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