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

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

Version: ~ [ linux-5.10-rc5 ] ~ [ linux-5.9.10 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.79 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.159 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.208 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.245 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.245 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.85 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 /*
  2  * Copyright (C) 2007 Oracle.  All rights reserved.
  3  *
  4  * This program is free software; you can redistribute it and/or
  5  * modify it under the terms of the GNU General Public
  6  * License v2 as published by the Free Software Foundation.
  7  *
  8  * This program is distributed in the hope that it will be useful,
  9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
 10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 11  * General Public License for more details.
 12  *
 13  * You should have received a copy of the GNU General Public
 14  * License along with this program; if not, write to the
 15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
 16  * Boston, MA 021110-1307, USA.
 17  */
 18 #include <linux/sched.h>
 19 #include <linux/sched/signal.h>
 20 #include <linux/pagemap.h>
 21 #include <linux/writeback.h>
 22 #include <linux/blkdev.h>
 23 #include <linux/sort.h>
 24 #include <linux/rcupdate.h>
 25 #include <linux/kthread.h>
 26 #include <linux/slab.h>
 27 #include <linux/ratelimit.h>
 28 #include <linux/percpu_counter.h>
 29 #include "hash.h"
 30 #include "tree-log.h"
 31 #include "disk-io.h"
 32 #include "print-tree.h"
 33 #include "volumes.h"
 34 #include "raid56.h"
 35 #include "locking.h"
 36 #include "free-space-cache.h"
 37 #include "free-space-tree.h"
 38 #include "math.h"
 39 #include "sysfs.h"
 40 #include "qgroup.h"
 41 
 42 #undef SCRAMBLE_DELAYED_REFS
 43 
 44 /*
 45  * control flags for do_chunk_alloc's force field
 46  * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
 47  * if we really need one.
 48  *
 49  * CHUNK_ALLOC_LIMITED means to only try and allocate one
 50  * if we have very few chunks already allocated.  This is
 51  * used as part of the clustering code to help make sure
 52  * we have a good pool of storage to cluster in, without
 53  * filling the FS with empty chunks
 54  *
 55  * CHUNK_ALLOC_FORCE means it must try to allocate one
 56  *
 57  */
 58 enum {
 59         CHUNK_ALLOC_NO_FORCE = 0,
 60         CHUNK_ALLOC_LIMITED = 1,
 61         CHUNK_ALLOC_FORCE = 2,
 62 };
 63 
 64 static int update_block_group(struct btrfs_trans_handle *trans,
 65                               struct btrfs_fs_info *fs_info, u64 bytenr,
 66                               u64 num_bytes, int alloc);
 67 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
 68                                struct btrfs_fs_info *fs_info,
 69                                 struct btrfs_delayed_ref_node *node, u64 parent,
 70                                 u64 root_objectid, u64 owner_objectid,
 71                                 u64 owner_offset, int refs_to_drop,
 72                                 struct btrfs_delayed_extent_op *extra_op);
 73 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
 74                                     struct extent_buffer *leaf,
 75                                     struct btrfs_extent_item *ei);
 76 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
 77                                       struct btrfs_fs_info *fs_info,
 78                                       u64 parent, u64 root_objectid,
 79                                       u64 flags, u64 owner, u64 offset,
 80                                       struct btrfs_key *ins, int ref_mod);
 81 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
 82                                      struct btrfs_fs_info *fs_info,
 83                                      u64 parent, u64 root_objectid,
 84                                      u64 flags, struct btrfs_disk_key *key,
 85                                      int level, struct btrfs_key *ins);
 86 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
 87                           struct btrfs_fs_info *fs_info, u64 flags,
 88                           int force);
 89 static int find_next_key(struct btrfs_path *path, int level,
 90                          struct btrfs_key *key);
 91 static void dump_space_info(struct btrfs_fs_info *fs_info,
 92                             struct btrfs_space_info *info, u64 bytes,
 93                             int dump_block_groups);
 94 static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache *cache,
 95                                     u64 ram_bytes, u64 num_bytes, int delalloc);
 96 static int btrfs_free_reserved_bytes(struct btrfs_block_group_cache *cache,
 97                                      u64 num_bytes, int delalloc);
 98 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
 99                                u64 num_bytes);
100 static int __reserve_metadata_bytes(struct btrfs_fs_info *fs_info,
101                                     struct btrfs_space_info *space_info,
102                                     u64 orig_bytes,
103                                     enum btrfs_reserve_flush_enum flush,
104                                     bool system_chunk);
105 static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info,
106                                      struct btrfs_space_info *space_info,
107                                      u64 num_bytes);
108 static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info,
109                                      struct btrfs_space_info *space_info,
110                                      u64 num_bytes);
111 
112 static noinline int
113 block_group_cache_done(struct btrfs_block_group_cache *cache)
114 {
115         smp_mb();
116         return cache->cached == BTRFS_CACHE_FINISHED ||
117                 cache->cached == BTRFS_CACHE_ERROR;
118 }
119 
120 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
121 {
122         return (cache->flags & bits) == bits;
123 }
124 
125 void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
126 {
127         atomic_inc(&cache->count);
128 }
129 
130 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
131 {
132         if (atomic_dec_and_test(&cache->count)) {
133                 WARN_ON(cache->pinned > 0);
134                 WARN_ON(cache->reserved > 0);
135 
136                 /*
137                  * If not empty, someone is still holding mutex of
138                  * full_stripe_lock, which can only be released by caller.
139                  * And it will definitely cause use-after-free when caller
140                  * tries to release full stripe lock.
141                  *
142                  * No better way to resolve, but only to warn.
143                  */
144                 WARN_ON(!RB_EMPTY_ROOT(&cache->full_stripe_locks_root.root));
145                 kfree(cache->free_space_ctl);
146                 kfree(cache);
147         }
148 }
149 
150 /*
151  * this adds the block group to the fs_info rb tree for the block group
152  * cache
153  */
154 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
155                                 struct btrfs_block_group_cache *block_group)
156 {
157         struct rb_node **p;
158         struct rb_node *parent = NULL;
159         struct btrfs_block_group_cache *cache;
160 
161         spin_lock(&info->block_group_cache_lock);
162         p = &info->block_group_cache_tree.rb_node;
163 
164         while (*p) {
165                 parent = *p;
166                 cache = rb_entry(parent, struct btrfs_block_group_cache,
167                                  cache_node);
168                 if (block_group->key.objectid < cache->key.objectid) {
169                         p = &(*p)->rb_left;
170                 } else if (block_group->key.objectid > cache->key.objectid) {
171                         p = &(*p)->rb_right;
172                 } else {
173                         spin_unlock(&info->block_group_cache_lock);
174                         return -EEXIST;
175                 }
176         }
177 
178         rb_link_node(&block_group->cache_node, parent, p);
179         rb_insert_color(&block_group->cache_node,
180                         &info->block_group_cache_tree);
181 
182         if (info->first_logical_byte > block_group->key.objectid)
183                 info->first_logical_byte = block_group->key.objectid;
184 
185         spin_unlock(&info->block_group_cache_lock);
186 
187         return 0;
188 }
189 
190 /*
191  * This will return the block group at or after bytenr if contains is 0, else
192  * it will return the block group that contains the bytenr
193  */
194 static struct btrfs_block_group_cache *
195 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
196                               int contains)
197 {
198         struct btrfs_block_group_cache *cache, *ret = NULL;
199         struct rb_node *n;
200         u64 end, start;
201 
202         spin_lock(&info->block_group_cache_lock);
203         n = info->block_group_cache_tree.rb_node;
204 
205         while (n) {
206                 cache = rb_entry(n, struct btrfs_block_group_cache,
207                                  cache_node);
208                 end = cache->key.objectid + cache->key.offset - 1;
209                 start = cache->key.objectid;
210 
211                 if (bytenr < start) {
212                         if (!contains && (!ret || start < ret->key.objectid))
213                                 ret = cache;
214                         n = n->rb_left;
215                 } else if (bytenr > start) {
216                         if (contains && bytenr <= end) {
217                                 ret = cache;
218                                 break;
219                         }
220                         n = n->rb_right;
221                 } else {
222                         ret = cache;
223                         break;
224                 }
225         }
226         if (ret) {
227                 btrfs_get_block_group(ret);
228                 if (bytenr == 0 && info->first_logical_byte > ret->key.objectid)
229                         info->first_logical_byte = ret->key.objectid;
230         }
231         spin_unlock(&info->block_group_cache_lock);
232 
233         return ret;
234 }
235 
236 static int add_excluded_extent(struct btrfs_fs_info *fs_info,
237                                u64 start, u64 num_bytes)
238 {
239         u64 end = start + num_bytes - 1;
240         set_extent_bits(&fs_info->freed_extents[0],
241                         start, end, EXTENT_UPTODATE);
242         set_extent_bits(&fs_info->freed_extents[1],
243                         start, end, EXTENT_UPTODATE);
244         return 0;
245 }
246 
247 static void free_excluded_extents(struct btrfs_fs_info *fs_info,
248                                   struct btrfs_block_group_cache *cache)
249 {
250         u64 start, end;
251 
252         start = cache->key.objectid;
253         end = start + cache->key.offset - 1;
254 
255         clear_extent_bits(&fs_info->freed_extents[0],
256                           start, end, EXTENT_UPTODATE);
257         clear_extent_bits(&fs_info->freed_extents[1],
258                           start, end, EXTENT_UPTODATE);
259 }
260 
261 static int exclude_super_stripes(struct btrfs_fs_info *fs_info,
262                                  struct btrfs_block_group_cache *cache)
263 {
264         u64 bytenr;
265         u64 *logical;
266         int stripe_len;
267         int i, nr, ret;
268 
269         if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
270                 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
271                 cache->bytes_super += stripe_len;
272                 ret = add_excluded_extent(fs_info, cache->key.objectid,
273                                           stripe_len);
274                 if (ret)
275                         return ret;
276         }
277 
278         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
279                 bytenr = btrfs_sb_offset(i);
280                 ret = btrfs_rmap_block(fs_info, cache->key.objectid,
281                                        bytenr, 0, &logical, &nr, &stripe_len);
282                 if (ret)
283                         return ret;
284 
285                 while (nr--) {
286                         u64 start, len;
287 
288                         if (logical[nr] > cache->key.objectid +
289                             cache->key.offset)
290                                 continue;
291 
292                         if (logical[nr] + stripe_len <= cache->key.objectid)
293                                 continue;
294 
295                         start = logical[nr];
296                         if (start < cache->key.objectid) {
297                                 start = cache->key.objectid;
298                                 len = (logical[nr] + stripe_len) - start;
299                         } else {
300                                 len = min_t(u64, stripe_len,
301                                             cache->key.objectid +
302                                             cache->key.offset - start);
303                         }
304 
305                         cache->bytes_super += len;
306                         ret = add_excluded_extent(fs_info, start, len);
307                         if (ret) {
308                                 kfree(logical);
309                                 return ret;
310                         }
311                 }
312 
313                 kfree(logical);
314         }
315         return 0;
316 }
317 
318 static struct btrfs_caching_control *
319 get_caching_control(struct btrfs_block_group_cache *cache)
320 {
321         struct btrfs_caching_control *ctl;
322 
323         spin_lock(&cache->lock);
324         if (!cache->caching_ctl) {
325                 spin_unlock(&cache->lock);
326                 return NULL;
327         }
328 
329         ctl = cache->caching_ctl;
330         refcount_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 (refcount_dec_and_test(&ctl->count))
338                 kfree(ctl);
339 }
340 
341 #ifdef CONFIG_BTRFS_DEBUG
342 static void fragment_free_space(struct btrfs_block_group_cache *block_group)
343 {
344         struct btrfs_fs_info *fs_info = block_group->fs_info;
345         u64 start = block_group->key.objectid;
346         u64 len = block_group->key.offset;
347         u64 chunk = block_group->flags & BTRFS_BLOCK_GROUP_METADATA ?
348                 fs_info->nodesize : fs_info->sectorsize;
349         u64 step = chunk << 1;
350 
351         while (len > chunk) {
352                 btrfs_remove_free_space(block_group, start, chunk);
353                 start += step;
354                 if (len < step)
355                         len = 0;
356                 else
357                         len -= step;
358         }
359 }
360 #endif
361 
362 /*
363  * this is only called by cache_block_group, since we could have freed extents
364  * we need to check the pinned_extents for any extents that can't be used yet
365  * since their free space will be released as soon as the transaction commits.
366  */
367 u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
368                        struct btrfs_fs_info *info, u64 start, u64 end)
369 {
370         u64 extent_start, extent_end, size, total_added = 0;
371         int ret;
372 
373         while (start < end) {
374                 ret = find_first_extent_bit(info->pinned_extents, start,
375                                             &extent_start, &extent_end,
376                                             EXTENT_DIRTY | EXTENT_UPTODATE,
377                                             NULL);
378                 if (ret)
379                         break;
380 
381                 if (extent_start <= start) {
382                         start = extent_end + 1;
383                 } else if (extent_start > start && extent_start < end) {
384                         size = extent_start - start;
385                         total_added += size;
386                         ret = btrfs_add_free_space(block_group, start,
387                                                    size);
388                         BUG_ON(ret); /* -ENOMEM or logic error */
389                         start = extent_end + 1;
390                 } else {
391                         break;
392                 }
393         }
394 
395         if (start < end) {
396                 size = end - start;
397                 total_added += size;
398                 ret = btrfs_add_free_space(block_group, start, size);
399                 BUG_ON(ret); /* -ENOMEM or logic error */
400         }
401 
402         return total_added;
403 }
404 
405 static int load_extent_tree_free(struct btrfs_caching_control *caching_ctl)
406 {
407         struct btrfs_block_group_cache *block_group = caching_ctl->block_group;
408         struct btrfs_fs_info *fs_info = block_group->fs_info;
409         struct btrfs_root *extent_root = fs_info->extent_root;
410         struct btrfs_path *path;
411         struct extent_buffer *leaf;
412         struct btrfs_key key;
413         u64 total_found = 0;
414         u64 last = 0;
415         u32 nritems;
416         int ret;
417         bool wakeup = true;
418 
419         path = btrfs_alloc_path();
420         if (!path)
421                 return -ENOMEM;
422 
423         last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
424 
425 #ifdef CONFIG_BTRFS_DEBUG
426         /*
427          * If we're fragmenting we don't want to make anybody think we can
428          * allocate from this block group until we've had a chance to fragment
429          * the free space.
430          */
431         if (btrfs_should_fragment_free_space(block_group))
432                 wakeup = false;
433 #endif
434         /*
435          * We don't want to deadlock with somebody trying to allocate a new
436          * extent for the extent root while also trying to search the extent
437          * root to add free space.  So we skip locking and search the commit
438          * root, since its read-only
439          */
440         path->skip_locking = 1;
441         path->search_commit_root = 1;
442         path->reada = READA_FORWARD;
443 
444         key.objectid = last;
445         key.offset = 0;
446         key.type = BTRFS_EXTENT_ITEM_KEY;
447 
448 next:
449         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
450         if (ret < 0)
451                 goto out;
452 
453         leaf = path->nodes[0];
454         nritems = btrfs_header_nritems(leaf);
455 
456         while (1) {
457                 if (btrfs_fs_closing(fs_info) > 1) {
458                         last = (u64)-1;
459                         break;
460                 }
461 
462                 if (path->slots[0] < nritems) {
463                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
464                 } else {
465                         ret = find_next_key(path, 0, &key);
466                         if (ret)
467                                 break;
468 
469                         if (need_resched() ||
470                             rwsem_is_contended(&fs_info->commit_root_sem)) {
471                                 if (wakeup)
472                                         caching_ctl->progress = last;
473                                 btrfs_release_path(path);
474                                 up_read(&fs_info->commit_root_sem);
475                                 mutex_unlock(&caching_ctl->mutex);
476                                 cond_resched();
477                                 mutex_lock(&caching_ctl->mutex);
478                                 down_read(&fs_info->commit_root_sem);
479                                 goto next;
480                         }
481 
482                         ret = btrfs_next_leaf(extent_root, path);
483                         if (ret < 0)
484                                 goto out;
485                         if (ret)
486                                 break;
487                         leaf = path->nodes[0];
488                         nritems = btrfs_header_nritems(leaf);
489                         continue;
490                 }
491 
492                 if (key.objectid < last) {
493                         key.objectid = last;
494                         key.offset = 0;
495                         key.type = BTRFS_EXTENT_ITEM_KEY;
496 
497                         if (wakeup)
498                                 caching_ctl->progress = last;
499                         btrfs_release_path(path);
500                         goto next;
501                 }
502 
503                 if (key.objectid < block_group->key.objectid) {
504                         path->slots[0]++;
505                         continue;
506                 }
507 
508                 if (key.objectid >= block_group->key.objectid +
509                     block_group->key.offset)
510                         break;
511 
512                 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
513                     key.type == BTRFS_METADATA_ITEM_KEY) {
514                         total_found += add_new_free_space(block_group,
515                                                           fs_info, last,
516                                                           key.objectid);
517                         if (key.type == BTRFS_METADATA_ITEM_KEY)
518                                 last = key.objectid +
519                                         fs_info->nodesize;
520                         else
521                                 last = key.objectid + key.offset;
522 
523                         if (total_found > CACHING_CTL_WAKE_UP) {
524                                 total_found = 0;
525                                 if (wakeup)
526                                         wake_up(&caching_ctl->wait);
527                         }
528                 }
529                 path->slots[0]++;
530         }
531         ret = 0;
532 
533         total_found += add_new_free_space(block_group, fs_info, last,
534                                           block_group->key.objectid +
535                                           block_group->key.offset);
536         caching_ctl->progress = (u64)-1;
537 
538 out:
539         btrfs_free_path(path);
540         return ret;
541 }
542 
543 static noinline void caching_thread(struct btrfs_work *work)
544 {
545         struct btrfs_block_group_cache *block_group;
546         struct btrfs_fs_info *fs_info;
547         struct btrfs_caching_control *caching_ctl;
548         struct btrfs_root *extent_root;
549         int ret;
550 
551         caching_ctl = container_of(work, struct btrfs_caching_control, work);
552         block_group = caching_ctl->block_group;
553         fs_info = block_group->fs_info;
554         extent_root = fs_info->extent_root;
555 
556         mutex_lock(&caching_ctl->mutex);
557         down_read(&fs_info->commit_root_sem);
558 
559         if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE))
560                 ret = load_free_space_tree(caching_ctl);
561         else
562                 ret = load_extent_tree_free(caching_ctl);
563 
564         spin_lock(&block_group->lock);
565         block_group->caching_ctl = NULL;
566         block_group->cached = ret ? BTRFS_CACHE_ERROR : BTRFS_CACHE_FINISHED;
567         spin_unlock(&block_group->lock);
568 
569 #ifdef CONFIG_BTRFS_DEBUG
570         if (btrfs_should_fragment_free_space(block_group)) {
571                 u64 bytes_used;
572 
573                 spin_lock(&block_group->space_info->lock);
574                 spin_lock(&block_group->lock);
575                 bytes_used = block_group->key.offset -
576                         btrfs_block_group_used(&block_group->item);
577                 block_group->space_info->bytes_used += bytes_used >> 1;
578                 spin_unlock(&block_group->lock);
579                 spin_unlock(&block_group->space_info->lock);
580                 fragment_free_space(block_group);
581         }
582 #endif
583 
584         caching_ctl->progress = (u64)-1;
585 
586         up_read(&fs_info->commit_root_sem);
587         free_excluded_extents(fs_info, block_group);
588         mutex_unlock(&caching_ctl->mutex);
589 
590         wake_up(&caching_ctl->wait);
591 
592         put_caching_control(caching_ctl);
593         btrfs_put_block_group(block_group);
594 }
595 
596 static int cache_block_group(struct btrfs_block_group_cache *cache,
597                              int load_cache_only)
598 {
599         DEFINE_WAIT(wait);
600         struct btrfs_fs_info *fs_info = cache->fs_info;
601         struct btrfs_caching_control *caching_ctl;
602         int ret = 0;
603 
604         caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
605         if (!caching_ctl)
606                 return -ENOMEM;
607 
608         INIT_LIST_HEAD(&caching_ctl->list);
609         mutex_init(&caching_ctl->mutex);
610         init_waitqueue_head(&caching_ctl->wait);
611         caching_ctl->block_group = cache;
612         caching_ctl->progress = cache->key.objectid;
613         refcount_set(&caching_ctl->count, 1);
614         btrfs_init_work(&caching_ctl->work, btrfs_cache_helper,
615                         caching_thread, NULL, NULL);
616 
617         spin_lock(&cache->lock);
618         /*
619          * This should be a rare occasion, but this could happen I think in the
620          * case where one thread starts to load the space cache info, and then
621          * some other thread starts a transaction commit which tries to do an
622          * allocation while the other thread is still loading the space cache
623          * info.  The previous loop should have kept us from choosing this block
624          * group, but if we've moved to the state where we will wait on caching
625          * block groups we need to first check if we're doing a fast load here,
626          * so we can wait for it to finish, otherwise we could end up allocating
627          * from a block group who's cache gets evicted for one reason or
628          * another.
629          */
630         while (cache->cached == BTRFS_CACHE_FAST) {
631                 struct btrfs_caching_control *ctl;
632 
633                 ctl = cache->caching_ctl;
634                 refcount_inc(&ctl->count);
635                 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
636                 spin_unlock(&cache->lock);
637 
638                 schedule();
639 
640                 finish_wait(&ctl->wait, &wait);
641                 put_caching_control(ctl);
642                 spin_lock(&cache->lock);
643         }
644 
645         if (cache->cached != BTRFS_CACHE_NO) {
646                 spin_unlock(&cache->lock);
647                 kfree(caching_ctl);
648                 return 0;
649         }
650         WARN_ON(cache->caching_ctl);
651         cache->caching_ctl = caching_ctl;
652         cache->cached = BTRFS_CACHE_FAST;
653         spin_unlock(&cache->lock);
654 
655         if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) {
656                 mutex_lock(&caching_ctl->mutex);
657                 ret = load_free_space_cache(fs_info, cache);
658 
659                 spin_lock(&cache->lock);
660                 if (ret == 1) {
661                         cache->caching_ctl = NULL;
662                         cache->cached = BTRFS_CACHE_FINISHED;
663                         cache->last_byte_to_unpin = (u64)-1;
664                         caching_ctl->progress = (u64)-1;
665                 } else {
666                         if (load_cache_only) {
667                                 cache->caching_ctl = NULL;
668                                 cache->cached = BTRFS_CACHE_NO;
669                         } else {
670                                 cache->cached = BTRFS_CACHE_STARTED;
671                                 cache->has_caching_ctl = 1;
672                         }
673                 }
674                 spin_unlock(&cache->lock);
675 #ifdef CONFIG_BTRFS_DEBUG
676                 if (ret == 1 &&
677                     btrfs_should_fragment_free_space(cache)) {
678                         u64 bytes_used;
679 
680                         spin_lock(&cache->space_info->lock);
681                         spin_lock(&cache->lock);
682                         bytes_used = cache->key.offset -
683                                 btrfs_block_group_used(&cache->item);
684                         cache->space_info->bytes_used += bytes_used >> 1;
685                         spin_unlock(&cache->lock);
686                         spin_unlock(&cache->space_info->lock);
687                         fragment_free_space(cache);
688                 }
689 #endif
690                 mutex_unlock(&caching_ctl->mutex);
691 
692                 wake_up(&caching_ctl->wait);
693                 if (ret == 1) {
694                         put_caching_control(caching_ctl);
695                         free_excluded_extents(fs_info, cache);
696                         return 0;
697                 }
698         } else {
699                 /*
700                  * We're either using the free space tree or no caching at all.
701                  * Set cached to the appropriate value and wakeup any waiters.
702                  */
703                 spin_lock(&cache->lock);
704                 if (load_cache_only) {
705                         cache->caching_ctl = NULL;
706                         cache->cached = BTRFS_CACHE_NO;
707                 } else {
708                         cache->cached = BTRFS_CACHE_STARTED;
709                         cache->has_caching_ctl = 1;
710                 }
711                 spin_unlock(&cache->lock);
712                 wake_up(&caching_ctl->wait);
713         }
714 
715         if (load_cache_only) {
716                 put_caching_control(caching_ctl);
717                 return 0;
718         }
719 
720         down_write(&fs_info->commit_root_sem);
721         refcount_inc(&caching_ctl->count);
722         list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
723         up_write(&fs_info->commit_root_sem);
724 
725         btrfs_get_block_group(cache);
726 
727         btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work);
728 
729         return ret;
730 }
731 
732 /*
733  * return the block group that starts at or after bytenr
734  */
735 static struct btrfs_block_group_cache *
736 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
737 {
738         return block_group_cache_tree_search(info, bytenr, 0);
739 }
740 
741 /*
742  * return the block group that contains the given bytenr
743  */
744 struct btrfs_block_group_cache *btrfs_lookup_block_group(
745                                                  struct btrfs_fs_info *info,
746                                                  u64 bytenr)
747 {
748         return block_group_cache_tree_search(info, bytenr, 1);
749 }
750 
751 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
752                                                   u64 flags)
753 {
754         struct list_head *head = &info->space_info;
755         struct btrfs_space_info *found;
756 
757         flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
758 
759         rcu_read_lock();
760         list_for_each_entry_rcu(found, head, list) {
761                 if (found->flags & flags) {
762                         rcu_read_unlock();
763                         return found;
764                 }
765         }
766         rcu_read_unlock();
767         return NULL;
768 }
769 
770 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, s64 num_bytes,
771                              u64 owner, u64 root_objectid)
772 {
773         struct btrfs_space_info *space_info;
774         u64 flags;
775 
776         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
777                 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
778                         flags = BTRFS_BLOCK_GROUP_SYSTEM;
779                 else
780                         flags = BTRFS_BLOCK_GROUP_METADATA;
781         } else {
782                 flags = BTRFS_BLOCK_GROUP_DATA;
783         }
784 
785         space_info = __find_space_info(fs_info, flags);
786         ASSERT(space_info);
787         percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
788 }
789 
790 /*
791  * after adding space to the filesystem, we need to clear the full flags
792  * on all the space infos.
793  */
794 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
795 {
796         struct list_head *head = &info->space_info;
797         struct btrfs_space_info *found;
798 
799         rcu_read_lock();
800         list_for_each_entry_rcu(found, head, list)
801                 found->full = 0;
802         rcu_read_unlock();
803 }
804 
805 /* simple helper to search for an existing data extent at a given offset */
806 int btrfs_lookup_data_extent(struct btrfs_fs_info *fs_info, u64 start, u64 len)
807 {
808         int ret;
809         struct btrfs_key key;
810         struct btrfs_path *path;
811 
812         path = btrfs_alloc_path();
813         if (!path)
814                 return -ENOMEM;
815 
816         key.objectid = start;
817         key.offset = len;
818         key.type = BTRFS_EXTENT_ITEM_KEY;
819         ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
820         btrfs_free_path(path);
821         return ret;
822 }
823 
824 /*
825  * helper function to lookup reference count and flags of a tree block.
826  *
827  * the head node for delayed ref is used to store the sum of all the
828  * reference count modifications queued up in the rbtree. the head
829  * node may also store the extent flags to set. This way you can check
830  * to see what the reference count and extent flags would be if all of
831  * the delayed refs are not processed.
832  */
833 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
834                              struct btrfs_fs_info *fs_info, u64 bytenr,
835                              u64 offset, int metadata, u64 *refs, u64 *flags)
836 {
837         struct btrfs_delayed_ref_head *head;
838         struct btrfs_delayed_ref_root *delayed_refs;
839         struct btrfs_path *path;
840         struct btrfs_extent_item *ei;
841         struct extent_buffer *leaf;
842         struct btrfs_key key;
843         u32 item_size;
844         u64 num_refs;
845         u64 extent_flags;
846         int ret;
847 
848         /*
849          * If we don't have skinny metadata, don't bother doing anything
850          * different
851          */
852         if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) {
853                 offset = fs_info->nodesize;
854                 metadata = 0;
855         }
856 
857         path = btrfs_alloc_path();
858         if (!path)
859                 return -ENOMEM;
860 
861         if (!trans) {
862                 path->skip_locking = 1;
863                 path->search_commit_root = 1;
864         }
865 
866 search_again:
867         key.objectid = bytenr;
868         key.offset = offset;
869         if (metadata)
870                 key.type = BTRFS_METADATA_ITEM_KEY;
871         else
872                 key.type = BTRFS_EXTENT_ITEM_KEY;
873 
874         ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0);
875         if (ret < 0)
876                 goto out_free;
877 
878         if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
879                 if (path->slots[0]) {
880                         path->slots[0]--;
881                         btrfs_item_key_to_cpu(path->nodes[0], &key,
882                                               path->slots[0]);
883                         if (key.objectid == bytenr &&
884                             key.type == BTRFS_EXTENT_ITEM_KEY &&
885                             key.offset == fs_info->nodesize)
886                                 ret = 0;
887                 }
888         }
889 
890         if (ret == 0) {
891                 leaf = path->nodes[0];
892                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
893                 if (item_size >= sizeof(*ei)) {
894                         ei = btrfs_item_ptr(leaf, path->slots[0],
895                                             struct btrfs_extent_item);
896                         num_refs = btrfs_extent_refs(leaf, ei);
897                         extent_flags = btrfs_extent_flags(leaf, ei);
898                 } else {
899 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
900                         struct btrfs_extent_item_v0 *ei0;
901                         BUG_ON(item_size != sizeof(*ei0));
902                         ei0 = btrfs_item_ptr(leaf, path->slots[0],
903                                              struct btrfs_extent_item_v0);
904                         num_refs = btrfs_extent_refs_v0(leaf, ei0);
905                         /* FIXME: this isn't correct for data */
906                         extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
907 #else
908                         BUG();
909 #endif
910                 }
911                 BUG_ON(num_refs == 0);
912         } else {
913                 num_refs = 0;
914                 extent_flags = 0;
915                 ret = 0;
916         }
917 
918         if (!trans)
919                 goto out;
920 
921         delayed_refs = &trans->transaction->delayed_refs;
922         spin_lock(&delayed_refs->lock);
923         head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
924         if (head) {
925                 if (!mutex_trylock(&head->mutex)) {
926                         refcount_inc(&head->node.refs);
927                         spin_unlock(&delayed_refs->lock);
928 
929                         btrfs_release_path(path);
930 
931                         /*
932                          * Mutex was contended, block until it's released and try
933                          * again
934                          */
935                         mutex_lock(&head->mutex);
936                         mutex_unlock(&head->mutex);
937                         btrfs_put_delayed_ref(&head->node);
938                         goto search_again;
939                 }
940                 spin_lock(&head->lock);
941                 if (head->extent_op && head->extent_op->update_flags)
942                         extent_flags |= head->extent_op->flags_to_set;
943                 else
944                         BUG_ON(num_refs == 0);
945 
946                 num_refs += head->node.ref_mod;
947                 spin_unlock(&head->lock);
948                 mutex_unlock(&head->mutex);
949         }
950         spin_unlock(&delayed_refs->lock);
951 out:
952         WARN_ON(num_refs == 0);
953         if (refs)
954                 *refs = num_refs;
955         if (flags)
956                 *flags = extent_flags;
957 out_free:
958         btrfs_free_path(path);
959         return ret;
960 }
961 
962 /*
963  * Back reference rules.  Back refs have three main goals:
964  *
965  * 1) differentiate between all holders of references to an extent so that
966  *    when a reference is dropped we can make sure it was a valid reference
967  *    before freeing the extent.
968  *
969  * 2) Provide enough information to quickly find the holders of an extent
970  *    if we notice a given block is corrupted or bad.
971  *
972  * 3) Make it easy to migrate blocks for FS shrinking or storage pool
973  *    maintenance.  This is actually the same as #2, but with a slightly
974  *    different use case.
975  *
976  * There are two kinds of back refs. The implicit back refs is optimized
977  * for pointers in non-shared tree blocks. For a given pointer in a block,
978  * back refs of this kind provide information about the block's owner tree
979  * and the pointer's key. These information allow us to find the block by
980  * b-tree searching. The full back refs is for pointers in tree blocks not
981  * referenced by their owner trees. The location of tree block is recorded
982  * in the back refs. Actually the full back refs is generic, and can be
983  * used in all cases the implicit back refs is used. The major shortcoming
984  * of the full back refs is its overhead. Every time a tree block gets
985  * COWed, we have to update back refs entry for all pointers in it.
986  *
987  * For a newly allocated tree block, we use implicit back refs for
988  * pointers in it. This means most tree related operations only involve
989  * implicit back refs. For a tree block created in old transaction, the
990  * only way to drop a reference to it is COW it. So we can detect the
991  * event that tree block loses its owner tree's reference and do the
992  * back refs conversion.
993  *
994  * When a tree block is COWed through a tree, there are four cases:
995  *
996  * The reference count of the block is one and the tree is the block's
997  * owner tree. Nothing to do in this case.
998  *
999  * The reference count of the block is one and the tree is not the
1000  * block's owner tree. In this case, full back refs is used for pointers
1001  * in the block. Remove these full back refs, add implicit back refs for
1002  * every pointers in the new block.
1003  *
1004  * The reference count of the block is greater than one and the tree is
1005  * the block's owner tree. In this case, implicit back refs is used for
1006  * pointers in the block. Add full back refs for every pointers in the
1007  * block, increase lower level extents' reference counts. The original
1008  * implicit back refs are entailed to the new block.
1009  *
1010  * The reference count of the block is greater than one and the tree is
1011  * not the block's owner tree. Add implicit back refs for every pointer in
1012  * the new block, increase lower level extents' reference count.
1013  *
1014  * Back Reference Key composing:
1015  *
1016  * The key objectid corresponds to the first byte in the extent,
1017  * The key type is used to differentiate between types of back refs.
1018  * There are different meanings of the key offset for different types
1019  * of back refs.
1020  *
1021  * File extents can be referenced by:
1022  *
1023  * - multiple snapshots, subvolumes, or different generations in one subvol
1024  * - different files inside a single subvolume
1025  * - different offsets inside a file (bookend extents in file.c)
1026  *
1027  * The extent ref structure for the implicit back refs has fields for:
1028  *
1029  * - Objectid of the subvolume root
1030  * - objectid of the file holding the reference
1031  * - original offset in the file
1032  * - how many bookend extents
1033  *
1034  * The key offset for the implicit back refs is hash of the first
1035  * three fields.
1036  *
1037  * The extent ref structure for the full back refs has field for:
1038  *
1039  * - number of pointers in the tree leaf
1040  *
1041  * The key offset for the implicit back refs is the first byte of
1042  * the tree leaf
1043  *
1044  * When a file extent is allocated, The implicit back refs is used.
1045  * the fields are filled in:
1046  *
1047  *     (root_key.objectid, inode objectid, offset in file, 1)
1048  *
1049  * When a file extent is removed file truncation, we find the
1050  * corresponding implicit back refs and check the following fields:
1051  *
1052  *     (btrfs_header_owner(leaf), inode objectid, offset in file)
1053  *
1054  * Btree extents can be referenced by:
1055  *
1056  * - Different subvolumes
1057  *
1058  * Both the implicit back refs and the full back refs for tree blocks
1059  * only consist of key. The key offset for the implicit back refs is
1060  * objectid of block's owner tree. The key offset for the full back refs
1061  * is the first byte of parent block.
1062  *
1063  * When implicit back refs is used, information about the lowest key and
1064  * level of the tree block are required. These information are stored in
1065  * tree block info structure.
1066  */
1067 
1068 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1069 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
1070                                   struct btrfs_fs_info *fs_info,
1071                                   struct btrfs_path *path,
1072                                   u64 owner, u32 extra_size)
1073 {
1074         struct btrfs_root *root = fs_info->extent_root;
1075         struct btrfs_extent_item *item;
1076         struct btrfs_extent_item_v0 *ei0;
1077         struct btrfs_extent_ref_v0 *ref0;
1078         struct btrfs_tree_block_info *bi;
1079         struct extent_buffer *leaf;
1080         struct btrfs_key key;
1081         struct btrfs_key found_key;
1082         u32 new_size = sizeof(*item);
1083         u64 refs;
1084         int ret;
1085 
1086         leaf = path->nodes[0];
1087         BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
1088 
1089         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1090         ei0 = btrfs_item_ptr(leaf, path->slots[0],
1091                              struct btrfs_extent_item_v0);
1092         refs = btrfs_extent_refs_v0(leaf, ei0);
1093 
1094         if (owner == (u64)-1) {
1095                 while (1) {
1096                         if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1097                                 ret = btrfs_next_leaf(root, path);
1098                                 if (ret < 0)
1099                                         return ret;
1100                                 BUG_ON(ret > 0); /* Corruption */
1101                                 leaf = path->nodes[0];
1102                         }
1103                         btrfs_item_key_to_cpu(leaf, &found_key,
1104                                               path->slots[0]);
1105                         BUG_ON(key.objectid != found_key.objectid);
1106                         if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
1107                                 path->slots[0]++;
1108                                 continue;
1109                         }
1110                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
1111                                               struct btrfs_extent_ref_v0);
1112                         owner = btrfs_ref_objectid_v0(leaf, ref0);
1113                         break;
1114                 }
1115         }
1116         btrfs_release_path(path);
1117 
1118         if (owner < BTRFS_FIRST_FREE_OBJECTID)
1119                 new_size += sizeof(*bi);
1120 
1121         new_size -= sizeof(*ei0);
1122         ret = btrfs_search_slot(trans, root, &key, path,
1123                                 new_size + extra_size, 1);
1124         if (ret < 0)
1125                 return ret;
1126         BUG_ON(ret); /* Corruption */
1127 
1128         btrfs_extend_item(fs_info, path, new_size);
1129 
1130         leaf = path->nodes[0];
1131         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1132         btrfs_set_extent_refs(leaf, item, refs);
1133         /* FIXME: get real generation */
1134         btrfs_set_extent_generation(leaf, item, 0);
1135         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1136                 btrfs_set_extent_flags(leaf, item,
1137                                        BTRFS_EXTENT_FLAG_TREE_BLOCK |
1138                                        BTRFS_BLOCK_FLAG_FULL_BACKREF);
1139                 bi = (struct btrfs_tree_block_info *)(item + 1);
1140                 /* FIXME: get first key of the block */
1141                 memzero_extent_buffer(leaf, (unsigned long)bi, sizeof(*bi));
1142                 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1143         } else {
1144                 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1145         }
1146         btrfs_mark_buffer_dirty(leaf);
1147         return 0;
1148 }
1149 #endif
1150 
1151 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1152 {
1153         u32 high_crc = ~(u32)0;
1154         u32 low_crc = ~(u32)0;
1155         __le64 lenum;
1156 
1157         lenum = cpu_to_le64(root_objectid);
1158         high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum));
1159         lenum = cpu_to_le64(owner);
1160         low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1161         lenum = cpu_to_le64(offset);
1162         low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1163 
1164         return ((u64)high_crc << 31) ^ (u64)low_crc;
1165 }
1166 
1167 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1168                                      struct btrfs_extent_data_ref *ref)
1169 {
1170         return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1171                                     btrfs_extent_data_ref_objectid(leaf, ref),
1172                                     btrfs_extent_data_ref_offset(leaf, ref));
1173 }
1174 
1175 static int match_extent_data_ref(struct extent_buffer *leaf,
1176                                  struct btrfs_extent_data_ref *ref,
1177                                  u64 root_objectid, u64 owner, u64 offset)
1178 {
1179         if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1180             btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1181             btrfs_extent_data_ref_offset(leaf, ref) != offset)
1182                 return 0;
1183         return 1;
1184 }
1185 
1186 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1187                                            struct btrfs_fs_info *fs_info,
1188                                            struct btrfs_path *path,
1189                                            u64 bytenr, u64 parent,
1190                                            u64 root_objectid,
1191                                            u64 owner, u64 offset)
1192 {
1193         struct btrfs_root *root = fs_info->extent_root;
1194         struct btrfs_key key;
1195         struct btrfs_extent_data_ref *ref;
1196         struct extent_buffer *leaf;
1197         u32 nritems;
1198         int ret;
1199         int recow;
1200         int err = -ENOENT;
1201 
1202         key.objectid = bytenr;
1203         if (parent) {
1204                 key.type = BTRFS_SHARED_DATA_REF_KEY;
1205                 key.offset = parent;
1206         } else {
1207                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1208                 key.offset = hash_extent_data_ref(root_objectid,
1209                                                   owner, offset);
1210         }
1211 again:
1212         recow = 0;
1213         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1214         if (ret < 0) {
1215                 err = ret;
1216                 goto fail;
1217         }
1218 
1219         if (parent) {
1220                 if (!ret)
1221                         return 0;
1222 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1223                 key.type = BTRFS_EXTENT_REF_V0_KEY;
1224                 btrfs_release_path(path);
1225                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1226                 if (ret < 0) {
1227                         err = ret;
1228                         goto fail;
1229                 }
1230                 if (!ret)
1231                         return 0;
1232 #endif
1233                 goto fail;
1234         }
1235 
1236         leaf = path->nodes[0];
1237         nritems = btrfs_header_nritems(leaf);
1238         while (1) {
1239                 if (path->slots[0] >= nritems) {
1240                         ret = btrfs_next_leaf(root, path);
1241                         if (ret < 0)
1242                                 err = ret;
1243                         if (ret)
1244                                 goto fail;
1245 
1246                         leaf = path->nodes[0];
1247                         nritems = btrfs_header_nritems(leaf);
1248                         recow = 1;
1249                 }
1250 
1251                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1252                 if (key.objectid != bytenr ||
1253                     key.type != BTRFS_EXTENT_DATA_REF_KEY)
1254                         goto fail;
1255 
1256                 ref = btrfs_item_ptr(leaf, path->slots[0],
1257                                      struct btrfs_extent_data_ref);
1258 
1259                 if (match_extent_data_ref(leaf, ref, root_objectid,
1260                                           owner, offset)) {
1261                         if (recow) {
1262                                 btrfs_release_path(path);
1263                                 goto again;
1264                         }
1265                         err = 0;
1266                         break;
1267                 }
1268                 path->slots[0]++;
1269         }
1270 fail:
1271         return err;
1272 }
1273 
1274 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1275                                            struct btrfs_fs_info *fs_info,
1276                                            struct btrfs_path *path,
1277                                            u64 bytenr, u64 parent,
1278                                            u64 root_objectid, u64 owner,
1279                                            u64 offset, int refs_to_add)
1280 {
1281         struct btrfs_root *root = fs_info->extent_root;
1282         struct btrfs_key key;
1283         struct extent_buffer *leaf;
1284         u32 size;
1285         u32 num_refs;
1286         int ret;
1287 
1288         key.objectid = bytenr;
1289         if (parent) {
1290                 key.type = BTRFS_SHARED_DATA_REF_KEY;
1291                 key.offset = parent;
1292                 size = sizeof(struct btrfs_shared_data_ref);
1293         } else {
1294                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1295                 key.offset = hash_extent_data_ref(root_objectid,
1296                                                   owner, offset);
1297                 size = sizeof(struct btrfs_extent_data_ref);
1298         }
1299 
1300         ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1301         if (ret && ret != -EEXIST)
1302                 goto fail;
1303 
1304         leaf = path->nodes[0];
1305         if (parent) {
1306                 struct btrfs_shared_data_ref *ref;
1307                 ref = btrfs_item_ptr(leaf, path->slots[0],
1308                                      struct btrfs_shared_data_ref);
1309                 if (ret == 0) {
1310                         btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1311                 } else {
1312                         num_refs = btrfs_shared_data_ref_count(leaf, ref);
1313                         num_refs += refs_to_add;
1314                         btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1315                 }
1316         } else {
1317                 struct btrfs_extent_data_ref *ref;
1318                 while (ret == -EEXIST) {
1319                         ref = btrfs_item_ptr(leaf, path->slots[0],
1320                                              struct btrfs_extent_data_ref);
1321                         if (match_extent_data_ref(leaf, ref, root_objectid,
1322                                                   owner, offset))
1323                                 break;
1324                         btrfs_release_path(path);
1325                         key.offset++;
1326                         ret = btrfs_insert_empty_item(trans, root, path, &key,
1327                                                       size);
1328                         if (ret && ret != -EEXIST)
1329                                 goto fail;
1330 
1331                         leaf = path->nodes[0];
1332                 }
1333                 ref = btrfs_item_ptr(leaf, path->slots[0],
1334                                      struct btrfs_extent_data_ref);
1335                 if (ret == 0) {
1336                         btrfs_set_extent_data_ref_root(leaf, ref,
1337                                                        root_objectid);
1338                         btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1339                         btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1340                         btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1341                 } else {
1342                         num_refs = btrfs_extent_data_ref_count(leaf, ref);
1343                         num_refs += refs_to_add;
1344                         btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1345                 }
1346         }
1347         btrfs_mark_buffer_dirty(leaf);
1348         ret = 0;
1349 fail:
1350         btrfs_release_path(path);
1351         return ret;
1352 }
1353 
1354 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1355                                            struct btrfs_fs_info *fs_info,
1356                                            struct btrfs_path *path,
1357                                            int refs_to_drop, int *last_ref)
1358 {
1359         struct btrfs_key key;
1360         struct btrfs_extent_data_ref *ref1 = NULL;
1361         struct btrfs_shared_data_ref *ref2 = NULL;
1362         struct extent_buffer *leaf;
1363         u32 num_refs = 0;
1364         int ret = 0;
1365 
1366         leaf = path->nodes[0];
1367         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1368 
1369         if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1370                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1371                                       struct btrfs_extent_data_ref);
1372                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1373         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1374                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1375                                       struct btrfs_shared_data_ref);
1376                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1377 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1378         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1379                 struct btrfs_extent_ref_v0 *ref0;
1380                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1381                                       struct btrfs_extent_ref_v0);
1382                 num_refs = btrfs_ref_count_v0(leaf, ref0);
1383 #endif
1384         } else {
1385                 BUG();
1386         }
1387 
1388         BUG_ON(num_refs < refs_to_drop);
1389         num_refs -= refs_to_drop;
1390 
1391         if (num_refs == 0) {
1392                 ret = btrfs_del_item(trans, fs_info->extent_root, path);
1393                 *last_ref = 1;
1394         } else {
1395                 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1396                         btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1397                 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1398                         btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1399 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1400                 else {
1401                         struct btrfs_extent_ref_v0 *ref0;
1402                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
1403                                         struct btrfs_extent_ref_v0);
1404                         btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1405                 }
1406 #endif
1407                 btrfs_mark_buffer_dirty(leaf);
1408         }
1409         return ret;
1410 }
1411 
1412 static noinline u32 extent_data_ref_count(struct btrfs_path *path,
1413                                           struct btrfs_extent_inline_ref *iref)
1414 {
1415         struct btrfs_key key;
1416         struct extent_buffer *leaf;
1417         struct btrfs_extent_data_ref *ref1;
1418         struct btrfs_shared_data_ref *ref2;
1419         u32 num_refs = 0;
1420 
1421         leaf = path->nodes[0];
1422         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1423         if (iref) {
1424                 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1425                     BTRFS_EXTENT_DATA_REF_KEY) {
1426                         ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1427                         num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1428                 } else {
1429                         ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1430                         num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1431                 }
1432         } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1433                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1434                                       struct btrfs_extent_data_ref);
1435                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1436         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1437                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1438                                       struct btrfs_shared_data_ref);
1439                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1440 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1441         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1442                 struct btrfs_extent_ref_v0 *ref0;
1443                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1444                                       struct btrfs_extent_ref_v0);
1445                 num_refs = btrfs_ref_count_v0(leaf, ref0);
1446 #endif
1447         } else {
1448                 WARN_ON(1);
1449         }
1450         return num_refs;
1451 }
1452 
1453 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1454                                           struct btrfs_fs_info *fs_info,
1455                                           struct btrfs_path *path,
1456                                           u64 bytenr, u64 parent,
1457                                           u64 root_objectid)
1458 {
1459         struct btrfs_root *root = fs_info->extent_root;
1460         struct btrfs_key key;
1461         int ret;
1462 
1463         key.objectid = bytenr;
1464         if (parent) {
1465                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1466                 key.offset = parent;
1467         } else {
1468                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1469                 key.offset = root_objectid;
1470         }
1471 
1472         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1473         if (ret > 0)
1474                 ret = -ENOENT;
1475 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1476         if (ret == -ENOENT && parent) {
1477                 btrfs_release_path(path);
1478                 key.type = BTRFS_EXTENT_REF_V0_KEY;
1479                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1480                 if (ret > 0)
1481                         ret = -ENOENT;
1482         }
1483 #endif
1484         return ret;
1485 }
1486 
1487 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1488                                           struct btrfs_fs_info *fs_info,
1489                                           struct btrfs_path *path,
1490                                           u64 bytenr, u64 parent,
1491                                           u64 root_objectid)
1492 {
1493         struct btrfs_key key;
1494         int ret;
1495 
1496         key.objectid = bytenr;
1497         if (parent) {
1498                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1499                 key.offset = parent;
1500         } else {
1501                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1502                 key.offset = root_objectid;
1503         }
1504 
1505         ret = btrfs_insert_empty_item(trans, fs_info->extent_root,
1506                                       path, &key, 0);
1507         btrfs_release_path(path);
1508         return ret;
1509 }
1510 
1511 static inline int extent_ref_type(u64 parent, u64 owner)
1512 {
1513         int type;
1514         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1515                 if (parent > 0)
1516                         type = BTRFS_SHARED_BLOCK_REF_KEY;
1517                 else
1518                         type = BTRFS_TREE_BLOCK_REF_KEY;
1519         } else {
1520                 if (parent > 0)
1521                         type = BTRFS_SHARED_DATA_REF_KEY;
1522                 else
1523                         type = BTRFS_EXTENT_DATA_REF_KEY;
1524         }
1525         return type;
1526 }
1527 
1528 static int find_next_key(struct btrfs_path *path, int level,
1529                          struct btrfs_key *key)
1530 
1531 {
1532         for (; level < BTRFS_MAX_LEVEL; level++) {
1533                 if (!path->nodes[level])
1534                         break;
1535                 if (path->slots[level] + 1 >=
1536                     btrfs_header_nritems(path->nodes[level]))
1537                         continue;
1538                 if (level == 0)
1539                         btrfs_item_key_to_cpu(path->nodes[level], key,
1540                                               path->slots[level] + 1);
1541                 else
1542                         btrfs_node_key_to_cpu(path->nodes[level], key,
1543                                               path->slots[level] + 1);
1544                 return 0;
1545         }
1546         return 1;
1547 }
1548 
1549 /*
1550  * look for inline back ref. if back ref is found, *ref_ret is set
1551  * to the address of inline back ref, and 0 is returned.
1552  *
1553  * if back ref isn't found, *ref_ret is set to the address where it
1554  * should be inserted, and -ENOENT is returned.
1555  *
1556  * if insert is true and there are too many inline back refs, the path
1557  * points to the extent item, and -EAGAIN is returned.
1558  *
1559  * NOTE: inline back refs are ordered in the same way that back ref
1560  *       items in the tree are ordered.
1561  */
1562 static noinline_for_stack
1563 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1564                                  struct btrfs_fs_info *fs_info,
1565                                  struct btrfs_path *path,
1566                                  struct btrfs_extent_inline_ref **ref_ret,
1567                                  u64 bytenr, u64 num_bytes,
1568                                  u64 parent, u64 root_objectid,
1569                                  u64 owner, u64 offset, int insert)
1570 {
1571         struct btrfs_root *root = fs_info->extent_root;
1572         struct btrfs_key key;
1573         struct extent_buffer *leaf;
1574         struct btrfs_extent_item *ei;
1575         struct btrfs_extent_inline_ref *iref;
1576         u64 flags;
1577         u64 item_size;
1578         unsigned long ptr;
1579         unsigned long end;
1580         int extra_size;
1581         int type;
1582         int want;
1583         int ret;
1584         int err = 0;
1585         bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
1586 
1587         key.objectid = bytenr;
1588         key.type = BTRFS_EXTENT_ITEM_KEY;
1589         key.offset = num_bytes;
1590 
1591         want = extent_ref_type(parent, owner);
1592         if (insert) {
1593                 extra_size = btrfs_extent_inline_ref_size(want);
1594                 path->keep_locks = 1;
1595         } else
1596                 extra_size = -1;
1597 
1598         /*
1599          * Owner is our parent level, so we can just add one to get the level
1600          * for the block we are interested in.
1601          */
1602         if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
1603                 key.type = BTRFS_METADATA_ITEM_KEY;
1604                 key.offset = owner;
1605         }
1606 
1607 again:
1608         ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1609         if (ret < 0) {
1610                 err = ret;
1611                 goto out;
1612         }
1613 
1614         /*
1615          * We may be a newly converted file system which still has the old fat
1616          * extent entries for metadata, so try and see if we have one of those.
1617          */
1618         if (ret > 0 && skinny_metadata) {
1619                 skinny_metadata = false;
1620                 if (path->slots[0]) {
1621                         path->slots[0]--;
1622                         btrfs_item_key_to_cpu(path->nodes[0], &key,
1623                                               path->slots[0]);
1624                         if (key.objectid == bytenr &&
1625                             key.type == BTRFS_EXTENT_ITEM_KEY &&
1626                             key.offset == num_bytes)
1627                                 ret = 0;
1628                 }
1629                 if (ret) {
1630                         key.objectid = bytenr;
1631                         key.type = BTRFS_EXTENT_ITEM_KEY;
1632                         key.offset = num_bytes;
1633                         btrfs_release_path(path);
1634                         goto again;
1635                 }
1636         }
1637 
1638         if (ret && !insert) {
1639                 err = -ENOENT;
1640                 goto out;
1641         } else if (WARN_ON(ret)) {
1642                 err = -EIO;
1643                 goto out;
1644         }
1645 
1646         leaf = path->nodes[0];
1647         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1648 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1649         if (item_size < sizeof(*ei)) {
1650                 if (!insert) {
1651                         err = -ENOENT;
1652                         goto out;
1653                 }
1654                 ret = convert_extent_item_v0(trans, fs_info, path, owner,
1655                                              extra_size);
1656                 if (ret < 0) {
1657                         err = ret;
1658                         goto out;
1659                 }
1660                 leaf = path->nodes[0];
1661                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1662         }
1663 #endif
1664         BUG_ON(item_size < sizeof(*ei));
1665 
1666         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1667         flags = btrfs_extent_flags(leaf, ei);
1668 
1669         ptr = (unsigned long)(ei + 1);
1670         end = (unsigned long)ei + item_size;
1671 
1672         if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1673                 ptr += sizeof(struct btrfs_tree_block_info);
1674                 BUG_ON(ptr > end);
1675         }
1676 
1677         err = -ENOENT;
1678         while (1) {
1679                 if (ptr >= end) {
1680                         WARN_ON(ptr > end);
1681                         break;
1682                 }
1683                 iref = (struct btrfs_extent_inline_ref *)ptr;
1684                 type = btrfs_extent_inline_ref_type(leaf, iref);
1685                 if (want < type)
1686                         break;
1687                 if (want > type) {
1688                         ptr += btrfs_extent_inline_ref_size(type);
1689                         continue;
1690                 }
1691 
1692                 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1693                         struct btrfs_extent_data_ref *dref;
1694                         dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1695                         if (match_extent_data_ref(leaf, dref, root_objectid,
1696                                                   owner, offset)) {
1697                                 err = 0;
1698                                 break;
1699                         }
1700                         if (hash_extent_data_ref_item(leaf, dref) <
1701                             hash_extent_data_ref(root_objectid, owner, offset))
1702                                 break;
1703                 } else {
1704                         u64 ref_offset;
1705                         ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1706                         if (parent > 0) {
1707                                 if (parent == ref_offset) {
1708                                         err = 0;
1709                                         break;
1710                                 }
1711                                 if (ref_offset < parent)
1712                                         break;
1713                         } else {
1714                                 if (root_objectid == ref_offset) {
1715                                         err = 0;
1716                                         break;
1717                                 }
1718                                 if (ref_offset < root_objectid)
1719                                         break;
1720                         }
1721                 }
1722                 ptr += btrfs_extent_inline_ref_size(type);
1723         }
1724         if (err == -ENOENT && insert) {
1725                 if (item_size + extra_size >=
1726                     BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1727                         err = -EAGAIN;
1728                         goto out;
1729                 }
1730                 /*
1731                  * To add new inline back ref, we have to make sure
1732                  * there is no corresponding back ref item.
1733                  * For simplicity, we just do not add new inline back
1734                  * ref if there is any kind of item for this block
1735                  */
1736                 if (find_next_key(path, 0, &key) == 0 &&
1737                     key.objectid == bytenr &&
1738                     key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1739                         err = -EAGAIN;
1740                         goto out;
1741                 }
1742         }
1743         *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1744 out:
1745         if (insert) {
1746                 path->keep_locks = 0;
1747                 btrfs_unlock_up_safe(path, 1);
1748         }
1749         return err;
1750 }
1751 
1752 /*
1753  * helper to add new inline back ref
1754  */
1755 static noinline_for_stack
1756 void setup_inline_extent_backref(struct btrfs_fs_info *fs_info,
1757                                  struct btrfs_path *path,
1758                                  struct btrfs_extent_inline_ref *iref,
1759                                  u64 parent, u64 root_objectid,
1760                                  u64 owner, u64 offset, int refs_to_add,
1761                                  struct btrfs_delayed_extent_op *extent_op)
1762 {
1763         struct extent_buffer *leaf;
1764         struct btrfs_extent_item *ei;
1765         unsigned long ptr;
1766         unsigned long end;
1767         unsigned long item_offset;
1768         u64 refs;
1769         int size;
1770         int type;
1771 
1772         leaf = path->nodes[0];
1773         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1774         item_offset = (unsigned long)iref - (unsigned long)ei;
1775 
1776         type = extent_ref_type(parent, owner);
1777         size = btrfs_extent_inline_ref_size(type);
1778 
1779         btrfs_extend_item(fs_info, path, size);
1780 
1781         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1782         refs = btrfs_extent_refs(leaf, ei);
1783         refs += refs_to_add;
1784         btrfs_set_extent_refs(leaf, ei, refs);
1785         if (extent_op)
1786                 __run_delayed_extent_op(extent_op, leaf, ei);
1787 
1788         ptr = (unsigned long)ei + item_offset;
1789         end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1790         if (ptr < end - size)
1791                 memmove_extent_buffer(leaf, ptr + size, ptr,
1792                                       end - size - ptr);
1793 
1794         iref = (struct btrfs_extent_inline_ref *)ptr;
1795         btrfs_set_extent_inline_ref_type(leaf, iref, type);
1796         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1797                 struct btrfs_extent_data_ref *dref;
1798                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1799                 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1800                 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1801                 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1802                 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1803         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1804                 struct btrfs_shared_data_ref *sref;
1805                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1806                 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1807                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1808         } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1809                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1810         } else {
1811                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1812         }
1813         btrfs_mark_buffer_dirty(leaf);
1814 }
1815 
1816 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1817                                  struct btrfs_fs_info *fs_info,
1818                                  struct btrfs_path *path,
1819                                  struct btrfs_extent_inline_ref **ref_ret,
1820                                  u64 bytenr, u64 num_bytes, u64 parent,
1821                                  u64 root_objectid, u64 owner, u64 offset)
1822 {
1823         int ret;
1824 
1825         ret = lookup_inline_extent_backref(trans, fs_info, path, ref_ret,
1826                                            bytenr, num_bytes, parent,
1827                                            root_objectid, owner, offset, 0);
1828         if (ret != -ENOENT)
1829                 return ret;
1830 
1831         btrfs_release_path(path);
1832         *ref_ret = NULL;
1833 
1834         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1835                 ret = lookup_tree_block_ref(trans, fs_info, path, bytenr,
1836                                             parent, root_objectid);
1837         } else {
1838                 ret = lookup_extent_data_ref(trans, fs_info, path, bytenr,
1839                                              parent, root_objectid, owner,
1840                                              offset);
1841         }
1842         return ret;
1843 }
1844 
1845 /*
1846  * helper to update/remove inline back ref
1847  */
1848 static noinline_for_stack
1849 void update_inline_extent_backref(struct btrfs_fs_info *fs_info,
1850                                   struct btrfs_path *path,
1851                                   struct btrfs_extent_inline_ref *iref,
1852                                   int refs_to_mod,
1853                                   struct btrfs_delayed_extent_op *extent_op,
1854                                   int *last_ref)
1855 {
1856         struct extent_buffer *leaf;
1857         struct btrfs_extent_item *ei;
1858         struct btrfs_extent_data_ref *dref = NULL;
1859         struct btrfs_shared_data_ref *sref = NULL;
1860         unsigned long ptr;
1861         unsigned long end;
1862         u32 item_size;
1863         int size;
1864         int type;
1865         u64 refs;
1866 
1867         leaf = path->nodes[0];
1868         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1869         refs = btrfs_extent_refs(leaf, ei);
1870         WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1871         refs += refs_to_mod;
1872         btrfs_set_extent_refs(leaf, ei, refs);
1873         if (extent_op)
1874                 __run_delayed_extent_op(extent_op, leaf, ei);
1875 
1876         type = btrfs_extent_inline_ref_type(leaf, iref);
1877 
1878         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1879                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1880                 refs = btrfs_extent_data_ref_count(leaf, dref);
1881         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1882                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1883                 refs = btrfs_shared_data_ref_count(leaf, sref);
1884         } else {
1885                 refs = 1;
1886                 BUG_ON(refs_to_mod != -1);
1887         }
1888 
1889         BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1890         refs += refs_to_mod;
1891 
1892         if (refs > 0) {
1893                 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1894                         btrfs_set_extent_data_ref_count(leaf, dref, refs);
1895                 else
1896                         btrfs_set_shared_data_ref_count(leaf, sref, refs);
1897         } else {
1898                 *last_ref = 1;
1899                 size =  btrfs_extent_inline_ref_size(type);
1900                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1901                 ptr = (unsigned long)iref;
1902                 end = (unsigned long)ei + item_size;
1903                 if (ptr + size < end)
1904                         memmove_extent_buffer(leaf, ptr, ptr + size,
1905                                               end - ptr - size);
1906                 item_size -= size;
1907                 btrfs_truncate_item(fs_info, path, item_size, 1);
1908         }
1909         btrfs_mark_buffer_dirty(leaf);
1910 }
1911 
1912 static noinline_for_stack
1913 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1914                                  struct btrfs_fs_info *fs_info,
1915                                  struct btrfs_path *path,
1916                                  u64 bytenr, u64 num_bytes, u64 parent,
1917                                  u64 root_objectid, u64 owner,
1918                                  u64 offset, int refs_to_add,
1919                                  struct btrfs_delayed_extent_op *extent_op)
1920 {
1921         struct btrfs_extent_inline_ref *iref;
1922         int ret;
1923 
1924         ret = lookup_inline_extent_backref(trans, fs_info, path, &iref,
1925                                            bytenr, num_bytes, parent,
1926                                            root_objectid, owner, offset, 1);
1927         if (ret == 0) {
1928                 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1929                 update_inline_extent_backref(fs_info, path, iref,
1930                                              refs_to_add, extent_op, NULL);
1931         } else if (ret == -ENOENT) {
1932                 setup_inline_extent_backref(fs_info, path, iref, parent,
1933                                             root_objectid, owner, offset,
1934                                             refs_to_add, extent_op);
1935                 ret = 0;
1936         }
1937         return ret;
1938 }
1939 
1940 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1941                                  struct btrfs_fs_info *fs_info,
1942                                  struct btrfs_path *path,
1943                                  u64 bytenr, u64 parent, u64 root_objectid,
1944                                  u64 owner, u64 offset, int refs_to_add)
1945 {
1946         int ret;
1947         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1948                 BUG_ON(refs_to_add != 1);
1949                 ret = insert_tree_block_ref(trans, fs_info, path, bytenr,
1950                                             parent, root_objectid);
1951         } else {
1952                 ret = insert_extent_data_ref(trans, fs_info, path, bytenr,
1953                                              parent, root_objectid,
1954                                              owner, offset, refs_to_add);
1955         }
1956         return ret;
1957 }
1958 
1959 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1960                                  struct btrfs_fs_info *fs_info,
1961                                  struct btrfs_path *path,
1962                                  struct btrfs_extent_inline_ref *iref,
1963                                  int refs_to_drop, int is_data, int *last_ref)
1964 {
1965         int ret = 0;
1966 
1967         BUG_ON(!is_data && refs_to_drop != 1);
1968         if (iref) {
1969                 update_inline_extent_backref(fs_info, path, iref,
1970                                              -refs_to_drop, NULL, last_ref);
1971         } else if (is_data) {
1972                 ret = remove_extent_data_ref(trans, fs_info, path, refs_to_drop,
1973                                              last_ref);
1974         } else {
1975                 *last_ref = 1;
1976                 ret = btrfs_del_item(trans, fs_info->extent_root, path);
1977         }
1978         return ret;
1979 }
1980 
1981 #define in_range(b, first, len)        ((b) >= (first) && (b) < (first) + (len))
1982 static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len,
1983                                u64 *discarded_bytes)
1984 {
1985         int j, ret = 0;
1986         u64 bytes_left, end;
1987         u64 aligned_start = ALIGN(start, 1 << 9);
1988 
1989         if (WARN_ON(start != aligned_start)) {
1990                 len -= aligned_start - start;
1991                 len = round_down(len, 1 << 9);
1992                 start = aligned_start;
1993         }
1994 
1995         *discarded_bytes = 0;
1996 
1997         if (!len)
1998                 return 0;
1999 
2000         end = start + len;
2001         bytes_left = len;
2002 
2003         /* Skip any superblocks on this device. */
2004         for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) {
2005                 u64 sb_start = btrfs_sb_offset(j);
2006                 u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE;
2007                 u64 size = sb_start - start;
2008 
2009                 if (!in_range(sb_start, start, bytes_left) &&
2010                     !in_range(sb_end, start, bytes_left) &&
2011                     !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE))
2012                         continue;
2013 
2014                 /*
2015                  * Superblock spans beginning of range.  Adjust start and
2016                  * try again.
2017                  */
2018                 if (sb_start <= start) {
2019                         start += sb_end - start;
2020                         if (start > end) {
2021                                 bytes_left = 0;
2022                                 break;
2023                         }
2024                         bytes_left = end - start;
2025                         continue;
2026                 }
2027 
2028                 if (size) {
2029                         ret = blkdev_issue_discard(bdev, start >> 9, size >> 9,
2030                                                    GFP_NOFS, 0);
2031                         if (!ret)
2032                                 *discarded_bytes += size;
2033                         else if (ret != -EOPNOTSUPP)
2034                                 return ret;
2035                 }
2036 
2037                 start = sb_end;
2038                 if (start > end) {
2039                         bytes_left = 0;
2040                         break;
2041                 }
2042                 bytes_left = end - start;
2043         }
2044 
2045         if (bytes_left) {
2046                 ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9,
2047                                            GFP_NOFS, 0);
2048                 if (!ret)
2049                         *discarded_bytes += bytes_left;
2050         }
2051         return ret;
2052 }
2053 
2054 int btrfs_discard_extent(struct btrfs_fs_info *fs_info, u64 bytenr,
2055                          u64 num_bytes, u64 *actual_bytes)
2056 {
2057         int ret;
2058         u64 discarded_bytes = 0;
2059         struct btrfs_bio *bbio = NULL;
2060 
2061 
2062         /*
2063          * Avoid races with device replace and make sure our bbio has devices
2064          * associated to its stripes that don't go away while we are discarding.
2065          */
2066         btrfs_bio_counter_inc_blocked(fs_info);
2067         /* Tell the block device(s) that the sectors can be discarded */
2068         ret = btrfs_map_block(fs_info, BTRFS_MAP_DISCARD, bytenr, &num_bytes,
2069                               &bbio, 0);
2070         /* Error condition is -ENOMEM */
2071         if (!ret) {
2072                 struct btrfs_bio_stripe *stripe = bbio->stripes;
2073                 int i;
2074 
2075 
2076                 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
2077                         u64 bytes;
2078                         if (!stripe->dev->can_discard)
2079                                 continue;
2080 
2081                         ret = btrfs_issue_discard(stripe->dev->bdev,
2082                                                   stripe->physical,
2083                                                   stripe->length,
2084                                                   &bytes);
2085                         if (!ret)
2086                                 discarded_bytes += bytes;
2087                         else if (ret != -EOPNOTSUPP)
2088                                 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
2089 
2090                         /*
2091                          * Just in case we get back EOPNOTSUPP for some reason,
2092                          * just ignore the return value so we don't screw up
2093                          * people calling discard_extent.
2094                          */
2095                         ret = 0;
2096                 }
2097                 btrfs_put_bbio(bbio);
2098         }
2099         btrfs_bio_counter_dec(fs_info);
2100 
2101         if (actual_bytes)
2102                 *actual_bytes = discarded_bytes;
2103 
2104 
2105         if (ret == -EOPNOTSUPP)
2106                 ret = 0;
2107         return ret;
2108 }
2109 
2110 /* Can return -ENOMEM */
2111 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2112                          struct btrfs_fs_info *fs_info,
2113                          u64 bytenr, u64 num_bytes, u64 parent,
2114                          u64 root_objectid, u64 owner, u64 offset)
2115 {
2116         int old_ref_mod, new_ref_mod;
2117         int ret;
2118 
2119         BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
2120                root_objectid == BTRFS_TREE_LOG_OBJECTID);
2121 
2122         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
2123                 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
2124                                                  num_bytes, parent,
2125                                                  root_objectid, (int)owner,
2126                                                  BTRFS_ADD_DELAYED_REF, NULL,
2127                                                  &old_ref_mod, &new_ref_mod);
2128         } else {
2129                 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
2130                                                  num_bytes, parent,
2131                                                  root_objectid, owner, offset,
2132                                                  0, BTRFS_ADD_DELAYED_REF,
2133                                                  &old_ref_mod, &new_ref_mod);
2134         }
2135 
2136         if (ret == 0 && old_ref_mod < 0 && new_ref_mod >= 0)
2137                 add_pinned_bytes(fs_info, -num_bytes, owner, root_objectid);
2138 
2139         return ret;
2140 }
2141 
2142 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2143                                   struct btrfs_fs_info *fs_info,
2144                                   struct btrfs_delayed_ref_node *node,
2145                                   u64 parent, u64 root_objectid,
2146                                   u64 owner, u64 offset, int refs_to_add,
2147                                   struct btrfs_delayed_extent_op *extent_op)
2148 {
2149         struct btrfs_path *path;
2150         struct extent_buffer *leaf;
2151         struct btrfs_extent_item *item;
2152         struct btrfs_key key;
2153         u64 bytenr = node->bytenr;
2154         u64 num_bytes = node->num_bytes;
2155         u64 refs;
2156         int ret;
2157 
2158         path = btrfs_alloc_path();
2159         if (!path)
2160                 return -ENOMEM;
2161 
2162         path->reada = READA_FORWARD;
2163         path->leave_spinning = 1;
2164         /* this will setup the path even if it fails to insert the back ref */
2165         ret = insert_inline_extent_backref(trans, fs_info, path, bytenr,
2166                                            num_bytes, parent, root_objectid,
2167                                            owner, offset,
2168                                            refs_to_add, extent_op);
2169         if ((ret < 0 && ret != -EAGAIN) || !ret)
2170                 goto out;
2171 
2172         /*
2173          * Ok we had -EAGAIN which means we didn't have space to insert and
2174          * inline extent ref, so just update the reference count and add a
2175          * normal backref.
2176          */
2177         leaf = path->nodes[0];
2178         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2179         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2180         refs = btrfs_extent_refs(leaf, item);
2181         btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2182         if (extent_op)
2183                 __run_delayed_extent_op(extent_op, leaf, item);
2184 
2185         btrfs_mark_buffer_dirty(leaf);
2186         btrfs_release_path(path);
2187 
2188         path->reada = READA_FORWARD;
2189         path->leave_spinning = 1;
2190         /* now insert the actual backref */
2191         ret = insert_extent_backref(trans, fs_info, path, bytenr, parent,
2192                                     root_objectid, owner, offset, refs_to_add);
2193         if (ret)
2194                 btrfs_abort_transaction(trans, ret);
2195 out:
2196         btrfs_free_path(path);
2197         return ret;
2198 }
2199 
2200 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2201                                 struct btrfs_fs_info *fs_info,
2202                                 struct btrfs_delayed_ref_node *node,
2203                                 struct btrfs_delayed_extent_op *extent_op,
2204                                 int insert_reserved)
2205 {
2206         int ret = 0;
2207         struct btrfs_delayed_data_ref *ref;
2208         struct btrfs_key ins;
2209         u64 parent = 0;
2210         u64 ref_root = 0;
2211         u64 flags = 0;
2212 
2213         ins.objectid = node->bytenr;
2214         ins.offset = node->num_bytes;
2215         ins.type = BTRFS_EXTENT_ITEM_KEY;
2216 
2217         ref = btrfs_delayed_node_to_data_ref(node);
2218         trace_run_delayed_data_ref(fs_info, node, ref, node->action);
2219 
2220         if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2221                 parent = ref->parent;
2222         ref_root = ref->root;
2223 
2224         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2225                 if (extent_op)
2226                         flags |= extent_op->flags_to_set;
2227                 ret = alloc_reserved_file_extent(trans, fs_info,
2228                                                  parent, ref_root, flags,
2229                                                  ref->objectid, ref->offset,
2230                                                  &ins, node->ref_mod);
2231         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2232                 ret = __btrfs_inc_extent_ref(trans, fs_info, node, parent,
2233                                              ref_root, ref->objectid,
2234                                              ref->offset, node->ref_mod,
2235                                              extent_op);
2236         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2237                 ret = __btrfs_free_extent(trans, fs_info, node, parent,
2238                                           ref_root, ref->objectid,
2239                                           ref->offset, node->ref_mod,
2240                                           extent_op);
2241         } else {
2242                 BUG();
2243         }
2244         return ret;
2245 }
2246 
2247 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2248                                     struct extent_buffer *leaf,
2249                                     struct btrfs_extent_item *ei)
2250 {
2251         u64 flags = btrfs_extent_flags(leaf, ei);
2252         if (extent_op->update_flags) {
2253                 flags |= extent_op->flags_to_set;
2254                 btrfs_set_extent_flags(leaf, ei, flags);
2255         }
2256 
2257         if (extent_op->update_key) {
2258                 struct btrfs_tree_block_info *bi;
2259                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2260                 bi = (struct btrfs_tree_block_info *)(ei + 1);
2261                 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2262         }
2263 }
2264 
2265 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2266                                  struct btrfs_fs_info *fs_info,
2267                                  struct btrfs_delayed_ref_node *node,
2268                                  struct btrfs_delayed_extent_op *extent_op)
2269 {
2270         struct btrfs_key key;
2271         struct btrfs_path *path;
2272         struct btrfs_extent_item *ei;
2273         struct extent_buffer *leaf;
2274         u32 item_size;
2275         int ret;
2276         int err = 0;
2277         int metadata = !extent_op->is_data;
2278 
2279         if (trans->aborted)
2280                 return 0;
2281 
2282         if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA))
2283                 metadata = 0;
2284 
2285         path = btrfs_alloc_path();
2286         if (!path)
2287                 return -ENOMEM;
2288 
2289         key.objectid = node->bytenr;
2290 
2291         if (metadata) {
2292                 key.type = BTRFS_METADATA_ITEM_KEY;
2293                 key.offset = extent_op->level;
2294         } else {
2295                 key.type = BTRFS_EXTENT_ITEM_KEY;
2296                 key.offset = node->num_bytes;
2297         }
2298 
2299 again:
2300         path->reada = READA_FORWARD;
2301         path->leave_spinning = 1;
2302         ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 1);
2303         if (ret < 0) {
2304                 err = ret;
2305                 goto out;
2306         }
2307         if (ret > 0) {
2308                 if (metadata) {
2309                         if (path->slots[0] > 0) {
2310                                 path->slots[0]--;
2311                                 btrfs_item_key_to_cpu(path->nodes[0], &key,
2312                                                       path->slots[0]);
2313                                 if (key.objectid == node->bytenr &&
2314                                     key.type == BTRFS_EXTENT_ITEM_KEY &&
2315                                     key.offset == node->num_bytes)
2316                                         ret = 0;
2317                         }
2318                         if (ret > 0) {
2319                                 btrfs_release_path(path);
2320                                 metadata = 0;
2321 
2322                                 key.objectid = node->bytenr;
2323                                 key.offset = node->num_bytes;
2324                                 key.type = BTRFS_EXTENT_ITEM_KEY;
2325                                 goto again;
2326                         }
2327                 } else {
2328                         err = -EIO;
2329                         goto out;
2330                 }
2331         }
2332 
2333         leaf = path->nodes[0];
2334         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2335 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2336         if (item_size < sizeof(*ei)) {
2337                 ret = convert_extent_item_v0(trans, fs_info, path, (u64)-1, 0);
2338                 if (ret < 0) {
2339                         err = ret;
2340                         goto out;
2341                 }
2342                 leaf = path->nodes[0];
2343                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2344         }
2345 #endif
2346         BUG_ON(item_size < sizeof(*ei));
2347         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2348         __run_delayed_extent_op(extent_op, leaf, ei);
2349 
2350         btrfs_mark_buffer_dirty(leaf);
2351 out:
2352         btrfs_free_path(path);
2353         return err;
2354 }
2355 
2356 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2357                                 struct btrfs_fs_info *fs_info,
2358                                 struct btrfs_delayed_ref_node *node,
2359                                 struct btrfs_delayed_extent_op *extent_op,
2360                                 int insert_reserved)
2361 {
2362         int ret = 0;
2363         struct btrfs_delayed_tree_ref *ref;
2364         struct btrfs_key ins;
2365         u64 parent = 0;
2366         u64 ref_root = 0;
2367         bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
2368 
2369         ref = btrfs_delayed_node_to_tree_ref(node);
2370         trace_run_delayed_tree_ref(fs_info, node, ref, node->action);
2371 
2372         if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2373                 parent = ref->parent;
2374         ref_root = ref->root;
2375 
2376         ins.objectid = node->bytenr;
2377         if (skinny_metadata) {
2378                 ins.offset = ref->level;
2379                 ins.type = BTRFS_METADATA_ITEM_KEY;
2380         } else {
2381                 ins.offset = node->num_bytes;
2382                 ins.type = BTRFS_EXTENT_ITEM_KEY;
2383         }
2384 
2385         if (node->ref_mod != 1) {
2386                 btrfs_err(fs_info,
2387         "btree block(%llu) has %d references rather than 1: action %d ref_root %llu parent %llu",
2388                           node->bytenr, node->ref_mod, node->action, ref_root,
2389                           parent);
2390                 return -EIO;
2391         }
2392         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2393                 BUG_ON(!extent_op || !extent_op->update_flags);
2394                 ret = alloc_reserved_tree_block(trans, fs_info,
2395                                                 parent, ref_root,
2396                                                 extent_op->flags_to_set,
2397                                                 &extent_op->key,
2398                                                 ref->level, &ins);
2399         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2400                 ret = __btrfs_inc_extent_ref(trans, fs_info, node,
2401                                              parent, ref_root,
2402                                              ref->level, 0, 1,
2403                                              extent_op);
2404         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2405                 ret = __btrfs_free_extent(trans, fs_info, node,
2406                                           parent, ref_root,
2407                                           ref->level, 0, 1, extent_op);
2408         } else {
2409                 BUG();
2410         }
2411         return ret;
2412 }
2413 
2414 /* helper function to actually process a single delayed ref entry */
2415 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2416                                struct btrfs_fs_info *fs_info,
2417                                struct btrfs_delayed_ref_node *node,
2418                                struct btrfs_delayed_extent_op *extent_op,
2419                                int insert_reserved)
2420 {
2421         int ret = 0;
2422 
2423         if (trans->aborted) {
2424                 if (insert_reserved)
2425                         btrfs_pin_extent(fs_info, node->bytenr,
2426                                          node->num_bytes, 1);
2427                 return 0;
2428         }
2429 
2430         if (btrfs_delayed_ref_is_head(node)) {
2431                 struct btrfs_delayed_ref_head *head;
2432                 /*
2433                  * we've hit the end of the chain and we were supposed
2434                  * to insert this extent into the tree.  But, it got
2435                  * deleted before we ever needed to insert it, so all
2436                  * we have to do is clean up the accounting
2437                  */
2438                 BUG_ON(extent_op);
2439                 head = btrfs_delayed_node_to_head(node);
2440                 trace_run_delayed_ref_head(fs_info, node, head, node->action);
2441 
2442                 if (head->total_ref_mod < 0) {
2443                         struct btrfs_block_group_cache *cache;
2444 
2445                         cache = btrfs_lookup_block_group(fs_info, node->bytenr);
2446                         ASSERT(cache);
2447                         percpu_counter_add(&cache->space_info->total_bytes_pinned,
2448                                            -node->num_bytes);
2449                         btrfs_put_block_group(cache);
2450                 }
2451 
2452                 if (insert_reserved) {
2453                         btrfs_pin_extent(fs_info, node->bytenr,
2454                                          node->num_bytes, 1);
2455                         if (head->is_data) {
2456                                 ret = btrfs_del_csums(trans, fs_info,
2457                                                       node->bytenr,
2458                                                       node->num_bytes);
2459                         }
2460                 }
2461 
2462                 /* Also free its reserved qgroup space */
2463                 btrfs_qgroup_free_delayed_ref(fs_info, head->qgroup_ref_root,
2464                                               head->qgroup_reserved);
2465                 return ret;
2466         }
2467 
2468         if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2469             node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2470                 ret = run_delayed_tree_ref(trans, fs_info, node, extent_op,
2471                                            insert_reserved);
2472         else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2473                  node->type == BTRFS_SHARED_DATA_REF_KEY)
2474                 ret = run_delayed_data_ref(trans, fs_info, node, extent_op,
2475                                            insert_reserved);
2476         else
2477                 BUG();
2478         return ret;
2479 }
2480 
2481 static inline struct btrfs_delayed_ref_node *
2482 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2483 {
2484         struct btrfs_delayed_ref_node *ref;
2485 
2486         if (list_empty(&head->ref_list))
2487                 return NULL;
2488 
2489         /*
2490          * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
2491          * This is to prevent a ref count from going down to zero, which deletes
2492          * the extent item from the extent tree, when there still are references
2493          * to add, which would fail because they would not find the extent item.
2494          */
2495         if (!list_empty(&head->ref_add_list))
2496                 return list_first_entry(&head->ref_add_list,
2497                                 struct btrfs_delayed_ref_node, add_list);
2498 
2499         ref = list_first_entry(&head->ref_list, struct btrfs_delayed_ref_node,
2500                                list);
2501         ASSERT(list_empty(&ref->add_list));
2502         return ref;
2503 }
2504 
2505 /*
2506  * Returns 0 on success or if called with an already aborted transaction.
2507  * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2508  */
2509 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2510                                              struct btrfs_fs_info *fs_info,
2511                                              unsigned long nr)
2512 {
2513         struct btrfs_delayed_ref_root *delayed_refs;
2514         struct btrfs_delayed_ref_node *ref;
2515         struct btrfs_delayed_ref_head *locked_ref = NULL;
2516         struct btrfs_delayed_extent_op *extent_op;
2517         ktime_t start = ktime_get();
2518         int ret;
2519         unsigned long count = 0;
2520         unsigned long actual_count = 0;
2521         int must_insert_reserved = 0;
2522 
2523         delayed_refs = &trans->transaction->delayed_refs;
2524         while (1) {
2525                 if (!locked_ref) {
2526                         if (count >= nr)
2527                                 break;
2528 
2529                         spin_lock(&delayed_refs->lock);
2530                         locked_ref = btrfs_select_ref_head(trans);
2531                         if (!locked_ref) {
2532                                 spin_unlock(&delayed_refs->lock);
2533                                 break;
2534                         }
2535 
2536                         /* grab the lock that says we are going to process
2537                          * all the refs for this head */
2538                         ret = btrfs_delayed_ref_lock(trans, locked_ref);
2539                         spin_unlock(&delayed_refs->lock);
2540                         /*
2541                          * we may have dropped the spin lock to get the head
2542                          * mutex lock, and that might have given someone else
2543                          * time to free the head.  If that's true, it has been
2544                          * removed from our list and we can move on.
2545                          */
2546                         if (ret == -EAGAIN) {
2547                                 locked_ref = NULL;
2548                                 count++;
2549                                 continue;
2550                         }
2551                 }
2552 
2553                 /*
2554                  * We need to try and merge add/drops of the same ref since we
2555                  * can run into issues with relocate dropping the implicit ref
2556                  * and then it being added back again before the drop can
2557                  * finish.  If we merged anything we need to re-loop so we can
2558                  * get a good ref.
2559                  * Or we can get node references of the same type that weren't
2560                  * merged when created due to bumps in the tree mod seq, and
2561                  * we need to merge them to prevent adding an inline extent
2562                  * backref before dropping it (triggering a BUG_ON at
2563                  * insert_inline_extent_backref()).
2564                  */
2565                 spin_lock(&locked_ref->lock);
2566                 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2567                                          locked_ref);
2568 
2569                 /*
2570                  * locked_ref is the head node, so we have to go one
2571                  * node back for any delayed ref updates
2572                  */
2573                 ref = select_delayed_ref(locked_ref);
2574 
2575                 if (ref && ref->seq &&
2576                     btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2577                         spin_unlock(&locked_ref->lock);
2578                         spin_lock(&delayed_refs->lock);
2579                         locked_ref->processing = 0;
2580                         delayed_refs->num_heads_ready++;
2581                         spin_unlock(&delayed_refs->lock);
2582                         btrfs_delayed_ref_unlock(locked_ref);
2583                         locked_ref = NULL;
2584                         cond_resched();
2585                         count++;
2586                         continue;
2587                 }
2588 
2589                 /*
2590                  * record the must insert reserved flag before we
2591                  * drop the spin lock.
2592                  */
2593                 must_insert_reserved = locked_ref->must_insert_reserved;
2594                 locked_ref->must_insert_reserved = 0;
2595 
2596                 extent_op = locked_ref->extent_op;
2597                 locked_ref->extent_op = NULL;
2598 
2599                 if (!ref) {
2600 
2601 
2602                         /* All delayed refs have been processed, Go ahead
2603                          * and send the head node to run_one_delayed_ref,
2604                          * so that any accounting fixes can happen
2605                          */
2606                         ref = &locked_ref->node;
2607 
2608                         if (extent_op && must_insert_reserved) {
2609                                 btrfs_free_delayed_extent_op(extent_op);
2610                                 extent_op = NULL;
2611                         }
2612 
2613                         if (extent_op) {
2614                                 spin_unlock(&locked_ref->lock);
2615                                 ret = run_delayed_extent_op(trans, fs_info,
2616                                                             ref, extent_op);
2617                                 btrfs_free_delayed_extent_op(extent_op);
2618 
2619                                 if (ret) {
2620                                         /*
2621                                          * Need to reset must_insert_reserved if
2622                                          * there was an error so the abort stuff
2623                                          * can cleanup the reserved space
2624                                          * properly.
2625                                          */
2626                                         if (must_insert_reserved)
2627                                                 locked_ref->must_insert_reserved = 1;
2628                                         spin_lock(&delayed_refs->lock);
2629                                         locked_ref->processing = 0;
2630                                         delayed_refs->num_heads_ready++;
2631                                         spin_unlock(&delayed_refs->lock);
2632                                         btrfs_debug(fs_info,
2633                                                     "run_delayed_extent_op returned %d",
2634                                                     ret);
2635                                         btrfs_delayed_ref_unlock(locked_ref);
2636                                         return ret;
2637                                 }
2638                                 continue;
2639                         }
2640 
2641                         /*
2642                          * Need to drop our head ref lock and re-acquire the
2643                          * delayed ref lock and then re-check to make sure
2644                          * nobody got added.
2645                          */
2646                         spin_unlock(&locked_ref->lock);
2647                         spin_lock(&delayed_refs->lock);
2648                         spin_lock(&locked_ref->lock);
2649                         if (!list_empty(&locked_ref->ref_list) ||
2650                             locked_ref->extent_op) {
2651                                 spin_unlock(&locked_ref->lock);
2652                                 spin_unlock(&delayed_refs->lock);
2653                                 continue;
2654                         }
2655                         ref->in_tree = 0;
2656                         delayed_refs->num_heads--;
2657                         rb_erase(&locked_ref->href_node,
2658                                  &delayed_refs->href_root);
2659                         spin_unlock(&delayed_refs->lock);
2660                 } else {
2661                         actual_count++;
2662                         ref->in_tree = 0;
2663                         list_del(&ref->list);
2664                         if (!list_empty(&ref->add_list))
2665                                 list_del(&ref->add_list);
2666                 }
2667                 atomic_dec(&delayed_refs->num_entries);
2668 
2669                 if (!btrfs_delayed_ref_is_head(ref)) {
2670                         /*
2671                          * when we play the delayed ref, also correct the
2672                          * ref_mod on head
2673                          */
2674                         switch (ref->action) {
2675                         case BTRFS_ADD_DELAYED_REF:
2676                         case BTRFS_ADD_DELAYED_EXTENT:
2677                                 locked_ref->node.ref_mod -= ref->ref_mod;
2678                                 break;
2679                         case BTRFS_DROP_DELAYED_REF:
2680                                 locked_ref->node.ref_mod += ref->ref_mod;
2681                                 break;
2682                         default:
2683                                 WARN_ON(1);
2684                         }
2685                 }
2686                 spin_unlock(&locked_ref->lock);
2687 
2688                 ret = run_one_delayed_ref(trans, fs_info, ref, extent_op,
2689                                           must_insert_reserved);
2690 
2691                 btrfs_free_delayed_extent_op(extent_op);
2692                 if (ret) {
2693                         spin_lock(&delayed_refs->lock);
2694                         locked_ref->processing = 0;
2695                         delayed_refs->num_heads_ready++;
2696                         spin_unlock(&delayed_refs->lock);
2697                         btrfs_delayed_ref_unlock(locked_ref);
2698                         btrfs_put_delayed_ref(ref);
2699                         btrfs_debug(fs_info, "run_one_delayed_ref returned %d",
2700                                     ret);
2701                         return ret;
2702                 }
2703 
2704                 /*
2705                  * If this node is a head, that means all the refs in this head
2706                  * have been dealt with, and we will pick the next head to deal
2707                  * with, so we must unlock the head and drop it from the cluster
2708                  * list before we release it.
2709                  */
2710                 if (btrfs_delayed_ref_is_head(ref)) {
2711                         if (locked_ref->is_data &&
2712                             locked_ref->total_ref_mod < 0) {
2713                                 spin_lock(&delayed_refs->lock);
2714                                 delayed_refs->pending_csums -= ref->num_bytes;
2715                                 spin_unlock(&delayed_refs->lock);
2716                         }
2717                         btrfs_delayed_ref_unlock(locked_ref);
2718                         locked_ref = NULL;
2719                 }
2720                 btrfs_put_delayed_ref(ref);
2721                 count++;
2722                 cond_resched();
2723         }
2724 
2725         /*
2726          * We don't want to include ref heads since we can have empty ref heads
2727          * and those will drastically skew our runtime down since we just do
2728          * accounting, no actual extent tree updates.
2729          */
2730         if (actual_count > 0) {
2731                 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2732                 u64 avg;
2733 
2734                 /*
2735                  * We weigh the current average higher than our current runtime
2736                  * to avoid large swings in the average.
2737                  */
2738                 spin_lock(&delayed_refs->lock);
2739                 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2740                 fs_info->avg_delayed_ref_runtime = avg >> 2;    /* div by 4 */
2741                 spin_unlock(&delayed_refs->lock);
2742         }
2743         return 0;
2744 }
2745 
2746 #ifdef SCRAMBLE_DELAYED_REFS
2747 /*
2748  * Normally delayed refs get processed in ascending bytenr order. This
2749  * correlates in most cases to the order added. To expose dependencies on this
2750  * order, we start to process the tree in the middle instead of the beginning
2751  */
2752 static u64 find_middle(struct rb_root *root)
2753 {
2754         struct rb_node *n = root->rb_node;
2755         struct btrfs_delayed_ref_node *entry;
2756         int alt = 1;
2757         u64 middle;
2758         u64 first = 0, last = 0;
2759 
2760         n = rb_first(root);
2761         if (n) {
2762                 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2763                 first = entry->bytenr;
2764         }
2765         n = rb_last(root);
2766         if (n) {
2767                 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2768                 last = entry->bytenr;
2769         }
2770         n = root->rb_node;
2771 
2772         while (n) {
2773                 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2774                 WARN_ON(!entry->in_tree);
2775 
2776                 middle = entry->bytenr;
2777 
2778                 if (alt)
2779                         n = n->rb_left;
2780                 else
2781                         n = n->rb_right;
2782 
2783                 alt = 1 - alt;
2784         }
2785         return middle;
2786 }
2787 #endif
2788 
2789 static inline u64 heads_to_leaves(struct btrfs_fs_info *fs_info, u64 heads)
2790 {
2791         u64 num_bytes;
2792 
2793         num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2794                              sizeof(struct btrfs_extent_inline_ref));
2795         if (!btrfs_fs_incompat(fs_info, SKINNY_METADATA))
2796                 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2797 
2798         /*
2799          * We don't ever fill up leaves all the way so multiply by 2 just to be
2800          * closer to what we're really going to want to use.
2801          */
2802         return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(fs_info));
2803 }
2804 
2805 /*
2806  * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2807  * would require to store the csums for that many bytes.
2808  */
2809 u64 btrfs_csum_bytes_to_leaves(struct btrfs_fs_info *fs_info, u64 csum_bytes)
2810 {
2811         u64 csum_size;
2812         u64 num_csums_per_leaf;
2813         u64 num_csums;
2814 
2815         csum_size = BTRFS_MAX_ITEM_SIZE(fs_info);
2816         num_csums_per_leaf = div64_u64(csum_size,
2817                         (u64)btrfs_super_csum_size(fs_info->super_copy));
2818         num_csums = div64_u64(csum_bytes, fs_info->sectorsize);
2819         num_csums += num_csums_per_leaf - 1;
2820         num_csums = div64_u64(num_csums, num_csums_per_leaf);
2821         return num_csums;
2822 }
2823 
2824 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans,
2825                                        struct btrfs_fs_info *fs_info)
2826 {
2827         struct btrfs_block_rsv *global_rsv;
2828         u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2829         u64 csum_bytes = trans->transaction->delayed_refs.pending_csums;
2830         u64 num_dirty_bgs = trans->transaction->num_dirty_bgs;
2831         u64 num_bytes, num_dirty_bgs_bytes;
2832         int ret = 0;
2833 
2834         num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
2835         num_heads = heads_to_leaves(fs_info, num_heads);
2836         if (num_heads > 1)
2837                 num_bytes += (num_heads - 1) * fs_info->nodesize;
2838         num_bytes <<= 1;
2839         num_bytes += btrfs_csum_bytes_to_leaves(fs_info, csum_bytes) *
2840                                                         fs_info->nodesize;
2841         num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(fs_info,
2842                                                              num_dirty_bgs);
2843         global_rsv = &fs_info->global_block_rsv;
2844 
2845         /*
2846          * If we can't allocate any more chunks lets make sure we have _lots_ of
2847          * wiggle room since running delayed refs can create more delayed refs.
2848          */
2849         if (global_rsv->space_info->full) {
2850                 num_dirty_bgs_bytes <<= 1;
2851                 num_bytes <<= 1;
2852         }
2853 
2854         spin_lock(&global_rsv->lock);
2855         if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes)
2856                 ret = 1;
2857         spin_unlock(&global_rsv->lock);
2858         return ret;
2859 }
2860 
2861 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2862                                        struct btrfs_fs_info *fs_info)
2863 {
2864         u64 num_entries =
2865                 atomic_read(&trans->transaction->delayed_refs.num_entries);
2866         u64 avg_runtime;
2867         u64 val;
2868 
2869         smp_mb();
2870         avg_runtime = fs_info->avg_delayed_ref_runtime;
2871         val = num_entries * avg_runtime;
2872         if (val >= NSEC_PER_SEC)
2873                 return 1;
2874         if (val >= NSEC_PER_SEC / 2)
2875                 return 2;
2876 
2877         return btrfs_check_space_for_delayed_refs(trans, fs_info);
2878 }
2879 
2880 struct async_delayed_refs {
2881         struct btrfs_root *root;
2882         u64 transid;
2883         int count;
2884         int error;
2885         int sync;
2886         struct completion wait;
2887         struct btrfs_work work;
2888 };
2889 
2890 static inline struct async_delayed_refs *
2891 to_async_delayed_refs(struct btrfs_work *work)
2892 {
2893         return container_of(work, struct async_delayed_refs, work);
2894 }
2895 
2896 static void delayed_ref_async_start(struct btrfs_work *work)
2897 {
2898         struct async_delayed_refs *async = to_async_delayed_refs(work);
2899         struct btrfs_trans_handle *trans;
2900         struct btrfs_fs_info *fs_info = async->root->fs_info;
2901         int ret;
2902 
2903         /* if the commit is already started, we don't need to wait here */
2904         if (btrfs_transaction_blocked(fs_info))
2905                 goto done;
2906 
2907         trans = btrfs_join_transaction(async->root);
2908         if (IS_ERR(trans)) {
2909                 async->error = PTR_ERR(trans);
2910                 goto done;
2911         }
2912 
2913         /*
2914          * trans->sync means that when we call end_transaction, we won't
2915          * wait on delayed refs
2916          */
2917         trans->sync = true;
2918 
2919         /* Don't bother flushing if we got into a different transaction */
2920         if (trans->transid > async->transid)
2921                 goto end;
2922 
2923         ret = btrfs_run_delayed_refs(trans, fs_info, async->count);
2924         if (ret)
2925                 async->error = ret;
2926 end:
2927         ret = btrfs_end_transaction(trans);
2928         if (ret && !async->error)
2929                 async->error = ret;
2930 done:
2931         if (async->sync)
2932                 complete(&async->wait);
2933         else
2934                 kfree(async);
2935 }
2936 
2937 int btrfs_async_run_delayed_refs(struct btrfs_fs_info *fs_info,
2938                                  unsigned long count, u64 transid, int wait)
2939 {
2940         struct async_delayed_refs *async;
2941         int ret;
2942 
2943         async = kmalloc(sizeof(*async), GFP_NOFS);
2944         if (!async)
2945                 return -ENOMEM;
2946 
2947         async->root = fs_info->tree_root;
2948         async->count = count;
2949         async->error = 0;
2950         async->transid = transid;
2951         if (wait)
2952                 async->sync = 1;
2953         else
2954                 async->sync = 0;
2955         init_completion(&async->wait);
2956 
2957         btrfs_init_work(&async->work, btrfs_extent_refs_helper,
2958                         delayed_ref_async_start, NULL, NULL);
2959 
2960         btrfs_queue_work(fs_info->extent_workers, &async->work);
2961 
2962         if (wait) {
2963                 wait_for_completion(&async->wait);
2964                 ret = async->error;
2965                 kfree(async);
2966                 return ret;
2967         }
2968         return 0;
2969 }
2970 
2971 /*
2972  * this starts processing the delayed reference count updates and
2973  * extent insertions we have queued up so far.  count can be
2974  * 0, which means to process everything in the tree at the start
2975  * of the run (but not newly added entries), or it can be some target
2976  * number you'd like to process.
2977  *
2978  * Returns 0 on success or if called with an aborted transaction
2979  * Returns <0 on error and aborts the transaction
2980  */
2981 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2982                            struct btrfs_fs_info *fs_info, unsigned long count)
2983 {
2984         struct rb_node *node;
2985         struct btrfs_delayed_ref_root *delayed_refs;
2986         struct btrfs_delayed_ref_head *head;
2987         int ret;
2988         int run_all = count == (unsigned long)-1;
2989         bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
2990 
2991         /* We'll clean this up in btrfs_cleanup_transaction */
2992         if (trans->aborted)
2993                 return 0;
2994 
2995         if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags))
2996                 return 0;
2997 
2998         delayed_refs = &trans->transaction->delayed_refs;
2999         if (count == 0)
3000                 count = atomic_read(&delayed_refs->num_entries) * 2;
3001 
3002 again:
3003 #ifdef SCRAMBLE_DELAYED_REFS
3004         delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
3005 #endif
3006         trans->can_flush_pending_bgs = false;
3007         ret = __btrfs_run_delayed_refs(trans, fs_info, count);
3008         if (ret < 0) {
3009                 btrfs_abort_transaction(trans, ret);
3010                 return ret;
3011         }
3012 
3013         if (run_all) {
3014                 if (!list_empty(&trans->new_bgs))
3015                         btrfs_create_pending_block_groups(trans, fs_info);
3016 
3017                 spin_lock(&delayed_refs->lock);
3018                 node = rb_first(&delayed_refs->href_root);
3019                 if (!node) {
3020                         spin_unlock(&delayed_refs->lock);
3021                         goto out;
3022                 }
3023 
3024                 while (node) {
3025                         head = rb_entry(node, struct btrfs_delayed_ref_head,
3026                                         href_node);
3027                         if (btrfs_delayed_ref_is_head(&head->node)) {
3028                                 struct btrfs_delayed_ref_node *ref;
3029 
3030                                 ref = &head->node;
3031                                 refcount_inc(&ref->refs);
3032 
3033                                 spin_unlock(&delayed_refs->lock);
3034                                 /*
3035                                  * Mutex was contended, block until it's
3036                                  * released and try again
3037                                  */
3038                                 mutex_lock(&head->mutex);
3039                                 mutex_unlock(&head->mutex);
3040 
3041                                 btrfs_put_delayed_ref(ref);
3042                                 cond_resched();
3043                                 goto again;
3044                         } else {
3045                                 WARN_ON(1);
3046                         }
3047                         node = rb_next(node);
3048                 }
3049                 spin_unlock(&delayed_refs->lock);
3050                 cond_resched();
3051                 goto again;
3052         }
3053 out:
3054         trans->can_flush_pending_bgs = can_flush_pending_bgs;
3055         return 0;
3056 }
3057 
3058 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
3059                                 struct btrfs_fs_info *fs_info,
3060                                 u64 bytenr, u64 num_bytes, u64 flags,
3061                                 int level, int is_data)
3062 {
3063         struct btrfs_delayed_extent_op *extent_op;
3064         int ret;
3065 
3066         extent_op = btrfs_alloc_delayed_extent_op();
3067         if (!extent_op)
3068                 return -ENOMEM;
3069 
3070         extent_op->flags_to_set = flags;
3071         extent_op->update_flags = true;
3072         extent_op->update_key = false;
3073         extent_op->is_data = is_data ? true : false;
3074         extent_op->level = level;
3075 
3076         ret = btrfs_add_delayed_extent_op(fs_info, trans, bytenr,
3077                                           num_bytes, extent_op);
3078         if (ret)
3079                 btrfs_free_delayed_extent_op(extent_op);
3080         return ret;
3081 }
3082 
3083 static noinline int check_delayed_ref(struct btrfs_root *root,
3084                                       struct btrfs_path *path,
3085                                       u64 objectid, u64 offset, u64 bytenr)
3086 {
3087         struct btrfs_delayed_ref_head *head;
3088         struct btrfs_delayed_ref_node *ref;
3089         struct btrfs_delayed_data_ref *data_ref;
3090         struct btrfs_delayed_ref_root *delayed_refs;
3091         struct btrfs_transaction *cur_trans;
3092         int ret = 0;
3093 
3094         cur_trans = root->fs_info->running_transaction;
3095         if (!cur_trans)
3096                 return 0;
3097 
3098         delayed_refs = &cur_trans->delayed_refs;
3099         spin_lock(&delayed_refs->lock);
3100         head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
3101         if (!head) {
3102                 spin_unlock(&delayed_refs->lock);
3103                 return 0;
3104         }
3105 
3106         if (!mutex_trylock(&head->mutex)) {
3107                 refcount_inc(&head->node.refs);
3108                 spin_unlock(&delayed_refs->lock);
3109 
3110                 btrfs_release_path(path);
3111 
3112                 /*
3113                  * Mutex was contended, block until it's released and let
3114                  * caller try again
3115                  */
3116                 mutex_lock(&head->mutex);
3117                 mutex_unlock(&head->mutex);
3118                 btrfs_put_delayed_ref(&head->node);
3119                 return -EAGAIN;
3120         }
3121         spin_unlock(&delayed_refs->lock);
3122 
3123         spin_lock(&head->lock);
3124         list_for_each_entry(ref, &head->ref_list, list) {
3125                 /* If it's a shared ref we know a cross reference exists */
3126                 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
3127                         ret = 1;
3128                         break;
3129                 }
3130 
3131                 data_ref = btrfs_delayed_node_to_data_ref(ref);
3132 
3133                 /*
3134                  * If our ref doesn't match the one we're currently looking at
3135                  * then we have a cross reference.
3136                  */
3137                 if (data_ref->root != root->root_key.objectid ||
3138                     data_ref->objectid != objectid ||
3139                     data_ref->offset != offset) {
3140                         ret = 1;
3141                         break;
3142                 }
3143         }
3144         spin_unlock(&head->lock);
3145         mutex_unlock(&head->mutex);
3146         return ret;
3147 }
3148 
3149 static noinline int check_committed_ref(struct btrfs_root *root,
3150                                         struct btrfs_path *path,
3151                                         u64 objectid, u64 offset, u64 bytenr)
3152 {
3153         struct btrfs_fs_info *fs_info = root->fs_info;
3154         struct btrfs_root *extent_root = fs_info->extent_root;
3155         struct extent_buffer *leaf;
3156         struct btrfs_extent_data_ref *ref;
3157         struct btrfs_extent_inline_ref *iref;
3158         struct btrfs_extent_item *ei;
3159         struct btrfs_key key;
3160         u32 item_size;
3161         int ret;
3162 
3163         key.objectid = bytenr;
3164         key.offset = (u64)-1;
3165         key.type = BTRFS_EXTENT_ITEM_KEY;
3166 
3167         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
3168         if (ret < 0)
3169                 goto out;
3170         BUG_ON(ret == 0); /* Corruption */
3171 
3172         ret = -ENOENT;
3173         if (path->slots[0] == 0)
3174                 goto out;
3175 
3176         path->slots[0]--;
3177         leaf = path->nodes[0];
3178         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3179 
3180         if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
3181                 goto out;
3182 
3183         ret = 1;
3184         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3185 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3186         if (item_size < sizeof(*ei)) {
3187                 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
3188                 goto out;
3189         }
3190 #endif
3191         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
3192 
3193         if (item_size != sizeof(*ei) +
3194             btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
3195                 goto out;
3196 
3197         if (btrfs_extent_generation(leaf, ei) <=
3198             btrfs_root_last_snapshot(&root->root_item))
3199                 goto out;
3200 
3201         iref = (struct btrfs_extent_inline_ref *)(ei + 1);
3202         if (btrfs_extent_inline_ref_type(leaf, iref) !=
3203             BTRFS_EXTENT_DATA_REF_KEY)
3204                 goto out;
3205 
3206         ref = (struct btrfs_extent_data_ref *)(&iref->offset);
3207         if (btrfs_extent_refs(leaf, ei) !=
3208             btrfs_extent_data_ref_count(leaf, ref) ||
3209             btrfs_extent_data_ref_root(leaf, ref) !=
3210             root->root_key.objectid ||
3211             btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
3212             btrfs_extent_data_ref_offset(leaf, ref) != offset)
3213                 goto out;
3214 
3215         ret = 0;
3216 out:
3217         return ret;
3218 }
3219 
3220 int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset,
3221                           u64 bytenr)
3222 {
3223         struct btrfs_path *path;
3224         int ret;
3225         int ret2;
3226 
3227         path = btrfs_alloc_path();
3228         if (!path)
3229                 return -ENOENT;
3230 
3231         do {
3232                 ret = check_committed_ref(root, path, objectid,
3233                                           offset, bytenr);
3234                 if (ret && ret != -ENOENT)
3235                         goto out;
3236 
3237                 ret2 = check_delayed_ref(root, path, objectid,
3238                                          offset, bytenr);
3239         } while (ret2 == -EAGAIN);
3240 
3241         if (ret2 && ret2 != -ENOENT) {
3242                 ret = ret2;
3243                 goto out;
3244         }
3245 
3246         if (ret != -ENOENT || ret2 != -ENOENT)
3247                 ret = 0;
3248 out:
3249         btrfs_free_path(path);
3250         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3251                 WARN_ON(ret > 0);
3252         return ret;
3253 }
3254 
3255 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3256                            struct btrfs_root *root,
3257                            struct extent_buffer *buf,
3258                            int full_backref, int inc)
3259 {
3260         struct btrfs_fs_info *fs_info = root->fs_info;
3261         u64 bytenr;
3262         u64 num_bytes;
3263         u64 parent;
3264         u64 ref_root;
3265         u32 nritems;
3266         struct btrfs_key key;
3267         struct btrfs_file_extent_item *fi;
3268         int i;
3269         int level;
3270         int ret = 0;
3271         int (*process_func)(struct btrfs_trans_handle *,
3272                             struct btrfs_fs_info *,
3273                             u64, u64, u64, u64, u64, u64);
3274 
3275 
3276         if (btrfs_is_testing(fs_info))
3277                 return 0;
3278 
3279         ref_root = btrfs_header_owner(buf);
3280         nritems = btrfs_header_nritems(buf);
3281         level = btrfs_header_level(buf);
3282 
3283         if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0)
3284                 return 0;
3285 
3286         if (inc)
3287                 process_func = btrfs_inc_extent_ref;
3288         else
3289                 process_func = btrfs_free_extent;
3290 
3291         if (full_backref)
3292                 parent = buf->start;
3293         else
3294                 parent = 0;
3295 
3296         for (i = 0; i < nritems; i++) {
3297                 if (level == 0) {
3298                         btrfs_item_key_to_cpu(buf, &key, i);
3299                         if (key.type != BTRFS_EXTENT_DATA_KEY)
3300                                 continue;
3301                         fi = btrfs_item_ptr(buf, i,
3302                                             struct btrfs_file_extent_item);
3303                         if (btrfs_file_extent_type(buf, fi) ==
3304                             BTRFS_FILE_EXTENT_INLINE)
3305                                 continue;
3306                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3307                         if (bytenr == 0)
3308                                 continue;
3309 
3310                         num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3311                         key.offset -= btrfs_file_extent_offset(buf, fi);
3312                         ret = process_func(trans, fs_info, bytenr, num_bytes,
3313                                            parent, ref_root, key.objectid,
3314                                            key.offset);
3315                         if (ret)
3316                                 goto fail;
3317                 } else {
3318                         bytenr = btrfs_node_blockptr(buf, i);
3319                         num_bytes = fs_info->nodesize;
3320                         ret = process_func(trans, fs_info, bytenr, num_bytes,
3321                                            parent, ref_root, level - 1, 0);
3322                         if (ret)
3323                                 goto fail;
3324                 }
3325         }
3326         return 0;
3327 fail:
3328         return ret;
3329 }
3330 
3331 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3332                   struct extent_buffer *buf, int full_backref)
3333 {
3334         return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
3335 }
3336 
3337 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3338                   struct extent_buffer *buf, int full_backref)
3339 {
3340         return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
3341 }
3342 
3343 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3344                                  struct btrfs_fs_info *fs_info,
3345                                  struct btrfs_path *path,
3346                                  struct btrfs_block_group_cache *cache)
3347 {
3348         int ret;
3349         struct btrfs_root *extent_root = fs_info->extent_root;
3350         unsigned long bi;
3351         struct extent_buffer *leaf;
3352 
3353         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3354         if (ret) {
3355                 if (ret > 0)
3356                         ret = -ENOENT;
3357                 goto fail;
3358         }
3359 
3360         leaf = path->nodes[0];
3361         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3362         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3363         btrfs_mark_buffer_dirty(leaf);
3364 fail:
3365         btrfs_release_path(path);
3366         return ret;
3367 
3368 }
3369 
3370 static struct btrfs_block_group_cache *
3371 next_block_group(struct btrfs_fs_info *fs_info,
3372                  struct btrfs_block_group_cache *cache)
3373 {
3374         struct rb_node *node;
3375 
3376         spin_lock(&fs_info->block_group_cache_lock);
3377 
3378         /* If our block group was removed, we need a full search. */
3379         if (RB_EMPTY_NODE(&cache->cache_node)) {
3380                 const u64 next_bytenr = cache->key.objectid + cache->key.offset;
3381 
3382                 spin_unlock(&fs_info->block_group_cache_lock);
3383                 btrfs_put_block_group(cache);
3384                 cache = btrfs_lookup_first_block_group(fs_info, next_bytenr); return cache;
3385         }
3386         node = rb_next(&cache->cache_node);
3387         btrfs_put_block_group(cache);
3388         if (node) {
3389                 cache = rb_entry(node, struct btrfs_block_group_cache,
3390                                  cache_node);
3391                 btrfs_get_block_group(cache);
3392         } else
3393                 cache = NULL;
3394         spin_unlock(&fs_info->block_group_cache_lock);
3395         return cache;
3396 }
3397 
3398 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3399                             struct btrfs_trans_handle *trans,
3400                             struct btrfs_path *path)
3401 {
3402         struct btrfs_fs_info *fs_info = block_group->fs_info;
3403         struct btrfs_root *root = fs_info->tree_root;
3404         struct inode *inode = NULL;
3405         struct extent_changeset *data_reserved = NULL;
3406         u64 alloc_hint = 0;
3407         int dcs = BTRFS_DC_ERROR;
3408         u64 num_pages = 0;
3409         int retries = 0;
3410         int ret = 0;
3411 
3412         /*
3413          * If this block group is smaller than 100 megs don't bother caching the
3414          * block group.
3415          */
3416         if (block_group->key.offset < (100 * SZ_1M)) {
3417                 spin_lock(&block_group->lock);
3418                 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3419                 spin_unlock(&block_group->lock);
3420                 return 0;
3421         }
3422 
3423         if (trans->aborted)
3424                 return 0;
3425 again:
3426         inode = lookup_free_space_inode(fs_info, block_group, path);
3427         if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3428                 ret = PTR_ERR(inode);
3429                 btrfs_release_path(path);
3430                 goto out;
3431         }
3432 
3433         if (IS_ERR(inode)) {
3434                 BUG_ON(retries);
3435                 retries++;
3436 
3437                 if (block_group->ro)
3438                         goto out_free;
3439 
3440                 ret = create_free_space_inode(fs_info, trans, block_group,
3441                                               path);
3442                 if (ret)
3443                         goto out_free;
3444                 goto again;
3445         }
3446 
3447         /* We've already setup this transaction, go ahead and exit */
3448         if (block_group->cache_generation == trans->transid &&
3449             i_size_read(inode)) {
3450                 dcs = BTRFS_DC_SETUP;
3451                 goto out_put;
3452         }
3453 
3454         /*
3455          * We want to set the generation to 0, that way if anything goes wrong
3456          * from here on out we know not to trust this cache when we load up next
3457          * time.
3458          */
3459         BTRFS_I(inode)->generation = 0;
3460         ret = btrfs_update_inode(trans, root, inode);
3461         if (ret) {
3462                 /*
3463                  * So theoretically we could recover from this, simply set the
3464                  * super cache generation to 0 so we know to invalidate the
3465                  * cache, but then we'd have to keep track of the block groups
3466                  * that fail this way so we know we _have_ to reset this cache
3467                  * before the next commit or risk reading stale cache.  So to
3468                  * limit our exposure to horrible edge cases lets just abort the
3469                  * transaction, this only happens in really bad situations
3470                  * anyway.
3471                  */
3472                 btrfs_abort_transaction(trans, ret);
3473                 goto out_put;
3474         }
3475         WARN_ON(ret);
3476 
3477         if (i_size_read(inode) > 0) {
3478                 ret = btrfs_check_trunc_cache_free_space(fs_info,
3479                                         &fs_info->global_block_rsv);
3480                 if (ret)
3481                         goto out_put;
3482 
3483                 ret = btrfs_truncate_free_space_cache(trans, NULL, inode);
3484                 if (ret)
3485                         goto out_put;
3486         }
3487 
3488         spin_lock(&block_group->lock);
3489         if (block_group->cached != BTRFS_CACHE_FINISHED ||
3490             !btrfs_test_opt(fs_info, SPACE_CACHE)) {
3491                 /*
3492                  * don't bother trying to write stuff out _if_
3493                  * a) we're not cached,
3494                  * b) we're with nospace_cache mount option,
3495                  * c) we're with v2 space_cache (FREE_SPACE_TREE).
3496                  */
3497                 dcs = BTRFS_DC_WRITTEN;
3498                 spin_unlock(&block_group->lock);
3499                 goto out_put;
3500         }
3501         spin_unlock(&block_group->lock);
3502 
3503         /*
3504          * We hit an ENOSPC when setting up the cache in this transaction, just
3505          * skip doing the setup, we've already cleared the cache so we're safe.
3506          */
3507         if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) {
3508                 ret = -ENOSPC;
3509                 goto out_put;
3510         }
3511 
3512         /*
3513          * Try to preallocate enough space based on how big the block group is.
3514          * Keep in mind this has to include any pinned space which could end up
3515          * taking up quite a bit since it's not folded into the other space
3516          * cache.
3517          */
3518         num_pages = div_u64(block_group->key.offset, SZ_256M);
3519         if (!num_pages)
3520                 num_pages = 1;
3521 
3522         num_pages *= 16;
3523         num_pages *= PAGE_SIZE;
3524 
3525         ret = btrfs_check_data_free_space(inode, &data_reserved, 0, num_pages);
3526         if (ret)
3527                 goto out_put;
3528 
3529         ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3530                                               num_pages, num_pages,
3531                                               &alloc_hint);
3532         /*
3533          * Our cache requires contiguous chunks so that we don't modify a bunch
3534          * of metadata or split extents when writing the cache out, which means
3535          * we can enospc if we are heavily fragmented in addition to just normal
3536          * out of space conditions.  So if we hit this just skip setting up any
3537          * other block groups for this transaction, maybe we'll unpin enough
3538          * space the next time around.
3539          */
3540         if (!ret)
3541                 dcs = BTRFS_DC_SETUP;
3542         else if (ret == -ENOSPC)
3543                 set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags);
3544 
3545 out_put:
3546         iput(inode);
3547 out_free:
3548         btrfs_release_path(path);
3549 out:
3550         spin_lock(&block_group->lock);
3551         if (!ret && dcs == BTRFS_DC_SETUP)
3552                 block_group->cache_generation = trans->transid;
3553         block_group->disk_cache_state = dcs;
3554         spin_unlock(&block_group->lock);
3555 
3556         extent_changeset_free(data_reserved);
3557         return ret;
3558 }
3559 
3560 int btrfs_setup_space_cache(struct btrfs_trans_handle *trans,
3561                             struct btrfs_fs_info *fs_info)
3562 {
3563         struct btrfs_block_group_cache *cache, *tmp;
3564         struct btrfs_transaction *cur_trans = trans->transaction;
3565         struct btrfs_path *path;
3566 
3567         if (list_empty(&cur_trans->dirty_bgs) ||
3568             !btrfs_test_opt(fs_info, SPACE_CACHE))
3569                 return 0;
3570 
3571         path = btrfs_alloc_path();
3572         if (!path)
3573                 return -ENOMEM;
3574 
3575         /* Could add new block groups, use _safe just in case */
3576         list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs,
3577                                  dirty_list) {
3578                 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3579                         cache_save_setup(cache, trans, path);
3580         }
3581 
3582         btrfs_free_path(path);
3583         return 0;
3584 }
3585 
3586 /*
3587  * transaction commit does final block group cache writeback during a
3588  * critical section where nothing is allowed to change the FS.  This is
3589  * required in order for the cache to actually match the block group,
3590  * but can introduce a lot of latency into the commit.
3591  *
3592  * So, btrfs_start_dirty_block_groups is here to kick off block group
3593  * cache IO.  There's a chance we'll have to redo some of it if the
3594  * block group changes again during the commit, but it greatly reduces
3595  * the commit latency by getting rid of the easy block groups while
3596  * we're still allowing others to join the commit.
3597  */
3598 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans,
3599                                    struct btrfs_fs_info *fs_info)
3600 {
3601         struct btrfs_block_group_cache *cache;
3602         struct btrfs_transaction *cur_trans = trans->transaction;
3603         int ret = 0;
3604         int should_put;
3605         struct btrfs_path *path = NULL;
3606         LIST_HEAD(dirty);
3607         struct list_head *io = &cur_trans->io_bgs;
3608         int num_started = 0;
3609         int loops = 0;
3610 
3611         spin_lock(&cur_trans->dirty_bgs_lock);
3612         if (list_empty(&cur_trans->dirty_bgs)) {
3613                 spin_unlock(&cur_trans->dirty_bgs_lock);
3614                 return 0;
3615         }
3616         list_splice_init(&cur_trans->dirty_bgs, &dirty);
3617         spin_unlock(&cur_trans->dirty_bgs_lock);
3618 
3619 again:
3620         /*
3621          * make sure all the block groups on our dirty list actually
3622          * exist
3623          */
3624         btrfs_create_pending_block_groups(trans, fs_info);
3625 
3626         if (!path) {
3627                 path = btrfs_alloc_path();
3628                 if (!path)
3629                         return -ENOMEM;
3630         }
3631 
3632         /*
3633          * cache_write_mutex is here only to save us from balance or automatic
3634          * removal of empty block groups deleting this block group while we are
3635          * writing out the cache
3636          */
3637         mutex_lock(&trans->transaction->cache_write_mutex);
3638         while (!list_empty(&dirty)) {
3639                 cache = list_first_entry(&dirty,
3640                                          struct btrfs_block_group_cache,
3641                                          dirty_list);
3642                 /*
3643                  * this can happen if something re-dirties a block
3644                  * group that is already under IO.  Just wait for it to
3645                  * finish and then do it all again
3646                  */
3647                 if (!list_empty(&cache->io_list)) {
3648                         list_del_init(&cache->io_list);
3649                         btrfs_wait_cache_io(trans, cache, path);
3650                         btrfs_put_block_group(cache);
3651                 }
3652 
3653 
3654                 /*
3655                  * btrfs_wait_cache_io uses the cache->dirty_list to decide
3656                  * if it should update the cache_state.  Don't delete
3657                  * until after we wait.
3658                  *
3659                  * Since we're not running in the commit critical section
3660                  * we need the dirty_bgs_lock to protect from update_block_group
3661                  */
3662                 spin_lock(&cur_trans->dirty_bgs_lock);
3663                 list_del_init(&cache->dirty_list);
3664                 spin_unlock(&cur_trans->dirty_bgs_lock);
3665 
3666                 should_put = 1;
3667 
3668                 cache_save_setup(cache, trans, path);
3669 
3670                 if (cache->disk_cache_state == BTRFS_DC_SETUP) {
3671                         cache->io_ctl.inode = NULL;
3672                         ret = btrfs_write_out_cache(fs_info, trans,
3673                                                     cache, path);
3674                         if (ret == 0 && cache->io_ctl.inode) {
3675                                 num_started++;
3676                                 should_put = 0;
3677 
3678                                 /*
3679                                  * the cache_write_mutex is protecting
3680                                  * the io_list
3681                                  */
3682                                 list_add_tail(&cache->io_list, io);
3683                         } else {
3684                                 /*
3685                                  * if we failed to write the cache, the
3686                                  * generation will be bad and life goes on
3687                                  */
3688                                 ret = 0;
3689                         }
3690                 }
3691                 if (!ret) {
3692                         ret = write_one_cache_group(trans, fs_info,
3693                                                     path, cache);
3694                         /*
3695                          * Our block group might still be attached to the list
3696                          * of new block groups in the transaction handle of some
3697                          * other task (struct btrfs_trans_handle->new_bgs). This
3698                          * means its block group item isn't yet in the extent
3699                          * tree. If this happens ignore the error, as we will
3700                          * try again later in the critical section of the
3701                          * transaction commit.
3702                          */
3703                         if (ret == -ENOENT) {
3704                                 ret = 0;
3705                                 spin_lock(&cur_trans->dirty_bgs_lock);
3706                                 if (list_empty(&cache->dirty_list)) {
3707                                         list_add_tail(&cache->dirty_list,
3708                                                       &cur_trans->dirty_bgs);
3709                                         btrfs_get_block_group(cache);
3710                                 }
3711                                 spin_unlock(&cur_trans->dirty_bgs_lock);
3712                         } else if (ret) {
3713                                 btrfs_abort_transaction(trans, ret);
3714                         }
3715                 }
3716 
3717                 /* if its not on the io list, we need to put the block group */
3718                 if (should_put)
3719                         btrfs_put_block_group(cache);
3720 
3721                 if (ret)
3722                         break;
3723 
3724                 /*
3725                  * Avoid blocking other tasks for too long. It might even save
3726                  * us from writing caches for block groups that are going to be
3727                  * removed.
3728                  */
3729                 mutex_unlock(&trans->transaction->cache_write_mutex);
3730                 mutex_lock(&trans->transaction->cache_write_mutex);
3731         }
3732         mutex_unlock(&trans->transaction->cache_write_mutex);
3733 
3734         /*
3735          * go through delayed refs for all the stuff we've just kicked off
3736          * and then loop back (just once)
3737          */
3738         ret = btrfs_run_delayed_refs(trans, fs_info, 0);
3739         if (!ret && loops == 0) {
3740                 loops++;
3741                 spin_lock(&cur_trans->dirty_bgs_lock);
3742                 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3743                 /*
3744                  * dirty_bgs_lock protects us from concurrent block group
3745                  * deletes too (not just cache_write_mutex).
3746                  */
3747                 if (!list_empty(&dirty)) {
3748                         spin_unlock(&cur_trans->dirty_bgs_lock);
3749                         goto again;
3750                 }
3751                 spin_unlock(&cur_trans->dirty_bgs_lock);
3752         } else if (ret < 0) {
3753                 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
3754         }
3755 
3756         btrfs_free_path(path);
3757         return ret;
3758 }
3759 
3760 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3761                                    struct btrfs_fs_info *fs_info)
3762 {
3763         struct btrfs_block_group_cache *cache;
3764         struct btrfs_transaction *cur_trans = trans->transaction;
3765         int ret = 0;
3766         int should_put;
3767         struct btrfs_path *path;
3768         struct list_head *io = &cur_trans->io_bgs;
3769         int num_started = 0;
3770 
3771         path = btrfs_alloc_path();
3772         if (!path)
3773                 return -ENOMEM;
3774 
3775         /*
3776          * Even though we are in the critical section of the transaction commit,
3777          * we can still have concurrent tasks adding elements to this
3778          * transaction's list of dirty block groups. These tasks correspond to
3779          * endio free space workers started when writeback finishes for a
3780          * space cache, which run inode.c:btrfs_finish_ordered_io(), and can
3781          * allocate new block groups as a result of COWing nodes of the root
3782          * tree when updating the free space inode. The writeback for the space
3783          * caches is triggered by an earlier call to
3784          * btrfs_start_dirty_block_groups() and iterations of the following
3785          * loop.
3786          * Also we want to do the cache_save_setup first and then run the
3787          * delayed refs to make sure we have the best chance at doing this all
3788          * in one shot.
3789          */
3790         spin_lock(&cur_trans->dirty_bgs_lock);
3791         while (!list_empty(&cur_trans->dirty_bgs)) {
3792                 cache = list_first_entry(&cur_trans->dirty_bgs,
3793                                          struct btrfs_block_group_cache,
3794                                          dirty_list);
3795 
3796                 /*
3797                  * this can happen if cache_save_setup re-dirties a block
3798                  * group that is already under IO.  Just wait for it to
3799                  * finish and then do it all again
3800                  */
3801                 if (!list_empty(&cache->io_list)) {
3802                         spin_unlock(&cur_trans->dirty_bgs_lock);
3803                         list_del_init(&cache->io_list);
3804                         btrfs_wait_cache_io(trans, cache, path);
3805                         btrfs_put_block_group(cache);
3806                         spin_lock(&cur_trans->dirty_bgs_lock);
3807                 }
3808 
3809                 /*
3810                  * don't remove from the dirty list until after we've waited
3811                  * on any pending IO
3812                  */
3813                 list_del_init(&cache->dirty_list);
3814                 spin_unlock(&cur_trans->dirty_bgs_lock);
3815                 should_put = 1;
3816 
3817                 cache_save_setup(cache, trans, path);
3818 
3819                 if (!ret)
3820                         ret = btrfs_run_delayed_refs(trans, fs_info,
3821                                                      (unsigned long) -1);
3822 
3823                 if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) {
3824                         cache->io_ctl.inode = NULL;
3825                         ret = btrfs_write_out_cache(fs_info, trans,
3826                                                     cache, path);
3827                         if (ret == 0 && cache->io_ctl.inode) {
3828                                 num_started++;
3829                                 should_put = 0;
3830                                 list_add_tail(&cache->io_list, io);
3831                         } else {
3832                                 /*
3833                                  * if we failed to write the cache, the
3834                                  * generation will be bad and life goes on
3835                                  */
3836                                 ret = 0;
3837                         }
3838                 }
3839                 if (!ret) {
3840                         ret = write_one_cache_group(trans, fs_info,
3841                                                     path, cache);
3842                         /*
3843                          * One of the free space endio workers might have
3844                          * created a new block group while updating a free space
3845                          * cache's inode (at inode.c:btrfs_finish_ordered_io())
3846                          * and hasn't released its transaction handle yet, in
3847                          * which case the new block group is still attached to
3848                          * its transaction handle and its creation has not
3849                          * finished yet (no block group item in the extent tree
3850                          * yet, etc). If this is the case, wait for all free
3851                          * space endio workers to finish and retry. This is a
3852                          * a very rare case so no need for a more efficient and
3853                          * complex approach.
3854                          */
3855                         if (ret == -ENOENT) {
3856                                 wait_event(cur_trans->writer_wait,
3857                                    atomic_read(&cur_trans->num_writers) == 1);
3858                                 ret = write_one_cache_group(trans, fs_info,
3859                                                             path, cache);
3860                         }
3861                         if (ret)
3862                                 btrfs_abort_transaction(trans, ret);
3863                 }
3864 
3865                 /* if its not on the io list, we need to put the block group */
3866                 if (should_put)
3867                         btrfs_put_block_group(cache);
3868                 spin_lock(&cur_trans->dirty_bgs_lock);
3869         }
3870         spin_unlock(&cur_trans->dirty_bgs_lock);
3871 
3872         while (!list_empty(io)) {
3873                 cache = list_first_entry(io, struct btrfs_block_group_cache,
3874                                          io_list);
3875                 list_del_init(&cache->io_list);
3876                 btrfs_wait_cache_io(trans, cache, path);
3877                 btrfs_put_block_group(cache);
3878         }
3879 
3880         btrfs_free_path(path);
3881         return ret;
3882 }
3883 
3884 int btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr)
3885 {
3886         struct btrfs_block_group_cache *block_group;
3887         int readonly = 0;
3888 
3889         block_group = btrfs_lookup_block_group(fs_info, bytenr);
3890         if (!block_group || block_group->ro)
3891                 readonly = 1;
3892         if (block_group)
3893                 btrfs_put_block_group(block_group);
3894         return readonly;
3895 }
3896 
3897 bool btrfs_inc_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
3898 {
3899         struct btrfs_block_group_cache *bg;
3900         bool ret = true;
3901 
3902         bg = btrfs_lookup_block_group(fs_info, bytenr);
3903         if (!bg)
3904                 return false;
3905 
3906         spin_lock(&bg->lock);
3907         if (bg->ro)
3908                 ret = false;
3909         else
3910                 atomic_inc(&bg->nocow_writers);
3911         spin_unlock(&bg->lock);
3912 
3913         /* no put on block group, done by btrfs_dec_nocow_writers */
3914         if (!ret)
3915                 btrfs_put_block_group(bg);
3916 
3917         return ret;
3918 
3919 }
3920 
3921 void btrfs_dec_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
3922 {
3923         struct btrfs_block_group_cache *bg;
3924 
3925         bg = btrfs_lookup_block_group(fs_info, bytenr);
3926         ASSERT(bg);
3927         if (atomic_dec_and_test(&bg->nocow_writers))
3928                 wake_up_atomic_t(&bg->nocow_writers);
3929         /*
3930          * Once for our lookup and once for the lookup done by a previous call
3931          * to btrfs_inc_nocow_writers()
3932          */
3933         btrfs_put_block_group(bg);
3934         btrfs_put_block_group(bg);
3935 }
3936 
3937 static int btrfs_wait_nocow_writers_atomic_t(atomic_t *a)
3938 {
3939         schedule();
3940         return 0;
3941 }
3942 
3943 void btrfs_wait_nocow_writers(struct btrfs_block_group_cache *bg)
3944 {
3945         wait_on_atomic_t(&bg->nocow_writers,
3946                          btrfs_wait_nocow_writers_atomic_t,
3947                          TASK_UNINTERRUPTIBLE);
3948 }
3949 
3950 static const char *alloc_name(u64 flags)
3951 {
3952         switch (flags) {
3953         case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
3954                 return "mixed";
3955         case BTRFS_BLOCK_GROUP_METADATA:
3956                 return "metadata";
3957         case BTRFS_BLOCK_GROUP_DATA:
3958                 return "data";
3959         case BTRFS_BLOCK_GROUP_SYSTEM:
3960                 return "system";
3961         default:
3962                 WARN_ON(1);
3963                 return "invalid-combination";
3964         };
3965 }
3966 
3967 static int create_space_info(struct btrfs_fs_info *info, u64 flags,
3968                              struct btrfs_space_info **new)
3969 {
3970 
3971         struct btrfs_space_info *space_info;
3972         int i;
3973         int ret;
3974 
3975         space_info = kzalloc(sizeof(*space_info), GFP_NOFS);
3976         if (!space_info)
3977                 return -ENOMEM;
3978 
3979         ret = percpu_counter_init(&space_info->total_bytes_pinned, 0,
3980                                  GFP_KERNEL);
3981         if (ret) {
3982                 kfree(space_info);
3983                 return ret;
3984         }
3985 
3986         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3987                 INIT_LIST_HEAD(&space_info->block_groups[i]);
3988         init_rwsem(&space_info->groups_sem);
3989         spin_lock_init(&space_info->lock);
3990         space_info->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3991         space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3992         init_waitqueue_head(&space_info->wait);
3993         INIT_LIST_HEAD(&space_info->ro_bgs);
3994         INIT_LIST_HEAD(&space_info->tickets);
3995         INIT_LIST_HEAD(&space_info->priority_tickets);
3996 
3997         ret = kobject_init_and_add(&space_info->kobj, &space_info_ktype,
3998                                     info->space_info_kobj, "%s",
3999                                     alloc_name(space_info->flags));
4000         if (ret) {
4001                 percpu_counter_destroy(&space_info->total_bytes_pinned);
4002                 kfree(space_info);
4003                 return ret;
4004         }
4005 
4006         *new = space_info;
4007         list_add_rcu(&space_info->list, &info->space_info);
4008         if (flags & BTRFS_BLOCK_GROUP_DATA)
4009                 info->data_sinfo = space_info;
4010 
4011         return ret;
4012 }
4013 
4014 static void update_space_info(struct btrfs_fs_info *info, u64 flags,
4015                              u64 total_bytes, u64 bytes_used,
4016                              u64 bytes_readonly,
4017                              struct btrfs_space_info **space_info)
4018 {
4019         struct btrfs_space_info *found;
4020         int factor;
4021 
4022         if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
4023                      BTRFS_BLOCK_GROUP_RAID10))
4024                 factor = 2;
4025         else
4026                 factor = 1;
4027 
4028         found = __find_space_info(info, flags);
4029         ASSERT(found);
4030         spin_lock(&found->lock);
4031         found->total_bytes += total_bytes;
4032         found->disk_total += total_bytes * factor;
4033         found->bytes_used += bytes_used;
4034         found->disk_used += bytes_used * factor;
4035         found->bytes_readonly += bytes_readonly;
4036         if (total_bytes > 0)
4037                 found->full = 0;
4038         space_info_add_new_bytes(info, found, total_bytes -
4039                                  bytes_used - bytes_readonly);
4040         spin_unlock(&found->lock);
4041         *space_info = found;
4042 }
4043 
4044 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
4045 {
4046         u64 extra_flags = chunk_to_extended(flags) &
4047                                 BTRFS_EXTENDED_PROFILE_MASK;
4048 
4049         write_seqlock(&fs_info->profiles_lock);
4050         if (flags & BTRFS_BLOCK_GROUP_DATA)
4051                 fs_info->avail_data_alloc_bits |= extra_flags;
4052         if (flags & BTRFS_BLOCK_GROUP_METADATA)
4053                 fs_info->avail_metadata_alloc_bits |= extra_flags;
4054         if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
4055                 fs_info->avail_system_alloc_bits |= extra_flags;
4056         write_sequnlock(&fs_info->profiles_lock);
4057 }
4058 
4059 /*
4060  * returns target flags in extended format or 0 if restripe for this
4061  * chunk_type is not in progress
4062  *
4063  * should be called with either volume_mutex or balance_lock held
4064  */
4065 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
4066 {
4067         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4068         u64 target = 0;
4069 
4070         if (!bctl)
4071                 return 0;
4072 
4073         if (flags & BTRFS_BLOCK_GROUP_DATA &&
4074             bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4075                 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
4076         } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
4077                    bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4078                 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
4079         } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
4080                    bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4081                 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
4082         }
4083 
4084         return target;
4085 }
4086 
4087 /*
4088  * @flags: available profiles in extended format (see ctree.h)
4089  *
4090  * Returns reduced profile in chunk format.  If profile changing is in
4091  * progress (either running or paused) picks the target profile (if it's
4092  * already available), otherwise falls back to plain reducing.
4093  */
4094 static u64 btrfs_reduce_alloc_profile(struct btrfs_fs_info *fs_info, u64 flags)
4095 {
4096         u64 num_devices = fs_info->fs_devices->rw_devices;
4097         u64 target;
4098         u64 raid_type;
4099         u64 allowed = 0;
4100 
4101         /*
4102          * see if restripe for this chunk_type is in progress, if so
4103          * try to reduce to the target profile
4104          */
4105         spin_lock(&fs_info->balance_lock);
4106         target = get_restripe_target(fs_info, flags);
4107         if (target) {
4108                 /* pick target profile only if it's already available */
4109                 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
4110                         spin_unlock(&fs_info->balance_lock);
4111                         return extended_to_chunk(target);
4112                 }
4113         }
4114         spin_unlock(&fs_info->balance_lock);
4115 
4116         /* First, mask out the RAID levels which aren't possible */
4117         for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
4118                 if (num_devices >= btrfs_raid_array[raid_type].devs_min)
4119                         allowed |= btrfs_raid_group[raid_type];
4120         }
4121         allowed &= flags;
4122 
4123         if (allowed & BTRFS_BLOCK_GROUP_RAID6)
4124                 allowed = BTRFS_BLOCK_GROUP_RAID6;
4125         else if (allowed & BTRFS_BLOCK_GROUP_RAID5)
4126                 allowed = BTRFS_BLOCK_GROUP_RAID5;
4127         else if (allowed & BTRFS_BLOCK_GROUP_RAID10)
4128                 allowed = BTRFS_BLOCK_GROUP_RAID10;
4129         else if (allowed & BTRFS_BLOCK_GROUP_RAID1)
4130                 allowed = BTRFS_BLOCK_GROUP_RAID1;
4131         else if (allowed & BTRFS_BLOCK_GROUP_RAID0)
4132                 allowed = BTRFS_BLOCK_GROUP_RAID0;
4133 
4134         flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK;
4135 
4136         return extended_to_chunk(flags | allowed);
4137 }
4138 
4139 static u64 get_alloc_profile(struct btrfs_fs_info *fs_info, u64 orig_flags)
4140 {
4141         unsigned seq;
4142         u64 flags;
4143 
4144         do {
4145                 flags = orig_flags;
4146                 seq = read_seqbegin(&fs_info->profiles_lock);
4147 
4148                 if (flags & BTRFS_BLOCK_GROUP_DATA)
4149                         flags |= fs_info->avail_data_alloc_bits;
4150                 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
4151                         flags |= fs_info->avail_system_alloc_bits;
4152                 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
4153                         flags |= fs_info->avail_metadata_alloc_bits;
4154         } while (read_seqretry(&fs_info->profiles_lock, seq));
4155 
4156         return btrfs_reduce_alloc_profile(fs_info, flags);
4157 }
4158 
4159 static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data)
4160 {
4161         struct btrfs_fs_info *fs_info = root->fs_info;
4162         u64 flags;
4163         u64 ret;
4164 
4165         if (data)
4166                 flags = BTRFS_BLOCK_GROUP_DATA;
4167         else if (root == fs_info->chunk_root)
4168                 flags = BTRFS_BLOCK_GROUP_SYSTEM;
4169         else
4170                 flags = BTRFS_BLOCK_GROUP_METADATA;
4171 
4172         ret = get_alloc_profile(fs_info, flags);
4173         return ret;
4174 }
4175 
4176 u64 btrfs_data_alloc_profile(struct btrfs_fs_info *fs_info)
4177 {
4178         return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_DATA);
4179 }
4180 
4181 u64 btrfs_metadata_alloc_profile(struct btrfs_fs_info *fs_info)
4182 {
4183         return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4184 }
4185 
4186 u64 btrfs_system_alloc_profile(struct btrfs_fs_info *fs_info)
4187 {
4188         return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4189 }
4190 
4191 static u64 btrfs_space_info_used(struct btrfs_space_info *s_info,
4192                                  bool may_use_included)
4193 {
4194         ASSERT(s_info);
4195         return s_info->bytes_used + s_info->bytes_reserved +
4196                 s_info->bytes_pinned + s_info->bytes_readonly +
4197                 (may_use_included ? s_info->bytes_may_use : 0);
4198 }
4199 
4200 int btrfs_alloc_data_chunk_ondemand(struct btrfs_inode *inode, u64 bytes)
4201 {
4202         struct btrfs_space_info *data_sinfo;
4203         struct btrfs_root *root = inode->root;
4204         struct btrfs_fs_info *fs_info = root->fs_info;
4205         u64 used;
4206         int ret = 0;
4207         int need_commit = 2;
4208         int have_pinned_space;
4209 
4210         /* make sure bytes are sectorsize aligned */
4211         bytes = ALIGN(bytes, fs_info->sectorsize);
4212 
4213         if (btrfs_is_free_space_inode(inode)) {
4214                 need_commit = 0;
4215                 ASSERT(current->journal_info);
4216         }
4217 
4218         data_sinfo = fs_info->data_sinfo;
4219         if (!data_sinfo)
4220                 goto alloc;
4221 
4222 again:
4223         /* make sure we have enough space to handle the data first */
4224         spin_lock(&data_sinfo->lock);
4225         used = btrfs_space_info_used(data_sinfo, true);
4226 
4227         if (used + bytes > data_sinfo->total_bytes) {
4228                 struct btrfs_trans_handle *trans;
4229 
4230                 /*
4231                  * if we don't have enough free bytes in this space then we need
4232                  * to alloc a new chunk.
4233                  */
4234                 if (!data_sinfo->full) {
4235                         u64 alloc_target;
4236 
4237                         data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
4238                         spin_unlock(&data_sinfo->lock);
4239 alloc:
4240                         alloc_target = btrfs_data_alloc_profile(fs_info);
4241                         /*
4242                          * It is ugly that we don't call nolock join
4243                          * transaction for the free space inode case here.
4244                          * But it is safe because we only do the data space
4245                          * reservation for the free space cache in the
4246                          * transaction context, the common join transaction
4247                          * just increase the counter of the current transaction
4248                          * handler, doesn't try to acquire the trans_lock of
4249                          * the fs.
4250                          */
4251                         trans = btrfs_join_transaction(root);
4252                         if (IS_ERR(trans))
4253                                 return PTR_ERR(trans);
4254 
4255                         ret = do_chunk_alloc(trans, fs_info, alloc_target,
4256                                              CHUNK_ALLOC_NO_FORCE);
4257                         btrfs_end_transaction(trans);
4258                         if (ret < 0) {
4259                                 if (ret != -ENOSPC)
4260                                         return ret;
4261                                 else {
4262                                         have_pinned_space = 1;
4263                                         goto commit_trans;
4264                                 }
4265                         }
4266 
4267                         if (!data_sinfo)
4268                                 data_sinfo = fs_info->data_sinfo;
4269 
4270                         goto again;
4271                 }
4272 
4273                 /*
4274                  * If we don't have enough pinned space to deal with this
4275                  * allocation, and no removed chunk in current transaction,
4276                  * don't bother committing the transaction.
4277                  */
4278                 have_pinned_space = percpu_counter_compare(
4279                         &data_sinfo->total_bytes_pinned,
4280                         used + bytes - data_sinfo->total_bytes);
4281                 spin_unlock(&data_sinfo->lock);
4282 
4283                 /* commit the current transaction and try again */
4284 commit_trans:
4285                 if (need_commit &&
4286                     !atomic_read(&fs_info->open_ioctl_trans)) {
4287                         need_commit--;
4288 
4289                         if (need_commit > 0) {
4290                                 btrfs_start_delalloc_roots(fs_info, 0, -1);
4291                                 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0,
4292                                                          (u64)-1);
4293                         }
4294 
4295                         trans = btrfs_join_transaction(root);
4296                         if (IS_ERR(trans))
4297                                 return PTR_ERR(trans);
4298                         if (have_pinned_space >= 0 ||
4299                             test_bit(BTRFS_TRANS_HAVE_FREE_BGS,
4300                                      &trans->transaction->flags) ||
4301                             need_commit > 0) {
4302                                 ret = btrfs_commit_transaction(trans);
4303                                 if (ret)
4304                                         return ret;
4305                                 /*
4306                                  * The cleaner kthread might still be doing iput
4307                                  * operations. Wait for it to finish so that
4308                                  * more space is released.
4309                                  */
4310                                 mutex_lock(&fs_info->cleaner_delayed_iput_mutex);
4311                                 mutex_unlock(&fs_info->cleaner_delayed_iput_mutex);
4312                                 goto again;
4313                         } else {
4314                                 btrfs_end_transaction(trans);
4315                         }
4316                 }
4317 
4318                 trace_btrfs_space_reservation(fs_info,
4319                                               "space_info:enospc",
4320                                               data_sinfo->flags, bytes, 1);
4321                 return -ENOSPC;
4322         }
4323         data_sinfo->bytes_may_use += bytes;
4324         trace_btrfs_space_reservation(fs_info, "space_info",
4325                                       data_sinfo->flags, bytes, 1);
4326         spin_unlock(&data_sinfo->lock);
4327 
4328         return ret;
4329 }
4330 
4331 int btrfs_check_data_free_space(struct inode *inode,
4332                         struct extent_changeset **reserved, u64 start, u64 len)
4333 {
4334         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4335         int ret;
4336 
4337         /* align the range */
4338         len = round_up(start + len, fs_info->sectorsize) -
4339               round_down(start, fs_info->sectorsize);
4340         start = round_down(start, fs_info->sectorsize);
4341 
4342         ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode), len);
4343         if (ret < 0)
4344                 return ret;
4345 
4346         /* Use new btrfs_qgroup_reserve_data to reserve precious data space. */
4347         ret = btrfs_qgroup_reserve_data(inode, reserved, start, len);
4348         if (ret < 0)
4349                 btrfs_free_reserved_data_space_noquota(inode, start, len);
4350         else
4351                 ret = 0;
4352         return ret;
4353 }
4354 
4355 /*
4356  * Called if we need to clear a data reservation for this inode
4357  * Normally in a error case.
4358  *
4359  * This one will *NOT* use accurate qgroup reserved space API, just for case
4360  * which we can't sleep and is sure it won't affect qgroup reserved space.
4361  * Like clear_bit_hook().
4362  */
4363 void btrfs_free_reserved_data_space_noquota(struct inode *inode, u64 start,
4364                                             u64 len)
4365 {
4366         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4367         struct btrfs_space_info *data_sinfo;
4368 
4369         /* Make sure the range is aligned to sectorsize */
4370         len = round_up(start + len, fs_info->sectorsize) -
4371               round_down(start, fs_info->sectorsize);
4372         start = round_down(start, fs_info->sectorsize);
4373 
4374         data_sinfo = fs_info->data_sinfo;
4375         spin_lock(&data_sinfo->lock);
4376         if (WARN_ON(data_sinfo->bytes_may_use < len))
4377                 data_sinfo->bytes_may_use = 0;
4378         else
4379                 data_sinfo->bytes_may_use -= len;
4380         trace_btrfs_space_reservation(fs_info, "space_info",
4381                                       data_sinfo->flags, len, 0);
4382         spin_unlock(&data_sinfo->lock);
4383 }
4384 
4385 /*
4386  * Called if we need to clear a data reservation for this inode
4387  * Normally in a error case.
4388  *
4389  * This one will handle the per-inode data rsv map for accurate reserved
4390  * space framework.
4391  */
4392 void btrfs_free_reserved_data_space(struct inode *inode,
4393                         struct extent_changeset *reserved, u64 start, u64 len)
4394 {
4395         struct btrfs_root *root = BTRFS_I(inode)->root;
4396 
4397         /* Make sure the range is aligned to sectorsize */
4398         len = round_up(start + len, root->fs_info->sectorsize) -
4399               round_down(start, root->fs_info->sectorsize);
4400         start = round_down(start, root->fs_info->sectorsize);
4401 
4402         btrfs_free_reserved_data_space_noquota(inode, start, len);
4403         btrfs_qgroup_free_data(inode, reserved, start, len);
4404 }
4405 
4406 static void force_metadata_allocation(struct btrfs_fs_info *info)
4407 {
4408         struct list_head *head = &info->space_info;
4409         struct btrfs_space_info *found;
4410 
4411         rcu_read_lock();
4412         list_for_each_entry_rcu(found, head, list) {
4413                 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
4414                         found->force_alloc = CHUNK_ALLOC_FORCE;
4415         }
4416         rcu_read_unlock();
4417 }
4418 
4419 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
4420 {
4421         return (global->size << 1);
4422 }
4423 
4424 static int should_alloc_chunk(struct btrfs_fs_info *fs_info,
4425                               struct btrfs_space_info *sinfo, int force)
4426 {
4427         struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4428         u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
4429         u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
4430         u64 thresh;
4431 
4432         if (force == CHUNK_ALLOC_FORCE)
4433                 return 1;
4434 
4435         /*
4436          * We need to take into account the global rsv because for all intents
4437          * and purposes it's used space.  Don't worry about locking the
4438          * global_rsv, it doesn't change except when the transaction commits.
4439          */
4440         if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
4441                 num_allocated += calc_global_rsv_need_space(global_rsv);
4442 
4443         /*
4444          * in limited mode, we want to have some free space up to
4445          * about 1% of the FS size.
4446          */
4447         if (force == CHUNK_ALLOC_LIMITED) {
4448                 thresh = btrfs_super_total_bytes(fs_info->super_copy);
4449                 thresh = max_t(u64, SZ_64M, div_factor_fine(thresh, 1));
4450 
4451                 if (num_bytes - num_allocated < thresh)
4452                         return 1;
4453         }
4454 
4455         if (num_allocated + SZ_2M < div_factor(num_bytes, 8))
4456                 return 0;
4457         return 1;
4458 }
4459 
4460 static u64 get_profile_num_devs(struct btrfs_fs_info *fs_info, u64 type)
4461 {
4462         u64 num_dev;
4463 
4464         if (type & (BTRFS_BLOCK_GROUP_RAID10 |
4465                     BTRFS_BLOCK_GROUP_RAID0 |
4466                     BTRFS_BLOCK_GROUP_RAID5 |
4467                     BTRFS_BLOCK_GROUP_RAID6))
4468                 num_dev = fs_info->fs_devices->rw_devices;
4469         else if (type & BTRFS_BLOCK_GROUP_RAID1)
4470                 num_dev = 2;
4471         else
4472                 num_dev = 1;    /* DUP or single */
4473 
4474         return num_dev;
4475 }
4476 
4477 /*
4478  * If @is_allocation is true, reserve space in the system space info necessary
4479  * for allocating a chunk, otherwise if it's false, reserve space necessary for
4480  * removing a chunk.
4481  */
4482 void check_system_chunk(struct btrfs_trans_handle *trans,
4483                         struct btrfs_fs_info *fs_info, u64 type)
4484 {
4485         struct btrfs_space_info *info;
4486         u64 left;
4487         u64 thresh;
4488         int ret = 0;
4489         u64 num_devs;
4490 
4491         /*
4492          * Needed because we can end up allocating a system chunk and for an
4493          * atomic and race free space reservation in the chunk block reserve.
4494          */
4495         ASSERT(mutex_is_locked(&fs_info->chunk_mutex));
4496 
4497         info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4498         spin_lock(&info->lock);
4499         left = info->total_bytes - btrfs_space_info_used(info, true);
4500         spin_unlock(&info->lock);
4501 
4502         num_devs = get_profile_num_devs(fs_info, type);
4503 
4504         /* num_devs device items to update and 1 chunk item to add or remove */
4505         thresh = btrfs_calc_trunc_metadata_size(fs_info, num_devs) +
4506                 btrfs_calc_trans_metadata_size(fs_info, 1);
4507 
4508         if (left < thresh && btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
4509                 btrfs_info(fs_info, "left=%llu, need=%llu, flags=%llu",
4510                            left, thresh, type);
4511                 dump_space_info(fs_info, info, 0, 0);
4512         }
4513 
4514         if (left < thresh) {
4515                 u64 flags = btrfs_system_alloc_profile(fs_info);
4516 
4517                 /*
4518                  * Ignore failure to create system chunk. We might end up not
4519                  * needing it, as we might not need to COW all nodes/leafs from
4520                  * the paths we visit in the chunk tree (they were already COWed
4521                  * or created in the current transaction for example).
4522                  */
4523                 ret = btrfs_alloc_chunk(trans, fs_info, flags);
4524         }
4525 
4526         if (!ret) {
4527                 ret = btrfs_block_rsv_add(fs_info->chunk_root,
4528                                           &fs_info->chunk_block_rsv,
4529                                           thresh, BTRFS_RESERVE_NO_FLUSH);
4530                 if (!ret)
4531                         trans->chunk_bytes_reserved += thresh;
4532         }
4533 }
4534 
4535 /*
4536  * If force is CHUNK_ALLOC_FORCE:
4537  *    - return 1 if it successfully allocates a chunk,
4538  *    - return errors including -ENOSPC otherwise.
4539  * If force is NOT CHUNK_ALLOC_FORCE:
4540  *    - return 0 if it doesn't need to allocate a new chunk,
4541  *    - return 1 if it successfully allocates a chunk,
4542  *    - return errors including -ENOSPC otherwise.
4543  */
4544 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
4545                           struct btrfs_fs_info *fs_info, u64 flags, int force)
4546 {
4547         struct btrfs_space_info *space_info;
4548         int wait_for_alloc = 0;
4549         int ret = 0;
4550 
4551         /* Don't re-enter if we're already allocating a chunk */
4552         if (trans->allocating_chunk)
4553                 return -ENOSPC;
4554 
4555         space_info = __find_space_info(fs_info, flags);
4556         if (!space_info) {
4557                 ret = create_space_info(fs_info, flags, &space_info);
4558                 if (ret)
4559                         return ret;
4560         }
4561 
4562 again:
4563         spin_lock(&space_info->lock);
4564         if (force < space_info->force_alloc)
4565                 force = space_info->force_alloc;
4566         if (space_info->full) {
4567                 if (should_alloc_chunk(fs_info, space_info, force))
4568                         ret = -ENOSPC;
4569                 else
4570                         ret = 0;
4571                 spin_unlock(&space_info->lock);
4572                 return ret;
4573         }
4574 
4575         if (!should_alloc_chunk(fs_info, space_info, force)) {
4576                 spin_unlock(&space_info->lock);
4577                 return 0;
4578         } else if (space_info->chunk_alloc) {
4579                 wait_for_alloc = 1;
4580         } else {
4581                 space_info->chunk_alloc = 1;
4582         }
4583 
4584         spin_unlock(&space_info->lock);
4585 
4586         mutex_lock(&fs_info->chunk_mutex);
4587 
4588         /*
4589          * The chunk_mutex is held throughout the entirety of a chunk
4590          * allocation, so once we've acquired the chunk_mutex we know that the
4591          * other guy is done and we need to recheck and see if we should
4592          * allocate.
4593          */
4594         if (wait_for_alloc) {
4595                 mutex_unlock(&fs_info->chunk_mutex);
4596                 wait_for_alloc = 0;
4597                 goto again;
4598         }
4599 
4600         trans->allocating_chunk = true;
4601 
4602         /*
4603          * If we have mixed data/metadata chunks we want to make sure we keep
4604          * allocating mixed chunks instead of individual chunks.
4605          */
4606         if (btrfs_mixed_space_info(space_info))
4607                 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
4608 
4609         /*
4610          * if we're doing a data chunk, go ahead and make sure that
4611          * we keep a reasonable number of metadata chunks allocated in the
4612          * FS as well.
4613          */
4614         if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
4615                 fs_info->data_chunk_allocations++;
4616                 if (!(fs_info->data_chunk_allocations %
4617                       fs_info->metadata_ratio))
4618                         force_metadata_allocation(fs_info);
4619         }
4620 
4621         /*
4622          * Check if we have enough space in SYSTEM chunk because we may need
4623          * to update devices.
4624          */
4625         check_system_chunk(trans, fs_info, flags);
4626 
4627         ret = btrfs_alloc_chunk(trans, fs_info, flags);
4628         trans->allocating_chunk = false;
4629 
4630         spin_lock(&space_info->lock);
4631         if (ret < 0 && ret != -ENOSPC)
4632                 goto out;
4633         if (ret)
4634                 space_info->full = 1;
4635         else
4636                 ret = 1;
4637 
4638         space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4639 out:
4640         space_info->chunk_alloc = 0;
4641         spin_unlock(&space_info->lock);
4642         mutex_unlock(&fs_info->chunk_mutex);
4643         /*
4644          * When we allocate a new chunk we reserve space in the chunk block
4645          * reserve to make sure we can COW nodes/leafs in the chunk tree or
4646          * add new nodes/leafs to it if we end up needing to do it when
4647          * inserting the chunk item and updating device items as part of the
4648          * second phase of chunk allocation, performed by
4649          * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a
4650          * large number of new block groups to create in our transaction
4651          * handle's new_bgs list to avoid exhausting the chunk block reserve
4652          * in extreme cases - like having a single transaction create many new
4653          * block groups when starting to write out the free space caches of all
4654          * the block groups that were made dirty during the lifetime of the
4655          * transaction.
4656          */
4657         if (trans->can_flush_pending_bgs &&
4658             trans->chunk_bytes_reserved >= (u64)SZ_2M) {
4659                 btrfs_create_pending_block_groups(trans, fs_info);
4660                 btrfs_trans_release_chunk_metadata(trans);
4661         }
4662         return ret;
4663 }
4664 
4665 static int can_overcommit(struct btrfs_fs_info *fs_info,
4666                           struct btrfs_space_info *space_info, u64 bytes,
4667                           enum btrfs_reserve_flush_enum flush,
4668                           bool system_chunk)
4669 {
4670         struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4671         u64 profile;
4672         u64 space_size;
4673         u64 avail;
4674         u64 used;
4675 
4676         /* Don't overcommit when in mixed mode. */
4677         if (space_info->flags & BTRFS_BLOCK_GROUP_DATA)
4678                 return 0;
4679 
4680         if (system_chunk)
4681                 profile = btrfs_system_alloc_profile(fs_info);
4682         else
4683                 profile = btrfs_metadata_alloc_profile(fs_info);
4684 
4685         used = btrfs_space_info_used(space_info, false);
4686 
4687         /*
4688          * We only want to allow over committing if we have lots of actual space
4689          * free, but if we don't have enough space to handle the global reserve
4690          * space then we could end up having a real enospc problem when trying
4691          * to allocate a chunk or some other such important allocation.
4692          */
4693         spin_lock(&global_rsv->lock);
4694         space_size = calc_global_rsv_need_space(global_rsv);
4695         spin_unlock(&global_rsv->lock);
4696         if (used + space_size >= space_info->total_bytes)
4697                 return 0;
4698 
4699         used += space_info->bytes_may_use;
4700 
4701         avail = atomic64_read(&fs_info->free_chunk_space);
4702 
4703         /*
4704          * If we have dup, raid1 or raid10 then only half of the free
4705          * space is actually useable.  For raid56, the space info used
4706          * doesn't include the parity drive, so we don't have to
4707          * change the math
4708          */
4709         if (profile & (BTRFS_BLOCK_GROUP_DUP |
4710                        BTRFS_BLOCK_GROUP_RAID1 |
4711                        BTRFS_BLOCK_GROUP_RAID10))
4712                 avail >>= 1;
4713 
4714         /*
4715          * If we aren't flushing all things, let us overcommit up to
4716          * 1/2th of the space. If we can flush, don't let us overcommit
4717          * too much, let it overcommit up to 1/8 of the space.
4718          */
4719         if (flush == BTRFS_RESERVE_FLUSH_ALL)
4720                 avail >>= 3;
4721         else
4722                 avail >>= 1;
4723 
4724         if (used + bytes < space_info->total_bytes + avail)
4725                 return 1;
4726         return 0;
4727 }
4728 
4729 static void btrfs_writeback_inodes_sb_nr(struct btrfs_fs_info *fs_info,
4730                                          unsigned long nr_pages, int nr_items)
4731 {
4732         struct super_block *sb = fs_info->sb;
4733 
4734         if (down_read_trylock(&sb->s_umount)) {
4735                 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4736                 up_read(&sb->s_umount);
4737         } else {
4738                 /*
4739                  * We needn't worry the filesystem going from r/w to r/o though
4740                  * we don't acquire ->s_umount mutex, because the filesystem
4741                  * should guarantee the delalloc inodes list be empty after
4742                  * the filesystem is readonly(all dirty pages are written to
4743                  * the disk).
4744                  */
4745                 btrfs_start_delalloc_roots(fs_info, 0, nr_items);
4746                 if (!current->journal_info)
4747                         btrfs_wait_ordered_roots(fs_info, nr_items, 0, (u64)-1);
4748         }
4749 }
4750 
4751 static inline u64 calc_reclaim_items_nr(struct btrfs_fs_info *fs_info,
4752                                         u64 to_reclaim)
4753 {
4754         u64 bytes;
4755         u64 nr;
4756 
4757         bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
4758         nr = div64_u64(to_reclaim, bytes);
4759         if (!nr)
4760                 nr = 1;
4761         return nr;
4762 }
4763 
4764 #define EXTENT_SIZE_PER_ITEM    SZ_256K
4765 
4766 /*
4767  * shrink metadata reservation for delalloc
4768  */
4769 static void shrink_delalloc(struct btrfs_fs_info *fs_info, u64 to_reclaim,
4770                             u64 orig, bool wait_ordered)
4771 {
4772         struct btrfs_block_rsv *block_rsv;
4773         struct btrfs_space_info *space_info;
4774         struct btrfs_trans_handle *trans;
4775         u64 delalloc_bytes;
4776         u64 max_reclaim;
4777         u64 items;
4778         long time_left;
4779         unsigned long nr_pages;
4780         int loops;
4781         enum btrfs_reserve_flush_enum flush;
4782 
4783         /* Calc the number of the pages we need flush for space reservation */
4784         items = calc_reclaim_items_nr(fs_info, to_reclaim);
4785         to_reclaim = items * EXTENT_SIZE_PER_ITEM;
4786 
4787         trans = (struct btrfs_trans_handle *)current->journal_info;
4788         block_rsv = &fs_info->delalloc_block_rsv;
4789         space_info = block_rsv->space_info;
4790 
4791         delalloc_bytes = percpu_counter_sum_positive(
4792                                                 &fs_info->delalloc_bytes);
4793         if (delalloc_bytes == 0) {
4794                 if (trans)
4795                         return;
4796                 if (wait_ordered)
4797                         btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1);
4798                 return;
4799         }
4800 
4801         loops = 0;
4802         while (delalloc_bytes && loops < 3) {
4803                 max_reclaim = min(delalloc_bytes, to_reclaim);
4804                 nr_pages = max_reclaim >> PAGE_SHIFT;
4805                 btrfs_writeback_inodes_sb_nr(fs_info, nr_pages, items);
4806                 /*
4807                  * We need to wait for the async pages to actually start before
4808                  * we do anything.
4809                  */
4810                 max_reclaim = atomic_read(&fs_info->async_delalloc_pages);
4811                 if (!max_reclaim)
4812                         goto skip_async;
4813 
4814                 if (max_reclaim <= nr_pages)
4815                         max_reclaim = 0;
4816                 else
4817                         max_reclaim -= nr_pages;
4818 
4819                 wait_event(fs_info->async_submit_wait,
4820                            atomic_read(&fs_info->async_delalloc_pages) <=
4821                            (int)max_reclaim);
4822 skip_async:
4823                 if (!trans)
4824                         flush = BTRFS_RESERVE_FLUSH_ALL;
4825                 else
4826                         flush = BTRFS_RESERVE_NO_FLUSH;
4827                 spin_lock(&space_info->lock);
4828                 if (list_empty(&space_info->tickets) &&
4829                     list_empty(&space_info->priority_tickets)) {
4830                         spin_unlock(&space_info->lock);
4831                         break;
4832                 }
4833                 spin_unlock(&space_info->lock);
4834 
4835                 loops++;
4836                 if (wait_ordered && !trans) {
4837                         btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1);
4838                 } else {
4839                         time_left = schedule_timeout_killable(1);
4840                         if (time_left)
4841                                 break;
4842                 }
4843                 delalloc_bytes = percpu_counter_sum_positive(
4844                                                 &fs_info->delalloc_bytes);
4845         }
4846 }
4847 
4848 /**
4849  * maybe_commit_transaction - possibly commit the transaction if its ok to
4850  * @root - the root we're allocating for
4851  * @bytes - the number of bytes we want to reserve
4852  * @force - force the commit
4853  *
4854  * This will check to make sure that committing the transaction will actually
4855  * get us somewhere and then commit the transaction if it does.  Otherwise it
4856  * will return -ENOSPC.
4857  */
4858 static int may_commit_transaction(struct btrfs_fs_info *fs_info,
4859                                   struct btrfs_space_info *space_info,
4860                                   u64 bytes, int force)
4861 {
4862         struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_block_rsv;
4863         struct btrfs_trans_handle *trans;
4864 
4865         trans = (struct btrfs_trans_handle *)current->journal_info;
4866         if (trans)
4867                 return -EAGAIN;
4868 
4869         if (force)
4870                 goto commit;
4871 
4872         /* See if there is enough pinned space to make this reservation */
4873         if (percpu_counter_compare(&space_info->total_bytes_pinned,
4874                                    bytes) >= 0)
4875                 goto commit;
4876 
4877         /*
4878          * See if there is some space in the delayed insertion reservation for
4879          * this reservation.
4880          */
4881         if (space_info != delayed_rsv->space_info)
4882                 return -ENOSPC;
4883 
4884         spin_lock(&delayed_rsv->lock);
4885         if (percpu_counter_compare(&space_info->total_bytes_pinned,
4886                                    bytes - delayed_rsv->size) < 0) {
4887                 spin_unlock(&delayed_rsv->lock);
4888                 return -ENOSPC;
4889         }
4890         spin_unlock(&delayed_rsv->lock);
4891 
4892 commit:
4893         trans = btrfs_join_transaction(fs_info->extent_root);
4894         if (IS_ERR(trans))
4895                 return -ENOSPC;
4896 
4897         return btrfs_commit_transaction(trans);
4898 }
4899 
4900 struct reserve_ticket {
4901         u64 bytes;
4902         int error;
4903         struct list_head list;
4904         wait_queue_head_t wait;
4905 };
4906 
4907 static int flush_space(struct btrfs_fs_info *fs_info,
4908                        struct btrfs_space_info *space_info, u64 num_bytes,
4909                        u64 orig_bytes, int state)
4910 {
4911         struct btrfs_root *root = fs_info->extent_root;
4912         struct btrfs_trans_handle *trans;
4913         int nr;
4914         int ret = 0;
4915 
4916         switch (state) {
4917         case FLUSH_DELAYED_ITEMS_NR:
4918         case FLUSH_DELAYED_ITEMS:
4919                 if (state == FLUSH_DELAYED_ITEMS_NR)
4920                         nr = calc_reclaim_items_nr(fs_info, num_bytes) * 2;
4921                 else
4922                         nr = -1;
4923 
4924                 trans = btrfs_join_transaction(root);
4925                 if (IS_ERR(trans)) {
4926                         ret = PTR_ERR(trans);
4927                         break;
4928                 }
4929                 ret = btrfs_run_delayed_items_nr(trans, fs_info, nr);
4930                 btrfs_end_transaction(trans);
4931                 break;
4932         case FLUSH_DELALLOC:
4933         case FLUSH_DELALLOC_WAIT:
4934                 shrink_delalloc(fs_info, num_bytes * 2, orig_bytes,
4935                                 state == FLUSH_DELALLOC_WAIT);
4936                 break;
4937         case ALLOC_CHUNK:
4938                 trans = btrfs_join_transaction(root);
4939                 if (IS_ERR(trans)) {
4940                         ret = PTR_ERR(trans);
4941                         break;
4942                 }
4943                 ret = do_chunk_alloc(trans, fs_info,
4944                                      btrfs_metadata_alloc_profile(fs_info),
4945                                      CHUNK_ALLOC_NO_FORCE);
4946                 btrfs_end_transaction(trans);
4947                 if (ret > 0 || ret == -ENOSPC)
4948                         ret = 0;
4949                 break;
4950         case COMMIT_TRANS:
4951                 ret = may_commit_transaction(fs_info, space_info,
4952                                              orig_bytes, 0);
4953                 break;
4954         default:
4955                 ret = -ENOSPC;
4956                 break;
4957         }
4958 
4959         trace_btrfs_flush_space(fs_info, space_info->flags, num_bytes,
4960                                 orig_bytes, state, ret);
4961         return ret;
4962 }
4963 
4964 static inline u64
4965 btrfs_calc_reclaim_metadata_size(struct btrfs_fs_info *fs_info,
4966                                  struct btrfs_space_info *space_info,
4967                                  bool system_chunk)
4968 {
4969         struct reserve_ticket *ticket;
4970         u64 used;
4971         u64 expected;
4972         u64 to_reclaim = 0;
4973 
4974         list_for_each_entry(ticket, &space_info->tickets, list)
4975                 to_reclaim += ticket->bytes;
4976         list_for_each_entry(ticket, &space_info->priority_tickets, list)
4977                 to_reclaim += ticket->bytes;
4978         if (to_reclaim)
4979                 return to_reclaim;
4980 
4981         to_reclaim = min_t(u64, num_online_cpus() * SZ_1M, SZ_16M);
4982         if (can_overcommit(fs_info, space_info, to_reclaim,
4983                            BTRFS_RESERVE_FLUSH_ALL, system_chunk))
4984                 return 0;
4985 
4986         used = btrfs_space_info_used(space_info, true);
4987 
4988         if (can_overcommit(fs_info, space_info, SZ_1M,
4989                            BTRFS_RESERVE_FLUSH_ALL, system_chunk))
4990                 expected = div_factor_fine(space_info->total_bytes, 95);
4991         else
4992                 expected = div_factor_fine(space_info->total_bytes, 90);
4993 
4994         if (used > expected)
4995                 to_reclaim = used - expected;
4996         else
4997                 to_reclaim = 0;
4998         to_reclaim = min(to_reclaim, space_info->bytes_may_use +
4999                                      space_info->bytes_reserved);
5000         return to_reclaim;
5001 }
5002 
5003 static inline int need_do_async_reclaim(struct btrfs_fs_info *fs_info,
5004                                         struct btrfs_space_info *space_info,
5005                                         u64 used, bool system_chunk)
5006 {
5007         u64 thresh = div_factor_fine(space_info->total_bytes, 98);
5008 
5009         /* If we're just plain full then async reclaim just slows us down. */
5010         if ((space_info->bytes_used + space_info->bytes_reserved) >= thresh)
5011                 return 0;
5012 
5013         if (!btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5014                                               system_chunk))
5015                 return 0;
5016 
5017         return (used >= thresh && !btrfs_fs_closing(fs_info) &&
5018                 !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state));
5019 }
5020 
5021 static void wake_all_tickets(struct list_head *head)
5022 {
5023         struct reserve_ticket *ticket;
5024 
5025         while (!list_empty(head)) {
5026                 ticket = list_first_entry(head, struct reserve_ticket, list);
5027                 list_del_init(&ticket->list);
5028                 ticket->error = -ENOSPC;
5029                 wake_up(&ticket->wait);
5030         }
5031 }
5032 
5033 /*
5034  * This is for normal flushers, we can wait all goddamned day if we want to.  We
5035  * will loop and continuously try to flush as long as we are making progress.
5036  * We count progress as clearing off tickets each time we have to loop.
5037  */
5038 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
5039 {
5040         struct btrfs_fs_info *fs_info;
5041         struct btrfs_space_info *space_info;
5042         u64 to_reclaim;
5043         int flush_state;
5044         int commit_cycles = 0;
5045         u64 last_tickets_id;
5046 
5047         fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
5048         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5049 
5050         spin_lock(&space_info->lock);
5051         to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5052                                                       false);
5053         if (!to_reclaim) {
5054                 space_info->flush = 0;
5055                 spin_unlock(&space_info->lock);
5056                 return;
5057         }
5058         last_tickets_id = space_info->tickets_id;
5059         spin_unlock(&space_info->lock);
5060 
5061         flush_state = FLUSH_DELAYED_ITEMS_NR;
5062         do {
5063                 struct reserve_ticket *ticket;
5064                 int ret;
5065 
5066                 ret = flush_space(fs_info, space_info, to_reclaim, to_reclaim,
5067                                   flush_state);
5068                 spin_lock(&space_info->lock);
5069                 if (list_empty(&space_info->tickets)) {
5070                         space_info->flush = 0;
5071                         spin_unlock(&space_info->lock);
5072                         return;
5073                 }
5074                 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info,
5075                                                               space_info,
5076                                                               false);
5077                 ticket = list_first_entry(&space_info->tickets,
5078                                           struct reserve_ticket, list);
5079                 if (last_tickets_id == space_info->tickets_id) {
5080                         flush_state++;
5081                 } else {
5082                         last_tickets_id = space_info->tickets_id;
5083                         flush_state = FLUSH_DELAYED_ITEMS_NR;
5084                         if (commit_cycles)
5085                                 commit_cycles--;
5086                 }
5087 
5088                 if (flush_state > COMMIT_TRANS) {
5089                         commit_cycles++;
5090                         if (commit_cycles > 2) {
5091                                 wake_all_tickets(&space_info->tickets);
5092                                 space_info->flush = 0;
5093                         } else {
5094                                 flush_state = FLUSH_DELAYED_ITEMS_NR;
5095                         }
5096                 }
5097                 spin_unlock(&space_info->lock);
5098         } while (flush_state <= COMMIT_TRANS);
5099 }
5100 
5101 void btrfs_init_async_reclaim_work(struct work_struct *work)
5102 {
5103         INIT_WORK(work, btrfs_async_reclaim_metadata_space);
5104 }
5105 
5106 static void priority_reclaim_metadata_space(struct btrfs_fs_info *fs_info,
5107                                             struct btrfs_space_info *space_info,
5108                                             struct reserve_ticket *ticket)
5109 {
5110         u64 to_reclaim;
5111         int flush_state = FLUSH_DELAYED_ITEMS_NR;
5112 
5113         spin_lock(&space_info->lock);
5114         to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5115                                                       false);
5116         if (!to_reclaim) {
5117                 spin_unlock(&space_info->lock);
5118                 return;
5119         }
5120         spin_unlock(&space_info->lock);
5121 
5122         do {
5123                 flush_space(fs_info, space_info, to_reclaim, to_reclaim,
5124                             flush_state);
5125                 flush_state++;
5126                 spin_lock(&space_info->lock);
5127                 if (ticket->bytes == 0) {
5128                         spin_unlock(&space_info->lock);
5129                         return;
5130                 }
5131                 spin_unlock(&space_info->lock);
5132 
5133                 /*
5134                  * Priority flushers can't wait on delalloc without
5135                  * deadlocking.
5136                  */
5137                 if (flush_state == FLUSH_DELALLOC ||
5138                     flush_state == FLUSH_DELALLOC_WAIT)
5139                         flush_state = ALLOC_CHUNK;
5140         } while (flush_state < COMMIT_TRANS);
5141 }
5142 
5143 static int wait_reserve_ticket(struct btrfs_fs_info *fs_info,
5144                                struct btrfs_space_info *space_info,
5145                                struct reserve_ticket *ticket, u64 orig_bytes)
5146 
5147 {
5148         DEFINE_WAIT(wait);
5149         int ret = 0;
5150 
5151         spin_lock(&space_info->lock);
5152         while (ticket->bytes > 0 && ticket->error == 0) {
5153                 ret = prepare_to_wait_event(&ticket->wait, &wait, TASK_KILLABLE);
5154                 if (ret) {
5155                         ret = -EINTR;
5156                         break;
5157                 }
5158                 spin_unlock(&space_info->lock);
5159 
5160                 schedule();
5161 
5162                 finish_wait(&ticket->wait, &wait);
5163                 spin_lock(&space_info->lock);
5164         }
5165         if (!ret)
5166                 ret = ticket->error;
5167         if (!list_empty(&ticket->list))
5168                 list_del_init(&ticket->list);
5169         if (ticket->bytes && ticket->bytes < orig_bytes) {
5170                 u64 num_bytes = orig_bytes - ticket->bytes;
5171                 space_info->bytes_may_use -= num_bytes;
5172                 trace_btrfs_space_reservation(fs_info, "space_info",
5173                                               space_info->flags, num_bytes, 0);
5174         }
5175         spin_unlock(&space_info->lock);
5176 
5177         return ret;
5178 }
5179 
5180 /**
5181  * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5182  * @root - the root we're allocating for
5183  * @space_info - the space info we want to allocate from
5184  * @orig_bytes - the number of bytes we want
5185  * @flush - whether or not we can flush to make our reservation
5186  *
5187  * This will reserve orig_bytes number of bytes from the space info associated
5188  * with the block_rsv.  If there is not enough space it will make an attempt to
5189  * flush out space to make room.  It will do this by flushing delalloc if
5190  * possible or committing the transaction.  If flush is 0 then no attempts to
5191  * regain reservations will be made and this will fail if there is not enough
5192  * space already.
5193  */
5194 static int __reserve_metadata_bytes(struct btrfs_fs_info *fs_info,
5195                                     struct btrfs_space_info *space_info,
5196                                     u64 orig_bytes,
5197                                     enum btrfs_reserve_flush_enum flush,
5198                                     bool system_chunk)
5199 {
5200         struct reserve_ticket ticket;
5201         u64 used;
5202         int ret = 0;
5203 
5204         ASSERT(orig_bytes);
5205         ASSERT(!current->journal_info || flush != BTRFS_RESERVE_FLUSH_ALL);
5206 
5207         spin_lock(&space_info->lock);
5208         ret = -ENOSPC;
5209         used = btrfs_space_info_used(space_info, true);
5210 
5211         /*
5212          * If we have enough space then hooray, make our reservation and carry
5213          * on.  If not see if we can overcommit, and if we can, hooray carry on.
5214          * If not things get more complicated.
5215          */
5216         if (used + orig_bytes <= space_info->total_bytes) {
5217                 space_info->bytes_may_use += orig_bytes;
5218                 trace_btrfs_space_reservation(fs_info, "space_info",
5219                                               space_info->flags, orig_bytes, 1);
5220                 ret = 0;
5221         } else if (can_overcommit(fs_info, space_info, orig_bytes, flush,
5222                                   system_chunk)) {
5223                 space_info->bytes_may_use += orig_bytes;
5224                 trace_btrfs_space_reservation(fs_info, "space_info",
5225                                               space_info->flags, orig_bytes, 1);
5226                 ret = 0;
5227         }
5228 
5229         /*
5230          * If we couldn't make a reservation then setup our reservation ticket
5231          * and kick the async worker if it's not already running.
5232          *
5233          * If we are a priority flusher then we just need to add our ticket to
5234          * the list and we will do our own flushing further down.
5235          */
5236         if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
5237                 ticket.bytes = orig_bytes;
5238                 ticket.error = 0;
5239                 init_waitqueue_head(&ticket.wait);
5240                 if (flush == BTRFS_RESERVE_FLUSH_ALL) {
5241                         list_add_tail(&ticket.list, &space_info->tickets);
5242                         if (!space_info->flush) {
5243                                 space_info->flush = 1;
5244                                 trace_btrfs_trigger_flush(fs_info,
5245                                                           space_info->flags,
5246                                                           orig_bytes, flush,
5247                                                           "enospc");
5248                                 queue_work(system_unbound_wq,
5249                                            &fs_info->async_reclaim_work);
5250                         }
5251                 } else {
5252                         list_add_tail(&ticket.list,
5253                                       &space_info->priority_tickets);
5254                 }
5255         } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
5256                 used += orig_bytes;
5257                 /*
5258                  * We will do the space reservation dance during log replay,
5259                  * which means we won't have fs_info->fs_root set, so don't do
5260                  * the async reclaim as we will panic.
5261                  */
5262                 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags) &&
5263                     need_do_async_reclaim(fs_info, space_info,
5264                                           used, system_chunk) &&
5265                     !work_busy(&fs_info->async_reclaim_work)) {
5266                         trace_btrfs_trigger_flush(fs_info, space_info->flags,
5267                                                   orig_bytes, flush, "preempt");
5268                         queue_work(system_unbound_wq,
5269                                    &fs_info->async_reclaim_work);
5270                 }
5271         }
5272         spin_unlock(&space_info->lock);
5273         if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
5274                 return ret;
5275 
5276         if (flush == BTRFS_RESERVE_FLUSH_ALL)
5277                 return wait_reserve_ticket(fs_info, space_info, &ticket,
5278                                            orig_bytes);
5279 
5280         ret = 0;
5281         priority_reclaim_metadata_space(fs_info, space_info, &ticket);
5282         spin_lock(&space_info->lock);
5283         if (ticket.bytes) {
5284                 if (ticket.bytes < orig_bytes) {
5285                         u64 num_bytes = orig_bytes - ticket.bytes;
5286                         space_info->bytes_may_use -= num_bytes;
5287                         trace_btrfs_space_reservation(fs_info, "space_info",
5288                                                       space_info->flags,
5289                                                       num_bytes, 0);
5290 
5291                 }
5292                 list_del_init(&ticket.list);
5293                 ret = -ENOSPC;
5294         }
5295         spin_unlock(&space_info->lock);
5296         ASSERT(list_empty(&ticket.list));
5297         return ret;
5298 }
5299 
5300 /**
5301  * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5302  * @root - the root we're allocating for
5303  * @block_rsv - the block_rsv we're allocating for
5304  * @orig_bytes - the number of bytes we want
5305  * @flush - whether or not we can flush to make our reservation
5306  *
5307  * This will reserve orgi_bytes number of bytes from the space info associated
5308  * with the block_rsv.  If there is not enough space it will make an attempt to
5309  * flush out space to make room.  It will do this by flushing delalloc if
5310  * possible or committing the transaction.  If flush is 0 then no attempts to
5311  * regain reservations will be made and this will fail if there is not enough
5312  * space already.
5313  */
5314 static int reserve_metadata_bytes(struct btrfs_root *root,
5315                                   struct btrfs_block_rsv *block_rsv,
5316                                   u64 orig_bytes,
5317                                   enum btrfs_reserve_flush_enum flush)
5318 {
5319         struct btrfs_fs_info *fs_info = root->fs_info;
5320         struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5321         int ret;
5322         bool system_chunk = (root == fs_info->chunk_root);
5323 
5324         ret = __reserve_metadata_bytes(fs_info, block_rsv->space_info,
5325                                        orig_bytes, flush, system_chunk);
5326         if (ret == -ENOSPC &&
5327             unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
5328                 if (block_rsv != global_rsv &&
5329                     !block_rsv_use_bytes(global_rsv, orig_bytes))
5330                         ret = 0;
5331         }
5332         if (ret == -ENOSPC)
5333                 trace_btrfs_space_reservation(fs_info, "space_info:enospc",
5334                                               block_rsv->space_info->flags,
5335                                               orig_bytes, 1);
5336         return ret;
5337 }
5338 
5339 static struct btrfs_block_rsv *get_block_rsv(
5340                                         const struct btrfs_trans_handle *trans,
5341                                         const struct btrfs_root *root)
5342 {
5343         struct btrfs_fs_info *fs_info = root->fs_info;
5344         struct btrfs_block_rsv *block_rsv = NULL;
5345 
5346         if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
5347             (root == fs_info->csum_root && trans->adding_csums) ||
5348             (root == fs_info->uuid_root))
5349                 block_rsv = trans->block_rsv;
5350 
5351         if (!block_rsv)
5352                 block_rsv = root->block_rsv;
5353 
5354         if (!block_rsv)
5355                 block_rsv = &fs_info->empty_block_rsv;
5356 
5357         return block_rsv;
5358 }
5359 
5360 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
5361                                u64 num_bytes)
5362 {
5363         int ret = -ENOSPC;
5364         spin_lock(&block_rsv->lock);
5365         if (block_rsv->reserved >= num_bytes) {
5366                 block_rsv->reserved -= num_bytes;
5367                 if (block_rsv->reserved < block_rsv->size)
5368                         block_rsv->full = 0;
5369                 ret = 0;
5370         }
5371         spin_unlock(&block_rsv->lock);
5372         return ret;
5373 }
5374 
5375 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
5376                                 u64 num_bytes, int update_size)
5377 {
5378         spin_lock(&block_rsv->lock);
5379         block_rsv->reserved += num_bytes;
5380         if (update_size)
5381                 block_rsv->size += num_bytes;
5382         else if (block_rsv->reserved >= block_rsv->size)
5383                 block_rsv->full = 1;
5384         spin_unlock(&block_rsv->lock);
5385 }
5386 
5387 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
5388                              struct btrfs_block_rsv *dest, u64 num_bytes,
5389                              int min_factor)
5390 {
5391         struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5392         u64 min_bytes;
5393 
5394         if (global_rsv->space_info != dest->space_info)
5395                 return -ENOSPC;
5396 
5397         spin_lock(&global_rsv->lock);
5398         min_bytes = div_factor(global_rsv->size, min_factor);
5399         if (global_rsv->reserved < min_bytes + num_bytes) {
5400                 spin_unlock(&global_rsv->lock);
5401                 return -ENOSPC;
5402         }
5403         global_rsv->reserved -= num_bytes;
5404         if (global_rsv->reserved < global_rsv->size)
5405                 global_rsv->full = 0;
5406         spin_unlock(&global_rsv->lock);
5407 
5408         block_rsv_add_bytes(dest, num_bytes, 1);
5409         return 0;
5410 }
5411 
5412 /*
5413  * This is for space we already have accounted in space_info->bytes_may_use, so
5414  * basically when we're returning space from block_rsv's.
5415  */
5416 static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info,
5417                                      struct btrfs_space_info *space_info,
5418                                      u64 num_bytes)
5419 {
5420         struct reserve_ticket *ticket;
5421         struct list_head *head;
5422         u64 used;
5423         enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_NO_FLUSH;
5424         bool check_overcommit = false;
5425 
5426         spin_lock(&space_info->lock);
5427         head = &space_info->priority_tickets;
5428 
5429         /*
5430          * If we are over our limit then we need to check and see if we can
5431          * overcommit, and if we can't then we just need to free up our space
5432          * and not satisfy any requests.
5433          */
5434         used = btrfs_space_info_used(space_info, true);
5435         if (used - num_bytes >= space_info->total_bytes)
5436                 check_overcommit = true;
5437 again:
5438         while (!list_empty(head) && num_bytes) {
5439                 ticket = list_first_entry(head, struct reserve_ticket,
5440                                           list);
5441                 /*
5442                  * We use 0 bytes because this space is already reserved, so
5443                  * adding the ticket space would be a double count.
5444                  */
5445                 if (check_overcommit &&
5446                     !can_overcommit(fs_info, space_info, 0, flush, false))
5447                         break;
5448                 if (num_bytes >= ticket->bytes) {
5449                         list_del_init(&ticket->list);
5450                         num_bytes -= ticket->bytes;
5451                         ticket->bytes = 0;
5452                         space_info->tickets_id++;
5453                         wake_up(&ticket->wait);
5454                 } else {
5455                         ticket->bytes -= num_bytes;
5456                         num_bytes = 0;
5457                 }
5458         }
5459 
5460         if (num_bytes && head == &space_info->priority_tickets) {
5461                 head = &space_info->tickets;
5462                 flush = BTRFS_RESERVE_FLUSH_ALL;
5463                 goto again;
5464         }
5465         space_info->bytes_may_use -= num_bytes;
5466         trace_btrfs_space_reservation(fs_info, "space_info",
5467                                       space_info->flags, num_bytes, 0);
5468         spin_unlock(&space_info->lock);
5469 }
5470 
5471 /*
5472  * This is for newly allocated space that isn't accounted in
5473  * space_info->bytes_may_use yet.  So if we allocate a chunk or unpin an extent
5474  * we use this helper.
5475  */
5476 static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info,
5477                                      struct btrfs_space_info *space_info,
5478                                      u64 num_bytes)
5479 {
5480         struct reserve_ticket *ticket;
5481         struct list_head *head = &space_info->priority_tickets;
5482 
5483 again:
5484         while (!list_empty(head) && num_bytes) {
5485                 ticket = list_first_entry(head, struct reserve_ticket,
5486                                           list);
5487                 if (num_bytes >= ticket->bytes) {
5488                         trace_btrfs_space_reservation(fs_info, "space_info",
5489                                                       space_info->flags,
5490                                                       ticket->bytes, 1);
5491                         list_del_init(&ticket->list);
5492                         num_bytes -= ticket->bytes;
5493                         space_info->bytes_may_use += ticket->bytes;
5494                         ticket->bytes = 0;
5495                         space_info->tickets_id++;
5496                         wake_up(&ticket->wait);
5497                 } else {
5498                         trace_btrfs_space_reservation(fs_info, "space_info",
5499                                                       space_info->flags,
5500                                                       num_bytes, 1);
5501                         space_info->bytes_may_use += num_bytes;
5502                         ticket->bytes -= num_bytes;
5503                         num_bytes = 0;
5504                 }
5505         }
5506 
5507         if (num_bytes && head == &space_info->priority_tickets) {
5508                 head = &space_info->tickets;
5509                 goto again;
5510         }
5511 }
5512 
5513 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
5514                                     struct btrfs_block_rsv *block_rsv,
5515                                     struct btrfs_block_rsv *dest, u64 num_bytes)
5516 {
5517         struct btrfs_space_info *space_info = block_rsv->space_info;
5518 
5519         spin_lock(&block_rsv->lock);
5520         if (num_bytes == (u64)-1)
5521                 num_bytes = block_rsv->size;
5522         block_rsv->size -= num_bytes;
5523         if (block_rsv->reserved >= block_rsv->size) {
5524                 num_bytes = block_rsv->reserved - block_rsv->size;
5525                 block_rsv->reserved = block_rsv->size;
5526                 block_rsv->full = 1;
5527         } else {
5528                 num_bytes = 0;
5529         }
5530         spin_unlock(&block_rsv->lock);
5531 
5532         if (num_bytes > 0) {
5533                 if (dest) {
5534                         spin_lock(&dest->lock);
5535                         if (!dest->full) {
5536                                 u64 bytes_to_add;
5537 
5538                                 bytes_to_add = dest->size - dest->reserved;
5539                                 bytes_to_add = min(num_bytes, bytes_to_add);
5540                                 dest->reserved += bytes_to_add;
5541                                 if (dest->reserved >= dest->size)
5542                                         dest->full = 1;
5543                                 num_bytes -= bytes_to_add;
5544                         }
5545                         spin_unlock(&dest->lock);
5546                 }
5547                 if (num_bytes)
5548                         space_info_add_old_bytes(fs_info, space_info,
5549                                                  num_bytes);
5550         }
5551 }
5552 
5553 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src,
5554                             struct btrfs_block_rsv *dst, u64 num_bytes,
5555                             int update_size)
5556 {
5557         int ret;
5558 
5559         ret = block_rsv_use_bytes(src, num_bytes);
5560         if (ret)
5561                 return ret;
5562 
5563         block_rsv_add_bytes(dst, num_bytes, update_size);
5564         return 0;
5565 }
5566 
5567 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
5568 {
5569         memset(rsv, 0, sizeof(*rsv));
5570         spin_lock_init(&rsv->lock);
5571         rsv->type = type;
5572 }
5573 
5574 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_fs_info *fs_info,
5575                                               unsigned short type)
5576 {
5577         struct btrfs_block_rsv *block_rsv;
5578 
5579         block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
5580         if (!block_rsv)
5581                 return NULL;
5582 
5583         btrfs_init_block_rsv(block_rsv, type);
5584         block_rsv->space_info = __find_space_info(fs_info,
5585                                                   BTRFS_BLOCK_GROUP_METADATA);
5586         return block_rsv;
5587 }
5588 
5589 void btrfs_free_block_rsv(struct btrfs_fs_info *fs_info,
5590                           struct btrfs_block_rsv *rsv)
5591 {
5592         if (!rsv)
5593                 return;
5594         btrfs_block_rsv_release(fs_info, rsv, (u64)-1);
5595         kfree(rsv);
5596 }
5597 
5598 void __btrfs_free_block_rsv(struct btrfs_block_rsv *rsv)
5599 {
5600         kfree(rsv);
5601 }
5602 
5603 int btrfs_block_rsv_add(struct btrfs_root *root,
5604                         struct btrfs_block_rsv *block_rsv, u64 num_bytes,
5605                         enum btrfs_reserve_flush_enum flush)
5606 {
5607         int ret;
5608 
5609         if (num_bytes == 0)
5610                 return 0;
5611 
5612         ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5613         if (!ret) {
5614                 block_rsv_add_bytes(block_rsv, num_bytes, 1);
5615                 return 0;
5616         }
5617 
5618         return ret;
5619 }
5620 
5621 int btrfs_block_rsv_check(struct btrfs_block_rsv *block_rsv, int min_factor)
5622 {
5623         u64 num_bytes = 0;
5624         int ret = -ENOSPC;
5625 
5626         if (!block_rsv)
5627                 return 0;
5628 
5629         spin_lock(&block_rsv->lock);
5630         num_bytes = div_factor(block_rsv->size, min_factor);
5631         if (block_rsv->reserved >= num_bytes)
5632                 ret = 0;
5633         spin_unlock(&block_rsv->lock);
5634 
5635         return ret;
5636 }
5637 
5638 int btrfs_block_rsv_refill(struct btrfs_root *root,
5639                            struct btrfs_block_rsv *block_rsv, u64 min_reserved,
5640                            enum btrfs_reserve_flush_enum flush)
5641 {
5642         u64 num_bytes = 0;
5643         int ret = -ENOSPC;
5644 
5645         if (!block_rsv)
5646                 return 0;
5647 
5648         spin_lock(&block_rsv->lock);
5649         num_bytes = min_reserved;
5650         if (block_rsv->reserved >= num_bytes)
5651                 ret = 0;
5652         else
5653                 num_bytes -= block_rsv->reserved;
5654         spin_unlock(&block_rsv->lock);
5655 
5656         if (!ret)
5657                 return 0;
5658 
5659         ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5660         if (!ret) {
5661                 block_rsv_add_bytes(block_rsv, num_bytes, 0);
5662                 return 0;
5663         }
5664 
5665         return ret;
5666 }
5667 
5668 void btrfs_block_rsv_release(struct btrfs_fs_info *fs_info,
5669                              struct btrfs_block_rsv *block_rsv,
5670                              u64 num_bytes)
5671 {
5672         struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5673 
5674         if (global_rsv == block_rsv ||
5675             block_rsv->space_info != global_rsv->space_info)
5676                 global_rsv = NULL;
5677         block_rsv_release_bytes(fs_info, block_rsv, global_rsv, num_bytes);
5678 }
5679 
5680 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
5681 {
5682         struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
5683         struct btrfs_space_info *sinfo = block_rsv->space_info;
5684         u64 num_bytes;
5685 
5686         /*
5687          * The global block rsv is based on the size of the extent tree, the
5688          * checksum tree and the root tree.  If the fs is empty we want to set
5689          * it to a minimal amount for safety.
5690          */
5691         num_bytes = btrfs_root_used(&fs_info->extent_root->root_item) +
5692                 btrfs_root_used(&fs_info->csum_root->root_item) +
5693                 btrfs_root_used(&fs_info->tree_root->root_item);
5694         num_bytes = max_t(u64, num_bytes, SZ_16M);
5695 
5696         spin_lock(&sinfo->lock);
5697         spin_lock(&block_rsv->lock);
5698 
5699         block_rsv->size = min_t(u64, num_bytes, SZ_512M);
5700 
5701         if (block_rsv->reserved < block_rsv->size) {
5702                 num_bytes = btrfs_space_info_used(sinfo, true);
5703                 if (sinfo->total_bytes > num_bytes) {
5704                         num_bytes = sinfo->total_bytes - num_bytes;
5705                         num_bytes = min(num_bytes,
5706                                         block_rsv->size - block_rsv->reserved);
5707                         block_rsv->reserved += num_bytes;
5708                         sinfo->bytes_may_use += num_bytes;
5709                         trace_btrfs_space_reservation(fs_info, "space_info",
5710                                                       sinfo->flags, num_bytes,
5711                                                       1);
5712                 }
5713         } else if (block_rsv->reserved > block_rsv->size) {
5714                 num_bytes = block_rsv->reserved - block_rsv->size;
5715                 sinfo->bytes_may_use -= num_bytes;
5716                 trace_btrfs_space_reservation(fs_info, "space_info",
5717                                       sinfo->flags, num_bytes, 0);
5718                 block_rsv->reserved = block_rsv->size;
5719         }
5720 
5721         if (block_rsv->reserved == block_rsv->size)
5722                 block_rsv->full = 1;
5723         else
5724                 block_rsv->full = 0;
5725 
5726         spin_unlock(&block_rsv->lock);
5727         spin_unlock(&sinfo->lock);
5728 }
5729 
5730 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
5731 {
5732         struct btrfs_space_info *space_info;
5733 
5734         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
5735         fs_info->chunk_block_rsv.space_info = space_info;
5736 
5737         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5738         fs_info->global_block_rsv.space_info = space_info;
5739         fs_info->delalloc_block_rsv.space_info = space_info;
5740         fs_info->trans_block_rsv.space_info = space_info;
5741         fs_info->empty_block_rsv.space_info = space_info;
5742         fs_info->delayed_block_rsv.space_info = space_info;
5743 
5744         fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
5745         fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
5746         fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
5747         fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
5748         if (fs_info->quota_root)
5749                 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
5750         fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
5751 
5752         update_global_block_rsv(fs_info);
5753 }
5754 
5755 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
5756 {
5757         block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
5758                                 (u64)-1);
5759         WARN_ON(fs_info->delalloc_block_rsv.size > 0);
5760         WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
5761         WARN_ON(fs_info->trans_block_rsv.size > 0);
5762         WARN_ON(fs_info->trans_block_rsv.reserved > 0);
5763         WARN_ON(fs_info->chunk_block_rsv.size > 0);
5764         WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
5765         WARN_ON(fs_info->delayed_block_rsv.size > 0);
5766         WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
5767 }
5768 
5769 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
5770                                   struct btrfs_fs_info *fs_info)
5771 {
5772         if (!trans->block_rsv)
5773                 return;
5774 
5775         if (!trans->bytes_reserved)
5776                 return;
5777 
5778         trace_btrfs_space_reservation(fs_info, "transaction",
5779                                       trans->transid, trans->bytes_reserved, 0);
5780         btrfs_block_rsv_release(fs_info, trans->block_rsv,
5781                                 trans->bytes_reserved);
5782         trans->bytes_reserved = 0;
5783 }
5784 
5785 /*
5786  * To be called after all the new block groups attached to the transaction
5787  * handle have been created (btrfs_create_pending_block_groups()).
5788  */
5789 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans)
5790 {
5791         struct btrfs_fs_info *fs_info = trans->fs_info;
5792 
5793         if (!trans->chunk_bytes_reserved)
5794                 return;
5795 
5796         WARN_ON_ONCE(!list_empty(&trans->new_bgs));
5797 
5798         block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL,
5799                                 trans->chunk_bytes_reserved);
5800         trans->chunk_bytes_reserved = 0;
5801 }
5802 
5803 /* Can only return 0 or -ENOSPC */
5804 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
5805                                   struct btrfs_inode *inode)
5806 {
5807         struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
5808         struct btrfs_root *root = inode->root;
5809         /*
5810          * We always use trans->block_rsv here as we will have reserved space
5811          * for our orphan when starting the transaction, using get_block_rsv()
5812          * here will sometimes make us choose the wrong block rsv as we could be
5813          * doing a reloc inode for a non refcounted root.
5814          */
5815         struct btrfs_block_rsv *src_rsv = trans->block_rsv;
5816         struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
5817 
5818         /*
5819          * We need to hold space in order to delete our orphan item once we've
5820          * added it, so this takes the reservation so we can release it later
5821          * when we are truly done with the orphan item.
5822          */
5823         u64 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
5824 
5825         trace_btrfs_space_reservation(fs_info, "orphan", btrfs_ino(inode), 
5826                         num_bytes, 1);
5827         return btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, 1);
5828 }
5829 
5830 void btrfs_orphan_release_metadata(struct btrfs_inode *inode)
5831 {
5832         struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
5833         struct btrfs_root *root = inode->root;
5834         u64 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
5835 
5836         trace_btrfs_space_reservation(fs_info, "orphan", btrfs_ino(inode),
5837                         num_bytes, 0);
5838         btrfs_block_rsv_release(fs_info, root->orphan_block_rsv, num_bytes);
5839 }
5840 
5841 /*
5842  * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5843  * root: the root of the parent directory
5844  * rsv: block reservation
5845  * items: the number of items that we need do reservation
5846  * qgroup_reserved: used to return the reserved size in qgroup
5847  *
5848  * This function is used to reserve the space for snapshot/subvolume
5849  * creation and deletion. Those operations are different with the
5850  * common file/directory operations, they change two fs/file trees
5851  * and root tree, the number of items that the qgroup reserves is
5852  * different with the free space reservation. So we can not use
5853  * the space reservation mechanism in start_transaction().
5854  */
5855 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
5856                                      struct btrfs_block_rsv *rsv,
5857                                      int items,
5858                                      u64 *qgroup_reserved,
5859                                      bool use_global_rsv)
5860 {
5861         u64 num_bytes;
5862         int ret;
5863         struct btrfs_fs_info *fs_info = root->fs_info;
5864         struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5865 
5866         if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
5867                 /* One for parent inode, two for dir entries */
5868                 num_bytes = 3 * fs_info->nodesize;
5869                 ret = btrfs_qgroup_reserve_meta(root, num_bytes, true);
5870                 if (ret)
5871                         return ret;
5872         } else {
5873                 num_bytes = 0;
5874         }
5875 
5876         *qgroup_reserved = num_bytes;
5877 
5878         num_bytes = btrfs_calc_trans_metadata_size(fs_info, items);
5879         rsv->space_info = __find_space_info(fs_info,
5880                                             BTRFS_BLOCK_GROUP_METADATA);
5881         ret = btrfs_block_rsv_add(root, rsv, num_bytes,
5882                                   BTRFS_RESERVE_FLUSH_ALL);
5883 
5884         if (ret == -ENOSPC && use_global_rsv)
5885                 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes, 1);
5886 
5887         if (ret && *qgroup_reserved)
5888                 btrfs_qgroup_free_meta(root, *qgroup_reserved);
5889 
5890         return ret;
5891 }
5892 
5893 void btrfs_subvolume_release_metadata(struct btrfs_fs_info *fs_info,
5894                                       struct btrfs_block_rsv *rsv)
5895 {
5896         btrfs_block_rsv_release(fs_info, rsv, (u64)-1);
5897 }
5898 
5899 /**
5900  * drop_outstanding_extent - drop an outstanding extent
5901  * @inode: the inode we're dropping the extent for
5902  * @num_bytes: the number of bytes we're releasing.
5903  *
5904  * This is called when we are freeing up an outstanding extent, either called
5905  * after an error or after an extent is written.  This will return the number of
5906  * reserved extents that need to be freed.  This must be called with
5907  * BTRFS_I(inode)->lock held.
5908  */
5909 static unsigned drop_outstanding_extent(struct btrfs_inode *inode,
5910                 u64 num_bytes)
5911 {
5912         unsigned drop_inode_space = 0;
5913         unsigned dropped_extents = 0;
5914         unsigned num_extents;
5915 
5916         num_extents = count_max_extents(num_bytes);
5917         ASSERT(num_extents);
5918         ASSERT(inode->outstanding_extents >= num_extents);
5919         inode->outstanding_extents -= num_extents;
5920 
5921         if (inode->outstanding_extents == 0 &&
5922             test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5923                                &inode->runtime_flags))
5924                 drop_inode_space = 1;
5925 
5926         /*
5927          * If we have more or the same amount of outstanding extents than we have
5928          * reserved then we need to leave the reserved extents count alone.
5929          */
5930         if (inode->outstanding_extents >= inode->reserved_extents)
5931                 return drop_inode_space;
5932 
5933         dropped_extents = inode->reserved_extents - inode->outstanding_extents;
5934         inode->reserved_extents -= dropped_extents;
5935         return dropped_extents + drop_inode_space;
5936 }
5937 
5938 /**
5939  * calc_csum_metadata_size - return the amount of metadata space that must be
5940  *      reserved/freed for the given bytes.
5941  * @inode: the inode we're manipulating
5942  * @num_bytes: the number of bytes in question
5943  * @reserve: 1 if we are reserving space, 0 if we are freeing space
5944  *
5945  * This adjusts the number of csum_bytes in the inode and then returns the
5946  * correct amount of metadata that must either be reserved or freed.  We
5947  * calculate how many checksums we can fit into one leaf and then divide the
5948  * number of bytes that will need to be checksumed by this value to figure out
5949  * how many checksums will be required.  If we are adding bytes then the number