Merge git://git.kernel.org/pub/scm/linux/kernel/git/mason/btrfs-unstable
[pandora-kernel.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 <linux/slab.h>
26 #include "compat.h"
27 #include "hash.h"
28 #include "ctree.h"
29 #include "disk-io.h"
30 #include "print-tree.h"
31 #include "transaction.h"
32 #include "volumes.h"
33 #include "locking.h"
34 #include "free-space-cache.h"
35
36 static int update_block_group(struct btrfs_trans_handle *trans,
37                               struct btrfs_root *root,
38                               u64 bytenr, u64 num_bytes, int alloc);
39 static int update_reserved_bytes(struct btrfs_block_group_cache *cache,
40                                  u64 num_bytes, int reserve, int sinfo);
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 find_next_key(struct btrfs_path *path, int level,
64                          struct btrfs_key *key);
65 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
66                             int dump_block_groups);
67
68 static noinline int
69 block_group_cache_done(struct btrfs_block_group_cache *cache)
70 {
71         smp_mb();
72         return cache->cached == BTRFS_CACHE_FINISHED;
73 }
74
75 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
76 {
77         return (cache->flags & bits) == bits;
78 }
79
80 void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
81 {
82         atomic_inc(&cache->count);
83 }
84
85 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
86 {
87         if (atomic_dec_and_test(&cache->count)) {
88                 WARN_ON(cache->pinned > 0);
89                 WARN_ON(cache->reserved > 0);
90                 WARN_ON(cache->reserved_pinned > 0);
91                 kfree(cache);
92         }
93 }
94
95 /*
96  * this adds the block group to the fs_info rb tree for the block group
97  * cache
98  */
99 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
100                                 struct btrfs_block_group_cache *block_group)
101 {
102         struct rb_node **p;
103         struct rb_node *parent = NULL;
104         struct btrfs_block_group_cache *cache;
105
106         spin_lock(&info->block_group_cache_lock);
107         p = &info->block_group_cache_tree.rb_node;
108
109         while (*p) {
110                 parent = *p;
111                 cache = rb_entry(parent, struct btrfs_block_group_cache,
112                                  cache_node);
113                 if (block_group->key.objectid < cache->key.objectid) {
114                         p = &(*p)->rb_left;
115                 } else if (block_group->key.objectid > cache->key.objectid) {
116                         p = &(*p)->rb_right;
117                 } else {
118                         spin_unlock(&info->block_group_cache_lock);
119                         return -EEXIST;
120                 }
121         }
122
123         rb_link_node(&block_group->cache_node, parent, p);
124         rb_insert_color(&block_group->cache_node,
125                         &info->block_group_cache_tree);
126         spin_unlock(&info->block_group_cache_lock);
127
128         return 0;
129 }
130
131 /*
132  * This will return the block group at or after bytenr if contains is 0, else
133  * it will return the block group that contains the bytenr
134  */
135 static struct btrfs_block_group_cache *
136 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
137                               int contains)
138 {
139         struct btrfs_block_group_cache *cache, *ret = NULL;
140         struct rb_node *n;
141         u64 end, start;
142
143         spin_lock(&info->block_group_cache_lock);
144         n = info->block_group_cache_tree.rb_node;
145
146         while (n) {
147                 cache = rb_entry(n, struct btrfs_block_group_cache,
148                                  cache_node);
149                 end = cache->key.objectid + cache->key.offset - 1;
150                 start = cache->key.objectid;
151
152                 if (bytenr < start) {
153                         if (!contains && (!ret || start < ret->key.objectid))
154                                 ret = cache;
155                         n = n->rb_left;
156                 } else if (bytenr > start) {
157                         if (contains && bytenr <= end) {
158                                 ret = cache;
159                                 break;
160                         }
161                         n = n->rb_right;
162                 } else {
163                         ret = cache;
164                         break;
165                 }
166         }
167         if (ret)
168                 btrfs_get_block_group(ret);
169         spin_unlock(&info->block_group_cache_lock);
170
171         return ret;
172 }
173
174 static int add_excluded_extent(struct btrfs_root *root,
175                                u64 start, u64 num_bytes)
176 {
177         u64 end = start + num_bytes - 1;
178         set_extent_bits(&root->fs_info->freed_extents[0],
179                         start, end, EXTENT_UPTODATE, GFP_NOFS);
180         set_extent_bits(&root->fs_info->freed_extents[1],
181                         start, end, EXTENT_UPTODATE, GFP_NOFS);
182         return 0;
183 }
184
185 static void free_excluded_extents(struct btrfs_root *root,
186                                   struct btrfs_block_group_cache *cache)
187 {
188         u64 start, end;
189
190         start = cache->key.objectid;
191         end = start + cache->key.offset - 1;
192
193         clear_extent_bits(&root->fs_info->freed_extents[0],
194                           start, end, EXTENT_UPTODATE, GFP_NOFS);
195         clear_extent_bits(&root->fs_info->freed_extents[1],
196                           start, end, EXTENT_UPTODATE, GFP_NOFS);
197 }
198
199 static int exclude_super_stripes(struct btrfs_root *root,
200                                  struct btrfs_block_group_cache *cache)
201 {
202         u64 bytenr;
203         u64 *logical;
204         int stripe_len;
205         int i, nr, ret;
206
207         if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
208                 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
209                 cache->bytes_super += stripe_len;
210                 ret = add_excluded_extent(root, cache->key.objectid,
211                                           stripe_len);
212                 BUG_ON(ret);
213         }
214
215         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
216                 bytenr = btrfs_sb_offset(i);
217                 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
218                                        cache->key.objectid, bytenr,
219                                        0, &logical, &nr, &stripe_len);
220                 BUG_ON(ret);
221
222                 while (nr--) {
223                         cache->bytes_super += stripe_len;
224                         ret = add_excluded_extent(root, logical[nr],
225                                                   stripe_len);
226                         BUG_ON(ret);
227                 }
228
229                 kfree(logical);
230         }
231         return 0;
232 }
233
234 static struct btrfs_caching_control *
235 get_caching_control(struct btrfs_block_group_cache *cache)
236 {
237         struct btrfs_caching_control *ctl;
238
239         spin_lock(&cache->lock);
240         if (cache->cached != BTRFS_CACHE_STARTED) {
241                 spin_unlock(&cache->lock);
242                 return NULL;
243         }
244
245         /* We're loading it the fast way, so we don't have a caching_ctl. */
246         if (!cache->caching_ctl) {
247                 spin_unlock(&cache->lock);
248                 return NULL;
249         }
250
251         ctl = cache->caching_ctl;
252         atomic_inc(&ctl->count);
253         spin_unlock(&cache->lock);
254         return ctl;
255 }
256
257 static void put_caching_control(struct btrfs_caching_control *ctl)
258 {
259         if (atomic_dec_and_test(&ctl->count))
260                 kfree(ctl);
261 }
262
263 /*
264  * this is only called by cache_block_group, since we could have freed extents
265  * we need to check the pinned_extents for any extents that can't be used yet
266  * since their free space will be released as soon as the transaction commits.
267  */
268 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
269                               struct btrfs_fs_info *info, u64 start, u64 end)
270 {
271         u64 extent_start, extent_end, size, total_added = 0;
272         int ret;
273
274         while (start < end) {
275                 ret = find_first_extent_bit(info->pinned_extents, start,
276                                             &extent_start, &extent_end,
277                                             EXTENT_DIRTY | EXTENT_UPTODATE);
278                 if (ret)
279                         break;
280
281                 if (extent_start <= start) {
282                         start = extent_end + 1;
283                 } else if (extent_start > start && extent_start < end) {
284                         size = extent_start - start;
285                         total_added += size;
286                         ret = btrfs_add_free_space(block_group, start,
287                                                    size);
288                         BUG_ON(ret);
289                         start = extent_end + 1;
290                 } else {
291                         break;
292                 }
293         }
294
295         if (start < end) {
296                 size = end - start;
297                 total_added += size;
298                 ret = btrfs_add_free_space(block_group, start, size);
299                 BUG_ON(ret);
300         }
301
302         return total_added;
303 }
304
305 static int caching_kthread(void *data)
306 {
307         struct btrfs_block_group_cache *block_group = data;
308         struct btrfs_fs_info *fs_info = block_group->fs_info;
309         struct btrfs_caching_control *caching_ctl = block_group->caching_ctl;
310         struct btrfs_root *extent_root = fs_info->extent_root;
311         struct btrfs_path *path;
312         struct extent_buffer *leaf;
313         struct btrfs_key key;
314         u64 total_found = 0;
315         u64 last = 0;
316         u32 nritems;
317         int ret = 0;
318
319         path = btrfs_alloc_path();
320         if (!path)
321                 return -ENOMEM;
322
323         last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
324
325         /*
326          * We don't want to deadlock with somebody trying to allocate a new
327          * extent for the extent root while also trying to search the extent
328          * root to add free space.  So we skip locking and search the commit
329          * root, since its read-only
330          */
331         path->skip_locking = 1;
332         path->search_commit_root = 1;
333         path->reada = 2;
334
335         key.objectid = last;
336         key.offset = 0;
337         key.type = BTRFS_EXTENT_ITEM_KEY;
338 again:
339         mutex_lock(&caching_ctl->mutex);
340         /* need to make sure the commit_root doesn't disappear */
341         down_read(&fs_info->extent_commit_sem);
342
343         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
344         if (ret < 0)
345                 goto err;
346
347         leaf = path->nodes[0];
348         nritems = btrfs_header_nritems(leaf);
349
350         while (1) {
351                 smp_mb();
352                 if (fs_info->closing > 1) {
353                         last = (u64)-1;
354                         break;
355                 }
356
357                 if (path->slots[0] < nritems) {
358                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
359                 } else {
360                         ret = find_next_key(path, 0, &key);
361                         if (ret)
362                                 break;
363
364                         caching_ctl->progress = last;
365                         btrfs_release_path(extent_root, path);
366                         up_read(&fs_info->extent_commit_sem);
367                         mutex_unlock(&caching_ctl->mutex);
368                         if (btrfs_transaction_in_commit(fs_info))
369                                 schedule_timeout(1);
370                         else
371                                 cond_resched();
372                         goto again;
373                 }
374
375                 if (key.objectid < block_group->key.objectid) {
376                         path->slots[0]++;
377                         continue;
378                 }
379
380                 if (key.objectid >= block_group->key.objectid +
381                     block_group->key.offset)
382                         break;
383
384                 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
385                         total_found += add_new_free_space(block_group,
386                                                           fs_info, last,
387                                                           key.objectid);
388                         last = key.objectid + key.offset;
389
390                         if (total_found > (1024 * 1024 * 2)) {
391                                 total_found = 0;
392                                 wake_up(&caching_ctl->wait);
393                         }
394                 }
395                 path->slots[0]++;
396         }
397         ret = 0;
398
399         total_found += add_new_free_space(block_group, fs_info, last,
400                                           block_group->key.objectid +
401                                           block_group->key.offset);
402         caching_ctl->progress = (u64)-1;
403
404         spin_lock(&block_group->lock);
405         block_group->caching_ctl = NULL;
406         block_group->cached = BTRFS_CACHE_FINISHED;
407         spin_unlock(&block_group->lock);
408
409 err:
410         btrfs_free_path(path);
411         up_read(&fs_info->extent_commit_sem);
412
413         free_excluded_extents(extent_root, block_group);
414
415         mutex_unlock(&caching_ctl->mutex);
416         wake_up(&caching_ctl->wait);
417
418         put_caching_control(caching_ctl);
419         atomic_dec(&block_group->space_info->caching_threads);
420         btrfs_put_block_group(block_group);
421
422         return 0;
423 }
424
425 static int cache_block_group(struct btrfs_block_group_cache *cache,
426                              struct btrfs_trans_handle *trans,
427                              struct btrfs_root *root,
428                              int load_cache_only)
429 {
430         struct btrfs_fs_info *fs_info = cache->fs_info;
431         struct btrfs_caching_control *caching_ctl;
432         struct task_struct *tsk;
433         int ret = 0;
434
435         smp_mb();
436         if (cache->cached != BTRFS_CACHE_NO)
437                 return 0;
438
439         /*
440          * We can't do the read from on-disk cache during a commit since we need
441          * to have the normal tree locking.  Also if we are currently trying to
442          * allocate blocks for the tree root we can't do the fast caching since
443          * we likely hold important locks.
444          */
445         if (!trans->transaction->in_commit &&
446             (root && root != root->fs_info->tree_root)) {
447                 spin_lock(&cache->lock);
448                 if (cache->cached != BTRFS_CACHE_NO) {
449                         spin_unlock(&cache->lock);
450                         return 0;
451                 }
452                 cache->cached = BTRFS_CACHE_STARTED;
453                 spin_unlock(&cache->lock);
454
455                 ret = load_free_space_cache(fs_info, cache);
456
457                 spin_lock(&cache->lock);
458                 if (ret == 1) {
459                         cache->cached = BTRFS_CACHE_FINISHED;
460                         cache->last_byte_to_unpin = (u64)-1;
461                 } else {
462                         cache->cached = BTRFS_CACHE_NO;
463                 }
464                 spin_unlock(&cache->lock);
465                 if (ret == 1) {
466                         free_excluded_extents(fs_info->extent_root, cache);
467                         return 0;
468                 }
469         }
470
471         if (load_cache_only)
472                 return 0;
473
474         caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_KERNEL);
475         BUG_ON(!caching_ctl);
476
477         INIT_LIST_HEAD(&caching_ctl->list);
478         mutex_init(&caching_ctl->mutex);
479         init_waitqueue_head(&caching_ctl->wait);
480         caching_ctl->block_group = cache;
481         caching_ctl->progress = cache->key.objectid;
482         /* one for caching kthread, one for caching block group list */
483         atomic_set(&caching_ctl->count, 2);
484
485         spin_lock(&cache->lock);
486         if (cache->cached != BTRFS_CACHE_NO) {
487                 spin_unlock(&cache->lock);
488                 kfree(caching_ctl);
489                 return 0;
490         }
491         cache->caching_ctl = caching_ctl;
492         cache->cached = BTRFS_CACHE_STARTED;
493         spin_unlock(&cache->lock);
494
495         down_write(&fs_info->extent_commit_sem);
496         list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
497         up_write(&fs_info->extent_commit_sem);
498
499         atomic_inc(&cache->space_info->caching_threads);
500         btrfs_get_block_group(cache);
501
502         tsk = kthread_run(caching_kthread, cache, "btrfs-cache-%llu\n",
503                           cache->key.objectid);
504         if (IS_ERR(tsk)) {
505                 ret = PTR_ERR(tsk);
506                 printk(KERN_ERR "error running thread %d\n", ret);
507                 BUG();
508         }
509
510         return ret;
511 }
512
513 /*
514  * return the block group that starts at or after bytenr
515  */
516 static struct btrfs_block_group_cache *
517 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
518 {
519         struct btrfs_block_group_cache *cache;
520
521         cache = block_group_cache_tree_search(info, bytenr, 0);
522
523         return cache;
524 }
525
526 /*
527  * return the block group that contains the given bytenr
528  */
529 struct btrfs_block_group_cache *btrfs_lookup_block_group(
530                                                  struct btrfs_fs_info *info,
531                                                  u64 bytenr)
532 {
533         struct btrfs_block_group_cache *cache;
534
535         cache = block_group_cache_tree_search(info, bytenr, 1);
536
537         return cache;
538 }
539
540 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
541                                                   u64 flags)
542 {
543         struct list_head *head = &info->space_info;
544         struct btrfs_space_info *found;
545
546         flags &= BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_SYSTEM |
547                  BTRFS_BLOCK_GROUP_METADATA;
548
549         rcu_read_lock();
550         list_for_each_entry_rcu(found, head, list) {
551                 if (found->flags & flags) {
552                         rcu_read_unlock();
553                         return found;
554                 }
555         }
556         rcu_read_unlock();
557         return NULL;
558 }
559
560 /*
561  * after adding space to the filesystem, we need to clear the full flags
562  * on all the space infos.
563  */
564 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
565 {
566         struct list_head *head = &info->space_info;
567         struct btrfs_space_info *found;
568
569         rcu_read_lock();
570         list_for_each_entry_rcu(found, head, list)
571                 found->full = 0;
572         rcu_read_unlock();
573 }
574
575 static u64 div_factor(u64 num, int factor)
576 {
577         if (factor == 10)
578                 return num;
579         num *= factor;
580         do_div(num, 10);
581         return num;
582 }
583
584 static u64 div_factor_fine(u64 num, int factor)
585 {
586         if (factor == 100)
587                 return num;
588         num *= factor;
589         do_div(num, 100);
590         return num;
591 }
592
593 u64 btrfs_find_block_group(struct btrfs_root *root,
594                            u64 search_start, u64 search_hint, int owner)
595 {
596         struct btrfs_block_group_cache *cache;
597         u64 used;
598         u64 last = max(search_hint, search_start);
599         u64 group_start = 0;
600         int full_search = 0;
601         int factor = 9;
602         int wrapped = 0;
603 again:
604         while (1) {
605                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
606                 if (!cache)
607                         break;
608
609                 spin_lock(&cache->lock);
610                 last = cache->key.objectid + cache->key.offset;
611                 used = btrfs_block_group_used(&cache->item);
612
613                 if ((full_search || !cache->ro) &&
614                     block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
615                         if (used + cache->pinned + cache->reserved <
616                             div_factor(cache->key.offset, factor)) {
617                                 group_start = cache->key.objectid;
618                                 spin_unlock(&cache->lock);
619                                 btrfs_put_block_group(cache);
620                                 goto found;
621                         }
622                 }
623                 spin_unlock(&cache->lock);
624                 btrfs_put_block_group(cache);
625                 cond_resched();
626         }
627         if (!wrapped) {
628                 last = search_start;
629                 wrapped = 1;
630                 goto again;
631         }
632         if (!full_search && factor < 10) {
633                 last = search_start;
634                 full_search = 1;
635                 factor = 10;
636                 goto again;
637         }
638 found:
639         return group_start;
640 }
641
642 /* simple helper to search for an existing extent at a given offset */
643 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
644 {
645         int ret;
646         struct btrfs_key key;
647         struct btrfs_path *path;
648
649         path = btrfs_alloc_path();
650         BUG_ON(!path);
651         key.objectid = start;
652         key.offset = len;
653         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
654         ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
655                                 0, 0);
656         btrfs_free_path(path);
657         return ret;
658 }
659
660 /*
661  * helper function to lookup reference count and flags of extent.
662  *
663  * the head node for delayed ref is used to store the sum of all the
664  * reference count modifications queued up in the rbtree. the head
665  * node may also store the extent flags to set. This way you can check
666  * to see what the reference count and extent flags would be if all of
667  * the delayed refs are not processed.
668  */
669 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
670                              struct btrfs_root *root, u64 bytenr,
671                              u64 num_bytes, u64 *refs, u64 *flags)
672 {
673         struct btrfs_delayed_ref_head *head;
674         struct btrfs_delayed_ref_root *delayed_refs;
675         struct btrfs_path *path;
676         struct btrfs_extent_item *ei;
677         struct extent_buffer *leaf;
678         struct btrfs_key key;
679         u32 item_size;
680         u64 num_refs;
681         u64 extent_flags;
682         int ret;
683
684         path = btrfs_alloc_path();
685         if (!path)
686                 return -ENOMEM;
687
688         key.objectid = bytenr;
689         key.type = BTRFS_EXTENT_ITEM_KEY;
690         key.offset = num_bytes;
691         if (!trans) {
692                 path->skip_locking = 1;
693                 path->search_commit_root = 1;
694         }
695 again:
696         ret = btrfs_search_slot(trans, root->fs_info->extent_root,
697                                 &key, path, 0, 0);
698         if (ret < 0)
699                 goto out_free;
700
701         if (ret == 0) {
702                 leaf = path->nodes[0];
703                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
704                 if (item_size >= sizeof(*ei)) {
705                         ei = btrfs_item_ptr(leaf, path->slots[0],
706                                             struct btrfs_extent_item);
707                         num_refs = btrfs_extent_refs(leaf, ei);
708                         extent_flags = btrfs_extent_flags(leaf, ei);
709                 } else {
710 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
711                         struct btrfs_extent_item_v0 *ei0;
712                         BUG_ON(item_size != sizeof(*ei0));
713                         ei0 = btrfs_item_ptr(leaf, path->slots[0],
714                                              struct btrfs_extent_item_v0);
715                         num_refs = btrfs_extent_refs_v0(leaf, ei0);
716                         /* FIXME: this isn't correct for data */
717                         extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
718 #else
719                         BUG();
720 #endif
721                 }
722                 BUG_ON(num_refs == 0);
723         } else {
724                 num_refs = 0;
725                 extent_flags = 0;
726                 ret = 0;
727         }
728
729         if (!trans)
730                 goto out;
731
732         delayed_refs = &trans->transaction->delayed_refs;
733         spin_lock(&delayed_refs->lock);
734         head = btrfs_find_delayed_ref_head(trans, bytenr);
735         if (head) {
736                 if (!mutex_trylock(&head->mutex)) {
737                         atomic_inc(&head->node.refs);
738                         spin_unlock(&delayed_refs->lock);
739
740                         btrfs_release_path(root->fs_info->extent_root, path);
741
742                         mutex_lock(&head->mutex);
743                         mutex_unlock(&head->mutex);
744                         btrfs_put_delayed_ref(&head->node);
745                         goto again;
746                 }
747                 if (head->extent_op && head->extent_op->update_flags)
748                         extent_flags |= head->extent_op->flags_to_set;
749                 else
750                         BUG_ON(num_refs == 0);
751
752                 num_refs += head->node.ref_mod;
753                 mutex_unlock(&head->mutex);
754         }
755         spin_unlock(&delayed_refs->lock);
756 out:
757         WARN_ON(num_refs == 0);
758         if (refs)
759                 *refs = num_refs;
760         if (flags)
761                 *flags = extent_flags;
762 out_free:
763         btrfs_free_path(path);
764         return ret;
765 }
766
767 /*
768  * Back reference rules.  Back refs have three main goals:
769  *
770  * 1) differentiate between all holders of references to an extent so that
771  *    when a reference is dropped we can make sure it was a valid reference
772  *    before freeing the extent.
773  *
774  * 2) Provide enough information to quickly find the holders of an extent
775  *    if we notice a given block is corrupted or bad.
776  *
777  * 3) Make it easy to migrate blocks for FS shrinking or storage pool
778  *    maintenance.  This is actually the same as #2, but with a slightly
779  *    different use case.
780  *
781  * There are two kinds of back refs. The implicit back refs is optimized
782  * for pointers in non-shared tree blocks. For a given pointer in a block,
783  * back refs of this kind provide information about the block's owner tree
784  * and the pointer's key. These information allow us to find the block by
785  * b-tree searching. The full back refs is for pointers in tree blocks not
786  * referenced by their owner trees. The location of tree block is recorded
787  * in the back refs. Actually the full back refs is generic, and can be
788  * used in all cases the implicit back refs is used. The major shortcoming
789  * of the full back refs is its overhead. Every time a tree block gets
790  * COWed, we have to update back refs entry for all pointers in it.
791  *
792  * For a newly allocated tree block, we use implicit back refs for
793  * pointers in it. This means most tree related operations only involve
794  * implicit back refs. For a tree block created in old transaction, the
795  * only way to drop a reference to it is COW it. So we can detect the
796  * event that tree block loses its owner tree's reference and do the
797  * back refs conversion.
798  *
799  * When a tree block is COW'd through a tree, there are four cases:
800  *
801  * The reference count of the block is one and the tree is the block's
802  * owner tree. Nothing to do in this case.
803  *
804  * The reference count of the block is one and the tree is not the
805  * block's owner tree. In this case, full back refs is used for pointers
806  * in the block. Remove these full back refs, add implicit back refs for
807  * every pointers in the new block.
808  *
809  * The reference count of the block is greater than one and the tree is
810  * the block's owner tree. In this case, implicit back refs is used for
811  * pointers in the block. Add full back refs for every pointers in the
812  * block, increase lower level extents' reference counts. The original
813  * implicit back refs are entailed to the new block.
814  *
815  * The reference count of the block is greater than one and the tree is
816  * not the block's owner tree. Add implicit back refs for every pointer in
817  * the new block, increase lower level extents' reference count.
818  *
819  * Back Reference Key composing:
820  *
821  * The key objectid corresponds to the first byte in the extent,
822  * The key type is used to differentiate between types of back refs.
823  * There are different meanings of the key offset for different types
824  * of back refs.
825  *
826  * File extents can be referenced by:
827  *
828  * - multiple snapshots, subvolumes, or different generations in one subvol
829  * - different files inside a single subvolume
830  * - different offsets inside a file (bookend extents in file.c)
831  *
832  * The extent ref structure for the implicit back refs has fields for:
833  *
834  * - Objectid of the subvolume root
835  * - objectid of the file holding the reference
836  * - original offset in the file
837  * - how many bookend extents
838  *
839  * The key offset for the implicit back refs is hash of the first
840  * three fields.
841  *
842  * The extent ref structure for the full back refs has field for:
843  *
844  * - number of pointers in the tree leaf
845  *
846  * The key offset for the implicit back refs is the first byte of
847  * the tree leaf
848  *
849  * When a file extent is allocated, The implicit back refs is used.
850  * the fields are filled in:
851  *
852  *     (root_key.objectid, inode objectid, offset in file, 1)
853  *
854  * When a file extent is removed file truncation, we find the
855  * corresponding implicit back refs and check the following fields:
856  *
857  *     (btrfs_header_owner(leaf), inode objectid, offset in file)
858  *
859  * Btree extents can be referenced by:
860  *
861  * - Different subvolumes
862  *
863  * Both the implicit back refs and the full back refs for tree blocks
864  * only consist of key. The key offset for the implicit back refs is
865  * objectid of block's owner tree. The key offset for the full back refs
866  * is the first byte of parent block.
867  *
868  * When implicit back refs is used, information about the lowest key and
869  * level of the tree block are required. These information are stored in
870  * tree block info structure.
871  */
872
873 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
874 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
875                                   struct btrfs_root *root,
876                                   struct btrfs_path *path,
877                                   u64 owner, u32 extra_size)
878 {
879         struct btrfs_extent_item *item;
880         struct btrfs_extent_item_v0 *ei0;
881         struct btrfs_extent_ref_v0 *ref0;
882         struct btrfs_tree_block_info *bi;
883         struct extent_buffer *leaf;
884         struct btrfs_key key;
885         struct btrfs_key found_key;
886         u32 new_size = sizeof(*item);
887         u64 refs;
888         int ret;
889
890         leaf = path->nodes[0];
891         BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
892
893         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
894         ei0 = btrfs_item_ptr(leaf, path->slots[0],
895                              struct btrfs_extent_item_v0);
896         refs = btrfs_extent_refs_v0(leaf, ei0);
897
898         if (owner == (u64)-1) {
899                 while (1) {
900                         if (path->slots[0] >= btrfs_header_nritems(leaf)) {
901                                 ret = btrfs_next_leaf(root, path);
902                                 if (ret < 0)
903                                         return ret;
904                                 BUG_ON(ret > 0);
905                                 leaf = path->nodes[0];
906                         }
907                         btrfs_item_key_to_cpu(leaf, &found_key,
908                                               path->slots[0]);
909                         BUG_ON(key.objectid != found_key.objectid);
910                         if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
911                                 path->slots[0]++;
912                                 continue;
913                         }
914                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
915                                               struct btrfs_extent_ref_v0);
916                         owner = btrfs_ref_objectid_v0(leaf, ref0);
917                         break;
918                 }
919         }
920         btrfs_release_path(root, path);
921
922         if (owner < BTRFS_FIRST_FREE_OBJECTID)
923                 new_size += sizeof(*bi);
924
925         new_size -= sizeof(*ei0);
926         ret = btrfs_search_slot(trans, root, &key, path,
927                                 new_size + extra_size, 1);
928         if (ret < 0)
929                 return ret;
930         BUG_ON(ret);
931
932         ret = btrfs_extend_item(trans, root, path, new_size);
933         BUG_ON(ret);
934
935         leaf = path->nodes[0];
936         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
937         btrfs_set_extent_refs(leaf, item, refs);
938         /* FIXME: get real generation */
939         btrfs_set_extent_generation(leaf, item, 0);
940         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
941                 btrfs_set_extent_flags(leaf, item,
942                                        BTRFS_EXTENT_FLAG_TREE_BLOCK |
943                                        BTRFS_BLOCK_FLAG_FULL_BACKREF);
944                 bi = (struct btrfs_tree_block_info *)(item + 1);
945                 /* FIXME: get first key of the block */
946                 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
947                 btrfs_set_tree_block_level(leaf, bi, (int)owner);
948         } else {
949                 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
950         }
951         btrfs_mark_buffer_dirty(leaf);
952         return 0;
953 }
954 #endif
955
956 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
957 {
958         u32 high_crc = ~(u32)0;
959         u32 low_crc = ~(u32)0;
960         __le64 lenum;
961
962         lenum = cpu_to_le64(root_objectid);
963         high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
964         lenum = cpu_to_le64(owner);
965         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
966         lenum = cpu_to_le64(offset);
967         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
968
969         return ((u64)high_crc << 31) ^ (u64)low_crc;
970 }
971
972 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
973                                      struct btrfs_extent_data_ref *ref)
974 {
975         return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
976                                     btrfs_extent_data_ref_objectid(leaf, ref),
977                                     btrfs_extent_data_ref_offset(leaf, ref));
978 }
979
980 static int match_extent_data_ref(struct extent_buffer *leaf,
981                                  struct btrfs_extent_data_ref *ref,
982                                  u64 root_objectid, u64 owner, u64 offset)
983 {
984         if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
985             btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
986             btrfs_extent_data_ref_offset(leaf, ref) != offset)
987                 return 0;
988         return 1;
989 }
990
991 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
992                                            struct btrfs_root *root,
993                                            struct btrfs_path *path,
994                                            u64 bytenr, u64 parent,
995                                            u64 root_objectid,
996                                            u64 owner, u64 offset)
997 {
998         struct btrfs_key key;
999         struct btrfs_extent_data_ref *ref;
1000         struct extent_buffer *leaf;
1001         u32 nritems;
1002         int ret;
1003         int recow;
1004         int err = -ENOENT;
1005
1006         key.objectid = bytenr;
1007         if (parent) {
1008                 key.type = BTRFS_SHARED_DATA_REF_KEY;
1009                 key.offset = parent;
1010         } else {
1011                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1012                 key.offset = hash_extent_data_ref(root_objectid,
1013                                                   owner, offset);
1014         }
1015 again:
1016         recow = 0;
1017         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1018         if (ret < 0) {
1019                 err = ret;
1020                 goto fail;
1021         }
1022
1023         if (parent) {
1024                 if (!ret)
1025                         return 0;
1026 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1027                 key.type = BTRFS_EXTENT_REF_V0_KEY;
1028                 btrfs_release_path(root, path);
1029                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1030                 if (ret < 0) {
1031                         err = ret;
1032                         goto fail;
1033                 }
1034                 if (!ret)
1035                         return 0;
1036 #endif
1037                 goto fail;
1038         }
1039
1040         leaf = path->nodes[0];
1041         nritems = btrfs_header_nritems(leaf);
1042         while (1) {
1043                 if (path->slots[0] >= nritems) {
1044                         ret = btrfs_next_leaf(root, path);
1045                         if (ret < 0)
1046                                 err = ret;
1047                         if (ret)
1048                                 goto fail;
1049
1050                         leaf = path->nodes[0];
1051                         nritems = btrfs_header_nritems(leaf);
1052                         recow = 1;
1053                 }
1054
1055                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1056                 if (key.objectid != bytenr ||
1057                     key.type != BTRFS_EXTENT_DATA_REF_KEY)
1058                         goto fail;
1059
1060                 ref = btrfs_item_ptr(leaf, path->slots[0],
1061                                      struct btrfs_extent_data_ref);
1062
1063                 if (match_extent_data_ref(leaf, ref, root_objectid,
1064                                           owner, offset)) {
1065                         if (recow) {
1066                                 btrfs_release_path(root, path);
1067                                 goto again;
1068                         }
1069                         err = 0;
1070                         break;
1071                 }
1072                 path->slots[0]++;
1073         }
1074 fail:
1075         return err;
1076 }
1077
1078 static noinline int insert_extent_data_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, u64 owner,
1083                                            u64 offset, int refs_to_add)
1084 {
1085         struct btrfs_key key;
1086         struct extent_buffer *leaf;
1087         u32 size;
1088         u32 num_refs;
1089         int ret;
1090
1091         key.objectid = bytenr;
1092         if (parent) {
1093                 key.type = BTRFS_SHARED_DATA_REF_KEY;
1094                 key.offset = parent;
1095                 size = sizeof(struct btrfs_shared_data_ref);
1096         } else {
1097                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1098                 key.offset = hash_extent_data_ref(root_objectid,
1099                                                   owner, offset);
1100                 size = sizeof(struct btrfs_extent_data_ref);
1101         }
1102
1103         ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1104         if (ret && ret != -EEXIST)
1105                 goto fail;
1106
1107         leaf = path->nodes[0];
1108         if (parent) {
1109                 struct btrfs_shared_data_ref *ref;
1110                 ref = btrfs_item_ptr(leaf, path->slots[0],
1111                                      struct btrfs_shared_data_ref);
1112                 if (ret == 0) {
1113                         btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1114                 } else {
1115                         num_refs = btrfs_shared_data_ref_count(leaf, ref);
1116                         num_refs += refs_to_add;
1117                         btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1118                 }
1119         } else {
1120                 struct btrfs_extent_data_ref *ref;
1121                 while (ret == -EEXIST) {
1122                         ref = btrfs_item_ptr(leaf, path->slots[0],
1123                                              struct btrfs_extent_data_ref);
1124                         if (match_extent_data_ref(leaf, ref, root_objectid,
1125                                                   owner, offset))
1126                                 break;
1127                         btrfs_release_path(root, path);
1128                         key.offset++;
1129                         ret = btrfs_insert_empty_item(trans, root, path, &key,
1130                                                       size);
1131                         if (ret && ret != -EEXIST)
1132                                 goto fail;
1133
1134                         leaf = path->nodes[0];
1135                 }
1136                 ref = btrfs_item_ptr(leaf, path->slots[0],
1137                                      struct btrfs_extent_data_ref);
1138                 if (ret == 0) {
1139                         btrfs_set_extent_data_ref_root(leaf, ref,
1140                                                        root_objectid);
1141                         btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1142                         btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1143                         btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1144                 } else {
1145                         num_refs = btrfs_extent_data_ref_count(leaf, ref);
1146                         num_refs += refs_to_add;
1147                         btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1148                 }
1149         }
1150         btrfs_mark_buffer_dirty(leaf);
1151         ret = 0;
1152 fail:
1153         btrfs_release_path(root, path);
1154         return ret;
1155 }
1156
1157 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1158                                            struct btrfs_root *root,
1159                                            struct btrfs_path *path,
1160                                            int refs_to_drop)
1161 {
1162         struct btrfs_key key;
1163         struct btrfs_extent_data_ref *ref1 = NULL;
1164         struct btrfs_shared_data_ref *ref2 = NULL;
1165         struct extent_buffer *leaf;
1166         u32 num_refs = 0;
1167         int ret = 0;
1168
1169         leaf = path->nodes[0];
1170         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1171
1172         if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1173                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1174                                       struct btrfs_extent_data_ref);
1175                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1176         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1177                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1178                                       struct btrfs_shared_data_ref);
1179                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1180 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1181         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1182                 struct btrfs_extent_ref_v0 *ref0;
1183                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1184                                       struct btrfs_extent_ref_v0);
1185                 num_refs = btrfs_ref_count_v0(leaf, ref0);
1186 #endif
1187         } else {
1188                 BUG();
1189         }
1190
1191         BUG_ON(num_refs < refs_to_drop);
1192         num_refs -= refs_to_drop;
1193
1194         if (num_refs == 0) {
1195                 ret = btrfs_del_item(trans, root, path);
1196         } else {
1197                 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1198                         btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1199                 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1200                         btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1201 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1202                 else {
1203                         struct btrfs_extent_ref_v0 *ref0;
1204                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
1205                                         struct btrfs_extent_ref_v0);
1206                         btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1207                 }
1208 #endif
1209                 btrfs_mark_buffer_dirty(leaf);
1210         }
1211         return ret;
1212 }
1213
1214 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1215                                           struct btrfs_path *path,
1216                                           struct btrfs_extent_inline_ref *iref)
1217 {
1218         struct btrfs_key key;
1219         struct extent_buffer *leaf;
1220         struct btrfs_extent_data_ref *ref1;
1221         struct btrfs_shared_data_ref *ref2;
1222         u32 num_refs = 0;
1223
1224         leaf = path->nodes[0];
1225         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1226         if (iref) {
1227                 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1228                     BTRFS_EXTENT_DATA_REF_KEY) {
1229                         ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1230                         num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1231                 } else {
1232                         ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1233                         num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1234                 }
1235         } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1236                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1237                                       struct btrfs_extent_data_ref);
1238                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1239         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1240                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1241                                       struct btrfs_shared_data_ref);
1242                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1243 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1244         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1245                 struct btrfs_extent_ref_v0 *ref0;
1246                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1247                                       struct btrfs_extent_ref_v0);
1248                 num_refs = btrfs_ref_count_v0(leaf, ref0);
1249 #endif
1250         } else {
1251                 WARN_ON(1);
1252         }
1253         return num_refs;
1254 }
1255
1256 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1257                                           struct btrfs_root *root,
1258                                           struct btrfs_path *path,
1259                                           u64 bytenr, u64 parent,
1260                                           u64 root_objectid)
1261 {
1262         struct btrfs_key key;
1263         int ret;
1264
1265         key.objectid = bytenr;
1266         if (parent) {
1267                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1268                 key.offset = parent;
1269         } else {
1270                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1271                 key.offset = root_objectid;
1272         }
1273
1274         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1275         if (ret > 0)
1276                 ret = -ENOENT;
1277 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1278         if (ret == -ENOENT && parent) {
1279                 btrfs_release_path(root, path);
1280                 key.type = BTRFS_EXTENT_REF_V0_KEY;
1281                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1282                 if (ret > 0)
1283                         ret = -ENOENT;
1284         }
1285 #endif
1286         return ret;
1287 }
1288
1289 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1290                                           struct btrfs_root *root,
1291                                           struct btrfs_path *path,
1292                                           u64 bytenr, u64 parent,
1293                                           u64 root_objectid)
1294 {
1295         struct btrfs_key key;
1296         int ret;
1297
1298         key.objectid = bytenr;
1299         if (parent) {
1300                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1301                 key.offset = parent;
1302         } else {
1303                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1304                 key.offset = root_objectid;
1305         }
1306
1307         ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1308         btrfs_release_path(root, path);
1309         return ret;
1310 }
1311
1312 static inline int extent_ref_type(u64 parent, u64 owner)
1313 {
1314         int type;
1315         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1316                 if (parent > 0)
1317                         type = BTRFS_SHARED_BLOCK_REF_KEY;
1318                 else
1319                         type = BTRFS_TREE_BLOCK_REF_KEY;
1320         } else {
1321                 if (parent > 0)
1322                         type = BTRFS_SHARED_DATA_REF_KEY;
1323                 else
1324                         type = BTRFS_EXTENT_DATA_REF_KEY;
1325         }
1326         return type;
1327 }
1328
1329 static int find_next_key(struct btrfs_path *path, int level,
1330                          struct btrfs_key *key)
1331
1332 {
1333         for (; level < BTRFS_MAX_LEVEL; level++) {
1334                 if (!path->nodes[level])
1335                         break;
1336                 if (path->slots[level] + 1 >=
1337                     btrfs_header_nritems(path->nodes[level]))
1338                         continue;
1339                 if (level == 0)
1340                         btrfs_item_key_to_cpu(path->nodes[level], key,
1341                                               path->slots[level] + 1);
1342                 else
1343                         btrfs_node_key_to_cpu(path->nodes[level], key,
1344                                               path->slots[level] + 1);
1345                 return 0;
1346         }
1347         return 1;
1348 }
1349
1350 /*
1351  * look for inline back ref. if back ref is found, *ref_ret is set
1352  * to the address of inline back ref, and 0 is returned.
1353  *
1354  * if back ref isn't found, *ref_ret is set to the address where it
1355  * should be inserted, and -ENOENT is returned.
1356  *
1357  * if insert is true and there are too many inline back refs, the path
1358  * points to the extent item, and -EAGAIN is returned.
1359  *
1360  * NOTE: inline back refs are ordered in the same way that back ref
1361  *       items in the tree are ordered.
1362  */
1363 static noinline_for_stack
1364 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1365                                  struct btrfs_root *root,
1366                                  struct btrfs_path *path,
1367                                  struct btrfs_extent_inline_ref **ref_ret,
1368                                  u64 bytenr, u64 num_bytes,
1369                                  u64 parent, u64 root_objectid,
1370                                  u64 owner, u64 offset, int insert)
1371 {
1372         struct btrfs_key key;
1373         struct extent_buffer *leaf;
1374         struct btrfs_extent_item *ei;
1375         struct btrfs_extent_inline_ref *iref;
1376         u64 flags;
1377         u64 item_size;
1378         unsigned long ptr;
1379         unsigned long end;
1380         int extra_size;
1381         int type;
1382         int want;
1383         int ret;
1384         int err = 0;
1385
1386         key.objectid = bytenr;
1387         key.type = BTRFS_EXTENT_ITEM_KEY;
1388         key.offset = num_bytes;
1389
1390         want = extent_ref_type(parent, owner);
1391         if (insert) {
1392                 extra_size = btrfs_extent_inline_ref_size(want);
1393                 path->keep_locks = 1;
1394         } else
1395                 extra_size = -1;
1396         ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1397         if (ret < 0) {
1398                 err = ret;
1399                 goto out;
1400         }
1401         BUG_ON(ret);
1402
1403         leaf = path->nodes[0];
1404         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1405 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1406         if (item_size < sizeof(*ei)) {
1407                 if (!insert) {
1408                         err = -ENOENT;
1409                         goto out;
1410                 }
1411                 ret = convert_extent_item_v0(trans, root, path, owner,
1412                                              extra_size);
1413                 if (ret < 0) {
1414                         err = ret;
1415                         goto out;
1416                 }
1417                 leaf = path->nodes[0];
1418                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1419         }
1420 #endif
1421         BUG_ON(item_size < sizeof(*ei));
1422
1423         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1424         flags = btrfs_extent_flags(leaf, ei);
1425
1426         ptr = (unsigned long)(ei + 1);
1427         end = (unsigned long)ei + item_size;
1428
1429         if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1430                 ptr += sizeof(struct btrfs_tree_block_info);
1431                 BUG_ON(ptr > end);
1432         } else {
1433                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1434         }
1435
1436         err = -ENOENT;
1437         while (1) {
1438                 if (ptr >= end) {
1439                         WARN_ON(ptr > end);
1440                         break;
1441                 }
1442                 iref = (struct btrfs_extent_inline_ref *)ptr;
1443                 type = btrfs_extent_inline_ref_type(leaf, iref);
1444                 if (want < type)
1445                         break;
1446                 if (want > type) {
1447                         ptr += btrfs_extent_inline_ref_size(type);
1448                         continue;
1449                 }
1450
1451                 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1452                         struct btrfs_extent_data_ref *dref;
1453                         dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1454                         if (match_extent_data_ref(leaf, dref, root_objectid,
1455                                                   owner, offset)) {
1456                                 err = 0;
1457                                 break;
1458                         }
1459                         if (hash_extent_data_ref_item(leaf, dref) <
1460                             hash_extent_data_ref(root_objectid, owner, offset))
1461                                 break;
1462                 } else {
1463                         u64 ref_offset;
1464                         ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1465                         if (parent > 0) {
1466                                 if (parent == ref_offset) {
1467                                         err = 0;
1468                                         break;
1469                                 }
1470                                 if (ref_offset < parent)
1471                                         break;
1472                         } else {
1473                                 if (root_objectid == ref_offset) {
1474                                         err = 0;
1475                                         break;
1476                                 }
1477                                 if (ref_offset < root_objectid)
1478                                         break;
1479                         }
1480                 }
1481                 ptr += btrfs_extent_inline_ref_size(type);
1482         }
1483         if (err == -ENOENT && insert) {
1484                 if (item_size + extra_size >=
1485                     BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1486                         err = -EAGAIN;
1487                         goto out;
1488                 }
1489                 /*
1490                  * To add new inline back ref, we have to make sure
1491                  * there is no corresponding back ref item.
1492                  * For simplicity, we just do not add new inline back
1493                  * ref if there is any kind of item for this block
1494                  */
1495                 if (find_next_key(path, 0, &key) == 0 &&
1496                     key.objectid == bytenr &&
1497                     key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1498                         err = -EAGAIN;
1499                         goto out;
1500                 }
1501         }
1502         *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1503 out:
1504         if (insert) {
1505                 path->keep_locks = 0;
1506                 btrfs_unlock_up_safe(path, 1);
1507         }
1508         return err;
1509 }
1510
1511 /*
1512  * helper to add new inline back ref
1513  */
1514 static noinline_for_stack
1515 int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1516                                 struct btrfs_root *root,
1517                                 struct btrfs_path *path,
1518                                 struct btrfs_extent_inline_ref *iref,
1519                                 u64 parent, u64 root_objectid,
1520                                 u64 owner, u64 offset, int refs_to_add,
1521                                 struct btrfs_delayed_extent_op *extent_op)
1522 {
1523         struct extent_buffer *leaf;
1524         struct btrfs_extent_item *ei;
1525         unsigned long ptr;
1526         unsigned long end;
1527         unsigned long item_offset;
1528         u64 refs;
1529         int size;
1530         int type;
1531         int ret;
1532
1533         leaf = path->nodes[0];
1534         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1535         item_offset = (unsigned long)iref - (unsigned long)ei;
1536
1537         type = extent_ref_type(parent, owner);
1538         size = btrfs_extent_inline_ref_size(type);
1539
1540         ret = btrfs_extend_item(trans, root, path, size);
1541         BUG_ON(ret);
1542
1543         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1544         refs = btrfs_extent_refs(leaf, ei);
1545         refs += refs_to_add;
1546         btrfs_set_extent_refs(leaf, ei, refs);
1547         if (extent_op)
1548                 __run_delayed_extent_op(extent_op, leaf, ei);
1549
1550         ptr = (unsigned long)ei + item_offset;
1551         end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1552         if (ptr < end - size)
1553                 memmove_extent_buffer(leaf, ptr + size, ptr,
1554                                       end - size - ptr);
1555
1556         iref = (struct btrfs_extent_inline_ref *)ptr;
1557         btrfs_set_extent_inline_ref_type(leaf, iref, type);
1558         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1559                 struct btrfs_extent_data_ref *dref;
1560                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1561                 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1562                 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1563                 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1564                 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1565         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1566                 struct btrfs_shared_data_ref *sref;
1567                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1568                 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1569                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1570         } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1571                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1572         } else {
1573                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1574         }
1575         btrfs_mark_buffer_dirty(leaf);
1576         return 0;
1577 }
1578
1579 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1580                                  struct btrfs_root *root,
1581                                  struct btrfs_path *path,
1582                                  struct btrfs_extent_inline_ref **ref_ret,
1583                                  u64 bytenr, u64 num_bytes, u64 parent,
1584                                  u64 root_objectid, u64 owner, u64 offset)
1585 {
1586         int ret;
1587
1588         ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1589                                            bytenr, num_bytes, parent,
1590                                            root_objectid, owner, offset, 0);
1591         if (ret != -ENOENT)
1592                 return ret;
1593
1594         btrfs_release_path(root, path);
1595         *ref_ret = NULL;
1596
1597         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1598                 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1599                                             root_objectid);
1600         } else {
1601                 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1602                                              root_objectid, owner, offset);
1603         }
1604         return ret;
1605 }
1606
1607 /*
1608  * helper to update/remove inline back ref
1609  */
1610 static noinline_for_stack
1611 int update_inline_extent_backref(struct btrfs_trans_handle *trans,
1612                                  struct btrfs_root *root,
1613                                  struct btrfs_path *path,
1614                                  struct btrfs_extent_inline_ref *iref,
1615                                  int refs_to_mod,
1616                                  struct btrfs_delayed_extent_op *extent_op)
1617 {
1618         struct extent_buffer *leaf;
1619         struct btrfs_extent_item *ei;
1620         struct btrfs_extent_data_ref *dref = NULL;
1621         struct btrfs_shared_data_ref *sref = NULL;
1622         unsigned long ptr;
1623         unsigned long end;
1624         u32 item_size;
1625         int size;
1626         int type;
1627         int ret;
1628         u64 refs;
1629
1630         leaf = path->nodes[0];
1631         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1632         refs = btrfs_extent_refs(leaf, ei);
1633         WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1634         refs += refs_to_mod;
1635         btrfs_set_extent_refs(leaf, ei, refs);
1636         if (extent_op)
1637                 __run_delayed_extent_op(extent_op, leaf, ei);
1638
1639         type = btrfs_extent_inline_ref_type(leaf, iref);
1640
1641         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1642                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1643                 refs = btrfs_extent_data_ref_count(leaf, dref);
1644         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1645                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1646                 refs = btrfs_shared_data_ref_count(leaf, sref);
1647         } else {
1648                 refs = 1;
1649                 BUG_ON(refs_to_mod != -1);
1650         }
1651
1652         BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1653         refs += refs_to_mod;
1654
1655         if (refs > 0) {
1656                 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1657                         btrfs_set_extent_data_ref_count(leaf, dref, refs);
1658                 else
1659                         btrfs_set_shared_data_ref_count(leaf, sref, refs);
1660         } else {
1661                 size =  btrfs_extent_inline_ref_size(type);
1662                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1663                 ptr = (unsigned long)iref;
1664                 end = (unsigned long)ei + item_size;
1665                 if (ptr + size < end)
1666                         memmove_extent_buffer(leaf, ptr, ptr + size,
1667                                               end - ptr - size);
1668                 item_size -= size;
1669                 ret = btrfs_truncate_item(trans, root, path, item_size, 1);
1670                 BUG_ON(ret);
1671         }
1672         btrfs_mark_buffer_dirty(leaf);
1673         return 0;
1674 }
1675
1676 static noinline_for_stack
1677 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1678                                  struct btrfs_root *root,
1679                                  struct btrfs_path *path,
1680                                  u64 bytenr, u64 num_bytes, u64 parent,
1681                                  u64 root_objectid, u64 owner,
1682                                  u64 offset, int refs_to_add,
1683                                  struct btrfs_delayed_extent_op *extent_op)
1684 {
1685         struct btrfs_extent_inline_ref *iref;
1686         int ret;
1687
1688         ret = lookup_inline_extent_backref(trans, root, path, &iref,
1689                                            bytenr, num_bytes, parent,
1690                                            root_objectid, owner, offset, 1);
1691         if (ret == 0) {
1692                 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1693                 ret = update_inline_extent_backref(trans, root, path, iref,
1694                                                    refs_to_add, extent_op);
1695         } else if (ret == -ENOENT) {
1696                 ret = setup_inline_extent_backref(trans, root, path, iref,
1697                                                   parent, root_objectid,
1698                                                   owner, offset, refs_to_add,
1699                                                   extent_op);
1700         }
1701         return ret;
1702 }
1703
1704 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1705                                  struct btrfs_root *root,
1706                                  struct btrfs_path *path,
1707                                  u64 bytenr, u64 parent, u64 root_objectid,
1708                                  u64 owner, u64 offset, int refs_to_add)
1709 {
1710         int ret;
1711         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1712                 BUG_ON(refs_to_add != 1);
1713                 ret = insert_tree_block_ref(trans, root, path, bytenr,
1714                                             parent, root_objectid);
1715         } else {
1716                 ret = insert_extent_data_ref(trans, root, path, bytenr,
1717                                              parent, root_objectid,
1718                                              owner, offset, refs_to_add);
1719         }
1720         return ret;
1721 }
1722
1723 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1724                                  struct btrfs_root *root,
1725                                  struct btrfs_path *path,
1726                                  struct btrfs_extent_inline_ref *iref,
1727                                  int refs_to_drop, int is_data)
1728 {
1729         int ret;
1730
1731         BUG_ON(!is_data && refs_to_drop != 1);
1732         if (iref) {
1733                 ret = update_inline_extent_backref(trans, root, path, iref,
1734                                                    -refs_to_drop, NULL);
1735         } else if (is_data) {
1736                 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1737         } else {
1738                 ret = btrfs_del_item(trans, root, path);
1739         }
1740         return ret;
1741 }
1742
1743 static void btrfs_issue_discard(struct block_device *bdev,
1744                                 u64 start, u64 len)
1745 {
1746         blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL, 0);
1747 }
1748
1749 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1750                                 u64 num_bytes)
1751 {
1752         int ret;
1753         u64 map_length = num_bytes;
1754         struct btrfs_multi_bio *multi = NULL;
1755
1756         if (!btrfs_test_opt(root, DISCARD))
1757                 return 0;
1758
1759         /* Tell the block device(s) that the sectors can be discarded */
1760         ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
1761                               bytenr, &map_length, &multi, 0);
1762         if (!ret) {
1763                 struct btrfs_bio_stripe *stripe = multi->stripes;
1764                 int i;
1765
1766                 if (map_length > num_bytes)
1767                         map_length = num_bytes;
1768
1769                 for (i = 0; i < multi->num_stripes; i++, stripe++) {
1770                         btrfs_issue_discard(stripe->dev->bdev,
1771                                             stripe->physical,
1772                                             map_length);
1773                 }
1774                 kfree(multi);
1775         }
1776
1777         return ret;
1778 }
1779
1780 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1781                          struct btrfs_root *root,
1782                          u64 bytenr, u64 num_bytes, u64 parent,
1783                          u64 root_objectid, u64 owner, u64 offset)
1784 {
1785         int ret;
1786         BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1787                root_objectid == BTRFS_TREE_LOG_OBJECTID);
1788
1789         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1790                 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
1791                                         parent, root_objectid, (int)owner,
1792                                         BTRFS_ADD_DELAYED_REF, NULL);
1793         } else {
1794                 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
1795                                         parent, root_objectid, owner, offset,
1796                                         BTRFS_ADD_DELAYED_REF, NULL);
1797         }
1798         return ret;
1799 }
1800
1801 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1802                                   struct btrfs_root *root,
1803                                   u64 bytenr, u64 num_bytes,
1804                                   u64 parent, u64 root_objectid,
1805                                   u64 owner, u64 offset, int refs_to_add,
1806                                   struct btrfs_delayed_extent_op *extent_op)
1807 {
1808         struct btrfs_path *path;
1809         struct extent_buffer *leaf;
1810         struct btrfs_extent_item *item;
1811         u64 refs;
1812         int ret;
1813         int err = 0;
1814
1815         path = btrfs_alloc_path();
1816         if (!path)
1817                 return -ENOMEM;
1818
1819         path->reada = 1;
1820         path->leave_spinning = 1;
1821         /* this will setup the path even if it fails to insert the back ref */
1822         ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1823                                            path, bytenr, num_bytes, parent,
1824                                            root_objectid, owner, offset,
1825                                            refs_to_add, extent_op);
1826         if (ret == 0)
1827                 goto out;
1828
1829         if (ret != -EAGAIN) {
1830                 err = ret;
1831                 goto out;
1832         }
1833
1834         leaf = path->nodes[0];
1835         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1836         refs = btrfs_extent_refs(leaf, item);
1837         btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1838         if (extent_op)
1839                 __run_delayed_extent_op(extent_op, leaf, item);
1840
1841         btrfs_mark_buffer_dirty(leaf);
1842         btrfs_release_path(root->fs_info->extent_root, path);
1843
1844         path->reada = 1;
1845         path->leave_spinning = 1;
1846
1847         /* now insert the actual backref */
1848         ret = insert_extent_backref(trans, root->fs_info->extent_root,
1849                                     path, bytenr, parent, root_objectid,
1850                                     owner, offset, refs_to_add);
1851         BUG_ON(ret);
1852 out:
1853         btrfs_free_path(path);
1854         return err;
1855 }
1856
1857 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1858                                 struct btrfs_root *root,
1859                                 struct btrfs_delayed_ref_node *node,
1860                                 struct btrfs_delayed_extent_op *extent_op,
1861                                 int insert_reserved)
1862 {
1863         int ret = 0;
1864         struct btrfs_delayed_data_ref *ref;
1865         struct btrfs_key ins;
1866         u64 parent = 0;
1867         u64 ref_root = 0;
1868         u64 flags = 0;
1869
1870         ins.objectid = node->bytenr;
1871         ins.offset = node->num_bytes;
1872         ins.type = BTRFS_EXTENT_ITEM_KEY;
1873
1874         ref = btrfs_delayed_node_to_data_ref(node);
1875         if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1876                 parent = ref->parent;
1877         else
1878                 ref_root = ref->root;
1879
1880         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1881                 if (extent_op) {
1882                         BUG_ON(extent_op->update_key);
1883                         flags |= extent_op->flags_to_set;
1884                 }
1885                 ret = alloc_reserved_file_extent(trans, root,
1886                                                  parent, ref_root, flags,
1887                                                  ref->objectid, ref->offset,
1888                                                  &ins, node->ref_mod);
1889         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1890                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1891                                              node->num_bytes, parent,
1892                                              ref_root, ref->objectid,
1893                                              ref->offset, node->ref_mod,
1894                                              extent_op);
1895         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1896                 ret = __btrfs_free_extent(trans, root, node->bytenr,
1897                                           node->num_bytes, parent,
1898                                           ref_root, ref->objectid,
1899                                           ref->offset, node->ref_mod,
1900                                           extent_op);
1901         } else {
1902                 BUG();
1903         }
1904         return ret;
1905 }
1906
1907 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1908                                     struct extent_buffer *leaf,
1909                                     struct btrfs_extent_item *ei)
1910 {
1911         u64 flags = btrfs_extent_flags(leaf, ei);
1912         if (extent_op->update_flags) {
1913                 flags |= extent_op->flags_to_set;
1914                 btrfs_set_extent_flags(leaf, ei, flags);
1915         }
1916
1917         if (extent_op->update_key) {
1918                 struct btrfs_tree_block_info *bi;
1919                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
1920                 bi = (struct btrfs_tree_block_info *)(ei + 1);
1921                 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
1922         }
1923 }
1924
1925 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
1926                                  struct btrfs_root *root,
1927                                  struct btrfs_delayed_ref_node *node,
1928                                  struct btrfs_delayed_extent_op *extent_op)
1929 {
1930         struct btrfs_key key;
1931         struct btrfs_path *path;
1932         struct btrfs_extent_item *ei;
1933         struct extent_buffer *leaf;
1934         u32 item_size;
1935         int ret;
1936         int err = 0;
1937
1938         path = btrfs_alloc_path();
1939         if (!path)
1940                 return -ENOMEM;
1941
1942         key.objectid = node->bytenr;
1943         key.type = BTRFS_EXTENT_ITEM_KEY;
1944         key.offset = node->num_bytes;
1945
1946         path->reada = 1;
1947         path->leave_spinning = 1;
1948         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
1949                                 path, 0, 1);
1950         if (ret < 0) {
1951                 err = ret;
1952                 goto out;
1953         }
1954         if (ret > 0) {
1955                 err = -EIO;
1956                 goto out;
1957         }
1958
1959         leaf = path->nodes[0];
1960         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1961 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1962         if (item_size < sizeof(*ei)) {
1963                 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
1964                                              path, (u64)-1, 0);
1965                 if (ret < 0) {
1966                         err = ret;
1967                         goto out;
1968                 }
1969                 leaf = path->nodes[0];
1970                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1971         }
1972 #endif
1973         BUG_ON(item_size < sizeof(*ei));
1974         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1975         __run_delayed_extent_op(extent_op, leaf, ei);
1976
1977         btrfs_mark_buffer_dirty(leaf);
1978 out:
1979         btrfs_free_path(path);
1980         return err;
1981 }
1982
1983 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
1984                                 struct btrfs_root *root,
1985                                 struct btrfs_delayed_ref_node *node,
1986                                 struct btrfs_delayed_extent_op *extent_op,
1987                                 int insert_reserved)
1988 {
1989         int ret = 0;
1990         struct btrfs_delayed_tree_ref *ref;
1991         struct btrfs_key ins;
1992         u64 parent = 0;
1993         u64 ref_root = 0;
1994
1995         ins.objectid = node->bytenr;
1996         ins.offset = node->num_bytes;
1997         ins.type = BTRFS_EXTENT_ITEM_KEY;
1998
1999         ref = btrfs_delayed_node_to_tree_ref(node);
2000         if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2001                 parent = ref->parent;
2002         else
2003                 ref_root = ref->root;
2004
2005         BUG_ON(node->ref_mod != 1);
2006         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2007                 BUG_ON(!extent_op || !extent_op->update_flags ||
2008                        !extent_op->update_key);
2009                 ret = alloc_reserved_tree_block(trans, root,
2010                                                 parent, ref_root,
2011                                                 extent_op->flags_to_set,
2012                                                 &extent_op->key,
2013                                                 ref->level, &ins);
2014         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2015                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2016                                              node->num_bytes, parent, ref_root,
2017                                              ref->level, 0, 1, extent_op);
2018         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2019                 ret = __btrfs_free_extent(trans, root, node->bytenr,
2020                                           node->num_bytes, parent, ref_root,
2021                                           ref->level, 0, 1, extent_op);
2022         } else {
2023                 BUG();
2024         }
2025         return ret;
2026 }
2027
2028 /* helper function to actually process a single delayed ref entry */
2029 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2030                                struct btrfs_root *root,
2031                                struct btrfs_delayed_ref_node *node,
2032                                struct btrfs_delayed_extent_op *extent_op,
2033                                int insert_reserved)
2034 {
2035         int ret;
2036         if (btrfs_delayed_ref_is_head(node)) {
2037                 struct btrfs_delayed_ref_head *head;
2038                 /*
2039                  * we've hit the end of the chain and we were supposed
2040                  * to insert this extent into the tree.  But, it got
2041                  * deleted before we ever needed to insert it, so all
2042                  * we have to do is clean up the accounting
2043                  */
2044                 BUG_ON(extent_op);
2045                 head = btrfs_delayed_node_to_head(node);
2046                 if (insert_reserved) {
2047                         btrfs_pin_extent(root, node->bytenr,
2048                                          node->num_bytes, 1);
2049                         if (head->is_data) {
2050                                 ret = btrfs_del_csums(trans, root,
2051                                                       node->bytenr,
2052                                                       node->num_bytes);
2053                                 BUG_ON(ret);
2054                         }
2055                 }
2056                 mutex_unlock(&head->mutex);
2057                 return 0;
2058         }
2059
2060         if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2061             node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2062                 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2063                                            insert_reserved);
2064         else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2065                  node->type == BTRFS_SHARED_DATA_REF_KEY)
2066                 ret = run_delayed_data_ref(trans, root, node, extent_op,
2067                                            insert_reserved);
2068         else
2069                 BUG();
2070         return ret;
2071 }
2072
2073 static noinline struct btrfs_delayed_ref_node *
2074 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2075 {
2076         struct rb_node *node;
2077         struct btrfs_delayed_ref_node *ref;
2078         int action = BTRFS_ADD_DELAYED_REF;
2079 again:
2080         /*
2081          * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2082          * this prevents ref count from going down to zero when
2083          * there still are pending delayed ref.
2084          */
2085         node = rb_prev(&head->node.rb_node);
2086         while (1) {
2087                 if (!node)
2088                         break;
2089                 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2090                                 rb_node);
2091                 if (ref->bytenr != head->node.bytenr)
2092                         break;
2093                 if (ref->action == action)
2094                         return ref;
2095                 node = rb_prev(node);
2096         }
2097         if (action == BTRFS_ADD_DELAYED_REF) {
2098                 action = BTRFS_DROP_DELAYED_REF;
2099                 goto again;
2100         }
2101         return NULL;
2102 }
2103
2104 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2105                                        struct btrfs_root *root,
2106                                        struct list_head *cluster)
2107 {
2108         struct btrfs_delayed_ref_root *delayed_refs;
2109         struct btrfs_delayed_ref_node *ref;
2110         struct btrfs_delayed_ref_head *locked_ref = NULL;
2111         struct btrfs_delayed_extent_op *extent_op;
2112         int ret;
2113         int count = 0;
2114         int must_insert_reserved = 0;
2115
2116         delayed_refs = &trans->transaction->delayed_refs;
2117         while (1) {
2118                 if (!locked_ref) {
2119                         /* pick a new head ref from the cluster list */
2120                         if (list_empty(cluster))
2121                                 break;
2122
2123                         locked_ref = list_entry(cluster->next,
2124                                      struct btrfs_delayed_ref_head, cluster);
2125
2126                         /* grab the lock that says we are going to process
2127                          * all the refs for this head */
2128                         ret = btrfs_delayed_ref_lock(trans, locked_ref);
2129
2130                         /*
2131                          * we may have dropped the spin lock to get the head
2132                          * mutex lock, and that might have given someone else
2133                          * time to free the head.  If that's true, it has been
2134                          * removed from our list and we can move on.
2135                          */
2136                         if (ret == -EAGAIN) {
2137                                 locked_ref = NULL;
2138                                 count++;
2139                                 continue;
2140                         }
2141                 }
2142
2143                 /*
2144                  * record the must insert reserved flag before we
2145                  * drop the spin lock.
2146                  */
2147                 must_insert_reserved = locked_ref->must_insert_reserved;
2148                 locked_ref->must_insert_reserved = 0;
2149
2150                 extent_op = locked_ref->extent_op;
2151                 locked_ref->extent_op = NULL;
2152
2153                 /*
2154                  * locked_ref is the head node, so we have to go one
2155                  * node back for any delayed ref updates
2156                  */
2157                 ref = select_delayed_ref(locked_ref);
2158                 if (!ref) {
2159                         /* All delayed refs have been processed, Go ahead
2160                          * and send the head node to run_one_delayed_ref,
2161                          * so that any accounting fixes can happen
2162                          */
2163                         ref = &locked_ref->node;
2164
2165                         if (extent_op && must_insert_reserved) {
2166                                 kfree(extent_op);
2167                                 extent_op = NULL;
2168                         }
2169
2170                         if (extent_op) {
2171                                 spin_unlock(&delayed_refs->lock);
2172
2173                                 ret = run_delayed_extent_op(trans, root,
2174                                                             ref, extent_op);
2175                                 BUG_ON(ret);
2176                                 kfree(extent_op);
2177
2178                                 cond_resched();
2179                                 spin_lock(&delayed_refs->lock);
2180                                 continue;
2181                         }
2182
2183                         list_del_init(&locked_ref->cluster);
2184                         locked_ref = NULL;
2185                 }
2186
2187                 ref->in_tree = 0;
2188                 rb_erase(&ref->rb_node, &delayed_refs->root);
2189                 delayed_refs->num_entries--;
2190
2191                 spin_unlock(&delayed_refs->lock);
2192
2193                 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2194                                           must_insert_reserved);
2195                 BUG_ON(ret);
2196
2197                 btrfs_put_delayed_ref(ref);
2198                 kfree(extent_op);
2199                 count++;
2200
2201                 cond_resched();
2202                 spin_lock(&delayed_refs->lock);
2203         }
2204         return count;
2205 }
2206
2207 /*
2208  * this starts processing the delayed reference count updates and
2209  * extent insertions we have queued up so far.  count can be
2210  * 0, which means to process everything in the tree at the start
2211  * of the run (but not newly added entries), or it can be some target
2212  * number you'd like to process.
2213  */
2214 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2215                            struct btrfs_root *root, unsigned long count)
2216 {
2217         struct rb_node *node;
2218         struct btrfs_delayed_ref_root *delayed_refs;
2219         struct btrfs_delayed_ref_node *ref;
2220         struct list_head cluster;
2221         int ret;
2222         int run_all = count == (unsigned long)-1;
2223         int run_most = 0;
2224
2225         if (root == root->fs_info->extent_root)
2226                 root = root->fs_info->tree_root;
2227
2228         delayed_refs = &trans->transaction->delayed_refs;
2229         INIT_LIST_HEAD(&cluster);
2230 again:
2231         spin_lock(&delayed_refs->lock);
2232         if (count == 0) {
2233                 count = delayed_refs->num_entries * 2;
2234                 run_most = 1;
2235         }
2236         while (1) {
2237                 if (!(run_all || run_most) &&
2238                     delayed_refs->num_heads_ready < 64)
2239                         break;
2240
2241                 /*
2242                  * go find something we can process in the rbtree.  We start at
2243                  * the beginning of the tree, and then build a cluster
2244                  * of refs to process starting at the first one we are able to
2245                  * lock
2246                  */
2247                 ret = btrfs_find_ref_cluster(trans, &cluster,
2248                                              delayed_refs->run_delayed_start);
2249                 if (ret)
2250                         break;
2251
2252                 ret = run_clustered_refs(trans, root, &cluster);
2253                 BUG_ON(ret < 0);
2254
2255                 count -= min_t(unsigned long, ret, count);
2256
2257                 if (count == 0)
2258                         break;
2259         }
2260
2261         if (run_all) {
2262                 node = rb_first(&delayed_refs->root);
2263                 if (!node)
2264                         goto out;
2265                 count = (unsigned long)-1;
2266
2267                 while (node) {
2268                         ref = rb_entry(node, struct btrfs_delayed_ref_node,
2269                                        rb_node);
2270                         if (btrfs_delayed_ref_is_head(ref)) {
2271                                 struct btrfs_delayed_ref_head *head;
2272
2273                                 head = btrfs_delayed_node_to_head(ref);
2274                                 atomic_inc(&ref->refs);
2275
2276                                 spin_unlock(&delayed_refs->lock);
2277                                 mutex_lock(&head->mutex);
2278                                 mutex_unlock(&head->mutex);
2279
2280                                 btrfs_put_delayed_ref(ref);
2281                                 cond_resched();
2282                                 goto again;
2283                         }
2284                         node = rb_next(node);
2285                 }
2286                 spin_unlock(&delayed_refs->lock);
2287                 schedule_timeout(1);
2288                 goto again;
2289         }
2290 out:
2291         spin_unlock(&delayed_refs->lock);
2292         return 0;
2293 }
2294
2295 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2296                                 struct btrfs_root *root,
2297                                 u64 bytenr, u64 num_bytes, u64 flags,
2298                                 int is_data)
2299 {
2300         struct btrfs_delayed_extent_op *extent_op;
2301         int ret;
2302
2303         extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2304         if (!extent_op)
2305                 return -ENOMEM;
2306
2307         extent_op->flags_to_set = flags;
2308         extent_op->update_flags = 1;
2309         extent_op->update_key = 0;
2310         extent_op->is_data = is_data ? 1 : 0;
2311
2312         ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op);
2313         if (ret)
2314                 kfree(extent_op);
2315         return ret;
2316 }
2317
2318 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2319                                       struct btrfs_root *root,
2320                                       struct btrfs_path *path,
2321                                       u64 objectid, u64 offset, u64 bytenr)
2322 {
2323         struct btrfs_delayed_ref_head *head;
2324         struct btrfs_delayed_ref_node *ref;
2325         struct btrfs_delayed_data_ref *data_ref;
2326         struct btrfs_delayed_ref_root *delayed_refs;
2327         struct rb_node *node;
2328         int ret = 0;
2329
2330         ret = -ENOENT;
2331         delayed_refs = &trans->transaction->delayed_refs;
2332         spin_lock(&delayed_refs->lock);
2333         head = btrfs_find_delayed_ref_head(trans, bytenr);
2334         if (!head)
2335                 goto out;
2336
2337         if (!mutex_trylock(&head->mutex)) {
2338                 atomic_inc(&head->node.refs);
2339                 spin_unlock(&delayed_refs->lock);
2340
2341                 btrfs_release_path(root->fs_info->extent_root, path);
2342
2343                 mutex_lock(&head->mutex);
2344                 mutex_unlock(&head->mutex);
2345                 btrfs_put_delayed_ref(&head->node);
2346                 return -EAGAIN;
2347         }
2348
2349         node = rb_prev(&head->node.rb_node);
2350         if (!node)
2351                 goto out_unlock;
2352
2353         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2354
2355         if (ref->bytenr != bytenr)
2356                 goto out_unlock;
2357
2358         ret = 1;
2359         if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2360                 goto out_unlock;
2361
2362         data_ref = btrfs_delayed_node_to_data_ref(ref);
2363
2364         node = rb_prev(node);
2365         if (node) {
2366                 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2367                 if (ref->bytenr == bytenr)
2368                         goto out_unlock;
2369         }
2370
2371         if (data_ref->root != root->root_key.objectid ||
2372             data_ref->objectid != objectid || data_ref->offset != offset)
2373                 goto out_unlock;
2374
2375         ret = 0;
2376 out_unlock:
2377         mutex_unlock(&head->mutex);
2378 out:
2379         spin_unlock(&delayed_refs->lock);
2380         return ret;
2381 }
2382
2383 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2384                                         struct btrfs_root *root,
2385                                         struct btrfs_path *path,
2386                                         u64 objectid, u64 offset, u64 bytenr)
2387 {
2388         struct btrfs_root *extent_root = root->fs_info->extent_root;
2389         struct extent_buffer *leaf;
2390         struct btrfs_extent_data_ref *ref;
2391         struct btrfs_extent_inline_ref *iref;
2392         struct btrfs_extent_item *ei;
2393         struct btrfs_key key;
2394         u32 item_size;
2395         int ret;
2396
2397         key.objectid = bytenr;
2398         key.offset = (u64)-1;
2399         key.type = BTRFS_EXTENT_ITEM_KEY;
2400
2401         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2402         if (ret < 0)
2403                 goto out;
2404         BUG_ON(ret == 0);
2405
2406         ret = -ENOENT;
2407         if (path->slots[0] == 0)
2408                 goto out;
2409
2410         path->slots[0]--;
2411         leaf = path->nodes[0];
2412         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2413
2414         if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2415                 goto out;
2416
2417         ret = 1;
2418         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2419 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2420         if (item_size < sizeof(*ei)) {
2421                 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2422                 goto out;
2423         }
2424 #endif
2425         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2426
2427         if (item_size != sizeof(*ei) +
2428             btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2429                 goto out;
2430
2431         if (btrfs_extent_generation(leaf, ei) <=
2432             btrfs_root_last_snapshot(&root->root_item))
2433                 goto out;
2434
2435         iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2436         if (btrfs_extent_inline_ref_type(leaf, iref) !=
2437             BTRFS_EXTENT_DATA_REF_KEY)
2438                 goto out;
2439
2440         ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2441         if (btrfs_extent_refs(leaf, ei) !=
2442             btrfs_extent_data_ref_count(leaf, ref) ||
2443             btrfs_extent_data_ref_root(leaf, ref) !=
2444             root->root_key.objectid ||
2445             btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2446             btrfs_extent_data_ref_offset(leaf, ref) != offset)
2447                 goto out;
2448
2449         ret = 0;
2450 out:
2451         return ret;
2452 }
2453
2454 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2455                           struct btrfs_root *root,
2456                           u64 objectid, u64 offset, u64 bytenr)
2457 {
2458         struct btrfs_path *path;
2459         int ret;
2460         int ret2;
2461
2462         path = btrfs_alloc_path();
2463         if (!path)
2464                 return -ENOENT;
2465
2466         do {
2467                 ret = check_committed_ref(trans, root, path, objectid,
2468                                           offset, bytenr);
2469                 if (ret && ret != -ENOENT)
2470                         goto out;
2471
2472                 ret2 = check_delayed_ref(trans, root, path, objectid,
2473                                          offset, bytenr);
2474         } while (ret2 == -EAGAIN);
2475
2476         if (ret2 && ret2 != -ENOENT) {
2477                 ret = ret2;
2478                 goto out;
2479         }
2480
2481         if (ret != -ENOENT || ret2 != -ENOENT)
2482                 ret = 0;
2483 out:
2484         btrfs_free_path(path);
2485         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2486                 WARN_ON(ret > 0);
2487         return ret;
2488 }
2489
2490 #if 0
2491 int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2492                     struct extent_buffer *buf, u32 nr_extents)
2493 {
2494         struct btrfs_key key;
2495         struct btrfs_file_extent_item *fi;
2496         u64 root_gen;
2497         u32 nritems;
2498         int i;
2499         int level;
2500         int ret = 0;
2501         int shared = 0;
2502
2503         if (!root->ref_cows)
2504                 return 0;
2505
2506         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
2507                 shared = 0;
2508                 root_gen = root->root_key.offset;
2509         } else {
2510                 shared = 1;
2511                 root_gen = trans->transid - 1;
2512         }
2513
2514         level = btrfs_header_level(buf);
2515         nritems = btrfs_header_nritems(buf);
2516
2517         if (level == 0) {
2518                 struct btrfs_leaf_ref *ref;
2519                 struct btrfs_extent_info *info;
2520
2521                 ref = btrfs_alloc_leaf_ref(root, nr_extents);
2522                 if (!ref) {
2523                         ret = -ENOMEM;
2524                         goto out;
2525                 }
2526
2527                 ref->root_gen = root_gen;
2528                 ref->bytenr = buf->start;
2529                 ref->owner = btrfs_header_owner(buf);
2530                 ref->generation = btrfs_header_generation(buf);
2531                 ref->nritems = nr_extents;
2532                 info = ref->extents;
2533
2534                 for (i = 0; nr_extents > 0 && i < nritems; i++) {
2535                         u64 disk_bytenr;
2536                         btrfs_item_key_to_cpu(buf, &key, i);
2537                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2538                                 continue;
2539                         fi = btrfs_item_ptr(buf, i,
2540                                             struct btrfs_file_extent_item);
2541                         if (btrfs_file_extent_type(buf, fi) ==
2542                             BTRFS_FILE_EXTENT_INLINE)
2543                                 continue;
2544                         disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2545                         if (disk_bytenr == 0)
2546                                 continue;
2547
2548                         info->bytenr = disk_bytenr;
2549                         info->num_bytes =
2550                                 btrfs_file_extent_disk_num_bytes(buf, fi);
2551                         info->objectid = key.objectid;
2552                         info->offset = key.offset;
2553                         info++;
2554                 }
2555
2556                 ret = btrfs_add_leaf_ref(root, ref, shared);
2557                 if (ret == -EEXIST && shared) {
2558                         struct btrfs_leaf_ref *old;
2559                         old = btrfs_lookup_leaf_ref(root, ref->bytenr);
2560                         BUG_ON(!old);
2561                         btrfs_remove_leaf_ref(root, old);
2562                         btrfs_free_leaf_ref(root, old);
2563                         ret = btrfs_add_leaf_ref(root, ref, shared);
2564                 }
2565                 WARN_ON(ret);
2566                 btrfs_free_leaf_ref(root, ref);
2567         }
2568 out:
2569         return ret;
2570 }
2571
2572 /* when a block goes through cow, we update the reference counts of
2573  * everything that block points to.  The internal pointers of the block
2574  * can be in just about any order, and it is likely to have clusters of
2575  * things that are close together and clusters of things that are not.
2576  *
2577  * To help reduce the seeks that come with updating all of these reference
2578  * counts, sort them by byte number before actual updates are done.
2579  *
2580  * struct refsort is used to match byte number to slot in the btree block.
2581  * we sort based on the byte number and then use the slot to actually
2582  * find the item.
2583  *
2584  * struct refsort is smaller than strcut btrfs_item and smaller than
2585  * struct btrfs_key_ptr.  Since we're currently limited to the page size
2586  * for a btree block, there's no way for a kmalloc of refsorts for a
2587  * single node to be bigger than a page.
2588  */
2589 struct refsort {
2590         u64 bytenr;
2591         u32 slot;
2592 };
2593
2594 /*
2595  * for passing into sort()
2596  */
2597 static int refsort_cmp(const void *a_void, const void *b_void)
2598 {
2599         const struct refsort *a = a_void;
2600         const struct refsort *b = b_void;
2601
2602         if (a->bytenr < b->bytenr)
2603                 return -1;
2604         if (a->bytenr > b->bytenr)
2605                 return 1;
2606         return 0;
2607 }
2608 #endif
2609
2610 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2611                            struct btrfs_root *root,
2612                            struct extent_buffer *buf,
2613                            int full_backref, int inc)
2614 {
2615         u64 bytenr;
2616         u64 num_bytes;
2617         u64 parent;
2618         u64 ref_root;
2619         u32 nritems;
2620         struct btrfs_key key;
2621         struct btrfs_file_extent_item *fi;
2622         int i;
2623         int level;
2624         int ret = 0;
2625         int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2626                             u64, u64, u64, u64, u64, u64);
2627
2628         ref_root = btrfs_header_owner(buf);
2629         nritems = btrfs_header_nritems(buf);
2630         level = btrfs_header_level(buf);
2631
2632         if (!root->ref_cows && level == 0)
2633                 return 0;
2634
2635         if (inc)
2636                 process_func = btrfs_inc_extent_ref;
2637         else
2638                 process_func = btrfs_free_extent;
2639
2640         if (full_backref)
2641                 parent = buf->start;
2642         else
2643                 parent = 0;
2644
2645         for (i = 0; i < nritems; i++) {
2646                 if (level == 0) {
2647                         btrfs_item_key_to_cpu(buf, &key, i);
2648                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2649                                 continue;
2650                         fi = btrfs_item_ptr(buf, i,
2651                                             struct btrfs_file_extent_item);
2652                         if (btrfs_file_extent_type(buf, fi) ==
2653                             BTRFS_FILE_EXTENT_INLINE)
2654                                 continue;
2655                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2656                         if (bytenr == 0)
2657                                 continue;
2658
2659                         num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2660                         key.offset -= btrfs_file_extent_offset(buf, fi);
2661                         ret = process_func(trans, root, bytenr, num_bytes,
2662                                            parent, ref_root, key.objectid,
2663                                            key.offset);
2664                         if (ret)
2665                                 goto fail;
2666                 } else {
2667                         bytenr = btrfs_node_blockptr(buf, i);
2668                         num_bytes = btrfs_level_size(root, level - 1);
2669                         ret = process_func(trans, root, bytenr, num_bytes,
2670                                            parent, ref_root, level - 1, 0);
2671                         if (ret)
2672                                 goto fail;
2673                 }
2674         }
2675         return 0;
2676 fail:
2677         BUG();
2678         return ret;
2679 }
2680
2681 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2682                   struct extent_buffer *buf, int full_backref)
2683 {
2684         return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
2685 }
2686
2687 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2688                   struct extent_buffer *buf, int full_backref)
2689 {
2690         return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
2691 }
2692
2693 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2694                                  struct btrfs_root *root,
2695                                  struct btrfs_path *path,
2696                                  struct btrfs_block_group_cache *cache)
2697 {
2698         int ret;
2699         struct btrfs_root *extent_root = root->fs_info->extent_root;
2700         unsigned long bi;
2701         struct extent_buffer *leaf;
2702
2703         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2704         if (ret < 0)
2705                 goto fail;
2706         BUG_ON(ret);
2707
2708         leaf = path->nodes[0];
2709         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2710         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2711         btrfs_mark_buffer_dirty(leaf);
2712         btrfs_release_path(extent_root, path);
2713 fail:
2714         if (ret)
2715                 return ret;
2716         return 0;
2717
2718 }
2719
2720 static struct btrfs_block_group_cache *
2721 next_block_group(struct btrfs_root *root,
2722                  struct btrfs_block_group_cache *cache)
2723 {
2724         struct rb_node *node;
2725         spin_lock(&root->fs_info->block_group_cache_lock);
2726         node = rb_next(&cache->cache_node);
2727         btrfs_put_block_group(cache);
2728         if (node) {
2729                 cache = rb_entry(node, struct btrfs_block_group_cache,
2730                                  cache_node);
2731                 btrfs_get_block_group(cache);
2732         } else
2733                 cache = NULL;
2734         spin_unlock(&root->fs_info->block_group_cache_lock);
2735         return cache;
2736 }
2737
2738 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
2739                             struct btrfs_trans_handle *trans,
2740                             struct btrfs_path *path)
2741 {
2742         struct btrfs_root *root = block_group->fs_info->tree_root;
2743         struct inode *inode = NULL;
2744         u64 alloc_hint = 0;
2745         int dcs = BTRFS_DC_ERROR;
2746         int num_pages = 0;
2747         int retries = 0;
2748         int ret = 0;
2749
2750         /*
2751          * If this block group is smaller than 100 megs don't bother caching the
2752          * block group.
2753          */
2754         if (block_group->key.offset < (100 * 1024 * 1024)) {
2755                 spin_lock(&block_group->lock);
2756                 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
2757                 spin_unlock(&block_group->lock);
2758                 return 0;
2759         }
2760
2761 again:
2762         inode = lookup_free_space_inode(root, block_group, path);
2763         if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
2764                 ret = PTR_ERR(inode);
2765                 btrfs_release_path(root, path);
2766                 goto out;
2767         }
2768
2769         if (IS_ERR(inode)) {
2770                 BUG_ON(retries);
2771                 retries++;
2772
2773                 if (block_group->ro)
2774                         goto out_free;
2775
2776                 ret = create_free_space_inode(root, trans, block_group, path);
2777                 if (ret)
2778                         goto out_free;
2779                 goto again;
2780         }
2781
2782         /*
2783          * We want to set the generation to 0, that way if anything goes wrong
2784          * from here on out we know not to trust this cache when we load up next
2785          * time.
2786          */
2787         BTRFS_I(inode)->generation = 0;
2788         ret = btrfs_update_inode(trans, root, inode);
2789         WARN_ON(ret);
2790
2791         if (i_size_read(inode) > 0) {
2792                 ret = btrfs_truncate_free_space_cache(root, trans, path,
2793                                                       inode);
2794                 if (ret)
2795                         goto out_put;
2796         }
2797
2798         spin_lock(&block_group->lock);
2799         if (block_group->cached != BTRFS_CACHE_FINISHED) {
2800                 /* We're not cached, don't bother trying to write stuff out */
2801                 dcs = BTRFS_DC_WRITTEN;
2802                 spin_unlock(&block_group->lock);
2803                 goto out_put;
2804         }
2805         spin_unlock(&block_group->lock);
2806
2807         num_pages = (int)div64_u64(block_group->key.offset, 1024 * 1024 * 1024);
2808         if (!num_pages)
2809                 num_pages = 1;
2810
2811         /*
2812          * Just to make absolutely sure we have enough space, we're going to
2813          * preallocate 12 pages worth of space for each block group.  In
2814          * practice we ought to use at most 8, but we need extra space so we can
2815          * add our header and have a terminator between the extents and the
2816          * bitmaps.
2817          */
2818         num_pages *= 16;
2819         num_pages *= PAGE_CACHE_SIZE;
2820
2821         ret = btrfs_check_data_free_space(inode, num_pages);
2822         if (ret)
2823                 goto out_put;
2824
2825         ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
2826                                               num_pages, num_pages,
2827                                               &alloc_hint);
2828         if (!ret)
2829                 dcs = BTRFS_DC_SETUP;
2830         btrfs_free_reserved_data_space(inode, num_pages);
2831 out_put:
2832         iput(inode);
2833 out_free:
2834         btrfs_release_path(root, path);
2835 out:
2836         spin_lock(&block_group->lock);
2837         block_group->disk_cache_state = dcs;
2838         spin_unlock(&block_group->lock);
2839
2840         return ret;
2841 }
2842
2843 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
2844                                    struct btrfs_root *root)
2845 {
2846         struct btrfs_block_group_cache *cache;
2847         int err = 0;
2848         struct btrfs_path *path;
2849         u64 last = 0;
2850
2851         path = btrfs_alloc_path();
2852         if (!path)
2853                 return -ENOMEM;
2854
2855 again:
2856         while (1) {
2857                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2858                 while (cache) {
2859                         if (cache->disk_cache_state == BTRFS_DC_CLEAR)
2860                                 break;
2861                         cache = next_block_group(root, cache);
2862                 }
2863                 if (!cache) {
2864                         if (last == 0)
2865                                 break;
2866                         last = 0;
2867                         continue;
2868                 }
2869                 err = cache_save_setup(cache, trans, path);
2870                 last = cache->key.objectid + cache->key.offset;
2871                 btrfs_put_block_group(cache);
2872         }
2873
2874         while (1) {
2875                 if (last == 0) {
2876                         err = btrfs_run_delayed_refs(trans, root,
2877                                                      (unsigned long)-1);
2878                         BUG_ON(err);
2879                 }
2880
2881                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2882                 while (cache) {
2883                         if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
2884                                 btrfs_put_block_group(cache);
2885                                 goto again;
2886                         }
2887
2888                         if (cache->dirty)
2889                                 break;
2890                         cache = next_block_group(root, cache);
2891                 }
2892                 if (!cache) {
2893                         if (last == 0)
2894                                 break;
2895                         last = 0;
2896                         continue;
2897                 }
2898
2899                 if (cache->disk_cache_state == BTRFS_DC_SETUP)
2900                         cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
2901                 cache->dirty = 0;
2902                 last = cache->key.objectid + cache->key.offset;
2903
2904                 err = write_one_cache_group(trans, root, path, cache);
2905                 BUG_ON(err);
2906                 btrfs_put_block_group(cache);
2907         }
2908
2909         while (1) {
2910                 /*
2911                  * I don't think this is needed since we're just marking our
2912                  * preallocated extent as written, but just in case it can't
2913                  * hurt.
2914                  */
2915                 if (last == 0) {
2916                         err = btrfs_run_delayed_refs(trans, root,
2917                                                      (unsigned long)-1);
2918                         BUG_ON(err);
2919                 }
2920
2921                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2922                 while (cache) {
2923                         /*
2924                          * Really this shouldn't happen, but it could if we
2925                          * couldn't write the entire preallocated extent and
2926                          * splitting the extent resulted in a new block.
2927                          */
2928                         if (cache->dirty) {
2929                                 btrfs_put_block_group(cache);
2930                                 goto again;
2931                         }
2932                         if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2933                                 break;
2934                         cache = next_block_group(root, cache);
2935                 }
2936                 if (!cache) {
2937                         if (last == 0)
2938                                 break;
2939                         last = 0;
2940                         continue;
2941                 }
2942
2943                 btrfs_write_out_cache(root, trans, cache, path);
2944
2945                 /*
2946                  * If we didn't have an error then the cache state is still
2947                  * NEED_WRITE, so we can set it to WRITTEN.
2948                  */
2949                 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2950                         cache->disk_cache_state = BTRFS_DC_WRITTEN;
2951                 last = cache->key.objectid + cache->key.offset;
2952                 btrfs_put_block_group(cache);
2953         }
2954
2955         btrfs_free_path(path);
2956         return 0;
2957 }
2958
2959 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
2960 {
2961         struct btrfs_block_group_cache *block_group;
2962         int readonly = 0;
2963
2964         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
2965         if (!block_group || block_group->ro)
2966                 readonly = 1;
2967         if (block_group)
2968                 btrfs_put_block_group(block_group);
2969         return readonly;
2970 }
2971
2972 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
2973                              u64 total_bytes, u64 bytes_used,
2974                              struct btrfs_space_info **space_info)
2975 {
2976         struct btrfs_space_info *found;
2977         int i;
2978         int factor;
2979
2980         if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
2981                      BTRFS_BLOCK_GROUP_RAID10))
2982                 factor = 2;
2983         else
2984                 factor = 1;
2985
2986         found = __find_space_info(info, flags);
2987         if (found) {
2988                 spin_lock(&found->lock);
2989                 found->total_bytes += total_bytes;
2990                 found->disk_total += total_bytes * factor;
2991                 found->bytes_used += bytes_used;
2992                 found->disk_used += bytes_used * factor;
2993                 found->full = 0;
2994                 spin_unlock(&found->lock);
2995                 *space_info = found;
2996                 return 0;
2997         }
2998         found = kzalloc(sizeof(*found), GFP_NOFS);
2999         if (!found)
3000                 return -ENOMEM;
3001
3002         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3003                 INIT_LIST_HEAD(&found->block_groups[i]);
3004         init_rwsem(&found->groups_sem);
3005         spin_lock_init(&found->lock);
3006         found->flags = flags & (BTRFS_BLOCK_GROUP_DATA |
3007                                 BTRFS_BLOCK_GROUP_SYSTEM |
3008                                 BTRFS_BLOCK_GROUP_METADATA);
3009         found->total_bytes = total_bytes;
3010         found->disk_total = total_bytes * factor;
3011         found->bytes_used = bytes_used;
3012         found->disk_used = bytes_used * factor;
3013         found->bytes_pinned = 0;
3014         found->bytes_reserved = 0;
3015         found->bytes_readonly = 0;
3016         found->bytes_may_use = 0;
3017         found->full = 0;
3018         found->force_alloc = 0;
3019         *space_info = found;
3020         list_add_rcu(&found->list, &info->space_info);
3021         atomic_set(&found->caching_threads, 0);
3022         return 0;
3023 }
3024
3025 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3026 {
3027         u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
3028                                    BTRFS_BLOCK_GROUP_RAID1 |
3029                                    BTRFS_BLOCK_GROUP_RAID10 |
3030                                    BTRFS_BLOCK_GROUP_DUP);
3031         if (extra_flags) {
3032                 if (flags & BTRFS_BLOCK_GROUP_DATA)
3033                         fs_info->avail_data_alloc_bits |= extra_flags;
3034                 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3035                         fs_info->avail_metadata_alloc_bits |= extra_flags;
3036                 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3037                         fs_info->avail_system_alloc_bits |= extra_flags;
3038         }
3039 }
3040
3041 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3042 {
3043         /*
3044          * we add in the count of missing devices because we want
3045          * to make sure that any RAID levels on a degraded FS
3046          * continue to be honored.
3047          */
3048         u64 num_devices = root->fs_info->fs_devices->rw_devices +
3049                 root->fs_info->fs_devices->missing_devices;
3050
3051         if (num_devices == 1)
3052                 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
3053         if (num_devices < 4)
3054                 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3055
3056         if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
3057             (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3058                       BTRFS_BLOCK_GROUP_RAID10))) {
3059                 flags &= ~BTRFS_BLOCK_GROUP_DUP;
3060         }
3061
3062         if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
3063             (flags & BTRFS_BLOCK_GROUP_RAID10)) {
3064                 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
3065         }
3066
3067         if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
3068             ((flags & BTRFS_BLOCK_GROUP_RAID1) |
3069              (flags & BTRFS_BLOCK_GROUP_RAID10) |
3070              (flags & BTRFS_BLOCK_GROUP_DUP)))
3071                 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
3072         return flags;
3073 }
3074
3075 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3076 {
3077         if (flags & BTRFS_BLOCK_GROUP_DATA)
3078                 flags |= root->fs_info->avail_data_alloc_bits &
3079                          root->fs_info->data_alloc_profile;
3080         else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3081                 flags |= root->fs_info->avail_system_alloc_bits &
3082                          root->fs_info->system_alloc_profile;
3083         else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3084                 flags |= root->fs_info->avail_metadata_alloc_bits &
3085                          root->fs_info->metadata_alloc_profile;
3086         return btrfs_reduce_alloc_profile(root, flags);
3087 }
3088
3089 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3090 {
3091         u64 flags;
3092
3093         if (data)
3094                 flags = BTRFS_BLOCK_GROUP_DATA;
3095         else if (root == root->fs_info->chunk_root)
3096                 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3097         else
3098                 flags = BTRFS_BLOCK_GROUP_METADATA;
3099
3100         return get_alloc_profile(root, flags);
3101 }
3102
3103 void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
3104 {
3105         BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
3106                                                        BTRFS_BLOCK_GROUP_DATA);
3107 }
3108
3109 /*
3110  * This will check the space that the inode allocates from to make sure we have
3111  * enough space for bytes.
3112  */
3113 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3114 {
3115         struct btrfs_space_info *data_sinfo;
3116         struct btrfs_root *root = BTRFS_I(inode)->root;
3117         u64 used;
3118         int ret = 0, committed = 0, alloc_chunk = 1;
3119
3120         /* make sure bytes are sectorsize aligned */
3121         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3122
3123         if (root == root->fs_info->tree_root) {
3124                 alloc_chunk = 0;
3125                 committed = 1;
3126         }
3127
3128         data_sinfo = BTRFS_I(inode)->space_info;
3129         if (!data_sinfo)
3130                 goto alloc;
3131
3132 again:
3133         /* make sure we have enough space to handle the data first */
3134         spin_lock(&data_sinfo->lock);
3135         used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3136                 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3137                 data_sinfo->bytes_may_use;
3138
3139         if (used + bytes > data_sinfo->total_bytes) {
3140                 struct btrfs_trans_handle *trans;
3141
3142                 /*
3143                  * if we don't have enough free bytes in this space then we need
3144                  * to alloc a new chunk.
3145                  */
3146                 if (!data_sinfo->full && alloc_chunk) {
3147                         u64 alloc_target;
3148
3149                         data_sinfo->force_alloc = 1;
3150                         spin_unlock(&data_sinfo->lock);
3151 alloc:
3152                         alloc_target = btrfs_get_alloc_profile(root, 1);
3153                         trans = btrfs_join_transaction(root, 1);
3154                         if (IS_ERR(trans))
3155                                 return PTR_ERR(trans);
3156
3157                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3158                                              bytes + 2 * 1024 * 1024,
3159                                              alloc_target, 0);
3160                         btrfs_end_transaction(trans, root);
3161                         if (ret < 0) {
3162                                 if (ret != -ENOSPC)
3163                                         return ret;
3164                                 else
3165                                         goto commit_trans;
3166                         }
3167
3168                         if (!data_sinfo) {
3169                                 btrfs_set_inode_space_info(root, inode);
3170                                 data_sinfo = BTRFS_I(inode)->space_info;
3171                         }
3172                         goto again;
3173                 }
3174                 spin_unlock(&data_sinfo->lock);
3175
3176                 /* commit the current transaction and try again */
3177 commit_trans:
3178                 if (!committed && !root->fs_info->open_ioctl_trans) {
3179                         committed = 1;
3180                         trans = btrfs_join_transaction(root, 1);
3181                         if (IS_ERR(trans))
3182                                 return PTR_ERR(trans);
3183                         ret = btrfs_commit_transaction(trans, root);
3184                         if (ret)
3185                                 return ret;
3186                         goto again;
3187                 }
3188
3189 #if 0 /* I hope we never need this code again, just in case */
3190                 printk(KERN_ERR "no space left, need %llu, %llu bytes_used, "
3191                        "%llu bytes_reserved, " "%llu bytes_pinned, "
3192                        "%llu bytes_readonly, %llu may use %llu total\n",
3193                        (unsigned long long)bytes,
3194                        (unsigned long long)data_sinfo->bytes_used,
3195                        (unsigned long long)data_sinfo->bytes_reserved,
3196                        (unsigned long long)data_sinfo->bytes_pinned,
3197                        (unsigned long long)data_sinfo->bytes_readonly,
3198                        (unsigned long long)data_sinfo->bytes_may_use,
3199                        (unsigned long long)data_sinfo->total_bytes);
3200 #endif
3201                 return -ENOSPC;
3202         }
3203         data_sinfo->bytes_may_use += bytes;
3204         BTRFS_I(inode)->reserved_bytes += bytes;
3205         spin_unlock(&data_sinfo->lock);
3206
3207         return 0;
3208 }
3209
3210 /*
3211  * called when we are clearing an delalloc extent from the
3212  * inode's io_tree or there was an error for whatever reason
3213  * after calling btrfs_check_data_free_space
3214  */
3215 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3216 {
3217         struct btrfs_root *root = BTRFS_I(inode)->root;
3218         struct btrfs_space_info *data_sinfo;
3219
3220         /* make sure bytes are sectorsize aligned */
3221         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3222
3223         data_sinfo = BTRFS_I(inode)->space_info;
3224         spin_lock(&data_sinfo->lock);
3225         data_sinfo->bytes_may_use -= bytes;
3226         BTRFS_I(inode)->reserved_bytes -= bytes;
3227         spin_unlock(&data_sinfo->lock);
3228 }
3229
3230 static void force_metadata_allocation(struct btrfs_fs_info *info)
3231 {
3232         struct list_head *head = &info->space_info;
3233         struct btrfs_space_info *found;
3234
3235         rcu_read_lock();
3236         list_for_each_entry_rcu(found, head, list) {
3237                 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3238                         found->force_alloc = 1;
3239         }
3240         rcu_read_unlock();
3241 }
3242
3243 static int should_alloc_chunk(struct btrfs_root *root,
3244                               struct btrfs_space_info *sinfo, u64 alloc_bytes)
3245 {
3246         u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3247         u64 thresh;
3248
3249         if (sinfo->bytes_used + sinfo->bytes_reserved +
3250             alloc_bytes + 256 * 1024 * 1024 < num_bytes)
3251                 return 0;
3252
3253         if (sinfo->bytes_used + sinfo->bytes_reserved +
3254             alloc_bytes < div_factor(num_bytes, 8))
3255                 return 0;
3256
3257         thresh = btrfs_super_total_bytes(&root->fs_info->super_copy);
3258         thresh = max_t(u64, 256 * 1024 * 1024, div_factor_fine(thresh, 5));
3259
3260         if (num_bytes > thresh && sinfo->bytes_used < div_factor(num_bytes, 3))
3261                 return 0;
3262
3263         return 1;
3264 }
3265
3266 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3267                           struct btrfs_root *extent_root, u64 alloc_bytes,
3268                           u64 flags, int force)
3269 {
3270         struct btrfs_space_info *space_info;
3271         struct btrfs_fs_info *fs_info = extent_root->fs_info;
3272         int ret = 0;
3273
3274         mutex_lock(&fs_info->chunk_mutex);
3275
3276         flags = btrfs_reduce_alloc_profile(extent_root, flags);
3277
3278         space_info = __find_space_info(extent_root->fs_info, flags);
3279         if (!space_info) {
3280                 ret = update_space_info(extent_root->fs_info, flags,
3281                                         0, 0, &space_info);
3282                 BUG_ON(ret);
3283         }
3284         BUG_ON(!space_info);
3285
3286         spin_lock(&space_info->lock);
3287         if (space_info->force_alloc)
3288                 force = 1;
3289         if (space_info->full) {
3290                 spin_unlock(&space_info->lock);
3291                 goto out;
3292         }
3293
3294         if (!force && !should_alloc_chunk(extent_root, space_info,
3295                                           alloc_bytes)) {
3296                 spin_unlock(&space_info->lock);
3297                 goto out;
3298         }
3299         spin_unlock(&space_info->lock);
3300
3301         /*
3302          * If we have mixed data/metadata chunks we want to make sure we keep
3303          * allocating mixed chunks instead of individual chunks.
3304          */
3305         if (btrfs_mixed_space_info(space_info))
3306                 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3307
3308         /*
3309          * if we're doing a data chunk, go ahead and make sure that
3310          * we keep a reasonable number of metadata chunks allocated in the
3311          * FS as well.
3312          */
3313         if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3314                 fs_info->data_chunk_allocations++;
3315                 if (!(fs_info->data_chunk_allocations %
3316                       fs_info->metadata_ratio))
3317                         force_metadata_allocation(fs_info);
3318         }
3319
3320         ret = btrfs_alloc_chunk(trans, extent_root, flags);
3321         spin_lock(&space_info->lock);
3322         if (ret)
3323                 space_info->full = 1;
3324         else
3325                 ret = 1;
3326         space_info->force_alloc = 0;
3327         spin_unlock(&space_info->lock);
3328 out:
3329         mutex_unlock(&extent_root->fs_info->chunk_mutex);
3330         return ret;
3331 }
3332
3333 /*
3334  * shrink metadata reservation for delalloc
3335  */
3336 static int shrink_delalloc(struct btrfs_trans_handle *trans,
3337                            struct btrfs_root *root, u64 to_reclaim, int sync)
3338 {
3339         struct btrfs_block_rsv *block_rsv;
3340         struct btrfs_space_info *space_info;
3341         u64 reserved;
3342         u64 max_reclaim;
3343         u64 reclaimed = 0;
3344         long time_left;
3345         int nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
3346         int loops = 0;
3347         unsigned long progress;
3348
3349         block_rsv = &root->fs_info->delalloc_block_rsv;
3350         space_info = block_rsv->space_info;
3351
3352         smp_mb();
3353         reserved = space_info->bytes_reserved;
3354         progress = space_info->reservation_progress;
3355
3356         if (reserved == 0)
3357                 return 0;
3358
3359         max_reclaim = min(reserved, to_reclaim);
3360
3361         while (loops < 1024) {
3362                 /* have the flusher threads jump in and do some IO */
3363                 smp_mb();
3364                 nr_pages = min_t(unsigned long, nr_pages,
3365                        root->fs_info->delalloc_bytes >> PAGE_CACHE_SHIFT);
3366                 writeback_inodes_sb_nr_if_idle(root->fs_info->sb, nr_pages);
3367
3368                 spin_lock(&space_info->lock);
3369                 if (reserved > space_info->bytes_reserved)
3370                         reclaimed += reserved - space_info->bytes_reserved;
3371                 reserved = space_info->bytes_reserved;
3372                 spin_unlock(&space_info->lock);
3373
3374                 loops++;
3375
3376                 if (reserved == 0 || reclaimed >= max_reclaim)
3377                         break;
3378
3379                 if (trans && trans->transaction->blocked)
3380                         return -EAGAIN;
3381
3382                 time_left = schedule_timeout_interruptible(1);
3383
3384                 /* We were interrupted, exit */
3385                 if (time_left)
3386                         break;
3387
3388                 /* we've kicked the IO a few times, if anything has been freed,
3389                  * exit.  There is no sense in looping here for a long time
3390                  * when we really need to commit the transaction, or there are
3391                  * just too many writers without enough free space
3392                  */
3393
3394                 if (loops > 3) {
3395                         smp_mb();
3396                         if (progress != space_info->reservation_progress)
3397                                 break;
3398                 }
3399
3400         }
3401         return reclaimed >= to_reclaim;
3402 }
3403
3404 /*
3405  * Retries tells us how many times we've called reserve_metadata_bytes.  The
3406  * idea is if this is the first call (retries == 0) then we will add to our
3407  * reserved count if we can't make the allocation in order to hold our place
3408  * while we go and try and free up space.  That way for retries > 1 we don't try
3409  * and add space, we just check to see if the amount of unused space is >= the
3410  * total space, meaning that our reservation is valid.
3411  *
3412  * However if we don't intend to retry this reservation, pass -1 as retries so
3413  * that it short circuits this logic.
3414  */
3415 static int reserve_metadata_bytes(struct btrfs_trans_handle *trans,
3416                                   struct btrfs_root *root,
3417                                   struct btrfs_block_rsv *block_rsv,
3418                                   u64 orig_bytes, int flush)
3419 {
3420         struct btrfs_space_info *space_info = block_rsv->space_info;
3421         u64 unused;
3422         u64 num_bytes = orig_bytes;
3423         int retries = 0;
3424         int ret = 0;
3425         bool reserved = false;
3426         bool committed = false;
3427
3428 again:
3429         ret = -ENOSPC;
3430         if (reserved)
3431                 num_bytes = 0;
3432
3433         spin_lock(&space_info->lock);
3434         unused = space_info->bytes_used + space_info->bytes_reserved +
3435                  space_info->bytes_pinned + space_info->bytes_readonly +
3436                  space_info->bytes_may_use;
3437
3438         /*
3439          * The idea here is that we've not already over-reserved the block group
3440          * then we can go ahead and save our reservation first and then start
3441          * flushing if we need to.  Otherwise if we've already overcommitted
3442          * lets start flushing stuff first and then come back and try to make
3443          * our reservation.
3444          */
3445         if (unused <= space_info->total_bytes) {
3446                 unused = space_info->total_bytes - unused;
3447                 if (unused >= num_bytes) {
3448                         if (!reserved)
3449                                 space_info->bytes_reserved += orig_bytes;
3450                         ret = 0;
3451                 } else {
3452                         /*
3453                          * Ok set num_bytes to orig_bytes since we aren't
3454                          * overocmmitted, this way we only try and reclaim what
3455                          * we need.
3456                          */
3457                         num_bytes = orig_bytes;
3458                 }
3459         } else {
3460                 /*
3461                  * Ok we're over committed, set num_bytes to the overcommitted
3462                  * amount plus the amount of bytes that we need for this
3463                  * reservation.
3464                  */
3465                 num_bytes = unused - space_info->total_bytes +
3466                         (orig_bytes * (retries + 1));
3467         }
3468
3469         /*
3470          * Couldn't make our reservation, save our place so while we're trying
3471          * to reclaim space we can actually use it instead of somebody else
3472          * stealing it from us.
3473          */
3474         if (ret && !reserved) {
3475                 space_info->bytes_reserved += orig_bytes;
3476                 reserved = true;
3477         }
3478
3479         spin_unlock(&space_info->lock);
3480
3481         if (!ret)
3482                 return 0;
3483
3484         if (!flush)
3485                 goto out;
3486
3487         /*
3488          * We do synchronous shrinking since we don't actually unreserve
3489          * metadata until after the IO is completed.
3490          */
3491         ret = shrink_delalloc(trans, root, num_bytes, 1);
3492         if (ret > 0)
3493                 return 0;
3494         else if (ret < 0)
3495                 goto out;
3496
3497         /*
3498          * So if we were overcommitted it's possible that somebody else flushed
3499          * out enough space and we simply didn't have enough space to reclaim,
3500          * so go back around and try again.
3501          */
3502         if (retries < 2) {
3503                 retries++;
3504                 goto again;
3505         }
3506
3507         spin_lock(&space_info->lock);
3508         /*
3509          * Not enough space to be reclaimed, don't bother committing the
3510          * transaction.
3511          */
3512         if (space_info->bytes_pinned < orig_bytes)
3513                 ret = -ENOSPC;
3514         spin_unlock(&space_info->lock);
3515         if (ret)
3516                 goto out;
3517
3518         ret = -EAGAIN;
3519         if (trans || committed)
3520                 goto out;
3521
3522         ret = -ENOSPC;
3523         trans = btrfs_join_transaction(root, 1);
3524         if (IS_ERR(trans))
3525                 goto out;
3526         ret = btrfs_commit_transaction(trans, root);
3527         if (!ret) {
3528                 trans = NULL;
3529                 committed = true;
3530                 goto again;
3531         }
3532
3533 out:
3534         if (reserved) {
3535                 spin_lock(&space_info->lock);
3536                 space_info->bytes_reserved -= orig_bytes;
3537                 spin_unlock(&space_info->lock);
3538         }
3539
3540         return ret;
3541 }
3542
3543 static struct btrfs_block_rsv *get_block_rsv(struct btrfs_trans_handle *trans,
3544                                              struct btrfs_root *root)
3545 {
3546         struct btrfs_block_rsv *block_rsv;
3547         if (root->ref_cows)
3548                 block_rsv = trans->block_rsv;
3549         else
3550                 block_rsv = root->block_rsv;
3551
3552         if (!block_rsv)
3553                 block_rsv = &root->fs_info->empty_block_rsv;
3554
3555         return block_rsv;
3556 }
3557
3558 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
3559                                u64 num_bytes)
3560 {
3561         int ret = -ENOSPC;
3562         spin_lock(&block_rsv->lock);
3563         if (block_rsv->reserved >= num_bytes) {
3564                 block_rsv->reserved -= num_bytes;
3565                 if (block_rsv->reserved < block_rsv->size)
3566                         block_rsv->full = 0;
3567                 ret = 0;
3568         }
3569         spin_unlock(&block_rsv->lock);
3570         return ret;
3571 }
3572
3573 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
3574                                 u64 num_bytes, int update_size)
3575 {
3576         spin_lock(&block_rsv->lock);
3577         block_rsv->reserved += num_bytes;
3578         if (update_size)
3579                 block_rsv->size += num_bytes;
3580         else if (block_rsv->reserved >= block_rsv->size)
3581                 block_rsv->full = 1;
3582         spin_unlock(&block_rsv->lock);
3583 }
3584
3585 void block_rsv_release_bytes(struct btrfs_block_rsv *block_rsv,
3586                              struct btrfs_block_rsv *dest, u64 num_bytes)
3587 {
3588         struct btrfs_space_info *space_info = block_rsv->space_info;
3589
3590         spin_lock(&block_rsv->lock);
3591         if (num_bytes == (u64)-1)
3592                 num_bytes = block_rsv->size;
3593         block_rsv->size -= num_bytes;
3594         if (block_rsv->reserved >= block_rsv->size) {
3595                 num_bytes = block_rsv->reserved - block_rsv->size;
3596                 block_rsv->reserved = block_rsv->size;
3597                 block_rsv->full = 1;
3598         } else {
3599                 num_bytes = 0;
3600         }
3601         spin_unlock(&block_rsv->lock);
3602
3603         if (num_bytes > 0) {
3604                 if (dest) {
3605                         spin_lock(&dest->lock);
3606                         if (!dest->full) {
3607                                 u64 bytes_to_add;
3608
3609                                 bytes_to_add = dest->size - dest->reserved;
3610                                 bytes_to_add = min(num_bytes, bytes_to_add);
3611                                 dest->reserved += bytes_to_add;
3612                                 if (dest->reserved >= dest->size)
3613                                         dest->full = 1;
3614                                 num_bytes -= bytes_to_add;
3615                         }
3616                         spin_unlock(&dest->lock);
3617                 }
3618                 if (num_bytes) {
3619                         spin_lock(&space_info->lock);
3620                         space_info->bytes_reserved -= num_bytes;
3621                         space_info->reservation_progress++;
3622                         spin_unlock(&space_info->lock);
3623                 }
3624         }
3625 }
3626
3627 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
3628                                    struct btrfs_block_rsv *dst, u64 num_bytes)
3629 {
3630         int ret;
3631
3632         ret = block_rsv_use_bytes(src, num_bytes);
3633         if (ret)
3634                 return ret;
3635
3636         block_rsv_add_bytes(dst, num_bytes, 1);
3637         return 0;
3638 }
3639
3640 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv)
3641 {
3642         memset(rsv, 0, sizeof(*rsv));
3643         spin_lock_init(&rsv->lock);
3644         atomic_set(&rsv->usage, 1);
3645         rsv->priority = 6;
3646         INIT_LIST_HEAD(&rsv->list);
3647 }
3648
3649 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root)
3650 {
3651         struct btrfs_block_rsv *block_rsv;
3652         struct btrfs_fs_info *fs_info = root->fs_info;
3653
3654         block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
3655         if (!block_rsv)
3656                 return NULL;
3657
3658         btrfs_init_block_rsv(block_rsv);
3659         block_rsv->space_info = __find_space_info(fs_info,
3660                                                   BTRFS_BLOCK_GROUP_METADATA);
3661         return block_rsv;
3662 }
3663
3664 void btrfs_free_block_rsv(struct btrfs_root *root,
3665                           struct btrfs_block_rsv *rsv)
3666 {
3667         if (rsv && atomic_dec_and_test(&rsv->usage)) {
3668                 btrfs_block_rsv_release(root, rsv, (u64)-1);
3669                 if (!rsv->durable)
3670                         kfree(rsv);
3671         }
3672 }
3673
3674 /*
3675  * make the block_rsv struct be able to capture freed space.
3676  * the captured space will re-add to the the block_rsv struct
3677  * after transaction commit
3678  */
3679 void btrfs_add_durable_block_rsv(struct btrfs_fs_info *fs_info,
3680                                  struct btrfs_block_rsv *block_rsv)
3681 {
3682         block_rsv->durable = 1;
3683         mutex_lock(&fs_info->durable_block_rsv_mutex);
3684         list_add_tail(&block_rsv->list, &fs_info->durable_block_rsv_list);
3685         mutex_unlock(&fs_info->durable_block_rsv_mutex);
3686 }
3687
3688 int btrfs_block_rsv_add(struct btrfs_trans_handle *trans,
3689                         struct btrfs_root *root,
3690                         struct btrfs_block_rsv *block_rsv,
3691                         u64 num_bytes)
3692 {
3693         int ret;
3694
3695         if (num_bytes == 0)
3696                 return 0;
3697
3698         ret = reserve_metadata_bytes(trans, root, block_rsv, num_bytes, 1);
3699         if (!ret) {
3700                 block_rsv_add_bytes(block_rsv, num_bytes, 1);
3701                 return 0;
3702         }
3703
3704         return ret;
3705 }
3706
3707 int btrfs_block_rsv_check(struct btrfs_trans_handle *trans,
3708                           struct btrfs_root *root,
3709                           struct btrfs_block_rsv *block_rsv,
3710                           u64 min_reserved, int min_factor)
3711 {
3712         u64 num_bytes = 0;
3713         int commit_trans = 0;
3714         int ret = -ENOSPC;
3715
3716         if (!block_rsv)
3717                 return 0;
3718
3719         spin_lock(&block_rsv->lock);
3720         if (min_factor > 0)
3721                 num_bytes = div_factor(block_rsv->size, min_factor);
3722         if (min_reserved > num_bytes)
3723                 num_bytes = min_reserved;
3724
3725         if (block_rsv->reserved >= num_bytes) {
3726                 ret = 0;
3727         } else {
3728                 num_bytes -= block_rsv->reserved;
3729                 if (block_rsv->durable &&
3730                     block_rsv->freed[0] + block_rsv->freed[1] >= num_bytes)
3731                         commit_trans = 1;
3732         }
3733         spin_unlock(&block_rsv->lock);
3734         if (!ret)
3735                 return 0;
3736
3737         if (block_rsv->refill_used) {
3738                 ret = reserve_metadata_bytes(trans, root, block_rsv,
3739                                              num_bytes, 0);
3740                 if (!ret) {
3741                         block_rsv_add_bytes(block_rsv, num_bytes, 0);
3742                         return 0;
3743                 }
3744         }
3745
3746         if (commit_trans) {
3747                 if (trans)
3748                         return -EAGAIN;
3749
3750                 trans = btrfs_join_transaction(root, 1);
3751                 BUG_ON(IS_ERR(trans));
3752                 ret = btrfs_commit_transaction(trans, root);
3753                 return 0;
3754         }
3755
3756         return -ENOSPC;
3757 }
3758
3759 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
3760                             struct btrfs_block_rsv *dst_rsv,
3761                             u64 num_bytes)
3762 {
3763         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3764 }
3765
3766 void btrfs_block_rsv_release(struct btrfs_root *root,
3767                              struct btrfs_block_rsv *block_rsv,
3768                              u64 num_bytes)
3769 {
3770         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3771         if (global_rsv->full || global_rsv == block_rsv ||
3772             block_rsv->space_info != global_rsv->space_info)
3773                 global_rsv = NULL;
3774         block_rsv_release_bytes(block_rsv, global_rsv, num_bytes);
3775 }
3776
3777 /*
3778  * helper to calculate size of global block reservation.
3779  * the desired value is sum of space used by extent tree,
3780  * checksum tree and root tree
3781  */
3782 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
3783 {
3784         struct btrfs_space_info *sinfo;
3785         u64 num_bytes;
3786         u64 meta_used;
3787         u64 data_used;
3788         int csum_size = btrfs_super_csum_size(&fs_info->super_copy);
3789 #if 0
3790         /*
3791          * per tree used space accounting can be inaccuracy, so we
3792          * can't rely on it.
3793          */
3794         spin_lock(&fs_info->extent_root->accounting_lock);
3795         num_bytes = btrfs_root_used(&fs_info->extent_root->root_item);
3796         spin_unlock(&fs_info->extent_root->accounting_lock);
3797
3798         spin_lock(&fs_info->csum_root->accounting_lock);
3799         num_bytes += btrfs_root_used(&fs_info->csum_root->root_item);
3800         spin_unlock(&fs_info->csum_root->accounting_lock);
3801
3802         spin_lock(&fs_info->tree_root->accounting_lock);
3803         num_bytes += btrfs_root_used(&fs_info->tree_root->root_item);
3804         spin_unlock(&fs_info->tree_root->accounting_lock);
3805 #endif
3806         sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
3807         spin_lock(&sinfo->lock);
3808         data_used = sinfo->bytes_used;
3809         spin_unlock(&sinfo->lock);
3810
3811         sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3812         spin_lock(&sinfo->lock);
3813         if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
3814                 data_used = 0;
3815         meta_used = sinfo->bytes_used;
3816         spin_unlock(&sinfo->lock);
3817
3818         num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
3819                     csum_size * 2;
3820         num_bytes += div64_u64(data_used + meta_used, 50);
3821
3822         if (num_bytes * 3 > meta_used)
3823                 num_bytes = div64_u64(meta_used, 3);
3824
3825         return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
3826 }
3827
3828 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
3829 {
3830         struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
3831         struct btrfs_space_info *sinfo = block_rsv->space_info;
3832         u64 num_bytes;
3833
3834         num_bytes = calc_global_metadata_size(fs_info);
3835
3836         spin_lock(&block_rsv->lock);
3837         spin_lock(&sinfo->lock);
3838
3839         block_rsv->size = num_bytes;
3840
3841         num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
3842                     sinfo->bytes_reserved + sinfo->bytes_readonly +
3843                     sinfo->bytes_may_use;
3844
3845         if (sinfo->total_bytes > num_bytes) {
3846                 num_bytes = sinfo->total_bytes - num_bytes;
3847                 block_rsv->reserved += num_bytes;
3848                 sinfo->bytes_reserved += num_bytes;
3849         }
3850
3851         if (block_rsv->reserved >= block_rsv->size) {
3852                 num_bytes = block_rsv->reserved - block_rsv->size;
3853                 sinfo->bytes_reserved -= num_bytes;
3854                 sinfo->reservation_progress++;
3855                 block_rsv->reserved = block_rsv->size;
3856                 block_rsv->full = 1;
3857         }
3858 #if 0
3859         printk(KERN_INFO"global block rsv size %llu reserved %llu\n",
3860                 block_rsv->size, block_rsv->reserved);
3861 #endif
3862         spin_unlock(&sinfo->lock);
3863         spin_unlock(&block_rsv->lock);
3864 }
3865
3866 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
3867 {
3868         struct btrfs_space_info *space_info;
3869
3870         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3871         fs_info->chunk_block_rsv.space_info = space_info;
3872         fs_info->chunk_block_rsv.priority = 10;
3873
3874         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3875         fs_info->global_block_rsv.space_info = space_info;
3876         fs_info->global_block_rsv.priority = 10;
3877         fs_info->global_block_rsv.refill_used = 1;
3878         fs_info->delalloc_block_rsv.space_info = space_info;
3879         fs_info->trans_block_rsv.space_info = space_info;
3880         fs_info->empty_block_rsv.space_info = space_info;
3881         fs_info->empty_block_rsv.priority = 10;
3882
3883         fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
3884         fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
3885         fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
3886         fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
3887         fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
3888
3889         btrfs_add_durable_block_rsv(fs_info, &fs_info->global_block_rsv);
3890
3891         btrfs_add_durable_block_rsv(fs_info, &fs_info->delalloc_block_rsv);
3892
3893         update_global_block_rsv(fs_info);
3894 }
3895
3896 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
3897 {
3898         block_rsv_release_bytes(&fs_info->global_block_rsv, NULL, (u64)-1);
3899         WARN_ON(fs_info->delalloc_block_rsv.size > 0);
3900         WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
3901         WARN_ON(fs_info->trans_block_rsv.size > 0);
3902         WARN_ON(fs_info->trans_block_rsv.reserved > 0);
3903         WARN_ON(fs_info->chunk_block_rsv.size > 0);
3904         WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
3905 }
3906
3907 static u64 calc_trans_metadata_size(struct btrfs_root *root, int num_items)
3908 {
3909         return (root->leafsize + root->nodesize * (BTRFS_MAX_LEVEL - 1)) *
3910                 3 * num_items;
3911 }
3912
3913 int btrfs_trans_reserve_metadata(struct btrfs_trans_handle *trans,
3914                                  struct btrfs_root *root,
3915                                  int num_items)
3916 {
3917         u64 num_bytes;
3918         int ret;
3919
3920         if (num_items == 0 || root->fs_info->chunk_root == root)
3921                 return 0;
3922
3923         num_bytes = calc_trans_metadata_size(root, num_items);
3924         ret = btrfs_block_rsv_add(trans, root, &root->fs_info->trans_block_rsv,
3925                                   num_bytes);
3926         if (!ret) {
3927                 trans->bytes_reserved += num_bytes;
3928                 trans->block_rsv = &root->fs_info->trans_block_rsv;
3929         }
3930         return ret;
3931 }
3932
3933 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
3934                                   struct btrfs_root *root)
3935 {
3936         if (!trans->bytes_reserved)
3937                 return;
3938
3939         BUG_ON(trans->block_rsv != &root->fs_info->trans_block_rsv);
3940         btrfs_block_rsv_release(root, trans->block_rsv,
3941                                 trans->bytes_reserved);
3942         trans->bytes_reserved = 0;
3943 }
3944
3945 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
3946                                   struct inode *inode)
3947 {
3948         struct btrfs_root *root = BTRFS_I(inode)->root;
3949         struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
3950         struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
3951
3952         /*
3953          * one for deleting orphan item, one for updating inode and
3954          * two for calling btrfs_truncate_inode_items.
3955          *
3956          * btrfs_truncate_inode_items is a delete operation, it frees
3957          * more space than it uses in most cases. So two units of
3958          * metadata space should be enough for calling it many times.
3959          * If all of the metadata space is used, we can commit
3960          * transaction and use space it freed.
3961          */
3962         u64 num_bytes = calc_trans_metadata_size(root, 4);
3963         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3964 }
3965
3966 void btrfs_orphan_release_metadata(struct inode *inode)
3967 {
3968         struct btrfs_root *root = BTRFS_I(inode)->root;
3969         u64 num_bytes = calc_trans_metadata_size(root, 4);
3970         btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
3971 }
3972
3973 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
3974                                 struct btrfs_pending_snapshot *pending)
3975 {
3976         struct btrfs_root *root = pending->root;
3977         struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
3978         struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
3979         /*
3980          * two for root back/forward refs, two for directory entries
3981          * and one for root of the snapshot.
3982          */
3983         u64 num_bytes = calc_trans_metadata_size(root, 5);
3984         dst_rsv->space_info = src_rsv->space_info;
3985         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3986 }
3987
3988 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes)
3989 {
3990         return num_bytes >>= 3;
3991 }
3992
3993 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
3994 {
3995         struct btrfs_root *root = BTRFS_I(inode)->root;
3996         struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
3997         u64 to_reserve;
3998         int nr_extents;
3999         int ret;
4000
4001         if (btrfs_transaction_in_commit(root->fs_info))
4002                 schedule_timeout(1);
4003
4004         num_bytes = ALIGN(num_bytes, root->sectorsize);
4005
4006         spin_lock(&BTRFS_I(inode)->accounting_lock);
4007         nr_extents = atomic_read(&BTRFS_I(inode)->outstanding_extents) + 1;
4008         if (nr_extents > BTRFS_I(inode)->reserved_extents) {
4009                 nr_extents -= BTRFS_I(inode)->reserved_extents;
4010                 to_reserve = calc_trans_metadata_size(root, nr_extents);
4011         } else {
4012                 nr_extents = 0;
4013                 to_reserve = 0;
4014         }
4015         spin_unlock(&BTRFS_I(inode)->accounting_lock);
4016         to_reserve += calc_csum_metadata_size(inode, num_bytes);
4017         ret = reserve_metadata_bytes(NULL, root, block_rsv, to_reserve, 1);
4018         if (ret)
4019                 return ret;
4020
4021         spin_lock(&BTRFS_I(inode)->accounting_lock);
4022         BTRFS_I(inode)->reserved_extents += nr_extents;
4023         atomic_inc(&BTRFS_I(inode)->outstanding_extents);
4024         spin_unlock(&BTRFS_I(inode)->accounting_lock);
4025
4026         block_rsv_add_bytes(block_rsv, to_reserve, 1);
4027
4028         if (block_rsv->size > 512 * 1024 * 1024)
4029                 shrink_delalloc(NULL, root, to_reserve, 0);
4030
4031         return 0;
4032 }
4033
4034 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
4035 {
4036         struct btrfs_root *root = BTRFS_I(inode)->root;
4037         u64 to_free;
4038         int nr_extents;
4039
4040         num_bytes = ALIGN(num_bytes, root->sectorsize);
4041         atomic_dec(&BTRFS_I(inode)->outstanding_extents);
4042         WARN_ON(atomic_read(&BTRFS_I(inode)->outstanding_extents) < 0);
4043
4044         spin_lock(&BTRFS_I(inode)->accounting_lock);
4045         nr_extents = atomic_read(&BTRFS_I(inode)->outstanding_extents);
4046         if (nr_extents < BTRFS_I(inode)->reserved_extents) {
4047                 nr_extents = BTRFS_I(inode)->reserved_extents - nr_extents;
4048                 BTRFS_I(inode)->reserved_extents -= nr_extents;
4049         } else {
4050                 nr_extents = 0;
4051         }
4052         spin_unlock(&BTRFS_I(inode)->accounting_lock);
4053
4054         to_free = calc_csum_metadata_size(inode, num_bytes);
4055         if (nr_extents > 0)
4056                 to_free += calc_trans_metadata_size(root, nr_extents);
4057
4058         btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
4059                                 to_free);
4060 }
4061
4062 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
4063 {
4064         int ret;
4065
4066         ret = btrfs_check_data_free_space(inode, num_bytes);
4067         if (ret)
4068                 return ret;
4069
4070         ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
4071         if (ret) {
4072                 btrfs_free_reserved_data_space(inode, num_bytes);
4073                 return ret;
4074         }
4075
4076         return 0;
4077 }
4078
4079 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
4080 {
4081         btrfs_delalloc_release_metadata(inode, num_bytes);
4082         btrfs_free_reserved_data_space(inode, num_bytes);
4083 }
4084
4085 static int update_block_group(struct btrfs_trans_handle *trans,
4086                               struct btrfs_root *root,
4087                               u64 bytenr, u64 num_bytes, int alloc)
4088 {
4089         struct btrfs_block_group_cache *cache = NULL;
4090         struct btrfs_fs_info *info = root->fs_info;
4091         u64 total = num_bytes;
4092         u64 old_val;
4093         u64 byte_in_group;
4094         int factor;
4095
4096         /* block accounting for super block */
4097         spin_lock(&info->delalloc_lock);
4098         old_val = btrfs_super_bytes_used(&info->super_copy);
4099         if (alloc)
4100                 old_val += num_bytes;
4101         else
4102                 old_val -= num_bytes;
4103         btrfs_set_super_bytes_used(&info->super_copy, old_val);
4104         spin_unlock(&info->delalloc_lock);
4105
4106         while (total) {
4107                 cache = btrfs_lookup_block_group(info, bytenr);
4108                 if (!cache)
4109                         return -1;
4110                 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
4111                                     BTRFS_BLOCK_GROUP_RAID1 |
4112                                     BTRFS_BLOCK_GROUP_RAID10))
4113                         factor = 2;
4114                 else
4115                         factor = 1;
4116                 /*
4117                  * If this block group has free space cache written out, we
4118                  * need to make sure to load it if we are removing space.  This
4119                  * is because we need the unpinning stage to actually add the
4120                  * space back to the block group, otherwise we will leak space.
4121                  */
4122                 if (!alloc && cache->cached == BTRFS_CACHE_NO)
4123                         cache_block_group(cache, trans, NULL, 1);
4124
4125                 byte_in_group = bytenr - cache->key.objectid;
4126                 WARN_ON(byte_in_group > cache->key.offset);
4127
4128                 spin_lock(&cache->space_info->lock);
4129                 spin_lock(&cache->lock);
4130
4131                 if (btrfs_super_cache_generation(&info->super_copy) != 0 &&
4132                     cache->disk_cache_state < BTRFS_DC_CLEAR)
4133                         cache->disk_cache_state = BTRFS_DC_CLEAR;
4134
4135                 cache->dirty = 1;
4136                 old_val = btrfs_block_group_used(&cache->item);
4137                 num_bytes = min(total, cache->key.offset - byte_in_group);
4138                 if (alloc) {
4139                         old_val += num_bytes;
4140                         btrfs_set_block_group_used(&cache->item, old_val);
4141                         cache->reserved -= num_bytes;
4142                         cache->space_info->bytes_reserved -= num_bytes;
4143                         cache->space_info->reservation_progress++;
4144                         cache->space_info->bytes_used += num_bytes;
4145                         cache->space_info->disk_used += num_bytes * factor;
4146                         spin_unlock(&cache->lock);
4147                         spin_unlock(&cache->space_info->lock);
4148                 } else {
4149                         old_val -= num_bytes;
4150                         btrfs_set_block_group_used(&cache->item, old_val);
4151                         cache->pinned += num_bytes;
4152                         cache->space_info->bytes_pinned += num_bytes;
4153                         cache->space_info->bytes_used -= num_bytes;
4154                         cache->space_info->disk_used -= num_bytes * factor;
4155                         spin_unlock(&cache->lock);
4156                         spin_unlock(&cache->space_info->lock);
4157
4158                         set_extent_dirty(info->pinned_extents,
4159                                          bytenr, bytenr + num_bytes - 1,
4160                                          GFP_NOFS | __GFP_NOFAIL);
4161                 }
4162                 btrfs_put_block_group(cache);
4163                 total -= num_bytes;
4164                 bytenr += num_bytes;
4165         }
4166         return 0;
4167 }
4168
4169 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
4170 {
4171         struct btrfs_block_group_cache *cache;
4172         u64 bytenr;
4173
4174         cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
4175         if (!cache)
4176                 return 0;
4177
4178         bytenr = cache->key.objectid;
4179         btrfs_put_block_group(cache);
4180
4181         return bytenr;
4182 }
4183
4184 static int pin_down_extent(struct btrfs_root *root,
4185                            struct btrfs_block_group_cache *cache,
4186                            u64 bytenr, u64 num_bytes, int reserved)
4187 {
4188         spin_lock(&cache->space_info->lock);
4189         spin_lock(&cache->lock);
4190         cache->pinned += num_bytes;
4191         cache->space_info->bytes_pinned += num_bytes;
4192         if (reserved) {
4193                 cache->reserved -= num_bytes;
4194                 cache->space_info->bytes_reserved -= num_bytes;
4195                 cache->space_info->reservation_progress++;
4196         }
4197         spin_unlock(&cache->lock);
4198         spin_unlock(&cache->space_info->lock);
4199
4200         set_extent_dirty(root->fs_info->pinned_extents, bytenr,
4201                          bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
4202         return 0;
4203 }
4204
4205 /*
4206  * this function must be called within transaction
4207  */
4208 int btrfs_pin_extent(struct btrfs_root *root,
4209                      u64 bytenr, u64 num_bytes, int reserved)
4210 {
4211         struct btrfs_block_group_cache *cache;
4212
4213         cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4214         BUG_ON(!cache);
4215
4216         pin_down_extent(root, cache, bytenr, num_bytes, reserved);
4217
4218         btrfs_put_block_group(cache);
4219         return 0;
4220 }
4221
4222 /*
4223  * update size of reserved extents. this function may return -EAGAIN
4224  * if 'reserve' is true or 'sinfo' is false.
4225  */
4226 static int update_reserved_bytes(struct btrfs_block_group_cache *cache,
4227                                  u64 num_bytes, int reserve, int sinfo)
4228 {
4229         int ret = 0;
4230         if (sinfo) {
4231                 struct btrfs_space_info *space_info = cache->space_info;
4232                 spin_lock(&space_info->lock);
4233                 spin_lock(&cache->lock);
4234                 if (reserve) {
4235                         if (cache->ro) {
4236                                 ret = -EAGAIN;
4237                         } else {
4238                                 cache->reserved += num_bytes;
4239                                 space_info->bytes_reserved += num_bytes;
4240                         }
4241                 } else {
4242                         if (cache->ro)
4243                                 space_info->bytes_readonly += num_bytes;
4244                         cache->reserved -= num_bytes;
4245                         space_info->bytes_reserved -= num_bytes;
4246                         space_info->reservation_progress++;
4247                 }
4248                 spin_unlock(&cache->lock);
4249                 spin_unlock(&space_info->lock);
4250         } else {
4251                 spin_lock(&cache->lock);
4252                 if (cache->ro) {
4253                         ret = -EAGAIN;
4254                 } else {
4255                         if (reserve)
4256                                 cache->reserved += num_bytes;
4257                         else
4258                                 cache->reserved -= num_bytes;
4259                 }
4260                 spin_unlock(&cache->lock);
4261         }
4262         return ret;
4263 }
4264
4265 int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
4266                                 struct btrfs_root *root)
4267 {
4268         struct btrfs_fs_info *fs_info = root->fs_info;
4269         struct btrfs_caching_control *next;
4270         struct btrfs_caching_control *caching_ctl;
4271         struct btrfs_block_group_cache *cache;
4272
4273         down_write(&fs_info->extent_commit_sem);
4274
4275         list_for_each_entry_safe(caching_ctl, next,
4276                                  &fs_info->caching_block_groups, list) {
4277                 cache = caching_ctl->block_group;
4278                 if (block_group_cache_done(cache)) {
4279                         cache->last_byte_to_unpin = (u64)-1;
4280                         list_del_init(&caching_ctl->list);
4281                         put_caching_control(caching_ctl);
4282                 } else {
4283                         cache->last_byte_to_unpin = caching_ctl->progress;
4284                 }
4285         }
4286
4287         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4288                 fs_info->pinned_extents = &fs_info->freed_extents[1];
4289         else
4290                 fs_info->pinned_extents = &fs_info->freed_extents[0];
4291
4292         up_write(&fs_info->extent_commit_sem);
4293
4294         update_global_block_rsv(fs_info);
4295         return 0;
4296 }
4297
4298 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
4299 {
4300         struct btrfs_fs_info *fs_info = root->fs_info;
4301         struct btrfs_block_group_cache *cache = NULL;
4302         u64 len;
4303
4304         while (start <= end) {
4305                 if (!cache ||
4306                     start >= cache->key.objectid + cache->key.offset) {
4307                         if (cache)
4308                                 btrfs_put_block_group(cache);
4309                         cache = btrfs_lookup_block_group(fs_info, start);
4310                         BUG_ON(!cache);
4311                 }
4312
4313                 len = cache->key.objectid + cache->key.offset - start;
4314                 len = min(len, end + 1 - start);
4315
4316                 if (start < cache->last_byte_to_unpin) {
4317                         len = min(len, cache->last_byte_to_unpin - start);
4318                         btrfs_add_free_space(cache, start, len);
4319                 }
4320
4321                 start += len;
4322
4323                 spin_lock(&cache->space_info->lock);
4324                 spin_lock(&cache->lock);
4325                 cache->pinned -= len;
4326                 cache->space_info->bytes_pinned -= len;
4327                 if (cache->ro) {
4328                         cache->space_info->bytes_readonly += len;
4329                 } else if (cache->reserved_pinned > 0) {
4330                         len = min(len, cache->reserved_pinned);
4331                         cache->reserved_pinned -= len;
4332                         cache->space_info->bytes_reserved += len;
4333                 }
4334                 spin_unlock(&cache->lock);
4335                 spin_unlock(&cache->space_info->lock);
4336         }
4337
4338         if (cache)
4339                 btrfs_put_block_group(cache);
4340         return 0;
4341 }
4342
4343 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
4344                                struct btrfs_root *root)
4345 {
4346         struct btrfs_fs_info *fs_info = root->fs_info;
4347         struct extent_io_tree *unpin;
4348         struct btrfs_block_rsv *block_rsv;
4349         struct btrfs_block_rsv *next_rsv;
4350         u64 start;
4351         u64 end;
4352         int idx;
4353         int ret;
4354
4355         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4356                 unpin = &fs_info->freed_extents[1];
4357         else
4358                 unpin = &fs_info->freed_extents[0];
4359
4360         while (1) {
4361                 ret = find_first_extent_bit(unpin, 0, &start, &end,
4362                                             EXTENT_DIRTY);
4363                 if (ret)
4364                         break;
4365
4366                 ret = btrfs_discard_extent(root, start, end + 1 - start);
4367
4368                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4369                 unpin_extent_range(root, start, end);
4370                 cond_resched();
4371         }
4372
4373         mutex_lock(&fs_info->durable_block_rsv_mutex);
4374         list_for_each_entry_safe(block_rsv, next_rsv,
4375                                  &fs_info->durable_block_rsv_list, list) {
4376
4377                 idx = trans->transid & 0x1;
4378                 if (block_rsv->freed[idx] > 0) {
4379                         block_rsv_add_bytes(block_rsv,
4380                                             block_rsv->freed[idx], 0);
4381                         block_rsv->freed[idx] = 0;
4382                 }
4383                 if (atomic_read(&block_rsv->usage) == 0) {
4384                         btrfs_block_rsv_release(root, block_rsv, (u64)-1);
4385
4386                         if (block_rsv->freed[0] == 0 &&
4387                             block_rsv->freed[1] == 0) {
4388                                 list_del_init(&block_rsv->list);
4389                                 kfree(block_rsv);
4390                         }
4391                 } else {
4392                         btrfs_block_rsv_release(root, block_rsv, 0);
4393                 }
4394         }
4395         mutex_unlock(&fs_info->durable_block_rsv_mutex);
4396
4397         return 0;
4398 }
4399
4400 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
4401                                 struct btrfs_root *root,
4402                                 u64 bytenr, u64 num_bytes, u64 parent,
4403                                 u64 root_objectid, u64 owner_objectid,
4404                                 u64 owner_offset, int refs_to_drop,
4405                                 struct btrfs_delayed_extent_op *extent_op)
4406 {
4407         struct btrfs_key key;
4408         struct btrfs_path *path;
4409         struct btrfs_fs_info *info = root->fs_info;
4410         struct btrfs_root *extent_root = info->extent_root;
4411         struct extent_buffer *leaf;
4412         struct btrfs_extent_item *ei;
4413         struct btrfs_extent_inline_ref *iref;
4414         int ret;
4415         int is_data;
4416         int extent_slot = 0;
4417         int found_extent = 0;
4418         int num_to_del = 1;
4419         u32 item_size;
4420         u64 refs;
4421
4422         path = btrfs_alloc_path();
4423         if (!path)
4424                 return -ENOMEM;
4425
4426         path->reada = 1;
4427         path->leave_spinning = 1;
4428
4429         is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
4430         BUG_ON(!is_data && refs_to_drop != 1);
4431
4432         ret = lookup_extent_backref(trans, extent_root, path, &iref,
4433                                     bytenr, num_bytes, parent,
4434                                     root_objectid, owner_objectid,
4435                                     owner_offset);
4436         if (ret == 0) {
4437                 extent_slot = path->slots[0];
4438                 while (extent_slot >= 0) {
4439                         btrfs_item_key_to_cpu(path->nodes[0], &key,
4440                                               extent_slot);
4441                         if (key.objectid != bytenr)
4442                                 break;
4443                         if (key.type == BTRFS_EXTENT_ITEM_KEY &&
4444                             key.offset == num_bytes) {
4445                                 found_extent = 1;
4446                                 break;
4447                         }
4448                         if (path->slots[0] - extent_slot > 5)
4449                                 break;
4450                         extent_slot--;
4451                 }
4452 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4453                 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
4454                 if (found_extent && item_size < sizeof(*ei))
4455                         found_extent = 0;
4456 #endif
4457                 if (!found_extent) {
4458                         BUG_ON(iref);
4459                         ret = remove_extent_backref(trans, extent_root, path,
4460                                                     NULL, refs_to_drop,
4461                                                     is_data);
4462                         BUG_ON(ret);
4463                         btrfs_release_path(extent_root, path);
4464                         path->leave_spinning = 1;
4465
4466                         key.objectid = bytenr;
4467                         key.type = BTRFS_EXTENT_ITEM_KEY;
4468                         key.offset = num_bytes;
4469
4470                         ret = btrfs_search_slot(trans, extent_root,
4471                                                 &key, path, -1, 1);
4472                         if (ret) {
4473                                 printk(KERN_ERR "umm, got %d back from search"
4474                                        ", was looking for %llu\n", ret,
4475                                        (unsigned long long)bytenr);
4476                                 btrfs_print_leaf(extent_root, path->nodes[0]);
4477                         }
4478                         BUG_ON(ret);
4479                         extent_slot = path->slots[0];
4480                 }
4481         } else {
4482                 btrfs_print_leaf(extent_root, path->nodes[0]);
4483                 WARN_ON(1);
4484                 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
4485                        "parent %llu root %llu  owner %llu offset %llu\n",
4486                        (unsigned long long)bytenr,
4487                        (unsigned long long)parent,
4488                        (unsigned long long)root_objectid,
4489                        (unsigned long long)owner_objectid,
4490                        (unsigned long long)owner_offset);
4491         }
4492
4493         leaf = path->nodes[0];
4494         item_size = btrfs_item_size_nr(leaf, extent_slot);
4495 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4496         if (item_size < sizeof(*ei)) {
4497                 BUG_ON(found_extent || extent_slot != path->slots[0]);
4498                 ret = convert_extent_item_v0(trans, extent_root, path,
4499                                              owner_objectid, 0);
4500                 BUG_ON(ret < 0);
4501
4502                 btrfs_release_path(extent_root, path);
4503                 path->leave_spinning = 1;
4504
4505                 key.objectid = bytenr;
4506                 key.type = BTRFS_EXTENT_ITEM_KEY;
4507                 key.offset = num_bytes;
4508
4509                 ret = btrfs_search_slot(trans, extent_root, &key, path,
4510                                         -1, 1);
4511                 if (ret) {
4512                         printk(KERN_ERR "umm, got %d back from search"
4513                                ", was looking for %llu\n", ret,
4514                                (unsigned long long)bytenr);
4515                         btrfs_print_leaf(extent_root, path->nodes[0]);
4516                 }
4517                 BUG_ON(ret);
4518                 extent_slot = path->slots[0];
4519                 leaf = path->nodes[0];
4520                 item_size = btrfs_item_size_nr(leaf, extent_slot);
4521         }
4522 #endif
4523         BUG_ON(item_size < sizeof(*ei));
4524         ei = btrfs_item_ptr(leaf, extent_slot,
4525                             struct btrfs_extent_item);
4526         if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4527                 struct btrfs_tree_block_info *bi;
4528                 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
4529                 bi = (struct btrfs_tree_block_info *)(ei + 1);
4530                 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
4531         }
4532
4533         refs = btrfs_extent_refs(leaf, ei);
4534         BUG_ON(refs < refs_to_drop);
4535         refs -= refs_to_drop;
4536
4537         if (refs > 0) {
4538                 if (extent_op)
4539                         __run_delayed_extent_op(extent_op, leaf, ei);
4540                 /*
4541                  * In the case of inline back ref, reference count will
4542                  * be updated by remove_extent_backref
4543                  */
4544                 if (iref) {
4545                         BUG_ON(!found_extent);
4546                 } else {
4547                         btrfs_set_extent_refs(leaf, ei, refs);
4548                         btrfs_mark_buffer_dirty(leaf);
4549                 }
4550                 if (found_extent) {
4551                         ret = remove_extent_backref(trans, extent_root, path,
4552                                                     iref, refs_to_drop,
4553                                                     is_data);
4554                         BUG_ON(ret);
4555                 }
4556         } else {
4557                 if (found_extent) {
4558                         BUG_ON(is_data && refs_to_drop !=
4559                                extent_data_ref_count(root, path, iref));
4560                         if (iref) {
4561                                 BUG_ON(path->slots[0] != extent_slot);
4562                         } else {
4563                                 BUG_ON(path->slots[0] != extent_slot + 1);
4564                                 path->slots[0] = extent_slot;
4565                                 num_to_del = 2;
4566                         }
4567                 }
4568
4569                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
4570                                       num_to_del);
4571                 BUG_ON(ret);
4572                 btrfs_release_path(extent_root, path);
4573
4574                 if (is_data) {
4575                         ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
4576                         BUG_ON(ret);
4577                 } else {
4578                         invalidate_mapping_pages(info->btree_inode->i_mapping,
4579                              bytenr >> PAGE_CACHE_SHIFT,
4580                              (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
4581                 }
4582
4583                 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
4584                 BUG_ON(ret);
4585         }
4586         btrfs_free_path(path);
4587         return ret;
4588 }
4589
4590 /*
4591  * when we free an block, it is possible (and likely) that we free the last
4592  * delayed ref for that extent as well.  This searches the delayed ref tree for
4593  * a given extent, and if there are no other delayed refs to be processed, it
4594  * removes it from the tree.
4595  */
4596 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
4597                                       struct btrfs_root *root, u64 bytenr)
4598 {
4599         struct btrfs_delayed_ref_head *head;
4600         struct btrfs_delayed_ref_root *delayed_refs;
4601         struct btrfs_delayed_ref_node *ref;
4602         struct rb_node *node;
4603         int ret = 0;
4604
4605         delayed_refs = &trans->transaction->delayed_refs;
4606         spin_lock(&delayed_refs->lock);
4607         head = btrfs_find_delayed_ref_head(trans, bytenr);
4608         if (!head)
4609                 goto out;
4610
4611         node = rb_prev(&head->node.rb_node);
4612         if (!node)
4613                 goto out;
4614
4615         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
4616
4617         /* there are still entries for this ref, we can't drop it */
4618         if (ref->bytenr == bytenr)
4619                 goto out;
4620
4621         if (head->extent_op) {
4622                 if (!head->must_insert_reserved)
4623                         goto out;
4624                 kfree(head->extent_op);
4625                 head->extent_op = NULL;
4626         }
4627
4628         /*
4629          * waiting for the lock here would deadlock.  If someone else has it
4630          * locked they are already in the process of dropping it anyway
4631          */
4632         if (!mutex_trylock(&head->mutex))
4633                 goto out;
4634
4635         /*
4636          * at this point we have a head with no other entries.  Go
4637          * ahead and process it.
4638          */
4639         head->node.in_tree = 0;
4640         rb_erase(&head->node.rb_node, &delayed_refs->root);
4641
4642         delayed_refs->num_entries--;
4643
4644         /*
4645          * we don't take a ref on the node because we're removing it from the
4646          * tree, so we just steal the ref the tree was holding.
4647          */
4648         delayed_refs->num_heads--;
4649         if (list_empty(&head->cluster))
4650                 delayed_refs->num_heads_ready--;
4651
4652         list_del_init(&head->cluster);
4653         spin_unlock(&delayed_refs->lock);
4654
4655         BUG_ON(head->extent_op);
4656         if (head->must_insert_reserved)
4657                 ret = 1;
4658
4659         mutex_unlock(&head->mutex);
4660         btrfs_put_delayed_ref(&head->node);
4661         return ret;
4662 out:
4663         spin_unlock(&delayed_refs->lock);
4664         return 0;
4665 }
4666
4667 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
4668                            struct btrfs_root *root,
4669                            struct extent_buffer *buf,
4670                            u64 parent, int last_ref)
4671 {
4672         struct btrfs_block_rsv *block_rsv;
4673         struct btrfs_block_group_cache *cache = NULL;
4674         int ret;
4675
4676         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4677                 ret = btrfs_add_delayed_tree_ref(trans, buf->start, buf->len,
4678                                                 parent, root->root_key.objectid,
4679                                                 btrfs_header_level(buf),
4680                                                 BTRFS_DROP_DELAYED_REF, NULL);
4681                 BUG_ON(ret);
4682         }
4683
4684         if (!last_ref)
4685                 return;
4686
4687         block_rsv = get_block_rsv(trans, root);
4688         cache = btrfs_lookup_block_group(root->fs_info, buf->start);
4689         if (block_rsv->space_info != cache->space_info)
4690                 goto out;
4691
4692         if (btrfs_header_generation(buf) == trans->transid) {
4693                 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4694                         ret = check_ref_cleanup(trans, root, buf->start);
4695                         if (!ret)
4696                                 goto pin;
4697                 }
4698
4699                 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
4700                         pin_down_extent(root, cache, buf->start, buf->len, 1);
4701                         goto pin;
4702                 }
4703
4704                 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
4705
4706                 btrfs_add_free_space(cache, buf->start, buf->len);
4707                 ret = update_reserved_bytes(cache, buf->len, 0, 0);
4708                 if (ret == -EAGAIN) {
4709                         /* block group became read-only */
4710                         update_reserved_bytes(cache, buf->len, 0, 1);
4711                         goto out;
4712                 }
4713
4714                 ret = 1;
4715                 spin_lock(&block_rsv->lock);
4716                 if (block_rsv->reserved < block_rsv->size) {
4717                         block_rsv->reserved += buf->len;
4718                         ret = 0;
4719                 }
4720                 spin_unlock(&block_rsv->lock);
4721
4722                 if (ret) {
4723                         spin_lock(&cache->space_info->lock);
4724                         cache->space_info->bytes_reserved -= buf->len;
4725                         cache->space_info->reservation_progress++;
4726                         spin_unlock(&cache->space_info->lock);
4727                 }
4728                 goto out;
4729         }
4730 pin:
4731         if (block_rsv->durable && !cache->ro) {
4732                 ret = 0;
4733                 spin_lock(&cache->lock);
4734                 if (!cache->ro) {
4735                         cache->reserved_pinned += buf->len;
4736                         ret = 1;
4737                 }
4738                 spin_unlock(&cache->lock);
4739
4740                 if (ret) {
4741                         spin_lock(&block_rsv->lock);
4742                         block_rsv->freed[trans->transid & 0x1] += buf->len;
4743                         spin_unlock(&block_rsv->lock);
4744                 }
4745         }
4746 out:
4747         btrfs_put_block_group(cache);
4748 }
4749
4750 int btrfs_free_extent(struct btrfs_trans_handle *trans,
4751                       struct btrfs_root *root,
4752                       u64 bytenr, u64 num_bytes, u64 parent,
4753                       u64 root_objectid, u64 owner, u64 offset)
4754 {
4755         int ret;
4756
4757         /*
4758          * tree log blocks never actually go into the extent allocation
4759          * tree, just update pinning info and exit early.
4760          */
4761         if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
4762                 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
4763                 /* unlocks the pinned mutex */
4764                 btrfs_pin_extent(root, bytenr, num_bytes, 1);
4765                 ret = 0;
4766         } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
4767                 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
4768                                         parent, root_objectid, (int)owner,
4769                                         BTRFS_DROP_DELAYED_REF, NULL);
4770                 BUG_ON(ret);
4771         } else {
4772                 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
4773                                         parent, root_objectid, owner,
4774                                         offset, BTRFS_DROP_DELAYED_REF, NULL);
4775                 BUG_ON(ret);
4776         }
4777         return ret;
4778 }
4779
4780 static u64 stripe_align(struct btrfs_root *root, u64 val)
4781 {
4782         u64 mask = ((u64)root->stripesize - 1);
4783         u64 ret = (val + mask) & ~mask;
4784         return ret;
4785 }
4786
4787 /*
4788  * when we wait for progress in the block group caching, its because
4789  * our allocation attempt failed at least once.  So, we must sleep
4790  * and let some progress happen before we try again.
4791  *
4792  * This function will sleep at least once waiting for new free space to
4793  * show up, and then it will check the block group free space numbers
4794  * for our min num_bytes.  Another option is to have it go ahead
4795  * and look in the rbtree for a free extent of a given size, but this
4796  * is a good start.
4797  */
4798 static noinline int
4799 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
4800                                 u64 num_bytes)
4801 {
4802         struct btrfs_caching_control *caching_ctl;
4803         DEFINE_WAIT(wait);
4804
4805         caching_ctl = get_caching_control(cache);
4806         if (!caching_ctl)
4807                 return 0;
4808
4809         wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
4810                    (cache->free_space >= num_bytes));
4811
4812         put_caching_control(caching_ctl);
4813         return 0;
4814 }
4815
4816 static noinline int
4817 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
4818 {
4819         struct btrfs_caching_control *caching_ctl;
4820         DEFINE_WAIT(wait);
4821
4822         caching_ctl = get_caching_control(cache);
4823         if (!caching_ctl)
4824                 return 0;
4825
4826         wait_event(caching_ctl->wait, block_group_cache_done(cache));
4827
4828         put_caching_control(caching_ctl);
4829         return 0;
4830 }
4831
4832 static int get_block_group_index(struct btrfs_block_group_cache *cache)
4833 {
4834         int index;
4835         if (cache->flags & BTRFS_BLOCK_GROUP_RAID10)
4836                 index = 0;
4837         else if (cache->flags & BTRFS_BLOCK_GROUP_RAID1)
4838                 index = 1;
4839         else if (cache->flags & BTRFS_BLOCK_GROUP_DUP)
4840                 index = 2;
4841         else if (cache->flags & BTRFS_BLOCK_GROUP_RAID0)
4842                 index = 3;
4843         else
4844                 index = 4;
4845         return index;
4846 }
4847
4848 enum btrfs_loop_type {
4849         LOOP_FIND_IDEAL = 0,
4850         LOOP_CACHING_NOWAIT = 1,
4851         LOOP_CACHING_WAIT = 2,
4852         LOOP_ALLOC_CHUNK = 3,
4853         LOOP_NO_EMPTY_SIZE = 4,
4854 };
4855
4856 /*
4857  * walks the btree of allocated extents and find a hole of a given size.
4858  * The key ins is changed to record the hole:
4859  * ins->objectid == block start
4860  * ins->flags = BTRFS_EXTENT_ITEM_KEY
4861  * ins->offset == number of blocks
4862  * Any available blocks before search_start are skipped.
4863  */
4864 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
4865                                      struct btrfs_root *orig_root,
4866                                      u64 num_bytes, u64 empty_size,
4867                                      u64 search_start, u64 search_end,
4868                                      u64 hint_byte, struct btrfs_key *ins,
4869                                      int data)
4870 {
4871         int ret = 0;
4872         struct btrfs_root *root = orig_root->fs_info->extent_root;
4873         struct btrfs_free_cluster *last_ptr = NULL;
4874         struct btrfs_block_group_cache *block_group = NULL;
4875         int empty_cluster = 2 * 1024 * 1024;
4876         int allowed_chunk_alloc = 0;
4877         int done_chunk_alloc = 0;
4878         struct btrfs_space_info *space_info;
4879         int last_ptr_loop = 0;
4880         int loop = 0;
4881         int index = 0;
4882         bool found_uncached_bg = false;
4883         bool failed_cluster_refill = false;
4884         bool failed_alloc = false;
4885         bool use_cluster = true;
4886         u64 ideal_cache_percent = 0;
4887         u64 ideal_cache_offset = 0;
4888
4889         WARN_ON(num_bytes < root->sectorsize);
4890         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
4891         ins->objectid = 0;
4892         ins->offset = 0;
4893
4894         space_info = __find_space_info(root->fs_info, data);
4895         if (!space_info) {
4896                 printk(KERN_ERR "No space info for %d\n", data);
4897                 return -ENOSPC;
4898         }
4899
4900         /*
4901          * If the space info is for both data and metadata it means we have a
4902          * small filesystem and we can't use the clustering stuff.
4903          */
4904         if (btrfs_mixed_space_info(space_info))
4905                 use_cluster = false;
4906
4907         if (orig_root->ref_cows || empty_size)
4908                 allowed_chunk_alloc = 1;
4909
4910         if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
4911                 last_ptr = &root->fs_info->meta_alloc_cluster;
4912                 if (!btrfs_test_opt(root, SSD))
4913                         empty_cluster = 64 * 1024;
4914         }
4915
4916         if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
4917             btrfs_test_opt(root, SSD)) {
4918                 last_ptr = &root->fs_info->data_alloc_cluster;
4919         }
4920
4921         if (last_ptr) {
4922                 spin_lock(&last_ptr->lock);
4923                 if (last_ptr->block_group)
4924                         hint_byte = last_ptr->window_start;
4925                 spin_unlock(&last_ptr->lock);
4926         }
4927
4928         search_start = max(search_start, first_logical_byte(root, 0));
4929         search_start = max(search_start, hint_byte);
4930
4931         if (!last_ptr)
4932                 empty_cluster = 0;
4933
4934         if (search_start == hint_byte) {
4935 ideal_cache:
4936                 block_group = btrfs_lookup_block_group(root->fs_info,
4937                                                        search_start);
4938                 /*
4939                  * we don't want to use the block group if it doesn't match our
4940                  * allocation bits, or if its not cached.
4941                  *
4942                  * However if we are re-searching with an ideal block group
4943                  * picked out then we don't care that the block group is cached.
4944                  */
4945                 if (block_group && block_group_bits(block_group, data) &&
4946                     (block_group->cached != BTRFS_CACHE_NO ||
4947                      search_start == ideal_cache_offset)) {
4948                         down_read(&space_info->groups_sem);
4949                         if (list_empty(&block_group->list) ||
4950                             block_group->ro) {
4951                                 /*
4952                                  * someone is removing this block group,
4953                                  * we can't jump into the have_block_group
4954                                  * target because our list pointers are not
4955                                  * valid
4956                                  */
4957                                 btrfs_put_block_group(block_group);
4958                                 up_read(&space_info->groups_sem);
4959                         } else {
4960                                 index = get_block_group_index(block_group);
4961                                 goto have_block_group;
4962                         }
4963                 } else if (block_group) {
4964                         btrfs_put_block_group(block_group);
4965                 }
4966         }
4967 search:
4968         down_read(&space_info->groups_sem);
4969         list_for_each_entry(block_group, &space_info->block_groups[index],
4970                             list) {
4971                 u64 offset;
4972                 int cached;
4973
4974                 btrfs_get_block_group(block_group);
4975                 search_start = block_group->key.objectid;
4976
4977                 /*
4978                  * this can happen if we end up cycling through all the
4979                  * raid types, but we want to make sure we only allocate
4980                  * for the proper type.
4981                  */
4982                 if (!block_group_bits(block_group, data)) {
4983                     u64 extra = BTRFS_BLOCK_GROUP_DUP |
4984                                 BTRFS_BLOCK_GROUP_RAID1 |
4985                                 BTRFS_BLOCK_GROUP_RAID10;
4986
4987                         /*
4988                          * if they asked for extra copies and this block group
4989                          * doesn't provide them, bail.  This does allow us to
4990                          * fill raid0 from raid1.
4991                          */
4992                         if ((data & extra) && !(block_group->flags & extra))
4993                                 goto loop;
4994                 }
4995
4996 have_block_group:
4997                 if (unlikely(block_group->cached == BTRFS_CACHE_NO)) {
4998                         u64 free_percent;
4999
5000                         ret = cache_block_group(block_group, trans,
5001                                                 orig_root, 1);
5002                         if (block_group->cached == BTRFS_CACHE_FINISHED)
5003                                 goto have_block_group;
5004
5005                         free_percent = btrfs_block_group_used(&block_group->item);
5006                         free_percent *= 100;
5007                         free_percent = div64_u64(free_percent,
5008                                                  block_group->key.offset);
5009                         free_percent = 100 - free_percent;
5010                         if (free_percent > ideal_cache_percent &&
5011                             likely(!block_group->ro)) {
5012                                 ideal_cache_offset = block_group->key.objectid;
5013                                 ideal_cache_percent = free_percent;
5014                         }
5015
5016                         /*
5017                          * We only want to start kthread caching if we are at
5018                          * the point where we will wait for caching to make
5019                          * progress, or if our ideal search is over and we've
5020                          * found somebody to start caching.
5021                          */
5022                         if (loop > LOOP_CACHING_NOWAIT ||
5023                             (loop > LOOP_FIND_IDEAL &&
5024                              atomic_read(&space_info->caching_threads) < 2)) {
5025                                 ret = cache_block_group(block_group, trans,
5026                                                         orig_root, 0);
5027                                 BUG_ON(ret);
5028                         }
5029                         found_uncached_bg = true;
5030
5031                         /*
5032                          * If loop is set for cached only, try the next block
5033                          * group.
5034                          */
5035                         if (loop == LOOP_FIND_IDEAL)
5036                                 goto loop;
5037                 }
5038
5039                 cached = block_group_cache_done(block_group);
5040                 if (unlikely(!cached))
5041                         found_uncached_bg = true;
5042
5043                 if (unlikely(block_group->ro))
5044                         goto loop;
5045
5046                 /*
5047                  * Ok we want to try and use the cluster allocator, so lets look
5048                  * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will
5049                  * have tried the cluster allocator plenty of times at this
5050                  * point and not have found anything, so we are likely way too
5051                  * fragmented for the clustering stuff to find anything, so lets
5052                  * just skip it and let the allocator find whatever block it can
5053                  * find
5054                  */
5055                 if (last_ptr && loop < LOOP_NO_EMPTY_SIZE) {
5056                         /*
5057                          * the refill lock keeps out other
5058                          * people trying to start a new cluster
5059                          */
5060                         spin_lock(&last_ptr->refill_lock);
5061                         if (last_ptr->block_group &&
5062                             (last_ptr->block_group->ro ||
5063                             !block_group_bits(last_ptr->block_group, data))) {
5064                                 offset = 0;
5065                                 goto refill_cluster;
5066                         }
5067
5068                         offset = btrfs_alloc_from_cluster(block_group, last_ptr,
5069                                                  num_bytes, search_start);
5070                         if (offset) {
5071                                 /* we have a block, we're done */
5072                                 spin_unlock(&last_ptr->refill_lock);
5073                                 goto checks;
5074                         }
5075
5076                         spin_lock(&last_ptr->lock);
5077                         /*
5078                          * whoops, this cluster doesn't actually point to
5079                          * this block group.  Get a ref on the block
5080                          * group is does point to and try again
5081                          */
5082                         if (!last_ptr_loop && last_ptr->block_group &&
5083                             last_ptr->block_group != block_group) {
5084
5085                                 btrfs_put_block_group(block_group);
5086                                 block_group = last_ptr->block_group;
5087                                 btrfs_get_block_group(block_group);
5088                                 spin_unlock(&last_ptr->lock);
5089                                 spin_unlock(&last_ptr->refill_lock);
5090
5091                                 last_ptr_loop = 1;
5092                                 search_start = block_group->key.objectid;
5093                                 /*
5094                                  * we know this block group is properly
5095                                  * in the list because
5096                                  * btrfs_remove_block_group, drops the
5097                                  * cluster before it removes the block
5098                                  * group from the list
5099                                  */
5100                                 goto have_block_group;
5101                         }
5102                         spin_unlock(&last_ptr->lock);
5103 refill_cluster:
5104                         /*
5105                          * this cluster didn't work out, free it and
5106                          * start over
5107                          */
5108                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
5109
5110                         last_ptr_loop = 0;
5111
5112                         /* allocate a cluster in this block group */
5113                         ret = btrfs_find_space_cluster(trans, root,
5114                                                block_group, last_ptr,
5115                                                offset, num_bytes,
5116                                                empty_cluster + empty_size);
5117                         if (ret == 0) {
5118                                 /*
5119                                  * now pull our allocation out of this
5120                                  * cluster
5121                                  */
5122                                 offset = btrfs_alloc_from_cluster(block_group,
5123                                                   last_ptr, num_bytes,
5124                                                   search_start);
5125                                 if (offset) {
5126                                         /* we found one, proceed */
5127                                         spin_unlock(&last_ptr->refill_lock);
5128                                         goto checks;
5129                                 }
5130                         } else if (!cached && loop > LOOP_CACHING_NOWAIT
5131                                    && !failed_cluster_refill) {
5132                                 spin_unlock(&last_ptr->refill_lock);
5133
5134                                 failed_cluster_refill = true;
5135                                 wait_block_group_cache_progress(block_group,
5136                                        num_bytes + empty_cluster + empty_size);
5137                                 goto have_block_group;
5138                         }
5139
5140                         /*
5141                          * at this point we either didn't find a cluster
5142                          * or we weren't able to allocate a block from our
5143                          * cluster.  Free the cluster we've been trying
5144                          * to use, and go to the next block group
5145                          */
5146                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
5147                         spin_unlock(&last_ptr->refill_lock);
5148                         goto loop;
5149                 }
5150
5151                 offset = btrfs_find_space_for_alloc(block_group, search_start,
5152                                                     num_bytes, empty_size);
5153                 /*
5154                  * If we didn't find a chunk, and we haven't failed on this
5155                  * block group before, and this block group is in the middle of
5156                  * caching and we are ok with waiting, then go ahead and wait
5157                  * for progress to be made, and set failed_alloc to true.
5158                  *
5159                  * If failed_alloc is true then we've already waited on this
5160                  * block group once and should move on to the next block group.
5161                  */
5162                 if (!offset && !failed_alloc && !cached &&
5163                     loop > LOOP_CACHING_NOWAIT) {
5164                         wait_block_group_cache_progress(block_group,
5165                                                 num_bytes + empty_size);
5166                         failed_alloc = true;
5167                         goto have_block_group;
5168                 } else if (!offset) {
5169                         goto loop;
5170                 }
5171 checks:
5172                 search_start = stripe_align(root, offset);
5173                 /* move on to the next group */
5174                 if (search_start + num_bytes >= search_end) {
5175                         btrfs_add_free_space(block_group, offset, num_bytes);
5176                         goto loop;
5177                 }
5178
5179                 /* move on to the next group */
5180                 if (search_start + num_bytes >
5181                     block_group->key.objectid + block_group->key.offset) {
5182                         btrfs_add_free_space(block_group, offset, num_bytes);
5183                         goto loop;
5184                 }
5185
5186                 ins->objectid = search_start;
5187                 ins->offset = num_bytes;
5188
5189                 if (offset < search_start)
5190                         btrfs_add_free_space(block_group, offset,
5191                                              search_start - offset);
5192                 BUG_ON(offset > search_start);
5193
5194                 ret = update_reserved_bytes(block_group, num_bytes, 1,
5195                                             (data & BTRFS_BLOCK_GROUP_DATA));
5196                 if (ret == -EAGAIN) {
5197                         btrfs_add_free_space(block_group, offset, num_bytes);
5198                         goto loop;
5199                 }
5200
5201                 /* we are all good, lets return */
5202                 ins->objectid = search_start;
5203                 ins->offset = num_bytes;
5204
5205                 if (offset < search_start)
5206                         btrfs_add_free_space(block_group, offset,
5207                                              search_start - offset);
5208                 BUG_ON(offset > search_start);
5209                 break;
5210 loop:
5211                 failed_cluster_refill = false;
5212                 failed_alloc = false;
5213                 BUG_ON(index != get_block_group_index(block_group));
5214                 btrfs_put_block_group(block_group);
5215         }
5216         up_read(&space_info->groups_sem);
5217
5218         if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
5219                 goto search;
5220
5221         /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
5222          *                      for them to make caching progress.  Also
5223          *                      determine the best possible bg to cache
5224          * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5225          *                      caching kthreads as we move along
5226          * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5227          * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5228          * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5229          *                      again
5230          */
5231         if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE &&
5232             (found_uncached_bg || empty_size || empty_cluster ||
5233              allowed_chunk_alloc)) {
5234                 index = 0;
5235                 if (loop == LOOP_FIND_IDEAL && found_uncached_bg) {
5236                         found_uncached_bg = false;
5237                         loop++;
5238                         if (!ideal_cache_percent &&
5239                             atomic_read(&space_info->caching_threads))
5240                                 goto search;
5241
5242                         /*
5243                          * 1 of the following 2 things have happened so far
5244                          *
5245                          * 1) We found an ideal block group for caching that
5246                          * is mostly full and will cache quickly, so we might
5247                          * as well wait for it.
5248                          *
5249                          * 2) We searched for cached only and we didn't find
5250                          * anything, and we didn't start any caching kthreads
5251                          * either, so chances are we will loop through and
5252                          * start a couple caching kthreads, and then come back
5253                          * around and just wait for them.  This will be slower
5254                          * because we will have 2 caching kthreads reading at
5255                          * the same time when we could have just started one
5256                          * and waited for it to get far enough to give us an
5257                          * allocation, so go ahead and go to the wait caching
5258                          * loop.
5259                          */
5260                         loop = LOOP_CACHING_WAIT;
5261                         search_start = ideal_cache_offset;
5262                         ideal_cache_percent = 0;
5263                         goto ideal_cache;
5264                 } else if (loop == LOOP_FIND_IDEAL) {
5265                         /*
5266                          * Didn't find a uncached bg, wait on anything we find
5267                          * next.
5268                          */
5269                         loop = LOOP_CACHING_WAIT;
5270                         goto search;
5271                 }
5272
5273                 if (loop < LOOP_CACHING_WAIT) {
5274                         loop++;
5275                         goto search;
5276                 }
5277
5278                 if (loop == LOOP_ALLOC_CHUNK) {
5279                         empty_size = 0;
5280                         empty_cluster = 0;
5281                 }
5282
5283                 if (allowed_chunk_alloc) {
5284                         ret = do_chunk_alloc(trans, root, num_bytes +
5285                                              2 * 1024 * 1024, data, 1);
5286                         allowed_chunk_alloc = 0;
5287                         done_chunk_alloc = 1;
5288                 } else if (!done_chunk_alloc) {
5289                         space_info->force_alloc = 1;
5290                 }
5291
5292                 if (loop < LOOP_NO_EMPTY_SIZE) {
5293                         loop++;
5294                         goto search;
5295                 }
5296                 ret = -ENOSPC;
5297         } else if (!ins->objectid) {
5298                 ret = -ENOSPC;
5299         }
5300
5301         /* we found what we needed */
5302         if (ins->objectid) {
5303                 if (!(data & BTRFS_BLOCK_GROUP_DATA))
5304                         trans->block_group = block_group->key.objectid;
5305
5306                 btrfs_put_block_group(block_group);
5307                 ret = 0;
5308         }
5309
5310         return ret;
5311 }
5312
5313 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
5314                             int dump_block_groups)
5315 {
5316         struct btrfs_block_group_cache *cache;
5317         int index = 0;
5318
5319         spin_lock(&info->lock);
5320         printk(KERN_INFO "space_info has %llu free, is %sfull\n",
5321                (unsigned long long)(info->total_bytes - info->bytes_used -
5322                                     info->bytes_pinned - info->bytes_reserved -
5323                                     info->bytes_readonly),
5324                (info->full) ? "" : "not ");
5325         printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
5326                "reserved=%llu, may_use=%llu, readonly=%llu\n",
5327                (unsigned long long)info->total_bytes,
5328                (unsigned long long)info->bytes_used,
5329                (unsigned long long)info->bytes_pinned,
5330                (unsigned long long)info->bytes_reserved,
5331                (unsigned long long)info->bytes_may_use,
5332                (unsigned long long)info->bytes_readonly);
5333         spin_unlock(&info->lock);
5334
5335         if (!dump_block_groups)
5336                 return;
5337
5338         down_read(&info->groups_sem);
5339 again:
5340         list_for_each_entry(cache, &info->block_groups[index], list) {
5341                 spin_lock(&cache->lock);
5342                 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
5343                        "%llu pinned %llu reserved\n",
5344                        (unsigned long long)cache->key.objectid,
5345                        (unsigned long long)cache->key.offset,
5346                        (unsigned long long)btrfs_block_group_used(&cache->item),
5347                        (unsigned long long)cache->pinned,
5348                        (unsigned long long)cache->reserved);
5349                 btrfs_dump_free_space(cache, bytes);
5350                 spin_unlock(&cache->lock);
5351         }
5352         if (++index < BTRFS_NR_RAID_TYPES)
5353                 goto again;
5354         up_read(&info->groups_sem);
5355 }
5356
5357 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
5358                          struct btrfs_root *root,
5359                          u64 num_bytes, u64 min_alloc_size,
5360                          u64 empty_size, u64 hint_byte,
5361                          u64 search_end, struct btrfs_key *ins,
5362                          u64 data)
5363 {
5364         int ret;
5365         u64 search_start = 0;
5366
5367         data = btrfs_get_alloc_profile(root, data);
5368 again:
5369         /*
5370          * the only place that sets empty_size is btrfs_realloc_node, which
5371          * is not called recursively on allocations
5372          */
5373         if (empty_size || root->ref_cows)
5374                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
5375                                      num_bytes + 2 * 1024 * 1024, data, 0);
5376
5377         WARN_ON(num_bytes < root->sectorsize);
5378         ret = find_free_extent(trans, root, num_bytes, empty_size,
5379                                search_start, search_end, hint_byte,
5380                                ins, data);
5381
5382         if (ret == -ENOSPC && num_bytes > min_alloc_size) {
5383                 num_bytes = num_bytes >> 1;
5384                 num_bytes = num_bytes & ~(root->sectorsize - 1);
5385                 num_bytes = max(num_bytes, min_alloc_size);
5386                 do_chunk_alloc(trans, root->fs_info->extent_root,
5387                                num_bytes, data, 1);
5388                 goto again;
5389         }
5390         if (ret == -ENOSPC && btrfs_test_opt(root, ENOSPC_DEBUG)) {
5391                 struct btrfs_space_info *sinfo;
5392
5393                 sinfo = __find_space_info(root->fs_info, data);
5394                 printk(KERN_ERR "btrfs allocation failed flags %llu, "
5395                        "wanted %llu\n", (unsigned long long)data,
5396                        (unsigned long long)num_bytes);
5397                 dump_space_info(sinfo, num_bytes, 1);
5398         }
5399
5400         return ret;
5401 }
5402
5403 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
5404 {
5405         struct btrfs_block_group_cache *cache;
5406         int ret = 0;
5407
5408         cache = btrfs_lookup_block_group(root->fs_info, start);
5409         if (!cache) {
5410                 printk(KERN_ERR "Unable to find block group for %llu\n",
5411                        (unsigned long long)start);
5412                 return -ENOSPC;
5413         }
5414
5415         ret = btrfs_discard_extent(root, start, len);
5416
5417         btrfs_add_free_space(cache, start, len);
5418         update_reserved_bytes(cache, len, 0, 1);
5419         btrfs_put_block_group(cache);
5420
5421         return ret;
5422 }
5423
5424 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5425                                       struct btrfs_root *root,
5426                                       u64 parent, u64 root_objectid,
5427                                       u64 flags, u64 owner, u64 offset,
5428                                       struct btrfs_key *ins, int ref_mod)
5429 {
5430         int ret;
5431         struct btrfs_fs_info *fs_info = root->fs_info;
5432         struct btrfs_extent_item *extent_item;
5433         struct btrfs_extent_inline_ref *iref;
5434         struct btrfs_path *path;
5435         struct extent_buffer *leaf;
5436         int type;
5437         u32 size;
5438
5439         if (parent > 0)
5440                 type = BTRFS_SHARED_DATA_REF_KEY;
5441         else
5442                 type = BTRFS_EXTENT_DATA_REF_KEY;
5443
5444         size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
5445
5446         path = btrfs_alloc_path();
5447         BUG_ON(!path);
5448
5449         path->leave_spinning = 1;
5450         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5451                                       ins, size);
5452         BUG_ON(ret);
5453
5454         leaf = path->nodes[0];
5455         extent_item = btrfs_item_ptr(leaf, path->slots[0],
5456                                      struct btrfs_extent_item);
5457         btrfs_set_extent_refs(leaf, extent_item, ref_mod);
5458         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5459         btrfs_set_extent_flags(leaf, extent_item,
5460                                flags | BTRFS_EXTENT_FLAG_DATA);
5461
5462         iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
5463         btrfs_set_extent_inline_ref_type(leaf, iref, type);
5464         if (parent > 0) {
5465                 struct btrfs_shared_data_ref *ref;
5466                 ref = (struct btrfs_shared_data_ref *)(iref + 1);
5467                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5468                 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
5469         } else {
5470                 struct btrfs_extent_data_ref *ref;
5471                 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
5472                 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
5473                 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
5474                 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
5475                 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
5476         }
5477
5478         btrfs_mark_buffer_dirty(path->nodes[0]);
5479         btrfs_free_path(path);
5480
5481         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5482         if (ret) {
5483                 printk(KERN_ERR "btrfs update block group failed for %llu "
5484                        "%llu\n", (unsigned long long)ins->objectid,
5485                        (unsigned long long)ins->offset);
5486                 BUG();
5487         }
5488         return ret;
5489 }
5490
5491 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
5492                                      struct btrfs_root *root,
5493                                      u64 parent, u64 root_objectid,
5494                                      u64 flags, struct btrfs_disk_key *key,
5495                                      int level, struct btrfs_key *ins)
5496 {
5497         int ret;
5498         struct btrfs_fs_info *fs_info = root->fs_info;
5499         struct btrfs_extent_item *extent_item;
5500         struct btrfs_tree_block_info *block_info;
5501         struct btrfs_extent_inline_ref *iref;
5502         struct btrfs_path *path;
5503         struct extent_buffer *leaf;
5504         u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
5505
5506         path = btrfs_alloc_path();
5507         BUG_ON(!path);
5508
5509         path->leave_spinning = 1;
5510         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5511                                       ins, size);
5512         BUG_ON(ret);
5513
5514         leaf = path->nodes[0];
5515         extent_item = btrfs_item_ptr(leaf, path->slots[0],
5516                                      struct btrfs_extent_item);
5517         btrfs_set_extent_refs(leaf, extent_item, 1);
5518         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5519         btrfs_set_extent_flags(leaf, extent_item,
5520                                flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
5521         block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
5522
5523         btrfs_set_tree_block_key(leaf, block_info, key);
5524         btrfs_set_tree_block_level(leaf, block_info, level);
5525
5526         iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
5527         if (parent > 0) {
5528                 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
5529                 btrfs_set_extent_inline_ref_type(leaf, iref,
5530                                                  BTRFS_SHARED_BLOCK_REF_KEY);
5531                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5532         } else {
5533                 btrfs_set_extent_inline_ref_type(leaf, iref,
5534                                                  BTRFS_TREE_BLOCK_REF_KEY);
5535                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
5536         }
5537
5538         btrfs_mark_buffer_dirty(leaf);
5539         btrfs_free_path(path);
5540
5541         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5542         if (ret) {
5543                 printk(KERN_ERR "btrfs update block group failed for %llu "
5544                        "%llu\n", (unsigned long long)ins->objectid,
5545                        (unsigned long long)ins->offset);
5546                 BUG();
5547         }
5548         return ret;
5549 }
5550
5551 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5552                                      struct btrfs_root *root,
5553                                      u64 root_objectid, u64 owner,
5554                                      u64 offset, struct btrfs_key *ins)
5555 {
5556         int ret;
5557
5558         BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
5559
5560         ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset,
5561                                          0, root_objectid, owner, offset,
5562                                          BTRFS_ADD_DELAYED_EXTENT, NULL);
5563         return ret;
5564 }
5565
5566 /*
5567  * this is used by the tree logging recovery code.  It records that
5568  * an extent has been allocated and makes sure to clear the free
5569  * space cache bits as well
5570  */
5571 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
5572                                    struct btrfs_root *root,
5573                                    u64 root_objectid, u64 owner, u64 offset,
5574                                    struct btrfs_key *ins)
5575 {
5576         int ret;
5577         struct btrfs_block_group_cache *block_group;
5578         struct btrfs_caching_control *caching_ctl;
5579         u64 start = ins->objectid;
5580         u64 num_bytes = ins->offset;
5581
5582         block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
5583         cache_block_group(block_group, trans, NULL, 0);
5584         caching_ctl = get_caching_control(block_group);
5585
5586         if (!caching_ctl) {
5587                 BUG_ON(!block_group_cache_done(block_group));
5588                 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5589                 BUG_ON(ret);
5590         } else {
5591                 mutex_lock(&caching_ctl->mutex);
5592
5593                 if (start >= caching_ctl->progress) {
5594                         ret = add_excluded_extent(root, start, num_bytes);
5595                         BUG_ON(ret);
5596                 } else if (start + num_bytes <= caching_ctl->progress) {
5597                         ret = btrfs_remove_free_space(block_group,
5598                                                       start, num_bytes);
5599                         BUG_ON(ret);
5600                 } else {
5601                         num_bytes = caching_ctl->progress - start;
5602                         ret = btrfs_remove_free_space(block_group,
5603                                                       start, num_bytes);
5604                         BUG_ON(ret);
5605
5606                         start = caching_ctl->progress;
5607                         num_bytes = ins->objectid + ins->offset -
5608                                     caching_ctl->progress;
5609                         ret = add_excluded_extent(root, start, num_bytes);
5610                         BUG_ON(ret);
5611                 }
5612
5613                 mutex_unlock(&caching_ctl->mutex);
5614                 put_caching_control(caching_ctl);
5615         }
5616
5617         ret = update_reserved_bytes(block_group, ins->offset, 1, 1);
5618         BUG_ON(ret);
5619         btrfs_put_block_group(block_group);
5620         ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
5621                                          0, owner, offset, ins, 1);
5622         return ret;
5623 }
5624
5625 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
5626                                             struct btrfs_root *root,
5627                                             u64 bytenr, u32 blocksize,
5628                                             int level)
5629 {
5630         struct extent_buffer *buf;
5631
5632         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5633         if (!buf)
5634                 return ERR_PTR(-ENOMEM);
5635         btrfs_set_header_generation(buf, trans->transid);
5636         btrfs_set_buffer_lockdep_class(buf, level);
5637         btrfs_tree_lock(buf);
5638         clean_tree_block(trans, root, buf);
5639
5640         btrfs_set_lock_blocking(buf);
5641         btrfs_set_buffer_uptodate(buf);
5642
5643         if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
5644                 /*
5645                  * we allow two log transactions at a time, use different
5646                  * EXENT bit to differentiate dirty pages.
5647                  */
5648                 if (root->log_transid % 2 == 0)
5649                         set_extent_dirty(&root->dirty_log_pages, buf->start,
5650                                         buf->start + buf->len - 1, GFP_NOFS);
5651                 else
5652                         set_extent_new(&root->dirty_log_pages, buf->start,
5653                                         buf->start + buf->len - 1, GFP_NOFS);
5654         } else {
5655                 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
5656                          buf->start + buf->len - 1, GFP_NOFS);
5657         }
5658         trans->blocks_used++;
5659         /* this returns a buffer locked for blocking */
5660         return buf;
5661 }
5662
5663 static struct btrfs_block_rsv *
5664 use_block_rsv(struct btrfs_trans_handle *trans,
5665               struct btrfs_root *root, u32 blocksize)
5666 {
5667         struct btrfs_block_rsv *block_rsv;
5668         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5669         int ret;
5670
5671         block_rsv = get_block_rsv(trans, root);
5672
5673         if (block_rsv->size == 0) {
5674                 ret = reserve_metadata_bytes(trans, root, block_rsv,
5675                                              blocksize, 0);
5676                 /*
5677                  * If we couldn't reserve metadata bytes try and use some from
5678                  * the global reserve.
5679                  */
5680                 if (ret && block_rsv != global_rsv) {
5681                         ret = block_rsv_use_bytes(global_rsv, blocksize);
5682                         if (!ret)
5683                                 return global_rsv;
5684                         return ERR_PTR(ret);
5685                 } else if (ret) {
5686                         return ERR_PTR(ret);
5687                 }
5688                 return block_rsv;
5689         }
5690
5691         ret = block_rsv_use_bytes(block_rsv, blocksize);
5692         if (!ret)
5693                 return block_rsv;
5694         if (ret) {
5695                 WARN_ON(1);
5696                 ret = reserve_metadata_bytes(trans, root, block_rsv, blocksize,
5697                                              0);
5698                 if (!ret) {
5699                         spin_lock(&block_rsv->lock);
5700                         block_rsv->size += blocksize;
5701                         spin_unlock(&block_rsv->lock);
5702                         return block_rsv;
5703                 } else if (ret && block_rsv != global_rsv) {
5704                         ret = block_rsv_use_bytes(global_rsv, blocksize);
5705                         if (!ret)
5706                                 return global_rsv;
5707                 }
5708         }
5709
5710         return ERR_PTR(-ENOSPC);
5711 }
5712
5713 static void unuse_block_rsv(struct btrfs_block_rsv *block_rsv, u32 blocksize)
5714 {
5715         block_rsv_add_bytes(block_rsv, blocksize, 0);
5716         block_rsv_release_bytes(block_rsv, NULL, 0);
5717 }
5718
5719 /*
5720  * finds a free extent and does all the dirty work required for allocation
5721  * returns the key for the extent through ins, and a tree buffer for
5722  * the first block of the extent through buf.
5723  *
5724  * returns the tree buffer or NULL.
5725  */
5726 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
5727                                         struct btrfs_root *root, u32 blocksize,
5728                                         u64 parent, u64 root_objectid,
5729                                         struct btrfs_disk_key *key, int level,
5730                                         u64 hint, u64 empty_size)
5731 {
5732         struct btrfs_key ins;
5733         struct btrfs_block_rsv *block_rsv;
5734         struct extent_buffer *buf;
5735         u64 flags = 0;
5736         int ret;
5737
5738
5739         block_rsv = use_block_rsv(trans, root, blocksize);
5740         if (IS_ERR(block_rsv))
5741                 return ERR_CAST(block_rsv);
5742
5743         ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
5744                                    empty_size, hint, (u64)-1, &ins, 0);
5745         if (ret) {
5746                 unuse_block_rsv(block_rsv, blocksize);
5747                 return ERR_PTR(ret);
5748         }
5749
5750         buf = btrfs_init_new_buffer(trans, root, ins.objectid,
5751                                     blocksize, level);
5752         BUG_ON(IS_ERR(buf));
5753
5754         if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
5755                 if (parent == 0)
5756                         parent = ins.objectid;
5757                 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
5758         } else
5759                 BUG_ON(parent > 0);
5760
5761         if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
5762                 struct btrfs_delayed_extent_op *extent_op;
5763                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
5764                 BUG_ON(!extent_op);
5765                 if (key)
5766                         memcpy(&extent_op->key, key, sizeof(extent_op->key));
5767                 else
5768                         memset(&extent_op->key, 0, sizeof(extent_op->key));
5769                 extent_op->flags_to_set = flags;
5770                 extent_op->update_key = 1;
5771                 extent_op->update_flags = 1;
5772                 extent_op->is_data = 0;
5773
5774                 ret = btrfs_add_delayed_tree_ref(trans, ins.objectid,
5775                                         ins.offset, parent, root_objectid,
5776                                         level, BTRFS_ADD_DELAYED_EXTENT,
5777                                         extent_op);
5778                 BUG_ON(ret);
5779         }
5780         return buf;
5781 }
5782
5783 struct walk_control {
5784         u64 refs[BTRFS_MAX_LEVEL];
5785         u64 flags[BTRFS_MAX_LEVEL];
5786         struct btrfs_key update_progress;
5787         int stage;
5788         int level;
5789         int shared_level;
5790         int update_ref;
5791         int keep_locks;
5792         int reada_slot;
5793         int reada_count;
5794 };
5795
5796 #define DROP_REFERENCE  1
5797 #define UPDATE_BACKREF  2
5798
5799 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
5800                                      struct btrfs_root *root,
5801                                      struct walk_control *wc,
5802                                      struct btrfs_path *path)
5803 {
5804         u64 bytenr;
5805         u64 generation;
5806         u64 refs;
5807         u64 flags;
5808         u32 nritems;
5809         u32 blocksize;
5810         struct btrfs_key key;
5811         struct extent_buffer *eb;
5812         int ret;
5813         int slot;
5814         int nread = 0;
5815
5816         if (path->slots[wc->level] < wc->reada_slot) {
5817                 wc->reada_count = wc->reada_count * 2 / 3;
5818                 wc->reada_count = max(wc->reada_count, 2);
5819         } else {
5820                 wc->reada_count = wc->reada_count * 3 / 2;
5821                 wc->reada_count = min_t(int, wc->reada_count,
5822                                         BTRFS_NODEPTRS_PER_BLOCK(root));
5823         }
5824
5825         eb = path->nodes[wc->level];
5826         nritems = btrfs_header_nritems(eb);
5827         blocksize = btrfs_level_size(root, wc->level - 1);
5828
5829         for (slot = path->slots[wc->level]; slot < nritems; slot++) {
5830                 if (nread >= wc->reada_count)
5831                         break;
5832
5833                 cond_resched();
5834                 bytenr = btrfs_node_blockptr(eb, slot);
5835                 generation = btrfs_node_ptr_generation(eb, slot);
5836
5837                 if (slot == path->slots[wc->level])
5838                         goto reada;
5839
5840                 if (wc->stage == UPDATE_BACKREF &&
5841                     generation <= root->root_key.offset)
5842                         continue;
5843
5844                 /* We don't lock the tree block, it's OK to be racy here */
5845                 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
5846                                                &refs, &flags);
5847                 BUG_ON(ret);
5848                 BUG_ON(refs == 0);
5849
5850                 if (wc->stage == DROP_REFERENCE) {
5851                         if (refs == 1)
5852                                 goto reada;
5853
5854                         if (wc->level == 1 &&
5855                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5856                                 continue;
5857                         if (!wc->update_ref ||
5858                             generation <= root->root_key.offset)
5859                                 continue;
5860                         btrfs_node_key_to_cpu(eb, &key, slot);
5861                         ret = btrfs_comp_cpu_keys(&key,
5862                                                   &wc->update_progress);
5863                         if (ret < 0)
5864                                 continue;
5865                 } else {
5866                         if (wc->level == 1 &&
5867                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5868                                 continue;
5869                 }
5870 reada:
5871                 ret = readahead_tree_block(root, bytenr, blocksize,
5872                                            generation);
5873                 if (ret)
5874                         break;
5875                 nread++;
5876         }
5877         wc->reada_slot = slot;
5878 }
5879
5880 /*
5881  * hepler to process tree block while walking down the tree.
5882  *
5883  * when wc->stage == UPDATE_BACKREF, this function updates
5884  * back refs for pointers in the block.
5885  *
5886  * NOTE: return value 1 means we should stop walking down.
5887  */
5888 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
5889                                    struct btrfs_root *root,
5890                                    struct btrfs_path *path,
5891                                    struct walk_control *wc, int lookup_info)
5892 {
5893         int level = wc->level;
5894         struct extent_buffer *eb = path->nodes[level];
5895         u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5896         int ret;
5897
5898         if (wc->stage == UPDATE_BACKREF &&
5899             btrfs_header_owner(eb) != root->root_key.objectid)
5900                 return 1;
5901
5902         /*
5903          * when reference count of tree block is 1, it won't increase
5904          * again. once full backref flag is set, we never clear it.
5905          */
5906         if (lookup_info &&
5907             ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
5908              (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
5909                 BUG_ON(!path->locks[level]);
5910                 ret = btrfs_lookup_extent_info(trans, root,
5911                                                eb->start, eb->len,
5912                                                &wc->refs[level],
5913                                                &wc->flags[level]);
5914                 BUG_ON(ret);
5915                 BUG_ON(wc->refs[level] == 0);
5916         }
5917
5918         if (wc->stage == DROP_REFERENCE) {
5919                 if (wc->refs[level] > 1)
5920                         return 1;
5921
5922                 if (path->locks[level] && !wc->keep_locks) {
5923                         btrfs_tree_unlock(eb);
5924                         path->locks[level] = 0;
5925                 }
5926                 return 0;
5927         }
5928
5929         /* wc->stage == UPDATE_BACKREF */
5930         if (!(wc->flags[level] & flag)) {
5931                 BUG_ON(!path->locks[level]);
5932                 ret = btrfs_inc_ref(trans, root, eb, 1);
5933                 BUG_ON(ret);
5934                 ret = btrfs_dec_ref(trans, root, eb, 0);
5935                 BUG_ON(ret);
5936                 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
5937                                                   eb->len, flag, 0);
5938                 BUG_ON(ret);
5939                 wc->flags[level] |= flag;
5940         }
5941
5942         /*
5943          * the block is shared by multiple trees, so it's not good to
5944          * keep the tree lock
5945          */
5946         if (path->locks[level] && level > 0) {
5947                 btrfs_tree_unlock(eb);
5948                 path->locks[level] = 0;
5949         }
5950         return 0;
5951 }
5952
5953 /*
5954  * hepler to process tree block pointer.
5955  *
5956  * when wc->stage == DROP_REFERENCE, this function checks
5957  * reference count of the block pointed to. if the block
5958  * is shared and we need update back refs for the subtree
5959  * rooted at the block, this function changes wc->stage to
5960  * UPDATE_BACKREF. if the block is shared and there is no
5961  * need to update back, this function drops the reference
5962  * to the block.
5963  *
5964  * NOTE: return value 1 means we should stop walking down.
5965  */
5966 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
5967                                  struct btrfs_root *root,
5968                                  struct btrfs_path *path,
5969                                  struct walk_control *wc, int *lookup_info)
5970 {
5971         u64 bytenr;
5972         u64 generation;
5973         u64 parent;
5974         u32 blocksize;
5975         struct btrfs_key key;
5976         struct extent_buffer *next;
5977         int level = wc->level;
5978         int reada = 0;
5979         int ret = 0;
5980
5981         generation = btrfs_node_ptr_generation(path->nodes[level],
5982                                                path->slots[level]);
5983         /*
5984          * if the lower level block was created before the snapshot
5985          * was created, we know there is no need to update back refs
5986          * for the subtree
5987          */
5988         if (wc->stage == UPDATE_BACKREF &&
5989             generation <= root->root_key.offset) {
5990                 *lookup_info = 1;
5991                 return 1;
5992         }
5993
5994         bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
5995         blocksize = btrfs_level_size(root, level - 1);
5996
5997         next = btrfs_find_tree_block(root, bytenr, blocksize);
5998         if (!next) {
5999                 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
6000                 if (!next)
6001                         return -ENOMEM;
6002                 reada = 1;
6003         }
6004         btrfs_tree_lock(next);
6005         btrfs_set_lock_blocking(next);
6006
6007         ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6008                                        &wc->refs[level - 1],
6009                                        &wc->flags[level - 1]);
6010         BUG_ON(ret);
6011         BUG_ON(wc->refs[level - 1] == 0);
6012         *lookup_info = 0;
6013
6014         if (wc->stage == DROP_REFERENCE) {
6015                 if (wc->refs[level - 1] > 1) {
6016                         if (level == 1 &&
6017                             (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6018                                 goto skip;
6019
6020                         if (!wc->update_ref ||
6021                             generation <= root->root_key.offset)
6022                                 goto skip;
6023
6024                         btrfs_node_key_to_cpu(path->nodes[level], &key,
6025                                               path->slots[level]);
6026                         ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
6027                         if (ret < 0)
6028                                 goto skip;
6029
6030                         wc->stage = UPDATE_BACKREF;
6031                         wc->shared_level = level - 1;
6032                 }
6033         } else {
6034                 if (level == 1 &&
6035                     (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6036                         goto skip;
6037         }
6038
6039         if (!btrfs_buffer_uptodate(next, generation)) {
6040                 btrfs_tree_unlock(next);
6041                 free_extent_buffer(next);
6042                 next = NULL;
6043                 *lookup_info = 1;
6044         }
6045
6046         if (!next) {
6047                 if (reada && level == 1)
6048                         reada_walk_down(trans, root, wc, path);
6049                 next = read_tree_block(root, bytenr, blocksize, generation);
6050                 btrfs_tree_lock(next);
6051                 btrfs_set_lock_blocking(next);
6052         }
6053
6054         level--;
6055         BUG_ON(level != btrfs_header_level(next));
6056         path->nodes[level] = next;
6057         path->slots[level] = 0;
6058         path->locks[level] = 1;
6059         wc->level = level;
6060         if (wc->level == 1)
6061                 wc->reada_slot = 0;
6062         return 0;
6063 skip:
6064         wc->refs[level - 1] = 0;
6065         wc->flags[level - 1] = 0;
6066         if (wc->stage == DROP_REFERENCE) {
6067                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
6068                         parent = path->nodes[level]->start;
6069                 } else {
6070                         BUG_ON(root->root_key.objectid !=
6071                                btrfs_header_owner(path->nodes[level]));
6072                         parent = 0;
6073                 }
6074
6075                 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
6076                                         root->root_key.objectid, level - 1, 0);
6077                 BUG_ON(ret);
6078         }
6079         btrfs_tree_unlock(next);
6080         free_extent_buffer(next);
6081         *lookup_info = 1;
6082         return 1;
6083 }
6084
6085 /*
6086  * hepler to process tree block while walking up the tree.
6087  *
6088  * when wc->stage == DROP_REFERENCE, this function drops
6089  * reference count on the block.
6090  *
6091  * when wc->stage == UPDATE_BACKREF, this function changes
6092  * wc->stage back to DROP_REFERENCE if we changed wc->stage
6093  * to UPDATE_BACKREF previously while processing the block.
6094  *
6095  * NOTE: return value 1 means we should stop walking up.
6096  */
6097 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
6098                                  struct btrfs_root *root,
6099                                  struct btrfs_path *path,
6100                                  struct walk_control *wc)
6101 {
6102         int ret;
6103         int level = wc->level;
6104         struct extent_buffer *eb = path->nodes[level];
6105         u64 parent = 0;
6106
6107         if (wc->stage == UPDATE_BACKREF) {
6108                 BUG_ON(wc->shared_level < level);
6109                 if (level < wc->shared_level)
6110                         goto out;
6111
6112                 ret = find_next_key(path, level + 1, &wc->update_progress);
6113                 if (ret > 0)
6114                         wc->update_ref = 0;
6115
6116                 wc->stage = DROP_REFERENCE;
6117                 wc->shared_level = -1;
6118                 path->slots[level] = 0;
6119
6120                 /*
6121                  * check reference count again if the block isn't locked.
6122                  * we should start walking down the tree again if reference
6123                  * count is one.
6124                  */
6125                 if (!path->locks[level]) {
6126                         BUG_ON(level == 0);
6127                         btrfs_tree_lock(eb);
6128                         btrfs_set_lock_blocking(eb);
6129                         path->locks[level] = 1;
6130
6131                         ret = btrfs_lookup_extent_info(trans, root,
6132                                                        eb->start, eb->len,
6133                                                        &wc->refs[level],
6134                                                        &wc->flags[level]);
6135                         BUG_ON(ret);
6136                         BUG_ON(wc->refs[level] == 0);
6137                         if (wc->refs[level] == 1) {
6138                                 btrfs_tree_unlock(eb);
6139                                 path->locks[level] = 0;
6140                                 return 1;
6141                         }
6142                 }
6143         }
6144
6145         /* wc->stage == DROP_REFERENCE */
6146         BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
6147
6148         if (wc->refs[level] == 1) {
6149                 if (level == 0) {
6150                         if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6151                                 ret = btrfs_dec_ref(trans, root, eb, 1);
6152                         else
6153                                 ret = btrfs_dec_ref(trans, root, eb, 0);
6154                         BUG_ON(ret);
6155                 }
6156                 /* make block locked assertion in clean_tree_block happy */
6157                 if (!path->locks[level] &&
6158                     btrfs_header_generation(eb) == trans->transid) {
6159                         btrfs_tree_lock(eb);
6160                         btrfs_set_lock_blocking(eb);
6161                         path->locks[level] = 1;
6162                 }
6163                 clean_tree_block(trans, root, eb);
6164         }
6165
6166         if (eb == root->node) {
6167                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6168                         parent = eb->start;
6169                 else
6170                         BUG_ON(root->root_key.objectid !=
6171                                btrfs_header_owner(eb));
6172         } else {
6173                 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6174                         parent = path->nodes[level + 1]->start;
6175                 else
6176                         BUG_ON(root->root_key.objectid !=
6177                                btrfs_header_owner(path->nodes[level + 1]));
6178         }
6179
6180         btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
6181 out:
6182         wc->refs[level] = 0;
6183         wc->flags[level] = 0;
6184         return 0;
6185 }
6186
6187 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
6188                                    struct btrfs_root *root,
6189                                    struct btrfs_path *path,
6190                                    struct walk_control *wc)
6191 {
6192         int level = wc->level;
6193         int lookup_info = 1;
6194         int ret;
6195
6196         while (level >= 0) {
6197                 ret = walk_down_proc(trans, root, path, wc, lookup_info);
6198                 if (ret > 0)
6199                         break;
6200
6201                 if (level == 0)
6202                         break;
6203
6204                 if (path->slots[level] >=
6205                     btrfs_header_nritems(path->nodes[level]))
6206                         break;
6207
6208                 ret = do_walk_down(trans, root, path, wc, &lookup_info);
6209                 if (ret > 0) {
6210                         path->slots[level]++;
6211                         continue;
6212                 } else if (ret < 0)
6213                         return ret;
6214                 level = wc->level;
6215         }
6216         return 0;
6217 }
6218
6219 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
6220                                  struct btrfs_root *root,
6221                                  struct btrfs_path *path,
6222                                  struct walk_control *wc, int max_level)
6223 {
6224         int level = wc->level;
6225         int ret;
6226
6227         path->slots[level] = btrfs_header_nritems(path->nodes[level]);
6228         while (level < max_level && path->nodes[level]) {
6229                 wc->level = level;
6230                 if (path->slots[level] + 1 <
6231                     btrfs_header_nritems(path->nodes[level])) {
6232                         path->slots[level]++;
6233                         return 0;
6234                 } else {
6235                         ret = walk_up_proc(trans, root, path, wc);
6236                         if (ret > 0)
6237                                 return 0;
6238
6239                         if (path->locks[level]) {
6240                                 btrfs_tree_unlock(path->nodes[level]);
6241                                 path->locks[level] = 0;
6242                         }
6243                         free_extent_buffer(path->nodes[level]);
6244                         path->nodes[level] = NULL;
6245                         level++;
6246                 }
6247         }
6248         return 1;
6249 }
6250
6251 /*
6252  * drop a subvolume tree.
6253  *
6254  * this function traverses the tree freeing any blocks that only
6255  * referenced by the tree.
6256  *
6257  * when a shared tree block is found. this function decreases its
6258  * reference count by one. if update_ref is true, this function
6259  * also make sure backrefs for the shared block and all lower level
6260  * blocks are properly updated.
6261  */
6262 int btrfs_drop_snapshot(struct btrfs_root *root,
6263                         struct btrfs_block_rsv *block_rsv, int update_ref)
6264 {
6265         struct btrfs_path *path;
6266         struct btrfs_trans_handle *trans;
6267         struct btrfs_root *tree_root = root->fs_info->tree_root;
6268         struct btrfs_root_item *root_item = &root->root_item;
6269         struct walk_control *wc;
6270         struct btrfs_key key;
6271         int err = 0;
6272         int ret;
6273         int level;
6274
6275         path = btrfs_alloc_path();
6276         BUG_ON(!path);
6277
6278         wc = kzalloc(sizeof(*wc), GFP_NOFS);
6279         BUG_ON(!wc);
6280
6281         trans = btrfs_start_transaction(tree_root, 0);
6282         BUG_ON(IS_ERR(trans));
6283
6284         if (block_rsv)
6285                 trans->block_rsv = block_rsv;
6286
6287         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
6288                 level = btrfs_header_level(root->node);
6289                 path->nodes[level] = btrfs_lock_root_node(root);
6290                 btrfs_set_lock_blocking(path->nodes[level]);
6291                 path->slots[level] = 0;
6292                 path->locks[level] = 1;
6293                 memset(&wc->update_progress, 0,
6294                        sizeof(wc->update_progress));
6295         } else {
6296                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
6297                 memcpy(&wc->update_progress, &key,
6298                        sizeof(wc->update_progress));
6299
6300                 level = root_item->drop_level;
6301                 BUG_ON(level == 0);
6302                 path->lowest_level = level;
6303                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6304                 path->lowest_level = 0;
6305                 if (ret < 0) {
6306                         err = ret;
6307                         goto out;
6308                 }
6309                 WARN_ON(ret > 0);
6310
6311                 /*
6312                  * unlock our path, this is safe because only this
6313                  * function is allowed to delete this snapshot
6314                  */
6315                 btrfs_unlock_up_safe(path, 0);
6316
6317                 level = btrfs_header_level(root->node);
6318                 while (1) {
6319                         btrfs_tree_lock(path->nodes[level]);
6320                         btrfs_set_lock_blocking(path->nodes[level]);
6321
6322                         ret = btrfs_lookup_extent_info(trans, root,
6323                                                 path->nodes[level]->start,
6324                                                 path->nodes[level]->len,
6325                                                 &wc->refs[level],
6326                                                 &wc->flags[level]);
6327                         BUG_ON(ret);
6328                         BUG_ON(wc->refs[level] == 0);
6329
6330                         if (level == root_item->drop_level)
6331                                 break;
6332
6333                         btrfs_tree_unlock(path->nodes[level]);
6334                         WARN_ON(wc->refs[level] != 1);
6335                         level--;
6336                 }
6337         }
6338
6339         wc->level = level;
6340         wc->shared_level = -1;
6341         wc->stage = DROP_REFERENCE;
6342         wc->update_ref = update_ref;
6343         wc->keep_locks = 0;
6344         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6345
6346         while (1) {
6347                 ret = walk_down_tree(trans, root, path, wc);
6348                 if (ret < 0) {
6349                         err = ret;
6350                         break;
6351                 }
6352
6353                 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
6354                 if (ret < 0) {
6355                         err = ret;
6356                         break;
6357                 }
6358
6359                 if (ret > 0) {
6360                         BUG_ON(wc->stage != DROP_REFERENCE);
6361                         break;
6362                 }
6363
6364                 if (wc->stage == DROP_REFERENCE) {
6365                         level = wc->level;
6366                         btrfs_node_key(path->nodes[level],
6367                                        &root_item->drop_progress,
6368                                        path->slots[level]);
6369                         root_item->drop_level = level;
6370                 }
6371
6372                 BUG_ON(wc->level == 0);
6373                 if (btrfs_should_end_transaction(trans, tree_root)) {
6374                         ret = btrfs_update_root(trans, tree_root,
6375                                                 &root->root_key,
6376                                                 root_item);
6377                         BUG_ON(ret);
6378
6379                         btrfs_end_transaction_throttle(trans, tree_root);
6380                         trans = btrfs_start_transaction(tree_root, 0);
6381                         BUG_ON(IS_ERR(trans));
6382                         if (block_rsv)
6383                                 trans->block_rsv = block_rsv;
6384                 }
6385         }
6386         btrfs_release_path(root, path);
6387         BUG_ON(err);
6388
6389         ret = btrfs_del_root(trans, tree_root, &root->root_key);
6390         BUG_ON(ret);
6391
6392         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
6393                 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
6394                                            NULL, NULL);
6395                 BUG_ON(ret < 0);
6396                 if (ret > 0) {
6397                         /* if we fail to delete the orphan item this time
6398                          * around, it'll get picked up the next time.
6399                          *
6400                          * The most common failure here is just -ENOENT.
6401                          */
6402                         btrfs_del_orphan_item(trans, tree_root,
6403                                               root->root_key.objectid);
6404                 }
6405         }
6406
6407         if (root->in_radix) {
6408                 btrfs_free_fs_root(tree_root->fs_info, root);
6409         } else {
6410                 free_extent_buffer(root->node);
6411                 free_extent_buffer(root->commit_root);
6412                 kfree(root);
6413         }
6414 out:
6415         btrfs_end_transaction_throttle(trans, tree_root);
6416         kfree(wc);
6417         btrfs_free_path(path);
6418         return err;
6419 }
6420
6421 /*
6422  * drop subtree rooted at tree block 'node'.
6423  *
6424  * NOTE: this function will unlock and release tree block 'node'
6425  */
6426 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
6427                         struct btrfs_root *root,
6428                         struct extent_buffer *node,
6429                         struct extent_buffer *parent)
6430 {
6431         struct btrfs_path *path;
6432         struct walk_control *wc;
6433         int level;
6434         int parent_level;
6435         int ret = 0;
6436         int wret;
6437
6438         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6439
6440         path = btrfs_alloc_path();
6441         BUG_ON(!path);
6442
6443         wc = kzalloc(sizeof(*wc), GFP_NOFS);
6444         BUG_ON(!wc);
6445
6446         btrfs_assert_tree_locked(parent);
6447         parent_level = btrfs_header_level(parent);
6448         extent_buffer_get(parent);
6449         path->nodes[parent_level] = parent;
6450         path->slots[parent_level] = btrfs_header_nritems(parent);
6451
6452         btrfs_assert_tree_locked(node);
6453         level = btrfs_header_level(node);
6454         path->nodes[level] = node;
6455         path->slots[level] = 0;
6456         path->locks[level] = 1;
6457
6458         wc->refs[parent_level] = 1;
6459         wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6460         wc->level = level;
6461         wc->shared_level = -1;
6462         wc->stage = DROP_REFERENCE;
6463         wc->update_ref = 0;
6464         wc->keep_locks = 1;
6465         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6466
6467         while (1) {
6468                 wret = walk_down_tree(trans, root, path, wc);
6469                 if (wret < 0) {
6470                         ret = wret;
6471                         break;
6472                 }
6473
6474                 wret = walk_up_tree(trans, root, path, wc, parent_level);
6475                 if (wret < 0)
6476                         ret = wret;
6477                 if (wret != 0)
6478                         break;
6479         }
6480
6481         kfree(wc);
6482         btrfs_free_path(path);
6483         return ret;
6484 }
6485
6486 #if 0
6487 static unsigned long calc_ra(unsigned long start, unsigned long last,
6488                              unsigned long nr)
6489 {
6490         return min(last, start + nr - 1);
6491 }
6492
6493 static noinline int relocate_inode_pages(struct inode *inode, u64 start,
6494                                          u64 len)
6495 {
6496         u64 page_start;
6497         u64 page_end;
6498         unsigned long first_index;
6499         unsigned long last_index;
6500         unsigned long i;
6501         struct page *page;
6502         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6503         struct file_ra_state *ra;
6504         struct btrfs_ordered_extent *ordered;
6505         unsigned int total_read = 0;
6506         unsigned int total_dirty = 0;
6507         int ret = 0;
6508
6509         ra = kzalloc(sizeof(*ra), GFP_NOFS);
6510         if (!ra)
6511                 return -ENOMEM;
6512
6513         mutex_lock(&inode->i_mutex);
6514         first_index = start >> PAGE_CACHE_SHIFT;
6515         last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
6516
6517         /* make sure the dirty trick played by the caller work */
6518         ret = invalidate_inode_pages2_range(inode->i_mapping,
6519                                             first_index, last_index);
6520         if (ret)
6521                 goto out_unlock;
6522
6523         file_ra_state_init(ra, inode->i_mapping);
6524
6525         for (i = first_index ; i <= last_index; i++) {
6526                 if (total_read % ra->ra_pages == 0) {
6527                         btrfs_force_ra(inode->i_mapping, ra, NULL, i,
6528                                        calc_ra(i, last_index, ra->ra_pages));
6529                 }
6530                 total_read++;
6531 again:
6532                 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
6533                         BUG_ON(1);
6534                 page = grab_cache_page(inode->i_mapping, i);
6535                 if (!page) {
6536                         ret = -ENOMEM;
6537                         goto out_unlock;
6538                 }
6539                 if (!PageUptodate(page)) {
6540                         btrfs_readpage(NULL, page);
6541                         lock_page(page);
6542                         if (!PageUptodate(page)) {
6543                                 unlock_page(page);
6544                                 page_cache_release(page);
6545                                 ret = -EIO;
6546                                 goto out_unlock;
6547                         }
6548                 }
6549                 wait_on_page_writeback(page);
6550
6551                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
6552                 page_end = page_start + PAGE_CACHE_SIZE - 1;
6553                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
6554
6555                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6556                 if (ordered) {
6557                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
6558                         unlock_page(page);
6559                         page_cache_release(page);
6560                         btrfs_start_ordered_extent(inode, ordered, 1);
6561                         btrfs_put_ordered_extent(ordered);
6562                         goto again;
6563                 }
6564                 set_page_extent_mapped(page);
6565
6566                 if (i == first_index)
6567                         set_extent_bits(io_tree, page_start, page_end,
6568                                         EXTENT_BOUNDARY, GFP_NOFS);
6569                 btrfs_set_extent_delalloc(inode, page_start, page_end);
6570
6571                 set_page_dirty(page);
6572                 total_dirty++;
6573
6574                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
6575                 unlock_page(page);
6576                 page_cache_release(page);
6577         }
6578
6579 out_unlock:
6580         kfree(ra);
6581         mutex_unlock(&inode->i_mutex);
6582         balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
6583         return ret;
6584 }
6585
6586 static noinline int relocate_data_extent(struct inode *reloc_inode,
6587                                          struct btrfs_key *extent_key,
6588                                          u64 offset)
6589 {
6590         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
6591         struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
6592         struct extent_map *em;
6593         u64 start = extent_key->objectid - offset;
6594         u64 end = start + extent_key->offset - 1;
6595
6596         em = alloc_extent_map(GFP_NOFS);
6597         BUG_ON(!em);
6598
6599         em->start = start;
6600         em->len = extent_key->offset;
6601         em->block_len = extent_key->offset;
6602         em->block_start = extent_key->objectid;
6603         em->bdev = root->fs_info->fs_devices->latest_bdev;
6604         set_bit(EXTENT_FLAG_PINNED, &em->flags);
6605
6606         /* setup extent map to cheat btrfs_readpage */
6607         lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
6608         while (1) {
6609                 int ret;
6610                 write_lock(&em_tree->lock);
6611                 ret = add_extent_mapping(em_tree, em);
6612                 write_unlock(&em_tree->lock);
6613                 if (ret != -EEXIST) {
6614                         free_extent_map(em);
6615                         break;
6616                 }
6617                 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
6618         }
6619         unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
6620
6621         return relocate_inode_pages(reloc_inode, start, extent_key->offset);
6622 }
6623
6624 struct btrfs_ref_path {
6625         u64 extent_start;
6626         u64 nodes[BTRFS_MAX_LEVEL];
6627         u64 root_objectid;
6628         u64 root_generation;
6629         u64 owner_objectid;
6630         u32 num_refs;
6631         int lowest_level;
6632         int current_level;
6633         int shared_level;
6634
6635         struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
6636         u64 new_nodes[BTRFS_MAX_LEVEL];
6637 };
6638
6639 struct disk_extent {
6640         u64 ram_bytes;
6641         u64 disk_bytenr;
6642         u64 disk_num_bytes;
6643         u64 offset;
6644         u64 num_bytes;
6645         u8 compression;
6646         u8 encryption;
6647         u16 other_encoding;
6648 };
6649
6650 static int is_cowonly_root(u64 root_objectid)
6651 {
6652         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
6653             root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
6654             root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
6655             root_objectid == BTRFS_DEV_TREE_OBJECTID ||
6656             root_objectid == BTRFS_TREE_LOG_OBJECTID ||
6657             root_objectid == BTRFS_CSUM_TREE_OBJECTID)
6658                 return 1;
6659         return 0;
6660 }
6661
6662 static noinline int __next_ref_path(struct btrfs_trans_handle *trans,
6663                                     struct btrfs_root *extent_root,
6664                                     struct btrfs_ref_path *ref_path,
6665                                     int first_time)
6666 {
6667         struct extent_buffer *leaf;
6668         struct btrfs_path *path;
6669         struct btrfs_extent_ref *ref;
6670         struct btrfs_key key;
6671         struct btrfs_key found_key;
6672         u64 bytenr;
6673         u32 nritems;
6674         int level;
6675         int ret = 1;
6676
6677         path = btrfs_alloc_path();
6678         if (!path)
6679                 return -ENOMEM;
6680
6681         if (first_time) {
6682                 ref_path->lowest_level = -1;
6683                 ref_path->current_level = -1;
6684                 ref_path->shared_level = -1;
6685                 goto walk_up;
6686         }
6687 walk_down:
6688         level = ref_path->current_level - 1;
6689         while (level >= -1) {
6690                 u64 parent;
6691                 if (level < ref_path->lowest_level)
6692                         break;
6693
6694                 if (level >= 0)
6695                         bytenr = ref_path->nodes[level];
6696                 else
6697                         bytenr = ref_path->extent_start;
6698                 BUG_ON(bytenr == 0);
6699
6700                 parent = ref_path->nodes[level + 1];
6701                 ref_path->nodes[level + 1] = 0;
6702                 ref_path->current_level = level;
6703                 BUG_ON(parent == 0);
6704
6705                 key.objectid = bytenr;
6706                 key.offset = parent + 1;
6707                 key.type = BTRFS_EXTENT_REF_KEY;
6708
6709                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
6710                 if (ret < 0)
6711                         goto out;
6712                 BUG_ON(ret == 0);
6713
6714                 leaf = path->nodes[0];
6715                 nritems = btrfs_header_nritems(leaf);
6716                 if (path->slots[0] >= nritems) {
6717                         ret = btrfs_next_leaf(extent_root, path);
6718                         if (ret < 0)
6719                                 goto out;
6720                         if (ret > 0)
6721                                 goto next;
6722                         leaf = path->nodes[0];
6723                 }
6724
6725                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6726                 if (found_key.objectid == bytenr &&
6727                     found_key.type == BTRFS_EXTENT_REF_KEY) {
6728                         if (level < ref_path->shared_level)
6729                                 ref_path->shared_level = level;
6730                         goto found;
6731                 }
6732 next:
6733                 level--;
6734                 btrfs_release_path(extent_root, path);
6735                 cond_resched();
6736         }
6737         /* reached lowest level */
6738         ret = 1;
6739         goto out;
6740 walk_up:
6741         level = ref_path->current_level;
6742         while (level < BTRFS_MAX_LEVEL - 1) {
6743                 u64 ref_objectid;
6744
6745                 if (level >= 0)
6746                         bytenr = ref_path->nodes[level];
6747                 else
6748                         bytenr = ref_path->extent_start;
6749
6750                 BUG_ON(bytenr == 0);
6751
6752                 key.objectid = bytenr;
6753                 key.offset = 0;
6754                 key.type = BTRFS_EXTENT_REF_KEY;
6755
6756                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
6757                 if (ret < 0)
6758                         goto out;
6759
6760                 leaf = path->nodes[0];
6761                 nritems = btrfs_header_nritems(leaf);
6762                 if (path->slots[0] >= nritems) {
6763                         ret = btrfs_next_leaf(extent_root, path);
6764                         if (ret < 0)
6765                                 goto out;
6766                         if (ret > 0) {
6767                                 /* the extent was freed by someone */
6768                                 if (ref_path->lowest_level == level)
6769                                         goto out;
6770                                 btrfs_release_path(extent_root, path);
6771                                 goto walk_down;
6772                         }
6773                         leaf = path->nodes[0];
6774                 }
6775
6776                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6777                 if (found_key.objectid != bytenr ||
6778                                 found_key.type != BTRFS_EXTENT_REF_KEY) {
6779                         /* the extent was freed by someone */
6780                         if (ref_path->lowest_level == level) {
6781                                 ret = 1;
6782                                 goto out;
6783                         }
6784                         btrfs_release_path(extent_root, path);
6785                         goto walk_down;
6786                 }
6787 found:
6788                 ref = btrfs_item_ptr(leaf, path->slots[0],
6789                                 struct btrfs_extent_ref);
6790                 ref_objectid = btrfs_ref_objectid(leaf, ref);
6791                 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
6792                         if (first_time) {
6793                                 level = (int)ref_objectid;
6794                                 BUG_ON(level >= BTRFS_MAX_LEVEL);
6795                                 ref_path->lowest_level = level;
6796                                 ref_path->current_level = level;
6797                                 ref_path->nodes[level] = bytenr;
6798                         } else {
6799                                 WARN_ON(ref_objectid != level);
6800                         }
6801                 } else {
6802                         WARN_ON(level != -1);
6803                 }
6804                 first_time = 0;
6805
6806                 if (ref_path->lowest_level == level) {
6807                         ref_path->owner_objectid = ref_objectid;
6808                         ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
6809                 }
6810
6811                 /*
6812                  * the block is tree root or the block isn't in reference
6813                  * counted tree.
6814                  */
6815                 if (found_key.objectid == found_key.offset ||
6816                     is_cowonly_root(btrfs_ref_root(leaf, ref))) {
6817                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
6818                         ref_path->root_generation =
6819                                 btrfs_ref_generation(leaf, ref);
6820                         if (level < 0) {
6821                                 /* special reference from the tree log */
6822                                 ref_path->nodes[0] = found_key.offset;
6823                                 ref_path->current_level = 0;
6824                         }
6825                         ret = 0;
6826                         goto out;
6827                 }
6828
6829                 level++;
6830                 BUG_ON(ref_path->nodes[level] != 0);
6831                 ref_path->nodes[level] = found_key.offset;
6832                 ref_path->current_level = level;
6833
6834                 /*
6835                  * the reference was created in the running transaction,
6836                  * no need to continue walking up.
6837                  */
6838                 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
6839                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
6840                         ref_path->root_generation =
6841                                 btrfs_ref_generation(leaf, ref);
6842                         ret = 0;
6843                         goto out;
6844                 }
6845
6846                 btrfs_release_path(extent_root, path);
6847                 cond_resched();
6848         }
6849         /* reached max tree level, but no tree root found. */
6850         BUG();
6851 out:
6852         btrfs_free_path(path);
6853         return ret;
6854 }
6855
6856 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
6857                                 struct btrfs_root *extent_root,
6858                                 struct btrfs_ref_path *ref_path,
6859                                 u64 extent_start)
6860 {
6861         memset(ref_path, 0, sizeof(*ref_path));
6862         ref_path->extent_start = extent_start;
6863
6864         return __next_ref_path(trans, extent_root, ref_path, 1);
6865 }
6866
6867 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
6868                                struct btrfs_root *extent_root,
6869                                struct btrfs_ref_path *ref_path)
6870 {
6871         return __next_ref_path(trans, extent_root, ref_path, 0);
6872 }
6873
6874 static noinline int get_new_locations(struct inode *reloc_inode,
6875                                       struct btrfs_key *extent_key,
6876                                       u64 offset, int no_fragment,
6877                                       struct disk_extent **extents,
6878                                       int *nr_extents)
6879 {
6880         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
6881         struct btrfs_path *path;
6882         struct btrfs_file_extent_item *fi;
6883         struct extent_buffer *leaf;
6884         struct disk_extent *exts = *extents;
6885         struct btrfs_key found_key;
6886         u64 cur_pos;
6887         u64 last_byte;
6888         u32 nritems;
6889         int nr = 0;
6890         int max = *nr_extents;
6891         int ret;
6892
6893         WARN_ON(!no_fragment && *extents);
6894         if (!exts) {
6895                 max = 1;
6896                 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
6897                 if (!exts)
6898                         return -ENOMEM;
6899         }
6900
6901         path = btrfs_alloc_path();
6902         BUG_ON(!path);
6903
6904         cur_pos = extent_key->objectid - offset;
6905         last_byte = extent_key->objectid + extent_key->offset;
6906         ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
6907                                        cur_pos, 0);
6908         if (ret < 0)
6909                 goto out;
6910         if (ret > 0) {
6911                 ret = -ENOENT;
6912                 goto out;
6913         }
6914
6915         while (1) {
6916                 leaf = path->nodes[0];
6917                 nritems = btrfs_header_nritems(leaf);
6918                 if (path->slots[0] >= nritems) {
6919                         ret = btrfs_next_leaf(root, path);
6920                         if (ret < 0)
6921                                 goto out;
6922                         if (ret > 0)
6923                                 break;
6924                         leaf = path->nodes[0];
6925                 }
6926
6927                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6928                 if (found_key.offset != cur_pos ||
6929                     found_key.type != BTRFS_EXTENT_DATA_KEY ||
6930                     found_key.objectid != reloc_inode->i_ino)
6931                         break;
6932
6933                 fi = btrfs_item_ptr(leaf, path->slots[0],
6934                                     struct btrfs_file_extent_item);
6935                 if (btrfs_file_extent_type(leaf, fi) !=
6936                     BTRFS_FILE_EXTENT_REG ||
6937                     btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
6938                         break;
6939
6940                 if (nr == max) {
6941                         struct disk_extent *old = exts;
6942                         max *= 2;
6943                         exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
6944                         memcpy(exts, old, sizeof(*exts) * nr);
6945                         if (old != *extents)
6946                                 kfree(old);
6947                 }
6948
6949                 exts[nr].disk_bytenr =
6950                         btrfs_file_extent_disk_bytenr(leaf, fi);
6951                 exts[nr].disk_num_bytes =
6952                         btrfs_file_extent_disk_num_bytes(leaf, fi);
6953                 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
6954                 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
6955                 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6956                 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
6957                 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
6958                 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
6959                                                                            fi);
6960                 BUG_ON(exts[nr].offset > 0);
6961                 BUG_ON(exts[nr].compression || exts[nr].encryption);
6962                 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
6963
6964                 cur_pos += exts[nr].num_bytes;
6965                 nr++;
6966
6967                 if (cur_pos + offset >= last_byte)
6968                         break;
6969
6970                 if (no_fragment) {
6971                         ret = 1;
6972                         goto out;
6973                 }
6974                 path->slots[0]++;
6975         }
6976
6977         BUG_ON(cur_pos + offset > last_byte);
6978         if (cur_pos + offset < last_byte) {
6979                 ret = -ENOENT;
6980                 goto out;
6981         }
6982         ret = 0;
6983 out:
6984         btrfs_free_path(path);
6985         if (ret) {
6986                 if (exts != *extents)
6987                         kfree(exts);
6988         } else {
6989                 *extents = exts;
6990                 *nr_extents = nr;
6991         }
6992         return ret;
6993 }
6994
6995 static noinline int replace_one_extent(struct btrfs_trans_handle *trans,
6996                                         struct btrfs_root *root,
6997                                         struct btrfs_path *path,
6998                                         struct btrfs_key *extent_key,
6999                                         struct btrfs_key *leaf_key,
7000                                         struct btrfs_ref_path *ref_path,
7001                                         struct disk_extent *new_extents,
7002                                         int nr_extents)
7003 {
7004         struct extent_buffer *leaf;
7005         struct btrfs_file_extent_item *fi;
7006         struct inode *inode = NULL;
7007         struct btrfs_key key;
7008         u64 lock_start = 0;
7009         u64 lock_end = 0;
7010         u64 num_bytes;
7011         u64 ext_offset;
7012         u64 search_end = (u64)-1;
7013         u32 nritems;
7014         int nr_scaned = 0;
7015         int extent_locked = 0;
7016         int extent_type;
7017         int ret;
7018
7019         memcpy(&key, leaf_key, sizeof(key));
7020         if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
7021                 if (key.objectid < ref_path->owner_objectid ||
7022                     (key.objectid == ref_path->owner_objectid &&
7023                      key.type < BTRFS_EXTENT_DATA_KEY)) {
7024                         key.objectid = ref_path->owner_objectid;
7025                         key.type = BTRFS_EXTENT_DATA_KEY;
7026                         key.offset = 0;
7027                 }
7028         }
7029
7030         while (1) {
7031                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
7032                 if (ret < 0)
7033                         goto out;
7034
7035                 leaf = path->nodes[0];
7036                 nritems = btrfs_header_nritems(leaf);
7037 next:
7038                 if (extent_locked && ret > 0) {
7039                         /*
7040                          * the file extent item was modified by someone
7041                          * before the extent got locked.
7042                          */
7043                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7044                                       lock_end, GFP_NOFS);
7045                         extent_locked = 0;
7046                 }
7047
7048                 if (path->slots[0] >= nritems) {
7049                         if (++nr_scaned > 2)
7050                                 break;
7051
7052                         BUG_ON(extent_locked);
7053                         ret = btrfs_next_leaf(root, path);
7054                         if (ret < 0)
7055                                 goto out;
7056                         if (ret > 0)
7057                                 break;
7058                         leaf = path->nodes[0];
7059                         nritems = btrfs_header_nritems(leaf);
7060                 }
7061
7062                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
7063
7064                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
7065                         if ((key.objectid > ref_path->owner_objectid) ||
7066                             (key.objectid == ref_path->owner_objectid &&
7067                              key.type > BTRFS_EXTENT_DATA_KEY) ||
7068                             key.offset >= search_end)
7069                                 break;
7070                 }
7071
7072                 if (inode && key.objectid != inode->i_ino) {
7073                         BUG_ON(extent_locked);
7074                         btrfs_release_path(root, path);
7075                         mutex_unlock(&inode->i_mutex);
7076                         iput(inode);
7077                         inode = NULL;
7078                         continue;
7079                 }
7080
7081                 if (key.type != BTRFS_EXTENT_DATA_KEY) {
7082                         path->slots[0]++;
7083                         ret = 1;
7084                         goto next;
7085                 }
7086                 fi = btrfs_item_ptr(leaf, path->slots[0],
7087                                     struct btrfs_file_extent_item);
7088                 extent_type = btrfs_file_extent_type(leaf, fi);
7089                 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
7090                      extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
7091                     (btrfs_file_extent_disk_bytenr(leaf, fi) !=
7092                      extent_key->objectid)) {
7093                         path->slots[0]++;
7094                         ret = 1;
7095                         goto next;
7096                 }
7097
7098                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
7099                 ext_offset = btrfs_file_extent_offset(leaf, fi);
7100
7101                 if (search_end == (u64)-1) {
7102                         search_end = key.offset - ext_offset +
7103                                 btrfs_file_extent_ram_bytes(leaf, fi);
7104                 }
7105
7106                 if (!extent_locked) {
7107                         lock_start = key.offset;
7108                         lock_end = lock_start + num_bytes - 1;
7109                 } else {
7110                         if (lock_start > key.offset ||
7111                             lock_end + 1 < key.offset + num_bytes) {
7112                                 unlock_extent(&BTRFS_I(inode)->io_tree,
7113                                               lock_start, lock_end, GFP_NOFS);
7114                                 extent_locked = 0;
7115                         }
7116                 }
7117
7118                 if (!inode) {
7119                         btrfs_release_path(root, path);
7120
7121                         inode = btrfs_iget_locked(root->fs_info->sb,
7122                                                   key.objectid, root);
7123                         if (inode->i_state & I_NEW) {
7124                                 BTRFS_I(inode)->root = root;
7125                                 BTRFS_I(inode)->location.objectid =
7126                                         key.objectid;
7127                                 BTRFS_I(inode)->location.type =
7128                                         BTRFS_INODE_ITEM_KEY;
7129                                 BTRFS_I(inode)->location.offset = 0;
7130                                 btrfs_read_locked_inode(inode);
7131                                 unlock_new_inode(inode);
7132                         }
7133                         /*
7134                          * some code call btrfs_commit_transaction while
7135                          * holding the i_mutex, so we can't use mutex_lock
7136                          * here.
7137                          */
7138                         if (is_bad_inode(inode) ||
7139                             !mutex_trylock(&inode->i_mutex)) {
7140                                 iput(inode);
7141                                 inode = NULL;
7142                                 key.offset = (u64)-1;
7143                                 goto skip;
7144                         }
7145                 }
7146
7147                 if (!extent_locked) {
7148                         struct btrfs_ordered_extent *ordered;
7149
7150                         btrfs_release_path(root, path);
7151
7152                         lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7153                                     lock_end, GFP_NOFS);
7154                         ordered = btrfs_lookup_first_ordered_extent(inode,
7155                                                                     lock_end);
7156                         if (ordered &&
7157                             ordered->file_offset <= lock_end &&
7158                             ordered->file_offset + ordered->len > lock_start) {
7159                                 unlock_extent(&BTRFS_I(inode)->io_tree,
7160                                               lock_start, lock_end, GFP_NOFS);
7161                                 btrfs_start_ordered_extent(inode, ordered, 1);
7162                                 btrfs_put_ordered_extent(ordered);
7163                                 key.offset += num_bytes;
7164                                 goto skip;
7165                         }
7166                         if (ordered)
7167                                 btrfs_put_ordered_extent(ordered);
7168
7169                         extent_locked = 1;
7170                         continue;
7171                 }
7172
7173                 if (nr_extents == 1) {
7174                         /* update extent pointer in place */
7175                         btrfs_set_file_extent_disk_bytenr(leaf, fi,
7176                                                 new_extents[0].disk_bytenr);
7177                         btrfs_set_file_extent_disk_num_bytes(leaf, fi,
7178                                                 new_extents[0].disk_num_bytes);
7179                         btrfs_mark_buffer_dirty(leaf);
7180
7181                         btrfs_drop_extent_cache(inode, key.offset,
7182                                                 key.offset + num_bytes - 1, 0);
7183
7184                         ret = btrfs_inc_extent_ref(trans, root,
7185                                                 new_extents[0].disk_bytenr,
7186                                                 new_extents[0].disk_num_bytes,
7187                                                 leaf->start,
7188                                                 root->root_key.objectid,
7189                                                 trans->transid,
7190                                                 key.objectid);
7191                         BUG_ON(ret);
7192
7193                         ret = btrfs_free_extent(trans, root,
7194                                                 extent_key->objectid,
7195                                                 extent_key->offset,
7196                                                 leaf->start,
7197                                                 btrfs_header_owner(leaf),
7198                                                 btrfs_header_generation(leaf),
7199                                                 key.objectid, 0);
7200                         BUG_ON(ret);
7201
7202                         btrfs_release_path(root, path);
7203                         key.offset += num_bytes;
7204                 } else {
7205                         BUG_ON(1);
7206 #if 0
7207                         u64 alloc_hint;
7208                         u64 extent_len;
7209                         int i;
7210                         /*
7211                          * drop old extent pointer at first, then insert the
7212                          * new pointers one bye one
7213                          */
7214                         btrfs_release_path(root, path);
7215                         ret = btrfs_drop_extents(trans, root, inode, key.offset,
7216                                                  key.offset + num_bytes,
7217                                                  key.offset, &alloc_hint);
7218                         BUG_ON(ret);
7219
7220                         for (i = 0; i < nr_extents; i++) {
7221                                 if (ext_offset >= new_extents[i].num_bytes) {
7222                                         ext_offset -= new_extents[i].num_bytes;
7223                                         continue;
7224                                 }
7225                                 extent_len = min(new_extents[i].num_bytes -
7226                                                  ext_offset, num_bytes);
7227
7228                                 ret = btrfs_insert_empty_item(trans, root,
7229                                                               path, &key,
7230                                                               sizeof(*fi));
7231                                 BUG_ON(ret);
7232
7233                                 leaf = path->nodes[0];
7234                                 fi = btrfs_item_ptr(leaf, path->slots[0],
7235                                                 struct btrfs_file_extent_item);
7236                                 btrfs_set_file_extent_generation(leaf, fi,
7237                                                         trans->transid);
7238                                 btrfs_set_file_extent_type(leaf, fi,
7239                                                         BTRFS_FILE_EXTENT_REG);
7240                                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
7241                                                 new_extents[i].disk_bytenr);
7242                                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
7243                                                 new_extents[i].disk_num_bytes);
7244                                 btrfs_set_file_extent_ram_bytes(leaf, fi,
7245                                                 new_extents[i].ram_bytes);
7246
7247                                 btrfs_set_file_extent_compression(leaf, fi,
7248                                                 new_extents[i].compression);
7249                                 btrfs_set_file_extent_encryption(leaf, fi,
7250                                                 new_extents[i].encryption);
7251                                 btrfs_set_file_extent_other_encoding(leaf, fi,
7252                                                 new_extents[i].other_encoding);
7253
7254                                 btrfs_set_file_extent_num_bytes(leaf, fi,
7255                                                         extent_len);
7256                                 ext_offset += new_extents[i].offset;
7257                                 btrfs_set_file_extent_offset(leaf, fi,
7258                                                         ext_offset);
7259                                 btrfs_mark_buffer_dirty(leaf);
7260
7261                                 btrfs_drop_extent_cache(inode, key.offset,
7262                                                 key.offset + extent_len - 1, 0);
7263
7264                                 ret = btrfs_inc_extent_ref(trans, root,
7265                                                 new_extents[i].disk_bytenr,
7266                                                 new_extents[i].disk_num_bytes,
7267                                                 leaf->start,
7268                                                 root->root_key.objectid,
7269                                                 trans->transid, key.objectid);
7270                                 BUG_ON(ret);
7271                                 btrfs_release_path(root, path);
7272
7273                                 inode_add_bytes(inode, extent_len);
7274
7275                                 ext_offset = 0;
7276                                 num_bytes -= extent_len;
7277                                 key.offset += extent_len;
7278
7279                                 if (num_bytes == 0)
7280                                         break;
7281                         }
7282                         BUG_ON(i >= nr_extents);
7283 #endif
7284                 }
7285
7286                 if (extent_locked) {
7287                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7288                                       lock_end, GFP_NOFS);
7289                         extent_locked = 0;
7290                 }
7291 skip:
7292                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
7293                     key.offset >= search_end)
7294                         break;
7295
7296                 cond_resched();
7297         }
7298         ret = 0;
7299 out:
7300         btrfs_release_path(root, path);
7301         if (inode) {
7302                 mutex_unlock(&inode->i_mutex);
7303                 if (extent_locked) {
7304                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7305                                       lock_end, GFP_NOFS);
7306                 }
7307                 iput(inode);
7308         }
7309         return ret;
7310 }
7311
7312 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
7313                                struct btrfs_root *root,
7314                                struct extent_buffer *buf, u64 orig_start)
7315 {
7316         int level;
7317         int ret;
7318
7319         BUG_ON(btrfs_header_generation(buf) != trans->transid);
7320         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
7321
7322         level = btrfs_header_level(buf);
7323         if (level == 0) {
7324                 struct btrfs_leaf_ref *ref;
7325                 struct btrfs_leaf_ref *orig_ref;
7326
7327                 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
7328                 if (!orig_ref)
7329                         return -ENOENT;
7330
7331                 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
7332                 if (!ref) {
7333                         btrfs_free_leaf_ref(root, orig_ref);
7334                         return -ENOMEM;
7335                 }
7336
7337                 ref->nritems = orig_ref->nritems;
7338                 memcpy(ref->extents, orig_ref->extents,
7339                         sizeof(ref->extents[0]) * ref->nritems);
7340
7341                 btrfs_free_leaf_ref(root, orig_ref);
7342
7343                 ref->root_gen = trans->transid;
7344                 ref->bytenr = buf->start;
7345                 ref->owner = btrfs_header_owner(buf);
7346                 ref->generation = btrfs_header_generation(buf);
7347
7348                 ret = btrfs_add_leaf_ref(root, ref, 0);
7349                 WARN_ON(ret);
7350                 btrfs_free_leaf_ref(root, ref);
7351         }
7352         return 0;
7353 }
7354
7355 static noinline int invalidate_extent_cache(struct btrfs_root *root,
7356                                         struct extent_buffer *leaf,
7357                                         struct btrfs_block_group_cache *group,
7358                                         struct btrfs_root *target_root)
7359 {
7360         struct btrfs_key key;
7361         struct inode *inode = NULL;
7362         struct btrfs_file_extent_item *fi;
7363         struct extent_state *cached_state = NULL;
7364         u64 num_bytes;
7365         u64 skip_objectid = 0;
7366         u32 nritems;
7367         u32 i;
7368
7369         nritems = btrfs_header_nritems(leaf);
7370         for (i = 0; i < nritems; i++) {
7371                 btrfs_item_key_to_cpu(leaf, &key, i);
7372                 if (key.objectid == skip_objectid ||
7373                     key.type != BTRFS_EXTENT_DATA_KEY)
7374                         continue;
7375                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
7376                 if (btrfs_file_extent_type(leaf, fi) ==
7377                     BTRFS_FILE_EXTENT_INLINE)
7378                         continue;
7379                 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
7380                         continue;
7381                 if (!inode || inode->i_ino != key.objectid) {
7382                         iput(inode);
7383                         inode = btrfs_ilookup(target_root->fs_info->sb,
7384                                               key.objectid, target_root, 1);
7385                 }
7386                 if (!inode) {
7387                         skip_objectid = key.objectid;
7388                         continue;
7389                 }
7390                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
7391
7392                 lock_extent_bits(&BTRFS_I(inode)->io_tree, key.offset,
7393                                  key.offset + num_bytes - 1, 0, &cached_state,
7394                                  GFP_NOFS);
7395                 btrfs_drop_extent_cache(inode, key.offset,
7396                                         key.offset + num_bytes - 1, 1);
7397                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, key.offset,
7398                                      key.offset + num_bytes - 1, &cached_state,
7399                                      GFP_NOFS);
7400                 cond_resched();
7401         }
7402         iput(inode);
7403         return 0;
7404 }
7405
7406 static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans,
7407                                         struct btrfs_root *root,
7408                                         struct extent_buffer *leaf,
7409                                         struct btrfs_block_group_cache *group,
7410                                         struct inode *reloc_inode)
7411 {
7412         struct btrfs_key key;
7413         struct btrfs_key extent_key;
7414         struct btrfs_file_extent_item *fi;
7415         struct btrfs_leaf_ref *ref;
7416         struct disk_extent *new_extent;
7417         u64 bytenr;
7418         u64 num_bytes;
7419         u32 nritems;
7420         u32 i;
7421         int ext_index;
7422         int nr_extent;
7423         int ret;
7424
7425         new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
7426         BUG_ON(!new_extent);
7427
7428         ref = btrfs_lookup_leaf_ref(root, leaf->start);
7429         BUG_ON(!ref);
7430
7431         ext_index = -1;
7432         nritems = btrfs_header_nritems(leaf);
7433         for (i = 0; i < nritems; i++) {
7434                 btrfs_item_key_to_cpu(leaf, &key, i);
7435                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
7436                         continue;
7437                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
7438                 if (btrfs_file_extent_type(leaf, fi) ==
7439                     BTRFS_FILE_EXTENT_INLINE)
7440                         continue;
7441                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7442                 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
7443                 if (bytenr == 0)
7444                         continue;
7445
7446                 ext_index++;
7447                 if (bytenr >= group->key.objectid + group->key.offset ||
7448                     bytenr + num_bytes <= group->key.objectid)
7449                         continue;
7450
7451                 extent_key.objectid = bytenr;
7452                 extent_key.offset = num_bytes;
7453                 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
7454                 nr_extent = 1;
7455                 ret = get_new_locations(reloc_inode, &extent_key,
7456                                         group->key.objectid, 1,
7457                                         &new_extent, &nr_extent);
7458                 if (ret > 0)
7459                         continue;
7460                 BUG_ON(ret < 0);
7461
7462                 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
7463                 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
7464                 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
7465                 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
7466
7467                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
7468                                                 new_extent->disk_bytenr);
7469                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
7470                                                 new_extent->disk_num_bytes);
7471                 btrfs_mark_buffer_dirty(leaf);
7472
7473                 ret = btrfs_inc_extent_ref(trans, root,
7474                                         new_extent->disk_bytenr,
7475                                         new_extent->disk_num_bytes,
7476                                         leaf->start,
7477                                         root->root_key.objectid,
7478                                         trans->transid, key.objectid);
7479                 BUG_ON(ret);
7480
7481                 ret = btrfs_free_extent(trans, root,
7482                                         bytenr, num_bytes, leaf->start,
7483                                         btrfs_header_owner(leaf),
7484                                         btrfs_header_generation(leaf),
7485                                         key.objectid, 0);
7486                 BUG_ON(ret);
7487                 cond_resched();
7488         }
7489         kfree(new_extent);
7490         BUG_ON(ext_index + 1 != ref->nritems);
7491         btrfs_free_leaf_ref(root, ref);
7492         return 0;
7493 }
7494
7495 int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
7496                           struct btrfs_root *root)
7497 {
7498         struct btrfs_root *reloc_root;
7499         int ret;
7500
7501         if (root->reloc_root) {
7502                 reloc_root = root->reloc_root;
7503                 root->reloc_root = NULL;
7504                 list_add(&reloc_root->dead_list,
7505                          &root->fs_info->dead_reloc_roots);
7506
7507                 btrfs_set_root_bytenr(&reloc_root->root_item,
7508                                       reloc_root->node->start);
7509                 btrfs_set_root_level(&root->root_item,
7510                                      btrfs_header_level(reloc_root->node));
7511                 memset(&reloc_root->root_item.drop_progress, 0,
7512                         sizeof(struct btrfs_disk_key));
7513                 reloc_root->root_item.drop_level = 0;
7514
7515                 ret = btrfs_update_root(trans, root->fs_info->tree_root,
7516                                         &reloc_root->root_key,
7517                                         &reloc_root->root_item);
7518                 BUG_ON(ret);
7519         }
7520         return 0;
7521 }
7522
7523 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
7524 {
7525         struct btrfs_trans_handle *trans;
7526         struct btrfs_root *reloc_root;
7527         struct btrfs_root *prev_root = NULL;
7528         struct list_head dead_roots;
7529         int ret;
7530         unsigned long nr;
7531
7532         INIT_LIST_HEAD(&dead_roots);
7533         list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
7534
7535         while (!list_empty(&dead_roots)) {
7536                 reloc_root = list_entry(dead_roots.prev,
7537                                         struct btrfs_root, dead_list);
7538                 list_del_init(&reloc_root->dead_list);
7539
7540                 BUG_ON(reloc_root->commit_root != NULL);
7541                 while (1) {
7542                         trans = btrfs_join_transaction(root, 1);
7543                         BUG_ON(IS_ERR(trans));
7544
7545                         mutex_lock(&root->fs_info->drop_mutex);
7546                         ret = btrfs_drop_snapshot(trans, reloc_root);
7547                         if (ret != -EAGAIN)
7548                                 break;
7549                         mutex_unlock(&root->fs_info->drop_mutex);
7550
7551                         nr = trans->blocks_used;
7552                         ret = btrfs_end_transaction(trans, root);
7553                         BUG_ON(ret);
7554                         btrfs_btree_balance_dirty(root, nr);
7555                 }
7556
7557                 free_extent_buffer(reloc_root->node);
7558
7559                 ret = btrfs_del_root(trans, root->fs_info->tree_root,
7560                                      &reloc_root->root_key);
7561                 BUG_ON(ret);
7562                 mutex_unlock(&root->fs_info->drop_mutex);
7563
7564                 nr = trans->blocks_used;
7565                 ret = btrfs_end_transaction(trans, root);
7566                 BUG_ON(ret);
7567                 btrfs_btree_balance_dirty(root, nr);
7568
7569                 kfree(prev_root);
7570                 prev_root = reloc_root;
7571         }
7572         if (prev_root) {
7573                 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
7574                 kfree(prev_root);
7575         }
7576         return 0;
7577 }
7578
7579 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
7580 {
7581         list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
7582         return 0;
7583 }
7584
7585 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
7586 {
7587         struct btrfs_root *reloc_root;
7588         struct btrfs_trans_handle *trans;
7589         struct btrfs_key location;
7590         int found;
7591         int ret;
7592
7593         mutex_lock(&root->fs_info->tree_reloc_mutex);
7594         ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
7595         BUG_ON(ret);
7596         found = !list_empty(&root->fs_info->dead_reloc_roots);
7597         mutex_unlock(&root->fs_info->tree_reloc_mutex);
7598
7599         if (found) {
7600                 trans = btrfs_start_transaction(root, 1);
7601                 BUG_ON(IS_ERR(trans));
7602                 ret = btrfs_commit_transaction(trans, root);
7603                 BUG_ON(ret);
7604         }
7605
7606         location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
7607         location.offset = (u64)-1;
7608         location.type = BTRFS_ROOT_ITEM_KEY;
7609
7610         reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
7611         BUG_ON(!reloc_root);
7612         btrfs_orphan_cleanup(reloc_root);
7613         return 0;
7614 }
7615
7616 static noinline int init_reloc_tree(struct btrfs_trans_handle *trans,
7617                                     struct btrfs_root *root)
7618 {
7619         struct btrfs_root *reloc_root;
7620         struct extent_buffer *eb;
7621         struct btrfs_root_item *root_item;
7622         struct btrfs_key root_key;
7623         int ret;
7624
7625         BUG_ON(!root->ref_cows);
7626         if (root->reloc_root)
7627                 return 0;
7628
7629         root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
7630         BUG_ON(!root_item);
7631
7632         ret = btrfs_copy_root(trans, root, root->commit_root,
7633                               &eb, BTRFS_TREE_RELOC_OBJECTID);
7634         BUG_ON(ret);
7635
7636         root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
7637         root_key.offset = root->root_key.objectid;
7638         root_key.type = BTRFS_ROOT_ITEM_KEY;
7639
7640         memcpy(root_item, &root->root_item, sizeof(root_item));
7641         btrfs_set_root_refs(root_item, 0);
7642         btrfs_set_root_bytenr(root_item, eb->start);
7643         btrfs_set_root_level(root_item, btrfs_header_level(eb));
7644         btrfs_set_root_generation(root_item, trans->transid);
7645
7646         btrfs_tree_unlock(eb);
7647         free_extent_buffer(eb);
7648
7649         ret = btrfs_insert_root(trans, root->fs_info->tree_root,
7650                                 &root_key, root_item);
7651         BUG_ON(ret);
7652         kfree(root_item);
7653
7654         reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
7655                                                  &root_key);
7656         BUG_ON(!reloc_root);
7657         reloc_root->last_trans = trans->transid;
7658         reloc_root->commit_root = NULL;
7659         reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
7660
7661         root->reloc_root = reloc_root;
7662         return 0;
7663 }
7664
7665 /*
7666  * Core function of space balance.
7667  *
7668  * The idea is using reloc trees to relocate tree blocks in reference
7669  * counted roots. There is one reloc tree for each subvol, and all
7670  * reloc trees share same root key objectid. Reloc trees are snapshots
7671  * of the latest committed roots of subvols (root->commit_root).
7672  *
7673  * To relocate a tree block referenced by a subvol, there are two steps.
7674  * COW the block through subvol's reloc tree, then update block pointer
7675  * in the subvol to point to the new block. Since all reloc trees share
7676  * same root key objectid, doing special handing for tree blocks owned
7677  * by them is easy. Once a tree block has been COWed in one reloc tree,
7678  * we can use the resulting new block directly when the same block is
7679  * required to COW again through other reloc trees. By this way, relocated
7680  * tree blocks are shared between reloc trees, so they are also shared
7681  * between subvols.
7682  */
7683 static noinline int relocate_one_path(struct btrfs_trans_handle *trans,
7684                                       struct btrfs_root *root,
7685                                       struct btrfs_path *path,
7686                                       struct btrfs_key *first_key,
7687                                       struct btrfs_ref_path *ref_path,
7688                                       struct btrfs_block_group_cache *group,
7689                                       struct inode *reloc_inode)
7690 {
7691         struct btrfs_root *reloc_root;
7692         struct extent_buffer *eb = NULL;
7693         struct btrfs_key *keys;
7694         u64 *nodes;
7695         int level;
7696         int shared_level;
7697         int lowest_level = 0;
7698         int ret;
7699
7700         if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
7701                 lowest_level = ref_path->owner_objectid;
7702
7703         if (!root->ref_cows) {
7704                 path->lowest_level = lowest_level;
7705                 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
7706                 BUG_ON(ret < 0);
7707                 path->lowest_level = 0;
7708                 btrfs_release_path(root, path);
7709                 return 0;
7710         }
7711
7712         mutex_lock(&root->fs_info->tree_reloc_mutex);
7713         ret = init_reloc_tree(trans, root);
7714         BUG_ON(ret);
7715         reloc_root = root->reloc_root;
7716
7717         shared_level = ref_path->shared_level;
7718         ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
7719
7720         keys = ref_path->node_keys;
7721         nodes = ref_path->new_nodes;
7722         memset(&keys[shared_level + 1], 0,
7723                sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
7724         memset(&nodes[shared_level + 1], 0,
7725                sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
7726
7727         if (nodes[lowest_level] == 0) {
7728                 path->lowest_level = lowest_level;
7729                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
7730                                         0, 1);
7731                 BUG_ON(ret);
7732                 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
7733                         eb = path->nodes[level];
7734                         if (!eb || eb == reloc_root->node)
7735                                 break;
7736                         nodes[level] = eb->start;
7737                         if (level == 0)
7738                                 btrfs_item_key_to_cpu(eb, &keys[level], 0);
7739                         else
7740                                 btrfs_node_key_to_cpu(eb, &keys[level], 0);
7741                 }
7742                 if (nodes[0] &&
7743                     ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7744                         eb = path->nodes[0];
7745                         ret = replace_extents_in_leaf(trans, reloc_root, eb,
7746                                                       group, reloc_inode);
7747                         BUG_ON(ret);
7748                 }
7749                 btrfs_release_path(reloc_root, path);
7750         } else {
7751                 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
7752                                        lowest_level);
7753                 BUG_ON(ret);
7754         }
7755
7756         /*
7757          * replace tree blocks in the fs tree with tree blocks in
7758          * the reloc tree.
7759          */
7760         ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
7761         BUG_ON(ret < 0);
7762
7763         if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7764                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
7765                                         0, 0);
7766                 BUG_ON(ret);
7767                 extent_buffer_get(path->nodes[0]);
7768                 eb = path->nodes[0];
7769                 btrfs_release_path(reloc_root, path);
7770                 ret = invalidate_extent_cache(reloc_root, eb, group, root);
7771                 BUG_ON(ret);
7772                 free_extent_buffer(eb);
7773         }
7774
7775         mutex_unlock(&root->fs_info->tree_reloc_mutex);
7776         path->lowest_level = 0;
7777         return 0;
7778 }
7779
7780 static noinline int relocate_tree_block(struct btrfs_trans_handle *trans,
7781                                         struct btrfs_root *root,
7782                                         struct btrfs_path *path,
7783                                         struct btrfs_key *first_key,
7784                                         struct btrfs_ref_path *ref_path)
7785 {
7786         int ret;
7787
7788         ret = relocate_one_path(trans, root, path, first_key,
7789                                 ref_path, NULL, NULL);
7790         BUG_ON(ret);
7791
7792         return 0;
7793 }
7794
7795 static noinline int del_extent_zero(struct btrfs_trans_handle *trans,
7796                                     struct btrfs_root *extent_root,
7797                                     struct btrfs_path *path,
7798                                     struct btrfs_key *extent_key)
7799 {
7800         int ret;
7801
7802         ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
7803         if (ret)
7804                 goto out;
7805         ret = btrfs_del_item(trans, extent_root, path);
7806 out:
7807         btrfs_release_path(extent_root, path);
7808         return ret;
7809 }
7810
7811 static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info,
7812                                                 struct btrfs_ref_path *ref_path)
7813 {
7814         struct btrfs_key root_key;
7815
7816         root_key.objectid = ref_path->root_objectid;
7817         root_key.type = BTRFS_ROOT_ITEM_KEY;
7818         if (is_cowonly_root(ref_path->root_objectid))
7819                 root_key.offset = 0;
7820         else
7821                 root_key.offset = (u64)-1;
7822
7823         return btrfs_read_fs_root_no_name(fs_info, &root_key);
7824 }
7825
7826 static noinline int relocate_one_extent(struct btrfs_root *extent_root,
7827                                         struct btrfs_path *path,
7828                                         struct btrfs_key *extent_key,
7829                                         struct btrfs_block_group_cache *group,
7830                                         struct inode *reloc_inode, int pass)
7831 {
7832         struct btrfs_trans_handle *trans;
7833         struct btrfs_root *found_root;
7834         struct btrfs_ref_path *ref_path = NULL;
7835         struct disk_extent *new_extents = NULL;
7836         int nr_extents = 0;
7837         int loops;
7838         int ret;
7839         int level;
7840         struct btrfs_key first_key;
7841         u64 prev_block = 0;
7842
7843
7844         trans = btrfs_start_transaction(extent_root, 1);
7845         BUG_ON(IS_ERR(trans));
7846
7847         if (extent_key->objectid == 0) {
7848                 ret = del_extent_zero(trans, extent_root, path, extent_key);
7849                 goto out;
7850         }
7851
7852         ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
7853         if (!ref_path) {
7854                 ret = -ENOMEM;
7855                 goto out;
7856         }
7857
7858         for (loops = 0; ; loops++) {
7859                 if (loops == 0) {
7860                         ret = btrfs_first_ref_path(trans, extent_root, ref_path,
7861                                                    extent_key->objectid);
7862                 } else {
7863                         ret = btrfs_next_ref_path(trans, extent_root, ref_path);
7864                 }
7865                 if (ret < 0)
7866                         goto out;
7867                 if (ret > 0)
7868                         break;
7869
7870                 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
7871                     ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
7872                         continue;
7873
7874                 found_root = read_ref_root(extent_root->fs_info, ref_path);
7875                 BUG_ON(!found_root);
7876                 /*
7877                  * for reference counted tree, only process reference paths
7878                  * rooted at the latest committed root.
7879                  */
7880                 if (found_root->ref_cows &&
7881                     ref_path->root_generation != found_root->root_key.offset)
7882                         continue;
7883
7884                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7885                         if (pass == 0) {
7886                                 /*
7887                                  * copy data extents to new locations
7888                                  */
7889                                 u64 group_start = group->key.objectid;
7890                                 ret = relocate_data_extent(reloc_inode,
7891                                                            extent_key,
7892                                                            group_start);
7893                                 if (ret < 0)
7894                                         goto out;
7895                                 break;
7896                         }
7897                         level = 0;
7898                 } else {
7899                         level = ref_path->owner_objectid;
7900                 }
7901
7902                 if (prev_block != ref_path->nodes[level]) {
7903                         struct extent_buffer *eb;
7904                         u64 block_start = ref_path->nodes[level];
7905                         u64 block_size = btrfs_level_size(found_root, level);
7906
7907                         eb = read_tree_block(found_root, block_start,
7908                                              block_size, 0);
7909                         btrfs_tree_lock(eb);
7910                         BUG_ON(level != btrfs_header_level(eb));
7911
7912                         if (level == 0)
7913                                 btrfs_item_key_to_cpu(eb, &first_key, 0);
7914                         else
7915                                 btrfs_node_key_to_cpu(eb, &first_key, 0);
7916
7917                         btrfs_tree_unlock(eb);
7918                         free_extent_buffer(eb);
7919                         prev_block = block_start;
7920                 }
7921
7922                 mutex_lock(&extent_root->fs_info->trans_mutex);
7923                 btrfs_record_root_in_trans(found_root);
7924                 mutex_unlock(&extent_root->fs_info->trans_mutex);
7925                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7926                         /*
7927                          * try to update data extent references while
7928                          * keeping metadata shared between snapshots.
7929                          */
7930                         if (pass == 1) {
7931                                 ret = relocate_one_path(trans, found_root,
7932                                                 path, &first_key, ref_path,
7933                                                 group, reloc_inode);
7934                                 if (ret < 0)
7935                                         goto out;
7936                                 continue;
7937                         }
7938                         /*
7939                          * use fallback method to process the remaining
7940                          * references.
7941                          */
7942                         if (!new_extents) {
7943                                 u64 group_start = group->key.objectid;
7944                                 new_extents = kmalloc(sizeof(*new_extents),
7945                                                       GFP_NOFS);
7946                                 nr_extents = 1;
7947                                 ret = get_new_locations(reloc_inode,
7948                                                         extent_key,
7949                                                         group_start, 1,
7950                                                         &new_extents,
7951                                                         &nr_extents);
7952                                 if (ret)
7953                                         goto out;
7954                         }
7955                         ret = replace_one_extent(trans, found_root,
7956                                                 path, extent_key,
7957                                                 &first_key, ref_path,
7958                                                 new_extents, nr_extents);
7959                 } else {
7960                         ret = relocate_tree_block(trans, found_root, path,
7961                                                   &first_key, ref_path);
7962                 }
7963                 if (ret < 0)
7964                         goto out;
7965         }
7966         ret = 0;
7967 out:
7968         btrfs_end_transaction(trans, extent_root);
7969         kfree(new_extents);
7970         kfree(ref_path);
7971         return ret;
7972 }
7973 #endif
7974
7975 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7976 {
7977         u64 num_devices;
7978         u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
7979                 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7980
7981         /*
7982          * we add in the count of missing devices because we want
7983          * to make sure that any RAID levels on a degraded FS
7984          * continue to be honored.
7985          */
7986         num_devices = root->fs_info->fs_devices->rw_devices +
7987                 root->fs_info->fs_devices->missing_devices;
7988
7989         if (num_devices == 1) {
7990                 stripped |= BTRFS_BLOCK_GROUP_DUP;
7991                 stripped = flags & ~stripped;
7992
7993                 /* turn raid0 into single device chunks */
7994                 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7995                         return stripped;
7996
7997                 /* turn mirroring into duplication */
7998                 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7999                              BTRFS_BLOCK_GROUP_RAID10))
8000                         return stripped | BTRFS_BLOCK_GROUP_DUP;
8001                 return flags;
8002         } else {
8003                 /* they already had raid on here, just return */
8004                 if (flags & stripped)
8005                         return flags;
8006
8007                 stripped |= BTRFS_BLOCK_GROUP_DUP;
8008                 stripped = flags & ~stripped;
8009
8010                 /* switch duplicated blocks with raid1 */
8011                 if (flags & BTRFS_BLOCK_GROUP_DUP)
8012                         return stripped | BTRFS_BLOCK_GROUP_RAID1;
8013
8014                 /* turn single device chunks into raid0 */
8015                 return stripped | BTRFS_BLOCK_GROUP_RAID0;
8016         }
8017         return flags;
8018 }
8019
8020 static int set_block_group_ro(struct btrfs_block_group_cache *cache)
8021 {
8022         struct btrfs_space_info *sinfo = cache->space_info;
8023         u64 num_bytes;
8024         int ret = -ENOSPC;
8025
8026         if (cache->ro)
8027                 return 0;
8028
8029         spin_lock(&sinfo->lock);
8030         spin_lock(&cache->lock);
8031         num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8032                     cache->bytes_super - btrfs_block_group_used(&cache->item);
8033
8034         if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
8035             sinfo->bytes_may_use + sinfo->bytes_readonly +
8036             cache->reserved_pinned + num_bytes <= sinfo->total_bytes) {
8037                 sinfo->bytes_readonly += num_bytes;
8038                 sinfo->bytes_reserved += cache->reserved_pinned;
8039                 cache->reserved_pinned = 0;
8040                 cache->ro = 1;
8041                 ret = 0;
8042         }
8043
8044         spin_unlock(&cache->lock);
8045         spin_unlock(&sinfo->lock);
8046         return ret;
8047 }
8048
8049 int btrfs_set_block_group_ro(struct btrfs_root *root,
8050                              struct btrfs_block_group_cache *cache)
8051
8052 {
8053         struct btrfs_trans_handle *trans;
8054         u64 alloc_flags;
8055         int ret;
8056
8057         BUG_ON(cache->ro);
8058
8059         trans = btrfs_join_transaction(root, 1);
8060         BUG_ON(IS_ERR(trans));
8061
8062         alloc_flags = update_block_group_flags(root, cache->flags);
8063         if (alloc_flags != cache->flags)
8064                 do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, 1);
8065
8066         ret = set_block_group_ro(cache);
8067         if (!ret)
8068                 goto out;
8069         alloc_flags = get_alloc_profile(root, cache->space_info->flags);
8070         ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, 1);
8071         if (ret < 0)
8072                 goto out;
8073         ret = set_block_group_ro(cache);
8074 out:
8075         btrfs_end_transaction(trans, root);
8076         return ret;
8077 }
8078
8079 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
8080                             struct btrfs_root *root, u64 type)
8081 {
8082         u64 alloc_flags = get_alloc_profile(root, type);
8083         return do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, 1);
8084 }
8085
8086 /*
8087  * helper to account the unused space of all the readonly block group in the
8088  * list. takes mirrors into account.
8089  */
8090 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
8091 {
8092         struct btrfs_block_group_cache *block_group;
8093         u64 free_bytes = 0;
8094         int factor;
8095
8096         list_for_each_entry(block_group, groups_list, list) {
8097                 spin_lock(&block_group->lock);
8098
8099                 if (!block_group->ro) {
8100                         spin_unlock(&block_group->lock);
8101                         continue;
8102                 }
8103
8104                 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
8105                                           BTRFS_BLOCK_GROUP_RAID10 |
8106                                           BTRFS_BLOCK_GROUP_DUP))
8107                         factor = 2;
8108                 else
8109                         factor = 1;
8110
8111                 free_bytes += (block_group->key.offset -
8112                                btrfs_block_group_used(&block_group->item)) *
8113                                factor;
8114
8115                 spin_unlock(&block_group->lock);
8116         }
8117
8118         return free_bytes;
8119 }
8120
8121 /*
8122  * helper to account the unused space of all the readonly block group in the
8123  * space_info. takes mirrors into account.
8124  */
8125 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
8126 {
8127         int i;
8128         u64 free_bytes = 0;
8129
8130         spin_lock(&sinfo->lock);
8131
8132         for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
8133                 if (!list_empty(&sinfo->block_groups[i]))
8134                         free_bytes += __btrfs_get_ro_block_group_free_space(
8135                                                 &sinfo->block_groups[i]);
8136
8137         spin_unlock(&sinfo->lock);
8138
8139         return free_bytes;
8140 }
8141
8142 int btrfs_set_block_group_rw(struct btrfs_root *root,
8143                               struct btrfs_block_group_cache *cache)
8144 {
8145         struct btrfs_space_info *sinfo = cache->space_info;
8146         u64 num_bytes;
8147
8148         BUG_ON(!cache->ro);
8149
8150         spin_lock(&sinfo->lock);
8151         spin_lock(&cache->lock);
8152         num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8153                     cache->bytes_super - btrfs_block_group_used(&cache->item);
8154         sinfo->bytes_readonly -= num_bytes;
8155         cache->ro = 0;
8156         spin_unlock(&cache->lock);
8157         spin_unlock(&sinfo->lock);
8158         return 0;
8159 }
8160
8161 /*
8162  * checks to see if its even possible to relocate this block group.
8163  *
8164  * @return - -1 if it's not a good idea to relocate this block group, 0 if its
8165  * ok to go ahead and try.
8166  */
8167 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
8168 {
8169         struct btrfs_block_group_cache *block_group;
8170         struct btrfs_space_info *space_info;
8171         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
8172         struct btrfs_device *device;
8173         int full = 0;
8174         int ret = 0;
8175
8176         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
8177
8178         /* odd, couldn't find the block group, leave it alone */
8179         if (!block_group)
8180                 return -1;
8181
8182         /* no bytes used, we're good */
8183         if (!btrfs_block_group_used(&block_group->item))
8184                 goto out;
8185
8186         space_info = block_group->space_info;
8187         spin_lock(&space_info->lock);
8188
8189         full = space_info->full;
8190
8191         /*
8192          * if this is the last block group we have in this space, we can't
8193          * relocate it unless we're able to allocate a new chunk below.
8194          *
8195          * Otherwise, we need to make sure we have room in the space to handle
8196          * all of the extents from this block group.  If we can, we're good
8197          */
8198         if ((space_info->total_bytes != block_group->key.offset) &&
8199            (space_info->bytes_used + space_info->bytes_reserved +
8200             space_info->bytes_pinned + space_info->bytes_readonly +
8201             btrfs_block_group_used(&block_group->item) <
8202             space_info->total_bytes)) {
8203                 spin_unlock(&space_info->lock);
8204                 goto out;
8205         }
8206         spin_unlock(&space_info->lock);
8207
8208         /*
8209          * ok we don't have enough space, but maybe we have free space on our
8210          * devices to allocate new chunks for relocation, so loop through our
8211          * alloc devices and guess if we have enough space.  However, if we
8212          * were marked as full, then we know there aren't enough chunks, and we
8213          * can just return.
8214          */
8215         ret = -1;
8216         if (full)
8217                 goto out;
8218
8219         mutex_lock(&root->fs_info->chunk_mutex);
8220         list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
8221                 u64 min_free = btrfs_block_group_used(&block_group->item);
8222                 u64 dev_offset;
8223
8224                 /*
8225                  * check to make sure we can actually find a chunk with enough
8226                  * space to fit our block group in.
8227                  */
8228                 if (device->total_bytes > device->bytes_used + min_free) {
8229                         ret = find_free_dev_extent(NULL, device, min_free,
8230                                                    &dev_offset, NULL);
8231                         if (!ret)
8232                                 break;
8233                         ret = -1;
8234                 }
8235         }
8236         mutex_unlock(&root->fs_info->chunk_mutex);
8237 out:
8238         btrfs_put_block_group(block_group);
8239         return ret;
8240 }
8241
8242 static int find_first_block_group(struct btrfs_root *root,
8243                 struct btrfs_path *path, struct btrfs_key *key)
8244 {
8245         int ret = 0;
8246         struct btrfs_key found_key;
8247         struct extent_buffer *leaf;
8248         int slot;
8249
8250         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
8251         if (ret < 0)
8252                 goto out;
8253
8254         while (1) {
8255                 slot = path->slots[0];
8256                 leaf = path->nodes[0];
8257                 if (slot >= btrfs_header_nritems(leaf)) {
8258                         ret = btrfs_next_leaf(root, path);
8259                         if (ret == 0)
8260                                 continue;
8261                         if (ret < 0)
8262                                 goto out;
8263                         break;
8264                 }
8265                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
8266
8267                 if (found_key.objectid >= key->objectid &&
8268                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
8269                         ret = 0;
8270                         goto out;
8271                 }
8272                 path->slots[0]++;
8273         }
8274 out:
8275         return ret;
8276 }
8277
8278 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
8279 {
8280         struct btrfs_block_group_cache *block_group;
8281         u64 last = 0;
8282
8283         while (1) {
8284                 struct inode *inode;
8285
8286                 block_group = btrfs_lookup_first_block_group(info, last);
8287                 while (block_group) {
8288                         spin_lock(&block_group->lock);
8289                         if (block_group->iref)
8290                                 break;
8291                         spin_unlock(&block_group->lock);
8292                         block_group = next_block_group(info->tree_root,
8293                                                        block_group);
8294                 }
8295                 if (!block_group) {
8296                         if (last == 0)
8297                                 break;
8298                         last = 0;
8299                         continue;
8300                 }
8301
8302                 inode = block_group->inode;
8303                 block_group->iref = 0;
8304                 block_group->inode = NULL;
8305                 spin_unlock(&block_group->lock);
8306                 iput(inode);
8307                 last = block_group->key.objectid + block_group->key.offset;
8308                 btrfs_put_block_group(block_group);
8309         }
8310 }
8311
8312 int btrfs_free_block_groups(struct btrfs_fs_info *info)
8313 {
8314         struct btrfs_block_group_cache *block_group;
8315         struct btrfs_space_info *space_info;
8316         struct btrfs_caching_control *caching_ctl;
8317         struct rb_node *n;
8318
8319         down_write(&info->extent_commit_sem);
8320         while (!list_empty(&info->caching_block_groups)) {
8321                 caching_ctl = list_entry(info->caching_block_groups.next,
8322                                          struct btrfs_caching_control, list);
8323                 list_del(&caching_ctl->list);
8324                 put_caching_control(caching_ctl);
8325         }
8326         up_write(&info->extent_commit_sem);
8327
8328         spin_lock(&info->block_group_cache_lock);
8329         while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
8330                 block_group = rb_entry(n, struct btrfs_block_group_cache,
8331                                        cache_node);
8332                 rb_erase(&block_group->cache_node,
8333                          &info->block_group_cache_tree);
8334                 spin_unlock(&info->block_group_cache_lock);
8335
8336                 down_write(&block_group->space_info->groups_sem);
8337                 list_del(&block_group->list);
8338                 up_write(&block_group->space_info->groups_sem);
8339
8340                 if (block_group->cached == BTRFS_CACHE_STARTED)
8341                         wait_block_group_cache_done(block_group);
8342
8343                 /*
8344                  * We haven't cached this block group, which means we could
8345                  * possibly have excluded extents on this block group.
8346                  */
8347                 if (block_group->cached == BTRFS_CACHE_NO)
8348                         free_excluded_extents(info->extent_root, block_group);
8349
8350                 btrfs_remove_free_space_cache(block_group);
8351                 btrfs_put_block_group(block_group);
8352
8353                 spin_lock(&info->block_group_cache_lock);
8354         }
8355         spin_unlock(&info->block_group_cache_lock);
8356
8357         /* now that all the block groups are freed, go through and
8358          * free all the space_info structs.  This is only called during
8359          * the final stages of unmount, and so we know nobody is
8360          * using them.  We call synchronize_rcu() once before we start,
8361          * just to be on the safe side.
8362          */
8363         synchronize_rcu();
8364
8365         release_global_block_rsv(info);
8366
8367         while(!list_empty(&info->space_info)) {
8368                 space_info = list_entry(info->space_info.next,
8369                                         struct btrfs_space_info,
8370                                         list);
8371                 if (space_info->bytes_pinned > 0 ||
8372                     space_info->bytes_reserved > 0) {
8373                         WARN_ON(1);
8374                         dump_space_info(space_info, 0, 0);
8375                 }
8376                 list_del(&space_info->list);
8377                 kfree(space_info);
8378         }
8379         return 0;
8380 }
8381
8382 static void __link_block_group(struct btrfs_space_info *space_info,
8383                                struct btrfs_block_group_cache *cache)
8384 {
8385         int index = get_block_group_index(cache);
8386
8387         down_write(&space_info->groups_sem);
8388         list_add_tail(&cache->list, &space_info->block_groups[index]);
8389         up_write(&space_info->groups_sem);
8390 }
8391
8392 int btrfs_read_block_groups(struct btrfs_root *root)
8393 {
8394         struct btrfs_path *path;
8395         int ret;
8396         struct btrfs_block_group_cache *cache;
8397         struct btrfs_fs_info *info = root->fs_info;
8398         struct btrfs_space_info *space_info;
8399         struct btrfs_key key;
8400         struct btrfs_key found_key;
8401         struct extent_buffer *leaf;
8402         int need_clear = 0;
8403         u64 cache_gen;
8404
8405         root = info->extent_root;
8406         key.objectid = 0;
8407         key.offset = 0;
8408         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
8409         path = btrfs_alloc_path();
8410         if (!path)
8411                 return -ENOMEM;
8412
8413         cache_gen = btrfs_super_cache_generation(&root->fs_info->super_copy);
8414         if (cache_gen != 0 &&
8415             btrfs_super_generation(&root->fs_info->super_copy) != cache_gen)
8416                 need_clear = 1;
8417         if (btrfs_test_opt(root, CLEAR_CACHE))
8418                 need_clear = 1;
8419         if (!btrfs_test_opt(root, SPACE_CACHE) && cache_gen)
8420                 printk(KERN_INFO "btrfs: disk space caching is enabled\n");
8421
8422         while (1) {
8423                 ret = find_first_block_group(root, path, &key);
8424                 if (ret > 0)
8425                         break;
8426                 if (ret != 0)
8427                         goto error;
8428                 leaf = path->nodes[0];
8429                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
8430                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8431                 if (!cache) {
8432                         ret = -ENOMEM;
8433                         goto error;
8434                 }
8435
8436                 atomic_set(&cache->count, 1);
8437                 spin_lock_init(&cache->lock);
8438                 spin_lock_init(&cache->tree_lock);
8439                 cache->fs_info = info;
8440                 INIT_LIST_HEAD(&cache->list);
8441                 INIT_LIST_HEAD(&cache->cluster_list);
8442
8443                 if (need_clear)
8444                         cache->disk_cache_state = BTRFS_DC_CLEAR;
8445
8446                 /*
8447                  * we only want to have 32k of ram per block group for keeping
8448                  * track of free space, and if we pass 1/2 of that we want to
8449                  * start converting things over to using bitmaps
8450                  */
8451                 cache->extents_thresh = ((1024 * 32) / 2) /
8452                         sizeof(struct btrfs_free_space);
8453
8454                 read_extent_buffer(leaf, &cache->item,
8455                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
8456                                    sizeof(cache->item));
8457                 memcpy(&cache->key, &found_key, sizeof(found_key));
8458
8459                 key.objectid = found_key.objectid + found_key.offset;
8460                 btrfs_release_path(root, path);
8461                 cache->flags = btrfs_block_group_flags(&cache->item);
8462                 cache->sectorsize = root->sectorsize;
8463
8464                 /*
8465                  * We need to exclude the super stripes now so that the space
8466                  * info has super bytes accounted for, otherwise we'll think
8467                  * we have more space than we actually do.
8468                  */
8469                 exclude_super_stripes(root, cache);
8470
8471                 /*
8472                  * check for two cases, either we are full, and therefore
8473                  * don't need to bother with the caching work since we won't
8474                  * find any space, or we are empty, and we can just add all
8475                  * the space in and be done with it.  This saves us _alot_ of
8476                  * time, particularly in the full case.
8477                  */
8478                 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
8479                         cache->last_byte_to_unpin = (u64)-1;
8480                         cache->cached = BTRFS_CACHE_FINISHED;
8481                         free_excluded_extents(root, cache);
8482                 } else if (btrfs_block_group_used(&cache->item) == 0) {
8483                         cache->last_byte_to_unpin = (u64)-1;
8484                         cache->cached = BTRFS_CACHE_FINISHED;
8485                         add_new_free_space(cache, root->fs_info,
8486                                            found_key.objectid,
8487                                            found_key.objectid +
8488                                            found_key.offset);
8489                         free_excluded_extents(root, cache);
8490                 }
8491
8492                 ret = update_space_info(info, cache->flags, found_key.offset,
8493                                         btrfs_block_group_used(&cache->item),
8494                                         &space_info);
8495                 BUG_ON(ret);
8496                 cache->space_info = space_info;
8497                 spin_lock(&cache->space_info->lock);
8498                 cache->space_info->bytes_readonly += cache->bytes_super;
8499                 spin_unlock(&cache->space_info->lock);
8500
8501                 __link_block_group(space_info, cache);
8502
8503                 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8504                 BUG_ON(ret);
8505
8506                 set_avail_alloc_bits(root->fs_info, cache->flags);
8507                 if (btrfs_chunk_readonly(root, cache->key.objectid))
8508                         set_block_group_ro(cache);
8509         }
8510
8511         list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
8512                 if (!(get_alloc_profile(root, space_info->flags) &
8513                       (BTRFS_BLOCK_GROUP_RAID10 |
8514                        BTRFS_BLOCK_GROUP_RAID1 |
8515                        BTRFS_BLOCK_GROUP_DUP)))
8516                         continue;
8517                 /*
8518                  * avoid allocating from un-mirrored block group if there are
8519                  * mirrored block groups.
8520                  */
8521                 list_for_each_entry(cache, &space_info->block_groups[3], list)
8522                         set_block_group_ro(cache);
8523                 list_for_each_entry(cache, &space_info->block_groups[4], list)
8524                         set_block_group_ro(cache);
8525         }
8526
8527         init_global_block_rsv(info);
8528         ret = 0;
8529 error:
8530         btrfs_free_path(path);
8531         return ret;
8532 }
8533
8534 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
8535                            struct btrfs_root *root, u64 bytes_used,
8536                            u64 type, u64 chunk_objectid, u64 chunk_offset,
8537                            u64 size)
8538 {
8539         int ret;
8540         struct btrfs_root *extent_root;
8541         struct btrfs_block_group_cache *cache;
8542
8543         extent_root = root->fs_info->extent_root;
8544
8545         root->fs_info->last_trans_log_full_commit = trans->transid;
8546
8547         cache = kzalloc(sizeof(*cache), GFP_NOFS);
8548         if (!cache)
8549                 return -ENOMEM;
8550
8551         cache->key.objectid = chunk_offset;
8552         cache->key.offset = size;
8553         cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
8554         cache->sectorsize = root->sectorsize;
8555         cache->fs_info = root->fs_info;
8556
8557         /*
8558          * we only want to have 32k of ram per block group for keeping track
8559          * of free space, and if we pass 1/2 of that we want to start
8560          * converting things over to using bitmaps
8561          */
8562         cache->extents_thresh = ((1024 * 32) / 2) /
8563                 sizeof(struct btrfs_free_space);
8564         atomic_set(&cache->count, 1);
8565         spin_lock_init(&cache->lock);
8566         spin_lock_init(&cache->tree_lock);
8567         INIT_LIST_HEAD(&cache->list);
8568         INIT_LIST_HEAD(&cache->cluster_list);
8569
8570         btrfs_set_block_group_used(&cache->item, bytes_used);
8571         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
8572         cache->flags = type;
8573         btrfs_set_block_group_flags(&cache->item, type);
8574
8575         cache->last_byte_to_unpin = (u64)-1;
8576         cache->cached = BTRFS_CACHE_FINISHED;
8577         exclude_super_stripes(root, cache);
8578
8579         add_new_free_space(cache, root->fs_info, chunk_offset,
8580                            chunk_offset + size);
8581
8582         free_excluded_extents(root, cache);
8583
8584         ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
8585                                 &cache->space_info);
8586         BUG_ON(ret);
8587
8588         spin_lock(&cache->space_info->lock);
8589         cache->space_info->bytes_readonly += cache->bytes_super;
8590         spin_unlock(&cache->space_info->lock);
8591
8592         __link_block_group(cache->space_info, cache);
8593
8594         ret = btrfs_add_block_group_cache(root->fs_info, cache);
8595         BUG_ON(ret);
8596
8597         ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
8598                                 sizeof(cache->item));
8599         BUG_ON(ret);
8600
8601         set_avail_alloc_bits(extent_root->fs_info, type);
8602
8603         return 0;
8604 }
8605
8606 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
8607                              struct btrfs_root *root, u64 group_start)
8608 {
8609         struct btrfs_path *path;
8610         struct btrfs_block_group_cache *block_group;
8611         struct btrfs_free_cluster *cluster;
8612         struct btrfs_root *tree_root = root->fs_info->tree_root;
8613         struct btrfs_key key;
8614         struct inode *inode;
8615         int ret;
8616         int factor;
8617
8618         root = root->fs_info->extent_root;
8619
8620         block_group = btrfs_lookup_block_group(root->fs_info, group_start);
8621         BUG_ON(!block_group);
8622         BUG_ON(!block_group->ro);
8623
8624         memcpy(&key, &block_group->key, sizeof(key));
8625         if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
8626                                   BTRFS_BLOCK_GROUP_RAID1 |
8627                                   BTRFS_BLOCK_GROUP_RAID10))
8628                 factor = 2;
8629         else
8630                 factor = 1;
8631
8632         /* make sure this block group isn't part of an allocation cluster */
8633         cluster = &root->fs_info->data_alloc_cluster;
8634         spin_lock(&cluster->refill_lock);
8635         btrfs_return_cluster_to_free_space(block_group, cluster);
8636         spin_unlock(&cluster->refill_lock);
8637
8638         /*
8639          * make sure this block group isn't part of a metadata
8640          * allocation cluster
8641          */
8642         cluster = &root->fs_info->meta_alloc_cluster;
8643         spin_lock(&cluster->refill_lock);
8644         btrfs_return_cluster_to_free_space(block_group, cluster);
8645         spin_unlock(&cluster->refill_lock);
8646
8647         path = btrfs_alloc_path();
8648         BUG_ON(!path);
8649
8650         inode = lookup_free_space_inode(root, block_group, path);
8651         if (!IS_ERR(inode)) {
8652                 btrfs_orphan_add(trans, inode);
8653                 clear_nlink(inode);
8654                 /* One for the block groups ref */
8655                 spin_lock(&block_group->lock);
8656                 if (block_group->iref) {
8657                         block_group->iref = 0;
8658                         block_group->inode = NULL;
8659                         spin_unlock(&block_group->lock);
8660                         iput(inode);
8661                 } else {
8662                         spin_unlock(&block_group->lock);
8663                 }
8664                 /* One for our lookup ref */
8665                 iput(inode);
8666         }
8667
8668         key.objectid = BTRFS_FREE_SPACE_OBJECTID;
8669         key.offset = block_group->key.objectid;
8670         key.type = 0;
8671
8672         ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
8673         if (ret < 0)
8674                 goto out;
8675         if (ret > 0)
8676                 btrfs_release_path(tree_root, path);
8677         if (ret == 0) {
8678                 ret = btrfs_del_item(trans, tree_root, path);
8679                 if (ret)
8680                         goto out;
8681                 btrfs_release_path(tree_root, path);
8682         }
8683
8684         spin_lock(&root->fs_info->block_group_cache_lock);
8685         rb_erase(&block_group->cache_node,
8686                  &root->fs_info->block_group_cache_tree);
8687         spin_unlock(&root->fs_info->block_group_cache_lock);
8688
8689         down_write(&block_group->space_info->groups_sem);
8690         /*
8691          * we must use list_del_init so people can check to see if they
8692          * are still on the list after taking the semaphore
8693          */
8694         list_del_init(&block_group->list);
8695         up_write(&block_group->space_info->groups_sem);
8696
8697         if (block_group->cached == BTRFS_CACHE_STARTED)
8698                 wait_block_group_cache_done(block_group);
8699
8700         btrfs_remove_free_space_cache(block_group);
8701
8702         spin_lock(&block_group->space_info->lock);
8703         block_group->space_info->total_bytes -= block_group->key.offset;
8704         block_group->space_info->bytes_readonly -= block_group->key.offset;
8705         block_group->space_info->disk_total -= block_group->key.offset * factor;
8706         spin_unlock(&block_group->space_info->lock);
8707
8708         memcpy(&key, &block_group->key, sizeof(key));
8709
8710         btrfs_clear_space_info_full(root->fs_info);
8711
8712         btrfs_put_block_group(block_group);
8713         btrfs_put_block_group(block_group);
8714
8715         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8716         if (ret > 0)
8717                 ret = -EIO;
8718         if (ret < 0)
8719                 goto out;
8720
8721         ret = btrfs_del_item(trans, root, path);
8722 out:
8723         btrfs_free_path(path);
8724         return ret;
8725 }
8726
8727 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
8728 {
8729         return unpin_extent_range(root, start, end);
8730 }
8731
8732 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
8733                                u64 num_bytes)
8734 {
8735         return btrfs_discard_extent(root, bytenr, num_bytes);
8736 }