]> git.openfabrics.org - ~shefty/rdma-dev.git/blob - fs/btrfs/extent-tree.c
Merge git://git.kernel.org/pub/scm/linux/kernel/git/hch/xfs-icache-races
[~shefty/rdma-dev.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_reserved_extents(struct btrfs_root *root,
36                                    u64 bytenr, u64 num, int reserve);
37 static int update_block_group(struct btrfs_trans_handle *trans,
38                               struct btrfs_root *root,
39                               u64 bytenr, u64 num_bytes, int alloc,
40                               int mark_free);
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
61 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
62                           struct btrfs_root *extent_root, u64 alloc_bytes,
63                           u64 flags, int force);
64
65 static noinline int
66 block_group_cache_done(struct btrfs_block_group_cache *cache)
67 {
68         smp_mb();
69         return cache->cached == BTRFS_CACHE_FINISHED;
70 }
71
72 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
73 {
74         return (cache->flags & bits) == bits;
75 }
76
77 /*
78  * this adds the block group to the fs_info rb tree for the block group
79  * cache
80  */
81 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
82                                 struct btrfs_block_group_cache *block_group)
83 {
84         struct rb_node **p;
85         struct rb_node *parent = NULL;
86         struct btrfs_block_group_cache *cache;
87
88         spin_lock(&info->block_group_cache_lock);
89         p = &info->block_group_cache_tree.rb_node;
90
91         while (*p) {
92                 parent = *p;
93                 cache = rb_entry(parent, struct btrfs_block_group_cache,
94                                  cache_node);
95                 if (block_group->key.objectid < cache->key.objectid) {
96                         p = &(*p)->rb_left;
97                 } else if (block_group->key.objectid > cache->key.objectid) {
98                         p = &(*p)->rb_right;
99                 } else {
100                         spin_unlock(&info->block_group_cache_lock);
101                         return -EEXIST;
102                 }
103         }
104
105         rb_link_node(&block_group->cache_node, parent, p);
106         rb_insert_color(&block_group->cache_node,
107                         &info->block_group_cache_tree);
108         spin_unlock(&info->block_group_cache_lock);
109
110         return 0;
111 }
112
113 /*
114  * This will return the block group at or after bytenr if contains is 0, else
115  * it will return the block group that contains the bytenr
116  */
117 static struct btrfs_block_group_cache *
118 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
119                               int contains)
120 {
121         struct btrfs_block_group_cache *cache, *ret = NULL;
122         struct rb_node *n;
123         u64 end, start;
124
125         spin_lock(&info->block_group_cache_lock);
126         n = info->block_group_cache_tree.rb_node;
127
128         while (n) {
129                 cache = rb_entry(n, struct btrfs_block_group_cache,
130                                  cache_node);
131                 end = cache->key.objectid + cache->key.offset - 1;
132                 start = cache->key.objectid;
133
134                 if (bytenr < start) {
135                         if (!contains && (!ret || start < ret->key.objectid))
136                                 ret = cache;
137                         n = n->rb_left;
138                 } else if (bytenr > start) {
139                         if (contains && bytenr <= end) {
140                                 ret = cache;
141                                 break;
142                         }
143                         n = n->rb_right;
144                 } else {
145                         ret = cache;
146                         break;
147                 }
148         }
149         if (ret)
150                 atomic_inc(&ret->count);
151         spin_unlock(&info->block_group_cache_lock);
152
153         return ret;
154 }
155
156 /*
157  * We always set EXTENT_LOCKED for the super mirror extents so we don't
158  * overwrite them, so those bits need to be unset.  Also, if we are unmounting
159  * with pinned extents still sitting there because we had a block group caching,
160  * we need to clear those now, since we are done.
161  */
162 void btrfs_free_pinned_extents(struct btrfs_fs_info *info)
163 {
164         u64 start, end, last = 0;
165         int ret;
166
167         while (1) {
168                 ret = find_first_extent_bit(&info->pinned_extents, last,
169                                             &start, &end,
170                                             EXTENT_LOCKED|EXTENT_DIRTY);
171                 if (ret)
172                         break;
173
174                 clear_extent_bits(&info->pinned_extents, start, end,
175                                   EXTENT_LOCKED|EXTENT_DIRTY, GFP_NOFS);
176                 last = end+1;
177         }
178 }
179
180 static int remove_sb_from_cache(struct btrfs_root *root,
181                                 struct btrfs_block_group_cache *cache)
182 {
183         struct btrfs_fs_info *fs_info = root->fs_info;
184         u64 bytenr;
185         u64 *logical;
186         int stripe_len;
187         int i, nr, ret;
188
189         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
190                 bytenr = btrfs_sb_offset(i);
191                 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
192                                        cache->key.objectid, bytenr,
193                                        0, &logical, &nr, &stripe_len);
194                 BUG_ON(ret);
195                 while (nr--) {
196                         try_lock_extent(&fs_info->pinned_extents,
197                                         logical[nr],
198                                         logical[nr] + stripe_len - 1, GFP_NOFS);
199                 }
200                 kfree(logical);
201         }
202
203         return 0;
204 }
205
206 /*
207  * this is only called by cache_block_group, since we could have freed extents
208  * we need to check the pinned_extents for any extents that can't be used yet
209  * since their free space will be released as soon as the transaction commits.
210  */
211 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
212                               struct btrfs_fs_info *info, u64 start, u64 end)
213 {
214         u64 extent_start, extent_end, size, total_added = 0;
215         int ret;
216
217         while (start < end) {
218                 ret = find_first_extent_bit(&info->pinned_extents, start,
219                                             &extent_start, &extent_end,
220                                             EXTENT_DIRTY|EXTENT_LOCKED);
221                 if (ret)
222                         break;
223
224                 if (extent_start == start) {
225                         start = extent_end + 1;
226                 } else if (extent_start > start && extent_start < end) {
227                         size = extent_start - start;
228                         total_added += size;
229                         ret = btrfs_add_free_space(block_group, start,
230                                                    size);
231                         BUG_ON(ret);
232                         start = extent_end + 1;
233                 } else {
234                         break;
235                 }
236         }
237
238         if (start < end) {
239                 size = end - start;
240                 total_added += size;
241                 ret = btrfs_add_free_space(block_group, start, size);
242                 BUG_ON(ret);
243         }
244
245         return total_added;
246 }
247
248 static int caching_kthread(void *data)
249 {
250         struct btrfs_block_group_cache *block_group = data;
251         struct btrfs_fs_info *fs_info = block_group->fs_info;
252         u64 last = 0;
253         struct btrfs_path *path;
254         int ret = 0;
255         struct btrfs_key key;
256         struct extent_buffer *leaf;
257         int slot;
258         u64 total_found = 0;
259
260         BUG_ON(!fs_info);
261
262         path = btrfs_alloc_path();
263         if (!path)
264                 return -ENOMEM;
265
266         atomic_inc(&block_group->space_info->caching_threads);
267         last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
268 again:
269         /* need to make sure the commit_root doesn't disappear */
270         down_read(&fs_info->extent_commit_sem);
271
272         /*
273          * We don't want to deadlock with somebody trying to allocate a new
274          * extent for the extent root while also trying to search the extent
275          * root to add free space.  So we skip locking and search the commit
276          * root, since its read-only
277          */
278         path->skip_locking = 1;
279         path->search_commit_root = 1;
280         path->reada = 2;
281
282         key.objectid = last;
283         key.offset = 0;
284         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
285         ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
286         if (ret < 0)
287                 goto err;
288
289         while (1) {
290                 smp_mb();
291                 if (block_group->fs_info->closing > 1) {
292                         last = (u64)-1;
293                         break;
294                 }
295
296                 leaf = path->nodes[0];
297                 slot = path->slots[0];
298                 if (slot >= btrfs_header_nritems(leaf)) {
299                         ret = btrfs_next_leaf(fs_info->extent_root, path);
300                         if (ret < 0)
301                                 goto err;
302                         else if (ret)
303                                 break;
304
305                         if (need_resched() ||
306                             btrfs_transaction_in_commit(fs_info)) {
307                                 btrfs_release_path(fs_info->extent_root, path);
308                                 up_read(&fs_info->extent_commit_sem);
309                                 schedule_timeout(1);
310                                 goto again;
311                         }
312
313                         continue;
314                 }
315                 btrfs_item_key_to_cpu(leaf, &key, slot);
316                 if (key.objectid < block_group->key.objectid)
317                         goto next;
318
319                 if (key.objectid >= block_group->key.objectid +
320                     block_group->key.offset)
321                         break;
322
323                 if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
324                         total_found += add_new_free_space(block_group,
325                                                           fs_info, last,
326                                                           key.objectid);
327                         last = key.objectid + key.offset;
328                 }
329
330                 if (total_found > (1024 * 1024 * 2)) {
331                         total_found = 0;
332                         wake_up(&block_group->caching_q);
333                 }
334 next:
335                 path->slots[0]++;
336         }
337         ret = 0;
338
339         total_found += add_new_free_space(block_group, fs_info, last,
340                                           block_group->key.objectid +
341                                           block_group->key.offset);
342
343         spin_lock(&block_group->lock);
344         block_group->cached = BTRFS_CACHE_FINISHED;
345         spin_unlock(&block_group->lock);
346
347 err:
348         btrfs_free_path(path);
349         up_read(&fs_info->extent_commit_sem);
350         atomic_dec(&block_group->space_info->caching_threads);
351         wake_up(&block_group->caching_q);
352
353         return 0;
354 }
355
356 static int cache_block_group(struct btrfs_block_group_cache *cache)
357 {
358         struct task_struct *tsk;
359         int ret = 0;
360
361         spin_lock(&cache->lock);
362         if (cache->cached != BTRFS_CACHE_NO) {
363                 spin_unlock(&cache->lock);
364                 return ret;
365         }
366         cache->cached = BTRFS_CACHE_STARTED;
367         spin_unlock(&cache->lock);
368
369         tsk = kthread_run(caching_kthread, cache, "btrfs-cache-%llu\n",
370                           cache->key.objectid);
371         if (IS_ERR(tsk)) {
372                 ret = PTR_ERR(tsk);
373                 printk(KERN_ERR "error running thread %d\n", ret);
374                 BUG();
375         }
376
377         return ret;
378 }
379
380 /*
381  * return the block group that starts at or after bytenr
382  */
383 static struct btrfs_block_group_cache *
384 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
385 {
386         struct btrfs_block_group_cache *cache;
387
388         cache = block_group_cache_tree_search(info, bytenr, 0);
389
390         return cache;
391 }
392
393 /*
394  * return the block group that contains the given bytenr
395  */
396 struct btrfs_block_group_cache *btrfs_lookup_block_group(
397                                                  struct btrfs_fs_info *info,
398                                                  u64 bytenr)
399 {
400         struct btrfs_block_group_cache *cache;
401
402         cache = block_group_cache_tree_search(info, bytenr, 1);
403
404         return cache;
405 }
406
407 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
408 {
409         if (atomic_dec_and_test(&cache->count))
410                 kfree(cache);
411 }
412
413 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
414                                                   u64 flags)
415 {
416         struct list_head *head = &info->space_info;
417         struct btrfs_space_info *found;
418
419         rcu_read_lock();
420         list_for_each_entry_rcu(found, head, list) {
421                 if (found->flags == flags) {
422                         rcu_read_unlock();
423                         return found;
424                 }
425         }
426         rcu_read_unlock();
427         return NULL;
428 }
429
430 /*
431  * after adding space to the filesystem, we need to clear the full flags
432  * on all the space infos.
433  */
434 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
435 {
436         struct list_head *head = &info->space_info;
437         struct btrfs_space_info *found;
438
439         rcu_read_lock();
440         list_for_each_entry_rcu(found, head, list)
441                 found->full = 0;
442         rcu_read_unlock();
443 }
444
445 static u64 div_factor(u64 num, int factor)
446 {
447         if (factor == 10)
448                 return num;
449         num *= factor;
450         do_div(num, 10);
451         return num;
452 }
453
454 u64 btrfs_find_block_group(struct btrfs_root *root,
455                            u64 search_start, u64 search_hint, int owner)
456 {
457         struct btrfs_block_group_cache *cache;
458         u64 used;
459         u64 last = max(search_hint, search_start);
460         u64 group_start = 0;
461         int full_search = 0;
462         int factor = 9;
463         int wrapped = 0;
464 again:
465         while (1) {
466                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
467                 if (!cache)
468                         break;
469
470                 spin_lock(&cache->lock);
471                 last = cache->key.objectid + cache->key.offset;
472                 used = btrfs_block_group_used(&cache->item);
473
474                 if ((full_search || !cache->ro) &&
475                     block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
476                         if (used + cache->pinned + cache->reserved <
477                             div_factor(cache->key.offset, factor)) {
478                                 group_start = cache->key.objectid;
479                                 spin_unlock(&cache->lock);
480                                 btrfs_put_block_group(cache);
481                                 goto found;
482                         }
483                 }
484                 spin_unlock(&cache->lock);
485                 btrfs_put_block_group(cache);
486                 cond_resched();
487         }
488         if (!wrapped) {
489                 last = search_start;
490                 wrapped = 1;
491                 goto again;
492         }
493         if (!full_search && factor < 10) {
494                 last = search_start;
495                 full_search = 1;
496                 factor = 10;
497                 goto again;
498         }
499 found:
500         return group_start;
501 }
502
503 /* simple helper to search for an existing extent at a given offset */
504 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
505 {
506         int ret;
507         struct btrfs_key key;
508         struct btrfs_path *path;
509
510         path = btrfs_alloc_path();
511         BUG_ON(!path);
512         key.objectid = start;
513         key.offset = len;
514         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
515         ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
516                                 0, 0);
517         btrfs_free_path(path);
518         return ret;
519 }
520
521 /*
522  * Back reference rules.  Back refs have three main goals:
523  *
524  * 1) differentiate between all holders of references to an extent so that
525  *    when a reference is dropped we can make sure it was a valid reference
526  *    before freeing the extent.
527  *
528  * 2) Provide enough information to quickly find the holders of an extent
529  *    if we notice a given block is corrupted or bad.
530  *
531  * 3) Make it easy to migrate blocks for FS shrinking or storage pool
532  *    maintenance.  This is actually the same as #2, but with a slightly
533  *    different use case.
534  *
535  * There are two kinds of back refs. The implicit back refs is optimized
536  * for pointers in non-shared tree blocks. For a given pointer in a block,
537  * back refs of this kind provide information about the block's owner tree
538  * and the pointer's key. These information allow us to find the block by
539  * b-tree searching. The full back refs is for pointers in tree blocks not
540  * referenced by their owner trees. The location of tree block is recorded
541  * in the back refs. Actually the full back refs is generic, and can be
542  * used in all cases the implicit back refs is used. The major shortcoming
543  * of the full back refs is its overhead. Every time a tree block gets
544  * COWed, we have to update back refs entry for all pointers in it.
545  *
546  * For a newly allocated tree block, we use implicit back refs for
547  * pointers in it. This means most tree related operations only involve
548  * implicit back refs. For a tree block created in old transaction, the
549  * only way to drop a reference to it is COW it. So we can detect the
550  * event that tree block loses its owner tree's reference and do the
551  * back refs conversion.
552  *
553  * When a tree block is COW'd through a tree, there are four cases:
554  *
555  * The reference count of the block is one and the tree is the block's
556  * owner tree. Nothing to do in this case.
557  *
558  * The reference count of the block is one and the tree is not the
559  * block's owner tree. In this case, full back refs is used for pointers
560  * in the block. Remove these full back refs, add implicit back refs for
561  * every pointers in the new block.
562  *
563  * The reference count of the block is greater than one and the tree is
564  * the block's owner tree. In this case, implicit back refs is used for
565  * pointers in the block. Add full back refs for every pointers in the
566  * block, increase lower level extents' reference counts. The original
567  * implicit back refs are entailed to the new block.
568  *
569  * The reference count of the block is greater than one and the tree is
570  * not the block's owner tree. Add implicit back refs for every pointer in
571  * the new block, increase lower level extents' reference count.
572  *
573  * Back Reference Key composing:
574  *
575  * The key objectid corresponds to the first byte in the extent,
576  * The key type is used to differentiate between types of back refs.
577  * There are different meanings of the key offset for different types
578  * of back refs.
579  *
580  * File extents can be referenced by:
581  *
582  * - multiple snapshots, subvolumes, or different generations in one subvol
583  * - different files inside a single subvolume
584  * - different offsets inside a file (bookend extents in file.c)
585  *
586  * The extent ref structure for the implicit back refs has fields for:
587  *
588  * - Objectid of the subvolume root
589  * - objectid of the file holding the reference
590  * - original offset in the file
591  * - how many bookend extents
592  *
593  * The key offset for the implicit back refs is hash of the first
594  * three fields.
595  *
596  * The extent ref structure for the full back refs has field for:
597  *
598  * - number of pointers in the tree leaf
599  *
600  * The key offset for the implicit back refs is the first byte of
601  * the tree leaf
602  *
603  * When a file extent is allocated, The implicit back refs is used.
604  * the fields are filled in:
605  *
606  *     (root_key.objectid, inode objectid, offset in file, 1)
607  *
608  * When a file extent is removed file truncation, we find the
609  * corresponding implicit back refs and check the following fields:
610  *
611  *     (btrfs_header_owner(leaf), inode objectid, offset in file)
612  *
613  * Btree extents can be referenced by:
614  *
615  * - Different subvolumes
616  *
617  * Both the implicit back refs and the full back refs for tree blocks
618  * only consist of key. The key offset for the implicit back refs is
619  * objectid of block's owner tree. The key offset for the full back refs
620  * is the first byte of parent block.
621  *
622  * When implicit back refs is used, information about the lowest key and
623  * level of the tree block are required. These information are stored in
624  * tree block info structure.
625  */
626
627 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
628 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
629                                   struct btrfs_root *root,
630                                   struct btrfs_path *path,
631                                   u64 owner, u32 extra_size)
632 {
633         struct btrfs_extent_item *item;
634         struct btrfs_extent_item_v0 *ei0;
635         struct btrfs_extent_ref_v0 *ref0;
636         struct btrfs_tree_block_info *bi;
637         struct extent_buffer *leaf;
638         struct btrfs_key key;
639         struct btrfs_key found_key;
640         u32 new_size = sizeof(*item);
641         u64 refs;
642         int ret;
643
644         leaf = path->nodes[0];
645         BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
646
647         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
648         ei0 = btrfs_item_ptr(leaf, path->slots[0],
649                              struct btrfs_extent_item_v0);
650         refs = btrfs_extent_refs_v0(leaf, ei0);
651
652         if (owner == (u64)-1) {
653                 while (1) {
654                         if (path->slots[0] >= btrfs_header_nritems(leaf)) {
655                                 ret = btrfs_next_leaf(root, path);
656                                 if (ret < 0)
657                                         return ret;
658                                 BUG_ON(ret > 0);
659                                 leaf = path->nodes[0];
660                         }
661                         btrfs_item_key_to_cpu(leaf, &found_key,
662                                               path->slots[0]);
663                         BUG_ON(key.objectid != found_key.objectid);
664                         if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
665                                 path->slots[0]++;
666                                 continue;
667                         }
668                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
669                                               struct btrfs_extent_ref_v0);
670                         owner = btrfs_ref_objectid_v0(leaf, ref0);
671                         break;
672                 }
673         }
674         btrfs_release_path(root, path);
675
676         if (owner < BTRFS_FIRST_FREE_OBJECTID)
677                 new_size += sizeof(*bi);
678
679         new_size -= sizeof(*ei0);
680         ret = btrfs_search_slot(trans, root, &key, path,
681                                 new_size + extra_size, 1);
682         if (ret < 0)
683                 return ret;
684         BUG_ON(ret);
685
686         ret = btrfs_extend_item(trans, root, path, new_size);
687         BUG_ON(ret);
688
689         leaf = path->nodes[0];
690         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
691         btrfs_set_extent_refs(leaf, item, refs);
692         /* FIXME: get real generation */
693         btrfs_set_extent_generation(leaf, item, 0);
694         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
695                 btrfs_set_extent_flags(leaf, item,
696                                        BTRFS_EXTENT_FLAG_TREE_BLOCK |
697                                        BTRFS_BLOCK_FLAG_FULL_BACKREF);
698                 bi = (struct btrfs_tree_block_info *)(item + 1);
699                 /* FIXME: get first key of the block */
700                 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
701                 btrfs_set_tree_block_level(leaf, bi, (int)owner);
702         } else {
703                 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
704         }
705         btrfs_mark_buffer_dirty(leaf);
706         return 0;
707 }
708 #endif
709
710 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
711 {
712         u32 high_crc = ~(u32)0;
713         u32 low_crc = ~(u32)0;
714         __le64 lenum;
715
716         lenum = cpu_to_le64(root_objectid);
717         high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
718         lenum = cpu_to_le64(owner);
719         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
720         lenum = cpu_to_le64(offset);
721         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
722
723         return ((u64)high_crc << 31) ^ (u64)low_crc;
724 }
725
726 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
727                                      struct btrfs_extent_data_ref *ref)
728 {
729         return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
730                                     btrfs_extent_data_ref_objectid(leaf, ref),
731                                     btrfs_extent_data_ref_offset(leaf, ref));
732 }
733
734 static int match_extent_data_ref(struct extent_buffer *leaf,
735                                  struct btrfs_extent_data_ref *ref,
736                                  u64 root_objectid, u64 owner, u64 offset)
737 {
738         if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
739             btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
740             btrfs_extent_data_ref_offset(leaf, ref) != offset)
741                 return 0;
742         return 1;
743 }
744
745 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
746                                            struct btrfs_root *root,
747                                            struct btrfs_path *path,
748                                            u64 bytenr, u64 parent,
749                                            u64 root_objectid,
750                                            u64 owner, u64 offset)
751 {
752         struct btrfs_key key;
753         struct btrfs_extent_data_ref *ref;
754         struct extent_buffer *leaf;
755         u32 nritems;
756         int ret;
757         int recow;
758         int err = -ENOENT;
759
760         key.objectid = bytenr;
761         if (parent) {
762                 key.type = BTRFS_SHARED_DATA_REF_KEY;
763                 key.offset = parent;
764         } else {
765                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
766                 key.offset = hash_extent_data_ref(root_objectid,
767                                                   owner, offset);
768         }
769 again:
770         recow = 0;
771         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
772         if (ret < 0) {
773                 err = ret;
774                 goto fail;
775         }
776
777         if (parent) {
778                 if (!ret)
779                         return 0;
780 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
781                 key.type = BTRFS_EXTENT_REF_V0_KEY;
782                 btrfs_release_path(root, path);
783                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
784                 if (ret < 0) {
785                         err = ret;
786                         goto fail;
787                 }
788                 if (!ret)
789                         return 0;
790 #endif
791                 goto fail;
792         }
793
794         leaf = path->nodes[0];
795         nritems = btrfs_header_nritems(leaf);
796         while (1) {
797                 if (path->slots[0] >= nritems) {
798                         ret = btrfs_next_leaf(root, path);
799                         if (ret < 0)
800                                 err = ret;
801                         if (ret)
802                                 goto fail;
803
804                         leaf = path->nodes[0];
805                         nritems = btrfs_header_nritems(leaf);
806                         recow = 1;
807                 }
808
809                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
810                 if (key.objectid != bytenr ||
811                     key.type != BTRFS_EXTENT_DATA_REF_KEY)
812                         goto fail;
813
814                 ref = btrfs_item_ptr(leaf, path->slots[0],
815                                      struct btrfs_extent_data_ref);
816
817                 if (match_extent_data_ref(leaf, ref, root_objectid,
818                                           owner, offset)) {
819                         if (recow) {
820                                 btrfs_release_path(root, path);
821                                 goto again;
822                         }
823                         err = 0;
824                         break;
825                 }
826                 path->slots[0]++;
827         }
828 fail:
829         return err;
830 }
831
832 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
833                                            struct btrfs_root *root,
834                                            struct btrfs_path *path,
835                                            u64 bytenr, u64 parent,
836                                            u64 root_objectid, u64 owner,
837                                            u64 offset, int refs_to_add)
838 {
839         struct btrfs_key key;
840         struct extent_buffer *leaf;
841         u32 size;
842         u32 num_refs;
843         int ret;
844
845         key.objectid = bytenr;
846         if (parent) {
847                 key.type = BTRFS_SHARED_DATA_REF_KEY;
848                 key.offset = parent;
849                 size = sizeof(struct btrfs_shared_data_ref);
850         } else {
851                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
852                 key.offset = hash_extent_data_ref(root_objectid,
853                                                   owner, offset);
854                 size = sizeof(struct btrfs_extent_data_ref);
855         }
856
857         ret = btrfs_insert_empty_item(trans, root, path, &key, size);
858         if (ret && ret != -EEXIST)
859                 goto fail;
860
861         leaf = path->nodes[0];
862         if (parent) {
863                 struct btrfs_shared_data_ref *ref;
864                 ref = btrfs_item_ptr(leaf, path->slots[0],
865                                      struct btrfs_shared_data_ref);
866                 if (ret == 0) {
867                         btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
868                 } else {
869                         num_refs = btrfs_shared_data_ref_count(leaf, ref);
870                         num_refs += refs_to_add;
871                         btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
872                 }
873         } else {
874                 struct btrfs_extent_data_ref *ref;
875                 while (ret == -EEXIST) {
876                         ref = btrfs_item_ptr(leaf, path->slots[0],
877                                              struct btrfs_extent_data_ref);
878                         if (match_extent_data_ref(leaf, ref, root_objectid,
879                                                   owner, offset))
880                                 break;
881                         btrfs_release_path(root, path);
882                         key.offset++;
883                         ret = btrfs_insert_empty_item(trans, root, path, &key,
884                                                       size);
885                         if (ret && ret != -EEXIST)
886                                 goto fail;
887
888                         leaf = path->nodes[0];
889                 }
890                 ref = btrfs_item_ptr(leaf, path->slots[0],
891                                      struct btrfs_extent_data_ref);
892                 if (ret == 0) {
893                         btrfs_set_extent_data_ref_root(leaf, ref,
894                                                        root_objectid);
895                         btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
896                         btrfs_set_extent_data_ref_offset(leaf, ref, offset);
897                         btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
898                 } else {
899                         num_refs = btrfs_extent_data_ref_count(leaf, ref);
900                         num_refs += refs_to_add;
901                         btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
902                 }
903         }
904         btrfs_mark_buffer_dirty(leaf);
905         ret = 0;
906 fail:
907         btrfs_release_path(root, path);
908         return ret;
909 }
910
911 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
912                                            struct btrfs_root *root,
913                                            struct btrfs_path *path,
914                                            int refs_to_drop)
915 {
916         struct btrfs_key key;
917         struct btrfs_extent_data_ref *ref1 = NULL;
918         struct btrfs_shared_data_ref *ref2 = NULL;
919         struct extent_buffer *leaf;
920         u32 num_refs = 0;
921         int ret = 0;
922
923         leaf = path->nodes[0];
924         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
925
926         if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
927                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
928                                       struct btrfs_extent_data_ref);
929                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
930         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
931                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
932                                       struct btrfs_shared_data_ref);
933                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
934 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
935         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
936                 struct btrfs_extent_ref_v0 *ref0;
937                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
938                                       struct btrfs_extent_ref_v0);
939                 num_refs = btrfs_ref_count_v0(leaf, ref0);
940 #endif
941         } else {
942                 BUG();
943         }
944
945         BUG_ON(num_refs < refs_to_drop);
946         num_refs -= refs_to_drop;
947
948         if (num_refs == 0) {
949                 ret = btrfs_del_item(trans, root, path);
950         } else {
951                 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
952                         btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
953                 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
954                         btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
955 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
956                 else {
957                         struct btrfs_extent_ref_v0 *ref0;
958                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
959                                         struct btrfs_extent_ref_v0);
960                         btrfs_set_ref_count_v0(leaf, ref0, num_refs);
961                 }
962 #endif
963                 btrfs_mark_buffer_dirty(leaf);
964         }
965         return ret;
966 }
967
968 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
969                                           struct btrfs_path *path,
970                                           struct btrfs_extent_inline_ref *iref)
971 {
972         struct btrfs_key key;
973         struct extent_buffer *leaf;
974         struct btrfs_extent_data_ref *ref1;
975         struct btrfs_shared_data_ref *ref2;
976         u32 num_refs = 0;
977
978         leaf = path->nodes[0];
979         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
980         if (iref) {
981                 if (btrfs_extent_inline_ref_type(leaf, iref) ==
982                     BTRFS_EXTENT_DATA_REF_KEY) {
983                         ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
984                         num_refs = btrfs_extent_data_ref_count(leaf, ref1);
985                 } else {
986                         ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
987                         num_refs = btrfs_shared_data_ref_count(leaf, ref2);
988                 }
989         } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
990                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
991                                       struct btrfs_extent_data_ref);
992                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
993         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
994                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
995                                       struct btrfs_shared_data_ref);
996                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
997 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
998         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
999                 struct btrfs_extent_ref_v0 *ref0;
1000                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1001                                       struct btrfs_extent_ref_v0);
1002                 num_refs = btrfs_ref_count_v0(leaf, ref0);
1003 #endif
1004         } else {
1005                 WARN_ON(1);
1006         }
1007         return num_refs;
1008 }
1009
1010 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1011                                           struct btrfs_root *root,
1012                                           struct btrfs_path *path,
1013                                           u64 bytenr, u64 parent,
1014                                           u64 root_objectid)
1015 {
1016         struct btrfs_key key;
1017         int ret;
1018
1019         key.objectid = bytenr;
1020         if (parent) {
1021                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1022                 key.offset = parent;
1023         } else {
1024                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1025                 key.offset = root_objectid;
1026         }
1027
1028         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1029         if (ret > 0)
1030                 ret = -ENOENT;
1031 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1032         if (ret == -ENOENT && parent) {
1033                 btrfs_release_path(root, path);
1034                 key.type = BTRFS_EXTENT_REF_V0_KEY;
1035                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1036                 if (ret > 0)
1037                         ret = -ENOENT;
1038         }
1039 #endif
1040         return ret;
1041 }
1042
1043 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1044                                           struct btrfs_root *root,
1045                                           struct btrfs_path *path,
1046                                           u64 bytenr, u64 parent,
1047                                           u64 root_objectid)
1048 {
1049         struct btrfs_key key;
1050         int ret;
1051
1052         key.objectid = bytenr;
1053         if (parent) {
1054                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1055                 key.offset = parent;
1056         } else {
1057                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1058                 key.offset = root_objectid;
1059         }
1060
1061         ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1062         btrfs_release_path(root, path);
1063         return ret;
1064 }
1065
1066 static inline int extent_ref_type(u64 parent, u64 owner)
1067 {
1068         int type;
1069         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1070                 if (parent > 0)
1071                         type = BTRFS_SHARED_BLOCK_REF_KEY;
1072                 else
1073                         type = BTRFS_TREE_BLOCK_REF_KEY;
1074         } else {
1075                 if (parent > 0)
1076                         type = BTRFS_SHARED_DATA_REF_KEY;
1077                 else
1078                         type = BTRFS_EXTENT_DATA_REF_KEY;
1079         }
1080         return type;
1081 }
1082
1083 static int find_next_key(struct btrfs_path *path, int level,
1084                          struct btrfs_key *key)
1085
1086 {
1087         for (; level < BTRFS_MAX_LEVEL; level++) {
1088                 if (!path->nodes[level])
1089                         break;
1090                 if (path->slots[level] + 1 >=
1091                     btrfs_header_nritems(path->nodes[level]))
1092                         continue;
1093                 if (level == 0)
1094                         btrfs_item_key_to_cpu(path->nodes[level], key,
1095                                               path->slots[level] + 1);
1096                 else
1097                         btrfs_node_key_to_cpu(path->nodes[level], key,
1098                                               path->slots[level] + 1);
1099                 return 0;
1100         }
1101         return 1;
1102 }
1103
1104 /*
1105  * look for inline back ref. if back ref is found, *ref_ret is set
1106  * to the address of inline back ref, and 0 is returned.
1107  *
1108  * if back ref isn't found, *ref_ret is set to the address where it
1109  * should be inserted, and -ENOENT is returned.
1110  *
1111  * if insert is true and there are too many inline back refs, the path
1112  * points to the extent item, and -EAGAIN is returned.
1113  *
1114  * NOTE: inline back refs are ordered in the same way that back ref
1115  *       items in the tree are ordered.
1116  */
1117 static noinline_for_stack
1118 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1119                                  struct btrfs_root *root,
1120                                  struct btrfs_path *path,
1121                                  struct btrfs_extent_inline_ref **ref_ret,
1122                                  u64 bytenr, u64 num_bytes,
1123                                  u64 parent, u64 root_objectid,
1124                                  u64 owner, u64 offset, int insert)
1125 {
1126         struct btrfs_key key;
1127         struct extent_buffer *leaf;
1128         struct btrfs_extent_item *ei;
1129         struct btrfs_extent_inline_ref *iref;
1130         u64 flags;
1131         u64 item_size;
1132         unsigned long ptr;
1133         unsigned long end;
1134         int extra_size;
1135         int type;
1136         int want;
1137         int ret;
1138         int err = 0;
1139
1140         key.objectid = bytenr;
1141         key.type = BTRFS_EXTENT_ITEM_KEY;
1142         key.offset = num_bytes;
1143
1144         want = extent_ref_type(parent, owner);
1145         if (insert) {
1146                 extra_size = btrfs_extent_inline_ref_size(want);
1147                 path->keep_locks = 1;
1148         } else
1149                 extra_size = -1;
1150         ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1151         if (ret < 0) {
1152                 err = ret;
1153                 goto out;
1154         }
1155         BUG_ON(ret);
1156
1157         leaf = path->nodes[0];
1158         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1159 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1160         if (item_size < sizeof(*ei)) {
1161                 if (!insert) {
1162                         err = -ENOENT;
1163                         goto out;
1164                 }
1165                 ret = convert_extent_item_v0(trans, root, path, owner,
1166                                              extra_size);
1167                 if (ret < 0) {
1168                         err = ret;
1169                         goto out;
1170                 }
1171                 leaf = path->nodes[0];
1172                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1173         }
1174 #endif
1175         BUG_ON(item_size < sizeof(*ei));
1176
1177         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1178         flags = btrfs_extent_flags(leaf, ei);
1179
1180         ptr = (unsigned long)(ei + 1);
1181         end = (unsigned long)ei + item_size;
1182
1183         if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1184                 ptr += sizeof(struct btrfs_tree_block_info);
1185                 BUG_ON(ptr > end);
1186         } else {
1187                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1188         }
1189
1190         err = -ENOENT;
1191         while (1) {
1192                 if (ptr >= end) {
1193                         WARN_ON(ptr > end);
1194                         break;
1195                 }
1196                 iref = (struct btrfs_extent_inline_ref *)ptr;
1197                 type = btrfs_extent_inline_ref_type(leaf, iref);
1198                 if (want < type)
1199                         break;
1200                 if (want > type) {
1201                         ptr += btrfs_extent_inline_ref_size(type);
1202                         continue;
1203                 }
1204
1205                 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1206                         struct btrfs_extent_data_ref *dref;
1207                         dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1208                         if (match_extent_data_ref(leaf, dref, root_objectid,
1209                                                   owner, offset)) {
1210                                 err = 0;
1211                                 break;
1212                         }
1213                         if (hash_extent_data_ref_item(leaf, dref) <
1214                             hash_extent_data_ref(root_objectid, owner, offset))
1215                                 break;
1216                 } else {
1217                         u64 ref_offset;
1218                         ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1219                         if (parent > 0) {
1220                                 if (parent == ref_offset) {
1221                                         err = 0;
1222                                         break;
1223                                 }
1224                                 if (ref_offset < parent)
1225                                         break;
1226                         } else {
1227                                 if (root_objectid == ref_offset) {
1228                                         err = 0;
1229                                         break;
1230                                 }
1231                                 if (ref_offset < root_objectid)
1232                                         break;
1233                         }
1234                 }
1235                 ptr += btrfs_extent_inline_ref_size(type);
1236         }
1237         if (err == -ENOENT && insert) {
1238                 if (item_size + extra_size >=
1239                     BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1240                         err = -EAGAIN;
1241                         goto out;
1242                 }
1243                 /*
1244                  * To add new inline back ref, we have to make sure
1245                  * there is no corresponding back ref item.
1246                  * For simplicity, we just do not add new inline back
1247                  * ref if there is any kind of item for this block
1248                  */
1249                 if (find_next_key(path, 0, &key) == 0 &&
1250                     key.objectid == bytenr &&
1251                     key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1252                         err = -EAGAIN;
1253                         goto out;
1254                 }
1255         }
1256         *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1257 out:
1258         if (insert) {
1259                 path->keep_locks = 0;
1260                 btrfs_unlock_up_safe(path, 1);
1261         }
1262         return err;
1263 }
1264
1265 /*
1266  * helper to add new inline back ref
1267  */
1268 static noinline_for_stack
1269 int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1270                                 struct btrfs_root *root,
1271                                 struct btrfs_path *path,
1272                                 struct btrfs_extent_inline_ref *iref,
1273                                 u64 parent, u64 root_objectid,
1274                                 u64 owner, u64 offset, int refs_to_add,
1275                                 struct btrfs_delayed_extent_op *extent_op)
1276 {
1277         struct extent_buffer *leaf;
1278         struct btrfs_extent_item *ei;
1279         unsigned long ptr;
1280         unsigned long end;
1281         unsigned long item_offset;
1282         u64 refs;
1283         int size;
1284         int type;
1285         int ret;
1286
1287         leaf = path->nodes[0];
1288         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1289         item_offset = (unsigned long)iref - (unsigned long)ei;
1290
1291         type = extent_ref_type(parent, owner);
1292         size = btrfs_extent_inline_ref_size(type);
1293
1294         ret = btrfs_extend_item(trans, root, path, size);
1295         BUG_ON(ret);
1296
1297         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1298         refs = btrfs_extent_refs(leaf, ei);
1299         refs += refs_to_add;
1300         btrfs_set_extent_refs(leaf, ei, refs);
1301         if (extent_op)
1302                 __run_delayed_extent_op(extent_op, leaf, ei);
1303
1304         ptr = (unsigned long)ei + item_offset;
1305         end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1306         if (ptr < end - size)
1307                 memmove_extent_buffer(leaf, ptr + size, ptr,
1308                                       end - size - ptr);
1309
1310         iref = (struct btrfs_extent_inline_ref *)ptr;
1311         btrfs_set_extent_inline_ref_type(leaf, iref, type);
1312         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1313                 struct btrfs_extent_data_ref *dref;
1314                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1315                 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1316                 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1317                 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1318                 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1319         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1320                 struct btrfs_shared_data_ref *sref;
1321                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1322                 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1323                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1324         } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1325                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1326         } else {
1327                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1328         }
1329         btrfs_mark_buffer_dirty(leaf);
1330         return 0;
1331 }
1332
1333 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1334                                  struct btrfs_root *root,
1335                                  struct btrfs_path *path,
1336                                  struct btrfs_extent_inline_ref **ref_ret,
1337                                  u64 bytenr, u64 num_bytes, u64 parent,
1338                                  u64 root_objectid, u64 owner, u64 offset)
1339 {
1340         int ret;
1341
1342         ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1343                                            bytenr, num_bytes, parent,
1344                                            root_objectid, owner, offset, 0);
1345         if (ret != -ENOENT)
1346                 return ret;
1347
1348         btrfs_release_path(root, path);
1349         *ref_ret = NULL;
1350
1351         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1352                 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1353                                             root_objectid);
1354         } else {
1355                 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1356                                              root_objectid, owner, offset);
1357         }
1358         return ret;
1359 }
1360
1361 /*
1362  * helper to update/remove inline back ref
1363  */
1364 static noinline_for_stack
1365 int update_inline_extent_backref(struct btrfs_trans_handle *trans,
1366                                  struct btrfs_root *root,
1367                                  struct btrfs_path *path,
1368                                  struct btrfs_extent_inline_ref *iref,
1369                                  int refs_to_mod,
1370                                  struct btrfs_delayed_extent_op *extent_op)
1371 {
1372         struct extent_buffer *leaf;
1373         struct btrfs_extent_item *ei;
1374         struct btrfs_extent_data_ref *dref = NULL;
1375         struct btrfs_shared_data_ref *sref = NULL;
1376         unsigned long ptr;
1377         unsigned long end;
1378         u32 item_size;
1379         int size;
1380         int type;
1381         int ret;
1382         u64 refs;
1383
1384         leaf = path->nodes[0];
1385         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1386         refs = btrfs_extent_refs(leaf, ei);
1387         WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1388         refs += refs_to_mod;
1389         btrfs_set_extent_refs(leaf, ei, refs);
1390         if (extent_op)
1391                 __run_delayed_extent_op(extent_op, leaf, ei);
1392
1393         type = btrfs_extent_inline_ref_type(leaf, iref);
1394
1395         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1396                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1397                 refs = btrfs_extent_data_ref_count(leaf, dref);
1398         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1399                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1400                 refs = btrfs_shared_data_ref_count(leaf, sref);
1401         } else {
1402                 refs = 1;
1403                 BUG_ON(refs_to_mod != -1);
1404         }
1405
1406         BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1407         refs += refs_to_mod;
1408
1409         if (refs > 0) {
1410                 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1411                         btrfs_set_extent_data_ref_count(leaf, dref, refs);
1412                 else
1413                         btrfs_set_shared_data_ref_count(leaf, sref, refs);
1414         } else {
1415                 size =  btrfs_extent_inline_ref_size(type);
1416                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1417                 ptr = (unsigned long)iref;
1418                 end = (unsigned long)ei + item_size;
1419                 if (ptr + size < end)
1420                         memmove_extent_buffer(leaf, ptr, ptr + size,
1421                                               end - ptr - size);
1422                 item_size -= size;
1423                 ret = btrfs_truncate_item(trans, root, path, item_size, 1);
1424                 BUG_ON(ret);
1425         }
1426         btrfs_mark_buffer_dirty(leaf);
1427         return 0;
1428 }
1429
1430 static noinline_for_stack
1431 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1432                                  struct btrfs_root *root,
1433                                  struct btrfs_path *path,
1434                                  u64 bytenr, u64 num_bytes, u64 parent,
1435                                  u64 root_objectid, u64 owner,
1436                                  u64 offset, int refs_to_add,
1437                                  struct btrfs_delayed_extent_op *extent_op)
1438 {
1439         struct btrfs_extent_inline_ref *iref;
1440         int ret;
1441
1442         ret = lookup_inline_extent_backref(trans, root, path, &iref,
1443                                            bytenr, num_bytes, parent,
1444                                            root_objectid, owner, offset, 1);
1445         if (ret == 0) {
1446                 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1447                 ret = update_inline_extent_backref(trans, root, path, iref,
1448                                                    refs_to_add, extent_op);
1449         } else if (ret == -ENOENT) {
1450                 ret = setup_inline_extent_backref(trans, root, path, iref,
1451                                                   parent, root_objectid,
1452                                                   owner, offset, refs_to_add,
1453                                                   extent_op);
1454         }
1455         return ret;
1456 }
1457
1458 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1459                                  struct btrfs_root *root,
1460                                  struct btrfs_path *path,
1461                                  u64 bytenr, u64 parent, u64 root_objectid,
1462                                  u64 owner, u64 offset, int refs_to_add)
1463 {
1464         int ret;
1465         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1466                 BUG_ON(refs_to_add != 1);
1467                 ret = insert_tree_block_ref(trans, root, path, bytenr,
1468                                             parent, root_objectid);
1469         } else {
1470                 ret = insert_extent_data_ref(trans, root, path, bytenr,
1471                                              parent, root_objectid,
1472                                              owner, offset, refs_to_add);
1473         }
1474         return ret;
1475 }
1476
1477 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1478                                  struct btrfs_root *root,
1479                                  struct btrfs_path *path,
1480                                  struct btrfs_extent_inline_ref *iref,
1481                                  int refs_to_drop, int is_data)
1482 {
1483         int ret;
1484
1485         BUG_ON(!is_data && refs_to_drop != 1);
1486         if (iref) {
1487                 ret = update_inline_extent_backref(trans, root, path, iref,
1488                                                    -refs_to_drop, NULL);
1489         } else if (is_data) {
1490                 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1491         } else {
1492                 ret = btrfs_del_item(trans, root, path);
1493         }
1494         return ret;
1495 }
1496
1497 #ifdef BIO_RW_DISCARD
1498 static void btrfs_issue_discard(struct block_device *bdev,
1499                                 u64 start, u64 len)
1500 {
1501         blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL);
1502 }
1503 #endif
1504
1505 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1506                                 u64 num_bytes)
1507 {
1508 #ifdef BIO_RW_DISCARD
1509         int ret;
1510         u64 map_length = num_bytes;
1511         struct btrfs_multi_bio *multi = NULL;
1512
1513         /* Tell the block device(s) that the sectors can be discarded */
1514         ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
1515                               bytenr, &map_length, &multi, 0);
1516         if (!ret) {
1517                 struct btrfs_bio_stripe *stripe = multi->stripes;
1518                 int i;
1519
1520                 if (map_length > num_bytes)
1521                         map_length = num_bytes;
1522
1523                 for (i = 0; i < multi->num_stripes; i++, stripe++) {
1524                         btrfs_issue_discard(stripe->dev->bdev,
1525                                             stripe->physical,
1526                                             map_length);
1527                 }
1528                 kfree(multi);
1529         }
1530
1531         return ret;
1532 #else
1533         return 0;
1534 #endif
1535 }
1536
1537 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1538                          struct btrfs_root *root,
1539                          u64 bytenr, u64 num_bytes, u64 parent,
1540                          u64 root_objectid, u64 owner, u64 offset)
1541 {
1542         int ret;
1543         BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1544                root_objectid == BTRFS_TREE_LOG_OBJECTID);
1545
1546         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1547                 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
1548                                         parent, root_objectid, (int)owner,
1549                                         BTRFS_ADD_DELAYED_REF, NULL);
1550         } else {
1551                 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
1552                                         parent, root_objectid, owner, offset,
1553                                         BTRFS_ADD_DELAYED_REF, NULL);
1554         }
1555         return ret;
1556 }
1557
1558 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1559                                   struct btrfs_root *root,
1560                                   u64 bytenr, u64 num_bytes,
1561                                   u64 parent, u64 root_objectid,
1562                                   u64 owner, u64 offset, int refs_to_add,
1563                                   struct btrfs_delayed_extent_op *extent_op)
1564 {
1565         struct btrfs_path *path;
1566         struct extent_buffer *leaf;
1567         struct btrfs_extent_item *item;
1568         u64 refs;
1569         int ret;
1570         int err = 0;
1571
1572         path = btrfs_alloc_path();
1573         if (!path)
1574                 return -ENOMEM;
1575
1576         path->reada = 1;
1577         path->leave_spinning = 1;
1578         /* this will setup the path even if it fails to insert the back ref */
1579         ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1580                                            path, bytenr, num_bytes, parent,
1581                                            root_objectid, owner, offset,
1582                                            refs_to_add, extent_op);
1583         if (ret == 0)
1584                 goto out;
1585
1586         if (ret != -EAGAIN) {
1587                 err = ret;
1588                 goto out;
1589         }
1590
1591         leaf = path->nodes[0];
1592         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1593         refs = btrfs_extent_refs(leaf, item);
1594         btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1595         if (extent_op)
1596                 __run_delayed_extent_op(extent_op, leaf, item);
1597
1598         btrfs_mark_buffer_dirty(leaf);
1599         btrfs_release_path(root->fs_info->extent_root, path);
1600
1601         path->reada = 1;
1602         path->leave_spinning = 1;
1603
1604         /* now insert the actual backref */
1605         ret = insert_extent_backref(trans, root->fs_info->extent_root,
1606                                     path, bytenr, parent, root_objectid,
1607                                     owner, offset, refs_to_add);
1608         BUG_ON(ret);
1609 out:
1610         btrfs_free_path(path);
1611         return err;
1612 }
1613
1614 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1615                                 struct btrfs_root *root,
1616                                 struct btrfs_delayed_ref_node *node,
1617                                 struct btrfs_delayed_extent_op *extent_op,
1618                                 int insert_reserved)
1619 {
1620         int ret = 0;
1621         struct btrfs_delayed_data_ref *ref;
1622         struct btrfs_key ins;
1623         u64 parent = 0;
1624         u64 ref_root = 0;
1625         u64 flags = 0;
1626
1627         ins.objectid = node->bytenr;
1628         ins.offset = node->num_bytes;
1629         ins.type = BTRFS_EXTENT_ITEM_KEY;
1630
1631         ref = btrfs_delayed_node_to_data_ref(node);
1632         if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1633                 parent = ref->parent;
1634         else
1635                 ref_root = ref->root;
1636
1637         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1638                 if (extent_op) {
1639                         BUG_ON(extent_op->update_key);
1640                         flags |= extent_op->flags_to_set;
1641                 }
1642                 ret = alloc_reserved_file_extent(trans, root,
1643                                                  parent, ref_root, flags,
1644                                                  ref->objectid, ref->offset,
1645                                                  &ins, node->ref_mod);
1646                 update_reserved_extents(root, ins.objectid, ins.offset, 0);
1647         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1648                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1649                                              node->num_bytes, parent,
1650                                              ref_root, ref->objectid,
1651                                              ref->offset, node->ref_mod,
1652                                              extent_op);
1653         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1654                 ret = __btrfs_free_extent(trans, root, node->bytenr,
1655                                           node->num_bytes, parent,
1656                                           ref_root, ref->objectid,
1657                                           ref->offset, node->ref_mod,
1658                                           extent_op);
1659         } else {
1660                 BUG();
1661         }
1662         return ret;
1663 }
1664
1665 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1666                                     struct extent_buffer *leaf,
1667                                     struct btrfs_extent_item *ei)
1668 {
1669         u64 flags = btrfs_extent_flags(leaf, ei);
1670         if (extent_op->update_flags) {
1671                 flags |= extent_op->flags_to_set;
1672                 btrfs_set_extent_flags(leaf, ei, flags);
1673         }
1674
1675         if (extent_op->update_key) {
1676                 struct btrfs_tree_block_info *bi;
1677                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
1678                 bi = (struct btrfs_tree_block_info *)(ei + 1);
1679                 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
1680         }
1681 }
1682
1683 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
1684                                  struct btrfs_root *root,
1685                                  struct btrfs_delayed_ref_node *node,
1686                                  struct btrfs_delayed_extent_op *extent_op)
1687 {
1688         struct btrfs_key key;
1689         struct btrfs_path *path;
1690         struct btrfs_extent_item *ei;
1691         struct extent_buffer *leaf;
1692         u32 item_size;
1693         int ret;
1694         int err = 0;
1695
1696         path = btrfs_alloc_path();
1697         if (!path)
1698                 return -ENOMEM;
1699
1700         key.objectid = node->bytenr;
1701         key.type = BTRFS_EXTENT_ITEM_KEY;
1702         key.offset = node->num_bytes;
1703
1704         path->reada = 1;
1705         path->leave_spinning = 1;
1706         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
1707                                 path, 0, 1);
1708         if (ret < 0) {
1709                 err = ret;
1710                 goto out;
1711         }
1712         if (ret > 0) {
1713                 err = -EIO;
1714                 goto out;
1715         }
1716
1717         leaf = path->nodes[0];
1718         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1719 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1720         if (item_size < sizeof(*ei)) {
1721                 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
1722                                              path, (u64)-1, 0);
1723                 if (ret < 0) {
1724                         err = ret;
1725                         goto out;
1726                 }
1727                 leaf = path->nodes[0];
1728                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1729         }
1730 #endif
1731         BUG_ON(item_size < sizeof(*ei));
1732         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1733         __run_delayed_extent_op(extent_op, leaf, ei);
1734
1735         btrfs_mark_buffer_dirty(leaf);
1736 out:
1737         btrfs_free_path(path);
1738         return err;
1739 }
1740
1741 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
1742                                 struct btrfs_root *root,
1743                                 struct btrfs_delayed_ref_node *node,
1744                                 struct btrfs_delayed_extent_op *extent_op,
1745                                 int insert_reserved)
1746 {
1747         int ret = 0;
1748         struct btrfs_delayed_tree_ref *ref;
1749         struct btrfs_key ins;
1750         u64 parent = 0;
1751         u64 ref_root = 0;
1752
1753         ins.objectid = node->bytenr;
1754         ins.offset = node->num_bytes;
1755         ins.type = BTRFS_EXTENT_ITEM_KEY;
1756
1757         ref = btrfs_delayed_node_to_tree_ref(node);
1758         if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1759                 parent = ref->parent;
1760         else
1761                 ref_root = ref->root;
1762
1763         BUG_ON(node->ref_mod != 1);
1764         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1765                 BUG_ON(!extent_op || !extent_op->update_flags ||
1766                        !extent_op->update_key);
1767                 ret = alloc_reserved_tree_block(trans, root,
1768                                                 parent, ref_root,
1769                                                 extent_op->flags_to_set,
1770                                                 &extent_op->key,
1771                                                 ref->level, &ins);
1772                 update_reserved_extents(root, ins.objectid, ins.offset, 0);
1773         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1774                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1775                                              node->num_bytes, parent, ref_root,
1776                                              ref->level, 0, 1, extent_op);
1777         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1778                 ret = __btrfs_free_extent(trans, root, node->bytenr,
1779                                           node->num_bytes, parent, ref_root,
1780                                           ref->level, 0, 1, extent_op);
1781         } else {
1782                 BUG();
1783         }
1784         return ret;
1785 }
1786
1787
1788 /* helper function to actually process a single delayed ref entry */
1789 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
1790                                struct btrfs_root *root,
1791                                struct btrfs_delayed_ref_node *node,
1792                                struct btrfs_delayed_extent_op *extent_op,
1793                                int insert_reserved)
1794 {
1795         int ret;
1796         if (btrfs_delayed_ref_is_head(node)) {
1797                 struct btrfs_delayed_ref_head *head;
1798                 /*
1799                  * we've hit the end of the chain and we were supposed
1800                  * to insert this extent into the tree.  But, it got
1801                  * deleted before we ever needed to insert it, so all
1802                  * we have to do is clean up the accounting
1803                  */
1804                 BUG_ON(extent_op);
1805                 head = btrfs_delayed_node_to_head(node);
1806                 if (insert_reserved) {
1807                         if (head->is_data) {
1808                                 ret = btrfs_del_csums(trans, root,
1809                                                       node->bytenr,
1810                                                       node->num_bytes);
1811                                 BUG_ON(ret);
1812                         }
1813                         btrfs_update_pinned_extents(root, node->bytenr,
1814                                                     node->num_bytes, 1);
1815                         update_reserved_extents(root, node->bytenr,
1816                                                 node->num_bytes, 0);
1817                 }
1818                 mutex_unlock(&head->mutex);
1819                 return 0;
1820         }
1821
1822         if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
1823             node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1824                 ret = run_delayed_tree_ref(trans, root, node, extent_op,
1825                                            insert_reserved);
1826         else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
1827                  node->type == BTRFS_SHARED_DATA_REF_KEY)
1828                 ret = run_delayed_data_ref(trans, root, node, extent_op,
1829                                            insert_reserved);
1830         else
1831                 BUG();
1832         return ret;
1833 }
1834
1835 static noinline struct btrfs_delayed_ref_node *
1836 select_delayed_ref(struct btrfs_delayed_ref_head *head)
1837 {
1838         struct rb_node *node;
1839         struct btrfs_delayed_ref_node *ref;
1840         int action = BTRFS_ADD_DELAYED_REF;
1841 again:
1842         /*
1843          * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
1844          * this prevents ref count from going down to zero when
1845          * there still are pending delayed ref.
1846          */
1847         node = rb_prev(&head->node.rb_node);
1848         while (1) {
1849                 if (!node)
1850                         break;
1851                 ref = rb_entry(node, struct btrfs_delayed_ref_node,
1852                                 rb_node);
1853                 if (ref->bytenr != head->node.bytenr)
1854                         break;
1855                 if (ref->action == action)
1856                         return ref;
1857                 node = rb_prev(node);
1858         }
1859         if (action == BTRFS_ADD_DELAYED_REF) {
1860                 action = BTRFS_DROP_DELAYED_REF;
1861                 goto again;
1862         }
1863         return NULL;
1864 }
1865
1866 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
1867                                        struct btrfs_root *root,
1868                                        struct list_head *cluster)
1869 {
1870         struct btrfs_delayed_ref_root *delayed_refs;
1871         struct btrfs_delayed_ref_node *ref;
1872         struct btrfs_delayed_ref_head *locked_ref = NULL;
1873         struct btrfs_delayed_extent_op *extent_op;
1874         int ret;
1875         int count = 0;
1876         int must_insert_reserved = 0;
1877
1878         delayed_refs = &trans->transaction->delayed_refs;
1879         while (1) {
1880                 if (!locked_ref) {
1881                         /* pick a new head ref from the cluster list */
1882                         if (list_empty(cluster))
1883                                 break;
1884
1885                         locked_ref = list_entry(cluster->next,
1886                                      struct btrfs_delayed_ref_head, cluster);
1887
1888                         /* grab the lock that says we are going to process
1889                          * all the refs for this head */
1890                         ret = btrfs_delayed_ref_lock(trans, locked_ref);
1891
1892                         /*
1893                          * we may have dropped the spin lock to get the head
1894                          * mutex lock, and that might have given someone else
1895                          * time to free the head.  If that's true, it has been
1896                          * removed from our list and we can move on.
1897                          */
1898                         if (ret == -EAGAIN) {
1899                                 locked_ref = NULL;
1900                                 count++;
1901                                 continue;
1902                         }
1903                 }
1904
1905                 /*
1906                  * record the must insert reserved flag before we
1907                  * drop the spin lock.
1908                  */
1909                 must_insert_reserved = locked_ref->must_insert_reserved;
1910                 locked_ref->must_insert_reserved = 0;
1911
1912                 extent_op = locked_ref->extent_op;
1913                 locked_ref->extent_op = NULL;
1914
1915                 /*
1916                  * locked_ref is the head node, so we have to go one
1917                  * node back for any delayed ref updates
1918                  */
1919                 ref = select_delayed_ref(locked_ref);
1920                 if (!ref) {
1921                         /* All delayed refs have been processed, Go ahead
1922                          * and send the head node to run_one_delayed_ref,
1923                          * so that any accounting fixes can happen
1924                          */
1925                         ref = &locked_ref->node;
1926
1927                         if (extent_op && must_insert_reserved) {
1928                                 kfree(extent_op);
1929                                 extent_op = NULL;
1930                         }
1931
1932                         if (extent_op) {
1933                                 spin_unlock(&delayed_refs->lock);
1934
1935                                 ret = run_delayed_extent_op(trans, root,
1936                                                             ref, extent_op);
1937                                 BUG_ON(ret);
1938                                 kfree(extent_op);
1939
1940                                 cond_resched();
1941                                 spin_lock(&delayed_refs->lock);
1942                                 continue;
1943                         }
1944
1945                         list_del_init(&locked_ref->cluster);
1946                         locked_ref = NULL;
1947                 }
1948
1949                 ref->in_tree = 0;
1950                 rb_erase(&ref->rb_node, &delayed_refs->root);
1951                 delayed_refs->num_entries--;
1952
1953                 spin_unlock(&delayed_refs->lock);
1954
1955                 ret = run_one_delayed_ref(trans, root, ref, extent_op,
1956                                           must_insert_reserved);
1957                 BUG_ON(ret);
1958
1959                 btrfs_put_delayed_ref(ref);
1960                 kfree(extent_op);
1961                 count++;
1962
1963                 cond_resched();
1964                 spin_lock(&delayed_refs->lock);
1965         }
1966         return count;
1967 }
1968
1969 /*
1970  * this starts processing the delayed reference count updates and
1971  * extent insertions we have queued up so far.  count can be
1972  * 0, which means to process everything in the tree at the start
1973  * of the run (but not newly added entries), or it can be some target
1974  * number you'd like to process.
1975  */
1976 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
1977                            struct btrfs_root *root, unsigned long count)
1978 {
1979         struct rb_node *node;
1980         struct btrfs_delayed_ref_root *delayed_refs;
1981         struct btrfs_delayed_ref_node *ref;
1982         struct list_head cluster;
1983         int ret;
1984         int run_all = count == (unsigned long)-1;
1985         int run_most = 0;
1986
1987         if (root == root->fs_info->extent_root)
1988                 root = root->fs_info->tree_root;
1989
1990         delayed_refs = &trans->transaction->delayed_refs;
1991         INIT_LIST_HEAD(&cluster);
1992 again:
1993         spin_lock(&delayed_refs->lock);
1994         if (count == 0) {
1995                 count = delayed_refs->num_entries * 2;
1996                 run_most = 1;
1997         }
1998         while (1) {
1999                 if (!(run_all || run_most) &&
2000                     delayed_refs->num_heads_ready < 64)
2001                         break;
2002
2003                 /*
2004                  * go find something we can process in the rbtree.  We start at
2005                  * the beginning of the tree, and then build a cluster
2006                  * of refs to process starting at the first one we are able to
2007                  * lock
2008                  */
2009                 ret = btrfs_find_ref_cluster(trans, &cluster,
2010                                              delayed_refs->run_delayed_start);
2011                 if (ret)
2012                         break;
2013
2014                 ret = run_clustered_refs(trans, root, &cluster);
2015                 BUG_ON(ret < 0);
2016
2017                 count -= min_t(unsigned long, ret, count);
2018
2019                 if (count == 0)
2020                         break;
2021         }
2022
2023         if (run_all) {
2024                 node = rb_first(&delayed_refs->root);
2025                 if (!node)
2026                         goto out;
2027                 count = (unsigned long)-1;
2028
2029                 while (node) {
2030                         ref = rb_entry(node, struct btrfs_delayed_ref_node,
2031                                        rb_node);
2032                         if (btrfs_delayed_ref_is_head(ref)) {
2033                                 struct btrfs_delayed_ref_head *head;
2034
2035                                 head = btrfs_delayed_node_to_head(ref);
2036                                 atomic_inc(&ref->refs);
2037
2038                                 spin_unlock(&delayed_refs->lock);
2039                                 mutex_lock(&head->mutex);
2040                                 mutex_unlock(&head->mutex);
2041
2042                                 btrfs_put_delayed_ref(ref);
2043                                 cond_resched();
2044                                 goto again;
2045                         }
2046                         node = rb_next(node);
2047                 }
2048                 spin_unlock(&delayed_refs->lock);
2049                 schedule_timeout(1);
2050                 goto again;
2051         }
2052 out:
2053         spin_unlock(&delayed_refs->lock);
2054         return 0;
2055 }
2056
2057 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2058                                 struct btrfs_root *root,
2059                                 u64 bytenr, u64 num_bytes, u64 flags,
2060                                 int is_data)
2061 {
2062         struct btrfs_delayed_extent_op *extent_op;
2063         int ret;
2064
2065         extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2066         if (!extent_op)
2067                 return -ENOMEM;
2068
2069         extent_op->flags_to_set = flags;
2070         extent_op->update_flags = 1;
2071         extent_op->update_key = 0;
2072         extent_op->is_data = is_data ? 1 : 0;
2073
2074         ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op);
2075         if (ret)
2076                 kfree(extent_op);
2077         return ret;
2078 }
2079
2080 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2081                                       struct btrfs_root *root,
2082                                       struct btrfs_path *path,
2083                                       u64 objectid, u64 offset, u64 bytenr)
2084 {
2085         struct btrfs_delayed_ref_head *head;
2086         struct btrfs_delayed_ref_node *ref;
2087         struct btrfs_delayed_data_ref *data_ref;
2088         struct btrfs_delayed_ref_root *delayed_refs;
2089         struct rb_node *node;
2090         int ret = 0;
2091
2092         ret = -ENOENT;
2093         delayed_refs = &trans->transaction->delayed_refs;
2094         spin_lock(&delayed_refs->lock);
2095         head = btrfs_find_delayed_ref_head(trans, bytenr);
2096         if (!head)
2097                 goto out;
2098
2099         if (!mutex_trylock(&head->mutex)) {
2100                 atomic_inc(&head->node.refs);
2101                 spin_unlock(&delayed_refs->lock);
2102
2103                 btrfs_release_path(root->fs_info->extent_root, path);
2104
2105                 mutex_lock(&head->mutex);
2106                 mutex_unlock(&head->mutex);
2107                 btrfs_put_delayed_ref(&head->node);
2108                 return -EAGAIN;
2109         }
2110
2111         node = rb_prev(&head->node.rb_node);
2112         if (!node)
2113                 goto out_unlock;
2114
2115         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2116
2117         if (ref->bytenr != bytenr)
2118                 goto out_unlock;
2119
2120         ret = 1;
2121         if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2122                 goto out_unlock;
2123
2124         data_ref = btrfs_delayed_node_to_data_ref(ref);
2125
2126         node = rb_prev(node);
2127         if (node) {
2128                 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2129                 if (ref->bytenr == bytenr)
2130                         goto out_unlock;
2131         }
2132
2133         if (data_ref->root != root->root_key.objectid ||
2134             data_ref->objectid != objectid || data_ref->offset != offset)
2135                 goto out_unlock;
2136
2137         ret = 0;
2138 out_unlock:
2139         mutex_unlock(&head->mutex);
2140 out:
2141         spin_unlock(&delayed_refs->lock);
2142         return ret;
2143 }
2144
2145 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2146                                         struct btrfs_root *root,
2147                                         struct btrfs_path *path,
2148                                         u64 objectid, u64 offset, u64 bytenr)
2149 {
2150         struct btrfs_root *extent_root = root->fs_info->extent_root;
2151         struct extent_buffer *leaf;
2152         struct btrfs_extent_data_ref *ref;
2153         struct btrfs_extent_inline_ref *iref;
2154         struct btrfs_extent_item *ei;
2155         struct btrfs_key key;
2156         u32 item_size;
2157         int ret;
2158
2159         key.objectid = bytenr;
2160         key.offset = (u64)-1;
2161         key.type = BTRFS_EXTENT_ITEM_KEY;
2162
2163         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2164         if (ret < 0)
2165                 goto out;
2166         BUG_ON(ret == 0);
2167
2168         ret = -ENOENT;
2169         if (path->slots[0] == 0)
2170                 goto out;
2171
2172         path->slots[0]--;
2173         leaf = path->nodes[0];
2174         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2175
2176         if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2177                 goto out;
2178
2179         ret = 1;
2180         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2181 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2182         if (item_size < sizeof(*ei)) {
2183                 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2184                 goto out;
2185         }
2186 #endif
2187         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2188
2189         if (item_size != sizeof(*ei) +
2190             btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2191                 goto out;
2192
2193         if (btrfs_extent_generation(leaf, ei) <=
2194             btrfs_root_last_snapshot(&root->root_item))
2195                 goto out;
2196
2197         iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2198         if (btrfs_extent_inline_ref_type(leaf, iref) !=
2199             BTRFS_EXTENT_DATA_REF_KEY)
2200                 goto out;
2201
2202         ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2203         if (btrfs_extent_refs(leaf, ei) !=
2204             btrfs_extent_data_ref_count(leaf, ref) ||
2205             btrfs_extent_data_ref_root(leaf, ref) !=
2206             root->root_key.objectid ||
2207             btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2208             btrfs_extent_data_ref_offset(leaf, ref) != offset)
2209                 goto out;
2210
2211         ret = 0;
2212 out:
2213         return ret;
2214 }
2215
2216 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2217                           struct btrfs_root *root,
2218                           u64 objectid, u64 offset, u64 bytenr)
2219 {
2220         struct btrfs_path *path;
2221         int ret;
2222         int ret2;
2223
2224         path = btrfs_alloc_path();
2225         if (!path)
2226                 return -ENOENT;
2227
2228         do {
2229                 ret = check_committed_ref(trans, root, path, objectid,
2230                                           offset, bytenr);
2231                 if (ret && ret != -ENOENT)
2232                         goto out;
2233
2234                 ret2 = check_delayed_ref(trans, root, path, objectid,
2235                                          offset, bytenr);
2236         } while (ret2 == -EAGAIN);
2237
2238         if (ret2 && ret2 != -ENOENT) {
2239                 ret = ret2;
2240                 goto out;
2241         }
2242
2243         if (ret != -ENOENT || ret2 != -ENOENT)
2244                 ret = 0;
2245 out:
2246         btrfs_free_path(path);
2247         return ret;
2248 }
2249
2250 #if 0
2251 int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2252                     struct extent_buffer *buf, u32 nr_extents)
2253 {
2254         struct btrfs_key key;
2255         struct btrfs_file_extent_item *fi;
2256         u64 root_gen;
2257         u32 nritems;
2258         int i;
2259         int level;
2260         int ret = 0;
2261         int shared = 0;
2262
2263         if (!root->ref_cows)
2264                 return 0;
2265
2266         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
2267                 shared = 0;
2268                 root_gen = root->root_key.offset;
2269         } else {
2270                 shared = 1;
2271                 root_gen = trans->transid - 1;
2272         }
2273
2274         level = btrfs_header_level(buf);
2275         nritems = btrfs_header_nritems(buf);
2276
2277         if (level == 0) {
2278                 struct btrfs_leaf_ref *ref;
2279                 struct btrfs_extent_info *info;
2280
2281                 ref = btrfs_alloc_leaf_ref(root, nr_extents);
2282                 if (!ref) {
2283                         ret = -ENOMEM;
2284                         goto out;
2285                 }
2286
2287                 ref->root_gen = root_gen;
2288                 ref->bytenr = buf->start;
2289                 ref->owner = btrfs_header_owner(buf);
2290                 ref->generation = btrfs_header_generation(buf);
2291                 ref->nritems = nr_extents;
2292                 info = ref->extents;
2293
2294                 for (i = 0; nr_extents > 0 && i < nritems; i++) {
2295                         u64 disk_bytenr;
2296                         btrfs_item_key_to_cpu(buf, &key, i);
2297                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2298                                 continue;
2299                         fi = btrfs_item_ptr(buf, i,
2300                                             struct btrfs_file_extent_item);
2301                         if (btrfs_file_extent_type(buf, fi) ==
2302                             BTRFS_FILE_EXTENT_INLINE)
2303                                 continue;
2304                         disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2305                         if (disk_bytenr == 0)
2306                                 continue;
2307
2308                         info->bytenr = disk_bytenr;
2309                         info->num_bytes =
2310                                 btrfs_file_extent_disk_num_bytes(buf, fi);
2311                         info->objectid = key.objectid;
2312                         info->offset = key.offset;
2313                         info++;
2314                 }
2315
2316                 ret = btrfs_add_leaf_ref(root, ref, shared);
2317                 if (ret == -EEXIST && shared) {
2318                         struct btrfs_leaf_ref *old;
2319                         old = btrfs_lookup_leaf_ref(root, ref->bytenr);
2320                         BUG_ON(!old);
2321                         btrfs_remove_leaf_ref(root, old);
2322                         btrfs_free_leaf_ref(root, old);
2323                         ret = btrfs_add_leaf_ref(root, ref, shared);
2324                 }
2325                 WARN_ON(ret);
2326                 btrfs_free_leaf_ref(root, ref);
2327         }
2328 out:
2329         return ret;
2330 }
2331
2332 /* when a block goes through cow, we update the reference counts of
2333  * everything that block points to.  The internal pointers of the block
2334  * can be in just about any order, and it is likely to have clusters of
2335  * things that are close together and clusters of things that are not.
2336  *
2337  * To help reduce the seeks that come with updating all of these reference
2338  * counts, sort them by byte number before actual updates are done.
2339  *
2340  * struct refsort is used to match byte number to slot in the btree block.
2341  * we sort based on the byte number and then use the slot to actually
2342  * find the item.
2343  *
2344  * struct refsort is smaller than strcut btrfs_item and smaller than
2345  * struct btrfs_key_ptr.  Since we're currently limited to the page size
2346  * for a btree block, there's no way for a kmalloc of refsorts for a
2347  * single node to be bigger than a page.
2348  */
2349 struct refsort {
2350         u64 bytenr;
2351         u32 slot;
2352 };
2353
2354 /*
2355  * for passing into sort()
2356  */
2357 static int refsort_cmp(const void *a_void, const void *b_void)
2358 {
2359         const struct refsort *a = a_void;
2360         const struct refsort *b = b_void;
2361
2362         if (a->bytenr < b->bytenr)
2363                 return -1;
2364         if (a->bytenr > b->bytenr)
2365                 return 1;
2366         return 0;
2367 }
2368 #endif
2369
2370 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2371                            struct btrfs_root *root,
2372                            struct extent_buffer *buf,
2373                            int full_backref, int inc)
2374 {
2375         u64 bytenr;
2376         u64 num_bytes;
2377         u64 parent;
2378         u64 ref_root;
2379         u32 nritems;
2380         struct btrfs_key key;
2381         struct btrfs_file_extent_item *fi;
2382         int i;
2383         int level;
2384         int ret = 0;
2385         int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2386                             u64, u64, u64, u64, u64, u64);
2387
2388         ref_root = btrfs_header_owner(buf);
2389         nritems = btrfs_header_nritems(buf);
2390         level = btrfs_header_level(buf);
2391
2392         if (!root->ref_cows && level == 0)
2393                 return 0;
2394
2395         if (inc)
2396                 process_func = btrfs_inc_extent_ref;
2397         else
2398                 process_func = btrfs_free_extent;
2399
2400         if (full_backref)
2401                 parent = buf->start;
2402         else
2403                 parent = 0;
2404
2405         for (i = 0; i < nritems; i++) {
2406                 if (level == 0) {
2407                         btrfs_item_key_to_cpu(buf, &key, i);
2408                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2409                                 continue;
2410                         fi = btrfs_item_ptr(buf, i,
2411                                             struct btrfs_file_extent_item);
2412                         if (btrfs_file_extent_type(buf, fi) ==
2413                             BTRFS_FILE_EXTENT_INLINE)
2414                                 continue;
2415                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2416                         if (bytenr == 0)
2417                                 continue;
2418
2419                         num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2420                         key.offset -= btrfs_file_extent_offset(buf, fi);
2421                         ret = process_func(trans, root, bytenr, num_bytes,
2422                                            parent, ref_root, key.objectid,
2423                                            key.offset);
2424                         if (ret)
2425                                 goto fail;
2426                 } else {
2427                         bytenr = btrfs_node_blockptr(buf, i);
2428                         num_bytes = btrfs_level_size(root, level - 1);
2429                         ret = process_func(trans, root, bytenr, num_bytes,
2430                                            parent, ref_root, level - 1, 0);
2431                         if (ret)
2432                                 goto fail;
2433                 }
2434         }
2435         return 0;
2436 fail:
2437         BUG();
2438         return ret;
2439 }
2440
2441 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2442                   struct extent_buffer *buf, int full_backref)
2443 {
2444         return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
2445 }
2446
2447 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2448                   struct extent_buffer *buf, int full_backref)
2449 {
2450         return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
2451 }
2452
2453 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2454                                  struct btrfs_root *root,
2455                                  struct btrfs_path *path,
2456                                  struct btrfs_block_group_cache *cache)
2457 {
2458         int ret;
2459         struct btrfs_root *extent_root = root->fs_info->extent_root;
2460         unsigned long bi;
2461         struct extent_buffer *leaf;
2462
2463         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2464         if (ret < 0)
2465                 goto fail;
2466         BUG_ON(ret);
2467
2468         leaf = path->nodes[0];
2469         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2470         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2471         btrfs_mark_buffer_dirty(leaf);
2472         btrfs_release_path(extent_root, path);
2473 fail:
2474         if (ret)
2475                 return ret;
2476         return 0;
2477
2478 }
2479
2480 static struct btrfs_block_group_cache *
2481 next_block_group(struct btrfs_root *root,
2482                  struct btrfs_block_group_cache *cache)
2483 {
2484         struct rb_node *node;
2485         spin_lock(&root->fs_info->block_group_cache_lock);
2486         node = rb_next(&cache->cache_node);
2487         btrfs_put_block_group(cache);
2488         if (node) {
2489                 cache = rb_entry(node, struct btrfs_block_group_cache,
2490                                  cache_node);
2491                 atomic_inc(&cache->count);
2492         } else
2493                 cache = NULL;
2494         spin_unlock(&root->fs_info->block_group_cache_lock);
2495         return cache;
2496 }
2497
2498 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
2499                                    struct btrfs_root *root)
2500 {
2501         struct btrfs_block_group_cache *cache;
2502         int err = 0;
2503         struct btrfs_path *path;
2504         u64 last = 0;
2505
2506         path = btrfs_alloc_path();
2507         if (!path)
2508                 return -ENOMEM;
2509
2510         while (1) {
2511                 if (last == 0) {
2512                         err = btrfs_run_delayed_refs(trans, root,
2513                                                      (unsigned long)-1);
2514                         BUG_ON(err);
2515                 }
2516
2517                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2518                 while (cache) {
2519                         if (cache->dirty)
2520                                 break;
2521                         cache = next_block_group(root, cache);
2522                 }
2523                 if (!cache) {
2524                         if (last == 0)
2525                                 break;
2526                         last = 0;
2527                         continue;
2528                 }
2529
2530                 cache->dirty = 0;
2531                 last = cache->key.objectid + cache->key.offset;
2532
2533                 err = write_one_cache_group(trans, root, path, cache);
2534                 BUG_ON(err);
2535                 btrfs_put_block_group(cache);
2536         }
2537
2538         btrfs_free_path(path);
2539         return 0;
2540 }
2541
2542 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
2543 {
2544         struct btrfs_block_group_cache *block_group;
2545         int readonly = 0;
2546
2547         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
2548         if (!block_group || block_group->ro)
2549                 readonly = 1;
2550         if (block_group)
2551                 btrfs_put_block_group(block_group);
2552         return readonly;
2553 }
2554
2555 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
2556                              u64 total_bytes, u64 bytes_used,
2557                              struct btrfs_space_info **space_info)
2558 {
2559         struct btrfs_space_info *found;
2560
2561         found = __find_space_info(info, flags);
2562         if (found) {
2563                 spin_lock(&found->lock);
2564                 found->total_bytes += total_bytes;
2565                 found->bytes_used += bytes_used;
2566                 found->full = 0;
2567                 spin_unlock(&found->lock);
2568                 *space_info = found;
2569                 return 0;
2570         }
2571         found = kzalloc(sizeof(*found), GFP_NOFS);
2572         if (!found)
2573                 return -ENOMEM;
2574
2575         INIT_LIST_HEAD(&found->block_groups);
2576         init_rwsem(&found->groups_sem);
2577         spin_lock_init(&found->lock);
2578         found->flags = flags;
2579         found->total_bytes = total_bytes;
2580         found->bytes_used = bytes_used;
2581         found->bytes_pinned = 0;
2582         found->bytes_reserved = 0;
2583         found->bytes_readonly = 0;
2584         found->bytes_delalloc = 0;
2585         found->full = 0;
2586         found->force_alloc = 0;
2587         *space_info = found;
2588         list_add_rcu(&found->list, &info->space_info);
2589         atomic_set(&found->caching_threads, 0);
2590         return 0;
2591 }
2592
2593 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
2594 {
2595         u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
2596                                    BTRFS_BLOCK_GROUP_RAID1 |
2597                                    BTRFS_BLOCK_GROUP_RAID10 |
2598                                    BTRFS_BLOCK_GROUP_DUP);
2599         if (extra_flags) {
2600                 if (flags & BTRFS_BLOCK_GROUP_DATA)
2601                         fs_info->avail_data_alloc_bits |= extra_flags;
2602                 if (flags & BTRFS_BLOCK_GROUP_METADATA)
2603                         fs_info->avail_metadata_alloc_bits |= extra_flags;
2604                 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
2605                         fs_info->avail_system_alloc_bits |= extra_flags;
2606         }
2607 }
2608
2609 static void set_block_group_readonly(struct btrfs_block_group_cache *cache)
2610 {
2611         spin_lock(&cache->space_info->lock);
2612         spin_lock(&cache->lock);
2613         if (!cache->ro) {
2614                 cache->space_info->bytes_readonly += cache->key.offset -
2615                                         btrfs_block_group_used(&cache->item);
2616                 cache->ro = 1;
2617         }
2618         spin_unlock(&cache->lock);
2619         spin_unlock(&cache->space_info->lock);
2620 }
2621
2622 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
2623 {
2624         u64 num_devices = root->fs_info->fs_devices->rw_devices;
2625
2626         if (num_devices == 1)
2627                 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
2628         if (num_devices < 4)
2629                 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
2630
2631         if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
2632             (flags & (BTRFS_BLOCK_GROUP_RAID1 |
2633                       BTRFS_BLOCK_GROUP_RAID10))) {
2634                 flags &= ~BTRFS_BLOCK_GROUP_DUP;
2635         }
2636
2637         if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
2638             (flags & BTRFS_BLOCK_GROUP_RAID10)) {
2639                 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
2640         }
2641
2642         if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
2643             ((flags & BTRFS_BLOCK_GROUP_RAID1) |
2644              (flags & BTRFS_BLOCK_GROUP_RAID10) |
2645              (flags & BTRFS_BLOCK_GROUP_DUP)))
2646                 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
2647         return flags;
2648 }
2649
2650 static u64 btrfs_get_alloc_profile(struct btrfs_root *root, u64 data)
2651 {
2652         struct btrfs_fs_info *info = root->fs_info;
2653         u64 alloc_profile;
2654
2655         if (data) {
2656                 alloc_profile = info->avail_data_alloc_bits &
2657                         info->data_alloc_profile;
2658                 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
2659         } else if (root == root->fs_info->chunk_root) {
2660                 alloc_profile = info->avail_system_alloc_bits &
2661                         info->system_alloc_profile;
2662                 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
2663         } else {
2664                 alloc_profile = info->avail_metadata_alloc_bits &
2665                         info->metadata_alloc_profile;
2666                 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
2667         }
2668
2669         return btrfs_reduce_alloc_profile(root, data);
2670 }
2671
2672 void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
2673 {
2674         u64 alloc_target;
2675
2676         alloc_target = btrfs_get_alloc_profile(root, 1);
2677         BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
2678                                                        alloc_target);
2679 }
2680
2681 /*
2682  * for now this just makes sure we have at least 5% of our metadata space free
2683  * for use.
2684  */
2685 int btrfs_check_metadata_free_space(struct btrfs_root *root)
2686 {
2687         struct btrfs_fs_info *info = root->fs_info;
2688         struct btrfs_space_info *meta_sinfo;
2689         u64 alloc_target, thresh;
2690         int committed = 0, ret;
2691
2692         /* get the space info for where the metadata will live */
2693         alloc_target = btrfs_get_alloc_profile(root, 0);
2694         meta_sinfo = __find_space_info(info, alloc_target);
2695
2696 again:
2697         spin_lock(&meta_sinfo->lock);
2698         if (!meta_sinfo->full)
2699                 thresh = meta_sinfo->total_bytes * 80;
2700         else
2701                 thresh = meta_sinfo->total_bytes * 95;
2702
2703         do_div(thresh, 100);
2704
2705         if (meta_sinfo->bytes_used + meta_sinfo->bytes_reserved +
2706             meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly > thresh) {
2707                 struct btrfs_trans_handle *trans;
2708                 if (!meta_sinfo->full) {
2709                         meta_sinfo->force_alloc = 1;
2710                         spin_unlock(&meta_sinfo->lock);
2711
2712                         trans = btrfs_start_transaction(root, 1);
2713                         if (!trans)
2714                                 return -ENOMEM;
2715
2716                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2717                                              2 * 1024 * 1024, alloc_target, 0);
2718                         btrfs_end_transaction(trans, root);
2719                         goto again;
2720                 }
2721                 spin_unlock(&meta_sinfo->lock);
2722
2723                 if (!committed) {
2724                         committed = 1;
2725                         trans = btrfs_join_transaction(root, 1);
2726                         if (!trans)
2727                                 return -ENOMEM;
2728                         ret = btrfs_commit_transaction(trans, root);
2729                         if (ret)
2730                                 return ret;
2731                         goto again;
2732                 }
2733                 return -ENOSPC;
2734         }
2735         spin_unlock(&meta_sinfo->lock);
2736
2737         return 0;
2738 }
2739
2740 /*
2741  * This will check the space that the inode allocates from to make sure we have
2742  * enough space for bytes.
2743  */
2744 int btrfs_check_data_free_space(struct btrfs_root *root, struct inode *inode,
2745                                 u64 bytes)
2746 {
2747         struct btrfs_space_info *data_sinfo;
2748         int ret = 0, committed = 0;
2749
2750         /* make sure bytes are sectorsize aligned */
2751         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
2752
2753         data_sinfo = BTRFS_I(inode)->space_info;
2754 again:
2755         /* make sure we have enough space to handle the data first */
2756         spin_lock(&data_sinfo->lock);
2757         if (data_sinfo->total_bytes - data_sinfo->bytes_used -
2758             data_sinfo->bytes_delalloc - data_sinfo->bytes_reserved -
2759             data_sinfo->bytes_pinned - data_sinfo->bytes_readonly -
2760             data_sinfo->bytes_may_use < bytes) {
2761                 struct btrfs_trans_handle *trans;
2762
2763                 /*
2764                  * if we don't have enough free bytes in this space then we need
2765                  * to alloc a new chunk.
2766                  */
2767                 if (!data_sinfo->full) {
2768                         u64 alloc_target;
2769
2770                         data_sinfo->force_alloc = 1;
2771                         spin_unlock(&data_sinfo->lock);
2772
2773                         alloc_target = btrfs_get_alloc_profile(root, 1);
2774                         trans = btrfs_start_transaction(root, 1);
2775                         if (!trans)
2776                                 return -ENOMEM;
2777
2778                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2779                                              bytes + 2 * 1024 * 1024,
2780                                              alloc_target, 0);
2781                         btrfs_end_transaction(trans, root);
2782                         if (ret)
2783                                 return ret;
2784                         goto again;
2785                 }
2786                 spin_unlock(&data_sinfo->lock);
2787
2788                 /* commit the current transaction and try again */
2789                 if (!committed) {
2790                         committed = 1;
2791                         trans = btrfs_join_transaction(root, 1);
2792                         if (!trans)
2793                                 return -ENOMEM;
2794                         ret = btrfs_commit_transaction(trans, root);
2795                         if (ret)
2796                                 return ret;
2797                         goto again;
2798                 }
2799
2800                 printk(KERN_ERR "no space left, need %llu, %llu delalloc bytes"
2801                        ", %llu bytes_used, %llu bytes_reserved, "
2802                        "%llu bytes_pinned, %llu bytes_readonly, %llu may use "
2803                        "%llu total\n", (unsigned long long)bytes,
2804                        (unsigned long long)data_sinfo->bytes_delalloc,
2805                        (unsigned long long)data_sinfo->bytes_used,
2806                        (unsigned long long)data_sinfo->bytes_reserved,
2807                        (unsigned long long)data_sinfo->bytes_pinned,
2808                        (unsigned long long)data_sinfo->bytes_readonly,
2809                        (unsigned long long)data_sinfo->bytes_may_use,
2810                        (unsigned long long)data_sinfo->total_bytes);
2811                 return -ENOSPC;
2812         }
2813         data_sinfo->bytes_may_use += bytes;
2814         BTRFS_I(inode)->reserved_bytes += bytes;
2815         spin_unlock(&data_sinfo->lock);
2816
2817         return btrfs_check_metadata_free_space(root);
2818 }
2819
2820 /*
2821  * if there was an error for whatever reason after calling
2822  * btrfs_check_data_free_space, call this so we can cleanup the counters.
2823  */
2824 void btrfs_free_reserved_data_space(struct btrfs_root *root,
2825                                     struct inode *inode, u64 bytes)
2826 {
2827         struct btrfs_space_info *data_sinfo;
2828
2829         /* make sure bytes are sectorsize aligned */
2830         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
2831
2832         data_sinfo = BTRFS_I(inode)->space_info;
2833         spin_lock(&data_sinfo->lock);
2834         data_sinfo->bytes_may_use -= bytes;
2835         BTRFS_I(inode)->reserved_bytes -= bytes;
2836         spin_unlock(&data_sinfo->lock);
2837 }
2838
2839 /* called when we are adding a delalloc extent to the inode's io_tree */
2840 void btrfs_delalloc_reserve_space(struct btrfs_root *root, struct inode *inode,
2841                                   u64 bytes)
2842 {
2843         struct btrfs_space_info *data_sinfo;
2844
2845         /* get the space info for where this inode will be storing its data */
2846         data_sinfo = BTRFS_I(inode)->space_info;
2847
2848         /* make sure we have enough space to handle the data first */
2849         spin_lock(&data_sinfo->lock);
2850         data_sinfo->bytes_delalloc += bytes;
2851
2852         /*
2853          * we are adding a delalloc extent without calling
2854          * btrfs_check_data_free_space first.  This happens on a weird
2855          * writepage condition, but shouldn't hurt our accounting
2856          */
2857         if (unlikely(bytes > BTRFS_I(inode)->reserved_bytes)) {
2858                 data_sinfo->bytes_may_use -= BTRFS_I(inode)->reserved_bytes;
2859                 BTRFS_I(inode)->reserved_bytes = 0;
2860         } else {
2861                 data_sinfo->bytes_may_use -= bytes;
2862                 BTRFS_I(inode)->reserved_bytes -= bytes;
2863         }
2864
2865         spin_unlock(&data_sinfo->lock);
2866 }
2867
2868 /* called when we are clearing an delalloc extent from the inode's io_tree */
2869 void btrfs_delalloc_free_space(struct btrfs_root *root, struct inode *inode,
2870                               u64 bytes)
2871 {
2872         struct btrfs_space_info *info;
2873
2874         info = BTRFS_I(inode)->space_info;
2875
2876         spin_lock(&info->lock);
2877         info->bytes_delalloc -= bytes;
2878         spin_unlock(&info->lock);
2879 }
2880
2881 static void force_metadata_allocation(struct btrfs_fs_info *info)
2882 {
2883         struct list_head *head = &info->space_info;
2884         struct btrfs_space_info *found;
2885
2886         rcu_read_lock();
2887         list_for_each_entry_rcu(found, head, list) {
2888                 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
2889                         found->force_alloc = 1;
2890         }
2891         rcu_read_unlock();
2892 }
2893
2894 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
2895                           struct btrfs_root *extent_root, u64 alloc_bytes,
2896                           u64 flags, int force)
2897 {
2898         struct btrfs_space_info *space_info;
2899         struct btrfs_fs_info *fs_info = extent_root->fs_info;
2900         u64 thresh;
2901         int ret = 0;
2902
2903         mutex_lock(&fs_info->chunk_mutex);
2904
2905         flags = btrfs_reduce_alloc_profile(extent_root, flags);
2906
2907         space_info = __find_space_info(extent_root->fs_info, flags);
2908         if (!space_info) {
2909                 ret = update_space_info(extent_root->fs_info, flags,
2910                                         0, 0, &space_info);
2911                 BUG_ON(ret);
2912         }
2913         BUG_ON(!space_info);
2914
2915         spin_lock(&space_info->lock);
2916         if (space_info->force_alloc) {
2917                 force = 1;
2918                 space_info->force_alloc = 0;
2919         }
2920         if (space_info->full) {
2921                 spin_unlock(&space_info->lock);
2922                 goto out;
2923         }
2924
2925         thresh = space_info->total_bytes - space_info->bytes_readonly;
2926         thresh = div_factor(thresh, 6);
2927         if (!force &&
2928            (space_info->bytes_used + space_info->bytes_pinned +
2929             space_info->bytes_reserved + alloc_bytes) < thresh) {
2930                 spin_unlock(&space_info->lock);
2931                 goto out;
2932         }
2933         spin_unlock(&space_info->lock);
2934
2935         /*
2936          * if we're doing a data chunk, go ahead and make sure that
2937          * we keep a reasonable number of metadata chunks allocated in the
2938          * FS as well.
2939          */
2940         if (flags & BTRFS_BLOCK_GROUP_DATA) {
2941                 fs_info->data_chunk_allocations++;
2942                 if (!(fs_info->data_chunk_allocations %
2943                       fs_info->metadata_ratio))
2944                         force_metadata_allocation(fs_info);
2945         }
2946
2947         ret = btrfs_alloc_chunk(trans, extent_root, flags);
2948         if (ret)
2949                 space_info->full = 1;
2950 out:
2951         mutex_unlock(&extent_root->fs_info->chunk_mutex);
2952         return ret;
2953 }
2954
2955 static int update_block_group(struct btrfs_trans_handle *trans,
2956                               struct btrfs_root *root,
2957                               u64 bytenr, u64 num_bytes, int alloc,
2958                               int mark_free)
2959 {
2960         struct btrfs_block_group_cache *cache;
2961         struct btrfs_fs_info *info = root->fs_info;
2962         u64 total = num_bytes;
2963         u64 old_val;
2964         u64 byte_in_group;
2965
2966         /* block accounting for super block */
2967         spin_lock(&info->delalloc_lock);
2968         old_val = btrfs_super_bytes_used(&info->super_copy);
2969         if (alloc)
2970                 old_val += num_bytes;
2971         else
2972                 old_val -= num_bytes;
2973         btrfs_set_super_bytes_used(&info->super_copy, old_val);
2974
2975         /* block accounting for root item */
2976         old_val = btrfs_root_used(&root->root_item);
2977         if (alloc)
2978                 old_val += num_bytes;
2979         else
2980                 old_val -= num_bytes;
2981         btrfs_set_root_used(&root->root_item, old_val);
2982         spin_unlock(&info->delalloc_lock);
2983
2984         while (total) {
2985                 cache = btrfs_lookup_block_group(info, bytenr);
2986                 if (!cache)
2987                         return -1;
2988                 byte_in_group = bytenr - cache->key.objectid;
2989                 WARN_ON(byte_in_group > cache->key.offset);
2990
2991                 spin_lock(&cache->space_info->lock);
2992                 spin_lock(&cache->lock);
2993                 cache->dirty = 1;
2994                 old_val = btrfs_block_group_used(&cache->item);
2995                 num_bytes = min(total, cache->key.offset - byte_in_group);
2996                 if (alloc) {
2997                         old_val += num_bytes;
2998                         cache->space_info->bytes_used += num_bytes;
2999                         if (cache->ro)
3000                                 cache->space_info->bytes_readonly -= num_bytes;
3001                         btrfs_set_block_group_used(&cache->item, old_val);
3002                         spin_unlock(&cache->lock);
3003                         spin_unlock(&cache->space_info->lock);
3004                 } else {
3005                         old_val -= num_bytes;
3006                         cache->space_info->bytes_used -= num_bytes;
3007                         if (cache->ro)
3008                                 cache->space_info->bytes_readonly += num_bytes;
3009                         btrfs_set_block_group_used(&cache->item, old_val);
3010                         spin_unlock(&cache->lock);
3011                         spin_unlock(&cache->space_info->lock);
3012                         if (mark_free) {
3013                                 int ret;
3014
3015                                 ret = btrfs_discard_extent(root, bytenr,
3016                                                            num_bytes);
3017                                 WARN_ON(ret);
3018
3019                                 ret = btrfs_add_free_space(cache, bytenr,
3020                                                            num_bytes);
3021                                 WARN_ON(ret);
3022                         }
3023                 }
3024                 btrfs_put_block_group(cache);
3025                 total -= num_bytes;
3026                 bytenr += num_bytes;
3027         }
3028         return 0;
3029 }
3030
3031 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
3032 {
3033         struct btrfs_block_group_cache *cache;
3034         u64 bytenr;
3035
3036         cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
3037         if (!cache)
3038                 return 0;
3039
3040         bytenr = cache->key.objectid;
3041         btrfs_put_block_group(cache);
3042
3043         return bytenr;
3044 }
3045
3046 int btrfs_update_pinned_extents(struct btrfs_root *root,
3047                                 u64 bytenr, u64 num, int pin)
3048 {
3049         u64 len;
3050         struct btrfs_block_group_cache *cache;
3051         struct btrfs_fs_info *fs_info = root->fs_info;
3052
3053         if (pin)
3054                 set_extent_dirty(&fs_info->pinned_extents,
3055                                 bytenr, bytenr + num - 1, GFP_NOFS);
3056
3057         while (num > 0) {
3058                 cache = btrfs_lookup_block_group(fs_info, bytenr);
3059                 BUG_ON(!cache);
3060                 len = min(num, cache->key.offset -
3061                           (bytenr - cache->key.objectid));
3062                 if (pin) {
3063                         spin_lock(&cache->space_info->lock);
3064                         spin_lock(&cache->lock);
3065                         cache->pinned += len;
3066                         cache->space_info->bytes_pinned += len;
3067                         spin_unlock(&cache->lock);
3068                         spin_unlock(&cache->space_info->lock);
3069                         fs_info->total_pinned += len;
3070                 } else {
3071                         int unpin = 0;
3072
3073                         /*
3074                          * in order to not race with the block group caching, we
3075                          * only want to unpin the extent if we are cached.  If
3076                          * we aren't cached, we want to start async caching this
3077                          * block group so we can free the extent the next time
3078                          * around.
3079                          */
3080                         spin_lock(&cache->space_info->lock);
3081                         spin_lock(&cache->lock);
3082                         unpin = (cache->cached == BTRFS_CACHE_FINISHED);
3083                         if (likely(unpin)) {
3084                                 cache->pinned -= len;
3085                                 cache->space_info->bytes_pinned -= len;
3086                                 fs_info->total_pinned -= len;
3087                         }
3088                         spin_unlock(&cache->lock);
3089                         spin_unlock(&cache->space_info->lock);
3090
3091                         if (likely(unpin))
3092                                 clear_extent_dirty(&fs_info->pinned_extents,
3093                                                    bytenr, bytenr + len -1,
3094                                                    GFP_NOFS);
3095                         else
3096                                 cache_block_group(cache);
3097
3098                         if (unpin)
3099                                 btrfs_add_free_space(cache, bytenr, len);
3100                 }
3101                 btrfs_put_block_group(cache);
3102                 bytenr += len;
3103                 num -= len;
3104         }
3105         return 0;
3106 }
3107
3108 static int update_reserved_extents(struct btrfs_root *root,
3109                                    u64 bytenr, u64 num, int reserve)
3110 {
3111         u64 len;
3112         struct btrfs_block_group_cache *cache;
3113         struct btrfs_fs_info *fs_info = root->fs_info;
3114
3115         while (num > 0) {
3116                 cache = btrfs_lookup_block_group(fs_info, bytenr);
3117                 BUG_ON(!cache);
3118                 len = min(num, cache->key.offset -
3119                           (bytenr - cache->key.objectid));
3120
3121                 spin_lock(&cache->space_info->lock);
3122                 spin_lock(&cache->lock);
3123                 if (reserve) {
3124                         cache->reserved += len;
3125                         cache->space_info->bytes_reserved += len;
3126                 } else {
3127                         cache->reserved -= len;
3128                         cache->space_info->bytes_reserved -= len;
3129                 }
3130                 spin_unlock(&cache->lock);
3131                 spin_unlock(&cache->space_info->lock);
3132                 btrfs_put_block_group(cache);
3133                 bytenr += len;
3134                 num -= len;
3135         }
3136         return 0;
3137 }
3138
3139 int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
3140 {
3141         u64 last = 0;
3142         u64 start;
3143         u64 end;
3144         struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
3145         int ret;
3146
3147         while (1) {
3148                 ret = find_first_extent_bit(pinned_extents, last,
3149                                             &start, &end, EXTENT_DIRTY);
3150                 if (ret)
3151                         break;
3152
3153                 set_extent_dirty(copy, start, end, GFP_NOFS);
3154                 last = end + 1;
3155         }
3156         return 0;
3157 }
3158
3159 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
3160                                struct btrfs_root *root,
3161                                struct extent_io_tree *unpin)
3162 {
3163         u64 start;
3164         u64 end;
3165         int ret;
3166
3167         while (1) {
3168                 ret = find_first_extent_bit(unpin, 0, &start, &end,
3169                                             EXTENT_DIRTY);
3170                 if (ret)
3171                         break;
3172
3173                 ret = btrfs_discard_extent(root, start, end + 1 - start);
3174
3175                 /* unlocks t