Btrfs: enable discard support
[linux-2.6.git] / fs / btrfs / extent-tree.c
1 /*
2  * Copyright (C) 2007 Oracle.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18 #include <linux/sched.h>
19 #include <linux/pagemap.h>
20 #include <linux/writeback.h>
21 #include <linux/blkdev.h>
22 #include <linux/sort.h>
23 #include <linux/rcupdate.h>
24 #include <linux/kthread.h>
25 #include "compat.h"
26 #include "hash.h"
27 #include "ctree.h"
28 #include "disk-io.h"
29 #include "print-tree.h"
30 #include "transaction.h"
31 #include "volumes.h"
32 #include "locking.h"
33 #include "free-space-cache.h"
34
35 static int update_block_group(struct btrfs_trans_handle *trans,
36                               struct btrfs_root *root,
37                               u64 bytenr, u64 num_bytes, int alloc,
38                               int mark_free);
39 static int update_reserved_extents(struct btrfs_block_group_cache *cache,
40                                    u64 num_bytes, int reserve);
41 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
42                                 struct btrfs_root *root,
43                                 u64 bytenr, u64 num_bytes, u64 parent,
44                                 u64 root_objectid, u64 owner_objectid,
45                                 u64 owner_offset, int refs_to_drop,
46                                 struct btrfs_delayed_extent_op *extra_op);
47 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
48                                     struct extent_buffer *leaf,
49                                     struct btrfs_extent_item *ei);
50 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
51                                       struct btrfs_root *root,
52                                       u64 parent, u64 root_objectid,
53                                       u64 flags, u64 owner, u64 offset,
54                                       struct btrfs_key *ins, int ref_mod);
55 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
56                                      struct btrfs_root *root,
57                                      u64 parent, u64 root_objectid,
58                                      u64 flags, struct btrfs_disk_key *key,
59                                      int level, struct btrfs_key *ins);
60 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
61                           struct btrfs_root *extent_root, u64 alloc_bytes,
62                           u64 flags, int force);
63 static int pin_down_bytes(struct btrfs_trans_handle *trans,
64                           struct btrfs_root *root,
65                           struct btrfs_path *path,
66                           u64 bytenr, u64 num_bytes,
67                           int is_data, int reserved,
68                           struct extent_buffer **must_clean);
69 static int find_next_key(struct btrfs_path *path, int level,
70                          struct btrfs_key *key);
71 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
72                             int dump_block_groups);
73
74 static noinline int
75 block_group_cache_done(struct btrfs_block_group_cache *cache)
76 {
77         smp_mb();
78         return cache->cached == BTRFS_CACHE_FINISHED;
79 }
80
81 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
82 {
83         return (cache->flags & bits) == bits;
84 }
85
86 /*
87  * this adds the block group to the fs_info rb tree for the block group
88  * cache
89  */
90 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
91                                 struct btrfs_block_group_cache *block_group)
92 {
93         struct rb_node **p;
94         struct rb_node *parent = NULL;
95         struct btrfs_block_group_cache *cache;
96
97         spin_lock(&info->block_group_cache_lock);
98         p = &info->block_group_cache_tree.rb_node;
99
100         while (*p) {
101                 parent = *p;
102                 cache = rb_entry(parent, struct btrfs_block_group_cache,
103                                  cache_node);
104                 if (block_group->key.objectid < cache->key.objectid) {
105                         p = &(*p)->rb_left;
106                 } else if (block_group->key.objectid > cache->key.objectid) {
107                         p = &(*p)->rb_right;
108                 } else {
109                         spin_unlock(&info->block_group_cache_lock);
110                         return -EEXIST;
111                 }
112         }
113
114         rb_link_node(&block_group->cache_node, parent, p);
115         rb_insert_color(&block_group->cache_node,
116                         &info->block_group_cache_tree);
117         spin_unlock(&info->block_group_cache_lock);
118
119         return 0;
120 }
121
122 /*
123  * This will return the block group at or after bytenr if contains is 0, else
124  * it will return the block group that contains the bytenr
125  */
126 static struct btrfs_block_group_cache *
127 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
128                               int contains)
129 {
130         struct btrfs_block_group_cache *cache, *ret = NULL;
131         struct rb_node *n;
132         u64 end, start;
133
134         spin_lock(&info->block_group_cache_lock);
135         n = info->block_group_cache_tree.rb_node;
136
137         while (n) {
138                 cache = rb_entry(n, struct btrfs_block_group_cache,
139                                  cache_node);
140                 end = cache->key.objectid + cache->key.offset - 1;
141                 start = cache->key.objectid;
142
143                 if (bytenr < start) {
144                         if (!contains && (!ret || start < ret->key.objectid))
145                                 ret = cache;
146                         n = n->rb_left;
147                 } else if (bytenr > start) {
148                         if (contains && bytenr <= end) {
149                                 ret = cache;
150                                 break;
151                         }
152                         n = n->rb_right;
153                 } else {
154                         ret = cache;
155                         break;
156                 }
157         }
158         if (ret)
159                 atomic_inc(&ret->count);
160         spin_unlock(&info->block_group_cache_lock);
161
162         return ret;
163 }
164
165 static int add_excluded_extent(struct btrfs_root *root,
166                                u64 start, u64 num_bytes)
167 {
168         u64 end = start + num_bytes - 1;
169         set_extent_bits(&root->fs_info->freed_extents[0],
170                         start, end, EXTENT_UPTODATE, GFP_NOFS);
171         set_extent_bits(&root->fs_info->freed_extents[1],
172                         start, end, EXTENT_UPTODATE, GFP_NOFS);
173         return 0;
174 }
175
176 static void free_excluded_extents(struct btrfs_root *root,
177                                   struct btrfs_block_group_cache *cache)
178 {
179         u64 start, end;
180
181         start = cache->key.objectid;
182         end = start + cache->key.offset - 1;
183
184         clear_extent_bits(&root->fs_info->freed_extents[0],
185                           start, end, EXTENT_UPTODATE, GFP_NOFS);
186         clear_extent_bits(&root->fs_info->freed_extents[1],
187                           start, end, EXTENT_UPTODATE, GFP_NOFS);
188 }
189
190 static int exclude_super_stripes(struct btrfs_root *root,
191                                  struct btrfs_block_group_cache *cache)
192 {
193         u64 bytenr;
194         u64 *logical;
195         int stripe_len;
196         int i, nr, ret;
197
198         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
199                 bytenr = btrfs_sb_offset(i);
200                 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
201                                        cache->key.objectid, bytenr,
202                                        0, &logical, &nr, &stripe_len);
203                 BUG_ON(ret);
204
205                 while (nr--) {
206                         cache->bytes_super += stripe_len;
207                         ret = add_excluded_extent(root, logical[nr],
208                                                   stripe_len);
209                         BUG_ON(ret);
210                 }
211
212                 kfree(logical);
213         }
214         return 0;
215 }
216
217 static struct btrfs_caching_control *
218 get_caching_control(struct btrfs_block_group_cache *cache)
219 {
220         struct btrfs_caching_control *ctl;
221
222         spin_lock(&cache->lock);
223         if (cache->cached != BTRFS_CACHE_STARTED) {
224                 spin_unlock(&cache->lock);
225                 return NULL;
226         }
227
228         ctl = cache->caching_ctl;
229         atomic_inc(&ctl->count);
230         spin_unlock(&cache->lock);
231         return ctl;
232 }
233
234 static void put_caching_control(struct btrfs_caching_control *ctl)
235 {
236         if (atomic_dec_and_test(&ctl->count))
237                 kfree(ctl);
238 }
239
240 /*
241  * this is only called by cache_block_group, since we could have freed extents
242  * we need to check the pinned_extents for any extents that can't be used yet
243  * since their free space will be released as soon as the transaction commits.
244  */
245 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
246                               struct btrfs_fs_info *info, u64 start, u64 end)
247 {
248         u64 extent_start, extent_end, size, total_added = 0;
249         int ret;
250
251         while (start < end) {
252                 ret = find_first_extent_bit(info->pinned_extents, start,
253                                             &extent_start, &extent_end,
254                                             EXTENT_DIRTY | EXTENT_UPTODATE);
255                 if (ret)
256                         break;
257
258                 if (extent_start == start) {
259                         start = extent_end + 1;
260                 } else if (extent_start > start && extent_start < end) {
261                         size = extent_start - start;
262                         total_added += size;
263                         ret = btrfs_add_free_space(block_group, start,
264                                                    size);
265                         BUG_ON(ret);
266                         start = extent_end + 1;
267                 } else {
268                         break;
269                 }
270         }
271
272         if (start < end) {
273                 size = end - start;
274                 total_added += size;
275                 ret = btrfs_add_free_space(block_group, start, size);
276                 BUG_ON(ret);
277         }
278
279         return total_added;
280 }
281
282 static int caching_kthread(void *data)
283 {
284         struct btrfs_block_group_cache *block_group = data;
285         struct btrfs_fs_info *fs_info = block_group->fs_info;
286         struct btrfs_caching_control *caching_ctl = block_group->caching_ctl;
287         struct btrfs_root *extent_root = fs_info->extent_root;
288         struct btrfs_path *path;
289         struct extent_buffer *leaf;
290         struct btrfs_key key;
291         u64 total_found = 0;
292         u64 last = 0;
293         u32 nritems;
294         int ret = 0;
295
296         path = btrfs_alloc_path();
297         if (!path)
298                 return -ENOMEM;
299
300         exclude_super_stripes(extent_root, block_group);
301         spin_lock(&block_group->space_info->lock);
302         block_group->space_info->bytes_super += block_group->bytes_super;
303         spin_unlock(&block_group->space_info->lock);
304
305         last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
306
307         /*
308          * We don't want to deadlock with somebody trying to allocate a new
309          * extent for the extent root while also trying to search the extent
310          * root to add free space.  So we skip locking and search the commit
311          * root, since its read-only
312          */
313         path->skip_locking = 1;
314         path->search_commit_root = 1;
315         path->reada = 2;
316
317         key.objectid = last;
318         key.offset = 0;
319         key.type = BTRFS_EXTENT_ITEM_KEY;
320 again:
321         mutex_lock(&caching_ctl->mutex);
322         /* need to make sure the commit_root doesn't disappear */
323         down_read(&fs_info->extent_commit_sem);
324
325         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
326         if (ret < 0)
327                 goto err;
328
329         leaf = path->nodes[0];
330         nritems = btrfs_header_nritems(leaf);
331
332         while (1) {
333                 smp_mb();
334                 if (fs_info->closing > 1) {
335                         last = (u64)-1;
336                         break;
337                 }
338
339                 if (path->slots[0] < nritems) {
340                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
341                 } else {
342                         ret = find_next_key(path, 0, &key);
343                         if (ret)
344                                 break;
345
346                         caching_ctl->progress = last;
347                         btrfs_release_path(extent_root, path);
348                         up_read(&fs_info->extent_commit_sem);
349                         mutex_unlock(&caching_ctl->mutex);
350                         if (btrfs_transaction_in_commit(fs_info))
351                                 schedule_timeout(1);
352                         else
353                                 cond_resched();
354                         goto again;
355                 }
356
357                 if (key.objectid < block_group->key.objectid) {
358                         path->slots[0]++;
359                         continue;
360                 }
361
362                 if (key.objectid >= block_group->key.objectid +
363                     block_group->key.offset)
364                         break;
365
366                 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
367                         total_found += add_new_free_space(block_group,
368                                                           fs_info, last,
369                                                           key.objectid);
370                         last = key.objectid + key.offset;
371
372                         if (total_found > (1024 * 1024 * 2)) {
373                                 total_found = 0;
374                                 wake_up(&caching_ctl->wait);
375                         }
376                 }
377                 path->slots[0]++;
378         }
379         ret = 0;
380
381         total_found += add_new_free_space(block_group, fs_info, last,
382                                           block_group->key.objectid +
383                                           block_group->key.offset);
384         caching_ctl->progress = (u64)-1;
385
386         spin_lock(&block_group->lock);
387         block_group->caching_ctl = NULL;
388         block_group->cached = BTRFS_CACHE_FINISHED;
389         spin_unlock(&block_group->lock);
390
391 err:
392         btrfs_free_path(path);
393         up_read(&fs_info->extent_commit_sem);
394
395         free_excluded_extents(extent_root, block_group);
396
397         mutex_unlock(&caching_ctl->mutex);
398         wake_up(&caching_ctl->wait);
399
400         put_caching_control(caching_ctl);
401         atomic_dec(&block_group->space_info->caching_threads);
402         return 0;
403 }
404
405 static int cache_block_group(struct btrfs_block_group_cache *cache)
406 {
407         struct btrfs_fs_info *fs_info = cache->fs_info;
408         struct btrfs_caching_control *caching_ctl;
409         struct task_struct *tsk;
410         int ret = 0;
411
412         smp_mb();
413         if (cache->cached != BTRFS_CACHE_NO)
414                 return 0;
415
416         caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_KERNEL);
417         BUG_ON(!caching_ctl);
418
419         INIT_LIST_HEAD(&caching_ctl->list);
420         mutex_init(&caching_ctl->mutex);
421         init_waitqueue_head(&caching_ctl->wait);
422         caching_ctl->block_group = cache;
423         caching_ctl->progress = cache->key.objectid;
424         /* one for caching kthread, one for caching block group list */
425         atomic_set(&caching_ctl->count, 2);
426
427         spin_lock(&cache->lock);
428         if (cache->cached != BTRFS_CACHE_NO) {
429                 spin_unlock(&cache->lock);
430                 kfree(caching_ctl);
431                 return 0;
432         }
433         cache->caching_ctl = caching_ctl;
434         cache->cached = BTRFS_CACHE_STARTED;
435         spin_unlock(&cache->lock);
436
437         down_write(&fs_info->extent_commit_sem);
438         list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
439         up_write(&fs_info->extent_commit_sem);
440
441         atomic_inc(&cache->space_info->caching_threads);
442
443         tsk = kthread_run(caching_kthread, cache, "btrfs-cache-%llu\n",
444                           cache->key.objectid);
445         if (IS_ERR(tsk)) {
446                 ret = PTR_ERR(tsk);
447                 printk(KERN_ERR "error running thread %d\n", ret);
448                 BUG();
449         }
450
451         return ret;
452 }
453
454 /*
455  * return the block group that starts at or after bytenr
456  */
457 static struct btrfs_block_group_cache *
458 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
459 {
460         struct btrfs_block_group_cache *cache;
461
462         cache = block_group_cache_tree_search(info, bytenr, 0);
463
464         return cache;
465 }
466
467 /*
468  * return the block group that contains the given bytenr
469  */
470 struct btrfs_block_group_cache *btrfs_lookup_block_group(
471                                                  struct btrfs_fs_info *info,
472                                                  u64 bytenr)
473 {
474         struct btrfs_block_group_cache *cache;
475
476         cache = block_group_cache_tree_search(info, bytenr, 1);
477
478         return cache;
479 }
480
481 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
482 {
483         if (atomic_dec_and_test(&cache->count))
484                 kfree(cache);
485 }
486
487 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
488                                                   u64 flags)
489 {
490         struct list_head *head = &info->space_info;
491         struct btrfs_space_info *found;
492
493         rcu_read_lock();
494         list_for_each_entry_rcu(found, head, list) {
495                 if (found->flags == flags) {
496                         rcu_read_unlock();
497                         return found;
498                 }
499         }
500         rcu_read_unlock();
501         return NULL;
502 }
503
504 /*
505  * after adding space to the filesystem, we need to clear the full flags
506  * on all the space infos.
507  */
508 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
509 {
510         struct list_head *head = &info->space_info;
511         struct btrfs_space_info *found;
512
513         rcu_read_lock();
514         list_for_each_entry_rcu(found, head, list)
515                 found->full = 0;
516         rcu_read_unlock();
517 }
518
519 static u64 div_factor(u64 num, int factor)
520 {
521         if (factor == 10)
522                 return num;
523         num *= factor;
524         do_div(num, 10);
525         return num;
526 }
527
528 u64 btrfs_find_block_group(struct btrfs_root *root,
529                            u64 search_start, u64 search_hint, int owner)
530 {
531         struct btrfs_block_group_cache *cache;
532         u64 used;
533         u64 last = max(search_hint, search_start);
534         u64 group_start = 0;
535         int full_search = 0;
536         int factor = 9;
537         int wrapped = 0;
538 again:
539         while (1) {
540                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
541                 if (!cache)
542                         break;
543
544                 spin_lock(&cache->lock);
545                 last = cache->key.objectid + cache->key.offset;
546                 used = btrfs_block_group_used(&cache->item);
547
548                 if ((full_search || !cache->ro) &&
549                     block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
550                         if (used + cache->pinned + cache->reserved <
551                             div_factor(cache->key.offset, factor)) {
552                                 group_start = cache->key.objectid;
553                                 spin_unlock(&cache->lock);
554                                 btrfs_put_block_group(cache);
555                                 goto found;
556                         }
557                 }
558                 spin_unlock(&cache->lock);
559                 btrfs_put_block_group(cache);
560                 cond_resched();
561         }
562         if (!wrapped) {
563                 last = search_start;
564                 wrapped = 1;
565                 goto again;
566         }
567         if (!full_search && factor < 10) {
568                 last = search_start;
569                 full_search = 1;
570                 factor = 10;
571                 goto again;
572         }
573 found:
574         return group_start;
575 }
576
577 /* simple helper to search for an existing extent at a given offset */
578 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
579 {
580         int ret;
581         struct btrfs_key key;
582         struct btrfs_path *path;
583
584         path = btrfs_alloc_path();
585         BUG_ON(!path);
586         key.objectid = start;
587         key.offset = len;
588         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
589         ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
590                                 0, 0);
591         btrfs_free_path(path);
592         return ret;
593 }
594
595 /*
596  * Back reference rules.  Back refs have three main goals:
597  *
598  * 1) differentiate between all holders of references to an extent so that
599  *    when a reference is dropped we can make sure it was a valid reference
600  *    before freeing the extent.
601  *
602  * 2) Provide enough information to quickly find the holders of an extent
603  *    if we notice a given block is corrupted or bad.
604  *
605  * 3) Make it easy to migrate blocks for FS shrinking or storage pool
606  *    maintenance.  This is actually the same as #2, but with a slightly
607  *    different use case.
608  *
609  * There are two kinds of back refs. The implicit back refs is optimized
610  * for pointers in non-shared tree blocks. For a given pointer in a block,
611  * back refs of this kind provide information about the block's owner tree
612  * and the pointer's key. These information allow us to find the block by
613  * b-tree searching. The full back refs is for pointers in tree blocks not
614  * referenced by their owner trees. The location of tree block is recorded
615  * in the back refs. Actually the full back refs is generic, and can be
616  * used in all cases the implicit back refs is used. The major shortcoming
617  * of the full back refs is its overhead. Every time a tree block gets
618  * COWed, we have to update back refs entry for all pointers in it.
619  *
620  * For a newly allocated tree block, we use implicit back refs for
621  * pointers in it. This means most tree related operations only involve
622  * implicit back refs. For a tree block created in old transaction, the
623  * only way to drop a reference to it is COW it. So we can detect the
624  * event that tree block loses its owner tree's reference and do the
625  * back refs conversion.
626  *
627  * When a tree block is COW'd through a tree, there are four cases:
628  *
629  * The reference count of the block is one and the tree is the block's
630  * owner tree. Nothing to do in this case.
631  *
632  * The reference count of the block is one and the tree is not the
633  * block's owner tree. In this case, full back refs is used for pointers
634  * in the block. Remove these full back refs, add implicit back refs for
635  * every pointers in the new block.
636  *
637  * The reference count of the block is greater than one and the tree is
638  * the block's owner tree. In this case, implicit back refs is used for
639  * pointers in the block. Add full back refs for every pointers in the
640  * block, increase lower level extents' reference counts. The original
641  * implicit back refs are entailed to the new block.
642  *
643  * The reference count of the block is greater than one and the tree is
644  * not the block's owner tree. Add implicit back refs for every pointer in
645  * the new block, increase lower level extents' reference count.
646  *
647  * Back Reference Key composing:
648  *
649  * The key objectid corresponds to the first byte in the extent,
650  * The key type is used to differentiate between types of back refs.
651  * There are different meanings of the key offset for different types
652  * of back refs.
653  *
654  * File extents can be referenced by:
655  *
656  * - multiple snapshots, subvolumes, or different generations in one subvol
657  * - different files inside a single subvolume
658  * - different offsets inside a file (bookend extents in file.c)
659  *
660  * The extent ref structure for the implicit back refs has fields for:
661  *
662  * - Objectid of the subvolume root
663  * - objectid of the file holding the reference
664  * - original offset in the file
665  * - how many bookend extents
666  *
667  * The key offset for the implicit back refs is hash of the first
668  * three fields.
669  *
670  * The extent ref structure for the full back refs has field for:
671  *
672  * - number of pointers in the tree leaf
673  *
674  * The key offset for the implicit back refs is the first byte of
675  * the tree leaf
676  *
677  * When a file extent is allocated, The implicit back refs is used.
678  * the fields are filled in:
679  *
680  *     (root_key.objectid, inode objectid, offset in file, 1)
681  *
682  * When a file extent is removed file truncation, we find the
683  * corresponding implicit back refs and check the following fields:
684  *
685  *     (btrfs_header_owner(leaf), inode objectid, offset in file)
686  *
687  * Btree extents can be referenced by:
688  *
689  * - Different subvolumes
690  *
691  * Both the implicit back refs and the full back refs for tree blocks
692  * only consist of key. The key offset for the implicit back refs is
693  * objectid of block's owner tree. The key offset for the full back refs
694  * is the first byte of parent block.
695  *
696  * When implicit back refs is used, information about the lowest key and
697  * level of the tree block are required. These information are stored in
698  * tree block info structure.
699  */
700
701 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
702 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
703                                   struct btrfs_root *root,
704                                   struct btrfs_path *path,
705                                   u64 owner, u32 extra_size)
706 {
707         struct btrfs_extent_item *item;
708         struct btrfs_extent_item_v0 *ei0;
709         struct btrfs_extent_ref_v0 *ref0;
710         struct btrfs_tree_block_info *bi;
711         struct extent_buffer *leaf;
712         struct btrfs_key key;
713         struct btrfs_key found_key;
714         u32 new_size = sizeof(*item);
715         u64 refs;
716         int ret;
717
718         leaf = path->nodes[0];
719         BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
720
721         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
722         ei0 = btrfs_item_ptr(leaf, path->slots[0],
723                              struct btrfs_extent_item_v0);
724         refs = btrfs_extent_refs_v0(leaf, ei0);
725
726         if (owner == (u64)-1) {
727                 while (1) {
728                         if (path->slots[0] >= btrfs_header_nritems(leaf)) {
729                                 ret = btrfs_next_leaf(root, path);
730                                 if (ret < 0)
731                                         return ret;
732                                 BUG_ON(ret > 0);
733                                 leaf = path->nodes[0];
734                         }
735                         btrfs_item_key_to_cpu(leaf, &found_key,
736                                               path->slots[0]);
737                         BUG_ON(key.objectid != found_key.objectid);
738                         if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
739                                 path->slots[0]++;
740                                 continue;
741                         }
742                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
743                                               struct btrfs_extent_ref_v0);
744                         owner = btrfs_ref_objectid_v0(leaf, ref0);
745                         break;
746                 }
747         }
748         btrfs_release_path(root, path);
749
750         if (owner < BTRFS_FIRST_FREE_OBJECTID)
751                 new_size += sizeof(*bi);
752
753         new_size -= sizeof(*ei0);
754         ret = btrfs_search_slot(trans, root, &key, path,
755                                 new_size + extra_size, 1);
756         if (ret < 0)
757                 return ret;
758         BUG_ON(ret);
759
760         ret = btrfs_extend_item(trans, root, path, new_size);
761         BUG_ON(ret);
762
763         leaf = path->nodes[0];
764         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
765         btrfs_set_extent_refs(leaf, item, refs);
766         /* FIXME: get real generation */
767         btrfs_set_extent_generation(leaf, item, 0);
768         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
769                 btrfs_set_extent_flags(leaf, item,
770                                        BTRFS_EXTENT_FLAG_TREE_BLOCK |
771                                        BTRFS_BLOCK_FLAG_FULL_BACKREF);
772                 bi = (struct btrfs_tree_block_info *)(item + 1);
773                 /* FIXME: get first key of the block */
774                 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
775                 btrfs_set_tree_block_level(leaf, bi, (int)owner);
776         } else {
777                 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
778         }
779         btrfs_mark_buffer_dirty(leaf);
780         return 0;
781 }
782 #endif
783
784 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
785 {
786         u32 high_crc = ~(u32)0;
787         u32 low_crc = ~(u32)0;
788         __le64 lenum;
789
790         lenum = cpu_to_le64(root_objectid);
791         high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
792         lenum = cpu_to_le64(owner);
793         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
794         lenum = cpu_to_le64(offset);
795         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
796
797         return ((u64)high_crc << 31) ^ (u64)low_crc;
798 }
799
800 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
801                                      struct btrfs_extent_data_ref *ref)
802 {
803         return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
804                                     btrfs_extent_data_ref_objectid(leaf, ref),
805                                     btrfs_extent_data_ref_offset(leaf, ref));
806 }
807
808 static int match_extent_data_ref(struct extent_buffer *leaf,
809                                  struct btrfs_extent_data_ref *ref,
810                                  u64 root_objectid, u64 owner, u64 offset)
811 {
812         if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
813             btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
814             btrfs_extent_data_ref_offset(leaf, ref) != offset)
815                 return 0;
816         return 1;
817 }
818
819 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
820                                            struct btrfs_root *root,
821                                            struct btrfs_path *path,
822                                            u64 bytenr, u64 parent,
823                                            u64 root_objectid,
824                                            u64 owner, u64 offset)
825 {
826         struct btrfs_key key;
827         struct btrfs_extent_data_ref *ref;
828         struct extent_buffer *leaf;
829         u32 nritems;
830         int ret;
831         int recow;
832         int err = -ENOENT;
833
834         key.objectid = bytenr;
835         if (parent) {
836                 key.type = BTRFS_SHARED_DATA_REF_KEY;
837                 key.offset = parent;
838         } else {
839                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
840                 key.offset = hash_extent_data_ref(root_objectid,
841                                                   owner, offset);
842         }
843 again:
844         recow = 0;
845         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
846         if (ret < 0) {
847                 err = ret;
848                 goto fail;
849         }
850
851         if (parent) {
852                 if (!ret)
853                         return 0;
854 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
855                 key.type = BTRFS_EXTENT_REF_V0_KEY;
856                 btrfs_release_path(root, path);
857                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
858                 if (ret < 0) {
859                         err = ret;
860                         goto fail;
861                 }
862                 if (!ret)
863                         return 0;
864 #endif
865                 goto fail;
866         }
867
868         leaf = path->nodes[0];
869         nritems = btrfs_header_nritems(leaf);
870         while (1) {
871                 if (path->slots[0] >= nritems) {
872                         ret = btrfs_next_leaf(root, path);
873                         if (ret < 0)
874                                 err = ret;
875                         if (ret)
876                                 goto fail;
877
878                         leaf = path->nodes[0];
879                         nritems = btrfs_header_nritems(leaf);
880                         recow = 1;
881                 }
882
883                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
884                 if (key.objectid != bytenr ||
885                     key.type != BTRFS_EXTENT_DATA_REF_KEY)
886                         goto fail;
887
888                 ref = btrfs_item_ptr(leaf, path->slots[0],
889                                      struct btrfs_extent_data_ref);
890
891                 if (match_extent_data_ref(leaf, ref, root_objectid,
892                                           owner, offset)) {
893                         if (recow) {
894                                 btrfs_release_path(root, path);
895                                 goto again;
896                         }
897                         err = 0;
898                         break;
899                 }
900                 path->slots[0]++;
901         }
902 fail:
903         return err;
904 }
905
906 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
907                                            struct btrfs_root *root,
908                                            struct btrfs_path *path,
909                                            u64 bytenr, u64 parent,
910                                            u64 root_objectid, u64 owner,
911                                            u64 offset, int refs_to_add)
912 {
913         struct btrfs_key key;
914         struct extent_buffer *leaf;
915         u32 size;
916         u32 num_refs;
917         int ret;
918
919         key.objectid = bytenr;
920         if (parent) {
921                 key.type = BTRFS_SHARED_DATA_REF_KEY;
922                 key.offset = parent;
923                 size = sizeof(struct btrfs_shared_data_ref);
924         } else {
925                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
926                 key.offset = hash_extent_data_ref(root_objectid,
927                                                   owner, offset);
928                 size = sizeof(struct btrfs_extent_data_ref);
929         }
930
931         ret = btrfs_insert_empty_item(trans, root, path, &key, size);
932         if (ret && ret != -EEXIST)
933                 goto fail;
934
935         leaf = path->nodes[0];
936         if (parent) {
937                 struct btrfs_shared_data_ref *ref;
938                 ref = btrfs_item_ptr(leaf, path->slots[0],
939                                      struct btrfs_shared_data_ref);
940                 if (ret == 0) {
941                         btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
942                 } else {
943                         num_refs = btrfs_shared_data_ref_count(leaf, ref);
944                         num_refs += refs_to_add;
945                         btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
946                 }
947         } else {
948                 struct btrfs_extent_data_ref *ref;
949                 while (ret == -EEXIST) {
950                         ref = btrfs_item_ptr(leaf, path->slots[0],
951                                              struct btrfs_extent_data_ref);
952                         if (match_extent_data_ref(leaf, ref, root_objectid,
953                                                   owner, offset))
954                                 break;
955                         btrfs_release_path(root, path);
956                         key.offset++;
957                         ret = btrfs_insert_empty_item(trans, root, path, &key,
958                                                       size);
959                         if (ret && ret != -EEXIST)
960                                 goto fail;
961
962                         leaf = path->nodes[0];
963                 }
964                 ref = btrfs_item_ptr(leaf, path->slots[0],
965                                      struct btrfs_extent_data_ref);
966                 if (ret == 0) {
967                         btrfs_set_extent_data_ref_root(leaf, ref,
968                                                        root_objectid);
969                         btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
970                         btrfs_set_extent_data_ref_offset(leaf, ref, offset);
971                         btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
972                 } else {
973                         num_refs = btrfs_extent_data_ref_count(leaf, ref);
974                         num_refs += refs_to_add;
975                         btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
976                 }
977         }
978         btrfs_mark_buffer_dirty(leaf);
979         ret = 0;
980 fail:
981         btrfs_release_path(root, path);
982         return ret;
983 }
984
985 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
986                                            struct btrfs_root *root,
987                                            struct btrfs_path *path,
988                                            int refs_to_drop)
989 {
990         struct btrfs_key key;
991         struct btrfs_extent_data_ref *ref1 = NULL;
992         struct btrfs_shared_data_ref *ref2 = NULL;
993         struct extent_buffer *leaf;
994         u32 num_refs = 0;
995         int ret = 0;
996
997         leaf = path->nodes[0];
998         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
999
1000         if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1001                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1002                                       struct btrfs_extent_data_ref);
1003                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1004         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1005                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1006                                       struct btrfs_shared_data_ref);
1007                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1008 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1009         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1010                 struct btrfs_extent_ref_v0 *ref0;
1011                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1012                                       struct btrfs_extent_ref_v0);
1013                 num_refs = btrfs_ref_count_v0(leaf, ref0);
1014 #endif
1015         } else {
1016                 BUG();
1017         }
1018
1019         BUG_ON(num_refs < refs_to_drop);
1020         num_refs -= refs_to_drop;
1021
1022         if (num_refs == 0) {
1023                 ret = btrfs_del_item(trans, root, path);
1024         } else {
1025                 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1026                         btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1027                 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1028                         btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1029 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1030                 else {
1031                         struct btrfs_extent_ref_v0 *ref0;
1032                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
1033                                         struct btrfs_extent_ref_v0);
1034                         btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1035                 }
1036 #endif
1037                 btrfs_mark_buffer_dirty(leaf);
1038         }
1039         return ret;
1040 }
1041
1042 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1043                                           struct btrfs_path *path,
1044                                           struct btrfs_extent_inline_ref *iref)
1045 {
1046         struct btrfs_key key;
1047         struct extent_buffer *leaf;
1048         struct btrfs_extent_data_ref *ref1;
1049         struct btrfs_shared_data_ref *ref2;
1050         u32 num_refs = 0;
1051
1052         leaf = path->nodes[0];
1053         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1054         if (iref) {
1055                 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1056                     BTRFS_EXTENT_DATA_REF_KEY) {
1057                         ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1058                         num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1059                 } else {
1060                         ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1061                         num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1062                 }
1063         } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1064                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1065                                       struct btrfs_extent_data_ref);
1066                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1067         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1068                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1069                                       struct btrfs_shared_data_ref);
1070                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1071 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1072         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1073                 struct btrfs_extent_ref_v0 *ref0;
1074                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1075                                       struct btrfs_extent_ref_v0);
1076                 num_refs = btrfs_ref_count_v0(leaf, ref0);
1077 #endif
1078         } else {
1079                 WARN_ON(1);
1080         }
1081         return num_refs;
1082 }
1083
1084 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1085                                           struct btrfs_root *root,
1086                                           struct btrfs_path *path,
1087                                           u64 bytenr, u64 parent,
1088                                           u64 root_objectid)
1089 {
1090         struct btrfs_key key;
1091         int ret;
1092
1093         key.objectid = bytenr;
1094         if (parent) {
1095                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1096                 key.offset = parent;
1097         } else {
1098                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1099                 key.offset = root_objectid;
1100         }
1101
1102         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1103         if (ret > 0)
1104                 ret = -ENOENT;
1105 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1106         if (ret == -ENOENT && parent) {
1107                 btrfs_release_path(root, path);
1108                 key.type = BTRFS_EXTENT_REF_V0_KEY;
1109                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1110                 if (ret > 0)
1111                         ret = -ENOENT;
1112         }
1113 #endif
1114         return ret;
1115 }
1116
1117 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1118                                           struct btrfs_root *root,
1119                                           struct btrfs_path *path,
1120                                           u64 bytenr, u64 parent,
1121                                           u64 root_objectid)
1122 {
1123         struct btrfs_key key;
1124         int ret;
1125
1126         key.objectid = bytenr;
1127         if (parent) {
1128                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1129                 key.offset = parent;
1130         } else {
1131                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1132                 key.offset = root_objectid;
1133         }
1134
1135         ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1136         btrfs_release_path(root, path);
1137         return ret;
1138 }
1139
1140 static inline int extent_ref_type(u64 parent, u64 owner)
1141 {
1142         int type;
1143         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1144                 if (parent > 0)
1145                         type = BTRFS_SHARED_BLOCK_REF_KEY;
1146                 else
1147                         type = BTRFS_TREE_BLOCK_REF_KEY;
1148         } else {
1149                 if (parent > 0)
1150                         type = BTRFS_SHARED_DATA_REF_KEY;
1151                 else
1152                         type = BTRFS_EXTENT_DATA_REF_KEY;
1153         }
1154         return type;
1155 }
1156
1157 static int find_next_key(struct btrfs_path *path, int level,
1158                          struct btrfs_key *key)
1159
1160 {
1161         for (; level < BTRFS_MAX_LEVEL; level++) {
1162                 if (!path->nodes[level])
1163                         break;
1164                 if (path->slots[level] + 1 >=
1165                     btrfs_header_nritems(path->nodes[level]))
1166                         continue;
1167                 if (level == 0)
1168                         btrfs_item_key_to_cpu(path->nodes[level], key,
1169                                               path->slots[level] + 1);
1170                 else
1171                         btrfs_node_key_to_cpu(path->nodes[level], key,
1172                                               path->slots[level] + 1);
1173                 return 0;
1174         }
1175         return 1;
1176 }
1177
1178 /*
1179  * look for inline back ref. if back ref is found, *ref_ret is set
1180  * to the address of inline back ref, and 0 is returned.
1181  *
1182  * if back ref isn't found, *ref_ret is set to the address where it
1183  * should be inserted, and -ENOENT is returned.
1184  *
1185  * if insert is true and there are too many inline back refs, the path
1186  * points to the extent item, and -EAGAIN is returned.
1187  *
1188  * NOTE: inline back refs are ordered in the same way that back ref
1189  *       items in the tree are ordered.
1190  */
1191 static noinline_for_stack
1192 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1193                                  struct btrfs_root *root,
1194                                  struct btrfs_path *path,
1195                                  struct btrfs_extent_inline_ref **ref_ret,
1196                                  u64 bytenr, u64 num_bytes,
1197                                  u64 parent, u64 root_objectid,
1198                                  u64 owner, u64 offset, int insert)
1199 {
1200         struct btrfs_key key;
1201         struct extent_buffer *leaf;
1202         struct btrfs_extent_item *ei;
1203         struct btrfs_extent_inline_ref *iref;
1204         u64 flags;
1205         u64 item_size;
1206         unsigned long ptr;
1207         unsigned long end;
1208         int extra_size;
1209         int type;
1210         int want;
1211         int ret;
1212         int err = 0;
1213
1214         key.objectid = bytenr;
1215         key.type = BTRFS_EXTENT_ITEM_KEY;
1216         key.offset = num_bytes;
1217
1218         want = extent_ref_type(parent, owner);
1219         if (insert) {
1220                 extra_size = btrfs_extent_inline_ref_size(want);
1221                 path->keep_locks = 1;
1222         } else
1223                 extra_size = -1;
1224         ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1225         if (ret < 0) {
1226                 err = ret;
1227                 goto out;
1228         }
1229         BUG_ON(ret);
1230
1231         leaf = path->nodes[0];
1232         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1233 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1234         if (item_size < sizeof(*ei)) {
1235                 if (!insert) {
1236                         err = -ENOENT;
1237                         goto out;
1238                 }
1239                 ret = convert_extent_item_v0(trans, root, path, owner,
1240                                              extra_size);
1241                 if (ret < 0) {
1242                         err = ret;
1243                         goto out;
1244                 }
1245                 leaf = path->nodes[0];
1246                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1247         }
1248 #endif
1249         BUG_ON(item_size < sizeof(*ei));
1250
1251         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1252         flags = btrfs_extent_flags(leaf, ei);
1253
1254         ptr = (unsigned long)(ei + 1);
1255         end = (unsigned long)ei + item_size;
1256
1257         if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1258                 ptr += sizeof(struct btrfs_tree_block_info);
1259                 BUG_ON(ptr > end);
1260         } else {
1261                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1262         }
1263
1264         err = -ENOENT;
1265         while (1) {
1266                 if (ptr >= end) {
1267                         WARN_ON(ptr > end);
1268                         break;
1269                 }
1270                 iref = (struct btrfs_extent_inline_ref *)ptr;
1271                 type = btrfs_extent_inline_ref_type(leaf, iref);
1272                 if (want < type)
1273                         break;
1274                 if (want > type) {
1275                         ptr += btrfs_extent_inline_ref_size(type);
1276                         continue;
1277                 }
1278
1279                 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1280                         struct btrfs_extent_data_ref *dref;
1281                         dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1282                         if (match_extent_data_ref(leaf, dref, root_objectid,
1283                                                   owner, offset)) {
1284                                 err = 0;
1285                                 break;
1286                         }
1287                         if (hash_extent_data_ref_item(leaf, dref) <
1288                             hash_extent_data_ref(root_objectid, owner, offset))
1289                                 break;
1290                 } else {
1291                         u64 ref_offset;
1292                         ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1293                         if (parent > 0) {
1294                                 if (parent == ref_offset) {
1295                                         err = 0;
1296                                         break;
1297                                 }
1298                                 if (ref_offset < parent)
1299                                         break;
1300                         } else {
1301                                 if (root_objectid == ref_offset) {
1302                                         err = 0;
1303                                         break;
1304                                 }
1305                                 if (ref_offset < root_objectid)
1306                                         break;
1307                         }
1308                 }
1309                 ptr += btrfs_extent_inline_ref_size(type);
1310         }
1311         if (err == -ENOENT && insert) {
1312                 if (item_size + extra_size >=
1313                     BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1314                         err = -EAGAIN;
1315                         goto out;
1316                 }
1317                 /*
1318                  * To add new inline back ref, we have to make sure
1319                  * there is no corresponding back ref item.
1320                  * For simplicity, we just do not add new inline back
1321                  * ref if there is any kind of item for this block
1322                  */
1323                 if (find_next_key(path, 0, &key) == 0 &&
1324                     key.objectid == bytenr &&
1325                     key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1326                         err = -EAGAIN;
1327                         goto out;
1328                 }
1329         }
1330         *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1331 out:
1332         if (insert) {
1333                 path->keep_locks = 0;
1334                 btrfs_unlock_up_safe(path, 1);
1335         }
1336         return err;
1337 }
1338
1339 /*
1340  * helper to add new inline back ref
1341  */
1342 static noinline_for_stack
1343 int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1344                                 struct btrfs_root *root,
1345                                 struct btrfs_path *path,
1346                                 struct btrfs_extent_inline_ref *iref,
1347                                 u64 parent, u64 root_objectid,
1348                                 u64 owner, u64 offset, int refs_to_add,
1349                                 struct btrfs_delayed_extent_op *extent_op)
1350 {
1351         struct extent_buffer *leaf;
1352         struct btrfs_extent_item *ei;
1353         unsigned long ptr;
1354         unsigned long end;
1355         unsigned long item_offset;
1356         u64 refs;
1357         int size;
1358         int type;
1359         int ret;
1360
1361         leaf = path->nodes[0];
1362         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1363         item_offset = (unsigned long)iref - (unsigned long)ei;
1364
1365         type = extent_ref_type(parent, owner);
1366         size = btrfs_extent_inline_ref_size(type);
1367
1368         ret = btrfs_extend_item(trans, root, path, size);
1369         BUG_ON(ret);
1370
1371         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1372         refs = btrfs_extent_refs(leaf, ei);
1373         refs += refs_to_add;
1374         btrfs_set_extent_refs(leaf, ei, refs);
1375         if (extent_op)
1376                 __run_delayed_extent_op(extent_op, leaf, ei);
1377
1378         ptr = (unsigned long)ei + item_offset;
1379         end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1380         if (ptr < end - size)
1381                 memmove_extent_buffer(leaf, ptr + size, ptr,
1382                                       end - size - ptr);
1383
1384         iref = (struct btrfs_extent_inline_ref *)ptr;
1385         btrfs_set_extent_inline_ref_type(leaf, iref, type);
1386         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1387                 struct btrfs_extent_data_ref *dref;
1388                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1389                 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1390                 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1391                 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1392                 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1393         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1394                 struct btrfs_shared_data_ref *sref;
1395                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1396                 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1397                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1398         } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1399                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1400         } else {
1401                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1402         }
1403         btrfs_mark_buffer_dirty(leaf);
1404         return 0;
1405 }
1406
1407 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1408                                  struct btrfs_root *root,
1409                                  struct btrfs_path *path,
1410                                  struct btrfs_extent_inline_ref **ref_ret,
1411                                  u64 bytenr, u64 num_bytes, u64 parent,
1412                                  u64 root_objectid, u64 owner, u64 offset)
1413 {
1414         int ret;
1415
1416         ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1417                                            bytenr, num_bytes, parent,
1418                                            root_objectid, owner, offset, 0);
1419         if (ret != -ENOENT)
1420                 return ret;
1421
1422         btrfs_release_path(root, path);
1423         *ref_ret = NULL;
1424
1425         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1426                 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1427                                             root_objectid);
1428         } else {
1429                 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1430                                              root_objectid, owner, offset);
1431         }
1432         return ret;
1433 }
1434
1435 /*
1436  * helper to update/remove inline back ref
1437  */
1438 static noinline_for_stack
1439 int update_inline_extent_backref(struct btrfs_trans_handle *trans,
1440                                  struct btrfs_root *root,
1441                                  struct btrfs_path *path,
1442                                  struct btrfs_extent_inline_ref *iref,
1443                                  int refs_to_mod,
1444                                  struct btrfs_delayed_extent_op *extent_op)
1445 {
1446         struct extent_buffer *leaf;
1447         struct btrfs_extent_item *ei;
1448         struct btrfs_extent_data_ref *dref = NULL;
1449         struct btrfs_shared_data_ref *sref = NULL;
1450         unsigned long ptr;
1451         unsigned long end;
1452         u32 item_size;
1453         int size;
1454         int type;
1455         int ret;
1456         u64 refs;
1457
1458         leaf = path->nodes[0];
1459         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1460         refs = btrfs_extent_refs(leaf, ei);
1461         WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1462         refs += refs_to_mod;
1463         btrfs_set_extent_refs(leaf, ei, refs);
1464         if (extent_op)
1465                 __run_delayed_extent_op(extent_op, leaf, ei);
1466
1467         type = btrfs_extent_inline_ref_type(leaf, iref);
1468
1469         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1470                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1471                 refs = btrfs_extent_data_ref_count(leaf, dref);
1472         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1473                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1474                 refs = btrfs_shared_data_ref_count(leaf, sref);
1475         } else {
1476                 refs = 1;
1477                 BUG_ON(refs_to_mod != -1);
1478         }
1479
1480         BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1481         refs += refs_to_mod;
1482
1483         if (refs > 0) {
1484                 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1485                         btrfs_set_extent_data_ref_count(leaf, dref, refs);
1486                 else
1487                         btrfs_set_shared_data_ref_count(leaf, sref, refs);
1488         } else {
1489                 size =  btrfs_extent_inline_ref_size(type);
1490                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1491                 ptr = (unsigned long)iref;
1492                 end = (unsigned long)ei + item_size;
1493                 if (ptr + size < end)
1494                         memmove_extent_buffer(leaf, ptr, ptr + size,
1495                                               end - ptr - size);
1496                 item_size -= size;
1497                 ret = btrfs_truncate_item(trans, root, path, item_size, 1);
1498                 BUG_ON(ret);
1499         }
1500         btrfs_mark_buffer_dirty(leaf);
1501         return 0;
1502 }
1503
1504 static noinline_for_stack
1505 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1506                                  struct btrfs_root *root,
1507                                  struct btrfs_path *path,
1508                                  u64 bytenr, u64 num_bytes, u64 parent,
1509                                  u64 root_objectid, u64 owner,
1510                                  u64 offset, int refs_to_add,
1511                                  struct btrfs_delayed_extent_op *extent_op)
1512 {
1513         struct btrfs_extent_inline_ref *iref;
1514         int ret;
1515
1516         ret = lookup_inline_extent_backref(trans, root, path, &iref,
1517                                            bytenr, num_bytes, parent,
1518                                            root_objectid, owner, offset, 1);
1519         if (ret == 0) {
1520                 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1521                 ret = update_inline_extent_backref(trans, root, path, iref,
1522                                                    refs_to_add, extent_op);
1523         } else if (ret == -ENOENT) {
1524                 ret = setup_inline_extent_backref(trans, root, path, iref,
1525                                                   parent, root_objectid,
1526                                                   owner, offset, refs_to_add,
1527                                                   extent_op);
1528         }
1529         return ret;
1530 }
1531
1532 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1533                                  struct btrfs_root *root,
1534                                  struct btrfs_path *path,
1535                                  u64 bytenr, u64 parent, u64 root_objectid,
1536                                  u64 owner, u64 offset, int refs_to_add)
1537 {
1538         int ret;
1539         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1540                 BUG_ON(refs_to_add != 1);
1541                 ret = insert_tree_block_ref(trans, root, path, bytenr,
1542                                             parent, root_objectid);
1543         } else {
1544                 ret = insert_extent_data_ref(trans, root, path, bytenr,
1545                                              parent, root_objectid,
1546                                              owner, offset, refs_to_add);
1547         }
1548         return ret;
1549 }
1550
1551 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1552                                  struct btrfs_root *root,
1553                                  struct btrfs_path *path,
1554                                  struct btrfs_extent_inline_ref *iref,
1555                                  int refs_to_drop, int is_data)
1556 {
1557         int ret;
1558
1559         BUG_ON(!is_data && refs_to_drop != 1);
1560         if (iref) {
1561                 ret = update_inline_extent_backref(trans, root, path, iref,
1562                                                    -refs_to_drop, NULL);
1563         } else if (is_data) {
1564                 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1565         } else {
1566                 ret = btrfs_del_item(trans, root, path);
1567         }
1568         return ret;
1569 }
1570
1571 static void btrfs_issue_discard(struct block_device *bdev,
1572                                 u64 start, u64 len)
1573 {
1574         blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL);
1575 }
1576
1577 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1578                                 u64 num_bytes)
1579 {
1580         int ret;
1581         u64 map_length = num_bytes;
1582         struct btrfs_multi_bio *multi = NULL;
1583
1584         if (!btrfs_test_opt(root, DISCARD))
1585                 return 0;
1586
1587         /* Tell the block device(s) that the sectors can be discarded */
1588         ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
1589                               bytenr, &map_length, &multi, 0);
1590         if (!ret) {
1591                 struct btrfs_bio_stripe *stripe = multi->stripes;
1592                 int i;
1593
1594                 if (map_length > num_bytes)
1595                         map_length = num_bytes;
1596
1597                 for (i = 0; i < multi->num_stripes; i++, stripe++) {
1598                         btrfs_issue_discard(stripe->dev->bdev,
1599                                             stripe->physical,
1600                                             map_length);
1601                 }
1602                 kfree(multi);
1603         }
1604
1605         return ret;
1606 }
1607
1608 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1609                          struct btrfs_root *root,
1610                          u64 bytenr, u64 num_bytes, u64 parent,
1611                          u64 root_objectid, u64 owner, u64 offset)
1612 {
1613         int ret;
1614         BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1615                root_objectid == BTRFS_TREE_LOG_OBJECTID);
1616
1617         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1618                 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
1619                                         parent, root_objectid, (int)owner,
1620                                         BTRFS_ADD_DELAYED_REF, NULL);
1621         } else {
1622                 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
1623                                         parent, root_objectid, owner, offset,
1624                                         BTRFS_ADD_DELAYED_REF, NULL);
1625         }
1626         return ret;
1627 }
1628
1629 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1630                                   struct btrfs_root *root,
1631                                   u64 bytenr, u64 num_bytes,
1632                                   u64 parent, u64 root_objectid,
1633                                   u64 owner, u64 offset, int refs_to_add,
1634                                   struct btrfs_delayed_extent_op *extent_op)
1635 {
1636         struct btrfs_path *path;
1637         struct extent_buffer *leaf;
1638         struct btrfs_extent_item *item;
1639         u64 refs;
1640         int ret;
1641         int err = 0;
1642
1643         path = btrfs_alloc_path();
1644         if (!path)
1645                 return -ENOMEM;
1646
1647         path->reada = 1;
1648         path->leave_spinning = 1;
1649         /* this will setup the path even if it fails to insert the back ref */
1650         ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1651                                            path, bytenr, num_bytes, parent,
1652                                            root_objectid, owner, offset,
1653                                            refs_to_add, extent_op);
1654         if (ret == 0)
1655                 goto out;
1656
1657         if (ret != -EAGAIN) {
1658                 err = ret;
1659                 goto out;
1660         }
1661
1662         leaf = path->nodes[0];
1663         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1664         refs = btrfs_extent_refs(leaf, item);
1665         btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1666         if (extent_op)
1667                 __run_delayed_extent_op(extent_op, leaf, item);
1668
1669         btrfs_mark_buffer_dirty(leaf);
1670         btrfs_release_path(root->fs_info->extent_root, path);
1671
1672         path->reada = 1;
1673         path->leave_spinning = 1;
1674
1675         /* now insert the actual backref */
1676         ret = insert_extent_backref(trans, root->fs_info->extent_root,
1677                                     path, bytenr, parent, root_objectid,
1678                                     owner, offset, refs_to_add);
1679         BUG_ON(ret);
1680 out:
1681         btrfs_free_path(path);
1682         return err;
1683 }
1684
1685 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1686                                 struct btrfs_root *root,
1687                                 struct btrfs_delayed_ref_node *node,
1688                                 struct btrfs_delayed_extent_op *extent_op,
1689                                 int insert_reserved)
1690 {
1691         int ret = 0;
1692         struct btrfs_delayed_data_ref *ref;
1693         struct btrfs_key ins;
1694         u64 parent = 0;
1695         u64 ref_root = 0;
1696         u64 flags = 0;
1697
1698         ins.objectid = node->bytenr;
1699         ins.offset = node->num_bytes;
1700         ins.type = BTRFS_EXTENT_ITEM_KEY;
1701
1702         ref = btrfs_delayed_node_to_data_ref(node);
1703         if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1704                 parent = ref->parent;
1705         else
1706                 ref_root = ref->root;
1707
1708         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1709                 if (extent_op) {
1710                         BUG_ON(extent_op->update_key);
1711                         flags |= extent_op->flags_to_set;
1712                 }
1713                 ret = alloc_reserved_file_extent(trans, root,
1714                                                  parent, ref_root, flags,
1715                                                  ref->objectid, ref->offset,
1716                                                  &ins, node->ref_mod);
1717         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1718                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1719                                              node->num_bytes, parent,
1720                                              ref_root, ref->objectid,
1721                                              ref->offset, node->ref_mod,
1722                                              extent_op);
1723         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1724                 ret = __btrfs_free_extent(trans, root, node->bytenr,
1725                                           node->num_bytes, parent,
1726                                           ref_root, ref->objectid,
1727                                           ref->offset, node->ref_mod,
1728                                           extent_op);
1729         } else {
1730                 BUG();
1731         }
1732         return ret;
1733 }
1734
1735 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1736                                     struct extent_buffer *leaf,
1737                                     struct btrfs_extent_item *ei)
1738 {
1739         u64 flags = btrfs_extent_flags(leaf, ei);
1740         if (extent_op->update_flags) {
1741                 flags |= extent_op->flags_to_set;
1742                 btrfs_set_extent_flags(leaf, ei, flags);
1743         }
1744
1745         if (extent_op->update_key) {
1746                 struct btrfs_tree_block_info *bi;
1747                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
1748                 bi = (struct btrfs_tree_block_info *)(ei + 1);
1749                 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
1750         }
1751 }
1752
1753 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
1754                                  struct btrfs_root *root,
1755                                  struct btrfs_delayed_ref_node *node,
1756                                  struct btrfs_delayed_extent_op *extent_op)
1757 {
1758         struct btrfs_key key;
1759         struct btrfs_path *path;
1760         struct btrfs_extent_item *ei;
1761         struct extent_buffer *leaf;
1762         u32 item_size;
1763         int ret;
1764         int err = 0;
1765
1766         path = btrfs_alloc_path();
1767         if (!path)
1768                 return -ENOMEM;
1769
1770         key.objectid = node->bytenr;
1771         key.type = BTRFS_EXTENT_ITEM_KEY;
1772         key.offset = node->num_bytes;
1773
1774         path->reada = 1;
1775         path->leave_spinning = 1;
1776         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
1777                                 path, 0, 1);
1778         if (ret < 0) {
1779                 err = ret;
1780                 goto out;
1781         }
1782         if (ret > 0) {
1783                 err = -EIO;
1784                 goto out;
1785         }
1786
1787         leaf = path->nodes[0];
1788         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1789 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1790         if (item_size < sizeof(*ei)) {
1791                 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
1792                                              path, (u64)-1, 0);
1793                 if (ret < 0) {
1794                         err = ret;
1795                         goto out;
1796                 }
1797                 leaf = path->nodes[0];
1798                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1799         }
1800 #endif
1801         BUG_ON(item_size < sizeof(*ei));
1802         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1803         __run_delayed_extent_op(extent_op, leaf, ei);
1804
1805         btrfs_mark_buffer_dirty(leaf);
1806 out:
1807         btrfs_free_path(path);
1808         return err;
1809 }
1810
1811 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
1812                                 struct btrfs_root *root,
1813                                 struct btrfs_delayed_ref_node *node,
1814                                 struct btrfs_delayed_extent_op *extent_op,
1815                                 int insert_reserved)
1816 {
1817         int ret = 0;
1818         struct btrfs_delayed_tree_ref *ref;
1819         struct btrfs_key ins;
1820         u64 parent = 0;
1821         u64 ref_root = 0;
1822
1823         ins.objectid = node->bytenr;
1824         ins.offset = node->num_bytes;
1825         ins.type = BTRFS_EXTENT_ITEM_KEY;
1826
1827         ref = btrfs_delayed_node_to_tree_ref(node);
1828         if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1829                 parent = ref->parent;
1830         else
1831                 ref_root = ref->root;
1832
1833         BUG_ON(node->ref_mod != 1);
1834         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1835                 BUG_ON(!extent_op || !extent_op->update_flags ||
1836                        !extent_op->update_key);
1837                 ret = alloc_reserved_tree_block(trans, root,
1838                                                 parent, ref_root,
1839                                                 extent_op->flags_to_set,
1840                                                 &extent_op->key,
1841                                                 ref->level, &ins);
1842         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1843                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1844                                              node->num_bytes, parent, ref_root,
1845                                              ref->level, 0, 1, extent_op);
1846         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1847                 ret = __btrfs_free_extent(trans, root, node->bytenr,
1848                                           node->num_bytes, parent, ref_root,
1849                                           ref->level, 0, 1, extent_op);
1850         } else {
1851                 BUG();
1852         }
1853         return ret;
1854 }
1855
1856
1857 /* helper function to actually process a single delayed ref entry */
1858 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
1859                                struct btrfs_root *root,
1860                                struct btrfs_delayed_ref_node *node,
1861                                struct btrfs_delayed_extent_op *extent_op,
1862                                int insert_reserved)
1863 {
1864         int ret;
1865         if (btrfs_delayed_ref_is_head(node)) {
1866                 struct btrfs_delayed_ref_head *head;
1867                 /*
1868                  * we've hit the end of the chain and we were supposed
1869                  * to insert this extent into the tree.  But, it got
1870                  * deleted before we ever needed to insert it, so all
1871                  * we have to do is clean up the accounting
1872                  */
1873                 BUG_ON(extent_op);
1874                 head = btrfs_delayed_node_to_head(node);
1875                 if (insert_reserved) {
1876                         int mark_free = 0;
1877                         struct extent_buffer *must_clean = NULL;
1878
1879                         ret = pin_down_bytes(trans, root, NULL,
1880                                              node->bytenr, node->num_bytes,
1881                                              head->is_data, 1, &must_clean);
1882                         if (ret > 0)
1883                                 mark_free = 1;
1884
1885                         if (must_clean) {
1886                                 clean_tree_block(NULL, root, must_clean);
1887                                 btrfs_tree_unlock(must_clean);
1888                                 free_extent_buffer(must_clean);
1889                         }
1890                         if (head->is_data) {
1891                                 ret = btrfs_del_csums(trans, root,
1892                                                       node->bytenr,
1893                                                       node->num_bytes);
1894                                 BUG_ON(ret);
1895                         }
1896                         if (mark_free) {
1897                                 ret = btrfs_free_reserved_extent(root,
1898                                                         node->bytenr,
1899                                                         node->num_bytes);
1900                                 BUG_ON(ret);
1901                         }
1902                 }
1903                 mutex_unlock(&head->mutex);
1904                 return 0;
1905         }
1906
1907         if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
1908             node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1909                 ret = run_delayed_tree_ref(trans, root, node, extent_op,
1910                                            insert_reserved);
1911         else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
1912                  node->type == BTRFS_SHARED_DATA_REF_KEY)
1913                 ret = run_delayed_data_ref(trans, root, node, extent_op,
1914                                            insert_reserved);
1915         else
1916                 BUG();
1917         return ret;
1918 }
1919
1920 static noinline struct btrfs_delayed_ref_node *
1921 select_delayed_ref(struct btrfs_delayed_ref_head *head)
1922 {
1923         struct rb_node *node;
1924         struct btrfs_delayed_ref_node *ref;
1925         int action = BTRFS_ADD_DELAYED_REF;
1926 again:
1927         /*
1928          * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
1929          * this prevents ref count from going down to zero when
1930          * there still are pending delayed ref.
1931          */
1932         node = rb_prev(&head->node.rb_node);
1933         while (1) {
1934                 if (!node)
1935                         break;
1936                 ref = rb_entry(node, struct btrfs_delayed_ref_node,
1937                                 rb_node);
1938                 if (ref->bytenr != head->node.bytenr)
1939                         break;
1940                 if (ref->action == action)
1941                         return ref;
1942                 node = rb_prev(node);
1943         }
1944         if (action == BTRFS_ADD_DELAYED_REF) {
1945                 action = BTRFS_DROP_DELAYED_REF;
1946                 goto again;
1947         }
1948         return NULL;
1949 }
1950
1951 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
1952                                        struct btrfs_root *root,
1953                                        struct list_head *cluster)
1954 {
1955         struct btrfs_delayed_ref_root *delayed_refs;
1956         struct btrfs_delayed_ref_node *ref;
1957         struct btrfs_delayed_ref_head *locked_ref = NULL;
1958         struct btrfs_delayed_extent_op *extent_op;
1959         int ret;
1960         int count = 0;
1961         int must_insert_reserved = 0;
1962
1963         delayed_refs = &trans->transaction->delayed_refs;
1964         while (1) {
1965                 if (!locked_ref) {
1966                         /* pick a new head ref from the cluster list */
1967                         if (list_empty(cluster))
1968                                 break;
1969
1970                         locked_ref = list_entry(cluster->next,
1971                                      struct btrfs_delayed_ref_head, cluster);
1972
1973                         /* grab the lock that says we are going to process
1974                          * all the refs for this head */
1975                         ret = btrfs_delayed_ref_lock(trans, locked_ref);
1976
1977                         /*
1978                          * we may have dropped the spin lock to get the head
1979                          * mutex lock, and that might have given someone else
1980                          * time to free the head.  If that's true, it has been
1981                          * removed from our list and we can move on.
1982                          */
1983                         if (ret == -EAGAIN) {
1984                                 locked_ref = NULL;
1985                                 count++;
1986                                 continue;
1987                         }
1988                 }
1989
1990                 /*
1991                  * record the must insert reserved flag before we
1992                  * drop the spin lock.
1993                  */
1994                 must_insert_reserved = locked_ref->must_insert_reserved;
1995                 locked_ref->must_insert_reserved = 0;
1996
1997                 extent_op = locked_ref->extent_op;
1998                 locked_ref->extent_op = NULL;
1999
2000                 /*
2001                  * locked_ref is the head node, so we have to go one
2002                  * node back for any delayed ref updates
2003                  */
2004                 ref = select_delayed_ref(locked_ref);
2005                 if (!ref) {
2006                         /* All delayed refs have been processed, Go ahead
2007                          * and send the head node to run_one_delayed_ref,
2008                          * so that any accounting fixes can happen
2009                          */
2010                         ref = &locked_ref->node;
2011
2012                         if (extent_op && must_insert_reserved) {
2013                                 kfree(extent_op);
2014                                 extent_op = NULL;
2015                         }
2016
2017                         if (extent_op) {
2018                                 spin_unlock(&delayed_refs->lock);
2019
2020                                 ret = run_delayed_extent_op(trans, root,
2021                                                             ref, extent_op);
2022                                 BUG_ON(ret);
2023                                 kfree(extent_op);
2024
2025                                 cond_resched();
2026                                 spin_lock(&delayed_refs->lock);
2027                                 continue;
2028                         }
2029
2030                         list_del_init(&locked_ref->cluster);
2031                         locked_ref = NULL;
2032                 }
2033
2034                 ref->in_tree = 0;
2035                 rb_erase(&ref->rb_node, &delayed_refs->root);
2036                 delayed_refs->num_entries--;
2037
2038                 spin_unlock(&delayed_refs->lock);
2039
2040                 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2041                                           must_insert_reserved);
2042                 BUG_ON(ret);
2043
2044                 btrfs_put_delayed_ref(ref);
2045                 kfree(extent_op);
2046                 count++;
2047
2048                 cond_resched();
2049                 spin_lock(&delayed_refs->lock);
2050         }
2051         return count;
2052 }
2053
2054 /*
2055  * this starts processing the delayed reference count updates and
2056  * extent insertions we have queued up so far.  count can be
2057  * 0, which means to process everything in the tree at the start
2058  * of the run (but not newly added entries), or it can be some target
2059  * number you'd like to process.
2060  */
2061 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2062                            struct btrfs_root *root, unsigned long count)
2063 {
2064         struct rb_node *node;
2065         struct btrfs_delayed_ref_root *delayed_refs;
2066         struct btrfs_delayed_ref_node *ref;
2067         struct list_head cluster;
2068         int ret;
2069         int run_all = count == (unsigned long)-1;
2070         int run_most = 0;
2071
2072         if (root == root->fs_info->extent_root)
2073                 root = root->fs_info->tree_root;
2074
2075         delayed_refs = &trans->transaction->delayed_refs;
2076         INIT_LIST_HEAD(&cluster);
2077 again:
2078         spin_lock(&delayed_refs->lock);
2079         if (count == 0) {
2080                 count = delayed_refs->num_entries * 2;
2081                 run_most = 1;
2082         }
2083         while (1) {
2084                 if (!(run_all || run_most) &&
2085                     delayed_refs->num_heads_ready < 64)
2086                         break;
2087
2088                 /*
2089                  * go find something we can process in the rbtree.  We start at
2090                  * the beginning of the tree, and then build a cluster
2091                  * of refs to process starting at the first one we are able to
2092                  * lock
2093                  */
2094                 ret = btrfs_find_ref_cluster(trans, &cluster,
2095                                              delayed_refs->run_delayed_start);
2096                 if (ret)
2097                         break;
2098
2099                 ret = run_clustered_refs(trans, root, &cluster);
2100                 BUG_ON(ret < 0);
2101
2102                 count -= min_t(unsigned long, ret, count);
2103
2104                 if (count == 0)
2105                         break;
2106         }
2107
2108         if (run_all) {
2109                 node = rb_first(&delayed_refs->root);
2110                 if (!node)
2111                         goto out;
2112                 count = (unsigned long)-1;
2113
2114                 while (node) {
2115                         ref = rb_entry(node, struct btrfs_delayed_ref_node,
2116                                        rb_node);
2117                         if (btrfs_delayed_ref_is_head(ref)) {
2118                                 struct btrfs_delayed_ref_head *head;
2119
2120                                 head = btrfs_delayed_node_to_head(ref);
2121                                 atomic_inc(&ref->refs);
2122
2123                                 spin_unlock(&delayed_refs->lock);
2124                                 mutex_lock(&head->mutex);
2125                                 mutex_unlock(&head->mutex);
2126
2127                                 btrfs_put_delayed_ref(ref);
2128                                 cond_resched();
2129                                 goto again;
2130                         }
2131                         node = rb_next(node);
2132                 }
2133                 spin_unlock(&delayed_refs->lock);
2134                 schedule_timeout(1);
2135                 goto again;
2136         }
2137 out:
2138         spin_unlock(&delayed_refs->lock);
2139         return 0;
2140 }
2141
2142 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2143                                 struct btrfs_root *root,
2144                                 u64 bytenr, u64 num_bytes, u64 flags,
2145                                 int is_data)
2146 {
2147         struct btrfs_delayed_extent_op *extent_op;
2148         int ret;
2149
2150         extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2151         if (!extent_op)
2152                 return -ENOMEM;
2153
2154         extent_op->flags_to_set = flags;
2155         extent_op->update_flags = 1;
2156         extent_op->update_key = 0;
2157         extent_op->is_data = is_data ? 1 : 0;
2158
2159         ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op);
2160         if (ret)
2161                 kfree(extent_op);
2162         return ret;
2163 }
2164
2165 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2166                                       struct btrfs_root *root,
2167                                       struct btrfs_path *path,
2168                                       u64 objectid, u64 offset, u64 bytenr)
2169 {
2170         struct btrfs_delayed_ref_head *head;
2171         struct btrfs_delayed_ref_node *ref;
2172         struct btrfs_delayed_data_ref *data_ref;
2173         struct btrfs_delayed_ref_root *delayed_refs;
2174         struct rb_node *node;
2175         int ret = 0;
2176
2177         ret = -ENOENT;
2178         delayed_refs = &trans->transaction->delayed_refs;
2179         spin_lock(&delayed_refs->lock);
2180         head = btrfs_find_delayed_ref_head(trans, bytenr);
2181         if (!head)
2182                 goto out;
2183
2184         if (!mutex_trylock(&head->mutex)) {
2185                 atomic_inc(&head->node.refs);
2186                 spin_unlock(&delayed_refs->lock);
2187
2188                 btrfs_release_path(root->fs_info->extent_root, path);
2189
2190                 mutex_lock(&head->mutex);
2191                 mutex_unlock(&head->mutex);
2192                 btrfs_put_delayed_ref(&head->node);
2193                 return -EAGAIN;
2194         }
2195
2196         node = rb_prev(&head->node.rb_node);
2197         if (!node)
2198                 goto out_unlock;
2199
2200         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2201
2202         if (ref->bytenr != bytenr)
2203                 goto out_unlock;
2204
2205         ret = 1;
2206         if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2207                 goto out_unlock;
2208
2209         data_ref = btrfs_delayed_node_to_data_ref(ref);
2210
2211         node = rb_prev(node);
2212         if (node) {
2213                 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2214                 if (ref->bytenr == bytenr)
2215                         goto out_unlock;
2216         }
2217
2218         if (data_ref->root != root->root_key.objectid ||
2219             data_ref->objectid != objectid || data_ref->offset != offset)
2220                 goto out_unlock;
2221
2222         ret = 0;
2223 out_unlock:
2224         mutex_unlock(&head->mutex);
2225 out:
2226         spin_unlock(&delayed_refs->lock);
2227         return ret;
2228 }
2229
2230 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2231                                         struct btrfs_root *root,
2232                                         struct btrfs_path *path,
2233                                         u64 objectid, u64 offset, u64 bytenr)
2234 {
2235         struct btrfs_root *extent_root = root->fs_info->extent_root;
2236         struct extent_buffer *leaf;
2237         struct btrfs_extent_data_ref *ref;
2238         struct btrfs_extent_inline_ref *iref;
2239         struct btrfs_extent_item *ei;
2240         struct btrfs_key key;
2241         u32 item_size;
2242         int ret;
2243
2244         key.objectid = bytenr;
2245         key.offset = (u64)-1;
2246         key.type = BTRFS_EXTENT_ITEM_KEY;
2247
2248         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2249         if (ret < 0)
2250                 goto out;
2251         BUG_ON(ret == 0);
2252
2253         ret = -ENOENT;
2254         if (path->slots[0] == 0)
2255                 goto out;
2256
2257         path->slots[0]--;
2258         leaf = path->nodes[0];
2259         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2260
2261         if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2262                 goto out;
2263
2264         ret = 1;
2265         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2266 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2267         if (item_size < sizeof(*ei)) {
2268                 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2269                 goto out;
2270         }
2271 #endif
2272         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2273
2274         if (item_size != sizeof(*ei) +
2275             btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2276                 goto out;
2277
2278         if (btrfs_extent_generation(leaf, ei) <=
2279             btrfs_root_last_snapshot(&root->root_item))
2280                 goto out;
2281
2282         iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2283         if (btrfs_extent_inline_ref_type(leaf, iref) !=
2284             BTRFS_EXTENT_DATA_REF_KEY)
2285                 goto out;
2286
2287         ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2288         if (btrfs_extent_refs(leaf, ei) !=
2289             btrfs_extent_data_ref_count(leaf, ref) ||
2290             btrfs_extent_data_ref_root(leaf, ref) !=
2291             root->root_key.objectid ||
2292             btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2293             btrfs_extent_data_ref_offset(leaf, ref) != offset)
2294                 goto out;
2295
2296         ret = 0;
2297 out:
2298         return ret;
2299 }
2300
2301 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2302                           struct btrfs_root *root,
2303                           u64 objectid, u64 offset, u64 bytenr)
2304 {
2305         struct btrfs_path *path;
2306         int ret;
2307         int ret2;
2308
2309         path = btrfs_alloc_path();
2310         if (!path)
2311                 return -ENOENT;
2312
2313         do {
2314                 ret = check_committed_ref(trans, root, path, objectid,
2315                                           offset, bytenr);
2316                 if (ret && ret != -ENOENT)
2317                         goto out;
2318
2319                 ret2 = check_delayed_ref(trans, root, path, objectid,
2320                                          offset, bytenr);
2321         } while (ret2 == -EAGAIN);
2322
2323         if (ret2 && ret2 != -ENOENT) {
2324                 ret = ret2;
2325                 goto out;
2326         }
2327
2328         if (ret != -ENOENT || ret2 != -ENOENT)
2329                 ret = 0;
2330 out:
2331         btrfs_free_path(path);
2332         return ret;
2333 }
2334
2335 #if 0
2336 int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2337                     struct extent_buffer *buf, u32 nr_extents)
2338 {
2339         struct btrfs_key key;
2340         struct btrfs_file_extent_item *fi;
2341         u64 root_gen;
2342         u32 nritems;
2343         int i;
2344         int level;
2345         int ret = 0;
2346         int shared = 0;
2347
2348         if (!root->ref_cows)
2349                 return 0;
2350
2351         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
2352                 shared = 0;
2353                 root_gen = root->root_key.offset;
2354         } else {
2355                 shared = 1;
2356                 root_gen = trans->transid - 1;
2357         }
2358
2359         level = btrfs_header_level(buf);
2360         nritems = btrfs_header_nritems(buf);
2361
2362         if (level == 0) {
2363                 struct btrfs_leaf_ref *ref;
2364                 struct btrfs_extent_info *info;
2365
2366                 ref = btrfs_alloc_leaf_ref(root, nr_extents);
2367                 if (!ref) {
2368                         ret = -ENOMEM;
2369                         goto out;
2370                 }
2371
2372                 ref->root_gen = root_gen;
2373                 ref->bytenr = buf->start;
2374                 ref->owner = btrfs_header_owner(buf);
2375                 ref->generation = btrfs_header_generation(buf);
2376                 ref->nritems = nr_extents;
2377                 info = ref->extents;
2378
2379                 for (i = 0; nr_extents > 0 && i < nritems; i++) {
2380                         u64 disk_bytenr;
2381                         btrfs_item_key_to_cpu(buf, &key, i);
2382                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2383                                 continue;
2384                         fi = btrfs_item_ptr(buf, i,
2385                                             struct btrfs_file_extent_item);
2386                         if (btrfs_file_extent_type(buf, fi) ==
2387                             BTRFS_FILE_EXTENT_INLINE)
2388                                 continue;
2389                         disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2390                         if (disk_bytenr == 0)
2391                                 continue;
2392
2393                         info->bytenr = disk_bytenr;
2394                         info->num_bytes =
2395                                 btrfs_file_extent_disk_num_bytes(buf, fi);
2396                         info->objectid = key.objectid;
2397                         info->offset = key.offset;
2398                         info++;
2399                 }
2400
2401                 ret = btrfs_add_leaf_ref(root, ref, shared);
2402                 if (ret == -EEXIST && shared) {
2403                         struct btrfs_leaf_ref *old;
2404                         old = btrfs_lookup_leaf_ref(root, ref->bytenr);
2405                         BUG_ON(!old);
2406                         btrfs_remove_leaf_ref(root, old);
2407                         btrfs_free_leaf_ref(root, old);
2408                         ret = btrfs_add_leaf_ref(root, ref, shared);
2409                 }
2410                 WARN_ON(ret);
2411                 btrfs_free_leaf_ref(root, ref);
2412         }
2413 out:
2414         return ret;
2415 }
2416
2417 /* when a block goes through cow, we update the reference counts of
2418  * everything that block points to.  The internal pointers of the block
2419  * can be in just about any order, and it is likely to have clusters of
2420  * things that are close together and clusters of things that are not.
2421  *
2422  * To help reduce the seeks that come with updating all of these reference
2423  * counts, sort them by byte number before actual updates are done.
2424  *
2425  * struct refsort is used to match byte number to slot in the btree block.
2426  * we sort based on the byte number and then use the slot to actually
2427  * find the item.
2428  *
2429  * struct refsort is smaller than strcut btrfs_item and smaller than
2430  * struct btrfs_key_ptr.  Since we're currently limited to the page size
2431  * for a btree block, there's no way for a kmalloc of refsorts for a
2432  * single node to be bigger than a page.
2433  */
2434 struct refsort {
2435         u64 bytenr;
2436         u32 slot;
2437 };
2438
2439 /*
2440  * for passing into sort()
2441  */
2442 static int refsort_cmp(const void *a_void, const void *b_void)
2443 {
2444         const struct refsort *a = a_void;
2445         const struct refsort *b = b_void;
2446
2447         if (a->bytenr < b->bytenr)
2448                 return -1;
2449         if (a->bytenr > b->bytenr)
2450                 return 1;
2451         return 0;
2452 }
2453 #endif
2454
2455 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2456                            struct btrfs_root *root,
2457                            struct extent_buffer *buf,
2458                            int full_backref, int inc)
2459 {
2460         u64 bytenr;
2461         u64 num_bytes;
2462         u64 parent;
2463         u64 ref_root;
2464         u32 nritems;
2465         struct btrfs_key key;
2466         struct btrfs_file_extent_item *fi;
2467         int i;
2468         int level;
2469         int ret = 0;
2470         int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2471                             u64, u64, u64, u64, u64, u64);
2472
2473         ref_root = btrfs_header_owner(buf);
2474         nritems = btrfs_header_nritems(buf);
2475         level = btrfs_header_level(buf);
2476
2477         if (!root->ref_cows && level == 0)
2478                 return 0;
2479
2480         if (inc)
2481                 process_func = btrfs_inc_extent_ref;
2482         else
2483                 process_func = btrfs_free_extent;
2484
2485         if (full_backref)
2486                 parent = buf->start;
2487         else
2488                 parent = 0;
2489
2490         for (i = 0; i < nritems; i++) {
2491                 if (level == 0) {
2492                         btrfs_item_key_to_cpu(buf, &key, i);
2493                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2494                                 continue;
2495                         fi = btrfs_item_ptr(buf, i,
2496                                             struct btrfs_file_extent_item);
2497                         if (btrfs_file_extent_type(buf, fi) ==
2498                             BTRFS_FILE_EXTENT_INLINE)
2499                                 continue;
2500                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2501                         if (bytenr == 0)
2502                                 continue;
2503
2504                         num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2505                         key.offset -= btrfs_file_extent_offset(buf, fi);
2506                         ret = process_func(trans, root, bytenr, num_bytes,
2507                                            parent, ref_root, key.objectid,
2508                                            key.offset);
2509                         if (ret)
2510                                 goto fail;
2511                 } else {
2512                         bytenr = btrfs_node_blockptr(buf, i);
2513                         num_bytes = btrfs_level_size(root, level - 1);
2514                         ret = process_func(trans, root, bytenr, num_bytes,
2515                                            parent, ref_root, level - 1, 0);
2516                         if (ret)
2517                                 goto fail;
2518                 }
2519         }
2520         return 0;
2521 fail:
2522         BUG();
2523         return ret;
2524 }
2525
2526 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2527                   struct extent_buffer *buf, int full_backref)
2528 {
2529         return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
2530 }
2531
2532 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2533                   struct extent_buffer *buf, int full_backref)
2534 {
2535         return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
2536 }
2537
2538 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2539                                  struct btrfs_root *root,
2540                                  struct btrfs_path *path,
2541                                  struct btrfs_block_group_cache *cache)
2542 {
2543         int ret;
2544         struct btrfs_root *extent_root = root->fs_info->extent_root;
2545         unsigned long bi;
2546         struct extent_buffer *leaf;
2547
2548         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2549         if (ret < 0)
2550                 goto fail;
2551         BUG_ON(ret);
2552
2553         leaf = path->nodes[0];
2554         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2555         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2556         btrfs_mark_buffer_dirty(leaf);
2557         btrfs_release_path(extent_root, path);
2558 fail:
2559         if (ret)
2560                 return ret;
2561         return 0;
2562
2563 }
2564
2565 static struct btrfs_block_group_cache *
2566 next_block_group(struct btrfs_root *root,
2567                  struct btrfs_block_group_cache *cache)
2568 {
2569         struct rb_node *node;
2570         spin_lock(&root->fs_info->block_group_cache_lock);
2571         node = rb_next(&cache->cache_node);
2572         btrfs_put_block_group(cache);
2573         if (node) {
2574                 cache = rb_entry(node, struct btrfs_block_group_cache,
2575                                  cache_node);
2576                 atomic_inc(&cache->count);
2577         } else
2578                 cache = NULL;
2579         spin_unlock(&root->fs_info->block_group_cache_lock);
2580         return cache;
2581 }
2582
2583 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
2584                                    struct btrfs_root *root)
2585 {
2586         struct btrfs_block_group_cache *cache;
2587         int err = 0;
2588         struct btrfs_path *path;
2589         u64 last = 0;
2590
2591         path = btrfs_alloc_path();
2592         if (!path)
2593                 return -ENOMEM;
2594
2595         while (1) {
2596                 if (last == 0) {
2597                         err = btrfs_run_delayed_refs(trans, root,
2598                                                      (unsigned long)-1);
2599                         BUG_ON(err);
2600                 }
2601
2602                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2603                 while (cache) {
2604                         if (cache->dirty)
2605                                 break;
2606                         cache = next_block_group(root, cache);
2607                 }
2608                 if (!cache) {
2609                         if (last == 0)
2610                                 break;
2611                         last = 0;
2612                         continue;
2613                 }
2614
2615                 cache->dirty = 0;
2616                 last = cache->key.objectid + cache->key.offset;
2617
2618                 err = write_one_cache_group(trans, root, path, cache);
2619                 BUG_ON(err);
2620                 btrfs_put_block_group(cache);
2621         }
2622
2623         btrfs_free_path(path);
2624         return 0;
2625 }
2626
2627 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
2628 {
2629         struct btrfs_block_group_cache *block_group;
2630         int readonly = 0;
2631
2632         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
2633         if (!block_group || block_group->ro)
2634                 readonly = 1;
2635         if (block_group)
2636                 btrfs_put_block_group(block_group);
2637         return readonly;
2638 }
2639
2640 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
2641                              u64 total_bytes, u64 bytes_used,
2642                              struct btrfs_space_info **space_info)
2643 {
2644         struct btrfs_space_info *found;
2645
2646         found = __find_space_info(info, flags);
2647         if (found) {
2648                 spin_lock(&found->lock);
2649                 found->total_bytes += total_bytes;
2650                 found->bytes_used += bytes_used;
2651                 found->full = 0;
2652                 spin_unlock(&found->lock);
2653                 *space_info = found;
2654                 return 0;
2655         }
2656         found = kzalloc(sizeof(*found), GFP_NOFS);
2657         if (!found)
2658                 return -ENOMEM;
2659
2660         INIT_LIST_HEAD(&found->block_groups);
2661         init_rwsem(&found->groups_sem);
2662         spin_lock_init(&found->lock);
2663         found->flags = flags;
2664         found->total_bytes = total_bytes;
2665         found->bytes_used = bytes_used;
2666         found->bytes_pinned = 0;
2667         found->bytes_reserved = 0;
2668         found->bytes_readonly = 0;
2669         found->bytes_delalloc = 0;
2670         found->full = 0;
2671         found->force_alloc = 0;
2672         *space_info = found;
2673         list_add_rcu(&found->list, &info->space_info);
2674         atomic_set(&found->caching_threads, 0);
2675         return 0;
2676 }
2677
2678 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
2679 {
2680         u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
2681                                    BTRFS_BLOCK_GROUP_RAID1 |
2682                                    BTRFS_BLOCK_GROUP_RAID10 |
2683                                    BTRFS_BLOCK_GROUP_DUP);
2684         if (extra_flags) {
2685                 if (flags & BTRFS_BLOCK_GROUP_DATA)
2686                         fs_info->avail_data_alloc_bits |= extra_flags;
2687                 if (flags & BTRFS_BLOCK_GROUP_METADATA)
2688                         fs_info->avail_metadata_alloc_bits |= extra_flags;
2689                 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
2690                         fs_info->avail_system_alloc_bits |= extra_flags;
2691         }
2692 }
2693
2694 static void set_block_group_readonly(struct btrfs_block_group_cache *cache)
2695 {
2696         spin_lock(&cache->space_info->lock);
2697         spin_lock(&cache->lock);
2698         if (!cache->ro) {
2699                 cache->space_info->bytes_readonly += cache->key.offset -
2700                                         btrfs_block_group_used(&cache->item);
2701                 cache->ro = 1;
2702         }
2703         spin_unlock(&cache->lock);
2704         spin_unlock(&cache->space_info->lock);
2705 }
2706
2707 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
2708 {
2709         u64 num_devices = root->fs_info->fs_devices->rw_devices;
2710
2711         if (num_devices == 1)
2712                 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
2713         if (num_devices < 4)
2714                 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
2715
2716         if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
2717             (flags & (BTRFS_BLOCK_GROUP_RAID1 |
2718                       BTRFS_BLOCK_GROUP_RAID10))) {
2719                 flags &= ~BTRFS_BLOCK_GROUP_DUP;
2720         }
2721
2722         if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
2723             (flags & BTRFS_BLOCK_GROUP_RAID10)) {
2724                 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
2725         }
2726
2727         if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
2728             ((flags & BTRFS_BLOCK_GROUP_RAID1) |
2729              (flags & BTRFS_BLOCK_GROUP_RAID10) |
2730              (flags & BTRFS_BLOCK_GROUP_DUP)))
2731                 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
2732         return flags;
2733 }
2734
2735 static u64 btrfs_get_alloc_profile(struct btrfs_root *root, u64 data)
2736 {
2737         struct btrfs_fs_info *info = root->fs_info;
2738         u64 alloc_profile;
2739
2740         if (data) {
2741                 alloc_profile = info->avail_data_alloc_bits &
2742                         info->data_alloc_profile;
2743                 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
2744         } else if (root == root->fs_info->chunk_root) {
2745                 alloc_profile = info->avail_system_alloc_bits &
2746                         info->system_alloc_profile;
2747                 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
2748         } else {
2749                 alloc_profile = info->avail_metadata_alloc_bits &
2750                         info->metadata_alloc_profile;
2751                 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
2752         }
2753
2754         return btrfs_reduce_alloc_profile(root, data);
2755 }
2756
2757 void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
2758 {
2759         u64 alloc_target;
2760
2761         alloc_target = btrfs_get_alloc_profile(root, 1);
2762         BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
2763                                                        alloc_target);
2764 }
2765
2766 static u64 calculate_bytes_needed(struct btrfs_root *root, int num_items)
2767 {
2768         u64 num_bytes;
2769         int level;
2770
2771         level = BTRFS_MAX_LEVEL - 2;
2772         /*
2773          * NOTE: these calculations are absolutely the worst possible case.
2774          * This assumes that _every_ item we insert will require a new leaf, and
2775          * that the tree has grown to its maximum level size.
2776          */
2777
2778         /*
2779          * for every item we insert we could insert both an extent item and a
2780          * extent ref item.  Then for ever item we insert, we will need to cow
2781          * both the original leaf, plus the leaf to the left and right of it.
2782          *
2783          * Unless we are talking about the extent root, then we just want the
2784          * number of items * 2, since we just need the extent item plus its ref.
2785          */
2786         if (root == root->fs_info->extent_root)
2787                 num_bytes = num_items * 2;
2788         else
2789                 num_bytes = (num_items + (2 * num_items)) * 3;
2790
2791         /*
2792          * num_bytes is total number of leaves we could need times the leaf
2793          * size, and then for every leaf we could end up cow'ing 2 nodes per
2794          * level, down to the leaf level.
2795          */
2796         num_bytes = (num_bytes * root->leafsize) +
2797                 (num_bytes * (level * 2)) * root->nodesize;
2798
2799         return num_bytes;
2800 }
2801
2802 /*
2803  * Unreserve metadata space for delalloc.  If we have less reserved credits than
2804  * we have extents, this function does nothing.
2805  */
2806 int btrfs_unreserve_metadata_for_delalloc(struct btrfs_root *root,
2807                                           struct inode *inode, int num_items)
2808 {
2809         struct btrfs_fs_info *info = root->fs_info;
2810         struct btrfs_space_info *meta_sinfo;
2811         u64 num_bytes;
2812         u64 alloc_target;
2813         bool bug = false;
2814
2815         /* get the space info for where the metadata will live */
2816         alloc_target = btrfs_get_alloc_profile(root, 0);
2817         meta_sinfo = __find_space_info(info, alloc_target);
2818
2819         num_bytes = calculate_bytes_needed(root->fs_info->extent_root,
2820                                            num_items);
2821
2822         spin_lock(&meta_sinfo->lock);
2823         spin_lock(&BTRFS_I(inode)->accounting_lock);
2824         if (BTRFS_I(inode)->reserved_extents <=
2825             BTRFS_I(inode)->outstanding_extents) {
2826                 spin_unlock(&BTRFS_I(inode)->accounting_lock);
2827                 spin_unlock(&meta_sinfo->lock);
2828                 return 0;
2829         }
2830         spin_unlock(&BTRFS_I(inode)->accounting_lock);
2831
2832         BTRFS_I(inode)->reserved_extents--;
2833         BUG_ON(BTRFS_I(inode)->reserved_extents < 0);
2834
2835         if (meta_sinfo->bytes_delalloc < num_bytes) {
2836                 bug = true;
2837                 meta_sinfo->bytes_delalloc = 0;
2838         } else {
2839                 meta_sinfo->bytes_delalloc -= num_bytes;
2840         }
2841         spin_unlock(&meta_sinfo->lock);
2842
2843         BUG_ON(bug);
2844
2845         return 0;
2846 }
2847
2848 static void check_force_delalloc(struct btrfs_space_info *meta_sinfo)
2849 {
2850         u64 thresh;
2851
2852         thresh = meta_sinfo->bytes_used + meta_sinfo->bytes_reserved +
2853                 meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly +
2854                 meta_sinfo->bytes_super + meta_sinfo->bytes_root +
2855                 meta_sinfo->bytes_may_use;
2856
2857         thresh = meta_sinfo->total_bytes - thresh;
2858         thresh *= 80;
2859         do_div(thresh, 100);
2860         if (thresh <= meta_sinfo->bytes_delalloc)
2861                 meta_sinfo->force_delalloc = 1;
2862         else
2863                 meta_sinfo->force_delalloc = 0;
2864 }
2865
2866 struct async_flush {
2867         struct btrfs_root *root;
2868         struct btrfs_space_info *info;
2869         struct btrfs_work work;
2870 };
2871
2872 static noinline void flush_delalloc_async(struct btrfs_work *work)
2873 {
2874         struct async_flush *async;
2875         struct btrfs_root *root;
2876         struct btrfs_space_info *info;
2877
2878         async = container_of(work, struct async_flush, work);
2879         root = async->root;
2880         info = async->info;
2881
2882         btrfs_start_delalloc_inodes(root);
2883         wake_up(&info->flush_wait);
2884         btrfs_wait_ordered_extents(root, 0);
2885
2886         spin_lock(&info->lock);
2887         info->flushing = 0;
2888         spin_unlock(&info->lock);
2889         wake_up(&info->flush_wait);
2890
2891         kfree(async);
2892 }
2893
2894 static void wait_on_flush(struct btrfs_space_info *info)
2895 {
2896         DEFINE_WAIT(wait);
2897         u64 used;
2898
2899         while (1) {
2900                 prepare_to_wait(&info->flush_wait, &wait,
2901                                 TASK_UNINTERRUPTIBLE);
2902                 spin_lock(&info->lock);
2903                 if (!info->flushing) {
2904                         spin_unlock(&info->lock);
2905                         break;
2906                 }
2907
2908                 used = info->bytes_used + info->bytes_reserved +
2909                         info->bytes_pinned + info->bytes_readonly +
2910                         info->bytes_super + info->bytes_root +
2911                         info->bytes_may_use + info->bytes_delalloc;
2912                 if (used < info->total_bytes) {
2913                         spin_unlock(&info->lock);
2914                         break;
2915                 }
2916                 spin_unlock(&info->lock);
2917                 schedule();
2918         }
2919         finish_wait(&info->flush_wait, &wait);
2920 }
2921
2922 static void flush_delalloc(struct btrfs_root *root,
2923                                  struct btrfs_space_info *info)
2924 {
2925         struct async_flush *async;
2926         bool wait = false;
2927
2928         spin_lock(&info->lock);
2929
2930         if (!info->flushing) {
2931                 info->flushing = 1;
2932                 init_waitqueue_head(&info->flush_wait);
2933         } else {
2934                 wait = true;
2935         }
2936
2937         spin_unlock(&info->lock);
2938
2939         if (wait) {
2940                 wait_on_flush(info);
2941                 return;
2942         }
2943
2944         async = kzalloc(sizeof(*async), GFP_NOFS);
2945         if (!async)
2946                 goto flush;
2947
2948         async->root = root;
2949         async->info = info;
2950         async->work.func = flush_delalloc_async;
2951
2952         btrfs_queue_worker(&root->fs_info->enospc_workers,
2953                            &async->work);
2954         wait_on_flush(info);
2955         return;
2956
2957 flush:
2958         btrfs_start_delalloc_inodes(root);
2959         btrfs_wait_ordered_extents(root, 0);
2960
2961         spin_lock(&info->lock);
2962         info->flushing = 0;
2963         spin_unlock(&info->lock);
2964         wake_up(&info->flush_wait);
2965 }
2966
2967 static int maybe_allocate_chunk(struct btrfs_root *root,
2968                                  struct btrfs_space_info *info)
2969 {
2970         struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
2971         struct btrfs_trans_handle *trans;
2972         bool wait = false;
2973         int ret = 0;
2974         u64 min_metadata;
2975         u64 free_space;
2976
2977         free_space = btrfs_super_total_bytes(disk_super);
2978         /*
2979          * we allow the metadata to grow to a max of either 5gb or 5% of the
2980          * space in the volume.
2981          */
2982         min_metadata = min((u64)5 * 1024 * 1024 * 1024,
2983                              div64_u64(free_space * 5, 100));
2984         if (info->total_bytes >= min_metadata) {
2985                 spin_unlock(&info->lock);
2986                 return 0;
2987         }
2988
2989         if (info->full) {
2990                 spin_unlock(&info->lock);
2991                 return 0;
2992         }
2993
2994         if (!info->allocating_chunk) {
2995                 info->force_alloc = 1;
2996                 info->allocating_chunk = 1;
2997                 init_waitqueue_head(&info->allocate_wait);
2998         } else {
2999                 wait = true;
3000         }
3001
3002         spin_unlock(&info->lock);
3003
3004         if (wait) {
3005                 wait_event(info->allocate_wait,
3006                            !info->allocating_chunk);
3007                 return 1;
3008         }
3009
3010         trans = btrfs_start_transaction(root, 1);
3011         if (!trans) {
3012                 ret = -ENOMEM;
3013                 goto out;
3014         }
3015
3016         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3017                              4096 + 2 * 1024 * 1024,
3018                              info->flags, 0);
3019         btrfs_end_transaction(trans, root);
3020         if (ret)
3021                 goto out;
3022 out:
3023         spin_lock(&info->lock);
3024         info->allocating_chunk = 0;
3025         spin_unlock(&info->lock);
3026         wake_up(&info->allocate_wait);
3027
3028         if (ret)
3029                 return 0;
3030         return 1;
3031 }
3032
3033 /*
3034  * Reserve metadata space for delalloc.
3035  */
3036 int btrfs_reserve_metadata_for_delalloc(struct btrfs_root *root,
3037                                         struct inode *inode, int num_items)
3038 {
3039         struct btrfs_fs_info *info = root->fs_info;
3040         struct btrfs_space_info *meta_sinfo;
3041         u64 num_bytes;
3042         u64 used;
3043         u64 alloc_target;
3044         int flushed = 0;
3045         int force_delalloc;
3046
3047         /* get the space info for where the metadata will live */
3048         alloc_target = btrfs_get_alloc_profile(root, 0);
3049         meta_sinfo = __find_space_info(info, alloc_target);
3050
3051         num_bytes = calculate_bytes_needed(root->fs_info->extent_root,
3052                                            num_items);
3053 again:
3054         spin_lock(&meta_sinfo->lock);
3055
3056         force_delalloc = meta_sinfo->force_delalloc;
3057
3058         if (unlikely(!meta_sinfo->bytes_root))
3059                 meta_sinfo->bytes_root = calculate_bytes_needed(root, 6);
3060
3061         if (!flushed)
3062                 meta_sinfo->bytes_delalloc += num_bytes;
3063
3064         used = meta_sinfo->bytes_used + meta_sinfo->bytes_reserved +
3065                 meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly +
3066                 meta_sinfo->bytes_super + meta_sinfo->bytes_root +
3067                 meta_sinfo->bytes_may_use + meta_sinfo->bytes_delalloc;
3068
3069         if (used > meta_sinfo->total_bytes) {
3070                 flushed++;
3071
3072                 if (flushed == 1) {
3073                         if (maybe_allocate_chunk(root, meta_sinfo))
3074                                 goto again;
3075                         flushed++;
3076                 } else {
3077                         spin_unlock(&meta_sinfo->lock);
3078                 }
3079
3080                 if (flushed == 2) {
3081                         filemap_flush(inode->i_mapping);
3082                         goto again;
3083                 } else if (flushed == 3) {
3084                         flush_delalloc(root, meta_sinfo);
3085                         goto again;
3086                 }
3087                 spin_lock(&meta_sinfo->lock);
3088                 meta_sinfo->bytes_delalloc -= num_bytes;
3089                 spin_unlock(&meta_sinfo->lock);
3090                 printk(KERN_ERR "enospc, has %d, reserved %d\n",
3091                        BTRFS_I(inode)->outstanding_extents,
3092                        BTRFS_I(inode)->reserved_extents);
3093                 dump_space_info(meta_sinfo, 0, 0);
3094                 return -ENOSPC;
3095         }
3096
3097         BTRFS_I(inode)->reserved_extents++;
3098         check_force_delalloc(meta_sinfo);
3099         spin_unlock(&meta_sinfo->lock);
3100
3101         if (!flushed && force_delalloc)
3102                 filemap_flush(inode->i_mapping);
3103
3104         return 0;
3105 }
3106
3107 /*
3108  * unreserve num_items number of items worth of metadata space.  This needs to
3109  * be paired with btrfs_reserve_metadata_space.
3110  *
3111  * NOTE: if you have the option, run this _AFTER_ you do a
3112  * btrfs_end_transaction, since btrfs_end_transaction will run delayed ref
3113  * oprations which will result in more used metadata, so we want to make sure we
3114  * can do that without issue.
3115  */
3116 int btrfs_unreserve_metadata_space(struct btrfs_root *root, int num_items)
3117 {
3118         struct btrfs_fs_info *info = root->fs_info;
3119         struct btrfs_space_info *meta_sinfo;
3120         u64 num_bytes;
3121         u64 alloc_target;
3122         bool bug = false;
3123
3124         /* get the space info for where the metadata will live */
3125         alloc_target = btrfs_get_alloc_profile(root, 0);
3126         meta_sinfo = __find_space_info(info, alloc_target);
3127
3128         num_bytes = calculate_bytes_needed(root, num_items);
3129
3130         spin_lock(&meta_sinfo->lock);
3131         if (meta_sinfo->bytes_may_use < num_bytes) {
3132                 bug = true;
3133                 meta_sinfo->bytes_may_use = 0;
3134         } else {
3135                 meta_sinfo->bytes_may_use -= num_bytes;
3136         }
3137         spin_unlock(&meta_sinfo->lock);
3138
3139         BUG_ON(bug);
3140
3141         return 0;
3142 }
3143
3144 /*
3145  * Reserve some metadata space for use.  We'll calculate the worste case number
3146  * of bytes that would be needed to modify num_items number of items.  If we
3147  * have space, fantastic, if not, you get -ENOSPC.  Please call
3148  * btrfs_unreserve_metadata_space when you are done for the _SAME_ number of
3149  * items you reserved, since whatever metadata you needed should have already
3150  * been allocated.
3151  *
3152  * This will commit the transaction to make more space if we don't have enough
3153  * metadata space.  THe only time we don't do this is if we're reserving space
3154  * inside of a transaction, then we will just return -ENOSPC and it is the
3155  * callers responsibility to handle it properly.
3156  */
3157 int btrfs_reserve_metadata_space(struct btrfs_root *root, int num_items)
3158 {
3159         struct btrfs_fs_info *info = root->fs_info;
3160         struct btrfs_space_info *meta_sinfo;
3161         u64 num_bytes;
3162         u64 used;
3163         u64 alloc_target;
3164         int retries = 0;
3165
3166         /* get the space info for where the metadata will live */
3167         alloc_target = btrfs_get_alloc_profile(root, 0);
3168         meta_sinfo = __find_space_info(info, alloc_target);
3169
3170         num_bytes = calculate_bytes_needed(root, num_items);
3171 again:
3172         spin_lock(&meta_sinfo->lock);
3173
3174         if (unlikely(!meta_sinfo->bytes_root))
3175                 meta_sinfo->bytes_root = calculate_bytes_needed(root, 6);
3176
3177         if (!retries)
3178                 meta_sinfo->bytes_may_use += num_bytes;
3179
3180         used = meta_sinfo->bytes_used + meta_sinfo->bytes_reserved +
3181                 meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly +
3182                 meta_sinfo->bytes_super + meta_sinfo->bytes_root +
3183                 meta_sinfo->bytes_may_use + meta_sinfo->bytes_delalloc;
3184
3185         if (used > meta_sinfo->total_bytes) {
3186                 retries++;
3187                 if (retries == 1) {
3188                         if (maybe_allocate_chunk(root, meta_sinfo))
3189                                 goto again;
3190                         retries++;
3191                 } else {
3192                         spin_unlock(&meta_sinfo->lock);
3193                 }
3194
3195                 if (retries == 2) {
3196                         flush_delalloc(root, meta_sinfo);
3197                         goto again;
3198                 }
3199                 spin_lock(&meta_sinfo->lock);
3200                 meta_sinfo->bytes_may_use -= num_bytes;
3201                 spin_unlock(&meta_sinfo->lock);
3202
3203                 dump_space_info(meta_sinfo, 0, 0);
3204                 return -ENOSPC;
3205         }
3206
3207         check_force_delalloc(meta_sinfo);
3208         spin_unlock(&meta_sinfo->lock);
3209
3210         return 0;
3211 }
3212
3213 /*
3214  * This will check the space that the inode allocates from to make sure we have
3215  * enough space for bytes.
3216  */
3217 int btrfs_check_data_free_space(struct btrfs_root *root, struct inode *inode,
3218                                 u64 bytes)
3219 {
3220         struct btrfs_space_info *data_sinfo;
3221         int ret = 0, committed = 0;
3222
3223         /* make sure bytes are sectorsize aligned */
3224         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3225
3226         data_sinfo = BTRFS_I(inode)->space_info;
3227         if (!data_sinfo)
3228                 goto alloc;
3229
3230 again:
3231         /* make sure we have enough space to handle the data first */
3232         spin_lock(&data_sinfo->lock);
3233         if (data_sinfo->total_bytes - data_sinfo->bytes_used -
3234             data_sinfo->bytes_delalloc - data_sinfo->bytes_reserved -
3235             data_sinfo->bytes_pinned - data_sinfo->bytes_readonly -
3236             data_sinfo->bytes_may_use - data_sinfo->bytes_super < bytes) {
3237                 struct btrfs_trans_handle *trans;
3238
3239                 /*
3240                  * if we don't have enough free bytes in this space then we need
3241                  * to alloc a new chunk.
3242                  */
3243                 if (!data_sinfo->full) {
3244                         u64 alloc_target;
3245
3246                         data_sinfo->force_alloc = 1;
3247                         spin_unlock(&data_sinfo->lock);
3248 alloc:
3249                         alloc_target = btrfs_get_alloc_profile(root, 1);
3250                         trans = btrfs_start_transaction(root, 1);
3251                         if (!trans)
3252                                 return -ENOMEM;
3253
3254                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3255                                              bytes + 2 * 1024 * 1024,
3256                                              alloc_target, 0);
3257                         btrfs_end_transaction(trans, root);
3258                         if (ret)
3259                                 return ret;
3260
3261                         if (!data_sinfo) {
3262                                 btrfs_set_inode_space_info(root, inode);
3263                                 data_sinfo = BTRFS_I(inode)->space_info;
3264                         }
3265                         goto again;
3266                 }
3267                 spin_unlock(&data_sinfo->lock);
3268
3269                 /* commit the current transaction and try again */
3270                 if (!committed && !root->fs_info->open_ioctl_trans) {
3271                         committed = 1;
3272                         trans = btrfs_join_transaction(root, 1);
3273                         if (!trans)
3274                                 return -ENOMEM;
3275                         ret = btrfs_commit_transaction(trans, root);
3276                         if (ret)
3277                                 return ret;
3278                         goto again;
3279                 }
3280
3281                 printk(KERN_ERR "no space left, need %llu, %llu delalloc bytes"
3282                        ", %llu bytes_used, %llu bytes_reserved, "
3283                        "%llu bytes_pinned, %llu bytes_readonly, %llu may use "
3284                        "%llu total\n", (unsigned long long)bytes,
3285                        (unsigned long long)data_sinfo->bytes_delalloc,
3286                        (unsigned long long)data_sinfo->bytes_used,
3287                        (unsigned long long)data_sinfo->bytes_reserved,
3288                        (unsigned long long)data_sinfo->bytes_pinned,
3289                        (unsigned long long)data_sinfo->bytes_readonly,
3290                        (unsigned long long)data_sinfo->bytes_may_use,
3291                        (unsigned long long)data_sinfo->total_bytes);
3292                 return -ENOSPC;
3293         }
3294         data_sinfo->bytes_may_use += bytes;
3295         BTRFS_I(inode)->reserved_bytes += bytes;
3296         spin_unlock(&data_sinfo->lock);
3297
3298         return 0;
3299 }
3300
3301 /*
3302  * if there was an error for whatever reason after calling
3303  * btrfs_check_data_free_space, call this so we can cleanup the counters.
3304  */
3305 void btrfs_free_reserved_data_space(struct btrfs_root *root,
3306                                     struct inode *inode, u64 bytes)
3307 {
3308         struct btrfs_space_info *data_sinfo;
3309
3310         /* make sure bytes are sectorsize aligned */
3311         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3312
3313         data_sinfo = BTRFS_I(inode)->space_info;
3314         spin_lock(&data_sinfo->lock);
3315         data_sinfo->bytes_may_use -= bytes;
3316         BTRFS_I(inode)->reserved_bytes -= bytes;
3317         spin_unlock(&data_sinfo->lock);
3318 }
3319
3320 /* called when we are adding a delalloc extent to the inode's io_tree */
3321 void btrfs_delalloc_reserve_space(struct btrfs_root *root, struct inode *inode,
3322                                   u64 bytes)
3323 {
3324         struct btrfs_space_info *data_sinfo;
3325
3326         /* get the space info for where this inode will be storing its data */
3327         data_sinfo = BTRFS_I(inode)->space_info;
3328
3329         /* make sure we have enough space to handle the data first */
3330         spin_lock(&data_sinfo->lock);
3331         data_sinfo->bytes_delalloc += bytes;
3332
3333         /*
3334          * we are adding a delalloc extent without calling
3335          * btrfs_check_data_free_space first.  This happens on a weird
3336          * writepage condition, but shouldn't hurt our accounting
3337          */
3338         if (unlikely(bytes > BTRFS_I(inode)->reserved_bytes)) {
3339                 data_sinfo->bytes_may_use -= BTRFS_I(inode)->reserved_bytes;
3340                 BTRFS_I(inode)->reserved_bytes = 0;
3341         } else {
3342                 data_sinfo->bytes_may_use -= bytes;
3343                 BTRFS_I(inode)->reserved_bytes -= bytes;
3344         }
3345
3346         spin_unlock(&data_sinfo->lock);
3347 }
3348
3349 /* called when we are clearing an delalloc extent from the inode's io_tree */
3350 void btrfs_delalloc_free_space(struct btrfs_root *root, struct inode *inode,
3351                               u64 bytes)
3352 {
3353         struct btrfs_space_info *info;
3354
3355         info = BTRFS_I(inode)->space_info;
3356
3357         spin_lock(&info->lock);
3358         info->bytes_delalloc -= bytes;
3359         spin_unlock(&info->lock);
3360 }
3361
3362 static void force_metadata_allocation(struct btrfs_fs_info *info)
3363 {
3364         struct list_head *head = &info->space_info;
3365         struct btrfs_space_info *found;
3366
3367         rcu_read_lock();
3368         list_for_each_entry_rcu(found, head, list) {
3369                 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3370                         found->force_alloc = 1;
3371         }
3372         rcu_read_unlock();
3373 }
3374
3375 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3376                           struct btrfs_root *extent_root, u64 alloc_bytes,
3377                           u64 flags, int force)
3378 {
3379         struct btrfs_space_info *space_info;
3380         struct btrfs_fs_info *fs_info = extent_root->fs_info;
3381         u64 thresh;
3382         int ret = 0;
3383
3384         mutex_lock(&fs_info->chunk_mutex);
3385
3386         flags = btrfs_reduce_alloc_profile(extent_root, flags);
3387
3388         space_info = __find_space_info(extent_root->fs_info, flags);
3389         if (!space_info) {
3390                 ret = update_space_info(extent_root->fs_info, flags,
3391                                         0, 0, &space_info);
3392                 BUG_ON(ret);
3393         }
3394         BUG_ON(!space_info);
3395
3396         spin_lock(&space_info->lock);
3397         if (space_info->force_alloc)
3398                 force = 1;
3399         if (space_info->full) {
3400                 spin_unlock(&space_info->lock);
3401                 goto out;
3402         }
3403
3404         thresh = space_info->total_bytes - space_info->bytes_readonly;
3405         thresh = div_factor(thresh, 8);
3406         if (!force &&
3407            (space_info->bytes_used + space_info->bytes_pinned +
3408             space_info->bytes_reserved + alloc_bytes) < thresh) {
3409                 spin_unlock(&space_info->lock);
3410                 goto out;
3411         }
3412         spin_unlock(&space_info->lock);
3413
3414         /*
3415          * if we're doing a data chunk, go ahead and make sure that
3416          * we keep a reasonable number of metadata chunks allocated in the
3417          * FS as well.
3418          */
3419         if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3420                 fs_info->data_chunk_allocations++;
3421                 if (!(fs_info->data_chunk_allocations %
3422                       fs_info->metadata_ratio))
3423                         force_metadata_allocation(fs_info);
3424         }
3425
3426         ret = btrfs_alloc_chunk(trans, extent_root, flags);
3427         spin_lock(&space_info->lock);
3428         if (ret)
3429                 space_info->full = 1;
3430         space_info->force_alloc = 0;
3431         spin_unlock(&space_info->lock);
3432 out:
3433         mutex_unlock(&extent_root->fs_info->chunk_mutex);
3434         return ret;
3435 }
3436
3437 static int update_block_group(struct btrfs_trans_handle *trans,
3438                               struct btrfs_root *root,
3439                               u64 bytenr, u64 num_bytes, int alloc,
3440                               int mark_free)
3441 {
3442         struct btrfs_block_group_cache *cache;
3443         struct btrfs_fs_info *info = root->fs_info;
3444         u64 total = num_bytes;
3445         u64 old_val;
3446         u64 byte_in_group;
3447
3448         /* block accounting for super block */
3449         spin_lock(&info->delalloc_lock);
3450         old_val = btrfs_super_bytes_used(&info->super_copy);
3451         if (alloc)
3452                 old_val += num_bytes;
3453         else
3454                 old_val -= num_bytes;
3455         btrfs_set_super_bytes_used(&info->super_copy, old_val);
3456
3457         /* block accounting for root item */
3458         old_val = btrfs_root_used(&root->root_item);
3459         if (alloc)
3460                 old_val += num_bytes;
3461         else
3462                 old_val -= num_bytes;
3463         btrfs_set_root_used(&root->root_item, old_val);
3464         spin_unlock(&info->delalloc_lock);
3465
3466         while (total) {
3467                 cache = btrfs_lookup_block_group(info, bytenr);
3468                 if (!cache)
3469                         return -1;
3470                 byte_in_group = bytenr - cache->key.objectid;
3471                 WARN_ON(byte_in_group > cache->key.offset);
3472
3473                 spin_lock(&cache->space_info->lock);
3474                 spin_lock(&cache->lock);
3475                 cache->dirty = 1;
3476                 old_val = btrfs_block_group_used(&cache->item);
3477                 num_bytes = min(total, cache->key.offset - byte_in_group);
3478                 if (alloc) {
3479                         old_val += num_bytes;
3480                         btrfs_set_block_group_used(&cache->item, old_val);
3481                         cache->reserved -= num_bytes;
3482                         cache->space_info->bytes_used += num_bytes;
3483                         cache->space_info->bytes_reserved -= num_bytes;
3484                         if (cache->ro)
3485                                 cache->space_info->bytes_readonly -= num_bytes;
3486                         spin_unlock(&cache->lock);
3487                         spin_unlock(&cache->space_info->lock);
3488                 } else {
3489                         old_val -= num_bytes;
3490                         cache->space_info->bytes_used -= num_bytes;
3491                         if (cache->ro)
3492                                 cache->space_info->bytes_readonly += num_bytes;
3493                         btrfs_set_block_group_used(&cache->item, old_val);
3494                         spin_unlock(&cache->lock);
3495                         spin_unlock(&cache->space_info->lock);
3496                         if (mark_free) {
3497                                 int ret;
3498
3499                                 ret = btrfs_discard_extent(root, bytenr,
3500                                                            num_bytes);
3501                                 WARN_ON(ret);
3502
3503                                 ret = btrfs_add_free_space(cache, bytenr,
3504                                                            num_bytes);
3505                                 WARN_ON(ret);
3506                         }
3507                 }
3508                 btrfs_put_block_group(cache);
3509                 total -= num_bytes;
3510                 bytenr += num_bytes;
3511         }
3512         return 0;
3513 }
3514
3515 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
3516 {
3517         struct btrfs_block_group_cache *cache;
3518         u64 bytenr;
3519
3520         cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
3521         if (!cache)
3522                 return 0;
3523
3524         bytenr = cache->key.objectid;
3525         btrfs_put_block_group(cache);
3526
3527         return bytenr;
3528 }
3529
3530 /*
3531  * this function must be called within transaction
3532  */
3533 int btrfs_pin_extent(struct btrfs_root *root,
3534                      u64 bytenr, u64 num_bytes, int reserved)
3535 {
3536         struct btrfs_fs_info *fs_info = root->fs_info;
3537         struct btrfs_block_group_cache *cache;
3538
3539         cache = btrfs_lookup_block_group(fs_info, bytenr);
3540         BUG_ON(!cache);
3541
3542         spin_lock(&cache->space_info->lock);
3543         spin_lock(&cache->lock);
3544         cache->pinned += num_bytes;
3545         cache->space_info->bytes_pinned += num_bytes;
3546         if (reserved) {
3547                 cache->reserved -= num_bytes;
3548                 cache->space_info->bytes_reserved -= num_bytes;
3549         }
3550         spin_unlock(&cache->lock);
3551         spin_unlock(&cache->space_info->lock);
3552
3553         btrfs_put_block_group(cache);
3554
3555         set_extent_dirty(fs_info->pinned_extents,
3556                          bytenr, bytenr + num_bytes - 1, GFP_NOFS);
3557         return 0;
3558 }
3559
3560 static int update_reserved_extents(struct btrfs_block_group_cache *cache,
3561                                    u64 num_bytes, int reserve)
3562 {
3563         spin_lock(&cache->space_info->lock);
3564         spin_lock(&cache->lock);
3565         if (reserve) {
3566                 cache->reserved += num_bytes;
3567                 cache->space_info->bytes_reserved += num_bytes;
3568         } else {
3569                 cache->reserved -= num_bytes;
3570                 cache->space_info->bytes_reserved -= num_bytes;
3571         }
3572         spin_unlock(&cache->lock);
3573         spin_unlock(&cache->space_info->lock);
3574         return 0;
3575 }
3576
3577 int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
3578                                 struct btrfs_root *root)
3579 {
3580         struct btrfs_fs_info *fs_info = root->fs_info;
3581         struct btrfs_caching_control *next;
3582         struct btrfs_caching_control *caching_ctl;
3583         struct btrfs_block_group_cache *cache;
3584
3585         down_write(&fs_info->extent_commit_sem);
3586
3587         list_for_each_entry_safe(caching_ctl, next,
3588                                  &fs_info->caching_block_groups, list) {
3589                 cache = caching_ctl->block_group;
3590                 if (block_group_cache_done(cache)) {
3591                         cache->last_byte_to_unpin = (u64)-1;
3592                         list_del_init(&caching_ctl->list);
3593                         put_caching_control(caching_ctl);
3594                 } else {
3595                         cache->last_byte_to_unpin = caching_ctl->progress;
3596                 }
3597         }
3598
3599         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
3600                 fs_info->pinned_extents = &fs_info->freed_extents[1];
3601         else
3602                 fs_info->pinned_extents = &fs_info->freed_extents[0];
3603
3604         up_write(&fs_info->extent_commit_sem);
3605         return 0;
3606 }
3607
3608 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
3609 {
3610         struct btrfs_fs_info *fs_info = root->fs_info;
3611         struct btrfs_block_group_cache *cache = NULL;
3612         u64 len;
3613
3614         while (start <= end) {
3615                 if (!cache ||
3616                     start >= cache->key.objectid + cache->key.offset) {
3617                         if (cache)
3618                                 btrfs_put_block_group(cache);
3619                         cache = btrfs_lookup_block_group(fs_info, start);
3620                         BUG_ON(!cache);
3621                 }
3622
3623                 len = cache->key.objectid + cache->key.offset - start;
3624                 len = min(len, end + 1 - start);
3625
3626                 if (start < cache->last_byte_to_unpin) {
3627                         len = min(len, cache->last_byte_to_unpin - start);
3628                         btrfs_add_free_space(cache, start, len);
3629                 }
3630
3631                 spin_lock(&cache->space_info->lock);
3632                 spin_lock(&cache->lock);
3633                 cache->pinned -= len;
3634                 cache->space_info->bytes_pinned -= len;
3635                 spin_unlock(&cache->lock);
3636                 spin_unlock(&cache->space_info->lock);
3637
3638                 start += len;
3639         }
3640
3641         if (cache)
3642                 btrfs_put_block_group(cache);
3643         return 0;
3644 }
3645
3646 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
3647                                struct btrfs_root *root)
3648 {
3649         struct btrfs_fs_info *fs_info = root->fs_info;
3650         struct extent_io_tree *unpin;
3651         u64 start;
3652         u64 end;
3653         int ret;
3654
3655         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
3656                 unpin = &fs_info->freed_extents[1];
3657         else
3658                 unpin = &fs_info->freed_extents[0];
3659
3660         while (1) {
3661                 ret = find_first_extent_bit(unpin, 0, &start, &end,
3662                                             EXTENT_DIRTY);
3663                 if (ret)
3664                         break;
3665
3666                 ret = btrfs_discard_extent(root, start, end + 1 - start);
3667
3668                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3669                 unpin_extent_range(root, start, end);
3670                 cond_resched();
3671         }
3672
3673         return ret;
3674 }
3675
3676 static int pin_down_bytes(struct btrfs_trans_handle *trans,
3677                           struct btrfs_root *root,
3678                           struct btrfs_path *path,
3679                           u64 bytenr, u64 num_bytes,
3680                           int is_data, int reserved,
3681                           struct extent_buffer **must_clean)
3682 {
3683         int err = 0;
3684         struct extent_buffer *buf;
3685
3686         if (is_data)
3687                 goto pinit;
3688
3689         buf = btrfs_find_tree_block(root, bytenr, num_bytes);
3690         if (!buf)
3691                 goto pinit;
3692
3693         /* we can reuse a block if it hasn't been written
3694          * and it is from this transaction.  We can't
3695          * reuse anything from the tree log root because
3696          * it has tiny sub-transactions.
3697          */
3698         if (btrfs_buffer_uptodate(buf, 0) &&
3699             btrfs_try_tree_lock(buf)) {
3700                 u64 header_owner = btrfs_header_owner(buf);
3701                 u64 header_transid = btrfs_header_generation(buf);
3702                 if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
3703                     header_transid == trans->transid &&
3704                     !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
3705                         *must_clean = buf;
3706                         return 1;
3707                 }
3708                 btrfs_tree_unlock(buf);
3709         }
3710         free_extent_buffer(buf);
3711 pinit:
3712         if (path)
3713                 btrfs_set_path_blocking(path);
3714         /* unlocks the pinned mutex */
3715         btrfs_pin_extent(root, bytenr, num_bytes, reserved);
3716
3717         BUG_ON(err < 0);
3718         return 0;
3719 }
3720
3721 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
3722                                 struct btrfs_root *root,
3723                                 u64 bytenr, u64 num_bytes, u64 parent,
3724                                 u64 root_objectid, u64 owner_objectid,
3725                                 u64 owner_offset, int refs_to_drop,
3726                                 struct btrfs_delayed_extent_op *extent_op)
3727 {
3728         struct btrfs_key key;
3729         struct btrfs_path *path;
3730         struct btrfs_fs_info *info = root->fs_info;
3731         struct btrfs_root *extent_root = info->extent_root;
3732         struct extent_buffer *leaf;
3733         struct btrfs_extent_item *ei;
3734         struct btrfs_extent_inline_ref *iref;
3735         int ret;
3736         int is_data;
3737         int extent_slot = 0;
3738         int found_extent = 0;
3739         int num_to_del = 1;
3740         u32 item_size;
3741         u64 refs;
3742
3743         path = btrfs_alloc_path();
3744         if (!path)
3745                 return -ENOMEM;
3746
3747         path->reada = 1;
3748         path->leave_spinning = 1;
3749
3750         is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
3751         BUG_ON(!is_data && refs_to_drop != 1);
3752
3753         ret = lookup_extent_backref(trans, extent_root, path, &iref,
3754                                     bytenr, num_bytes, parent,
3755                                     root_objectid, owner_objectid,
3756                                     owner_offset);
3757         if (ret == 0) {
3758                 extent_slot = path->slots[0];
3759                 while (extent_slot >= 0) {
3760                         btrfs_item_key_to_cpu(path->nodes[0], &key,
3761                                               extent_slot);
3762                         if (key.objectid != bytenr)
3763                                 break;
3764                         if (key.type == BTRFS_EXTENT_ITEM_KEY &&
3765                             key.offset == num_bytes) {
3766                                 found_extent = 1;
3767                                 break;
3768                         }
3769                         if (path->slots[0] - extent_slot > 5)
3770                                 break;
3771                         extent_slot--;
3772                 }
3773 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3774                 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
3775                 if (found_extent && item_size < sizeof(*ei))
3776                         found_extent = 0;
3777 #endif
3778                 if (!found_extent) {
3779                         BUG_ON(iref);
3780                         ret = remove_extent_backref(trans, extent_root, path,
3781                                                     NULL, refs_to_drop,
3782                                                     is_data);
3783                         BUG_ON(ret);
3784                         btrfs_release_path(extent_root, path);
3785                         path->leave_spinning = 1;
3786
3787                         key.objectid = bytenr;
3788                         key.type = BTRFS_EXTENT_ITEM_KEY;
3789                         key.offset = num_bytes;
3790
3791                         ret = btrfs_search_slot(trans, extent_root,
3792                                                 &key, path, -1, 1);
3793                         if (ret) {
3794                                 printk(KERN_ERR "umm, got %d back from search"
3795                                        ", was looking for %llu\n", ret,
3796                                        (unsigned long long)bytenr);
3797                                 btrfs_print_leaf(extent_root, path->nodes[0]);
3798                         }
3799                         BUG_ON(ret);
3800                         extent_slot = path->slots[0];
3801                 }
3802         } else {
3803                 btrfs_print_leaf(extent_root, path->nodes[0]);
3804                 WARN_ON(1);
3805                 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
3806                        "parent %llu root %llu  owner %llu offset %llu\n",
3807                        (unsigned long long)bytenr,
3808                        (unsigned long long)parent,
3809                        (unsigned long long)root_objectid,
3810                        (unsigned long long)owner_objectid,
3811                        (unsigned long long)owner_offset);
3812         }
3813
3814         leaf = path->nodes[0];
3815         item_size = btrfs_item_size_nr(leaf, extent_slot);
3816 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3817         if (item_size < sizeof(*ei)) {
3818                 BUG_ON(found_extent || extent_slot != path->slots[0]);
3819                 ret = convert_extent_item_v0(trans, extent_root, path,
3820                                              owner_objectid, 0);
3821                 BUG_ON(ret < 0);
3822
3823                 btrfs_release_path(extent_root, path);
3824                 path->leave_spinning = 1;
3825
3826                 key.objectid = bytenr;
3827                 key.type = BTRFS_EXTENT_ITEM_KEY;
3828                 key.offset = num_bytes;
3829
3830                 ret = btrfs_search_slot(trans, extent_root, &key, path,
3831                                         -1, 1);
3832                 if (ret) {
3833                         printk(KERN_ERR "umm, got %d back from search"
3834                                ", was looking for %llu\n", ret,
3835                                (unsigned long long)bytenr);
3836                         btrfs_print_leaf(extent_root, path->nodes[0]);
3837                 }
3838                 BUG_ON(ret);
3839                 extent_slot = path->slots[0];
3840                 leaf = path->nodes[0];
3841                 item_size = btrfs_item_size_nr(leaf, extent_slot);
3842         }
3843 #endif
3844         BUG_ON(item_size < sizeof(*ei));
3845         ei = btrfs_item_ptr(leaf, extent_slot,
3846                             struct btrfs_extent_item);
3847         if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
3848                 struct btrfs_tree_block_info *bi;
3849                 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
3850                 bi = (struct btrfs_tree_block_info *)(ei + 1);
3851                 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
3852         }
3853
3854         refs = btrfs_extent_refs(leaf, ei);
3855         BUG_ON(refs < refs_to_drop);
3856         refs -= refs_to_drop;
3857
3858         if (refs > 0) {
3859                 if (extent_op)
3860                         __run_delayed_extent_op(extent_op, leaf, ei);
3861                 /*
3862                  * In the case of inline back ref, reference count will
3863                  * be updated by remove_extent_backref
3864                  */
3865                 if (iref) {
3866                         BUG_ON(!found_extent);
3867                 } else {
3868                         btrfs_set_extent_refs(leaf, ei, refs);
3869                         btrfs_mark_buffer_dirty(leaf);
3870                 }
3871                 if (found_extent) {
3872                         ret = remove_extent_backref(trans, extent_root, path,
3873                                                     iref, refs_to_drop,
3874                                                     is_data);
3875                         BUG_ON(ret);
3876                 }
3877         } else {
3878                 int mark_free = 0;
3879                 struct extent_buffer *must_clean = NULL;
3880
3881                 if (found_extent) {
3882                         BUG_ON(is_data && refs_to_drop !=
3883                                extent_data_ref_count(root, path, iref));
3884                         if (iref) {
3885                                 BUG_ON(path->slots[0] != extent_slot);
3886                         } else {
3887                                 BUG_ON(path->slots[0] != extent_slot + 1);
3888                                 path->slots[0] = extent_slot;
3889                                 num_to_del = 2;
3890                         }
3891                 }
3892
3893                 ret = pin_down_bytes(trans, root, path, bytenr,
3894                                      num_bytes, is_data, 0, &must_clean);
3895                 if (ret > 0)
3896                         mark_free = 1;
3897                 BUG_ON(ret < 0);
3898                 /*
3899                  * it is going to be very rare for someone to be waiting
3900                  * on the block we're freeing.  del_items might need to
3901                  * schedule, so rather than get fancy, just force it
3902                  * to blocking here
3903                  */
3904                 if (must_clean)
3905                         btrfs_set_lock_blocking(must_clean);
3906
3907                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
3908                                       num_to_del);
3909                 BUG_ON(ret);
3910                 btrfs_release_path(extent_root, path);
3911
3912                 if (must_clean) {
3913                         clean_tree_block(NULL, root, must_clean);
3914                         btrfs_tree_unlock(must_clean);
3915                         free_extent_buffer(must_clean);
3916                 }
3917
3918                 if (is_data) {
3919                         ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
3920                         BUG_ON(ret);
3921                 } else {
3922                         invalidate_mapping_pages(info->btree_inode->i_mapping,
3923                              bytenr >> PAGE_CACHE_SHIFT,
3924                              (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
3925                 }
3926
3927                 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
3928                                          mark_free);
3929                 BUG_ON(ret);
3930         }
3931         btrfs_free_path(path);
3932         return ret;
3933 }
3934
3935 /*
3936  * when we free an extent, it is possible (and likely) that we free the last
3937  * delayed ref for that extent as well.  This searches the delayed ref tree for
3938  * a given extent, and if there are no other delayed refs to be processed, it
3939  * removes it from the tree.
3940  */
3941 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
3942                                       struct btrfs_root *root, u64 bytenr)
3943 {
3944         struct btrfs_delayed_ref_head *head;
3945         struct btrfs_delayed_ref_root *delayed_refs;
3946         struct btrfs_delayed_ref_node *ref;
3947         struct rb_node *node;
3948         int ret;
3949
3950         delayed_refs = &trans->transaction->delayed_refs;
3951         spin_lock(&delayed_refs->lock);
3952         head = btrfs_find_delayed_ref_head(trans, bytenr);
3953         if (!head)
3954                 goto out;
3955
3956         node = rb_prev(&head->node.rb_node);
3957         if (!node)
3958                 goto out;
3959
3960         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
3961
3962         /* there are still entries for this ref, we can't drop it */
3963         if (ref->bytenr == bytenr)
3964                 goto out;
3965
3966         if (head->extent_op) {
3967                 if (!head->must_insert_reserved)
3968                         goto out;
3969                 kfree(head->extent_op);
3970                 head->extent_op = NULL;
3971         }
3972
3973         /*
3974          * waiting for the lock here would deadlock.  If someone else has it
3975          * locked they are already in the process of dropping it anyway
3976          */
3977         if (!mutex_trylock(&head->mutex))
3978                 goto out;
3979
3980         /*
3981          * at this point we have a head with no other entries.  Go
3982          * ahead and process it.
3983          */
3984         head->node.in_tree = 0;
3985         rb_erase(&head->node.rb_node, &delayed_refs->root);
3986
3987         delayed_refs->num_entries--;
3988
3989         /*
3990          * we don't take a ref on the node because we're removing it from the
3991          * tree, so we just steal the ref the tree was holding.
3992          */
3993         delayed_refs->num_heads--;
3994         if (list_empty(&head->cluster))
3995                 delayed_refs->num_heads_ready--;
3996
3997         list_del_init(&head->cluster);
3998         spin_unlock(&delayed_refs->lock);
3999
4000         ret = run_one_delayed_ref(trans, root->fs_info->tree_root,
4001                                   &head->node, head->extent_op,
4002                                   head->must_insert_reserved);
4003         BUG_ON(ret);
4004         btrfs_put_delayed_ref(&head->node);
4005         return 0;
4006 out:
4007         spin_unlock(&delayed_refs->lock);
4008         return 0;
4009 }
4010
4011 int btrfs_free_extent(struct btrfs_trans_handle *trans,
4012                       struct btrfs_root *root,
4013                       u64 bytenr, u64 num_bytes, u64 parent,
4014                       u64 root_objectid, u64 owner, u64 offset)
4015 {
4016         int ret;
4017
4018         /*
4019          * tree log blocks never actually go into the extent allocation
4020          * tree, just update pinning info and exit early.
4021          */
4022         if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
4023                 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
4024                 /* unlocks the pinned mutex */
4025                 btrfs_pin_extent(root, bytenr, num_bytes, 1);
4026                 ret = 0;
4027         } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
4028                 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
4029                                         parent, root_objectid, (int)owner,
4030                                         BTRFS_DROP_DELAYED_REF, NULL);
4031                 BUG_ON(ret);
4032                 ret = check_ref_cleanup(trans, root, bytenr);
4033                 BUG_ON(ret);
4034         } else {
4035                 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
4036                                         parent, root_objectid, owner,
4037                                         offset, BTRFS_DROP_DELAYED_REF, NULL);
4038                 BUG_ON(ret);
4039         }
4040         return ret;
4041 }
4042
4043 static u64 stripe_align(struct btrfs_root *root, u64 val)
4044 {
4045         u64 mask = ((u64)root->stripesize - 1);
4046         u64 ret = (val + mask) & ~mask;
4047         return ret;
4048 }
4049
4050 /*
4051  * when we wait for progress in the block group caching, its because
4052  * our allocation attempt failed at least once.  So, we must sleep
4053  * and let some progress happen before we try again.
4054  *
4055  * This function will sleep at least once waiting for new free space to
4056  * show up, and then it will check the block group free space numbers
4057  * for our min num_bytes.  Another option is to have it go ahead
4058  * and look in the rbtree for a free extent of a given size, but this
4059  * is a good start.
4060  */
4061 static noinline int
4062 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
4063                                 u64 num_bytes)
4064 {
4065         struct btrfs_caching_control *caching_ctl;
4066         DEFINE_WAIT(wait);
4067
4068         caching_ctl = get_caching_control(cache);
4069         if (!caching_ctl)
4070                 return 0;
4071
4072         wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
4073                    (cache->free_space >= num_bytes));
4074
4075         put_caching_control(caching_ctl);
4076         return 0;
4077 }
4078
4079 static noinline int
4080 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
4081 {
4082         struct btrfs_caching_control *caching_ctl;
4083         DEFINE_WAIT(wait);
4084
4085         caching_ctl = get_caching_control(cache);
4086         if (!caching_ctl)
4087                 return 0;
4088
4089         wait_event(caching_ctl->wait, block_group_cache_done(cache));
4090
4091         put_caching_control(caching_ctl);
4092         return 0;
4093 }
4094
4095 enum btrfs_loop_type {
4096         LOOP_CACHED_ONLY = 0,
4097         LOOP_CACHING_NOWAIT = 1,
4098         LOOP_CACHING_WAIT = 2,
4099         LOOP_ALLOC_CHUNK = 3,
4100         LOOP_NO_EMPTY_SIZE = 4,
4101 };
4102
4103 /*
4104  * walks the btree of allocated extents and find a hole of a given size.
4105  * The key ins is changed to record the hole:
4106  * ins->objectid == block start
4107  * ins->flags = BTRFS_EXTENT_ITEM_KEY
4108  * ins->offset == number of blocks
4109  * Any available blocks before search_start are skipped.
4110  */
4111 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
4112                                      struct btrfs_root *orig_root,
4113                                      u64 num_bytes, u64 empty_size,
4114                                      u64 search_start, u64 search_end,
4115                                      u64 hint_byte, struct btrfs_key *ins,
4116                                      u64 exclude_start, u64 exclude_nr,
4117                                      int data)
4118 {
4119         int ret = 0;
4120         struct btrfs_root *root = orig_root->fs_info->extent_root;
4121         struct btrfs_free_cluster *last_ptr = NULL;
4122         struct btrfs_block_group_cache *block_group = NULL;
4123         int empty_cluster = 2 * 1024 * 1024;
4124         int allowed_chunk_alloc = 0;
4125         struct btrfs_space_info *space_info;
4126         int last_ptr_loop = 0;
4127         int loop = 0;
4128         bool found_uncached_bg = false;
4129         bool failed_cluster_refill = false;
4130         bool failed_alloc = false;
4131
4132         WARN_ON(num_bytes < root->sectorsize);
4133         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
4134         ins->objectid = 0;
4135         ins->offset = 0;
4136
4137         space_info = __find_space_info(root->fs_info, data);
4138
4139         if (orig_root->ref_cows || empty_size)
4140                 allowed_chunk_alloc = 1;
4141
4142         if (data & BTRFS_BLOCK_GROUP_METADATA) {
4143                 last_ptr = &root->fs_info->meta_alloc_cluster;
4144                 if (!btrfs_test_opt(root, SSD))
4145                         empty_cluster = 64 * 1024;
4146         }
4147
4148         if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD)) {
4149                 last_ptr = &root->fs_info->data_alloc_cluster;
4150         }
4151
4152         if (last_ptr) {
4153                 spin_lock(&last_ptr->lock);
4154                 if (last_ptr->block_group)
4155                         hint_byte = last_ptr->window_start;
4156                 spin_unlock(&last_ptr->lock);
4157         }
4158
4159         search_start = max(search_start, first_logical_byte(root, 0));
4160         search_start = max(search_start, hint_byte);
4161
4162         if (!last_ptr)
4163                 empty_cluster = 0;
4164
4165         if (search_start == hint_byte) {
4166                 block_group = btrfs_lookup_block_group(root->fs_info,
4167                                                        search_start);
4168                 /*
4169                  * we don't want to use the block group if it doesn't match our
4170                  * allocation bits, or if its not cached.
4171                  */
4172                 if (block_group && block_group_bits(block_group, data) &&
4173                     block_group_cache_done(block_group)) {
4174                         down_read(&space_info->groups_sem);
4175                         if (list_empty(&block_group->list) ||
4176                             block_group->ro) {
4177                                 /*
4178                                  * someone is removing this block group,
4179                                  * we can't jump into the have_block_group
4180                                  * target because our list pointers are not
4181                                  * valid
4182                                  */
4183                                 btrfs_put_block_group(block_group);
4184                                 up_read(&space_info->groups_sem);
4185                         } else
4186                                 goto have_block_group;
4187                 } else if (block_group) {
4188                         btrfs_put_block_group(block_group);
4189                 }
4190         }
4191
4192 search:
4193         down_read(&space_info->groups_sem);
4194         list_for_each_entry(block_group, &space_info->block_groups, list) {
4195                 u64 offset;
4196                 int cached;
4197
4198                 atomic_inc(&block_group->count);
4199                 search_start = block_group->key.objectid;
4200
4201 have_block_group:
4202                 if (unlikely(block_group->cached == BTRFS_CACHE_NO)) {
4203                         /*
4204                          * we want to start caching kthreads, but not too many
4205                          * right off the bat so we don't overwhelm the system,
4206                          * so only start them if there are less than 2 and we're
4207                          * in the initial allocation phase.
4208                          */
4209                         if (loop > LOOP_CACHING_NOWAIT ||
4210                             atomic_read(&space_info->caching_threads) < 2) {
4211                                 ret = cache_block_group(block_group);
4212                                 BUG_ON(ret);
4213                         }
4214                 }
4215
4216                 cached = block_group_cache_done(block_group);
4217                 if (unlikely(!cached)) {
4218                         found_uncached_bg = true;
4219
4220                         /* if we only want cached bgs, loop */
4221                         if (loop == LOOP_CACHED_ONLY)
4222                                 goto loop;
4223                 }
4224
4225                 if (unlikely(block_group->ro))
4226                         goto loop;
4227
4228                 /*
4229                  * Ok we want to try and use the cluster allocator, so lets look
4230                  * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will
4231                  * have tried the cluster allocator plenty of times at this
4232                  * point and not have found anything, so we are likely way too
4233                  * fragmented for the clustering stuff to find anything, so lets
4234                  * just skip it and let the allocator find whatever block it can
4235                  * find
4236                  */
4237                 if (last_ptr && loop < LOOP_NO_EMPTY_SIZE) {
4238                         /*
4239                          * the refill lock keeps out other
4240                          * people trying to start a new cluster
4241                          */
4242                         spin_lock(&last_ptr->refill_lock);
4243                         if (last_ptr->block_group &&
4244                             (last_ptr->block_group->ro ||
4245                             !block_group_bits(last_ptr->block_group, data))) {
4246                                 offset = 0;
4247                                 goto refill_cluster;
4248                         }
4249
4250                         offset = btrfs_alloc_from_cluster(block_group, last_ptr,
4251                                                  num_bytes, search_start);
4252                         if (offset) {
4253                                 /* we have a block, we're done */
4254                                 spin_unlock(&last_ptr->refill_lock);
4255                                 goto checks;
4256                         }
4257
4258                         spin_lock(&last_ptr->lock);
4259                         /*
4260                          * whoops, this cluster doesn't actually point to
4261                          * this block group.  Get a ref on the block
4262                          * group is does point to and try again
4263                          */
4264                         if (!last_ptr_loop && last_ptr->block_group &&
4265                             last_ptr->block_group != block_group) {
4266
4267                                 btrfs_put_block_group(block_group);
4268                                 block_group = last_ptr->block_group;
4269                                 atomic_inc(&block_group->count);
4270                                 spin_unlock(&last_ptr->lock);
4271                                 spin_unlock(&last_ptr->refill_lock);
4272
4273                                 last_ptr_loop = 1;
4274                                 search_start = block_group->key.objectid;
4275                                 /*
4276                                  * we know this block group is properly
4277                                  * in the list because
4278                                  * btrfs_remove_block_group, drops the
4279                                  * cluster before it removes the block
4280                                  * group from the list
4281                                  */
4282                                 goto have_block_group;
4283                         }
4284                         spin_unlock(&last_ptr->lock);
4285 refill_cluster:
4286                         /*
4287                          * this cluster didn't work out, free it and
4288                          * start over
4289                          */
4290                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
4291
4292                         last_ptr_loop = 0;
4293
4294                         /* allocate a cluster in this block group */
4295                         ret = btrfs_find_space_cluster(trans, root,
4296                                                block_group, last_ptr,
4297                                                offset, num_bytes,
4298                                                empty_cluster + empty_size);
4299                         if (ret == 0) {
4300                                 /*
4301                                  * now pull our allocation out of this
4302                                  * cluster
4303                                  */
4304                                 offset = btrfs_alloc_from_cluster(block_group,
4305                                                   last_ptr, num_bytes,
4306                                                   search_start);
4307                                 if (offset) {
4308                                         /* we found one, proceed */
4309                                         spin_unlock(&last_ptr->refill_lock);
4310                                         goto checks;
4311                                 }
4312                         } else if (!cached && loop > LOOP_CACHING_NOWAIT
4313                                    && !failed_cluster_refill) {
4314                                 spin_unlock(&last_ptr->refill_lock);
4315
4316                                 failed_cluster_refill = true;
4317                                 wait_block_group_cache_progress(block_group,
4318                                        num_bytes + empty_cluster + empty_size);
4319                                 goto have_block_group;
4320                         }
4321
4322                         /*
4323                          * at this point we either didn't find a cluster
4324                          * or we weren't able to allocate a block from our
4325                          * cluster.  Free the cluster we've been trying
4326                          * to use, and go to the next block group
4327                          */
4328                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
4329                         spin_unlock(&last_ptr->refill_lock);
4330                         goto loop;
4331                 }
4332
4333                 offset = btrfs_find_space_for_alloc(block_group, search_start,
4334                                                     num_bytes, empty_size);
4335                 /*
4336                  * If we didn't find a chunk, and we haven't failed on this
4337                  * block group before, and this block group is in the middle of
4338                  * caching and we are ok with waiting, then go ahead and wait
4339                  * for progress to be made, and set failed_alloc to true.
4340                  *
4341                  * If failed_alloc is true then we've already waited on this
4342                  * block group once and should move on to the next block group.
4343                  */
4344                 if (!offset && !failed_alloc && !cached &&
4345                     loop > LOOP_CACHING_NOWAIT) {
4346                         wait_block_group_cache_progress(block_group,
4347                                                 num_bytes + empty_size);
4348                         failed_alloc = true;
4349                         goto have_block_group;
4350                 } else if (!offset) {
4351                         goto loop;
4352                 }
4353 checks:
4354                 search_start = stripe_align(root, offset);
4355                 /* move on to the next group */
4356                 if (search_start + num_bytes >= search_end) {
4357                         btrfs_add_free_space(block_group, offset, num_bytes);
4358                         goto loop;
4359                 }
4360
4361                 /* move on to the next group */
4362                 if (search_start + num_bytes >
4363                     block_group->key.objectid + block_group->key.offset) {
4364                         btrfs_add_free_space(block_group, offset, num_bytes);
4365                         goto loop;
4366                 }
4367
4368                 if (exclude_nr > 0 &&
4369                     (search_start + num_bytes > exclude_start &&
4370                      search_start < exclude_start + exclude_nr)) {
4371                         search_start = exclude_start + exclude_nr;
4372
4373                         btrfs_add_free_space(block_group, offset, num_bytes);
4374                         /*
4375                          * if search_start is still in this block group
4376                          * then we just re-search this block group
4377                          */
4378                         if (search_start >= block_group->key.objectid &&
4379                             search_start < (block_group->key.objectid +
4380                                             block_group->key.offset))
4381                                 goto have_block_group;
4382                         goto loop;
4383                 }
4384
4385                 ins->objectid = search_start;
4386                 ins->offset = num_bytes;
4387
4388                 if (offset < search_start)
4389                         btrfs_add_free_space(block_group, offset,
4390                                              search_start - offset);
4391                 BUG_ON(offset > search_start);
4392
4393                 update_reserved_extents(block_group, num_bytes, 1);
4394
4395                 /* we are all good, lets return */
4396                 break;
4397 loop:
4398                 failed_cluster_refill = false;
4399                 failed_alloc = false;
4400                 btrfs_put_block_group(block_group);
4401         }
4402         up_read(&space_info->groups_sem);
4403
4404         /* LOOP_CACHED_ONLY, only search fully cached block groups
4405          * LOOP_CACHING_NOWAIT, search partially cached block groups, but
4406          *                      dont wait foR them to finish caching
4407          * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
4408          * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
4409          * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
4410          *                      again
4411          */
4412         if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE &&
4413             (found_uncached_bg || empty_size || empty_cluster ||
4414              allowed_chunk_alloc)) {
4415                 if (found_uncached_bg) {
4416                         found_uncached_bg = false;
4417                         if (loop < LOOP_CACHING_WAIT) {
4418                                 loop++;
4419                                 goto search;
4420                         }
4421                 }
4422
4423                 if (loop == LOOP_ALLOC_CHUNK) {
4424                         empty_size = 0;
4425                         empty_cluster = 0;
4426                 }
4427
4428                 if (allowed_chunk_alloc) {
4429                         ret = do_chunk_alloc(trans, root, num_bytes +
4430                                              2 * 1024 * 1024, data, 1);
4431                         allowed_chunk_alloc = 0;
4432                 } else {
4433                         space_info->force_alloc = 1;
4434                 }
4435
4436                 if (loop < LOOP_NO_EMPTY_SIZE) {
4437                         loop++;
4438                         goto search;
4439                 }
4440                 ret = -ENOSPC;
4441         } else if (!ins->objectid) {
4442                 ret = -ENOSPC;
4443         }
4444
4445         /* we found what we needed */
4446         if (ins->objectid) {
4447                 if (!(data & BTRFS_BLOCK_GROUP_DATA))
4448                         trans->block_group = block_group->key.objectid;
4449
4450                 btrfs_put_block_group(block_group);
4451                 ret = 0;
4452         }
4453
4454         return ret;
4455 }
4456
4457 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
4458                             int dump_block_groups)
4459 {
4460         struct btrfs_block_group_cache *cache;
4461
4462         spin_lock(&info->lock);
4463         printk(KERN_INFO "space_info has %llu free, is %sfull\n",
4464                (unsigned long long)(info->total_bytes - info->bytes_used -
4465                                     info->bytes_pinned - info->bytes_reserved -
4466                                     info->bytes_super),
4467                (info->full) ? "" : "not ");
4468         printk(KERN_INFO "space_info total=%llu, pinned=%llu, delalloc=%llu,"
4469                " may_use=%llu, used=%llu, root=%llu, super=%llu, reserved=%llu"
4470                "\n",
4471                (unsigned long long)info->total_bytes,
4472                (unsigned long long)info->bytes_pinned,
4473                (unsigned long long)info->bytes_delalloc,
4474                (unsigned long long)info->bytes_may_use,
4475                (unsigned long long)info->bytes_used,
4476                (unsigned long long)info->bytes_root,
4477                (unsigned long long)info->bytes_super,
4478                (unsigned long long)info->bytes_reserved);
4479         spin_unlock(&info->lock);
4480
4481         if (!dump_block_groups)
4482                 return;
4483
4484         down_read(&info->groups_sem);
4485         list_for_each_entry(cache, &info->block_groups, list) {
4486                 spin_lock(&cache->lock);
4487                 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
4488                        "%llu pinned %llu reserved\n",
4489                        (unsigned long long)cache->key.objectid,
4490                        (unsigned long long)cache->key.offset,
4491                        (unsigned long long)btrfs_block_group_used(&cache->item),
4492                        (unsigned long long)cache->pinned,
4493                        (unsigned long long)cache->reserved);
4494                 btrfs_dump_free_space(cache, bytes);
4495                 spin_unlock(&cache->lock);
4496         }
4497         up_read(&info->groups_sem);
4498 }
4499
4500 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
4501                          struct btrfs_root *root,
4502                          u64 num_bytes, u64 min_alloc_size,
4503                          u64 empty_size, u64 hint_byte,
4504                          u64 search_end, struct btrfs_key *ins,
4505                          u64 data)
4506 {
4507         int ret;
4508         u64 search_start = 0;
4509         struct btrfs_fs_info *info = root->fs_info;
4510
4511         data = btrfs_get_alloc_profile(root, data);
4512 again:
4513         /*
4514          * the only place that sets empty_size is btrfs_realloc_node, which
4515          * is not called recursively on allocations
4516          */
4517         if (empty_size || root->ref_cows) {
4518                 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
4519                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4520                                      2 * 1024 * 1024,
4521                                      BTRFS_BLOCK_GROUP_METADATA |
4522                                      (info->metadata_alloc_profile &
4523                                       info->avail_metadata_alloc_bits), 0);
4524                 }
4525                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4526                                      num_bytes + 2 * 1024 * 1024, data, 0);
4527         }
4528
4529         WARN_ON(num_bytes < root->sectorsize);
4530         ret = find_free_extent(trans, root, num_bytes, empty_size,
4531                                search_start, search_end, hint_byte, ins,
4532                                trans->alloc_exclude_start,
4533                                trans->alloc_exclude_nr, data);
4534
4535         if (ret == -ENOSPC && num_bytes > min_alloc_size) {
4536                 num_bytes = num_bytes >> 1;
4537                 num_bytes = num_bytes & ~(root->sectorsize - 1);
4538                 num_bytes = max(num_bytes, min_alloc_size);
4539                 do_chunk_alloc(trans, root->fs_info->extent_root,
4540                                num_bytes, data, 1);
4541                 goto again;
4542         }
4543         if (ret == -ENOSPC) {
4544                 struct btrfs_space_info *sinfo;
4545
4546                 sinfo = __find_space_info(root->fs_info, data);
4547                 printk(KERN_ERR "btrfs allocation failed flags %llu, "
4548                        "wanted %llu\n", (unsigned long long)data,
4549                        (unsigned long long)num_bytes);
4550                 dump_space_info(sinfo, num_bytes, 1);
4551         }
4552
4553         return ret;
4554 }
4555
4556 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
4557 {
4558         struct btrfs_block_group_cache *cache;
4559         int ret = 0;
4560
4561         cache = btrfs_lookup_block_group(root->fs_info, start);
4562         if (!cache) {
4563                 printk(KERN_ERR "Unable to find block group for %llu\n",
4564                        (unsigned long long)start);
4565                 return -ENOSPC;
4566         }
4567
4568         ret = btrfs_discard_extent(root, start, len);
4569
4570         btrfs_add_free_space(cache, start, len);
4571         update_reserved_extents(cache, len, 0);
4572         btrfs_put_block_group(cache);
4573
4574         return ret;
4575 }
4576
4577 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
4578                                       struct btrfs_root *root,
4579                                       u64 parent, u64 root_objectid,
4580                                       u64 flags, u64 owner, u64 offset,
4581                                       struct btrfs_key *ins, int ref_mod)
4582 {
4583         int ret;
4584         struct btrfs_fs_info *fs_info = root->fs_info;
4585         struct btrfs_extent_item *extent_item;
4586         struct btrfs_extent_inline_ref *iref;
4587         struct btrfs_path *path;
4588         struct extent_buffer *leaf;
4589         int type;
4590         u32 size;
4591
4592         if (parent > 0)
4593                 type = BTRFS_SHARED_DATA_REF_KEY;
4594         else
4595                 type = BTRFS_EXTENT_DATA_REF_KEY;
4596
4597         size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
4598
4599         path = btrfs_alloc_path();
4600         BUG_ON(!path);
4601
4602         path->leave_spinning = 1;
4603         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
4604                                       ins, size);
4605         BUG_ON(ret);
4606
4607         leaf = path->nodes[0];
4608         extent_item = btrfs_item_ptr(leaf, path->slots[0],
4609                                      struct btrfs_extent_item);
4610         btrfs_set_extent_refs(leaf, extent_item, ref_mod);
4611         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
4612         btrfs_set_extent_flags(leaf, extent_item,
4613                                flags | BTRFS_EXTENT_FLAG_DATA);
4614
4615         iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
4616         btrfs_set_extent_inline_ref_type(leaf, iref, type);
4617         if (parent > 0) {
4618                 struct btrfs_shared_data_ref *ref;
4619                 ref = (struct btrfs_shared_data_ref *)(iref + 1);
4620                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
4621                 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
4622         } else {
4623                 struct btrfs_extent_data_ref *ref;
4624                 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
4625                 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
4626                 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
4627                 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
4628                 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
4629         }
4630
4631         btrfs_mark_buffer_dirty(path->nodes[0]);
4632         btrfs_free_path(path);
4633
4634         ret = update_block_group(trans, root, ins->objectid, ins->offset,
4635                                  1, 0);
4636         if (ret) {
4637                 printk(KERN_ERR "btrfs update block group failed for %llu "
4638                        "%llu\n", (unsigned long long)ins->objectid,
4639                        (unsigned long long)ins->offset);
4640                 BUG();
4641         }
4642         return ret;
4643 }
4644
4645 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
4646                                      struct btrfs_root *root,
4647                                      u64 parent, u64 root_objectid,
4648                                      u64 flags, struct btrfs_disk_key *key,
4649                                      int level, struct btrfs_key *ins)
4650 {
4651         int ret;
4652         struct btrfs_fs_info *fs_info = root->fs_info;
4653         struct btrfs_extent_item *extent_item;
4654         struct btrfs_tree_block_info *block_info;
4655         struct btrfs_extent_inline_ref *iref;
4656         struct btrfs_path *path;
4657         struct extent_buffer *leaf;
4658         u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
4659
4660         path = btrfs_alloc_path();
4661         BUG_ON(!path);
4662
4663         path->leave_spinning = 1;
4664         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
4665                                       ins, size);
4666         BUG_ON(ret);
4667
4668         leaf = path->nodes[0];
4669         extent_item = btrfs_item_ptr(leaf, path->slots[0],
4670                                      struct btrfs_extent_item);
4671         btrfs_set_extent_refs(leaf, extent_item, 1);
4672         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
4673         btrfs_set_extent_flags(leaf, extent_item,
4674                                flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
4675         block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
4676
4677         btrfs_set_tree_block_key(leaf, block_info, key);
4678         btrfs_set_tree_block_level(leaf, block_info, level);
4679
4680         iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
4681         if (parent > 0) {
4682                 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
4683                 btrfs_set_extent_inline_ref_type(leaf, iref,
4684                                                  BTRFS_SHARED_BLOCK_REF_KEY);
4685                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
4686         } else {
4687                 btrfs_set_extent_inline_ref_type(leaf, iref,
4688                                                  BTRFS_TREE_BLOCK_REF_KEY);
4689                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
4690         }
4691
4692         btrfs_mark_buffer_dirty(leaf);
4693         btrfs_free_path(path);
4694
4695         ret = update_block_group(trans, root, ins->objectid, ins->offset,
4696                                  1, 0);
4697         if (ret) {
4698                 printk(KERN_ERR "btrfs update block group failed for %llu "
4699                        "%llu\n", (unsigned long long)ins->objectid,
4700                        (unsigned long long)ins->offset);
4701                 BUG();
4702         }
4703         return ret;
4704 }
4705
4706 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
4707                                      struct btrfs_root *root,
4708                                      u64 root_objectid, u64 owner,
4709                                      u64 offset, struct btrfs_key *ins)
4710 {
4711         int ret;
4712
4713         BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
4714
4715         ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset,
4716                                          0, root_objectid, owner, offset,
4717                                          BTRFS_ADD_DELAYED_EXTENT, NULL);
4718         return ret;
4719 }
4720
4721 /*
4722  * this is used by the tree logging recovery code.  It records that
4723  * an extent has been allocated and makes sure to clear the free
4724  * space cache bits as well
4725  */
4726 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
4727                                    struct btrfs_root *root,
4728                                    u64 root_objectid, u64 owner, u64 offset,
4729                                    struct btrfs_key *ins)
4730 {
4731         int ret;
4732         struct btrfs_block_group_cache *block_group;
4733         struct btrfs_caching_control *caching_ctl;
4734         u64 start = ins->objectid;
4735         u64 num_bytes = ins->offset;
4736
4737         block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
4738         cache_block_group(block_group);
4739         caching_ctl = get_caching_control(block_group);
4740
4741         if (!caching_ctl) {
4742                 BUG_ON(!block_group_cache_done(block_group));
4743                 ret = btrfs_remove_free_space(block_group, start, num_bytes);
4744                 BUG_ON(ret);
4745         } else {
4746                 mutex_lock(&caching_ctl->mutex);
4747
4748                 if (start >= caching_ctl->progress) {
4749                         ret = add_excluded_extent(root, start, num_bytes);
4750                         BUG_ON(ret);
4751                 } else if (start + num_bytes <= caching_ctl->progress) {
4752                         ret = btrfs_remove_free_space(block_group,
4753                                                       start, num_bytes);
4754                         BUG_ON(ret);
4755                 } else {
4756                         num_bytes = caching_ctl->progress - start;
4757                         ret = btrfs_remove_free_space(block_group,
4758                                                       start, num_bytes);
4759                         BUG_ON(ret);
4760
4761                         start = caching_ctl->progress;
4762                         num_bytes = ins->objectid + ins->offset -
4763                                     caching_ctl->progress;
4764                         ret = add_excluded_extent(root, start, num_bytes);
4765                         BUG_ON(ret);
4766                 }
4767
4768                 mutex_unlock(&caching_ctl->mutex);
4769                 put_caching_control(caching_ctl);
4770         }
4771
4772         update_reserved_extents(block_group, ins->offset, 1);
4773         btrfs_put_block_group(block_group);
4774         ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
4775                                          0, owner, offset, ins, 1);
4776         return ret;
4777 }
4778
4779 /*
4780  * finds a free extent and does all the dirty work required for allocation
4781  * returns the key for the extent through ins, and a tree buffer for
4782  * the first block of the extent through buf.
4783  *
4784  * returns 0 if everything worked, non-zero otherwise.
4785  */
4786 static int alloc_tree_block(struct btrfs_trans_handle *trans,
4787                             struct btrfs_root *root,
4788                             u64 num_bytes, u64 parent, u64 root_objectid,
4789                             struct btrfs_disk_key *key, int level,
4790                             u64 empty_size, u64 hint_byte, u64 search_end,
4791                             struct btrfs_key *ins)
4792 {
4793         int ret;
4794         u64 flags = 0;
4795
4796         ret = btrfs_reserve_extent(trans, root, num_bytes, num_bytes,
4797                                    empty_size, hint_byte, search_end,
4798                                    ins, 0);
4799         if (ret)
4800                 return ret;
4801
4802         if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
4803                 if (parent == 0)
4804                         parent = ins->objectid;
4805                 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
4806         } else
4807                 BUG_ON(parent > 0);
4808
4809         if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
4810                 struct btrfs_delayed_extent_op *extent_op;
4811                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
4812                 BUG_ON(!extent_op);
4813                 if (key)
4814                         memcpy(&extent_op->key, key, sizeof(extent_op->key));
4815                 else
4816                         memset(&extent_op->key, 0, sizeof(extent_op->key));
4817                 extent_op->flags_to_set = flags;
4818                 extent_op->update_key = 1;
4819                 extent_op->update_flags = 1;
4820                 extent_op->is_data = 0;
4821
4822                 ret = btrfs_add_delayed_tree_ref(trans, ins->objectid,
4823                                         ins->offset, parent, root_objectid,
4824                                         level, BTRFS_ADD_DELAYED_EXTENT,
4825                                         extent_op);
4826                 BUG_ON(ret);
4827         }
4828         return ret;
4829 }
4830
4831 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
4832                                             struct btrfs_root *root,
4833                                             u64 bytenr, u32 blocksize,
4834                                             int level)
4835 {
4836         struct extent_buffer *buf;
4837
4838         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
4839         if (!buf)
4840                 return ERR_PTR(-ENOMEM);
4841         btrfs_set_header_generation(buf, trans->transid);
4842         btrfs_set_buffer_lockdep_class(buf, level);
4843         btrfs_tree_lock(buf);
4844         clean_tree_block(trans, root, buf);
4845
4846         btrfs_set_lock_blocking(buf);
4847         btrfs_set_buffer_uptodate(buf);
4848
4849         if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
4850                 set_extent_dirty(&root->dirty_log_pages, buf->start,
4851                          buf->start + buf->len - 1, GFP_NOFS);
4852         } else {
4853                 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
4854                          buf->start + buf->len - 1, GFP_NOFS);
4855         }
4856         trans->blocks_used++;
4857         /* this returns a buffer locked for blocking */
4858         return buf;
4859 }
4860
4861 /*
4862  * helper function to allocate a block for a given tree
4863  * returns the tree buffer or NULL.
4864  */
4865 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
4866                                         struct btrfs_root *root, u32 blocksize,
4867                                         u64 parent, u64 root_objectid,
4868                                         struct btrfs_disk_key *key, int level,
4869                                         u64 hint, u64 empty_size)
4870 {
4871         struct btrfs_key ins;
4872         int ret;
4873         struct extent_buffer *buf;
4874
4875         ret = alloc_tree_block(trans, root, blocksize, parent, root_objectid,
4876                                key, level, empty_size, hint, (u64)-1, &ins);
4877         if (ret) {
4878                 BUG_ON(ret > 0);
4879                 return ERR_PTR(ret);
4880         }
4881
4882         buf = btrfs_init_new_buffer(trans, root, ins.objectid,
4883                                     blocksize, level);
4884         return buf;
4885 }
4886
4887 struct walk_control {
4888         u64 refs[BTRFS_MAX_LEVEL];
4889         u64 flags[BTRFS_MAX_LEVEL];
4890         struct btrfs_key update_progress;
4891         int stage;
4892         int level;
4893         int shared_level;
4894         int update_ref;
4895         int keep_locks;
4896         int reada_slot;
4897         int reada_count;
4898 };
4899
4900 #define DROP_REFERENCE  1
4901 #define UPDATE_BACKREF  2
4902
4903 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
4904                                      struct btrfs_root *root,
4905                                      struct walk_control *wc,
4906                                      struct btrfs_path *path)
4907 {
4908         u64 bytenr;
4909         u64 generation;
4910         u64 refs;
4911         u64 flags;
4912         u64 last = 0;
4913         u32 nritems;
4914         u32 blocksize;
4915         struct btrfs_key key;
4916         struct extent_buffer *eb;
4917         int ret;
4918         int slot;
4919         int nread = 0;
4920
4921         if (path->slots[wc->level] < wc->reada_slot) {
4922                 wc->reada_count = wc->reada_count * 2 / 3;
4923                 wc->reada_count = max(wc->reada_count, 2);
4924         } else {
4925                 wc->reada_count = wc->reada_count * 3 / 2;
4926                 wc->reada_count = min_t(int, wc->reada_count,
4927                                         BTRFS_NODEPTRS_PER_BLOCK(root));
4928         }
4929
4930         eb = path->nodes[wc->level];
4931         nritems = btrfs_header_nritems(eb);
4932         blocksize = btrfs_level_size(root, wc->level - 1);
4933
4934         for (slot = path->slots[wc->level]; slot < nritems; slot++) {
4935                 if (nread >= wc->reada_count)
4936                         break;
4937
4938                 cond_resched();
4939                 bytenr = btrfs_node_blockptr(eb, slot);
4940                 generation = btrfs_node_ptr_generation(eb, slot);
4941
4942                 if (slot == path->slots[wc->level])
4943                         goto reada;
4944
4945                 if (wc->stage == UPDATE_BACKREF &&
4946                     generation <= root->root_key.offset)
4947                         continue;
4948
4949                 /* We don't lock the tree block, it's OK to be racy here */
4950                 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
4951                                                &refs, &flags);
4952                 BUG_ON(ret);
4953                 BUG_ON(refs == 0);
4954
4955                 if (wc->stage == DROP_REFERENCE) {
4956                         if (refs == 1)
4957                                 goto reada;
4958
4959                         if (wc->level == 1 &&
4960                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
4961                                 continue;
4962                         if (!wc->update_ref ||
4963                             generation <= root->root_key.offset)
4964                                 continue;
4965                         btrfs_node_key_to_cpu(eb, &key, slot);
4966                         ret = btrfs_comp_cpu_keys(&key,
4967                                                   &wc->update_progress);
4968                         if (ret < 0)
4969                                 continue;
4970                 } else {
4971                         if (wc->level == 1 &&
4972                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
4973                                 continue;
4974                 }
4975 reada:
4976                 ret = readahead_tree_block(root, bytenr, blocksize,
4977                                            generation);
4978                 if (ret)
4979                         break;
4980                 last = bytenr + blocksize;
4981                 nread++;
4982         }
4983         wc->reada_slot = slot;
4984 }
4985
4986 /*
4987  * hepler to process tree block while walking down the tree.
4988  *
4989  * when wc->stage == UPDATE_BACKREF, this function updates
4990  * back refs for pointers in the block.
4991  *
4992  * NOTE: return value 1 means we should stop walking down.
4993  */
4994 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
4995                                    struct btrfs_root *root,
4996                                    struct btrfs_path *path,
4997                                    struct walk_control *wc, int lookup_info)
4998 {
4999         int level = wc->level;
5000         struct extent_buffer *eb = path->nodes[level];
5001         u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5002         int ret;
5003
5004         if (wc->stage == UPDATE_BACKREF &&
5005             btrfs_header_owner(eb) != root->root_key.objectid)
5006                 return 1;
5007
5008         /*
5009          * when reference count of tree block is 1, it won't increase
5010          * again. once full backref flag is set, we never clear it.
5011          */
5012         if (lookup_info &&
5013             ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
5014              (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
5015                 BUG_ON(!path->locks[level]);
5016                 ret = btrfs_lookup_extent_info(trans, root,
5017                                                eb->start, eb->len,
5018                                                &wc->refs[level],
5019                                                &wc->flags[level]);
5020                 BUG_ON(ret);
5021                 BUG_ON(wc->refs[level] == 0);
5022         }
5023
5024         if (wc->stage == DROP_REFERENCE) {
5025                 if (wc->refs[level] > 1)
5026                         return 1;
5027
5028                 if (path->locks[level] && !wc->keep_locks) {
5029                         btrfs_tree_unlock(eb);
5030                         path->locks[level] = 0;
5031                 }
5032                 return 0;
5033         }
5034
5035         /* wc->stage == UPDATE_BACKREF */
5036         if (!(wc->flags[level] & flag)) {
5037                 BUG_ON(!path->locks[level]);
5038                 ret = btrfs_inc_ref(trans, root, eb, 1);
5039                 BUG_ON(ret);
5040                 ret = btrfs_dec_ref(trans, root, eb, 0);
5041                 BUG_ON(ret);
5042                 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
5043                                                   eb->len, flag, 0);
5044                 BUG_ON(ret);
5045                 wc->flags[level] |= flag;
5046         }
5047
5048         /*
5049          * the block is shared by multiple trees, so it's not good to
5050          * keep the tree lock
5051          */
5052         if (path->locks[level] && level > 0) {
5053                 btrfs_tree_unlock(eb);
5054                 path->locks[level] = 0;
5055         }
5056         return 0;
5057 }
5058
5059 /*
5060  * hepler to process tree block pointer.
5061  *
5062  * when wc->stage == DROP_REFERENCE, this function checks
5063  * reference count of the block pointed to. if the block
5064  * is shared and we need update back refs for the subtree
5065  * rooted at the block, this function changes wc->stage to
5066  * UPDATE_BACKREF. if the block is shared and there is no
5067  * need to update back, this function drops the reference
5068  * to the block.
5069  *
5070  * NOTE: return value 1 means we should stop walking down.
5071  */
5072 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
5073                                  struct btrfs_root *root,
5074                                  struct btrfs_path *path,
5075                                  struct walk_control *wc, int *lookup_info)
5076 {
5077         u64 bytenr;
5078         u64 generation;
5079         u64 parent;
5080         u32 blocksize;
5081         struct btrfs_key key;
5082         struct extent_buffer *next;
5083         int level = wc->level;
5084         int reada = 0;
5085         int ret = 0;
5086
5087         generation = btrfs_node_ptr_generation(path->nodes[level],
5088                                                path->slots[level]);
5089         /*
5090          * if the lower level block was created before the snapshot
5091          * was created, we know there is no need to update back refs
5092          * for the subtree
5093          */
5094         if (wc->stage == UPDATE_BACKREF &&
5095             generation <= root->root_key.offset) {
5096                 *lookup_info = 1;
5097                 return 1;
5098         }
5099
5100         bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
5101         blocksize = btrfs_level_size(root, level - 1);
5102
5103         next = btrfs_find_tree_block(root, bytenr, blocksize);
5104         if (!next) {
5105                 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
5106                 reada = 1;
5107         }
5108         btrfs_tree_lock(next);
5109         btrfs_set_lock_blocking(next);
5110
5111         ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
5112                                        &wc->refs[level - 1],
5113                                        &wc->flags[level - 1]);
5114         BUG_ON(ret);
5115         BUG_ON(wc->refs[level - 1] == 0);
5116         *lookup_info = 0;
5117
5118         if (wc->stage == DROP_REFERENCE) {
5119                 if (wc->refs[level - 1] > 1) {
5120                         if (level == 1 &&
5121                             (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5122                                 goto skip;
5123
5124                         if (!wc->update_ref ||
5125                             generation <= root->root_key.offset)
5126                                 goto skip;
5127
5128                         btrfs_node_key_to_cpu(path->nodes[level], &key,
5129                                               path->slots[level]);
5130                         ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
5131                         if (ret < 0)
5132                                 goto skip;
5133
5134                         wc->stage = UPDATE_BACKREF;
5135                         wc->shared_level = level - 1;
5136                 }
5137         } else {
5138                 if (level == 1 &&
5139                     (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5140                         goto skip;
5141         }
5142
5143         if (!btrfs_buffer_uptodate(next, generation)) {
5144                 btrfs_tree_unlock(next);
5145                 free_extent_buffer(next);
5146                 next = NULL;
5147                 *lookup_info = 1;
5148         }
5149
5150         if (!next) {
5151                 if (reada && level == 1)
5152                         reada_walk_down(trans, root, wc, path);
5153                 next = read_tree_block(root, bytenr, blocksize, generation);
5154                 btrfs_tree_lock(next);
5155                 btrfs_set_lock_blocking(next);
5156         }
5157
5158         level--;
5159         BUG_ON(level != btrfs_header_level(next));
5160         path->nodes[level] = next;
5161         path->slots[level] = 0;
5162         path->locks[level] = 1;
5163         wc->level = level;
5164         if (wc->level == 1)
5165                 wc->reada_slot = 0;
5166         return 0;
5167 skip:
5168         wc->refs[level - 1] = 0;
5169         wc->flags[level - 1] = 0;
5170         if (wc->stage == DROP_REFERENCE) {
5171                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
5172                         parent = path->nodes[level]->start;
5173                 } else {
5174                         BUG_ON(root->root_key.objectid !=
5175                                btrfs_header_owner(path->nodes[level]));
5176                         parent = 0;
5177                 }
5178
5179                 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
5180                                         root->root_key.objectid, level - 1, 0);
5181                 BUG_ON(ret);
5182         }
5183         btrfs_tree_unlock(next);
5184         free_extent_buffer(next);
5185         *lookup_info = 1;
5186         return 1;
5187 }
5188
5189 /*
5190  * hepler to process tree block while walking up the tree.
5191  *
5192  * when wc->stage == DROP_REFERENCE, this function drops
5193  * reference count on the block.
5194  *
5195  * when wc->stage == UPDATE_BACKREF, this function changes
5196  * wc->stage back to DROP_REFERENCE if we changed wc->stage
5197  * to UPDATE_BACKREF previously while processing the block.
5198  *
5199  * NOTE: return value 1 means we should stop walking up.
5200  */
5201 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
5202                                  struct btrfs_root *root,
5203                                  struct btrfs_path *path,
5204                                  struct walk_control *wc)
5205 {
5206         int ret = 0;
5207         int level = wc->level;
5208         struct extent_buffer *eb = path->nodes[level];
5209         u64 parent = 0;
5210
5211         if (wc->stage == UPDATE_BACKREF) {
5212                 BUG_ON(wc->shared_level < level);
5213                 if (level < wc->shared_level)
5214                         goto out;
5215
5216                 ret = find_next_key(path, level + 1, &wc->update_progress);
5217                 if (ret > 0)
5218                         wc->update_ref = 0;
5219
5220                 wc->stage = DROP_REFERENCE;
5221                 wc->shared_level = -1;
5222                 path->slots[level] = 0;
5223
5224                 /*
5225                  * check reference count again if the block isn't locked.
5226                  * we should start walking down the tree again if reference
5227                  * count is one.
5228                  */
5229                 if (!path->locks[level]) {
5230                         BUG_ON(level == 0);
5231                         btrfs_tree_lock(eb);
5232                         btrfs_set_lock_blocking(eb);
5233                         path->locks[level] = 1;
5234
5235                         ret = btrfs_lookup_extent_info(trans, root,
5236                                                        eb->start, eb->len,
5237                                                        &wc->refs[level],
5238                                                        &wc->flags[level]);
5239                         BUG_ON(ret);
5240                         BUG_ON(wc->refs[level] == 0);
5241                         if (wc->refs[level] == 1) {
5242                                 btrfs_tree_unlock(eb);
5243                                 path->locks[level] = 0;
5244                                 return 1;
5245                         }
5246                 }
5247         }
5248
5249         /* wc->stage == DROP_REFERENCE */
5250         BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
5251
5252         if (wc->refs[level] == 1) {
5253                 if (level == 0) {
5254                         if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5255                                 ret = btrfs_dec_ref(trans, root, eb, 1);
5256                         else
5257                                 ret = btrfs_dec_ref(trans, root, eb, 0);
5258                         BUG_ON(ret);
5259                 }
5260                 /* make block locked assertion in clean_tree_block happy */
5261                 if (!path->locks[level] &&
5262                     btrfs_header_generation(eb) == trans->transid) {
5263                         btrfs_tree_lock(eb);
5264                         btrfs_set_lock_blocking(eb);
5265                         path->locks[level] = 1;
5266                 }
5267                 clean_tree_block(trans, root, eb);
5268         }
5269
5270         if (eb == root->node) {
5271                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5272                         parent = eb->start;
5273                 else
5274                         BUG_ON(root->root_key.objectid !=
5275                                btrfs_header_owner(eb));
5276         } else {
5277                 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5278                         parent = path->nodes[level + 1]->start;
5279                 else
5280                         BUG_ON(root->root_key.objectid !=
5281                                btrfs_header_owner(path->nodes[level + 1]));
5282         }
5283
5284         ret = btrfs_free_extent(trans, root, eb->start, eb->len, parent,
5285                                 root->root_key.objectid, level, 0);
5286         BUG_ON(ret);
5287 out:
5288         wc->refs[level] = 0;
5289         wc->flags[level] = 0;
5290         return ret;
5291 }
5292
5293 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
5294                                    struct btrfs_root *root,
5295                                    struct btrfs_path *path,
5296                                    struct walk_control *wc)
5297 {
5298         int level = wc->level;
5299         int lookup_info = 1;
5300         int ret;
5301
5302         while (level >= 0) {
5303                 if (path->slots[level] >=
5304                     btrfs_header_nritems(path->nodes[level]))
5305                         break;
5306
5307                 ret = walk_down_proc(trans, root, path, wc, lookup_info);
5308                 if (ret > 0)
5309                         break;
5310
5311                 if (level == 0)
5312                         break;
5313
5314                 ret = do_walk_down(trans, root, path, wc, &lookup_info);
5315                 if (ret > 0) {
5316                         path->slots[level]++;
5317                         continue;
5318                 }
5319                 level = wc->level;
5320         }
5321         return 0;
5322 }
5323
5324 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
5325                                  struct btrfs_root *root,
5326                                  struct btrfs_path *path,
5327                                  struct walk_control *wc, int max_level)
5328 {
5329         int level = wc->level;
5330         int ret;
5331
5332         path->slots[level] = btrfs_header_nritems(path->nodes[level]);
5333         while (level < max_level && path->nodes[level]) {
5334                 wc->level = level;
5335                 if (path->slots[level] + 1 <
5336                     btrfs_header_nritems(path->nodes[level])) {
5337                         path->slots[level]++;
5338                         return 0;
5339                 } else {
5340                         ret = walk_up_proc(trans, root, path, wc);
5341                         if (ret > 0)
5342                                 return 0;
5343
5344                         if (path->locks[level]) {
5345                                 btrfs_tree_unlock(path->nodes[level]);
5346                                 path->locks[level] = 0;
5347                         }
5348                         free_extent_buffer(path->nodes[level]);
5349                         path->nodes[level] = NULL;
5350                         level++;
5351                 }
5352         }
5353         return 1;
5354 }
5355
5356 /*
5357  * drop a subvolume tree.
5358  *
5359  * this function traverses the tree freeing any blocks that only
5360  * referenced by the tree.
5361  *
5362  * when a shared tree block is found. this function decreases its
5363  * reference count by one. if update_ref is true, this function
5364  * also make sure backrefs for the shared block and all lower level
5365  * blocks are properly updated.
5366  */
5367 int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref)
5368 {
5369         struct btrfs_path *path;
5370         struct btrfs_trans_handle *trans;
5371         struct btrfs_root *tree_root = root->fs_info->tree_root;
5372         struct btrfs_root_item *root_item = &root->root_item;
5373         struct walk_control *wc;
5374         struct btrfs_key key;
5375         int err = 0;
5376         int ret;
5377         int level;
5378
5379         path = btrfs_alloc_path();
5380         BUG_ON(!path);
5381
5382         wc = kzalloc(sizeof(*wc), GFP_NOFS);
5383         BUG_ON(!wc);
5384
5385         trans = btrfs_start_transaction(tree_root, 1);
5386
5387         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
5388                 level = btrfs_header_level(root->node);
5389                 path->nodes[level] = btrfs_lock_root_node(root);
5390                 btrfs_set_lock_blocking(path->nodes[level]);
5391                 path->slots[level] = 0;
5392                 path->locks[level] = 1;
5393                 memset(&wc->update_progress, 0,
5394                        sizeof(wc->update_progress));
5395         } else {
5396                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
5397                 memcpy(&wc->update_progress, &key,
5398                        sizeof(wc->update_progress));
5399
5400                 level = root_item->drop_level;
5401                 BUG_ON(level == 0);
5402                 path->lowest_level = level;
5403                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5404                 path->lowest_level = 0;
5405                 if (ret < 0) {
5406                         err = ret;
5407                         goto out;
5408                 }
5409                 WARN_ON(ret > 0);
5410
5411                 /*
5412                  * unlock our path, this is safe because only this
5413                  * function is allowed to delete this snapshot
5414                  */
5415                 btrfs_unlock_up_safe(path, 0);
5416
5417                 level = btrfs_header_level(root->node);
5418                 while (1) {
5419                         btrfs_tree_lock(path->nodes[level]);
5420                         btrfs_set_lock_blocking(path->nodes[level]);
5421
5422                         ret = btrfs_lookup_extent_info(trans, root,
5423                                                 path->nodes[level]->start,
5424                                                 path->nodes[level]->len,
5425                                                 &wc->refs[level],
5426                                                 &wc->flags[level]);
5427                         BUG_ON(ret);
5428                         BUG_ON(wc->refs[level] == 0);
5429
5430                         if (level == root_item->drop_level)
5431                                 break;
5432
5433                         btrfs_tree_unlock(path->nodes[level]);
5434                         WARN_ON(wc->refs[level] != 1);
5435                         level--;
5436                 }
5437         }
5438
5439         wc->level = level;
5440         wc->shared_level = -1;
5441         wc->stage = DROP_REFERENCE;
5442         wc->update_ref = update_ref;
5443         wc->keep_locks = 0;
5444         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
5445
5446         while (1) {
5447                 ret = walk_down_tree(trans, root, path, wc);
5448                 if (ret < 0) {
5449                         err = ret;
5450                         break;
5451                 }
5452
5453                 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
5454                 if (ret < 0) {
5455                         err = ret;
5456                         break;
5457                 }
5458
5459                 if (ret > 0) {
5460                         BUG_ON(wc->stage != DROP_REFERENCE);
5461                         break;
5462                 }
5463
5464                 if (wc->stage == DROP_REFERENCE) {
5465                         level = wc->level;
5466                         btrfs_node_key(path->nodes[level],
5467                                        &root_item->drop_progress,
5468                                        path->slots[level]);
5469                         root_item->drop_level = level;
5470                 }
5471
5472                 BUG_ON(wc->level == 0);
5473                 if (trans->transaction->in_commit ||
5474                     trans->transaction->delayed_refs.flushing) {
5475                         ret = btrfs_update_root(trans, tree_root,
5476                                                 &root->root_key,
5477                                                 root_item);
5478                         BUG_ON(ret);
5479
5480                         btrfs_end_transaction(trans, tree_root);
5481                         trans = btrfs_start_transaction(tree_root, 1);
5482                 } else {
5483                         unsigned long update;
5484                         update = trans->delayed_ref_updates;
5485                         trans->delayed_ref_updates = 0;
5486                         if (update)
5487                                 btrfs_run_delayed_refs(trans, tree_root,
5488                                                        update);
5489                 }
5490         }
5491         btrfs_release_path(root, path);
5492         BUG_ON(err);
5493
5494         ret = btrfs_del_root(trans, tree_root, &root->root_key);
5495         BUG_ON(ret);
5496
5497         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
5498                 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
5499                                            NULL, NULL);
5500                 BUG_ON(ret < 0);
5501                 if (ret > 0) {
5502                         ret = btrfs_del_orphan_item(trans, tree_root,
5503                                                     root->root_key.objectid);
5504                         BUG_ON(ret);
5505                 }
5506         }
5507
5508         if (root->in_radix) {
5509                 btrfs_free_fs_root(tree_root->fs_info, root);
5510         } else {
5511                 free_extent_buffer(root->node);
5512                 free_extent_buffer(root->commit_root);
5513                 kfree(root);
5514         }
5515 out:
5516         btrfs_end_transaction(trans, tree_root);
5517         kfree(wc);
5518         btrfs_free_path(path);
5519         return err;
5520 }
5521
5522 /*
5523  * drop subtree rooted at tree block 'node'.
5524  *
5525  * NOTE: this function will unlock and release tree block 'node'
5526  */
5527 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
5528                         struct btrfs_root *root,
5529                         struct extent_buffer *node,
5530                         struct extent_buffer *parent)
5531 {
5532         struct btrfs_path *path;
5533         struct walk_control *wc;
5534         int level;
5535         int parent_level;
5536         int ret = 0;
5537         int wret;
5538
5539         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
5540
5541         path = btrfs_alloc_path();
5542         BUG_ON(!path);
5543
5544         wc = kzalloc(sizeof(*wc), GFP_NOFS);
5545         BUG_ON(!wc);
5546
5547         btrfs_assert_tree_locked(parent);
5548         parent_level = btrfs_header_level(parent);
5549         extent_buffer_get(parent);
5550         path->nodes[parent_level] = parent;
5551         path->slots[parent_level] = btrfs_header_nritems(parent);
5552
5553         btrfs_assert_tree_locked(node);
5554         level = btrfs_header_level(node);
5555         path->nodes[level] = node;
5556         path->slots[level] = 0;
5557         path->locks[level] = 1;
5558
5559         wc->refs[parent_level] = 1;
5560         wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5561         wc->level = level;
5562         wc->shared_level = -1;
5563         wc->stage = DROP_REFERENCE;
5564         wc->update_ref = 0;
5565         wc->keep_locks = 1;
5566         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
5567
5568         while (1) {
5569                 wret = walk_down_tree(trans, root, path, wc);
5570                 if (wret < 0) {
5571                         ret = wret;
5572                         break;
5573                 }
5574
5575                 wret = walk_up_tree(trans, root, path, wc, parent_level);
5576                 if (wret < 0)
5577                         ret = wret;
5578                 if (wret != 0)
5579                         break;
5580         }
5581