~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/fs/btrfs/extent-tree.c

Version: ~ [ linux-5.18-rc6 ] ~ [ linux-5.17.6 ] ~ [ linux-5.16.20 ] ~ [ linux-5.15.38 ] ~ [ linux-5.14.21 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.114 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.192 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.241 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.277 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.312 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.302 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.9 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 /*
  2  * Copyright (C) 2007 Oracle.  All rights reserved.
  3  *
  4  * This program is free software; you can redistribute it and/or
  5  * modify it under the terms of the GNU General Public
  6  * License v2 as published by the Free Software Foundation.
  7  *
  8  * This program is distributed in the hope that it will be useful,
  9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
 10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 11  * General Public License for more details.
 12  *
 13  * You should have received a copy of the GNU General Public
 14  * License along with this program; if not, write to the
 15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
 16  * Boston, MA 021110-1307, USA.
 17  */
 18 #include <linux/sched.h>
 19 #include <linux/pagemap.h>
 20 #include <linux/writeback.h>
 21 #include <linux/blkdev.h>
 22 #include <linux/sort.h>
 23 #include <linux/rcupdate.h>
 24 #include <linux/kthread.h>
 25 #include "compat.h"
 26 #include "hash.h"
 27 #include "ctree.h"
 28 #include "disk-io.h"
 29 #include "print-tree.h"
 30 #include "transaction.h"
 31 #include "volumes.h"
 32 #include "locking.h"
 33 #include "free-space-cache.h"
 34 
 35 static int update_block_group(struct btrfs_trans_handle *trans,
 36                               struct btrfs_root *root,
 37                               u64 bytenr, u64 num_bytes, int alloc,
 38                               int mark_free);
 39 static int update_reserved_extents(struct btrfs_block_group_cache *cache,
 40                                    u64 num_bytes, int reserve);
 41 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
 42                                 struct btrfs_root *root,
 43                                 u64 bytenr, u64 num_bytes, u64 parent,
 44                                 u64 root_objectid, u64 owner_objectid,
 45                                 u64 owner_offset, int refs_to_drop,
 46                                 struct btrfs_delayed_extent_op *extra_op);
 47 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
 48                                     struct extent_buffer *leaf,
 49                                     struct btrfs_extent_item *ei);
 50 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
 51                                       struct btrfs_root *root,
 52                                       u64 parent, u64 root_objectid,
 53                                       u64 flags, u64 owner, u64 offset,
 54                                       struct btrfs_key *ins, int ref_mod);
 55 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
 56                                      struct btrfs_root *root,
 57                                      u64 parent, u64 root_objectid,
 58                                      u64 flags, struct btrfs_disk_key *key,
 59                                      int level, struct btrfs_key *ins);
 60 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
 61                           struct btrfs_root *extent_root, u64 alloc_bytes,
 62                           u64 flags, int force);
 63 static int pin_down_bytes(struct btrfs_trans_handle *trans,
 64                           struct btrfs_root *root,
 65                           struct btrfs_path *path,
 66                           u64 bytenr, u64 num_bytes,
 67                           int is_data, int reserved,
 68                           struct extent_buffer **must_clean);
 69 static int find_next_key(struct btrfs_path *path, int level,
 70                          struct btrfs_key *key);
 71 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
 72                             int dump_block_groups);
 73 
 74 static noinline int
 75 block_group_cache_done(struct btrfs_block_group_cache *cache)
 76 {
 77         smp_mb();
 78         return cache->cached == BTRFS_CACHE_FINISHED;
 79 }
 80 
 81 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
 82 {
 83         return (cache->flags & bits) == bits;
 84 }
 85 
 86 void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
 87 {
 88         atomic_inc(&cache->count);
 89 }
 90 
 91 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
 92 {
 93         if (atomic_dec_and_test(&cache->count))
 94                 kfree(cache);
 95 }
 96 
 97 /*
 98  * this adds the block group to the fs_info rb tree for the block group
 99  * cache
100  */
101 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
102                                 struct btrfs_block_group_cache *block_group)
103 {
104         struct rb_node **p;
105         struct rb_node *parent = NULL;
106         struct btrfs_block_group_cache *cache;
107 
108         spin_lock(&info->block_group_cache_lock);
109         p = &info->block_group_cache_tree.rb_node;
110 
111         while (*p) {
112                 parent = *p;
113                 cache = rb_entry(parent, struct btrfs_block_group_cache,
114                                  cache_node);
115                 if (block_group->key.objectid < cache->key.objectid) {
116                         p = &(*p)->rb_left;
117                 } else if (block_group->key.objectid > cache->key.objectid) {
118                         p = &(*p)->rb_right;
119                 } else {
120                         spin_unlock(&info->block_group_cache_lock);
121                         return -EEXIST;
122                 }
123         }
124 
125         rb_link_node(&block_group->cache_node, parent, p);
126         rb_insert_color(&block_group->cache_node,
127                         &info->block_group_cache_tree);
128         spin_unlock(&info->block_group_cache_lock);
129 
130         return 0;
131 }
132 
133 /*
134  * This will return the block group at or after bytenr if contains is 0, else
135  * it will return the block group that contains the bytenr
136  */
137 static struct btrfs_block_group_cache *
138 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
139                               int contains)
140 {
141         struct btrfs_block_group_cache *cache, *ret = NULL;
142         struct rb_node *n;
143         u64 end, start;
144 
145         spin_lock(&info->block_group_cache_lock);
146         n = info->block_group_cache_tree.rb_node;
147 
148         while (n) {
149                 cache = rb_entry(n, struct btrfs_block_group_cache,
150                                  cache_node);
151                 end = cache->key.objectid + cache->key.offset - 1;
152                 start = cache->key.objectid;
153 
154                 if (bytenr < start) {
155                         if (!contains && (!ret || start < ret->key.objectid))
156                                 ret = cache;
157                         n = n->rb_left;
158                 } else if (bytenr > start) {
159                         if (contains && bytenr <= end) {
160                                 ret = cache;
161                                 break;
162                         }
163                         n = n->rb_right;
164                 } else {
165                         ret = cache;
166                         break;
167                 }
168         }
169         if (ret)
170                 btrfs_get_block_group(ret);
171         spin_unlock(&info->block_group_cache_lock);
172 
173         return ret;
174 }
175 
176 static int add_excluded_extent(struct btrfs_root *root,
177                                u64 start, u64 num_bytes)
178 {
179         u64 end = start + num_bytes - 1;
180         set_extent_bits(&root->fs_info->freed_extents[0],
181                         start, end, EXTENT_UPTODATE, GFP_NOFS);
182         set_extent_bits(&root->fs_info->freed_extents[1],
183                         start, end, EXTENT_UPTODATE, GFP_NOFS);
184         return 0;
185 }
186 
187 static void free_excluded_extents(struct btrfs_root *root,
188                                   struct btrfs_block_group_cache *cache)
189 {
190         u64 start, end;
191 
192         start = cache->key.objectid;
193         end = start + cache->key.offset - 1;
194 
195         clear_extent_bits(&root->fs_info->freed_extents[0],
196                           start, end, EXTENT_UPTODATE, GFP_NOFS);
197         clear_extent_bits(&root->fs_info->freed_extents[1],
198                           start, end, EXTENT_UPTODATE, GFP_NOFS);
199 }
200 
201 static int exclude_super_stripes(struct btrfs_root *root,
202                                  struct btrfs_block_group_cache *cache)
203 {
204         u64 bytenr;
205         u64 *logical;
206         int stripe_len;
207         int i, nr, ret;
208 
209         if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
210                 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
211                 cache->bytes_super += stripe_len;
212                 ret = add_excluded_extent(root, cache->key.objectid,
213                                           stripe_len);
214                 BUG_ON(ret);
215         }
216 
217         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
218                 bytenr = btrfs_sb_offset(i);
219                 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
220                                        cache->key.objectid, bytenr,
221                                        0, &logical, &nr, &stripe_len);
222                 BUG_ON(ret);
223 
224                 while (nr--) {
225                         cache->bytes_super += stripe_len;
226                         ret = add_excluded_extent(root, logical[nr],
227                                                   stripe_len);
228                         BUG_ON(ret);
229                 }
230 
231                 kfree(logical);
232         }
233         return 0;
234 }
235 
236 static struct btrfs_caching_control *
237 get_caching_control(struct btrfs_block_group_cache *cache)
238 {
239         struct btrfs_caching_control *ctl;
240 
241         spin_lock(&cache->lock);
242         if (cache->cached != BTRFS_CACHE_STARTED) {
243                 spin_unlock(&cache->lock);
244                 return NULL;
245         }
246 
247         ctl = cache->caching_ctl;
248         atomic_inc(&ctl->count);
249         spin_unlock(&cache->lock);
250         return ctl;
251 }
252 
253 static void put_caching_control(struct btrfs_caching_control *ctl)
254 {
255         if (atomic_dec_and_test(&ctl->count))
256                 kfree(ctl);
257 }
258 
259 /*
260  * this is only called by cache_block_group, since we could have freed extents
261  * we need to check the pinned_extents for any extents that can't be used yet
262  * since their free space will be released as soon as the transaction commits.
263  */
264 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
265                               struct btrfs_fs_info *info, u64 start, u64 end)
266 {
267         u64 extent_start, extent_end, size, total_added = 0;
268         int ret;
269 
270         while (start < end) {
271                 ret = find_first_extent_bit(info->pinned_extents, start,
272                                             &extent_start, &extent_end,
273                                             EXTENT_DIRTY | EXTENT_UPTODATE);
274                 if (ret)
275                         break;
276 
277                 if (extent_start <= start) {
278                         start = extent_end + 1;
279                 } else if (extent_start > start && extent_start < end) {
280                         size = extent_start - start;
281                         total_added += size;
282                         ret = btrfs_add_free_space(block_group, start,
283                                                    size);
284                         BUG_ON(ret);
285                         start = extent_end + 1;
286                 } else {
287                         break;
288                 }
289         }
290 
291         if (start < end) {
292                 size = end - start;
293                 total_added += size;
294                 ret = btrfs_add_free_space(block_group, start, size);
295                 BUG_ON(ret);
296         }
297 
298         return total_added;
299 }
300 
301 static int caching_kthread(void *data)
302 {
303         struct btrfs_block_group_cache *block_group = data;
304         struct btrfs_fs_info *fs_info = block_group->fs_info;
305         struct btrfs_caching_control *caching_ctl = block_group->caching_ctl;
306         struct btrfs_root *extent_root = fs_info->extent_root;
307         struct btrfs_path *path;
308         struct extent_buffer *leaf;
309         struct btrfs_key key;
310         u64 total_found = 0;
311         u64 last = 0;
312         u32 nritems;
313         int ret = 0;
314 
315         path = btrfs_alloc_path();
316         if (!path)
317                 return -ENOMEM;
318 
319         exclude_super_stripes(extent_root, block_group);
320         spin_lock(&block_group->space_info->lock);
321         block_group->space_info->bytes_super += block_group->bytes_super;
322         spin_unlock(&block_group->space_info->lock);
323 
324         last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
325 
326         /*
327          * We don't want to deadlock with somebody trying to allocate a new
328          * extent for the extent root while also trying to search the extent
329          * root to add free space.  So we skip locking and search the commit
330          * root, since its read-only
331          */
332         path->skip_locking = 1;
333         path->search_commit_root = 1;
334         path->reada = 2;
335 
336         key.objectid = last;
337         key.offset = 0;
338         key.type = BTRFS_EXTENT_ITEM_KEY;
339 again:
340         mutex_lock(&caching_ctl->mutex);
341         /* need to make sure the commit_root doesn't disappear */
342         down_read(&fs_info->extent_commit_sem);
343 
344         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
345         if (ret < 0)
346                 goto err;
347 
348         leaf = path->nodes[0];
349         nritems = btrfs_header_nritems(leaf);
350 
351         while (1) {
352                 smp_mb();
353                 if (fs_info->closing > 1) {
354                         last = (u64)-1;
355                         break;
356                 }
357 
358                 if (path->slots[0] < nritems) {
359                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
360                 } else {
361                         ret = find_next_key(path, 0, &key);
362                         if (ret)
363                                 break;
364 
365                         caching_ctl->progress = last;
366                         btrfs_release_path(extent_root, path);
367                         up_read(&fs_info->extent_commit_sem);
368                         mutex_unlock(&caching_ctl->mutex);
369                         if (btrfs_transaction_in_commit(fs_info))
370                                 schedule_timeout(1);
371                         else
372                                 cond_resched();
373                         goto again;
374                 }
375 
376                 if (key.objectid < block_group->key.objectid) {
377                         path->slots[0]++;
378                         continue;
379                 }
380 
381                 if (key.objectid >= block_group->key.objectid +
382                     block_group->key.offset)
383                         break;
384 
385                 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
386                         total_found += add_new_free_space(block_group,
387                                                           fs_info, last,
388                                                           key.objectid);
389                         last = key.objectid + key.offset;
390 
391                         if (total_found > (1024 * 1024 * 2)) {
392                                 total_found = 0;
393                                 wake_up(&caching_ctl->wait);
394                         }
395                 }
396                 path->slots[0]++;
397         }
398         ret = 0;
399 
400         total_found += add_new_free_space(block_group, fs_info, last,
401                                           block_group->key.objectid +
402                                           block_group->key.offset);
403         caching_ctl->progress = (u64)-1;
404 
405         spin_lock(&block_group->lock);
406         block_group->caching_ctl = NULL;
407         block_group->cached = BTRFS_CACHE_FINISHED;
408         spin_unlock(&block_group->lock);
409 
410 err:
411         btrfs_free_path(path);
412         up_read(&fs_info->extent_commit_sem);
413 
414         free_excluded_extents(extent_root, block_group);
415 
416         mutex_unlock(&caching_ctl->mutex);
417         wake_up(&caching_ctl->wait);
418 
419         put_caching_control(caching_ctl);
420         atomic_dec(&block_group->space_info->caching_threads);
421         btrfs_put_block_group(block_group);
422 
423         return 0;
424 }
425 
426 static int cache_block_group(struct btrfs_block_group_cache *cache)
427 {
428         struct btrfs_fs_info *fs_info = cache->fs_info;
429         struct btrfs_caching_control *caching_ctl;
430         struct task_struct *tsk;
431         int ret = 0;
432 
433         smp_mb();
434         if (cache->cached != BTRFS_CACHE_NO)
435                 return 0;
436 
437         caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_KERNEL);
438         BUG_ON(!caching_ctl);
439 
440         INIT_LIST_HEAD(&caching_ctl->list);
441         mutex_init(&caching_ctl->mutex);
442         init_waitqueue_head(&caching_ctl->wait);
443         caching_ctl->block_group = cache;
444         caching_ctl->progress = cache->key.objectid;
445         /* one for caching kthread, one for caching block group list */
446         atomic_set(&caching_ctl->count, 2);
447 
448         spin_lock(&cache->lock);
449         if (cache->cached != BTRFS_CACHE_NO) {
450                 spin_unlock(&cache->lock);
451                 kfree(caching_ctl);
452                 return 0;
453         }
454         cache->caching_ctl = caching_ctl;
455         cache->cached = BTRFS_CACHE_STARTED;
456         spin_unlock(&cache->lock);
457 
458         down_write(&fs_info->extent_commit_sem);
459         list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
460         up_write(&fs_info->extent_commit_sem);
461 
462         atomic_inc(&cache->space_info->caching_threads);
463         btrfs_get_block_group(cache);
464 
465         tsk = kthread_run(caching_kthread, cache, "btrfs-cache-%llu\n",
466                           cache->key.objectid);
467         if (IS_ERR(tsk)) {
468                 ret = PTR_ERR(tsk);
469                 printk(KERN_ERR "error running thread %d\n", ret);
470                 BUG();
471         }
472 
473         return ret;
474 }
475 
476 /*
477  * return the block group that starts at or after bytenr
478  */
479 static struct btrfs_block_group_cache *
480 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
481 {
482         struct btrfs_block_group_cache *cache;
483 
484         cache = block_group_cache_tree_search(info, bytenr, 0);
485 
486         return cache;
487 }
488 
489 /*
490  * return the block group that contains the given bytenr
491  */
492 struct btrfs_block_group_cache *btrfs_lookup_block_group(
493                                                  struct btrfs_fs_info *info,
494                                                  u64 bytenr)
495 {
496         struct btrfs_block_group_cache *cache;
497 
498         cache = block_group_cache_tree_search(info, bytenr, 1);
499 
500         return cache;
501 }
502 
503 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
504                                                   u64 flags)
505 {
506         struct list_head *head = &info->space_info;
507         struct btrfs_space_info *found;
508 
509         rcu_read_lock();
510         list_for_each_entry_rcu(found, head, list) {
511                 if (found->flags == flags) {
512                         rcu_read_unlock();
513                         return found;
514                 }
515         }
516         rcu_read_unlock();
517         return NULL;
518 }
519 
520 /*
521  * after adding space to the filesystem, we need to clear the full flags
522  * on all the space infos.
523  */
524 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
525 {
526         struct list_head *head = &info->space_info;
527         struct btrfs_space_info *found;
528 
529         rcu_read_lock();
530         list_for_each_entry_rcu(found, head, list)
531                 found->full = 0;
532         rcu_read_unlock();
533 }
534 
535 static u64 div_factor(u64 num, int factor)
536 {
537         if (factor == 10)
538                 return num;
539         num *= factor;
540         do_div(num, 10);
541         return num;
542 }
543 
544 u64 btrfs_find_block_group(struct btrfs_root *root,
545                            u64 search_start, u64 search_hint, int owner)
546 {
547         struct btrfs_block_group_cache *cache;
548         u64 used;
549         u64 last = max(search_hint, search_start);
550         u64 group_start = 0;
551         int full_search = 0;
552         int factor = 9;
553         int wrapped = 0;
554 again:
555         while (1) {
556                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
557                 if (!cache)
558                         break;
559 
560                 spin_lock(&cache->lock);
561                 last = cache->key.objectid + cache->key.offset;
562                 used = btrfs_block_group_used(&cache->item);
563 
564                 if ((full_search || !cache->ro) &&
565                     block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
566                         if (used + cache->pinned + cache->reserved <
567                             div_factor(cache->key.offset, factor)) {
568                                 group_start = cache->key.objectid;
569                                 spin_unlock(&cache->lock);
570                                 btrfs_put_block_group(cache);
571                                 goto found;
572                         }
573                 }
574                 spin_unlock(&cache->lock);
575                 btrfs_put_block_group(cache);
576                 cond_resched();
577         }
578         if (!wrapped) {
579                 last = search_start;
580                 wrapped = 1;
581                 goto again;
582         }
583         if (!full_search && factor < 10) {
584                 last = search_start;
585                 full_search = 1;
586                 factor = 10;
587                 goto again;
588         }
589 found:
590         return group_start;
591 }
592 
593 /* simple helper to search for an existing extent at a given offset */
594 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
595 {
596         int ret;
597         struct btrfs_key key;
598         struct btrfs_path *path;
599 
600         path = btrfs_alloc_path();
601         BUG_ON(!path);
602         key.objectid = start;
603         key.offset = len;
604         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
605         ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
606                                 0, 0);
607         btrfs_free_path(path);
608         return ret;
609 }
610 
611 /*
612  * Back reference rules.  Back refs have three main goals:
613  *
614  * 1) differentiate between all holders of references to an extent so that
615  *    when a reference is dropped we can make sure it was a valid reference
616  *    before freeing the extent.
617  *
618  * 2) Provide enough information to quickly find the holders of an extent
619  *    if we notice a given block is corrupted or bad.
620  *
621  * 3) Make it easy to migrate blocks for FS shrinking or storage pool
622  *    maintenance.  This is actually the same as #2, but with a slightly
623  *    different use case.
624  *
625  * There are two kinds of back refs. The implicit back refs is optimized
626  * for pointers in non-shared tree blocks. For a given pointer in a block,
627  * back refs of this kind provide information about the block's owner tree
628  * and the pointer's key. These information allow us to find the block by
629  * b-tree searching. The full back refs is for pointers in tree blocks not
630  * referenced by their owner trees. The location of tree block is recorded
631  * in the back refs. Actually the full back refs is generic, and can be
632  * used in all cases the implicit back refs is used. The major shortcoming
633  * of the full back refs is its overhead. Every time a tree block gets
634  * COWed, we have to update back refs entry for all pointers in it.
635  *
636  * For a newly allocated tree block, we use implicit back refs for
637  * pointers in it. This means most tree related operations only involve
638  * implicit back refs. For a tree block created in old transaction, the
639  * only way to drop a reference to it is COW it. So we can detect the
640  * event that tree block loses its owner tree's reference and do the
641  * back refs conversion.
642  *
643  * When a tree block is COW'd through a tree, there are four cases:
644  *
645  * The reference count of the block is one and the tree is the block's
646  * owner tree. Nothing to do in this case.
647  *
648  * The reference count of the block is one and the tree is not the
649  * block's owner tree. In this case, full back refs is used for pointers
650  * in the block. Remove these full back refs, add implicit back refs for
651  * every pointers in the new block.
652  *
653  * The reference count of the block is greater than one and the tree is
654  * the block's owner tree. In this case, implicit back refs is used for
655  * pointers in the block. Add full back refs for every pointers in the
656  * block, increase lower level extents' reference counts. The original
657  * implicit back refs are entailed to the new block.
658  *
659  * The reference count of the block is greater than one and the tree is
660  * not the block's owner tree. Add implicit back refs for every pointer in
661  * the new block, increase lower level extents' reference count.
662  *
663  * Back Reference Key composing:
664  *
665  * The key objectid corresponds to the first byte in the extent,
666  * The key type is used to differentiate between types of back refs.
667  * There are different meanings of the key offset for different types
668  * of back refs.
669  *
670  * File extents can be referenced by:
671  *
672  * - multiple snapshots, subvolumes, or different generations in one subvol
673  * - different files inside a single subvolume
674  * - different offsets inside a file (bookend extents in file.c)
675  *
676  * The extent ref structure for the implicit back refs has fields for:
677  *
678  * - Objectid of the subvolume root
679  * - objectid of the file holding the reference
680  * - original offset in the file
681  * - how many bookend extents
682  *
683  * The key offset for the implicit back refs is hash of the first
684  * three fields.
685  *
686  * The extent ref structure for the full back refs has field for:
687  *
688  * - number of pointers in the tree leaf
689  *
690  * The key offset for the implicit back refs is the first byte of
691  * the tree leaf
692  *
693  * When a file extent is allocated, The implicit back refs is used.
694  * the fields are filled in:
695  *
696  *     (root_key.objectid, inode objectid, offset in file, 1)
697  *
698  * When a file extent is removed file truncation, we find the
699  * corresponding implicit back refs and check the following fields:
700  *
701  *     (btrfs_header_owner(leaf), inode objectid, offset in file)
702  *
703  * Btree extents can be referenced by:
704  *
705  * - Different subvolumes
706  *
707  * Both the implicit back refs and the full back refs for tree blocks
708  * only consist of key. The key offset for the implicit back refs is
709  * objectid of block's owner tree. The key offset for the full back refs
710  * is the first byte of parent block.
711  *
712  * When implicit back refs is used, information about the lowest key and
713  * level of the tree block are required. These information are stored in
714  * tree block info structure.
715  */
716 
717 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
718 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
719                                   struct btrfs_root *root,
720                                   struct btrfs_path *path,
721                                   u64 owner, u32 extra_size)
722 {
723         struct btrfs_extent_item *item;
724         struct btrfs_extent_item_v0 *ei0;
725         struct btrfs_extent_ref_v0 *ref0;
726         struct btrfs_tree_block_info *bi;
727         struct extent_buffer *leaf;
728         struct btrfs_key key;
729         struct btrfs_key found_key;
730         u32 new_size = sizeof(*item);
731         u64 refs;
732         int ret;
733 
734         leaf = path->nodes[0];
735         BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
736 
737         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
738         ei0 = btrfs_item_ptr(leaf, path->slots[0],
739                              struct btrfs_extent_item_v0);
740         refs = btrfs_extent_refs_v0(leaf, ei0);
741 
742         if (owner == (u64)-1) {
743                 while (1) {
744                         if (path->slots[0] >= btrfs_header_nritems(leaf)) {
745                                 ret = btrfs_next_leaf(root, path);
746                                 if (ret < 0)
747                                         return ret;
748                                 BUG_ON(ret > 0);
749                                 leaf = path->nodes[0];
750                         }
751                         btrfs_item_key_to_cpu(leaf, &found_key,
752                                               path->slots[0]);
753                         BUG_ON(key.objectid != found_key.objectid);
754                         if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
755                                 path->slots[0]++;
756                                 continue;
757                         }
758                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
759                                               struct btrfs_extent_ref_v0);
760                         owner = btrfs_ref_objectid_v0(leaf, ref0);
761                         break;
762                 }
763         }
764         btrfs_release_path(root, path);
765 
766         if (owner < BTRFS_FIRST_FREE_OBJECTID)
767                 new_size += sizeof(*bi);
768 
769         new_size -= sizeof(*ei0);
770         ret = btrfs_search_slot(trans, root, &key, path,
771                                 new_size + extra_size, 1);
772         if (ret < 0)
773                 return ret;
774         BUG_ON(ret);
775 
776         ret = btrfs_extend_item(trans, root, path, new_size);
777         BUG_ON(ret);
778 
779         leaf = path->nodes[0];
780         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
781         btrfs_set_extent_refs(leaf, item, refs);
782         /* FIXME: get real generation */
783         btrfs_set_extent_generation(leaf, item, 0);
784         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
785                 btrfs_set_extent_flags(leaf, item,
786                                        BTRFS_EXTENT_FLAG_TREE_BLOCK |
787                                        BTRFS_BLOCK_FLAG_FULL_BACKREF);
788                 bi = (struct btrfs_tree_block_info *)(item + 1);
789                 /* FIXME: get first key of the block */
790                 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
791                 btrfs_set_tree_block_level(leaf, bi, (int)owner);
792         } else {
793                 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
794         }
795         btrfs_mark_buffer_dirty(leaf);
796         return 0;
797 }
798 #endif
799 
800 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
801 {
802         u32 high_crc = ~(u32)0;
803         u32 low_crc = ~(u32)0;
804         __le64 lenum;
805 
806         lenum = cpu_to_le64(root_objectid);
807         high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
808         lenum = cpu_to_le64(owner);
809         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
810         lenum = cpu_to_le64(offset);
811         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
812 
813         return ((u64)high_crc << 31) ^ (u64)low_crc;
814 }
815 
816 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
817                                      struct btrfs_extent_data_ref *ref)
818 {
819         return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
820                                     btrfs_extent_data_ref_objectid(leaf, ref),
821                                     btrfs_extent_data_ref_offset(leaf, ref));
822 }
823 
824 static int match_extent_data_ref(struct extent_buffer *leaf,
825                                  struct btrfs_extent_data_ref *ref,
826                                  u64 root_objectid, u64 owner, u64 offset)
827 {
828         if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
829             btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
830             btrfs_extent_data_ref_offset(leaf, ref) != offset)
831                 return 0;
832         return 1;
833 }
834 
835 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
836                                            struct btrfs_root *root,
837                                            struct btrfs_path *path,
838                                            u64 bytenr, u64 parent,
839                                            u64 root_objectid,
840                                            u64 owner, u64 offset)
841 {
842         struct btrfs_key key;
843         struct btrfs_extent_data_ref *ref;
844         struct extent_buffer *leaf;
845         u32 nritems;
846         int ret;
847         int recow;
848         int err = -ENOENT;
849 
850         key.objectid = bytenr;
851         if (parent) {
852                 key.type = BTRFS_SHARED_DATA_REF_KEY;
853                 key.offset = parent;
854         } else {
855                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
856                 key.offset = hash_extent_data_ref(root_objectid,
857                                                   owner, offset);
858         }
859 again:
860         recow = 0;
861         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
862         if (ret < 0) {
863                 err = ret;
864                 goto fail;
865         }
866 
867         if (parent) {
868                 if (!ret)
869                         return 0;
870 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
871                 key.type = BTRFS_EXTENT_REF_V0_KEY;
872                 btrfs_release_path(root, path);
873                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
874                 if (ret < 0) {
875                         err = ret;
876                         goto fail;
877                 }
878                 if (!ret)
879                         return 0;
880 #endif
881                 goto fail;
882         }
883 
884         leaf = path->nodes[0];
885         nritems = btrfs_header_nritems(leaf);
886         while (1) {
887                 if (path->slots[0] >= nritems) {
888                         ret = btrfs_next_leaf(root, path);
889                         if (ret < 0)
890                                 err = ret;
891                         if (ret)
892                                 goto fail;
893 
894                         leaf = path->nodes[0];
895                         nritems = btrfs_header_nritems(leaf);
896                         recow = 1;
897                 }
898 
899                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
900                 if (key.objectid != bytenr ||
901                     key.type != BTRFS_EXTENT_DATA_REF_KEY)
902                         goto fail;
903 
904                 ref = btrfs_item_ptr(leaf, path->slots[0],
905                                      struct btrfs_extent_data_ref);
906 
907                 if (match_extent_data_ref(leaf, ref, root_objectid,
908                                           owner, offset)) {
909                         if (recow) {
910                                 btrfs_release_path(root, path);
911                                 goto again;
912                         }
913                         err = 0;
914                         break;
915                 }
916                 path->slots[0]++;
917         }
918 fail:
919         return err;
920 }
921 
922 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
923                                            struct btrfs_root *root,
924                                            struct btrfs_path *path,
925                                            u64 bytenr, u64 parent,
926                                            u64 root_objectid, u64 owner,
927                                            u64 offset, int refs_to_add)
928 {
929         struct btrfs_key key;
930         struct extent_buffer *leaf;
931         u32 size;
932         u32 num_refs;
933         int ret;
934 
935         key.objectid = bytenr;
936         if (parent) {
937                 key.type = BTRFS_SHARED_DATA_REF_KEY;
938                 key.offset = parent;
939                 size = sizeof(struct btrfs_shared_data_ref);
940         } else {
941                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
942                 key.offset = hash_extent_data_ref(root_objectid,
943                                                   owner, offset);
944                 size = sizeof(struct btrfs_extent_data_ref);
945         }
946 
947         ret = btrfs_insert_empty_item(trans, root, path, &key, size);
948         if (ret && ret != -EEXIST)
949                 goto fail;
950 
951         leaf = path->nodes[0];
952         if (parent) {
953                 struct btrfs_shared_data_ref *ref;
954                 ref = btrfs_item_ptr(leaf, path->slots[0],
955                                      struct btrfs_shared_data_ref);
956                 if (ret == 0) {
957                         btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
958                 } else {
959                         num_refs = btrfs_shared_data_ref_count(leaf, ref);
960                         num_refs += refs_to_add;
961                         btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
962                 }
963         } else {
964                 struct btrfs_extent_data_ref *ref;
965                 while (ret == -EEXIST) {
966                         ref = btrfs_item_ptr(leaf, path->slots[0],
967                                              struct btrfs_extent_data_ref);
968                         if (match_extent_data_ref(leaf, ref, root_objectid,
969                                                   owner, offset))
970                                 break;
971                         btrfs_release_path(root, path);
972                         key.offset++;
973                         ret = btrfs_insert_empty_item(trans, root, path, &key,
974                                                       size);
975                         if (ret && ret != -EEXIST)
976                                 goto fail;
977 
978                         leaf = path->nodes[0];
979                 }
980                 ref = btrfs_item_ptr(leaf, path->slots[0],
981                                      struct btrfs_extent_data_ref);
982                 if (ret == 0) {
983                         btrfs_set_extent_data_ref_root(leaf, ref,
984                                                        root_objectid);
985                         btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
986                         btrfs_set_extent_data_ref_offset(leaf, ref, offset);
987                         btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
988                 } else {
989                         num_refs = btrfs_extent_data_ref_count(leaf, ref);
990                         num_refs += refs_to_add;
991                         btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
992                 }
993         }
994         btrfs_mark_buffer_dirty(leaf);
995         ret = 0;
996 fail:
997         btrfs_release_path(root, path);
998         return ret;
999 }
1000 
1001 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1002                                            struct btrfs_root *root,
1003                                            struct btrfs_path *path,
1004                                            int refs_to_drop)
1005 {
1006         struct btrfs_key key;
1007         struct btrfs_extent_data_ref *ref1 = NULL;
1008         struct btrfs_shared_data_ref *ref2 = NULL;
1009         struct extent_buffer *leaf;
1010         u32 num_refs = 0;
1011         int ret = 0;
1012 
1013         leaf = path->nodes[0];
1014         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1015 
1016         if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1017                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1018                                       struct btrfs_extent_data_ref);
1019                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1020         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1021                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1022                                       struct btrfs_shared_data_ref);
1023                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1024 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1025         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1026                 struct btrfs_extent_ref_v0 *ref0;
1027                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1028                                       struct btrfs_extent_ref_v0);
1029                 num_refs = btrfs_ref_count_v0(leaf, ref0);
1030 #endif
1031         } else {
1032                 BUG();
1033         }
1034 
1035         BUG_ON(num_refs < refs_to_drop);
1036         num_refs -= refs_to_drop;
1037 
1038         if (num_refs == 0) {
1039                 ret = btrfs_del_item(trans, root, path);
1040         } else {
1041                 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1042                         btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1043                 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1044                         btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1045 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1046                 else {
1047                         struct btrfs_extent_ref_v0 *ref0;
1048                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
1049                                         struct btrfs_extent_ref_v0);
1050                         btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1051                 }
1052 #endif
1053                 btrfs_mark_buffer_dirty(leaf);
1054         }
1055         return ret;
1056 }
1057 
1058 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1059                                           struct btrfs_path *path,
1060                                           struct btrfs_extent_inline_ref *iref)
1061 {
1062         struct btrfs_key key;
1063         struct extent_buffer *leaf;
1064         struct btrfs_extent_data_ref *ref1;
1065         struct btrfs_shared_data_ref *ref2;
1066         u32 num_refs = 0;
1067 
1068         leaf = path->nodes[0];
1069         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1070         if (iref) {
1071                 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1072                     BTRFS_EXTENT_DATA_REF_KEY) {
1073                         ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1074                         num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1075                 } else {
1076                         ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1077                         num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1078                 }
1079         } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1080                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1081                                       struct btrfs_extent_data_ref);
1082                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1083         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1084                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1085                                       struct btrfs_shared_data_ref);
1086                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1087 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1088         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1089                 struct btrfs_extent_ref_v0 *ref0;
1090                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1091                                       struct btrfs_extent_ref_v0);
1092                 num_refs = btrfs_ref_count_v0(leaf, ref0);
1093 #endif
1094         } else {
1095                 WARN_ON(1);
1096         }
1097         return num_refs;
1098 }
1099 
1100 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1101                                           struct btrfs_root *root,
1102                                           struct btrfs_path *path,
1103                                           u64 bytenr, u64 parent,
1104                                           u64 root_objectid)
1105 {
1106         struct btrfs_key key;
1107         int ret;
1108 
1109         key.objectid = bytenr;
1110         if (parent) {
1111                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1112                 key.offset = parent;
1113         } else {
1114                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1115                 key.offset = root_objectid;
1116         }
1117 
1118         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1119         if (ret > 0)
1120                 ret = -ENOENT;
1121 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1122         if (ret == -ENOENT && parent) {
1123                 btrfs_release_path(root, path);
1124                 key.type = BTRFS_EXTENT_REF_V0_KEY;
1125                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1126                 if (ret > 0)
1127                         ret = -ENOENT;
1128         }
1129 #endif
1130         return ret;
1131 }
1132 
1133 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1134                                           struct btrfs_root *root,
1135                                           struct btrfs_path *path,
1136                                           u64 bytenr, u64 parent,
1137                                           u64 root_objectid)
1138 {
1139         struct btrfs_key key;
1140         int ret;
1141 
1142         key.objectid = bytenr;
1143         if (parent) {
1144                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1145                 key.offset = parent;
1146         } else {
1147                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1148                 key.offset = root_objectid;
1149         }
1150 
1151         ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1152         btrfs_release_path(root, path);
1153         return ret;
1154 }
1155 
1156 static inline int extent_ref_type(u64 parent, u64 owner)
1157 {
1158         int type;
1159         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1160                 if (parent > 0)
1161                         type = BTRFS_SHARED_BLOCK_REF_KEY;
1162                 else
1163                         type = BTRFS_TREE_BLOCK_REF_KEY;
1164         } else {
1165                 if (parent > 0)
1166                         type = BTRFS_SHARED_DATA_REF_KEY;
1167                 else
1168                         type = BTRFS_EXTENT_DATA_REF_KEY;
1169         }
1170         return type;
1171 }
1172 
1173 static int find_next_key(struct btrfs_path *path, int level,
1174                          struct btrfs_key *key)
1175 
1176 {
1177         for (; level < BTRFS_MAX_LEVEL; level++) {
1178                 if (!path->nodes[level])
1179                         break;
1180                 if (path->slots[level] + 1 >=
1181                     btrfs_header_nritems(path->nodes[level]))
1182                         continue;
1183                 if (level == 0)
1184                         btrfs_item_key_to_cpu(path->nodes[level], key,
1185                                               path->slots[level] + 1);
1186                 else
1187                         btrfs_node_key_to_cpu(path->nodes[level], key,
1188                                               path->slots[level] + 1);
1189                 return 0;
1190         }
1191         return 1;
1192 }
1193 
1194 /*
1195  * look for inline back ref. if back ref is found, *ref_ret is set
1196  * to the address of inline back ref, and 0 is returned.
1197  *
1198  * if back ref isn't found, *ref_ret is set to the address where it
1199  * should be inserted, and -ENOENT is returned.
1200  *
1201  * if insert is true and there are too many inline back refs, the path
1202  * points to the extent item, and -EAGAIN is returned.
1203  *
1204  * NOTE: inline back refs are ordered in the same way that back ref
1205  *       items in the tree are ordered.
1206  */
1207 static noinline_for_stack
1208 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1209                                  struct btrfs_root *root,
1210                                  struct btrfs_path *path,
1211                                  struct btrfs_extent_inline_ref **ref_ret,
1212                                  u64 bytenr, u64 num_bytes,
1213                                  u64 parent, u64 root_objectid,
1214                                  u64 owner, u64 offset, int insert)
1215 {
1216         struct btrfs_key key;
1217         struct extent_buffer *leaf;
1218         struct btrfs_extent_item *ei;
1219         struct btrfs_extent_inline_ref *iref;
1220         u64 flags;
1221         u64 item_size;
1222         unsigned long ptr;
1223         unsigned long end;
1224         int extra_size;
1225         int type;
1226         int want;
1227         int ret;
1228         int err = 0;
1229 
1230         key.objectid = bytenr;
1231         key.type = BTRFS_EXTENT_ITEM_KEY;
1232         key.offset = num_bytes;
1233 
1234         want = extent_ref_type(parent, owner);
1235         if (insert) {
1236                 extra_size = btrfs_extent_inline_ref_size(want);
1237                 path->keep_locks = 1;
1238         } else
1239                 extra_size = -1;
1240         ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1241         if (ret < 0) {
1242                 err = ret;
1243                 goto out;
1244         }
1245         BUG_ON(ret);
1246 
1247         leaf = path->nodes[0];
1248         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1249 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1250         if (item_size < sizeof(*ei)) {
1251                 if (!insert) {
1252                         err = -ENOENT;
1253                         goto out;
1254                 }
1255                 ret = convert_extent_item_v0(trans, root, path, owner,
1256                                              extra_size);
1257                 if (ret < 0) {
1258                         err = ret;
1259                         goto out;
1260                 }
1261                 leaf = path->nodes[0];
1262                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1263         }
1264 #endif
1265         BUG_ON(item_size < sizeof(*ei));
1266 
1267         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1268         flags = btrfs_extent_flags(leaf, ei);
1269 
1270         ptr = (unsigned long)(ei + 1);
1271         end = (unsigned long)ei + item_size;
1272 
1273         if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1274                 ptr += sizeof(struct btrfs_tree_block_info);
1275                 BUG_ON(ptr > end);
1276         } else {
1277                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1278         }
1279 
1280         err = -ENOENT;
1281         while (1) {
1282                 if (ptr >= end) {
1283                         WARN_ON(ptr > end);
1284                         break;
1285                 }
1286                 iref = (struct btrfs_extent_inline_ref *)ptr;
1287                 type = btrfs_extent_inline_ref_type(leaf, iref);
1288                 if (want < type)
1289                         break;
1290                 if (want > type) {
1291                         ptr += btrfs_extent_inline_ref_size(type);
1292                         continue;
1293                 }
1294 
1295                 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1296                         struct btrfs_extent_data_ref *dref;
1297                         dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1298                         if (match_extent_data_ref(leaf, dref, root_objectid,
1299                                                   owner, offset)) {
1300                                 err = 0;
1301                                 break;
1302                         }
1303                         if (hash_extent_data_ref_item(leaf, dref) <
1304                             hash_extent_data_ref(root_objectid, owner, offset))
1305                                 break;
1306                 } else {
1307                         u64 ref_offset;
1308                         ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1309                         if (parent > 0) {
1310                                 if (parent == ref_offset) {
1311                                         err = 0;
1312                                         break;
1313                                 }
1314                                 if (ref_offset < parent)
1315                                         break;
1316                         } else {
1317                                 if (root_objectid == ref_offset) {
1318                                         err = 0;
1319                                         break;
1320                                 }
1321                                 if (ref_offset < root_objectid)
1322                                         break;
1323                         }
1324                 }
1325                 ptr += btrfs_extent_inline_ref_size(type);
1326         }
1327         if (err == -ENOENT && insert) {
1328                 if (item_size + extra_size >=
1329                     BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1330                         err = -EAGAIN;
1331                         goto out;
1332                 }
1333                 /*
1334                  * To add new inline back ref, we have to make sure
1335                  * there is no corresponding back ref item.
1336                  * For simplicity, we just do not add new inline back
1337                  * ref if there is any kind of item for this block
1338                  */
1339                 if (find_next_key(path, 0, &key) == 0 &&
1340                     key.objectid == bytenr &&
1341                     key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1342                         err = -EAGAIN;
1343                         goto out;
1344                 }
1345         }
1346         *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1347 out:
1348         if (insert) {
1349                 path->keep_locks = 0;
1350                 btrfs_unlock_up_safe(path, 1);
1351         }
1352         return err;
1353 }
1354 
1355 /*
1356  * helper to add new inline back ref
1357  */
1358 static noinline_for_stack
1359 int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1360                                 struct btrfs_root *root,
1361                                 struct btrfs_path *path,
1362                                 struct btrfs_extent_inline_ref *iref,
1363                                 u64 parent, u64 root_objectid,
1364                                 u64 owner, u64 offset, int refs_to_add,
1365                                 struct btrfs_delayed_extent_op *extent_op)
1366 {
1367         struct extent_buffer *leaf;
1368         struct btrfs_extent_item *ei;
1369         unsigned long ptr;
1370         unsigned long end;
1371         unsigned long item_offset;
1372         u64 refs;
1373         int size;
1374         int type;
1375         int ret;
1376 
1377         leaf = path->nodes[0];
1378         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1379         item_offset = (unsigned long)iref - (unsigned long)ei;
1380 
1381         type = extent_ref_type(parent, owner);
1382         size = btrfs_extent_inline_ref_size(type);
1383 
1384         ret = btrfs_extend_item(trans, root, path, size);
1385         BUG_ON(ret);
1386 
1387         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1388         refs = btrfs_extent_refs(leaf, ei);
1389         refs += refs_to_add;
1390         btrfs_set_extent_refs(leaf, ei, refs);
1391         if (extent_op)
1392                 __run_delayed_extent_op(extent_op, leaf, ei);
1393 
1394         ptr = (unsigned long)ei + item_offset;
1395         end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1396         if (ptr < end - size)
1397                 memmove_extent_buffer(leaf, ptr + size, ptr,
1398                                       end - size - ptr);
1399 
1400         iref = (struct btrfs_extent_inline_ref *)ptr;
1401         btrfs_set_extent_inline_ref_type(leaf, iref, type);
1402         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1403                 struct btrfs_extent_data_ref *dref;
1404                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1405                 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1406                 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1407                 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1408                 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1409         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1410                 struct btrfs_shared_data_ref *sref;
1411                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1412                 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1413                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1414         } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1415                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1416         } else {
1417                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1418         }
1419         btrfs_mark_buffer_dirty(leaf);
1420         return 0;
1421 }
1422 
1423 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1424                                  struct btrfs_root *root,
1425                                  struct btrfs_path *path,
1426                                  struct btrfs_extent_inline_ref **ref_ret,
1427                                  u64 bytenr, u64 num_bytes, u64 parent,
1428                                  u64 root_objectid, u64 owner, u64 offset)
1429 {
1430         int ret;
1431 
1432         ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1433                                            bytenr, num_bytes, parent,
1434                                            root_objectid, owner, offset, 0);
1435         if (ret != -ENOENT)
1436                 return ret;
1437 
1438         btrfs_release_path(root, path);
1439         *ref_ret = NULL;
1440 
1441         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1442                 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1443                                             root_objectid);
1444         } else {
1445                 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1446                                              root_objectid, owner, offset);
1447         }
1448         return ret;
1449 }
1450 
1451 /*
1452  * helper to update/remove inline back ref
1453  */
1454 static noinline_for_stack
1455 int update_inline_extent_backref(struct btrfs_trans_handle *trans,
1456                                  struct btrfs_root *root,
1457                                  struct btrfs_path *path,
1458                                  struct btrfs_extent_inline_ref *iref,
1459                                  int refs_to_mod,
1460                                  struct btrfs_delayed_extent_op *extent_op)
1461 {
1462         struct extent_buffer *leaf;
1463         struct btrfs_extent_item *ei;
1464         struct btrfs_extent_data_ref *dref = NULL;
1465         struct btrfs_shared_data_ref *sref = NULL;
1466         unsigned long ptr;
1467         unsigned long end;
1468         u32 item_size;
1469         int size;
1470         int type;
1471         int ret;
1472         u64 refs;
1473 
1474         leaf = path->nodes[0];
1475         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1476         refs = btrfs_extent_refs(leaf, ei);
1477         WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1478         refs += refs_to_mod;
1479         btrfs_set_extent_refs(leaf, ei, refs);
1480         if (extent_op)
1481                 __run_delayed_extent_op(extent_op, leaf, ei);
1482 
1483         type = btrfs_extent_inline_ref_type(leaf, iref);
1484 
1485         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1486                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1487                 refs = btrfs_extent_data_ref_count(leaf, dref);
1488         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1489                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1490                 refs = btrfs_shared_data_ref_count(leaf, sref);
1491         } else {
1492                 refs = 1;
1493                 BUG_ON(refs_to_mod != -1);
1494         }
1495 
1496         BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1497         refs += refs_to_mod;
1498 
1499         if (refs > 0) {
1500                 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1501                         btrfs_set_extent_data_ref_count(leaf, dref, refs);
1502                 else
1503                         btrfs_set_shared_data_ref_count(leaf, sref, refs);
1504         } else {
1505                 size =  btrfs_extent_inline_ref_size(type);
1506                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1507                 ptr = (unsigned long)iref;
1508                 end = (unsigned long)ei + item_size;
1509                 if (ptr + size < end)
1510                         memmove_extent_buffer(leaf, ptr, ptr + size,
1511                                               end - ptr - size);
1512                 item_size -= size;
1513                 ret = btrfs_truncate_item(trans, root, path, item_size, 1);
1514                 BUG_ON(ret);
1515         }
1516         btrfs_mark_buffer_dirty(leaf);
1517         return 0;
1518 }
1519 
1520 static noinline_for_stack
1521 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1522                                  struct btrfs_root *root,
1523                                  struct btrfs_path *path,
1524                                  u64 bytenr, u64 num_bytes, u64 parent,
1525                                  u64 root_objectid, u64 owner,
1526                                  u64 offset, int refs_to_add,
1527                                  struct btrfs_delayed_extent_op *extent_op)
1528 {
1529         struct btrfs_extent_inline_ref *iref;
1530         int ret;
1531 
1532         ret = lookup_inline_extent_backref(trans, root, path, &iref,
1533                                            bytenr, num_bytes, parent,
1534                                            root_objectid, owner, offset, 1);
1535         if (ret == 0) {
1536                 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1537                 ret = update_inline_extent_backref(trans, root, path, iref,
1538                                                    refs_to_add, extent_op);
1539         } else if (ret == -ENOENT) {
1540                 ret = setup_inline_extent_backref(trans, root, path, iref,
1541                                                   parent, root_objectid,
1542                                                   owner, offset, refs_to_add,
1543                                                   extent_op);
1544         }
1545         return ret;
1546 }
1547 
1548 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1549                                  struct btrfs_root *root,
1550                                  struct btrfs_path *path,
1551                                  u64 bytenr, u64 parent, u64 root_objectid,
1552                                  u64 owner, u64 offset, int refs_to_add)
1553 {
1554         int ret;
1555         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1556                 BUG_ON(refs_to_add != 1);
1557                 ret = insert_tree_block_ref(trans, root, path, bytenr,
1558                                             parent, root_objectid);
1559         } else {
1560                 ret = insert_extent_data_ref(trans, root, path, bytenr,
1561                                              parent, root_objectid,
1562                                              owner, offset, refs_to_add);
1563         }
1564         return ret;
1565 }
1566 
1567 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1568                                  struct btrfs_root *root,
1569                                  struct btrfs_path *path,
1570                                  struct btrfs_extent_inline_ref *iref,
1571                                  int refs_to_drop, int is_data)
1572 {
1573         int ret;
1574 
1575         BUG_ON(!is_data && refs_to_drop != 1);
1576         if (iref) {
1577                 ret = update_inline_extent_backref(trans, root, path, iref,
1578                                                    -refs_to_drop, NULL);
1579         } else if (is_data) {
1580                 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1581         } else {
1582                 ret = btrfs_del_item(trans, root, path);
1583         }
1584         return ret;
1585 }
1586 
1587 static void btrfs_issue_discard(struct block_device *bdev,
1588                                 u64 start, u64 len)
1589 {
1590         blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL,
1591                              DISCARD_FL_BARRIER);
1592 }
1593 
1594 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1595                                 u64 num_bytes)
1596 {
1597         int ret;
1598         u64 map_length = num_bytes;
1599         struct btrfs_multi_bio *multi = NULL;
1600 
1601         if (!btrfs_test_opt(root, DISCARD))
1602                 return 0;
1603 
1604         /* Tell the block device(s) that the sectors can be discarded */
1605         ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
1606                               bytenr, &map_length, &multi, 0);
1607         if (!ret) {
1608                 struct btrfs_bio_stripe *stripe = multi->stripes;
1609                 int i;
1610 
1611                 if (map_length > num_bytes)
1612                         map_length = num_bytes;
1613 
1614                 for (i = 0; i < multi->num_stripes; i++, stripe++) {
1615                         btrfs_issue_discard(stripe->dev->bdev,
1616                                             stripe->physical,
1617                                             map_length);
1618                 }
1619                 kfree(multi);
1620         }
1621 
1622         return ret;
1623 }
1624 
1625 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1626                          struct btrfs_root *root,
1627                          u64 bytenr, u64 num_bytes, u64 parent,
1628                          u64 root_objectid, u64 owner, u64 offset)
1629 {
1630         int ret;
1631         BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1632                root_objectid == BTRFS_TREE_LOG_OBJECTID);
1633 
1634         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1635                 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
1636                                         parent, root_objectid, (int)owner,
1637                                         BTRFS_ADD_DELAYED_REF, NULL);
1638         } else {
1639                 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
1640                                         parent, root_objectid, owner, offset,
1641                                         BTRFS_ADD_DELAYED_REF, NULL);
1642         }
1643         return ret;
1644 }
1645 
1646 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1647                                   struct btrfs_root *root,
1648                                   u64 bytenr, u64 num_bytes,
1649                                   u64 parent, u64 root_objectid,
1650                                   u64 owner, u64 offset, int refs_to_add,
1651                                   struct btrfs_delayed_extent_op *extent_op)
1652 {
1653         struct btrfs_path *path;
1654         struct extent_buffer *leaf;
1655         struct btrfs_extent_item *item;
1656         u64 refs;
1657         int ret;
1658         int err = 0;
1659 
1660         path = btrfs_alloc_path();
1661         if (!path)
1662                 return -ENOMEM;
1663 
1664         path->reada = 1;
1665         path->leave_spinning = 1;
1666         /* this will setup the path even if it fails to insert the back ref */
1667         ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1668                                            path, bytenr, num_bytes, parent,
1669                                            root_objectid, owner, offset,
1670                                            refs_to_add, extent_op);
1671         if (ret == 0)
1672                 goto out;
1673 
1674         if (ret != -EAGAIN) {
1675                 err = ret;
1676                 goto out;
1677         }
1678 
1679         leaf = path->nodes[0];
1680         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1681         refs = btrfs_extent_refs(leaf, item);
1682         btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1683         if (extent_op)
1684                 __run_delayed_extent_op(extent_op, leaf, item);
1685 
1686         btrfs_mark_buffer_dirty(leaf);
1687         btrfs_release_path(root->fs_info->extent_root, path);
1688 
1689         path->reada = 1;
1690         path->leave_spinning = 1;
1691 
1692         /* now insert the actual backref */
1693         ret = insert_extent_backref(trans, root->fs_info->extent_root,
1694                                     path, bytenr, parent, root_objectid,
1695                                     owner, offset, refs_to_add);
1696         BUG_ON(ret);
1697 out:
1698         btrfs_free_path(path);
1699         return err;
1700 }
1701 
1702 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1703                                 struct btrfs_root *root,
1704                                 struct btrfs_delayed_ref_node *node,
1705                                 struct btrfs_delayed_extent_op *extent_op,
1706                                 int insert_reserved)
1707 {
1708         int ret = 0;
1709         struct btrfs_delayed_data_ref *ref;
1710         struct btrfs_key ins;
1711         u64 parent = 0;
1712         u64 ref_root = 0;
1713         u64 flags = 0;
1714 
1715         ins.objectid = node->bytenr;
1716         ins.offset = node->num_bytes;
1717         ins.type = BTRFS_EXTENT_ITEM_KEY;
1718 
1719         ref = btrfs_delayed_node_to_data_ref(node);
1720         if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1721                 parent = ref->parent;
1722         else
1723                 ref_root = ref->root;
1724 
1725         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1726                 if (extent_op) {
1727                         BUG_ON(extent_op->update_key);
1728                         flags |= extent_op->flags_to_set;
1729                 }
1730                 ret = alloc_reserved_file_extent(trans, root,
1731                                                  parent, ref_root, flags,
1732                                                  ref->objectid, ref->offset,
1733                                                  &ins, node->ref_mod);
1734         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1735                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1736                                              node->num_bytes, parent,
1737                                              ref_root, ref->objectid,
1738                                              ref->offset, node->ref_mod,
1739                                              extent_op);
1740         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1741                 ret = __btrfs_free_extent(trans, root, node->bytenr,
1742                                           node->num_bytes, parent,
1743                                           ref_root, ref->objectid,
1744                                           ref->offset, node->ref_mod,
1745                                           extent_op);
1746         } else {
1747                 BUG();
1748         }
1749         return ret;
1750 }
1751 
1752 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1753                                     struct extent_buffer *leaf,
1754                                     struct btrfs_extent_item *ei)
1755 {
1756         u64 flags = btrfs_extent_flags(leaf, ei);
1757         if (extent_op->update_flags) {
1758                 flags |= extent_op->flags_to_set;
1759                 btrfs_set_extent_flags(leaf, ei, flags);
1760         }
1761 
1762         if (extent_op->update_key) {
1763                 struct btrfs_tree_block_info *bi;
1764                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
1765                 bi = (struct btrfs_tree_block_info *)(ei + 1);
1766                 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
1767         }
1768 }
1769 
1770 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
1771                                  struct btrfs_root *root,
1772                                  struct btrfs_delayed_ref_node *node,
1773                                  struct btrfs_delayed_extent_op *extent_op)
1774 {
1775         struct btrfs_key key;
1776         struct btrfs_path *path;
1777         struct btrfs_extent_item *ei;
1778         struct extent_buffer *leaf;
1779         u32 item_size;
1780         int ret;
1781         int err = 0;
1782 
1783         path = btrfs_alloc_path();
1784         if (!path)
1785                 return -ENOMEM;
1786 
1787         key.objectid = node->bytenr;
1788         key.type = BTRFS_EXTENT_ITEM_KEY;
1789         key.offset = node->num_bytes;
1790 
1791         path->reada = 1;
1792         path->leave_spinning = 1;
1793         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
1794                                 path, 0, 1);
1795         if (ret < 0) {
1796                 err = ret;
1797                 goto out;
1798         }
1799         if (ret > 0) {
1800                 err = -EIO;
1801                 goto out;
1802         }
1803 
1804         leaf = path->nodes[0];
1805         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1806 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1807         if (item_size < sizeof(*ei)) {
1808                 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
1809                                              path, (u64)-1, 0);
1810                 if (ret < 0) {
1811                         err = ret;
1812                         goto out;
1813                 }
1814                 leaf = path->nodes[0];
1815                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1816         }
1817 #endif
1818         BUG_ON(item_size < sizeof(*ei));
1819         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1820         __run_delayed_extent_op(extent_op, leaf, ei);
1821 
1822         btrfs_mark_buffer_dirty(leaf);
1823 out:
1824         btrfs_free_path(path);
1825         return err;
1826 }
1827 
1828 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
1829                                 struct btrfs_root *root,
1830                                 struct btrfs_delayed_ref_node *node,
1831                                 struct btrfs_delayed_extent_op *extent_op,
1832                                 int insert_reserved)
1833 {
1834         int ret = 0;
1835         struct btrfs_delayed_tree_ref *ref;
1836         struct btrfs_key ins;
1837         u64 parent = 0;
1838         u64 ref_root = 0;
1839 
1840         ins.objectid = node->bytenr;
1841         ins.offset = node->num_bytes;
1842         ins.type = BTRFS_EXTENT_ITEM_KEY;
1843 
1844         ref = btrfs_delayed_node_to_tree_ref(node);
1845         if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1846                 parent = ref->parent;
1847         else
1848                 ref_root = ref->root;
1849 
1850         BUG_ON(node->ref_mod != 1);
1851         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1852                 BUG_ON(!extent_op || !extent_op->update_flags ||
1853                        !extent_op->update_key);
1854                 ret = alloc_reserved_tree_block(trans, root,
1855                                                 parent, ref_root,
1856                                                 extent_op->flags_to_set,
1857                                                 &extent_op->key,
1858                                                 ref->level, &ins);
1859         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1860                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1861                                              node->num_bytes, parent, ref_root,
1862                                              ref->level, 0, 1, extent_op);
1863         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1864                 ret = __btrfs_free_extent(trans, root, node->bytenr,
1865                                           node->num_bytes, parent, ref_root,
1866                                           ref->level, 0, 1, extent_op);
1867         } else {
1868                 BUG();
1869         }
1870         return ret;
1871 }
1872 
1873 
1874 /* helper function to actually process a single delayed ref entry */
1875 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
1876                                struct btrfs_root *root,
1877                                struct btrfs_delayed_ref_node *node,
1878                                struct btrfs_delayed_extent_op *extent_op,
1879                                int insert_reserved)
1880 {
1881         int ret;
1882         if (btrfs_delayed_ref_is_head(node)) {
1883                 struct btrfs_delayed_ref_head *head;
1884                 /*
1885                  * we've hit the end of the chain and we were supposed
1886                  * to insert this extent into the tree.  But, it got
1887                  * deleted before we ever needed to insert it, so all
1888                  * we have to do is clean up the accounting
1889                  */
1890                 BUG_ON(extent_op);
1891                 head = btrfs_delayed_node_to_head(node);
1892                 if (insert_reserved) {
1893                         int mark_free = 0;
1894                         struct extent_buffer *must_clean = NULL;
1895 
1896                         ret = pin_down_bytes(trans, root, NULL,
1897                                              node->bytenr, node->num_bytes,
1898                                              head->is_data, 1, &must_clean);
1899                         if (ret > 0)
1900                                 mark_free = 1;
1901 
1902                         if (must_clean) {
1903                                 clean_tree_block(NULL, root, must_clean);
1904                                 btrfs_tree_unlock(must_clean);
1905                                 free_extent_buffer(must_clean);
1906                         }
1907                         if (head->is_data) {
1908                                 ret = btrfs_del_csums(trans, root,
1909                                                       node->bytenr,
1910                                                       node->num_bytes);
1911                                 BUG_ON(ret);
1912                         }
1913                         if (mark_free) {
1914                                 ret = btrfs_free_reserved_extent(root,
1915                                                         node->bytenr,
1916                                                         node->num_bytes);
1917                                 BUG_ON(ret);
1918                         }
1919                 }
1920                 mutex_unlock(&head->mutex);
1921                 return 0;
1922         }
1923 
1924         if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
1925             node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1926                 ret = run_delayed_tree_ref(trans, root, node, extent_op,
1927                                            insert_reserved);
1928         else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
1929                  node->type == BTRFS_SHARED_DATA_REF_KEY)
1930                 ret = run_delayed_data_ref(trans, root, node, extent_op,
1931                                            insert_reserved);
1932         else
1933                 BUG();
1934         return ret;
1935 }
1936 
1937 static noinline struct btrfs_delayed_ref_node *
1938 select_delayed_ref(struct btrfs_delayed_ref_head *head)
1939 {
1940         struct rb_node *node;
1941         struct btrfs_delayed_ref_node *ref;
1942         int action = BTRFS_ADD_DELAYED_REF;
1943 again:
1944         /*
1945          * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
1946          * this prevents ref count from going down to zero when
1947          * there still are pending delayed ref.
1948          */
1949         node = rb_prev(&head->node.rb_node);
1950         while (1) {
1951                 if (!node)
1952                         break;
1953                 ref = rb_entry(node, struct btrfs_delayed_ref_node,
1954                                 rb_node);
1955                 if (ref->bytenr != head->node.bytenr)
1956                         break;
1957                 if (ref->action == action)
1958                         return ref;
1959                 node = rb_prev(node);
1960         }
1961         if (action == BTRFS_ADD_DELAYED_REF) {
1962                 action = BTRFS_DROP_DELAYED_REF;
1963                 goto again;
1964         }
1965         return NULL;
1966 }
1967 
1968 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
1969                                        struct btrfs_root *root,
1970                                        struct list_head *cluster)
1971 {
1972         struct btrfs_delayed_ref_root *delayed_refs;
1973         struct btrfs_delayed_ref_node *ref;
1974         struct btrfs_delayed_ref_head *locked_ref = NULL;
1975         struct btrfs_delayed_extent_op *extent_op;
1976         int ret;
1977         int count = 0;
1978         int must_insert_reserved = 0;
1979 
1980         delayed_refs = &trans->transaction->delayed_refs;
1981         while (1) {
1982                 if (!locked_ref) {
1983                         /* pick a new head ref from the cluster list */
1984                         if (list_empty(cluster))
1985                                 break;
1986 
1987                         locked_ref = list_entry(cluster->next,
1988                                      struct btrfs_delayed_ref_head, cluster);
1989 
1990                         /* grab the lock that says we are going to process
1991                          * all the refs for this head */
1992                         ret = btrfs_delayed_ref_lock(trans, locked_ref);
1993 
1994                         /*
1995                          * we may have dropped the spin lock to get the head
1996                          * mutex lock, and that might have given someone else
1997                          * time to free the head.  If that's true, it has been
1998                          * removed from our list and we can move on.
1999                          */
2000                         if (ret == -EAGAIN) {
2001                                 locked_ref = NULL;
2002                                 count++;
2003                                 continue;
2004                         }
2005                 }
2006 
2007                 /*
2008                  * record the must insert reserved flag before we
2009                  * drop the spin lock.
2010                  */
2011                 must_insert_reserved = locked_ref->must_insert_reserved;
2012                 locked_ref->must_insert_reserved = 0;
2013 
2014                 extent_op = locked_ref->extent_op;
2015                 locked_ref->extent_op = NULL;
2016 
2017                 /*
2018                  * locked_ref is the head node, so we have to go one
2019                  * node back for any delayed ref updates
2020                  */
2021                 ref = select_delayed_ref(locked_ref);
2022                 if (!ref) {
2023                         /* All delayed refs have been processed, Go ahead
2024                          * and send the head node to run_one_delayed_ref,
2025                          * so that any accounting fixes can happen
2026                          */
2027                         ref = &locked_ref->node;
2028 
2029                         if (extent_op && must_insert_reserved) {
2030                                 kfree(extent_op);
2031                                 extent_op = NULL;
2032                         }
2033 
2034                         if (extent_op) {
2035                                 spin_unlock(&delayed_refs->lock);
2036 
2037                                 ret = run_delayed_extent_op(trans, root,
2038                                                             ref, extent_op);
2039                                 BUG_ON(ret);
2040                                 kfree(extent_op);
2041 
2042                                 cond_resched();
2043                                 spin_lock(&delayed_refs->lock);
2044                                 continue;
2045                         }
2046 
2047                         list_del_init(&locked_ref->cluster);
2048                         locked_ref = NULL;
2049                 }
2050 
2051                 ref->in_tree = 0;
2052                 rb_erase(&ref->rb_node, &delayed_refs->root);
2053                 delayed_refs->num_entries--;
2054 
2055                 spin_unlock(&delayed_refs->lock);
2056 
2057                 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2058                                           must_insert_reserved);
2059                 BUG_ON(ret);
2060 
2061                 btrfs_put_delayed_ref(ref);
2062                 kfree(extent_op);
2063                 count++;
2064 
2065                 cond_resched();
2066                 spin_lock(&delayed_refs->lock);
2067         }
2068         return count;
2069 }
2070 
2071 /*
2072  * this starts processing the delayed reference count updates and
2073  * extent insertions we have queued up so far.  count can be
2074  * 0, which means to process everything in the tree at the start
2075  * of the run (but not newly added entries), or it can be some target
2076  * number you'd like to process.
2077  */
2078 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2079                            struct btrfs_root *root, unsigned long count)
2080 {
2081         struct rb_node *node;
2082         struct btrfs_delayed_ref_root *delayed_refs;
2083         struct btrfs_delayed_ref_node *ref;
2084         struct list_head cluster;
2085         int ret;
2086         int run_all = count == (unsigned long)-1;
2087         int run_most = 0;
2088 
2089         if (root == root->fs_info->extent_root)
2090                 root = root->fs_info->tree_root;
2091 
2092         delayed_refs = &trans->transaction->delayed_refs;
2093         INIT_LIST_HEAD(&cluster);
2094 again:
2095         spin_lock(&delayed_refs->lock);
2096         if (count == 0) {
2097                 count = delayed_refs->num_entries * 2;
2098                 run_most = 1;
2099         }
2100         while (1) {
2101                 if (!(run_all || run_most) &&
2102                     delayed_refs->num_heads_ready < 64)
2103                         break;
2104 
2105                 /*
2106                  * go find something we can process in the rbtree.  We start at
2107                  * the beginning of the tree, and then build a cluster
2108                  * of refs to process starting at the first one we are able to
2109                  * lock
2110                  */
2111                 ret = btrfs_find_ref_cluster(trans, &cluster,
2112                                              delayed_refs->run_delayed_start);
2113                 if (ret)
2114                         break;
2115 
2116                 ret = run_clustered_refs(trans, root, &cluster);
2117                 BUG_ON(ret < 0);
2118 
2119                 count -= min_t(unsigned long, ret, count);
2120 
2121                 if (count == 0)
2122                         break;
2123         }
2124 
2125         if (run_all) {
2126                 node = rb_first(&delayed_refs->root);
2127                 if (!node)
2128                         goto out;
2129                 count = (unsigned long)-1;
2130 
2131                 while (node) {
2132                         ref = rb_entry(node, struct btrfs_delayed_ref_node,
2133                                        rb_node);
2134                         if (btrfs_delayed_ref_is_head(ref)) {
2135                                 struct btrfs_delayed_ref_head *head;
2136 
2137                                 head = btrfs_delayed_node_to_head(ref);
2138                                 atomic_inc(&ref->refs);
2139 
2140                                 spin_unlock(&delayed_refs->lock);
2141                                 mutex_lock(&head->mutex);
2142                                 mutex_unlock(&head->mutex);
2143 
2144                                 btrfs_put_delayed_ref(ref);
2145                                 cond_resched();
2146                                 goto again;
2147                         }
2148                         node = rb_next(node);
2149                 }
2150                 spin_unlock(&delayed_refs->lock);
2151                 schedule_timeout(1);
2152                 goto again;
2153         }
2154 out:
2155         spin_unlock(&delayed_refs->lock);
2156         return 0;
2157 }
2158 
2159 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2160                                 struct btrfs_root *root,
2161                                 u64 bytenr, u64 num_bytes, u64 flags,
2162                                 int is_data)
2163 {
2164         struct btrfs_delayed_extent_op *extent_op;
2165         int ret;
2166 
2167         extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2168         if (!extent_op)
2169                 return -ENOMEM;
2170 
2171         extent_op->flags_to_set = flags;
2172         extent_op->update_flags = 1;
2173         extent_op->update_key = 0;
2174         extent_op->is_data = is_data ? 1 : 0;
2175 
2176         ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op);
2177         if (ret)
2178                 kfree(extent_op);
2179         return ret;
2180 }
2181 
2182 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2183                                       struct btrfs_root *root,
2184                                       struct btrfs_path *path,
2185                                       u64 objectid, u64 offset, u64 bytenr)
2186 {
2187         struct btrfs_delayed_ref_head *head;
2188         struct btrfs_delayed_ref_node *ref;
2189         struct btrfs_delayed_data_ref *data_ref;
2190         struct btrfs_delayed_ref_root *delayed_refs;
2191         struct rb_node *node;
2192         int ret = 0;
2193 
2194         ret = -ENOENT;
2195         delayed_refs = &trans->transaction->delayed_refs;
2196         spin_lock(&delayed_refs->lock);
2197         head = btrfs_find_delayed_ref_head(trans, bytenr);
2198         if (!head)
2199                 goto out;
2200 
2201         if (!mutex_trylock(&head->mutex)) {
2202                 atomic_inc(&head->node.refs);
2203                 spin_unlock(&delayed_refs->lock);
2204 
2205                 btrfs_release_path(root->fs_info->extent_root, path);
2206 
2207                 mutex_lock(&head->mutex);
2208                 mutex_unlock(&head->mutex);
2209                 btrfs_put_delayed_ref(&head->node);
2210                 return -EAGAIN;
2211         }
2212 
2213         node = rb_prev(&head->node.rb_node);
2214         if (!node)
2215                 goto out_unlock;
2216 
2217         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2218 
2219         if (ref->bytenr != bytenr)
2220                 goto out_unlock;
2221 
2222         ret = 1;
2223         if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2224                 goto out_unlock;
2225 
2226         data_ref = btrfs_delayed_node_to_data_ref(ref);
2227 
2228         node = rb_prev(node);
2229         if (node) {
2230                 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2231                 if (ref->bytenr == bytenr)
2232                         goto out_unlock;
2233         }
2234 
2235         if (data_ref->root != root->root_key.objectid ||
2236             data_ref->objectid != objectid || data_ref->offset != offset)
2237                 goto out_unlock;
2238 
2239         ret = 0;
2240 out_unlock:
2241         mutex_unlock(&head->mutex);
2242 out:
2243         spin_unlock(&delayed_refs->lock);
2244         return ret;
2245 }
2246 
2247 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2248                                         struct btrfs_root *root,
2249                                         struct btrfs_path *path,
2250                                         u64 objectid, u64 offset, u64 bytenr)
2251 {
2252         struct btrfs_root *extent_root = root->fs_info->extent_root;
2253         struct extent_buffer *leaf;
2254         struct btrfs_extent_data_ref *ref;
2255         struct btrfs_extent_inline_ref *iref;
2256         struct btrfs_extent_item *ei;
2257         struct btrfs_key key;
2258         u32 item_size;
2259         int ret;
2260 
2261         key.objectid = bytenr;
2262         key.offset = (u64)-1;
2263         key.type = BTRFS_EXTENT_ITEM_KEY;
2264 
2265         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2266         if (ret < 0)
2267                 goto out;
2268         BUG_ON(ret == 0);
2269 
2270         ret = -ENOENT;
2271         if (path->slots[0] == 0)
2272                 goto out;
2273 
2274         path->slots[0]--;
2275         leaf = path->nodes[0];
2276         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2277 
2278         if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2279                 goto out;
2280 
2281         ret = 1;
2282         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2283 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2284         if (item_size < sizeof(*ei)) {
2285                 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2286                 goto out;
2287         }
2288 #endif
2289         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2290 
2291         if (item_size != sizeof(*ei) +
2292             btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2293                 goto out;
2294 
2295         if (btrfs_extent_generation(leaf, ei) <=
2296             btrfs_root_last_snapshot(&root->root_item))
2297                 goto out;
2298 
2299         iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2300         if (btrfs_extent_inline_ref_type(leaf, iref) !=
2301             BTRFS_EXTENT_DATA_REF_KEY)
2302                 goto out;
2303 
2304         ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2305         if (btrfs_extent_refs(leaf, ei) !=
2306             btrfs_extent_data_ref_count(leaf, ref) ||
2307             btrfs_extent_data_ref_root(leaf, ref) !=
2308             root->root_key.objectid ||
2309             btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2310             btrfs_extent_data_ref_offset(leaf, ref) != offset)
2311                 goto out;
2312 
2313         ret = 0;
2314 out:
2315         return ret;
2316 }
2317 
2318 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2319                           struct btrfs_root *root,
2320                           u64 objectid, u64 offset, u64 bytenr)
2321 {
2322         struct btrfs_path *path;
2323         int ret;
2324         int ret2;
2325 
2326         path = btrfs_alloc_path();
2327         if (!path)
2328                 return -ENOENT;
2329 
2330         do {
2331                 ret = check_committed_ref(trans, root, path, objectid,
2332                                           offset, bytenr);
2333                 if (ret && ret != -ENOENT)
2334                         goto out;
2335 
2336                 ret2 = check_delayed_ref(trans, root, path, objectid,
2337                                          offset, bytenr);
2338         } while (ret2 == -EAGAIN);
2339 
2340         if (ret2 && ret2 != -ENOENT) {
2341                 ret = ret2;
2342                 goto out;
2343         }
2344 
2345         if (ret != -ENOENT || ret2 != -ENOENT)
2346                 ret = 0;
2347 out:
2348         btrfs_free_path(path);
2349         return ret;
2350 }
2351 
2352 #if 0
2353 int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2354                     struct extent_buffer *buf, u32 nr_extents)
2355 {
2356         struct btrfs_key key;
2357         struct btrfs_file_extent_item *fi;
2358         u64 root_gen;
2359         u32 nritems;
2360         int i;
2361         int level;
2362         int ret = 0;
2363         int shared = 0;
2364 
2365         if (!root->ref_cows)
2366                 return 0;
2367 
2368         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
2369                 shared = 0;
2370                 root_gen = root->root_key.offset;
2371         } else {
2372                 shared = 1;
2373                 root_gen = trans->transid - 1;
2374         }
2375 
2376         level = btrfs_header_level(buf);
2377         nritems = btrfs_header_nritems(buf);
2378 
2379         if (level == 0) {
2380                 struct btrfs_leaf_ref *ref;
2381                 struct btrfs_extent_info *info;
2382 
2383                 ref = btrfs_alloc_leaf_ref(root, nr_extents);
2384                 if (!ref) {
2385                         ret = -ENOMEM;
2386                         goto out;
2387                 }
2388 
2389                 ref->root_gen = root_gen;
2390                 ref->bytenr = buf->start;
2391                 ref->owner = btrfs_header_owner(buf);
2392                 ref->generation = btrfs_header_generation(buf);
2393                 ref->nritems = nr_extents;
2394                 info = ref->extents;
2395 
2396                 for (i = 0; nr_extents > 0 && i < nritems; i++) {
2397                         u64 disk_bytenr;
2398                         btrfs_item_key_to_cpu(buf, &key, i);
2399                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2400                                 continue;
2401                         fi = btrfs_item_ptr(buf, i,
2402                                             struct btrfs_file_extent_item);
2403                         if (btrfs_file_extent_type(buf, fi) ==
2404                             BTRFS_FILE_EXTENT_INLINE)
2405                                 continue;
2406                         disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2407                         if (disk_bytenr == 0)
2408                                 continue;
2409 
2410                         info->bytenr = disk_bytenr;
2411                         info->num_bytes =
2412                                 btrfs_file_extent_disk_num_bytes(buf, fi);
2413                         info->objectid = key.objectid;
2414                         info->offset = key.offset;
2415                         info++;
2416                 }
2417 
2418                 ret = btrfs_add_leaf_ref(root, ref, shared);
2419                 if (ret == -EEXIST && shared) {
2420                         struct btrfs_leaf_ref *old;
2421                         old = btrfs_lookup_leaf_ref(root, ref->bytenr);
2422                         BUG_ON(!old);
2423                         btrfs_remove_leaf_ref(root, old);
2424                         btrfs_free_leaf_ref(root, old);
2425                         ret = btrfs_add_leaf_ref(root, ref, shared);
2426                 }
2427                 WARN_ON(ret);
2428                 btrfs_free_leaf_ref(root, ref);
2429         }
2430 out:
2431         return ret;
2432 }
2433 
2434 /* when a block goes through cow, we update the reference counts of
2435  * everything that block points to.  The internal pointers of the block
2436  * can be in just about any order, and it is likely to have clusters of
2437  * things that are close together and clusters of things that are not.
2438  *
2439  * To help reduce the seeks that come with updating all of these reference
2440  * counts, sort them by byte number before actual updates are done.
2441  *
2442  * struct refsort is used to match byte number to slot in the btree block.
2443  * we sort based on the byte number and then use the slot to actually
2444  * find the item.
2445  *
2446  * struct refsort is smaller than strcut btrfs_item and smaller than
2447  * struct btrfs_key_ptr.  Since we're currently limited to the page size
2448  * for a btree block, there's no way for a kmalloc of refsorts for a
2449  * single node to be bigger than a page.
2450  */
2451 struct refsort {
2452         u64 bytenr;
2453         u32 slot;
2454 };
2455 
2456 /*
2457  * for passing into sort()
2458  */
2459 static int refsort_cmp(const void *a_void, const void *b_void)
2460 {
2461         const struct refsort *a = a_void;
2462         const struct refsort *b = b_void;
2463 
2464         if (a->bytenr < b->bytenr)
2465                 return -1;
2466         if (a->bytenr > b->bytenr)
2467                 return 1;
2468         return 0;
2469 }
2470 #endif
2471 
2472 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2473                            struct btrfs_root *root,
2474                            struct extent_buffer *buf,
2475                            int full_backref, int inc)
2476 {
2477         u64 bytenr;
2478         u64 num_bytes;
2479         u64 parent;
2480         u64 ref_root;
2481         u32 nritems;
2482         struct btrfs_key key;
2483         struct btrfs_file_extent_item *fi;
2484         int i;
2485         int level;
2486         int ret = 0;
2487         int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2488                             u64, u64, u64, u64, u64, u64);
2489 
2490         ref_root = btrfs_header_owner(buf);
2491         nritems = btrfs_header_nritems(buf);
2492         level = btrfs_header_level(buf);
2493 
2494         if (!root->ref_cows && level == 0)
2495                 return 0;
2496 
2497         if (inc)
2498                 process_func = btrfs_inc_extent_ref;
2499         else
2500                 process_func = btrfs_free_extent;
2501 
2502         if (full_backref)
2503                 parent = buf->start;
2504         else
2505                 parent = 0;
2506 
2507         for (i = 0; i < nritems; i++) {
2508                 if (level == 0) {
2509                         btrfs_item_key_to_cpu(buf, &key, i);
2510                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2511                                 continue;
2512                         fi = btrfs_item_ptr(buf, i,
2513                                             struct btrfs_file_extent_item);
2514                         if (btrfs_file_extent_type(buf, fi) ==
2515                             BTRFS_FILE_EXTENT_INLINE)
2516                                 continue;
2517                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2518                         if (bytenr == 0)
2519                                 continue;
2520 
2521                         num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2522                         key.offset -= btrfs_file_extent_offset(buf, fi);
2523                         ret = process_func(trans, root, bytenr, num_bytes,
2524                                            parent, ref_root, key.objectid,
2525                                            key.offset);
2526                         if (ret)
2527                                 goto fail;
2528                 } else {
2529                         bytenr = btrfs_node_blockptr(buf, i);
2530                         num_bytes = btrfs_level_size(root, level - 1);
2531                         ret = process_func(trans, root, bytenr, num_bytes,
2532                                            parent, ref_root, level - 1, 0);
2533                         if (ret)
2534                                 goto fail;
2535                 }
2536         }
2537         return 0;
2538 fail:
2539         BUG();
2540         return ret;
2541 }
2542 
2543 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2544                   struct extent_buffer *buf, int full_backref)
2545 {
2546         return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
2547 }
2548 
2549 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2550                   struct extent_buffer *buf, int full_backref)
2551 {
2552         return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
2553 }
2554 
2555 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2556                                  struct btrfs_root *root,
2557                                  struct btrfs_path *path,
2558                                  struct btrfs_block_group_cache *cache)
2559 {
2560         int ret;
2561         struct btrfs_root *extent_root = root->fs_info->extent_root;
2562         unsigned long bi;
2563         struct extent_buffer *leaf;
2564 
2565         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2566         if (ret < 0)
2567                 goto fail;
2568         BUG_ON(ret);
2569 
2570         leaf = path->nodes[0];
2571         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2572         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2573         btrfs_mark_buffer_dirty(leaf);
2574         btrfs_release_path(extent_root, path);
2575 fail:
2576         if (ret)
2577                 return ret;
2578         return 0;
2579 
2580 }
2581 
2582 static struct btrfs_block_group_cache *
2583 next_block_group(struct btrfs_root *root,
2584                  struct btrfs_block_group_cache *cache)
2585 {
2586         struct rb_node *node;
2587         spin_lock(&root->fs_info->block_group_cache_lock);
2588         node = rb_next(&cache->cache_node);
2589         btrfs_put_block_group(cache);
2590         if (node) {
2591                 cache = rb_entry(node, struct btrfs_block_group_cache,
2592                                  cache_node);
2593                 btrfs_get_block_group(cache);
2594         } else
2595                 cache = NULL;
2596         spin_unlock(&root->fs_info->block_group_cache_lock);
2597         return cache;
2598 }
2599 
2600 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
2601                                    struct btrfs_root *root)
2602 {
2603         struct btrfs_block_group_cache *cache;
2604         int err = 0;
2605         struct btrfs_path *path;
2606         u64 last = 0;
2607 
2608         path = btrfs_alloc_path();
2609         if (!path)
2610                 return -ENOMEM;
2611 
2612         while (1) {
2613                 if (last == 0) {
2614                         err = btrfs_run_delayed_refs(trans, root,
2615                                                      (unsigned long)-1);
2616                         BUG_ON(err);
2617                 }
2618 
2619                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2620                 while (cache) {
2621                         if (cache->dirty)
2622                                 break;
2623                         cache = next_block_group(root, cache);
2624                 }
2625                 if (!cache) {
2626                         if (last == 0)
2627                                 break;
2628                         last = 0;
2629                         continue;
2630                 }
2631 
2632                 cache->dirty = 0;
2633                 last = cache->key.objectid + cache->key.offset;
2634 
2635                 err = write_one_cache_group(trans, root, path, cache);
2636                 BUG_ON(err);
2637                 btrfs_put_block_group(cache);
2638         }
2639 
2640         btrfs_free_path(path);
2641         return 0;
2642 }
2643 
2644 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
2645 {
2646         struct btrfs_block_group_cache *block_group;
2647         int readonly = 0;
2648 
2649         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
2650         if (!block_group || block_group->ro)
2651                 readonly = 1;
2652         if (block_group)
2653                 btrfs_put_block_group(block_group);
2654         return readonly;
2655 }
2656 
2657 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
2658                              u64 total_bytes, u64 bytes_used,
2659                              struct btrfs_space_info **space_info)
2660 {
2661         struct btrfs_space_info *found;
2662 
2663         found = __find_space_info(info, flags);
2664         if (found) {
2665                 spin_lock(&found->lock);
2666                 found->total_bytes += total_bytes;
2667                 found->bytes_used += bytes_used;
2668                 found->full = 0;
2669                 spin_unlock(&found->lock);
2670                 *space_info = found;
2671                 return 0;
2672         }
2673         found = kzalloc(sizeof(*found), GFP_NOFS);
2674         if (!found)
2675                 return -ENOMEM;
2676 
2677         INIT_LIST_HEAD(&found->block_groups);
2678         init_rwsem(&found->groups_sem);
2679         spin_lock_init(&found->lock);
2680         found->flags = flags;
2681         found->total_bytes = total_bytes;
2682         found->bytes_used = bytes_used;
2683         found->bytes_pinned = 0;
2684         found->bytes_reserved = 0;
2685         found->bytes_readonly = 0;
2686         found->bytes_delalloc = 0;
2687         found->full = 0;
2688         found->force_alloc = 0;
2689         *space_info = found;
2690         list_add_rcu(&found->list, &info->space_info);
2691         atomic_set(&found->caching_threads, 0);
2692         return 0;
2693 }
2694 
2695 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
2696 {
2697         u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
2698                                    BTRFS_BLOCK_GROUP_RAID1 |
2699                                    BTRFS_BLOCK_GROUP_RAID10 |
2700                                    BTRFS_BLOCK_GROUP_DUP);
2701         if (extra_flags) {
2702                 if (flags & BTRFS_BLOCK_GROUP_DATA)
2703                         fs_info->avail_data_alloc_bits |= extra_flags;
2704                 if (flags & BTRFS_BLOCK_GROUP_METADATA)
2705                         fs_info->avail_metadata_alloc_bits |= extra_flags;
2706                 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
2707                         fs_info->avail_system_alloc_bits |= extra_flags;
2708         }
2709 }
2710 
2711 static void set_block_group_readonly(struct btrfs_block_group_cache *cache)
2712 {
2713         spin_lock(&cache->space_info->lock);
2714         spin_lock(&cache->lock);
2715         if (!cache->ro) {
2716                 cache->space_info->bytes_readonly += cache->key.offset -
2717                                         btrfs_block_group_used(&cache->item);
2718                 cache->ro = 1;
2719         }
2720         spin_unlock(&cache->lock);
2721         spin_unlock(&cache->space_info->lock);
2722 }
2723 
2724 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
2725 {
2726         u64 num_devices = root->fs_info->fs_devices->rw_devices;
2727 
2728         if (num_devices == 1)
2729                 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
2730         if (num_devices < 4)
2731                 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
2732 
2733         if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
2734             (flags & (BTRFS_BLOCK_GROUP_RAID1 |
2735                       BTRFS_BLOCK_GROUP_RAID10))) {
2736                 flags &= ~BTRFS_BLOCK_GROUP_DUP;
2737         }
2738 
2739         if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
2740             (flags & BTRFS_BLOCK_GROUP_RAID10)) {
2741                 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
2742         }
2743 
2744         if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
2745             ((flags & BTRFS_BLOCK_GROUP_RAID1) |
2746              (flags & BTRFS_BLOCK_GROUP_RAID10) |
2747              (flags & BTRFS_BLOCK_GROUP_DUP)))
2748                 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
2749         return flags;
2750 }
2751 
2752 static u64 btrfs_get_alloc_profile(struct btrfs_root *root, u64 data)
2753 {
2754         struct btrfs_fs_info *info = root->fs_info;
2755         u64 alloc_profile;
2756 
2757         if (data) {
2758                 alloc_profile = info->avail_data_alloc_bits &
2759                         info->data_alloc_profile;
2760                 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
2761         } else if (root == root->fs_info->chunk_root) {
2762                 alloc_profile = info->avail_system_alloc_bits &
2763                         info->system_alloc_profile;
2764                 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
2765         } else {
2766                 alloc_profile = info->avail_metadata_alloc_bits &
2767                         info->metadata_alloc_profile;
2768                 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
2769         }
2770 
2771         return btrfs_reduce_alloc_profile(root, data);
2772 }
2773 
2774 void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
2775 {
2776         u64 alloc_target;
2777 
2778         alloc_target = btrfs_get_alloc_profile(root, 1);
2779         BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
2780                                                        alloc_target);
2781 }
2782 
2783 static u64 calculate_bytes_needed(struct btrfs_root *root, int num_items)
2784 {
2785         u64 num_bytes;
2786         int level;
2787 
2788         level = BTRFS_MAX_LEVEL - 2;
2789         /*
2790          * NOTE: these calculations are absolutely the worst possible case.
2791          * This assumes that _every_ item we insert will require a new leaf, and
2792          * that the tree has grown to its maximum level size.
2793          */
2794 
2795         /*
2796          * for every item we insert we could insert both an extent item and a
2797          * extent ref item.  Then for ever item we insert, we will need to cow
2798          * both the original leaf, plus the leaf to the left and right of it.
2799          *
2800          * Unless we are talking about the extent root, then we just want the
2801          * number of items * 2, since we just need the extent item plus its ref.
2802          */
2803         if (root == root->fs_info->extent_root)
2804                 num_bytes = num_items * 2;
2805         else
2806                 num_bytes = (num_items + (2 * num_items)) * 3;
2807 
2808         /*
2809          * num_bytes is total number of leaves we could need times the leaf
2810          * size, and then for every leaf we could end up cow'ing 2 nodes per
2811          * level, down to the leaf level.
2812          */
2813         num_bytes = (num_bytes * root->leafsize) +
2814                 (num_bytes * (level * 2)) * root->nodesize;
2815 
2816         return num_bytes;
2817 }
2818 
2819 /*
2820  * Unreserve metadata space for delalloc.  If we have less reserved credits than
2821  * we have extents, this function does nothing.
2822  */
2823 int btrfs_unreserve_metadata_for_delalloc(struct btrfs_root *root,
2824                                           struct inode *inode, int num_items)
2825 {
2826         struct btrfs_fs_info *info = root->fs_info;
2827         struct btrfs_space_info *meta_sinfo;
2828         u64 num_bytes;
2829         u64 alloc_target;
2830         bool bug = false;
2831 
2832         /* get the space info for where the metadata will live */
2833         alloc_target = btrfs_get_alloc_profile(root, 0);
2834         meta_sinfo = __find_space_info(info, alloc_target);
2835 
2836         num_bytes = calculate_bytes_needed(root->fs_info->extent_root,
2837                                            num_items);
2838 
2839         spin_lock(&meta_sinfo->lock);
2840         spin_lock(&BTRFS_I(inode)->accounting_lock);
2841         if (BTRFS_I(inode)->reserved_extents <=
2842             BTRFS_I(inode)->outstanding_extents) {
2843                 spin_unlock(&BTRFS_I(inode)->accounting_lock);
2844                 spin_unlock(&meta_sinfo->lock);
2845                 return 0;
2846         }
2847         spin_unlock(&BTRFS_I(inode)->accounting_lock);
2848 
2849         BTRFS_I(inode)->reserved_extents--;
2850         BUG_ON(BTRFS_I(inode)->reserved_extents < 0);
2851 
2852         if (meta_sinfo->bytes_delalloc < num_bytes) {
2853                 bug = true;
2854                 meta_sinfo->bytes_delalloc = 0;
2855         } else {
2856                 meta_sinfo->bytes_delalloc -= num_bytes;
2857         }
2858         spin_unlock(&meta_sinfo->lock);
2859 
2860         BUG_ON(bug);
2861 
2862         return 0;
2863 }
2864 
2865 static void check_force_delalloc(struct btrfs_space_info *meta_sinfo)
2866 {
2867         u64 thresh;
2868 
2869         thresh = meta_sinfo->bytes_used + meta_sinfo->bytes_reserved +
2870                 meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly +
2871                 meta_sinfo->bytes_super + meta_sinfo->bytes_root +
2872                 meta_sinfo->bytes_may_use;
2873 
2874         thresh = meta_sinfo->total_bytes - thresh;
2875         thresh *= 80;
2876         do_div(thresh, 100);
2877         if (thresh <= meta_sinfo->bytes_delalloc)
2878                 meta_sinfo->force_delalloc = 1;
2879         else
2880                 meta_sinfo->force_delalloc = 0;
2881 }
2882 
2883 struct async_flush {
2884         struct btrfs_root *root;
2885         struct btrfs_space_info *info;
2886         struct btrfs_work work;
2887 };
2888 
2889 static noinline void flush_delalloc_async(struct btrfs_work *work)
2890 {
2891         struct async_flush *async;
2892         struct btrfs_root *root;
2893         struct btrfs_space_info *info;
2894 
2895         async = container_of(work, struct async_flush, work);
2896         root = async->root;
2897         info = async->info;
2898 
2899         btrfs_start_delalloc_inodes(root, 0);
2900         wake_up(&info->flush_wait);
2901         btrfs_wait_ordered_extents(root, 0, 0);
2902 
2903         spin_lock(&info->lock);
2904         info->flushing = 0;
2905         spin_unlock(&info->lock);
2906         wake_up(&info->flush_wait);
2907 
2908         kfree(async);
2909 }
2910 
2911 static void wait_on_flush(struct btrfs_space_info *info)
2912 {
2913         DEFINE_WAIT(wait);
2914         u64 used;
2915 
2916         while (1) {
2917                 prepare_to_wait(&info->flush_wait, &wait,
2918                                 TASK_UNINTERRUPTIBLE);
2919                 spin_lock(&info->lock);
2920                 if (!info->flushing) {
2921                         spin_unlock(&info->lock);
2922                         break;
2923                 }
2924 
2925                 used = info->bytes_used + info->bytes_reserved +
2926                         info->bytes_pinned + info->bytes_readonly +
2927                         info->bytes_super + info->bytes_root +
2928                         info->bytes_may_use + info->bytes_delalloc;
2929                 if (used < info->total_bytes) {
2930                         spin_unlock(&info->lock);
2931                         break;
2932                 }
2933                 spin_unlock(&info->lock);
2934                 schedule();
2935         }
2936         finish_wait(&info->flush_wait, &wait);
2937 }
2938 
2939 static void flush_delalloc(struct btrfs_root *root,
2940                                  struct btrfs_space_info *info)
2941 {
2942         struct async_flush *async;
2943         bool wait = false;
2944 
2945         spin_lock(&info->lock);
2946 
2947         if (!info->flushing) {
2948                 info->flushing = 1;
2949                 init_waitqueue_head(&info->flush_wait);
2950         } else {
2951                 wait = true;
2952         }
2953 
2954         spin_unlock(&info->lock);
2955 
2956         if (wait) {
2957                 wait_on_flush(info);
2958                 return;
2959         }
2960 
2961         async = kzalloc(sizeof(*async), GFP_NOFS);
2962         if (!async)
2963                 goto flush;
2964 
2965         async->root = root;
2966         async->info = info;
2967         async->work.func = flush_delalloc_async;
2968 
2969         btrfs_queue_worker(&root->fs_info->enospc_workers,
2970                            &async->work);
2971         wait_on_flush(info);
2972         return;
2973 
2974 flush:
2975         btrfs_start_delalloc_inodes(root, 0);
2976         btrfs_wait_ordered_extents(root, 0, 0);
2977 
2978         spin_lock(&info->lock);
2979         info->flushing = 0;
2980         spin_unlock(&info->lock);
2981         wake_up(&info->flush_wait);
2982 }
2983 
2984 static int maybe_allocate_chunk(struct btrfs_root *root,
2985                                  struct btrfs_space_info *info)
2986 {
2987         struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
2988         struct btrfs_trans_handle *trans;
2989         bool wait = false;
2990         int ret = 0;
2991         u64 min_metadata;
2992         u64 free_space;
2993 
2994         free_space = btrfs_super_total_bytes(disk_super);
2995         /*
2996          * we allow the metadata to grow to a max of either 10gb or 5% of the
2997          * space in the volume.
2998          */
2999         min_metadata = min((u64)10 * 1024 * 1024 * 1024,
3000                              div64_u64(free_space * 5, 100));
3001         if (info->total_bytes >= min_metadata) {
3002                 spin_unlock(&info->lock);
3003                 return 0;
3004         }
3005 
3006         if (info->full) {
3007                 spin_unlock(&info->lock);
3008                 return 0;
3009         }
3010 
3011         if (!info->allocating_chunk) {
3012                 info->force_alloc = 1;
3013                 info->allocating_chunk = 1;
3014                 init_waitqueue_head(&info->allocate_wait);
3015         } else {
3016                 wait = true;
3017         }
3018 
3019         spin_unlock(&info->lock);
3020 
3021         if (wait) {
3022                 wait_event(info->allocate_wait,
3023                            !info->allocating_chunk);
3024                 return 1;
3025         }
3026 
3027         trans = btrfs_start_transaction(root, 1);
3028         if (!trans) {
3029                 ret = -ENOMEM;
3030                 goto out;
3031         }
3032 
3033         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3034                              4096 + 2 * 1024 * 1024,
3035                              info->flags, 0);
3036         btrfs_end_transaction(trans, root);
3037         if (ret)
3038                 goto out;
3039 out:
3040         spin_lock(&info->lock);
3041         info->allocating_chunk = 0;
3042         spin_unlock(&info->lock);
3043         wake_up(&info->allocate_wait);
3044 
3045         if (ret)
3046                 return 0;
3047         return 1;
3048 }
3049 
3050 /*
3051  * Reserve metadata space for delalloc.
3052  */
3053 int btrfs_reserve_metadata_for_delalloc(struct btrfs_root *root,
3054                                         struct inode *inode, int num_items)
3055 {
3056         struct btrfs_fs_info *info = root->fs_info;
3057         struct btrfs_space_info *meta_sinfo;
3058         u64 num_bytes;
3059         u64 used;
3060         u64 alloc_target;
3061         int flushed = 0;
3062         int force_delalloc;
3063 
3064         /* get the space info for where the metadata will live */
3065         alloc_target = btrfs_get_alloc_profile(root, 0);
3066         meta_sinfo = __find_space_info(info, alloc_target);
3067 
3068         num_bytes = calculate_bytes_needed(root->fs_info->extent_root,
3069                                            num_items);
3070 again:
3071         spin_lock(&meta_sinfo->lock);
3072 
3073         force_delalloc = meta_sinfo->force_delalloc;
3074 
3075         if (unlikely(!meta_sinfo->bytes_root))
3076                 meta_sinfo->bytes_root = calculate_bytes_needed(root, 6);
3077 
3078         if (!flushed)
3079                 meta_sinfo->bytes_delalloc += num_bytes;
3080 
3081         used = meta_sinfo->bytes_used + meta_sinfo->bytes_reserved +
3082                 meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly +
3083                 meta_sinfo->bytes_super + meta_sinfo->bytes_root +
3084                 meta_sinfo->bytes_may_use + meta_sinfo->bytes_delalloc;
3085 
3086         if (used > meta_sinfo->total_bytes) {
3087                 flushed++;
3088 
3089                 if (flushed == 1) {
3090                         if (maybe_allocate_chunk(root, meta_sinfo))
3091                                 goto again;
3092                         flushed++;
3093                 } else {
3094                         spin_unlock(&meta_sinfo->lock);
3095                 }
3096 
3097                 if (flushed == 2) {
3098                         filemap_flush(inode->i_mapping);
3099                         goto again;
3100                 } else if (flushed == 3) {
3101                         flush_delalloc(root, meta_sinfo);
3102                         goto again;
3103                 }
3104                 spin_lock(&meta_sinfo->lock);
3105                 meta_sinfo->bytes_delalloc -= num_bytes;
3106                 spin_unlock(&meta_sinfo->lock);
3107                 printk(KERN_ERR "enospc, has %d, reserved %d\n",
3108                        BTRFS_I(inode)->outstanding_extents,
3109                        BTRFS_I(inode)->reserved_extents);
3110                 dump_space_info(meta_sinfo, 0, 0);
3111                 return -ENOSPC;
3112         }
3113 
3114         BTRFS_I(inode)->reserved_extents++;
3115         check_force_delalloc(meta_sinfo);
3116         spin_unlock(&meta_sinfo->lock);
3117 
3118         if (!flushed && force_delalloc)
3119                 filemap_flush(inode->i_mapping);
3120 
3121         return 0;
3122 }
3123 
3124 /*
3125  * unreserve num_items number of items worth of metadata space.  This needs to
3126  * be paired with btrfs_reserve_metadata_space.
3127  *
3128  * NOTE: if you have the option, run this _AFTER_ you do a
3129  * btrfs_end_transaction, since btrfs_end_transaction will run delayed ref
3130  * oprations which will result in more used metadata, so we want to make sure we
3131  * can do that without issue.
3132  */
3133 int btrfs_unreserve_metadata_space(struct btrfs_root *root, int num_items)
3134 {
3135         struct btrfs_fs_info *info = root->fs_info;
3136         struct btrfs_space_info *meta_sinfo;
3137         u64 num_bytes;
3138         u64 alloc_target;
3139         bool bug = false;
3140 
3141         /* get the space info for where the metadata will live */
3142         alloc_target = btrfs_get_alloc_profile(root, 0);
3143         meta_sinfo = __find_space_info(info, alloc_target);
3144 
3145         num_bytes = calculate_bytes_needed(root, num_items);
3146 
3147         spin_lock(&meta_sinfo->lock);
3148         if (meta_sinfo->bytes_may_use < num_bytes) {
3149                 bug = true;
3150                 meta_sinfo->bytes_may_use = 0;
3151         } else {
3152                 meta_sinfo->bytes_may_use -= num_bytes;
3153         }
3154         spin_unlock(&meta_sinfo->lock);
3155 
3156         BUG_ON(bug);
3157 
3158         return 0;
3159 }
3160 
3161 /*
3162  * Reserve some metadata space for use.  We'll calculate the worste case number
3163  * of bytes that would be needed to modify num_items number of items.  If we
3164  * have space, fantastic, if not, you get -ENOSPC.  Please call
3165  * btrfs_unreserve_metadata_space when you are done for the _SAME_ number of
3166  * items you reserved, since whatever metadata you needed should have already
3167  * been allocated.
3168  *
3169  * This will commit the transaction to make more space if we don't have enough
3170  * metadata space.  THe only time we don't do this is if we're reserving space
3171  * inside of a transaction, then we will just return -ENOSPC and it is the
3172  * callers responsibility to handle it properly.
3173  */
3174 int btrfs_reserve_metadata_space(struct btrfs_root *root, int num_items)
3175 {
3176         struct btrfs_fs_info *info = root->fs_info;
3177         struct btrfs_space_info *meta_sinfo;
3178         u64 num_bytes;
3179         u64 used;
3180         u64 alloc_target;
3181         int retries = 0;
3182 
3183         /* get the space info for where the metadata will live */
3184         alloc_target = btrfs_get_alloc_profile(root, 0);
3185         meta_sinfo = __find_space_info(info, alloc_target);
3186 
3187         num_bytes = calculate_bytes_needed(root, num_items);
3188 again:
3189         spin_lock(&meta_sinfo->lock);
3190 
3191         if (unlikely(!meta_sinfo->bytes_root))
3192                 meta_sinfo->bytes_root = calculate_bytes_needed(root, 6);
3193 
3194         if (!retries)
3195                 meta_sinfo->bytes_may_use += num_bytes;
3196 
3197         used = meta_sinfo->bytes_used + meta_sinfo->bytes_reserved +
3198                 meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly +
3199                 meta_sinfo->bytes_super + meta_sinfo->bytes_root +
3200                 meta_sinfo->bytes_may_use + meta_sinfo->bytes_delalloc;
3201 
3202         if (used > meta_sinfo->total_bytes) {
3203                 retries++;
3204                 if (retries == 1) {
3205                         if (maybe_allocate_chunk(root, meta_sinfo))
3206                                 goto again;
3207                         retries++;
3208                 } else {
3209                         spin_unlock(&meta_sinfo->lock);
3210                 }
3211 
3212                 if (retries == 2) {
3213                         flush_delalloc(root, meta_sinfo);
3214                         goto again;
3215                 }
3216                 spin_lock(&meta_sinfo->lock);
3217                 meta_sinfo->bytes_may_use -= num_bytes;
3218                 spin_unlock(&meta_sinfo->lock);
3219 
3220                 dump_space_info(meta_sinfo, 0, 0);
3221                 return -ENOSPC;
3222         }
3223 
3224         check_force_delalloc(meta_sinfo);
3225         spin_unlock(&meta_sinfo->lock);
3226 
3227         return 0;
3228 }
3229 
3230 /*
3231  * This will check the space that the inode allocates from to make sure we have
3232  * enough space for bytes.
3233  */
3234 int btrfs_check_data_free_space(struct btrfs_root *root, struct inode *inode,
3235                                 u64 bytes)
3236 {
3237         struct btrfs_space_info *data_sinfo;
3238         int ret = 0, committed = 0;
3239 
3240         /* make sure bytes are sectorsize aligned */
3241         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3242 
3243         data_sinfo = BTRFS_I(inode)->space_info;
3244         if (!data_sinfo)
3245                 goto alloc;
3246 
3247 again:
3248         /* make sure we have enough space to handle the data first */
3249         spin_lock(&data_sinfo->lock);
3250         if (data_sinfo->total_bytes - data_sinfo->bytes_used -
3251             data_sinfo->bytes_delalloc - data_sinfo->bytes_reserved -
3252             data_sinfo->bytes_pinned - data_sinfo->bytes_readonly -
3253             data_sinfo->bytes_may_use - data_sinfo->bytes_super < bytes) {
3254                 struct btrfs_trans_handle *trans;
3255 
3256                 /*
3257                  * if we don't have enough free bytes in this space then we need
3258                  * to alloc a new chunk.
3259                  */
3260                 if (!data_sinfo->full) {
3261                         u64 alloc_target;
3262 
3263                         data_sinfo->force_alloc = 1;
3264                         spin_unlock(&data_sinfo->lock);
3265 alloc:
3266                         alloc_target = btrfs_get_alloc_profile(root, 1);
3267                         trans = btrfs_start_transaction(root, 1);
3268                         if (!trans)
3269                                 return -ENOMEM;
3270 
3271                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3272                                              bytes + 2 * 1024 * 1024,
3273                                              alloc_target, 0);
3274                         btrfs_end_transaction(trans, root);
3275                         if (ret)
3276                                 return ret;
3277 
3278                         if (!data_sinfo) {
3279                                 btrfs_set_inode_space_info(root, inode);
3280                                 data_sinfo = BTRFS_I(inode)->space_info;
3281                         }
3282                         goto again;
3283                 }
3284                 spin_unlock(&data_sinfo->lock);
3285 
3286                 /* commit the current transaction and try again */
3287                 if (!committed && !root->fs_info->open_ioctl_trans) {
3288                         committed = 1;
3289                         trans = btrfs_join_transaction(root, 1);
3290                         if (!trans)
3291                                 return -ENOMEM;
3292                         ret = btrfs_commit_transaction(trans, root);
3293                         if (ret)
3294                                 return ret;
3295                         goto again;
3296                 }
3297 
3298                 printk(KERN_ERR "no space left, need %llu, %llu delalloc bytes"
3299                        ", %llu bytes_used, %llu bytes_reserved, "
3300                        "%llu bytes_pinned, %llu bytes_readonly, %llu may use "
3301                        "%llu total\n", (unsigned long long)bytes,
3302                        (unsigned long long)data_sinfo->bytes_delalloc,
3303                        (unsigned long long)data_sinfo->bytes_used,
3304                        (unsigned long long)data_sinfo->bytes_reserved,
3305                        (unsigned long long)data_sinfo->bytes_pinned,
3306                        (unsigned long long)data_sinfo->bytes_readonly,
3307                        (unsigned long long)data_sinfo->bytes_may_use,
3308                        (unsigned long long)data_sinfo->total_bytes);
3309                 return -ENOSPC;
3310         }
3311         data_sinfo->bytes_may_use += bytes;
3312         BTRFS_I(inode)->reserved_bytes += bytes;
3313         spin_unlock(&data_sinfo->lock);
3314 
3315         return 0;
3316 }
3317 
3318 /*
3319  * if there was an error for whatever reason after calling
3320  * btrfs_check_data_free_space, call this so we can cleanup the counters.
3321  */
3322 void btrfs_free_reserved_data_space(struct btrfs_root *root,
3323                                     struct inode *inode, u64 bytes)
3324 {
3325         struct btrfs_space_info *data_sinfo;
3326 
3327         /* make sure bytes are sectorsize aligned */
3328         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3329 
3330         data_sinfo = BTRFS_I(inode)->space_info;
3331         spin_lock(&data_sinfo->lock);
3332         data_sinfo->bytes_may_use -= bytes;
3333         BTRFS_I(inode)->reserved_bytes -= bytes;
3334         spin_unlock(&data_sinfo->lock);
3335 }
3336 
3337 /* called when we are adding a delalloc extent to the inode's io_tree */
3338 void btrfs_delalloc_reserve_space(struct btrfs_root *root, struct inode *inode,
3339                                   u64 bytes)
3340 {
3341         struct btrfs_space_info *data_sinfo;
3342 
3343         /* get the space info for where this inode will be storing its data */
3344         data_sinfo = BTRFS_I(inode)->space_info;
3345 
3346         /* make sure we have enough space to handle the data first */
3347         spin_lock(&data_sinfo->lock);
3348         data_sinfo->bytes_delalloc += bytes;
3349 
3350         /*
3351          * we are adding a delalloc extent without calling
3352          * btrfs_check_data_free_space first.  This happens on a weird
3353          * writepage condition, but shouldn't hurt our accounting
3354          */
3355         if (unlikely(bytes > BTRFS_I(inode)->reserved_bytes)) {
3356                 data_sinfo->bytes_may_use -= BTRFS_I(inode)->reserved_bytes;
3357                 BTRFS_I(inode)->reserved_bytes = 0;
3358         } else {
3359                 data_sinfo->bytes_may_use -= bytes;
3360                 BTRFS_I(inode)->reserved_bytes -= bytes;
3361         }
3362 
3363         spin_unlock(&data_sinfo->lock);
3364 }
3365 
3366 /* called when we are clearing an delalloc extent from the inode's io_tree */
3367 void btrfs_delalloc_free_space(struct btrfs_root *root, struct inode *inode,
3368                               u64 bytes)
3369 {
3370         struct btrfs_space_info *info;
3371 
3372         info = BTRFS_I(inode)->space_info;
3373 
3374         spin_lock(&info->lock);
3375         info->bytes_delalloc -= bytes;
3376         spin_unlock(&info->lock);
3377 }
3378 
3379 static void force_metadata_allocation(struct btrfs_fs_info *info)
3380 {
3381         struct list_head *head = &info->space_info;
3382         struct btrfs_space_info *found;
3383 
3384         rcu_read_lock();
3385         list_for_each_entry_rcu(found, head, list) {
3386                 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3387                         found->force_alloc = 1;
3388         }
3389         rcu_read_unlock();
3390 }
3391 
3392 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3393                           struct btrfs_root *extent_root, u64 alloc_bytes,
3394                           u64 flags, int force)
3395 {
3396         struct btrfs_space_info *space_info;
3397         struct btrfs_fs_info *fs_info = extent_root->fs_info;
3398         u64 thresh;
3399         int ret = 0;
3400 
3401         mutex_lock(&fs_info->chunk_mutex);
3402 
3403         flags = btrfs_reduce_alloc_profile(extent_root, flags);
3404 
3405         space_info = __find_space_info(extent_root->fs_info, flags);
3406         if (!space_info) {
3407                 ret = update_space_info(extent_root->fs_info, flags,
3408                                         0, 0, &space_info);
3409                 BUG_ON(ret);
3410         }
3411         BUG_ON(!space_info);
3412 
3413         spin_lock(&space_info->lock);
3414         if (space_info->force_alloc)
3415                 force = 1;
3416         if (space_info->full) {
3417                 spin_unlock(&space_info->lock);
3418                 goto out;
3419         }
3420 
3421         thresh = space_info->total_bytes - space_info->bytes_readonly;
3422         thresh = div_factor(thresh, 8);
3423         if (!force &&
3424            (space_info->bytes_used + space_info->bytes_pinned +
3425             space_info->bytes_reserved + alloc_bytes) < thresh) {
3426                 spin_unlock(&space_info->lock);
3427                 goto out;
3428         }
3429         spin_unlock(&space_info->lock);
3430 
3431         /*
3432          * if we're doing a data chunk, go ahead and make sure that
3433          * we keep a reasonable number of metadata chunks allocated in the
3434          * FS as well.
3435          */
3436         if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3437                 fs_info->data_chunk_allocations++;
3438                 if (!(fs_info->data_chunk_allocations %
3439                       fs_info->metadata_ratio))
3440                         force_metadata_allocation(fs_info);
3441         }
3442 
3443         ret = btrfs_alloc_chunk(trans, extent_root, flags);
3444         spin_lock(&space_info->lock);
3445         if (ret)
3446                 space_info->full = 1;
3447         space_info->force_alloc = 0;
3448         spin_unlock(&space_info->lock);
3449 out:
3450         mutex_unlock(&extent_root->fs_info->chunk_mutex);
3451         return ret;
3452 }
3453 
3454 static int update_block_group(struct btrfs_trans_handle *trans,
3455                               struct btrfs_root *root,
3456                               u64 bytenr, u64 num_bytes, int alloc,
3457                               int mark_free)
3458 {
3459         struct btrfs_block_group_cache *cache;
3460         struct btrfs_fs_info *info = root->fs_info;
3461         u64 total = num_bytes;
3462         u64 old_val;
3463         u64 byte_in_group;
3464 
3465         /* block accounting for super block */
3466         spin_lock(&info->delalloc_lock);
3467         old_val = btrfs_super_bytes_used(&info->super_copy);
3468         if (alloc)
3469                 old_val += num_bytes;
3470         else
3471                 old_val -= num_bytes;
3472         btrfs_set_super_bytes_used(&info->super_copy, old_val);
3473         spin_unlock(&info->delalloc_lock);
3474 
3475         while (total) {
3476                 cache = btrfs_lookup_block_group(info, bytenr);
3477                 if (!cache)
3478                         return -1;
3479                 byte_in_group = bytenr - cache->key.objectid;
3480                 WARN_ON(byte_in_group > cache->key.offset);
3481 
3482                 spin_lock(&cache->space_info->lock);
3483                 spin_lock(&cache->lock);
3484                 cache->dirty = 1;
3485                 old_val = btrfs_block_group_used(&cache->item);
3486                 num_bytes = min(total, cache->key.offset - byte_in_group);
3487                 if (alloc) {
3488                         old_val += num_bytes;
3489                         btrfs_set_block_group_used(&cache->item, old_val);
3490                         cache->reserved -= num_bytes;
3491                         cache->space_info->bytes_used += num_bytes;
3492                         cache->space_info->bytes_reserved -= num_bytes;
3493                         if (cache->ro)
3494                                 cache->space_info->bytes_readonly -= num_bytes;
3495                         spin_unlock(&cache->lock);
3496                         spin_unlock(&cache->space_info->lock);
3497                 } else {
3498                         old_val -= num_bytes;
3499                         cache->space_info->bytes_used -= num_bytes;
3500                         if (cache->ro)
3501                                 cache->space_info->bytes_readonly += num_bytes;
3502                         btrfs_set_block_group_used(&cache->item, old_val);
3503                         spin_unlock(&cache->lock);
3504                         spin_unlock(&cache->space_info->lock);
3505                         if (mark_free) {
3506                                 int ret;
3507 
3508                                 ret = btrfs_discard_extent(root, bytenr,
3509                                                            num_bytes);
3510                                 WARN_ON(ret);
3511 
3512                                 ret = btrfs_add_free_space(cache, bytenr,
3513                                                            num_bytes);
3514                                 WARN_ON(ret);
3515                         }
3516                 }
3517                 btrfs_put_block_group(cache);
3518                 total -= num_bytes;
3519                 bytenr += num_bytes;
3520         }
3521         return 0;
3522 }
3523 
3524 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
3525 {
3526         struct btrfs_block_group_cache *cache;
3527         u64 bytenr;
3528 
3529         cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
3530         if (!cache)
3531                 return 0;
3532 
3533         bytenr = cache->key.objectid;
3534         btrfs_put_block_group(cache);
3535 
3536         return bytenr;
3537 }
3538 
3539 /*
3540  * this function must be called within transaction
3541  */
3542 int btrfs_pin_extent(struct btrfs_root *root,
3543                      u64 bytenr, u64 num_bytes, int reserved)
3544 {
3545         struct btrfs_fs_info *fs_info = root->fs_info;
3546         struct btrfs_block_group_cache *cache;
3547 
3548         cache = btrfs_lookup_block_group(fs_info, bytenr);
3549         BUG_ON(!cache);
3550 
3551         spin_lock(&cache->space_info->lock);
3552         spin_lock(&cache->lock);
3553         cache->pinned += num_bytes;
3554         cache->space_info->bytes_pinned += num_bytes;
3555         if (reserved) {
3556                 cache->reserved -= num_bytes;
3557                 cache->space_info->bytes_reserved -= num_bytes;
3558         }
3559         spin_unlock(&cache->lock);
3560         spin_unlock(&cache->space_info->lock);
3561 
3562         btrfs_put_block_group(cache);
3563 
3564         set_extent_dirty(fs_info->pinned_extents,
3565                          bytenr, bytenr + num_bytes - 1, GFP_NOFS);
3566         return 0;
3567 }
3568 
3569 static int update_reserved_extents(struct btrfs_block_group_cache *cache,
3570                                    u64 num_bytes, int reserve)
3571 {
3572         spin_lock(&cache->space_info->lock);
3573         spin_lock(&cache->lock);
3574         if (reserve) {
3575                 cache->reserved += num_bytes;
3576                 cache->space_info->bytes_reserved += num_bytes;
3577         } else {
3578                 cache->reserved -= num_bytes;
3579                 cache->space_info->bytes_reserved -= num_bytes;
3580         }
3581         spin_unlock(&cache->lock);
3582         spin_unlock(&cache->space_info->lock);
3583         return 0;
3584 }
3585 
3586 int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
3587                                 struct btrfs_root *root)
3588 {
3589         struct btrfs_fs_info *fs_info = root->fs_info;
3590         struct btrfs_caching_control *next;
3591         struct btrfs_caching_control *caching_ctl;
3592         struct btrfs_block_group_cache *cache;
3593 
3594         down_write(&fs_info->extent_commit_sem);
3595 
3596         list_for_each_entry_safe(caching_ctl, next,
3597                                  &fs_info->caching_block_groups, list) {
3598                 cache = caching_ctl->block_group;
3599                 if (block_group_cache_done(cache)) {
3600                         cache->last_byte_to_unpin = (u64)-1;
3601                         list_del_init(&caching_ctl->list);
3602                         put_caching_control(caching_ctl);
3603                 } else {
3604                         cache->last_byte_to_unpin = caching_ctl->progress;
3605                 }
3606         }
3607 
3608         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
3609                 fs_info->pinned_extents = &fs_info->freed_extents[1];
3610         else
3611                 fs_info->pinned_extents = &fs_info->freed_extents[0];
3612 
3613         up_write(&fs_info->extent_commit_sem);
3614         return 0;
3615 }
3616 
3617 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
3618 {
3619         struct btrfs_fs_info *fs_info = root->fs_info;
3620         struct btrfs_block_group_cache *cache = NULL;
3621         u64 len;
3622 
3623         while (start <= end) {
3624                 if (!cache ||
3625                     start >= cache->key.objectid + cache->key.offset) {
3626                         if (cache)
3627                                 btrfs_put_block_group(cache);
3628                         cache = btrfs_lookup_block_group(fs_info, start);
3629                         BUG_ON(!cache);
3630                 }
3631 
3632                 len = cache->key.objectid + cache->key.offset - start;
3633                 len = min(len, end + 1 - start);
3634 
3635                 if (start < cache->last_byte_to_unpin) {
3636                         len = min(len, cache->last_byte_to_unpin - start);
3637                         btrfs_add_free_space(cache, start, len);
3638                 }
3639 
3640                 spin_lock(&cache->space_info->lock);
3641                 spin_lock(&cache->lock);
3642                 cache->pinned -= len;
3643                 cache->space_info->bytes_pinned -= len;
3644                 spin_unlock(&cache->lock);
3645                 spin_unlock(&cache->space_info->lock);
3646 
3647                 start += len;
3648         }
3649 
3650         if (cache)
3651                 btrfs_put_block_group(cache);
3652         return 0;
3653 }
3654 
3655 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
3656                                struct btrfs_root *root)
3657 {
3658         struct btrfs_fs_info *fs_info = root->fs_info;
3659         struct extent_io_tree *unpin;
3660         u64 start;
3661         u64 end;
3662         int ret;
3663 
3664         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
3665                 unpin = &fs_info->freed_extents[1];
3666         else
3667                 unpin = &fs_info->freed_extents[0];
3668 
3669         while (1) {
3670                 ret = find_first_extent_bit(unpin, 0, &start, &end,
3671                                             EXTENT_DIRTY);
3672                 if (ret)
3673                         break;
3674 
3675                 ret = btrfs_discard_extent(root, start, end + 1 - start);
3676 
3677                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3678                 unpin_extent_range(root, start, end);
3679                 cond_resched();
3680         }
3681 
3682         return ret;
3683 }
3684 
3685 static int pin_down_bytes(struct btrfs_trans_handle *trans,
3686                           struct btrfs_root *root,
3687                           struct btrfs_path *path,
3688                           u64 bytenr, u64 num_bytes,
3689                           int is_data, int reserved,
3690                           struct extent_buffer **must_clean)
3691 {
3692         int err = 0;
3693         struct extent_buffer *buf;
3694 
3695         if (is_data)
3696                 goto pinit;
3697 
3698         /*
3699          * discard is sloooow, and so triggering discards on
3700          * individual btree blocks isn't a good plan.  Just
3701          * pin everything in discard mode.
3702          */
3703         if (btrfs_test_opt(root, DISCARD))
3704                 goto pinit;
3705 
3706         buf = btrfs_find_tree_block(root, bytenr, num_bytes);
3707         if (!buf)
3708                 goto pinit;
3709 
3710         /* we can reuse a block if it hasn't been written
3711          * and it is from this transaction.  We can't
3712          * reuse anything from the tree log root because
3713          * it has tiny sub-transactions.
3714          */
3715         if (btrfs_buffer_uptodate(buf, 0) &&
3716             btrfs_try_tree_lock(buf)) {
3717                 u64 header_owner = btrfs_header_owner(buf);
3718                 u64 header_transid = btrfs_header_generation(buf);
3719                 if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
3720                     header_transid == trans->transid &&
3721                     !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
3722                         *must_clean = buf;
3723                         return 1;
3724                 }
3725                 btrfs_tree_unlock(buf);
3726         }
3727         free_extent_buffer(buf);
3728 pinit:
3729         if (path)
3730                 btrfs_set_path_blocking(path);
3731         /* unlocks the pinned mutex */
3732         btrfs_pin_extent(root, bytenr, num_bytes, reserved);
3733 
3734         BUG_ON(err < 0);
3735         return 0;
3736 }
3737 
3738 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
3739                                 struct btrfs_root *root,
3740                                 u64 bytenr, u64 num_bytes, u64 parent,
3741                                 u64 root_objectid, u64 owner_objectid,
3742                                 u64 owner_offset, int refs_to_drop,
3743                                 struct btrfs_delayed_extent_op *extent_op)
3744 {
3745         struct btrfs_key key;
3746         struct btrfs_path *path;
3747         struct btrfs_fs_info *info = root->fs_info;
3748         struct btrfs_root *extent_root = info->extent_root;
3749         struct extent_buffer *leaf;
3750         struct btrfs_extent_item *ei;
3751         struct btrfs_extent_inline_ref *iref;
3752         int ret;
3753         int is_data;
3754         int extent_slot = 0;
3755         int found_extent = 0;
3756         int num_to_del = 1;
3757         u32 item_size;
3758         u64 refs;
3759 
3760         path = btrfs_alloc_path();
3761         if (!path)
3762                 return -ENOMEM;
3763 
3764         path->reada = 1;
3765         path->leave_spinning = 1;
3766 
3767         is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
3768         BUG_ON(!is_data && refs_to_drop != 1);
3769 
3770         ret = lookup_extent_backref(trans, extent_root, path, &iref,
3771                                     bytenr, num_bytes, parent,
3772                                     root_objectid, owner_objectid,
3773                                     owner_offset);
3774         if (ret == 0) {
3775                 extent_slot = path->slots[0];
3776                 while (extent_slot >= 0) {
3777                         btrfs_item_key_to_cpu(path->nodes[0], &key,
3778                                               extent_slot);
3779                         if (key.objectid != bytenr)
3780                                 break;
3781                         if (key.type == BTRFS_EXTENT_ITEM_KEY &&
3782                             key.offset == num_bytes) {
3783                                 found_extent = 1;
3784                                 break;
3785                         }
3786                         if (path->slots[0] - extent_slot > 5)
3787                                 break;
3788                         extent_slot--;
3789                 }
3790 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3791                 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
3792                 if (found_extent && item_size < sizeof(*ei))
3793                         found_extent = 0;
3794 #endif
3795                 if (!found_extent) {
3796                         BUG_ON(iref);
3797                         ret = remove_extent_backref(trans, extent_root, path,
3798                                                     NULL, refs_to_drop,
3799                                                     is_data);
3800                         BUG_ON(ret);
3801                         btrfs_release_path(extent_root, path);
3802                         path->leave_spinning = 1;
3803 
3804                         key.objectid = bytenr;
3805                         key.type = BTRFS_EXTENT_ITEM_KEY;
3806                         key.offset = num_bytes;
3807 
3808                         ret = btrfs_search_slot(trans, extent_root,
3809                                                 &key, path, -1, 1);
3810                         if (ret) {
3811                                 printk(KERN_ERR "umm, got %d back from search"
3812                                        ", was looking for %llu\n", ret,
3813                                        (unsigned long long)bytenr);
3814                                 btrfs_print_leaf(extent_root, path->nodes[0]);
3815                         }
3816                         BUG_ON(ret);
3817                         extent_slot = path->slots[0];
3818                 }
3819         } else {
3820                 btrfs_print_leaf(extent_root, path->nodes[0]);
3821                 WARN_ON(1);
3822                 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
3823                        "parent %llu root %llu  owner %llu offset %llu\n",
3824                        (unsigned long long)bytenr,
3825                        (unsigned long long)parent,
3826                        (unsigned long long)root_objectid,
3827                        (unsigned long long)owner_objectid,
3828                        (unsigned long long)owner_offset);
3829         }
3830 
3831         leaf = path->nodes[0];
3832         item_size = btrfs_item_size_nr(leaf, extent_slot);
3833 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3834         if (item_size < sizeof(*ei)) {
3835                 BUG_ON(found_extent || extent_slot != path->slots[0]);
3836                 ret = convert_extent_item_v0(trans, extent_root, path,
3837                                              owner_objectid, 0);
3838                 BUG_ON(ret < 0);
3839 
3840                 btrfs_release_path(extent_root, path);
3841                 path->leave_spinning = 1;
3842 
3843                 key.objectid = bytenr;
3844                 key.type = BTRFS_EXTENT_ITEM_KEY;
3845                 key.offset = num_bytes;
3846 
3847                 ret = btrfs_search_slot(trans, extent_root, &key, path,
3848                                         -1, 1);
3849                 if (ret) {
3850                         printk(KERN_ERR "umm, got %d back from search"
3851                                ", was looking for %llu\n", ret,
3852                                (unsigned long long)bytenr);
3853                         btrfs_print_leaf(extent_root, path->nodes[0]);
3854                 }
3855                 BUG_ON(ret);
3856                 extent_slot = path->slots[0];
3857                 leaf = path->nodes[0];
3858                 item_size = btrfs_item_size_nr(leaf, extent_slot);
3859         }
3860 #endif
3861         BUG_ON(item_size < sizeof(*ei));
3862         ei = btrfs_item_ptr(leaf, extent_slot,
3863                             struct btrfs_extent_item);
3864         if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
3865                 struct btrfs_tree_block_info *bi;
3866                 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
3867                 bi = (struct btrfs_tree_block_info *)(ei + 1);
3868                 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
3869         }
3870 
3871         refs = btrfs_extent_refs(leaf, ei);
3872         BUG_ON(refs < refs_to_drop);
3873         refs -= refs_to_drop;
3874 
3875         if (refs > 0) {
3876                 if (extent_op)
3877                         __run_delayed_extent_op(extent_op, leaf, ei);
3878                 /*
3879                  * In the case of inline back ref, reference count will
3880                  * be updated by remove_extent_backref
3881                  */
3882                 if (iref) {
3883                         BUG_ON(!found_extent);
3884                 } else {
3885                         btrfs_set_extent_refs(leaf, ei, refs);
3886                         btrfs_mark_buffer_dirty(leaf);
3887                 }
3888                 if (found_extent) {
3889                         ret = remove_extent_backref(trans, extent_root, path,
3890                                                     iref, refs_to_drop,
3891                                                     is_data);
3892                         BUG_ON(ret);
3893                 }
3894         } else {
3895                 int mark_free = 0;
3896                 struct extent_buffer *must_clean = NULL;
3897 
3898                 if (found_extent) {
3899                         BUG_ON(is_data && refs_to_drop !=
3900                                extent_data_ref_count(root, path, iref));
3901                         if (iref) {
3902                                 BUG_ON(path->slots[0] != extent_slot);
3903                         } else {
3904                                 BUG_ON(path->slots[0] != extent_slot + 1);
3905                                 path->slots[0] = extent_slot;
3906                                 num_to_del = 2;
3907                         }
3908                 }
3909 
3910                 ret = pin_down_bytes(trans, root, path, bytenr,
3911                                      num_bytes, is_data, 0, &must_clean);
3912                 if (ret > 0)
3913                         mark_free = 1;
3914                 BUG_ON(ret < 0);
3915                 /*
3916                  * it is going to be very rare for someone to be waiting
3917                  * on the block we're freeing.  del_items might need to
3918                  * schedule, so rather than get fancy, just force it
3919                  * to blocking here
3920                  */
3921                 if (must_clean)
3922                         btrfs_set_lock_blocking(must_clean);
3923 
3924                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
3925                                       num_to_del);
3926                 BUG_ON(ret);
3927                 btrfs_release_path(extent_root, path);
3928 
3929                 if (must_clean) {
3930                         clean_tree_block(NULL, root, must_clean);
3931                         btrfs_tree_unlock(must_clean);
3932                         free_extent_buffer(must_clean);
3933                 }
3934 
3935                 if (is_data) {
3936                         ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
3937                         BUG_ON(ret);
3938                 } else {
3939                         invalidate_mapping_pages(info->btree_inode->i_mapping,
3940                              bytenr >> PAGE_CACHE_SHIFT,
3941                              (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
3942                 }
3943 
3944                 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
3945                                          mark_free);
3946                 BUG_ON(ret);
3947         }
3948         btrfs_free_path(path);
3949         return ret;
3950 }
3951 
3952 /*
3953  * when we free an extent, it is possible (and likely) that we free the last
3954  * delayed ref for that extent as well.  This searches the delayed ref tree for
3955  * a given extent, and if there are no other delayed refs to be processed, it
3956  * removes it from the tree.
3957  */
3958 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
3959                                       struct btrfs_root *root, u64 bytenr)
3960 {
3961         struct btrfs_delayed_ref_head *head;
3962         struct btrfs_delayed_ref_root *delayed_refs;
3963         struct btrfs_delayed_ref_node *ref;
3964         struct rb_node *node;
3965         int ret;
3966 
3967         delayed_refs = &trans->transaction->delayed_refs;
3968         spin_lock(&delayed_refs->lock);
3969         head = btrfs_find_delayed_ref_head(trans, bytenr);
3970         if (!head)
3971                 goto out;
3972 
3973         node = rb_prev(&head->node.rb_node);
3974         if (!node)
3975                 goto out;
3976 
3977         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
3978 
3979         /* there are still entries for this ref, we can't drop it */
3980         if (ref->bytenr == bytenr)
3981                 goto out;
3982 
3983         if (head->extent_op) {
3984                 if (!head->must_insert_reserved)
3985                         goto out;
3986                 kfree(head->extent_op);
3987                 head->extent_op = NULL;
3988         }
3989 
3990         /*
3991          * waiting for the lock here would deadlock.  If someone else has it
3992          * locked they are already in the process of dropping it anyway
3993          */
3994         if (!mutex_trylock(&head->mutex))
3995                 goto out;
3996 
3997         /*
3998          * at this point we have a head with no other entries.  Go
3999          * ahead and process it.
4000          */
4001         head->node.in_tree = 0;
4002         rb_erase(&head->node.rb_node, &delayed_refs->root);
4003 
4004         delayed_refs->num_entries--;
4005 
4006         /*
4007          * we don't take a ref on the node because we're removing it from the
4008          * tree, so we just steal the ref the tree was holding.
4009          */
4010         delayed_refs->num_heads--;
4011         if (list_empty(&head->cluster))
4012                 delayed_refs->num_heads_ready--;
4013 
4014         list_del_init(&head->cluster);
4015         spin_unlock(&delayed_refs->lock);
4016 
4017         ret = run_one_delayed_ref(trans, root->fs_info->tree_root,
4018                                   &head->node, head->extent_op,
4019                                   head->must_insert_reserved);
4020         BUG_ON(ret);
4021         btrfs_put_delayed_ref(&head->node);
4022         return 0;
4023 out:
4024         spin_unlock(&delayed_refs->lock);
4025         return 0;
4026 }
4027 
4028 int btrfs_free_extent(struct btrfs_trans_handle *trans,
4029                       struct btrfs_root *root,
4030                       u64 bytenr, u64 num_bytes, u64 parent,
4031                       u64 root_objectid, u64 owner, u64 offset)
4032 {
4033         int ret;
4034 
4035         /*
4036          * tree log blocks never actually go into the extent allocation
4037          * tree, just update pinning info and exit early.
4038          */
4039         if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
4040                 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
4041                 /* unlocks the pinned mutex */
4042                 btrfs_pin_extent(root, bytenr, num_bytes, 1);
4043                 ret = 0;
4044         } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
4045                 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
4046                                         parent, root_objectid, (int)owner,
4047                                         BTRFS_DROP_DELAYED_REF, NULL);
4048                 BUG_ON(ret);
4049                 ret = check_ref_cleanup(trans, root, bytenr);
4050                 BUG_ON(ret);
4051         } else {
4052                 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
4053                                         parent, root_objectid, owner,
4054                                         offset, BTRFS_DROP_DELAYED_REF, NULL);
4055                 BUG_ON(ret);
4056         }
4057         return ret;
4058 }
4059 
4060 int btrfs_free_tree_block(struct btrfs_trans_handle *trans,
4061                           struct btrfs_root *root,
4062                           u64 bytenr, u32 blocksize,
4063                           u64 parent, u64 root_objectid, int level)
4064 {
4065         u64 used;
4066         spin_lock(&root->node_lock);
4067         used = btrfs_root_used(&root->root_item) - blocksize;
4068         btrfs_set_root_used(&root->root_item, used);
4069         spin_unlock(&root->node_lock);
4070 
4071         return btrfs_free_extent(trans, root, bytenr, blocksize,
4072                                  parent, root_objectid, level, 0);
4073 }
4074 
4075 static u64 stripe_align(struct btrfs_root *root, u64 val)
4076 {
4077         u64 mask = ((u64)root->stripesize - 1);
4078         u64 ret = (val + mask) & ~mask;
4079         return ret;
4080 }
4081 
4082 /*
4083  * when we wait for progress in the block group caching, its because
4084  * our allocation attempt failed at least once.  So, we must sleep
4085  * and let some progress happen before we try again.
4086  *
4087  * This function will sleep at least once waiting for new free space to
4088  * show up, and then it will check the block group free space numbers
4089  * for our min num_bytes.  Another option is to have it go ahead
4090  * and look in the rbtree for a free extent of a given size, but this
4091  * is a good start.
4092  */
4093 static noinline int
4094 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
4095                                 u64 num_bytes)
4096 {
4097         struct btrfs_caching_control *caching_ctl;
4098         DEFINE_WAIT(wait);
4099 
4100         caching_ctl = get_caching_control(cache);
4101         if (!caching_ctl)
4102                 return 0;
4103 
4104         wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
4105                    (cache->free_space >= num_bytes));
4106 
4107         put_caching_control(caching_ctl);
4108         return 0;
4109 }
4110 
4111 static noinline int
4112 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
4113 {
4114         struct btrfs_caching_control *caching_ctl;
4115         DEFINE_WAIT(wait);
4116 
4117         caching_ctl = get_caching_control(cache);
4118         if (!caching_ctl)
4119                 return 0;
4120 
4121         wait_event(caching_ctl->wait, block_group_cache_done(cache));
4122 
4123         put_caching_control(caching_ctl);
4124         return 0;
4125 }
4126 
4127 enum btrfs_loop_type {
4128         LOOP_FIND_IDEAL = 0,
4129         LOOP_CACHING_NOWAIT = 1,
4130         LOOP_CACHING_WAIT = 2,
4131         LOOP_ALLOC_CHUNK = 3,
4132         LOOP_NO_EMPTY_SIZE = 4,
4133 };
4134 
4135 /*
4136  * walks the btree of allocated extents and find a hole of a given size.
4137  * The key ins is changed to record the hole:
4138  * ins->objectid == block start
4139  * ins->flags = BTRFS_EXTENT_ITEM_KEY
4140  * ins->offset == number of blocks
4141  * Any available blocks before search_start are skipped.
4142  */
4143 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
4144                                      struct btrfs_root *orig_root,
4145                                      u64 num_bytes, u64 empty_size,
4146                                      u64 search_start, u64 search_end,
4147                                      u64 hint_byte, struct btrfs_key *ins,
4148                                      u64 exclude_start, u64 exclude_nr,
4149                                      int data)
4150 {
4151         int ret = 0;
4152         struct btrfs_root *root = orig_root->fs_info->extent_root;
4153         struct btrfs_free_cluster *last_ptr = NULL;
4154         struct btrfs_block_group_cache *block_group = NULL;
4155         int empty_cluster = 2 * 1024 * 1024;
4156         int allowed_chunk_alloc = 0;
4157         int done_chunk_alloc = 0;
4158         struct btrfs_space_info *space_info;
4159         int last_ptr_loop = 0;
4160         int loop = 0;
4161         bool found_uncached_bg = false;
4162         bool failed_cluster_refill = false;
4163         bool failed_alloc = false;
4164         u64 ideal_cache_percent = 0;
4165         u64 ideal_cache_offset = 0;
4166 
4167         WARN_ON(num_bytes < root->sectorsize);
4168         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
4169         ins->objectid = 0;
4170         ins->offset = 0;
4171 
4172         space_info = __find_space_info(root->fs_info, data);
4173 
4174         if (orig_root->ref_cows || empty_size)
4175                 allowed_chunk_alloc = 1;
4176 
4177         if (data & BTRFS_BLOCK_GROUP_METADATA) {
4178                 last_ptr = &root->fs_info->meta_alloc_cluster;
4179                 if (!btrfs_test_opt(root, SSD))
4180                         empty_cluster = 64 * 1024;
4181         }
4182 
4183         if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD)) {
4184                 last_ptr = &root->fs_info->data_alloc_cluster;
4185         }
4186 
4187         if (last_ptr) {
4188                 spin_lock(&last_ptr->lock);
4189                 if (last_ptr->block_group)
4190                         hint_byte = last_ptr->window_start;
4191                 spin_unlock(&last_ptr->lock);
4192         }
4193 
4194         search_start = max(search_start, first_logical_byte(root, 0));
4195         search_start = max(search_start, hint_byte);
4196 
4197         if (!last_ptr)
4198                 empty_cluster = 0;
4199 
4200         if (search_start == hint_byte) {
4201 ideal_cache:
4202                 block_group = btrfs_lookup_block_group(root->fs_info,
4203                                                        search_start);
4204                 /*
4205                  * we don't want to use the block group if it doesn't match our
4206                  * allocation bits, or if its not cached.
4207                  *
4208                  * However if we are re-searching with an ideal block group
4209                  * picked out then we don't care that the block group is cached.
4210                  */
4211                 if (block_group && block_group_bits(block_group, data) &&
4212                     (block_group->cached != BTRFS_CACHE_NO ||
4213                      search_start == ideal_cache_offset)) {
4214                         down_read(&space_info->groups_sem);
4215                         if (list_empty(&block_group->list) ||
4216                             block_group->ro) {
4217                                 /*
4218                                  * someone is removing this block group,
4219                                  * we can't jump into the have_block_group
4220                                  * target because our list pointers are not
4221                                  * valid
4222                                  */
4223                                 btrfs_put_block_group(block_group);
4224                                 up_read(&space_info->groups_sem);
4225                         } else {
4226                                 goto have_block_group;
4227                         }
4228                 } else if (block_group) {
4229                         btrfs_put_block_group(block_group);
4230                 }
4231         }
4232 search:
4233         down_read(&space_info->groups_sem);
4234         list_for_each_entry(block_group, &space_info->block_groups, list) {
4235                 u64 offset;
4236                 int cached;
4237 
4238                 btrfs_get_block_group(block_group);
4239                 search_start = block_group->key.objectid;
4240 
4241 have_block_group:
4242                 if (unlikely(block_group->cached == BTRFS_CACHE_NO)) {
4243                         u64 free_percent;
4244 
4245                         free_percent = btrfs_block_group_used(&block_group->item);
4246                         free_percent *= 100;
4247                         free_percent = div64_u64(free_percent,
4248                                                  block_group->key.offset);
4249                         free_percent = 100 - free_percent;
4250                         if (free_percent > ideal_cache_percent &&
4251                             likely(!block_group->ro)) {
4252                                 ideal_cache_offset = block_group->key.objectid;
4253                                 ideal_cache_percent = free_percent;
4254                         }
4255 
4256                         /*
4257                          * We only want to start kthread caching if we are at
4258                          * the point where we will wait for caching to make
4259                          * progress, or if our ideal search is over and we've
4260                          * found somebody to start caching.
4261                          */
4262                         if (loop > LOOP_CACHING_NOWAIT ||
4263                             (loop > LOOP_FIND_IDEAL &&
4264                              atomic_read(&space_info->caching_threads) < 2)) {
4265                                 ret = cache_block_group(block_group);
4266                                 BUG_ON(ret);
4267                         }
4268                         found_uncached_bg = true;
4269 
4270                         /*
4271                          * If loop is set for cached only, try the next block
4272                          * group.
4273                          */
4274                         if (loop == LOOP_FIND_IDEAL)
4275                                 goto loop;
4276                 }
4277 
4278                 cached = block_group_cache_done(block_group);
4279                 if (unlikely(!cached))
4280                         found_uncached_bg = true;
4281 
4282                 if (unlikely(block_group->ro))
4283                         goto loop;
4284 
4285                 /*
4286                  * Ok we want to try and use the cluster allocator, so lets look
4287                  * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will
4288                  * have tried the cluster allocator plenty of times at this
4289                  * point and not have found anything, so we are likely way too
4290                  * fragmented for the clustering stuff to find anything, so lets
4291                  * just skip it and let the allocator find whatever block it can
4292                  * find
4293                  */
4294                 if (last_ptr && loop < LOOP_NO_EMPTY_SIZE) {
4295                         /*
4296                          * the refill lock keeps out other
4297                          * people trying to start a new cluster
4298                          */
4299                         spin_lock(&last_ptr->refill_lock);
4300                         if (last_ptr->block_group &&
4301                             (last_ptr->block_group->ro ||
4302                             !block_group_bits(last_ptr->block_group, data))) {
4303                                 offset = 0;
4304                                 goto refill_cluster;
4305                         }
4306 
4307                         offset = btrfs_alloc_from_cluster(block_group, last_ptr,
4308                                                  num_bytes, search_start);
4309                         if (offset) {
4310                                 /* we have a block, we're done */
4311                                 spin_unlock(&last_ptr->refill_lock);
4312                                 goto checks;
4313                         }
4314 
4315                         spin_lock(&last_ptr->lock);
4316                         /*
4317                          * whoops, this cluster doesn't actually point to
4318                          * this block group.  Get a ref on the block
4319                          * group is does point to and try again
4320                          */
4321                         if (!last_ptr_loop && last_ptr->block_group &&
4322                             last_ptr->block_group != block_group) {
4323 
4324                                 btrfs_put_block_group(block_group);
4325                                 block_group = last_ptr->block_group;
4326                                 btrfs_get_block_group(block_group);
4327                                 spin_unlock(&last_ptr->lock);
4328                                 spin_unlock(&last_ptr->refill_lock);
4329 
4330                                 last_ptr_loop = 1;
4331                                 search_start = block_group->key.objectid;
4332                                 /*
4333                                  * we know this block group is properly
4334                                  * in the list because
4335                                  * btrfs_remove_block_group, drops the
4336                                  * cluster before it removes the block
4337                                  * group from the list
4338                                  */
4339                                 goto have_block_group;
4340                         }
4341                         spin_unlock(&last_ptr->lock);
4342 refill_cluster:
4343                         /*
4344                          * this cluster didn't work out, free it and
4345                          * start over
4346                          */
4347                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
4348 
4349                         last_ptr_loop = 0;
4350 
4351                         /* allocate a cluster in this block group */
4352                         ret = btrfs_find_space_cluster(trans, root,
4353                                                block_group, last_ptr,
4354                                                offset, num_bytes,
4355                                                empty_cluster + empty_size);
4356                         if (ret == 0) {
4357                                 /*
4358                                  * now pull our allocation out of this
4359                                  * cluster
4360                                  */
4361                                 offset = btrfs_alloc_from_cluster(block_group,
4362                                                   last_ptr, num_bytes,
4363                                                   search_start);
4364                                 if (offset) {
4365                                         /* we found one, proceed */
4366                                         spin_unlock(&last_ptr->refill_lock);
4367                                         goto checks;
4368                                 }
4369                         } else if (!cached && loop > LOOP_CACHING_NOWAIT
4370                                    && !failed_cluster_refill) {
4371                                 spin_unlock(&last_ptr->refill_lock);
4372 
4373                                 failed_cluster_refill = true;
4374                                 wait_block_group_cache_progress(block_group,
4375                                        num_bytes + empty_cluster + empty_size);
4376                                 goto have_block_group;
4377                         }
4378 
4379                         /*
4380                          * at this point we either didn't find a cluster
4381                          * or we weren't able to allocate a block from our
4382                          * cluster.  Free the cluster we've been trying
4383                          * to use, and go to the next block group
4384                          */
4385                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
4386                         spin_unlock(&last_ptr->refill_lock);
4387                         goto loop;
4388                 }
4389 
4390                 offset = btrfs_find_space_for_alloc(block_group, search_start,
4391                                                     num_bytes, empty_size);
4392                 /*
4393                  * If we didn't find a chunk, and we haven't failed on this
4394                  * block group before, and this block group is in the middle of
4395                  * caching and we are ok with waiting, then go ahead and wait
4396                  * for progress to be made, and set failed_alloc to true.
4397                  *
4398                  * If failed_alloc is true then we've already waited on this
4399                  * block group once and should move on to the next block group.
4400                  */
4401                 if (!offset && !failed_alloc && !cached &&
4402                     loop > LOOP_CACHING_NOWAIT) {
4403                         wait_block_group_cache_progress(block_group,
4404                                                 num_bytes + empty_size);
4405                         failed_alloc = true;
4406                         goto have_block_group;
4407                 } else if (!offset) {
4408                         goto loop;
4409                 }
4410 checks:
4411                 search_start = stripe_align(root, offset);
4412                 /* move on to the next group */
4413                 if (search_start + num_bytes >= search_end) {
4414                         btrfs_add_free_space(block_group, offset, num_bytes);
4415                         goto loop;
4416                 }
4417 
4418                 /* move on to the next group */
4419                 if (search_start + num_bytes >
4420                     block_group->key.objectid + block_group->key.offset) {
4421                         btrfs_add_free_space(block_group, offset, num_bytes);
4422                         goto loop;
4423                 }
4424 
4425                 if (exclude_nr > 0 &&
4426                     (search_start + num_bytes > exclude_start &&
4427                      search_start < exclude_start + exclude_nr)) {
4428                         search_start = exclude_start + exclude_nr;
4429 
4430                         btrfs_add_free_space(block_group, offset, num_bytes);
4431                         /*
4432                          * if search_start is still in this block group
4433                          * then we just re-search this block group
4434                          */
4435                         if (search_start >= block_group->key.objectid &&
4436                             search_start < (block_group->key.objectid +
4437                                             block_group->key.offset))
4438                                 goto have_block_group;
4439                         goto loop;
4440                 }
4441 
4442                 ins->objectid = search_start;
4443                 ins->offset = num_bytes;
4444 
4445                 if (offset < search_start)
4446                         btrfs_add_free_space(block_group, offset,
4447                                              search_start - offset);
4448                 BUG_ON(offset > search_start);
4449 
4450                 update_reserved_extents(block_group, num_bytes, 1);
4451 
4452                 /* we are all good, lets return */
4453                 break;
4454 loop:
4455                 failed_cluster_refill = false;
4456                 failed_alloc = false;
4457                 btrfs_put_block_group(block_group);
4458         }
4459         up_read(&space_info->groups_sem);
4460 
4461         /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
4462          *                      for them to make caching progress.  Also
4463          *                      determine the best possible bg to cache
4464          * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
4465          *                      caching kthreads as we move along
4466          * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
4467          * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
4468          * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
4469          *                      again
4470          */
4471         if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE &&
4472             (found_uncached_bg || empty_size || empty_cluster ||
4473              allowed_chunk_alloc)) {
4474                 if (loop == LOOP_FIND_IDEAL && found_uncached_bg) {
4475                         found_uncached_bg = false;
4476                         loop++;
4477                         if (!ideal_cache_percent &&
4478                             atomic_read(&space_info->caching_threads))
4479                                 goto search;
4480 
4481                         /*
4482                          * 1 of the following 2 things have happened so far
4483                          *
4484                          * 1) We found an ideal block group for caching that
4485                          * is mostly full and will cache quickly, so we might
4486                          * as well wait for it.
4487                          *
4488                          * 2) We searched for cached only and we didn't find
4489                          * anything, and we didn't start any caching kthreads
4490                          * either, so chances are we will loop through and
4491                          * start a couple caching kthreads, and then come back
4492                          * around and just wait for them.  This will be slower
4493                          * because we will have 2 caching kthreads reading at
4494                          * the same time when we could have just started one
4495                          * and waited for it to get far enough to give us an
4496                          * allocation, so go ahead and go to the wait caching
4497                          * loop.
4498                          */
4499                         loop = LOOP_CACHING_WAIT;
4500                         search_start = ideal_cache_offset;
4501                         ideal_cache_percent = 0;
4502                         goto ideal_cache;
4503                 } else if (loop == LOOP_FIND_IDEAL) {
4504                         /*
4505                          * Didn't find a uncached bg, wait on anything we find
4506                          * next.
4507                          */
4508                         loop = LOOP_CACHING_WAIT;
4509                         goto search;
4510                 }
4511 
4512                 if (loop < LOOP_CACHING_WAIT) {
4513                         loop++;
4514                         goto search;
4515                 }
4516 
4517                 if (loop == LOOP_ALLOC_CHUNK) {
4518                         empty_size = 0;
4519                         empty_cluster = 0;
4520                 }
4521 
4522                 if (allowed_chunk_alloc) {
4523                         ret = do_chunk_alloc(trans, root, num_bytes +
4524                                              2 * 1024 * 1024, data, 1);
4525                         allowed_chunk_alloc = 0;
4526                         done_chunk_alloc = 1;
4527                 } else if (!done_chunk_alloc) {
4528                         space_info->force_alloc = 1;
4529                 }
4530 
4531                 if (loop < LOOP_NO_EMPTY_SIZE) {
4532                         loop++;
4533                         goto search;
4534                 }
4535                 ret = -ENOSPC;
4536         } else if (!ins->objectid) {
4537                 ret = -ENOSPC;
4538         }
4539 
4540         /* we found what we needed */
4541         if (ins->objectid) {
4542                 if (!(data & BTRFS_BLOCK_GROUP_DATA))
4543                         trans->block_group = block_group->key.objectid;
4544 
4545                 btrfs_put_block_group(block_group);
4546                 ret = 0;
4547         }
4548 
4549         return ret;
4550 }
4551 
4552 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
4553                             int dump_block_groups)
4554 {
4555         struct btrfs_block_group_cache *cache;
4556 
4557         spin_lock(&info->lock);
4558         printk(KERN_INFO "space_info has %llu free, is %sfull\n",
4559                (unsigned long long)(info->total_bytes - info->bytes_used -
4560                                     info->bytes_pinned - info->bytes_reserved -
4561                                     info->bytes_super),
4562                (info->full) ? "" : "not ");
4563         printk(KERN_INFO "space_info total=%llu, pinned=%llu, delalloc=%llu,"
4564                " may_use=%llu, used=%llu, root=%llu, super=%llu, reserved=%llu"
4565                "\n",
4566                (unsigned long long)info->total_bytes,
4567                (unsigned long long)info->bytes_pinned,
4568                (unsigned long long)info->bytes_delalloc,
4569                (unsigned long long)info->bytes_may_use,
4570                (unsigned long long)info->bytes_used,
4571                (unsigned long long)info->bytes_root,
4572                (unsigned long long)info->bytes_super,
4573                (unsigned long long)info->bytes_reserved);
4574         spin_unlock(&info->lock);
4575 
4576         if (!dump_block_groups)
4577                 return;
4578 
4579         down_read(&info->groups_sem);
4580         list_for_each_entry(cache, &info->block_groups, list) {
4581                 spin_lock(&cache->lock);
4582                 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
4583                        "%llu pinned %llu reserved\n",
4584                        (unsigned long long)cache->key.objectid,
4585                        (unsigned long long)cache->key.offset,
4586                        (unsigned long long)btrfs_block_group_used(&cache->item),
4587                        (unsigned long long)cache->pinned,
4588                        (unsigned long long)cache->reserved);
4589                 btrfs_dump_free_space(cache, bytes);
4590                 spin_unlock(&cache->lock);
4591         }
4592         up_read(&info->groups_sem);
4593 }
4594 
4595 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
4596                          struct btrfs_root *root,
4597                          u64 num_bytes, u64 min_alloc_size,
4598                          u64 empty_size, u64 hint_byte,
4599                          u64 search_end, struct btrfs_key *ins,
4600                          u64 data)
4601 {
4602         int ret;
4603         u64 search_start = 0;
4604 
4605         data = btrfs_get_alloc_profile(root, data);
4606 again:
4607         /*
4608          * the only place that sets empty_size is btrfs_realloc_node, which
4609          * is not called recursively on allocations
4610          */
4611         if (empty_size || root->ref_cows)
4612                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4613                                      num_bytes + 2 * 1024 * 1024, data, 0);
4614 
4615         WARN_ON(num_bytes < root->sectorsize);
4616         ret = find_free_extent(trans, root, num_bytes, empty_size,
4617                                search_start, search_end, hint_byte, ins,
4618                                trans->alloc_exclude_start,
4619                                trans->alloc_exclude_nr, data);
4620 
4621         if (ret == -ENOSPC && num_bytes > min_alloc_size) {
4622                 num_bytes = num_bytes >> 1;
4623                 num_bytes = num_bytes & ~(root->sectorsize - 1);
4624                 num_bytes = max(num_bytes, min_alloc_size);
4625                 do_chunk_alloc(trans, root->fs_info->extent_root,
4626                                num_bytes, data, 1);
4627                 goto again;
4628         }
4629         if (ret == -ENOSPC) {
4630                 struct btrfs_space_info *sinfo;
4631 
4632                 sinfo = __find_space_info(root->fs_info, data);
4633                 printk(KERN_ERR "btrfs allocation failed flags %llu, "
4634                        "wanted %llu\n", (unsigned long long)data,
4635                        (unsigned long long)num_bytes);
4636                 dump_space_info(sinfo, num_bytes, 1);
4637         }
4638 
4639         return ret;
4640 }
4641 
4642 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
4643 {
4644         struct btrfs_block_group_cache *cache;
4645         int ret = 0;
4646 
4647         cache = btrfs_lookup_block_group(root->fs_info, start);
4648         if (!cache) {
4649                 printk(KERN_ERR "Unable to find block group for %llu\n",
4650                        (unsigned long long)start);
4651                 return -ENOSPC;
4652         }
4653 
4654         ret = btrfs_discard_extent(root, start, len);
4655 
4656         btrfs_add_free_space(cache, start, len);
4657         update_reserved_extents(cache, len, 0);
4658         btrfs_put_block_group(cache);
4659 
4660         return ret;
4661 }
4662 
4663 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
4664                                       struct btrfs_root *root,
4665                                       u64 parent, u64 root_objectid,
4666                                       u64 flags, u64 owner, u64 offset,
4667                                       struct btrfs_key *ins, int ref_mod)
4668 {
4669         int ret;
4670         struct btrfs_fs_info *fs_info = root->fs_info;
4671         struct btrfs_extent_item *extent_item;
4672         struct btrfs_extent_inline_ref *iref;
4673         struct btrfs_path *path;
4674         struct extent_buffer *leaf;
4675         int type;
4676         u32 size;
4677 
4678         if (parent > 0)
4679                 type = BTRFS_SHARED_DATA_REF_KEY;
4680         else
4681                 type = BTRFS_EXTENT_DATA_REF_KEY;
4682 
4683         size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
4684 
4685         path = btrfs_alloc_path();
4686         BUG_ON(!path);
4687 
4688         path->leave_spinning = 1;
4689         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
4690                                       ins, size);
4691         BUG_ON(ret);
4692 
4693         leaf = path->nodes[0];
4694         extent_item = btrfs_item_ptr(leaf, path->slots[0],
4695                                      struct btrfs_extent_item);
4696         btrfs_set_extent_refs(leaf, extent_item, ref_mod);
4697         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
4698         btrfs_set_extent_flags(leaf, extent_item,
4699                                flags | BTRFS_EXTENT_FLAG_DATA);
4700 
4701         iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
4702         btrfs_set_extent_inline_ref_type(leaf, iref, type);
4703         if (parent > 0) {
4704                 struct btrfs_shared_data_ref *ref;
4705                 ref = (struct btrfs_shared_data_ref *)(iref + 1);
4706                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
4707                 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
4708         } else {
4709                 struct btrfs_extent_data_ref *ref;
4710                 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
4711                 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
4712                 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
4713                 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
4714                 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
4715         }
4716 
4717         btrfs_mark_buffer_dirty(path->nodes[0]);
4718         btrfs_free_path(path);
4719 
4720         ret = update_block_group(trans, root, ins->objectid, ins->offset,
4721                                  1, 0);
4722         if (ret) {
4723                 printk(KERN_ERR "btrfs update block group failed for %llu "
4724                        "%llu\n", (unsigned long long)ins->objectid,
4725                        (unsigned long long)ins->offset);
4726                 BUG();
4727         }
4728         return ret;
4729 }
4730 
4731 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
4732                                      struct btrfs_root *root,
4733                                      u64 parent, u64 root_objectid,
4734                                      u64 flags, struct btrfs_disk_key *key,
4735                                      int level, struct btrfs_key *ins)
4736 {
4737         int ret;
4738         struct btrfs_fs_info *fs_info = root->fs_info;
4739         struct btrfs_extent_item *extent_item;
4740         struct btrfs_tree_block_info *block_info;
4741         struct btrfs_extent_inline_ref *iref;
4742         struct btrfs_path *path;
4743         struct extent_buffer *leaf;
4744         u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
4745 
4746         path = btrfs_alloc_path();
4747         BUG_ON(!path);
4748 
4749         path->leave_spinning = 1;
4750         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
4751                                       ins, size);
4752         BUG_ON(ret);
4753 
4754         leaf = path->nodes[0];
4755         extent_item = btrfs_item_ptr(leaf, path->slots[0],
4756                                      struct btrfs_extent_item);
4757         btrfs_set_extent_refs(leaf, extent_item, 1);
4758         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
4759         btrfs_set_extent_flags(leaf, extent_item,
4760                                flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
4761         block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
4762 
4763         btrfs_set_tree_block_key(leaf, block_info, key);
4764         btrfs_set_tree_block_level(leaf, block_info, level);
4765 
4766         iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
4767         if (parent > 0) {
4768                 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
4769                 btrfs_set_extent_inline_ref_type(leaf, iref,
4770                                                  BTRFS_SHARED_BLOCK_REF_KEY);
4771                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
4772         } else {
4773                 btrfs_set_extent_inline_ref_type(leaf, iref,
4774                                                  BTRFS_TREE_BLOCK_REF_KEY);
4775                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
4776         }
4777 
4778         btrfs_mark_buffer_dirty(leaf);
4779         btrfs_free_path(path);
4780 
4781         ret = update_block_group(trans, root, ins->objectid, ins->offset,
4782                                  1, 0);
4783         if (ret) {
4784                 printk(KERN_ERR "btrfs update block group failed for %llu "
4785                        "%llu\n", (unsigned long long)ins->objectid,
4786                        (unsigned long long)ins->offset);
4787                 BUG();
4788         }
4789         return ret;
4790 }
4791 
4792 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
4793                                      struct btrfs_root *root,
4794                                      u64 root_objectid, u64 owner,
4795                                      u64 offset, struct btrfs_key *ins)
4796 {
4797         int ret;
4798 
4799         BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
4800 
4801         ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset,
4802                                          0, root_objectid, owner, offset,
4803                                          BTRFS_ADD_DELAYED_EXTENT, NULL);
4804         return ret;
4805 }
4806 
4807 /*
4808  * this is used by the tree logging recovery code.  It records that
4809  * an extent has been allocated and makes sure to clear the free
4810  * space cache bits as well
4811  */
4812 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
4813                                    struct btrfs_root *root,
4814                                    u64 root_objectid, u64 owner, u64 offset,
4815                                    struct btrfs_key *ins)
4816 {
4817         int ret;
4818         struct btrfs_block_group_cache *block_group;
4819         struct btrfs_caching_control *caching_ctl;
4820         u64 start = ins->objectid;
4821         u64 num_bytes = ins->offset;
4822 
4823         block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
4824         cache_block_group(block_group);
4825         caching_ctl = get_caching_control(block_group);
4826 
4827         if (!caching_ctl) {
4828                 BUG_ON(!block_group_cache_done(block_group));
4829                 ret = btrfs_remove_free_space(block_group, start, num_bytes);
4830                 BUG_ON(ret);
4831         } else {
4832                 mutex_lock(&caching_ctl->mutex);
4833 
4834                 if (start >= caching_ctl->progress) {
4835                         ret = add_excluded_extent(root, start, num_bytes);
4836                         BUG_ON(ret);
4837                 } else if (start + num_bytes <= caching_ctl->progress) {
4838                         ret = btrfs_remove_free_space(block_group,
4839                                                       start, num_bytes);
4840                         BUG_ON(ret);
4841                 } else {
4842                         num_bytes = caching_ctl->progress - start;
4843                         ret = btrfs_remove_free_space(block_group,
4844                                                       start, num_bytes);
4845                         BUG_ON(ret);
4846 
4847                         start = caching_ctl->progress;
4848                         num_bytes = ins->objectid + ins->offset -
4849                                     caching_ctl->progress;
4850                         ret = add_excluded_extent(root, start, num_bytes);
4851                         BUG_ON(ret);
4852                 }
4853 
4854                 mutex_unlock(&caching_ctl->mutex);
4855                 put_caching_control(caching_ctl);
4856         }
4857 
4858         update_reserved_extents(block_group, ins->offset, 1);
4859         btrfs_put_block_group(block_group);
4860         ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
4861                                          0, owner, offset, ins, 1);
4862         return ret;
4863 }
4864 
4865 /*
4866  * finds a free extent and does all the dirty work required for allocation
4867  * returns the key for the extent through ins, and a tree buffer for
4868  * the first block of the extent through buf.
4869  *
4870  * returns 0 if everything worked, non-zero otherwise.
4871  */
4872 static int alloc_tree_block(struct btrfs_trans_handle *trans,
4873                             struct btrfs_root *root,
4874                             u64 num_bytes, u64 parent, u64 root_objectid,
4875                             struct btrfs_disk_key *key, int level,
4876                             u64 empty_size, u64 hint_byte, u64 search_end,
4877                             struct btrfs_key *ins)
4878 {
4879         int ret;
4880         u64 flags = 0;
4881 
4882         ret = btrfs_reserve_extent(trans, root, num_bytes, num_bytes,
4883                                    empty_size, hint_byte, search_end,
4884                                    ins, 0);
4885         if (ret)
4886                 return ret;
4887 
4888         if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
4889                 if (parent == 0)
4890                         parent = ins->objectid;
4891                 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
4892         } else
4893                 BUG_ON(parent > 0);
4894 
4895         if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
4896                 struct btrfs_delayed_extent_op *extent_op;
4897                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
4898                 BUG_ON(!extent_op);
4899                 if (key)
4900                         memcpy(&extent_op->key, key, sizeof(extent_op->key));
4901                 else
4902                         memset(&extent_op->key, 0, sizeof(extent_op->key));
4903                 extent_op->flags_to_set = flags;
4904                 extent_op->update_key = 1;
4905                 extent_op->update_flags = 1;
4906                 extent_op->is_data = 0;
4907 
4908                 ret = btrfs_add_delayed_tree_ref(trans, ins->objectid,
4909                                         ins->offset, parent, root_objectid,
4910                                         level, BTRFS_ADD_DELAYED_EXTENT,
4911                                         extent_op);
4912                 BUG_ON(ret);
4913         }
4914 
4915         if (root_objectid == root->root_key.objectid) {
4916                 u64 used;
4917                 spin_lock(&root->node_lock);
4918                 used = btrfs_root_used(&root->root_item) + num_bytes;
4919                 btrfs_set_root_used(&root->root_item, used);
4920                 spin_unlock(&root->node_lock);
4921         }
4922         return ret;
4923 }
4924 
4925 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
4926                                             struct btrfs_root *root,
4927                                             u64 bytenr, u32 blocksize,
4928                                             int level)
4929 {
4930         struct extent_buffer *buf;
4931 
4932         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
4933         if (!buf)
4934                 return ERR_PTR(-ENOMEM);
4935         btrfs_set_header_generation(buf, trans->transid);
4936         btrfs_set_buffer_lockdep_class(buf, level);
4937         btrfs_tree_lock(buf);
4938         clean_tree_block(trans, root, buf);
4939 
4940         btrfs_set_lock_blocking(buf);
4941         btrfs_set_buffer_uptodate(buf);
4942 
4943         if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
4944                 /*
4945                  * we allow two log transactions at a time, use different
4946                  * EXENT bit to differentiate dirty pages.
4947                  */
4948                 if (root->log_transid % 2 == 0)
4949                         set_extent_dirty(&root->dirty_log_pages, buf->start,
4950                                         buf->start + buf->len - 1, GFP_NOFS);
4951                 else
4952                         set_extent_new(&root->dirty_log_pages, buf->start,
4953                                         buf->start + buf->len - 1, GFP_NOFS);
4954         } else {
4955                 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
4956                          buf->start + buf->len - 1, GFP_NOFS);
4957         }
4958         trans->blocks_used++;
4959         /* this returns a buffer locked for blocking */
4960         return buf;
4961 }
4962 
4963 /*
4964  * helper function to allocate a block for a given tree
4965  * returns the tree buffer or NULL.
4966  */
4967 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
4968                                         struct btrfs_root *root, u32 blocksize,
4969                                         u64 parent, u64 root_objectid,
4970                                         struct btrfs_disk_key *key, int level,
4971                                         u64 hint, u64 empty_size)
4972 {
4973         struct btrfs_key ins;
4974         int ret;
4975         struct extent_buffer *buf;
4976 
4977         ret = alloc_tree_block(trans, root, blocksize, parent, root_objectid,
4978                                key, level, empty_size, hint, (u64)-1, &ins);
4979         if (ret) {
4980                 BUG_ON(ret > 0);
4981                 return ERR_PTR(ret);
4982         }
4983 
4984         buf = btrfs_init_new_buffer(trans, root, ins.objectid,
4985                                     blocksize, level);
4986         return buf;
4987 }
4988 
4989 struct walk_control {
4990         u64 refs[BTRFS_MAX_LEVEL];
4991         u64 flags[BTRFS_MAX_LEVEL];
4992         struct btrfs_key update_progress;
4993         int stage;
4994         int level;
4995         int shared_level;
4996         int update_ref;
4997         int keep_locks;
4998         int reada_slot;
4999         int reada_count;
5000 };
5001 
5002 #define DROP_REFERENCE  1
5003 #define UPDATE_BACKREF  2
5004 
5005 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
5006                                      struct btrfs_root *root,
5007                                      struct walk_control *wc,
5008                                      struct btrfs_path *path)
5009 {
5010         u64 bytenr;
5011         u64 generation;
5012         u64 refs;
5013         u64 flags;
5014         u64 last = 0;
5015         u32 nritems;
5016         u32 blocksize;
5017         struct btrfs_key key;
5018         struct extent_buffer *eb;
5019         int ret;
5020         int slot;
5021         int nread = 0;
5022 
5023         if (path->slots[wc->level] < wc->reada_slot) {
5024                 wc->reada_count = wc->reada_count * 2 / 3;
5025                 wc->reada_count = max(wc->reada_count, 2);
5026         } else {
5027                 wc->reada_count = wc->reada_count * 3 / 2;
5028                 wc->reada_count = min_t(int, wc->reada_count,
5029                                         BTRFS_NODEPTRS_PER_BLOCK(root));
5030         }
5031 
5032         eb = path->nodes[wc->level];
5033         nritems = btrfs_header_nritems(eb);
5034         blocksize = btrfs_level_size(root, wc->level - 1);
5035 
5036         for (slot = path->slots[wc->level]; slot < nritems; slot++) {
5037                 if (nread >= wc->reada_count)
5038                         break;
5039 
5040                 cond_resched();
5041                 bytenr = btrfs_node_blockptr(eb, slot);
5042                 generation = btrfs_node_ptr_generation(eb, slot);
5043 
5044                 if (slot == path->slots[wc->level])
5045                         goto reada;
5046 
5047                 if (wc->stage == UPDATE_BACKREF &&
5048                     generation <= root->root_key.offset)
5049                         continue;
5050 
5051                 /* We don't lock the tree block, it's OK to be racy here */
5052                 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
5053                                                &refs, &flags);
5054                 BUG_ON(ret);
5055                 BUG_ON(refs == 0);
5056 
5057                 if (wc->stage == DROP_REFERENCE) {
5058                         if (refs == 1)
5059                                 goto reada;
5060 
5061                         if (wc->level == 1 &&
5062                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5063                                 continue;
5064                         if (!wc->update_ref ||
5065                             generation <= root->root_key.offset)
5066                                 continue;
5067                         btrfs_node_key_to_cpu(eb, &key, slot);
5068                         ret = btrfs_comp_cpu_keys(&key,
5069                                                   &wc->update_progress);
5070                         if (ret < 0)
5071                                 continue;
5072                 } else {
5073                         if (wc->level == 1 &&
5074                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5075                                 continue;
5076                 }
5077 reada:
5078                 ret = readahead_tree_block(root, bytenr, blocksize,
5079                                            generation);
5080                 if (ret)
5081                         break;
5082                 last = bytenr + blocksize;
5083                 nread++;
5084         }
5085         wc->reada_slot = slot;
5086 }
5087 
5088 /*
5089  * hepler to process tree block while walking down the tree.
5090  *
5091  * when wc->stage == UPDATE_BACKREF, this function updates
5092  * back refs for pointers in the block.
5093  *
5094  * NOTE: return value 1 means we should stop walking down.
5095  */
5096 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
5097                                    struct btrfs_root *root,
5098                                    struct btrfs_path *path,
5099                                    struct walk_control *wc, int lookup_info)
5100 {
5101         int level = wc->level;
5102         struct extent_buffer *eb = path->nodes[level];
5103         u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5104         int ret;
5105 
5106         if (wc->stage == UPDATE_BACKREF &&
5107             btrfs_header_owner(eb) != root->root_key.objectid)
5108                 return 1;
5109 
5110         /*
5111          * when reference count of tree block is 1, it won't increase
5112          * again. once full backref flag is set, we never clear it.
5113          */
5114         if (lookup_info &&
5115             ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
5116              (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
5117                 BUG_ON(!path->locks[level]);
5118                 ret = btrfs_lookup_extent_info(trans, root,
5119                                                eb->start, eb->len,
5120                                                &wc->refs[level],
5121                                                &wc->flags[level]);
5122                 BUG_ON(ret);
5123                 BUG_ON(wc->refs[level] == 0);
5124         }
5125 
5126         if (wc->stage == DROP_REFERENCE) {
5127                 if (wc->refs[level] > 1)
5128                         return 1;
5129 
5130                 if (path->locks[level] && !wc->keep_locks) {
5131                         btrfs_tree_unlock(eb);
5132                         path->locks[level] = 0;
5133                 }
5134                 return 0;
5135         }
5136 
5137         /* wc->stage == UPDATE_BACKREF */
5138         if (!(wc->flags[level] & flag)) {
5139                 BUG_ON(!path->locks[level]);
5140                 ret = btrfs_inc_ref(trans, root, eb, 1);
5141                 BUG_ON(ret);
5142                 ret = btrfs_dec_ref(trans, root, eb, 0);
5143                 BUG_ON(ret);
5144                 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
5145                                                   eb->len, flag, 0);
5146                 BUG_ON(ret);
5147                 wc->flags[level] |= flag;
5148         }
5149 
5150         /*
5151          * the block is shared by multiple trees, so it's not good to
5152          * keep the tree lock
5153          */
5154         if (path->locks[level] && level > 0) {
5155                 btrfs_tree_unlock(eb);
5156                 path->locks[level] = 0;
5157         }
5158         return 0;
5159 }
5160 
5161 /*
5162  * hepler to process tree block pointer.
5163  *
5164  * when wc->stage == DROP_REFERENCE, this function checks
5165  * reference count of the block pointed to. if the block
5166  * is shared and we need update back refs for the subtree
5167  * rooted at the block, this function changes wc->stage to
5168  * UPDATE_BACKREF. if the block is shared and there is no
5169  * need to update back, this function drops the reference
5170  * to the block.
5171  *
5172  * NOTE: return value 1 means we should stop walking down.
5173  */
5174 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
5175                                  struct btrfs_root *root,
5176                                  struct btrfs_path *path,
5177                                  struct walk_control *wc, int *lookup_info)
5178 {
5179         u64 bytenr;
5180         u64 generation;
5181         u64 parent;
5182         u32 blocksize;
5183         struct btrfs_key key;
5184         struct extent_buffer *next;
5185         int level = wc->level;
5186         int reada = 0;
5187         int ret = 0;
5188 
5189         generation = btrfs_node_ptr_generation(path->nodes[level],
5190                                                path->slots[level]);
5191         /*
5192          * if the lower level block was created before the snapshot
5193          * was created, we know there is no need to update back refs
5194          * for the subtree
5195          */
5196         if (wc->stage == UPDATE_BACKREF &&
5197             generation <= root->root_key.offset) {
5198                 *lookup_info = 1;
5199                 return 1;
5200         }
5201 
5202         bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
5203         blocksize = btrfs_level_size(root, level - 1);
5204 
5205         next = btrfs_find_tree_block(root, bytenr, blocksize);
5206         if (!next) {
5207                 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
5208                 reada = 1;
5209         }
5210         btrfs_tree_lock(next);
5211         btrfs_set_lock_blocking(next);
5212 
5213         ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
5214                                        &wc->refs[level - 1],
5215                                        &wc->flags[level - 1]);
5216         BUG_ON(ret);
5217         BUG_ON(wc->refs[level - 1] == 0);
5218         *lookup_info = 0;
5219 
5220         if (wc->stage == DROP_REFERENCE) {
5221                 if (wc->refs[level - 1] > 1) {
5222                         if (level == 1 &&
5223                             (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5224                                 goto skip;
5225 
5226                         if (!wc->update_ref ||
5227                             generation <= root->root_key.offset)
5228                                 goto skip;
5229 
5230                         btrfs_node_key_to_cpu(path->nodes[level], &key,
5231                                               path->slots[level]);
5232                         ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
5233                         if (ret < 0)
5234                                 goto skip;
5235 
5236                         wc->stage = UPDATE_BACKREF;
5237                         wc->shared_level = level - 1;
5238                 }
5239         } else {
5240                 if (level == 1 &&
5241                     (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5242                         goto skip;
5243         }
5244 
5245         if (!btrfs_buffer_uptodate(next, generation)) {
5246                 btrfs_tree_unlock(next);
5247                 free_extent_buffer(next);
5248                 next = NULL;
5249                 *lookup_info = 1;
5250         }
5251 
5252         if (!next) {
5253                 if (reada && level == 1)
5254                         reada_walk_down(trans, root, wc, path);
5255                 next = read_tree_block(root, bytenr, blocksize, generation);
5256                 btrfs_tree_lock(next);
5257                 btrfs_set_lock_blocking(next);
5258         }
5259 
5260         level--;
5261         BUG_ON(level != btrfs_header_level(next));
5262         path->nodes[level] = next;
5263         path->slots[level] = 0;
5264         path->locks[level] = 1;
5265         wc->level = level;
5266         if (wc->level == 1)
5267                 wc->reada_slot = 0;
5268         return 0;
5269 skip:
5270         wc->refs[level - 1] = 0;
5271         wc->flags[level - 1] = 0;
5272         if (wc->stage == DROP_REFERENCE) {
5273                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
5274                         parent = path->nodes[level]->start;
5275                 } else {
5276                         BUG_ON(root->root_key.objectid !=
5277                                btrfs_header_owner(path->nodes[level]));
5278                         parent = 0;
5279                 }
5280 
5281                 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
5282                                         root->root_key.objectid, level - 1, 0);
5283                 BUG_ON(ret);
5284         }
5285         btrfs_tree_unlock(next);
5286         free_extent_buffer(next);
5287         *lookup_info = 1;
5288         return 1;
5289 }
5290 
5291 /*
5292  * hepler to process tree block while walking up the tree.
5293  *
5294  * when wc->stage == DROP_REFERENCE, this function drops
5295  * reference count on the block.
5296  *
5297  * when wc->stage == UPDATE_BACKREF, this function changes
5298  * wc->stage back to DROP_REFERENCE if we changed wc->stage
5299  * to UPDATE_BACKREF previously while processing the block.
5300  *
5301  * NOTE: return value 1 means we should stop walking up.
5302  */
5303 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
5304                                  struct btrfs_root *root,
5305                                  struct btrfs_path *path,
5306                                  struct walk_control *wc)
5307 {
5308         int ret = 0;
5309         int level = wc->level;
5310         struct extent_buffer *eb = path->nodes[level];
5311         u64 parent = 0;
5312 
5313         if (wc->stage == UPDATE_BACKREF) {
5314                 BUG_ON(wc->shared_level < level);
5315                 if (level < wc->shared_level)
5316                         goto out;
5317 
5318                 ret = find_next_key(path, level + 1, &wc->update_progress);
5319                 if (ret > 0)
5320                         wc->update_ref = 0;
5321 
5322                 wc->stage = DROP_REFERENCE;
5323                 wc->shared_level = -1;
5324                 path->slots[level] = 0;
5325 
5326                 /*
5327                  * check reference count again if the block isn't locked.
5328                  * we should start walking down the tree again if reference
5329                  * count is one.
5330                  */
5331                 if (!path->locks[level]) {
5332                         BUG_ON(level == 0);
5333                         btrfs_tree_lock(eb);
5334                         btrfs_set_lock_blocking(eb);
5335                         path->locks[level] = 1;
5336 
5337                         ret = btrfs_lookup_extent_info(trans, root,
5338                                                        eb->start, eb->len,
5339                                                        &wc->refs[level],
5340                                                        &wc->flags[level]);
5341                         BUG_ON(ret);
5342                         BUG_ON(wc->refs[level] == 0);
5343                         if (wc->refs[level] == 1) {
5344                                 btrfs_tree_unlock(eb);
5345                                 path->locks[level] = 0;
5346                                 return 1;
5347                         }
5348                 }
5349         }
5350 
5351         /* wc->stage == DROP_REFERENCE */
5352         BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
5353 
5354         if (wc->refs[level] == 1) {
5355                 if (level == 0) {
5356                         if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5357                                 ret = btrfs_dec_ref(trans, root, eb, 1);
5358                         else
5359                                 ret = btrfs_dec_ref(trans, root, eb, 0);
5360                         BUG_ON(ret);
5361                 }
5362                 /* make block locked assertion in clean_tree_block happy */
5363                 if (!path->locks[level] &&
5364                     btrfs_header_generation(eb) == trans->transid) {
5365                         btrfs_tree_lock(eb);
5366                         btrfs_set_lock_blocking(eb);
5367                         path->locks[level] = 1;
5368                 }
5369                 clean_tree_block(trans, root, eb);
5370         }
5371 
5372         if (eb == root->node) {
5373                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5374                         parent = eb->start;
5375                 else
5376                         BUG_ON(root->root_key.objectid !=
5377                                btrfs_header_owner(eb));
5378         } else {
5379                 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5380                         parent = path->nodes[level + 1]->start;
5381                 else
5382                         BUG_ON(root->root_key.objectid !=
5383                                btrfs_header_owner(path->nodes[level + 1]));
5384         }
5385 
5386         ret = btrfs_free_extent(trans, root, eb->start, eb->len, parent,
5387                                 root->root_key.objectid, level, 0);
5388         BUG_ON(ret);
5389 out:
5390         wc->refs[level] = 0;
5391         wc->flags[level] = 0;
5392         return ret;
5393 }
5394 
5395 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
5396                                    struct btrfs_root *root,
5397                                    struct btrfs_path *path,
5398                                    struct walk_control *wc)
5399 {
5400         int level = wc->level;
5401         int lookup_info = 1;
5402         int ret;
5403 
5404         while (level >= 0) {
5405                 ret = walk_down_proc(trans, root, path, wc, lookup_info);
5406                 if (ret > 0)
5407                         break;
5408 
5409                 if (level == 0)
5410                         break;
5411 
5412                 if (path->slots[level] >=
5413                     btrfs_header_nritems(path->nodes[level]))
5414                         break;
5415 
5416                 ret = do_walk_down(trans, root, path, wc, &lookup_info);
5417                 if (ret > 0) {
5418                         path->slots[level]++;
5419                         continue;
5420                 }
5421                 level = wc->level;
5422         }
5423         return 0;
5424 }
5425 
5426 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
5427                                  struct btrfs_root *root,
5428                                  struct btrfs_path *path,
5429                                  struct walk_control *wc, int max_level)
5430 {
5431         int level = wc->level;
5432         int ret;
5433 
5434         path->slots[level] = btrfs_header_nritems(path->nodes[level]);
5435         while (level < max_level && path->nodes[level]) {
5436                 wc->level = level;
5437                 if (path->slots[level] + 1 <
5438                     btrfs_header_nritems(path->nodes[level])) {
5439                         path->slots[level]++;
5440                         return 0;
5441                 } else {
5442                         ret = walk_up_proc(trans, root, path, wc);
5443                         if (ret > 0)
5444                                 return 0;
5445 
5446                         if (path->locks[level]) {
5447                                 btrfs_tree_unlock(path->nodes[level]);
5448                                 path->locks[level] = 0;
5449                         }
5450                         free_extent_buffer(path->nodes[level]);
5451                         path->nodes[level] = NULL;
5452                         level++;
5453                 }
5454         }
5455         return 1;
5456 }
5457 
5458 /*
5459  * drop a subvolume tree.
5460  *
5461  * this function traverses the tree freeing any blocks that only
5462  * referenced by the tree.
5463  *
5464  * when a shared tree block is found. this function decreases its
5465  * reference count by one. if update_ref is true, this function
5466  * also make sure backrefs for the shared block and all lower level
5467  * blocks are properly updated.
5468  */
5469 int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref)
5470 {
5471         struct btrfs_path *path;
5472         struct btrfs_trans_handle *trans;
5473         struct btrfs_root *tree_root = root->fs_info->tree_root;
5474         struct btrfs_root_item *root_item = &root->root_item;
5475         struct walk_control *wc;
5476         struct btrfs_key key;
5477         int err = 0;
5478         int ret;
5479         int level;
5480 
5481         path = btrfs_alloc_path();
5482         BUG_ON(!path);
5483 
5484         wc = kzalloc(sizeof(*wc), GFP_NOFS);
5485         BUG_ON(!wc);
5486 
5487         trans = btrfs_start_transaction(tree_root, 1);
5488 
5489         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
5490                 level = btrfs_header_level(root->node);
5491                 path->nodes[level] = btrfs_lock_root_node(root);
5492                 btrfs_set_lock_blocking(path->nodes[level]);
5493                 path->slots[level] = 0;
5494                 path->locks[level] = 1;
5495                 memset(&wc->update_progress, 0,
5496                        sizeof(wc->update_progress));
5497         } else {
5498                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
5499                 memcpy(&wc->update_progress, &key,
5500                        sizeof(wc->update_progress));
5501 
5502                 level = root_item->drop_level;
5503                 BUG_ON(level == 0);
5504                 path->lowest_level = level;
5505                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5506                 path->lowest_level = 0;
5507                 if (ret < 0) {
5508                         err = ret;
5509                         goto out;
5510                 }
5511                 WARN_ON(ret > 0);
5512 
5513                 /*
5514                  * unlock our path, this is safe because only this
5515                  * function is allowed to delete this snapshot
5516                  */
5517                 btrfs_unlock_up_safe(path, 0);
5518 
5519                 level = btrfs_header_level(root->node);
5520                 while (1) {
5521                         btrfs_tree_lock(path->nodes[level]);
5522                         btrfs_set_lock_blocking(path->nodes[level]);
5523 
5524                         ret = btrfs_lookup_extent_info(trans, root,
5525                                                 path->nodes[level]->start,
5526                                                 path->nodes[level]->len,
5527                                                 &wc->refs[level],
5528                                                 &wc->flags[level]);
5529                         BUG_ON(ret);
5530                         BUG_ON(wc->refs[level] == 0);
5531 
5532                         if (level == root_item->drop_level)
5533                                 break;
5534 
5535                         btrfs_tree_unlock(path->nodes[level]);
5536                         WARN_ON(wc->refs[level] != 1);
5537                         level--;
5538                 }
5539         }
5540 
5541         wc->level = level;
5542         wc->shared_level = -1;
5543         wc->stage = DROP_REFERENCE;
5544         wc->update_ref = update_ref;
5545         wc->keep_locks = 0;
5546         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
5547 
5548         while (1) {
5549                 ret = walk_down_tree(trans, root, path, wc);
5550                 if (ret < 0) {
5551                         err = ret;
5552                         break;
5553                 }
5554 
5555                 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
5556                 if (ret < 0) {
5557                         err = ret;
5558                         break;
5559                 }
5560 
5561                 if (ret > 0) {
5562                         BUG_ON(wc->stage != DROP_REFERENCE);
5563                         break;
5564                 }
5565 
5566                 if (wc->stage == DROP_REFERENCE) {
5567                         level = wc->level;
5568                         btrfs_node_key(path->nodes[level],
5569                                        &root_item->drop_progress,
5570                                        path->slots[level]);
5571                         root_item->drop_level = level;
5572                 }
5573 
5574                 BUG_ON(wc->level == 0);
5575                 if (trans->transaction->in_commit ||
5576                     trans->transaction->delayed_refs.flushing) {
5577                         ret = btrfs_update_root(trans, tree_root,
5578                                                 &root->root_key,
5579                                                 root_item);
5580                         BUG_ON(ret);
5581 
5582                         btrfs_end_transaction(trans, tree_root);
5583                         trans = btrfs_start_transaction(tree_root, 1);
5584                 } else {
5585                         unsigned long update;
5586                         update = trans->delayed_ref_updates;
5587                         trans->delayed_ref_updates = 0;
5588                         if (update)
5589                                 btrfs_run_delayed_refs(trans, tree_root,
5590                                                        update);
5591                 }
5592         }
5593         btrfs_release_path(root, path);
5594         BUG_ON(err);
5595 
5596         ret = btrfs_del_root(trans, tree_root, &root->root_key);
5597         BUG_ON(ret);
5598 
5599         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
5600                 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
5601                                            NULL, NULL);
5602                 BUG_ON(ret < 0);
5603                 if (ret > 0) {
5604                         ret = btrfs_del_orphan_item(trans, tree_root,
5605                                                     root->root_key.objectid);
5606                         BUG_ON(ret);
5607                 }
5608         }
5609 
5610         if (root->in_radix) {
5611                 btrfs_free_fs_root(tree_root->fs_info, root);
5612         } else {
5613                 free_extent_buffer(root->node);
5614                 free_extent_buffer(root->commit_root);
5615                 kfree(root);
5616         }
5617 out:
5618         btrfs_end_transaction(trans, tree_root);
5619         kfree(wc);
5620         btrfs_free_path(path);
5621         return err;
5622 }
5623 
5624 /*
5625  * drop subtree rooted at tree block 'node'.
5626  *
5627  * NOTE: this function will unlock and release tree block 'node'
5628  */
5629 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
5630                         struct btrfs_root *root,
5631                         struct extent_buffer *node,
5632                         struct extent_buffer *parent)
5633 {
5634         struct btrfs_path *path;
5635         struct walk_control *wc;
5636         int level;
5637         int parent_level;
5638         int ret = 0;
5639         int wret;
5640 
5641         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
5642 
5643         path = btrfs_alloc_path();
5644         BUG_ON(!path);
5645 
5646         wc = kzalloc(sizeof(*wc), GFP_NOFS);
5647         BUG_ON(!wc);
5648 
5649         btrfs_assert_tree_locked(parent);
5650         parent_level = btrfs_header_level(parent);
5651         extent_buffer_get(parent);
5652         path->nodes[parent_level] = parent;
5653         path->slots[parent_level] = btrfs_header_nritems(parent);
5654 
5655         btrfs_assert_tree_locked(node);
5656         level = btrfs_header_level(node);
5657         path->nodes[level] = node;
5658         path->slots[level] = 0;
5659         path->locks[level] = 1;
5660 
5661         wc->refs[parent_level] = 1;
5662         wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5663         wc->level = level;
5664         wc->shared_level = -1;
5665         wc->stage = DROP_REFERENCE;
5666         wc->update_ref = 0;
5667         wc->keep_locks = 1;
5668         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
5669 
5670         while (1) {
5671                 wret = walk_down_tree(trans, root, path, wc);
5672                 if (wret < 0) {
5673                         ret = wret;
5674                         break;
5675                 }
5676 
5677                 wret = walk_up_tree(trans, root, path, wc, parent_level);
5678                 if (wret < 0)
5679                         ret = wret;
5680                 if (wret != 0)
5681                         break;
5682         }
5683 
5684         kfree(wc);
5685         btrfs_free_path(path);
5686         return ret;
5687 }
5688 
5689 #if 0
5690 static unsigned long calc_ra(unsigned long start, unsigned long last,
5691                              unsigned long nr)
5692 {
5693         return min(last, start + nr - 1);
5694 }
5695 
5696 static noinline int relocate_inode_pages(struct inode *inode, u64 start,
5697                                          u64 len)
5698 {
5699         u64 page_start;
5700         u64 page_end;
5701         unsigned long first_index;
5702         unsigned long last_index;
5703         unsigned long i;
5704         struct page *page;
5705         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5706         struct file_ra_state *ra;
5707         struct btrfs_ordered_extent *ordered;
5708         unsigned int total_read = 0;
5709         unsigned int total_dirty = 0;
5710         int ret = 0;
5711 
5712         ra = kzalloc(sizeof(*ra), GFP_NOFS);
5713 
5714         mutex_lock(&inode->i_mutex);
5715         first_index = start >> PAGE_CACHE_SHIFT;
5716         last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
5717 
5718         /* make sure the dirty trick played by the caller work */
5719         ret = invalidate_inode_pages2_range(inode->i_mapping,
5720                                             first_index, last_index);
5721         if (ret)
5722                 goto out_unlock;
5723 
5724         file_ra_state_init(ra, inode->i_mapping);
5725 
5726         for (i = first_index ; i <= last_index; i++) {
5727                 if (total_read % ra->ra_pages == 0) {
5728                         btrfs_force_ra(inode->i_mapping, ra, NULL, i,
5729                                        calc_ra(i, last_index, ra->ra_pages));
5730                 }
5731                 total_read++;
5732 again:
5733                 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
5734                         BUG_ON(1);
5735                 page = grab_cache_page(inode->i_mapping, i);
5736                 if (!page) {
5737                         ret = -ENOMEM;
5738                         goto out_unlock;
5739                 }
5740                 if (!PageUptodate(page)) {
5741                         btrfs_readpage(NULL, page);
5742                         lock_page(page);
5743                         if (!PageUptodate(page)) {
5744                                 unlock_page(page);
5745                                 page_cache_release(page);
5746                                 ret = -EIO;
5747                                 goto out_unlock;
5748                         }
5749                 }
5750                 wait_on_page_writeback(page);
5751 
5752                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
5753                 page_end = page_start + PAGE_CACHE_SIZE - 1;
5754                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
5755 
5756                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
5757                 if (ordered) {
5758                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
5759                         unlock_page(page);
5760                         page_cache_release(page);
5761                         btrfs_start_ordered_extent(inode, ordered, 1);
5762                         btrfs_put_ordered_extent(ordered);
5763                         goto again;
5764                 }
5765                 set_page_extent_mapped(page);
5766 
5767                 if (i == first_index)
5768                         set_extent_bits(io_tree, page_start, page_end,
5769                                         EXTENT_BOUNDARY, GFP_NOFS);
5770                 btrfs_set_extent_delalloc(inode, page_start, page_end);
5771 
5772                 set_page_dirty(page);
5773                 total_dirty++;
5774 
5775                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
5776                 unlock_page(page);
5777                 page_cache_release(page);
5778         }
5779 
5780 out_unlock:
5781         kfree(ra);
5782         mutex_unlock(&inode->i_mutex);
5783         balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
5784         return ret;
5785 }
5786 
5787 static noinline int relocate_data_extent(struct inode *reloc_inode,
5788                                          struct btrfs_key *extent_key,
5789                                          u64 offset)
5790 {
5791         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
5792         struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
5793         struct extent_map *em;
5794         u64 start = extent_key->objectid - offset;
5795         u64 end = start + extent_key->offset - 1;
5796 
5797         em = alloc_extent_map(GFP_NOFS);
5798         BUG_ON(!em || IS_ERR(em));
5799 
5800         em->start = start;
5801         em->len = extent_key->offset;
5802         em->block_len = extent_key->offset;
5803         em->block_start = extent_key->objectid;
5804         em->bdev = root->fs_info->fs_devices->latest_bdev;
5805         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5806 
5807         /* setup extent map to cheat btrfs_readpage */
5808         lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
5809         while (1) {
5810                 int ret;
5811                 write_lock(&em_tree->lock);
5812                 ret = add_extent_mapping(em_tree, em);
5813                 write_unlock(&em_tree->lock);
5814                 if (ret != -EEXIST) {
5815                         free_extent_map(em);
5816                         break;
5817                 }
5818                 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
5819         }
5820         unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
5821 
5822         return relocate_inode_pages(reloc_inode, start, extent_key->offset);
5823 }
5824 
5825 struct btrfs_ref_path {
5826         u64 extent_start;
5827         u64 nodes[BTRFS_MAX_LEVEL];
5828         u64 root_objectid;
5829         u64 root_generation;
5830         u64 owner_objectid;
5831         u32 num_refs;
5832         int lowest_level;
5833         int current_level;
5834         int shared_level;
5835 
5836         struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
5837         u64 new_nodes[BTRFS_MAX_LEVEL];
5838 };
5839 
5840 struct disk_extent {
5841         u64 ram_bytes;
5842         u64 disk_bytenr;
5843         u64 disk_num_bytes;
5844         u64 offset;
5845         u64 num_bytes;
5846         u8 compression;
5847         u8 encryption;
5848         u16 other_encoding;
5849 };
5850 
5851 static int is_cowonly_root(u64 root_objectid)
5852 {
5853         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
5854             root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
5855             root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
5856             root_objectid == BTRFS_DEV_TREE_OBJECTID ||
5857             root_objectid == BTRFS_TREE_LOG_OBJECTID ||
5858             root_objectid == BTRFS_CSUM_TREE_OBJECTID)
5859                 return 1;
5860         return 0;
5861 }
5862 
5863 static noinline int __next_ref_path(struct btrfs_trans_handle *trans,
5864                                     struct btrfs_root *extent_root,
5865                                     struct btrfs_ref_path *ref_path,
5866                                     int first_time)
5867 {
5868         struct extent_buffer *leaf;
5869         struct btrfs_path *path;
5870         struct btrfs_extent_ref *ref;
5871         struct btrfs_key key;
5872         struct btrfs_key found_key;
5873         u64 bytenr;
5874         u32 nritems;
5875         int level;
5876         int ret = 1;
5877 
5878         path = btrfs_alloc_path();
5879         if (!path)
5880                 return -ENOMEM;
5881 
5882         if (first_time) {
5883                 ref_path->lowest_level = -1;
5884                 ref_path->current_level = -1;
5885                 ref_path->shared_level = -1;
5886                 goto walk_up;
5887         }
5888 walk_down:
5889         level = ref_path->current_level - 1;
5890         while (level >= -1) {
5891                 u64 parent;
5892                 if (level < ref_path->lowest_level)
5893                         break;
5894 
5895                 if (level >= 0)
5896                         bytenr = ref_path->nodes[level];
5897                 else
5898                         bytenr = ref_path->extent_start;
5899                 BUG_ON(bytenr == 0);
5900 
5901                 parent = ref_path->nodes[level + 1];
5902                 ref_path->nodes[level + 1] = 0;
5903                 ref_path->current_level = level;
5904                 BUG_ON(parent == 0);
5905 
5906                 key.objectid = bytenr;
5907                 key.offset = parent + 1;
5908                 key.type = BTRFS_EXTENT_REF_KEY;
5909 
5910                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
5911                 if (ret < 0)
5912                         goto out;
5913                 BUG_ON(ret == 0);
5914 
5915                 leaf = path->nodes[0];
5916                 nritems = btrfs_header_nritems(leaf);
5917                 if (path->slots[0] >= nritems) {
5918                         ret = btrfs_next_leaf(extent_root, path);
5919                         if (ret < 0)
5920                                 goto out;
5921                         if (ret > 0)
5922                                 goto next;
5923                         leaf = path->nodes[0];
5924                 }
5925 
5926                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5927                 if (found_key.objectid == bytenr &&
5928                     found_key.type == BTRFS_EXTENT_REF_KEY) {
5929                         if (level < ref_path->shared_level)
5930                                 ref_path->shared_level = level;
5931                         goto found;
5932                 }
5933 next:
5934                 level--;
5935                 btrfs_release_path(extent_root, path);
5936                 cond_resched();
5937         }
5938         /* reached lowest level */
5939         ret = 1;
5940         goto out;
5941 walk_up:
5942         level = ref_path->current_level;
5943         while (level < BTRFS_MAX_LEVEL - 1) {
5944                 u64 ref_objectid;
5945 
5946                 if (level >= 0)
5947                         bytenr = ref_path->nodes[level];
5948                 else
5949                         bytenr = ref_path->extent_start;
5950 
5951                 BUG_ON(bytenr == 0);
5952 
5953                 key.objectid = bytenr;
5954                 key.offset = 0;
5955                 key.type = BTRFS_EXTENT_REF_KEY;
5956 
5957                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);