?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Btrfs: fix stop searching test in replace_one_extent
[linux-2.6/mini2440.git] / fs / btrfs / extent-tree.c
blob1d7f043152b039f626ea78da3f4d1c06c7690cfb
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 "compat.h"
23 #include "hash.h"
24 #include "crc32c.h"
25 #include "ctree.h"
26 #include "disk-io.h"
27 #include "print-tree.h"
28 #include "transaction.h"
29 #include "volumes.h"
30 #include "locking.h"
31 #include "ref-cache.h"
32
33 #define PENDING_EXTENT_INSERT 0
34 #define PENDING_EXTENT_DELETE 1
35 #define PENDING_BACKREF_UPDATE 2
36
37 struct pending_extent_op {
38 int type;
39 u64 bytenr;
40 u64 num_bytes;
41 u64 parent;
42 u64 orig_parent;
43 u64 generation;
44 u64 orig_generation;
45 int level;
46 struct list_head list;
47 int del;
48 };
49
50 static int finish_current_insert(struct btrfs_trans_handle *trans,
51 struct btrfs_root *extent_root, int all);
52 static int del_pending_extents(struct btrfs_trans_handle *trans,
53 struct btrfs_root *extent_root, int all);
54 static int pin_down_bytes(struct btrfs_trans_handle *trans,
55 struct btrfs_root *root,
56 u64 bytenr, u64 num_bytes, int is_data);
57 static int update_block_group(struct btrfs_trans_handle *trans,
58 struct btrfs_root *root,
59 u64 bytenr, u64 num_bytes, int alloc,
60 int mark_free);
61
62 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
63 {
64 return (cache->flags & bits) == bits;
65 }
66
67 /*
68 * this adds the block group to the fs_info rb tree for the block group
69 * cache
70 */
71 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
72 struct btrfs_block_group_cache *block_group)
73 {
74 struct rb_node **p;
75 struct rb_node *parent = NULL;
76 struct btrfs_block_group_cache *cache;
77
78 spin_lock(&info->block_group_cache_lock);
79 p = &info->block_group_cache_tree.rb_node;
80
81 while (*p) {
82 parent = *p;
83 cache = rb_entry(parent, struct btrfs_block_group_cache,
84 cache_node);
85 if (block_group->key.objectid < cache->key.objectid) {
86 p = &(*p)->rb_left;
87 } else if (block_group->key.objectid > cache->key.objectid) {
88 p = &(*p)->rb_right;
89 } else {
90 spin_unlock(&info->block_group_cache_lock);
91 return -EEXIST;
92 }
93 }
94
95 rb_link_node(&block_group->cache_node, parent, p);
96 rb_insert_color(&block_group->cache_node,
97 &info->block_group_cache_tree);
98 spin_unlock(&info->block_group_cache_lock);
99
100 return 0;
101 }
102
103 /*
104 * This will return the block group at or after bytenr if contains is 0, else
105 * it will return the block group that contains the bytenr
106 */
107 static struct btrfs_block_group_cache *
108 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
109 int contains)
110 {
111 struct btrfs_block_group_cache *cache, *ret = NULL;
112 struct rb_node *n;
113 u64 end, start;
114
115 spin_lock(&info->block_group_cache_lock);
116 n = info->block_group_cache_tree.rb_node;
117
118 while (n) {
119 cache = rb_entry(n, struct btrfs_block_group_cache,
120 cache_node);
121 end = cache->key.objectid + cache->key.offset - 1;
122 start = cache->key.objectid;
123
124 if (bytenr < start) {
125 if (!contains && (!ret || start < ret->key.objectid))
126 ret = cache;
127 n = n->rb_left;
128 } else if (bytenr > start) {
129 if (contains && bytenr <= end) {
130 ret = cache;
131 break;
132 }
133 n = n->rb_right;
134 } else {
135 ret = cache;
136 break;
137 }
138 }
139 if (ret)
140 atomic_inc(&ret->count);
141 spin_unlock(&info->block_group_cache_lock);
142
143 return ret;
144 }
145
146 /*
147 * this is only called by cache_block_group, since we could have freed extents
148 * we need to check the pinned_extents for any extents that can't be used yet
149 * since their free space will be released as soon as the transaction commits.
150 */
151 static int add_new_free_space(struct btrfs_block_group_cache *block_group,
152 struct btrfs_fs_info *info, u64 start, u64 end)
153 {
154 u64 extent_start, extent_end, size;
155 int ret;
156
157 mutex_lock(&info->pinned_mutex);
158 while (start < end) {
159 ret = find_first_extent_bit(&info->pinned_extents, start,
160 &extent_start, &extent_end,
161 EXTENT_DIRTY);
162 if (ret)
163 break;
164
165 if (extent_start == start) {
166 start = extent_end + 1;
167 } else if (extent_start > start && extent_start < end) {
168 size = extent_start - start;
169 ret = btrfs_add_free_space(block_group, start,
170 size);
171 BUG_ON(ret);
172 start = extent_end + 1;
173 } else {
174 break;
175 }
176 }
177
178 if (start < end) {
179 size = end - start;
180 ret = btrfs_add_free_space(block_group, start, size);
181 BUG_ON(ret);
182 }
183 mutex_unlock(&info->pinned_mutex);
184
185 return 0;
186 }
187
188 static int remove_sb_from_cache(struct btrfs_root *root,
189 struct btrfs_block_group_cache *cache)
190 {
191 u64 bytenr;
192 u64 *logical;
193 int stripe_len;
194 int i, nr, ret;
195
196 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
197 bytenr = btrfs_sb_offset(i);
198 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
199 cache->key.objectid, bytenr, 0,
200 &logical, &nr, &stripe_len);
201 BUG_ON(ret);
202 while (nr--) {
203 btrfs_remove_free_space(cache, logical[nr],
204 stripe_len);
205 }
206 kfree(logical);
207 }
208 return 0;
209 }
210
211 static int cache_block_group(struct btrfs_root *root,
212 struct btrfs_block_group_cache *block_group)
213 {
214 struct btrfs_path *path;
215 int ret = 0;
216 struct btrfs_key key;
217 struct extent_buffer *leaf;
218 int slot;
219 u64 last;
220
221 if (!block_group)
222 return 0;
223
224 root = root->fs_info->extent_root;
225
226 if (block_group->cached)
227 return 0;
228
229 path = btrfs_alloc_path();
230 if (!path)
231 return -ENOMEM;
232
233 path->reada = 2;
234 /*
235 * we get into deadlocks with paths held by callers of this function.
236 * since the alloc_mutex is protecting things right now, just
237 * skip the locking here
238 */
239 path->skip_locking = 1;
240 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
241 key.objectid = last;
242 key.offset = 0;
243 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
244 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
245 if (ret < 0)
246 goto err;
247
248 while (1) {
249 leaf = path->nodes[0];
250 slot = path->slots[0];
251 if (slot >= btrfs_header_nritems(leaf)) {
252 ret = btrfs_next_leaf(root, path);
253 if (ret < 0)
254 goto err;
255 if (ret == 0)
256 continue;
257 else
258 break;
259 }
260 btrfs_item_key_to_cpu(leaf, &key, slot);
261 if (key.objectid < block_group->key.objectid)
262 goto next;
263
264 if (key.objectid >= block_group->key.objectid +
265 block_group->key.offset)
266 break;
267
268 if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
269 add_new_free_space(block_group, root->fs_info, last,
270 key.objectid);
271
272 last = key.objectid + key.offset;
273 }
274 next:
275 path->slots[0]++;
276 }
277
278 add_new_free_space(block_group, root->fs_info, last,
279 block_group->key.objectid +
280 block_group->key.offset);
281
282 remove_sb_from_cache(root, block_group);
283 block_group->cached = 1;
284 ret = 0;
285 err:
286 btrfs_free_path(path);
287 return ret;
288 }
289
290 /*
291 * return the block group that starts at or after bytenr
292 */
293 static struct btrfs_block_group_cache *
294 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
295 {
296 struct btrfs_block_group_cache *cache;
297
298 cache = block_group_cache_tree_search(info, bytenr, 0);
299
300 return cache;
301 }
302
303 /*
304 * return the block group that contains teh given bytenr
305 */
306 struct btrfs_block_group_cache *btrfs_lookup_block_group(
307 struct btrfs_fs_info *info,
308 u64 bytenr)
309 {
310 struct btrfs_block_group_cache *cache;
311
312 cache = block_group_cache_tree_search(info, bytenr, 1);
313
314 return cache;
315 }
316
317 static inline void put_block_group(struct btrfs_block_group_cache *cache)
318 {
319 if (atomic_dec_and_test(&cache->count))
320 kfree(cache);
321 }
322
323 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
324 u64 flags)
325 {
326 struct list_head *head = &info->space_info;
327 struct btrfs_space_info *found;
328 list_for_each_entry(found, head, list) {
329 if (found->flags == flags)
330 return found;
331 }
332 return NULL;
333 }
334
335 static u64 div_factor(u64 num, int factor)
336 {
337 if (factor == 10)
338 return num;
339 num *= factor;
340 do_div(num, 10);
341 return num;
342 }
343
344 u64 btrfs_find_block_group(struct btrfs_root *root,
345 u64 search_start, u64 search_hint, int owner)
346 {
347 struct btrfs_block_group_cache *cache;
348 u64 used;
349 u64 last = max(search_hint, search_start);
350 u64 group_start = 0;
351 int full_search = 0;
352 int factor = 9;
353 int wrapped = 0;
354 again:
355 while (1) {
356 cache = btrfs_lookup_first_block_group(root->fs_info, last);
357 if (!cache)
358 break;
359
360 spin_lock(&cache->lock);
361 last = cache->key.objectid + cache->key.offset;
362 used = btrfs_block_group_used(&cache->item);
363
364 if ((full_search || !cache->ro) &&
365 block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
366 if (used + cache->pinned + cache->reserved <
367 div_factor(cache->key.offset, factor)) {
368 group_start = cache->key.objectid;
369 spin_unlock(&cache->lock);
370 put_block_group(cache);
371 goto found;
372 }
373 }
374 spin_unlock(&cache->lock);
375 put_block_group(cache);
376 cond_resched();
377 }
378 if (!wrapped) {
379 last = search_start;
380 wrapped = 1;
381 goto again;
382 }
383 if (!full_search && factor < 10) {
384 last = search_start;
385 full_search = 1;
386 factor = 10;
387 goto again;
388 }
389 found:
390 return group_start;
391 }
392
393 /* simple helper to search for an existing extent at a given offset */
394 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
395 {
396 int ret;
397 struct btrfs_key key;
398 struct btrfs_path *path;
399
400 path = btrfs_alloc_path();
401 BUG_ON(!path);
402 key.objectid = start;
403 key.offset = len;
404 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
405 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
406 0, 0);
407 btrfs_free_path(path);
408 return ret;
409 }
410
411 /*
412 * Back reference rules. Back refs have three main goals:
413 *
414 * 1) differentiate between all holders of references to an extent so that
415 * when a reference is dropped we can make sure it was a valid reference
416 * before freeing the extent.
417 *
418 * 2) Provide enough information to quickly find the holders of an extent
419 * if we notice a given block is corrupted or bad.
420 *
421 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
422 * maintenance. This is actually the same as #2, but with a slightly
423 * different use case.
424 *
425 * File extents can be referenced by:
426 *
427 * - multiple snapshots, subvolumes, or different generations in one subvol
428 * - different files inside a single subvolume
429 * - different offsets inside a file (bookend extents in file.c)
430 *
431 * The extent ref structure has fields for:
432 *
433 * - Objectid of the subvolume root
434 * - Generation number of the tree holding the reference
435 * - objectid of the file holding the reference
436 * - number of references holding by parent node (alway 1 for tree blocks)
437 *
438 * Btree leaf may hold multiple references to a file extent. In most cases,
439 * these references are from same file and the corresponding offsets inside
440 * the file are close together.
441 *
442 * When a file extent is allocated the fields are filled in:
443 * (root_key.objectid, trans->transid, inode objectid, 1)
444 *
445 * When a leaf is cow'd new references are added for every file extent found
446 * in the leaf. It looks similar to the create case, but trans->transid will
447 * be different when the block is cow'd.
448 *
449 * (root_key.objectid, trans->transid, inode objectid,
450 * number of references in the leaf)
451 *
452 * When a file extent is removed either during snapshot deletion or
453 * file truncation, we find the corresponding back reference and check
454 * the following fields:
455 *
456 * (btrfs_header_owner(leaf), btrfs_header_generation(leaf),
457 * inode objectid)
458 *
459 * Btree extents can be referenced by:
460 *
461 * - Different subvolumes
462 * - Different generations of the same subvolume
463 *
464 * When a tree block is created, back references are inserted:
465 *
466 * (root->root_key.objectid, trans->transid, level, 1)
467 *
468 * When a tree block is cow'd, new back references are added for all the
469 * blocks it points to. If the tree block isn't in reference counted root,
470 * the old back references are removed. These new back references are of
471 * the form (trans->transid will have increased since creation):
472 *
473 * (root->root_key.objectid, trans->transid, level, 1)
474 *
475 * When a backref is in deleting, the following fields are checked:
476 *
477 * if backref was for a tree root:
478 * (btrfs_header_owner(itself), btrfs_header_generation(itself), level)
479 * else
480 * (btrfs_header_owner(parent), btrfs_header_generation(parent), level)
481 *
482 * Back Reference Key composing:
483 *
484 * The key objectid corresponds to the first byte in the extent, the key
485 * type is set to BTRFS_EXTENT_REF_KEY, and the key offset is the first
486 * byte of parent extent. If a extent is tree root, the key offset is set
487 * to the key objectid.
488 */
489
490 static noinline int lookup_extent_backref(struct btrfs_trans_handle *trans,
491 struct btrfs_root *root,
492 struct btrfs_path *path,
493 u64 bytenr, u64 parent,
494 u64 ref_root, u64 ref_generation,
495 u64 owner_objectid, int del)
496 {
497 struct btrfs_key key;
498 struct btrfs_extent_ref *ref;
499 struct extent_buffer *leaf;
500 u64 ref_objectid;
501 int ret;
502
503 key.objectid = bytenr;
504 key.type = BTRFS_EXTENT_REF_KEY;
505 key.offset = parent;
506
507 ret = btrfs_search_slot(trans, root, &key, path, del ? -1 : 0, 1);
508 if (ret < 0)
509 goto out;
510 if (ret > 0) {
511 ret = -ENOENT;
512 goto out;
513 }
514
515 leaf = path->nodes[0];
516 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
517 ref_objectid = btrfs_ref_objectid(leaf, ref);
518 if (btrfs_ref_root(leaf, ref) != ref_root ||
519 btrfs_ref_generation(leaf, ref) != ref_generation ||
520 (ref_objectid != owner_objectid &&
521 ref_objectid != BTRFS_MULTIPLE_OBJECTIDS)) {
522 ret = -EIO;
523 WARN_ON(1);
524 goto out;
525 }
526 ret = 0;
527 out:
528 return ret;
529 }
530
531 /*
532 * updates all the backrefs that are pending on update_list for the
533 * extent_root
534 */
535 static noinline int update_backrefs(struct btrfs_trans_handle *trans,
536 struct btrfs_root *extent_root,
537 struct btrfs_path *path,
538 struct list_head *update_list)
539 {
540 struct btrfs_key key;
541 struct btrfs_extent_ref *ref;
542 struct btrfs_fs_info *info = extent_root->fs_info;
543 struct pending_extent_op *op;
544 struct extent_buffer *leaf;
545 int ret = 0;
546 struct list_head *cur = update_list->next;
547 u64 ref_objectid;
548 u64 ref_root = extent_root->root_key.objectid;
549
550 op = list_entry(cur, struct pending_extent_op, list);
551
552 search:
553 key.objectid = op->bytenr;
554 key.type = BTRFS_EXTENT_REF_KEY;
555 key.offset = op->orig_parent;
556
557 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 1);
558 BUG_ON(ret);
559
560 leaf = path->nodes[0];
561
562 loop:
563 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
564
565 ref_objectid = btrfs_ref_objectid(leaf, ref);
566
567 if (btrfs_ref_root(leaf, ref) != ref_root ||
568 btrfs_ref_generation(leaf, ref) != op->orig_generation ||
569 (ref_objectid != op->level &&
570 ref_objectid != BTRFS_MULTIPLE_OBJECTIDS)) {
571 printk(KERN_ERR "btrfs couldn't find %llu, parent %llu, "
572 "root %llu, owner %u\n",
573 (unsigned long long)op->bytenr,
574 (unsigned long long)op->orig_parent,
575 (unsigned long long)ref_root, op->level);
576 btrfs_print_leaf(extent_root, leaf);
577 BUG();
578 }
579
580 key.objectid = op->bytenr;
581 key.offset = op->parent;
582 key.type = BTRFS_EXTENT_REF_KEY;
583 ret = btrfs_set_item_key_safe(trans, extent_root, path, &key);
584 BUG_ON(ret);
585 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
586 btrfs_set_ref_generation(leaf, ref, op->generation);
587
588 cur = cur->next;
589
590 list_del_init(&op->list);
591 unlock_extent(&info->extent_ins, op->bytenr,
592 op->bytenr + op->num_bytes - 1, GFP_NOFS);
593 kfree(op);
594
595 if (cur == update_list) {
596 btrfs_mark_buffer_dirty(path->nodes[0]);
597 btrfs_release_path(extent_root, path);
598 goto out;
599 }
600
601 op = list_entry(cur, struct pending_extent_op, list);
602
603 path->slots[0]++;
604 while (path->slots[0] < btrfs_header_nritems(leaf)) {
605 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
606 if (key.objectid == op->bytenr &&
607 key.type == BTRFS_EXTENT_REF_KEY)
608 goto loop;
609 path->slots[0]++;
610 }
611
612 btrfs_mark_buffer_dirty(path->nodes[0]);
613 btrfs_release_path(extent_root, path);
614 goto search;
615
616 out:
617 return 0;
618 }
619
620 static noinline int insert_extents(struct btrfs_trans_handle *trans,
621 struct btrfs_root *extent_root,
622 struct btrfs_path *path,
623 struct list_head *insert_list, int nr)
624 {
625 struct btrfs_key *keys;
626 u32 *data_size;
627 struct pending_extent_op *op;
628 struct extent_buffer *leaf;
629 struct list_head *cur = insert_list->next;
630 struct btrfs_fs_info *info = extent_root->fs_info;
631 u64 ref_root = extent_root->root_key.objectid;
632 int i = 0, last = 0, ret;
633 int total = nr * 2;
634
635 if (!nr)
636 return 0;
637
638 keys = kzalloc(total * sizeof(struct btrfs_key), GFP_NOFS);
639 if (!keys)
640 return -ENOMEM;
641
642 data_size = kzalloc(total * sizeof(u32), GFP_NOFS);
643 if (!data_size) {
644 kfree(keys);
645 return -ENOMEM;
646 }
647
648 list_for_each_entry(op, insert_list, list) {
649 keys[i].objectid = op->bytenr;
650 keys[i].offset = op->num_bytes;
651 keys[i].type = BTRFS_EXTENT_ITEM_KEY;
652 data_size[i] = sizeof(struct btrfs_extent_item);
653 i++;
654
655 keys[i].objectid = op->bytenr;
656 keys[i].offset = op->parent;
657 keys[i].type = BTRFS_EXTENT_REF_KEY;
658 data_size[i] = sizeof(struct btrfs_extent_ref);
659 i++;
660 }
661
662 op = list_entry(cur, struct pending_extent_op, list);
663 i = 0;
664 while (i < total) {
665 int c;
666 ret = btrfs_insert_some_items(trans, extent_root, path,
667 keys+i, data_size+i, total-i);
668 BUG_ON(ret < 0);
669
670 if (last && ret > 1)
671 BUG();
672
673 leaf = path->nodes[0];
674 for (c = 0; c < ret; c++) {
675 int ref_first = keys[i].type == BTRFS_EXTENT_REF_KEY;
676
677 /*
678 * if the first item we inserted was a backref, then
679 * the EXTENT_ITEM will be the odd c's, else it will
680 * be the even c's
681 */
682 if ((ref_first && (c % 2)) ||
683 (!ref_first && !(c % 2))) {
684 struct btrfs_extent_item *itm;
685
686 itm = btrfs_item_ptr(leaf, path->slots[0] + c,
687 struct btrfs_extent_item);
688 btrfs_set_extent_refs(path->nodes[0], itm, 1);
689 op->del++;
690 } else {
691 struct btrfs_extent_ref *ref;
692
693 ref = btrfs_item_ptr(leaf, path->slots[0] + c,
694 struct btrfs_extent_ref);
695 btrfs_set_ref_root(leaf, ref, ref_root);
696 btrfs_set_ref_generation(leaf, ref,
697 op->generation);
698 btrfs_set_ref_objectid(leaf, ref, op->level);
699 btrfs_set_ref_num_refs(leaf, ref, 1);
700 op->del++;
701 }
702
703 /*
704 * using del to see when its ok to free up the
705 * pending_extent_op. In the case where we insert the
706 * last item on the list in order to help do batching
707 * we need to not free the extent op until we actually
708 * insert the extent_item
709 */
710 if (op->del == 2) {
711 unlock_extent(&info->extent_ins, op->bytenr,
712 op->bytenr + op->num_bytes - 1,
713 GFP_NOFS);
714 cur = cur->next;
715 list_del_init(&op->list);
716 kfree(op);
717 if (cur != insert_list)
718 op = list_entry(cur,
719 struct pending_extent_op,
720 list);
721 }
722 }
723 btrfs_mark_buffer_dirty(leaf);
724 btrfs_release_path(extent_root, path);
725
726 /*
727 * Ok backref's and items usually go right next to eachother,
728 * but if we could only insert 1 item that means that we
729 * inserted on the end of a leaf, and we have no idea what may
730 * be on the next leaf so we just play it safe. In order to
731 * try and help this case we insert the last thing on our
732 * insert list so hopefully it will end up being the last
733 * thing on the leaf and everything else will be before it,
734 * which will let us insert a whole bunch of items at the same
735 * time.
736 */
737 if (ret == 1 && !last && (i + ret < total)) {
738 /*
739 * last: where we will pick up the next time around
740 * i: our current key to insert, will be total - 1
741 * cur: the current op we are screwing with
742 * op: duh
743 */
744 last = i + ret;
745 i = total - 1;
746 cur = insert_list->prev;
747 op = list_entry(cur, struct pending_extent_op, list);
748 } else if (last) {
749 /*
750 * ok we successfully inserted the last item on the
751 * list, lets reset everything
752 *
753 * i: our current key to insert, so where we left off
754 * last time
755 * last: done with this
756 * cur: the op we are messing with
757 * op: duh
758 * total: since we inserted the last key, we need to
759 * decrement total so we dont overflow
760 */
761 i = last;
762 last = 0;
763 total--;
764 if (i < total) {
765 cur = insert_list->next;
766 op = list_entry(cur, struct pending_extent_op,
767 list);
768 }
769 } else {
770 i += ret;
771 }
772
773 cond_resched();
774 }
775 ret = 0;
776 kfree(keys);
777 kfree(data_size);
778 return ret;
779 }
780
781 static noinline int insert_extent_backref(struct btrfs_trans_handle *trans,
782 struct btrfs_root *root,
783 struct btrfs_path *path,
784 u64 bytenr, u64 parent,
785 u64 ref_root, u64 ref_generation,
786 u64 owner_objectid)
787 {
788 struct btrfs_key key;
789 struct extent_buffer *leaf;
790 struct btrfs_extent_ref *ref;
791 u32 num_refs;
792 int ret;
793
794 key.objectid = bytenr;
795 key.type = BTRFS_EXTENT_REF_KEY;
796 key.offset = parent;
797
798 ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*ref));
799 if (ret == 0) {
800 leaf = path->nodes[0];
801 ref = btrfs_item_ptr(leaf, path->slots[0],
802 struct btrfs_extent_ref);
803 btrfs_set_ref_root(leaf, ref, ref_root);
804 btrfs_set_ref_generation(leaf, ref, ref_generation);
805 btrfs_set_ref_objectid(leaf, ref, owner_objectid);
806 btrfs_set_ref_num_refs(leaf, ref, 1);
807 } else if (ret == -EEXIST) {
808 u64 existing_owner;
809 BUG_ON(owner_objectid < BTRFS_FIRST_FREE_OBJECTID);
810 leaf = path->nodes[0];
811 ref = btrfs_item_ptr(leaf, path->slots[0],
812 struct btrfs_extent_ref);
813 if (btrfs_ref_root(leaf, ref) != ref_root ||
814 btrfs_ref_generation(leaf, ref) != ref_generation) {
815 ret = -EIO;
816 WARN_ON(1);
817 goto out;
818 }
819
820 num_refs = btrfs_ref_num_refs(leaf, ref);
821 BUG_ON(num_refs == 0);
822 btrfs_set_ref_num_refs(leaf, ref, num_refs + 1);
823
824 existing_owner = btrfs_ref_objectid(leaf, ref);
825 if (existing_owner != owner_objectid &&
826 existing_owner != BTRFS_MULTIPLE_OBJECTIDS) {
827 btrfs_set_ref_objectid(leaf, ref,
828 BTRFS_MULTIPLE_OBJECTIDS);
829 }
830 ret = 0;
831 } else {
832 goto out;
833 }
834 btrfs_mark_buffer_dirty(path->nodes[0]);
835 out:
836 btrfs_release_path(root, path);
837 return ret;
838 }
839
840 static noinline int remove_extent_backref(struct btrfs_trans_handle *trans,
841 struct btrfs_root *root,
842 struct btrfs_path *path)
843 {
844 struct extent_buffer *leaf;
845 struct btrfs_extent_ref *ref;
846 u32 num_refs;
847 int ret = 0;
848
849 leaf = path->nodes[0];
850 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
851 num_refs = btrfs_ref_num_refs(leaf, ref);
852 BUG_ON(num_refs == 0);
853 num_refs -= 1;
854 if (num_refs == 0) {
855 ret = btrfs_del_item(trans, root, path);
856 } else {
857 btrfs_set_ref_num_refs(leaf, ref, num_refs);
858 btrfs_mark_buffer_dirty(leaf);
859 }
860 btrfs_release_path(root, path);
861 return ret;
862 }
863
864 #ifdef BIO_RW_DISCARD
865 static void btrfs_issue_discard(struct block_device *bdev,
866 u64 start, u64 len)
867 {
868 blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL);
869 }
870 #endif
871
872 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
873 u64 num_bytes)
874 {
875 #ifdef BIO_RW_DISCARD
876 int ret;
877 u64 map_length = num_bytes;
878 struct btrfs_multi_bio *multi = NULL;
879
880 /* Tell the block device(s) that the sectors can be discarded */
881 ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
882 bytenr, &map_length, &multi, 0);
883 if (!ret) {
884 struct btrfs_bio_stripe *stripe = multi->stripes;
885 int i;
886
887 if (map_length > num_bytes)
888 map_length = num_bytes;
889
890 for (i = 0; i < multi->num_stripes; i++, stripe++) {
891 btrfs_issue_discard(stripe->dev->bdev,
892 stripe->physical,
893 map_length);
894 }
895 kfree(multi);
896 }
897
898 return ret;
899 #else
900 return 0;
901 #endif
902 }
903
904 static noinline int free_extents(struct btrfs_trans_handle *trans,
905 struct btrfs_root *extent_root,
906 struct list_head *del_list)
907 {
908 struct btrfs_fs_info *info = extent_root->fs_info;
909 struct btrfs_path *path;
910 struct btrfs_key key, found_key;
911 struct extent_buffer *leaf;
912 struct list_head *cur;
913 struct pending_extent_op *op;
914 struct btrfs_extent_item *ei;
915 int ret, num_to_del, extent_slot = 0, found_extent = 0;
916 u32 refs;
917 u64 bytes_freed = 0;
918
919 path = btrfs_alloc_path();
920 if (!path)
921 return -ENOMEM;
922 path->reada = 1;
923
924 search:
925 /* search for the backref for the current ref we want to delete */
926 cur = del_list->next;
927 op = list_entry(cur, struct pending_extent_op, list);
928 ret = lookup_extent_backref(trans, extent_root, path, op->bytenr,
929 op->orig_parent,
930 extent_root->root_key.objectid,
931 op->orig_generation, op->level, 1);
932 if (ret) {
933 printk(KERN_ERR "btrfs unable to find backref byte nr %llu "
934 "root %llu gen %llu owner %u\n",
935 (unsigned long long)op->bytenr,
936 (unsigned long long)extent_root->root_key.objectid,
937 (unsigned long long)op->orig_generation, op->level);
938 btrfs_print_leaf(extent_root, path->nodes[0]);
939 WARN_ON(1);
940 goto out;
941 }
942
943 extent_slot = path->slots[0];
944 num_to_del = 1;
945 found_extent = 0;
946
947 /*
948 * if we aren't the first item on the leaf we can move back one and see
949 * if our ref is right next to our extent item
950 */
951 if (likely(extent_slot)) {
952 extent_slot--;
953 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
954 extent_slot);
955 if (found_key.objectid == op->bytenr &&
956 found_key.type == BTRFS_EXTENT_ITEM_KEY &&
957 found_key.offset == op->num_bytes) {
958 num_to_del++;
959 found_extent = 1;
960 }
961 }
962
963 /*
964 * if we didn't find the extent we need to delete the backref and then
965 * search for the extent item key so we can update its ref count
966 */
967 if (!found_extent) {
968 key.objectid = op->bytenr;
969 key.type = BTRFS_EXTENT_ITEM_KEY;
970 key.offset = op->num_bytes;
971
972 ret = remove_extent_backref(trans, extent_root, path);
973 BUG_ON(ret);
974 btrfs_release_path(extent_root, path);
975 ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1);
976 BUG_ON(ret);
977 extent_slot = path->slots[0];
978 }
979
980 /* this is where we update the ref count for the extent */
981 leaf = path->nodes[0];
982 ei = btrfs_item_ptr(leaf, extent_slot, struct btrfs_extent_item);
983 refs = btrfs_extent_refs(leaf, ei);
984 BUG_ON(refs == 0);
985 refs--;
986 btrfs_set_extent_refs(leaf, ei, refs);
987
988 btrfs_mark_buffer_dirty(leaf);
989
990 /*
991 * This extent needs deleting. The reason cur_slot is extent_slot +
992 * num_to_del is because extent_slot points to the slot where the extent
993 * is, and if the backref was not right next to the extent we will be
994 * deleting at least 1 item, and will want to start searching at the
995 * slot directly next to extent_slot. However if we did find the
996 * backref next to the extent item them we will be deleting at least 2
997 * items and will want to start searching directly after the ref slot
998 */
999 if (!refs) {
1000 struct list_head *pos, *n, *end;
1001 int cur_slot = extent_slot+num_to_del;
1002 u64 super_used;
1003 u64 root_used;
1004
1005 path->slots[0] = extent_slot;
1006 bytes_freed = op->num_bytes;
1007
1008 mutex_lock(&info->pinned_mutex);
1009 ret = pin_down_bytes(trans, extent_root, op->bytenr,
1010 op->num_bytes, op->level >=
1011 BTRFS_FIRST_FREE_OBJECTID);
1012 mutex_unlock(&info->pinned_mutex);
1013 BUG_ON(ret < 0);
1014 op->del = ret;
1015
1016 /*
1017 * we need to see if we can delete multiple things at once, so
1018 * start looping through the list of extents we are wanting to
1019 * delete and see if their extent/backref's are right next to
1020 * eachother and the extents only have 1 ref
1021 */
1022 for (pos = cur->next; pos != del_list; pos = pos->next) {
1023 struct pending_extent_op *tmp;
1024
1025 tmp = list_entry(pos, struct pending_extent_op, list);
1026
1027 /* we only want to delete extent+ref at this stage */
1028 if (cur_slot >= btrfs_header_nritems(leaf) - 1)
1029 break;
1030
1031 btrfs_item_key_to_cpu(leaf, &found_key, cur_slot);
1032 if (found_key.objectid != tmp->bytenr ||
1033 found_key.type != BTRFS_EXTENT_ITEM_KEY ||
1034 found_key.offset != tmp->num_bytes)
1035 break;
1036
1037 /* check to make sure this extent only has one ref */
1038 ei = btrfs_item_ptr(leaf, cur_slot,
1039 struct btrfs_extent_item);
1040 if (btrfs_extent_refs(leaf, ei) != 1)
1041 break;
1042
1043 btrfs_item_key_to_cpu(leaf, &found_key, cur_slot+1);
1044 if (found_key.objectid != tmp->bytenr ||
1045 found_key.type != BTRFS_EXTENT_REF_KEY ||
1046 found_key.offset != tmp->orig_parent)
1047 break;
1048
1049 /*
1050 * the ref is right next to the extent, we can set the
1051 * ref count to 0 since we will delete them both now
1052 */
1053 btrfs_set_extent_refs(leaf, ei, 0);
1054
1055 /* pin down the bytes for this extent */
1056 mutex_lock(&info->pinned_mutex);
1057 ret = pin_down_bytes(trans, extent_root, tmp->bytenr,
1058 tmp->num_bytes, tmp->level >=
1059 BTRFS_FIRST_FREE_OBJECTID);
1060 mutex_unlock(&info->pinned_mutex);
1061 BUG_ON(ret < 0);
1062
1063 /*
1064 * use the del field to tell if we need to go ahead and
1065 * free up the extent when we delete the item or not.
1066 */
1067 tmp->del = ret;
1068 bytes_freed += tmp->num_bytes;
1069
1070 num_to_del += 2;
1071 cur_slot += 2;
1072 }
1073 end = pos;
1074
1075 /* update the free space counters */
1076 spin_lock(&info->delalloc_lock);
1077 super_used = btrfs_super_bytes_used(&info->super_copy);
1078 btrfs_set_super_bytes_used(&info->super_copy,
1079 super_used - bytes_freed);
1080
1081 root_used = btrfs_root_used(&extent_root->root_item);
1082 btrfs_set_root_used(&extent_root->root_item,
1083 root_used - bytes_freed);
1084 spin_unlock(&info->delalloc_lock);
1085
1086 /* delete the items */
1087 ret = btrfs_del_items(trans, extent_root, path,
1088 path->slots[0], num_to_del);
1089 BUG_ON(ret);
1090
1091 /*
1092 * loop through the extents we deleted and do the cleanup work
1093 * on them
1094 */
1095 for (pos = cur, n = pos->next; pos != end;
1096 pos = n, n = pos->next) {
1097 struct pending_extent_op *tmp;
1098 tmp = list_entry(pos, struct pending_extent_op, list);
1099
1100 /*
1101 * remember tmp->del tells us wether or not we pinned
1102 * down the extent
1103 */
1104 ret = update_block_group(trans, extent_root,
1105 tmp->bytenr, tmp->num_bytes, 0,
1106 tmp->del);
1107 BUG_ON(ret);
1108
1109 list_del_init(&tmp->list);
1110 unlock_extent(&info->extent_ins, tmp->bytenr,
1111 tmp->bytenr + tmp->num_bytes - 1,
1112 GFP_NOFS);
1113 kfree(tmp);
1114 }
1115 } else if (refs && found_extent) {
1116 /*
1117 * the ref and extent were right next to eachother, but the
1118 * extent still has a ref, so just free the backref and keep
1119 * going
1120 */
1121 ret = remove_extent_backref(trans, extent_root, path);
1122 BUG_ON(ret);
1123
1124 list_del_init(&op->list);
1125 unlock_extent(&info->extent_ins, op->bytenr,
1126 op->bytenr + op->num_bytes - 1, GFP_NOFS);
1127 kfree(op);
1128 } else {
1129 /*
1130 * the extent has multiple refs and the backref we were looking
1131 * for was not right next to it, so just unlock and go next,
1132 * we're good to go
1133 */
1134 list_del_init(&op->list);
1135 unlock_extent(&info->extent_ins, op->bytenr,
1136 op->bytenr + op->num_bytes - 1, GFP_NOFS);
1137 kfree(op);
1138 }
1139
1140 btrfs_release_path(extent_root, path);
1141 if (!list_empty(del_list))
1142 goto search;
1143
1144 out:
1145 btrfs_free_path(path);
1146 return ret;
1147 }
1148
1149 static int __btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
1150 struct btrfs_root *root, u64 bytenr,
1151 u64 orig_parent, u64 parent,
1152 u64 orig_root, u64 ref_root,
1153 u64 orig_generation, u64 ref_generation,
1154 u64 owner_objectid)
1155 {
1156 int ret;
1157 struct btrfs_root *extent_root = root->fs_info->extent_root;
1158 struct btrfs_path *path;
1159
1160 if (root == root->fs_info->extent_root) {
1161 struct pending_extent_op *extent_op;
1162 u64 num_bytes;
1163
1164 BUG_ON(owner_objectid >= BTRFS_MAX_LEVEL);
1165 num_bytes = btrfs_level_size(root, (int)owner_objectid);
1166 mutex_lock(&root->fs_info->extent_ins_mutex);
1167 if (test_range_bit(&root->fs_info->extent_ins, bytenr,
1168 bytenr + num_bytes - 1, EXTENT_WRITEBACK, 0)) {
1169 u64 priv;
1170 ret = get_state_private(&root->fs_info->extent_ins,
1171 bytenr, &priv);
1172 BUG_ON(ret);
1173 extent_op = (struct pending_extent_op *)
1174 (unsigned long)priv;
1175 BUG_ON(extent_op->parent != orig_parent);
1176 BUG_ON(extent_op->generation != orig_generation);
1177
1178 extent_op->parent = parent;
1179 extent_op->generation = ref_generation;
1180 } else {
1181 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
1182 BUG_ON(!extent_op);
1183
1184 extent_op->type = PENDING_BACKREF_UPDATE;
1185 extent_op->bytenr = bytenr;
1186 extent_op->num_bytes = num_bytes;
1187 extent_op->parent = parent;
1188 extent_op->orig_parent = orig_parent;
1189 extent_op->generation = ref_generation;
1190 extent_op->orig_generation = orig_generation;
1191 extent_op->level = (int)owner_objectid;
1192 INIT_LIST_HEAD(&extent_op->list);
1193 extent_op->del = 0;
1194
1195 set_extent_bits(&root->fs_info->extent_ins,
1196 bytenr, bytenr + num_bytes - 1,
1197 EXTENT_WRITEBACK, GFP_NOFS);
1198 set_state_private(&root->fs_info->extent_ins,
1199 bytenr, (unsigned long)extent_op);
1200 }
1201 mutex_unlock(&root->fs_info->extent_ins_mutex);
1202 return 0;
1203 }
1204
1205 path = btrfs_alloc_path();
1206 if (!path)
1207 return -ENOMEM;
1208 ret = lookup_extent_backref(trans, extent_root, path,
1209 bytenr, orig_parent, orig_root,
1210 orig_generation, owner_objectid, 1);
1211 if (ret)
1212 goto out;
1213 ret = remove_extent_backref(trans, extent_root, path);
1214 if (ret)
1215 goto out;
1216 ret = insert_extent_backref(trans, extent_root, path, bytenr,
1217 parent, ref_root, ref_generation,
1218 owner_objectid);
1219 BUG_ON(ret);
1220 finish_current_insert(trans, extent_root, 0);
1221 del_pending_extents(trans, extent_root, 0);
1222 out:
1223 btrfs_free_path(path);
1224 return ret;
1225 }
1226
1227 int btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
1228 struct btrfs_root *root, u64 bytenr,
1229 u64 orig_parent, u64 parent,
1230 u64 ref_root, u64 ref_generation,
1231 u64 owner_objectid)
1232 {
1233 int ret;
1234 if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
1235 owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
1236 return 0;
1237 ret = __btrfs_update_extent_ref(trans, root, bytenr, orig_parent,
1238 parent, ref_root, ref_root,
1239 ref_generation, ref_generation,
1240 owner_objectid);
1241 return ret;
1242 }
1243
1244 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1245 struct btrfs_root *root, u64 bytenr,
1246 u64 orig_parent, u64 parent,
1247 u64 orig_root, u64 ref_root,
1248 u64 orig_generation, u64 ref_generation,
1249 u64 owner_objectid)
1250 {
1251 struct btrfs_path *path;
1252 int ret;
1253 struct btrfs_key key;
1254 struct extent_buffer *l;
1255 struct btrfs_extent_item *item;
1256 u32 refs;
1257
1258 path = btrfs_alloc_path();
1259 if (!path)
1260 return -ENOMEM;
1261
1262 path->reada = 1;
1263 key.objectid = bytenr;
1264 key.type = BTRFS_EXTENT_ITEM_KEY;
1265 key.offset = (u64)-1;
1266
1267 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
1268 0, 1);
1269 if (ret < 0)
1270 return ret;
1271 BUG_ON(ret == 0 || path->slots[0] == 0);
1272
1273 path->slots[0]--;
1274 l = path->nodes[0];
1275
1276 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
1277 if (key.objectid != bytenr) {
1278 btrfs_print_leaf(root->fs_info->extent_root, path->nodes[0]);
1279 printk(KERN_ERR "btrfs wanted %llu found %llu\n",
1280 (unsigned long long)bytenr,
1281 (unsigned long long)key.objectid);
1282 BUG();
1283 }
1284 BUG_ON(key.type != BTRFS_EXTENT_ITEM_KEY);
1285
1286 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
1287 refs = btrfs_extent_refs(l, item);
1288 btrfs_set_extent_refs(l, item, refs + 1);
1289 btrfs_mark_buffer_dirty(path->nodes[0]);
1290
1291 btrfs_release_path(root->fs_info->extent_root, path);
1292
1293 path->reada = 1;
1294 ret = insert_extent_backref(trans, root->fs_info->extent_root,
1295 path, bytenr, parent,
1296 ref_root, ref_generation,
1297 owner_objectid);
1298 BUG_ON(ret);
1299 finish_current_insert(trans, root->fs_info->extent_root, 0);
1300 del_pending_extents(trans, root->fs_info->extent_root, 0);
1301
1302 btrfs_free_path(path);
1303 return 0;
1304 }
1305
1306 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1307 struct btrfs_root *root,
1308 u64 bytenr, u64 num_bytes, u64 parent,
1309 u64 ref_root, u64 ref_generation,
1310 u64 owner_objectid)
1311 {
1312 int ret;
1313 if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
1314 owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
1315 return 0;
1316 ret = __btrfs_inc_extent_ref(trans, root, bytenr, 0, parent,
1317 0, ref_root, 0, ref_generation,
1318 owner_objectid);
1319 return ret;
1320 }
1321
1322 int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
1323 struct btrfs_root *root)
1324 {
1325 finish_current_insert(trans, root->fs_info->extent_root, 1);
1326 del_pending_extents(trans, root->fs_info->extent_root, 1);
1327 return 0;
1328 }
1329
1330 int btrfs_lookup_extent_ref(struct btrfs_trans_handle *trans,
1331 struct btrfs_root *root, u64 bytenr,
1332 u64 num_bytes, u32 *refs)
1333 {
1334 struct btrfs_path *path;
1335 int ret;
1336 struct btrfs_key key;
1337 struct extent_buffer *l;
1338 struct btrfs_extent_item *item;
1339
1340 WARN_ON(num_bytes < root->sectorsize);
1341 path = btrfs_alloc_path();
1342 path->reada = 1;
1343 key.objectid = bytenr;
1344 key.offset = num_bytes;
1345 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
1346 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
1347 0, 0);
1348 if (ret < 0)
1349 goto out;
1350 if (ret != 0) {
1351 btrfs_print_leaf(root, path->nodes[0]);
1352 printk(KERN_INFO "btrfs failed to find block number %llu\n",
1353 (unsigned long long)bytenr);
1354 BUG();
1355 }
1356 l = path->nodes[0];
1357 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
1358 *refs = btrfs_extent_refs(l, item);
1359 out:
1360 btrfs_free_path(path);
1361 return 0;
1362 }
1363
1364 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
1365 struct btrfs_root *root, u64 objectid, u64 bytenr)
1366 {
1367 struct btrfs_root *extent_root = root->fs_info->extent_root;
1368 struct btrfs_path *path;
1369 struct extent_buffer *leaf;
1370 struct btrfs_extent_ref *ref_item;
1371 struct btrfs_key key;
1372 struct btrfs_key found_key;
1373 u64 ref_root;
1374 u64 last_snapshot;
1375 u32 nritems;
1376 int ret;
1377
1378 key.objectid = bytenr;
1379 key.offset = (u64)-1;
1380 key.type = BTRFS_EXTENT_ITEM_KEY;
1381
1382 path = btrfs_alloc_path();
1383 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
1384 if (ret < 0)
1385 goto out;
1386 BUG_ON(ret == 0);
1387
1388 ret = -ENOENT;
1389 if (path->slots[0] == 0)
1390 goto out;
1391
1392 path->slots[0]--;
1393 leaf = path->nodes[0];
1394 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1395
1396 if (found_key.objectid != bytenr ||
1397 found_key.type != BTRFS_EXTENT_ITEM_KEY)
1398 goto out;
1399
1400 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
1401 while (1) {
1402 leaf = path->nodes[0];
1403 nritems = btrfs_header_nritems(leaf);
1404 if (path->slots[0] >= nritems) {
1405 ret = btrfs_next_leaf(extent_root, path);
1406 if (ret < 0)
1407 goto out;
1408 if (ret == 0)
1409 continue;
1410 break;
1411 }
1412 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1413 if (found_key.objectid != bytenr)
1414 break;
1415
1416 if (found_key.type != BTRFS_EXTENT_REF_KEY) {
1417 path->slots[0]++;
1418 continue;
1419 }
1420
1421 ref_item = btrfs_item_ptr(leaf, path->slots[0],
1422 struct btrfs_extent_ref);
1423 ref_root = btrfs_ref_root(leaf, ref_item);
1424 if ((ref_root != root->root_key.objectid &&
1425 ref_root != BTRFS_TREE_LOG_OBJECTID) ||
1426 objectid != btrfs_ref_objectid(leaf, ref_item)) {
1427 ret = 1;
1428 goto out;
1429 }
1430 if (btrfs_ref_generation(leaf, ref_item) <= last_snapshot) {
1431 ret = 1;
1432 goto out;
1433 }
1434
1435 path->slots[0]++;
1436 }
1437 ret = 0;
1438 out:
1439 btrfs_free_path(path);
1440 return ret;
1441 }
1442
1443 int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1444 struct extent_buffer *buf, u32 nr_extents)
1445 {
1446 struct btrfs_key key;
1447 struct btrfs_file_extent_item *fi;
1448 u64 root_gen;
1449 u32 nritems;
1450 int i;
1451 int level;
1452 int ret = 0;
1453 int shared = 0;
1454
1455 if (!root->ref_cows)
1456 return 0;
1457
1458 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
1459 shared = 0;
1460 root_gen = root->root_key.offset;
1461 } else {
1462 shared = 1;
1463 root_gen = trans->transid - 1;
1464 }
1465
1466 level = btrfs_header_level(buf);
1467 nritems = btrfs_header_nritems(buf);
1468
1469 if (level == 0) {
1470 struct btrfs_leaf_ref *ref;
1471 struct btrfs_extent_info *info;
1472
1473 ref = btrfs_alloc_leaf_ref(root, nr_extents);
1474 if (!ref) {
1475 ret = -ENOMEM;
1476 goto out;
1477 }
1478
1479 ref->root_gen = root_gen;
1480 ref->bytenr = buf->start;
1481 ref->owner = btrfs_header_owner(buf);
1482 ref->generation = btrfs_header_generation(buf);
1483 ref->nritems = nr_extents;
1484 info = ref->extents;
1485
1486 for (i = 0; nr_extents > 0 && i < nritems; i++) {
1487 u64 disk_bytenr;
1488 btrfs_item_key_to_cpu(buf, &key, i);
1489 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1490 continue;
1491 fi = btrfs_item_ptr(buf, i,
1492 struct btrfs_file_extent_item);
1493 if (btrfs_file_extent_type(buf, fi) ==
1494 BTRFS_FILE_EXTENT_INLINE)
1495 continue;
1496 disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1497 if (disk_bytenr == 0)
1498 continue;
1499
1500 info->bytenr = disk_bytenr;
1501 info->num_bytes =
1502 btrfs_file_extent_disk_num_bytes(buf, fi);
1503 info->objectid = key.objectid;
1504 info->offset = key.offset;
1505 info++;
1506 }
1507
1508 ret = btrfs_add_leaf_ref(root, ref, shared);
1509 if (ret == -EEXIST && shared) {
1510 struct btrfs_leaf_ref *old;
1511 old = btrfs_lookup_leaf_ref(root, ref->bytenr);
1512 BUG_ON(!old);
1513 btrfs_remove_leaf_ref(root, old);
1514 btrfs_free_leaf_ref(root, old);
1515 ret = btrfs_add_leaf_ref(root, ref, shared);
1516 }
1517 WARN_ON(ret);
1518 btrfs_free_leaf_ref(root, ref);
1519 }
1520 out:
1521 return ret;
1522 }
1523
1524 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1525 struct extent_buffer *orig_buf, struct extent_buffer *buf,
1526 u32 *nr_extents)
1527 {
1528 u64 bytenr;
1529 u64 ref_root;
1530 u64 orig_root;
1531 u64 ref_generation;
1532 u64 orig_generation;
1533 u32 nritems;
1534 u32 nr_file_extents = 0;
1535 struct btrfs_key key;
1536 struct btrfs_file_extent_item *fi;
1537 int i;
1538 int level;
1539 int ret = 0;
1540 int faili = 0;
1541 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
1542 u64, u64, u64, u64, u64, u64, u64, u64);
1543
1544 ref_root = btrfs_header_owner(buf);
1545 ref_generation = btrfs_header_generation(buf);
1546 orig_root = btrfs_header_owner(orig_buf);
1547 orig_generation = btrfs_header_generation(orig_buf);
1548
1549 nritems = btrfs_header_nritems(buf);
1550 level = btrfs_header_level(buf);
1551
1552 if (root->ref_cows) {
1553 process_func = __btrfs_inc_extent_ref;
1554 } else {
1555 if (level == 0 &&
1556 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
1557 goto out;
1558 if (level != 0 &&
1559 root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
1560 goto out;
1561 process_func = __btrfs_update_extent_ref;
1562 }
1563
1564 for (i = 0; i < nritems; i++) {
1565 cond_resched();
1566 if (level == 0) {
1567 btrfs_item_key_to_cpu(buf, &key, i);
1568 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1569 continue;
1570 fi = btrfs_item_ptr(buf, i,
1571 struct btrfs_file_extent_item);
1572 if (btrfs_file_extent_type(buf, fi) ==
1573 BTRFS_FILE_EXTENT_INLINE)
1574 continue;
1575 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1576 if (bytenr == 0)
1577 continue;
1578
1579 nr_file_extents++;
1580
1581 ret = process_func(trans, root, bytenr,
1582 orig_buf->start, buf->start,
1583 orig_root, ref_root,
1584 orig_generation, ref_generation,
1585 key.objectid);
1586
1587 if (ret) {
1588 faili = i;
1589 WARN_ON(1);
1590 goto fail;
1591 }
1592 } else {
1593 bytenr = btrfs_node_blockptr(buf, i);
1594 ret = process_func(trans, root, bytenr,
1595 orig_buf->start, buf->start,
1596 orig_root, ref_root,
1597 orig_generation, ref_generation,
1598 level - 1);
1599 if (ret) {
1600 faili = i;
1601 WARN_ON(1);
1602 goto fail;
1603 }
1604 }
1605 }
1606 out:
1607 if (nr_extents) {
1608 if (level == 0)
1609 *nr_extents = nr_file_extents;
1610 else
1611 *nr_extents = nritems;
1612 }
1613 return 0;
1614 fail:
1615 WARN_ON(1);
1616 return ret;
1617 }
1618
1619 int btrfs_update_ref(struct btrfs_trans_handle *trans,
1620 struct btrfs_root *root, struct extent_buffer *orig_buf,
1621 struct extent_buffer *buf, int start_slot, int nr)
1622
1623 {
1624 u64 bytenr;
1625 u64 ref_root;
1626 u64 orig_root;
1627 u64 ref_generation;
1628 u64 orig_generation;
1629 struct btrfs_key key;
1630 struct btrfs_file_extent_item *fi;
1631 int i;
1632 int ret;
1633 int slot;
1634 int level;
1635
1636 BUG_ON(start_slot < 0);
1637 BUG_ON(start_slot + nr > btrfs_header_nritems(buf));
1638
1639 ref_root = btrfs_header_owner(buf);
1640 ref_generation = btrfs_header_generation(buf);
1641 orig_root = btrfs_header_owner(orig_buf);
1642 orig_generation = btrfs_header_generation(orig_buf);
1643 level = btrfs_header_level(buf);
1644
1645 if (!root->ref_cows) {
1646 if (level == 0 &&
1647 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
1648 return 0;
1649 if (level != 0 &&
1650 root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
1651 return 0;
1652 }
1653
1654 for (i = 0, slot = start_slot; i < nr; i++, slot++) {
1655 cond_resched();
1656 if (level == 0) {
1657 btrfs_item_key_to_cpu(buf, &key, slot);
1658 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1659 continue;
1660 fi = btrfs_item_ptr(buf, slot,
1661 struct btrfs_file_extent_item);
1662 if (btrfs_file_extent_type(buf, fi) ==
1663 BTRFS_FILE_EXTENT_INLINE)
1664 continue;
1665 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1666 if (bytenr == 0)
1667 continue;
1668 ret = __btrfs_update_extent_ref(trans, root, bytenr,
1669 orig_buf->start, buf->start,
1670 orig_root, ref_root,
1671 orig_generation, ref_generation,
1672 key.objectid);
1673 if (ret)
1674 goto fail;
1675 } else {
1676 bytenr = btrfs_node_blockptr(buf, slot);
1677 ret = __btrfs_update_extent_ref(trans, root, bytenr,
1678 orig_buf->start, buf->start,
1679 orig_root, ref_root,
1680 orig_generation, ref_generation,
1681 level - 1);
1682 if (ret)
1683 goto fail;
1684 }
1685 }
1686 return 0;
1687 fail:
1688 WARN_ON(1);
1689 return -1;
1690 }
1691
1692 static int write_one_cache_group(struct btrfs_trans_handle *trans,
1693 struct btrfs_root *root,
1694 struct btrfs_path *path,
1695 struct btrfs_block_group_cache *cache)
1696 {
1697 int ret;
1698 int pending_ret;
1699 struct btrfs_root *extent_root = root->fs_info->extent_root;
1700 unsigned long bi;
1701 struct extent_buffer *leaf;
1702
1703 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
1704 if (ret < 0)
1705 goto fail;
1706 BUG_ON(ret);
1707
1708 leaf = path->nodes[0];
1709 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
1710 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
1711 btrfs_mark_buffer_dirty(leaf);
1712 btrfs_release_path(extent_root, path);
1713 fail:
1714 finish_current_insert(trans, extent_root, 0);
1715 pending_ret = del_pending_extents(trans, extent_root, 0);
1716 if (ret)
1717 return ret;
1718 if (pending_ret)
1719 return pending_ret;
1720 return 0;
1721
1722 }
1723
1724 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
1725 struct btrfs_root *root)
1726 {
1727 struct btrfs_block_group_cache *cache, *entry;
1728 struct rb_node *n;
1729 int err = 0;
1730 int werr = 0;
1731 struct btrfs_path *path;
1732 u64 last = 0;
1733
1734 path = btrfs_alloc_path();
1735 if (!path)
1736 return -ENOMEM;
1737
1738 while (1) {
1739 cache = NULL;
1740 spin_lock(&root->fs_info->block_group_cache_lock);
1741 for (n = rb_first(&root->fs_info->block_group_cache_tree);
1742 n; n = rb_next(n)) {
1743 entry = rb_entry(n, struct btrfs_block_group_cache,
1744 cache_node);
1745 if (entry->dirty) {
1746 cache = entry;
1747 break;
1748 }
1749 }
1750 spin_unlock(&root->fs_info->block_group_cache_lock);
1751
1752 if (!cache)
1753 break;
1754
1755 cache->dirty = 0;
1756 last += cache->key.offset;
1757
1758 err = write_one_cache_group(trans, root,
1759 path, cache);
1760 /*
1761 * if we fail to write the cache group, we want
1762 * to keep it marked dirty in hopes that a later
1763 * write will work
1764 */
1765 if (err) {
1766 werr = err;
1767 continue;
1768 }
1769 }
1770 btrfs_free_path(path);
1771 return werr;
1772 }
1773
1774 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
1775 {
1776 struct btrfs_block_group_cache *block_group;
1777 int readonly = 0;
1778
1779 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
1780 if (!block_group || block_group->ro)
1781 readonly = 1;
1782 if (block_group)
1783 put_block_group(block_group);
1784 return readonly;
1785 }
1786
1787 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
1788 u64 total_bytes, u64 bytes_used,
1789 struct btrfs_space_info **space_info)
1790 {
1791 struct btrfs_space_info *found;
1792
1793 found = __find_space_info(info, flags);
1794 if (found) {
1795 spin_lock(&found->lock);
1796 found->total_bytes += total_bytes;
1797 found->bytes_used += bytes_used;
1798 found->full = 0;
1799 spin_unlock(&found->lock);
1800 *space_info = found;
1801 return 0;
1802 }
1803 found = kzalloc(sizeof(*found), GFP_NOFS);
1804 if (!found)
1805 return -ENOMEM;
1806
1807 list_add(&found->list, &info->space_info);
1808 INIT_LIST_HEAD(&found->block_groups);
1809 init_rwsem(&found->groups_sem);
1810 spin_lock_init(&found->lock);
1811 found->flags = flags;
1812 found->total_bytes = total_bytes;
1813 found->bytes_used = bytes_used;
1814 found->bytes_pinned = 0;
1815 found->bytes_reserved = 0;
1816 found->bytes_readonly = 0;
1817 found->full = 0;
1818 found->force_alloc = 0;
1819 *space_info = found;
1820 return 0;
1821 }
1822
1823 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
1824 {
1825 u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
1826 BTRFS_BLOCK_GROUP_RAID1 |
1827 BTRFS_BLOCK_GROUP_RAID10 |
1828 BTRFS_BLOCK_GROUP_DUP);
1829 if (extra_flags) {
1830 if (flags & BTRFS_BLOCK_GROUP_DATA)
1831 fs_info->avail_data_alloc_bits |= extra_flags;
1832 if (flags & BTRFS_BLOCK_GROUP_METADATA)
1833 fs_info->avail_metadata_alloc_bits |= extra_flags;
1834 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
1835 fs_info->avail_system_alloc_bits |= extra_flags;
1836 }
1837 }
1838
1839 static void set_block_group_readonly(struct btrfs_block_group_cache *cache)
1840 {
1841 spin_lock(&cache->space_info->lock);
1842 spin_lock(&cache->lock);
1843 if (!cache->ro) {
1844 cache->space_info->bytes_readonly += cache->key.offset -
1845 btrfs_block_group_used(&cache->item);
1846 cache->ro = 1;
1847 }
1848 spin_unlock(&cache->lock);
1849 spin_unlock(&cache->space_info->lock);
1850 }
1851
1852 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
1853 {
1854 u64 num_devices = root->fs_info->fs_devices->rw_devices;
1855
1856 if (num_devices == 1)
1857 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
1858 if (num_devices < 4)
1859 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
1860
1861 if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
1862 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
1863 BTRFS_BLOCK_GROUP_RAID10))) {
1864 flags &= ~BTRFS_BLOCK_GROUP_DUP;
1865 }
1866
1867 if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
1868 (flags & BTRFS_BLOCK_GROUP_RAID10)) {
1869 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
1870 }
1871
1872 if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
1873 ((flags & BTRFS_BLOCK_GROUP_RAID1) |
1874 (flags & BTRFS_BLOCK_GROUP_RAID10) |
1875 (flags & BTRFS_BLOCK_GROUP_DUP)))
1876 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
1877 return flags;
1878 }
1879
1880 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
1881 struct btrfs_root *extent_root, u64 alloc_bytes,
1882 u64 flags, int force)
1883 {
1884 struct btrfs_space_info *space_info;
1885 u64 thresh;
1886 int ret = 0;
1887
1888 mutex_lock(&extent_root->fs_info->chunk_mutex);
1889
1890 flags = btrfs_reduce_alloc_profile(extent_root, flags);
1891
1892 space_info = __find_space_info(extent_root->fs_info, flags);
1893 if (!space_info) {
1894 ret = update_space_info(extent_root->fs_info, flags,
1895 0, 0, &space_info);
1896 BUG_ON(ret);
1897 }
1898 BUG_ON(!space_info);
1899
1900 spin_lock(&space_info->lock);
1901 if (space_info->force_alloc) {
1902 force = 1;
1903 space_info->force_alloc = 0;
1904 }
1905 if (space_info->full) {
1906 spin_unlock(&space_info->lock);
1907 goto out;
1908 }
1909
1910 thresh = space_info->total_bytes - space_info->bytes_readonly;
1911 thresh = div_factor(thresh, 6);
1912 if (!force &&
1913 (space_info->bytes_used + space_info->bytes_pinned +
1914 space_info->bytes_reserved + alloc_bytes) < thresh) {
1915 spin_unlock(&space_info->lock);
1916 goto out;
1917 }
1918 spin_unlock(&space_info->lock);
1919
1920 ret = btrfs_alloc_chunk(trans, extent_root, flags);
1921 if (ret)
1922 space_info->full = 1;
1923 out:
1924 mutex_unlock(&extent_root->fs_info->chunk_mutex);
1925 return ret;
1926 }
1927
1928 static int update_block_group(struct btrfs_trans_handle *trans,
1929 struct btrfs_root *root,
1930 u64 bytenr, u64 num_bytes, int alloc,
1931 int mark_free)
1932 {
1933 struct btrfs_block_group_cache *cache;
1934 struct btrfs_fs_info *info = root->fs_info;
1935 u64 total = num_bytes;
1936 u64 old_val;
1937 u64 byte_in_group;
1938
1939 while (total) {
1940 cache = btrfs_lookup_block_group(info, bytenr);
1941 if (!cache)
1942 return -1;
1943 byte_in_group = bytenr - cache->key.objectid;
1944 WARN_ON(byte_in_group > cache->key.offset);
1945
1946 spin_lock(&cache->space_info->lock);
1947 spin_lock(&cache->lock);
1948 cache->dirty = 1;
1949 old_val = btrfs_block_group_used(&cache->item);
1950 num_bytes = min(total, cache->key.offset - byte_in_group);
1951 if (alloc) {
1952 old_val += num_bytes;
1953 cache->space_info->bytes_used += num_bytes;
1954 if (cache->ro)
1955 cache->space_info->bytes_readonly -= num_bytes;
1956 btrfs_set_block_group_used(&cache->item, old_val);
1957 spin_unlock(&cache->lock);
1958 spin_unlock(&cache->space_info->lock);
1959 } else {
1960 old_val -= num_bytes;
1961 cache->space_info->bytes_used -= num_bytes;
1962 if (cache->ro)
1963 cache->space_info->bytes_readonly += num_bytes;
1964 btrfs_set_block_group_used(&cache->item, old_val);
1965 spin_unlock(&cache->lock);
1966 spin_unlock(&cache->space_info->lock);
1967 if (mark_free) {
1968 int ret;
1969
1970 ret = btrfs_discard_extent(root, bytenr,
1971 num_bytes);
1972 WARN_ON(ret);
1973
1974 ret = btrfs_add_free_space(cache, bytenr,
1975 num_bytes);
1976 WARN_ON(ret);
1977 }
1978 }
1979 put_block_group(cache);
1980 total -= num_bytes;
1981 bytenr += num_bytes;
1982 }
1983 return 0;
1984 }
1985
1986 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
1987 {
1988 struct btrfs_block_group_cache *cache;
1989 u64 bytenr;
1990
1991 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
1992 if (!cache)
1993 return 0;
1994
1995 bytenr = cache->key.objectid;
1996 put_block_group(cache);
1997
1998 return bytenr;
1999 }
2000
2001 int btrfs_update_pinned_extents(struct btrfs_root *root,
2002 u64 bytenr, u64 num, int pin)
2003 {
2004 u64 len;
2005 struct btrfs_block_group_cache *cache;
2006 struct btrfs_fs_info *fs_info = root->fs_info;
2007
2008 WARN_ON(!mutex_is_locked(&root->fs_info->pinned_mutex));
2009 if (pin) {
2010 set_extent_dirty(&fs_info->pinned_extents,
2011 bytenr, bytenr + num - 1, GFP_NOFS);
2012 } else {
2013 clear_extent_dirty(&fs_info->pinned_extents,
2014 bytenr, bytenr + num - 1, GFP_NOFS);
2015 }
2016 while (num > 0) {
2017 cache = btrfs_lookup_block_group(fs_info, bytenr);
2018 BUG_ON(!cache);
2019 len = min(num, cache->key.offset -
2020 (bytenr - cache->key.objectid));
2021 if (pin) {
2022 spin_lock(&cache->space_info->lock);
2023 spin_lock(&cache->lock);
2024 cache->pinned += len;
2025 cache->space_info->bytes_pinned += len;
2026 spin_unlock(&cache->lock);
2027 spin_unlock(&cache->space_info->lock);
2028 fs_info->total_pinned += len;
2029 } else {
2030 spin_lock(&cache->space_info->lock);
2031 spin_lock(&cache->lock);
2032 cache->pinned -= len;
2033 cache->space_info->bytes_pinned -= len;
2034 spin_unlock(&cache->lock);
2035 spin_unlock(&cache->space_info->lock);
2036 fs_info->total_pinned -= len;
2037 if (cache->cached)
2038 btrfs_add_free_space(cache, bytenr, len);
2039 }
2040 put_block_group(cache);
2041 bytenr += len;
2042 num -= len;
2043 }
2044 return 0;
2045 }
2046
2047 static int update_reserved_extents(struct btrfs_root *root,
2048 u64 bytenr, u64 num, int reserve)
2049 {
2050 u64 len;
2051 struct btrfs_block_group_cache *cache;
2052 struct btrfs_fs_info *fs_info = root->fs_info;
2053
2054 while (num > 0) {
2055 cache = btrfs_lookup_block_group(fs_info, bytenr);
2056 BUG_ON(!cache);
2057 len = min(num, cache->key.offset -
2058 (bytenr - cache->key.objectid));
2059
2060 spin_lock(&cache->space_info->lock);
2061 spin_lock(&cache->lock);
2062 if (reserve) {
2063 cache->reserved += len;
2064 cache->space_info->bytes_reserved += len;
2065 } else {
2066 cache->reserved -= len;
2067 cache->space_info->bytes_reserved -= len;
2068 }
2069 spin_unlock(&cache->lock);
2070 spin_unlock(&cache->space_info->lock);
2071 put_block_group(cache);
2072 bytenr += len;
2073 num -= len;
2074 }
2075 return 0;
2076 }
2077
2078 int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
2079 {
2080 u64 last = 0;
2081 u64 start;
2082 u64 end;
2083 struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
2084 int ret;
2085
2086 mutex_lock(&root->fs_info->pinned_mutex);
2087 while (1) {
2088 ret = find_first_extent_bit(pinned_extents, last,
2089 &start, &end, EXTENT_DIRTY);
2090 if (ret)
2091 break;
2092 set_extent_dirty(copy, start, end, GFP_NOFS);
2093 last = end + 1;
2094 }
2095 mutex_unlock(&root->fs_info->pinned_mutex);
2096 return 0;
2097 }
2098
2099 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
2100 struct btrfs_root *root,
2101 struct extent_io_tree *unpin)
2102 {
2103 u64 start;
2104 u64 end;
2105 int ret;
2106
2107 mutex_lock(&root->fs_info->pinned_mutex);
2108 while (1) {
2109 ret = find_first_extent_bit(unpin, 0, &start, &end,
2110 EXTENT_DIRTY);
2111 if (ret)
2112 break;
2113
2114 ret = btrfs_discard_extent(root, start, end + 1 - start);
2115
2116 btrfs_update_pinned_extents(root, start, end + 1 - start, 0);
2117 clear_extent_dirty(unpin, start, end, GFP_NOFS);
2118
2119 if (need_resched()) {
2120 mutex_unlock(&root->fs_info->pinned_mutex);
2121 cond_resched();
2122 mutex_lock(&root->fs_info->pinned_mutex);
2123 }
2124 }
2125 mutex_unlock(&root->fs_info->pinned_mutex);
2126 return ret;
2127 }
2128
2129 static int finish_current_insert(struct btrfs_trans_handle *trans,
2130 struct btrfs_root *extent_root, int all)
2131 {
2132 u64 start;
2133 u64 end;
2134 u64 priv;
2135 u64 search = 0;
2136 u64 skipped = 0;
2137 struct btrfs_fs_info *info = extent_root->fs_info;
2138 struct btrfs_path *path;
2139 struct pending_extent_op *extent_op, *tmp;
2140 struct list_head insert_list, update_list;
2141 int ret;
2142 int num_inserts = 0, max_inserts;
2143
2144 path = btrfs_alloc_path();
2145 INIT_LIST_HEAD(&insert_list);
2146 INIT_LIST_HEAD(&update_list);
2147
2148 max_inserts = extent_root->leafsize /
2149 (2 * sizeof(struct btrfs_key) + 2 * sizeof(struct btrfs_item) +
2150 sizeof(struct btrfs_extent_ref) +
2151 sizeof(struct btrfs_extent_item));
2152 again:
2153 mutex_lock(&info->extent_ins_mutex);
2154 while (1) {
2155 ret = find_first_extent_bit(&info->extent_ins, search, &start,
2156 &end, EXTENT_WRITEBACK);
2157 if (ret) {
2158 if (skipped && all && !num_inserts &&
2159 list_empty(&update_list)) {
2160 skipped = 0;
2161 search = 0;
2162 continue;
2163 }
2164 mutex_unlock(&info->extent_ins_mutex);
2165 break;
2166 }
2167
2168 ret = try_lock_extent(&info->extent_ins, start, end, GFP_NOFS);
2169 if (!ret) {
2170 skipped = 1;
2171 search = end + 1;
2172 if (need_resched()) {
2173 mutex_unlock(&info->extent_ins_mutex);
2174 cond_resched();
2175 mutex_lock(&info->extent_ins_mutex);
2176 }
2177 continue;
2178 }
2179
2180 ret = get_state_private(&info->extent_ins, start, &priv);
2181 BUG_ON(ret);
2182 extent_op = (struct pending_extent_op *)(unsigned long) priv;
2183
2184 if (extent_op->type == PENDING_EXTENT_INSERT) {
2185 num_inserts++;
2186 list_add_tail(&extent_op->list, &insert_list);
2187 search = end + 1;
2188 if (num_inserts == max_inserts) {
2189 mutex_unlock(&info->extent_ins_mutex);
2190 break;
2191 }
2192 } else if (extent_op->type == PENDING_BACKREF_UPDATE) {
2193 list_add_tail(&extent_op->list, &update_list);
2194 search = end + 1;
2195 } else {
2196 BUG();
2197 }
2198 }
2199
2200 /*
2201 * process the update list, clear the writeback bit for it, and if
2202 * somebody marked this thing for deletion then just unlock it and be
2203 * done, the free_extents will handle it
2204 */
2205 mutex_lock(&info->extent_ins_mutex);
2206 list_for_each_entry_safe(extent_op, tmp, &update_list, list) {
2207 clear_extent_bits(&info->extent_ins, extent_op->bytenr,
2208 extent_op->bytenr + extent_op->num_bytes - 1,
2209 EXTENT_WRITEBACK, GFP_NOFS);
2210 if (extent_op->del) {
2211 list_del_init(&extent_op->list);
2212 unlock_extent(&info->extent_ins, extent_op->bytenr,
2213 extent_op->bytenr + extent_op->num_bytes
2214 - 1, GFP_NOFS);
2215 kfree(extent_op);
2216 }
2217 }
2218 mutex_unlock(&info->extent_ins_mutex);
2219
2220 /*
2221 * still have things left on the update list, go ahead an update
2222 * everything
2223 */
2224 if (!list_empty(&update_list)) {
2225 ret = update_backrefs(trans, extent_root, path, &update_list);
2226 BUG_ON(ret);
2227 }
2228
2229 /*
2230 * if no inserts need to be done, but we skipped some extents and we
2231 * need to make sure everything is cleaned then reset everything and
2232 * go back to the beginning
2233 */
2234 if (!num_inserts && all && skipped) {
2235 search = 0;
2236 skipped = 0;
2237 INIT_LIST_HEAD(&update_list);
2238 INIT_LIST_HEAD(&insert_list);
2239 goto again;
2240 } else if (!num_inserts) {
2241 goto out;
2242 }
2243
2244 /*
2245 * process the insert extents list. Again if we are deleting this
2246 * extent, then just unlock it, pin down the bytes if need be, and be
2247 * done with it. Saves us from having to actually insert the extent
2248 * into the tree and then subsequently come along and delete it
2249 */
2250 mutex_lock(&info->extent_ins_mutex);
2251 list_for_each_entry_safe(extent_op, tmp, &insert_list, list) {
2252 clear_extent_bits(&info->extent_ins, extent_op->bytenr,
2253 extent_op->bytenr + extent_op->num_bytes - 1,
2254 EXTENT_WRITEBACK, GFP_NOFS);
2255 if (extent_op->del) {
2256 u64 used;
2257 list_del_init(&extent_op->list);
2258 unlock_extent(&info->extent_ins, extent_op->bytenr,
2259 extent_op->bytenr + extent_op->num_bytes
2260 - 1, GFP_NOFS);
2261
2262 mutex_lock(&extent_root->fs_info->pinned_mutex);
2263 ret = pin_down_bytes(trans, extent_root,
2264 extent_op->bytenr,
2265 extent_op->num_bytes, 0);
2266 mutex_unlock(&extent_root->fs_info->pinned_mutex);
2267
2268 spin_lock(&info->delalloc_lock);
2269 used = btrfs_super_bytes_used(&info->super_copy);
2270 btrfs_set_super_bytes_used(&info->super_copy,
2271 used - extent_op->num_bytes);
2272 used = btrfs_root_used(&extent_root->root_item);
2273 btrfs_set_root_used(&extent_root->root_item,
2274 used - extent_op->num_bytes);
2275 spin_unlock(&info->delalloc_lock);
2276
2277 ret = update_block_group(trans, extent_root,
2278 extent_op->bytenr,
2279 extent_op->num_bytes,
2280 0, ret > 0);
2281 BUG_ON(ret);
2282 kfree(extent_op);
2283 num_inserts--;
2284 }
2285 }
2286 mutex_unlock(&info->extent_ins_mutex);
2287
2288 ret = insert_extents(trans, extent_root, path, &insert_list,
2289 num_inserts);
2290 BUG_ON(ret);
2291
2292 /*
2293 * if we broke out of the loop in order to insert stuff because we hit
2294 * the maximum number of inserts at a time we can handle, then loop
2295 * back and pick up where we left off
2296 */
2297 if (num_inserts == max_inserts) {
2298 INIT_LIST_HEAD(&insert_list);
2299 INIT_LIST_HEAD(&update_list);
2300 num_inserts = 0;
2301 goto again;
2302 }
2303
2304 /*
2305 * again, if we need to make absolutely sure there are no more pending
2306 * extent operations left and we know that we skipped some, go back to
2307 * the beginning and do it all again
2308 */
2309 if (all && skipped) {
2310 INIT_LIST_HEAD(&insert_list);
2311 INIT_LIST_HEAD(&update_list);
2312 search = 0;
2313 skipped = 0;
2314 num_inserts = 0;
2315 goto again;
2316 }
2317 out:
2318 btrfs_free_path(path);
2319 return 0;
2320 }
2321
2322 static int pin_down_bytes(struct btrfs_trans_handle *trans,
2323 struct btrfs_root *root,
2324 u64 bytenr, u64 num_bytes, int is_data)
2325 {
2326 int err = 0;
2327 struct extent_buffer *buf;
2328
2329 if (is_data)
2330 goto pinit;
2331
2332 buf = btrfs_find_tree_block(root, bytenr, num_bytes);
2333 if (!buf)
2334 goto pinit;
2335
2336 /* we can reuse a block if it hasn't been written
2337 * and it is from this transaction. We can't
2338 * reuse anything from the tree log root because
2339 * it has tiny sub-transactions.
2340 */
2341 if (btrfs_buffer_uptodate(buf, 0) &&
2342 btrfs_try_tree_lock(buf)) {
2343 u64 header_owner = btrfs_header_owner(buf);
2344 u64 header_transid = btrfs_header_generation(buf);
2345 if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
2346 header_owner != BTRFS_TREE_RELOC_OBJECTID &&
2347 header_transid == trans->transid &&
2348 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
2349 clean_tree_block(NULL, root, buf);
2350 btrfs_tree_unlock(buf);
2351 free_extent_buffer(buf);
2352 return 1;
2353 }
2354 btrfs_tree_unlock(buf);
2355 }
2356 free_extent_buffer(buf);
2357 pinit:
2358 btrfs_update_pinned_extents(root, bytenr, num_bytes, 1);
2359
2360 BUG_ON(err < 0);
2361 return 0;
2362 }
2363
2364 /*
2365 * remove an extent from the root, returns 0 on success
2366 */
2367 static int __free_extent(struct btrfs_trans_handle *trans,
2368 struct btrfs_root *root,
2369 u64 bytenr, u64 num_bytes, u64 parent,
2370 u64 root_objectid, u64 ref_generation,
2371 u64 owner_objectid, int pin, int mark_free)
2372 {
2373 struct btrfs_path *path;
2374 struct btrfs_key key;
2375 struct btrfs_fs_info *info = root->fs_info;
2376 struct btrfs_root *extent_root = info->extent_root;
2377 struct extent_buffer *leaf;
2378 int ret;
2379 int extent_slot = 0;
2380 int found_extent = 0;
2381 int num_to_del = 1;
2382 struct btrfs_extent_item *ei;
2383 u32 refs;
2384
2385 key.objectid = bytenr;
2386 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
2387 key.offset = num_bytes;
2388 path = btrfs_alloc_path();
2389 if (!path)
2390 return -ENOMEM;
2391
2392 path->reada = 1;
2393 ret = lookup_extent_backref(trans, extent_root, path,
2394 bytenr, parent, root_objectid,
2395 ref_generation, owner_objectid, 1);
2396 if (ret == 0) {
2397 struct btrfs_key found_key;
2398 extent_slot = path->slots[0];
2399 while (extent_slot > 0) {
2400 extent_slot--;
2401 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2402 extent_slot);
2403 if (found_key.objectid != bytenr)
2404 break;
2405 if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
2406 found_key.offset == num_bytes) {
2407 found_extent = 1;
2408 break;
2409 }
2410 if (path->slots[0] - extent_slot > 5)
2411 break;
2412 }
2413 if (!found_extent) {
2414 ret = remove_extent_backref(trans, extent_root, path);
2415 BUG_ON(ret);
2416 btrfs_release_path(extent_root, path);
2417 ret = btrfs_search_slot(trans, extent_root,
2418 &key, path, -1, 1);
2419 if (ret) {
2420 printk(KERN_ERR "umm, got %d back from search"
2421 ", was looking for %llu\n", ret,
2422 (unsigned long long)bytenr);
2423 btrfs_print_leaf(extent_root, path->nodes[0]);
2424 }
2425 BUG_ON(ret);
2426 extent_slot = path->slots[0];
2427 }
2428 } else {
2429 btrfs_print_leaf(extent_root, path->nodes[0]);
2430 WARN_ON(1);
2431 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
2432 "root %llu gen %llu owner %llu\n",
2433 (unsigned long long)bytenr,
2434 (unsigned long long)root_objectid,
2435 (unsigned long long)ref_generation,
2436 (unsigned long long)owner_objectid);
2437 }
2438
2439 leaf = path->nodes[0];
2440 ei = btrfs_item_ptr(leaf, extent_slot,
2441 struct btrfs_extent_item);
2442 refs = btrfs_extent_refs(leaf, ei);
2443 BUG_ON(refs == 0);
2444 refs -= 1;
2445 btrfs_set_extent_refs(leaf, ei, refs);
2446
2447 btrfs_mark_buffer_dirty(leaf);
2448
2449 if (refs == 0 && found_extent && path->slots[0] == extent_slot + 1) {
2450 struct btrfs_extent_ref *ref;
2451 ref = btrfs_item_ptr(leaf, path->slots[0],
2452 struct btrfs_extent_ref);
2453 BUG_ON(btrfs_ref_num_refs(leaf, ref) != 1);
2454 /* if the back ref and the extent are next to each other
2455 * they get deleted below in one shot
2456 */
2457 path->slots[0] = extent_slot;
2458 num_to_del = 2;
2459 } else if (found_extent) {
2460 /* otherwise delete the extent back ref */
2461 ret = remove_extent_backref(trans, extent_root, path);
2462 BUG_ON(ret);
2463 /* if refs are 0, we need to setup the path for deletion */
2464 if (refs == 0) {
2465 btrfs_release_path(extent_root, path);
2466 ret = btrfs_search_slot(trans, extent_root, &key, path,
2467 -1, 1);
2468 BUG_ON(ret);
2469 }
2470 }
2471
2472 if (refs == 0) {
2473 u64 super_used;
2474 u64 root_used;
2475
2476 if (pin) {
2477 mutex_lock(&root->fs_info->pinned_mutex);
2478 ret = pin_down_bytes(trans, root, bytenr, num_bytes,
2479 owner_objectid >= BTRFS_FIRST_FREE_OBJECTID);
2480 mutex_unlock(&root->fs_info->pinned_mutex);
2481 if (ret > 0)
2482 mark_free = 1;
2483 BUG_ON(ret < 0);
2484 }
2485 /* block accounting for super block */
2486 spin_lock(&info->delalloc_lock);
2487 super_used = btrfs_super_bytes_used(&info->super_copy);
2488 btrfs_set_super_bytes_used(&info->super_copy,
2489 super_used - num_bytes);
2490
2491 /* block accounting for root item */
2492 root_used = btrfs_root_used(&root->root_item);
2493 btrfs_set_root_used(&root->root_item,
2494 root_used - num_bytes);
2495 spin_unlock(&info->delalloc_lock);
2496 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
2497 num_to_del);
2498 BUG_ON(ret);
2499 btrfs_release_path(extent_root, path);
2500
2501 if (owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
2502 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
2503 BUG_ON(ret);
2504 }
2505
2506 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
2507 mark_free);
2508 BUG_ON(ret);
2509 }
2510 btrfs_free_path(path);
2511 finish_current_insert(trans, extent_root, 0);
2512 return ret;
2513 }
2514
2515 /*
2516 * find all the blocks marked as pending in the radix tree and remove
2517 * them from the extent map
2518 */
2519 static int del_pending_extents(struct btrfs_trans_handle *trans,
2520 struct btrfs_root *extent_root, int all)
2521 {
2522 int ret;
2523 int err = 0;
2524 u64 start;
2525 u64 end;
2526 u64 priv;
2527 u64 search = 0;
2528 int nr = 0, skipped = 0;
2529 struct extent_io_tree *pending_del;
2530 struct extent_io_tree *extent_ins;
2531 struct pending_extent_op *extent_op;
2532 struct btrfs_fs_info *info = extent_root->fs_info;
2533 struct list_head delete_list;
2534
2535 INIT_LIST_HEAD(&delete_list);
2536 extent_ins = &extent_root->fs_info->extent_ins;
2537 pending_del = &extent_root->fs_info->pending_del;
2538
2539 again:
2540 mutex_lock(&info->extent_ins_mutex);
2541 while (1) {
2542 ret = find_first_extent_bit(pending_del, search, &start, &end,
2543 EXTENT_WRITEBACK);
2544 if (ret) {
2545 if (all && skipped && !nr) {
2546 search = 0;
2547 skipped = 0;
2548 continue;
2549 }
2550 mutex_unlock(&info->extent_ins_mutex);
2551 break;
2552 }
2553
2554 ret = try_lock_extent(extent_ins, start, end, GFP_NOFS);
2555 if (!ret) {
2556 search = end+1;
2557 skipped = 1;
2558
2559 if (need_resched()) {
2560 mutex_unlock(&info->extent_ins_mutex);
2561 cond_resched();
2562 mutex_lock(&info->extent_ins_mutex);
2563 }
2564
2565 continue;
2566 }
2567 BUG_ON(ret < 0);
2568
2569 ret = get_state_private(pending_del, start, &priv);
2570 BUG_ON(ret);
2571 extent_op = (struct pending_extent_op *)(unsigned long)priv;
2572
2573 clear_extent_bits(pending_del, start, end, EXTENT_WRITEBACK,
2574 GFP_NOFS);
2575 if (!test_range_bit(extent_ins, start, end,
2576 EXTENT_WRITEBACK, 0)) {
2577 list_add_tail(&extent_op->list, &delete_list);
2578 nr++;
2579 } else {
2580 kfree(extent_op);
2581
2582 ret = get_state_private(&info->extent_ins, start,
2583 &priv);
2584 BUG_ON(ret);
2585 extent_op = (struct pending_extent_op *)
2586 (unsigned long)priv;
2587
2588 clear_extent_bits(&info->extent_ins, start, end,
2589 EXTENT_WRITEBACK, GFP_NOFS);
2590
2591 if (extent_op->type == PENDING_BACKREF_UPDATE) {
2592 list_add_tail(&extent_op->list, &delete_list);
2593 search = end + 1;
2594 nr++;
2595 continue;
2596 }
2597
2598 mutex_lock(&extent_root->fs_info->pinned_mutex);
2599 ret = pin_down_bytes(trans, extent_root, start,
2600 end + 1 - start, 0);
2601 mutex_unlock(&extent_root->fs_info->pinned_mutex);
2602
2603 ret = update_block_group(trans, extent_root, start,
2604 end + 1 - start, 0, ret > 0);
2605
2606 unlock_extent(extent_ins, start, end, GFP_NOFS);
2607 BUG_ON(ret);
2608 kfree(extent_op);
2609 }
2610 if (ret)
2611 err = ret;
2612
2613 search = end + 1;
2614
2615 if (need_resched()) {
2616 mutex_unlock(&info->extent_ins_mutex);
2617 cond_resched();
2618 mutex_lock(&info->extent_ins_mutex);
2619 }
2620 }
2621
2622 if (nr) {
2623 ret = free_extents(trans, extent_root, &delete_list);
2624 BUG_ON(ret);
2625 }
2626
2627 if (all && skipped) {
2628 INIT_LIST_HEAD(&delete_list);
2629 search = 0;
2630 nr = 0;
2631 goto again;
2632 }
2633
2634 return err;
2635 }
2636
2637 /*
2638 * remove an extent from the root, returns 0 on success
2639 */
2640 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
2641 struct btrfs_root *root,
2642 u64 bytenr, u64 num_bytes, u64 parent,
2643 u64 root_objectid, u64 ref_generation,
2644 u64 owner_objectid, int pin)
2645 {
2646 struct btrfs_root *extent_root = root->fs_info->extent_root;
2647 int pending_ret;
2648 int ret;
2649
2650 WARN_ON(num_bytes < root->sectorsize);
2651 if (root == extent_root) {
2652 struct pending_extent_op *extent_op = NULL;
2653
2654 mutex_lock(&root->fs_info->extent_ins_mutex);
2655 if (test_range_bit(&root->fs_info->extent_ins, bytenr,
2656 bytenr + num_bytes - 1, EXTENT_WRITEBACK, 0)) {
2657 u64 priv;
2658 ret = get_state_private(&root->fs_info->extent_ins,
2659 bytenr, &priv);
2660 BUG_ON(ret);
2661 extent_op = (struct pending_extent_op *)
2662 (unsigned long)priv;
2663
2664 extent_op->del = 1;
2665 if (extent_op->type == PENDING_EXTENT_INSERT) {
2666 mutex_unlock(&root->fs_info->extent_ins_mutex);
2667 return 0;
2668 }
2669 }
2670
2671 if (extent_op) {
2672 ref_generation = extent_op->orig_generation;
2673 parent = extent_op->orig_parent;
2674 }
2675
2676 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2677 BUG_ON(!extent_op);
2678
2679 extent_op->type = PENDING_EXTENT_DELETE;
2680 extent_op->bytenr = bytenr;
2681 extent_op->num_bytes = num_bytes;
2682 extent_op->parent = parent;
2683 extent_op->orig_parent = parent;
2684 extent_op->generation = ref_generation;
2685 extent_op->orig_generation = ref_generation;
2686 extent_op->level = (int)owner_objectid;
2687 INIT_LIST_HEAD(&extent_op->list);
2688 extent_op->del = 0;
2689
2690 set_extent_bits(&root->fs_info->pending_del,
2691 bytenr, bytenr + num_bytes - 1,
2692 EXTENT_WRITEBACK, GFP_NOFS);
2693 set_state_private(&root->fs_info->pending_del,
2694 bytenr, (unsigned long)extent_op);
2695 mutex_unlock(&root->fs_info->extent_ins_mutex);
2696 return 0;
2697 }
2698 /* if metadata always pin */
2699 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
2700 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
2701 struct btrfs_block_group_cache *cache;
2702
2703 /* btrfs_free_reserved_extent */
2704 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
2705 BUG_ON(!cache);
2706 btrfs_add_free_space(cache, bytenr, num_bytes);
2707 put_block_group(cache);
2708 update_reserved_extents(root, bytenr, num_bytes, 0);
2709 return 0;
2710 }
2711 pin = 1;
2712 }
2713
2714 /* if data pin when any transaction has committed this */
2715 if (ref_generation != trans->transid)
2716 pin = 1;
2717
2718 ret = __free_extent(trans, root, bytenr, num_bytes, parent,
2719 root_objectid, ref_generation,
2720 owner_objectid, pin, pin == 0);
2721
2722 finish_current_insert(trans, root->fs_info->extent_root, 0);
2723 pending_ret = del_pending_extents(trans, root->fs_info->extent_root, 0);
2724 return ret ? ret : pending_ret;
2725 }
2726
2727 int btrfs_free_extent(struct btrfs_trans_handle *trans,
2728 struct btrfs_root *root,
2729 u64 bytenr, u64 num_bytes, u64 parent,
2730 u64 root_objectid, u64 ref_generation,
2731 u64 owner_objectid, int pin)
2732 {
2733 int ret;
2734
2735 ret = __btrfs_free_extent(trans, root, bytenr, num_bytes, parent,
2736 root_objectid, ref_generation,
2737 owner_objectid, pin);
2738 return ret;
2739 }
2740
2741 static u64 stripe_align(struct btrfs_root *root, u64 val)
2742 {
2743 u64 mask = ((u64)root->stripesize - 1);
2744 u64 ret = (val + mask) & ~mask;
2745 return ret;
2746 }
2747
2748 /*
2749 * walks the btree of allocated extents and find a hole of a given size.
2750 * The key ins is changed to record the hole:
2751 * ins->objectid == block start
2752 * ins->flags = BTRFS_EXTENT_ITEM_KEY
2753 * ins->offset == number of blocks
2754 * Any available blocks before search_start are skipped.
2755 */
2756 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
2757 struct btrfs_root *orig_root,
2758 u64 num_bytes, u64 empty_size,
2759 u64 search_start, u64 search_end,
2760 u64 hint_byte, struct btrfs_key *ins,
2761 u64 exclude_start, u64 exclude_nr,
2762 int data)
2763 {
2764 int ret = 0;
2765 struct btrfs_root *root = orig_root->fs_info->extent_root;
2766 u64 total_needed = num_bytes;
2767 u64 *last_ptr = NULL;
2768 u64 last_wanted = 0;
2769 struct btrfs_block_group_cache *block_group = NULL;
2770 int chunk_alloc_done = 0;
2771 int empty_cluster = 2 * 1024 * 1024;
2772 int allowed_chunk_alloc = 0;
2773 struct list_head *head = NULL, *cur = NULL;
2774 int loop = 0;
2775 int extra_loop = 0;
2776 struct btrfs_space_info *space_info;
2777
2778 WARN_ON(num_bytes < root->sectorsize);
2779 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
2780 ins->objectid = 0;
2781 ins->offset = 0;
2782
2783 if (orig_root->ref_cows || empty_size)
2784 allowed_chunk_alloc = 1;
2785
2786 if (data & BTRFS_BLOCK_GROUP_METADATA) {
2787 last_ptr = &root->fs_info->last_alloc;
2788 empty_cluster = 64 * 1024;
2789 }
2790
2791 if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD))
2792 last_ptr = &root->fs_info->last_data_alloc;
2793
2794 if (last_ptr) {
2795 if (*last_ptr) {
2796 hint_byte = *last_ptr;
2797 last_wanted = *last_ptr;
2798 } else
2799 empty_size += empty_cluster;
2800 } else {
2801 empty_cluster = 0;
2802 }
2803 search_start = max(search_start, first_logical_byte(root, 0));
2804 search_start = max(search_start, hint_byte);
2805
2806 if (last_wanted && search_start != last_wanted) {
2807 last_wanted = 0;
2808 empty_size += empty_cluster;
2809 }
2810
2811 total_needed += empty_size;
2812 block_group = btrfs_lookup_block_group(root->fs_info, search_start);
2813 if (!block_group)
2814 block_group = btrfs_lookup_first_block_group(root->fs_info,
2815 search_start);
2816 space_info = __find_space_info(root->fs_info, data);
2817
2818 down_read(&space_info->groups_sem);
2819 while (1) {
2820 struct btrfs_free_space *free_space;
2821 /*
2822 * the only way this happens if our hint points to a block
2823 * group thats not of the proper type, while looping this
2824 * should never happen
2825 */
2826 if (empty_size)
2827 extra_loop = 1;
2828
2829 if (!block_group)
2830 goto new_group_no_lock;
2831
2832 if (unlikely(!block_group->cached)) {
2833 mutex_lock(&block_group->cache_mutex);
2834 ret = cache_block_group(root, block_group);
2835 mutex_unlock(&block_group->cache_mutex);
2836 if (ret)
2837 break;
2838 }
2839
2840 mutex_lock(&block_group->alloc_mutex);
2841 if (unlikely(!block_group_bits(block_group, data)))
2842 goto new_group;
2843
2844 if (unlikely(block_group->ro))
2845 goto new_group;
2846
2847 free_space = btrfs_find_free_space(block_group, search_start,
2848 total_needed);
2849 if (free_space) {
2850 u64 start = block_group->key.objectid;
2851 u64 end = block_group->key.objectid +
2852 block_group->key.offset;
2853
2854 search_start = stripe_align(root, free_space->offset);
2855
2856 /* move on to the next group */
2857 if (search_start + num_bytes >= search_end)
2858 goto new_group;
2859
2860 /* move on to the next group */
2861 if (search_start + num_bytes > end)
2862 goto new_group;
2863
2864 if (last_wanted && search_start != last_wanted) {
2865 total_needed += empty_cluster;
2866 empty_size += empty_cluster;
2867 last_wanted = 0;
2868 /*
2869 * if search_start is still in this block group
2870 * then we just re-search this block group
2871 */
2872 if (search_start >= start &&
2873 search_start < end) {
2874 mutex_unlock(&block_group->alloc_mutex);
2875 continue;
2876 }
2877
2878 /* else we go to the next block group */
2879 goto new_group;
2880 }
2881
2882 if (exclude_nr > 0 &&
2883 (search_start + num_bytes > exclude_start &&
2884 search_start < exclude_start + exclude_nr)) {
2885 search_start = exclude_start + exclude_nr;
2886 /*
2887 * if search_start is still in this block group
2888 * then we just re-search this block group
2889 */
2890 if (search_start >= start &&
2891 search_start < end) {
2892 mutex_unlock(&block_group->alloc_mutex);
2893 last_wanted = 0;
2894 continue;
2895 }
2896
2897 /* else we go to the next block group */
2898 goto new_group;
2899 }
2900
2901 ins->objectid = search_start;
2902 ins->offset = num_bytes;
2903
2904 btrfs_remove_free_space_lock(block_group, search_start,
2905 num_bytes);
2906 /* we are all good, lets return */
2907 mutex_unlock(&block_group->alloc_mutex);
2908 break;
2909 }
2910 new_group:
2911 mutex_unlock(&block_group->alloc_mutex);
2912 put_block_group(block_group);
2913 block_group = NULL;
2914 new_group_no_lock:
2915 /* don't try to compare new allocations against the
2916 * last allocation any more
2917 */
2918 last_wanted = 0;
2919
2920 /*
2921 * Here's how this works.
2922 * loop == 0: we were searching a block group via a hint
2923 * and didn't find anything, so we start at
2924 * the head of the block groups and keep searching
2925 * loop == 1: we're searching through all of the block groups
2926 * if we hit the head again we have searched
2927 * all of the block groups for this space and we
2928 * need to try and allocate, if we cant error out.
2929 * loop == 2: we allocated more space and are looping through
2930 * all of the block groups again.
2931 */
2932 if (loop == 0) {
2933 head = &space_info->block_groups;
2934 cur = head->next;
2935 loop++;
2936 } else if (loop == 1 && cur == head) {
2937 int keep_going;
2938
2939 /* at this point we give up on the empty_size
2940 * allocations and just try to allocate the min
2941 * space.
2942 *
2943 * The extra_loop field was set if an empty_size
2944 * allocation was attempted above, and if this
2945 * is try we need to try the loop again without
2946 * the additional empty_size.
2947 */
2948 total_needed -= empty_size;
2949 empty_size = 0;
2950 keep_going = extra_loop;
2951 loop++;
2952
2953 if (allowed_chunk_alloc && !chunk_alloc_done) {
2954 up_read(&space_info->groups_sem);
2955 ret = do_chunk_alloc(trans, root, num_bytes +
2956 2 * 1024 * 1024, data, 1);
2957 down_read(&space_info->groups_sem);
2958 if (ret < 0)
2959 goto loop_check;
2960 head = &space_info->block_groups;
2961 /*
2962 * we've allocated a new chunk, keep
2963 * trying
2964 */
2965 keep_going = 1;
2966 chunk_alloc_done = 1;
2967 } else if (!allowed_chunk_alloc) {
2968 space_info->force_alloc = 1;
2969 }
2970 loop_check:
2971 if (keep_going) {
2972 cur = head->next;
2973 extra_loop = 0;
2974 } else {
2975 break;
2976 }
2977 } else if (cur == head) {
2978 break;
2979 }
2980
2981 block_group = list_entry(cur, struct btrfs_block_group_cache,
2982 list);
2983 atomic_inc(&block_group->count);
2984
2985 search_start = block_group->key.objectid;
2986 cur = cur->next;
2987 }
2988
2989 /* we found what we needed */
2990 if (ins->objectid) {
2991 if (!(data & BTRFS_BLOCK_GROUP_DATA))
2992 trans->block_group = block_group->key.objectid;
2993
2994 if (last_ptr)
2995 *last_ptr = ins->objectid + ins->offset;
2996 ret = 0;
2997 } else if (!ret) {
2998 printk(KERN_ERR "btrfs searching for %llu bytes, "
2999 "num_bytes %llu, loop %d, allowed_alloc %d\n",
3000 (unsigned long long)total_needed,
3001 (unsigned long long)num_bytes,
3002 loop, allowed_chunk_alloc);
3003 ret = -ENOSPC;
3004 }
3005 if (block_group)
3006 put_block_group(block_group);
3007
3008 up_read(&space_info->groups_sem);
3009 return ret;
3010 }
3011
3012 static void dump_space_info(struct btrfs_space_info *info, u64 bytes)
3013 {
3014 struct btrfs_block_group_cache *cache;
3015
3016 printk(KERN_INFO "space_info has %llu free, is %sfull\n",
3017 (unsigned long long)(info->total_bytes - info->bytes_used -
3018 info->bytes_pinned - info->bytes_reserved),
3019 (info->full) ? "" : "not ");
3020
3021 down_read(&info->groups_sem);
3022 list_for_each_entry(cache, &info->block_groups, list) {
3023 spin_lock(&cache->lock);
3024 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
3025 "%llu pinned %llu reserved\n",
3026 (unsigned long long)cache->key.objectid,
3027 (unsigned long long)cache->key.offset,
3028 (unsigned long long)btrfs_block_group_used(&cache->item),
3029 (unsigned long long)cache->pinned,
3030 (unsigned long long)cache->reserved);
3031 btrfs_dump_free_space(cache, bytes);
3032 spin_unlock(&cache->lock);
3033 }
3034 up_read(&info->groups_sem);
3035 }
3036
3037 static int __btrfs_reserve_extent(struct btrfs_trans_handle *trans,
3038 struct btrfs_root *root,
3039 u64 num_bytes, u64 min_alloc_size,
3040 u64 empty_size, u64 hint_byte,
3041 u64 search_end, struct btrfs_key *ins,
3042 u64 data)
3043 {
3044 int ret;
3045 u64 search_start = 0;
3046 u64 alloc_profile;
3047 struct btrfs_fs_info *info = root->fs_info;
3048
3049 if (data) {
3050 alloc_profile = info->avail_data_alloc_bits &
3051 info->data_alloc_profile;
3052 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
3053 } else if (root == root->fs_info->chunk_root) {
3054 alloc_profile = info->avail_system_alloc_bits &
3055 info->system_alloc_profile;
3056 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
3057 } else {
3058 alloc_profile = info->avail_metadata_alloc_bits &
3059 info->metadata_alloc_profile;
3060 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
3061 }
3062 again:
3063 data = btrfs_reduce_alloc_profile(root, data);
3064 /*
3065 * the only place that sets empty_size is btrfs_realloc_node, which
3066 * is not called recursively on allocations
3067 */
3068 if (empty_size || root->ref_cows) {
3069 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
3070 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3071 2 * 1024 * 1024,
3072 BTRFS_BLOCK_GROUP_METADATA |
3073 (info->metadata_alloc_profile &
3074 info->avail_metadata_alloc_bits), 0);
3075 }
3076 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3077 num_bytes + 2 * 1024 * 1024, data, 0);
3078 }
3079
3080 WARN_ON(num_bytes < root->sectorsize);
3081 ret = find_free_extent(trans, root, num_bytes, empty_size,
3082 search_start, search_end, hint_byte, ins,
3083 trans->alloc_exclude_start,
3084 trans->alloc_exclude_nr, data);
3085
3086 if (ret == -ENOSPC && num_bytes > min_alloc_size) {
3087 num_bytes = num_bytes >> 1;
3088 num_bytes = num_bytes & ~(root->sectorsize - 1);
3089 num_bytes = max(num_bytes, min_alloc_size);
3090 do_chunk_alloc(trans, root->fs_info->extent_root,
3091 num_bytes, data, 1);
3092 goto again;
3093 }
3094 if (ret) {
3095 struct btrfs_space_info *sinfo;
3096
3097 sinfo = __find_space_info(root->fs_info, data);
3098 printk(KERN_ERR "btrfs allocation failed flags %llu, "
3099 "wanted %llu\n", (unsigned long long)data,
3100 (unsigned long long)num_bytes);
3101 dump_space_info(sinfo, num_bytes);
3102 BUG();
3103 }
3104
3105 return ret;
3106 }
3107
3108 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
3109 {
3110 struct btrfs_block_group_cache *cache;
3111 int ret = 0;
3112
3113 cache = btrfs_lookup_block_group(root->fs_info, start);
3114 if (!cache) {
3115 printk(KERN_ERR "Unable to find block group for %llu\n",
3116 (unsigned long long)start);
3117 return -ENOSPC;
3118 }
3119
3120 ret = btrfs_discard_extent(root, start, len);
3121
3122 btrfs_add_free_space(cache, start, len);
3123 put_block_group(cache);
3124 update_reserved_extents(root, start, len, 0);
3125
3126 return ret;
3127 }
3128
3129 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
3130 struct btrfs_root *root,
3131 u64 num_bytes, u64 min_alloc_size,
3132 u64 empty_size, u64 hint_byte,
3133 u64 search_end, struct btrfs_key *ins,
3134 u64 data)
3135 {
3136 int ret;
3137 ret = __btrfs_reserve_extent(trans, root, num_bytes, min_alloc_size,
3138 empty_size, hint_byte, search_end, ins,
3139 data);
3140 update_reserved_extents(root, ins->objectid, ins->offset, 1);
3141 return ret;
3142 }
3143
3144 static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
3145 struct btrfs_root *root, u64 parent,
3146 u64 root_objectid, u64 ref_generation,
3147 u64 owner, struct btrfs_key *ins)
3148 {
3149 int ret;
3150 int pending_ret;
3151 u64 super_used;
3152 u64 root_used;
3153 u64 num_bytes = ins->offset;
3154 u32 sizes[2];
3155 struct btrfs_fs_info *info = root->fs_info;
3156 struct btrfs_root *extent_root = info->extent_root;
3157 struct btrfs_extent_item *extent_item;
3158 struct btrfs_extent_ref *ref;
3159 struct btrfs_path *path;
3160 struct btrfs_key keys[2];
3161
3162 if (parent == 0)
3163 parent = ins->objectid;
3164
3165 /* block accounting for super block */
3166 spin_lock(&info->delalloc_lock);
3167 super_used = btrfs_super_bytes_used(&info->super_copy);
3168 btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes);
3169
3170 /* block accounting for root item */
3171 root_used = btrfs_root_used(&root->root_item);
3172 btrfs_set_root_used(&root->root_item, root_used + num_bytes);
3173 spin_unlock(&info->delalloc_lock);
3174
3175 if (root == extent_root) {
3176 struct pending_extent_op *extent_op;
3177
3178 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
3179 BUG_ON(!extent_op);
3180
3181 extent_op->type = PENDING_EXTENT_INSERT;
3182 extent_op->bytenr = ins->objectid;
3183 extent_op->num_bytes = ins->offset;
3184 extent_op->parent = parent;
3185 extent_op->orig_parent = 0;
3186 extent_op->generation = ref_generation;
3187 extent_op->orig_generation = 0;
3188 extent_op->level = (int)owner;
3189 INIT_LIST_HEAD(&extent_op->list);
3190 extent_op->del = 0;
3191
3192 mutex_lock(&root->fs_info->extent_ins_mutex);
3193 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
3194 ins->objectid + ins->offset - 1,
3195 EXTENT_WRITEBACK, GFP_NOFS);
3196 set_state_private(&root->fs_info->extent_ins,
3197 ins->objectid, (unsigned long)extent_op);
3198 mutex_unlock(&root->fs_info->extent_ins_mutex);
3199 goto update_block;
3200 }
3201
3202 memcpy(&keys[0], ins, sizeof(*ins));
3203 keys[1].objectid = ins->objectid;
3204 keys[1].type = BTRFS_EXTENT_REF_KEY;
3205 keys[1].offset = parent;
3206 sizes[0] = sizeof(*extent_item);
3207 sizes[1] = sizeof(*ref);
3208
3209 path = btrfs_alloc_path();
3210 BUG_ON(!path);
3211
3212 ret = btrfs_insert_empty_items(trans, extent_root, path, keys,
3213 sizes, 2);
3214 BUG_ON(ret);
3215
3216 extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3217 struct btrfs_extent_item);
3218 btrfs_set_extent_refs(path->nodes[0], extent_item, 1);
3219 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
3220 struct btrfs_extent_ref);
3221
3222 btrfs_set_ref_root(path->nodes[0], ref, root_objectid);
3223 btrfs_set_ref_generation(path->nodes[0], ref, ref_generation);
3224 btrfs_set_ref_objectid(path->nodes[0], ref, owner);
3225 btrfs_set_ref_num_refs(path->nodes[0], ref, 1);
3226
3227 btrfs_mark_buffer_dirty(path->nodes[0]);
3228
3229 trans->alloc_exclude_start = 0;
3230 trans->alloc_exclude_nr = 0;
3231 btrfs_free_path(path);
3232 finish_current_insert(trans, extent_root, 0);
3233 pending_ret = del_pending_extents(trans, extent_root, 0);
3234
3235 if (ret)
3236 goto out;
3237 if (pending_ret) {
3238 ret = pending_ret;
3239 goto out;
3240 }
3241
3242 update_block:
3243 ret = update_block_group(trans, root, ins->objectid,
3244 ins->offset, 1, 0);
3245 if (ret) {
3246 printk(KERN_ERR "btrfs update block group failed for %llu "
3247 "%llu\n", (unsigned long long)ins->objectid,
3248 (unsigned long long)ins->offset);
3249 BUG();
3250 }
3251 out:
3252 return ret;
3253 }
3254
3255 int btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
3256 struct btrfs_root *root, u64 parent,
3257 u64 root_objectid, u64 ref_generation,
3258 u64 owner, struct btrfs_key *ins)
3259 {
3260 int ret;
3261
3262 if (root_objectid == BTRFS_TREE_LOG_OBJECTID)
3263 return 0;
3264 ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
3265 ref_generation, owner, ins);
3266 update_reserved_extents(root, ins->objectid, ins->offset, 0);
3267 return ret;
3268 }
3269
3270 /*
3271 * this is used by the tree logging recovery code. It records that
3272 * an extent has been allocated and makes sure to clear the free
3273 * space cache bits as well
3274 */
3275 int btrfs_alloc_logged_extent(struct btrfs_trans_handle *trans,
3276 struct btrfs_root *root, u64 parent,
3277 u64 root_objectid, u64 ref_generation,
3278 u64 owner, struct btrfs_key *ins)
3279 {
3280 int ret;
3281 struct btrfs_block_group_cache *block_group;
3282
3283 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
3284 mutex_lock(&block_group->cache_mutex);
3285 cache_block_group(root, block_group);
3286 mutex_unlock(&block_group->cache_mutex);
3287
3288 ret = btrfs_remove_free_space(block_group, ins->objectid,
3289 ins->offset);
3290 BUG_ON(ret);
3291 put_block_group(block_group);
3292 ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
3293 ref_generation, owner, ins);
3294 return ret;
3295 }
3296
3297 /*
3298 * finds a free extent and does all the dirty work required for allocation
3299 * returns the key for the extent through ins, and a tree buffer for
3300 * the first block of the extent through buf.
3301 *
3302 * returns 0 if everything worked, non-zero otherwise.
3303 */
3304 int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
3305 struct btrfs_root *root,
3306 u64 num_bytes, u64 parent, u64 min_alloc_size,
3307 u64 root_objectid, u64 ref_generation,
3308 u64 owner_objectid, u64 empty_size, u64 hint_byte,
3309 u64 search_end, struct btrfs_key *ins, u64 data)
3310 {
3311 int ret;
3312
3313 ret = __btrfs_reserve_extent(trans, root, num_bytes,
3314 min_alloc_size, empty_size, hint_byte,
3315 search_end, ins, data);
3316 BUG_ON(ret);
3317 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
3318 ret = __btrfs_alloc_reserved_extent(trans, root, parent,
3319 root_objectid, ref_generation,
3320 owner_objectid, ins);
3321 BUG_ON(ret);
3322
3323 } else {
3324 update_reserved_extents(root, ins->objectid, ins->offset, 1);
3325 }
3326 return ret;
3327 }
3328
3329 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
3330 struct btrfs_root *root,
3331 u64 bytenr, u32 blocksize)
3332 {
3333 struct extent_buffer *buf;
3334
3335 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
3336 if (!buf)
3337 return ERR_PTR(-ENOMEM);
3338 btrfs_set_header_generation(buf, trans->transid);
3339 btrfs_tree_lock(buf);
3340 clean_tree_block(trans, root, buf);
3341 btrfs_set_buffer_uptodate(buf);
3342 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
3343 set_extent_dirty(&root->dirty_log_pages, buf->start,
3344 buf->start + buf->len - 1, GFP_NOFS);
3345 } else {
3346 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
3347 buf->start + buf->len - 1, GFP_NOFS);
3348 }
3349 trans->blocks_used++;
3350 return buf;
3351 }
3352
3353 /*
3354 * helper function to allocate a block for a given tree
3355 * returns the tree buffer or NULL.
3356 */
3357 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
3358 struct btrfs_root *root,
3359 u32 blocksize, u64 parent,
3360 u64 root_objectid,
3361 u64 ref_generation,
3362 int level,
3363 u64 hint,
3364 u64 empty_size)
3365 {
3366 struct btrfs_key ins;
3367 int ret;
3368 struct extent_buffer *buf;
3369
3370 ret = btrfs_alloc_extent(trans, root, blocksize, parent, blocksize,
3371 root_objectid, ref_generation, level,
3372 empty_size, hint, (u64)-1, &ins, 0);
3373 if (ret) {
3374 BUG_ON(ret > 0);
3375 return ERR_PTR(ret);
3376 }
3377
3378 buf = btrfs_init_new_buffer(trans, root, ins.objectid, blocksize);
3379 return buf;
3380 }
3381
3382 int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans,
3383 struct btrfs_root *root, struct extent_buffer *leaf)
3384 {
3385 u64 leaf_owner;
3386 u64 leaf_generation;
3387 struct btrfs_key key;
3388 struct btrfs_file_extent_item *fi;
3389 int i;
3390 int nritems;
3391 int ret;
3392
3393 BUG_ON(!btrfs_is_leaf(leaf));
3394 nritems = btrfs_header_nritems(leaf);
3395 leaf_owner = btrfs_header_owner(leaf);
3396 leaf_generation = btrfs_header_generation(leaf);
3397
3398 for (i = 0; i < nritems; i++) {
3399 u64 disk_bytenr;
3400 cond_resched();
3401
3402 btrfs_item_key_to_cpu(leaf, &key, i);
3403 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
3404 continue;
3405 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
3406 if (btrfs_file_extent_type(leaf, fi) ==
3407 BTRFS_FILE_EXTENT_INLINE)
3408 continue;
3409 /*
3410 * FIXME make sure to insert a trans record that
3411 * repeats the snapshot del on crash
3412 */
3413 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
3414 if (disk_bytenr == 0)
3415 continue;
3416
3417 ret = __btrfs_free_extent(trans, root, disk_bytenr,
3418 btrfs_file_extent_disk_num_bytes(leaf, fi),
3419 leaf->start, leaf_owner, leaf_generation,
3420 key.objectid, 0);
3421 BUG_ON(ret);
3422
3423 atomic_inc(&root->fs_info->throttle_gen);
3424 wake_up(&root->fs_info->transaction_throttle);
3425 cond_resched();
3426 }
3427 return 0;
3428 }
3429
3430 static noinline int cache_drop_leaf_ref(struct btrfs_trans_handle *trans,
3431 struct btrfs_root *root,
3432 struct btrfs_leaf_ref *ref)
3433 {
3434 int i;
3435 int ret;
3436 struct btrfs_extent_info *info = ref->extents;
3437
3438 for (i = 0; i < ref->nritems; i++) {
3439 ret = __btrfs_free_extent(trans, root, info->bytenr,
3440 info->num_bytes, ref->bytenr,
3441 ref->owner, ref->generation,
3442 info->objectid, 0);
3443
3444 atomic_inc(&root->fs_info->throttle_gen);
3445 wake_up(&root->fs_info->transaction_throttle);
3446 cond_resched();
3447
3448 BUG_ON(ret);
3449 info++;
3450 }
3451
3452 return 0;
3453 }
3454
3455 static int drop_snap_lookup_refcount(struct btrfs_root *root, u64 start,
3456 u64 len, u32 *refs)
3457 {
3458 int ret;
3459
3460 ret = btrfs_lookup_extent_ref(NULL, root, start, len, refs);
3461 BUG_ON(ret);
3462
3463 #if 0 /* some debugging code in case we see problems here */
3464 /* if the refs count is one, it won't get increased again. But
3465 * if the ref count is > 1, someone may be decreasing it at
3466 * the same time we are.
3467 */
3468 if (*refs != 1) {
3469 struct extent_buffer *eb = NULL;
3470 eb = btrfs_find_create_tree_block(root, start, len);
3471 if (eb)
3472 btrfs_tree_lock(eb);
3473
3474 mutex_lock(&root->fs_info->alloc_mutex);
3475 ret = lookup_extent_ref(NULL, root, start, len, refs);
3476 BUG_ON(ret);
3477 mutex_unlock(&root->fs_info->alloc_mutex);
3478
3479 if (eb) {
3480 btrfs_tree_unlock(eb);
3481 free_extent_buffer(eb);
3482 }
3483 if (*refs == 1) {
3484 printk(KERN_ERR "btrfs block %llu went down to one "
3485 "during drop_snap\n", (unsigned long long)start);
3486 }
3487
3488 }
3489 #endif
3490
3491 cond_resched();
3492 return ret;
3493 }
3494
3495 /*
3496 * helper function for drop_snapshot, this walks down the tree dropping ref
3497 * counts as it goes.
3498 */
3499 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
3500 struct btrfs_root *root,
3501 struct btrfs_path *path, int *level)
3502 {
3503 u64 root_owner;
3504 u64 root_gen;
3505 u64 bytenr;
3506 u64 ptr_gen;
3507 struct extent_buffer *next;
3508 struct extent_buffer *cur;
3509 struct extent_buffer *parent;
3510 struct btrfs_leaf_ref *ref;
3511 u32 blocksize;
3512 int ret;
3513 u32 refs;
3514
3515 WARN_ON(*level < 0);
3516 WARN_ON(*level >= BTRFS_MAX_LEVEL);
3517 ret = drop_snap_lookup_refcount(root, path->nodes[*level]->start,
3518 path->nodes[*level]->len, &refs);
3519 BUG_ON(ret);
3520 if (refs > 1)
3521 goto out;
3522
3523 /*
3524 * walk down to the last node level and free all the leaves
3525 */
3526 while (*level >= 0) {
3527 WARN_ON(*level < 0);
3528 WARN_ON(*level >= BTRFS_MAX_LEVEL);
3529 cur = path->nodes[*level];
3530
3531 if (btrfs_header_level(cur) != *level)
3532 WARN_ON(1);
3533
3534 if (path->slots[*level] >=
3535 btrfs_header_nritems(cur))
3536 break;
3537 if (*level == 0) {
3538 ret = btrfs_drop_leaf_ref(trans, root, cur);
3539 BUG_ON(ret);
3540 break;
3541 }
3542 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
3543 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
3544 blocksize = btrfs_level_size(root, *level - 1);
3545
3546 ret = drop_snap_lookup_refcount(root, bytenr, blocksize, &refs);
3547 BUG_ON(ret);
3548 if (refs != 1) {
3549 parent = path->nodes[*level];
3550 root_owner = btrfs_header_owner(parent);
3551 root_gen = btrfs_header_generation(parent);
3552 path->slots[*level]++;
3553
3554 ret = __btrfs_free_extent(trans, root, bytenr,
3555 blocksize, parent->start,
3556 root_owner, root_gen,
3557 *level - 1, 1);
3558 BUG_ON(ret);
3559
3560 atomic_inc(&root->fs_info->throttle_gen);
3561 wake_up(&root->fs_info->transaction_throttle);
3562 cond_resched();
3563
3564 continue;
3565 }
3566 /*
3567 * at this point, we have a single ref, and since the
3568 * only place referencing this extent is a dead root
3569 * the reference count should never go higher.
3570 * So, we don't need to check it again
3571 */
3572 if (*level == 1) {
3573 ref = btrfs_lookup_leaf_ref(root, bytenr);
3574 if (ref && ref->generation != ptr_gen) {
3575 btrfs_free_leaf_ref(root, ref);
3576 ref = NULL;
3577 }
3578 if (ref) {
3579 ret = cache_drop_leaf_ref(trans, root, ref);
3580 BUG_ON(ret);
3581 btrfs_remove_leaf_ref(root, ref);
3582 btrfs_free_leaf_ref(root, ref);
3583 *level = 0;
3584 break;
3585 }
3586 }
3587 next = btrfs_find_tree_block(root, bytenr, blocksize);
3588 if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
3589 free_extent_buffer(next);
3590
3591 next = read_tree_block(root, bytenr, blocksize,
3592 ptr_gen);
3593 cond_resched();
3594 #if 0
3595 /*
3596 * this is a debugging check and can go away
3597 * the ref should never go all the way down to 1
3598 * at this point
3599 */
3600 ret = lookup_extent_ref(NULL, root, bytenr, blocksize,
3601 &refs);
3602 BUG_ON(ret);
3603 WARN_ON(refs != 1);
3604 #endif
3605 }
3606 WARN_ON(*level <= 0);
3607 if (path->nodes[*level-1])
3608 free_extent_buffer(path->nodes[*level-1]);
3609 path->nodes[*level-1] = next;
3610 *level = btrfs_header_level(next);
3611 path->slots[*level] = 0;
3612 cond_resched();
3613 }
3614 out:
3615 WARN_ON(*level < 0);
3616 WARN_ON(*level >= BTRFS_MAX_LEVEL);
3617
3618 if (path->nodes[*level] == root->node) {
3619 parent = path->nodes[*level];
3620 bytenr = path->nodes[*level]->start;
3621 } else {
3622 parent = path->nodes[*level + 1];
3623 bytenr = btrfs_node_blockptr(parent, path->slots[*level + 1]);
3624 }
3625
3626 blocksize = btrfs_level_size(root, *level);
3627 root_owner = btrfs_header_owner(parent);
3628 root_gen = btrfs_header_generation(parent);
3629
3630 ret = __btrfs_free_extent(trans, root, bytenr, blocksize,
3631 parent->start, root_owner, root_gen,
3632 *level, 1);
3633 free_extent_buffer(path->nodes[*level]);
3634 path->nodes[*level] = NULL;
3635 *level += 1;
3636 BUG_ON(ret);
3637
3638 cond_resched();
3639 return 0;
3640 }
3641
3642 /*
3643 * helper function for drop_subtree, this function is similar to
3644 * walk_down_tree. The main difference is that it checks reference
3645 * counts while tree blocks are locked.
3646 */
3647 static noinline int walk_down_subtree(struct btrfs_trans_handle *trans,
3648 struct btrfs_root *root,
3649 struct btrfs_path *path, int *level)
3650 {
3651 struct extent_buffer *next;
3652 struct extent_buffer *cur;
3653 struct extent_buffer *parent;
3654 u64 bytenr;
3655 u64 ptr_gen;
3656 u32 blocksize;
3657 u32 refs;
3658 int ret;
3659
3660 cur = path->nodes[*level];
3661 ret = btrfs_lookup_extent_ref(trans, root, cur->start, cur->len,
3662 &refs);
3663 BUG_ON(ret);
3664 if (refs > 1)
3665 goto out;
3666
3667 while (*level >= 0) {
3668 cur = path->nodes[*level];
3669 if (*level == 0) {
3670 ret = btrfs_drop_leaf_ref(trans, root, cur);
3671 BUG_ON(ret);
3672 clean_tree_block(trans, root, cur);
3673 break;
3674 }
3675 if (path->slots[*level] >= btrfs_header_nritems(cur)) {
3676 clean_tree_block(trans, root, cur);
3677 break;
3678 }
3679
3680 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
3681 blocksize = btrfs_level_size(root, *level - 1);
3682 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
3683
3684 next = read_tree_block(root, bytenr, blocksize, ptr_gen);
3685 btrfs_tree_lock(next);
3686
3687 ret = btrfs_lookup_extent_ref(trans, root, bytenr, blocksize,
3688 &refs);
3689 BUG_ON(ret);
3690 if (refs > 1) {
3691 parent = path->nodes[*level];
3692 ret = btrfs_free_extent(trans, root, bytenr,
3693 blocksize, parent->start,
3694 btrfs_header_owner(parent),
3695 btrfs_header_generation(parent),
3696 *level - 1, 1);
3697 BUG_ON(ret);
3698 path->slots[*level]++;
3699 btrfs_tree_unlock(next);
3700 free_extent_buffer(next);
3701 continue;
3702 }
3703
3704 *level = btrfs_header_level(next);
3705 path->nodes[*level] = next;
3706 path->slots[*level] = 0;
3707 path->locks[*level] = 1;
3708 cond_resched();
3709 }
3710 out:
3711 parent = path->nodes[*level + 1];
3712 bytenr = path->nodes[*level]->start;
3713 blocksize = path->nodes[*level]->len;
3714
3715 ret = btrfs_free_extent(trans, root, bytenr, blocksize,
3716 parent->start, btrfs_header_owner(parent),
3717 btrfs_header_generation(parent), *level, 1);
3718 BUG_ON(ret);
3719
3720 if (path->locks[*level]) {
3721 btrfs_tree_unlock(path->nodes[*level]);
3722 path->locks[*level] = 0;
3723 }
3724 free_extent_buffer(path->nodes[*level]);
3725 path->nodes[*level] = NULL;
3726 *level += 1;
3727 cond_resched();
3728 return 0;
3729 }
3730
3731 /*
3732 * helper for dropping snapshots. This walks back up the tree in the path
3733 * to find the first node higher up where we haven't yet gone through
3734 * all the slots
3735 */
3736 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
3737 struct btrfs_root *root,
3738 struct btrfs_path *path,
3739 int *level, int max_level)
3740 {
3741 u64 root_owner;
3742 u64 root_gen;
3743 struct btrfs_root_item *root_item = &root->root_item;
3744 int i;
3745 int slot;
3746 int ret;
3747
3748 for (i = *level; i < max_level && path->nodes[i]; i++) {
3749 slot = path->slots[i];
3750 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
3751 struct extent_buffer *node;
3752 struct btrfs_disk_key disk_key;
3753 node = path->nodes[i];
3754 path->slots[i]++;
3755 *level = i;
3756 WARN_ON(*level == 0);
3757 btrfs_node_key(node, &disk_key, path->slots[i]);
3758 memcpy(&root_item->drop_progress,
3759 &disk_key, sizeof(disk_key));
3760 root_item->drop_level = i;
3761 return 0;
3762 } else {
3763 struct extent_buffer *parent;
3764 if (path->nodes[*level] == root->node)
3765 parent = path->nodes[*level];
3766 else
3767 parent = path->nodes[*level + 1];
3768
3769 root_owner = btrfs_header_owner(parent);
3770 root_gen = btrfs_header_generation(parent);
3771
3772 clean_tree_block(trans, root, path->nodes[*level]);
3773 ret = btrfs_free_extent(trans, root,
3774 path->nodes[*level]->start,
3775 path->nodes[*level]->len,
3776 parent->start, root_owner,
3777 root_gen, *level, 1);
3778 BUG_ON(ret);
3779 if (path->locks[*level]) {
3780 btrfs_tree_unlock(path->nodes[*level]);
3781 path->locks[*level] = 0;
3782 }
3783 free_extent_buffer(path->nodes[*level]);
3784 path->nodes[*level] = NULL;
3785 *level = i + 1;
3786 }
3787 }
3788 return 1;
3789 }
3790
3791 /*
3792 * drop the reference count on the tree rooted at 'snap'. This traverses
3793 * the tree freeing any blocks that have a ref count of zero after being
3794 * decremented.
3795 */
3796 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
3797 *root)
3798 {
3799 int ret = 0;
3800 int wret;
3801 int level;
3802 struct btrfs_path *path;
3803 int i;
3804 int orig_level;
3805 struct btrfs_root_item *root_item = &root->root_item;
3806
3807 WARN_ON(!mutex_is_locked(&root->fs_info->drop_mutex));
3808 path = btrfs_alloc_path();
3809 BUG_ON(!path);
3810
3811 level = btrfs_header_level(root->node);
3812 orig_level = level;
3813 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
3814 path->nodes[level] = root->node;
3815 extent_buffer_get(root->node);
3816 path->slots[level] = 0;
3817 } else {
3818 struct btrfs_key key;
3819 struct btrfs_disk_key found_key;
3820 struct extent_buffer *node;
3821
3822 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
3823 level = root_item->drop_level;
3824 path->lowest_level = level;
3825 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3826 if (wret < 0) {
3827 ret = wret;
3828 goto out;
3829 }
3830 node = path->nodes[level];
3831 btrfs_node_key(node, &found_key, path->slots[level]);
3832 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
3833 sizeof(found_key)));
3834 /*
3835 * unlock our path, this is safe because only this
3836 * function is allowed to delete this snapshot
3837 */
3838 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
3839 if (path->nodes[i] && path->locks[i]) {
3840 path->locks[i] = 0;
3841 btrfs_tree_unlock(path->nodes[i]);
3842 }
3843 }
3844 }
3845 while (1) {
3846 wret = walk_down_tree(trans, root, path, &level);
3847 if (wret > 0)
3848 break;
3849 if (wret < 0)
3850 ret = wret;
3851
3852 wret = walk_up_tree(trans, root, path, &level,
3853 BTRFS_MAX_LEVEL);
3854 if (wret > 0)
3855 break;
3856 if (wret < 0)
3857 ret = wret;
3858 if (trans->transaction->in_commit) {
3859 ret = -EAGAIN;
3860 break;
3861 }
3862 atomic_inc(&root->fs_info->throttle_gen);
3863 wake_up(&root->fs_info->transaction_throttle);
3864 }
3865 for (i = 0; i <= orig_level; i++) {
3866 if (path->nodes[i]) {
3867 free_extent_buffer(path->nodes[i]);
3868 path->nodes[i] = NULL;
3869 }
3870 }
3871 out:
3872 btrfs_free_path(path);
3873 return ret;
3874 }
3875
3876 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
3877 struct btrfs_root *root,
3878 struct extent_buffer *node,
3879 struct extent_buffer *parent)
3880 {
3881 struct btrfs_path *path;
3882 int level;
3883 int parent_level;
3884 int ret = 0;
3885 int wret;
3886
3887 path = btrfs_alloc_path();
3888 BUG_ON(!path);
3889
3890 BUG_ON(!btrfs_tree_locked(parent));
3891 parent_level = btrfs_header_level(parent);
3892 extent_buffer_get(parent);
3893 path->nodes[parent_level] = parent;
3894 path->slots[parent_level] = btrfs_header_nritems(parent);
3895
3896 BUG_ON(!btrfs_tree_locked(node));
3897 level = btrfs_header_level(node);
3898 extent_buffer_get(node);
3899 path->nodes[level] = node;
3900 path->slots[level] = 0;
3901
3902 while (1) {
3903 wret = walk_down_subtree(trans, root, path, &level);
3904 if (wret < 0)
3905 ret = wret;
3906 if (wret != 0)
3907 break;
3908
3909 wret = walk_up_tree(trans, root, path, &level, parent_level);
3910 if (wret < 0)
3911 ret = wret;
3912 if (wret != 0)
3913 break;
3914 }
3915
3916 btrfs_free_path(path);
3917 return ret;
3918 }
3919
3920 static unsigned long calc_ra(unsigned long start, unsigned long last,
3921 unsigned long nr)
3922 {
3923 return min(last, start + nr - 1);
3924 }
3925
3926 static noinline int relocate_inode_pages(struct inode *inode, u64 start,
3927 u64 len)
3928 {
3929 u64 page_start;
3930 u64 page_end;
3931 unsigned long first_index;
3932 unsigned long last_index;
3933 unsigned long i;
3934 struct page *page;
3935 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3936 struct file_ra_state *ra;
3937 struct btrfs_ordered_extent *ordered;
3938 unsigned int total_read = 0;
3939 unsigned int total_dirty = 0;
3940 int ret = 0;
3941
3942 ra = kzalloc(sizeof(*ra), GFP_NOFS);
3943
3944 mutex_lock(&inode->i_mutex);
3945 first_index = start >> PAGE_CACHE_SHIFT;
3946 last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
3947
3948 /* make sure the dirty trick played by the caller work */
3949 ret = invalidate_inode_pages2_range(inode->i_mapping,
3950 first_index, last_index);
3951 if (ret)
3952 goto out_unlock;
3953
3954 file_ra_state_init(ra, inode->i_mapping);
3955
3956 for (i = first_index ; i <= last_index; i++) {
3957 if (total_read % ra->ra_pages == 0) {
3958 btrfs_force_ra(inode->i_mapping, ra, NULL, i,
3959 calc_ra(i, last_index, ra->ra_pages));
3960 }
3961 total_read++;
3962 again:
3963 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
3964 BUG_ON(1);
3965 page = grab_cache_page(inode->i_mapping, i);
3966 if (!page) {
3967 ret = -ENOMEM;
3968 goto out_unlock;
3969 }
3970 if (!PageUptodate(page)) {
3971 btrfs_readpage(NULL, page);
3972 lock_page(page);
3973 if (!PageUptodate(page)) {
3974 unlock_page(page);
3975 page_cache_release(page);
3976 ret = -EIO;
3977 goto out_unlock;
3978 }
3979 }
3980 wait_on_page_writeback(page);
3981
3982 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
3983 page_end = page_start + PAGE_CACHE_SIZE - 1;
3984 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
3985
3986 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3987 if (ordered) {
3988 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3989 unlock_page(page);
3990 page_cache_release(page);
3991 btrfs_start_ordered_extent(inode, ordered, 1);
3992 btrfs_put_ordered_extent(ordered);
3993 goto again;
3994 }
3995 set_page_extent_mapped(page);
3996
3997 if (i == first_index)
3998 set_extent_bits(io_tree, page_start, page_end,
3999 EXTENT_BOUNDARY, GFP_NOFS);
4000 btrfs_set_extent_delalloc(inode, page_start, page_end);
4001
4002 set_page_dirty(page);
4003 total_dirty++;
4004
4005 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4006 unlock_page(page);
4007 page_cache_release(page);
4008 }
4009
4010 out_unlock:
4011 kfree(ra);
4012 mutex_unlock(&inode->i_mutex);
4013 balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
4014 return ret;
4015 }
4016
4017 static noinline int relocate_data_extent(struct inode *reloc_inode,
4018 struct btrfs_key *extent_key,
4019 u64 offset)
4020 {
4021 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
4022 struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
4023 struct extent_map *em;
4024 u64 start = extent_key->objectid - offset;
4025 u64 end = start + extent_key->offset - 1;
4026
4027 em = alloc_extent_map(GFP_NOFS);
4028 BUG_ON(!em || IS_ERR(em));
4029
4030 em->start = start;
4031 em->len = extent_key->offset;
4032 em->block_len = extent_key->offset;
4033 em->block_start = extent_key->objectid;
4034 em->bdev = root->fs_info->fs_devices->latest_bdev;
4035 set_bit(EXTENT_FLAG_PINNED, &em->flags);
4036
4037 /* setup extent map to cheat btrfs_readpage */
4038 lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
4039 while (1) {
4040 int ret;
4041 spin_lock(&em_tree->lock);
4042 ret = add_extent_mapping(em_tree, em);
4043 spin_unlock(&em_tree->lock);
4044 if (ret != -EEXIST) {
4045 free_extent_map(em);
4046 break;
4047 }
4048 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
4049 }
4050 unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
4051
4052 return relocate_inode_pages(reloc_inode, start, extent_key->offset);
4053 }
4054
4055 struct btrfs_ref_path {
4056 u64 extent_start;
4057 u64 nodes[BTRFS_MAX_LEVEL];
4058 u64 root_objectid;
4059 u64 root_generation;
4060 u64 owner_objectid;
4061 u32 num_refs;
4062 int lowest_level;
4063 int current_level;
4064 int shared_level;
4065
4066 struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
4067 u64 new_nodes[BTRFS_MAX_LEVEL];
4068 };
4069
4070 struct disk_extent {
4071 u64 ram_bytes;
4072 u64 disk_bytenr;
4073 u64 disk_num_bytes;
4074 u64 offset;
4075 u64 num_bytes;
4076 u8 compression;
4077 u8 encryption;
4078 u16 other_encoding;
4079 };
4080
4081 static int is_cowonly_root(u64 root_objectid)
4082 {
4083 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
4084 root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
4085 root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
4086 root_objectid == BTRFS_DEV_TREE_OBJECTID ||
4087 root_objectid == BTRFS_TREE_LOG_OBJECTID ||
4088 root_objectid == BTRFS_CSUM_TREE_OBJECTID)
4089 return 1;
4090 return 0;
4091 }
4092
4093 static noinline int __next_ref_path(struct btrfs_trans_handle *trans,
4094 struct btrfs_root *extent_root,
4095 struct btrfs_ref_path *ref_path,
4096 int first_time)
4097 {
4098 struct extent_buffer *leaf;
4099 struct btrfs_path *path;
4100 struct btrfs_extent_ref *ref;
4101 struct btrfs_key key;
4102 struct btrfs_key found_key;
4103 u64 bytenr;
4104 u32 nritems;
4105 int level;
4106 int ret = 1;
4107
4108 path = btrfs_alloc_path();
4109 if (!path)
4110 return -ENOMEM;
4111
4112 if (first_time) {
4113 ref_path->lowest_level = -1;
4114 ref_path->current_level = -1;
4115 ref_path->shared_level = -1;
4116 goto walk_up;
4117 }
4118 walk_down:
4119 level = ref_path->current_level - 1;
4120 while (level >= -1) {
4121 u64 parent;
4122 if (level < ref_path->lowest_level)
4123 break;
4124
4125 if (level >= 0)
4126 bytenr = ref_path->nodes[level];
4127 else
4128 bytenr = ref_path->extent_start;
4129 BUG_ON(bytenr == 0);
4130
4131 parent = ref_path->nodes[level + 1];
4132 ref_path->nodes[level + 1] = 0;
4133 ref_path->current_level = level;
4134 BUG_ON(parent == 0);
4135
4136 key.objectid = bytenr;
4137 key.offset = parent + 1;
4138 key.type = BTRFS_EXTENT_REF_KEY;
4139
4140 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
4141 if (ret < 0)
4142 goto out;
4143 BUG_ON(ret == 0);
4144
4145 leaf = path->nodes[0];
4146 nritems = btrfs_header_nritems(leaf);
4147 if (path->slots[0] >= nritems) {
4148 ret = btrfs_next_leaf(extent_root, path);
4149 if (ret < 0)
4150 goto out;
4151 if (ret > 0)
4152 goto next;
4153 leaf = path->nodes[0];
4154 }
4155
4156 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4157 if (found_key.objectid == bytenr &&
4158 found_key.type == BTRFS_EXTENT_REF_KEY) {
4159 if (level < ref_path->shared_level)
4160 ref_path->shared_level = level;
4161 goto found;
4162 }
4163 next:
4164 level--;
4165 btrfs_release_path(extent_root, path);
4166 cond_resched();
4167 }
4168 /* reached lowest level */
4169 ret = 1;
4170 goto out;
4171 walk_up:
4172 level = ref_path->current_level;
4173 while (level < BTRFS_MAX_LEVEL - 1) {
4174 u64 ref_objectid;
4175
4176 if (level >= 0)
4177 bytenr = ref_path->nodes[level];
4178 else
4179 bytenr = ref_path->extent_start;
4180
4181 BUG_ON(bytenr == 0);
4182
4183 key.objectid = bytenr;
4184 key.offset = 0;
4185 key.type = BTRFS_EXTENT_REF_KEY;
4186
4187 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
4188 if (ret < 0)
4189 goto out;
4190
4191 leaf = path->nodes[0];
4192 nritems = btrfs_header_nritems(leaf);
4193 if (path->slots[0] >= nritems) {
4194 ret = btrfs_next_leaf(extent_root, path);
4195 if (ret < 0)
4196 goto out;
4197 if (ret > 0) {
4198 /* the extent was freed by someone */
4199 if (ref_path->lowest_level == level)
4200 goto out;
4201 btrfs_release_path(extent_root, path);
4202 goto walk_down;
4203 }
4204 leaf = path->nodes[0];
4205 }
4206
4207 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4208 if (found_key.objectid != bytenr ||
4209 found_key.type != BTRFS_EXTENT_REF_KEY) {
4210 /* the extent was freed by someone */
4211 if (ref_path->lowest_level == level) {
4212 ret = 1;
4213 goto out;
4214 }
4215 btrfs_release_path(extent_root, path);
4216 goto walk_down;
4217 }
4218 found:
4219 ref = btrfs_item_ptr(leaf, path->slots[0],
4220 struct btrfs_extent_ref);
4221 ref_objectid = btrfs_ref_objectid(leaf, ref);
4222 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4223 if (first_time) {
4224 level = (int)ref_objectid;
4225 BUG_ON(level >= BTRFS_MAX_LEVEL);
4226 ref_path->lowest_level = level;
4227 ref_path->current_level = level;
4228 ref_path->nodes[level] = bytenr;
4229 } else {
4230 WARN_ON(ref_objectid != level);
4231 }
4232 } else {
4233 WARN_ON(level != -1);
4234 }
4235 first_time = 0;
4236
4237 if (ref_path->lowest_level == level) {
4238 ref_path->owner_objectid = ref_objectid;
4239 ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
4240 }
4241
4242 /*
4243 * the block is tree root or the block isn't in reference
4244 * counted tree.
4245 */
4246 if (found_key.objectid == found_key.offset ||
4247 is_cowonly_root(btrfs_ref_root(leaf, ref))) {
4248 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
4249 ref_path->root_generation =
4250 btrfs_ref_generation(leaf, ref);
4251 if (level < 0) {
4252 /* special reference from the tree log */
4253 ref_path->nodes[0] = found_key.offset;
4254 ref_path->current_level = 0;
4255 }
4256 ret = 0;
4257 goto out;
4258 }
4259
4260 level++;
4261 BUG_ON(ref_path->nodes[level] != 0);
4262 ref_path->nodes[level] = found_key.offset;
4263 ref_path->current_level = level;
4264
4265 /*
4266 * the reference was created in the running transaction,
4267 * no need to continue walking up.
4268 */
4269 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
4270 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
4271 ref_path->root_generation =
4272 btrfs_ref_generation(leaf, ref);
4273 ret = 0;
4274 goto out;
4275 }
4276
4277 btrfs_release_path(extent_root, path);
4278 cond_resched();
4279 }
4280 /* reached max tree level, but no tree root found. */
4281 BUG();
4282 out:
4283 btrfs_free_path(path);
4284 return ret;
4285 }
4286
4287 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
4288 struct btrfs_root *extent_root,
4289 struct btrfs_ref_path *ref_path,
4290 u64 extent_start)
4291 {
4292 memset(ref_path, 0, sizeof(*ref_path));
4293 ref_path->extent_start = extent_start;
4294
4295 return __next_ref_path(trans, extent_root, ref_path, 1);
4296 }
4297
4298 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
4299 struct btrfs_root *extent_root,
4300 struct btrfs_ref_path *ref_path)
4301 {
4302 return __next_ref_path(trans, extent_root, ref_path, 0);
4303 }
4304
4305 static noinline int get_new_locations(struct inode *reloc_inode,
4306 struct btrfs_key *extent_key,
4307 u64 offset, int no_fragment,
4308 struct disk_extent **extents,
4309 int *nr_extents)
4310 {
4311 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
4312 struct btrfs_path *path;
4313 struct btrfs_file_extent_item *fi;
4314 struct extent_buffer *leaf;
4315 struct disk_extent *exts = *extents;
4316 struct btrfs_key found_key;
4317 u64 cur_pos;
4318 u64 last_byte;
4319 u32 nritems;
4320 int nr = 0;
4321 int max = *nr_extents;
4322 int ret;
4323
4324 WARN_ON(!no_fragment && *extents);
4325 if (!exts) {
4326 max = 1;
4327 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
4328 if (!exts)
4329 return -ENOMEM;
4330 }
4331
4332 path = btrfs_alloc_path();
4333 BUG_ON(!path);
4334
4335 cur_pos = extent_key->objectid - offset;
4336 last_byte = extent_key->objectid + extent_key->offset;
4337 ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
4338 cur_pos, 0);
4339 if (ret < 0)
4340 goto out;
4341 if (ret > 0) {
4342 ret = -ENOENT;
4343 goto out;
4344 }
4345
4346 while (1) {
4347 leaf = path->nodes[0];
4348 nritems = btrfs_header_nritems(leaf);
4349 if (path->slots[0] >= nritems) {
4350 ret = btrfs_next_leaf(root, path);
4351 if (ret < 0)
4352 goto out;
4353 if (ret > 0)
4354 break;
4355 leaf = path->nodes[0];
4356 }
4357
4358 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4359 if (found_key.offset != cur_pos ||
4360 found_key.type != BTRFS_EXTENT_DATA_KEY ||
4361 found_key.objectid != reloc_inode->i_ino)
4362 break;
4363
4364 fi = btrfs_item_ptr(leaf, path->slots[0],
4365 struct btrfs_file_extent_item);
4366 if (btrfs_file_extent_type(leaf, fi) !=
4367 BTRFS_FILE_EXTENT_REG ||
4368 btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
4369 break;
4370
4371 if (nr == max) {
4372 struct disk_extent *old = exts;
4373 max *= 2;
4374 exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
4375 memcpy(exts, old, sizeof(*exts) * nr);
4376 if (old != *extents)
4377 kfree(old);
4378 }
4379
4380 exts[nr].disk_bytenr =
4381 btrfs_file_extent_disk_bytenr(leaf, fi);
4382 exts[nr].disk_num_bytes =
4383 btrfs_file_extent_disk_num_bytes(leaf, fi);
4384 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
4385 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4386 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
4387 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
4388 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
4389 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
4390 fi);
4391 BUG_ON(exts[nr].offset > 0);
4392 BUG_ON(exts[nr].compression || exts[nr].encryption);
4393 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
4394
4395 cur_pos += exts[nr].num_bytes;
4396 nr++;
4397
4398 if (cur_pos + offset >= last_byte)
4399 break;
4400
4401 if (no_fragment) {
4402 ret = 1;
4403 goto out;
4404 }
4405 path->slots[0]++;
4406 }
4407
4408 BUG_ON(cur_pos + offset > last_byte);
4409 if (cur_pos + offset < last_byte) {
4410 ret = -ENOENT;
4411 goto out;
4412 }
4413 ret = 0;
4414 out:
4415 btrfs_free_path(path);
4416 if (ret) {
4417 if (exts != *extents)
4418 kfree(exts);
4419 } else {
4420 *extents = exts;
4421 *nr_extents = nr;
4422 }
4423 return ret;
4424 }
4425
4426 static noinline int replace_one_extent(struct btrfs_trans_handle *trans,
4427 struct btrfs_root *root,
4428 struct btrfs_path *path,
4429 struct btrfs_key *extent_key,
4430 struct btrfs_key *leaf_key,
4431 struct btrfs_ref_path *ref_path,
4432 struct disk_extent *new_extents,
4433 int nr_extents)
4434 {
4435 struct extent_buffer *leaf;
4436 struct btrfs_file_extent_item *fi;
4437 struct inode *inode = NULL;
4438 struct btrfs_key key;
4439 u64 lock_start = 0;
4440 u64 lock_end = 0;
4441 u64 num_bytes;
4442 u64 ext_offset;
4443 u64 search_end = (u64)-1;
4444 u32 nritems;
4445 int nr_scaned = 0;
4446 int extent_locked = 0;
4447 int extent_type;
4448 int ret;
4449
4450 memcpy(&key, leaf_key, sizeof(key));
4451 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
4452 if (key.objectid < ref_path->owner_objectid ||
4453 (key.objectid == ref_path->owner_objectid &&
4454 key.type < BTRFS_EXTENT_DATA_KEY)) {
4455 key.objectid = ref_path->owner_objectid;
4456 key.type = BTRFS_EXTENT_DATA_KEY;
4457 key.offset = 0;
4458 }
4459 }
4460
4461 while (1) {
4462 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
4463 if (ret < 0)
4464 goto out;
4465
4466 leaf = path->nodes[0];
4467 nritems = btrfs_header_nritems(leaf);
4468 next:
4469 if (extent_locked && ret > 0) {
4470 /*
4471 * the file extent item was modified by someone
4472 * before the extent got locked.
4473 */
4474 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4475 lock_end, GFP_NOFS);
4476 extent_locked = 0;
4477 }
4478
4479 if (path->slots[0] >= nritems) {
4480 if (++nr_scaned > 2)
4481 break;
4482
4483 BUG_ON(extent_locked);
4484 ret = btrfs_next_leaf(root, path);
4485 if (ret < 0)
4486 goto out;
4487 if (ret > 0)
4488 break;
4489 leaf = path->nodes[0];
4490 nritems = btrfs_header_nritems(leaf);
4491 }
4492
4493 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4494
4495 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
4496 if ((key.objectid > ref_path->owner_objectid) ||
4497 (key.objectid == ref_path->owner_objectid &&
4498 key.type > BTRFS_EXTENT_DATA_KEY) ||
4499 key.offset >= search_end)
4500 break;
4501 }
4502
4503 if (inode && key.objectid != inode->i_ino) {
4504 BUG_ON(extent_locked);
4505 btrfs_release_path(root, path);
4506 mutex_unlock(&inode->i_mutex);
4507 iput(inode);
4508 inode = NULL;
4509 continue;
4510 }
4511
4512 if (key.type != BTRFS_EXTENT_DATA_KEY) {
4513 path->slots[0]++;
4514 ret = 1;
4515 goto next;
4516 }
4517 fi = btrfs_item_ptr(leaf, path->slots[0],
4518 struct btrfs_file_extent_item);
4519 extent_type = btrfs_file_extent_type(leaf, fi);
4520 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
4521 extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
4522 (btrfs_file_extent_disk_bytenr(leaf, fi) !=
4523 extent_key->objectid)) {
4524 path->slots[0]++;
4525 ret = 1;
4526 goto next;
4527 }
4528
4529 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4530 ext_offset = btrfs_file_extent_offset(leaf, fi);
4531
4532 if (search_end == (u64)-1) {
4533 search_end = key.offset - ext_offset +
4534 btrfs_file_extent_ram_bytes(leaf, fi);
4535 }
4536
4537 if (!extent_locked) {
4538 lock_start = key.offset;
4539 lock_end = lock_start + num_bytes - 1;
4540 } else {
4541 if (lock_start > key.offset ||
4542 lock_end + 1 < key.offset + num_bytes) {
4543 unlock_extent(&BTRFS_I(inode)->io_tree,
4544 lock_start, lock_end, GFP_NOFS);
4545 extent_locked = 0;
4546 }
4547 }
4548
4549 if (!inode) {
4550 btrfs_release_path(root, path);
4551
4552 inode = btrfs_iget_locked(root->fs_info->sb,
4553 key.objectid, root);
4554 if (inode->i_state & I_NEW) {
4555 BTRFS_I(inode)->root = root;
4556 BTRFS_I(inode)->location.objectid =
4557 key.objectid;
4558 BTRFS_I(inode)->location.type =
4559 BTRFS_INODE_ITEM_KEY;
4560 BTRFS_I(inode)->location.offset = 0;
4561 btrfs_read_locked_inode(inode);
4562 unlock_new_inode(inode);
4563 }
4564 /*
4565 * some code call btrfs_commit_transaction while
4566 * holding the i_mutex, so we can't use mutex_lock
4567 * here.
4568 */
4569 if (is_bad_inode(inode) ||
4570 !mutex_trylock(&inode->i_mutex)) {
4571 iput(inode);
4572 inode = NULL;
4573 key.offset = (u64)-1;
4574 goto skip;
4575 }
4576 }
4577
4578 if (!extent_locked) {
4579 struct btrfs_ordered_extent *ordered;
4580
4581 btrfs_release_path(root, path);
4582
4583 lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4584 lock_end, GFP_NOFS);
4585 ordered = btrfs_lookup_first_ordered_extent(inode,
4586 lock_end);
4587 if (ordered &&
4588 ordered->file_offset <= lock_end &&
4589 ordered->file_offset + ordered->len > lock_start) {
4590 unlock_extent(&BTRFS_I(inode)->io_tree,
4591 lock_start, lock_end, GFP_NOFS);
4592 btrfs_start_ordered_extent(inode, ordered, 1);
4593 btrfs_put_ordered_extent(ordered);
4594 key.offset += num_bytes;
4595 goto skip;
4596 }
4597 if (ordered)
4598 btrfs_put_ordered_extent(ordered);
4599
4600 extent_locked = 1;
4601 continue;
4602 }
4603
4604 if (nr_extents == 1) {
4605 /* update extent pointer in place */
4606 btrfs_set_file_extent_disk_bytenr(leaf, fi,
4607 new_extents[0].disk_bytenr);
4608 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4609 new_extents[0].disk_num_bytes);
4610 btrfs_mark_buffer_dirty(leaf);
4611
4612 btrfs_drop_extent_cache(inode, key.offset,
4613 key.offset + num_bytes - 1, 0);
4614
4615 ret = btrfs_inc_extent_ref(trans, root,
4616 new_extents[0].disk_bytenr,
4617 new_extents[0].disk_num_bytes,
4618 leaf->start,
4619 root->root_key.objectid,
4620 trans->transid,
4621 key.objectid);
4622 BUG_ON(ret);
4623
4624 ret = btrfs_free_extent(trans, root,
4625 extent_key->objectid,
4626 extent_key->offset,
4627 leaf->start,
4628 btrfs_header_owner(leaf),
4629 btrfs_header_generation(leaf),
4630 key.objectid, 0);
4631 BUG_ON(ret);
4632
4633 btrfs_release_path(root, path);
4634 key.offset += num_bytes;
4635 } else {
4636 BUG_ON(1);
4637 #if 0
4638 u64 alloc_hint;
4639 u64 extent_len;
4640 int i;
4641 /*
4642 * drop old extent pointer at first, then insert the
4643 * new pointers one bye one
4644 */
4645 btrfs_release_path(root, path);
4646 ret = btrfs_drop_extents(trans, root, inode, key.offset,
4647 key.offset + num_bytes,
4648 key.offset, &alloc_hint);
4649 BUG_ON(ret);
4650
4651 for (i = 0; i < nr_extents; i++) {
4652 if (ext_offset >= new_extents[i].num_bytes) {
4653 ext_offset -= new_extents[i].num_bytes;
4654 continue;
4655 }
4656 extent_len = min(new_extents[i].num_bytes -
4657 ext_offset, num_bytes);
4658
4659 ret = btrfs_insert_empty_item(trans, root,
4660 path, &key,
4661 sizeof(*fi));
4662 BUG_ON(ret);
4663
4664 leaf = path->nodes[0];
4665 fi = btrfs_item_ptr(leaf, path->slots[0],
4666 struct btrfs_file_extent_item);
4667 btrfs_set_file_extent_generation(leaf, fi,
4668 trans->transid);
4669 btrfs_set_file_extent_type(leaf, fi,
4670 BTRFS_FILE_EXTENT_REG);
4671 btrfs_set_file_extent_disk_bytenr(leaf, fi,
4672 new_extents[i].disk_bytenr);
4673 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4674 new_extents[i].disk_num_bytes);
4675 btrfs_set_file_extent_ram_bytes(leaf, fi,
4676 new_extents[i].ram_bytes);
4677
4678 btrfs_set_file_extent_compression(leaf, fi,
4679 new_extents[i].compression);
4680 btrfs_set_file_extent_encryption(leaf, fi,
4681 new_extents[i].encryption);
4682 btrfs_set_file_extent_other_encoding(leaf, fi,
4683 new_extents[i].other_encoding);
4684
4685 btrfs_set_file_extent_num_bytes(leaf, fi,
4686 extent_len);
4687 ext_offset += new_extents[i].offset;
4688 btrfs_set_file_extent_offset(leaf, fi,
4689 ext_offset);
4690 btrfs_mark_buffer_dirty(leaf);
4691
4692 btrfs_drop_extent_cache(inode, key.offset,
4693 key.offset + extent_len - 1, 0);
4694
4695 ret = btrfs_inc_extent_ref(trans, root,
4696 new_extents[i].disk_bytenr,
4697 new_extents[i].disk_num_bytes,
4698 leaf->start,
4699 root->root_key.objectid,
4700 trans->transid, key.objectid);
4701 BUG_ON(ret);
4702 btrfs_release_path(root, path);
4703
4704 inode_add_bytes(inode, extent_len);
4705
4706 ext_offset = 0;
4707 num_bytes -= extent_len;
4708 key.offset += extent_len;
4709
4710 if (num_bytes == 0)
4711 break;
4712 }
4713 BUG_ON(i >= nr_extents);
4714 #endif
4715 }
4716
4717 if (extent_locked) {
4718 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4719 lock_end, GFP_NOFS);
4720 extent_locked = 0;
4721 }
4722 skip:
4723 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
4724 key.offset >= search_end)
4725 break;
4726
4727 cond_resched();
4728 }
4729 ret = 0;
4730 out:
4731 btrfs_release_path(root, path);
4732 if (inode) {
4733 mutex_unlock(&inode->i_mutex);
4734 if (extent_locked) {
4735 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4736 lock_end, GFP_NOFS);
4737 }
4738 iput(inode);
4739 }
4740 return ret;
4741 }
4742
4743 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
4744 struct btrfs_root *root,
4745 struct extent_buffer *buf, u64 orig_start)
4746 {
4747 int level;
4748 int ret;
4749
4750 BUG_ON(btrfs_header_generation(buf) != trans->transid);
4751 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
4752
4753 level = btrfs_header_level(buf);
4754 if (level == 0) {
4755 struct btrfs_leaf_ref *ref;
4756 struct btrfs_leaf_ref *orig_ref;
4757
4758 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
4759 if (!orig_ref)
4760 return -ENOENT;
4761
4762 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
4763 if (!ref) {
4764 btrfs_free_leaf_ref(root, orig_ref);
4765 return -ENOMEM;
4766 }
4767
4768 ref->nritems = orig_ref->nritems;
4769 memcpy(ref->extents, orig_ref->extents,
4770 sizeof(ref->extents[0]) * ref->nritems);
4771
4772 btrfs_free_leaf_ref(root, orig_ref);
4773
4774 ref->root_gen = trans->transid;
4775 ref->bytenr = buf->start;
4776 ref->owner = btrfs_header_owner(buf);
4777 ref->generation = btrfs_header_generation(buf);
4778 ret = btrfs_add_leaf_ref(root, ref, 0);
4779 WARN_ON(ret);
4780 btrfs_free_leaf_ref(root, ref);
4781 }
4782 return 0;
4783 }
4784
4785 static noinline int invalidate_extent_cache(struct btrfs_root *root,
4786 struct extent_buffer *leaf,
4787 struct btrfs_block_group_cache *group,
4788 struct btrfs_root *target_root)
4789 {
4790 struct btrfs_key key;
4791 struct inode *inode = NULL;
4792 struct btrfs_file_extent_item *fi;
4793 u64 num_bytes;
4794 u64 skip_objectid = 0;
4795 u32 nritems;
4796 u32 i;
4797
4798 nritems = btrfs_header_nritems(leaf);
4799 for (i = 0; i < nritems; i++) {
4800 btrfs_item_key_to_cpu(leaf, &key, i);
4801 if (key.objectid == skip_objectid ||
4802 key.type != BTRFS_EXTENT_DATA_KEY)
4803 continue;
4804 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4805 if (btrfs_file_extent_type(leaf, fi) ==
4806 BTRFS_FILE_EXTENT_INLINE)
4807 continue;
4808 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
4809 continue;
4810 if (!inode || inode->i_ino != key.objectid) {
4811 iput(inode);
4812 inode = btrfs_ilookup(target_root->fs_info->sb,
4813 key.objectid, target_root, 1);
4814 }
4815 if (!inode) {
4816 skip_objectid = key.objectid;
4817 continue;
4818 }
4819 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4820
4821 lock_extent(&BTRFS_I(inode)->io_tree, key.offset,
4822 key.offset + num_bytes - 1, GFP_NOFS);
4823 btrfs_drop_extent_cache(inode, key.offset,
4824 key.offset + num_bytes - 1, 1);
4825 unlock_extent(&BTRFS_I(inode)->io_tree, key.offset,
4826 key.offset + num_bytes - 1, GFP_NOFS);
4827 cond_resched();
4828 }
4829 iput(inode);
4830 return 0;
4831 }
4832
4833 static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans,
4834 struct btrfs_root *root,
4835 struct extent_buffer *leaf,
4836 struct btrfs_block_group_cache *group,
4837 struct inode *reloc_inode)
4838 {
4839 struct btrfs_key key;
4840 struct btrfs_key extent_key;
4841 struct btrfs_file_extent_item *fi;
4842 struct btrfs_leaf_ref *ref;
4843 struct disk_extent *new_extent;
4844 u64 bytenr;
4845 u64 num_bytes;
4846 u32 nritems;
4847 u32 i;
4848 int ext_index;
4849 int nr_extent;
4850 int ret;
4851
4852 new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
4853 BUG_ON(!new_extent);
4854
4855 ref = btrfs_lookup_leaf_ref(root, leaf->start);
4856 BUG_ON(!ref);
4857
4858 ext_index = -1;
4859 nritems = btrfs_header_nritems(leaf);
4860 for (i = 0; i < nritems; i++) {
4861 btrfs_item_key_to_cpu(leaf, &key, i);
4862 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
4863 continue;
4864 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4865 if (btrfs_file_extent_type(leaf, fi) ==
4866 BTRFS_FILE_EXTENT_INLINE)
4867 continue;
4868 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
4869 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
4870 if (bytenr == 0)
4871 continue;
4872
4873 ext_index++;
4874 if (bytenr >= group->key.objectid + group->key.offset ||
4875 bytenr + num_bytes <= group->key.objectid)
4876 continue;
4877
4878 extent_key.objectid = bytenr;
4879 extent_key.offset = num_bytes;
4880 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
4881 nr_extent = 1;
4882 ret = get_new_locations(reloc_inode, &extent_key,
4883 group->key.objectid, 1,
4884 &new_extent, &nr_extent);
4885 if (ret > 0)
4886 continue;
4887 BUG_ON(ret < 0);
4888
4889 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
4890 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
4891 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
4892 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
4893
4894 btrfs_set_file_extent_disk_bytenr(leaf, fi,
4895 new_extent->disk_bytenr);
4896 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4897 new_extent->disk_num_bytes);
4898 btrfs_mark_buffer_dirty(leaf);
4899
4900 ret = btrfs_inc_extent_ref(trans, root,
4901 new_extent->disk_bytenr,
4902 new_extent->disk_num_bytes,
4903 leaf->start,
4904 root->root_key.objectid,
4905 trans->transid, key.objectid);
4906 BUG_ON(ret);
4907 ret = btrfs_free_extent(trans, root,
4908 bytenr, num_bytes, leaf->start,
4909 btrfs_header_owner(leaf),
4910 btrfs_header_generation(leaf),
4911 key.objectid, 0);
4912 BUG_ON(ret);
4913 cond_resched();
4914 }
4915 kfree(new_extent);
4916 BUG_ON(ext_index + 1 != ref->nritems);
4917 btrfs_free_leaf_ref(root, ref);
4918 return 0;
4919 }
4920
4921 int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
4922 struct btrfs_root *root)
4923 {
4924 struct btrfs_root *reloc_root;
4925 int ret;
4926
4927 if (root->reloc_root) {
4928 reloc_root = root->reloc_root;
4929 root->reloc_root = NULL;
4930 list_add(&reloc_root->dead_list,
4931 &root->fs_info->dead_reloc_roots);
4932
4933 btrfs_set_root_bytenr(&reloc_root->root_item,
4934 reloc_root->node->start);
4935 btrfs_set_root_level(&root->root_item,
4936 btrfs_header_level(reloc_root->node));
4937 memset(&reloc_root->root_item.drop_progress, 0,
4938 sizeof(struct btrfs_disk_key));
4939 reloc_root->root_item.drop_level = 0;
4940
4941 ret = btrfs_update_root(trans, root->fs_info->tree_root,
4942 &reloc_root->root_key,
4943 &reloc_root->root_item);
4944 BUG_ON(ret);
4945 }
4946 return 0;
4947 }
4948
4949 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
4950 {
4951 struct btrfs_trans_handle *trans;
4952 struct btrfs_root *reloc_root;
4953 struct btrfs_root *prev_root = NULL;
4954 struct list_head dead_roots;
4955 int ret;
4956 unsigned long nr;
4957
4958 INIT_LIST_HEAD(&dead_roots);
4959 list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
4960
4961 while (!list_empty(&dead_roots)) {
4962 reloc_root = list_entry(dead_roots.prev,
4963 struct btrfs_root, dead_list);
4964 list_del_init(&reloc_root->dead_list);
4965
4966 BUG_ON(reloc_root->commit_root != NULL);
4967 while (1) {
4968 trans = btrfs_join_transaction(root, 1);
4969 BUG_ON(!trans);
4970
4971 mutex_lock(&root->fs_info->drop_mutex);
4972 ret = btrfs_drop_snapshot(trans, reloc_root);
4973 if (ret != -EAGAIN)
4974 break;
4975 mutex_unlock(&root->fs_info->drop_mutex);
4976
4977 nr = trans->blocks_used;
4978 ret = btrfs_end_transaction(trans, root);
4979 BUG_ON(ret);
4980 btrfs_btree_balance_dirty(root, nr);
4981 }
4982
4983 free_extent_buffer(reloc_root->node);
4984
4985 ret = btrfs_del_root(trans, root->fs_info->tree_root,
4986 &reloc_root->root_key);
4987 BUG_ON(ret);
4988 mutex_unlock(&root->fs_info->drop_mutex);
4989
4990 nr = trans->blocks_used;
4991 ret = btrfs_end_transaction(trans, root);
4992 BUG_ON(ret);
4993 btrfs_btree_balance_dirty(root, nr);
4994
4995 kfree(prev_root);
4996 prev_root = reloc_root;
4997 }
4998 if (prev_root) {
4999 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
5000 kfree(prev_root);
5001 }
5002 return 0;
5003 }
5004
5005 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
5006 {
5007 list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
5008 return 0;
5009 }
5010
5011 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
5012 {
5013 struct btrfs_root *reloc_root;
5014 struct btrfs_trans_handle *trans;
5015 struct btrfs_key location;
5016 int found;
5017 int ret;
5018
5019 mutex_lock(&root->fs_info->tree_reloc_mutex);
5020 ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
5021 BUG_ON(ret);
5022 found = !list_empty(&root->fs_info->dead_reloc_roots);
5023 mutex_unlock(&root->fs_info->tree_reloc_mutex);
5024
5025 if (found) {
5026 trans = btrfs_start_transaction(root, 1);
5027 BUG_ON(!trans);
5028 ret = btrfs_commit_transaction(trans, root);
5029 BUG_ON(ret);
5030 }
5031
5032 location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
5033 location.offset = (u64)-1;
5034 location.type = BTRFS_ROOT_ITEM_KEY;
5035
5036 reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
5037 BUG_ON(!reloc_root);
5038 btrfs_orphan_cleanup(reloc_root);
5039 return 0;
5040 }
5041
5042 static noinline int init_reloc_tree(struct btrfs_trans_handle *trans,
5043 struct btrfs_root *root)
5044 {
5045 struct btrfs_root *reloc_root;
5046 struct extent_buffer *eb;
5047 struct btrfs_root_item *root_item;
5048 struct btrfs_key root_key;
5049 int ret;
5050
5051 BUG_ON(!root->ref_cows);
5052 if (root->reloc_root)
5053 return 0;
5054
5055 root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
5056 BUG_ON(!root_item);
5057
5058 ret = btrfs_copy_root(trans, root, root->commit_root,
5059 &eb, BTRFS_TREE_RELOC_OBJECTID);
5060 BUG_ON(ret);
5061
5062 root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
5063 root_key.offset = root->root_key.objectid;
5064 root_key.type = BTRFS_ROOT_ITEM_KEY;
5065
5066 memcpy(root_item, &root->root_item, sizeof(root_item));
5067 btrfs_set_root_refs(root_item, 0);
5068 btrfs_set_root_bytenr(root_item, eb->start);
5069 btrfs_set_root_level(root_item, btrfs_header_level(eb));
5070 btrfs_set_root_generation(root_item, trans->transid);
5071
5072 btrfs_tree_unlock(eb);
5073 free_extent_buffer(eb);
5074
5075 ret = btrfs_insert_root(trans, root->fs_info->tree_root,
5076 &root_key, root_item);
5077 BUG_ON(ret);
5078 kfree(root_item);
5079
5080 reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
5081 &root_key);
5082 BUG_ON(!reloc_root);
5083 reloc_root->last_trans = trans->transid;
5084 reloc_root->commit_root = NULL;
5085 reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
5086
5087 root->reloc_root = reloc_root;
5088 return 0;
5089 }
5090
5091 /*
5092 * Core function of space balance.
5093 *
5094 * The idea is using reloc trees to relocate tree blocks in reference
5095 * counted roots. There is one reloc tree for each subvol, and all
5096 * reloc trees share same root key objectid. Reloc trees are snapshots
5097 * of the latest committed roots of subvols (root->commit_root).
5098 *
5099 * To relocate a tree block referenced by a subvol, there are two steps.
5100 * COW the block through subvol's reloc tree, then update block pointer
5101 * in the subvol to point to the new block. Since all reloc trees share
5102 * same root key objectid, doing special handing for tree blocks owned
5103 * by them is easy. Once a tree block has been COWed in one reloc tree,
5104 * we can use the resulting new block directly when the same block is
5105 * required to COW again through other reloc trees. By this way, relocated
5106 * tree blocks are shared between reloc trees, so they are also shared
5107 * between subvols.
5108 */
5109 static noinline int relocate_one_path(struct btrfs_trans_handle *trans,
5110 struct btrfs_root *root,
5111 struct btrfs_path *path,
5112 struct btrfs_key *first_key,
5113 struct btrfs_ref_path *ref_path,
5114 struct btrfs_block_group_cache *group,
5115 struct inode *reloc_inode)
5116 {
5117 struct btrfs_root *reloc_root;
5118 struct extent_buffer *eb = NULL;
5119 struct btrfs_key *keys;
5120 u64 *nodes;
5121 int level;
5122 int shared_level;
5123 int lowest_level = 0;
5124 int ret;
5125
5126 if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
5127 lowest_level = ref_path->owner_objectid;
5128
5129 if (!root->ref_cows) {
5130 path->lowest_level = lowest_level;
5131 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
5132 BUG_ON(ret < 0);
5133 path->lowest_level = 0;
5134 btrfs_release_path(root, path);
5135 return 0;
5136 }
5137
5138 mutex_lock(&root->fs_info->tree_reloc_mutex);
5139 ret = init_reloc_tree(trans, root);
5140 BUG_ON(ret);
5141 reloc_root = root->reloc_root;
5142
5143 shared_level = ref_path->shared_level;
5144 ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
5145
5146 keys = ref_path->node_keys;
5147 nodes = ref_path->new_nodes;
5148 memset(&keys[shared_level + 1], 0,
5149 sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
5150 memset(&nodes[shared_level + 1], 0,
5151 sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
5152
5153 if (nodes[lowest_level] == 0) {
5154 path->lowest_level = lowest_level;
5155 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
5156 0, 1);
5157 BUG_ON(ret);
5158 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
5159 eb = path->nodes[level];
5160 if (!eb || eb == reloc_root->node)
5161 break;
5162 nodes[level] = eb->start;
5163 if (level == 0)
5164 btrfs_item_key_to_cpu(eb, &keys[level], 0);
5165 else
5166 btrfs_node_key_to_cpu(eb, &keys[level], 0);
5167 }
5168 if (nodes[0] &&
5169 ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5170 eb = path->nodes[0];
5171 ret = replace_extents_in_leaf(trans, reloc_root, eb,
5172 group, reloc_inode);
5173 BUG_ON(ret);
5174 }
5175 btrfs_release_path(reloc_root, path);
5176 } else {
5177 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
5178 lowest_level);
5179 BUG_ON(ret);
5180 }
5181
5182 /*
5183 * replace tree blocks in the fs tree with tree blocks in
5184 * the reloc tree.
5185 */
5186 ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
5187 BUG_ON(ret < 0);
5188
5189 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5190 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
5191 0, 0);
5192 BUG_ON(ret);
5193 extent_buffer_get(path->nodes[0]);
5194 eb = path->nodes[0];
5195 btrfs_release_path(reloc_root, path);
5196 ret = invalidate_extent_cache(reloc_root, eb, group, root);
5197 BUG_ON(ret);
5198 free_extent_buffer(eb);
5199 }
5200
5201 mutex_unlock(&root->fs_info->tree_reloc_mutex);
5202 path->lowest_level = 0;
5203 return 0;
5204 }
5205
5206 static noinline int relocate_tree_block(struct btrfs_trans_handle *trans,
5207 struct btrfs_root *root,
5208 struct btrfs_path *path,
5209 struct btrfs_key *first_key,
5210 struct btrfs_ref_path *ref_path)
5211 {
5212 int ret;
5213
5214 ret = relocate_one_path(trans, root, path, first_key,
5215 ref_path, NULL, NULL);
5216 BUG_ON(ret);
5217
5218 if (root == root->fs_info->extent_root)
5219 btrfs_extent_post_op(trans, root);
5220
5221 return 0;
5222 }
5223
5224 static noinline int del_extent_zero(struct btrfs_trans_handle *trans,
5225 struct btrfs_root *extent_root,
5226 struct btrfs_path *path,
5227 struct btrfs_key *extent_key)
5228 {
5229 int ret;
5230
5231 ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
5232 if (ret)
5233 goto out;
5234 ret = btrfs_del_item(trans, extent_root, path);
5235 out:
5236 btrfs_release_path(extent_root, path);
5237 return ret;
5238 }
5239
5240 static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info,
5241 struct btrfs_ref_path *ref_path)
5242 {
5243 struct btrfs_key root_key;
5244
5245 root_key.objectid = ref_path->root_objectid;
5246 root_key.type = BTRFS_ROOT_ITEM_KEY;
5247 if (is_cowonly_root(ref_path->root_objectid))
5248 root_key.offset = 0;
5249 else
5250 root_key.offset = (u64)-1;
5251
5252 return btrfs_read_fs_root_no_name(fs_info, &root_key);
5253 }
5254
5255 static noinline int relocate_one_extent(struct btrfs_root *extent_root,
5256 struct btrfs_path *path,
5257 struct btrfs_key *extent_key,
5258 struct btrfs_block_group_cache *group,
5259 struct inode *reloc_inode, int pass)
5260 {
5261 struct btrfs_trans_handle *trans;
5262 struct btrfs_root *found_root;
5263 struct btrfs_ref_path *ref_path = NULL;
5264 struct disk_extent *new_extents = NULL;
5265 int nr_extents = 0;
5266 int loops;
5267 int ret;
5268 int level;
5269 struct btrfs_key first_key;
5270 u64 prev_block = 0;
5271
5272
5273 trans = btrfs_start_transaction(extent_root, 1);
5274 BUG_ON(!trans);
5275
5276 if (extent_key->objectid == 0) {
5277 ret = del_extent_zero(trans, extent_root, path, extent_key);
5278 goto out;
5279 }
5280
5281 ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
5282 if (!ref_path) {
5283 ret = -ENOMEM;
5284 goto out;
5285 }
5286
5287 for (loops = 0; ; loops++) {
5288 if (loops == 0) {
5289 ret = btrfs_first_ref_path(trans, extent_root, ref_path,
5290 extent_key->objectid);
5291 } else {
5292 ret = btrfs_next_ref_path(trans, extent_root, ref_path);
5293 }
5294 if (ret < 0)
5295 goto out;
5296 if (ret > 0)
5297 break;
5298
5299 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
5300 ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
5301 continue;
5302
5303 found_root = read_ref_root(extent_root->fs_info, ref_path);
5304 BUG_ON(!found_root);
5305 /*
5306 * for reference counted tree, only process reference paths
5307 * rooted at the latest committed root.
5308 */
5309 if (found_root->ref_cows &&
5310 ref_path->root_generation != found_root->root_key.offset)
5311 continue;
5312
5313 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5314 if (pass == 0) {
5315 /*
5316 * copy data extents to new locations
5317 */
5318 u64 group_start = group->key.objectid;
5319 ret = relocate_data_extent(reloc_inode,
5320 extent_key,
5321 group_start);
5322 if (ret < 0)
5323 goto out;
5324 break;
5325 }
5326 level = 0;
5327 } else {
5328 level = ref_path->owner_objectid;
5329 }
5330
5331 if (prev_block != ref_path->nodes[level]) {
5332 struct extent_buffer *eb;
5333 u64 block_start = ref_path->nodes[level];
5334 u64 block_size = btrfs_level_size(found_root, level);
5335
5336 eb = read_tree_block(found_root, block_start,
5337 block_size, 0);
5338 btrfs_tree_lock(eb);
5339 BUG_ON(level != btrfs_header_level(eb));
5340
5341 if (level == 0)
5342 btrfs_item_key_to_cpu(eb, &first_key, 0);
5343 else
5344 btrfs_node_key_to_cpu(eb, &first_key, 0);
5345
5346 btrfs_tree_unlock(eb);
5347 free_extent_buffer(eb);
5348 prev_block = block_start;
5349 }
5350
5351 btrfs_record_root_in_trans(found_root);
5352 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5353 /*
5354 * try to update data extent references while
5355 * keeping metadata shared between snapshots.
5356 */
5357 if (pass == 1) {
5358 ret = relocate_one_path(trans, found_root,
5359 path, &first_key, ref_path,
5360 group, reloc_inode);
5361 if (ret < 0)
5362 goto out;
5363 continue;
5364 }
5365 /*
5366 * use fallback method to process the remaining
5367 * references.
5368 */
5369 if (!new_extents) {
5370 u64 group_start = group->key.objectid;
5371 new_extents = kmalloc(sizeof(*new_extents),
5372 GFP_NOFS);
5373 nr_extents = 1;
5374 ret = get_new_locations(reloc_inode,
5375 extent_key,
5376 group_start, 1,
5377 &new_extents,
5378 &nr_extents);
5379 if (ret)
5380 goto out;
5381 }
5382 ret = replace_one_extent(trans, found_root,
5383 path, extent_key,
5384 &first_key, ref_path,
5385 new_extents, nr_extents);
5386 } else {
5387 ret = relocate_tree_block(trans, found_root, path,
5388 &first_key, ref_path);
5389 }
5390 if (ret < 0)
5391 goto out;
5392 }
5393 ret = 0;
5394 out:
5395 btrfs_end_transaction(trans, extent_root);
5396 kfree(new_extents);
5397 kfree(ref_path);
5398 return ret;
5399 }
5400
5401 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
5402 {
5403 u64 num_devices;
5404 u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
5405 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
5406
5407 num_devices = root->fs_info->fs_devices->rw_devices;
5408 if (num_devices == 1) {
5409 stripped |= BTRFS_BLOCK_GROUP_DUP;
5410 stripped = flags & ~stripped;
5411
5412 /* turn raid0 into single device chunks */
5413 if (flags & BTRFS_BLOCK_GROUP_RAID0)
5414 return stripped;
5415
5416 /* turn mirroring into duplication */
5417 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
5418 BTRFS_BLOCK_GROUP_RAID10))
5419 return stripped | BTRFS_BLOCK_GROUP_DUP;
5420 return flags;
5421 } else {
5422 /* they already had raid on here, just return */
5423 if (flags & stripped)
5424 return flags;
5425
5426 stripped |= BTRFS_BLOCK_GROUP_DUP;
5427 stripped = flags & ~stripped;
5428
5429 /* switch duplicated blocks with raid1 */
5430 if (flags & BTRFS_BLOCK_GROUP_DUP)
5431 return stripped | BTRFS_BLOCK_GROUP_RAID1;
5432
5433 /* turn single device chunks into raid0 */
5434 return stripped | BTRFS_BLOCK_GROUP_RAID0;
5435 }
5436 return flags;
5437 }
5438
5439 static int __alloc_chunk_for_shrink(struct btrfs_root *root,
5440 struct btrfs_block_group_cache *shrink_block_group,
5441 int force)
5442 {
5443 struct btrfs_trans_handle *trans;
5444 u64 new_alloc_flags;
5445 u64 calc;
5446
5447 spin_lock(&shrink_block_group->lock);
5448 if (btrfs_block_group_used(&shrink_block_group->item) > 0) {
5449 spin_unlock(&shrink_block_group->lock);
5450
5451 trans = btrfs_start_transaction(root, 1);
5452 spin_lock(&shrink_block_group->lock);
5453
5454 new_alloc_flags = update_block_group_flags(root,
5455 shrink_block_group->flags);
5456 if (new_alloc_flags != shrink_block_group->flags) {
5457 calc =
5458 btrfs_block_group_used(&shrink_block_group->item);
5459 } else {
5460 calc = shrink_block_group->key.offset;
5461 }
5462 spin_unlock(&shrink_block_group->lock);
5463
5464 do_chunk_alloc(trans, root->fs_info->extent_root,
5465 calc + 2 * 1024 * 1024, new_alloc_flags, force);
5466
5467 btrfs_end_transaction(trans, root);
5468 } else
5469 spin_unlock(&shrink_block_group->lock);
5470 return 0;
5471 }
5472
5473 static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
5474 struct btrfs_root *root,
5475 u64 objectid, u64 size)
5476 {
5477 struct btrfs_path *path;
5478 struct btrfs_inode_item *item;
5479 struct extent_buffer *leaf;
5480 int ret;
5481
5482 path = btrfs_alloc_path();
5483 if (!path)
5484 return -ENOMEM;
5485
5486 ret = btrfs_insert_empty_inode(trans, root, path, objectid);
5487 if (ret)
5488 goto out;
5489
5490 leaf = path->nodes[0];
5491 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
5492 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
5493 btrfs_set_inode_generation(leaf, item, 1);
5494 btrfs_set_inode_size(leaf, item, size);
5495 btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
5496 btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS);
5497 btrfs_mark_buffer_dirty(leaf);
5498 btrfs_release_path(root, path);
5499 out:
5500 btrfs_free_path(path);
5501 return ret;
5502 }
5503
5504 static noinline struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info,
5505 struct btrfs_block_group_cache *group)
5506 {
5507 struct inode *inode = NULL;
5508 struct btrfs_trans_handle *trans;
5509 struct btrfs_root *root;
5510 struct btrfs_key root_key;
5511 u64 objectid = BTRFS_FIRST_FREE_OBJECTID;
5512 int err = 0;
5513
5514 root_key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
5515 root_key.type = BTRFS_ROOT_ITEM_KEY;
5516 root_key.offset = (u64)-1;
5517 root = btrfs_read_fs_root_no_name(fs_info, &root_key);
5518 if (IS_ERR(root))
5519 return ERR_CAST(root);
5520
5521 trans = btrfs_start_transaction(root, 1);
5522 BUG_ON(!trans);
5523
5524 err = btrfs_find_free_objectid(trans, root, objectid, &objectid);
5525 if (err)
5526 goto out;
5527
5528 err = __insert_orphan_inode(trans, root, objectid, group->key.offset);
5529 BUG_ON(err);
5530
5531 err = btrfs_insert_file_extent(trans, root, objectid, 0, 0, 0,
5532 group->key.offset, 0, group->key.offset,
5533 0, 0, 0);
5534 BUG_ON(err);
5535
5536 inode = btrfs_iget_locked(root->fs_info->sb, objectid, root);
5537 if (inode->i_state & I_NEW) {
5538 BTRFS_I(inode)->root = root;
5539 BTRFS_I(inode)->location.objectid = objectid;
5540 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
5541 BTRFS_I(inode)->location.offset = 0;
5542 btrfs_read_locked_inode(inode);
5543 unlock_new_inode(inode);
5544 BUG_ON(is_bad_inode(inode));
5545 } else {
5546 BUG_ON(1);
5547 }
5548 BTRFS_I(inode)->index_cnt = group->key.objectid;
5549
5550 err = btrfs_orphan_add(trans, inode);
5551 out:
5552 btrfs_end_transaction(trans, root);
5553 if (err) {
5554 if (inode)
5555 iput(inode);
5556 inode = ERR_PTR(err);
5557 }
5558 return inode;
5559 }
5560
5561 int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len)
5562 {
5563
5564 struct btrfs_ordered_sum *sums;
5565 struct btrfs_sector_sum *sector_sum;
5566 struct btrfs_ordered_extent *ordered;
5567 struct btrfs_root *root = BTRFS_I(inode)->root;
5568 struct list_head list;
5569 size_t offset;
5570 int ret;
5571 u64 disk_bytenr;
5572
5573 INIT_LIST_HEAD(&list);
5574
5575 ordered = btrfs_lookup_ordered_extent(inode, file_pos);
5576 BUG_ON(ordered->file_offset != file_pos || ordered->len != len);
5577
5578 disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt;
5579 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, disk_bytenr,
5580 disk_bytenr + len - 1, &list);
5581
5582 while (!list_empty(&list)) {
5583 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
5584 list_del_init(&sums->list);
5585
5586 sector_sum = sums->sums;
5587 sums->bytenr = ordered->start;
5588
5589 offset = 0;
5590 while (offset < sums->len) {
5591 sector_sum->bytenr += ordered->start - disk_bytenr;
5592 sector_sum++;
5593 offset += root->sectorsize;
5594 }
5595
5596 btrfs_add_ordered_sum(inode, ordered, sums);
5597 }
5598 btrfs_put_ordered_extent(ordered);
5599 return 0;
5600 }
5601
5602 int btrfs_relocate_block_group(struct btrfs_root *root, u64 group_start)
5603 {
5604 struct btrfs_trans_handle *trans;
5605 struct btrfs_path *path;
5606 struct btrfs_fs_info *info = root->fs_info;
5607 struct extent_buffer *leaf;
5608 struct inode *reloc_inode;
5609 struct btrfs_block_group_cache *block_group;
5610 struct btrfs_key key;
5611 u64 skipped;
5612 u64 cur_byte;
5613 u64 total_found;
5614 u32 nritems;
5615 int ret;
5616 int progress;
5617 int pass = 0;
5618
5619 root = root->fs_info->extent_root;
5620
5621 block_group = btrfs_lookup_block_group(info, group_start);
5622 BUG_ON(!block_group);
5623
5624 printk(KERN_INFO "btrfs relocating block group %llu flags %llu\n",
5625 (unsigned long long)block_group->key.objectid,
5626 (unsigned long long)block_group->flags);
5627
5628 path = btrfs_alloc_path();
5629 BUG_ON(!path);
5630
5631 reloc_inode = create_reloc_inode(info, block_group);
5632 BUG_ON(IS_ERR(reloc_inode));
5633
5634 __alloc_chunk_for_shrink(root, block_group, 1);
5635 set_block_group_readonly(block_group);
5636
5637 btrfs_start_delalloc_inodes(info->tree_root);
5638 btrfs_wait_ordered_extents(info->tree_root, 0);
5639 again:
5640 skipped = 0;
5641 total_found = 0;
5642 progress = 0;
5643 key.objectid = block_group->key.objectid;
5644 key.offset = 0;
5645 key.type = 0;
5646 cur_byte = key.objectid;
5647
5648 trans = btrfs_start_transaction(info->tree_root, 1);
5649 btrfs_commit_transaction(trans, info->tree_root);
5650
5651 mutex_lock(&root->fs_info->cleaner_mutex);
5652 btrfs_clean_old_snapshots(info->tree_root);
5653 btrfs_remove_leaf_refs(info->tree_root, (u64)-1, 1);
5654 mutex_unlock(&root->fs_info->cleaner_mutex);
5655
5656 while (1) {
5657 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5658 if (ret < 0)
5659 goto out;
5660 next:
5661 leaf = path->nodes[0];
5662 nritems = btrfs_header_nritems(leaf);
5663 if (path->slots[0] >= nritems) {
5664 ret = btrfs_next_leaf(root, path);
5665 if (ret < 0)
5666 goto out;
5667 if (ret == 1) {
5668 ret = 0;
5669 break;
5670 }
5671 leaf = path->nodes[0];
5672 nritems = btrfs_header_nritems(leaf);
5673 }
5674
5675 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
5676
5677 if (key.objectid >= block_group->key.objectid +
5678 block_group->key.offset)
5679 break;
5680
5681 if (progress && need_resched()) {
5682 btrfs_release_path(root, path);
5683 cond_resched();
5684 progress = 0;
5685 continue;
5686 }
5687 progress = 1;
5688
5689 if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY ||
5690 key.objectid + key.offset <= cur_byte) {
5691 path->slots[0]++;
5692 goto next;
5693 }
5694
5695 total_found++;
5696 cur_byte = key.objectid + key.offset;
5697 btrfs_release_path(root, path);
5698
5699 __alloc_chunk_for_shrink(root, block_group, 0);
5700 ret = relocate_one_extent(root, path, &key, block_group,
5701 reloc_inode, pass);
5702 BUG_ON(ret < 0);
5703 if (ret > 0)
5704 skipped++;
5705
5706 key.objectid = cur_byte;
5707 key.type = 0;
5708 key.offset = 0;
5709 }
5710
5711 btrfs_release_path(root, path);
5712
5713 if (pass == 0) {
5714 btrfs_wait_ordered_range(reloc_inode, 0, (u64)-1);
5715 invalidate_mapping_pages(reloc_inode->i_mapping, 0, -1);
5716 }
5717
5718 if (total_found > 0) {
5719 printk(KERN_INFO "btrfs found %llu extents in pass %d\n",
5720 (unsigned long long)total_found, pass);
5721 pass++;
5722 if (total_found == skipped && pass > 2) {
5723 iput(reloc_inode);
5724 reloc_inode = create_reloc_inode(info, block_group);
5725 pass = 0;
5726 }
5727 goto again;
5728 }
5729
5730 /* delete reloc_inode */
5731 iput(reloc_inode);
5732
5733 /* unpin extents in this range */
5734 trans = btrfs_start_transaction(info->tree_root, 1);
5735 btrfs_commit_transaction(trans, info->tree_root);
5736
5737 spin_lock(&block_group->lock);
5738 WARN_ON(block_group->pinned > 0);
5739 WARN_ON(block_group->reserved > 0);
5740 WARN_ON(btrfs_block_group_used(&block_group->item) > 0);
5741 spin_unlock(&block_group->lock);
5742 put_block_group(block_group);
5743 ret = 0;
5744 out:
5745 btrfs_free_path(path);
5746 return ret;
5747 }
5748
5749 static int find_first_block_group(struct btrfs_root *root,
5750 struct btrfs_path *path, struct btrfs_key *key)
5751 {
5752 int ret = 0;
5753 struct btrfs_key found_key;
5754 struct extent_buffer *leaf;
5755 int slot;
5756
5757 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
5758 if (ret < 0)
5759 goto out;
5760
5761 while (1) {
5762 slot = path->slots[0];
5763 leaf = path->nodes[0];
5764 if (slot >= btrfs_header_nritems(leaf)) {
5765 ret = btrfs_next_leaf(root, path);
5766 if (ret == 0)
5767 continue;
5768 if (ret < 0)
5769 goto out;
5770 break;
5771 }
5772 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5773
5774 if (found_key.objectid >= key->objectid &&
5775 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
5776 ret = 0;
5777 goto out;
5778 }
5779 path->slots[0]++;
5780 }
5781 ret = -ENOENT;
5782 out:
5783 return ret;
5784 }
5785
5786 int btrfs_free_block_groups(struct btrfs_fs_info *info)
5787 {
5788 struct btrfs_block_group_cache *block_group;
5789 struct rb_node *n;
5790
5791 spin_lock(&info->block_group_cache_lock);
5792 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
5793 block_group = rb_entry(n, struct btrfs_block_group_cache,
5794 cache_node);
5795 rb_erase(&block_group->cache_node,
5796 &info->block_group_cache_tree);
5797 spin_unlock(&info->block_group_cache_lock);
5798
5799 btrfs_remove_free_space_cache(block_group);
5800 down_write(&block_group->space_info->groups_sem);
5801 list_del(&block_group->list);
5802 up_write(&block_group->space_info->groups_sem);
5803
5804 WARN_ON(atomic_read(&block_group->count) != 1);
5805 kfree(block_group);
5806
5807 spin_lock(&info->block_group_cache_lock);
5808 }
5809 spin_unlock(&info->block_group_cache_lock);
5810 return 0;
5811 }
5812
5813 int btrfs_read_block_groups(struct btrfs_root *root)
5814 {
5815 struct btrfs_path *path;
5816 int ret;
5817 struct btrfs_block_group_cache *cache;
5818 struct btrfs_fs_info *info = root->fs_info;
5819 struct btrfs_space_info *space_info;
5820 struct btrfs_key key;
5821 struct btrfs_key found_key;
5822 struct extent_buffer *leaf;
5823
5824 root = info->extent_root;
5825 key.objectid = 0;
5826 key.offset = 0;
5827 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
5828 path = btrfs_alloc_path();
5829 if (!path)
5830 return -ENOMEM;
5831
5832 while (1) {
5833 ret = find_first_block_group(root, path, &key);
5834 if (ret > 0) {
5835 ret = 0;
5836 goto error;
5837 }
5838 if (ret != 0)
5839 goto error;
5840
5841 leaf = path->nodes[0];
5842 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5843 cache = kzalloc(sizeof(*cache), GFP_NOFS);
5844 if (!cache) {
5845 ret = -ENOMEM;
5846 break;
5847 }
5848
5849 atomic_set(&cache->count, 1);
5850 spin_lock_init(&cache->lock);
5851 mutex_init(&cache->alloc_mutex);
5852 mutex_init(&cache->cache_mutex);
5853 INIT_LIST_HEAD(&cache->list);
5854 read_extent_buffer(leaf, &cache->item,
5855 btrfs_item_ptr_offset(leaf, path->slots[0]),
5856 sizeof(cache->item));
5857 memcpy(&cache->key, &found_key, sizeof(found_key));
5858
5859 key.objectid = found_key.objectid + found_key.offset;
5860 btrfs_release_path(root, path);
5861 cache->flags = btrfs_block_group_flags(&cache->item);
5862
5863 ret = update_space_info(info, cache->flags, found_key.offset,
5864 btrfs_block_group_used(&cache->item),
5865 &space_info);
5866 BUG_ON(ret);
5867 cache->space_info = space_info;
5868 down_write(&space_info->groups_sem);
5869 list_add_tail(&cache->list, &space_info->block_groups);
5870 up_write(&space_info->groups_sem);
5871
5872 ret = btrfs_add_block_group_cache(root->fs_info, cache);
5873 BUG_ON(ret);
5874
5875 set_avail_alloc_bits(root->fs_info, cache->flags);
5876 if (btrfs_chunk_readonly(root, cache->key.objectid))
5877 set_block_group_readonly(cache);
5878 }
5879 ret = 0;
5880 error:
5881 btrfs_free_path(path);
5882 return ret;
5883 }
5884
5885 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
5886 struct btrfs_root *root, u64 bytes_used,
5887 u64 type, u64 chunk_objectid, u64 chunk_offset,
5888 u64 size)
5889 {
5890 int ret;
5891 struct btrfs_root *extent_root;
5892 struct btrfs_block_group_cache *cache;
5893
5894 extent_root = root->fs_info->extent_root;
5895
5896 root->fs_info->last_trans_new_blockgroup = trans->transid;
5897
5898 cache = kzalloc(sizeof(*cache), GFP_NOFS);
5899 if (!cache)
5900 return -ENOMEM;
5901
5902 cache->key.objectid = chunk_offset;
5903 cache->key.offset = size;
5904 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
5905 atomic_set(&cache->count, 1);
5906 spin_lock_init(&cache->lock);
5907 mutex_init(&cache->alloc_mutex);
5908 mutex_init(&cache->cache_mutex);
5909 INIT_LIST_HEAD(&cache->list);
5910
5911 btrfs_set_block_group_used(&cache->item, bytes_used);
5912 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
5913 cache->flags = type;
5914 btrfs_set_block_group_flags(&cache->item, type);
5915
5916 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
5917 &cache->space_info);
5918 BUG_ON(ret);
5919 down_write(&cache->space_info->groups_sem);
5920 list_add_tail(&cache->list, &cache->space_info->block_groups);
5921 up_write(&cache->space_info->groups_sem);
5922
5923 ret = btrfs_add_block_group_cache(root->fs_info, cache);
5924 BUG_ON(ret);
5925
5926 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
5927 sizeof(cache->item));
5928 BUG_ON(ret);
5929
5930 finish_current_insert(trans, extent_root, 0);
5931 ret = del_pending_extents(trans, extent_root, 0);
5932 BUG_ON(ret);
5933 set_avail_alloc_bits(extent_root->fs_info, type);
5934
5935 return 0;
5936 }
5937
5938 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
5939 struct btrfs_root *root, u64 group_start)
5940 {
5941 struct btrfs_path *path;
5942 struct btrfs_block_group_cache *block_group;
5943 struct btrfs_key key;
5944 int ret;
5945
5946 root = root->fs_info->extent_root;
5947
5948 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
5949 BUG_ON(!block_group);
5950 BUG_ON(!block_group->ro);
5951
5952 memcpy(&key, &block_group->key, sizeof(key));
5953
5954 path = btrfs_alloc_path();
5955 BUG_ON(!path);
5956
5957 spin_lock(&root->fs_info->block_group_cache_lock);
5958 rb_erase(&block_group->cache_node,
5959 &root->fs_info->block_group_cache_tree);
5960 spin_unlock(&root->fs_info->block_group_cache_lock);
5961 btrfs_remove_free_space_cache(block_group);
5962 down_write(&block_group->space_info->groups_sem);
5963 list_del(&block_group->list);
5964 up_write(&block_group->space_info->groups_sem);
5965
5966 spin_lock(&block_group->space_info->lock);
5967 block_group->space_info->total_bytes -= block_group->key.offset;
5968 block_group->space_info->bytes_readonly -= block_group->key.offset;
5969 spin_unlock(&block_group->space_info->lock);
5970 block_group->space_info->full = 0;
5971
5972 put_block_group(block_group);
5973 put_block_group(block_group);
5974
5975 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
5976 if (ret > 0)
5977 ret = -EIO;
5978 if (ret < 0)
5979 goto out;
5980
5981 ret = btrfs_del_item(trans, root, path);
5982 out:
5983 btrfs_free_path(path);
5984 return ret;
5985 }
Support for the Chinese Samsung S3C2440 based development boards