?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Driver Core: devtmpfs - kernel-maintained tmpfs-based /dev
[linux/fpc-iii.git] / fs / btrfs / extent-tree.c
blob535f85ba104f41fea1c28820533da7e1af1512c3
1 /*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18 #include <linux/sched.h>
19 #include <linux/pagemap.h>
20 #include <linux/writeback.h>
21 #include <linux/blkdev.h>
22 #include <linux/sort.h>
23 #include <linux/rcupdate.h>
24 #include <linux/kthread.h>
25 #include "compat.h"
26 #include "hash.h"
27 #include "ctree.h"
28 #include "disk-io.h"
29 #include "print-tree.h"
30 #include "transaction.h"
31 #include "volumes.h"
32 #include "locking.h"
33 #include "free-space-cache.h"
34
35 static int update_reserved_extents(struct btrfs_root *root,
36 u64 bytenr, u64 num, int reserve);
37 static int update_block_group(struct btrfs_trans_handle *trans,
38 struct btrfs_root *root,
39 u64 bytenr, u64 num_bytes, int alloc,
40 int mark_free);
41 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
42 struct btrfs_root *root,
43 u64 bytenr, u64 num_bytes, u64 parent,
44 u64 root_objectid, u64 owner_objectid,
45 u64 owner_offset, int refs_to_drop,
46 struct btrfs_delayed_extent_op *extra_op);
47 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
48 struct extent_buffer *leaf,
49 struct btrfs_extent_item *ei);
50 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
51 struct btrfs_root *root,
52 u64 parent, u64 root_objectid,
53 u64 flags, u64 owner, u64 offset,
54 struct btrfs_key *ins, int ref_mod);
55 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
56 struct btrfs_root *root,
57 u64 parent, u64 root_objectid,
58 u64 flags, struct btrfs_disk_key *key,
59 int level, struct btrfs_key *ins);
60
61 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
62 struct btrfs_root *extent_root, u64 alloc_bytes,
63 u64 flags, int force);
64
65 static noinline int
66 block_group_cache_done(struct btrfs_block_group_cache *cache)
67 {
68 smp_mb();
69 return cache->cached == BTRFS_CACHE_FINISHED;
70 }
71
72 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
73 {
74 return (cache->flags & bits) == bits;
75 }
76
77 /*
78 * this adds the block group to the fs_info rb tree for the block group
79 * cache
80 */
81 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
82 struct btrfs_block_group_cache *block_group)
83 {
84 struct rb_node **p;
85 struct rb_node *parent = NULL;
86 struct btrfs_block_group_cache *cache;
87
88 spin_lock(&info->block_group_cache_lock);
89 p = &info->block_group_cache_tree.rb_node;
90
91 while (*p) {
92 parent = *p;
93 cache = rb_entry(parent, struct btrfs_block_group_cache,
94 cache_node);
95 if (block_group->key.objectid < cache->key.objectid) {
96 p = &(*p)->rb_left;
97 } else if (block_group->key.objectid > cache->key.objectid) {
98 p = &(*p)->rb_right;
99 } else {
100 spin_unlock(&info->block_group_cache_lock);
101 return -EEXIST;
102 }
103 }
104
105 rb_link_node(&block_group->cache_node, parent, p);
106 rb_insert_color(&block_group->cache_node,
107 &info->block_group_cache_tree);
108 spin_unlock(&info->block_group_cache_lock);
109
110 return 0;
111 }
112
113 /*
114 * This will return the block group at or after bytenr if contains is 0, else
115 * it will return the block group that contains the bytenr
116 */
117 static struct btrfs_block_group_cache *
118 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
119 int contains)
120 {
121 struct btrfs_block_group_cache *cache, *ret = NULL;
122 struct rb_node *n;
123 u64 end, start;
124
125 spin_lock(&info->block_group_cache_lock);
126 n = info->block_group_cache_tree.rb_node;
127
128 while (n) {
129 cache = rb_entry(n, struct btrfs_block_group_cache,
130 cache_node);
131 end = cache->key.objectid + cache->key.offset - 1;
132 start = cache->key.objectid;
133
134 if (bytenr < start) {
135 if (!contains && (!ret || start < ret->key.objectid))
136 ret = cache;
137 n = n->rb_left;
138 } else if (bytenr > start) {
139 if (contains && bytenr <= end) {
140 ret = cache;
141 break;
142 }
143 n = n->rb_right;
144 } else {
145 ret = cache;
146 break;
147 }
148 }
149 if (ret)
150 atomic_inc(&ret->count);
151 spin_unlock(&info->block_group_cache_lock);
152
153 return ret;
154 }
155
156 /*
157 * We always set EXTENT_LOCKED for the super mirror extents so we don't
158 * overwrite them, so those bits need to be unset. Also, if we are unmounting
159 * with pinned extents still sitting there because we had a block group caching,
160 * we need to clear those now, since we are done.
161 */
162 void btrfs_free_pinned_extents(struct btrfs_fs_info *info)
163 {
164 u64 start, end, last = 0;
165 int ret;
166
167 while (1) {
168 ret = find_first_extent_bit(&info->pinned_extents, last,
169 &start, &end,
170 EXTENT_LOCKED|EXTENT_DIRTY);
171 if (ret)
172 break;
173
174 clear_extent_bits(&info->pinned_extents, start, end,
175 EXTENT_LOCKED|EXTENT_DIRTY, GFP_NOFS);
176 last = end+1;
177 }
178 }
179
180 static int remove_sb_from_cache(struct btrfs_root *root,
181 struct btrfs_block_group_cache *cache)
182 {
183 struct btrfs_fs_info *fs_info = root->fs_info;
184 u64 bytenr;
185 u64 *logical;
186 int stripe_len;
187 int i, nr, ret;
188
189 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
190 bytenr = btrfs_sb_offset(i);
191 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
192 cache->key.objectid, bytenr,
193 0, &logical, &nr, &stripe_len);
194 BUG_ON(ret);
195 while (nr--) {
196 try_lock_extent(&fs_info->pinned_extents,
197 logical[nr],
198 logical[nr] + stripe_len - 1, GFP_NOFS);
199 }
200 kfree(logical);
201 }
202
203 return 0;
204 }
205
206 /*
207 * this is only called by cache_block_group, since we could have freed extents
208 * we need to check the pinned_extents for any extents that can't be used yet
209 * since their free space will be released as soon as the transaction commits.
210 */
211 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
212 struct btrfs_fs_info *info, u64 start, u64 end)
213 {
214 u64 extent_start, extent_end, size, total_added = 0;
215 int ret;
216
217 while (start < end) {
218 ret = find_first_extent_bit(&info->pinned_extents, start,
219 &extent_start, &extent_end,
220 EXTENT_DIRTY|EXTENT_LOCKED);
221 if (ret)
222 break;
223
224 if (extent_start == start) {
225 start = extent_end + 1;
226 } else if (extent_start > start && extent_start < end) {
227 size = extent_start - start;
228 total_added += size;
229 ret = btrfs_add_free_space(block_group, start,
230 size);
231 BUG_ON(ret);
232 start = extent_end + 1;
233 } else {
234 break;
235 }
236 }
237
238 if (start < end) {
239 size = end - start;
240 total_added += size;
241 ret = btrfs_add_free_space(block_group, start, size);
242 BUG_ON(ret);
243 }
244
245 return total_added;
246 }
247
248 static int caching_kthread(void *data)
249 {
250 struct btrfs_block_group_cache *block_group = data;
251 struct btrfs_fs_info *fs_info = block_group->fs_info;
252 u64 last = 0;
253 struct btrfs_path *path;
254 int ret = 0;
255 struct btrfs_key key;
256 struct extent_buffer *leaf;
257 int slot;
258 u64 total_found = 0;
259
260 BUG_ON(!fs_info);
261
262 path = btrfs_alloc_path();
263 if (!path)
264 return -ENOMEM;
265
266 atomic_inc(&block_group->space_info->caching_threads);
267 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
268 /*
269 * We don't want to deadlock with somebody trying to allocate a new
270 * extent for the extent root while also trying to search the extent
271 * root to add free space. So we skip locking and search the commit
272 * root, since its read-only
273 */
274 path->skip_locking = 1;
275 path->search_commit_root = 1;
276 path->reada = 2;
277
278 key.objectid = last;
279 key.offset = 0;
280 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
281 again:
282 /* need to make sure the commit_root doesn't disappear */
283 down_read(&fs_info->extent_commit_sem);
284
285 ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
286 if (ret < 0)
287 goto err;
288
289 while (1) {
290 smp_mb();
291 if (block_group->fs_info->closing > 1) {
292 last = (u64)-1;
293 break;
294 }
295
296 leaf = path->nodes[0];
297 slot = path->slots[0];
298 if (slot >= btrfs_header_nritems(leaf)) {
299 ret = btrfs_next_leaf(fs_info->extent_root, path);
300 if (ret < 0)
301 goto err;
302 else if (ret)
303 break;
304
305 if (need_resched() ||
306 btrfs_transaction_in_commit(fs_info)) {
307 leaf = path->nodes[0];
308
309 /* this shouldn't happen, but if the
310 * leaf is empty just move on.
311 */
312 if (btrfs_header_nritems(leaf) == 0)
313 break;
314 /*
315 * we need to copy the key out so that
316 * we are sure the next search advances
317 * us forward in the btree.
318 */
319 btrfs_item_key_to_cpu(leaf, &key, 0);
320 btrfs_release_path(fs_info->extent_root, path);
321 up_read(&fs_info->extent_commit_sem);
322 schedule_timeout(1);
323 goto again;
324 }
325
326 continue;
327 }
328 btrfs_item_key_to_cpu(leaf, &key, slot);
329 if (key.objectid < block_group->key.objectid)
330 goto next;
331
332 if (key.objectid >= block_group->key.objectid +
333 block_group->key.offset)
334 break;
335
336 if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
337 total_found += add_new_free_space(block_group,
338 fs_info, last,
339 key.objectid);
340 last = key.objectid + key.offset;
341 }
342
343 if (total_found > (1024 * 1024 * 2)) {
344 total_found = 0;
345 wake_up(&block_group->caching_q);
346 }
347 next:
348 path->slots[0]++;
349 }
350 ret = 0;
351
352 total_found += add_new_free_space(block_group, fs_info, last,
353 block_group->key.objectid +
354 block_group->key.offset);
355
356 spin_lock(&block_group->lock);
357 block_group->cached = BTRFS_CACHE_FINISHED;
358 spin_unlock(&block_group->lock);
359
360 err:
361 btrfs_free_path(path);
362 up_read(&fs_info->extent_commit_sem);
363 atomic_dec(&block_group->space_info->caching_threads);
364 wake_up(&block_group->caching_q);
365
366 return 0;
367 }
368
369 static int cache_block_group(struct btrfs_block_group_cache *cache)
370 {
371 struct task_struct *tsk;
372 int ret = 0;
373
374 spin_lock(&cache->lock);
375 if (cache->cached != BTRFS_CACHE_NO) {
376 spin_unlock(&cache->lock);
377 return ret;
378 }
379 cache->cached = BTRFS_CACHE_STARTED;
380 spin_unlock(&cache->lock);
381
382 tsk = kthread_run(caching_kthread, cache, "btrfs-cache-%llu\n",
383 cache->key.objectid);
384 if (IS_ERR(tsk)) {
385 ret = PTR_ERR(tsk);
386 printk(KERN_ERR "error running thread %d\n", ret);
387 BUG();
388 }
389
390 return ret;
391 }
392
393 /*
394 * return the block group that starts at or after bytenr
395 */
396 static struct btrfs_block_group_cache *
397 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
398 {
399 struct btrfs_block_group_cache *cache;
400
401 cache = block_group_cache_tree_search(info, bytenr, 0);
402
403 return cache;
404 }
405
406 /*
407 * return the block group that contains the given bytenr
408 */
409 struct btrfs_block_group_cache *btrfs_lookup_block_group(
410 struct btrfs_fs_info *info,
411 u64 bytenr)
412 {
413 struct btrfs_block_group_cache *cache;
414
415 cache = block_group_cache_tree_search(info, bytenr, 1);
416
417 return cache;
418 }
419
420 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
421 {
422 if (atomic_dec_and_test(&cache->count))
423 kfree(cache);
424 }
425
426 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
427 u64 flags)
428 {
429 struct list_head *head = &info->space_info;
430 struct btrfs_space_info *found;
431
432 rcu_read_lock();
433 list_for_each_entry_rcu(found, head, list) {
434 if (found->flags == flags) {
435 rcu_read_unlock();
436 return found;
437 }
438 }
439 rcu_read_unlock();
440 return NULL;
441 }
442
443 /*
444 * after adding space to the filesystem, we need to clear the full flags
445 * on all the space infos.
446 */
447 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
448 {
449 struct list_head *head = &info->space_info;
450 struct btrfs_space_info *found;
451
452 rcu_read_lock();
453 list_for_each_entry_rcu(found, head, list)
454 found->full = 0;
455 rcu_read_unlock();
456 }
457
458 static u64 div_factor(u64 num, int factor)
459 {
460 if (factor == 10)
461 return num;
462 num *= factor;
463 do_div(num, 10);
464 return num;
465 }
466
467 u64 btrfs_find_block_group(struct btrfs_root *root,
468 u64 search_start, u64 search_hint, int owner)
469 {
470 struct btrfs_block_group_cache *cache;
471 u64 used;
472 u64 last = max(search_hint, search_start);
473 u64 group_start = 0;
474 int full_search = 0;
475 int factor = 9;
476 int wrapped = 0;
477 again:
478 while (1) {
479 cache = btrfs_lookup_first_block_group(root->fs_info, last);
480 if (!cache)
481 break;
482
483 spin_lock(&cache->lock);
484 last = cache->key.objectid + cache->key.offset;
485 used = btrfs_block_group_used(&cache->item);
486
487 if ((full_search || !cache->ro) &&
488 block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
489 if (used + cache->pinned + cache->reserved <
490 div_factor(cache->key.offset, factor)) {
491 group_start = cache->key.objectid;
492 spin_unlock(&cache->lock);
493 btrfs_put_block_group(cache);
494 goto found;
495 }
496 }
497 spin_unlock(&cache->lock);
498 btrfs_put_block_group(cache);
499 cond_resched();
500 }
501 if (!wrapped) {
502 last = search_start;
503 wrapped = 1;
504 goto again;
505 }
506 if (!full_search && factor < 10) {
507 last = search_start;
508 full_search = 1;
509 factor = 10;
510 goto again;
511 }
512 found:
513 return group_start;
514 }
515
516 /* simple helper to search for an existing extent at a given offset */
517 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
518 {
519 int ret;
520 struct btrfs_key key;
521 struct btrfs_path *path;
522
523 path = btrfs_alloc_path();
524 BUG_ON(!path);
525 key.objectid = start;
526 key.offset = len;
527 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
528 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
529 0, 0);
530 btrfs_free_path(path);
531 return ret;
532 }
533
534 /*
535 * Back reference rules. Back refs have three main goals:
536 *
537 * 1) differentiate between all holders of references to an extent so that
538 * when a reference is dropped we can make sure it was a valid reference
539 * before freeing the extent.
540 *
541 * 2) Provide enough information to quickly find the holders of an extent
542 * if we notice a given block is corrupted or bad.
543 *
544 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
545 * maintenance. This is actually the same as #2, but with a slightly
546 * different use case.
547 *
548 * There are two kinds of back refs. The implicit back refs is optimized
549 * for pointers in non-shared tree blocks. For a given pointer in a block,
550 * back refs of this kind provide information about the block's owner tree
551 * and the pointer's key. These information allow us to find the block by
552 * b-tree searching. The full back refs is for pointers in tree blocks not
553 * referenced by their owner trees. The location of tree block is recorded
554 * in the back refs. Actually the full back refs is generic, and can be
555 * used in all cases the implicit back refs is used. The major shortcoming
556 * of the full back refs is its overhead. Every time a tree block gets
557 * COWed, we have to update back refs entry for all pointers in it.
558 *
559 * For a newly allocated tree block, we use implicit back refs for
560 * pointers in it. This means most tree related operations only involve
561 * implicit back refs. For a tree block created in old transaction, the
562 * only way to drop a reference to it is COW it. So we can detect the
563 * event that tree block loses its owner tree's reference and do the
564 * back refs conversion.
565 *
566 * When a tree block is COW'd through a tree, there are four cases:
567 *
568 * The reference count of the block is one and the tree is the block's
569 * owner tree. Nothing to do in this case.
570 *
571 * The reference count of the block is one and the tree is not the
572 * block's owner tree. In this case, full back refs is used for pointers
573 * in the block. Remove these full back refs, add implicit back refs for
574 * every pointers in the new block.
575 *
576 * The reference count of the block is greater than one and the tree is
577 * the block's owner tree. In this case, implicit back refs is used for
578 * pointers in the block. Add full back refs for every pointers in the
579 * block, increase lower level extents' reference counts. The original
580 * implicit back refs are entailed to the new block.
581 *
582 * The reference count of the block is greater than one and the tree is
583 * not the block's owner tree. Add implicit back refs for every pointer in
584 * the new block, increase lower level extents' reference count.
585 *
586 * Back Reference Key composing:
587 *
588 * The key objectid corresponds to the first byte in the extent,
589 * The key type is used to differentiate between types of back refs.
590 * There are different meanings of the key offset for different types
591 * of back refs.
592 *
593 * File extents can be referenced by:
594 *
595 * - multiple snapshots, subvolumes, or different generations in one subvol
596 * - different files inside a single subvolume
597 * - different offsets inside a file (bookend extents in file.c)
598 *
599 * The extent ref structure for the implicit back refs has fields for:
600 *
601 * - Objectid of the subvolume root
602 * - objectid of the file holding the reference
603 * - original offset in the file
604 * - how many bookend extents
605 *
606 * The key offset for the implicit back refs is hash of the first
607 * three fields.
608 *
609 * The extent ref structure for the full back refs has field for:
610 *
611 * - number of pointers in the tree leaf
612 *
613 * The key offset for the implicit back refs is the first byte of
614 * the tree leaf
615 *
616 * When a file extent is allocated, The implicit back refs is used.
617 * the fields are filled in:
618 *
619 * (root_key.objectid, inode objectid, offset in file, 1)
620 *
621 * When a file extent is removed file truncation, we find the
622 * corresponding implicit back refs and check the following fields:
623 *
624 * (btrfs_header_owner(leaf), inode objectid, offset in file)
625 *
626 * Btree extents can be referenced by:
627 *
628 * - Different subvolumes
629 *
630 * Both the implicit back refs and the full back refs for tree blocks
631 * only consist of key. The key offset for the implicit back refs is
632 * objectid of block's owner tree. The key offset for the full back refs
633 * is the first byte of parent block.
634 *
635 * When implicit back refs is used, information about the lowest key and
636 * level of the tree block are required. These information are stored in
637 * tree block info structure.
638 */
639
640 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
641 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
642 struct btrfs_root *root,
643 struct btrfs_path *path,
644 u64 owner, u32 extra_size)
645 {
646 struct btrfs_extent_item *item;
647 struct btrfs_extent_item_v0 *ei0;
648 struct btrfs_extent_ref_v0 *ref0;
649 struct btrfs_tree_block_info *bi;
650 struct extent_buffer *leaf;
651 struct btrfs_key key;
652 struct btrfs_key found_key;
653 u32 new_size = sizeof(*item);
654 u64 refs;
655 int ret;
656
657 leaf = path->nodes[0];
658 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
659
660 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
661 ei0 = btrfs_item_ptr(leaf, path->slots[0],
662 struct btrfs_extent_item_v0);
663 refs = btrfs_extent_refs_v0(leaf, ei0);
664
665 if (owner == (u64)-1) {
666 while (1) {
667 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
668 ret = btrfs_next_leaf(root, path);
669 if (ret < 0)
670 return ret;
671 BUG_ON(ret > 0);
672 leaf = path->nodes[0];
673 }
674 btrfs_item_key_to_cpu(leaf, &found_key,
675 path->slots[0]);
676 BUG_ON(key.objectid != found_key.objectid);
677 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
678 path->slots[0]++;
679 continue;
680 }
681 ref0 = btrfs_item_ptr(leaf, path->slots[0],
682 struct btrfs_extent_ref_v0);
683 owner = btrfs_ref_objectid_v0(leaf, ref0);
684 break;
685 }
686 }
687 btrfs_release_path(root, path);
688
689 if (owner < BTRFS_FIRST_FREE_OBJECTID)
690 new_size += sizeof(*bi);
691
692 new_size -= sizeof(*ei0);
693 ret = btrfs_search_slot(trans, root, &key, path,
694 new_size + extra_size, 1);
695 if (ret < 0)
696 return ret;
697 BUG_ON(ret);
698
699 ret = btrfs_extend_item(trans, root, path, new_size);
700 BUG_ON(ret);
701
702 leaf = path->nodes[0];
703 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
704 btrfs_set_extent_refs(leaf, item, refs);
705 /* FIXME: get real generation */
706 btrfs_set_extent_generation(leaf, item, 0);
707 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
708 btrfs_set_extent_flags(leaf, item,
709 BTRFS_EXTENT_FLAG_TREE_BLOCK |
710 BTRFS_BLOCK_FLAG_FULL_BACKREF);
711 bi = (struct btrfs_tree_block_info *)(item + 1);
712 /* FIXME: get first key of the block */
713 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
714 btrfs_set_tree_block_level(leaf, bi, (int)owner);
715 } else {
716 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
717 }
718 btrfs_mark_buffer_dirty(leaf);
719 return 0;
720 }
721 #endif
722
723 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
724 {
725 u32 high_crc = ~(u32)0;
726 u32 low_crc = ~(u32)0;
727 __le64 lenum;
728
729 lenum = cpu_to_le64(root_objectid);
730 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
731 lenum = cpu_to_le64(owner);
732 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
733 lenum = cpu_to_le64(offset);
734 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
735
736 return ((u64)high_crc << 31) ^ (u64)low_crc;
737 }
738
739 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
740 struct btrfs_extent_data_ref *ref)
741 {
742 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
743 btrfs_extent_data_ref_objectid(leaf, ref),
744 btrfs_extent_data_ref_offset(leaf, ref));
745 }
746
747 static int match_extent_data_ref(struct extent_buffer *leaf,
748 struct btrfs_extent_data_ref *ref,
749 u64 root_objectid, u64 owner, u64 offset)
750 {
751 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
752 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
753 btrfs_extent_data_ref_offset(leaf, ref) != offset)
754 return 0;
755 return 1;
756 }
757
758 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
759 struct btrfs_root *root,
760 struct btrfs_path *path,
761 u64 bytenr, u64 parent,
762 u64 root_objectid,
763 u64 owner, u64 offset)
764 {
765 struct btrfs_key key;
766 struct btrfs_extent_data_ref *ref;
767 struct extent_buffer *leaf;
768 u32 nritems;
769 int ret;
770 int recow;
771 int err = -ENOENT;
772
773 key.objectid = bytenr;
774 if (parent) {
775 key.type = BTRFS_SHARED_DATA_REF_KEY;
776 key.offset = parent;
777 } else {
778 key.type = BTRFS_EXTENT_DATA_REF_KEY;
779 key.offset = hash_extent_data_ref(root_objectid,
780 owner, offset);
781 }
782 again:
783 recow = 0;
784 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
785 if (ret < 0) {
786 err = ret;
787 goto fail;
788 }
789
790 if (parent) {
791 if (!ret)
792 return 0;
793 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
794 key.type = BTRFS_EXTENT_REF_V0_KEY;
795 btrfs_release_path(root, path);
796 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
797 if (ret < 0) {
798 err = ret;
799 goto fail;
800 }
801 if (!ret)
802 return 0;
803 #endif
804 goto fail;
805 }
806
807 leaf = path->nodes[0];
808 nritems = btrfs_header_nritems(leaf);
809 while (1) {
810 if (path->slots[0] >= nritems) {
811 ret = btrfs_next_leaf(root, path);
812 if (ret < 0)
813 err = ret;
814 if (ret)
815 goto fail;
816
817 leaf = path->nodes[0];
818 nritems = btrfs_header_nritems(leaf);
819 recow = 1;
820 }
821
822 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
823 if (key.objectid != bytenr ||
824 key.type != BTRFS_EXTENT_DATA_REF_KEY)
825 goto fail;
826
827 ref = btrfs_item_ptr(leaf, path->slots[0],
828 struct btrfs_extent_data_ref);
829
830 if (match_extent_data_ref(leaf, ref, root_objectid,
831 owner, offset)) {
832 if (recow) {
833 btrfs_release_path(root, path);
834 goto again;
835 }
836 err = 0;
837 break;
838 }
839 path->slots[0]++;
840 }
841 fail:
842 return err;
843 }
844
845 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
846 struct btrfs_root *root,
847 struct btrfs_path *path,
848 u64 bytenr, u64 parent,
849 u64 root_objectid, u64 owner,
850 u64 offset, int refs_to_add)
851 {
852 struct btrfs_key key;
853 struct extent_buffer *leaf;
854 u32 size;
855 u32 num_refs;
856 int ret;
857
858 key.objectid = bytenr;
859 if (parent) {
860 key.type = BTRFS_SHARED_DATA_REF_KEY;
861 key.offset = parent;
862 size = sizeof(struct btrfs_shared_data_ref);
863 } else {
864 key.type = BTRFS_EXTENT_DATA_REF_KEY;
865 key.offset = hash_extent_data_ref(root_objectid,
866 owner, offset);
867 size = sizeof(struct btrfs_extent_data_ref);
868 }
869
870 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
871 if (ret && ret != -EEXIST)
872 goto fail;
873
874 leaf = path->nodes[0];
875 if (parent) {
876 struct btrfs_shared_data_ref *ref;
877 ref = btrfs_item_ptr(leaf, path->slots[0],
878 struct btrfs_shared_data_ref);
879 if (ret == 0) {
880 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
881 } else {
882 num_refs = btrfs_shared_data_ref_count(leaf, ref);
883 num_refs += refs_to_add;
884 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
885 }
886 } else {
887 struct btrfs_extent_data_ref *ref;
888 while (ret == -EEXIST) {
889 ref = btrfs_item_ptr(leaf, path->slots[0],
890 struct btrfs_extent_data_ref);
891 if (match_extent_data_ref(leaf, ref, root_objectid,
892 owner, offset))
893 break;
894 btrfs_release_path(root, path);
895 key.offset++;
896 ret = btrfs_insert_empty_item(trans, root, path, &key,
897 size);
898 if (ret && ret != -EEXIST)
899 goto fail;
900
901 leaf = path->nodes[0];
902 }
903 ref = btrfs_item_ptr(leaf, path->slots[0],
904 struct btrfs_extent_data_ref);
905 if (ret == 0) {
906 btrfs_set_extent_data_ref_root(leaf, ref,
907 root_objectid);
908 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
909 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
910 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
911 } else {
912 num_refs = btrfs_extent_data_ref_count(leaf, ref);
913 num_refs += refs_to_add;
914 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
915 }
916 }
917 btrfs_mark_buffer_dirty(leaf);
918 ret = 0;
919 fail:
920 btrfs_release_path(root, path);
921 return ret;
922 }
923
924 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
925 struct btrfs_root *root,
926 struct btrfs_path *path,
927 int refs_to_drop)
928 {
929 struct btrfs_key key;
930 struct btrfs_extent_data_ref *ref1 = NULL;
931 struct btrfs_shared_data_ref *ref2 = NULL;
932 struct extent_buffer *leaf;
933 u32 num_refs = 0;
934 int ret = 0;
935
936 leaf = path->nodes[0];
937 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
938
939 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
940 ref1 = btrfs_item_ptr(leaf, path->slots[0],
941 struct btrfs_extent_data_ref);
942 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
943 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
944 ref2 = btrfs_item_ptr(leaf, path->slots[0],
945 struct btrfs_shared_data_ref);
946 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
947 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
948 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
949 struct btrfs_extent_ref_v0 *ref0;
950 ref0 = btrfs_item_ptr(leaf, path->slots[0],
951 struct btrfs_extent_ref_v0);
952 num_refs = btrfs_ref_count_v0(leaf, ref0);
953 #endif
954 } else {
955 BUG();
956 }
957
958 BUG_ON(num_refs < refs_to_drop);
959 num_refs -= refs_to_drop;
960
961 if (num_refs == 0) {
962 ret = btrfs_del_item(trans, root, path);
963 } else {
964 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
965 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
966 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
967 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
968 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
969 else {
970 struct btrfs_extent_ref_v0 *ref0;
971 ref0 = btrfs_item_ptr(leaf, path->slots[0],
972 struct btrfs_extent_ref_v0);
973 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
974 }
975 #endif
976 btrfs_mark_buffer_dirty(leaf);
977 }
978 return ret;
979 }
980
981 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
982 struct btrfs_path *path,
983 struct btrfs_extent_inline_ref *iref)
984 {
985 struct btrfs_key key;
986 struct extent_buffer *leaf;
987 struct btrfs_extent_data_ref *ref1;
988 struct btrfs_shared_data_ref *ref2;
989 u32 num_refs = 0;
990
991 leaf = path->nodes[0];
992 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
993 if (iref) {
994 if (btrfs_extent_inline_ref_type(leaf, iref) ==
995 BTRFS_EXTENT_DATA_REF_KEY) {
996 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
997 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
998 } else {
999 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1000 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1001 }
1002 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1003 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1004 struct btrfs_extent_data_ref);
1005 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1006 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1007 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1008 struct btrfs_shared_data_ref);
1009 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1010 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1011 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1012 struct btrfs_extent_ref_v0 *ref0;
1013 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1014 struct btrfs_extent_ref_v0);
1015 num_refs = btrfs_ref_count_v0(leaf, ref0);
1016 #endif
1017 } else {
1018 WARN_ON(1);
1019 }
1020 return num_refs;
1021 }
1022
1023 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1024 struct btrfs_root *root,
1025 struct btrfs_path *path,
1026 u64 bytenr, u64 parent,
1027 u64 root_objectid)
1028 {
1029 struct btrfs_key key;
1030 int ret;
1031
1032 key.objectid = bytenr;
1033 if (parent) {
1034 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1035 key.offset = parent;
1036 } else {
1037 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1038 key.offset = root_objectid;
1039 }
1040
1041 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1042 if (ret > 0)
1043 ret = -ENOENT;
1044 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1045 if (ret == -ENOENT && parent) {
1046 btrfs_release_path(root, path);
1047 key.type = BTRFS_EXTENT_REF_V0_KEY;
1048 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1049 if (ret > 0)
1050 ret = -ENOENT;
1051 }
1052 #endif
1053 return ret;
1054 }
1055
1056 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1057 struct btrfs_root *root,
1058 struct btrfs_path *path,
1059 u64 bytenr, u64 parent,
1060 u64 root_objectid)
1061 {
1062 struct btrfs_key key;
1063 int ret;
1064
1065 key.objectid = bytenr;
1066 if (parent) {
1067 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1068 key.offset = parent;
1069 } else {
1070 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1071 key.offset = root_objectid;
1072 }
1073
1074 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1075 btrfs_release_path(root, path);
1076 return ret;
1077 }
1078
1079 static inline int extent_ref_type(u64 parent, u64 owner)
1080 {
1081 int type;
1082 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1083 if (parent > 0)
1084 type = BTRFS_SHARED_BLOCK_REF_KEY;
1085 else
1086 type = BTRFS_TREE_BLOCK_REF_KEY;
1087 } else {
1088 if (parent > 0)
1089 type = BTRFS_SHARED_DATA_REF_KEY;
1090 else
1091 type = BTRFS_EXTENT_DATA_REF_KEY;
1092 }
1093 return type;
1094 }
1095
1096 static int find_next_key(struct btrfs_path *path, int level,
1097 struct btrfs_key *key)
1098
1099 {
1100 for (; level < BTRFS_MAX_LEVEL; level++) {
1101 if (!path->nodes[level])
1102 break;
1103 if (path->slots[level] + 1 >=
1104 btrfs_header_nritems(path->nodes[level]))
1105 continue;
1106 if (level == 0)
1107 btrfs_item_key_to_cpu(path->nodes[level], key,
1108 path->slots[level] + 1);
1109 else
1110 btrfs_node_key_to_cpu(path->nodes[level], key,
1111 path->slots[level] + 1);
1112 return 0;
1113 }
1114 return 1;
1115 }
1116
1117 /*
1118 * look for inline back ref. if back ref is found, *ref_ret is set
1119 * to the address of inline back ref, and 0 is returned.
1120 *
1121 * if back ref isn't found, *ref_ret is set to the address where it
1122 * should be inserted, and -ENOENT is returned.
1123 *
1124 * if insert is true and there are too many inline back refs, the path
1125 * points to the extent item, and -EAGAIN is returned.
1126 *
1127 * NOTE: inline back refs are ordered in the same way that back ref
1128 * items in the tree are ordered.
1129 */
1130 static noinline_for_stack
1131 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1132 struct btrfs_root *root,
1133 struct btrfs_path *path,
1134 struct btrfs_extent_inline_ref **ref_ret,
1135 u64 bytenr, u64 num_bytes,
1136 u64 parent, u64 root_objectid,
1137 u64 owner, u64 offset, int insert)
1138 {
1139 struct btrfs_key key;
1140 struct extent_buffer *leaf;
1141 struct btrfs_extent_item *ei;
1142 struct btrfs_extent_inline_ref *iref;
1143 u64 flags;
1144 u64 item_size;
1145 unsigned long ptr;
1146 unsigned long end;
1147 int extra_size;
1148 int type;
1149 int want;
1150 int ret;
1151 int err = 0;
1152
1153 key.objectid = bytenr;
1154 key.type = BTRFS_EXTENT_ITEM_KEY;
1155 key.offset = num_bytes;
1156
1157 want = extent_ref_type(parent, owner);
1158 if (insert) {
1159 extra_size = btrfs_extent_inline_ref_size(want);
1160 path->keep_locks = 1;
1161 } else
1162 extra_size = -1;
1163 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1164 if (ret < 0) {
1165 err = ret;
1166 goto out;
1167 }
1168 BUG_ON(ret);
1169
1170 leaf = path->nodes[0];
1171 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1172 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1173 if (item_size < sizeof(*ei)) {
1174 if (!insert) {
1175 err = -ENOENT;
1176 goto out;
1177 }
1178 ret = convert_extent_item_v0(trans, root, path, owner,
1179 extra_size);
1180 if (ret < 0) {
1181 err = ret;
1182 goto out;
1183 }
1184 leaf = path->nodes[0];
1185 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1186 }
1187 #endif
1188 BUG_ON(item_size < sizeof(*ei));
1189
1190 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1191 flags = btrfs_extent_flags(leaf, ei);
1192
1193 ptr = (unsigned long)(ei + 1);
1194 end = (unsigned long)ei + item_size;
1195
1196 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1197 ptr += sizeof(struct btrfs_tree_block_info);
1198 BUG_ON(ptr > end);
1199 } else {
1200 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1201 }
1202
1203 err = -ENOENT;
1204 while (1) {
1205 if (ptr >= end) {
1206 WARN_ON(ptr > end);
1207 break;
1208 }
1209 iref = (struct btrfs_extent_inline_ref *)ptr;
1210 type = btrfs_extent_inline_ref_type(leaf, iref);
1211 if (want < type)
1212 break;
1213 if (want > type) {
1214 ptr += btrfs_extent_inline_ref_size(type);
1215 continue;
1216 }
1217
1218 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1219 struct btrfs_extent_data_ref *dref;
1220 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1221 if (match_extent_data_ref(leaf, dref, root_objectid,
1222 owner, offset)) {
1223 err = 0;
1224 break;
1225 }
1226 if (hash_extent_data_ref_item(leaf, dref) <
1227 hash_extent_data_ref(root_objectid, owner, offset))
1228 break;
1229 } else {
1230 u64 ref_offset;
1231 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1232 if (parent > 0) {
1233 if (parent == ref_offset) {
1234 err = 0;
1235 break;
1236 }
1237 if (ref_offset < parent)
1238 break;
1239 } else {
1240 if (root_objectid == ref_offset) {
1241 err = 0;
1242 break;
1243 }
1244 if (ref_offset < root_objectid)
1245 break;
1246 }
1247 }
1248 ptr += btrfs_extent_inline_ref_size(type);
1249 }
1250 if (err == -ENOENT && insert) {
1251 if (item_size + extra_size >=
1252 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1253 err = -EAGAIN;
1254 goto out;
1255 }
1256 /*
1257 * To add new inline back ref, we have to make sure
1258 * there is no corresponding back ref item.
1259 * For simplicity, we just do not add new inline back
1260 * ref if there is any kind of item for this block
1261 */
1262 if (find_next_key(path, 0, &key) == 0 &&
1263 key.objectid == bytenr &&
1264 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1265 err = -EAGAIN;
1266 goto out;
1267 }
1268 }
1269 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1270 out:
1271 if (insert) {
1272 path->keep_locks = 0;
1273 btrfs_unlock_up_safe(path, 1);
1274 }
1275 return err;
1276 }
1277
1278 /*
1279 * helper to add new inline back ref
1280 */
1281 static noinline_for_stack
1282 int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1283 struct btrfs_root *root,
1284 struct btrfs_path *path,
1285 struct btrfs_extent_inline_ref *iref,
1286 u64 parent, u64 root_objectid,
1287 u64 owner, u64 offset, int refs_to_add,
1288 struct btrfs_delayed_extent_op *extent_op)
1289 {
1290 struct extent_buffer *leaf;
1291 struct btrfs_extent_item *ei;
1292 unsigned long ptr;
1293 unsigned long end;
1294 unsigned long item_offset;
1295 u64 refs;
1296 int size;
1297 int type;
1298 int ret;
1299
1300 leaf = path->nodes[0];
1301 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1302 item_offset = (unsigned long)iref - (unsigned long)ei;
1303
1304 type = extent_ref_type(parent, owner);
1305 size = btrfs_extent_inline_ref_size(type);
1306
1307 ret = btrfs_extend_item(trans, root, path, size);
1308 BUG_ON(ret);
1309
1310 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1311 refs = btrfs_extent_refs(leaf, ei);
1312 refs += refs_to_add;
1313 btrfs_set_extent_refs(leaf, ei, refs);
1314 if (extent_op)
1315 __run_delayed_extent_op(extent_op, leaf, ei);
1316
1317 ptr = (unsigned long)ei + item_offset;
1318 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1319 if (ptr < end - size)
1320 memmove_extent_buffer(leaf, ptr + size, ptr,
1321 end - size - ptr);
1322
1323 iref = (struct btrfs_extent_inline_ref *)ptr;
1324 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1325 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1326 struct btrfs_extent_data_ref *dref;
1327 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1328 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1329 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1330 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1331 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1332 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1333 struct btrfs_shared_data_ref *sref;
1334 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1335 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1336 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1337 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1338 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1339 } else {
1340 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1341 }
1342 btrfs_mark_buffer_dirty(leaf);
1343 return 0;
1344 }
1345
1346 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1347 struct btrfs_root *root,
1348 struct btrfs_path *path,
1349 struct btrfs_extent_inline_ref **ref_ret,
1350 u64 bytenr, u64 num_bytes, u64 parent,
1351 u64 root_objectid, u64 owner, u64 offset)
1352 {
1353 int ret;
1354
1355 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1356 bytenr, num_bytes, parent,
1357 root_objectid, owner, offset, 0);
1358 if (ret != -ENOENT)
1359 return ret;
1360
1361 btrfs_release_path(root, path);
1362 *ref_ret = NULL;
1363
1364 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1365 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1366 root_objectid);
1367 } else {
1368 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1369 root_objectid, owner, offset);
1370 }
1371 return ret;
1372 }
1373
1374 /*
1375 * helper to update/remove inline back ref
1376 */
1377 static noinline_for_stack
1378 int update_inline_extent_backref(struct btrfs_trans_handle *trans,
1379 struct btrfs_root *root,
1380 struct btrfs_path *path,
1381 struct btrfs_extent_inline_ref *iref,
1382 int refs_to_mod,
1383 struct btrfs_delayed_extent_op *extent_op)
1384 {
1385 struct extent_buffer *leaf;
1386 struct btrfs_extent_item *ei;
1387 struct btrfs_extent_data_ref *dref = NULL;
1388 struct btrfs_shared_data_ref *sref = NULL;
1389 unsigned long ptr;
1390 unsigned long end;
1391 u32 item_size;
1392 int size;
1393 int type;
1394 int ret;
1395 u64 refs;
1396
1397 leaf = path->nodes[0];
1398 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1399 refs = btrfs_extent_refs(leaf, ei);
1400 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1401 refs += refs_to_mod;
1402 btrfs_set_extent_refs(leaf, ei, refs);
1403 if (extent_op)
1404 __run_delayed_extent_op(extent_op, leaf, ei);
1405
1406 type = btrfs_extent_inline_ref_type(leaf, iref);
1407
1408 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1409 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1410 refs = btrfs_extent_data_ref_count(leaf, dref);
1411 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1412 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1413 refs = btrfs_shared_data_ref_count(leaf, sref);
1414 } else {
1415 refs = 1;
1416 BUG_ON(refs_to_mod != -1);
1417 }
1418
1419 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1420 refs += refs_to_mod;
1421
1422 if (refs > 0) {
1423 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1424 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1425 else
1426 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1427 } else {
1428 size = btrfs_extent_inline_ref_size(type);
1429 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1430 ptr = (unsigned long)iref;
1431 end = (unsigned long)ei + item_size;
1432 if (ptr + size < end)
1433 memmove_extent_buffer(leaf, ptr, ptr + size,
1434 end - ptr - size);
1435 item_size -= size;
1436 ret = btrfs_truncate_item(trans, root, path, item_size, 1);
1437 BUG_ON(ret);
1438 }
1439 btrfs_mark_buffer_dirty(leaf);
1440 return 0;
1441 }
1442
1443 static noinline_for_stack
1444 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1445 struct btrfs_root *root,
1446 struct btrfs_path *path,
1447 u64 bytenr, u64 num_bytes, u64 parent,
1448 u64 root_objectid, u64 owner,
1449 u64 offset, int refs_to_add,
1450 struct btrfs_delayed_extent_op *extent_op)
1451 {
1452 struct btrfs_extent_inline_ref *iref;
1453 int ret;
1454
1455 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1456 bytenr, num_bytes, parent,
1457 root_objectid, owner, offset, 1);
1458 if (ret == 0) {
1459 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1460 ret = update_inline_extent_backref(trans, root, path, iref,
1461 refs_to_add, extent_op);
1462 } else if (ret == -ENOENT) {
1463 ret = setup_inline_extent_backref(trans, root, path, iref,
1464 parent, root_objectid,
1465 owner, offset, refs_to_add,
1466 extent_op);
1467 }
1468 return ret;
1469 }
1470
1471 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1472 struct btrfs_root *root,
1473 struct btrfs_path *path,
1474 u64 bytenr, u64 parent, u64 root_objectid,
1475 u64 owner, u64 offset, int refs_to_add)
1476 {
1477 int ret;
1478 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1479 BUG_ON(refs_to_add != 1);
1480 ret = insert_tree_block_ref(trans, root, path, bytenr,
1481 parent, root_objectid);
1482 } else {
1483 ret = insert_extent_data_ref(trans, root, path, bytenr,
1484 parent, root_objectid,
1485 owner, offset, refs_to_add);
1486 }
1487 return ret;
1488 }
1489
1490 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1491 struct btrfs_root *root,
1492 struct btrfs_path *path,
1493 struct btrfs_extent_inline_ref *iref,
1494 int refs_to_drop, int is_data)
1495 {
1496 int ret;
1497
1498 BUG_ON(!is_data && refs_to_drop != 1);
1499 if (iref) {
1500 ret = update_inline_extent_backref(trans, root, path, iref,
1501 -refs_to_drop, NULL);
1502 } else if (is_data) {
1503 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1504 } else {
1505 ret = btrfs_del_item(trans, root, path);
1506 }
1507 return ret;
1508 }
1509
1510 #ifdef BIO_RW_DISCARD
1511 static void btrfs_issue_discard(struct block_device *bdev,
1512 u64 start, u64 len)
1513 {
1514 blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL,
1515 DISCARD_FL_BARRIER);
1516 }
1517 #endif
1518
1519 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1520 u64 num_bytes)
1521 {
1522 #ifdef BIO_RW_DISCARD
1523 int ret;
1524 u64 map_length = num_bytes;
1525 struct btrfs_multi_bio *multi = NULL;
1526
1527 /* Tell the block device(s) that the sectors can be discarded */
1528 ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
1529 bytenr, &map_length, &multi, 0);
1530 if (!ret) {
1531 struct btrfs_bio_stripe *stripe = multi->stripes;
1532 int i;
1533
1534 if (map_length > num_bytes)
1535 map_length = num_bytes;
1536
1537 for (i = 0; i < multi->num_stripes; i++, stripe++) {
1538 btrfs_issue_discard(stripe->dev->bdev,
1539 stripe->physical,
1540 map_length);
1541 }
1542 kfree(multi);
1543 }
1544
1545 return ret;
1546 #else
1547 return 0;
1548 #endif
1549 }
1550
1551 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1552 struct btrfs_root *root,
1553 u64 bytenr, u64 num_bytes, u64 parent,
1554 u64 root_objectid, u64 owner, u64 offset)
1555 {
1556 int ret;
1557 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1558 root_objectid == BTRFS_TREE_LOG_OBJECTID);
1559
1560 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1561 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
1562 parent, root_objectid, (int)owner,
1563 BTRFS_ADD_DELAYED_REF, NULL);
1564 } else {
1565 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
1566 parent, root_objectid, owner, offset,
1567 BTRFS_ADD_DELAYED_REF, NULL);
1568 }
1569 return ret;
1570 }
1571
1572 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1573 struct btrfs_root *root,
1574 u64 bytenr, u64 num_bytes,
1575 u64 parent, u64 root_objectid,
1576 u64 owner, u64 offset, int refs_to_add,
1577 struct btrfs_delayed_extent_op *extent_op)
1578 {
1579 struct btrfs_path *path;
1580 struct extent_buffer *leaf;
1581 struct btrfs_extent_item *item;
1582 u64 refs;
1583 int ret;
1584 int err = 0;
1585
1586 path = btrfs_alloc_path();
1587 if (!path)
1588 return -ENOMEM;
1589
1590 path->reada = 1;
1591 path->leave_spinning = 1;
1592 /* this will setup the path even if it fails to insert the back ref */
1593 ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1594 path, bytenr, num_bytes, parent,
1595 root_objectid, owner, offset,
1596 refs_to_add, extent_op);
1597 if (ret == 0)
1598 goto out;
1599
1600 if (ret != -EAGAIN) {
1601 err = ret;
1602 goto out;
1603 }
1604
1605 leaf = path->nodes[0];
1606 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1607 refs = btrfs_extent_refs(leaf, item);
1608 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1609 if (extent_op)
1610 __run_delayed_extent_op(extent_op, leaf, item);
1611
1612 btrfs_mark_buffer_dirty(leaf);
1613 btrfs_release_path(root->fs_info->extent_root, path);
1614
1615 path->reada = 1;
1616 path->leave_spinning = 1;
1617
1618 /* now insert the actual backref */
1619 ret = insert_extent_backref(trans, root->fs_info->extent_root,
1620 path, bytenr, parent, root_objectid,
1621 owner, offset, refs_to_add);
1622 BUG_ON(ret);
1623 out:
1624 btrfs_free_path(path);
1625 return err;
1626 }
1627
1628 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1629 struct btrfs_root *root,
1630 struct btrfs_delayed_ref_node *node,
1631 struct btrfs_delayed_extent_op *extent_op,
1632 int insert_reserved)
1633 {
1634 int ret = 0;
1635 struct btrfs_delayed_data_ref *ref;
1636 struct btrfs_key ins;
1637 u64 parent = 0;
1638 u64 ref_root = 0;
1639 u64 flags = 0;
1640
1641 ins.objectid = node->bytenr;
1642 ins.offset = node->num_bytes;
1643 ins.type = BTRFS_EXTENT_ITEM_KEY;
1644
1645 ref = btrfs_delayed_node_to_data_ref(node);
1646 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1647 parent = ref->parent;
1648 else
1649 ref_root = ref->root;
1650
1651 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1652 if (extent_op) {
1653 BUG_ON(extent_op->update_key);
1654 flags |= extent_op->flags_to_set;
1655 }
1656 ret = alloc_reserved_file_extent(trans, root,
1657 parent, ref_root, flags,
1658 ref->objectid, ref->offset,
1659 &ins, node->ref_mod);
1660 update_reserved_extents(root, ins.objectid, ins.offset, 0);
1661 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1662 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1663 node->num_bytes, parent,
1664 ref_root, ref->objectid,
1665 ref->offset, node->ref_mod,
1666 extent_op);
1667 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1668 ret = __btrfs_free_extent(trans, root, node->bytenr,
1669 node->num_bytes, parent,
1670 ref_root, ref->objectid,
1671 ref->offset, node->ref_mod,
1672 extent_op);
1673 } else {
1674 BUG();
1675 }
1676 return ret;
1677 }
1678
1679 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1680 struct extent_buffer *leaf,
1681 struct btrfs_extent_item *ei)
1682 {
1683 u64 flags = btrfs_extent_flags(leaf, ei);
1684 if (extent_op->update_flags) {
1685 flags |= extent_op->flags_to_set;
1686 btrfs_set_extent_flags(leaf, ei, flags);
1687 }
1688
1689 if (extent_op->update_key) {
1690 struct btrfs_tree_block_info *bi;
1691 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
1692 bi = (struct btrfs_tree_block_info *)(ei + 1);
1693 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
1694 }
1695 }
1696
1697 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
1698 struct btrfs_root *root,
1699 struct btrfs_delayed_ref_node *node,
1700 struct btrfs_delayed_extent_op *extent_op)
1701 {
1702 struct btrfs_key key;
1703 struct btrfs_path *path;
1704 struct btrfs_extent_item *ei;
1705 struct extent_buffer *leaf;
1706 u32 item_size;
1707 int ret;
1708 int err = 0;
1709
1710 path = btrfs_alloc_path();
1711 if (!path)
1712 return -ENOMEM;
1713
1714 key.objectid = node->bytenr;
1715 key.type = BTRFS_EXTENT_ITEM_KEY;
1716 key.offset = node->num_bytes;
1717
1718 path->reada = 1;
1719 path->leave_spinning = 1;
1720 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
1721 path, 0, 1);
1722 if (ret < 0) {
1723 err = ret;
1724 goto out;
1725 }
1726 if (ret > 0) {
1727 err = -EIO;
1728 goto out;
1729 }
1730
1731 leaf = path->nodes[0];
1732 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1733 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1734 if (item_size < sizeof(*ei)) {
1735 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
1736 path, (u64)-1, 0);
1737 if (ret < 0) {
1738 err = ret;
1739 goto out;
1740 }
1741 leaf = path->nodes[0];
1742 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1743 }
1744 #endif
1745 BUG_ON(item_size < sizeof(*ei));
1746 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1747 __run_delayed_extent_op(extent_op, leaf, ei);
1748
1749 btrfs_mark_buffer_dirty(leaf);
1750 out:
1751 btrfs_free_path(path);
1752 return err;
1753 }
1754
1755 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
1756 struct btrfs_root *root,
1757 struct btrfs_delayed_ref_node *node,
1758 struct btrfs_delayed_extent_op *extent_op,
1759 int insert_reserved)
1760 {
1761 int ret = 0;
1762 struct btrfs_delayed_tree_ref *ref;
1763 struct btrfs_key ins;
1764 u64 parent = 0;
1765 u64 ref_root = 0;
1766
1767 ins.objectid = node->bytenr;
1768 ins.offset = node->num_bytes;
1769 ins.type = BTRFS_EXTENT_ITEM_KEY;
1770
1771 ref = btrfs_delayed_node_to_tree_ref(node);
1772 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1773 parent = ref->parent;
1774 else
1775 ref_root = ref->root;
1776
1777 BUG_ON(node->ref_mod != 1);
1778 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1779 BUG_ON(!extent_op || !extent_op->update_flags ||
1780 !extent_op->update_key);
1781 ret = alloc_reserved_tree_block(trans, root,
1782 parent, ref_root,
1783 extent_op->flags_to_set,
1784 &extent_op->key,
1785 ref->level, &ins);
1786 update_reserved_extents(root, ins.objectid, ins.offset, 0);
1787 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1788 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1789 node->num_bytes, parent, ref_root,
1790 ref->level, 0, 1, extent_op);
1791 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1792 ret = __btrfs_free_extent(trans, root, node->bytenr,
1793 node->num_bytes, parent, ref_root,
1794 ref->level, 0, 1, extent_op);
1795 } else {
1796 BUG();
1797 }
1798 return ret;
1799 }
1800
1801
1802 /* helper function to actually process a single delayed ref entry */
1803 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
1804 struct btrfs_root *root,
1805 struct btrfs_delayed_ref_node *node,
1806 struct btrfs_delayed_extent_op *extent_op,
1807 int insert_reserved)
1808 {
1809 int ret;
1810 if (btrfs_delayed_ref_is_head(node)) {
1811 struct btrfs_delayed_ref_head *head;
1812 /*
1813 * we've hit the end of the chain and we were supposed
1814 * to insert this extent into the tree. But, it got
1815 * deleted before we ever needed to insert it, so all
1816 * we have to do is clean up the accounting
1817 */
1818 BUG_ON(extent_op);
1819 head = btrfs_delayed_node_to_head(node);
1820 if (insert_reserved) {
1821 if (head->is_data) {
1822 ret = btrfs_del_csums(trans, root,
1823 node->bytenr,
1824 node->num_bytes);
1825 BUG_ON(ret);
1826 }
1827 btrfs_update_pinned_extents(root, node->bytenr,
1828 node->num_bytes, 1);
1829 update_reserved_extents(root, node->bytenr,
1830 node->num_bytes, 0);
1831 }
1832 mutex_unlock(&head->mutex);
1833 return 0;
1834 }
1835
1836 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
1837 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1838 ret = run_delayed_tree_ref(trans, root, node, extent_op,
1839 insert_reserved);
1840 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
1841 node->type == BTRFS_SHARED_DATA_REF_KEY)
1842 ret = run_delayed_data_ref(trans, root, node, extent_op,
1843 insert_reserved);
1844 else
1845 BUG();
1846 return ret;
1847 }
1848
1849 static noinline struct btrfs_delayed_ref_node *
1850 select_delayed_ref(struct btrfs_delayed_ref_head *head)
1851 {
1852 struct rb_node *node;
1853 struct btrfs_delayed_ref_node *ref;
1854 int action = BTRFS_ADD_DELAYED_REF;
1855 again:
1856 /*
1857 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
1858 * this prevents ref count from going down to zero when
1859 * there still are pending delayed ref.
1860 */
1861 node = rb_prev(&head->node.rb_node);
1862 while (1) {
1863 if (!node)
1864 break;
1865 ref = rb_entry(node, struct btrfs_delayed_ref_node,
1866 rb_node);
1867 if (ref->bytenr != head->node.bytenr)
1868 break;
1869 if (ref->action == action)
1870 return ref;
1871 node = rb_prev(node);
1872 }
1873 if (action == BTRFS_ADD_DELAYED_REF) {
1874 action = BTRFS_DROP_DELAYED_REF;
1875 goto again;
1876 }
1877 return NULL;
1878 }
1879
1880 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
1881 struct btrfs_root *root,
1882 struct list_head *cluster)
1883 {
1884 struct btrfs_delayed_ref_root *delayed_refs;
1885 struct btrfs_delayed_ref_node *ref;
1886 struct btrfs_delayed_ref_head *locked_ref = NULL;
1887 struct btrfs_delayed_extent_op *extent_op;
1888 int ret;
1889 int count = 0;
1890 int must_insert_reserved = 0;
1891
1892 delayed_refs = &trans->transaction->delayed_refs;
1893 while (1) {
1894 if (!locked_ref) {
1895 /* pick a new head ref from the cluster list */
1896 if (list_empty(cluster))
1897 break;
1898
1899 locked_ref = list_entry(cluster->next,
1900 struct btrfs_delayed_ref_head, cluster);
1901
1902 /* grab the lock that says we are going to process
1903 * all the refs for this head */
1904 ret = btrfs_delayed_ref_lock(trans, locked_ref);
1905
1906 /*
1907 * we may have dropped the spin lock to get the head
1908 * mutex lock, and that might have given someone else
1909 * time to free the head. If that's true, it has been
1910 * removed from our list and we can move on.
1911 */
1912 if (ret == -EAGAIN) {
1913 locked_ref = NULL;
1914 count++;
1915 continue;
1916 }
1917 }
1918
1919 /*
1920 * record the must insert reserved flag before we
1921 * drop the spin lock.
1922 */
1923 must_insert_reserved = locked_ref->must_insert_reserved;
1924 locked_ref->must_insert_reserved = 0;
1925
1926 extent_op = locked_ref->extent_op;
1927 locked_ref->extent_op = NULL;
1928
1929 /*
1930 * locked_ref is the head node, so we have to go one
1931 * node back for any delayed ref updates
1932 */
1933 ref = select_delayed_ref(locked_ref);
1934 if (!ref) {
1935 /* All delayed refs have been processed, Go ahead
1936 * and send the head node to run_one_delayed_ref,
1937 * so that any accounting fixes can happen
1938 */
1939 ref = &locked_ref->node;
1940
1941 if (extent_op && must_insert_reserved) {
1942 kfree(extent_op);
1943 extent_op = NULL;
1944 }
1945
1946 if (extent_op) {
1947 spin_unlock(&delayed_refs->lock);
1948
1949 ret = run_delayed_extent_op(trans, root,
1950 ref, extent_op);
1951 BUG_ON(ret);
1952 kfree(extent_op);
1953
1954 cond_resched();
1955 spin_lock(&delayed_refs->lock);
1956 continue;
1957 }
1958
1959 list_del_init(&locked_ref->cluster);
1960 locked_ref = NULL;
1961 }
1962
1963 ref->in_tree = 0;
1964 rb_erase(&ref->rb_node, &delayed_refs->root);
1965 delayed_refs->num_entries--;
1966
1967 spin_unlock(&delayed_refs->lock);
1968
1969 ret = run_one_delayed_ref(trans, root, ref, extent_op,
1970 must_insert_reserved);
1971 BUG_ON(ret);
1972
1973 btrfs_put_delayed_ref(ref);
1974 kfree(extent_op);
1975 count++;
1976
1977 cond_resched();
1978 spin_lock(&delayed_refs->lock);
1979 }
1980 return count;
1981 }
1982
1983 /*
1984 * this starts processing the delayed reference count updates and
1985 * extent insertions we have queued up so far. count can be
1986 * 0, which means to process everything in the tree at the start
1987 * of the run (but not newly added entries), or it can be some target
1988 * number you'd like to process.
1989 */
1990 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
1991 struct btrfs_root *root, unsigned long count)
1992 {
1993 struct rb_node *node;
1994 struct btrfs_delayed_ref_root *delayed_refs;
1995 struct btrfs_delayed_ref_node *ref;
1996 struct list_head cluster;
1997 int ret;
1998 int run_all = count == (unsigned long)-1;
1999 int run_most = 0;
2000
2001 if (root == root->fs_info->extent_root)
2002 root = root->fs_info->tree_root;
2003
2004 delayed_refs = &trans->transaction->delayed_refs;
2005 INIT_LIST_HEAD(&cluster);
2006 again:
2007 spin_lock(&delayed_refs->lock);
2008 if (count == 0) {
2009 count = delayed_refs->num_entries * 2;
2010 run_most = 1;
2011 }
2012 while (1) {
2013 if (!(run_all || run_most) &&
2014 delayed_refs->num_heads_ready < 64)
2015 break;
2016
2017 /*
2018 * go find something we can process in the rbtree. We start at
2019 * the beginning of the tree, and then build a cluster
2020 * of refs to process starting at the first one we are able to
2021 * lock
2022 */
2023 ret = btrfs_find_ref_cluster(trans, &cluster,
2024 delayed_refs->run_delayed_start);
2025 if (ret)
2026 break;
2027
2028 ret = run_clustered_refs(trans, root, &cluster);
2029 BUG_ON(ret < 0);
2030
2031 count -= min_t(unsigned long, ret, count);
2032
2033 if (count == 0)
2034 break;
2035 }
2036
2037 if (run_all) {
2038 node = rb_first(&delayed_refs->root);
2039 if (!node)
2040 goto out;
2041 count = (unsigned long)-1;
2042
2043 while (node) {
2044 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2045 rb_node);
2046 if (btrfs_delayed_ref_is_head(ref)) {
2047 struct btrfs_delayed_ref_head *head;
2048
2049 head = btrfs_delayed_node_to_head(ref);
2050 atomic_inc(&ref->refs);
2051
2052 spin_unlock(&delayed_refs->lock);
2053 mutex_lock(&head->mutex);
2054 mutex_unlock(&head->mutex);
2055
2056 btrfs_put_delayed_ref(ref);
2057 cond_resched();
2058 goto again;
2059 }
2060 node = rb_next(node);
2061 }
2062 spin_unlock(&delayed_refs->lock);
2063 schedule_timeout(1);
2064 goto again;
2065 }
2066 out:
2067 spin_unlock(&delayed_refs->lock);
2068 return 0;
2069 }
2070
2071 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2072 struct btrfs_root *root,
2073 u64 bytenr, u64 num_bytes, u64 flags,
2074 int is_data)
2075 {
2076 struct btrfs_delayed_extent_op *extent_op;
2077 int ret;
2078
2079 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2080 if (!extent_op)
2081 return -ENOMEM;
2082
2083 extent_op->flags_to_set = flags;
2084 extent_op->update_flags = 1;
2085 extent_op->update_key = 0;
2086 extent_op->is_data = is_data ? 1 : 0;
2087
2088 ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op);
2089 if (ret)
2090 kfree(extent_op);
2091 return ret;
2092 }
2093
2094 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2095 struct btrfs_root *root,
2096 struct btrfs_path *path,
2097 u64 objectid, u64 offset, u64 bytenr)
2098 {
2099 struct btrfs_delayed_ref_head *head;
2100 struct btrfs_delayed_ref_node *ref;
2101 struct btrfs_delayed_data_ref *data_ref;
2102 struct btrfs_delayed_ref_root *delayed_refs;
2103 struct rb_node *node;
2104 int ret = 0;
2105
2106 ret = -ENOENT;
2107 delayed_refs = &trans->transaction->delayed_refs;
2108 spin_lock(&delayed_refs->lock);
2109 head = btrfs_find_delayed_ref_head(trans, bytenr);
2110 if (!head)
2111 goto out;
2112
2113 if (!mutex_trylock(&head->mutex)) {
2114 atomic_inc(&head->node.refs);
2115 spin_unlock(&delayed_refs->lock);
2116
2117 btrfs_release_path(root->fs_info->extent_root, path);
2118
2119 mutex_lock(&head->mutex);
2120 mutex_unlock(&head->mutex);
2121 btrfs_put_delayed_ref(&head->node);
2122 return -EAGAIN;
2123 }
2124
2125 node = rb_prev(&head->node.rb_node);
2126 if (!node)
2127 goto out_unlock;
2128
2129 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2130
2131 if (ref->bytenr != bytenr)
2132 goto out_unlock;
2133
2134 ret = 1;
2135 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2136 goto out_unlock;
2137
2138 data_ref = btrfs_delayed_node_to_data_ref(ref);
2139
2140 node = rb_prev(node);
2141 if (node) {
2142 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2143 if (ref->bytenr == bytenr)
2144 goto out_unlock;
2145 }
2146
2147 if (data_ref->root != root->root_key.objectid ||
2148 data_ref->objectid != objectid || data_ref->offset != offset)
2149 goto out_unlock;
2150
2151 ret = 0;
2152 out_unlock:
2153 mutex_unlock(&head->mutex);
2154 out:
2155 spin_unlock(&delayed_refs->lock);
2156 return ret;
2157 }
2158
2159 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2160 struct btrfs_root *root,
2161 struct btrfs_path *path,
2162 u64 objectid, u64 offset, u64 bytenr)
2163 {
2164 struct btrfs_root *extent_root = root->fs_info->extent_root;
2165 struct extent_buffer *leaf;
2166 struct btrfs_extent_data_ref *ref;
2167 struct btrfs_extent_inline_ref *iref;
2168 struct btrfs_extent_item *ei;
2169 struct btrfs_key key;
2170 u32 item_size;
2171 int ret;
2172
2173 key.objectid = bytenr;
2174 key.offset = (u64)-1;
2175 key.type = BTRFS_EXTENT_ITEM_KEY;
2176
2177 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2178 if (ret < 0)
2179 goto out;
2180 BUG_ON(ret == 0);
2181
2182 ret = -ENOENT;
2183 if (path->slots[0] == 0)
2184 goto out;
2185
2186 path->slots[0]--;
2187 leaf = path->nodes[0];
2188 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2189
2190 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2191 goto out;
2192
2193 ret = 1;
2194 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2195 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2196 if (item_size < sizeof(*ei)) {
2197 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2198 goto out;
2199 }
2200 #endif
2201 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2202
2203 if (item_size != sizeof(*ei) +
2204 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2205 goto out;
2206
2207 if (btrfs_extent_generation(leaf, ei) <=
2208 btrfs_root_last_snapshot(&root->root_item))
2209 goto out;
2210
2211 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2212 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2213 BTRFS_EXTENT_DATA_REF_KEY)
2214 goto out;
2215
2216 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2217 if (btrfs_extent_refs(leaf, ei) !=
2218 btrfs_extent_data_ref_count(leaf, ref) ||
2219 btrfs_extent_data_ref_root(leaf, ref) !=
2220 root->root_key.objectid ||
2221 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2222 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2223 goto out;
2224
2225 ret = 0;
2226 out:
2227 return ret;
2228 }
2229
2230 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2231 struct btrfs_root *root,
2232 u64 objectid, u64 offset, u64 bytenr)
2233 {
2234 struct btrfs_path *path;
2235 int ret;
2236 int ret2;
2237
2238 path = btrfs_alloc_path();
2239 if (!path)
2240 return -ENOENT;
2241
2242 do {
2243 ret = check_committed_ref(trans, root, path, objectid,
2244 offset, bytenr);
2245 if (ret && ret != -ENOENT)
2246 goto out;
2247
2248 ret2 = check_delayed_ref(trans, root, path, objectid,
2249 offset, bytenr);
2250 } while (ret2 == -EAGAIN);
2251
2252 if (ret2 && ret2 != -ENOENT) {
2253 ret = ret2;
2254 goto out;
2255 }
2256
2257 if (ret != -ENOENT || ret2 != -ENOENT)
2258 ret = 0;
2259 out:
2260 btrfs_free_path(path);
2261 return ret;
2262 }
2263
2264 #if 0
2265 int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2266 struct extent_buffer *buf, u32 nr_extents)
2267 {
2268 struct btrfs_key key;
2269 struct btrfs_file_extent_item *fi;
2270 u64 root_gen;
2271 u32 nritems;
2272 int i;
2273 int level;
2274 int ret = 0;
2275 int shared = 0;
2276
2277 if (!root->ref_cows)
2278 return 0;
2279
2280 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
2281 shared = 0;
2282 root_gen = root->root_key.offset;
2283 } else {
2284 shared = 1;
2285 root_gen = trans->transid - 1;
2286 }
2287
2288 level = btrfs_header_level(buf);
2289 nritems = btrfs_header_nritems(buf);
2290
2291 if (level == 0) {
2292 struct btrfs_leaf_ref *ref;
2293 struct btrfs_extent_info *info;
2294
2295 ref = btrfs_alloc_leaf_ref(root, nr_extents);
2296 if (!ref) {
2297 ret = -ENOMEM;
2298 goto out;
2299 }
2300
2301 ref->root_gen = root_gen;
2302 ref->bytenr = buf->start;
2303 ref->owner = btrfs_header_owner(buf);
2304 ref->generation = btrfs_header_generation(buf);
2305 ref->nritems = nr_extents;
2306 info = ref->extents;
2307
2308 for (i = 0; nr_extents > 0 && i < nritems; i++) {
2309 u64 disk_bytenr;
2310 btrfs_item_key_to_cpu(buf, &key, i);
2311 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2312 continue;
2313 fi = btrfs_item_ptr(buf, i,
2314 struct btrfs_file_extent_item);
2315 if (btrfs_file_extent_type(buf, fi) ==
2316 BTRFS_FILE_EXTENT_INLINE)
2317 continue;
2318 disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2319 if (disk_bytenr == 0)
2320 continue;
2321
2322 info->bytenr = disk_bytenr;
2323 info->num_bytes =
2324 btrfs_file_extent_disk_num_bytes(buf, fi);
2325 info->objectid = key.objectid;
2326 info->offset = key.offset;
2327 info++;
2328 }
2329
2330 ret = btrfs_add_leaf_ref(root, ref, shared);
2331 if (ret == -EEXIST && shared) {
2332 struct btrfs_leaf_ref *old;
2333 old = btrfs_lookup_leaf_ref(root, ref->bytenr);
2334 BUG_ON(!old);
2335 btrfs_remove_leaf_ref(root, old);
2336 btrfs_free_leaf_ref(root, old);
2337 ret = btrfs_add_leaf_ref(root, ref, shared);
2338 }
2339 WARN_ON(ret);
2340 btrfs_free_leaf_ref(root, ref);
2341 }
2342 out:
2343 return ret;
2344 }
2345
2346 /* when a block goes through cow, we update the reference counts of
2347 * everything that block points to. The internal pointers of the block
2348 * can be in just about any order, and it is likely to have clusters of
2349 * things that are close together and clusters of things that are not.
2350 *
2351 * To help reduce the seeks that come with updating all of these reference
2352 * counts, sort them by byte number before actual updates are done.
2353 *
2354 * struct refsort is used to match byte number to slot in the btree block.
2355 * we sort based on the byte number and then use the slot to actually
2356 * find the item.
2357 *
2358 * struct refsort is smaller than strcut btrfs_item and smaller than
2359 * struct btrfs_key_ptr. Since we're currently limited to the page size
2360 * for a btree block, there's no way for a kmalloc of refsorts for a
2361 * single node to be bigger than a page.
2362 */
2363 struct refsort {
2364 u64 bytenr;
2365 u32 slot;
2366 };
2367
2368 /*
2369 * for passing into sort()
2370 */
2371 static int refsort_cmp(const void *a_void, const void *b_void)
2372 {
2373 const struct refsort *a = a_void;
2374 const struct refsort *b = b_void;
2375
2376 if (a->bytenr < b->bytenr)
2377 return -1;
2378 if (a->bytenr > b->bytenr)
2379 return 1;
2380 return 0;
2381 }
2382 #endif
2383
2384 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2385 struct btrfs_root *root,
2386 struct extent_buffer *buf,
2387 int full_backref, int inc)
2388 {
2389 u64 bytenr;
2390 u64 num_bytes;
2391 u64 parent;
2392 u64 ref_root;
2393 u32 nritems;
2394 struct btrfs_key key;
2395 struct btrfs_file_extent_item *fi;
2396 int i;
2397 int level;
2398 int ret = 0;
2399 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2400 u64, u64, u64, u64, u64, u64);
2401
2402 ref_root = btrfs_header_owner(buf);
2403 nritems = btrfs_header_nritems(buf);
2404 level = btrfs_header_level(buf);
2405
2406 if (!root->ref_cows && level == 0)
2407 return 0;
2408
2409 if (inc)
2410 process_func = btrfs_inc_extent_ref;
2411 else
2412 process_func = btrfs_free_extent;
2413
2414 if (full_backref)
2415 parent = buf->start;
2416 else
2417 parent = 0;
2418
2419 for (i = 0; i < nritems; i++) {
2420 if (level == 0) {
2421 btrfs_item_key_to_cpu(buf, &key, i);
2422 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2423 continue;
2424 fi = btrfs_item_ptr(buf, i,
2425 struct btrfs_file_extent_item);
2426 if (btrfs_file_extent_type(buf, fi) ==
2427 BTRFS_FILE_EXTENT_INLINE)
2428 continue;
2429 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2430 if (bytenr == 0)
2431 continue;
2432
2433 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2434 key.offset -= btrfs_file_extent_offset(buf, fi);
2435 ret = process_func(trans, root, bytenr, num_bytes,
2436 parent, ref_root, key.objectid,
2437 key.offset);
2438 if (ret)
2439 goto fail;
2440 } else {
2441 bytenr = btrfs_node_blockptr(buf, i);
2442 num_bytes = btrfs_level_size(root, level - 1);
2443 ret = process_func(trans, root, bytenr, num_bytes,
2444 parent, ref_root, level - 1, 0);
2445 if (ret)
2446 goto fail;
2447 }
2448 }
2449 return 0;
2450 fail:
2451 BUG();
2452 return ret;
2453 }
2454
2455 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2456 struct extent_buffer *buf, int full_backref)
2457 {
2458 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
2459 }
2460
2461 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2462 struct extent_buffer *buf, int full_backref)
2463 {
2464 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
2465 }
2466
2467 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2468 struct btrfs_root *root,
2469 struct btrfs_path *path,
2470 struct btrfs_block_group_cache *cache)
2471 {
2472 int ret;
2473 struct btrfs_root *extent_root = root->fs_info->extent_root;
2474 unsigned long bi;
2475 struct extent_buffer *leaf;
2476
2477 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2478 if (ret < 0)
2479 goto fail;
2480 BUG_ON(ret);
2481
2482 leaf = path->nodes[0];
2483 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2484 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2485 btrfs_mark_buffer_dirty(leaf);
2486 btrfs_release_path(extent_root, path);
2487 fail:
2488 if (ret)
2489 return ret;
2490 return 0;
2491
2492 }
2493
2494 static struct btrfs_block_group_cache *
2495 next_block_group(struct btrfs_root *root,
2496 struct btrfs_block_group_cache *cache)
2497 {
2498 struct rb_node *node;
2499 spin_lock(&root->fs_info->block_group_cache_lock);
2500 node = rb_next(&cache->cache_node);
2501 btrfs_put_block_group(cache);
2502 if (node) {
2503 cache = rb_entry(node, struct btrfs_block_group_cache,
2504 cache_node);
2505 atomic_inc(&cache->count);
2506 } else
2507 cache = NULL;
2508 spin_unlock(&root->fs_info->block_group_cache_lock);
2509 return cache;
2510 }
2511
2512 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
2513 struct btrfs_root *root)
2514 {
2515 struct btrfs_block_group_cache *cache;
2516 int err = 0;
2517 struct btrfs_path *path;
2518 u64 last = 0;
2519
2520 path = btrfs_alloc_path();
2521 if (!path)
2522 return -ENOMEM;
2523
2524 while (1) {
2525 if (last == 0) {
2526 err = btrfs_run_delayed_refs(trans, root,
2527 (unsigned long)-1);
2528 BUG_ON(err);
2529 }
2530
2531 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2532 while (cache) {
2533 if (cache->dirty)
2534 break;
2535 cache = next_block_group(root, cache);
2536 }
2537 if (!cache) {
2538 if (last == 0)
2539 break;
2540 last = 0;
2541 continue;
2542 }
2543
2544 cache->dirty = 0;
2545 last = cache->key.objectid + cache->key.offset;
2546
2547 err = write_one_cache_group(trans, root, path, cache);
2548 BUG_ON(err);
2549 btrfs_put_block_group(cache);
2550 }
2551
2552 btrfs_free_path(path);
2553 return 0;
2554 }
2555
2556 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
2557 {
2558 struct btrfs_block_group_cache *block_group;
2559 int readonly = 0;
2560
2561 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
2562 if (!block_group || block_group->ro)
2563 readonly = 1;
2564 if (block_group)
2565 btrfs_put_block_group(block_group);
2566 return readonly;
2567 }
2568
2569 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
2570 u64 total_bytes, u64 bytes_used,
2571 struct btrfs_space_info **space_info)
2572 {
2573 struct btrfs_space_info *found;
2574
2575 found = __find_space_info(info, flags);
2576 if (found) {
2577 spin_lock(&found->lock);
2578 found->total_bytes += total_bytes;
2579 found->bytes_used += bytes_used;
2580 found->full = 0;
2581 spin_unlock(&found->lock);
2582 *space_info = found;
2583 return 0;
2584 }
2585 found = kzalloc(sizeof(*found), GFP_NOFS);
2586 if (!found)
2587 return -ENOMEM;
2588
2589 INIT_LIST_HEAD(&found->block_groups);
2590 init_rwsem(&found->groups_sem);
2591 spin_lock_init(&found->lock);
2592 found->flags = flags;
2593 found->total_bytes = total_bytes;
2594 found->bytes_used = bytes_used;
2595 found->bytes_pinned = 0;
2596 found->bytes_reserved = 0;
2597 found->bytes_readonly = 0;
2598 found->bytes_delalloc = 0;
2599 found->full = 0;
2600 found->force_alloc = 0;
2601 *space_info = found;
2602 list_add_rcu(&found->list, &info->space_info);
2603 atomic_set(&found->caching_threads, 0);
2604 return 0;
2605 }
2606
2607 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
2608 {
2609 u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
2610 BTRFS_BLOCK_GROUP_RAID1 |
2611 BTRFS_BLOCK_GROUP_RAID10 |
2612 BTRFS_BLOCK_GROUP_DUP);
2613 if (extra_flags) {
2614 if (flags & BTRFS_BLOCK_GROUP_DATA)
2615 fs_info->avail_data_alloc_bits |= extra_flags;
2616 if (flags & BTRFS_BLOCK_GROUP_METADATA)
2617 fs_info->avail_metadata_alloc_bits |= extra_flags;
2618 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
2619 fs_info->avail_system_alloc_bits |= extra_flags;
2620 }
2621 }
2622
2623 static void set_block_group_readonly(struct btrfs_block_group_cache *cache)
2624 {
2625 spin_lock(&cache->space_info->lock);
2626 spin_lock(&cache->lock);
2627 if (!cache->ro) {
2628 cache->space_info->bytes_readonly += cache->key.offset -
2629 btrfs_block_group_used(&cache->item);
2630 cache->ro = 1;
2631 }
2632 spin_unlock(&cache->lock);
2633 spin_unlock(&cache->space_info->lock);
2634 }
2635
2636 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
2637 {
2638 u64 num_devices = root->fs_info->fs_devices->rw_devices;
2639
2640 if (num_devices == 1)
2641 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
2642 if (num_devices < 4)
2643 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
2644
2645 if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
2646 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
2647 BTRFS_BLOCK_GROUP_RAID10))) {
2648 flags &= ~BTRFS_BLOCK_GROUP_DUP;
2649 }
2650
2651 if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
2652 (flags & BTRFS_BLOCK_GROUP_RAID10)) {
2653 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
2654 }
2655
2656 if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
2657 ((flags & BTRFS_BLOCK_GROUP_RAID1) |
2658 (flags & BTRFS_BLOCK_GROUP_RAID10) |
2659 (flags & BTRFS_BLOCK_GROUP_DUP)))
2660 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
2661 return flags;
2662 }
2663
2664 static u64 btrfs_get_alloc_profile(struct btrfs_root *root, u64 data)
2665 {
2666 struct btrfs_fs_info *info = root->fs_info;
2667 u64 alloc_profile;
2668
2669 if (data) {
2670 alloc_profile = info->avail_data_alloc_bits &
2671 info->data_alloc_profile;
2672 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
2673 } else if (root == root->fs_info->chunk_root) {
2674 alloc_profile = info->avail_system_alloc_bits &
2675 info->system_alloc_profile;
2676 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
2677 } else {
2678 alloc_profile = info->avail_metadata_alloc_bits &
2679 info->metadata_alloc_profile;
2680 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
2681 }
2682
2683 return btrfs_reduce_alloc_profile(root, data);
2684 }
2685
2686 void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
2687 {
2688 u64 alloc_target;
2689
2690 alloc_target = btrfs_get_alloc_profile(root, 1);
2691 BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
2692 alloc_target);
2693 }
2694
2695 /*
2696 * for now this just makes sure we have at least 5% of our metadata space free
2697 * for use.
2698 */
2699 int btrfs_check_metadata_free_space(struct btrfs_root *root)
2700 {
2701 struct btrfs_fs_info *info = root->fs_info;
2702 struct btrfs_space_info *meta_sinfo;
2703 u64 alloc_target, thresh;
2704 int committed = 0, ret;
2705
2706 /* get the space info for where the metadata will live */
2707 alloc_target = btrfs_get_alloc_profile(root, 0);
2708 meta_sinfo = __find_space_info(info, alloc_target);
2709
2710 again:
2711 spin_lock(&meta_sinfo->lock);
2712 if (!meta_sinfo->full)
2713 thresh = meta_sinfo->total_bytes * 80;
2714 else
2715 thresh = meta_sinfo->total_bytes * 95;
2716
2717 do_div(thresh, 100);
2718
2719 if (meta_sinfo->bytes_used + meta_sinfo->bytes_reserved +
2720 meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly > thresh) {
2721 struct btrfs_trans_handle *trans;
2722 if (!meta_sinfo->full) {
2723 meta_sinfo->force_alloc = 1;
2724 spin_unlock(&meta_sinfo->lock);
2725
2726 trans = btrfs_start_transaction(root, 1);
2727 if (!trans)
2728 return -ENOMEM;
2729
2730 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2731 2 * 1024 * 1024, alloc_target, 0);
2732 btrfs_end_transaction(trans, root);
2733 goto again;
2734 }
2735 spin_unlock(&meta_sinfo->lock);
2736
2737 if (!committed) {
2738 committed = 1;
2739 trans = btrfs_join_transaction(root, 1);
2740 if (!trans)
2741 return -ENOMEM;
2742 ret = btrfs_commit_transaction(trans, root);
2743 if (ret)
2744 return ret;
2745 goto again;
2746 }
2747 return -ENOSPC;
2748 }
2749 spin_unlock(&meta_sinfo->lock);
2750
2751 return 0;
2752 }
2753
2754 /*
2755 * This will check the space that the inode allocates from to make sure we have
2756 * enough space for bytes.
2757 */
2758 int btrfs_check_data_free_space(struct btrfs_root *root, struct inode *inode,
2759 u64 bytes)
2760 {
2761 struct btrfs_space_info *data_sinfo;
2762 int ret = 0, committed = 0;
2763
2764 /* make sure bytes are sectorsize aligned */
2765 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
2766
2767 data_sinfo = BTRFS_I(inode)->space_info;
2768 again:
2769 /* make sure we have enough space to handle the data first */
2770 spin_lock(&data_sinfo->lock);
2771 if (data_sinfo->total_bytes - data_sinfo->bytes_used -
2772 data_sinfo->bytes_delalloc - data_sinfo->bytes_reserved -
2773 data_sinfo->bytes_pinned - data_sinfo->bytes_readonly -
2774 data_sinfo->bytes_may_use < bytes) {
2775 struct btrfs_trans_handle *trans;
2776
2777 /*
2778 * if we don't have enough free bytes in this space then we need
2779 * to alloc a new chunk.
2780 */
2781 if (!data_sinfo->full) {
2782 u64 alloc_target;
2783
2784 data_sinfo->force_alloc = 1;
2785 spin_unlock(&data_sinfo->lock);
2786
2787 alloc_target = btrfs_get_alloc_profile(root, 1);
2788 trans = btrfs_start_transaction(root, 1);
2789 if (!trans)
2790 return -ENOMEM;
2791
2792 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2793 bytes + 2 * 1024 * 1024,
2794 alloc_target, 0);
2795 btrfs_end_transaction(trans, root);
2796 if (ret)
2797 return ret;
2798 goto again;
2799 }
2800 spin_unlock(&data_sinfo->lock);
2801
2802 /* commit the current transaction and try again */
2803 if (!committed) {
2804 committed = 1;
2805 trans = btrfs_join_transaction(root, 1);
2806 if (!trans)
2807 return -ENOMEM;
2808 ret = btrfs_commit_transaction(trans, root);
2809 if (ret)
2810 return ret;
2811 goto again;
2812 }
2813
2814 printk(KERN_ERR "no space left, need %llu, %llu delalloc bytes"
2815 ", %llu bytes_used, %llu bytes_reserved, "
2816 "%llu bytes_pinned, %llu bytes_readonly, %llu may use "
2817 "%llu total\n", (unsigned long long)bytes,
2818 (unsigned long long)data_sinfo->bytes_delalloc,
2819 (unsigned long long)data_sinfo->bytes_used,
2820 (unsigned long long)data_sinfo->bytes_reserved,
2821 (unsigned long long)data_sinfo->bytes_pinned,
2822 (unsigned long long)data_sinfo->bytes_readonly,
2823 (unsigned long long)data_sinfo->bytes_may_use,
2824 (unsigned long long)data_sinfo->total_bytes);
2825 return -ENOSPC;
2826 }
2827 data_sinfo->bytes_may_use += bytes;
2828 BTRFS_I(inode)->reserved_bytes += bytes;
2829 spin_unlock(&data_sinfo->lock);
2830
2831 return btrfs_check_metadata_free_space(root);
2832 }
2833
2834 /*
2835 * if there was an error for whatever reason after calling
2836 * btrfs_check_data_free_space, call this so we can cleanup the counters.
2837 */
2838 void btrfs_free_reserved_data_space(struct btrfs_root *root,
2839 struct inode *inode, u64 bytes)
2840 {
2841 struct btrfs_space_info *data_sinfo;
2842
2843 /* make sure bytes are sectorsize aligned */
2844 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
2845
2846 data_sinfo = BTRFS_I(inode)->space_info;
2847 spin_lock(&data_sinfo->lock);
2848 data_sinfo->bytes_may_use -= bytes;
2849 BTRFS_I(inode)->reserved_bytes -= bytes;
2850 spin_unlock(&data_sinfo->lock);
2851 }
2852
2853 /* called when we are adding a delalloc extent to the inode's io_tree */
2854 void btrfs_delalloc_reserve_space(struct btrfs_root *root, struct inode *inode,
2855 u64 bytes)
2856 {
2857 struct btrfs_space_info *data_sinfo;
2858
2859 /* get the space info for where this inode will be storing its data */
2860 data_sinfo = BTRFS_I(inode)->space_info;
2861
2862 /* make sure we have enough space to handle the data first */
2863 spin_lock(&data_sinfo->lock);
2864 data_sinfo->bytes_delalloc += bytes;
2865
2866 /*
2867 * we are adding a delalloc extent without calling
2868 * btrfs_check_data_free_space first. This happens on a weird
2869 * writepage condition, but shouldn't hurt our accounting
2870 */
2871 if (unlikely(bytes > BTRFS_I(inode)->reserved_bytes)) {
2872 data_sinfo->bytes_may_use -= BTRFS_I(inode)->reserved_bytes;
2873 BTRFS_I(inode)->reserved_bytes = 0;
2874 } else {
2875 data_sinfo->bytes_may_use -= bytes;
2876 BTRFS_I(inode)->reserved_bytes -= bytes;
2877 }
2878
2879 spin_unlock(&data_sinfo->lock);
2880 }
2881
2882 /* called when we are clearing an delalloc extent from the inode's io_tree */
2883 void btrfs_delalloc_free_space(struct btrfs_root *root, struct inode *inode,
2884 u64 bytes)
2885 {
2886 struct btrfs_space_info *info;
2887
2888 info = BTRFS_I(inode)->space_info;
2889
2890 spin_lock(&info->lock);
2891 info->bytes_delalloc -= bytes;
2892 spin_unlock(&info->lock);
2893 }
2894
2895 static void force_metadata_allocation(struct btrfs_fs_info *info)
2896 {
2897 struct list_head *head = &info->space_info;
2898 struct btrfs_space_info *found;
2899
2900 rcu_read_lock();
2901 list_for_each_entry_rcu(found, head, list) {
2902 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
2903 found->force_alloc = 1;
2904 }
2905 rcu_read_unlock();
2906 }
2907
2908 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
2909 struct btrfs_root *extent_root, u64 alloc_bytes,
2910 u64 flags, int force)
2911 {
2912 struct btrfs_space_info *space_info;
2913 struct btrfs_fs_info *fs_info = extent_root->fs_info;
2914 u64 thresh;
2915 int ret = 0;
2916
2917 mutex_lock(&fs_info->chunk_mutex);
2918
2919 flags = btrfs_reduce_alloc_profile(extent_root, flags);
2920
2921 space_info = __find_space_info(extent_root->fs_info, flags);
2922 if (!space_info) {
2923 ret = update_space_info(extent_root->fs_info, flags,
2924 0, 0, &space_info);
2925 BUG_ON(ret);
2926 }
2927 BUG_ON(!space_info);
2928
2929 spin_lock(&space_info->lock);
2930 if (space_info->force_alloc) {
2931 force = 1;
2932 space_info->force_alloc = 0;
2933 }
2934 if (space_info->full) {
2935 spin_unlock(&space_info->lock);
2936 goto out;
2937 }
2938
2939 thresh = space_info->total_bytes - space_info->bytes_readonly;
2940 thresh = div_factor(thresh, 6);
2941 if (!force &&
2942 (space_info->bytes_used + space_info->bytes_pinned +
2943 space_info->bytes_reserved + alloc_bytes) < thresh) {
2944 spin_unlock(&space_info->lock);
2945 goto out;
2946 }
2947 spin_unlock(&space_info->lock);
2948
2949 /*
2950 * if we're doing a data chunk, go ahead and make sure that
2951 * we keep a reasonable number of metadata chunks allocated in the
2952 * FS as well.
2953 */
2954 if (flags & BTRFS_BLOCK_GROUP_DATA) {
2955 fs_info->data_chunk_allocations++;
2956 if (!(fs_info->data_chunk_allocations %
2957 fs_info->metadata_ratio))
2958 force_metadata_allocation(fs_info);
2959 }
2960
2961 ret = btrfs_alloc_chunk(trans, extent_root, flags);
2962 if (ret)
2963 space_info->full = 1;
2964 out:
2965 mutex_unlock(&extent_root->fs_info->chunk_mutex);
2966 return ret;
2967 }
2968
2969 static int update_block_group(struct btrfs_trans_handle *trans,
2970 struct btrfs_root *root,
2971 u64 bytenr, u64 num_bytes, int alloc,
2972 int mark_free)
2973 {
2974 struct btrfs_block_group_cache *cache;
2975 struct btrfs_fs_info *info = root->fs_info;
2976 u64 total = num_bytes;
2977 u64 old_val;
2978 u64 byte_in_group;
2979
2980 /* block accounting for super block */
2981 spin_lock(&info->delalloc_lock);
2982 old_val = btrfs_super_bytes_used(&info->super_copy);
2983 if (alloc)
2984 old_val += num_bytes;
2985 else
2986 old_val -= num_bytes;
2987 btrfs_set_super_bytes_used(&info->super_copy, old_val);
2988
2989 /* block accounting for root item */
2990 old_val = btrfs_root_used(&root->root_item);
2991 if (alloc)
2992 old_val += num_bytes;
2993 else
2994 old_val -= num_bytes;
2995 btrfs_set_root_used(&root->root_item, old_val);
2996 spin_unlock(&info->delalloc_lock);
2997
2998 while (total) {
2999 cache = btrfs_lookup_block_group(info, bytenr);
3000 if (!cache)
3001 return -1;
3002 byte_in_group = bytenr - cache->key.objectid;
3003 WARN_ON(byte_in_group > cache->key.offset);
3004
3005 spin_lock(&cache->space_info->lock);
3006 spin_lock(&cache->lock);
3007 cache->dirty = 1;
3008 old_val = btrfs_block_group_used(&cache->item);
3009 num_bytes = min(total, cache->key.offset - byte_in_group);
3010 if (alloc) {
3011 old_val += num_bytes;
3012 cache->space_info->bytes_used += num_bytes;
3013 if (cache->ro)
3014 cache->space_info->bytes_readonly -= num_bytes;
3015 btrfs_set_block_group_used(&cache->item, old_val);
3016 spin_unlock(&cache->lock);
3017 spin_unlock(&cache->space_info->lock);
3018 } else {
3019 old_val -= num_bytes;
3020 cache->space_info->bytes_used -= num_bytes;
3021 if (cache->ro)
3022 cache->space_info->bytes_readonly += num_bytes;
3023 btrfs_set_block_group_used(&cache->item, old_val);
3024 spin_unlock(&cache->lock);
3025 spin_unlock(&cache->space_info->lock);
3026 if (mark_free) {
3027 int ret;
3028
3029 ret = btrfs_discard_extent(root, bytenr,
3030 num_bytes);
3031 WARN_ON(ret);
3032
3033 ret = btrfs_add_free_space(cache, bytenr,
3034 num_bytes);
3035 WARN_ON(ret);
3036 }
3037 }
3038 btrfs_put_block_group(cache);
3039 total -= num_bytes;
3040 bytenr += num_bytes;
3041 }
3042 return 0;
3043 }
3044
3045 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
3046 {
3047 struct btrfs_block_group_cache *cache;
3048 u64 bytenr;
3049
3050 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
3051 if (!cache)
3052 return 0;
3053
3054 bytenr = cache->key.objectid;
3055 btrfs_put_block_group(cache);
3056
3057 return bytenr;
3058 }
3059
3060 int btrfs_update_pinned_extents(struct btrfs_root *root,
3061 u64 bytenr, u64 num, int pin)
3062 {
3063 u64 len;
3064 struct btrfs_block_group_cache *cache;
3065 struct btrfs_fs_info *fs_info = root->fs_info;
3066
3067 if (pin)
3068 set_extent_dirty(&fs_info->pinned_extents,
3069 bytenr, bytenr + num - 1, GFP_NOFS);
3070
3071 while (num > 0) {
3072 cache = btrfs_lookup_block_group(fs_info, bytenr);
3073 BUG_ON(!cache);
3074 len = min(num, cache->key.offset -
3075 (bytenr - cache->key.objectid));
3076 if (pin) {
3077 spin_lock(&cache->space_info->lock);
3078 spin_lock(&cache->lock);
3079 cache->pinned += len;
3080 cache->space_info->bytes_pinned += len;
3081 spin_unlock(&cache->lock);
3082 spin_unlock(&cache->space_info->lock);
3083 fs_info->total_pinned += len;
3084 } else {
3085 int unpin = 0;
3086
3087 /*
3088 * in order to not race with the block group caching, we
3089 * only want to unpin the extent if we are cached. If
3090 * we aren't cached, we want to start async caching this
3091 * block group so we can free the extent the next time
3092 * around.
3093 */
3094 spin_lock(&cache->space_info->lock);
3095 spin_lock(&cache->lock);
3096 unpin = (cache->cached == BTRFS_CACHE_FINISHED);
3097 if (likely(unpin)) {
3098 cache->pinned -= len;
3099 cache->space_info->bytes_pinned -= len;
3100 fs_info->total_pinned -= len;
3101 }
3102 spin_unlock(&cache->lock);
3103 spin_unlock(&cache->space_info->lock);
3104
3105 if (likely(unpin))
3106 clear_extent_dirty(&fs_info->pinned_extents,
3107 bytenr, bytenr + len -1,
3108 GFP_NOFS);
3109 else
3110 cache_block_group(cache);
3111
3112 if (unpin)
3113 btrfs_add_free_space(cache, bytenr, len);
3114 }
3115 btrfs_put_block_group(cache);
3116 bytenr += len;
3117 num -= len;
3118 }
3119 return 0;
3120 }
3121
3122 static int update_reserved_extents(struct btrfs_root *root,
3123 u64 bytenr, u64 num, int reserve)
3124 {
3125 u64 len;
3126 struct btrfs_block_group_cache *cache;
3127 struct btrfs_fs_info *fs_info = root->fs_info;
3128
3129 while (num > 0) {
3130 cache = btrfs_lookup_block_group(fs_info, bytenr);
3131 BUG_ON(!cache);
3132 len = min(num, cache->key.offset -
3133 (bytenr - cache->key.objectid));
3134
3135 spin_lock(&cache->space_info->lock);
3136 spin_lock(&cache->lock);
3137 if (reserve) {
3138 cache->reserved += len;
3139 cache->space_info->bytes_reserved += len;
3140 } else {
3141 cache->reserved -= len;
3142 cache->space_info->bytes_reserved -= len;
3143 }
3144 spin_unlock(&cache->lock);
3145 spin_unlock(&cache->space_info->lock);
3146 btrfs_put_block_group(cache);
3147 bytenr += len;
3148 num -= len;
3149 }
3150 return 0;
3151 }
3152
3153 int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
3154 {
3155 u64 last = 0;
3156 u64 start;
3157 u64 end;
3158 struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
3159 int ret;
3160
3161 while (1) {
3162 ret = find_first_extent_bit(pinned_extents, last,
3163 &start, &end, EXTENT_DIRTY);
3164 if (ret)
3165 break;
3166
3167 set_extent_dirty(copy, start, end, GFP_NOFS);
3168 last = end + 1;
3169 }
3170 return 0;
3171 }
3172
3173 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
3174 struct btrfs_root *root,
3175 struct extent_io_tree *unpin)
3176 {
3177 u64 start;
3178 u64 end;
3179 int ret;
3180
3181 while (1) {
3182 ret = find_first_extent_bit(unpin, 0, &start, &end,
3183 EXTENT_DIRTY);
3184 if (ret)
3185 break;
3186
3187 ret = btrfs_discard_extent(root, start, end + 1 - start);
3188
3189 /* unlocks the pinned mutex */
3190 btrfs_update_pinned_extents(root, start, end + 1 - start, 0);
3191 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3192
3193 cond_resched();
3194 }
3195
3196 return ret;
3197 }
3198
3199 static int pin_down_bytes(struct btrfs_trans_handle *trans,
3200 struct btrfs_root *root,
3201 struct btrfs_path *path,
3202 u64 bytenr, u64 num_bytes, int is_data,
3203 struct extent_buffer **must_clean)
3204 {
3205 int err = 0;
3206 struct extent_buffer *buf;
3207
3208 if (is_data)
3209 goto pinit;
3210
3211 buf = btrfs_find_tree_block(root, bytenr, num_bytes);
3212 if (!buf)
3213 goto pinit;
3214
3215 /* we can reuse a block if it hasn't been written
3216 * and it is from this transaction. We can't
3217 * reuse anything from the tree log root because
3218 * it has tiny sub-transactions.
3219 */
3220 if (btrfs_buffer_uptodate(buf, 0) &&
3221 btrfs_try_tree_lock(buf)) {
3222 u64 header_owner = btrfs_header_owner(buf);
3223 u64 header_transid = btrfs_header_generation(buf);
3224 if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
3225 header_transid == trans->transid &&
3226 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
3227 *must_clean = buf;
3228 return 1;
3229 }
3230 btrfs_tree_unlock(buf);
3231 }
3232 free_extent_buffer(buf);
3233 pinit:
3234 btrfs_set_path_blocking(path);
3235 /* unlocks the pinned mutex */
3236 btrfs_update_pinned_extents(root, bytenr, num_bytes, 1);
3237
3238 BUG_ON(err < 0);
3239 return 0;
3240 }
3241
3242
3243 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
3244 struct btrfs_root *root,
3245 u64 bytenr, u64 num_bytes, u64 parent,
3246 u64 root_objectid, u64 owner_objectid,
3247 u64 owner_offset, int refs_to_drop,
3248 struct btrfs_delayed_extent_op *extent_op)
3249 {
3250 struct btrfs_key key;
3251 struct btrfs_path *path;
3252 struct btrfs_fs_info *info = root->fs_info;
3253 struct btrfs_root *extent_root = info->extent_root;
3254 struct extent_buffer *leaf;
3255 struct btrfs_extent_item *ei;
3256 struct btrfs_extent_inline_ref *iref;
3257 int ret;
3258 int is_data;
3259 int extent_slot = 0;
3260 int found_extent = 0;
3261 int num_to_del = 1;
3262 u32 item_size;
3263 u64 refs;
3264
3265 path = btrfs_alloc_path();
3266 if (!path)
3267 return -ENOMEM;
3268
3269 path->reada = 1;
3270 path->leave_spinning = 1;
3271
3272 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
3273 BUG_ON(!is_data && refs_to_drop != 1);
3274
3275 ret = lookup_extent_backref(trans, extent_root, path, &iref,
3276 bytenr, num_bytes, parent,
3277 root_objectid, owner_objectid,
3278 owner_offset);
3279 if (ret == 0) {
3280 extent_slot = path->slots[0];
3281 while (extent_slot >= 0) {
3282 btrfs_item_key_to_cpu(path->nodes[0], &key,
3283 extent_slot);
3284 if (key.objectid != bytenr)
3285 break;
3286 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
3287 key.offset == num_bytes) {
3288 found_extent = 1;
3289 break;
3290 }
3291 if (path->slots[0] - extent_slot > 5)
3292 break;
3293 extent_slot--;
3294 }
3295 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3296 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
3297 if (found_extent && item_size < sizeof(*ei))
3298 found_extent = 0;
3299 #endif
3300 if (!found_extent) {
3301 BUG_ON(iref);
3302 ret = remove_extent_backref(trans, extent_root, path,
3303 NULL, refs_to_drop,
3304 is_data);
3305 BUG_ON(ret);
3306 btrfs_release_path(extent_root, path);
3307 path->leave_spinning = 1;
3308
3309 key.objectid = bytenr;
3310 key.type = BTRFS_EXTENT_ITEM_KEY;
3311 key.offset = num_bytes;
3312
3313 ret = btrfs_search_slot(trans, extent_root,
3314 &key, path, -1, 1);
3315 if (ret) {
3316 printk(KERN_ERR "umm, got %d back from search"
3317 ", was looking for %llu\n", ret,
3318 (unsigned long long)bytenr);
3319 btrfs_print_leaf(extent_root, path->nodes[0]);
3320 }
3321 BUG_ON(ret);
3322 extent_slot = path->slots[0];
3323 }
3324 } else {
3325 btrfs_print_leaf(extent_root, path->nodes[0]);
3326 WARN_ON(1);
3327 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
3328 "parent %llu root %llu owner %llu offset %llu\n",
3329 (unsigned long long)bytenr,
3330 (unsigned long long)parent,
3331 (unsigned long long)root_objectid,
3332 (unsigned long long)owner_objectid,
3333 (unsigned long long)owner_offset);
3334 }
3335
3336 leaf = path->nodes[0];
3337 item_size = btrfs_item_size_nr(leaf, extent_slot);
3338 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3339 if (item_size < sizeof(*ei)) {
3340 BUG_ON(found_extent || extent_slot != path->slots[0]);
3341 ret = convert_extent_item_v0(trans, extent_root, path,
3342 owner_objectid, 0);
3343 BUG_ON(ret < 0);
3344
3345 btrfs_release_path(extent_root, path);
3346 path->leave_spinning = 1;
3347
3348 key.objectid = bytenr;
3349 key.type = BTRFS_EXTENT_ITEM_KEY;
3350 key.offset = num_bytes;
3351
3352 ret = btrfs_search_slot(trans, extent_root, &key, path,
3353 -1, 1);
3354 if (ret) {
3355 printk(KERN_ERR "umm, got %d back from search"
3356 ", was looking for %llu\n", ret,
3357 (unsigned long long)bytenr);
3358 btrfs_print_leaf(extent_root, path->nodes[0]);
3359 }
3360 BUG_ON(ret);
3361 extent_slot = path->slots[0];
3362 leaf = path->nodes[0];
3363 item_size = btrfs_item_size_nr(leaf, extent_slot);
3364 }
3365 #endif
3366 BUG_ON(item_size < sizeof(*ei));
3367 ei = btrfs_item_ptr(leaf, extent_slot,
3368 struct btrfs_extent_item);
3369 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
3370 struct btrfs_tree_block_info *bi;
3371 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
3372 bi = (struct btrfs_tree_block_info *)(ei + 1);
3373 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
3374 }
3375
3376 refs = btrfs_extent_refs(leaf, ei);
3377 BUG_ON(refs < refs_to_drop);
3378 refs -= refs_to_drop;
3379
3380 if (refs > 0) {
3381 if (extent_op)
3382 __run_delayed_extent_op(extent_op, leaf, ei);
3383 /*
3384 * In the case of inline back ref, reference count will
3385 * be updated by remove_extent_backref
3386 */
3387 if (iref) {
3388 BUG_ON(!found_extent);
3389 } else {
3390 btrfs_set_extent_refs(leaf, ei, refs);
3391 btrfs_mark_buffer_dirty(leaf);
3392 }
3393 if (found_extent) {
3394 ret = remove_extent_backref(trans, extent_root, path,
3395 iref, refs_to_drop,
3396 is_data);
3397 BUG_ON(ret);
3398 }
3399 } else {
3400 int mark_free = 0;
3401 struct extent_buffer *must_clean = NULL;
3402
3403 if (found_extent) {
3404 BUG_ON(is_data && refs_to_drop !=
3405 extent_data_ref_count(root, path, iref));
3406 if (iref) {
3407 BUG_ON(path->slots[0] != extent_slot);
3408 } else {
3409 BUG_ON(path->slots[0] != extent_slot + 1);
3410 path->slots[0] = extent_slot;
3411 num_to_del = 2;
3412 }
3413 }
3414
3415 ret = pin_down_bytes(trans, root, path, bytenr,
3416 num_bytes, is_data, &must_clean);
3417 if (ret > 0)
3418 mark_free = 1;
3419 BUG_ON(ret < 0);
3420 /*
3421 * it is going to be very rare for someone to be waiting
3422 * on the block we're freeing. del_items might need to
3423 * schedule, so rather than get fancy, just force it
3424 * to blocking here
3425 */
3426 if (must_clean)
3427 btrfs_set_lock_blocking(must_clean);
3428
3429 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
3430 num_to_del);
3431 BUG_ON(ret);
3432 btrfs_release_path(extent_root, path);
3433
3434 if (must_clean) {
3435 clean_tree_block(NULL, root, must_clean);
3436 btrfs_tree_unlock(must_clean);
3437 free_extent_buffer(must_clean);
3438 }
3439
3440 if (is_data) {
3441 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
3442 BUG_ON(ret);
3443 } else {
3444 invalidate_mapping_pages(info->btree_inode->i_mapping,
3445 bytenr >> PAGE_CACHE_SHIFT,
3446 (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
3447 }
3448
3449 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
3450 mark_free);
3451 BUG_ON(ret);
3452 }
3453 btrfs_free_path(path);
3454 return ret;
3455 }
3456
3457 /*
3458 * when we free an extent, it is possible (and likely) that we free the last
3459 * delayed ref for that extent as well. This searches the delayed ref tree for
3460 * a given extent, and if there are no other delayed refs to be processed, it
3461 * removes it from the tree.
3462 */
3463 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
3464 struct btrfs_root *root, u64 bytenr)
3465 {
3466 struct btrfs_delayed_ref_head *head;
3467 struct btrfs_delayed_ref_root *delayed_refs;
3468 struct btrfs_delayed_ref_node *ref;
3469 struct rb_node *node;
3470 int ret;
3471
3472 delayed_refs = &trans->transaction->delayed_refs;
3473 spin_lock(&delayed_refs->lock);
3474 head = btrfs_find_delayed_ref_head(trans, bytenr);
3475 if (!head)
3476 goto out;
3477
3478 node = rb_prev(&head->node.rb_node);
3479 if (!node)
3480 goto out;
3481
3482 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
3483
3484 /* there are still entries for this ref, we can't drop it */
3485 if (ref->bytenr == bytenr)
3486 goto out;
3487
3488 if (head->extent_op) {
3489 if (!head->must_insert_reserved)
3490 goto out;
3491 kfree(head->extent_op);
3492 head->extent_op = NULL;
3493 }
3494
3495 /*
3496 * waiting for the lock here would deadlock. If someone else has it
3497 * locked they are already in the process of dropping it anyway
3498 */
3499 if (!mutex_trylock(&head->mutex))
3500 goto out;
3501
3502 /*
3503 * at this point we have a head with no other entries. Go
3504 * ahead and process it.
3505 */
3506 head->node.in_tree = 0;
3507 rb_erase(&head->node.rb_node, &delayed_refs->root);
3508
3509 delayed_refs->num_entries--;
3510
3511 /*
3512 * we don't take a ref on the node because we're removing it from the
3513 * tree, so we just steal the ref the tree was holding.
3514 */
3515 delayed_refs->num_heads--;
3516 if (list_empty(&head->cluster))
3517 delayed_refs->num_heads_ready--;
3518
3519 list_del_init(&head->cluster);
3520 spin_unlock(&delayed_refs->lock);
3521
3522 ret = run_one_delayed_ref(trans, root->fs_info->tree_root,
3523 &head->node, head->extent_op,
3524 head->must_insert_reserved);
3525 BUG_ON(ret);
3526 btrfs_put_delayed_ref(&head->node);
3527 return 0;
3528 out:
3529 spin_unlock(&delayed_refs->lock);
3530 return 0;
3531 }
3532
3533 int btrfs_free_extent(struct btrfs_trans_handle *trans,
3534 struct btrfs_root *root,
3535 u64 bytenr, u64 num_bytes, u64 parent,
3536 u64 root_objectid, u64 owner, u64 offset)
3537 {
3538 int ret;
3539
3540 /*
3541 * tree log blocks never actually go into the extent allocation
3542 * tree, just update pinning info and exit early.
3543 */
3544 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
3545 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
3546 /* unlocks the pinned mutex */
3547 btrfs_update_pinned_extents(root, bytenr, num_bytes, 1);
3548 update_reserved_extents(root, bytenr, num_bytes, 0);
3549 ret = 0;
3550 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
3551 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
3552 parent, root_objectid, (int)owner,
3553 BTRFS_DROP_DELAYED_REF, NULL);
3554 BUG_ON(ret);
3555 ret = check_ref_cleanup(trans, root, bytenr);
3556 BUG_ON(ret);
3557 } else {
3558 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
3559 parent, root_objectid, owner,
3560 offset, BTRFS_DROP_DELAYED_REF, NULL);
3561 BUG_ON(ret);
3562 }
3563 return ret;
3564 }
3565
3566 static u64 stripe_align(struct btrfs_root *root, u64 val)
3567 {
3568 u64 mask = ((u64)root->stripesize - 1);
3569 u64 ret = (val + mask) & ~mask;
3570 return ret;
3571 }
3572
3573 /*
3574 * when we wait for progress in the block group caching, its because
3575 * our allocation attempt failed at least once. So, we must sleep
3576 * and let some progress happen before we try again.
3577 *
3578 * This function will sleep at least once waiting for new free space to
3579 * show up, and then it will check the block group free space numbers
3580 * for our min num_bytes. Another option is to have it go ahead
3581 * and look in the rbtree for a free extent of a given size, but this
3582 * is a good start.
3583 */
3584 static noinline int
3585 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
3586 u64 num_bytes)
3587 {
3588 DEFINE_WAIT(wait);
3589
3590 prepare_to_wait(&cache->caching_q, &wait, TASK_UNINTERRUPTIBLE);
3591
3592 if (block_group_cache_done(cache)) {
3593 finish_wait(&cache->caching_q, &wait);
3594 return 0;
3595 }
3596 schedule();
3597 finish_wait(&cache->caching_q, &wait);
3598
3599 wait_event(cache->caching_q, block_group_cache_done(cache) ||
3600 (cache->free_space >= num_bytes));
3601 return 0;
3602 }
3603
3604 enum btrfs_loop_type {
3605 LOOP_CACHED_ONLY = 0,
3606 LOOP_CACHING_NOWAIT = 1,
3607 LOOP_CACHING_WAIT = 2,
3608 LOOP_ALLOC_CHUNK = 3,
3609 LOOP_NO_EMPTY_SIZE = 4,
3610 };
3611
3612 /*
3613 * walks the btree of allocated extents and find a hole of a given size.
3614 * The key ins is changed to record the hole:
3615 * ins->objectid == block start
3616 * ins->flags = BTRFS_EXTENT_ITEM_KEY
3617 * ins->offset == number of blocks
3618 * Any available blocks before search_start are skipped.
3619 */
3620 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
3621 struct btrfs_root *orig_root,
3622 u64 num_bytes, u64 empty_size,
3623 u64 search_start, u64 search_end,
3624 u64 hint_byte, struct btrfs_key *ins,
3625 u64 exclude_start, u64 exclude_nr,
3626 int data)
3627 {
3628 int ret = 0;
3629 struct btrfs_root *root = orig_root->fs_info->extent_root;
3630 struct btrfs_free_cluster *last_ptr = NULL;
3631 struct btrfs_block_group_cache *block_group = NULL;
3632 int empty_cluster = 2 * 1024 * 1024;
3633 int allowed_chunk_alloc = 0;
3634 struct btrfs_space_info *space_info;
3635 int last_ptr_loop = 0;
3636 int loop = 0;
3637 bool found_uncached_bg = false;
3638
3639 WARN_ON(num_bytes < root->sectorsize);
3640 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
3641 ins->objectid = 0;
3642 ins->offset = 0;
3643
3644 space_info = __find_space_info(root->fs_info, data);
3645
3646 if (orig_root->ref_cows || empty_size)
3647 allowed_chunk_alloc = 1;
3648
3649 if (data & BTRFS_BLOCK_GROUP_METADATA) {
3650 last_ptr = &root->fs_info->meta_alloc_cluster;
3651 if (!btrfs_test_opt(root, SSD))
3652 empty_cluster = 64 * 1024;
3653 }
3654
3655 if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD)) {
3656 last_ptr = &root->fs_info->data_alloc_cluster;
3657 }
3658
3659 if (last_ptr) {
3660 spin_lock(&last_ptr->lock);
3661 if (last_ptr->block_group)
3662 hint_byte = last_ptr->window_start;
3663 spin_unlock(&last_ptr->lock);
3664 }
3665
3666 search_start = max(search_start, first_logical_byte(root, 0));
3667 search_start = max(search_start, hint_byte);
3668
3669 if (!last_ptr)
3670 empty_cluster = 0;
3671
3672 if (search_start == hint_byte) {
3673 block_group = btrfs_lookup_block_group(root->fs_info,
3674 search_start);
3675 /*
3676 * we don't want to use the block group if it doesn't match our
3677 * allocation bits, or if its not cached.
3678 */
3679 if (block_group && block_group_bits(block_group, data) &&
3680 block_group_cache_done(block_group)) {
3681 down_read(&space_info->groups_sem);
3682 if (list_empty(&block_group->list) ||
3683 block_group->ro) {
3684 /*
3685 * someone is removing this block group,
3686 * we can't jump into the have_block_group
3687 * target because our list pointers are not
3688 * valid
3689 */
3690 btrfs_put_block_group(block_group);
3691 up_read(&space_info->groups_sem);
3692 } else
3693 goto have_block_group;
3694 } else if (block_group) {
3695 btrfs_put_block_group(block_group);
3696 }
3697 }
3698
3699 search:
3700 down_read(&space_info->groups_sem);
3701 list_for_each_entry(block_group, &space_info->block_groups, list) {
3702 u64 offset;
3703 int cached;
3704
3705 atomic_inc(&block_group->count);
3706 search_start = block_group->key.objectid;
3707
3708 have_block_group:
3709 if (unlikely(block_group->cached == BTRFS_CACHE_NO)) {
3710 /*
3711 * we want to start caching kthreads, but not too many
3712 * right off the bat so we don't overwhelm the system,
3713 * so only start them if there are less than 2 and we're
3714 * in the initial allocation phase.
3715 */
3716 if (loop > LOOP_CACHING_NOWAIT ||
3717 atomic_read(&space_info->caching_threads) < 2) {
3718 ret = cache_block_group(block_group);
3719 BUG_ON(ret);
3720 }
3721 }
3722
3723 cached = block_group_cache_done(block_group);
3724 if (unlikely(!cached)) {
3725 found_uncached_bg = true;
3726
3727 /* if we only want cached bgs, loop */
3728 if (loop == LOOP_CACHED_ONLY)
3729 goto loop;
3730 }
3731
3732 if (unlikely(block_group->ro))
3733 goto loop;
3734
3735 if (last_ptr) {
3736 /*
3737 * the refill lock keeps out other
3738 * people trying to start a new cluster
3739 */
3740 spin_lock(&last_ptr->refill_lock);
3741 if (last_ptr->block_group &&
3742 (last_ptr->block_group->ro ||
3743 !block_group_bits(last_ptr->block_group, data))) {
3744 offset = 0;
3745 goto refill_cluster;
3746 }
3747
3748 offset = btrfs_alloc_from_cluster(block_group, last_ptr,
3749 num_bytes, search_start);
3750 if (offset) {
3751 /* we have a block, we're done */
3752 spin_unlock(&last_ptr->refill_lock);
3753 goto checks;
3754 }
3755
3756 spin_lock(&last_ptr->lock);
3757 /*
3758 * whoops, this cluster doesn't actually point to
3759 * this block group. Get a ref on the block
3760 * group is does point to and try again
3761 */
3762 if (!last_ptr_loop && last_ptr->block_group &&
3763 last_ptr->block_group != block_group) {
3764
3765 btrfs_put_block_group(block_group);
3766 block_group = last_ptr->block_group;
3767 atomic_inc(&block_group->count);
3768 spin_unlock(&last_ptr->lock);
3769 spin_unlock(&last_ptr->refill_lock);
3770
3771 last_ptr_loop = 1;
3772 search_start = block_group->key.objectid;
3773 /*
3774 * we know this block group is properly
3775 * in the list because
3776 * btrfs_remove_block_group, drops the
3777 * cluster before it removes the block
3778 * group from the list
3779 */
3780 goto have_block_group;
3781 }
3782 spin_unlock(&last_ptr->lock);
3783 refill_cluster:
3784 /*
3785 * this cluster didn't work out, free it and
3786 * start over
3787 */
3788 btrfs_return_cluster_to_free_space(NULL, last_ptr);
3789
3790 last_ptr_loop = 0;
3791
3792 /* allocate a cluster in this block group */
3793 ret = btrfs_find_space_cluster(trans, root,
3794 block_group, last_ptr,
3795 offset, num_bytes,
3796 empty_cluster + empty_size);
3797 if (ret == 0) {
3798 /*
3799 * now pull our allocation out of this
3800 * cluster
3801 */
3802 offset = btrfs_alloc_from_cluster(block_group,
3803 last_ptr, num_bytes,
3804 search_start);
3805 if (offset) {
3806 /* we found one, proceed */
3807 spin_unlock(&last_ptr->refill_lock);
3808 goto checks;
3809 }
3810 } else if (!cached && loop > LOOP_CACHING_NOWAIT) {
3811 spin_unlock(&last_ptr->refill_lock);
3812
3813 wait_block_group_cache_progress(block_group,
3814 num_bytes + empty_cluster + empty_size);
3815 goto have_block_group;
3816 }
3817
3818 /*
3819 * at this point we either didn't find a cluster
3820 * or we weren't able to allocate a block from our
3821 * cluster. Free the cluster we've been trying
3822 * to use, and go to the next block group
3823 */
3824 if (loop < LOOP_NO_EMPTY_SIZE) {
3825 btrfs_return_cluster_to_free_space(NULL,
3826 last_ptr);
3827 spin_unlock(&last_ptr->refill_lock);
3828 goto loop;
3829 }
3830 spin_unlock(&last_ptr->refill_lock);
3831 }
3832
3833 offset = btrfs_find_space_for_alloc(block_group, search_start,
3834 num_bytes, empty_size);
3835 if (!offset && (cached || (!cached &&
3836 loop == LOOP_CACHING_NOWAIT))) {
3837 goto loop;
3838 } else if (!offset && (!cached &&
3839 loop > LOOP_CACHING_NOWAIT)) {
3840 wait_block_group_cache_progress(block_group,
3841 num_bytes + empty_size);
3842 goto have_block_group;
3843 }
3844 checks:
3845 search_start = stripe_align(root, offset);
3846 /* move on to the next group */
3847 if (search_start + num_bytes >= search_end) {
3848 btrfs_add_free_space(block_group, offset, num_bytes);
3849 goto loop;
3850 }
3851
3852 /* move on to the next group */
3853 if (search_start + num_bytes >
3854 block_group->key.objectid + block_group->key.offset) {
3855 btrfs_add_free_space(block_group, offset, num_bytes);
3856 goto loop;
3857 }
3858
3859 if (exclude_nr > 0 &&
3860 (search_start + num_bytes > exclude_start &&
3861 search_start < exclude_start + exclude_nr)) {
3862 search_start = exclude_start + exclude_nr;
3863
3864 btrfs_add_free_space(block_group, offset, num_bytes);
3865 /*
3866 * if search_start is still in this block group
3867 * then we just re-search this block group
3868 */
3869 if (search_start >= block_group->key.objectid &&
3870 search_start < (block_group->key.objectid +
3871 block_group->key.offset))
3872 goto have_block_group;
3873 goto loop;
3874 }
3875
3876 ins->objectid = search_start;
3877 ins->offset = num_bytes;
3878
3879 if (offset < search_start)
3880 btrfs_add_free_space(block_group, offset,
3881 search_start - offset);
3882 BUG_ON(offset > search_start);
3883
3884 /* we are all good, lets return */
3885 break;
3886 loop:
3887 btrfs_put_block_group(block_group);
3888 }
3889 up_read(&space_info->groups_sem);
3890
3891 /* LOOP_CACHED_ONLY, only search fully cached block groups
3892 * LOOP_CACHING_NOWAIT, search partially cached block groups, but
3893 * dont wait foR them to finish caching
3894 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
3895 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
3896 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
3897 * again
3898 */
3899 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE &&
3900 (found_uncached_bg || empty_size || empty_cluster ||
3901 allowed_chunk_alloc)) {
3902 if (found_uncached_bg) {
3903 found_uncached_bg = false;
3904 if (loop < LOOP_CACHING_WAIT) {
3905 loop++;
3906 goto search;
3907 }
3908 }
3909
3910 if (loop == LOOP_ALLOC_CHUNK) {
3911 empty_size = 0;
3912 empty_cluster = 0;
3913 }
3914
3915 if (allowed_chunk_alloc) {
3916 ret = do_chunk_alloc(trans, root, num_bytes +
3917 2 * 1024 * 1024, data, 1);
3918 allowed_chunk_alloc = 0;
3919 } else {
3920 space_info->force_alloc = 1;
3921 }
3922
3923 if (loop < LOOP_NO_EMPTY_SIZE) {
3924 loop++;
3925 goto search;
3926 }
3927 ret = -ENOSPC;
3928 } else if (!ins->objectid) {
3929 ret = -ENOSPC;
3930 }
3931
3932 /* we found what we needed */
3933 if (ins->objectid) {
3934 if (!(data & BTRFS_BLOCK_GROUP_DATA))
3935 trans->block_group = block_group->key.objectid;
3936
3937 btrfs_put_block_group(block_group);
3938 ret = 0;
3939 }
3940
3941 return ret;
3942 }
3943
3944 static void dump_space_info(struct btrfs_space_info *info, u64 bytes)
3945 {
3946 struct btrfs_block_group_cache *cache;
3947
3948 printk(KERN_INFO "space_info has %llu free, is %sfull\n",
3949 (unsigned long long)(info->total_bytes - info->bytes_used -
3950 info->bytes_pinned - info->bytes_reserved),
3951 (info->full) ? "" : "not ");
3952 printk(KERN_INFO "space_info total=%llu, pinned=%llu, delalloc=%llu,"
3953 " may_use=%llu, used=%llu\n",
3954 (unsigned long long)info->total_bytes,
3955 (unsigned long long)info->bytes_pinned,
3956 (unsigned long long)info->bytes_delalloc,
3957 (unsigned long long)info->bytes_may_use,
3958 (unsigned long long)info->bytes_used);
3959
3960 down_read(&info->groups_sem);
3961 list_for_each_entry(cache, &info->block_groups, list) {
3962 spin_lock(&cache->lock);
3963 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
3964 "%llu pinned %llu reserved\n",
3965 (unsigned long long)cache->key.objectid,
3966 (unsigned long long)cache->key.offset,
3967 (unsigned long long)btrfs_block_group_used(&cache->item),
3968 (unsigned long long)cache->pinned,
3969 (unsigned long long)cache->reserved);
3970 btrfs_dump_free_space(cache, bytes);
3971 spin_unlock(&cache->lock);
3972 }
3973 up_read(&info->groups_sem);
3974 }
3975
3976 static int __btrfs_reserve_extent(struct btrfs_trans_handle *trans,
3977 struct btrfs_root *root,
3978 u64 num_bytes, u64 min_alloc_size,
3979 u64 empty_size, u64 hint_byte,
3980 u64 search_end, struct btrfs_key *ins,
3981 u64 data)
3982 {
3983 int ret;
3984 u64 search_start = 0;
3985 struct btrfs_fs_info *info = root->fs_info;
3986
3987 data = btrfs_get_alloc_profile(root, data);
3988 again:
3989 /*
3990 * the only place that sets empty_size is btrfs_realloc_node, which
3991 * is not called recursively on allocations
3992 */
3993 if (empty_size || root->ref_cows) {
3994 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
3995 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3996 2 * 1024 * 1024,
3997 BTRFS_BLOCK_GROUP_METADATA |
3998 (info->metadata_alloc_profile &
3999 info->avail_metadata_alloc_bits), 0);
4000 }
4001 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4002 num_bytes + 2 * 1024 * 1024, data, 0);
4003 }
4004
4005 WARN_ON(num_bytes < root->sectorsize);
4006 ret = find_free_extent(trans, root, num_bytes, empty_size,
4007 search_start, search_end, hint_byte, ins,
4008 trans->alloc_exclude_start,
4009 trans->alloc_exclude_nr, data);
4010
4011 if (ret == -ENOSPC && num_bytes > min_alloc_size) {
4012 num_bytes = num_bytes >> 1;
4013 num_bytes = num_bytes & ~(root->sectorsize - 1);
4014 num_bytes = max(num_bytes, min_alloc_size);
4015 do_chunk_alloc(trans, root->fs_info->extent_root,
4016 num_bytes, data, 1);
4017 goto again;
4018 }
4019 if (ret == -ENOSPC) {
4020 struct btrfs_space_info *sinfo;
4021
4022 sinfo = __find_space_info(root->fs_info, data);
4023 printk(KERN_ERR "btrfs allocation failed flags %llu, "
4024 "wanted %llu\n", (unsigned long long)data,
4025 (unsigned long long)num_bytes);
4026 dump_space_info(sinfo, num_bytes);
4027 }
4028
4029 return ret;
4030 }
4031
4032 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
4033 {
4034 struct btrfs_block_group_cache *cache;
4035 int ret = 0;
4036
4037 cache = btrfs_lookup_block_group(root->fs_info, start);
4038 if (!cache) {
4039 printk(KERN_ERR "Unable to find block group for %llu\n",
4040 (unsigned long long)start);
4041 return -ENOSPC;
4042 }
4043
4044 ret = btrfs_discard_extent(root, start, len);
4045
4046 btrfs_add_free_space(cache, start, len);
4047 btrfs_put_block_group(cache);
4048 update_reserved_extents(root, start, len, 0);
4049
4050 return ret;
4051 }
4052
4053 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
4054 struct btrfs_root *root,
4055 u64 num_bytes, u64 min_alloc_size,
4056 u64 empty_size, u64 hint_byte,
4057 u64 search_end, struct btrfs_key *ins,
4058 u64 data)
4059 {
4060 int ret;
4061 ret = __btrfs_reserve_extent(trans, root, num_bytes, min_alloc_size,
4062 empty_size, hint_byte, search_end, ins,
4063 data);
4064 if (!ret)
4065 update_reserved_extents(root, ins->objectid, ins->offset, 1);
4066
4067 return ret;
4068 }
4069
4070 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
4071 struct btrfs_root *root,
4072 u64 parent, u64 root_objectid,
4073 u64 flags, u64 owner, u64 offset,
4074 struct btrfs_key *ins, int ref_mod)
4075 {
4076 int ret;
4077 struct btrfs_fs_info *fs_info = root->fs_info;
4078 struct btrfs_extent_item *extent_item;
4079 struct btrfs_extent_inline_ref *iref;
4080 struct btrfs_path *path;
4081 struct extent_buffer *leaf;
4082 int type;
4083 u32 size;
4084
4085 if (parent > 0)
4086 type = BTRFS_SHARED_DATA_REF_KEY;
4087 else
4088 type = BTRFS_EXTENT_DATA_REF_KEY;
4089
4090 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
4091
4092 path = btrfs_alloc_path();
4093 BUG_ON(!path);
4094
4095 path->leave_spinning = 1;
4096 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
4097 ins, size);
4098 BUG_ON(ret);
4099
4100 leaf = path->nodes[0];
4101 extent_item = btrfs_item_ptr(leaf, path->slots[0],
4102 struct btrfs_extent_item);
4103 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
4104 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
4105 btrfs_set_extent_flags(leaf, extent_item,
4106 flags | BTRFS_EXTENT_FLAG_DATA);
4107
4108 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
4109 btrfs_set_extent_inline_ref_type(leaf, iref, type);
4110 if (parent > 0) {
4111 struct btrfs_shared_data_ref *ref;
4112 ref = (struct btrfs_shared_data_ref *)(iref + 1);
4113 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
4114 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
4115 } else {
4116 struct btrfs_extent_data_ref *ref;
4117 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
4118 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
4119 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
4120 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
4121 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
4122 }
4123
4124 btrfs_mark_buffer_dirty(path->nodes[0]);
4125 btrfs_free_path(path);
4126
4127 ret = update_block_group(trans, root, ins->objectid, ins->offset,
4128 1, 0);
4129 if (ret) {
4130 printk(KERN_ERR "btrfs update block group failed for %llu "
4131 "%llu\n", (unsigned long long)ins->objectid,
4132 (unsigned long long)ins->offset);
4133 BUG();
4134 }
4135 return ret;
4136 }
4137
4138 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
4139 struct btrfs_root *root,
4140 u64 parent, u64 root_objectid,
4141 u64 flags, struct btrfs_disk_key *key,
4142 int level, struct btrfs_key *ins)
4143 {
4144 int ret;
4145 struct btrfs_fs_info *fs_info = root->fs_info;
4146 struct btrfs_extent_item *extent_item;
4147 struct btrfs_tree_block_info *block_info;
4148 struct btrfs_extent_inline_ref *iref;
4149 struct btrfs_path *path;
4150 struct extent_buffer *leaf;
4151 u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
4152
4153 path = btrfs_alloc_path();
4154 BUG_ON(!path);
4155
4156 path->leave_spinning = 1;
4157 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
4158 ins, size);
4159 BUG_ON(ret);
4160
4161 leaf = path->nodes[0];
4162 extent_item = btrfs_item_ptr(leaf, path->slots[0],
4163 struct btrfs_extent_item);
4164 btrfs_set_extent_refs(leaf, extent_item, 1);
4165 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
4166 btrfs_set_extent_flags(leaf, extent_item,
4167 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
4168 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
4169
4170 btrfs_set_tree_block_key(leaf, block_info, key);
4171 btrfs_set_tree_block_level(leaf, block_info, level);
4172
4173 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
4174 if (parent > 0) {
4175 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
4176 btrfs_set_extent_inline_ref_type(leaf, iref,
4177 BTRFS_SHARED_BLOCK_REF_KEY);
4178 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
4179 } else {
4180 btrfs_set_extent_inline_ref_type(leaf, iref,
4181 BTRFS_TREE_BLOCK_REF_KEY);
4182 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
4183 }
4184
4185 btrfs_mark_buffer_dirty(leaf);
4186 btrfs_free_path(path);
4187
4188 ret = update_block_group(trans, root, ins->objectid, ins->offset,
4189 1, 0);
4190 if (ret) {
4191 printk(KERN_ERR "btrfs update block group failed for %llu "
4192 "%llu\n", (unsigned long long)ins->objectid,
4193 (unsigned long long)ins->offset);
4194 BUG();
4195 }
4196 return ret;
4197 }
4198
4199 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
4200 struct btrfs_root *root,
4201 u64 root_objectid, u64 owner,
4202 u64 offset, struct btrfs_key *ins)
4203 {
4204 int ret;
4205
4206 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
4207
4208 ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset,
4209 0, root_objectid, owner, offset,
4210 BTRFS_ADD_DELAYED_EXTENT, NULL);
4211 return ret;
4212 }
4213
4214 /*
4215 * this is used by the tree logging recovery code. It records that
4216 * an extent has been allocated and makes sure to clear the free
4217 * space cache bits as well
4218 */
4219 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
4220 struct btrfs_root *root,
4221 u64 root_objectid, u64 owner, u64 offset,
4222 struct btrfs_key *ins)
4223 {
4224 int ret;
4225 struct btrfs_block_group_cache *block_group;
4226
4227 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
4228 cache_block_group(block_group);
4229 wait_event(block_group->caching_q,
4230 block_group_cache_done(block_group));
4231
4232 ret = btrfs_remove_free_space(block_group, ins->objectid,
4233 ins->offset);
4234 BUG_ON(ret);
4235 btrfs_put_block_group(block_group);
4236 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
4237 0, owner, offset, ins, 1);
4238 return ret;
4239 }
4240
4241 /*
4242 * finds a free extent and does all the dirty work required for allocation
4243 * returns the key for the extent through ins, and a tree buffer for
4244 * the first block of the extent through buf.
4245 *
4246 * returns 0 if everything worked, non-zero otherwise.
4247 */
4248 static int alloc_tree_block(struct btrfs_trans_handle *trans,
4249 struct btrfs_root *root,
4250 u64 num_bytes, u64 parent, u64 root_objectid,
4251 struct btrfs_disk_key *key, int level,
4252 u64 empty_size, u64 hint_byte, u64 search_end,
4253 struct btrfs_key *ins)
4254 {
4255 int ret;
4256 u64 flags = 0;
4257
4258 ret = __btrfs_reserve_extent(trans, root, num_bytes, num_bytes,
4259 empty_size, hint_byte, search_end,
4260 ins, 0);
4261 if (ret)
4262 return ret;
4263
4264 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
4265 if (parent == 0)
4266 parent = ins->objectid;
4267 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
4268 } else
4269 BUG_ON(parent > 0);
4270
4271 update_reserved_extents(root, ins->objectid, ins->offset, 1);
4272 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
4273 struct btrfs_delayed_extent_op *extent_op;
4274 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
4275 BUG_ON(!extent_op);
4276 if (key)
4277 memcpy(&extent_op->key, key, sizeof(extent_op->key));
4278 else
4279 memset(&extent_op->key, 0, sizeof(extent_op->key));
4280 extent_op->flags_to_set = flags;
4281 extent_op->update_key = 1;
4282 extent_op->update_flags = 1;
4283 extent_op->is_data = 0;
4284
4285 ret = btrfs_add_delayed_tree_ref(trans, ins->objectid,
4286 ins->offset, parent, root_objectid,
4287 level, BTRFS_ADD_DELAYED_EXTENT,
4288 extent_op);
4289 BUG_ON(ret);
4290 }
4291 return ret;
4292 }
4293
4294 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
4295 struct btrfs_root *root,
4296 u64 bytenr, u32 blocksize,
4297 int level)
4298 {
4299 struct extent_buffer *buf;
4300
4301 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
4302 if (!buf)
4303 return ERR_PTR(-ENOMEM);
4304 btrfs_set_header_generation(buf, trans->transid);
4305 btrfs_set_buffer_lockdep_class(buf, level);
4306 btrfs_tree_lock(buf);
4307 clean_tree_block(trans, root, buf);
4308
4309 btrfs_set_lock_blocking(buf);
4310 btrfs_set_buffer_uptodate(buf);
4311
4312 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
4313 set_extent_dirty(&root->dirty_log_pages, buf->start,
4314 buf->start + buf->len - 1, GFP_NOFS);
4315 } else {
4316 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
4317 buf->start + buf->len - 1, GFP_NOFS);
4318 }
4319 trans->blocks_used++;
4320 /* this returns a buffer locked for blocking */
4321 return buf;
4322 }
4323
4324 /*
4325 * helper function to allocate a block for a given tree
4326 * returns the tree buffer or NULL.
4327 */
4328 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
4329 struct btrfs_root *root, u32 blocksize,
4330 u64 parent, u64 root_objectid,
4331 struct btrfs_disk_key *key, int level,
4332 u64 hint, u64 empty_size)
4333 {
4334 struct btrfs_key ins;
4335 int ret;
4336 struct extent_buffer *buf;
4337
4338 ret = alloc_tree_block(trans, root, blocksize, parent, root_objectid,
4339 key, level, empty_size, hint, (u64)-1, &ins);
4340 if (ret) {
4341 BUG_ON(ret > 0);
4342 return ERR_PTR(ret);
4343 }
4344
4345 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
4346 blocksize, level);
4347 return buf;
4348 }
4349
4350 #if 0
4351 int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans,
4352 struct btrfs_root *root, struct extent_buffer *leaf)
4353 {
4354 u64 disk_bytenr;
4355 u64 num_bytes;
4356 struct btrfs_key key;
4357 struct btrfs_file_extent_item *fi;
4358 u32 nritems;
4359 int i;
4360 int ret;
4361
4362 BUG_ON(!btrfs_is_leaf(leaf));
4363 nritems = btrfs_header_nritems(leaf);
4364
4365 for (i = 0; i < nritems; i++) {
4366 cond_resched();
4367 btrfs_item_key_to_cpu(leaf, &key, i);
4368
4369 /* only extents have references, skip everything else */
4370 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
4371 continue;
4372
4373 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4374
4375 /* inline extents live in the btree, they don't have refs */
4376 if (btrfs_file_extent_type(leaf, fi) ==
4377 BTRFS_FILE_EXTENT_INLINE)
4378 continue;
4379
4380 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
4381
4382 /* holes don't have refs */
4383 if (disk_bytenr == 0)
4384 continue;
4385
4386 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
4387 ret = btrfs_free_extent(trans, root, disk_bytenr, num_bytes,
4388 leaf->start, 0, key.objectid, 0);
4389 BUG_ON(ret);
4390 }
4391 return 0;
4392 }
4393
4394 static noinline int cache_drop_leaf_ref(struct btrfs_trans_handle *trans,
4395 struct btrfs_root *root,
4396 struct btrfs_leaf_ref *ref)
4397 {
4398 int i;
4399 int ret;
4400 struct btrfs_extent_info *info;
4401 struct refsort *sorted;
4402
4403 if (ref->nritems == 0)
4404 return 0;
4405
4406 sorted = kmalloc(sizeof(*sorted) * ref->nritems, GFP_NOFS);
4407 for (i = 0; i < ref->nritems; i++) {
4408 sorted[i].bytenr = ref->extents[i].bytenr;
4409 sorted[i].slot = i;
4410 }
4411 sort(sorted, ref->nritems, sizeof(struct refsort), refsort_cmp, NULL);
4412
4413 /*
4414 * the items in the ref were sorted when the ref was inserted
4415 * into the ref cache, so this is already in order
4416 */
4417 for (i = 0; i < ref->nritems; i++) {
4418 info = ref->extents + sorted[i].slot;
4419 ret = btrfs_free_extent(trans, root, info->bytenr,
4420 info->num_bytes, ref->bytenr,
4421 ref->owner, ref->generation,
4422 info->objectid, 0);
4423
4424 atomic_inc(&root->fs_info->throttle_gen);
4425 wake_up(&root->fs_info->transaction_throttle);
4426 cond_resched();
4427
4428 BUG_ON(ret);
4429 info++;
4430 }
4431
4432 kfree(sorted);
4433 return 0;
4434 }
4435
4436
4437 static int drop_snap_lookup_refcount(struct btrfs_trans_handle *trans,
4438 struct btrfs_root *root, u64 start,
4439 u64 len, u32 *refs)
4440 {
4441 int ret;
4442
4443 ret = btrfs_lookup_extent_refs(trans, root, start, len, refs);
4444 BUG_ON(ret);
4445
4446 #if 0 /* some debugging code in case we see problems here */
4447 /* if the refs count is one, it won't get increased again. But
4448 * if the ref count is > 1, someone may be decreasing it at
4449 * the same time we are.
4450 */
4451 if (*refs != 1) {
4452 struct extent_buffer *eb = NULL;
4453 eb = btrfs_find_create_tree_block(root, start, len);
4454 if (eb)
4455 btrfs_tree_lock(eb);
4456
4457 mutex_lock(&root->fs_info->alloc_mutex);
4458 ret = lookup_extent_ref(NULL, root, start, len, refs);
4459 BUG_ON(ret);
4460 mutex_unlock(&root->fs_info->alloc_mutex);
4461
4462 if (eb) {
4463 btrfs_tree_unlock(eb);
4464 free_extent_buffer(eb);
4465 }
4466 if (*refs == 1) {
4467 printk(KERN_ERR "btrfs block %llu went down to one "
4468 "during drop_snap\n", (unsigned long long)start);
4469 }
4470
4471 }
4472 #endif
4473
4474 cond_resched();
4475 return ret;
4476 }
4477
4478
4479 /*
4480 * this is used while deleting old snapshots, and it drops the refs
4481 * on a whole subtree starting from a level 1 node.
4482 *
4483 * The idea is to sort all the leaf pointers, and then drop the
4484 * ref on all the leaves in order. Most of the time the leaves
4485 * will have ref cache entries, so no leaf IOs will be required to
4486 * find the extents they have references on.
4487 *
4488 * For each leaf, any references it has are also dropped in order
4489 *
4490 * This ends up dropping the references in something close to optimal
4491 * order for reading and modifying the extent allocation tree.
4492 */
4493 static noinline int drop_level_one_refs(struct btrfs_trans_handle *trans,
4494 struct btrfs_root *root,
4495 struct btrfs_path *path)
4496 {
4497 u64 bytenr;
4498 u64 root_owner;
4499 u64 root_gen;
4500 struct extent_buffer *eb = path->nodes[1];
4501 struct extent_buffer *leaf;
4502 struct btrfs_leaf_ref *ref;
4503 struct refsort *sorted = NULL;
4504 int nritems = btrfs_header_nritems(eb);
4505 int ret;
4506 int i;
4507 int refi = 0;
4508 int slot = path->slots[1];
4509 u32 blocksize = btrfs_level_size(root, 0);
4510 u32 refs;
4511
4512 if (nritems == 0)
4513 goto out;
4514
4515 root_owner = btrfs_header_owner(eb);
4516 root_gen = btrfs_header_generation(eb);
4517 sorted = kmalloc(sizeof(*sorted) * nritems, GFP_NOFS);
4518
4519 /*
4520 * step one, sort all the leaf pointers so we don't scribble
4521 * randomly into the extent allocation tree
4522 */
4523 for (i = slot; i < nritems; i++) {
4524 sorted[refi].bytenr = btrfs_node_blockptr(eb, i);
4525 sorted[refi].slot = i;
4526 refi++;
4527 }
4528
4529 /*
4530 * nritems won't be zero, but if we're picking up drop_snapshot
4531 * after a crash, slot might be > 0, so double check things
4532 * just in case.
4533 */
4534 if (refi == 0)
4535 goto out;
4536
4537 sort(sorted, refi, sizeof(struct refsort), refsort_cmp, NULL);
4538
4539 /*
4540 * the first loop frees everything the leaves point to
4541 */
4542 for (i = 0; i < refi; i++) {
4543 u64 ptr_gen;
4544
4545 bytenr = sorted[i].bytenr;
4546
4547 /*
4548 * check the reference count on this leaf. If it is > 1
4549 * we just decrement it below and don't update any
4550 * of the refs the leaf points to.
4551 */
4552 ret = drop_snap_lookup_refcount(trans, root, bytenr,
4553 blocksize, &refs);
4554 BUG_ON(ret);
4555 if (refs != 1)
4556 continue;
4557
4558 ptr_gen = btrfs_node_ptr_generation(eb, sorted[i].slot);
4559
4560 /*
4561 * the leaf only had one reference, which means the
4562 * only thing pointing to this leaf is the snapshot
4563 * we're deleting. It isn't possible for the reference
4564 * count to increase again later
4565 *
4566 * The reference cache is checked for the leaf,
4567 * and if found we'll be able to drop any refs held by
4568 * the leaf without needing to read it in.
4569 */
4570 ref = btrfs_lookup_leaf_ref(root, bytenr);
4571 if (ref && ref->generation != ptr_gen) {
4572 btrfs_free_leaf_ref(root, ref);
4573 ref = NULL;
4574 }
4575 if (ref) {
4576 ret = cache_drop_leaf_ref(trans, root, ref);
4577 BUG_ON(ret);
4578 btrfs_remove_leaf_ref(root, ref);
4579 btrfs_free_leaf_ref(root, ref);
4580 } else {
4581 /*
4582 * the leaf wasn't in the reference cache, so
4583 * we have to read it.
4584 */
4585 leaf = read_tree_block(root, bytenr, blocksize,
4586 ptr_gen);
4587 ret = btrfs_drop_leaf_ref(trans, root, leaf);
4588 BUG_ON(ret);
4589 free_extent_buffer(leaf);
4590 }
4591 atomic_inc(&root->fs_info->throttle_gen);
4592 wake_up(&root->fs_info->transaction_throttle);
4593 cond_resched();
4594 }
4595
4596 /*
4597 * run through the loop again to free the refs on the leaves.
4598 * This is faster than doing it in the loop above because
4599 * the leaves are likely to be clustered together. We end up
4600 * working in nice chunks on the extent allocation tree.
4601 */
4602 for (i = 0; i < refi; i++) {
4603 bytenr = sorted[i].bytenr;
4604 ret = btrfs_free_extent(trans, root, bytenr,
4605 blocksize, eb->start,
4606 root_owner, root_gen, 0, 1);
4607 BUG_ON(ret);
4608
4609 atomic_inc(&root->fs_info->throttle_gen);
4610 wake_up(&root->fs_info->transaction_throttle);
4611 cond_resched();
4612 }
4613 out:
4614 kfree(sorted);
4615
4616 /*
4617 * update the path to show we've processed the entire level 1
4618 * node. This will get saved into the root's drop_snapshot_progress
4619 * field so these drops are not repeated again if this transaction
4620 * commits.
4621 */
4622 path->slots[1] = nritems;
4623 return 0;
4624 }
4625
4626 /*
4627 * helper function for drop_snapshot, this walks down the tree dropping ref
4628 * counts as it goes.
4629 */
4630 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
4631 struct btrfs_root *root,
4632 struct btrfs_path *path, int *level)
4633 {
4634 u64 root_owner;
4635 u64 root_gen;
4636 u64 bytenr;
4637 u64 ptr_gen;
4638 struct extent_buffer *next;
4639 struct extent_buffer *cur;
4640 struct extent_buffer *parent;
4641 u32 blocksize;
4642 int ret;
4643 u32 refs;
4644
4645 WARN_ON(*level < 0);
4646 WARN_ON(*level >= BTRFS_MAX_LEVEL);
4647 ret = drop_snap_lookup_refcount(trans, root, path->nodes[*level]->start,
4648 path->nodes[*level]->len, &refs);
4649 BUG_ON(ret);
4650 if (refs > 1)
4651 goto out;
4652
4653 /*
4654 * walk down to the last node level and free all the leaves
4655 */
4656 while (*level >= 0) {
4657 WARN_ON(*level < 0);
4658 WARN_ON(*level >= BTRFS_MAX_LEVEL);
4659 cur = path->nodes[*level];
4660
4661 if (btrfs_header_level(cur) != *level)
4662 WARN_ON(1);
4663
4664 if (path->slots[*level] >=
4665 btrfs_header_nritems(cur))
4666 break;
4667
4668 /* the new code goes down to level 1 and does all the
4669 * leaves pointed to that node in bulk. So, this check
4670 * for level 0 will always be false.
4671 *
4672 * But, the disk format allows the drop_snapshot_progress
4673 * field in the root to leave things in a state where
4674 * a leaf will need cleaning up here. If someone crashes
4675 * with the old code and then boots with the new code,
4676 * we might find a leaf here.
4677 */
4678 if (*level == 0) {
4679 ret = btrfs_drop_leaf_ref(trans, root, cur);
4680 BUG_ON(ret);
4681 break;
4682 }
4683
4684 /*
4685 * once we get to level one, process the whole node
4686 * at once, including everything below it.
4687 */
4688 if (*level == 1) {
4689 ret = drop_level_one_refs(trans, root, path);
4690 BUG_ON(ret);
4691 break;
4692 }
4693
4694 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
4695 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
4696 blocksize = btrfs_level_size(root, *level - 1);
4697
4698 ret = drop_snap_lookup_refcount(trans, root, bytenr,
4699 blocksize, &refs);
4700 BUG_ON(ret);
4701
4702 /*
4703 * if there is more than one reference, we don't need
4704 * to read that node to drop any references it has. We
4705 * just drop the ref we hold on that node and move on to the
4706 * next slot in this level.
4707 */
4708 if (refs != 1) {
4709 parent = path->nodes[*level];
4710 root_owner = btrfs_header_owner(parent);
4711 root_gen = btrfs_header_generation(parent);
4712 path->slots[*level]++;
4713
4714 ret = btrfs_free_extent(trans, root, bytenr,
4715 blocksize, parent->start,
4716 root_owner, root_gen,
4717 *level - 1, 1);
4718 BUG_ON(ret);
4719
4720 atomic_inc(&root->fs_info->throttle_gen);
4721 wake_up(&root->fs_info->transaction_throttle);
4722 cond_resched();
4723
4724 continue;
4725 }
4726
4727 /*
4728 * we need to keep freeing things in the next level down.
4729 * read the block and loop around to process it
4730 */
4731 next = read_tree_block(root, bytenr, blocksize, ptr_gen);
4732 WARN_ON(*level <= 0);
4733 if (path->nodes[*level-1])
4734 free_extent_buffer(path->nodes[*level-1]);
4735 path->nodes[*level-1] = next;
4736 *level = btrfs_header_level(next);
4737 path->slots[*level] = 0;
4738 cond_resched();
4739 }
4740 out:
4741 WARN_ON(*level < 0);
4742 WARN_ON(*level >= BTRFS_MAX_LEVEL);
4743
4744 if (path->nodes[*level] == root->node) {
4745 parent = path->nodes[*level];
4746 bytenr = path->nodes[*level]->start;
4747 } else {
4748 parent = path->nodes[*level + 1];
4749 bytenr = btrfs_node_blockptr(parent, path->slots[*level + 1]);
4750 }
4751
4752 blocksize = btrfs_level_size(root, *level);
4753 root_owner = btrfs_header_owner(parent);
4754 root_gen = btrfs_header_generation(parent);
4755
4756 /*
4757 * cleanup and free the reference on the last node
4758 * we processed
4759 */
4760 ret = btrfs_free_extent(trans, root, bytenr, blocksize,
4761 parent->start, root_owner, root_gen,
4762 *level, 1);
4763 free_extent_buffer(path->nodes[*level]);
4764 path->nodes[*level] = NULL;
4765
4766 *level += 1;
4767 BUG_ON(ret);
4768
4769 cond_resched();
4770 return 0;
4771 }
4772 #endif
4773
4774 struct walk_control {
4775 u64 refs[BTRFS_MAX_LEVEL];
4776 u64 flags[BTRFS_MAX_LEVEL];
4777 struct btrfs_key update_progress;
4778 int stage;
4779 int level;
4780 int shared_level;
4781 int update_ref;
4782 int keep_locks;
4783 };
4784
4785 #define DROP_REFERENCE 1
4786 #define UPDATE_BACKREF 2
4787
4788 /*
4789 * hepler to process tree block while walking down the tree.
4790 *
4791 * when wc->stage == DROP_REFERENCE, this function checks
4792 * reference count of the block. if the block is shared and
4793 * we need update back refs for the subtree rooted at the
4794 * block, this function changes wc->stage to UPDATE_BACKREF
4795 *
4796 * when wc->stage == UPDATE_BACKREF, this function updates
4797 * back refs for pointers in the block.
4798 *
4799 * NOTE: return value 1 means we should stop walking down.
4800 */
4801 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
4802 struct btrfs_root *root,
4803 struct btrfs_path *path,
4804 struct walk_control *wc)
4805 {
4806 int level = wc->level;
4807 struct extent_buffer *eb = path->nodes[level];
4808 struct btrfs_key key;
4809 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
4810 int ret;
4811
4812 if (wc->stage == UPDATE_BACKREF &&
4813 btrfs_header_owner(eb) != root->root_key.objectid)
4814 return 1;
4815
4816 /*
4817 * when reference count of tree block is 1, it won't increase
4818 * again. once full backref flag is set, we never clear it.
4819 */
4820 if ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
4821 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag))) {
4822 BUG_ON(!path->locks[level]);
4823 ret = btrfs_lookup_extent_info(trans, root,
4824 eb->start, eb->len,
4825 &wc->refs[level],
4826 &wc->flags[level]);
4827 BUG_ON(ret);
4828 BUG_ON(wc->refs[level] == 0);
4829 }
4830
4831 if (wc->stage == DROP_REFERENCE &&
4832 wc->update_ref && wc->refs[level] > 1) {
4833 BUG_ON(eb == root->node);
4834 BUG_ON(path->slots[level] > 0);
4835 if (level == 0)
4836 btrfs_item_key_to_cpu(eb, &key, path->slots[level]);
4837 else
4838 btrfs_node_key_to_cpu(eb, &key, path->slots[level]);
4839 if (btrfs_header_owner(eb) == root->root_key.objectid &&
4840 btrfs_comp_cpu_keys(&key, &wc->update_progress) >= 0) {
4841 wc->stage = UPDATE_BACKREF;
4842 wc->shared_level = level;
4843 }
4844 }
4845
4846 if (wc->stage == DROP_REFERENCE) {
4847 if (wc->refs[level] > 1)
4848 return 1;
4849
4850 if (path->locks[level] && !wc->keep_locks) {
4851 btrfs_tree_unlock(eb);
4852 path->locks[level] = 0;
4853 }
4854 return 0;
4855 }
4856
4857 /* wc->stage == UPDATE_BACKREF */
4858 if (!(wc->flags[level] & flag)) {
4859 BUG_ON(!path->locks[level]);
4860 ret = btrfs_inc_ref(trans, root, eb, 1);
4861 BUG_ON(ret);
4862 ret = btrfs_dec_ref(trans, root, eb, 0);
4863 BUG_ON(ret);
4864 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
4865 eb->len, flag, 0);
4866 BUG_ON(ret);
4867 wc->flags[level] |= flag;
4868 }
4869
4870 /*
4871 * the block is shared by multiple trees, so it's not good to
4872 * keep the tree lock
4873 */
4874 if (path->locks[level] && level > 0) {
4875 btrfs_tree_unlock(eb);
4876 path->locks[level] = 0;
4877 }
4878 return 0;
4879 }
4880
4881 /*
4882 * hepler to process tree block while walking up the tree.
4883 *
4884 * when wc->stage == DROP_REFERENCE, this function drops
4885 * reference count on the block.
4886 *
4887 * when wc->stage == UPDATE_BACKREF, this function changes
4888 * wc->stage back to DROP_REFERENCE if we changed wc->stage
4889 * to UPDATE_BACKREF previously while processing the block.
4890 *
4891 * NOTE: return value 1 means we should stop walking up.
4892 */
4893 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
4894 struct btrfs_root *root,
4895 struct btrfs_path *path,
4896 struct walk_control *wc)
4897 {
4898 int ret = 0;
4899 int level = wc->level;
4900 struct extent_buffer *eb = path->nodes[level];
4901 u64 parent = 0;
4902
4903 if (wc->stage == UPDATE_BACKREF) {
4904 BUG_ON(wc->shared_level < level);
4905 if (level < wc->shared_level)
4906 goto out;
4907
4908 BUG_ON(wc->refs[level] <= 1);
4909 ret = find_next_key(path, level + 1, &wc->update_progress);
4910 if (ret > 0)
4911 wc->update_ref = 0;
4912
4913 wc->stage = DROP_REFERENCE;
4914 wc->shared_level = -1;
4915 path->slots[level] = 0;
4916
4917 /*
4918 * check reference count again if the block isn't locked.
4919 * we should start walking down the tree again if reference
4920 * count is one.
4921 */
4922 if (!path->locks[level]) {
4923 BUG_ON(level == 0);
4924 btrfs_tree_lock(eb);
4925 btrfs_set_lock_blocking(eb);
4926 path->locks[level] = 1;
4927
4928 ret = btrfs_lookup_extent_info(trans, root,
4929 eb->start, eb->len,
4930 &wc->refs[level],
4931 &wc->flags[level]);
4932 BUG_ON(ret);
4933 BUG_ON(wc->refs[level] == 0);
4934 if (wc->refs[level] == 1) {
4935 btrfs_tree_unlock(eb);
4936 path->locks[level] = 0;
4937 return 1;
4938 }
4939 } else {
4940 BUG_ON(level != 0);
4941 }
4942 }
4943
4944 /* wc->stage == DROP_REFERENCE */
4945 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
4946
4947 if (wc->refs[level] == 1) {
4948 if (level == 0) {
4949 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
4950 ret = btrfs_dec_ref(trans, root, eb, 1);
4951 else
4952 ret = btrfs_dec_ref(trans, root, eb, 0);
4953 BUG_ON(ret);
4954 }
4955 /* make block locked assertion in clean_tree_block happy */
4956 if (!path->locks[level] &&
4957 btrfs_header_generation(eb) == trans->transid) {
4958 btrfs_tree_lock(eb);
4959 btrfs_set_lock_blocking(eb);
4960 path->locks[level] = 1;
4961 }
4962 clean_tree_block(trans, root, eb);
4963 }
4964
4965 if (eb == root->node) {
4966 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
4967 parent = eb->start;
4968 else
4969 BUG_ON(root->root_key.objectid !=
4970 btrfs_header_owner(eb));
4971 } else {
4972 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
4973 parent = path->nodes[level + 1]->start;
4974 else
4975 BUG_ON(root->root_key.objectid !=
4976 btrfs_header_owner(path->nodes[level + 1]));
4977 }
4978
4979 ret = btrfs_free_extent(trans, root, eb->start, eb->len, parent,
4980 root->root_key.objectid, level, 0);
4981 BUG_ON(ret);
4982 out:
4983 wc->refs[level] = 0;
4984 wc->flags[level] = 0;
4985 return ret;
4986 }
4987
4988 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
4989 struct btrfs_root *root,
4990 struct btrfs_path *path,
4991 struct walk_control *wc)
4992 {
4993 struct extent_buffer *next;
4994 struct extent_buffer *cur;
4995 u64 bytenr;
4996 u64 ptr_gen;
4997 u32 blocksize;
4998 int level = wc->level;
4999 int ret;
5000
5001 while (level >= 0) {
5002 cur = path->nodes[level];
5003 BUG_ON(path->slots[level] >= btrfs_header_nritems(cur));
5004
5005 ret = walk_down_proc(trans, root, path, wc);
5006 if (ret > 0)
5007 break;
5008
5009 if (level == 0)
5010 break;
5011
5012 bytenr = btrfs_node_blockptr(cur, path->slots[level]);
5013 blocksize = btrfs_level_size(root, level - 1);
5014 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[level]);
5015
5016 next = read_tree_block(root, bytenr, blocksize, ptr_gen);
5017 btrfs_tree_lock(next);
5018 btrfs_set_lock_blocking(next);
5019
5020 level--;
5021 BUG_ON(level != btrfs_header_level(next));
5022 path->nodes[level] = next;
5023 path->slots[level] = 0;
5024 path->locks[level] = 1;
5025 wc->level = level;
5026 }
5027 return 0;
5028 }
5029
5030 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
5031 struct btrfs_root *root,
5032 struct btrfs_path *path,
5033 struct walk_control *wc, int max_level)
5034 {
5035 int level = wc->level;
5036 int ret;
5037
5038 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
5039 while (level < max_level && path->nodes[level]) {
5040 wc->level = level;
5041 if (path->slots[level] + 1 <
5042 btrfs_header_nritems(path->nodes[level])) {
5043 path->slots[level]++;
5044 return 0;
5045 } else {
5046 ret = walk_up_proc(trans, root, path, wc);
5047 if (ret > 0)
5048 return 0;
5049
5050 if (path->locks[level]) {
5051 btrfs_tree_unlock(path->nodes[level]);
5052 path->locks[level] = 0;
5053 }
5054 free_extent_buffer(path->nodes[level]);
5055 path->nodes[level] = NULL;
5056 level++;
5057 }
5058 }
5059 return 1;
5060 }
5061
5062 /*
5063 * drop a subvolume tree.
5064 *
5065 * this function traverses the tree freeing any blocks that only
5066 * referenced by the tree.
5067 *
5068 * when a shared tree block is found. this function decreases its
5069 * reference count by one. if update_ref is true, this function
5070 * also make sure backrefs for the shared block and all lower level
5071 * blocks are properly updated.
5072 */
5073 int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref)
5074 {
5075 struct btrfs_path *path;
5076 struct btrfs_trans_handle *trans;
5077 struct btrfs_root *tree_root = root->fs_info->tree_root;
5078 struct btrfs_root_item *root_item = &root->root_item;
5079 struct walk_control *wc;
5080 struct btrfs_key key;
5081 int err = 0;
5082 int ret;
5083 int level;
5084
5085 path = btrfs_alloc_path();
5086 BUG_ON(!path);
5087
5088 wc = kzalloc(sizeof(*wc), GFP_NOFS);
5089 BUG_ON(!wc);
5090
5091 trans = btrfs_start_transaction(tree_root, 1);
5092
5093 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
5094 level = btrfs_header_level(root->node);
5095 path->nodes[level] = btrfs_lock_root_node(root);
5096 btrfs_set_lock_blocking(path->nodes[level]);
5097 path->slots[level] = 0;
5098 path->locks[level] = 1;
5099 memset(&wc->update_progress, 0,
5100 sizeof(wc->update_progress));
5101 } else {
5102 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
5103 memcpy(&wc->update_progress, &key,
5104 sizeof(wc->update_progress));
5105
5106 level = root_item->drop_level;
5107 BUG_ON(level == 0);
5108 path->lowest_level = level;
5109 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5110 path->lowest_level = 0;
5111 if (ret < 0) {
5112 err = ret;
5113 goto out;
5114 }
5115 btrfs_node_key_to_cpu(path->nodes[level], &key,
5116 path->slots[level]);
5117 WARN_ON(memcmp(&key, &wc->update_progress, sizeof(key)));
5118
5119 /*
5120 * unlock our path, this is safe because only this
5121 * function is allowed to delete this snapshot
5122 */
5123 btrfs_unlock_up_safe(path, 0);
5124
5125 level = btrfs_header_level(root->node);
5126 while (1) {
5127 btrfs_tree_lock(path->nodes[level]);
5128 btrfs_set_lock_blocking(path->nodes[level]);
5129
5130 ret = btrfs_lookup_extent_info(trans, root,
5131 path->nodes[level]->start,
5132 path->nodes[level]->len,
5133 &wc->refs[level],
5134 &wc->flags[level]);
5135 BUG_ON(ret);
5136 BUG_ON(wc->refs[level] == 0);
5137
5138 if (level == root_item->drop_level)
5139 break;
5140
5141 btrfs_tree_unlock(path->nodes[level]);
5142 WARN_ON(wc->refs[level] != 1);
5143 level--;
5144 }
5145 }
5146
5147 wc->level = level;
5148 wc->shared_level = -1;
5149 wc->stage = DROP_REFERENCE;
5150 wc->update_ref = update_ref;
5151 wc->keep_locks = 0;
5152
5153 while (1) {
5154 ret = walk_down_tree(trans, root, path, wc);
5155 if (ret < 0) {
5156 err = ret;
5157 break;
5158 }
5159
5160 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
5161 if (ret < 0) {
5162 err = ret;
5163 break;
5164 }
5165
5166 if (ret > 0) {
5167 BUG_ON(wc->stage != DROP_REFERENCE);
5168 break;
5169 }
5170
5171 if (wc->stage == DROP_REFERENCE) {
5172 level = wc->level;
5173 btrfs_node_key(path->nodes[level],
5174 &root_item->drop_progress,
5175 path->slots[level]);
5176 root_item->drop_level = level;
5177 }
5178
5179 BUG_ON(wc->level == 0);
5180 if (trans->transaction->in_commit ||
5181 trans->transaction->delayed_refs.flushing) {
5182 ret = btrfs_update_root(trans, tree_root,
5183 &root->root_key,
5184 root_item);
5185 BUG_ON(ret);
5186
5187 btrfs_end_transaction(trans, tree_root);
5188 trans = btrfs_start_transaction(tree_root, 1);
5189 } else {
5190 unsigned long update;
5191 update = trans->delayed_ref_updates;
5192 trans->delayed_ref_updates = 0;
5193 if (update)
5194 btrfs_run_delayed_refs(trans, tree_root,
5195 update);
5196 }
5197 }
5198 btrfs_release_path(root, path);
5199 BUG_ON(err);
5200
5201 ret = btrfs_del_root(trans, tree_root, &root->root_key);
5202 BUG_ON(ret);
5203
5204 free_extent_buffer(root->node);
5205 free_extent_buffer(root->commit_root);
5206 kfree(root);
5207 out:
5208 btrfs_end_transaction(trans, tree_root);
5209 kfree(wc);
5210 btrfs_free_path(path);
5211 return err;
5212 }
5213
5214 /*
5215 * drop subtree rooted at tree block 'node'.
5216 *
5217 * NOTE: this function will unlock and release tree block 'node'
5218 */
5219 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
5220 struct btrfs_root *root,
5221 struct extent_buffer *node,
5222 struct extent_buffer *parent)
5223 {
5224 struct btrfs_path *path;
5225 struct walk_control *wc;
5226 int level;
5227 int parent_level;
5228 int ret = 0;
5229 int wret;
5230
5231 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
5232
5233 path = btrfs_alloc_path();
5234 BUG_ON(!path);
5235
5236 wc = kzalloc(sizeof(*wc), GFP_NOFS);
5237 BUG_ON(!wc);
5238
5239 btrfs_assert_tree_locked(parent);
5240 parent_level = btrfs_header_level(parent);
5241 extent_buffer_get(parent);
5242 path->nodes[parent_level] = parent;
5243 path->slots[parent_level] = btrfs_header_nritems(parent);
5244
5245 btrfs_assert_tree_locked(node);
5246 level = btrfs_header_level(node);
5247 path->nodes[level] = node;
5248 path->slots[level] = 0;
5249 path->locks[level] = 1;
5250
5251 wc->refs[parent_level] = 1;
5252 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5253 wc->level = level;
5254 wc->shared_level = -1;
5255 wc->stage = DROP_REFERENCE;
5256 wc->update_ref = 0;
5257 wc->keep_locks = 1;
5258
5259 while (1) {
5260 wret = walk_down_tree(trans, root, path, wc);
5261 if (wret < 0) {
5262 ret = wret;
5263 break;
5264 }
5265
5266 wret = walk_up_tree(trans, root, path, wc, parent_level);
5267 if (wret < 0)
5268 ret = wret;
5269 if (wret != 0)
5270 break;
5271 }
5272
5273 kfree(wc);
5274 btrfs_free_path(path);
5275 return ret;
5276 }
5277
5278 #if 0
5279 static unsigned long calc_ra(unsigned long start, unsigned long last,
5280 unsigned long nr)
5281 {
5282 return min(last, start + nr - 1);
5283 }
5284
5285 static noinline int relocate_inode_pages(struct inode *inode, u64 start,
5286 u64 len)
5287 {
5288 u64 page_start;
5289 u64 page_end;
5290 unsigned long first_index;
5291 unsigned long last_index;
5292 unsigned long i;
5293 struct page *page;
5294 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5295 struct file_ra_state *ra;
5296 struct btrfs_ordered_extent *ordered;
5297 unsigned int total_read = 0;
5298 unsigned int total_dirty = 0;
5299 int ret = 0;
5300
5301 ra = kzalloc(sizeof(*ra), GFP_NOFS);
5302
5303 mutex_lock(&inode->i_mutex);
5304 first_index = start >> PAGE_CACHE_SHIFT;
5305 last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
5306
5307 /* make sure the dirty trick played by the caller work */
5308 ret = invalidate_inode_pages2_range(inode->i_mapping,
5309 first_index, last_index);
5310 if (ret)
5311 goto out_unlock;
5312
5313 file_ra_state_init(ra, inode->i_mapping);
5314
5315 for (i = first_index ; i <= last_index; i++) {
5316 if (total_read % ra->ra_pages == 0) {
5317 btrfs_force_ra(inode->i_mapping, ra, NULL, i,
5318 calc_ra(i, last_index, ra->ra_pages));
5319 }
5320 total_read++;
5321 again:
5322 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
5323 BUG_ON(1);
5324 page = grab_cache_page(inode->i_mapping, i);
5325 if (!page) {
5326 ret = -ENOMEM;
5327 goto out_unlock;
5328 }
5329 if (!PageUptodate(page)) {
5330 btrfs_readpage(NULL, page);
5331 lock_page(page);
5332 if (!PageUptodate(page)) {
5333 unlock_page(page);
5334 page_cache_release(page);
5335 ret = -EIO;
5336 goto out_unlock;
5337 }
5338 }
5339 wait_on_page_writeback(page);
5340
5341 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
5342 page_end = page_start + PAGE_CACHE_SIZE - 1;
5343 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
5344
5345 ordered = btrfs_lookup_ordered_extent(inode, page_start);
5346 if (ordered) {
5347 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
5348 unlock_page(page);
5349 page_cache_release(page);
5350 btrfs_start_ordered_extent(inode, ordered, 1);
5351 btrfs_put_ordered_extent(ordered);
5352 goto again;
5353 }
5354 set_page_extent_mapped(page);
5355
5356 if (i == first_index)
5357 set_extent_bits(io_tree, page_start, page_end,
5358 EXTENT_BOUNDARY, GFP_NOFS);
5359 btrfs_set_extent_delalloc(inode, page_start, page_end);
5360
5361 set_page_dirty(page);
5362 total_dirty++;
5363
5364 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
5365 unlock_page(page);
5366 page_cache_release(page);
5367 }
5368
5369 out_unlock:
5370 kfree(ra);
5371 mutex_unlock(&inode->i_mutex);
5372 balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
5373 return ret;
5374 }
5375
5376 static noinline int relocate_data_extent(struct inode *reloc_inode,
5377 struct btrfs_key *extent_key,
5378 u64 offset)
5379 {
5380 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
5381 struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
5382 struct extent_map *em;
5383 u64 start = extent_key->objectid - offset;
5384 u64 end = start + extent_key->offset - 1;
5385
5386 em = alloc_extent_map(GFP_NOFS);
5387 BUG_ON(!em || IS_ERR(em));
5388
5389 em->start = start;
5390 em->len = extent_key->offset;
5391 em->block_len = extent_key->offset;
5392 em->block_start = extent_key->objectid;
5393 em->bdev = root->fs_info->fs_devices->latest_bdev;
5394 set_bit(EXTENT_FLAG_PINNED, &em->flags);
5395
5396 /* setup extent map to cheat btrfs_readpage */
5397 lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
5398 while (1) {
5399 int ret;
5400 spin_lock(&em_tree->lock);
5401 ret = add_extent_mapping(em_tree, em);
5402 spin_unlock(&em_tree->lock);
5403 if (ret != -EEXIST) {
5404 free_extent_map(em);
5405 break;
5406 }
5407 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
5408 }
5409 unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
5410
5411 return relocate_inode_pages(reloc_inode, start, extent_key->offset);
5412 }
5413
5414 struct btrfs_ref_path {
5415 u64 extent_start;
5416 u64 nodes[BTRFS_MAX_LEVEL];
5417 u64 root_objectid;
5418 u64 root_generation;
5419 u64 owner_objectid;
5420 u32 num_refs;
5421 int lowest_level;
5422 int current_level;
5423 int shared_level;
5424
5425 struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
5426 u64 new_nodes[BTRFS_MAX_LEVEL];
5427 };
5428
5429 struct disk_extent {
5430 u64 ram_bytes;
5431 u64 disk_bytenr;
5432 u64 disk_num_bytes;
5433 u64 offset;
5434 u64 num_bytes;
5435 u8 compression;
5436 u8 encryption;
5437 u16 other_encoding;
5438 };
5439
5440 static int is_cowonly_root(u64 root_objectid)
5441 {
5442 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
5443 root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
5444 root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
5445 root_objectid == BTRFS_DEV_TREE_OBJECTID ||
5446 root_objectid == BTRFS_TREE_LOG_OBJECTID ||
5447 root_objectid == BTRFS_CSUM_TREE_OBJECTID)
5448 return 1;
5449 return 0;
5450 }
5451
5452 static noinline int __next_ref_path(struct btrfs_trans_handle *trans,
5453 struct btrfs_root *extent_root,
5454 struct btrfs_ref_path *ref_path,
5455 int first_time)
5456 {
5457 struct extent_buffer *leaf;
5458 struct btrfs_path *path;
5459 struct btrfs_extent_ref *ref;
5460 struct btrfs_key key;
5461 struct btrfs_key found_key;
5462 u64 bytenr;
5463 u32 nritems;
5464 int level;
5465 int ret = 1;
5466
5467 path = btrfs_alloc_path();
5468 if (!path)
5469 return -ENOMEM;
5470
5471 if (first_time) {
5472 ref_path->lowest_level = -1;
5473 ref_path->current_level = -1;
5474 ref_path->shared_level = -1;
5475 goto walk_up;
5476 }
5477 walk_down:
5478 level = ref_path->current_level - 1;
5479 while (level >= -1) {
5480 u64 parent;
5481 if (level < ref_path->lowest_level)
5482 break;
5483
5484 if (level >= 0)
5485 bytenr = ref_path->nodes[level];
5486 else
5487 bytenr = ref_path->extent_start;
5488 BUG_ON(bytenr == 0);
5489
5490 parent = ref_path->nodes[level + 1];
5491 ref_path->nodes[level + 1] = 0;
5492 ref_path->current_level = level;
5493 BUG_ON(parent == 0);
5494
5495 key.objectid = bytenr;
5496 key.offset = parent + 1;
5497 key.type = BTRFS_EXTENT_REF_KEY;
5498
5499 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
5500 if (ret < 0)
5501 goto out;
5502 BUG_ON(ret == 0);
5503
5504 leaf = path->nodes[0];
5505 nritems = btrfs_header_nritems(leaf);
5506 if (path->slots[0] >= nritems) {
5507 ret = btrfs_next_leaf(extent_root, path);
5508 if (ret < 0)
5509 goto out;
5510 if (ret > 0)
5511 goto next;
5512 leaf = path->nodes[0];
5513 }
5514
5515 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5516 if (found_key.objectid == bytenr &&
5517 found_key.type == BTRFS_EXTENT_REF_KEY) {
5518 if (level < ref_path->shared_level)
5519 ref_path->shared_level = level;
5520 goto found;
5521 }
5522 next:
5523 level--;
5524 btrfs_release_path(extent_root, path);
5525 cond_resched();
5526 }
5527 /* reached lowest level */
5528 ret = 1;
5529 goto out;
5530 walk_up:
5531 level = ref_path->current_level;
5532 while (level < BTRFS_MAX_LEVEL - 1) {
5533 u64 ref_objectid;
5534
5535 if (level >= 0)
5536 bytenr = ref_path->nodes[level];
5537 else
5538 bytenr = ref_path->extent_start;
5539
5540 BUG_ON(bytenr == 0);
5541
5542 key.objectid = bytenr;
5543 key.offset = 0;
5544 key.type = BTRFS_EXTENT_REF_KEY;
5545
5546 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
5547 if (ret < 0)
5548 goto out;
5549
5550 leaf = path->nodes[0];
5551 nritems = btrfs_header_nritems(leaf);
5552 if (path->slots[0] >= nritems) {
5553 ret = btrfs_next_leaf(extent_root, path);
5554 if (ret < 0)
5555 goto out;
5556 if (ret > 0) {
5557 /* the extent was freed by someone */
5558 if (ref_path->lowest_level == level)
5559 goto out;
5560 btrfs_release_path(extent_root, path);
5561 goto walk_down;
5562 }
5563 leaf = path->nodes[0];
5564 }
5565
5566 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5567 if (found_key.objectid != bytenr ||
5568 found_key.type != BTRFS_EXTENT_REF_KEY) {
5569 /* the extent was freed by someone */
5570 if (ref_path->lowest_level == level) {
5571 ret = 1;
5572 goto out;
5573 }
5574 btrfs_release_path(extent_root, path);
5575 goto walk_down;
5576 }
5577 found:
5578 ref = btrfs_item_ptr(leaf, path->slots[0],
5579 struct btrfs_extent_ref);
5580 ref_objectid = btrfs_ref_objectid(leaf, ref);
5581 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
5582 if (first_time) {
5583 level = (int)ref_objectid;
5584 BUG_ON(level >= BTRFS_MAX_LEVEL);
5585 ref_path->lowest_level = level;
5586 ref_path->current_level = level;
5587 ref_path->nodes[level] = bytenr;
5588 } else {
5589 WARN_ON(ref_objectid != level);
5590 }
5591 } else {
5592 WARN_ON(level != -1);
5593 }
5594 first_time = 0;
5595
5596 if (ref_path->lowest_level == level) {
5597 ref_path->owner_objectid = ref_objectid;
5598 ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
5599 }
5600
5601 /*
5602 * the block is tree root or the block isn't in reference
5603 * counted tree.
5604 */
5605 if (found_key.objectid == found_key.offset ||
5606 is_cowonly_root(btrfs_ref_root(leaf, ref))) {
5607 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
5608 ref_path->root_generation =
5609 btrfs_ref_generation(leaf, ref);
5610 if (level < 0) {
5611 /* special reference from the tree log */
5612 ref_path->nodes[0] = found_key.offset;
5613 ref_path->current_level = 0;
5614 }
5615 ret = 0;
5616 goto out;
5617 }
5618
5619 level++;
5620 BUG_ON(ref_path->nodes[level] != 0);
5621 ref_path->nodes[level] = found_key.offset;
5622 ref_path->current_level = level;
5623
5624 /*
5625 * the reference was created in the running transaction,
5626 * no need to continue walking up.
5627 */
5628 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
5629 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
5630 ref_path->root_generation =
5631 btrfs_ref_generation(leaf, ref);
5632 ret = 0;
5633 goto out;
5634 }
5635
5636 btrfs_release_path(extent_root, path);
5637 cond_resched();
5638 }
5639 /* reached max tree level, but no tree root found. */
5640 BUG();
5641 out:
5642 btrfs_free_path(path);
5643 return ret;
5644 }
5645
5646 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
5647 struct btrfs_root *extent_root,
5648 struct btrfs_ref_path *ref_path,
5649 u64 extent_start)
5650 {
5651 memset(ref_path, 0, sizeof(*ref_path));
5652 ref_path->extent_start = extent_start;
5653
5654 return __next_ref_path(trans, extent_root, ref_path, 1);
5655 }
5656
5657 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
5658 struct btrfs_root *extent_root,
5659 struct btrfs_ref_path *ref_path)
5660 {
5661 return __next_ref_path(trans, extent_root, ref_path, 0);
5662 }
5663
5664 static noinline int get_new_locations(struct inode *reloc_inode,
5665 struct btrfs_key *extent_key,
5666 u64 offset, int no_fragment,
5667 struct disk_extent **extents,
5668 int *nr_extents)
5669 {
5670 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
5671 struct btrfs_path *path;
5672 struct btrfs_file_extent_item *fi;
5673 struct extent_buffer *leaf;
5674 struct disk_extent *exts = *extents;
5675 struct btrfs_key found_key;
5676 u64 cur_pos;
5677 u64 last_byte;
5678 u32 nritems;
5679 int nr = 0;
5680 int max = *nr_extents;
5681 int ret;
5682
5683 WARN_ON(!no_fragment && *extents);
5684 if (!exts) {
5685 max = 1;
5686 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
5687 if (!exts)
5688 return -ENOMEM;
5689 }
5690
5691 path = btrfs_alloc_path();
5692 BUG_ON(!path);
5693
5694 cur_pos = extent_key->objectid - offset;
5695 last_byte = extent_key->objectid + extent_key->offset;
5696 ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
5697 cur_pos, 0);
5698 if (ret < 0)
5699 goto out;
5700 if (ret > 0) {
5701 ret = -ENOENT;
5702 goto out;
5703 }
5704
5705 while (1) {
5706 leaf = path->nodes[0];
5707 nritems = btrfs_header_nritems(leaf);
5708 if (path->slots[0] >= nritems) {
5709 ret = btrfs_next_leaf(root, path);
5710 if (ret < 0)
5711 goto out;
5712 if (ret > 0)
5713 break;
5714 leaf = path->nodes[0];
5715 }
5716
5717 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5718 if (found_key.offset != cur_pos ||
5719 found_key.type != BTRFS_EXTENT_DATA_KEY ||
5720 found_key.objectid != reloc_inode->i_ino)
5721 break;
5722
5723 fi = btrfs_item_ptr(leaf, path->slots[0],
5724 struct btrfs_file_extent_item);
5725 if (btrfs_file_extent_type(leaf, fi) !=
5726 BTRFS_FILE_EXTENT_REG ||
5727 btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
5728 break;
5729
5730 if (nr == max) {
5731 struct disk_extent *old = exts;
5732 max *= 2;
5733 exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
5734 memcpy(exts, old, sizeof(*exts) * nr);
5735 if (old != *extents)
5736 kfree(old);
5737 }
5738
5739 exts[nr].disk_bytenr =
5740 btrfs_file_extent_disk_bytenr(leaf, fi);
5741 exts[nr].disk_num_bytes =
5742 btrfs_file_extent_disk_num_bytes(leaf, fi);
5743 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
5744 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
5745 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
5746 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
5747 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
5748 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
5749 fi);
5750 BUG_ON(exts[nr].offset > 0);
5751 BUG_ON(exts[nr].compression || exts[nr].encryption);
5752 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
5753
5754 cur_pos += exts[nr].num_bytes;
5755 nr++;
5756
5757 if (cur_pos + offset >= last_byte)
5758 break;
5759
5760 if (no_fragment) {
5761 ret = 1;
5762 goto out;
5763 }
5764 path->slots[0]++;
5765 }
5766
5767 BUG_ON(cur_pos + offset > last_byte);
5768 if (cur_pos + offset < last_byte) {
5769 ret = -ENOENT;
5770 goto out;
5771 }
5772 ret = 0;
5773 out:
5774 btrfs_free_path(path);
5775 if (ret) {
5776 if (exts != *extents)
5777 kfree(exts);
5778 } else {
5779 *extents = exts;
5780 *nr_extents = nr;
5781 }
5782 return ret;
5783 }
5784
5785 static noinline int replace_one_extent(struct btrfs_trans_handle *trans,
5786 struct btrfs_root *root,
5787 struct btrfs_path *path,
5788 struct btrfs_key *extent_key,
5789 struct btrfs_key *leaf_key,
5790 struct btrfs_ref_path *ref_path,
5791 struct disk_extent *new_extents,
5792 int nr_extents)
5793 {
5794 struct extent_buffer *leaf;
5795 struct btrfs_file_extent_item *fi;
5796 struct inode *inode = NULL;
5797 struct btrfs_key key;
5798 u64 lock_start = 0;
5799 u64 lock_end = 0;
5800 u64 num_bytes;
5801 u64 ext_offset;
5802 u64 search_end = (u64)-1;
5803 u32 nritems;
5804 int nr_scaned = 0;
5805 int extent_locked = 0;
5806 int extent_type;
5807 int ret;
5808
5809 memcpy(&key, leaf_key, sizeof(key));
5810 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
5811 if (key.objectid < ref_path->owner_objectid ||
5812 (key.objectid == ref_path->owner_objectid &&
5813 key.type < BTRFS_EXTENT_DATA_KEY)) {
5814 key.objectid = ref_path->owner_objectid;
5815 key.type = BTRFS_EXTENT_DATA_KEY;
5816 key.offset = 0;
5817 }
5818 }
5819
5820 while (1) {
5821 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
5822 if (ret < 0)
5823 goto out;
5824
5825 leaf = path->nodes[0];
5826 nritems = btrfs_header_nritems(leaf);
5827 next:
5828 if (extent_locked && ret > 0) {
5829 /*
5830 * the file extent item was modified by someone
5831 * before the extent got locked.
5832 */
5833 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
5834 lock_end, GFP_NOFS);
5835 extent_locked = 0;
5836 }
5837
5838 if (path->slots[0] >= nritems) {
5839 if (++nr_scaned > 2)
5840 break;
5841
5842 BUG_ON(extent_locked);
5843 ret = btrfs_next_leaf(root, path);
5844 if (ret < 0)
5845 goto out;
5846 if (ret > 0)
5847 break;
5848 leaf = path->nodes[0];
5849 nritems = btrfs_header_nritems(leaf);
5850 }
5851
5852 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
5853
5854 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
5855 if ((key.objectid > ref_path->owner_objectid) ||
5856 (key.objectid == ref_path->owner_objectid &&
5857 key.type > BTRFS_EXTENT_DATA_KEY) ||
5858 key.offset >= search_end)
5859 break;
5860 }
5861
5862 if (inode && key.objectid != inode->i_ino) {
5863 BUG_ON(extent_locked);
5864 btrfs_release_path(root, path);
5865 mutex_unlock(&inode->i_mutex);
5866 iput(inode);
5867 inode = NULL;
5868 continue;
5869 }
5870
5871 if (key.type != BTRFS_EXTENT_DATA_KEY) {
5872 path->slots[0]++;
5873 ret = 1;
5874 goto next;
5875 }
5876 fi = btrfs_item_ptr(leaf, path->slots[0],
5877 struct btrfs_file_extent_item);
5878 extent_type = btrfs_file_extent_type(leaf, fi);
5879 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
5880 extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
5881 (btrfs_file_extent_disk_bytenr(leaf, fi) !=
5882 extent_key->objectid)) {
5883 path->slots[0]++;
5884 ret = 1;
5885 goto next;
5886 }
5887
5888 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
5889 ext_offset = btrfs_file_extent_offset(leaf, fi);
5890
5891 if (search_end == (u64)-1) {
5892 search_end = key.offset - ext_offset +
5893 btrfs_file_extent_ram_bytes(leaf, fi);
5894 }
5895
5896 if (!extent_locked) {
5897 lock_start = key.offset;
5898 lock_end = lock_start + num_bytes - 1;
5899 } else {
5900 if (lock_start > key.offset ||
5901 lock_end + 1 < key.offset + num_bytes) {
5902 unlock_extent(&BTRFS_I(inode)->io_tree,
5903 lock_start, lock_end, GFP_NOFS);
5904 extent_locked = 0;
5905 }
5906 }
5907
5908 if (!inode) {
5909 btrfs_release_path(root, path);
5910
5911 inode = btrfs_iget_locked(root->fs_info->sb,
5912 key.objectid, root);
5913 if (inode->i_state & I_NEW) {
5914 BTRFS_I(inode)->root = root;
5915 BTRFS_I(inode)->location.objectid =
5916 key.objectid;
5917 BTRFS_I(inode)->location.type =
5918 BTRFS_INODE_ITEM_KEY;
5919 BTRFS_I(inode)->location.offset = 0;
5920 btrfs_read_locked_inode(inode);
5921 unlock_new_inode(inode);
5922 }
5923 /*
5924 * some code call btrfs_commit_transaction while
5925 * holding the i_mutex, so we can't use mutex_lock
5926 * here.
5927 */
5928 if (is_bad_inode(inode) ||
5929 !mutex_trylock(&inode->i_mutex)) {
5930 iput(inode);
5931 inode = NULL;
5932 key.offset = (u64)-1;
5933 goto skip;
5934 }
5935 }
5936
5937 if (!extent_locked) {
5938 struct btrfs_ordered_extent *ordered;
5939
5940 btrfs_release_path(root, path);
5941
5942 lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
5943 lock_end, GFP_NOFS);
5944 ordered = btrfs_lookup_first_ordered_extent(inode,
5945 lock_end);
5946 if (ordered &&
5947 ordered->file_offset <= lock_end &&
5948 ordered->file_offset + ordered->len > lock_start) {
5949 unlock_extent(&BTRFS_I(inode)->io_tree,
5950 lock_start, lock_end, GFP_NOFS);
5951 btrfs_start_ordered_extent(inode, ordered, 1);
5952 btrfs_put_ordered_extent(ordered);
5953 key.offset += num_bytes;
5954 goto skip;
5955 }
5956 if (ordered)
5957 btrfs_put_ordered_extent(ordered);
5958
5959 extent_locked = 1;
5960 continue;
5961 }
5962
5963 if (nr_extents == 1) {
5964 /* update extent pointer in place */
5965 btrfs_set_file_extent_disk_bytenr(leaf, fi,
5966 new_extents[0].disk_bytenr);
5967 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
5968 new_extents[0].disk_num_bytes);
5969 btrfs_mark_buffer_dirty(leaf);
5970
5971 btrfs_drop_extent_cache(inode, key.offset,
5972 key.offset + num_bytes - 1, 0);
5973
5974 ret = btrfs_inc_extent_ref(trans, root,
5975 new_extents[0].disk_bytenr,
5976 new_extents[0].disk_num_bytes,
5977 leaf->start,
5978 root->root_key.objectid,
5979 trans->transid,
5980 key.objectid);
5981 BUG_ON(ret);
5982
5983 ret = btrfs_free_extent(trans, root,
5984 extent_key->objectid,
5985 extent_key->offset,
5986 leaf->start,
5987 btrfs_header_owner(leaf),
5988 btrfs_header_generation(leaf),
5989 key.objectid, 0);
5990 BUG_ON(ret);
5991
5992 btrfs_release_path(root, path);
5993 key.offset += num_bytes;
5994 } else {
5995 BUG_ON(1);
5996 #if 0
5997 u64 alloc_hint;
5998 u64 extent_len;
5999 int i;
6000 /*
6001 * drop old extent pointer at first, then insert the
6002 * new pointers one bye one
6003 */
6004 btrfs_release_path(root, path);
6005 ret = btrfs_drop_extents(trans, root, inode, key.offset,
6006 key.offset + num_bytes,
6007 key.offset, &alloc_hint);
6008 BUG_ON(ret);
6009
6010 for (i = 0; i < nr_extents; i++) {
6011 if (ext_offset >= new_extents[i].num_bytes) {
6012 ext_offset -= new_extents[i].num_bytes;
6013 continue;
6014 }
6015 extent_len = min(new_extents[i].num_bytes -
6016 ext_offset, num_bytes);
6017
6018 ret = btrfs_insert_empty_item(trans, root,
6019 path, &key,
6020 sizeof(*fi));
6021 BUG_ON(ret);
6022
6023 leaf = path->nodes[0];
6024 fi = btrfs_item_ptr(leaf, path->slots[0],
6025 struct btrfs_file_extent_item);
6026 btrfs_set_file_extent_generation(leaf, fi,
6027 trans->transid);
6028 btrfs_set_file_extent_type(leaf, fi,
6029 BTRFS_FILE_EXTENT_REG);
6030 btrfs_set_file_extent_disk_bytenr(leaf, fi,
6031 new_extents[i].disk_bytenr);
6032 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
6033 new_extents[i].disk_num_bytes);
6034 btrfs_set_file_extent_ram_bytes(leaf, fi,
6035 new_extents[i].ram_bytes);
6036
6037 btrfs_set_file_extent_compression(leaf, fi,
6038 new_extents[i].compression);
6039 btrfs_set_file_extent_encryption(leaf, fi,
6040 new_extents[i].encryption);
6041 btrfs_set_file_extent_other_encoding(leaf, fi,
6042 new_extents[i].other_encoding);
6043
6044 btrfs_set_file_extent_num_bytes(leaf, fi,
6045 extent_len);
6046 ext_offset += new_extents[i].offset;
6047 btrfs_set_file_extent_offset(leaf, fi,
6048 ext_offset);
6049 btrfs_mark_buffer_dirty(leaf);
6050
6051 btrfs_drop_extent_cache(inode, key.offset,
6052 key.offset + extent_len - 1, 0);
6053
6054 ret = btrfs_inc_extent_ref(trans, root,
6055 new_extents[i].disk_bytenr,
6056 new_extents[i].disk_num_bytes,
6057 leaf->start,
6058 root->root_key.objectid,
6059 trans->transid, key.objectid);
6060 BUG_ON(ret);
6061 btrfs_release_path(root, path);
6062
6063 inode_add_bytes(inode, extent_len);
6064
6065 ext_offset = 0;
6066 num_bytes -= extent_len;
6067 key.offset += extent_len;
6068
6069 if (num_bytes == 0)
6070 break;
6071 }
6072 BUG_ON(i >= nr_extents);
6073 #endif
6074 }
6075
6076 if (extent_locked) {
6077 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6078 lock_end, GFP_NOFS);
6079 extent_locked = 0;
6080 }
6081 skip:
6082 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
6083 key.offset >= search_end)
6084 break;
6085
6086 cond_resched();
6087 }
6088 ret = 0;
6089 out:
6090 btrfs_release_path(root, path);
6091 if (inode) {
6092 mutex_unlock(&inode->i_mutex);
6093 if (extent_locked) {
6094 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6095 lock_end, GFP_NOFS);
6096 }
6097 iput(inode);
6098 }
6099 return ret;
6100 }
6101
6102 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
6103 struct btrfs_root *root,
6104 struct extent_buffer *buf, u64 orig_start)
6105 {
6106 int level;
6107 int ret;
6108
6109 BUG_ON(btrfs_header_generation(buf) != trans->transid);
6110 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6111
6112 level = btrfs_header_level(buf);
6113 if (level == 0) {
6114 struct btrfs_leaf_ref *ref;
6115 struct btrfs_leaf_ref *orig_ref;
6116
6117 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
6118 if (!orig_ref)
6119 return -ENOENT;
6120
6121 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
6122 if (!ref) {
6123 btrfs_free_leaf_ref(root, orig_ref);
6124 return -ENOMEM;
6125 }
6126
6127 ref->nritems = orig_ref->nritems;
6128 memcpy(ref->extents, orig_ref->extents,
6129 sizeof(ref->extents[0]) * ref->nritems);
6130
6131 btrfs_free_leaf_ref(root, orig_ref);
6132
6133 ref->root_gen = trans->transid;
6134 ref->bytenr = buf->start;
6135 ref->owner = btrfs_header_owner(buf);
6136 ref->generation = btrfs_header_generation(buf);
6137
6138 ret = btrfs_add_leaf_ref(root, ref, 0);
6139 WARN_ON(ret);
6140 btrfs_free_leaf_ref(root, ref);
6141 }
6142 return 0;
6143 }
6144
6145 static noinline int invalidate_extent_cache(struct btrfs_root *root,
6146 struct extent_buffer *leaf,
6147 struct btrfs_block_group_cache *group,
6148 struct btrfs_root *target_root)
6149 {
6150 struct btrfs_key key;
6151 struct inode *inode = NULL;
6152 struct btrfs_file_extent_item *fi;
6153 u64 num_bytes;
6154 u64 skip_objectid = 0;
6155 u32 nritems;
6156 u32 i;
6157
6158 nritems = btrfs_header_nritems(leaf);
6159 for (i = 0; i < nritems; i++) {
6160 btrfs_item_key_to_cpu(leaf, &key, i);
6161 if (key.objectid == skip_objectid ||
6162 key.type != BTRFS_EXTENT_DATA_KEY)
6163 continue;
6164 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
6165 if (btrfs_file_extent_type(leaf, fi) ==
6166 BTRFS_FILE_EXTENT_INLINE)
6167 continue;
6168 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
6169 continue;
6170 if (!inode || inode->i_ino != key.objectid) {
6171 iput(inode);
6172 inode = btrfs_ilookup(target_root->fs_info->sb,
6173 key.objectid, target_root, 1);
6174 }
6175 if (!inode) {
6176 skip_objectid = key.objectid;
6177 continue;
6178 }
6179 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
6180
6181 lock_extent(&BTRFS_I(inode)->io_tree, key.offset,
6182 key.offset + num_bytes - 1, GFP_NOFS);
6183 btrfs_drop_extent_cache(inode, key.offset,
6184 key.offset + num_bytes - 1, 1);
6185 unlock_extent(&BTRFS_I(inode)->io_tree, key.offset,
6186 key.offset + num_bytes - 1, GFP_NOFS);
6187 cond_resched();
6188 }
6189 iput(inode);
6190 return 0;
6191 }
6192
6193 static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans,
6194 struct btrfs_root *root,
6195 struct extent_buffer *leaf,
6196 struct btrfs_block_group_cache *group,
6197 struct inode *reloc_inode)
6198 {
6199 struct btrfs_key key;
6200 struct btrfs_key extent_key;
6201 struct btrfs_file_extent_item *fi;
6202 struct btrfs_leaf_ref *ref;
6203 struct disk_extent *new_extent;
6204 u64 bytenr;
6205 u64 num_bytes;
6206 u32 nritems;
6207 u32 i;
6208 int ext_index;
6209 int nr_extent;
6210 int ret;
6211
6212 new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
6213 BUG_ON(!new_extent);
6214
6215 ref = btrfs_lookup_leaf_ref(root, leaf->start);
6216 BUG_ON(!ref);
6217
6218 ext_index = -1;
6219 nritems = btrfs_header_nritems(leaf);
6220 for (i = 0; i < nritems; i++) {
6221 btrfs_item_key_to_cpu(leaf, &key, i);
6222 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
6223 continue;
6224 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
6225 if (btrfs_file_extent_type(leaf, fi) ==
6226 BTRFS_FILE_EXTENT_INLINE)
6227 continue;
6228 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
6229 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
6230 if (bytenr == 0)
6231 continue;
6232
6233 ext_index++;
6234 if (bytenr >= group->key.objectid + group->key.offset ||
6235 bytenr + num_bytes <= group->key.objectid)
6236 continue;
6237
6238 extent_key.objectid = bytenr;
6239 extent_key.offset = num_bytes;
6240 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
6241 nr_extent = 1;
6242 ret = get_new_locations(reloc_inode, &extent_key,
6243 group->key.objectid, 1,
6244 &new_extent, &nr_extent);
6245 if (ret > 0)
6246 continue;
6247 BUG_ON(ret < 0);
6248
6249 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
6250 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
6251 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
6252 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
6253
6254 btrfs_set_file_extent_disk_bytenr(leaf, fi,
6255 new_extent->disk_bytenr);
6256 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
6257 new_extent->disk_num_bytes);
6258 btrfs_mark_buffer_dirty(leaf);
6259
6260 ret = btrfs_inc_extent_ref(trans, root,
6261 new_extent->disk_bytenr,
6262 new_extent->disk_num_bytes,
6263 leaf->start,
6264 root->root_key.objectid,
6265 trans->transid, key.objectid);
6266 BUG_ON(ret);
6267
6268 ret = btrfs_free_extent(trans, root,
6269 bytenr, num_bytes, leaf->start,
6270 btrfs_header_owner(leaf),
6271 btrfs_header_generation(leaf),
6272 key.objectid, 0);
6273 BUG_ON(ret);
6274 cond_resched();
6275 }
6276 kfree(new_extent);
6277 BUG_ON(ext_index + 1 != ref->nritems);
6278 btrfs_free_leaf_ref(root, ref);
6279 return 0;
6280 }
6281
6282 int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
6283 struct btrfs_root *root)
6284 {
6285 struct btrfs_root *reloc_root;
6286 int ret;
6287
6288 if (root->reloc_root) {
6289 reloc_root = root->reloc_root;
6290 root->reloc_root = NULL;
6291 list_add(&reloc_root->dead_list,
6292 &root->fs_info->dead_reloc_roots);
6293
6294 btrfs_set_root_bytenr(&reloc_root->root_item,
6295 reloc_root->node->start);
6296 btrfs_set_root_level(&root->root_item,
6297 btrfs_header_level(reloc_root->node));
6298 memset(&reloc_root->root_item.drop_progress, 0,
6299 sizeof(struct btrfs_disk_key));
6300 reloc_root->root_item.drop_level = 0;
6301
6302 ret = btrfs_update_root(trans, root->fs_info->tree_root,
6303 &reloc_root->root_key,
6304 &reloc_root->root_item);
6305 BUG_ON(ret);
6306 }
6307 return 0;
6308 }
6309
6310 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
6311 {
6312 struct btrfs_trans_handle *trans;
6313 struct btrfs_root *reloc_root;
6314 struct btrfs_root *prev_root = NULL;
6315 struct list_head dead_roots;
6316 int ret;
6317 unsigned long nr;
6318
6319 INIT_LIST_HEAD(&dead_roots);
6320 list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
6321
6322 while (!list_empty(&dead_roots)) {
6323 reloc_root = list_entry(dead_roots.prev,
6324 struct btrfs_root, dead_list);
6325 list_del_init(&reloc_root->dead_list);
6326
6327 BUG_ON(reloc_root->commit_root != NULL);
6328 while (1) {
6329 trans = btrfs_join_transaction(root, 1);
6330 BUG_ON(!trans);
6331
6332 mutex_lock(&root->fs_info->drop_mutex);
6333 ret = btrfs_drop_snapshot(trans, reloc_root);
6334 if (ret != -EAGAIN)
6335 break;
6336 mutex_unlock(&root->fs_info->drop_mutex);
6337
6338 nr = trans->blocks_used;
6339 ret = btrfs_end_transaction(trans, root);
6340 BUG_ON(ret);
6341 btrfs_btree_balance_dirty(root, nr);
6342 }
6343
6344 free_extent_buffer(reloc_root->node);
6345
6346 ret = btrfs_del_root(trans, root->fs_info->tree_root,
6347 &reloc_root->root_key);
6348 BUG_ON(ret);
6349 mutex_unlock(&root->fs_info->drop_mutex);
6350
6351 nr = trans->blocks_used;
6352 ret = btrfs_end_transaction(trans, root);
6353 BUG_ON(ret);
6354 btrfs_btree_balance_dirty(root, nr);
6355
6356 kfree(prev_root);
6357 prev_root = reloc_root;
6358 }
6359 if (prev_root) {
6360 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
6361 kfree(prev_root);
6362 }
6363 return 0;
6364 }
6365
6366 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
6367 {
6368 list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
6369 return 0;
6370 }
6371
6372 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
6373 {
6374 struct btrfs_root *reloc_root;
6375 struct btrfs_trans_handle *trans;
6376 struct btrfs_key location;
6377 int found;
6378 int ret;
6379
6380 mutex_lock(&root->fs_info->tree_reloc_mutex);
6381 ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
6382 BUG_ON(ret);
6383 found = !list_empty(&root->fs_info->dead_reloc_roots);
6384 mutex_unlock(&root->fs_info->tree_reloc_mutex);
6385
6386 if (found) {
6387 trans = btrfs_start_transaction(root, 1);
6388 BUG_ON(!trans);
6389 ret = btrfs_commit_transaction(trans, root);
6390 BUG_ON(ret);
6391 }
6392
6393 location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
6394 location.offset = (u64)-1;
6395 location.type = BTRFS_ROOT_ITEM_KEY;
6396
6397 reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
6398 BUG_ON(!reloc_root);
6399 btrfs_orphan_cleanup(reloc_root);
6400 return 0;
6401 }
6402
6403 static noinline int init_reloc_tree(struct btrfs_trans_handle *trans,
6404 struct btrfs_root *root)
6405 {
6406 struct btrfs_root *reloc_root;
6407 struct extent_buffer *eb;
6408 struct btrfs_root_item *root_item;
6409 struct btrfs_key root_key;
6410 int ret;
6411
6412 BUG_ON(!root->ref_cows);
6413 if (root->reloc_root)
6414 return 0;
6415
6416 root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
6417 BUG_ON(!root_item);
6418
6419 ret = btrfs_copy_root(trans, root, root->commit_root,
6420 &eb, BTRFS_TREE_RELOC_OBJECTID);
6421 BUG_ON(ret);
6422
6423 root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
6424 root_key.offset = root->root_key.objectid;
6425 root_key.type = BTRFS_ROOT_ITEM_KEY;
6426
6427 memcpy(root_item, &root->root_item, sizeof(root_item));
6428 btrfs_set_root_refs(root_item, 0);
6429 btrfs_set_root_bytenr(root_item, eb->start);
6430 btrfs_set_root_level(root_item, btrfs_header_level(eb));
6431 btrfs_set_root_generation(root_item, trans->transid);
6432
6433 btrfs_tree_unlock(eb);
6434 free_extent_buffer(eb);
6435
6436 ret = btrfs_insert_root(trans, root->fs_info->tree_root,
6437 &root_key, root_item);
6438 BUG_ON(ret);
6439 kfree(root_item);
6440
6441 reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
6442 &root_key);
6443 BUG_ON(!reloc_root);
6444 reloc_root->last_trans = trans->transid;
6445 reloc_root->commit_root = NULL;
6446 reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
6447
6448 root->reloc_root = reloc_root;
6449 return 0;
6450 }
6451
6452 /*
6453 * Core function of space balance.
6454 *
6455 * The idea is using reloc trees to relocate tree blocks in reference
6456 * counted roots. There is one reloc tree for each subvol, and all
6457 * reloc trees share same root key objectid. Reloc trees are snapshots
6458 * of the latest committed roots of subvols (root->commit_root).
6459 *
6460 * To relocate a tree block referenced by a subvol, there are two steps.
6461 * COW the block through subvol's reloc tree, then update block pointer
6462 * in the subvol to point to the new block. Since all reloc trees share
6463 * same root key objectid, doing special handing for tree blocks owned
6464 * by them is easy. Once a tree block has been COWed in one reloc tree,
6465 * we can use the resulting new block directly when the same block is
6466 * required to COW again through other reloc trees. By this way, relocated
6467 * tree blocks are shared between reloc trees, so they are also shared
6468 * between subvols.
6469 */
6470 static noinline int relocate_one_path(struct btrfs_trans_handle *trans,
6471 struct btrfs_root *root,
6472 struct btrfs_path *path,
6473 struct btrfs_key *first_key,
6474 struct btrfs_ref_path *ref_path,
6475 struct btrfs_block_group_cache *group,
6476 struct inode *reloc_inode)
6477 {
6478 struct btrfs_root *reloc_root;
6479 struct extent_buffer *eb = NULL;
6480 struct btrfs_key *keys;
6481 u64 *nodes;
6482 int level;
6483 int shared_level;
6484 int lowest_level = 0;
6485 int ret;
6486
6487 if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
6488 lowest_level = ref_path->owner_objectid;
6489
6490 if (!root->ref_cows) {
6491 path->lowest_level = lowest_level;
6492 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
6493 BUG_ON(ret < 0);
6494 path->lowest_level = 0;
6495 btrfs_release_path(root, path);
6496 return 0;
6497 }
6498
6499 mutex_lock(&root->fs_info->tree_reloc_mutex);
6500 ret = init_reloc_tree(trans, root);
6501 BUG_ON(ret);
6502 reloc_root = root->reloc_root;
6503
6504 shared_level = ref_path->shared_level;
6505 ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
6506
6507 keys = ref_path->node_keys;
6508 nodes = ref_path->new_nodes;
6509 memset(&keys[shared_level + 1], 0,
6510 sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
6511 memset(&nodes[shared_level + 1], 0,
6512 sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
6513
6514 if (nodes[lowest_level] == 0) {
6515 path->lowest_level = lowest_level;
6516 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
6517 0, 1);
6518 BUG_ON(ret);
6519 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
6520 eb = path->nodes[level];
6521 if (!eb || eb == reloc_root->node)
6522 break;
6523 nodes[level] = eb->start;
6524 if (level == 0)
6525 btrfs_item_key_to_cpu(eb, &keys[level], 0);
6526 else
6527 btrfs_node_key_to_cpu(eb, &keys[level], 0);
6528 }
6529 if (nodes[0] &&
6530 ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
6531 eb = path->nodes[0];
6532 ret = replace_extents_in_leaf(trans, reloc_root, eb,
6533 group, reloc_inode);
6534 BUG_ON(ret);
6535 }
6536 btrfs_release_path(reloc_root, path);
6537 } else {
6538 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
6539 lowest_level);
6540 BUG_ON(ret);
6541 }
6542
6543 /*
6544 * replace tree blocks in the fs tree with tree blocks in
6545 * the reloc tree.
6546 */
6547 ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
6548 BUG_ON(ret < 0);
6549
6550 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
6551 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
6552 0, 0);
6553 BUG_ON(ret);
6554 extent_buffer_get(path->nodes[0]);
6555 eb = path->nodes[0];
6556 btrfs_release_path(reloc_root, path);
6557 ret = invalidate_extent_cache(reloc_root, eb, group, root);
6558 BUG_ON(ret);
6559 free_extent_buffer(eb);
6560 }
6561
6562 mutex_unlock(&root->fs_info->tree_reloc_mutex);
6563 path->lowest_level = 0;
6564 return 0;
6565 }
6566
6567 static noinline int relocate_tree_block(struct btrfs_trans_handle *trans,
6568 struct btrfs_root *root,
6569 struct btrfs_path *path,
6570 struct btrfs_key *first_key,
6571 struct btrfs_ref_path *ref_path)
6572 {
6573 int ret;
6574
6575 ret = relocate_one_path(trans, root, path, first_key,
6576 ref_path, NULL, NULL);
6577 BUG_ON(ret);
6578
6579 return 0;
6580 }
6581
6582 static noinline int del_extent_zero(struct btrfs_trans_handle *trans,
6583 struct btrfs_root *extent_root,
6584 struct btrfs_path *path,
6585 struct btrfs_key *extent_key)
6586 {
6587 int ret;
6588
6589 ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
6590 if (ret)
6591 goto out;
6592 ret = btrfs_del_item(trans, extent_root, path);
6593 out:
6594 btrfs_release_path(extent_root, path);
6595 return ret;
6596 }
6597
6598 static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info,
6599 struct btrfs_ref_path *ref_path)
6600 {
6601 struct btrfs_key root_key;
6602
6603 root_key.objectid = ref_path->root_objectid;
6604 root_key.type = BTRFS_ROOT_ITEM_KEY;
6605 if (is_cowonly_root(ref_path->root_objectid))
6606 root_key.offset = 0;
6607 else
6608 root_key.offset = (u64)-1;
6609
6610 return btrfs_read_fs_root_no_name(fs_info, &root_key);
6611 }
6612
6613 static noinline int relocate_one_extent(struct btrfs_root *extent_root,
6614 struct btrfs_path *path,
6615 struct btrfs_key *extent_key,
6616 struct btrfs_block_group_cache *group,
6617 struct inode *reloc_inode, int pass)
6618 {
6619 struct btrfs_trans_handle *trans;
6620 struct btrfs_root *found_root;
6621 struct btrfs_ref_path *ref_path = NULL;
6622 struct disk_extent *new_extents = NULL;
6623 int nr_extents = 0;
6624 int loops;
6625 int ret;
6626 int level;
6627 struct btrfs_key first_key;
6628 u64 prev_block = 0;
6629
6630
6631 trans = btrfs_start_transaction(extent_root, 1);
6632 BUG_ON(!trans);
6633
6634 if (extent_key->objectid == 0) {
6635 ret = del_extent_zero(trans, extent_root, path, extent_key);
6636 goto out;
6637 }
6638
6639 ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
6640 if (!ref_path) {
6641 ret = -ENOMEM;
6642 goto out;
6643 }
6644
6645 for (loops = 0; ; loops++) {
6646 if (loops == 0) {
6647 ret = btrfs_first_ref_path(trans, extent_root, ref_path,
6648 extent_key->objectid);
6649 } else {
6650 ret = btrfs_next_ref_path(trans, extent_root, ref_path);
6651 }
6652 if (ret < 0)
6653 goto out;
6654 if (ret > 0)
6655 break;
6656
6657 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
6658 ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
6659 continue;
6660
6661 found_root = read_ref_root(extent_root->fs_info, ref_path);
6662 BUG_ON(!found_root);
6663 /*
6664 * for reference counted tree, only process reference paths
6665 * rooted at the latest committed root.
6666 */
6667 if (found_root->ref_cows &&
6668 ref_path->root_generation != found_root->root_key.offset)
6669 continue;
6670
6671 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
6672 if (pass == 0) {
6673 /*
6674 * copy data extents to new locations
6675 */
6676 u64 group_start = group->key.objectid;
6677 ret = relocate_data_extent(reloc_inode,
6678 extent_key,
6679 group_start);
6680 if (ret < 0)
6681 goto out;
6682 break;
6683 }
6684 level = 0;
6685 } else {
6686 level = ref_path->owner_objectid;
6687 }
6688
6689 if (prev_block != ref_path->nodes[level]) {
6690 struct extent_buffer *eb;
6691 u64 block_start = ref_path->nodes[level];
6692 u64 block_size = btrfs_level_size(found_root, level);
6693
6694 eb = read_tree_block(found_root, block_start,
6695 block_size, 0);
6696 btrfs_tree_lock(eb);
6697 BUG_ON(level != btrfs_header_level(eb));
6698
6699 if (level == 0)
6700 btrfs_item_key_to_cpu(eb, &first_key, 0);
6701 else
6702 btrfs_node_key_to_cpu(eb, &first_key, 0);
6703
6704 btrfs_tree_unlock(eb);
6705 free_extent_buffer(eb);
6706 prev_block = block_start;
6707 }
6708
6709 mutex_lock(&extent_root->fs_info->trans_mutex);
6710 btrfs_record_root_in_trans(found_root);
6711 mutex_unlock(&extent_root->fs_info->trans_mutex);
6712 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
6713 /*
6714 * try to update data extent references while
6715 * keeping metadata shared between snapshots.
6716 */
6717 if (pass == 1) {
6718 ret = relocate_one_path(trans, found_root,
6719 path, &first_key, ref_path,
6720 group, reloc_inode);
6721 if (ret < 0)
6722 goto out;
6723 continue;
6724 }
6725 /*
6726 * use fallback method to process the remaining
6727 * references.
6728 */
6729 if (!new_extents) {
6730 u64 group_start = group->key.objectid;
6731 new_extents = kmalloc(sizeof(*new_extents),
6732 GFP_NOFS);
6733 nr_extents = 1;
6734 ret = get_new_locations(reloc_inode,
6735 extent_key,
6736 group_start, 1,
6737 &new_extents,
6738 &nr_extents);
6739 if (ret)
6740 goto out;
6741 }
6742 ret = replace_one_extent(trans, found_root,
6743 path, extent_key,
6744 &first_key, ref_path,
6745 new_extents, nr_extents);
6746 } else {
6747 ret = relocate_tree_block(trans, found_root, path,
6748 &first_key, ref_path);
6749 }
6750 if (ret < 0)
6751 goto out;
6752 }
6753 ret = 0;
6754 out:
6755 btrfs_end_transaction(trans, extent_root);
6756 kfree(new_extents);
6757 kfree(ref_path);
6758 return ret;
6759 }
6760 #endif
6761
6762 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
6763 {
6764 u64 num_devices;
6765 u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
6766 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
6767
6768 num_devices = root->fs_info->fs_devices->rw_devices;
6769 if (num_devices == 1) {
6770 stripped |= BTRFS_BLOCK_GROUP_DUP;
6771 stripped = flags & ~stripped;
6772
6773 /* turn raid0 into single device chunks */
6774 if (flags & BTRFS_BLOCK_GROUP_RAID0)
6775 return stripped;
6776
6777 /* turn mirroring into duplication */
6778 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
6779 BTRFS_BLOCK_GROUP_RAID10))
6780 return stripped | BTRFS_BLOCK_GROUP_DUP;
6781 return flags;
6782 } else {
6783 /* they already had raid on here, just return */
6784 if (flags & stripped)
6785 return flags;
6786
6787 stripped |= BTRFS_BLOCK_GROUP_DUP;
6788 stripped = flags & ~stripped;
6789
6790 /* switch duplicated blocks with raid1 */
6791 if (flags & BTRFS_BLOCK_GROUP_DUP)
6792 return stripped | BTRFS_BLOCK_GROUP_RAID1;
6793
6794 /* turn single device chunks into raid0 */
6795 return stripped | BTRFS_BLOCK_GROUP_RAID0;
6796 }
6797 return flags;
6798 }
6799
6800 static int __alloc_chunk_for_shrink(struct btrfs_root *root,
6801 struct btrfs_block_group_cache *shrink_block_group,
6802 int force)
6803 {
6804 struct btrfs_trans_handle *trans;
6805 u64 new_alloc_flags;
6806 u64 calc;
6807
6808 spin_lock(&shrink_block_group->lock);
6809 if (btrfs_block_group_used(&shrink_block_group->item) +
6810 shrink_block_group->reserved > 0) {
6811 spin_unlock(&shrink_block_group->lock);
6812
6813 trans = btrfs_start_transaction(root, 1);
6814 spin_lock(&shrink_block_group->lock);
6815
6816 new_alloc_flags = update_block_group_flags(root,
6817 shrink_block_group->flags);
6818 if (new_alloc_flags != shrink_block_group->flags) {
6819 calc =
6820 btrfs_block_group_used(&shrink_block_group->item);
6821 } else {
6822 calc = shrink_block_group->key.offset;
6823 }
6824 spin_unlock(&shrink_block_group->lock);
6825
6826 do_chunk_alloc(trans, root->fs_info->extent_root,
6827 calc + 2 * 1024 * 1024, new_alloc_flags, force);
6828
6829 btrfs_end_transaction(trans, root);
6830 } else
6831 spin_unlock(&shrink_block_group->lock);
6832 return 0;
6833 }
6834
6835
6836 int btrfs_prepare_block_group_relocation(struct btrfs_root *root,
6837 struct btrfs_block_group_cache *group)
6838
6839 {
6840 __alloc_chunk_for_shrink(root, group, 1);
6841 set_block_group_readonly(group);
6842 return 0;
6843 }
6844
6845 #if 0
6846 static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
6847 struct btrfs_root *root,
6848 u64 objectid, u64 size)
6849 {
6850 struct btrfs_path *path;
6851 struct btrfs_inode_item *item;
6852 struct extent_buffer *leaf;
6853 int ret;
6854
6855 path = btrfs_alloc_path();
6856 if (!path)
6857 return -ENOMEM;
6858
6859 path->leave_spinning = 1;
6860 ret = btrfs_insert_empty_inode(trans, root, path, objectid);
6861 if (ret)
6862 goto out;
6863
6864 leaf = path->nodes[0];
6865 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
6866 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
6867 btrfs_set_inode_generation(leaf, item, 1);
6868 btrfs_set_inode_size(leaf, item, size);
6869 btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
6870 btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS);
6871 btrfs_mark_buffer_dirty(leaf);
6872 btrfs_release_path(root, path);
6873 out:
6874 btrfs_free_path(path);
6875 return ret;
6876 }
6877
6878 static noinline struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info,
6879 struct btrfs_block_group_cache *group)
6880 {
6881 struct inode *inode = NULL;
6882 struct btrfs_trans_handle *trans;
6883 struct btrfs_root *root;
6884 struct btrfs_key root_key;
6885 u64 objectid = BTRFS_FIRST_FREE_OBJECTID;
6886 int err = 0;
6887
6888 root_key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
6889 root_key.type = BTRFS_ROOT_ITEM_KEY;
6890 root_key.offset = (u64)-1;
6891 root = btrfs_read_fs_root_no_name(fs_info, &root_key);
6892 if (IS_ERR(root))
6893 return ERR_CAST(root);
6894
6895 trans = btrfs_start_transaction(root, 1);
6896 BUG_ON(!trans);
6897
6898 err = btrfs_find_free_objectid(trans, root, objectid, &objectid);
6899 if (err)
6900 goto out;
6901
6902 err = __insert_orphan_inode(trans, root, objectid, group->key.offset);
6903 BUG_ON(err);
6904
6905 err = btrfs_insert_file_extent(trans, root, objectid, 0, 0, 0,
6906 group->key.offset, 0, group->key.offset,
6907 0, 0, 0);
6908 BUG_ON(err);
6909
6910 inode = btrfs_iget_locked(root->fs_info->sb, objectid, root);
6911 if (inode->i_state & I_NEW) {
6912 BTRFS_I(inode)->root = root;
6913 BTRFS_I(inode)->location.objectid = objectid;
6914 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
6915 BTRFS_I(inode)->location.offset = 0;
6916 btrfs_read_locked_inode(inode);
6917 unlock_new_inode(inode);
6918 BUG_ON(is_bad_inode(inode));
6919 } else {
6920 BUG_ON(1);
6921 }
6922 BTRFS_I(inode)->index_cnt = group->key.objectid;
6923
6924 err = btrfs_orphan_add(trans, inode);
6925 out:
6926 btrfs_end_transaction(trans, root);
6927 if (err) {
6928 if (inode)
6929 iput(inode);
6930 inode = ERR_PTR(err);
6931 }
6932 return inode;
6933 }
6934
6935 int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len)
6936 {
6937
6938 struct btrfs_ordered_sum *sums;
6939 struct btrfs_sector_sum *sector_sum;
6940 struct btrfs_ordered_extent *ordered;
6941 struct btrfs_root *root = BTRFS_I(inode)->root;
6942 struct list_head list;
6943 size_t offset;
6944 int ret;
6945 u64 disk_bytenr;
6946
6947 INIT_LIST_HEAD(&list);
6948
6949 ordered = btrfs_lookup_ordered_extent(inode, file_pos);
6950 BUG_ON(ordered->file_offset != file_pos || ordered->len != len);
6951
6952 disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt;
6953 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, disk_bytenr,
6954 disk_bytenr + len - 1, &list);
6955
6956 while (!list_empty(&list)) {
6957 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
6958 list_del_init(&sums->list);
6959
6960 sector_sum = sums->sums;
6961 sums->bytenr = ordered->start;
6962
6963 offset = 0;
6964 while (offset < sums->len) {
6965 sector_sum->bytenr += ordered->start - disk_bytenr;
6966 sector_sum++;
6967 offset += root->sectorsize;
6968 }
6969
6970 btrfs_add_ordered_sum(inode, ordered, sums);
6971 }
6972 btrfs_put_ordered_extent(ordered);
6973 return 0;
6974 }
6975
6976 int btrfs_relocate_block_group(struct btrfs_root *root, u64 group_start)
6977 {
6978 struct btrfs_trans_handle *trans;
6979 struct btrfs_path *path;
6980 struct btrfs_fs_info *info = root->fs_info;
6981 struct extent_buffer *leaf;
6982 struct inode *reloc_inode;
6983 struct btrfs_block_group_cache *block_group;
6984 struct btrfs_key key;
6985 u64 skipped;
6986 u64 cur_byte;
6987 u64 total_found;
6988 u32 nritems;
6989 int ret;
6990 int progress;
6991 int pass = 0;
6992
6993 root = root->fs_info->extent_root;
6994
6995 block_group = btrfs_lookup_block_group(info, group_start);
6996 BUG_ON(!block_group);
6997
6998 printk(KERN_INFO "btrfs relocating block group %llu flags %llu\n",
6999 (unsigned long long)block_group->key.objectid,
7000 (unsigned long long)block_group->flags);
7001
7002 path = btrfs_alloc_path();
7003 BUG_ON(!path);
7004
7005 reloc_inode = create_reloc_inode(info, block_group);
7006 BUG_ON(IS_ERR(reloc_inode));
7007
7008 __alloc_chunk_for_shrink(root, block_group, 1);
7009 set_block_group_readonly(block_group);
7010
7011 btrfs_start_delalloc_inodes(info->tree_root);
7012 btrfs_wait_ordered_extents(info->tree_root, 0);
7013 again:
7014 skipped = 0;
7015 total_found = 0;
7016 progress = 0;
7017 key.objectid = block_group->key.objectid;
7018 key.offset = 0;
7019 key.type = 0;
7020 cur_byte = key.objectid;
7021
7022 trans = btrfs_start_transaction(info->tree_root, 1);
7023 btrfs_commit_transaction(trans, info->tree_root);
7024
7025 mutex_lock(&root->fs_info->cleaner_mutex);
7026 btrfs_clean_old_snapshots(info->tree_root);
7027 btrfs_remove_leaf_refs(info->tree_root, (u64)-1, 1);
7028 mutex_unlock(&root->fs_info->cleaner_mutex);
7029
7030 trans = btrfs_start_transaction(info->tree_root, 1);
7031 btrfs_commit_transaction(trans, info->tree_root);
7032
7033 while (1) {
7034 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7035 if (ret < 0)
7036 goto out;
7037 next:
7038 leaf = path->nodes[0];
7039 nritems = btrfs_header_nritems(leaf);
7040 if (path->slots[0] >= nritems) {
7041 ret = btrfs_next_leaf(root, path);
7042 if (ret < 0)
7043 goto out;
7044 if (ret == 1) {
7045 ret = 0;
7046 break;
7047 }
7048 leaf = path->nodes[0];
7049 nritems = btrfs_header_nritems(leaf);
7050 }
7051
7052 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
7053
7054 if (key.objectid >= block_group->key.objectid +
7055 block_group->key.offset)
7056 break;
7057
7058 if (progress && need_resched()) {
7059 btrfs_release_path(root, path);
7060 cond_resched();
7061 progress = 0;
7062 continue;
7063 }
7064 progress = 1;
7065
7066 if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY ||
7067 key.objectid + key.offset <= cur_byte) {
7068 path->slots[0]++;
7069 goto next;
7070 }
7071
7072 total_found++;
7073 cur_byte = key.objectid + key.offset;
7074 btrfs_release_path(root, path);
7075
7076 __alloc_chunk_for_shrink(root, block_group, 0);
7077 ret = relocate_one_extent(root, path, &key, block_group,
7078 reloc_inode, pass);
7079 BUG_ON(ret < 0);
7080 if (ret > 0)
7081 skipped++;
7082
7083 key.objectid = cur_byte;
7084 key.type = 0;
7085 key.offset = 0;
7086 }
7087
7088 btrfs_release_path(root, path);
7089
7090 if (pass == 0) {
7091 btrfs_wait_ordered_range(reloc_inode, 0, (u64)-1);
7092 invalidate_mapping_pages(reloc_inode->i_mapping, 0, -1);
7093 }
7094
7095 if (total_found > 0) {
7096 printk(KERN_INFO "btrfs found %llu extents in pass %d\n",
7097 (unsigned long long)total_found, pass);
7098 pass++;
7099 if (total_found == skipped && pass > 2) {
7100 iput(reloc_inode);
7101 reloc_inode = create_reloc_inode(info, block_group);
7102 pass = 0;
7103 }
7104 goto again;
7105 }
7106
7107 /* delete reloc_inode */
7108 iput(reloc_inode);
7109
7110 /* unpin extents in this range */
7111 trans = btrfs_start_transaction(info->tree_root, 1);
7112 btrfs_commit_transaction(trans, info->tree_root);
7113
7114 spin_lock(&block_group->lock);
7115 WARN_ON(block_group->pinned > 0);
7116 WARN_ON(block_group->reserved > 0);
7117 WARN_ON(btrfs_block_group_used(&block_group->item) > 0);
7118 spin_unlock(&block_group->lock);
7119 btrfs_put_block_group(block_group);
7120 ret = 0;
7121 out:
7122 btrfs_free_path(path);
7123 return ret;
7124 }
7125 #endif
7126
7127 static int find_first_block_group(struct btrfs_root *root,
7128 struct btrfs_path *path, struct btrfs_key *key)
7129 {
7130 int ret = 0;
7131 struct btrfs_key found_key;
7132 struct extent_buffer *leaf;
7133 int slot;
7134
7135 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
7136 if (ret < 0)
7137 goto out;
7138
7139 while (1) {
7140 slot = path->slots[0];
7141 leaf = path->nodes[0];
7142 if (slot >= btrfs_header_nritems(leaf)) {
7143 ret = btrfs_next_leaf(root, path);
7144 if (ret == 0)
7145 continue;
7146 if (ret < 0)
7147 goto out;
7148 break;
7149 }
7150 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7151
7152 if (found_key.objectid >= key->objectid &&
7153 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
7154 ret = 0;
7155 goto out;
7156 }
7157 path->slots[0]++;
7158 }
7159 ret = -ENOENT;
7160 out:
7161 return ret;
7162 }
7163
7164 int btrfs_free_block_groups(struct btrfs_fs_info *info)
7165 {
7166 struct btrfs_block_group_cache *block_group;
7167 struct btrfs_space_info *space_info;
7168 struct rb_node *n;
7169
7170 spin_lock(&info->block_group_cache_lock);
7171 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
7172 block_group = rb_entry(n, struct btrfs_block_group_cache,
7173 cache_node);
7174 rb_erase(&block_group->cache_node,
7175 &info->block_group_cache_tree);
7176 spin_unlock(&info->block_group_cache_lock);
7177
7178 down_write(&block_group->space_info->groups_sem);
7179 list_del(&block_group->list);
7180 up_write(&block_group->space_info->groups_sem);
7181
7182 if (block_group->cached == BTRFS_CACHE_STARTED)
7183 wait_event(block_group->caching_q,
7184 block_group_cache_done(block_group));
7185
7186 btrfs_remove_free_space_cache(block_group);
7187
7188 WARN_ON(atomic_read(&block_group->count) != 1);
7189 kfree(block_group);
7190
7191 spin_lock(&info->block_group_cache_lock);
7192 }
7193 spin_unlock(&info->block_group_cache_lock);
7194
7195 /* now that all the block groups are freed, go through and
7196 * free all the space_info structs. This is only called during
7197 * the final stages of unmount, and so we know nobody is
7198 * using them. We call synchronize_rcu() once before we start,
7199 * just to be on the safe side.
7200 */
7201 synchronize_rcu();
7202
7203 while(!list_empty(&info->space_info)) {
7204 space_info = list_entry(info->space_info.next,
7205 struct btrfs_space_info,
7206 list);
7207
7208 list_del(&space_info->list);
7209 kfree(space_info);
7210 }
7211 return 0;
7212 }
7213
7214 int btrfs_read_block_groups(struct btrfs_root *root)
7215 {
7216 struct btrfs_path *path;
7217 int ret;
7218 struct btrfs_block_group_cache *cache;
7219 struct btrfs_fs_info *info = root->fs_info;
7220 struct btrfs_space_info *space_info;
7221 struct btrfs_key key;
7222 struct btrfs_key found_key;
7223 struct extent_buffer *leaf;
7224
7225 root = info->extent_root;
7226 key.objectid = 0;
7227 key.offset = 0;
7228 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
7229 path = btrfs_alloc_path();
7230 if (!path)
7231 return -ENOMEM;
7232
7233 while (1) {
7234 ret = find_first_block_group(root, path, &key);
7235 if (ret > 0) {
7236 ret = 0;
7237 goto error;
7238 }
7239 if (ret != 0)
7240 goto error;
7241
7242 leaf = path->nodes[0];
7243 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7244 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7245 if (!cache) {
7246 ret = -ENOMEM;
7247 break;
7248 }
7249
7250 atomic_set(&cache->count, 1);
7251 spin_lock_init(&cache->lock);
7252 spin_lock_init(&cache->tree_lock);
7253 cache->fs_info = info;
7254 init_waitqueue_head(&cache->caching_q);
7255 INIT_LIST_HEAD(&cache->list);
7256 INIT_LIST_HEAD(&cache->cluster_list);
7257
7258 /*
7259 * we only want to have 32k of ram per block group for keeping
7260 * track of free space, and if we pass 1/2 of that we want to
7261 * start converting things over to using bitmaps
7262 */
7263 cache->extents_thresh = ((1024 * 32) / 2) /
7264 sizeof(struct btrfs_free_space);
7265
7266 read_extent_buffer(leaf, &cache->item,
7267 btrfs_item_ptr_offset(leaf, path->slots[0]),
7268 sizeof(cache->item));
7269 memcpy(&cache->key, &found_key, sizeof(found_key));
7270
7271 key.objectid = found_key.objectid + found_key.offset;
7272 btrfs_release_path(root, path);
7273 cache->flags = btrfs_block_group_flags(&cache->item);
7274 cache->sectorsize = root->sectorsize;
7275
7276 remove_sb_from_cache(root, cache);
7277
7278 /*
7279 * check for two cases, either we are full, and therefore
7280 * don't need to bother with the caching work since we won't
7281 * find any space, or we are empty, and we can just add all
7282 * the space in and be done with it. This saves us _alot_ of
7283 * time, particularly in the full case.
7284 */
7285 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
7286 cache->cached = BTRFS_CACHE_FINISHED;
7287 } else if (btrfs_block_group_used(&cache->item) == 0) {
7288 cache->cached = BTRFS_CACHE_FINISHED;
7289 add_new_free_space(cache, root->fs_info,
7290 found_key.objectid,
7291 found_key.objectid +
7292 found_key.offset);
7293 }
7294
7295 ret = update_space_info(info, cache->flags, found_key.offset,
7296 btrfs_block_group_used(&cache->item),
7297 &space_info);
7298 BUG_ON(ret);
7299 cache->space_info = space_info;
7300 down_write(&space_info->groups_sem);
7301 list_add_tail(&cache->list, &space_info->block_groups);
7302 up_write(&space_info->groups_sem);
7303
7304 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7305 BUG_ON(ret);
7306
7307 set_avail_alloc_bits(root->fs_info, cache->flags);
7308 if (btrfs_chunk_readonly(root, cache->key.objectid))
7309 set_block_group_readonly(cache);
7310 }
7311 ret = 0;
7312 error:
7313 btrfs_free_path(path);
7314 return ret;
7315 }
7316
7317 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
7318 struct btrfs_root *root, u64 bytes_used,
7319 u64 type, u64 chunk_objectid, u64 chunk_offset,
7320 u64 size)
7321 {
7322 int ret;
7323 struct btrfs_root *extent_root;
7324 struct btrfs_block_group_cache *cache;
7325
7326 extent_root = root->fs_info->extent_root;
7327
7328 root->fs_info->last_trans_log_full_commit = trans->transid;
7329
7330 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7331 if (!cache)
7332 return -ENOMEM;
7333
7334 cache->key.objectid = chunk_offset;
7335 cache->key.offset = size;
7336 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
7337 cache->sectorsize = root->sectorsize;
7338
7339 /*
7340 * we only want to have 32k of ram per block group for keeping track
7341 * of free space, and if we pass 1/2 of that we want to start
7342 * converting things over to using bitmaps
7343 */
7344 cache->extents_thresh = ((1024 * 32) / 2) /
7345 sizeof(struct btrfs_free_space);
7346 atomic_set(&cache->count, 1);
7347 spin_lock_init(&cache->lock);
7348 spin_lock_init(&cache->tree_lock);
7349 init_waitqueue_head(&cache->caching_q);
7350 INIT_LIST_HEAD(&cache->list);
7351 INIT_LIST_HEAD(&cache->cluster_list);
7352
7353 btrfs_set_block_group_used(&cache->item, bytes_used);
7354 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
7355 cache->flags = type;
7356 btrfs_set_block_group_flags(&cache->item, type);
7357
7358 cache->cached = BTRFS_CACHE_FINISHED;
7359 remove_sb_from_cache(root, cache);
7360
7361 add_new_free_space(cache, root->fs_info, chunk_offset,
7362 chunk_offset + size);
7363
7364 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
7365 &cache->space_info);
7366 BUG_ON(ret);
7367 down_write(&cache->space_info->groups_sem);
7368 list_add_tail(&cache->list, &cache->space_info->block_groups);
7369 up_write(&cache->space_info->groups_sem);
7370
7371 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7372 BUG_ON(ret);
7373
7374 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
7375 sizeof(cache->item));
7376 BUG_ON(ret);
7377
7378 set_avail_alloc_bits(extent_root->fs_info, type);
7379
7380 return 0;
7381 }
7382
7383 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
7384 struct btrfs_root *root, u64 group_start)
7385 {
7386 struct btrfs_path *path;
7387 struct btrfs_block_group_cache *block_group;
7388 struct btrfs_free_cluster *cluster;
7389 struct btrfs_key key;
7390 int ret;
7391
7392 root = root->fs_info->extent_root;
7393
7394 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
7395 BUG_ON(!block_group);
7396 BUG_ON(!block_group->ro);
7397
7398 memcpy(&key, &block_group->key, sizeof(key));
7399
7400 /* make sure this block group isn't part of an allocation cluster */
7401 cluster = &root->fs_info->data_alloc_cluster;
7402 spin_lock(&cluster->refill_lock);
7403 btrfs_return_cluster_to_free_space(block_group, cluster);
7404 spin_unlock(&cluster->refill_lock);
7405
7406 /*
7407 * make sure this block group isn't part of a metadata
7408 * allocation cluster
7409 */
7410 cluster = &root->fs_info->meta_alloc_cluster;
7411 spin_lock(&cluster->refill_lock);
7412 btrfs_return_cluster_to_free_space(block_group, cluster);
7413 spin_unlock(&cluster->refill_lock);
7414
7415 path = btrfs_alloc_path();
7416 BUG_ON(!path);
7417
7418 spin_lock(&root->fs_info->block_group_cache_lock);
7419 rb_erase(&block_group->cache_node,
7420 &root->fs_info->block_group_cache_tree);
7421 spin_unlock(&root->fs_info->block_group_cache_lock);
7422
7423 down_write(&block_group->space_info->groups_sem);
7424 /*
7425 * we must use list_del_init so people can check to see if they
7426 * are still on the list after taking the semaphore
7427 */
7428 list_del_init(&block_group->list);
7429 up_write(&block_group->space_info->groups_sem);
7430
7431 if (block_group->cached == BTRFS_CACHE_STARTED)
7432 wait_event(block_group->caching_q,
7433 block_group_cache_done(block_group));
7434
7435 btrfs_remove_free_space_cache(block_group);
7436
7437 spin_lock(&block_group->space_info->lock);
7438 block_group->space_info->total_bytes -= block_group->key.offset;
7439 block_group->space_info->bytes_readonly -= block_group->key.offset;
7440 spin_unlock(&block_group->space_info->lock);
7441
7442 btrfs_clear_space_info_full(root->fs_info);
7443
7444 btrfs_put_block_group(block_group);
7445 btrfs_put_block_group(block_group);
7446
7447 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
7448 if (ret > 0)
7449 ret = -EIO;
7450 if (ret < 0)
7451 goto out;
7452
7453 ret = btrfs_del_item(trans, root, path);
7454 out:
7455 btrfs_free_path(path);
7456 return ret;
7457 }
Linux with added FPC-III support