lu root %Lu " | |
1674 | " gen %Lu owner %Lu offset %Lu\n", bytenr, | |
1675 | root_objectid, ref_generation, owner_objectid, | |
1676 | owner_offset); | |
1677 | } | |
1678 | if (!found_extent) { | |
1679 | btrfs_release_path(extent_root, path); | |
1680 | ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1); | |
1681 | if (ret < 0) | |
1682 | return ret; | |
1683 | BUG_ON(ret); | |
1684 | extent_slot = path->slots[0]; | |
1685 | } | |
1686 | ||
1687 | leaf = path->nodes[0]; | |
1688 | ei = btrfs_item_ptr(leaf, extent_slot, | |
1689 | struct btrfs_extent_item); | |
1690 | refs = btrfs_extent_refs(leaf, ei); | |
1691 | BUG_ON(refs == 0); | |
1692 | refs -= 1; | |
1693 | btrfs_set_extent_refs(leaf, ei, refs); | |
1694 | ||
1695 | btrfs_mark_buffer_dirty(leaf); | |
1696 | ||
1697 | if (refs == 0 && found_extent && path->slots[0] == extent_slot + 1) { | |
1698 | /* if the back ref and the extent are next to each other | |
1699 | * they get deleted below in one shot | |
1700 | */ | |
1701 | path->slots[0] = extent_slot; | |
1702 | num_to_del = 2; | |
1703 | } else if (found_extent) { | |
1704 | /* otherwise delete the extent back ref */ | |
1705 | ret = btrfs_del_item(trans, extent_root, path); | |
1706 | BUG_ON(ret); | |
1707 | /* if refs are 0, we need to setup the path for deletion */ | |
1708 | if (refs == 0) { | |
1709 | btrfs_release_path(extent_root, path); | |
1710 | ret = btrfs_search_slot(trans, extent_root, &key, path, | |
1711 | -1, 1); | |
1712 | if (ret < 0) | |
1713 | return ret; | |
1714 | BUG_ON(ret); | |
1715 | } | |
1716 | } | |
1717 | ||
1718 | if (refs == 0) { | |
1719 | u64 super_used; | |
1720 | u64 root_used; | |
1721 | ||
1722 | if (pin) { | |
1723 | ret = pin_down_bytes(root, bytenr, num_bytes, 0); | |
1724 | if (ret > 0) | |
1725 | mark_free = 1; | |
1726 | BUG_ON(ret < 0); | |
1727 | } | |
1728 | ||
1729 | /* block accounting for super block */ | |
1730 | spin_lock_irq(&info->delalloc_lock); | |
1731 | super_used = btrfs_super_bytes_used(&info->super_copy); | |
1732 | btrfs_set_super_bytes_used(&info->super_copy, | |
1733 | super_used - num_bytes); | |
1734 | spin_unlock_irq(&info->delalloc_lock); | |
1735 | ||
1736 | /* block accounting for root item */ | |
1737 | root_used = btrfs_root_used(&root->root_item); | |
1738 | btrfs_set_root_used(&root->root_item, | |
1739 | root_used - num_bytes); | |
1740 | ret = btrfs_del_items(trans, extent_root, path, path->slots[0], | |
1741 | num_to_del); | |
1742 | if (ret) { | |
1743 | return ret; | |
1744 | } | |
1745 | ret = update_block_group(trans, root, bytenr, num_bytes, 0, | |
1746 | mark_free); | |
1747 | BUG_ON(ret); | |
1748 | } | |
1749 | btrfs_free_path(path); | |
1750 | finish_current_insert(trans, extent_root); | |
1751 | return ret; | |
1752 | } | |
1753 | ||
1754 | /* | |
1755 | * find all the blocks marked as pending in the radix tree and remove | |
1756 | * them from the extent map | |
1757 | */ | |
1758 | static int del_pending_extents(struct btrfs_trans_handle *trans, struct | |
1759 | btrfs_root *extent_root) | |
1760 | { | |
1761 | int ret; | |
1762 | int err = 0; | |
1763 | u64 start; | |
1764 | u64 end; | |
1765 | struct extent_io_tree *pending_del; | |
1766 | struct extent_io_tree *pinned_extents; | |
1767 | ||
1768 | WARN_ON(!mutex_is_locked(&extent_root->fs_info->alloc_mutex)); | |
1769 | pending_del = &extent_root->fs_info->pending_del; | |
1770 | pinned_extents = &extent_root->fs_info->pinned_extents; | |
1771 | ||
1772 | while(1) { | |
1773 | ret = find_first_extent_bit(pending_del, 0, &start, &end, | |
1774 | EXTENT_LOCKED); | |
1775 | if (ret) | |
1776 | break; | |
1777 | clear_extent_bits(pending_del, start, end, EXTENT_LOCKED, | |
1778 | GFP_NOFS); | |
1779 | if (!test_range_bit(&extent_root->fs_info->extent_ins, | |
1780 | start, end, EXTENT_LOCKED, 0)) { | |
1781 | update_pinned_extents(extent_root, start, | |
1782 | end + 1 - start, 1); | |
1783 | ret = __free_extent(trans, extent_root, | |
1784 | start, end + 1 - start, | |
1785 | extent_root->root_key.objectid, | |
1786 | 0, 0, 0, 0, 0); | |
1787 | } else { | |
1788 | clear_extent_bits(&extent_root->fs_info->extent_ins, | |
1789 | start, end, EXTENT_LOCKED, GFP_NOFS); | |
1790 | } | |
1791 | if (ret) | |
1792 | err = ret; | |
1793 | ||
1794 | if (need_resched()) { | |
1795 | mutex_unlock(&extent_root->fs_info->alloc_mutex); | |
1796 | cond_resched(); | |
1797 | mutex_lock(&extent_root->fs_info->alloc_mutex); | |
1798 | } | |
1799 | } | |
1800 | return err; | |
1801 | } | |
1802 | ||
1803 | /* | |
1804 | * remove an extent from the root, returns 0 on success | |
1805 | */ | |
1806 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, | |
1807 | struct btrfs_root *root, u64 bytenr, | |
1808 | u64 num_bytes, u64 root_objectid, | |
1809 | u64 ref_generation, u64 owner_objectid, | |
1810 | u64 owner_offset, int pin) | |
1811 | { | |
1812 | struct btrfs_root *extent_root = root->fs_info->extent_root; | |
1813 | int pending_ret; | |
1814 | int ret; | |
1815 | ||
1816 | WARN_ON(num_bytes < root->sectorsize); | |
1817 | if (!root->ref_cows) | |
1818 | ref_generation = 0; | |
1819 | ||
1820 | if (root == extent_root) { | |
1821 | pin_down_bytes(root, bytenr, num_bytes, 1); | |
1822 | return 0; | |
1823 | } | |
1824 | ret = __free_extent(trans, root, bytenr, num_bytes, root_objectid, | |
1825 | ref_generation, owner_objectid, owner_offset, | |
1826 | pin, pin == 0); | |
1827 | ||
1828 | finish_current_insert(trans, root->fs_info->extent_root); | |
1829 | pending_ret = del_pending_extents(trans, root->fs_info->extent_root); | |
1830 | return ret ? ret : pending_ret; | |
1831 | } | |
1832 | ||
1833 | int btrfs_free_extent(struct btrfs_trans_handle *trans, | |
1834 | struct btrfs_root *root, u64 bytenr, | |
1835 | u64 num_bytes, u64 root_objectid, | |
1836 | u64 ref_generation, u64 owner_objectid, | |
1837 | u64 owner_offset, int pin) | |
1838 | { | |
1839 | int ret; | |
1840 | ||
1841 | maybe_lock_mutex(root); | |
1842 | ret = __btrfs_free_extent(trans, root, bytenr, num_bytes, | |
1843 | root_objectid, ref_generation, | |
1844 | owner_objectid, owner_offset, pin); | |
1845 | maybe_unlock_mutex(root); | |
1846 | return ret; | |
1847 | } | |
1848 | ||
1849 | static u64 stripe_align(struct btrfs_root *root, u64 val) | |
1850 | { | |
1851 | u64 mask = ((u64)root->stripesize - 1); | |
1852 | u64 ret = (val + mask) & ~mask; | |
1853 | return ret; | |
1854 | } | |
1855 | ||
1856 | /* | |
1857 | * walks the btree of allocated extents and find a hole of a given size. | |
1858 | * The key ins is changed to record the hole: | |
1859 | * ins->objectid == block start | |
1860 | * ins->flags = BTRFS_EXTENT_ITEM_KEY | |
1861 | * ins->offset == number of blocks | |
1862 | * Any available blocks before search_start are skipped. | |
1863 | */ | |
1864 | static int noinline find_free_extent(struct btrfs_trans_handle *trans, | |
1865 | struct btrfs_root *orig_root, | |
1866 | u64 num_bytes, u64 empty_size, | |
1867 | u64 search_start, u64 search_end, | |
1868 | u64 hint_byte, struct btrfs_key *ins, | |
1869 | u64 exclude_start, u64 exclude_nr, | |
1870 | int data) | |
1871 | { | |
1872 | int ret; | |
1873 | u64 orig_search_start; | |
1874 | struct btrfs_root * root = orig_root->fs_info->extent_root; | |
1875 | struct btrfs_fs_info *info = root->fs_info; | |
1876 | u64 total_needed = num_bytes; | |
1877 | u64 *last_ptr = NULL; | |
1878 | struct btrfs_block_group_cache *block_group; | |
1879 | int full_scan = 0; | |
1880 | int wrapped = 0; | |
1881 | int chunk_alloc_done = 0; | |
1882 | int empty_cluster = 2 * 1024 * 1024; | |
1883 | int allowed_chunk_alloc = 0; | |
1884 | ||
1885 | WARN_ON(num_bytes < root->sectorsize); | |
1886 | btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY); | |
1887 | ||
1888 | if (orig_root->ref_cows || empty_size) | |
1889 | allowed_chunk_alloc = 1; | |
1890 | ||
1891 | if (data & BTRFS_BLOCK_GROUP_METADATA) { | |
1892 | last_ptr = &root->fs_info->last_alloc; | |
1893 | empty_cluster = 256 * 1024; | |
1894 | } | |
1895 | ||
1896 | if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD)) { | |
1897 | last_ptr = &root->fs_info->last_data_alloc; | |
1898 | } | |
1899 | ||
1900 | if (last_ptr) { | |
1901 | if (*last_ptr) | |
1902 | hint_byte = *last_ptr; | |
1903 | else { | |
1904 | empty_size += empty_cluster; | |
1905 | } | |
1906 | } | |
1907 | ||
1908 | search_start = max(search_start, first_logical_byte(root, 0)); | |
1909 | orig_search_start = search_start; | |
1910 | ||
1911 | if (search_end == (u64)-1) | |
1912 | search_end = btrfs_super_total_bytes(&info->super_copy); | |
1913 | ||
1914 | if (hint_byte) { | |
1915 | block_group = btrfs_lookup_first_block_group(info, hint_byte); | |
1916 | if (!block_group) | |
1917 | hint_byte = search_start; | |
1918 | block_group = btrfs_find_block_group(root, block_group, | |
1919 | hint_byte, data, 1); | |
1920 | if (last_ptr && *last_ptr == 0 && block_group) | |
1921 | hint_byte = block_group->key.objectid; | |
1922 | } else { | |
1923 | block_group = btrfs_find_block_group(root, | |
1924 | trans->block_group, | |
1925 | search_start, data, 1); | |
1926 | } | |
1927 | search_start = max(search_start, hint_byte); | |
1928 | ||
1929 | total_needed += empty_size; | |
1930 | ||
1931 | check_failed: | |
1932 | if (!block_group) { | |
1933 | block_group = btrfs_lookup_first_block_group(info, | |
1934 | search_start); | |
1935 | if (!block_group) | |
1936 | block_group = btrfs_lookup_first_block_group(info, | |
1937 | orig_search_start); | |
1938 | } | |
1939 | if (full_scan && !chunk_alloc_done) { | |
1940 | if (allowed_chunk_alloc) { | |
1941 | do_chunk_alloc(trans, root, | |
1942 | num_bytes + 2 * 1024 * 1024, data, 1); | |
1943 | allowed_chunk_alloc = 0; | |
1944 | } else if (block_group && block_group_bits(block_group, data)) { | |
1945 | block_group->space_info->force_alloc = 1; | |
1946 | } | |
1947 | chunk_alloc_done = 1; | |
1948 | } | |
1949 | ret = find_search_start(root, &block_group, &search_start, | |
1950 | total_needed, data); | |
1951 | if (ret == -ENOSPC && last_ptr && *last_ptr) { | |
1952 | *last_ptr = 0; | |
1953 | block_group = btrfs_lookup_first_block_group(info, | |
1954 | orig_search_start); | |
1955 | search_start = orig_search_start; | |
1956 | ret = find_search_start(root, &block_group, &search_start, | |
1957 | total_needed, data); | |
1958 | } | |
1959 | if (ret == -ENOSPC) | |
1960 | goto enospc; | |
1961 | if (ret) | |
1962 | goto error; | |
1963 | ||
1964 | if (last_ptr && *last_ptr && search_start != *last_ptr) { | |
1965 | *last_ptr = 0; | |
1966 | if (!empty_size) { | |
1967 | empty_size += empty_cluster; | |
1968 | total_needed += empty_size; | |
1969 | } | |
1970 | block_group = btrfs_lookup_first_block_group(info, | |
1971 | orig_search_start); | |
1972 | search_start = orig_search_start; | |
1973 | ret = find_search_start(root, &block_group, | |
1974 | &search_start, total_needed, data); | |
1975 | if (ret == -ENOSPC) | |
1976 | goto enospc; | |
1977 | if (ret) | |
1978 | goto error; | |
1979 | } | |
1980 | ||
1981 | search_start = stripe_align(root, search_start); | |
1982 | ins->objectid = search_start; | |
1983 | ins->offset = num_bytes; | |
1984 | ||
1985 | if (ins->objectid + num_bytes >= search_end) | |
1986 | goto enospc; | |
1987 | ||
1988 | if (ins->objectid + num_bytes > | |
1989 | block_group->key.objectid + block_group->key.offset) { | |
1990 | search_start = block_group->key.objectid + | |
1991 | block_group->key.offset; | |
1992 | goto new_group; | |
1993 | } | |
1994 | ||
1995 | if (test_range_bit(&info->extent_ins, ins->objectid, | |
1996 | ins->objectid + num_bytes -1, EXTENT_LOCKED, 0)) { | |
1997 | search_start = ins->objectid + num_bytes; | |
1998 | goto new_group; | |
1999 | } | |
2000 | ||
2001 | if (test_range_bit(&info->pinned_extents, ins->objectid, | |
2002 | ins->objectid + num_bytes -1, EXTENT_DIRTY, 0)) { | |
2003 | search_start = ins->objectid + num_bytes; | |
2004 | goto new_group; | |
2005 | } | |
2006 | ||
2007 | if (exclude_nr > 0 && (ins->objectid + num_bytes > exclude_start && | |
2008 | ins->objectid < exclude_start + exclude_nr)) { | |
2009 | search_start = exclude_start + exclude_nr; | |
2010 | goto new_group; | |
2011 | } | |
2012 | ||
2013 | if (!(data & BTRFS_BLOCK_GROUP_DATA)) { | |
2014 | block_group = btrfs_lookup_block_group(info, ins->objectid); | |
2015 | if (block_group) | |
2016 | trans->block_group = block_group; | |
2017 | } | |
2018 | ins->offset = num_bytes; | |
2019 | if (last_ptr) { | |
2020 | *last_ptr = ins->objectid + ins->offset; | |
2021 | if (*last_ptr == | |
2022 | btrfs_super_total_bytes(&root->fs_info->super_copy)) { | |
2023 | *last_ptr = 0; | |
2024 | } | |
2025 | } | |
2026 | return 0; | |
2027 | ||
2028 | new_group: | |
2029 | if (search_start + num_bytes >= search_end) { | |
2030 | enospc: | |
2031 | search_start = orig_search_start; | |
2032 | if (full_scan) { | |
2033 | ret = -ENOSPC; | |
2034 | goto error; | |
2035 | } | |
2036 | if (wrapped) { | |
2037 | if (!full_scan) | |
2038 | total_needed -= empty_size; | |
2039 | full_scan = 1; | |
2040 | } else | |
2041 | wrapped = 1; | |
2042 | } | |
2043 | block_group = btrfs_lookup_first_block_group(info, search_start); | |
2044 | cond_resched(); | |
2045 | block_group = btrfs_find_block_group(root, block_group, | |
2046 | search_start, data, 0); | |
2047 | goto check_failed; | |
2048 | ||
2049 | error: | |
2050 | return ret; | |
2051 | } | |
2052 | ||
2053 | static int __btrfs_reserve_extent(struct btrfs_trans_handle *trans, | |
2054 | struct btrfs_root *root, | |
2055 | u64 num_bytes, u64 min_alloc_size, | |
2056 | u64 empty_size, u64 hint_byte, | |
2057 | u64 search_end, struct btrfs_key *ins, | |
2058 | u64 data) | |
2059 | { | |
2060 | int ret; | |
2061 | u64 search_start = 0; | |
2062 | u64 alloc_profile; | |
2063 | struct btrfs_fs_info *info = root->fs_info; | |
2064 | ||
2065 | if (data) { | |
2066 | alloc_profile = info->avail_data_alloc_bits & | |
2067 | info->data_alloc_profile; | |
2068 | data = BTRFS_BLOCK_GROUP_DATA | alloc_profile; | |
2069 | } else if (root == root->fs_info->chunk_root) { | |
2070 | alloc_profile = info->avail_system_alloc_bits & | |
2071 | info->system_alloc_profile; | |
2072 | data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile; | |
2073 | } else { | |
2074 | alloc_profile = info->avail_metadata_alloc_bits & | |
2075 | info->metadata_alloc_profile; | |
2076 | data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile; | |
2077 | } | |
2078 | again: | |
2079 | data = reduce_alloc_profile(root, data); | |
2080 | /* | |
2081 | * the only place that sets empty_size is btrfs_realloc_node, which | |
2082 | * is not called recursively on allocations | |
2083 | */ | |
2084 | if (empty_size || root->ref_cows) { | |
2085 | if (!(data & BTRFS_BLOCK_GROUP_METADATA)) { | |
2086 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, | |
2087 | 2 * 1024 * 1024, | |
2088 | BTRFS_BLOCK_GROUP_METADATA | | |
2089 | (info->metadata_alloc_profile & | |
2090 | info->avail_metadata_alloc_bits), 0); | |
2091 | BUG_ON(ret); | |
2092 | } | |
2093 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, | |
2094 | num_bytes + 2 * 1024 * 1024, data, 0); | |
2095 | BUG_ON(ret); | |
2096 | } | |
2097 | ||
2098 | WARN_ON(num_bytes < root->sectorsize); | |
2099 | ret = find_free_extent(trans, root, num_bytes, empty_size, | |
2100 | search_start, search_end, hint_byte, ins, | |
2101 | trans->alloc_exclude_start, | |
2102 | trans->alloc_exclude_nr, data); | |
2103 | ||
2104 | if (ret == -ENOSPC && num_bytes > min_alloc_size) { | |
2105 | num_bytes = num_bytes >> 1; | |
2106 | num_bytes = max(num_bytes, min_alloc_size); | |
2107 | do_chunk_alloc(trans, root->fs_info->extent_root, | |
2108 | num_bytes, data, 1); | |
2109 | goto again; | |
2110 | } | |
2111 | if (ret) { | |
2112 | printk("allocation failed flags %Lu\n", data); | |
2113 | BUG(); | |
2114 | } | |
2115 | clear_extent_dirty(&root->fs_info->free_space_cache, | |
2116 | ins->objectid, ins->objectid + ins->offset - 1, | |
2117 | GFP_NOFS); | |
2118 | return 0; | |
2119 | } | |
2120 | ||
2121 | int btrfs_reserve_extent(struct btrfs_trans_handle *trans, | |
2122 | struct btrfs_root *root, | |
2123 | u64 num_bytes, u64 min_alloc_size, | |
2124 | u64 empty_size, u64 hint_byte, | |
2125 | u64 search_end, struct btrfs_key *ins, | |
2126 | u64 data) | |
2127 | { | |
2128 | int ret; | |
2129 | maybe_lock_mutex(root); | |
2130 | ret = __btrfs_reserve_extent(trans, root, num_bytes, min_alloc_size, | |
2131 | empty_size, hint_byte, search_end, ins, | |
2132 | data); | |
2133 | maybe_unlock_mutex(root); | |
2134 | return ret; | |
2135 | } | |
2136 | ||
2137 | static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans, | |
2138 | struct btrfs_root *root, | |
2139 | u64 root_objectid, u64 ref_generation, | |
2140 | u64 owner, u64 owner_offset, | |
2141 | struct btrfs_key *ins) | |
2142 | { | |
2143 | int ret; | |
2144 | int pending_ret; | |
2145 | u64 super_used; | |
2146 | u64 root_used; | |
2147 | u64 num_bytes = ins->offset; | |
2148 | u32 sizes[2]; | |
2149 | struct btrfs_fs_info *info = root->fs_info; | |
2150 | struct btrfs_root *extent_root = info->extent_root; | |
2151 | struct btrfs_extent_item *extent_item; | |
2152 | struct btrfs_extent_ref *ref; | |
2153 | struct btrfs_path *path; | |
2154 | struct btrfs_key keys[2]; | |
2155 | ||
2156 | /* block accounting for super block */ | |
2157 | spin_lock_irq(&info->delalloc_lock); | |
2158 | super_used = btrfs_super_bytes_used(&info->super_copy); | |
2159 | btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes); | |
2160 | spin_unlock_irq(&info->delalloc_lock); | |
2161 | ||
2162 | /* block accounting for root item */ | |
2163 | root_used = btrfs_root_used(&root->root_item); | |
2164 | btrfs_set_root_used(&root->root_item, root_used + num_bytes); | |
2165 | ||
2166 | if (root == extent_root) { | |
2167 | set_extent_bits(&root->fs_info->extent_ins, ins->objectid, | |
2168 | ins->objectid + ins->offset - 1, | |
2169 | EXTENT_LOCKED, GFP_NOFS); | |
2170 | goto update_block; | |
2171 | } | |
2172 | ||
2173 | memcpy(&keys[0], ins, sizeof(*ins)); | |
2174 | keys[1].offset = hash_extent_ref(root_objectid, ref_generation, | |
2175 | owner, owner_offset); | |
2176 | keys[1].objectid = ins->objectid; | |
2177 | keys[1].type = BTRFS_EXTENT_REF_KEY; | |
2178 | sizes[0] = sizeof(*extent_item); | |
2179 | sizes[1] = sizeof(*ref); | |
2180 | ||
2181 | path = btrfs_alloc_path(); | |
2182 | BUG_ON(!path); | |
2183 | ||
2184 | ret = btrfs_insert_empty_items(trans, extent_root, path, keys, | |
2185 | sizes, 2); | |
2186 | ||
2187 | BUG_ON(ret); | |
2188 | extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
2189 | struct btrfs_extent_item); | |
2190 | btrfs_set_extent_refs(path->nodes[0], extent_item, 1); | |
2191 | ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1, | |
2192 | struct btrfs_extent_ref); | |
2193 | ||
2194 | btrfs_set_ref_root(path->nodes[0], ref, root_objectid); | |
2195 | btrfs_set_ref_generation(path->nodes[0], ref, ref_generation); | |
2196 | btrfs_set_ref_objectid(path->nodes[0], ref, owner); | |
2197 | btrfs_set_ref_offset(path->nodes[0], ref, owner_offset); | |
2198 | ||
2199 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
2200 | ||
2201 | trans->alloc_exclude_start = 0; | |
2202 | trans->alloc_exclude_nr = 0; | |
2203 | btrfs_free_path(path); | |
2204 | finish_current_insert(trans, extent_root); | |
2205 | pending_ret = del_pending_extents(trans, extent_root); | |
2206 | ||
2207 | if (ret) | |
2208 | goto out; | |
2209 | if (pending_ret) { | |
2210 | ret = pending_ret; | |
2211 | goto out; | |
2212 | } | |
2213 | ||
2214 | update_block: | |
2215 | ret = update_block_group(trans, root, ins->objectid, ins->offset, 1, 0); | |
2216 | if (ret) { | |
2217 | printk("update block group failed for %Lu %Lu\n", | |
2218 | ins->objectid, ins->offset); | |
2219 | BUG(); | |
2220 | } | |
2221 | out: | |
2222 | return ret; | |
2223 | } | |
2224 | ||
2225 | int btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans, | |
2226 | struct btrfs_root *root, | |
2227 | u64 root_objectid, u64 ref_generation, | |
2228 | u64 owner, u64 owner_offset, | |
2229 | struct btrfs_key *ins) | |
2230 | { | |
2231 | int ret; | |
2232 | maybe_lock_mutex(root); | |
2233 | ret = __btrfs_alloc_reserved_extent(trans, root, root_objectid, | |
2234 | ref_generation, owner, | |
2235 | owner_offset, ins); | |
2236 | maybe_unlock_mutex(root); | |
2237 | return ret; | |
2238 | } | |
2239 | /* | |
2240 | * finds a free extent and does all the dirty work required for allocation | |
2241 | * returns the key for the extent through ins, and a tree buffer for | |
2242 | * the first block of the extent through buf. | |
2243 | * | |
2244 | * returns 0 if everything worked, non-zero otherwise. | |
2245 | */ | |
2246 | int btrfs_alloc_extent(struct btrfs_trans_handle *trans, | |
2247 | struct btrfs_root *root, | |
2248 | u64 num_bytes, u64 min_alloc_size, | |
2249 | u64 root_objectid, u64 ref_generation, | |
2250 | u64 owner, u64 owner_offset, | |
2251 | u64 empty_size, u64 hint_byte, | |
2252 | u64 search_end, struct btrfs_key *ins, u64 data) | |
2253 | { | |
2254 | int ret; | |
2255 | ||
2256 | maybe_lock_mutex(root); | |
2257 | ||
2258 | ret = __btrfs_reserve_extent(trans, root, num_bytes, | |
2259 | min_alloc_size, empty_size, hint_byte, | |
2260 | search_end, ins, data); | |
2261 | BUG_ON(ret); | |
2262 | ret = __btrfs_alloc_reserved_extent(trans, root, root_objectid, | |
2263 | ref_generation, owner, | |
2264 | owner_offset, ins); | |
2265 | BUG_ON(ret); | |
2266 | ||
2267 | maybe_unlock_mutex(root); | |
2268 | return ret; | |
2269 | } | |
2270 | /* | |
2271 | * helper function to allocate a block for a given tree | |
2272 | * returns the tree buffer or NULL. | |
2273 | */ | |
2274 | struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans, | |
2275 | struct btrfs_root *root, | |
2276 | u32 blocksize, | |
2277 | u64 root_objectid, | |
2278 | u64 ref_generation, | |
2279 | u64 first_objectid, | |
2280 | int level, | |
2281 | u64 hint, | |
2282 | u64 empty_size) | |
2283 | { | |
2284 | struct btrfs_key ins; | |
2285 | int ret; | |
2286 | struct extent_buffer *buf; | |
2287 | ||
2288 | ret = btrfs_alloc_extent(trans, root, blocksize, blocksize, | |
2289 | root_objectid, ref_generation, | |
2290 | level, first_objectid, empty_size, hint, | |
2291 | (u64)-1, &ins, 0); | |
2292 | if (ret) { | |
2293 | BUG_ON(ret > 0); | |
2294 | return ERR_PTR(ret); | |
2295 | } | |
2296 | buf = btrfs_find_create_tree_block(root, ins.objectid, blocksize); | |
2297 | if (!buf) { | |
2298 | btrfs_free_extent(trans, root, ins.objectid, blocksize, | |
2299 | root->root_key.objectid, ref_generation, | |
2300 | 0, 0, 0); | |
2301 | return ERR_PTR(-ENOMEM); | |
2302 | } | |
2303 | btrfs_set_header_generation(buf, trans->transid); | |
2304 | btrfs_tree_lock(buf); | |
2305 | clean_tree_block(trans, root, buf); | |
2306 | btrfs_set_buffer_uptodate(buf); | |
2307 | ||
2308 | if (PageDirty(buf->first_page)) { | |
2309 | printk("page %lu dirty\n", buf->first_page->index); | |
2310 | WARN_ON(1); | |
2311 | } | |
2312 | ||
2313 | set_extent_dirty(&trans->transaction->dirty_pages, buf->start, | |
2314 | buf->start + buf->len - 1, GFP_NOFS); | |
2315 | trans->blocks_used++; | |
2316 | return buf; | |
2317 | } | |
2318 | ||
2319 | static int noinline drop_leaf_ref_no_cache(struct btrfs_trans_handle *trans, | |
2320 | struct btrfs_root *root, | |
2321 | struct extent_buffer *leaf) | |
2322 | { | |
2323 | u64 leaf_owner; | |
2324 | u64 leaf_generation; | |
2325 | struct btrfs_key key; | |
2326 | struct btrfs_file_extent_item *fi; | |
2327 | int i; | |
2328 | int nritems; | |
2329 | int ret; | |
2330 | ||
2331 | BUG_ON(!btrfs_is_leaf(leaf)); | |
2332 | nritems = btrfs_header_nritems(leaf); | |
2333 | leaf_owner = btrfs_header_owner(leaf); | |
2334 | leaf_generation = btrfs_header_generation(leaf); | |
2335 | ||
2336 | for (i = 0; i < nritems; i++) { | |
2337 | u64 disk_bytenr; | |
2338 | cond_resched(); | |
2339 | ||
2340 | btrfs_item_key_to_cpu(leaf, &key, i); | |
2341 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) | |
2342 | continue; | |
2343 | fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); | |
2344 | if (btrfs_file_extent_type(leaf, fi) == | |
2345 | BTRFS_FILE_EXTENT_INLINE) | |
2346 | continue; | |
2347 | /* | |
2348 | * FIXME make sure to insert a trans record that | |
2349 | * repeats the snapshot del on crash | |
2350 | */ | |
2351 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); | |
2352 | if (disk_bytenr == 0) | |
2353 | continue; | |
2354 | ||
2355 | mutex_lock(&root->fs_info->alloc_mutex); | |
2356 | ret = __btrfs_free_extent(trans, root, disk_bytenr, | |
2357 | btrfs_file_extent_disk_num_bytes(leaf, fi), | |
2358 | leaf_owner, leaf_generation, | |
2359 | key.objectid, key.offset, 0); | |
2360 | mutex_unlock(&root->fs_info->alloc_mutex); | |
2361 | BUG_ON(ret); | |
2362 | } | |
2363 | return 0; | |
2364 | } | |
2365 | ||
2366 | static int noinline drop_leaf_ref(struct btrfs_trans_handle *trans, | |
2367 | struct btrfs_root *root, | |
2368 | struct btrfs_leaf_ref *ref) | |
2369 | { | |
2370 | int i; | |
2371 | int ret; | |
2372 | struct btrfs_extent_info *info = ref->extents; | |
2373 | ||
2374 | for (i = 0; i < ref->nritems; i++) { | |
2375 | mutex_lock(&root->fs_info->alloc_mutex); | |
2376 | ret = __btrfs_free_extent(trans, root, | |
2377 | info->bytenr, info->num_bytes, | |
2378 | ref->owner, ref->generation, | |
2379 | info->objectid, info->offset, 0); | |
2380 | mutex_unlock(&root->fs_info->alloc_mutex); | |
2381 | BUG_ON(ret); | |
2382 | info++; | |
2383 | } | |
2384 | ||
2385 | return 0; | |
2386 | } | |
2387 | ||
2388 | static void noinline reada_walk_down(struct btrfs_root *root, | |
2389 | struct extent_buffer *node, | |
2390 | int slot) | |
2391 | { | |
2392 | u64 bytenr; | |
2393 | u64 last = 0; | |
2394 | u32 nritems; | |
2395 | u32 refs; | |
2396 | u32 blocksize; | |
2397 | int ret; | |
2398 | int i; | |
2399 | int level; | |
2400 | int skipped = 0; | |
2401 | ||
2402 | nritems = btrfs_header_nritems(node); | |
2403 | level = btrfs_header_level(node); | |
2404 | if (level) | |
2405 | return; | |
2406 | ||
2407 | for (i = slot; i < nritems && skipped < 32; i++) { | |
2408 | bytenr = btrfs_node_blockptr(node, i); | |
2409 | if (last && ((bytenr > last && bytenr - last > 32 * 1024) || | |
2410 | (last > bytenr && last - bytenr > 32 * 1024))) { | |
2411 | skipped++; | |
2412 | continue; | |
2413 | } | |
2414 | blocksize = btrfs_level_size(root, level - 1); | |
2415 | if (i != slot) { | |
2416 | ret = lookup_extent_ref(NULL, root, bytenr, | |
2417 | blocksize, &refs); | |
2418 | BUG_ON(ret); | |
2419 | if (refs != 1) { | |
2420 | skipped++; | |
2421 | continue; | |
2422 | } | |
2423 | } | |
2424 | ret = readahead_tree_block(root, bytenr, blocksize, | |
2425 | btrfs_node_ptr_generation(node, i)); | |
2426 | last = bytenr + blocksize; | |
2427 | cond_resched(); | |
2428 | if (ret) | |
2429 | break; | |
2430 | } | |
2431 | } | |
2432 | ||
2433 | int drop_snap_lookup_refcount(struct btrfs_root *root, u64 start, u64 len, | |
2434 | u32 *refs) | |
2435 | { | |
2436 | int ret; | |
2437 | ||
2438 | ret = lookup_extent_ref(NULL, root, start, len, refs); | |
2439 | BUG_ON(ret); | |
2440 | ||
2441 | #if 0 // some debugging code in case we see problems here | |
2442 | /* if the refs count is one, it won't get increased again. But | |
2443 | * if the ref count is > 1, someone may be decreasing it at | |
2444 | * the same time we are. | |
2445 | */ | |
2446 | if (*refs != 1) { | |
2447 | struct extent_buffer *eb = NULL; | |
2448 | eb = btrfs_find_create_tree_block(root, start, len); | |
2449 | if (eb) | |
2450 | btrfs_tree_lock(eb); | |
2451 | ||
2452 | mutex_lock(&root->fs_info->alloc_mutex); | |
2453 | ret = lookup_extent_ref(NULL, root, start, len, refs); | |
2454 | BUG_ON(ret); | |
2455 | mutex_unlock(&root->fs_info->alloc_mutex); | |
2456 | ||
2457 | if (eb) { | |
2458 | btrfs_tree_unlock(eb); | |
2459 | free_extent_buffer(eb); | |
2460 | } | |
2461 | if (*refs == 1) { | |
2462 | printk("block %llu went down to one during drop_snap\n", | |
2463 | (unsigned long long)start); | |
2464 | } | |
2465 | ||
2466 | } | |
2467 | #endif | |
2468 | ||
2469 | cond_resched(); | |
2470 | return ret; | |
2471 | } | |
2472 | ||
2473 | /* | |
2474 | * helper function for drop_snapshot, this walks down the tree dropping ref | |
2475 | * counts as it goes. | |
2476 | */ | |
2477 | static int noinline walk_down_tree(struct btrfs_trans_handle *trans, | |
2478 | struct btrfs_root *root, | |
2479 | struct btrfs_path *path, int *level) | |
2480 | { | |
2481 | u64 root_owner; | |
2482 | u64 root_gen; | |
2483 | u64 bytenr; | |
2484 | u64 ptr_gen; | |
2485 | struct extent_buffer *next; | |
2486 | struct extent_buffer *cur; | |
2487 | struct extent_buffer *parent; | |
2488 | struct btrfs_leaf_ref *ref; | |
2489 | u32 blocksize; | |
2490 | int ret; | |
2491 | u32 refs; | |
2492 | ||
2493 | WARN_ON(*level < 0); | |
2494 | WARN_ON(*level >= BTRFS_MAX_LEVEL); | |
2495 | ret = drop_snap_lookup_refcount(root, path->nodes[*level]->start, | |
2496 | path->nodes[*level]->len, &refs); | |
2497 | BUG_ON(ret); | |
2498 | if (refs > 1) | |
2499 | goto out; | |
2500 | ||
2501 | /* | |
2502 | * walk down to the last node level and free all the leaves | |
2503 | */ | |
2504 | while(*level >= 0) { | |
2505 | WARN_ON(*level < 0); | |
2506 | WARN_ON(*level >= BTRFS_MAX_LEVEL); | |
2507 | cur = path->nodes[*level]; | |
2508 | ||
2509 | if (btrfs_header_level(cur) != *level) | |
2510 | WARN_ON(1); | |
2511 | ||
2512 | if (path->slots[*level] >= | |
2513 | btrfs_header_nritems(cur)) | |
2514 | break; | |
2515 | if (*level == 0) { | |
2516 | ret = drop_leaf_ref_no_cache(trans, root, cur); | |
2517 | BUG_ON(ret); | |
2518 | break; | |
2519 | } | |
2520 | bytenr = btrfs_node_blockptr(cur, path->slots[*level]); | |
2521 | ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]); | |
2522 | blocksize = btrfs_level_size(root, *level - 1); | |
2523 | ||
2524 | ret = drop_snap_lookup_refcount(root, bytenr, blocksize, &refs); | |
2525 | BUG_ON(ret); | |
2526 | if (refs != 1) { | |
2527 | parent = path->nodes[*level]; | |
2528 | root_owner = btrfs_header_owner(parent); | |
2529 | root_gen = btrfs_header_generation(parent); | |
2530 | path->slots[*level]++; | |
2531 | ||
2532 | mutex_lock(&root->fs_info->alloc_mutex); | |
2533 | ret = __btrfs_free_extent(trans, root, bytenr, | |
2534 | blocksize, root_owner, | |
2535 | root_gen, 0, 0, 1); | |
2536 | BUG_ON(ret); | |
2537 | mutex_unlock(&root->fs_info->alloc_mutex); | |
2538 | ||
2539 | atomic_inc(&root->fs_info->throttle_gen); | |
2540 | wake_up(&root->fs_info->transaction_throttle); | |
2541 | ||
2542 | continue; | |
2543 | } | |
2544 | /* | |
2545 | * at this point, we have a single ref, and since the | |
2546 | * only place referencing this extent is a dead root | |
2547 | * the reference count should never go higher. | |
2548 | * So, we don't need to check it again | |
2549 | */ | |
2550 | if (*level == 1) { | |
2551 | struct btrfs_key key; | |
2552 | btrfs_node_key_to_cpu(cur, &key, path->slots[*level]); | |
2553 | ref = btrfs_lookup_leaf_ref(root, bytenr); | |
2554 | if (ref) { | |
2555 | ret = drop_leaf_ref(trans, root, ref); | |
2556 | BUG_ON(ret); | |
2557 | btrfs_remove_leaf_ref(root, ref); | |
2558 | btrfs_free_leaf_ref(root, ref); | |
2559 | *level = 0; | |
2560 | break; | |
2561 | } | |
2562 | if (printk_ratelimit()) | |
2563 | printk("leaf ref miss for bytenr %llu\n", | |
2564 | (unsigned long long)bytenr); | |
2565 | } | |
2566 | next = btrfs_find_tree_block(root, bytenr, blocksize); | |
2567 | if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) { | |
2568 | free_extent_buffer(next); | |
2569 | ||
2570 | if (path->slots[*level] == 0) | |
2571 | reada_walk_down(root, cur, path->slots[*level]); | |
2572 | next = read_tree_block(root, bytenr, blocksize, | |
2573 | ptr_gen); | |
2574 | cond_resched(); | |
2575 | #if 0 | |
2576 | /* | |
2577 | * this is a debugging check and can go away | |
2578 | * the ref should never go all the way down to 1 | |
2579 | * at this point | |
2580 | */ | |
2581 | ret = lookup_extent_ref(NULL, root, bytenr, blocksize, | |
2582 | &refs); | |
2583 | BUG_ON(ret); | |
2584 | WARN_ON(refs != 1); | |
2585 | #endif | |
2586 | } | |
2587 | WARN_ON(*level <= 0); | |
2588 | if (path->nodes[*level-1]) | |
2589 | free_extent_buffer(path->nodes[*level-1]); | |
2590 | path->nodes[*level-1] = next; | |
2591 | *level = btrfs_header_level(next); | |
2592 | path->slots[*level] = 0; | |
2593 | } | |
2594 | out: | |
2595 | WARN_ON(*level < 0); | |
2596 | WARN_ON(*level >= BTRFS_MAX_LEVEL); | |
2597 | ||
2598 | if (path->nodes[*level] == root->node) { | |
2599 | parent = path->nodes[*level]; | |
2600 | bytenr = path->nodes[*level]->start; | |
2601 | } else { | |
2602 | parent = path->nodes[*level + 1]; | |
2603 | bytenr = btrfs_node_blockptr(parent, path->slots[*level + 1]); | |
2604 | } | |
2605 | ||
2606 | blocksize = btrfs_level_size(root, *level); | |
2607 | root_owner = btrfs_header_owner(parent); | |
2608 | root_gen = btrfs_header_generation(parent); | |
2609 | ||
2610 | mutex_lock(&root->fs_info->alloc_mutex); | |
2611 | ret = __btrfs_free_extent(trans, root, bytenr, blocksize, | |
2612 | root_owner, root_gen, 0, 0, 1); | |
2613 | free_extent_buffer(path->nodes[*level]); | |
2614 | path->nodes[*level] = NULL; | |
2615 | *level += 1; | |
2616 | BUG_ON(ret); | |
2617 | mutex_unlock(&root->fs_info->alloc_mutex); | |
2618 | ||
2619 | cond_resched(); | |
2620 | return 0; | |
2621 | } | |
2622 | ||
2623 | /* | |
2624 | * helper for dropping snapshots. This walks back up the tree in the path | |
2625 | * to find the first node higher up where we haven't yet gone through | |
2626 | * all the slots | |
2627 | */ | |
2628 | static int noinline walk_up_tree(struct btrfs_trans_handle *trans, | |
2629 | struct btrfs_root *root, | |
2630 | struct btrfs_path *path, int *level) | |
2631 | { | |
2632 | u64 root_owner; | |
2633 | u64 root_gen; | |
2634 | struct btrfs_root_item *root_item = &root->root_item; | |
2635 | int i; | |
2636 | int slot; | |
2637 | int ret; | |
2638 | ||
2639 | for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) { | |
2640 | slot = path->slots[i]; | |
2641 | if (slot < btrfs_header_nritems(path->nodes[i]) - 1) { | |
2642 | struct extent_buffer *node; | |
2643 | struct btrfs_disk_key disk_key; | |
2644 | node = path->nodes[i]; | |
2645 | path->slots[i]++; | |
2646 | *level = i; | |
2647 | WARN_ON(*level == 0); | |
2648 | btrfs_node_key(node, &disk_key, path->slots[i]); | |
2649 | memcpy(&root_item->drop_progress, | |
2650 | &disk_key, sizeof(disk_key)); | |
2651 | root_item->drop_level = i; | |
2652 | return 0; | |
2653 | } else { | |
2654 | if (path->nodes[*level] == root->node) { | |
2655 | root_owner = root->root_key.objectid; | |
2656 | root_gen = | |
2657 | btrfs_header_generation(path->nodes[*level]); | |
2658 | } else { | |
2659 | struct extent_buffer *node; | |
2660 | node = path->nodes[*level + 1]; | |
2661 | root_owner = btrfs_header_owner(node); | |
2662 | root_gen = btrfs_header_generation(node); | |
2663 | } | |
2664 | ret = btrfs_free_extent(trans, root, | |
2665 | path->nodes[*level]->start, | |
2666 | path->nodes[*level]->len, | |
2667 | root_owner, root_gen, 0, 0, 1); | |
2668 | BUG_ON(ret); | |
2669 | free_extent_buffer(path->nodes[*level]); | |
2670 | path->nodes[*level] = NULL; | |
2671 | *level = i + 1; | |
2672 | } | |
2673 | } | |
2674 | return 1; | |
2675 | } | |
2676 | ||
2677 | /* | |
2678 | * drop the reference count on the tree rooted at 'snap'. This traverses | |
2679 | * the tree freeing any blocks that have a ref count of zero after being | |
2680 | * decremented. | |
2681 | */ | |
2682 | int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root | |
2683 | *root) | |
2684 | { | |
2685 | int ret = 0; | |
2686 | int wret; | |
2687 | int level; | |
2688 | struct btrfs_path *path; | |
2689 | int i; | |
2690 | int orig_level; | |
2691 | struct btrfs_root_item *root_item = &root->root_item; | |
2692 | ||
2693 | WARN_ON(!mutex_is_locked(&root->fs_info->drop_mutex)); | |
2694 | path = btrfs_alloc_path(); | |
2695 | BUG_ON(!path); | |
2696 | ||
2697 | level = btrfs_header_level(root->node); | |
2698 | orig_level = level; | |
2699 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { | |
2700 | path->nodes[level] = root->node; | |
2701 | extent_buffer_get(root->node); | |
2702 | path->slots[level] = 0; | |
2703 | } else { | |
2704 | struct btrfs_key key; | |
2705 | struct btrfs_disk_key found_key; | |
2706 | struct extent_buffer *node; | |
2707 | ||
2708 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); | |
2709 | level = root_item->drop_level; | |
2710 | path->lowest_level = level; | |
2711 | wret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2712 | if (wret < 0) { | |
2713 | ret = wret; | |
2714 | goto out; | |
2715 | } | |
2716 | node = path->nodes[level]; | |
2717 | btrfs_node_key(node, &found_key, path->slots[level]); | |
2718 | WARN_ON(memcmp(&found_key, &root_item->drop_progress, | |
2719 | sizeof(found_key))); | |
2720 | /* | |
2721 | * unlock our path, this is safe because only this | |
2722 | * function is allowed to delete this snapshot | |
2723 | */ | |
2724 | for (i = 0; i < BTRFS_MAX_LEVEL; i++) { | |
2725 | if (path->nodes[i] && path->locks[i]) { | |
2726 | path->locks[i] = 0; | |
2727 | btrfs_tree_unlock(path->nodes[i]); | |
2728 | } | |
2729 | } | |
2730 | } | |
2731 | while(1) { | |
2732 | wret = walk_down_tree(trans, root, path, &level); | |
2733 | if (wret > 0) | |
2734 | break; | |
2735 | if (wret < 0) | |
2736 | ret = wret; | |
2737 | ||
2738 | wret = walk_up_tree(trans, root, path, &level); | |
2739 | if (wret > 0) | |
2740 | break; | |
2741 | if (wret < 0) | |
2742 | ret = wret; | |
2743 | if (trans->transaction->in_commit) { | |
2744 | ret = -EAGAIN; | |
2745 | break; | |
2746 | } | |
2747 | atomic_inc(&root->fs_info->throttle_gen); | |
2748 | wake_up(&root->fs_info->transaction_throttle); | |
2749 | } | |
2750 | for (i = 0; i <= orig_level; i++) { | |
2751 | if (path->nodes[i]) { | |
2752 | free_extent_buffer(path->nodes[i]); | |
2753 | path->nodes[i] = NULL; | |
2754 | } | |
2755 | } | |
2756 | out: | |
2757 | btrfs_free_path(path); | |
2758 | return ret; | |
2759 | } | |
2760 | ||
2761 | int btrfs_free_block_groups(struct btrfs_fs_info *info) | |
2762 | { | |
2763 | u64 start; | |
2764 | u64 end; | |
2765 | u64 ptr; | |
2766 | int ret; | |
2767 | ||
2768 | mutex_lock(&info->alloc_mutex); | |
2769 | while(1) { | |
2770 | ret = find_first_extent_bit(&info->block_group_cache, 0, | |
2771 | &start, &end, (unsigned int)-1); | |
2772 | if (ret) | |
2773 | break; | |
2774 | ret = get_state_private(&info->block_group_cache, start, &ptr); | |
2775 | if (!ret) | |
2776 | kfree((void *)(unsigned long)ptr); | |
2777 | clear_extent_bits(&info->block_group_cache, start, | |
2778 | end, (unsigned int)-1, GFP_NOFS); | |
2779 | } | |
2780 | while(1) { | |
2781 | ret = find_first_extent_bit(&info->free_space_cache, 0, | |
2782 | &start, &end, EXTENT_DIRTY); | |
2783 | if (ret) | |
2784 | break; | |
2785 | clear_extent_dirty(&info->free_space_cache, start, | |
2786 | end, GFP_NOFS); | |
2787 | } | |
2788 | mutex_unlock(&info->alloc_mutex); | |
2789 | return 0; | |
2790 | } | |
2791 | ||
2792 | static unsigned long calc_ra(unsigned long start, unsigned long last, | |
2793 | unsigned long nr) | |
2794 | { | |
2795 | return min(last, start + nr - 1); | |
2796 | } | |
2797 | ||
2798 | static int noinline relocate_inode_pages(struct inode *inode, u64 start, | |
2799 | u64 len) | |
2800 | { | |
2801 | u64 page_start; | |
2802 | u64 page_end; | |
2803 | unsigned long last_index; | |
2804 | unsigned long i; | |
2805 | struct page *page; | |
2806 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
2807 | struct file_ra_state *ra; | |
2808 | unsigned long total_read = 0; | |
2809 | unsigned long ra_pages; | |
2810 | struct btrfs_ordered_extent *ordered; | |
2811 | struct btrfs_trans_handle *trans; | |
2812 | ||
2813 | ra = kzalloc(sizeof(*ra), GFP_NOFS); | |
2814 | ||
2815 | mutex_lock(&inode->i_mutex); | |
2816 | i = start >> PAGE_CACHE_SHIFT; | |
2817 | last_index = (start + len - 1) >> PAGE_CACHE_SHIFT; | |
2818 | ||
2819 | ra_pages = BTRFS_I(inode)->root->fs_info->bdi.ra_pages; | |
2820 | ||
2821 | file_ra_state_init(ra, inode->i_mapping); | |
2822 | ||
2823 | for (; i <= last_index; i++) { | |
2824 | if (total_read % ra_pages == 0) { | |
2825 | btrfs_force_ra(inode->i_mapping, ra, NULL, i, | |
2826 | calc_ra(i, last_index, ra_pages)); | |
2827 | } | |
2828 | total_read++; | |
2829 | again: | |
2830 | if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode)) | |
2831 | goto truncate_racing; | |
2832 | page = grab_cache_page(inode->i_mapping, i); | |
2833 | if (!page) { | |
2834 | goto out_unlock; | |
2835 | } | |
2836 | if (!PageUptodate(page)) { | |
2837 | btrfs_readpage(NULL, page); | |
2838 | lock_page(page); | |
2839 | if (!PageUptodate(page)) { | |
2840 | unlock_page(page); | |
2841 | page_cache_release(page); | |
2842 | goto out_unlock; | |
2843 | } | |
2844 | } | |
2845 | wait_on_page_writeback(page); | |
2846 | ||
2847 | page_start = (u64)page->index << PAGE_CACHE_SHIFT; | |
2848 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
2849 | lock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
2850 | ||
2851 | ordered = btrfs_lookup_ordered_extent(inode, page_start); | |
2852 | if (ordered) { | |
2853 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
2854 | unlock_page(page); | |
2855 | page_cache_release(page); | |
2856 | btrfs_start_ordered_extent(inode, ordered, 1); | |
2857 | btrfs_put_ordered_extent(ordered); | |
2858 | goto again; | |
2859 | } | |
2860 | set_page_extent_mapped(page); | |
2861 | ||
2862 | /* | |
2863 | * make sure page_mkwrite is called for this page if userland | |
2864 | * wants to change it from mmap | |
2865 | */ | |
2866 | clear_page_dirty_for_io(page); | |
2867 | ||
2868 | set_extent_delalloc(io_tree, page_start, | |
2869 | page_end, GFP_NOFS); | |
2870 | set_page_dirty(page); | |
2871 | ||
2872 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); | |
2873 | unlock_page(page); | |
2874 | page_cache_release(page); | |
2875 | } | |
2876 | ||
2877 | out_unlock: | |
2878 | /* we have to start the IO in order to get the ordered extents | |
2879 | * instantiated. This allows the relocation to code to wait | |
2880 | * for all the ordered extents to hit the disk. | |
2881 | * | |
2882 | * Otherwise, it would constantly loop over the same extents | |
2883 | * because the old ones don't get deleted until the IO is | |
2884 | * started | |
2885 | */ | |
2886 | btrfs_fdatawrite_range(inode->i_mapping, start, start + len - 1, | |
2887 | WB_SYNC_NONE); | |
2888 | kfree(ra); | |
2889 | trans = btrfs_start_transaction(BTRFS_I(inode)->root, 1); | |
2890 | if (trans) { | |
2891 | btrfs_end_transaction(trans, BTRFS_I(inode)->root); | |
2892 | mark_inode_dirty(inode); | |
2893 | } | |
2894 | mutex_unlock(&inode->i_mutex); | |
2895 | return 0; | |
2896 | ||
2897 | truncate_racing: | |
2898 | vmtruncate(inode, inode->i_size); | |
2899 | balance_dirty_pages_ratelimited_nr(inode->i_mapping, | |
2900 | total_read); | |
2901 | goto out_unlock; | |
2902 | } | |
2903 | ||
2904 | /* | |
2905 | * The back references tell us which tree holds a ref on a block, | |
2906 | * but it is possible for the tree root field in the reference to | |
2907 | * reflect the original root before a snapshot was made. In this | |
2908 | * case we should search through all the children of a given root | |
2909 | * to find potential holders of references on a block. | |
2910 | * | |
2911 | * Instead, we do something a little less fancy and just search | |
2912 | * all the roots for a given key/block combination. | |
2913 | */ | |
2914 | static int find_root_for_ref(struct btrfs_root *root, | |
2915 | struct btrfs_path *path, | |
2916 | struct btrfs_key *key0, | |
2917 | int level, | |
2918 | int file_key, | |
2919 | struct btrfs_root **found_root, | |
2920 | u64 bytenr) | |
2921 | { | |
2922 | struct btrfs_key root_location; | |
2923 | struct btrfs_root *cur_root = *found_root; | |
2924 | struct btrfs_file_extent_item *file_extent; | |
2925 | u64 root_search_start = BTRFS_FS_TREE_OBJECTID; | |
2926 | u64 found_bytenr; | |
2927 | int ret; | |
2928 | ||
2929 | root_location.offset = (u64)-1; | |
2930 | root_location.type = BTRFS_ROOT_ITEM_KEY; | |
2931 | path->lowest_level = level; | |
2932 | path->reada = 0; | |
2933 | while(1) { | |
2934 | ret = btrfs_search_slot(NULL, cur_root, key0, path, 0, 0); | |
2935 | found_bytenr = 0; | |
2936 | if (ret == 0 && file_key) { | |
2937 | struct extent_buffer *leaf = path->nodes[0]; | |
2938 | file_extent = btrfs_item_ptr(leaf, path->slots[0], | |
2939 | struct btrfs_file_extent_item); | |
2940 | if (btrfs_file_extent_type(leaf, file_extent) == | |
2941 | BTRFS_FILE_EXTENT_REG) { | |
2942 | found_bytenr = | |
2943 | btrfs_file_extent_disk_bytenr(leaf, | |
2944 | file_extent); | |
2945 | } | |
2946 | } else if (!file_key) { | |
2947 | if (path->nodes[level]) | |
2948 | found_bytenr = path->nodes[level]->start; | |
2949 | } | |
2950 | ||
2951 | btrfs_release_path(cur_root, path); | |
2952 | ||
2953 | if (found_bytenr == bytenr) { | |
2954 | *found_root = cur_root; | |
2955 | ret = 0; | |
2956 | goto out; | |
2957 | } | |
2958 | ret = btrfs_search_root(root->fs_info->tree_root, | |
2959 | root_search_start, &root_search_start); | |
2960 | if (ret) | |
2961 | break; | |
2962 | ||
2963 | root_location.objectid = root_search_start; | |
2964 | cur_root = btrfs_read_fs_root_no_name(root->fs_info, | |
2965 | &root_location); | |
2966 | if (!cur_root) { | |
2967 | ret = 1; | |
2968 | break; | |
2969 | } | |
2970 | } | |
2971 | out: | |
2972 | path->lowest_level = 0; | |
2973 | return ret; | |
2974 | } | |
2975 | ||
2976 | /* | |
2977 | * note, this releases the path | |
2978 | */ | |
2979 | static int noinline relocate_one_reference(struct btrfs_root *extent_root, | |
2980 | struct btrfs_path *path, | |
2981 | struct btrfs_key *extent_key, | |
2982 | u64 *last_file_objectid, | |
2983 | u64 *last_file_offset, | |
2984 | u64 *last_file_root, | |
2985 | u64 last_extent) | |
2986 | { | |
2987 | struct inode *inode; | |
2988 | struct btrfs_root *found_root; | |
2989 | struct btrfs_key root_location; | |
2990 | struct btrfs_key found_key; | |
2991 | struct btrfs_extent_ref *ref; | |
2992 | u64 ref_root; | |
2993 | u64 ref_gen; | |
2994 | u64 ref_objectid; | |
2995 | u64 ref_offset; | |
2996 | int ret; | |
2997 | int level; | |
2998 | ||
2999 | WARN_ON(!mutex_is_locked(&extent_root->fs_info->alloc_mutex)); | |
3000 | ||
3001 | ref = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
3002 | struct btrfs_extent_ref); | |
3003 | ref_root = btrfs_ref_root(path->nodes[0], ref); | |
3004 | ref_gen = btrfs_ref_generation(path->nodes[0], ref); | |
3005 | ref_objectid = btrfs_ref_objectid(path->nodes[0], ref); | |
3006 | ref_offset = btrfs_ref_offset(path->nodes[0], ref); | |
3007 | btrfs_release_path(extent_root, path); | |
3008 | ||
3009 | root_location.objectid = ref_root; | |
3010 | if (ref_gen == 0) | |
3011 | root_location.offset = 0; | |
3012 | else | |
3013 | root_location.offset = (u64)-1; | |
3014 | root_location.type = BTRFS_ROOT_ITEM_KEY; | |
3015 | ||
3016 | found_root = btrfs_read_fs_root_no_name(extent_root->fs_info, | |
3017 | &root_location); | |
3018 | BUG_ON(!found_root); | |
3019 | mutex_unlock(&extent_root->fs_info->alloc_mutex); | |
3020 | ||
3021 | if (ref_objectid >= BTRFS_FIRST_FREE_OBJECTID) { | |
3022 | found_key.objectid = ref_objectid; | |
3023 | found_key.type = BTRFS_EXTENT_DATA_KEY; | |
3024 | found_key.offset = ref_offset; | |
3025 | level = 0; | |
3026 | ||
3027 | if (last_extent == extent_key->objectid && | |
3028 | *last_file_objectid == ref_objectid && | |
3029 | *last_file_offset == ref_offset && | |
3030 | *last_file_root == ref_root) | |
3031 | goto out; | |
3032 | ||
3033 | ret = find_root_for_ref(extent_root, path, &found_key, | |
3034 | level, 1, &found_root, | |
3035 | extent_key->objectid); | |
3036 | ||
3037 | if (ret) | |
3038 | goto out; | |
3039 | ||
3040 | if (last_extent == extent_key->objectid && | |
3041 | *last_file_objectid == ref_objectid && | |
3042 | *last_file_offset == ref_offset && | |
3043 | *last_file_root == ref_root) | |
3044 | goto out; | |
3045 | ||
3046 | inode = btrfs_iget_locked(extent_root->fs_info->sb, | |
3047 | ref_objectid, found_root); | |
3048 | if (inode->i_state & I_NEW) { | |
3049 | /* the inode and parent dir are two different roots */ | |
3050 | BTRFS_I(inode)->root = found_root; | |
3051 | BTRFS_I(inode)->location.objectid = ref_objectid; | |
3052 | BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY; | |
3053 | BTRFS_I(inode)->location.offset = 0; | |
3054 | btrfs_read_locked_inode(inode); | |
3055 | unlock_new_inode(inode); | |
3056 | ||
3057 | } | |
3058 | /* this can happen if the reference is not against | |
3059 | * the latest version of the tree root | |
3060 | */ | |
3061 | if (is_bad_inode(inode)) | |
3062 | goto out; | |
3063 | ||
3064 | *last_file_objectid = inode->i_ino; | |
3065 | *last_file_root = found_root->root_key.objectid; | |
3066 | *last_file_offset = ref_offset; | |
3067 | ||
3068 | relocate_inode_pages(inode, ref_offset, extent_key->offset); | |
3069 | iput(inode); | |
3070 | } else { | |
3071 | struct btrfs_trans_handle *trans; | |
3072 | struct extent_buffer *eb; | |
3073 | int needs_lock = 0; | |
3074 | ||
3075 | eb = read_tree_block(found_root, extent_key->objectid, | |
3076 | extent_key->offset, 0); | |
3077 | btrfs_tree_lock(eb); | |
3078 | level = btrfs_header_level(eb); | |
3079 | ||
3080 | if (level == 0) | |
3081 | btrfs_item_key_to_cpu(eb, &found_key, 0); | |
3082 | else | |
3083 | btrfs_node_key_to_cpu(eb, &found_key, 0); | |
3084 | ||
3085 | btrfs_tree_unlock(eb); | |
3086 | free_extent_buffer(eb); | |
3087 | ||
3088 | ret = find_root_for_ref(extent_root, path, &found_key, | |
3089 | level, 0, &found_root, | |
3090 | extent_key->objectid); | |
3091 | ||
3092 | if (ret) | |
3093 | goto out; | |
3094 | ||
3095 | /* | |
3096 | * right here almost anything could happen to our key, | |
3097 | * but that's ok. The cow below will either relocate it | |
3098 | * or someone else will have relocated it. Either way, | |
3099 | * it is in a different spot than it was before and | |
3100 | * we're happy. | |
3101 | */ | |
3102 | ||
3103 | trans = btrfs_start_transaction(found_root, 1); | |
3104 | ||
3105 | if (found_root == extent_root->fs_info->extent_root || | |
3106 | found_root == extent_root->fs_info->chunk_root || | |
3107 | found_root == extent_root->fs_info->dev_root) { | |
3108 | needs_lock = 1; | |
3109 | mutex_lock(&extent_root->fs_info->alloc_mutex); | |
3110 | } | |
3111 | ||
3112 | path->lowest_level = level; | |
3113 | path->reada = 2; | |
3114 | ret = btrfs_search_slot(trans, found_root, &found_key, path, | |
3115 | 0, 1); | |
3116 | path->lowest_level = 0; | |
3117 | btrfs_release_path(found_root, path); | |
3118 | ||
3119 | if (found_root == found_root->fs_info->extent_root) | |
3120 | btrfs_extent_post_op(trans, found_root); | |
3121 | if (needs_lock) | |
3122 | mutex_unlock(&extent_root->fs_info->alloc_mutex); | |
3123 | ||
3124 | btrfs_end_transaction(trans, found_root); | |
3125 | ||
3126 | } | |
3127 | out: | |
3128 | mutex_lock(&extent_root->fs_info->alloc_mutex); | |
3129 | return 0; | |
3130 | } | |
3131 | ||
3132 | static int noinline del_extent_zero(struct btrfs_root *extent_root, | |
3133 | struct btrfs_path *path, | |
3134 | struct btrfs_key *extent_key) | |
3135 | { | |
3136 | int ret; | |
3137 | struct btrfs_trans_handle *trans; | |
3138 | ||
3139 | trans = btrfs_start_transaction(extent_root, 1); | |
3140 | ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1); | |
3141 | if (ret > 0) { | |
3142 | ret = -EIO; | |
3143 | goto out; | |
3144 | } | |
3145 | if (ret < 0) | |
3146 | goto out; | |
3147 | ret = btrfs_del_item(trans, extent_root, path); | |
3148 | out: | |
3149 | btrfs_end_transaction(trans, extent_root); | |
3150 | return ret; | |
3151 | } | |
3152 | ||
3153 | static int noinline relocate_one_extent(struct btrfs_root *extent_root, | |
3154 | struct btrfs_path *path, | |
3155 | struct btrfs_key *extent_key) | |
3156 | { | |
3157 | struct btrfs_key key; | |
3158 | struct btrfs_key found_key; | |
3159 | struct extent_buffer *leaf; | |
3160 | u64 last_file_objectid = 0; | |
3161 | u64 last_file_root = 0; | |
3162 | u64 last_file_offset = (u64)-1; | |
3163 | u64 last_extent = 0; | |
3164 | u32 nritems; | |
3165 | u32 item_size; | |
3166 | int ret = 0; | |
3167 | ||
3168 | if (extent_key->objectid == 0) { | |
3169 | ret = del_extent_zero(extent_root, path, extent_key); | |
3170 | goto out; | |
3171 | } | |
3172 | key.objectid = extent_key->objectid; | |
3173 | key.type = BTRFS_EXTENT_REF_KEY; | |
3174 | key.offset = 0; | |
3175 | ||
3176 | while(1) { | |
3177 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); | |
3178 | ||
3179 | if (ret < 0) | |
3180 | goto out; | |
3181 | ||
3182 | ret = 0; | |
3183 | leaf = path->nodes[0]; | |
3184 | nritems = btrfs_header_nritems(leaf); | |
3185 | if (path->slots[0] == nritems) { | |
3186 | ret = btrfs_next_leaf(extent_root, path); | |
3187 | if (ret > 0) { | |
3188 | ret = 0; | |
3189 | goto out; | |
3190 | } | |
3191 | if (ret < 0) | |
3192 | goto out; | |
3193 | leaf = path->nodes[0]; | |
3194 | } | |
3195 | ||
3196 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
3197 | if (found_key.objectid != extent_key->objectid) { | |
3198 | break; | |
3199 | } | |
3200 | ||
3201 | if (found_key.type != BTRFS_EXTENT_REF_KEY) { | |
3202 | break; | |
3203 | } | |
3204 | ||
3205 | key.offset = found_key.offset + 1; | |
3206 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
3207 | ||
3208 | ret = relocate_one_reference(extent_root, path, extent_key, | |
3209 | &last_file_objectid, | |
3210 | &last_file_offset, | |
3211 | &last_file_root, last_extent); | |
3212 | if (ret) | |
3213 | goto out; | |
3214 | last_extent = extent_key->objectid; | |
3215 | } | |
3216 | ret = 0; | |
3217 | out: | |
3218 | btrfs_release_path(extent_root, path); | |
3219 | return ret; | |
3220 | } | |
3221 | ||
3222 | static u64 update_block_group_flags(struct btrfs_root *root, u64 flags) | |
3223 | { | |
3224 | u64 num_devices; | |
3225 | u64 stripped = BTRFS_BLOCK_GROUP_RAID0 | | |
3226 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10; | |
3227 | ||
3228 | num_devices = root->fs_info->fs_devices->num_devices; | |
3229 | if (num_devices == 1) { | |
3230 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
3231 | stripped = flags & ~stripped; | |
3232 | ||
3233 | /* turn raid0 into single device chunks */ | |
3234 | if (flags & BTRFS_BLOCK_GROUP_RAID0) | |
3235 | return stripped; | |
3236 | ||
3237 | /* turn mirroring into duplication */ | |
3238 | if (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
3239 | BTRFS_BLOCK_GROUP_RAID10)) | |
3240 | return stripped | BTRFS_BLOCK_GROUP_DUP; | |
3241 | return flags; | |
3242 | } else { | |
3243 | /* they already had raid on here, just return */ | |
3244 | if (flags & stripped) | |
3245 | return flags; | |
3246 | ||
3247 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
3248 | stripped = flags & ~stripped; | |
3249 | ||
3250 | /* switch duplicated blocks with raid1 */ | |
3251 | if (flags & BTRFS_BLOCK_GROUP_DUP) | |
3252 | return stripped | BTRFS_BLOCK_GROUP_RAID1; | |
3253 | ||
3254 | /* turn single device chunks into raid0 */ | |
3255 | return stripped | BTRFS_BLOCK_GROUP_RAID0; | |
3256 | } | |
3257 | return flags; | |
3258 | } | |
3259 | ||
3260 | int __alloc_chunk_for_shrink(struct btrfs_root *root, | |
3261 | struct btrfs_block_group_cache *shrink_block_group, | |
3262 | int force) | |
3263 | { | |
3264 | struct btrfs_trans_handle *trans; | |
3265 | u64 new_alloc_flags; | |
3266 | u64 calc; | |
3267 | ||
3268 | spin_lock(&shrink_block_group->lock); | |
3269 | if (btrfs_block_group_used(&shrink_block_group->item) > 0) { | |
3270 | spin_unlock(&shrink_block_group->lock); | |
3271 | mutex_unlock(&root->fs_info->alloc_mutex); | |
3272 | ||
3273 | trans = btrfs_start_transaction(root, 1); | |
3274 | mutex_lock(&root->fs_info->alloc_mutex); | |
3275 | spin_lock(&shrink_block_group->lock); | |
3276 | ||
3277 | new_alloc_flags = update_block_group_flags(root, | |
3278 | shrink_block_group->flags); | |
3279 | if (new_alloc_flags != shrink_block_group->flags) { | |
3280 | calc = | |
3281 | btrfs_block_group_used(&shrink_block_group->item); | |
3282 | } else { | |
3283 | calc = shrink_block_group->key.offset; | |
3284 | } | |
3285 | spin_unlock(&shrink_block_group->lock); | |
3286 | ||
3287 | do_chunk_alloc(trans, root->fs_info->extent_root, | |
3288 | calc + 2 * 1024 * 1024, new_alloc_flags, force); | |
3289 | ||
3290 | mutex_unlock(&root->fs_info->alloc_mutex); | |
3291 | btrfs_end_transaction(trans, root); | |
3292 | mutex_lock(&root->fs_info->alloc_mutex); | |
3293 | } else | |
3294 | spin_unlock(&shrink_block_group->lock); | |
3295 | return 0; | |
3296 | } | |
3297 | ||
3298 | int btrfs_shrink_extent_tree(struct btrfs_root *root, u64 shrink_start) | |
3299 | { | |
3300 | struct btrfs_trans_handle *trans; | |
3301 | struct btrfs_root *tree_root = root->fs_info->tree_root; | |
3302 | struct btrfs_path *path; | |
3303 | u64 cur_byte; | |
3304 | u64 total_found; | |
3305 | u64 shrink_last_byte; | |
3306 | struct btrfs_block_group_cache *shrink_block_group; | |
3307 | struct btrfs_fs_info *info = root->fs_info; | |
3308 | struct btrfs_key key; | |
3309 | struct btrfs_key found_key; | |
3310 | struct extent_buffer *leaf; | |
3311 | u32 nritems; | |
3312 | int ret; | |
3313 | int progress; | |
3314 | ||
3315 | mutex_lock(&root->fs_info->alloc_mutex); | |
3316 | shrink_block_group = btrfs_lookup_block_group(root->fs_info, | |
3317 | shrink_start); | |
3318 | BUG_ON(!shrink_block_group); | |
3319 | ||
3320 | shrink_last_byte = shrink_block_group->key.objectid + | |
3321 | shrink_block_group->key.offset; | |
3322 | ||
3323 | shrink_block_group->space_info->total_bytes -= | |
3324 | shrink_block_group->key.offset; | |
3325 | path = btrfs_alloc_path(); | |
3326 | root = root->fs_info->extent_root; | |
3327 | path->reada = 2; | |
3328 | ||
3329 | printk("btrfs relocating block group %llu flags %llu\n", | |
3330 | (unsigned long long)shrink_start, | |
3331 | (unsigned long long)shrink_block_group->flags); | |
3332 | ||
3333 | __alloc_chunk_for_shrink(root, shrink_block_group, 1); | |
3334 | ||
3335 | again: | |
3336 | ||
3337 | shrink_block_group->ro = 1; | |
3338 | ||
3339 | total_found = 0; | |
3340 | progress = 0; | |
3341 | key.objectid = shrink_start; | |
3342 | key.offset = 0; | |
3343 | key.type = 0; | |
3344 | cur_byte = key.objectid; | |
3345 | ||
3346 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
3347 | if (ret < 0) | |
3348 | goto out; | |
3349 | ||
3350 | ret = btrfs_previous_item(root, path, 0, BTRFS_EXTENT_ITEM_KEY); | |
3351 | if (ret < 0) | |
3352 | goto out; | |
3353 | ||
3354 | if (ret == 0) { | |
3355 | leaf = path->nodes[0]; | |
3356 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
3357 | if (found_key.objectid + found_key.offset > shrink_start && | |
3358 | found_key.objectid < shrink_last_byte) { | |
3359 | cur_byte = found_key.objectid; | |
3360 | key.objectid = cur_byte; | |
3361 | } | |
3362 | } | |
3363 | btrfs_release_path(root, path); | |
3364 | ||
3365 | while(1) { | |
3366 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
3367 | if (ret < 0) | |
3368 | goto out; | |
3369 | ||
3370 | next: | |
3371 | leaf = path->nodes[0]; | |
3372 | nritems = btrfs_header_nritems(leaf); | |
3373 | if (path->slots[0] >= nritems) { | |
3374 | ret = btrfs_next_leaf(root, path); | |
3375 | if (ret < 0) | |
3376 | goto out; | |
3377 | if (ret == 1) { | |
3378 | ret = 0; | |
3379 | break; | |
3380 | } | |
3381 | leaf = path->nodes[0]; | |
3382 | nritems = btrfs_header_nritems(leaf); | |
3383 | } | |
3384 | ||
3385 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
3386 | ||
3387 | if (found_key.objectid >= shrink_last_byte) | |
3388 | break; | |
3389 | ||
3390 | if (progress && need_resched()) { | |
3391 | memcpy(&key, &found_key, sizeof(key)); | |
3392 | cond_resched(); | |
3393 | btrfs_release_path(root, path); | |
3394 | btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
3395 | progress = 0; | |
3396 | goto next; | |
3397 | } | |
3398 | progress = 1; | |
3399 | ||
3400 | if (btrfs_key_type(&found_key) != BTRFS_EXTENT_ITEM_KEY || | |
3401 | found_key.objectid + found_key.offset <= cur_byte) { | |
3402 | memcpy(&key, &found_key, sizeof(key)); | |
3403 | key.offset++; | |
3404 | path->slots[0]++; | |
3405 | goto next; | |
3406 | } | |
3407 | ||
3408 | total_found++; | |
3409 | cur_byte = found_key.objectid + found_key.offset; | |
3410 | key.objectid = cur_byte; | |
3411 | btrfs_release_path(root, path); | |
3412 | ret = relocate_one_extent(root, path, &found_key); | |
3413 | __alloc_chunk_for_shrink(root, shrink_block_group, 0); | |
3414 | } | |
3415 | ||
3416 | btrfs_release_path(root, path); | |
3417 | ||
3418 | if (total_found > 0) { | |
3419 | printk("btrfs relocate found %llu last extent was %llu\n", | |
3420 | (unsigned long long)total_found, | |
3421 | (unsigned long long)found_key.objectid); | |
3422 | mutex_unlock(&root->fs_info->alloc_mutex); | |
3423 | trans = btrfs_start_transaction(tree_root, 1); | |
3424 | btrfs_commit_transaction(trans, tree_root); | |
3425 | ||
3426 | btrfs_clean_old_snapshots(tree_root); | |
3427 | ||
3428 | btrfs_wait_ordered_extents(tree_root); | |
3429 | ||
3430 | trans = btrfs_start_transaction(tree_root, 1); | |
3431 | btrfs_commit_transaction(trans, tree_root); | |
3432 | mutex_lock(&root->fs_info->alloc_mutex); | |
3433 | goto again; | |
3434 | } | |
3435 | ||
3436 | /* | |
3437 | * we've freed all the extents, now remove the block | |
3438 | * group item from the tree | |
3439 | */ | |
3440 | mutex_unlock(&root->fs_info->alloc_mutex); | |
3441 | ||
3442 | trans = btrfs_start_transaction(root, 1); | |
3443 | ||
3444 | mutex_lock(&root->fs_info->alloc_mutex); | |
3445 | memcpy(&key, &shrink_block_group->key, sizeof(key)); | |
3446 | ||
3447 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
3448 | if (ret > 0) | |
3449 | ret = -EIO; | |
3450 | if (ret < 0) { | |
3451 | btrfs_end_transaction(trans, root); | |
3452 | goto out; | |
3453 | } | |
3454 | ||
3455 | clear_extent_bits(&info->block_group_cache, key.objectid, | |
3456 | key.objectid + key.offset - 1, | |
3457 | (unsigned int)-1, GFP_NOFS); | |
3458 | ||
3459 | ||
3460 | clear_extent_bits(&info->free_space_cache, | |
3461 | key.objectid, key.objectid + key.offset - 1, | |
3462 | (unsigned int)-1, GFP_NOFS); | |
3463 | ||
3464 | memset(shrink_block_group, 0, sizeof(*shrink_block_group)); | |
3465 | kfree(shrink_block_group); | |
3466 | ||
3467 | btrfs_del_item(trans, root, path); | |
3468 | btrfs_release_path(root, path); | |
3469 | mutex_unlock(&root->fs_info->alloc_mutex); | |
3470 | btrfs_commit_transaction(trans, root); | |
3471 | ||
3472 | mutex_lock(&root->fs_info->alloc_mutex); | |
3473 | ||
3474 | /* the code to unpin extents might set a few bits in the free | |
3475 | * space cache for this range again | |
3476 | */ | |
3477 | clear_extent_bits(&info->free_space_cache, | |
3478 | key.objectid, key.objectid + key.offset - 1, | |
3479 | (unsigned int)-1, GFP_NOFS); | |
3480 | out: | |
3481 | btrfs_free_path(path); | |
3482 | mutex_unlock(&root->fs_info->alloc_mutex); | |
3483 | return ret; | |
3484 | } | |
3485 | ||
3486 | int find_first_block_group(struct btrfs_root *root, struct btrfs_path *path, | |
3487 | struct btrfs_key *key) | |
3488 | { | |
3489 | int ret = 0; | |
3490 | struct btrfs_key found_key; | |
3491 | struct extent_buffer *leaf; | |
3492 | int slot; | |
3493 | ||
3494 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); | |
3495 | if (ret < 0) | |
3496 | goto out; | |
3497 | ||
3498 | while(1) { | |
3499 | slot = path->slots[0]; | |
3500 | leaf = path->nodes[0]; | |
3501 | if (slot >= btrfs_header_nritems(leaf)) { | |
3502 | ret = btrfs_next_leaf(root, path); | |
3503 | if (ret == 0) | |
3504 | continue; | |
3505 | if (ret < 0) | |
3506 | goto out; | |
3507 | break; | |
3508 | } | |
3509 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
3510 | ||
3511 | if (found_key.objectid >= key->objectid && | |
3512 | found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) { | |
3513 | ret = 0; | |
3514 | goto out; | |
3515 | } | |
3516 | path->slots[0]++; | |
3517 | } | |
3518 | ret = -ENOENT; | |
3519 | out: | |
3520 | return ret; | |
3521 | } | |
3522 | ||
3523 | int btrfs_read_block_groups(struct btrfs_root *root) | |
3524 | { | |
3525 | struct btrfs_path *path; | |
3526 | int ret; | |
3527 | int bit; | |
3528 | struct btrfs_block_group_cache *cache; | |
3529 | struct btrfs_fs_info *info = root->fs_info; | |
3530 | struct btrfs_space_info *space_info; | |
3531 | struct extent_io_tree *block_group_cache; | |
3532 | struct btrfs_key key; | |
3533 | struct btrfs_key found_key; | |
3534 | struct extent_buffer *leaf; | |
3535 | ||
3536 | block_group_cache = &info->block_group_cache; | |
3537 | root = info->extent_root; | |
3538 | key.objectid = 0; | |
3539 | key.offset = 0; | |
3540 | btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY); | |
3541 | path = btrfs_alloc_path(); | |
3542 | if (!path) | |
3543 | return -ENOMEM; | |
3544 | ||
3545 | mutex_lock(&root->fs_info->alloc_mutex); | |
3546 | while(1) { | |
3547 | ret = find_first_block_group(root, path, &key); | |
3548 | if (ret > 0) { | |
3549 | ret = 0; | |
3550 | goto error; | |
3551 | } | |
3552 | if (ret != 0) | |
3553 | goto error; | |
3554 | ||
3555 | leaf = path->nodes[0]; | |
3556 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
3557 | cache = kzalloc(sizeof(*cache), GFP_NOFS); | |
3558 | if (!cache) { | |
3559 | ret = -ENOMEM; | |
3560 | break; | |
3561 | } | |
3562 | ||
3563 | spin_lock_init(&cache->lock); | |
3564 | read_extent_buffer(leaf, &cache->item, | |
3565 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
3566 | sizeof(cache->item)); | |
3567 | memcpy(&cache->key, &found_key, sizeof(found_key)); | |
3568 | ||
3569 | key.objectid = found_key.objectid + found_key.offset; | |
3570 | btrfs_release_path(root, path); | |
3571 | cache->flags = btrfs_block_group_flags(&cache->item); | |
3572 | bit = 0; | |
3573 | if (cache->flags & BTRFS_BLOCK_GROUP_DATA) { | |
3574 | bit = BLOCK_GROUP_DATA; | |
3575 | } else if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) { | |
3576 | bit = BLOCK_GROUP_SYSTEM; | |
3577 | } else if (cache->flags & BTRFS_BLOCK_GROUP_METADATA) { | |
3578 | bit = BLOCK_GROUP_METADATA; | |
3579 | } | |
3580 | set_avail_alloc_bits(info, cache->flags); | |
3581 | ||
3582 | ret = update_space_info(info, cache->flags, found_key.offset, | |
3583 | btrfs_block_group_used(&cache->item), | |
3584 | &space_info); | |
3585 | BUG_ON(ret); | |
3586 | cache->space_info = space_info; | |
3587 | ||
3588 | /* use EXTENT_LOCKED to prevent merging */ | |
3589 | set_extent_bits(block_group_cache, found_key.objectid, | |
3590 | found_key.objectid + found_key.offset - 1, | |
3591 | EXTENT_LOCKED, GFP_NOFS); | |
3592 | set_state_private(block_group_cache, found_key.objectid, | |
3593 | (unsigned long)cache); | |
3594 | set_extent_bits(block_group_cache, found_key.objectid, | |
3595 | found_key.objectid + found_key.offset - 1, | |
3596 | bit | EXTENT_LOCKED, GFP_NOFS); | |
3597 | if (key.objectid >= | |
3598 | btrfs_super_total_bytes(&info->super_copy)) | |
3599 | break; | |
3600 | } | |
3601 | ret = 0; | |
3602 | error: | |
3603 | btrfs_free_path(path); | |
3604 | mutex_unlock(&root->fs_info->alloc_mutex); | |
3605 | return ret; | |
3606 | } | |
3607 | ||
3608 | int btrfs_make_block_group(struct btrfs_trans_handle *trans, | |
3609 | struct btrfs_root *root, u64 bytes_used, | |
3610 | u64 type, u64 chunk_objectid, u64 chunk_offset, | |
3611 | u64 size) | |
3612 | { | |
3613 | int ret; | |
3614 | int bit = 0; | |
3615 | struct btrfs_root *extent_root; | |
3616 | struct btrfs_block_group_cache *cache; | |
3617 | struct extent_io_tree *block_group_cache; | |
3618 | ||
3619 | WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex)); | |
3620 | extent_root = root->fs_info->extent_root; | |
3621 | block_group_cache = &root->fs_info->block_group_cache; | |
3622 | ||
3623 | cache = kzalloc(sizeof(*cache), GFP_NOFS); | |
3624 | BUG_ON(!cache); | |
3625 | cache->key.objectid = chunk_offset; | |
3626 | cache->key.offset = size; | |
3627 | spin_lock_init(&cache->lock); | |
3628 | btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY); | |
3629 | ||
3630 | btrfs_set_block_group_used(&cache->item, bytes_used); | |
3631 | btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid); | |
3632 | cache->flags = type; | |
3633 | btrfs_set_block_group_flags(&cache->item, type); | |
3634 | ||
3635 | ret = update_space_info(root->fs_info, cache->flags, size, bytes_used, | |
3636 | &cache->space_info); | |
3637 | BUG_ON(ret); | |
3638 | ||
3639 | bit = block_group_state_bits(type); | |
3640 | set_extent_bits(block_group_cache, chunk_offset, | |
3641 | chunk_offset + size - 1, | |
3642 | EXTENT_LOCKED, GFP_NOFS); | |
3643 | set_state_private(block_group_cache, chunk_offset, | |
3644 | (unsigned long)cache); | |
3645 | set_extent_bits(block_group_cache, chunk_offset, | |
3646 | chunk_offset + size - 1, | |
3647 | bit | EXTENT_LOCKED, GFP_NOFS); | |
3648 | ||
3649 | ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item, | |
3650 | sizeof(cache->item)); | |
3651 | BUG_ON(ret); | |
3652 | ||
3653 | finish_current_insert(trans, extent_root); | |
3654 | ret = del_pending_extents(trans, extent_root); | |
3655 | BUG_ON(ret); | |
3656 | set_avail_alloc_bits(extent_root->fs_info, type); | |
3657 | ||
3658 | return 0; | |
3659 | } |