projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Btrfs: change/remove typedef
[linux-2.6-block.git] / fs / btrfs / extent-tree.c
CommitLineData
6cbd5570
CM		 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 */
ec6b910f 18#include <linux/sched.h>
edbd8d4e 19#include <linux/pagemap.h>
ec44a35c 20#include <linux/writeback.h>
21af804c 21#include <linux/blkdev.h>
4b4e25f2 22#include "compat.h"
74493f7a 23#include "hash.h"
a5eb62e3 24#include "crc32c.h"
fec577fb
CM		 25#include "ctree.h"
26#include "disk-io.h"
27#include "print-tree.h"
e089f05c 28#include "transaction.h"
0b86a832 29#include "volumes.h"
925baedd 30#include "locking.h"
31153d81 31#include "ref-cache.h"
fec577fb 32
31840ae1
ZY		 33#define PENDING_EXTENT_INSERT 0
34#define PENDING_EXTENT_DELETE 1
35#define PENDING_BACKREF_UPDATE 2
36
37struct pending_extent_op {
38 int type;
39 u64 bytenr;
40 u64 num_bytes;
41 u64 parent;
42 u64 orig_parent;
43 u64 generation;
44 u64 orig_generation;
45 int level;
f3465ca4
JB		 46 struct list_head list;
47 int del;
31840ae1
ZY		 48};
49
d397712b
CM		 50static int finish_current_insert(struct btrfs_trans_handle *trans,
51 struct btrfs_root *extent_root, int all);
52static int del_pending_extents(struct btrfs_trans_handle *trans,
53 struct btrfs_root *extent_root, int all);
f3465ca4
JB		 54static int pin_down_bytes(struct btrfs_trans_handle *trans,
55 struct btrfs_root *root,
56 u64 bytenr, u64 num_bytes, int is_data);
57static int update_block_group(struct btrfs_trans_handle *trans,
58 struct btrfs_root *root,
59 u64 bytenr, u64 num_bytes, int alloc,
60 int mark_free);
d548ee51 61
0f9dd46c
JB		 62static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
63{
64 return (cache->flags & bits) == bits;
65}
66
67/*
68 * this adds the block group to the fs_info rb tree for the block group
69 * cache
70 */
b2950863 71static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
0f9dd46c
JB		 72 struct btrfs_block_group_cache *block_group)
73{
74 struct rb_node **p;
75 struct rb_node *parent = NULL;
76 struct btrfs_block_group_cache *cache;
77
78 spin_lock(&info->block_group_cache_lock);
79 p = &info->block_group_cache_tree.rb_node;
80
81 while (*p) {
82 parent = *p;
83 cache = rb_entry(parent, struct btrfs_block_group_cache,
84 cache_node);
85 if (block_group->key.objectid < cache->key.objectid) {
86 p = &(*p)->rb_left;
87 } else if (block_group->key.objectid > cache->key.objectid) {
88 p = &(*p)->rb_right;
89 } else {
90 spin_unlock(&info->block_group_cache_lock);
91 return -EEXIST;
92 }
93 }
94
95 rb_link_node(&block_group->cache_node, parent, p);
96 rb_insert_color(&block_group->cache_node,
97 &info->block_group_cache_tree);
98 spin_unlock(&info->block_group_cache_lock);
99
100 return 0;
101}
102
103/*
104 * This will return the block group at or after bytenr if contains is 0, else
105 * it will return the block group that contains the bytenr
106 */
107static struct btrfs_block_group_cache *
108block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
109 int contains)
110{
111 struct btrfs_block_group_cache *cache, *ret = NULL;
112 struct rb_node *n;
113 u64 end, start;
114
115 spin_lock(&info->block_group_cache_lock);
116 n = info->block_group_cache_tree.rb_node;
117
118 while (n) {
119 cache = rb_entry(n, struct btrfs_block_group_cache,
120 cache_node);
121 end = cache->key.objectid + cache->key.offset - 1;
122 start = cache->key.objectid;
123
124 if (bytenr < start) {
125 if (!contains && (!ret || start < ret->key.objectid))
126 ret = cache;
127 n = n->rb_left;
128 } else if (bytenr > start) {
129 if (contains && bytenr <= end) {
130 ret = cache;
131 break;
132 }
133 n = n->rb_right;
134 } else {
135 ret = cache;
136 break;
137 }
138 }
d2fb3437
YZ		 139 if (ret)
140 atomic_inc(&ret->count);
0f9dd46c
JB		 141 spin_unlock(&info->block_group_cache_lock);
142
143 return ret;
144}
145
146/*
147 * this is only called by cache_block_group, since we could have freed extents
148 * we need to check the pinned_extents for any extents that can't be used yet
149 * since their free space will be released as soon as the transaction commits.
150 */
151static int add_new_free_space(struct btrfs_block_group_cache *block_group,
152 struct btrfs_fs_info *info, u64 start, u64 end)
153{
154 u64 extent_start, extent_end, size;
155 int ret;
156
25179201 157 mutex_lock(&info->pinned_mutex);
0f9dd46c
JB		 158 while (start < end) {
159 ret = find_first_extent_bit(&info->pinned_extents, start,
160 &extent_start, &extent_end,
161 EXTENT_DIRTY);
162 if (ret)
163 break;
164
165 if (extent_start == start) {
166 start = extent_end + 1;
167 } else if (extent_start > start && extent_start < end) {
168 size = extent_start - start;
ea6a478e
JB		 169 ret = btrfs_add_free_space(block_group, start,
170 size);
0f9dd46c
JB		 171 BUG_ON(ret);
172 start = extent_end + 1;
173 } else {
174 break;
175 }
176 }
177
178 if (start < end) {
179 size = end - start;
ea6a478e 180 ret = btrfs_add_free_space(block_group, start, size);
0f9dd46c
JB		 181 BUG_ON(ret);
182 }
25179201 183 mutex_unlock(&info->pinned_mutex);
0f9dd46c
JB		 184
185 return 0;
186}
187
a512bbf8
YZ		 188static int remove_sb_from_cache(struct btrfs_root *root,
189 struct btrfs_block_group_cache *cache)
190{
191 u64 bytenr;
192 u64 *logical;
193 int stripe_len;
194 int i, nr, ret;
195
196 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
197 bytenr = btrfs_sb_offset(i);
198 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
199 cache->key.objectid, bytenr, 0,
200 &logical, &nr, &stripe_len);
201 BUG_ON(ret);
202 while (nr--) {
203 btrfs_remove_free_space(cache, logical[nr],
204 stripe_len);
205 }
206 kfree(logical);
207 }
208 return 0;
209}
210
e37c9e69
CM		 211static int cache_block_group(struct btrfs_root *root,
212 struct btrfs_block_group_cache *block_group)
213{
214 struct btrfs_path *path;
ef8bbdfe 215 int ret = 0;
e37c9e69 216 struct btrfs_key key;
5f39d397 217 struct extent_buffer *leaf;
e37c9e69 218 int slot;
e4404d6e 219 u64 last;
e37c9e69 220
00f5c795
CM		 221 if (!block_group)
222 return 0;
223
e37c9e69 224 root = root->fs_info->extent_root;
e37c9e69
CM		 225
226 if (block_group->cached)
227 return 0;
f510cfec 228
e37c9e69
CM		 229 path = btrfs_alloc_path();
230 if (!path)
231 return -ENOMEM;
7d7d6068 232
2cc58cf2 233 path->reada = 2;
5cd57b2c
CM		 234 /*
235 * we get into deadlocks with paths held by callers of this function.
236 * since the alloc_mutex is protecting things right now, just
237 * skip the locking here
238 */
239 path->skip_locking = 1;
e4404d6e
YZ		 240 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
241 key.objectid = last;
e37c9e69
CM		 242 key.offset = 0;
243 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
244 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
245 if (ret < 0)
ef8bbdfe 246 goto err;
a512bbf8 247
d397712b 248 while (1) {
5f39d397 249 leaf = path->nodes[0];
e37c9e69 250 slot = path->slots[0];
5f39d397 251 if (slot >= btrfs_header_nritems(leaf)) {
e37c9e69 252 ret = btrfs_next_leaf(root, path);
54aa1f4d
CM		 253 if (ret < 0)
254 goto err;
0f9dd46c 255 if (ret == 0)
e37c9e69 256 continue;
0f9dd46c 257 else
e37c9e69 258 break;
e37c9e69 259 }
5f39d397 260 btrfs_item_key_to_cpu(leaf, &key, slot);
0f9dd46c 261 if (key.objectid < block_group->key.objectid)
7d7d6068 262 goto next;
0f9dd46c 263
e37c9e69 264 if (key.objectid >= block_group->key.objectid +
0f9dd46c 265 block_group->key.offset)
e37c9e69 266 break;
7d7d6068 267
e37c9e69 268 if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
0f9dd46c
JB		 269 add_new_free_space(block_group, root->fs_info, last,
270 key.objectid);
271
7d7d6068 272 last = key.objectid + key.offset;
e37c9e69 273 }
7d7d6068 274next:
e37c9e69
CM		 275 path->slots[0]++;
276 }
277
0f9dd46c
JB		 278 add_new_free_space(block_group, root->fs_info, last,
279 block_group->key.objectid +
280 block_group->key.offset);
281
a512bbf8 282 remove_sb_from_cache(root, block_group);
e37c9e69 283 block_group->cached = 1;
ef8bbdfe 284 ret = 0;
54aa1f4d 285err:
e37c9e69 286 btrfs_free_path(path);
ef8bbdfe 287 return ret;
e37c9e69
CM		 288}
289
0f9dd46c
JB		 290/*
291 * return the block group that starts at or after bytenr
292 */
d397712b
CM		 293static struct btrfs_block_group_cache *
294btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
0ef3e66b 295{
0f9dd46c 296 struct btrfs_block_group_cache *cache;
0ef3e66b 297
0f9dd46c 298 cache = block_group_cache_tree_search(info, bytenr, 0);
0ef3e66b 299
0f9dd46c 300 return cache;
0ef3e66b
CM		 301}
302
0f9dd46c
JB		 303/*
304 * return the block group that contains teh given bytenr
305 */
d397712b
CM		 306struct btrfs_block_group_cache *btrfs_lookup_block_group(
307 struct btrfs_fs_info *info,
308 u64 bytenr)
be744175 309{
0f9dd46c 310 struct btrfs_block_group_cache *cache;
be744175 311
0f9dd46c 312 cache = block_group_cache_tree_search(info, bytenr, 1);
96b5179d 313
0f9dd46c 314 return cache;
be744175 315}
0b86a832 316
d2fb3437
YZ		 317static inline void put_block_group(struct btrfs_block_group_cache *cache)
318{
319 if (atomic_dec_and_test(&cache->count))
320 kfree(cache);
321}
322
0f9dd46c
JB		 323static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
324 u64 flags)
6324fbf3 325{
0f9dd46c 326 struct list_head *head = &info->space_info;
0f9dd46c 327 struct btrfs_space_info *found;
c6e30871 328 list_for_each_entry(found, head, list) {
0f9dd46c
JB		 329 if (found->flags == flags)
330 return found;
331 }
332 return NULL;
6324fbf3
CM		 333}
334
80eb234a
JB		 335static u64 div_factor(u64 num, int factor)
336{
337 if (factor == 10)
338 return num;
339 num *= factor;
340 do_div(num, 10);
341 return num;
342}
343
d2fb3437
YZ		 344u64 btrfs_find_block_group(struct btrfs_root *root,
345 u64 search_start, u64 search_hint, int owner)
cd1bc465 346{
96b5179d 347 struct btrfs_block_group_cache *cache;
cd1bc465 348 u64 used;
d2fb3437
YZ		 349 u64 last = max(search_hint, search_start);
350 u64 group_start = 0;
31f3c99b 351 int full_search = 0;
d2fb3437 352 int factor = 9;
0ef3e66b 353 int wrapped = 0;
31f3c99b 354again:
e8569813
ZY		 355 while (1) {
356 cache = btrfs_lookup_first_block_group(root->fs_info, last);
0f9dd46c
JB		 357 if (!cache)
358 break;
96b5179d 359
c286ac48 360 spin_lock(&cache->lock);
96b5179d
CM		 361 last = cache->key.objectid + cache->key.offset;
362 used = btrfs_block_group_used(&cache->item);
363
d2fb3437
YZ		 364 if ((full_search || !cache->ro) &&
365 block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
e8569813 366 if (used + cache->pinned + cache->reserved <
d2fb3437
YZ		 367 div_factor(cache->key.offset, factor)) {
368 group_start = cache->key.objectid;
c286ac48 369 spin_unlock(&cache->lock);
d2fb3437 370 put_block_group(cache);
8790d502
CM		 371 goto found;
372 }
6324fbf3 373 }
c286ac48 374 spin_unlock(&cache->lock);
d2fb3437 375 put_block_group(cache);
de428b63 376 cond_resched();
cd1bc465 377 }
0ef3e66b
CM		 378 if (!wrapped) {
379 last = search_start;
380 wrapped = 1;
381 goto again;
382 }
383 if (!full_search && factor < 10) {
be744175 384 last = search_start;
31f3c99b 385 full_search = 1;
0ef3e66b 386 factor = 10;
31f3c99b
CM		 387 goto again;
388 }
be744175 389found:
d2fb3437 390 return group_start;
925baedd 391}
0f9dd46c 392
e02119d5 393/* simple helper to search for an existing extent at a given offset */
31840ae1 394int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
e02119d5
CM		 395{
396 int ret;
397 struct btrfs_key key;
31840ae1 398 struct btrfs_path *path;
e02119d5 399
31840ae1
ZY		 400 path = btrfs_alloc_path();
401 BUG_ON(!path);
e02119d5
CM		 402 key.objectid = start;
403 key.offset = len;
404 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
405 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
406 0, 0);
31840ae1 407 btrfs_free_path(path);
7bb86316
CM		 408 return ret;
409}
410
d8d5f3e1
CM		 411/*
412 * Back reference rules. Back refs have three main goals:
413 *
414 * 1) differentiate between all holders of references to an extent so that
415 * when a reference is dropped we can make sure it was a valid reference
416 * before freeing the extent.
417 *
418 * 2) Provide enough information to quickly find the holders of an extent
419 * if we notice a given block is corrupted or bad.
420 *
421 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
422 * maintenance. This is actually the same as #2, but with a slightly
423 * different use case.
424 *
425 * File extents can be referenced by:
426 *
427 * - multiple snapshots, subvolumes, or different generations in one subvol
31840ae1 428 * - different files inside a single subvolume
d8d5f3e1
CM		 429 * - different offsets inside a file (bookend extents in file.c)
430 *
431 * The extent ref structure has fields for:
432 *
433 * - Objectid of the subvolume root
434 * - Generation number of the tree holding the reference
435 * - objectid of the file holding the reference
31840ae1
ZY		 436 * - number of references holding by parent node (alway 1 for tree blocks)
437 *
438 * Btree leaf may hold multiple references to a file extent. In most cases,
439 * these references are from same file and the corresponding offsets inside
3bb1a1bc 440 * the file are close together.
d8d5f3e1
CM		 441 *
442 * When a file extent is allocated the fields are filled in:
3bb1a1bc 443 * (root_key.objectid, trans->transid, inode objectid, 1)
d8d5f3e1
CM		 444 *
445 * When a leaf is cow'd new references are added for every file extent found
31840ae1
ZY		 446 * in the leaf. It looks similar to the create case, but trans->transid will
447 * be different when the block is cow'd.
d8d5f3e1 448 *
3bb1a1bc 449 * (root_key.objectid, trans->transid, inode objectid,
31840ae1 450 * number of references in the leaf)
d8d5f3e1 451 *
3bb1a1bc
YZ		 452 * When a file extent is removed either during snapshot deletion or
453 * file truncation, we find the corresponding back reference and check
454 * the following fields:
d8d5f3e1 455 *
3bb1a1bc
YZ		 456 * (btrfs_header_owner(leaf), btrfs_header_generation(leaf),
457 * inode objectid)
d8d5f3e1
CM		 458 *
459 * Btree extents can be referenced by:
460 *
461 * - Different subvolumes
462 * - Different generations of the same subvolume
463 *
d8d5f3e1
CM		 464 * When a tree block is created, back references are inserted:
465 *
3bb1a1bc 466 * (root->root_key.objectid, trans->transid, level, 1)
d8d5f3e1 467 *
31840ae1
ZY		 468 * When a tree block is cow'd, new back references are added for all the
469 * blocks it points to. If the tree block isn't in reference counted root,
470 * the old back references are removed. These new back references are of
471 * the form (trans->transid will have increased since creation):
d8d5f3e1 472 *
3bb1a1bc 473 * (root->root_key.objectid, trans->transid, level, 1)
d8d5f3e1 474 *
31840ae1 475 * When a backref is in deleting, the following fields are checked:
d8d5f3e1
CM		 476 *
477 * if backref was for a tree root:
3bb1a1bc 478 * (btrfs_header_owner(itself), btrfs_header_generation(itself), level)
d8d5f3e1 479 * else
3bb1a1bc 480 * (btrfs_header_owner(parent), btrfs_header_generation(parent), level)
d8d5f3e1 481 *
31840ae1 482 * Back Reference Key composing:
d8d5f3e1 483 *
31840ae1
ZY		 484 * The key objectid corresponds to the first byte in the extent, the key
485 * type is set to BTRFS_EXTENT_REF_KEY, and the key offset is the first
486 * byte of parent extent. If a extent is tree root, the key offset is set
487 * to the key objectid.
d8d5f3e1 488 */
31840ae1 489
d397712b 490static noinline int lookup_extent_backref(struct btrfs_trans_handle *trans,
31840ae1 491 struct btrfs_root *root,
3bb1a1bc
YZ		 492 struct btrfs_path *path,
493 u64 bytenr, u64 parent,
494 u64 ref_root, u64 ref_generation,
495 u64 owner_objectid, int del)
7bb86316 496{
7bb86316 497 struct btrfs_key key;
31840ae1
ZY		 498 struct btrfs_extent_ref *ref;
499 struct extent_buffer *leaf;
3bb1a1bc 500 u64 ref_objectid;
74493f7a
CM		 501 int ret;
502
31840ae1
ZY		 503 key.objectid = bytenr;
504 key.type = BTRFS_EXTENT_REF_KEY;
505 key.offset = parent;
506
507 ret = btrfs_search_slot(trans, root, &key, path, del ? -1 : 0, 1);
508 if (ret < 0)
509 goto out;
510 if (ret > 0) {
511 ret = -ENOENT;
512 goto out;
513 }
514
515 leaf = path->nodes[0];
516 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
3bb1a1bc 517 ref_objectid = btrfs_ref_objectid(leaf, ref);
31840ae1 518 if (btrfs_ref_root(leaf, ref) != ref_root ||
3bb1a1bc
YZ		 519 btrfs_ref_generation(leaf, ref) != ref_generation ||
520 (ref_objectid != owner_objectid &&
521 ref_objectid != BTRFS_MULTIPLE_OBJECTIDS)) {
31840ae1
ZY		 522 ret = -EIO;
523 WARN_ON(1);
524 goto out;
525 }
526 ret = 0;
527out:
528 return ret;
529}
530
f3465ca4
JB		 531/*
532 * updates all the backrefs that are pending on update_list for the
533 * extent_root
534 */
d397712b 535static noinline int update_backrefs(struct btrfs_trans_handle *trans,
f3465ca4
JB		 536 struct btrfs_root *extent_root,
537 struct btrfs_path *path,
538 struct list_head *update_list)
539{
540 struct btrfs_key key;
541 struct btrfs_extent_ref *ref;
542 struct btrfs_fs_info *info = extent_root->fs_info;
543 struct pending_extent_op *op;
544 struct extent_buffer *leaf;
545 int ret = 0;
546 struct list_head *cur = update_list->next;
547 u64 ref_objectid;
548 u64 ref_root = extent_root->root_key.objectid;
549
550 op = list_entry(cur, struct pending_extent_op, list);
551
552search:
553 key.objectid = op->bytenr;
554 key.type = BTRFS_EXTENT_REF_KEY;
555 key.offset = op->orig_parent;
556
557 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 1);
558 BUG_ON(ret);
559
560 leaf = path->nodes[0];
561
562loop:
563 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
564
565 ref_objectid = btrfs_ref_objectid(leaf, ref);
566
567 if (btrfs_ref_root(leaf, ref) != ref_root ||
568 btrfs_ref_generation(leaf, ref) != op->orig_generation ||
569 (ref_objectid != op->level &&
570 ref_objectid != BTRFS_MULTIPLE_OBJECTIDS)) {
d397712b
CM		 571 printk(KERN_ERR "btrfs couldn't find %llu, parent %llu, "
572 "root %llu, owner %u\n",
573 (unsigned long long)op->bytenr,
574 (unsigned long long)op->orig_parent,
575 (unsigned long long)ref_root, op->level);
f3465ca4
JB		 576 btrfs_print_leaf(extent_root, leaf);
577 BUG();
578 }
579
580 key.objectid = op->bytenr;
581 key.offset = op->parent;
582 key.type = BTRFS_EXTENT_REF_KEY;
583 ret = btrfs_set_item_key_safe(trans, extent_root, path, &key);
584 BUG_ON(ret);
585 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
586 btrfs_set_ref_generation(leaf, ref, op->generation);
587
588 cur = cur->next;
589
590 list_del_init(&op->list);
591 unlock_extent(&info->extent_ins, op->bytenr,
592 op->bytenr + op->num_bytes - 1, GFP_NOFS);
593 kfree(op);
594
595 if (cur == update_list) {
596 btrfs_mark_buffer_dirty(path->nodes[0]);
597 btrfs_release_path(extent_root, path);
598 goto out;
599 }
600
601 op = list_entry(cur, struct pending_extent_op, list);
602
603 path->slots[0]++;
604 while (path->slots[0] < btrfs_header_nritems(leaf)) {
605 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
606 if (key.objectid == op->bytenr &&
607 key.type == BTRFS_EXTENT_REF_KEY)
608 goto loop;
609 path->slots[0]++;
610 }
611
612 btrfs_mark_buffer_dirty(path->nodes[0]);
613 btrfs_release_path(extent_root, path);
614 goto search;
615
616out:
617 return 0;
618}
619
d397712b 620static noinline int insert_extents(struct btrfs_trans_handle *trans,
f3465ca4
JB		 621 struct btrfs_root *extent_root,
622 struct btrfs_path *path,
623 struct list_head *insert_list, int nr)
624{
625 struct btrfs_key *keys;
626 u32 *data_size;
627 struct pending_extent_op *op;
628 struct extent_buffer *leaf;
629 struct list_head *cur = insert_list->next;
630 struct btrfs_fs_info *info = extent_root->fs_info;
631 u64 ref_root = extent_root->root_key.objectid;
632 int i = 0, last = 0, ret;
633 int total = nr * 2;
634
635 if (!nr)
636 return 0;
637
638 keys = kzalloc(total * sizeof(struct btrfs_key), GFP_NOFS);
639 if (!keys)
640 return -ENOMEM;
641
642 data_size = kzalloc(total * sizeof(u32), GFP_NOFS);
643 if (!data_size) {
644 kfree(keys);
645 return -ENOMEM;
646 }
647
648 list_for_each_entry(op, insert_list, list) {
649 keys[i].objectid = op->bytenr;
650 keys[i].offset = op->num_bytes;
651 keys[i].type = BTRFS_EXTENT_ITEM_KEY;
652 data_size[i] = sizeof(struct btrfs_extent_item);
653 i++;
654
655 keys[i].objectid = op->bytenr;
656 keys[i].offset = op->parent;
657 keys[i].type = BTRFS_EXTENT_REF_KEY;
658 data_size[i] = sizeof(struct btrfs_extent_ref);
659 i++;
660 }
661
662 op = list_entry(cur, struct pending_extent_op, list);
663 i = 0;
664 while (i < total) {
665 int c;
666 ret = btrfs_insert_some_items(trans, extent_root, path,
667 keys+i, data_size+i, total-i);
668 BUG_ON(ret < 0);
669
670 if (last && ret > 1)
671 BUG();
672
673 leaf = path->nodes[0];
674 for (c = 0; c < ret; c++) {
675 int ref_first = keys[i].type == BTRFS_EXTENT_REF_KEY;
676
677 /*
678 * if the first item we inserted was a backref, then
679 * the EXTENT_ITEM will be the odd c's, else it will
680 * be the even c's
681 */
682 if ((ref_first && (c % 2)) ||
683 (!ref_first && !(c % 2))) {
684 struct btrfs_extent_item *itm;
685
686 itm = btrfs_item_ptr(leaf, path->slots[0] + c,
687 struct btrfs_extent_item);
688 btrfs_set_extent_refs(path->nodes[0], itm, 1);
689 op->del++;
690 } else {
691 struct btrfs_extent_ref *ref;
692
693 ref = btrfs_item_ptr(leaf, path->slots[0] + c,
694 struct btrfs_extent_ref);
695 btrfs_set_ref_root(leaf, ref, ref_root);
696 btrfs_set_ref_generation(leaf, ref,
697 op->generation);
698 btrfs_set_ref_objectid(leaf, ref, op->level);
699 btrfs_set_ref_num_refs(leaf, ref, 1);
700 op->del++;
701 }
702
703 /*
704 * using del to see when its ok to free up the
705 * pending_extent_op. In the case where we insert the
706 * last item on the list in order to help do batching
707 * we need to not free the extent op until we actually
708 * insert the extent_item
709 */
710 if (op->del == 2) {
711 unlock_extent(&info->extent_ins, op->bytenr,
712 op->bytenr + op->num_bytes - 1,
713 GFP_NOFS);
714 cur = cur->next;
715 list_del_init(&op->list);
716 kfree(op);
717 if (cur != insert_list)
718 op = list_entry(cur,
719 struct pending_extent_op,
720 list);
721 }
722 }
723 btrfs_mark_buffer_dirty(leaf);
724 btrfs_release_path(extent_root, path);
725
726 /*
727 * Ok backref's and items usually go right next to eachother,
728 * but if we could only insert 1 item that means that we
729 * inserted on the end of a leaf, and we have no idea what may
730 * be on the next leaf so we just play it safe. In order to
731 * try and help this case we insert the last thing on our
732 * insert list so hopefully it will end up being the last
733 * thing on the leaf and everything else will be before it,
734 * which will let us insert a whole bunch of items at the same
735 * time.
736 */
737 if (ret == 1 && !last && (i + ret < total)) {
738 /*
739 * last: where we will pick up the next time around
740 * i: our current key to insert, will be total - 1
741 * cur: the current op we are screwing with
742 * op: duh
743 */
744 last = i + ret;
745 i = total - 1;
746 cur = insert_list->prev;
747 op = list_entry(cur, struct pending_extent_op, list);
748 } else if (last) {
749 /*
750 * ok we successfully inserted the last item on the
751 * list, lets reset everything
752 *
753 * i: our current key to insert, so where we left off
754 * last time
755 * last: done with this
756 * cur: the op we are messing with
757 * op: duh
758 * total: since we inserted the last key, we need to
759 * decrement total so we dont overflow
760 */
761 i = last;
762 last = 0;
f3465ca4 763 total--;
b4eec2ca
LH		 764 if (i < total) {
765 cur = insert_list->next;
766 op = list_entry(cur, struct pending_extent_op,
767 list);
768 }
f3465ca4
JB		 769 } else {
770 i += ret;
771 }
772
773 cond_resched();
774 }
775 ret = 0;
776 kfree(keys);
777 kfree(data_size);
778 return ret;
779}
780
d397712b 781static noinline int insert_extent_backref(struct btrfs_trans_handle *trans,
31840ae1
ZY		 782 struct btrfs_root *root,
783 struct btrfs_path *path,
784 u64 bytenr, u64 parent,
785 u64 ref_root, u64 ref_generation,
3bb1a1bc 786 u64 owner_objectid)
31840ae1
ZY		 787{
788 struct btrfs_key key;
789 struct extent_buffer *leaf;
790 struct btrfs_extent_ref *ref;
791 u32 num_refs;
792 int ret;
74493f7a 793
74493f7a
CM		 794 key.objectid = bytenr;
795 key.type = BTRFS_EXTENT_REF_KEY;
31840ae1 796 key.offset = parent;
74493f7a 797
31840ae1
ZY		 798 ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*ref));
799 if (ret == 0) {
800 leaf = path->nodes[0];
801 ref = btrfs_item_ptr(leaf, path->slots[0],
802 struct btrfs_extent_ref);
803 btrfs_set_ref_root(leaf, ref, ref_root);
804 btrfs_set_ref_generation(leaf, ref, ref_generation);
805 btrfs_set_ref_objectid(leaf, ref, owner_objectid);
31840ae1
ZY		 806 btrfs_set_ref_num_refs(leaf, ref, 1);
807 } else if (ret == -EEXIST) {
808 u64 existing_owner;
809 BUG_ON(owner_objectid < BTRFS_FIRST_FREE_OBJECTID);
810 leaf = path->nodes[0];
811 ref = btrfs_item_ptr(leaf, path->slots[0],
812 struct btrfs_extent_ref);
813 if (btrfs_ref_root(leaf, ref) != ref_root ||
814 btrfs_ref_generation(leaf, ref) != ref_generation) {
815 ret = -EIO;
816 WARN_ON(1);
7bb86316 817 goto out;
31840ae1
ZY		 818 }
819
820 num_refs = btrfs_ref_num_refs(leaf, ref);
821 BUG_ON(num_refs == 0);
822 btrfs_set_ref_num_refs(leaf, ref, num_refs + 1);
823
824 existing_owner = btrfs_ref_objectid(leaf, ref);
3bb1a1bc
YZ		 825 if (existing_owner != owner_objectid &&
826 existing_owner != BTRFS_MULTIPLE_OBJECTIDS) {
31840ae1
ZY		 827 btrfs_set_ref_objectid(leaf, ref,
828 BTRFS_MULTIPLE_OBJECTIDS);
31840ae1
ZY		 829 }
830 ret = 0;
831 } else {
7bb86316 832 goto out;
31840ae1 833 }
7bb86316
CM		 834 btrfs_mark_buffer_dirty(path->nodes[0]);
835out:
836 btrfs_release_path(root, path);
837 return ret;
74493f7a
CM		 838}
839
d397712b 840static noinline int remove_extent_backref(struct btrfs_trans_handle *trans,
31840ae1
ZY		 841 struct btrfs_root *root,
842 struct btrfs_path *path)
843{
844 struct extent_buffer *leaf;
845 struct btrfs_extent_ref *ref;
846 u32 num_refs;
847 int ret = 0;
848
849 leaf = path->nodes[0];
850 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
851 num_refs = btrfs_ref_num_refs(leaf, ref);
852 BUG_ON(num_refs == 0);
853 num_refs -= 1;
854 if (num_refs == 0) {
855 ret = btrfs_del_item(trans, root, path);
856 } else {
857 btrfs_set_ref_num_refs(leaf, ref, num_refs);
858 btrfs_mark_buffer_dirty(leaf);
859 }
860 btrfs_release_path(root, path);
861 return ret;
862}
863
4b4e25f2 864#ifdef BIO_RW_DISCARD
15916de8
CM		 865static void btrfs_issue_discard(struct block_device *bdev,
866 u64 start, u64 len)
867{
15916de8 868 blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL);
15916de8 869}
4b4e25f2 870#endif
15916de8 871
1f3c79a2
LH		 872static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
873 u64 num_bytes)
874{
875#ifdef BIO_RW_DISCARD
876 int ret;
877 u64 map_length = num_bytes;
878 struct btrfs_multi_bio *multi = NULL;
879
880 /* Tell the block device(s) that the sectors can be discarded */
881 ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
882 bytenr, &map_length, &multi, 0);
883 if (!ret) {
884 struct btrfs_bio_stripe *stripe = multi->stripes;
885 int i;
886
887 if (map_length > num_bytes)
888 map_length = num_bytes;
889
890 for (i = 0; i < multi->num_stripes; i++, stripe++) {
891 btrfs_issue_discard(stripe->dev->bdev,
892 stripe->physical,
893 map_length);
894 }
895 kfree(multi);
896 }
897
898 return ret;
899#else
900 return 0;
901#endif
902}
903
d397712b 904static noinline int free_extents(struct btrfs_trans_handle *trans,
f3465ca4
JB		 905 struct btrfs_root *extent_root,
906 struct list_head *del_list)
907{
908 struct btrfs_fs_info *info = extent_root->fs_info;
909 struct btrfs_path *path;
910 struct btrfs_key key, found_key;
911 struct extent_buffer *leaf;
912 struct list_head *cur;
913 struct pending_extent_op *op;
914 struct btrfs_extent_item *ei;
915 int ret, num_to_del, extent_slot = 0, found_extent = 0;
916 u32 refs;
917 u64 bytes_freed = 0;
918
919 path = btrfs_alloc_path();
920 if (!path)
921 return -ENOMEM;
922 path->reada = 1;
923
924search:
925 /* search for the backref for the current ref we want to delete */
926 cur = del_list->next;
927 op = list_entry(cur, struct pending_extent_op, list);
928 ret = lookup_extent_backref(trans, extent_root, path, op->bytenr,
929 op->orig_parent,
930 extent_root->root_key.objectid,
931 op->orig_generation, op->level, 1);
932 if (ret) {
d397712b
CM		 933 printk(KERN_ERR "btrfs unable to find backref byte nr %llu "
934 "root %llu gen %llu owner %u\n",
935 (unsigned long long)op->bytenr,
936 (unsigned long long)extent_root->root_key.objectid,
937 (unsigned long long)op->orig_generation, op->level);
f3465ca4
JB		 938 btrfs_print_leaf(extent_root, path->nodes[0]);
939 WARN_ON(1);
940 goto out;
941 }
942
943 extent_slot = path->slots[0];
944 num_to_del = 1;
945 found_extent = 0;
946
947 /*
948 * if we aren't the first item on the leaf we can move back one and see
949 * if our ref is right next to our extent item
950 */
951 if (likely(extent_slot)) {
952 extent_slot--;
953 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
954 extent_slot);
955 if (found_key.objectid == op->bytenr &&
956 found_key.type == BTRFS_EXTENT_ITEM_KEY &&
957 found_key.offset == op->num_bytes) {
958 num_to_del++;
959 found_extent = 1;
960 }
961 }
962
963 /*
964 * if we didn't find the extent we need to delete the backref and then
965 * search for the extent item key so we can update its ref count
966 */
967 if (!found_extent) {
968 key.objectid = op->bytenr;
969 key.type = BTRFS_EXTENT_ITEM_KEY;
970 key.offset = op->num_bytes;
971
972 ret = remove_extent_backref(trans, extent_root, path);
973 BUG_ON(ret);
974 btrfs_release_path(extent_root, path);
975 ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1);
976 BUG_ON(ret);
977 extent_slot = path->slots[0];
978 }
979
980 /* this is where we update the ref count for the extent */
981 leaf = path->nodes[0];
982 ei = btrfs_item_ptr(leaf, extent_slot, struct btrfs_extent_item);
983 refs = btrfs_extent_refs(leaf, ei);
984 BUG_ON(refs == 0);
985 refs--;
986 btrfs_set_extent_refs(leaf, ei, refs);
987
988 btrfs_mark_buffer_dirty(leaf);
989
990 /*
991 * This extent needs deleting. The reason cur_slot is extent_slot +
992 * num_to_del is because extent_slot points to the slot where the extent
993 * is, and if the backref was not right next to the extent we will be
994 * deleting at least 1 item, and will want to start searching at the
995 * slot directly next to extent_slot. However if we did find the
996 * backref next to the extent item them we will be deleting at least 2
997 * items and will want to start searching directly after the ref slot
998 */
999 if (!refs) {
1000 struct list_head *pos, *n, *end;
1001 int cur_slot = extent_slot+num_to_del;
1002 u64 super_used;
1003 u64 root_used;
1004
1005 path->slots[0] = extent_slot;
1006 bytes_freed = op->num_bytes;
1007
e3e469f8
JB		 1008 mutex_lock(&info->pinned_mutex);
1009 ret = pin_down_bytes(trans, extent_root, op->bytenr,
1010 op->num_bytes, op->level >=
1011 BTRFS_FIRST_FREE_OBJECTID);
1012 mutex_unlock(&info->pinned_mutex);
1013 BUG_ON(ret < 0);
1014 op->del = ret;
1015
f3465ca4
JB		 1016 /*
1017 * we need to see if we can delete multiple things at once, so
1018 * start looping through the list of extents we are wanting to
1019 * delete and see if their extent/backref's are right next to
1020 * eachother and the extents only have 1 ref
1021 */
1022 for (pos = cur->next; pos != del_list; pos = pos->next) {
1023 struct pending_extent_op *tmp;
1024
1025 tmp = list_entry(pos, struct pending_extent_op, list);
1026
1027 /* we only want to delete extent+ref at this stage */
1028 if (cur_slot >= btrfs_header_nritems(leaf) - 1)
1029 break;
1030
1031 btrfs_item_key_to_cpu(leaf, &found_key, cur_slot);
1032 if (found_key.objectid != tmp->bytenr ||
1033 found_key.type != BTRFS_EXTENT_ITEM_KEY ||
1034 found_key.offset != tmp->num_bytes)
1035 break;
1036
1037 /* check to make sure this extent only has one ref */
1038 ei = btrfs_item_ptr(leaf, cur_slot,
1039 struct btrfs_extent_item);
1040 if (btrfs_extent_refs(leaf, ei) != 1)
1041 break;
1042
1043 btrfs_item_key_to_cpu(leaf, &found_key, cur_slot+1);
1044 if (found_key.objectid != tmp->bytenr ||
1045 found_key.type != BTRFS_EXTENT_REF_KEY ||
1046 found_key.offset != tmp->orig_parent)
1047 break;
1048
1049 /*
1050 * the ref is right next to the extent, we can set the
1051 * ref count to 0 since we will delete them both now
1052 */
1053 btrfs_set_extent_refs(leaf, ei, 0);
1054
1055 /* pin down the bytes for this extent */
1056 mutex_lock(&info->pinned_mutex);
1057 ret = pin_down_bytes(trans, extent_root, tmp->bytenr,
1058 tmp->num_bytes, tmp->level >=
1059 BTRFS_FIRST_FREE_OBJECTID);
1060 mutex_unlock(&info->pinned_mutex);
1061 BUG_ON(ret < 0);
1062
1063 /*
1064 * use the del field to tell if we need to go ahead and
1065 * free up the extent when we delete the item or not.
1066 */
1067 tmp->del = ret;
1068 bytes_freed += tmp->num_bytes;
1069
1070 num_to_del += 2;
1071 cur_slot += 2;
1072 }
1073 end = pos;
1074
1075 /* update the free space counters */
75eff68e 1076 spin_lock(&info->delalloc_lock);
f3465ca4
JB		 1077 super_used = btrfs_super_bytes_used(&info->super_copy);
1078 btrfs_set_super_bytes_used(&info->super_copy,
1079 super_used - bytes_freed);
f3465ca4
JB		 1080
1081 root_used = btrfs_root_used(&extent_root->root_item);
1082 btrfs_set_root_used(&extent_root->root_item,
1083 root_used - bytes_freed);
34bf63c4 1084 spin_unlock(&info->delalloc_lock);
f3465ca4
JB		 1085
1086 /* delete the items */
1087 ret = btrfs_del_items(trans, extent_root, path,
1088 path->slots[0], num_to_del);
1089 BUG_ON(ret);
1090
1091 /*
1092 * loop through the extents we deleted and do the cleanup work
1093 * on them
1094 */
1095 for (pos = cur, n = pos->next; pos != end;
1096 pos = n, n = pos->next) {
1097 struct pending_extent_op *tmp;
f3465ca4
JB		 1098 tmp = list_entry(pos, struct pending_extent_op, list);
1099
1100 /*
1101 * remember tmp->del tells us wether or not we pinned
1102 * down the extent
1103 */
1104 ret = update_block_group(trans, extent_root,
1105 tmp->bytenr, tmp->num_bytes, 0,
1106 tmp->del);
1107 BUG_ON(ret);
1108
f3465ca4
JB		 1109 list_del_init(&tmp->list);
1110 unlock_extent(&info->extent_ins, tmp->bytenr,
1111 tmp->bytenr + tmp->num_bytes - 1,
1112 GFP_NOFS);
1113 kfree(tmp);
1114 }
1115 } else if (refs && found_extent) {
1116 /*
1117 * the ref and extent were right next to eachother, but the
1118 * extent still has a ref, so just free the backref and keep
1119 * going
1120 */
1121 ret = remove_extent_backref(trans, extent_root, path);
1122 BUG_ON(ret);
1123
1124 list_del_init(&op->list);
1125 unlock_extent(&info->extent_ins, op->bytenr,
1126 op->bytenr + op->num_bytes - 1, GFP_NOFS);
1127 kfree(op);
1128 } else {
1129 /*
1130 * the extent has multiple refs and the backref we were looking
1131 * for was not right next to it, so just unlock and go next,
1132 * we're good to go
1133 */
1134 list_del_init(&op->list);
1135 unlock_extent(&info->extent_ins, op->bytenr,
1136 op->bytenr + op->num_bytes - 1, GFP_NOFS);
1137 kfree(op);
1138 }
1139
1140 btrfs_release_path(extent_root, path);
1141 if (!list_empty(del_list))
1142 goto search;
1143
1144out:
1145 btrfs_free_path(path);
1146 return ret;
1147}
1148
31840ae1
ZY		 1149static int __btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
1150 struct btrfs_root *root, u64 bytenr,
1151 u64 orig_parent, u64 parent,
1152 u64 orig_root, u64 ref_root,
1153 u64 orig_generation, u64 ref_generation,
3bb1a1bc 1154 u64 owner_objectid)
31840ae1
ZY		 1155{
1156 int ret;
1157 struct btrfs_root *extent_root = root->fs_info->extent_root;
1158 struct btrfs_path *path;
1159
1160 if (root == root->fs_info->extent_root) {
1161 struct pending_extent_op *extent_op;
1162 u64 num_bytes;
1163
1164 BUG_ON(owner_objectid >= BTRFS_MAX_LEVEL);
1165 num_bytes = btrfs_level_size(root, (int)owner_objectid);
25179201 1166 mutex_lock(&root->fs_info->extent_ins_mutex);
31840ae1 1167 if (test_range_bit(&root->fs_info->extent_ins, bytenr,
25179201 1168 bytenr + num_bytes - 1, EXTENT_WRITEBACK, 0)) {
31840ae1
ZY		 1169 u64 priv;
1170 ret = get_state_private(&root->fs_info->extent_ins,
1171 bytenr, &priv);
1172 BUG_ON(ret);
1173 extent_op = (struct pending_extent_op *)
1174 (unsigned long)priv;
1175 BUG_ON(extent_op->parent != orig_parent);
1176 BUG_ON(extent_op->generation != orig_generation);
25179201 1177
31840ae1
ZY		 1178 extent_op->parent = parent;
1179 extent_op->generation = ref_generation;
1180 } else {
1181 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
1182 BUG_ON(!extent_op);
1183
1184 extent_op->type = PENDING_BACKREF_UPDATE;
1185 extent_op->bytenr = bytenr;
1186 extent_op->num_bytes = num_bytes;
1187 extent_op->parent = parent;
1188 extent_op->orig_parent = orig_parent;
1189 extent_op->generation = ref_generation;
1190 extent_op->orig_generation = orig_generation;
1191 extent_op->level = (int)owner_objectid;
f3465ca4
JB		 1192 INIT_LIST_HEAD(&extent_op->list);
1193 extent_op->del = 0;
31840ae1
ZY		 1194
1195 set_extent_bits(&root->fs_info->extent_ins,
1196 bytenr, bytenr + num_bytes - 1,
25179201 1197 EXTENT_WRITEBACK, GFP_NOFS);
31840ae1
ZY		 1198 set_state_private(&root->fs_info->extent_ins,
1199 bytenr, (unsigned long)extent_op);
1200 }
25179201 1201 mutex_unlock(&root->fs_info->extent_ins_mutex);
31840ae1
ZY		 1202 return 0;
1203 }
1204
1205 path = btrfs_alloc_path();
1206 if (!path)
1207 return -ENOMEM;
1208 ret = lookup_extent_backref(trans, extent_root, path,
1209 bytenr, orig_parent, orig_root,
3bb1a1bc 1210 orig_generation, owner_objectid, 1);
31840ae1
ZY		 1211 if (ret)
1212 goto out;
1213 ret = remove_extent_backref(trans, extent_root, path);
1214 if (ret)
1215 goto out;
1216 ret = insert_extent_backref(trans, extent_root, path, bytenr,
1217 parent, ref_root, ref_generation,
3bb1a1bc 1218 owner_objectid);
31840ae1 1219 BUG_ON(ret);
87ef2bb4
CM		 1220 finish_current_insert(trans, extent_root, 0);
1221 del_pending_extents(trans, extent_root, 0);
31840ae1
ZY		 1222out:
1223 btrfs_free_path(path);
1224 return ret;
1225}
1226
1227int btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
1228 struct btrfs_root *root, u64 bytenr,
1229 u64 orig_parent, u64 parent,
1230 u64 ref_root, u64 ref_generation,
3bb1a1bc 1231 u64 owner_objectid)
31840ae1
ZY		 1232{
1233 int ret;
1234 if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
1235 owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
1236 return 0;
31840ae1
ZY		 1237 ret = __btrfs_update_extent_ref(trans, root, bytenr, orig_parent,
1238 parent, ref_root, ref_root,
1239 ref_generation, ref_generation,
3bb1a1bc 1240 owner_objectid);
31840ae1
ZY		 1241 return ret;
1242}
1243
925baedd 1244static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
31840ae1
ZY		 1245 struct btrfs_root *root, u64 bytenr,
1246 u64 orig_parent, u64 parent,
1247 u64 orig_root, u64 ref_root,
1248 u64 orig_generation, u64 ref_generation,
3bb1a1bc 1249 u64 owner_objectid)
02217ed2 1250{
5caf2a00 1251 struct btrfs_path *path;
02217ed2 1252 int ret;
e2fa7227 1253 struct btrfs_key key;
5f39d397 1254 struct extent_buffer *l;
234b63a0 1255 struct btrfs_extent_item *item;
cf27e1ee 1256 u32 refs;
037e6390 1257
5caf2a00 1258 path = btrfs_alloc_path();
54aa1f4d
CM		 1259 if (!path)
1260 return -ENOMEM;
26b8003f 1261
3c12ac72 1262 path->reada = 1;
db94535d 1263 key.objectid = bytenr;
31840ae1
ZY		 1264 key.type = BTRFS_EXTENT_ITEM_KEY;
1265 key.offset = (u64)-1;
1266
5caf2a00 1267 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
9f5fae2f 1268 0, 1);
54aa1f4d
CM		 1269 if (ret < 0)
1270 return ret;
31840ae1
ZY		 1271 BUG_ON(ret == 0 || path->slots[0] == 0);
1272
1273 path->slots[0]--;
5f39d397 1274 l = path->nodes[0];
31840ae1
ZY		 1275
1276 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
771ed689
CM		 1277 if (key.objectid != bytenr) {
1278 btrfs_print_leaf(root->fs_info->extent_root, path->nodes[0]);
d397712b
CM		 1279 printk(KERN_ERR "btrfs wanted %llu found %llu\n",
1280 (unsigned long long)bytenr,
1281 (unsigned long long)key.objectid);
771ed689
CM		 1282 BUG();
1283 }
31840ae1
ZY		 1284 BUG_ON(key.type != BTRFS_EXTENT_ITEM_KEY);
1285
5caf2a00 1286 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
5f39d397
CM		 1287 refs = btrfs_extent_refs(l, item);
1288 btrfs_set_extent_refs(l, item, refs + 1);
5caf2a00 1289 btrfs_mark_buffer_dirty(path->nodes[0]);
a28ec197 1290
5caf2a00 1291 btrfs_release_path(root->fs_info->extent_root, path);
7bb86316 1292
3c12ac72 1293 path->reada = 1;
31840ae1
ZY		 1294 ret = insert_extent_backref(trans, root->fs_info->extent_root,
1295 path, bytenr, parent,
1296 ref_root, ref_generation,
3bb1a1bc 1297 owner_objectid);
7bb86316 1298 BUG_ON(ret);
87ef2bb4
CM		 1299 finish_current_insert(trans, root->fs_info->extent_root, 0);
1300 del_pending_extents(trans, root->fs_info->extent_root, 0);
74493f7a
CM		 1301
1302 btrfs_free_path(path);
02217ed2
CM		 1303 return 0;
1304}
1305
925baedd 1306int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
31840ae1
ZY		 1307 struct btrfs_root *root,
1308 u64 bytenr, u64 num_bytes, u64 parent,
1309 u64 ref_root, u64 ref_generation,
3bb1a1bc 1310 u64 owner_objectid)
925baedd
CM		 1311{
1312 int ret;
31840ae1
ZY		 1313 if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
1314 owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
1315 return 0;
31840ae1
ZY		 1316 ret = __btrfs_inc_extent_ref(trans, root, bytenr, 0, parent,
1317 0, ref_root, 0, ref_generation,
3bb1a1bc 1318 owner_objectid);
925baedd
CM		 1319 return ret;
1320}
1321
e9d0b13b
CM		 1322int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
1323 struct btrfs_root *root)
1324{
87ef2bb4
CM		 1325 finish_current_insert(trans, root->fs_info->extent_root, 1);
1326 del_pending_extents(trans, root->fs_info->extent_root, 1);
e9d0b13b
CM		 1327 return 0;
1328}
1329
31840ae1
ZY		 1330int btrfs_lookup_extent_ref(struct btrfs_trans_handle *trans,
1331 struct btrfs_root *root, u64 bytenr,
1332 u64 num_bytes, u32 *refs)
a28ec197 1333{
5caf2a00 1334 struct btrfs_path *path;
a28ec197 1335 int ret;
e2fa7227 1336 struct btrfs_key key;
5f39d397 1337 struct extent_buffer *l;
234b63a0 1338 struct btrfs_extent_item *item;
5caf2a00 1339
db94535d 1340 WARN_ON(num_bytes < root->sectorsize);
5caf2a00 1341 path = btrfs_alloc_path();
3c12ac72 1342 path->reada = 1;
db94535d
CM		 1343 key.objectid = bytenr;
1344 key.offset = num_bytes;
62e2749e 1345 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
5caf2a00 1346 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
9f5fae2f 1347 0, 0);
54aa1f4d
CM		 1348 if (ret < 0)
1349 goto out;
5f39d397
CM		 1350 if (ret != 0) {
1351 btrfs_print_leaf(root, path->nodes[0]);
d397712b
CM		 1352 printk(KERN_INFO "btrfs failed to find block number %llu\n",
1353 (unsigned long long)bytenr);
a28ec197 1354 BUG();
5f39d397
CM		 1355 }
1356 l = path->nodes[0];
5caf2a00 1357 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
5f39d397 1358 *refs = btrfs_extent_refs(l, item);
54aa1f4d 1359out:
5caf2a00 1360 btrfs_free_path(path);
a28ec197
CM		 1361 return 0;
1362}
1363
80ff3856 1364int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
17d217fe 1365 struct btrfs_root *root, u64 objectid, u64 bytenr)
be20aa9d
CM		 1366{
1367 struct btrfs_root *extent_root = root->fs_info->extent_root;
1368 struct btrfs_path *path;
f321e491
YZ		 1369 struct extent_buffer *leaf;
1370 struct btrfs_extent_ref *ref_item;
1371 struct btrfs_key key;
1372 struct btrfs_key found_key;
80ff3856
YZ		 1373 u64 ref_root;
1374 u64 last_snapshot;
be20aa9d
CM		 1375 u32 nritems;
1376 int ret;
925baedd 1377
be20aa9d 1378 key.objectid = bytenr;
31840ae1 1379 key.offset = (u64)-1;
f321e491 1380 key.type = BTRFS_EXTENT_ITEM_KEY;
be20aa9d 1381
f321e491 1382 path = btrfs_alloc_path();
be20aa9d
CM		 1383 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
1384 if (ret < 0)
1385 goto out;
1386 BUG_ON(ret == 0);
80ff3856
YZ		 1387
1388 ret = -ENOENT;
1389 if (path->slots[0] == 0)
31840ae1 1390 goto out;
be20aa9d 1391
31840ae1 1392 path->slots[0]--;
f321e491
YZ		 1393 leaf = path->nodes[0];
1394 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
be20aa9d
CM		 1395
1396 if (found_key.objectid != bytenr ||
80ff3856 1397 found_key.type != BTRFS_EXTENT_ITEM_KEY)
be20aa9d 1398 goto out;
f321e491 1399
80ff3856 1400 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
be20aa9d 1401 while (1) {
f321e491
YZ		 1402 leaf = path->nodes[0];
1403 nritems = btrfs_header_nritems(leaf);
be20aa9d
CM		 1404 if (path->slots[0] >= nritems) {
1405 ret = btrfs_next_leaf(extent_root, path);
f321e491
YZ		 1406 if (ret < 0)
1407 goto out;
be20aa9d
CM		 1408 if (ret == 0)
1409 continue;
1410 break;
1411 }
f321e491 1412 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
be20aa9d
CM		 1413 if (found_key.objectid != bytenr)
1414 break;
bd09835d 1415
be20aa9d
CM		 1416 if (found_key.type != BTRFS_EXTENT_REF_KEY) {
1417 path->slots[0]++;
1418 continue;
1419 }
1420
f321e491 1421 ref_item = btrfs_item_ptr(leaf, path->slots[0],
be20aa9d 1422 struct btrfs_extent_ref);
80ff3856 1423 ref_root = btrfs_ref_root(leaf, ref_item);
17d217fe
YZ		 1424 if ((ref_root != root->root_key.objectid &&
1425 ref_root != BTRFS_TREE_LOG_OBJECTID) ||
1426 objectid != btrfs_ref_objectid(leaf, ref_item)) {
80ff3856 1427 ret = 1;
f321e491 1428 goto out;
80ff3856
YZ		 1429 }
1430 if (btrfs_ref_generation(leaf, ref_item) <= last_snapshot) {
f321e491
YZ		 1431 ret = 1;
1432 goto out;
1433 }
80ff3856
YZ		 1434
1435 path->slots[0]++;
f321e491
YZ		 1436 }
1437 ret = 0;
be20aa9d 1438out:
80ff3856 1439 btrfs_free_path(path);
f321e491 1440 return ret;
be20aa9d 1441}
c5739bba 1442
31840ae1
ZY		 1443int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1444 struct extent_buffer *buf, u32 nr_extents)
02217ed2 1445{
5f39d397 1446 struct btrfs_key key;
6407bf6d 1447 struct btrfs_file_extent_item *fi;
e4657689
ZY		 1448 u64 root_gen;
1449 u32 nritems;
02217ed2 1450 int i;
db94535d 1451 int level;
31840ae1 1452 int ret = 0;
e4657689 1453 int shared = 0;
a28ec197 1454
3768f368 1455 if (!root->ref_cows)
a28ec197 1456 return 0;
5f39d397 1457
e4657689
ZY		 1458 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
1459 shared = 0;
1460 root_gen = root->root_key.offset;
1461 } else {
1462 shared = 1;
1463 root_gen = trans->transid - 1;
1464 }
1465
db94535d 1466 level = btrfs_header_level(buf);
5f39d397 1467 nritems = btrfs_header_nritems(buf);
4a096752 1468
31840ae1 1469 if (level == 0) {
31153d81
YZ		 1470 struct btrfs_leaf_ref *ref;
1471 struct btrfs_extent_info *info;
1472
31840ae1 1473 ref = btrfs_alloc_leaf_ref(root, nr_extents);
31153d81 1474 if (!ref) {
31840ae1 1475 ret = -ENOMEM;
31153d81
YZ		 1476 goto out;
1477 }
1478
e4657689 1479 ref->root_gen = root_gen;
31153d81
YZ		 1480 ref->bytenr = buf->start;
1481 ref->owner = btrfs_header_owner(buf);
1482 ref->generation = btrfs_header_generation(buf);
31840ae1 1483 ref->nritems = nr_extents;
31153d81 1484 info = ref->extents;
bcc63abb 1485
31840ae1 1486 for (i = 0; nr_extents > 0 && i < nritems; i++) {
31153d81
YZ		 1487 u64 disk_bytenr;
1488 btrfs_item_key_to_cpu(buf, &key, i);
1489 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1490 continue;
1491 fi = btrfs_item_ptr(buf, i,
1492 struct btrfs_file_extent_item);
1493 if (btrfs_file_extent_type(buf, fi) ==
1494 BTRFS_FILE_EXTENT_INLINE)
1495 continue;
1496 disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1497 if (disk_bytenr == 0)
1498 continue;
1499
1500 info->bytenr = disk_bytenr;
1501 info->num_bytes =
1502 btrfs_file_extent_disk_num_bytes(buf, fi);
1503 info->objectid = key.objectid;
1504 info->offset = key.offset;
1505 info++;
1506 }
1507
e4657689 1508 ret = btrfs_add_leaf_ref(root, ref, shared);
5b84e8d6
YZ		 1509 if (ret == -EEXIST && shared) {
1510 struct btrfs_leaf_ref *old;
1511 old = btrfs_lookup_leaf_ref(root, ref->bytenr);
1512 BUG_ON(!old);
1513 btrfs_remove_leaf_ref(root, old);
1514 btrfs_free_leaf_ref(root, old);
1515 ret = btrfs_add_leaf_ref(root, ref, shared);
1516 }
31153d81 1517 WARN_ON(ret);
bcc63abb 1518 btrfs_free_leaf_ref(root, ref);
31153d81
YZ		 1519 }
1520out:
31840ae1
ZY		 1521 return ret;
1522}
1523
1524int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1525 struct extent_buffer *orig_buf, struct extent_buffer *buf,
1526 u32 *nr_extents)
1527{
1528 u64 bytenr;
1529 u64 ref_root;
1530 u64 orig_root;
1531 u64 ref_generation;
1532 u64 orig_generation;
1533 u32 nritems;
1534 u32 nr_file_extents = 0;
1535 struct btrfs_key key;
1536 struct btrfs_file_extent_item *fi;
1537 int i;
1538 int level;
1539 int ret = 0;
1540 int faili = 0;
1541 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
3bb1a1bc 1542 u64, u64, u64, u64, u64, u64, u64, u64);
31840ae1
ZY		 1543
1544 ref_root = btrfs_header_owner(buf);
1545 ref_generation = btrfs_header_generation(buf);
1546 orig_root = btrfs_header_owner(orig_buf);
1547 orig_generation = btrfs_header_generation(orig_buf);
1548
1549 nritems = btrfs_header_nritems(buf);
1550 level = btrfs_header_level(buf);
1551
1552 if (root->ref_cows) {
1553 process_func = __btrfs_inc_extent_ref;
1554 } else {
1555 if (level == 0 &&
1556 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
1557 goto out;
1558 if (level != 0 &&
1559 root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
1560 goto out;
1561 process_func = __btrfs_update_extent_ref;
1562 }
1563
1564 for (i = 0; i < nritems; i++) {
1565 cond_resched();
db94535d 1566 if (level == 0) {
5f39d397
CM		 1567 btrfs_item_key_to_cpu(buf, &key, i);
1568 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
54aa1f4d 1569 continue;
5f39d397 1570 fi = btrfs_item_ptr(buf, i,
54aa1f4d 1571 struct btrfs_file_extent_item);
5f39d397 1572 if (btrfs_file_extent_type(buf, fi) ==
54aa1f4d
CM		 1573 BTRFS_FILE_EXTENT_INLINE)
1574 continue;
31840ae1
ZY		 1575 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1576 if (bytenr == 0)
54aa1f4d 1577 continue;
31840ae1
ZY		 1578
1579 nr_file_extents++;
1580
31840ae1
ZY		 1581 ret = process_func(trans, root, bytenr,
1582 orig_buf->start, buf->start,
1583 orig_root, ref_root,
1584 orig_generation, ref_generation,
3bb1a1bc 1585 key.objectid);
31840ae1
ZY		 1586
1587 if (ret) {
1588 faili = i;
1589 WARN_ON(1);
1590 goto fail;
1591 }
54aa1f4d 1592 } else {
db94535d 1593 bytenr = btrfs_node_blockptr(buf, i);
31840ae1
ZY		 1594 ret = process_func(trans, root, bytenr,
1595 orig_buf->start, buf->start,
1596 orig_root, ref_root,
1597 orig_generation, ref_generation,
3bb1a1bc 1598 level - 1);
31840ae1
ZY		 1599 if (ret) {
1600 faili = i;
1601 WARN_ON(1);
1602 goto fail;
1603 }
54aa1f4d
CM		 1604 }
1605 }
31840ae1
ZY		 1606out:
1607 if (nr_extents) {
1608 if (level == 0)
1609 *nr_extents = nr_file_extents;
1610 else
1611 *nr_extents = nritems;
1612 }
1613 return 0;
1614fail:
1615 WARN_ON(1);
54aa1f4d 1616 return ret;
02217ed2
CM		 1617}
1618
31840ae1
ZY		 1619int btrfs_update_ref(struct btrfs_trans_handle *trans,
1620 struct btrfs_root *root, struct extent_buffer *orig_buf,
1621 struct extent_buffer *buf, int start_slot, int nr)
1622
1623{
1624 u64 bytenr;
1625 u64 ref_root;
1626 u64 orig_root;
1627 u64 ref_generation;
1628 u64 orig_generation;
1629 struct btrfs_key key;
1630 struct btrfs_file_extent_item *fi;
1631 int i;
1632 int ret;
1633 int slot;
1634 int level;
1635
1636 BUG_ON(start_slot < 0);
1637 BUG_ON(start_slot + nr > btrfs_header_nritems(buf));
1638
1639 ref_root = btrfs_header_owner(buf);
1640 ref_generation = btrfs_header_generation(buf);
1641 orig_root = btrfs_header_owner(orig_buf);
1642 orig_generation = btrfs_header_generation(orig_buf);
1643 level = btrfs_header_level(buf);
1644
1645 if (!root->ref_cows) {
1646 if (level == 0 &&
1647 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
1648 return 0;
1649 if (level != 0 &&
1650 root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
1651 return 0;
1652 }
1653
1654 for (i = 0, slot = start_slot; i < nr; i++, slot++) {
1655 cond_resched();
1656 if (level == 0) {
1657 btrfs_item_key_to_cpu(buf, &key, slot);
1658 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1659 continue;
1660 fi = btrfs_item_ptr(buf, slot,
1661 struct btrfs_file_extent_item);
1662 if (btrfs_file_extent_type(buf, fi) ==
1663 BTRFS_FILE_EXTENT_INLINE)
1664 continue;
1665 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1666 if (bytenr == 0)
1667 continue;
31840ae1
ZY		 1668 ret = __btrfs_update_extent_ref(trans, root, bytenr,
1669 orig_buf->start, buf->start,
1670 orig_root, ref_root,
1671 orig_generation, ref_generation,
3bb1a1bc 1672 key.objectid);
31840ae1
ZY		 1673 if (ret)
1674 goto fail;
1675 } else {
1676 bytenr = btrfs_node_blockptr(buf, slot);
31840ae1
ZY		 1677 ret = __btrfs_update_extent_ref(trans, root, bytenr,
1678 orig_buf->start, buf->start,
1679 orig_root, ref_root,
1680 orig_generation, ref_generation,
3bb1a1bc 1681 level - 1);
31840ae1
ZY		 1682 if (ret)
1683 goto fail;
1684 }
1685 }
1686 return 0;
1687fail:
1688 WARN_ON(1);
1689 return -1;
1690}
1691
9078a3e1
CM		 1692static int write_one_cache_group(struct btrfs_trans_handle *trans,
1693 struct btrfs_root *root,
1694 struct btrfs_path *path,
1695 struct btrfs_block_group_cache *cache)
1696{
1697 int ret;
1698 int pending_ret;
1699 struct btrfs_root *extent_root = root->fs_info->extent_root;
5f39d397
CM		 1700 unsigned long bi;
1701 struct extent_buffer *leaf;
9078a3e1 1702
9078a3e1 1703 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
54aa1f4d
CM		 1704 if (ret < 0)
1705 goto fail;
9078a3e1 1706 BUG_ON(ret);
5f39d397
CM		 1707
1708 leaf = path->nodes[0];
1709 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
1710 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
1711 btrfs_mark_buffer_dirty(leaf);
9078a3e1 1712 btrfs_release_path(extent_root, path);
54aa1f4d 1713fail:
87ef2bb4
CM		 1714 finish_current_insert(trans, extent_root, 0);
1715 pending_ret = del_pending_extents(trans, extent_root, 0);
9078a3e1
CM		 1716 if (ret)
1717 return ret;
1718 if (pending_ret)
1719 return pending_ret;
1720 return 0;
1721
1722}
1723
96b5179d
CM		 1724int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
1725 struct btrfs_root *root)
9078a3e1 1726{
0f9dd46c
JB		 1727 struct btrfs_block_group_cache *cache, *entry;
1728 struct rb_node *n;
9078a3e1
CM		 1729 int err = 0;
1730 int werr = 0;
9078a3e1 1731 struct btrfs_path *path;
96b5179d 1732 u64 last = 0;
9078a3e1
CM		 1733
1734 path = btrfs_alloc_path();
1735 if (!path)
1736 return -ENOMEM;
1737
d397712b 1738 while (1) {
0f9dd46c
JB		 1739 cache = NULL;
1740 spin_lock(&root->fs_info->block_group_cache_lock);
1741 for (n = rb_first(&root->fs_info->block_group_cache_tree);
1742 n; n = rb_next(n)) {
1743 entry = rb_entry(n, struct btrfs_block_group_cache,
1744 cache_node);
1745 if (entry->dirty) {
1746 cache = entry;
1747 break;
1748 }
1749 }
1750 spin_unlock(&root->fs_info->block_group_cache_lock);
54aa1f4d 1751
0f9dd46c 1752 if (!cache)
96b5179d 1753 break;
0f9dd46c 1754
e8569813 1755 cache->dirty = 0;
0f9dd46c
JB		 1756 last += cache->key.offset;
1757
96b5179d
CM		 1758 err = write_one_cache_group(trans, root,
1759 path, cache);
1760 /*
1761 * if we fail to write the cache group, we want
1762 * to keep it marked dirty in hopes that a later
1763 * write will work
1764 */
1765 if (err) {
1766 werr = err;
1767 continue;
9078a3e1
CM		 1768 }
1769 }
1770 btrfs_free_path(path);
1771 return werr;
1772}
1773
d2fb3437
YZ		 1774int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
1775{
1776 struct btrfs_block_group_cache *block_group;
1777 int readonly = 0;
1778
1779 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
1780 if (!block_group || block_group->ro)
1781 readonly = 1;
1782 if (block_group)
1783 put_block_group(block_group);
1784 return readonly;
1785}
1786
593060d7
CM		 1787static int update_space_info(struct btrfs_fs_info *info, u64 flags,
1788 u64 total_bytes, u64 bytes_used,
1789 struct btrfs_space_info **space_info)
1790{
1791 struct btrfs_space_info *found;
1792
1793 found = __find_space_info(info, flags);
1794 if (found) {
25179201 1795 spin_lock(&found->lock);
593060d7
CM		 1796 found->total_bytes += total_bytes;
1797 found->bytes_used += bytes_used;
8f18cf13 1798 found->full = 0;
25179201 1799 spin_unlock(&found->lock);
593060d7
CM		 1800 *space_info = found;
1801 return 0;
1802 }
c146afad 1803 found = kzalloc(sizeof(*found), GFP_NOFS);
593060d7
CM		 1804 if (!found)
1805 return -ENOMEM;
1806
1807 list_add(&found->list, &info->space_info);
0f9dd46c 1808 INIT_LIST_HEAD(&found->block_groups);
80eb234a 1809 init_rwsem(&found->groups_sem);
0f9dd46c 1810 spin_lock_init(&found->lock);
593060d7
CM		 1811 found->flags = flags;
1812 found->total_bytes = total_bytes;
1813 found->bytes_used = bytes_used;
1814 found->bytes_pinned = 0;
e8569813 1815 found->bytes_reserved = 0;
c146afad 1816 found->bytes_readonly = 0;
593060d7 1817 found->full = 0;
0ef3e66b 1818 found->force_alloc = 0;
593060d7
CM		 1819 *space_info = found;
1820 return 0;
1821}
1822
8790d502
CM		 1823static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
1824{
1825 u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
611f0e00 1826 BTRFS_BLOCK_GROUP_RAID1 |
321aecc6 1827 BTRFS_BLOCK_GROUP_RAID10 |
611f0e00 1828 BTRFS_BLOCK_GROUP_DUP);
8790d502
CM		 1829 if (extra_flags) {
1830 if (flags & BTRFS_BLOCK_GROUP_DATA)
1831 fs_info->avail_data_alloc_bits |= extra_flags;
1832 if (flags & BTRFS_BLOCK_GROUP_METADATA)
1833 fs_info->avail_metadata_alloc_bits |= extra_flags;
1834 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
1835 fs_info->avail_system_alloc_bits |= extra_flags;
1836 }
1837}
593060d7 1838
c146afad
YZ		 1839static void set_block_group_readonly(struct btrfs_block_group_cache *cache)
1840{
1841 spin_lock(&cache->space_info->lock);
1842 spin_lock(&cache->lock);
1843 if (!cache->ro) {
1844 cache->space_info->bytes_readonly += cache->key.offset -
1845 btrfs_block_group_used(&cache->item);
1846 cache->ro = 1;
1847 }
1848 spin_unlock(&cache->lock);
1849 spin_unlock(&cache->space_info->lock);
1850}
1851
2b82032c 1852u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
ec44a35c 1853{
2b82032c 1854 u64 num_devices = root->fs_info->fs_devices->rw_devices;
a061fc8d
CM		 1855
1856 if (num_devices == 1)
1857 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
1858 if (num_devices < 4)
1859 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
1860
ec44a35c
CM		 1861 if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
1862 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
a061fc8d 1863 BTRFS_BLOCK_GROUP_RAID10))) {
ec44a35c 1864 flags &= ~BTRFS_BLOCK_GROUP_DUP;
a061fc8d 1865 }
ec44a35c
CM		 1866
1867 if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
a061fc8d 1868 (flags & BTRFS_BLOCK_GROUP_RAID10)) {
ec44a35c 1869 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
a061fc8d 1870 }
ec44a35c
CM		 1871
1872 if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
1873 ((flags & BTRFS_BLOCK_GROUP_RAID1) |
1874 (flags & BTRFS_BLOCK_GROUP_RAID10) |
1875 (flags & BTRFS_BLOCK_GROUP_DUP)))
1876 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
1877 return flags;
1878}
1879
6324fbf3
CM		 1880static int do_chunk_alloc(struct btrfs_trans_handle *trans,
1881 struct btrfs_root *extent_root, u64 alloc_bytes,
0ef3e66b 1882 u64 flags, int force)
6324fbf3
CM		 1883{
1884 struct btrfs_space_info *space_info;
1885 u64 thresh;
c146afad
YZ		 1886 int ret = 0;
1887
1888 mutex_lock(&extent_root->fs_info->chunk_mutex);
6324fbf3 1889
2b82032c 1890 flags = btrfs_reduce_alloc_profile(extent_root, flags);
ec44a35c 1891
6324fbf3 1892 space_info = __find_space_info(extent_root->fs_info, flags);
593060d7
CM		 1893 if (!space_info) {
1894 ret = update_space_info(extent_root->fs_info, flags,
1895 0, 0, &space_info);
1896 BUG_ON(ret);
1897 }
6324fbf3
CM		 1898 BUG_ON(!space_info);
1899
25179201 1900 spin_lock(&space_info->lock);
0ef3e66b
CM		 1901 if (space_info->force_alloc) {
1902 force = 1;
1903 space_info->force_alloc = 0;
1904 }
25179201
JB		 1905 if (space_info->full) {
1906 spin_unlock(&space_info->lock);
925baedd 1907 goto out;
25179201 1908 }
6324fbf3 1909
c146afad
YZ		 1910 thresh = space_info->total_bytes - space_info->bytes_readonly;
1911 thresh = div_factor(thresh, 6);
0ef3e66b 1912 if (!force &&
e8569813 1913 (space_info->bytes_used + space_info->bytes_pinned +
25179201
JB		 1914 space_info->bytes_reserved + alloc_bytes) < thresh) {
1915 spin_unlock(&space_info->lock);
925baedd 1916 goto out;
25179201 1917 }
25179201
JB		 1918 spin_unlock(&space_info->lock);
1919
2b82032c 1920 ret = btrfs_alloc_chunk(trans, extent_root, flags);
d397712b 1921 if (ret)
6324fbf3 1922 space_info->full = 1;
a74a4b97 1923out:
c146afad 1924 mutex_unlock(&extent_root->fs_info->chunk_mutex);
0f9dd46c 1925 return ret;
6324fbf3
CM		 1926}
1927
9078a3e1
CM		 1928static int update_block_group(struct btrfs_trans_handle *trans,
1929 struct btrfs_root *root,
db94535d 1930 u64 bytenr, u64 num_bytes, int alloc,
0b86a832 1931 int mark_free)
9078a3e1
CM		 1932{
1933 struct btrfs_block_group_cache *cache;
1934 struct btrfs_fs_info *info = root->fs_info;
db94535d 1935 u64 total = num_bytes;
9078a3e1 1936 u64 old_val;
db94535d 1937 u64 byte_in_group;
3e1ad54f 1938
d397712b 1939 while (total) {
db94535d 1940 cache = btrfs_lookup_block_group(info, bytenr);
f3465ca4 1941 if (!cache)
9078a3e1 1942 return -1;
db94535d
CM		 1943 byte_in_group = bytenr - cache->key.objectid;
1944 WARN_ON(byte_in_group > cache->key.offset);
9078a3e1 1945
25179201 1946 spin_lock(&cache->space_info->lock);
c286ac48 1947 spin_lock(&cache->lock);
0f9dd46c 1948 cache->dirty = 1;
9078a3e1 1949 old_val = btrfs_block_group_used(&cache->item);
db94535d 1950 num_bytes = min(total, cache->key.offset - byte_in_group);
cd1bc465 1951 if (alloc) {
db94535d 1952 old_val += num_bytes;
6324fbf3 1953 cache->space_info->bytes_used += num_bytes;
a512bbf8 1954 if (cache->ro)
c146afad 1955 cache->space_info->bytes_readonly -= num_bytes;
c286ac48
CM		 1956 btrfs_set_block_group_used(&cache->item, old_val);
1957 spin_unlock(&cache->lock);
25179201 1958 spin_unlock(&cache->space_info->lock);
cd1bc465 1959 } else {
db94535d 1960 old_val -= num_bytes;
6324fbf3 1961 cache->space_info->bytes_used -= num_bytes;
c146afad
YZ		 1962 if (cache->ro)
1963 cache->space_info->bytes_readonly += num_bytes;
c286ac48
CM		 1964 btrfs_set_block_group_used(&cache->item, old_val);
1965 spin_unlock(&cache->lock);
25179201 1966 spin_unlock(&cache->space_info->lock);
f510cfec 1967 if (mark_free) {
0f9dd46c 1968 int ret;
1f3c79a2
LH		 1969
1970 ret = btrfs_discard_extent(root, bytenr,
1971 num_bytes);
1972 WARN_ON(ret);
1973
0f9dd46c
JB		 1974 ret = btrfs_add_free_space(cache, bytenr,
1975 num_bytes);
d2fb3437 1976 WARN_ON(ret);
e37c9e69 1977 }
cd1bc465 1978 }
d2fb3437 1979 put_block_group(cache);
db94535d
CM		 1980 total -= num_bytes;
1981 bytenr += num_bytes;
9078a3e1
CM		 1982 }
1983 return 0;
1984}
6324fbf3 1985
a061fc8d
CM		 1986static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
1987{
0f9dd46c 1988 struct btrfs_block_group_cache *cache;
d2fb3437 1989 u64 bytenr;
0f9dd46c
JB		 1990
1991 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
1992 if (!cache)
a061fc8d 1993 return 0;
0f9dd46c 1994
d2fb3437
YZ		 1995 bytenr = cache->key.objectid;
1996 put_block_group(cache);
1997
1998 return bytenr;
a061fc8d
CM		 1999}
2000
e02119d5 2001int btrfs_update_pinned_extents(struct btrfs_root *root,
324ae4df
Y		 2002 u64 bytenr, u64 num, int pin)
2003{
2004 u64 len;
2005 struct btrfs_block_group_cache *cache;
2006 struct btrfs_fs_info *fs_info = root->fs_info;
2007
25179201 2008 WARN_ON(!mutex_is_locked(&root->fs_info->pinned_mutex));
324ae4df
Y		 2009 if (pin) {
2010 set_extent_dirty(&fs_info->pinned_extents,
2011 bytenr, bytenr + num - 1, GFP_NOFS);
2012 } else {
2013 clear_extent_dirty(&fs_info->pinned_extents,
2014 bytenr, bytenr + num - 1, GFP_NOFS);
2015 }
2016 while (num > 0) {
2017 cache = btrfs_lookup_block_group(fs_info, bytenr);
e8569813
ZY		 2018 BUG_ON(!cache);
2019 len = min(num, cache->key.offset -
2020 (bytenr - cache->key.objectid));
324ae4df 2021 if (pin) {
25179201 2022 spin_lock(&cache->space_info->lock);
e8569813
ZY		 2023 spin_lock(&cache->lock);
2024 cache->pinned += len;
2025 cache->space_info->bytes_pinned += len;
2026 spin_unlock(&cache->lock);
25179201 2027 spin_unlock(&cache->space_info->lock);
324ae4df
Y		 2028 fs_info->total_pinned += len;
2029 } else {
25179201 2030 spin_lock(&cache->space_info->lock);
e8569813
ZY		 2031 spin_lock(&cache->lock);
2032 cache->pinned -= len;
2033 cache->space_info->bytes_pinned -= len;
2034 spin_unlock(&cache->lock);
25179201 2035 spin_unlock(&cache->space_info->lock);
324ae4df 2036 fs_info->total_pinned -= len;
07103a3c
JB		 2037 if (cache->cached)
2038 btrfs_add_free_space(cache, bytenr, len);
324ae4df 2039 }
d2fb3437 2040 put_block_group(cache);
324ae4df
Y		 2041 bytenr += len;
2042 num -= len;
2043 }
2044 return 0;
2045}
9078a3e1 2046
e8569813
ZY		 2047static int update_reserved_extents(struct btrfs_root *root,
2048 u64 bytenr, u64 num, int reserve)
2049{
2050 u64 len;
2051 struct btrfs_block_group_cache *cache;
2052 struct btrfs_fs_info *fs_info = root->fs_info;
2053
e8569813
ZY		 2054 while (num > 0) {
2055 cache = btrfs_lookup_block_group(fs_info, bytenr);
2056 BUG_ON(!cache);
2057 len = min(num, cache->key.offset -
2058 (bytenr - cache->key.objectid));
25179201
JB		 2059
2060 spin_lock(&cache->space_info->lock);
2061 spin_lock(&cache->lock);
e8569813 2062 if (reserve) {
e8569813
ZY		 2063 cache->reserved += len;
2064 cache->space_info->bytes_reserved += len;
e8569813 2065 } else {
e8569813
ZY		 2066 cache->reserved -= len;
2067 cache->space_info->bytes_reserved -= len;
e8569813 2068 }
25179201
JB		 2069 spin_unlock(&cache->lock);
2070 spin_unlock(&cache->space_info->lock);
d2fb3437 2071 put_block_group(cache);
e8569813
ZY		 2072 bytenr += len;
2073 num -= len;
2074 }
2075 return 0;
2076}
2077
d1310b2e 2078int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
ccd467d6 2079{
ccd467d6 2080 u64 last = 0;
1a5bc167
CM		 2081 u64 start;
2082 u64 end;
d1310b2e 2083 struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
ccd467d6 2084 int ret;
ccd467d6 2085
25179201 2086 mutex_lock(&root->fs_info->pinned_mutex);
d397712b 2087 while (1) {
1a5bc167
CM		 2088 ret = find_first_extent_bit(pinned_extents, last,
2089 &start, &end, EXTENT_DIRTY);
2090 if (ret)
ccd467d6 2091 break;
1a5bc167
CM		 2092 set_extent_dirty(copy, start, end, GFP_NOFS);
2093 last = end + 1;
ccd467d6 2094 }
25179201 2095 mutex_unlock(&root->fs_info->pinned_mutex);
ccd467d6
CM		 2096 return 0;
2097}
2098
2099int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
2100 struct btrfs_root *root,
d1310b2e 2101 struct extent_io_tree *unpin)
a28ec197 2102{
1a5bc167
CM		 2103 u64 start;
2104 u64 end;
a28ec197 2105 int ret;
a28ec197 2106
25179201 2107 mutex_lock(&root->fs_info->pinned_mutex);
d397712b 2108 while (1) {
1a5bc167
CM		 2109 ret = find_first_extent_bit(unpin, 0, &start, &end,
2110 EXTENT_DIRTY);
2111 if (ret)
a28ec197 2112 break;
1f3c79a2
LH		 2113
2114 ret = btrfs_discard_extent(root, start, end + 1 - start);
2115
e02119d5 2116 btrfs_update_pinned_extents(root, start, end + 1 - start, 0);
1a5bc167 2117 clear_extent_dirty(unpin, start, end, GFP_NOFS);
1f3c79a2 2118
c286ac48 2119 if (need_resched()) {
25179201 2120 mutex_unlock(&root->fs_info->pinned_mutex);
c286ac48 2121 cond_resched();
25179201 2122 mutex_lock(&root->fs_info->pinned_mutex);
c286ac48 2123 }
a28ec197 2124 }
25179201 2125 mutex_unlock(&root->fs_info->pinned_mutex);
1f3c79a2 2126 return ret;
a28ec197
CM		 2127}
2128
98ed5174 2129static int finish_current_insert(struct btrfs_trans_handle *trans,
87ef2bb4 2130 struct btrfs_root *extent_root, int all)
037e6390 2131{
7bb86316
CM		 2132 u64 start;
2133 u64 end;
31840ae1 2134 u64 priv;
25179201 2135 u64 search = 0;
f3465ca4 2136 u64 skipped = 0;
7bb86316
CM		 2137 struct btrfs_fs_info *info = extent_root->fs_info;
2138 struct btrfs_path *path;
f3465ca4
JB		 2139 struct pending_extent_op *extent_op, *tmp;
2140 struct list_head insert_list, update_list;
037e6390 2141 int ret;
f3465ca4 2142 int num_inserts = 0, max_inserts;
037e6390 2143
7bb86316 2144 path = btrfs_alloc_path();
f3465ca4
JB		 2145 INIT_LIST_HEAD(&insert_list);
2146 INIT_LIST_HEAD(&update_list);
037e6390 2147
f3465ca4
JB		 2148 max_inserts = extent_root->leafsize /
2149 (2 * sizeof(struct btrfs_key) + 2 * sizeof(struct btrfs_item) +
2150 sizeof(struct btrfs_extent_ref) +
2151 sizeof(struct btrfs_extent_item));
2152again:
2153 mutex_lock(&info->extent_ins_mutex);
2154 while (1) {
25179201
JB		 2155 ret = find_first_extent_bit(&info->extent_ins, search, &start,
2156 &end, EXTENT_WRITEBACK);
2157 if (ret) {
5a7be515
YZ		 2158 if (skipped && all && !num_inserts &&
2159 list_empty(&update_list)) {
f3465ca4 2160 skipped = 0;
b4eec2ca 2161 search = 0;
25179201
JB		 2162 continue;
2163 }
f3465ca4 2164 mutex_unlock(&info->extent_ins_mutex);
26b8003f 2165 break;
25179201
JB		 2166 }
2167
2168 ret = try_lock_extent(&info->extent_ins, start, end, GFP_NOFS);
2169 if (!ret) {
f3465ca4 2170 skipped = 1;
87ef2bb4 2171 search = end + 1;
f3465ca4
JB		 2172 if (need_resched()) {
2173 mutex_unlock(&info->extent_ins_mutex);
2174 cond_resched();
2175 mutex_lock(&info->extent_ins_mutex);
2176 }
25179201
JB		 2177 continue;
2178 }
26b8003f 2179
31840ae1
ZY		 2180 ret = get_state_private(&info->extent_ins, start, &priv);
2181 BUG_ON(ret);
f3465ca4 2182 extent_op = (struct pending_extent_op *)(unsigned long) priv;
25179201 2183
31840ae1 2184 if (extent_op->type == PENDING_EXTENT_INSERT) {
f3465ca4
JB		 2185 num_inserts++;
2186 list_add_tail(&extent_op->list, &insert_list);
2187 search = end + 1;
2188 if (num_inserts == max_inserts) {
2189 mutex_unlock(&info->extent_ins_mutex);
2190 break;
2191 }
2192 } else if (extent_op->type == PENDING_BACKREF_UPDATE) {
2193 list_add_tail(&extent_op->list, &update_list);
2194 search = end + 1;
2195 } else {
2196 BUG();
2197 }
2198 }
c286ac48 2199
f3465ca4 2200 /*
b4eec2ca 2201 * process the update list, clear the writeback bit for it, and if
f3465ca4
JB		 2202 * somebody marked this thing for deletion then just unlock it and be
2203 * done, the free_extents will handle it
2204 */
2205 mutex_lock(&info->extent_ins_mutex);
2206 list_for_each_entry_safe(extent_op, tmp, &update_list, list) {
2207 clear_extent_bits(&info->extent_ins, extent_op->bytenr,
2208 extent_op->bytenr + extent_op->num_bytes - 1,
2209 EXTENT_WRITEBACK, GFP_NOFS);
2210 if (extent_op->del) {
2211 list_del_init(&extent_op->list);
2212 unlock_extent(&info->extent_ins, extent_op->bytenr,
2213 extent_op->bytenr + extent_op->num_bytes
2214 - 1, GFP_NOFS);
2215 kfree(extent_op);
2216 }
2217 }
2218 mutex_unlock(&info->extent_ins_mutex);
c286ac48 2219
f3465ca4
JB		 2220 /*
2221 * still have things left on the update list, go ahead an update
2222 * everything
2223 */
2224 if (!list_empty(&update_list)) {
2225 ret = update_backrefs(trans, extent_root, path, &update_list);
2226 BUG_ON(ret);
2227 }
c286ac48 2228
f3465ca4
JB		 2229 /*
2230 * if no inserts need to be done, but we skipped some extents and we
2231 * need to make sure everything is cleaned then reset everything and
2232 * go back to the beginning
2233 */
2234 if (!num_inserts && all && skipped) {
2235 search = 0;
2236 skipped = 0;
2237 INIT_LIST_HEAD(&update_list);
2238 INIT_LIST_HEAD(&insert_list);
2239 goto again;
2240 } else if (!num_inserts) {
2241 goto out;
2242 }
31840ae1 2243
f3465ca4
JB		 2244 /*
2245 * process the insert extents list. Again if we are deleting this
2246 * extent, then just unlock it, pin down the bytes if need be, and be
2247 * done with it. Saves us from having to actually insert the extent
2248 * into the tree and then subsequently come along and delete it
2249 */
2250 mutex_lock(&info->extent_ins_mutex);
2251 list_for_each_entry_safe(extent_op, tmp, &insert_list, list) {
2252 clear_extent_bits(&info->extent_ins, extent_op->bytenr,
2253 extent_op->bytenr + extent_op->num_bytes - 1,
2254 EXTENT_WRITEBACK, GFP_NOFS);
2255 if (extent_op->del) {
34bf63c4 2256 u64 used;
f3465ca4
JB		 2257 list_del_init(&extent_op->list);
2258 unlock_extent(&info->extent_ins, extent_op->bytenr,
2259 extent_op->bytenr + extent_op->num_bytes
2260 - 1, GFP_NOFS);
2261
2262 mutex_lock(&extent_root->fs_info->pinned_mutex);
2263 ret = pin_down_bytes(trans, extent_root,
2264 extent_op->bytenr,
2265 extent_op->num_bytes, 0);
2266 mutex_unlock(&extent_root->fs_info->pinned_mutex);
2267
34bf63c4
YZ		 2268 spin_lock(&info->delalloc_lock);
2269 used = btrfs_super_bytes_used(&info->super_copy);
2270 btrfs_set_super_bytes_used(&info->super_copy,
2271 used - extent_op->num_bytes);
2272 used = btrfs_root_used(&extent_root->root_item);
2273 btrfs_set_root_used(&extent_root->root_item,
2274 used - extent_op->num_bytes);
2275 spin_unlock(&info->delalloc_lock);
2276
f3465ca4
JB		 2277 ret = update_block_group(trans, extent_root,
2278 extent_op->bytenr,
2279 extent_op->num_bytes,
2280 0, ret > 0);
2281 BUG_ON(ret);
2282 kfree(extent_op);
2283 num_inserts--;
d8d5f3e1 2284 }
f3465ca4
JB		 2285 }
2286 mutex_unlock(&info->extent_ins_mutex);
31840ae1 2287
f3465ca4
JB		 2288 ret = insert_extents(trans, extent_root, path, &insert_list,
2289 num_inserts);
2290 BUG_ON(ret);
2291
2292 /*
2293 * if we broke out of the loop in order to insert stuff because we hit
2294 * the maximum number of inserts at a time we can handle, then loop
2295 * back and pick up where we left off
2296 */
2297 if (num_inserts == max_inserts) {
2298 INIT_LIST_HEAD(&insert_list);
2299 INIT_LIST_HEAD(&update_list);
2300 num_inserts = 0;
2301 goto again;
2302 }
2303
2304 /*
2305 * again, if we need to make absolutely sure there are no more pending
2306 * extent operations left and we know that we skipped some, go back to
2307 * the beginning and do it all again
2308 */
2309 if (all && skipped) {
2310 INIT_LIST_HEAD(&insert_list);
2311 INIT_LIST_HEAD(&update_list);
2312 search = 0;
2313 skipped = 0;
2314 num_inserts = 0;
2315 goto again;
037e6390 2316 }
f3465ca4 2317out:
7bb86316 2318 btrfs_free_path(path);
037e6390
CM		 2319 return 0;
2320}
2321
31840ae1
ZY		 2322static int pin_down_bytes(struct btrfs_trans_handle *trans,
2323 struct btrfs_root *root,
2324 u64 bytenr, u64 num_bytes, int is_data)
e20d96d6 2325{
1a5bc167 2326 int err = 0;
31840ae1 2327 struct extent_buffer *buf;
8ef97622 2328
31840ae1
ZY		 2329 if (is_data)
2330 goto pinit;
2331
2332 buf = btrfs_find_tree_block(root, bytenr, num_bytes);
2333 if (!buf)
2334 goto pinit;
2335
2336 /* we can reuse a block if it hasn't been written
2337 * and it is from this transaction. We can't
2338 * reuse anything from the tree log root because
2339 * it has tiny sub-transactions.
2340 */
2341 if (btrfs_buffer_uptodate(buf, 0) &&
2342 btrfs_try_tree_lock(buf)) {
2343 u64 header_owner = btrfs_header_owner(buf);
2344 u64 header_transid = btrfs_header_generation(buf);
2345 if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
1a40e23b 2346 header_owner != BTRFS_TREE_RELOC_OBJECTID &&
31840ae1
ZY		 2347 header_transid == trans->transid &&
2348 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
2349 clean_tree_block(NULL, root, buf);
2350 btrfs_tree_unlock(buf);
5f39d397 2351 free_extent_buffer(buf);
31840ae1 2352 return 1;
8ef97622 2353 }
31840ae1 2354 btrfs_tree_unlock(buf);
f4b9aa8d 2355 }
31840ae1
ZY		 2356 free_extent_buffer(buf);
2357pinit:
2358 btrfs_update_pinned_extents(root, bytenr, num_bytes, 1);
2359
be744175 2360 BUG_ON(err < 0);
e20d96d6
CM		 2361 return 0;
2362}
2363
fec577fb 2364/*
a28ec197 2365 * remove an extent from the root, returns 0 on success
fec577fb 2366 */
31840ae1
ZY		 2367static int __free_extent(struct btrfs_trans_handle *trans,
2368 struct btrfs_root *root,
2369 u64 bytenr, u64 num_bytes, u64 parent,
7bb86316 2370 u64 root_objectid, u64 ref_generation,
3bb1a1bc 2371 u64 owner_objectid, int pin, int mark_free)
a28ec197 2372{
5caf2a00 2373 struct btrfs_path *path;
e2fa7227 2374 struct btrfs_key key;
1261ec42
CM		 2375 struct btrfs_fs_info *info = root->fs_info;
2376 struct btrfs_root *extent_root = info->extent_root;
5f39d397 2377 struct extent_buffer *leaf;
a28ec197 2378 int ret;
952fccac
CM		 2379 int extent_slot = 0;
2380 int found_extent = 0;
2381 int num_to_del = 1;
234b63a0 2382 struct btrfs_extent_item *ei;
cf27e1ee 2383 u32 refs;
037e6390 2384
db94535d 2385 key.objectid = bytenr;
62e2749e 2386 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
db94535d 2387 key.offset = num_bytes;
5caf2a00 2388 path = btrfs_alloc_path();
54aa1f4d
CM		 2389 if (!path)
2390 return -ENOMEM;
5f26f772 2391
3c12ac72 2392 path->reada = 1;
3bb1a1bc
YZ		 2393 ret = lookup_extent_backref(trans, extent_root, path,
2394 bytenr, parent, root_objectid,
2395 ref_generation, owner_objectid, 1);
7bb86316 2396 if (ret == 0) {
952fccac
CM		 2397 struct btrfs_key found_key;
2398 extent_slot = path->slots[0];
d397712b 2399 while (extent_slot > 0) {
952fccac
CM		 2400 extent_slot--;
2401 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2402 extent_slot);
2403 if (found_key.objectid != bytenr)
2404 break;
2405 if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
2406 found_key.offset == num_bytes) {
2407 found_extent = 1;
2408 break;
2409 }
2410 if (path->slots[0] - extent_slot > 5)
2411 break;
2412 }
31840ae1
ZY		 2413 if (!found_extent) {
2414 ret = remove_extent_backref(trans, extent_root, path);
2415 BUG_ON(ret);
2416 btrfs_release_path(extent_root, path);
2417 ret = btrfs_search_slot(trans, extent_root,
2418 &key, path, -1, 1);
f3465ca4
JB		 2419 if (ret) {
2420 printk(KERN_ERR "umm, got %d back from search"
d397712b
CM		 2421 ", was looking for %llu\n", ret,
2422 (unsigned long long)bytenr);
f3465ca4
JB		 2423 btrfs_print_leaf(extent_root, path->nodes[0]);
2424 }
31840ae1
ZY		 2425 BUG_ON(ret);
2426 extent_slot = path->slots[0];
2427 }
7bb86316
CM		 2428 } else {
2429 btrfs_print_leaf(extent_root, path->nodes[0]);
2430 WARN_ON(1);
d397712b
CM		 2431 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
2432 "root %llu gen %llu owner %llu\n",
2433 (unsigned long long)bytenr,
2434 (unsigned long long)root_objectid,
2435 (unsigned long long)ref_generation,
2436 (unsigned long long)owner_objectid);
7bb86316 2437 }
5f39d397
CM		 2438
2439 leaf = path->nodes[0];
952fccac 2440 ei = btrfs_item_ptr(leaf, extent_slot,
123abc88 2441 struct btrfs_extent_item);
5f39d397
CM		 2442 refs = btrfs_extent_refs(leaf, ei);
2443 BUG_ON(refs == 0);
2444 refs -= 1;
2445 btrfs_set_extent_refs(leaf, ei, refs);
952fccac 2446
5f39d397
CM		 2447 btrfs_mark_buffer_dirty(leaf);
2448
952fccac 2449 if (refs == 0 && found_extent && path->slots[0] == extent_slot + 1) {
31840ae1
ZY		 2450 struct btrfs_extent_ref *ref;
2451 ref = btrfs_item_ptr(leaf, path->slots[0],
2452 struct btrfs_extent_ref);
2453 BUG_ON(btrfs_ref_num_refs(leaf, ref) != 1);
952fccac
CM		 2454 /* if the back ref and the extent are next to each other
2455 * they get deleted below in one shot
2456 */
2457 path->slots[0] = extent_slot;
2458 num_to_del = 2;
2459 } else if (found_extent) {
2460 /* otherwise delete the extent back ref */
31840ae1 2461 ret = remove_extent_backref(trans, extent_root, path);
952fccac
CM		 2462 BUG_ON(ret);
2463 /* if refs are 0, we need to setup the path for deletion */
2464 if (refs == 0) {
2465 btrfs_release_path(extent_root, path);
2466 ret = btrfs_search_slot(trans, extent_root, &key, path,
2467 -1, 1);
952fccac
CM		 2468 BUG_ON(ret);
2469 }
2470 }
2471
cf27e1ee 2472 if (refs == 0) {
db94535d
CM		 2473 u64 super_used;
2474 u64 root_used;
78fae27e
CM		 2475
2476 if (pin) {
25179201 2477 mutex_lock(&root->fs_info->pinned_mutex);
31840ae1
ZY		 2478 ret = pin_down_bytes(trans, root, bytenr, num_bytes,
2479 owner_objectid >= BTRFS_FIRST_FREE_OBJECTID);
25179201 2480 mutex_unlock(&root->fs_info->pinned_mutex);
c549228f
Y		 2481 if (ret > 0)
2482 mark_free = 1;
2483 BUG_ON(ret < 0);
78fae27e 2484 }
58176a96 2485 /* block accounting for super block */
75eff68e 2486 spin_lock(&info->delalloc_lock);
db94535d
CM		 2487 super_used = btrfs_super_bytes_used(&info->super_copy);
2488 btrfs_set_super_bytes_used(&info->super_copy,
2489 super_used - num_bytes);
58176a96
JB		 2490
2491 /* block accounting for root item */
db94535d 2492 root_used = btrfs_root_used(&root->root_item);
5f39d397 2493 btrfs_set_root_used(&root->root_item,
db94535d 2494 root_used - num_bytes);
34bf63c4 2495 spin_unlock(&info->delalloc_lock);
952fccac
CM		 2496 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
2497 num_to_del);
31840ae1 2498 BUG_ON(ret);
25179201 2499 btrfs_release_path(extent_root, path);
21af804c 2500
459931ec
CM		 2501 if (owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
2502 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
2503 BUG_ON(ret);
2504 }
2505
dcbdd4dc
CM		 2506 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
2507 mark_free);
2508 BUG_ON(ret);
a28ec197 2509 }
5caf2a00 2510 btrfs_free_path(path);
87ef2bb4 2511 finish_current_insert(trans, extent_root, 0);
a28ec197
CM		 2512 return ret;
2513}
2514
a28ec197
CM		 2515/*
2516 * find all the blocks marked as pending in the radix tree and remove
2517 * them from the extent map
2518 */
d397712b
CM		 2519static int del_pending_extents(struct btrfs_trans_handle *trans,
2520 struct btrfs_root *extent_root, int all)
a28ec197
CM		 2521{
2522 int ret;
e20d96d6 2523 int err = 0;
1a5bc167
CM		 2524 u64 start;
2525 u64 end;
31840ae1 2526 u64 priv;
25179201 2527 u64 search = 0;
f3465ca4 2528 int nr = 0, skipped = 0;
d1310b2e 2529 struct extent_io_tree *pending_del;
31840ae1
ZY		 2530 struct extent_io_tree *extent_ins;
2531 struct pending_extent_op *extent_op;
25179201 2532 struct btrfs_fs_info *info = extent_root->fs_info;
f3465ca4 2533 struct list_head delete_list;
8ef97622 2534
f3465ca4 2535 INIT_LIST_HEAD(&delete_list);
31840ae1 2536 extent_ins = &extent_root->fs_info->extent_ins;
1a5bc167 2537 pending_del = &extent_root->fs_info->pending_del;
a28ec197 2538
f3465ca4
JB		 2539again:
2540 mutex_lock(&info->extent_ins_mutex);
d397712b 2541 while (1) {
25179201
JB		 2542 ret = find_first_extent_bit(pending_del, search, &start, &end,
2543 EXTENT_WRITEBACK);
2544 if (ret) {
f3465ca4 2545 if (all && skipped && !nr) {
25179201 2546 search = 0;
5a7be515 2547 skipped = 0;
25179201
JB		 2548 continue;
2549 }
f3465ca4 2550 mutex_unlock(&info->extent_ins_mutex);
a28ec197 2551 break;
25179201
JB		 2552 }
2553
2554 ret = try_lock_extent(extent_ins, start, end, GFP_NOFS);
2555 if (!ret) {
2556 search = end+1;
f3465ca4
JB		 2557 skipped = 1;
2558
2559 if (need_resched()) {
2560 mutex_unlock(&info->extent_ins_mutex);
2561 cond_resched();
2562 mutex_lock(&info->extent_ins_mutex);
2563 }
2564
25179201
JB		 2565 continue;
2566 }
2567 BUG_ON(ret < 0);
31840ae1
ZY		 2568
2569 ret = get_state_private(pending_del, start, &priv);
2570 BUG_ON(ret);
2571 extent_op = (struct pending_extent_op *)(unsigned long)priv;
2572
25179201 2573 clear_extent_bits(pending_del, start, end, EXTENT_WRITEBACK,
1a5bc167 2574 GFP_NOFS);
31840ae1 2575 if (!test_range_bit(extent_ins, start, end,
25179201 2576 EXTENT_WRITEBACK, 0)) {
f3465ca4
JB		 2577 list_add_tail(&extent_op->list, &delete_list);
2578 nr++;
c286ac48 2579 } else {
31840ae1 2580 kfree(extent_op);
25179201
JB		 2581
2582 ret = get_state_private(&info->extent_ins, start,
2583 &priv);
31840ae1
ZY		 2584 BUG_ON(ret);
2585 extent_op = (struct pending_extent_op *)
25179201
JB		 2586 (unsigned long)priv;
2587
2588 clear_extent_bits(&info->extent_ins, start, end,
2589 EXTENT_WRITEBACK, GFP_NOFS);
31840ae1 2590
f3465ca4
JB		 2591 if (extent_op->type == PENDING_BACKREF_UPDATE) {
2592 list_add_tail(&extent_op->list, &delete_list);
2593 search = end + 1;
2594 nr++;
2595 continue;
2596 }
31840ae1 2597
25179201
JB		 2598 mutex_lock(&extent_root->fs_info->pinned_mutex);
2599 ret = pin_down_bytes(trans, extent_root, start,
2600 end + 1 - start, 0);
2601 mutex_unlock(&extent_root->fs_info->pinned_mutex);
2602
31840ae1 2603 ret = update_block_group(trans, extent_root, start,
25179201
JB		 2604 end + 1 - start, 0, ret > 0);
2605
f3465ca4 2606 unlock_extent(extent_ins, start, end, GFP_NOFS);
31840ae1
ZY		 2607 BUG_ON(ret);
2608 kfree(extent_op);
c286ac48 2609 }
1a5bc167
CM		 2610 if (ret)
2611 err = ret;
c286ac48 2612
f3465ca4
JB		 2613 search = end + 1;
2614
2615 if (need_resched()) {
2616 mutex_unlock(&info->extent_ins_mutex);
2617 cond_resched();
2618 mutex_lock(&info->extent_ins_mutex);
2619 }
2620 }
2621
2622 if (nr) {
2623 ret = free_extents(trans, extent_root, &delete_list);
2624 BUG_ON(ret);
2625 }
2626
2627 if (all && skipped) {
2628 INIT_LIST_HEAD(&delete_list);
2629 search = 0;
2630 nr = 0;
2631 goto again;
fec577fb 2632 }
f3465ca4 2633
e20d96d6 2634 return err;
fec577fb
CM		 2635}
2636
2637/*
2638 * remove an extent from the root, returns 0 on success
2639 */
925baedd 2640static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
31840ae1
ZY		 2641 struct btrfs_root *root,
2642 u64 bytenr, u64 num_bytes, u64 parent,
2643 u64 root_objectid, u64 ref_generation,
3bb1a1bc 2644 u64 owner_objectid, int pin)
fec577fb 2645{
9f5fae2f 2646 struct btrfs_root *extent_root = root->fs_info->extent_root;
fec577fb
CM		 2647 int pending_ret;
2648 int ret;
a28ec197 2649
db94535d 2650 WARN_ON(num_bytes < root->sectorsize);
fec577fb 2651 if (root == extent_root) {
f3465ca4
JB		 2652 struct pending_extent_op *extent_op = NULL;
2653
2654 mutex_lock(&root->fs_info->extent_ins_mutex);
2655 if (test_range_bit(&root->fs_info->extent_ins, bytenr,
2656 bytenr + num_bytes - 1, EXTENT_WRITEBACK, 0)) {
2657 u64 priv;
2658 ret = get_state_private(&root->fs_info->extent_ins,
2659 bytenr, &priv);
2660 BUG_ON(ret);
2661 extent_op = (struct pending_extent_op *)
2662 (unsigned long)priv;
2663
2664 extent_op->del = 1;
2665 if (extent_op->type == PENDING_EXTENT_INSERT) {
2666 mutex_unlock(&root->fs_info->extent_ins_mutex);
2667 return 0;
2668 }
2669 }
2670
2671 if (extent_op) {
2672 ref_generation = extent_op->orig_generation;
2673 parent = extent_op->orig_parent;
2674 }
31840ae1
ZY		 2675
2676 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2677 BUG_ON(!extent_op);
2678
2679 extent_op->type = PENDING_EXTENT_DELETE;
2680 extent_op->bytenr = bytenr;
2681 extent_op->num_bytes = num_bytes;
2682 extent_op->parent = parent;
2683 extent_op->orig_parent = parent;
2684 extent_op->generation = ref_generation;
2685 extent_op->orig_generation = ref_generation;
2686 extent_op->level = (int)owner_objectid;
f3465ca4
JB		 2687 INIT_LIST_HEAD(&extent_op->list);
2688 extent_op->del = 0;
31840ae1
ZY		 2689
2690 set_extent_bits(&root->fs_info->pending_del,
2691 bytenr, bytenr + num_bytes - 1,
25179201 2692 EXTENT_WRITEBACK, GFP_NOFS);
31840ae1
ZY		 2693 set_state_private(&root->fs_info->pending_del,
2694 bytenr, (unsigned long)extent_op);
25179201 2695 mutex_unlock(&root->fs_info->extent_ins_mutex);
fec577fb
CM		 2696 return 0;
2697 }
4bef0848 2698 /* if metadata always pin */
d00aff00
CM		 2699 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
2700 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
0f9dd46c
JB		 2701 struct btrfs_block_group_cache *cache;
2702
d00aff00 2703 /* btrfs_free_reserved_extent */
0f9dd46c
JB		 2704 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
2705 BUG_ON(!cache);
2706 btrfs_add_free_space(cache, bytenr, num_bytes);
d2fb3437 2707 put_block_group(cache);
e8569813 2708 update_reserved_extents(root, bytenr, num_bytes, 0);
d00aff00
CM		 2709 return 0;
2710 }
4bef0848 2711 pin = 1;
d00aff00 2712 }
4bef0848
CM		 2713
2714 /* if data pin when any transaction has committed this */
2715 if (ref_generation != trans->transid)
2716 pin = 1;
2717
31840ae1 2718 ret = __free_extent(trans, root, bytenr, num_bytes, parent,
3bb1a1bc
YZ		 2719 root_objectid, ref_generation,
2720 owner_objectid, pin, pin == 0);
ee6e6504 2721
87ef2bb4
CM		 2722 finish_current_insert(trans, root->fs_info->extent_root, 0);
2723 pending_ret = del_pending_extents(trans, root->fs_info->extent_root, 0);
fec577fb
CM		 2724 return ret ? ret : pending_ret;
2725}
2726
925baedd 2727int btrfs_free_extent(struct btrfs_trans_handle *trans,
31840ae1
ZY		 2728 struct btrfs_root *root,
2729 u64 bytenr, u64 num_bytes, u64 parent,
2730 u64 root_objectid, u64 ref_generation,
3bb1a1bc 2731 u64 owner_objectid, int pin)
925baedd
CM		 2732{
2733 int ret;
2734
31840ae1 2735 ret = __btrfs_free_extent(trans, root, bytenr, num_bytes, parent,
925baedd 2736 root_objectid, ref_generation,
3bb1a1bc 2737 owner_objectid, pin);
925baedd
CM		 2738 return ret;
2739}
2740
87ee04eb
CM		 2741static u64 stripe_align(struct btrfs_root *root, u64 val)
2742{
2743 u64 mask = ((u64)root->stripesize - 1);
2744 u64 ret = (val + mask) & ~mask;
2745 return ret;
2746}
2747
fec577fb
CM		 2748/*
2749 * walks the btree of allocated extents and find a hole of a given size.
2750 * The key ins is changed to record the hole:
2751 * ins->objectid == block start
62e2749e 2752 * ins->flags = BTRFS_EXTENT_ITEM_KEY
fec577fb
CM		 2753 * ins->offset == number of blocks
2754 * Any available blocks before search_start are skipped.
2755 */
d397712b 2756static noinline int find_free_extent(struct btrfs_trans_handle *trans,
98ed5174
CM		 2757 struct btrfs_root *orig_root,
2758 u64 num_bytes, u64 empty_size,
2759 u64 search_start, u64 search_end,
2760 u64 hint_byte, struct btrfs_key *ins,
2761 u64 exclude_start, u64 exclude_nr,
2762 int data)
fec577fb 2763{
80eb234a 2764 int ret = 0;
d397712b 2765 struct btrfs_root *root = orig_root->fs_info->extent_root;
db94535d 2766 u64 total_needed = num_bytes;
239b14b3 2767 u64 *last_ptr = NULL;
4366211c 2768 u64 last_wanted = 0;
80eb234a 2769 struct btrfs_block_group_cache *block_group = NULL;
0ef3e66b 2770 int chunk_alloc_done = 0;
239b14b3 2771 int empty_cluster = 2 * 1024 * 1024;
0ef3e66b 2772 int allowed_chunk_alloc = 0;
80eb234a
JB		 2773 struct list_head *head = NULL, *cur = NULL;
2774 int loop = 0;
f5a31e16 2775 int extra_loop = 0;
80eb234a 2776 struct btrfs_space_info *space_info;
fec577fb 2777
db94535d 2778 WARN_ON(num_bytes < root->sectorsize);
b1a4d965 2779 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
80eb234a
JB		 2780 ins->objectid = 0;
2781 ins->offset = 0;
b1a4d965 2782
0ef3e66b
CM		 2783 if (orig_root->ref_cows || empty_size)
2784 allowed_chunk_alloc = 1;
2785
239b14b3
CM		 2786 if (data & BTRFS_BLOCK_GROUP_METADATA) {
2787 last_ptr = &root->fs_info->last_alloc;
4366211c 2788 empty_cluster = 64 * 1024;
239b14b3
CM		 2789 }
2790
0f9dd46c 2791 if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD))
239b14b3 2792 last_ptr = &root->fs_info->last_data_alloc;
0f9dd46c 2793
239b14b3 2794 if (last_ptr) {
4366211c 2795 if (*last_ptr) {
239b14b3 2796 hint_byte = *last_ptr;
4366211c
CM		 2797 last_wanted = *last_ptr;
2798 } else
239b14b3 2799 empty_size += empty_cluster;
4366211c
CM		 2800 } else {
2801 empty_cluster = 0;
239b14b3 2802 }
a061fc8d 2803 search_start = max(search_start, first_logical_byte(root, 0));
239b14b3 2804 search_start = max(search_start, hint_byte);
0b86a832 2805
42e70e7a 2806 if (last_wanted && search_start != last_wanted) {
4366211c 2807 last_wanted = 0;
42e70e7a
CM		 2808 empty_size += empty_cluster;
2809 }
4366211c 2810
42e70e7a 2811 total_needed += empty_size;
80eb234a 2812 block_group = btrfs_lookup_block_group(root->fs_info, search_start);
d899e052
YZ		 2813 if (!block_group)
2814 block_group = btrfs_lookup_first_block_group(root->fs_info,
2815 search_start);
80eb234a 2816 space_info = __find_space_info(root->fs_info, data);
0f9dd46c 2817
80eb234a
JB		 2818 down_read(&space_info->groups_sem);
2819 while (1) {
2820 struct btrfs_free_space *free_space;
0f9dd46c 2821 /*
80eb234a
JB		 2822 * the only way this happens if our hint points to a block
2823 * group thats not of the proper type, while looping this
2824 * should never happen
0f9dd46c 2825 */
8a1413a2
CM		 2826 if (empty_size)
2827 extra_loop = 1;
2828
42e70e7a
CM		 2829 if (!block_group)
2830 goto new_group_no_lock;
2831
ea6a478e
JB		 2832 if (unlikely(!block_group->cached)) {
2833 mutex_lock(&block_group->cache_mutex);
2834 ret = cache_block_group(root, block_group);
2835 mutex_unlock(&block_group->cache_mutex);
2836 if (ret)
2837 break;
2838 }
2839
25179201 2840 mutex_lock(&block_group->alloc_mutex);
80eb234a 2841 if (unlikely(!block_group_bits(block_group, data)))
0f9dd46c 2842 goto new_group;
0f9dd46c 2843
ea6a478e 2844 if (unlikely(block_group->ro))
80eb234a 2845 goto new_group;
0f9dd46c 2846
80eb234a
JB		 2847 free_space = btrfs_find_free_space(block_group, search_start,
2848 total_needed);
2849 if (free_space) {
2850 u64 start = block_group->key.objectid;
2851 u64 end = block_group->key.objectid +
0f9dd46c 2852 block_group->key.offset;
239b14b3 2853
80eb234a 2854 search_start = stripe_align(root, free_space->offset);
0f9dd46c 2855
80eb234a
JB		 2856 /* move on to the next group */
2857 if (search_start + num_bytes >= search_end)
2858 goto new_group;
e37c9e69 2859
80eb234a
JB		 2860 /* move on to the next group */
2861 if (search_start + num_bytes > end)
2862 goto new_group;
2863
4366211c 2864 if (last_wanted && search_start != last_wanted) {
3b7885bf 2865 total_needed += empty_cluster;
8a1413a2 2866 empty_size += empty_cluster;
4366211c 2867 last_wanted = 0;
3b7885bf
CM		 2868 /*
2869 * if search_start is still in this block group
2870 * then we just re-search this block group
2871 */
2872 if (search_start >= start &&
2873 search_start < end) {
2874 mutex_unlock(&block_group->alloc_mutex);
2875 continue;
2876 }
2877
2878 /* else we go to the next block group */
2879 goto new_group;
2880 }
2881
80eb234a
JB		 2882 if (exclude_nr > 0 &&
2883 (search_start + num_bytes > exclude_start &&
2884 search_start < exclude_start + exclude_nr)) {
2885 search_start = exclude_start + exclude_nr;
2886 /*
2887 * if search_start is still in this block group
2888 * then we just re-search this block group
2889 */
2890 if (search_start >= start &&
25179201
JB		 2891 search_start < end) {
2892 mutex_unlock(&block_group->alloc_mutex);
4366211c 2893 last_wanted = 0;
80eb234a 2894 continue;
25179201 2895 }
80eb234a
JB		 2896
2897 /* else we go to the next block group */
2898 goto new_group;
2899 }
2900
2901 ins->objectid = search_start;
2902 ins->offset = num_bytes;
25179201
JB		 2903
2904 btrfs_remove_free_space_lock(block_group, search_start,
2905 num_bytes);
80eb234a 2906 /* we are all good, lets return */
25179201 2907 mutex_unlock(&block_group->alloc_mutex);
80eb234a 2908 break;
0f9dd46c 2909 }
80eb234a 2910new_group:
42e70e7a 2911 mutex_unlock(&block_group->alloc_mutex);
d2fb3437
YZ		 2912 put_block_group(block_group);
2913 block_group = NULL;
42e70e7a 2914new_group_no_lock:
f5a31e16
CM		 2915 /* don't try to compare new allocations against the
2916 * last allocation any more
2917 */
4366211c 2918 last_wanted = 0;
f5a31e16 2919
80eb234a
JB		 2920 /*
2921 * Here's how this works.
2922 * loop == 0: we were searching a block group via a hint
2923 * and didn't find anything, so we start at
2924 * the head of the block groups and keep searching
2925 * loop == 1: we're searching through all of the block groups
2926 * if we hit the head again we have searched
2927 * all of the block groups for this space and we
2928 * need to try and allocate, if we cant error out.
2929 * loop == 2: we allocated more space and are looping through
2930 * all of the block groups again.
2931 */
2932 if (loop == 0) {
2933 head = &space_info->block_groups;
2934 cur = head->next;
80eb234a
JB		 2935 loop++;
2936 } else if (loop == 1 && cur == head) {
f5a31e16
CM		 2937 int keep_going;
2938
2939 /* at this point we give up on the empty_size
2940 * allocations and just try to allocate the min
2941 * space.
2942 *
2943 * The extra_loop field was set if an empty_size
2944 * allocation was attempted above, and if this
2945 * is try we need to try the loop again without
2946 * the additional empty_size.
2947 */
5b7c3fcc
CM		 2948 total_needed -= empty_size;
2949 empty_size = 0;
f5a31e16
CM		 2950 keep_going = extra_loop;
2951 loop++;
42e70e7a 2952
80eb234a
JB		 2953 if (allowed_chunk_alloc && !chunk_alloc_done) {
2954 up_read(&space_info->groups_sem);
2955 ret = do_chunk_alloc(trans, root, num_bytes +
2956 2 * 1024 * 1024, data, 1);
80eb234a 2957 down_read(&space_info->groups_sem);
2ed6d664
CM		 2958 if (ret < 0)
2959 goto loop_check;
80eb234a 2960 head = &space_info->block_groups;
f5a31e16
CM		 2961 /*
2962 * we've allocated a new chunk, keep
2963 * trying
2964 */
2965 keep_going = 1;
80eb234a
JB		 2966 chunk_alloc_done = 1;
2967 } else if (!allowed_chunk_alloc) {
2968 space_info->force_alloc = 1;
f5a31e16 2969 }
2ed6d664 2970loop_check:
f5a31e16
CM		 2971 if (keep_going) {
2972 cur = head->next;
2973 extra_loop = 0;
80eb234a
JB		 2974 } else {
2975 break;
2976 }
2977 } else if (cur == head) {
2978 break;
0f9dd46c 2979 }
0b86a832 2980
80eb234a
JB		 2981 block_group = list_entry(cur, struct btrfs_block_group_cache,
2982 list);
d2fb3437
YZ		 2983 atomic_inc(&block_group->count);
2984
80eb234a
JB		 2985 search_start = block_group->key.objectid;
2986 cur = cur->next;
f2654de4 2987 }
0b86a832 2988
80eb234a
JB		 2989 /* we found what we needed */
2990 if (ins->objectid) {
2991 if (!(data & BTRFS_BLOCK_GROUP_DATA))
d2fb3437 2992 trans->block_group = block_group->key.objectid;
0f9dd46c 2993
80eb234a
JB		 2994 if (last_ptr)
2995 *last_ptr = ins->objectid + ins->offset;
2996 ret = 0;
2997 } else if (!ret) {
d397712b
CM		 2998 printk(KERN_ERR "btrfs searching for %llu bytes, "
2999 "num_bytes %llu, loop %d, allowed_alloc %d\n",
3000 (unsigned long long)total_needed,
3001 (unsigned long long)num_bytes,
4ce4cb52 3002 loop, allowed_chunk_alloc);
80eb234a 3003 ret = -ENOSPC;
be744175 3004 }
d2fb3437
YZ		 3005 if (block_group)
3006 put_block_group(block_group);
be744175 3007
80eb234a 3008 up_read(&space_info->groups_sem);
0f70abe2 3009 return ret;
fec577fb 3010}
ec44a35c 3011
0f9dd46c
JB		 3012static void dump_space_info(struct btrfs_space_info *info, u64 bytes)
3013{
3014 struct btrfs_block_group_cache *cache;
0f9dd46c 3015
d397712b
CM		 3016 printk(KERN_INFO "space_info has %llu free, is %sfull\n",
3017 (unsigned long long)(info->total_bytes - info->bytes_used -
3018 info->bytes_pinned - info->bytes_reserved),
3019 (info->full) ? "" : "not ");
0f9dd46c 3020
80eb234a 3021 down_read(&info->groups_sem);
c6e30871 3022 list_for_each_entry(cache, &info->block_groups, list) {
0f9dd46c 3023 spin_lock(&cache->lock);
d397712b
CM		 3024 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
3025 "%llu pinned %llu reserved\n",
3026 (unsigned long long)cache->key.objectid,
3027 (unsigned long long)cache->key.offset,
3028 (unsigned long long)btrfs_block_group_used(&cache->item),
3029 (unsigned long long)cache->pinned,
3030 (unsigned long long)cache->reserved);
0f9dd46c
JB		 3031 btrfs_dump_free_space(cache, bytes);
3032 spin_unlock(&cache->lock);
3033 }
80eb234a 3034 up_read(&info->groups_sem);
0f9dd46c 3035}
e8569813 3036
e6dcd2dc
CM		 3037static int __btrfs_reserve_extent(struct btrfs_trans_handle *trans,
3038 struct btrfs_root *root,
3039 u64 num_bytes, u64 min_alloc_size,
3040 u64 empty_size, u64 hint_byte,
3041 u64 search_end, struct btrfs_key *ins,
3042 u64 data)
fec577fb
CM		 3043{
3044 int ret;
fbdc762b 3045 u64 search_start = 0;
8790d502 3046 u64 alloc_profile;
1261ec42 3047 struct btrfs_fs_info *info = root->fs_info;
925baedd 3048
6324fbf3 3049 if (data) {
8790d502 3050 alloc_profile = info->avail_data_alloc_bits &
d397712b 3051 info->data_alloc_profile;
8790d502 3052 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
6324fbf3 3053 } else if (root == root->fs_info->chunk_root) {
8790d502 3054 alloc_profile = info->avail_system_alloc_bits &
d397712b 3055 info->system_alloc_profile;
8790d502 3056 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
6324fbf3 3057 } else {
8790d502 3058 alloc_profile = info->avail_metadata_alloc_bits &
d397712b 3059 info->metadata_alloc_profile;
8790d502 3060 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
6324fbf3 3061 }
98d20f67 3062again:
2b82032c 3063 data = btrfs_reduce_alloc_profile(root, data);
0ef3e66b
CM		 3064 /*
3065 * the only place that sets empty_size is btrfs_realloc_node, which
3066 * is not called recursively on allocations
3067 */
3068 if (empty_size || root->ref_cows) {
593060d7 3069 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
6324fbf3 3070 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
0ef3e66b
CM		 3071 2 * 1024 * 1024,
3072 BTRFS_BLOCK_GROUP_METADATA |
3073 (info->metadata_alloc_profile &
3074 info->avail_metadata_alloc_bits), 0);
6324fbf3
CM		 3075 }
3076 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
0ef3e66b 3077 num_bytes + 2 * 1024 * 1024, data, 0);
6324fbf3 3078 }
0b86a832 3079
db94535d
CM		 3080 WARN_ON(num_bytes < root->sectorsize);
3081 ret = find_free_extent(trans, root, num_bytes, empty_size,
3082 search_start, search_end, hint_byte, ins,
26b8003f
CM		 3083 trans->alloc_exclude_start,
3084 trans->alloc_exclude_nr, data);
3b951516 3085
98d20f67
CM		 3086 if (ret == -ENOSPC && num_bytes > min_alloc_size) {
3087 num_bytes = num_bytes >> 1;
0f9dd46c 3088 num_bytes = num_bytes & ~(root->sectorsize - 1);
98d20f67 3089 num_bytes = max(num_bytes, min_alloc_size);
0ef3e66b
CM		 3090 do_chunk_alloc(trans, root->fs_info->extent_root,
3091 num_bytes, data, 1);
98d20f67
CM		 3092 goto again;
3093 }
ec44a35c 3094 if (ret) {
0f9dd46c
JB		 3095 struct btrfs_space_info *sinfo;
3096
3097 sinfo = __find_space_info(root->fs_info, data);
d397712b
CM		 3098 printk(KERN_ERR "btrfs allocation failed flags %llu, "
3099 "wanted %llu\n", (unsigned long long)data,
3100 (unsigned long long)num_bytes);
0f9dd46c 3101 dump_space_info(sinfo, num_bytes);
925baedd 3102 BUG();
925baedd 3103 }
0f9dd46c
JB		 3104
3105 return ret;
e6dcd2dc
CM		 3106}
3107
65b51a00
CM		 3108int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
3109{
0f9dd46c 3110 struct btrfs_block_group_cache *cache;
1f3c79a2 3111 int ret = 0;
0f9dd46c 3112
0f9dd46c
JB		 3113 cache = btrfs_lookup_block_group(root->fs_info, start);
3114 if (!cache) {
d397712b
CM		 3115 printk(KERN_ERR "Unable to find block group for %llu\n",
3116 (unsigned long long)start);
0f9dd46c
JB		 3117 return -ENOSPC;
3118 }
1f3c79a2
LH		 3119
3120 ret = btrfs_discard_extent(root, start, len);
3121
0f9dd46c 3122 btrfs_add_free_space(cache, start, len);
d2fb3437 3123 put_block_group(cache);
1a40e23b 3124 update_reserved_extents(root, start, len, 0);
1f3c79a2
LH		 3125
3126 return ret;
65b51a00
CM		 3127}
3128
e6dcd2dc
CM		 3129int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
3130 struct btrfs_root *root,
3131 u64 num_bytes, u64 min_alloc_size,
3132 u64 empty_size, u64 hint_byte,
3133 u64 search_end, struct btrfs_key *ins,
3134 u64 data)
3135{
3136 int ret;
e6dcd2dc
CM		 3137 ret = __btrfs_reserve_extent(trans, root, num_bytes, min_alloc_size,
3138 empty_size, hint_byte, search_end, ins,
3139 data);
e8569813 3140 update_reserved_extents(root, ins->objectid, ins->offset, 1);
e6dcd2dc
CM		 3141 return ret;
3142}
3143
3144static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
31840ae1 3145 struct btrfs_root *root, u64 parent,
e6dcd2dc 3146 u64 root_objectid, u64 ref_generation,
3bb1a1bc 3147 u64 owner, struct btrfs_key *ins)
e6dcd2dc
CM		 3148{
3149 int ret;
3150 int pending_ret;
3151 u64 super_used;
3152 u64 root_used;
3153 u64 num_bytes = ins->offset;
3154 u32 sizes[2];
3155 struct btrfs_fs_info *info = root->fs_info;
3156 struct btrfs_root *extent_root = info->extent_root;
3157 struct btrfs_extent_item *extent_item;
3158 struct btrfs_extent_ref *ref;
3159 struct btrfs_path *path;
3160 struct btrfs_key keys[2];
fec577fb 3161
31840ae1
ZY		 3162 if (parent == 0)
3163 parent = ins->objectid;
3164
58176a96 3165 /* block accounting for super block */
75eff68e 3166 spin_lock(&info->delalloc_lock);
db94535d
CM		 3167 super_used = btrfs_super_bytes_used(&info->super_copy);
3168 btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes);
26b8003f 3169
58176a96 3170 /* block accounting for root item */
db94535d
CM		 3171 root_used = btrfs_root_used(&root->root_item);
3172 btrfs_set_root_used(&root->root_item, root_used + num_bytes);
34bf63c4 3173 spin_unlock(&info->delalloc_lock);
58176a96 3174
26b8003f 3175 if (root == extent_root) {
31840ae1
ZY		 3176 struct pending_extent_op *extent_op;
3177
3178 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
3179 BUG_ON(!extent_op);
3180
3181 extent_op->type = PENDING_EXTENT_INSERT;
3182 extent_op->bytenr = ins->objectid;
3183 extent_op->num_bytes = ins->offset;
3184 extent_op->parent = parent;
3185 extent_op->orig_parent = 0;
3186 extent_op->generation = ref_generation;
3187 extent_op->orig_generation = 0;
3188 extent_op->level = (int)owner;
f3465ca4
JB		 3189 INIT_LIST_HEAD(&extent_op->list);
3190 extent_op->del = 0;
31840ae1 3191
25179201 3192 mutex_lock(&root->fs_info->extent_ins_mutex);
1a5bc167
CM		 3193 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
3194 ins->objectid + ins->offset - 1,
25179201 3195 EXTENT_WRITEBACK, GFP_NOFS);
31840ae1
ZY		 3196 set_state_private(&root->fs_info->extent_ins,
3197 ins->objectid, (unsigned long)extent_op);
25179201 3198 mutex_unlock(&root->fs_info->extent_ins_mutex);
26b8003f
CM		 3199 goto update_block;
3200 }
3201
47e4bb98 3202 memcpy(&keys[0], ins, sizeof(*ins));
47e4bb98
CM		 3203 keys[1].objectid = ins->objectid;
3204 keys[1].type = BTRFS_EXTENT_REF_KEY;
31840ae1 3205 keys[1].offset = parent;
47e4bb98
CM		 3206 sizes[0] = sizeof(*extent_item);
3207 sizes[1] = sizeof(*ref);
7bb86316
CM		 3208
3209 path = btrfs_alloc_path();
3210 BUG_ON(!path);
47e4bb98
CM		 3211
3212 ret = btrfs_insert_empty_items(trans, extent_root, path, keys,
3213 sizes, 2);
ccd467d6 3214 BUG_ON(ret);
0f9dd46c 3215
47e4bb98
CM		 3216 extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3217 struct btrfs_extent_item);
3218 btrfs_set_extent_refs(path->nodes[0], extent_item, 1);
3219 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
3220 struct btrfs_extent_ref);
3221
3222 btrfs_set_ref_root(path->nodes[0], ref, root_objectid);
3223 btrfs_set_ref_generation(path->nodes[0], ref, ref_generation);
3224 btrfs_set_ref_objectid(path->nodes[0], ref, owner);
31840ae1 3225 btrfs_set_ref_num_refs(path->nodes[0], ref, 1);
47e4bb98
CM		 3226
3227 btrfs_mark_buffer_dirty(path->nodes[0]);
3228
3229 trans->alloc_exclude_start = 0;
3230 trans->alloc_exclude_nr = 0;
7bb86316 3231 btrfs_free_path(path);
87ef2bb4
CM		 3232 finish_current_insert(trans, extent_root, 0);
3233 pending_ret = del_pending_extents(trans, extent_root, 0);
f510cfec 3234
925baedd
CM		 3235 if (ret)
3236 goto out;
e37c9e69 3237 if (pending_ret) {
925baedd
CM		 3238 ret = pending_ret;
3239 goto out;
e37c9e69 3240 }
26b8003f
CM		 3241
3242update_block:
d397712b
CM		 3243 ret = update_block_group(trans, root, ins->objectid,
3244 ins->offset, 1, 0);
f5947066 3245 if (ret) {
d397712b
CM		 3246 printk(KERN_ERR "btrfs update block group failed for %llu "
3247 "%llu\n", (unsigned long long)ins->objectid,
3248 (unsigned long long)ins->offset);
f5947066
CM		 3249 BUG();
3250 }
925baedd 3251out:
e6dcd2dc
CM		 3252 return ret;
3253}
3254
3255int btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
31840ae1 3256 struct btrfs_root *root, u64 parent,
e6dcd2dc 3257 u64 root_objectid, u64 ref_generation,
3bb1a1bc 3258 u64 owner, struct btrfs_key *ins)
e6dcd2dc
CM		 3259{
3260 int ret;
1c2308f8
CM		 3261
3262 if (root_objectid == BTRFS_TREE_LOG_OBJECTID)
3263 return 0;
3bb1a1bc
YZ		 3264 ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
3265 ref_generation, owner, ins);
e8569813 3266 update_reserved_extents(root, ins->objectid, ins->offset, 0);
e6dcd2dc
CM		 3267 return ret;
3268}
e02119d5
CM		 3269
3270/*
3271 * this is used by the tree logging recovery code. It records that
3272 * an extent has been allocated and makes sure to clear the free
3273 * space cache bits as well
3274 */
3275int btrfs_alloc_logged_extent(struct btrfs_trans_handle *trans,
31840ae1 3276 struct btrfs_root *root, u64 parent,
e02119d5 3277 u64 root_objectid, u64 ref_generation,
3bb1a1bc 3278 u64 owner, struct btrfs_key *ins)
e02119d5
CM		 3279{
3280 int ret;
3281 struct btrfs_block_group_cache *block_group;
3282
e02119d5 3283 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
ea6a478e 3284 mutex_lock(&block_group->cache_mutex);
e02119d5 3285 cache_block_group(root, block_group);
ea6a478e 3286 mutex_unlock(&block_group->cache_mutex);
e02119d5 3287
ea6a478e
JB		 3288 ret = btrfs_remove_free_space(block_group, ins->objectid,
3289 ins->offset);
0f9dd46c 3290 BUG_ON(ret);
d2fb3437 3291 put_block_group(block_group);
3bb1a1bc
YZ		 3292 ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
3293 ref_generation, owner, ins);
e02119d5
CM		 3294 return ret;
3295}
3296
e6dcd2dc
CM		 3297/*
3298 * finds a free extent and does all the dirty work required for allocation
3299 * returns the key for the extent through ins, and a tree buffer for
3300 * the first block of the extent through buf.
3301 *
3302 * returns 0 if everything worked, non-zero otherwise.
3303 */
3304int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
3305 struct btrfs_root *root,
31840ae1 3306 u64 num_bytes, u64 parent, u64 min_alloc_size,
e6dcd2dc 3307 u64 root_objectid, u64 ref_generation,
3bb1a1bc 3308 u64 owner_objectid, u64 empty_size, u64 hint_byte,
e6dcd2dc
CM		 3309 u64 search_end, struct btrfs_key *ins, u64 data)
3310{
3311 int ret;
3312
e6dcd2dc
CM		 3313 ret = __btrfs_reserve_extent(trans, root, num_bytes,
3314 min_alloc_size, empty_size, hint_byte,
3315 search_end, ins, data);
3316 BUG_ON(ret);
d00aff00 3317 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
31840ae1
ZY		 3318 ret = __btrfs_alloc_reserved_extent(trans, root, parent,
3319 root_objectid, ref_generation,
3bb1a1bc 3320 owner_objectid, ins);
d00aff00 3321 BUG_ON(ret);
e6dcd2dc 3322
e8569813
ZY		 3323 } else {
3324 update_reserved_extents(root, ins->objectid, ins->offset, 1);
d00aff00 3325 }
925baedd 3326 return ret;
fec577fb 3327}
65b51a00
CM		 3328
3329struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
3330 struct btrfs_root *root,
3331 u64 bytenr, u32 blocksize)
3332{
3333 struct extent_buffer *buf;
3334
3335 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
3336 if (!buf)
3337 return ERR_PTR(-ENOMEM);
3338 btrfs_set_header_generation(buf, trans->transid);
3339 btrfs_tree_lock(buf);
3340 clean_tree_block(trans, root, buf);
3341 btrfs_set_buffer_uptodate(buf);
d0c803c4
CM		 3342 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
3343 set_extent_dirty(&root->dirty_log_pages, buf->start,
3344 buf->start + buf->len - 1, GFP_NOFS);
3345 } else {
3346 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
65b51a00 3347 buf->start + buf->len - 1, GFP_NOFS);
d0c803c4 3348 }
65b51a00
CM		 3349 trans->blocks_used++;
3350 return buf;
3351}
3352
fec577fb
CM		 3353/*
3354 * helper function to allocate a block for a given tree
3355 * returns the tree buffer or NULL.
3356 */
5f39d397 3357struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
7bb86316 3358 struct btrfs_root *root,
31840ae1 3359 u32 blocksize, u64 parent,
7bb86316
CM		 3360 u64 root_objectid,
3361 u64 ref_generation,
7bb86316
CM		 3362 int level,
3363 u64 hint,
5f39d397 3364 u64 empty_size)
fec577fb 3365{
e2fa7227 3366 struct btrfs_key ins;
fec577fb 3367 int ret;
5f39d397 3368 struct extent_buffer *buf;
fec577fb 3369
31840ae1 3370 ret = btrfs_alloc_extent(trans, root, blocksize, parent, blocksize,
3bb1a1bc 3371 root_objectid, ref_generation, level,
31840ae1 3372 empty_size, hint, (u64)-1, &ins, 0);
fec577fb 3373 if (ret) {
54aa1f4d
CM		 3374 BUG_ON(ret > 0);
3375 return ERR_PTR(ret);
fec577fb 3376 }
55c69072 3377
65b51a00 3378 buf = btrfs_init_new_buffer(trans, root, ins.objectid, blocksize);
fec577fb
CM		 3379 return buf;
3380}
a28ec197 3381
e02119d5
CM		 3382int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans,
3383 struct btrfs_root *root, struct extent_buffer *leaf)
6407bf6d 3384{
7bb86316
CM		 3385 u64 leaf_owner;
3386 u64 leaf_generation;
5f39d397 3387 struct btrfs_key key;
6407bf6d
CM		 3388 struct btrfs_file_extent_item *fi;
3389 int i;
3390 int nritems;
3391 int ret;
3392
5f39d397
CM		 3393 BUG_ON(!btrfs_is_leaf(leaf));
3394 nritems = btrfs_header_nritems(leaf);
7bb86316
CM		 3395 leaf_owner = btrfs_header_owner(leaf);
3396 leaf_generation = btrfs_header_generation(leaf);
3397
6407bf6d 3398 for (i = 0; i < nritems; i++) {
db94535d 3399 u64 disk_bytenr;
e34a5b4f 3400 cond_resched();
5f39d397
CM		 3401
3402 btrfs_item_key_to_cpu(leaf, &key, i);
3403 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
6407bf6d
CM		 3404 continue;
3405 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
5f39d397
CM		 3406 if (btrfs_file_extent_type(leaf, fi) ==
3407 BTRFS_FILE_EXTENT_INLINE)
236454df 3408 continue;
6407bf6d
CM		 3409 /*
3410 * FIXME make sure to insert a trans record that
3411 * repeats the snapshot del on crash
3412 */
db94535d
CM		 3413 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
3414 if (disk_bytenr == 0)
3a686375 3415 continue;
4a096752 3416
925baedd 3417 ret = __btrfs_free_extent(trans, root, disk_bytenr,
7bb86316 3418 btrfs_file_extent_disk_num_bytes(leaf, fi),
31840ae1 3419 leaf->start, leaf_owner, leaf_generation,
3bb1a1bc 3420 key.objectid, 0);
31840ae1 3421 BUG_ON(ret);
2dd3e67b
CM		 3422
3423 atomic_inc(&root->fs_info->throttle_gen);
3424 wake_up(&root->fs_info->transaction_throttle);
3425 cond_resched();
6407bf6d
CM		 3426 }
3427 return 0;
3428}
3429
d397712b 3430static noinline int cache_drop_leaf_ref(struct btrfs_trans_handle *trans,
e02119d5
CM		 3431 struct btrfs_root *root,
3432 struct btrfs_leaf_ref *ref)
31153d81
YZ		 3433{
3434 int i;
3435 int ret;
3436 struct btrfs_extent_info *info = ref->extents;
3437
31153d81 3438 for (i = 0; i < ref->nritems; i++) {
31840ae1
ZY		 3439 ret = __btrfs_free_extent(trans, root, info->bytenr,
3440 info->num_bytes, ref->bytenr,
3441 ref->owner, ref->generation,
3bb1a1bc 3442 info->objectid, 0);
2dd3e67b
CM		 3443
3444 atomic_inc(&root->fs_info->throttle_gen);
3445 wake_up(&root->fs_info->transaction_throttle);
3446 cond_resched();
3447
31153d81
YZ		 3448 BUG_ON(ret);
3449 info++;
3450 }
31153d81
YZ		 3451
3452 return 0;
3453}
3454
d397712b
CM		 3455static int drop_snap_lookup_refcount(struct btrfs_root *root, u64 start,
3456 u64 len, u32 *refs)
333db94c 3457{
017e5369 3458 int ret;
f87f057b 3459
31840ae1 3460 ret = btrfs_lookup_extent_ref(NULL, root, start, len, refs);
f87f057b
CM		 3461 BUG_ON(ret);
3462
d397712b 3463#if 0 /* some debugging code in case we see problems here */
f87f057b
CM		 3464 /* if the refs count is one, it won't get increased again. But
3465 * if the ref count is > 1, someone may be decreasing it at
3466 * the same time we are.
3467 */
3468 if (*refs != 1) {
3469 struct extent_buffer *eb = NULL;
3470 eb = btrfs_find_create_tree_block(root, start, len);
3471 if (eb)
3472 btrfs_tree_lock(eb);
3473
3474 mutex_lock(&root->fs_info->alloc_mutex);
3475 ret = lookup_extent_ref(NULL, root, start, len, refs);
3476 BUG_ON(ret);
3477 mutex_unlock(&root->fs_info->alloc_mutex);
3478
3479 if (eb) {
3480 btrfs_tree_unlock(eb);
3481 free_extent_buffer(eb);
3482 }
3483 if (*refs == 1) {
d397712b
CM		 3484 printk(KERN_ERR "btrfs block %llu went down to one "
3485 "during drop_snap\n", (unsigned long long)start);
f87f057b
CM		 3486 }
3487
3488 }
3489#endif
3490
e7a84565 3491 cond_resched();
017e5369 3492 return ret;
333db94c
CM		 3493}
3494
9aca1d51
CM		 3495/*
3496 * helper function for drop_snapshot, this walks down the tree dropping ref
3497 * counts as it goes.
3498 */
d397712b 3499static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
98ed5174
CM		 3500 struct btrfs_root *root,
3501 struct btrfs_path *path, int *level)
20524f02 3502{
7bb86316
CM		 3503 u64 root_owner;
3504 u64 root_gen;
3505 u64 bytenr;
ca7a79ad 3506 u64 ptr_gen;
5f39d397
CM		 3507 struct extent_buffer *next;
3508 struct extent_buffer *cur;
7bb86316 3509 struct extent_buffer *parent;
31153d81 3510 struct btrfs_leaf_ref *ref;
db94535d 3511 u32 blocksize;
20524f02
CM		 3512 int ret;
3513 u32 refs;
3514
5caf2a00
CM		 3515 WARN_ON(*level < 0);
3516 WARN_ON(*level >= BTRFS_MAX_LEVEL);
333db94c 3517 ret = drop_snap_lookup_refcount(root, path->nodes[*level]->start,
db94535d 3518 path->nodes[*level]->len, &refs);
20524f02
CM		 3519 BUG_ON(ret);
3520 if (refs > 1)
3521 goto out;
e011599b 3522
9aca1d51
CM		 3523 /*
3524 * walk down to the last node level and free all the leaves
3525 */
d397712b 3526 while (*level >= 0) {
5caf2a00
CM		 3527 WARN_ON(*level < 0);
3528 WARN_ON(*level >= BTRFS_MAX_LEVEL);
20524f02 3529 cur = path->nodes[*level];
e011599b 3530
5f39d397 3531 if (btrfs_header_level(cur) != *level)
2c90e5d6 3532 WARN_ON(1);
e011599b 3533
7518a238 3534 if (path->slots[*level] >=
5f39d397 3535 btrfs_header_nritems(cur))
20524f02 3536 break;
6407bf6d 3537 if (*level == 0) {
e02119d5 3538 ret = btrfs_drop_leaf_ref(trans, root, cur);
6407bf6d
CM		 3539 BUG_ON(ret);
3540 break;
3541 }
db94535d 3542 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
ca7a79ad 3543 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
db94535d 3544 blocksize = btrfs_level_size(root, *level - 1);
925baedd 3545
333db94c 3546 ret = drop_snap_lookup_refcount(root, bytenr, blocksize, &refs);
6407bf6d
CM		 3547 BUG_ON(ret);
3548 if (refs != 1) {
7bb86316
CM		 3549 parent = path->nodes[*level];
3550 root_owner = btrfs_header_owner(parent);
3551 root_gen = btrfs_header_generation(parent);
20524f02 3552 path->slots[*level]++;
f87f057b 3553
925baedd 3554 ret = __btrfs_free_extent(trans, root, bytenr,
31840ae1 3555 blocksize, parent->start,
3bb1a1bc
YZ		 3556 root_owner, root_gen,
3557 *level - 1, 1);
20524f02 3558 BUG_ON(ret);
18e35e0a
CM		 3559
3560 atomic_inc(&root->fs_info->throttle_gen);
3561 wake_up(&root->fs_info->transaction_throttle);
2dd3e67b 3562 cond_resched();
18e35e0a 3563
20524f02
CM		 3564 continue;
3565 }
f87f057b
CM		 3566 /*
3567 * at this point, we have a single ref, and since the
3568 * only place referencing this extent is a dead root
3569 * the reference count should never go higher.
3570 * So, we don't need to check it again
3571 */
31153d81 3572 if (*level == 1) {
017e5369 3573 ref = btrfs_lookup_leaf_ref(root, bytenr);
1a40e23b
ZY		 3574 if (ref && ref->generation != ptr_gen) {
3575 btrfs_free_leaf_ref(root, ref);
3576 ref = NULL;
3577 }
31153d81 3578 if (ref) {
e02119d5 3579 ret = cache_drop_leaf_ref(trans, root, ref);
31153d81
YZ		 3580 BUG_ON(ret);
3581 btrfs_remove_leaf_ref(root, ref);
bcc63abb 3582 btrfs_free_leaf_ref(root, ref);
31153d81
YZ		 3583 *level = 0;
3584 break;
3585 }
3586 }
db94535d 3587 next = btrfs_find_tree_block(root, bytenr, blocksize);
1259ab75 3588 if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
5f39d397 3589 free_extent_buffer(next);
333db94c 3590
ca7a79ad
CM		 3591 next = read_tree_block(root, bytenr, blocksize,
3592 ptr_gen);
e7a84565 3593 cond_resched();
f87f057b
CM		 3594#if 0
3595 /*
3596 * this is a debugging check and can go away
3597 * the ref should never go all the way down to 1
3598 * at this point
3599 */
e6dcd2dc
CM		 3600 ret = lookup_extent_ref(NULL, root, bytenr, blocksize,
3601 &refs);
e9d0b13b 3602 BUG_ON(ret);
f87f057b
CM		 3603 WARN_ON(refs != 1);
3604#endif
e9d0b13b 3605 }
5caf2a00 3606 WARN_ON(*level <= 0);
83e15a28 3607 if (path->nodes[*level-1])
5f39d397 3608 free_extent_buffer(path->nodes[*level-1]);
20524f02 3609 path->nodes[*level-1] = next;
5f39d397 3610 *level = btrfs_header_level(next);
20524f02 3611 path->slots[*level] = 0;
2dd3e67b 3612 cond_resched();
20524f02
CM		 3613 }
3614out:
5caf2a00
CM		 3615 WARN_ON(*level < 0);
3616 WARN_ON(*level >= BTRFS_MAX_LEVEL);
7bb86316
CM		 3617
3618 if (path->nodes[*level] == root->node) {
7bb86316 3619 parent = path->nodes[*level];
31153d81 3620 bytenr = path->nodes[*level]->start;
7bb86316
CM		 3621 } else {
3622 parent = path->nodes[*level + 1];
31153d81 3623 bytenr = btrfs_node_blockptr(parent, path->slots[*level + 1]);
7bb86316
CM		 3624 }
3625
31153d81
YZ		 3626 blocksize = btrfs_level_size(root, *level);
3627 root_owner = btrfs_header_owner(parent);
7bb86316 3628 root_gen = btrfs_header_generation(parent);
31153d81
YZ		 3629
3630 ret = __btrfs_free_extent(trans, root, bytenr, blocksize,
31840ae1 3631 parent->start, root_owner, root_gen,
3bb1a1bc 3632 *level, 1);
5f39d397 3633 free_extent_buffer(path->nodes[*level]);
20524f02
CM		 3634 path->nodes[*level] = NULL;
3635 *level += 1;
3636 BUG_ON(ret);
f87f057b 3637
e7a84565 3638 cond_resched();
20524f02
CM		 3639 return 0;
3640}
3641
f82d02d9
YZ		 3642/*
3643 * helper function for drop_subtree, this function is similar to
3644 * walk_down_tree. The main difference is that it checks reference
3645 * counts while tree blocks are locked.
3646 */
d397712b 3647static noinline int walk_down_subtree(struct btrfs_trans_handle *trans,
f82d02d9
YZ		 3648 struct btrfs_root *root,
3649 struct btrfs_path *path, int *level)
3650{
3651 struct extent_buffer *next;
3652 struct extent_buffer *cur;
3653 struct extent_buffer *parent;
3654 u64 bytenr;
3655 u64 ptr_gen;
3656 u32 blocksize;
3657 u32 refs;
3658 int ret;
3659
3660 cur = path->nodes[*level];
3661 ret = btrfs_lookup_extent_ref(trans, root, cur->start, cur->len,
3662 &refs);
3663 BUG_ON(ret);
3664 if (refs > 1)
3665 goto out;
3666
3667 while (*level >= 0) {
3668 cur = path->nodes[*level];
3669 if (*level == 0) {
3670 ret = btrfs_drop_leaf_ref(trans, root, cur);
3671 BUG_ON(ret);
3672 clean_tree_block(trans, root, cur);
3673 break;
3674 }
3675 if (path->slots[*level] >= btrfs_header_nritems(cur)) {
3676 clean_tree_block(trans, root, cur);
3677 break;
3678 }
3679
3680 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
3681 blocksize = btrfs_level_size(root, *level - 1);
3682 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
3683
3684 next = read_tree_block(root, bytenr, blocksize, ptr_gen);
3685 btrfs_tree_lock(next);
3686
3687 ret = btrfs_lookup_extent_ref(trans, root, bytenr, blocksize,
3688 &refs);
3689 BUG_ON(ret);
3690 if (refs > 1) {
3691 parent = path->nodes[*level];
3692 ret = btrfs_free_extent(trans, root, bytenr,
3693 blocksize, parent->start,
3694 btrfs_header_owner(parent),
3695 btrfs_header_generation(parent),
3696 *level - 1, 1);
3697 BUG_ON(ret);
3698 path->slots[*level]++;
3699 btrfs_tree_unlock(next);
3700 free_extent_buffer(next);
3701 continue;
3702 }
3703
3704 *level = btrfs_header_level(next);
3705 path->nodes[*level] = next;
3706 path->slots[*level] = 0;
3707 path->locks[*level] = 1;
3708 cond_resched();
3709 }
3710out:
3711 parent = path->nodes[*level + 1];
3712 bytenr = path->nodes[*level]->start;
3713 blocksize = path->nodes[*level]->len;
3714
3715 ret = btrfs_free_extent(trans, root, bytenr, blocksize,
3716 parent->start, btrfs_header_owner(parent),
3717 btrfs_header_generation(parent), *level, 1);
3718 BUG_ON(ret);
3719
3720 if (path->locks[*level]) {
3721 btrfs_tree_unlock(path->nodes[*level]);
3722 path->locks[*level] = 0;
3723 }
3724 free_extent_buffer(path->nodes[*level]);
3725 path->nodes[*level] = NULL;
3726 *level += 1;
3727 cond_resched();
3728 return 0;
3729}
3730
9aca1d51
CM		 3731/*
3732 * helper for dropping snapshots. This walks back up the tree in the path
3733 * to find the first node higher up where we haven't yet gone through
3734 * all the slots
3735 */
d397712b 3736static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
98ed5174 3737 struct btrfs_root *root,
f82d02d9
YZ		 3738 struct btrfs_path *path,
3739 int *level, int max_level)
20524f02 3740{
7bb86316
CM		 3741 u64 root_owner;
3742 u64 root_gen;
3743 struct btrfs_root_item *root_item = &root->root_item;
20524f02
CM		 3744 int i;
3745 int slot;
3746 int ret;
9f3a7427 3747
f82d02d9 3748 for (i = *level; i < max_level && path->nodes[i]; i++) {
20524f02 3749 slot = path->slots[i];
5f39d397
CM		 3750 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
3751 struct extent_buffer *node;
3752 struct btrfs_disk_key disk_key;
3753 node = path->nodes[i];
20524f02
CM		 3754 path->slots[i]++;
3755 *level = i;
9f3a7427 3756 WARN_ON(*level == 0);
5f39d397 3757 btrfs_node_key(node, &disk_key, path->slots[i]);
9f3a7427 3758 memcpy(&root_item->drop_progress,
5f39d397 3759 &disk_key, sizeof(disk_key));
9f3a7427 3760 root_item->drop_level = i;
20524f02
CM		 3761 return 0;
3762 } else {
31840ae1
ZY		 3763 struct extent_buffer *parent;
3764 if (path->nodes[*level] == root->node)
3765 parent = path->nodes[*level];
3766 else
3767 parent = path->nodes[*level + 1];
3768
3769 root_owner = btrfs_header_owner(parent);
3770 root_gen = btrfs_header_generation(parent);
f82d02d9
YZ		 3771
3772 clean_tree_block(trans, root, path->nodes[*level]);
e089f05c 3773 ret = btrfs_free_extent(trans, root,
db94535d 3774 path->nodes[*level]->start,
7bb86316 3775 path->nodes[*level]->len,
3bb1a1bc
YZ		 3776 parent->start, root_owner,
3777 root_gen, *level, 1);
6407bf6d 3778 BUG_ON(ret);
f82d02d9
YZ		 3779 if (path->locks[*level]) {
3780 btrfs_tree_unlock(path->nodes[*level]);
3781 path->locks[*level] = 0;
3782 }
5f39d397 3783 free_extent_buffer(path->nodes[*level]);
83e15a28 3784 path->nodes[*level] = NULL;
20524f02 3785 *level = i + 1;
20524f02
CM		 3786 }
3787 }
3788 return 1;
3789}
3790
9aca1d51
CM		 3791/*
3792 * drop the reference count on the tree rooted at 'snap'. This traverses
3793 * the tree freeing any blocks that have a ref count of zero after being
3794 * decremented.
3795 */
e089f05c 3796int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
9f3a7427 3797 *root)
20524f02 3798{
3768f368 3799 int ret = 0;
9aca1d51 3800 int wret;
20524f02 3801 int level;
5caf2a00 3802 struct btrfs_path *path;
20524f02
CM		 3803 int i;
3804 int orig_level;
9f3a7427 3805 struct btrfs_root_item *root_item = &root->root_item;
20524f02 3806
a2135011 3807 WARN_ON(!mutex_is_locked(&root->fs_info->drop_mutex));
5caf2a00
CM		 3808 path = btrfs_alloc_path();
3809 BUG_ON(!path);
20524f02 3810
5f39d397 3811 level = btrfs_header_level(root->node);
20524f02 3812 orig_level = level;
9f3a7427
CM		 3813 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
3814 path->nodes[level] = root->node;
f510cfec 3815 extent_buffer_get(root->node);
9f3a7427
CM		 3816 path->slots[level] = 0;
3817 } else {
3818 struct btrfs_key key;
5f39d397
CM		 3819 struct btrfs_disk_key found_key;
3820 struct extent_buffer *node;
6702ed49 3821
9f3a7427 3822 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
6702ed49
CM		 3823 level = root_item->drop_level;
3824 path->lowest_level = level;
9f3a7427 3825 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6702ed49 3826 if (wret < 0) {
9f3a7427
CM		 3827 ret = wret;
3828 goto out;
3829 }
5f39d397
CM		 3830 node = path->nodes[level];
3831 btrfs_node_key(node, &found_key, path->slots[level]);
3832 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
3833 sizeof(found_key)));
7d9eb12c
CM		 3834 /*
3835 * unlock our path, this is safe because only this
3836 * function is allowed to delete this snapshot
3837 */
925baedd
CM		 3838 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
3839 if (path->nodes[i] && path->locks[i]) {
3840 path->locks[i] = 0;
3841 btrfs_tree_unlock(path->nodes[i]);
3842 }
3843 }
9f3a7427 3844 }
d397712b 3845 while (1) {
5caf2a00 3846 wret = walk_down_tree(trans, root, path, &level);
9aca1d51 3847 if (wret > 0)
20524f02 3848 break;
9aca1d51
CM		 3849 if (wret < 0)
3850 ret = wret;
3851
f82d02d9
YZ		 3852 wret = walk_up_tree(trans, root, path, &level,
3853 BTRFS_MAX_LEVEL);
9aca1d51 3854 if (wret > 0)
20524f02 3855 break;
9aca1d51
CM		 3856 if (wret < 0)
3857 ret = wret;
e7a84565
CM		 3858 if (trans->transaction->in_commit) {
3859 ret = -EAGAIN;
3860 break;
3861 }
18e35e0a 3862 atomic_inc(&root->fs_info->throttle_gen);
017e5369 3863 wake_up(&root->fs_info->transaction_throttle);
20524f02 3864 }
83e15a28 3865 for (i = 0; i <= orig_level; i++) {
5caf2a00 3866 if (path->nodes[i]) {
5f39d397 3867 free_extent_buffer(path->nodes[i]);
0f82731f 3868 path->nodes[i] = NULL;
83e15a28 3869 }
20524f02 3870 }
9f3a7427 3871out:
5caf2a00 3872 btrfs_free_path(path);
9aca1d51 3873 return ret;
20524f02 3874}
9078a3e1 3875
f82d02d9
YZ		 3876int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
3877 struct btrfs_root *root,
3878 struct extent_buffer *node,
3879 struct extent_buffer *parent)
3880{
3881 struct btrfs_path *path;
3882 int level;
3883 int parent_level;
3884 int ret = 0;
3885 int wret;
3886
3887 path = btrfs_alloc_path();
3888 BUG_ON(!path);
3889
3890 BUG_ON(!btrfs_tree_locked(parent));
3891 parent_level = btrfs_header_level(parent);
3892 extent_buffer_get(parent);
3893 path->nodes[parent_level] = parent;
3894 path->slots[parent_level] = btrfs_header_nritems(parent);
3895
3896 BUG_ON(!btrfs_tree_locked(node));
3897 level = btrfs_header_level(node);
3898 extent_buffer_get(node);
3899 path->nodes[level] = node;
3900 path->slots[level] = 0;
3901
3902 while (1) {
3903 wret = walk_down_subtree(trans, root, path, &level);
3904 if (wret < 0)
3905 ret = wret;
3906 if (wret != 0)
3907 break;
3908
3909 wret = walk_up_tree(trans, root, path, &level, parent_level);
3910 if (wret < 0)
3911 ret = wret;
3912 if (wret != 0)
3913 break;
3914 }
3915
3916 btrfs_free_path(path);
3917 return ret;
3918}
3919
8e7bf94f
CM		 3920static unsigned long calc_ra(unsigned long start, unsigned long last,
3921 unsigned long nr)
3922{
3923 return min(last, start + nr - 1);
3924}
3925
d397712b 3926static noinline int relocate_inode_pages(struct inode *inode, u64 start,
98ed5174 3927 u64 len)
edbd8d4e
CM		 3928{
3929 u64 page_start;
3930 u64 page_end;
1a40e23b 3931 unsigned long first_index;
edbd8d4e 3932 unsigned long last_index;
edbd8d4e
CM		 3933 unsigned long i;
3934 struct page *page;
d1310b2e 3935 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4313b399 3936 struct file_ra_state *ra;
3eaa2885 3937 struct btrfs_ordered_extent *ordered;
1a40e23b
ZY		 3938 unsigned int total_read = 0;
3939 unsigned int total_dirty = 0;
3940 int ret = 0;
4313b399
CM		 3941
3942 ra = kzalloc(sizeof(*ra), GFP_NOFS);
edbd8d4e
CM		 3943
3944 mutex_lock(&inode->i_mutex);
1a40e23b 3945 first_index = start >> PAGE_CACHE_SHIFT;
edbd8d4e
CM		 3946 last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
3947
1a40e23b
ZY		 3948 /* make sure the dirty trick played by the caller work */
3949 ret = invalidate_inode_pages2_range(inode->i_mapping,
3950 first_index, last_index);
3951 if (ret)
3952 goto out_unlock;
8e7bf94f 3953
4313b399 3954 file_ra_state_init(ra, inode->i_mapping);
edbd8d4e 3955
1a40e23b
ZY		 3956 for (i = first_index ; i <= last_index; i++) {
3957 if (total_read % ra->ra_pages == 0) {
8e7bf94f 3958 btrfs_force_ra(inode->i_mapping, ra, NULL, i,
1a40e23b 3959 calc_ra(i, last_index, ra->ra_pages));
8e7bf94f
CM		 3960 }
3961 total_read++;
3eaa2885
CM		 3962again:
3963 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
1a40e23b 3964 BUG_ON(1);
edbd8d4e 3965 page = grab_cache_page(inode->i_mapping, i);
a061fc8d 3966 if (!page) {
1a40e23b 3967 ret = -ENOMEM;
edbd8d4e 3968 goto out_unlock;
a061fc8d 3969 }
edbd8d4e
CM		 3970 if (!PageUptodate(page)) {
3971 btrfs_readpage(NULL, page);
3972 lock_page(page);
3973 if (!PageUptodate(page)) {
3974 unlock_page(page);
3975 page_cache_release(page);
1a40e23b 3976 ret = -EIO;
edbd8d4e
CM		 3977 goto out_unlock;
3978 }
3979 }
ec44a35c 3980 wait_on_page_writeback(page);
3eaa2885 3981
edbd8d4e
CM		 3982 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
3983 page_end = page_start + PAGE_CACHE_SIZE - 1;
d1310b2e 3984 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
edbd8d4e 3985
3eaa2885
CM		 3986 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3987 if (ordered) {
3988 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3989 unlock_page(page);
3990 page_cache_release(page);
3991 btrfs_start_ordered_extent(inode, ordered, 1);
3992 btrfs_put_ordered_extent(ordered);
3993 goto again;
3994 }
3995 set_page_extent_mapped(page);
3996
1a40e23b
ZY		 3997 if (i == first_index)
3998 set_extent_bits(io_tree, page_start, page_end,
3999 EXTENT_BOUNDARY, GFP_NOFS);
1f80e4db 4000 btrfs_set_extent_delalloc(inode, page_start, page_end);
1a40e23b 4001
a061fc8d 4002 set_page_dirty(page);
1a40e23b 4003 total_dirty++;
edbd8d4e 4004
d1310b2e 4005 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
edbd8d4e
CM		 4006 unlock_page(page);
4007 page_cache_release(page);
4008 }
4009
4010out_unlock:
ec44a35c 4011 kfree(ra);
edbd8d4e 4012 mutex_unlock(&inode->i_mutex);
1a40e23b
ZY		 4013 balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
4014 return ret;
edbd8d4e
CM		 4015}
4016
d397712b 4017static noinline int relocate_data_extent(struct inode *reloc_inode,
1a40e23b
ZY		 4018 struct btrfs_key *extent_key,
4019 u64 offset)
4020{
4021 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
4022 struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
4023 struct extent_map *em;
6643558d
YZ		 4024 u64 start = extent_key->objectid - offset;
4025 u64 end = start + extent_key->offset - 1;
bf4ef679 4026
1a40e23b
ZY		 4027 em = alloc_extent_map(GFP_NOFS);
4028 BUG_ON(!em || IS_ERR(em));
bf4ef679 4029
6643558d 4030 em->start = start;
1a40e23b 4031 em->len = extent_key->offset;
c8b97818 4032 em->block_len = extent_key->offset;
1a40e23b
ZY		 4033 em->block_start = extent_key->objectid;
4034 em->bdev = root->fs_info->fs_devices->latest_bdev;
4035 set_bit(EXTENT_FLAG_PINNED, &em->flags);
4036
4037 /* setup extent map to cheat btrfs_readpage */
6643558d 4038 lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
1a40e23b
ZY		 4039 while (1) {
4040 int ret;
4041 spin_lock(&em_tree->lock);
4042 ret = add_extent_mapping(em_tree, em);
4043 spin_unlock(&em_tree->lock);
4044 if (ret != -EEXIST) {
4045 free_extent_map(em);
bf4ef679
CM		 4046 break;
4047 }
6643558d 4048 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
bf4ef679 4049 }
6643558d 4050 unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
bf4ef679 4051
6643558d 4052 return relocate_inode_pages(reloc_inode, start, extent_key->offset);
1a40e23b 4053}
edbd8d4e 4054
1a40e23b
ZY		 4055struct btrfs_ref_path {
4056 u64 extent_start;
4057 u64 nodes[BTRFS_MAX_LEVEL];
4058 u64 root_objectid;
4059 u64 root_generation;
4060 u64 owner_objectid;
1a40e23b
ZY		 4061 u32 num_refs;
4062 int lowest_level;
4063 int current_level;
f82d02d9
YZ		 4064 int shared_level;
4065
4066 struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
4067 u64 new_nodes[BTRFS_MAX_LEVEL];
1a40e23b 4068};
7d9eb12c 4069
1a40e23b 4070struct disk_extent {
c8b97818 4071 u64 ram_bytes;
1a40e23b
ZY		 4072 u64 disk_bytenr;
4073 u64 disk_num_bytes;
4074 u64 offset;
4075 u64 num_bytes;
c8b97818
CM		 4076 u8 compression;
4077 u8 encryption;
4078 u16 other_encoding;
1a40e23b 4079};
4313b399 4080
1a40e23b
ZY		 4081static int is_cowonly_root(u64 root_objectid)
4082{
4083 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
4084 root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
4085 root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
4086 root_objectid == BTRFS_DEV_TREE_OBJECTID ||
0403e47e
YZ		 4087 root_objectid == BTRFS_TREE_LOG_OBJECTID ||
4088 root_objectid == BTRFS_CSUM_TREE_OBJECTID)
1a40e23b
ZY		 4089 return 1;
4090 return 0;
4091}
edbd8d4e 4092
d397712b 4093static noinline int __next_ref_path(struct btrfs_trans_handle *trans,
1a40e23b
ZY		 4094 struct btrfs_root *extent_root,
4095 struct btrfs_ref_path *ref_path,
4096 int first_time)
4097{
4098 struct extent_buffer *leaf;
4099 struct btrfs_path *path;
4100 struct btrfs_extent_ref *ref;
4101 struct btrfs_key key;
4102 struct btrfs_key found_key;
4103 u64 bytenr;
4104 u32 nritems;
4105 int level;
4106 int ret = 1;
edbd8d4e 4107
1a40e23b
ZY		 4108 path = btrfs_alloc_path();
4109 if (!path)
4110 return -ENOMEM;
bf4ef679 4111
1a40e23b
ZY		 4112 if (first_time) {
4113 ref_path->lowest_level = -1;
4114 ref_path->current_level = -1;
f82d02d9 4115 ref_path->shared_level = -1;
1a40e23b
ZY		 4116 goto walk_up;
4117 }
4118walk_down:
4119 level = ref_path->current_level - 1;
4120 while (level >= -1) {
4121 u64 parent;
4122 if (level < ref_path->lowest_level)
4123 break;
bf4ef679 4124
d397712b 4125 if (level >= 0)
1a40e23b 4126 bytenr = ref_path->nodes[level];
d397712b 4127 else
1a40e23b 4128 bytenr = ref_path->extent_start;
1a40e23b 4129 BUG_ON(bytenr == 0);
bf4ef679 4130
1a40e23b
ZY		 4131 parent = ref_path->nodes[level + 1];
4132 ref_path->nodes[level + 1] = 0;
4133 ref_path->current_level = level;
4134 BUG_ON(parent == 0);
0ef3e66b 4135
1a40e23b
ZY		 4136 key.objectid = bytenr;
4137 key.offset = parent + 1;
4138 key.type = BTRFS_EXTENT_REF_KEY;
edbd8d4e 4139
1a40e23b
ZY		 4140 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
4141 if (ret < 0)
edbd8d4e 4142 goto out;
1a40e23b 4143 BUG_ON(ret == 0);
7d9eb12c 4144
1a40e23b
ZY		 4145 leaf = path->nodes[0];
4146 nritems = btrfs_header_nritems(leaf);
4147 if (path->slots[0] >= nritems) {
4148 ret = btrfs_next_leaf(extent_root, path);
4149 if (ret < 0)
4150 goto out;
4151 if (ret > 0)
4152 goto next;
4153 leaf = path->nodes[0];
4154 }
0ef3e66b 4155
1a40e23b
ZY		 4156 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4157 if (found_key.objectid == bytenr &&
f82d02d9
YZ		 4158 found_key.type == BTRFS_EXTENT_REF_KEY) {
4159 if (level < ref_path->shared_level)
4160 ref_path->shared_level = level;
1a40e23b 4161 goto found;
f82d02d9 4162 }
1a40e23b
ZY		 4163next:
4164 level--;
4165 btrfs_release_path(extent_root, path);
d899e052 4166 cond_resched();
1a40e23b
ZY		 4167 }
4168 /* reached lowest level */
4169 ret = 1;
4170 goto out;
4171walk_up:
4172 level = ref_path->current_level;
4173 while (level < BTRFS_MAX_LEVEL - 1) {
4174 u64 ref_objectid;
d397712b
CM		 4175
4176 if (level >= 0)
1a40e23b 4177 bytenr = ref_path->nodes[level];
d397712b 4178 else
1a40e23b 4179 bytenr = ref_path->extent_start;
d397712b 4180
1a40e23b 4181 BUG_ON(bytenr == 0);
edbd8d4e 4182
1a40e23b
ZY		 4183 key.objectid = bytenr;
4184 key.offset = 0;
4185 key.type = BTRFS_EXTENT_REF_KEY;
edbd8d4e 4186
1a40e23b
ZY		 4187 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
4188 if (ret < 0)
4189 goto out;
edbd8d4e 4190
1a40e23b
ZY		 4191 leaf = path->nodes[0];
4192 nritems = btrfs_header_nritems(leaf);
4193 if (path->slots[0] >= nritems) {
4194 ret = btrfs_next_leaf(extent_root, path);
4195 if (ret < 0)
4196 goto out;
4197 if (ret > 0) {
4198 /* the extent was freed by someone */
4199 if (ref_path->lowest_level == level)
4200 goto out;
4201 btrfs_release_path(extent_root, path);
4202 goto walk_down;
4203 }
4204 leaf = path->nodes[0];
4205 }
edbd8d4e 4206
1a40e23b
ZY		 4207 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4208 if (found_key.objectid != bytenr ||
4209 found_key.type != BTRFS_EXTENT_REF_KEY) {
4210 /* the extent was freed by someone */
4211 if (ref_path->lowest_level == level) {
4212 ret = 1;
4213 goto out;
4214 }
4215 btrfs_release_path(extent_root, path);
4216 goto walk_down;
4217 }
4218found:
4219 ref = btrfs_item_ptr(leaf, path->slots[0],
4220 struct btrfs_extent_ref);
4221 ref_objectid = btrfs_ref_objectid(leaf, ref);
4222 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4223 if (first_time) {
4224 level = (int)ref_objectid;
4225 BUG_ON(level >= BTRFS_MAX_LEVEL);
4226 ref_path->lowest_level = level;
4227 ref_path->current_level = level;
4228 ref_path->nodes[level] = bytenr;
4229 } else {
4230 WARN_ON(ref_objectid != level);
4231 }
4232 } else {
4233 WARN_ON(level != -1);
4234 }
4235 first_time = 0;
bf4ef679 4236
1a40e23b
ZY		 4237 if (ref_path->lowest_level == level) {
4238 ref_path->owner_objectid = ref_objectid;
1a40e23b
ZY		 4239 ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
4240 }
bf4ef679 4241
7d9eb12c 4242 /*
1a40e23b
ZY		 4243 * the block is tree root or the block isn't in reference
4244 * counted tree.
7d9eb12c 4245 */
1a40e23b
ZY		 4246 if (found_key.objectid == found_key.offset ||
4247 is_cowonly_root(btrfs_ref_root(leaf, ref))) {
4248 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
4249 ref_path->root_generation =
4250 btrfs_ref_generation(leaf, ref);
4251 if (level < 0) {
4252 /* special reference from the tree log */
4253 ref_path->nodes[0] = found_key.offset;
4254 ref_path->current_level = 0;
4255 }
4256 ret = 0;
4257 goto out;
4258 }
7d9eb12c 4259
1a40e23b
ZY		 4260 level++;
4261 BUG_ON(ref_path->nodes[level] != 0);
4262 ref_path->nodes[level] = found_key.offset;
4263 ref_path->current_level = level;
bf4ef679 4264
1a40e23b
ZY		 4265 /*
4266 * the reference was created in the running transaction,
4267 * no need to continue walking up.
4268 */
4269 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
4270 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
4271 ref_path->root_generation =
4272 btrfs_ref_generation(leaf, ref);
4273 ret = 0;
4274 goto out;
7d9eb12c
CM		 4275 }
4276
1a40e23b 4277 btrfs_release_path(extent_root, path);
d899e052 4278 cond_resched();
7d9eb12c 4279 }
1a40e23b
ZY		 4280 /* reached max tree level, but no tree root found. */
4281 BUG();
edbd8d4e 4282out:
1a40e23b
ZY		 4283 btrfs_free_path(path);
4284 return ret;
edbd8d4e
CM		 4285}
4286
1a40e23b
ZY		 4287static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
4288 struct btrfs_root *extent_root,
4289 struct btrfs_ref_path *ref_path,
4290 u64 extent_start)
a061fc8d 4291{
1a40e23b
ZY		 4292 memset(ref_path, 0, sizeof(*ref_path));
4293 ref_path->extent_start = extent_start;
a061fc8d 4294
1a40e23b 4295 return __next_ref_path(trans, extent_root, ref_path, 1);
a061fc8d
CM		 4296}
4297
1a40e23b
ZY		 4298static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
4299 struct btrfs_root *extent_root,
4300 struct btrfs_ref_path *ref_path)
edbd8d4e 4301{
1a40e23b
ZY		 4302 return __next_ref_path(trans, extent_root, ref_path, 0);
4303}
4304
d397712b 4305static noinline int get_new_locations(struct inode *reloc_inode,
1a40e23b
ZY		 4306 struct btrfs_key *extent_key,
4307 u64 offset, int no_fragment,
4308 struct disk_extent **extents,
4309 int *nr_extents)
4310{
4311 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
4312 struct btrfs_path *path;
4313 struct btrfs_file_extent_item *fi;
edbd8d4e 4314 struct extent_buffer *leaf;
1a40e23b
ZY		 4315 struct disk_extent *exts = *extents;
4316 struct btrfs_key found_key;
4317 u64 cur_pos;
4318 u64 last_byte;
edbd8d4e 4319 u32 nritems;
1a40e23b
ZY		 4320 int nr = 0;
4321 int max = *nr_extents;
4322 int ret;
edbd8d4e 4323
1a40e23b
ZY		 4324 WARN_ON(!no_fragment && *extents);
4325 if (!exts) {
4326 max = 1;
4327 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
4328 if (!exts)
4329 return -ENOMEM;
a061fc8d 4330 }
edbd8d4e 4331
1a40e23b
ZY		 4332 path = btrfs_alloc_path();
4333 BUG_ON(!path);
edbd8d4e 4334
1a40e23b
ZY		 4335 cur_pos = extent_key->objectid - offset;
4336 last_byte = extent_key->objectid + extent_key->offset;
4337 ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
4338 cur_pos, 0);
4339 if (ret < 0)
4340 goto out;
4341 if (ret > 0) {
4342 ret = -ENOENT;
4343 goto out;
4344 }
edbd8d4e 4345
1a40e23b 4346 while (1) {
edbd8d4e
CM		 4347 leaf = path->nodes[0];
4348 nritems = btrfs_header_nritems(leaf);
1a40e23b
ZY		 4349 if (path->slots[0] >= nritems) {
4350 ret = btrfs_next_leaf(root, path);
a061fc8d
CM		 4351 if (ret < 0)
4352 goto out;
1a40e23b
ZY		 4353 if (ret > 0)
4354 break;
bf4ef679 4355 leaf = path->nodes[0];
a061fc8d 4356 }
edbd8d4e
CM		 4357
4358 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1a40e23b
ZY		 4359 if (found_key.offset != cur_pos ||
4360 found_key.type != BTRFS_EXTENT_DATA_KEY ||
4361 found_key.objectid != reloc_inode->i_ino)
edbd8d4e
CM		 4362 break;
4363
1a40e23b
ZY		 4364 fi = btrfs_item_ptr(leaf, path->slots[0],
4365 struct btrfs_file_extent_item);
4366 if (btrfs_file_extent_type(leaf, fi) !=
4367 BTRFS_FILE_EXTENT_REG ||
4368 btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
edbd8d4e 4369 break;
1a40e23b
ZY		 4370
4371 if (nr == max) {
4372 struct disk_extent *old = exts;
4373 max *= 2;
4374 exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
4375 memcpy(exts, old, sizeof(*exts) * nr);
4376 if (old != *extents)
4377 kfree(old);
a061fc8d 4378 }
edbd8d4e 4379
1a40e23b
ZY		 4380 exts[nr].disk_bytenr =
4381 btrfs_file_extent_disk_bytenr(leaf, fi);
4382 exts[nr].disk_num_bytes =
4383 btrfs_file_extent_disk_num_bytes(leaf, fi);
4384 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
4385 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
c8b97818
CM		 4386 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
4387 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
4388 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
4389 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
4390 fi);
d899e052
YZ		 4391 BUG_ON(exts[nr].offset > 0);
4392 BUG_ON(exts[nr].compression || exts[nr].encryption);
4393 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
edbd8d4e 4394
1a40e23b
ZY		 4395 cur_pos += exts[nr].num_bytes;
4396 nr++;
4397
4398 if (cur_pos + offset >= last_byte)
4399 break;
4400
4401 if (no_fragment) {
4402 ret = 1;
edbd8d4e 4403 goto out;
1a40e23b
ZY		 4404 }
4405 path->slots[0]++;
4406 }
4407
1f80e4db 4408 BUG_ON(cur_pos + offset > last_byte);
1a40e23b
ZY		 4409 if (cur_pos + offset < last_byte) {
4410 ret = -ENOENT;
4411 goto out;
edbd8d4e
CM		 4412 }
4413 ret = 0;
4414out:
1a40e23b
ZY		 4415 btrfs_free_path(path);
4416 if (ret) {
4417 if (exts != *extents)
4418 kfree(exts);
4419 } else {
4420 *extents = exts;
4421 *nr_extents = nr;
4422 }
4423 return ret;
4424}
4425
d397712b 4426static noinline int replace_one_extent(struct btrfs_trans_handle *trans,
1a40e23b
ZY		 4427 struct btrfs_root *root,
4428 struct btrfs_path *path,
4429 struct btrfs_key *extent_key,
4430 struct btrfs_key *leaf_key,
4431 struct btrfs_ref_path *ref_path,
4432 struct disk_extent *new_extents,
4433 int nr_extents)
4434{
4435 struct extent_buffer *leaf;
4436 struct btrfs_file_extent_item *fi;
4437 struct inode *inode = NULL;
4438 struct btrfs_key key;
4439 u64 lock_start = 0;
4440 u64 lock_end = 0;
4441 u64 num_bytes;
4442 u64 ext_offset;
4443 u64 first_pos;
4444 u32 nritems;
3bb1a1bc 4445 int nr_scaned = 0;
1a40e23b 4446 int extent_locked = 0;
d899e052 4447 int extent_type;
1a40e23b
ZY		 4448 int ret;
4449
3bb1a1bc
YZ		 4450 memcpy(&key, leaf_key, sizeof(key));
4451 first_pos = INT_LIMIT(loff_t) - extent_key->offset;
1a40e23b 4452 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
3bb1a1bc
YZ		 4453 if (key.objectid < ref_path->owner_objectid ||
4454 (key.objectid == ref_path->owner_objectid &&
4455 key.type < BTRFS_EXTENT_DATA_KEY)) {
4456 key.objectid = ref_path->owner_objectid;
4457 key.type = BTRFS_EXTENT_DATA_KEY;
4458 key.offset = 0;
4459 }
1a40e23b
ZY		 4460 }
4461
4462 while (1) {
4463 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
4464 if (ret < 0)
4465 goto out;
4466
4467 leaf = path->nodes[0];
4468 nritems = btrfs_header_nritems(leaf);
4469next:
4470 if (extent_locked && ret > 0) {
4471 /*
4472 * the file extent item was modified by someone
4473 * before the extent got locked.
4474 */
1a40e23b
ZY		 4475 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4476 lock_end, GFP_NOFS);
4477 extent_locked = 0;
4478 }
4479
4480 if (path->slots[0] >= nritems) {
3bb1a1bc 4481 if (++nr_scaned > 2)
1a40e23b
ZY		 4482 break;
4483
4484 BUG_ON(extent_locked);
4485 ret = btrfs_next_leaf(root, path);
4486 if (ret < 0)
4487 goto out;
4488 if (ret > 0)
4489 break;
4490 leaf = path->nodes[0];
4491 nritems = btrfs_header_nritems(leaf);
4492 }
4493
4494 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4495
4496 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
4497 if ((key.objectid > ref_path->owner_objectid) ||
4498 (key.objectid == ref_path->owner_objectid &&
4499 key.type > BTRFS_EXTENT_DATA_KEY) ||
4500 (key.offset >= first_pos + extent_key->offset))
4501 break;
4502 }
4503
4504 if (inode && key.objectid != inode->i_ino) {
4505 BUG_ON(extent_locked);
4506 btrfs_release_path(root, path);
4507 mutex_unlock(&inode->i_mutex);
4508 iput(inode);
4509 inode = NULL;
4510 continue;
4511 }
4512
4513 if (key.type != BTRFS_EXTENT_DATA_KEY) {
4514 path->slots[0]++;
4515 ret = 1;
4516 goto next;
4517 }
4518 fi = btrfs_item_ptr(leaf, path->slots[0],
4519 struct btrfs_file_extent_item);
d899e052
YZ		 4520 extent_type = btrfs_file_extent_type(leaf, fi);
4521 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
4522 extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
1a40e23b
ZY		 4523 (btrfs_file_extent_disk_bytenr(leaf, fi) !=
4524 extent_key->objectid)) {
4525 path->slots[0]++;
4526 ret = 1;
4527 goto next;
4528 }
4529
4530 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4531 ext_offset = btrfs_file_extent_offset(leaf, fi);
4532
4533 if (first_pos > key.offset - ext_offset)
4534 first_pos = key.offset - ext_offset;
4535
4536 if (!extent_locked) {
4537 lock_start = key.offset;
4538 lock_end = lock_start + num_bytes - 1;
4539 } else {
6643558d
YZ		 4540 if (lock_start > key.offset ||
4541 lock_end + 1 < key.offset + num_bytes) {
4542 unlock_extent(&BTRFS_I(inode)->io_tree,
4543 lock_start, lock_end, GFP_NOFS);
4544 extent_locked = 0;
4545 }
1a40e23b
ZY		 4546 }
4547
4548 if (!inode) {
4549 btrfs_release_path(root, path);
4550
4551 inode = btrfs_iget_locked(root->fs_info->sb,
4552 key.objectid, root);
4553 if (inode->i_state & I_NEW) {
4554 BTRFS_I(inode)->root = root;
4555 BTRFS_I(inode)->location.objectid =
4556 key.objectid;
4557 BTRFS_I(inode)->location.type =
4558 BTRFS_INODE_ITEM_KEY;
4559 BTRFS_I(inode)->location.offset = 0;
4560 btrfs_read_locked_inode(inode);
4561 unlock_new_inode(inode);
4562 }
4563 /*
4564 * some code call btrfs_commit_transaction while
4565 * holding the i_mutex, so we can't use mutex_lock
4566 * here.
4567 */
4568 if (is_bad_inode(inode) ||
4569 !mutex_trylock(&inode->i_mutex)) {
4570 iput(inode);
4571 inode = NULL;
4572 key.offset = (u64)-1;
4573 goto skip;
4574 }
4575 }
4576
4577 if (!extent_locked) {
4578 struct btrfs_ordered_extent *ordered;
4579
4580 btrfs_release_path(root, path);
4581
4582 lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4583 lock_end, GFP_NOFS);
4584 ordered = btrfs_lookup_first_ordered_extent(inode,
4585 lock_end);
4586 if (ordered &&
4587 ordered->file_offset <= lock_end &&
4588 ordered->file_offset + ordered->len > lock_start) {
4589 unlock_extent(&BTRFS_I(inode)->io_tree,
4590 lock_start, lock_end, GFP_NOFS);
4591 btrfs_start_ordered_extent(inode, ordered, 1);
4592 btrfs_put_ordered_extent(ordered);
4593 key.offset += num_bytes;
4594 goto skip;
4595 }
4596 if (ordered)
4597 btrfs_put_ordered_extent(ordered);
4598
1a40e23b
ZY		 4599 extent_locked = 1;
4600 continue;
4601 }
4602
4603 if (nr_extents == 1) {
4604 /* update extent pointer in place */
1a40e23b
ZY		 4605 btrfs_set_file_extent_disk_bytenr(leaf, fi,
4606 new_extents[0].disk_bytenr);
4607 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4608 new_extents[0].disk_num_bytes);
1a40e23b
ZY		 4609 btrfs_mark_buffer_dirty(leaf);
4610
4611 btrfs_drop_extent_cache(inode, key.offset,
4612 key.offset + num_bytes - 1, 0);
4613
4614 ret = btrfs_inc_extent_ref(trans, root,
4615 new_extents[0].disk_bytenr,
4616 new_extents[0].disk_num_bytes,
4617 leaf->start,
4618 root->root_key.objectid,
4619 trans->transid,
3bb1a1bc 4620 key.objectid);
1a40e23b
ZY		 4621 BUG_ON(ret);
4622
4623 ret = btrfs_free_extent(trans, root,
4624 extent_key->objectid,
4625 extent_key->offset,
4626 leaf->start,
4627 btrfs_header_owner(leaf),
4628 btrfs_header_generation(leaf),
3bb1a1bc 4629 key.objectid, 0);
1a40e23b
ZY		 4630 BUG_ON(ret);
4631
4632 btrfs_release_path(root, path);
4633 key.offset += num_bytes;
4634 } else {
d899e052
YZ		 4635 BUG_ON(1);
4636#if 0
1a40e23b
ZY		 4637 u64 alloc_hint;
4638 u64 extent_len;
4639 int i;
4640 /*
4641 * drop old extent pointer at first, then insert the
4642 * new pointers one bye one
4643 */
4644 btrfs_release_path(root, path);
4645 ret = btrfs_drop_extents(trans, root, inode, key.offset,
4646 key.offset + num_bytes,
4647 key.offset, &alloc_hint);
4648 BUG_ON(ret);
4649
4650 for (i = 0; i < nr_extents; i++) {
4651 if (ext_offset >= new_extents[i].num_bytes) {
4652 ext_offset -= new_extents[i].num_bytes;
4653 continue;
4654 }
4655 extent_len = min(new_extents[i].num_bytes -
4656 ext_offset, num_bytes);
4657
4658 ret = btrfs_insert_empty_item(trans, root,
4659 path, &key,
4660 sizeof(*fi));
4661 BUG_ON(ret);
4662
4663 leaf = path->nodes[0];
4664 fi = btrfs_item_ptr(leaf, path->slots[0],
4665 struct btrfs_file_extent_item);
4666 btrfs_set_file_extent_generation(leaf, fi,
4667 trans->transid);
4668 btrfs_set_file_extent_type(leaf, fi,
4669 BTRFS_FILE_EXTENT_REG);
4670 btrfs_set_file_extent_disk_bytenr(leaf, fi,
4671 new_extents[i].disk_bytenr);
4672 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4673 new_extents[i].disk_num_bytes);
c8b97818
CM		 4674 btrfs_set_file_extent_ram_bytes(leaf, fi,
4675 new_extents[i].ram_bytes);
4676
4677 btrfs_set_file_extent_compression(leaf, fi,
4678 new_extents[i].compression);
4679 btrfs_set_file_extent_encryption(leaf, fi,
4680 new_extents[i].encryption);
4681 btrfs_set_file_extent_other_encoding(leaf, fi,
4682 new_extents[i].other_encoding);
4683
1a40e23b
ZY		 4684 btrfs_set_file_extent_num_bytes(leaf, fi,
4685 extent_len);
4686 ext_offset += new_extents[i].offset;
4687 btrfs_set_file_extent_offset(leaf, fi,
4688 ext_offset);
4689 btrfs_mark_buffer_dirty(leaf);
4690
4691 btrfs_drop_extent_cache(inode, key.offset,
4692 key.offset + extent_len - 1, 0);
4693
4694 ret = btrfs_inc_extent_ref(trans, root,
4695 new_extents[i].disk_bytenr,
4696 new_extents[i].disk_num_bytes,
4697 leaf->start,
4698 root->root_key.objectid,
3bb1a1bc 4699 trans->transid, key.objectid);
1a40e23b
ZY		 4700 BUG_ON(ret);
4701 btrfs_release_path(root, path);
4702
a76a3cd4 4703 inode_add_bytes(inode, extent_len);
1a40e23b
ZY		 4704
4705 ext_offset = 0;
4706 num_bytes -= extent_len;
4707 key.offset += extent_len;
4708
4709 if (num_bytes == 0)
4710 break;
4711 }
4712 BUG_ON(i >= nr_extents);
d899e052 4713#endif
1a40e23b
ZY		 4714 }
4715
4716 if (extent_locked) {
1a40e23b
ZY		 4717 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4718 lock_end, GFP_NOFS);
4719 extent_locked = 0;
4720 }
4721skip:
4722 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
4723 key.offset >= first_pos + extent_key->offset)
4724 break;
4725
4726 cond_resched();
4727 }
4728 ret = 0;
4729out:
4730 btrfs_release_path(root, path);
4731 if (inode) {
4732 mutex_unlock(&inode->i_mutex);
4733 if (extent_locked) {
1a40e23b
ZY		 4734 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4735 lock_end, GFP_NOFS);
4736 }
4737 iput(inode);
4738 }
4739 return ret;
4740}
4741
1a40e23b
ZY		 4742int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
4743 struct btrfs_root *root,
4744 struct extent_buffer *buf, u64 orig_start)
4745{
4746 int level;
4747 int ret;
4748
4749 BUG_ON(btrfs_header_generation(buf) != trans->transid);
4750 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
4751
4752 level = btrfs_header_level(buf);
4753 if (level == 0) {
4754 struct btrfs_leaf_ref *ref;
4755 struct btrfs_leaf_ref *orig_ref;
4756
4757 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
4758 if (!orig_ref)
4759 return -ENOENT;
4760
4761 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
4762 if (!ref) {
4763 btrfs_free_leaf_ref(root, orig_ref);
4764 return -ENOMEM;
4765 }
4766
4767 ref->nritems = orig_ref->nritems;
4768 memcpy(ref->extents, orig_ref->extents,
4769 sizeof(ref->extents[0]) * ref->nritems);
4770
4771 btrfs_free_leaf_ref(root, orig_ref);
4772
4773 ref->root_gen = trans->transid;
4774 ref->bytenr = buf->start;
4775 ref->owner = btrfs_header_owner(buf);
4776 ref->generation = btrfs_header_generation(buf);
4777 ret = btrfs_add_leaf_ref(root, ref, 0);
4778 WARN_ON(ret);
4779 btrfs_free_leaf_ref(root, ref);
4780 }
4781 return 0;
4782}
4783
d397712b 4784static noinline int invalidate_extent_cache(struct btrfs_root *root,
1a40e23b
ZY		 4785 struct extent_buffer *leaf,
4786 struct btrfs_block_group_cache *group,
4787 struct btrfs_root *target_root)
4788{
4789 struct btrfs_key key;
4790 struct inode *inode = NULL;
4791 struct btrfs_file_extent_item *fi;
4792 u64 num_bytes;
4793 u64 skip_objectid = 0;
4794 u32 nritems;
4795 u32 i;
4796
4797 nritems = btrfs_header_nritems(leaf);
4798 for (i = 0; i < nritems; i++) {
4799 btrfs_item_key_to_cpu(leaf, &key, i);
4800 if (key.objectid == skip_objectid ||
4801 key.type != BTRFS_EXTENT_DATA_KEY)
4802 continue;
4803 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4804 if (btrfs_file_extent_type(leaf, fi) ==
4805 BTRFS_FILE_EXTENT_INLINE)
4806 continue;
4807 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
4808 continue;
4809 if (!inode || inode->i_ino != key.objectid) {
4810 iput(inode);
4811 inode = btrfs_ilookup(target_root->fs_info->sb,
4812 key.objectid, target_root, 1);
4813 }
4814 if (!inode) {
4815 skip_objectid = key.objectid;
4816 continue;
4817 }
4818 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4819
4820 lock_extent(&BTRFS_I(inode)->io_tree, key.offset,
4821 key.offset + num_bytes - 1, GFP_NOFS);
1a40e23b
ZY		 4822 btrfs_drop_extent_cache(inode, key.offset,
4823 key.offset + num_bytes - 1, 1);
1a40e23b
ZY		 4824 unlock_extent(&BTRFS_I(inode)->io_tree, key.offset,
4825 key.offset + num_bytes - 1, GFP_NOFS);
4826 cond_resched();
4827 }
4828 iput(inode);
4829 return 0;
4830}
4831
d397712b 4832static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans,
1a40e23b
ZY		 4833 struct btrfs_root *root,
4834 struct extent_buffer *leaf,
4835 struct btrfs_block_group_cache *group,
4836 struct inode *reloc_inode)
4837{
4838 struct btrfs_key key;
4839 struct btrfs_key extent_key;
4840 struct btrfs_file_extent_item *fi;
4841 struct btrfs_leaf_ref *ref;
4842 struct disk_extent *new_extent;
4843 u64 bytenr;
4844 u64 num_bytes;
4845 u32 nritems;
4846 u32 i;
4847 int ext_index;
4848 int nr_extent;
4849 int ret;
4850
4851 new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
4852 BUG_ON(!new_extent);
4853
4854 ref = btrfs_lookup_leaf_ref(root, leaf->start);
4855 BUG_ON(!ref);
4856
4857 ext_index = -1;
4858 nritems = btrfs_header_nritems(leaf);
4859 for (i = 0; i < nritems; i++) {
4860 btrfs_item_key_to_cpu(leaf, &key, i);
4861 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
4862 continue;
4863 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4864 if (btrfs_file_extent_type(leaf, fi) ==
4865 BTRFS_FILE_EXTENT_INLINE)
4866 continue;
4867 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
4868 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
4869 if (bytenr == 0)
4870 continue;
4871
4872 ext_index++;
4873 if (bytenr >= group->key.objectid + group->key.offset ||
4874 bytenr + num_bytes <= group->key.objectid)
4875 continue;
4876
4877 extent_key.objectid = bytenr;
4878 extent_key.offset = num_bytes;
4879 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
4880 nr_extent = 1;
4881 ret = get_new_locations(reloc_inode, &extent_key,
4882 group->key.objectid, 1,
4883 &new_extent, &nr_extent);
4884 if (ret > 0)
4885 continue;
4886 BUG_ON(ret < 0);
4887
4888 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
4889 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
4890 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
4891 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
4892
1a40e23b
ZY		 4893 btrfs_set_file_extent_disk_bytenr(leaf, fi,
4894 new_extent->disk_bytenr);
4895 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4896 new_extent->disk_num_bytes);
1a40e23b
ZY		 4897 btrfs_mark_buffer_dirty(leaf);
4898
4899 ret = btrfs_inc_extent_ref(trans, root,
4900 new_extent->disk_bytenr,
4901 new_extent->disk_num_bytes,
4902 leaf->start,
4903 root->root_key.objectid,
3bb1a1bc 4904 trans->transid, key.objectid);
1a40e23b
ZY		 4905 BUG_ON(ret);
4906 ret = btrfs_free_extent(trans, root,
4907 bytenr, num_bytes, leaf->start,
4908 btrfs_header_owner(leaf),
4909 btrfs_header_generation(leaf),
3bb1a1bc 4910 key.objectid, 0);
1a40e23b
ZY		 4911 BUG_ON(ret);
4912 cond_resched();
4913 }
4914 kfree(new_extent);
4915 BUG_ON(ext_index + 1 != ref->nritems);
4916 btrfs_free_leaf_ref(root, ref);
4917 return 0;
4918}
4919
f82d02d9
YZ		 4920int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
4921 struct btrfs_root *root)
1a40e23b
ZY		 4922{
4923 struct btrfs_root *reloc_root;
f82d02d9 4924 int ret;
1a40e23b
ZY		 4925
4926 if (root->reloc_root) {
4927 reloc_root = root->reloc_root;
4928 root->reloc_root = NULL;
4929 list_add(&reloc_root->dead_list,
4930 &root->fs_info->dead_reloc_roots);
f82d02d9
YZ		 4931
4932 btrfs_set_root_bytenr(&reloc_root->root_item,
4933 reloc_root->node->start);
4934 btrfs_set_root_level(&root->root_item,
4935 btrfs_header_level(reloc_root->node));
4936 memset(&reloc_root->root_item.drop_progress, 0,
4937 sizeof(struct btrfs_disk_key));
4938 reloc_root->root_item.drop_level = 0;
4939
4940 ret = btrfs_update_root(trans, root->fs_info->tree_root,
4941 &reloc_root->root_key,
4942 &reloc_root->root_item);
4943 BUG_ON(ret);
1a40e23b
ZY		 4944 }
4945 return 0;
4946}
4947
4948int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
4949{
4950 struct btrfs_trans_handle *trans;
4951 struct btrfs_root *reloc_root;
4952 struct btrfs_root *prev_root = NULL;
4953 struct list_head dead_roots;
4954 int ret;
4955 unsigned long nr;
4956
4957 INIT_LIST_HEAD(&dead_roots);
4958 list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
4959
4960 while (!list_empty(&dead_roots)) {
4961 reloc_root = list_entry(dead_roots.prev,
4962 struct btrfs_root, dead_list);
4963 list_del_init(&reloc_root->dead_list);
4964
4965 BUG_ON(reloc_root->commit_root != NULL);
4966 while (1) {
4967 trans = btrfs_join_transaction(root, 1);
4968 BUG_ON(!trans);
4969
4970 mutex_lock(&root->fs_info->drop_mutex);
4971 ret = btrfs_drop_snapshot(trans, reloc_root);
4972 if (ret != -EAGAIN)
4973 break;
4974 mutex_unlock(&root->fs_info->drop_mutex);
4975
4976 nr = trans->blocks_used;
4977 ret = btrfs_end_transaction(trans, root);
4978 BUG_ON(ret);
4979 btrfs_btree_balance_dirty(root, nr);
4980 }
4981
4982 free_extent_buffer(reloc_root->node);
4983
4984 ret = btrfs_del_root(trans, root->fs_info->tree_root,
4985 &reloc_root->root_key);
4986 BUG_ON(ret);
4987 mutex_unlock(&root->fs_info->drop_mutex);
4988
4989 nr = trans->blocks_used;
4990 ret = btrfs_end_transaction(trans, root);
4991 BUG_ON(ret);
4992 btrfs_btree_balance_dirty(root, nr);
4993
4994 kfree(prev_root);
4995 prev_root = reloc_root;
4996 }
4997 if (prev_root) {
4998 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
4999 kfree(prev_root);
5000 }
5001 return 0;
5002}
5003
5004int btrfs_add_dead_reloc_root(struct btrfs_root *root)
5005{
5006 list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
5007 return 0;
5008}
5009
5010int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
5011{
5012 struct btrfs_root *reloc_root;
5013 struct btrfs_trans_handle *trans;
5014 struct btrfs_key location;
5015 int found;
5016 int ret;
5017
5018 mutex_lock(&root->fs_info->tree_reloc_mutex);
5019 ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
5020 BUG_ON(ret);
5021 found = !list_empty(&root->fs_info->dead_reloc_roots);
5022 mutex_unlock(&root->fs_info->tree_reloc_mutex);
5023
5024 if (found) {
5025 trans = btrfs_start_transaction(root, 1);
5026 BUG_ON(!trans);
5027 ret = btrfs_commit_transaction(trans, root);
5028 BUG_ON(ret);
5029 }
5030
5031 location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
5032 location.offset = (u64)-1;
5033 location.type = BTRFS_ROOT_ITEM_KEY;
5034
5035 reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
5036 BUG_ON(!reloc_root);
5037 btrfs_orphan_cleanup(reloc_root);
5038 return 0;
5039}
5040
d397712b 5041static noinline int init_reloc_tree(struct btrfs_trans_handle *trans,
1a40e23b
ZY		 5042 struct btrfs_root *root)
5043{
5044 struct btrfs_root *reloc_root;
5045 struct extent_buffer *eb;
5046 struct btrfs_root_item *root_item;
5047 struct btrfs_key root_key;
5048 int ret;
5049
5050 BUG_ON(!root->ref_cows);
5051 if (root->reloc_root)
5052 return 0;
5053
5054 root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
5055 BUG_ON(!root_item);
5056
5057 ret = btrfs_copy_root(trans, root, root->commit_root,
5058 &eb, BTRFS_TREE_RELOC_OBJECTID);
5059 BUG_ON(ret);
5060
5061 root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
5062 root_key.offset = root->root_key.objectid;
5063 root_key.type = BTRFS_ROOT_ITEM_KEY;
5064
5065 memcpy(root_item, &root->root_item, sizeof(root_item));
5066 btrfs_set_root_refs(root_item, 0);
5067 btrfs_set_root_bytenr(root_item, eb->start);
5068 btrfs_set_root_level(root_item, btrfs_header_level(eb));
84234f3a 5069 btrfs_set_root_generation(root_item, trans->transid);
1a40e23b
ZY		 5070
5071 btrfs_tree_unlock(eb);
5072 free_extent_buffer(eb);
5073
5074 ret = btrfs_insert_root(trans, root->fs_info->tree_root,
5075 &root_key, root_item);
5076 BUG_ON(ret);
5077 kfree(root_item);
5078
5079 reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
5080 &root_key);
5081 BUG_ON(!reloc_root);
5082 reloc_root->last_trans = trans->transid;
5083 reloc_root->commit_root = NULL;
5084 reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
5085
5086 root->reloc_root = reloc_root;
5087 return 0;
5088}
5089
5090/*
5091 * Core function of space balance.
5092 *
5093 * The idea is using reloc trees to relocate tree blocks in reference
f82d02d9
YZ		 5094 * counted roots. There is one reloc tree for each subvol, and all
5095 * reloc trees share same root key objectid. Reloc trees are snapshots
5096 * of the latest committed roots of subvols (root->commit_root).
5097 *
5098 * To relocate a tree block referenced by a subvol, there are two steps.
5099 * COW the block through subvol's reloc tree, then update block pointer
5100 * in the subvol to point to the new block. Since all reloc trees share
5101 * same root key objectid, doing special handing for tree blocks owned
5102 * by them is easy. Once a tree block has been COWed in one reloc tree,
5103 * we can use the resulting new block directly when the same block is
5104 * required to COW again through other reloc trees. By this way, relocated
5105 * tree blocks are shared between reloc trees, so they are also shared
5106 * between subvols.
1a40e23b 5107 */
d397712b 5108static noinline int relocate_one_path(struct btrfs_trans_handle *trans,
1a40e23b
ZY		 5109 struct btrfs_root *root,
5110 struct btrfs_path *path,
5111 struct btrfs_key *first_key,
5112 struct btrfs_ref_path *ref_path,
5113 struct btrfs_block_group_cache *group,
5114 struct inode *reloc_inode)
5115{
5116 struct btrfs_root *reloc_root;
5117 struct extent_buffer *eb = NULL;
5118 struct btrfs_key *keys;
5119 u64 *nodes;
5120 int level;
f82d02d9 5121 int shared_level;
1a40e23b 5122 int lowest_level = 0;
1a40e23b
ZY		 5123 int ret;
5124
5125 if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
5126 lowest_level = ref_path->owner_objectid;
5127
f82d02d9 5128 if (!root->ref_cows) {
1a40e23b
ZY		 5129 path->lowest_level = lowest_level;
5130 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
5131 BUG_ON(ret < 0);
5132 path->lowest_level = 0;
5133 btrfs_release_path(root, path);
5134 return 0;
5135 }
5136
1a40e23b
ZY		 5137 mutex_lock(&root->fs_info->tree_reloc_mutex);
5138 ret = init_reloc_tree(trans, root);
5139 BUG_ON(ret);
5140 reloc_root = root->reloc_root;
5141
f82d02d9
YZ		 5142 shared_level = ref_path->shared_level;
5143 ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
1a40e23b 5144
f82d02d9
YZ		 5145 keys = ref_path->node_keys;
5146 nodes = ref_path->new_nodes;
5147 memset(&keys[shared_level + 1], 0,
5148 sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
5149 memset(&nodes[shared_level + 1], 0,
5150 sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
1a40e23b 5151
f82d02d9
YZ		 5152 if (nodes[lowest_level] == 0) {
5153 path->lowest_level = lowest_level;
5154 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
5155 0, 1);
5156 BUG_ON(ret);
5157 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
5158 eb = path->nodes[level];
5159 if (!eb || eb == reloc_root->node)
5160 break;
5161 nodes[level] = eb->start;
5162 if (level == 0)
5163 btrfs_item_key_to_cpu(eb, &keys[level], 0);
5164 else
5165 btrfs_node_key_to_cpu(eb, &keys[level], 0);
5166 }
2b82032c
YZ		 5167 if (nodes[0] &&
5168 ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
f82d02d9
YZ		 5169 eb = path->nodes[0];
5170 ret = replace_extents_in_leaf(trans, reloc_root, eb,
5171 group, reloc_inode);
5172 BUG_ON(ret);
5173 }
5174 btrfs_release_path(reloc_root, path);
5175 } else {
1a40e23b 5176 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
f82d02d9 5177 lowest_level);
1a40e23b
ZY		 5178 BUG_ON(ret);
5179 }
5180
1a40e23b
ZY		 5181 /*
5182 * replace tree blocks in the fs tree with tree blocks in
5183 * the reloc tree.
5184 */
5185 ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
5186 BUG_ON(ret < 0);
5187
5188 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
f82d02d9
YZ		 5189 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
5190 0, 0);
5191 BUG_ON(ret);
5192 extent_buffer_get(path->nodes[0]);
5193 eb = path->nodes[0];
5194 btrfs_release_path(reloc_root, path);
1a40e23b
ZY		 5195 ret = invalidate_extent_cache(reloc_root, eb, group, root);
5196 BUG_ON(ret);
5197 free_extent_buffer(eb);
5198 }
1a40e23b 5199
f82d02d9 5200 mutex_unlock(&root->fs_info->tree_reloc_mutex);
1a40e23b 5201 path->lowest_level = 0;
1a40e23b
ZY		 5202 return 0;
5203}
5204
d397712b 5205static noinline int relocate_tree_block(struct btrfs_trans_handle *trans,
1a40e23b
ZY		 5206 struct btrfs_root *root,
5207 struct btrfs_path *path,
5208 struct btrfs_key *first_key,
5209 struct btrfs_ref_path *ref_path)
5210{
5211 int ret;
1a40e23b
ZY		 5212
5213 ret = relocate_one_path(trans, root, path, first_key,
5214 ref_path, NULL, NULL);
5215 BUG_ON(ret);
5216
5217 if (root == root->fs_info->extent_root)
5218 btrfs_extent_post_op(trans, root);
1a40e23b
ZY		 5219
5220 return 0;
5221}
5222
d397712b 5223static noinline int del_extent_zero(struct btrfs_trans_handle *trans,
1a40e23b
ZY		 5224 struct btrfs_root *extent_root,
5225 struct btrfs_path *path,
5226 struct btrfs_key *extent_key)
5227{
5228 int ret;
5229
1a40e23b
ZY		 5230 ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
5231 if (ret)
5232 goto out;
5233 ret = btrfs_del_item(trans, extent_root, path);
5234out:
5235 btrfs_release_path(extent_root, path);
1a40e23b
ZY		 5236 return ret;
5237}
5238
d397712b 5239static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info,
1a40e23b
ZY		 5240 struct btrfs_ref_path *ref_path)
5241{
5242 struct btrfs_key root_key;
5243
5244 root_key.objectid = ref_path->root_objectid;
5245 root_key.type = BTRFS_ROOT_ITEM_KEY;
5246 if (is_cowonly_root(ref_path->root_objectid))
5247 root_key.offset = 0;
5248 else
5249 root_key.offset = (u64)-1;
5250
5251 return btrfs_read_fs_root_no_name(fs_info, &root_key);
5252}
5253
d397712b 5254static noinline int relocate_one_extent(struct btrfs_root *extent_root,
1a40e23b
ZY		 5255 struct btrfs_path *path,
5256 struct btrfs_key *extent_key,
5257 struct btrfs_block_group_cache *group,
5258 struct inode *reloc_inode, int pass)
5259{
5260 struct btrfs_trans_handle *trans;
5261 struct btrfs_root *found_root;
5262 struct btrfs_ref_path *ref_path = NULL;
5263 struct disk_extent *new_extents = NULL;
5264 int nr_extents = 0;
5265 int loops;
5266 int ret;
5267 int level;
5268 struct btrfs_key first_key;
5269 u64 prev_block = 0;
5270
1a40e23b
ZY		 5271
5272 trans = btrfs_start_transaction(extent_root, 1);
5273 BUG_ON(!trans);
5274
5275 if (extent_key->objectid == 0) {
5276 ret = del_extent_zero(trans, extent_root, path, extent_key);
5277 goto out;
5278 }
5279
5280 ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
5281 if (!ref_path) {
d397712b
CM		 5282 ret = -ENOMEM;
5283 goto out;
1a40e23b
ZY		 5284 }
5285
5286 for (loops = 0; ; loops++) {
5287 if (loops == 0) {
5288 ret = btrfs_first_ref_path(trans, extent_root, ref_path,
5289 extent_key->objectid);
5290 } else {
5291 ret = btrfs_next_ref_path(trans, extent_root, ref_path);
5292 }
5293 if (ret < 0)
5294 goto out;
5295 if (ret > 0)
5296 break;
5297
5298 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
5299 ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
5300 continue;
5301
5302 found_root = read_ref_root(extent_root->fs_info, ref_path);
5303 BUG_ON(!found_root);
5304 /*
5305 * for reference counted tree, only process reference paths
5306 * rooted at the latest committed root.
5307 */
5308 if (found_root->ref_cows &&
5309 ref_path->root_generation != found_root->root_key.offset)
5310 continue;
5311
5312 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5313 if (pass == 0) {
5314 /*
5315 * copy data extents to new locations
5316 */
5317 u64 group_start = group->key.objectid;
5318 ret = relocate_data_extent(reloc_inode,
5319 extent_key,
5320 group_start);
5321 if (ret < 0)
5322 goto out;
5323 break;
5324 }
5325 level = 0;
5326 } else {
5327 level = ref_path->owner_objectid;
5328 }
5329
5330 if (prev_block != ref_path->nodes[level]) {
5331 struct extent_buffer *eb;
5332 u64 block_start = ref_path->nodes[level];
5333 u64 block_size = btrfs_level_size(found_root, level);
5334
5335 eb = read_tree_block(found_root, block_start,
5336 block_size, 0);
5337 btrfs_tree_lock(eb);
5338 BUG_ON(level != btrfs_header_level(eb));
5339
5340 if (level == 0)
5341 btrfs_item_key_to_cpu(eb, &first_key, 0);
5342 else
5343 btrfs_node_key_to_cpu(eb, &first_key, 0);
5344
5345 btrfs_tree_unlock(eb);
5346 free_extent_buffer(eb);
5347 prev_block = block_start;
5348 }
5349
e4404d6e
YZ		 5350 btrfs_record_root_in_trans(found_root);
5351 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5352 /*
5353 * try to update data extent references while
5354 * keeping metadata shared between snapshots.
5355 */
5356 if (pass == 1) {
5357 ret = relocate_one_path(trans, found_root,
5358 path, &first_key, ref_path,
5359 group, reloc_inode);
5360 if (ret < 0)
5361 goto out;
5362 continue;
5363 }
1a40e23b
ZY		 5364 /*
5365 * use fallback method to process the remaining
5366 * references.
5367 */
5368 if (!new_extents) {
5369 u64 group_start = group->key.objectid;
d899e052
YZ		 5370 new_extents = kmalloc(sizeof(*new_extents),
5371 GFP_NOFS);
5372 nr_extents = 1;
1a40e23b
ZY		 5373 ret = get_new_locations(reloc_inode,
5374 extent_key,
d899e052 5375 group_start, 1,
1a40e23b
ZY		 5376 &new_extents,
5377 &nr_extents);
d899e052 5378 if (ret)
1a40e23b
ZY		 5379 goto out;
5380 }
1a40e23b
ZY		 5381 ret = replace_one_extent(trans, found_root,
5382 path, extent_key,
5383 &first_key, ref_path,
5384 new_extents, nr_extents);
e4404d6e 5385 } else {
1a40e23b
ZY		 5386 ret = relocate_tree_block(trans, found_root, path,
5387 &first_key, ref_path);
1a40e23b
ZY		 5388 }
5389 if (ret < 0)
5390 goto out;
5391 }
5392 ret = 0;
5393out:
5394 btrfs_end_transaction(trans, extent_root);
5395 kfree(new_extents);
5396 kfree(ref_path);
1a40e23b
ZY		 5397 return ret;
5398}
5399
ec44a35c
CM		 5400static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
5401{
5402 u64 num_devices;
5403 u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
5404 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
5405
2b82032c 5406 num_devices = root->fs_info->fs_devices->rw_devices;
ec44a35c
CM		 5407 if (num_devices == 1) {
5408 stripped |= BTRFS_BLOCK_GROUP_DUP;
5409 stripped = flags & ~stripped;
5410
5411 /* turn raid0 into single device chunks */
5412 if (flags & BTRFS_BLOCK_GROUP_RAID0)
5413 return stripped;
5414
5415 /* turn mirroring into duplication */
5416 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
5417 BTRFS_BLOCK_GROUP_RAID10))
5418 return stripped | BTRFS_BLOCK_GROUP_DUP;
5419 return flags;
5420 } else {
5421 /* they already had raid on here, just return */
ec44a35c
CM		 5422 if (flags & stripped)
5423 return flags;
5424
5425 stripped |= BTRFS_BLOCK_GROUP_DUP;
5426 stripped = flags & ~stripped;
5427
5428 /* switch duplicated blocks with raid1 */
5429 if (flags & BTRFS_BLOCK_GROUP_DUP)
5430 return stripped | BTRFS_BLOCK_GROUP_RAID1;
5431
5432 /* turn single device chunks into raid0 */
5433 return stripped | BTRFS_BLOCK_GROUP_RAID0;
5434 }
5435 return flags;
5436}
5437
b2950863 5438static int __alloc_chunk_for_shrink(struct btrfs_root *root,
0ef3e66b
CM		 5439 struct btrfs_block_group_cache *shrink_block_group,
5440 int force)
5441{
5442 struct btrfs_trans_handle *trans;
5443 u64 new_alloc_flags;
5444 u64 calc;
5445
c286ac48 5446 spin_lock(&shrink_block_group->lock);
0ef3e66b 5447 if (btrfs_block_group_used(&shrink_block_group->item) > 0) {
c286ac48 5448 spin_unlock(&shrink_block_group->lock);
c286ac48 5449
0ef3e66b 5450 trans = btrfs_start_transaction(root, 1);
c286ac48 5451 spin_lock(&shrink_block_group->lock);
7d9eb12c 5452
0ef3e66b
CM		 5453 new_alloc_flags = update_block_group_flags(root,
5454 shrink_block_group->flags);
5455 if (new_alloc_flags != shrink_block_group->flags) {
5456 calc =
5457 btrfs_block_group_used(&shrink_block_group->item);
5458 } else {
5459 calc = shrink_block_group->key.offset;
5460 }
c286ac48
CM		 5461 spin_unlock(&shrink_block_group->lock);
5462
0ef3e66b
CM		 5463 do_chunk_alloc(trans, root->fs_info->extent_root,
5464 calc + 2 * 1024 * 1024, new_alloc_flags, force);
7d9eb12c 5465
0ef3e66b 5466 btrfs_end_transaction(trans, root);
c286ac48
CM		 5467 } else
5468 spin_unlock(&shrink_block_group->lock);
0ef3e66b
CM		 5469 return 0;
5470}
5471
1a40e23b
ZY		 5472static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
5473 struct btrfs_root *root,
5474 u64 objectid, u64 size)
5475{
5476 struct btrfs_path *path;
5477 struct btrfs_inode_item *item;
5478 struct extent_buffer *leaf;
5479 int ret;
5480
5481 path = btrfs_alloc_path();
5482 if (!path)
5483 return -ENOMEM;
5484
5485 ret = btrfs_insert_empty_inode(trans, root, path, objectid);
5486 if (ret)
5487 goto out;
5488
5489 leaf = path->nodes[0];
5490 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
5491 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
5492 btrfs_set_inode_generation(leaf, item, 1);
5493 btrfs_set_inode_size(leaf, item, size);
5494 btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
0403e47e 5495 btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS);
1a40e23b
ZY		 5496 btrfs_mark_buffer_dirty(leaf);
5497 btrfs_release_path(root, path);
5498out:
5499 btrfs_free_path(path);
5500 return ret;
5501}
5502
d397712b 5503static noinline struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info,
1a40e23b
ZY		 5504 struct btrfs_block_group_cache *group)
5505{
5506 struct inode *inode = NULL;
5507 struct btrfs_trans_handle *trans;
5508 struct btrfs_root *root;
5509 struct btrfs_key root_key;
5510 u64 objectid = BTRFS_FIRST_FREE_OBJECTID;
5511 int err = 0;
5512
5513 root_key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
5514 root_key.type = BTRFS_ROOT_ITEM_KEY;
5515 root_key.offset = (u64)-1;
5516 root = btrfs_read_fs_root_no_name(fs_info, &root_key);
5517 if (IS_ERR(root))
5518 return ERR_CAST(root);
5519
5520 trans = btrfs_start_transaction(root, 1);
5521 BUG_ON(!trans);
5522
5523 err = btrfs_find_free_objectid(trans, root, objectid, &objectid);
5524 if (err)
5525 goto out;
5526
5527 err = __insert_orphan_inode(trans, root, objectid, group->key.offset);
5528 BUG_ON(err);
5529
5530 err = btrfs_insert_file_extent(trans, root, objectid, 0, 0, 0,
c8b97818
CM		 5531 group->key.offset, 0, group->key.offset,
5532 0, 0, 0);
1a40e23b
ZY		 5533 BUG_ON(err);
5534
5535 inode = btrfs_iget_locked(root->fs_info->sb, objectid, root);
5536 if (inode->i_state & I_NEW) {
5537 BTRFS_I(inode)->root = root;
5538 BTRFS_I(inode)->location.objectid = objectid;
5539 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
5540 BTRFS_I(inode)->location.offset = 0;
5541 btrfs_read_locked_inode(inode);
5542 unlock_new_inode(inode);
5543 BUG_ON(is_bad_inode(inode));
5544 } else {
5545 BUG_ON(1);
5546 }
17d217fe 5547 BTRFS_I(inode)->index_cnt = group->key.objectid;
1a40e23b
ZY		 5548
5549 err = btrfs_orphan_add(trans, inode);
5550out:
5551 btrfs_end_transaction(trans, root);
5552 if (err) {
5553 if (inode)
5554 iput(inode);
5555 inode = ERR_PTR(err);
5556 }
5557 return inode;
5558}
5559
17d217fe
YZ		 5560int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len)
5561{
5562
5563 struct btrfs_ordered_sum *sums;
5564 struct btrfs_sector_sum *sector_sum;
5565 struct btrfs_ordered_extent *ordered;
5566 struct btrfs_root *root = BTRFS_I(inode)->root;
5567 struct list_head list;
5568 size_t offset;
5569 int ret;
5570 u64 disk_bytenr;
5571
5572 INIT_LIST_HEAD(&list);
5573
5574 ordered = btrfs_lookup_ordered_extent(inode, file_pos);
5575 BUG_ON(ordered->file_offset != file_pos || ordered->len != len);
5576
5577 disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt;
07d400a6 5578 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, disk_bytenr,
17d217fe
YZ		 5579 disk_bytenr + len - 1, &list);
5580
5581 while (!list_empty(&list)) {
5582 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
5583 list_del_init(&sums->list);
5584
5585 sector_sum = sums->sums;
5586 sums->bytenr = ordered->start;
5587
5588 offset = 0;
5589 while (offset < sums->len) {
5590 sector_sum->bytenr += ordered->start - disk_bytenr;
5591 sector_sum++;
5592 offset += root->sectorsize;
5593 }
5594
5595 btrfs_add_ordered_sum(inode, ordered, sums);
5596 }
5597 btrfs_put_ordered_extent(ordered);
5598 return 0;
5599}
5600
1a40e23b 5601int btrfs_relocate_block_group(struct btrfs_root *root, u64 group_start)
edbd8d4e
CM		 5602{
5603 struct btrfs_trans_handle *trans;
edbd8d4e 5604 struct btrfs_path *path;
1a40e23b
ZY		 5605 struct btrfs_fs_info *info = root->fs_info;
5606 struct extent_buffer *leaf;
5607 struct inode *reloc_inode;
5608 struct btrfs_block_group_cache *block_group;
5609 struct btrfs_key key;
d899e052 5610 u64 skipped;
edbd8d4e
CM		 5611 u64 cur_byte;
5612 u64 total_found;
edbd8d4e
CM		 5613 u32 nritems;
5614 int ret;
a061fc8d 5615 int progress;
1a40e23b 5616 int pass = 0;
edbd8d4e 5617
1a40e23b
ZY		 5618 root = root->fs_info->extent_root;
5619
5620 block_group = btrfs_lookup_block_group(info, group_start);
5621 BUG_ON(!block_group);
8f18cf13 5622
d397712b 5623 printk(KERN_INFO "btrfs relocating block group %llu flags %llu\n",
1a40e23b
ZY		 5624 (unsigned long long)block_group->key.objectid,
5625 (unsigned long long)block_group->flags);
8f18cf13 5626
edbd8d4e 5627 path = btrfs_alloc_path();
1a40e23b 5628 BUG_ON(!path);
edbd8d4e 5629
1a40e23b
ZY		 5630 reloc_inode = create_reloc_inode(info, block_group);
5631 BUG_ON(IS_ERR(reloc_inode));
323da79c 5632
1a40e23b 5633 __alloc_chunk_for_shrink(root, block_group, 1);
c146afad 5634 set_block_group_readonly(block_group);
323da79c 5635
1a40e23b
ZY		 5636 btrfs_start_delalloc_inodes(info->tree_root);
5637 btrfs_wait_ordered_extents(info->tree_root, 0);
5638again:
d899e052 5639 skipped = 0;
edbd8d4e 5640 total_found = 0;
a061fc8d 5641 progress = 0;
1a40e23b 5642 key.objectid = block_group->key.objectid;
edbd8d4e
CM		 5643 key.offset = 0;
5644 key.type = 0;
73e48b27 5645 cur_byte = key.objectid;
4313b399 5646
1a40e23b
ZY		 5647 trans = btrfs_start_transaction(info->tree_root, 1);
5648 btrfs_commit_transaction(trans, info->tree_root);
ea8c2819 5649
1a40e23b
ZY		 5650 mutex_lock(&root->fs_info->cleaner_mutex);
5651 btrfs_clean_old_snapshots(info->tree_root);
5652 btrfs_remove_leaf_refs(info->tree_root, (u64)-1, 1);
5653 mutex_unlock(&root->fs_info->cleaner_mutex);
ea8c2819 5654
d397712b 5655 while (1) {
edbd8d4e
CM		 5656 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5657 if (ret < 0)
5658 goto out;
7d9eb12c 5659next:
edbd8d4e 5660 leaf = path->nodes[0];
73e48b27 5661 nritems = btrfs_header_nritems(leaf);
73e48b27
Y		 5662 if (path->slots[0] >= nritems) {
5663 ret = btrfs_next_leaf(root, path);
5664 if (ret < 0)
5665 goto out;
5666 if (ret == 1) {
5667 ret = 0;
5668 break;
edbd8d4e 5669 }
73e48b27
Y		 5670 leaf = path->nodes[0];
5671 nritems = btrfs_header_nritems(leaf);
edbd8d4e 5672 }
73e48b27 5673
1a40e23b 5674 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
725c8463 5675
1a40e23b
ZY		 5676 if (key.objectid >= block_group->key.objectid +
5677 block_group->key.offset)
8f18cf13
CM		 5678 break;
5679
725c8463 5680 if (progress && need_resched()) {
725c8463 5681 btrfs_release_path(root, path);
1a40e23b 5682 cond_resched();
725c8463 5683 progress = 0;
1a40e23b 5684 continue;
725c8463
CM		 5685 }
5686 progress = 1;
5687
1a40e23b
ZY		 5688 if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY ||
5689 key.objectid + key.offset <= cur_byte) {
edbd8d4e 5690 path->slots[0]++;
edbd8d4e
CM		 5691 goto next;
5692 }
73e48b27 5693
edbd8d4e 5694 total_found++;
1a40e23b 5695 cur_byte = key.objectid + key.offset;
edbd8d4e 5696 btrfs_release_path(root, path);
edbd8d4e 5697
1a40e23b
ZY		 5698 __alloc_chunk_for_shrink(root, block_group, 0);
5699 ret = relocate_one_extent(root, path, &key, block_group,
5700 reloc_inode, pass);
5701 BUG_ON(ret < 0);
d899e052
YZ		 5702 if (ret > 0)
5703 skipped++;
edbd8d4e 5704
1a40e23b
ZY		 5705 key.objectid = cur_byte;
5706 key.type = 0;
5707 key.offset = 0;
edbd8d4e
CM		 5708 }
5709
1a40e23b 5710 btrfs_release_path(root, path);
7d9eb12c 5711
1a40e23b
ZY		 5712 if (pass == 0) {
5713 btrfs_wait_ordered_range(reloc_inode, 0, (u64)-1);
5714 invalidate_mapping_pages(reloc_inode->i_mapping, 0, -1);
8e8a1e31 5715 }
73e48b27 5716
1a40e23b 5717 if (total_found > 0) {
d397712b 5718 printk(KERN_INFO "btrfs found %llu extents in pass %d\n",
1a40e23b
ZY		 5719 (unsigned long long)total_found, pass);
5720 pass++;
d899e052
YZ		 5721 if (total_found == skipped && pass > 2) {
5722 iput(reloc_inode);
5723 reloc_inode = create_reloc_inode(info, block_group);
5724 pass = 0;
5725 }
1a40e23b 5726 goto again;
0f9dd46c 5727 }
0ef3e66b 5728
1a40e23b
ZY		 5729 /* delete reloc_inode */
5730 iput(reloc_inode);
5731
5732 /* unpin extents in this range */
5733 trans = btrfs_start_transaction(info->tree_root, 1);
5734 btrfs_commit_transaction(trans, info->tree_root);
0ef3e66b 5735
1a40e23b
ZY		 5736 spin_lock(&block_group->lock);
5737 WARN_ON(block_group->pinned > 0);
5738 WARN_ON(block_group->reserved > 0);
5739 WARN_ON(btrfs_block_group_used(&block_group->item) > 0);
5740 spin_unlock(&block_group->lock);
d2fb3437 5741 put_block_group(block_group);
1a40e23b 5742 ret = 0;
edbd8d4e 5743out:
1a40e23b 5744 btrfs_free_path(path);
edbd8d4e
CM		 5745 return ret;
5746}
5747
b2950863
CH		 5748static int find_first_block_group(struct btrfs_root *root,
5749 struct btrfs_path *path, struct btrfs_key *key)
0b86a832 5750{
925baedd 5751 int ret = 0;
0b86a832
CM		 5752 struct btrfs_key found_key;
5753 struct extent_buffer *leaf;
5754 int slot;
edbd8d4e 5755
0b86a832
CM		 5756 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
5757 if (ret < 0)
925baedd
CM		 5758 goto out;
5759
d397712b 5760 while (1) {
0b86a832 5761 slot = path->slots[0];
edbd8d4e 5762 leaf = path->nodes[0];
0b86a832
CM		 5763 if (slot >= btrfs_header_nritems(leaf)) {
5764 ret = btrfs_next_leaf(root, path);
5765 if (ret == 0)
5766 continue;
5767 if (ret < 0)
925baedd 5768 goto out;
0b86a832 5769 break;
edbd8d4e 5770 }
0b86a832 5771 btrfs_item_key_to_cpu(leaf, &found_key, slot);
edbd8d4e 5772
0b86a832 5773 if (found_key.objectid >= key->objectid &&
925baedd
CM		 5774 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
5775 ret = 0;
5776 goto out;
5777 }
0b86a832 5778 path->slots[0]++;
edbd8d4e 5779 }
0b86a832 5780 ret = -ENOENT;
925baedd 5781out:
0b86a832 5782 return ret;
edbd8d4e
CM		 5783}
5784
1a40e23b
ZY		 5785int btrfs_free_block_groups(struct btrfs_fs_info *info)
5786{
5787 struct btrfs_block_group_cache *block_group;
5788 struct rb_node *n;
5789
1a40e23b
ZY		 5790 spin_lock(&info->block_group_cache_lock);
5791 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
5792 block_group = rb_entry(n, struct btrfs_block_group_cache,
5793 cache_node);
1a40e23b
ZY		 5794 rb_erase(&block_group->cache_node,
5795 &info->block_group_cache_tree);
d899e052
YZ		 5796 spin_unlock(&info->block_group_cache_lock);
5797
5798 btrfs_remove_free_space_cache(block_group);
80eb234a 5799 down_write(&block_group->space_info->groups_sem);
1a40e23b 5800 list_del(&block_group->list);
80eb234a 5801 up_write(&block_group->space_info->groups_sem);
d2fb3437
YZ		 5802
5803 WARN_ON(atomic_read(&block_group->count) != 1);
1a40e23b 5804 kfree(block_group);
d899e052
YZ		 5805
5806 spin_lock(&info->block_group_cache_lock);
1a40e23b
ZY		 5807 }
5808 spin_unlock(&info->block_group_cache_lock);
1a40e23b
ZY		 5809 return 0;
5810}
5811
9078a3e1
CM		 5812int btrfs_read_block_groups(struct btrfs_root *root)
5813{
5814 struct btrfs_path *path;
5815 int ret;
9078a3e1 5816 struct btrfs_block_group_cache *cache;
be744175 5817 struct btrfs_fs_info *info = root->fs_info;
6324fbf3 5818 struct btrfs_space_info *space_info;
9078a3e1
CM		 5819 struct btrfs_key key;
5820 struct btrfs_key found_key;
5f39d397 5821 struct extent_buffer *leaf;
96b5179d 5822
be744175 5823 root = info->extent_root;
9078a3e1 5824 key.objectid = 0;
0b86a832 5825 key.offset = 0;
9078a3e1 5826 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
9078a3e1
CM		 5827 path = btrfs_alloc_path();
5828 if (!path)
5829 return -ENOMEM;
5830
d397712b 5831 while (1) {
0b86a832
CM		 5832 ret = find_first_block_group(root, path, &key);
5833 if (ret > 0) {
5834 ret = 0;
5835 goto error;
9078a3e1 5836 }
0b86a832
CM		 5837 if (ret != 0)
5838 goto error;
5839
5f39d397
CM		 5840 leaf = path->nodes[0];
5841 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
8f18cf13 5842 cache = kzalloc(sizeof(*cache), GFP_NOFS);
9078a3e1 5843 if (!cache) {
0b86a832 5844 ret = -ENOMEM;
9078a3e1
CM		 5845 break;
5846 }
3e1ad54f 5847
d2fb3437 5848 atomic_set(&cache->count, 1);
c286ac48 5849 spin_lock_init(&cache->lock);
25179201 5850 mutex_init(&cache->alloc_mutex);
ea6a478e 5851 mutex_init(&cache->cache_mutex);
0f9dd46c 5852 INIT_LIST_HEAD(&cache->list);
5f39d397
CM		 5853 read_extent_buffer(leaf, &cache->item,
5854 btrfs_item_ptr_offset(leaf, path->slots[0]),
5855 sizeof(cache->item));
9078a3e1 5856 memcpy(&cache->key, &found_key, sizeof(found_key));
0b86a832 5857
9078a3e1
CM		 5858 key.objectid = found_key.objectid + found_key.offset;
5859 btrfs_release_path(root, path);
0b86a832 5860 cache->flags = btrfs_block_group_flags(&cache->item);
96b5179d 5861
6324fbf3
CM		 5862 ret = update_space_info(info, cache->flags, found_key.offset,
5863 btrfs_block_group_used(&cache->item),
5864 &space_info);
5865 BUG_ON(ret);
5866 cache->space_info = space_info;
80eb234a
JB		 5867 down_write(&space_info->groups_sem);
5868 list_add_tail(&cache->list, &space_info->block_groups);
5869 up_write(&space_info->groups_sem);
0f9dd46c
JB		 5870
5871 ret = btrfs_add_block_group_cache(root->fs_info, cache);
5872 BUG_ON(ret);
75ccf47d
CM		 5873
5874 set_avail_alloc_bits(root->fs_info, cache->flags);
2b82032c
YZ		 5875 if (btrfs_chunk_readonly(root, cache->key.objectid))
5876 set_block_group_readonly(cache);
9078a3e1 5877 }
0b86a832
CM		 5878 ret = 0;
5879error:
9078a3e1 5880 btrfs_free_path(path);
0b86a832 5881 return ret;
9078a3e1 5882}
6324fbf3
CM		 5883
5884int btrfs_make_block_group(struct btrfs_trans_handle *trans,
5885 struct btrfs_root *root, u64 bytes_used,
e17cade2 5886 u64 type, u64 chunk_objectid, u64 chunk_offset,
6324fbf3
CM		 5887 u64 size)
5888{
5889 int ret;
6324fbf3
CM		 5890 struct btrfs_root *extent_root;
5891 struct btrfs_block_group_cache *cache;
6324fbf3
CM		 5892
5893 extent_root = root->fs_info->extent_root;
6324fbf3 5894
e02119d5
CM		 5895 root->fs_info->last_trans_new_blockgroup = trans->transid;
5896
8f18cf13 5897 cache = kzalloc(sizeof(*cache), GFP_NOFS);
0f9dd46c
JB		 5898 if (!cache)
5899 return -ENOMEM;
5900
e17cade2 5901 cache->key.objectid = chunk_offset;
6324fbf3 5902 cache->key.offset = size;
d2fb3437
YZ		 5903 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
5904 atomic_set(&cache->count, 1);
c286ac48 5905 spin_lock_init(&cache->lock);
25179201 5906 mutex_init(&cache->alloc_mutex);
ea6a478e 5907 mutex_init(&cache->cache_mutex);
0f9dd46c 5908 INIT_LIST_HEAD(&cache->list);
0ef3e66b 5909
6324fbf3 5910 btrfs_set_block_group_used(&cache->item, bytes_used);
6324fbf3
CM		 5911 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
5912 cache->flags = type;
5913 btrfs_set_block_group_flags(&cache->item, type);
5914
5915 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
5916 &cache->space_info);
5917 BUG_ON(ret);
80eb234a
JB		 5918 down_write(&cache->space_info->groups_sem);
5919 list_add_tail(&cache->list, &cache->space_info->block_groups);
5920 up_write(&cache->space_info->groups_sem);
6324fbf3 5921
0f9dd46c
JB		 5922 ret = btrfs_add_block_group_cache(root->fs_info, cache);
5923 BUG_ON(ret);
c286ac48 5924
6324fbf3
CM		 5925 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
5926 sizeof(cache->item));
5927 BUG_ON(ret);
5928
87ef2bb4
CM		 5929 finish_current_insert(trans, extent_root, 0);
5930 ret = del_pending_extents(trans, extent_root, 0);
6324fbf3 5931 BUG_ON(ret);
d18a2c44 5932 set_avail_alloc_bits(extent_root->fs_info, type);
925baedd 5933
6324fbf3
CM		 5934 return 0;
5935}
1a40e23b
ZY		 5936
5937int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
5938 struct btrfs_root *root, u64 group_start)
5939{
5940 struct btrfs_path *path;
5941 struct btrfs_block_group_cache *block_group;
5942 struct btrfs_key key;
5943 int ret;
5944
1a40e23b
ZY		 5945 root = root->fs_info->extent_root;
5946
5947 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
5948 BUG_ON(!block_group);
c146afad 5949 BUG_ON(!block_group->ro);
1a40e23b
ZY		 5950
5951 memcpy(&key, &block_group->key, sizeof(key));
5952
5953 path = btrfs_alloc_path();
5954 BUG_ON(!path);
5955
3dfdb934 5956 spin_lock(&root->fs_info->block_group_cache_lock);
1a40e23b
ZY		 5957 rb_erase(&block_group->cache_node,
5958 &root->fs_info->block_group_cache_tree);
3dfdb934
YZ		 5959 spin_unlock(&root->fs_info->block_group_cache_lock);
5960 btrfs_remove_free_space_cache(block_group);
80eb234a 5961 down_write(&block_group->space_info->groups_sem);
1a40e23b 5962 list_del(&block_group->list);
80eb234a 5963 up_write(&block_group->space_info->groups_sem);
1a40e23b 5964
c146afad
YZ		 5965 spin_lock(&block_group->space_info->lock);
5966 block_group->space_info->total_bytes -= block_group->key.offset;
5967 block_group->space_info->bytes_readonly -= block_group->key.offset;
5968 spin_unlock(&block_group->space_info->lock);
2b82032c 5969 block_group->space_info->full = 0;
c146afad 5970
d2fb3437
YZ		 5971 put_block_group(block_group);
5972 put_block_group(block_group);
1a40e23b
ZY		 5973
5974 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
5975 if (ret > 0)
5976 ret = -EIO;
5977 if (ret < 0)
5978 goto out;
5979
5980 ret = btrfs_del_item(trans, root, path);
5981out:
5982 btrfs_free_path(path);
5983 return ret;
5984}
Linux 6.x block layer and io_uring tree(s)