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

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

Version: ~ [ linux-5.15-rc5 ] ~ [ linux-5.14.11 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.72 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.152 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.210 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.250 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.286 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.288 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.18.140 ] ~ [ linux-3.16.85 ] ~ [ linux-3.14.79 ] ~ [ linux-3.12.74 ] ~ [ 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_root *root,
 78                               u64 bytenr, u64 num_bytes, int alloc);
 79 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
 80                                 struct btrfs_root *root,
 81                                 u64 bytenr, u64 num_bytes, u64 parent,
 82                                 u64 root_objectid, u64 owner_objectid,
 83                                 u64 owner_offset, int refs_to_drop,
 84                                 struct btrfs_delayed_extent_op *extra_op,
 85                                 int no_quota);
 86 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
 87                                     struct extent_buffer *leaf,
 88                                     struct btrfs_extent_item *ei);
 89 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
 90                                       struct btrfs_root *root,
 91                                       u64 parent, u64 root_objectid,
 92                                       u64 flags, u64 owner, u64 offset,
 93                                       struct btrfs_key *ins, int ref_mod);
 94 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
 95                                      struct btrfs_root *root,
 96                                      u64 parent, u64 root_objectid,
 97                                      u64 flags, struct btrfs_disk_key *key,
 98                                      int level, struct btrfs_key *ins,
 99                                      int no_quota);
100 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
101                           struct btrfs_root *extent_root, u64 flags,
102                           int force);
103 static int find_next_key(struct btrfs_path *path, int level,
104                          struct btrfs_key *key);
105 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
106                             int dump_block_groups);
107 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
108                                        u64 num_bytes, int reserve,
109                                        int delalloc);
110 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
111                                u64 num_bytes);
112 int btrfs_pin_extent(struct btrfs_root *root,
113                      u64 bytenr, u64 num_bytes, int reserved);
114 
115 static noinline int
116 block_group_cache_done(struct btrfs_block_group_cache *cache)
117 {
118         smp_mb();
119         return cache->cached == BTRFS_CACHE_FINISHED ||
120                 cache->cached == BTRFS_CACHE_ERROR;
121 }
122 
123 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
124 {
125         return (cache->flags & bits) == bits;
126 }
127 
128 static void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
129 {
130         atomic_inc(&cache->count);
131 }
132 
133 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
134 {
135         if (atomic_dec_and_test(&cache->count)) {
136                 WARN_ON(cache->pinned > 0);
137                 WARN_ON(cache->reserved > 0);
138                 kfree(cache->free_space_ctl);
139                 kfree(cache);
140         }
141 }
142 
143 /*
144  * this adds the block group to the fs_info rb tree for the block group
145  * cache
146  */
147 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
148                                 struct btrfs_block_group_cache *block_group)
149 {
150         struct rb_node **p;
151         struct rb_node *parent = NULL;
152         struct btrfs_block_group_cache *cache;
153 
154         spin_lock(&info->block_group_cache_lock);
155         p = &info->block_group_cache_tree.rb_node;
156 
157         while (*p) {
158                 parent = *p;
159                 cache = rb_entry(parent, struct btrfs_block_group_cache,
160                                  cache_node);
161                 if (block_group->key.objectid < cache->key.objectid) {
162                         p = &(*p)->rb_left;
163                 } else if (block_group->key.objectid > cache->key.objectid) {
164                         p = &(*p)->rb_right;
165                 } else {
166                         spin_unlock(&info->block_group_cache_lock);
167                         return -EEXIST;
168                 }
169         }
170 
171         rb_link_node(&block_group->cache_node, parent, p);
172         rb_insert_color(&block_group->cache_node,
173                         &info->block_group_cache_tree);
174 
175         if (info->first_logical_byte > block_group->key.objectid)
176                 info->first_logical_byte = block_group->key.objectid;
177 
178         spin_unlock(&info->block_group_cache_lock);
179 
180         return 0;
181 }
182 
183 /*
184  * This will return the block group at or after bytenr if contains is 0, else
185  * it will return the block group that contains the bytenr
186  */
187 static struct btrfs_block_group_cache *
188 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
189                               int contains)
190 {
191         struct btrfs_block_group_cache *cache, *ret = NULL;
192         struct rb_node *n;
193         u64 end, start;
194 
195         spin_lock(&info->block_group_cache_lock);
196         n = info->block_group_cache_tree.rb_node;
197 
198         while (n) {
199                 cache = rb_entry(n, struct btrfs_block_group_cache,
200                                  cache_node);
201                 end = cache->key.objectid + cache->key.offset - 1;
202                 start = cache->key.objectid;
203 
204                 if (bytenr < start) {
205                         if (!contains && (!ret || start < ret->key.objectid))
206                                 ret = cache;
207                         n = n->rb_left;
208                 } else if (bytenr > start) {
209                         if (contains && bytenr <= end) {
210                                 ret = cache;
211                                 break;
212                         }
213                         n = n->rb_right;
214                 } else {
215                         ret = cache;
216                         break;
217                 }
218         }
219         if (ret) {
220                 btrfs_get_block_group(ret);
221                 if (bytenr == 0 && info->first_logical_byte > ret->key.objectid)
222                         info->first_logical_byte = ret->key.objectid;
223         }
224         spin_unlock(&info->block_group_cache_lock);
225 
226         return ret;
227 }
228 
229 static int add_excluded_extent(struct btrfs_root *root,
230                                u64 start, u64 num_bytes)
231 {
232         u64 end = start + num_bytes - 1;
233         set_extent_bits(&root->fs_info->freed_extents[0],
234                         start, end, EXTENT_UPTODATE, GFP_NOFS);
235         set_extent_bits(&root->fs_info->freed_extents[1],
236                         start, end, EXTENT_UPTODATE, GFP_NOFS);
237         return 0;
238 }
239 
240 static void free_excluded_extents(struct btrfs_root *root,
241                                   struct btrfs_block_group_cache *cache)
242 {
243         u64 start, end;
244 
245         start = cache->key.objectid;
246         end = start + cache->key.offset - 1;
247 
248         clear_extent_bits(&root->fs_info->freed_extents[0],
249                           start, end, EXTENT_UPTODATE, GFP_NOFS);
250         clear_extent_bits(&root->fs_info->freed_extents[1],
251                           start, end, EXTENT_UPTODATE, GFP_NOFS);
252 }
253 
254 static int exclude_super_stripes(struct btrfs_root *root,
255                                  struct btrfs_block_group_cache *cache)
256 {
257         u64 bytenr;
258         u64 *logical;
259         int stripe_len;
260         int i, nr, ret;
261 
262         if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
263                 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
264                 cache->bytes_super += stripe_len;
265                 ret = add_excluded_extent(root, cache->key.objectid,
266                                           stripe_len);
267                 if (ret)
268                         return ret;
269         }
270 
271         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
272                 bytenr = btrfs_sb_offset(i);
273                 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
274                                        cache->key.objectid, bytenr,
275                                        0, &logical, &nr, &stripe_len);
276                 if (ret)
277                         return ret;
278 
279                 while (nr--) {
280                         u64 start, len;
281 
282                         if (logical[nr] > cache->key.objectid +
283                             cache->key.offset)
284                                 continue;
285 
286                         if (logical[nr] + stripe_len <= cache->key.objectid)
287                                 continue;
288 
289                         start = logical[nr];
290                         if (start < cache->key.objectid) {
291                                 start = cache->key.objectid;
292                                 len = (logical[nr] + stripe_len) - start;
293                         } else {
294                                 len = min_t(u64, stripe_len,
295                                             cache->key.objectid +
296                                             cache->key.offset - start);
297                         }
298 
299                         cache->bytes_super += len;
300                         ret = add_excluded_extent(root, start, len);
301                         if (ret) {
302                                 kfree(logical);
303                                 return ret;
304                         }
305                 }
306 
307                 kfree(logical);
308         }
309         return 0;
310 }
311 
312 static struct btrfs_caching_control *
313 get_caching_control(struct btrfs_block_group_cache *cache)
314 {
315         struct btrfs_caching_control *ctl;
316 
317         spin_lock(&cache->lock);
318         if (cache->cached != BTRFS_CACHE_STARTED) {
319                 spin_unlock(&cache->lock);
320                 return NULL;
321         }
322 
323         /* We're loading it the fast way, so we don't have a caching_ctl. */
324         if (!cache->caching_ctl) {
325                 spin_unlock(&cache->lock);
326                 return NULL;
327         }
328 
329         ctl = cache->caching_ctl;
330         atomic_inc(&ctl->count);
331         spin_unlock(&cache->lock);
332         return ctl;
333 }
334 
335 static void put_caching_control(struct btrfs_caching_control *ctl)
336 {
337         if (atomic_dec_and_test(&ctl->count))
338                 kfree(ctl);
339 }
340 
341 /*
342  * this is only called by cache_block_group, since we could have freed extents
343  * we need to check the pinned_extents for any extents that can't be used yet
344  * since their free space will be released as soon as the transaction commits.
345  */
346 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
347                               struct btrfs_fs_info *info, u64 start, u64 end)
348 {
349         u64 extent_start, extent_end, size, total_added = 0;
350         int ret;
351 
352         while (start < end) {
353                 ret = find_first_extent_bit(info->pinned_extents, start,
354                                             &extent_start, &extent_end,
355                                             EXTENT_DIRTY | EXTENT_UPTODATE,
356                                             NULL);
357                 if (ret)
358                         break;
359 
360                 if (extent_start <= start) {
361                         start = extent_end + 1;
362                 } else if (extent_start > start && extent_start < end) {
363                         size = extent_start - start;
364                         total_added += size;
365                         ret = btrfs_add_free_space(block_group, start,
366                                                    size);
367                         BUG_ON(ret); /* -ENOMEM or logic error */
368                         start = extent_end + 1;
369                 } else {
370                         break;
371                 }
372         }
373 
374         if (start < end) {
375                 size = end - start;
376                 total_added += size;
377                 ret = btrfs_add_free_space(block_group, start, size);
378                 BUG_ON(ret); /* -ENOMEM or logic error */
379         }
380 
381         return total_added;
382 }
383 
384 static noinline void caching_thread(struct btrfs_work *work)
385 {
386         struct btrfs_block_group_cache *block_group;
387         struct btrfs_fs_info *fs_info;
388         struct btrfs_caching_control *caching_ctl;
389         struct btrfs_root *extent_root;
390         struct btrfs_path *path;
391         struct extent_buffer *leaf;
392         struct btrfs_key key;
393         u64 total_found = 0;
394         u64 last = 0;
395         u32 nritems;
396         int ret = -ENOMEM;
397 
398         caching_ctl = container_of(work, struct btrfs_caching_control, work);
399         block_group = caching_ctl->block_group;
400         fs_info = block_group->fs_info;
401         extent_root = fs_info->extent_root;
402 
403         path = btrfs_alloc_path();
404         if (!path)
405                 goto out;
406 
407         last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
408 
409         /*
410          * We don't want to deadlock with somebody trying to allocate a new
411          * extent for the extent root while also trying to search the extent
412          * root to add free space.  So we skip locking and search the commit
413          * root, since its read-only
414          */
415         path->skip_locking = 1;
416         path->search_commit_root = 1;
417         path->reada = 1;
418 
419         key.objectid = last;
420         key.offset = 0;
421         key.type = BTRFS_EXTENT_ITEM_KEY;
422 again:
423         mutex_lock(&caching_ctl->mutex);
424         /* need to make sure the commit_root doesn't disappear */
425         down_read(&fs_info->commit_root_sem);
426 
427 next:
428         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
429         if (ret < 0)
430                 goto err;
431 
432         leaf = path->nodes[0];
433         nritems = btrfs_header_nritems(leaf);
434 
435         while (1) {
436                 if (btrfs_fs_closing(fs_info) > 1) {
437                         last = (u64)-1;
438                         break;
439                 }
440 
441                 if (path->slots[0] < nritems) {
442                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
443                 } else {
444                         ret = find_next_key(path, 0, &key);
445                         if (ret)
446                                 break;
447 
448                         if (need_resched() ||
449                             rwsem_is_contended(&fs_info->commit_root_sem)) {
450                                 caching_ctl->progress = last;
451                                 btrfs_release_path(path);
452                                 up_read(&fs_info->commit_root_sem);
453                                 mutex_unlock(&caching_ctl->mutex);
454                                 cond_resched();
455                                 goto again;
456                         }
457 
458                         ret = btrfs_next_leaf(extent_root, path);
459                         if (ret < 0)
460                                 goto err;
461                         if (ret)
462                                 break;
463                         leaf = path->nodes[0];
464                         nritems = btrfs_header_nritems(leaf);
465                         continue;
466                 }
467 
468                 if (key.objectid < last) {
469                         key.objectid = last;
470                         key.offset = 0;
471                         key.type = BTRFS_EXTENT_ITEM_KEY;
472 
473                         caching_ctl->progress = last;
474                         btrfs_release_path(path);
475                         goto next;
476                 }
477 
478                 if (key.objectid < block_group->key.objectid) {
479                         path->slots[0]++;
480                         continue;
481                 }
482 
483                 if (key.objectid >= block_group->key.objectid +
484                     block_group->key.offset)
485                         break;
486 
487                 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
488                     key.type == BTRFS_METADATA_ITEM_KEY) {
489                         total_found += add_new_free_space(block_group,
490                                                           fs_info, last,
491                                                           key.objectid);
492                         if (key.type == BTRFS_METADATA_ITEM_KEY)
493                                 last = key.objectid +
494                                         fs_info->tree_root->nodesize;
495                         else
496                                 last = key.objectid + key.offset;
497 
498                         if (total_found > (1024 * 1024 * 2)) {
499                                 total_found = 0;
500                                 wake_up(&caching_ctl->wait);
501                         }
502                 }
503                 path->slots[0]++;
504         }
505         ret = 0;
506 
507         total_found += add_new_free_space(block_group, fs_info, last,
508                                           block_group->key.objectid +
509                                           block_group->key.offset);
510         caching_ctl->progress = (u64)-1;
511 
512         spin_lock(&block_group->lock);
513         block_group->caching_ctl = NULL;
514         block_group->cached = BTRFS_CACHE_FINISHED;
515         spin_unlock(&block_group->lock);
516 
517 err:
518         btrfs_free_path(path);
519         up_read(&fs_info->commit_root_sem);
520 
521         free_excluded_extents(extent_root, block_group);
522 
523         mutex_unlock(&caching_ctl->mutex);
524 out:
525         if (ret) {
526                 spin_lock(&block_group->lock);
527                 block_group->caching_ctl = NULL;
528                 block_group->cached = BTRFS_CACHE_ERROR;
529                 spin_unlock(&block_group->lock);
530         }
531         wake_up(&caching_ctl->wait);
532 
533         put_caching_control(caching_ctl);
534         btrfs_put_block_group(block_group);
535 }
536 
537 static int cache_block_group(struct btrfs_block_group_cache *cache,
538                              int load_cache_only)
539 {
540         DEFINE_WAIT(wait);
541         struct btrfs_fs_info *fs_info = cache->fs_info;
542         struct btrfs_caching_control *caching_ctl;
543         int ret = 0;
544 
545         caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
546         if (!caching_ctl)
547                 return -ENOMEM;
548 
549         INIT_LIST_HEAD(&caching_ctl->list);
550         mutex_init(&caching_ctl->mutex);
551         init_waitqueue_head(&caching_ctl->wait);
552         caching_ctl->block_group = cache;
553         caching_ctl->progress = cache->key.objectid;
554         atomic_set(&caching_ctl->count, 1);
555         btrfs_init_work(&caching_ctl->work, btrfs_cache_helper,
556                         caching_thread, NULL, NULL);
557 
558         spin_lock(&cache->lock);
559         /*
560          * This should be a rare occasion, but this could happen I think in the
561          * case where one thread starts to load the space cache info, and then
562          * some other thread starts a transaction commit which tries to do an
563          * allocation while the other thread is still loading the space cache
564          * info.  The previous loop should have kept us from choosing this block
565          * group, but if we've moved to the state where we will wait on caching
566          * block groups we need to first check if we're doing a fast load here,
567          * so we can wait for it to finish, otherwise we could end up allocating
568          * from a block group who's cache gets evicted for one reason or
569          * another.
570          */
571         while (cache->cached == BTRFS_CACHE_FAST) {
572                 struct btrfs_caching_control *ctl;
573 
574                 ctl = cache->caching_ctl;
575                 atomic_inc(&ctl->count);
576                 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
577                 spin_unlock(&cache->lock);
578 
579                 schedule();
580 
581                 finish_wait(&ctl->wait, &wait);
582                 put_caching_control(ctl);
583                 spin_lock(&cache->lock);
584         }
585 
586         if (cache->cached != BTRFS_CACHE_NO) {
587                 spin_unlock(&cache->lock);
588                 kfree(caching_ctl);
589                 return 0;
590         }
591         WARN_ON(cache->caching_ctl);
592         cache->caching_ctl = caching_ctl;
593         cache->cached = BTRFS_CACHE_FAST;
594         spin_unlock(&cache->lock);
595 
596         if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) {
597                 ret = load_free_space_cache(fs_info, cache);
598 
599                 spin_lock(&cache->lock);
600                 if (ret == 1) {
601                         cache->caching_ctl = NULL;
602                         cache->cached = BTRFS_CACHE_FINISHED;
603                         cache->last_byte_to_unpin = (u64)-1;
604                 } else {
605                         if (load_cache_only) {
606                                 cache->caching_ctl = NULL;
607                                 cache->cached = BTRFS_CACHE_NO;
608                         } else {
609                                 cache->cached = BTRFS_CACHE_STARTED;
610                         }
611                 }
612                 spin_unlock(&cache->lock);
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                 }
631                 spin_unlock(&cache->lock);
632                 wake_up(&caching_ctl->wait);
633         }
634 
635         if (load_cache_only) {
636                 put_caching_control(caching_ctl);
637                 return 0;
638         }
639 
640         down_write(&fs_info->commit_root_sem);
641         atomic_inc(&caching_ctl->count);
642         list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
643         up_write(&fs_info->commit_root_sem);
644 
645         btrfs_get_block_group(cache);
646 
647         btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work);
648 
649         return ret;
650 }
651 
652 /*
653  * return the block group that starts at or after bytenr
654  */
655 static struct btrfs_block_group_cache *
656 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
657 {
658         struct btrfs_block_group_cache *cache;
659 
660         cache = block_group_cache_tree_search(info, bytenr, 0);
661 
662         return cache;
663 }
664 
665 /*
666  * return the block group that contains the given bytenr
667  */
668 struct btrfs_block_group_cache *btrfs_lookup_block_group(
669                                                  struct btrfs_fs_info *info,
670                                                  u64 bytenr)
671 {
672         struct btrfs_block_group_cache *cache;
673 
674         cache = block_group_cache_tree_search(info, bytenr, 1);
675 
676         return cache;
677 }
678 
679 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
680                                                   u64 flags)
681 {
682         struct list_head *head = &info->space_info;
683         struct btrfs_space_info *found;
684 
685         flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
686 
687         rcu_read_lock();
688         list_for_each_entry_rcu(found, head, list) {
689                 if (found->flags & flags) {
690                         rcu_read_unlock();
691                         return found;
692                 }
693         }
694         rcu_read_unlock();
695         return NULL;
696 }
697 
698 /*
699  * after adding space to the filesystem, we need to clear the full flags
700  * on all the space infos.
701  */
702 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
703 {
704         struct list_head *head = &info->space_info;
705         struct btrfs_space_info *found;
706 
707         rcu_read_lock();
708         list_for_each_entry_rcu(found, head, list)
709                 found->full = 0;
710         rcu_read_unlock();
711 }
712 
713 /* simple helper to search for an existing data extent at a given offset */
714 int btrfs_lookup_data_extent(struct btrfs_root *root, u64 start, u64 len)
715 {
716         int ret;
717         struct btrfs_key key;
718         struct btrfs_path *path;
719 
720         path = btrfs_alloc_path();
721         if (!path)
722                 return -ENOMEM;
723 
724         key.objectid = start;
725         key.offset = len;
726         key.type = BTRFS_EXTENT_ITEM_KEY;
727         ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
728                                 0, 0);
729         btrfs_free_path(path);
730         return ret;
731 }
732 
733 /*
734  * helper function to lookup reference count and flags of a tree block.
735  *
736  * the head node for delayed ref is used to store the sum of all the
737  * reference count modifications queued up in the rbtree. the head
738  * node may also store the extent flags to set. This way you can check
739  * to see what the reference count and extent flags would be if all of
740  * the delayed refs are not processed.
741  */
742 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
743                              struct btrfs_root *root, u64 bytenr,
744                              u64 offset, int metadata, u64 *refs, u64 *flags)
745 {
746         struct btrfs_delayed_ref_head *head;
747         struct btrfs_delayed_ref_root *delayed_refs;
748         struct btrfs_path *path;
749         struct btrfs_extent_item *ei;
750         struct extent_buffer *leaf;
751         struct btrfs_key key;
752         u32 item_size;
753         u64 num_refs;
754         u64 extent_flags;
755         int ret;
756 
757         /*
758          * If we don't have skinny metadata, don't bother doing anything
759          * different
760          */
761         if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) {
762                 offset = root->nodesize;
763                 metadata = 0;
764         }
765 
766         path = btrfs_alloc_path();
767         if (!path)
768                 return -ENOMEM;
769 
770         if (!trans) {
771                 path->skip_locking = 1;
772                 path->search_commit_root = 1;
773         }
774 
775 search_again:
776         key.objectid = bytenr;
777         key.offset = offset;
778         if (metadata)
779                 key.type = BTRFS_METADATA_ITEM_KEY;
780         else
781                 key.type = BTRFS_EXTENT_ITEM_KEY;
782 
783         ret = btrfs_search_slot(trans, root->fs_info->extent_root,
784                                 &key, path, 0, 0);
785         if (ret < 0)
786                 goto out_free;
787 
788         if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
789                 if (path->slots[0]) {
790                         path->slots[0]--;
791                         btrfs_item_key_to_cpu(path->nodes[0], &key,
792                                               path->slots[0]);
793                         if (key.objectid == bytenr &&
794                             key.type == BTRFS_EXTENT_ITEM_KEY &&
795                             key.offset == root->nodesize)
796                                 ret = 0;
797                 }
798         }
799 
800         if (ret == 0) {
801                 leaf = path->nodes[0];
802                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
803                 if (item_size >= sizeof(*ei)) {
804                         ei = btrfs_item_ptr(leaf, path->slots[0],
805                                             struct btrfs_extent_item);
806                         num_refs = btrfs_extent_refs(leaf, ei);
807                         extent_flags = btrfs_extent_flags(leaf, ei);
808                 } else {
809 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
810                         struct btrfs_extent_item_v0 *ei0;
811                         BUG_ON(item_size != sizeof(*ei0));
812                         ei0 = btrfs_item_ptr(leaf, path->slots[0],
813                                              struct btrfs_extent_item_v0);
814                         num_refs = btrfs_extent_refs_v0(leaf, ei0);
815                         /* FIXME: this isn't correct for data */
816                         extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
817 #else
818                         BUG();
819 #endif
820                 }
821                 BUG_ON(num_refs == 0);
822         } else {
823                 num_refs = 0;
824                 extent_flags = 0;
825                 ret = 0;
826         }
827 
828         if (!trans)
829                 goto out;
830 
831         delayed_refs = &trans->transaction->delayed_refs;
832         spin_lock(&delayed_refs->lock);
833         head = btrfs_find_delayed_ref_head(trans, bytenr);
834         if (head) {
835                 if (!mutex_trylock(&head->mutex)) {
836                         atomic_inc(&head->node.refs);
837                         spin_unlock(&delayed_refs->lock);
838 
839                         btrfs_release_path(path);
840 
841                         /*
842                          * Mutex was contended, block until it's released and try
843                          * again
844                          */
845                         mutex_lock(&head->mutex);
846                         mutex_unlock(&head->mutex);
847                         btrfs_put_delayed_ref(&head->node);
848                         goto search_again;
849                 }
850                 spin_lock(&head->lock);
851                 if (head->extent_op && head->extent_op->update_flags)
852                         extent_flags |= head->extent_op->flags_to_set;
853                 else
854                         BUG_ON(num_refs == 0);
855 
856                 num_refs += head->node.ref_mod;
857                 spin_unlock(&head->lock);
858                 mutex_unlock(&head->mutex);
859         }
860         spin_unlock(&delayed_refs->lock);
861 out:
862         WARN_ON(num_refs == 0);
863         if (refs)
864                 *refs = num_refs;
865         if (flags)
866                 *flags = extent_flags;
867 out_free:
868         btrfs_free_path(path);
869         return ret;
870 }
871 
872 /*
873  * Back reference rules.  Back refs have three main goals:
874  *
875  * 1) differentiate between all holders of references to an extent so that
876  *    when a reference is dropped we can make sure it was a valid reference
877  *    before freeing the extent.
878  *
879  * 2) Provide enough information to quickly find the holders of an extent
880  *    if we notice a given block is corrupted or bad.
881  *
882  * 3) Make it easy to migrate blocks for FS shrinking or storage pool
883  *    maintenance.  This is actually the same as #2, but with a slightly
884  *    different use case.
885  *
886  * There are two kinds of back refs. The implicit back refs is optimized
887  * for pointers in non-shared tree blocks. For a given pointer in a block,
888  * back refs of this kind provide information about the block's owner tree
889  * and the pointer's key. These information allow us to find the block by
890  * b-tree searching. The full back refs is for pointers in tree blocks not
891  * referenced by their owner trees. The location of tree block is recorded
892  * in the back refs. Actually the full back refs is generic, and can be
893  * used in all cases the implicit back refs is used. The major shortcoming
894  * of the full back refs is its overhead. Every time a tree block gets
895  * COWed, we have to update back refs entry for all pointers in it.
896  *
897  * For a newly allocated tree block, we use implicit back refs for
898  * pointers in it. This means most tree related operations only involve
899  * implicit back refs. For a tree block created in old transaction, the
900  * only way to drop a reference to it is COW it. So we can detect the
901  * event that tree block loses its owner tree's reference and do the
902  * back refs conversion.
903  *
904  * When a tree block is COW'd through a tree, there are four cases:
905  *
906  * The reference count of the block is one and the tree is the block's
907  * owner tree. Nothing to do in this case.
908  *
909  * The reference count of the block is one and the tree is not the
910  * block's owner tree. In this case, full back refs is used for pointers
911  * in the block. Remove these full back refs, add implicit back refs for
912  * every pointers in the new block.
913  *
914  * The reference count of the block is greater than one and the tree is
915  * the block's owner tree. In this case, implicit back refs is used for
916  * pointers in the block. Add full back refs for every pointers in the
917  * block, increase lower level extents' reference counts. The original
918  * implicit back refs are entailed to the new block.
919  *
920  * The reference count of the block is greater than one and the tree is
921  * not the block's owner tree. Add implicit back refs for every pointer in
922  * the new block, increase lower level extents' reference count.
923  *
924  * Back Reference Key composing:
925  *
926  * The key objectid corresponds to the first byte in the extent,
927  * The key type is used to differentiate between types of back refs.
928  * There are different meanings of the key offset for different types
929  * of back refs.
930  *
931  * File extents can be referenced by:
932  *
933  * - multiple snapshots, subvolumes, or different generations in one subvol
934  * - different files inside a single subvolume
935  * - different offsets inside a file (bookend extents in file.c)
936  *
937  * The extent ref structure for the implicit back refs has fields for:
938  *
939  * - Objectid of the subvolume root
940  * - objectid of the file holding the reference
941  * - original offset in the file
942  * - how many bookend extents
943  *
944  * The key offset for the implicit back refs is hash of the first
945  * three fields.
946  *
947  * The extent ref structure for the full back refs has field for:
948  *
949  * - number of pointers in the tree leaf
950  *
951  * The key offset for the implicit back refs is the first byte of
952  * the tree leaf
953  *
954  * When a file extent is allocated, The implicit back refs is used.
955  * the fields are filled in:
956  *
957  *     (root_key.objectid, inode objectid, offset in file, 1)
958  *
959  * When a file extent is removed file truncation, we find the
960  * corresponding implicit back refs and check the following fields:
961  *
962  *     (btrfs_header_owner(leaf), inode objectid, offset in file)
963  *
964  * Btree extents can be referenced by:
965  *
966  * - Different subvolumes
967  *
968  * Both the implicit back refs and the full back refs for tree blocks
969  * only consist of key. The key offset for the implicit back refs is
970  * objectid of block's owner tree. The key offset for the full back refs
971  * is the first byte of parent block.
972  *
973  * When implicit back refs is used, information about the lowest key and
974  * level of the tree block are required. These information are stored in
975  * tree block info structure.
976  */
977 
978 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
979 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
980                                   struct btrfs_root *root,
981                                   struct btrfs_path *path,
982                                   u64 owner, u32 extra_size)
983 {
984         struct btrfs_extent_item *item;
985         struct btrfs_extent_item_v0 *ei0;
986         struct btrfs_extent_ref_v0 *ref0;
987         struct btrfs_tree_block_info *bi;
988         struct extent_buffer *leaf;
989         struct btrfs_key key;
990         struct btrfs_key found_key;
991         u32 new_size = sizeof(*item);
992         u64 refs;
993         int ret;
994 
995         leaf = path->nodes[0];
996         BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
997 
998         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
999         ei0 = btrfs_item_ptr(leaf, path->slots[0],
1000                              struct btrfs_extent_item_v0);
1001         refs = btrfs_extent_refs_v0(leaf, ei0);
1002 
1003         if (owner == (u64)-1) {
1004                 while (1) {
1005                         if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1006                                 ret = btrfs_next_leaf(root, path);
1007                                 if (ret < 0)
1008                                         return ret;
1009                                 BUG_ON(ret > 0); /* Corruption */
1010                                 leaf = path->nodes[0];
1011                         }
1012                         btrfs_item_key_to_cpu(leaf, &found_key,
1013                                               path->slots[0]);
1014                         BUG_ON(key.objectid != found_key.objectid);
1015                         if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
1016                                 path->slots[0]++;
1017                                 continue;
1018                         }
1019                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
1020                                               struct btrfs_extent_ref_v0);
1021                         owner = btrfs_ref_objectid_v0(leaf, ref0);
1022                         break;
1023                 }
1024         }
1025         btrfs_release_path(path);
1026 
1027         if (owner < BTRFS_FIRST_FREE_OBJECTID)
1028                 new_size += sizeof(*bi);
1029 
1030         new_size -= sizeof(*ei0);
1031         ret = btrfs_search_slot(trans, root, &key, path,
1032                                 new_size + extra_size, 1);
1033         if (ret < 0)
1034                 return ret;
1035         BUG_ON(ret); /* Corruption */
1036 
1037         btrfs_extend_item(root, path, new_size);
1038 
1039         leaf = path->nodes[0];
1040         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1041         btrfs_set_extent_refs(leaf, item, refs);
1042         /* FIXME: get real generation */
1043         btrfs_set_extent_generation(leaf, item, 0);
1044         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1045                 btrfs_set_extent_flags(leaf, item,
1046                                        BTRFS_EXTENT_FLAG_TREE_BLOCK |
1047                                        BTRFS_BLOCK_FLAG_FULL_BACKREF);
1048                 bi = (struct btrfs_tree_block_info *)(item + 1);
1049                 /* FIXME: get first key of the block */
1050                 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
1051                 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1052         } else {
1053                 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1054         }
1055         btrfs_mark_buffer_dirty(leaf);
1056         return 0;
1057 }
1058 #endif
1059 
1060 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1061 {
1062         u32 high_crc = ~(u32)0;
1063         u32 low_crc = ~(u32)0;
1064         __le64 lenum;
1065 
1066         lenum = cpu_to_le64(root_objectid);
1067         high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum));
1068         lenum = cpu_to_le64(owner);
1069         low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1070         lenum = cpu_to_le64(offset);
1071         low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1072 
1073         return ((u64)high_crc << 31) ^ (u64)low_crc;
1074 }
1075 
1076 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1077                                      struct btrfs_extent_data_ref *ref)
1078 {
1079         return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1080                                     btrfs_extent_data_ref_objectid(leaf, ref),
1081                                     btrfs_extent_data_ref_offset(leaf, ref));
1082 }
1083 
1084 static int match_extent_data_ref(struct extent_buffer *leaf,
1085                                  struct btrfs_extent_data_ref *ref,
1086                                  u64 root_objectid, u64 owner, u64 offset)
1087 {
1088         if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1089             btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1090             btrfs_extent_data_ref_offset(leaf, ref) != offset)
1091                 return 0;
1092         return 1;
1093 }
1094 
1095 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1096                                            struct btrfs_root *root,
1097                                            struct btrfs_path *path,
1098                                            u64 bytenr, u64 parent,
1099                                            u64 root_objectid,
1100                                            u64 owner, u64 offset)
1101 {
1102         struct btrfs_key key;
1103         struct btrfs_extent_data_ref *ref;
1104         struct extent_buffer *leaf;
1105         u32 nritems;
1106         int ret;
1107         int recow;
1108         int err = -ENOENT;
1109 
1110         key.objectid = bytenr;
1111         if (parent) {
1112                 key.type = BTRFS_SHARED_DATA_REF_KEY;
1113                 key.offset = parent;
1114         } else {
1115                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1116                 key.offset = hash_extent_data_ref(root_objectid,
1117                                                   owner, offset);
1118         }
1119 again:
1120         recow = 0;
1121         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1122         if (ret < 0) {
1123                 err = ret;
1124                 goto fail;
1125         }
1126 
1127         if (parent) {
1128                 if (!ret)
1129                         return 0;
1130 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1131                 key.type = BTRFS_EXTENT_REF_V0_KEY;
1132                 btrfs_release_path(path);
1133                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1134                 if (ret < 0) {
1135                         err = ret;
1136                         goto fail;
1137                 }
1138                 if (!ret)
1139                         return 0;
1140 #endif
1141                 goto fail;
1142         }
1143 
1144         leaf = path->nodes[0];
1145         nritems = btrfs_header_nritems(leaf);
1146         while (1) {
1147                 if (path->slots[0] >= nritems) {
1148                         ret = btrfs_next_leaf(root, path);
1149                         if (ret < 0)
1150                                 err = ret;
1151                         if (ret)
1152                                 goto fail;
1153 
1154                         leaf = path->nodes[0];
1155                         nritems = btrfs_header_nritems(leaf);
1156                         recow = 1;
1157                 }
1158 
1159                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1160                 if (key.objectid != bytenr ||
1161                     key.type != BTRFS_EXTENT_DATA_REF_KEY)
1162                         goto fail;
1163 
1164                 ref = btrfs_item_ptr(leaf, path->slots[0],
1165                                      struct btrfs_extent_data_ref);
1166 
1167                 if (match_extent_data_ref(leaf, ref, root_objectid,
1168                                           owner, offset)) {
1169                         if (recow) {
1170                                 btrfs_release_path(path);
1171                                 goto again;
1172                         }
1173                         err = 0;
1174                         break;
1175                 }
1176                 path->slots[0]++;
1177         }
1178 fail:
1179         return err;
1180 }
1181 
1182 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1183                                            struct btrfs_root *root,
1184                                            struct btrfs_path *path,
1185                                            u64 bytenr, u64 parent,
1186                                            u64 root_objectid, u64 owner,
1187                                            u64 offset, int refs_to_add)
1188 {
1189         struct btrfs_key key;
1190         struct extent_buffer *leaf;
1191         u32 size;
1192         u32 num_refs;
1193         int ret;
1194 
1195         key.objectid = bytenr;
1196         if (parent) {
1197                 key.type = BTRFS_SHARED_DATA_REF_KEY;
1198                 key.offset = parent;
1199                 size = sizeof(struct btrfs_shared_data_ref);
1200         } else {
1201                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1202                 key.offset = hash_extent_data_ref(root_objectid,
1203                                                   owner, offset);
1204                 size = sizeof(struct btrfs_extent_data_ref);
1205         }
1206 
1207         ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1208         if (ret && ret != -EEXIST)
1209                 goto fail;
1210 
1211         leaf = path->nodes[0];
1212         if (parent) {
1213                 struct btrfs_shared_data_ref *ref;
1214                 ref = btrfs_item_ptr(leaf, path->slots[0],
1215                                      struct btrfs_shared_data_ref);
1216                 if (ret == 0) {
1217                         btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1218                 } else {
1219                         num_refs = btrfs_shared_data_ref_count(leaf, ref);
1220                         num_refs += refs_to_add;
1221                         btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1222                 }
1223         } else {
1224                 struct btrfs_extent_data_ref *ref;
1225                 while (ret == -EEXIST) {
1226                         ref = btrfs_item_ptr(leaf, path->slots[0],
1227                                              struct btrfs_extent_data_ref);
1228                         if (match_extent_data_ref(leaf, ref, root_objectid,
1229                                                   owner, offset))
1230                                 break;
1231                         btrfs_release_path(path);
1232                         key.offset++;
1233                         ret = btrfs_insert_empty_item(trans, root, path, &key,
1234                                                       size);
1235                         if (ret && ret != -EEXIST)
1236                                 goto fail;
1237 
1238                         leaf = path->nodes[0];
1239                 }
1240                 ref = btrfs_item_ptr(leaf, path->slots[0],
1241                                      struct btrfs_extent_data_ref);
1242                 if (ret == 0) {
1243                         btrfs_set_extent_data_ref_root(leaf, ref,
1244                                                        root_objectid);
1245                         btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1246                         btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1247                         btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1248                 } else {
1249                         num_refs = btrfs_extent_data_ref_count(leaf, ref);
1250                         num_refs += refs_to_add;
1251                         btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1252                 }
1253         }
1254         btrfs_mark_buffer_dirty(leaf);
1255         ret = 0;
1256 fail:
1257         btrfs_release_path(path);
1258         return ret;
1259 }
1260 
1261 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1262                                            struct btrfs_root *root,
1263                                            struct btrfs_path *path,
1264                                            int refs_to_drop, int *last_ref)
1265 {
1266         struct btrfs_key key;
1267         struct btrfs_extent_data_ref *ref1 = NULL;
1268         struct btrfs_shared_data_ref *ref2 = NULL;
1269         struct extent_buffer *leaf;
1270         u32 num_refs = 0;
1271         int ret = 0;
1272 
1273         leaf = path->nodes[0];
1274         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1275 
1276         if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1277                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1278                                       struct btrfs_extent_data_ref);
1279                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1280         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1281                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1282                                       struct btrfs_shared_data_ref);
1283                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1284 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1285         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1286                 struct btrfs_extent_ref_v0 *ref0;
1287                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1288                                       struct btrfs_extent_ref_v0);
1289                 num_refs = btrfs_ref_count_v0(leaf, ref0);
1290 #endif
1291         } else {
1292                 BUG();
1293         }
1294 
1295         BUG_ON(num_refs < refs_to_drop);
1296         num_refs -= refs_to_drop;
1297 
1298         if (num_refs == 0) {
1299                 ret = btrfs_del_item(trans, root, path);
1300                 *last_ref = 1;
1301         } else {
1302                 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1303                         btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1304                 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1305                         btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1306 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1307                 else {
1308                         struct btrfs_extent_ref_v0 *ref0;
1309                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
1310                                         struct btrfs_extent_ref_v0);
1311                         btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1312                 }
1313 #endif
1314                 btrfs_mark_buffer_dirty(leaf);
1315         }
1316         return ret;
1317 }
1318 
1319 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1320                                           struct btrfs_path *path,
1321                                           struct btrfs_extent_inline_ref *iref)
1322 {
1323         struct btrfs_key key;
1324         struct extent_buffer *leaf;
1325         struct btrfs_extent_data_ref *ref1;
1326         struct btrfs_shared_data_ref *ref2;
1327         u32 num_refs = 0;
1328 
1329         leaf = path->nodes[0];
1330         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1331         if (iref) {
1332                 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1333                     BTRFS_EXTENT_DATA_REF_KEY) {
1334                         ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1335                         num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1336                 } else {
1337                         ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1338                         num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1339                 }
1340         } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1341                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1342                                       struct btrfs_extent_data_ref);
1343                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1344         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1345                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1346                                       struct btrfs_shared_data_ref);
1347                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1348 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1349         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1350                 struct btrfs_extent_ref_v0 *ref0;
1351                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1352                                       struct btrfs_extent_ref_v0);
1353                 num_refs = btrfs_ref_count_v0(leaf, ref0);
1354 #endif
1355         } else {
1356                 WARN_ON(1);
1357         }
1358         return num_refs;
1359 }
1360 
1361 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1362                                           struct btrfs_root *root,
1363                                           struct btrfs_path *path,
1364                                           u64 bytenr, u64 parent,
1365                                           u64 root_objectid)
1366 {
1367         struct btrfs_key key;
1368         int ret;
1369 
1370         key.objectid = bytenr;
1371         if (parent) {
1372                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1373                 key.offset = parent;
1374         } else {
1375                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1376                 key.offset = root_objectid;
1377         }
1378 
1379         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1380         if (ret > 0)
1381                 ret = -ENOENT;
1382 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1383         if (ret == -ENOENT && parent) {
1384                 btrfs_release_path(path);
1385                 key.type = BTRFS_EXTENT_REF_V0_KEY;
1386                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1387                 if (ret > 0)
1388                         ret = -ENOENT;
1389         }
1390 #endif
1391         return ret;
1392 }
1393 
1394 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1395                                           struct btrfs_root *root,
1396                                           struct btrfs_path *path,
1397                                           u64 bytenr, u64 parent,
1398                                           u64 root_objectid)
1399 {
1400         struct btrfs_key key;
1401         int ret;
1402 
1403         key.objectid = bytenr;
1404         if (parent) {
1405                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1406                 key.offset = parent;
1407         } else {
1408                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1409                 key.offset = root_objectid;
1410         }
1411 
1412         ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1413         btrfs_release_path(path);
1414         return ret;
1415 }
1416 
1417 static inline int extent_ref_type(u64 parent, u64 owner)
1418 {
1419         int type;
1420         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1421                 if (parent > 0)
1422                         type = BTRFS_SHARED_BLOCK_REF_KEY;
1423                 else
1424                         type = BTRFS_TREE_BLOCK_REF_KEY;
1425         } else {
1426                 if (parent > 0)
1427                         type = BTRFS_SHARED_DATA_REF_KEY;
1428                 else
1429                         type = BTRFS_EXTENT_DATA_REF_KEY;
1430         }
1431         return type;
1432 }
1433 
1434 static int find_next_key(struct btrfs_path *path, int level,
1435                          struct btrfs_key *key)
1436 
1437 {
1438         for (; level < BTRFS_MAX_LEVEL; level++) {
1439                 if (!path->nodes[level])
1440                         break;
1441                 if (path->slots[level] + 1 >=
1442                     btrfs_header_nritems(path->nodes[level]))
1443                         continue;
1444                 if (level == 0)
1445                         btrfs_item_key_to_cpu(path->nodes[level], key,
1446                                               path->slots[level] + 1);
1447                 else
1448                         btrfs_node_key_to_cpu(path->nodes[level], key,
1449                                               path->slots[level] + 1);
1450                 return 0;
1451         }
1452         return 1;
1453 }
1454 
1455 /*
1456  * look for inline back ref. if back ref is found, *ref_ret is set
1457  * to the address of inline back ref, and 0 is returned.
1458  *
1459  * if back ref isn't found, *ref_ret is set to the address where it
1460  * should be inserted, and -ENOENT is returned.
1461  *
1462  * if insert is true and there are too many inline back refs, the path
1463  * points to the extent item, and -EAGAIN is returned.
1464  *
1465  * NOTE: inline back refs are ordered in the same way that back ref
1466  *       items in the tree are ordered.
1467  */
1468 static noinline_for_stack
1469 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1470                                  struct btrfs_root *root,
1471                                  struct btrfs_path *path,
1472                                  struct btrfs_extent_inline_ref **ref_ret,
1473                                  u64 bytenr, u64 num_bytes,
1474                                  u64 parent, u64 root_objectid,
1475                                  u64 owner, u64 offset, int insert)
1476 {
1477         struct btrfs_key key;
1478         struct extent_buffer *leaf;
1479         struct btrfs_extent_item *ei;
1480         struct btrfs_extent_inline_ref *iref;
1481         u64 flags;
1482         u64 item_size;
1483         unsigned long ptr;
1484         unsigned long end;
1485         int extra_size;
1486         int type;
1487         int want;
1488         int ret;
1489         int err = 0;
1490         bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
1491                                                  SKINNY_METADATA);
1492 
1493         key.objectid = bytenr;
1494         key.type = BTRFS_EXTENT_ITEM_KEY;
1495         key.offset = num_bytes;
1496 
1497         want = extent_ref_type(parent, owner);
1498         if (insert) {
1499                 extra_size = btrfs_extent_inline_ref_size(want);
1500                 path->keep_locks = 1;
1501         } else
1502                 extra_size = -1;
1503 
1504         /*
1505          * Owner is our parent level, so we can just add one to get the level
1506          * for the block we are interested in.
1507          */
1508         if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
1509                 key.type = BTRFS_METADATA_ITEM_KEY;
1510                 key.offset = owner;
1511         }
1512 
1513 again:
1514         ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1515         if (ret < 0) {
1516                 err = ret;
1517                 goto out;
1518         }
1519 
1520         /*
1521          * We may be a newly converted file system which still has the old fat
1522          * extent entries for metadata, so try and see if we have one of those.
1523          */
1524         if (ret > 0 && skinny_metadata) {
1525                 skinny_metadata = false;
1526                 if (path->slots[0]) {
1527                         path->slots[0]--;
1528                         btrfs_item_key_to_cpu(path->nodes[0], &key,
1529                                               path->slots[0]);
1530                         if (key.objectid == bytenr &&
1531                             key.type == BTRFS_EXTENT_ITEM_KEY &&
1532                             key.offset == num_bytes)
1533                                 ret = 0;
1534                 }
1535                 if (ret) {
1536                         key.objectid = bytenr;
1537                         key.type = BTRFS_EXTENT_ITEM_KEY;
1538                         key.offset = num_bytes;
1539                         btrfs_release_path(path);
1540                         goto again;
1541                 }
1542         }
1543 
1544         if (ret && !insert) {
1545                 err = -ENOENT;
1546                 goto out;
1547         } else if (WARN_ON(ret)) {
1548                 err = -EIO;
1549                 goto out;
1550         }
1551 
1552         leaf = path->nodes[0];
1553         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1554 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1555         if (item_size < sizeof(*ei)) {
1556                 if (!insert) {
1557                         err = -ENOENT;
1558                         goto out;
1559                 }
1560                 ret = convert_extent_item_v0(trans, root, path, owner,
1561                                              extra_size);
1562                 if (ret < 0) {
1563                         err = ret;
1564                         goto out;
1565                 }
1566                 leaf = path->nodes[0];
1567                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1568         }
1569 #endif
1570         BUG_ON(item_size < sizeof(*ei));
1571 
1572         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1573         flags = btrfs_extent_flags(leaf, ei);
1574 
1575         ptr = (unsigned long)(ei + 1);
1576         end = (unsigned long)ei + item_size;
1577 
1578         if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1579                 ptr += sizeof(struct btrfs_tree_block_info);
1580                 BUG_ON(ptr > end);
1581         }
1582 
1583         err = -ENOENT;
1584         while (1) {
1585                 if (ptr >= end) {
1586                         WARN_ON(ptr > end);
1587                         break;
1588                 }
1589                 iref = (struct btrfs_extent_inline_ref *)ptr;
1590                 type = btrfs_extent_inline_ref_type(leaf, iref);
1591                 if (want < type)
1592                         break;
1593                 if (want > type) {
1594                         ptr += btrfs_extent_inline_ref_size(type);
1595                         continue;
1596                 }
1597 
1598                 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1599                         struct btrfs_extent_data_ref *dref;
1600                         dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1601                         if (match_extent_data_ref(leaf, dref, root_objectid,
1602                                                   owner, offset)) {
1603                                 err = 0;
1604                                 break;
1605                         }
1606                         if (hash_extent_data_ref_item(leaf, dref) <
1607                             hash_extent_data_ref(root_objectid, owner, offset))
1608                                 break;
1609                 } else {
1610                         u64 ref_offset;
1611                         ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1612                         if (parent > 0) {
1613                                 if (parent == ref_offset) {
1614                                         err = 0;
1615                                         break;
1616                                 }
1617                                 if (ref_offset < parent)
1618                                         break;
1619                         } else {
1620                                 if (root_objectid == ref_offset) {
1621                                         err = 0;
1622                                         break;
1623                                 }
1624                                 if (ref_offset < root_objectid)
1625                                         break;
1626                         }
1627                 }
1628                 ptr += btrfs_extent_inline_ref_size(type);
1629         }
1630         if (err == -ENOENT && insert) {
1631                 if (item_size + extra_size >=
1632                     BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1633                         err = -EAGAIN;
1634                         goto out;
1635                 }
1636                 /*
1637                  * To add new inline back ref, we have to make sure
1638                  * there is no corresponding back ref item.
1639                  * For simplicity, we just do not add new inline back
1640                  * ref if there is any kind of item for this block
1641                  */
1642                 if (find_next_key(path, 0, &key) == 0 &&
1643                     key.objectid == bytenr &&
1644                     key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1645                         err = -EAGAIN;
1646                         goto out;
1647                 }
1648         }
1649         *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1650 out:
1651         if (insert) {
1652                 path->keep_locks = 0;
1653                 btrfs_unlock_up_safe(path, 1);
1654         }
1655         return err;
1656 }
1657 
1658 /*
1659  * helper to add new inline back ref
1660  */
1661 static noinline_for_stack
1662 void setup_inline_extent_backref(struct btrfs_root *root,
1663                                  struct btrfs_path *path,
1664                                  struct btrfs_extent_inline_ref *iref,
1665                                  u64 parent, u64 root_objectid,
1666                                  u64 owner, u64 offset, int refs_to_add,
1667                                  struct btrfs_delayed_extent_op *extent_op)
1668 {
1669         struct extent_buffer *leaf;
1670         struct btrfs_extent_item *ei;
1671         unsigned long ptr;
1672         unsigned long end;
1673         unsigned long item_offset;
1674         u64 refs;
1675         int size;
1676         int type;
1677 
1678         leaf = path->nodes[0];
1679         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1680         item_offset = (unsigned long)iref - (unsigned long)ei;
1681 
1682         type = extent_ref_type(parent, owner);
1683         size = btrfs_extent_inline_ref_size(type);
1684 
1685         btrfs_extend_item(root, path, size);
1686 
1687         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1688         refs = btrfs_extent_refs(leaf, ei);
1689         refs += refs_to_add;
1690         btrfs_set_extent_refs(leaf, ei, refs);
1691         if (extent_op)
1692                 __run_delayed_extent_op(extent_op, leaf, ei);
1693 
1694         ptr = (unsigned long)ei + item_offset;
1695         end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1696         if (ptr < end - size)
1697                 memmove_extent_buffer(leaf, ptr + size, ptr,
1698                                       end - size - ptr);
1699 
1700         iref = (struct btrfs_extent_inline_ref *)ptr;
1701         btrfs_set_extent_inline_ref_type(leaf, iref, type);
1702         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1703                 struct btrfs_extent_data_ref *dref;
1704                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1705                 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1706                 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1707                 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1708                 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1709         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1710                 struct btrfs_shared_data_ref *sref;
1711                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1712                 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1713                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1714         } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1715                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1716         } else {
1717                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1718         }
1719         btrfs_mark_buffer_dirty(leaf);
1720 }
1721 
1722 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1723                                  struct btrfs_root *root,
1724                                  struct btrfs_path *path,
1725                                  struct btrfs_extent_inline_ref **ref_ret,
1726                                  u64 bytenr, u64 num_bytes, u64 parent,
1727                                  u64 root_objectid, u64 owner, u64 offset)
1728 {
1729         int ret;
1730 
1731         ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1732                                            bytenr, num_bytes, parent,
1733                                            root_objectid, owner, offset, 0);
1734         if (ret != -ENOENT)
1735                 return ret;
1736 
1737         btrfs_release_path(path);
1738         *ref_ret = NULL;
1739 
1740         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1741                 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1742                                             root_objectid);
1743         } else {
1744                 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1745                                              root_objectid, owner, offset);
1746         }
1747         return ret;
1748 }
1749 
1750 /*
1751  * helper to update/remove inline back ref
1752  */
1753 static noinline_for_stack
1754 void update_inline_extent_backref(struct btrfs_root *root,
1755                                   struct btrfs_path *path,
1756                                   struct btrfs_extent_inline_ref *iref,
1757                                   int refs_to_mod,
1758                                   struct btrfs_delayed_extent_op *extent_op,
1759                                   int *last_ref)
1760 {
1761         struct extent_buffer *leaf;
1762         struct btrfs_extent_item *ei;
1763         struct btrfs_extent_data_ref *dref = NULL;
1764         struct btrfs_shared_data_ref *sref = NULL;
1765         unsigned long ptr;
1766         unsigned long end;
1767         u32 item_size;
1768         int size;
1769         int type;
1770         u64 refs;
1771 
1772         leaf = path->nodes[0];
1773         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1774         refs = btrfs_extent_refs(leaf, ei);
1775         WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1776         refs += refs_to_mod;
1777         btrfs_set_extent_refs(leaf, ei, refs);
1778         if (extent_op)
1779                 __run_delayed_extent_op(extent_op, leaf, ei);
1780 
1781         type = btrfs_extent_inline_ref_type(leaf, iref);
1782 
1783         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1784                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1785                 refs = btrfs_extent_data_ref_count(leaf, dref);
1786         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1787                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1788                 refs = btrfs_shared_data_ref_count(leaf, sref);
1789         } else {
1790                 refs = 1;
1791                 BUG_ON(refs_to_mod != -1);
1792         }
1793 
1794         BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1795         refs += refs_to_mod;
1796 
1797         if (refs > 0) {
1798                 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1799                         btrfs_set_extent_data_ref_count(leaf, dref, refs);
1800                 else
1801                         btrfs_set_shared_data_ref_count(leaf, sref, refs);
1802         } else {
1803                 *last_ref = 1;
1804                 size =  btrfs_extent_inline_ref_size(type);
1805                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1806                 ptr = (unsigned long)iref;
1807                 end = (unsigned long)ei + item_size;
1808                 if (ptr + size < end)
1809                         memmove_extent_buffer(leaf, ptr, ptr + size,
1810                                               end - ptr - size);
1811                 item_size -= size;
1812                 btrfs_truncate_item(root, path, item_size, 1);
1813         }
1814         btrfs_mark_buffer_dirty(leaf);
1815 }
1816 
1817 static noinline_for_stack
1818 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1819                                  struct btrfs_root *root,
1820                                  struct btrfs_path *path,
1821                                  u64 bytenr, u64 num_bytes, u64 parent,
1822                                  u64 root_objectid, u64 owner,
1823                                  u64 offset, int refs_to_add,
1824                                  struct btrfs_delayed_extent_op *extent_op)
1825 {
1826         struct btrfs_extent_inline_ref *iref;
1827         int ret;
1828 
1829         ret = lookup_inline_extent_backref(trans, root, path, &iref,
1830                                            bytenr, num_bytes, parent,
1831                                            root_objectid, owner, offset, 1);
1832         if (ret == 0) {
1833                 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1834                 update_inline_extent_backref(root, path, iref,
1835                                              refs_to_add, extent_op, NULL);
1836         } else if (ret == -ENOENT) {
1837                 setup_inline_extent_backref(root, path, iref, parent,
1838                                             root_objectid, owner, offset,
1839                                             refs_to_add, extent_op);
1840                 ret = 0;
1841         }
1842         return ret;
1843 }
1844 
1845 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1846                                  struct btrfs_root *root,
1847                                  struct btrfs_path *path,
1848                                  u64 bytenr, u64 parent, u64 root_objectid,
1849                                  u64 owner, u64 offset, int refs_to_add)
1850 {
1851         int ret;
1852         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1853                 BUG_ON(refs_to_add != 1);
1854                 ret = insert_tree_block_ref(trans, root, path, bytenr,
1855                                             parent, root_objectid);
1856         } else {
1857                 ret = insert_extent_data_ref(trans, root, path, bytenr,
1858                                              parent, root_objectid,
1859                                              owner, offset, refs_to_add);
1860         }
1861         return ret;
1862 }
1863 
1864 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1865                                  struct btrfs_root *root,
1866                                  struct btrfs_path *path,
1867                                  struct btrfs_extent_inline_ref *iref,
1868                                  int refs_to_drop, int is_data, int *last_ref)
1869 {
1870         int ret = 0;
1871 
1872         BUG_ON(!is_data && refs_to_drop != 1);
1873         if (iref) {
1874                 update_inline_extent_backref(root, path, iref,
1875                                              -refs_to_drop, NULL, last_ref);
1876         } else if (is_data) {
1877                 ret = remove_extent_data_ref(trans, root, path, refs_to_drop,
1878                                              last_ref);
1879         } else {
1880                 *last_ref = 1;
1881                 ret = btrfs_del_item(trans, root, path);
1882         }
1883         return ret;
1884 }
1885 
1886 static int btrfs_issue_discard(struct block_device *bdev,
1887                                 u64 start, u64 len)
1888 {
1889         return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
1890 }
1891 
1892 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1893                                 u64 num_bytes, u64 *actual_bytes)
1894 {
1895         int ret;
1896         u64 discarded_bytes = 0;
1897         struct btrfs_bio *bbio = NULL;
1898 
1899 
1900         /* Tell the block device(s) that the sectors can be discarded */
1901         ret = btrfs_map_block(root->fs_info, REQ_DISCARD,
1902                               bytenr, &num_bytes, &bbio, 0);
1903         /* Error condition is -ENOMEM */
1904         if (!ret) {
1905                 struct btrfs_bio_stripe *stripe = bbio->stripes;
1906                 int i;
1907 
1908 
1909                 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
1910                         if (!stripe->dev->can_discard)
1911                                 continue;
1912 
1913                         ret = btrfs_issue_discard(stripe->dev->bdev,
1914                                                   stripe->physical,
1915                                                   stripe->length);
1916                         if (!ret)
1917                                 discarded_bytes += stripe->length;
1918                         else if (ret != -EOPNOTSUPP)
1919                                 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
1920 
1921                         /*
1922                          * Just in case we get back EOPNOTSUPP for some reason,
1923                          * just ignore the return value so we don't screw up
1924                          * people calling discard_extent.
1925                          */
1926                         ret = 0;
1927                 }
1928                 kfree(bbio);
1929         }
1930 
1931         if (actual_bytes)
1932                 *actual_bytes = discarded_bytes;
1933 
1934 
1935         if (ret == -EOPNOTSUPP)
1936                 ret = 0;
1937         return ret;
1938 }
1939 
1940 /* Can return -ENOMEM */
1941 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1942                          struct btrfs_root *root,
1943                          u64 bytenr, u64 num_bytes, u64 parent,
1944                          u64 root_objectid, u64 owner, u64 offset,
1945                          int no_quota)
1946 {
1947         int ret;
1948         struct btrfs_fs_info *fs_info = root->fs_info;
1949 
1950         BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1951                root_objectid == BTRFS_TREE_LOG_OBJECTID);
1952 
1953         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1954                 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
1955                                         num_bytes,
1956                                         parent, root_objectid, (int)owner,
1957                                         BTRFS_ADD_DELAYED_REF, NULL, no_quota);
1958         } else {
1959                 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
1960                                         num_bytes,
1961                                         parent, root_objectid, owner, offset,
1962                                         BTRFS_ADD_DELAYED_REF, NULL, no_quota);
1963         }
1964         return ret;
1965 }
1966 
1967 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1968                                   struct btrfs_root *root,
1969                                   u64 bytenr, u64 num_bytes,
1970                                   u64 parent, u64 root_objectid,
1971                                   u64 owner, u64 offset, int refs_to_add,
1972                                   int no_quota,
1973                                   struct btrfs_delayed_extent_op *extent_op)
1974 {
1975         struct btrfs_fs_info *fs_info = root->fs_info;
1976         struct btrfs_path *path;
1977         struct extent_buffer *leaf;
1978         struct btrfs_extent_item *item;
1979         struct btrfs_key key;
1980         u64 refs;
1981         int ret;
1982         enum btrfs_qgroup_operation_type type = BTRFS_QGROUP_OPER_ADD_EXCL;
1983 
1984         path = btrfs_alloc_path();
1985         if (!path)
1986                 return -ENOMEM;
1987 
1988         if (!is_fstree(root_objectid) || !root->fs_info->quota_enabled)
1989                 no_quota = 1;
1990 
1991         path->reada = 1;
1992         path->leave_spinning = 1;
1993         /* this will setup the path even if it fails to insert the back ref */
1994         ret = insert_inline_extent_backref(trans, fs_info->extent_root, path,
1995                                            bytenr, num_bytes, parent,
1996                                            root_objectid, owner, offset,
1997                                            refs_to_add, extent_op);
1998         if ((ret < 0 && ret != -EAGAIN) || (!ret && no_quota))
1999                 goto out;
2000         /*
2001          * Ok we were able to insert an inline extent and it appears to be a new
2002          * reference, deal with the qgroup accounting.
2003          */
2004         if (!ret && !no_quota) {
2005                 ASSERT(root->fs_info->quota_enabled);
2006                 leaf = path->nodes[0];
2007                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2008                 item = btrfs_item_ptr(leaf, path->slots[0],
2009                                       struct btrfs_extent_item);
2010                 if (btrfs_extent_refs(leaf, item) > (u64)refs_to_add)
2011                         type = BTRFS_QGROUP_OPER_ADD_SHARED;
2012                 btrfs_release_path(path);
2013 
2014                 ret = btrfs_qgroup_record_ref(trans, fs_info, root_objectid,
2015                                               bytenr, num_bytes, type, 0);
2016                 goto out;
2017         }
2018 
2019         /*
2020          * Ok we had -EAGAIN which means we didn't have space to insert and
2021          * inline extent ref, so just update the reference count and add a
2022          * normal backref.
2023          */
2024         leaf = path->nodes[0];
2025         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2026         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2027         refs = btrfs_extent_refs(leaf, item);
2028         if (refs)
2029                 type = BTRFS_QGROUP_OPER_ADD_SHARED;
2030         btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2031         if (extent_op)
2032                 __run_delayed_extent_op(extent_op, leaf, item);
2033 
2034         btrfs_mark_buffer_dirty(leaf);
2035         btrfs_release_path(path);
2036 
2037         if (!no_quota) {
2038                 ret = btrfs_qgroup_record_ref(trans, fs_info, root_objectid,
2039                                               bytenr, num_bytes, type, 0);
2040                 if (ret)
2041                         goto out;
2042         }
2043 
2044         path->reada = 1;
2045         path->leave_spinning = 1;
2046         /* now insert the actual backref */
2047         ret = insert_extent_backref(trans, root->fs_info->extent_root,
2048                                     path, bytenr, parent, root_objectid,
2049                                     owner, offset, refs_to_add);
2050         if (ret)
2051                 btrfs_abort_transaction(trans, root, ret);
2052 out:
2053         btrfs_free_path(path);
2054         return ret;
2055 }
2056 
2057 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2058                                 struct btrfs_root *root,
2059                                 struct btrfs_delayed_ref_node *node,
2060                                 struct btrfs_delayed_extent_op *extent_op,
2061                                 int insert_reserved)
2062 {
2063         int ret = 0;
2064         struct btrfs_delayed_data_ref *ref;
2065         struct btrfs_key ins;
2066         u64 parent = 0;
2067         u64 ref_root = 0;
2068         u64 flags = 0;
2069 
2070         ins.objectid = node->bytenr;
2071         ins.offset = node->num_bytes;
2072         ins.type = BTRFS_EXTENT_ITEM_KEY;
2073 
2074         ref = btrfs_delayed_node_to_data_ref(node);
2075         trace_run_delayed_data_ref(node, ref, node->action);
2076 
2077         if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2078                 parent = ref->parent;
2079         ref_root = ref->root;
2080 
2081         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2082                 if (extent_op)
2083                         flags |= extent_op->flags_to_set;
2084                 ret = alloc_reserved_file_extent(trans, root,
2085                                                  parent, ref_root, flags,
2086                                                  ref->objectid, ref->offset,
2087                                                  &ins, node->ref_mod);
2088         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2089                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2090                                              node->num_bytes, parent,
2091                                              ref_root, ref->objectid,
2092                                              ref->offset, node->ref_mod,
2093                                              node->no_quota, extent_op);
2094         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2095                 ret = __btrfs_free_extent(trans, root, node->bytenr,
2096                                           node->num_bytes, parent,
2097                                           ref_root, ref->objectid,
2098                                           ref->offset, node->ref_mod,
2099                                           extent_op, node->no_quota);
2100         } else {
2101                 BUG();
2102         }
2103         return ret;
2104 }
2105 
2106 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2107                                     struct extent_buffer *leaf,
2108                                     struct btrfs_extent_item *ei)
2109 {
2110         u64 flags = btrfs_extent_flags(leaf, ei);
2111         if (extent_op->update_flags) {
2112                 flags |= extent_op->flags_to_set;
2113                 btrfs_set_extent_flags(leaf, ei, flags);
2114         }
2115 
2116         if (extent_op->update_key) {
2117                 struct btrfs_tree_block_info *bi;
2118                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2119                 bi = (struct btrfs_tree_block_info *)(ei + 1);
2120                 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2121         }
2122 }
2123 
2124 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2125                                  struct btrfs_root *root,
2126                                  struct btrfs_delayed_ref_node *node,
2127                                  struct btrfs_delayed_extent_op *extent_op)
2128 {
2129         struct btrfs_key key;
2130         struct btrfs_path *path;
2131         struct btrfs_extent_item *ei;
2132         struct extent_buffer *leaf;
2133         u32 item_size;
2134         int ret;
2135         int err = 0;
2136         int metadata = !extent_op->is_data;
2137 
2138         if (trans->aborted)
2139                 return 0;
2140 
2141         if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2142                 metadata = 0;
2143 
2144         path = btrfs_alloc_path();
2145         if (!path)
2146                 return -ENOMEM;
2147 
2148         key.objectid = node->bytenr;
2149 
2150         if (metadata) {
2151                 key.type = BTRFS_METADATA_ITEM_KEY;
2152                 key.offset = extent_op->level;
2153         } else {
2154                 key.type = BTRFS_EXTENT_ITEM_KEY;
2155                 key.offset = node->num_bytes;
2156         }
2157 
2158 again:
2159         path->reada = 1;
2160         path->leave_spinning = 1;
2161         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2162                                 path, 0, 1);
2163         if (ret < 0) {
2164                 err = ret;
2165                 goto out;
2166         }
2167         if (ret > 0) {
2168                 if (metadata) {
2169                         if (path->slots[0] > 0) {
2170                                 path->slots[0]--;
2171                                 btrfs_item_key_to_cpu(path->nodes[0], &key,
2172                                                       path->slots[0]);
2173                                 if (key.objectid == node->bytenr &&
2174                                     key.type == BTRFS_EXTENT_ITEM_KEY &&
2175                                     key.offset == node->num_bytes)
2176                                         ret = 0;
2177                         }
2178                         if (ret > 0) {
2179                                 btrfs_release_path(path);
2180                                 metadata = 0;
2181 
2182                                 key.objectid = node->bytenr;
2183                                 key.offset = node->num_bytes;
2184                                 key.type = BTRFS_EXTENT_ITEM_KEY;
2185                                 goto again;
2186                         }
2187                 } else {
2188                         err = -EIO;
2189                         goto out;
2190                 }
2191         }
2192 
2193         leaf = path->nodes[0];
2194         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2195 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2196         if (item_size < sizeof(*ei)) {
2197                 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2198                                              path, (u64)-1, 0);
2199                 if (ret < 0) {
2200                         err = ret;
2201                         goto out;
2202                 }
2203                 leaf = path->nodes[0];
2204                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2205         }
2206 #endif
2207         BUG_ON(item_size < sizeof(*ei));
2208         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2209         __run_delayed_extent_op(extent_op, leaf, ei);
2210 
2211         btrfs_mark_buffer_dirty(leaf);
2212 out:
2213         btrfs_free_path(path);
2214         return err;
2215 }
2216 
2217 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2218                                 struct btrfs_root *root,
2219                                 struct btrfs_delayed_ref_node *node,
2220                                 struct btrfs_delayed_extent_op *extent_op,
2221                                 int insert_reserved)
2222 {
2223         int ret = 0;
2224         struct btrfs_delayed_tree_ref *ref;
2225         struct btrfs_key ins;
2226         u64 parent = 0;
2227         u64 ref_root = 0;
2228         bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
2229                                                  SKINNY_METADATA);
2230 
2231         ref = btrfs_delayed_node_to_tree_ref(node);
2232         trace_run_delayed_tree_ref(node, ref, node->action);
2233 
2234         if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2235                 parent = ref->parent;
2236         ref_root = ref->root;
2237 
2238         ins.objectid = node->bytenr;
2239         if (skinny_metadata) {
2240                 ins.offset = ref->level;
2241                 ins.type = BTRFS_METADATA_ITEM_KEY;
2242         } else {
2243                 ins.offset = node->num_bytes;
2244                 ins.type = BTRFS_EXTENT_ITEM_KEY;
2245         }
2246 
2247         BUG_ON(node->ref_mod != 1);
2248         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2249                 BUG_ON(!extent_op || !extent_op->update_flags);
2250                 ret = alloc_reserved_tree_block(trans, root,
2251                                                 parent, ref_root,
2252                                                 extent_op->flags_to_set,
2253                                                 &extent_op->key,
2254                                                 ref->level, &ins,
2255                                                 node->no_quota);
2256         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2257                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2258                                              node->num_bytes, parent, ref_root,
2259                                              ref->level, 0, 1, node->no_quota,
2260                                              extent_op);
2261         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2262                 ret = __btrfs_free_extent(trans, root, node->bytenr,
2263                                           node->num_bytes, parent, ref_root,
2264                                           ref->level, 0, 1, extent_op,
2265                                           node->no_quota);
2266         } else {
2267                 BUG();
2268         }
2269         return ret;
2270 }
2271 
2272 /* helper function to actually process a single delayed ref entry */
2273 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2274                                struct btrfs_root *root,
2275                                struct btrfs_delayed_ref_node *node,
2276                                struct btrfs_delayed_extent_op *extent_op,
2277                                int insert_reserved)
2278 {
2279         int ret = 0;
2280 
2281         if (trans->aborted) {
2282                 if (insert_reserved)
2283                         btrfs_pin_extent(root, node->bytenr,
2284                                          node->num_bytes, 1);
2285                 return 0;
2286         }
2287 
2288         if (btrfs_delayed_ref_is_head(node)) {
2289                 struct btrfs_delayed_ref_head *head;
2290                 /*
2291                  * we've hit the end of the chain and we were supposed
2292                  * to insert this extent into the tree.  But, it got
2293                  * deleted before we ever needed to insert it, so all
2294                  * we have to do is clean up the accounting
2295                  */
2296                 BUG_ON(extent_op);
2297                 head = btrfs_delayed_node_to_head(node);
2298                 trace_run_delayed_ref_head(node, head, node->action);
2299 
2300                 if (insert_reserved) {
2301                         btrfs_pin_extent(root, node->bytenr,
2302                                          node->num_bytes, 1);
2303                         if (head->is_data) {
2304                                 ret = btrfs_del_csums(trans, root,
2305                                                       node->bytenr,
2306                                                       node->num_bytes);
2307                         }
2308                 }
2309                 return ret;
2310         }
2311 
2312         if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2313             node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2314                 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2315                                            insert_reserved);
2316         else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2317                  node->type == BTRFS_SHARED_DATA_REF_KEY)
2318                 ret = run_delayed_data_ref(trans, root, node, extent_op,
2319                                            insert_reserved);
2320         else
2321                 BUG();
2322         return ret;
2323 }
2324 
2325 static noinline struct btrfs_delayed_ref_node *
2326 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2327 {
2328         struct rb_node *node;
2329         struct btrfs_delayed_ref_node *ref, *last = NULL;;
2330 
2331         /*
2332          * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2333          * this prevents ref count from going down to zero when
2334          * there still are pending delayed ref.
2335          */
2336         node = rb_first(&head->ref_root);
2337         while (node) {
2338                 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2339                                 rb_node);
2340                 if (ref->action == BTRFS_ADD_DELAYED_REF)
2341                         return ref;
2342                 else if (last == NULL)
2343                         last = ref;
2344                 node = rb_next(node);
2345         }
2346         return last;
2347 }
2348 
2349 /*
2350  * Returns 0 on success or if called with an already aborted transaction.
2351  * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2352  */
2353 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2354                                              struct btrfs_root *root,
2355                                              unsigned long nr)
2356 {
2357         struct btrfs_delayed_ref_root *delayed_refs;
2358         struct btrfs_delayed_ref_node *ref;
2359         struct btrfs_delayed_ref_head *locked_ref = NULL;
2360         struct btrfs_delayed_extent_op *extent_op;
2361         struct btrfs_fs_info *fs_info = root->fs_info;
2362         ktime_t start = ktime_get();
2363         int ret;
2364         unsigned long count = 0;
2365         unsigned long actual_count = 0;
2366         int must_insert_reserved = 0;
2367 
2368         delayed_refs = &trans->transaction->delayed_refs;
2369         while (1) {
2370                 if (!locked_ref) {
2371                         if (count >= nr)
2372                                 break;
2373 
2374                         spin_lock(&delayed_refs->lock);
2375                         locked_ref = btrfs_select_ref_head(trans);
2376                         if (!locked_ref) {
2377                                 spin_unlock(&delayed_refs->lock);
2378                                 break;
2379                         }
2380 
2381                         /* grab the lock that says we are going to process
2382                          * all the refs for this head */
2383                         ret = btrfs_delayed_ref_lock(trans, locked_ref);
2384                         spin_unlock(&delayed_refs->lock);
2385                         /*
2386                          * we may have dropped the spin lock to get the head
2387                          * mutex lock, and that might have given someone else
2388                          * time to free the head.  If that's true, it has been
2389                          * removed from our list and we can move on.
2390                          */
2391                         if (ret == -EAGAIN) {
2392                                 locked_ref = NULL;
2393                                 count++;
2394                                 continue;
2395                         }
2396                 }
2397 
2398                 /*
2399                  * We need to try and merge add/drops of the same ref since we
2400                  * can run into issues with relocate dropping the implicit ref
2401                  * and then it being added back again before the drop can
2402                  * finish.  If we merged anything we need to re-loop so we can
2403                  * get a good ref.
2404                  */
2405                 spin_lock(&locked_ref->lock);
2406                 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2407                                          locked_ref);
2408 
2409                 /*
2410                  * locked_ref is the head node, so we have to go one
2411                  * node back for any delayed ref updates
2412                  */
2413                 ref = select_delayed_ref(locked_ref);
2414 
2415                 if (ref && ref->seq &&
2416                     btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2417                         spin_unlock(&locked_ref->lock);
2418                         btrfs_delayed_ref_unlock(locked_ref);
2419                         spin_lock(&delayed_refs->lock);
2420                         locked_ref->processing = 0;
2421                         delayed_refs->num_heads_ready++;
2422                         spin_unlock(&delayed_refs->lock);
2423                         locked_ref = NULL;
2424                         cond_resched();
2425                         count++;
2426                         continue;
2427                 }
2428 
2429                 /*
2430                  * record the must insert reserved flag before we
2431                  * drop the spin lock.
2432                  */
2433                 must_insert_reserved = locked_ref->must_insert_reserved;
2434                 locked_ref->must_insert_reserved = 0;
2435 
2436                 extent_op = locked_ref->extent_op;
2437                 locked_ref->extent_op = NULL;
2438 
2439                 if (!ref) {
2440 
2441 
2442                         /* All delayed refs have been processed, Go ahead
2443                          * and send the head node to run_one_delayed_ref,
2444                          * so that any accounting fixes can happen
2445                          */
2446                         ref = &locked_ref->node;
2447 
2448                         if (extent_op && must_insert_reserved) {
2449                                 btrfs_free_delayed_extent_op(extent_op);
2450                                 extent_op = NULL;
2451                         }
2452 
2453                         if (extent_op) {
2454                                 spin_unlock(&locked_ref->lock);
2455                                 ret = run_delayed_extent_op(trans, root,
2456                                                             ref, extent_op);
2457                                 btrfs_free_delayed_extent_op(extent_op);
2458 
2459                                 if (ret) {
2460                                         /*
2461                                          * Need to reset must_insert_reserved if
2462                                          * there was an error so the abort stuff
2463                                          * can cleanup the reserved space
2464                                          * properly.
2465                                          */
2466                                         if (must_insert_reserved)
2467                                                 locked_ref->must_insert_reserved = 1;
2468                                         locked_ref->processing = 0;
2469                                         btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2470                                         btrfs_delayed_ref_unlock(locked_ref);
2471                                         return ret;
2472                                 }
2473                                 continue;
2474                         }
2475 
2476                         /*
2477                          * Need to drop our head ref lock and re-aqcuire the
2478                          * delayed ref lock and then re-check to make sure
2479                          * nobody got added.
2480                          */
2481                         spin_unlock(&locked_ref->lock);
2482                         spin_lock(&delayed_refs->lock);
2483                         spin_lock(&locked_ref->lock);
2484                         if (rb_first(&locked_ref->ref_root) ||
2485                             locked_ref->extent_op) {
2486                                 spin_unlock(&locked_ref->lock);
2487                                 spin_unlock(&delayed_refs->lock);
2488                                 continue;
2489                         }
2490                         ref->in_tree = 0;
2491                         delayed_refs->num_heads--;
2492                         rb_erase(&locked_ref->href_node,
2493                                  &delayed_refs->href_root);
2494                         spin_unlock(&delayed_refs->lock);
2495                 } else {
2496                         actual_count++;
2497                         ref->in_tree = 0;
2498                         rb_erase(&ref->rb_node, &locked_ref->ref_root);
2499                 }
2500                 atomic_dec(&delayed_refs->num_entries);
2501 
2502                 if (!btrfs_delayed_ref_is_head(ref)) {
2503                         /*
2504                          * when we play the delayed ref, also correct the
2505                          * ref_mod on head
2506                          */
2507                         switch (ref->action) {
2508                         case BTRFS_ADD_DELAYED_REF:
2509                         case BTRFS_ADD_DELAYED_EXTENT:
2510                                 locked_ref->node.ref_mod -= ref->ref_mod;
2511                                 break;
2512                         case BTRFS_DROP_DELAYED_REF:
2513                                 locked_ref->node.ref_mod += ref->ref_mod;
2514                                 break;
2515                         default:
2516                                 WARN_ON(1);
2517                         }
2518                 }
2519                 spin_unlock(&locked_ref->lock);
2520 
2521                 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2522                                           must_insert_reserved);
2523 
2524                 btrfs_free_delayed_extent_op(extent_op);
2525                 if (ret) {
2526                         locked_ref->processing = 0;
2527                         btrfs_delayed_ref_unlock(locked_ref);
2528                         btrfs_put_delayed_ref(ref);
2529                         btrfs_debug(fs_info, "run_one_delayed_ref returned %d", ret);
2530                         return ret;
2531                 }
2532 
2533                 /*
2534                  * If this node is a head, that means all the refs in this head
2535                  * have been dealt with, and we will pick the next head to deal
2536                  * with, so we must unlock the head and drop it from the cluster
2537                  * list before we release it.
2538                  */
2539                 if (btrfs_delayed_ref_is_head(ref)) {
2540                         btrfs_delayed_ref_unlock(locked_ref);
2541                         locked_ref = NULL;
2542                 }
2543                 btrfs_put_delayed_ref(ref);
2544                 count++;
2545                 cond_resched();
2546         }
2547 
2548         /*
2549          * We don't want to include ref heads since we can have empty ref heads
2550          * and those will drastically skew our runtime down since we just do
2551          * accounting, no actual extent tree updates.
2552          */
2553         if (actual_count > 0) {
2554                 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2555                 u64 avg;
2556 
2557                 /*
2558                  * We weigh the current average higher than our current runtime
2559                  * to avoid large swings in the average.
2560                  */
2561                 spin_lock(&delayed_refs->lock);
2562                 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2563                 avg = div64_u64(avg, 4);
2564                 fs_info->avg_delayed_ref_runtime = avg;
2565                 spin_unlock(&delayed_refs->lock);
2566         }
2567         return 0;
2568 }
2569 
2570 #ifdef SCRAMBLE_DELAYED_REFS
2571 /*
2572  * Normally delayed refs get processed in ascending bytenr order. This
2573  * correlates in most cases to the order added. To expose dependencies on this
2574  * order, we start to process the tree in the middle instead of the beginning
2575  */
2576 static u64 find_middle(struct rb_root *root)
2577 {
2578         struct rb_node *n = root->rb_node;
2579         struct btrfs_delayed_ref_node *entry;
2580         int alt = 1;
2581         u64 middle;
2582         u64 first = 0, last = 0;
2583 
2584         n = rb_first(root);
2585         if (n) {
2586                 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2587                 first = entry->bytenr;
2588         }
2589         n = rb_last(root);
2590         if (n) {
2591                 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2592                 last = entry->bytenr;
2593         }
2594         n = root->rb_node;
2595 
2596         while (n) {
2597                 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2598                 WARN_ON(!entry->in_tree);
2599 
2600                 middle = entry->bytenr;
2601 
2602                 if (alt)
2603                         n = n->rb_left;
2604                 else
2605                         n = n->rb_right;
2606 
2607                 alt = 1 - alt;
2608         }
2609         return middle;
2610 }
2611 #endif
2612 
2613 static inline u64 heads_to_leaves(struct btrfs_root *root, u64 heads)
2614 {
2615         u64 num_bytes;
2616 
2617         num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2618                              sizeof(struct btrfs_extent_inline_ref));
2619         if (!btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2620                 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2621 
2622         /*
2623          * We don't ever fill up leaves all the way so multiply by 2 just to be
2624          * closer to what we're really going to want to ouse.
2625          */
2626         return div64_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(root));
2627 }
2628 
2629 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans,
2630                                        struct btrfs_root *root)
2631 {
2632         struct btrfs_block_rsv *global_rsv;
2633         u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2634         u64 num_bytes;
2635         int ret = 0;
2636 
2637         num_bytes = btrfs_calc_trans_metadata_size(root, 1);
2638         num_heads = heads_to_leaves(root, num_heads);
2639         if (num_heads > 1)
2640                 num_bytes += (num_heads - 1) * root->nodesize;
2641         num_bytes <<= 1;
2642         global_rsv = &root->fs_info->global_block_rsv;
2643 
2644         /*
2645          * If we can't allocate any more chunks lets make sure we have _lots_ of
2646          * wiggle room since running delayed refs can create more delayed refs.
2647          */
2648         if (global_rsv->space_info->full)
2649                 num_bytes <<= 1;
2650 
2651         spin_lock(&global_rsv->lock);
2652         if (global_rsv->reserved <= num_bytes)
2653                 ret = 1;
2654         spin_unlock(&global_rsv->lock);
2655         return ret;
2656 }
2657 
2658 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2659                                        struct btrfs_root *root)
2660 {
2661         struct btrfs_fs_info *fs_info = root->fs_info;
2662         u64 num_entries =
2663                 atomic_read(&trans->transaction->delayed_refs.num_entries);
2664         u64 avg_runtime;
2665         u64 val;
2666 
2667         smp_mb();
2668         avg_runtime = fs_info->avg_delayed_ref_runtime;
2669         val = num_entries * avg_runtime;
2670         if (num_entries * avg_runtime >= NSEC_PER_SEC)
2671                 return 1;
2672         if (val >= NSEC_PER_SEC / 2)
2673                 return 2;
2674 
2675         return btrfs_check_space_for_delayed_refs(trans, root);
2676 }
2677 
2678 struct async_delayed_refs {
2679         struct btrfs_root *root;
2680         int count;
2681         int error;
2682         int sync;
2683         struct completion wait;
2684         struct btrfs_work work;
2685 };
2686 
2687 static void delayed_ref_async_start(struct btrfs_work *work)
2688 {
2689         struct async_delayed_refs *async;
2690         struct btrfs_trans_handle *trans;
2691         int ret;
2692 
2693         async = container_of(work, struct async_delayed_refs, work);
2694 
2695         trans = btrfs_join_transaction(async->root);
2696         if (IS_ERR(trans)) {
2697                 async->error = PTR_ERR(trans);
2698                 goto done;
2699         }
2700 
2701         /*
2702          * trans->sync means that when we call end_transaciton, we won't
2703          * wait on delayed refs
2704          */
2705         trans->sync = true;
2706         ret = btrfs_run_delayed_refs(trans, async->root, async->count);
2707         if (ret)
2708                 async->error = ret;
2709 
2710         ret = btrfs_end_transaction(trans, async->root);
2711         if (ret && !async->error)
2712                 async->error = ret;
2713 done:
2714         if (async->sync)
2715                 complete(&async->wait);
2716         else
2717                 kfree(async);
2718 }
2719 
2720 int btrfs_async_run_delayed_refs(struct btrfs_root *root,
2721                                  unsigned long count, int wait)
2722 {
2723         struct async_delayed_refs *async;
2724         int ret;
2725 
2726         async = kmalloc(sizeof(*async), GFP_NOFS);
2727         if (!async)
2728                 return -ENOMEM;
2729 
2730         async->root = root->fs_info->tree_root;
2731         async->count = count;
2732         async->error = 0;
2733         if (wait)
2734                 async->sync = 1;
2735         else
2736                 async->sync = 0;
2737         init_completion(&async->wait);
2738 
2739         btrfs_init_work(&async->work, btrfs_extent_refs_helper,
2740                         delayed_ref_async_start, NULL, NULL);
2741 
2742         btrfs_queue_work(root->fs_info->extent_workers, &async->work);
2743 
2744         if (wait) {
2745                 wait_for_completion(&async->wait);
2746                 ret = async->error;
2747                 kfree(async);
2748                 return ret;
2749         }
2750         return 0;
2751 }
2752 
2753 /*
2754  * this starts processing the delayed reference count updates and
2755  * extent insertions we have queued up so far.  count can be
2756  * 0, which means to process everything in the tree at the start
2757  * of the run (but not newly added entries), or it can be some target
2758  * number you'd like to process.
2759  *
2760  * Returns 0 on success or if called with an aborted transaction
2761  * Returns <0 on error and aborts the transaction
2762  */
2763 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2764                            struct btrfs_root *root, unsigned long count)
2765 {
2766         struct rb_node *node;
2767         struct btrfs_delayed_ref_root *delayed_refs;
2768         struct btrfs_delayed_ref_head *head;
2769         int ret;
2770         int run_all = count == (unsigned long)-1;
2771         int run_most = 0;
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                 run_most = 1;
2784         }
2785 
2786 again:
2787 #ifdef SCRAMBLE_DELAYED_REFS
2788         delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2789 #endif
2790         ret = __btrfs_run_delayed_refs(trans, root, count);
2791         if (ret < 0) {
2792                 btrfs_abort_transaction(trans, root, ret);
2793                 return ret;
2794         }
2795 
2796         if (run_all) {
2797                 if (!list_empty(&trans->new_bgs))
2798                         btrfs_create_pending_block_groups(trans, root);
2799 
2800                 spin_lock(&delayed_refs->lock);
2801                 node = rb_first(&delayed_refs->href_root);
2802                 if (!node) {
2803                         spin_unlock(&delayed_refs->lock);
2804                         goto out;
2805                 }
2806                 count = (unsigned long)-1;
2807 
2808                 while (node) {
2809                         head = rb_entry(node, struct btrfs_delayed_ref_head,
2810                                         href_node);
2811                         if (btrfs_delayed_ref_is_head(&head->node)) {
2812                                 struct btrfs_delayed_ref_node *ref;
2813 
2814                                 ref = &head->node;
2815                                 atomic_inc(&ref->refs);
2816 
2817                                 spin_unlock(&delayed_refs->lock);
2818                                 /*
2819                                  * Mutex was contended, block until it's
2820                                  * released and try again
2821                                  */
2822                                 mutex_lock(&head->mutex);
2823                                 mutex_unlock(&head->mutex);
2824 
2825                                 btrfs_put_delayed_ref(ref);
2826                                 cond_resched();
2827                                 goto again;
2828                         } else {
2829                                 WARN_ON(1);
2830                         }
2831                         node = rb_next(node);
2832                 }
2833                 spin_unlock(&delayed_refs->lock);
2834                 cond_resched();
2835                 goto again;
2836         }
2837 out:
2838         ret = btrfs_delayed_qgroup_accounting(trans, root->fs_info);
2839         if (ret)
2840                 return ret;
2841         assert_qgroups_uptodate(trans);
2842         return 0;
2843 }
2844 
2845 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2846                                 struct btrfs_root *root,
2847                                 u64 bytenr, u64 num_bytes, u64 flags,
2848                                 int level, int is_data)
2849 {
2850         struct btrfs_delayed_extent_op *extent_op;
2851         int ret;
2852 
2853         extent_op = btrfs_alloc_delayed_extent_op();
2854         if (!extent_op)
2855                 return -ENOMEM;
2856 
2857         extent_op->flags_to_set = flags;
2858         extent_op->update_flags = 1;
2859         extent_op->update_key = 0;
2860         extent_op->is_data = is_data ? 1 : 0;
2861         extent_op->level = level;
2862 
2863         ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2864                                           num_bytes, extent_op);
2865         if (ret)
2866                 btrfs_free_delayed_extent_op(extent_op);
2867         return ret;
2868 }
2869 
2870 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2871                                       struct btrfs_root *root,
2872                                       struct btrfs_path *path,
2873                                       u64 objectid, u64 offset, u64 bytenr)
2874 {
2875         struct btrfs_delayed_ref_head *head;
2876         struct btrfs_delayed_ref_node *ref;
2877         struct btrfs_delayed_data_ref *data_ref;
2878         struct btrfs_delayed_ref_root *delayed_refs;
2879         struct rb_node *node;
2880         int ret = 0;
2881 
2882         delayed_refs = &trans->transaction->delayed_refs;
2883         spin_lock(&delayed_refs->lock);
2884         head = btrfs_find_delayed_ref_head(trans, bytenr);
2885         if (!head) {
2886                 spin_unlock(&delayed_refs->lock);
2887                 return 0;
2888         }
2889 
2890         if (!mutex_trylock(&head->mutex)) {
2891                 atomic_inc(&head->node.refs);
2892                 spin_unlock(&delayed_refs->lock);
2893 
2894                 btrfs_release_path(path);
2895 
2896                 /*
2897                  * Mutex was contended, block until it's released and let
2898                  * caller try again
2899                  */
2900                 mutex_lock(&head->mutex);
2901                 mutex_unlock(&head->mutex);
2902                 btrfs_put_delayed_ref(&head->node);
2903                 return -EAGAIN;
2904         }
2905         spin_unlock(&delayed_refs->lock);
2906 
2907         spin_lock(&head->lock);
2908         node = rb_first(&head->ref_root);
2909         while (node) {
2910                 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2911                 node = rb_next(node);
2912 
2913                 /* If it's a shared ref we know a cross reference exists */
2914                 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
2915                         ret = 1;
2916                         break;
2917                 }
2918 
2919                 data_ref = btrfs_delayed_node_to_data_ref(ref);
2920 
2921                 /*
2922                  * If our ref doesn't match the one we're currently looking at
2923                  * then we have a cross reference.
2924                  */
2925                 if (data_ref->root != root->root_key.objectid ||
2926                     data_ref->objectid != objectid ||
2927                     data_ref->offset != offset) {
2928                         ret = 1;
2929                         break;
2930                 }
2931         }
2932         spin_unlock(&head->lock);
2933         mutex_unlock(&head->mutex);
2934         return ret;
2935 }
2936 
2937 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2938                                         struct btrfs_root *root,
2939                                         struct btrfs_path *path,
2940                                         u64 objectid, u64 offset, u64 bytenr)
2941 {
2942         struct btrfs_root *extent_root = root->fs_info->extent_root;
2943         struct extent_buffer *leaf;
2944         struct btrfs_extent_data_ref *ref;
2945         struct btrfs_extent_inline_ref *iref;
2946         struct btrfs_extent_item *ei;
2947         struct btrfs_key key;
2948         u32 item_size;
2949         int ret;
2950 
2951         key.objectid = bytenr;
2952         key.offset = (u64)-1;
2953         key.type = BTRFS_EXTENT_ITEM_KEY;
2954 
2955         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2956         if (ret < 0)
2957                 goto out;
2958         BUG_ON(ret == 0); /* Corruption */
2959 
2960         ret = -ENOENT;
2961         if (path->slots[0] == 0)
2962                 goto out;
2963 
2964         path->slots[0]--;
2965         leaf = path->nodes[0];
2966         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2967 
2968         if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2969                 goto out;
2970 
2971         ret = 1;
2972         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2973 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2974         if (item_size < sizeof(*ei)) {
2975                 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2976                 goto out;
2977         }
2978 #endif
2979         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2980 
2981         if (item_size != sizeof(*ei) +
2982             btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2983                 goto out;
2984 
2985         if (btrfs_extent_generation(leaf, ei) <=
2986             btrfs_root_last_snapshot(&root->root_item))
2987                 goto out;
2988 
2989         iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2990         if (btrfs_extent_inline_ref_type(leaf, iref) !=
2991             BTRFS_EXTENT_DATA_REF_KEY)
2992                 goto out;
2993 
2994         ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2995         if (btrfs_extent_refs(leaf, ei) !=
2996             btrfs_extent_data_ref_count(leaf, ref) ||
2997             btrfs_extent_data_ref_root(leaf, ref) !=
2998             root->root_key.objectid ||
2999             btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
3000             btrfs_extent_data_ref_offset(leaf, ref) != offset)
3001                 goto out;
3002 
3003         ret = 0;
3004 out:
3005         return ret;
3006 }
3007 
3008 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
3009                           struct btrfs_root *root,
3010                           u64 objectid, u64 offset, u64 bytenr)
3011 {
3012         struct btrfs_path *path;
3013         int ret;
3014         int ret2;
3015 
3016         path = btrfs_alloc_path();
3017         if (!path)
3018                 return -ENOENT;
3019 
3020         do {
3021                 ret = check_committed_ref(trans, root, path, objectid,
3022                                           offset, bytenr);
3023                 if (ret && ret != -ENOENT)
3024                         goto out;
3025 
3026                 ret2 = check_delayed_ref(trans, root, path, objectid,
3027                                          offset, bytenr);
3028         } while (ret2 == -EAGAIN);
3029 
3030         if (ret2 && ret2 != -ENOENT) {
3031                 ret = ret2;
3032                 goto out;
3033         }
3034 
3035         if (ret != -ENOENT || ret2 != -ENOENT)
3036                 ret = 0;
3037 out:
3038         btrfs_free_path(path);
3039         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3040                 WARN_ON(ret > 0);
3041         return ret;
3042 }
3043 
3044 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3045                            struct btrfs_root *root,
3046                            struct extent_buffer *buf,
3047                            int full_backref, int inc)
3048 {
3049         u64 bytenr;
3050         u64 num_bytes;
3051         u64 parent;
3052         u64 ref_root;
3053         u32 nritems;
3054         struct btrfs_key key;
3055         struct btrfs_file_extent_item *fi;
3056         int i;
3057         int level;
3058         int ret = 0;
3059         int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
3060                             u64, u64, u64, u64, u64, u64, int);
3061 
3062 
3063         if (btrfs_test_is_dummy_root(root))
3064                 return 0;
3065 
3066         ref_root = btrfs_header_owner(buf);
3067         nritems = btrfs_header_nritems(buf);
3068         level = btrfs_header_level(buf);
3069 
3070         if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0)
3071                 return 0;
3072 
3073         if (inc)
3074                 process_func = btrfs_inc_extent_ref;
3075         else
3076                 process_func = btrfs_free_extent;
3077 
3078         if (full_backref)
3079                 parent = buf->start;
3080         else
3081                 parent = 0;
3082 
3083         for (i = 0; i < nritems; i++) {
3084                 if (level == 0) {
3085                         btrfs_item_key_to_cpu(buf, &key, i);
3086                         if (key.type != BTRFS_EXTENT_DATA_KEY)
3087                                 continue;
3088                         fi = btrfs_item_ptr(buf, i,
3089                                             struct btrfs_file_extent_item);
3090                         if (btrfs_file_extent_type(buf, fi) ==
3091                             BTRFS_FILE_EXTENT_INLINE)
3092                                 continue;
3093                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3094                         if (bytenr == 0)
3095                                 continue;
3096 
3097                         num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3098                         key.offset -= btrfs_file_extent_offset(buf, fi);
3099                         ret = process_func(trans, root, bytenr, num_bytes,
3100                                            parent, ref_root, key.objectid,
3101                                            key.offset, 1);
3102                         if (ret)
3103                                 goto fail;
3104                 } else {
3105                         bytenr = btrfs_node_blockptr(buf, i);
3106                         num_bytes = root->nodesize;
3107                         ret = process_func(trans, root, bytenr, num_bytes,
3108                                            parent, ref_root, level - 1, 0,
3109                                            1);
3110                         if (ret)
3111                                 goto fail;
3112                 }
3113         }
3114         return 0;
3115 fail:
3116         return ret;
3117 }
3118 
3119 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3120                   struct extent_buffer *buf, int full_backref)
3121 {
3122         return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
3123 }
3124 
3125 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3126                   struct extent_buffer *buf, int full_backref)
3127 {
3128         return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
3129 }
3130 
3131 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3132                                  struct btrfs_root *root,
3133                                  struct btrfs_path *path,
3134                                  struct btrfs_block_group_cache *cache)
3135 {
3136         int ret;
3137         struct btrfs_root *extent_root = root->fs_info->extent_root;
3138         unsigned long bi;
3139         struct extent_buffer *leaf;
3140 
3141         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3142         if (ret < 0)
3143                 goto fail;
3144         BUG_ON(ret); /* Corruption */
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         return 0;
3157 
3158 }
3159 
3160 static struct btrfs_block_group_cache *
3161 next_block_group(struct btrfs_root *root,
3162                  struct btrfs_block_group_cache *cache)
3163 {
3164         struct rb_node *node;
3165         spin_lock(&root->fs_info->block_group_cache_lock);
3166         node = rb_next(&cache->cache_node);
3167         btrfs_put_block_group(cache);
3168         if (node) {
3169                 cache = rb_entry(node, struct btrfs_block_group_cache,
3170                                  cache_node);
3171                 btrfs_get_block_group(cache);
3172         } else
3173                 cache = NULL;
3174         spin_unlock(&root->fs_info->block_group_cache_lock);
3175         return cache;
3176 }
3177 
3178 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3179                             struct btrfs_trans_handle *trans,
3180                             struct btrfs_path *path)
3181 {
3182         struct btrfs_root *root = block_group->fs_info->tree_root;
3183         struct inode *inode = NULL;
3184         u64 alloc_hint = 0;
3185         int dcs = BTRFS_DC_ERROR;
3186         int num_pages = 0;
3187         int retries = 0;
3188         int ret = 0;
3189 
3190         /*
3191          * If this block group is smaller than 100 megs don't bother caching the
3192          * block group.
3193          */
3194         if (block_group->key.offset < (100 * 1024 * 1024)) {
3195                 spin_lock(&block_group->lock);
3196                 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3197                 spin_unlock(&block_group->lock);
3198                 return 0;
3199         }
3200 
3201 again:
3202         inode = lookup_free_space_inode(root, block_group, path);
3203         if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3204                 ret = PTR_ERR(inode);
3205                 btrfs_release_path(path);
3206                 goto out;
3207         }
3208 
3209         if (IS_ERR(inode)) {
3210                 BUG_ON(retries);
3211                 retries++;
3212 
3213                 if (block_group->ro)
3214                         goto out_free;
3215 
3216                 ret = create_free_space_inode(root, trans, block_group, path);
3217                 if (ret)
3218                         goto out_free;
3219                 goto again;
3220         }
3221 
3222         /*
3223          * We want to set the generation to 0, that way if anything goes wrong
3224          * from here on out we know not to trust this cache when we load up next
3225          * time.
3226          */
3227         BTRFS_I(inode)->generation = 0;
3228         ret = btrfs_update_inode(trans, root, inode);
3229         WARN_ON(ret);
3230 
3231         /* We've already setup this transaction, go ahead and exit */
3232         if (block_group->cache_generation == trans->transid &&
3233             i_size_read(inode)) {
3234                 dcs = BTRFS_DC_SETUP;
3235                 goto out_put;
3236         }
3237 
3238         if (i_size_read(inode) > 0) {
3239                 ret = btrfs_check_trunc_cache_free_space(root,
3240                                         &root->fs_info->global_block_rsv);
3241                 if (ret)
3242                         goto out_put;
3243 
3244                 ret = btrfs_truncate_free_space_cache(root, trans, inode);
3245                 if (ret)
3246                         goto out_put;
3247         }
3248 
3249         spin_lock(&block_group->lock);
3250         if (block_group->cached != BTRFS_CACHE_FINISHED ||
3251             !btrfs_test_opt(root, SPACE_CACHE) ||
3252             block_group->delalloc_bytes) {
3253                 /*
3254                  * don't bother trying to write stuff out _if_
3255                  * a) we're not cached,
3256                  * b) we're with nospace_cache mount option.
3257                  */
3258                 dcs = BTRFS_DC_WRITTEN;
3259                 spin_unlock(&block_group->lock);
3260                 goto out_put;
3261         }
3262         spin_unlock(&block_group->lock);
3263 
3264         /*
3265          * Try to preallocate enough space based on how big the block group is.
3266          * Keep in mind this has to include any pinned space which could end up
3267          * taking up quite a bit since it's not folded into the other space
3268          * cache.
3269          */
3270         num_pages = (int)div64_u64(block_group->key.offset, 256 * 1024 * 1024);
3271         if (!num_pages)
3272                 num_pages = 1;
3273 
3274         num_pages *= 16;
3275         num_pages *= PAGE_CACHE_SIZE;
3276 
3277         ret = btrfs_check_data_free_space(inode, num_pages);
3278         if (ret)
3279                 goto out_put;
3280 
3281         ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3282                                               num_pages, num_pages,
3283                                               &alloc_hint);
3284         if (!ret)
3285                 dcs = BTRFS_DC_SETUP;
3286         btrfs_free_reserved_data_space(inode, num_pages);
3287 
3288 out_put:
3289         iput(inode);
3290 out_free:
3291         btrfs_release_path(path);
3292 out:
3293         spin_lock(&block_group->lock);
3294         if (!ret && dcs == BTRFS_DC_SETUP)
3295                 block_group->cache_generation = trans->transid;
3296         block_group->disk_cache_state = dcs;
3297         spin_unlock(&block_group->lock);
3298 
3299         return ret;
3300 }
3301 
3302 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3303                                    struct btrfs_root *root)
3304 {
3305         struct btrfs_block_group_cache *cache;
3306         int err = 0;
3307         struct btrfs_path *path;
3308         u64 last = 0;
3309 
3310         path = btrfs_alloc_path();
3311         if (!path)
3312                 return -ENOMEM;
3313 
3314 again:
3315         while (1) {
3316                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3317                 while (cache) {
3318                         if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3319                                 break;
3320                         cache = next_block_group(root, cache);
3321                 }
3322                 if (!cache) {
3323                         if (last == 0)
3324                                 break;
3325                         last = 0;
3326                         continue;
3327                 }
3328                 err = cache_save_setup(cache, trans, path);
3329                 last = cache->key.objectid + cache->key.offset;
3330                 btrfs_put_block_group(cache);
3331         }
3332 
3333         while (1) {
3334                 if (last == 0) {
3335                         err = btrfs_run_delayed_refs(trans, root,
3336                                                      (unsigned long)-1);
3337                         if (err) /* File system offline */
3338                                 goto out;
3339                 }
3340 
3341                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3342                 while (cache) {
3343                         if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
3344                                 btrfs_put_block_group(cache);
3345                                 goto again;
3346                         }
3347 
3348                         if (cache->dirty)
3349                                 break;
3350                         cache = next_block_group(root, cache);
3351                 }
3352                 if (!cache) {
3353                         if (last == 0)
3354                                 break;
3355                         last = 0;
3356                         continue;
3357                 }
3358 
3359                 if (cache->disk_cache_state == BTRFS_DC_SETUP)
3360                         cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
3361                 cache->dirty = 0;
3362                 last = cache->key.objectid + cache->key.offset;
3363 
3364                 err = write_one_cache_group(trans, root, path, cache);
3365                 btrfs_put_block_group(cache);
3366                 if (err) /* File system offline */
3367                         goto out;
3368         }
3369 
3370         while (1) {
3371                 /*
3372                  * I don't think this is needed since we're just marking our
3373                  * preallocated extent as written, but just in case it can't
3374                  * hurt.
3375                  */
3376                 if (last == 0) {
3377                         err = btrfs_run_delayed_refs(trans, root,
3378                                                      (unsigned long)-1);
3379                         if (err) /* File system offline */
3380                                 goto out;
3381                 }
3382 
3383                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3384                 while (cache) {
3385                         /*
3386                          * Really this shouldn't happen, but it could if we
3387                          * couldn't write the entire preallocated extent and
3388                          * splitting the extent resulted in a new block.
3389                          */
3390                         if (cache->dirty) {
3391                                 btrfs_put_block_group(cache);
3392                                 goto again;
3393                         }
3394                         if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3395                                 break;
3396                         cache = next_block_group(root, cache);
3397                 }
3398                 if (!cache) {
3399                         if (last == 0)
3400                                 break;
3401                         last = 0;
3402                         continue;
3403                 }
3404 
3405                 err = btrfs_write_out_cache(root, trans, cache, path);
3406 
3407                 /*
3408                  * If we didn't have an error then the cache state is still
3409                  * NEED_WRITE, so we can set it to WRITTEN.
3410                  */
3411                 if (!err && cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3412                         cache->disk_cache_state = BTRFS_DC_WRITTEN;
3413                 last = cache->key.objectid + cache->key.offset;
3414                 btrfs_put_block_group(cache);
3415         }
3416 out:
3417 
3418         btrfs_free_path(path);
3419         return err;
3420 }
3421 
3422 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3423 {
3424         struct btrfs_block_group_cache *block_group;
3425         int readonly = 0;
3426 
3427         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3428         if (!block_group || block_group->ro)
3429                 readonly = 1;
3430         if (block_group)
3431                 btrfs_put_block_group(block_group);
3432         return readonly;
3433 }
3434 
3435 static const char *alloc_name(u64 flags)
3436 {
3437         switch (flags) {
3438         case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
3439                 return "mixed";
3440         case BTRFS_BLOCK_GROUP_METADATA:
3441                 return "metadata";
3442         case BTRFS_BLOCK_GROUP_DATA:
3443                 return "data";
3444         case BTRFS_BLOCK_GROUP_SYSTEM:
3445                 return "system";
3446         default:
3447                 WARN_ON(1);
3448                 return "invalid-combination";
3449         };
3450 }
3451 
3452 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3453                              u64 total_bytes, u64 bytes_used,
3454                              struct btrfs_space_info **space_info)
3455 {
3456         struct btrfs_space_info *found;
3457         int i;
3458         int factor;
3459         int ret;
3460 
3461         if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3462                      BTRFS_BLOCK_GROUP_RAID10))
3463                 factor = 2;
3464         else
3465                 factor = 1;
3466 
3467         found = __find_space_info(info, flags);
3468         if (found) {
3469                 spin_lock(&found->lock);
3470                 found->total_bytes += total_bytes;
3471                 found->disk_total += total_bytes * factor;
3472                 found->bytes_used += bytes_used;
3473                 found->disk_used += bytes_used * factor;
3474                 found->full = 0;
3475                 spin_unlock(&found->lock);
3476                 *space_info = found;
3477                 return 0;
3478         }
3479         found = kzalloc(sizeof(*found), GFP_NOFS);
3480         if (!found)
3481                 return -ENOMEM;
3482 
3483         ret = percpu_counter_init(&found->total_bytes_pinned, 0, GFP_KERNEL);
3484         if (ret) {
3485                 kfree(found);
3486                 return ret;
3487         }
3488 
3489         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3490                 INIT_LIST_HEAD(&found->block_groups[i]);
3491         init_rwsem(&found->groups_sem);
3492         spin_lock_init(&found->lock);
3493         found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3494         found->total_bytes = total_bytes;
3495         found->disk_total = total_bytes * factor;
3496         found->bytes_used = bytes_used;
3497         found->disk_used = bytes_used * factor;
3498         found->bytes_pinned = 0;
3499         found->bytes_reserved = 0;
3500         found->bytes_readonly = 0;
3501         found->bytes_may_use = 0;
3502         found->full = 0;
3503         found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3504         found->chunk_alloc = 0;
3505         found->flush = 0;
3506         init_waitqueue_head(&found->wait);
3507 
3508         ret = kobject_init_and_add(&found->kobj, &space_info_ktype,
3509                                     info->space_info_kobj, "%s",
3510                                     alloc_name(found->flags));
3511         if (ret) {
3512                 percpu_counter_destroy(&found->total_bytes_pinned);
3513                 kfree(found);
3514                 return ret;
3515         }
3516 
3517         *space_info = found;
3518         list_add_rcu(&found->list, &info->space_info);
3519         if (flags & BTRFS_BLOCK_GROUP_DATA)
3520                 info->data_sinfo = found;
3521 
3522         return ret;
3523 }
3524 
3525 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3526 {
3527         u64 extra_flags = chunk_to_extended(flags) &
3528                                 BTRFS_EXTENDED_PROFILE_MASK;
3529 
3530         write_seqlock(&fs_info->profiles_lock);
3531         if (flags & BTRFS_BLOCK_GROUP_DATA)
3532                 fs_info->avail_data_alloc_bits |= extra_flags;
3533         if (flags & BTRFS_BLOCK_GROUP_METADATA)
3534                 fs_info->avail_metadata_alloc_bits |= extra_flags;
3535         if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3536                 fs_info->avail_system_alloc_bits |= extra_flags;
3537         write_sequnlock(&fs_info->profiles_lock);
3538 }
3539 
3540 /*
3541  * returns target flags in extended format or 0 if restripe for this
3542  * chunk_type is not in progress
3543  *
3544  * should be called with either volume_mutex or balance_lock held
3545  */
3546 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3547 {
3548         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3549         u64 target = 0;
3550 
3551         if (!bctl)
3552                 return 0;
3553 
3554         if (flags & BTRFS_BLOCK_GROUP_DATA &&
3555             bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3556                 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3557         } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3558                    bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3559                 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3560         } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3561                    bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3562                 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3563         }
3564 
3565         return target;
3566 }
3567 
3568 /*
3569  * @flags: available profiles in extended format (see ctree.h)
3570  *
3571  * Returns reduced profile in chunk format.  If profile changing is in
3572  * progress (either running or paused) picks the target profile (if it's
3573  * already available), otherwise falls back to plain reducing.
3574  */
3575 static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3576 {
3577         u64 num_devices = root->fs_info->fs_devices->rw_devices;
3578         u64 target;
3579         u64 tmp;
3580 
3581         /*
3582          * see if restripe for this chunk_type is in progress, if so
3583          * try to reduce to the target profile
3584          */
3585         spin_lock(&root->fs_info->balance_lock);
3586         target = get_restripe_target(root->fs_info, flags);
3587         if (target) {
3588                 /* pick target profile only if it's already available */
3589                 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3590                         spin_unlock(&root->fs_info->balance_lock);
3591                         return extended_to_chunk(target);
3592                 }
3593         }
3594         spin_unlock(&root->fs_info->balance_lock);
3595 
3596         /* First, mask out the RAID levels which aren't possible */
3597         if (num_devices == 1)
3598                 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0 |
3599                            BTRFS_BLOCK_GROUP_RAID5);
3600         if (num_devices < 3)
3601                 flags &= ~BTRFS_BLOCK_GROUP_RAID6;
3602         if (num_devices < 4)
3603                 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3604 
3605         tmp = flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3606                        BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID5 |
3607                        BTRFS_BLOCK_GROUP_RAID6 | BTRFS_BLOCK_GROUP_RAID10);
3608         flags &= ~tmp;
3609 
3610         if (tmp & BTRFS_BLOCK_GROUP_RAID6)
3611                 tmp = BTRFS_BLOCK_GROUP_RAID6;
3612         else if (tmp & BTRFS_BLOCK_GROUP_RAID5)
3613                 tmp = BTRFS_BLOCK_GROUP_RAID5;
3614         else if (tmp & BTRFS_BLOCK_GROUP_RAID10)
3615                 tmp = BTRFS_BLOCK_GROUP_RAID10;
3616         else if (tmp & BTRFS_BLOCK_GROUP_RAID1)
3617                 tmp = BTRFS_BLOCK_GROUP_RAID1;
3618         else if (tmp & BTRFS_BLOCK_GROUP_RAID0)
3619                 tmp = BTRFS_BLOCK_GROUP_RAID0;
3620 
3621         return extended_to_chunk(flags | tmp);
3622 }
3623 
3624 static u64 get_alloc_profile(struct btrfs_root *root, u64 orig_flags)
3625 {
3626         unsigned seq;
3627         u64 flags;
3628 
3629         do {
3630                 flags = orig_flags;
3631                 seq = read_seqbegin(&root->fs_info->profiles_lock);
3632 
3633                 if (flags & BTRFS_BLOCK_GROUP_DATA)
3634                         flags |= root->fs_info->avail_data_alloc_bits;
3635                 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3636                         flags |= root->fs_info->avail_system_alloc_bits;
3637                 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3638                         flags |= root->fs_info->avail_metadata_alloc_bits;
3639         } while (read_seqretry(&root->fs_info->profiles_lock, seq));
3640 
3641         return btrfs_reduce_alloc_profile(root, flags);
3642 }
3643 
3644 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3645 {
3646         u64 flags;
3647         u64 ret;
3648 
3649         if (data)
3650                 flags = BTRFS_BLOCK_GROUP_DATA;
3651         else if (root == root->fs_info->chunk_root)
3652                 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3653         else
3654                 flags = BTRFS_BLOCK_GROUP_METADATA;
3655 
3656         ret = get_alloc_profile(root, flags);
3657         return ret;
3658 }
3659 
3660 /*
3661  * This will check the space that the inode allocates from to make sure we have
3662  * enough space for bytes.
3663  */
3664 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3665 {
3666         struct btrfs_space_info *data_sinfo;
3667         struct btrfs_root *root = BTRFS_I(inode)->root;
3668         struct btrfs_fs_info *fs_info = root->fs_info;
3669         u64 used;
3670         int ret = 0, committed = 0, alloc_chunk = 1;
3671 
3672         /* make sure bytes are sectorsize aligned */
3673         bytes = ALIGN(bytes, root->sectorsize);
3674 
3675         if (btrfs_is_free_space_inode(inode)) {
3676                 committed = 1;
3677                 ASSERT(current->journal_info);
3678         }
3679 
3680         data_sinfo = fs_info->data_sinfo;
3681         if (!data_sinfo)
3682                 goto alloc;
3683 
3684 again:
3685         /* make sure we have enough space to handle the data first */
3686         spin_lock(&data_sinfo->lock);
3687         used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3688                 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3689                 data_sinfo->bytes_may_use;
3690 
3691         if (used + bytes > data_sinfo->total_bytes) {
3692                 struct btrfs_trans_handle *trans;
3693 
3694                 /*
3695                  * if we don't have enough free bytes in this space then we need
3696                  * to alloc a new chunk.
3697                  */
3698                 if (!data_sinfo->full && alloc_chunk) {
3699                         u64 alloc_target;
3700 
3701                         data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3702                         spin_unlock(&data_sinfo->lock);
3703 alloc:
3704                         alloc_target = btrfs_get_alloc_profile(root, 1);
3705                         /*
3706                          * It is ugly that we don't call nolock join
3707                          * transaction for the free space inode case here.
3708                          * But it is safe because we only do the data space
3709                          * reservation for the free space cache in the
3710                          * transaction context, the common join transaction
3711                          * just increase the counter of the current transaction
3712                          * handler, doesn't try to acquire the trans_lock of
3713                          * the fs.
3714                          */
3715                         trans = btrfs_join_transaction(root);
3716                         if (IS_ERR(trans))
3717                                 return PTR_ERR(trans);
3718 
3719                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3720                                              alloc_target,
3721                                              CHUNK_ALLOC_NO_FORCE);
3722                         btrfs_end_transaction(trans, root);
3723                         if (ret < 0) {
3724                                 if (ret != -ENOSPC)
3725                                         return ret;
3726                                 else
3727                                         goto commit_trans;
3728                         }
3729 
3730                         if (!data_sinfo)
3731                                 data_sinfo = fs_info->data_sinfo;
3732 
3733                         goto again;
3734                 }
3735 
3736                 /*
3737                  * If we don't have enough pinned space to deal with this
3738                  * allocation don't bother committing the transaction.
3739                  */
3740                 if (percpu_counter_compare(&data_sinfo->total_bytes_pinned,
3741                                            bytes) < 0)
3742                         committed = 1;
3743                 spin_unlock(&data_sinfo->lock);
3744 
3745                 /* commit the current transaction and try again */
3746 commit_trans:
3747                 if (!committed &&
3748                     !atomic_read(&root->fs_info->open_ioctl_trans)) {
3749                         committed = 1;
3750 
3751                         trans = btrfs_join_transaction(root);
3752                         if (IS_ERR(trans))
3753                                 return PTR_ERR(trans);
3754                         ret = btrfs_commit_transaction(trans, root);
3755                         if (ret)
3756                                 return ret;
3757                         goto again;
3758                 }
3759 
3760                 trace_btrfs_space_reservation(root->fs_info,
3761                                               "space_info:enospc",
3762                                               data_sinfo->flags, bytes, 1);
3763                 return -ENOSPC;
3764         }
3765         data_sinfo->bytes_may_use += bytes;
3766         trace_btrfs_space_reservation(root->fs_info, "space_info",
3767                                       data_sinfo->flags, bytes, 1);
3768         spin_unlock(&data_sinfo->lock);
3769 
3770         return 0;
3771 }
3772 
3773 /*
3774  * Called if we need to clear a data reservation for this inode.
3775  */
3776 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3777 {
3778         struct btrfs_root *root = BTRFS_I(inode)->root;
3779         struct btrfs_space_info *data_sinfo;
3780 
3781         /* make sure bytes are sectorsize aligned */
3782         bytes = ALIGN(bytes, root->sectorsize);
3783 
3784         data_sinfo = root->fs_info->data_sinfo;
3785         spin_lock(&data_sinfo->lock);
3786         WARN_ON(data_sinfo->bytes_may_use < bytes);
3787         data_sinfo->bytes_may_use -= bytes;
3788         trace_btrfs_space_reservation(root->fs_info, "space_info",
3789                                       data_sinfo->flags, bytes, 0);
3790         spin_unlock(&data_sinfo->lock);
3791 }
3792 
3793 static void force_metadata_allocation(struct btrfs_fs_info *info)
3794 {
3795         struct list_head *head = &info->space_info;
3796         struct btrfs_space_info *found;
3797 
3798         rcu_read_lock();
3799         list_for_each_entry_rcu(found, head, list) {
3800                 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3801                         found->force_alloc = CHUNK_ALLOC_FORCE;
3802         }
3803         rcu_read_unlock();
3804 }
3805 
3806 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
3807 {
3808         return (global->size << 1);
3809 }
3810 
3811 static int should_alloc_chunk(struct btrfs_root *root,
3812                               struct btrfs_space_info *sinfo, int force)
3813 {
3814         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3815         u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3816         u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3817         u64 thresh;
3818 
3819         if (force == CHUNK_ALLOC_FORCE)
3820                 return 1;
3821 
3822         /*
3823          * We need to take into account the global rsv because for all intents
3824          * and purposes it's used space.  Don't worry about locking the
3825          * global_rsv, it doesn't change except when the transaction commits.
3826          */
3827         if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
3828                 num_allocated += calc_global_rsv_need_space(global_rsv);
3829 
3830         /*
3831          * in limited mode, we want to have some free space up to
3832          * about 1% of the FS size.
3833          */
3834         if (force == CHUNK_ALLOC_LIMITED) {
3835                 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3836                 thresh = max_t(u64, 64 * 1024 * 1024,
3837                                div_factor_fine(thresh, 1));
3838 
3839                 if (num_bytes - num_allocated < thresh)
3840                         return 1;
3841         }
3842 
3843         if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8))
3844                 return 0;
3845         return 1;
3846 }
3847 
3848 static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type)
3849 {
3850         u64 num_dev;
3851 
3852         if (type & (BTRFS_BLOCK_GROUP_RAID10 |
3853                     BTRFS_BLOCK_GROUP_RAID0 |
3854                     BTRFS_BLOCK_GROUP_RAID5 |
3855                     BTRFS_BLOCK_GROUP_RAID6))
3856                 num_dev = root->fs_info->fs_devices->rw_devices;
3857         else if (type & BTRFS_BLOCK_GROUP_RAID1)
3858                 num_dev = 2;
3859         else
3860                 num_dev = 1;    /* DUP or single */
3861 
3862         /* metadata for updaing devices and chunk tree */
3863         return btrfs_calc_trans_metadata_size(root, num_dev + 1);
3864 }
3865 
3866 static void check_system_chunk(struct btrfs_trans_handle *trans,
3867                                struct btrfs_root *root, u64 type)
3868 {
3869         struct btrfs_space_info *info;
3870         u64 left;
3871         u64 thresh;
3872 
3873         info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3874         spin_lock(&info->lock);
3875         left = info->total_bytes - info->bytes_used - info->bytes_pinned -
3876                 info->bytes_reserved - info->bytes_readonly;
3877         spin_unlock(&info->lock);
3878 
3879         thresh = get_system_chunk_thresh(root, type);
3880         if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
3881                 btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu",
3882                         left, thresh, type);
3883                 dump_space_info(info, 0, 0);
3884         }
3885 
3886         if (left < thresh) {
3887                 u64 flags;
3888 
3889                 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
3890                 btrfs_alloc_chunk(trans, root, flags);
3891         }
3892 }
3893 
3894 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3895                           struct btrfs_root *extent_root, u64 flags, int force)
3896 {
3897         struct btrfs_space_info *space_info;
3898         struct btrfs_fs_info *fs_info = extent_root->fs_info;
3899         int wait_for_alloc = 0;
3900         int ret = 0;
3901 
3902         /* Don't re-enter if we're already allocating a chunk */
3903         if (trans->allocating_chunk)
3904                 return -ENOSPC;
3905 
3906         space_info = __find_space_info(extent_root->fs_info, flags);
3907         if (!space_info) {
3908                 ret = update_space_info(extent_root->fs_info, flags,
3909                                         0, 0, &space_info);
3910                 BUG_ON(ret); /* -ENOMEM */
3911         }
3912         BUG_ON(!space_info); /* Logic error */
3913 
3914 again:
3915         spin_lock(&space_info->lock);
3916         if (force < space_info->force_alloc)
3917                 force = space_info->force_alloc;
3918         if (space_info->full) {
3919                 if (should_alloc_chunk(extent_root, space_info, force))
3920                         ret = -ENOSPC;
3921                 else
3922                         ret = 0;
3923                 spin_unlock(&space_info->lock);
3924                 return ret;
3925         }
3926 
3927         if (!should_alloc_chunk(extent_root, space_info, force)) {
3928                 spin_unlock(&space_info->lock);
3929                 return 0;
3930         } else if (space_info->chunk_alloc) {
3931                 wait_for_alloc = 1;
3932         } else {
3933                 space_info->chunk_alloc = 1;
3934         }
3935 
3936         spin_unlock(&space_info->lock);
3937 
3938         mutex_lock(&fs_info->chunk_mutex);
3939 
3940         /*
3941          * The chunk_mutex is held throughout the entirety of a chunk
3942          * allocation, so once we've acquired the chunk_mutex we know that the
3943          * other guy is done and we need to recheck and see if we should
3944          * allocate.
3945          */
3946         if (wait_for_alloc) {
3947                 mutex_unlock(&fs_info->chunk_mutex);
3948                 wait_for_alloc = 0;
3949                 cond_resched();
3950                 goto again;
3951         }
3952 
3953         trans->allocating_chunk = true;
3954 
3955         /*
3956          * If we have mixed data/metadata chunks we want to make sure we keep
3957          * allocating mixed chunks instead of individual chunks.
3958          */
3959         if (btrfs_mixed_space_info(space_info))
3960                 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3961 
3962         /*
3963          * if we're doing a data chunk, go ahead and make sure that
3964          * we keep a reasonable number of metadata chunks allocated in the
3965          * FS as well.
3966          */
3967         if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3968                 fs_info->data_chunk_allocations++;
3969                 if (!(fs_info->data_chunk_allocations %
3970                       fs_info->metadata_ratio))
3971                         force_metadata_allocation(fs_info);
3972         }
3973 
3974         /*
3975          * Check if we have enough space in SYSTEM chunk because we may need
3976          * to update devices.
3977          */
3978         check_system_chunk(trans, extent_root, flags);
3979 
3980         ret = btrfs_alloc_chunk(trans, extent_root, flags);
3981         trans->allocating_chunk = false;
3982 
3983         spin_lock(&space_info->lock);
3984         if (ret < 0 && ret != -ENOSPC)
3985                 goto out;
3986         if (ret)
3987                 space_info->full = 1;
3988         else
3989                 ret = 1;
3990 
3991         space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3992 out:
3993         space_info->chunk_alloc = 0;
3994         spin_unlock(&space_info->lock);
3995         mutex_unlock(&fs_info->chunk_mutex);
3996         return ret;
3997 }
3998 
3999 static int can_overcommit(struct btrfs_root *root,
4000                           struct btrfs_space_info *space_info, u64 bytes,
4001                           enum btrfs_reserve_flush_enum flush)
4002 {
4003         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4004         u64 profile = btrfs_get_alloc_profile(root, 0);
4005         u64 space_size;
4006         u64 avail;
4007         u64 used;
4008 
4009         used = space_info->bytes_used + space_info->bytes_reserved +
4010                 space_info->bytes_pinned + space_info->bytes_readonly;
4011 
4012         /*
4013          * We only want to allow over committing if we have lots of actual space
4014          * free, but if we don't have enough space to handle the global reserve
4015          * space then we could end up having a real enospc problem when trying
4016          * to allocate a chunk or some other such important allocation.
4017          */
4018         spin_lock(&global_rsv->lock);
4019         space_size = calc_global_rsv_need_space(global_rsv);
4020         spin_unlock(&global_rsv->lock);
4021         if (used + space_size >= space_info->total_bytes)
4022                 return 0;
4023 
4024         used += space_info->bytes_may_use;
4025 
4026         spin_lock(&root->fs_info->free_chunk_lock);
4027         avail = root->fs_info->free_chunk_space;
4028         spin_unlock(&root->fs_info->free_chunk_lock);
4029 
4030         /*
4031          * If we have dup, raid1 or raid10 then only half of the free
4032          * space is actually useable.  For raid56, the space info used
4033          * doesn't include the parity drive, so we don't have to
4034          * change the math
4035          */
4036         if (profile & (BTRFS_BLOCK_GROUP_DUP |
4037                        BTRFS_BLOCK_GROUP_RAID1 |
4038                        BTRFS_BLOCK_GROUP_RAID10))
4039                 avail >>= 1;
4040 
4041         /*
4042          * If we aren't flushing all things, let us overcommit up to
4043          * 1/2th of the space. If we can flush, don't let us overcommit
4044          * too much, let it overcommit up to 1/8 of the space.
4045          */
4046         if (flush == BTRFS_RESERVE_FLUSH_ALL)
4047                 avail >>= 3;
4048         else
4049                 avail >>= 1;
4050 
4051         if (used + bytes < space_info->total_bytes + avail)
4052                 return 1;
4053         return 0;
4054 }
4055 
4056 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
4057                                          unsigned long nr_pages, int nr_items)
4058 {
4059         struct super_block *sb = root->fs_info->sb;
4060 
4061         if (down_read_trylock(&sb->s_umount)) {
4062                 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4063                 up_read(&sb->s_umount);
4064         } else {
4065                 /*
4066                  * We needn't worry the filesystem going from r/w to r/o though
4067                  * we don't acquire ->s_umount mutex, because the filesystem
4068                  * should guarantee the delalloc inodes list be empty after
4069                  * the filesystem is readonly(all dirty pages are written to
4070                  * the disk).
4071                  */
4072                 btrfs_start_delalloc_roots(root->fs_info, 0, nr_items);
4073                 if (!current->journal_info)
4074                         btrfs_wait_ordered_roots(root->fs_info, nr_items);
4075         }
4076 }
4077 
4078 static inline int calc_reclaim_items_nr(struct btrfs_root *root, u64 to_reclaim)
4079 {
4080         u64 bytes;
4081         int nr;
4082 
4083         bytes = btrfs_calc_trans_metadata_size(root, 1);
4084         nr = (int)div64_u64(to_reclaim, bytes);
4085         if (!nr)
4086                 nr = 1;
4087         return nr;
4088 }
4089 
4090 #define EXTENT_SIZE_PER_ITEM    (256 * 1024)
4091 
4092 /*
4093  * shrink metadata reservation for delalloc
4094  */
4095 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
4096                             bool wait_ordered)
4097 {
4098         struct btrfs_block_rsv *block_rsv;
4099         struct btrfs_space_info *space_info;
4100         struct btrfs_trans_handle *trans;
4101         u64 delalloc_bytes;
4102         u64 max_reclaim;
4103         long time_left;
4104         unsigned long nr_pages;
4105         int loops;
4106         int items;
4107         enum btrfs_reserve_flush_enum flush;
4108 
4109         /* Calc the number of the pages we need flush for space reservation */
4110         items = calc_reclaim_items_nr(root, to_reclaim);
4111         to_reclaim = items * EXTENT_SIZE_PER_ITEM;
4112 
4113         trans = (struct btrfs_trans_handle *)current->journal_info;
4114         block_rsv = &root->fs_info->delalloc_block_rsv;
4115         space_info = block_rsv->space_info;
4116 
4117         delalloc_bytes = percpu_counter_sum_positive(
4118                                                 &root->fs_info->delalloc_bytes);
4119         if (delalloc_bytes == 0) {
4120                 if (trans)
4121                         return;
4122                 if (wait_ordered)
4123                         btrfs_wait_ordered_roots(root->fs_info, items);
4124                 return;
4125         }
4126 
4127         loops = 0;
4128         while (delalloc_bytes && loops < 3) {
4129                 max_reclaim = min(delalloc_bytes, to_reclaim);
4130                 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
4131                 btrfs_writeback_inodes_sb_nr(root, nr_pages, items);
4132                 /*
4133                  * We need to wait for the async pages to actually start before
4134                  * we do anything.
4135                  */
4136                 max_reclaim = atomic_read(&root->fs_info->async_delalloc_pages);
4137                 if (!max_reclaim)
4138                         goto skip_async;
4139 
4140                 if (max_reclaim <= nr_pages)
4141                         max_reclaim = 0;
4142                 else
4143                         max_reclaim -= nr_pages;
4144 
4145                 wait_event(root->fs_info->async_submit_wait,
4146                            atomic_read(&root->fs_info->async_delalloc_pages) <=
4147                            (int)max_reclaim);
4148 skip_async:
4149                 if (!trans)
4150                         flush = BTRFS_RESERVE_FLUSH_ALL;
4151                 else
4152                         flush = BTRFS_RESERVE_NO_FLUSH;
4153                 spin_lock(&space_info->lock);
4154                 if (can_overcommit(root, space_info, orig, flush)) {
4155                         spin_unlock(&space_info->lock);
4156                         break;
4157                 }
4158                 spin_unlock(&space_info->lock);
4159 
4160                 loops++;
4161                 if (wait_ordered && !trans) {
4162                         btrfs_wait_ordered_roots(root->fs_info, items);
4163                 } else {
4164                         time_left = schedule_timeout_killable(1);
4165                         if (time_left)
4166                                 break;
4167                 }
4168                 delalloc_bytes = percpu_counter_sum_positive(
4169                                                 &root->fs_info->delalloc_bytes);
4170         }
4171 }
4172 
4173 /**
4174  * maybe_commit_transaction - possibly commit the transaction if its ok to
4175  * @root - the root we're allocating for
4176  * @bytes - the number of bytes we want to reserve
4177  * @force - force the commit
4178  *
4179  * This will check to make sure that committing the transaction will actually
4180  * get us somewhere and then commit the transaction if it does.  Otherwise it
4181  * will return -ENOSPC.
4182  */
4183 static int may_commit_transaction(struct btrfs_root *root,
4184                                   struct btrfs_space_info *space_info,
4185                                   u64 bytes, int force)
4186 {
4187         struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
4188         struct btrfs_trans_handle *trans;
4189 
4190         trans = (struct btrfs_trans_handle *)current->journal_info;
4191         if (trans)
4192                 return -EAGAIN;
4193 
4194         if (force)
4195                 goto commit;
4196 
4197         /* See if there is enough pinned space to make this reservation */
4198         if (percpu_counter_compare(&space_info->total_bytes_pinned,
4199                                    bytes) >= 0)
4200                 goto commit;
4201 
4202         /*
4203          * See if there is some space in the delayed insertion reservation for
4204          * this reservation.
4205          */
4206         if (space_info != delayed_rsv->space_info)
4207                 return -ENOSPC;
4208 
4209         spin_lock(&delayed_rsv->lock);
4210         if (percpu_counter_compare(&space_info->total_bytes_pinned,
4211                                    bytes - delayed_rsv->size) >= 0) {
4212                 spin_unlock(&delayed_rsv->lock);
4213                 return -ENOSPC;
4214         }
4215         spin_unlock(&delayed_rsv->lock);
4216 
4217 commit:
4218         trans = btrfs_join_transaction(root);
4219         if (IS_ERR(trans))
4220                 return -ENOSPC;
4221 
4222         return btrfs_commit_transaction(trans, root);
4223 }
4224 
4225 enum flush_state {
4226         FLUSH_DELAYED_ITEMS_NR  =       1,
4227         FLUSH_DELAYED_ITEMS     =       2,
4228         FLUSH_DELALLOC          =       3,
4229         FLUSH_DELALLOC_WAIT     =       4,
4230         ALLOC_CHUNK             =       5,
4231         COMMIT_TRANS            =       6,
4232 };
4233 
4234 static int flush_space(struct btrfs_root *root,
4235                        struct btrfs_space_info *space_info, u64 num_bytes,
4236                        u64 orig_bytes, int state)
4237 {
4238         struct btrfs_trans_handle *trans;
4239         int nr;
4240         int ret = 0;
4241 
4242         switch (state) {
4243         case FLUSH_DELAYED_ITEMS_NR:
4244         case FLUSH_DELAYED_ITEMS:
4245                 if (state == FLUSH_DELAYED_ITEMS_NR)
4246                         nr = calc_reclaim_items_nr(root, num_bytes) * 2;
4247                 else
4248                         nr = -1;
4249 
4250                 trans = btrfs_join_transaction(root);
4251                 if (IS_ERR(trans)) {
4252                         ret = PTR_ERR(trans);
4253                         break;
4254                 }
4255                 ret = btrfs_run_delayed_items_nr(trans, root, nr);
4256                 btrfs_end_transaction(trans, root);
4257                 break;
4258         case FLUSH_DELALLOC:
4259         case FLUSH_DELALLOC_WAIT:
4260                 shrink_delalloc(root, num_bytes * 2, orig_bytes,
4261                                 state == FLUSH_DELALLOC_WAIT);
4262                 break;
4263         case ALLOC_CHUNK:
4264                 trans = btrfs_join_transaction(root);
4265                 if (IS_ERR(trans)) {
4266                         ret = PTR_ERR(trans);
4267                         break;
4268                 }
4269                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4270                                      btrfs_get_alloc_profile(root, 0),
4271                                      CHUNK_ALLOC_NO_FORCE);
4272                 btrfs_end_transaction(trans, root);
4273                 if (ret == -ENOSPC)
4274                         ret = 0;
4275                 break;
4276         case COMMIT_TRANS:
4277                 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
4278                 break;
4279         default:
4280                 ret = -ENOSPC;
4281                 break;
4282         }
4283 
4284         return 0;
4285 }
4286 
4287 static inline u64
4288 btrfs_calc_reclaim_metadata_size(struct btrfs_root *root,
4289                                  struct btrfs_space_info *space_info)
4290 {
4291         u64 used;
4292         u64 expected;
4293         u64 to_reclaim;
4294 
4295         to_reclaim = min_t(u64, num_online_cpus() * 1024 * 1024,
4296                                 16 * 1024 * 1024);
4297         spin_lock(&space_info->lock);
4298         if (can_overcommit(root, space_info, to_reclaim,
4299                            BTRFS_RESERVE_FLUSH_ALL)) {
4300                 to_reclaim = 0;
4301                 goto out;
4302         }
4303 
4304         used = space_info->bytes_used + space_info->bytes_reserved +
4305                space_info->bytes_pinned + space_info->bytes_readonly +
4306                space_info->bytes_may_use;
4307         if (can_overcommit(root, space_info, 1024 * 1024,
4308                            BTRFS_RESERVE_FLUSH_ALL))
4309                 expected = div_factor_fine(space_info->total_bytes, 95);
4310         else
4311                 expected = div_factor_fine(space_info->total_bytes, 90);
4312 
4313         if (used > expected)
4314                 to_reclaim = used - expected;
4315         else
4316                 to_reclaim = 0;
4317         to_reclaim = min(to_reclaim, space_info->bytes_may_use +
4318                                      space_info->bytes_reserved);
4319 out:
4320         spin_unlock(&space_info->lock);
4321 
4322         return to_reclaim;
4323 }
4324 
4325 static inline int need_do_async_reclaim(struct btrfs_space_info *space_info,
4326                                         struct btrfs_fs_info *fs_info, u64 used)
4327 {
4328         return (used >= div_factor_fine(space_info->total_bytes, 98) &&
4329                 !btrfs_fs_closing(fs_info) &&
4330                 !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state));
4331 }
4332 
4333 static int btrfs_need_do_async_reclaim(struct btrfs_space_info *space_info,
4334                                        struct btrfs_fs_info *fs_info,
4335                                        int flush_state)
4336 {
4337         u64 used;
4338 
4339         spin_lock(&space_info->lock);
4340         /*
4341          * We run out of space and have not got any free space via flush_space,
4342          * so don't bother doing async reclaim.
4343          */
4344         if (flush_state > COMMIT_TRANS && space_info->full) {
4345                 spin_unlock(&space_info->lock);
4346                 return 0;
4347         }
4348 
4349         used = space_info->bytes_used + space_info->bytes_reserved +
4350                space_info->bytes_pinned + space_info->bytes_readonly +
4351                space_info->bytes_may_use;
4352         if (need_do_async_reclaim(space_info, fs_info, used)) {
4353                 spin_unlock(&space_info->lock);
4354                 return 1;
4355         }
4356         spin_unlock(&space_info->lock);
4357 
4358         return 0;
4359 }
4360 
4361 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
4362 {
4363         struct btrfs_fs_info *fs_info;
4364         struct btrfs_space_info *space_info;
4365         u64 to_reclaim;
4366         int flush_state;
4367 
4368         fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
4369         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4370 
4371         to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root,
4372                                                       space_info);
4373         if (!to_reclaim)
4374                 return;
4375 
4376         flush_state = FLUSH_DELAYED_ITEMS_NR;
4377         do {
4378                 flush_space(fs_info->fs_root, space_info, to_reclaim,
4379                             to_reclaim, flush_state);
4380                 flush_state++;
4381                 if (!btrfs_need_do_async_reclaim(space_info, fs_info,
4382                                                  flush_state))
4383                         return;
4384         } while (flush_state <= COMMIT_TRANS);
4385 
4386         if (btrfs_need_do_async_reclaim(space_info, fs_info, flush_state))
4387                 queue_work(system_unbound_wq, work);
4388 }
4389 
4390 void btrfs_init_async_reclaim_work(struct work_struct *work)
4391 {
4392         INIT_WORK(work, btrfs_async_reclaim_metadata_space);
4393 }
4394 
4395 /**
4396  * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4397  * @root - the root we're allocating for
4398  * @block_rsv - the block_rsv we're allocating for
4399  * @orig_bytes - the number of bytes we want
4400  * @flush - whether or not we can flush to make our reservation
4401  *
4402  * This will reserve orgi_bytes number of bytes from the space info associated
4403  * with the block_rsv.  If there is not enough space it will make an attempt to
4404  * flush out space to make room.  It will do this by flushing delalloc if
4405  * possible or committing the transaction.  If flush is 0 then no attempts to
4406  * regain reservations will be made and this will fail if there is not enough
4407  * space already.
4408  */
4409 static int reserve_metadata_bytes(struct btrfs_root *root,
4410                                   struct btrfs_block_rsv *block_rsv,
4411                                   u64 orig_bytes,
4412                                   enum btrfs_reserve_flush_enum flush)
4413 {
4414         struct btrfs_space_info *space_info = block_rsv->space_info;
4415         u64 used;
4416         u64 num_bytes = orig_bytes;
4417         int flush_state = FLUSH_DELAYED_ITEMS_NR;
4418         int ret = 0;
4419         bool flushing = false;
4420 
4421 again:
4422         ret = 0;
4423         spin_lock(&space_info->lock);
4424         /*
4425          * We only want to wait if somebody other than us is flushing and we
4426          * are actually allowed to flush all things.
4427          */
4428         while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
4429                space_info->flush) {
4430                 spin_unlock(&space_info->lock);
4431                 /*
4432                  * If we have a trans handle we can't wait because the flusher
4433                  * may have to commit the transaction, which would mean we would
4434                  * deadlock since we are waiting for the flusher to finish, but
4435                  * hold the current transaction open.
4436                  */
4437                 if (current->journal_info)
4438                         return -EAGAIN;
4439                 ret = wait_event_killable(space_info->wait, !space_info->flush);
4440                 /* Must have been killed, return */
4441                 if (ret)
4442                         return -EINTR;
4443 
4444                 spin_lock(&space_info->lock);
4445         }
4446 
4447         ret = -ENOSPC;
4448         used = space_info->bytes_used + space_info->bytes_reserved +
4449                 space_info->bytes_pinned + space_info->bytes_readonly +
4450                 space_info->bytes_may_use;
4451 
4452         /*
4453          * The idea here is that we've not already over-reserved the block group
4454          * then we can go ahead and save our reservation first and then start
4455          * flushing if we need to.  Otherwise if we've already overcommitted
4456          * lets start flushing stuff first and then come back and try to make
4457          * our reservation.
4458          */
4459         if (used <= space_info->total_bytes) {
4460                 if (used + orig_bytes <= space_info->total_bytes) {
4461                         space_info->bytes_may_use += orig_bytes;
4462                         trace_btrfs_space_reservation(root->fs_info,
4463                                 "space_info", space_info->flags, orig_bytes, 1);
4464                         ret = 0;
4465                 } else {
4466                         /*
4467                          * Ok set num_bytes to orig_bytes since we aren't
4468                          * overocmmitted, this way we only try and reclaim what
4469                          * we need.
4470                          */
4471                         num_bytes = orig_bytes;
4472                 }
4473         } else {
4474                 /*
4475                  * Ok we're over committed, set num_bytes to the overcommitted
4476                  * amount plus the amount of bytes that we need for this
4477                  * reservation.
4478                  */
4479                 num_bytes = used - space_info->total_bytes +
4480                         (orig_bytes * 2);
4481         }
4482 
4483         if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
4484                 space_info->bytes_may_use += orig_bytes;
4485                 trace_btrfs_space_reservation(root->fs_info, "space_info",
4486                                               space_info->flags, orig_bytes,
4487                                               1);
4488                 ret = 0;
4489         }
4490 
4491         /*
4492          * Couldn't make our reservation, save our place so while we're trying
4493          * to reclaim space we can actually use it instead of somebody else
4494          * stealing it from us.
4495          *
4496          * We make the other tasks wait for the flush only when we can flush
4497          * all things.
4498          */
4499         if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
4500                 flushing = true;
4501                 space_info->flush = 1;
4502         } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
4503                 used += orig_bytes;
4504                 /*
4505                  * We will do the space reservation dance during log replay,
4506                  * which means we won't have fs_info->fs_root set, so don't do
4507                  * the async reclaim as we will panic.
4508                  */
4509                 if (!root->fs_info->log_root_recovering &&
4510                     need_do_async_reclaim(space_info, root->fs_info, used) &&
4511                     !work_busy(&root->fs_info->async_reclaim_work))
4512                         queue_work(system_unbound_wq,
4513                                    &root->fs_info->async_reclaim_work);
4514         }
4515         spin_unlock(&space_info->lock);
4516 
4517         if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
4518                 goto out;
4519 
4520         ret = flush_space(root, space_info, num_bytes, orig_bytes,
4521                           flush_state);
4522         flush_state++;
4523 
4524         /*
4525          * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4526          * would happen. So skip delalloc flush.
4527          */
4528         if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4529             (flush_state == FLUSH_DELALLOC ||
4530              flush_state == FLUSH_DELALLOC_WAIT))
4531                 flush_state = ALLOC_CHUNK;
4532 
4533         if (!ret)
4534                 goto again;
4535         else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4536                  flush_state < COMMIT_TRANS)
4537                 goto again;
4538         else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
4539                  flush_state <= COMMIT_TRANS)
4540                 goto again;
4541 
4542 out:
4543         if (ret == -ENOSPC &&
4544             unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
4545                 struct btrfs_block_rsv *global_rsv =
4546                         &root->fs_info->global_block_rsv;
4547 
4548                 if (block_rsv != global_rsv &&
4549                     !block_rsv_use_bytes(global_rsv, orig_bytes))
4550                         ret = 0;
4551         }
4552         if (ret == -ENOSPC)
4553                 trace_btrfs_space_reservation(root->fs_info,
4554                                               "space_info:enospc",
4555                                               space_info->flags, orig_bytes, 1);
4556         if (flushing) {
4557                 spin_lock(&space_info->lock);
4558                 space_info->flush = 0;
4559                 wake_up_all(&space_info->wait);
4560                 spin_unlock(&space_info->lock);
4561         }
4562         return ret;
4563 }
4564 
4565 static struct btrfs_block_rsv *get_block_rsv(
4566                                         const struct btrfs_trans_handle *trans,
4567                                         const struct btrfs_root *root)
4568 {
4569         struct btrfs_block_rsv *block_rsv = NULL;
4570 
4571         if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4572                 block_rsv = trans->block_rsv;
4573 
4574         if (root == root->fs_info->csum_root && trans->adding_csums)
4575                 block_rsv = trans->block_rsv;
4576 
4577         if (root == root->fs_info->uuid_root)
4578                 block_rsv = trans->block_rsv;
4579 
4580         if (!block_rsv)
4581                 block_rsv = root->block_rsv;
4582 
4583         if (!block_rsv)
4584                 block_rsv = &root->fs_info->empty_block_rsv;
4585 
4586         return block_rsv;
4587 }
4588 
4589 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
4590                                u64 num_bytes)
4591 {
4592         int ret = -ENOSPC;
4593         spin_lock(&block_rsv->lock);
4594         if (block_rsv->reserved >= num_bytes) {
4595                 block_rsv->reserved -= num_bytes;
4596                 if (block_rsv->reserved < block_rsv->size)
4597                         block_rsv->full = 0;
4598                 ret = 0;
4599         }
4600         spin_unlock(&block_rsv->lock);
4601         return ret;
4602 }
4603 
4604 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
4605                                 u64 num_bytes, int update_size)
4606 {
4607         spin_lock(&block_rsv->lock);
4608         block_rsv->reserved += num_bytes;
4609         if (update_size)
4610                 block_rsv->size += num_bytes;
4611         else if (block_rsv->reserved >= block_rsv->size)
4612                 block_rsv->full = 1;
4613         spin_unlock(&block_rsv->lock);
4614 }
4615 
4616 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
4617                              struct btrfs_block_rsv *dest, u64 num_bytes,
4618                              int min_factor)
4619 {
4620         struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4621         u64 min_bytes;
4622 
4623         if (global_rsv->space_info != dest->space_info)
4624                 return -ENOSPC;
4625 
4626         spin_lock(&global_rsv->lock);
4627         min_bytes = div_factor(global_rsv->size, min_factor);
4628         if (global_rsv->reserved < min_bytes + num_bytes) {
4629                 spin_unlock(&global_rsv->lock);
4630                 return -ENOSPC;
4631         }
4632         global_rsv->reserved -= num_bytes;
4633         if (global_rsv->reserved < global_rsv->size)
4634                 global_rsv->full = 0;
4635         spin_unlock(&global_rsv->lock);
4636 
4637         block_rsv_add_bytes(dest, num_bytes, 1);
4638         return 0;
4639 }
4640 
4641 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
4642                                     struct btrfs_block_rsv *block_rsv,
4643                                     struct btrfs_block_rsv *dest, u64 num_bytes)
4644 {
4645         struct btrfs_space_info *space_info = block_rsv->space_info;
4646 
4647         spin_lock(&block_rsv->lock);
4648         if (num_bytes == (u64)-1)
4649                 num_bytes = block_rsv->size;
4650         block_rsv->size -= num_bytes;
4651         if (block_rsv->reserved >= block_rsv->size) {
4652                 num_bytes = block_rsv->reserved - block_rsv->size;
4653                 block_rsv->reserved = block_rsv->size;
4654                 block_rsv->full = 1;
4655         } else {
4656                 num_bytes = 0;
4657         }
4658         spin_unlock(&block_rsv->lock);
4659 
4660         if (num_bytes > 0) {
4661                 if (dest) {
4662                         spin_lock(&dest->lock);
4663                         if (!dest->full) {
4664                                 u64 bytes_to_add;
4665 
4666                                 bytes_to_add = dest->size - dest->reserved;
4667                                 bytes_to_add = min(num_bytes, bytes_to_add);
4668                                 dest->reserved += bytes_to_add;
4669                                 if (dest->reserved >= dest->size)
4670                                         dest->full = 1;
4671                                 num_bytes -= bytes_to_add;
4672                         }
4673                         spin_unlock(&dest->lock);
4674                 }
4675                 if (num_bytes) {
4676                         spin_lock(&space_info->lock);
4677                         space_info->bytes_may_use -= num_bytes;
4678                         trace_btrfs_space_reservation(fs_info, "space_info",
4679                                         space_info->flags, num_bytes, 0);
4680                         spin_unlock(&space_info->lock);
4681                 }
4682         }
4683 }
4684 
4685 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
4686                                    struct btrfs_block_rsv *dst, u64 num_bytes)
4687 {
4688         int ret;
4689 
4690         ret = block_rsv_use_bytes(src, num_bytes);
4691         if (ret)
4692                 return ret;
4693 
4694         block_rsv_add_bytes(dst, num_bytes, 1);
4695         return 0;
4696 }
4697 
4698 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
4699 {
4700         memset(rsv, 0, sizeof(*rsv));
4701         spin_lock_init(&rsv->lock);
4702         rsv->type = type;
4703 }
4704 
4705 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
4706                                               unsigned short type)
4707 {
4708         struct btrfs_block_rsv *block_rsv;
4709         struct btrfs_fs_info *fs_info = root->fs_info;
4710 
4711         block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
4712         if (!block_rsv)
4713                 return NULL;
4714 
4715         btrfs_init_block_rsv(block_rsv, type);
4716         block_rsv->space_info = __find_space_info(fs_info,
4717                                                   BTRFS_BLOCK_GROUP_METADATA);
4718         return block_rsv;
4719 }
4720 
4721 void btrfs_free_block_rsv(struct btrfs_root *root,
4722                           struct btrfs_block_rsv *rsv)
4723 {
4724         if (!rsv)
4725                 return;
4726         btrfs_block_rsv_release(root, rsv, (u64)-1);
4727         kfree(rsv);
4728 }
4729 
4730 int btrfs_block_rsv_add(struct btrfs_root *root,
4731                         struct btrfs_block_rsv *block_rsv, u64 num_bytes,
4732                         enum btrfs_reserve_flush_enum flush)
4733 {
4734         int ret;
4735 
4736         if (num_bytes == 0)
4737                 return 0;
4738 
4739         ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4740         if (!ret) {
4741                 block_rsv_add_bytes(block_rsv, num_bytes, 1);
4742                 return 0;
4743         }
4744 
4745         return ret;
4746 }
4747 
4748 int btrfs_block_rsv_check(struct btrfs_root *root,
4749                           struct btrfs_block_rsv *block_rsv, int min_factor)
4750 {
4751         u64 num_bytes = 0;
4752         int ret = -ENOSPC;
4753 
4754         if (!block_rsv)
4755                 return 0;
4756 
4757         spin_lock(&block_rsv->lock);
4758         num_bytes = div_factor(block_rsv->size, min_factor);
4759         if (block_rsv->reserved >= num_bytes)
4760                 ret = 0;
4761         spin_unlock(&block_rsv->lock);
4762 
4763         return ret;
4764 }
4765 
4766 int btrfs_block_rsv_refill(struct btrfs_root *root,
4767                            struct btrfs_block_rsv *block_rsv, u64 min_reserved,
4768                            enum btrfs_reserve_flush_enum flush)
4769 {
4770         u64 num_bytes = 0;
4771         int ret = -ENOSPC;
4772 
4773         if (!block_rsv)
4774                 return 0;
4775 
4776         spin_lock(&block_rsv->lock);
4777         num_bytes = min_reserved;
4778         if (block_rsv->reserved >= num_bytes)
4779                 ret = 0;
4780         else
4781                 num_bytes -= block_rsv->reserved;
4782         spin_unlock(&block_rsv->lock);
4783 
4784         if (!ret)
4785                 return 0;
4786 
4787         ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4788         if (!ret) {
4789                 block_rsv_add_bytes(block_rsv, num_bytes, 0);
4790                 return 0;
4791         }
4792 
4793         return ret;
4794 }
4795 
4796 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
4797                             struct btrfs_block_rsv *dst_rsv,
4798                             u64 num_bytes)
4799 {
4800         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4801 }
4802 
4803 void btrfs_block_rsv_release(struct btrfs_root *root,
4804                              struct btrfs_block_rsv *block_rsv,
4805                              u64 num_bytes)
4806 {
4807         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4808         if (global_rsv == block_rsv ||
4809             block_rsv->space_info != global_rsv->space_info)
4810                 global_rsv = NULL;
4811         block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
4812                                 num_bytes);
4813 }
4814 
4815 /*
4816  * helper to calculate size of global block reservation.
4817  * the desired value is sum of space used by extent tree,
4818  * checksum tree and root tree
4819  */
4820 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
4821 {
4822         struct btrfs_space_info *sinfo;
4823         u64 num_bytes;
4824         u64 meta_used;
4825         u64 data_used;
4826         int csum_size = btrfs_super_csum_size(fs_info->super_copy);
4827 
4828         sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
4829         spin_lock(&sinfo->lock);
4830         data_used = sinfo->bytes_used;
4831         spin_unlock(&sinfo->lock);
4832 
4833         sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4834         spin_lock(&sinfo->lock);
4835         if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
4836                 data_used = 0;
4837         meta_used = sinfo->bytes_used;
4838         spin_unlock(&sinfo->lock);
4839 
4840         num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
4841                     csum_size * 2;
4842         num_bytes += div64_u64(data_used + meta_used, 50);
4843 
4844         if (num_bytes * 3 > meta_used)
4845                 num_bytes = div64_u64(meta_used, 3);
4846 
4847         return ALIGN(num_bytes, fs_info->extent_root->nodesize << 10);
4848 }
4849 
4850 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
4851 {
4852         struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4853         struct btrfs_space_info *sinfo = block_rsv->space_info;
4854         u64 num_bytes;
4855 
4856         num_bytes = calc_global_metadata_size(fs_info);
4857 
4858         spin_lock(&sinfo->lock);
4859         spin_lock(&block_rsv->lock);
4860 
4861         block_rsv->size = min_t(u64, num_bytes, 512 * 1024 * 1024);
4862 
4863         num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
4864                     sinfo->bytes_reserved + sinfo->bytes_readonly +
4865                     sinfo->bytes_may_use;
4866 
4867         if (sinfo->total_bytes > num_bytes) {
4868                 num_bytes = sinfo->total_bytes - num_bytes;
4869                 block_rsv->reserved += num_bytes;
4870                 sinfo->bytes_may_use += num_bytes;
4871                 trace_btrfs_space_reservation(fs_info, "space_info",
4872                                       sinfo->flags, num_bytes, 1);
4873         }
4874 
4875         if (block_rsv->reserved >= block_rsv->size) {
4876                 num_bytes = block_rsv->reserved - block_rsv->size;
4877                 sinfo->bytes_may_use -= num_bytes;
4878                 trace_btrfs_space_reservation(fs_info, "space_info",
4879                                       sinfo->flags, num_bytes, 0);
4880                 block_rsv->reserved = block_rsv->size;
4881                 block_rsv->full = 1;
4882         }
4883 
4884         spin_unlock(&block_rsv->lock);
4885         spin_unlock(&sinfo->lock);
4886 }
4887 
4888 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
4889 {
4890         struct btrfs_space_info *space_info;
4891 
4892         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4893         fs_info->chunk_block_rsv.space_info = space_info;
4894 
4895         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4896         fs_info->global_block_rsv.space_info = space_info;
4897         fs_info->delalloc_block_rsv.space_info = space_info;
4898         fs_info->trans_block_rsv.space_info = space_info;
4899         fs_info->empty_block_rsv.space_info = space_info;
4900         fs_info->delayed_block_rsv.space_info = space_info;
4901 
4902         fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
4903         fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
4904         fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
4905         fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
4906         if (fs_info->quota_root)
4907                 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
4908         fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
4909 
4910         update_global_block_rsv(fs_info);
4911 }
4912 
4913 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
4914 {
4915         block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
4916                                 (u64)-1);
4917         WARN_ON(fs_info->delalloc_block_rsv.size > 0);
4918         WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
4919         WARN_ON(fs_info->trans_block_rsv.size > 0);
4920         WARN_ON(fs_info->trans_block_rsv.reserved > 0);
4921         WARN_ON(fs_info->chunk_block_rsv.size > 0);
4922         WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
4923         WARN_ON(fs_info->delayed_block_rsv.size > 0);
4924         WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
4925 }
4926 
4927 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4928                                   struct btrfs_root *root)
4929 {
4930         if (!trans->block_rsv)
4931                 return;
4932 
4933         if (!trans->bytes_reserved)
4934                 return;
4935 
4936         trace_btrfs_space_reservation(root->fs_info, "transaction",
4937                                       trans->transid, trans->bytes_reserved, 0);
4938         btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
4939         trans->bytes_reserved = 0;
4940 }
4941 
4942 /* Can only return 0 or -ENOSPC */
4943 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4944                                   struct inode *inode)
4945 {
4946         struct btrfs_root *root = BTRFS_I(inode)->root;
4947         struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4948         struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4949 
4950         /*
4951          * We need to hold space in order to delete our orphan item once we've
4952          * added it, so this takes the reservation so we can release it later
4953          * when we are truly done with the orphan item.
4954          */
4955         u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4956         trace_btrfs_space_reservation(root->fs_info, "orphan",
4957                                       btrfs_ino(inode), num_bytes, 1);
4958         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4959 }
4960 
4961 void btrfs_orphan_release_metadata(struct inode *inode)
4962 {
4963         struct btrfs_root *root = BTRFS_I(inode)->root;
4964         u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4965         trace_btrfs_space_reservation(root->fs_info, "orphan",
4966                                       btrfs_ino(inode), num_bytes, 0);
4967         btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4968 }
4969 
4970 /*
4971  * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
4972  * root: the root of the parent directory
4973  * rsv: block reservation
4974  * items: the number of items that we need do reservation
4975  * qgroup_reserved: used to return the reserved size in qgroup
4976  *
4977  * This function is used to reserve the space for snapshot/subvolume
4978  * creation and deletion. Those operations are different with the
4979  * common file/directory operations, they change two fs/file trees
4980  * and root tree, the number of items that the qgroup reserves is
4981  * different with the free space reservation. So we can not use
4982  * the space reseravtion mechanism in start_transaction().
4983  */
4984 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
4985                                      struct btrfs_block_rsv *rsv,
4986                                      int items,
4987                                      u64 *qgroup_reserved,
4988                                      bool use_global_rsv)
4989 {
4990         u64 num_bytes;
4991         int ret;
4992         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4993 
4994         if (root->fs_info->quota_enabled) {
4995                 /* One for parent inode, two for dir entries */
4996                 num_bytes = 3 * root->nodesize;
4997                 ret = btrfs_qgroup_reserve(root, num_bytes);
4998                 if (ret)
4999                         return ret;
5000         } else {
5001                 num_bytes = 0;
5002         }
5003 
5004         *qgroup_reserved = num_bytes;
5005 
5006         num_bytes = btrfs_calc_trans_metadata_size(root, items);
5007         rsv->space_info = __find_space_info(root->fs_info,
5008                                             BTRFS_BLOCK_GROUP_METADATA);
5009         ret = btrfs_block_rsv_add(root, rsv, num_bytes,
5010                                   BTRFS_RESERVE_FLUSH_ALL);
5011 
5012         if (ret == -ENOSPC && use_global_rsv)
5013                 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes);
5014 
5015         if (ret) {
5016                 if (*qgroup_reserved)
5017                         btrfs_qgroup_free(root, *qgroup_reserved);
5018         }
5019 
5020         return ret;
5021 }
5022 
5023 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
5024                                       struct btrfs_block_rsv *rsv,
5025                                       u64 qgroup_reserved)
5026 {
5027         btrfs_block_rsv_release(root, rsv, (u64)-1);
5028         if (qgroup_reserved)
5029                 btrfs_qgroup_free(root, qgroup_reserved);
5030 }
5031 
5032 /**
5033  * drop_outstanding_extent - drop an outstanding extent
5034  * @inode: the inode we're dropping the extent for
5035  *
5036  * This is called when we are freeing up an outstanding extent, either called
5037  * after an error or after an extent is written.  This will return the number of
5038  * reserved extents that need to be freed.  This must be called with
5039  * BTRFS_I(inode)->lock held.
5040  */
5041 static unsigned drop_outstanding_extent(struct inode *inode)
5042 {
5043         unsigned drop_inode_space = 0;
5044         unsigned dropped_extents = 0;
5045 
5046         BUG_ON(!BTRFS_I(inode)->outstanding_extents);
5047         BTRFS_I(inode)->outstanding_extents--;
5048 
5049         if (BTRFS_I(inode)->outstanding_extents == 0 &&
5050             test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5051                                &BTRFS_I(inode)->runtime_flags))
5052                 drop_inode_space = 1;
5053 
5054         /*
5055          * If we have more or the same amount of outsanding extents than we have
5056          * reserved then we need to leave the reserved extents count alone.
5057          */
5058         if (BTRFS_I(inode)->outstanding_extents >=
5059             BTRFS_I(inode)->reserved_extents)
5060                 return drop_inode_space;
5061 
5062         dropped_extents = BTRFS_I(inode)->reserved_extents -
5063                 BTRFS_I(inode)->outstanding_extents;
5064         BTRFS_I(inode)->reserved_extents -= dropped_extents;
5065         return dropped_extents + drop_inode_space;
5066 }
5067 
5068 /**
5069  * calc_csum_metadata_size - return the amount of metada space that must be
5070  *      reserved/free'd for the given bytes.
5071  * @inode: the inode we're manipulating
5072  * @num_bytes: the number of bytes in question
5073  * @reserve: 1 if we are reserving space, 0 if we are freeing space
5074  *
5075  * This adjusts the number of csum_bytes in the inode and then returns the
5076  * correct amount of metadata that must either be reserved or freed.  We
5077  * calculate how many checksums we can fit into one leaf and then divide the
5078  * number of bytes that will need to be checksumed by this value to figure out
5079  * how many checksums will be required.  If we are adding bytes then the number
5080  * may go up and we will return the number of additional bytes that must be
5081  * reserved.  If it is going down we will return the number of bytes that must
5082  * be freed.
5083  *
5084  * This must be called with BTRFS_I(inode)->lock held.
5085  */
5086 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
5087                                    int reserve)
5088 {
5089         struct btrfs_root *root = BTRFS_I(inode)->root;
5090         u64 csum_size;
5091         int num_csums_per_leaf;
5092         int num_csums;
5093         int old_csums;
5094 
5095         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
5096             BTRFS_I(inode)->csum_bytes == 0)
5097                 return 0;
5098 
5099         old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
5100         if (reserve)
5101                 BTRFS_I(inode)->csum_bytes += num_bytes;
5102         else
5103                 BTRFS_I(inode)->csum_bytes -= num_bytes;
5104         csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
5105         num_csums_per_leaf = (int)div64_u64(csum_size,
5106                                             sizeof(struct btrfs_csum_item) +
5107                                             sizeof(struct btrfs_disk_key));
5108         num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
5109         num_csums = num_csums + num_csums_per_leaf - 1;
5110         num_csums = num_csums / num_csums_per_leaf;
5111 
5112         old_csums = old_csums + num_csums_per_leaf - 1;
5113         old_csums = old_csums / num_csums_per_leaf;
5114 
5115         /* No change, no need to reserve more */
5116         if (old_csums == num_csums)
5117                 return 0;
5118 
5119         if (reserve)
5120                 return btrfs_calc_trans_metadata_size(root,
5121                                                       num_csums - old_csums);
5122 
5123         return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
5124 }
5125 
5126 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
5127 {
5128         struct btrfs_root *root = BTRFS_I(inode)->root;
5129         struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
5130         u64 to_reserve = 0;
5131         u64 csum_bytes;
5132         unsigned nr_extents = 0;
5133         int extra_reserve = 0;
5134         enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
5135         int ret = 0;
5136         bool delalloc_lock = true;
5137         u64 to_free = 0;
5138         unsigned dropped;
5139 
5140         /* If we are a free space inode we need to not flush since we will be in
5141          * the middle of a transaction commit.  We also don't need the delalloc
5142          * mutex since we won't race with anybody.  We need this mostly to make
5143          * lockdep shut its filthy mouth.
5144          */
5145         if (btrfs_is_free_space_inode(inode)) {
5146                 flush = BTRFS_RESERVE_NO_FLUSH;
5147                 delalloc_lock = false;
5148         }
5149 
5150         if (flush != BTRFS_RESERVE_NO_FLUSH &&
5151             btrfs_transaction_in_commit(root->fs_info))
5152                 schedule_timeout(1);
5153 
5154         if (delalloc_lock)
5155                 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
5156 
5157         num_bytes = ALIGN(num_bytes, root->sectorsize);
5158 
5159         spin_lock(&BTRFS_I(inode)->lock);
5160         BTRFS_I(inode)->outstanding_extents++;
5161 
5162         if (BTRFS_I(inode)->outstanding_extents >
5163             BTRFS_I(inode)->reserved_extents)
5164                 nr_extents = BTRFS_I(inode)->outstanding_extents -
5165                         BTRFS_I(inode)->reserved_extents;
5166 
5167         /*
5168          * Add an item to reserve for updating the inode when we complete the
5169          * delalloc io.
5170          */
5171         if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5172                       &BTRFS_I(inode)->runtime_flags)) {
5173                 nr_extents++;
5174                 extra_reserve = 1;
5175         }
5176 
5177         to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
5178         to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
5179         csum_bytes = BTRFS_I(inode)->csum_bytes;
5180         spin_unlock(&BTRFS_I(inode)->lock);
5181 
5182         if (root->fs_info->quota_enabled) {
5183                 ret = btrfs_qgroup_reserve(root, num_bytes +
5184                                            nr_extents * root->nodesize);
5185                 if (ret)
5186                         goto out_fail;
5187         }
5188 
5189         ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
5190         if (unlikely(ret)) {
5191                 if (root->fs_info->quota_enabled)
5192                         btrfs_qgroup_free(root, num_bytes +
5193                                                 nr_extents * root->nodesize);
5194                 goto out_fail;
5195         }
5196 
5197         spin_lock(&BTRFS_I(inode)->lock);
5198         if (extra_reserve) {
5199                 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5200                         &BTRFS_I(inode)->runtime_flags);
5201                 nr_extents--;
5202         }
5203         BTRFS_I(inode)->reserved_extents += nr_extents;
5204         spin_unlock(&BTRFS_I(inode)->lock);
5205 
5206         if (delalloc_lock)
5207                 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5208 
5209         if (to_reserve)
5210                 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5211                                               btrfs_ino(inode), to_reserve, 1);
5212         block_rsv_add_bytes(block_rsv, to_reserve, 1);
5213 
5214         return 0;
5215 
5216 out_fail:
5217         spin_lock(&BTRFS_I(inode)->lock);
5218         dropped = drop_outstanding_extent(inode);
5219         /*
5220          * If the inodes csum_bytes is the same as the original
5221          * csum_bytes then we know we haven't raced with any free()ers
5222          * so we can just reduce our inodes csum bytes and carry on.
5223          */
5224         if (BTRFS_I(inode)->csum_bytes == csum_bytes) {
5225                 calc_csum_metadata_size(inode, num_bytes, 0);
5226         } else {
5227                 u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes;
5228                 u64 bytes;
5229 
5230                 /*
5231                  * This is tricky, but first we need to figure out how much we
5232                  * free'd from any free-ers that occured during this
5233                  * reservation, so we reset ->csum_bytes to the csum_bytes
5234                  * before we dropped our lock, and then call the free for the
5235                  * number of bytes that were freed while we were trying our
5236                  * reservation.
5237                  */
5238                 bytes = csum_bytes - BTRFS_I(inode)->csum_bytes;
5239                 BTRFS_I(inode)->csum_bytes = csum_bytes;
5240                 to_free = calc_csum_metadata_size(inode, bytes, 0);
5241 
5242 
5243                 /*
5244                  * Now we need to see how much we would have freed had we not
5245                  * been making this reservation and our ->csum_bytes were not
5246                  * artificially inflated.
5247                  */
5248                 BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes;
5249                 bytes = csum_bytes - orig_csum_bytes;
5250                 bytes = calc_csum_metadata_size(inode, bytes, 0);
5251 
5252                 /*
5253                  * Now reset ->csum_bytes to what it should be.  If bytes is
5254                  * more than to_free then we would have free'd more space had we
5255                  * not had an artificially high ->csum_bytes, so we need to free
5256                  * the remainder.  If bytes is the same or less then we don't
5257                  * need to do anything, the other free-ers did the correct
5258                  * thing.
5259                  */
5260                 BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes;
5261                 if (bytes > to_free)
5262                         to_free = bytes - to_free;
5263                 else
5264                         to_free = 0;
5265         }
5266         spin_unlock(&BTRFS_I(inode)->lock);
5267         if (dropped)
5268                 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5269 
5270         if (to_free) {
5271                 btrfs_block_rsv_release(root, block_rsv, to_free);
5272                 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5273                                               btrfs_ino(inode), to_free, 0);
5274         }
5275         if (delalloc_lock)
5276                 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5277         return ret;
5278 }
5279 
5280 /**
5281  * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5282  * @inode: the inode to release the reservation for
5283  * @num_bytes: the number of bytes we're releasing
5284  *
5285  * This will release the metadata reservation for an inode.  This can be called
5286  * once we complete IO for a given set of bytes to release their metadata
5287  * reservations.
5288  */
5289 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
5290 {
5291         struct btrfs_root *root = BTRFS_I(inode)->root;
5292         u64 to_free = 0;
5293         unsigned dropped;
5294 
5295         num_bytes = ALIGN(num_bytes, root->sectorsize);
5296         spin_lock(&BTRFS_I(inode)->lock);
5297         dropped = drop_outstanding_extent(inode);
5298 
5299         if (num_bytes)
5300                 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
5301         spin_unlock(&BTRFS_I(inode)->lock);
5302         if (dropped > 0)
5303                 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5304 
5305         trace_btrfs_space_reservation(root->fs_info, "delalloc",
5306                                       btrfs_ino(inode), to_free, 0);
5307         if (root->fs_info->quota_enabled) {
5308                 btrfs_qgroup_free(root, num_bytes +
5309                                         dropped * root->nodesize);
5310         }
5311 
5312         btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
5313                                 to_free);
5314 }
5315 
5316 /**
5317  * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
5318  * @inode: inode we're writing to
5319  * @num_bytes: the number of bytes we want to allocate
5320  *
5321  * This will do the following things
5322  *
5323  * o reserve space in the data space info for num_bytes
5324  * o reserve space in the metadata space info based on number of outstanding
5325  *   extents and how much csums will be needed
5326  * o add to the inodes ->delalloc_bytes
5327  * o add it to the fs_info's delalloc inodes list.
5328  *
5329  * This will return 0 for success and -ENOSPC if there is no space left.
5330  */
5331 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
5332 {
5333         int ret;
5334 
5335         ret = btrfs_check_data_free_space(inode, num_bytes);
5336         if (ret)
5337                 return ret;
5338 
5339         ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
5340         if (ret) {
5341                 btrfs_free_reserved_data_space(inode, num_bytes);
5342                 return ret;
5343         }
5344 
5345         return 0;
5346 }
5347 
5348 /**
5349  * btrfs_delalloc_release_space - release data and metadata space for delalloc
5350  * @inode: inode we're releasing space for
5351  * @num_bytes: the number of bytes we want to free up
5352  *
5353  * This must be matched with a call to btrfs_delalloc_reserve_space.  This is
5354  * called in the case that we don't need the metadata AND data reservations
5355  * anymore.  So if there is an error or we insert an inline extent.
5356  *
5357  * This function will release the metadata space that was not used and will
5358  * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5359  * list if there are no delalloc bytes left.
5360  */
5361 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
5362 {
5363         btrfs_delalloc_release_metadata(inode, num_bytes);
5364         btrfs_free_reserved_data_space(inode, num_bytes);
5365 }
5366 
5367 static int update_block_group(struct btrfs_root *root,
5368                               u64 bytenr, u64 num_bytes, int alloc)
5369 {
5370         struct btrfs_block_group_cache *cache = NULL;
5371         struct btrfs_fs_info *info = root->fs_info;
5372         u64 total = num_bytes;
5373         u64 old_val;
5374         u64 byte_in_group;
5375         int factor;
5376 
5377         /* block accounting for super block */
5378         spin_lock(&info->delalloc_root_lock);
5379         old_val = btrfs_super_bytes_used(info->super_copy);
5380         if (alloc)
5381                 old_val += num_bytes;
5382         else
5383                 old_val -= num_bytes;
5384         btrfs_set_super_bytes_used(info->super_copy, old_val);
5385         spin_unlock(&info->delalloc_root_lock);
5386 
5387         while (total) {
5388                 cache = btrfs_lookup_block_group(info, bytenr);
5389                 if (!cache)
5390                         return -ENOENT;
5391                 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
5392                                     BTRFS_BLOCK_GROUP_RAID1 |
5393                                     BTRFS_BLOCK_GROUP_RAID10))
5394                         factor = 2;
5395                 else
5396                         factor = 1;
5397                 /*
5398                  * If this block group has free space cache written out, we
5399                  * need to make sure to load it if we are removing space.  This
5400                  * is because we need the unpinning stage to actually add the
5401                  * space back to the block group, otherwise we will leak space.
5402                  */
5403                 if (!alloc && cache->cached == BTRFS_CACHE_NO)
5404                         cache_block_group(cache, 1);
5405 
5406                 byte_in_group = bytenr - cache->key.objectid;
5407                 WARN_ON(byte_in_group > cache->key.offset);
5408 
5409                 spin_lock(&cache->space_info->lock);
5410                 spin_lock(&cache->lock);
5411 
5412                 if (btrfs_test_opt(root, SPACE_CACHE) &&
5413                     cache->disk_cache_state < BTRFS_DC_CLEAR)
5414                         cache->disk_cache_state = BTRFS_DC_CLEAR;
5415 
5416                 cache->dirty = 1;
5417                 old_val = btrfs_block_group_used(&cache->item);
5418                 num_bytes = min(total, cache->key.offset - byte_in_group);
5419                 if (alloc) {
5420                         old_val += num_bytes;
5421                         btrfs_set_block_group_used(&cache->item, old_val);
5422                         cache->reserved -= num_bytes;
5423                         cache->space_info->bytes_reserved -= num_bytes;
5424                         cache->space_info->bytes_used += num_bytes;
5425                         cache->space_info->disk_used += num_bytes * factor;
5426                         spin_unlock(&cache->lock);
5427                         spin_unlock(&cache->space_info->lock);
5428                 } else {
5429                         old_val -= num_bytes;
5430 
5431                         /*
5432                          * No longer have used bytes in this block group, queue
5433                          * it for deletion.
5434                          */
5435                         if (old_val == 0) {
5436                                 spin_lock(&info->unused_bgs_lock);
5437                                 if (list_empty(&cache->bg_list)) {
5438                                         btrfs_get_block_group(cache);
5439                                         list_add_tail(&cache->bg_list,
5440                                                       &info->unused_bgs);
5441                                 }
5442                                 spin_unlock(&info->unused_bgs_lock);
5443                         }
5444                         btrfs_set_block_group_used(&cache->item, old_val);
5445                         cache->pinned += num_bytes;
5446                         cache->space_info->bytes_pinned += num_bytes;
5447                         cache->space_info->bytes_used -= num_bytes;
5448                         cache->space_info->disk_used -= num_bytes * factor;
5449                         spin_unlock(&cache->lock);
5450                         spin_unlock(&cache->space_info->lock);
5451 
5452                         set_extent_dirty(info->pinned_extents,
5453                                          bytenr, bytenr + num_bytes - 1,
5454                                          GFP_NOFS | __GFP_NOFAIL);
5455                 }
5456                 btrfs_put_block_group(cache);
5457                 total -= num_bytes;
5458                 bytenr += num_bytes;
5459         }
5460         return 0;
5461 }
5462 
5463 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
5464 {
5465         struct btrfs_block_group_cache *cache;
5466         u64 bytenr;
5467 
5468         spin_lock(&root->fs_info->block_group_cache_lock);
5469         bytenr = root->fs_info->first_logical_byte;
5470         spin_unlock(&root->fs_info->block_group_cache_lock);
5471 
5472         if (bytenr < (u64)-1)
5473                 return bytenr;
5474 
5475         cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
5476         if (!cache)
5477                 return 0;
5478 
5479         bytenr = cache->key.objectid;
5480         btrfs_put_block_group(cache);
5481 
5482         return bytenr;
5483 }
5484 
5485 static int pin_down_extent(struct btrfs_root *root,
5486                            struct btrfs_block_group_cache *cache,
5487                            u64 bytenr, u64 num_bytes, int reserved)
5488 {
5489         spin_lock(&cache->space_info->lock);
5490         spin_lock(&cache->lock);
5491         cache->pinned += num_bytes;
5492         cache->space_info->bytes_pinned += num_bytes;
5493         if (reserved) {
5494                 cache->reserved -= num_bytes;
5495                 cache->space_info->bytes_reserved -= num_bytes;
5496         }
5497         spin_unlock(&cache->lock);
5498         spin_unlock(&cache->space_info->lock);
5499 
5500         set_extent_dirty(root->fs_info->pinned_extents, bytenr,
5501                          bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
5502         if (reserved)
5503                 trace_btrfs_reserved_extent_free(root, bytenr, num_bytes);
5504         return 0;
5505 }
5506 
5507 /*
5508  * this function must be called within transaction
5509  */
5510 int btrfs_pin_extent(struct btrfs_root *root,
5511                      u64 bytenr, u64 num_bytes, int reserved)
5512 {
5513         struct btrfs_block_group_cache *cache;
5514 
5515         cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5516         BUG_ON(!cache); /* Logic error */
5517 
5518         pin_down_extent(root, cache, bytenr, num_bytes, reserved);
5519 
5520         btrfs_put_block_group(cache);
5521         return 0;
5522 }
5523 
5524 /*
5525  * this function must be called within transaction
5526  */
5527 int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
5528                                     u64 bytenr, u64 num_bytes)
5529 {
5530         struct btrfs_block_group_cache *cache;
5531         int ret;
5532 
5533         cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5534         if (!cache)
5535                 return -EINVAL;
5536 
5537         /*
5538          * pull in the free space cache (if any) so that our pin
5539          * removes the free space from the cache.  We have load_only set
5540          * to one because the slow code to read in the free extents does check
5541          * the pinned extents.
5542          */
5543         cache_block_group(cache, 1);
5544 
5545         pin_down_extent(root, cache, bytenr, num_bytes, 0);
5546 
5547         /* remove us from the free space cache (if we're there at all) */
5548         ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
5549         btrfs_put_block_group(cache);
5550         return ret;
5551 }
5552 
5553 static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes)
5554 {
5555         int ret;
5556         struct btrfs_block_group_cache *block_group;
5557         struct btrfs_caching_control *caching_ctl;
5558 
5559         block_group = btrfs_lookup_block_group(root->fs_info, start);
5560         if (!block_group)
5561                 return -EINVAL;
5562 
5563         cache_block_group(block_group, 0);
5564         caching_ctl = get_caching_control(block_group);
5565 
5566         if (!caching_ctl) {
5567                 /* Logic error */
5568                 BUG_ON(!block_group_cache_done(block_group));
5569                 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5570         } else {
5571                 mutex_lock(&caching_ctl->mutex);
5572 
5573                 if (start >= caching_ctl->progress) {
5574                         ret = add_excluded_extent(root, start, num_bytes);
5575                 } else if (start + num_bytes <= caching_ctl->progress) {
5576                         ret = btrfs_remove_free_space(block_group,
5577                                                       start, num_bytes);
5578                 } else {
5579                         num_bytes = caching_ctl->progress - start;
5580                         ret = btrfs_remove_free_space(block_group,
5581                                                       start, num_bytes);
5582                         if (ret)
5583                                 goto out_lock;
5584 
5585                         num_bytes = (start + num_bytes) -
5586                                 caching_ctl->progress;
5587                         start = caching_ctl->progress;
5588                         ret = add_excluded_extent(root, start, num_bytes);
5589                 }
5590 out_lock:
5591                 mutex_unlock(&caching_ctl->mutex);
5592                 put_caching_control(caching_ctl);
5593         }
5594         btrfs_put_block_group(block_group);
5595         return ret;
5596 }
5597 
5598 int btrfs_exclude_logged_extents(struct btrfs_root *log,
5599                                  struct extent_buffer *eb)
5600 {
5601         struct btrfs_file_extent_item *item;
5602         struct btrfs_key key;
5603         int found_type;
5604         int i;
5605 
5606         if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS))
5607                 return 0;
5608 
5609         for (i = 0; i < btrfs_header_nritems(eb); i++) {
5610                 btrfs_item_key_to_cpu(eb, &key, i);
5611                 if (key.type != BTRFS_EXTENT_DATA_KEY)
5612                         continue;
5613                 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
5614                 found_type = btrfs_file_extent_type(eb, item);
5615                 if (found_type == BTRFS_FILE_EXTENT_INLINE)
5616                         continue;
5617                 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
5618                         continue;
5619                 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
5620                 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
5621                 __exclude_logged_extent(log, key.objectid, key.offset);
5622         }
5623 
5624         return 0;
5625 }
5626 
5627 /**
5628  * btrfs_update_reserved_bytes - update the block_group and space info counters
5629  * @cache:      The cache we are manipulating
5630  * @num_bytes:  The number of bytes in question
5631  * @reserve:    One of the reservation enums
5632  * @delalloc:   The blocks are allocated for the delalloc write
5633  *
5634  * This is called by the allocator when it reserves space, or by somebody who is
5635  * freeing space that was never actually used on disk.  For example if you
5636  * reserve some space for a new leaf in transaction A and before transaction A
5637  * commits you free that leaf, you call this with reserve set to 0 in order to
5638  * clear the reservation.
5639  *
5640  * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
5641  * ENOSPC accounting.  For data we handle the reservation through clearing the
5642  * delalloc bits in the io_tree.  We have to do this since we could end up
5643  * allocating less disk space for the amount of data we have reserved in the
5644  * case of compression.
5645  *
5646  * If this is a reservation and the block group has become read only we cannot
5647  * make the reservation and return -EAGAIN, otherwise this function always
5648  * succeeds.
5649  */
5650 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
5651                                        u64 num_bytes, int reserve, int delalloc)
5652 {
5653         struct btrfs_space_info *space_info = cache->space_info;
5654         int ret = 0;
5655 
5656         spin_lock(&space_info->lock);
5657         spin_lock(&cache->lock);
5658         if (reserve != RESERVE_FREE) {
5659                 if (cache->ro) {
5660                         ret = -EAGAIN;
5661                 } else {
5662                         cache->reserved += num_bytes;
5663                         space_info->bytes_reserved += num_bytes;
5664                         if (reserve == RESERVE_ALLOC) {
5665                                 trace_btrfs_space_reservation(cache->fs_info,
5666                                                 "space_info", space_info->flags,
5667                                                 num_bytes, 0);
5668                                 space_info->bytes_may_use -= num_bytes;
5669                         }
5670 
5671                         if (delalloc)
5672                                 cache->delalloc_bytes += num_bytes;
5673                 }
5674         } else {
5675                 if (cache->ro)
5676                         space_info->bytes_readonly += num_bytes;
5677                 cache->reserved -= num_bytes;
5678                 space_info->bytes_reserved -= num_bytes;
5679 
5680                 if (delalloc)
5681                         cache->delalloc_bytes -= num_bytes;
5682         }
5683         spin_unlock(&cache->lock);
5684         spin_unlock(&space_info->lock);
5685         return ret;
5686 }
5687 
5688 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
5689                                 struct btrfs_root *root)
5690 {
5691         struct btrfs_fs_info *fs_info = root->fs_info;
5692         struct btrfs_caching_control *next;
5693         struct btrfs_caching_control *caching_ctl;
5694         struct btrfs_block_group_cache *cache;
5695 
5696         down_write(&fs_info->commit_root_sem);
5697 
5698         list_for_each_entry_safe(caching_ctl, next,
5699                                  &fs_info->caching_block_groups, list) {
5700                 cache = caching_ctl->block_group;
5701                 if (block_group_cache_done(cache)) {
5702                         cache->last_byte_to_unpin = (u64)-1;
5703                         list_del_init(&caching_ctl->list);
5704                         put_caching_control(caching_ctl);
5705                 } else {
5706                         cache->last_byte_to_unpin = caching_ctl->progress;
5707                 }
5708         }
5709 
5710         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5711                 fs_info->pinned_extents = &fs_info->freed_extents[1];
5712         else
5713                 fs_info->pinned_extents = &fs_info->freed_extents[0];
5714 
5715         up_write(&fs_info->commit_root_sem);
5716 
5717         update_global_block_rsv(fs_info);
5718 }
5719 
5720 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end,
5721                               const bool return_free_space)
5722 {
5723         struct btrfs_fs_info *fs_info = root->fs_info;
5724         struct btrfs_block_group_cache *cache = NULL;
5725         struct btrfs_space_info *space_info;
5726         struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5727         u64 len;
5728         bool readonly;
5729 
5730         while (start <= end) {
5731                 readonly = false;
5732                 if (!cache ||
5733                     start >= cache->key.objectid + cache->key.offset) {
5734                         if (cache)
5735                                 btrfs_put_block_group(cache);
5736                         cache = btrfs_lookup_block_group(fs_info, start);
5737                         BUG_ON(!cache); /* Logic error */
5738                 }
5739 
5740                 len = cache->key.objectid + cache->key.offset - start;
5741                 len = min(len, end + 1 - start);
5742 
5743                 if (start < cache->last_byte_to_unpin) {
5744                         len = min(len, cache->last_byte_to_unpin - start);
5745                         if (return_free_space)
5746                                 btrfs_add_free_space(cache, start, len);
5747                 }
5748 
5749                 start += len;
5750                 space_info = cache->space_info;
5751 
5752                 spin_lock(&space_info->lock);
5753                 spin_lock(&cache->lock);
5754                 cache->pinned -= len;
5755                 space_info->bytes_pinned -= len;
5756                 percpu_counter_add(&space_info->total_bytes_pinned, -len);
5757                 if (cache->ro) {
5758                         space_info->bytes_readonly += len;
5759                         readonly = true;
5760                 }
5761                 spin_unlock(&cache->lock);
5762                 if (!readonly && global_rsv->space_info == space_info) {
5763                         spin_lock(&global_rsv->lock);
5764                         if (!global_rsv->full) {
5765                                 len = min(len, global_rsv->size -
5766                                           global_rsv->reserved);
5767                                 global_rsv->reserved += len;
5768                                 space_info->bytes_may_use += len;
5769                                 if (global_rsv->reserved >= global_rsv->size)
5770                                         global_rsv->full = 1;
5771                         }
5772                         spin_unlock(&global_rsv->lock);
5773                 }
5774                 spin_unlock(&space_info->lock);
5775         }
5776 
5777         if (cache)
5778                 btrfs_put_block_group(cache);
5779         return 0;
5780 }
5781 
5782 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
5783                                struct btrfs_root *root)
5784 {
5785         struct btrfs_fs_info *fs_info = root->fs_info;
5786         struct extent_io_tree *unpin;
5787         u64 start;
5788         u64 end;
5789         int ret;
5790 
5791         if (trans->aborted)
5792                 return 0;
5793 
5794         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5795                 unpin = &fs_info->freed_extents[1];
5796         else
5797                 unpin = &fs_info->freed_extents[0];
5798 
5799         while (1) {
5800                 ret = find_first_extent_bit(unpin, 0, &start, &end,
5801                                             EXTENT_DIRTY, NULL);
5802                 if (ret)
5803                         break;
5804 
5805                 if (btrfs_test_opt(root, DISCARD))
5806                         ret = btrfs_discard_extent(root, start,
5807                                                    end + 1 - start, NULL);
5808 
5809                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
5810                 unpin_extent_range(root, start, end, true);
5811                 cond_resched();
5812         }
5813 
5814         return 0;
5815 }
5816 
5817 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes,
5818                              u64 owner, u64 root_objectid)
5819 {
5820         struct btrfs_space_info *space_info;
5821         u64 flags;
5822 
5823         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
5824                 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
5825                         flags = BTRFS_BLOCK_GROUP_SYSTEM;
5826                 else
5827                         flags = BTRFS_BLOCK_GROUP_METADATA;
5828         } else {
5829                 flags = BTRFS_BLOCK_GROUP_DATA;
5830         }
5831 
5832         space_info = __find_space_info(fs_info, flags);
5833         BUG_ON(!space_info); /* Logic bug */
5834         percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
5835 }
5836 
5837 
5838 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
5839                                 struct btrfs_root *root,
5840                                 u64 bytenr, u64 num_bytes, u64 parent,
5841                                 u64 root_objectid, u64 owner_objectid,
5842                                 u64 owner_offset, int refs_to_drop,
5843                                 struct btrfs_delayed_extent_op *extent_op,
5844                                 int no_quota)
5845 {
5846         struct btrfs_key key;
5847         struct btrfs_path *path;
5848         struct btrfs_fs_info *info = root->fs_info;
5849         struct btrfs_root *extent_root = info->extent_root;
5850         struct extent_buffer *leaf;
5851         struct btrfs_extent_item *ei;
5852         struct btrfs_extent_inline_ref *iref;
5853         int ret;
5854         int is_data;
5855         int extent_slot = 0;
5856         int found_extent = 0;
5857         int num_to_del = 1;
5858         u32 item_size;
5859         u64 refs;
5860         int last_ref = 0;
5861         enum btrfs_qgroup_operation_type type = BTRFS_QGROUP_OPER_SUB_EXCL;
5862         bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
5863                                                  SKINNY_METADATA);
5864 
5865         if (!info->quota_enabled || !is_fstree(root_objectid))
5866                 no_quota = 1;
5867 
5868         path = btrfs_alloc_path();
5869         if (!path)
5870                 return -ENOMEM;
5871 
5872         path->reada = 1;
5873         path->leave_spinning = 1;
5874 
5875         is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
5876         BUG_ON(!is_data && refs_to_drop != 1);
5877 
5878         if (is_data)
5879                 skinny_metadata = 0;
5880 
5881         ret = lookup_extent_backref(trans, extent_root, path, &iref,
5882                                     bytenr, num_bytes, parent,
5883                                     root_objectid, owner_objectid,
5884                                     owner_offset);
5885         if (ret == 0) {
5886                 extent_slot = path->slots[0];
5887                 while (extent_slot >= 0) {
5888                         btrfs_item_key_to_cpu(path->nodes[0], &key,
5889                                               extent_slot);
5890                         if (key.objectid != bytenr)
5891                                 break;
5892                         if (key.type == BTRFS_EXTENT_ITEM_KEY &&
5893                             key.offset == num_bytes) {
5894                                 found_extent = 1;
5895                                 break;
5896                         }
5897                         if (key.type == BTRFS_METADATA_ITEM_KEY &&
5898                             key.offset == owner_objectid) {
5899                                 found_extent = 1;
5900                                 break;
5901                         }
5902                         if (path->slots[0] - extent_slot > 5)
5903                                 break;
5904                         extent_slot--;
5905                 }
5906 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5907                 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
5908                 if (found_extent && item_size < sizeof(*ei))
5909                         found_extent = 0;
5910 #endif
5911                 if (!found_extent) {
5912                         BUG_ON(iref);
5913                         ret = remove_extent_backref(trans, extent_root, path,
5914                                                     NULL, refs_to_drop,
5915                                                     is_data, &last_ref);
5916                         if (ret) {
5917                                 btrfs_abort_transaction(trans, extent_root, ret);
5918                                 goto out;
5919                         }
5920                         btrfs_release_path(path);
5921                         path->leave_spinning = 1;
5922 
5923                         key.objectid = bytenr;
5924                         key.type = BTRFS_EXTENT_ITEM_KEY;
5925                         key.offset = num_bytes;
5926 
5927                         if (!is_data && skinny_metadata) {
5928                                 key.type = BTRFS_METADATA_ITEM_KEY;
5929                                 key.offset = owner_objectid;
5930                         }
5931 
5932                         ret = btrfs_search_slot(trans, extent_root,
5933                                                 &key, path, -1, 1);
5934                         if (ret > 0 && skinny_metadata && path->slots[0]) {
5935                                 /*
5936                                  * Couldn't find our skinny metadata item,
5937                                  * see if we have ye olde extent item.
5938                                  */
5939                                 path->slots[0]--;
5940                                 btrfs_item_key_to_cpu(path->nodes[0], &key,
5941                                                       path->slots[0]);
5942                                 if (key.objectid == bytenr &&
5943                                     key.type == BTRFS_EXTENT_ITEM_KEY &&
5944                                     key.offset == num_bytes)
5945                                         ret = 0;
5946                         }
5947 
5948                         if (ret > 0 && skinny_metadata) {
5949                                 skinny_metadata = false;
5950                                 key.objectid = bytenr;
5951                                 key.type = BTRFS_EXTENT_ITEM_KEY;
5952                                 key.offset = num_bytes;
5953                                 btrfs_release_path(path);
5954                                 ret = btrfs_search_slot(trans, extent_root,
5955                                                         &key, path, -1, 1);
5956                         }
5957 
5958                         if (ret) {
5959                                 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
5960                                         ret, bytenr);
5961                                 if (ret > 0)
5962                                         btrfs_print_leaf(extent_root,
5963                                                          path->nodes[0]);
5964                         }
5965                         if (ret < 0) {
5966                                 btrfs_abort_transaction(trans, extent_root, ret);