projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Btrfs: add snapshot/subvolume destroy ioctl
[deliverable/linux.git] / fs / btrfs / extent-tree.c
1 /*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18 #include <linux/sched.h>
19 #include <linux/pagemap.h>
20 #include <linux/writeback.h>
21 #include <linux/blkdev.h>
22 #include <linux/sort.h>
23 #include <linux/rcupdate.h>
24 #include <linux/kthread.h>
25 #include "compat.h"
26 #include "hash.h"
27 #include "ctree.h"
28 #include "disk-io.h"
29 #include "print-tree.h"
30 #include "transaction.h"
31 #include "volumes.h"
32 #include "locking.h"
33 #include "free-space-cache.h"
34
35 static int update_block_group(struct btrfs_trans_handle *trans,
36 struct btrfs_root *root,
37 u64 bytenr, u64 num_bytes, int alloc,
38 int mark_free);
39 static int update_reserved_extents(struct btrfs_block_group_cache *cache,
40 u64 num_bytes, int reserve);
41 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
42 struct btrfs_root *root,
43 u64 bytenr, u64 num_bytes, u64 parent,
44 u64 root_objectid, u64 owner_objectid,
45 u64 owner_offset, int refs_to_drop,
46 struct btrfs_delayed_extent_op *extra_op);
47 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
48 struct extent_buffer *leaf,
49 struct btrfs_extent_item *ei);
50 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
51 struct btrfs_root *root,
52 u64 parent, u64 root_objectid,
53 u64 flags, u64 owner, u64 offset,
54 struct btrfs_key *ins, int ref_mod);
55 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
56 struct btrfs_root *root,
57 u64 parent, u64 root_objectid,
58 u64 flags, struct btrfs_disk_key *key,
59 int level, struct btrfs_key *ins);
60 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
61 struct btrfs_root *extent_root, u64 alloc_bytes,
62 u64 flags, int force);
63 static int pin_down_bytes(struct btrfs_trans_handle *trans,
64 struct btrfs_root *root,
65 struct btrfs_path *path,
66 u64 bytenr, u64 num_bytes,
67 int is_data, int reserved,
68 struct extent_buffer **must_clean);
69 static int find_next_key(struct btrfs_path *path, int level,
70 struct btrfs_key *key);
71
72 static noinline int
73 block_group_cache_done(struct btrfs_block_group_cache *cache)
74 {
75 smp_mb();
76 return cache->cached == BTRFS_CACHE_FINISHED;
77 }
78
79 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
80 {
81 return (cache->flags & bits) == bits;
82 }
83
84 /*
85 * this adds the block group to the fs_info rb tree for the block group
86 * cache
87 */
88 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
89 struct btrfs_block_group_cache *block_group)
90 {
91 struct rb_node **p;
92 struct rb_node *parent = NULL;
93 struct btrfs_block_group_cache *cache;
94
95 spin_lock(&info->block_group_cache_lock);
96 p = &info->block_group_cache_tree.rb_node;
97
98 while (*p) {
99 parent = *p;
100 cache = rb_entry(parent, struct btrfs_block_group_cache,
101 cache_node);
102 if (block_group->key.objectid < cache->key.objectid) {
103 p = &(*p)->rb_left;
104 } else if (block_group->key.objectid > cache->key.objectid) {
105 p = &(*p)->rb_right;
106 } else {
107 spin_unlock(&info->block_group_cache_lock);
108 return -EEXIST;
109 }
110 }
111
112 rb_link_node(&block_group->cache_node, parent, p);
113 rb_insert_color(&block_group->cache_node,
114 &info->block_group_cache_tree);
115 spin_unlock(&info->block_group_cache_lock);
116
117 return 0;
118 }
119
120 /*
121 * This will return the block group at or after bytenr if contains is 0, else
122 * it will return the block group that contains the bytenr
123 */
124 static struct btrfs_block_group_cache *
125 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
126 int contains)
127 {
128 struct btrfs_block_group_cache *cache, *ret = NULL;
129 struct rb_node *n;
130 u64 end, start;
131
132 spin_lock(&info->block_group_cache_lock);
133 n = info->block_group_cache_tree.rb_node;
134
135 while (n) {
136 cache = rb_entry(n, struct btrfs_block_group_cache,
137 cache_node);
138 end = cache->key.objectid + cache->key.offset - 1;
139 start = cache->key.objectid;
140
141 if (bytenr < start) {
142 if (!contains && (!ret || start < ret->key.objectid))
143 ret = cache;
144 n = n->rb_left;
145 } else if (bytenr > start) {
146 if (contains && bytenr <= end) {
147 ret = cache;
148 break;
149 }
150 n = n->rb_right;
151 } else {
152 ret = cache;
153 break;
154 }
155 }
156 if (ret)
157 atomic_inc(&ret->count);
158 spin_unlock(&info->block_group_cache_lock);
159
160 return ret;
161 }
162
163 static int add_excluded_extent(struct btrfs_root *root,
164 u64 start, u64 num_bytes)
165 {
166 u64 end = start + num_bytes - 1;
167 set_extent_bits(&root->fs_info->freed_extents[0],
168 start, end, EXTENT_UPTODATE, GFP_NOFS);
169 set_extent_bits(&root->fs_info->freed_extents[1],
170 start, end, EXTENT_UPTODATE, GFP_NOFS);
171 return 0;
172 }
173
174 static void free_excluded_extents(struct btrfs_root *root,
175 struct btrfs_block_group_cache *cache)
176 {
177 u64 start, end;
178
179 start = cache->key.objectid;
180 end = start + cache->key.offset - 1;
181
182 clear_extent_bits(&root->fs_info->freed_extents[0],
183 start, end, EXTENT_UPTODATE, GFP_NOFS);
184 clear_extent_bits(&root->fs_info->freed_extents[1],
185 start, end, EXTENT_UPTODATE, GFP_NOFS);
186 }
187
188 static int exclude_super_stripes(struct btrfs_root *root,
189 struct btrfs_block_group_cache *cache)
190 {
191 u64 bytenr;
192 u64 *logical;
193 int stripe_len;
194 int i, nr, ret;
195
196 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
197 bytenr = btrfs_sb_offset(i);
198 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
199 cache->key.objectid, bytenr,
200 0, &logical, &nr, &stripe_len);
201 BUG_ON(ret);
202
203 while (nr--) {
204 ret = add_excluded_extent(root, logical[nr],
205 stripe_len);
206 BUG_ON(ret);
207 }
208
209 kfree(logical);
210 }
211 return 0;
212 }
213
214 static struct btrfs_caching_control *
215 get_caching_control(struct btrfs_block_group_cache *cache)
216 {
217 struct btrfs_caching_control *ctl;
218
219 spin_lock(&cache->lock);
220 if (cache->cached != BTRFS_CACHE_STARTED) {
221 spin_unlock(&cache->lock);
222 return NULL;
223 }
224
225 ctl = cache->caching_ctl;
226 atomic_inc(&ctl->count);
227 spin_unlock(&cache->lock);
228 return ctl;
229 }
230
231 static void put_caching_control(struct btrfs_caching_control *ctl)
232 {
233 if (atomic_dec_and_test(&ctl->count))
234 kfree(ctl);
235 }
236
237 /*
238 * this is only called by cache_block_group, since we could have freed extents
239 * we need to check the pinned_extents for any extents that can't be used yet
240 * since their free space will be released as soon as the transaction commits.
241 */
242 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
243 struct btrfs_fs_info *info, u64 start, u64 end)
244 {
245 u64 extent_start, extent_end, size, total_added = 0;
246 int ret;
247
248 while (start < end) {
249 ret = find_first_extent_bit(info->pinned_extents, start,
250 &extent_start, &extent_end,
251 EXTENT_DIRTY | EXTENT_UPTODATE);
252 if (ret)
253 break;
254
255 if (extent_start == start) {
256 start = extent_end + 1;
257 } else if (extent_start > start && extent_start < end) {
258 size = extent_start - start;
259 total_added += size;
260 ret = btrfs_add_free_space(block_group, start,
261 size);
262 BUG_ON(ret);
263 start = extent_end + 1;
264 } else {
265 break;
266 }
267 }
268
269 if (start < end) {
270 size = end - start;
271 total_added += size;
272 ret = btrfs_add_free_space(block_group, start, size);
273 BUG_ON(ret);
274 }
275
276 return total_added;
277 }
278
279 static int caching_kthread(void *data)
280 {
281 struct btrfs_block_group_cache *block_group = data;
282 struct btrfs_fs_info *fs_info = block_group->fs_info;
283 struct btrfs_caching_control *caching_ctl = block_group->caching_ctl;
284 struct btrfs_root *extent_root = fs_info->extent_root;
285 struct btrfs_path *path;
286 struct extent_buffer *leaf;
287 struct btrfs_key key;
288 u64 total_found = 0;
289 u64 last = 0;
290 u32 nritems;
291 int ret = 0;
292
293 path = btrfs_alloc_path();
294 if (!path)
295 return -ENOMEM;
296
297 exclude_super_stripes(extent_root, block_group);
298
299 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
300
301 /*
302 * We don't want to deadlock with somebody trying to allocate a new
303 * extent for the extent root while also trying to search the extent
304 * root to add free space. So we skip locking and search the commit
305 * root, since its read-only
306 */
307 path->skip_locking = 1;
308 path->search_commit_root = 1;
309 path->reada = 2;
310
311 key.objectid = last;
312 key.offset = 0;
313 key.type = BTRFS_EXTENT_ITEM_KEY;
314 again:
315 mutex_lock(&caching_ctl->mutex);
316 /* need to make sure the commit_root doesn't disappear */
317 down_read(&fs_info->extent_commit_sem);
318
319 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
320 if (ret < 0)
321 goto err;
322
323 leaf = path->nodes[0];
324 nritems = btrfs_header_nritems(leaf);
325
326 while (1) {
327 smp_mb();
328 if (fs_info->closing > 1) {
329 last = (u64)-1;
330 break;
331 }
332
333 if (path->slots[0] < nritems) {
334 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
335 } else {
336 ret = find_next_key(path, 0, &key);
337 if (ret)
338 break;
339
340 caching_ctl->progress = last;
341 btrfs_release_path(extent_root, path);
342 up_read(&fs_info->extent_commit_sem);
343 mutex_unlock(&caching_ctl->mutex);
344 if (btrfs_transaction_in_commit(fs_info))
345 schedule_timeout(1);
346 else
347 cond_resched();
348 goto again;
349 }
350
351 if (key.objectid < block_group->key.objectid) {
352 path->slots[0]++;
353 continue;
354 }
355
356 if (key.objectid >= block_group->key.objectid +
357 block_group->key.offset)
358 break;
359
360 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
361 total_found += add_new_free_space(block_group,
362 fs_info, last,
363 key.objectid);
364 last = key.objectid + key.offset;
365
366 if (total_found > (1024 * 1024 * 2)) {
367 total_found = 0;
368 wake_up(&caching_ctl->wait);
369 }
370 }
371 path->slots[0]++;
372 }
373 ret = 0;
374
375 total_found += add_new_free_space(block_group, fs_info, last,
376 block_group->key.objectid +
377 block_group->key.offset);
378 caching_ctl->progress = (u64)-1;
379
380 spin_lock(&block_group->lock);
381 block_group->caching_ctl = NULL;
382 block_group->cached = BTRFS_CACHE_FINISHED;
383 spin_unlock(&block_group->lock);
384
385 err:
386 btrfs_free_path(path);
387 up_read(&fs_info->extent_commit_sem);
388
389 free_excluded_extents(extent_root, block_group);
390
391 mutex_unlock(&caching_ctl->mutex);
392 wake_up(&caching_ctl->wait);
393
394 put_caching_control(caching_ctl);
395 atomic_dec(&block_group->space_info->caching_threads);
396 return 0;
397 }
398
399 static int cache_block_group(struct btrfs_block_group_cache *cache)
400 {
401 struct btrfs_fs_info *fs_info = cache->fs_info;
402 struct btrfs_caching_control *caching_ctl;
403 struct task_struct *tsk;
404 int ret = 0;
405
406 smp_mb();
407 if (cache->cached != BTRFS_CACHE_NO)
408 return 0;
409
410 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_KERNEL);
411 BUG_ON(!caching_ctl);
412
413 INIT_LIST_HEAD(&caching_ctl->list);
414 mutex_init(&caching_ctl->mutex);
415 init_waitqueue_head(&caching_ctl->wait);
416 caching_ctl->block_group = cache;
417 caching_ctl->progress = cache->key.objectid;
418 /* one for caching kthread, one for caching block group list */
419 atomic_set(&caching_ctl->count, 2);
420
421 spin_lock(&cache->lock);
422 if (cache->cached != BTRFS_CACHE_NO) {
423 spin_unlock(&cache->lock);
424 kfree(caching_ctl);
425 return 0;
426 }
427 cache->caching_ctl = caching_ctl;
428 cache->cached = BTRFS_CACHE_STARTED;
429 spin_unlock(&cache->lock);
430
431 down_write(&fs_info->extent_commit_sem);
432 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
433 up_write(&fs_info->extent_commit_sem);
434
435 atomic_inc(&cache->space_info->caching_threads);
436
437 tsk = kthread_run(caching_kthread, cache, "btrfs-cache-%llu\n",
438 cache->key.objectid);
439 if (IS_ERR(tsk)) {
440 ret = PTR_ERR(tsk);
441 printk(KERN_ERR "error running thread %d\n", ret);
442 BUG();
443 }
444
445 return ret;
446 }
447
448 /*
449 * return the block group that starts at or after bytenr
450 */
451 static struct btrfs_block_group_cache *
452 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
453 {
454 struct btrfs_block_group_cache *cache;
455
456 cache = block_group_cache_tree_search(info, bytenr, 0);
457
458 return cache;
459 }
460
461 /*
462 * return the block group that contains the given bytenr
463 */
464 struct btrfs_block_group_cache *btrfs_lookup_block_group(
465 struct btrfs_fs_info *info,
466 u64 bytenr)
467 {
468 struct btrfs_block_group_cache *cache;
469
470 cache = block_group_cache_tree_search(info, bytenr, 1);
471
472 return cache;
473 }
474
475 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
476 {
477 if (atomic_dec_and_test(&cache->count))
478 kfree(cache);
479 }
480
481 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
482 u64 flags)
483 {
484 struct list_head *head = &info->space_info;
485 struct btrfs_space_info *found;
486
487 rcu_read_lock();
488 list_for_each_entry_rcu(found, head, list) {
489 if (found->flags == flags) {
490 rcu_read_unlock();
491 return found;
492 }
493 }
494 rcu_read_unlock();
495 return NULL;
496 }
497
498 /*
499 * after adding space to the filesystem, we need to clear the full flags
500 * on all the space infos.
501 */
502 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
503 {
504 struct list_head *head = &info->space_info;
505 struct btrfs_space_info *found;
506
507 rcu_read_lock();
508 list_for_each_entry_rcu(found, head, list)
509 found->full = 0;
510 rcu_read_unlock();
511 }
512
513 static u64 div_factor(u64 num, int factor)
514 {
515 if (factor == 10)
516 return num;
517 num *= factor;
518 do_div(num, 10);
519 return num;
520 }
521
522 u64 btrfs_find_block_group(struct btrfs_root *root,
523 u64 search_start, u64 search_hint, int owner)
524 {
525 struct btrfs_block_group_cache *cache;
526 u64 used;
527 u64 last = max(search_hint, search_start);
528 u64 group_start = 0;
529 int full_search = 0;
530 int factor = 9;
531 int wrapped = 0;
532 again:
533 while (1) {
534 cache = btrfs_lookup_first_block_group(root->fs_info, last);
535 if (!cache)
536 break;
537
538 spin_lock(&cache->lock);
539 last = cache->key.objectid + cache->key.offset;
540 used = btrfs_block_group_used(&cache->item);
541
542 if ((full_search || !cache->ro) &&
543 block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
544 if (used + cache->pinned + cache->reserved <
545 div_factor(cache->key.offset, factor)) {
546 group_start = cache->key.objectid;
547 spin_unlock(&cache->lock);
548 btrfs_put_block_group(cache);
549 goto found;
550 }
551 }
552 spin_unlock(&cache->lock);
553 btrfs_put_block_group(cache);
554 cond_resched();
555 }
556 if (!wrapped) {
557 last = search_start;
558 wrapped = 1;
559 goto again;
560 }
561 if (!full_search && factor < 10) {
562 last = search_start;
563 full_search = 1;
564 factor = 10;
565 goto again;
566 }
567 found:
568 return group_start;
569 }
570
571 /* simple helper to search for an existing extent at a given offset */
572 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
573 {
574 int ret;
575 struct btrfs_key key;
576 struct btrfs_path *path;
577
578 path = btrfs_alloc_path();
579 BUG_ON(!path);
580 key.objectid = start;
581 key.offset = len;
582 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
583 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
584 0, 0);
585 btrfs_free_path(path);
586 return ret;
587 }
588
589 /*
590 * Back reference rules. Back refs have three main goals:
591 *
592 * 1) differentiate between all holders of references to an extent so that
593 * when a reference is dropped we can make sure it was a valid reference
594 * before freeing the extent.
595 *
596 * 2) Provide enough information to quickly find the holders of an extent
597 * if we notice a given block is corrupted or bad.
598 *
599 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
600 * maintenance. This is actually the same as #2, but with a slightly
601 * different use case.
602 *
603 * There are two kinds of back refs. The implicit back refs is optimized
604 * for pointers in non-shared tree blocks. For a given pointer in a block,
605 * back refs of this kind provide information about the block's owner tree
606 * and the pointer's key. These information allow us to find the block by
607 * b-tree searching. The full back refs is for pointers in tree blocks not
608 * referenced by their owner trees. The location of tree block is recorded
609 * in the back refs. Actually the full back refs is generic, and can be
610 * used in all cases the implicit back refs is used. The major shortcoming
611 * of the full back refs is its overhead. Every time a tree block gets
612 * COWed, we have to update back refs entry for all pointers in it.
613 *
614 * For a newly allocated tree block, we use implicit back refs for
615 * pointers in it. This means most tree related operations only involve
616 * implicit back refs. For a tree block created in old transaction, the
617 * only way to drop a reference to it is COW it. So we can detect the
618 * event that tree block loses its owner tree's reference and do the
619 * back refs conversion.
620 *
621 * When a tree block is COW'd through a tree, there are four cases:
622 *
623 * The reference count of the block is one and the tree is the block's
624 * owner tree. Nothing to do in this case.
625 *
626 * The reference count of the block is one and the tree is not the
627 * block's owner tree. In this case, full back refs is used for pointers
628 * in the block. Remove these full back refs, add implicit back refs for
629 * every pointers in the new block.
630 *
631 * The reference count of the block is greater than one and the tree is
632 * the block's owner tree. In this case, implicit back refs is used for
633 * pointers in the block. Add full back refs for every pointers in the
634 * block, increase lower level extents' reference counts. The original
635 * implicit back refs are entailed to the new block.
636 *
637 * The reference count of the block is greater than one and the tree is
638 * not the block's owner tree. Add implicit back refs for every pointer in
639 * the new block, increase lower level extents' reference count.
640 *
641 * Back Reference Key composing:
642 *
643 * The key objectid corresponds to the first byte in the extent,
644 * The key type is used to differentiate between types of back refs.
645 * There are different meanings of the key offset for different types
646 * of back refs.
647 *
648 * File extents can be referenced by:
649 *
650 * - multiple snapshots, subvolumes, or different generations in one subvol
651 * - different files inside a single subvolume
652 * - different offsets inside a file (bookend extents in file.c)
653 *
654 * The extent ref structure for the implicit back refs has fields for:
655 *
656 * - Objectid of the subvolume root
657 * - objectid of the file holding the reference
658 * - original offset in the file
659 * - how many bookend extents
660 *
661 * The key offset for the implicit back refs is hash of the first
662 * three fields.
663 *
664 * The extent ref structure for the full back refs has field for:
665 *
666 * - number of pointers in the tree leaf
667 *
668 * The key offset for the implicit back refs is the first byte of
669 * the tree leaf
670 *
671 * When a file extent is allocated, The implicit back refs is used.
672 * the fields are filled in:
673 *
674 * (root_key.objectid, inode objectid, offset in file, 1)
675 *
676 * When a file extent is removed file truncation, we find the
677 * corresponding implicit back refs and check the following fields:
678 *
679 * (btrfs_header_owner(leaf), inode objectid, offset in file)
680 *
681 * Btree extents can be referenced by:
682 *
683 * - Different subvolumes
684 *
685 * Both the implicit back refs and the full back refs for tree blocks
686 * only consist of key. The key offset for the implicit back refs is
687 * objectid of block's owner tree. The key offset for the full back refs
688 * is the first byte of parent block.
689 *
690 * When implicit back refs is used, information about the lowest key and
691 * level of the tree block are required. These information are stored in
692 * tree block info structure.
693 */
694
695 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
696 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
697 struct btrfs_root *root,
698 struct btrfs_path *path,
699 u64 owner, u32 extra_size)
700 {
701 struct btrfs_extent_item *item;
702 struct btrfs_extent_item_v0 *ei0;
703 struct btrfs_extent_ref_v0 *ref0;
704 struct btrfs_tree_block_info *bi;
705 struct extent_buffer *leaf;
706 struct btrfs_key key;
707 struct btrfs_key found_key;
708 u32 new_size = sizeof(*item);
709 u64 refs;
710 int ret;
711
712 leaf = path->nodes[0];
713 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
714
715 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
716 ei0 = btrfs_item_ptr(leaf, path->slots[0],
717 struct btrfs_extent_item_v0);
718 refs = btrfs_extent_refs_v0(leaf, ei0);
719
720 if (owner == (u64)-1) {
721 while (1) {
722 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
723 ret = btrfs_next_leaf(root, path);
724 if (ret < 0)
725 return ret;
726 BUG_ON(ret > 0);
727 leaf = path->nodes[0];
728 }
729 btrfs_item_key_to_cpu(leaf, &found_key,
730 path->slots[0]);
731 BUG_ON(key.objectid != found_key.objectid);
732 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
733 path->slots[0]++;
734 continue;
735 }
736 ref0 = btrfs_item_ptr(leaf, path->slots[0],
737 struct btrfs_extent_ref_v0);
738 owner = btrfs_ref_objectid_v0(leaf, ref0);
739 break;
740 }
741 }
742 btrfs_release_path(root, path);
743
744 if (owner < BTRFS_FIRST_FREE_OBJECTID)
745 new_size += sizeof(*bi);
746
747 new_size -= sizeof(*ei0);
748 ret = btrfs_search_slot(trans, root, &key, path,
749 new_size + extra_size, 1);
750 if (ret < 0)
751 return ret;
752 BUG_ON(ret);
753
754 ret = btrfs_extend_item(trans, root, path, new_size);
755 BUG_ON(ret);
756
757 leaf = path->nodes[0];
758 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
759 btrfs_set_extent_refs(leaf, item, refs);
760 /* FIXME: get real generation */
761 btrfs_set_extent_generation(leaf, item, 0);
762 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
763 btrfs_set_extent_flags(leaf, item,
764 BTRFS_EXTENT_FLAG_TREE_BLOCK |
765 BTRFS_BLOCK_FLAG_FULL_BACKREF);
766 bi = (struct btrfs_tree_block_info *)(item + 1);
767 /* FIXME: get first key of the block */
768 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
769 btrfs_set_tree_block_level(leaf, bi, (int)owner);
770 } else {
771 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
772 }
773 btrfs_mark_buffer_dirty(leaf);
774 return 0;
775 }
776 #endif
777
778 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
779 {
780 u32 high_crc = ~(u32)0;
781 u32 low_crc = ~(u32)0;
782 __le64 lenum;
783
784 lenum = cpu_to_le64(root_objectid);
785 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
786 lenum = cpu_to_le64(owner);
787 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
788 lenum = cpu_to_le64(offset);
789 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
790
791 return ((u64)high_crc << 31) ^ (u64)low_crc;
792 }
793
794 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
795 struct btrfs_extent_data_ref *ref)
796 {
797 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
798 btrfs_extent_data_ref_objectid(leaf, ref),
799 btrfs_extent_data_ref_offset(leaf, ref));
800 }
801
802 static int match_extent_data_ref(struct extent_buffer *leaf,
803 struct btrfs_extent_data_ref *ref,
804 u64 root_objectid, u64 owner, u64 offset)
805 {
806 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
807 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
808 btrfs_extent_data_ref_offset(leaf, ref) != offset)
809 return 0;
810 return 1;
811 }
812
813 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
814 struct btrfs_root *root,
815 struct btrfs_path *path,
816 u64 bytenr, u64 parent,
817 u64 root_objectid,
818 u64 owner, u64 offset)
819 {
820 struct btrfs_key key;
821 struct btrfs_extent_data_ref *ref;
822 struct extent_buffer *leaf;
823 u32 nritems;
824 int ret;
825 int recow;
826 int err = -ENOENT;
827
828 key.objectid = bytenr;
829 if (parent) {
830 key.type = BTRFS_SHARED_DATA_REF_KEY;
831 key.offset = parent;
832 } else {
833 key.type = BTRFS_EXTENT_DATA_REF_KEY;
834 key.offset = hash_extent_data_ref(root_objectid,
835 owner, offset);
836 }
837 again:
838 recow = 0;
839 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
840 if (ret < 0) {
841 err = ret;
842 goto fail;
843 }
844
845 if (parent) {
846 if (!ret)
847 return 0;
848 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
849 key.type = BTRFS_EXTENT_REF_V0_KEY;
850 btrfs_release_path(root, path);
851 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
852 if (ret < 0) {
853 err = ret;
854 goto fail;
855 }
856 if (!ret)
857 return 0;
858 #endif
859 goto fail;
860 }
861
862 leaf = path->nodes[0];
863 nritems = btrfs_header_nritems(leaf);
864 while (1) {
865 if (path->slots[0] >= nritems) {
866 ret = btrfs_next_leaf(root, path);
867 if (ret < 0)
868 err = ret;
869 if (ret)
870 goto fail;
871
872 leaf = path->nodes[0];
873 nritems = btrfs_header_nritems(leaf);
874 recow = 1;
875 }
876
877 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
878 if (key.objectid != bytenr ||
879 key.type != BTRFS_EXTENT_DATA_REF_KEY)
880 goto fail;
881
882 ref = btrfs_item_ptr(leaf, path->slots[0],
883 struct btrfs_extent_data_ref);
884
885 if (match_extent_data_ref(leaf, ref, root_objectid,
886 owner, offset)) {
887 if (recow) {
888 btrfs_release_path(root, path);
889 goto again;
890 }
891 err = 0;
892 break;
893 }
894 path->slots[0]++;
895 }
896 fail:
897 return err;
898 }
899
900 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
901 struct btrfs_root *root,
902 struct btrfs_path *path,
903 u64 bytenr, u64 parent,
904 u64 root_objectid, u64 owner,
905 u64 offset, int refs_to_add)
906 {
907 struct btrfs_key key;
908 struct extent_buffer *leaf;
909 u32 size;
910 u32 num_refs;
911 int ret;
912
913 key.objectid = bytenr;
914 if (parent) {
915 key.type = BTRFS_SHARED_DATA_REF_KEY;
916 key.offset = parent;
917 size = sizeof(struct btrfs_shared_data_ref);
918 } else {
919 key.type = BTRFS_EXTENT_DATA_REF_KEY;
920 key.offset = hash_extent_data_ref(root_objectid,
921 owner, offset);
922 size = sizeof(struct btrfs_extent_data_ref);
923 }
924
925 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
926 if (ret && ret != -EEXIST)
927 goto fail;
928
929 leaf = path->nodes[0];
930 if (parent) {
931 struct btrfs_shared_data_ref *ref;
932 ref = btrfs_item_ptr(leaf, path->slots[0],
933 struct btrfs_shared_data_ref);
934 if (ret == 0) {
935 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
936 } else {
937 num_refs = btrfs_shared_data_ref_count(leaf, ref);
938 num_refs += refs_to_add;
939 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
940 }
941 } else {
942 struct btrfs_extent_data_ref *ref;
943 while (ret == -EEXIST) {
944 ref = btrfs_item_ptr(leaf, path->slots[0],
945 struct btrfs_extent_data_ref);
946 if (match_extent_data_ref(leaf, ref, root_objectid,
947 owner, offset))
948 break;
949 btrfs_release_path(root, path);
950 key.offset++;
951 ret = btrfs_insert_empty_item(trans, root, path, &key,
952 size);
953 if (ret && ret != -EEXIST)
954 goto fail;
955
956 leaf = path->nodes[0];
957 }
958 ref = btrfs_item_ptr(leaf, path->slots[0],
959 struct btrfs_extent_data_ref);
960 if (ret == 0) {
961 btrfs_set_extent_data_ref_root(leaf, ref,
962 root_objectid);
963 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
964 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
965 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
966 } else {
967 num_refs = btrfs_extent_data_ref_count(leaf, ref);
968 num_refs += refs_to_add;
969 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
970 }
971 }
972 btrfs_mark_buffer_dirty(leaf);
973 ret = 0;
974 fail:
975 btrfs_release_path(root, path);
976 return ret;
977 }
978
979 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
980 struct btrfs_root *root,
981 struct btrfs_path *path,
982 int refs_to_drop)
983 {
984 struct btrfs_key key;
985 struct btrfs_extent_data_ref *ref1 = NULL;
986 struct btrfs_shared_data_ref *ref2 = NULL;
987 struct extent_buffer *leaf;
988 u32 num_refs = 0;
989 int ret = 0;
990
991 leaf = path->nodes[0];
992 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
993
994 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
995 ref1 = btrfs_item_ptr(leaf, path->slots[0],
996 struct btrfs_extent_data_ref);
997 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
998 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
999 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1000 struct btrfs_shared_data_ref);
1001 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1002 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1003 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1004 struct btrfs_extent_ref_v0 *ref0;
1005 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1006 struct btrfs_extent_ref_v0);
1007 num_refs = btrfs_ref_count_v0(leaf, ref0);
1008 #endif
1009 } else {
1010 BUG();
1011 }
1012
1013 BUG_ON(num_refs < refs_to_drop);
1014 num_refs -= refs_to_drop;
1015
1016 if (num_refs == 0) {
1017 ret = btrfs_del_item(trans, root, path);
1018 } else {
1019 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1020 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1021 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1022 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1023 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1024 else {
1025 struct btrfs_extent_ref_v0 *ref0;
1026 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1027 struct btrfs_extent_ref_v0);
1028 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1029 }
1030 #endif
1031 btrfs_mark_buffer_dirty(leaf);
1032 }
1033 return ret;
1034 }
1035
1036 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1037 struct btrfs_path *path,
1038 struct btrfs_extent_inline_ref *iref)
1039 {
1040 struct btrfs_key key;
1041 struct extent_buffer *leaf;
1042 struct btrfs_extent_data_ref *ref1;
1043 struct btrfs_shared_data_ref *ref2;
1044 u32 num_refs = 0;
1045
1046 leaf = path->nodes[0];
1047 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1048 if (iref) {
1049 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1050 BTRFS_EXTENT_DATA_REF_KEY) {
1051 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1052 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1053 } else {
1054 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1055 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1056 }
1057 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1058 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1059 struct btrfs_extent_data_ref);
1060 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1061 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1062 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1063 struct btrfs_shared_data_ref);
1064 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1065 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1066 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1067 struct btrfs_extent_ref_v0 *ref0;
1068 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1069 struct btrfs_extent_ref_v0);
1070 num_refs = btrfs_ref_count_v0(leaf, ref0);
1071 #endif
1072 } else {
1073 WARN_ON(1);
1074 }
1075 return num_refs;
1076 }
1077
1078 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1079 struct btrfs_root *root,
1080 struct btrfs_path *path,
1081 u64 bytenr, u64 parent,
1082 u64 root_objectid)
1083 {
1084 struct btrfs_key key;
1085 int ret;
1086
1087 key.objectid = bytenr;
1088 if (parent) {
1089 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1090 key.offset = parent;
1091 } else {
1092 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1093 key.offset = root_objectid;
1094 }
1095
1096 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1097 if (ret > 0)
1098 ret = -ENOENT;
1099 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1100 if (ret == -ENOENT && parent) {
1101 btrfs_release_path(root, path);
1102 key.type = BTRFS_EXTENT_REF_V0_KEY;
1103 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1104 if (ret > 0)
1105 ret = -ENOENT;
1106 }
1107 #endif
1108 return ret;
1109 }
1110
1111 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1112 struct btrfs_root *root,
1113 struct btrfs_path *path,
1114 u64 bytenr, u64 parent,
1115 u64 root_objectid)
1116 {
1117 struct btrfs_key key;
1118 int ret;
1119
1120 key.objectid = bytenr;
1121 if (parent) {
1122 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1123 key.offset = parent;
1124 } else {
1125 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1126 key.offset = root_objectid;
1127 }
1128
1129 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1130 btrfs_release_path(root, path);
1131 return ret;
1132 }
1133
1134 static inline int extent_ref_type(u64 parent, u64 owner)
1135 {
1136 int type;
1137 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1138 if (parent > 0)
1139 type = BTRFS_SHARED_BLOCK_REF_KEY;
1140 else
1141 type = BTRFS_TREE_BLOCK_REF_KEY;
1142 } else {
1143 if (parent > 0)
1144 type = BTRFS_SHARED_DATA_REF_KEY;
1145 else
1146 type = BTRFS_EXTENT_DATA_REF_KEY;
1147 }
1148 return type;
1149 }
1150
1151 static int find_next_key(struct btrfs_path *path, int level,
1152 struct btrfs_key *key)
1153
1154 {
1155 for (; level < BTRFS_MAX_LEVEL; level++) {
1156 if (!path->nodes[level])
1157 break;
1158 if (path->slots[level] + 1 >=
1159 btrfs_header_nritems(path->nodes[level]))
1160 continue;
1161 if (level == 0)
1162 btrfs_item_key_to_cpu(path->nodes[level], key,
1163 path->slots[level] + 1);
1164 else
1165 btrfs_node_key_to_cpu(path->nodes[level], key,
1166 path->slots[level] + 1);
1167 return 0;
1168 }
1169 return 1;
1170 }
1171
1172 /*
1173 * look for inline back ref. if back ref is found, *ref_ret is set
1174 * to the address of inline back ref, and 0 is returned.
1175 *
1176 * if back ref isn't found, *ref_ret is set to the address where it
1177 * should be inserted, and -ENOENT is returned.
1178 *
1179 * if insert is true and there are too many inline back refs, the path
1180 * points to the extent item, and -EAGAIN is returned.
1181 *
1182 * NOTE: inline back refs are ordered in the same way that back ref
1183 * items in the tree are ordered.
1184 */
1185 static noinline_for_stack
1186 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1187 struct btrfs_root *root,
1188 struct btrfs_path *path,
1189 struct btrfs_extent_inline_ref **ref_ret,
1190 u64 bytenr, u64 num_bytes,
1191 u64 parent, u64 root_objectid,
1192 u64 owner, u64 offset, int insert)
1193 {
1194 struct btrfs_key key;
1195 struct extent_buffer *leaf;
1196 struct btrfs_extent_item *ei;
1197 struct btrfs_extent_inline_ref *iref;
1198 u64 flags;
1199 u64 item_size;
1200 unsigned long ptr;
1201 unsigned long end;
1202 int extra_size;
1203 int type;
1204 int want;
1205 int ret;
1206 int err = 0;
1207
1208 key.objectid = bytenr;
1209 key.type = BTRFS_EXTENT_ITEM_KEY;
1210 key.offset = num_bytes;
1211
1212 want = extent_ref_type(parent, owner);
1213 if (insert) {
1214 extra_size = btrfs_extent_inline_ref_size(want);
1215 path->keep_locks = 1;
1216 } else
1217 extra_size = -1;
1218 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1219 if (ret < 0) {
1220 err = ret;
1221 goto out;
1222 }
1223 BUG_ON(ret);
1224
1225 leaf = path->nodes[0];
1226 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1227 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1228 if (item_size < sizeof(*ei)) {
1229 if (!insert) {
1230 err = -ENOENT;
1231 goto out;
1232 }
1233 ret = convert_extent_item_v0(trans, root, path, owner,
1234 extra_size);
1235 if (ret < 0) {
1236 err = ret;
1237 goto out;
1238 }
1239 leaf = path->nodes[0];
1240 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1241 }
1242 #endif
1243 BUG_ON(item_size < sizeof(*ei));
1244
1245 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1246 flags = btrfs_extent_flags(leaf, ei);
1247
1248 ptr = (unsigned long)(ei + 1);
1249 end = (unsigned long)ei + item_size;
1250
1251 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1252 ptr += sizeof(struct btrfs_tree_block_info);
1253 BUG_ON(ptr > end);
1254 } else {
1255 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1256 }
1257
1258 err = -ENOENT;
1259 while (1) {
1260 if (ptr >= end) {
1261 WARN_ON(ptr > end);
1262 break;
1263 }
1264 iref = (struct btrfs_extent_inline_ref *)ptr;
1265 type = btrfs_extent_inline_ref_type(leaf, iref);
1266 if (want < type)
1267 break;
1268 if (want > type) {
1269 ptr += btrfs_extent_inline_ref_size(type);
1270 continue;
1271 }
1272
1273 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1274 struct btrfs_extent_data_ref *dref;
1275 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1276 if (match_extent_data_ref(leaf, dref, root_objectid,
1277 owner, offset)) {
1278 err = 0;
1279 break;
1280 }
1281 if (hash_extent_data_ref_item(leaf, dref) <
1282 hash_extent_data_ref(root_objectid, owner, offset))
1283 break;
1284 } else {
1285 u64 ref_offset;
1286 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1287 if (parent > 0) {
1288 if (parent == ref_offset) {
1289 err = 0;
1290 break;
1291 }
1292 if (ref_offset < parent)
1293 break;
1294 } else {
1295 if (root_objectid == ref_offset) {
1296 err = 0;
1297 break;
1298 }
1299 if (ref_offset < root_objectid)
1300 break;
1301 }
1302 }
1303 ptr += btrfs_extent_inline_ref_size(type);
1304 }
1305 if (err == -ENOENT && insert) {
1306 if (item_size + extra_size >=
1307 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1308 err = -EAGAIN;
1309 goto out;
1310 }
1311 /*
1312 * To add new inline back ref, we have to make sure
1313 * there is no corresponding back ref item.
1314 * For simplicity, we just do not add new inline back
1315 * ref if there is any kind of item for this block
1316 */
1317 if (find_next_key(path, 0, &key) == 0 &&
1318 key.objectid == bytenr &&
1319 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1320 err = -EAGAIN;
1321 goto out;
1322 }
1323 }
1324 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1325 out:
1326 if (insert) {
1327 path->keep_locks = 0;
1328 btrfs_unlock_up_safe(path, 1);
1329 }
1330 return err;
1331 }
1332
1333 /*
1334 * helper to add new inline back ref
1335 */
1336 static noinline_for_stack
1337 int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1338 struct btrfs_root *root,
1339 struct btrfs_path *path,
1340 struct btrfs_extent_inline_ref *iref,
1341 u64 parent, u64 root_objectid,
1342 u64 owner, u64 offset, int refs_to_add,
1343 struct btrfs_delayed_extent_op *extent_op)
1344 {
1345 struct extent_buffer *leaf;
1346 struct btrfs_extent_item *ei;
1347 unsigned long ptr;
1348 unsigned long end;
1349 unsigned long item_offset;
1350 u64 refs;
1351 int size;
1352 int type;
1353 int ret;
1354
1355 leaf = path->nodes[0];
1356 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1357 item_offset = (unsigned long)iref - (unsigned long)ei;
1358
1359 type = extent_ref_type(parent, owner);
1360 size = btrfs_extent_inline_ref_size(type);
1361
1362 ret = btrfs_extend_item(trans, root, path, size);
1363 BUG_ON(ret);
1364
1365 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1366 refs = btrfs_extent_refs(leaf, ei);
1367 refs += refs_to_add;
1368 btrfs_set_extent_refs(leaf, ei, refs);
1369 if (extent_op)
1370 __run_delayed_extent_op(extent_op, leaf, ei);
1371
1372 ptr = (unsigned long)ei + item_offset;
1373 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1374 if (ptr < end - size)
1375 memmove_extent_buffer(leaf, ptr + size, ptr,
1376 end - size - ptr);
1377
1378 iref = (struct btrfs_extent_inline_ref *)ptr;
1379 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1380 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1381 struct btrfs_extent_data_ref *dref;
1382 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1383 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1384 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1385 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1386 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1387 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1388 struct btrfs_shared_data_ref *sref;
1389 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1390 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1391 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1392 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1393 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1394 } else {
1395 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1396 }
1397 btrfs_mark_buffer_dirty(leaf);
1398 return 0;
1399 }
1400
1401 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1402 struct btrfs_root *root,
1403 struct btrfs_path *path,
1404 struct btrfs_extent_inline_ref **ref_ret,
1405 u64 bytenr, u64 num_bytes, u64 parent,
1406 u64 root_objectid, u64 owner, u64 offset)
1407 {
1408 int ret;
1409
1410 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1411 bytenr, num_bytes, parent,
1412 root_objectid, owner, offset, 0);
1413 if (ret != -ENOENT)
1414 return ret;
1415
1416 btrfs_release_path(root, path);
1417 *ref_ret = NULL;
1418
1419 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1420 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1421 root_objectid);
1422 } else {
1423 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1424 root_objectid, owner, offset);
1425 }
1426 return ret;
1427 }
1428
1429 /*
1430 * helper to update/remove inline back ref
1431 */
1432 static noinline_for_stack
1433 int update_inline_extent_backref(struct btrfs_trans_handle *trans,
1434 struct btrfs_root *root,
1435 struct btrfs_path *path,
1436 struct btrfs_extent_inline_ref *iref,
1437 int refs_to_mod,
1438 struct btrfs_delayed_extent_op *extent_op)
1439 {
1440 struct extent_buffer *leaf;
1441 struct btrfs_extent_item *ei;
1442 struct btrfs_extent_data_ref *dref = NULL;
1443 struct btrfs_shared_data_ref *sref = NULL;
1444 unsigned long ptr;
1445 unsigned long end;
1446 u32 item_size;
1447 int size;
1448 int type;
1449 int ret;
1450 u64 refs;
1451
1452 leaf = path->nodes[0];
1453 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1454 refs = btrfs_extent_refs(leaf, ei);
1455 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1456 refs += refs_to_mod;
1457 btrfs_set_extent_refs(leaf, ei, refs);
1458 if (extent_op)
1459 __run_delayed_extent_op(extent_op, leaf, ei);
1460
1461 type = btrfs_extent_inline_ref_type(leaf, iref);
1462
1463 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1464 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1465 refs = btrfs_extent_data_ref_count(leaf, dref);
1466 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1467 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1468 refs = btrfs_shared_data_ref_count(leaf, sref);
1469 } else {
1470 refs = 1;
1471 BUG_ON(refs_to_mod != -1);
1472 }
1473
1474 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1475 refs += refs_to_mod;
1476
1477 if (refs > 0) {
1478 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1479 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1480 else
1481 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1482 } else {
1483 size = btrfs_extent_inline_ref_size(type);
1484 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1485 ptr = (unsigned long)iref;
1486 end = (unsigned long)ei + item_size;
1487 if (ptr + size < end)
1488 memmove_extent_buffer(leaf, ptr, ptr + size,
1489 end - ptr - size);
1490 item_size -= size;
1491 ret = btrfs_truncate_item(trans, root, path, item_size, 1);
1492 BUG_ON(ret);
1493 }
1494 btrfs_mark_buffer_dirty(leaf);
1495 return 0;
1496 }
1497
1498 static noinline_for_stack
1499 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1500 struct btrfs_root *root,
1501 struct btrfs_path *path,
1502 u64 bytenr, u64 num_bytes, u64 parent,
1503 u64 root_objectid, u64 owner,
1504 u64 offset, int refs_to_add,
1505 struct btrfs_delayed_extent_op *extent_op)
1506 {
1507 struct btrfs_extent_inline_ref *iref;
1508 int ret;
1509
1510 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1511 bytenr, num_bytes, parent,
1512 root_objectid, owner, offset, 1);
1513 if (ret == 0) {
1514 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1515 ret = update_inline_extent_backref(trans, root, path, iref,
1516 refs_to_add, extent_op);
1517 } else if (ret == -ENOENT) {
1518 ret = setup_inline_extent_backref(trans, root, path, iref,
1519 parent, root_objectid,
1520 owner, offset, refs_to_add,
1521 extent_op);
1522 }
1523 return ret;
1524 }
1525
1526 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1527 struct btrfs_root *root,
1528 struct btrfs_path *path,
1529 u64 bytenr, u64 parent, u64 root_objectid,
1530 u64 owner, u64 offset, int refs_to_add)
1531 {
1532 int ret;
1533 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1534 BUG_ON(refs_to_add != 1);
1535 ret = insert_tree_block_ref(trans, root, path, bytenr,
1536 parent, root_objectid);
1537 } else {
1538 ret = insert_extent_data_ref(trans, root, path, bytenr,
1539 parent, root_objectid,
1540 owner, offset, refs_to_add);
1541 }
1542 return ret;
1543 }
1544
1545 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1546 struct btrfs_root *root,
1547 struct btrfs_path *path,
1548 struct btrfs_extent_inline_ref *iref,
1549 int refs_to_drop, int is_data)
1550 {
1551 int ret;
1552
1553 BUG_ON(!is_data && refs_to_drop != 1);
1554 if (iref) {
1555 ret = update_inline_extent_backref(trans, root, path, iref,
1556 -refs_to_drop, NULL);
1557 } else if (is_data) {
1558 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1559 } else {
1560 ret = btrfs_del_item(trans, root, path);
1561 }
1562 return ret;
1563 }
1564
1565 #ifdef BIO_RW_DISCARD
1566 static void btrfs_issue_discard(struct block_device *bdev,
1567 u64 start, u64 len)
1568 {
1569 blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL);
1570 }
1571 #endif
1572
1573 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1574 u64 num_bytes)
1575 {
1576 #ifdef BIO_RW_DISCARD
1577 int ret;
1578 u64 map_length = num_bytes;
1579 struct btrfs_multi_bio *multi = NULL;
1580
1581 /* Tell the block device(s) that the sectors can be discarded */
1582 ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
1583 bytenr, &map_length, &multi, 0);
1584 if (!ret) {
1585 struct btrfs_bio_stripe *stripe = multi->stripes;
1586 int i;
1587
1588 if (map_length > num_bytes)
1589 map_length = num_bytes;
1590
1591 for (i = 0; i < multi->num_stripes; i++, stripe++) {
1592 btrfs_issue_discard(stripe->dev->bdev,
1593 stripe->physical,
1594 map_length);
1595 }
1596 kfree(multi);
1597 }
1598
1599 return ret;
1600 #else
1601 return 0;
1602 #endif
1603 }
1604
1605 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1606 struct btrfs_root *root,
1607 u64 bytenr, u64 num_bytes, u64 parent,
1608 u64 root_objectid, u64 owner, u64 offset)
1609 {
1610 int ret;
1611 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1612 root_objectid == BTRFS_TREE_LOG_OBJECTID);
1613
1614 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1615 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
1616 parent, root_objectid, (int)owner,
1617 BTRFS_ADD_DELAYED_REF, NULL);
1618 } else {
1619 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
1620 parent, root_objectid, owner, offset,
1621 BTRFS_ADD_DELAYED_REF, NULL);
1622 }
1623 return ret;
1624 }
1625
1626 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1627 struct btrfs_root *root,
1628 u64 bytenr, u64 num_bytes,
1629 u64 parent, u64 root_objectid,
1630 u64 owner, u64 offset, int refs_to_add,
1631 struct btrfs_delayed_extent_op *extent_op)
1632 {
1633 struct btrfs_path *path;
1634 struct extent_buffer *leaf;
1635 struct btrfs_extent_item *item;
1636 u64 refs;
1637 int ret;
1638 int err = 0;
1639
1640 path = btrfs_alloc_path();
1641 if (!path)
1642 return -ENOMEM;
1643
1644 path->reada = 1;
1645 path->leave_spinning = 1;
1646 /* this will setup the path even if it fails to insert the back ref */
1647 ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1648 path, bytenr, num_bytes, parent,
1649 root_objectid, owner, offset,
1650 refs_to_add, extent_op);
1651 if (ret == 0)
1652 goto out;
1653
1654 if (ret != -EAGAIN) {
1655 err = ret;
1656 goto out;
1657 }
1658
1659 leaf = path->nodes[0];
1660 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1661 refs = btrfs_extent_refs(leaf, item);
1662 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1663 if (extent_op)
1664 __run_delayed_extent_op(extent_op, leaf, item);
1665
1666 btrfs_mark_buffer_dirty(leaf);
1667 btrfs_release_path(root->fs_info->extent_root, path);
1668
1669 path->reada = 1;
1670 path->leave_spinning = 1;
1671
1672 /* now insert the actual backref */
1673 ret = insert_extent_backref(trans, root->fs_info->extent_root,
1674 path, bytenr, parent, root_objectid,
1675 owner, offset, refs_to_add);
1676 BUG_ON(ret);
1677 out:
1678 btrfs_free_path(path);
1679 return err;
1680 }
1681
1682 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1683 struct btrfs_root *root,
1684 struct btrfs_delayed_ref_node *node,
1685 struct btrfs_delayed_extent_op *extent_op,
1686 int insert_reserved)
1687 {
1688 int ret = 0;
1689 struct btrfs_delayed_data_ref *ref;
1690 struct btrfs_key ins;
1691 u64 parent = 0;
1692 u64 ref_root = 0;
1693 u64 flags = 0;
1694
1695 ins.objectid = node->bytenr;
1696 ins.offset = node->num_bytes;
1697 ins.type = BTRFS_EXTENT_ITEM_KEY;
1698
1699 ref = btrfs_delayed_node_to_data_ref(node);
1700 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1701 parent = ref->parent;
1702 else
1703 ref_root = ref->root;
1704
1705 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1706 if (extent_op) {
1707 BUG_ON(extent_op->update_key);
1708 flags |= extent_op->flags_to_set;
1709 }
1710 ret = alloc_reserved_file_extent(trans, root,
1711 parent, ref_root, flags,
1712 ref->objectid, ref->offset,
1713 &ins, node->ref_mod);
1714 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1715 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1716 node->num_bytes, parent,
1717 ref_root, ref->objectid,
1718 ref->offset, node->ref_mod,
1719 extent_op);
1720 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1721 ret = __btrfs_free_extent(trans, root, node->bytenr,
1722 node->num_bytes, parent,
1723 ref_root, ref->objectid,
1724 ref->offset, node->ref_mod,
1725 extent_op);
1726 } else {
1727 BUG();
1728 }
1729 return ret;
1730 }
1731
1732 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1733 struct extent_buffer *leaf,
1734 struct btrfs_extent_item *ei)
1735 {
1736 u64 flags = btrfs_extent_flags(leaf, ei);
1737 if (extent_op->update_flags) {
1738 flags |= extent_op->flags_to_set;
1739 btrfs_set_extent_flags(leaf, ei, flags);
1740 }
1741
1742 if (extent_op->update_key) {
1743 struct btrfs_tree_block_info *bi;
1744 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
1745 bi = (struct btrfs_tree_block_info *)(ei + 1);
1746 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
1747 }
1748 }
1749
1750 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
1751 struct btrfs_root *root,
1752 struct btrfs_delayed_ref_node *node,
1753 struct btrfs_delayed_extent_op *extent_op)
1754 {
1755 struct btrfs_key key;
1756 struct btrfs_path *path;
1757 struct btrfs_extent_item *ei;
1758 struct extent_buffer *leaf;
1759 u32 item_size;
1760 int ret;
1761 int err = 0;
1762
1763 path = btrfs_alloc_path();
1764 if (!path)
1765 return -ENOMEM;
1766
1767 key.objectid = node->bytenr;
1768 key.type = BTRFS_EXTENT_ITEM_KEY;
1769 key.offset = node->num_bytes;
1770
1771 path->reada = 1;
1772 path->leave_spinning = 1;
1773 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
1774 path, 0, 1);
1775 if (ret < 0) {
1776 err = ret;
1777 goto out;
1778 }
1779 if (ret > 0) {
1780 err = -EIO;
1781 goto out;
1782 }
1783
1784 leaf = path->nodes[0];
1785 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1786 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1787 if (item_size < sizeof(*ei)) {
1788 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
1789 path, (u64)-1, 0);
1790 if (ret < 0) {
1791 err = ret;
1792 goto out;
1793 }
1794 leaf = path->nodes[0];
1795 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1796 }
1797 #endif
1798 BUG_ON(item_size < sizeof(*ei));
1799 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1800 __run_delayed_extent_op(extent_op, leaf, ei);
1801
1802 btrfs_mark_buffer_dirty(leaf);
1803 out:
1804 btrfs_free_path(path);
1805 return err;
1806 }
1807
1808 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
1809 struct btrfs_root *root,
1810 struct btrfs_delayed_ref_node *node,
1811 struct btrfs_delayed_extent_op *extent_op,
1812 int insert_reserved)
1813 {
1814 int ret = 0;
1815 struct btrfs_delayed_tree_ref *ref;
1816 struct btrfs_key ins;
1817 u64 parent = 0;
1818 u64 ref_root = 0;
1819
1820 ins.objectid = node->bytenr;
1821 ins.offset = node->num_bytes;
1822 ins.type = BTRFS_EXTENT_ITEM_KEY;
1823
1824 ref = btrfs_delayed_node_to_tree_ref(node);
1825 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1826 parent = ref->parent;
1827 else
1828 ref_root = ref->root;
1829
1830 BUG_ON(node->ref_mod != 1);
1831 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1832 BUG_ON(!extent_op || !extent_op->update_flags ||
1833 !extent_op->update_key);
1834 ret = alloc_reserved_tree_block(trans, root,
1835 parent, ref_root,
1836 extent_op->flags_to_set,
1837 &extent_op->key,
1838 ref->level, &ins);
1839 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1840 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1841 node->num_bytes, parent, ref_root,
1842 ref->level, 0, 1, extent_op);
1843 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1844 ret = __btrfs_free_extent(trans, root, node->bytenr,
1845 node->num_bytes, parent, ref_root,
1846 ref->level, 0, 1, extent_op);
1847 } else {
1848 BUG();
1849 }
1850 return ret;
1851 }
1852
1853
1854 /* helper function to actually process a single delayed ref entry */
1855 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
1856 struct btrfs_root *root,
1857 struct btrfs_delayed_ref_node *node,
1858 struct btrfs_delayed_extent_op *extent_op,
1859 int insert_reserved)
1860 {
1861 int ret;
1862 if (btrfs_delayed_ref_is_head(node)) {
1863 struct btrfs_delayed_ref_head *head;
1864 /*
1865 * we've hit the end of the chain and we were supposed
1866 * to insert this extent into the tree. But, it got
1867 * deleted before we ever needed to insert it, so all
1868 * we have to do is clean up the accounting
1869 */
1870 BUG_ON(extent_op);
1871 head = btrfs_delayed_node_to_head(node);
1872 if (insert_reserved) {
1873 int mark_free = 0;
1874 struct extent_buffer *must_clean = NULL;
1875
1876 ret = pin_down_bytes(trans, root, NULL,
1877 node->bytenr, node->num_bytes,
1878 head->is_data, 1, &must_clean);
1879 if (ret > 0)
1880 mark_free = 1;
1881
1882 if (must_clean) {
1883 clean_tree_block(NULL, root, must_clean);
1884 btrfs_tree_unlock(must_clean);
1885 free_extent_buffer(must_clean);
1886 }
1887 if (head->is_data) {
1888 ret = btrfs_del_csums(trans, root,
1889 node->bytenr,
1890 node->num_bytes);
1891 BUG_ON(ret);
1892 }
1893 if (mark_free) {
1894 ret = btrfs_free_reserved_extent(root,
1895 node->bytenr,
1896 node->num_bytes);
1897 BUG_ON(ret);
1898 }
1899 }
1900 mutex_unlock(&head->mutex);
1901 return 0;
1902 }
1903
1904 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
1905 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1906 ret = run_delayed_tree_ref(trans, root, node, extent_op,
1907 insert_reserved);
1908 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
1909 node->type == BTRFS_SHARED_DATA_REF_KEY)
1910 ret = run_delayed_data_ref(trans, root, node, extent_op,
1911 insert_reserved);
1912 else
1913 BUG();
1914 return ret;
1915 }
1916
1917 static noinline struct btrfs_delayed_ref_node *
1918 select_delayed_ref(struct btrfs_delayed_ref_head *head)
1919 {
1920 struct rb_node *node;
1921 struct btrfs_delayed_ref_node *ref;
1922 int action = BTRFS_ADD_DELAYED_REF;
1923 again:
1924 /*
1925 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
1926 * this prevents ref count from going down to zero when
1927 * there still are pending delayed ref.
1928 */
1929 node = rb_prev(&head->node.rb_node);
1930 while (1) {
1931 if (!node)
1932 break;
1933 ref = rb_entry(node, struct btrfs_delayed_ref_node,
1934 rb_node);
1935 if (ref->bytenr != head->node.bytenr)
1936 break;
1937 if (ref->action == action)
1938 return ref;
1939 node = rb_prev(node);
1940 }
1941 if (action == BTRFS_ADD_DELAYED_REF) {
1942 action = BTRFS_DROP_DELAYED_REF;
1943 goto again;
1944 }
1945 return NULL;
1946 }
1947
1948 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
1949 struct btrfs_root *root,
1950 struct list_head *cluster)
1951 {
1952 struct btrfs_delayed_ref_root *delayed_refs;
1953 struct btrfs_delayed_ref_node *ref;
1954 struct btrfs_delayed_ref_head *locked_ref = NULL;
1955 struct btrfs_delayed_extent_op *extent_op;
1956 int ret;
1957 int count = 0;
1958 int must_insert_reserved = 0;
1959
1960 delayed_refs = &trans->transaction->delayed_refs;
1961 while (1) {
1962 if (!locked_ref) {
1963 /* pick a new head ref from the cluster list */
1964 if (list_empty(cluster))
1965 break;
1966
1967 locked_ref = list_entry(cluster->next,
1968 struct btrfs_delayed_ref_head, cluster);
1969
1970 /* grab the lock that says we are going to process
1971 * all the refs for this head */
1972 ret = btrfs_delayed_ref_lock(trans, locked_ref);
1973
1974 /*
1975 * we may have dropped the spin lock to get the head
1976 * mutex lock, and that might have given someone else
1977 * time to free the head. If that's true, it has been
1978 * removed from our list and we can move on.
1979 */
1980 if (ret == -EAGAIN) {
1981 locked_ref = NULL;
1982 count++;
1983 continue;
1984 }
1985 }
1986
1987 /*
1988 * record the must insert reserved flag before we
1989 * drop the spin lock.
1990 */
1991 must_insert_reserved = locked_ref->must_insert_reserved;
1992 locked_ref->must_insert_reserved = 0;
1993
1994 extent_op = locked_ref->extent_op;
1995 locked_ref->extent_op = NULL;
1996
1997 /*
1998 * locked_ref is the head node, so we have to go one
1999 * node back for any delayed ref updates
2000 */
2001 ref = select_delayed_ref(locked_ref);
2002 if (!ref) {
2003 /* All delayed refs have been processed, Go ahead
2004 * and send the head node to run_one_delayed_ref,
2005 * so that any accounting fixes can happen
2006 */
2007 ref = &locked_ref->node;
2008
2009 if (extent_op && must_insert_reserved) {
2010 kfree(extent_op);
2011 extent_op = NULL;
2012 }
2013
2014 if (extent_op) {
2015 spin_unlock(&delayed_refs->lock);
2016
2017 ret = run_delayed_extent_op(trans, root,
2018 ref, extent_op);
2019 BUG_ON(ret);
2020 kfree(extent_op);
2021
2022 cond_resched();
2023 spin_lock(&delayed_refs->lock);
2024 continue;
2025 }
2026
2027 list_del_init(&locked_ref->cluster);
2028 locked_ref = NULL;
2029 }
2030
2031 ref->in_tree = 0;
2032 rb_erase(&ref->rb_node, &delayed_refs->root);
2033 delayed_refs->num_entries--;
2034
2035 spin_unlock(&delayed_refs->lock);
2036
2037 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2038 must_insert_reserved);
2039 BUG_ON(ret);
2040
2041 btrfs_put_delayed_ref(ref);
2042 kfree(extent_op);
2043 count++;
2044
2045 cond_resched();
2046 spin_lock(&delayed_refs->lock);
2047 }
2048 return count;
2049 }
2050
2051 /*
2052 * this starts processing the delayed reference count updates and
2053 * extent insertions we have queued up so far. count can be
2054 * 0, which means to process everything in the tree at the start
2055 * of the run (but not newly added entries), or it can be some target
2056 * number you'd like to process.
2057 */
2058 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2059 struct btrfs_root *root, unsigned long count)
2060 {
2061 struct rb_node *node;
2062 struct btrfs_delayed_ref_root *delayed_refs;
2063 struct btrfs_delayed_ref_node *ref;
2064 struct list_head cluster;
2065 int ret;
2066 int run_all = count == (unsigned long)-1;
2067 int run_most = 0;
2068
2069 if (root == root->fs_info->extent_root)
2070 root = root->fs_info->tree_root;
2071
2072 delayed_refs = &trans->transaction->delayed_refs;
2073 INIT_LIST_HEAD(&cluster);
2074 again:
2075 spin_lock(&delayed_refs->lock);
2076 if (count == 0) {
2077 count = delayed_refs->num_entries * 2;
2078 run_most = 1;
2079 }
2080 while (1) {
2081 if (!(run_all || run_most) &&
2082 delayed_refs->num_heads_ready < 64)
2083 break;
2084
2085 /*
2086 * go find something we can process in the rbtree. We start at
2087 * the beginning of the tree, and then build a cluster
2088 * of refs to process starting at the first one we are able to
2089 * lock
2090 */
2091 ret = btrfs_find_ref_cluster(trans, &cluster,
2092 delayed_refs->run_delayed_start);
2093 if (ret)
2094 break;
2095
2096 ret = run_clustered_refs(trans, root, &cluster);
2097 BUG_ON(ret < 0);
2098
2099 count -= min_t(unsigned long, ret, count);
2100
2101 if (count == 0)
2102 break;
2103 }
2104
2105 if (run_all) {
2106 node = rb_first(&delayed_refs->root);
2107 if (!node)
2108 goto out;
2109 count = (unsigned long)-1;
2110
2111 while (node) {
2112 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2113 rb_node);
2114 if (btrfs_delayed_ref_is_head(ref)) {
2115 struct btrfs_delayed_ref_head *head;
2116
2117 head = btrfs_delayed_node_to_head(ref);
2118 atomic_inc(&ref->refs);
2119
2120 spin_unlock(&delayed_refs->lock);
2121 mutex_lock(&head->mutex);
2122 mutex_unlock(&head->mutex);
2123
2124 btrfs_put_delayed_ref(ref);
2125 cond_resched();
2126 goto again;
2127 }
2128 node = rb_next(node);
2129 }
2130 spin_unlock(&delayed_refs->lock);
2131 schedule_timeout(1);
2132 goto again;
2133 }
2134 out:
2135 spin_unlock(&delayed_refs->lock);
2136 return 0;
2137 }
2138
2139 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2140 struct btrfs_root *root,
2141 u64 bytenr, u64 num_bytes, u64 flags,
2142 int is_data)
2143 {
2144 struct btrfs_delayed_extent_op *extent_op;
2145 int ret;
2146
2147 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2148 if (!extent_op)
2149 return -ENOMEM;
2150
2151 extent_op->flags_to_set = flags;
2152 extent_op->update_flags = 1;
2153 extent_op->update_key = 0;
2154 extent_op->is_data = is_data ? 1 : 0;
2155
2156 ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op);
2157 if (ret)
2158 kfree(extent_op);
2159 return ret;
2160 }
2161
2162 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2163 struct btrfs_root *root,
2164 struct btrfs_path *path,
2165 u64 objectid, u64 offset, u64 bytenr)
2166 {
2167 struct btrfs_delayed_ref_head *head;
2168 struct btrfs_delayed_ref_node *ref;
2169 struct btrfs_delayed_data_ref *data_ref;
2170 struct btrfs_delayed_ref_root *delayed_refs;
2171 struct rb_node *node;
2172 int ret = 0;
2173
2174 ret = -ENOENT;
2175 delayed_refs = &trans->transaction->delayed_refs;
2176 spin_lock(&delayed_refs->lock);
2177 head = btrfs_find_delayed_ref_head(trans, bytenr);
2178 if (!head)
2179 goto out;
2180
2181 if (!mutex_trylock(&head->mutex)) {
2182 atomic_inc(&head->node.refs);
2183 spin_unlock(&delayed_refs->lock);
2184
2185 btrfs_release_path(root->fs_info->extent_root, path);
2186
2187 mutex_lock(&head->mutex);
2188 mutex_unlock(&head->mutex);
2189 btrfs_put_delayed_ref(&head->node);
2190 return -EAGAIN;
2191 }
2192
2193 node = rb_prev(&head->node.rb_node);
2194 if (!node)
2195 goto out_unlock;
2196
2197 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2198
2199 if (ref->bytenr != bytenr)
2200 goto out_unlock;
2201
2202 ret = 1;
2203 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2204 goto out_unlock;
2205
2206 data_ref = btrfs_delayed_node_to_data_ref(ref);
2207
2208 node = rb_prev(node);
2209 if (node) {
2210 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2211 if (ref->bytenr == bytenr)
2212 goto out_unlock;
2213 }
2214
2215 if (data_ref->root != root->root_key.objectid ||
2216 data_ref->objectid != objectid || data_ref->offset != offset)
2217 goto out_unlock;
2218
2219 ret = 0;
2220 out_unlock:
2221 mutex_unlock(&head->mutex);
2222 out:
2223 spin_unlock(&delayed_refs->lock);
2224 return ret;
2225 }
2226
2227 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2228 struct btrfs_root *root,
2229 struct btrfs_path *path,
2230 u64 objectid, u64 offset, u64 bytenr)
2231 {
2232 struct btrfs_root *extent_root = root->fs_info->extent_root;
2233 struct extent_buffer *leaf;
2234 struct btrfs_extent_data_ref *ref;
2235 struct btrfs_extent_inline_ref *iref;
2236 struct btrfs_extent_item *ei;
2237 struct btrfs_key key;
2238 u32 item_size;
2239 int ret;
2240
2241 key.objectid = bytenr;
2242 key.offset = (u64)-1;
2243 key.type = BTRFS_EXTENT_ITEM_KEY;
2244
2245 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2246 if (ret < 0)
2247 goto out;
2248 BUG_ON(ret == 0);
2249
2250 ret = -ENOENT;
2251 if (path->slots[0] == 0)
2252 goto out;
2253
2254 path->slots[0]--;
2255 leaf = path->nodes[0];
2256 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2257
2258 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2259 goto out;
2260
2261 ret = 1;
2262 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2263 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2264 if (item_size < sizeof(*ei)) {
2265 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2266 goto out;
2267 }
2268 #endif
2269 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2270
2271 if (item_size != sizeof(*ei) +
2272 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2273 goto out;
2274
2275 if (btrfs_extent_generation(leaf, ei) <=
2276 btrfs_root_last_snapshot(&root->root_item))
2277 goto out;
2278
2279 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2280 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2281 BTRFS_EXTENT_DATA_REF_KEY)
2282 goto out;
2283
2284 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2285 if (btrfs_extent_refs(leaf, ei) !=
2286 btrfs_extent_data_ref_count(leaf, ref) ||
2287 btrfs_extent_data_ref_root(leaf, ref) !=
2288 root->root_key.objectid ||
2289 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2290 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2291 goto out;
2292
2293 ret = 0;
2294 out:
2295 return ret;
2296 }
2297
2298 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2299 struct btrfs_root *root,
2300 u64 objectid, u64 offset, u64 bytenr)
2301 {
2302 struct btrfs_path *path;
2303 int ret;
2304 int ret2;
2305
2306 path = btrfs_alloc_path();
2307 if (!path)
2308 return -ENOENT;
2309
2310 do {
2311 ret = check_committed_ref(trans, root, path, objectid,
2312 offset, bytenr);
2313 if (ret && ret != -ENOENT)
2314 goto out;
2315
2316 ret2 = check_delayed_ref(trans, root, path, objectid,
2317 offset, bytenr);
2318 } while (ret2 == -EAGAIN);
2319
2320 if (ret2 && ret2 != -ENOENT) {
2321 ret = ret2;
2322 goto out;
2323 }
2324
2325 if (ret != -ENOENT || ret2 != -ENOENT)
2326 ret = 0;
2327 out:
2328 btrfs_free_path(path);
2329 return ret;
2330 }
2331
2332 #if 0
2333 int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2334 struct extent_buffer *buf, u32 nr_extents)
2335 {
2336 struct btrfs_key key;
2337 struct btrfs_file_extent_item *fi;
2338 u64 root_gen;
2339 u32 nritems;
2340 int i;
2341 int level;
2342 int ret = 0;
2343 int shared = 0;
2344
2345 if (!root->ref_cows)
2346 return 0;
2347
2348 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
2349 shared = 0;
2350 root_gen = root->root_key.offset;
2351 } else {
2352 shared = 1;
2353 root_gen = trans->transid - 1;
2354 }
2355
2356 level = btrfs_header_level(buf);
2357 nritems = btrfs_header_nritems(buf);
2358
2359 if (level == 0) {
2360 struct btrfs_leaf_ref *ref;
2361 struct btrfs_extent_info *info;
2362
2363 ref = btrfs_alloc_leaf_ref(root, nr_extents);
2364 if (!ref) {
2365 ret = -ENOMEM;
2366 goto out;
2367 }
2368
2369 ref->root_gen = root_gen;
2370 ref->bytenr = buf->start;
2371 ref->owner = btrfs_header_owner(buf);
2372 ref->generation = btrfs_header_generation(buf);
2373 ref->nritems = nr_extents;
2374 info = ref->extents;
2375
2376 for (i = 0; nr_extents > 0 && i < nritems; i++) {
2377 u64 disk_bytenr;
2378 btrfs_item_key_to_cpu(buf, &key, i);
2379 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2380 continue;
2381 fi = btrfs_item_ptr(buf, i,
2382 struct btrfs_file_extent_item);
2383 if (btrfs_file_extent_type(buf, fi) ==
2384 BTRFS_FILE_EXTENT_INLINE)
2385 continue;
2386 disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2387 if (disk_bytenr == 0)
2388 continue;
2389
2390 info->bytenr = disk_bytenr;
2391 info->num_bytes =
2392 btrfs_file_extent_disk_num_bytes(buf, fi);
2393 info->objectid = key.objectid;
2394 info->offset = key.offset;
2395 info++;
2396 }
2397
2398 ret = btrfs_add_leaf_ref(root, ref, shared);
2399 if (ret == -EEXIST && shared) {
2400 struct btrfs_leaf_ref *old;
2401 old = btrfs_lookup_leaf_ref(root, ref->bytenr);
2402 BUG_ON(!old);
2403 btrfs_remove_leaf_ref(root, old);
2404 btrfs_free_leaf_ref(root, old);
2405 ret = btrfs_add_leaf_ref(root, ref, shared);
2406 }
2407 WARN_ON(ret);
2408 btrfs_free_leaf_ref(root, ref);
2409 }
2410 out:
2411 return ret;
2412 }
2413
2414 /* when a block goes through cow, we update the reference counts of
2415 * everything that block points to. The internal pointers of the block
2416 * can be in just about any order, and it is likely to have clusters of
2417 * things that are close together and clusters of things that are not.
2418 *
2419 * To help reduce the seeks that come with updating all of these reference
2420 * counts, sort them by byte number before actual updates are done.
2421 *
2422 * struct refsort is used to match byte number to slot in the btree block.
2423 * we sort based on the byte number and then use the slot to actually
2424 * find the item.
2425 *
2426 * struct refsort is smaller than strcut btrfs_item and smaller than
2427 * struct btrfs_key_ptr. Since we're currently limited to the page size
2428 * for a btree block, there's no way for a kmalloc of refsorts for a
2429 * single node to be bigger than a page.
2430 */
2431 struct refsort {
2432 u64 bytenr;
2433 u32 slot;
2434 };
2435
2436 /*
2437 * for passing into sort()
2438 */
2439 static int refsort_cmp(const void *a_void, const void *b_void)
2440 {
2441 const struct refsort *a = a_void;
2442 const struct refsort *b = b_void;
2443
2444 if (a->bytenr < b->bytenr)
2445 return -1;
2446 if (a->bytenr > b->bytenr)
2447 return 1;
2448 return 0;
2449 }
2450 #endif
2451
2452 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2453 struct btrfs_root *root,
2454 struct extent_buffer *buf,
2455 int full_backref, int inc)
2456 {
2457 u64 bytenr;
2458 u64 num_bytes;
2459 u64 parent;
2460 u64 ref_root;
2461 u32 nritems;
2462 struct btrfs_key key;
2463 struct btrfs_file_extent_item *fi;
2464 int i;
2465 int level;
2466 int ret = 0;
2467 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2468 u64, u64, u64, u64, u64, u64);
2469
2470 ref_root = btrfs_header_owner(buf);
2471 nritems = btrfs_header_nritems(buf);
2472 level = btrfs_header_level(buf);
2473
2474 if (!root->ref_cows && level == 0)
2475 return 0;
2476
2477 if (inc)
2478 process_func = btrfs_inc_extent_ref;
2479 else
2480 process_func = btrfs_free_extent;
2481
2482 if (full_backref)
2483 parent = buf->start;
2484 else
2485 parent = 0;
2486
2487 for (i = 0; i < nritems; i++) {
2488 if (level == 0) {
2489 btrfs_item_key_to_cpu(buf, &key, i);
2490 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2491 continue;
2492 fi = btrfs_item_ptr(buf, i,
2493 struct btrfs_file_extent_item);
2494 if (btrfs_file_extent_type(buf, fi) ==
2495 BTRFS_FILE_EXTENT_INLINE)
2496 continue;
2497 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2498 if (bytenr == 0)
2499 continue;
2500
2501 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2502 key.offset -= btrfs_file_extent_offset(buf, fi);
2503 ret = process_func(trans, root, bytenr, num_bytes,
2504 parent, ref_root, key.objectid,
2505 key.offset);
2506 if (ret)
2507 goto fail;
2508 } else {
2509 bytenr = btrfs_node_blockptr(buf, i);
2510 num_bytes = btrfs_level_size(root, level - 1);
2511 ret = process_func(trans, root, bytenr, num_bytes,
2512 parent, ref_root, level - 1, 0);
2513 if (ret)
2514 goto fail;
2515 }
2516 }
2517 return 0;
2518 fail:
2519 BUG();
2520 return ret;
2521 }
2522
2523 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2524 struct extent_buffer *buf, int full_backref)
2525 {
2526 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
2527 }
2528
2529 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2530 struct extent_buffer *buf, int full_backref)
2531 {
2532 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
2533 }
2534
2535 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2536 struct btrfs_root *root,
2537 struct btrfs_path *path,
2538 struct btrfs_block_group_cache *cache)
2539 {
2540 int ret;
2541 struct btrfs_root *extent_root = root->fs_info->extent_root;
2542 unsigned long bi;
2543 struct extent_buffer *leaf;
2544
2545 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2546 if (ret < 0)
2547 goto fail;
2548 BUG_ON(ret);
2549
2550 leaf = path->nodes[0];
2551 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2552 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2553 btrfs_mark_buffer_dirty(leaf);
2554 btrfs_release_path(extent_root, path);
2555 fail:
2556 if (ret)
2557 return ret;
2558 return 0;
2559
2560 }
2561
2562 static struct btrfs_block_group_cache *
2563 next_block_group(struct btrfs_root *root,
2564 struct btrfs_block_group_cache *cache)
2565 {
2566 struct rb_node *node;
2567 spin_lock(&root->fs_info->block_group_cache_lock);
2568 node = rb_next(&cache->cache_node);
2569 btrfs_put_block_group(cache);
2570 if (node) {
2571 cache = rb_entry(node, struct btrfs_block_group_cache,
2572 cache_node);
2573 atomic_inc(&cache->count);
2574 } else
2575 cache = NULL;
2576 spin_unlock(&root->fs_info->block_group_cache_lock);
2577 return cache;
2578 }
2579
2580 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
2581 struct btrfs_root *root)
2582 {
2583 struct btrfs_block_group_cache *cache;
2584 int err = 0;
2585 struct btrfs_path *path;
2586 u64 last = 0;
2587
2588 path = btrfs_alloc_path();
2589 if (!path)
2590 return -ENOMEM;
2591
2592 while (1) {
2593 if (last == 0) {
2594 err = btrfs_run_delayed_refs(trans, root,
2595 (unsigned long)-1);
2596 BUG_ON(err);
2597 }
2598
2599 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2600 while (cache) {
2601 if (cache->dirty)
2602 break;
2603 cache = next_block_group(root, cache);
2604 }
2605 if (!cache) {
2606 if (last == 0)
2607 break;
2608 last = 0;
2609 continue;
2610 }
2611
2612 cache->dirty = 0;
2613 last = cache->key.objectid + cache->key.offset;
2614
2615 err = write_one_cache_group(trans, root, path, cache);
2616 BUG_ON(err);
2617 btrfs_put_block_group(cache);
2618 }
2619
2620 btrfs_free_path(path);
2621 return 0;
2622 }
2623
2624 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
2625 {
2626 struct btrfs_block_group_cache *block_group;
2627 int readonly = 0;
2628
2629 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
2630 if (!block_group || block_group->ro)
2631 readonly = 1;
2632 if (block_group)
2633 btrfs_put_block_group(block_group);
2634 return readonly;
2635 }
2636
2637 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
2638 u64 total_bytes, u64 bytes_used,
2639 struct btrfs_space_info **space_info)
2640 {
2641 struct btrfs_space_info *found;
2642
2643 found = __find_space_info(info, flags);
2644 if (found) {
2645 spin_lock(&found->lock);
2646 found->total_bytes += total_bytes;
2647 found->bytes_used += bytes_used;
2648 found->full = 0;
2649 spin_unlock(&found->lock);
2650 *space_info = found;
2651 return 0;
2652 }
2653 found = kzalloc(sizeof(*found), GFP_NOFS);
2654 if (!found)
2655 return -ENOMEM;
2656
2657 INIT_LIST_HEAD(&found->block_groups);
2658 init_rwsem(&found->groups_sem);
2659 spin_lock_init(&found->lock);
2660 found->flags = flags;
2661 found->total_bytes = total_bytes;
2662 found->bytes_used = bytes_used;
2663 found->bytes_pinned = 0;
2664 found->bytes_reserved = 0;
2665 found->bytes_readonly = 0;
2666 found->bytes_delalloc = 0;
2667 found->full = 0;
2668 found->force_alloc = 0;
2669 *space_info = found;
2670 list_add_rcu(&found->list, &info->space_info);
2671 atomic_set(&found->caching_threads, 0);
2672 return 0;
2673 }
2674
2675 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
2676 {
2677 u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
2678 BTRFS_BLOCK_GROUP_RAID1 |
2679 BTRFS_BLOCK_GROUP_RAID10 |
2680 BTRFS_BLOCK_GROUP_DUP);
2681 if (extra_flags) {
2682 if (flags & BTRFS_BLOCK_GROUP_DATA)
2683 fs_info->avail_data_alloc_bits |= extra_flags;
2684 if (flags & BTRFS_BLOCK_GROUP_METADATA)
2685 fs_info->avail_metadata_alloc_bits |= extra_flags;
2686 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
2687 fs_info->avail_system_alloc_bits |= extra_flags;
2688 }
2689 }
2690
2691 static void set_block_group_readonly(struct btrfs_block_group_cache *cache)
2692 {
2693 spin_lock(&cache->space_info->lock);
2694 spin_lock(&cache->lock);
2695 if (!cache->ro) {
2696 cache->space_info->bytes_readonly += cache->key.offset -
2697 btrfs_block_group_used(&cache->item);
2698 cache->ro = 1;
2699 }
2700 spin_unlock(&cache->lock);
2701 spin_unlock(&cache->space_info->lock);
2702 }
2703
2704 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
2705 {
2706 u64 num_devices = root->fs_info->fs_devices->rw_devices;
2707
2708 if (num_devices == 1)
2709 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
2710 if (num_devices < 4)
2711 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
2712
2713 if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
2714 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
2715 BTRFS_BLOCK_GROUP_RAID10))) {
2716 flags &= ~BTRFS_BLOCK_GROUP_DUP;
2717 }
2718
2719 if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
2720 (flags & BTRFS_BLOCK_GROUP_RAID10)) {
2721 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
2722 }
2723
2724 if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
2725 ((flags & BTRFS_BLOCK_GROUP_RAID1) |
2726 (flags & BTRFS_BLOCK_GROUP_RAID10) |
2727 (flags & BTRFS_BLOCK_GROUP_DUP)))
2728 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
2729 return flags;
2730 }
2731
2732 static u64 btrfs_get_alloc_profile(struct btrfs_root *root, u64 data)
2733 {
2734 struct btrfs_fs_info *info = root->fs_info;
2735 u64 alloc_profile;
2736
2737 if (data) {
2738 alloc_profile = info->avail_data_alloc_bits &
2739 info->data_alloc_profile;
2740 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
2741 } else if (root == root->fs_info->chunk_root) {
2742 alloc_profile = info->avail_system_alloc_bits &
2743 info->system_alloc_profile;
2744 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
2745 } else {
2746 alloc_profile = info->avail_metadata_alloc_bits &
2747 info->metadata_alloc_profile;
2748 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
2749 }
2750
2751 return btrfs_reduce_alloc_profile(root, data);
2752 }
2753
2754 void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
2755 {
2756 u64 alloc_target;
2757
2758 alloc_target = btrfs_get_alloc_profile(root, 1);
2759 BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
2760 alloc_target);
2761 }
2762
2763 /*
2764 * for now this just makes sure we have at least 5% of our metadata space free
2765 * for use.
2766 */
2767 int btrfs_check_metadata_free_space(struct btrfs_root *root)
2768 {
2769 struct btrfs_fs_info *info = root->fs_info;
2770 struct btrfs_space_info *meta_sinfo;
2771 u64 alloc_target, thresh;
2772 int committed = 0, ret;
2773
2774 /* get the space info for where the metadata will live */
2775 alloc_target = btrfs_get_alloc_profile(root, 0);
2776 meta_sinfo = __find_space_info(info, alloc_target);
2777
2778 again:
2779 spin_lock(&meta_sinfo->lock);
2780 if (!meta_sinfo->full)
2781 thresh = meta_sinfo->total_bytes * 80;
2782 else
2783 thresh = meta_sinfo->total_bytes * 95;
2784
2785 do_div(thresh, 100);
2786
2787 if (meta_sinfo->bytes_used + meta_sinfo->bytes_reserved +
2788 meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly > thresh) {
2789 struct btrfs_trans_handle *trans;
2790 if (!meta_sinfo->full) {
2791 meta_sinfo->force_alloc = 1;
2792 spin_unlock(&meta_sinfo->lock);
2793
2794 trans = btrfs_start_transaction(root, 1);
2795 if (!trans)
2796 return -ENOMEM;
2797
2798 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2799 2 * 1024 * 1024, alloc_target, 0);
2800 btrfs_end_transaction(trans, root);
2801 goto again;
2802 }
2803 spin_unlock(&meta_sinfo->lock);
2804
2805 if (!committed) {
2806 committed = 1;
2807 trans = btrfs_join_transaction(root, 1);
2808 if (!trans)
2809 return -ENOMEM;
2810 ret = btrfs_commit_transaction(trans, root);
2811 if (ret)
2812 return ret;
2813 goto again;
2814 }
2815 return -ENOSPC;
2816 }
2817 spin_unlock(&meta_sinfo->lock);
2818
2819 return 0;
2820 }
2821
2822 /*
2823 * This will check the space that the inode allocates from to make sure we have
2824 * enough space for bytes.
2825 */
2826 int btrfs_check_data_free_space(struct btrfs_root *root, struct inode *inode,
2827 u64 bytes)
2828 {
2829 struct btrfs_space_info *data_sinfo;
2830 int ret = 0, committed = 0;
2831
2832 /* make sure bytes are sectorsize aligned */
2833 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
2834
2835 data_sinfo = BTRFS_I(inode)->space_info;
2836 again:
2837 /* make sure we have enough space to handle the data first */
2838 spin_lock(&data_sinfo->lock);
2839 if (data_sinfo->total_bytes - data_sinfo->bytes_used -
2840 data_sinfo->bytes_delalloc - data_sinfo->bytes_reserved -
2841 data_sinfo->bytes_pinned - data_sinfo->bytes_readonly -
2842 data_sinfo->bytes_may_use < bytes) {
2843 struct btrfs_trans_handle *trans;
2844
2845 /*
2846 * if we don't have enough free bytes in this space then we need
2847 * to alloc a new chunk.
2848 */
2849 if (!data_sinfo->full) {
2850 u64 alloc_target;
2851
2852 data_sinfo->force_alloc = 1;
2853 spin_unlock(&data_sinfo->lock);
2854
2855 alloc_target = btrfs_get_alloc_profile(root, 1);
2856 trans = btrfs_start_transaction(root, 1);
2857 if (!trans)
2858 return -ENOMEM;
2859
2860 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2861 bytes + 2 * 1024 * 1024,
2862 alloc_target, 0);
2863 btrfs_end_transaction(trans, root);
2864 if (ret)
2865 return ret;
2866 goto again;
2867 }
2868 spin_unlock(&data_sinfo->lock);
2869
2870 /* commit the current transaction and try again */
2871 if (!committed) {
2872 committed = 1;
2873 trans = btrfs_join_transaction(root, 1);
2874 if (!trans)
2875 return -ENOMEM;
2876 ret = btrfs_commit_transaction(trans, root);
2877 if (ret)
2878 return ret;
2879 goto again;
2880 }
2881
2882 printk(KERN_ERR "no space left, need %llu, %llu delalloc bytes"
2883 ", %llu bytes_used, %llu bytes_reserved, "
2884 "%llu bytes_pinned, %llu bytes_readonly, %llu may use "
2885 "%llu total\n", (unsigned long long)bytes,
2886 (unsigned long long)data_sinfo->bytes_delalloc,
2887 (unsigned long long)data_sinfo->bytes_used,
2888 (unsigned long long)data_sinfo->bytes_reserved,
2889 (unsigned long long)data_sinfo->bytes_pinned,
2890 (unsigned long long)data_sinfo->bytes_readonly,
2891 (unsigned long long)data_sinfo->bytes_may_use,
2892 (unsigned long long)data_sinfo->total_bytes);
2893 return -ENOSPC;
2894 }
2895 data_sinfo->bytes_may_use += bytes;
2896 BTRFS_I(inode)->reserved_bytes += bytes;
2897 spin_unlock(&data_sinfo->lock);
2898
2899 return btrfs_check_metadata_free_space(root);
2900 }
2901
2902 /*
2903 * if there was an error for whatever reason after calling
2904 * btrfs_check_data_free_space, call this so we can cleanup the counters.
2905 */
2906 void btrfs_free_reserved_data_space(struct btrfs_root *root,
2907 struct inode *inode, u64 bytes)
2908 {
2909 struct btrfs_space_info *data_sinfo;
2910
2911 /* make sure bytes are sectorsize aligned */
2912 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
2913
2914 data_sinfo = BTRFS_I(inode)->space_info;
2915 spin_lock(&data_sinfo->lock);
2916 data_sinfo->bytes_may_use -= bytes;
2917 BTRFS_I(inode)->reserved_bytes -= bytes;
2918 spin_unlock(&data_sinfo->lock);
2919 }
2920
2921 /* called when we are adding a delalloc extent to the inode's io_tree */
2922 void btrfs_delalloc_reserve_space(struct btrfs_root *root, struct inode *inode,
2923 u64 bytes)
2924 {
2925 struct btrfs_space_info *data_sinfo;
2926
2927 /* get the space info for where this inode will be storing its data */
2928 data_sinfo = BTRFS_I(inode)->space_info;
2929
2930 /* make sure we have enough space to handle the data first */
2931 spin_lock(&data_sinfo->lock);
2932 data_sinfo->bytes_delalloc += bytes;
2933
2934 /*
2935 * we are adding a delalloc extent without calling
2936 * btrfs_check_data_free_space first. This happens on a weird
2937 * writepage condition, but shouldn't hurt our accounting
2938 */
2939 if (unlikely(bytes > BTRFS_I(inode)->reserved_bytes)) {
2940 data_sinfo->bytes_may_use -= BTRFS_I(inode)->reserved_bytes;
2941 BTRFS_I(inode)->reserved_bytes = 0;
2942 } else {
2943 data_sinfo->bytes_may_use -= bytes;
2944 BTRFS_I(inode)->reserved_bytes -= bytes;
2945 }
2946
2947 spin_unlock(&data_sinfo->lock);
2948 }
2949
2950 /* called when we are clearing an delalloc extent from the inode's io_tree */
2951 void btrfs_delalloc_free_space(struct btrfs_root *root, struct inode *inode,
2952 u64 bytes)
2953 {
2954 struct btrfs_space_info *info;
2955
2956 info = BTRFS_I(inode)->space_info;
2957
2958 spin_lock(&info->lock);
2959 info->bytes_delalloc -= bytes;
2960 spin_unlock(&info->lock);
2961 }
2962
2963 static void force_metadata_allocation(struct btrfs_fs_info *info)
2964 {
2965 struct list_head *head = &info->space_info;
2966 struct btrfs_space_info *found;
2967
2968 rcu_read_lock();
2969 list_for_each_entry_rcu(found, head, list) {
2970 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
2971 found->force_alloc = 1;
2972 }
2973 rcu_read_unlock();
2974 }
2975
2976 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
2977 struct btrfs_root *extent_root, u64 alloc_bytes,
2978 u64 flags, int force)
2979 {
2980 struct btrfs_space_info *space_info;
2981 struct btrfs_fs_info *fs_info = extent_root->fs_info;
2982 u64 thresh;
2983 int ret = 0;
2984
2985 mutex_lock(&fs_info->chunk_mutex);
2986
2987 flags = btrfs_reduce_alloc_profile(extent_root, flags);
2988
2989 space_info = __find_space_info(extent_root->fs_info, flags);
2990 if (!space_info) {
2991 ret = update_space_info(extent_root->fs_info, flags,
2992 0, 0, &space_info);
2993 BUG_ON(ret);
2994 }
2995 BUG_ON(!space_info);
2996
2997 spin_lock(&space_info->lock);
2998 if (space_info->force_alloc) {
2999 force = 1;
3000 space_info->force_alloc = 0;
3001 }
3002 if (space_info->full) {
3003 spin_unlock(&space_info->lock);
3004 goto out;
3005 }
3006
3007 thresh = space_info->total_bytes - space_info->bytes_readonly;
3008 thresh = div_factor(thresh, 6);
3009 if (!force &&
3010 (space_info->bytes_used + space_info->bytes_pinned +
3011 space_info->bytes_reserved + alloc_bytes) < thresh) {
3012 spin_unlock(&space_info->lock);
3013 goto out;
3014 }
3015 spin_unlock(&space_info->lock);
3016
3017 /*
3018 * if we're doing a data chunk, go ahead and make sure that
3019 * we keep a reasonable number of metadata chunks allocated in the
3020 * FS as well.
3021 */
3022 if (flags & BTRFS_BLOCK_GROUP_DATA) {
3023 fs_info->data_chunk_allocations++;
3024 if (!(fs_info->data_chunk_allocations %
3025 fs_info->metadata_ratio))
3026 force_metadata_allocation(fs_info);
3027 }
3028
3029 ret = btrfs_alloc_chunk(trans, extent_root, flags);
3030 if (ret)
3031 space_info->full = 1;
3032 out:
3033 mutex_unlock(&extent_root->fs_info->chunk_mutex);
3034 return ret;
3035 }
3036
3037 static int update_block_group(struct btrfs_trans_handle *trans,
3038 struct btrfs_root *root,
3039 u64 bytenr, u64 num_bytes, int alloc,
3040 int mark_free)
3041 {
3042 struct btrfs_block_group_cache *cache;
3043 struct btrfs_fs_info *info = root->fs_info;
3044 u64 total = num_bytes;
3045 u64 old_val;
3046 u64 byte_in_group;
3047
3048 /* block accounting for super block */
3049 spin_lock(&info->delalloc_lock);
3050 old_val = btrfs_super_bytes_used(&info->super_copy);
3051 if (alloc)
3052 old_val += num_bytes;
3053 else
3054 old_val -= num_bytes;
3055 btrfs_set_super_bytes_used(&info->super_copy, old_val);
3056
3057 /* block accounting for root item */
3058 old_val = btrfs_root_used(&root->root_item);
3059 if (alloc)
3060 old_val += num_bytes;
3061 else
3062 old_val -= num_bytes;
3063 btrfs_set_root_used(&root->root_item, old_val);
3064 spin_unlock(&info->delalloc_lock);
3065
3066 while (total) {
3067 cache = btrfs_lookup_block_group(info, bytenr);
3068 if (!cache)
3069 return -1;
3070 byte_in_group = bytenr - cache->key.objectid;
3071 WARN_ON(byte_in_group > cache->key.offset);
3072
3073 spin_lock(&cache->space_info->lock);
3074 spin_lock(&cache->lock);
3075 cache->dirty = 1;
3076 old_val = btrfs_block_group_used(&cache->item);
3077 num_bytes = min(total, cache->key.offset - byte_in_group);
3078 if (alloc) {
3079 old_val += num_bytes;
3080 btrfs_set_block_group_used(&cache->item, old_val);
3081 cache->reserved -= num_bytes;
3082 cache->space_info->bytes_used += num_bytes;
3083 cache->space_info->bytes_reserved -= num_bytes;
3084 if (cache->ro)
3085 cache->space_info->bytes_readonly -= num_bytes;
3086 spin_unlock(&cache->lock);
3087 spin_unlock(&cache->space_info->lock);
3088 } else {
3089 old_val -= num_bytes;
3090 cache->space_info->bytes_used -= num_bytes;
3091 if (cache->ro)
3092 cache->space_info->bytes_readonly += num_bytes;
3093 btrfs_set_block_group_used(&cache->item, old_val);
3094 spin_unlock(&cache->lock);
3095 spin_unlock(&cache->space_info->lock);
3096 if (mark_free) {
3097 int ret;
3098
3099 ret = btrfs_discard_extent(root, bytenr,
3100 num_bytes);
3101 WARN_ON(ret);
3102
3103 ret = btrfs_add_free_space(cache, bytenr,
3104 num_bytes);
3105 WARN_ON(ret);
3106 }
3107 }
3108 btrfs_put_block_group(cache);
3109 total -= num_bytes;
3110 bytenr += num_bytes;
3111 }
3112 return 0;
3113 }
3114
3115 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
3116 {
3117 struct btrfs_block_group_cache *cache;
3118 u64 bytenr;
3119
3120 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
3121 if (!cache)
3122 return 0;
3123
3124 bytenr = cache->key.objectid;
3125 btrfs_put_block_group(cache);
3126
3127 return bytenr;
3128 }
3129
3130 /*
3131 * this function must be called within transaction
3132 */
3133 int btrfs_pin_extent(struct btrfs_root *root,
3134 u64 bytenr, u64 num_bytes, int reserved)
3135 {
3136 struct btrfs_fs_info *fs_info = root->fs_info;
3137 struct btrfs_block_group_cache *cache;
3138
3139 cache = btrfs_lookup_block_group(fs_info, bytenr);
3140 BUG_ON(!cache);
3141
3142 spin_lock(&cache->space_info->lock);
3143 spin_lock(&cache->lock);
3144 cache->pinned += num_bytes;
3145 cache->space_info->bytes_pinned += num_bytes;
3146 if (reserved) {
3147 cache->reserved -= num_bytes;
3148 cache->space_info->bytes_reserved -= num_bytes;
3149 }
3150 spin_unlock(&cache->lock);
3151 spin_unlock(&cache->space_info->lock);
3152
3153 btrfs_put_block_group(cache);
3154
3155 set_extent_dirty(fs_info->pinned_extents,
3156 bytenr, bytenr + num_bytes - 1, GFP_NOFS);
3157 return 0;
3158 }
3159
3160 static int update_reserved_extents(struct btrfs_block_group_cache *cache,
3161 u64 num_bytes, int reserve)
3162 {
3163 spin_lock(&cache->space_info->lock);
3164 spin_lock(&cache->lock);
3165 if (reserve) {
3166 cache->reserved += num_bytes;
3167 cache->space_info->bytes_reserved += num_bytes;
3168 } else {
3169 cache->reserved -= num_bytes;
3170 cache->space_info->bytes_reserved -= num_bytes;
3171 }
3172 spin_unlock(&cache->lock);
3173 spin_unlock(&cache->space_info->lock);
3174 return 0;
3175 }
3176
3177 int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
3178 struct btrfs_root *root)
3179 {
3180 struct btrfs_fs_info *fs_info = root->fs_info;
3181 struct btrfs_caching_control *next;
3182 struct btrfs_caching_control *caching_ctl;
3183 struct btrfs_block_group_cache *cache;
3184
3185 down_write(&fs_info->extent_commit_sem);
3186
3187 list_for_each_entry_safe(caching_ctl, next,
3188 &fs_info->caching_block_groups, list) {
3189 cache = caching_ctl->block_group;
3190 if (block_group_cache_done(cache)) {
3191 cache->last_byte_to_unpin = (u64)-1;
3192 list_del_init(&caching_ctl->list);
3193 put_caching_control(caching_ctl);
3194 } else {
3195 cache->last_byte_to_unpin = caching_ctl->progress;
3196 }
3197 }
3198
3199 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
3200 fs_info->pinned_extents = &fs_info->freed_extents[1];
3201 else
3202 fs_info->pinned_extents = &fs_info->freed_extents[0];
3203
3204 up_write(&fs_info->extent_commit_sem);
3205 return 0;
3206 }
3207
3208 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
3209 {
3210 struct btrfs_fs_info *fs_info = root->fs_info;
3211 struct btrfs_block_group_cache *cache = NULL;
3212 u64 len;
3213
3214 while (start <= end) {
3215 if (!cache ||
3216 start >= cache->key.objectid + cache->key.offset) {
3217 if (cache)
3218 btrfs_put_block_group(cache);
3219 cache = btrfs_lookup_block_group(fs_info, start);
3220 BUG_ON(!cache);
3221 }
3222
3223 len = cache->key.objectid + cache->key.offset - start;
3224 len = min(len, end + 1 - start);
3225
3226 if (start < cache->last_byte_to_unpin) {
3227 len = min(len, cache->last_byte_to_unpin - start);
3228 btrfs_add_free_space(cache, start, len);
3229 }
3230
3231 spin_lock(&cache->space_info->lock);
3232 spin_lock(&cache->lock);
3233 cache->pinned -= len;
3234 cache->space_info->bytes_pinned -= len;
3235 spin_unlock(&cache->lock);
3236 spin_unlock(&cache->space_info->lock);
3237
3238 start += len;
3239 }
3240
3241 if (cache)
3242 btrfs_put_block_group(cache);
3243 return 0;
3244 }
3245
3246 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
3247 struct btrfs_root *root)
3248 {
3249 struct btrfs_fs_info *fs_info = root->fs_info;
3250 struct extent_io_tree *unpin;
3251 u64 start;
3252 u64 end;
3253 int ret;
3254
3255 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
3256 unpin = &fs_info->freed_extents[1];
3257 else
3258 unpin = &fs_info->freed_extents[0];
3259
3260 while (1) {
3261 ret = find_first_extent_bit(unpin, 0, &start, &end,
3262 EXTENT_DIRTY);
3263 if (ret)
3264 break;
3265
3266 ret = btrfs_discard_extent(root, start, end + 1 - start);
3267
3268 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3269 unpin_extent_range(root, start, end);
3270 cond_resched();
3271 }
3272
3273 return ret;
3274 }
3275
3276 static int pin_down_bytes(struct btrfs_trans_handle *trans,
3277 struct btrfs_root *root,
3278 struct btrfs_path *path,
3279 u64 bytenr, u64 num_bytes,
3280 int is_data, int reserved,
3281 struct extent_buffer **must_clean)
3282 {
3283 int err = 0;
3284 struct extent_buffer *buf;
3285
3286 if (is_data)
3287 goto pinit;
3288
3289 buf = btrfs_find_tree_block(root, bytenr, num_bytes);
3290 if (!buf)
3291 goto pinit;
3292
3293 /* we can reuse a block if it hasn't been written
3294 * and it is from this transaction. We can't
3295 * reuse anything from the tree log root because
3296 * it has tiny sub-transactions.
3297 */
3298 if (btrfs_buffer_uptodate(buf, 0) &&
3299 btrfs_try_tree_lock(buf)) {
3300 u64 header_owner = btrfs_header_owner(buf);
3301 u64 header_transid = btrfs_header_generation(buf);
3302 if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
3303 header_transid == trans->transid &&
3304 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
3305 *must_clean = buf;
3306 return 1;
3307 }
3308 btrfs_tree_unlock(buf);
3309 }
3310 free_extent_buffer(buf);
3311 pinit:
3312 if (path)
3313 btrfs_set_path_blocking(path);
3314 /* unlocks the pinned mutex */
3315 btrfs_pin_extent(root, bytenr, num_bytes, reserved);
3316
3317 BUG_ON(err < 0);
3318 return 0;
3319 }
3320
3321 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
3322 struct btrfs_root *root,
3323 u64 bytenr, u64 num_bytes, u64 parent,
3324 u64 root_objectid, u64 owner_objectid,
3325 u64 owner_offset, int refs_to_drop,
3326 struct btrfs_delayed_extent_op *extent_op)
3327 {
3328 struct btrfs_key key;
3329 struct btrfs_path *path;
3330 struct btrfs_fs_info *info = root->fs_info;
3331 struct btrfs_root *extent_root = info->extent_root;
3332 struct extent_buffer *leaf;
3333 struct btrfs_extent_item *ei;
3334 struct btrfs_extent_inline_ref *iref;
3335 int ret;
3336 int is_data;
3337 int extent_slot = 0;
3338 int found_extent = 0;
3339 int num_to_del = 1;
3340 u32 item_size;
3341 u64 refs;
3342
3343 path = btrfs_alloc_path();
3344 if (!path)
3345 return -ENOMEM;
3346
3347 path->reada = 1;
3348 path->leave_spinning = 1;
3349
3350 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
3351 BUG_ON(!is_data && refs_to_drop != 1);
3352
3353 ret = lookup_extent_backref(trans, extent_root, path, &iref,
3354 bytenr, num_bytes, parent,
3355 root_objectid, owner_objectid,
3356 owner_offset);
3357 if (ret == 0) {
3358 extent_slot = path->slots[0];
3359 while (extent_slot >= 0) {
3360 btrfs_item_key_to_cpu(path->nodes[0], &key,
3361 extent_slot);
3362 if (key.objectid != bytenr)
3363 break;
3364 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
3365 key.offset == num_bytes) {
3366 found_extent = 1;
3367 break;
3368 }
3369 if (path->slots[0] - extent_slot > 5)
3370 break;
3371 extent_slot--;
3372 }
3373 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3374 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
3375 if (found_extent && item_size < sizeof(*ei))
3376 found_extent = 0;
3377 #endif
3378 if (!found_extent) {
3379 BUG_ON(iref);
3380 ret = remove_extent_backref(trans, extent_root, path,
3381 NULL, refs_to_drop,
3382 is_data);
3383 BUG_ON(ret);
3384 btrfs_release_path(extent_root, path);
3385 path->leave_spinning = 1;
3386
3387 key.objectid = bytenr;
3388 key.type = BTRFS_EXTENT_ITEM_KEY;
3389 key.offset = num_bytes;
3390
3391 ret = btrfs_search_slot(trans, extent_root,
3392 &key, path, -1, 1);
3393 if (ret) {
3394 printk(KERN_ERR "umm, got %d back from search"
3395 ", was looking for %llu\n", ret,
3396 (unsigned long long)bytenr);
3397 btrfs_print_leaf(extent_root, path->nodes[0]);
3398 }
3399 BUG_ON(ret);
3400 extent_slot = path->slots[0];
3401 }
3402 } else {
3403 btrfs_print_leaf(extent_root, path->nodes[0]);
3404 WARN_ON(1);
3405 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
3406 "parent %llu root %llu owner %llu offset %llu\n",
3407 (unsigned long long)bytenr,
3408 (unsigned long long)parent,
3409 (unsigned long long)root_objectid,
3410 (unsigned long long)owner_objectid,
3411 (unsigned long long)owner_offset);
3412 }
3413
3414 leaf = path->nodes[0];
3415 item_size = btrfs_item_size_nr(leaf, extent_slot);
3416 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3417 if (item_size < sizeof(*ei)) {
3418 BUG_ON(found_extent || extent_slot != path->slots[0]);
3419 ret = convert_extent_item_v0(trans, extent_root, path,
3420 owner_objectid, 0);
3421 BUG_ON(ret < 0);
3422
3423 btrfs_release_path(extent_root, path);
3424 path->leave_spinning = 1;
3425
3426 key.objectid = bytenr;
3427 key.type = BTRFS_EXTENT_ITEM_KEY;
3428 key.offset = num_bytes;
3429
3430 ret = btrfs_search_slot(trans, extent_root, &key, path,
3431 -1, 1);
3432 if (ret) {
3433 printk(KERN_ERR "umm, got %d back from search"
3434 ", was looking for %llu\n", ret,
3435 (unsigned long long)bytenr);
3436 btrfs_print_leaf(extent_root, path->nodes[0]);
3437 }
3438 BUG_ON(ret);
3439 extent_slot = path->slots[0];
3440 leaf = path->nodes[0];
3441 item_size = btrfs_item_size_nr(leaf, extent_slot);
3442 }
3443 #endif
3444 BUG_ON(item_size < sizeof(*ei));
3445 ei = btrfs_item_ptr(leaf, extent_slot,
3446 struct btrfs_extent_item);
3447 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
3448 struct btrfs_tree_block_info *bi;
3449 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
3450 bi = (struct btrfs_tree_block_info *)(ei + 1);
3451 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
3452 }
3453
3454 refs = btrfs_extent_refs(leaf, ei);
3455 BUG_ON(refs < refs_to_drop);
3456 refs -= refs_to_drop;
3457
3458 if (refs > 0) {
3459 if (extent_op)
3460 __run_delayed_extent_op(extent_op, leaf, ei);
3461 /*
3462 * In the case of inline back ref, reference count will
3463 * be updated by remove_extent_backref
3464 */
3465 if (iref) {
3466 BUG_ON(!found_extent);
3467 } else {
3468 btrfs_set_extent_refs(leaf, ei, refs);
3469 btrfs_mark_buffer_dirty(leaf);
3470 }
3471 if (found_extent) {
3472 ret = remove_extent_backref(trans, extent_root, path,
3473 iref, refs_to_drop,
3474 is_data);
3475 BUG_ON(ret);
3476 }
3477 } else {
3478 int mark_free = 0;
3479 struct extent_buffer *must_clean = NULL;
3480
3481 if (found_extent) {
3482 BUG_ON(is_data && refs_to_drop !=
3483 extent_data_ref_count(root, path, iref));
3484 if (iref) {
3485 BUG_ON(path->slots[0] != extent_slot);
3486 } else {
3487 BUG_ON(path->slots[0] != extent_slot + 1);
3488 path->slots[0] = extent_slot;
3489 num_to_del = 2;
3490 }
3491 }
3492
3493 ret = pin_down_bytes(trans, root, path, bytenr,
3494 num_bytes, is_data, 0, &must_clean);
3495 if (ret > 0)
3496 mark_free = 1;
3497 BUG_ON(ret < 0);
3498 /*
3499 * it is going to be very rare for someone to be waiting
3500 * on the block we're freeing. del_items might need to
3501 * schedule, so rather than get fancy, just force it
3502 * to blocking here
3503 */
3504 if (must_clean)
3505 btrfs_set_lock_blocking(must_clean);
3506
3507 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
3508 num_to_del);
3509 BUG_ON(ret);
3510 btrfs_release_path(extent_root, path);
3511
3512 if (must_clean) {
3513 clean_tree_block(NULL, root, must_clean);
3514 btrfs_tree_unlock(must_clean);
3515 free_extent_buffer(must_clean);
3516 }
3517
3518 if (is_data) {
3519 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
3520 BUG_ON(ret);
3521 } else {
3522 invalidate_mapping_pages(info->btree_inode->i_mapping,
3523 bytenr >> PAGE_CACHE_SHIFT,
3524 (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
3525 }
3526
3527 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
3528 mark_free);
3529 BUG_ON(ret);
3530 }
3531 btrfs_free_path(path);
3532 return ret;
3533 }
3534
3535 /*
3536 * when we free an extent, it is possible (and likely) that we free the last
3537 * delayed ref for that extent as well. This searches the delayed ref tree for
3538 * a given extent, and if there are no other delayed refs to be processed, it
3539 * removes it from the tree.
3540 */
3541 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
3542 struct btrfs_root *root, u64 bytenr)
3543 {
3544 struct btrfs_delayed_ref_head *head;
3545 struct btrfs_delayed_ref_root *delayed_refs;
3546 struct btrfs_delayed_ref_node *ref;
3547 struct rb_node *node;
3548 int ret;
3549
3550 delayed_refs = &trans->transaction->delayed_refs;
3551 spin_lock(&delayed_refs->lock);
3552 head = btrfs_find_delayed_ref_head(trans, bytenr);
3553 if (!head)
3554 goto out;
3555
3556 node = rb_prev(&head->node.rb_node);
3557 if (!node)
3558 goto out;
3559
3560 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
3561
3562 /* there are still entries for this ref, we can't drop it */
3563 if (ref->bytenr == bytenr)
3564 goto out;
3565
3566 if (head->extent_op) {
3567 if (!head->must_insert_reserved)
3568 goto out;
3569 kfree(head->extent_op);
3570 head->extent_op = NULL;
3571 }
3572
3573 /*
3574 * waiting for the lock here would deadlock. If someone else has it
3575 * locked they are already in the process of dropping it anyway
3576 */
3577 if (!mutex_trylock(&head->mutex))
3578 goto out;
3579
3580 /*
3581 * at this point we have a head with no other entries. Go
3582 * ahead and process it.
3583 */
3584 head->node.in_tree = 0;
3585 rb_erase(&head->node.rb_node, &delayed_refs->root);
3586
3587 delayed_refs->num_entries--;
3588
3589 /*
3590 * we don't take a ref on the node because we're removing it from the
3591 * tree, so we just steal the ref the tree was holding.
3592 */
3593 delayed_refs->num_heads--;
3594 if (list_empty(&head->cluster))
3595 delayed_refs->num_heads_ready--;
3596
3597 list_del_init(&head->cluster);
3598 spin_unlock(&delayed_refs->lock);
3599
3600 ret = run_one_delayed_ref(trans, root->fs_info->tree_root,
3601 &head->node, head->extent_op,
3602 head->must_insert_reserved);
3603 BUG_ON(ret);
3604 btrfs_put_delayed_ref(&head->node);
3605 return 0;
3606 out:
3607 spin_unlock(&delayed_refs->lock);
3608 return 0;
3609 }
3610
3611 int btrfs_free_extent(struct btrfs_trans_handle *trans,
3612 struct btrfs_root *root,
3613 u64 bytenr, u64 num_bytes, u64 parent,
3614 u64 root_objectid, u64 owner, u64 offset)
3615 {
3616 int ret;
3617
3618 /*
3619 * tree log blocks never actually go into the extent allocation
3620 * tree, just update pinning info and exit early.
3621 */
3622 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
3623 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
3624 /* unlocks the pinned mutex */
3625 btrfs_pin_extent(root, bytenr, num_bytes, 1);
3626 ret = 0;
3627 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
3628 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
3629 parent, root_objectid, (int)owner,
3630 BTRFS_DROP_DELAYED_REF, NULL);
3631 BUG_ON(ret);
3632 ret = check_ref_cleanup(trans, root, bytenr);
3633 BUG_ON(ret);
3634 } else {
3635 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
3636 parent, root_objectid, owner,
3637 offset, BTRFS_DROP_DELAYED_REF, NULL);
3638 BUG_ON(ret);
3639 }
3640 return ret;
3641 }
3642
3643 static u64 stripe_align(struct btrfs_root *root, u64 val)
3644 {
3645 u64 mask = ((u64)root->stripesize - 1);
3646 u64 ret = (val + mask) & ~mask;
3647 return ret;
3648 }
3649
3650 /*
3651 * when we wait for progress in the block group caching, its because
3652 * our allocation attempt failed at least once. So, we must sleep
3653 * and let some progress happen before we try again.
3654 *
3655 * This function will sleep at least once waiting for new free space to
3656 * show up, and then it will check the block group free space numbers
3657 * for our min num_bytes. Another option is to have it go ahead
3658 * and look in the rbtree for a free extent of a given size, but this
3659 * is a good start.
3660 */
3661 static noinline int
3662 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
3663 u64 num_bytes)
3664 {
3665 struct btrfs_caching_control *caching_ctl;
3666 DEFINE_WAIT(wait);
3667
3668 caching_ctl = get_caching_control(cache);
3669 if (!caching_ctl)
3670 return 0;
3671
3672 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
3673 (cache->free_space >= num_bytes));
3674
3675 put_caching_control(caching_ctl);
3676 return 0;
3677 }
3678
3679 static noinline int
3680 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
3681 {
3682 struct btrfs_caching_control *caching_ctl;
3683 DEFINE_WAIT(wait);
3684
3685 caching_ctl = get_caching_control(cache);
3686 if (!caching_ctl)
3687 return 0;
3688
3689 wait_event(caching_ctl->wait, block_group_cache_done(cache));
3690
3691 put_caching_control(caching_ctl);
3692 return 0;
3693 }
3694
3695 enum btrfs_loop_type {
3696 LOOP_CACHED_ONLY = 0,
3697 LOOP_CACHING_NOWAIT = 1,
3698 LOOP_CACHING_WAIT = 2,
3699 LOOP_ALLOC_CHUNK = 3,
3700 LOOP_NO_EMPTY_SIZE = 4,
3701 };
3702
3703 /*
3704 * walks the btree of allocated extents and find a hole of a given size.
3705 * The key ins is changed to record the hole:
3706 * ins->objectid == block start
3707 * ins->flags = BTRFS_EXTENT_ITEM_KEY
3708 * ins->offset == number of blocks
3709 * Any available blocks before search_start are skipped.
3710 */
3711 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
3712 struct btrfs_root *orig_root,
3713 u64 num_bytes, u64 empty_size,
3714 u64 search_start, u64 search_end,
3715 u64 hint_byte, struct btrfs_key *ins,
3716 u64 exclude_start, u64 exclude_nr,
3717 int data)
3718 {
3719 int ret = 0;
3720 struct btrfs_root *root = orig_root->fs_info->extent_root;
3721 struct btrfs_free_cluster *last_ptr = NULL;
3722 struct btrfs_block_group_cache *block_group = NULL;
3723 int empty_cluster = 2 * 1024 * 1024;
3724 int allowed_chunk_alloc = 0;
3725 struct btrfs_space_info *space_info;
3726 int last_ptr_loop = 0;
3727 int loop = 0;
3728 bool found_uncached_bg = false;
3729
3730 WARN_ON(num_bytes < root->sectorsize);
3731 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
3732 ins->objectid = 0;
3733 ins->offset = 0;
3734
3735 space_info = __find_space_info(root->fs_info, data);
3736
3737 if (orig_root->ref_cows || empty_size)
3738 allowed_chunk_alloc = 1;
3739
3740 if (data & BTRFS_BLOCK_GROUP_METADATA) {
3741 last_ptr = &root->fs_info->meta_alloc_cluster;
3742 if (!btrfs_test_opt(root, SSD))
3743 empty_cluster = 64 * 1024;
3744 }
3745
3746 if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD)) {
3747 last_ptr = &root->fs_info->data_alloc_cluster;
3748 }
3749
3750 if (last_ptr) {
3751 spin_lock(&last_ptr->lock);
3752 if (last_ptr->block_group)
3753 hint_byte = last_ptr->window_start;
3754 spin_unlock(&last_ptr->lock);
3755 }
3756
3757 search_start = max(search_start, first_logical_byte(root, 0));
3758 search_start = max(search_start, hint_byte);
3759
3760 if (!last_ptr)
3761 empty_cluster = 0;
3762
3763 if (search_start == hint_byte) {
3764 block_group = btrfs_lookup_block_group(root->fs_info,
3765 search_start);
3766 /*
3767 * we don't want to use the block group if it doesn't match our
3768 * allocation bits, or if its not cached.
3769 */
3770 if (block_group && block_group_bits(block_group, data) &&
3771 block_group_cache_done(block_group)) {
3772 down_read(&space_info->groups_sem);
3773 if (list_empty(&block_group->list) ||
3774 block_group->ro) {
3775 /*
3776 * someone is removing this block group,
3777 * we can't jump into the have_block_group
3778 * target because our list pointers are not
3779 * valid
3780 */
3781 btrfs_put_block_group(block_group);
3782 up_read(&space_info->groups_sem);
3783 } else
3784 goto have_block_group;
3785 } else if (block_group) {
3786 btrfs_put_block_group(block_group);
3787 }
3788 }
3789
3790 search:
3791 down_read(&space_info->groups_sem);
3792 list_for_each_entry(block_group, &space_info->block_groups, list) {
3793 u64 offset;
3794 int cached;
3795
3796 atomic_inc(&block_group->count);
3797 search_start = block_group->key.objectid;
3798
3799 have_block_group:
3800 if (unlikely(block_group->cached == BTRFS_CACHE_NO)) {
3801 /*
3802 * we want to start caching kthreads, but not too many
3803 * right off the bat so we don't overwhelm the system,
3804 * so only start them if there are less than 2 and we're
3805 * in the initial allocation phase.
3806 */
3807 if (loop > LOOP_CACHING_NOWAIT ||
3808 atomic_read(&space_info->caching_threads) < 2) {
3809 ret = cache_block_group(block_group);
3810 BUG_ON(ret);
3811 }
3812 }
3813
3814 cached = block_group_cache_done(block_group);
3815 if (unlikely(!cached)) {
3816 found_uncached_bg = true;
3817
3818 /* if we only want cached bgs, loop */
3819 if (loop == LOOP_CACHED_ONLY)
3820 goto loop;
3821 }
3822
3823 if (unlikely(block_group->ro))
3824 goto loop;
3825
3826 if (last_ptr) {
3827 /*
3828 * the refill lock keeps out other
3829 * people trying to start a new cluster
3830 */
3831 spin_lock(&last_ptr->refill_lock);
3832 if (last_ptr->block_group &&
3833 (last_ptr->block_group->ro ||
3834 !block_group_bits(last_ptr->block_group, data))) {
3835 offset = 0;
3836 goto refill_cluster;
3837 }
3838
3839 offset = btrfs_alloc_from_cluster(block_group, last_ptr,
3840 num_bytes, search_start);
3841 if (offset) {
3842 /* we have a block, we're done */
3843 spin_unlock(&last_ptr->refill_lock);
3844 goto checks;
3845 }
3846
3847 spin_lock(&last_ptr->lock);
3848 /*
3849 * whoops, this cluster doesn't actually point to
3850 * this block group. Get a ref on the block
3851 * group is does point to and try again
3852 */
3853 if (!last_ptr_loop && last_ptr->block_group &&
3854 last_ptr->block_group != block_group) {
3855
3856 btrfs_put_block_group(block_group);
3857 block_group = last_ptr->block_group;
3858 atomic_inc(&block_group->count);
3859 spin_unlock(&last_ptr->lock);
3860 spin_unlock(&last_ptr->refill_lock);
3861
3862 last_ptr_loop = 1;
3863 search_start = block_group->key.objectid;
3864 /*
3865 * we know this block group is properly
3866 * in the list because
3867 * btrfs_remove_block_group, drops the
3868 * cluster before it removes the block
3869 * group from the list
3870 */
3871 goto have_block_group;
3872 }
3873 spin_unlock(&last_ptr->lock);
3874 refill_cluster:
3875 /*
3876 * this cluster didn't work out, free it and
3877 * start over
3878 */
3879 btrfs_return_cluster_to_free_space(NULL, last_ptr);
3880
3881 last_ptr_loop = 0;
3882
3883 /* allocate a cluster in this block group */
3884 ret = btrfs_find_space_cluster(trans, root,
3885 block_group, last_ptr,
3886 offset, num_bytes,
3887 empty_cluster + empty_size);
3888 if (ret == 0) {
3889 /*
3890 * now pull our allocation out of this
3891 * cluster
3892 */
3893 offset = btrfs_alloc_from_cluster(block_group,
3894 last_ptr, num_bytes,
3895 search_start);
3896 if (offset) {
3897 /* we found one, proceed */
3898 spin_unlock(&last_ptr->refill_lock);
3899 goto checks;
3900 }
3901 } else if (!cached && loop > LOOP_CACHING_NOWAIT) {
3902 spin_unlock(&last_ptr->refill_lock);
3903
3904 wait_block_group_cache_progress(block_group,
3905 num_bytes + empty_cluster + empty_size);
3906 goto have_block_group;
3907 }
3908
3909 /*
3910 * at this point we either didn't find a cluster
3911 * or we weren't able to allocate a block from our
3912 * cluster. Free the cluster we've been trying
3913 * to use, and go to the next block group
3914 */
3915 if (loop < LOOP_NO_EMPTY_SIZE) {
3916 btrfs_return_cluster_to_free_space(NULL,
3917 last_ptr);
3918 spin_unlock(&last_ptr->refill_lock);
3919 goto loop;
3920 }
3921 spin_unlock(&last_ptr->refill_lock);
3922 }
3923
3924 offset = btrfs_find_space_for_alloc(block_group, search_start,
3925 num_bytes, empty_size);
3926 if (!offset && (cached || (!cached &&
3927 loop == LOOP_CACHING_NOWAIT))) {
3928 goto loop;
3929 } else if (!offset && (!cached &&
3930 loop > LOOP_CACHING_NOWAIT)) {
3931 wait_block_group_cache_progress(block_group,
3932 num_bytes + empty_size);
3933 goto have_block_group;
3934 }
3935 checks:
3936 search_start = stripe_align(root, offset);
3937 /* move on to the next group */
3938 if (search_start + num_bytes >= search_end) {
3939 btrfs_add_free_space(block_group, offset, num_bytes);
3940 goto loop;
3941 }
3942
3943 /* move on to the next group */
3944 if (search_start + num_bytes >
3945 block_group->key.objectid + block_group->key.offset) {
3946 btrfs_add_free_space(block_group, offset, num_bytes);
3947 goto loop;
3948 }
3949
3950 if (exclude_nr > 0 &&
3951 (search_start + num_bytes > exclude_start &&
3952 search_start < exclude_start + exclude_nr)) {
3953 search_start = exclude_start + exclude_nr;
3954
3955 btrfs_add_free_space(block_group, offset, num_bytes);
3956 /*
3957 * if search_start is still in this block group
3958 * then we just re-search this block group
3959 */
3960 if (search_start >= block_group->key.objectid &&
3961 search_start < (block_group->key.objectid +
3962 block_group->key.offset))
3963 goto have_block_group;
3964 goto loop;
3965 }
3966
3967 ins->objectid = search_start;
3968 ins->offset = num_bytes;
3969
3970 if (offset < search_start)
3971 btrfs_add_free_space(block_group, offset,
3972 search_start - offset);
3973 BUG_ON(offset > search_start);
3974
3975 update_reserved_extents(block_group, num_bytes, 1);
3976
3977 /* we are all good, lets return */
3978 break;
3979 loop:
3980 btrfs_put_block_group(block_group);
3981 }
3982 up_read(&space_info->groups_sem);
3983
3984 /* LOOP_CACHED_ONLY, only search fully cached block groups
3985 * LOOP_CACHING_NOWAIT, search partially cached block groups, but
3986 * dont wait foR them to finish caching
3987 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
3988 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
3989 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
3990 * again
3991 */
3992 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE &&
3993 (found_uncached_bg || empty_size || empty_cluster ||
3994 allowed_chunk_alloc)) {
3995 if (found_uncached_bg) {
3996 found_uncached_bg = false;
3997 if (loop < LOOP_CACHING_WAIT) {
3998 loop++;
3999 goto search;
4000 }
4001 }
4002
4003 if (loop == LOOP_ALLOC_CHUNK) {
4004 empty_size = 0;
4005 empty_cluster = 0;
4006 }
4007
4008 if (allowed_chunk_alloc) {
4009 ret = do_chunk_alloc(trans, root, num_bytes +
4010 2 * 1024 * 1024, data, 1);
4011 allowed_chunk_alloc = 0;
4012 } else {
4013 space_info->force_alloc = 1;
4014 }
4015
4016 if (loop < LOOP_NO_EMPTY_SIZE) {
4017 loop++;
4018 goto search;
4019 }
4020 ret = -ENOSPC;
4021 } else if (!ins->objectid) {
4022 ret = -ENOSPC;
4023 }
4024
4025 /* we found what we needed */
4026 if (ins->objectid) {
4027 if (!(data & BTRFS_BLOCK_GROUP_DATA))
4028 trans->block_group = block_group->key.objectid;
4029
4030 btrfs_put_block_group(block_group);
4031 ret = 0;
4032 }
4033
4034 return ret;
4035 }
4036
4037 static void dump_space_info(struct btrfs_space_info *info, u64 bytes)
4038 {
4039 struct btrfs_block_group_cache *cache;
4040
4041 printk(KERN_INFO "space_info has %llu free, is %sfull\n",
4042 (unsigned long long)(info->total_bytes - info->bytes_used -
4043 info->bytes_pinned - info->bytes_reserved),
4044 (info->full) ? "" : "not ");
4045 printk(KERN_INFO "space_info total=%llu, pinned=%llu, delalloc=%llu,"
4046 " may_use=%llu, used=%llu\n",
4047 (unsigned long long)info->total_bytes,
4048 (unsigned long long)info->bytes_pinned,
4049 (unsigned long long)info->bytes_delalloc,
4050 (unsigned long long)info->bytes_may_use,
4051 (unsigned long long)info->bytes_used);
4052
4053 down_read(&info->groups_sem);
4054 list_for_each_entry(cache, &info->block_groups, list) {
4055 spin_lock(&cache->lock);
4056 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
4057 "%llu pinned %llu reserved\n",
4058 (unsigned long long)cache->key.objectid,
4059 (unsigned long long)cache->key.offset,
4060 (unsigned long long)btrfs_block_group_used(&cache->item),
4061 (unsigned long long)cache->pinned,
4062 (unsigned long long)cache->reserved);
4063 btrfs_dump_free_space(cache, bytes);
4064 spin_unlock(&cache->lock);
4065 }
4066 up_read(&info->groups_sem);
4067 }
4068
4069 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
4070 struct btrfs_root *root,
4071 u64 num_bytes, u64 min_alloc_size,
4072 u64 empty_size, u64 hint_byte,
4073 u64 search_end, struct btrfs_key *ins,
4074 u64 data)
4075 {
4076 int ret;
4077 u64 search_start = 0;
4078 struct btrfs_fs_info *info = root->fs_info;
4079
4080 data = btrfs_get_alloc_profile(root, data);
4081 again:
4082 /*
4083 * the only place that sets empty_size is btrfs_realloc_node, which
4084 * is not called recursively on allocations
4085 */
4086 if (empty_size || root->ref_cows) {
4087 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
4088 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4089 2 * 1024 * 1024,
4090 BTRFS_BLOCK_GROUP_METADATA |
4091 (info->metadata_alloc_profile &
4092 info->avail_metadata_alloc_bits), 0);
4093 }
4094 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4095 num_bytes + 2 * 1024 * 1024, data, 0);
4096 }
4097
4098 WARN_ON(num_bytes < root->sectorsize);
4099 ret = find_free_extent(trans, root, num_bytes, empty_size,
4100 search_start, search_end, hint_byte, ins,
4101 trans->alloc_exclude_start,
4102 trans->alloc_exclude_nr, data);
4103
4104 if (ret == -ENOSPC && num_bytes > min_alloc_size) {
4105 num_bytes = num_bytes >> 1;
4106 num_bytes = num_bytes & ~(root->sectorsize - 1);
4107 num_bytes = max(num_bytes, min_alloc_size);
4108 do_chunk_alloc(trans, root->fs_info->extent_root,
4109 num_bytes, data, 1);
4110 goto again;
4111 }
4112 if (ret == -ENOSPC) {
4113 struct btrfs_space_info *sinfo;
4114
4115 sinfo = __find_space_info(root->fs_info, data);
4116 printk(KERN_ERR "btrfs allocation failed flags %llu, "
4117 "wanted %llu\n", (unsigned long long)data,
4118 (unsigned long long)num_bytes);
4119 dump_space_info(sinfo, num_bytes);
4120 }
4121
4122 return ret;
4123 }
4124
4125 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
4126 {
4127 struct btrfs_block_group_cache *cache;
4128 int ret = 0;
4129
4130 cache = btrfs_lookup_block_group(root->fs_info, start);
4131 if (!cache) {
4132 printk(KERN_ERR "Unable to find block group for %llu\n",
4133 (unsigned long long)start);
4134 return -ENOSPC;
4135 }
4136
4137 ret = btrfs_discard_extent(root, start, len);
4138
4139 btrfs_add_free_space(cache, start, len);
4140 update_reserved_extents(cache, len, 0);
4141 btrfs_put_block_group(cache);
4142
4143 return ret;
4144 }
4145
4146 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
4147 struct btrfs_root *root,
4148 u64 parent, u64 root_objectid,
4149 u64 flags, u64 owner, u64 offset,
4150 struct btrfs_key *ins, int ref_mod)
4151 {
4152 int ret;
4153 struct btrfs_fs_info *fs_info = root->fs_info;
4154 struct btrfs_extent_item *extent_item;
4155 struct btrfs_extent_inline_ref *iref;
4156 struct btrfs_path *path;
4157 struct extent_buffer *leaf;
4158 int type;
4159 u32 size;
4160
4161 if (parent > 0)
4162 type = BTRFS_SHARED_DATA_REF_KEY;
4163 else
4164 type = BTRFS_EXTENT_DATA_REF_KEY;
4165
4166 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
4167
4168 path = btrfs_alloc_path();
4169 BUG_ON(!path);
4170
4171 path->leave_spinning = 1;
4172 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
4173 ins, size);
4174 BUG_ON(ret);
4175
4176 leaf = path->nodes[0];
4177 extent_item = btrfs_item_ptr(leaf, path->slots[0],
4178 struct btrfs_extent_item);
4179 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
4180 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
4181 btrfs_set_extent_flags(leaf, extent_item,
4182 flags | BTRFS_EXTENT_FLAG_DATA);
4183
4184 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
4185 btrfs_set_extent_inline_ref_type(leaf, iref, type);
4186 if (parent > 0) {
4187 struct btrfs_shared_data_ref *ref;
4188 ref = (struct btrfs_shared_data_ref *)(iref + 1);
4189 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
4190 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
4191 } else {
4192 struct btrfs_extent_data_ref *ref;
4193 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
4194 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
4195 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
4196 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
4197 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
4198 }
4199
4200 btrfs_mark_buffer_dirty(path->nodes[0]);
4201 btrfs_free_path(path);
4202
4203 ret = update_block_group(trans, root, ins->objectid, ins->offset,
4204 1, 0);
4205 if (ret) {
4206 printk(KERN_ERR "btrfs update block group failed for %llu "
4207 "%llu\n", (unsigned long long)ins->objectid,
4208 (unsigned long long)ins->offset);
4209 BUG();
4210 }
4211 return ret;
4212 }
4213
4214 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
4215 struct btrfs_root *root,
4216 u64 parent, u64 root_objectid,
4217 u64 flags, struct btrfs_disk_key *key,
4218 int level, struct btrfs_key *ins)
4219 {
4220 int ret;
4221 struct btrfs_fs_info *fs_info = root->fs_info;
4222 struct btrfs_extent_item *extent_item;
4223 struct btrfs_tree_block_info *block_info;
4224 struct btrfs_extent_inline_ref *iref;
4225 struct btrfs_path *path;
4226 struct extent_buffer *leaf;
4227 u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
4228
4229 path = btrfs_alloc_path();
4230 BUG_ON(!path);
4231
4232 path->leave_spinning = 1;
4233 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
4234 ins, size);
4235 BUG_ON(ret);
4236
4237 leaf = path->nodes[0];
4238 extent_item = btrfs_item_ptr(leaf, path->slots[0],
4239 struct btrfs_extent_item);
4240 btrfs_set_extent_refs(leaf, extent_item, 1);
4241 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
4242 btrfs_set_extent_flags(leaf, extent_item,
4243 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
4244 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
4245
4246 btrfs_set_tree_block_key(leaf, block_info, key);
4247 btrfs_set_tree_block_level(leaf, block_info, level);
4248
4249 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
4250 if (parent > 0) {
4251 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
4252 btrfs_set_extent_inline_ref_type(leaf, iref,
4253 BTRFS_SHARED_BLOCK_REF_KEY);
4254 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
4255 } else {
4256 btrfs_set_extent_inline_ref_type(leaf, iref,
4257 BTRFS_TREE_BLOCK_REF_KEY);
4258 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
4259 }
4260
4261 btrfs_mark_buffer_dirty(leaf);
4262 btrfs_free_path(path);
4263
4264 ret = update_block_group(trans, root, ins->objectid, ins->offset,
4265 1, 0);
4266 if (ret) {
4267 printk(KERN_ERR "btrfs update block group failed for %llu "
4268 "%llu\n", (unsigned long long)ins->objectid,
4269 (unsigned long long)ins->offset);
4270 BUG();
4271 }
4272 return ret;
4273 }
4274
4275 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
4276 struct btrfs_root *root,
4277 u64 root_objectid, u64 owner,
4278 u64 offset, struct btrfs_key *ins)
4279 {
4280 int ret;
4281
4282 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
4283
4284 ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset,
4285 0, root_objectid, owner, offset,
4286 BTRFS_ADD_DELAYED_EXTENT, NULL);
4287 return ret;
4288 }
4289
4290 /*
4291 * this is used by the tree logging recovery code. It records that
4292 * an extent has been allocated and makes sure to clear the free
4293 * space cache bits as well
4294 */
4295 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
4296 struct btrfs_root *root,
4297 u64 root_objectid, u64 owner, u64 offset,
4298 struct btrfs_key *ins)
4299 {
4300 int ret;
4301 struct btrfs_block_group_cache *block_group;
4302 struct btrfs_caching_control *caching_ctl;
4303 u64 start = ins->objectid;
4304 u64 num_bytes = ins->offset;
4305
4306 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
4307 cache_block_group(block_group);
4308 caching_ctl = get_caching_control(block_group);
4309
4310 if (!caching_ctl) {
4311 BUG_ON(!block_group_cache_done(block_group));
4312 ret = btrfs_remove_free_space(block_group, start, num_bytes);
4313 BUG_ON(ret);
4314 } else {
4315 mutex_lock(&caching_ctl->mutex);
4316
4317 if (start >= caching_ctl->progress) {
4318 ret = add_excluded_extent(root, start, num_bytes);
4319 BUG_ON(ret);
4320 } else if (start + num_bytes <= caching_ctl->progress) {
4321 ret = btrfs_remove_free_space(block_group,
4322 start, num_bytes);
4323 BUG_ON(ret);
4324 } else {
4325 num_bytes = caching_ctl->progress - start;
4326 ret = btrfs_remove_free_space(block_group,
4327 start, num_bytes);
4328 BUG_ON(ret);
4329
4330 start = caching_ctl->progress;
4331 num_bytes = ins->objectid + ins->offset -
4332 caching_ctl->progress;
4333 ret = add_excluded_extent(root, start, num_bytes);
4334 BUG_ON(ret);
4335 }
4336
4337 mutex_unlock(&caching_ctl->mutex);
4338 put_caching_control(caching_ctl);
4339 }
4340
4341 update_reserved_extents(block_group, ins->offset, 1);
4342 btrfs_put_block_group(block_group);
4343 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
4344 0, owner, offset, ins, 1);
4345 return ret;
4346 }
4347
4348 /*
4349 * finds a free extent and does all the dirty work required for allocation
4350 * returns the key for the extent through ins, and a tree buffer for
4351 * the first block of the extent through buf.
4352 *
4353 * returns 0 if everything worked, non-zero otherwise.
4354 */
4355 static int alloc_tree_block(struct btrfs_trans_handle *trans,
4356 struct btrfs_root *root,
4357 u64 num_bytes, u64 parent, u64 root_objectid,
4358 struct btrfs_disk_key *key, int level,
4359 u64 empty_size, u64 hint_byte, u64 search_end,
4360 struct btrfs_key *ins)
4361 {
4362 int ret;
4363 u64 flags = 0;
4364
4365 ret = btrfs_reserve_extent(trans, root, num_bytes, num_bytes,
4366 empty_size, hint_byte, search_end,
4367 ins, 0);
4368 if (ret)
4369 return ret;
4370
4371 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
4372 if (parent == 0)
4373 parent = ins->objectid;
4374 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
4375 } else
4376 BUG_ON(parent > 0);
4377
4378 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
4379 struct btrfs_delayed_extent_op *extent_op;
4380 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
4381 BUG_ON(!extent_op);
4382 if (key)
4383 memcpy(&extent_op->key, key, sizeof(extent_op->key));
4384 else
4385 memset(&extent_op->key, 0, sizeof(extent_op->key));
4386 extent_op->flags_to_set = flags;
4387 extent_op->update_key = 1;
4388 extent_op->update_flags = 1;
4389 extent_op->is_data = 0;
4390
4391 ret = btrfs_add_delayed_tree_ref(trans, ins->objectid,
4392 ins->offset, parent, root_objectid,
4393 level, BTRFS_ADD_DELAYED_EXTENT,
4394 extent_op);
4395 BUG_ON(ret);
4396 }
4397 return ret;
4398 }
4399
4400 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
4401 struct btrfs_root *root,
4402 u64 bytenr, u32 blocksize,
4403 int level)
4404 {
4405 struct extent_buffer *buf;
4406
4407 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
4408 if (!buf)
4409 return ERR_PTR(-ENOMEM);
4410 btrfs_set_header_generation(buf, trans->transid);
4411 btrfs_set_buffer_lockdep_class(buf, level);
4412 btrfs_tree_lock(buf);
4413 clean_tree_block(trans, root, buf);
4414
4415 btrfs_set_lock_blocking(buf);
4416 btrfs_set_buffer_uptodate(buf);
4417
4418 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
4419 set_extent_dirty(&root->dirty_log_pages, buf->start,
4420 buf->start + buf->len - 1, GFP_NOFS);
4421 } else {
4422 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
4423 buf->start + buf->len - 1, GFP_NOFS);
4424 }
4425 trans->blocks_used++;
4426 /* this returns a buffer locked for blocking */
4427 return buf;
4428 }
4429
4430 /*
4431 * helper function to allocate a block for a given tree
4432 * returns the tree buffer or NULL.
4433 */
4434 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
4435 struct btrfs_root *root, u32 blocksize,
4436 u64 parent, u64 root_objectid,
4437 struct btrfs_disk_key *key, int level,
4438 u64 hint, u64 empty_size)
4439 {
4440 struct btrfs_key ins;
4441 int ret;
4442 struct extent_buffer *buf;
4443
4444 ret = alloc_tree_block(trans, root, blocksize, parent, root_objectid,
4445 key, level, empty_size, hint, (u64)-1, &ins);
4446 if (ret) {
4447 BUG_ON(ret > 0);
4448 return ERR_PTR(ret);
4449 }
4450
4451 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
4452 blocksize, level);
4453 return buf;
4454 }
4455
4456 #if 0
4457 int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans,
4458 struct btrfs_root *root, struct extent_buffer *leaf)
4459 {
4460 u64 disk_bytenr;
4461 u64 num_bytes;
4462 struct btrfs_key key;
4463 struct btrfs_file_extent_item *fi;
4464 u32 nritems;
4465 int i;
4466 int ret;
4467
4468 BUG_ON(!btrfs_is_leaf(leaf));
4469 nritems = btrfs_header_nritems(leaf);
4470
4471 for (i = 0; i < nritems; i++) {
4472 cond_resched();
4473 btrfs_item_key_to_cpu(leaf, &key, i);
4474
4475 /* only extents have references, skip everything else */
4476 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
4477 continue;
4478
4479 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4480
4481 /* inline extents live in the btree, they don't have refs */
4482 if (btrfs_file_extent_type(leaf, fi) ==
4483 BTRFS_FILE_EXTENT_INLINE)
4484 continue;
4485
4486 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
4487
4488 /* holes don't have refs */
4489 if (disk_bytenr == 0)
4490 continue;
4491
4492 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
4493 ret = btrfs_free_extent(trans, root, disk_bytenr, num_bytes,
4494 leaf->start, 0, key.objectid, 0);
4495 BUG_ON(ret);
4496 }
4497 return 0;
4498 }
4499
4500 static noinline int cache_drop_leaf_ref(struct btrfs_trans_handle *trans,
4501 struct btrfs_root *root,
4502 struct btrfs_leaf_ref *ref)
4503 {
4504 int i;
4505 int ret;
4506 struct btrfs_extent_info *info;
4507 struct refsort *sorted;
4508
4509 if (ref->nritems == 0)
4510 return 0;
4511
4512 sorted = kmalloc(sizeof(*sorted) * ref->nritems, GFP_NOFS);
4513 for (i = 0; i < ref->nritems; i++) {
4514 sorted[i].bytenr = ref->extents[i].bytenr;
4515 sorted[i].slot = i;
4516 }
4517 sort(sorted, ref->nritems, sizeof(struct refsort), refsort_cmp, NULL);
4518
4519 /*
4520 * the items in the ref were sorted when the ref was inserted
4521 * into the ref cache, so this is already in order
4522 */
4523 for (i = 0; i < ref->nritems; i++) {
4524 info = ref->extents + sorted[i].slot;
4525 ret = btrfs_free_extent(trans, root, info->bytenr,
4526 info->num_bytes, ref->bytenr,
4527 ref->owner, ref->generation,
4528 info->objectid, 0);
4529
4530 atomic_inc(&root->fs_info->throttle_gen);
4531 wake_up(&root->fs_info->transaction_throttle);
4532 cond_resched();
4533
4534 BUG_ON(ret);
4535 info++;
4536 }
4537
4538 kfree(sorted);
4539 return 0;
4540 }
4541
4542
4543 static int drop_snap_lookup_refcount(struct btrfs_trans_handle *trans,
4544 struct btrfs_root *root, u64 start,
4545 u64 len, u32 *refs)
4546 {
4547 int ret;
4548
4549 ret = btrfs_lookup_extent_refs(trans, root, start, len, refs);
4550 BUG_ON(ret);
4551
4552 #if 0 /* some debugging code in case we see problems here */
4553 /* if the refs count is one, it won't get increased again. But
4554 * if the ref count is > 1, someone may be decreasing it at
4555 * the same time we are.
4556 */
4557 if (*refs != 1) {
4558 struct extent_buffer *eb = NULL;
4559 eb = btrfs_find_create_tree_block(root, start, len);
4560 if (eb)
4561 btrfs_tree_lock(eb);
4562
4563 mutex_lock(&root->fs_info->alloc_mutex);
4564 ret = lookup_extent_ref(NULL, root, start, len, refs);
4565 BUG_ON(ret);
4566 mutex_unlock(&root->fs_info->alloc_mutex);
4567
4568 if (eb) {
4569 btrfs_tree_unlock(eb);
4570 free_extent_buffer(eb);
4571 }
4572 if (*refs == 1) {
4573 printk(KERN_ERR "btrfs block %llu went down to one "
4574 "during drop_snap\n", (unsigned long long)start);
4575 }
4576
4577 }
4578 #endif
4579
4580 cond_resched();
4581 return ret;
4582 }
4583
4584
4585 /*
4586 * this is used while deleting old snapshots, and it drops the refs
4587 * on a whole subtree starting from a level 1 node.
4588 *
4589 * The idea is to sort all the leaf pointers, and then drop the
4590 * ref on all the leaves in order. Most of the time the leaves
4591 * will have ref cache entries, so no leaf IOs will be required to
4592 * find the extents they have references on.
4593 *
4594 * For each leaf, any references it has are also dropped in order
4595 *
4596 * This ends up dropping the references in something close to optimal
4597 * order for reading and modifying the extent allocation tree.
4598 */
4599 static noinline int drop_level_one_refs(struct btrfs_trans_handle *trans,
4600 struct btrfs_root *root,
4601 struct btrfs_path *path)
4602 {
4603 u64 bytenr;
4604 u64 root_owner;
4605 u64 root_gen;
4606 struct extent_buffer *eb = path->nodes[1];
4607 struct extent_buffer *leaf;
4608 struct btrfs_leaf_ref *ref;
4609 struct refsort *sorted = NULL;
4610 int nritems = btrfs_header_nritems(eb);
4611 int ret;
4612 int i;
4613 int refi = 0;
4614 int slot = path->slots[1];
4615 u32 blocksize = btrfs_level_size(root, 0);
4616 u32 refs;
4617
4618 if (nritems == 0)
4619 goto out;
4620
4621 root_owner = btrfs_header_owner(eb);
4622 root_gen = btrfs_header_generation(eb);
4623 sorted = kmalloc(sizeof(*sorted) * nritems, GFP_NOFS);
4624
4625 /*
4626 * step one, sort all the leaf pointers so we don't scribble
4627 * randomly into the extent allocation tree
4628 */
4629 for (i = slot; i < nritems; i++) {
4630 sorted[refi].bytenr = btrfs_node_blockptr(eb, i);
4631 sorted[refi].slot = i;
4632 refi++;
4633 }
4634
4635 /*
4636 * nritems won't be zero, but if we're picking up drop_snapshot
4637 * after a crash, slot might be > 0, so double check things
4638 * just in case.
4639 */
4640 if (refi == 0)
4641 goto out;
4642
4643 sort(sorted, refi, sizeof(struct refsort), refsort_cmp, NULL);
4644
4645 /*
4646 * the first loop frees everything the leaves point to
4647 */
4648 for (i = 0; i < refi; i++) {
4649 u64 ptr_gen;
4650
4651 bytenr = sorted[i].bytenr;
4652
4653 /*
4654 * check the reference count on this leaf. If it is > 1
4655 * we just decrement it below and don't update any
4656 * of the refs the leaf points to.
4657 */
4658 ret = drop_snap_lookup_refcount(trans, root, bytenr,
4659 blocksize, &refs);
4660 BUG_ON(ret);
4661 if (refs != 1)
4662 continue;
4663
4664 ptr_gen = btrfs_node_ptr_generation(eb, sorted[i].slot);
4665
4666 /*
4667 * the leaf only had one reference, which means the
4668 * only thing pointing to this leaf is the snapshot
4669 * we're deleting. It isn't possible for the reference
4670 * count to increase again later
4671 *
4672 * The reference cache is checked for the leaf,
4673 * and if found we'll be able to drop any refs held by
4674 * the leaf without needing to read it in.
4675 */
4676 ref = btrfs_lookup_leaf_ref(root, bytenr);
4677 if (ref && ref->generation != ptr_gen) {
4678 btrfs_free_leaf_ref(root, ref);
4679 ref = NULL;
4680 }
4681 if (ref) {
4682 ret = cache_drop_leaf_ref(trans, root, ref);
4683 BUG_ON(ret);
4684 btrfs_remove_leaf_ref(root, ref);
4685 btrfs_free_leaf_ref(root, ref);
4686 } else {
4687 /*
4688 * the leaf wasn't in the reference cache, so
4689 * we have to read it.
4690 */
4691 leaf = read_tree_block(root, bytenr, blocksize,
4692 ptr_gen);
4693 ret = btrfs_drop_leaf_ref(trans, root, leaf);
4694 BUG_ON(ret);
4695 free_extent_buffer(leaf);
4696 }
4697 atomic_inc(&root->fs_info->throttle_gen);
4698 wake_up(&root->fs_info->transaction_throttle);
4699 cond_resched();
4700 }
4701
4702 /*
4703 * run through the loop again to free the refs on the leaves.
4704 * This is faster than doing it in the loop above because
4705 * the leaves are likely to be clustered together. We end up
4706 * working in nice chunks on the extent allocation tree.
4707 */
4708 for (i = 0; i < refi; i++) {
4709 bytenr = sorted[i].bytenr;
4710 ret = btrfs_free_extent(trans, root, bytenr,
4711 blocksize, eb->start,
4712 root_owner, root_gen, 0, 1);
4713 BUG_ON(ret);
4714
4715 atomic_inc(&root->fs_info->throttle_gen);
4716 wake_up(&root->fs_info->transaction_throttle);
4717 cond_resched();
4718 }
4719 out:
4720 kfree(sorted);
4721
4722 /*
4723 * update the path to show we've processed the entire level 1
4724 * node. This will get saved into the root's drop_snapshot_progress
4725 * field so these drops are not repeated again if this transaction
4726 * commits.
4727 */
4728 path->slots[1] = nritems;
4729 return 0;
4730 }
4731
4732 /*
4733 * helper function for drop_snapshot, this walks down the tree dropping ref
4734 * counts as it goes.
4735 */
4736 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
4737 struct btrfs_root *root,
4738 struct btrfs_path *path, int *level)
4739 {
4740 u64 root_owner;
4741 u64 root_gen;
4742 u64 bytenr;
4743 u64 ptr_gen;
4744 struct extent_buffer *next;
4745 struct extent_buffer *cur;
4746 struct extent_buffer *parent;
4747 u32 blocksize;
4748 int ret;
4749 u32 refs;
4750
4751 WARN_ON(*level < 0);
4752 WARN_ON(*level >= BTRFS_MAX_LEVEL);
4753 ret = drop_snap_lookup_refcount(trans, root, path->nodes[*level]->start,
4754 path->nodes[*level]->len, &refs);
4755 BUG_ON(ret);
4756 if (refs > 1)
4757 goto out;
4758
4759 /*
4760 * walk down to the last node level and free all the leaves
4761 */
4762 while (*level >= 0) {
4763 WARN_ON(*level < 0);
4764 WARN_ON(*level >= BTRFS_MAX_LEVEL);
4765 cur = path->nodes[*level];
4766
4767 if (btrfs_header_level(cur) != *level)
4768 WARN_ON(1);
4769
4770 if (path->slots[*level] >=
4771 btrfs_header_nritems(cur))
4772 break;
4773
4774 /* the new code goes down to level 1 and does all the
4775 * leaves pointed to that node in bulk. So, this check
4776 * for level 0 will always be false.
4777 *
4778 * But, the disk format allows the drop_snapshot_progress
4779 * field in the root to leave things in a state where
4780 * a leaf will need cleaning up here. If someone crashes
4781 * with the old code and then boots with the new code,
4782 * we might find a leaf here.
4783 */
4784 if (*level == 0) {
4785 ret = btrfs_drop_leaf_ref(trans, root, cur);
4786 BUG_ON(ret);
4787 break;
4788 }
4789
4790 /*
4791 * once we get to level one, process the whole node
4792 * at once, including everything below it.
4793 */
4794 if (*level == 1) {
4795 ret = drop_level_one_refs(trans, root, path);
4796 BUG_ON(ret);
4797 break;
4798 }
4799
4800 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
4801 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
4802 blocksize = btrfs_level_size(root, *level - 1);
4803
4804 ret = drop_snap_lookup_refcount(trans, root, bytenr,
4805 blocksize, &refs);
4806 BUG_ON(ret);
4807
4808 /*
4809 * if there is more than one reference, we don't need
4810 * to read that node to drop any references it has. We
4811 * just drop the ref we hold on that node and move on to the
4812 * next slot in this level.
4813 */
4814 if (refs != 1) {
4815 parent = path->nodes[*level];
4816 root_owner = btrfs_header_owner(parent);
4817 root_gen = btrfs_header_generation(parent);
4818 path->slots[*level]++;
4819
4820 ret = btrfs_free_extent(trans, root, bytenr,
4821 blocksize, parent->start,
4822 root_owner, root_gen,
4823 *level - 1, 1);
4824 BUG_ON(ret);
4825
4826 atomic_inc(&root->fs_info->throttle_gen);
4827 wake_up(&root->fs_info->transaction_throttle);
4828 cond_resched();
4829
4830 continue;
4831 }
4832
4833 /*
4834 * we need to keep freeing things in the next level down.
4835 * read the block and loop around to process it
4836 */
4837 next = read_tree_block(root, bytenr, blocksize, ptr_gen);
4838 WARN_ON(*level <= 0);
4839 if (path->nodes[*level-1])
4840 free_extent_buffer(path->nodes[*level-1]);
4841 path->nodes[*level-1] = next;
4842 *level = btrfs_header_level(next);
4843 path->slots[*level] = 0;
4844 cond_resched();
4845 }
4846 out:
4847 WARN_ON(*level < 0);
4848 WARN_ON(*level >= BTRFS_MAX_LEVEL);
4849
4850 if (path->nodes[*level] == root->node) {
4851 parent = path->nodes[*level];
4852 bytenr = path->nodes[*level]->start;
4853 } else {
4854 parent = path->nodes[*level + 1];
4855 bytenr = btrfs_node_blockptr(parent, path->slots[*level + 1]);
4856 }
4857
4858 blocksize = btrfs_level_size(root, *level);
4859 root_owner = btrfs_header_owner(parent);
4860 root_gen = btrfs_header_generation(parent);
4861
4862 /*
4863 * cleanup and free the reference on the last node
4864 * we processed
4865 */
4866 ret = btrfs_free_extent(trans, root, bytenr, blocksize,
4867 parent->start, root_owner, root_gen,
4868 *level, 1);
4869 free_extent_buffer(path->nodes[*level]);
4870 path->nodes[*level] = NULL;
4871
4872 *level += 1;
4873 BUG_ON(ret);
4874
4875 cond_resched();
4876 return 0;
4877 }
4878 #endif
4879
4880 struct walk_control {
4881 u64 refs[BTRFS_MAX_LEVEL];
4882 u64 flags[BTRFS_MAX_LEVEL];
4883 struct btrfs_key update_progress;
4884 int stage;
4885 int level;
4886 int shared_level;
4887 int update_ref;
4888 int keep_locks;
4889 int reada_slot;
4890 int reada_count;
4891 };
4892
4893 #define DROP_REFERENCE 1
4894 #define UPDATE_BACKREF 2
4895
4896 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
4897 struct btrfs_root *root,
4898 struct walk_control *wc,
4899 struct btrfs_path *path)
4900 {
4901 u64 bytenr;
4902 u64 generation;
4903 u64 refs;
4904 u64 last = 0;
4905 u32 nritems;
4906 u32 blocksize;
4907 struct btrfs_key key;
4908 struct extent_buffer *eb;
4909 int ret;
4910 int slot;
4911 int nread = 0;
4912
4913 if (path->slots[wc->level] < wc->reada_slot) {
4914 wc->reada_count = wc->reada_count * 2 / 3;
4915 wc->reada_count = max(wc->reada_count, 2);
4916 } else {
4917 wc->reada_count = wc->reada_count * 3 / 2;
4918 wc->reada_count = min_t(int, wc->reada_count,
4919 BTRFS_NODEPTRS_PER_BLOCK(root));
4920 }
4921
4922 eb = path->nodes[wc->level];
4923 nritems = btrfs_header_nritems(eb);
4924 blocksize = btrfs_level_size(root, wc->level - 1);
4925
4926 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
4927 if (nread >= wc->reada_count)
4928 break;
4929
4930 cond_resched();
4931 bytenr = btrfs_node_blockptr(eb, slot);
4932 generation = btrfs_node_ptr_generation(eb, slot);
4933
4934 if (slot == path->slots[wc->level])
4935 goto reada;
4936
4937 if (wc->stage == UPDATE_BACKREF &&
4938 generation <= root->root_key.offset)
4939 continue;
4940
4941 if (wc->stage == DROP_REFERENCE) {
4942 ret = btrfs_lookup_extent_info(trans, root,
4943 bytenr, blocksize,
4944 &refs, NULL);
4945 BUG_ON(ret);
4946 BUG_ON(refs == 0);
4947 if (refs == 1)
4948 goto reada;
4949
4950 if (!wc->update_ref ||
4951 generation <= root->root_key.offset)
4952 continue;
4953 btrfs_node_key_to_cpu(eb, &key, slot);
4954 ret = btrfs_comp_cpu_keys(&key,
4955 &wc->update_progress);
4956 if (ret < 0)
4957 continue;
4958 }
4959 reada:
4960 ret = readahead_tree_block(root, bytenr, blocksize,
4961 generation);
4962 if (ret)
4963 break;
4964 last = bytenr + blocksize;
4965 nread++;
4966 }
4967 wc->reada_slot = slot;
4968 }
4969
4970 /*
4971 * hepler to process tree block while walking down the tree.
4972 *
4973 * when wc->stage == UPDATE_BACKREF, this function updates
4974 * back refs for pointers in the block.
4975 *
4976 * NOTE: return value 1 means we should stop walking down.
4977 */
4978 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
4979 struct btrfs_root *root,
4980 struct btrfs_path *path,
4981 struct walk_control *wc)
4982 {
4983 int level = wc->level;
4984 struct extent_buffer *eb = path->nodes[level];
4985 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
4986 int ret;
4987
4988 if (wc->stage == UPDATE_BACKREF &&
4989 btrfs_header_owner(eb) != root->root_key.objectid)
4990 return 1;
4991
4992 /*
4993 * when reference count of tree block is 1, it won't increase
4994 * again. once full backref flag is set, we never clear it.
4995 */
4996 if ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
4997 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag))) {
4998 BUG_ON(!path->locks[level]);
4999 ret = btrfs_lookup_extent_info(trans, root,
5000 eb->start, eb->len,
5001 &wc->refs[level],
5002 &wc->flags[level]);
5003 BUG_ON(ret);
5004 BUG_ON(wc->refs[level] == 0);
5005 }
5006
5007 if (wc->stage == DROP_REFERENCE) {
5008 if (wc->refs[level] > 1)
5009 return 1;
5010
5011 if (path->locks[level] && !wc->keep_locks) {
5012 btrfs_tree_unlock(eb);
5013 path->locks[level] = 0;
5014 }
5015 return 0;
5016 }
5017
5018 /* wc->stage == UPDATE_BACKREF */
5019 if (!(wc->flags[level] & flag)) {
5020 BUG_ON(!path->locks[level]);
5021 ret = btrfs_inc_ref(trans, root, eb, 1);
5022 BUG_ON(ret);
5023 ret = btrfs_dec_ref(trans, root, eb, 0);
5024 BUG_ON(ret);
5025 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
5026 eb->len, flag, 0);
5027 BUG_ON(ret);
5028 wc->flags[level] |= flag;
5029 }
5030
5031 /*
5032 * the block is shared by multiple trees, so it's not good to
5033 * keep the tree lock
5034 */
5035 if (path->locks[level] && level > 0) {
5036 btrfs_tree_unlock(eb);
5037 path->locks[level] = 0;
5038 }
5039 return 0;
5040 }
5041
5042 /*
5043 * hepler to process tree block pointer.
5044 *
5045 * when wc->stage == DROP_REFERENCE, this function checks
5046 * reference count of the block pointed to. if the block
5047 * is shared and we need update back refs for the subtree
5048 * rooted at the block, this function changes wc->stage to
5049 * UPDATE_BACKREF. if the block is shared and there is no
5050 * need to update back, this function drops the reference
5051 * to the block.
5052 *
5053 * NOTE: return value 1 means we should stop walking down.
5054 */
5055 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
5056 struct btrfs_root *root,
5057 struct btrfs_path *path,
5058 struct walk_control *wc)
5059 {
5060 u64 bytenr;
5061 u64 generation;
5062 u64 parent;
5063 u32 blocksize;
5064 struct btrfs_key key;
5065 struct extent_buffer *next;
5066 int level = wc->level;
5067 int reada = 0;
5068 int ret = 0;
5069
5070 generation = btrfs_node_ptr_generation(path->nodes[level],
5071 path->slots[level]);
5072 /*
5073 * if the lower level block was created before the snapshot
5074 * was created, we know there is no need to update back refs
5075 * for the subtree
5076 */
5077 if (wc->stage == UPDATE_BACKREF &&
5078 generation <= root->root_key.offset)
5079 return 1;
5080
5081 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
5082 blocksize = btrfs_level_size(root, level - 1);
5083
5084 next = btrfs_find_tree_block(root, bytenr, blocksize);
5085 if (!next) {
5086 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
5087 reada = 1;
5088 }
5089 btrfs_tree_lock(next);
5090 btrfs_set_lock_blocking(next);
5091
5092 if (wc->stage == DROP_REFERENCE) {
5093 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
5094 &wc->refs[level - 1],
5095 &wc->flags[level - 1]);
5096 BUG_ON(ret);
5097 BUG_ON(wc->refs[level - 1] == 0);
5098
5099 if (wc->refs[level - 1] > 1) {
5100 if (!wc->update_ref ||
5101 generation <= root->root_key.offset)
5102 goto skip;
5103
5104 btrfs_node_key_to_cpu(path->nodes[level], &key,
5105 path->slots[level]);
5106 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
5107 if (ret < 0)
5108 goto skip;
5109
5110 wc->stage = UPDATE_BACKREF;
5111 wc->shared_level = level - 1;
5112 }
5113 }
5114
5115 if (!btrfs_buffer_uptodate(next, generation)) {
5116 btrfs_tree_unlock(next);
5117 free_extent_buffer(next);
5118 next = NULL;
5119 }
5120
5121 if (!next) {
5122 if (reada && level == 1)
5123 reada_walk_down(trans, root, wc, path);
5124 next = read_tree_block(root, bytenr, blocksize, generation);
5125 btrfs_tree_lock(next);
5126 btrfs_set_lock_blocking(next);
5127 }
5128
5129 level--;
5130 BUG_ON(level != btrfs_header_level(next));
5131 path->nodes[level] = next;
5132 path->slots[level] = 0;
5133 path->locks[level] = 1;
5134 wc->level = level;
5135 if (wc->level == 1)
5136 wc->reada_slot = 0;
5137 return 0;
5138 skip:
5139 wc->refs[level - 1] = 0;
5140 wc->flags[level - 1] = 0;
5141
5142 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
5143 parent = path->nodes[level]->start;
5144 } else {
5145 BUG_ON(root->root_key.objectid !=
5146 btrfs_header_owner(path->nodes[level]));
5147 parent = 0;
5148 }
5149
5150 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
5151 root->root_key.objectid, level - 1, 0);
5152 BUG_ON(ret);
5153
5154 btrfs_tree_unlock(next);
5155 free_extent_buffer(next);
5156 return 1;
5157 }
5158
5159 /*
5160 * hepler to process tree block while walking up the tree.
5161 *
5162 * when wc->stage == DROP_REFERENCE, this function drops
5163 * reference count on the block.
5164 *
5165 * when wc->stage == UPDATE_BACKREF, this function changes
5166 * wc->stage back to DROP_REFERENCE if we changed wc->stage
5167 * to UPDATE_BACKREF previously while processing the block.
5168 *
5169 * NOTE: return value 1 means we should stop walking up.
5170 */
5171 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
5172 struct btrfs_root *root,
5173 struct btrfs_path *path,
5174 struct walk_control *wc)
5175 {
5176 int ret = 0;
5177 int level = wc->level;
5178 struct extent_buffer *eb = path->nodes[level];
5179 u64 parent = 0;
5180
5181 if (wc->stage == UPDATE_BACKREF) {
5182 BUG_ON(wc->shared_level < level);
5183 if (level < wc->shared_level)
5184 goto out;
5185
5186 ret = find_next_key(path, level + 1, &wc->update_progress);
5187 if (ret > 0)
5188 wc->update_ref = 0;
5189
5190 wc->stage = DROP_REFERENCE;
5191 wc->shared_level = -1;
5192 path->slots[level] = 0;
5193
5194 /*
5195 * check reference count again if the block isn't locked.
5196 * we should start walking down the tree again if reference
5197 * count is one.
5198 */
5199 if (!path->locks[level]) {
5200 BUG_ON(level == 0);
5201 btrfs_tree_lock(eb);
5202 btrfs_set_lock_blocking(eb);
5203 path->locks[level] = 1;
5204
5205 ret = btrfs_lookup_extent_info(trans, root,
5206 eb->start, eb->len,
5207 &wc->refs[level],
5208 &wc->flags[level]);
5209 BUG_ON(ret);
5210 BUG_ON(wc->refs[level] == 0);
5211 if (wc->refs[level] == 1) {
5212 btrfs_tree_unlock(eb);
5213 path->locks[level] = 0;
5214 return 1;
5215 }
5216 }
5217 }
5218
5219 /* wc->stage == DROP_REFERENCE */
5220 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
5221
5222 if (wc->refs[level] == 1) {
5223 if (level == 0) {
5224 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5225 ret = btrfs_dec_ref(trans, root, eb, 1);
5226 else
5227 ret = btrfs_dec_ref(trans, root, eb, 0);
5228 BUG_ON(ret);
5229 }
5230 /* make block locked assertion in clean_tree_block happy */
5231 if (!path->locks[level] &&
5232 btrfs_header_generation(eb) == trans->transid) {
5233 btrfs_tree_lock(eb);
5234 btrfs_set_lock_blocking(eb);
5235 path->locks[level] = 1;
5236 }
5237 clean_tree_block(trans, root, eb);
5238 }
5239
5240 if (eb == root->node) {
5241 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5242 parent = eb->start;
5243 else
5244 BUG_ON(root->root_key.objectid !=
5245 btrfs_header_owner(eb));
5246 } else {
5247 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5248 parent = path->nodes[level + 1]->start;
5249 else
5250 BUG_ON(root->root_key.objectid !=
5251 btrfs_header_owner(path->nodes[level + 1]));
5252 }
5253
5254 ret = btrfs_free_extent(trans, root, eb->start, eb->len, parent,
5255 root->root_key.objectid, level, 0);
5256 BUG_ON(ret);
5257 out:
5258 wc->refs[level] = 0;
5259 wc->flags[level] = 0;
5260 return ret;
5261 }
5262
5263 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
5264 struct btrfs_root *root,
5265 struct btrfs_path *path,
5266 struct walk_control *wc)
5267 {
5268 int level = wc->level;
5269 int ret;
5270
5271 while (level >= 0) {
5272 if (path->slots[level] >=
5273 btrfs_header_nritems(path->nodes[level]))
5274 break;
5275
5276 ret = walk_down_proc(trans, root, path, wc);
5277 if (ret > 0)
5278 break;
5279
5280 if (level == 0)
5281 break;
5282
5283 ret = do_walk_down(trans, root, path, wc);
5284 if (ret > 0) {
5285 path->slots[level]++;
5286 continue;
5287 }
5288 level = wc->level;
5289 }
5290 return 0;
5291 }
5292
5293 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
5294 struct btrfs_root *root,
5295 struct btrfs_path *path,
5296 struct walk_control *wc, int max_level)
5297 {
5298 int level = wc->level;
5299 int ret;
5300
5301 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
5302 while (level < max_level && path->nodes[level]) {
5303 wc->level = level;
5304 if (path->slots[level] + 1 <
5305 btrfs_header_nritems(path->nodes[level])) {
5306 path->slots[level]++;
5307 return 0;
5308 } else {
5309 ret = walk_up_proc(trans, root, path, wc);
5310 if (ret > 0)
5311 return 0;
5312
5313 if (path->locks[level]) {
5314 btrfs_tree_unlock(path->nodes[level]);
5315 path->locks[level] = 0;
5316 }
5317 free_extent_buffer(path->nodes[level]);
5318 path->nodes[level] = NULL;
5319 level++;
5320 }
5321 }
5322 return 1;
5323 }
5324
5325 /*
5326 * drop a subvolume tree.
5327 *
5328 * this function traverses the tree freeing any blocks that only
5329 * referenced by the tree.
5330 *
5331 * when a shared tree block is found. this function decreases its
5332 * reference count by one. if update_ref is true, this function
5333 * also make sure backrefs for the shared block and all lower level
5334 * blocks are properly updated.
5335 */
5336 int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref)
5337 {
5338 struct btrfs_path *path;
5339 struct btrfs_trans_handle *trans;
5340 struct btrfs_root *tree_root = root->fs_info->tree_root;
5341 struct btrfs_root_item *root_item = &root->root_item;
5342 struct walk_control *wc;
5343 struct btrfs_key key;
5344 int err = 0;
5345 int ret;
5346 int level;
5347
5348 path = btrfs_alloc_path();
5349 BUG_ON(!path);
5350
5351 wc = kzalloc(sizeof(*wc), GFP_NOFS);
5352 BUG_ON(!wc);
5353
5354 trans = btrfs_start_transaction(tree_root, 1);
5355
5356 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
5357 level = btrfs_header_level(root->node);
5358 path->nodes[level] = btrfs_lock_root_node(root);
5359 btrfs_set_lock_blocking(path->nodes[level]);
5360 path->slots[level] = 0;
5361 path->locks[level] = 1;
5362 memset(&wc->update_progress, 0,
5363 sizeof(wc->update_progress));
5364 } else {
5365 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
5366 memcpy(&wc->update_progress, &key,
5367 sizeof(wc->update_progress));
5368
5369 level = root_item->drop_level;
5370 BUG_ON(level == 0);
5371 path->lowest_level = level;
5372 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5373 path->lowest_level = 0;
5374 if (ret < 0) {
5375 err = ret;
5376 goto out;
5377 }
5378 WARN_ON(ret > 0);
5379
5380 /*
5381 * unlock our path, this is safe because only this
5382 * function is allowed to delete this snapshot
5383 */
5384 btrfs_unlock_up_safe(path, 0);
5385
5386 level = btrfs_header_level(root->node);
5387 while (1) {
5388 btrfs_tree_lock(path->nodes[level]);
5389 btrfs_set_lock_blocking(path->nodes[level]);
5390
5391 ret = btrfs_lookup_extent_info(trans, root,
5392 path->nodes[level]->start,
5393 path->nodes[level]->len,
5394 &wc->refs[level],
5395 &wc->flags[level]);
5396 BUG_ON(ret);
5397 BUG_ON(wc->refs[level] == 0);
5398
5399 if (level == root_item->drop_level)
5400 break;
5401
5402 btrfs_tree_unlock(path->nodes[level]);
5403 WARN_ON(wc->refs[level] != 1);
5404 level--;
5405 }
5406 }
5407
5408 wc->level = level;
5409 wc->shared_level = -1;
5410 wc->stage = DROP_REFERENCE;
5411 wc->update_ref = update_ref;
5412 wc->keep_locks = 0;
5413 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
5414
5415 while (1) {
5416 ret = walk_down_tree(trans, root, path, wc);
5417 if (ret < 0) {
5418 err = ret;
5419 break;
5420 }
5421
5422 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
5423 if (ret < 0) {
5424 err = ret;
5425 break;
5426 }
5427
5428 if (ret > 0) {
5429 BUG_ON(wc->stage != DROP_REFERENCE);
5430 break;
5431 }
5432
5433 if (wc->stage == DROP_REFERENCE) {
5434 level = wc->level;
5435 btrfs_node_key(path->nodes[level],
5436 &root_item->drop_progress,
5437 path->slots[level]);
5438 root_item->drop_level = level;
5439 }
5440
5441 BUG_ON(wc->level == 0);
5442 if (trans->transaction->in_commit ||
5443 trans->transaction->delayed_refs.flushing) {
5444 ret = btrfs_update_root(trans, tree_root,
5445 &root->root_key,
5446 root_item);
5447 BUG_ON(ret);
5448
5449 btrfs_end_transaction(trans, tree_root);
5450 trans = btrfs_start_transaction(tree_root, 1);
5451 } else {
5452 unsigned long update;
5453 update = trans->delayed_ref_updates;
5454 trans->delayed_ref_updates = 0;
5455 if (update)
5456 btrfs_run_delayed_refs(trans, tree_root,
5457 update);
5458 }
5459 }
5460 btrfs_release_path(root, path);
5461 BUG_ON(err);
5462
5463 ret = btrfs_del_root(trans, tree_root, &root->root_key);
5464 BUG_ON(ret);
5465
5466 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
5467 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
5468 NULL, NULL);
5469 BUG_ON(ret < 0);
5470 if (ret > 0) {
5471 ret = btrfs_del_orphan_item(trans, tree_root,
5472 root->root_key.objectid);
5473 BUG_ON(ret);
5474 }
5475 }
5476
5477 if (root->in_radix) {
5478 btrfs_free_fs_root(tree_root->fs_info, root);
5479 } else {
5480 free_extent_buffer(root->node);
5481 free_extent_buffer(root->commit_root);
5482 kfree(root);
5483 }
5484 out:
5485 btrfs_end_transaction(trans, tree_root);
5486 kfree(wc);
5487 btrfs_free_path(path);
5488 return err;
5489 }
5490
5491 /*
5492 * drop subtree rooted at tree block 'node'.
5493 *
5494 * NOTE: this function will unlock and release tree block 'node'
5495 */
5496 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
5497 struct btrfs_root *root,
5498 struct extent_buffer *node,
5499 struct extent_buffer *parent)
5500 {
5501 struct btrfs_path *path;
5502 struct walk_control *wc;
5503 int level;
5504 int parent_level;
5505 int ret = 0;
5506 int wret;
5507
5508 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
5509
5510 path = btrfs_alloc_path();
5511 BUG_ON(!path);
5512
5513 wc = kzalloc(sizeof(*wc), GFP_NOFS);
5514 BUG_ON(!wc);
5515
5516 btrfs_assert_tree_locked(parent);
5517 parent_level = btrfs_header_level(parent);
5518 extent_buffer_get(parent);
5519 path->nodes[parent_level] = parent;
5520 path->slots[parent_level] = btrfs_header_nritems(parent);
5521
5522 btrfs_assert_tree_locked(node);
5523 level = btrfs_header_level(node);
5524 path->nodes[level] = node;
5525 path->slots[level] = 0;
5526 path->locks[level] = 1;
5527
5528 wc->refs[parent_level] = 1;
5529 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5530 wc->level = level;
5531 wc->shared_level = -1;
5532 wc->stage = DROP_REFERENCE;
5533 wc->update_ref = 0;
5534 wc->keep_locks = 1;
5535 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
5536
5537 while (1) {
5538 wret = walk_down_tree(trans, root, path, wc);
5539 if (wret < 0) {
5540 ret = wret;
5541 break;
5542 }
5543
5544 wret = walk_up_tree(trans, root, path, wc, parent_level);
5545 if (wret < 0)
5546 ret = wret;
5547 if (wret != 0)
5548 break;
5549 }
5550
5551 kfree(wc);
5552 btrfs_free_path(path);
5553 return ret;
5554 }
5555
5556 #if 0
5557 static unsigned long calc_ra(unsigned long start, unsigned long last,
5558 unsigned long nr)
5559 {
5560 return min(last, start + nr - 1);
5561 }
5562
5563 static noinline int relocate_inode_pages(struct inode *inode, u64 start,
5564 u64 len)
5565 {
5566 u64 page_start;
5567 u64 page_end;
5568 unsigned long first_index;
5569 unsigned long last_index;
5570 unsigned long i;
5571 struct page *page;
5572 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5573 struct file_ra_state *ra;
5574 struct btrfs_ordered_extent *ordered;
5575 unsigned int total_read = 0;
5576 unsigned int total_dirty = 0;
5577 int ret = 0;
5578
5579 ra = kzalloc(sizeof(*ra), GFP_NOFS);
5580
5581 mutex_lock(&inode->i_mutex);
5582 first_index = start >> PAGE_CACHE_SHIFT;
5583 last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
5584
5585 /* make sure the dirty trick played by the caller work */
5586 ret = invalidate_inode_pages2_range(inode->i_mapping,
5587 first_index, last_index);
5588 if (ret)
5589 goto out_unlock;
5590
5591 file_ra_state_init(ra, inode->i_mapping);
5592
5593 for (i = first_index ; i <= last_index; i++) {
5594 if (total_read % ra->ra_pages == 0) {
5595 btrfs_force_ra(inode->i_mapping, ra, NULL, i,
5596 calc_ra(i, last_index, ra->ra_pages));
5597 }
5598 total_read++;
5599 again:
5600 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
5601 BUG_ON(1);
5602 page = grab_cache_page(inode->i_mapping, i);
5603 if (!page) {
5604 ret = -ENOMEM;
5605 goto out_unlock;
5606 }
5607 if (!PageUptodate(page)) {
5608 btrfs_readpage(NULL, page);
5609 lock_page(page);
5610 if (!PageUptodate(page)) {
5611 unlock_page(page);
5612 page_cache_release(page);
5613 ret = -EIO;
5614 goto out_unlock;
5615 }
5616 }
5617 wait_on_page_writeback(page);
5618
5619 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
5620 page_end = page_start + PAGE_CACHE_SIZE - 1;
5621 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
5622
5623 ordered = btrfs_lookup_ordered_extent(inode, page_start);
5624 if (ordered) {
5625 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
5626 unlock_page(page);
5627 page_cache_release(page);
5628 btrfs_start_ordered_extent(inode, ordered, 1);
5629 btrfs_put_ordered_extent(ordered);
5630 goto again;
5631 }
5632 set_page_extent_mapped(page);
5633
5634 if (i == first_index)
5635 set_extent_bits(io_tree, page_start, page_end,
5636 EXTENT_BOUNDARY, GFP_NOFS);
5637 btrfs_set_extent_delalloc(inode, page_start, page_end);
5638
5639 set_page_dirty(page);
5640 total_dirty++;
5641
5642 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
5643 unlock_page(page);
5644 page_cache_release(page);
5645 }
5646
5647 out_unlock:
5648 kfree(ra);
5649 mutex_unlock(&inode->i_mutex);
5650 balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
5651 return ret;
5652 }
5653
5654 static noinline int relocate_data_extent(struct inode *reloc_inode,
5655 struct btrfs_key *extent_key,
5656 u64 offset)
5657 {
5658 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
5659 struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
5660 struct extent_map *em;
5661 u64 start = extent_key->objectid - offset;
5662 u64 end = start + extent_key->offset - 1;
5663
5664 em = alloc_extent_map(GFP_NOFS);
5665 BUG_ON(!em || IS_ERR(em));
5666
5667 em->start = start;
5668 em->len = extent_key->offset;
5669 em->block_len = extent_key->offset;
5670 em->block_start = extent_key->objectid;
5671 em->bdev = root->fs_info->fs_devices->latest_bdev;
5672 set_bit(EXTENT_FLAG_PINNED, &em->flags);
5673
5674 /* setup extent map to cheat btrfs_readpage */
5675 lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
5676 while (1) {
5677 int ret;
5678 write_lock(&em_tree->lock);
5679 ret = add_extent_mapping(em_tree, em);
5680 write_unlock(&em_tree->lock);
5681 if (ret != -EEXIST) {
5682 free_extent_map(em);
5683 break;
5684 }
5685 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
5686 }
5687 unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
5688
5689 return relocate_inode_pages(reloc_inode, start, extent_key->offset);
5690 }
5691
5692 struct btrfs_ref_path {
5693 u64 extent_start;
5694 u64 nodes[BTRFS_MAX_LEVEL];
5695 u64 root_objectid;
5696 u64 root_generation;
5697 u64 owner_objectid;
5698 u32 num_refs;
5699 int lowest_level;
5700 int current_level;
5701 int shared_level;
5702
5703 struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
5704 u64 new_nodes[BTRFS_MAX_LEVEL];
5705 };
5706
5707 struct disk_extent {
5708 u64 ram_bytes;
5709 u64 disk_bytenr;
5710 u64 disk_num_bytes;
5711 u64 offset;
5712 u64 num_bytes;
5713 u8 compression;
5714 u8 encryption;
5715 u16 other_encoding;
5716 };
5717
5718 static int is_cowonly_root(u64 root_objectid)
5719 {
5720 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
5721 root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
5722 root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
5723 root_objectid == BTRFS_DEV_TREE_OBJECTID ||
5724 root_objectid == BTRFS_TREE_LOG_OBJECTID ||
5725 root_objectid == BTRFS_CSUM_TREE_OBJECTID)
5726 return 1;
5727 return 0;
5728 }
5729
5730 static noinline int __next_ref_path(struct btrfs_trans_handle *trans,
5731 struct btrfs_root *extent_root,
5732 struct btrfs_ref_path *ref_path,
5733 int first_time)
5734 {
5735 struct extent_buffer *leaf;
5736 struct btrfs_path *path;
5737 struct btrfs_extent_ref *ref;
5738 struct btrfs_key key;
5739 struct btrfs_key found_key;
5740 u64 bytenr;
5741 u32 nritems;
5742 int level;
5743 int ret = 1;
5744
5745 path = btrfs_alloc_path();
5746 if (!path)
5747 return -ENOMEM;
5748
5749 if (first_time) {
5750 ref_path->lowest_level = -1;
5751 ref_path->current_level = -1;
5752 ref_path->shared_level = -1;
5753 goto walk_up;
5754 }
5755 walk_down:
5756 level = ref_path->current_level - 1;
5757 while (level >= -1) {
5758 u64 parent;
5759 if (level < ref_path->lowest_level)
5760 break;
5761
5762 if (level >= 0)
5763 bytenr = ref_path->nodes[level];
5764 else
5765 bytenr = ref_path->extent_start;
5766 BUG_ON(bytenr == 0);
5767
5768 parent = ref_path->nodes[level + 1];
5769 ref_path->nodes[level + 1] = 0;
5770 ref_path->current_level = level;
5771 BUG_ON(parent == 0);
5772
5773 key.objectid = bytenr;
5774 key.offset = parent + 1;
5775 key.type = BTRFS_EXTENT_REF_KEY;
5776
5777 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
5778 if (ret < 0)
5779 goto out;
5780 BUG_ON(ret == 0);
5781
5782 leaf = path->nodes[0];
5783 nritems = btrfs_header_nritems(leaf);
5784 if (path->slots[0] >= nritems) {
5785 ret = btrfs_next_leaf(extent_root, path);
5786 if (ret < 0)
5787 goto out;
5788 if (ret > 0)
5789 goto next;
5790 leaf = path->nodes[0];
5791 }
5792
5793 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5794 if (found_key.objectid == bytenr &&
5795 found_key.type == BTRFS_EXTENT_REF_KEY) {
5796 if (level < ref_path->shared_level)
5797 ref_path->shared_level = level;
5798 goto found;
5799 }
5800 next:
5801 level--;
5802 btrfs_release_path(extent_root, path);
5803 cond_resched();
5804 }
5805 /* reached lowest level */
5806 ret = 1;
5807 goto out;
5808 walk_up:
5809 level = ref_path->current_level;
5810 while (level < BTRFS_MAX_LEVEL - 1) {
5811 u64 ref_objectid;
5812
5813 if (level >= 0)
5814 bytenr = ref_path->nodes[level];
5815 else
5816 bytenr = ref_path->extent_start;
5817
5818 BUG_ON(bytenr == 0);
5819
5820 key.objectid = bytenr;
5821 key.offset = 0;
5822 key.type = BTRFS_EXTENT_REF_KEY;
5823
5824 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
5825 if (ret < 0)
5826 goto out;
5827
5828 leaf = path->nodes[0];
5829 nritems = btrfs_header_nritems(leaf);
5830 if (path->slots[0] >= nritems) {
5831 ret = btrfs_next_leaf(extent_root, path);
5832 if (ret < 0)
5833 goto out;
5834 if (ret > 0) {
5835 /* the extent was freed by someone */
5836 if (ref_path->lowest_level == level)
5837 goto out;
5838 btrfs_release_path(extent_root, path);
5839 goto walk_down;
5840 }
5841 leaf = path->nodes[0];
5842 }
5843
5844 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5845 if (found_key.objectid != bytenr ||
5846 found_key.type != BTRFS_EXTENT_REF_KEY) {
5847 /* the extent was freed by someone */
5848 if (ref_path->lowest_level == level) {
5849 ret = 1;
5850 goto out;
5851 }
5852 btrfs_release_path(extent_root, path);
5853 goto walk_down;
5854 }
5855 found:
5856 ref = btrfs_item_ptr(leaf, path->slots[0],
5857 struct btrfs_extent_ref);
5858 ref_objectid = btrfs_ref_objectid(leaf, ref);
5859 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
5860 if (first_time) {
5861 level = (int)ref_objectid;
5862 BUG_ON(level >= BTRFS_MAX_LEVEL);
5863 ref_path->lowest_level = level;
5864 ref_path->current_level = level;
5865 ref_path->nodes[level] = bytenr;
5866 } else {
5867 WARN_ON(ref_objectid != level);
5868 }
5869 } else {
5870 WARN_ON(level != -1);
5871 }
5872 first_time = 0;
5873
5874 if (ref_path->lowest_level == level) {
5875 ref_path->owner_objectid = ref_objectid;
5876 ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
5877 }
5878
5879 /*
5880 * the block is tree root or the block isn't in reference
5881 * counted tree.
5882 */
5883 if (found_key.objectid == found_key.offset ||
5884 is_cowonly_root(btrfs_ref_root(leaf, ref))) {
5885 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
5886 ref_path->root_generation =
5887 btrfs_ref_generation(leaf, ref);
5888 if (level < 0) {
5889 /* special reference from the tree log */
5890 ref_path->nodes[0] = found_key.offset;
5891 ref_path->current_level = 0;
5892 }
5893 ret = 0;
5894 goto out;
5895 }
5896
5897 level++;
5898 BUG_ON(ref_path->nodes[level] != 0);
5899 ref_path->nodes[level] = found_key.offset;
5900 ref_path->current_level = level;
5901
5902 /*
5903 * the reference was created in the running transaction,
5904 * no need to continue walking up.
5905 */
5906 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
5907 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
5908 ref_path->root_generation =
5909 btrfs_ref_generation(leaf, ref);
5910 ret = 0;
5911 goto out;
5912 }
5913
5914 btrfs_release_path(extent_root, path);
5915 cond_resched();
5916 }
5917 /* reached max tree level, but no tree root found. */
5918 BUG();
5919 out:
5920 btrfs_free_path(path);
5921 return ret;
5922 }
5923
5924 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
5925 struct btrfs_root *extent_root,
5926 struct btrfs_ref_path *ref_path,
5927 u64 extent_start)
5928 {
5929 memset(ref_path, 0, sizeof(*ref_path));
5930 ref_path->extent_start = extent_start;
5931
5932 return __next_ref_path(trans, extent_root, ref_path, 1);
5933 }
5934
5935 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
5936 struct btrfs_root *extent_root,
5937 struct btrfs_ref_path *ref_path)
5938 {
5939 return __next_ref_path(trans, extent_root, ref_path, 0);
5940 }
5941
5942 static noinline int get_new_locations(struct inode *reloc_inode,
5943 struct btrfs_key *extent_key,
5944 u64 offset, int no_fragment,
5945 struct disk_extent **extents,
5946 int *nr_extents)
5947 {
5948 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
5949 struct btrfs_path *path;
5950 struct btrfs_file_extent_item *fi;
5951 struct extent_buffer *leaf;
5952 struct disk_extent *exts = *extents;
5953 struct btrfs_key found_key;
5954 u64 cur_pos;
5955 u64 last_byte;
5956 u32 nritems;
5957 int nr = 0;
5958 int max = *nr_extents;
5959 int ret;
5960
5961 WARN_ON(!no_fragment && *extents);
5962 if (!exts) {
5963 max = 1;
5964 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
5965 if (!exts)
5966 return -ENOMEM;
5967 }
5968
5969 path = btrfs_alloc_path();
5970 BUG_ON(!path);
5971
5972 cur_pos = extent_key->objectid - offset;
5973 last_byte = extent_key->objectid + extent_key->offset;
5974 ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
5975 cur_pos, 0);
5976 if (ret < 0)
5977 goto out;
5978 if (ret > 0) {
5979 ret = -ENOENT;
5980 goto out;
5981 }
5982
5983 while (1) {
5984 leaf = path->nodes[0];
5985 nritems = btrfs_header_nritems(leaf);
5986 if (path->slots[0] >= nritems) {
5987 ret = btrfs_next_leaf(root, path);
5988 if (ret < 0)
5989 goto out;
5990 if (ret > 0)
5991 break;
5992 leaf = path->nodes[0];
5993 }
5994
5995 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5996 if (found_key.offset != cur_pos ||
5997 found_key.type != BTRFS_EXTENT_DATA_KEY ||
5998 found_key.objectid != reloc_inode->i_ino)
5999 break;
6000
6001 fi = btrfs_item_ptr(leaf, path->slots[0],
6002 struct btrfs_file_extent_item);
6003 if (btrfs_file_extent_type(leaf, fi) !=
6004 BTRFS_FILE_EXTENT_REG ||
6005 btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
6006 break;
6007
6008 if (nr == max) {
6009 struct disk_extent *old = exts;
6010 max *= 2;
6011 exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
6012 memcpy(exts, old, sizeof(*exts) * nr);
6013 if (old != *extents)
6014 kfree(old);
6015 }
6016
6017 exts[nr].disk_bytenr =
6018 btrfs_file_extent_disk_bytenr(leaf, fi);
6019 exts[nr].disk_num_bytes =
6020 btrfs_file_extent_disk_num_bytes(leaf, fi);
6021 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
6022 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
6023 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6024 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
6025 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
6026 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
6027 fi);
6028 BUG_ON(exts[nr].offset > 0);
6029 BUG_ON(exts[nr].compression || exts[nr].encryption);
6030 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
6031
6032 cur_pos += exts[nr].num_bytes;
6033 nr++;
6034
6035 if (cur_pos + offset >= last_byte)
6036 break;
6037
6038 if (no_fragment) {
6039 ret = 1;
6040 goto out;
6041 }
6042 path->slots[0]++;
6043 }
6044
6045 BUG_ON(cur_pos + offset > last_byte);
6046 if (cur_pos + offset < last_byte) {
6047 ret = -ENOENT;
6048 goto out;
6049 }
6050 ret = 0;
6051 out:
6052 btrfs_free_path(path);
6053 if (ret) {
6054 if (exts != *extents)
6055 kfree(exts);
6056 } else {
6057 *extents = exts;
6058 *nr_extents = nr;
6059 }
6060 return ret;
6061 }
6062
6063 static noinline int replace_one_extent(struct btrfs_trans_handle *trans,
6064 struct btrfs_root *root,
6065 struct btrfs_path *path,
6066 struct btrfs_key *extent_key,
6067 struct btrfs_key *leaf_key,
6068 struct btrfs_ref_path *ref_path,
6069 struct disk_extent *new_extents,
6070 int nr_extents)
6071 {
6072 struct extent_buffer *leaf;
6073 struct btrfs_file_extent_item *fi;
6074 struct inode *inode = NULL;
6075 struct btrfs_key key;
6076 u64 lock_start = 0;
6077 u64 lock_end = 0;
6078 u64 num_bytes;
6079 u64 ext_offset;
6080 u64 search_end = (u64)-1;
6081 u32 nritems;
6082 int nr_scaned = 0;
6083 int extent_locked = 0;
6084 int extent_type;
6085 int ret;
6086
6087 memcpy(&key, leaf_key, sizeof(key));
6088 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
6089 if (key.objectid < ref_path->owner_objectid ||
6090 (key.objectid == ref_path->owner_objectid &&
6091 key.type < BTRFS_EXTENT_DATA_KEY)) {
6092 key.objectid = ref_path->owner_objectid;
6093 key.type = BTRFS_EXTENT_DATA_KEY;
6094 key.offset = 0;
6095 }
6096 }
6097
6098 while (1) {
6099 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
6100 if (ret < 0)
6101 goto out;
6102
6103 leaf = path->nodes[0];
6104 nritems = btrfs_header_nritems(leaf);
6105 next:
6106 if (extent_locked && ret > 0) {
6107 /*
6108 * the file extent item was modified by someone
6109 * before the extent got locked.
6110 */
6111 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6112 lock_end, GFP_NOFS);
6113 extent_locked = 0;
6114 }
6115
6116 if (path->slots[0] >= nritems) {
6117 if (++nr_scaned > 2)
6118 break;
6119
6120 BUG_ON(extent_locked);
6121 ret = btrfs_next_leaf(root, path);
6122 if (ret < 0)
6123 goto out;
6124 if (ret > 0)
6125 break;
6126 leaf = path->nodes[0];
6127 nritems = btrfs_header_nritems(leaf);
6128 }
6129
6130 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
6131
6132 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
6133 if ((key.objectid > ref_path->owner_objectid) ||
6134 (key.objectid == ref_path->owner_objectid &&
6135 key.type > BTRFS_EXTENT_DATA_KEY) ||
6136 key.offset >= search_end)
6137 break;
6138 }
6139
6140 if (inode && key.objectid != inode->i_ino) {
6141 BUG_ON(extent_locked);
6142 btrfs_release_path(root, path);
6143 mutex_unlock(&inode->i_mutex);
6144 iput(inode);
6145 inode = NULL;
6146 continue;
6147 }
6148
6149 if (key.type != BTRFS_EXTENT_DATA_KEY) {
6150 path->slots[0]++;
6151 ret = 1;
6152 goto next;
6153 }
6154 fi = btrfs_item_ptr(leaf, path->slots[0],
6155 struct btrfs_file_extent_item);
6156 extent_type = btrfs_file_extent_type(leaf, fi);
6157 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
6158 extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
6159 (btrfs_file_extent_disk_bytenr(leaf, fi) !=
6160 extent_key->objectid)) {
6161 path->slots[0]++;
6162 ret = 1;
6163 goto next;
6164 }
6165
6166 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
6167 ext_offset = btrfs_file_extent_offset(leaf, fi);
6168
6169 if (search_end == (u64)-1) {
6170 search_end = key.offset - ext_offset +
6171 btrfs_file_extent_ram_bytes(leaf, fi);
6172 }
6173
6174 if (!extent_locked) {
6175 lock_start = key.offset;
6176 lock_end = lock_start + num_bytes - 1;
6177 } else {
6178 if (lock_start > key.offset ||
6179 lock_end + 1 < key.offset + num_bytes) {
6180 unlock_extent(&BTRFS_I(inode)->io_tree,
6181 lock_start, lock_end, GFP_NOFS);
6182 extent_locked = 0;
6183 }
6184 }
6185
6186 if (!inode) {
6187 btrfs_release_path(root, path);
6188
6189 inode = btrfs_iget_locked(root->fs_info->sb,
6190 key.objectid, root);
6191 if (inode->i_state & I_NEW) {
6192 BTRFS_I(inode)->root = root;
6193 BTRFS_I(inode)->location.objectid =
6194 key.objectid;
6195 BTRFS_I(inode)->location.type =
6196 BTRFS_INODE_ITEM_KEY;
6197 BTRFS_I(inode)->location.offset = 0;
6198 btrfs_read_locked_inode(inode);
6199 unlock_new_inode(inode);
6200 }
6201 /*
6202 * some code call btrfs_commit_transaction while
6203 * holding the i_mutex, so we can't use mutex_lock
6204 * here.
6205 */
6206 if (is_bad_inode(inode) ||
6207 !mutex_trylock(&inode->i_mutex)) {
6208 iput(inode);
6209 inode = NULL;
6210 key.offset = (u64)-1;
6211 goto skip;
6212 }
6213 }
6214
6215 if (!extent_locked) {
6216 struct btrfs_ordered_extent *ordered;
6217
6218 btrfs_release_path(root, path);
6219
6220 lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6221 lock_end, GFP_NOFS);
6222 ordered = btrfs_lookup_first_ordered_extent(inode,
6223 lock_end);
6224 if (ordered &&
6225 ordered->file_offset <= lock_end &&
6226 ordered->file_offset + ordered->len > lock_start) {
6227 unlock_extent(&BTRFS_I(inode)->io_tree,
6228 lock_start, lock_end, GFP_NOFS);
6229 btrfs_start_ordered_extent(inode, ordered, 1);
6230 btrfs_put_ordered_extent(ordered);
6231 key.offset += num_bytes;
6232 goto skip;
6233 }
6234 if (ordered)
6235 btrfs_put_ordered_extent(ordered);
6236
6237 extent_locked = 1;
6238 continue;
6239 }
6240
6241 if (nr_extents == 1) {
6242 /* update extent pointer in place */
6243 btrfs_set_file_extent_disk_bytenr(leaf, fi,
6244 new_extents[0].disk_bytenr);
6245 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
6246 new_extents[0].disk_num_bytes);
6247 btrfs_mark_buffer_dirty(leaf);
6248
6249 btrfs_drop_extent_cache(inode, key.offset,
6250 key.offset + num_bytes - 1, 0);
6251
6252 ret = btrfs_inc_extent_ref(trans, root,
6253 new_extents[0].disk_bytenr,
6254 new_extents[0].disk_num_bytes,
6255 leaf->start,
6256 root->root_key.objectid,
6257 trans->transid,
6258 key.objectid);
6259 BUG_ON(ret);
6260
6261 ret = btrfs_free_extent(trans, root,
6262 extent_key->objectid,
6263 extent_key->offset,
6264 leaf->start,
6265 btrfs_header_owner(leaf),
6266 btrfs_header_generation(leaf),
6267 key.objectid, 0);
6268 BUG_ON(ret);
6269
6270 btrfs_release_path(root, path);
6271 key.offset += num_bytes;
6272 } else {
6273 BUG_ON(1);
6274 #if 0
6275 u64 alloc_hint;
6276 u64 extent_len;
6277 int i;
6278 /*
6279 * drop old extent pointer at first, then insert the
6280 * new pointers one bye one
6281 */
6282 btrfs_release_path(root, path);
6283 ret = btrfs_drop_extents(trans, root, inode, key.offset,
6284 key.offset + num_bytes,
6285 key.offset, &alloc_hint);
6286 BUG_ON(ret);
6287
6288 for (i = 0; i < nr_extents; i++) {
6289 if (ext_offset >= new_extents[i].num_bytes) {
6290 ext_offset -= new_extents[i].num_bytes;
6291 continue;
6292 }
6293 extent_len = min(new_extents[i].num_bytes -
6294 ext_offset, num_bytes);
6295
6296 ret = btrfs_insert_empty_item(trans, root,
6297 path, &key,
6298 sizeof(*fi));
6299 BUG_ON(ret);
6300
6301 leaf = path->nodes[0];
6302 fi = btrfs_item_ptr(leaf, path->slots[0],
6303 struct btrfs_file_extent_item);
6304 btrfs_set_file_extent_generation(leaf, fi,
6305 trans->transid);
6306 btrfs_set_file_extent_type(leaf, fi,
6307 BTRFS_FILE_EXTENT_REG);
6308 btrfs_set_file_extent_disk_bytenr(leaf, fi,
6309 new_extents[i].disk_bytenr);
6310 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
6311 new_extents[i].disk_num_bytes);
6312 btrfs_set_file_extent_ram_bytes(leaf, fi,
6313 new_extents[i].ram_bytes);
6314
6315 btrfs_set_file_extent_compression(leaf, fi,
6316 new_extents[i].compression);
6317 btrfs_set_file_extent_encryption(leaf, fi,
6318 new_extents[i].encryption);
6319 btrfs_set_file_extent_other_encoding(leaf, fi,
6320 new_extents[i].other_encoding);
6321
6322 btrfs_set_file_extent_num_bytes(leaf, fi,
6323 extent_len);
6324 ext_offset += new_extents[i].offset;
6325 btrfs_set_file_extent_offset(leaf, fi,
6326 ext_offset);
6327 btrfs_mark_buffer_dirty(leaf);
6328
6329 btrfs_drop_extent_cache(inode, key.offset,
6330 key.offset + extent_len - 1, 0);
6331
6332 ret = btrfs_inc_extent_ref(trans, root,
6333 new_extents[i].disk_bytenr,
6334 new_extents[i].disk_num_bytes,
6335 leaf->start,
6336 root->root_key.objectid,
6337 trans->transid, key.objectid);
6338 BUG_ON(ret);
6339 btrfs_release_path(root, path);
6340
6341 inode_add_bytes(inode, extent_len);
6342
6343 ext_offset = 0;
6344 num_bytes -= extent_len;
6345 key.offset += extent_len;
6346
6347 if (num_bytes == 0)
6348 break;
6349 }
6350 BUG_ON(i >= nr_extents);
6351 #endif
6352 }
6353
6354 if (extent_locked) {
6355 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6356 lock_end, GFP_NOFS);
6357 extent_locked = 0;
6358 }
6359 skip:
6360 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
6361 key.offset >= search_end)
6362 break;
6363
6364 cond_resched();
6365 }
6366 ret = 0;
6367 out:
6368 btrfs_release_path(root, path);
6369 if (inode) {
6370 mutex_unlock(&inode->i_mutex);
6371 if (extent_locked) {
6372 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6373 lock_end, GFP_NOFS);
6374 }
6375 iput(inode);
6376 }
6377 return ret;
6378 }
6379
6380 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
6381 struct btrfs_root *root,
6382 struct extent_buffer *buf, u64 orig_start)
6383 {
6384 int level;
6385 int ret;
6386
6387 BUG_ON(btrfs_header_generation(buf) != trans->transid);
6388 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6389
6390 level = btrfs_header_level(buf);
6391 if (level == 0) {
6392 struct btrfs_leaf_ref *ref;
6393 struct btrfs_leaf_ref *orig_ref;
6394
6395 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
6396 if (!orig_ref)
6397 return -ENOENT;
6398
6399 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
6400 if (!ref) {
6401 btrfs_free_leaf_ref(root, orig_ref);
6402 return -ENOMEM;
6403 }
6404
6405 ref->nritems = orig_ref->nritems;
6406 memcpy(ref->extents, orig_ref->extents,
6407 sizeof(ref->extents[0]) * ref->nritems);
6408
6409 btrfs_free_leaf_ref(root, orig_ref);
6410
6411 ref->root_gen = trans->transid;
6412 ref->bytenr = buf->start;
6413 ref->owner = btrfs_header_owner(buf);
6414 ref->generation = btrfs_header_generation(buf);
6415
6416 ret = btrfs_add_leaf_ref(root, ref, 0);
6417 WARN_ON(ret);
6418 btrfs_free_leaf_ref(root, ref);
6419 }
6420 return 0;
6421 }
6422
6423 static noinline int invalidate_extent_cache(struct btrfs_root *root,
6424 struct extent_buffer *leaf,
6425 struct btrfs_block_group_cache *group,
6426 struct btrfs_root *target_root)
6427 {
6428 struct btrfs_key key;
6429 struct inode *inode = NULL;
6430 struct btrfs_file_extent_item *fi;
6431 u64 num_bytes;
6432 u64 skip_objectid = 0;
6433 u32 nritems;
6434 u32 i;
6435
6436 nritems = btrfs_header_nritems(leaf);
6437 for (i = 0; i < nritems; i++) {
6438 btrfs_item_key_to_cpu(leaf, &key, i);
6439 if (key.objectid == skip_objectid ||
6440 key.type != BTRFS_EXTENT_DATA_KEY)
6441 continue;
6442 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
6443 if (btrfs_file_extent_type(leaf, fi) ==
6444 BTRFS_FILE_EXTENT_INLINE)
6445 continue;
6446 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
6447 continue;
6448 if (!inode || inode->i_ino != key.objectid) {
6449 iput(inode);
6450 inode = btrfs_ilookup(target_root->fs_info->sb,
6451 key.objectid, target_root, 1);
6452 }
6453 if (!inode) {
6454 skip_objectid = key.objectid;
6455 continue;
6456 }
6457 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
6458
6459 lock_extent(&BTRFS_I(inode)->io_tree, key.offset,
6460 key.offset + num_bytes - 1, GFP_NOFS);
6461 btrfs_drop_extent_cache(inode, key.offset,
6462 key.offset + num_bytes - 1, 1);
6463 unlock_extent(&BTRFS_I(inode)->io_tree, key.offset,
6464 key.offset + num_bytes - 1, GFP_NOFS);
6465 cond_resched();
6466 }
6467 iput(inode);
6468 return 0;
6469 }
6470
6471 static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans,
6472 struct btrfs_root *root,
6473 struct extent_buffer *leaf,
6474 struct btrfs_block_group_cache *group,
6475 struct inode *reloc_inode)
6476 {
6477 struct btrfs_key key;
6478 struct btrfs_key extent_key;
6479 struct btrfs_file_extent_item *fi;
6480 struct btrfs_leaf_ref *ref;
6481 struct disk_extent *new_extent;
6482 u64 bytenr;
6483 u64 num_bytes;
6484 u32 nritems;
6485 u32 i;
6486 int ext_index;
6487 int nr_extent;
6488 int ret;
6489
6490 new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
6491 BUG_ON(!new_extent);
6492
6493 ref = btrfs_lookup_leaf_ref(root, leaf->start);
6494 BUG_ON(!ref);
6495
6496 ext_index = -1;
6497 nritems = btrfs_header_nritems(leaf);
6498 for (i = 0; i < nritems; i++) {
6499 btrfs_item_key_to_cpu(leaf, &key, i);
6500 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
6501 continue;
6502 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
6503 if (btrfs_file_extent_type(leaf, fi) ==
6504 BTRFS_FILE_EXTENT_INLINE)
6505 continue;
6506 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
6507 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
6508 if (bytenr == 0)
6509 continue;
6510
6511 ext_index++;
6512 if (bytenr >= group->key.objectid + group->key.offset ||
6513 bytenr + num_bytes <= group->key.objectid)
6514 continue;
6515
6516 extent_key.objectid = bytenr;
6517 extent_key.offset = num_bytes;
6518 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
6519 nr_extent = 1;
6520 ret = get_new_locations(reloc_inode, &extent_key,
6521 group->key.objectid, 1,
6522 &new_extent, &nr_extent);
6523 if (ret > 0)
6524 continue;
6525 BUG_ON(ret < 0);
6526
6527 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
6528 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
6529 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
6530 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
6531
6532 btrfs_set_file_extent_disk_bytenr(leaf, fi,
6533 new_extent->disk_bytenr);
6534 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
6535 new_extent->disk_num_bytes);
6536 btrfs_mark_buffer_dirty(leaf);
6537
6538 ret = btrfs_inc_extent_ref(trans, root,
6539 new_extent->disk_bytenr,
6540 new_extent->disk_num_bytes,
6541 leaf->start,
6542 root->root_key.objectid,
6543 trans->transid, key.objectid);
6544 BUG_ON(ret);
6545
6546 ret = btrfs_free_extent(trans, root,
6547 bytenr, num_bytes, leaf->start,
6548 btrfs_header_owner(leaf),
6549 btrfs_header_generation(leaf),
6550 key.objectid, 0);
6551 BUG_ON(ret);
6552 cond_resched();
6553 }
6554 kfree(new_extent);
6555 BUG_ON(ext_index + 1 != ref->nritems);
6556 btrfs_free_leaf_ref(root, ref);
6557 return 0;
6558 }
6559
6560 int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
6561 struct btrfs_root *root)
6562 {
6563 struct btrfs_root *reloc_root;
6564 int ret;
6565
6566 if (root->reloc_root) {
6567 reloc_root = root->reloc_root;
6568 root->reloc_root = NULL;
6569 list_add(&reloc_root->dead_list,
6570 &root->fs_info->dead_reloc_roots);
6571
6572 btrfs_set_root_bytenr(&reloc_root->root_item,
6573 reloc_root->node->start);
6574 btrfs_set_root_level(&root->root_item,
6575 btrfs_header_level(reloc_root->node));
6576 memset(&reloc_root->root_item.drop_progress, 0,
6577 sizeof(struct btrfs_disk_key));
6578 reloc_root->root_item.drop_level = 0;
6579
6580 ret = btrfs_update_root(trans, root->fs_info->tree_root,
6581 &reloc_root->root_key,
6582 &reloc_root->root_item);
6583 BUG_ON(ret);
6584 }
6585 return 0;
6586 }
6587
6588 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
6589 {
6590 struct btrfs_trans_handle *trans;
6591 struct btrfs_root *reloc_root;
6592 struct btrfs_root *prev_root = NULL;
6593 struct list_head dead_roots;
6594 int ret;
6595 unsigned long nr;
6596
6597 INIT_LIST_HEAD(&dead_roots);
6598 list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
6599
6600 while (!list_empty(&dead_roots)) {
6601 reloc_root = list_entry(dead_roots.prev,
6602 struct btrfs_root, dead_list);
6603 list_del_init(&reloc_root->dead_list);
6604
6605 BUG_ON(reloc_root->commit_root != NULL);
6606 while (1) {
6607 trans = btrfs_join_transaction(root, 1);
6608 BUG_ON(!trans);
6609
6610 mutex_lock(&root->fs_info->drop_mutex);
6611 ret = btrfs_drop_snapshot(trans, reloc_root);
6612 if (ret != -EAGAIN)
6613 break;
6614 mutex_unlock(&root->fs_info->drop_mutex);
6615
6616 nr = trans->blocks_used;
6617 ret = btrfs_end_transaction(trans, root);
6618 BUG_ON(ret);
6619 btrfs_btree_balance_dirty(root, nr);
6620 }
6621
6622 free_extent_buffer(reloc_root->node);
6623
6624 ret = btrfs_del_root(trans, root->fs_info->tree_root,
6625 &reloc_root->root_key);
6626 BUG_ON(ret);
6627 mutex_unlock(&root->fs_info->drop_mutex);
6628
6629 nr = trans->blocks_used;
6630 ret = btrfs_end_transaction(trans, root);
6631 BUG_ON(ret);
6632 btrfs_btree_balance_dirty(root, nr);
6633
6634 kfree(prev_root);
6635 prev_root = reloc_root;
6636 }
6637 if (prev_root) {
6638 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
6639 kfree(prev_root);
6640 }
6641 return 0;
6642 }
6643
6644 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
6645 {
6646 list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
6647 return 0;
6648 }
6649
6650 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
6651 {
6652 struct btrfs_root *reloc_root;
6653 struct btrfs_trans_handle *trans;
6654 struct btrfs_key location;
6655 int found;
6656 int ret;
6657
6658 mutex_lock(&root->fs_info->tree_reloc_mutex);
6659 ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
6660 BUG_ON(ret);
6661 found = !list_empty(&root->fs_info->dead_reloc_roots);
6662 mutex_unlock(&root->fs_info->tree_reloc_mutex);
6663
6664 if (found) {
6665 trans = btrfs_start_transaction(root, 1);
6666 BUG_ON(!trans);
6667 ret = btrfs_commit_transaction(trans, root);
6668 BUG_ON(ret);
6669 }
6670
6671 location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
6672 location.offset = (u64)-1;
6673 location.type = BTRFS_ROOT_ITEM_KEY;
6674
6675 reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
6676 BUG_ON(!reloc_root);
6677 btrfs_orphan_cleanup(reloc_root);
6678 return 0;
6679 }
6680
6681 static noinline int init_reloc_tree(struct btrfs_trans_handle *trans,
6682 struct btrfs_root *root)
6683 {
6684 struct btrfs_root *reloc_root;
6685 struct extent_buffer *eb;
6686 struct btrfs_root_item *root_item;
6687 struct btrfs_key root_key;
6688 int ret;
6689
6690 BUG_ON(!root->ref_cows);
6691 if (root->reloc_root)
6692 return 0;
6693
6694 root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
6695 BUG_ON(!root_item);
6696
6697 ret = btrfs_copy_root(trans, root, root->commit_root,
6698 &eb, BTRFS_TREE_RELOC_OBJECTID);
6699 BUG_ON(ret);
6700
6701 root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
6702 root_key.offset = root->root_key.objectid;
6703 root_key.type = BTRFS_ROOT_ITEM_KEY;
6704
6705 memcpy(root_item, &root->root_item, sizeof(root_item));
6706 btrfs_set_root_refs(root_item, 0);
6707 btrfs_set_root_bytenr(root_item, eb->start);
6708 btrfs_set_root_level(root_item, btrfs_header_level(eb));
6709 btrfs_set_root_generation(root_item, trans->transid);
6710
6711 btrfs_tree_unlock(eb);
6712 free_extent_buffer(eb);
6713
6714 ret = btrfs_insert_root(trans, root->fs_info->tree_root,
6715 &root_key, root_item);
6716 BUG_ON(ret);
6717 kfree(root_item);
6718
6719 reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
6720 &root_key);
6721 BUG_ON(!reloc_root);
6722 reloc_root->last_trans = trans->transid;
6723 reloc_root->commit_root = NULL;
6724 reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
6725
6726 root->reloc_root = reloc_root;
6727 return 0;
6728 }
6729
6730 /*
6731 * Core function of space balance.
6732 *
6733 * The idea is using reloc trees to relocate tree blocks in reference
6734 * counted roots. There is one reloc tree for each subvol, and all
6735 * reloc trees share same root key objectid. Reloc trees are snapshots
6736 * of the latest committed roots of subvols (root->commit_root).
6737 *
6738 * To relocate a tree block referenced by a subvol, there are two steps.
6739 * COW the block through subvol's reloc tree, then update block pointer
6740 * in the subvol to point to the new block. Since all reloc trees share
6741 * same root key objectid, doing special handing for tree blocks owned
6742 * by them is easy. Once a tree block has been COWed in one reloc tree,
6743 * we can use the resulting new block directly when the same block is
6744 * required to COW again through other reloc trees. By this way, relocated
6745 * tree blocks are shared between reloc trees, so they are also shared
6746 * between subvols.
6747 */
6748 static noinline int relocate_one_path(struct btrfs_trans_handle *trans,
6749 struct btrfs_root *root,
6750 struct btrfs_path *path,
6751 struct btrfs_key *first_key,
6752 struct btrfs_ref_path *ref_path,
6753 struct btrfs_block_group_cache *group,
6754 struct inode *reloc_inode)
6755 {
6756 struct btrfs_root *reloc_root;
6757 struct extent_buffer *eb = NULL;
6758 struct btrfs_key *keys;
6759 u64 *nodes;
6760 int level;
6761 int shared_level;
6762 int lowest_level = 0;
6763 int ret;
6764
6765 if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
6766 lowest_level = ref_path->owner_objectid;
6767
6768 if (!root->ref_cows) {
6769 path->lowest_level = lowest_level;
6770 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
6771 BUG_ON(ret < 0);
6772 path->lowest_level = 0;
6773 btrfs_release_path(root, path);
6774 return 0;
6775 }
6776
6777 mutex_lock(&root->fs_info->tree_reloc_mutex);
6778 ret = init_reloc_tree(trans, root);
6779 BUG_ON(ret);
6780 reloc_root = root->reloc_root;
6781
6782 shared_level = ref_path->shared_level;
6783 ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
6784
6785 keys = ref_path->node_keys;
6786 nodes = ref_path->new_nodes;
6787 memset(&keys[shared_level + 1], 0,
6788 sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
6789 memset(&nodes[shared_level + 1], 0,
6790 sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
6791
6792 if (nodes[lowest_level] == 0) {
6793 path->lowest_level = lowest_level;
6794 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
6795 0, 1);
6796 BUG_ON(ret);
6797 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
6798 eb = path->nodes[level];
6799 if (!eb || eb == reloc_root->node)
6800 break;
6801 nodes[level] = eb->start;
6802 if (level == 0)
6803 btrfs_item_key_to_cpu(eb, &keys[level], 0);
6804 else
6805 btrfs_node_key_to_cpu(eb, &keys[level], 0);
6806 }
6807 if (nodes[0] &&
6808 ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
6809 eb = path->nodes[0];
6810 ret = replace_extents_in_leaf(trans, reloc_root, eb,
6811 group, reloc_inode);
6812 BUG_ON(ret);
6813 }
6814 btrfs_release_path(reloc_root, path);
6815 } else {
6816 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
6817 lowest_level);
6818 BUG_ON(ret);
6819 }
6820
6821 /*
6822 * replace tree blocks in the fs tree with tree blocks in
6823 * the reloc tree.
6824 */
6825 ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
6826 BUG_ON(ret < 0);
6827
6828 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
6829 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
6830 0, 0);
6831 BUG_ON(ret);
6832 extent_buffer_get(path->nodes[0]);
6833 eb = path->nodes[0];
6834 btrfs_release_path(reloc_root, path);
6835 ret = invalidate_extent_cache(reloc_root, eb, group, root);
6836 BUG_ON(ret);
6837 free_extent_buffer(eb);
6838 }
6839
6840 mutex_unlock(&root->fs_info->tree_reloc_mutex);
6841 path->lowest_level = 0;
6842 return 0;
6843 }
6844
6845 static noinline int relocate_tree_block(struct btrfs_trans_handle *trans,
6846 struct btrfs_root *root,
6847 struct btrfs_path *path,
6848 struct btrfs_key *first_key,
6849 struct btrfs_ref_path *ref_path)
6850 {
6851 int ret;
6852
6853 ret = relocate_one_path(trans, root, path, first_key,
6854 ref_path, NULL, NULL);
6855 BUG_ON(ret);
6856
6857 return 0;
6858 }
6859
6860 static noinline int del_extent_zero(struct btrfs_trans_handle *trans,
6861 struct btrfs_root *extent_root,
6862 struct btrfs_path *path,
6863 struct btrfs_key *extent_key)
6864 {
6865 int ret;
6866
6867 ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
6868 if (ret)
6869 goto out;
6870 ret = btrfs_del_item(trans, extent_root, path);
6871 out:
6872 btrfs_release_path(extent_root, path);
6873 return ret;
6874 }
6875
6876 static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info,
6877 struct btrfs_ref_path *ref_path)
6878 {
6879 struct btrfs_key root_key;
6880
6881 root_key.objectid = ref_path->root_objectid;
6882 root_key.type = BTRFS_ROOT_ITEM_KEY;
6883 if (is_cowonly_root(ref_path->root_objectid))
6884 root_key.offset = 0;
6885 else
6886 root_key.offset = (u64)-1;
6887
6888 return btrfs_read_fs_root_no_name(fs_info, &root_key);
6889 }
6890
6891 static noinline int relocate_one_extent(struct btrfs_root *extent_root,
6892 struct btrfs_path *path,
6893 struct btrfs_key *extent_key,
6894 struct btrfs_block_group_cache *group,
6895 struct inode *reloc_inode, int pass)
6896 {
6897 struct btrfs_trans_handle *trans;
6898 struct btrfs_root *found_root;
6899 struct btrfs_ref_path *ref_path = NULL;
6900 struct disk_extent *new_extents = NULL;
6901 int nr_extents = 0;
6902 int loops;
6903 int ret;
6904 int level;
6905 struct btrfs_key first_key;
6906 u64 prev_block = 0;
6907
6908
6909 trans = btrfs_start_transaction(extent_root, 1);
6910 BUG_ON(!trans);
6911
6912 if (extent_key->objectid == 0) {
6913 ret = del_extent_zero(trans, extent_root, path, extent_key);
6914 goto out;
6915 }
6916
6917 ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
6918 if (!ref_path) {
6919 ret = -ENOMEM;
6920 goto out;
6921 }
6922
6923 for (loops = 0; ; loops++) {
6924 if (loops == 0) {
6925 ret = btrfs_first_ref_path(trans, extent_root, ref_path,
6926 extent_key->objectid);
6927 } else {
6928 ret = btrfs_next_ref_path(trans, extent_root, ref_path);
6929 }
6930 if (ret < 0)
6931 goto out;
6932 if (ret > 0)
6933 break;
6934
6935 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
6936 ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
6937 continue;
6938
6939 found_root = read_ref_root(extent_root->fs_info, ref_path);
6940 BUG_ON(!found_root);
6941 /*
6942 * for reference counted tree, only process reference paths
6943 * rooted at the latest committed root.
6944 */
6945 if (found_root->ref_cows &&
6946 ref_path->root_generation != found_root->root_key.offset)
6947 continue;
6948
6949 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
6950 if (pass == 0) {
6951 /*
6952 * copy data extents to new locations
6953 */
6954 u64 group_start = group->key.objectid;
6955 ret = relocate_data_extent(reloc_inode,
6956 extent_key,
6957 group_start);
6958 if (ret < 0)
6959 goto out;
6960 break;
6961 }
6962 level = 0;
6963 } else {
6964 level = ref_path->owner_objectid;
6965 }
6966
6967 if (prev_block != ref_path->nodes[level]) {
6968 struct extent_buffer *eb;
6969 u64 block_start = ref_path->nodes[level];
6970 u64 block_size = btrfs_level_size(found_root, level);
6971
6972 eb = read_tree_block(found_root, block_start,
6973 block_size, 0);
6974 btrfs_tree_lock(eb);
6975 BUG_ON(level != btrfs_header_level(eb));
6976
6977 if (level == 0)
6978 btrfs_item_key_to_cpu(eb, &first_key, 0);
6979 else
6980 btrfs_node_key_to_cpu(eb, &first_key, 0);
6981
6982 btrfs_tree_unlock(eb);
6983 free_extent_buffer(eb);
6984 prev_block = block_start;
6985 }
6986
6987 mutex_lock(&extent_root->fs_info->trans_mutex);
6988 btrfs_record_root_in_trans(found_root);
6989 mutex_unlock(&extent_root->fs_info->trans_mutex);
6990 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
6991 /*
6992 * try to update data extent references while
6993 * keeping metadata shared between snapshots.
6994 */
6995 if (pass == 1) {
6996 ret = relocate_one_path(trans, found_root,
6997 path, &first_key, ref_path,
6998 group, reloc_inode);
6999 if (ret < 0)
7000 goto out;
7001 continue;
7002 }
7003 /*
7004 * use fallback method to process the remaining
7005 * references.
7006 */
7007 if (!new_extents) {
7008 u64 group_start = group->key.objectid;
7009 new_extents = kmalloc(sizeof(*new_extents),
7010 GFP_NOFS);
7011 nr_extents = 1;
7012 ret = get_new_locations(reloc_inode,
7013 extent_key,
7014 group_start, 1,
7015 &new_extents,
7016 &nr_extents);
7017 if (ret)
7018 goto out;
7019 }
7020 ret = replace_one_extent(trans, found_root,
7021 path, extent_key,
7022 &first_key, ref_path,
7023 new_extents, nr_extents);
7024 } else {
7025 ret = relocate_tree_block(trans, found_root, path,
7026 &first_key, ref_path);
7027 }
7028 if (ret < 0)
7029 goto out;
7030 }
7031 ret = 0;
7032 out:
7033 btrfs_end_transaction(trans, extent_root);
7034 kfree(new_extents);
7035 kfree(ref_path);
7036 return ret;
7037 }
7038 #endif
7039
7040 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7041 {
7042 u64 num_devices;
7043 u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
7044 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7045
7046 num_devices = root->fs_info->fs_devices->rw_devices;
7047 if (num_devices == 1) {
7048 stripped |= BTRFS_BLOCK_GROUP_DUP;
7049 stripped = flags & ~stripped;
7050
7051 /* turn raid0 into single device chunks */
7052 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7053 return stripped;
7054
7055 /* turn mirroring into duplication */
7056 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7057 BTRFS_BLOCK_GROUP_RAID10))
7058 return stripped | BTRFS_BLOCK_GROUP_DUP;
7059 return flags;
7060 } else {
7061 /* they already had raid on here, just return */
7062 if (flags & stripped)
7063 return flags;
7064
7065 stripped |= BTRFS_BLOCK_GROUP_DUP;
7066 stripped = flags & ~stripped;
7067
7068 /* switch duplicated blocks with raid1 */
7069 if (flags & BTRFS_BLOCK_GROUP_DUP)
7070 return stripped | BTRFS_BLOCK_GROUP_RAID1;
7071
7072 /* turn single device chunks into raid0 */
7073 return stripped | BTRFS_BLOCK_GROUP_RAID0;
7074 }
7075 return flags;
7076 }
7077
7078 static int __alloc_chunk_for_shrink(struct btrfs_root *root,
7079 struct btrfs_block_group_cache *shrink_block_group,
7080 int force)
7081 {
7082 struct btrfs_trans_handle *trans;
7083 u64 new_alloc_flags;
7084 u64 calc;
7085
7086 spin_lock(&shrink_block_group->lock);
7087 if (btrfs_block_group_used(&shrink_block_group->item) +
7088 shrink_block_group->reserved > 0) {
7089 spin_unlock(&shrink_block_group->lock);
7090
7091 trans = btrfs_start_transaction(root, 1);
7092 spin_lock(&shrink_block_group->lock);
7093
7094 new_alloc_flags = update_block_group_flags(root,
7095 shrink_block_group->flags);
7096 if (new_alloc_flags != shrink_block_group->flags) {
7097 calc =
7098 btrfs_block_group_used(&shrink_block_group->item);
7099 } else {
7100 calc = shrink_block_group->key.offset;
7101 }
7102 spin_unlock(&shrink_block_group->lock);
7103
7104 do_chunk_alloc(trans, root->fs_info->extent_root,
7105 calc + 2 * 1024 * 1024, new_alloc_flags, force);
7106
7107 btrfs_end_transaction(trans, root);
7108 } else
7109 spin_unlock(&shrink_block_group->lock);
7110 return 0;
7111 }
7112
7113
7114 int btrfs_prepare_block_group_relocation(struct btrfs_root *root,
7115 struct btrfs_block_group_cache *group)
7116
7117 {
7118 __alloc_chunk_for_shrink(root, group, 1);
7119 set_block_group_readonly(group);
7120 return 0;
7121 }
7122
7123 #if 0
7124 static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
7125 struct btrfs_root *root,
7126 u64 objectid, u64 size)
7127 {
7128 struct btrfs_path *path;
7129 struct btrfs_inode_item *item;
7130 struct extent_buffer *leaf;
7131 int ret;
7132
7133 path = btrfs_alloc_path();
7134 if (!path)
7135 return -ENOMEM;
7136
7137 path->leave_spinning = 1;
7138 ret = btrfs_insert_empty_inode(trans, root, path, objectid);
7139 if (ret)
7140 goto out;
7141
7142 leaf = path->nodes[0];
7143 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
7144 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
7145 btrfs_set_inode_generation(leaf, item, 1);
7146 btrfs_set_inode_size(leaf, item, size);
7147 btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
7148 btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS);
7149 btrfs_mark_buffer_dirty(leaf);
7150 btrfs_release_path(root, path);
7151 out:
7152 btrfs_free_path(path);
7153 return ret;
7154 }
7155
7156 static noinline struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info,
7157 struct btrfs_block_group_cache *group)
7158 {
7159 struct inode *inode = NULL;
7160 struct btrfs_trans_handle *trans;
7161 struct btrfs_root *root;
7162 struct btrfs_key root_key;
7163 u64 objectid = BTRFS_FIRST_FREE_OBJECTID;
7164 int err = 0;
7165
7166 root_key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
7167 root_key.type = BTRFS_ROOT_ITEM_KEY;
7168 root_key.offset = (u64)-1;
7169 root = btrfs_read_fs_root_no_name(fs_info, &root_key);
7170 if (IS_ERR(root))
7171 return ERR_CAST(root);
7172
7173 trans = btrfs_start_transaction(root, 1);
7174 BUG_ON(!trans);
7175
7176 err = btrfs_find_free_objectid(trans, root, objectid, &objectid);
7177 if (err)
7178 goto out;
7179
7180 err = __insert_orphan_inode(trans, root, objectid, group->key.offset);
7181 BUG_ON(err);
7182
7183 err = btrfs_insert_file_extent(trans, root, objectid, 0, 0, 0,
7184 group->key.offset, 0, group->key.offset,
7185 0, 0, 0);
7186 BUG_ON(err);
7187
7188 inode = btrfs_iget_locked(root->fs_info->sb, objectid, root);
7189 if (inode->i_state & I_NEW) {
7190 BTRFS_I(inode)->root = root;
7191 BTRFS_I(inode)->location.objectid = objectid;
7192 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
7193 BTRFS_I(inode)->location.offset = 0;
7194 btrfs_read_locked_inode(inode);
7195 unlock_new_inode(inode);
7196 BUG_ON(is_bad_inode(inode));
7197 } else {
7198 BUG_ON(1);
7199 }
7200 BTRFS_I(inode)->index_cnt = group->key.objectid;
7201
7202 err = btrfs_orphan_add(trans, inode);
7203 out:
7204 btrfs_end_transaction(trans, root);
7205 if (err) {
7206 if (inode)
7207 iput(inode);
7208 inode = ERR_PTR(err);
7209 }
7210 return inode;
7211 }
7212
7213 int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len)
7214 {
7215
7216 struct btrfs_ordered_sum *sums;
7217 struct btrfs_sector_sum *sector_sum;
7218 struct btrfs_ordered_extent *ordered;
7219 struct btrfs_root *root = BTRFS_I(inode)->root;
7220 struct list_head list;
7221 size_t offset;
7222 int ret;
7223 u64 disk_bytenr;
7224
7225 INIT_LIST_HEAD(&list);
7226
7227 ordered = btrfs_lookup_ordered_extent(inode, file_pos);
7228 BUG_ON(ordered->file_offset != file_pos || ordered->len != len);
7229
7230 disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt;
7231 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, disk_bytenr,
7232 disk_bytenr + len - 1, &list);
7233
7234 while (!list_empty(&list)) {
7235 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
7236 list_del_init(&sums->list);
7237
7238 sector_sum = sums->sums;
7239 sums->bytenr = ordered->start;
7240
7241 offset = 0;
7242 while (offset < sums->len) {
7243 sector_sum->bytenr += ordered->start - disk_bytenr;
7244 sector_sum++;
7245 offset += root->sectorsize;
7246 }
7247
7248 btrfs_add_ordered_sum(inode, ordered, sums);
7249 }
7250 btrfs_put_ordered_extent(ordered);
7251 return 0;
7252 }
7253
7254 int btrfs_relocate_block_group(struct btrfs_root *root, u64 group_start)
7255 {
7256 struct btrfs_trans_handle *trans;
7257 struct btrfs_path *path;
7258 struct btrfs_fs_info *info = root->fs_info;
7259 struct extent_buffer *leaf;
7260 struct inode *reloc_inode;
7261 struct btrfs_block_group_cache *block_group;
7262 struct btrfs_key key;
7263 u64 skipped;
7264 u64 cur_byte;
7265 u64 total_found;
7266 u32 nritems;
7267 int ret;
7268 int progress;
7269 int pass = 0;
7270
7271 root = root->fs_info->extent_root;
7272
7273 block_group = btrfs_lookup_block_group(info, group_start);
7274 BUG_ON(!block_group);
7275
7276 printk(KERN_INFO "btrfs relocating block group %llu flags %llu\n",
7277 (unsigned long long)block_group->key.objectid,
7278 (unsigned long long)block_group->flags);
7279
7280 path = btrfs_alloc_path();
7281 BUG_ON(!path);
7282
7283 reloc_inode = create_reloc_inode(info, block_group);
7284 BUG_ON(IS_ERR(reloc_inode));
7285
7286 __alloc_chunk_for_shrink(root, block_group, 1);
7287 set_block_group_readonly(block_group);
7288
7289 btrfs_start_delalloc_inodes(info->tree_root);
7290 btrfs_wait_ordered_extents(info->tree_root, 0);
7291 again:
7292 skipped = 0;
7293 total_found = 0;
7294 progress = 0;
7295 key.objectid = block_group->key.objectid;
7296 key.offset = 0;
7297 key.type = 0;
7298 cur_byte = key.objectid;
7299
7300 trans = btrfs_start_transaction(info->tree_root, 1);
7301 btrfs_commit_transaction(trans, info->tree_root);
7302
7303 mutex_lock(&root->fs_info->cleaner_mutex);
7304 btrfs_clean_old_snapshots(info->tree_root);
7305 btrfs_remove_leaf_refs(info->tree_root, (u64)-1, 1);
7306 mutex_unlock(&root->fs_info->cleaner_mutex);
7307
7308 trans = btrfs_start_transaction(info->tree_root, 1);
7309 btrfs_commit_transaction(trans, info->tree_root);
7310
7311 while (1) {
7312 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7313 if (ret < 0)
7314 goto out;
7315 next:
7316 leaf = path->nodes[0];
7317 nritems = btrfs_header_nritems(leaf);
7318 if (path->slots[0] >= nritems) {
7319 ret = btrfs_next_leaf(root, path);
7320 if (ret < 0)
7321 goto out;
7322 if (ret == 1) {
7323 ret = 0;
7324 break;
7325 }
7326 leaf = path->nodes[0];
7327 nritems = btrfs_header_nritems(leaf);
7328 }
7329
7330 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
7331
7332 if (key.objectid >= block_group->key.objectid +
7333 block_group->key.offset)
7334 break;
7335
7336 if (progress && need_resched()) {
7337 btrfs_release_path(root, path);
7338 cond_resched();
7339 progress = 0;
7340 continue;
7341 }
7342 progress = 1;
7343
7344 if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY ||
7345 key.objectid + key.offset <= cur_byte) {
7346 path->slots[0]++;
7347 goto next;
7348 }
7349
7350 total_found++;
7351 cur_byte = key.objectid + key.offset;
7352 btrfs_release_path(root, path);
7353
7354 __alloc_chunk_for_shrink(root, block_group, 0);
7355 ret = relocate_one_extent(root, path, &key, block_group,
7356 reloc_inode, pass);
7357 BUG_ON(ret < 0);
7358 if (ret > 0)
7359 skipped++;
7360
7361 key.objectid = cur_byte;
7362 key.type = 0;
7363 key.offset = 0;
7364 }
7365
7366 btrfs_release_path(root, path);
7367
7368 if (pass == 0) {
7369 btrfs_wait_ordered_range(reloc_inode, 0, (u64)-1);
7370 invalidate_mapping_pages(reloc_inode->i_mapping, 0, -1);
7371 }
7372
7373 if (total_found > 0) {
7374 printk(KERN_INFO "btrfs found %llu extents in pass %d\n",
7375 (unsigned long long)total_found, pass);
7376 pass++;
7377 if (total_found == skipped && pass > 2) {
7378 iput(reloc_inode);
7379 reloc_inode = create_reloc_inode(info, block_group);
7380 pass = 0;
7381 }
7382 goto again;
7383 }
7384
7385 /* delete reloc_inode */
7386 iput(reloc_inode);
7387
7388 /* unpin extents in this range */
7389 trans = btrfs_start_transaction(info->tree_root, 1);
7390 btrfs_commit_transaction(trans, info->tree_root);
7391
7392 spin_lock(&block_group->lock);
7393 WARN_ON(block_group->pinned > 0);
7394 WARN_ON(block_group->reserved > 0);
7395 WARN_ON(btrfs_block_group_used(&block_group->item) > 0);
7396 spin_unlock(&block_group->lock);
7397 btrfs_put_block_group(block_group);
7398 ret = 0;
7399 out:
7400 btrfs_free_path(path);
7401 return ret;
7402 }
7403 #endif
7404
7405 static int find_first_block_group(struct btrfs_root *root,
7406 struct btrfs_path *path, struct btrfs_key *key)
7407 {
7408 int ret = 0;
7409 struct btrfs_key found_key;
7410 struct extent_buffer *leaf;
7411 int slot;
7412
7413 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
7414 if (ret < 0)
7415 goto out;
7416
7417 while (1) {
7418 slot = path->slots[0];
7419 leaf = path->nodes[0];
7420 if (slot >= btrfs_header_nritems(leaf)) {
7421 ret = btrfs_next_leaf(root, path);
7422 if (ret == 0)
7423 continue;
7424 if (ret < 0)
7425 goto out;
7426 break;
7427 }
7428 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7429
7430 if (found_key.objectid >= key->objectid &&
7431 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
7432 ret = 0;
7433 goto out;
7434 }
7435 path->slots[0]++;
7436 }
7437 ret = -ENOENT;
7438 out:
7439 return ret;
7440 }
7441
7442 int btrfs_free_block_groups(struct btrfs_fs_info *info)
7443 {
7444 struct btrfs_block_group_cache *block_group;
7445 struct btrfs_space_info *space_info;
7446 struct btrfs_caching_control *caching_ctl;
7447 struct rb_node *n;
7448
7449 down_write(&info->extent_commit_sem);
7450 while (!list_empty(&info->caching_block_groups)) {
7451 caching_ctl = list_entry(info->caching_block_groups.next,
7452 struct btrfs_caching_control, list);
7453 list_del(&caching_ctl->list);
7454 put_caching_control(caching_ctl);
7455 }
7456 up_write(&info->extent_commit_sem);
7457
7458 spin_lock(&info->block_group_cache_lock);
7459 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
7460 block_group = rb_entry(n, struct btrfs_block_group_cache,
7461 cache_node);
7462 rb_erase(&block_group->cache_node,
7463 &info->block_group_cache_tree);
7464 spin_unlock(&info->block_group_cache_lock);
7465
7466 down_write(&block_group->space_info->groups_sem);
7467 list_del(&block_group->list);
7468 up_write(&block_group->space_info->groups_sem);
7469
7470 if (block_group->cached == BTRFS_CACHE_STARTED)
7471 wait_block_group_cache_done(block_group);
7472
7473 btrfs_remove_free_space_cache(block_group);
7474
7475 WARN_ON(atomic_read(&block_group->count) != 1);
7476 kfree(block_group);
7477
7478 spin_lock(&info->block_group_cache_lock);
7479 }
7480 spin_unlock(&info->block_group_cache_lock);
7481
7482 /* now that all the block groups are freed, go through and
7483 * free all the space_info structs. This is only called during
7484 * the final stages of unmount, and so we know nobody is
7485 * using them. We call synchronize_rcu() once before we start,
7486 * just to be on the safe side.
7487 */
7488 synchronize_rcu();
7489
7490 while(!list_empty(&info->space_info)) {
7491 space_info = list_entry(info->space_info.next,
7492 struct btrfs_space_info,
7493 list);
7494
7495 list_del(&space_info->list);
7496 kfree(space_info);
7497 }
7498 return 0;
7499 }
7500
7501 int btrfs_read_block_groups(struct btrfs_root *root)
7502 {
7503 struct btrfs_path *path;
7504 int ret;
7505 struct btrfs_block_group_cache *cache;
7506 struct btrfs_fs_info *info = root->fs_info;
7507 struct btrfs_space_info *space_info;
7508 struct btrfs_key key;
7509 struct btrfs_key found_key;
7510 struct extent_buffer *leaf;
7511
7512 root = info->extent_root;
7513 key.objectid = 0;
7514 key.offset = 0;
7515 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
7516 path = btrfs_alloc_path();
7517 if (!path)
7518 return -ENOMEM;
7519
7520 while (1) {
7521 ret = find_first_block_group(root, path, &key);
7522 if (ret > 0) {
7523 ret = 0;
7524 goto error;
7525 }
7526 if (ret != 0)
7527 goto error;
7528
7529 leaf = path->nodes[0];
7530 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7531 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7532 if (!cache) {
7533 ret = -ENOMEM;
7534 break;
7535 }
7536
7537 atomic_set(&cache->count, 1);
7538 spin_lock_init(&cache->lock);
7539 spin_lock_init(&cache->tree_lock);
7540 cache->fs_info = info;
7541 INIT_LIST_HEAD(&cache->list);
7542 INIT_LIST_HEAD(&cache->cluster_list);
7543
7544 /*
7545 * we only want to have 32k of ram per block group for keeping
7546 * track of free space, and if we pass 1/2 of that we want to
7547 * start converting things over to using bitmaps
7548 */
7549 cache->extents_thresh = ((1024 * 32) / 2) /
7550 sizeof(struct btrfs_free_space);
7551
7552 read_extent_buffer(leaf, &cache->item,
7553 btrfs_item_ptr_offset(leaf, path->slots[0]),
7554 sizeof(cache->item));
7555 memcpy(&cache->key, &found_key, sizeof(found_key));
7556
7557 key.objectid = found_key.objectid + found_key.offset;
7558 btrfs_release_path(root, path);
7559 cache->flags = btrfs_block_group_flags(&cache->item);
7560 cache->sectorsize = root->sectorsize;
7561
7562 /*
7563 * check for two cases, either we are full, and therefore
7564 * don't need to bother with the caching work since we won't
7565 * find any space, or we are empty, and we can just add all
7566 * the space in and be done with it. This saves us _alot_ of
7567 * time, particularly in the full case.
7568 */
7569 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
7570 cache->last_byte_to_unpin = (u64)-1;
7571 cache->cached = BTRFS_CACHE_FINISHED;
7572 } else if (btrfs_block_group_used(&cache->item) == 0) {
7573 exclude_super_stripes(root, cache);
7574 cache->last_byte_to_unpin = (u64)-1;
7575 cache->cached = BTRFS_CACHE_FINISHED;
7576 add_new_free_space(cache, root->fs_info,
7577 found_key.objectid,
7578 found_key.objectid +
7579 found_key.offset);
7580 free_excluded_extents(root, cache);
7581 }
7582
7583 ret = update_space_info(info, cache->flags, found_key.offset,
7584 btrfs_block_group_used(&cache->item),
7585 &space_info);
7586 BUG_ON(ret);
7587 cache->space_info = space_info;
7588 down_write(&space_info->groups_sem);
7589 list_add_tail(&cache->list, &space_info->block_groups);
7590 up_write(&space_info->groups_sem);
7591
7592 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7593 BUG_ON(ret);
7594
7595 set_avail_alloc_bits(root->fs_info, cache->flags);
7596 if (btrfs_chunk_readonly(root, cache->key.objectid))
7597 set_block_group_readonly(cache);
7598 }
7599 ret = 0;
7600 error:
7601 btrfs_free_path(path);
7602 return ret;
7603 }
7604
7605 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
7606 struct btrfs_root *root, u64 bytes_used,
7607 u64 type, u64 chunk_objectid, u64 chunk_offset,
7608 u64 size)
7609 {
7610 int ret;
7611 struct btrfs_root *extent_root;
7612 struct btrfs_block_group_cache *cache;
7613
7614 extent_root = root->fs_info->extent_root;
7615
7616 root->fs_info->last_trans_log_full_commit = trans->transid;
7617
7618 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7619 if (!cache)
7620 return -ENOMEM;
7621
7622 cache->key.objectid = chunk_offset;
7623 cache->key.offset = size;
7624 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
7625 cache->sectorsize = root->sectorsize;
7626
7627 /*
7628 * we only want to have 32k of ram per block group for keeping track
7629 * of free space, and if we pass 1/2 of that we want to start
7630 * converting things over to using bitmaps
7631 */
7632 cache->extents_thresh = ((1024 * 32) / 2) /
7633 sizeof(struct btrfs_free_space);
7634 atomic_set(&cache->count, 1);
7635 spin_lock_init(&cache->lock);
7636 spin_lock_init(&cache->tree_lock);
7637 INIT_LIST_HEAD(&cache->list);
7638 INIT_LIST_HEAD(&cache->cluster_list);
7639
7640 btrfs_set_block_group_used(&cache->item, bytes_used);
7641 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
7642 cache->flags = type;
7643 btrfs_set_block_group_flags(&cache->item, type);
7644
7645 cache->last_byte_to_unpin = (u64)-1;
7646 cache->cached = BTRFS_CACHE_FINISHED;
7647 exclude_super_stripes(root, cache);
7648
7649 add_new_free_space(cache, root->fs_info, chunk_offset,
7650 chunk_offset + size);
7651
7652 free_excluded_extents(root, cache);
7653
7654 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
7655 &cache->space_info);
7656 BUG_ON(ret);
7657 down_write(&cache->space_info->groups_sem);
7658 list_add_tail(&cache->list, &cache->space_info->block_groups);
7659 up_write(&cache->space_info->groups_sem);
7660
7661 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7662 BUG_ON(ret);
7663
7664 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
7665 sizeof(cache->item));
7666 BUG_ON(ret);
7667
7668 set_avail_alloc_bits(extent_root->fs_info, type);
7669
7670 return 0;
7671 }
7672
7673 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
7674 struct btrfs_root *root, u64 group_start)
7675 {
7676 struct btrfs_path *path;
7677 struct btrfs_block_group_cache *block_group;
7678 struct btrfs_free_cluster *cluster;
7679 struct btrfs_key key;
7680 int ret;
7681
7682 root = root->fs_info->extent_root;
7683
7684 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
7685 BUG_ON(!block_group);
7686 BUG_ON(!block_group->ro);
7687
7688 memcpy(&key, &block_group->key, sizeof(key));
7689
7690 /* make sure this block group isn't part of an allocation cluster */
7691 cluster = &root->fs_info->data_alloc_cluster;
7692 spin_lock(&cluster->refill_lock);
7693 btrfs_return_cluster_to_free_space(block_group, cluster);
7694 spin_unlock(&cluster->refill_lock);
7695
7696 /*
7697 * make sure this block group isn't part of a metadata
7698 * allocation cluster
7699 */
7700 cluster = &root->fs_info->meta_alloc_cluster;
7701 spin_lock(&cluster->refill_lock);
7702 btrfs_return_cluster_to_free_space(block_group, cluster);
7703 spin_unlock(&cluster->refill_lock);
7704
7705 path = btrfs_alloc_path();
7706 BUG_ON(!path);
7707
7708 spin_lock(&root->fs_info->block_group_cache_lock);
7709 rb_erase(&block_group->cache_node,
7710 &root->fs_info->block_group_cache_tree);
7711 spin_unlock(&root->fs_info->block_group_cache_lock);
7712
7713 down_write(&block_group->space_info->groups_sem);
7714 /*
7715 * we must use list_del_init so people can check to see if they
7716 * are still on the list after taking the semaphore
7717 */
7718 list_del_init(&block_group->list);
7719 up_write(&block_group->space_info->groups_sem);
7720
7721 if (block_group->cached == BTRFS_CACHE_STARTED)
7722 wait_block_group_cache_done(block_group);
7723
7724 btrfs_remove_free_space_cache(block_group);
7725
7726 spin_lock(&block_group->space_info->lock);
7727 block_group->space_info->total_bytes -= block_group->key.offset;
7728 block_group->space_info->bytes_readonly -= block_group->key.offset;
7729 spin_unlock(&block_group->space_info->lock);
7730
7731 btrfs_clear_space_info_full(root->fs_info);
7732
7733 btrfs_put_block_group(block_group);
7734 btrfs_put_block_group(block_group);
7735
7736 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
7737 if (ret > 0)
7738 ret = -EIO;
7739 if (ret < 0)
7740 goto out;
7741
7742 ret = btrfs_del_item(trans, root, path);
7743 out:
7744 btrfs_free_path(path);
7745 return ret;
7746 }
Linux kernel - Deliverables
This page took 0.494159 seconds and 5 git commands to generate.