]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - fs/btrfs/extent-tree.c
projects / mirror_ubuntu-artful-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Btrfs: enforce metadata allocation clustering
[mirror_ubuntu-artful-kernel.git] / fs / btrfs / extent-tree.c
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 "hash.h"
23 #include "crc32c.h"
24 #include "ctree.h"
25 #include "disk-io.h"
26 #include "print-tree.h"
27 #include "transaction.h"
28 #include "volumes.h"
29 #include "locking.h"
30 #include "ref-cache.h"
31
32 #define PENDING_EXTENT_INSERT 0
33 #define PENDING_EXTENT_DELETE 1
34 #define PENDING_BACKREF_UPDATE 2
35
36 struct pending_extent_op {
37 int type;
38 u64 bytenr;
39 u64 num_bytes;
40 u64 parent;
41 u64 orig_parent;
42 u64 generation;
43 u64 orig_generation;
44 int level;
45 };
46
47 static int finish_current_insert(struct btrfs_trans_handle *trans, struct
48 btrfs_root *extent_root, int all);
49 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
50 btrfs_root *extent_root, int all);
51 static struct btrfs_block_group_cache *
52 __btrfs_find_block_group(struct btrfs_root *root,
53 struct btrfs_block_group_cache *hint,
54 u64 search_start, int data, int owner);
55
56 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
57 {
58 return (cache->flags & bits) == bits;
59 }
60
61 /*
62 * this adds the block group to the fs_info rb tree for the block group
63 * cache
64 */
65 int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
66 struct btrfs_block_group_cache *block_group)
67 {
68 struct rb_node **p;
69 struct rb_node *parent = NULL;
70 struct btrfs_block_group_cache *cache;
71
72 spin_lock(&info->block_group_cache_lock);
73 p = &info->block_group_cache_tree.rb_node;
74
75 while (*p) {
76 parent = *p;
77 cache = rb_entry(parent, struct btrfs_block_group_cache,
78 cache_node);
79 if (block_group->key.objectid < cache->key.objectid) {
80 p = &(*p)->rb_left;
81 } else if (block_group->key.objectid > cache->key.objectid) {
82 p = &(*p)->rb_right;
83 } else {
84 spin_unlock(&info->block_group_cache_lock);
85 return -EEXIST;
86 }
87 }
88
89 rb_link_node(&block_group->cache_node, parent, p);
90 rb_insert_color(&block_group->cache_node,
91 &info->block_group_cache_tree);
92 spin_unlock(&info->block_group_cache_lock);
93
94 return 0;
95 }
96
97 /*
98 * This will return the block group at or after bytenr if contains is 0, else
99 * it will return the block group that contains the bytenr
100 */
101 static struct btrfs_block_group_cache *
102 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
103 int contains)
104 {
105 struct btrfs_block_group_cache *cache, *ret = NULL;
106 struct rb_node *n;
107 u64 end, start;
108
109 spin_lock(&info->block_group_cache_lock);
110 n = info->block_group_cache_tree.rb_node;
111
112 while (n) {
113 cache = rb_entry(n, struct btrfs_block_group_cache,
114 cache_node);
115 end = cache->key.objectid + cache->key.offset - 1;
116 start = cache->key.objectid;
117
118 if (bytenr < start) {
119 if (!contains && (!ret || start < ret->key.objectid))
120 ret = cache;
121 n = n->rb_left;
122 } else if (bytenr > start) {
123 if (contains && bytenr <= end) {
124 ret = cache;
125 break;
126 }
127 n = n->rb_right;
128 } else {
129 ret = cache;
130 break;
131 }
132 }
133 spin_unlock(&info->block_group_cache_lock);
134
135 return ret;
136 }
137
138 /*
139 * this is only called by cache_block_group, since we could have freed extents
140 * we need to check the pinned_extents for any extents that can't be used yet
141 * since their free space will be released as soon as the transaction commits.
142 */
143 static int add_new_free_space(struct btrfs_block_group_cache *block_group,
144 struct btrfs_fs_info *info, u64 start, u64 end)
145 {
146 u64 extent_start, extent_end, size;
147 int ret;
148
149 mutex_lock(&info->pinned_mutex);
150 while (start < end) {
151 ret = find_first_extent_bit(&info->pinned_extents, start,
152 &extent_start, &extent_end,
153 EXTENT_DIRTY);
154 if (ret)
155 break;
156
157 if (extent_start == start) {
158 start = extent_end + 1;
159 } else if (extent_start > start && extent_start < end) {
160 size = extent_start - start;
161 ret = btrfs_add_free_space_lock(block_group, start,
162 size);
163 BUG_ON(ret);
164 start = extent_end + 1;
165 } else {
166 break;
167 }
168 }
169
170 if (start < end) {
171 size = end - start;
172 ret = btrfs_add_free_space_lock(block_group, start, size);
173 BUG_ON(ret);
174 }
175 mutex_unlock(&info->pinned_mutex);
176
177 return 0;
178 }
179
180 static int cache_block_group(struct btrfs_root *root,
181 struct btrfs_block_group_cache *block_group)
182 {
183 struct btrfs_path *path;
184 int ret = 0;
185 struct btrfs_key key;
186 struct extent_buffer *leaf;
187 int slot;
188 u64 last = 0;
189 u64 first_free;
190 int found = 0;
191
192 if (!block_group)
193 return 0;
194
195 root = root->fs_info->extent_root;
196
197 if (block_group->cached)
198 return 0;
199
200 path = btrfs_alloc_path();
201 if (!path)
202 return -ENOMEM;
203
204 path->reada = 2;
205 /*
206 * we get into deadlocks with paths held by callers of this function.
207 * since the alloc_mutex is protecting things right now, just
208 * skip the locking here
209 */
210 path->skip_locking = 1;
211 first_free = max_t(u64, block_group->key.objectid,
212 BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE);
213 key.objectid = block_group->key.objectid;
214 key.offset = 0;
215 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
216 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
217 if (ret < 0)
218 goto err;
219 ret = btrfs_previous_item(root, path, 0, BTRFS_EXTENT_ITEM_KEY);
220 if (ret < 0)
221 goto err;
222 if (ret == 0) {
223 leaf = path->nodes[0];
224 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
225 if (key.objectid + key.offset > first_free)
226 first_free = key.objectid + key.offset;
227 }
228 while(1) {
229 leaf = path->nodes[0];
230 slot = path->slots[0];
231 if (slot >= btrfs_header_nritems(leaf)) {
232 ret = btrfs_next_leaf(root, path);
233 if (ret < 0)
234 goto err;
235 if (ret == 0)
236 continue;
237 else
238 break;
239 }
240 btrfs_item_key_to_cpu(leaf, &key, slot);
241 if (key.objectid < block_group->key.objectid)
242 goto next;
243
244 if (key.objectid >= block_group->key.objectid +
245 block_group->key.offset)
246 break;
247
248 if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
249 if (!found) {
250 last = first_free;
251 found = 1;
252 }
253
254 add_new_free_space(block_group, root->fs_info, last,
255 key.objectid);
256
257 last = key.objectid + key.offset;
258 }
259 next:
260 path->slots[0]++;
261 }
262
263 if (!found)
264 last = first_free;
265
266 add_new_free_space(block_group, root->fs_info, last,
267 block_group->key.objectid +
268 block_group->key.offset);
269
270 block_group->cached = 1;
271 ret = 0;
272 err:
273 btrfs_free_path(path);
274 return ret;
275 }
276
277 /*
278 * return the block group that starts at or after bytenr
279 */
280 struct btrfs_block_group_cache *btrfs_lookup_first_block_group(struct
281 btrfs_fs_info *info,
282 u64 bytenr)
283 {
284 struct btrfs_block_group_cache *cache;
285
286 cache = block_group_cache_tree_search(info, bytenr, 0);
287
288 return cache;
289 }
290
291 /*
292 * return the block group that contains teh given bytenr
293 */
294 struct btrfs_block_group_cache *btrfs_lookup_block_group(struct
295 btrfs_fs_info *info,
296 u64 bytenr)
297 {
298 struct btrfs_block_group_cache *cache;
299
300 cache = block_group_cache_tree_search(info, bytenr, 1);
301
302 return cache;
303 }
304
305 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
306 u64 flags)
307 {
308 struct list_head *head = &info->space_info;
309 struct list_head *cur;
310 struct btrfs_space_info *found;
311 list_for_each(cur, head) {
312 found = list_entry(cur, struct btrfs_space_info, list);
313 if (found->flags == flags)
314 return found;
315 }
316 return NULL;
317 }
318
319 static u64 div_factor(u64 num, int factor)
320 {
321 if (factor == 10)
322 return num;
323 num *= factor;
324 do_div(num, 10);
325 return num;
326 }
327
328 static struct btrfs_block_group_cache *
329 __btrfs_find_block_group(struct btrfs_root *root,
330 struct btrfs_block_group_cache *hint,
331 u64 search_start, int data, int owner)
332 {
333 struct btrfs_block_group_cache *cache;
334 struct btrfs_block_group_cache *found_group = NULL;
335 struct btrfs_fs_info *info = root->fs_info;
336 u64 used;
337 u64 last = 0;
338 u64 free_check;
339 int full_search = 0;
340 int factor = 10;
341 int wrapped = 0;
342
343 if (data & BTRFS_BLOCK_GROUP_METADATA)
344 factor = 9;
345
346 if (search_start) {
347 struct btrfs_block_group_cache *shint;
348 shint = btrfs_lookup_first_block_group(info, search_start);
349 if (shint && block_group_bits(shint, data) && !shint->ro) {
350 spin_lock(&shint->lock);
351 used = btrfs_block_group_used(&shint->item);
352 if (used + shint->pinned + shint->reserved <
353 div_factor(shint->key.offset, factor)) {
354 spin_unlock(&shint->lock);
355 return shint;
356 }
357 spin_unlock(&shint->lock);
358 }
359 }
360 if (hint && !hint->ro && block_group_bits(hint, data)) {
361 spin_lock(&hint->lock);
362 used = btrfs_block_group_used(&hint->item);
363 if (used + hint->pinned + hint->reserved <
364 div_factor(hint->key.offset, factor)) {
365 spin_unlock(&hint->lock);
366 return hint;
367 }
368 spin_unlock(&hint->lock);
369 last = hint->key.objectid + hint->key.offset;
370 } else {
371 if (hint)
372 last = max(hint->key.objectid, search_start);
373 else
374 last = search_start;
375 }
376 again:
377 while (1) {
378 cache = btrfs_lookup_first_block_group(root->fs_info, last);
379 if (!cache)
380 break;
381
382 spin_lock(&cache->lock);
383 last = cache->key.objectid + cache->key.offset;
384 used = btrfs_block_group_used(&cache->item);
385
386 if (!cache->ro && block_group_bits(cache, data)) {
387 free_check = div_factor(cache->key.offset, factor);
388 if (used + cache->pinned + cache->reserved <
389 free_check) {
390 found_group = cache;
391 spin_unlock(&cache->lock);
392 goto found;
393 }
394 }
395 spin_unlock(&cache->lock);
396 cond_resched();
397 }
398 if (!wrapped) {
399 last = search_start;
400 wrapped = 1;
401 goto again;
402 }
403 if (!full_search && factor < 10) {
404 last = search_start;
405 full_search = 1;
406 factor = 10;
407 goto again;
408 }
409 found:
410 return found_group;
411 }
412
413 struct btrfs_block_group_cache *btrfs_find_block_group(struct btrfs_root *root,
414 struct btrfs_block_group_cache
415 *hint, u64 search_start,
416 int data, int owner)
417 {
418
419 struct btrfs_block_group_cache *ret;
420 ret = __btrfs_find_block_group(root, hint, search_start, data, owner);
421 return ret;
422 }
423
424 /* simple helper to search for an existing extent at a given offset */
425 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
426 {
427 int ret;
428 struct btrfs_key key;
429 struct btrfs_path *path;
430
431 path = btrfs_alloc_path();
432 BUG_ON(!path);
433 key.objectid = start;
434 key.offset = len;
435 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
436 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
437 0, 0);
438 btrfs_free_path(path);
439 return ret;
440 }
441
442 /*
443 * Back reference rules. Back refs have three main goals:
444 *
445 * 1) differentiate between all holders of references to an extent so that
446 * when a reference is dropped we can make sure it was a valid reference
447 * before freeing the extent.
448 *
449 * 2) Provide enough information to quickly find the holders of an extent
450 * if we notice a given block is corrupted or bad.
451 *
452 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
453 * maintenance. This is actually the same as #2, but with a slightly
454 * different use case.
455 *
456 * File extents can be referenced by:
457 *
458 * - multiple snapshots, subvolumes, or different generations in one subvol
459 * - different files inside a single subvolume
460 * - different offsets inside a file (bookend extents in file.c)
461 *
462 * The extent ref structure has fields for:
463 *
464 * - Objectid of the subvolume root
465 * - Generation number of the tree holding the reference
466 * - objectid of the file holding the reference
467 * - number of references holding by parent node (alway 1 for tree blocks)
468 *
469 * Btree leaf may hold multiple references to a file extent. In most cases,
470 * these references are from same file and the corresponding offsets inside
471 * the file are close together.
472 *
473 * When a file extent is allocated the fields are filled in:
474 * (root_key.objectid, trans->transid, inode objectid, 1)
475 *
476 * When a leaf is cow'd new references are added for every file extent found
477 * in the leaf. It looks similar to the create case, but trans->transid will
478 * be different when the block is cow'd.
479 *
480 * (root_key.objectid, trans->transid, inode objectid,
481 * number of references in the leaf)
482 *
483 * When a file extent is removed either during snapshot deletion or
484 * file truncation, we find the corresponding back reference and check
485 * the following fields:
486 *
487 * (btrfs_header_owner(leaf), btrfs_header_generation(leaf),
488 * inode objectid)
489 *
490 * Btree extents can be referenced by:
491 *
492 * - Different subvolumes
493 * - Different generations of the same subvolume
494 *
495 * When a tree block is created, back references are inserted:
496 *
497 * (root->root_key.objectid, trans->transid, level, 1)
498 *
499 * When a tree block is cow'd, new back references are added for all the
500 * blocks it points to. If the tree block isn't in reference counted root,
501 * the old back references are removed. These new back references are of
502 * the form (trans->transid will have increased since creation):
503 *
504 * (root->root_key.objectid, trans->transid, level, 1)
505 *
506 * When a backref is in deleting, the following fields are checked:
507 *
508 * if backref was for a tree root:
509 * (btrfs_header_owner(itself), btrfs_header_generation(itself), level)
510 * else
511 * (btrfs_header_owner(parent), btrfs_header_generation(parent), level)
512 *
513 * Back Reference Key composing:
514 *
515 * The key objectid corresponds to the first byte in the extent, the key
516 * type is set to BTRFS_EXTENT_REF_KEY, and the key offset is the first
517 * byte of parent extent. If a extent is tree root, the key offset is set
518 * to the key objectid.
519 */
520
521 static int noinline lookup_extent_backref(struct btrfs_trans_handle *trans,
522 struct btrfs_root *root,
523 struct btrfs_path *path,
524 u64 bytenr, u64 parent,
525 u64 ref_root, u64 ref_generation,
526 u64 owner_objectid, int del)
527 {
528 struct btrfs_key key;
529 struct btrfs_extent_ref *ref;
530 struct extent_buffer *leaf;
531 u64 ref_objectid;
532 int ret;
533
534 key.objectid = bytenr;
535 key.type = BTRFS_EXTENT_REF_KEY;
536 key.offset = parent;
537
538 ret = btrfs_search_slot(trans, root, &key, path, del ? -1 : 0, 1);
539 if (ret < 0)
540 goto out;
541 if (ret > 0) {
542 ret = -ENOENT;
543 goto out;
544 }
545
546 leaf = path->nodes[0];
547 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
548 ref_objectid = btrfs_ref_objectid(leaf, ref);
549 if (btrfs_ref_root(leaf, ref) != ref_root ||
550 btrfs_ref_generation(leaf, ref) != ref_generation ||
551 (ref_objectid != owner_objectid &&
552 ref_objectid != BTRFS_MULTIPLE_OBJECTIDS)) {
553 ret = -EIO;
554 WARN_ON(1);
555 goto out;
556 }
557 ret = 0;
558 out:
559 return ret;
560 }
561
562 static int noinline insert_extent_backref(struct btrfs_trans_handle *trans,
563 struct btrfs_root *root,
564 struct btrfs_path *path,
565 u64 bytenr, u64 parent,
566 u64 ref_root, u64 ref_generation,
567 u64 owner_objectid)
568 {
569 struct btrfs_key key;
570 struct extent_buffer *leaf;
571 struct btrfs_extent_ref *ref;
572 u32 num_refs;
573 int ret;
574
575 key.objectid = bytenr;
576 key.type = BTRFS_EXTENT_REF_KEY;
577 key.offset = parent;
578
579 ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*ref));
580 if (ret == 0) {
581 leaf = path->nodes[0];
582 ref = btrfs_item_ptr(leaf, path->slots[0],
583 struct btrfs_extent_ref);
584 btrfs_set_ref_root(leaf, ref, ref_root);
585 btrfs_set_ref_generation(leaf, ref, ref_generation);
586 btrfs_set_ref_objectid(leaf, ref, owner_objectid);
587 btrfs_set_ref_num_refs(leaf, ref, 1);
588 } else if (ret == -EEXIST) {
589 u64 existing_owner;
590 BUG_ON(owner_objectid < BTRFS_FIRST_FREE_OBJECTID);
591 leaf = path->nodes[0];
592 ref = btrfs_item_ptr(leaf, path->slots[0],
593 struct btrfs_extent_ref);
594 if (btrfs_ref_root(leaf, ref) != ref_root ||
595 btrfs_ref_generation(leaf, ref) != ref_generation) {
596 ret = -EIO;
597 WARN_ON(1);
598 goto out;
599 }
600
601 num_refs = btrfs_ref_num_refs(leaf, ref);
602 BUG_ON(num_refs == 0);
603 btrfs_set_ref_num_refs(leaf, ref, num_refs + 1);
604
605 existing_owner = btrfs_ref_objectid(leaf, ref);
606 if (existing_owner != owner_objectid &&
607 existing_owner != BTRFS_MULTIPLE_OBJECTIDS) {
608 btrfs_set_ref_objectid(leaf, ref,
609 BTRFS_MULTIPLE_OBJECTIDS);
610 }
611 ret = 0;
612 } else {
613 goto out;
614 }
615 btrfs_mark_buffer_dirty(path->nodes[0]);
616 out:
617 btrfs_release_path(root, path);
618 return ret;
619 }
620
621 static int noinline remove_extent_backref(struct btrfs_trans_handle *trans,
622 struct btrfs_root *root,
623 struct btrfs_path *path)
624 {
625 struct extent_buffer *leaf;
626 struct btrfs_extent_ref *ref;
627 u32 num_refs;
628 int ret = 0;
629
630 leaf = path->nodes[0];
631 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
632 num_refs = btrfs_ref_num_refs(leaf, ref);
633 BUG_ON(num_refs == 0);
634 num_refs -= 1;
635 if (num_refs == 0) {
636 ret = btrfs_del_item(trans, root, path);
637 } else {
638 btrfs_set_ref_num_refs(leaf, ref, num_refs);
639 btrfs_mark_buffer_dirty(leaf);
640 }
641 btrfs_release_path(root, path);
642 return ret;
643 }
644
645 static int __btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
646 struct btrfs_root *root, u64 bytenr,
647 u64 orig_parent, u64 parent,
648 u64 orig_root, u64 ref_root,
649 u64 orig_generation, u64 ref_generation,
650 u64 owner_objectid)
651 {
652 int ret;
653 struct btrfs_root *extent_root = root->fs_info->extent_root;
654 struct btrfs_path *path;
655
656 if (root == root->fs_info->extent_root) {
657 struct pending_extent_op *extent_op;
658 u64 num_bytes;
659
660 BUG_ON(owner_objectid >= BTRFS_MAX_LEVEL);
661 num_bytes = btrfs_level_size(root, (int)owner_objectid);
662 mutex_lock(&root->fs_info->extent_ins_mutex);
663 if (test_range_bit(&root->fs_info->extent_ins, bytenr,
664 bytenr + num_bytes - 1, EXTENT_WRITEBACK, 0)) {
665 u64 priv;
666 ret = get_state_private(&root->fs_info->extent_ins,
667 bytenr, &priv);
668 BUG_ON(ret);
669 extent_op = (struct pending_extent_op *)
670 (unsigned long)priv;
671 BUG_ON(extent_op->parent != orig_parent);
672 BUG_ON(extent_op->generation != orig_generation);
673
674 extent_op->parent = parent;
675 extent_op->generation = ref_generation;
676 } else {
677 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
678 BUG_ON(!extent_op);
679
680 extent_op->type = PENDING_BACKREF_UPDATE;
681 extent_op->bytenr = bytenr;
682 extent_op->num_bytes = num_bytes;
683 extent_op->parent = parent;
684 extent_op->orig_parent = orig_parent;
685 extent_op->generation = ref_generation;
686 extent_op->orig_generation = orig_generation;
687 extent_op->level = (int)owner_objectid;
688
689 set_extent_bits(&root->fs_info->extent_ins,
690 bytenr, bytenr + num_bytes - 1,
691 EXTENT_WRITEBACK, GFP_NOFS);
692 set_state_private(&root->fs_info->extent_ins,
693 bytenr, (unsigned long)extent_op);
694 }
695 mutex_unlock(&root->fs_info->extent_ins_mutex);
696 return 0;
697 }
698
699 path = btrfs_alloc_path();
700 if (!path)
701 return -ENOMEM;
702 ret = lookup_extent_backref(trans, extent_root, path,
703 bytenr, orig_parent, orig_root,
704 orig_generation, owner_objectid, 1);
705 if (ret)
706 goto out;
707 ret = remove_extent_backref(trans, extent_root, path);
708 if (ret)
709 goto out;
710 ret = insert_extent_backref(trans, extent_root, path, bytenr,
711 parent, ref_root, ref_generation,
712 owner_objectid);
713 BUG_ON(ret);
714 finish_current_insert(trans, extent_root, 0);
715 del_pending_extents(trans, extent_root, 0);
716 out:
717 btrfs_free_path(path);
718 return ret;
719 }
720
721 int btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
722 struct btrfs_root *root, u64 bytenr,
723 u64 orig_parent, u64 parent,
724 u64 ref_root, u64 ref_generation,
725 u64 owner_objectid)
726 {
727 int ret;
728 if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
729 owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
730 return 0;
731 ret = __btrfs_update_extent_ref(trans, root, bytenr, orig_parent,
732 parent, ref_root, ref_root,
733 ref_generation, ref_generation,
734 owner_objectid);
735 return ret;
736 }
737
738 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
739 struct btrfs_root *root, u64 bytenr,
740 u64 orig_parent, u64 parent,
741 u64 orig_root, u64 ref_root,
742 u64 orig_generation, u64 ref_generation,
743 u64 owner_objectid)
744 {
745 struct btrfs_path *path;
746 int ret;
747 struct btrfs_key key;
748 struct extent_buffer *l;
749 struct btrfs_extent_item *item;
750 u32 refs;
751
752 path = btrfs_alloc_path();
753 if (!path)
754 return -ENOMEM;
755
756 path->reada = 1;
757 key.objectid = bytenr;
758 key.type = BTRFS_EXTENT_ITEM_KEY;
759 key.offset = (u64)-1;
760
761 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
762 0, 1);
763 if (ret < 0)
764 return ret;
765 BUG_ON(ret == 0 || path->slots[0] == 0);
766
767 path->slots[0]--;
768 l = path->nodes[0];
769
770 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
771 if (key.objectid != bytenr) {
772 btrfs_print_leaf(root->fs_info->extent_root, path->nodes[0]);
773 printk("wanted %Lu found %Lu\n", bytenr, key.objectid);
774 BUG();
775 }
776 BUG_ON(key.type != BTRFS_EXTENT_ITEM_KEY);
777
778 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
779 refs = btrfs_extent_refs(l, item);
780 btrfs_set_extent_refs(l, item, refs + 1);
781 btrfs_mark_buffer_dirty(path->nodes[0]);
782
783 btrfs_release_path(root->fs_info->extent_root, path);
784
785 path->reada = 1;
786 ret = insert_extent_backref(trans, root->fs_info->extent_root,
787 path, bytenr, parent,
788 ref_root, ref_generation,
789 owner_objectid);
790 BUG_ON(ret);
791 finish_current_insert(trans, root->fs_info->extent_root, 0);
792 del_pending_extents(trans, root->fs_info->extent_root, 0);
793
794 btrfs_free_path(path);
795 return 0;
796 }
797
798 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
799 struct btrfs_root *root,
800 u64 bytenr, u64 num_bytes, u64 parent,
801 u64 ref_root, u64 ref_generation,
802 u64 owner_objectid)
803 {
804 int ret;
805 if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
806 owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
807 return 0;
808 ret = __btrfs_inc_extent_ref(trans, root, bytenr, 0, parent,
809 0, ref_root, 0, ref_generation,
810 owner_objectid);
811 return ret;
812 }
813
814 int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
815 struct btrfs_root *root)
816 {
817 finish_current_insert(trans, root->fs_info->extent_root, 1);
818 del_pending_extents(trans, root->fs_info->extent_root, 1);
819 return 0;
820 }
821
822 int btrfs_lookup_extent_ref(struct btrfs_trans_handle *trans,
823 struct btrfs_root *root, u64 bytenr,
824 u64 num_bytes, u32 *refs)
825 {
826 struct btrfs_path *path;
827 int ret;
828 struct btrfs_key key;
829 struct extent_buffer *l;
830 struct btrfs_extent_item *item;
831
832 WARN_ON(num_bytes < root->sectorsize);
833 path = btrfs_alloc_path();
834 path->reada = 1;
835 key.objectid = bytenr;
836 key.offset = num_bytes;
837 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
838 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
839 0, 0);
840 if (ret < 0)
841 goto out;
842 if (ret != 0) {
843 btrfs_print_leaf(root, path->nodes[0]);
844 printk("failed to find block number %Lu\n", bytenr);
845 BUG();
846 }
847 l = path->nodes[0];
848 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
849 *refs = btrfs_extent_refs(l, item);
850 out:
851 btrfs_free_path(path);
852 return 0;
853 }
854
855 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
856 struct btrfs_root *root, u64 bytenr)
857 {
858 struct btrfs_root *extent_root = root->fs_info->extent_root;
859 struct btrfs_path *path;
860 struct extent_buffer *leaf;
861 struct btrfs_extent_ref *ref_item;
862 struct btrfs_key key;
863 struct btrfs_key found_key;
864 u64 ref_root;
865 u64 last_snapshot;
866 u32 nritems;
867 int ret;
868
869 key.objectid = bytenr;
870 key.offset = (u64)-1;
871 key.type = BTRFS_EXTENT_ITEM_KEY;
872
873 path = btrfs_alloc_path();
874 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
875 if (ret < 0)
876 goto out;
877 BUG_ON(ret == 0);
878
879 ret = -ENOENT;
880 if (path->slots[0] == 0)
881 goto out;
882
883 path->slots[0]--;
884 leaf = path->nodes[0];
885 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
886
887 if (found_key.objectid != bytenr ||
888 found_key.type != BTRFS_EXTENT_ITEM_KEY)
889 goto out;
890
891 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
892 while (1) {
893 leaf = path->nodes[0];
894 nritems = btrfs_header_nritems(leaf);
895 if (path->slots[0] >= nritems) {
896 ret = btrfs_next_leaf(extent_root, path);
897 if (ret < 0)
898 goto out;
899 if (ret == 0)
900 continue;
901 break;
902 }
903 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
904 if (found_key.objectid != bytenr)
905 break;
906
907 if (found_key.type != BTRFS_EXTENT_REF_KEY) {
908 path->slots[0]++;
909 continue;
910 }
911
912 ref_item = btrfs_item_ptr(leaf, path->slots[0],
913 struct btrfs_extent_ref);
914 ref_root = btrfs_ref_root(leaf, ref_item);
915 if (ref_root != root->root_key.objectid &&
916 ref_root != BTRFS_TREE_LOG_OBJECTID) {
917 ret = 1;
918 goto out;
919 }
920 if (btrfs_ref_generation(leaf, ref_item) <= last_snapshot) {
921 ret = 1;
922 goto out;
923 }
924
925 path->slots[0]++;
926 }
927 ret = 0;
928 out:
929 btrfs_free_path(path);
930 return ret;
931 }
932
933 int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
934 struct extent_buffer *buf, u32 nr_extents)
935 {
936 struct btrfs_key key;
937 struct btrfs_file_extent_item *fi;
938 u64 root_gen;
939 u32 nritems;
940 int i;
941 int level;
942 int ret = 0;
943 int shared = 0;
944
945 if (!root->ref_cows)
946 return 0;
947
948 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
949 shared = 0;
950 root_gen = root->root_key.offset;
951 } else {
952 shared = 1;
953 root_gen = trans->transid - 1;
954 }
955
956 level = btrfs_header_level(buf);
957 nritems = btrfs_header_nritems(buf);
958
959 if (level == 0) {
960 struct btrfs_leaf_ref *ref;
961 struct btrfs_extent_info *info;
962
963 ref = btrfs_alloc_leaf_ref(root, nr_extents);
964 if (!ref) {
965 ret = -ENOMEM;
966 goto out;
967 }
968
969 ref->root_gen = root_gen;
970 ref->bytenr = buf->start;
971 ref->owner = btrfs_header_owner(buf);
972 ref->generation = btrfs_header_generation(buf);
973 ref->nritems = nr_extents;
974 info = ref->extents;
975
976 for (i = 0; nr_extents > 0 && i < nritems; i++) {
977 u64 disk_bytenr;
978 btrfs_item_key_to_cpu(buf, &key, i);
979 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
980 continue;
981 fi = btrfs_item_ptr(buf, i,
982 struct btrfs_file_extent_item);
983 if (btrfs_file_extent_type(buf, fi) ==
984 BTRFS_FILE_EXTENT_INLINE)
985 continue;
986 disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
987 if (disk_bytenr == 0)
988 continue;
989
990 info->bytenr = disk_bytenr;
991 info->num_bytes =
992 btrfs_file_extent_disk_num_bytes(buf, fi);
993 info->objectid = key.objectid;
994 info->offset = key.offset;
995 info++;
996 }
997
998 ret = btrfs_add_leaf_ref(root, ref, shared);
999 if (ret == -EEXIST && shared) {
1000 struct btrfs_leaf_ref *old;
1001 old = btrfs_lookup_leaf_ref(root, ref->bytenr);
1002 BUG_ON(!old);
1003 btrfs_remove_leaf_ref(root, old);
1004 btrfs_free_leaf_ref(root, old);
1005 ret = btrfs_add_leaf_ref(root, ref, shared);
1006 }
1007 WARN_ON(ret);
1008 btrfs_free_leaf_ref(root, ref);
1009 }
1010 out:
1011 return ret;
1012 }
1013
1014 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1015 struct extent_buffer *orig_buf, struct extent_buffer *buf,
1016 u32 *nr_extents)
1017 {
1018 u64 bytenr;
1019 u64 ref_root;
1020 u64 orig_root;
1021 u64 ref_generation;
1022 u64 orig_generation;
1023 u32 nritems;
1024 u32 nr_file_extents = 0;
1025 struct btrfs_key key;
1026 struct btrfs_file_extent_item *fi;
1027 int i;
1028 int level;
1029 int ret = 0;
1030 int faili = 0;
1031 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
1032 u64, u64, u64, u64, u64, u64, u64, u64);
1033
1034 ref_root = btrfs_header_owner(buf);
1035 ref_generation = btrfs_header_generation(buf);
1036 orig_root = btrfs_header_owner(orig_buf);
1037 orig_generation = btrfs_header_generation(orig_buf);
1038
1039 nritems = btrfs_header_nritems(buf);
1040 level = btrfs_header_level(buf);
1041
1042 if (root->ref_cows) {
1043 process_func = __btrfs_inc_extent_ref;
1044 } else {
1045 if (level == 0 &&
1046 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
1047 goto out;
1048 if (level != 0 &&
1049 root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
1050 goto out;
1051 process_func = __btrfs_update_extent_ref;
1052 }
1053
1054 for (i = 0; i < nritems; i++) {
1055 cond_resched();
1056 if (level == 0) {
1057 btrfs_item_key_to_cpu(buf, &key, i);
1058 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1059 continue;
1060 fi = btrfs_item_ptr(buf, i,
1061 struct btrfs_file_extent_item);
1062 if (btrfs_file_extent_type(buf, fi) ==
1063 BTRFS_FILE_EXTENT_INLINE)
1064 continue;
1065 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1066 if (bytenr == 0)
1067 continue;
1068
1069 nr_file_extents++;
1070
1071 ret = process_func(trans, root, bytenr,
1072 orig_buf->start, buf->start,
1073 orig_root, ref_root,
1074 orig_generation, ref_generation,
1075 key.objectid);
1076
1077 if (ret) {
1078 faili = i;
1079 WARN_ON(1);
1080 goto fail;
1081 }
1082 } else {
1083 bytenr = btrfs_node_blockptr(buf, i);
1084 ret = process_func(trans, root, bytenr,
1085 orig_buf->start, buf->start,
1086 orig_root, ref_root,
1087 orig_generation, ref_generation,
1088 level - 1);
1089 if (ret) {
1090 faili = i;
1091 WARN_ON(1);
1092 goto fail;
1093 }
1094 }
1095 }
1096 out:
1097 if (nr_extents) {
1098 if (level == 0)
1099 *nr_extents = nr_file_extents;
1100 else
1101 *nr_extents = nritems;
1102 }
1103 return 0;
1104 fail:
1105 WARN_ON(1);
1106 return ret;
1107 }
1108
1109 int btrfs_update_ref(struct btrfs_trans_handle *trans,
1110 struct btrfs_root *root, struct extent_buffer *orig_buf,
1111 struct extent_buffer *buf, int start_slot, int nr)
1112
1113 {
1114 u64 bytenr;
1115 u64 ref_root;
1116 u64 orig_root;
1117 u64 ref_generation;
1118 u64 orig_generation;
1119 struct btrfs_key key;
1120 struct btrfs_file_extent_item *fi;
1121 int i;
1122 int ret;
1123 int slot;
1124 int level;
1125
1126 BUG_ON(start_slot < 0);
1127 BUG_ON(start_slot + nr > btrfs_header_nritems(buf));
1128
1129 ref_root = btrfs_header_owner(buf);
1130 ref_generation = btrfs_header_generation(buf);
1131 orig_root = btrfs_header_owner(orig_buf);
1132 orig_generation = btrfs_header_generation(orig_buf);
1133 level = btrfs_header_level(buf);
1134
1135 if (!root->ref_cows) {
1136 if (level == 0 &&
1137 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
1138 return 0;
1139 if (level != 0 &&
1140 root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
1141 return 0;
1142 }
1143
1144 for (i = 0, slot = start_slot; i < nr; i++, slot++) {
1145 cond_resched();
1146 if (level == 0) {
1147 btrfs_item_key_to_cpu(buf, &key, slot);
1148 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1149 continue;
1150 fi = btrfs_item_ptr(buf, slot,
1151 struct btrfs_file_extent_item);
1152 if (btrfs_file_extent_type(buf, fi) ==
1153 BTRFS_FILE_EXTENT_INLINE)
1154 continue;
1155 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1156 if (bytenr == 0)
1157 continue;
1158 ret = __btrfs_update_extent_ref(trans, root, bytenr,
1159 orig_buf->start, buf->start,
1160 orig_root, ref_root,
1161 orig_generation, ref_generation,
1162 key.objectid);
1163 if (ret)
1164 goto fail;
1165 } else {
1166 bytenr = btrfs_node_blockptr(buf, slot);
1167 ret = __btrfs_update_extent_ref(trans, root, bytenr,
1168 orig_buf->start, buf->start,
1169 orig_root, ref_root,
1170 orig_generation, ref_generation,
1171 level - 1);
1172 if (ret)
1173 goto fail;
1174 }
1175 }
1176 return 0;
1177 fail:
1178 WARN_ON(1);
1179 return -1;
1180 }
1181
1182 static int write_one_cache_group(struct btrfs_trans_handle *trans,
1183 struct btrfs_root *root,
1184 struct btrfs_path *path,
1185 struct btrfs_block_group_cache *cache)
1186 {
1187 int ret;
1188 int pending_ret;
1189 struct btrfs_root *extent_root = root->fs_info->extent_root;
1190 unsigned long bi;
1191 struct extent_buffer *leaf;
1192
1193 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
1194 if (ret < 0)
1195 goto fail;
1196 BUG_ON(ret);
1197
1198 leaf = path->nodes[0];
1199 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
1200 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
1201 btrfs_mark_buffer_dirty(leaf);
1202 btrfs_release_path(extent_root, path);
1203 fail:
1204 finish_current_insert(trans, extent_root, 0);
1205 pending_ret = del_pending_extents(trans, extent_root, 0);
1206 if (ret)
1207 return ret;
1208 if (pending_ret)
1209 return pending_ret;
1210 return 0;
1211
1212 }
1213
1214 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
1215 struct btrfs_root *root)
1216 {
1217 struct btrfs_block_group_cache *cache, *entry;
1218 struct rb_node *n;
1219 int err = 0;
1220 int werr = 0;
1221 struct btrfs_path *path;
1222 u64 last = 0;
1223
1224 path = btrfs_alloc_path();
1225 if (!path)
1226 return -ENOMEM;
1227
1228 while(1) {
1229 cache = NULL;
1230 spin_lock(&root->fs_info->block_group_cache_lock);
1231 for (n = rb_first(&root->fs_info->block_group_cache_tree);
1232 n; n = rb_next(n)) {
1233 entry = rb_entry(n, struct btrfs_block_group_cache,
1234 cache_node);
1235 if (entry->dirty) {
1236 cache = entry;
1237 break;
1238 }
1239 }
1240 spin_unlock(&root->fs_info->block_group_cache_lock);
1241
1242 if (!cache)
1243 break;
1244
1245 cache->dirty = 0;
1246 last += cache->key.offset;
1247
1248 err = write_one_cache_group(trans, root,
1249 path, cache);
1250 /*
1251 * if we fail to write the cache group, we want
1252 * to keep it marked dirty in hopes that a later
1253 * write will work
1254 */
1255 if (err) {
1256 werr = err;
1257 continue;
1258 }
1259 }
1260 btrfs_free_path(path);
1261 return werr;
1262 }
1263
1264 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
1265 u64 total_bytes, u64 bytes_used,
1266 struct btrfs_space_info **space_info)
1267 {
1268 struct btrfs_space_info *found;
1269
1270 found = __find_space_info(info, flags);
1271 if (found) {
1272 spin_lock(&found->lock);
1273 found->total_bytes += total_bytes;
1274 found->bytes_used += bytes_used;
1275 found->full = 0;
1276 spin_unlock(&found->lock);
1277 *space_info = found;
1278 return 0;
1279 }
1280 found = kmalloc(sizeof(*found), GFP_NOFS);
1281 if (!found)
1282 return -ENOMEM;
1283
1284 list_add(&found->list, &info->space_info);
1285 INIT_LIST_HEAD(&found->block_groups);
1286 init_rwsem(&found->groups_sem);
1287 spin_lock_init(&found->lock);
1288 found->flags = flags;
1289 found->total_bytes = total_bytes;
1290 found->bytes_used = bytes_used;
1291 found->bytes_pinned = 0;
1292 found->bytes_reserved = 0;
1293 found->full = 0;
1294 found->force_alloc = 0;
1295 *space_info = found;
1296 return 0;
1297 }
1298
1299 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
1300 {
1301 u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
1302 BTRFS_BLOCK_GROUP_RAID1 |
1303 BTRFS_BLOCK_GROUP_RAID10 |
1304 BTRFS_BLOCK_GROUP_DUP);
1305 if (extra_flags) {
1306 if (flags & BTRFS_BLOCK_GROUP_DATA)
1307 fs_info->avail_data_alloc_bits |= extra_flags;
1308 if (flags & BTRFS_BLOCK_GROUP_METADATA)
1309 fs_info->avail_metadata_alloc_bits |= extra_flags;
1310 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
1311 fs_info->avail_system_alloc_bits |= extra_flags;
1312 }
1313 }
1314
1315 static u64 reduce_alloc_profile(struct btrfs_root *root, u64 flags)
1316 {
1317 u64 num_devices = root->fs_info->fs_devices->num_devices;
1318
1319 if (num_devices == 1)
1320 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
1321 if (num_devices < 4)
1322 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
1323
1324 if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
1325 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
1326 BTRFS_BLOCK_GROUP_RAID10))) {
1327 flags &= ~BTRFS_BLOCK_GROUP_DUP;
1328 }
1329
1330 if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
1331 (flags & BTRFS_BLOCK_GROUP_RAID10)) {
1332 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
1333 }
1334
1335 if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
1336 ((flags & BTRFS_BLOCK_GROUP_RAID1) |
1337 (flags & BTRFS_BLOCK_GROUP_RAID10) |
1338 (flags & BTRFS_BLOCK_GROUP_DUP)))
1339 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
1340 return flags;
1341 }
1342
1343 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
1344 struct btrfs_root *extent_root, u64 alloc_bytes,
1345 u64 flags, int force)
1346 {
1347 struct btrfs_space_info *space_info;
1348 u64 thresh;
1349 u64 start;
1350 u64 num_bytes;
1351 int ret = 0, waited = 0;
1352
1353 flags = reduce_alloc_profile(extent_root, flags);
1354
1355 space_info = __find_space_info(extent_root->fs_info, flags);
1356 if (!space_info) {
1357 ret = update_space_info(extent_root->fs_info, flags,
1358 0, 0, &space_info);
1359 BUG_ON(ret);
1360 }
1361 BUG_ON(!space_info);
1362
1363 spin_lock(&space_info->lock);
1364 if (space_info->force_alloc) {
1365 force = 1;
1366 space_info->force_alloc = 0;
1367 }
1368 if (space_info->full) {
1369 spin_unlock(&space_info->lock);
1370 goto out;
1371 }
1372
1373 thresh = div_factor(space_info->total_bytes, 6);
1374 if (!force &&
1375 (space_info->bytes_used + space_info->bytes_pinned +
1376 space_info->bytes_reserved + alloc_bytes) < thresh) {
1377 spin_unlock(&space_info->lock);
1378 goto out;
1379 }
1380
1381 spin_unlock(&space_info->lock);
1382
1383 ret = mutex_trylock(&extent_root->fs_info->chunk_mutex);
1384 if (!ret && !force) {
1385 goto out;
1386 } else if (!ret) {
1387 mutex_lock(&extent_root->fs_info->chunk_mutex);
1388 waited = 1;
1389 }
1390
1391 if (waited) {
1392 spin_lock(&space_info->lock);
1393 if (space_info->full) {
1394 spin_unlock(&space_info->lock);
1395 goto out_unlock;
1396 }
1397 spin_unlock(&space_info->lock);
1398 }
1399
1400 ret = btrfs_alloc_chunk(trans, extent_root, &start, &num_bytes, flags);
1401 if (ret) {
1402 printk("space info full %Lu\n", flags);
1403 space_info->full = 1;
1404 goto out_unlock;
1405 }
1406
1407 ret = btrfs_make_block_group(trans, extent_root, 0, flags,
1408 BTRFS_FIRST_CHUNK_TREE_OBJECTID, start, num_bytes);
1409 BUG_ON(ret);
1410 out_unlock:
1411 mutex_unlock(&extent_root->fs_info->chunk_mutex);
1412 out:
1413 return ret;
1414 }
1415
1416 static int update_block_group(struct btrfs_trans_handle *trans,
1417 struct btrfs_root *root,
1418 u64 bytenr, u64 num_bytes, int alloc,
1419 int mark_free)
1420 {
1421 struct btrfs_block_group_cache *cache;
1422 struct btrfs_fs_info *info = root->fs_info;
1423 u64 total = num_bytes;
1424 u64 old_val;
1425 u64 byte_in_group;
1426
1427 while(total) {
1428 cache = btrfs_lookup_block_group(info, bytenr);
1429 if (!cache) {
1430 return -1;
1431 }
1432 byte_in_group = bytenr - cache->key.objectid;
1433 WARN_ON(byte_in_group > cache->key.offset);
1434
1435 spin_lock(&cache->space_info->lock);
1436 spin_lock(&cache->lock);
1437 cache->dirty = 1;
1438 old_val = btrfs_block_group_used(&cache->item);
1439 num_bytes = min(total, cache->key.offset - byte_in_group);
1440 if (alloc) {
1441 old_val += num_bytes;
1442 cache->space_info->bytes_used += num_bytes;
1443 btrfs_set_block_group_used(&cache->item, old_val);
1444 spin_unlock(&cache->lock);
1445 spin_unlock(&cache->space_info->lock);
1446 } else {
1447 old_val -= num_bytes;
1448 cache->space_info->bytes_used -= num_bytes;
1449 btrfs_set_block_group_used(&cache->item, old_val);
1450 spin_unlock(&cache->lock);
1451 spin_unlock(&cache->space_info->lock);
1452 if (mark_free) {
1453 int ret;
1454 ret = btrfs_add_free_space(cache, bytenr,
1455 num_bytes);
1456 if (ret)
1457 return -1;
1458 }
1459 }
1460 total -= num_bytes;
1461 bytenr += num_bytes;
1462 }
1463 return 0;
1464 }
1465
1466 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
1467 {
1468 struct btrfs_block_group_cache *cache;
1469
1470 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
1471 if (!cache)
1472 return 0;
1473
1474 return cache->key.objectid;
1475 }
1476
1477 int btrfs_update_pinned_extents(struct btrfs_root *root,
1478 u64 bytenr, u64 num, int pin)
1479 {
1480 u64 len;
1481 struct btrfs_block_group_cache *cache;
1482 struct btrfs_fs_info *fs_info = root->fs_info;
1483
1484 WARN_ON(!mutex_is_locked(&root->fs_info->pinned_mutex));
1485 if (pin) {
1486 set_extent_dirty(&fs_info->pinned_extents,
1487 bytenr, bytenr + num - 1, GFP_NOFS);
1488 } else {
1489 clear_extent_dirty(&fs_info->pinned_extents,
1490 bytenr, bytenr + num - 1, GFP_NOFS);
1491 }
1492 while (num > 0) {
1493 cache = btrfs_lookup_block_group(fs_info, bytenr);
1494 BUG_ON(!cache);
1495 len = min(num, cache->key.offset -
1496 (bytenr - cache->key.objectid));
1497 if (pin) {
1498 spin_lock(&cache->space_info->lock);
1499 spin_lock(&cache->lock);
1500 cache->pinned += len;
1501 cache->space_info->bytes_pinned += len;
1502 spin_unlock(&cache->lock);
1503 spin_unlock(&cache->space_info->lock);
1504 fs_info->total_pinned += len;
1505 } else {
1506 spin_lock(&cache->space_info->lock);
1507 spin_lock(&cache->lock);
1508 cache->pinned -= len;
1509 cache->space_info->bytes_pinned -= len;
1510 spin_unlock(&cache->lock);
1511 spin_unlock(&cache->space_info->lock);
1512 fs_info->total_pinned -= len;
1513 }
1514 bytenr += len;
1515 num -= len;
1516 }
1517 return 0;
1518 }
1519
1520 static int update_reserved_extents(struct btrfs_root *root,
1521 u64 bytenr, u64 num, int reserve)
1522 {
1523 u64 len;
1524 struct btrfs_block_group_cache *cache;
1525 struct btrfs_fs_info *fs_info = root->fs_info;
1526
1527 while (num > 0) {
1528 cache = btrfs_lookup_block_group(fs_info, bytenr);
1529 BUG_ON(!cache);
1530 len = min(num, cache->key.offset -
1531 (bytenr - cache->key.objectid));
1532
1533 spin_lock(&cache->space_info->lock);
1534 spin_lock(&cache->lock);
1535 if (reserve) {
1536 cache->reserved += len;
1537 cache->space_info->bytes_reserved += len;
1538 } else {
1539 cache->reserved -= len;
1540 cache->space_info->bytes_reserved -= len;
1541 }
1542 spin_unlock(&cache->lock);
1543 spin_unlock(&cache->space_info->lock);
1544 bytenr += len;
1545 num -= len;
1546 }
1547 return 0;
1548 }
1549
1550 int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
1551 {
1552 u64 last = 0;
1553 u64 start;
1554 u64 end;
1555 struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
1556 int ret;
1557
1558 mutex_lock(&root->fs_info->pinned_mutex);
1559 while(1) {
1560 ret = find_first_extent_bit(pinned_extents, last,
1561 &start, &end, EXTENT_DIRTY);
1562 if (ret)
1563 break;
1564 set_extent_dirty(copy, start, end, GFP_NOFS);
1565 last = end + 1;
1566 }
1567 mutex_unlock(&root->fs_info->pinned_mutex);
1568 return 0;
1569 }
1570
1571 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
1572 struct btrfs_root *root,
1573 struct extent_io_tree *unpin)
1574 {
1575 u64 start;
1576 u64 end;
1577 int ret;
1578 struct btrfs_block_group_cache *cache;
1579
1580 mutex_lock(&root->fs_info->pinned_mutex);
1581 while(1) {
1582 ret = find_first_extent_bit(unpin, 0, &start, &end,
1583 EXTENT_DIRTY);
1584 if (ret)
1585 break;
1586 btrfs_update_pinned_extents(root, start, end + 1 - start, 0);
1587 clear_extent_dirty(unpin, start, end, GFP_NOFS);
1588 cache = btrfs_lookup_block_group(root->fs_info, start);
1589 if (cache->cached)
1590 btrfs_add_free_space(cache, start, end - start + 1);
1591 if (need_resched()) {
1592 mutex_unlock(&root->fs_info->pinned_mutex);
1593 cond_resched();
1594 mutex_lock(&root->fs_info->pinned_mutex);
1595 }
1596 }
1597 mutex_unlock(&root->fs_info->pinned_mutex);
1598 return 0;
1599 }
1600
1601 static int finish_current_insert(struct btrfs_trans_handle *trans,
1602 struct btrfs_root *extent_root, int all)
1603 {
1604 u64 start;
1605 u64 end;
1606 u64 priv;
1607 u64 search = 0;
1608 struct btrfs_fs_info *info = extent_root->fs_info;
1609 struct btrfs_path *path;
1610 struct btrfs_extent_ref *ref;
1611 struct pending_extent_op *extent_op;
1612 struct btrfs_key key;
1613 struct btrfs_extent_item extent_item;
1614 int ret;
1615 int err = 0;
1616
1617 btrfs_set_stack_extent_refs(&extent_item, 1);
1618 path = btrfs_alloc_path();
1619
1620 while(1) {
1621 mutex_lock(&info->extent_ins_mutex);
1622 ret = find_first_extent_bit(&info->extent_ins, search, &start,
1623 &end, EXTENT_WRITEBACK);
1624 if (ret) {
1625 mutex_unlock(&info->extent_ins_mutex);
1626 if (search && all) {
1627 search = 0;
1628 continue;
1629 }
1630 break;
1631 }
1632
1633 ret = try_lock_extent(&info->extent_ins, start, end, GFP_NOFS);
1634 if (!ret) {
1635 search = end + 1;
1636 mutex_unlock(&info->extent_ins_mutex);
1637 cond_resched();
1638 continue;
1639 }
1640 BUG_ON(ret < 0);
1641
1642 ret = get_state_private(&info->extent_ins, start, &priv);
1643 BUG_ON(ret);
1644 extent_op = (struct pending_extent_op *)(unsigned long)priv;
1645
1646 mutex_unlock(&info->extent_ins_mutex);
1647
1648 if (extent_op->type == PENDING_EXTENT_INSERT) {
1649 key.objectid = start;
1650 key.offset = end + 1 - start;
1651 key.type = BTRFS_EXTENT_ITEM_KEY;
1652 err = btrfs_insert_item(trans, extent_root, &key,
1653 &extent_item, sizeof(extent_item));
1654 BUG_ON(err);
1655
1656 mutex_lock(&info->extent_ins_mutex);
1657 clear_extent_bits(&info->extent_ins, start, end,
1658 EXTENT_WRITEBACK, GFP_NOFS);
1659 mutex_unlock(&info->extent_ins_mutex);
1660
1661 err = insert_extent_backref(trans, extent_root, path,
1662 start, extent_op->parent,
1663 extent_root->root_key.objectid,
1664 extent_op->generation,
1665 extent_op->level);
1666 BUG_ON(err);
1667 } else if (extent_op->type == PENDING_BACKREF_UPDATE) {
1668 err = lookup_extent_backref(trans, extent_root, path,
1669 start, extent_op->orig_parent,
1670 extent_root->root_key.objectid,
1671 extent_op->orig_generation,
1672 extent_op->level, 0);
1673 BUG_ON(err);
1674
1675 mutex_lock(&info->extent_ins_mutex);
1676 clear_extent_bits(&info->extent_ins, start, end,
1677 EXTENT_WRITEBACK, GFP_NOFS);
1678 mutex_unlock(&info->extent_ins_mutex);
1679
1680 key.objectid = start;
1681 key.offset = extent_op->parent;
1682 key.type = BTRFS_EXTENT_REF_KEY;
1683 err = btrfs_set_item_key_safe(trans, extent_root, path,
1684 &key);
1685 BUG_ON(err);
1686 ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1687 struct btrfs_extent_ref);
1688 btrfs_set_ref_generation(path->nodes[0], ref,
1689 extent_op->generation);
1690 btrfs_mark_buffer_dirty(path->nodes[0]);
1691 btrfs_release_path(extent_root, path);
1692 } else {
1693 BUG_ON(1);
1694 }
1695 kfree(extent_op);
1696 unlock_extent(&info->extent_ins, start, end, GFP_NOFS);
1697 if (all)
1698 search = 0;
1699 else
1700 search = end + 1;
1701
1702 cond_resched();
1703 }
1704 btrfs_free_path(path);
1705 return 0;
1706 }
1707
1708 static int pin_down_bytes(struct btrfs_trans_handle *trans,
1709 struct btrfs_root *root,
1710 u64 bytenr, u64 num_bytes, int is_data)
1711 {
1712 int err = 0;
1713 struct extent_buffer *buf;
1714
1715 if (is_data)
1716 goto pinit;
1717
1718 buf = btrfs_find_tree_block(root, bytenr, num_bytes);
1719 if (!buf)
1720 goto pinit;
1721
1722 /* we can reuse a block if it hasn't been written
1723 * and it is from this transaction. We can't
1724 * reuse anything from the tree log root because
1725 * it has tiny sub-transactions.
1726 */
1727 if (btrfs_buffer_uptodate(buf, 0) &&
1728 btrfs_try_tree_lock(buf)) {
1729 u64 header_owner = btrfs_header_owner(buf);
1730 u64 header_transid = btrfs_header_generation(buf);
1731 if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
1732 header_owner != BTRFS_TREE_RELOC_OBJECTID &&
1733 header_transid == trans->transid &&
1734 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
1735 clean_tree_block(NULL, root, buf);
1736 btrfs_tree_unlock(buf);
1737 free_extent_buffer(buf);
1738 return 1;
1739 }
1740 btrfs_tree_unlock(buf);
1741 }
1742 free_extent_buffer(buf);
1743 pinit:
1744 btrfs_update_pinned_extents(root, bytenr, num_bytes, 1);
1745
1746 BUG_ON(err < 0);
1747 return 0;
1748 }
1749
1750 /*
1751 * remove an extent from the root, returns 0 on success
1752 */
1753 static int __free_extent(struct btrfs_trans_handle *trans,
1754 struct btrfs_root *root,
1755 u64 bytenr, u64 num_bytes, u64 parent,
1756 u64 root_objectid, u64 ref_generation,
1757 u64 owner_objectid, int pin, int mark_free)
1758 {
1759 struct btrfs_path *path;
1760 struct btrfs_key key;
1761 struct btrfs_fs_info *info = root->fs_info;
1762 struct btrfs_root *extent_root = info->extent_root;
1763 struct extent_buffer *leaf;
1764 int ret;
1765 int extent_slot = 0;
1766 int found_extent = 0;
1767 int num_to_del = 1;
1768 struct btrfs_extent_item *ei;
1769 u32 refs;
1770
1771 key.objectid = bytenr;
1772 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
1773 key.offset = num_bytes;
1774 path = btrfs_alloc_path();
1775 if (!path)
1776 return -ENOMEM;
1777
1778 path->reada = 1;
1779 ret = lookup_extent_backref(trans, extent_root, path,
1780 bytenr, parent, root_objectid,
1781 ref_generation, owner_objectid, 1);
1782 if (ret == 0) {
1783 struct btrfs_key found_key;
1784 extent_slot = path->slots[0];
1785 while(extent_slot > 0) {
1786 extent_slot--;
1787 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1788 extent_slot);
1789 if (found_key.objectid != bytenr)
1790 break;
1791 if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
1792 found_key.offset == num_bytes) {
1793 found_extent = 1;
1794 break;
1795 }
1796 if (path->slots[0] - extent_slot > 5)
1797 break;
1798 }
1799 if (!found_extent) {
1800 ret = remove_extent_backref(trans, extent_root, path);
1801 BUG_ON(ret);
1802 btrfs_release_path(extent_root, path);
1803 ret = btrfs_search_slot(trans, extent_root,
1804 &key, path, -1, 1);
1805 BUG_ON(ret);
1806 extent_slot = path->slots[0];
1807 }
1808 } else {
1809 btrfs_print_leaf(extent_root, path->nodes[0]);
1810 WARN_ON(1);
1811 printk("Unable to find ref byte nr %Lu root %Lu "
1812 "gen %Lu owner %Lu\n", bytenr,
1813 root_objectid, ref_generation, owner_objectid);
1814 }
1815
1816 leaf = path->nodes[0];
1817 ei = btrfs_item_ptr(leaf, extent_slot,
1818 struct btrfs_extent_item);
1819 refs = btrfs_extent_refs(leaf, ei);
1820 BUG_ON(refs == 0);
1821 refs -= 1;
1822 btrfs_set_extent_refs(leaf, ei, refs);
1823
1824 btrfs_mark_buffer_dirty(leaf);
1825
1826 if (refs == 0 && found_extent && path->slots[0] == extent_slot + 1) {
1827 struct btrfs_extent_ref *ref;
1828 ref = btrfs_item_ptr(leaf, path->slots[0],
1829 struct btrfs_extent_ref);
1830 BUG_ON(btrfs_ref_num_refs(leaf, ref) != 1);
1831 /* if the back ref and the extent are next to each other
1832 * they get deleted below in one shot
1833 */
1834 path->slots[0] = extent_slot;
1835 num_to_del = 2;
1836 } else if (found_extent) {
1837 /* otherwise delete the extent back ref */
1838 ret = remove_extent_backref(trans, extent_root, path);
1839 BUG_ON(ret);
1840 /* if refs are 0, we need to setup the path for deletion */
1841 if (refs == 0) {
1842 btrfs_release_path(extent_root, path);
1843 ret = btrfs_search_slot(trans, extent_root, &key, path,
1844 -1, 1);
1845 BUG_ON(ret);
1846 }
1847 }
1848
1849 if (refs == 0) {
1850 u64 super_used;
1851 u64 root_used;
1852 #ifdef BIO_RW_DISCARD
1853 u64 map_length = num_bytes;
1854 struct btrfs_multi_bio *multi = NULL;
1855 #endif
1856
1857 if (pin) {
1858 mutex_lock(&root->fs_info->pinned_mutex);
1859 ret = pin_down_bytes(trans, root, bytenr, num_bytes,
1860 owner_objectid >= BTRFS_FIRST_FREE_OBJECTID);
1861 mutex_unlock(&root->fs_info->pinned_mutex);
1862 if (ret > 0)
1863 mark_free = 1;
1864 BUG_ON(ret < 0);
1865 }
1866
1867 /* block accounting for super block */
1868 spin_lock_irq(&info->delalloc_lock);
1869 super_used = btrfs_super_bytes_used(&info->super_copy);
1870 btrfs_set_super_bytes_used(&info->super_copy,
1871 super_used - num_bytes);
1872 spin_unlock_irq(&info->delalloc_lock);
1873
1874 /* block accounting for root item */
1875 root_used = btrfs_root_used(&root->root_item);
1876 btrfs_set_root_used(&root->root_item,
1877 root_used - num_bytes);
1878 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
1879 num_to_del);
1880 BUG_ON(ret);
1881 btrfs_release_path(extent_root, path);
1882 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
1883 mark_free);
1884 BUG_ON(ret);
1885
1886 #ifdef BIO_RW_DISCARD
1887 /* Tell the block device(s) that the sectors can be discarded */
1888 ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
1889 bytenr, &map_length, &multi, 0);
1890 if (!ret) {
1891 struct btrfs_bio_stripe *stripe = multi->stripes;
1892 int i;
1893
1894 if (map_length > num_bytes)
1895 map_length = num_bytes;
1896
1897 for (i = 0; i < multi->num_stripes; i++, stripe++) {
1898 blkdev_issue_discard(stripe->dev->bdev,
1899 stripe->physical >> 9,
1900 map_length >> 9);
1901 }
1902 kfree(multi);
1903 }
1904 #endif
1905 }
1906 btrfs_free_path(path);
1907 finish_current_insert(trans, extent_root, 0);
1908 return ret;
1909 }
1910
1911 /*
1912 * find all the blocks marked as pending in the radix tree and remove
1913 * them from the extent map
1914 */
1915 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
1916 btrfs_root *extent_root, int all)
1917 {
1918 int ret;
1919 int err = 0;
1920 u64 start;
1921 u64 end;
1922 u64 priv;
1923 u64 search = 0;
1924 struct extent_io_tree *pending_del;
1925 struct extent_io_tree *extent_ins;
1926 struct pending_extent_op *extent_op;
1927 struct btrfs_fs_info *info = extent_root->fs_info;
1928
1929 extent_ins = &extent_root->fs_info->extent_ins;
1930 pending_del = &extent_root->fs_info->pending_del;
1931
1932 while(1) {
1933 mutex_lock(&info->extent_ins_mutex);
1934 ret = find_first_extent_bit(pending_del, search, &start, &end,
1935 EXTENT_WRITEBACK);
1936 if (ret) {
1937 mutex_unlock(&info->extent_ins_mutex);
1938 if (all && search) {
1939 search = 0;
1940 continue;
1941 }
1942 break;
1943 }
1944
1945 ret = try_lock_extent(extent_ins, start, end, GFP_NOFS);
1946 if (!ret) {
1947 search = end+1;
1948 mutex_unlock(&info->extent_ins_mutex);
1949 cond_resched();
1950 continue;
1951 }
1952 BUG_ON(ret < 0);
1953
1954 ret = get_state_private(pending_del, start, &priv);
1955 BUG_ON(ret);
1956 extent_op = (struct pending_extent_op *)(unsigned long)priv;
1957
1958 clear_extent_bits(pending_del, start, end, EXTENT_WRITEBACK,
1959 GFP_NOFS);
1960 if (!test_range_bit(extent_ins, start, end,
1961 EXTENT_WRITEBACK, 0)) {
1962 mutex_unlock(&info->extent_ins_mutex);
1963 free_extent:
1964 ret = __free_extent(trans, extent_root,
1965 start, end + 1 - start,
1966 extent_op->orig_parent,
1967 extent_root->root_key.objectid,
1968 extent_op->orig_generation,
1969 extent_op->level, 1, 0);
1970 kfree(extent_op);
1971 } else {
1972 kfree(extent_op);
1973
1974 ret = get_state_private(&info->extent_ins, start,
1975 &priv);
1976 BUG_ON(ret);
1977 extent_op = (struct pending_extent_op *)
1978 (unsigned long)priv;
1979
1980 clear_extent_bits(&info->extent_ins, start, end,
1981 EXTENT_WRITEBACK, GFP_NOFS);
1982
1983 mutex_unlock(&info->extent_ins_mutex);
1984
1985 if (extent_op->type == PENDING_BACKREF_UPDATE)
1986 goto free_extent;
1987
1988 mutex_lock(&extent_root->fs_info->pinned_mutex);
1989 ret = pin_down_bytes(trans, extent_root, start,
1990 end + 1 - start, 0);
1991 mutex_unlock(&extent_root->fs_info->pinned_mutex);
1992
1993 ret = update_block_group(trans, extent_root, start,
1994 end + 1 - start, 0, ret > 0);
1995
1996 BUG_ON(ret);
1997 kfree(extent_op);
1998 }
1999 if (ret)
2000 err = ret;
2001 unlock_extent(extent_ins, start, end, GFP_NOFS);
2002
2003 if (all)
2004 search = 0;
2005 else
2006 search = end + 1;
2007 cond_resched();
2008 }
2009 return err;
2010 }
2011
2012 /*
2013 * remove an extent from the root, returns 0 on success
2014 */
2015 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
2016 struct btrfs_root *root,
2017 u64 bytenr, u64 num_bytes, u64 parent,
2018 u64 root_objectid, u64 ref_generation,
2019 u64 owner_objectid, int pin)
2020 {
2021 struct btrfs_root *extent_root = root->fs_info->extent_root;
2022 int pending_ret;
2023 int ret;
2024
2025 WARN_ON(num_bytes < root->sectorsize);
2026 if (root == extent_root) {
2027 struct pending_extent_op *extent_op;
2028
2029 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2030 BUG_ON(!extent_op);
2031
2032 extent_op->type = PENDING_EXTENT_DELETE;
2033 extent_op->bytenr = bytenr;
2034 extent_op->num_bytes = num_bytes;
2035 extent_op->parent = parent;
2036 extent_op->orig_parent = parent;
2037 extent_op->generation = ref_generation;
2038 extent_op->orig_generation = ref_generation;
2039 extent_op->level = (int)owner_objectid;
2040
2041 mutex_lock(&root->fs_info->extent_ins_mutex);
2042 set_extent_bits(&root->fs_info->pending_del,
2043 bytenr, bytenr + num_bytes - 1,
2044 EXTENT_WRITEBACK, GFP_NOFS);
2045 set_state_private(&root->fs_info->pending_del,
2046 bytenr, (unsigned long)extent_op);
2047 mutex_unlock(&root->fs_info->extent_ins_mutex);
2048 return 0;
2049 }
2050 /* if metadata always pin */
2051 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
2052 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
2053 struct btrfs_block_group_cache *cache;
2054
2055 /* btrfs_free_reserved_extent */
2056 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
2057 BUG_ON(!cache);
2058 btrfs_add_free_space(cache, bytenr, num_bytes);
2059 update_reserved_extents(root, bytenr, num_bytes, 0);
2060 return 0;
2061 }
2062 pin = 1;
2063 }
2064
2065 /* if data pin when any transaction has committed this */
2066 if (ref_generation != trans->transid)
2067 pin = 1;
2068
2069 ret = __free_extent(trans, root, bytenr, num_bytes, parent,
2070 root_objectid, ref_generation,
2071 owner_objectid, pin, pin == 0);
2072
2073 finish_current_insert(trans, root->fs_info->extent_root, 0);
2074 pending_ret = del_pending_extents(trans, root->fs_info->extent_root, 0);
2075 return ret ? ret : pending_ret;
2076 }
2077
2078 int btrfs_free_extent(struct btrfs_trans_handle *trans,
2079 struct btrfs_root *root,
2080 u64 bytenr, u64 num_bytes, u64 parent,
2081 u64 root_objectid, u64 ref_generation,
2082 u64 owner_objectid, int pin)
2083 {
2084 int ret;
2085
2086 ret = __btrfs_free_extent(trans, root, bytenr, num_bytes, parent,
2087 root_objectid, ref_generation,
2088 owner_objectid, pin);
2089 return ret;
2090 }
2091
2092 static u64 stripe_align(struct btrfs_root *root, u64 val)
2093 {
2094 u64 mask = ((u64)root->stripesize - 1);
2095 u64 ret = (val + mask) & ~mask;
2096 return ret;
2097 }
2098
2099 /*
2100 * walks the btree of allocated extents and find a hole of a given size.
2101 * The key ins is changed to record the hole:
2102 * ins->objectid == block start
2103 * ins->flags = BTRFS_EXTENT_ITEM_KEY
2104 * ins->offset == number of blocks
2105 * Any available blocks before search_start are skipped.
2106 */
2107 static int noinline find_free_extent(struct btrfs_trans_handle *trans,
2108 struct btrfs_root *orig_root,
2109 u64 num_bytes, u64 empty_size,
2110 u64 search_start, u64 search_end,
2111 u64 hint_byte, struct btrfs_key *ins,
2112 u64 exclude_start, u64 exclude_nr,
2113 int data)
2114 {
2115 int ret = 0;
2116 struct btrfs_root * root = orig_root->fs_info->extent_root;
2117 u64 total_needed = num_bytes;
2118 u64 *last_ptr = NULL;
2119 struct btrfs_block_group_cache *block_group = NULL;
2120 int chunk_alloc_done = 0;
2121 int empty_cluster = 2 * 1024 * 1024;
2122 int allowed_chunk_alloc = 0;
2123 struct list_head *head = NULL, *cur = NULL;
2124 int loop = 0;
2125 struct btrfs_space_info *space_info;
2126
2127 WARN_ON(num_bytes < root->sectorsize);
2128 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
2129 ins->objectid = 0;
2130 ins->offset = 0;
2131
2132 if (orig_root->ref_cows || empty_size)
2133 allowed_chunk_alloc = 1;
2134
2135 if (data & BTRFS_BLOCK_GROUP_METADATA) {
2136 last_ptr = &root->fs_info->last_alloc;
2137 empty_cluster = 256 * 1024;
2138 }
2139
2140 if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD))
2141 last_ptr = &root->fs_info->last_data_alloc;
2142
2143 if (last_ptr) {
2144 if (*last_ptr)
2145 hint_byte = *last_ptr;
2146 else
2147 empty_size += empty_cluster;
2148 }
2149 search_start = max(search_start, first_logical_byte(root, 0));
2150 search_start = max(search_start, hint_byte);
2151 total_needed += empty_size;
2152
2153 block_group = btrfs_lookup_block_group(root->fs_info, search_start);
2154 if (!block_group)
2155 block_group = btrfs_lookup_first_block_group(root->fs_info,
2156 search_start);
2157 space_info = __find_space_info(root->fs_info, data);
2158
2159 down_read(&space_info->groups_sem);
2160 while (1) {
2161 struct btrfs_free_space *free_space;
2162 /*
2163 * the only way this happens if our hint points to a block
2164 * group thats not of the proper type, while looping this
2165 * should never happen
2166 */
2167 WARN_ON(!block_group);
2168 mutex_lock(&block_group->alloc_mutex);
2169 if (unlikely(!block_group_bits(block_group, data)))
2170 goto new_group;
2171
2172 ret = cache_block_group(root, block_group);
2173 if (ret) {
2174 mutex_unlock(&block_group->alloc_mutex);
2175 break;
2176 }
2177
2178 if (block_group->ro)
2179 goto new_group;
2180
2181 free_space = btrfs_find_free_space(block_group, search_start,
2182 total_needed);
2183 if (free_space) {
2184 u64 start = block_group->key.objectid;
2185 u64 end = block_group->key.objectid +
2186 block_group->key.offset;
2187
2188 search_start = stripe_align(root, free_space->offset);
2189
2190 /* move on to the next group */
2191 if (search_start + num_bytes >= search_end)
2192 goto new_group;
2193
2194 /* move on to the next group */
2195 if (search_start + num_bytes > end)
2196 goto new_group;
2197
2198 if (last_ptr && *last_ptr && search_start != *last_ptr) {
2199 total_needed += empty_cluster;
2200 *last_ptr = 0;
2201 /*
2202 * if search_start is still in this block group
2203 * then we just re-search this block group
2204 */
2205 if (search_start >= start &&
2206 search_start < end) {
2207 mutex_unlock(&block_group->alloc_mutex);
2208 continue;
2209 }
2210
2211 /* else we go to the next block group */
2212 goto new_group;
2213 }
2214
2215 if (exclude_nr > 0 &&
2216 (search_start + num_bytes > exclude_start &&
2217 search_start < exclude_start + exclude_nr)) {
2218 search_start = exclude_start + exclude_nr;
2219 /*
2220 * if search_start is still in this block group
2221 * then we just re-search this block group
2222 */
2223 if (search_start >= start &&
2224 search_start < end) {
2225 mutex_unlock(&block_group->alloc_mutex);
2226 continue;
2227 }
2228
2229 /* else we go to the next block group */
2230 goto new_group;
2231 }
2232
2233 ins->objectid = search_start;
2234 ins->offset = num_bytes;
2235
2236 btrfs_remove_free_space_lock(block_group, search_start,
2237 num_bytes);
2238 /* we are all good, lets return */
2239 mutex_unlock(&block_group->alloc_mutex);
2240 break;
2241 }
2242 new_group:
2243 mutex_unlock(&block_group->alloc_mutex);
2244 /*
2245 * Here's how this works.
2246 * loop == 0: we were searching a block group via a hint
2247 * and didn't find anything, so we start at
2248 * the head of the block groups and keep searching
2249 * loop == 1: we're searching through all of the block groups
2250 * if we hit the head again we have searched
2251 * all of the block groups for this space and we
2252 * need to try and allocate, if we cant error out.
2253 * loop == 2: we allocated more space and are looping through
2254 * all of the block groups again.
2255 */
2256 if (loop == 0) {
2257 head = &space_info->block_groups;
2258 cur = head->next;
2259
2260 if (last_ptr && *last_ptr) {
2261 total_needed += empty_cluster;
2262 *last_ptr = 0;
2263 }
2264 loop++;
2265 } else if (loop == 1 && cur == head) {
2266 if (allowed_chunk_alloc && !chunk_alloc_done) {
2267 up_read(&space_info->groups_sem);
2268 ret = do_chunk_alloc(trans, root, num_bytes +
2269 2 * 1024 * 1024, data, 1);
2270 if (ret < 0)
2271 break;
2272 down_read(&space_info->groups_sem);
2273 loop++;
2274 head = &space_info->block_groups;
2275 cur = head->next;
2276 chunk_alloc_done = 1;
2277 } else if (!allowed_chunk_alloc) {
2278 space_info->force_alloc = 1;
2279 break;
2280 } else {
2281 break;
2282 }
2283 } else if (cur == head) {
2284 break;
2285 }
2286
2287 block_group = list_entry(cur, struct btrfs_block_group_cache,
2288 list);
2289 search_start = block_group->key.objectid;
2290 cur = cur->next;
2291 }
2292
2293 /* we found what we needed */
2294 if (ins->objectid) {
2295 if (!(data & BTRFS_BLOCK_GROUP_DATA))
2296 trans->block_group = block_group;
2297
2298 if (last_ptr)
2299 *last_ptr = ins->objectid + ins->offset;
2300 ret = 0;
2301 } else if (!ret) {
2302 ret = -ENOSPC;
2303 }
2304
2305 up_read(&space_info->groups_sem);
2306 return ret;
2307 }
2308
2309 static void dump_space_info(struct btrfs_space_info *info, u64 bytes)
2310 {
2311 struct btrfs_block_group_cache *cache;
2312 struct list_head *l;
2313
2314 printk(KERN_INFO "space_info has %Lu free, is %sfull\n",
2315 info->total_bytes - info->bytes_used - info->bytes_pinned -
2316 info->bytes_reserved, (info->full) ? "" : "not ");
2317
2318 down_read(&info->groups_sem);
2319 list_for_each(l, &info->block_groups) {
2320 cache = list_entry(l, struct btrfs_block_group_cache, list);
2321 spin_lock(&cache->lock);
2322 printk(KERN_INFO "block group %Lu has %Lu bytes, %Lu used "
2323 "%Lu pinned %Lu reserved\n",
2324 cache->key.objectid, cache->key.offset,
2325 btrfs_block_group_used(&cache->item),
2326 cache->pinned, cache->reserved);
2327 btrfs_dump_free_space(cache, bytes);
2328 spin_unlock(&cache->lock);
2329 }
2330 up_read(&info->groups_sem);
2331 }
2332
2333 static int __btrfs_reserve_extent(struct btrfs_trans_handle *trans,
2334 struct btrfs_root *root,
2335 u64 num_bytes, u64 min_alloc_size,
2336 u64 empty_size, u64 hint_byte,
2337 u64 search_end, struct btrfs_key *ins,
2338 u64 data)
2339 {
2340 int ret;
2341 u64 search_start = 0;
2342 u64 alloc_profile;
2343 struct btrfs_fs_info *info = root->fs_info;
2344
2345 if (data) {
2346 alloc_profile = info->avail_data_alloc_bits &
2347 info->data_alloc_profile;
2348 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
2349 } else if (root == root->fs_info->chunk_root) {
2350 alloc_profile = info->avail_system_alloc_bits &
2351 info->system_alloc_profile;
2352 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
2353 } else {
2354 alloc_profile = info->avail_metadata_alloc_bits &
2355 info->metadata_alloc_profile;
2356 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
2357 }
2358 again:
2359 data = reduce_alloc_profile(root, data);
2360 /*
2361 * the only place that sets empty_size is btrfs_realloc_node, which
2362 * is not called recursively on allocations
2363 */
2364 if (empty_size || root->ref_cows) {
2365 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
2366 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2367 2 * 1024 * 1024,
2368 BTRFS_BLOCK_GROUP_METADATA |
2369 (info->metadata_alloc_profile &
2370 info->avail_metadata_alloc_bits), 0);
2371 }
2372 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2373 num_bytes + 2 * 1024 * 1024, data, 0);
2374 }
2375
2376 WARN_ON(num_bytes < root->sectorsize);
2377 ret = find_free_extent(trans, root, num_bytes, empty_size,
2378 search_start, search_end, hint_byte, ins,
2379 trans->alloc_exclude_start,
2380 trans->alloc_exclude_nr, data);
2381
2382 if (ret == -ENOSPC && num_bytes > min_alloc_size) {
2383 num_bytes = num_bytes >> 1;
2384 num_bytes = num_bytes & ~(root->sectorsize - 1);
2385 num_bytes = max(num_bytes, min_alloc_size);
2386 do_chunk_alloc(trans, root->fs_info->extent_root,
2387 num_bytes, data, 1);
2388 goto again;
2389 }
2390 if (ret) {
2391 struct btrfs_space_info *sinfo;
2392
2393 sinfo = __find_space_info(root->fs_info, data);
2394 printk("allocation failed flags %Lu, wanted %Lu\n",
2395 data, num_bytes);
2396 dump_space_info(sinfo, num_bytes);
2397 BUG();
2398 }
2399
2400 return ret;
2401 }
2402
2403 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
2404 {
2405 struct btrfs_block_group_cache *cache;
2406
2407 cache = btrfs_lookup_block_group(root->fs_info, start);
2408 if (!cache) {
2409 printk(KERN_ERR "Unable to find block group for %Lu\n", start);
2410 return -ENOSPC;
2411 }
2412 btrfs_add_free_space(cache, start, len);
2413 update_reserved_extents(root, start, len, 0);
2414 return 0;
2415 }
2416
2417 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
2418 struct btrfs_root *root,
2419 u64 num_bytes, u64 min_alloc_size,
2420 u64 empty_size, u64 hint_byte,
2421 u64 search_end, struct btrfs_key *ins,
2422 u64 data)
2423 {
2424 int ret;
2425 ret = __btrfs_reserve_extent(trans, root, num_bytes, min_alloc_size,
2426 empty_size, hint_byte, search_end, ins,
2427 data);
2428 update_reserved_extents(root, ins->objectid, ins->offset, 1);
2429 return ret;
2430 }
2431
2432 static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
2433 struct btrfs_root *root, u64 parent,
2434 u64 root_objectid, u64 ref_generation,
2435 u64 owner, struct btrfs_key *ins)
2436 {
2437 int ret;
2438 int pending_ret;
2439 u64 super_used;
2440 u64 root_used;
2441 u64 num_bytes = ins->offset;
2442 u32 sizes[2];
2443 struct btrfs_fs_info *info = root->fs_info;
2444 struct btrfs_root *extent_root = info->extent_root;
2445 struct btrfs_extent_item *extent_item;
2446 struct btrfs_extent_ref *ref;
2447 struct btrfs_path *path;
2448 struct btrfs_key keys[2];
2449
2450 if (parent == 0)
2451 parent = ins->objectid;
2452
2453 /* block accounting for super block */
2454 spin_lock_irq(&info->delalloc_lock);
2455 super_used = btrfs_super_bytes_used(&info->super_copy);
2456 btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes);
2457 spin_unlock_irq(&info->delalloc_lock);
2458
2459 /* block accounting for root item */
2460 root_used = btrfs_root_used(&root->root_item);
2461 btrfs_set_root_used(&root->root_item, root_used + num_bytes);
2462
2463 if (root == extent_root) {
2464 struct pending_extent_op *extent_op;
2465
2466 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2467 BUG_ON(!extent_op);
2468
2469 extent_op->type = PENDING_EXTENT_INSERT;
2470 extent_op->bytenr = ins->objectid;
2471 extent_op->num_bytes = ins->offset;
2472 extent_op->parent = parent;
2473 extent_op->orig_parent = 0;
2474 extent_op->generation = ref_generation;
2475 extent_op->orig_generation = 0;
2476 extent_op->level = (int)owner;
2477
2478 mutex_lock(&root->fs_info->extent_ins_mutex);
2479 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
2480 ins->objectid + ins->offset - 1,
2481 EXTENT_WRITEBACK, GFP_NOFS);
2482 set_state_private(&root->fs_info->extent_ins,
2483 ins->objectid, (unsigned long)extent_op);
2484 mutex_unlock(&root->fs_info->extent_ins_mutex);
2485 goto update_block;
2486 }
2487
2488 memcpy(&keys[0], ins, sizeof(*ins));
2489 keys[1].objectid = ins->objectid;
2490 keys[1].type = BTRFS_EXTENT_REF_KEY;
2491 keys[1].offset = parent;
2492 sizes[0] = sizeof(*extent_item);
2493 sizes[1] = sizeof(*ref);
2494
2495 path = btrfs_alloc_path();
2496 BUG_ON(!path);
2497
2498 ret = btrfs_insert_empty_items(trans, extent_root, path, keys,
2499 sizes, 2);
2500 BUG_ON(ret);
2501
2502 extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2503 struct btrfs_extent_item);
2504 btrfs_set_extent_refs(path->nodes[0], extent_item, 1);
2505 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
2506 struct btrfs_extent_ref);
2507
2508 btrfs_set_ref_root(path->nodes[0], ref, root_objectid);
2509 btrfs_set_ref_generation(path->nodes[0], ref, ref_generation);
2510 btrfs_set_ref_objectid(path->nodes[0], ref, owner);
2511 btrfs_set_ref_num_refs(path->nodes[0], ref, 1);
2512
2513 btrfs_mark_buffer_dirty(path->nodes[0]);
2514
2515 trans->alloc_exclude_start = 0;
2516 trans->alloc_exclude_nr = 0;
2517 btrfs_free_path(path);
2518 finish_current_insert(trans, extent_root, 0);
2519 pending_ret = del_pending_extents(trans, extent_root, 0);
2520
2521 if (ret)
2522 goto out;
2523 if (pending_ret) {
2524 ret = pending_ret;
2525 goto out;
2526 }
2527
2528 update_block:
2529 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1, 0);
2530 if (ret) {
2531 printk("update block group failed for %Lu %Lu\n",
2532 ins->objectid, ins->offset);
2533 BUG();
2534 }
2535 out:
2536 return ret;
2537 }
2538
2539 int btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
2540 struct btrfs_root *root, u64 parent,
2541 u64 root_objectid, u64 ref_generation,
2542 u64 owner, struct btrfs_key *ins)
2543 {
2544 int ret;
2545
2546 if (root_objectid == BTRFS_TREE_LOG_OBJECTID)
2547 return 0;
2548 ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
2549 ref_generation, owner, ins);
2550 update_reserved_extents(root, ins->objectid, ins->offset, 0);
2551 return ret;
2552 }
2553
2554 /*
2555 * this is used by the tree logging recovery code. It records that
2556 * an extent has been allocated and makes sure to clear the free
2557 * space cache bits as well
2558 */
2559 int btrfs_alloc_logged_extent(struct btrfs_trans_handle *trans,
2560 struct btrfs_root *root, u64 parent,
2561 u64 root_objectid, u64 ref_generation,
2562 u64 owner, struct btrfs_key *ins)
2563 {
2564 int ret;
2565 struct btrfs_block_group_cache *block_group;
2566
2567 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
2568 mutex_lock(&block_group->alloc_mutex);
2569 cache_block_group(root, block_group);
2570
2571 ret = btrfs_remove_free_space_lock(block_group, ins->objectid,
2572 ins->offset);
2573 mutex_unlock(&block_group->alloc_mutex);
2574 BUG_ON(ret);
2575 ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
2576 ref_generation, owner, ins);
2577 return ret;
2578 }
2579
2580 /*
2581 * finds a free extent and does all the dirty work required for allocation
2582 * returns the key for the extent through ins, and a tree buffer for
2583 * the first block of the extent through buf.
2584 *
2585 * returns 0 if everything worked, non-zero otherwise.
2586 */
2587 int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
2588 struct btrfs_root *root,
2589 u64 num_bytes, u64 parent, u64 min_alloc_size,
2590 u64 root_objectid, u64 ref_generation,
2591 u64 owner_objectid, u64 empty_size, u64 hint_byte,
2592 u64 search_end, struct btrfs_key *ins, u64 data)
2593 {
2594 int ret;
2595
2596 ret = __btrfs_reserve_extent(trans, root, num_bytes,
2597 min_alloc_size, empty_size, hint_byte,
2598 search_end, ins, data);
2599 BUG_ON(ret);
2600 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
2601 ret = __btrfs_alloc_reserved_extent(trans, root, parent,
2602 root_objectid, ref_generation,
2603 owner_objectid, ins);
2604 BUG_ON(ret);
2605
2606 } else {
2607 update_reserved_extents(root, ins->objectid, ins->offset, 1);
2608 }
2609 return ret;
2610 }
2611
2612 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
2613 struct btrfs_root *root,
2614 u64 bytenr, u32 blocksize)
2615 {
2616 struct extent_buffer *buf;
2617
2618 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
2619 if (!buf)
2620 return ERR_PTR(-ENOMEM);
2621 btrfs_set_header_generation(buf, trans->transid);
2622 btrfs_tree_lock(buf);
2623 clean_tree_block(trans, root, buf);
2624 btrfs_set_buffer_uptodate(buf);
2625 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
2626 set_extent_dirty(&root->dirty_log_pages, buf->start,
2627 buf->start + buf->len - 1, GFP_NOFS);
2628 } else {
2629 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
2630 buf->start + buf->len - 1, GFP_NOFS);
2631 }
2632 trans->blocks_used++;
2633 return buf;
2634 }
2635
2636 /*
2637 * helper function to allocate a block for a given tree
2638 * returns the tree buffer or NULL.
2639 */
2640 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
2641 struct btrfs_root *root,
2642 u32 blocksize, u64 parent,
2643 u64 root_objectid,
2644 u64 ref_generation,
2645 int level,
2646 u64 hint,
2647 u64 empty_size)
2648 {
2649 struct btrfs_key ins;
2650 int ret;
2651 struct extent_buffer *buf;
2652
2653 ret = btrfs_alloc_extent(trans, root, blocksize, parent, blocksize,
2654 root_objectid, ref_generation, level,
2655 empty_size, hint, (u64)-1, &ins, 0);
2656 if (ret) {
2657 BUG_ON(ret > 0);
2658 return ERR_PTR(ret);
2659 }
2660
2661 buf = btrfs_init_new_buffer(trans, root, ins.objectid, blocksize);
2662 return buf;
2663 }
2664
2665 int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans,
2666 struct btrfs_root *root, struct extent_buffer *leaf)
2667 {
2668 u64 leaf_owner;
2669 u64 leaf_generation;
2670 struct btrfs_key key;
2671 struct btrfs_file_extent_item *fi;
2672 int i;
2673 int nritems;
2674 int ret;
2675
2676 BUG_ON(!btrfs_is_leaf(leaf));
2677 nritems = btrfs_header_nritems(leaf);
2678 leaf_owner = btrfs_header_owner(leaf);
2679 leaf_generation = btrfs_header_generation(leaf);
2680
2681 for (i = 0; i < nritems; i++) {
2682 u64 disk_bytenr;
2683 cond_resched();
2684
2685 btrfs_item_key_to_cpu(leaf, &key, i);
2686 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2687 continue;
2688 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
2689 if (btrfs_file_extent_type(leaf, fi) ==
2690 BTRFS_FILE_EXTENT_INLINE)
2691 continue;
2692 /*
2693 * FIXME make sure to insert a trans record that
2694 * repeats the snapshot del on crash
2695 */
2696 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
2697 if (disk_bytenr == 0)
2698 continue;
2699
2700 ret = __btrfs_free_extent(trans, root, disk_bytenr,
2701 btrfs_file_extent_disk_num_bytes(leaf, fi),
2702 leaf->start, leaf_owner, leaf_generation,
2703 key.objectid, 0);
2704 BUG_ON(ret);
2705
2706 atomic_inc(&root->fs_info->throttle_gen);
2707 wake_up(&root->fs_info->transaction_throttle);
2708 cond_resched();
2709 }
2710 return 0;
2711 }
2712
2713 static int noinline cache_drop_leaf_ref(struct btrfs_trans_handle *trans,
2714 struct btrfs_root *root,
2715 struct btrfs_leaf_ref *ref)
2716 {
2717 int i;
2718 int ret;
2719 struct btrfs_extent_info *info = ref->extents;
2720
2721 for (i = 0; i < ref->nritems; i++) {
2722 ret = __btrfs_free_extent(trans, root, info->bytenr,
2723 info->num_bytes, ref->bytenr,
2724 ref->owner, ref->generation,
2725 info->objectid, 0);
2726
2727 atomic_inc(&root->fs_info->throttle_gen);
2728 wake_up(&root->fs_info->transaction_throttle);
2729 cond_resched();
2730
2731 BUG_ON(ret);
2732 info++;
2733 }
2734
2735 return 0;
2736 }
2737
2738 int drop_snap_lookup_refcount(struct btrfs_root *root, u64 start, u64 len,
2739 u32 *refs)
2740 {
2741 int ret;
2742
2743 ret = btrfs_lookup_extent_ref(NULL, root, start, len, refs);
2744 BUG_ON(ret);
2745
2746 #if 0 // some debugging code in case we see problems here
2747 /* if the refs count is one, it won't get increased again. But
2748 * if the ref count is > 1, someone may be decreasing it at
2749 * the same time we are.
2750 */
2751 if (*refs != 1) {
2752 struct extent_buffer *eb = NULL;
2753 eb = btrfs_find_create_tree_block(root, start, len);
2754 if (eb)
2755 btrfs_tree_lock(eb);
2756
2757 mutex_lock(&root->fs_info->alloc_mutex);
2758 ret = lookup_extent_ref(NULL, root, start, len, refs);
2759 BUG_ON(ret);
2760 mutex_unlock(&root->fs_info->alloc_mutex);
2761
2762 if (eb) {
2763 btrfs_tree_unlock(eb);
2764 free_extent_buffer(eb);
2765 }
2766 if (*refs == 1) {
2767 printk("block %llu went down to one during drop_snap\n",
2768 (unsigned long long)start);
2769 }
2770
2771 }
2772 #endif
2773
2774 cond_resched();
2775 return ret;
2776 }
2777
2778 /*
2779 * helper function for drop_snapshot, this walks down the tree dropping ref
2780 * counts as it goes.
2781 */
2782 static int noinline walk_down_tree(struct btrfs_trans_handle *trans,
2783 struct btrfs_root *root,
2784 struct btrfs_path *path, int *level)
2785 {
2786 u64 root_owner;
2787 u64 root_gen;
2788 u64 bytenr;
2789 u64 ptr_gen;
2790 struct extent_buffer *next;
2791 struct extent_buffer *cur;
2792 struct extent_buffer *parent;
2793 struct btrfs_leaf_ref *ref;
2794 u32 blocksize;
2795 int ret;
2796 u32 refs;
2797
2798 WARN_ON(*level < 0);
2799 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2800 ret = drop_snap_lookup_refcount(root, path->nodes[*level]->start,
2801 path->nodes[*level]->len, &refs);
2802 BUG_ON(ret);
2803 if (refs > 1)
2804 goto out;
2805
2806 /*
2807 * walk down to the last node level and free all the leaves
2808 */
2809 while(*level >= 0) {
2810 WARN_ON(*level < 0);
2811 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2812 cur = path->nodes[*level];
2813
2814 if (btrfs_header_level(cur) != *level)
2815 WARN_ON(1);
2816
2817 if (path->slots[*level] >=
2818 btrfs_header_nritems(cur))
2819 break;
2820 if (*level == 0) {
2821 ret = btrfs_drop_leaf_ref(trans, root, cur);
2822 BUG_ON(ret);
2823 break;
2824 }
2825 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2826 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2827 blocksize = btrfs_level_size(root, *level - 1);
2828
2829 ret = drop_snap_lookup_refcount(root, bytenr, blocksize, &refs);
2830 BUG_ON(ret);
2831 if (refs != 1) {
2832 parent = path->nodes[*level];
2833 root_owner = btrfs_header_owner(parent);
2834 root_gen = btrfs_header_generation(parent);
2835 path->slots[*level]++;
2836
2837 ret = __btrfs_free_extent(trans, root, bytenr,
2838 blocksize, parent->start,
2839 root_owner, root_gen,
2840 *level - 1, 1);
2841 BUG_ON(ret);
2842
2843 atomic_inc(&root->fs_info->throttle_gen);
2844 wake_up(&root->fs_info->transaction_throttle);
2845 cond_resched();
2846
2847 continue;
2848 }
2849 /*
2850 * at this point, we have a single ref, and since the
2851 * only place referencing this extent is a dead root
2852 * the reference count should never go higher.
2853 * So, we don't need to check it again
2854 */
2855 if (*level == 1) {
2856 ref = btrfs_lookup_leaf_ref(root, bytenr);
2857 if (ref && ref->generation != ptr_gen) {
2858 btrfs_free_leaf_ref(root, ref);
2859 ref = NULL;
2860 }
2861 if (ref) {
2862 ret = cache_drop_leaf_ref(trans, root, ref);
2863 BUG_ON(ret);
2864 btrfs_remove_leaf_ref(root, ref);
2865 btrfs_free_leaf_ref(root, ref);
2866 *level = 0;
2867 break;
2868 }
2869 if (printk_ratelimit()) {
2870 printk("leaf ref miss for bytenr %llu\n",
2871 (unsigned long long)bytenr);
2872 }
2873 }
2874 next = btrfs_find_tree_block(root, bytenr, blocksize);
2875 if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
2876 free_extent_buffer(next);
2877
2878 next = read_tree_block(root, bytenr, blocksize,
2879 ptr_gen);
2880 cond_resched();
2881 #if 0
2882 /*
2883 * this is a debugging check and can go away
2884 * the ref should never go all the way down to 1
2885 * at this point
2886 */
2887 ret = lookup_extent_ref(NULL, root, bytenr, blocksize,
2888 &refs);
2889 BUG_ON(ret);
2890 WARN_ON(refs != 1);
2891 #endif
2892 }
2893 WARN_ON(*level <= 0);
2894 if (path->nodes[*level-1])
2895 free_extent_buffer(path->nodes[*level-1]);
2896 path->nodes[*level-1] = next;
2897 *level = btrfs_header_level(next);
2898 path->slots[*level] = 0;
2899 cond_resched();
2900 }
2901 out:
2902 WARN_ON(*level < 0);
2903 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2904
2905 if (path->nodes[*level] == root->node) {
2906 parent = path->nodes[*level];
2907 bytenr = path->nodes[*level]->start;
2908 } else {
2909 parent = path->nodes[*level + 1];
2910 bytenr = btrfs_node_blockptr(parent, path->slots[*level + 1]);
2911 }
2912
2913 blocksize = btrfs_level_size(root, *level);
2914 root_owner = btrfs_header_owner(parent);
2915 root_gen = btrfs_header_generation(parent);
2916
2917 ret = __btrfs_free_extent(trans, root, bytenr, blocksize,
2918 parent->start, root_owner, root_gen,
2919 *level, 1);
2920 free_extent_buffer(path->nodes[*level]);
2921 path->nodes[*level] = NULL;
2922 *level += 1;
2923 BUG_ON(ret);
2924
2925 cond_resched();
2926 return 0;
2927 }
2928
2929 /*
2930 * helper function for drop_subtree, this function is similar to
2931 * walk_down_tree. The main difference is that it checks reference
2932 * counts while tree blocks are locked.
2933 */
2934 static int noinline walk_down_subtree(struct btrfs_trans_handle *trans,
2935 struct btrfs_root *root,
2936 struct btrfs_path *path, int *level)
2937 {
2938 struct extent_buffer *next;
2939 struct extent_buffer *cur;
2940 struct extent_buffer *parent;
2941 u64 bytenr;
2942 u64 ptr_gen;
2943 u32 blocksize;
2944 u32 refs;
2945 int ret;
2946
2947 cur = path->nodes[*level];
2948 ret = btrfs_lookup_extent_ref(trans, root, cur->start, cur->len,
2949 &refs);
2950 BUG_ON(ret);
2951 if (refs > 1)
2952 goto out;
2953
2954 while (*level >= 0) {
2955 cur = path->nodes[*level];
2956 if (*level == 0) {
2957 ret = btrfs_drop_leaf_ref(trans, root, cur);
2958 BUG_ON(ret);
2959 clean_tree_block(trans, root, cur);
2960 break;
2961 }
2962 if (path->slots[*level] >= btrfs_header_nritems(cur)) {
2963 clean_tree_block(trans, root, cur);
2964 break;
2965 }
2966
2967 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2968 blocksize = btrfs_level_size(root, *level - 1);
2969 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2970
2971 next = read_tree_block(root, bytenr, blocksize, ptr_gen);
2972 btrfs_tree_lock(next);
2973
2974 ret = btrfs_lookup_extent_ref(trans, root, bytenr, blocksize,
2975 &refs);
2976 BUG_ON(ret);
2977 if (refs > 1) {
2978 parent = path->nodes[*level];
2979 ret = btrfs_free_extent(trans, root, bytenr,
2980 blocksize, parent->start,
2981 btrfs_header_owner(parent),
2982 btrfs_header_generation(parent),
2983 *level - 1, 1);
2984 BUG_ON(ret);
2985 path->slots[*level]++;
2986 btrfs_tree_unlock(next);
2987 free_extent_buffer(next);
2988 continue;
2989 }
2990
2991 *level = btrfs_header_level(next);
2992 path->nodes[*level] = next;
2993 path->slots[*level] = 0;
2994 path->locks[*level] = 1;
2995 cond_resched();
2996 }
2997 out:
2998 parent = path->nodes[*level + 1];
2999 bytenr = path->nodes[*level]->start;
3000 blocksize = path->nodes[*level]->len;
3001
3002 ret = btrfs_free_extent(trans, root, bytenr, blocksize,
3003 parent->start, btrfs_header_owner(parent),
3004 btrfs_header_generation(parent), *level, 1);
3005 BUG_ON(ret);
3006
3007 if (path->locks[*level]) {
3008 btrfs_tree_unlock(path->nodes[*level]);
3009 path->locks[*level] = 0;
3010 }
3011 free_extent_buffer(path->nodes[*level]);
3012 path->nodes[*level] = NULL;
3013 *level += 1;
3014 cond_resched();
3015 return 0;
3016 }
3017
3018 /*
3019 * helper for dropping snapshots. This walks back up the tree in the path
3020 * to find the first node higher up where we haven't yet gone through
3021 * all the slots
3022 */
3023 static int noinline walk_up_tree(struct btrfs_trans_handle *trans,
3024 struct btrfs_root *root,
3025 struct btrfs_path *path,
3026 int *level, int max_level)
3027 {
3028 u64 root_owner;
3029 u64 root_gen;
3030 struct btrfs_root_item *root_item = &root->root_item;
3031 int i;
3032 int slot;
3033 int ret;
3034
3035 for (i = *level; i < max_level && path->nodes[i]; i++) {
3036 slot = path->slots[i];
3037 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
3038 struct extent_buffer *node;
3039 struct btrfs_disk_key disk_key;
3040 node = path->nodes[i];
3041 path->slots[i]++;
3042 *level = i;
3043 WARN_ON(*level == 0);
3044 btrfs_node_key(node, &disk_key, path->slots[i]);
3045 memcpy(&root_item->drop_progress,
3046 &disk_key, sizeof(disk_key));
3047 root_item->drop_level = i;
3048 return 0;
3049 } else {
3050 struct extent_buffer *parent;
3051 if (path->nodes[*level] == root->node)
3052 parent = path->nodes[*level];
3053 else
3054 parent = path->nodes[*level + 1];
3055
3056 root_owner = btrfs_header_owner(parent);
3057 root_gen = btrfs_header_generation(parent);
3058
3059 clean_tree_block(trans, root, path->nodes[*level]);
3060 ret = btrfs_free_extent(trans, root,
3061 path->nodes[*level]->start,
3062 path->nodes[*level]->len,
3063 parent->start, root_owner,
3064 root_gen, *level, 1);
3065 BUG_ON(ret);
3066 if (path->locks[*level]) {
3067 btrfs_tree_unlock(path->nodes[*level]);
3068 path->locks[*level] = 0;
3069 }
3070 free_extent_buffer(path->nodes[*level]);
3071 path->nodes[*level] = NULL;
3072 *level = i + 1;
3073 }
3074 }
3075 return 1;
3076 }
3077
3078 /*
3079 * drop the reference count on the tree rooted at 'snap'. This traverses
3080 * the tree freeing any blocks that have a ref count of zero after being
3081 * decremented.
3082 */
3083 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
3084 *root)
3085 {
3086 int ret = 0;
3087 int wret;
3088 int level;
3089 struct btrfs_path *path;
3090 int i;
3091 int orig_level;
3092 struct btrfs_root_item *root_item = &root->root_item;
3093
3094 WARN_ON(!mutex_is_locked(&root->fs_info->drop_mutex));
3095 path = btrfs_alloc_path();
3096 BUG_ON(!path);
3097
3098 level = btrfs_header_level(root->node);
3099 orig_level = level;
3100 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
3101 path->nodes[level] = root->node;
3102 extent_buffer_get(root->node);
3103 path->slots[level] = 0;
3104 } else {
3105 struct btrfs_key key;
3106 struct btrfs_disk_key found_key;
3107 struct extent_buffer *node;
3108
3109 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
3110 level = root_item->drop_level;
3111 path->lowest_level = level;
3112 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3113 if (wret < 0) {
3114 ret = wret;
3115 goto out;
3116 }
3117 node = path->nodes[level];
3118 btrfs_node_key(node, &found_key, path->slots[level]);
3119 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
3120 sizeof(found_key)));
3121 /*
3122 * unlock our path, this is safe because only this
3123 * function is allowed to delete this snapshot
3124 */
3125 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
3126 if (path->nodes[i] && path->locks[i]) {
3127 path->locks[i] = 0;
3128 btrfs_tree_unlock(path->nodes[i]);
3129 }
3130 }
3131 }
3132 while(1) {
3133 wret = walk_down_tree(trans, root, path, &level);
3134 if (wret > 0)
3135 break;
3136 if (wret < 0)
3137 ret = wret;
3138
3139 wret = walk_up_tree(trans, root, path, &level,
3140 BTRFS_MAX_LEVEL);
3141 if (wret > 0)
3142 break;
3143 if (wret < 0)
3144 ret = wret;
3145 if (trans->transaction->in_commit) {
3146 ret = -EAGAIN;
3147 break;
3148 }
3149 atomic_inc(&root->fs_info->throttle_gen);
3150 wake_up(&root->fs_info->transaction_throttle);
3151 }
3152 for (i = 0; i <= orig_level; i++) {
3153 if (path->nodes[i]) {
3154 free_extent_buffer(path->nodes[i]);
3155 path->nodes[i] = NULL;
3156 }
3157 }
3158 out:
3159 btrfs_free_path(path);
3160 return ret;
3161 }
3162
3163 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
3164 struct btrfs_root *root,
3165 struct extent_buffer *node,
3166 struct extent_buffer *parent)
3167 {
3168 struct btrfs_path *path;
3169 int level;
3170 int parent_level;
3171 int ret = 0;
3172 int wret;
3173
3174 path = btrfs_alloc_path();
3175 BUG_ON(!path);
3176
3177 BUG_ON(!btrfs_tree_locked(parent));
3178 parent_level = btrfs_header_level(parent);
3179 extent_buffer_get(parent);
3180 path->nodes[parent_level] = parent;
3181 path->slots[parent_level] = btrfs_header_nritems(parent);
3182
3183 BUG_ON(!btrfs_tree_locked(node));
3184 level = btrfs_header_level(node);
3185 extent_buffer_get(node);
3186 path->nodes[level] = node;
3187 path->slots[level] = 0;
3188
3189 while (1) {
3190 wret = walk_down_subtree(trans, root, path, &level);
3191 if (wret < 0)
3192 ret = wret;
3193 if (wret != 0)
3194 break;
3195
3196 wret = walk_up_tree(trans, root, path, &level, parent_level);
3197 if (wret < 0)
3198 ret = wret;
3199 if (wret != 0)
3200 break;
3201 }
3202
3203 btrfs_free_path(path);
3204 return ret;
3205 }
3206
3207 static unsigned long calc_ra(unsigned long start, unsigned long last,
3208 unsigned long nr)
3209 {
3210 return min(last, start + nr - 1);
3211 }
3212
3213 static int noinline relocate_inode_pages(struct inode *inode, u64 start,
3214 u64 len)
3215 {
3216 u64 page_start;
3217 u64 page_end;
3218 unsigned long first_index;
3219 unsigned long last_index;
3220 unsigned long i;
3221 struct page *page;
3222 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3223 struct file_ra_state *ra;
3224 struct btrfs_ordered_extent *ordered;
3225 unsigned int total_read = 0;
3226 unsigned int total_dirty = 0;
3227 int ret = 0;
3228
3229 ra = kzalloc(sizeof(*ra), GFP_NOFS);
3230
3231 mutex_lock(&inode->i_mutex);
3232 first_index = start >> PAGE_CACHE_SHIFT;
3233 last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
3234
3235 /* make sure the dirty trick played by the caller work */
3236 ret = invalidate_inode_pages2_range(inode->i_mapping,
3237 first_index, last_index);
3238 if (ret)
3239 goto out_unlock;
3240
3241 file_ra_state_init(ra, inode->i_mapping);
3242
3243 for (i = first_index ; i <= last_index; i++) {
3244 if (total_read % ra->ra_pages == 0) {
3245 btrfs_force_ra(inode->i_mapping, ra, NULL, i,
3246 calc_ra(i, last_index, ra->ra_pages));
3247 }
3248 total_read++;
3249 again:
3250 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
3251 BUG_ON(1);
3252 page = grab_cache_page(inode->i_mapping, i);
3253 if (!page) {
3254 ret = -ENOMEM;
3255 goto out_unlock;
3256 }
3257 if (!PageUptodate(page)) {
3258 btrfs_readpage(NULL, page);
3259 lock_page(page);
3260 if (!PageUptodate(page)) {
3261 unlock_page(page);
3262 page_cache_release(page);
3263 ret = -EIO;
3264 goto out_unlock;
3265 }
3266 }
3267 wait_on_page_writeback(page);
3268
3269 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
3270 page_end = page_start + PAGE_CACHE_SIZE - 1;
3271 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
3272
3273 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3274 if (ordered) {
3275 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3276 unlock_page(page);
3277 page_cache_release(page);
3278 btrfs_start_ordered_extent(inode, ordered, 1);
3279 btrfs_put_ordered_extent(ordered);
3280 goto again;
3281 }
3282 set_page_extent_mapped(page);
3283
3284 btrfs_set_extent_delalloc(inode, page_start, page_end);
3285 if (i == first_index)
3286 set_extent_bits(io_tree, page_start, page_end,
3287 EXTENT_BOUNDARY, GFP_NOFS);
3288
3289 set_page_dirty(page);
3290 total_dirty++;
3291
3292 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3293 unlock_page(page);
3294 page_cache_release(page);
3295 }
3296
3297 out_unlock:
3298 kfree(ra);
3299 mutex_unlock(&inode->i_mutex);
3300 balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
3301 return ret;
3302 }
3303
3304 static int noinline relocate_data_extent(struct inode *reloc_inode,
3305 struct btrfs_key *extent_key,
3306 u64 offset)
3307 {
3308 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
3309 struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
3310 struct extent_map *em;
3311 u64 start = extent_key->objectid - offset;
3312 u64 end = start + extent_key->offset - 1;
3313
3314 em = alloc_extent_map(GFP_NOFS);
3315 BUG_ON(!em || IS_ERR(em));
3316
3317 em->start = start;
3318 em->len = extent_key->offset;
3319 em->block_len = extent_key->offset;
3320 em->block_start = extent_key->objectid;
3321 em->bdev = root->fs_info->fs_devices->latest_bdev;
3322 set_bit(EXTENT_FLAG_PINNED, &em->flags);
3323
3324 /* setup extent map to cheat btrfs_readpage */
3325 lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
3326 while (1) {
3327 int ret;
3328 spin_lock(&em_tree->lock);
3329 ret = add_extent_mapping(em_tree, em);
3330 spin_unlock(&em_tree->lock);
3331 if (ret != -EEXIST) {
3332 free_extent_map(em);
3333 break;
3334 }
3335 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
3336 }
3337 unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
3338
3339 return relocate_inode_pages(reloc_inode, start, extent_key->offset);
3340 }
3341
3342 struct btrfs_ref_path {
3343 u64 extent_start;
3344 u64 nodes[BTRFS_MAX_LEVEL];
3345 u64 root_objectid;
3346 u64 root_generation;
3347 u64 owner_objectid;
3348 u32 num_refs;
3349 int lowest_level;
3350 int current_level;
3351 int shared_level;
3352
3353 struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
3354 u64 new_nodes[BTRFS_MAX_LEVEL];
3355 };
3356
3357 struct disk_extent {
3358 u64 ram_bytes;
3359 u64 disk_bytenr;
3360 u64 disk_num_bytes;
3361 u64 offset;
3362 u64 num_bytes;
3363 u8 compression;
3364 u8 encryption;
3365 u16 other_encoding;
3366 };
3367
3368 static int is_cowonly_root(u64 root_objectid)
3369 {
3370 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
3371 root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
3372 root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
3373 root_objectid == BTRFS_DEV_TREE_OBJECTID ||
3374 root_objectid == BTRFS_TREE_LOG_OBJECTID)
3375 return 1;
3376 return 0;
3377 }
3378
3379 static int noinline __next_ref_path(struct btrfs_trans_handle *trans,
3380 struct btrfs_root *extent_root,
3381 struct btrfs_ref_path *ref_path,
3382 int first_time)
3383 {
3384 struct extent_buffer *leaf;
3385 struct btrfs_path *path;
3386 struct btrfs_extent_ref *ref;
3387 struct btrfs_key key;
3388 struct btrfs_key found_key;
3389 u64 bytenr;
3390 u32 nritems;
3391 int level;
3392 int ret = 1;
3393
3394 path = btrfs_alloc_path();
3395 if (!path)
3396 return -ENOMEM;
3397
3398 if (first_time) {
3399 ref_path->lowest_level = -1;
3400 ref_path->current_level = -1;
3401 ref_path->shared_level = -1;
3402 goto walk_up;
3403 }
3404 walk_down:
3405 level = ref_path->current_level - 1;
3406 while (level >= -1) {
3407 u64 parent;
3408 if (level < ref_path->lowest_level)
3409 break;
3410
3411 if (level >= 0) {
3412 bytenr = ref_path->nodes[level];
3413 } else {
3414 bytenr = ref_path->extent_start;
3415 }
3416 BUG_ON(bytenr == 0);
3417
3418 parent = ref_path->nodes[level + 1];
3419 ref_path->nodes[level + 1] = 0;
3420 ref_path->current_level = level;
3421 BUG_ON(parent == 0);
3422
3423 key.objectid = bytenr;
3424 key.offset = parent + 1;
3425 key.type = BTRFS_EXTENT_REF_KEY;
3426
3427 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
3428 if (ret < 0)
3429 goto out;
3430 BUG_ON(ret == 0);
3431
3432 leaf = path->nodes[0];
3433 nritems = btrfs_header_nritems(leaf);
3434 if (path->slots[0] >= nritems) {
3435 ret = btrfs_next_leaf(extent_root, path);
3436 if (ret < 0)
3437 goto out;
3438 if (ret > 0)
3439 goto next;
3440 leaf = path->nodes[0];
3441 }
3442
3443 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3444 if (found_key.objectid == bytenr &&
3445 found_key.type == BTRFS_EXTENT_REF_KEY) {
3446 if (level < ref_path->shared_level)
3447 ref_path->shared_level = level;
3448 goto found;
3449 }
3450 next:
3451 level--;
3452 btrfs_release_path(extent_root, path);
3453 cond_resched();
3454 }
3455 /* reached lowest level */
3456 ret = 1;
3457 goto out;
3458 walk_up:
3459 level = ref_path->current_level;
3460 while (level < BTRFS_MAX_LEVEL - 1) {
3461 u64 ref_objectid;
3462 if (level >= 0) {
3463 bytenr = ref_path->nodes[level];
3464 } else {
3465 bytenr = ref_path->extent_start;
3466 }
3467 BUG_ON(bytenr == 0);
3468
3469 key.objectid = bytenr;
3470 key.offset = 0;
3471 key.type = BTRFS_EXTENT_REF_KEY;
3472
3473 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
3474 if (ret < 0)
3475 goto out;
3476
3477 leaf = path->nodes[0];
3478 nritems = btrfs_header_nritems(leaf);
3479 if (path->slots[0] >= nritems) {
3480 ret = btrfs_next_leaf(extent_root, path);
3481 if (ret < 0)
3482 goto out;
3483 if (ret > 0) {
3484 /* the extent was freed by someone */
3485 if (ref_path->lowest_level == level)
3486 goto out;
3487 btrfs_release_path(extent_root, path);
3488 goto walk_down;
3489 }
3490 leaf = path->nodes[0];
3491 }
3492
3493 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3494 if (found_key.objectid != bytenr ||
3495 found_key.type != BTRFS_EXTENT_REF_KEY) {
3496 /* the extent was freed by someone */
3497 if (ref_path->lowest_level == level) {
3498 ret = 1;
3499 goto out;
3500 }
3501 btrfs_release_path(extent_root, path);
3502 goto walk_down;
3503 }
3504 found:
3505 ref = btrfs_item_ptr(leaf, path->slots[0],
3506 struct btrfs_extent_ref);
3507 ref_objectid = btrfs_ref_objectid(leaf, ref);
3508 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
3509 if (first_time) {
3510 level = (int)ref_objectid;
3511 BUG_ON(level >= BTRFS_MAX_LEVEL);
3512 ref_path->lowest_level = level;
3513 ref_path->current_level = level;
3514 ref_path->nodes[level] = bytenr;
3515 } else {
3516 WARN_ON(ref_objectid != level);
3517 }
3518 } else {
3519 WARN_ON(level != -1);
3520 }
3521 first_time = 0;
3522
3523 if (ref_path->lowest_level == level) {
3524 ref_path->owner_objectid = ref_objectid;
3525 ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
3526 }
3527
3528 /*
3529 * the block is tree root or the block isn't in reference
3530 * counted tree.
3531 */
3532 if (found_key.objectid == found_key.offset ||
3533 is_cowonly_root(btrfs_ref_root(leaf, ref))) {
3534 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
3535 ref_path->root_generation =
3536 btrfs_ref_generation(leaf, ref);
3537 if (level < 0) {
3538 /* special reference from the tree log */
3539 ref_path->nodes[0] = found_key.offset;
3540 ref_path->current_level = 0;
3541 }
3542 ret = 0;
3543 goto out;
3544 }
3545
3546 level++;
3547 BUG_ON(ref_path->nodes[level] != 0);
3548 ref_path->nodes[level] = found_key.offset;
3549 ref_path->current_level = level;
3550
3551 /*
3552 * the reference was created in the running transaction,
3553 * no need to continue walking up.
3554 */
3555 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
3556 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
3557 ref_path->root_generation =
3558 btrfs_ref_generation(leaf, ref);
3559 ret = 0;
3560 goto out;
3561 }
3562
3563 btrfs_release_path(extent_root, path);
3564 cond_resched();
3565 }
3566 /* reached max tree level, but no tree root found. */
3567 BUG();
3568 out:
3569 btrfs_free_path(path);
3570 return ret;
3571 }
3572
3573 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
3574 struct btrfs_root *extent_root,
3575 struct btrfs_ref_path *ref_path,
3576 u64 extent_start)
3577 {
3578 memset(ref_path, 0, sizeof(*ref_path));
3579 ref_path->extent_start = extent_start;
3580
3581 return __next_ref_path(trans, extent_root, ref_path, 1);
3582 }
3583
3584 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
3585 struct btrfs_root *extent_root,
3586 struct btrfs_ref_path *ref_path)
3587 {
3588 return __next_ref_path(trans, extent_root, ref_path, 0);
3589 }
3590
3591 static int noinline get_new_locations(struct inode *reloc_inode,
3592 struct btrfs_key *extent_key,
3593 u64 offset, int no_fragment,
3594 struct disk_extent **extents,
3595 int *nr_extents)
3596 {
3597 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
3598 struct btrfs_path *path;
3599 struct btrfs_file_extent_item *fi;
3600 struct extent_buffer *leaf;
3601 struct disk_extent *exts = *extents;
3602 struct btrfs_key found_key;
3603 u64 cur_pos;
3604 u64 last_byte;
3605 u32 nritems;
3606 int nr = 0;
3607 int max = *nr_extents;
3608 int ret;
3609
3610 WARN_ON(!no_fragment && *extents);
3611 if (!exts) {
3612 max = 1;
3613 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
3614 if (!exts)
3615 return -ENOMEM;
3616 }
3617
3618 path = btrfs_alloc_path();
3619 BUG_ON(!path);
3620
3621 cur_pos = extent_key->objectid - offset;
3622 last_byte = extent_key->objectid + extent_key->offset;
3623 ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
3624 cur_pos, 0);
3625 if (ret < 0)
3626 goto out;
3627 if (ret > 0) {
3628 ret = -ENOENT;
3629 goto out;
3630 }
3631
3632 while (1) {
3633 leaf = path->nodes[0];
3634 nritems = btrfs_header_nritems(leaf);
3635 if (path->slots[0] >= nritems) {
3636 ret = btrfs_next_leaf(root, path);
3637 if (ret < 0)
3638 goto out;
3639 if (ret > 0)
3640 break;
3641 leaf = path->nodes[0];
3642 }
3643
3644 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3645 if (found_key.offset != cur_pos ||
3646 found_key.type != BTRFS_EXTENT_DATA_KEY ||
3647 found_key.objectid != reloc_inode->i_ino)
3648 break;
3649
3650 fi = btrfs_item_ptr(leaf, path->slots[0],
3651 struct btrfs_file_extent_item);
3652 if (btrfs_file_extent_type(leaf, fi) !=
3653 BTRFS_FILE_EXTENT_REG ||
3654 btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
3655 break;
3656
3657 if (nr == max) {
3658 struct disk_extent *old = exts;
3659 max *= 2;
3660 exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
3661 memcpy(exts, old, sizeof(*exts) * nr);
3662 if (old != *extents)
3663 kfree(old);
3664 }
3665
3666 exts[nr].disk_bytenr =
3667 btrfs_file_extent_disk_bytenr(leaf, fi);
3668 exts[nr].disk_num_bytes =
3669 btrfs_file_extent_disk_num_bytes(leaf, fi);
3670 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
3671 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
3672 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
3673 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
3674 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
3675 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
3676 fi);
3677 BUG_ON(exts[nr].offset > 0);
3678 BUG_ON(exts[nr].compression || exts[nr].encryption);
3679 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
3680
3681 cur_pos += exts[nr].num_bytes;
3682 nr++;
3683
3684 if (cur_pos + offset >= last_byte)
3685 break;
3686
3687 if (no_fragment) {
3688 ret = 1;
3689 goto out;
3690 }
3691 path->slots[0]++;
3692 }
3693
3694 WARN_ON(cur_pos + offset > last_byte);
3695 if (cur_pos + offset < last_byte) {
3696 ret = -ENOENT;
3697 goto out;
3698 }
3699 ret = 0;
3700 out:
3701 btrfs_free_path(path);
3702 if (ret) {
3703 if (exts != *extents)
3704 kfree(exts);
3705 } else {
3706 *extents = exts;
3707 *nr_extents = nr;
3708 }
3709 return ret;
3710 }
3711
3712 static int noinline replace_one_extent(struct btrfs_trans_handle *trans,
3713 struct btrfs_root *root,
3714 struct btrfs_path *path,
3715 struct btrfs_key *extent_key,
3716 struct btrfs_key *leaf_key,
3717 struct btrfs_ref_path *ref_path,
3718 struct disk_extent *new_extents,
3719 int nr_extents)
3720 {
3721 struct extent_buffer *leaf;
3722 struct btrfs_file_extent_item *fi;
3723 struct inode *inode = NULL;
3724 struct btrfs_key key;
3725 u64 lock_start = 0;
3726 u64 lock_end = 0;
3727 u64 num_bytes;
3728 u64 ext_offset;
3729 u64 first_pos;
3730 u32 nritems;
3731 int nr_scaned = 0;
3732 int extent_locked = 0;
3733 int extent_type;
3734 int ret;
3735
3736 memcpy(&key, leaf_key, sizeof(key));
3737 first_pos = INT_LIMIT(loff_t) - extent_key->offset;
3738 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
3739 if (key.objectid < ref_path->owner_objectid ||
3740 (key.objectid == ref_path->owner_objectid &&
3741 key.type < BTRFS_EXTENT_DATA_KEY)) {
3742 key.objectid = ref_path->owner_objectid;
3743 key.type = BTRFS_EXTENT_DATA_KEY;
3744 key.offset = 0;
3745 }
3746 }
3747
3748 while (1) {
3749 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
3750 if (ret < 0)
3751 goto out;
3752
3753 leaf = path->nodes[0];
3754 nritems = btrfs_header_nritems(leaf);
3755 next:
3756 if (extent_locked && ret > 0) {
3757 /*
3758 * the file extent item was modified by someone
3759 * before the extent got locked.
3760 */
3761 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
3762 lock_end, GFP_NOFS);
3763 extent_locked = 0;
3764 }
3765
3766 if (path->slots[0] >= nritems) {
3767 if (++nr_scaned > 2)
3768 break;
3769
3770 BUG_ON(extent_locked);
3771 ret = btrfs_next_leaf(root, path);
3772 if (ret < 0)
3773 goto out;
3774 if (ret > 0)
3775 break;
3776 leaf = path->nodes[0];
3777 nritems = btrfs_header_nritems(leaf);
3778 }
3779
3780 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3781
3782 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
3783 if ((key.objectid > ref_path->owner_objectid) ||
3784 (key.objectid == ref_path->owner_objectid &&
3785 key.type > BTRFS_EXTENT_DATA_KEY) ||
3786 (key.offset >= first_pos + extent_key->offset))
3787 break;
3788 }
3789
3790 if (inode && key.objectid != inode->i_ino) {
3791 BUG_ON(extent_locked);
3792 btrfs_release_path(root, path);
3793 mutex_unlock(&inode->i_mutex);
3794 iput(inode);
3795 inode = NULL;
3796 continue;
3797 }
3798
3799 if (key.type != BTRFS_EXTENT_DATA_KEY) {
3800 path->slots[0]++;
3801 ret = 1;
3802 goto next;
3803 }
3804 fi = btrfs_item_ptr(leaf, path->slots[0],
3805 struct btrfs_file_extent_item);
3806 extent_type = btrfs_file_extent_type(leaf, fi);
3807 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
3808 extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
3809 (btrfs_file_extent_disk_bytenr(leaf, fi) !=
3810 extent_key->objectid)) {
3811 path->slots[0]++;
3812 ret = 1;
3813 goto next;
3814 }
3815
3816 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
3817 ext_offset = btrfs_file_extent_offset(leaf, fi);
3818
3819 if (first_pos > key.offset - ext_offset)
3820 first_pos = key.offset - ext_offset;
3821
3822 if (!extent_locked) {
3823 lock_start = key.offset;
3824 lock_end = lock_start + num_bytes - 1;
3825 } else {
3826 if (lock_start > key.offset ||
3827 lock_end + 1 < key.offset + num_bytes) {
3828 unlock_extent(&BTRFS_I(inode)->io_tree,
3829 lock_start, lock_end, GFP_NOFS);
3830 extent_locked = 0;
3831 }
3832 }
3833
3834 if (!inode) {
3835 btrfs_release_path(root, path);
3836
3837 inode = btrfs_iget_locked(root->fs_info->sb,
3838 key.objectid, root);
3839 if (inode->i_state & I_NEW) {
3840 BTRFS_I(inode)->root = root;
3841 BTRFS_I(inode)->location.objectid =
3842 key.objectid;
3843 BTRFS_I(inode)->location.type =
3844 BTRFS_INODE_ITEM_KEY;
3845 BTRFS_I(inode)->location.offset = 0;
3846 btrfs_read_locked_inode(inode);
3847 unlock_new_inode(inode);
3848 }
3849 /*
3850 * some code call btrfs_commit_transaction while
3851 * holding the i_mutex, so we can't use mutex_lock
3852 * here.
3853 */
3854 if (is_bad_inode(inode) ||
3855 !mutex_trylock(&inode->i_mutex)) {
3856 iput(inode);
3857 inode = NULL;
3858 key.offset = (u64)-1;
3859 goto skip;
3860 }
3861 }
3862
3863 if (!extent_locked) {
3864 struct btrfs_ordered_extent *ordered;
3865
3866 btrfs_release_path(root, path);
3867
3868 lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
3869 lock_end, GFP_NOFS);
3870 ordered = btrfs_lookup_first_ordered_extent(inode,
3871 lock_end);
3872 if (ordered &&
3873 ordered->file_offset <= lock_end &&
3874 ordered->file_offset + ordered->len > lock_start) {
3875 unlock_extent(&BTRFS_I(inode)->io_tree,
3876 lock_start, lock_end, GFP_NOFS);
3877 btrfs_start_ordered_extent(inode, ordered, 1);
3878 btrfs_put_ordered_extent(ordered);
3879 key.offset += num_bytes;
3880 goto skip;
3881 }
3882 if (ordered)
3883 btrfs_put_ordered_extent(ordered);
3884
3885 extent_locked = 1;
3886 continue;
3887 }
3888
3889 if (nr_extents == 1) {
3890 /* update extent pointer in place */
3891 btrfs_set_file_extent_disk_bytenr(leaf, fi,
3892 new_extents[0].disk_bytenr);
3893 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
3894 new_extents[0].disk_num_bytes);
3895 btrfs_mark_buffer_dirty(leaf);
3896
3897 btrfs_drop_extent_cache(inode, key.offset,
3898 key.offset + num_bytes - 1, 0);
3899
3900 ret = btrfs_inc_extent_ref(trans, root,
3901 new_extents[0].disk_bytenr,
3902 new_extents[0].disk_num_bytes,
3903 leaf->start,
3904 root->root_key.objectid,
3905 trans->transid,
3906 key.objectid);
3907 BUG_ON(ret);
3908
3909 ret = btrfs_free_extent(trans, root,
3910 extent_key->objectid,
3911 extent_key->offset,
3912 leaf->start,
3913 btrfs_header_owner(leaf),
3914 btrfs_header_generation(leaf),
3915 key.objectid, 0);
3916 BUG_ON(ret);
3917
3918 btrfs_release_path(root, path);
3919 key.offset += num_bytes;
3920 } else {
3921 BUG_ON(1);
3922 #if 0
3923 u64 alloc_hint;
3924 u64 extent_len;
3925 int i;
3926 /*
3927 * drop old extent pointer at first, then insert the
3928 * new pointers one bye one
3929 */
3930 btrfs_release_path(root, path);
3931 ret = btrfs_drop_extents(trans, root, inode, key.offset,
3932 key.offset + num_bytes,
3933 key.offset, &alloc_hint);
3934 BUG_ON(ret);
3935
3936 for (i = 0; i < nr_extents; i++) {
3937 if (ext_offset >= new_extents[i].num_bytes) {
3938 ext_offset -= new_extents[i].num_bytes;
3939 continue;
3940 }
3941 extent_len = min(new_extents[i].num_bytes -
3942 ext_offset, num_bytes);
3943
3944 ret = btrfs_insert_empty_item(trans, root,
3945 path, &key,
3946 sizeof(*fi));
3947 BUG_ON(ret);
3948
3949 leaf = path->nodes[0];
3950 fi = btrfs_item_ptr(leaf, path->slots[0],
3951 struct btrfs_file_extent_item);
3952 btrfs_set_file_extent_generation(leaf, fi,
3953 trans->transid);
3954 btrfs_set_file_extent_type(leaf, fi,
3955 BTRFS_FILE_EXTENT_REG);
3956 btrfs_set_file_extent_disk_bytenr(leaf, fi,
3957 new_extents[i].disk_bytenr);
3958 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
3959 new_extents[i].disk_num_bytes);
3960 btrfs_set_file_extent_ram_bytes(leaf, fi,
3961 new_extents[i].ram_bytes);
3962
3963 btrfs_set_file_extent_compression(leaf, fi,
3964 new_extents[i].compression);
3965 btrfs_set_file_extent_encryption(leaf, fi,
3966 new_extents[i].encryption);
3967 btrfs_set_file_extent_other_encoding(leaf, fi,
3968 new_extents[i].other_encoding);
3969
3970 btrfs_set_file_extent_num_bytes(leaf, fi,
3971 extent_len);
3972 ext_offset += new_extents[i].offset;
3973 btrfs_set_file_extent_offset(leaf, fi,
3974 ext_offset);
3975 btrfs_mark_buffer_dirty(leaf);
3976
3977 btrfs_drop_extent_cache(inode, key.offset,
3978 key.offset + extent_len - 1, 0);
3979
3980 ret = btrfs_inc_extent_ref(trans, root,
3981 new_extents[i].disk_bytenr,
3982 new_extents[i].disk_num_bytes,
3983 leaf->start,
3984 root->root_key.objectid,
3985 trans->transid, key.objectid);
3986 BUG_ON(ret);
3987 btrfs_release_path(root, path);
3988
3989 inode_add_bytes(inode, extent_len);
3990
3991 ext_offset = 0;
3992 num_bytes -= extent_len;
3993 key.offset += extent_len;
3994
3995 if (num_bytes == 0)
3996 break;
3997 }
3998 BUG_ON(i >= nr_extents);
3999 #endif
4000 }
4001
4002 if (extent_locked) {
4003 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4004 lock_end, GFP_NOFS);
4005 extent_locked = 0;
4006 }
4007 skip:
4008 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
4009 key.offset >= first_pos + extent_key->offset)
4010 break;
4011
4012 cond_resched();
4013 }
4014 ret = 0;
4015 out:
4016 btrfs_release_path(root, path);
4017 if (inode) {
4018 mutex_unlock(&inode->i_mutex);
4019 if (extent_locked) {
4020 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4021 lock_end, GFP_NOFS);
4022 }
4023 iput(inode);
4024 }
4025 return ret;
4026 }
4027
4028 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
4029 struct btrfs_root *root,
4030 struct extent_buffer *buf, u64 orig_start)
4031 {
4032 int level;
4033 int ret;
4034
4035 BUG_ON(btrfs_header_generation(buf) != trans->transid);
4036 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
4037
4038 level = btrfs_header_level(buf);
4039 if (level == 0) {
4040 struct btrfs_leaf_ref *ref;
4041 struct btrfs_leaf_ref *orig_ref;
4042
4043 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
4044 if (!orig_ref)
4045 return -ENOENT;
4046
4047 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
4048 if (!ref) {
4049 btrfs_free_leaf_ref(root, orig_ref);
4050 return -ENOMEM;
4051 }
4052
4053 ref->nritems = orig_ref->nritems;
4054 memcpy(ref->extents, orig_ref->extents,
4055 sizeof(ref->extents[0]) * ref->nritems);
4056
4057 btrfs_free_leaf_ref(root, orig_ref);
4058
4059 ref->root_gen = trans->transid;
4060 ref->bytenr = buf->start;
4061 ref->owner = btrfs_header_owner(buf);
4062 ref->generation = btrfs_header_generation(buf);
4063 ret = btrfs_add_leaf_ref(root, ref, 0);
4064 WARN_ON(ret);
4065 btrfs_free_leaf_ref(root, ref);
4066 }
4067 return 0;
4068 }
4069
4070 static int noinline invalidate_extent_cache(struct btrfs_root *root,
4071 struct extent_buffer *leaf,
4072 struct btrfs_block_group_cache *group,
4073 struct btrfs_root *target_root)
4074 {
4075 struct btrfs_key key;
4076 struct inode *inode = NULL;
4077 struct btrfs_file_extent_item *fi;
4078 u64 num_bytes;
4079 u64 skip_objectid = 0;
4080 u32 nritems;
4081 u32 i;
4082
4083 nritems = btrfs_header_nritems(leaf);
4084 for (i = 0; i < nritems; i++) {
4085 btrfs_item_key_to_cpu(leaf, &key, i);
4086 if (key.objectid == skip_objectid ||
4087 key.type != BTRFS_EXTENT_DATA_KEY)
4088 continue;
4089 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4090 if (btrfs_file_extent_type(leaf, fi) ==
4091 BTRFS_FILE_EXTENT_INLINE)
4092 continue;
4093 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
4094 continue;
4095 if (!inode || inode->i_ino != key.objectid) {
4096 iput(inode);
4097 inode = btrfs_ilookup(target_root->fs_info->sb,
4098 key.objectid, target_root, 1);
4099 }
4100 if (!inode) {
4101 skip_objectid = key.objectid;
4102 continue;
4103 }
4104 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4105
4106 lock_extent(&BTRFS_I(inode)->io_tree, key.offset,
4107 key.offset + num_bytes - 1, GFP_NOFS);
4108 btrfs_drop_extent_cache(inode, key.offset,
4109 key.offset + num_bytes - 1, 1);
4110 unlock_extent(&BTRFS_I(inode)->io_tree, key.offset,
4111 key.offset + num_bytes - 1, GFP_NOFS);
4112 cond_resched();
4113 }
4114 iput(inode);
4115 return 0;
4116 }
4117
4118 static int noinline replace_extents_in_leaf(struct btrfs_trans_handle *trans,
4119 struct btrfs_root *root,
4120 struct extent_buffer *leaf,
4121 struct btrfs_block_group_cache *group,
4122 struct inode *reloc_inode)
4123 {
4124 struct btrfs_key key;
4125 struct btrfs_key extent_key;
4126 struct btrfs_file_extent_item *fi;
4127 struct btrfs_leaf_ref *ref;
4128 struct disk_extent *new_extent;
4129 u64 bytenr;
4130 u64 num_bytes;
4131 u32 nritems;
4132 u32 i;
4133 int ext_index;
4134 int nr_extent;
4135 int ret;
4136
4137 new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
4138 BUG_ON(!new_extent);
4139
4140 ref = btrfs_lookup_leaf_ref(root, leaf->start);
4141 BUG_ON(!ref);
4142
4143 ext_index = -1;
4144 nritems = btrfs_header_nritems(leaf);
4145 for (i = 0; i < nritems; i++) {
4146 btrfs_item_key_to_cpu(leaf, &key, i);
4147 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
4148 continue;
4149 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4150 if (btrfs_file_extent_type(leaf, fi) ==
4151 BTRFS_FILE_EXTENT_INLINE)
4152 continue;
4153 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
4154 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
4155 if (bytenr == 0)
4156 continue;
4157
4158 ext_index++;
4159 if (bytenr >= group->key.objectid + group->key.offset ||
4160 bytenr + num_bytes <= group->key.objectid)
4161 continue;
4162
4163 extent_key.objectid = bytenr;
4164 extent_key.offset = num_bytes;
4165 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
4166 nr_extent = 1;
4167 ret = get_new_locations(reloc_inode, &extent_key,
4168 group->key.objectid, 1,
4169 &new_extent, &nr_extent);
4170 if (ret > 0)
4171 continue;
4172 BUG_ON(ret < 0);
4173
4174 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
4175 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
4176 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
4177 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
4178
4179 btrfs_set_file_extent_disk_bytenr(leaf, fi,
4180 new_extent->disk_bytenr);
4181 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4182 new_extent->disk_num_bytes);
4183 btrfs_mark_buffer_dirty(leaf);
4184
4185 ret = btrfs_inc_extent_ref(trans, root,
4186 new_extent->disk_bytenr,
4187 new_extent->disk_num_bytes,
4188 leaf->start,
4189 root->root_key.objectid,
4190 trans->transid, key.objectid);
4191 BUG_ON(ret);
4192 ret = btrfs_free_extent(trans, root,
4193 bytenr, num_bytes, leaf->start,
4194 btrfs_header_owner(leaf),
4195 btrfs_header_generation(leaf),
4196 key.objectid, 0);
4197 BUG_ON(ret);
4198 cond_resched();
4199 }
4200 kfree(new_extent);
4201 BUG_ON(ext_index + 1 != ref->nritems);
4202 btrfs_free_leaf_ref(root, ref);
4203 return 0;
4204 }
4205
4206 int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
4207 struct btrfs_root *root)
4208 {
4209 struct btrfs_root *reloc_root;
4210 int ret;
4211
4212 if (root->reloc_root) {
4213 reloc_root = root->reloc_root;
4214 root->reloc_root = NULL;
4215 list_add(&reloc_root->dead_list,
4216 &root->fs_info->dead_reloc_roots);
4217
4218 btrfs_set_root_bytenr(&reloc_root->root_item,
4219 reloc_root->node->start);
4220 btrfs_set_root_level(&root->root_item,
4221 btrfs_header_level(reloc_root->node));
4222 memset(&reloc_root->root_item.drop_progress, 0,
4223 sizeof(struct btrfs_disk_key));
4224 reloc_root->root_item.drop_level = 0;
4225
4226 ret = btrfs_update_root(trans, root->fs_info->tree_root,
4227 &reloc_root->root_key,
4228 &reloc_root->root_item);
4229 BUG_ON(ret);
4230 }
4231 return 0;
4232 }
4233
4234 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
4235 {
4236 struct btrfs_trans_handle *trans;
4237 struct btrfs_root *reloc_root;
4238 struct btrfs_root *prev_root = NULL;
4239 struct list_head dead_roots;
4240 int ret;
4241 unsigned long nr;
4242
4243 INIT_LIST_HEAD(&dead_roots);
4244 list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
4245
4246 while (!list_empty(&dead_roots)) {
4247 reloc_root = list_entry(dead_roots.prev,
4248 struct btrfs_root, dead_list);
4249 list_del_init(&reloc_root->dead_list);
4250
4251 BUG_ON(reloc_root->commit_root != NULL);
4252 while (1) {
4253 trans = btrfs_join_transaction(root, 1);
4254 BUG_ON(!trans);
4255
4256 mutex_lock(&root->fs_info->drop_mutex);
4257 ret = btrfs_drop_snapshot(trans, reloc_root);
4258 if (ret != -EAGAIN)
4259 break;
4260 mutex_unlock(&root->fs_info->drop_mutex);
4261
4262 nr = trans->blocks_used;
4263 ret = btrfs_end_transaction(trans, root);
4264 BUG_ON(ret);
4265 btrfs_btree_balance_dirty(root, nr);
4266 }
4267
4268 free_extent_buffer(reloc_root->node);
4269
4270 ret = btrfs_del_root(trans, root->fs_info->tree_root,
4271 &reloc_root->root_key);
4272 BUG_ON(ret);
4273 mutex_unlock(&root->fs_info->drop_mutex);
4274
4275 nr = trans->blocks_used;
4276 ret = btrfs_end_transaction(trans, root);
4277 BUG_ON(ret);
4278 btrfs_btree_balance_dirty(root, nr);
4279
4280 kfree(prev_root);
4281 prev_root = reloc_root;
4282 }
4283 if (prev_root) {
4284 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
4285 kfree(prev_root);
4286 }
4287 return 0;
4288 }
4289
4290 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
4291 {
4292 list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
4293 return 0;
4294 }
4295
4296 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
4297 {
4298 struct btrfs_root *reloc_root;
4299 struct btrfs_trans_handle *trans;
4300 struct btrfs_key location;
4301 int found;
4302 int ret;
4303
4304 mutex_lock(&root->fs_info->tree_reloc_mutex);
4305 ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
4306 BUG_ON(ret);
4307 found = !list_empty(&root->fs_info->dead_reloc_roots);
4308 mutex_unlock(&root->fs_info->tree_reloc_mutex);
4309
4310 if (found) {
4311 trans = btrfs_start_transaction(root, 1);
4312 BUG_ON(!trans);
4313 ret = btrfs_commit_transaction(trans, root);
4314 BUG_ON(ret);
4315 }
4316
4317 location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
4318 location.offset = (u64)-1;
4319 location.type = BTRFS_ROOT_ITEM_KEY;
4320
4321 reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
4322 BUG_ON(!reloc_root);
4323 btrfs_orphan_cleanup(reloc_root);
4324 return 0;
4325 }
4326
4327 static int noinline init_reloc_tree(struct btrfs_trans_handle *trans,
4328 struct btrfs_root *root)
4329 {
4330 struct btrfs_root *reloc_root;
4331 struct extent_buffer *eb;
4332 struct btrfs_root_item *root_item;
4333 struct btrfs_key root_key;
4334 int ret;
4335
4336 BUG_ON(!root->ref_cows);
4337 if (root->reloc_root)
4338 return 0;
4339
4340 root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
4341 BUG_ON(!root_item);
4342
4343 ret = btrfs_copy_root(trans, root, root->commit_root,
4344 &eb, BTRFS_TREE_RELOC_OBJECTID);
4345 BUG_ON(ret);
4346
4347 root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
4348 root_key.offset = root->root_key.objectid;
4349 root_key.type = BTRFS_ROOT_ITEM_KEY;
4350
4351 memcpy(root_item, &root->root_item, sizeof(root_item));
4352 btrfs_set_root_refs(root_item, 0);
4353 btrfs_set_root_bytenr(root_item, eb->start);
4354 btrfs_set_root_level(root_item, btrfs_header_level(eb));
4355 btrfs_set_root_generation(root_item, trans->transid);
4356
4357 btrfs_tree_unlock(eb);
4358 free_extent_buffer(eb);
4359
4360 ret = btrfs_insert_root(trans, root->fs_info->tree_root,
4361 &root_key, root_item);
4362 BUG_ON(ret);
4363 kfree(root_item);
4364
4365 reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
4366 &root_key);
4367 BUG_ON(!reloc_root);
4368 reloc_root->last_trans = trans->transid;
4369 reloc_root->commit_root = NULL;
4370 reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
4371
4372 root->reloc_root = reloc_root;
4373 return 0;
4374 }
4375
4376 /*
4377 * Core function of space balance.
4378 *
4379 * The idea is using reloc trees to relocate tree blocks in reference
4380 * counted roots. There is one reloc tree for each subvol, and all
4381 * reloc trees share same root key objectid. Reloc trees are snapshots
4382 * of the latest committed roots of subvols (root->commit_root).
4383 *
4384 * To relocate a tree block referenced by a subvol, there are two steps.
4385 * COW the block through subvol's reloc tree, then update block pointer
4386 * in the subvol to point to the new block. Since all reloc trees share
4387 * same root key objectid, doing special handing for tree blocks owned
4388 * by them is easy. Once a tree block has been COWed in one reloc tree,
4389 * we can use the resulting new block directly when the same block is
4390 * required to COW again through other reloc trees. By this way, relocated
4391 * tree blocks are shared between reloc trees, so they are also shared
4392 * between subvols.
4393 */
4394 static int noinline relocate_one_path(struct btrfs_trans_handle *trans,
4395 struct btrfs_root *root,
4396 struct btrfs_path *path,
4397 struct btrfs_key *first_key,
4398 struct btrfs_ref_path *ref_path,
4399 struct btrfs_block_group_cache *group,
4400 struct inode *reloc_inode)
4401 {
4402 struct btrfs_root *reloc_root;
4403 struct extent_buffer *eb = NULL;
4404 struct btrfs_key *keys;
4405 u64 *nodes;
4406 int level;
4407 int shared_level;
4408 int lowest_level = 0;
4409 int ret;
4410
4411 if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
4412 lowest_level = ref_path->owner_objectid;
4413
4414 if (!root->ref_cows) {
4415 path->lowest_level = lowest_level;
4416 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
4417 BUG_ON(ret < 0);
4418 path->lowest_level = 0;
4419 btrfs_release_path(root, path);
4420 return 0;
4421 }
4422
4423 mutex_lock(&root->fs_info->tree_reloc_mutex);
4424 ret = init_reloc_tree(trans, root);
4425 BUG_ON(ret);
4426 reloc_root = root->reloc_root;
4427
4428 shared_level = ref_path->shared_level;
4429 ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
4430
4431 keys = ref_path->node_keys;
4432 nodes = ref_path->new_nodes;
4433 memset(&keys[shared_level + 1], 0,
4434 sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
4435 memset(&nodes[shared_level + 1], 0,
4436 sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
4437
4438 if (nodes[lowest_level] == 0) {
4439 path->lowest_level = lowest_level;
4440 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
4441 0, 1);
4442 BUG_ON(ret);
4443 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
4444 eb = path->nodes[level];
4445 if (!eb || eb == reloc_root->node)
4446 break;
4447 nodes[level] = eb->start;
4448 if (level == 0)
4449 btrfs_item_key_to_cpu(eb, &keys[level], 0);
4450 else
4451 btrfs_node_key_to_cpu(eb, &keys[level], 0);
4452 }
4453 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
4454 eb = path->nodes[0];
4455 ret = replace_extents_in_leaf(trans, reloc_root, eb,
4456 group, reloc_inode);
4457 BUG_ON(ret);
4458 }
4459 btrfs_release_path(reloc_root, path);
4460 } else {
4461 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
4462 lowest_level);
4463 BUG_ON(ret);
4464 }
4465
4466 /*
4467 * replace tree blocks in the fs tree with tree blocks in
4468 * the reloc tree.
4469 */
4470 ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
4471 BUG_ON(ret < 0);
4472
4473 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
4474 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
4475 0, 0);
4476 BUG_ON(ret);
4477 extent_buffer_get(path->nodes[0]);
4478 eb = path->nodes[0];
4479 btrfs_release_path(reloc_root, path);
4480 ret = invalidate_extent_cache(reloc_root, eb, group, root);
4481 BUG_ON(ret);
4482 free_extent_buffer(eb);
4483 }
4484
4485 mutex_unlock(&root->fs_info->tree_reloc_mutex);
4486 path->lowest_level = 0;
4487 return 0;
4488 }
4489
4490 static int noinline relocate_tree_block(struct btrfs_trans_handle *trans,
4491 struct btrfs_root *root,
4492 struct btrfs_path *path,
4493 struct btrfs_key *first_key,
4494 struct btrfs_ref_path *ref_path)
4495 {
4496 int ret;
4497
4498 ret = relocate_one_path(trans, root, path, first_key,
4499 ref_path, NULL, NULL);
4500 BUG_ON(ret);
4501
4502 if (root == root->fs_info->extent_root)
4503 btrfs_extent_post_op(trans, root);
4504
4505 return 0;
4506 }
4507
4508 static int noinline del_extent_zero(struct btrfs_trans_handle *trans,
4509 struct btrfs_root *extent_root,
4510 struct btrfs_path *path,
4511 struct btrfs_key *extent_key)
4512 {
4513 int ret;
4514
4515 ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
4516 if (ret)
4517 goto out;
4518 ret = btrfs_del_item(trans, extent_root, path);
4519 out:
4520 btrfs_release_path(extent_root, path);
4521 return ret;
4522 }
4523
4524 static struct btrfs_root noinline *read_ref_root(struct btrfs_fs_info *fs_info,
4525 struct btrfs_ref_path *ref_path)
4526 {
4527 struct btrfs_key root_key;
4528
4529 root_key.objectid = ref_path->root_objectid;
4530 root_key.type = BTRFS_ROOT_ITEM_KEY;
4531 if (is_cowonly_root(ref_path->root_objectid))
4532 root_key.offset = 0;
4533 else
4534 root_key.offset = (u64)-1;
4535
4536 return btrfs_read_fs_root_no_name(fs_info, &root_key);
4537 }
4538
4539 static int noinline relocate_one_extent(struct btrfs_root *extent_root,
4540 struct btrfs_path *path,
4541 struct btrfs_key *extent_key,
4542 struct btrfs_block_group_cache *group,
4543 struct inode *reloc_inode, int pass)
4544 {
4545 struct btrfs_trans_handle *trans;
4546 struct btrfs_root *found_root;
4547 struct btrfs_ref_path *ref_path = NULL;
4548 struct disk_extent *new_extents = NULL;
4549 int nr_extents = 0;
4550 int loops;
4551 int ret;
4552 int level;
4553 struct btrfs_key first_key;
4554 u64 prev_block = 0;
4555
4556
4557 trans = btrfs_start_transaction(extent_root, 1);
4558 BUG_ON(!trans);
4559
4560 if (extent_key->objectid == 0) {
4561 ret = del_extent_zero(trans, extent_root, path, extent_key);
4562 goto out;
4563 }
4564
4565 ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
4566 if (!ref_path) {
4567 ret = -ENOMEM;
4568 goto out;
4569 }
4570
4571 for (loops = 0; ; loops++) {
4572 if (loops == 0) {
4573 ret = btrfs_first_ref_path(trans, extent_root, ref_path,
4574 extent_key->objectid);
4575 } else {
4576 ret = btrfs_next_ref_path(trans, extent_root, ref_path);
4577 }
4578 if (ret < 0)
4579 goto out;
4580 if (ret > 0)
4581 break;
4582
4583 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
4584 ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
4585 continue;
4586
4587 found_root = read_ref_root(extent_root->fs_info, ref_path);
4588 BUG_ON(!found_root);
4589 /*
4590 * for reference counted tree, only process reference paths
4591 * rooted at the latest committed root.
4592 */
4593 if (found_root->ref_cows &&
4594 ref_path->root_generation != found_root->root_key.offset)
4595 continue;
4596
4597 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
4598 if (pass == 0) {
4599 /*
4600 * copy data extents to new locations
4601 */
4602 u64 group_start = group->key.objectid;
4603 ret = relocate_data_extent(reloc_inode,
4604 extent_key,
4605 group_start);
4606 if (ret < 0)
4607 goto out;
4608 break;
4609 }
4610 level = 0;
4611 } else {
4612 level = ref_path->owner_objectid;
4613 }
4614
4615 if (prev_block != ref_path->nodes[level]) {
4616 struct extent_buffer *eb;
4617 u64 block_start = ref_path->nodes[level];
4618 u64 block_size = btrfs_level_size(found_root, level);
4619
4620 eb = read_tree_block(found_root, block_start,
4621 block_size, 0);
4622 btrfs_tree_lock(eb);
4623 BUG_ON(level != btrfs_header_level(eb));
4624
4625 if (level == 0)
4626 btrfs_item_key_to_cpu(eb, &first_key, 0);
4627 else
4628 btrfs_node_key_to_cpu(eb, &first_key, 0);
4629
4630 btrfs_tree_unlock(eb);
4631 free_extent_buffer(eb);
4632 prev_block = block_start;
4633 }
4634
4635 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID &&
4636 pass >= 2) {
4637 /*
4638 * use fallback method to process the remaining
4639 * references.
4640 */
4641 if (!new_extents) {
4642 u64 group_start = group->key.objectid;
4643 new_extents = kmalloc(sizeof(*new_extents),
4644 GFP_NOFS);
4645 nr_extents = 1;
4646 ret = get_new_locations(reloc_inode,
4647 extent_key,
4648 group_start, 1,
4649 &new_extents,
4650 &nr_extents);
4651 if (ret)
4652 goto out;
4653 }
4654 btrfs_record_root_in_trans(found_root);
4655 ret = replace_one_extent(trans, found_root,
4656 path, extent_key,
4657 &first_key, ref_path,
4658 new_extents, nr_extents);
4659 if (ret < 0)
4660 goto out;
4661 continue;
4662 }
4663
4664 btrfs_record_root_in_trans(found_root);
4665 if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4666 ret = relocate_tree_block(trans, found_root, path,
4667 &first_key, ref_path);
4668 } else {
4669 /*
4670 * try to update data extent references while
4671 * keeping metadata shared between snapshots.
4672 */
4673 ret = relocate_one_path(trans, found_root, path,
4674 &first_key, ref_path,
4675 group, reloc_inode);
4676 }
4677 if (ret < 0)
4678 goto out;
4679 }
4680 ret = 0;
4681 out:
4682 btrfs_end_transaction(trans, extent_root);
4683 kfree(new_extents);
4684 kfree(ref_path);
4685 return ret;
4686 }
4687
4688 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
4689 {
4690 u64 num_devices;
4691 u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
4692 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
4693
4694 num_devices = root->fs_info->fs_devices->num_devices;
4695 if (num_devices == 1) {
4696 stripped |= BTRFS_BLOCK_GROUP_DUP;
4697 stripped = flags & ~stripped;
4698
4699 /* turn raid0 into single device chunks */
4700 if (flags & BTRFS_BLOCK_GROUP_RAID0)
4701 return stripped;
4702
4703 /* turn mirroring into duplication */
4704 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
4705 BTRFS_BLOCK_GROUP_RAID10))
4706 return stripped | BTRFS_BLOCK_GROUP_DUP;
4707 return flags;
4708 } else {
4709 /* they already had raid on here, just return */
4710 if (flags & stripped)
4711 return flags;
4712
4713 stripped |= BTRFS_BLOCK_GROUP_DUP;
4714 stripped = flags & ~stripped;
4715
4716 /* switch duplicated blocks with raid1 */
4717 if (flags & BTRFS_BLOCK_GROUP_DUP)
4718 return stripped | BTRFS_BLOCK_GROUP_RAID1;
4719
4720 /* turn single device chunks into raid0 */
4721 return stripped | BTRFS_BLOCK_GROUP_RAID0;
4722 }
4723 return flags;
4724 }
4725
4726 int __alloc_chunk_for_shrink(struct btrfs_root *root,
4727 struct btrfs_block_group_cache *shrink_block_group,
4728 int force)
4729 {
4730 struct btrfs_trans_handle *trans;
4731 u64 new_alloc_flags;
4732 u64 calc;
4733
4734 spin_lock(&shrink_block_group->lock);
4735 if (btrfs_block_group_used(&shrink_block_group->item) > 0) {
4736 spin_unlock(&shrink_block_group->lock);
4737
4738 trans = btrfs_start_transaction(root, 1);
4739 spin_lock(&shrink_block_group->lock);
4740
4741 new_alloc_flags = update_block_group_flags(root,
4742 shrink_block_group->flags);
4743 if (new_alloc_flags != shrink_block_group->flags) {
4744 calc =
4745 btrfs_block_group_used(&shrink_block_group->item);
4746 } else {
4747 calc = shrink_block_group->key.offset;
4748 }
4749 spin_unlock(&shrink_block_group->lock);
4750
4751 do_chunk_alloc(trans, root->fs_info->extent_root,
4752 calc + 2 * 1024 * 1024, new_alloc_flags, force);
4753
4754 btrfs_end_transaction(trans, root);
4755 } else
4756 spin_unlock(&shrink_block_group->lock);
4757 return 0;
4758 }
4759
4760 static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
4761 struct btrfs_root *root,
4762 u64 objectid, u64 size)
4763 {
4764 struct btrfs_path *path;
4765 struct btrfs_inode_item *item;
4766 struct extent_buffer *leaf;
4767 int ret;
4768
4769 path = btrfs_alloc_path();
4770 if (!path)
4771 return -ENOMEM;
4772
4773 ret = btrfs_insert_empty_inode(trans, root, path, objectid);
4774 if (ret)
4775 goto out;
4776
4777 leaf = path->nodes[0];
4778 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
4779 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
4780 btrfs_set_inode_generation(leaf, item, 1);
4781 btrfs_set_inode_size(leaf, item, size);
4782 btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
4783 btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NODATASUM |
4784 BTRFS_INODE_NOCOMPRESS);
4785 btrfs_mark_buffer_dirty(leaf);
4786 btrfs_release_path(root, path);
4787 out:
4788 btrfs_free_path(path);
4789 return ret;
4790 }
4791
4792 static struct inode noinline *create_reloc_inode(struct btrfs_fs_info *fs_info,
4793 struct btrfs_block_group_cache *group)
4794 {
4795 struct inode *inode = NULL;
4796 struct btrfs_trans_handle *trans;
4797 struct btrfs_root *root;
4798 struct btrfs_key root_key;
4799 u64 objectid = BTRFS_FIRST_FREE_OBJECTID;
4800 int err = 0;
4801
4802 root_key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
4803 root_key.type = BTRFS_ROOT_ITEM_KEY;
4804 root_key.offset = (u64)-1;
4805 root = btrfs_read_fs_root_no_name(fs_info, &root_key);
4806 if (IS_ERR(root))
4807 return ERR_CAST(root);
4808
4809 trans = btrfs_start_transaction(root, 1);
4810 BUG_ON(!trans);
4811
4812 err = btrfs_find_free_objectid(trans, root, objectid, &objectid);
4813 if (err)
4814 goto out;
4815
4816 err = __insert_orphan_inode(trans, root, objectid, group->key.offset);
4817 BUG_ON(err);
4818
4819 err = btrfs_insert_file_extent(trans, root, objectid, 0, 0, 0,
4820 group->key.offset, 0, group->key.offset,
4821 0, 0, 0);
4822 BUG_ON(err);
4823
4824 inode = btrfs_iget_locked(root->fs_info->sb, objectid, root);
4825 if (inode->i_state & I_NEW) {
4826 BTRFS_I(inode)->root = root;
4827 BTRFS_I(inode)->location.objectid = objectid;
4828 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
4829 BTRFS_I(inode)->location.offset = 0;
4830 btrfs_read_locked_inode(inode);
4831 unlock_new_inode(inode);
4832 BUG_ON(is_bad_inode(inode));
4833 } else {
4834 BUG_ON(1);
4835 }
4836
4837 err = btrfs_orphan_add(trans, inode);
4838 out:
4839 btrfs_end_transaction(trans, root);
4840 if (err) {
4841 if (inode)
4842 iput(inode);
4843 inode = ERR_PTR(err);
4844 }
4845 return inode;
4846 }
4847
4848 int btrfs_relocate_block_group(struct btrfs_root *root, u64 group_start)
4849 {
4850 struct btrfs_trans_handle *trans;
4851 struct btrfs_path *path;
4852 struct btrfs_fs_info *info = root->fs_info;
4853 struct extent_buffer *leaf;
4854 struct inode *reloc_inode;
4855 struct btrfs_block_group_cache *block_group;
4856 struct btrfs_key key;
4857 u64 skipped;
4858 u64 cur_byte;
4859 u64 total_found;
4860 u32 nritems;
4861 int ret;
4862 int progress;
4863 int pass = 0;
4864
4865 root = root->fs_info->extent_root;
4866
4867 block_group = btrfs_lookup_block_group(info, group_start);
4868 BUG_ON(!block_group);
4869
4870 printk("btrfs relocating block group %llu flags %llu\n",
4871 (unsigned long long)block_group->key.objectid,
4872 (unsigned long long)block_group->flags);
4873
4874 path = btrfs_alloc_path();
4875 BUG_ON(!path);
4876
4877 reloc_inode = create_reloc_inode(info, block_group);
4878 BUG_ON(IS_ERR(reloc_inode));
4879
4880 __alloc_chunk_for_shrink(root, block_group, 1);
4881 block_group->ro = 1;
4882 block_group->space_info->total_bytes -= block_group->key.offset;
4883
4884 btrfs_start_delalloc_inodes(info->tree_root);
4885 btrfs_wait_ordered_extents(info->tree_root, 0);
4886 again:
4887 skipped = 0;
4888 total_found = 0;
4889 progress = 0;
4890 key.objectid = block_group->key.objectid;
4891 key.offset = 0;
4892 key.type = 0;
4893 cur_byte = key.objectid;
4894
4895 trans = btrfs_start_transaction(info->tree_root, 1);
4896 btrfs_commit_transaction(trans, info->tree_root);
4897
4898 mutex_lock(&root->fs_info->cleaner_mutex);
4899 btrfs_clean_old_snapshots(info->tree_root);
4900 btrfs_remove_leaf_refs(info->tree_root, (u64)-1, 1);
4901 mutex_unlock(&root->fs_info->cleaner_mutex);
4902
4903 while(1) {
4904 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4905 if (ret < 0)
4906 goto out;
4907 next:
4908 leaf = path->nodes[0];
4909 nritems = btrfs_header_nritems(leaf);
4910 if (path->slots[0] >= nritems) {
4911 ret = btrfs_next_leaf(root, path);
4912 if (ret < 0)
4913 goto out;
4914 if (ret == 1) {
4915 ret = 0;
4916 break;
4917 }
4918 leaf = path->nodes[0];
4919 nritems = btrfs_header_nritems(leaf);
4920 }
4921
4922 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4923
4924 if (key.objectid >= block_group->key.objectid +
4925 block_group->key.offset)
4926 break;
4927
4928 if (progress && need_resched()) {
4929 btrfs_release_path(root, path);
4930 cond_resched();
4931 progress = 0;
4932 continue;
4933 }
4934 progress = 1;
4935
4936 if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY ||
4937 key.objectid + key.offset <= cur_byte) {
4938 path->slots[0]++;
4939 goto next;
4940 }
4941
4942 total_found++;
4943 cur_byte = key.objectid + key.offset;
4944 btrfs_release_path(root, path);
4945
4946 __alloc_chunk_for_shrink(root, block_group, 0);
4947 ret = relocate_one_extent(root, path, &key, block_group,
4948 reloc_inode, pass);
4949 BUG_ON(ret < 0);
4950 if (ret > 0)
4951 skipped++;
4952
4953 key.objectid = cur_byte;
4954 key.type = 0;
4955 key.offset = 0;
4956 }
4957
4958 btrfs_release_path(root, path);
4959
4960 if (pass == 0) {
4961 btrfs_wait_ordered_range(reloc_inode, 0, (u64)-1);
4962 invalidate_mapping_pages(reloc_inode->i_mapping, 0, -1);
4963 WARN_ON(reloc_inode->i_mapping->nrpages);
4964 }
4965
4966 if (total_found > 0) {
4967 printk("btrfs found %llu extents in pass %d\n",
4968 (unsigned long long)total_found, pass);
4969 pass++;
4970 if (total_found == skipped && pass > 2) {
4971 iput(reloc_inode);
4972 reloc_inode = create_reloc_inode(info, block_group);
4973 pass = 0;
4974 }
4975 goto again;
4976 }
4977
4978 /* delete reloc_inode */
4979 iput(reloc_inode);
4980
4981 /* unpin extents in this range */
4982 trans = btrfs_start_transaction(info->tree_root, 1);
4983 btrfs_commit_transaction(trans, info->tree_root);
4984
4985 spin_lock(&block_group->lock);
4986 WARN_ON(block_group->pinned > 0);
4987 WARN_ON(block_group->reserved > 0);
4988 WARN_ON(btrfs_block_group_used(&block_group->item) > 0);
4989 spin_unlock(&block_group->lock);
4990 ret = 0;
4991 out:
4992 btrfs_free_path(path);
4993 return ret;
4994 }
4995
4996 int find_first_block_group(struct btrfs_root *root, struct btrfs_path *path,
4997 struct btrfs_key *key)
4998 {
4999 int ret = 0;
5000 struct btrfs_key found_key;
5001 struct extent_buffer *leaf;
5002 int slot;
5003
5004 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
5005 if (ret < 0)
5006 goto out;
5007
5008 while(1) {
5009 slot = path->slots[0];
5010 leaf = path->nodes[0];
5011 if (slot >= btrfs_header_nritems(leaf)) {
5012 ret = btrfs_next_leaf(root, path);
5013 if (ret == 0)
5014 continue;
5015 if (ret < 0)
5016 goto out;
5017 break;
5018 }
5019 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5020
5021 if (found_key.objectid >= key->objectid &&
5022 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
5023 ret = 0;
5024 goto out;
5025 }
5026 path->slots[0]++;
5027 }
5028 ret = -ENOENT;
5029 out:
5030 return ret;
5031 }
5032
5033 int btrfs_free_block_groups(struct btrfs_fs_info *info)
5034 {
5035 struct btrfs_block_group_cache *block_group;
5036 struct rb_node *n;
5037
5038 spin_lock(&info->block_group_cache_lock);
5039 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
5040 block_group = rb_entry(n, struct btrfs_block_group_cache,
5041 cache_node);
5042 rb_erase(&block_group->cache_node,
5043 &info->block_group_cache_tree);
5044 spin_unlock(&info->block_group_cache_lock);
5045
5046 btrfs_remove_free_space_cache(block_group);
5047 down_write(&block_group->space_info->groups_sem);
5048 list_del(&block_group->list);
5049 up_write(&block_group->space_info->groups_sem);
5050 kfree(block_group);
5051
5052 spin_lock(&info->block_group_cache_lock);
5053 }
5054 spin_unlock(&info->block_group_cache_lock);
5055 return 0;
5056 }
5057
5058 int btrfs_read_block_groups(struct btrfs_root *root)
5059 {
5060 struct btrfs_path *path;
5061 int ret;
5062 struct btrfs_block_group_cache *cache;
5063 struct btrfs_fs_info *info = root->fs_info;
5064 struct btrfs_space_info *space_info;
5065 struct btrfs_key key;
5066 struct btrfs_key found_key;
5067 struct extent_buffer *leaf;
5068
5069 root = info->extent_root;
5070 key.objectid = 0;
5071 key.offset = 0;
5072 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
5073 path = btrfs_alloc_path();
5074 if (!path)
5075 return -ENOMEM;
5076
5077 while(1) {
5078 ret = find_first_block_group(root, path, &key);
5079 if (ret > 0) {
5080 ret = 0;
5081 goto error;
5082 }
5083 if (ret != 0)
5084 goto error;
5085
5086 leaf = path->nodes[0];
5087 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5088 cache = kzalloc(sizeof(*cache), GFP_NOFS);
5089 if (!cache) {
5090 ret = -ENOMEM;
5091 break;
5092 }
5093
5094 spin_lock_init(&cache->lock);
5095 mutex_init(&cache->alloc_mutex);
5096 INIT_LIST_HEAD(&cache->list);
5097 read_extent_buffer(leaf, &cache->item,
5098 btrfs_item_ptr_offset(leaf, path->slots[0]),
5099 sizeof(cache->item));
5100 memcpy(&cache->key, &found_key, sizeof(found_key));
5101
5102 key.objectid = found_key.objectid + found_key.offset;
5103 btrfs_release_path(root, path);
5104 cache->flags = btrfs_block_group_flags(&cache->item);
5105
5106 ret = update_space_info(info, cache->flags, found_key.offset,
5107 btrfs_block_group_used(&cache->item),
5108 &space_info);
5109 BUG_ON(ret);
5110 cache->space_info = space_info;
5111 down_write(&space_info->groups_sem);
5112 list_add_tail(&cache->list, &space_info->block_groups);
5113 up_write(&space_info->groups_sem);
5114
5115 ret = btrfs_add_block_group_cache(root->fs_info, cache);
5116 BUG_ON(ret);
5117
5118 set_avail_alloc_bits(root->fs_info, cache->flags);
5119 }
5120 ret = 0;
5121 error:
5122 btrfs_free_path(path);
5123 return ret;
5124 }
5125
5126 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
5127 struct btrfs_root *root, u64 bytes_used,
5128 u64 type, u64 chunk_objectid, u64 chunk_offset,
5129 u64 size)
5130 {
5131 int ret;
5132 struct btrfs_root *extent_root;
5133 struct btrfs_block_group_cache *cache;
5134
5135 extent_root = root->fs_info->extent_root;
5136
5137 root->fs_info->last_trans_new_blockgroup = trans->transid;
5138
5139 cache = kzalloc(sizeof(*cache), GFP_NOFS);
5140 if (!cache)
5141 return -ENOMEM;
5142
5143 cache->key.objectid = chunk_offset;
5144 cache->key.offset = size;
5145 spin_lock_init(&cache->lock);
5146 mutex_init(&cache->alloc_mutex);
5147 INIT_LIST_HEAD(&cache->list);
5148 btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY);
5149
5150 btrfs_set_block_group_used(&cache->item, bytes_used);
5151 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
5152 cache->flags = type;
5153 btrfs_set_block_group_flags(&cache->item, type);
5154
5155 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
5156 &cache->space_info);
5157 BUG_ON(ret);
5158 down_write(&cache->space_info->groups_sem);
5159 list_add_tail(&cache->list, &cache->space_info->block_groups);
5160 up_write(&cache->space_info->groups_sem);
5161
5162 ret = btrfs_add_block_group_cache(root->fs_info, cache);
5163 BUG_ON(ret);
5164
5165 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
5166 sizeof(cache->item));
5167 BUG_ON(ret);
5168
5169 finish_current_insert(trans, extent_root, 0);
5170 ret = del_pending_extents(trans, extent_root, 0);
5171 BUG_ON(ret);
5172 set_avail_alloc_bits(extent_root->fs_info, type);
5173
5174 return 0;
5175 }
5176
5177 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
5178 struct btrfs_root *root, u64 group_start)
5179 {
5180 struct btrfs_path *path;
5181 struct btrfs_block_group_cache *block_group;
5182 struct btrfs_key key;
5183 int ret;
5184
5185 root = root->fs_info->extent_root;
5186
5187 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
5188 BUG_ON(!block_group);
5189
5190 memcpy(&key, &block_group->key, sizeof(key));
5191
5192 path = btrfs_alloc_path();
5193 BUG_ON(!path);
5194
5195 btrfs_remove_free_space_cache(block_group);
5196 rb_erase(&block_group->cache_node,
5197 &root->fs_info->block_group_cache_tree);
5198 down_write(&block_group->space_info->groups_sem);
5199 list_del(&block_group->list);
5200 up_write(&block_group->space_info->groups_sem);
5201
5202 /*
5203 memset(shrink_block_group, 0, sizeof(*shrink_block_group));
5204 kfree(shrink_block_group);
5205 */
5206
5207 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
5208 if (ret > 0)
5209 ret = -EIO;
5210 if (ret < 0)
5211 goto out;
5212
5213 ret = btrfs_del_item(trans, root, path);
5214 out:
5215 btrfs_free_path(path);
5216 return ret;
5217 }
Ubuntu Artful kernel mirror