projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Btrfs: add better -ENOSPC handling
[deliverable/linux.git] / fs / btrfs / extent-tree.c
1 /*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18 #include <linux/sched.h>
19 #include <linux/pagemap.h>
20 #include <linux/writeback.h>
21 #include <linux/blkdev.h>
22 #include <linux/sort.h>
23 #include "compat.h"
24 #include "hash.h"
25 #include "crc32c.h"
26 #include "ctree.h"
27 #include "disk-io.h"
28 #include "print-tree.h"
29 #include "transaction.h"
30 #include "volumes.h"
31 #include "locking.h"
32 #include "ref-cache.h"
33
34 #define PENDING_EXTENT_INSERT 0
35 #define PENDING_EXTENT_DELETE 1
36 #define PENDING_BACKREF_UPDATE 2
37
38 struct pending_extent_op {
39 int type;
40 u64 bytenr;
41 u64 num_bytes;
42 u64 parent;
43 u64 orig_parent;
44 u64 generation;
45 u64 orig_generation;
46 int level;
47 struct list_head list;
48 int del;
49 };
50
51 static int finish_current_insert(struct btrfs_trans_handle *trans,
52 struct btrfs_root *extent_root, int all);
53 static int del_pending_extents(struct btrfs_trans_handle *trans,
54 struct btrfs_root *extent_root, int all);
55 static int pin_down_bytes(struct btrfs_trans_handle *trans,
56 struct btrfs_root *root,
57 u64 bytenr, u64 num_bytes, int is_data);
58 static int update_block_group(struct btrfs_trans_handle *trans,
59 struct btrfs_root *root,
60 u64 bytenr, u64 num_bytes, int alloc,
61 int mark_free);
62
63 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
64 struct btrfs_root *extent_root, u64 alloc_bytes,
65 u64 flags, int force);
66
67 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
68 {
69 return (cache->flags & bits) == bits;
70 }
71
72 /*
73 * this adds the block group to the fs_info rb tree for the block group
74 * cache
75 */
76 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
77 struct btrfs_block_group_cache *block_group)
78 {
79 struct rb_node **p;
80 struct rb_node *parent = NULL;
81 struct btrfs_block_group_cache *cache;
82
83 spin_lock(&info->block_group_cache_lock);
84 p = &info->block_group_cache_tree.rb_node;
85
86 while (*p) {
87 parent = *p;
88 cache = rb_entry(parent, struct btrfs_block_group_cache,
89 cache_node);
90 if (block_group->key.objectid < cache->key.objectid) {
91 p = &(*p)->rb_left;
92 } else if (block_group->key.objectid > cache->key.objectid) {
93 p = &(*p)->rb_right;
94 } else {
95 spin_unlock(&info->block_group_cache_lock);
96 return -EEXIST;
97 }
98 }
99
100 rb_link_node(&block_group->cache_node, parent, p);
101 rb_insert_color(&block_group->cache_node,
102 &info->block_group_cache_tree);
103 spin_unlock(&info->block_group_cache_lock);
104
105 return 0;
106 }
107
108 /*
109 * This will return the block group at or after bytenr if contains is 0, else
110 * it will return the block group that contains the bytenr
111 */
112 static struct btrfs_block_group_cache *
113 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
114 int contains)
115 {
116 struct btrfs_block_group_cache *cache, *ret = NULL;
117 struct rb_node *n;
118 u64 end, start;
119
120 spin_lock(&info->block_group_cache_lock);
121 n = info->block_group_cache_tree.rb_node;
122
123 while (n) {
124 cache = rb_entry(n, struct btrfs_block_group_cache,
125 cache_node);
126 end = cache->key.objectid + cache->key.offset - 1;
127 start = cache->key.objectid;
128
129 if (bytenr < start) {
130 if (!contains && (!ret || start < ret->key.objectid))
131 ret = cache;
132 n = n->rb_left;
133 } else if (bytenr > start) {
134 if (contains && bytenr <= end) {
135 ret = cache;
136 break;
137 }
138 n = n->rb_right;
139 } else {
140 ret = cache;
141 break;
142 }
143 }
144 if (ret)
145 atomic_inc(&ret->count);
146 spin_unlock(&info->block_group_cache_lock);
147
148 return ret;
149 }
150
151 /*
152 * this is only called by cache_block_group, since we could have freed extents
153 * we need to check the pinned_extents for any extents that can't be used yet
154 * since their free space will be released as soon as the transaction commits.
155 */
156 static int add_new_free_space(struct btrfs_block_group_cache *block_group,
157 struct btrfs_fs_info *info, u64 start, u64 end)
158 {
159 u64 extent_start, extent_end, size;
160 int ret;
161
162 mutex_lock(&info->pinned_mutex);
163 while (start < end) {
164 ret = find_first_extent_bit(&info->pinned_extents, start,
165 &extent_start, &extent_end,
166 EXTENT_DIRTY);
167 if (ret)
168 break;
169
170 if (extent_start == start) {
171 start = extent_end + 1;
172 } else if (extent_start > start && extent_start < end) {
173 size = extent_start - start;
174 ret = btrfs_add_free_space(block_group, start,
175 size);
176 BUG_ON(ret);
177 start = extent_end + 1;
178 } else {
179 break;
180 }
181 }
182
183 if (start < end) {
184 size = end - start;
185 ret = btrfs_add_free_space(block_group, start, size);
186 BUG_ON(ret);
187 }
188 mutex_unlock(&info->pinned_mutex);
189
190 return 0;
191 }
192
193 static int remove_sb_from_cache(struct btrfs_root *root,
194 struct btrfs_block_group_cache *cache)
195 {
196 u64 bytenr;
197 u64 *logical;
198 int stripe_len;
199 int i, nr, ret;
200
201 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
202 bytenr = btrfs_sb_offset(i);
203 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
204 cache->key.objectid, bytenr, 0,
205 &logical, &nr, &stripe_len);
206 BUG_ON(ret);
207 while (nr--) {
208 btrfs_remove_free_space(cache, logical[nr],
209 stripe_len);
210 }
211 kfree(logical);
212 }
213 return 0;
214 }
215
216 static int cache_block_group(struct btrfs_root *root,
217 struct btrfs_block_group_cache *block_group)
218 {
219 struct btrfs_path *path;
220 int ret = 0;
221 struct btrfs_key key;
222 struct extent_buffer *leaf;
223 int slot;
224 u64 last;
225
226 if (!block_group)
227 return 0;
228
229 root = root->fs_info->extent_root;
230
231 if (block_group->cached)
232 return 0;
233
234 path = btrfs_alloc_path();
235 if (!path)
236 return -ENOMEM;
237
238 path->reada = 2;
239 /*
240 * we get into deadlocks with paths held by callers of this function.
241 * since the alloc_mutex is protecting things right now, just
242 * skip the locking here
243 */
244 path->skip_locking = 1;
245 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
246 key.objectid = last;
247 key.offset = 0;
248 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
249 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
250 if (ret < 0)
251 goto err;
252
253 while (1) {
254 leaf = path->nodes[0];
255 slot = path->slots[0];
256 if (slot >= btrfs_header_nritems(leaf)) {
257 ret = btrfs_next_leaf(root, path);
258 if (ret < 0)
259 goto err;
260 if (ret == 0)
261 continue;
262 else
263 break;
264 }
265 btrfs_item_key_to_cpu(leaf, &key, slot);
266 if (key.objectid < block_group->key.objectid)
267 goto next;
268
269 if (key.objectid >= block_group->key.objectid +
270 block_group->key.offset)
271 break;
272
273 if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
274 add_new_free_space(block_group, root->fs_info, last,
275 key.objectid);
276
277 last = key.objectid + key.offset;
278 }
279 next:
280 path->slots[0]++;
281 }
282
283 add_new_free_space(block_group, root->fs_info, last,
284 block_group->key.objectid +
285 block_group->key.offset);
286
287 remove_sb_from_cache(root, block_group);
288 block_group->cached = 1;
289 ret = 0;
290 err:
291 btrfs_free_path(path);
292 return ret;
293 }
294
295 /*
296 * return the block group that starts at or after bytenr
297 */
298 static struct btrfs_block_group_cache *
299 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
300 {
301 struct btrfs_block_group_cache *cache;
302
303 cache = block_group_cache_tree_search(info, bytenr, 0);
304
305 return cache;
306 }
307
308 /*
309 * return the block group that contains teh given bytenr
310 */
311 struct btrfs_block_group_cache *btrfs_lookup_block_group(
312 struct btrfs_fs_info *info,
313 u64 bytenr)
314 {
315 struct btrfs_block_group_cache *cache;
316
317 cache = block_group_cache_tree_search(info, bytenr, 1);
318
319 return cache;
320 }
321
322 static inline void put_block_group(struct btrfs_block_group_cache *cache)
323 {
324 if (atomic_dec_and_test(&cache->count))
325 kfree(cache);
326 }
327
328 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
329 u64 flags)
330 {
331 struct list_head *head = &info->space_info;
332 struct btrfs_space_info *found;
333 list_for_each_entry(found, head, list) {
334 if (found->flags == flags)
335 return found;
336 }
337 return NULL;
338 }
339
340 static u64 div_factor(u64 num, int factor)
341 {
342 if (factor == 10)
343 return num;
344 num *= factor;
345 do_div(num, 10);
346 return num;
347 }
348
349 u64 btrfs_find_block_group(struct btrfs_root *root,
350 u64 search_start, u64 search_hint, int owner)
351 {
352 struct btrfs_block_group_cache *cache;
353 u64 used;
354 u64 last = max(search_hint, search_start);
355 u64 group_start = 0;
356 int full_search = 0;
357 int factor = 9;
358 int wrapped = 0;
359 again:
360 while (1) {
361 cache = btrfs_lookup_first_block_group(root->fs_info, last);
362 if (!cache)
363 break;
364
365 spin_lock(&cache->lock);
366 last = cache->key.objectid + cache->key.offset;
367 used = btrfs_block_group_used(&cache->item);
368
369 if ((full_search || !cache->ro) &&
370 block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
371 if (used + cache->pinned + cache->reserved <
372 div_factor(cache->key.offset, factor)) {
373 group_start = cache->key.objectid;
374 spin_unlock(&cache->lock);
375 put_block_group(cache);
376 goto found;
377 }
378 }
379 spin_unlock(&cache->lock);
380 put_block_group(cache);
381 cond_resched();
382 }
383 if (!wrapped) {
384 last = search_start;
385 wrapped = 1;
386 goto again;
387 }
388 if (!full_search && factor < 10) {
389 last = search_start;
390 full_search = 1;
391 factor = 10;
392 goto again;
393 }
394 found:
395 return group_start;
396 }
397
398 /* simple helper to search for an existing extent at a given offset */
399 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
400 {
401 int ret;
402 struct btrfs_key key;
403 struct btrfs_path *path;
404
405 path = btrfs_alloc_path();
406 BUG_ON(!path);
407 key.objectid = start;
408 key.offset = len;
409 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
410 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
411 0, 0);
412 btrfs_free_path(path);
413 return ret;
414 }
415
416 /*
417 * Back reference rules. Back refs have three main goals:
418 *
419 * 1) differentiate between all holders of references to an extent so that
420 * when a reference is dropped we can make sure it was a valid reference
421 * before freeing the extent.
422 *
423 * 2) Provide enough information to quickly find the holders of an extent
424 * if we notice a given block is corrupted or bad.
425 *
426 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
427 * maintenance. This is actually the same as #2, but with a slightly
428 * different use case.
429 *
430 * File extents can be referenced by:
431 *
432 * - multiple snapshots, subvolumes, or different generations in one subvol
433 * - different files inside a single subvolume
434 * - different offsets inside a file (bookend extents in file.c)
435 *
436 * The extent ref structure has fields for:
437 *
438 * - Objectid of the subvolume root
439 * - Generation number of the tree holding the reference
440 * - objectid of the file holding the reference
441 * - number of references holding by parent node (alway 1 for tree blocks)
442 *
443 * Btree leaf may hold multiple references to a file extent. In most cases,
444 * these references are from same file and the corresponding offsets inside
445 * the file are close together.
446 *
447 * When a file extent is allocated the fields are filled in:
448 * (root_key.objectid, trans->transid, inode objectid, 1)
449 *
450 * When a leaf is cow'd new references are added for every file extent found
451 * in the leaf. It looks similar to the create case, but trans->transid will
452 * be different when the block is cow'd.
453 *
454 * (root_key.objectid, trans->transid, inode objectid,
455 * number of references in the leaf)
456 *
457 * When a file extent is removed either during snapshot deletion or
458 * file truncation, we find the corresponding back reference and check
459 * the following fields:
460 *
461 * (btrfs_header_owner(leaf), btrfs_header_generation(leaf),
462 * inode objectid)
463 *
464 * Btree extents can be referenced by:
465 *
466 * - Different subvolumes
467 * - Different generations of the same subvolume
468 *
469 * When a tree block is created, back references are inserted:
470 *
471 * (root->root_key.objectid, trans->transid, level, 1)
472 *
473 * When a tree block is cow'd, new back references are added for all the
474 * blocks it points to. If the tree block isn't in reference counted root,
475 * the old back references are removed. These new back references are of
476 * the form (trans->transid will have increased since creation):
477 *
478 * (root->root_key.objectid, trans->transid, level, 1)
479 *
480 * When a backref is in deleting, the following fields are checked:
481 *
482 * if backref was for a tree root:
483 * (btrfs_header_owner(itself), btrfs_header_generation(itself), level)
484 * else
485 * (btrfs_header_owner(parent), btrfs_header_generation(parent), level)
486 *
487 * Back Reference Key composing:
488 *
489 * The key objectid corresponds to the first byte in the extent, the key
490 * type is set to BTRFS_EXTENT_REF_KEY, and the key offset is the first
491 * byte of parent extent. If a extent is tree root, the key offset is set
492 * to the key objectid.
493 */
494
495 static noinline int lookup_extent_backref(struct btrfs_trans_handle *trans,
496 struct btrfs_root *root,
497 struct btrfs_path *path,
498 u64 bytenr, u64 parent,
499 u64 ref_root, u64 ref_generation,
500 u64 owner_objectid, int del)
501 {
502 struct btrfs_key key;
503 struct btrfs_extent_ref *ref;
504 struct extent_buffer *leaf;
505 u64 ref_objectid;
506 int ret;
507
508 key.objectid = bytenr;
509 key.type = BTRFS_EXTENT_REF_KEY;
510 key.offset = parent;
511
512 ret = btrfs_search_slot(trans, root, &key, path, del ? -1 : 0, 1);
513 if (ret < 0)
514 goto out;
515 if (ret > 0) {
516 ret = -ENOENT;
517 goto out;
518 }
519
520 leaf = path->nodes[0];
521 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
522 ref_objectid = btrfs_ref_objectid(leaf, ref);
523 if (btrfs_ref_root(leaf, ref) != ref_root ||
524 btrfs_ref_generation(leaf, ref) != ref_generation ||
525 (ref_objectid != owner_objectid &&
526 ref_objectid != BTRFS_MULTIPLE_OBJECTIDS)) {
527 ret = -EIO;
528 WARN_ON(1);
529 goto out;
530 }
531 ret = 0;
532 out:
533 return ret;
534 }
535
536 /*
537 * updates all the backrefs that are pending on update_list for the
538 * extent_root
539 */
540 static noinline int update_backrefs(struct btrfs_trans_handle *trans,
541 struct btrfs_root *extent_root,
542 struct btrfs_path *path,
543 struct list_head *update_list)
544 {
545 struct btrfs_key key;
546 struct btrfs_extent_ref *ref;
547 struct btrfs_fs_info *info = extent_root->fs_info;
548 struct pending_extent_op *op;
549 struct extent_buffer *leaf;
550 int ret = 0;
551 struct list_head *cur = update_list->next;
552 u64 ref_objectid;
553 u64 ref_root = extent_root->root_key.objectid;
554
555 op = list_entry(cur, struct pending_extent_op, list);
556
557 search:
558 key.objectid = op->bytenr;
559 key.type = BTRFS_EXTENT_REF_KEY;
560 key.offset = op->orig_parent;
561
562 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 1);
563 BUG_ON(ret);
564
565 leaf = path->nodes[0];
566
567 loop:
568 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
569
570 ref_objectid = btrfs_ref_objectid(leaf, ref);
571
572 if (btrfs_ref_root(leaf, ref) != ref_root ||
573 btrfs_ref_generation(leaf, ref) != op->orig_generation ||
574 (ref_objectid != op->level &&
575 ref_objectid != BTRFS_MULTIPLE_OBJECTIDS)) {
576 printk(KERN_ERR "btrfs couldn't find %llu, parent %llu, "
577 "root %llu, owner %u\n",
578 (unsigned long long)op->bytenr,
579 (unsigned long long)op->orig_parent,
580 (unsigned long long)ref_root, op->level);
581 btrfs_print_leaf(extent_root, leaf);
582 BUG();
583 }
584
585 key.objectid = op->bytenr;
586 key.offset = op->parent;
587 key.type = BTRFS_EXTENT_REF_KEY;
588 ret = btrfs_set_item_key_safe(trans, extent_root, path, &key);
589 BUG_ON(ret);
590 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
591 btrfs_set_ref_generation(leaf, ref, op->generation);
592
593 cur = cur->next;
594
595 list_del_init(&op->list);
596 unlock_extent(&info->extent_ins, op->bytenr,
597 op->bytenr + op->num_bytes - 1, GFP_NOFS);
598 kfree(op);
599
600 if (cur == update_list) {
601 btrfs_mark_buffer_dirty(path->nodes[0]);
602 btrfs_release_path(extent_root, path);
603 goto out;
604 }
605
606 op = list_entry(cur, struct pending_extent_op, list);
607
608 path->slots[0]++;
609 while (path->slots[0] < btrfs_header_nritems(leaf)) {
610 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
611 if (key.objectid == op->bytenr &&
612 key.type == BTRFS_EXTENT_REF_KEY)
613 goto loop;
614 path->slots[0]++;
615 }
616
617 btrfs_mark_buffer_dirty(path->nodes[0]);
618 btrfs_release_path(extent_root, path);
619 goto search;
620
621 out:
622 return 0;
623 }
624
625 static noinline int insert_extents(struct btrfs_trans_handle *trans,
626 struct btrfs_root *extent_root,
627 struct btrfs_path *path,
628 struct list_head *insert_list, int nr)
629 {
630 struct btrfs_key *keys;
631 u32 *data_size;
632 struct pending_extent_op *op;
633 struct extent_buffer *leaf;
634 struct list_head *cur = insert_list->next;
635 struct btrfs_fs_info *info = extent_root->fs_info;
636 u64 ref_root = extent_root->root_key.objectid;
637 int i = 0, last = 0, ret;
638 int total = nr * 2;
639
640 if (!nr)
641 return 0;
642
643 keys = kzalloc(total * sizeof(struct btrfs_key), GFP_NOFS);
644 if (!keys)
645 return -ENOMEM;
646
647 data_size = kzalloc(total * sizeof(u32), GFP_NOFS);
648 if (!data_size) {
649 kfree(keys);
650 return -ENOMEM;
651 }
652
653 list_for_each_entry(op, insert_list, list) {
654 keys[i].objectid = op->bytenr;
655 keys[i].offset = op->num_bytes;
656 keys[i].type = BTRFS_EXTENT_ITEM_KEY;
657 data_size[i] = sizeof(struct btrfs_extent_item);
658 i++;
659
660 keys[i].objectid = op->bytenr;
661 keys[i].offset = op->parent;
662 keys[i].type = BTRFS_EXTENT_REF_KEY;
663 data_size[i] = sizeof(struct btrfs_extent_ref);
664 i++;
665 }
666
667 op = list_entry(cur, struct pending_extent_op, list);
668 i = 0;
669 while (i < total) {
670 int c;
671 ret = btrfs_insert_some_items(trans, extent_root, path,
672 keys+i, data_size+i, total-i);
673 BUG_ON(ret < 0);
674
675 if (last && ret > 1)
676 BUG();
677
678 leaf = path->nodes[0];
679 for (c = 0; c < ret; c++) {
680 int ref_first = keys[i].type == BTRFS_EXTENT_REF_KEY;
681
682 /*
683 * if the first item we inserted was a backref, then
684 * the EXTENT_ITEM will be the odd c's, else it will
685 * be the even c's
686 */
687 if ((ref_first && (c % 2)) ||
688 (!ref_first && !(c % 2))) {
689 struct btrfs_extent_item *itm;
690
691 itm = btrfs_item_ptr(leaf, path->slots[0] + c,
692 struct btrfs_extent_item);
693 btrfs_set_extent_refs(path->nodes[0], itm, 1);
694 op->del++;
695 } else {
696 struct btrfs_extent_ref *ref;
697
698 ref = btrfs_item_ptr(leaf, path->slots[0] + c,
699 struct btrfs_extent_ref);
700 btrfs_set_ref_root(leaf, ref, ref_root);
701 btrfs_set_ref_generation(leaf, ref,
702 op->generation);
703 btrfs_set_ref_objectid(leaf, ref, op->level);
704 btrfs_set_ref_num_refs(leaf, ref, 1);
705 op->del++;
706 }
707
708 /*
709 * using del to see when its ok to free up the
710 * pending_extent_op. In the case where we insert the
711 * last item on the list in order to help do batching
712 * we need to not free the extent op until we actually
713 * insert the extent_item
714 */
715 if (op->del == 2) {
716 unlock_extent(&info->extent_ins, op->bytenr,
717 op->bytenr + op->num_bytes - 1,
718 GFP_NOFS);
719 cur = cur->next;
720 list_del_init(&op->list);
721 kfree(op);
722 if (cur != insert_list)
723 op = list_entry(cur,
724 struct pending_extent_op,
725 list);
726 }
727 }
728 btrfs_mark_buffer_dirty(leaf);
729 btrfs_release_path(extent_root, path);
730
731 /*
732 * Ok backref's and items usually go right next to eachother,
733 * but if we could only insert 1 item that means that we
734 * inserted on the end of a leaf, and we have no idea what may
735 * be on the next leaf so we just play it safe. In order to
736 * try and help this case we insert the last thing on our
737 * insert list so hopefully it will end up being the last
738 * thing on the leaf and everything else will be before it,
739 * which will let us insert a whole bunch of items at the same
740 * time.
741 */
742 if (ret == 1 && !last && (i + ret < total)) {
743 /*
744 * last: where we will pick up the next time around
745 * i: our current key to insert, will be total - 1
746 * cur: the current op we are screwing with
747 * op: duh
748 */
749 last = i + ret;
750 i = total - 1;
751 cur = insert_list->prev;
752 op = list_entry(cur, struct pending_extent_op, list);
753 } else if (last) {
754 /*
755 * ok we successfully inserted the last item on the
756 * list, lets reset everything
757 *
758 * i: our current key to insert, so where we left off
759 * last time
760 * last: done with this
761 * cur: the op we are messing with
762 * op: duh
763 * total: since we inserted the last key, we need to
764 * decrement total so we dont overflow
765 */
766 i = last;
767 last = 0;
768 total--;
769 if (i < total) {
770 cur = insert_list->next;
771 op = list_entry(cur, struct pending_extent_op,
772 list);
773 }
774 } else {
775 i += ret;
776 }
777
778 cond_resched();
779 }
780 ret = 0;
781 kfree(keys);
782 kfree(data_size);
783 return ret;
784 }
785
786 static noinline int insert_extent_backref(struct btrfs_trans_handle *trans,
787 struct btrfs_root *root,
788 struct btrfs_path *path,
789 u64 bytenr, u64 parent,
790 u64 ref_root, u64 ref_generation,
791 u64 owner_objectid)
792 {
793 struct btrfs_key key;
794 struct extent_buffer *leaf;
795 struct btrfs_extent_ref *ref;
796 u32 num_refs;
797 int ret;
798
799 key.objectid = bytenr;
800 key.type = BTRFS_EXTENT_REF_KEY;
801 key.offset = parent;
802
803 ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*ref));
804 if (ret == 0) {
805 leaf = path->nodes[0];
806 ref = btrfs_item_ptr(leaf, path->slots[0],
807 struct btrfs_extent_ref);
808 btrfs_set_ref_root(leaf, ref, ref_root);
809 btrfs_set_ref_generation(leaf, ref, ref_generation);
810 btrfs_set_ref_objectid(leaf, ref, owner_objectid);
811 btrfs_set_ref_num_refs(leaf, ref, 1);
812 } else if (ret == -EEXIST) {
813 u64 existing_owner;
814 BUG_ON(owner_objectid < BTRFS_FIRST_FREE_OBJECTID);
815 leaf = path->nodes[0];
816 ref = btrfs_item_ptr(leaf, path->slots[0],
817 struct btrfs_extent_ref);
818 if (btrfs_ref_root(leaf, ref) != ref_root ||
819 btrfs_ref_generation(leaf, ref) != ref_generation) {
820 ret = -EIO;
821 WARN_ON(1);
822 goto out;
823 }
824
825 num_refs = btrfs_ref_num_refs(leaf, ref);
826 BUG_ON(num_refs == 0);
827 btrfs_set_ref_num_refs(leaf, ref, num_refs + 1);
828
829 existing_owner = btrfs_ref_objectid(leaf, ref);
830 if (existing_owner != owner_objectid &&
831 existing_owner != BTRFS_MULTIPLE_OBJECTIDS) {
832 btrfs_set_ref_objectid(leaf, ref,
833 BTRFS_MULTIPLE_OBJECTIDS);
834 }
835 ret = 0;
836 } else {
837 goto out;
838 }
839 btrfs_mark_buffer_dirty(path->nodes[0]);
840 out:
841 btrfs_release_path(root, path);
842 return ret;
843 }
844
845 static noinline int remove_extent_backref(struct btrfs_trans_handle *trans,
846 struct btrfs_root *root,
847 struct btrfs_path *path)
848 {
849 struct extent_buffer *leaf;
850 struct btrfs_extent_ref *ref;
851 u32 num_refs;
852 int ret = 0;
853
854 leaf = path->nodes[0];
855 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
856 num_refs = btrfs_ref_num_refs(leaf, ref);
857 BUG_ON(num_refs == 0);
858 num_refs -= 1;
859 if (num_refs == 0) {
860 ret = btrfs_del_item(trans, root, path);
861 } else {
862 btrfs_set_ref_num_refs(leaf, ref, num_refs);
863 btrfs_mark_buffer_dirty(leaf);
864 }
865 btrfs_release_path(root, path);
866 return ret;
867 }
868
869 #ifdef BIO_RW_DISCARD
870 static void btrfs_issue_discard(struct block_device *bdev,
871 u64 start, u64 len)
872 {
873 blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL);
874 }
875 #endif
876
877 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
878 u64 num_bytes)
879 {
880 #ifdef BIO_RW_DISCARD
881 int ret;
882 u64 map_length = num_bytes;
883 struct btrfs_multi_bio *multi = NULL;
884
885 /* Tell the block device(s) that the sectors can be discarded */
886 ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
887 bytenr, &map_length, &multi, 0);
888 if (!ret) {
889 struct btrfs_bio_stripe *stripe = multi->stripes;
890 int i;
891
892 if (map_length > num_bytes)
893 map_length = num_bytes;
894
895 for (i = 0; i < multi->num_stripes; i++, stripe++) {
896 btrfs_issue_discard(stripe->dev->bdev,
897 stripe->physical,
898 map_length);
899 }
900 kfree(multi);
901 }
902
903 return ret;
904 #else
905 return 0;
906 #endif
907 }
908
909 static noinline int free_extents(struct btrfs_trans_handle *trans,
910 struct btrfs_root *extent_root,
911 struct list_head *del_list)
912 {
913 struct btrfs_fs_info *info = extent_root->fs_info;
914 struct btrfs_path *path;
915 struct btrfs_key key, found_key;
916 struct extent_buffer *leaf;
917 struct list_head *cur;
918 struct pending_extent_op *op;
919 struct btrfs_extent_item *ei;
920 int ret, num_to_del, extent_slot = 0, found_extent = 0;
921 u32 refs;
922 u64 bytes_freed = 0;
923
924 path = btrfs_alloc_path();
925 if (!path)
926 return -ENOMEM;
927 path->reada = 1;
928
929 search:
930 /* search for the backref for the current ref we want to delete */
931 cur = del_list->next;
932 op = list_entry(cur, struct pending_extent_op, list);
933 ret = lookup_extent_backref(trans, extent_root, path, op->bytenr,
934 op->orig_parent,
935 extent_root->root_key.objectid,
936 op->orig_generation, op->level, 1);
937 if (ret) {
938 printk(KERN_ERR "btrfs unable to find backref byte nr %llu "
939 "root %llu gen %llu owner %u\n",
940 (unsigned long long)op->bytenr,
941 (unsigned long long)extent_root->root_key.objectid,
942 (unsigned long long)op->orig_generation, op->level);
943 btrfs_print_leaf(extent_root, path->nodes[0]);
944 WARN_ON(1);
945 goto out;
946 }
947
948 extent_slot = path->slots[0];
949 num_to_del = 1;
950 found_extent = 0;
951
952 /*
953 * if we aren't the first item on the leaf we can move back one and see
954 * if our ref is right next to our extent item
955 */
956 if (likely(extent_slot)) {
957 extent_slot--;
958 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
959 extent_slot);
960 if (found_key.objectid == op->bytenr &&
961 found_key.type == BTRFS_EXTENT_ITEM_KEY &&
962 found_key.offset == op->num_bytes) {
963 num_to_del++;
964 found_extent = 1;
965 }
966 }
967
968 /*
969 * if we didn't find the extent we need to delete the backref and then
970 * search for the extent item key so we can update its ref count
971 */
972 if (!found_extent) {
973 key.objectid = op->bytenr;
974 key.type = BTRFS_EXTENT_ITEM_KEY;
975 key.offset = op->num_bytes;
976
977 ret = remove_extent_backref(trans, extent_root, path);
978 BUG_ON(ret);
979 btrfs_release_path(extent_root, path);
980 ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1);
981 BUG_ON(ret);
982 extent_slot = path->slots[0];
983 }
984
985 /* this is where we update the ref count for the extent */
986 leaf = path->nodes[0];
987 ei = btrfs_item_ptr(leaf, extent_slot, struct btrfs_extent_item);
988 refs = btrfs_extent_refs(leaf, ei);
989 BUG_ON(refs == 0);
990 refs--;
991 btrfs_set_extent_refs(leaf, ei, refs);
992
993 btrfs_mark_buffer_dirty(leaf);
994
995 /*
996 * This extent needs deleting. The reason cur_slot is extent_slot +
997 * num_to_del is because extent_slot points to the slot where the extent
998 * is, and if the backref was not right next to the extent we will be
999 * deleting at least 1 item, and will want to start searching at the
1000 * slot directly next to extent_slot. However if we did find the
1001 * backref next to the extent item them we will be deleting at least 2
1002 * items and will want to start searching directly after the ref slot
1003 */
1004 if (!refs) {
1005 struct list_head *pos, *n, *end;
1006 int cur_slot = extent_slot+num_to_del;
1007 u64 super_used;
1008 u64 root_used;
1009
1010 path->slots[0] = extent_slot;
1011 bytes_freed = op->num_bytes;
1012
1013 mutex_lock(&info->pinned_mutex);
1014 ret = pin_down_bytes(trans, extent_root, op->bytenr,
1015 op->num_bytes, op->level >=
1016 BTRFS_FIRST_FREE_OBJECTID);
1017 mutex_unlock(&info->pinned_mutex);
1018 BUG_ON(ret < 0);
1019 op->del = ret;
1020
1021 /*
1022 * we need to see if we can delete multiple things at once, so
1023 * start looping through the list of extents we are wanting to
1024 * delete and see if their extent/backref's are right next to
1025 * eachother and the extents only have 1 ref
1026 */
1027 for (pos = cur->next; pos != del_list; pos = pos->next) {
1028 struct pending_extent_op *tmp;
1029
1030 tmp = list_entry(pos, struct pending_extent_op, list);
1031
1032 /* we only want to delete extent+ref at this stage */
1033 if (cur_slot >= btrfs_header_nritems(leaf) - 1)
1034 break;
1035
1036 btrfs_item_key_to_cpu(leaf, &found_key, cur_slot);
1037 if (found_key.objectid != tmp->bytenr ||
1038 found_key.type != BTRFS_EXTENT_ITEM_KEY ||
1039 found_key.offset != tmp->num_bytes)
1040 break;
1041
1042 /* check to make sure this extent only has one ref */
1043 ei = btrfs_item_ptr(leaf, cur_slot,
1044 struct btrfs_extent_item);
1045 if (btrfs_extent_refs(leaf, ei) != 1)
1046 break;
1047
1048 btrfs_item_key_to_cpu(leaf, &found_key, cur_slot+1);
1049 if (found_key.objectid != tmp->bytenr ||
1050 found_key.type != BTRFS_EXTENT_REF_KEY ||
1051 found_key.offset != tmp->orig_parent)
1052 break;
1053
1054 /*
1055 * the ref is right next to the extent, we can set the
1056 * ref count to 0 since we will delete them both now
1057 */
1058 btrfs_set_extent_refs(leaf, ei, 0);
1059
1060 /* pin down the bytes for this extent */
1061 mutex_lock(&info->pinned_mutex);
1062 ret = pin_down_bytes(trans, extent_root, tmp->bytenr,
1063 tmp->num_bytes, tmp->level >=
1064 BTRFS_FIRST_FREE_OBJECTID);
1065 mutex_unlock(&info->pinned_mutex);
1066 BUG_ON(ret < 0);
1067
1068 /*
1069 * use the del field to tell if we need to go ahead and
1070 * free up the extent when we delete the item or not.
1071 */
1072 tmp->del = ret;
1073 bytes_freed += tmp->num_bytes;
1074
1075 num_to_del += 2;
1076 cur_slot += 2;
1077 }
1078 end = pos;
1079
1080 /* update the free space counters */
1081 spin_lock(&info->delalloc_lock);
1082 super_used = btrfs_super_bytes_used(&info->super_copy);
1083 btrfs_set_super_bytes_used(&info->super_copy,
1084 super_used - bytes_freed);
1085
1086 root_used = btrfs_root_used(&extent_root->root_item);
1087 btrfs_set_root_used(&extent_root->root_item,
1088 root_used - bytes_freed);
1089 spin_unlock(&info->delalloc_lock);
1090
1091 /* delete the items */
1092 ret = btrfs_del_items(trans, extent_root, path,
1093 path->slots[0], num_to_del);
1094 BUG_ON(ret);
1095
1096 /*
1097 * loop through the extents we deleted and do the cleanup work
1098 * on them
1099 */
1100 for (pos = cur, n = pos->next; pos != end;
1101 pos = n, n = pos->next) {
1102 struct pending_extent_op *tmp;
1103 tmp = list_entry(pos, struct pending_extent_op, list);
1104
1105 /*
1106 * remember tmp->del tells us wether or not we pinned
1107 * down the extent
1108 */
1109 ret = update_block_group(trans, extent_root,
1110 tmp->bytenr, tmp->num_bytes, 0,
1111 tmp->del);
1112 BUG_ON(ret);
1113
1114 list_del_init(&tmp->list);
1115 unlock_extent(&info->extent_ins, tmp->bytenr,
1116 tmp->bytenr + tmp->num_bytes - 1,
1117 GFP_NOFS);
1118 kfree(tmp);
1119 }
1120 } else if (refs && found_extent) {
1121 /*
1122 * the ref and extent were right next to eachother, but the
1123 * extent still has a ref, so just free the backref and keep
1124 * going
1125 */
1126 ret = remove_extent_backref(trans, extent_root, path);
1127 BUG_ON(ret);
1128
1129 list_del_init(&op->list);
1130 unlock_extent(&info->extent_ins, op->bytenr,
1131 op->bytenr + op->num_bytes - 1, GFP_NOFS);
1132 kfree(op);
1133 } else {
1134 /*
1135 * the extent has multiple refs and the backref we were looking
1136 * for was not right next to it, so just unlock and go next,
1137 * we're good to go
1138 */
1139 list_del_init(&op->list);
1140 unlock_extent(&info->extent_ins, op->bytenr,
1141 op->bytenr + op->num_bytes - 1, GFP_NOFS);
1142 kfree(op);
1143 }
1144
1145 btrfs_release_path(extent_root, path);
1146 if (!list_empty(del_list))
1147 goto search;
1148
1149 out:
1150 btrfs_free_path(path);
1151 return ret;
1152 }
1153
1154 static int __btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
1155 struct btrfs_root *root, u64 bytenr,
1156 u64 orig_parent, u64 parent,
1157 u64 orig_root, u64 ref_root,
1158 u64 orig_generation, u64 ref_generation,
1159 u64 owner_objectid)
1160 {
1161 int ret;
1162 struct btrfs_root *extent_root = root->fs_info->extent_root;
1163 struct btrfs_path *path;
1164
1165 if (root == root->fs_info->extent_root) {
1166 struct pending_extent_op *extent_op;
1167 u64 num_bytes;
1168
1169 BUG_ON(owner_objectid >= BTRFS_MAX_LEVEL);
1170 num_bytes = btrfs_level_size(root, (int)owner_objectid);
1171 mutex_lock(&root->fs_info->extent_ins_mutex);
1172 if (test_range_bit(&root->fs_info->extent_ins, bytenr,
1173 bytenr + num_bytes - 1, EXTENT_WRITEBACK, 0)) {
1174 u64 priv;
1175 ret = get_state_private(&root->fs_info->extent_ins,
1176 bytenr, &priv);
1177 BUG_ON(ret);
1178 extent_op = (struct pending_extent_op *)
1179 (unsigned long)priv;
1180 BUG_ON(extent_op->parent != orig_parent);
1181 BUG_ON(extent_op->generation != orig_generation);
1182
1183 extent_op->parent = parent;
1184 extent_op->generation = ref_generation;
1185 } else {
1186 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
1187 BUG_ON(!extent_op);
1188
1189 extent_op->type = PENDING_BACKREF_UPDATE;
1190 extent_op->bytenr = bytenr;
1191 extent_op->num_bytes = num_bytes;
1192 extent_op->parent = parent;
1193 extent_op->orig_parent = orig_parent;
1194 extent_op->generation = ref_generation;
1195 extent_op->orig_generation = orig_generation;
1196 extent_op->level = (int)owner_objectid;
1197 INIT_LIST_HEAD(&extent_op->list);
1198 extent_op->del = 0;
1199
1200 set_extent_bits(&root->fs_info->extent_ins,
1201 bytenr, bytenr + num_bytes - 1,
1202 EXTENT_WRITEBACK, GFP_NOFS);
1203 set_state_private(&root->fs_info->extent_ins,
1204 bytenr, (unsigned long)extent_op);
1205 }
1206 mutex_unlock(&root->fs_info->extent_ins_mutex);
1207 return 0;
1208 }
1209
1210 path = btrfs_alloc_path();
1211 if (!path)
1212 return -ENOMEM;
1213 ret = lookup_extent_backref(trans, extent_root, path,
1214 bytenr, orig_parent, orig_root,
1215 orig_generation, owner_objectid, 1);
1216 if (ret)
1217 goto out;
1218 ret = remove_extent_backref(trans, extent_root, path);
1219 if (ret)
1220 goto out;
1221 ret = insert_extent_backref(trans, extent_root, path, bytenr,
1222 parent, ref_root, ref_generation,
1223 owner_objectid);
1224 BUG_ON(ret);
1225 finish_current_insert(trans, extent_root, 0);
1226 del_pending_extents(trans, extent_root, 0);
1227 out:
1228 btrfs_free_path(path);
1229 return ret;
1230 }
1231
1232 int btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
1233 struct btrfs_root *root, u64 bytenr,
1234 u64 orig_parent, u64 parent,
1235 u64 ref_root, u64 ref_generation,
1236 u64 owner_objectid)
1237 {
1238 int ret;
1239 if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
1240 owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
1241 return 0;
1242 ret = __btrfs_update_extent_ref(trans, root, bytenr, orig_parent,
1243 parent, ref_root, ref_root,
1244 ref_generation, ref_generation,
1245 owner_objectid);
1246 return ret;
1247 }
1248
1249 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1250 struct btrfs_root *root, u64 bytenr,
1251 u64 orig_parent, u64 parent,
1252 u64 orig_root, u64 ref_root,
1253 u64 orig_generation, u64 ref_generation,
1254 u64 owner_objectid)
1255 {
1256 struct btrfs_path *path;
1257 int ret;
1258 struct btrfs_key key;
1259 struct extent_buffer *l;
1260 struct btrfs_extent_item *item;
1261 u32 refs;
1262
1263 path = btrfs_alloc_path();
1264 if (!path)
1265 return -ENOMEM;
1266
1267 path->reada = 1;
1268 key.objectid = bytenr;
1269 key.type = BTRFS_EXTENT_ITEM_KEY;
1270 key.offset = (u64)-1;
1271
1272 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
1273 0, 1);
1274 if (ret < 0)
1275 return ret;
1276 BUG_ON(ret == 0 || path->slots[0] == 0);
1277
1278 path->slots[0]--;
1279 l = path->nodes[0];
1280
1281 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
1282 if (key.objectid != bytenr) {
1283 btrfs_print_leaf(root->fs_info->extent_root, path->nodes[0]);
1284 printk(KERN_ERR "btrfs wanted %llu found %llu\n",
1285 (unsigned long long)bytenr,
1286 (unsigned long long)key.objectid);
1287 BUG();
1288 }
1289 BUG_ON(key.type != BTRFS_EXTENT_ITEM_KEY);
1290
1291 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
1292 refs = btrfs_extent_refs(l, item);
1293 btrfs_set_extent_refs(l, item, refs + 1);
1294 btrfs_mark_buffer_dirty(path->nodes[0]);
1295
1296 btrfs_release_path(root->fs_info->extent_root, path);
1297
1298 path->reada = 1;
1299 ret = insert_extent_backref(trans, root->fs_info->extent_root,
1300 path, bytenr, parent,
1301 ref_root, ref_generation,
1302 owner_objectid);
1303 BUG_ON(ret);
1304 finish_current_insert(trans, root->fs_info->extent_root, 0);
1305 del_pending_extents(trans, root->fs_info->extent_root, 0);
1306
1307 btrfs_free_path(path);
1308 return 0;
1309 }
1310
1311 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1312 struct btrfs_root *root,
1313 u64 bytenr, u64 num_bytes, u64 parent,
1314 u64 ref_root, u64 ref_generation,
1315 u64 owner_objectid)
1316 {
1317 int ret;
1318 if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
1319 owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
1320 return 0;
1321 ret = __btrfs_inc_extent_ref(trans, root, bytenr, 0, parent,
1322 0, ref_root, 0, ref_generation,
1323 owner_objectid);
1324 return ret;
1325 }
1326
1327 int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
1328 struct btrfs_root *root)
1329 {
1330 u64 start;
1331 u64 end;
1332 int ret;
1333
1334 while(1) {
1335 finish_current_insert(trans, root->fs_info->extent_root, 1);
1336 del_pending_extents(trans, root->fs_info->extent_root, 1);
1337
1338 /* is there more work to do? */
1339 ret = find_first_extent_bit(&root->fs_info->pending_del,
1340 0, &start, &end, EXTENT_WRITEBACK);
1341 if (!ret)
1342 continue;
1343 ret = find_first_extent_bit(&root->fs_info->extent_ins,
1344 0, &start, &end, EXTENT_WRITEBACK);
1345 if (!ret)
1346 continue;
1347 break;
1348 }
1349 return 0;
1350 }
1351
1352 int btrfs_lookup_extent_ref(struct btrfs_trans_handle *trans,
1353 struct btrfs_root *root, u64 bytenr,
1354 u64 num_bytes, u32 *refs)
1355 {
1356 struct btrfs_path *path;
1357 int ret;
1358 struct btrfs_key key;
1359 struct extent_buffer *l;
1360 struct btrfs_extent_item *item;
1361
1362 WARN_ON(num_bytes < root->sectorsize);
1363 path = btrfs_alloc_path();
1364 path->reada = 1;
1365 key.objectid = bytenr;
1366 key.offset = num_bytes;
1367 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
1368 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
1369 0, 0);
1370 if (ret < 0)
1371 goto out;
1372 if (ret != 0) {
1373 btrfs_print_leaf(root, path->nodes[0]);
1374 printk(KERN_INFO "btrfs failed to find block number %llu\n",
1375 (unsigned long long)bytenr);
1376 BUG();
1377 }
1378 l = path->nodes[0];
1379 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
1380 *refs = btrfs_extent_refs(l, item);
1381 out:
1382 btrfs_free_path(path);
1383 return 0;
1384 }
1385
1386 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
1387 struct btrfs_root *root, u64 objectid, u64 bytenr)
1388 {
1389 struct btrfs_root *extent_root = root->fs_info->extent_root;
1390 struct btrfs_path *path;
1391 struct extent_buffer *leaf;
1392 struct btrfs_extent_ref *ref_item;
1393 struct btrfs_key key;
1394 struct btrfs_key found_key;
1395 u64 ref_root;
1396 u64 last_snapshot;
1397 u32 nritems;
1398 int ret;
1399
1400 key.objectid = bytenr;
1401 key.offset = (u64)-1;
1402 key.type = BTRFS_EXTENT_ITEM_KEY;
1403
1404 path = btrfs_alloc_path();
1405 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
1406 if (ret < 0)
1407 goto out;
1408 BUG_ON(ret == 0);
1409
1410 ret = -ENOENT;
1411 if (path->slots[0] == 0)
1412 goto out;
1413
1414 path->slots[0]--;
1415 leaf = path->nodes[0];
1416 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1417
1418 if (found_key.objectid != bytenr ||
1419 found_key.type != BTRFS_EXTENT_ITEM_KEY)
1420 goto out;
1421
1422 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
1423 while (1) {
1424 leaf = path->nodes[0];
1425 nritems = btrfs_header_nritems(leaf);
1426 if (path->slots[0] >= nritems) {
1427 ret = btrfs_next_leaf(extent_root, path);
1428 if (ret < 0)
1429 goto out;
1430 if (ret == 0)
1431 continue;
1432 break;
1433 }
1434 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1435 if (found_key.objectid != bytenr)
1436 break;
1437
1438 if (found_key.type != BTRFS_EXTENT_REF_KEY) {
1439 path->slots[0]++;
1440 continue;
1441 }
1442
1443 ref_item = btrfs_item_ptr(leaf, path->slots[0],
1444 struct btrfs_extent_ref);
1445 ref_root = btrfs_ref_root(leaf, ref_item);
1446 if ((ref_root != root->root_key.objectid &&
1447 ref_root != BTRFS_TREE_LOG_OBJECTID) ||
1448 objectid != btrfs_ref_objectid(leaf, ref_item)) {
1449 ret = 1;
1450 goto out;
1451 }
1452 if (btrfs_ref_generation(leaf, ref_item) <= last_snapshot) {
1453 ret = 1;
1454 goto out;
1455 }
1456
1457 path->slots[0]++;
1458 }
1459 ret = 0;
1460 out:
1461 btrfs_free_path(path);
1462 return ret;
1463 }
1464
1465 int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1466 struct extent_buffer *buf, u32 nr_extents)
1467 {
1468 struct btrfs_key key;
1469 struct btrfs_file_extent_item *fi;
1470 u64 root_gen;
1471 u32 nritems;
1472 int i;
1473 int level;
1474 int ret = 0;
1475 int shared = 0;
1476
1477 if (!root->ref_cows)
1478 return 0;
1479
1480 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
1481 shared = 0;
1482 root_gen = root->root_key.offset;
1483 } else {
1484 shared = 1;
1485 root_gen = trans->transid - 1;
1486 }
1487
1488 level = btrfs_header_level(buf);
1489 nritems = btrfs_header_nritems(buf);
1490
1491 if (level == 0) {
1492 struct btrfs_leaf_ref *ref;
1493 struct btrfs_extent_info *info;
1494
1495 ref = btrfs_alloc_leaf_ref(root, nr_extents);
1496 if (!ref) {
1497 ret = -ENOMEM;
1498 goto out;
1499 }
1500
1501 ref->root_gen = root_gen;
1502 ref->bytenr = buf->start;
1503 ref->owner = btrfs_header_owner(buf);
1504 ref->generation = btrfs_header_generation(buf);
1505 ref->nritems = nr_extents;
1506 info = ref->extents;
1507
1508 for (i = 0; nr_extents > 0 && i < nritems; i++) {
1509 u64 disk_bytenr;
1510 btrfs_item_key_to_cpu(buf, &key, i);
1511 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1512 continue;
1513 fi = btrfs_item_ptr(buf, i,
1514 struct btrfs_file_extent_item);
1515 if (btrfs_file_extent_type(buf, fi) ==
1516 BTRFS_FILE_EXTENT_INLINE)
1517 continue;
1518 disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1519 if (disk_bytenr == 0)
1520 continue;
1521
1522 info->bytenr = disk_bytenr;
1523 info->num_bytes =
1524 btrfs_file_extent_disk_num_bytes(buf, fi);
1525 info->objectid = key.objectid;
1526 info->offset = key.offset;
1527 info++;
1528 }
1529
1530 ret = btrfs_add_leaf_ref(root, ref, shared);
1531 if (ret == -EEXIST && shared) {
1532 struct btrfs_leaf_ref *old;
1533 old = btrfs_lookup_leaf_ref(root, ref->bytenr);
1534 BUG_ON(!old);
1535 btrfs_remove_leaf_ref(root, old);
1536 btrfs_free_leaf_ref(root, old);
1537 ret = btrfs_add_leaf_ref(root, ref, shared);
1538 }
1539 WARN_ON(ret);
1540 btrfs_free_leaf_ref(root, ref);
1541 }
1542 out:
1543 return ret;
1544 }
1545
1546 /* when a block goes through cow, we update the reference counts of
1547 * everything that block points to. The internal pointers of the block
1548 * can be in just about any order, and it is likely to have clusters of
1549 * things that are close together and clusters of things that are not.
1550 *
1551 * To help reduce the seeks that come with updating all of these reference
1552 * counts, sort them by byte number before actual updates are done.
1553 *
1554 * struct refsort is used to match byte number to slot in the btree block.
1555 * we sort based on the byte number and then use the slot to actually
1556 * find the item.
1557 *
1558 * struct refsort is smaller than strcut btrfs_item and smaller than
1559 * struct btrfs_key_ptr. Since we're currently limited to the page size
1560 * for a btree block, there's no way for a kmalloc of refsorts for a
1561 * single node to be bigger than a page.
1562 */
1563 struct refsort {
1564 u64 bytenr;
1565 u32 slot;
1566 };
1567
1568 /*
1569 * for passing into sort()
1570 */
1571 static int refsort_cmp(const void *a_void, const void *b_void)
1572 {
1573 const struct refsort *a = a_void;
1574 const struct refsort *b = b_void;
1575
1576 if (a->bytenr < b->bytenr)
1577 return -1;
1578 if (a->bytenr > b->bytenr)
1579 return 1;
1580 return 0;
1581 }
1582
1583
1584 noinline int btrfs_inc_ref(struct btrfs_trans_handle *trans,
1585 struct btrfs_root *root,
1586 struct extent_buffer *orig_buf,
1587 struct extent_buffer *buf, u32 *nr_extents)
1588 {
1589 u64 bytenr;
1590 u64 ref_root;
1591 u64 orig_root;
1592 u64 ref_generation;
1593 u64 orig_generation;
1594 struct refsort *sorted;
1595 u32 nritems;
1596 u32 nr_file_extents = 0;
1597 struct btrfs_key key;
1598 struct btrfs_file_extent_item *fi;
1599 int i;
1600 int level;
1601 int ret = 0;
1602 int faili = 0;
1603 int refi = 0;
1604 int slot;
1605 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
1606 u64, u64, u64, u64, u64, u64, u64, u64);
1607
1608 ref_root = btrfs_header_owner(buf);
1609 ref_generation = btrfs_header_generation(buf);
1610 orig_root = btrfs_header_owner(orig_buf);
1611 orig_generation = btrfs_header_generation(orig_buf);
1612
1613 nritems = btrfs_header_nritems(buf);
1614 level = btrfs_header_level(buf);
1615
1616 sorted = kmalloc(sizeof(struct refsort) * nritems, GFP_NOFS);
1617 BUG_ON(!sorted);
1618
1619 if (root->ref_cows) {
1620 process_func = __btrfs_inc_extent_ref;
1621 } else {
1622 if (level == 0 &&
1623 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
1624 goto out;
1625 if (level != 0 &&
1626 root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
1627 goto out;
1628 process_func = __btrfs_update_extent_ref;
1629 }
1630
1631 /*
1632 * we make two passes through the items. In the first pass we
1633 * only record the byte number and slot. Then we sort based on
1634 * byte number and do the actual work based on the sorted results
1635 */
1636 for (i = 0; i < nritems; i++) {
1637 cond_resched();
1638 if (level == 0) {
1639 btrfs_item_key_to_cpu(buf, &key, i);
1640 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1641 continue;
1642 fi = btrfs_item_ptr(buf, i,
1643 struct btrfs_file_extent_item);
1644 if (btrfs_file_extent_type(buf, fi) ==
1645 BTRFS_FILE_EXTENT_INLINE)
1646 continue;
1647 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1648 if (bytenr == 0)
1649 continue;
1650
1651 nr_file_extents++;
1652 sorted[refi].bytenr = bytenr;
1653 sorted[refi].slot = i;
1654 refi++;
1655 } else {
1656 bytenr = btrfs_node_blockptr(buf, i);
1657 sorted[refi].bytenr = bytenr;
1658 sorted[refi].slot = i;
1659 refi++;
1660 }
1661 }
1662 /*
1663 * if refi == 0, we didn't actually put anything into the sorted
1664 * array and we're done
1665 */
1666 if (refi == 0)
1667 goto out;
1668
1669 sort(sorted, refi, sizeof(struct refsort), refsort_cmp, NULL);
1670
1671 for (i = 0; i < refi; i++) {
1672 cond_resched();
1673 slot = sorted[i].slot;
1674 bytenr = sorted[i].bytenr;
1675
1676 if (level == 0) {
1677 btrfs_item_key_to_cpu(buf, &key, slot);
1678
1679 ret = process_func(trans, root, bytenr,
1680 orig_buf->start, buf->start,
1681 orig_root, ref_root,
1682 orig_generation, ref_generation,
1683 key.objectid);
1684
1685 if (ret) {
1686 faili = slot;
1687 WARN_ON(1);
1688 goto fail;
1689 }
1690 } else {
1691 ret = process_func(trans, root, bytenr,
1692 orig_buf->start, buf->start,
1693 orig_root, ref_root,
1694 orig_generation, ref_generation,
1695 level - 1);
1696 if (ret) {
1697 faili = slot;
1698 WARN_ON(1);
1699 goto fail;
1700 }
1701 }
1702 }
1703 out:
1704 kfree(sorted);
1705 if (nr_extents) {
1706 if (level == 0)
1707 *nr_extents = nr_file_extents;
1708 else
1709 *nr_extents = nritems;
1710 }
1711 return 0;
1712 fail:
1713 kfree(sorted);
1714 WARN_ON(1);
1715 return ret;
1716 }
1717
1718 int btrfs_update_ref(struct btrfs_trans_handle *trans,
1719 struct btrfs_root *root, struct extent_buffer *orig_buf,
1720 struct extent_buffer *buf, int start_slot, int nr)
1721
1722 {
1723 u64 bytenr;
1724 u64 ref_root;
1725 u64 orig_root;
1726 u64 ref_generation;
1727 u64 orig_generation;
1728 struct btrfs_key key;
1729 struct btrfs_file_extent_item *fi;
1730 int i;
1731 int ret;
1732 int slot;
1733 int level;
1734
1735 BUG_ON(start_slot < 0);
1736 BUG_ON(start_slot + nr > btrfs_header_nritems(buf));
1737
1738 ref_root = btrfs_header_owner(buf);
1739 ref_generation = btrfs_header_generation(buf);
1740 orig_root = btrfs_header_owner(orig_buf);
1741 orig_generation = btrfs_header_generation(orig_buf);
1742 level = btrfs_header_level(buf);
1743
1744 if (!root->ref_cows) {
1745 if (level == 0 &&
1746 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
1747 return 0;
1748 if (level != 0 &&
1749 root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
1750 return 0;
1751 }
1752
1753 for (i = 0, slot = start_slot; i < nr; i++, slot++) {
1754 cond_resched();
1755 if (level == 0) {
1756 btrfs_item_key_to_cpu(buf, &key, slot);
1757 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1758 continue;
1759 fi = btrfs_item_ptr(buf, slot,
1760 struct btrfs_file_extent_item);
1761 if (btrfs_file_extent_type(buf, fi) ==
1762 BTRFS_FILE_EXTENT_INLINE)
1763 continue;
1764 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1765 if (bytenr == 0)
1766 continue;
1767 ret = __btrfs_update_extent_ref(trans, root, bytenr,
1768 orig_buf->start, buf->start,
1769 orig_root, ref_root,
1770 orig_generation, ref_generation,
1771 key.objectid);
1772 if (ret)
1773 goto fail;
1774 } else {
1775 bytenr = btrfs_node_blockptr(buf, slot);
1776 ret = __btrfs_update_extent_ref(trans, root, bytenr,
1777 orig_buf->start, buf->start,
1778 orig_root, ref_root,
1779 orig_generation, ref_generation,
1780 level - 1);
1781 if (ret)
1782 goto fail;
1783 }
1784 }
1785 return 0;
1786 fail:
1787 WARN_ON(1);
1788 return -1;
1789 }
1790
1791 static int write_one_cache_group(struct btrfs_trans_handle *trans,
1792 struct btrfs_root *root,
1793 struct btrfs_path *path,
1794 struct btrfs_block_group_cache *cache)
1795 {
1796 int ret;
1797 int pending_ret;
1798 struct btrfs_root *extent_root = root->fs_info->extent_root;
1799 unsigned long bi;
1800 struct extent_buffer *leaf;
1801
1802 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
1803 if (ret < 0)
1804 goto fail;
1805 BUG_ON(ret);
1806
1807 leaf = path->nodes[0];
1808 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
1809 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
1810 btrfs_mark_buffer_dirty(leaf);
1811 btrfs_release_path(extent_root, path);
1812 fail:
1813 finish_current_insert(trans, extent_root, 0);
1814 pending_ret = del_pending_extents(trans, extent_root, 0);
1815 if (ret)
1816 return ret;
1817 if (pending_ret)
1818 return pending_ret;
1819 return 0;
1820
1821 }
1822
1823 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
1824 struct btrfs_root *root)
1825 {
1826 struct btrfs_block_group_cache *cache, *entry;
1827 struct rb_node *n;
1828 int err = 0;
1829 int werr = 0;
1830 struct btrfs_path *path;
1831 u64 last = 0;
1832
1833 path = btrfs_alloc_path();
1834 if (!path)
1835 return -ENOMEM;
1836
1837 while (1) {
1838 cache = NULL;
1839 spin_lock(&root->fs_info->block_group_cache_lock);
1840 for (n = rb_first(&root->fs_info->block_group_cache_tree);
1841 n; n = rb_next(n)) {
1842 entry = rb_entry(n, struct btrfs_block_group_cache,
1843 cache_node);
1844 if (entry->dirty) {
1845 cache = entry;
1846 break;
1847 }
1848 }
1849 spin_unlock(&root->fs_info->block_group_cache_lock);
1850
1851 if (!cache)
1852 break;
1853
1854 cache->dirty = 0;
1855 last += cache->key.offset;
1856
1857 err = write_one_cache_group(trans, root,
1858 path, cache);
1859 /*
1860 * if we fail to write the cache group, we want
1861 * to keep it marked dirty in hopes that a later
1862 * write will work
1863 */
1864 if (err) {
1865 werr = err;
1866 continue;
1867 }
1868 }
1869 btrfs_free_path(path);
1870 return werr;
1871 }
1872
1873 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
1874 {
1875 struct btrfs_block_group_cache *block_group;
1876 int readonly = 0;
1877
1878 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
1879 if (!block_group || block_group->ro)
1880 readonly = 1;
1881 if (block_group)
1882 put_block_group(block_group);
1883 return readonly;
1884 }
1885
1886 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
1887 u64 total_bytes, u64 bytes_used,
1888 struct btrfs_space_info **space_info)
1889 {
1890 struct btrfs_space_info *found;
1891
1892 found = __find_space_info(info, flags);
1893 if (found) {
1894 spin_lock(&found->lock);
1895 found->total_bytes += total_bytes;
1896 found->bytes_used += bytes_used;
1897 found->full = 0;
1898 spin_unlock(&found->lock);
1899 *space_info = found;
1900 return 0;
1901 }
1902 found = kzalloc(sizeof(*found), GFP_NOFS);
1903 if (!found)
1904 return -ENOMEM;
1905
1906 list_add(&found->list, &info->space_info);
1907 INIT_LIST_HEAD(&found->block_groups);
1908 init_rwsem(&found->groups_sem);
1909 spin_lock_init(&found->lock);
1910 found->flags = flags;
1911 found->total_bytes = total_bytes;
1912 found->bytes_used = bytes_used;
1913 found->bytes_pinned = 0;
1914 found->bytes_reserved = 0;
1915 found->bytes_readonly = 0;
1916 found->bytes_delalloc = 0;
1917 found->full = 0;
1918 found->force_alloc = 0;
1919 *space_info = found;
1920 return 0;
1921 }
1922
1923 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
1924 {
1925 u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
1926 BTRFS_BLOCK_GROUP_RAID1 |
1927 BTRFS_BLOCK_GROUP_RAID10 |
1928 BTRFS_BLOCK_GROUP_DUP);
1929 if (extra_flags) {
1930 if (flags & BTRFS_BLOCK_GROUP_DATA)
1931 fs_info->avail_data_alloc_bits |= extra_flags;
1932 if (flags & BTRFS_BLOCK_GROUP_METADATA)
1933 fs_info->avail_metadata_alloc_bits |= extra_flags;
1934 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
1935 fs_info->avail_system_alloc_bits |= extra_flags;
1936 }
1937 }
1938
1939 static void set_block_group_readonly(struct btrfs_block_group_cache *cache)
1940 {
1941 spin_lock(&cache->space_info->lock);
1942 spin_lock(&cache->lock);
1943 if (!cache->ro) {
1944 cache->space_info->bytes_readonly += cache->key.offset -
1945 btrfs_block_group_used(&cache->item);
1946 cache->ro = 1;
1947 }
1948 spin_unlock(&cache->lock);
1949 spin_unlock(&cache->space_info->lock);
1950 }
1951
1952 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
1953 {
1954 u64 num_devices = root->fs_info->fs_devices->rw_devices;
1955
1956 if (num_devices == 1)
1957 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
1958 if (num_devices < 4)
1959 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
1960
1961 if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
1962 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
1963 BTRFS_BLOCK_GROUP_RAID10))) {
1964 flags &= ~BTRFS_BLOCK_GROUP_DUP;
1965 }
1966
1967 if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
1968 (flags & BTRFS_BLOCK_GROUP_RAID10)) {
1969 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
1970 }
1971
1972 if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
1973 ((flags & BTRFS_BLOCK_GROUP_RAID1) |
1974 (flags & BTRFS_BLOCK_GROUP_RAID10) |
1975 (flags & BTRFS_BLOCK_GROUP_DUP)))
1976 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
1977 return flags;
1978 }
1979
1980 static u64 btrfs_get_alloc_profile(struct btrfs_root *root, u64 data)
1981 {
1982 struct btrfs_fs_info *info = root->fs_info;
1983 u64 alloc_profile;
1984
1985 if (data) {
1986 alloc_profile = info->avail_data_alloc_bits &
1987 info->data_alloc_profile;
1988 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
1989 } else if (root == root->fs_info->chunk_root) {
1990 alloc_profile = info->avail_system_alloc_bits &
1991 info->system_alloc_profile;
1992 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
1993 } else {
1994 alloc_profile = info->avail_metadata_alloc_bits &
1995 info->metadata_alloc_profile;
1996 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
1997 }
1998
1999 return btrfs_reduce_alloc_profile(root, data);
2000 }
2001
2002 void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
2003 {
2004 u64 alloc_target;
2005
2006 alloc_target = btrfs_get_alloc_profile(root, 1);
2007 BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
2008 alloc_target);
2009 }
2010
2011 /*
2012 * for now this just makes sure we have at least 5% of our metadata space free
2013 * for use.
2014 */
2015 int btrfs_check_metadata_free_space(struct btrfs_root *root)
2016 {
2017 struct btrfs_fs_info *info = root->fs_info;
2018 struct btrfs_space_info *meta_sinfo;
2019 u64 alloc_target, thresh;
2020
2021 /* get the space info for where the metadata will live */
2022 alloc_target = btrfs_get_alloc_profile(root, 0);
2023 meta_sinfo = __find_space_info(info, alloc_target);
2024
2025 /*
2026 * if the metadata area isn't maxed out then there is no sense in
2027 * checking how much is used, since we can always allocate a new chunk
2028 */
2029 if (!meta_sinfo->full)
2030 return 0;
2031
2032 spin_lock(&meta_sinfo->lock);
2033 thresh = meta_sinfo->total_bytes * 95;
2034
2035 do_div(thresh, 100);
2036
2037 if (meta_sinfo->bytes_used + meta_sinfo->bytes_reserved +
2038 meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly > thresh) {
2039 spin_unlock(&meta_sinfo->lock);
2040 return -ENOSPC;
2041 }
2042 spin_unlock(&meta_sinfo->lock);
2043
2044 return 0;
2045 }
2046
2047 /*
2048 * This will check the space that the inode allocates from to make sure we have
2049 * enough space for bytes.
2050 */
2051 int btrfs_check_data_free_space(struct btrfs_root *root, struct inode *inode,
2052 u64 bytes)
2053 {
2054 struct btrfs_space_info *data_sinfo;
2055 int ret = 0;
2056
2057 /* make sure bytes are sectorsize aligned */
2058 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
2059
2060 data_sinfo = BTRFS_I(inode)->space_info;
2061 again:
2062 /* make sure we have enough space to handle the data first */
2063 spin_lock(&data_sinfo->lock);
2064 if (data_sinfo->total_bytes - data_sinfo->bytes_used -
2065 data_sinfo->bytes_delalloc - data_sinfo->bytes_reserved -
2066 data_sinfo->bytes_pinned - data_sinfo->bytes_readonly -
2067 data_sinfo->bytes_may_use < bytes) {
2068 /*
2069 * if we don't have enough free bytes in this space then we need
2070 * to alloc a new chunk.
2071 */
2072 if (!data_sinfo->full) {
2073 u64 alloc_target;
2074 struct btrfs_trans_handle *trans;
2075
2076 data_sinfo->force_alloc = 1;
2077 spin_unlock(&data_sinfo->lock);
2078
2079 alloc_target = btrfs_get_alloc_profile(root, 1);
2080 trans = btrfs_start_transaction(root, 1);
2081 if (!trans)
2082 return -ENOMEM;
2083
2084 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2085 bytes + 2 * 1024 * 1024,
2086 alloc_target, 0);
2087 btrfs_end_transaction(trans, root);
2088 if (ret)
2089 return ret;
2090 goto again;
2091 }
2092 spin_unlock(&data_sinfo->lock);
2093 printk(KERN_ERR "no space left, need %llu, %llu delalloc bytes"
2094 ", %llu bytes_used, %llu bytes_reserved, "
2095 "%llu bytes_pinned, %llu bytes_readonly, %llu may use"
2096 "%llu total\n", bytes, data_sinfo->bytes_delalloc,
2097 data_sinfo->bytes_used, data_sinfo->bytes_reserved,
2098 data_sinfo->bytes_pinned, data_sinfo->bytes_readonly,
2099 data_sinfo->bytes_may_use, data_sinfo->total_bytes);
2100 return -ENOSPC;
2101 }
2102 data_sinfo->bytes_may_use += bytes;
2103 BTRFS_I(inode)->reserved_bytes += bytes;
2104 spin_unlock(&data_sinfo->lock);
2105
2106 return btrfs_check_metadata_free_space(root);
2107 }
2108
2109 /*
2110 * if there was an error for whatever reason after calling
2111 * btrfs_check_data_free_space, call this so we can cleanup the counters.
2112 */
2113 void btrfs_free_reserved_data_space(struct btrfs_root *root,
2114 struct inode *inode, u64 bytes)
2115 {
2116 struct btrfs_space_info *data_sinfo;
2117
2118 /* make sure bytes are sectorsize aligned */
2119 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
2120
2121 data_sinfo = BTRFS_I(inode)->space_info;
2122 spin_lock(&data_sinfo->lock);
2123 data_sinfo->bytes_may_use -= bytes;
2124 BTRFS_I(inode)->reserved_bytes -= bytes;
2125 spin_unlock(&data_sinfo->lock);
2126 }
2127
2128 /* called when we are adding a delalloc extent to the inode's io_tree */
2129 void btrfs_delalloc_reserve_space(struct btrfs_root *root, struct inode *inode,
2130 u64 bytes)
2131 {
2132 struct btrfs_space_info *data_sinfo;
2133
2134 /* get the space info for where this inode will be storing its data */
2135 data_sinfo = BTRFS_I(inode)->space_info;
2136
2137 /* make sure we have enough space to handle the data first */
2138 spin_lock(&data_sinfo->lock);
2139 data_sinfo->bytes_delalloc += bytes;
2140
2141 /*
2142 * we are adding a delalloc extent without calling
2143 * btrfs_check_data_free_space first. This happens on a weird
2144 * writepage condition, but shouldn't hurt our accounting
2145 */
2146 if (unlikely(bytes > BTRFS_I(inode)->reserved_bytes)) {
2147 data_sinfo->bytes_may_use -= BTRFS_I(inode)->reserved_bytes;
2148 BTRFS_I(inode)->reserved_bytes = 0;
2149 } else {
2150 data_sinfo->bytes_may_use -= bytes;
2151 BTRFS_I(inode)->reserved_bytes -= bytes;
2152 }
2153
2154 spin_unlock(&data_sinfo->lock);
2155 }
2156
2157 /* called when we are clearing an delalloc extent from the inode's io_tree */
2158 void btrfs_delalloc_free_space(struct btrfs_root *root, struct inode *inode,
2159 u64 bytes)
2160 {
2161 struct btrfs_space_info *info;
2162
2163 info = BTRFS_I(inode)->space_info;
2164
2165 spin_lock(&info->lock);
2166 info->bytes_delalloc -= bytes;
2167 spin_unlock(&info->lock);
2168 }
2169
2170 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
2171 struct btrfs_root *extent_root, u64 alloc_bytes,
2172 u64 flags, int force)
2173 {
2174 struct btrfs_space_info *space_info;
2175 u64 thresh;
2176 int ret = 0;
2177
2178 mutex_lock(&extent_root->fs_info->chunk_mutex);
2179
2180 flags = btrfs_reduce_alloc_profile(extent_root, flags);
2181
2182 space_info = __find_space_info(extent_root->fs_info, flags);
2183 if (!space_info) {
2184 ret = update_space_info(extent_root->fs_info, flags,
2185 0, 0, &space_info);
2186 BUG_ON(ret);
2187 }
2188 BUG_ON(!space_info);
2189
2190 spin_lock(&space_info->lock);
2191 if (space_info->force_alloc) {
2192 force = 1;
2193 space_info->force_alloc = 0;
2194 }
2195 if (space_info->full) {
2196 spin_unlock(&space_info->lock);
2197 goto out;
2198 }
2199
2200 thresh = space_info->total_bytes - space_info->bytes_readonly;
2201 thresh = div_factor(thresh, 6);
2202 if (!force &&
2203 (space_info->bytes_used + space_info->bytes_pinned +
2204 space_info->bytes_reserved + alloc_bytes) < thresh) {
2205 spin_unlock(&space_info->lock);
2206 goto out;
2207 }
2208 spin_unlock(&space_info->lock);
2209
2210 ret = btrfs_alloc_chunk(trans, extent_root, flags);
2211 if (ret)
2212 space_info->full = 1;
2213 out:
2214 mutex_unlock(&extent_root->fs_info->chunk_mutex);
2215 return ret;
2216 }
2217
2218 static int update_block_group(struct btrfs_trans_handle *trans,
2219 struct btrfs_root *root,
2220 u64 bytenr, u64 num_bytes, int alloc,
2221 int mark_free)
2222 {
2223 struct btrfs_block_group_cache *cache;
2224 struct btrfs_fs_info *info = root->fs_info;
2225 u64 total = num_bytes;
2226 u64 old_val;
2227 u64 byte_in_group;
2228
2229 while (total) {
2230 cache = btrfs_lookup_block_group(info, bytenr);
2231 if (!cache)
2232 return -1;
2233 byte_in_group = bytenr - cache->key.objectid;
2234 WARN_ON(byte_in_group > cache->key.offset);
2235
2236 spin_lock(&cache->space_info->lock);
2237 spin_lock(&cache->lock);
2238 cache->dirty = 1;
2239 old_val = btrfs_block_group_used(&cache->item);
2240 num_bytes = min(total, cache->key.offset - byte_in_group);
2241 if (alloc) {
2242 old_val += num_bytes;
2243 cache->space_info->bytes_used += num_bytes;
2244 if (cache->ro)
2245 cache->space_info->bytes_readonly -= num_bytes;
2246 btrfs_set_block_group_used(&cache->item, old_val);
2247 spin_unlock(&cache->lock);
2248 spin_unlock(&cache->space_info->lock);
2249 } else {
2250 old_val -= num_bytes;
2251 cache->space_info->bytes_used -= num_bytes;
2252 if (cache->ro)
2253 cache->space_info->bytes_readonly += num_bytes;
2254 btrfs_set_block_group_used(&cache->item, old_val);
2255 spin_unlock(&cache->lock);
2256 spin_unlock(&cache->space_info->lock);
2257 if (mark_free) {
2258 int ret;
2259
2260 ret = btrfs_discard_extent(root, bytenr,
2261 num_bytes);
2262 WARN_ON(ret);
2263
2264 ret = btrfs_add_free_space(cache, bytenr,
2265 num_bytes);
2266 WARN_ON(ret);
2267 }
2268 }
2269 put_block_group(cache);
2270 total -= num_bytes;
2271 bytenr += num_bytes;
2272 }
2273 return 0;
2274 }
2275
2276 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
2277 {
2278 struct btrfs_block_group_cache *cache;
2279 u64 bytenr;
2280
2281 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
2282 if (!cache)
2283 return 0;
2284
2285 bytenr = cache->key.objectid;
2286 put_block_group(cache);
2287
2288 return bytenr;
2289 }
2290
2291 int btrfs_update_pinned_extents(struct btrfs_root *root,
2292 u64 bytenr, u64 num, int pin)
2293 {
2294 u64 len;
2295 struct btrfs_block_group_cache *cache;
2296 struct btrfs_fs_info *fs_info = root->fs_info;
2297
2298 WARN_ON(!mutex_is_locked(&root->fs_info->pinned_mutex));
2299 if (pin) {
2300 set_extent_dirty(&fs_info->pinned_extents,
2301 bytenr, bytenr + num - 1, GFP_NOFS);
2302 } else {
2303 clear_extent_dirty(&fs_info->pinned_extents,
2304 bytenr, bytenr + num - 1, GFP_NOFS);
2305 }
2306 while (num > 0) {
2307 cache = btrfs_lookup_block_group(fs_info, bytenr);
2308 BUG_ON(!cache);
2309 len = min(num, cache->key.offset -
2310 (bytenr - cache->key.objectid));
2311 if (pin) {
2312 spin_lock(&cache->space_info->lock);
2313 spin_lock(&cache->lock);
2314 cache->pinned += len;
2315 cache->space_info->bytes_pinned += len;
2316 spin_unlock(&cache->lock);
2317 spin_unlock(&cache->space_info->lock);
2318 fs_info->total_pinned += len;
2319 } else {
2320 spin_lock(&cache->space_info->lock);
2321 spin_lock(&cache->lock);
2322 cache->pinned -= len;
2323 cache->space_info->bytes_pinned -= len;
2324 spin_unlock(&cache->lock);
2325 spin_unlock(&cache->space_info->lock);
2326 fs_info->total_pinned -= len;
2327 if (cache->cached)
2328 btrfs_add_free_space(cache, bytenr, len);
2329 }
2330 put_block_group(cache);
2331 bytenr += len;
2332 num -= len;
2333 }
2334 return 0;
2335 }
2336
2337 static int update_reserved_extents(struct btrfs_root *root,
2338 u64 bytenr, u64 num, int reserve)
2339 {
2340 u64 len;
2341 struct btrfs_block_group_cache *cache;
2342 struct btrfs_fs_info *fs_info = root->fs_info;
2343
2344 while (num > 0) {
2345 cache = btrfs_lookup_block_group(fs_info, bytenr);
2346 BUG_ON(!cache);
2347 len = min(num, cache->key.offset -
2348 (bytenr - cache->key.objectid));
2349
2350 spin_lock(&cache->space_info->lock);
2351 spin_lock(&cache->lock);
2352 if (reserve) {
2353 cache->reserved += len;
2354 cache->space_info->bytes_reserved += len;
2355 } else {
2356 cache->reserved -= len;
2357 cache->space_info->bytes_reserved -= len;
2358 }
2359 spin_unlock(&cache->lock);
2360 spin_unlock(&cache->space_info->lock);
2361 put_block_group(cache);
2362 bytenr += len;
2363 num -= len;
2364 }
2365 return 0;
2366 }
2367
2368 int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
2369 {
2370 u64 last = 0;
2371 u64 start;
2372 u64 end;
2373 struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
2374 int ret;
2375
2376 mutex_lock(&root->fs_info->pinned_mutex);
2377 while (1) {
2378 ret = find_first_extent_bit(pinned_extents, last,
2379 &start, &end, EXTENT_DIRTY);
2380 if (ret)
2381 break;
2382 set_extent_dirty(copy, start, end, GFP_NOFS);
2383 last = end + 1;
2384 }
2385 mutex_unlock(&root->fs_info->pinned_mutex);
2386 return 0;
2387 }
2388
2389 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
2390 struct btrfs_root *root,
2391 struct extent_io_tree *unpin)
2392 {
2393 u64 start;
2394 u64 end;
2395 int ret;
2396
2397 mutex_lock(&root->fs_info->pinned_mutex);
2398 while (1) {
2399 ret = find_first_extent_bit(unpin, 0, &start, &end,
2400 EXTENT_DIRTY);
2401 if (ret)
2402 break;
2403
2404 ret = btrfs_discard_extent(root, start, end + 1 - start);
2405
2406 btrfs_update_pinned_extents(root, start, end + 1 - start, 0);
2407 clear_extent_dirty(unpin, start, end, GFP_NOFS);
2408
2409 if (need_resched()) {
2410 mutex_unlock(&root->fs_info->pinned_mutex);
2411 cond_resched();
2412 mutex_lock(&root->fs_info->pinned_mutex);
2413 }
2414 }
2415 mutex_unlock(&root->fs_info->pinned_mutex);
2416 return ret;
2417 }
2418
2419 static int finish_current_insert(struct btrfs_trans_handle *trans,
2420 struct btrfs_root *extent_root, int all)
2421 {
2422 u64 start;
2423 u64 end;
2424 u64 priv;
2425 u64 search = 0;
2426 struct btrfs_fs_info *info = extent_root->fs_info;
2427 struct btrfs_path *path;
2428 struct pending_extent_op *extent_op, *tmp;
2429 struct list_head insert_list, update_list;
2430 int ret;
2431 int num_inserts = 0, max_inserts, restart = 0;
2432
2433 path = btrfs_alloc_path();
2434 INIT_LIST_HEAD(&insert_list);
2435 INIT_LIST_HEAD(&update_list);
2436
2437 max_inserts = extent_root->leafsize /
2438 (2 * sizeof(struct btrfs_key) + 2 * sizeof(struct btrfs_item) +
2439 sizeof(struct btrfs_extent_ref) +
2440 sizeof(struct btrfs_extent_item));
2441 again:
2442 mutex_lock(&info->extent_ins_mutex);
2443 while (1) {
2444 ret = find_first_extent_bit(&info->extent_ins, search, &start,
2445 &end, EXTENT_WRITEBACK);
2446 if (ret) {
2447 if (restart && !num_inserts &&
2448 list_empty(&update_list)) {
2449 restart = 0;
2450 search = 0;
2451 continue;
2452 }
2453 break;
2454 }
2455
2456 ret = try_lock_extent(&info->extent_ins, start, end, GFP_NOFS);
2457 if (!ret) {
2458 if (all)
2459 restart = 1;
2460 search = end + 1;
2461 if (need_resched()) {
2462 mutex_unlock(&info->extent_ins_mutex);
2463 cond_resched();
2464 mutex_lock(&info->extent_ins_mutex);
2465 }
2466 continue;
2467 }
2468
2469 ret = get_state_private(&info->extent_ins, start, &priv);
2470 BUG_ON(ret);
2471 extent_op = (struct pending_extent_op *)(unsigned long) priv;
2472
2473 if (extent_op->type == PENDING_EXTENT_INSERT) {
2474 num_inserts++;
2475 list_add_tail(&extent_op->list, &insert_list);
2476 search = end + 1;
2477 if (num_inserts == max_inserts) {
2478 restart = 1;
2479 break;
2480 }
2481 } else if (extent_op->type == PENDING_BACKREF_UPDATE) {
2482 list_add_tail(&extent_op->list, &update_list);
2483 search = end + 1;
2484 } else {
2485 BUG();
2486 }
2487 }
2488
2489 /*
2490 * process the update list, clear the writeback bit for it, and if
2491 * somebody marked this thing for deletion then just unlock it and be
2492 * done, the free_extents will handle it
2493 */
2494 list_for_each_entry_safe(extent_op, tmp, &update_list, list) {
2495 clear_extent_bits(&info->extent_ins, extent_op->bytenr,
2496 extent_op->bytenr + extent_op->num_bytes - 1,
2497 EXTENT_WRITEBACK, GFP_NOFS);
2498 if (extent_op->del) {
2499 list_del_init(&extent_op->list);
2500 unlock_extent(&info->extent_ins, extent_op->bytenr,
2501 extent_op->bytenr + extent_op->num_bytes
2502 - 1, GFP_NOFS);
2503 kfree(extent_op);
2504 }
2505 }
2506 mutex_unlock(&info->extent_ins_mutex);
2507
2508 /*
2509 * still have things left on the update list, go ahead an update
2510 * everything
2511 */
2512 if (!list_empty(&update_list)) {
2513 ret = update_backrefs(trans, extent_root, path, &update_list);
2514 BUG_ON(ret);
2515
2516 /* we may have COW'ed new blocks, so lets start over */
2517 if (all)
2518 restart = 1;
2519 }
2520
2521 /*
2522 * if no inserts need to be done, but we skipped some extents and we
2523 * need to make sure everything is cleaned then reset everything and
2524 * go back to the beginning
2525 */
2526 if (!num_inserts && restart) {
2527 search = 0;
2528 restart = 0;
2529 INIT_LIST_HEAD(&update_list);
2530 INIT_LIST_HEAD(&insert_list);
2531 goto again;
2532 } else if (!num_inserts) {
2533 goto out;
2534 }
2535
2536 /*
2537 * process the insert extents list. Again if we are deleting this
2538 * extent, then just unlock it, pin down the bytes if need be, and be
2539 * done with it. Saves us from having to actually insert the extent
2540 * into the tree and then subsequently come along and delete it
2541 */
2542 mutex_lock(&info->extent_ins_mutex);
2543 list_for_each_entry_safe(extent_op, tmp, &insert_list, list) {
2544 clear_extent_bits(&info->extent_ins, extent_op->bytenr,
2545 extent_op->bytenr + extent_op->num_bytes - 1,
2546 EXTENT_WRITEBACK, GFP_NOFS);
2547 if (extent_op->del) {
2548 u64 used;
2549 list_del_init(&extent_op->list);
2550 unlock_extent(&info->extent_ins, extent_op->bytenr,
2551 extent_op->bytenr + extent_op->num_bytes
2552 - 1, GFP_NOFS);
2553
2554 mutex_lock(&extent_root->fs_info->pinned_mutex);
2555 ret = pin_down_bytes(trans, extent_root,
2556 extent_op->bytenr,
2557 extent_op->num_bytes, 0);
2558 mutex_unlock(&extent_root->fs_info->pinned_mutex);
2559
2560 spin_lock(&info->delalloc_lock);
2561 used = btrfs_super_bytes_used(&info->super_copy);
2562 btrfs_set_super_bytes_used(&info->super_copy,
2563 used - extent_op->num_bytes);
2564 used = btrfs_root_used(&extent_root->root_item);
2565 btrfs_set_root_used(&extent_root->root_item,
2566 used - extent_op->num_bytes);
2567 spin_unlock(&info->delalloc_lock);
2568
2569 ret = update_block_group(trans, extent_root,
2570 extent_op->bytenr,
2571 extent_op->num_bytes,
2572 0, ret > 0);
2573 BUG_ON(ret);
2574 kfree(extent_op);
2575 num_inserts--;
2576 }
2577 }
2578 mutex_unlock(&info->extent_ins_mutex);
2579
2580 ret = insert_extents(trans, extent_root, path, &insert_list,
2581 num_inserts);
2582 BUG_ON(ret);
2583
2584 /*
2585 * if restart is set for whatever reason we need to go back and start
2586 * searching through the pending list again.
2587 *
2588 * We just inserted some extents, which could have resulted in new
2589 * blocks being allocated, which would result in new blocks needing
2590 * updates, so if all is set we _must_ restart to get the updated
2591 * blocks.
2592 */
2593 if (restart || all) {
2594 INIT_LIST_HEAD(&insert_list);
2595 INIT_LIST_HEAD(&update_list);
2596 search = 0;
2597 restart = 0;
2598 num_inserts = 0;
2599 goto again;
2600 }
2601 out:
2602 btrfs_free_path(path);
2603 return 0;
2604 }
2605
2606 static int pin_down_bytes(struct btrfs_trans_handle *trans,
2607 struct btrfs_root *root,
2608 u64 bytenr, u64 num_bytes, int is_data)
2609 {
2610 int err = 0;
2611 struct extent_buffer *buf;
2612
2613 if (is_data)
2614 goto pinit;
2615
2616 buf = btrfs_find_tree_block(root, bytenr, num_bytes);
2617 if (!buf)
2618 goto pinit;
2619
2620 /* we can reuse a block if it hasn't been written
2621 * and it is from this transaction. We can't
2622 * reuse anything from the tree log root because
2623 * it has tiny sub-transactions.
2624 */
2625 if (btrfs_buffer_uptodate(buf, 0) &&
2626 btrfs_try_tree_lock(buf)) {
2627 u64 header_owner = btrfs_header_owner(buf);
2628 u64 header_transid = btrfs_header_generation(buf);
2629 if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
2630 header_owner != BTRFS_TREE_RELOC_OBJECTID &&
2631 header_transid == trans->transid &&
2632 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
2633 clean_tree_block(NULL, root, buf);
2634 btrfs_tree_unlock(buf);
2635 free_extent_buffer(buf);
2636 return 1;
2637 }
2638 btrfs_tree_unlock(buf);
2639 }
2640 free_extent_buffer(buf);
2641 pinit:
2642 btrfs_update_pinned_extents(root, bytenr, num_bytes, 1);
2643
2644 BUG_ON(err < 0);
2645 return 0;
2646 }
2647
2648 /*
2649 * remove an extent from the root, returns 0 on success
2650 */
2651 static int __free_extent(struct btrfs_trans_handle *trans,
2652 struct btrfs_root *root,
2653 u64 bytenr, u64 num_bytes, u64 parent,
2654 u64 root_objectid, u64 ref_generation,
2655 u64 owner_objectid, int pin, int mark_free)
2656 {
2657 struct btrfs_path *path;
2658 struct btrfs_key key;
2659 struct btrfs_fs_info *info = root->fs_info;
2660 struct btrfs_root *extent_root = info->extent_root;
2661 struct extent_buffer *leaf;
2662 int ret;
2663 int extent_slot = 0;
2664 int found_extent = 0;
2665 int num_to_del = 1;
2666 struct btrfs_extent_item *ei;
2667 u32 refs;
2668
2669 key.objectid = bytenr;
2670 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
2671 key.offset = num_bytes;
2672 path = btrfs_alloc_path();
2673 if (!path)
2674 return -ENOMEM;
2675
2676 path->reada = 1;
2677 ret = lookup_extent_backref(trans, extent_root, path,
2678 bytenr, parent, root_objectid,
2679 ref_generation, owner_objectid, 1);
2680 if (ret == 0) {
2681 struct btrfs_key found_key;
2682 extent_slot = path->slots[0];
2683 while (extent_slot > 0) {
2684 extent_slot--;
2685 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2686 extent_slot);
2687 if (found_key.objectid != bytenr)
2688 break;
2689 if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
2690 found_key.offset == num_bytes) {
2691 found_extent = 1;
2692 break;
2693 }
2694 if (path->slots[0] - extent_slot > 5)
2695 break;
2696 }
2697 if (!found_extent) {
2698 ret = remove_extent_backref(trans, extent_root, path);
2699 BUG_ON(ret);
2700 btrfs_release_path(extent_root, path);
2701 ret = btrfs_search_slot(trans, extent_root,
2702 &key, path, -1, 1);
2703 if (ret) {
2704 printk(KERN_ERR "umm, got %d back from search"
2705 ", was looking for %llu\n", ret,
2706 (unsigned long long)bytenr);
2707 btrfs_print_leaf(extent_root, path->nodes[0]);
2708 }
2709 BUG_ON(ret);
2710 extent_slot = path->slots[0];
2711 }
2712 } else {
2713 btrfs_print_leaf(extent_root, path->nodes[0]);
2714 WARN_ON(1);
2715 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
2716 "root %llu gen %llu owner %llu\n",
2717 (unsigned long long)bytenr,
2718 (unsigned long long)root_objectid,
2719 (unsigned long long)ref_generation,
2720 (unsigned long long)owner_objectid);
2721 }
2722
2723 leaf = path->nodes[0];
2724 ei = btrfs_item_ptr(leaf, extent_slot,
2725 struct btrfs_extent_item);
2726 refs = btrfs_extent_refs(leaf, ei);
2727 BUG_ON(refs == 0);
2728 refs -= 1;
2729 btrfs_set_extent_refs(leaf, ei, refs);
2730
2731 btrfs_mark_buffer_dirty(leaf);
2732
2733 if (refs == 0 && found_extent && path->slots[0] == extent_slot + 1) {
2734 struct btrfs_extent_ref *ref;
2735 ref = btrfs_item_ptr(leaf, path->slots[0],
2736 struct btrfs_extent_ref);
2737 BUG_ON(btrfs_ref_num_refs(leaf, ref) != 1);
2738 /* if the back ref and the extent are next to each other
2739 * they get deleted below in one shot
2740 */
2741 path->slots[0] = extent_slot;
2742 num_to_del = 2;
2743 } else if (found_extent) {
2744 /* otherwise delete the extent back ref */
2745 ret = remove_extent_backref(trans, extent_root, path);
2746 BUG_ON(ret);
2747 /* if refs are 0, we need to setup the path for deletion */
2748 if (refs == 0) {
2749 btrfs_release_path(extent_root, path);
2750 ret = btrfs_search_slot(trans, extent_root, &key, path,
2751 -1, 1);
2752 BUG_ON(ret);
2753 }
2754 }
2755
2756 if (refs == 0) {
2757 u64 super_used;
2758 u64 root_used;
2759
2760 if (pin) {
2761 mutex_lock(&root->fs_info->pinned_mutex);
2762 ret = pin_down_bytes(trans, root, bytenr, num_bytes,
2763 owner_objectid >= BTRFS_FIRST_FREE_OBJECTID);
2764 mutex_unlock(&root->fs_info->pinned_mutex);
2765 if (ret > 0)
2766 mark_free = 1;
2767 BUG_ON(ret < 0);
2768 }
2769 /* block accounting for super block */
2770 spin_lock(&info->delalloc_lock);
2771 super_used = btrfs_super_bytes_used(&info->super_copy);
2772 btrfs_set_super_bytes_used(&info->super_copy,
2773 super_used - num_bytes);
2774
2775 /* block accounting for root item */
2776 root_used = btrfs_root_used(&root->root_item);
2777 btrfs_set_root_used(&root->root_item,
2778 root_used - num_bytes);
2779 spin_unlock(&info->delalloc_lock);
2780 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
2781 num_to_del);
2782 BUG_ON(ret);
2783 btrfs_release_path(extent_root, path);
2784
2785 if (owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
2786 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
2787 BUG_ON(ret);
2788 }
2789
2790 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
2791 mark_free);
2792 BUG_ON(ret);
2793 }
2794 btrfs_free_path(path);
2795 finish_current_insert(trans, extent_root, 0);
2796 return ret;
2797 }
2798
2799 /*
2800 * find all the blocks marked as pending in the radix tree and remove
2801 * them from the extent map
2802 */
2803 static int del_pending_extents(struct btrfs_trans_handle *trans,
2804 struct btrfs_root *extent_root, int all)
2805 {
2806 int ret;
2807 int err = 0;
2808 u64 start;
2809 u64 end;
2810 u64 priv;
2811 u64 search = 0;
2812 int nr = 0, skipped = 0;
2813 struct extent_io_tree *pending_del;
2814 struct extent_io_tree *extent_ins;
2815 struct pending_extent_op *extent_op;
2816 struct btrfs_fs_info *info = extent_root->fs_info;
2817 struct list_head delete_list;
2818
2819 INIT_LIST_HEAD(&delete_list);
2820 extent_ins = &extent_root->fs_info->extent_ins;
2821 pending_del = &extent_root->fs_info->pending_del;
2822
2823 again:
2824 mutex_lock(&info->extent_ins_mutex);
2825 while (1) {
2826 ret = find_first_extent_bit(pending_del, search, &start, &end,
2827 EXTENT_WRITEBACK);
2828 if (ret) {
2829 if (all && skipped && !nr) {
2830 search = 0;
2831 skipped = 0;
2832 continue;
2833 }
2834 mutex_unlock(&info->extent_ins_mutex);
2835 break;
2836 }
2837
2838 ret = try_lock_extent(extent_ins, start, end, GFP_NOFS);
2839 if (!ret) {
2840 search = end+1;
2841 skipped = 1;
2842
2843 if (need_resched()) {
2844 mutex_unlock(&info->extent_ins_mutex);
2845 cond_resched();
2846 mutex_lock(&info->extent_ins_mutex);
2847 }
2848
2849 continue;
2850 }
2851 BUG_ON(ret < 0);
2852
2853 ret = get_state_private(pending_del, start, &priv);
2854 BUG_ON(ret);
2855 extent_op = (struct pending_extent_op *)(unsigned long)priv;
2856
2857 clear_extent_bits(pending_del, start, end, EXTENT_WRITEBACK,
2858 GFP_NOFS);
2859 if (!test_range_bit(extent_ins, start, end,
2860 EXTENT_WRITEBACK, 0)) {
2861 list_add_tail(&extent_op->list, &delete_list);
2862 nr++;
2863 } else {
2864 kfree(extent_op);
2865
2866 ret = get_state_private(&info->extent_ins, start,
2867 &priv);
2868 BUG_ON(ret);
2869 extent_op = (struct pending_extent_op *)
2870 (unsigned long)priv;
2871
2872 clear_extent_bits(&info->extent_ins, start, end,
2873 EXTENT_WRITEBACK, GFP_NOFS);
2874
2875 if (extent_op->type == PENDING_BACKREF_UPDATE) {
2876 list_add_tail(&extent_op->list, &delete_list);
2877 search = end + 1;
2878 nr++;
2879 continue;
2880 }
2881
2882 mutex_lock(&extent_root->fs_info->pinned_mutex);
2883 ret = pin_down_bytes(trans, extent_root, start,
2884 end + 1 - start, 0);
2885 mutex_unlock(&extent_root->fs_info->pinned_mutex);
2886
2887 ret = update_block_group(trans, extent_root, start,
2888 end + 1 - start, 0, ret > 0);
2889
2890 unlock_extent(extent_ins, start, end, GFP_NOFS);
2891 BUG_ON(ret);
2892 kfree(extent_op);
2893 }
2894 if (ret)
2895 err = ret;
2896
2897 search = end + 1;
2898
2899 if (need_resched()) {
2900 mutex_unlock(&info->extent_ins_mutex);
2901 cond_resched();
2902 mutex_lock(&info->extent_ins_mutex);
2903 }
2904 }
2905
2906 if (nr) {
2907 ret = free_extents(trans, extent_root, &delete_list);
2908 BUG_ON(ret);
2909 }
2910
2911 if (all && skipped) {
2912 INIT_LIST_HEAD(&delete_list);
2913 search = 0;
2914 nr = 0;
2915 goto again;
2916 }
2917
2918 if (!err)
2919 finish_current_insert(trans, extent_root, 0);
2920 return err;
2921 }
2922
2923 /*
2924 * remove an extent from the root, returns 0 on success
2925 */
2926 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
2927 struct btrfs_root *root,
2928 u64 bytenr, u64 num_bytes, u64 parent,
2929 u64 root_objectid, u64 ref_generation,
2930 u64 owner_objectid, int pin)
2931 {
2932 struct btrfs_root *extent_root = root->fs_info->extent_root;
2933 int pending_ret;
2934 int ret;
2935
2936 WARN_ON(num_bytes < root->sectorsize);
2937 if (root == extent_root) {
2938 struct pending_extent_op *extent_op = NULL;
2939
2940 mutex_lock(&root->fs_info->extent_ins_mutex);
2941 if (test_range_bit(&root->fs_info->extent_ins, bytenr,
2942 bytenr + num_bytes - 1, EXTENT_WRITEBACK, 0)) {
2943 u64 priv;
2944 ret = get_state_private(&root->fs_info->extent_ins,
2945 bytenr, &priv);
2946 BUG_ON(ret);
2947 extent_op = (struct pending_extent_op *)
2948 (unsigned long)priv;
2949
2950 extent_op->del = 1;
2951 if (extent_op->type == PENDING_EXTENT_INSERT) {
2952 mutex_unlock(&root->fs_info->extent_ins_mutex);
2953 return 0;
2954 }
2955 }
2956
2957 if (extent_op) {
2958 ref_generation = extent_op->orig_generation;
2959 parent = extent_op->orig_parent;
2960 }
2961
2962 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2963 BUG_ON(!extent_op);
2964
2965 extent_op->type = PENDING_EXTENT_DELETE;
2966 extent_op->bytenr = bytenr;
2967 extent_op->num_bytes = num_bytes;
2968 extent_op->parent = parent;
2969 extent_op->orig_parent = parent;
2970 extent_op->generation = ref_generation;
2971 extent_op->orig_generation = ref_generation;
2972 extent_op->level = (int)owner_objectid;
2973 INIT_LIST_HEAD(&extent_op->list);
2974 extent_op->del = 0;
2975
2976 set_extent_bits(&root->fs_info->pending_del,
2977 bytenr, bytenr + num_bytes - 1,
2978 EXTENT_WRITEBACK, GFP_NOFS);
2979 set_state_private(&root->fs_info->pending_del,
2980 bytenr, (unsigned long)extent_op);
2981 mutex_unlock(&root->fs_info->extent_ins_mutex);
2982 return 0;
2983 }
2984 /* if metadata always pin */
2985 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
2986 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
2987 mutex_lock(&root->fs_info->pinned_mutex);
2988 btrfs_update_pinned_extents(root, bytenr, num_bytes, 1);
2989 mutex_unlock(&root->fs_info->pinned_mutex);
2990 update_reserved_extents(root, bytenr, num_bytes, 0);
2991 return 0;
2992 }
2993 pin = 1;
2994 }
2995
2996 /* if data pin when any transaction has committed this */
2997 if (ref_generation != trans->transid)
2998 pin = 1;
2999
3000 ret = __free_extent(trans, root, bytenr, num_bytes, parent,
3001 root_objectid, ref_generation,
3002 owner_objectid, pin, pin == 0);
3003
3004 finish_current_insert(trans, root->fs_info->extent_root, 0);
3005 pending_ret = del_pending_extents(trans, root->fs_info->extent_root, 0);
3006 return ret ? ret : pending_ret;
3007 }
3008
3009 int btrfs_free_extent(struct btrfs_trans_handle *trans,
3010 struct btrfs_root *root,
3011 u64 bytenr, u64 num_bytes, u64 parent,
3012 u64 root_objectid, u64 ref_generation,
3013 u64 owner_objectid, int pin)
3014 {
3015 int ret;
3016
3017 ret = __btrfs_free_extent(trans, root, bytenr, num_bytes, parent,
3018 root_objectid, ref_generation,
3019 owner_objectid, pin);
3020 return ret;
3021 }
3022
3023 static u64 stripe_align(struct btrfs_root *root, u64 val)
3024 {
3025 u64 mask = ((u64)root->stripesize - 1);
3026 u64 ret = (val + mask) & ~mask;
3027 return ret;
3028 }
3029
3030 /*
3031 * walks the btree of allocated extents and find a hole of a given size.
3032 * The key ins is changed to record the hole:
3033 * ins->objectid == block start
3034 * ins->flags = BTRFS_EXTENT_ITEM_KEY
3035 * ins->offset == number of blocks
3036 * Any available blocks before search_start are skipped.
3037 */
3038 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
3039 struct btrfs_root *orig_root,
3040 u64 num_bytes, u64 empty_size,
3041 u64 search_start, u64 search_end,
3042 u64 hint_byte, struct btrfs_key *ins,
3043 u64 exclude_start, u64 exclude_nr,
3044 int data)
3045 {
3046 int ret = 0;
3047 struct btrfs_root *root = orig_root->fs_info->extent_root;
3048 u64 total_needed = num_bytes;
3049 u64 *last_ptr = NULL;
3050 u64 last_wanted = 0;
3051 struct btrfs_block_group_cache *block_group = NULL;
3052 int chunk_alloc_done = 0;
3053 int empty_cluster = 2 * 1024 * 1024;
3054 int allowed_chunk_alloc = 0;
3055 struct list_head *head = NULL, *cur = NULL;
3056 int loop = 0;
3057 int extra_loop = 0;
3058 struct btrfs_space_info *space_info;
3059
3060 WARN_ON(num_bytes < root->sectorsize);
3061 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
3062 ins->objectid = 0;
3063 ins->offset = 0;
3064
3065 if (orig_root->ref_cows || empty_size)
3066 allowed_chunk_alloc = 1;
3067
3068 if (data & BTRFS_BLOCK_GROUP_METADATA) {
3069 last_ptr = &root->fs_info->last_alloc;
3070 if (!btrfs_test_opt(root, SSD))
3071 empty_cluster = 64 * 1024;
3072 }
3073
3074 if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD))
3075 last_ptr = &root->fs_info->last_data_alloc;
3076
3077 if (last_ptr) {
3078 if (*last_ptr) {
3079 hint_byte = *last_ptr;
3080 last_wanted = *last_ptr;
3081 } else
3082 empty_size += empty_cluster;
3083 } else {
3084 empty_cluster = 0;
3085 }
3086 search_start = max(search_start, first_logical_byte(root, 0));
3087 search_start = max(search_start, hint_byte);
3088
3089 if (last_wanted && search_start != last_wanted) {
3090 last_wanted = 0;
3091 empty_size += empty_cluster;
3092 }
3093
3094 total_needed += empty_size;
3095 block_group = btrfs_lookup_block_group(root->fs_info, search_start);
3096 if (!block_group)
3097 block_group = btrfs_lookup_first_block_group(root->fs_info,
3098 search_start);
3099 space_info = __find_space_info(root->fs_info, data);
3100
3101 down_read(&space_info->groups_sem);
3102 while (1) {
3103 struct btrfs_free_space *free_space;
3104 /*
3105 * the only way this happens if our hint points to a block
3106 * group thats not of the proper type, while looping this
3107 * should never happen
3108 */
3109 if (empty_size)
3110 extra_loop = 1;
3111
3112 if (!block_group)
3113 goto new_group_no_lock;
3114
3115 if (unlikely(!block_group->cached)) {
3116 mutex_lock(&block_group->cache_mutex);
3117 ret = cache_block_group(root, block_group);
3118 mutex_unlock(&block_group->cache_mutex);
3119 if (ret)
3120 break;
3121 }
3122
3123 mutex_lock(&block_group->alloc_mutex);
3124 if (unlikely(!block_group_bits(block_group, data)))
3125 goto new_group;
3126
3127 if (unlikely(block_group->ro))
3128 goto new_group;
3129
3130 free_space = btrfs_find_free_space(block_group, search_start,
3131 total_needed);
3132 if (free_space) {
3133 u64 start = block_group->key.objectid;
3134 u64 end = block_group->key.objectid +
3135 block_group->key.offset;
3136
3137 search_start = stripe_align(root, free_space->offset);
3138
3139 /* move on to the next group */
3140 if (search_start + num_bytes >= search_end)
3141 goto new_group;
3142
3143 /* move on to the next group */
3144 if (search_start + num_bytes > end)
3145 goto new_group;
3146
3147 if (last_wanted && search_start != last_wanted) {
3148 total_needed += empty_cluster;
3149 empty_size += empty_cluster;
3150 last_wanted = 0;
3151 /*
3152 * if search_start is still in this block group
3153 * then we just re-search this block group
3154 */
3155 if (search_start >= start &&
3156 search_start < end) {
3157 mutex_unlock(&block_group->alloc_mutex);
3158 continue;
3159 }
3160
3161 /* else we go to the next block group */
3162 goto new_group;
3163 }
3164
3165 if (exclude_nr > 0 &&
3166 (search_start + num_bytes > exclude_start &&
3167 search_start < exclude_start + exclude_nr)) {
3168 search_start = exclude_start + exclude_nr;
3169 /*
3170 * if search_start is still in this block group
3171 * then we just re-search this block group
3172 */
3173 if (search_start >= start &&
3174 search_start < end) {
3175 mutex_unlock(&block_group->alloc_mutex);
3176 last_wanted = 0;
3177 continue;
3178 }
3179
3180 /* else we go to the next block group */
3181 goto new_group;
3182 }
3183
3184 ins->objectid = search_start;
3185 ins->offset = num_bytes;
3186
3187 btrfs_remove_free_space_lock(block_group, search_start,
3188 num_bytes);
3189 /* we are all good, lets return */
3190 mutex_unlock(&block_group->alloc_mutex);
3191 break;
3192 }
3193 new_group:
3194 mutex_unlock(&block_group->alloc_mutex);
3195 put_block_group(block_group);
3196 block_group = NULL;
3197 new_group_no_lock:
3198 /* don't try to compare new allocations against the
3199 * last allocation any more
3200 */
3201 last_wanted = 0;
3202
3203 /*
3204 * Here's how this works.
3205 * loop == 0: we were searching a block group via a hint
3206 * and didn't find anything, so we start at
3207 * the head of the block groups and keep searching
3208 * loop == 1: we're searching through all of the block groups
3209 * if we hit the head again we have searched
3210 * all of the block groups for this space and we
3211 * need to try and allocate, if we cant error out.
3212 * loop == 2: we allocated more space and are looping through
3213 * all of the block groups again.
3214 */
3215 if (loop == 0) {
3216 head = &space_info->block_groups;
3217 cur = head->next;
3218 loop++;
3219 } else if (loop == 1 && cur == head) {
3220 int keep_going;
3221
3222 /* at this point we give up on the empty_size
3223 * allocations and just try to allocate the min
3224 * space.
3225 *
3226 * The extra_loop field was set if an empty_size
3227 * allocation was attempted above, and if this
3228 * is try we need to try the loop again without
3229 * the additional empty_size.
3230 */
3231 total_needed -= empty_size;
3232 empty_size = 0;
3233 keep_going = extra_loop;
3234 loop++;
3235
3236 if (allowed_chunk_alloc && !chunk_alloc_done) {
3237 up_read(&space_info->groups_sem);
3238 ret = do_chunk_alloc(trans, root, num_bytes +
3239 2 * 1024 * 1024, data, 1);
3240 down_read(&space_info->groups_sem);
3241 if (ret < 0)
3242 goto loop_check;
3243 head = &space_info->block_groups;
3244 /*
3245 * we've allocated a new chunk, keep
3246 * trying
3247 */
3248 keep_going = 1;
3249 chunk_alloc_done = 1;
3250 } else if (!allowed_chunk_alloc) {
3251 space_info->force_alloc = 1;
3252 }
3253 loop_check:
3254 if (keep_going) {
3255 cur = head->next;
3256 extra_loop = 0;
3257 } else {
3258 break;
3259 }
3260 } else if (cur == head) {
3261 break;
3262 }
3263
3264 block_group = list_entry(cur, struct btrfs_block_group_cache,
3265 list);
3266 atomic_inc(&block_group->count);
3267
3268 search_start = block_group->key.objectid;
3269 cur = cur->next;
3270 }
3271
3272 /* we found what we needed */
3273 if (ins->objectid) {
3274 if (!(data & BTRFS_BLOCK_GROUP_DATA))
3275 trans->block_group = block_group->key.objectid;
3276
3277 if (last_ptr)
3278 *last_ptr = ins->objectid + ins->offset;
3279 ret = 0;
3280 } else if (!ret) {
3281 printk(KERN_ERR "btrfs searching for %llu bytes, "
3282 "num_bytes %llu, loop %d, allowed_alloc %d\n",
3283 (unsigned long long)total_needed,
3284 (unsigned long long)num_bytes,
3285 loop, allowed_chunk_alloc);
3286 ret = -ENOSPC;
3287 }
3288 if (block_group)
3289 put_block_group(block_group);
3290
3291 up_read(&space_info->groups_sem);
3292 return ret;
3293 }
3294
3295 static void dump_space_info(struct btrfs_space_info *info, u64 bytes)
3296 {
3297 struct btrfs_block_group_cache *cache;
3298
3299 printk(KERN_INFO "space_info has %llu free, is %sfull\n",
3300 (unsigned long long)(info->total_bytes - info->bytes_used -
3301 info->bytes_pinned - info->bytes_reserved),
3302 (info->full) ? "" : "not ");
3303 printk(KERN_INFO "space_info total=%llu, pinned=%llu, delalloc=%llu,"
3304 " may_use=%llu, used=%llu\n", info->total_bytes,
3305 info->bytes_pinned, info->bytes_delalloc, info->bytes_may_use,
3306 info->bytes_used);
3307
3308 down_read(&info->groups_sem);
3309 list_for_each_entry(cache, &info->block_groups, list) {
3310 spin_lock(&cache->lock);
3311 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
3312 "%llu pinned %llu reserved\n",
3313 (unsigned long long)cache->key.objectid,
3314 (unsigned long long)cache->key.offset,
3315 (unsigned long long)btrfs_block_group_used(&cache->item),
3316 (unsigned long long)cache->pinned,
3317 (unsigned long long)cache->reserved);
3318 btrfs_dump_free_space(cache, bytes);
3319 spin_unlock(&cache->lock);
3320 }
3321 up_read(&info->groups_sem);
3322 }
3323
3324 static int __btrfs_reserve_extent(struct btrfs_trans_handle *trans,
3325 struct btrfs_root *root,
3326 u64 num_bytes, u64 min_alloc_size,
3327 u64 empty_size, u64 hint_byte,
3328 u64 search_end, struct btrfs_key *ins,
3329 u64 data)
3330 {
3331 int ret;
3332 u64 search_start = 0;
3333 struct btrfs_fs_info *info = root->fs_info;
3334
3335 data = btrfs_get_alloc_profile(root, data);
3336 again:
3337 /*
3338 * the only place that sets empty_size is btrfs_realloc_node, which
3339 * is not called recursively on allocations
3340 */
3341 if (empty_size || root->ref_cows) {
3342 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
3343 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3344 2 * 1024 * 1024,
3345 BTRFS_BLOCK_GROUP_METADATA |
3346 (info->metadata_alloc_profile &
3347 info->avail_metadata_alloc_bits), 0);
3348 }
3349 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3350 num_bytes + 2 * 1024 * 1024, data, 0);
3351 }
3352
3353 WARN_ON(num_bytes < root->sectorsize);
3354 ret = find_free_extent(trans, root, num_bytes, empty_size,
3355 search_start, search_end, hint_byte, ins,
3356 trans->alloc_exclude_start,
3357 trans->alloc_exclude_nr, data);
3358
3359 if (ret == -ENOSPC && num_bytes > min_alloc_size) {
3360 num_bytes = num_bytes >> 1;
3361 num_bytes = num_bytes & ~(root->sectorsize - 1);
3362 num_bytes = max(num_bytes, min_alloc_size);
3363 do_chunk_alloc(trans, root->fs_info->extent_root,
3364 num_bytes, data, 1);
3365 goto again;
3366 }
3367 if (ret) {
3368 struct btrfs_space_info *sinfo;
3369
3370 sinfo = __find_space_info(root->fs_info, data);
3371 printk(KERN_ERR "btrfs allocation failed flags %llu, "
3372 "wanted %llu\n", (unsigned long long)data,
3373 (unsigned long long)num_bytes);
3374 dump_space_info(sinfo, num_bytes);
3375 BUG();
3376 }
3377
3378 return ret;
3379 }
3380
3381 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
3382 {
3383 struct btrfs_block_group_cache *cache;
3384 int ret = 0;
3385
3386 cache = btrfs_lookup_block_group(root->fs_info, start);
3387 if (!cache) {
3388 printk(KERN_ERR "Unable to find block group for %llu\n",
3389 (unsigned long long)start);
3390 return -ENOSPC;
3391 }
3392
3393 ret = btrfs_discard_extent(root, start, len);
3394
3395 btrfs_add_free_space(cache, start, len);
3396 put_block_group(cache);
3397 update_reserved_extents(root, start, len, 0);
3398
3399 return ret;
3400 }
3401
3402 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
3403 struct btrfs_root *root,
3404 u64 num_bytes, u64 min_alloc_size,
3405 u64 empty_size, u64 hint_byte,
3406 u64 search_end, struct btrfs_key *ins,
3407 u64 data)
3408 {
3409 int ret;
3410 ret = __btrfs_reserve_extent(trans, root, num_bytes, min_alloc_size,
3411 empty_size, hint_byte, search_end, ins,
3412 data);
3413 update_reserved_extents(root, ins->objectid, ins->offset, 1);
3414 return ret;
3415 }
3416
3417 static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
3418 struct btrfs_root *root, u64 parent,
3419 u64 root_objectid, u64 ref_generation,
3420 u64 owner, struct btrfs_key *ins)
3421 {
3422 int ret;
3423 int pending_ret;
3424 u64 super_used;
3425 u64 root_used;
3426 u64 num_bytes = ins->offset;
3427 u32 sizes[2];
3428 struct btrfs_fs_info *info = root->fs_info;
3429 struct btrfs_root *extent_root = info->extent_root;
3430 struct btrfs_extent_item *extent_item;
3431 struct btrfs_extent_ref *ref;
3432 struct btrfs_path *path;
3433 struct btrfs_key keys[2];
3434
3435 if (parent == 0)
3436 parent = ins->objectid;
3437
3438 /* block accounting for super block */
3439 spin_lock(&info->delalloc_lock);
3440 super_used = btrfs_super_bytes_used(&info->super_copy);
3441 btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes);
3442
3443 /* block accounting for root item */
3444 root_used = btrfs_root_used(&root->root_item);
3445 btrfs_set_root_used(&root->root_item, root_used + num_bytes);
3446 spin_unlock(&info->delalloc_lock);
3447
3448 if (root == extent_root) {
3449 struct pending_extent_op *extent_op;
3450
3451 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
3452 BUG_ON(!extent_op);
3453
3454 extent_op->type = PENDING_EXTENT_INSERT;
3455 extent_op->bytenr = ins->objectid;
3456 extent_op->num_bytes = ins->offset;
3457 extent_op->parent = parent;
3458 extent_op->orig_parent = 0;
3459 extent_op->generation = ref_generation;
3460 extent_op->orig_generation = 0;
3461 extent_op->level = (int)owner;
3462 INIT_LIST_HEAD(&extent_op->list);
3463 extent_op->del = 0;
3464
3465 mutex_lock(&root->fs_info->extent_ins_mutex);
3466 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
3467 ins->objectid + ins->offset - 1,
3468 EXTENT_WRITEBACK, GFP_NOFS);
3469 set_state_private(&root->fs_info->extent_ins,
3470 ins->objectid, (unsigned long)extent_op);
3471 mutex_unlock(&root->fs_info->extent_ins_mutex);
3472 goto update_block;
3473 }
3474
3475 memcpy(&keys[0], ins, sizeof(*ins));
3476 keys[1].objectid = ins->objectid;
3477 keys[1].type = BTRFS_EXTENT_REF_KEY;
3478 keys[1].offset = parent;
3479 sizes[0] = sizeof(*extent_item);
3480 sizes[1] = sizeof(*ref);
3481
3482 path = btrfs_alloc_path();
3483 BUG_ON(!path);
3484
3485 ret = btrfs_insert_empty_items(trans, extent_root, path, keys,
3486 sizes, 2);
3487 BUG_ON(ret);
3488
3489 extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3490 struct btrfs_extent_item);
3491 btrfs_set_extent_refs(path->nodes[0], extent_item, 1);
3492 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
3493 struct btrfs_extent_ref);
3494
3495 btrfs_set_ref_root(path->nodes[0], ref, root_objectid);
3496 btrfs_set_ref_generation(path->nodes[0], ref, ref_generation);
3497 btrfs_set_ref_objectid(path->nodes[0], ref, owner);
3498 btrfs_set_ref_num_refs(path->nodes[0], ref, 1);
3499
3500 btrfs_mark_buffer_dirty(path->nodes[0]);
3501
3502 trans->alloc_exclude_start = 0;
3503 trans->alloc_exclude_nr = 0;
3504 btrfs_free_path(path);
3505 finish_current_insert(trans, extent_root, 0);
3506 pending_ret = del_pending_extents(trans, extent_root, 0);
3507
3508 if (ret)
3509 goto out;
3510 if (pending_ret) {
3511 ret = pending_ret;
3512 goto out;
3513 }
3514
3515 update_block:
3516 ret = update_block_group(trans, root, ins->objectid,
3517 ins->offset, 1, 0);
3518 if (ret) {
3519 printk(KERN_ERR "btrfs update block group failed for %llu "
3520 "%llu\n", (unsigned long long)ins->objectid,
3521 (unsigned long long)ins->offset);
3522 BUG();
3523 }
3524 out:
3525 return ret;
3526 }
3527
3528 int btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
3529 struct btrfs_root *root, u64 parent,
3530 u64 root_objectid, u64 ref_generation,
3531 u64 owner, struct btrfs_key *ins)
3532 {
3533 int ret;
3534
3535 if (root_objectid == BTRFS_TREE_LOG_OBJECTID)
3536 return 0;
3537 ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
3538 ref_generation, owner, ins);
3539 update_reserved_extents(root, ins->objectid, ins->offset, 0);
3540 return ret;
3541 }
3542
3543 /*
3544 * this is used by the tree logging recovery code. It records that
3545 * an extent has been allocated and makes sure to clear the free
3546 * space cache bits as well
3547 */
3548 int btrfs_alloc_logged_extent(struct btrfs_trans_handle *trans,
3549 struct btrfs_root *root, u64 parent,
3550 u64 root_objectid, u64 ref_generation,
3551 u64 owner, struct btrfs_key *ins)
3552 {
3553 int ret;
3554 struct btrfs_block_group_cache *block_group;
3555
3556 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
3557 mutex_lock(&block_group->cache_mutex);
3558 cache_block_group(root, block_group);
3559 mutex_unlock(&block_group->cache_mutex);
3560
3561 ret = btrfs_remove_free_space(block_group, ins->objectid,
3562 ins->offset);
3563 BUG_ON(ret);
3564 put_block_group(block_group);
3565 ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
3566 ref_generation, owner, ins);
3567 return ret;
3568 }
3569
3570 /*
3571 * finds a free extent and does all the dirty work required for allocation
3572 * returns the key for the extent through ins, and a tree buffer for
3573 * the first block of the extent through buf.
3574 *
3575 * returns 0 if everything worked, non-zero otherwise.
3576 */
3577 int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
3578 struct btrfs_root *root,
3579 u64 num_bytes, u64 parent, u64 min_alloc_size,
3580 u64 root_objectid, u64 ref_generation,
3581 u64 owner_objectid, u64 empty_size, u64 hint_byte,
3582 u64 search_end, struct btrfs_key *ins, u64 data)
3583 {
3584 int ret;
3585
3586 ret = __btrfs_reserve_extent(trans, root, num_bytes,
3587 min_alloc_size, empty_size, hint_byte,
3588 search_end, ins, data);
3589 BUG_ON(ret);
3590 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
3591 ret = __btrfs_alloc_reserved_extent(trans, root, parent,
3592 root_objectid, ref_generation,
3593 owner_objectid, ins);
3594 BUG_ON(ret);
3595
3596 } else {
3597 update_reserved_extents(root, ins->objectid, ins->offset, 1);
3598 }
3599 return ret;
3600 }
3601
3602 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
3603 struct btrfs_root *root,
3604 u64 bytenr, u32 blocksize,
3605 int level)
3606 {
3607 struct extent_buffer *buf;
3608
3609 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
3610 if (!buf)
3611 return ERR_PTR(-ENOMEM);
3612 btrfs_set_header_generation(buf, trans->transid);
3613 btrfs_set_buffer_lockdep_class(buf, level);
3614 btrfs_tree_lock(buf);
3615 clean_tree_block(trans, root, buf);
3616
3617 btrfs_set_lock_blocking(buf);
3618 btrfs_set_buffer_uptodate(buf);
3619
3620 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
3621 set_extent_dirty(&root->dirty_log_pages, buf->start,
3622 buf->start + buf->len - 1, GFP_NOFS);
3623 } else {
3624 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
3625 buf->start + buf->len - 1, GFP_NOFS);
3626 }
3627 trans->blocks_used++;
3628 /* this returns a buffer locked for blocking */
3629 return buf;
3630 }
3631
3632 /*
3633 * helper function to allocate a block for a given tree
3634 * returns the tree buffer or NULL.
3635 */
3636 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
3637 struct btrfs_root *root,
3638 u32 blocksize, u64 parent,
3639 u64 root_objectid,
3640 u64 ref_generation,
3641 int level,
3642 u64 hint,
3643 u64 empty_size)
3644 {
3645 struct btrfs_key ins;
3646 int ret;
3647 struct extent_buffer *buf;
3648
3649 ret = btrfs_alloc_extent(trans, root, blocksize, parent, blocksize,
3650 root_objectid, ref_generation, level,
3651 empty_size, hint, (u64)-1, &ins, 0);
3652 if (ret) {
3653 BUG_ON(ret > 0);
3654 return ERR_PTR(ret);
3655 }
3656
3657 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
3658 blocksize, level);
3659 return buf;
3660 }
3661
3662 int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans,
3663 struct btrfs_root *root, struct extent_buffer *leaf)
3664 {
3665 u64 leaf_owner;
3666 u64 leaf_generation;
3667 struct refsort *sorted;
3668 struct btrfs_key key;
3669 struct btrfs_file_extent_item *fi;
3670 int i;
3671 int nritems;
3672 int ret;
3673 int refi = 0;
3674 int slot;
3675
3676 BUG_ON(!btrfs_is_leaf(leaf));
3677 nritems = btrfs_header_nritems(leaf);
3678 leaf_owner = btrfs_header_owner(leaf);
3679 leaf_generation = btrfs_header_generation(leaf);
3680
3681 sorted = kmalloc(sizeof(*sorted) * nritems, GFP_NOFS);
3682 /* we do this loop twice. The first time we build a list
3683 * of the extents we have a reference on, then we sort the list
3684 * by bytenr. The second time around we actually do the
3685 * extent freeing.
3686 */
3687 for (i = 0; i < nritems; i++) {
3688 u64 disk_bytenr;
3689 cond_resched();
3690
3691 btrfs_item_key_to_cpu(leaf, &key, i);
3692
3693 /* only extents have references, skip everything else */
3694 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
3695 continue;
3696
3697 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
3698
3699 /* inline extents live in the btree, they don't have refs */
3700 if (btrfs_file_extent_type(leaf, fi) ==
3701 BTRFS_FILE_EXTENT_INLINE)
3702 continue;
3703
3704 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
3705
3706 /* holes don't have refs */
3707 if (disk_bytenr == 0)
3708 continue;
3709
3710 sorted[refi].bytenr = disk_bytenr;
3711 sorted[refi].slot = i;
3712 refi++;
3713 }
3714
3715 if (refi == 0)
3716 goto out;
3717
3718 sort(sorted, refi, sizeof(struct refsort), refsort_cmp, NULL);
3719
3720 for (i = 0; i < refi; i++) {
3721 u64 disk_bytenr;
3722
3723 disk_bytenr = sorted[i].bytenr;
3724 slot = sorted[i].slot;
3725
3726 cond_resched();
3727
3728 btrfs_item_key_to_cpu(leaf, &key, slot);
3729 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
3730 continue;
3731
3732 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
3733
3734 ret = __btrfs_free_extent(trans, root, disk_bytenr,
3735 btrfs_file_extent_disk_num_bytes(leaf, fi),
3736 leaf->start, leaf_owner, leaf_generation,
3737 key.objectid, 0);
3738 BUG_ON(ret);
3739
3740 atomic_inc(&root->fs_info->throttle_gen);
3741 wake_up(&root->fs_info->transaction_throttle);
3742 cond_resched();
3743 }
3744 out:
3745 kfree(sorted);
3746 return 0;
3747 }
3748
3749 static noinline int cache_drop_leaf_ref(struct btrfs_trans_handle *trans,
3750 struct btrfs_root *root,
3751 struct btrfs_leaf_ref *ref)
3752 {
3753 int i;
3754 int ret;
3755 struct btrfs_extent_info *info;
3756 struct refsort *sorted;
3757
3758 if (ref->nritems == 0)
3759 return 0;
3760
3761 sorted = kmalloc(sizeof(*sorted) * ref->nritems, GFP_NOFS);
3762 for (i = 0; i < ref->nritems; i++) {
3763 sorted[i].bytenr = ref->extents[i].bytenr;
3764 sorted[i].slot = i;
3765 }
3766 sort(sorted, ref->nritems, sizeof(struct refsort), refsort_cmp, NULL);
3767
3768 /*
3769 * the items in the ref were sorted when the ref was inserted
3770 * into the ref cache, so this is already in order
3771 */
3772 for (i = 0; i < ref->nritems; i++) {
3773 info = ref->extents + sorted[i].slot;
3774 ret = __btrfs_free_extent(trans, root, info->bytenr,
3775 info->num_bytes, ref->bytenr,
3776 ref->owner, ref->generation,
3777 info->objectid, 0);
3778
3779 atomic_inc(&root->fs_info->throttle_gen);
3780 wake_up(&root->fs_info->transaction_throttle);
3781 cond_resched();
3782
3783 BUG_ON(ret);
3784 info++;
3785 }
3786
3787 kfree(sorted);
3788 return 0;
3789 }
3790
3791 static int drop_snap_lookup_refcount(struct btrfs_root *root, u64 start,
3792 u64 len, u32 *refs)
3793 {
3794 int ret;
3795
3796 ret = btrfs_lookup_extent_ref(NULL, root, start, len, refs);
3797 BUG_ON(ret);
3798
3799 #if 0 /* some debugging code in case we see problems here */
3800 /* if the refs count is one, it won't get increased again. But
3801 * if the ref count is > 1, someone may be decreasing it at
3802 * the same time we are.
3803 */
3804 if (*refs != 1) {
3805 struct extent_buffer *eb = NULL;
3806 eb = btrfs_find_create_tree_block(root, start, len);
3807 if (eb)
3808 btrfs_tree_lock(eb);
3809
3810 mutex_lock(&root->fs_info->alloc_mutex);
3811 ret = lookup_extent_ref(NULL, root, start, len, refs);
3812 BUG_ON(ret);
3813 mutex_unlock(&root->fs_info->alloc_mutex);
3814
3815 if (eb) {
3816 btrfs_tree_unlock(eb);
3817 free_extent_buffer(eb);
3818 }
3819 if (*refs == 1) {
3820 printk(KERN_ERR "btrfs block %llu went down to one "
3821 "during drop_snap\n", (unsigned long long)start);
3822 }
3823
3824 }
3825 #endif
3826
3827 cond_resched();
3828 return ret;
3829 }
3830
3831 /*
3832 * this is used while deleting old snapshots, and it drops the refs
3833 * on a whole subtree starting from a level 1 node.
3834 *
3835 * The idea is to sort all the leaf pointers, and then drop the
3836 * ref on all the leaves in order. Most of the time the leaves
3837 * will have ref cache entries, so no leaf IOs will be required to
3838 * find the extents they have references on.
3839 *
3840 * For each leaf, any references it has are also dropped in order
3841 *
3842 * This ends up dropping the references in something close to optimal
3843 * order for reading and modifying the extent allocation tree.
3844 */
3845 static noinline int drop_level_one_refs(struct btrfs_trans_handle *trans,
3846 struct btrfs_root *root,
3847 struct btrfs_path *path)
3848 {
3849 u64 bytenr;
3850 u64 root_owner;
3851 u64 root_gen;
3852 struct extent_buffer *eb = path->nodes[1];
3853 struct extent_buffer *leaf;
3854 struct btrfs_leaf_ref *ref;
3855 struct refsort *sorted = NULL;
3856 int nritems = btrfs_header_nritems(eb);
3857 int ret;
3858 int i;
3859 int refi = 0;
3860 int slot = path->slots[1];
3861 u32 blocksize = btrfs_level_size(root, 0);
3862 u32 refs;
3863
3864 if (nritems == 0)
3865 goto out;
3866
3867 root_owner = btrfs_header_owner(eb);
3868 root_gen = btrfs_header_generation(eb);
3869 sorted = kmalloc(sizeof(*sorted) * nritems, GFP_NOFS);
3870
3871 /*
3872 * step one, sort all the leaf pointers so we don't scribble
3873 * randomly into the extent allocation tree
3874 */
3875 for (i = slot; i < nritems; i++) {
3876 sorted[refi].bytenr = btrfs_node_blockptr(eb, i);
3877 sorted[refi].slot = i;
3878 refi++;
3879 }
3880
3881 /*
3882 * nritems won't be zero, but if we're picking up drop_snapshot
3883 * after a crash, slot might be > 0, so double check things
3884 * just in case.
3885 */
3886 if (refi == 0)
3887 goto out;
3888
3889 sort(sorted, refi, sizeof(struct refsort), refsort_cmp, NULL);
3890
3891 /*
3892 * the first loop frees everything the leaves point to
3893 */
3894 for (i = 0; i < refi; i++) {
3895 u64 ptr_gen;
3896
3897 bytenr = sorted[i].bytenr;
3898
3899 /*
3900 * check the reference count on this leaf. If it is > 1
3901 * we just decrement it below and don't update any
3902 * of the refs the leaf points to.
3903 */
3904 ret = drop_snap_lookup_refcount(root, bytenr, blocksize, &refs);
3905 BUG_ON(ret);
3906 if (refs != 1)
3907 continue;
3908
3909 ptr_gen = btrfs_node_ptr_generation(eb, sorted[i].slot);
3910
3911 /*
3912 * the leaf only had one reference, which means the
3913 * only thing pointing to this leaf is the snapshot
3914 * we're deleting. It isn't possible for the reference
3915 * count to increase again later
3916 *
3917 * The reference cache is checked for the leaf,
3918 * and if found we'll be able to drop any refs held by
3919 * the leaf without needing to read it in.
3920 */
3921 ref = btrfs_lookup_leaf_ref(root, bytenr);
3922 if (ref && ref->generation != ptr_gen) {
3923 btrfs_free_leaf_ref(root, ref);
3924 ref = NULL;
3925 }
3926 if (ref) {
3927 ret = cache_drop_leaf_ref(trans, root, ref);
3928 BUG_ON(ret);
3929 btrfs_remove_leaf_ref(root, ref);
3930 btrfs_free_leaf_ref(root, ref);
3931 } else {
3932 /*
3933 * the leaf wasn't in the reference cache, so
3934 * we have to read it.
3935 */
3936 leaf = read_tree_block(root, bytenr, blocksize,
3937 ptr_gen);
3938 ret = btrfs_drop_leaf_ref(trans, root, leaf);
3939 BUG_ON(ret);
3940 free_extent_buffer(leaf);
3941 }
3942 atomic_inc(&root->fs_info->throttle_gen);
3943 wake_up(&root->fs_info->transaction_throttle);
3944 cond_resched();
3945 }
3946
3947 /*
3948 * run through the loop again to free the refs on the leaves.
3949 * This is faster than doing it in the loop above because
3950 * the leaves are likely to be clustered together. We end up
3951 * working in nice chunks on the extent allocation tree.
3952 */
3953 for (i = 0; i < refi; i++) {
3954 bytenr = sorted[i].bytenr;
3955 ret = __btrfs_free_extent(trans, root, bytenr,
3956 blocksize, eb->start,
3957 root_owner, root_gen, 0, 1);
3958 BUG_ON(ret);
3959
3960 atomic_inc(&root->fs_info->throttle_gen);
3961 wake_up(&root->fs_info->transaction_throttle);
3962 cond_resched();
3963 }
3964 out:
3965 kfree(sorted);
3966
3967 /*
3968 * update the path to show we've processed the entire level 1
3969 * node. This will get saved into the root's drop_snapshot_progress
3970 * field so these drops are not repeated again if this transaction
3971 * commits.
3972 */
3973 path->slots[1] = nritems;
3974 return 0;
3975 }
3976
3977 /*
3978 * helper function for drop_snapshot, this walks down the tree dropping ref
3979 * counts as it goes.
3980 */
3981 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
3982 struct btrfs_root *root,
3983 struct btrfs_path *path, int *level)
3984 {
3985 u64 root_owner;
3986 u64 root_gen;
3987 u64 bytenr;
3988 u64 ptr_gen;
3989 struct extent_buffer *next;
3990 struct extent_buffer *cur;
3991 struct extent_buffer *parent;
3992 u32 blocksize;
3993 int ret;
3994 u32 refs;
3995
3996 WARN_ON(*level < 0);
3997 WARN_ON(*level >= BTRFS_MAX_LEVEL);
3998 ret = drop_snap_lookup_refcount(root, path->nodes[*level]->start,
3999 path->nodes[*level]->len, &refs);
4000 BUG_ON(ret);
4001 if (refs > 1)
4002 goto out;
4003
4004 /*
4005 * walk down to the last node level and free all the leaves
4006 */
4007 while (*level >= 0) {
4008 WARN_ON(*level < 0);
4009 WARN_ON(*level >= BTRFS_MAX_LEVEL);
4010 cur = path->nodes[*level];
4011
4012 if (btrfs_header_level(cur) != *level)
4013 WARN_ON(1);
4014
4015 if (path->slots[*level] >=
4016 btrfs_header_nritems(cur))
4017 break;
4018
4019 /* the new code goes down to level 1 and does all the
4020 * leaves pointed to that node in bulk. So, this check
4021 * for level 0 will always be false.
4022 *
4023 * But, the disk format allows the drop_snapshot_progress
4024 * field in the root to leave things in a state where
4025 * a leaf will need cleaning up here. If someone crashes
4026 * with the old code and then boots with the new code,
4027 * we might find a leaf here.
4028 */
4029 if (*level == 0) {
4030 ret = btrfs_drop_leaf_ref(trans, root, cur);
4031 BUG_ON(ret);
4032 break;
4033 }
4034
4035 /*
4036 * once we get to level one, process the whole node
4037 * at once, including everything below it.
4038 */
4039 if (*level == 1) {
4040 ret = drop_level_one_refs(trans, root, path);
4041 BUG_ON(ret);
4042 break;
4043 }
4044
4045 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
4046 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
4047 blocksize = btrfs_level_size(root, *level - 1);
4048
4049 ret = drop_snap_lookup_refcount(root, bytenr, blocksize, &refs);
4050 BUG_ON(ret);
4051
4052 /*
4053 * if there is more than one reference, we don't need
4054 * to read that node to drop any references it has. We
4055 * just drop the ref we hold on that node and move on to the
4056 * next slot in this level.
4057 */
4058 if (refs != 1) {
4059 parent = path->nodes[*level];
4060 root_owner = btrfs_header_owner(parent);
4061 root_gen = btrfs_header_generation(parent);
4062 path->slots[*level]++;
4063
4064 ret = __btrfs_free_extent(trans, root, bytenr,
4065 blocksize, parent->start,
4066 root_owner, root_gen,
4067 *level - 1, 1);
4068 BUG_ON(ret);
4069
4070 atomic_inc(&root->fs_info->throttle_gen);
4071 wake_up(&root->fs_info->transaction_throttle);
4072 cond_resched();
4073
4074 continue;
4075 }
4076
4077 /*
4078 * we need to keep freeing things in the next level down.
4079 * read the block and loop around to process it
4080 */
4081 next = read_tree_block(root, bytenr, blocksize, ptr_gen);
4082 WARN_ON(*level <= 0);
4083 if (path->nodes[*level-1])
4084 free_extent_buffer(path->nodes[*level-1]);
4085 path->nodes[*level-1] = next;
4086 *level = btrfs_header_level(next);
4087 path->slots[*level] = 0;
4088 cond_resched();
4089 }
4090 out:
4091 WARN_ON(*level < 0);
4092 WARN_ON(*level >= BTRFS_MAX_LEVEL);
4093
4094 if (path->nodes[*level] == root->node) {
4095 parent = path->nodes[*level];
4096 bytenr = path->nodes[*level]->start;
4097 } else {
4098 parent = path->nodes[*level + 1];
4099 bytenr = btrfs_node_blockptr(parent, path->slots[*level + 1]);
4100 }
4101
4102 blocksize = btrfs_level_size(root, *level);
4103 root_owner = btrfs_header_owner(parent);
4104 root_gen = btrfs_header_generation(parent);
4105
4106 /*
4107 * cleanup and free the reference on the last node
4108 * we processed
4109 */
4110 ret = __btrfs_free_extent(trans, root, bytenr, blocksize,
4111 parent->start, root_owner, root_gen,
4112 *level, 1);
4113 free_extent_buffer(path->nodes[*level]);
4114 path->nodes[*level] = NULL;
4115
4116 *level += 1;
4117 BUG_ON(ret);
4118
4119 cond_resched();
4120 return 0;
4121 }
4122
4123 /*
4124 * helper function for drop_subtree, this function is similar to
4125 * walk_down_tree. The main difference is that it checks reference
4126 * counts while tree blocks are locked.
4127 */
4128 static noinline int walk_down_subtree(struct btrfs_trans_handle *trans,
4129 struct btrfs_root *root,
4130 struct btrfs_path *path, int *level)
4131 {
4132 struct extent_buffer *next;
4133 struct extent_buffer *cur;
4134 struct extent_buffer *parent;
4135 u64 bytenr;
4136 u64 ptr_gen;
4137 u32 blocksize;
4138 u32 refs;
4139 int ret;
4140
4141 cur = path->nodes[*level];
4142 ret = btrfs_lookup_extent_ref(trans, root, cur->start, cur->len,
4143 &refs);
4144 BUG_ON(ret);
4145 if (refs > 1)
4146 goto out;
4147
4148 while (*level >= 0) {
4149 cur = path->nodes[*level];
4150 if (*level == 0) {
4151 ret = btrfs_drop_leaf_ref(trans, root, cur);
4152 BUG_ON(ret);
4153 clean_tree_block(trans, root, cur);
4154 break;
4155 }
4156 if (path->slots[*level] >= btrfs_header_nritems(cur)) {
4157 clean_tree_block(trans, root, cur);
4158 break;
4159 }
4160
4161 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
4162 blocksize = btrfs_level_size(root, *level - 1);
4163 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
4164
4165 next = read_tree_block(root, bytenr, blocksize, ptr_gen);
4166 btrfs_tree_lock(next);
4167 btrfs_set_lock_blocking(next);
4168
4169 ret = btrfs_lookup_extent_ref(trans, root, bytenr, blocksize,
4170 &refs);
4171 BUG_ON(ret);
4172 if (refs > 1) {
4173 parent = path->nodes[*level];
4174 ret = btrfs_free_extent(trans, root, bytenr,
4175 blocksize, parent->start,
4176 btrfs_header_owner(parent),
4177 btrfs_header_generation(parent),
4178 *level - 1, 1);
4179 BUG_ON(ret);
4180 path->slots[*level]++;
4181 btrfs_tree_unlock(next);
4182 free_extent_buffer(next);
4183 continue;
4184 }
4185
4186 *level = btrfs_header_level(next);
4187 path->nodes[*level] = next;
4188 path->slots[*level] = 0;
4189 path->locks[*level] = 1;
4190 cond_resched();
4191 }
4192 out:
4193 parent = path->nodes[*level + 1];
4194 bytenr = path->nodes[*level]->start;
4195 blocksize = path->nodes[*level]->len;
4196
4197 ret = btrfs_free_extent(trans, root, bytenr, blocksize,
4198 parent->start, btrfs_header_owner(parent),
4199 btrfs_header_generation(parent), *level, 1);
4200 BUG_ON(ret);
4201
4202 if (path->locks[*level]) {
4203 btrfs_tree_unlock(path->nodes[*level]);
4204 path->locks[*level] = 0;
4205 }
4206 free_extent_buffer(path->nodes[*level]);
4207 path->nodes[*level] = NULL;
4208 *level += 1;
4209 cond_resched();
4210 return 0;
4211 }
4212
4213 /*
4214 * helper for dropping snapshots. This walks back up the tree in the path
4215 * to find the first node higher up where we haven't yet gone through
4216 * all the slots
4217 */
4218 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
4219 struct btrfs_root *root,
4220 struct btrfs_path *path,
4221 int *level, int max_level)
4222 {
4223 u64 root_owner;
4224 u64 root_gen;
4225 struct btrfs_root_item *root_item = &root->root_item;
4226 int i;
4227 int slot;
4228 int ret;
4229
4230 for (i = *level; i < max_level && path->nodes[i]; i++) {
4231 slot = path->slots[i];
4232 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
4233 struct extent_buffer *node;
4234 struct btrfs_disk_key disk_key;
4235
4236 /*
4237 * there is more work to do in this level.
4238 * Update the drop_progress marker to reflect
4239 * the work we've done so far, and then bump
4240 * the slot number
4241 */
4242 node = path->nodes[i];
4243 path->slots[i]++;
4244 *level = i;
4245 WARN_ON(*level == 0);
4246 btrfs_node_key(node, &disk_key, path->slots[i]);
4247 memcpy(&root_item->drop_progress,
4248 &disk_key, sizeof(disk_key));
4249 root_item->drop_level = i;
4250 return 0;
4251 } else {
4252 struct extent_buffer *parent;
4253
4254 /*
4255 * this whole node is done, free our reference
4256 * on it and go up one level
4257 */
4258 if (path->nodes[*level] == root->node)
4259 parent = path->nodes[*level];
4260 else
4261 parent = path->nodes[*level + 1];
4262
4263 root_owner = btrfs_header_owner(parent);
4264 root_gen = btrfs_header_generation(parent);
4265
4266 clean_tree_block(trans, root, path->nodes[*level]);
4267 ret = btrfs_free_extent(trans, root,
4268 path->nodes[*level]->start,
4269 path->nodes[*level]->len,
4270 parent->start, root_owner,
4271 root_gen, *level, 1);
4272 BUG_ON(ret);
4273 if (path->locks[*level]) {
4274 btrfs_tree_unlock(path->nodes[*level]);
4275 path->locks[*level] = 0;
4276 }
4277 free_extent_buffer(path->nodes[*level]);
4278 path->nodes[*level] = NULL;
4279 *level = i + 1;
4280 }
4281 }
4282 return 1;
4283 }
4284
4285 /*
4286 * drop the reference count on the tree rooted at 'snap'. This traverses
4287 * the tree freeing any blocks that have a ref count of zero after being
4288 * decremented.
4289 */
4290 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
4291 *root)
4292 {
4293 int ret = 0;
4294 int wret;
4295 int level;
4296 struct btrfs_path *path;
4297 int i;
4298 int orig_level;
4299 struct btrfs_root_item *root_item = &root->root_item;
4300
4301 WARN_ON(!mutex_is_locked(&root->fs_info->drop_mutex));
4302 path = btrfs_alloc_path();
4303 BUG_ON(!path);
4304
4305 level = btrfs_header_level(root->node);
4306 orig_level = level;
4307 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
4308 path->nodes[level] = root->node;
4309 extent_buffer_get(root->node);
4310 path->slots[level] = 0;
4311 } else {
4312 struct btrfs_key key;
4313 struct btrfs_disk_key found_key;
4314 struct extent_buffer *node;
4315
4316 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
4317 level = root_item->drop_level;
4318 path->lowest_level = level;
4319 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4320 if (wret < 0) {
4321 ret = wret;
4322 goto out;
4323 }
4324 node = path->nodes[level];
4325 btrfs_node_key(node, &found_key, path->slots[level]);
4326 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
4327 sizeof(found_key)));
4328 /*
4329 * unlock our path, this is safe because only this
4330 * function is allowed to delete this snapshot
4331 */
4332 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
4333 if (path->nodes[i] && path->locks[i]) {
4334 path->locks[i] = 0;
4335 btrfs_tree_unlock(path->nodes[i]);
4336 }
4337 }
4338 }
4339 while (1) {
4340 wret = walk_down_tree(trans, root, path, &level);
4341 if (wret > 0)
4342 break;
4343 if (wret < 0)
4344 ret = wret;
4345
4346 wret = walk_up_tree(trans, root, path, &level,
4347 BTRFS_MAX_LEVEL);
4348 if (wret > 0)
4349 break;
4350 if (wret < 0)
4351 ret = wret;
4352 if (trans->transaction->in_commit) {
4353 ret = -EAGAIN;
4354 break;
4355 }
4356 atomic_inc(&root->fs_info->throttle_gen);
4357 wake_up(&root->fs_info->transaction_throttle);
4358 }
4359 for (i = 0; i <= orig_level; i++) {
4360 if (path->nodes[i]) {
4361 free_extent_buffer(path->nodes[i]);
4362 path->nodes[i] = NULL;
4363 }
4364 }
4365 out:
4366 btrfs_free_path(path);
4367 return ret;
4368 }
4369
4370 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
4371 struct btrfs_root *root,
4372 struct extent_buffer *node,
4373 struct extent_buffer *parent)
4374 {
4375 struct btrfs_path *path;
4376 int level;
4377 int parent_level;
4378 int ret = 0;
4379 int wret;
4380
4381 path = btrfs_alloc_path();
4382 BUG_ON(!path);
4383
4384 BUG_ON(!btrfs_tree_locked(parent));
4385 parent_level = btrfs_header_level(parent);
4386 extent_buffer_get(parent);
4387 path->nodes[parent_level] = parent;
4388 path->slots[parent_level] = btrfs_header_nritems(parent);
4389
4390 BUG_ON(!btrfs_tree_locked(node));
4391 level = btrfs_header_level(node);
4392 extent_buffer_get(node);
4393 path->nodes[level] = node;
4394 path->slots[level] = 0;
4395
4396 while (1) {
4397 wret = walk_down_subtree(trans, root, path, &level);
4398 if (wret < 0)
4399 ret = wret;
4400 if (wret != 0)
4401 break;
4402
4403 wret = walk_up_tree(trans, root, path, &level, parent_level);
4404 if (wret < 0)
4405 ret = wret;
4406 if (wret != 0)
4407 break;
4408 }
4409
4410 btrfs_free_path(path);
4411 return ret;
4412 }
4413
4414 static unsigned long calc_ra(unsigned long start, unsigned long last,
4415 unsigned long nr)
4416 {
4417 return min(last, start + nr - 1);
4418 }
4419
4420 static noinline int relocate_inode_pages(struct inode *inode, u64 start,
4421 u64 len)
4422 {
4423 u64 page_start;
4424 u64 page_end;
4425 unsigned long first_index;
4426 unsigned long last_index;
4427 unsigned long i;
4428 struct page *page;
4429 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4430 struct file_ra_state *ra;
4431 struct btrfs_ordered_extent *ordered;
4432 unsigned int total_read = 0;
4433 unsigned int total_dirty = 0;
4434 int ret = 0;
4435
4436 ra = kzalloc(sizeof(*ra), GFP_NOFS);
4437
4438 mutex_lock(&inode->i_mutex);
4439 first_index = start >> PAGE_CACHE_SHIFT;
4440 last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
4441
4442 /* make sure the dirty trick played by the caller work */
4443 ret = invalidate_inode_pages2_range(inode->i_mapping,
4444 first_index, last_index);
4445 if (ret)
4446 goto out_unlock;
4447
4448 file_ra_state_init(ra, inode->i_mapping);
4449
4450 for (i = first_index ; i <= last_index; i++) {
4451 if (total_read % ra->ra_pages == 0) {
4452 btrfs_force_ra(inode->i_mapping, ra, NULL, i,
4453 calc_ra(i, last_index, ra->ra_pages));
4454 }
4455 total_read++;
4456 again:
4457 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
4458 BUG_ON(1);
4459 page = grab_cache_page(inode->i_mapping, i);
4460 if (!page) {
4461 ret = -ENOMEM;
4462 goto out_unlock;
4463 }
4464 if (!PageUptodate(page)) {
4465 btrfs_readpage(NULL, page);
4466 lock_page(page);
4467 if (!PageUptodate(page)) {
4468 unlock_page(page);
4469 page_cache_release(page);
4470 ret = -EIO;
4471 goto out_unlock;
4472 }
4473 }
4474 wait_on_page_writeback(page);
4475
4476 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
4477 page_end = page_start + PAGE_CACHE_SIZE - 1;
4478 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
4479
4480 ordered = btrfs_lookup_ordered_extent(inode, page_start);
4481 if (ordered) {
4482 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4483 unlock_page(page);
4484 page_cache_release(page);
4485 btrfs_start_ordered_extent(inode, ordered, 1);
4486 btrfs_put_ordered_extent(ordered);
4487 goto again;
4488 }
4489 set_page_extent_mapped(page);
4490
4491 if (i == first_index)
4492 set_extent_bits(io_tree, page_start, page_end,
4493 EXTENT_BOUNDARY, GFP_NOFS);
4494 btrfs_set_extent_delalloc(inode, page_start, page_end);
4495
4496 set_page_dirty(page);
4497 total_dirty++;
4498
4499 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4500 unlock_page(page);
4501 page_cache_release(page);
4502 }
4503
4504 out_unlock:
4505 kfree(ra);
4506 mutex_unlock(&inode->i_mutex);
4507 balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
4508 return ret;
4509 }
4510
4511 static noinline int relocate_data_extent(struct inode *reloc_inode,
4512 struct btrfs_key *extent_key,
4513 u64 offset)
4514 {
4515 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
4516 struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
4517 struct extent_map *em;
4518 u64 start = extent_key->objectid - offset;
4519 u64 end = start + extent_key->offset - 1;
4520
4521 em = alloc_extent_map(GFP_NOFS);
4522 BUG_ON(!em || IS_ERR(em));
4523
4524 em->start = start;
4525 em->len = extent_key->offset;
4526 em->block_len = extent_key->offset;
4527 em->block_start = extent_key->objectid;
4528 em->bdev = root->fs_info->fs_devices->latest_bdev;
4529 set_bit(EXTENT_FLAG_PINNED, &em->flags);
4530
4531 /* setup extent map to cheat btrfs_readpage */
4532 lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
4533 while (1) {
4534 int ret;
4535 spin_lock(&em_tree->lock);
4536 ret = add_extent_mapping(em_tree, em);
4537 spin_unlock(&em_tree->lock);
4538 if (ret != -EEXIST) {
4539 free_extent_map(em);
4540 break;
4541 }
4542 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
4543 }
4544 unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
4545
4546 return relocate_inode_pages(reloc_inode, start, extent_key->offset);
4547 }
4548
4549 struct btrfs_ref_path {
4550 u64 extent_start;
4551 u64 nodes[BTRFS_MAX_LEVEL];
4552 u64 root_objectid;
4553 u64 root_generation;
4554 u64 owner_objectid;
4555 u32 num_refs;
4556 int lowest_level;
4557 int current_level;
4558 int shared_level;
4559
4560 struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
4561 u64 new_nodes[BTRFS_MAX_LEVEL];
4562 };
4563
4564 struct disk_extent {
4565 u64 ram_bytes;
4566 u64 disk_bytenr;
4567 u64 disk_num_bytes;
4568 u64 offset;
4569 u64 num_bytes;
4570 u8 compression;
4571 u8 encryption;
4572 u16 other_encoding;
4573 };
4574
4575 static int is_cowonly_root(u64 root_objectid)
4576 {
4577 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
4578 root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
4579 root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
4580 root_objectid == BTRFS_DEV_TREE_OBJECTID ||
4581 root_objectid == BTRFS_TREE_LOG_OBJECTID ||
4582 root_objectid == BTRFS_CSUM_TREE_OBJECTID)
4583 return 1;
4584 return 0;
4585 }
4586
4587 static noinline int __next_ref_path(struct btrfs_trans_handle *trans,
4588 struct btrfs_root *extent_root,
4589 struct btrfs_ref_path *ref_path,
4590 int first_time)
4591 {
4592 struct extent_buffer *leaf;
4593 struct btrfs_path *path;
4594 struct btrfs_extent_ref *ref;
4595 struct btrfs_key key;
4596 struct btrfs_key found_key;
4597 u64 bytenr;
4598 u32 nritems;
4599 int level;
4600 int ret = 1;
4601
4602 path = btrfs_alloc_path();
4603 if (!path)
4604 return -ENOMEM;
4605
4606 if (first_time) {
4607 ref_path->lowest_level = -1;
4608 ref_path->current_level = -1;
4609 ref_path->shared_level = -1;
4610 goto walk_up;
4611 }
4612 walk_down:
4613 level = ref_path->current_level - 1;
4614 while (level >= -1) {
4615 u64 parent;
4616 if (level < ref_path->lowest_level)
4617 break;
4618
4619 if (level >= 0)
4620 bytenr = ref_path->nodes[level];
4621 else
4622 bytenr = ref_path->extent_start;
4623 BUG_ON(bytenr == 0);
4624
4625 parent = ref_path->nodes[level + 1];
4626 ref_path->nodes[level + 1] = 0;
4627 ref_path->current_level = level;
4628 BUG_ON(parent == 0);
4629
4630 key.objectid = bytenr;
4631 key.offset = parent + 1;
4632 key.type = BTRFS_EXTENT_REF_KEY;
4633
4634 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
4635 if (ret < 0)
4636 goto out;
4637 BUG_ON(ret == 0);
4638
4639 leaf = path->nodes[0];
4640 nritems = btrfs_header_nritems(leaf);
4641 if (path->slots[0] >= nritems) {
4642 ret = btrfs_next_leaf(extent_root, path);
4643 if (ret < 0)
4644 goto out;
4645 if (ret > 0)
4646 goto next;
4647 leaf = path->nodes[0];
4648 }
4649
4650 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4651 if (found_key.objectid == bytenr &&
4652 found_key.type == BTRFS_EXTENT_REF_KEY) {
4653 if (level < ref_path->shared_level)
4654 ref_path->shared_level = level;
4655 goto found;
4656 }
4657 next:
4658 level--;
4659 btrfs_release_path(extent_root, path);
4660 cond_resched();
4661 }
4662 /* reached lowest level */
4663 ret = 1;
4664 goto out;
4665 walk_up:
4666 level = ref_path->current_level;
4667 while (level < BTRFS_MAX_LEVEL - 1) {
4668 u64 ref_objectid;
4669
4670 if (level >= 0)
4671 bytenr = ref_path->nodes[level];
4672 else
4673 bytenr = ref_path->extent_start;
4674
4675 BUG_ON(bytenr == 0);
4676
4677 key.objectid = bytenr;
4678 key.offset = 0;
4679 key.type = BTRFS_EXTENT_REF_KEY;
4680
4681 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
4682 if (ret < 0)
4683 goto out;
4684
4685 leaf = path->nodes[0];
4686 nritems = btrfs_header_nritems(leaf);
4687 if (path->slots[0] >= nritems) {
4688 ret = btrfs_next_leaf(extent_root, path);
4689 if (ret < 0)
4690 goto out;
4691 if (ret > 0) {
4692 /* the extent was freed by someone */
4693 if (ref_path->lowest_level == level)
4694 goto out;
4695 btrfs_release_path(extent_root, path);
4696 goto walk_down;
4697 }
4698 leaf = path->nodes[0];
4699 }
4700
4701 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4702 if (found_key.objectid != bytenr ||
4703 found_key.type != BTRFS_EXTENT_REF_KEY) {
4704 /* the extent was freed by someone */
4705 if (ref_path->lowest_level == level) {
4706 ret = 1;
4707 goto out;
4708 }
4709 btrfs_release_path(extent_root, path);
4710 goto walk_down;
4711 }
4712 found:
4713 ref = btrfs_item_ptr(leaf, path->slots[0],
4714 struct btrfs_extent_ref);
4715 ref_objectid = btrfs_ref_objectid(leaf, ref);
4716 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4717 if (first_time) {
4718 level = (int)ref_objectid;
4719 BUG_ON(level >= BTRFS_MAX_LEVEL);
4720 ref_path->lowest_level = level;
4721 ref_path->current_level = level;
4722 ref_path->nodes[level] = bytenr;
4723 } else {
4724 WARN_ON(ref_objectid != level);
4725 }
4726 } else {
4727 WARN_ON(level != -1);
4728 }
4729 first_time = 0;
4730
4731 if (ref_path->lowest_level == level) {
4732 ref_path->owner_objectid = ref_objectid;
4733 ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
4734 }
4735
4736 /*
4737 * the block is tree root or the block isn't in reference
4738 * counted tree.
4739 */
4740 if (found_key.objectid == found_key.offset ||
4741 is_cowonly_root(btrfs_ref_root(leaf, ref))) {
4742 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
4743 ref_path->root_generation =
4744 btrfs_ref_generation(leaf, ref);
4745 if (level < 0) {
4746 /* special reference from the tree log */
4747 ref_path->nodes[0] = found_key.offset;
4748 ref_path->current_level = 0;
4749 }
4750 ret = 0;
4751 goto out;
4752 }
4753
4754 level++;
4755 BUG_ON(ref_path->nodes[level] != 0);
4756 ref_path->nodes[level] = found_key.offset;
4757 ref_path->current_level = level;
4758
4759 /*
4760 * the reference was created in the running transaction,
4761 * no need to continue walking up.
4762 */
4763 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
4764 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
4765 ref_path->root_generation =
4766 btrfs_ref_generation(leaf, ref);
4767 ret = 0;
4768 goto out;
4769 }
4770
4771 btrfs_release_path(extent_root, path);
4772 cond_resched();
4773 }
4774 /* reached max tree level, but no tree root found. */
4775 BUG();
4776 out:
4777 btrfs_free_path(path);
4778 return ret;
4779 }
4780
4781 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
4782 struct btrfs_root *extent_root,
4783 struct btrfs_ref_path *ref_path,
4784 u64 extent_start)
4785 {
4786 memset(ref_path, 0, sizeof(*ref_path));
4787 ref_path->extent_start = extent_start;
4788
4789 return __next_ref_path(trans, extent_root, ref_path, 1);
4790 }
4791
4792 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
4793 struct btrfs_root *extent_root,
4794 struct btrfs_ref_path *ref_path)
4795 {
4796 return __next_ref_path(trans, extent_root, ref_path, 0);
4797 }
4798
4799 static noinline int get_new_locations(struct inode *reloc_inode,
4800 struct btrfs_key *extent_key,
4801 u64 offset, int no_fragment,
4802 struct disk_extent **extents,
4803 int *nr_extents)
4804 {
4805 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
4806 struct btrfs_path *path;
4807 struct btrfs_file_extent_item *fi;
4808 struct extent_buffer *leaf;
4809 struct disk_extent *exts = *extents;
4810 struct btrfs_key found_key;
4811 u64 cur_pos;
4812 u64 last_byte;
4813 u32 nritems;
4814 int nr = 0;
4815 int max = *nr_extents;
4816 int ret;
4817
4818 WARN_ON(!no_fragment && *extents);
4819 if (!exts) {
4820 max = 1;
4821 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
4822 if (!exts)
4823 return -ENOMEM;
4824 }
4825
4826 path = btrfs_alloc_path();
4827 BUG_ON(!path);
4828
4829 cur_pos = extent_key->objectid - offset;
4830 last_byte = extent_key->objectid + extent_key->offset;
4831 ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
4832 cur_pos, 0);
4833 if (ret < 0)
4834 goto out;
4835 if (ret > 0) {
4836 ret = -ENOENT;
4837 goto out;
4838 }
4839
4840 while (1) {
4841 leaf = path->nodes[0];
4842 nritems = btrfs_header_nritems(leaf);
4843 if (path->slots[0] >= nritems) {
4844 ret = btrfs_next_leaf(root, path);
4845 if (ret < 0)
4846 goto out;
4847 if (ret > 0)
4848 break;
4849 leaf = path->nodes[0];
4850 }
4851
4852 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4853 if (found_key.offset != cur_pos ||
4854 found_key.type != BTRFS_EXTENT_DATA_KEY ||
4855 found_key.objectid != reloc_inode->i_ino)
4856 break;
4857
4858 fi = btrfs_item_ptr(leaf, path->slots[0],
4859 struct btrfs_file_extent_item);
4860 if (btrfs_file_extent_type(leaf, fi) !=
4861 BTRFS_FILE_EXTENT_REG ||
4862 btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
4863 break;
4864
4865 if (nr == max) {
4866 struct disk_extent *old = exts;
4867 max *= 2;
4868 exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
4869 memcpy(exts, old, sizeof(*exts) * nr);
4870 if (old != *extents)
4871 kfree(old);
4872 }
4873
4874 exts[nr].disk_bytenr =
4875 btrfs_file_extent_disk_bytenr(leaf, fi);
4876 exts[nr].disk_num_bytes =
4877 btrfs_file_extent_disk_num_bytes(leaf, fi);
4878 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
4879 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4880 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
4881 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
4882 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
4883 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
4884 fi);
4885 BUG_ON(exts[nr].offset > 0);
4886 BUG_ON(exts[nr].compression || exts[nr].encryption);
4887 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
4888
4889 cur_pos += exts[nr].num_bytes;
4890 nr++;
4891
4892 if (cur_pos + offset >= last_byte)
4893 break;
4894
4895 if (no_fragment) {
4896 ret = 1;
4897 goto out;
4898 }
4899 path->slots[0]++;
4900 }
4901
4902 BUG_ON(cur_pos + offset > last_byte);
4903 if (cur_pos + offset < last_byte) {
4904 ret = -ENOENT;
4905 goto out;
4906 }
4907 ret = 0;
4908 out:
4909 btrfs_free_path(path);
4910 if (ret) {
4911 if (exts != *extents)
4912 kfree(exts);
4913 } else {
4914 *extents = exts;
4915 *nr_extents = nr;
4916 }
4917 return ret;
4918 }
4919
4920 static noinline int replace_one_extent(struct btrfs_trans_handle *trans,
4921 struct btrfs_root *root,
4922 struct btrfs_path *path,
4923 struct btrfs_key *extent_key,
4924 struct btrfs_key *leaf_key,
4925 struct btrfs_ref_path *ref_path,
4926 struct disk_extent *new_extents,
4927 int nr_extents)
4928 {
4929 struct extent_buffer *leaf;
4930 struct btrfs_file_extent_item *fi;
4931 struct inode *inode = NULL;
4932 struct btrfs_key key;
4933 u64 lock_start = 0;
4934 u64 lock_end = 0;
4935 u64 num_bytes;
4936 u64 ext_offset;
4937 u64 search_end = (u64)-1;
4938 u32 nritems;
4939 int nr_scaned = 0;
4940 int extent_locked = 0;
4941 int extent_type;
4942 int ret;
4943
4944 memcpy(&key, leaf_key, sizeof(key));
4945 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
4946 if (key.objectid < ref_path->owner_objectid ||
4947 (key.objectid == ref_path->owner_objectid &&
4948 key.type < BTRFS_EXTENT_DATA_KEY)) {
4949 key.objectid = ref_path->owner_objectid;
4950 key.type = BTRFS_EXTENT_DATA_KEY;
4951 key.offset = 0;
4952 }
4953 }
4954
4955 while (1) {
4956 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
4957 if (ret < 0)
4958 goto out;
4959
4960 leaf = path->nodes[0];
4961 nritems = btrfs_header_nritems(leaf);
4962 next:
4963 if (extent_locked && ret > 0) {
4964 /*
4965 * the file extent item was modified by someone
4966 * before the extent got locked.
4967 */
4968 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4969 lock_end, GFP_NOFS);
4970 extent_locked = 0;
4971 }
4972
4973 if (path->slots[0] >= nritems) {
4974 if (++nr_scaned > 2)
4975 break;
4976
4977 BUG_ON(extent_locked);
4978 ret = btrfs_next_leaf(root, path);
4979 if (ret < 0)
4980 goto out;
4981 if (ret > 0)
4982 break;
4983 leaf = path->nodes[0];
4984 nritems = btrfs_header_nritems(leaf);
4985 }
4986
4987 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4988
4989 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
4990 if ((key.objectid > ref_path->owner_objectid) ||
4991 (key.objectid == ref_path->owner_objectid &&
4992 key.type > BTRFS_EXTENT_DATA_KEY) ||
4993 key.offset >= search_end)
4994 break;
4995 }
4996
4997 if (inode && key.objectid != inode->i_ino) {
4998 BUG_ON(extent_locked);
4999 btrfs_release_path(root, path);
5000 mutex_unlock(&inode->i_mutex);
5001 iput(inode);
5002 inode = NULL;
5003 continue;
5004 }
5005
5006 if (key.type != BTRFS_EXTENT_DATA_KEY) {
5007 path->slots[0]++;
5008 ret = 1;
5009 goto next;
5010 }
5011 fi = btrfs_item_ptr(leaf, path->slots[0],
5012 struct btrfs_file_extent_item);
5013 extent_type = btrfs_file_extent_type(leaf, fi);
5014 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
5015 extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
5016 (btrfs_file_extent_disk_bytenr(leaf, fi) !=
5017 extent_key->objectid)) {
5018 path->slots[0]++;
5019 ret = 1;
5020 goto next;
5021 }
5022
5023 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
5024 ext_offset = btrfs_file_extent_offset(leaf, fi);
5025
5026 if (search_end == (u64)-1) {
5027 search_end = key.offset - ext_offset +
5028 btrfs_file_extent_ram_bytes(leaf, fi);
5029 }
5030
5031 if (!extent_locked) {
5032 lock_start = key.offset;
5033 lock_end = lock_start + num_bytes - 1;
5034 } else {
5035 if (lock_start > key.offset ||
5036 lock_end + 1 < key.offset + num_bytes) {
5037 unlock_extent(&BTRFS_I(inode)->io_tree,
5038 lock_start, lock_end, GFP_NOFS);
5039 extent_locked = 0;
5040 }
5041 }
5042
5043 if (!inode) {
5044 btrfs_release_path(root, path);
5045
5046 inode = btrfs_iget_locked(root->fs_info->sb,
5047 key.objectid, root);
5048 if (inode->i_state & I_NEW) {
5049 BTRFS_I(inode)->root = root;
5050 BTRFS_I(inode)->location.objectid =
5051 key.objectid;
5052 BTRFS_I(inode)->location.type =
5053 BTRFS_INODE_ITEM_KEY;
5054 BTRFS_I(inode)->location.offset = 0;
5055 btrfs_read_locked_inode(inode);
5056 unlock_new_inode(inode);
5057 }
5058 /*
5059 * some code call btrfs_commit_transaction while
5060 * holding the i_mutex, so we can't use mutex_lock
5061 * here.
5062 */
5063 if (is_bad_inode(inode) ||
5064 !mutex_trylock(&inode->i_mutex)) {
5065 iput(inode);
5066 inode = NULL;
5067 key.offset = (u64)-1;
5068 goto skip;
5069 }
5070 }
5071
5072 if (!extent_locked) {
5073 struct btrfs_ordered_extent *ordered;
5074
5075 btrfs_release_path(root, path);
5076
5077 lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
5078 lock_end, GFP_NOFS);
5079 ordered = btrfs_lookup_first_ordered_extent(inode,
5080 lock_end);
5081 if (ordered &&
5082 ordered->file_offset <= lock_end &&
5083 ordered->file_offset + ordered->len > lock_start) {
5084 unlock_extent(&BTRFS_I(inode)->io_tree,
5085 lock_start, lock_end, GFP_NOFS);
5086 btrfs_start_ordered_extent(inode, ordered, 1);
5087 btrfs_put_ordered_extent(ordered);
5088 key.offset += num_bytes;
5089 goto skip;
5090 }
5091 if (ordered)
5092 btrfs_put_ordered_extent(ordered);
5093
5094 extent_locked = 1;
5095 continue;
5096 }
5097
5098 if (nr_extents == 1) {
5099 /* update extent pointer in place */
5100 btrfs_set_file_extent_disk_bytenr(leaf, fi,
5101 new_extents[0].disk_bytenr);
5102 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
5103 new_extents[0].disk_num_bytes);
5104 btrfs_mark_buffer_dirty(leaf);
5105
5106 btrfs_drop_extent_cache(inode, key.offset,
5107 key.offset + num_bytes - 1, 0);
5108
5109 ret = btrfs_inc_extent_ref(trans, root,
5110 new_extents[0].disk_bytenr,
5111 new_extents[0].disk_num_bytes,
5112 leaf->start,
5113 root->root_key.objectid,
5114 trans->transid,
5115 key.objectid);
5116 BUG_ON(ret);
5117
5118 ret = btrfs_free_extent(trans, root,
5119 extent_key->objectid,
5120 extent_key->offset,
5121 leaf->start,
5122 btrfs_header_owner(leaf),
5123 btrfs_header_generation(leaf),
5124 key.objectid, 0);
5125 BUG_ON(ret);
5126
5127 btrfs_release_path(root, path);
5128 key.offset += num_bytes;
5129 } else {
5130 BUG_ON(1);
5131 #if 0
5132 u64 alloc_hint;
5133 u64 extent_len;
5134 int i;
5135 /*
5136 * drop old extent pointer at first, then insert the
5137 * new pointers one bye one
5138 */
5139 btrfs_release_path(root, path);
5140 ret = btrfs_drop_extents(trans, root, inode, key.offset,
5141 key.offset + num_bytes,
5142 key.offset, &alloc_hint);
5143 BUG_ON(ret);
5144
5145 for (i = 0; i < nr_extents; i++) {
5146 if (ext_offset >= new_extents[i].num_bytes) {
5147 ext_offset -= new_extents[i].num_bytes;
5148 continue;
5149 }
5150 extent_len = min(new_extents[i].num_bytes -
5151 ext_offset, num_bytes);
5152
5153 ret = btrfs_insert_empty_item(trans, root,
5154 path, &key,
5155 sizeof(*fi));
5156 BUG_ON(ret);
5157
5158 leaf = path->nodes[0];
5159 fi = btrfs_item_ptr(leaf, path->slots[0],
5160 struct btrfs_file_extent_item);
5161 btrfs_set_file_extent_generation(leaf, fi,
5162 trans->transid);
5163 btrfs_set_file_extent_type(leaf, fi,
5164 BTRFS_FILE_EXTENT_REG);
5165 btrfs_set_file_extent_disk_bytenr(leaf, fi,
5166 new_extents[i].disk_bytenr);
5167 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
5168 new_extents[i].disk_num_bytes);
5169 btrfs_set_file_extent_ram_bytes(leaf, fi,
5170 new_extents[i].ram_bytes);
5171
5172 btrfs_set_file_extent_compression(leaf, fi,
5173 new_extents[i].compression);
5174 btrfs_set_file_extent_encryption(leaf, fi,
5175 new_extents[i].encryption);
5176 btrfs_set_file_extent_other_encoding(leaf, fi,
5177 new_extents[i].other_encoding);
5178
5179 btrfs_set_file_extent_num_bytes(leaf, fi,
5180 extent_len);
5181 ext_offset += new_extents[i].offset;
5182 btrfs_set_file_extent_offset(leaf, fi,
5183 ext_offset);
5184 btrfs_mark_buffer_dirty(leaf);
5185
5186 btrfs_drop_extent_cache(inode, key.offset,
5187 key.offset + extent_len - 1, 0);
5188
5189 ret = btrfs_inc_extent_ref(trans, root,
5190 new_extents[i].disk_bytenr,
5191 new_extents[i].disk_num_bytes,
5192 leaf->start,
5193 root->root_key.objectid,
5194 trans->transid, key.objectid);
5195 BUG_ON(ret);
5196 btrfs_release_path(root, path);
5197
5198 inode_add_bytes(inode, extent_len);
5199
5200 ext_offset = 0;
5201 num_bytes -= extent_len;
5202 key.offset += extent_len;
5203
5204 if (num_bytes == 0)
5205 break;
5206 }
5207 BUG_ON(i >= nr_extents);
5208 #endif
5209 }
5210
5211 if (extent_locked) {
5212 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
5213 lock_end, GFP_NOFS);
5214 extent_locked = 0;
5215 }
5216 skip:
5217 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
5218 key.offset >= search_end)
5219 break;
5220
5221 cond_resched();
5222 }
5223 ret = 0;
5224 out:
5225 btrfs_release_path(root, path);
5226 if (inode) {
5227 mutex_unlock(&inode->i_mutex);
5228 if (extent_locked) {
5229 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
5230 lock_end, GFP_NOFS);
5231 }
5232 iput(inode);
5233 }
5234 return ret;
5235 }
5236
5237 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
5238 struct btrfs_root *root,
5239 struct extent_buffer *buf, u64 orig_start)
5240 {
5241 int level;
5242 int ret;
5243
5244 BUG_ON(btrfs_header_generation(buf) != trans->transid);
5245 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
5246
5247 level = btrfs_header_level(buf);
5248 if (level == 0) {
5249 struct btrfs_leaf_ref *ref;
5250 struct btrfs_leaf_ref *orig_ref;
5251
5252 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
5253 if (!orig_ref)
5254 return -ENOENT;
5255
5256 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
5257 if (!ref) {
5258 btrfs_free_leaf_ref(root, orig_ref);
5259 return -ENOMEM;
5260 }
5261
5262 ref->nritems = orig_ref->nritems;
5263 memcpy(ref->extents, orig_ref->extents,
5264 sizeof(ref->extents[0]) * ref->nritems);
5265
5266 btrfs_free_leaf_ref(root, orig_ref);
5267
5268 ref->root_gen = trans->transid;
5269 ref->bytenr = buf->start;
5270 ref->owner = btrfs_header_owner(buf);
5271 ref->generation = btrfs_header_generation(buf);
5272
5273 ret = btrfs_add_leaf_ref(root, ref, 0);
5274 WARN_ON(ret);
5275 btrfs_free_leaf_ref(root, ref);
5276 }
5277 return 0;
5278 }
5279
5280 static noinline int invalidate_extent_cache(struct btrfs_root *root,
5281 struct extent_buffer *leaf,
5282 struct btrfs_block_group_cache *group,
5283 struct btrfs_root *target_root)
5284 {
5285 struct btrfs_key key;
5286 struct inode *inode = NULL;
5287 struct btrfs_file_extent_item *fi;
5288 u64 num_bytes;
5289 u64 skip_objectid = 0;
5290 u32 nritems;
5291 u32 i;
5292
5293 nritems = btrfs_header_nritems(leaf);
5294 for (i = 0; i < nritems; i++) {
5295 btrfs_item_key_to_cpu(leaf, &key, i);
5296 if (key.objectid == skip_objectid ||
5297 key.type != BTRFS_EXTENT_DATA_KEY)
5298 continue;
5299 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
5300 if (btrfs_file_extent_type(leaf, fi) ==
5301 BTRFS_FILE_EXTENT_INLINE)
5302 continue;
5303 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
5304 continue;
5305 if (!inode || inode->i_ino != key.objectid) {
5306 iput(inode);
5307 inode = btrfs_ilookup(target_root->fs_info->sb,
5308 key.objectid, target_root, 1);
5309 }
5310 if (!inode) {
5311 skip_objectid = key.objectid;
5312 continue;
5313 }
5314 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
5315
5316 lock_extent(&BTRFS_I(inode)->io_tree, key.offset,
5317 key.offset + num_bytes - 1, GFP_NOFS);
5318 btrfs_drop_extent_cache(inode, key.offset,
5319 key.offset + num_bytes - 1, 1);
5320 unlock_extent(&BTRFS_I(inode)->io_tree, key.offset,
5321 key.offset + num_bytes - 1, GFP_NOFS);
5322 cond_resched();
5323 }
5324 iput(inode);
5325 return 0;
5326 }
5327
5328 static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans,
5329 struct btrfs_root *root,
5330 struct extent_buffer *leaf,
5331 struct btrfs_block_group_cache *group,
5332 struct inode *reloc_inode)
5333 {
5334 struct btrfs_key key;
5335 struct btrfs_key extent_key;
5336 struct btrfs_file_extent_item *fi;
5337 struct btrfs_leaf_ref *ref;
5338 struct disk_extent *new_extent;
5339 u64 bytenr;
5340 u64 num_bytes;
5341 u32 nritems;
5342 u32 i;
5343 int ext_index;
5344 int nr_extent;
5345 int ret;
5346
5347 new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
5348 BUG_ON(!new_extent);
5349
5350 ref = btrfs_lookup_leaf_ref(root, leaf->start);
5351 BUG_ON(!ref);
5352
5353 ext_index = -1;
5354 nritems = btrfs_header_nritems(leaf);
5355 for (i = 0; i < nritems; i++) {
5356 btrfs_item_key_to_cpu(leaf, &key, i);
5357 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
5358 continue;
5359 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
5360 if (btrfs_file_extent_type(leaf, fi) ==
5361 BTRFS_FILE_EXTENT_INLINE)
5362 continue;
5363 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5364 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
5365 if (bytenr == 0)
5366 continue;
5367
5368 ext_index++;
5369 if (bytenr >= group->key.objectid + group->key.offset ||
5370 bytenr + num_bytes <= group->key.objectid)
5371 continue;
5372
5373 extent_key.objectid = bytenr;
5374 extent_key.offset = num_bytes;
5375 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
5376 nr_extent = 1;
5377 ret = get_new_locations(reloc_inode, &extent_key,
5378 group->key.objectid, 1,
5379 &new_extent, &nr_extent);
5380 if (ret > 0)
5381 continue;
5382 BUG_ON(ret < 0);
5383
5384 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
5385 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
5386 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
5387 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
5388
5389 btrfs_set_file_extent_disk_bytenr(leaf, fi,
5390 new_extent->disk_bytenr);
5391 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
5392 new_extent->disk_num_bytes);
5393 btrfs_mark_buffer_dirty(leaf);
5394
5395 ret = btrfs_inc_extent_ref(trans, root,
5396 new_extent->disk_bytenr,
5397 new_extent->disk_num_bytes,
5398 leaf->start,
5399 root->root_key.objectid,
5400 trans->transid, key.objectid);
5401 BUG_ON(ret);
5402 ret = btrfs_free_extent(trans, root,
5403 bytenr, num_bytes, leaf->start,
5404 btrfs_header_owner(leaf),
5405 btrfs_header_generation(leaf),
5406 key.objectid, 0);
5407 BUG_ON(ret);
5408 cond_resched();
5409 }
5410 kfree(new_extent);
5411 BUG_ON(ext_index + 1 != ref->nritems);
5412 btrfs_free_leaf_ref(root, ref);
5413 return 0;
5414 }
5415
5416 int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
5417 struct btrfs_root *root)
5418 {
5419 struct btrfs_root *reloc_root;
5420 int ret;
5421
5422 if (root->reloc_root) {
5423 reloc_root = root->reloc_root;
5424 root->reloc_root = NULL;
5425 list_add(&reloc_root->dead_list,
5426 &root->fs_info->dead_reloc_roots);
5427
5428 btrfs_set_root_bytenr(&reloc_root->root_item,
5429 reloc_root->node->start);
5430 btrfs_set_root_level(&root->root_item,
5431 btrfs_header_level(reloc_root->node));
5432 memset(&reloc_root->root_item.drop_progress, 0,
5433 sizeof(struct btrfs_disk_key));
5434 reloc_root->root_item.drop_level = 0;
5435
5436 ret = btrfs_update_root(trans, root->fs_info->tree_root,
5437 &reloc_root->root_key,
5438 &reloc_root->root_item);
5439 BUG_ON(ret);
5440 }
5441 return 0;
5442 }
5443
5444 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
5445 {
5446 struct btrfs_trans_handle *trans;
5447 struct btrfs_root *reloc_root;
5448 struct btrfs_root *prev_root = NULL;
5449 struct list_head dead_roots;
5450 int ret;
5451 unsigned long nr;
5452
5453 INIT_LIST_HEAD(&dead_roots);
5454 list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
5455
5456 while (!list_empty(&dead_roots)) {
5457 reloc_root = list_entry(dead_roots.prev,
5458 struct btrfs_root, dead_list);
5459 list_del_init(&reloc_root->dead_list);
5460
5461 BUG_ON(reloc_root->commit_root != NULL);
5462 while (1) {
5463 trans = btrfs_join_transaction(root, 1);
5464 BUG_ON(!trans);
5465
5466 mutex_lock(&root->fs_info->drop_mutex);
5467 ret = btrfs_drop_snapshot(trans, reloc_root);
5468 if (ret != -EAGAIN)
5469 break;
5470 mutex_unlock(&root->fs_info->drop_mutex);
5471
5472 nr = trans->blocks_used;
5473 ret = btrfs_end_transaction(trans, root);
5474 BUG_ON(ret);
5475 btrfs_btree_balance_dirty(root, nr);
5476 }
5477
5478 free_extent_buffer(reloc_root->node);
5479
5480 ret = btrfs_del_root(trans, root->fs_info->tree_root,
5481 &reloc_root->root_key);
5482 BUG_ON(ret);
5483 mutex_unlock(&root->fs_info->drop_mutex);
5484
5485 nr = trans->blocks_used;
5486 ret = btrfs_end_transaction(trans, root);
5487 BUG_ON(ret);
5488 btrfs_btree_balance_dirty(root, nr);
5489
5490 kfree(prev_root);
5491 prev_root = reloc_root;
5492 }
5493 if (prev_root) {
5494 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
5495 kfree(prev_root);
5496 }
5497 return 0;
5498 }
5499
5500 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
5501 {
5502 list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
5503 return 0;
5504 }
5505
5506 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
5507 {
5508 struct btrfs_root *reloc_root;
5509 struct btrfs_trans_handle *trans;
5510 struct btrfs_key location;
5511 int found;
5512 int ret;
5513
5514 mutex_lock(&root->fs_info->tree_reloc_mutex);
5515 ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
5516 BUG_ON(ret);
5517 found = !list_empty(&root->fs_info->dead_reloc_roots);
5518 mutex_unlock(&root->fs_info->tree_reloc_mutex);
5519
5520 if (found) {
5521 trans = btrfs_start_transaction(root, 1);
5522 BUG_ON(!trans);
5523 ret = btrfs_commit_transaction(trans, root);
5524 BUG_ON(ret);
5525 }
5526
5527 location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
5528 location.offset = (u64)-1;
5529 location.type = BTRFS_ROOT_ITEM_KEY;
5530
5531 reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
5532 BUG_ON(!reloc_root);
5533 btrfs_orphan_cleanup(reloc_root);
5534 return 0;
5535 }
5536
5537 static noinline int init_reloc_tree(struct btrfs_trans_handle *trans,
5538 struct btrfs_root *root)
5539 {
5540 struct btrfs_root *reloc_root;
5541 struct extent_buffer *eb;
5542 struct btrfs_root_item *root_item;
5543 struct btrfs_key root_key;
5544 int ret;
5545
5546 BUG_ON(!root->ref_cows);
5547 if (root->reloc_root)
5548 return 0;
5549
5550 root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
5551 BUG_ON(!root_item);
5552
5553 ret = btrfs_copy_root(trans, root, root->commit_root,
5554 &eb, BTRFS_TREE_RELOC_OBJECTID);
5555 BUG_ON(ret);
5556
5557 root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
5558 root_key.offset = root->root_key.objectid;
5559 root_key.type = BTRFS_ROOT_ITEM_KEY;
5560
5561 memcpy(root_item, &root->root_item, sizeof(root_item));
5562 btrfs_set_root_refs(root_item, 0);
5563 btrfs_set_root_bytenr(root_item, eb->start);
5564 btrfs_set_root_level(root_item, btrfs_header_level(eb));
5565 btrfs_set_root_generation(root_item, trans->transid);
5566
5567 btrfs_tree_unlock(eb);
5568 free_extent_buffer(eb);
5569
5570 ret = btrfs_insert_root(trans, root->fs_info->tree_root,
5571 &root_key, root_item);
5572 BUG_ON(ret);
5573 kfree(root_item);
5574
5575 reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
5576 &root_key);
5577 BUG_ON(!reloc_root);
5578 reloc_root->last_trans = trans->transid;
5579 reloc_root->commit_root = NULL;
5580 reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
5581
5582 root->reloc_root = reloc_root;
5583 return 0;
5584 }
5585
5586 /*
5587 * Core function of space balance.
5588 *
5589 * The idea is using reloc trees to relocate tree blocks in reference
5590 * counted roots. There is one reloc tree for each subvol, and all
5591 * reloc trees share same root key objectid. Reloc trees are snapshots
5592 * of the latest committed roots of subvols (root->commit_root).
5593 *
5594 * To relocate a tree block referenced by a subvol, there are two steps.
5595 * COW the block through subvol's reloc tree, then update block pointer
5596 * in the subvol to point to the new block. Since all reloc trees share
5597 * same root key objectid, doing special handing for tree blocks owned
5598 * by them is easy. Once a tree block has been COWed in one reloc tree,
5599 * we can use the resulting new block directly when the same block is
5600 * required to COW again through other reloc trees. By this way, relocated
5601 * tree blocks are shared between reloc trees, so they are also shared
5602 * between subvols.
5603 */
5604 static noinline int relocate_one_path(struct btrfs_trans_handle *trans,
5605 struct btrfs_root *root,
5606 struct btrfs_path *path,
5607 struct btrfs_key *first_key,
5608 struct btrfs_ref_path *ref_path,
5609 struct btrfs_block_group_cache *group,
5610 struct inode *reloc_inode)
5611 {
5612 struct btrfs_root *reloc_root;
5613 struct extent_buffer *eb = NULL;
5614 struct btrfs_key *keys;
5615 u64 *nodes;
5616 int level;
5617 int shared_level;
5618 int lowest_level = 0;
5619 int ret;
5620
5621 if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
5622 lowest_level = ref_path->owner_objectid;
5623
5624 if (!root->ref_cows) {
5625 path->lowest_level = lowest_level;
5626 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
5627 BUG_ON(ret < 0);
5628 path->lowest_level = 0;
5629 btrfs_release_path(root, path);
5630 return 0;
5631 }
5632
5633 mutex_lock(&root->fs_info->tree_reloc_mutex);
5634 ret = init_reloc_tree(trans, root);
5635 BUG_ON(ret);
5636 reloc_root = root->reloc_root;
5637
5638 shared_level = ref_path->shared_level;
5639 ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
5640
5641 keys = ref_path->node_keys;
5642 nodes = ref_path->new_nodes;
5643 memset(&keys[shared_level + 1], 0,
5644 sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
5645 memset(&nodes[shared_level + 1], 0,
5646 sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
5647
5648 if (nodes[lowest_level] == 0) {
5649 path->lowest_level = lowest_level;
5650 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
5651 0, 1);
5652 BUG_ON(ret);
5653 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
5654 eb = path->nodes[level];
5655 if (!eb || eb == reloc_root->node)
5656 break;
5657 nodes[level] = eb->start;
5658 if (level == 0)
5659 btrfs_item_key_to_cpu(eb, &keys[level], 0);
5660 else
5661 btrfs_node_key_to_cpu(eb, &keys[level], 0);
5662 }
5663 if (nodes[0] &&
5664 ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5665 eb = path->nodes[0];
5666 ret = replace_extents_in_leaf(trans, reloc_root, eb,
5667 group, reloc_inode);
5668 BUG_ON(ret);
5669 }
5670 btrfs_release_path(reloc_root, path);
5671 } else {
5672 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
5673 lowest_level);
5674 BUG_ON(ret);
5675 }
5676
5677 /*
5678 * replace tree blocks in the fs tree with tree blocks in
5679 * the reloc tree.
5680 */
5681 ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
5682 BUG_ON(ret < 0);
5683
5684 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5685 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
5686 0, 0);
5687 BUG_ON(ret);
5688 extent_buffer_get(path->nodes[0]);
5689 eb = path->nodes[0];
5690 btrfs_release_path(reloc_root, path);
5691 ret = invalidate_extent_cache(reloc_root, eb, group, root);
5692 BUG_ON(ret);
5693 free_extent_buffer(eb);
5694 }
5695
5696 mutex_unlock(&root->fs_info->tree_reloc_mutex);
5697 path->lowest_level = 0;
5698 return 0;
5699 }
5700
5701 static noinline int relocate_tree_block(struct btrfs_trans_handle *trans,
5702 struct btrfs_root *root,
5703 struct btrfs_path *path,
5704 struct btrfs_key *first_key,
5705 struct btrfs_ref_path *ref_path)
5706 {
5707 int ret;
5708
5709 ret = relocate_one_path(trans, root, path, first_key,
5710 ref_path, NULL, NULL);
5711 BUG_ON(ret);
5712
5713 if (root == root->fs_info->extent_root)
5714 btrfs_extent_post_op(trans, root);
5715
5716 return 0;
5717 }
5718
5719 static noinline int del_extent_zero(struct btrfs_trans_handle *trans,
5720 struct btrfs_root *extent_root,
5721 struct btrfs_path *path,
5722 struct btrfs_key *extent_key)
5723 {
5724 int ret;
5725
5726 ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
5727 if (ret)
5728 goto out;
5729 ret = btrfs_del_item(trans, extent_root, path);
5730 out:
5731 btrfs_release_path(extent_root, path);
5732 return ret;
5733 }
5734
5735 static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info,
5736 struct btrfs_ref_path *ref_path)
5737 {
5738 struct btrfs_key root_key;
5739
5740 root_key.objectid = ref_path->root_objectid;
5741 root_key.type = BTRFS_ROOT_ITEM_KEY;
5742 if (is_cowonly_root(ref_path->root_objectid))
5743 root_key.offset = 0;
5744 else
5745 root_key.offset = (u64)-1;
5746
5747 return btrfs_read_fs_root_no_name(fs_info, &root_key);
5748 }
5749
5750 static noinline int relocate_one_extent(struct btrfs_root *extent_root,
5751 struct btrfs_path *path,
5752 struct btrfs_key *extent_key,
5753 struct btrfs_block_group_cache *group,
5754 struct inode *reloc_inode, int pass)
5755 {
5756 struct btrfs_trans_handle *trans;
5757 struct btrfs_root *found_root;
5758 struct btrfs_ref_path *ref_path = NULL;
5759 struct disk_extent *new_extents = NULL;
5760 int nr_extents = 0;
5761 int loops;
5762 int ret;
5763 int level;
5764 struct btrfs_key first_key;
5765 u64 prev_block = 0;
5766
5767
5768 trans = btrfs_start_transaction(extent_root, 1);
5769 BUG_ON(!trans);
5770
5771 if (extent_key->objectid == 0) {
5772 ret = del_extent_zero(trans, extent_root, path, extent_key);
5773 goto out;
5774 }
5775
5776 ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
5777 if (!ref_path) {
5778 ret = -ENOMEM;
5779 goto out;
5780 }
5781
5782 for (loops = 0; ; loops++) {
5783 if (loops == 0) {
5784 ret = btrfs_first_ref_path(trans, extent_root, ref_path,
5785 extent_key->objectid);
5786 } else {
5787 ret = btrfs_next_ref_path(trans, extent_root, ref_path);
5788 }
5789 if (ret < 0)
5790 goto out;
5791 if (ret > 0)
5792 break;
5793
5794 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
5795 ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
5796 continue;
5797
5798 found_root = read_ref_root(extent_root->fs_info, ref_path);
5799 BUG_ON(!found_root);
5800 /*
5801 * for reference counted tree, only process reference paths
5802 * rooted at the latest committed root.
5803 */
5804 if (found_root->ref_cows &&
5805 ref_path->root_generation != found_root->root_key.offset)
5806 continue;
5807
5808 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5809 if (pass == 0) {
5810 /*
5811 * copy data extents to new locations
5812 */
5813 u64 group_start = group->key.objectid;
5814 ret = relocate_data_extent(reloc_inode,
5815 extent_key,
5816 group_start);
5817 if (ret < 0)
5818 goto out;
5819 break;
5820 }
5821 level = 0;
5822 } else {
5823 level = ref_path->owner_objectid;
5824 }
5825
5826 if (prev_block != ref_path->nodes[level]) {
5827 struct extent_buffer *eb;
5828 u64 block_start = ref_path->nodes[level];
5829 u64 block_size = btrfs_level_size(found_root, level);
5830
5831 eb = read_tree_block(found_root, block_start,
5832 block_size, 0);
5833 btrfs_tree_lock(eb);
5834 BUG_ON(level != btrfs_header_level(eb));
5835
5836 if (level == 0)
5837 btrfs_item_key_to_cpu(eb, &first_key, 0);
5838 else
5839 btrfs_node_key_to_cpu(eb, &first_key, 0);
5840
5841 btrfs_tree_unlock(eb);
5842 free_extent_buffer(eb);
5843 prev_block = block_start;
5844 }
5845
5846 mutex_lock(&extent_root->fs_info->trans_mutex);
5847 btrfs_record_root_in_trans(found_root);
5848 mutex_unlock(&extent_root->fs_info->trans_mutex);
5849 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5850 /*
5851 * try to update data extent references while
5852 * keeping metadata shared between snapshots.
5853 */
5854 if (pass == 1) {
5855 ret = relocate_one_path(trans, found_root,
5856 path, &first_key, ref_path,
5857 group, reloc_inode);
5858 if (ret < 0)
5859 goto out;
5860 continue;
5861 }
5862 /*
5863 * use fallback method to process the remaining
5864 * references.
5865 */
5866 if (!new_extents) {
5867 u64 group_start = group->key.objectid;
5868 new_extents = kmalloc(sizeof(*new_extents),
5869 GFP_NOFS);
5870 nr_extents = 1;
5871 ret = get_new_locations(reloc_inode,
5872 extent_key,
5873 group_start, 1,
5874 &new_extents,
5875 &nr_extents);
5876 if (ret)
5877 goto out;
5878 }
5879 ret = replace_one_extent(trans, found_root,
5880 path, extent_key,
5881 &first_key, ref_path,
5882 new_extents, nr_extents);
5883 } else {
5884 ret = relocate_tree_block(trans, found_root, path,
5885 &first_key, ref_path);
5886 }
5887 if (ret < 0)
5888 goto out;
5889 }
5890 ret = 0;
5891 out:
5892 btrfs_end_transaction(trans, extent_root);
5893 kfree(new_extents);
5894 kfree(ref_path);
5895 return ret;
5896 }
5897
5898 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
5899 {
5900 u64 num_devices;
5901 u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
5902 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
5903
5904 num_devices = root->fs_info->fs_devices->rw_devices;
5905 if (num_devices == 1) {
5906 stripped |= BTRFS_BLOCK_GROUP_DUP;
5907 stripped = flags & ~stripped;
5908
5909 /* turn raid0 into single device chunks */
5910 if (flags & BTRFS_BLOCK_GROUP_RAID0)
5911 return stripped;
5912
5913 /* turn mirroring into duplication */
5914 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
5915 BTRFS_BLOCK_GROUP_RAID10))
5916 return stripped | BTRFS_BLOCK_GROUP_DUP;
5917 return flags;
5918 } else {
5919 /* they already had raid on here, just return */
5920 if (flags & stripped)
5921 return flags;
5922
5923 stripped |= BTRFS_BLOCK_GROUP_DUP;
5924 stripped = flags & ~stripped;
5925
5926 /* switch duplicated blocks with raid1 */
5927 if (flags & BTRFS_BLOCK_GROUP_DUP)
5928 return stripped | BTRFS_BLOCK_GROUP_RAID1;
5929
5930 /* turn single device chunks into raid0 */
5931 return stripped | BTRFS_BLOCK_GROUP_RAID0;
5932 }
5933 return flags;
5934 }
5935
5936 static int __alloc_chunk_for_shrink(struct btrfs_root *root,
5937 struct btrfs_block_group_cache *shrink_block_group,
5938 int force)
5939 {
5940 struct btrfs_trans_handle *trans;
5941 u64 new_alloc_flags;
5942 u64 calc;
5943
5944 spin_lock(&shrink_block_group->lock);
5945 if (btrfs_block_group_used(&shrink_block_group->item) > 0) {
5946 spin_unlock(&shrink_block_group->lock);
5947
5948 trans = btrfs_start_transaction(root, 1);
5949 spin_lock(&shrink_block_group->lock);
5950
5951 new_alloc_flags = update_block_group_flags(root,
5952 shrink_block_group->flags);
5953 if (new_alloc_flags != shrink_block_group->flags) {
5954 calc =
5955 btrfs_block_group_used(&shrink_block_group->item);
5956 } else {
5957 calc = shrink_block_group->key.offset;
5958 }
5959 spin_unlock(&shrink_block_group->lock);
5960
5961 do_chunk_alloc(trans, root->fs_info->extent_root,
5962 calc + 2 * 1024 * 1024, new_alloc_flags, force);
5963
5964 btrfs_end_transaction(trans, root);
5965 } else
5966 spin_unlock(&shrink_block_group->lock);
5967 return 0;
5968 }
5969
5970 static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
5971 struct btrfs_root *root,
5972 u64 objectid, u64 size)
5973 {
5974 struct btrfs_path *path;
5975 struct btrfs_inode_item *item;
5976 struct extent_buffer *leaf;
5977 int ret;
5978
5979 path = btrfs_alloc_path();
5980 if (!path)
5981 return -ENOMEM;
5982
5983 ret = btrfs_insert_empty_inode(trans, root, path, objectid);
5984 if (ret)
5985 goto out;
5986
5987 leaf = path->nodes[0];
5988 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
5989 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
5990 btrfs_set_inode_generation(leaf, item, 1);
5991 btrfs_set_inode_size(leaf, item, size);
5992 btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
5993 btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS);
5994 btrfs_mark_buffer_dirty(leaf);
5995 btrfs_release_path(root, path);
5996 out:
5997 btrfs_free_path(path);
5998 return ret;
5999 }
6000
6001 static noinline struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info,
6002 struct btrfs_block_group_cache *group)
6003 {
6004 struct inode *inode = NULL;
6005 struct btrfs_trans_handle *trans;
6006 struct btrfs_root *root;
6007 struct btrfs_key root_key;
6008 u64 objectid = BTRFS_FIRST_FREE_OBJECTID;
6009 int err = 0;
6010
6011 root_key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
6012 root_key.type = BTRFS_ROOT_ITEM_KEY;
6013 root_key.offset = (u64)-1;
6014 root = btrfs_read_fs_root_no_name(fs_info, &root_key);
6015 if (IS_ERR(root))
6016 return ERR_CAST(root);
6017
6018 trans = btrfs_start_transaction(root, 1);
6019 BUG_ON(!trans);
6020
6021 err = btrfs_find_free_objectid(trans, root, objectid, &objectid);
6022 if (err)
6023 goto out;
6024
6025 err = __insert_orphan_inode(trans, root, objectid, group->key.offset);
6026 BUG_ON(err);
6027
6028 err = btrfs_insert_file_extent(trans, root, objectid, 0, 0, 0,
6029 group->key.offset, 0, group->key.offset,
6030 0, 0, 0);
6031 BUG_ON(err);
6032
6033 inode = btrfs_iget_locked(root->fs_info->sb, objectid, root);
6034 if (inode->i_state & I_NEW) {
6035 BTRFS_I(inode)->root = root;
6036 BTRFS_I(inode)->location.objectid = objectid;
6037 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
6038 BTRFS_I(inode)->location.offset = 0;
6039 btrfs_read_locked_inode(inode);
6040 unlock_new_inode(inode);
6041 BUG_ON(is_bad_inode(inode));
6042 } else {
6043 BUG_ON(1);
6044 }
6045 BTRFS_I(inode)->index_cnt = group->key.objectid;
6046
6047 err = btrfs_orphan_add(trans, inode);
6048 out:
6049 btrfs_end_transaction(trans, root);
6050 if (err) {
6051 if (inode)
6052 iput(inode);
6053 inode = ERR_PTR(err);
6054 }
6055 return inode;
6056 }
6057
6058 int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len)
6059 {
6060
6061 struct btrfs_ordered_sum *sums;
6062 struct btrfs_sector_sum *sector_sum;
6063 struct btrfs_ordered_extent *ordered;
6064 struct btrfs_root *root = BTRFS_I(inode)->root;
6065 struct list_head list;
6066 size_t offset;
6067 int ret;
6068 u64 disk_bytenr;
6069
6070 INIT_LIST_HEAD(&list);
6071
6072 ordered = btrfs_lookup_ordered_extent(inode, file_pos);
6073 BUG_ON(ordered->file_offset != file_pos || ordered->len != len);
6074
6075 disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt;
6076 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, disk_bytenr,
6077 disk_bytenr + len - 1, &list);
6078
6079 while (!list_empty(&list)) {
6080 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
6081 list_del_init(&sums->list);
6082
6083 sector_sum = sums->sums;
6084 sums->bytenr = ordered->start;
6085
6086 offset = 0;
6087 while (offset < sums->len) {
6088 sector_sum->bytenr += ordered->start - disk_bytenr;
6089 sector_sum++;
6090 offset += root->sectorsize;
6091 }
6092
6093 btrfs_add_ordered_sum(inode, ordered, sums);
6094 }
6095 btrfs_put_ordered_extent(ordered);
6096 return 0;
6097 }
6098
6099 int btrfs_relocate_block_group(struct btrfs_root *root, u64 group_start)
6100 {
6101 struct btrfs_trans_handle *trans;
6102 struct btrfs_path *path;
6103 struct btrfs_fs_info *info = root->fs_info;
6104 struct extent_buffer *leaf;
6105 struct inode *reloc_inode;
6106 struct btrfs_block_group_cache *block_group;
6107 struct btrfs_key key;
6108 u64 skipped;
6109 u64 cur_byte;
6110 u64 total_found;
6111 u32 nritems;
6112 int ret;
6113 int progress;
6114 int pass = 0;
6115
6116 root = root->fs_info->extent_root;
6117
6118 block_group = btrfs_lookup_block_group(info, group_start);
6119 BUG_ON(!block_group);
6120
6121 printk(KERN_INFO "btrfs relocating block group %llu flags %llu\n",
6122 (unsigned long long)block_group->key.objectid,
6123 (unsigned long long)block_group->flags);
6124
6125 path = btrfs_alloc_path();
6126 BUG_ON(!path);
6127
6128 reloc_inode = create_reloc_inode(info, block_group);
6129 BUG_ON(IS_ERR(reloc_inode));
6130
6131 __alloc_chunk_for_shrink(root, block_group, 1);
6132 set_block_group_readonly(block_group);
6133
6134 btrfs_start_delalloc_inodes(info->tree_root);
6135 btrfs_wait_ordered_extents(info->tree_root, 0);
6136 again:
6137 skipped = 0;
6138 total_found = 0;
6139 progress = 0;
6140 key.objectid = block_group->key.objectid;
6141 key.offset = 0;
6142 key.type = 0;
6143 cur_byte = key.objectid;
6144
6145 trans = btrfs_start_transaction(info->tree_root, 1);
6146 btrfs_commit_transaction(trans, info->tree_root);
6147
6148 mutex_lock(&root->fs_info->cleaner_mutex);
6149 btrfs_clean_old_snapshots(info->tree_root);
6150 btrfs_remove_leaf_refs(info->tree_root, (u64)-1, 1);
6151 mutex_unlock(&root->fs_info->cleaner_mutex);
6152
6153 while (1) {
6154 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6155 if (ret < 0)
6156 goto out;
6157 next:
6158 leaf = path->nodes[0];
6159 nritems = btrfs_header_nritems(leaf);
6160 if (path->slots[0] >= nritems) {
6161 ret = btrfs_next_leaf(root, path);
6162 if (ret < 0)
6163 goto out;
6164 if (ret == 1) {
6165 ret = 0;
6166 break;
6167 }
6168 leaf = path->nodes[0];
6169 nritems = btrfs_header_nritems(leaf);
6170 }
6171
6172 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
6173
6174 if (key.objectid >= block_group->key.objectid +
6175 block_group->key.offset)
6176 break;
6177
6178 if (progress && need_resched()) {
6179 btrfs_release_path(root, path);
6180 cond_resched();
6181 progress = 0;
6182 continue;
6183 }
6184 progress = 1;
6185
6186 if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY ||
6187 key.objectid + key.offset <= cur_byte) {
6188 path->slots[0]++;
6189 goto next;
6190 }
6191
6192 total_found++;
6193 cur_byte = key.objectid + key.offset;
6194 btrfs_release_path(root, path);
6195
6196 __alloc_chunk_for_shrink(root, block_group, 0);
6197 ret = relocate_one_extent(root, path, &key, block_group,
6198 reloc_inode, pass);
6199 BUG_ON(ret < 0);
6200 if (ret > 0)
6201 skipped++;
6202
6203 key.objectid = cur_byte;
6204 key.type = 0;
6205 key.offset = 0;
6206 }
6207
6208 btrfs_release_path(root, path);
6209
6210 if (pass == 0) {
6211 btrfs_wait_ordered_range(reloc_inode, 0, (u64)-1);
6212 invalidate_mapping_pages(reloc_inode->i_mapping, 0, -1);
6213 }
6214
6215 if (total_found > 0) {
6216 printk(KERN_INFO "btrfs found %llu extents in pass %d\n",
6217 (unsigned long long)total_found, pass);
6218 pass++;
6219 if (total_found == skipped && pass > 2) {
6220 iput(reloc_inode);
6221 reloc_inode = create_reloc_inode(info, block_group);
6222 pass = 0;
6223 }
6224 goto again;
6225 }
6226
6227 /* delete reloc_inode */
6228 iput(reloc_inode);
6229
6230 /* unpin extents in this range */
6231 trans = btrfs_start_transaction(info->tree_root, 1);
6232 btrfs_commit_transaction(trans, info->tree_root);
6233
6234 spin_lock(&block_group->lock);
6235 WARN_ON(block_group->pinned > 0);
6236 WARN_ON(block_group->reserved > 0);
6237 WARN_ON(btrfs_block_group_used(&block_group->item) > 0);
6238 spin_unlock(&block_group->lock);
6239 put_block_group(block_group);
6240 ret = 0;
6241 out:
6242 btrfs_free_path(path);
6243 return ret;
6244 }
6245
6246 static int find_first_block_group(struct btrfs_root *root,
6247 struct btrfs_path *path, struct btrfs_key *key)
6248 {
6249 int ret = 0;
6250 struct btrfs_key found_key;
6251 struct extent_buffer *leaf;
6252 int slot;
6253
6254 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
6255 if (ret < 0)
6256 goto out;
6257
6258 while (1) {
6259 slot = path->slots[0];
6260 leaf = path->nodes[0];
6261 if (slot >= btrfs_header_nritems(leaf)) {
6262 ret = btrfs_next_leaf(root, path);
6263 if (ret == 0)
6264 continue;
6265 if (ret < 0)
6266 goto out;
6267 break;
6268 }
6269 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6270
6271 if (found_key.objectid >= key->objectid &&
6272 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
6273 ret = 0;
6274 goto out;
6275 }
6276 path->slots[0]++;
6277 }
6278 ret = -ENOENT;
6279 out:
6280 return ret;
6281 }
6282
6283 int btrfs_free_block_groups(struct btrfs_fs_info *info)
6284 {
6285 struct btrfs_block_group_cache *block_group;
6286 struct rb_node *n;
6287
6288 spin_lock(&info->block_group_cache_lock);
6289 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
6290 block_group = rb_entry(n, struct btrfs_block_group_cache,
6291 cache_node);
6292 rb_erase(&block_group->cache_node,
6293 &info->block_group_cache_tree);
6294 spin_unlock(&info->block_group_cache_lock);
6295
6296 btrfs_remove_free_space_cache(block_group);
6297 down_write(&block_group->space_info->groups_sem);
6298 list_del(&block_group->list);
6299 up_write(&block_group->space_info->groups_sem);
6300
6301 WARN_ON(atomic_read(&block_group->count) != 1);
6302 kfree(block_group);
6303
6304 spin_lock(&info->block_group_cache_lock);
6305 }
6306 spin_unlock(&info->block_group_cache_lock);
6307 return 0;
6308 }
6309
6310 int btrfs_read_block_groups(struct btrfs_root *root)
6311 {
6312 struct btrfs_path *path;
6313 int ret;
6314 struct btrfs_block_group_cache *cache;
6315 struct btrfs_fs_info *info = root->fs_info;
6316 struct btrfs_space_info *space_info;
6317 struct btrfs_key key;
6318 struct btrfs_key found_key;
6319 struct extent_buffer *leaf;
6320
6321 root = info->extent_root;
6322 key.objectid = 0;
6323 key.offset = 0;
6324 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
6325 path = btrfs_alloc_path();
6326 if (!path)
6327 return -ENOMEM;
6328
6329 while (1) {
6330 ret = find_first_block_group(root, path, &key);
6331 if (ret > 0) {
6332 ret = 0;
6333 goto error;
6334 }
6335 if (ret != 0)
6336 goto error;
6337
6338 leaf = path->nodes[0];
6339 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6340 cache = kzalloc(sizeof(*cache), GFP_NOFS);
6341 if (!cache) {
6342 ret = -ENOMEM;
6343 break;
6344 }
6345
6346 atomic_set(&cache->count, 1);
6347 spin_lock_init(&cache->lock);
6348 mutex_init(&cache->alloc_mutex);
6349 mutex_init(&cache->cache_mutex);
6350 INIT_LIST_HEAD(&cache->list);
6351 read_extent_buffer(leaf, &cache->item,
6352 btrfs_item_ptr_offset(leaf, path->slots[0]),
6353 sizeof(cache->item));
6354 memcpy(&cache->key, &found_key, sizeof(found_key));
6355
6356 key.objectid = found_key.objectid + found_key.offset;
6357 btrfs_release_path(root, path);
6358 cache->flags = btrfs_block_group_flags(&cache->item);
6359
6360 ret = update_space_info(info, cache->flags, found_key.offset,
6361 btrfs_block_group_used(&cache->item),
6362 &space_info);
6363 BUG_ON(ret);
6364 cache->space_info = space_info;
6365 down_write(&space_info->groups_sem);
6366 list_add_tail(&cache->list, &space_info->block_groups);
6367 up_write(&space_info->groups_sem);
6368
6369 ret = btrfs_add_block_group_cache(root->fs_info, cache);
6370 BUG_ON(ret);
6371
6372 set_avail_alloc_bits(root->fs_info, cache->flags);
6373 if (btrfs_chunk_readonly(root, cache->key.objectid))
6374 set_block_group_readonly(cache);
6375 }
6376 ret = 0;
6377 error:
6378 btrfs_free_path(path);
6379 return ret;
6380 }
6381
6382 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
6383 struct btrfs_root *root, u64 bytes_used,
6384 u64 type, u64 chunk_objectid, u64 chunk_offset,
6385 u64 size)
6386 {
6387 int ret;
6388 struct btrfs_root *extent_root;
6389 struct btrfs_block_group_cache *cache;
6390
6391 extent_root = root->fs_info->extent_root;
6392
6393 root->fs_info->last_trans_new_blockgroup = trans->transid;
6394
6395 cache = kzalloc(sizeof(*cache), GFP_NOFS);
6396 if (!cache)
6397 return -ENOMEM;
6398
6399 cache->key.objectid = chunk_offset;
6400 cache->key.offset = size;
6401 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
6402 atomic_set(&cache->count, 1);
6403 spin_lock_init(&cache->lock);
6404 mutex_init(&cache->alloc_mutex);
6405 mutex_init(&cache->cache_mutex);
6406 INIT_LIST_HEAD(&cache->list);
6407
6408 btrfs_set_block_group_used(&cache->item, bytes_used);
6409 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
6410 cache->flags = type;
6411 btrfs_set_block_group_flags(&cache->item, type);
6412
6413 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
6414 &cache->space_info);
6415 BUG_ON(ret);
6416 down_write(&cache->space_info->groups_sem);
6417 list_add_tail(&cache->list, &cache->space_info->block_groups);
6418 up_write(&cache->space_info->groups_sem);
6419
6420 ret = btrfs_add_block_group_cache(root->fs_info, cache);
6421 BUG_ON(ret);
6422
6423 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
6424 sizeof(cache->item));
6425 BUG_ON(ret);
6426
6427 finish_current_insert(trans, extent_root, 0);
6428 ret = del_pending_extents(trans, extent_root, 0);
6429 BUG_ON(ret);
6430 set_avail_alloc_bits(extent_root->fs_info, type);
6431
6432 return 0;
6433 }
6434
6435 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
6436 struct btrfs_root *root, u64 group_start)
6437 {
6438 struct btrfs_path *path;
6439 struct btrfs_block_group_cache *block_group;
6440 struct btrfs_key key;
6441 int ret;
6442
6443 root = root->fs_info->extent_root;
6444
6445 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
6446 BUG_ON(!block_group);
6447 BUG_ON(!block_group->ro);
6448
6449 memcpy(&key, &block_group->key, sizeof(key));
6450
6451 path = btrfs_alloc_path();
6452 BUG_ON(!path);
6453
6454 spin_lock(&root->fs_info->block_group_cache_lock);
6455 rb_erase(&block_group->cache_node,
6456 &root->fs_info->block_group_cache_tree);
6457 spin_unlock(&root->fs_info->block_group_cache_lock);
6458 btrfs_remove_free_space_cache(block_group);
6459 down_write(&block_group->space_info->groups_sem);
6460 list_del(&block_group->list);
6461 up_write(&block_group->space_info->groups_sem);
6462
6463 spin_lock(&block_group->space_info->lock);
6464 block_group->space_info->total_bytes -= block_group->key.offset;
6465 block_group->space_info->bytes_readonly -= block_group->key.offset;
6466 spin_unlock(&block_group->space_info->lock);
6467 block_group->space_info->full = 0;
6468
6469 put_block_group(block_group);
6470 put_block_group(block_group);
6471
6472 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
6473 if (ret > 0)
6474 ret = -EIO;
6475 if (ret < 0)
6476 goto out;
6477
6478 ret = btrfs_del_item(trans, root, path);
6479 out:
6480 btrfs_free_path(path);
6481 return ret;
6482 }
Linux kernel - Deliverables
This page took 0.166244 seconds and 5 git commands to generate.