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

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

Version: ~ [ linux-5.10-rc5 ] ~ [ linux-5.9.10 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.79 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.159 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.208 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.245 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.245 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.85 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ 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 <linux/slab.h>
 26 #include <linux/ratelimit.h>
 27 #include <linux/percpu_counter.h>
 28 #include "hash.h"
 29 #include "tree-log.h"
 30 #include "disk-io.h"
 31 #include "print-tree.h"
 32 #include "volumes.h"
 33 #include "raid56.h"
 34 #include "locking.h"
 35 #include "free-space-cache.h"
 36 #include "math.h"
 37 #include "sysfs.h"
 38 #include "qgroup.h"
 39 
 40 #undef SCRAMBLE_DELAYED_REFS
 41 
 42 /*
 43  * control flags for do_chunk_alloc's force field
 44  * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
 45  * if we really need one.
 46  *
 47  * CHUNK_ALLOC_LIMITED means to only try and allocate one
 48  * if we have very few chunks already allocated.  This is
 49  * used as part of the clustering code to help make sure
 50  * we have a good pool of storage to cluster in, without
 51  * filling the FS with empty chunks
 52  *
 53  * CHUNK_ALLOC_FORCE means it must try to allocate one
 54  *
 55  */
 56 enum {
 57         CHUNK_ALLOC_NO_FORCE = 0,
 58         CHUNK_ALLOC_LIMITED = 1,
 59         CHUNK_ALLOC_FORCE = 2,
 60 };
 61 
 62 /*
 63  * Control how reservations are dealt with.
 64  *
 65  * RESERVE_FREE - freeing a reservation.
 66  * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
 67  *   ENOSPC accounting
 68  * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
 69  *   bytes_may_use as the ENOSPC accounting is done elsewhere
 70  */
 71 enum {
 72         RESERVE_FREE = 0,
 73         RESERVE_ALLOC = 1,
 74         RESERVE_ALLOC_NO_ACCOUNT = 2,
 75 };
 76 
 77 static int update_block_group(struct btrfs_trans_handle *trans,
 78                               struct btrfs_root *root, u64 bytenr,
 79                               u64 num_bytes, int alloc);
 80 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
 81                                 struct btrfs_root *root,
 82                                 u64 bytenr, u64 num_bytes, u64 parent,
 83                                 u64 root_objectid, u64 owner_objectid,
 84                                 u64 owner_offset, int refs_to_drop,
 85                                 struct btrfs_delayed_extent_op *extra_op,
 86                                 int no_quota);
 87 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
 88                                     struct extent_buffer *leaf,
 89                                     struct btrfs_extent_item *ei);
 90 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
 91                                       struct btrfs_root *root,
 92                                       u64 parent, u64 root_objectid,
 93                                       u64 flags, u64 owner, u64 offset,
 94                                       struct btrfs_key *ins, int ref_mod);
 95 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
 96                                      struct btrfs_root *root,
 97                                      u64 parent, u64 root_objectid,
 98                                      u64 flags, struct btrfs_disk_key *key,
 99                                      int level, struct btrfs_key *ins,
100                                      int no_quota);
101 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
102                           struct btrfs_root *extent_root, u64 flags,
103                           int force);
104 static int find_next_key(struct btrfs_path *path, int level,
105                          struct btrfs_key *key);
106 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
107                             int dump_block_groups);
108 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
109                                        u64 num_bytes, int reserve,
110                                        int delalloc);
111 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
112                                u64 num_bytes);
113 int btrfs_pin_extent(struct btrfs_root *root,
114                      u64 bytenr, u64 num_bytes, int reserved);
115 
116 static noinline int
117 block_group_cache_done(struct btrfs_block_group_cache *cache)
118 {
119         smp_mb();
120         return cache->cached == BTRFS_CACHE_FINISHED ||
121                 cache->cached == BTRFS_CACHE_ERROR;
122 }
123 
124 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
125 {
126         return (cache->flags & bits) == bits;
127 }
128 
129 static void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
130 {
131         atomic_inc(&cache->count);
132 }
133 
134 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
135 {
136         if (atomic_dec_and_test(&cache->count)) {
137                 WARN_ON(cache->pinned > 0);
138                 WARN_ON(cache->reserved > 0);
139                 kfree(cache->free_space_ctl);
140                 kfree(cache);
141         }
142 }
143 
144 /*
145  * this adds the block group to the fs_info rb tree for the block group
146  * cache
147  */
148 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
149                                 struct btrfs_block_group_cache *block_group)
150 {
151         struct rb_node **p;
152         struct rb_node *parent = NULL;
153         struct btrfs_block_group_cache *cache;
154 
155         spin_lock(&info->block_group_cache_lock);
156         p = &info->block_group_cache_tree.rb_node;
157 
158         while (*p) {
159                 parent = *p;
160                 cache = rb_entry(parent, struct btrfs_block_group_cache,
161                                  cache_node);
162                 if (block_group->key.objectid < cache->key.objectid) {
163                         p = &(*p)->rb_left;
164                 } else if (block_group->key.objectid > cache->key.objectid) {
165                         p = &(*p)->rb_right;
166                 } else {
167                         spin_unlock(&info->block_group_cache_lock);
168                         return -EEXIST;
169                 }
170         }
171 
172         rb_link_node(&block_group->cache_node, parent, p);
173         rb_insert_color(&block_group->cache_node,
174                         &info->block_group_cache_tree);
175 
176         if (info->first_logical_byte > block_group->key.objectid)
177                 info->first_logical_byte = block_group->key.objectid;
178 
179         spin_unlock(&info->block_group_cache_lock);
180 
181         return 0;
182 }
183 
184 /*
185  * This will return the block group at or after bytenr if contains is 0, else
186  * it will return the block group that contains the bytenr
187  */
188 static struct btrfs_block_group_cache *
189 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
190                               int contains)
191 {
192         struct btrfs_block_group_cache *cache, *ret = NULL;
193         struct rb_node *n;
194         u64 end, start;
195 
196         spin_lock(&info->block_group_cache_lock);
197         n = info->block_group_cache_tree.rb_node;
198 
199         while (n) {
200                 cache = rb_entry(n, struct btrfs_block_group_cache,
201                                  cache_node);
202                 end = cache->key.objectid + cache->key.offset - 1;
203                 start = cache->key.objectid;
204 
205                 if (bytenr < start) {
206                         if (!contains && (!ret || start < ret->key.objectid))
207                                 ret = cache;
208                         n = n->rb_left;
209                 } else if (bytenr > start) {
210                         if (contains && bytenr <= end) {
211                                 ret = cache;
212                                 break;
213                         }
214                         n = n->rb_right;
215                 } else {
216                         ret = cache;
217                         break;
218                 }
219         }
220         if (ret) {
221                 btrfs_get_block_group(ret);
222                 if (bytenr == 0 && info->first_logical_byte > ret->key.objectid)
223                         info->first_logical_byte = ret->key.objectid;
224         }
225         spin_unlock(&info->block_group_cache_lock);
226 
227         return ret;
228 }
229 
230 static int add_excluded_extent(struct btrfs_root *root,
231                                u64 start, u64 num_bytes)
232 {
233         u64 end = start + num_bytes - 1;
234         set_extent_bits(&root->fs_info->freed_extents[0],
235                         start, end, EXTENT_UPTODATE, GFP_NOFS);
236         set_extent_bits(&root->fs_info->freed_extents[1],
237                         start, end, EXTENT_UPTODATE, GFP_NOFS);
238         return 0;
239 }
240 
241 static void free_excluded_extents(struct btrfs_root *root,
242                                   struct btrfs_block_group_cache *cache)
243 {
244         u64 start, end;
245 
246         start = cache->key.objectid;
247         end = start + cache->key.offset - 1;
248 
249         clear_extent_bits(&root->fs_info->freed_extents[0],
250                           start, end, EXTENT_UPTODATE, GFP_NOFS);
251         clear_extent_bits(&root->fs_info->freed_extents[1],
252                           start, end, EXTENT_UPTODATE, GFP_NOFS);
253 }
254 
255 static int exclude_super_stripes(struct btrfs_root *root,
256                                  struct btrfs_block_group_cache *cache)
257 {
258         u64 bytenr;
259         u64 *logical;
260         int stripe_len;
261         int i, nr, ret;
262 
263         if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
264                 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
265                 cache->bytes_super += stripe_len;
266                 ret = add_excluded_extent(root, cache->key.objectid,
267                                           stripe_len);
268                 if (ret)
269                         return ret;
270         }
271 
272         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
273                 bytenr = btrfs_sb_offset(i);
274                 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
275                                        cache->key.objectid, bytenr,
276                                        0, &logical, &nr, &stripe_len);
277                 if (ret)
278                         return ret;
279 
280                 while (nr--) {
281                         u64 start, len;
282 
283                         if (logical[nr] > cache->key.objectid +
284                             cache->key.offset)
285                                 continue;
286 
287                         if (logical[nr] + stripe_len <= cache->key.objectid)
288                                 continue;
289 
290                         start = logical[nr];
291                         if (start < cache->key.objectid) {
292                                 start = cache->key.objectid;
293                                 len = (logical[nr] + stripe_len) - start;
294                         } else {
295                                 len = min_t(u64, stripe_len,
296                                             cache->key.objectid +
297                                             cache->key.offset - start);
298                         }
299 
300                         cache->bytes_super += len;
301                         ret = add_excluded_extent(root, start, len);
302                         if (ret) {
303                                 kfree(logical);
304                                 return ret;
305                         }
306                 }
307 
308                 kfree(logical);
309         }
310         return 0;
311 }
312 
313 static struct btrfs_caching_control *
314 get_caching_control(struct btrfs_block_group_cache *cache)
315 {
316         struct btrfs_caching_control *ctl;
317 
318         spin_lock(&cache->lock);
319         if (!cache->caching_ctl) {
320                 spin_unlock(&cache->lock);
321                 return NULL;
322         }
323 
324         ctl = cache->caching_ctl;
325         atomic_inc(&ctl->count);
326         spin_unlock(&cache->lock);
327         return ctl;
328 }
329 
330 static void put_caching_control(struct btrfs_caching_control *ctl)
331 {
332         if (atomic_dec_and_test(&ctl->count))
333                 kfree(ctl);
334 }
335 
336 /*
337  * this is only called by cache_block_group, since we could have freed extents
338  * we need to check the pinned_extents for any extents that can't be used yet
339  * since their free space will be released as soon as the transaction commits.
340  */
341 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
342                               struct btrfs_fs_info *info, u64 start, u64 end)
343 {
344         u64 extent_start, extent_end, size, total_added = 0;
345         int ret;
346 
347         while (start < end) {
348                 ret = find_first_extent_bit(info->pinned_extents, start,
349                                             &extent_start, &extent_end,
350                                             EXTENT_DIRTY | EXTENT_UPTODATE,
351                                             NULL);
352                 if (ret)
353                         break;
354 
355                 if (extent_start <= start) {
356                         start = extent_end + 1;
357                 } else if (extent_start > start && extent_start < end) {
358                         size = extent_start - start;
359                         total_added += size;
360                         ret = btrfs_add_free_space(block_group, start,
361                                                    size);
362                         BUG_ON(ret); /* -ENOMEM or logic error */
363                         start = extent_end + 1;
364                 } else {
365                         break;
366                 }
367         }
368 
369         if (start < end) {
370                 size = end - start;
371                 total_added += size;
372                 ret = btrfs_add_free_space(block_group, start, size);
373                 BUG_ON(ret); /* -ENOMEM or logic error */
374         }
375 
376         return total_added;
377 }
378 
379 static noinline void caching_thread(struct btrfs_work *work)
380 {
381         struct btrfs_block_group_cache *block_group;
382         struct btrfs_fs_info *fs_info;
383         struct btrfs_caching_control *caching_ctl;
384         struct btrfs_root *extent_root;
385         struct btrfs_path *path;
386         struct extent_buffer *leaf;
387         struct btrfs_key key;
388         u64 total_found = 0;
389         u64 last = 0;
390         u32 nritems;
391         int ret = -ENOMEM;
392 
393         caching_ctl = container_of(work, struct btrfs_caching_control, work);
394         block_group = caching_ctl->block_group;
395         fs_info = block_group->fs_info;
396         extent_root = fs_info->extent_root;
397 
398         path = btrfs_alloc_path();
399         if (!path)
400                 goto out;
401 
402         last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
403 
404         /*
405          * We don't want to deadlock with somebody trying to allocate a new
406          * extent for the extent root while also trying to search the extent
407          * root to add free space.  So we skip locking and search the commit
408          * root, since its read-only
409          */
410         path->skip_locking = 1;
411         path->search_commit_root = 1;
412         path->reada = 1;
413 
414         key.objectid = last;
415         key.offset = 0;
416         key.type = BTRFS_EXTENT_ITEM_KEY;
417 again:
418         mutex_lock(&caching_ctl->mutex);
419         /* need to make sure the commit_root doesn't disappear */
420         down_read(&fs_info->commit_root_sem);
421 
422 next:
423         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
424         if (ret < 0)
425                 goto err;
426 
427         leaf = path->nodes[0];
428         nritems = btrfs_header_nritems(leaf);
429 
430         while (1) {
431                 if (btrfs_fs_closing(fs_info) > 1) {
432                         last = (u64)-1;
433                         break;
434                 }
435 
436                 if (path->slots[0] < nritems) {
437                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
438                 } else {
439                         ret = find_next_key(path, 0, &key);
440                         if (ret)
441                                 break;
442 
443                         if (need_resched() ||
444                             rwsem_is_contended(&fs_info->commit_root_sem)) {
445                                 caching_ctl->progress = last;
446                                 btrfs_release_path(path);
447                                 up_read(&fs_info->commit_root_sem);
448                                 mutex_unlock(&caching_ctl->mutex);
449                                 cond_resched();
450                                 goto again;
451                         }
452 
453                         ret = btrfs_next_leaf(extent_root, path);
454                         if (ret < 0)
455                                 goto err;
456                         if (ret)
457                                 break;
458                         leaf = path->nodes[0];
459                         nritems = btrfs_header_nritems(leaf);
460                         continue;
461                 }
462 
463                 if (key.objectid < last) {
464                         key.objectid = last;
465                         key.offset = 0;
466                         key.type = BTRFS_EXTENT_ITEM_KEY;
467 
468                         caching_ctl->progress = last;
469                         btrfs_release_path(path);
470                         goto next;
471                 }
472 
473                 if (key.objectid < block_group->key.objectid) {
474                         path->slots[0]++;
475                         continue;
476                 }
477 
478                 if (key.objectid >= block_group->key.objectid +
479                     block_group->key.offset)
480                         break;
481 
482                 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
483                     key.type == BTRFS_METADATA_ITEM_KEY) {
484                         total_found += add_new_free_space(block_group,
485                                                           fs_info, last,
486                                                           key.objectid);
487                         if (key.type == BTRFS_METADATA_ITEM_KEY)
488                                 last = key.objectid +
489                                         fs_info->tree_root->nodesize;
490                         else
491                                 last = key.objectid + key.offset;
492 
493                         if (total_found > (1024 * 1024 * 2)) {
494                                 total_found = 0;
495                                 wake_up(&caching_ctl->wait);
496                         }
497                 }
498                 path->slots[0]++;
499         }
500         ret = 0;
501 
502         total_found += add_new_free_space(block_group, fs_info, last,
503                                           block_group->key.objectid +
504                                           block_group->key.offset);
505         caching_ctl->progress = (u64)-1;
506 
507         spin_lock(&block_group->lock);
508         block_group->caching_ctl = NULL;
509         block_group->cached = BTRFS_CACHE_FINISHED;
510         spin_unlock(&block_group->lock);
511 
512 err:
513         btrfs_free_path(path);
514         up_read(&fs_info->commit_root_sem);
515 
516         free_excluded_extents(extent_root, block_group);
517 
518         mutex_unlock(&caching_ctl->mutex);
519 out:
520         if (ret) {
521                 spin_lock(&block_group->lock);
522                 block_group->caching_ctl = NULL;
523                 block_group->cached = BTRFS_CACHE_ERROR;
524                 spin_unlock(&block_group->lock);
525         }
526         wake_up(&caching_ctl->wait);
527 
528         put_caching_control(caching_ctl);
529         btrfs_put_block_group(block_group);
530 }
531 
532 static int cache_block_group(struct btrfs_block_group_cache *cache,
533                              int load_cache_only)
534 {
535         DEFINE_WAIT(wait);
536         struct btrfs_fs_info *fs_info = cache->fs_info;
537         struct btrfs_caching_control *caching_ctl;
538         int ret = 0;
539 
540         caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
541         if (!caching_ctl)
542                 return -ENOMEM;
543 
544         INIT_LIST_HEAD(&caching_ctl->list);
545         mutex_init(&caching_ctl->mutex);
546         init_waitqueue_head(&caching_ctl->wait);
547         caching_ctl->block_group = cache;
548         caching_ctl->progress = cache->key.objectid;
549         atomic_set(&caching_ctl->count, 1);
550         btrfs_init_work(&caching_ctl->work, btrfs_cache_helper,
551                         caching_thread, NULL, NULL);
552 
553         spin_lock(&cache->lock);
554         /*
555          * This should be a rare occasion, but this could happen I think in the
556          * case where one thread starts to load the space cache info, and then
557          * some other thread starts a transaction commit which tries to do an
558          * allocation while the other thread is still loading the space cache
559          * info.  The previous loop should have kept us from choosing this block
560          * group, but if we've moved to the state where we will wait on caching
561          * block groups we need to first check if we're doing a fast load here,
562          * so we can wait for it to finish, otherwise we could end up allocating
563          * from a block group who's cache gets evicted for one reason or
564          * another.
565          */
566         while (cache->cached == BTRFS_CACHE_FAST) {
567                 struct btrfs_caching_control *ctl;
568 
569                 ctl = cache->caching_ctl;
570                 atomic_inc(&ctl->count);
571                 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
572                 spin_unlock(&cache->lock);
573 
574                 schedule();
575 
576                 finish_wait(&ctl->wait, &wait);
577                 put_caching_control(ctl);
578                 spin_lock(&cache->lock);
579         }
580 
581         if (cache->cached != BTRFS_CACHE_NO) {
582                 spin_unlock(&cache->lock);
583                 kfree(caching_ctl);
584                 return 0;
585         }
586         WARN_ON(cache->caching_ctl);
587         cache->caching_ctl = caching_ctl;
588         cache->cached = BTRFS_CACHE_FAST;
589         spin_unlock(&cache->lock);
590 
591         if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) {
592                 mutex_lock(&caching_ctl->mutex);
593                 ret = load_free_space_cache(fs_info, cache);
594 
595                 spin_lock(&cache->lock);
596                 if (ret == 1) {
597                         cache->caching_ctl = NULL;
598                         cache->cached = BTRFS_CACHE_FINISHED;
599                         cache->last_byte_to_unpin = (u64)-1;
600                         caching_ctl->progress = (u64)-1;
601                 } else {
602                         if (load_cache_only) {
603                                 cache->caching_ctl = NULL;
604                                 cache->cached = BTRFS_CACHE_NO;
605                         } else {
606                                 cache->cached = BTRFS_CACHE_STARTED;
607                                 cache->has_caching_ctl = 1;
608                         }
609                 }
610                 spin_unlock(&cache->lock);
611                 mutex_unlock(&caching_ctl->mutex);
612 
613                 wake_up(&caching_ctl->wait);
614                 if (ret == 1) {
615                         put_caching_control(caching_ctl);
616                         free_excluded_extents(fs_info->extent_root, cache);
617                         return 0;
618                 }
619         } else {
620                 /*
621                  * We are not going to do the fast caching, set cached to the
622                  * appropriate value and wakeup any waiters.
623                  */
624                 spin_lock(&cache->lock);
625                 if (load_cache_only) {
626                         cache->caching_ctl = NULL;
627                         cache->cached = BTRFS_CACHE_NO;
628                 } else {
629                         cache->cached = BTRFS_CACHE_STARTED;
630                         cache->has_caching_ctl = 1;
631                 }
632                 spin_unlock(&cache->lock);
633                 wake_up(&caching_ctl->wait);
634         }
635 
636         if (load_cache_only) {
637                 put_caching_control(caching_ctl);
638                 return 0;
639         }
640 
641         down_write(&fs_info->commit_root_sem);
642         atomic_inc(&caching_ctl->count);
643         list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
644         up_write(&fs_info->commit_root_sem);
645 
646         btrfs_get_block_group(cache);
647 
648         btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work);
649 
650         return ret;
651 }
652 
653 /*
654  * return the block group that starts at or after bytenr
655  */
656 static struct btrfs_block_group_cache *
657 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
658 {
659         struct btrfs_block_group_cache *cache;
660 
661         cache = block_group_cache_tree_search(info, bytenr, 0);
662 
663         return cache;
664 }
665 
666 /*
667  * return the block group that contains the given bytenr
668  */
669 struct btrfs_block_group_cache *btrfs_lookup_block_group(
670                                                  struct btrfs_fs_info *info,
671                                                  u64 bytenr)
672 {
673         struct btrfs_block_group_cache *cache;
674 
675         cache = block_group_cache_tree_search(info, bytenr, 1);
676 
677         return cache;
678 }
679 
680 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
681                                                   u64 flags)
682 {
683         struct list_head *head = &info->space_info;
684         struct btrfs_space_info *found;
685 
686         flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
687 
688         rcu_read_lock();
689         list_for_each_entry_rcu(found, head, list) {
690                 if (found->flags & flags) {
691                         rcu_read_unlock();
692                         return found;
693                 }
694         }
695         rcu_read_unlock();
696         return NULL;
697 }
698 
699 /*
700  * after adding space to the filesystem, we need to clear the full flags
701  * on all the space infos.
702  */
703 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
704 {
705         struct list_head *head = &info->space_info;
706         struct btrfs_space_info *found;
707 
708         rcu_read_lock();
709         list_for_each_entry_rcu(found, head, list)
710                 found->full = 0;
711         rcu_read_unlock();
712 }
713 
714 /* simple helper to search for an existing data extent at a given offset */
715 int btrfs_lookup_data_extent(struct btrfs_root *root, u64 start, u64 len)
716 {
717         int ret;
718         struct btrfs_key key;
719         struct btrfs_path *path;
720 
721         path = btrfs_alloc_path();
722         if (!path)
723                 return -ENOMEM;
724 
725         key.objectid = start;
726         key.offset = len;
727         key.type = BTRFS_EXTENT_ITEM_KEY;
728         ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
729                                 0, 0);
730         btrfs_free_path(path);
731         return ret;
732 }
733 
734 /*
735  * helper function to lookup reference count and flags of a tree block.
736  *
737  * the head node for delayed ref is used to store the sum of all the
738  * reference count modifications queued up in the rbtree. the head
739  * node may also store the extent flags to set. This way you can check
740  * to see what the reference count and extent flags would be if all of
741  * the delayed refs are not processed.
742  */
743 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
744                              struct btrfs_root *root, u64 bytenr,
745                              u64 offset, int metadata, u64 *refs, u64 *flags)
746 {
747         struct btrfs_delayed_ref_head *head;
748         struct btrfs_delayed_ref_root *delayed_refs;
749         struct btrfs_path *path;
750         struct btrfs_extent_item *ei;
751         struct extent_buffer *leaf;
752         struct btrfs_key key;
753         u32 item_size;
754         u64 num_refs;
755         u64 extent_flags;
756         int ret;
757 
758         /*
759          * If we don't have skinny metadata, don't bother doing anything
760          * different
761          */
762         if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) {
763                 offset = root->nodesize;
764                 metadata = 0;
765         }
766 
767         path = btrfs_alloc_path();
768         if (!path)
769                 return -ENOMEM;
770 
771         if (!trans) {
772                 path->skip_locking = 1;
773                 path->search_commit_root = 1;
774         }
775 
776 search_again:
777         key.objectid = bytenr;
778         key.offset = offset;
779         if (metadata)
780                 key.type = BTRFS_METADATA_ITEM_KEY;
781         else
782                 key.type = BTRFS_EXTENT_ITEM_KEY;
783 
784         ret = btrfs_search_slot(trans, root->fs_info->extent_root,
785                                 &key, path, 0, 0);
786         if (ret < 0)
787                 goto out_free;
788 
789         if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
790                 if (path->slots[0]) {
791                         path->slots[0]--;
792                         btrfs_item_key_to_cpu(path->nodes[0], &key,
793                                               path->slots[0]);
794                         if (key.objectid == bytenr &&
795                             key.type == BTRFS_EXTENT_ITEM_KEY &&
796                             key.offset == root->nodesize)
797                                 ret = 0;
798                 }
799         }
800 
801         if (ret == 0) {
802                 leaf = path->nodes[0];
803                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
804                 if (item_size >= sizeof(*ei)) {
805                         ei = btrfs_item_ptr(leaf, path->slots[0],
806                                             struct btrfs_extent_item);
807                         num_refs = btrfs_extent_refs(leaf, ei);
808                         extent_flags = btrfs_extent_flags(leaf, ei);
809                 } else {
810 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
811                         struct btrfs_extent_item_v0 *ei0;
812                         BUG_ON(item_size != sizeof(*ei0));
813                         ei0 = btrfs_item_ptr(leaf, path->slots[0],
814                                              struct btrfs_extent_item_v0);
815                         num_refs = btrfs_extent_refs_v0(leaf, ei0);
816                         /* FIXME: this isn't correct for data */
817                         extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
818 #else
819                         BUG();
820 #endif
821                 }
822                 BUG_ON(num_refs == 0);
823         } else {
824                 num_refs = 0;
825                 extent_flags = 0;
826                 ret = 0;
827         }
828 
829         if (!trans)
830                 goto out;
831 
832         delayed_refs = &trans->transaction->delayed_refs;
833         spin_lock(&delayed_refs->lock);
834         head = btrfs_find_delayed_ref_head(trans, bytenr);
835         if (head) {
836                 if (!mutex_trylock(&head->mutex)) {
837                         atomic_inc(&head->node.refs);
838                         spin_unlock(&delayed_refs->lock);
839 
840                         btrfs_release_path(path);
841 
842                         /*
843                          * Mutex was contended, block until it's released and try
844                          * again
845                          */
846                         mutex_lock(&head->mutex);
847                         mutex_unlock(&head->mutex);
848                         btrfs_put_delayed_ref(&head->node);
849                         goto search_again;
850                 }
851                 spin_lock(&head->lock);
852                 if (head->extent_op && head->extent_op->update_flags)
853                         extent_flags |= head->extent_op->flags_to_set;
854                 else
855                         BUG_ON(num_refs == 0);
856 
857                 num_refs += head->node.ref_mod;
858                 spin_unlock(&head->lock);
859                 mutex_unlock(&head->mutex);
860         }
861         spin_unlock(&delayed_refs->lock);
862 out:
863         WARN_ON(num_refs == 0);
864         if (refs)
865                 *refs = num_refs;
866         if (flags)
867                 *flags = extent_flags;
868 out_free:
869         btrfs_free_path(path);
870         return ret;
871 }
872 
873 /*
874  * Back reference rules.  Back refs have three main goals:
875  *
876  * 1) differentiate between all holders of references to an extent so that
877  *    when a reference is dropped we can make sure it was a valid reference
878  *    before freeing the extent.
879  *
880  * 2) Provide enough information to quickly find the holders of an extent
881  *    if we notice a given block is corrupted or bad.
882  *
883  * 3) Make it easy to migrate blocks for FS shrinking or storage pool
884  *    maintenance.  This is actually the same as #2, but with a slightly
885  *    different use case.
886  *
887  * There are two kinds of back refs. The implicit back refs is optimized
888  * for pointers in non-shared tree blocks. For a given pointer in a block,
889  * back refs of this kind provide information about the block's owner tree
890  * and the pointer's key. These information allow us to find the block by
891  * b-tree searching. The full back refs is for pointers in tree blocks not
892  * referenced by their owner trees. The location of tree block is recorded
893  * in the back refs. Actually the full back refs is generic, and can be
894  * used in all cases the implicit back refs is used. The major shortcoming
895  * of the full back refs is its overhead. Every time a tree block gets
896  * COWed, we have to update back refs entry for all pointers in it.
897  *
898  * For a newly allocated tree block, we use implicit back refs for
899  * pointers in it. This means most tree related operations only involve
900  * implicit back refs. For a tree block created in old transaction, the
901  * only way to drop a reference to it is COW it. So we can detect the
902  * event that tree block loses its owner tree's reference and do the
903  * back refs conversion.
904  *
905  * When a tree block is COW'd through a tree, there are four cases:
906  *
907  * The reference count of the block is one and the tree is the block's
908  * owner tree. Nothing to do in this case.
909  *
910  * The reference count of the block is one and the tree is not the
911  * block's owner tree. In this case, full back refs is used for pointers
912  * in the block. Remove these full back refs, add implicit back refs for
913  * every pointers in the new block.
914  *
915  * The reference count of the block is greater than one and the tree is
916  * the block's owner tree. In this case, implicit back refs is used for
917  * pointers in the block. Add full back refs for every pointers in the
918  * block, increase lower level extents' reference counts. The original
919  * implicit back refs are entailed to the new block.
920  *
921  * The reference count of the block is greater than one and the tree is
922  * not the block's owner tree. Add implicit back refs for every pointer in
923  * the new block, increase lower level extents' reference count.
924  *
925  * Back Reference Key composing:
926  *
927  * The key objectid corresponds to the first byte in the extent,
928  * The key type is used to differentiate between types of back refs.
929  * There are different meanings of the key offset for different types
930  * of back refs.
931  *
932  * File extents can be referenced by:
933  *
934  * - multiple snapshots, subvolumes, or different generations in one subvol
935  * - different files inside a single subvolume
936  * - different offsets inside a file (bookend extents in file.c)
937  *
938  * The extent ref structure for the implicit back refs has fields for:
939  *
940  * - Objectid of the subvolume root
941  * - objectid of the file holding the reference
942  * - original offset in the file
943  * - how many bookend extents
944  *
945  * The key offset for the implicit back refs is hash of the first
946  * three fields.
947  *
948  * The extent ref structure for the full back refs has field for:
949  *
950  * - number of pointers in the tree leaf
951  *
952  * The key offset for the implicit back refs is the first byte of
953  * the tree leaf
954  *
955  * When a file extent is allocated, The implicit back refs is used.
956  * the fields are filled in:
957  *
958  *     (root_key.objectid, inode objectid, offset in file, 1)
959  *
960  * When a file extent is removed file truncation, we find the
961  * corresponding implicit back refs and check the following fields:
962  *
963  *     (btrfs_header_owner(leaf), inode objectid, offset in file)
964  *
965  * Btree extents can be referenced by:
966  *
967  * - Different subvolumes
968  *
969  * Both the implicit back refs and the full back refs for tree blocks
970  * only consist of key. The key offset for the implicit back refs is
971  * objectid of block's owner tree. The key offset for the full back refs
972  * is the first byte of parent block.
973  *
974  * When implicit back refs is used, information about the lowest key and
975  * level of the tree block are required. These information are stored in
976  * tree block info structure.
977  */
978 
979 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
980 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
981                                   struct btrfs_root *root,
982                                   struct btrfs_path *path,
983                                   u64 owner, u32 extra_size)
984 {
985         struct btrfs_extent_item *item;
986         struct btrfs_extent_item_v0 *ei0;
987         struct btrfs_extent_ref_v0 *ref0;
988         struct btrfs_tree_block_info *bi;
989         struct extent_buffer *leaf;
990         struct btrfs_key key;
991         struct btrfs_key found_key;
992         u32 new_size = sizeof(*item);
993         u64 refs;
994         int ret;
995 
996         leaf = path->nodes[0];
997         BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
998 
999         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1000         ei0 = btrfs_item_ptr(leaf, path->slots[0],
1001                              struct btrfs_extent_item_v0);
1002         refs = btrfs_extent_refs_v0(leaf, ei0);
1003 
1004         if (owner == (u64)-1) {
1005                 while (1) {
1006                         if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1007                                 ret = btrfs_next_leaf(root, path);
1008                                 if (ret < 0)
1009                                         return ret;
1010                                 BUG_ON(ret > 0); /* Corruption */
1011                                 leaf = path->nodes[0];
1012                         }
1013                         btrfs_item_key_to_cpu(leaf, &found_key,
1014                                               path->slots[0]);
1015                         BUG_ON(key.objectid != found_key.objectid);
1016                         if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
1017                                 path->slots[0]++;
1018                                 continue;
1019                         }
1020                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
1021                                               struct btrfs_extent_ref_v0);
1022                         owner = btrfs_ref_objectid_v0(leaf, ref0);
1023                         break;
1024                 }
1025         }
1026         btrfs_release_path(path);
1027 
1028         if (owner < BTRFS_FIRST_FREE_OBJECTID)
1029                 new_size += sizeof(*bi);
1030 
1031         new_size -= sizeof(*ei0);
1032         ret = btrfs_search_slot(trans, root, &key, path,
1033                                 new_size + extra_size, 1);
1034         if (ret < 0)
1035                 return ret;
1036         BUG_ON(ret); /* Corruption */
1037 
1038         btrfs_extend_item(root, path, new_size);
1039 
1040         leaf = path->nodes[0];
1041         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1042         btrfs_set_extent_refs(leaf, item, refs);
1043         /* FIXME: get real generation */
1044         btrfs_set_extent_generation(leaf, item, 0);
1045         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1046                 btrfs_set_extent_flags(leaf, item,
1047                                        BTRFS_EXTENT_FLAG_TREE_BLOCK |
1048                                        BTRFS_BLOCK_FLAG_FULL_BACKREF);
1049                 bi = (struct btrfs_tree_block_info *)(item + 1);
1050                 /* FIXME: get first key of the block */
1051                 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
1052                 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1053         } else {
1054                 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1055         }
1056         btrfs_mark_buffer_dirty(leaf);
1057         return 0;
1058 }
1059 #endif
1060 
1061 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1062 {
1063         u32 high_crc = ~(u32)0;
1064         u32 low_crc = ~(u32)0;
1065         __le64 lenum;
1066 
1067         lenum = cpu_to_le64(root_objectid);
1068         high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum));
1069         lenum = cpu_to_le64(owner);
1070         low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1071         lenum = cpu_to_le64(offset);
1072         low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1073 
1074         return ((u64)high_crc << 31) ^ (u64)low_crc;
1075 }
1076 
1077 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1078                                      struct btrfs_extent_data_ref *ref)
1079 {
1080         return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1081                                     btrfs_extent_data_ref_objectid(leaf, ref),
1082                                     btrfs_extent_data_ref_offset(leaf, ref));
1083 }
1084 
1085 static int match_extent_data_ref(struct extent_buffer *leaf,
1086                                  struct btrfs_extent_data_ref *ref,
1087                                  u64 root_objectid, u64 owner, u64 offset)
1088 {
1089         if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1090             btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1091             btrfs_extent_data_ref_offset(leaf, ref) != offset)
1092                 return 0;
1093         return 1;
1094 }
1095 
1096 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1097                                            struct btrfs_root *root,
1098                                            struct btrfs_path *path,
1099                                            u64 bytenr, u64 parent,
1100                                            u64 root_objectid,
1101                                            u64 owner, u64 offset)
1102 {
1103         struct btrfs_key key;
1104         struct btrfs_extent_data_ref *ref;
1105         struct extent_buffer *leaf;
1106         u32 nritems;
1107         int ret;
1108         int recow;
1109         int err = -ENOENT;
1110 
1111         key.objectid = bytenr;
1112         if (parent) {
1113                 key.type = BTRFS_SHARED_DATA_REF_KEY;
1114                 key.offset = parent;
1115         } else {
1116                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1117                 key.offset = hash_extent_data_ref(root_objectid,
1118                                                   owner, offset);
1119         }
1120 again:
1121         recow = 0;
1122         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1123         if (ret < 0) {
1124                 err = ret;
1125                 goto fail;
1126         }
1127 
1128         if (parent) {
1129                 if (!ret)
1130                         return 0;
1131 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1132                 key.type = BTRFS_EXTENT_REF_V0_KEY;
1133                 btrfs_release_path(path);
1134                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1135                 if (ret < 0) {
1136                         err = ret;
1137                         goto fail;
1138                 }
1139                 if (!ret)
1140                         return 0;
1141 #endif
1142                 goto fail;
1143         }
1144 
1145         leaf = path->nodes[0];
1146         nritems = btrfs_header_nritems(leaf);
1147         while (1) {
1148                 if (path->slots[0] >= nritems) {
1149                         ret = btrfs_next_leaf(root, path);
1150                         if (ret < 0)
1151                                 err = ret;
1152                         if (ret)
1153                                 goto fail;
1154 
1155                         leaf = path->nodes[0];
1156                         nritems = btrfs_header_nritems(leaf);
1157                         recow = 1;
1158                 }
1159 
1160                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1161                 if (key.objectid != bytenr ||
1162                     key.type != BTRFS_EXTENT_DATA_REF_KEY)
1163                         goto fail;
1164 
1165                 ref = btrfs_item_ptr(leaf, path->slots[0],
1166                                      struct btrfs_extent_data_ref);
1167 
1168                 if (match_extent_data_ref(leaf, ref, root_objectid,
1169                                           owner, offset)) {
1170                         if (recow) {
1171                                 btrfs_release_path(path);
1172                                 goto again;
1173                         }
1174                         err = 0;
1175                         break;
1176                 }
1177                 path->slots[0]++;
1178         }
1179 fail:
1180         return err;
1181 }
1182 
1183 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1184                                            struct btrfs_root *root,
1185                                            struct btrfs_path *path,
1186                                            u64 bytenr, u64 parent,
1187                                            u64 root_objectid, u64 owner,
1188                                            u64 offset, int refs_to_add)
1189 {
1190         struct btrfs_key key;
1191         struct extent_buffer *leaf;
1192         u32 size;
1193         u32 num_refs;
1194         int ret;
1195 
1196         key.objectid = bytenr;
1197         if (parent) {
1198                 key.type = BTRFS_SHARED_DATA_REF_KEY;
1199                 key.offset = parent;
1200                 size = sizeof(struct btrfs_shared_data_ref);
1201         } else {
1202                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1203                 key.offset = hash_extent_data_ref(root_objectid,
1204                                                   owner, offset);
1205                 size = sizeof(struct btrfs_extent_data_ref);
1206         }
1207 
1208         ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1209         if (ret && ret != -EEXIST)
1210                 goto fail;
1211 
1212         leaf = path->nodes[0];
1213         if (parent) {
1214                 struct btrfs_shared_data_ref *ref;
1215                 ref = btrfs_item_ptr(leaf, path->slots[0],
1216                                      struct btrfs_shared_data_ref);
1217                 if (ret == 0) {
1218                         btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1219                 } else {
1220                         num_refs = btrfs_shared_data_ref_count(leaf, ref);
1221                         num_refs += refs_to_add;
1222                         btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1223                 }
1224         } else {
1225                 struct btrfs_extent_data_ref *ref;
1226                 while (ret == -EEXIST) {
1227                         ref = btrfs_item_ptr(leaf, path->slots[0],
1228                                              struct btrfs_extent_data_ref);
1229                         if (match_extent_data_ref(leaf, ref, root_objectid,
1230                                                   owner, offset))
1231                                 break;
1232                         btrfs_release_path(path);
1233                         key.offset++;
1234                         ret = btrfs_insert_empty_item(trans, root, path, &key,
1235                                                       size);
1236                         if (ret && ret != -EEXIST)
1237                                 goto fail;
1238 
1239                         leaf = path->nodes[0];
1240                 }
1241                 ref = btrfs_item_ptr(leaf, path->slots[0],
1242                                      struct btrfs_extent_data_ref);
1243                 if (ret == 0) {
1244                         btrfs_set_extent_data_ref_root(leaf, ref,
1245                                                        root_objectid);
1246                         btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1247                         btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1248                         btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1249                 } else {
1250                         num_refs = btrfs_extent_data_ref_count(leaf, ref);
1251                         num_refs += refs_to_add;
1252                         btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1253                 }
1254         }
1255         btrfs_mark_buffer_dirty(leaf);
1256         ret = 0;
1257 fail:
1258         btrfs_release_path(path);
1259         return ret;
1260 }
1261 
1262 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1263                                            struct btrfs_root *root,
1264                                            struct btrfs_path *path,
1265                                            int refs_to_drop, int *last_ref)
1266 {
1267         struct btrfs_key key;
1268         struct btrfs_extent_data_ref *ref1 = NULL;
1269         struct btrfs_shared_data_ref *ref2 = NULL;
1270         struct extent_buffer *leaf;
1271         u32 num_refs = 0;
1272         int ret = 0;
1273 
1274         leaf = path->nodes[0];
1275         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1276 
1277         if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1278                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1279                                       struct btrfs_extent_data_ref);
1280                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1281         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1282                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1283                                       struct btrfs_shared_data_ref);
1284                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1285 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1286         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1287                 struct btrfs_extent_ref_v0 *ref0;
1288                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1289                                       struct btrfs_extent_ref_v0);
1290                 num_refs = btrfs_ref_count_v0(leaf, ref0);
1291 #endif
1292         } else {
1293                 BUG();
1294         }
1295 
1296         BUG_ON(num_refs < refs_to_drop);
1297         num_refs -= refs_to_drop;
1298 
1299         if (num_refs == 0) {
1300                 ret = btrfs_del_item(trans, root, path);
1301                 *last_ref = 1;
1302         } else {
1303                 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1304                         btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1305                 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1306                         btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1307 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1308                 else {
1309                         struct btrfs_extent_ref_v0 *ref0;
1310                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
1311                                         struct btrfs_extent_ref_v0);
1312                         btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1313                 }
1314 #endif
1315                 btrfs_mark_buffer_dirty(leaf);
1316         }
1317         return ret;
1318 }
1319 
1320 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1321                                           struct btrfs_path *path,
1322                                           struct btrfs_extent_inline_ref *iref)
1323 {
1324         struct btrfs_key key;
1325         struct extent_buffer *leaf;
1326         struct btrfs_extent_data_ref *ref1;
1327         struct btrfs_shared_data_ref *ref2;
1328         u32 num_refs = 0;
1329 
1330         leaf = path->nodes[0];
1331         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1332         if (iref) {
1333                 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1334                     BTRFS_EXTENT_DATA_REF_KEY) {
1335                         ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1336                         num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1337                 } else {
1338                         ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1339                         num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1340                 }
1341         } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1342                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1343                                       struct btrfs_extent_data_ref);
1344                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1345         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1346                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1347                                       struct btrfs_shared_data_ref);
1348                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1349 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1350         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1351                 struct btrfs_extent_ref_v0 *ref0;
1352                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1353                                       struct btrfs_extent_ref_v0);
1354                 num_refs = btrfs_ref_count_v0(leaf, ref0);
1355 #endif
1356         } else {
1357                 WARN_ON(1);
1358         }
1359         return num_refs;
1360 }
1361 
1362 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1363                                           struct btrfs_root *root,
1364                                           struct btrfs_path *path,
1365                                           u64 bytenr, u64 parent,
1366                                           u64 root_objectid)
1367 {
1368         struct btrfs_key key;
1369         int ret;
1370 
1371         key.objectid = bytenr;
1372         if (parent) {
1373                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1374                 key.offset = parent;
1375         } else {
1376                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1377                 key.offset = root_objectid;
1378         }
1379 
1380         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1381         if (ret > 0)
1382                 ret = -ENOENT;
1383 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1384         if (ret == -ENOENT && parent) {
1385                 btrfs_release_path(path);
1386                 key.type = BTRFS_EXTENT_REF_V0_KEY;
1387                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1388                 if (ret > 0)
1389                         ret = -ENOENT;
1390         }
1391 #endif
1392         return ret;
1393 }
1394 
1395 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1396                                           struct btrfs_root *root,
1397                                           struct btrfs_path *path,
1398                                           u64 bytenr, u64 parent,
1399                                           u64 root_objectid)
1400 {
1401         struct btrfs_key key;
1402         int ret;
1403 
1404         key.objectid = bytenr;
1405         if (parent) {
1406                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1407                 key.offset = parent;
1408         } else {
1409                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1410                 key.offset = root_objectid;
1411         }
1412 
1413         ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1414         btrfs_release_path(path);
1415         return ret;
1416 }
1417 
1418 static inline int extent_ref_type(u64 parent, u64 owner)
1419 {
1420         int type;
1421         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1422                 if (parent > 0)
1423                         type = BTRFS_SHARED_BLOCK_REF_KEY;
1424                 else
1425                         type = BTRFS_TREE_BLOCK_REF_KEY;
1426         } else {
1427                 if (parent > 0)
1428                         type = BTRFS_SHARED_DATA_REF_KEY;
1429                 else
1430                         type = BTRFS_EXTENT_DATA_REF_KEY;
1431         }
1432         return type;
1433 }
1434 
1435 static int find_next_key(struct btrfs_path *path, int level,
1436                          struct btrfs_key *key)
1437 
1438 {
1439         for (; level < BTRFS_MAX_LEVEL; level++) {
1440                 if (!path->nodes[level])
1441                         break;
1442                 if (path->slots[level] + 1 >=
1443                     btrfs_header_nritems(path->nodes[level]))
1444                         continue;
1445                 if (level == 0)
1446                         btrfs_item_key_to_cpu(path->nodes[level], key,
1447                                               path->slots[level] + 1);
1448                 else
1449                         btrfs_node_key_to_cpu(path->nodes[level], key,
1450                                               path->slots[level] + 1);
1451                 return 0;
1452         }
1453         return 1;
1454 }
1455 
1456 /*
1457  * look for inline back ref. if back ref is found, *ref_ret is set
1458  * to the address of inline back ref, and 0 is returned.
1459  *
1460  * if back ref isn't found, *ref_ret is set to the address where it
1461  * should be inserted, and -ENOENT is returned.
1462  *
1463  * if insert is true and there are too many inline back refs, the path
1464  * points to the extent item, and -EAGAIN is returned.
1465  *
1466  * NOTE: inline back refs are ordered in the same way that back ref
1467  *       items in the tree are ordered.
1468  */
1469 static noinline_for_stack
1470 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1471                                  struct btrfs_root *root,
1472                                  struct btrfs_path *path,
1473                                  struct btrfs_extent_inline_ref **ref_ret,
1474                                  u64 bytenr, u64 num_bytes,
1475                                  u64 parent, u64 root_objectid,
1476                                  u64 owner, u64 offset, int insert)
1477 {
1478         struct btrfs_key key;
1479         struct extent_buffer *leaf;
1480         struct btrfs_extent_item *ei;
1481         struct btrfs_extent_inline_ref *iref;
1482         u64 flags;
1483         u64 item_size;
1484         unsigned long ptr;
1485         unsigned long end;
1486         int extra_size;
1487         int type;
1488         int want;
1489         int ret;
1490         int err = 0;
1491         bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
1492                                                  SKINNY_METADATA);
1493 
1494         key.objectid = bytenr;
1495         key.type = BTRFS_EXTENT_ITEM_KEY;
1496         key.offset = num_bytes;
1497 
1498         want = extent_ref_type(parent, owner);
1499         if (insert) {
1500                 extra_size = btrfs_extent_inline_ref_size(want);
1501                 path->keep_locks = 1;
1502         } else
1503                 extra_size = -1;
1504 
1505         /*
1506          * Owner is our parent level, so we can just add one to get the level
1507          * for the block we are interested in.
1508          */
1509         if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
1510                 key.type = BTRFS_METADATA_ITEM_KEY;
1511                 key.offset = owner;
1512         }
1513 
1514 again:
1515         ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1516         if (ret < 0) {
1517                 err = ret;
1518                 goto out;
1519         }
1520 
1521         /*
1522          * We may be a newly converted file system which still has the old fat
1523          * extent entries for metadata, so try and see if we have one of those.
1524          */
1525         if (ret > 0 && skinny_metadata) {
1526                 skinny_metadata = false;
1527                 if (path->slots[0]) {
1528                         path->slots[0]--;
1529                         btrfs_item_key_to_cpu(path->nodes[0], &key,
1530                                               path->slots[0]);
1531                         if (key.objectid == bytenr &&
1532                             key.type == BTRFS_EXTENT_ITEM_KEY &&
1533                             key.offset == num_bytes)
1534                                 ret = 0;
1535                 }
1536                 if (ret) {
1537                         key.objectid = bytenr;
1538                         key.type = BTRFS_EXTENT_ITEM_KEY;
1539                         key.offset = num_bytes;
1540                         btrfs_release_path(path);
1541                         goto again;
1542                 }
1543         }
1544 
1545         if (ret && !insert) {
1546                 err = -ENOENT;
1547                 goto out;
1548         } else if (WARN_ON(ret)) {
1549                 err = -EIO;
1550                 goto out;
1551         }
1552 
1553         leaf = path->nodes[0];
1554         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1555 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1556         if (item_size < sizeof(*ei)) {
1557                 if (!insert) {
1558                         err = -ENOENT;
1559                         goto out;
1560                 }
1561                 ret = convert_extent_item_v0(trans, root, path, owner,
1562                                              extra_size);
1563                 if (ret < 0) {
1564                         err = ret;
1565                         goto out;
1566                 }
1567                 leaf = path->nodes[0];
1568                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1569         }
1570 #endif
1571         BUG_ON(item_size < sizeof(*ei));
1572 
1573         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1574         flags = btrfs_extent_flags(leaf, ei);
1575 
1576         ptr = (unsigned long)(ei + 1);
1577         end = (unsigned long)ei + item_size;
1578 
1579         if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1580                 ptr += sizeof(struct btrfs_tree_block_info);
1581                 BUG_ON(ptr > end);
1582         }
1583 
1584         err = -ENOENT;
1585         while (1) {
1586                 if (ptr >= end) {
1587                         WARN_ON(ptr > end);
1588                         break;
1589                 }
1590                 iref = (struct btrfs_extent_inline_ref *)ptr;
1591                 type = btrfs_extent_inline_ref_type(leaf, iref);
1592                 if (want < type)
1593                         break;
1594                 if (want > type) {
1595                         ptr += btrfs_extent_inline_ref_size(type);
1596                         continue;
1597                 }
1598 
1599                 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1600                         struct btrfs_extent_data_ref *dref;
1601                         dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1602                         if (match_extent_data_ref(leaf, dref, root_objectid,
1603                                                   owner, offset)) {
1604                                 err = 0;
1605                                 break;
1606                         }
1607                         if (hash_extent_data_ref_item(leaf, dref) <
1608                             hash_extent_data_ref(root_objectid, owner, offset))
1609                                 break;
1610                 } else {
1611                         u64 ref_offset;
1612                         ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1613                         if (parent > 0) {
1614                                 if (parent == ref_offset) {
1615                                         err = 0;
1616                                         break;
1617                                 }
1618                                 if (ref_offset < parent)
1619                                         break;
1620                         } else {
1621                                 if (root_objectid == ref_offset) {
1622                                         err = 0;
1623                                         break;
1624                                 }
1625                                 if (ref_offset < root_objectid)
1626                                         break;
1627                         }
1628                 }
1629                 ptr += btrfs_extent_inline_ref_size(type);
1630         }
1631         if (err == -ENOENT && insert) {
1632                 if (item_size + extra_size >=
1633                     BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1634                         err = -EAGAIN;
1635                         goto out;
1636                 }
1637                 /*
1638                  * To add new inline back ref, we have to make sure
1639                  * there is no corresponding back ref item.
1640                  * For simplicity, we just do not add new inline back
1641                  * ref if there is any kind of item for this block
1642                  */
1643                 if (find_next_key(path, 0, &key) == 0 &&
1644                     key.objectid == bytenr &&
1645                     key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1646                         err = -EAGAIN;
1647                         goto out;
1648                 }
1649         }
1650         *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1651 out:
1652         if (insert) {
1653                 path->keep_locks = 0;
1654                 btrfs_unlock_up_safe(path, 1);
1655         }
1656         return err;
1657 }
1658 
1659 /*
1660  * helper to add new inline back ref
1661  */
1662 static noinline_for_stack
1663 void setup_inline_extent_backref(struct btrfs_root *root,
1664                                  struct btrfs_path *path,
1665                                  struct btrfs_extent_inline_ref *iref,
1666                                  u64 parent, u64 root_objectid,
1667                                  u64 owner, u64 offset, int refs_to_add,
1668                                  struct btrfs_delayed_extent_op *extent_op)
1669 {
1670         struct extent_buffer *leaf;
1671         struct btrfs_extent_item *ei;
1672         unsigned long ptr;
1673         unsigned long end;
1674         unsigned long item_offset;
1675         u64 refs;
1676         int size;
1677         int type;
1678 
1679         leaf = path->nodes[0];
1680         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1681         item_offset = (unsigned long)iref - (unsigned long)ei;
1682 
1683         type = extent_ref_type(parent, owner);
1684         size = btrfs_extent_inline_ref_size(type);
1685 
1686         btrfs_extend_item(root, path, size);
1687 
1688         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1689         refs = btrfs_extent_refs(leaf, ei);
1690         refs += refs_to_add;
1691         btrfs_set_extent_refs(leaf, ei, refs);
1692         if (extent_op)
1693                 __run_delayed_extent_op(extent_op, leaf, ei);
1694 
1695         ptr = (unsigned long)ei + item_offset;
1696         end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1697         if (ptr < end - size)
1698                 memmove_extent_buffer(leaf, ptr + size, ptr,
1699                                       end - size - ptr);
1700 
1701         iref = (struct btrfs_extent_inline_ref *)ptr;
1702         btrfs_set_extent_inline_ref_type(leaf, iref, type);
1703         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1704                 struct btrfs_extent_data_ref *dref;
1705                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1706                 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1707                 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1708                 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1709                 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1710         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1711                 struct btrfs_shared_data_ref *sref;
1712                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1713                 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1714                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1715         } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1716                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1717         } else {
1718                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1719         }
1720         btrfs_mark_buffer_dirty(leaf);
1721 }
1722 
1723 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1724                                  struct btrfs_root *root,
1725                                  struct btrfs_path *path,
1726                                  struct btrfs_extent_inline_ref **ref_ret,
1727                                  u64 bytenr, u64 num_bytes, u64 parent,
1728                                  u64 root_objectid, u64 owner, u64 offset)
1729 {
1730         int ret;
1731 
1732         ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1733                                            bytenr, num_bytes, parent,
1734                                            root_objectid, owner, offset, 0);
1735         if (ret != -ENOENT)
1736                 return ret;
1737 
1738         btrfs_release_path(path);
1739         *ref_ret = NULL;
1740 
1741         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1742                 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1743                                             root_objectid);
1744         } else {
1745                 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1746                                              root_objectid, owner, offset);
1747         }
1748         return ret;
1749 }
1750 
1751 /*
1752  * helper to update/remove inline back ref
1753  */
1754 static noinline_for_stack
1755 void update_inline_extent_backref(struct btrfs_root *root,
1756                                   struct btrfs_path *path,
1757                                   struct btrfs_extent_inline_ref *iref,
1758                                   int refs_to_mod,
1759                                   struct btrfs_delayed_extent_op *extent_op,
1760                                   int *last_ref)
1761 {
1762         struct extent_buffer *leaf;
1763         struct btrfs_extent_item *ei;
1764         struct btrfs_extent_data_ref *dref = NULL;
1765         struct btrfs_shared_data_ref *sref = NULL;
1766         unsigned long ptr;
1767         unsigned long end;
1768         u32 item_size;
1769         int size;
1770         int type;
1771         u64 refs;
1772 
1773         leaf = path->nodes[0];
1774         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1775         refs = btrfs_extent_refs(leaf, ei);
1776         WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1777         refs += refs_to_mod;
1778         btrfs_set_extent_refs(leaf, ei, refs);
1779         if (extent_op)
1780                 __run_delayed_extent_op(extent_op, leaf, ei);
1781 
1782         type = btrfs_extent_inline_ref_type(leaf, iref);
1783 
1784         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1785                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1786                 refs = btrfs_extent_data_ref_count(leaf, dref);
1787         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1788                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1789                 refs = btrfs_shared_data_ref_count(leaf, sref);
1790         } else {
1791                 refs = 1;
1792                 BUG_ON(refs_to_mod != -1);
1793         }
1794 
1795         BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1796         refs += refs_to_mod;
1797 
1798         if (refs > 0) {
1799                 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1800                         btrfs_set_extent_data_ref_count(leaf, dref, refs);
1801                 else
1802                         btrfs_set_shared_data_ref_count(leaf, sref, refs);
1803         } else {
1804                 *last_ref = 1;
1805                 size =  btrfs_extent_inline_ref_size(type);
1806                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1807                 ptr = (unsigned long)iref;
1808                 end = (unsigned long)ei + item_size;
1809                 if (ptr + size < end)
1810                         memmove_extent_buffer(leaf, ptr, ptr + size,
1811                                               end - ptr - size);
1812                 item_size -= size;
1813                 btrfs_truncate_item(root, path, item_size, 1);
1814         }
1815         btrfs_mark_buffer_dirty(leaf);
1816 }
1817 
1818 static noinline_for_stack
1819 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1820                                  struct btrfs_root *root,
1821                                  struct btrfs_path *path,
1822                                  u64 bytenr, u64 num_bytes, u64 parent,
1823                                  u64 root_objectid, u64 owner,
1824                                  u64 offset, int refs_to_add,
1825                                  struct btrfs_delayed_extent_op *extent_op)
1826 {
1827         struct btrfs_extent_inline_ref *iref;
1828         int ret;
1829 
1830         ret = lookup_inline_extent_backref(trans, root, path, &iref,
1831                                            bytenr, num_bytes, parent,
1832                                            root_objectid, owner, offset, 1);
1833         if (ret == 0) {
1834                 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1835                 update_inline_extent_backref(root, path, iref,
1836                                              refs_to_add, extent_op, NULL);
1837         } else if (ret == -ENOENT) {
1838                 setup_inline_extent_backref(root, path, iref, parent,
1839                                             root_objectid, owner, offset,
1840                                             refs_to_add, extent_op);
1841                 ret = 0;
1842         }
1843         return ret;
1844 }
1845 
1846 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1847                                  struct btrfs_root *root,
1848                                  struct btrfs_path *path,
1849                                  u64 bytenr, u64 parent, u64 root_objectid,
1850                                  u64 owner, u64 offset, int refs_to_add)
1851 {
1852         int ret;
1853         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1854                 BUG_ON(refs_to_add != 1);
1855                 ret = insert_tree_block_ref(trans, root, path, bytenr,
1856                                             parent, root_objectid);
1857         } else {
1858                 ret = insert_extent_data_ref(trans, root, path, bytenr,
1859                                              parent, root_objectid,
1860                                              owner, offset, refs_to_add);
1861         }
1862         return ret;
1863 }
1864 
1865 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1866                                  struct btrfs_root *root,
1867                                  struct btrfs_path *path,
1868                                  struct btrfs_extent_inline_ref *iref,
1869                                  int refs_to_drop, int is_data, int *last_ref)
1870 {
1871         int ret = 0;
1872 
1873         BUG_ON(!is_data && refs_to_drop != 1);
1874         if (iref) {
1875                 update_inline_extent_backref(root, path, iref,
1876                                              -refs_to_drop, NULL, last_ref);
1877         } else if (is_data) {
1878                 ret = remove_extent_data_ref(trans, root, path, refs_to_drop,
1879                                              last_ref);
1880         } else {
1881                 *last_ref = 1;
1882                 ret = btrfs_del_item(trans, root, path);
1883         }
1884         return ret;
1885 }
1886 
1887 static int btrfs_issue_discard(struct block_device *bdev,
1888                                 u64 start, u64 len)
1889 {
1890         return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
1891 }
1892 
1893 int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1894                          u64 num_bytes, u64 *actual_bytes)
1895 {
1896         int ret;
1897         u64 discarded_bytes = 0;
1898         struct btrfs_bio *bbio = NULL;
1899 
1900 
1901         /* Tell the block device(s) that the sectors can be discarded */
1902         ret = btrfs_map_block(root->fs_info, REQ_DISCARD,
1903                               bytenr, &num_bytes, &bbio, 0);
1904         /* Error condition is -ENOMEM */
1905         if (!ret) {
1906                 struct btrfs_bio_stripe *stripe = bbio->stripes;
1907                 int i;
1908 
1909 
1910                 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
1911                         if (!stripe->dev->can_discard)
1912                                 continue;
1913 
1914                         ret = btrfs_issue_discard(stripe->dev->bdev,
1915                                                   stripe->physical,
1916                                                   stripe->length);
1917                         if (!ret)
1918                                 discarded_bytes += stripe->length;
1919                         else if (ret != -EOPNOTSUPP)
1920                                 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
1921 
1922                         /*
1923                          * Just in case we get back EOPNOTSUPP for some reason,
1924                          * just ignore the return value so we don't screw up
1925                          * people calling discard_extent.
1926                          */
1927                         ret = 0;
1928                 }
1929                 btrfs_put_bbio(bbio);
1930         }
1931 
1932         if (actual_bytes)
1933                 *actual_bytes = discarded_bytes;
1934 
1935 
1936         if (ret == -EOPNOTSUPP)
1937                 ret = 0;
1938         return ret;
1939 }
1940 
1941 /* Can return -ENOMEM */
1942 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1943                          struct btrfs_root *root,
1944                          u64 bytenr, u64 num_bytes, u64 parent,
1945                          u64 root_objectid, u64 owner, u64 offset,
1946                          int no_quota)
1947 {
1948         int ret;
1949         struct btrfs_fs_info *fs_info = root->fs_info;
1950 
1951         BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1952                root_objectid == BTRFS_TREE_LOG_OBJECTID);
1953 
1954         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1955                 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
1956                                         num_bytes,
1957                                         parent, root_objectid, (int)owner,
1958                                         BTRFS_ADD_DELAYED_REF, NULL, no_quota);
1959         } else {
1960                 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
1961                                         num_bytes,
1962                                         parent, root_objectid, owner, offset,
1963                                         BTRFS_ADD_DELAYED_REF, NULL, no_quota);
1964         }
1965         return ret;
1966 }
1967 
1968 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1969                                   struct btrfs_root *root,
1970                                   u64 bytenr, u64 num_bytes,
1971                                   u64 parent, u64 root_objectid,
1972                                   u64 owner, u64 offset, int refs_to_add,
1973                                   int no_quota,
1974                                   struct btrfs_delayed_extent_op *extent_op)
1975 {
1976         struct btrfs_fs_info *fs_info = root->fs_info;
1977         struct btrfs_path *path;
1978         struct extent_buffer *leaf;
1979         struct btrfs_extent_item *item;
1980         struct btrfs_key key;
1981         u64 refs;
1982         int ret;
1983         enum btrfs_qgroup_operation_type type = BTRFS_QGROUP_OPER_ADD_EXCL;
1984 
1985         path = btrfs_alloc_path();
1986         if (!path)
1987                 return -ENOMEM;
1988 
1989         if (!is_fstree(root_objectid) || !root->fs_info->quota_enabled)
1990                 no_quota = 1;
1991 
1992         path->reada = 1;
1993         path->leave_spinning = 1;
1994         /* this will setup the path even if it fails to insert the back ref */
1995         ret = insert_inline_extent_backref(trans, fs_info->extent_root, path,
1996                                            bytenr, num_bytes, parent,
1997                                            root_objectid, owner, offset,
1998                                            refs_to_add, extent_op);
1999         if ((ret < 0 && ret != -EAGAIN) || (!ret && no_quota))
2000                 goto out;
2001         /*
2002          * Ok we were able to insert an inline extent and it appears to be a new
2003          * reference, deal with the qgroup accounting.
2004          */
2005         if (!ret && !no_quota) {
2006                 ASSERT(root->fs_info->quota_enabled);
2007                 leaf = path->nodes[0];
2008                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2009                 item = btrfs_item_ptr(leaf, path->slots[0],
2010                                       struct btrfs_extent_item);
2011                 if (btrfs_extent_refs(leaf, item) > (u64)refs_to_add)
2012                         type = BTRFS_QGROUP_OPER_ADD_SHARED;
2013                 btrfs_release_path(path);
2014 
2015                 ret = btrfs_qgroup_record_ref(trans, fs_info, root_objectid,
2016                                               bytenr, num_bytes, type, 0);
2017                 goto out;
2018         }
2019 
2020         /*
2021          * Ok we had -EAGAIN which means we didn't have space to insert and
2022          * inline extent ref, so just update the reference count and add a
2023          * normal backref.
2024          */
2025         leaf = path->nodes[0];
2026         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2027         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2028         refs = btrfs_extent_refs(leaf, item);
2029         if (refs)
2030                 type = BTRFS_QGROUP_OPER_ADD_SHARED;
2031         btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2032         if (extent_op)
2033                 __run_delayed_extent_op(extent_op, leaf, item);
2034 
2035         btrfs_mark_buffer_dirty(leaf);
2036         btrfs_release_path(path);
2037 
2038         if (!no_quota) {
2039                 ret = btrfs_qgroup_record_ref(trans, fs_info, root_objectid,
2040                                               bytenr, num_bytes, type, 0);
2041                 if (ret)
2042                         goto out;
2043         }
2044 
2045         path->reada = 1;
2046         path->leave_spinning = 1;
2047         /* now insert the actual backref */
2048         ret = insert_extent_backref(trans, root->fs_info->extent_root,
2049                                     path, bytenr, parent, root_objectid,
2050                                     owner, offset, refs_to_add);
2051         if (ret)
2052                 btrfs_abort_transaction(trans, root, ret);
2053 out:
2054         btrfs_free_path(path);
2055         return ret;
2056 }
2057 
2058 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2059                                 struct btrfs_root *root,
2060                                 struct btrfs_delayed_ref_node *node,
2061                                 struct btrfs_delayed_extent_op *extent_op,
2062                                 int insert_reserved)
2063 {
2064         int ret = 0;
2065         struct btrfs_delayed_data_ref *ref;
2066         struct btrfs_key ins;
2067         u64 parent = 0;
2068         u64 ref_root = 0;
2069         u64 flags = 0;
2070 
2071         ins.objectid = node->bytenr;
2072         ins.offset = node->num_bytes;
2073         ins.type = BTRFS_EXTENT_ITEM_KEY;
2074 
2075         ref = btrfs_delayed_node_to_data_ref(node);
2076         trace_run_delayed_data_ref(node, ref, node->action);
2077 
2078         if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2079                 parent = ref->parent;
2080         ref_root = ref->root;
2081 
2082         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2083                 if (extent_op)
2084                         flags |= extent_op->flags_to_set;
2085                 ret = alloc_reserved_file_extent(trans, root,
2086                                                  parent, ref_root, flags,
2087                                                  ref->objectid, ref->offset,
2088                                                  &ins, node->ref_mod);
2089         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2090                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2091                                              node->num_bytes, parent,
2092                                              ref_root, ref->objectid,
2093                                              ref->offset, node->ref_mod,
2094                                              node->no_quota, extent_op);
2095         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2096                 ret = __btrfs_free_extent(trans, root, node->bytenr,
2097                                           node->num_bytes, parent,
2098                                           ref_root, ref->objectid,
2099                                           ref->offset, node->ref_mod,
2100                                           extent_op, node->no_quota);
2101         } else {
2102                 BUG();
2103         }
2104         return ret;
2105 }
2106 
2107 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2108                                     struct extent_buffer *leaf,
2109                                     struct btrfs_extent_item *ei)
2110 {
2111         u64 flags = btrfs_extent_flags(leaf, ei);
2112         if (extent_op->update_flags) {
2113                 flags |= extent_op->flags_to_set;
2114                 btrfs_set_extent_flags(leaf, ei, flags);
2115         }
2116 
2117         if (extent_op->update_key) {
2118                 struct btrfs_tree_block_info *bi;
2119                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2120                 bi = (struct btrfs_tree_block_info *)(ei + 1);
2121                 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2122         }
2123 }
2124 
2125 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2126                                  struct btrfs_root *root,
2127                                  struct btrfs_delayed_ref_node *node,
2128                                  struct btrfs_delayed_extent_op *extent_op)
2129 {
2130         struct btrfs_key key;
2131         struct btrfs_path *path;
2132         struct btrfs_extent_item *ei;
2133         struct extent_buffer *leaf;
2134         u32 item_size;
2135         int ret;
2136         int err = 0;
2137         int metadata = !extent_op->is_data;
2138 
2139         if (trans->aborted)
2140                 return 0;
2141 
2142         if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2143                 metadata = 0;
2144 
2145         path = btrfs_alloc_path();
2146         if (!path)
2147                 return -ENOMEM;
2148 
2149         key.objectid = node->bytenr;
2150 
2151         if (metadata) {
2152                 key.type = BTRFS_METADATA_ITEM_KEY;
2153                 key.offset = extent_op->level;
2154         } else {
2155                 key.type = BTRFS_EXTENT_ITEM_KEY;
2156                 key.offset = node->num_bytes;
2157         }
2158 
2159 again:
2160         path->reada = 1;
2161         path->leave_spinning = 1;
2162         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2163                                 path, 0, 1);
2164         if (ret < 0) {
2165                 err = ret;
2166                 goto out;
2167         }
2168         if (ret > 0) {
2169                 if (metadata) {
2170                         if (path->slots[0] > 0) {
2171                                 path->slots[0]--;
2172                                 btrfs_item_key_to_cpu(path->nodes[0], &key,
2173                                                       path->slots[0]);
2174                                 if (key.objectid == node->bytenr &&
2175                                     key.type == BTRFS_EXTENT_ITEM_KEY &&
2176                                     key.offset == node->num_bytes)
2177                                         ret = 0;
2178                         }
2179                         if (ret > 0) {
2180                                 btrfs_release_path(path);
2181                                 metadata = 0;
2182 
2183                                 key.objectid = node->bytenr;
2184                                 key.offset = node->num_bytes;
2185                                 key.type = BTRFS_EXTENT_ITEM_KEY;
2186                                 goto again;
2187                         }
2188                 } else {
2189                         err = -EIO;
2190                         goto out;
2191                 }
2192         }
2193 
2194         leaf = path->nodes[0];
2195         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2196 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2197         if (item_size < sizeof(*ei)) {
2198                 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2199                                              path, (u64)-1, 0);
2200                 if (ret < 0) {
2201                         err = ret;
2202                         goto out;
2203                 }
2204                 leaf = path->nodes[0];
2205                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2206         }
2207 #endif
2208         BUG_ON(item_size < sizeof(*ei));
2209         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2210         __run_delayed_extent_op(extent_op, leaf, ei);
2211 
2212         btrfs_mark_buffer_dirty(leaf);
2213 out:
2214         btrfs_free_path(path);
2215         return err;
2216 }
2217 
2218 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2219                                 struct btrfs_root *root,
2220                                 struct btrfs_delayed_ref_node *node,
2221                                 struct btrfs_delayed_extent_op *extent_op,
2222                                 int insert_reserved)
2223 {
2224         int ret = 0;
2225         struct btrfs_delayed_tree_ref *ref;
2226         struct btrfs_key ins;
2227         u64 parent = 0;
2228         u64 ref_root = 0;
2229         bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
2230                                                  SKINNY_METADATA);
2231 
2232         ref = btrfs_delayed_node_to_tree_ref(node);
2233         trace_run_delayed_tree_ref(node, ref, node->action);
2234 
2235         if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2236                 parent = ref->parent;
2237         ref_root = ref->root;
2238 
2239         ins.objectid = node->bytenr;
2240         if (skinny_metadata) {
2241                 ins.offset = ref->level;
2242                 ins.type = BTRFS_METADATA_ITEM_KEY;
2243         } else {
2244                 ins.offset = node->num_bytes;
2245                 ins.type = BTRFS_EXTENT_ITEM_KEY;
2246         }
2247 
2248         BUG_ON(node->ref_mod != 1);
2249         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2250                 BUG_ON(!extent_op || !extent_op->update_flags);
2251                 ret = alloc_reserved_tree_block(trans, root,
2252                                                 parent, ref_root,
2253                                                 extent_op->flags_to_set,
2254                                                 &extent_op->key,
2255                                                 ref->level, &ins,
2256                                                 node->no_quota);
2257         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2258                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2259                                              node->num_bytes, parent, ref_root,
2260                                              ref->level, 0, 1, node->no_quota,
2261                                              extent_op);
2262         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2263                 ret = __btrfs_free_extent(trans, root, node->bytenr,
2264                                           node->num_bytes, parent, ref_root,
2265                                           ref->level, 0, 1, extent_op,
2266                                           node->no_quota);
2267         } else {
2268                 BUG();
2269         }
2270         return ret;
2271 }
2272 
2273 /* helper function to actually process a single delayed ref entry */
2274 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2275                                struct btrfs_root *root,
2276                                struct btrfs_delayed_ref_node *node,
2277                                struct btrfs_delayed_extent_op *extent_op,
2278                                int insert_reserved)
2279 {
2280         int ret = 0;
2281 
2282         if (trans->aborted) {
2283                 if (insert_reserved)
2284                         btrfs_pin_extent(root, node->bytenr,
2285                                          node->num_bytes, 1);
2286                 return 0;
2287         }
2288 
2289         if (btrfs_delayed_ref_is_head(node)) {
2290                 struct btrfs_delayed_ref_head *head;
2291                 /*
2292                  * we've hit the end of the chain and we were supposed
2293                  * to insert this extent into the tree.  But, it got
2294                  * deleted before we ever needed to insert it, so all
2295                  * we have to do is clean up the accounting
2296                  */
2297                 BUG_ON(extent_op);
2298                 head = btrfs_delayed_node_to_head(node);
2299                 trace_run_delayed_ref_head(node, head, node->action);
2300 
2301                 if (insert_reserved) {
2302                         btrfs_pin_extent(root, node->bytenr,
2303                                          node->num_bytes, 1);
2304                         if (head->is_data) {
2305                                 ret = btrfs_del_csums(trans, root,
2306                                                       node->bytenr,
2307                                                       node->num_bytes);
2308                         }
2309                 }
2310                 return ret;
2311         }
2312 
2313         if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2314             node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2315                 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2316                                            insert_reserved);
2317         else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2318                  node->type == BTRFS_SHARED_DATA_REF_KEY)
2319                 ret = run_delayed_data_ref(trans, root, node, extent_op,
2320                                            insert_reserved);
2321         else
2322                 BUG();
2323         return ret;
2324 }
2325 
2326 static noinline struct btrfs_delayed_ref_node *
2327 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2328 {
2329         struct rb_node *node;
2330         struct btrfs_delayed_ref_node *ref, *last = NULL;;
2331 
2332         /*
2333          * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2334          * this prevents ref count from going down to zero when
2335          * there still are pending delayed ref.
2336          */
2337         node = rb_first(&head->ref_root);
2338         while (node) {
2339                 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2340                                 rb_node);
2341                 if (ref->action == BTRFS_ADD_DELAYED_REF)
2342                         return ref;
2343                 else if (last == NULL)
2344                         last = ref;
2345                 node = rb_next(node);
2346         }
2347         return last;
2348 }
2349 
2350 /*
2351  * Returns 0 on success or if called with an already aborted transaction.
2352  * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2353  */
2354 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2355                                              struct btrfs_root *root,
2356                                              unsigned long nr)
2357 {
2358         struct btrfs_delayed_ref_root *delayed_refs;
2359         struct btrfs_delayed_ref_node *ref;
2360         struct btrfs_delayed_ref_head *locked_ref = NULL;
2361         struct btrfs_delayed_extent_op *extent_op;
2362         struct btrfs_fs_info *fs_info = root->fs_info;
2363         ktime_t start = ktime_get();
2364         int ret;
2365         unsigned long count = 0;
2366         unsigned long actual_count = 0;
2367         int must_insert_reserved = 0;
2368 
2369         delayed_refs = &trans->transaction->delayed_refs;
2370         while (1) {
2371                 if (!locked_ref) {
2372                         if (count >= nr)
2373                                 break;
2374 
2375                         spin_lock(&delayed_refs->lock);
2376                         locked_ref = btrfs_select_ref_head(trans);
2377                         if (!locked_ref) {
2378                                 spin_unlock(&delayed_refs->lock);
2379                                 break;
2380                         }
2381 
2382                         /* grab the lock that says we are going to process
2383                          * all the refs for this head */
2384                         ret = btrfs_delayed_ref_lock(trans, locked_ref);
2385                         spin_unlock(&delayed_refs->lock);
2386                         /*
2387                          * we may have dropped the spin lock to get the head
2388                          * mutex lock, and that might have given someone else
2389                          * time to free the head.  If that's true, it has been
2390                          * removed from our list and we can move on.
2391                          */
2392                         if (ret == -EAGAIN) {
2393                                 locked_ref = NULL;
2394                                 count++;
2395                                 continue;
2396                         }
2397                 }
2398 
2399                 /*
2400                  * We need to try and merge add/drops of the same ref since we
2401                  * can run into issues with relocate dropping the implicit ref
2402                  * and then it being added back again before the drop can
2403                  * finish.  If we merged anything we need to re-loop so we can
2404                  * get a good ref.
2405                  */
2406                 spin_lock(&locked_ref->lock);
2407                 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2408                                          locked_ref);
2409 
2410                 /*
2411                  * locked_ref is the head node, so we have to go one
2412                  * node back for any delayed ref updates
2413                  */
2414                 ref = select_delayed_ref(locked_ref);
2415 
2416                 if (ref && ref->seq &&
2417                     btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2418                         spin_unlock(&locked_ref->lock);
2419                         btrfs_delayed_ref_unlock(locked_ref);
2420                         spin_lock(&delayed_refs->lock);
2421                         locked_ref->processing = 0;
2422                         delayed_refs->num_heads_ready++;
2423                         spin_unlock(&delayed_refs->lock);
2424                         locked_ref = NULL;
2425                         cond_resched();
2426                         count++;
2427                         continue;
2428                 }
2429 
2430                 /*
2431                  * record the must insert reserved flag before we
2432                  * drop the spin lock.
2433                  */
2434                 must_insert_reserved = locked_ref->must_insert_reserved;
2435                 locked_ref->must_insert_reserved = 0;
2436 
2437                 extent_op = locked_ref->extent_op;
2438                 locked_ref->extent_op = NULL;
2439 
2440                 if (!ref) {
2441 
2442 
2443                         /* All delayed refs have been processed, Go ahead
2444                          * and send the head node to run_one_delayed_ref,
2445                          * so that any accounting fixes can happen
2446                          */
2447                         ref = &locked_ref->node;
2448 
2449                         if (extent_op && must_insert_reserved) {
2450                                 btrfs_free_delayed_extent_op(extent_op);
2451                                 extent_op = NULL;
2452                         }
2453 
2454                         if (extent_op) {
2455                                 spin_unlock(&locked_ref->lock);
2456                                 ret = run_delayed_extent_op(trans, root,
2457                                                             ref, extent_op);
2458                                 btrfs_free_delayed_extent_op(extent_op);
2459 
2460                                 if (ret) {
2461                                         /*
2462                                          * Need to reset must_insert_reserved if
2463                                          * there was an error so the abort stuff
2464                                          * can cleanup the reserved space
2465                                          * properly.
2466                                          */
2467                                         if (must_insert_reserved)
2468                                                 locked_ref->must_insert_reserved = 1;
2469                                         locked_ref->processing = 0;
2470                                         btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2471                                         btrfs_delayed_ref_unlock(locked_ref);
2472                                         return ret;
2473                                 }
2474                                 continue;
2475                         }
2476 
2477                         /*
2478                          * Need to drop our head ref lock and re-aqcuire the
2479                          * delayed ref lock and then re-check to make sure
2480                          * nobody got added.
2481                          */
2482                         spin_unlock(&locked_ref->lock);
2483                         spin_lock(&delayed_refs->lock);
2484                         spin_lock(&locked_ref->lock);
2485                         if (rb_first(&locked_ref->ref_root) ||
2486                             locked_ref->extent_op) {
2487                                 spin_unlock(&locked_ref->lock);
2488                                 spin_unlock(&delayed_refs->lock);
2489                                 continue;
2490                         }
2491                         ref->in_tree = 0;
2492                         delayed_refs->num_heads--;
2493                         rb_erase(&locked_ref->href_node,
2494                                  &delayed_refs->href_root);
2495                         spin_unlock(&delayed_refs->lock);
2496                 } else {
2497                         actual_count++;
2498                         ref->in_tree = 0;
2499                         rb_erase(&ref->rb_node, &locked_ref->ref_root);
2500                 }
2501                 atomic_dec(&delayed_refs->num_entries);
2502 
2503                 if (!btrfs_delayed_ref_is_head(ref)) {
2504                         /*
2505                          * when we play the delayed ref, also correct the
2506                          * ref_mod on head
2507                          */
2508                         switch (ref->action) {
2509                         case BTRFS_ADD_DELAYED_REF:
2510                         case BTRFS_ADD_DELAYED_EXTENT:
2511                                 locked_ref->node.ref_mod -= ref->ref_mod;
2512                                 break;
2513                         case BTRFS_DROP_DELAYED_REF:
2514                                 locked_ref->node.ref_mod += ref->ref_mod;
2515                                 break;
2516                         default:
2517                                 WARN_ON(1);
2518                         }
2519                 }
2520                 spin_unlock(&locked_ref->lock);
2521 
2522                 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2523                                           must_insert_reserved);
2524 
2525                 btrfs_free_delayed_extent_op(extent_op);
2526                 if (ret) {
2527                         locked_ref->processing = 0;
2528                         btrfs_delayed_ref_unlock(locked_ref);
2529                         btrfs_put_delayed_ref(ref);
2530                         btrfs_debug(fs_info, "run_one_delayed_ref returned %d", ret);
2531                         return ret;
2532                 }
2533 
2534                 /*
2535                  * If this node is a head, that means all the refs in this head
2536                  * have been dealt with, and we will pick the next head to deal
2537                  * with, so we must unlock the head and drop it from the cluster
2538                  * list before we release it.
2539                  */
2540                 if (btrfs_delayed_ref_is_head(ref)) {
2541                         btrfs_delayed_ref_unlock(locked_ref);
2542                         locked_ref = NULL;
2543                 }
2544                 btrfs_put_delayed_ref(ref);
2545                 count++;
2546                 cond_resched();
2547         }
2548 
2549         /*
2550          * We don't want to include ref heads since we can have empty ref heads
2551          * and those will drastically skew our runtime down since we just do
2552          * accounting, no actual extent tree updates.
2553          */
2554         if (actual_count > 0) {
2555                 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2556                 u64 avg;
2557 
2558                 /*
2559                  * We weigh the current average higher than our current runtime
2560                  * to avoid large swings in the average.
2561                  */
2562                 spin_lock(&delayed_refs->lock);
2563                 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2564                 avg = div64_u64(avg, 4);
2565                 fs_info->avg_delayed_ref_runtime = avg;
2566                 spin_unlock(&delayed_refs->lock);
2567         }
2568         return 0;
2569 }
2570 
2571 #ifdef SCRAMBLE_DELAYED_REFS
2572 /*
2573  * Normally delayed refs get processed in ascending bytenr order. This
2574  * correlates in most cases to the order added. To expose dependencies on this
2575  * order, we start to process the tree in the middle instead of the beginning
2576  */
2577 static u64 find_middle(struct rb_root *root)
2578 {
2579         struct rb_node *n = root->rb_node;
2580         struct btrfs_delayed_ref_node *entry;
2581         int alt = 1;
2582         u64 middle;
2583         u64 first = 0, last = 0;
2584 
2585         n = rb_first(root);
2586         if (n) {
2587                 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2588                 first = entry->bytenr;
2589         }
2590         n = rb_last(root);
2591         if (n) {
2592                 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2593                 last = entry->bytenr;
2594         }
2595         n = root->rb_node;
2596 
2597         while (n) {
2598                 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2599                 WARN_ON(!entry->in_tree);
2600 
2601                 middle = entry->bytenr;
2602 
2603                 if (alt)
2604                         n = n->rb_left;
2605                 else
2606                         n = n->rb_right;
2607 
2608                 alt = 1 - alt;
2609         }
2610         return middle;
2611 }
2612 #endif
2613 
2614 static inline u64 heads_to_leaves(struct btrfs_root *root, u64 heads)
2615 {
2616         u64 num_bytes;
2617 
2618         num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2619                              sizeof(struct btrfs_extent_inline_ref));
2620         if (!btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2621                 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2622 
2623         /*
2624          * We don't ever fill up leaves all the way so multiply by 2 just to be
2625          * closer to what we're really going to want to ouse.
2626          */
2627         return div64_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(root));
2628 }
2629 
2630 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans,
2631                                        struct btrfs_root *root)
2632 {
2633         struct btrfs_block_rsv *global_rsv;
2634         u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2635         u64 num_bytes;
2636         int ret = 0;
2637 
2638         num_bytes = btrfs_calc_trans_metadata_size(root, 1);
2639         num_heads = heads_to_leaves(root, num_heads);
2640         if (num_heads > 1)
2641                 num_bytes += (num_heads - 1) * root->nodesize;
2642         num_bytes <<= 1;
2643         global_rsv = &root->fs_info->global_block_rsv;
2644 
2645         /*
2646          * If we can't allocate any more chunks lets make sure we have _lots_ of
2647          * wiggle room since running delayed refs can create more delayed refs.
2648          */
2649         if (global_rsv->space_info->full)
2650                 num_bytes <<= 1;
2651 
2652         spin_lock(&global_rsv->lock);
2653         if (global_rsv->reserved <= num_bytes)
2654                 ret = 1;
2655         spin_unlock(&global_rsv->lock);
2656         return ret;
2657 }
2658 
2659 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2660                                        struct btrfs_root *root)
2661 {
2662         struct btrfs_fs_info *fs_info = root->fs_info;
2663         u64 num_entries =
2664                 atomic_read(&trans->transaction->delayed_refs.num_entries);
2665         u64 avg_runtime;
2666         u64 val;
2667 
2668         smp_mb();
2669         avg_runtime = fs_info->avg_delayed_ref_runtime;
2670         val = num_entries * avg_runtime;
2671         if (num_entries * avg_runtime >= NSEC_PER_SEC)
2672                 return 1;
2673         if (val >= NSEC_PER_SEC / 2)
2674                 return 2;
2675 
2676         return btrfs_check_space_for_delayed_refs(trans, root);
2677 }
2678 
2679 struct async_delayed_refs {
2680         struct btrfs_root *root;
2681         int count;
2682         int error;
2683         int sync;
2684         struct completion wait;
2685         struct btrfs_work work;
2686 };
2687 
2688 static void delayed_ref_async_start(struct btrfs_work *work)
2689 {
2690         struct async_delayed_refs *async;
2691         struct btrfs_trans_handle *trans;
2692         int ret;
2693 
2694         async = container_of(work, struct async_delayed_refs, work);
2695 
2696         trans = btrfs_join_transaction(async->root);
2697         if (IS_ERR(trans)) {
2698                 async->error = PTR_ERR(trans);
2699                 goto done;
2700         }
2701 
2702         /*
2703          * trans->sync means that when we call end_transaciton, we won't
2704          * wait on delayed refs
2705          */
2706         trans->sync = true;
2707         ret = btrfs_run_delayed_refs(trans, async->root, async->count);
2708         if (ret)
2709                 async->error = ret;
2710 
2711         ret = btrfs_end_transaction(trans, async->root);
2712         if (ret && !async->error)
2713                 async->error = ret;
2714 done:
2715         if (async->sync)
2716                 complete(&async->wait);
2717         else
2718                 kfree(async);
2719 }
2720 
2721 int btrfs_async_run_delayed_refs(struct btrfs_root *root,
2722                                  unsigned long count, int wait)
2723 {
2724         struct async_delayed_refs *async;
2725         int ret;
2726 
2727         async = kmalloc(sizeof(*async), GFP_NOFS);
2728         if (!async)
2729                 return -ENOMEM;
2730 
2731         async->root = root->fs_info->tree_root;
2732         async->count = count;
2733         async->error = 0;
2734         if (wait)
2735                 async->sync = 1;
2736         else
2737                 async->sync = 0;
2738         init_completion(&async->wait);
2739 
2740         btrfs_init_work(&async->work, btrfs_extent_refs_helper,
2741                         delayed_ref_async_start, NULL, NULL);
2742 
2743         btrfs_queue_work(root->fs_info->extent_workers, &async->work);
2744 
2745         if (wait) {
2746                 wait_for_completion(&async->wait);
2747                 ret = async->error;
2748                 kfree(async);
2749                 return ret;
2750         }
2751         return 0;
2752 }
2753 
2754 /*
2755  * this starts processing the delayed reference count updates and
2756  * extent insertions we have queued up so far.  count can be
2757  * 0, which means to process everything in the tree at the start
2758  * of the run (but not newly added entries), or it can be some target
2759  * number you'd like to process.
2760  *
2761  * Returns 0 on success or if called with an aborted transaction
2762  * Returns <0 on error and aborts the transaction
2763  */
2764 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2765                            struct btrfs_root *root, unsigned long count)
2766 {
2767         struct rb_node *node;
2768         struct btrfs_delayed_ref_root *delayed_refs;
2769         struct btrfs_delayed_ref_head *head;
2770         int ret;
2771         int run_all = count == (unsigned long)-1;
2772 
2773         /* We'll clean this up in btrfs_cleanup_transaction */
2774         if (trans->aborted)
2775                 return 0;
2776 
2777         if (root == root->fs_info->extent_root)
2778                 root = root->fs_info->tree_root;
2779 
2780         delayed_refs = &trans->transaction->delayed_refs;
2781         if (count == 0)
2782                 count = atomic_read(&delayed_refs->num_entries) * 2;
2783 
2784 again:
2785 #ifdef SCRAMBLE_DELAYED_REFS
2786         delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2787 #endif
2788         ret = __btrfs_run_delayed_refs(trans, root, count);
2789         if (ret < 0) {
2790                 btrfs_abort_transaction(trans, root, ret);
2791                 return ret;
2792         }
2793 
2794         if (run_all) {
2795                 if (!list_empty(&trans->new_bgs))
2796                         btrfs_create_pending_block_groups(trans, root);
2797 
2798                 spin_lock(&delayed_refs->lock);
2799                 node = rb_first(&delayed_refs->href_root);
2800                 if (!node) {
2801                         spin_unlock(&delayed_refs->lock);
2802                         goto out;
2803                 }
2804                 count = (unsigned long)-1;
2805 
2806                 while (node) {
2807                         head = rb_entry(node, struct btrfs_delayed_ref_head,
2808                                         href_node);
2809                         if (btrfs_delayed_ref_is_head(&head->node)) {
2810                                 struct btrfs_delayed_ref_node *ref;
2811 
2812                                 ref = &head->node;
2813                                 atomic_inc(&ref->refs);
2814 
2815                                 spin_unlock(&delayed_refs->lock);
2816                                 /*
2817                                  * Mutex was contended, block until it's
2818                                  * released and try again
2819                                  */
2820                                 mutex_lock(&head->mutex);
2821                                 mutex_unlock(&head->mutex);
2822 
2823                                 btrfs_put_delayed_ref(ref);
2824                                 cond_resched();
2825                                 goto again;
2826                         } else {
2827                                 WARN_ON(1);
2828                         }
2829                         node = rb_next(node);
2830                 }
2831                 spin_unlock(&delayed_refs->lock);
2832                 cond_resched();
2833                 goto again;
2834         }
2835 out:
2836         ret = btrfs_delayed_qgroup_accounting(trans, root->fs_info);
2837         if (ret)
2838                 return ret;
2839         assert_qgroups_uptodate(trans);
2840         return 0;
2841 }
2842 
2843 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2844                                 struct btrfs_root *root,
2845                                 u64 bytenr, u64 num_bytes, u64 flags,
2846                                 int level, int is_data)
2847 {
2848         struct btrfs_delayed_extent_op *extent_op;
2849         int ret;
2850 
2851         extent_op = btrfs_alloc_delayed_extent_op();
2852         if (!extent_op)
2853                 return -ENOMEM;
2854 
2855         extent_op->flags_to_set = flags;
2856         extent_op->update_flags = 1;
2857         extent_op->update_key = 0;
2858         extent_op->is_data = is_data ? 1 : 0;
2859         extent_op->level = level;
2860 
2861         ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2862                                           num_bytes, extent_op);
2863         if (ret)
2864                 btrfs_free_delayed_extent_op(extent_op);
2865         return ret;
2866 }
2867 
2868 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2869                                       struct btrfs_root *root,
2870                                       struct btrfs_path *path,
2871                                       u64 objectid, u64 offset, u64 bytenr)
2872 {
2873         struct btrfs_delayed_ref_head *head;
2874         struct btrfs_delayed_ref_node *ref;
2875         struct btrfs_delayed_data_ref *data_ref;
2876         struct btrfs_delayed_ref_root *delayed_refs;
2877         struct rb_node *node;
2878         int ret = 0;
2879 
2880         delayed_refs = &trans->transaction->delayed_refs;
2881         spin_lock(&delayed_refs->lock);
2882         head = btrfs_find_delayed_ref_head(trans, bytenr);
2883         if (!head) {
2884                 spin_unlock(&delayed_refs->lock);
2885                 return 0;
2886         }
2887 
2888         if (!mutex_trylock(&head->mutex)) {
2889                 atomic_inc(&head->node.refs);
2890                 spin_unlock(&delayed_refs->lock);
2891 
2892                 btrfs_release_path(path);
2893 
2894                 /*
2895                  * Mutex was contended, block until it's released and let
2896                  * caller try again
2897                  */
2898                 mutex_lock(&head->mutex);
2899                 mutex_unlock(&head->mutex);
2900                 btrfs_put_delayed_ref(&head->node);
2901                 return -EAGAIN;
2902         }
2903         spin_unlock(&delayed_refs->lock);
2904 
2905         spin_lock(&head->lock);
2906         node = rb_first(&head->ref_root);
2907         while (node) {
2908                 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2909                 node = rb_next(node);
2910 
2911                 /* If it's a shared ref we know a cross reference exists */
2912                 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
2913                         ret = 1;
2914                         break;
2915                 }
2916 
2917                 data_ref = btrfs_delayed_node_to_data_ref(ref);
2918 
2919                 /*
2920                  * If our ref doesn't match the one we're currently looking at
2921                  * then we have a cross reference.
2922                  */
2923                 if (data_ref->root != root->root_key.objectid ||
2924                     data_ref->objectid != objectid ||
2925                     data_ref->offset != offset) {
2926                         ret = 1;
2927                         break;
2928                 }
2929         }
2930         spin_unlock(&head->lock);
2931         mutex_unlock(&head->mutex);
2932         return ret;
2933 }
2934 
2935 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2936                                         struct btrfs_root *root,
2937                                         struct btrfs_path *path,
2938                                         u64 objectid, u64 offset, u64 bytenr)
2939 {
2940         struct btrfs_root *extent_root = root->fs_info->extent_root;
2941         struct extent_buffer *leaf;
2942         struct btrfs_extent_data_ref *ref;
2943         struct btrfs_extent_inline_ref *iref;
2944         struct btrfs_extent_item *ei;
2945         struct btrfs_key key;
2946         u32 item_size;
2947         int ret;
2948 
2949         key.objectid = bytenr;
2950         key.offset = (u64)-1;
2951         key.type = BTRFS_EXTENT_ITEM_KEY;
2952 
2953         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2954         if (ret < 0)
2955                 goto out;
2956         BUG_ON(ret == 0); /* Corruption */
2957 
2958         ret = -ENOENT;
2959         if (path->slots[0] == 0)
2960                 goto out;
2961 
2962         path->slots[0]--;
2963         leaf = path->nodes[0];
2964         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2965 
2966         if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2967                 goto out;
2968 
2969         ret = 1;
2970         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2971 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2972         if (item_size < sizeof(*ei)) {
2973                 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2974                 goto out;
2975         }
2976 #endif
2977         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2978 
2979         if (item_size != sizeof(*ei) +
2980             btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2981                 goto out;
2982 
2983         if (btrfs_extent_generation(leaf, ei) <=
2984             btrfs_root_last_snapshot(&root->root_item))
2985                 goto out;
2986 
2987         iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2988         if (btrfs_extent_inline_ref_type(leaf, iref) !=
2989             BTRFS_EXTENT_DATA_REF_KEY)
2990                 goto out;
2991 
2992         ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2993         if (btrfs_extent_refs(leaf, ei) !=
2994             btrfs_extent_data_ref_count(leaf, ref) ||
2995             btrfs_extent_data_ref_root(leaf, ref) !=
2996             root->root_key.objectid ||
2997             btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2998             btrfs_extent_data_ref_offset(leaf, ref) != offset)
2999                 goto out;
3000 
3001         ret = 0;
3002 out:
3003         return ret;
3004 }
3005 
3006 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
3007                           struct btrfs_root *root,
3008                           u64 objectid, u64 offset, u64 bytenr)
3009 {
3010         struct btrfs_path *path;
3011         int ret;
3012         int ret2;
3013 
3014         path = btrfs_alloc_path();
3015         if (!path)
3016                 return -ENOENT;
3017 
3018         do {
3019                 ret = check_committed_ref(trans, root, path, objectid,
3020                                           offset, bytenr);
3021                 if (ret && ret != -ENOENT)
3022                         goto out;
3023 
3024                 ret2 = check_delayed_ref(trans, root, path, objectid,
3025                                          offset, bytenr);
3026         } while (ret2 == -EAGAIN);
3027 
3028         if (ret2 && ret2 != -ENOENT) {
3029                 ret = ret2;
3030                 goto out;
3031         }
3032 
3033         if (ret != -ENOENT || ret2 != -ENOENT)
3034                 ret = 0;
3035 out:
3036         btrfs_free_path(path);
3037         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3038                 WARN_ON(ret > 0);
3039         return ret;
3040 }
3041 
3042 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3043                            struct btrfs_root *root,
3044                            struct extent_buffer *buf,
3045                            int full_backref, int inc)
3046 {
3047         u64 bytenr;
3048         u64 num_bytes;
3049         u64 parent;
3050         u64 ref_root;
3051         u32 nritems;
3052         struct btrfs_key key;
3053         struct btrfs_file_extent_item *fi;
3054         int i;
3055         int level;
3056         int ret = 0;
3057         int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
3058                             u64, u64, u64, u64, u64, u64, int);
3059 
3060 
3061         if (btrfs_test_is_dummy_root(root))
3062                 return 0;
3063 
3064         ref_root = btrfs_header_owner(buf);
3065         nritems = btrfs_header_nritems(buf);
3066         level = btrfs_header_level(buf);
3067 
3068         if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0)
3069                 return 0;
3070 
3071         if (inc)
3072                 process_func = btrfs_inc_extent_ref;
3073         else
3074                 process_func = btrfs_free_extent;
3075 
3076         if (full_backref)
3077                 parent = buf->start;
3078         else
3079                 parent = 0;
3080 
3081         for (i = 0; i < nritems; i++) {
3082                 if (level == 0) {
3083                         btrfs_item_key_to_cpu(buf, &key, i);
3084                         if (key.type != BTRFS_EXTENT_DATA_KEY)
3085                                 continue;
3086                         fi = btrfs_item_ptr(buf, i,
3087                                             struct btrfs_file_extent_item);
3088                         if (btrfs_file_extent_type(buf, fi) ==
3089                             BTRFS_FILE_EXTENT_INLINE)
3090                                 continue;
3091                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3092                         if (bytenr == 0)
3093                                 continue;
3094 
3095                         num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3096                         key.offset -= btrfs_file_extent_offset(buf, fi);
3097                         ret = process_func(trans, root, bytenr, num_bytes,
3098                                            parent, ref_root, key.objectid,
3099                                            key.offset, 1);
3100                         if (ret)
3101                                 goto fail;
3102                 } else {
3103                         bytenr = btrfs_node_blockptr(buf, i);
3104                         num_bytes = root->nodesize;
3105                         ret = process_func(trans, root, bytenr, num_bytes,
3106                                            parent, ref_root, level - 1, 0,
3107                                            1);
3108                         if (ret)
3109                                 goto fail;
3110                 }
3111         }
3112         return 0;
3113 fail:
3114         return ret;
3115 }
3116 
3117 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3118                   struct extent_buffer *buf, int full_backref)
3119 {
3120         return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
3121 }
3122 
3123 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3124                   struct extent_buffer *buf, int full_backref)
3125 {
3126         return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
3127 }
3128 
3129 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3130                                  struct btrfs_root *root,
3131                                  struct btrfs_path *path,
3132                                  struct btrfs_block_group_cache *cache)
3133 {
3134         int ret;
3135         struct btrfs_root *extent_root = root->fs_info->extent_root;
3136         unsigned long bi;
3137         struct extent_buffer *leaf;
3138 
3139         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3140         if (ret) {
3141                 if (ret > 0)
3142                         ret = -ENOENT;
3143                 goto fail;
3144         }
3145 
3146         leaf = path->nodes[0];
3147         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3148         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3149         btrfs_mark_buffer_dirty(leaf);
3150         btrfs_release_path(path);
3151 fail:
3152         if (ret)
3153                 btrfs_abort_transaction(trans, root, ret);
3154         return ret;
3155 
3156 }
3157 
3158 static struct btrfs_block_group_cache *
3159 next_block_group(struct btrfs_root *root,
3160                  struct btrfs_block_group_cache *cache)
3161 {
3162         struct rb_node *node;
3163 
3164         spin_lock(&root->fs_info->block_group_cache_lock);
3165 
3166         /* If our block group was removed, we need a full search. */
3167         if (RB_EMPTY_NODE(&cache->cache_node)) {
3168                 const u64 next_bytenr = cache->key.objectid + cache->key.offset;
3169 
3170                 spin_unlock(&root->fs_info->block_group_cache_lock);
3171                 btrfs_put_block_group(cache);
3172                 cache = btrfs_lookup_first_block_group(root->fs_info,
3173                                                        next_bytenr);
3174                 return cache;
3175         }
3176         node = rb_next(&cache->cache_node);
3177         btrfs_put_block_group(cache);
3178         if (node) {
3179                 cache = rb_entry(node, struct btrfs_block_group_cache,
3180                                  cache_node);
3181                 btrfs_get_block_group(cache);
3182         } else
3183                 cache = NULL;
3184         spin_unlock(&root->fs_info->block_group_cache_lock);
3185         return cache;
3186 }
3187 
3188 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3189                             struct btrfs_trans_handle *trans,
3190                             struct btrfs_path *path)
3191 {
3192         struct btrfs_root *root = block_group->fs_info->tree_root;
3193         struct inode *inode = NULL;
3194         u64 alloc_hint = 0;
3195         int dcs = BTRFS_DC_ERROR;
3196         int num_pages = 0;
3197         int retries = 0;
3198         int ret = 0;
3199 
3200         /*
3201          * If this block group is smaller than 100 megs don't bother caching the
3202          * block group.
3203          */
3204         if (block_group->key.offset < (100 * 1024 * 1024)) {
3205                 spin_lock(&block_group->lock);
3206                 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3207                 spin_unlock(&block_group->lock);
3208                 return 0;
3209         }
3210 
3211         if (trans->aborted)
3212                 return 0;
3213 again:
3214         inode = lookup_free_space_inode(root, block_group, path);
3215         if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3216                 ret = PTR_ERR(inode);
3217                 btrfs_release_path(path);
3218                 goto out;
3219         }
3220 
3221         if (IS_ERR(inode)) {
3222                 BUG_ON(retries);
3223                 retries++;
3224 
3225                 if (block_group->ro)
3226                         goto out_free;
3227 
3228                 ret = create_free_space_inode(root, trans, block_group, path);
3229                 if (ret)
3230                         goto out_free;
3231                 goto again;
3232         }
3233 
3234         /* We've already setup this transaction, go ahead and exit */
3235         if (block_group->cache_generation == trans->transid &&
3236             i_size_read(inode)) {
3237                 dcs = BTRFS_DC_SETUP;
3238                 goto out_put;
3239         }
3240 
3241         /*
3242          * We want to set the generation to 0, that way if anything goes wrong
3243          * from here on out we know not to trust this cache when we load up next
3244          * time.
3245          */
3246         BTRFS_I(inode)->generation = 0;
3247         ret = btrfs_update_inode(trans, root, inode);
3248         if (ret) {
3249                 /*
3250                  * So theoretically we could recover from this, simply set the
3251                  * super cache generation to 0 so we know to invalidate the
3252                  * cache, but then we'd have to keep track of the block groups
3253                  * that fail this way so we know we _have_ to reset this cache
3254                  * before the next commit or risk reading stale cache.  So to
3255                  * limit our exposure to horrible edge cases lets just abort the
3256                  * transaction, this only happens in really bad situations
3257                  * anyway.
3258                  */
3259                 btrfs_abort_transaction(trans, root, ret);
3260                 goto out_put;
3261         }
3262         WARN_ON(ret);
3263 
3264         if (i_size_read(inode) > 0) {
3265                 ret = btrfs_check_trunc_cache_free_space(root,
3266                                         &root->fs_info->global_block_rsv);
3267                 if (ret)
3268                         goto out_put;
3269 
3270                 ret = btrfs_truncate_free_space_cache(root, trans, inode);
3271                 if (ret)
3272                         goto out_put;
3273         }
3274 
3275         spin_lock(&block_group->lock);
3276         if (block_group->cached != BTRFS_CACHE_FINISHED ||
3277             !btrfs_test_opt(root, SPACE_CACHE) ||
3278             block_group->delalloc_bytes) {
3279                 /*
3280                  * don't bother trying to write stuff out _if_
3281                  * a) we're not cached,
3282                  * b) we're with nospace_cache mount option.
3283                  */
3284                 dcs = BTRFS_DC_WRITTEN;
3285                 spin_unlock(&block_group->lock);
3286                 goto out_put;
3287         }
3288         spin_unlock(&block_group->lock);
3289 
3290         /*
3291          * Try to preallocate enough space based on how big the block group is.
3292          * Keep in mind this has to include any pinned space which could end up
3293          * taking up quite a bit since it's not folded into the other space
3294          * cache.
3295          */
3296         num_pages = (int)div64_u64(block_group->key.offset, 256 * 1024 * 1024);
3297         if (!num_pages)
3298                 num_pages = 1;
3299 
3300         num_pages *= 16;
3301         num_pages *= PAGE_CACHE_SIZE;
3302 
3303         ret = btrfs_check_data_free_space(inode, num_pages);
3304         if (ret)
3305                 goto out_put;
3306 
3307         ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3308                                               num_pages, num_pages,
3309                                               &alloc_hint);
3310         if (!ret)
3311                 dcs = BTRFS_DC_SETUP;
3312         btrfs_free_reserved_data_space(inode, num_pages);
3313 
3314 out_put:
3315         iput(inode);
3316 out_free:
3317         btrfs_release_path(path);
3318 out:
3319         spin_lock(&block_group->lock);
3320         if (!ret && dcs == BTRFS_DC_SETUP)
3321                 block_group->cache_generation = trans->transid;
3322         block_group->disk_cache_state = dcs;
3323         spin_unlock(&block_group->lock);
3324 
3325         return ret;
3326 }
3327 
3328 int btrfs_setup_space_cache(struct btrfs_trans_handle *trans,
3329                             struct btrfs_root *root)
3330 {
3331         struct btrfs_block_group_cache *cache, *tmp;
3332         struct btrfs_transaction *cur_trans = trans->transaction;
3333         struct btrfs_path *path;
3334 
3335         if (list_empty(&cur_trans->dirty_bgs) ||
3336             !btrfs_test_opt(root, SPACE_CACHE))
3337                 return 0;
3338 
3339         path = btrfs_alloc_path();
3340         if (!path)
3341                 return -ENOMEM;
3342 
3343         /* Could add new block groups, use _safe just in case */
3344         list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs,
3345                                  dirty_list) {
3346                 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3347                         cache_save_setup(cache, trans, path);
3348         }
3349 
3350         btrfs_free_path(path);
3351         return 0;
3352 }
3353 
3354 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3355                                    struct btrfs_root *root)
3356 {
3357         struct btrfs_block_group_cache *cache;
3358         struct btrfs_transaction *cur_trans = trans->transaction;
3359         int ret = 0;
3360         struct btrfs_path *path;
3361 
3362         if (list_empty(&cur_trans->dirty_bgs))
3363                 return 0;
3364 
3365         path = btrfs_alloc_path();
3366         if (!path)
3367                 return -ENOMEM;
3368 
3369         /*
3370          * We don't need the lock here since we are protected by the transaction
3371          * commit.  We want to do the cache_save_setup first and then run the
3372          * delayed refs to make sure we have the best chance at doing this all
3373          * in one shot.
3374          */
3375         while (!list_empty(&cur_trans->dirty_bgs)) {
3376                 cache = list_first_entry(&cur_trans->dirty_bgs,
3377                                          struct btrfs_block_group_cache,
3378                                          dirty_list);
3379                 list_del_init(&cache->dirty_list);
3380                 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3381                         cache_save_setup(cache, trans, path);
3382                 if (!ret)
3383                         ret = btrfs_run_delayed_refs(trans, root,
3384                                                      (unsigned long) -1);
3385                 if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP)
3386                         btrfs_write_out_cache(root, trans, cache, path);
3387                 if (!ret)
3388                         ret = write_one_cache_group(trans, root, path, cache);
3389                 btrfs_put_block_group(cache);
3390         }
3391 
3392         btrfs_free_path(path);
3393         return ret;
3394 }
3395 
3396 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3397 {
3398         struct btrfs_block_group_cache *block_group;
3399         int readonly = 0;
3400 
3401         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3402         if (!block_group || block_group->ro)
3403                 readonly = 1;
3404         if (block_group)
3405                 btrfs_put_block_group(block_group);
3406         return readonly;
3407 }
3408 
3409 static const char *alloc_name(u64 flags)
3410 {
3411         switch (flags) {
3412         case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
3413                 return "mixed";
3414         case BTRFS_BLOCK_GROUP_METADATA:
3415                 return "metadata";
3416         case BTRFS_BLOCK_GROUP_DATA:
3417                 return "data";
3418         case BTRFS_BLOCK_GROUP_SYSTEM:
3419                 return "system";
3420         default:
3421                 WARN_ON(1);
3422                 return "invalid-combination";
3423         };
3424 }
3425 
3426 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3427                              u64 total_bytes, u64 bytes_used,
3428                              struct btrfs_space_info **space_info)
3429 {
3430         struct btrfs_space_info *found;
3431         int i;
3432         int factor;
3433         int ret;
3434 
3435         if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3436                      BTRFS_BLOCK_GROUP_RAID10))
3437                 factor = 2;
3438         else
3439                 factor = 1;
3440 
3441         found = __find_space_info(info, flags);
3442         if (found) {
3443                 spin_lock(&found->lock);
3444                 found->total_bytes += total_bytes;
3445                 found->disk_total += total_bytes * factor;
3446                 found->bytes_used += bytes_used;
3447                 found->disk_used += bytes_used * factor;
3448                 found->full = 0;
3449                 spin_unlock(&found->lock);
3450                 *space_info = found;
3451                 return 0;
3452         }
3453         found = kzalloc(sizeof(*found), GFP_NOFS);
3454         if (!found)
3455                 return -ENOMEM;
3456 
3457         ret = percpu_counter_init(&found->total_bytes_pinned, 0, GFP_KERNEL);
3458         if (ret) {
3459                 kfree(found);
3460                 return ret;
3461         }
3462 
3463         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3464                 INIT_LIST_HEAD(&found->block_groups[i]);
3465         init_rwsem(&found->groups_sem);
3466         spin_lock_init(&found->lock);
3467         found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3468         found->total_bytes = total_bytes;
3469         found->disk_total = total_bytes * factor;
3470         found->bytes_used = bytes_used;
3471         found->disk_used = bytes_used * factor;
3472         found->bytes_pinned = 0;
3473         found->bytes_reserved = 0;
3474         found->bytes_readonly = 0;
3475         found->bytes_may_use = 0;
3476         found->full = 0;
3477         found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3478         found->chunk_alloc = 0;
3479         found->flush = 0;
3480         init_waitqueue_head(&found->wait);
3481         INIT_LIST_HEAD(&found->ro_bgs);
3482 
3483         ret = kobject_init_and_add(&found->kobj, &space_info_ktype,
3484                                     info->space_info_kobj, "%s",
3485                                     alloc_name(found->flags));
3486         if (ret) {
3487                 kfree(found);
3488                 return ret;
3489         }
3490 
3491         *space_info = found;
3492         list_add_rcu(&found->list, &info->space_info);
3493         if (flags & BTRFS_BLOCK_GROUP_DATA)
3494                 info->data_sinfo = found;
3495 
3496         return ret;
3497 }
3498 
3499 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3500 {
3501         u64 extra_flags = chunk_to_extended(flags) &
3502                                 BTRFS_EXTENDED_PROFILE_MASK;
3503 
3504         write_seqlock(&fs_info->profiles_lock);
3505         if (flags & BTRFS_BLOCK_GROUP_DATA)
3506                 fs_info->avail_data_alloc_bits |= extra_flags;
3507         if (flags & BTRFS_BLOCK_GROUP_METADATA)
3508                 fs_info->avail_metadata_alloc_bits |= extra_flags;
3509         if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3510                 fs_info->avail_system_alloc_bits |= extra_flags;
3511         write_sequnlock(&fs_info->profiles_lock);
3512 }
3513 
3514 /*
3515  * returns target flags in extended format or 0 if restripe for this
3516  * chunk_type is not in progress
3517  *
3518  * should be called with either volume_mutex or balance_lock held
3519  */
3520 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3521 {
3522         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3523         u64 target = 0;
3524 
3525         if (!bctl)
3526                 return 0;
3527 
3528         if (flags & BTRFS_BLOCK_GROUP_DATA &&
3529             bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3530                 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3531         } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3532                    bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3533                 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3534         } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3535                    bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3536                 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3537         }
3538 
3539         return target;
3540 }
3541 
3542 /*
3543  * @flags: available profiles in extended format (see ctree.h)
3544  *
3545  * Returns reduced profile in chunk format.  If profile changing is in
3546  * progress (either running or paused) picks the target profile (if it's
3547  * already available), otherwise falls back to plain reducing.
3548  */
3549 static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3550 {
3551         u64 num_devices = root->fs_info->fs_devices->rw_devices;
3552         u64 target;
3553         u64 tmp;
3554 
3555         /*
3556          * see if restripe for this chunk_type is in progress, if so
3557          * try to reduce to the target profile
3558          */
3559         spin_lock(&root->fs_info->balance_lock);
3560         target = get_restripe_target(root->fs_info, flags);
3561         if (target) {
3562                 /* pick target profile only if it's already available */
3563                 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3564                         spin_unlock(&root->fs_info->balance_lock);
3565                         return extended_to_chunk(target);
3566                 }
3567         }
3568         spin_unlock(&root->fs_info->balance_lock);
3569 
3570         /* First, mask out the RAID levels which aren't possible */
3571         if (num_devices == 1)
3572                 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0 |
3573                            BTRFS_BLOCK_GROUP_RAID5);
3574         if (num_devices < 3)
3575                 flags &= ~BTRFS_BLOCK_GROUP_RAID6;
3576         if (num_devices < 4)
3577                 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3578 
3579         tmp = flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3580                        BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID5 |
3581                        BTRFS_BLOCK_GROUP_RAID6 | BTRFS_BLOCK_GROUP_RAID10);
3582         flags &= ~tmp;
3583 
3584         if (tmp & BTRFS_BLOCK_GROUP_RAID6)
3585                 tmp = BTRFS_BLOCK_GROUP_RAID6;
3586         else if (tmp & BTRFS_BLOCK_GROUP_RAID5)
3587                 tmp = BTRFS_BLOCK_GROUP_RAID5;
3588         else if (tmp & BTRFS_BLOCK_GROUP_RAID10)
3589                 tmp = BTRFS_BLOCK_GROUP_RAID10;
3590         else if (tmp & BTRFS_BLOCK_GROUP_RAID1)
3591                 tmp = BTRFS_BLOCK_GROUP_RAID1;
3592         else if (tmp & BTRFS_BLOCK_GROUP_RAID0)
3593                 tmp = BTRFS_BLOCK_GROUP_RAID0;
3594 
3595         return extended_to_chunk(flags | tmp);
3596 }
3597 
3598 static u64 get_alloc_profile(struct btrfs_root *root, u64 orig_flags)
3599 {
3600         unsigned seq;
3601         u64 flags;
3602 
3603         do {
3604                 flags = orig_flags;
3605                 seq = read_seqbegin(&root->fs_info->profiles_lock);
3606 
3607                 if (flags & BTRFS_BLOCK_GROUP_DATA)
3608                         flags |= root->fs_info->avail_data_alloc_bits;
3609                 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3610                         flags |= root->fs_info->avail_system_alloc_bits;
3611                 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3612                         flags |= root->fs_info->avail_metadata_alloc_bits;
3613         } while (read_seqretry(&root->fs_info->profiles_lock, seq));
3614 
3615         return btrfs_reduce_alloc_profile(root, flags);
3616 }
3617 
3618 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3619 {
3620         u64 flags;
3621         u64 ret;
3622 
3623         if (data)
3624                 flags = BTRFS_BLOCK_GROUP_DATA;
3625         else if (root == root->fs_info->chunk_root)
3626                 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3627         else
3628                 flags = BTRFS_BLOCK_GROUP_METADATA;
3629 
3630         ret = get_alloc_profile(root, flags);
3631         return ret;
3632 }
3633 
3634 /*
3635  * This will check the space that the inode allocates from to make sure we have
3636  * enough space for bytes.
3637  */
3638 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3639 {
3640         struct btrfs_space_info *data_sinfo;
3641         struct btrfs_root *root = BTRFS_I(inode)->root;
3642         struct btrfs_fs_info *fs_info = root->fs_info;
3643         u64 used;
3644         int ret = 0, committed = 0, alloc_chunk = 1;
3645 
3646         /* make sure bytes are sectorsize aligned */
3647         bytes = ALIGN(bytes, root->sectorsize);
3648 
3649         if (btrfs_is_free_space_inode(inode)) {
3650                 committed = 1;
3651                 ASSERT(current->journal_info);
3652         }
3653 
3654         data_sinfo = fs_info->data_sinfo;
3655         if (!data_sinfo)
3656                 goto alloc;
3657 
3658 again:
3659         /* make sure we have enough space to handle the data first */
3660         spin_lock(&data_sinfo->lock);
3661         used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3662                 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3663                 data_sinfo->bytes_may_use;
3664 
3665         if (used + bytes > data_sinfo->total_bytes) {
3666                 struct btrfs_trans_handle *trans;
3667 
3668                 /*
3669                  * if we don't have enough free bytes in this space then we need
3670                  * to alloc a new chunk.
3671                  */
3672                 if (!data_sinfo->full && alloc_chunk) {
3673                         u64 alloc_target;
3674 
3675                         data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3676                         spin_unlock(&data_sinfo->lock);
3677 alloc:
3678                         alloc_target = btrfs_get_alloc_profile(root, 1);
3679                         /*
3680                          * It is ugly that we don't call nolock join
3681                          * transaction for the free space inode case here.
3682                          * But it is safe because we only do the data space
3683                          * reservation for the free space cache in the
3684                          * transaction context, the common join transaction
3685                          * just increase the counter of the current transaction
3686                          * handler, doesn't try to acquire the trans_lock of
3687                          * the fs.
3688                          */
3689                         trans = btrfs_join_transaction(root);
3690                         if (IS_ERR(trans))
3691                                 return PTR_ERR(trans);
3692 
3693                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3694                                              alloc_target,
3695                                              CHUNK_ALLOC_NO_FORCE);
3696                         btrfs_end_transaction(trans, root);
3697                         if (ret < 0) {
3698                                 if (ret != -ENOSPC)
3699                                         return ret;
3700                                 else
3701                                         goto commit_trans;
3702                         }
3703 
3704                         if (!data_sinfo)
3705                                 data_sinfo = fs_info->data_sinfo;
3706 
3707                         goto again;
3708                 }
3709 
3710                 /*
3711                  * If we don't have enough pinned space to deal with this
3712                  * allocation don't bother committing the transaction.
3713                  */
3714                 if (percpu_counter_compare(&data_sinfo->total_bytes_pinned,
3715                                            bytes) < 0)
3716                         committed = 1;
3717                 spin_unlock(&data_sinfo->lock);
3718 
3719                 /* commit the current transaction and try again */
3720 commit_trans:
3721                 if (!committed &&
3722                     !atomic_read(&root->fs_info->open_ioctl_trans)) {
3723                         committed = 1;
3724 
3725                         trans = btrfs_join_transaction(root);
3726                         if (IS_ERR(trans))
3727                                 return PTR_ERR(trans);
3728                         ret = btrfs_commit_transaction(trans, root);
3729                         if (ret)
3730                                 return ret;
3731                         goto again;
3732                 }
3733 
3734                 trace_btrfs_space_reservation(root->fs_info,
3735                                               "space_info:enospc",
3736                                               data_sinfo->flags, bytes, 1);
3737                 return -ENOSPC;
3738         }
3739         data_sinfo->bytes_may_use += bytes;
3740         trace_btrfs_space_reservation(root->fs_info, "space_info",
3741                                       data_sinfo->flags, bytes, 1);
3742         spin_unlock(&data_sinfo->lock);
3743 
3744         return 0;
3745 }
3746 
3747 /*
3748  * Called if we need to clear a data reservation for this inode.
3749  */
3750 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3751 {
3752         struct btrfs_root *root = BTRFS_I(inode)->root;
3753         struct btrfs_space_info *data_sinfo;
3754 
3755         /* make sure bytes are sectorsize aligned */
3756         bytes = ALIGN(bytes, root->sectorsize);
3757 
3758         data_sinfo = root->fs_info->data_sinfo;
3759         spin_lock(&data_sinfo->lock);
3760         WARN_ON(data_sinfo->bytes_may_use < bytes);
3761         data_sinfo->bytes_may_use -= bytes;
3762         trace_btrfs_space_reservation(root->fs_info, "space_info",
3763                                       data_sinfo->flags, bytes, 0);
3764         spin_unlock(&data_sinfo->lock);
3765 }
3766 
3767 static void force_metadata_allocation(struct btrfs_fs_info *info)
3768 {
3769         struct list_head *head = &info->space_info;
3770         struct btrfs_space_info *found;
3771 
3772         rcu_read_lock();
3773         list_for_each_entry_rcu(found, head, list) {
3774                 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3775                         found->force_alloc = CHUNK_ALLOC_FORCE;
3776         }
3777         rcu_read_unlock();
3778 }
3779 
3780 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
3781 {
3782         return (global->size << 1);
3783 }
3784 
3785 static int should_alloc_chunk(struct btrfs_root *root,
3786                               struct btrfs_space_info *sinfo, int force)
3787 {
3788         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3789         u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3790         u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3791         u64 thresh;
3792 
3793         if (force == CHUNK_ALLOC_FORCE)
3794                 return 1;
3795 
3796         /*
3797          * We need to take into account the global rsv because for all intents
3798          * and purposes it's used space.  Don't worry about locking the
3799          * global_rsv, it doesn't change except when the transaction commits.
3800          */
3801         if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
3802                 num_allocated += calc_global_rsv_need_space(global_rsv);
3803 
3804         /*
3805          * in limited mode, we want to have some free space up to
3806          * about 1% of the FS size.
3807          */
3808         if (force == CHUNK_ALLOC_LIMITED) {
3809                 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3810                 thresh = max_t(u64, 64 * 1024 * 1024,
3811                                div_factor_fine(thresh, 1));
3812 
3813                 if (num_bytes - num_allocated < thresh)
3814                         return 1;
3815         }
3816 
3817         if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8))
3818                 return 0;
3819         return 1;
3820 }
3821 
3822 static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type)
3823 {
3824         u64 num_dev;
3825 
3826         if (type & (BTRFS_BLOCK_GROUP_RAID10 |
3827                     BTRFS_BLOCK_GROUP_RAID0 |
3828                     BTRFS_BLOCK_GROUP_RAID5 |
3829                     BTRFS_BLOCK_GROUP_RAID6))
3830                 num_dev = root->fs_info->fs_devices->rw_devices;
3831         else if (type & BTRFS_BLOCK_GROUP_RAID1)
3832                 num_dev = 2;
3833         else
3834                 num_dev = 1;    /* DUP or single */
3835 
3836         /* metadata for updaing devices and chunk tree */
3837         return btrfs_calc_trans_metadata_size(root, num_dev + 1);
3838 }
3839 
3840 static void check_system_chunk(struct btrfs_trans_handle *trans,
3841                                struct btrfs_root *root, u64 type)
3842 {
3843         struct btrfs_space_info *info;
3844         u64 left;
3845         u64 thresh;
3846 
3847         info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3848         spin_lock(&info->lock);
3849         left = info->total_bytes - info->bytes_used - info->bytes_pinned -
3850                 info->bytes_reserved - info->bytes_readonly;
3851         spin_unlock(&info->lock);
3852 
3853         thresh = get_system_chunk_thresh(root, type);
3854         if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
3855                 btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu",
3856                         left, thresh, type);
3857                 dump_space_info(info, 0, 0);
3858         }
3859 
3860         if (left < thresh) {
3861                 u64 flags;
3862 
3863                 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
3864                 btrfs_alloc_chunk(trans, root, flags);
3865         }
3866 }
3867 
3868 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3869                           struct btrfs_root *extent_root, u64 flags, int force)
3870 {
3871         struct btrfs_space_info *space_info;
3872         struct btrfs_fs_info *fs_info = extent_root->fs_info;
3873         int wait_for_alloc = 0;
3874         int ret = 0;
3875 
3876         /* Don't re-enter if we're already allocating a chunk */
3877         if (trans->allocating_chunk)
3878                 return -ENOSPC;
3879 
3880         space_info = __find_space_info(extent_root->fs_info, flags);
3881         if (!space_info) {
3882                 ret = update_space_info(extent_root->fs_info, flags,
3883                                         0, 0, &space_info);
3884                 BUG_ON(ret); /* -ENOMEM */
3885         }
3886         BUG_ON(!space_info); /* Logic error */
3887 
3888 again:
3889         spin_lock(&space_info->lock);
3890         if (force < space_info->force_alloc)
3891                 force = space_info->force_alloc;
3892         if (space_info->full) {
3893                 if (should_alloc_chunk(extent_root, space_info, force))
3894                         ret = -ENOSPC;
3895                 else
3896                         ret = 0;
3897                 spin_unlock(&space_info->lock);
3898                 return ret;
3899         }
3900 
3901         if (!should_alloc_chunk(extent_root, space_info, force)) {
3902                 spin_unlock(&space_info->lock);
3903                 return 0;
3904         } else if (space_info->chunk_alloc) {
3905                 wait_for_alloc = 1;
3906         } else {
3907                 space_info->chunk_alloc = 1;
3908         }
3909 
3910         spin_unlock(&space_info->lock);
3911 
3912         mutex_lock(&fs_info->chunk_mutex);
3913 
3914         /*
3915          * The chunk_mutex is held throughout the entirety of a chunk
3916          * allocation, so once we've acquired the chunk_mutex we know that the
3917          * other guy is done and we need to recheck and see if we should
3918          * allocate.
3919          */
3920         if (wait_for_alloc) {
3921                 mutex_unlock(&fs_info->chunk_mutex);
3922                 wait_for_alloc = 0;
3923                 goto again;
3924         }
3925 
3926         trans->allocating_chunk = true;
3927 
3928         /*
3929          * If we have mixed data/metadata chunks we want to make sure we keep
3930          * allocating mixed chunks instead of individual chunks.
3931          */
3932         if (btrfs_mixed_space_info(space_info))
3933                 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3934 
3935         /*
3936          * if we're doing a data chunk, go ahead and make sure that
3937          * we keep a reasonable number of metadata chunks allocated in the
3938          * FS as well.
3939          */
3940         if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3941                 fs_info->data_chunk_allocations++;
3942                 if (!(fs_info->data_chunk_allocations %
3943                       fs_info->metadata_ratio))
3944                         force_metadata_allocation(fs_info);
3945         }
3946 
3947         /*
3948          * Check if we have enough space in SYSTEM chunk because we may need
3949          * to update devices.
3950          */
3951         check_system_chunk(trans, extent_root, flags);
3952 
3953         ret = btrfs_alloc_chunk(trans, extent_root, flags);
3954         trans->allocating_chunk = false;
3955 
3956         spin_lock(&space_info->lock);
3957         if (ret < 0 && ret != -ENOSPC)
3958                 goto out;
3959         if (ret)
3960                 space_info->full = 1;
3961         else
3962                 ret = 1;
3963 
3964         space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3965 out:
3966         space_info->chunk_alloc = 0;
3967         spin_unlock(&space_info->lock);
3968         mutex_unlock(&fs_info->chunk_mutex);
3969         return ret;
3970 }
3971 
3972 static int can_overcommit(struct btrfs_root *root,
3973                           struct btrfs_space_info *space_info, u64 bytes,
3974                           enum btrfs_reserve_flush_enum flush)
3975 {
3976         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3977         u64 profile = btrfs_get_alloc_profile(root, 0);
3978         u64 space_size;
3979         u64 avail;
3980         u64 used;
3981 
3982         used = space_info->bytes_used + space_info->bytes_reserved +
3983                 space_info->bytes_pinned + space_info->bytes_readonly;
3984 
3985         /*
3986          * We only want to allow over committing if we have lots of actual space
3987          * free, but if we don't have enough space to handle the global reserve
3988          * space then we could end up having a real enospc problem when trying
3989          * to allocate a chunk or some other such important allocation.
3990          */
3991         spin_lock(&global_rsv->lock);
3992         space_size = calc_global_rsv_need_space(global_rsv);
3993         spin_unlock(&global_rsv->lock);
3994         if (used + space_size >= space_info->total_bytes)
3995                 return 0;
3996 
3997         used += space_info->bytes_may_use;
3998 
3999         spin_lock(&root->fs_info->free_chunk_lock);
4000         avail = root->fs_info->free_chunk_space;
4001         spin_unlock(&root->fs_info->free_chunk_lock);
4002 
4003         /*
4004          * If we have dup, raid1 or raid10 then only half of the free
4005          * space is actually useable.  For raid56, the space info used
4006          * doesn't include the parity drive, so we don't have to
4007          * change the math
4008          */
4009         if (profile & (BTRFS_BLOCK_GROUP_DUP |
4010                        BTRFS_BLOCK_GROUP_RAID1 |
4011                        BTRFS_BLOCK_GROUP_RAID10))
4012                 avail >>= 1;
4013 
4014         /*
4015          * If we aren't flushing all things, let us overcommit up to
4016          * 1/2th of the space. If we can flush, don't let us overcommit
4017          * too much, let it overcommit up to 1/8 of the space.
4018          */
4019         if (flush == BTRFS_RESERVE_FLUSH_ALL)
4020                 avail >>= 3;
4021         else
4022                 avail >>= 1;
4023 
4024         if (used + bytes < space_info->total_bytes + avail)
4025                 return 1;
4026         return 0;
4027 }
4028 
4029 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
4030                                          unsigned long nr_pages, int nr_items)
4031 {
4032         struct super_block *sb = root->fs_info->sb;
4033 
4034         if (down_read_trylock(&sb->s_umount)) {
4035                 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4036                 up_read(&sb->s_umount);
4037         } else {
4038                 /*
4039                  * We needn't worry the filesystem going from r/w to r/o though
4040                  * we don't acquire ->s_umount mutex, because the filesystem
4041                  * should guarantee the delalloc inodes list be empty after
4042                  * the filesystem is readonly(all dirty pages are written to
4043                  * the disk).
4044                  */
4045                 btrfs_start_delalloc_roots(root->fs_info, 0, nr_items);
4046                 if (!current->journal_info)
4047                         btrfs_wait_ordered_roots(root->fs_info, nr_items);
4048         }
4049 }
4050 
4051 static inline int calc_reclaim_items_nr(struct btrfs_root *root, u64 to_reclaim)
4052 {
4053         u64 bytes;
4054         int nr;
4055 
4056         bytes = btrfs_calc_trans_metadata_size(root, 1);
4057         nr = (int)div64_u64(to_reclaim, bytes);
4058         if (!nr)
4059                 nr = 1;
4060         return nr;
4061 }
4062 
4063 #define EXTENT_SIZE_PER_ITEM    (256 * 1024)
4064 
4065 /*
4066  * shrink metadata reservation for delalloc
4067  */
4068 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
4069                             bool wait_ordered)
4070 {
4071         struct btrfs_block_rsv *block_rsv;
4072         struct btrfs_space_info *space_info;
4073         struct btrfs_trans_handle *trans;
4074         u64 delalloc_bytes;
4075         u64 max_reclaim;
4076         long time_left;
4077         unsigned long nr_pages;
4078         int loops;
4079         int items;
4080         enum btrfs_reserve_flush_enum flush;
4081 
4082         /* Calc the number of the pages we need flush for space reservation */
4083         items = calc_reclaim_items_nr(root, to_reclaim);
4084         to_reclaim = items * EXTENT_SIZE_PER_ITEM;
4085 
4086         trans = (struct btrfs_trans_handle *)current->journal_info;
4087         block_rsv = &root->fs_info->delalloc_block_rsv;
4088         space_info = block_rsv->space_info;
4089 
4090         delalloc_bytes = percpu_counter_sum_positive(
4091                                                 &root->fs_info->delalloc_bytes);
4092         if (delalloc_bytes == 0) {
4093                 if (trans)
4094                         return;
4095                 if (wait_ordered)
4096                         btrfs_wait_ordered_roots(root->fs_info, items);
4097                 return;
4098         }
4099 
4100         loops = 0;
4101         while (delalloc_bytes && loops < 3) {
4102                 max_reclaim = min(delalloc_bytes, to_reclaim);
4103                 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
4104                 btrfs_writeback_inodes_sb_nr(root, nr_pages, items);
4105                 /*
4106                  * We need to wait for the async pages to actually start before
4107                  * we do anything.
4108                  */
4109                 max_reclaim = atomic_read(&root->fs_info->async_delalloc_pages);
4110                 if (!max_reclaim)
4111                         goto skip_async;
4112 
4113                 if (max_reclaim <= nr_pages)
4114                         max_reclaim = 0;
4115                 else
4116                         max_reclaim -= nr_pages;
4117 
4118                 wait_event(root->fs_info->async_submit_wait,
4119                            atomic_read(&root->fs_info->async_delalloc_pages) <=
4120                            (int)max_reclaim);
4121 skip_async:
4122                 if (!trans)
4123                         flush = BTRFS_RESERVE_FLUSH_ALL;
4124                 else
4125                         flush = BTRFS_RESERVE_NO_FLUSH;
4126                 spin_lock(&space_info->lock);
4127                 if (can_overcommit(root, space_info, orig, flush)) {
4128                         spin_unlock(&space_info->lock);
4129                         break;
4130                 }
4131                 spin_unlock(&space_info->lock);
4132 
4133                 loops++;
4134                 if (wait_ordered && !trans) {
4135                         btrfs_wait_ordered_roots(root->fs_info, items);
4136                 } else {
4137                         time_left = schedule_timeout_killable(1);
4138                         if (time_left)
4139                                 break;
4140                 }
4141                 delalloc_bytes = percpu_counter_sum_positive(
4142                                                 &root->fs_info->delalloc_bytes);
4143         }
4144 }
4145 
4146 /**
4147  * maybe_commit_transaction - possibly commit the transaction if its ok to
4148  * @root - the root we're allocating for
4149  * @bytes - the number of bytes we want to reserve
4150  * @force - force the commit
4151  *
4152  * This will check to make sure that committing the transaction will actually
4153  * get us somewhere and then commit the transaction if it does.  Otherwise it
4154  * will return -ENOSPC.
4155  */
4156 static int may_commit_transaction(struct btrfs_root *root,
4157                                   struct btrfs_space_info *space_info,
4158                                   u64 bytes, int force)
4159 {
4160         struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
4161         struct btrfs_trans_handle *trans;
4162 
4163         trans = (struct btrfs_trans_handle *)current->journal_info;
4164         if (trans)
4165                 return -EAGAIN;
4166 
4167         if (force)
4168                 goto commit;
4169 
4170         /* See if there is enough pinned space to make this reservation */
4171         if (percpu_counter_compare(&space_info->total_bytes_pinned,
4172                                    bytes) >= 0)
4173                 goto commit;
4174 
4175         /*
4176          * See if there is some space in the delayed insertion reservation for
4177          * this reservation.
4178          */
4179         if (space_info != delayed_rsv->space_info)
4180                 return -ENOSPC;
4181 
4182         spin_lock(&delayed_rsv->lock);
4183         if (percpu_counter_compare(&space_info->total_bytes_pinned,
4184                                    bytes - delayed_rsv->size) >= 0) {
4185                 spin_unlock(&delayed_rsv->lock);
4186                 return -ENOSPC;
4187         }
4188         spin_unlock(&delayed_rsv->lock);
4189 
4190 commit:
4191         trans = btrfs_join_transaction(root);
4192         if (IS_ERR(trans))
4193                 return -ENOSPC;
4194 
4195         return btrfs_commit_transaction(trans, root);
4196 }
4197 
4198 enum flush_state {
4199         FLUSH_DELAYED_ITEMS_NR  =       1,
4200         FLUSH_DELAYED_ITEMS     =       2,
4201         FLUSH_DELALLOC          =       3,
4202         FLUSH_DELALLOC_WAIT     =       4,
4203         ALLOC_CHUNK             =       5,
4204         COMMIT_TRANS            =       6,
4205 };
4206 
4207 static int flush_space(struct btrfs_root *root,
4208                        struct btrfs_space_info *space_info, u64 num_bytes,
4209                        u64 orig_bytes, int state)
4210 {
4211         struct btrfs_trans_handle *trans;
4212         int nr;
4213         int ret = 0;
4214 
4215         switch (state) {
4216         case FLUSH_DELAYED_ITEMS_NR:
4217         case FLUSH_DELAYED_ITEMS:
4218                 if (state == FLUSH_DELAYED_ITEMS_NR)
4219                         nr = calc_reclaim_items_nr(root, num_bytes) * 2;
4220                 else
4221                         nr = -1;
4222 
4223                 trans = btrfs_join_transaction(root);
4224                 if (IS_ERR(trans)) {
4225                         ret = PTR_ERR(trans);
4226                         break;
4227                 }
4228                 ret = btrfs_run_delayed_items_nr(trans, root, nr);
4229                 btrfs_end_transaction(trans, root);
4230                 break;
4231         case FLUSH_DELALLOC:
4232         case FLUSH_DELALLOC_WAIT:
4233                 shrink_delalloc(root, num_bytes * 2, orig_bytes,
4234                                 state == FLUSH_DELALLOC_WAIT);
4235                 break;
4236         case ALLOC_CHUNK:
4237                 trans = btrfs_join_transaction(root);
4238                 if (IS_ERR(trans)) {
4239                         ret = PTR_ERR(trans);
4240                         break;
4241                 }
4242                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4243                                      btrfs_get_alloc_profile(root, 0),
4244                                      CHUNK_ALLOC_NO_FORCE);
4245                 btrfs_end_transaction(trans, root);
4246                 if (ret == -ENOSPC)
4247                         ret = 0;
4248                 break;
4249         case COMMIT_TRANS:
4250                 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
4251                 break;
4252         default:
4253                 ret = -ENOSPC;
4254                 break;
4255         }
4256 
4257         return ret;
4258 }
4259 
4260 static inline u64
4261 btrfs_calc_reclaim_metadata_size(struct btrfs_root *root,
4262                                  struct btrfs_space_info *space_info)
4263 {
4264         u64 used;
4265         u64 expected;
4266         u64 to_reclaim;
4267 
4268         to_reclaim = min_t(u64, num_online_cpus() * 1024 * 1024,
4269                                 16 * 1024 * 1024);
4270         spin_lock(&space_info->lock);
4271         if (can_overcommit(root, space_info, to_reclaim,
4272                            BTRFS_RESERVE_FLUSH_ALL)) {
4273                 to_reclaim = 0;
4274                 goto out;
4275         }
4276 
4277         used = space_info->bytes_used + space_info->bytes_reserved +
4278                space_info->bytes_pinned + space_info->bytes_readonly +
4279                space_info->bytes_may_use;
4280         if (can_overcommit(root, space_info, 1024 * 1024,
4281                            BTRFS_RESERVE_FLUSH_ALL))
4282                 expected = div_factor_fine(space_info->total_bytes, 95);
4283         else
4284                 expected = div_factor_fine(space_info->total_bytes, 90);
4285 
4286         if (used > expected)
4287                 to_reclaim = used - expected;
4288         else
4289                 to_reclaim = 0;
4290         to_reclaim = min(to_reclaim, space_info->bytes_may_use +
4291                                      space_info->bytes_reserved);
4292 out:
4293         spin_unlock(&space_info->lock);
4294 
4295         return to_reclaim;
4296 }
4297 
4298 static inline int need_do_async_reclaim(struct btrfs_space_info *space_info,
4299                                         struct btrfs_fs_info *fs_info, u64 used)
4300 {
4301         return (used >= div_factor_fine(space_info->total_bytes, 98) &&
4302                 !btrfs_fs_closing(fs_info) &&
4303                 !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state));
4304 }
4305 
4306 static int btrfs_need_do_async_reclaim(struct btrfs_space_info *space_info,
4307                                        struct btrfs_fs_info *fs_info,
4308                                        int flush_state)
4309 {
4310         u64 used;
4311 
4312         spin_lock(&space_info->lock);
4313         /*
4314          * We run out of space and have not got any free space via flush_space,
4315          * so don't bother doing async reclaim.
4316          */
4317         if (flush_state > COMMIT_TRANS && space_info->full) {
4318                 spin_unlock(&space_info->lock);
4319                 return 0;
4320         }
4321 
4322         used = space_info->bytes_used + space_info->bytes_reserved +
4323                space_info->bytes_pinned + space_info->bytes_readonly +
4324                space_info->bytes_may_use;
4325         if (need_do_async_reclaim(space_info, fs_info, used)) {
4326                 spin_unlock(&space_info->lock);
4327                 return 1;
4328         }
4329         spin_unlock(&space_info->lock);
4330 
4331         return 0;
4332 }
4333 
4334 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
4335 {
4336         struct btrfs_fs_info *fs_info;
4337         struct btrfs_space_info *space_info;
4338         u64 to_reclaim;
4339         int flush_state;
4340 
4341         fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
4342         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4343 
4344         to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root,
4345                                                       space_info);
4346         if (!to_reclaim)
4347                 return;
4348 
4349         flush_state = FLUSH_DELAYED_ITEMS_NR;
4350         do {
4351                 flush_space(fs_info->fs_root, space_info, to_reclaim,
4352                             to_reclaim, flush_state);
4353                 flush_state++;
4354                 if (!btrfs_need_do_async_reclaim(space_info, fs_info,
4355                                                  flush_state))
4356                         return;
4357         } while (flush_state <= COMMIT_TRANS);
4358 
4359         if (btrfs_need_do_async_reclaim(space_info, fs_info, flush_state))
4360                 queue_work(system_unbound_wq, work);
4361 }
4362 
4363 void btrfs_init_async_reclaim_work(struct work_struct *work)
4364 {
4365         INIT_WORK(work, btrfs_async_reclaim_metadata_space);
4366 }
4367 
4368 /**
4369  * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4370  * @root - the root we're allocating for
4371  * @block_rsv - the block_rsv we're allocating for
4372  * @orig_bytes - the number of bytes we want
4373  * @flush - whether or not we can flush to make our reservation
4374  *
4375  * This will reserve orgi_bytes number of bytes from the space info associated
4376  * with the block_rsv.  If there is not enough space it will make an attempt to
4377  * flush out space to make room.  It will do this by flushing delalloc if
4378  * possible or committing the transaction.  If flush is 0 then no attempts to
4379  * regain reservations will be made and this will fail if there is not enough
4380  * space already.
4381  */
4382 static int reserve_metadata_bytes(struct btrfs_root *root,
4383                                   struct btrfs_block_rsv *block_rsv,
4384                                   u64 orig_bytes,
4385                                   enum btrfs_reserve_flush_enum flush)
4386 {
4387         struct btrfs_space_info *space_info = block_rsv->space_info;
4388         u64 used;
4389         u64 num_bytes = orig_bytes;
4390         int flush_state = FLUSH_DELAYED_ITEMS_NR;
4391         int ret = 0;
4392         bool flushing = false;
4393 
4394 again:
4395         ret = 0;
4396         spin_lock(&space_info->lock);
4397         /*
4398          * We only want to wait if somebody other than us is flushing and we
4399          * are actually allowed to flush all things.
4400          */
4401         while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
4402                space_info->flush) {
4403                 spin_unlock(&space_info->lock);
4404                 /*
4405                  * If we have a trans handle we can't wait because the flusher
4406                  * may have to commit the transaction, which would mean we would
4407                  * deadlock since we are waiting for the flusher to finish, but
4408                  * hold the current transaction open.
4409                  */
4410                 if (current->journal_info)
4411                         return -EAGAIN;
4412                 ret = wait_event_killable(space_info->wait, !space_info->flush);
4413                 /* Must have been killed, return */
4414                 if (ret)
4415                         return -EINTR;
4416 
4417                 spin_lock(&space_info->lock);
4418         }
4419 
4420         ret = -ENOSPC;
4421         used = space_info->bytes_used + space_info->bytes_reserved +
4422                 space_info->bytes_pinned + space_info->bytes_readonly +
4423                 space_info->bytes_may_use;
4424 
4425         /*
4426          * The idea here is that we've not already over-reserved the block group
4427          * then we can go ahead and save our reservation first and then start
4428          * flushing if we need to.  Otherwise if we've already overcommitted
4429          * lets start flushing stuff first and then come back and try to make
4430          * our reservation.
4431          */
4432         if (used <= space_info->total_bytes) {
4433                 if (used + orig_bytes <= space_info->total_bytes) {
4434                         space_info->bytes_may_use += orig_bytes;
4435                         trace_btrfs_space_reservation(root->fs_info,
4436                                 "space_info", space_info->flags, orig_bytes, 1);
4437                         ret = 0;
4438                 } else {
4439                         /*
4440                          * Ok set num_bytes to orig_bytes since we aren't
4441                          * overocmmitted, this way we only try and reclaim what
4442                          * we need.
4443                          */
4444                         num_bytes = orig_bytes;
4445                 }
4446         } else {
4447                 /*
4448                  * Ok we're over committed, set num_bytes to the overcommitted
4449                  * amount plus the amount of bytes that we need for this
4450                  * reservation.
4451                  */
4452                 num_bytes = used - space_info->total_bytes +
4453                         (orig_bytes * 2);
4454         }
4455 
4456         if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
4457                 space_info->bytes_may_use += orig_bytes;
4458                 trace_btrfs_space_reservation(root->fs_info, "space_info",
4459                                               space_info->flags, orig_bytes,
4460                                               1);
4461                 ret = 0;
4462         }
4463 
4464         /*
4465          * Couldn't make our reservation, save our place so while we're trying
4466          * to reclaim space we can actually use it instead of somebody else
4467          * stealing it from us.
4468          *
4469          * We make the other tasks wait for the flush only when we can flush
4470          * all things.
4471          */
4472         if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
4473                 flushing = true;
4474                 space_info->flush = 1;
4475         } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
4476                 used += orig_bytes;
4477                 /*
4478                  * We will do the space reservation dance during log replay,
4479                  * which means we won't have fs_info->fs_root set, so don't do
4480                  * the async reclaim as we will panic.
4481                  */
4482                 if (!root->fs_info->log_root_recovering &&
4483                     need_do_async_reclaim(space_info, root->fs_info, used) &&
4484                     !work_busy(&root->fs_info->async_reclaim_work))
4485                         queue_work(system_unbound_wq,
4486                                    &root->fs_info->async_reclaim_work);
4487         }
4488         spin_unlock(&space_info->lock);
4489 
4490         if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
4491                 goto out;
4492 
4493         ret = flush_space(root, space_info, num_bytes, orig_bytes,
4494                           flush_state);
4495         flush_state++;
4496 
4497         /*
4498          * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4499          * would happen. So skip delalloc flush.
4500          */
4501         if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4502             (flush_state == FLUSH_DELALLOC ||
4503              flush_state == FLUSH_DELALLOC_WAIT))
4504                 flush_state = ALLOC_CHUNK;
4505 
4506         if (!ret)
4507                 goto again;
4508         else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4509                  flush_state < COMMIT_TRANS)
4510                 goto again;
4511         else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
4512                  flush_state <= COMMIT_TRANS)
4513                 goto again;
4514 
4515 out:
4516         if (ret == -ENOSPC &&
4517             unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
4518                 struct btrfs_block_rsv *global_rsv =
4519                         &root->fs_info->global_block_rsv;
4520 
4521                 if (block_rsv != global_rsv &&
4522                     !block_rsv_use_bytes(global_rsv, orig_bytes))
4523                         ret = 0;
4524         }
4525         if (ret == -ENOSPC)
4526                 trace_btrfs_space_reservation(root->fs_info,
4527                                               "space_info:enospc",
4528                                               space_info->flags, orig_bytes, 1);
4529         if (flushing) {
4530                 spin_lock(&space_info->lock);
4531                 space_info->flush = 0;
4532                 wake_up_all(&space_info->wait);
4533                 spin_unlock(&space_info->lock);
4534         }
4535         return ret;
4536 }
4537 
4538 static struct btrfs_block_rsv *get_block_rsv(
4539                                         const struct btrfs_trans_handle *trans,
4540                                         const struct btrfs_root *root)
4541 {
4542         struct btrfs_block_rsv *block_rsv = NULL;
4543 
4544         if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4545                 block_rsv = trans->block_rsv;
4546 
4547         if (root == root->fs_info->csum_root && trans->adding_csums)
4548                 block_rsv = trans->block_rsv;
4549 
4550         if (root == root->fs_info->uuid_root)
4551                 block_rsv = trans->block_rsv;
4552 
4553         if (!block_rsv)
4554                 block_rsv = root->block_rsv;
4555 
4556         if (!block_rsv)
4557                 block_rsv = &root->fs_info->empty_block_rsv;
4558 
4559         return block_rsv;
4560 }
4561 
4562 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
4563                                u64 num_bytes)
4564 {
4565         int ret = -ENOSPC;
4566         spin_lock(&block_rsv->lock);
4567         if (block_rsv->reserved >= num_bytes) {
4568                 block_rsv->reserved -= num_bytes;
4569                 if (block_rsv->reserved < block_rsv->size)
4570                         block_rsv->full = 0;
4571                 ret = 0;
4572         }
4573         spin_unlock(&block_rsv->lock);
4574         return ret;
4575 }
4576 
4577 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
4578                                 u64 num_bytes, int update_size)
4579 {
4580         spin_lock(&block_rsv->lock);
4581         block_rsv->reserved += num_bytes;
4582         if (update_size)
4583                 block_rsv->size += num_bytes;
4584         else if (block_rsv->reserved >= block_rsv->size)
4585                 block_rsv->full = 1;
4586         spin_unlock(&block_rsv->lock);
4587 }
4588 
4589 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
4590                              struct btrfs_block_rsv *dest, u64 num_bytes,
4591                              int min_factor)
4592 {
4593         struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4594         u64 min_bytes;
4595 
4596         if (global_rsv->space_info != dest->space_info)
4597                 return -ENOSPC;
4598 
4599         spin_lock(&global_rsv->lock);
4600         min_bytes = div_factor(global_rsv->size, min_factor);
4601         if (global_rsv->reserved < min_bytes + num_bytes) {
4602                 spin_unlock(&global_rsv->lock);
4603                 return -ENOSPC;
4604         }
4605         global_rsv->reserved -= num_bytes;
4606         if (global_rsv->reserved < global_rsv->size)
4607                 global_rsv->full = 0;
4608         spin_unlock(&global_rsv->lock);
4609 
4610         block_rsv_add_bytes(dest, num_bytes, 1);
4611         return 0;
4612 }
4613 
4614 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
4615                                     struct btrfs_block_rsv *block_rsv,
4616                                     struct btrfs_block_rsv *dest, u64 num_bytes)
4617 {
4618         struct btrfs_space_info *space_info = block_rsv->space_info;
4619 
4620         spin_lock(&block_rsv->lock);
4621         if (num_bytes == (u64)-1)
4622                 num_bytes = block_rsv->size;
4623         block_rsv->size -= num_bytes;
4624         if (block_rsv->reserved >= block_rsv->size) {
4625                 num_bytes = block_rsv->reserved - block_rsv->size;
4626                 block_rsv->reserved = block_rsv->size;
4627                 block_rsv->full = 1;
4628         } else {
4629                 num_bytes = 0;
4630         }
4631         spin_unlock(&block_rsv->lock);
4632 
4633         if (num_bytes > 0) {
4634                 if (dest) {
4635                         spin_lock(&dest->lock);
4636                         if (!dest->full) {
4637                                 u64 bytes_to_add;
4638 
4639                                 bytes_to_add = dest->size - dest->reserved;
4640                                 bytes_to_add = min(num_bytes, bytes_to_add);
4641                                 dest->reserved += bytes_to_add;
4642                                 if (dest->reserved >= dest->size)
4643                                         dest->full = 1;
4644                                 num_bytes -= bytes_to_add;
4645                         }
4646                         spin_unlock(&dest->lock);
4647                 }
4648                 if (num_bytes) {
4649                         spin_lock(&space_info->lock);
4650                         space_info->bytes_may_use -= num_bytes;
4651                         trace_btrfs_space_reservation(fs_info, "space_info",
4652                                         space_info->flags, num_bytes, 0);
4653                         spin_unlock(&space_info->lock);
4654                 }
4655         }
4656 }
4657 
4658 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
4659                                    struct btrfs_block_rsv *dst, u64 num_bytes)
4660 {
4661         int ret;
4662 
4663         ret = block_rsv_use_bytes(src, num_bytes);
4664         if (ret)
4665                 return ret;
4666 
4667         block_rsv_add_bytes(dst, num_bytes, 1);
4668         return 0;
4669 }
4670 
4671 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
4672 {
4673         memset(rsv, 0, sizeof(*rsv));
4674         spin_lock_init(&rsv->lock);
4675         rsv->type = type;
4676 }
4677 
4678 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
4679                                               unsigned short type)
4680 {
4681         struct btrfs_block_rsv *block_rsv;
4682         struct btrfs_fs_info *fs_info = root->fs_info;
4683 
4684         block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
4685         if (!block_rsv)
4686                 return NULL;
4687 
4688         btrfs_init_block_rsv(block_rsv, type);
4689         block_rsv->space_info = __find_space_info(fs_info,
4690                                                   BTRFS_BLOCK_GROUP_METADATA);
4691         return block_rsv;
4692 }
4693 
4694 void btrfs_free_block_rsv(struct btrfs_root *root,
4695                           struct btrfs_block_rsv *rsv)
4696 {
4697         if (!rsv)
4698                 return;
4699         btrfs_block_rsv_release(root, rsv, (u64)-1);
4700         kfree(rsv);
4701 }
4702 
4703 int btrfs_block_rsv_add(struct btrfs_root *root,
4704                         struct btrfs_block_rsv *block_rsv, u64 num_bytes,
4705                         enum btrfs_reserve_flush_enum flush)
4706 {
4707         int ret;
4708 
4709         if (num_bytes == 0)
4710                 return 0;
4711 
4712         ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4713         if (!ret) {
4714                 block_rsv_add_bytes(block_rsv, num_bytes, 1);
4715                 return 0;
4716         }
4717 
4718         return ret;
4719 }
4720 
4721 int btrfs_block_rsv_check(struct btrfs_root *root,
4722                           struct btrfs_block_rsv *block_rsv, int min_factor)
4723 {
4724         u64 num_bytes = 0;
4725         int ret = -ENOSPC;
4726 
4727         if (!block_rsv)
4728                 return 0;
4729 
4730         spin_lock(&block_rsv->lock);
4731         num_bytes = div_factor(block_rsv->size, min_factor);
4732         if (block_rsv->reserved >= num_bytes)
4733                 ret = 0;
4734         spin_unlock(&block_rsv->lock);
4735 
4736         return ret;
4737 }
4738 
4739 int btrfs_block_rsv_refill(struct btrfs_root *root,
4740                            struct btrfs_block_rsv *block_rsv, u64 min_reserved,
4741                            enum btrfs_reserve_flush_enum flush)
4742 {
4743         u64 num_bytes = 0;
4744         int ret = -ENOSPC;
4745 
4746         if (!block_rsv)
4747                 return 0;
4748 
4749         spin_lock(&block_rsv->lock);
4750         num_bytes = min_reserved;
4751         if (block_rsv->reserved >= num_bytes)
4752                 ret = 0;
4753         else
4754                 num_bytes -= block_rsv->reserved;
4755         spin_unlock(&block_rsv->lock);
4756 
4757         if (!ret)
4758                 return 0;
4759 
4760         ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4761         if (!ret) {
4762                 block_rsv_add_bytes(block_rsv, num_bytes, 0);
4763                 return 0;
4764         }
4765 
4766         return ret;
4767 }
4768 
4769 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
4770                             struct btrfs_block_rsv *dst_rsv,
4771                             u64 num_bytes)
4772 {
4773         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4774 }
4775 
4776 void btrfs_block_rsv_release(struct btrfs_root *root,
4777                              struct btrfs_block_rsv *block_rsv,
4778                              u64 num_bytes)
4779 {
4780         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4781         if (global_rsv == block_rsv ||
4782             block_rsv->space_info != global_rsv->space_info)
4783                 global_rsv = NULL;
4784         block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
4785                                 num_bytes);
4786 }
4787 
4788 /*
4789  * helper to calculate size of global block reservation.
4790  * the desired value is sum of space used by extent tree,
4791  * checksum tree and root tree
4792  */
4793 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
4794 {
4795         struct btrfs_space_info *sinfo;
4796         u64 num_bytes;
4797         u64 meta_used;
4798         u64 data_used;
4799         int csum_size = btrfs_super_csum_size(fs_info->super_copy);
4800 
4801         sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
4802         spin_lock(&sinfo->lock);
4803         data_used = sinfo->bytes_used;
4804         spin_unlock(&sinfo->lock);
4805 
4806         sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4807         spin_lock(&sinfo->lock);
4808         if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
4809                 data_used = 0;
4810         meta_used = sinfo->bytes_used;
4811         spin_unlock(&sinfo->lock);
4812 
4813         num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
4814                     csum_size * 2;
4815         num_bytes += div64_u64(data_used + meta_used, 50);
4816 
4817         if (num_bytes * 3 > meta_used)
4818                 num_bytes = div64_u64(meta_used, 3);
4819 
4820         return ALIGN(num_bytes, fs_info->extent_root->nodesize << 10);
4821 }
4822 
4823 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
4824 {
4825         struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4826         struct btrfs_space_info *sinfo = block_rsv->space_info;
4827         u64 num_bytes;
4828 
4829         num_bytes = calc_global_metadata_size(fs_info);
4830 
4831         spin_lock(&sinfo->lock);
4832         spin_lock(&block_rsv->lock);
4833 
4834         block_rsv->size = min_t(u64, num_bytes, 512 * 1024 * 1024);
4835 
4836         num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
4837                     sinfo->bytes_reserved + sinfo->bytes_readonly +
4838                     sinfo->bytes_may_use;
4839 
4840         if (sinfo->total_bytes > num_bytes) {
4841                 num_bytes = sinfo->total_bytes - num_bytes;
4842                 block_rsv->reserved += num_bytes;
4843                 sinfo->bytes_may_use += num_bytes;
4844                 trace_btrfs_space_reservation(fs_info, "space_info",
4845                                       sinfo->flags, num_bytes, 1);
4846         }
4847 
4848         if (block_rsv->reserved >= block_rsv->size) {
4849                 num_bytes = block_rsv->reserved - block_rsv->size;
4850                 sinfo->bytes_may_use -= num_bytes;
4851                 trace_btrfs_space_reservation(fs_info, "space_info",
4852                                       sinfo->flags, num_bytes, 0);
4853                 block_rsv->reserved = block_rsv->size;
4854                 block_rsv->full = 1;
4855         }
4856 
4857         spin_unlock(&block_rsv->lock);
4858         spin_unlock(&sinfo->lock);
4859 }
4860 
4861 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
4862 {
4863         struct btrfs_space_info *space_info;
4864 
4865         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4866         fs_info->chunk_block_rsv.space_info = space_info;
4867 
4868         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4869         fs_info->global_block_rsv.space_info = space_info;
4870         fs_info->delalloc_block_rsv.space_info = space_info;
4871         fs_info->trans_block_rsv.space_info = space_info;
4872         fs_info->empty_block_rsv.space_info = space_info;
4873         fs_info->delayed_block_rsv.space_info = space_info;
4874 
4875         fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
4876         fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
4877         fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
4878         fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
4879         if (fs_info->quota_root)
4880                 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
4881         fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
4882 
4883         update_global_block_rsv(fs_info);
4884 }
4885 
4886 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
4887 {
4888         block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
4889                                 (u64)-1);
4890         WARN_ON(fs_info->delalloc_block_rsv.size > 0);
4891         WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
4892         WARN_ON(fs_info->trans_block_rsv.size > 0);
4893         WARN_ON(fs_info->trans_block_rsv.reserved > 0);
4894         WARN_ON(fs_info->chunk_block_rsv.size > 0);
4895         WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
4896         WARN_ON(fs_info->delayed_block_rsv.size > 0);
4897         WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
4898 }
4899 
4900 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4901                                   struct btrfs_root *root)
4902 {
4903         if (!trans->block_rsv)
4904                 return;
4905 
4906         if (!trans->bytes_reserved)
4907                 return;
4908 
4909         trace_btrfs_space_reservation(root->fs_info, "transaction",
4910                                       trans->transid, trans->bytes_reserved, 0);
4911         btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
4912         trans->bytes_reserved = 0;
4913 }
4914 
4915 /* Can only return 0 or -ENOSPC */
4916 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4917                                   struct inode *inode)
4918 {
4919         struct btrfs_root *root = BTRFS_I(inode)->root;
4920         struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4921         struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4922 
4923         /*
4924          * We need to hold space in order to delete our orphan item once we've
4925          * added it, so this takes the reservation so we can release it later
4926          * when we are truly done with the orphan item.
4927          */
4928         u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4929         trace_btrfs_space_reservation(root->fs_info, "orphan",
4930                                       btrfs_ino(inode), num_bytes, 1);
4931         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4932 }
4933 
4934 void btrfs_orphan_release_metadata(struct inode *inode)
4935 {
4936         struct btrfs_root *root = BTRFS_I(inode)->root;
4937         u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4938         trace_btrfs_space_reservation(root->fs_info, "orphan",
4939                                       btrfs_ino(inode), num_bytes, 0);
4940         btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4941 }
4942 
4943 /*
4944  * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
4945  * root: the root of the parent directory
4946  * rsv: block reservation
4947  * items: the number of items that we need do reservation
4948  * qgroup_reserved: used to return the reserved size in qgroup
4949  *
4950  * This function is used to reserve the space for snapshot/subvolume
4951  * creation and deletion. Those operations are different with the
4952  * common file/directory operations, they change two fs/file trees
4953  * and root tree, the number of items that the qgroup reserves is
4954  * different with the free space reservation. So we can not use
4955  * the space reseravtion mechanism in start_transaction().
4956  */
4957 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
4958                                      struct btrfs_block_rsv *rsv,
4959                                      int items,
4960                                      u64 *qgroup_reserved,
4961                                      bool use_global_rsv)
4962 {
4963         u64 num_bytes;
4964         int ret;
4965         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4966 
4967         if (root->fs_info->quota_enabled) {
4968                 /* One for parent inode, two for dir entries */
4969                 num_bytes = 3 * root->nodesize;
4970                 ret = btrfs_qgroup_reserve(root, num_bytes);
4971                 if (ret)
4972                         return ret;
4973         } else {
4974                 num_bytes = 0;
4975         }
4976 
4977         *qgroup_reserved = num_bytes;
4978 
4979         num_bytes = btrfs_calc_trans_metadata_size(root, items);
4980         rsv->space_info = __find_space_info(root->fs_info,
4981                                             BTRFS_BLOCK_GROUP_METADATA);
4982         ret = btrfs_block_rsv_add(root, rsv, num_bytes,
4983                                   BTRFS_RESERVE_FLUSH_ALL);
4984 
4985         if (ret == -ENOSPC && use_global_rsv)
4986                 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes);
4987 
4988         if (ret) {
4989                 if (*qgroup_reserved)
4990                         btrfs_qgroup_free(root, *qgroup_reserved);
4991         }
4992 
4993         return ret;
4994 }
4995 
4996 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
4997                                       struct btrfs_block_rsv *rsv,
4998                                       u64 qgroup_reserved)
4999 {
5000         btrfs_block_rsv_release(root, rsv, (u64)-1);
5001         if (qgroup_reserved)
5002                 btrfs_qgroup_free(root, qgroup_reserved);
5003 }
5004 
5005 /**
5006  * drop_outstanding_extent - drop an outstanding extent
5007  * @inode: the inode we're dropping the extent for
5008  * @num_bytes: the number of bytes we're relaseing.
5009  *
5010  * This is called when we are freeing up an outstanding extent, either called
5011  * after an error or after an extent is written.  This will return the number of
5012  * reserved extents that need to be freed.  This must be called with
5013  * BTRFS_I(inode)->lock held.
5014  */
5015 static unsigned drop_outstanding_extent(struct inode *inode, u64 num_bytes)
5016 {
5017         unsigned drop_inode_space = 0;
5018         unsigned dropped_extents = 0;
5019         unsigned num_extents = 0;
5020 
5021         num_extents = (unsigned)div64_u64(num_bytes +
5022                                           BTRFS_MAX_EXTENT_SIZE - 1,
5023                                           BTRFS_MAX_EXTENT_SIZE);
5024         ASSERT(num_extents);
5025         ASSERT(BTRFS_I(inode)->outstanding_extents >= num_extents);
5026         BTRFS_I(inode)->outstanding_extents -= num_extents;
5027 
5028         if (BTRFS_I(inode)->outstanding_extents == 0 &&
5029             test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5030                                &BTRFS_I(inode)->runtime_flags))
5031                 drop_inode_space = 1;
5032 
5033         /*
5034          * If we have more or the same amount of outsanding extents than we have
5035          * reserved then we need to leave the reserved extents count alone.
5036          */
5037         if (BTRFS_I(inode)->outstanding_extents >=
5038             BTRFS_I(inode)->reserved_extents)
5039                 return drop_inode_space;
5040 
5041         dropped_extents = BTRFS_I(inode)->reserved_extents -
5042                 BTRFS_I(inode)->outstanding_extents;
5043         BTRFS_I(inode)->reserved_extents -= dropped_extents;
5044         return dropped_extents + drop_inode_space;
5045 }
5046 
5047 /**
5048  * calc_csum_metadata_size - return the amount of metada space that must be
5049  *      reserved/free'd for the given bytes.
5050  * @inode: the inode we're manipulating
5051  * @num_bytes: the number of bytes in question
5052  * @reserve: 1 if we are reserving space, 0 if we are freeing space
5053  *
5054  * This adjusts the number of csum_bytes in the inode and then returns the
5055  * correct amount of metadata that must either be reserved or freed.  We
5056  * calculate how many checksums we can fit into one leaf and then divide the
5057  * number of bytes that will need to be checksumed by this value to figure out
5058  * how many checksums will be required.  If we are adding bytes then the number
5059  * may go up and we will return the number of additional bytes that must be
5060  * reserved.  If it is going down we will return the number of bytes that must
5061  * be freed.
5062  *
5063  * This must be called with BTRFS_I(inode)->lock held.
5064  */
5065 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
5066                                    int reserve)
5067 {
5068         struct btrfs_root *root = BTRFS_I(inode)->root;
5069         u64 csum_size;
5070         int num_csums_per_leaf;
5071         int num_csums;
5072         int old_csums;
5073 
5074         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
5075             BTRFS_I(inode)->csum_bytes == 0)
5076                 return 0;
5077 
5078         old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
5079         if (reserve)
5080                 BTRFS_I(inode)->csum_bytes += num_bytes;
5081         else
5082                 BTRFS_I(inode)->csum_bytes -= num_bytes;
5083         csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
5084         num_csums_per_leaf = (int)div64_u64(csum_size,
5085                                             sizeof(struct btrfs_csum_item) +
5086                                             sizeof(struct btrfs_disk_key));
5087         num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
5088         num_csums = num_csums + num_csums_per_leaf - 1;
5089         num_csums = num_csums / num_csums_per_leaf;
5090 
5091         old_csums = old_csums + num_csums_per_leaf - 1;
5092         old_csums = old_csums / num_csums_per_leaf;
5093 
5094         /* No change, no need to reserve more */
5095         if (old_csums == num_csums)
5096                 return 0;
5097 
5098         if (reserve)
5099                 return btrfs_calc_trans_metadata_size(root,
5100                                                       num_csums - old_csums);
5101 
5102         return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
5103 }
5104 
5105 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
5106 {
5107         struct btrfs_root *root = BTRFS_I(inode)->root;
5108         struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
5109         u64 to_reserve = 0;
5110         u64 csum_bytes;
5111         unsigned nr_extents = 0;
5112         int extra_reserve = 0;
5113         enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
5114         int ret = 0;
5115         bool delalloc_lock = true;
5116         u64 to_free = 0;
5117         unsigned dropped;
5118 
5119         /* If we are a free space inode we need to not flush since we will be in
5120          * the middle of a transaction commit.  We also don't need the delalloc
5121          * mutex since we won't race with anybody.  We need this mostly to make
5122          * lockdep shut its filthy mouth.
5123          */
5124         if (btrfs_is_free_space_inode(inode)) {
5125                 flush = BTRFS_RESERVE_NO_FLUSH;
5126                 delalloc_lock = false;
5127         }
5128 
5129         if (flush != BTRFS_RESERVE_NO_FLUSH &&
5130             btrfs_transaction_in_commit(root->fs_info))
5131                 schedule_timeout(1);
5132 
5133         if (delalloc_lock)
5134                 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
5135 
5136         num_bytes = ALIGN(num_bytes, root->sectorsize);
5137 
5138         spin_lock(&BTRFS_I(inode)->lock);
5139         nr_extents = (unsigned)div64_u64(num_bytes +
5140                                          BTRFS_MAX_EXTENT_SIZE - 1,
5141                                          BTRFS_MAX_EXTENT_SIZE);
5142         BTRFS_I(inode)->outstanding_extents += nr_extents;
5143         nr_extents = 0;
5144 
5145         if (BTRFS_I(inode)->outstanding_extents >
5146             BTRFS_I(inode)->reserved_extents)
5147                 nr_extents = BTRFS_I(inode)->outstanding_extents -
5148                         BTRFS_I(inode)->reserved_extents;
5149 
5150         /*
5151          * Add an item to reserve for updating the inode when we complete the
5152          * delalloc io.
5153          */
5154         if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5155                       &BTRFS_I(inode)->runtime_flags)) {
5156                 nr_extents++;
5157                 extra_reserve = 1;
5158         }
5159 
5160         to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
5161         to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
5162         csum_bytes = BTRFS_I(inode)->csum_bytes;
5163         spin_unlock(&BTRFS_I(inode)->lock);
5164 
5165         if (root->fs_info->quota_enabled) {
5166                 ret = btrfs_qgroup_reserve(root, num_bytes +
5167                                            nr_extents * root->nodesize);
5168                 if (ret)
5169                         goto out_fail;
5170         }
5171 
5172         ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
5173         if (unlikely(ret)) {
5174                 if (root->fs_info->quota_enabled)
5175                         btrfs_qgroup_free(root, num_bytes +
5176                                                 nr_extents * root->nodesize);
5177                 goto out_fail;
5178         }
5179 
5180         spin_lock(&BTRFS_I(inode)->lock);
5181         if (extra_reserve) {
5182                 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5183                         &BTRFS_I(inode)->runtime_flags);
5184                 nr_extents--;
5185         }
5186         BTRFS_I(inode)->reserved_extents += nr_extents;
5187         spin_unlock(&BTRFS_I(inode)->lock);
5188 
5189         if (delalloc_lock)
5190                 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5191 
5192         if (to_reserve)
5193                 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5194                                               btrfs_ino(inode), to_reserve, 1);
5195         block_rsv_add_bytes(block_rsv, to_reserve, 1);
5196 
5197         return 0;
5198 
5199 out_fail:
5200         spin_lock(&BTRFS_I(inode)->lock);
5201         dropped = drop_outstanding_extent(inode, num_bytes);
5202         /*
5203          * If the inodes csum_bytes is the same as the original
5204          * csum_bytes then we know we haven't raced with any free()ers
5205          * so we can just reduce our inodes csum bytes and carry on.
5206          */
5207         if (BTRFS_I(inode)->csum_bytes == csum_bytes) {
5208                 calc_csum_metadata_size(inode, num_bytes, 0);
5209         } else {
5210                 u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes;
5211                 u64 bytes;
5212 
5213                 /*
5214                  * This is tricky, but first we need to figure out how much we
5215                  * free'd from any free-ers that occured during this
5216                  * reservation, so we reset ->csum_bytes to the csum_bytes
5217                  * before we dropped our lock, and then call the free for the
5218                  * number of bytes that were freed while we were trying our
5219                  * reservation.
5220                  */
5221                 bytes = csum_bytes - BTRFS_I(inode)->csum_bytes;
5222                 BTRFS_I(inode)->csum_bytes = csum_bytes;
5223                 to_free = calc_csum_metadata_size(inode, bytes, 0);
5224 
5225 
5226                 /*
5227                  * Now we need to see how much we would have freed had we not
5228                  * been making this reservation and our ->csum_bytes were not
5229                  * artificially inflated.
5230                  */
5231                 BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes;
5232                 bytes = csum_bytes - orig_csum_bytes;
5233                 bytes = calc_csum_metadata_size(inode, bytes, 0);
5234 
5235                 /*
5236                  * Now reset ->csum_bytes to what it should be.  If bytes is
5237                  * more than to_free then we would have free'd more space had we
5238                  * not had an artificially high ->csum_bytes, so we need to free
5239                  * the remainder.  If bytes is the same or less then we don't
5240                  * need to do anything, the other free-ers did the correct
5241                  * thing.
5242                  */
5243                 BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes;
5244                 if (bytes > to_free)
5245                         to_free = bytes - to_free;
5246                 else
5247                         to_free = 0;
5248         }
5249         spin_unlock(&BTRFS_I(inode)->lock);
5250         if (dropped)
5251                 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5252 
5253         if (to_free) {
5254                 btrfs_block_rsv_release(root, block_rsv, to_free);
5255                 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5256                                               btrfs_ino(inode), to_free, 0);
5257         }
5258         if (delalloc_lock)
5259                 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5260         return ret;
5261 }
5262 
5263 /**
5264  * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5265  * @inode: the inode to release the reservation for
5266  * @num_bytes: the number of bytes we're releasing
5267  *
5268  * This will release the metadata reservation for an inode.  This can be called
5269  * once we complete IO for a given set of bytes to release their metadata
5270  * reservations.
5271  */
5272 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
5273 {
5274         struct btrfs_root *root = BTRFS_I(inode)->root;
5275         u64 to_free = 0;
5276         unsigned dropped;
5277 
5278         num_bytes = ALIGN(num_bytes, root->sectorsize);
5279         spin_lock(&BTRFS_I(inode)->lock);
5280         dropped = drop_outstanding_extent(inode, num_bytes);
5281 
5282         if (num_bytes)
5283                 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
5284         spin_unlock(&BTRFS_I(inode)->lock);
5285         if (dropped > 0)
5286                 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5287 
5288         if (btrfs_test_is_dummy_root(root))
5289                 return;
5290 
5291         trace_btrfs_space_reservation(root->fs_info, "delalloc",
5292                                       btrfs_ino(inode), to_free, 0);
5293         if (root->fs_info->quota_enabled) {
5294                 btrfs_qgroup_free(root, num_bytes +
5295                                         dropped * root->nodesize);
5296         }
5297 
5298         btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
5299                                 to_free);
5300 }
5301 
5302 /**
5303  * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
5304  * @inode: inode we're writing to
5305  * @num_bytes: the number of bytes we want to allocate
5306  *
5307  * This will do the following things
5308  *
5309  * o reserve space in the data space info for num_bytes
5310  * o reserve space in the metadata space info based on number of outstanding
5311  *   extents and how much csums will be needed
5312  * o add to the inodes ->delalloc_bytes
5313  * o add it to the fs_info's delalloc inodes list.
5314  *
5315  * This will return 0 for success and -ENOSPC if there is no space left.
5316  */
5317 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
5318 {
5319         int ret;
5320 
5321         ret = btrfs_check_data_free_space(inode, num_bytes);
5322         if (ret)
5323                 return ret;
5324 
5325         ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
5326         if (ret) {
5327                 btrfs_free_reserved_data_space(inode, num_bytes);
5328                 return ret;
5329         }
5330 
5331         return 0;
5332 }
5333 
5334 /**
5335  * btrfs_delalloc_release_space - release data and metadata space for delalloc
5336  * @inode: inode we're releasing space for
5337  * @num_bytes: the number of bytes we want to free up
5338  *
5339  * This must be matched with a call to btrfs_delalloc_reserve_space.  This is
5340  * called in the case that we don't need the metadata AND data reservations
5341  * anymore.  So if there is an error or we insert an inline extent.
5342  *
5343  * This function will release the metadata space that was not used and will
5344  * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5345  * list if there are no delalloc bytes left.
5346  */
5347 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
5348 {
5349         btrfs_delalloc_release_metadata(inode, num_bytes);
5350         btrfs_free_reserved_data_space(inode, num_bytes);
5351 }
5352 
5353 static int update_block_group(struct btrfs_trans_handle *trans,
5354                               struct btrfs_root *root, u64 bytenr,
5355                               u64 num_bytes, int alloc)
5356 {
5357         struct btrfs_block_group_cache *cache = NULL;
5358         struct btrfs_fs_info *info = root->fs_info;
5359         u64 total = num_bytes;
5360         u64 old_val;
5361         u64 byte_in_group;
5362         int factor;
5363 
5364         /* block accounting for super block */
5365         spin_lock(&info->delalloc_root_lock);
5366         old_val = btrfs_super_bytes_used(info->super_copy);
5367         if (alloc)
5368                 old_val += num_bytes;
5369         else
5370                 old_val -= num_bytes;
5371         btrfs_set_super_bytes_used(info->super_copy, old_val);
5372         spin_unlock(&info->delalloc_root_lock);
5373 
5374         while (total) {
5375                 cache = btrfs_lookup_block_group(info, bytenr);
5376                 if (!cache)
5377                         return -ENOENT;
5378                 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
5379                                     BTRFS_BLOCK_GROUP_RAID1 |
5380                                     BTRFS_BLOCK_GROUP_RAID10))
5381                         factor = 2;
5382                 else
5383                         factor = 1;
5384                 /*
5385                  * If this block group has free space cache written out, we
5386                  * need to make sure to load it if we are removing space.  This
5387                  * is because we need the unpinning stage to actually add the
5388                  * space back to the block group, otherwise we will leak space.
5389                  */
5390                 if (!alloc && cache->cached == BTRFS_CACHE_NO)
5391                         cache_block_group(cache, 1);
5392 
5393                 spin_lock(&trans->transaction->dirty_bgs_lock);
5394                 if (list_empty(&cache->dirty_list)) {
5395                         list_add_tail(&cache->dirty_list,
5396                                       &trans->transaction->dirty_bgs);
5397                         btrfs_get_block_group(cache);
5398                 }
5399                 spin_unlock(&trans->transaction->dirty_bgs_lock);
5400 
5401                 byte_in_group = bytenr - cache->key.objectid;
5402                 WARN_ON(byte_in_group > cache->key.offset);
5403 
5404                 spin_lock(&cache->space_info->lock);
5405                 spin_lock(&cache->lock);
5406 
5407                 if (btrfs_test_opt(root, SPACE_CACHE) &&
5408                     cache->disk_cache_state < BTRFS_DC_CLEAR)
5409                         cache->disk_cache_state = BTRFS_DC_CLEAR;
5410 
5411                 old_val = btrfs_block_group_used(&cache->item);
5412                 num_bytes = min(total, cache->key.offset - byte_in_group);
5413                 if (alloc) {
5414                         old_val += num_bytes;
5415                         btrfs_set_block_group_used(&cache->item, old_val);
5416                         cache->reserved -= num_bytes;
5417                         cache->space_info->bytes_reserved -= num_bytes;
5418                         cache->space_info->bytes_used += num_bytes;
5419                         cache->space_info->disk_used += num_bytes * factor;
5420                         spin_unlock(&cache->lock);
5421                         spin_unlock(&cache->space_info->lock);
5422                 } else {
5423                         old_val -= num_bytes;
5424                         btrfs_set_block_group_used(&cache->item, old_val);
5425                         cache->pinned += num_bytes;
5426                         cache->space_info->bytes_pinned += num_bytes;
5427                         cache->space_info->bytes_used -= num_bytes;
5428                         cache->space_info->disk_used -= num_bytes * factor;
5429                         spin_unlock(&cache->lock);
5430                         spin_unlock(&cache->space_info->lock);
5431 
5432                         set_extent_dirty(info->pinned_extents,
5433                                          bytenr, bytenr + num_bytes - 1,
5434                                          GFP_NOFS | __GFP_NOFAIL);
5435                         /*
5436                          * No longer have used bytes in this block group, queue
5437                          * it for deletion.
5438                          */
5439                         if (old_val == 0) {
5440                                 spin_lock(&info->unused_bgs_lock);
5441                                 if (list_empty(&cache->bg_list)) {
5442                                         btrfs_get_block_group(cache);
5443                                         list_add_tail(&cache->bg_list,
5444                                                       &info->unused_bgs);
5445                                 }
5446                                 spin_unlock(&info->unused_bgs_lock);
5447                         }
5448                 }
5449                 btrfs_put_block_group(cache);
5450                 total -= num_bytes;
5451                 bytenr += num_bytes;
5452         }
5453         return 0;
5454 }
5455 
5456 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
5457 {
5458         struct btrfs_block_group_cache *cache;
5459         u64 bytenr;
5460 
5461         spin_lock(&root->fs_info->block_group_cache_lock);
5462         bytenr = root->fs_info->first_logical_byte;
5463         spin_unlock(&root->fs_info->block_group_cache_lock);
5464 
5465         if (bytenr < (u64)-1)
5466                 return bytenr;
5467 
5468         cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
5469         if (!cache)
5470                 return 0;
5471 
5472         bytenr = cache->key.objectid;
5473         btrfs_put_block_group(cache);
5474 
5475         return bytenr;
5476 }
5477 
5478 static int pin_down_extent(struct btrfs_root *root,
5479                            struct btrfs_block_group_cache *cache,
5480                            u64 bytenr, u64 num_bytes, int reserved)
5481 {
5482         spin_lock(&cache->space_info->lock);
5483         spin_lock(&cache->lock);
5484         cache->pinned += num_bytes;
5485         cache->space_info->bytes_pinned += num_bytes;
5486         if (reserved) {
5487                 cache->reserved -= num_bytes;
5488                 cache->space_info->bytes_reserved -= num_bytes;
5489         }
5490         spin_unlock(&cache->lock);
5491         spin_unlock(&cache->space_info->lock);
5492 
5493         set_extent_dirty(root->fs_info->pinned_extents, bytenr,
5494                          bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
5495         if (reserved)
5496                 trace_btrfs_reserved_extent_free(root, bytenr, num_bytes);
5497         return 0;
5498 }
5499 
5500 /*
5501  * this function must be called within transaction
5502  */
5503 int btrfs_pin_extent(struct btrfs_root *root,
5504                      u64 bytenr, u64 num_bytes, int reserved)
5505 {
5506         struct btrfs_block_group_cache *cache;
5507 
5508         cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5509         BUG_ON(!cache); /* Logic error */
5510 
5511         pin_down_extent(root, cache, bytenr, num_bytes, reserved);
5512 
5513         btrfs_put_block_group(cache);
5514         return 0;
5515 }
5516 
5517 /*
5518  * this function must be called within transaction
5519  */
5520 int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
5521                                     u64 bytenr, u64 num_bytes)
5522 {
5523         struct btrfs_block_group_cache *cache;
5524         int ret;
5525 
5526         cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5527         if (!cache)
5528                 return -EINVAL;
5529 
5530         /*
5531          * pull in the free space cache (if any) so that our pin
5532          * removes the free space from the cache.  We have load_only set
5533          * to one because the slow code to read in the free extents does check
5534          * the pinned extents.
5535          */
5536         cache_block_group(cache, 1);
5537 
5538         pin_down_extent(root, cache, bytenr, num_bytes, 0);
5539 
5540         /* remove us from the free space cache (if we're there at all) */
5541         ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
5542         btrfs_put_block_group(cache);
5543         return ret;
5544 }
5545 
5546 static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes)
5547 {
5548         int ret;
5549         struct btrfs_block_group_cache *block_group;
5550         struct btrfs_caching_control *caching_ctl;
5551 
5552         block_group = btrfs_lookup_block_group(root->fs_info, start);
5553         if (!block_group)
5554                 return -EINVAL;
5555 
5556         cache_block_group(block_group, 0);
5557         caching_ctl = get_caching_control(block_group);
5558 
5559         if (!caching_ctl) {
5560                 /* Logic error */
5561                 BUG_ON(!block_group_cache_done(block_group));
5562                 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5563         } else {
5564                 mutex_lock(&caching_ctl->mutex);
5565 
5566                 if (start >= caching_ctl->progress) {
5567                         ret = add_excluded_extent(root, start, num_bytes);
5568                 } else if (start + num_bytes <= caching_ctl->progress) {
5569                         ret = btrfs_remove_free_space(block_group,
5570                                                       start, num_bytes);
5571                 } else {
5572                         num_bytes = caching_ctl->progress - start;
5573                         ret = btrfs_remove_free_space(block_group,
5574                                                       start, num_bytes);
5575                         if (ret)
5576                                 goto out_lock;
5577 
5578                         num_bytes = (start + num_bytes) -
5579                                 caching_ctl->progress;
5580                         start = caching_ctl->progress;
5581                         ret = add_excluded_extent(root, start, num_bytes);
5582                 }
5583 out_lock:
5584                 mutex_unlock(&caching_ctl->mutex);
5585                 put_caching_control(caching_ctl);
5586         }
5587         btrfs_put_block_group(block_group);
5588         return ret;
5589 }
5590 
5591 int btrfs_exclude_logged_extents(struct btrfs_root *log,
5592                                  struct extent_buffer *eb)
5593 {
5594         struct btrfs_file_extent_item *item;
5595         struct btrfs_key key;
5596         int found_type;
5597         int i;
5598 
5599         if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS))
5600                 return 0;
5601 
5602         for (i = 0; i < btrfs_header_nritems(eb); i++) {
5603                 btrfs_item_key_to_cpu(eb, &key, i);
5604                 if (key.type != BTRFS_EXTENT_DATA_KEY)
5605                         continue;
5606                 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
5607                 found_type = btrfs_file_extent_type(eb, item);
5608                 if (found_type == BTRFS_FILE_EXTENT_INLINE)
5609                         continue;
5610                 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
5611                         continue;
5612                 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
5613                 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
5614                 __exclude_logged_extent(log, key.objectid, key.offset);
5615         }
5616 
5617         return 0;
5618 }
5619 
5620 /**
5621  * btrfs_update_reserved_bytes - update the block_group and space info counters
5622  * @cache:      The cache we are manipulating
5623  * @num_bytes:  The number of bytes in question
5624  * @reserve:    One of the reservation enums
5625  * @delalloc:   The blocks are allocated for the delalloc write
5626  *
5627  * This is called by the allocator when it reserves space, or by somebody who is
5628  * freeing space that was never actually used on disk.  For example if you
5629  * reserve some space for a new leaf in transaction A and before transaction A
5630  * commits you free that leaf, you call this with reserve set to 0 in order to
5631  * clear the reservation.
5632  *
5633  * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
5634  * ENOSPC accounting.  For data we handle the reservation through clearing the
5635  * delalloc bits in the io_tree.  We have to do this since we could end up
5636  * allocating less disk space for the amount of data we have reserved in the
5637  * case of compression.
5638  *
5639  * If this is a reservation and the block group has become read only we cannot
5640  * make the reservation and return -EAGAIN, otherwise this function always
5641  * succeeds.
5642  */
5643 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
5644                                        u64 num_bytes, int reserve, int delalloc)
5645 {
5646         struct btrfs_space_info *space_info = cache->space_info;
5647         int ret = 0;
5648 
5649         spin_lock(&space_info->lock);
5650         spin_lock(&cache->lock);
5651         if (reserve != RESERVE_FREE) {
5652                 if (cache->ro) {
5653                         ret = -EAGAIN;
5654                 } else {
5655                         cache->reserved += num_bytes;
5656                         space_info->bytes_reserved += num_bytes;
5657                         if (reserve == RESERVE_ALLOC) {
5658                                 trace_btrfs_space_reservation(cache->fs_info,
5659                                                 "space_info", space_info->flags,
5660                                                 num_bytes, 0);
5661                                 space_info->bytes_may_use -= num_bytes;
5662                         }
5663 
5664                         if (delalloc)
5665                                 cache->delalloc_bytes += num_bytes;
5666                 }
5667         } else {
5668                 if (cache->ro)
5669                         space_info->bytes_readonly += num_bytes;
5670                 cache->reserved -= num_bytes;
5671                 space_info->bytes_reserved -= num_bytes;
5672 
5673                 if (delalloc)
5674                         cache->delalloc_bytes -= num_bytes;
5675         }
5676         spin_unlock(&cache->lock);
5677         spin_unlock(&space_info->lock);
5678         return ret;
5679 }
5680 
5681 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
5682                                 struct btrfs_root *root)
5683 {
5684         struct btrfs_fs_info *fs_info = root->fs_info;
5685         struct btrfs_caching_control *next;
5686         struct btrfs_caching_control *caching_ctl;
5687         struct btrfs_block_group_cache *cache;
5688 
5689         down_write(&fs_info->commit_root_sem);
5690 
5691         list_for_each_entry_safe(caching_ctl, next,
5692                                  &fs_info->caching_block_groups, list) {