projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Btrfs: Add zlib compression support
[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>
74493f7a 22#include "hash.h"
a5eb62e3 23#include "crc32c.h"
fec577fb
CM		 24#include "ctree.h"
25#include "disk-io.h"
26#include "print-tree.h"
e089f05c 27#include "transaction.h"
0b86a832 28#include "volumes.h"
925baedd 29#include "locking.h"
31153d81 30#include "ref-cache.h"
fec577fb 31
31840ae1
ZY		 32#define PENDING_EXTENT_INSERT 0
33#define PENDING_EXTENT_DELETE 1
34#define PENDING_BACKREF_UPDATE 2
35
36struct pending_extent_op {
37 int type;
38 u64 bytenr;
39 u64 num_bytes;
40 u64 parent;
41 u64 orig_parent;
42 u64 generation;
43 u64 orig_generation;
44 int level;
45};
46
e089f05c
CM		 47static int finish_current_insert(struct btrfs_trans_handle *trans, struct
48 btrfs_root *extent_root);
e20d96d6
CM		 49static int del_pending_extents(struct btrfs_trans_handle *trans, struct
50 btrfs_root *extent_root);
925baedd
CM		 51static struct btrfs_block_group_cache *
52__btrfs_find_block_group(struct btrfs_root *root,
53 struct btrfs_block_group_cache *hint,
54 u64 search_start, int data, int owner);
d548ee51 55
925baedd
CM		 56void maybe_lock_mutex(struct btrfs_root *root)
57{
58 if (root != root->fs_info->extent_root &&
59 root != root->fs_info->chunk_root &&
60 root != root->fs_info->dev_root) {
61 mutex_lock(&root->fs_info->alloc_mutex);
62 }
63}
64
65void maybe_unlock_mutex(struct btrfs_root *root)
66{
67 if (root != root->fs_info->extent_root &&
68 root != root->fs_info->chunk_root &&
69 root != root->fs_info->dev_root) {
70 mutex_unlock(&root->fs_info->alloc_mutex);
71 }
72}
fec577fb 73
0f9dd46c
JB		 74static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
75{
76 return (cache->flags & bits) == bits;
77}
78
79/*
80 * this adds the block group to the fs_info rb tree for the block group
81 * cache
82 */
83int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
84 struct btrfs_block_group_cache *block_group)
85{
86 struct rb_node **p;
87 struct rb_node *parent = NULL;
88 struct btrfs_block_group_cache *cache;
89
90 spin_lock(&info->block_group_cache_lock);
91 p = &info->block_group_cache_tree.rb_node;
92
93 while (*p) {
94 parent = *p;
95 cache = rb_entry(parent, struct btrfs_block_group_cache,
96 cache_node);
97 if (block_group->key.objectid < cache->key.objectid) {
98 p = &(*p)->rb_left;
99 } else if (block_group->key.objectid > cache->key.objectid) {
100 p = &(*p)->rb_right;
101 } else {
102 spin_unlock(&info->block_group_cache_lock);
103 return -EEXIST;
104 }
105 }
106
107 rb_link_node(&block_group->cache_node, parent, p);
108 rb_insert_color(&block_group->cache_node,
109 &info->block_group_cache_tree);
110 spin_unlock(&info->block_group_cache_lock);
111
112 return 0;
113}
114
115/*
116 * This will return the block group at or after bytenr if contains is 0, else
117 * it will return the block group that contains the bytenr
118 */
119static struct btrfs_block_group_cache *
120block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
121 int contains)
122{
123 struct btrfs_block_group_cache *cache, *ret = NULL;
124 struct rb_node *n;
125 u64 end, start;
126
127 spin_lock(&info->block_group_cache_lock);
128 n = info->block_group_cache_tree.rb_node;
129
130 while (n) {
131 cache = rb_entry(n, struct btrfs_block_group_cache,
132 cache_node);
133 end = cache->key.objectid + cache->key.offset - 1;
134 start = cache->key.objectid;
135
136 if (bytenr < start) {
137 if (!contains && (!ret || start < ret->key.objectid))
138 ret = cache;
139 n = n->rb_left;
140 } else if (bytenr > start) {
141 if (contains && bytenr <= end) {
142 ret = cache;
143 break;
144 }
145 n = n->rb_right;
146 } else {
147 ret = cache;
148 break;
149 }
150 }
151 spin_unlock(&info->block_group_cache_lock);
152
153 return ret;
154}
155
156/*
157 * this is only called by cache_block_group, since we could have freed extents
158 * we need to check the pinned_extents for any extents that can't be used yet
159 * since their free space will be released as soon as the transaction commits.
160 */
161static int add_new_free_space(struct btrfs_block_group_cache *block_group,
162 struct btrfs_fs_info *info, u64 start, u64 end)
163{
164 u64 extent_start, extent_end, size;
165 int ret;
166
167 while (start < end) {
168 ret = find_first_extent_bit(&info->pinned_extents, start,
169 &extent_start, &extent_end,
170 EXTENT_DIRTY);
171 if (ret)
172 break;
173
174 if (extent_start == start) {
175 start = extent_end + 1;
176 } else if (extent_start > start && extent_start < end) {
177 size = extent_start - start;
178 ret = btrfs_add_free_space(block_group, start, size);
179 BUG_ON(ret);
180 start = extent_end + 1;
181 } else {
182 break;
183 }
184 }
185
186 if (start < end) {
187 size = end - start;
188 ret = btrfs_add_free_space(block_group, start, size);
189 BUG_ON(ret);
190 }
191
192 return 0;
193}
194
e37c9e69
CM		 195static int cache_block_group(struct btrfs_root *root,
196 struct btrfs_block_group_cache *block_group)
197{
198 struct btrfs_path *path;
ef8bbdfe 199 int ret = 0;
e37c9e69 200 struct btrfs_key key;
5f39d397 201 struct extent_buffer *leaf;
e37c9e69 202 int slot;
e37c9e69 203 u64 last = 0;
7d7d6068 204 u64 first_free;
e37c9e69
CM		 205 int found = 0;
206
00f5c795
CM		 207 if (!block_group)
208 return 0;
209
e37c9e69 210 root = root->fs_info->extent_root;
e37c9e69
CM		 211
212 if (block_group->cached)
213 return 0;
f510cfec 214
e37c9e69
CM		 215 path = btrfs_alloc_path();
216 if (!path)
217 return -ENOMEM;
7d7d6068 218
2cc58cf2 219 path->reada = 2;
5cd57b2c
CM		 220 /*
221 * we get into deadlocks with paths held by callers of this function.
222 * since the alloc_mutex is protecting things right now, just
223 * skip the locking here
224 */
225 path->skip_locking = 1;
0f9dd46c
JB		 226 first_free = max_t(u64, block_group->key.objectid,
227 BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE);
e37c9e69 228 key.objectid = block_group->key.objectid;
e37c9e69
CM		 229 key.offset = 0;
230 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
231 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
232 if (ret < 0)
ef8bbdfe 233 goto err;
0b86a832 234 ret = btrfs_previous_item(root, path, 0, BTRFS_EXTENT_ITEM_KEY);
d548ee51 235 if (ret < 0)
ef8bbdfe 236 goto err;
d548ee51
Y		 237 if (ret == 0) {
238 leaf = path->nodes[0];
239 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
240 if (key.objectid + key.offset > first_free)
241 first_free = key.objectid + key.offset;
242 }
e37c9e69 243 while(1) {
5f39d397 244 leaf = path->nodes[0];
e37c9e69 245 slot = path->slots[0];
5f39d397 246 if (slot >= btrfs_header_nritems(leaf)) {
e37c9e69 247 ret = btrfs_next_leaf(root, path);
54aa1f4d
CM		 248 if (ret < 0)
249 goto err;
0f9dd46c 250 if (ret == 0)
e37c9e69 251 continue;
0f9dd46c 252 else
e37c9e69 253 break;
e37c9e69 254 }
5f39d397 255 btrfs_item_key_to_cpu(leaf, &key, slot);
0f9dd46c 256 if (key.objectid < block_group->key.objectid)
7d7d6068 257 goto next;
0f9dd46c 258
e37c9e69 259 if (key.objectid >= block_group->key.objectid +
0f9dd46c 260 block_group->key.offset)
e37c9e69 261 break;
7d7d6068 262
e37c9e69
CM		 263 if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
264 if (!found) {
7d7d6068 265 last = first_free;
e37c9e69 266 found = 1;
e37c9e69 267 }
0f9dd46c
JB		 268
269 add_new_free_space(block_group, root->fs_info, last,
270 key.objectid);
271
7d7d6068 272 last = key.objectid + key.offset;
e37c9e69 273 }
7d7d6068 274next:
e37c9e69
CM		 275 path->slots[0]++;
276 }
277
7d7d6068
Y		 278 if (!found)
279 last = first_free;
0f9dd46c
JB		 280
281 add_new_free_space(block_group, root->fs_info, last,
282 block_group->key.objectid +
283 block_group->key.offset);
284
e37c9e69 285 block_group->cached = 1;
ef8bbdfe 286 ret = 0;
54aa1f4d 287err:
e37c9e69 288 btrfs_free_path(path);
ef8bbdfe 289 return ret;
e37c9e69
CM		 290}
291
0f9dd46c
JB		 292/*
293 * return the block group that starts at or after bytenr
294 */
0ef3e66b
CM		 295struct btrfs_block_group_cache *btrfs_lookup_first_block_group(struct
296 btrfs_fs_info *info,
297 u64 bytenr)
298{
0f9dd46c 299 struct btrfs_block_group_cache *cache;
0ef3e66b 300
0f9dd46c 301 cache = block_group_cache_tree_search(info, bytenr, 0);
0ef3e66b 302
0f9dd46c 303 return cache;
0ef3e66b
CM		 304}
305
0f9dd46c
JB		 306/*
307 * return the block group that contains teh given bytenr
308 */
5276aeda
CM		 309struct btrfs_block_group_cache *btrfs_lookup_block_group(struct
310 btrfs_fs_info *info,
db94535d 311 u64 bytenr)
be744175 312{
0f9dd46c 313 struct btrfs_block_group_cache *cache;
be744175 314
0f9dd46c 315 cache = block_group_cache_tree_search(info, bytenr, 1);
96b5179d 316
0f9dd46c 317 return cache;
be744175 318}
0b86a832 319
0f9dd46c
JB		 320static int noinline find_free_space(struct btrfs_root *root,
321 struct btrfs_block_group_cache **cache_ret,
322 u64 *start_ret, u64 num, int data)
e37c9e69 323{
e37c9e69
CM		 324 int ret;
325 struct btrfs_block_group_cache *cache = *cache_ret;
0f9dd46c 326 struct btrfs_free_space *info = NULL;
e19caa5f 327 u64 last;
0b86a832 328 u64 search_start = *start_ret;
e37c9e69 329
7d9eb12c 330 WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex));
0ef3e66b
CM		 331 if (!cache)
332 goto out;
333
0f9dd46c
JB		 334 last = max(search_start, cache->key.objectid);
335
e37c9e69 336again:
54aa1f4d 337 ret = cache_block_group(root, cache);
0f9dd46c 338 if (ret)
54aa1f4d 339 goto out;
f84a8b36 340
0f9dd46c 341 if (cache->ro || !block_group_bits(cache, data))
0b86a832 342 goto new_group;
e19caa5f 343
0f9dd46c
JB		 344 info = btrfs_find_free_space(cache, last, num);
345 if (info) {
346 *start_ret = info->offset;
0b86a832 347 return 0;
8790d502 348 }
e37c9e69
CM		 349
350new_group:
e19caa5f 351 last = cache->key.objectid + cache->key.offset;
0f9dd46c 352
0ef3e66b 353 cache = btrfs_lookup_first_block_group(root->fs_info, last);
e8569813 354 if (!cache)
1a2b2ac7 355 goto out;
0f9dd46c 356
e37c9e69
CM		 357 *cache_ret = cache;
358 goto again;
0f9dd46c
JB		 359
360out:
361 return -ENOSPC;
e37c9e69
CM		 362}
363
84f54cfa
CM		 364static u64 div_factor(u64 num, int factor)
365{
257d0ce3
CM		 366 if (factor == 10)
367 return num;
84f54cfa
CM		 368 num *= factor;
369 do_div(num, 10);
370 return num;
371}
372
0f9dd46c
JB		 373static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
374 u64 flags)
6324fbf3 375{
0f9dd46c
JB		 376 struct list_head *head = &info->space_info;
377 struct list_head *cur;
378 struct btrfs_space_info *found;
379 list_for_each(cur, head) {
380 found = list_entry(cur, struct btrfs_space_info, list);
381 if (found->flags == flags)
382 return found;
383 }
384 return NULL;
6324fbf3
CM		 385}
386
925baedd
CM		 387static struct btrfs_block_group_cache *
388__btrfs_find_block_group(struct btrfs_root *root,
389 struct btrfs_block_group_cache *hint,
390 u64 search_start, int data, int owner)
cd1bc465 391{
96b5179d 392 struct btrfs_block_group_cache *cache;
31f3c99b 393 struct btrfs_block_group_cache *found_group = NULL;
cd1bc465
CM		 394 struct btrfs_fs_info *info = root->fs_info;
395 u64 used;
31f3c99b 396 u64 last = 0;
96b5179d 397 u64 free_check;
31f3c99b 398 int full_search = 0;
bce4eae9 399 int factor = 10;
0ef3e66b 400 int wrapped = 0;
de428b63 401
a236aed1
CM		 402 if (data & BTRFS_BLOCK_GROUP_METADATA)
403 factor = 9;
be744175 404
0ef3e66b 405 if (search_start) {
be744175 406 struct btrfs_block_group_cache *shint;
0ef3e66b 407 shint = btrfs_lookup_first_block_group(info, search_start);
8f18cf13 408 if (shint && block_group_bits(shint, data) && !shint->ro) {
c286ac48 409 spin_lock(&shint->lock);
be744175 410 used = btrfs_block_group_used(&shint->item);
e8569813 411 if (used + shint->pinned + shint->reserved <
324ae4df 412 div_factor(shint->key.offset, factor)) {
c286ac48 413 spin_unlock(&shint->lock);
be744175
CM		 414 return shint;
415 }
c286ac48 416 spin_unlock(&shint->lock);
be744175
CM		 417 }
418 }
0ef3e66b 419 if (hint && !hint->ro && block_group_bits(hint, data)) {
c286ac48 420 spin_lock(&hint->lock);
31f3c99b 421 used = btrfs_block_group_used(&hint->item);
e8569813 422 if (used + hint->pinned + hint->reserved <
324ae4df 423 div_factor(hint->key.offset, factor)) {
c286ac48 424 spin_unlock(&hint->lock);
31f3c99b
CM		 425 return hint;
426 }
c286ac48 427 spin_unlock(&hint->lock);
e19caa5f 428 last = hint->key.objectid + hint->key.offset;
31f3c99b 429 } else {
e37c9e69 430 if (hint)
0ef3e66b 431 last = max(hint->key.objectid, search_start);
e37c9e69 432 else
0ef3e66b 433 last = search_start;
31f3c99b 434 }
31f3c99b 435again:
e8569813
ZY		 436 while (1) {
437 cache = btrfs_lookup_first_block_group(root->fs_info, last);
0f9dd46c
JB		 438 if (!cache)
439 break;
96b5179d 440
c286ac48 441 spin_lock(&cache->lock);
96b5179d
CM		 442 last = cache->key.objectid + cache->key.offset;
443 used = btrfs_block_group_used(&cache->item);
444
8f18cf13 445 if (!cache->ro && block_group_bits(cache, data)) {
0ef3e66b 446 free_check = div_factor(cache->key.offset, factor);
e8569813
ZY		 447 if (used + cache->pinned + cache->reserved <
448 free_check) {
8790d502 449 found_group = cache;
c286ac48 450 spin_unlock(&cache->lock);
8790d502
CM		 451 goto found;
452 }
6324fbf3 453 }
c286ac48 454 spin_unlock(&cache->lock);
de428b63 455 cond_resched();
cd1bc465 456 }
0ef3e66b
CM		 457 if (!wrapped) {
458 last = search_start;
459 wrapped = 1;
460 goto again;
461 }
462 if (!full_search && factor < 10) {
be744175 463 last = search_start;
31f3c99b 464 full_search = 1;
0ef3e66b 465 factor = 10;
31f3c99b
CM		 466 goto again;
467 }
be744175 468found:
31f3c99b 469 return found_group;
cd1bc465
CM		 470}
471
925baedd
CM		 472struct btrfs_block_group_cache *btrfs_find_block_group(struct btrfs_root *root,
473 struct btrfs_block_group_cache
474 *hint, u64 search_start,
475 int data, int owner)
476{
477
478 struct btrfs_block_group_cache *ret;
925baedd 479 ret = __btrfs_find_block_group(root, hint, search_start, data, owner);
925baedd
CM		 480 return ret;
481}
0f9dd46c 482
e02119d5 483/* simple helper to search for an existing extent at a given offset */
31840ae1 484int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
e02119d5
CM		 485{
486 int ret;
487 struct btrfs_key key;
31840ae1 488 struct btrfs_path *path;
e02119d5 489
31840ae1
ZY		 490 path = btrfs_alloc_path();
491 BUG_ON(!path);
e02119d5
CM		 492 maybe_lock_mutex(root);
493 key.objectid = start;
494 key.offset = len;
495 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
496 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
497 0, 0);
498 maybe_unlock_mutex(root);
31840ae1 499 btrfs_free_path(path);
7bb86316
CM		 500 return ret;
501}
502
d8d5f3e1
CM		 503/*
504 * Back reference rules. Back refs have three main goals:
505 *
506 * 1) differentiate between all holders of references to an extent so that
507 * when a reference is dropped we can make sure it was a valid reference
508 * before freeing the extent.
509 *
510 * 2) Provide enough information to quickly find the holders of an extent
511 * if we notice a given block is corrupted or bad.
512 *
513 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
514 * maintenance. This is actually the same as #2, but with a slightly
515 * different use case.
516 *
517 * File extents can be referenced by:
518 *
519 * - multiple snapshots, subvolumes, or different generations in one subvol
31840ae1 520 * - different files inside a single subvolume
d8d5f3e1
CM		 521 * - different offsets inside a file (bookend extents in file.c)
522 *
523 * The extent ref structure has fields for:
524 *
525 * - Objectid of the subvolume root
526 * - Generation number of the tree holding the reference
527 * - objectid of the file holding the reference
31840ae1
ZY		 528 * - number of references holding by parent node (alway 1 for tree blocks)
529 *
530 * Btree leaf may hold multiple references to a file extent. In most cases,
531 * these references are from same file and the corresponding offsets inside
3bb1a1bc 532 * the file are close together.
d8d5f3e1
CM		 533 *
534 * When a file extent is allocated the fields are filled in:
3bb1a1bc 535 * (root_key.objectid, trans->transid, inode objectid, 1)
d8d5f3e1
CM		 536 *
537 * When a leaf is cow'd new references are added for every file extent found
31840ae1
ZY		 538 * in the leaf. It looks similar to the create case, but trans->transid will
539 * be different when the block is cow'd.
d8d5f3e1 540 *
3bb1a1bc 541 * (root_key.objectid, trans->transid, inode objectid,
31840ae1 542 * number of references in the leaf)
d8d5f3e1 543 *
3bb1a1bc
YZ		 544 * When a file extent is removed either during snapshot deletion or
545 * file truncation, we find the corresponding back reference and check
546 * the following fields:
d8d5f3e1 547 *
3bb1a1bc
YZ		 548 * (btrfs_header_owner(leaf), btrfs_header_generation(leaf),
549 * inode objectid)
d8d5f3e1
CM		 550 *
551 * Btree extents can be referenced by:
552 *
553 * - Different subvolumes
554 * - Different generations of the same subvolume
555 *
d8d5f3e1
CM		 556 * When a tree block is created, back references are inserted:
557 *
3bb1a1bc 558 * (root->root_key.objectid, trans->transid, level, 1)
d8d5f3e1 559 *
31840ae1
ZY		 560 * When a tree block is cow'd, new back references are added for all the
561 * blocks it points to. If the tree block isn't in reference counted root,
562 * the old back references are removed. These new back references are of
563 * the form (trans->transid will have increased since creation):
d8d5f3e1 564 *
3bb1a1bc 565 * (root->root_key.objectid, trans->transid, level, 1)
d8d5f3e1 566 *
31840ae1 567 * When a backref is in deleting, the following fields are checked:
d8d5f3e1
CM		 568 *
569 * if backref was for a tree root:
3bb1a1bc 570 * (btrfs_header_owner(itself), btrfs_header_generation(itself), level)
d8d5f3e1 571 * else
3bb1a1bc 572 * (btrfs_header_owner(parent), btrfs_header_generation(parent), level)
d8d5f3e1 573 *
31840ae1 574 * Back Reference Key composing:
d8d5f3e1 575 *
31840ae1
ZY		 576 * The key objectid corresponds to the first byte in the extent, the key
577 * type is set to BTRFS_EXTENT_REF_KEY, and the key offset is the first
578 * byte of parent extent. If a extent is tree root, the key offset is set
579 * to the key objectid.
d8d5f3e1 580 */
31840ae1
ZY		 581
582static int noinline lookup_extent_backref(struct btrfs_trans_handle *trans,
583 struct btrfs_root *root,
3bb1a1bc
YZ		 584 struct btrfs_path *path,
585 u64 bytenr, u64 parent,
586 u64 ref_root, u64 ref_generation,
587 u64 owner_objectid, int del)
7bb86316 588{
7bb86316 589 struct btrfs_key key;
31840ae1
ZY		 590 struct btrfs_extent_ref *ref;
591 struct extent_buffer *leaf;
3bb1a1bc 592 u64 ref_objectid;
74493f7a
CM		 593 int ret;
594
31840ae1
ZY		 595 key.objectid = bytenr;
596 key.type = BTRFS_EXTENT_REF_KEY;
597 key.offset = parent;
598
599 ret = btrfs_search_slot(trans, root, &key, path, del ? -1 : 0, 1);
600 if (ret < 0)
601 goto out;
602 if (ret > 0) {
603 ret = -ENOENT;
604 goto out;
605 }
606
607 leaf = path->nodes[0];
608 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
3bb1a1bc 609 ref_objectid = btrfs_ref_objectid(leaf, ref);
31840ae1 610 if (btrfs_ref_root(leaf, ref) != ref_root ||
3bb1a1bc
YZ		 611 btrfs_ref_generation(leaf, ref) != ref_generation ||
612 (ref_objectid != owner_objectid &&
613 ref_objectid != BTRFS_MULTIPLE_OBJECTIDS)) {
31840ae1
ZY		 614 ret = -EIO;
615 WARN_ON(1);
616 goto out;
617 }
618 ret = 0;
619out:
620 return ret;
621}
622
623static int noinline insert_extent_backref(struct btrfs_trans_handle *trans,
624 struct btrfs_root *root,
625 struct btrfs_path *path,
626 u64 bytenr, u64 parent,
627 u64 ref_root, u64 ref_generation,
3bb1a1bc 628 u64 owner_objectid)
31840ae1
ZY		 629{
630 struct btrfs_key key;
631 struct extent_buffer *leaf;
632 struct btrfs_extent_ref *ref;
633 u32 num_refs;
634 int ret;
74493f7a 635
74493f7a
CM		 636 key.objectid = bytenr;
637 key.type = BTRFS_EXTENT_REF_KEY;
31840ae1 638 key.offset = parent;
74493f7a 639
31840ae1
ZY		 640 ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*ref));
641 if (ret == 0) {
642 leaf = path->nodes[0];
643 ref = btrfs_item_ptr(leaf, path->slots[0],
644 struct btrfs_extent_ref);
645 btrfs_set_ref_root(leaf, ref, ref_root);
646 btrfs_set_ref_generation(leaf, ref, ref_generation);
647 btrfs_set_ref_objectid(leaf, ref, owner_objectid);
31840ae1
ZY		 648 btrfs_set_ref_num_refs(leaf, ref, 1);
649 } else if (ret == -EEXIST) {
650 u64 existing_owner;
651 BUG_ON(owner_objectid < BTRFS_FIRST_FREE_OBJECTID);
652 leaf = path->nodes[0];
653 ref = btrfs_item_ptr(leaf, path->slots[0],
654 struct btrfs_extent_ref);
655 if (btrfs_ref_root(leaf, ref) != ref_root ||
656 btrfs_ref_generation(leaf, ref) != ref_generation) {
657 ret = -EIO;
658 WARN_ON(1);
7bb86316 659 goto out;
31840ae1
ZY		 660 }
661
662 num_refs = btrfs_ref_num_refs(leaf, ref);
663 BUG_ON(num_refs == 0);
664 btrfs_set_ref_num_refs(leaf, ref, num_refs + 1);
665
666 existing_owner = btrfs_ref_objectid(leaf, ref);
3bb1a1bc
YZ		 667 if (existing_owner != owner_objectid &&
668 existing_owner != BTRFS_MULTIPLE_OBJECTIDS) {
31840ae1
ZY		 669 btrfs_set_ref_objectid(leaf, ref,
670 BTRFS_MULTIPLE_OBJECTIDS);
31840ae1
ZY		 671 }
672 ret = 0;
673 } else {
7bb86316 674 goto out;
31840ae1 675 }
7bb86316
CM		 676 btrfs_mark_buffer_dirty(path->nodes[0]);
677out:
678 btrfs_release_path(root, path);
679 return ret;
74493f7a
CM		 680}
681
31840ae1
ZY		 682static int noinline remove_extent_backref(struct btrfs_trans_handle *trans,
683 struct btrfs_root *root,
684 struct btrfs_path *path)
685{
686 struct extent_buffer *leaf;
687 struct btrfs_extent_ref *ref;
688 u32 num_refs;
689 int ret = 0;
690
691 leaf = path->nodes[0];
692 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
693 num_refs = btrfs_ref_num_refs(leaf, ref);
694 BUG_ON(num_refs == 0);
695 num_refs -= 1;
696 if (num_refs == 0) {
697 ret = btrfs_del_item(trans, root, path);
698 } else {
699 btrfs_set_ref_num_refs(leaf, ref, num_refs);
700 btrfs_mark_buffer_dirty(leaf);
701 }
702 btrfs_release_path(root, path);
703 return ret;
704}
705
706static int __btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
707 struct btrfs_root *root, u64 bytenr,
708 u64 orig_parent, u64 parent,
709 u64 orig_root, u64 ref_root,
710 u64 orig_generation, u64 ref_generation,
3bb1a1bc 711 u64 owner_objectid)
31840ae1
ZY		 712{
713 int ret;
714 struct btrfs_root *extent_root = root->fs_info->extent_root;
715 struct btrfs_path *path;
716
717 if (root == root->fs_info->extent_root) {
718 struct pending_extent_op *extent_op;
719 u64 num_bytes;
720
721 BUG_ON(owner_objectid >= BTRFS_MAX_LEVEL);
722 num_bytes = btrfs_level_size(root, (int)owner_objectid);
723 if (test_range_bit(&root->fs_info->extent_ins, bytenr,
724 bytenr + num_bytes - 1, EXTENT_LOCKED, 0)) {
725 u64 priv;
726 ret = get_state_private(&root->fs_info->extent_ins,
727 bytenr, &priv);
728 BUG_ON(ret);
729 extent_op = (struct pending_extent_op *)
730 (unsigned long)priv;
731 BUG_ON(extent_op->parent != orig_parent);
732 BUG_ON(extent_op->generation != orig_generation);
733 extent_op->parent = parent;
734 extent_op->generation = ref_generation;
735 } else {
736 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
737 BUG_ON(!extent_op);
738
739 extent_op->type = PENDING_BACKREF_UPDATE;
740 extent_op->bytenr = bytenr;
741 extent_op->num_bytes = num_bytes;
742 extent_op->parent = parent;
743 extent_op->orig_parent = orig_parent;
744 extent_op->generation = ref_generation;
745 extent_op->orig_generation = orig_generation;
746 extent_op->level = (int)owner_objectid;
747
748 set_extent_bits(&root->fs_info->extent_ins,
749 bytenr, bytenr + num_bytes - 1,
750 EXTENT_LOCKED, GFP_NOFS);
751 set_state_private(&root->fs_info->extent_ins,
752 bytenr, (unsigned long)extent_op);
753 }
754 return 0;
755 }
756
757 path = btrfs_alloc_path();
758 if (!path)
759 return -ENOMEM;
760 ret = lookup_extent_backref(trans, extent_root, path,
761 bytenr, orig_parent, orig_root,
3bb1a1bc 762 orig_generation, owner_objectid, 1);
31840ae1
ZY		 763 if (ret)
764 goto out;
765 ret = remove_extent_backref(trans, extent_root, path);
766 if (ret)
767 goto out;
768 ret = insert_extent_backref(trans, extent_root, path, bytenr,
769 parent, ref_root, ref_generation,
3bb1a1bc 770 owner_objectid);
31840ae1
ZY		 771 BUG_ON(ret);
772 finish_current_insert(trans, extent_root);
773 del_pending_extents(trans, extent_root);
774out:
775 btrfs_free_path(path);
776 return ret;
777}
778
779int btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
780 struct btrfs_root *root, u64 bytenr,
781 u64 orig_parent, u64 parent,
782 u64 ref_root, u64 ref_generation,
3bb1a1bc 783 u64 owner_objectid)
31840ae1
ZY		 784{
785 int ret;
786 if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
787 owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
788 return 0;
789 maybe_lock_mutex(root);
790 ret = __btrfs_update_extent_ref(trans, root, bytenr, orig_parent,
791 parent, ref_root, ref_root,
792 ref_generation, ref_generation,
3bb1a1bc 793 owner_objectid);
31840ae1
ZY		 794 maybe_unlock_mutex(root);
795 return ret;
796}
797
925baedd 798static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
31840ae1
ZY		 799 struct btrfs_root *root, u64 bytenr,
800 u64 orig_parent, u64 parent,
801 u64 orig_root, u64 ref_root,
802 u64 orig_generation, u64 ref_generation,
3bb1a1bc 803 u64 owner_objectid)
02217ed2 804{
5caf2a00 805 struct btrfs_path *path;
02217ed2 806 int ret;
e2fa7227 807 struct btrfs_key key;
5f39d397 808 struct extent_buffer *l;
234b63a0 809 struct btrfs_extent_item *item;
cf27e1ee 810 u32 refs;
037e6390 811
5caf2a00 812 path = btrfs_alloc_path();
54aa1f4d
CM		 813 if (!path)
814 return -ENOMEM;
26b8003f 815
3c12ac72 816 path->reada = 1;
db94535d 817 key.objectid = bytenr;
31840ae1
ZY		 818 key.type = BTRFS_EXTENT_ITEM_KEY;
819 key.offset = (u64)-1;
820
5caf2a00 821 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
9f5fae2f 822 0, 1);
54aa1f4d
CM		 823 if (ret < 0)
824 return ret;
31840ae1
ZY		 825 BUG_ON(ret == 0 || path->slots[0] == 0);
826
827 path->slots[0]--;
5f39d397 828 l = path->nodes[0];
31840ae1
ZY		 829
830 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
831 BUG_ON(key.objectid != bytenr);
832 BUG_ON(key.type != BTRFS_EXTENT_ITEM_KEY);
833
5caf2a00 834 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
5f39d397
CM		 835 refs = btrfs_extent_refs(l, item);
836 btrfs_set_extent_refs(l, item, refs + 1);
5caf2a00 837 btrfs_mark_buffer_dirty(path->nodes[0]);
a28ec197 838
5caf2a00 839 btrfs_release_path(root->fs_info->extent_root, path);
7bb86316 840
3c12ac72 841 path->reada = 1;
31840ae1
ZY		 842 ret = insert_extent_backref(trans, root->fs_info->extent_root,
843 path, bytenr, parent,
844 ref_root, ref_generation,
3bb1a1bc 845 owner_objectid);
7bb86316 846 BUG_ON(ret);
9f5fae2f 847 finish_current_insert(trans, root->fs_info->extent_root);
e20d96d6 848 del_pending_extents(trans, root->fs_info->extent_root);
74493f7a
CM		 849
850 btrfs_free_path(path);
02217ed2
CM		 851 return 0;
852}
853
925baedd 854int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
31840ae1
ZY		 855 struct btrfs_root *root,
856 u64 bytenr, u64 num_bytes, u64 parent,
857 u64 ref_root, u64 ref_generation,
3bb1a1bc 858 u64 owner_objectid)
925baedd
CM		 859{
860 int ret;
31840ae1
ZY		 861 if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
862 owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
863 return 0;
864 maybe_lock_mutex(root);
865 ret = __btrfs_inc_extent_ref(trans, root, bytenr, 0, parent,
866 0, ref_root, 0, ref_generation,
3bb1a1bc 867 owner_objectid);
31840ae1 868 maybe_unlock_mutex(root);
925baedd
CM		 869 return ret;
870}
871
e9d0b13b
CM		 872int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
873 struct btrfs_root *root)
874{
875 finish_current_insert(trans, root->fs_info->extent_root);
876 del_pending_extents(trans, root->fs_info->extent_root);
877 return 0;
878}
879
31840ae1
ZY		 880int btrfs_lookup_extent_ref(struct btrfs_trans_handle *trans,
881 struct btrfs_root *root, u64 bytenr,
882 u64 num_bytes, u32 *refs)
a28ec197 883{
5caf2a00 884 struct btrfs_path *path;
a28ec197 885 int ret;
e2fa7227 886 struct btrfs_key key;
5f39d397 887 struct extent_buffer *l;
234b63a0 888 struct btrfs_extent_item *item;
5caf2a00 889
db94535d 890 WARN_ON(num_bytes < root->sectorsize);
5caf2a00 891 path = btrfs_alloc_path();
3c12ac72 892 path->reada = 1;
db94535d
CM		 893 key.objectid = bytenr;
894 key.offset = num_bytes;
62e2749e 895 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
5caf2a00 896 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
9f5fae2f 897 0, 0);
54aa1f4d
CM		 898 if (ret < 0)
899 goto out;
5f39d397
CM		 900 if (ret != 0) {
901 btrfs_print_leaf(root, path->nodes[0]);
db94535d 902 printk("failed to find block number %Lu\n", bytenr);
a28ec197 903 BUG();
5f39d397
CM		 904 }
905 l = path->nodes[0];
5caf2a00 906 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
5f39d397 907 *refs = btrfs_extent_refs(l, item);
54aa1f4d 908out:
5caf2a00 909 btrfs_free_path(path);
a28ec197
CM		 910 return 0;
911}
912
f321e491
YZ		 913static int get_reference_status(struct btrfs_root *root, u64 bytenr,
914 u64 parent_gen, u64 ref_objectid,
915 u64 *min_generation, u32 *ref_count)
be20aa9d
CM		 916{
917 struct btrfs_root *extent_root = root->fs_info->extent_root;
918 struct btrfs_path *path;
f321e491
YZ		 919 struct extent_buffer *leaf;
920 struct btrfs_extent_ref *ref_item;
921 struct btrfs_key key;
922 struct btrfs_key found_key;
56b453c9 923 u64 root_objectid = root->root_key.objectid;
f321e491 924 u64 ref_generation;
be20aa9d
CM		 925 u32 nritems;
926 int ret;
925baedd 927
be20aa9d 928 key.objectid = bytenr;
31840ae1 929 key.offset = (u64)-1;
f321e491 930 key.type = BTRFS_EXTENT_ITEM_KEY;
be20aa9d 931
f321e491
YZ		 932 path = btrfs_alloc_path();
933 mutex_lock(&root->fs_info->alloc_mutex);
be20aa9d
CM		 934 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
935 if (ret < 0)
936 goto out;
937 BUG_ON(ret == 0);
31840ae1
ZY		 938 if (ret < 0 || path->slots[0] == 0)
939 goto out;
be20aa9d 940
31840ae1 941 path->slots[0]--;
f321e491
YZ		 942 leaf = path->nodes[0];
943 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
be20aa9d
CM		 944
945 if (found_key.objectid != bytenr ||
946 found_key.type != BTRFS_EXTENT_ITEM_KEY) {
f321e491 947 ret = 1;
be20aa9d
CM		 948 goto out;
949 }
950
f321e491
YZ		 951 *ref_count = 0;
952 *min_generation = (u64)-1;
953
be20aa9d 954 while (1) {
f321e491
YZ		 955 leaf = path->nodes[0];
956 nritems = btrfs_header_nritems(leaf);
be20aa9d
CM		 957 if (path->slots[0] >= nritems) {
958 ret = btrfs_next_leaf(extent_root, path);
f321e491
YZ		 959 if (ret < 0)
960 goto out;
be20aa9d
CM		 961 if (ret == 0)
962 continue;
963 break;
964 }
f321e491 965 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
be20aa9d
CM		 966 if (found_key.objectid != bytenr)
967 break;
bd09835d 968
be20aa9d
CM		 969 if (found_key.type != BTRFS_EXTENT_REF_KEY) {
970 path->slots[0]++;
971 continue;
972 }
973
f321e491 974 ref_item = btrfs_item_ptr(leaf, path->slots[0],
be20aa9d 975 struct btrfs_extent_ref);
f321e491
YZ		 976 ref_generation = btrfs_ref_generation(leaf, ref_item);
977 /*
31840ae1 978 * For (parent_gen > 0 && parent_gen > ref_generation):
f321e491 979 *
bcc63abb
Y		 980 * we reach here through the oldest root, therefore
981 * all other reference from same snapshot should have
f321e491
YZ		 982 * a larger generation.
983 */
984 if ((root_objectid != btrfs_ref_root(leaf, ref_item)) ||
985 (parent_gen > 0 && parent_gen > ref_generation) ||
986 (ref_objectid >= BTRFS_FIRST_FREE_OBJECTID &&
987 ref_objectid != btrfs_ref_objectid(leaf, ref_item))) {
31840ae1 988 *ref_count = 2;
f321e491 989 break;
a68d5933 990 }
f321e491
YZ		 991
992 *ref_count = 1;
993 if (*min_generation > ref_generation)
994 *min_generation = ref_generation;
995
be20aa9d
CM		 996 path->slots[0]++;
997 }
f321e491
YZ		 998 ret = 0;
999out:
1000 mutex_unlock(&root->fs_info->alloc_mutex);
1001 btrfs_free_path(path);
1002 return ret;
1003}
1004
7ea394f1
YZ		 1005int btrfs_cross_ref_exists(struct btrfs_trans_handle *trans,
1006 struct btrfs_root *root,
f321e491
YZ		 1007 struct btrfs_key *key, u64 bytenr)
1008{
f321e491
YZ		 1009 struct btrfs_root *old_root;
1010 struct btrfs_path *path = NULL;
1011 struct extent_buffer *eb;
1012 struct btrfs_file_extent_item *item;
1013 u64 ref_generation;
1014 u64 min_generation;
1015 u64 extent_start;
1016 u32 ref_count;
1017 int level;
1018 int ret;
1019
7ea394f1 1020 BUG_ON(trans == NULL);
f321e491
YZ		 1021 BUG_ON(key->type != BTRFS_EXTENT_DATA_KEY);
1022 ret = get_reference_status(root, bytenr, 0, key->objectid,
1023 &min_generation, &ref_count);
1024 if (ret)
1025 return ret;
1026
1027 if (ref_count != 1)
1028 return 1;
1029
f321e491
YZ		 1030 old_root = root->dirty_root->root;
1031 ref_generation = old_root->root_key.offset;
1032
1033 /* all references are created in running transaction */
1034 if (min_generation > ref_generation) {
1035 ret = 0;
bbaf549e
CM		 1036 goto out;
1037 }
f321e491
YZ		 1038
1039 path = btrfs_alloc_path();
1040 if (!path) {
1041 ret = -ENOMEM;
be20aa9d
CM		 1042 goto out;
1043 }
f321e491
YZ		 1044
1045 path->skip_locking = 1;
1046 /* if no item found, the extent is referenced by other snapshot */
1047 ret = btrfs_search_slot(NULL, old_root, key, path, 0, 0);
1048 if (ret)
be20aa9d 1049 goto out;
be20aa9d 1050
f321e491
YZ		 1051 eb = path->nodes[0];
1052 item = btrfs_item_ptr(eb, path->slots[0],
1053 struct btrfs_file_extent_item);
1054 if (btrfs_file_extent_type(eb, item) != BTRFS_FILE_EXTENT_REG ||
1055 btrfs_file_extent_disk_bytenr(eb, item) != bytenr) {
1056 ret = 1;
1057 goto out;
1058 }
1059
1060 for (level = BTRFS_MAX_LEVEL - 1; level >= -1; level--) {
1061 if (level >= 0) {
1062 eb = path->nodes[level];
1063 if (!eb)
1064 continue;
1065 extent_start = eb->start;
bcc63abb 1066 } else
f321e491
YZ		 1067 extent_start = bytenr;
1068
1069 ret = get_reference_status(root, extent_start, ref_generation,
1070 0, &min_generation, &ref_count);
1071 if (ret)
1072 goto out;
1073
1074 if (ref_count != 1) {
1075 ret = 1;
1076 goto out;
1077 }
1078 if (level >= 0)
1079 ref_generation = btrfs_header_generation(eb);
1080 }
1081 ret = 0;
be20aa9d 1082out:
f321e491
YZ		 1083 if (path)
1084 btrfs_free_path(path);
f321e491 1085 return ret;
be20aa9d 1086}
c5739bba 1087
31840ae1
ZY		 1088int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1089 struct extent_buffer *buf, u32 nr_extents)
02217ed2 1090{
5f39d397 1091 struct btrfs_key key;
6407bf6d 1092 struct btrfs_file_extent_item *fi;
e4657689
ZY		 1093 u64 root_gen;
1094 u32 nritems;
02217ed2 1095 int i;
db94535d 1096 int level;
31840ae1 1097 int ret = 0;
e4657689 1098 int shared = 0;
a28ec197 1099
3768f368 1100 if (!root->ref_cows)
a28ec197 1101 return 0;
5f39d397 1102
e4657689
ZY		 1103 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
1104 shared = 0;
1105 root_gen = root->root_key.offset;
1106 } else {
1107 shared = 1;
1108 root_gen = trans->transid - 1;
1109 }
1110
db94535d 1111 level = btrfs_header_level(buf);
5f39d397 1112 nritems = btrfs_header_nritems(buf);
4a096752 1113
31840ae1 1114 if (level == 0) {
31153d81
YZ		 1115 struct btrfs_leaf_ref *ref;
1116 struct btrfs_extent_info *info;
1117
31840ae1 1118 ref = btrfs_alloc_leaf_ref(root, nr_extents);
31153d81 1119 if (!ref) {
31840ae1 1120 ret = -ENOMEM;
31153d81
YZ		 1121 goto out;
1122 }
1123
e4657689 1124 ref->root_gen = root_gen;
31153d81
YZ		 1125 ref->bytenr = buf->start;
1126 ref->owner = btrfs_header_owner(buf);
1127 ref->generation = btrfs_header_generation(buf);
31840ae1 1128 ref->nritems = nr_extents;
31153d81 1129 info = ref->extents;
bcc63abb 1130
31840ae1 1131 for (i = 0; nr_extents > 0 && i < nritems; i++) {
31153d81
YZ		 1132 u64 disk_bytenr;
1133 btrfs_item_key_to_cpu(buf, &key, i);
1134 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1135 continue;
1136 fi = btrfs_item_ptr(buf, i,
1137 struct btrfs_file_extent_item);
1138 if (btrfs_file_extent_type(buf, fi) ==
1139 BTRFS_FILE_EXTENT_INLINE)
1140 continue;
1141 disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1142 if (disk_bytenr == 0)
1143 continue;
1144
1145 info->bytenr = disk_bytenr;
1146 info->num_bytes =
1147 btrfs_file_extent_disk_num_bytes(buf, fi);
1148 info->objectid = key.objectid;
1149 info->offset = key.offset;
1150 info++;
1151 }
1152
e4657689 1153 ret = btrfs_add_leaf_ref(root, ref, shared);
5b84e8d6
YZ		 1154 if (ret == -EEXIST && shared) {
1155 struct btrfs_leaf_ref *old;
1156 old = btrfs_lookup_leaf_ref(root, ref->bytenr);
1157 BUG_ON(!old);
1158 btrfs_remove_leaf_ref(root, old);
1159 btrfs_free_leaf_ref(root, old);
1160 ret = btrfs_add_leaf_ref(root, ref, shared);
1161 }
31153d81 1162 WARN_ON(ret);
bcc63abb 1163 btrfs_free_leaf_ref(root, ref);
31153d81
YZ		 1164 }
1165out:
31840ae1
ZY		 1166 return ret;
1167}
1168
1169int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1170 struct extent_buffer *orig_buf, struct extent_buffer *buf,
1171 u32 *nr_extents)
1172{
1173 u64 bytenr;
1174 u64 ref_root;
1175 u64 orig_root;
1176 u64 ref_generation;
1177 u64 orig_generation;
1178 u32 nritems;
1179 u32 nr_file_extents = 0;
1180 struct btrfs_key key;
1181 struct btrfs_file_extent_item *fi;
1182 int i;
1183 int level;
1184 int ret = 0;
1185 int faili = 0;
1186 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
3bb1a1bc 1187 u64, u64, u64, u64, u64, u64, u64, u64);
31840ae1
ZY		 1188
1189 ref_root = btrfs_header_owner(buf);
1190 ref_generation = btrfs_header_generation(buf);
1191 orig_root = btrfs_header_owner(orig_buf);
1192 orig_generation = btrfs_header_generation(orig_buf);
1193
1194 nritems = btrfs_header_nritems(buf);
1195 level = btrfs_header_level(buf);
1196
1197 if (root->ref_cows) {
1198 process_func = __btrfs_inc_extent_ref;
1199 } else {
1200 if (level == 0 &&
1201 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
1202 goto out;
1203 if (level != 0 &&
1204 root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
1205 goto out;
1206 process_func = __btrfs_update_extent_ref;
1207 }
1208
1209 for (i = 0; i < nritems; i++) {
1210 cond_resched();
db94535d 1211 if (level == 0) {
5f39d397
CM		 1212 btrfs_item_key_to_cpu(buf, &key, i);
1213 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
54aa1f4d 1214 continue;
5f39d397 1215 fi = btrfs_item_ptr(buf, i,
54aa1f4d 1216 struct btrfs_file_extent_item);
5f39d397 1217 if (btrfs_file_extent_type(buf, fi) ==
54aa1f4d
CM		 1218 BTRFS_FILE_EXTENT_INLINE)
1219 continue;
31840ae1
ZY		 1220 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1221 if (bytenr == 0)
54aa1f4d 1222 continue;
31840ae1
ZY		 1223
1224 nr_file_extents++;
1225
1226 maybe_lock_mutex(root);
1227 ret = process_func(trans, root, bytenr,
1228 orig_buf->start, buf->start,
1229 orig_root, ref_root,
1230 orig_generation, ref_generation,
3bb1a1bc 1231 key.objectid);
31840ae1
ZY		 1232 maybe_unlock_mutex(root);
1233
1234 if (ret) {
1235 faili = i;
1236 WARN_ON(1);
1237 goto fail;
1238 }
54aa1f4d 1239 } else {
db94535d 1240 bytenr = btrfs_node_blockptr(buf, i);
31840ae1
ZY		 1241 maybe_lock_mutex(root);
1242 ret = process_func(trans, root, bytenr,
1243 orig_buf->start, buf->start,
1244 orig_root, ref_root,
1245 orig_generation, ref_generation,
3bb1a1bc 1246 level - 1);
31840ae1
ZY		 1247 maybe_unlock_mutex(root);
1248 if (ret) {
1249 faili = i;
1250 WARN_ON(1);
1251 goto fail;
1252 }
54aa1f4d
CM		 1253 }
1254 }
31840ae1
ZY		 1255out:
1256 if (nr_extents) {
1257 if (level == 0)
1258 *nr_extents = nr_file_extents;
1259 else
1260 *nr_extents = nritems;
1261 }
1262 return 0;
1263fail:
1264 WARN_ON(1);
54aa1f4d 1265 return ret;
02217ed2
CM		 1266}
1267
31840ae1
ZY		 1268int btrfs_update_ref(struct btrfs_trans_handle *trans,
1269 struct btrfs_root *root, struct extent_buffer *orig_buf,
1270 struct extent_buffer *buf, int start_slot, int nr)
1271
1272{
1273 u64 bytenr;
1274 u64 ref_root;
1275 u64 orig_root;
1276 u64 ref_generation;
1277 u64 orig_generation;
1278 struct btrfs_key key;
1279 struct btrfs_file_extent_item *fi;
1280 int i;
1281 int ret;
1282 int slot;
1283 int level;
1284
1285 BUG_ON(start_slot < 0);
1286 BUG_ON(start_slot + nr > btrfs_header_nritems(buf));
1287
1288 ref_root = btrfs_header_owner(buf);
1289 ref_generation = btrfs_header_generation(buf);
1290 orig_root = btrfs_header_owner(orig_buf);
1291 orig_generation = btrfs_header_generation(orig_buf);
1292 level = btrfs_header_level(buf);
1293
1294 if (!root->ref_cows) {
1295 if (level == 0 &&
1296 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
1297 return 0;
1298 if (level != 0 &&
1299 root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
1300 return 0;
1301 }
1302
1303 for (i = 0, slot = start_slot; i < nr; i++, slot++) {
1304 cond_resched();
1305 if (level == 0) {
1306 btrfs_item_key_to_cpu(buf, &key, slot);
1307 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1308 continue;
1309 fi = btrfs_item_ptr(buf, slot,
1310 struct btrfs_file_extent_item);
1311 if (btrfs_file_extent_type(buf, fi) ==
1312 BTRFS_FILE_EXTENT_INLINE)
1313 continue;
1314 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1315 if (bytenr == 0)
1316 continue;
1317 maybe_lock_mutex(root);
1318 ret = __btrfs_update_extent_ref(trans, root, bytenr,
1319 orig_buf->start, buf->start,
1320 orig_root, ref_root,
1321 orig_generation, ref_generation,
3bb1a1bc 1322 key.objectid);
31840ae1
ZY		 1323 maybe_unlock_mutex(root);
1324 if (ret)
1325 goto fail;
1326 } else {
1327 bytenr = btrfs_node_blockptr(buf, slot);
1328 maybe_lock_mutex(root);
1329 ret = __btrfs_update_extent_ref(trans, root, bytenr,
1330 orig_buf->start, buf->start,
1331 orig_root, ref_root,
1332 orig_generation, ref_generation,
3bb1a1bc 1333 level - 1);
31840ae1
ZY		 1334 maybe_unlock_mutex(root);
1335 if (ret)
1336 goto fail;
1337 }
1338 }
1339 return 0;
1340fail:
1341 WARN_ON(1);
1342 return -1;
1343}
1344
9078a3e1
CM		 1345static int write_one_cache_group(struct btrfs_trans_handle *trans,
1346 struct btrfs_root *root,
1347 struct btrfs_path *path,
1348 struct btrfs_block_group_cache *cache)
1349{
1350 int ret;
1351 int pending_ret;
1352 struct btrfs_root *extent_root = root->fs_info->extent_root;
5f39d397
CM		 1353 unsigned long bi;
1354 struct extent_buffer *leaf;
9078a3e1 1355
9078a3e1 1356 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
54aa1f4d
CM		 1357 if (ret < 0)
1358 goto fail;
9078a3e1 1359 BUG_ON(ret);
5f39d397
CM		 1360
1361 leaf = path->nodes[0];
1362 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
1363 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
1364 btrfs_mark_buffer_dirty(leaf);
9078a3e1 1365 btrfs_release_path(extent_root, path);
54aa1f4d 1366fail:
9078a3e1
CM		 1367 finish_current_insert(trans, extent_root);
1368 pending_ret = del_pending_extents(trans, extent_root);
1369 if (ret)
1370 return ret;
1371 if (pending_ret)
1372 return pending_ret;
1373 return 0;
1374
1375}
1376
96b5179d
CM		 1377int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
1378 struct btrfs_root *root)
9078a3e1 1379{
0f9dd46c
JB		 1380 struct btrfs_block_group_cache *cache, *entry;
1381 struct rb_node *n;
9078a3e1
CM		 1382 int err = 0;
1383 int werr = 0;
9078a3e1 1384 struct btrfs_path *path;
96b5179d 1385 u64 last = 0;
9078a3e1
CM		 1386
1387 path = btrfs_alloc_path();
1388 if (!path)
1389 return -ENOMEM;
1390
925baedd 1391 mutex_lock(&root->fs_info->alloc_mutex);
9078a3e1 1392 while(1) {
0f9dd46c
JB		 1393 cache = NULL;
1394 spin_lock(&root->fs_info->block_group_cache_lock);
1395 for (n = rb_first(&root->fs_info->block_group_cache_tree);
1396 n; n = rb_next(n)) {
1397 entry = rb_entry(n, struct btrfs_block_group_cache,
1398 cache_node);
1399 if (entry->dirty) {
1400 cache = entry;
1401 break;
1402 }
1403 }
1404 spin_unlock(&root->fs_info->block_group_cache_lock);
54aa1f4d 1405
0f9dd46c 1406 if (!cache)
96b5179d 1407 break;
0f9dd46c 1408
e8569813 1409 cache->dirty = 0;
0f9dd46c
JB		 1410 last += cache->key.offset;
1411
96b5179d
CM		 1412 err = write_one_cache_group(trans, root,
1413 path, cache);
1414 /*
1415 * if we fail to write the cache group, we want
1416 * to keep it marked dirty in hopes that a later
1417 * write will work
1418 */
1419 if (err) {
1420 werr = err;
1421 continue;
9078a3e1
CM		 1422 }
1423 }
1424 btrfs_free_path(path);
925baedd 1425 mutex_unlock(&root->fs_info->alloc_mutex);
9078a3e1
CM		 1426 return werr;
1427}
1428
593060d7
CM		 1429static int update_space_info(struct btrfs_fs_info *info, u64 flags,
1430 u64 total_bytes, u64 bytes_used,
1431 struct btrfs_space_info **space_info)
1432{
1433 struct btrfs_space_info *found;
1434
1435 found = __find_space_info(info, flags);
1436 if (found) {
1437 found->total_bytes += total_bytes;
1438 found->bytes_used += bytes_used;
8f18cf13 1439 found->full = 0;
593060d7
CM		 1440 *space_info = found;
1441 return 0;
1442 }
1443 found = kmalloc(sizeof(*found), GFP_NOFS);
1444 if (!found)
1445 return -ENOMEM;
1446
1447 list_add(&found->list, &info->space_info);
0f9dd46c
JB		 1448 INIT_LIST_HEAD(&found->block_groups);
1449 spin_lock_init(&found->lock);
593060d7
CM		 1450 found->flags = flags;
1451 found->total_bytes = total_bytes;
1452 found->bytes_used = bytes_used;
1453 found->bytes_pinned = 0;
e8569813 1454 found->bytes_reserved = 0;
593060d7 1455 found->full = 0;
0ef3e66b 1456 found->force_alloc = 0;
593060d7
CM		 1457 *space_info = found;
1458 return 0;
1459}
1460
8790d502
CM		 1461static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
1462{
1463 u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
611f0e00 1464 BTRFS_BLOCK_GROUP_RAID1 |
321aecc6 1465 BTRFS_BLOCK_GROUP_RAID10 |
611f0e00 1466 BTRFS_BLOCK_GROUP_DUP);
8790d502
CM		 1467 if (extra_flags) {
1468 if (flags & BTRFS_BLOCK_GROUP_DATA)
1469 fs_info->avail_data_alloc_bits |= extra_flags;
1470 if (flags & BTRFS_BLOCK_GROUP_METADATA)
1471 fs_info->avail_metadata_alloc_bits |= extra_flags;
1472 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
1473 fs_info->avail_system_alloc_bits |= extra_flags;
1474 }
1475}
593060d7 1476
a061fc8d 1477static u64 reduce_alloc_profile(struct btrfs_root *root, u64 flags)
ec44a35c 1478{
a061fc8d
CM		 1479 u64 num_devices = root->fs_info->fs_devices->num_devices;
1480
1481 if (num_devices == 1)
1482 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
1483 if (num_devices < 4)
1484 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
1485
ec44a35c
CM		 1486 if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
1487 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
a061fc8d 1488 BTRFS_BLOCK_GROUP_RAID10))) {
ec44a35c 1489 flags &= ~BTRFS_BLOCK_GROUP_DUP;
a061fc8d 1490 }
ec44a35c
CM		 1491
1492 if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
a061fc8d 1493 (flags & BTRFS_BLOCK_GROUP_RAID10)) {
ec44a35c 1494 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
a061fc8d 1495 }
ec44a35c
CM		 1496
1497 if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
1498 ((flags & BTRFS_BLOCK_GROUP_RAID1) |
1499 (flags & BTRFS_BLOCK_GROUP_RAID10) |
1500 (flags & BTRFS_BLOCK_GROUP_DUP)))
1501 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
1502 return flags;
1503}
1504
6324fbf3
CM		 1505static int do_chunk_alloc(struct btrfs_trans_handle *trans,
1506 struct btrfs_root *extent_root, u64 alloc_bytes,
0ef3e66b 1507 u64 flags, int force)
6324fbf3
CM		 1508{
1509 struct btrfs_space_info *space_info;
1510 u64 thresh;
1511 u64 start;
1512 u64 num_bytes;
cf749823 1513 int ret = 0, waited = 0;
6324fbf3 1514
a061fc8d 1515 flags = reduce_alloc_profile(extent_root, flags);
ec44a35c 1516
6324fbf3 1517 space_info = __find_space_info(extent_root->fs_info, flags);
593060d7
CM		 1518 if (!space_info) {
1519 ret = update_space_info(extent_root->fs_info, flags,
1520 0, 0, &space_info);
1521 BUG_ON(ret);
1522 }
6324fbf3
CM		 1523 BUG_ON(!space_info);
1524
0ef3e66b
CM		 1525 if (space_info->force_alloc) {
1526 force = 1;
1527 space_info->force_alloc = 0;
1528 }
6324fbf3 1529 if (space_info->full)
925baedd 1530 goto out;
6324fbf3 1531
8790d502 1532 thresh = div_factor(space_info->total_bytes, 6);
0ef3e66b 1533 if (!force &&
e8569813
ZY		 1534 (space_info->bytes_used + space_info->bytes_pinned +
1535 space_info->bytes_reserved + alloc_bytes) < thresh)
925baedd 1536 goto out;
6324fbf3 1537
cf749823
JB		 1538 while (!mutex_trylock(&extent_root->fs_info->chunk_mutex)) {
1539 if (!force)
1540 goto out;
1541 mutex_unlock(&extent_root->fs_info->alloc_mutex);
1542 cond_resched();
1543 mutex_lock(&extent_root->fs_info->alloc_mutex);
1544 waited = 1;
1545 }
1546
1547 if (waited && space_info->full)
1548 goto out_unlock;
1549
6324fbf3
CM		 1550 ret = btrfs_alloc_chunk(trans, extent_root, &start, &num_bytes, flags);
1551 if (ret == -ENOSPC) {
1552printk("space info full %Lu\n", flags);
1553 space_info->full = 1;
a74a4b97 1554 goto out_unlock;
6324fbf3 1555 }
6324fbf3
CM		 1556 BUG_ON(ret);
1557
1558 ret = btrfs_make_block_group(trans, extent_root, 0, flags,
e17cade2 1559 BTRFS_FIRST_CHUNK_TREE_OBJECTID, start, num_bytes);
6324fbf3 1560 BUG_ON(ret);
0f9dd46c 1561
a74a4b97 1562out_unlock:
333db94c 1563 mutex_unlock(&extent_root->fs_info->chunk_mutex);
a74a4b97 1564out:
0f9dd46c 1565 return ret;
6324fbf3
CM		 1566}
1567
9078a3e1
CM		 1568static int update_block_group(struct btrfs_trans_handle *trans,
1569 struct btrfs_root *root,
db94535d 1570 u64 bytenr, u64 num_bytes, int alloc,
0b86a832 1571 int mark_free)
9078a3e1
CM		 1572{
1573 struct btrfs_block_group_cache *cache;
1574 struct btrfs_fs_info *info = root->fs_info;
db94535d 1575 u64 total = num_bytes;
9078a3e1 1576 u64 old_val;
db94535d 1577 u64 byte_in_group;
3e1ad54f 1578
7d9eb12c 1579 WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex));
9078a3e1 1580 while(total) {
db94535d 1581 cache = btrfs_lookup_block_group(info, bytenr);
3e1ad54f 1582 if (!cache) {
9078a3e1 1583 return -1;
cd1bc465 1584 }
db94535d
CM		 1585 byte_in_group = bytenr - cache->key.objectid;
1586 WARN_ON(byte_in_group > cache->key.offset);
9078a3e1 1587
c286ac48 1588 spin_lock(&cache->lock);
0f9dd46c 1589 cache->dirty = 1;
9078a3e1 1590 old_val = btrfs_block_group_used(&cache->item);
db94535d 1591 num_bytes = min(total, cache->key.offset - byte_in_group);
cd1bc465 1592 if (alloc) {
db94535d 1593 old_val += num_bytes;
6324fbf3 1594 cache->space_info->bytes_used += num_bytes;
c286ac48
CM		 1595 btrfs_set_block_group_used(&cache->item, old_val);
1596 spin_unlock(&cache->lock);
cd1bc465 1597 } else {
db94535d 1598 old_val -= num_bytes;
6324fbf3 1599 cache->space_info->bytes_used -= num_bytes;
c286ac48
CM		 1600 btrfs_set_block_group_used(&cache->item, old_val);
1601 spin_unlock(&cache->lock);
f510cfec 1602 if (mark_free) {
0f9dd46c
JB		 1603 int ret;
1604 ret = btrfs_add_free_space(cache, bytenr,
1605 num_bytes);
1606 if (ret)
1607 return -1;
e37c9e69 1608 }
cd1bc465 1609 }
db94535d
CM		 1610 total -= num_bytes;
1611 bytenr += num_bytes;
9078a3e1
CM		 1612 }
1613 return 0;
1614}
6324fbf3 1615
a061fc8d
CM		 1616static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
1617{
0f9dd46c
JB		 1618 struct btrfs_block_group_cache *cache;
1619
1620 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
1621 if (!cache)
a061fc8d 1622 return 0;
0f9dd46c
JB		 1623
1624 return cache->key.objectid;
a061fc8d
CM		 1625}
1626
e02119d5 1627int btrfs_update_pinned_extents(struct btrfs_root *root,
324ae4df
Y		 1628 u64 bytenr, u64 num, int pin)
1629{
1630 u64 len;
1631 struct btrfs_block_group_cache *cache;
1632 struct btrfs_fs_info *fs_info = root->fs_info;
1633
7d9eb12c 1634 WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex));
324ae4df
Y		 1635 if (pin) {
1636 set_extent_dirty(&fs_info->pinned_extents,
1637 bytenr, bytenr + num - 1, GFP_NOFS);
1638 } else {
1639 clear_extent_dirty(&fs_info->pinned_extents,
1640 bytenr, bytenr + num - 1, GFP_NOFS);
1641 }
1642 while (num > 0) {
1643 cache = btrfs_lookup_block_group(fs_info, bytenr);
e8569813
ZY		 1644 BUG_ON(!cache);
1645 len = min(num, cache->key.offset -
1646 (bytenr - cache->key.objectid));
324ae4df 1647 if (pin) {
e8569813
ZY		 1648 spin_lock(&cache->lock);
1649 cache->pinned += len;
1650 cache->space_info->bytes_pinned += len;
1651 spin_unlock(&cache->lock);
324ae4df
Y		 1652 fs_info->total_pinned += len;
1653 } else {
e8569813
ZY		 1654 spin_lock(&cache->lock);
1655 cache->pinned -= len;
1656 cache->space_info->bytes_pinned -= len;
1657 spin_unlock(&cache->lock);
324ae4df
Y		 1658 fs_info->total_pinned -= len;
1659 }
1660 bytenr += len;
1661 num -= len;
1662 }
1663 return 0;
1664}
9078a3e1 1665
e8569813
ZY		 1666static int update_reserved_extents(struct btrfs_root *root,
1667 u64 bytenr, u64 num, int reserve)
1668{
1669 u64 len;
1670 struct btrfs_block_group_cache *cache;
1671 struct btrfs_fs_info *fs_info = root->fs_info;
1672
1673 WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex));
1674 while (num > 0) {
1675 cache = btrfs_lookup_block_group(fs_info, bytenr);
1676 BUG_ON(!cache);
1677 len = min(num, cache->key.offset -
1678 (bytenr - cache->key.objectid));
1679 if (reserve) {
1680 spin_lock(&cache->lock);
1681 cache->reserved += len;
1682 cache->space_info->bytes_reserved += len;
1683 spin_unlock(&cache->lock);
1684 } else {
1685 spin_lock(&cache->lock);
1686 cache->reserved -= len;
1687 cache->space_info->bytes_reserved -= len;
1688 spin_unlock(&cache->lock);
1689 }
1690 bytenr += len;
1691 num -= len;
1692 }
1693 return 0;
1694}
1695
d1310b2e 1696int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
ccd467d6 1697{
ccd467d6 1698 u64 last = 0;
1a5bc167
CM		 1699 u64 start;
1700 u64 end;
d1310b2e 1701 struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
ccd467d6 1702 int ret;
ccd467d6
CM		 1703
1704 while(1) {
1a5bc167
CM		 1705 ret = find_first_extent_bit(pinned_extents, last,
1706 &start, &end, EXTENT_DIRTY);
1707 if (ret)
ccd467d6 1708 break;
1a5bc167
CM		 1709 set_extent_dirty(copy, start, end, GFP_NOFS);
1710 last = end + 1;
ccd467d6
CM		 1711 }
1712 return 0;
1713}
1714
1715int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
1716 struct btrfs_root *root,
d1310b2e 1717 struct extent_io_tree *unpin)
a28ec197 1718{
1a5bc167
CM		 1719 u64 start;
1720 u64 end;
a28ec197 1721 int ret;
0f9dd46c 1722 struct btrfs_block_group_cache *cache;
a28ec197 1723
925baedd 1724 mutex_lock(&root->fs_info->alloc_mutex);
a28ec197 1725 while(1) {
1a5bc167
CM		 1726 ret = find_first_extent_bit(unpin, 0, &start, &end,
1727 EXTENT_DIRTY);
1728 if (ret)
a28ec197 1729 break;
e02119d5 1730 btrfs_update_pinned_extents(root, start, end + 1 - start, 0);
1a5bc167 1731 clear_extent_dirty(unpin, start, end, GFP_NOFS);
0f9dd46c
JB		 1732 cache = btrfs_lookup_block_group(root->fs_info, start);
1733 if (cache->cached)
1734 btrfs_add_free_space(cache, start, end - start + 1);
c286ac48
CM		 1735 if (need_resched()) {
1736 mutex_unlock(&root->fs_info->alloc_mutex);
1737 cond_resched();
1738 mutex_lock(&root->fs_info->alloc_mutex);
1739 }
a28ec197 1740 }
925baedd 1741 mutex_unlock(&root->fs_info->alloc_mutex);
a28ec197
CM		 1742 return 0;
1743}
1744
98ed5174
CM		 1745static int finish_current_insert(struct btrfs_trans_handle *trans,
1746 struct btrfs_root *extent_root)
037e6390 1747{
7bb86316
CM		 1748 u64 start;
1749 u64 end;
31840ae1 1750 u64 priv;
7bb86316
CM		 1751 struct btrfs_fs_info *info = extent_root->fs_info;
1752 struct btrfs_path *path;
31840ae1
ZY		 1753 struct btrfs_extent_ref *ref;
1754 struct pending_extent_op *extent_op;
1755 struct btrfs_key key;
234b63a0 1756 struct btrfs_extent_item extent_item;
037e6390 1757 int ret;
1a5bc167 1758 int err = 0;
037e6390 1759
7d9eb12c 1760 WARN_ON(!mutex_is_locked(&extent_root->fs_info->alloc_mutex));
5f39d397 1761 btrfs_set_stack_extent_refs(&extent_item, 1);
7bb86316 1762 path = btrfs_alloc_path();
037e6390 1763
26b8003f 1764 while(1) {
1a5bc167
CM		 1765 ret = find_first_extent_bit(&info->extent_ins, 0, &start,
1766 &end, EXTENT_LOCKED);
1767 if (ret)
26b8003f
CM		 1768 break;
1769
31840ae1
ZY		 1770 ret = get_state_private(&info->extent_ins, start, &priv);
1771 BUG_ON(ret);
1772 extent_op = (struct pending_extent_op *)(unsigned long)priv;
1773
1774 if (extent_op->type == PENDING_EXTENT_INSERT) {
1775 key.objectid = start;
1776 key.offset = end + 1 - start;
1777 key.type = BTRFS_EXTENT_ITEM_KEY;
1778 err = btrfs_insert_item(trans, extent_root, &key,
1a5bc167 1779 &extent_item, sizeof(extent_item));
31840ae1 1780 BUG_ON(err);
c286ac48 1781
31840ae1
ZY		 1782 clear_extent_bits(&info->extent_ins, start, end,
1783 EXTENT_LOCKED, GFP_NOFS);
c286ac48 1784
31840ae1
ZY		 1785 err = insert_extent_backref(trans, extent_root, path,
1786 start, extent_op->parent,
1787 extent_root->root_key.objectid,
1788 extent_op->generation,
3bb1a1bc 1789 extent_op->level);
31840ae1
ZY		 1790 BUG_ON(err);
1791 } else if (extent_op->type == PENDING_BACKREF_UPDATE) {
1792 err = lookup_extent_backref(trans, extent_root, path,
1793 start, extent_op->orig_parent,
1794 extent_root->root_key.objectid,
3bb1a1bc
YZ		 1795 extent_op->orig_generation,
1796 extent_op->level, 0);
31840ae1 1797 BUG_ON(err);
c286ac48 1798
31840ae1
ZY		 1799 clear_extent_bits(&info->extent_ins, start, end,
1800 EXTENT_LOCKED, GFP_NOFS);
1801
1802 key.objectid = start;
1803 key.offset = extent_op->parent;
1804 key.type = BTRFS_EXTENT_REF_KEY;
1805 err = btrfs_set_item_key_safe(trans, extent_root, path,
1806 &key);
1807 BUG_ON(err);
1808 ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1809 struct btrfs_extent_ref);
1810 btrfs_set_ref_generation(path->nodes[0], ref,
1811 extent_op->generation);
1812 btrfs_mark_buffer_dirty(path->nodes[0]);
1813 btrfs_release_path(extent_root, path);
d8d5f3e1 1814 } else {
31840ae1 1815 BUG_ON(1);
d8d5f3e1 1816 }
31840ae1
ZY		 1817 kfree(extent_op);
1818
c286ac48
CM		 1819 if (need_resched()) {
1820 mutex_unlock(&extent_root->fs_info->alloc_mutex);
1821 cond_resched();
1822 mutex_lock(&extent_root->fs_info->alloc_mutex);
1823 }
037e6390 1824 }
7bb86316 1825 btrfs_free_path(path);
037e6390
CM		 1826 return 0;
1827}
1828
31840ae1
ZY		 1829static int pin_down_bytes(struct btrfs_trans_handle *trans,
1830 struct btrfs_root *root,
1831 u64 bytenr, u64 num_bytes, int is_data)
e20d96d6 1832{
1a5bc167 1833 int err = 0;
31840ae1 1834 struct extent_buffer *buf;
8ef97622 1835
7d9eb12c 1836 WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex));
31840ae1
ZY		 1837 if (is_data)
1838 goto pinit;
1839
1840 buf = btrfs_find_tree_block(root, bytenr, num_bytes);
1841 if (!buf)
1842 goto pinit;
1843
1844 /* we can reuse a block if it hasn't been written
1845 * and it is from this transaction. We can't
1846 * reuse anything from the tree log root because
1847 * it has tiny sub-transactions.
1848 */
1849 if (btrfs_buffer_uptodate(buf, 0) &&
1850 btrfs_try_tree_lock(buf)) {
1851 u64 header_owner = btrfs_header_owner(buf);
1852 u64 header_transid = btrfs_header_generation(buf);
1853 if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
1a40e23b 1854 header_owner != BTRFS_TREE_RELOC_OBJECTID &&
31840ae1
ZY		 1855 header_transid == trans->transid &&
1856 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
1857 clean_tree_block(NULL, root, buf);
1858 btrfs_tree_unlock(buf);
5f39d397 1859 free_extent_buffer(buf);
31840ae1 1860 return 1;
8ef97622 1861 }
31840ae1 1862 btrfs_tree_unlock(buf);
f4b9aa8d 1863 }
31840ae1
ZY		 1864 free_extent_buffer(buf);
1865pinit:
1866 btrfs_update_pinned_extents(root, bytenr, num_bytes, 1);
1867
be744175 1868 BUG_ON(err < 0);
e20d96d6
CM		 1869 return 0;
1870}
1871
fec577fb 1872/*
a28ec197 1873 * remove an extent from the root, returns 0 on success
fec577fb 1874 */
31840ae1
ZY		 1875static int __free_extent(struct btrfs_trans_handle *trans,
1876 struct btrfs_root *root,
1877 u64 bytenr, u64 num_bytes, u64 parent,
7bb86316 1878 u64 root_objectid, u64 ref_generation,
3bb1a1bc 1879 u64 owner_objectid, int pin, int mark_free)
a28ec197 1880{
5caf2a00 1881 struct btrfs_path *path;
e2fa7227 1882 struct btrfs_key key;
1261ec42
CM		 1883 struct btrfs_fs_info *info = root->fs_info;
1884 struct btrfs_root *extent_root = info->extent_root;
5f39d397 1885 struct extent_buffer *leaf;
a28ec197 1886 int ret;
952fccac
CM		 1887 int extent_slot = 0;
1888 int found_extent = 0;
1889 int num_to_del = 1;
234b63a0 1890 struct btrfs_extent_item *ei;
cf27e1ee 1891 u32 refs;
037e6390 1892
7d9eb12c 1893 WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex));
db94535d 1894 key.objectid = bytenr;
62e2749e 1895 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
db94535d 1896 key.offset = num_bytes;
5caf2a00 1897 path = btrfs_alloc_path();
54aa1f4d
CM		 1898 if (!path)
1899 return -ENOMEM;
5f26f772 1900
3c12ac72 1901 path->reada = 1;
3bb1a1bc
YZ		 1902 ret = lookup_extent_backref(trans, extent_root, path,
1903 bytenr, parent, root_objectid,
1904 ref_generation, owner_objectid, 1);
7bb86316 1905 if (ret == 0) {
952fccac
CM		 1906 struct btrfs_key found_key;
1907 extent_slot = path->slots[0];
1908 while(extent_slot > 0) {
1909 extent_slot--;
1910 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1911 extent_slot);
1912 if (found_key.objectid != bytenr)
1913 break;
1914 if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
1915 found_key.offset == num_bytes) {
1916 found_extent = 1;
1917 break;
1918 }
1919 if (path->slots[0] - extent_slot > 5)
1920 break;
1921 }
31840ae1
ZY		 1922 if (!found_extent) {
1923 ret = remove_extent_backref(trans, extent_root, path);
1924 BUG_ON(ret);
1925 btrfs_release_path(extent_root, path);
1926 ret = btrfs_search_slot(trans, extent_root,
1927 &key, path, -1, 1);
1928 BUG_ON(ret);
1929 extent_slot = path->slots[0];
1930 }
7bb86316
CM		 1931 } else {
1932 btrfs_print_leaf(extent_root, path->nodes[0]);
1933 WARN_ON(1);
1934 printk("Unable to find ref byte nr %Lu root %Lu "
3bb1a1bc
YZ		 1935 "gen %Lu owner %Lu\n", bytenr,
1936 root_objectid, ref_generation, owner_objectid);
7bb86316 1937 }
5f39d397
CM		 1938
1939 leaf = path->nodes[0];
952fccac 1940 ei = btrfs_item_ptr(leaf, extent_slot,
123abc88 1941 struct btrfs_extent_item);
5f39d397
CM		 1942 refs = btrfs_extent_refs(leaf, ei);
1943 BUG_ON(refs == 0);
1944 refs -= 1;
1945 btrfs_set_extent_refs(leaf, ei, refs);
952fccac 1946
5f39d397
CM		 1947 btrfs_mark_buffer_dirty(leaf);
1948
952fccac 1949 if (refs == 0 && found_extent && path->slots[0] == extent_slot + 1) {
31840ae1
ZY		 1950 struct btrfs_extent_ref *ref;
1951 ref = btrfs_item_ptr(leaf, path->slots[0],
1952 struct btrfs_extent_ref);
1953 BUG_ON(btrfs_ref_num_refs(leaf, ref) != 1);
952fccac
CM		 1954 /* if the back ref and the extent are next to each other
1955 * they get deleted below in one shot
1956 */
1957 path->slots[0] = extent_slot;
1958 num_to_del = 2;
1959 } else if (found_extent) {
1960 /* otherwise delete the extent back ref */
31840ae1 1961 ret = remove_extent_backref(trans, extent_root, path);
952fccac
CM		 1962 BUG_ON(ret);
1963 /* if refs are 0, we need to setup the path for deletion */
1964 if (refs == 0) {
1965 btrfs_release_path(extent_root, path);
1966 ret = btrfs_search_slot(trans, extent_root, &key, path,
1967 -1, 1);
952fccac
CM		 1968 BUG_ON(ret);
1969 }
1970 }
1971
cf27e1ee 1972 if (refs == 0) {
db94535d
CM		 1973 u64 super_used;
1974 u64 root_used;
21af804c
DW		 1975#ifdef BIO_RW_DISCARD
1976 u64 map_length = num_bytes;
1977 struct btrfs_multi_bio *multi = NULL;
1978#endif
78fae27e
CM		 1979
1980 if (pin) {
31840ae1
ZY		 1981 ret = pin_down_bytes(trans, root, bytenr, num_bytes,
1982 owner_objectid >= BTRFS_FIRST_FREE_OBJECTID);
c549228f
Y		 1983 if (ret > 0)
1984 mark_free = 1;
1985 BUG_ON(ret < 0);
78fae27e
CM		 1986 }
1987
58176a96 1988 /* block accounting for super block */
a2135011 1989 spin_lock_irq(&info->delalloc_lock);
db94535d
CM		 1990 super_used = btrfs_super_bytes_used(&info->super_copy);
1991 btrfs_set_super_bytes_used(&info->super_copy,
1992 super_used - num_bytes);
a2135011 1993 spin_unlock_irq(&info->delalloc_lock);
58176a96
JB		 1994
1995 /* block accounting for root item */
db94535d 1996 root_used = btrfs_root_used(&root->root_item);
5f39d397 1997 btrfs_set_root_used(&root->root_item,
db94535d 1998 root_used - num_bytes);
952fccac
CM		 1999 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
2000 num_to_del);
31840ae1 2001 BUG_ON(ret);
db94535d 2002 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
0b86a832 2003 mark_free);
9078a3e1 2004 BUG_ON(ret);
21af804c
DW		 2005
2006#ifdef BIO_RW_DISCARD
2007 /* Tell the block device(s) that the sectors can be discarded */
2008 ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
2009 bytenr, &map_length, &multi, 0);
2010 if (!ret) {
2011 struct btrfs_bio_stripe *stripe = multi->stripes;
2012 int i;
2013
2014 if (map_length > num_bytes)
2015 map_length = num_bytes;
2016
2017 for (i = 0; i < multi->num_stripes; i++, stripe++) {
2018 blkdev_issue_discard(stripe->dev->bdev,
2019 stripe->physical >> 9,
2020 map_length >> 9);
2021 }
2022 kfree(multi);
2023 }
2024#endif
a28ec197 2025 }
5caf2a00 2026 btrfs_free_path(path);
e089f05c 2027 finish_current_insert(trans, extent_root);
a28ec197
CM		 2028 return ret;
2029}
2030
a28ec197
CM		 2031/*
2032 * find all the blocks marked as pending in the radix tree and remove
2033 * them from the extent map
2034 */
e089f05c
CM		 2035static int del_pending_extents(struct btrfs_trans_handle *trans, struct
2036 btrfs_root *extent_root)
a28ec197
CM		 2037{
2038 int ret;
e20d96d6 2039 int err = 0;
31840ae1 2040 int mark_free = 0;
1a5bc167
CM		 2041 u64 start;
2042 u64 end;
31840ae1 2043 u64 priv;
d1310b2e 2044 struct extent_io_tree *pending_del;
31840ae1
ZY		 2045 struct extent_io_tree *extent_ins;
2046 struct pending_extent_op *extent_op;
8ef97622 2047
7d9eb12c 2048 WARN_ON(!mutex_is_locked(&extent_root->fs_info->alloc_mutex));
31840ae1 2049 extent_ins = &extent_root->fs_info->extent_ins;
1a5bc167 2050 pending_del = &extent_root->fs_info->pending_del;
a28ec197
CM		 2051
2052 while(1) {
1a5bc167
CM		 2053 ret = find_first_extent_bit(pending_del, 0, &start, &end,
2054 EXTENT_LOCKED);
2055 if (ret)
a28ec197 2056 break;
31840ae1
ZY		 2057
2058 ret = get_state_private(pending_del, start, &priv);
2059 BUG_ON(ret);
2060 extent_op = (struct pending_extent_op *)(unsigned long)priv;
2061
1a5bc167
CM		 2062 clear_extent_bits(pending_del, start, end, EXTENT_LOCKED,
2063 GFP_NOFS);
31840ae1
ZY		 2064
2065 ret = pin_down_bytes(trans, extent_root, start,
2066 end + 1 - start, 0);
2067 mark_free = ret > 0;
2068 if (!test_range_bit(extent_ins, start, end,
2069 EXTENT_LOCKED, 0)) {
2070free_extent:
c286ac48 2071 ret = __free_extent(trans, extent_root,
31840ae1
ZY		 2072 start, end + 1 - start,
2073 extent_op->orig_parent,
2074 extent_root->root_key.objectid,
2075 extent_op->orig_generation,
3bb1a1bc 2076 extent_op->level, 0, mark_free);
31840ae1 2077 kfree(extent_op);
c286ac48 2078 } else {
31840ae1
ZY		 2079 kfree(extent_op);
2080 ret = get_state_private(extent_ins, start, &priv);
2081 BUG_ON(ret);
2082 extent_op = (struct pending_extent_op *)
2083 (unsigned long)priv;
2084
2085 clear_extent_bits(extent_ins, start, end,
2086 EXTENT_LOCKED, GFP_NOFS);
2087
2088 if (extent_op->type == PENDING_BACKREF_UPDATE)
2089 goto free_extent;
2090
2091 ret = update_block_group(trans, extent_root, start,
2092 end + 1 - start, 0, mark_free);
2093 BUG_ON(ret);
2094 kfree(extent_op);
c286ac48 2095 }
1a5bc167
CM		 2096 if (ret)
2097 err = ret;
c286ac48
CM		 2098
2099 if (need_resched()) {
2100 mutex_unlock(&extent_root->fs_info->alloc_mutex);
2101 cond_resched();
2102 mutex_lock(&extent_root->fs_info->alloc_mutex);
2103 }
fec577fb 2104 }
e20d96d6 2105 return err;
fec577fb
CM		 2106}
2107
2108/*
2109 * remove an extent from the root, returns 0 on success
2110 */
925baedd 2111static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
31840ae1
ZY		 2112 struct btrfs_root *root,
2113 u64 bytenr, u64 num_bytes, u64 parent,
2114 u64 root_objectid, u64 ref_generation,
3bb1a1bc 2115 u64 owner_objectid, int pin)
fec577fb 2116{
9f5fae2f 2117 struct btrfs_root *extent_root = root->fs_info->extent_root;
fec577fb
CM		 2118 int pending_ret;
2119 int ret;
a28ec197 2120
db94535d 2121 WARN_ON(num_bytes < root->sectorsize);
fec577fb 2122 if (root == extent_root) {
31840ae1
ZY		 2123 struct pending_extent_op *extent_op;
2124
2125 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2126 BUG_ON(!extent_op);
2127
2128 extent_op->type = PENDING_EXTENT_DELETE;
2129 extent_op->bytenr = bytenr;
2130 extent_op->num_bytes = num_bytes;
2131 extent_op->parent = parent;
2132 extent_op->orig_parent = parent;
2133 extent_op->generation = ref_generation;
2134 extent_op->orig_generation = ref_generation;
2135 extent_op->level = (int)owner_objectid;
2136
2137 set_extent_bits(&root->fs_info->pending_del,
2138 bytenr, bytenr + num_bytes - 1,
2139 EXTENT_LOCKED, GFP_NOFS);
2140 set_state_private(&root->fs_info->pending_del,
2141 bytenr, (unsigned long)extent_op);
fec577fb
CM		 2142 return 0;
2143 }
4bef0848 2144 /* if metadata always pin */
d00aff00
CM		 2145 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
2146 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
0f9dd46c
JB		 2147 struct btrfs_block_group_cache *cache;
2148
d00aff00 2149 /* btrfs_free_reserved_extent */
0f9dd46c
JB		 2150 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
2151 BUG_ON(!cache);
2152 btrfs_add_free_space(cache, bytenr, num_bytes);
e8569813 2153 update_reserved_extents(root, bytenr, num_bytes, 0);
d00aff00
CM		 2154 return 0;
2155 }
4bef0848 2156 pin = 1;
d00aff00 2157 }
4bef0848
CM		 2158
2159 /* if data pin when any transaction has committed this */
2160 if (ref_generation != trans->transid)
2161 pin = 1;
2162
31840ae1 2163 ret = __free_extent(trans, root, bytenr, num_bytes, parent,
3bb1a1bc
YZ		 2164 root_objectid, ref_generation,
2165 owner_objectid, pin, pin == 0);
ee6e6504
CM		 2166
2167 finish_current_insert(trans, root->fs_info->extent_root);
e20d96d6 2168 pending_ret = del_pending_extents(trans, root->fs_info->extent_root);
fec577fb
CM		 2169 return ret ? ret : pending_ret;
2170}
2171
925baedd 2172int btrfs_free_extent(struct btrfs_trans_handle *trans,
31840ae1
ZY		 2173 struct btrfs_root *root,
2174 u64 bytenr, u64 num_bytes, u64 parent,
2175 u64 root_objectid, u64 ref_generation,
3bb1a1bc 2176 u64 owner_objectid, int pin)
925baedd
CM		 2177{
2178 int ret;
2179
2180 maybe_lock_mutex(root);
31840ae1 2181 ret = __btrfs_free_extent(trans, root, bytenr, num_bytes, parent,
925baedd 2182 root_objectid, ref_generation,
3bb1a1bc 2183 owner_objectid, pin);
925baedd
CM		 2184 maybe_unlock_mutex(root);
2185 return ret;
2186}
2187
87ee04eb
CM		 2188static u64 stripe_align(struct btrfs_root *root, u64 val)
2189{
2190 u64 mask = ((u64)root->stripesize - 1);
2191 u64 ret = (val + mask) & ~mask;
2192 return ret;
2193}
2194
fec577fb
CM		 2195/*
2196 * walks the btree of allocated extents and find a hole of a given size.
2197 * The key ins is changed to record the hole:
2198 * ins->objectid == block start
62e2749e 2199 * ins->flags = BTRFS_EXTENT_ITEM_KEY
fec577fb
CM		 2200 * ins->offset == number of blocks
2201 * Any available blocks before search_start are skipped.
2202 */
98ed5174
CM		 2203static int noinline find_free_extent(struct btrfs_trans_handle *trans,
2204 struct btrfs_root *orig_root,
2205 u64 num_bytes, u64 empty_size,
2206 u64 search_start, u64 search_end,
2207 u64 hint_byte, struct btrfs_key *ins,
2208 u64 exclude_start, u64 exclude_nr,
2209 int data)
fec577fb 2210{
87ee04eb 2211 int ret;
a061fc8d 2212 u64 orig_search_start;
9f5fae2f 2213 struct btrfs_root * root = orig_root->fs_info->extent_root;
f2458e1d 2214 struct btrfs_fs_info *info = root->fs_info;
db94535d 2215 u64 total_needed = num_bytes;
239b14b3 2216 u64 *last_ptr = NULL;
be08c1b9 2217 struct btrfs_block_group_cache *block_group;
0ef3e66b 2218 int chunk_alloc_done = 0;
239b14b3 2219 int empty_cluster = 2 * 1024 * 1024;
0ef3e66b 2220 int allowed_chunk_alloc = 0;
fec577fb 2221
db94535d 2222 WARN_ON(num_bytes < root->sectorsize);
b1a4d965
CM		 2223 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
2224
0ef3e66b
CM		 2225 if (orig_root->ref_cows || empty_size)
2226 allowed_chunk_alloc = 1;
2227
239b14b3
CM		 2228 if (data & BTRFS_BLOCK_GROUP_METADATA) {
2229 last_ptr = &root->fs_info->last_alloc;
8790d502 2230 empty_cluster = 256 * 1024;
239b14b3
CM		 2231 }
2232
0f9dd46c 2233 if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD))
239b14b3 2234 last_ptr = &root->fs_info->last_data_alloc;
0f9dd46c 2235
239b14b3
CM		 2236 if (last_ptr) {
2237 if (*last_ptr)
2238 hint_byte = *last_ptr;
0f9dd46c 2239 else
239b14b3 2240 empty_size += empty_cluster;
239b14b3
CM		 2241 }
2242
a061fc8d
CM		 2243 search_start = max(search_start, first_logical_byte(root, 0));
2244 orig_search_start = search_start;
2245
239b14b3 2246 search_start = max(search_start, hint_byte);
6702ed49 2247 total_needed += empty_size;
0b86a832 2248
0f9dd46c 2249new_group:
45b8c9a8
JB		 2250 block_group = btrfs_lookup_block_group(info, search_start);
2251 if (!block_group)
2252 block_group = btrfs_lookup_first_block_group(info,
2253 search_start);
0f9dd46c
JB		 2254
2255 /*
2256 * Ok this looks a little tricky, buts its really simple. First if we
2257 * didn't find a block group obviously we want to start over.
2258 * Secondly, if the block group we found does not match the type we
2259 * need, and we have a last_ptr and its not 0, chances are the last
2260 * allocation we made was at the end of the block group, so lets go
2261 * ahead and skip the looking through the rest of the block groups and
2262 * start at the beginning. This helps with metadata allocations,
2263 * since you are likely to have a bunch of data block groups to search
2264 * through first before you realize that you need to start over, so go
2265 * ahead and start over and save the time.
2266 */
2267 if (!block_group || (!block_group_bits(block_group, data) &&
2268 last_ptr && *last_ptr)) {
2269 if (search_start != orig_search_start) {
45b8c9a8
JB		 2270 if (last_ptr && *last_ptr) {
2271 total_needed += empty_cluster;
0f9dd46c 2272 *last_ptr = 0;
45b8c9a8 2273 }
0f9dd46c
JB		 2274 search_start = orig_search_start;
2275 goto new_group;
2276 } else if (!chunk_alloc_done && allowed_chunk_alloc) {
2277 ret = do_chunk_alloc(trans, root,
2278 num_bytes + 2 * 1024 * 1024,
2279 data, 1);
e8569813 2280 if (ret < 0)
0f9dd46c 2281 goto error;
0f9dd46c
JB		 2282 BUG_ON(ret);
2283 chunk_alloc_done = 1;
2284 search_start = orig_search_start;
2285 goto new_group;
2286 } else {
2287 ret = -ENOSPC;
2288 goto error;
0ef3e66b 2289 }
239b14b3 2290 }
e19caa5f 2291
0f9dd46c
JB		 2292 /*
2293 * this is going to seach through all of the existing block groups it
2294 * can find, so if we don't find something we need to see if we can
2295 * allocate what we need.
2296 */
2297 ret = find_free_space(root, &block_group, &search_start,
2298 total_needed, data);
2299 if (ret == -ENOSPC) {
2300 /*
2301 * instead of allocating, start at the original search start
2302 * and see if there is something to be found, if not then we
2303 * allocate
2304 */
2305 if (search_start != orig_search_start) {
2306 if (last_ptr && *last_ptr) {
2307 *last_ptr = 0;
2308 total_needed += empty_cluster;
2309 }
2310 search_start = orig_search_start;
2311 goto new_group;
239b14b3 2312 }
0f9dd46c
JB		 2313
2314 /*
2315 * we've already allocated, we're pretty screwed
2316 */
2317 if (chunk_alloc_done) {
239b14b3 2318 goto error;
0f9dd46c
JB		 2319 } else if (!allowed_chunk_alloc && block_group &&
2320 block_group_bits(block_group, data)) {
2321 block_group->space_info->force_alloc = 1;
2322 goto error;
2323 } else if (!allowed_chunk_alloc) {
2324 goto error;
2325 }
2326
2327 ret = do_chunk_alloc(trans, root, num_bytes + 2 * 1024 * 1024,
2328 data, 1);
2329 if (ret < 0)
2330 goto error;
2331
2332 BUG_ON(ret);
2333 chunk_alloc_done = 1;
2334 if (block_group)
2335 search_start = block_group->key.objectid +
2336 block_group->key.offset;
2337 else
2338 search_start = orig_search_start;
2339 goto new_group;
239b14b3
CM		 2340 }
2341
0f9dd46c
JB		 2342 if (ret)
2343 goto error;
2344
0b86a832
CM		 2345 search_start = stripe_align(root, search_start);
2346 ins->objectid = search_start;
2347 ins->offset = num_bytes;
e37c9e69 2348
0f9dd46c
JB		 2349 if (ins->objectid + num_bytes >= search_end) {
2350 search_start = orig_search_start;
2351 if (chunk_alloc_done) {
2352 ret = -ENOSPC;
2353 goto error;
2354 }
2355 goto new_group;
2356 }
0b86a832
CM		 2357
2358 if (ins->objectid + num_bytes >
2359 block_group->key.objectid + block_group->key.offset) {
0f9dd46c
JB		 2360 if (search_start == orig_search_start && chunk_alloc_done) {
2361 ret = -ENOSPC;
2362 goto error;
2363 }
e19caa5f
CM		 2364 search_start = block_group->key.objectid +
2365 block_group->key.offset;
2366 goto new_group;
2367 }
0b86a832 2368
db94535d 2369 if (exclude_nr > 0 && (ins->objectid + num_bytes > exclude_start &&
f2654de4
CM		 2370 ins->objectid < exclude_start + exclude_nr)) {
2371 search_start = exclude_start + exclude_nr;
2372 goto new_group;
2373 }
0b86a832 2374
0f9dd46c
JB		 2375 if (!(data & BTRFS_BLOCK_GROUP_DATA))
2376 trans->block_group = block_group;
2377
db94535d 2378 ins->offset = num_bytes;
239b14b3
CM		 2379 if (last_ptr) {
2380 *last_ptr = ins->objectid + ins->offset;
2381 if (*last_ptr ==
0f9dd46c 2382 btrfs_super_total_bytes(&root->fs_info->super_copy))
239b14b3 2383 *last_ptr = 0;
be744175 2384 }
be744175 2385
0f9dd46c 2386 ret = 0;
0f70abe2 2387error:
0f70abe2 2388 return ret;
fec577fb 2389}
ec44a35c 2390
0f9dd46c
JB		 2391static void dump_space_info(struct btrfs_space_info *info, u64 bytes)
2392{
2393 struct btrfs_block_group_cache *cache;
2394 struct list_head *l;
2395
2396 printk(KERN_INFO "space_info has %Lu free, is %sfull\n",
e8569813
ZY		 2397 info->total_bytes - info->bytes_used - info->bytes_pinned -
2398 info->bytes_reserved, (info->full) ? "" : "not ");
0f9dd46c
JB		 2399
2400 spin_lock(&info->lock);
2401 list_for_each(l, &info->block_groups) {
2402 cache = list_entry(l, struct btrfs_block_group_cache, list);
2403 spin_lock(&cache->lock);
2404 printk(KERN_INFO "block group %Lu has %Lu bytes, %Lu used "
e8569813 2405 "%Lu pinned %Lu reserved\n",
0f9dd46c 2406 cache->key.objectid, cache->key.offset,
e8569813
ZY		 2407 btrfs_block_group_used(&cache->item),
2408 cache->pinned, cache->reserved);
0f9dd46c
JB		 2409 btrfs_dump_free_space(cache, bytes);
2410 spin_unlock(&cache->lock);
2411 }
2412 spin_unlock(&info->lock);
2413}
e8569813 2414
e6dcd2dc
CM		 2415static int __btrfs_reserve_extent(struct btrfs_trans_handle *trans,
2416 struct btrfs_root *root,
2417 u64 num_bytes, u64 min_alloc_size,
2418 u64 empty_size, u64 hint_byte,
2419 u64 search_end, struct btrfs_key *ins,
2420 u64 data)
fec577fb
CM		 2421{
2422 int ret;
fbdc762b 2423 u64 search_start = 0;
8790d502 2424 u64 alloc_profile;
1261ec42 2425 struct btrfs_fs_info *info = root->fs_info;
0f9dd46c 2426 struct btrfs_block_group_cache *cache;
925baedd 2427
6324fbf3 2428 if (data) {
8790d502
CM		 2429 alloc_profile = info->avail_data_alloc_bits &
2430 info->data_alloc_profile;
2431 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
6324fbf3 2432 } else if (root == root->fs_info->chunk_root) {
8790d502
CM		 2433 alloc_profile = info->avail_system_alloc_bits &
2434 info->system_alloc_profile;
2435 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
6324fbf3 2436 } else {
8790d502
CM		 2437 alloc_profile = info->avail_metadata_alloc_bits &
2438 info->metadata_alloc_profile;
2439 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
6324fbf3 2440 }
98d20f67 2441again:
a061fc8d 2442 data = reduce_alloc_profile(root, data);
0ef3e66b
CM		 2443 /*
2444 * the only place that sets empty_size is btrfs_realloc_node, which
2445 * is not called recursively on allocations
2446 */
2447 if (empty_size || root->ref_cows) {
593060d7 2448 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
6324fbf3 2449 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
0ef3e66b
CM		 2450 2 * 1024 * 1024,
2451 BTRFS_BLOCK_GROUP_METADATA |
2452 (info->metadata_alloc_profile &
2453 info->avail_metadata_alloc_bits), 0);
6324fbf3
CM		 2454 }
2455 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
0ef3e66b 2456 num_bytes + 2 * 1024 * 1024, data, 0);
6324fbf3 2457 }
0b86a832 2458
db94535d
CM		 2459 WARN_ON(num_bytes < root->sectorsize);
2460 ret = find_free_extent(trans, root, num_bytes, empty_size,
2461 search_start, search_end, hint_byte, ins,
26b8003f
CM		 2462 trans->alloc_exclude_start,
2463 trans->alloc_exclude_nr, data);
3b951516 2464
98d20f67
CM		 2465 if (ret == -ENOSPC && num_bytes > min_alloc_size) {
2466 num_bytes = num_bytes >> 1;
0f9dd46c 2467 num_bytes = num_bytes & ~(root->sectorsize - 1);
98d20f67 2468 num_bytes = max(num_bytes, min_alloc_size);
0ef3e66b
CM		 2469 do_chunk_alloc(trans, root->fs_info->extent_root,
2470 num_bytes, data, 1);
98d20f67
CM		 2471 goto again;
2472 }
ec44a35c 2473 if (ret) {
0f9dd46c
JB		 2474 struct btrfs_space_info *sinfo;
2475
2476 sinfo = __find_space_info(root->fs_info, data);
2477 printk("allocation failed flags %Lu, wanted %Lu\n",
2478 data, num_bytes);
2479 dump_space_info(sinfo, num_bytes);
925baedd 2480 BUG();
925baedd 2481 }
0f9dd46c
JB		 2482 cache = btrfs_lookup_block_group(root->fs_info, ins->objectid);
2483 if (!cache) {
2484 printk(KERN_ERR "Unable to find block group for %Lu\n", ins->objectid);
2485 return -ENOSPC;
2486 }
2487
2488 ret = btrfs_remove_free_space(cache, ins->objectid, ins->offset);
2489
2490 return ret;
e6dcd2dc
CM		 2491}
2492
65b51a00
CM		 2493int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
2494{
0f9dd46c
JB		 2495 struct btrfs_block_group_cache *cache;
2496
65b51a00 2497 maybe_lock_mutex(root);
0f9dd46c
JB		 2498 cache = btrfs_lookup_block_group(root->fs_info, start);
2499 if (!cache) {
2500 printk(KERN_ERR "Unable to find block group for %Lu\n", start);
2501 maybe_unlock_mutex(root);
2502 return -ENOSPC;
2503 }
2504 btrfs_add_free_space(cache, start, len);
1a40e23b 2505 update_reserved_extents(root, start, len, 0);
65b51a00
CM		 2506 maybe_unlock_mutex(root);
2507 return 0;
2508}
2509
e6dcd2dc
CM		 2510int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
2511 struct btrfs_root *root,
2512 u64 num_bytes, u64 min_alloc_size,
2513 u64 empty_size, u64 hint_byte,
2514 u64 search_end, struct btrfs_key *ins,
2515 u64 data)
2516{
2517 int ret;
2518 maybe_lock_mutex(root);
2519 ret = __btrfs_reserve_extent(trans, root, num_bytes, min_alloc_size,
2520 empty_size, hint_byte, search_end, ins,
2521 data);
e8569813 2522 update_reserved_extents(root, ins->objectid, ins->offset, 1);
e6dcd2dc
CM		 2523 maybe_unlock_mutex(root);
2524 return ret;
2525}
2526
2527static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
31840ae1 2528 struct btrfs_root *root, u64 parent,
e6dcd2dc 2529 u64 root_objectid, u64 ref_generation,
3bb1a1bc 2530 u64 owner, struct btrfs_key *ins)
e6dcd2dc
CM		 2531{
2532 int ret;
2533 int pending_ret;
2534 u64 super_used;
2535 u64 root_used;
2536 u64 num_bytes = ins->offset;
2537 u32 sizes[2];
2538 struct btrfs_fs_info *info = root->fs_info;
2539 struct btrfs_root *extent_root = info->extent_root;
2540 struct btrfs_extent_item *extent_item;
2541 struct btrfs_extent_ref *ref;
2542 struct btrfs_path *path;
2543 struct btrfs_key keys[2];
fec577fb 2544
31840ae1
ZY		 2545 if (parent == 0)
2546 parent = ins->objectid;
2547
58176a96 2548 /* block accounting for super block */
a2135011 2549 spin_lock_irq(&info->delalloc_lock);
db94535d
CM		 2550 super_used = btrfs_super_bytes_used(&info->super_copy);
2551 btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes);
a2135011 2552 spin_unlock_irq(&info->delalloc_lock);
26b8003f 2553
58176a96 2554 /* block accounting for root item */
db94535d
CM		 2555 root_used = btrfs_root_used(&root->root_item);
2556 btrfs_set_root_used(&root->root_item, root_used + num_bytes);
58176a96 2557
26b8003f 2558 if (root == extent_root) {
31840ae1
ZY		 2559 struct pending_extent_op *extent_op;
2560
2561 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2562 BUG_ON(!extent_op);
2563
2564 extent_op->type = PENDING_EXTENT_INSERT;
2565 extent_op->bytenr = ins->objectid;
2566 extent_op->num_bytes = ins->offset;
2567 extent_op->parent = parent;
2568 extent_op->orig_parent = 0;
2569 extent_op->generation = ref_generation;
2570 extent_op->orig_generation = 0;
2571 extent_op->level = (int)owner;
2572
1a5bc167
CM		 2573 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
2574 ins->objectid + ins->offset - 1,
2575 EXTENT_LOCKED, GFP_NOFS);
31840ae1
ZY		 2576 set_state_private(&root->fs_info->extent_ins,
2577 ins->objectid, (unsigned long)extent_op);
26b8003f
CM		 2578 goto update_block;
2579 }
2580
47e4bb98 2581 memcpy(&keys[0], ins, sizeof(*ins));
47e4bb98
CM		 2582 keys[1].objectid = ins->objectid;
2583 keys[1].type = BTRFS_EXTENT_REF_KEY;
31840ae1 2584 keys[1].offset = parent;
47e4bb98
CM		 2585 sizes[0] = sizeof(*extent_item);
2586 sizes[1] = sizeof(*ref);
7bb86316
CM		 2587
2588 path = btrfs_alloc_path();
2589 BUG_ON(!path);
47e4bb98
CM		 2590
2591 ret = btrfs_insert_empty_items(trans, extent_root, path, keys,
2592 sizes, 2);
ccd467d6 2593 BUG_ON(ret);
0f9dd46c 2594
47e4bb98
CM		 2595 extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2596 struct btrfs_extent_item);
2597 btrfs_set_extent_refs(path->nodes[0], extent_item, 1);
2598 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
2599 struct btrfs_extent_ref);
2600
2601 btrfs_set_ref_root(path->nodes[0], ref, root_objectid);
2602 btrfs_set_ref_generation(path->nodes[0], ref, ref_generation);
2603 btrfs_set_ref_objectid(path->nodes[0], ref, owner);
31840ae1 2604 btrfs_set_ref_num_refs(path->nodes[0], ref, 1);
47e4bb98
CM		 2605
2606 btrfs_mark_buffer_dirty(path->nodes[0]);
2607
2608 trans->alloc_exclude_start = 0;
2609 trans->alloc_exclude_nr = 0;
7bb86316 2610 btrfs_free_path(path);
e089f05c 2611 finish_current_insert(trans, extent_root);
e20d96d6 2612 pending_ret = del_pending_extents(trans, extent_root);
f510cfec 2613
925baedd
CM		 2614 if (ret)
2615 goto out;
e37c9e69 2616 if (pending_ret) {
925baedd
CM		 2617 ret = pending_ret;
2618 goto out;
e37c9e69 2619 }
26b8003f
CM		 2620
2621update_block:
0b86a832 2622 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1, 0);
f5947066
CM		 2623 if (ret) {
2624 printk("update block group failed for %Lu %Lu\n",
2625 ins->objectid, ins->offset);
2626 BUG();
2627 }
925baedd 2628out:
e6dcd2dc
CM		 2629 return ret;
2630}
2631
2632int btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
31840ae1 2633 struct btrfs_root *root, u64 parent,
e6dcd2dc 2634 u64 root_objectid, u64 ref_generation,
3bb1a1bc 2635 u64 owner, struct btrfs_key *ins)
e6dcd2dc
CM		 2636{
2637 int ret;
1c2308f8
CM		 2638
2639 if (root_objectid == BTRFS_TREE_LOG_OBJECTID)
2640 return 0;
e6dcd2dc 2641 maybe_lock_mutex(root);
3bb1a1bc
YZ		 2642 ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
2643 ref_generation, owner, ins);
e8569813 2644 update_reserved_extents(root, ins->objectid, ins->offset, 0);
e6dcd2dc
CM		 2645 maybe_unlock_mutex(root);
2646 return ret;
2647}
e02119d5
CM		 2648
2649/*
2650 * this is used by the tree logging recovery code. It records that
2651 * an extent has been allocated and makes sure to clear the free
2652 * space cache bits as well
2653 */
2654int btrfs_alloc_logged_extent(struct btrfs_trans_handle *trans,
31840ae1 2655 struct btrfs_root *root, u64 parent,
e02119d5 2656 u64 root_objectid, u64 ref_generation,
3bb1a1bc 2657 u64 owner, struct btrfs_key *ins)
e02119d5
CM		 2658{
2659 int ret;
2660 struct btrfs_block_group_cache *block_group;
2661
2662 maybe_lock_mutex(root);
2663 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
2664 cache_block_group(root, block_group);
2665
0f9dd46c
JB		 2666 ret = btrfs_remove_free_space(block_group, ins->objectid, ins->offset);
2667 BUG_ON(ret);
3bb1a1bc
YZ		 2668 ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
2669 ref_generation, owner, ins);
e02119d5
CM		 2670 maybe_unlock_mutex(root);
2671 return ret;
2672}
2673
e6dcd2dc
CM		 2674/*
2675 * finds a free extent and does all the dirty work required for allocation
2676 * returns the key for the extent through ins, and a tree buffer for
2677 * the first block of the extent through buf.
2678 *
2679 * returns 0 if everything worked, non-zero otherwise.
2680 */
2681int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
2682 struct btrfs_root *root,
31840ae1 2683 u64 num_bytes, u64 parent, u64 min_alloc_size,
e6dcd2dc 2684 u64 root_objectid, u64 ref_generation,
3bb1a1bc 2685 u64 owner_objectid, u64 empty_size, u64 hint_byte,
e6dcd2dc
CM		 2686 u64 search_end, struct btrfs_key *ins, u64 data)
2687{
2688 int ret;
2689
2690 maybe_lock_mutex(root);
2691
2692 ret = __btrfs_reserve_extent(trans, root, num_bytes,
2693 min_alloc_size, empty_size, hint_byte,
2694 search_end, ins, data);
2695 BUG_ON(ret);
d00aff00 2696 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
31840ae1
ZY		 2697 ret = __btrfs_alloc_reserved_extent(trans, root, parent,
2698 root_objectid, ref_generation,
3bb1a1bc 2699 owner_objectid, ins);
d00aff00 2700 BUG_ON(ret);
e6dcd2dc 2701
e8569813
ZY		 2702 } else {
2703 update_reserved_extents(root, ins->objectid, ins->offset, 1);
d00aff00 2704 }
925baedd
CM		 2705 maybe_unlock_mutex(root);
2706 return ret;
fec577fb 2707}
65b51a00
CM		 2708
2709struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
2710 struct btrfs_root *root,
2711 u64 bytenr, u32 blocksize)
2712{
2713 struct extent_buffer *buf;
2714
2715 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
2716 if (!buf)
2717 return ERR_PTR(-ENOMEM);
2718 btrfs_set_header_generation(buf, trans->transid);
2719 btrfs_tree_lock(buf);
2720 clean_tree_block(trans, root, buf);
2721 btrfs_set_buffer_uptodate(buf);
d0c803c4
CM		 2722 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
2723 set_extent_dirty(&root->dirty_log_pages, buf->start,
2724 buf->start + buf->len - 1, GFP_NOFS);
2725 } else {
2726 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
65b51a00 2727 buf->start + buf->len - 1, GFP_NOFS);
d0c803c4 2728 }
65b51a00
CM		 2729 trans->blocks_used++;
2730 return buf;
2731}
2732
fec577fb
CM		 2733/*
2734 * helper function to allocate a block for a given tree
2735 * returns the tree buffer or NULL.
2736 */
5f39d397 2737struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
7bb86316 2738 struct btrfs_root *root,
31840ae1 2739 u32 blocksize, u64 parent,
7bb86316
CM		 2740 u64 root_objectid,
2741 u64 ref_generation,
7bb86316
CM		 2742 int level,
2743 u64 hint,
5f39d397 2744 u64 empty_size)
fec577fb 2745{
e2fa7227 2746 struct btrfs_key ins;
fec577fb 2747 int ret;
5f39d397 2748 struct extent_buffer *buf;
fec577fb 2749
31840ae1 2750 ret = btrfs_alloc_extent(trans, root, blocksize, parent, blocksize,
3bb1a1bc 2751 root_objectid, ref_generation, level,
31840ae1 2752 empty_size, hint, (u64)-1, &ins, 0);
fec577fb 2753 if (ret) {
54aa1f4d
CM		 2754 BUG_ON(ret > 0);
2755 return ERR_PTR(ret);
fec577fb 2756 }
55c69072 2757
65b51a00 2758 buf = btrfs_init_new_buffer(trans, root, ins.objectid, blocksize);
fec577fb
CM		 2759 return buf;
2760}
a28ec197 2761
e02119d5
CM		 2762int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans,
2763 struct btrfs_root *root, struct extent_buffer *leaf)
6407bf6d 2764{
7bb86316
CM		 2765 u64 leaf_owner;
2766 u64 leaf_generation;
5f39d397 2767 struct btrfs_key key;
6407bf6d
CM		 2768 struct btrfs_file_extent_item *fi;
2769 int i;
2770 int nritems;
2771 int ret;
2772
5f39d397
CM		 2773 BUG_ON(!btrfs_is_leaf(leaf));
2774 nritems = btrfs_header_nritems(leaf);
7bb86316
CM		 2775 leaf_owner = btrfs_header_owner(leaf);
2776 leaf_generation = btrfs_header_generation(leaf);
2777
6407bf6d 2778 for (i = 0; i < nritems; i++) {
db94535d 2779 u64 disk_bytenr;
e34a5b4f 2780 cond_resched();
5f39d397
CM		 2781
2782 btrfs_item_key_to_cpu(leaf, &key, i);
2783 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
6407bf6d
CM		 2784 continue;
2785 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
5f39d397
CM		 2786 if (btrfs_file_extent_type(leaf, fi) ==
2787 BTRFS_FILE_EXTENT_INLINE)
236454df 2788 continue;
6407bf6d
CM		 2789 /*
2790 * FIXME make sure to insert a trans record that
2791 * repeats the snapshot del on crash
2792 */
db94535d
CM		 2793 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
2794 if (disk_bytenr == 0)
3a686375 2795 continue;
4a096752
CM		 2796
2797 mutex_lock(&root->fs_info->alloc_mutex);
925baedd 2798 ret = __btrfs_free_extent(trans, root, disk_bytenr,
7bb86316 2799 btrfs_file_extent_disk_num_bytes(leaf, fi),
31840ae1 2800 leaf->start, leaf_owner, leaf_generation,
3bb1a1bc 2801 key.objectid, 0);
4a096752 2802 mutex_unlock(&root->fs_info->alloc_mutex);
31840ae1 2803 BUG_ON(ret);
2dd3e67b
CM		 2804
2805 atomic_inc(&root->fs_info->throttle_gen);
2806 wake_up(&root->fs_info->transaction_throttle);
2807 cond_resched();
6407bf6d
CM		 2808 }
2809 return 0;
2810}
2811
e02119d5
CM		 2812static int noinline cache_drop_leaf_ref(struct btrfs_trans_handle *trans,
2813 struct btrfs_root *root,
2814 struct btrfs_leaf_ref *ref)
31153d81
YZ		 2815{
2816 int i;
2817 int ret;
2818 struct btrfs_extent_info *info = ref->extents;
2819
31153d81
YZ		 2820 for (i = 0; i < ref->nritems; i++) {
2821 mutex_lock(&root->fs_info->alloc_mutex);
31840ae1
ZY		 2822 ret = __btrfs_free_extent(trans, root, info->bytenr,
2823 info->num_bytes, ref->bytenr,
2824 ref->owner, ref->generation,
3bb1a1bc 2825 info->objectid, 0);
31153d81 2826 mutex_unlock(&root->fs_info->alloc_mutex);
2dd3e67b
CM		 2827
2828 atomic_inc(&root->fs_info->throttle_gen);
2829 wake_up(&root->fs_info->transaction_throttle);
2830 cond_resched();
2831
31153d81
YZ		 2832 BUG_ON(ret);
2833 info++;
2834 }
31153d81
YZ		 2835
2836 return 0;
2837}
2838
333db94c
CM		 2839int drop_snap_lookup_refcount(struct btrfs_root *root, u64 start, u64 len,
2840 u32 *refs)
2841{
017e5369 2842 int ret;
f87f057b 2843
31840ae1 2844 ret = btrfs_lookup_extent_ref(NULL, root, start, len, refs);
f87f057b
CM		 2845 BUG_ON(ret);
2846
2847#if 0 // some debugging code in case we see problems here
2848 /* if the refs count is one, it won't get increased again. But
2849 * if the ref count is > 1, someone may be decreasing it at
2850 * the same time we are.
2851 */
2852 if (*refs != 1) {
2853 struct extent_buffer *eb = NULL;
2854 eb = btrfs_find_create_tree_block(root, start, len);
2855 if (eb)
2856 btrfs_tree_lock(eb);
2857
2858 mutex_lock(&root->fs_info->alloc_mutex);
2859 ret = lookup_extent_ref(NULL, root, start, len, refs);
2860 BUG_ON(ret);
2861 mutex_unlock(&root->fs_info->alloc_mutex);
2862
2863 if (eb) {
2864 btrfs_tree_unlock(eb);
2865 free_extent_buffer(eb);
2866 }
2867 if (*refs == 1) {
2868 printk("block %llu went down to one during drop_snap\n",
2869 (unsigned long long)start);
2870 }
2871
2872 }
2873#endif
2874
e7a84565 2875 cond_resched();
017e5369 2876 return ret;
333db94c
CM		 2877}
2878
9aca1d51
CM		 2879/*
2880 * helper function for drop_snapshot, this walks down the tree dropping ref
2881 * counts as it goes.
2882 */
98ed5174
CM		 2883static int noinline walk_down_tree(struct btrfs_trans_handle *trans,
2884 struct btrfs_root *root,
2885 struct btrfs_path *path, int *level)
20524f02 2886{
7bb86316
CM		 2887 u64 root_owner;
2888 u64 root_gen;
2889 u64 bytenr;
ca7a79ad 2890 u64 ptr_gen;
5f39d397
CM		 2891 struct extent_buffer *next;
2892 struct extent_buffer *cur;
7bb86316 2893 struct extent_buffer *parent;
31153d81 2894 struct btrfs_leaf_ref *ref;
db94535d 2895 u32 blocksize;
20524f02
CM		 2896 int ret;
2897 u32 refs;
2898
5caf2a00
CM		 2899 WARN_ON(*level < 0);
2900 WARN_ON(*level >= BTRFS_MAX_LEVEL);
333db94c 2901 ret = drop_snap_lookup_refcount(root, path->nodes[*level]->start,
db94535d 2902 path->nodes[*level]->len, &refs);
20524f02
CM		 2903 BUG_ON(ret);
2904 if (refs > 1)
2905 goto out;
e011599b 2906
9aca1d51
CM		 2907 /*
2908 * walk down to the last node level and free all the leaves
2909 */
6407bf6d 2910 while(*level >= 0) {
5caf2a00
CM		 2911 WARN_ON(*level < 0);
2912 WARN_ON(*level >= BTRFS_MAX_LEVEL);
20524f02 2913 cur = path->nodes[*level];
e011599b 2914
5f39d397 2915 if (btrfs_header_level(cur) != *level)
2c90e5d6 2916 WARN_ON(1);
e011599b 2917
7518a238 2918 if (path->slots[*level] >=
5f39d397 2919 btrfs_header_nritems(cur))
20524f02 2920 break;
6407bf6d 2921 if (*level == 0) {
e02119d5 2922 ret = btrfs_drop_leaf_ref(trans, root, cur);
6407bf6d
CM		 2923 BUG_ON(ret);
2924 break;
2925 }
db94535d 2926 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
ca7a79ad 2927 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
db94535d 2928 blocksize = btrfs_level_size(root, *level - 1);
925baedd 2929
333db94c 2930 ret = drop_snap_lookup_refcount(root, bytenr, blocksize, &refs);
6407bf6d
CM		 2931 BUG_ON(ret);
2932 if (refs != 1) {
7bb86316
CM		 2933 parent = path->nodes[*level];
2934 root_owner = btrfs_header_owner(parent);
2935 root_gen = btrfs_header_generation(parent);
20524f02 2936 path->slots[*level]++;
f87f057b
CM		 2937
2938 mutex_lock(&root->fs_info->alloc_mutex);
925baedd 2939 ret = __btrfs_free_extent(trans, root, bytenr,
31840ae1 2940 blocksize, parent->start,
3bb1a1bc
YZ		 2941 root_owner, root_gen,
2942 *level - 1, 1);
20524f02 2943 BUG_ON(ret);
f87f057b 2944 mutex_unlock(&root->fs_info->alloc_mutex);
18e35e0a
CM		 2945
2946 atomic_inc(&root->fs_info->throttle_gen);
2947 wake_up(&root->fs_info->transaction_throttle);
2dd3e67b 2948 cond_resched();
18e35e0a 2949
20524f02
CM		 2950 continue;
2951 }
f87f057b
CM		 2952 /*
2953 * at this point, we have a single ref, and since the
2954 * only place referencing this extent is a dead root
2955 * the reference count should never go higher.
2956 * So, we don't need to check it again
2957 */
31153d81 2958 if (*level == 1) {
017e5369 2959 ref = btrfs_lookup_leaf_ref(root, bytenr);
1a40e23b
ZY		 2960 if (ref && ref->generation != ptr_gen) {
2961 btrfs_free_leaf_ref(root, ref);
2962 ref = NULL;
2963 }
31153d81 2964 if (ref) {
e02119d5 2965 ret = cache_drop_leaf_ref(trans, root, ref);
31153d81
YZ		 2966 BUG_ON(ret);
2967 btrfs_remove_leaf_ref(root, ref);
bcc63abb 2968 btrfs_free_leaf_ref(root, ref);
31153d81
YZ		 2969 *level = 0;
2970 break;
2971 }
3bb1a1bc 2972 if (printk_ratelimit()) {
37d1aeee
CM		 2973 printk("leaf ref miss for bytenr %llu\n",
2974 (unsigned long long)bytenr);
3bb1a1bc 2975 }
31153d81 2976 }
db94535d 2977 next = btrfs_find_tree_block(root, bytenr, blocksize);
1259ab75 2978 if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
5f39d397 2979 free_extent_buffer(next);
333db94c 2980
ca7a79ad
CM		 2981 next = read_tree_block(root, bytenr, blocksize,
2982 ptr_gen);
e7a84565 2983 cond_resched();
f87f057b
CM		 2984#if 0
2985 /*
2986 * this is a debugging check and can go away
2987 * the ref should never go all the way down to 1
2988 * at this point
2989 */
e6dcd2dc
CM		 2990 ret = lookup_extent_ref(NULL, root, bytenr, blocksize,
2991 &refs);
e9d0b13b 2992 BUG_ON(ret);
f87f057b
CM		 2993 WARN_ON(refs != 1);
2994#endif
e9d0b13b 2995 }
5caf2a00 2996 WARN_ON(*level <= 0);
83e15a28 2997 if (path->nodes[*level-1])
5f39d397 2998 free_extent_buffer(path->nodes[*level-1]);
20524f02 2999 path->nodes[*level-1] = next;
5f39d397 3000 *level = btrfs_header_level(next);
20524f02 3001 path->slots[*level] = 0;
2dd3e67b 3002 cond_resched();
20524f02
CM		 3003 }
3004out:
5caf2a00
CM		 3005 WARN_ON(*level < 0);
3006 WARN_ON(*level >= BTRFS_MAX_LEVEL);
7bb86316
CM		 3007
3008 if (path->nodes[*level] == root->node) {
7bb86316 3009 parent = path->nodes[*level];
31153d81 3010 bytenr = path->nodes[*level]->start;
7bb86316
CM		 3011 } else {
3012 parent = path->nodes[*level + 1];
31153d81 3013 bytenr = btrfs_node_blockptr(parent, path->slots[*level + 1]);
7bb86316
CM		 3014 }
3015
31153d81
YZ		 3016 blocksize = btrfs_level_size(root, *level);
3017 root_owner = btrfs_header_owner(parent);
7bb86316 3018 root_gen = btrfs_header_generation(parent);
31153d81 3019
f87f057b 3020 mutex_lock(&root->fs_info->alloc_mutex);
31153d81 3021 ret = __btrfs_free_extent(trans, root, bytenr, blocksize,
31840ae1 3022 parent->start, root_owner, root_gen,
3bb1a1bc 3023 *level, 1);
31840ae1 3024 mutex_unlock(&root->fs_info->alloc_mutex);
5f39d397 3025 free_extent_buffer(path->nodes[*level]);
20524f02
CM		 3026 path->nodes[*level] = NULL;
3027 *level += 1;
3028 BUG_ON(ret);
f87f057b 3029
e7a84565 3030 cond_resched();
20524f02
CM		 3031 return 0;
3032}
3033
9aca1d51
CM		 3034/*
3035 * helper for dropping snapshots. This walks back up the tree in the path
3036 * to find the first node higher up where we haven't yet gone through
3037 * all the slots
3038 */
98ed5174
CM		 3039static int noinline walk_up_tree(struct btrfs_trans_handle *trans,
3040 struct btrfs_root *root,
3041 struct btrfs_path *path, int *level)
20524f02 3042{
7bb86316
CM		 3043 u64 root_owner;
3044 u64 root_gen;
3045 struct btrfs_root_item *root_item = &root->root_item;
20524f02
CM		 3046 int i;
3047 int slot;
3048 int ret;
9f3a7427 3049
234b63a0 3050 for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
20524f02 3051 slot = path->slots[i];
5f39d397
CM		 3052 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
3053 struct extent_buffer *node;
3054 struct btrfs_disk_key disk_key;
3055 node = path->nodes[i];
20524f02
CM		 3056 path->slots[i]++;
3057 *level = i;
9f3a7427 3058 WARN_ON(*level == 0);
5f39d397 3059 btrfs_node_key(node, &disk_key, path->slots[i]);
9f3a7427 3060 memcpy(&root_item->drop_progress,
5f39d397 3061 &disk_key, sizeof(disk_key));
9f3a7427 3062 root_item->drop_level = i;
20524f02
CM		 3063 return 0;
3064 } else {
31840ae1
ZY		 3065 struct extent_buffer *parent;
3066 if (path->nodes[*level] == root->node)
3067 parent = path->nodes[*level];
3068 else
3069 parent = path->nodes[*level + 1];
3070
3071 root_owner = btrfs_header_owner(parent);
3072 root_gen = btrfs_header_generation(parent);
e089f05c 3073 ret = btrfs_free_extent(trans, root,
db94535d 3074 path->nodes[*level]->start,
7bb86316 3075 path->nodes[*level]->len,
3bb1a1bc
YZ		 3076 parent->start, root_owner,
3077 root_gen, *level, 1);
6407bf6d 3078 BUG_ON(ret);
5f39d397 3079 free_extent_buffer(path->nodes[*level]);
83e15a28 3080 path->nodes[*level] = NULL;
20524f02 3081 *level = i + 1;
20524f02
CM		 3082 }
3083 }
3084 return 1;
3085}
3086
9aca1d51
CM		 3087/*
3088 * drop the reference count on the tree rooted at 'snap'. This traverses
3089 * the tree freeing any blocks that have a ref count of zero after being
3090 * decremented.
3091 */
e089f05c 3092int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
9f3a7427 3093 *root)
20524f02 3094{
3768f368 3095 int ret = 0;
9aca1d51 3096 int wret;
20524f02 3097 int level;
5caf2a00 3098 struct btrfs_path *path;
20524f02
CM		 3099 int i;
3100 int orig_level;
9f3a7427 3101 struct btrfs_root_item *root_item = &root->root_item;
20524f02 3102
a2135011 3103 WARN_ON(!mutex_is_locked(&root->fs_info->drop_mutex));
5caf2a00
CM		 3104 path = btrfs_alloc_path();
3105 BUG_ON(!path);
20524f02 3106
5f39d397 3107 level = btrfs_header_level(root->node);
20524f02 3108 orig_level = level;
9f3a7427
CM		 3109 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
3110 path->nodes[level] = root->node;
f510cfec 3111 extent_buffer_get(root->node);
9f3a7427
CM		 3112 path->slots[level] = 0;
3113 } else {
3114 struct btrfs_key key;
5f39d397
CM		 3115 struct btrfs_disk_key found_key;
3116 struct extent_buffer *node;
6702ed49 3117
9f3a7427 3118 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
6702ed49
CM		 3119 level = root_item->drop_level;
3120 path->lowest_level = level;
9f3a7427 3121 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6702ed49 3122 if (wret < 0) {
9f3a7427
CM		 3123 ret = wret;
3124 goto out;
3125 }
5f39d397
CM		 3126 node = path->nodes[level];
3127 btrfs_node_key(node, &found_key, path->slots[level]);
3128 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
3129 sizeof(found_key)));
7d9eb12c
CM		 3130 /*
3131 * unlock our path, this is safe because only this
3132 * function is allowed to delete this snapshot
3133 */
925baedd
CM		 3134 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
3135 if (path->nodes[i] && path->locks[i]) {
3136 path->locks[i] = 0;
3137 btrfs_tree_unlock(path->nodes[i]);
3138 }
3139 }
9f3a7427 3140 }
20524f02 3141 while(1) {
5caf2a00 3142 wret = walk_down_tree(trans, root, path, &level);
9aca1d51 3143 if (wret > 0)
20524f02 3144 break;
9aca1d51
CM		 3145 if (wret < 0)
3146 ret = wret;
3147
5caf2a00 3148 wret = walk_up_tree(trans, root, path, &level);
9aca1d51 3149 if (wret > 0)
20524f02 3150 break;
9aca1d51
CM		 3151 if (wret < 0)
3152 ret = wret;
e7a84565
CM		 3153 if (trans->transaction->in_commit) {
3154 ret = -EAGAIN;
3155 break;
3156 }
18e35e0a 3157 atomic_inc(&root->fs_info->throttle_gen);
017e5369 3158 wake_up(&root->fs_info->transaction_throttle);
20524f02 3159 }
83e15a28 3160 for (i = 0; i <= orig_level; i++) {
5caf2a00 3161 if (path->nodes[i]) {
5f39d397 3162 free_extent_buffer(path->nodes[i]);
0f82731f 3163 path->nodes[i] = NULL;
83e15a28 3164 }
20524f02 3165 }
9f3a7427 3166out:
5caf2a00 3167 btrfs_free_path(path);
9aca1d51 3168 return ret;
20524f02 3169}
9078a3e1 3170
8e7bf94f
CM		 3171static unsigned long calc_ra(unsigned long start, unsigned long last,
3172 unsigned long nr)
3173{
3174 return min(last, start + nr - 1);
3175}
3176
98ed5174
CM		 3177static int noinline relocate_inode_pages(struct inode *inode, u64 start,
3178 u64 len)
edbd8d4e
CM		 3179{
3180 u64 page_start;
3181 u64 page_end;
1a40e23b 3182 unsigned long first_index;
edbd8d4e 3183 unsigned long last_index;
edbd8d4e
CM		 3184 unsigned long i;
3185 struct page *page;
d1310b2e 3186 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4313b399 3187 struct file_ra_state *ra;
3eaa2885 3188 struct btrfs_ordered_extent *ordered;
1a40e23b
ZY		 3189 unsigned int total_read = 0;
3190 unsigned int total_dirty = 0;
3191 int ret = 0;
4313b399
CM		 3192
3193 ra = kzalloc(sizeof(*ra), GFP_NOFS);
edbd8d4e
CM		 3194
3195 mutex_lock(&inode->i_mutex);
1a40e23b 3196 first_index = start >> PAGE_CACHE_SHIFT;
edbd8d4e
CM		 3197 last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
3198
1a40e23b
ZY		 3199 /* make sure the dirty trick played by the caller work */
3200 ret = invalidate_inode_pages2_range(inode->i_mapping,
3201 first_index, last_index);
3202 if (ret)
3203 goto out_unlock;
8e7bf94f 3204
4313b399 3205 file_ra_state_init(ra, inode->i_mapping);
edbd8d4e 3206
1a40e23b
ZY		 3207 for (i = first_index ; i <= last_index; i++) {
3208 if (total_read % ra->ra_pages == 0) {
8e7bf94f 3209 btrfs_force_ra(inode->i_mapping, ra, NULL, i,
1a40e23b 3210 calc_ra(i, last_index, ra->ra_pages));
8e7bf94f
CM		 3211 }
3212 total_read++;
3eaa2885
CM		 3213again:
3214 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
1a40e23b 3215 BUG_ON(1);
edbd8d4e 3216 page = grab_cache_page(inode->i_mapping, i);
a061fc8d 3217 if (!page) {
1a40e23b 3218 ret = -ENOMEM;
edbd8d4e 3219 goto out_unlock;
a061fc8d 3220 }
edbd8d4e
CM		 3221 if (!PageUptodate(page)) {
3222 btrfs_readpage(NULL, page);
3223 lock_page(page);
3224 if (!PageUptodate(page)) {
3225 unlock_page(page);
3226 page_cache_release(page);
1a40e23b 3227 ret = -EIO;
edbd8d4e
CM		 3228 goto out_unlock;
3229 }
3230 }
ec44a35c 3231 wait_on_page_writeback(page);
3eaa2885 3232
edbd8d4e
CM		 3233 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
3234 page_end = page_start + PAGE_CACHE_SIZE - 1;
d1310b2e 3235 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
edbd8d4e 3236
3eaa2885
CM		 3237 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3238 if (ordered) {
3239 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3240 unlock_page(page);
3241 page_cache_release(page);
3242 btrfs_start_ordered_extent(inode, ordered, 1);
3243 btrfs_put_ordered_extent(ordered);
3244 goto again;
3245 }
3246 set_page_extent_mapped(page);
3247
ea8c2819 3248 btrfs_set_extent_delalloc(inode, page_start, page_end);
1a40e23b
ZY		 3249 if (i == first_index)
3250 set_extent_bits(io_tree, page_start, page_end,
3251 EXTENT_BOUNDARY, GFP_NOFS);
3252
a061fc8d 3253 set_page_dirty(page);
1a40e23b 3254 total_dirty++;
edbd8d4e 3255
d1310b2e 3256 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
edbd8d4e
CM		 3257 unlock_page(page);
3258 page_cache_release(page);
3259 }
3260
3261out_unlock:
ec44a35c 3262 kfree(ra);
edbd8d4e 3263 mutex_unlock(&inode->i_mutex);
1a40e23b
ZY		 3264 balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
3265 return ret;
edbd8d4e
CM		 3266}
3267
1a40e23b
ZY		 3268static int noinline relocate_data_extent(struct inode *reloc_inode,
3269 struct btrfs_key *extent_key,
3270 u64 offset)
3271{
3272 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
3273 struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
3274 struct extent_map *em;
bf4ef679 3275
1a40e23b
ZY		 3276 em = alloc_extent_map(GFP_NOFS);
3277 BUG_ON(!em || IS_ERR(em));
bf4ef679 3278
1a40e23b
ZY		 3279 em->start = extent_key->objectid - offset;
3280 em->len = extent_key->offset;
c8b97818 3281 em->block_len = extent_key->offset;
1a40e23b
ZY		 3282 em->block_start = extent_key->objectid;
3283 em->bdev = root->fs_info->fs_devices->latest_bdev;
3284 set_bit(EXTENT_FLAG_PINNED, &em->flags);
3285
3286 /* setup extent map to cheat btrfs_readpage */
3287 mutex_lock(&BTRFS_I(reloc_inode)->extent_mutex);
3288 while (1) {
3289 int ret;
3290 spin_lock(&em_tree->lock);
3291 ret = add_extent_mapping(em_tree, em);
3292 spin_unlock(&em_tree->lock);
3293 if (ret != -EEXIST) {
3294 free_extent_map(em);
bf4ef679
CM		 3295 break;
3296 }
1a40e23b
ZY		 3297 btrfs_drop_extent_cache(reloc_inode, em->start,
3298 em->start + em->len - 1, 0);
bf4ef679 3299 }
1a40e23b 3300 mutex_unlock(&BTRFS_I(reloc_inode)->extent_mutex);
bf4ef679 3301
1a40e23b
ZY		 3302 return relocate_inode_pages(reloc_inode, extent_key->objectid - offset,
3303 extent_key->offset);
3304}
edbd8d4e 3305
1a40e23b
ZY		 3306struct btrfs_ref_path {
3307 u64 extent_start;
3308 u64 nodes[BTRFS_MAX_LEVEL];
3309 u64 root_objectid;
3310 u64 root_generation;
3311 u64 owner_objectid;
1a40e23b
ZY		 3312 u32 num_refs;
3313 int lowest_level;
3314 int current_level;
3315};
7d9eb12c 3316
1a40e23b 3317struct disk_extent {
c8b97818 3318 u64 ram_bytes;
1a40e23b
ZY		 3319 u64 disk_bytenr;
3320 u64 disk_num_bytes;
3321 u64 offset;
3322 u64 num_bytes;
c8b97818
CM		 3323 u8 compression;
3324 u8 encryption;
3325 u16 other_encoding;
1a40e23b 3326};
4313b399 3327
1a40e23b
ZY		 3328static int is_cowonly_root(u64 root_objectid)
3329{
3330 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
3331 root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
3332 root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
3333 root_objectid == BTRFS_DEV_TREE_OBJECTID ||
3334 root_objectid == BTRFS_TREE_LOG_OBJECTID)
3335 return 1;
3336 return 0;
3337}
edbd8d4e 3338
1a40e23b
ZY		 3339static int noinline __next_ref_path(struct btrfs_trans_handle *trans,
3340 struct btrfs_root *extent_root,
3341 struct btrfs_ref_path *ref_path,
3342 int first_time)
3343{
3344 struct extent_buffer *leaf;
3345 struct btrfs_path *path;
3346 struct btrfs_extent_ref *ref;
3347 struct btrfs_key key;
3348 struct btrfs_key found_key;
3349 u64 bytenr;
3350 u32 nritems;
3351 int level;
3352 int ret = 1;
edbd8d4e 3353
1a40e23b
ZY		 3354 path = btrfs_alloc_path();
3355 if (!path)
3356 return -ENOMEM;
bf4ef679 3357
1a40e23b 3358 mutex_lock(&extent_root->fs_info->alloc_mutex);
0ef3e66b 3359
1a40e23b
ZY		 3360 if (first_time) {
3361 ref_path->lowest_level = -1;
3362 ref_path->current_level = -1;
3363 goto walk_up;
3364 }
3365walk_down:
3366 level = ref_path->current_level - 1;
3367 while (level >= -1) {
3368 u64 parent;
3369 if (level < ref_path->lowest_level)
3370 break;
bf4ef679 3371
1a40e23b
ZY		 3372 if (level >= 0) {
3373 bytenr = ref_path->nodes[level];
3374 } else {
3375 bytenr = ref_path->extent_start;
3376 }
3377 BUG_ON(bytenr == 0);
bf4ef679 3378
1a40e23b
ZY		 3379 parent = ref_path->nodes[level + 1];
3380 ref_path->nodes[level + 1] = 0;
3381 ref_path->current_level = level;
3382 BUG_ON(parent == 0);
0ef3e66b 3383
1a40e23b
ZY		 3384 key.objectid = bytenr;
3385 key.offset = parent + 1;
3386 key.type = BTRFS_EXTENT_REF_KEY;
edbd8d4e 3387
1a40e23b
ZY		 3388 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
3389 if (ret < 0)
edbd8d4e 3390 goto out;
1a40e23b 3391 BUG_ON(ret == 0);
7d9eb12c 3392
1a40e23b
ZY		 3393 leaf = path->nodes[0];
3394 nritems = btrfs_header_nritems(leaf);
3395 if (path->slots[0] >= nritems) {
3396 ret = btrfs_next_leaf(extent_root, path);
3397 if (ret < 0)
3398 goto out;
3399 if (ret > 0)
3400 goto next;
3401 leaf = path->nodes[0];
3402 }
0ef3e66b 3403
1a40e23b
ZY		 3404 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3405 if (found_key.objectid == bytenr &&
3406 found_key.type == BTRFS_EXTENT_REF_KEY)
3407 goto found;
3408next:
3409 level--;
3410 btrfs_release_path(extent_root, path);
3411 if (need_resched()) {
3412 mutex_unlock(&extent_root->fs_info->alloc_mutex);
3413 cond_resched();
3414 mutex_lock(&extent_root->fs_info->alloc_mutex);
3415 }
3416 }
3417 /* reached lowest level */
3418 ret = 1;
3419 goto out;
3420walk_up:
3421 level = ref_path->current_level;
3422 while (level < BTRFS_MAX_LEVEL - 1) {
3423 u64 ref_objectid;
3424 if (level >= 0) {
3425 bytenr = ref_path->nodes[level];
3426 } else {
3427 bytenr = ref_path->extent_start;
3428 }
3429 BUG_ON(bytenr == 0);
edbd8d4e 3430
1a40e23b
ZY		 3431 key.objectid = bytenr;
3432 key.offset = 0;
3433 key.type = BTRFS_EXTENT_REF_KEY;
edbd8d4e 3434
1a40e23b
ZY		 3435 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
3436 if (ret < 0)
3437 goto out;
edbd8d4e 3438
1a40e23b
ZY		 3439 leaf = path->nodes[0];
3440 nritems = btrfs_header_nritems(leaf);
3441 if (path->slots[0] >= nritems) {
3442 ret = btrfs_next_leaf(extent_root, path);
3443 if (ret < 0)
3444 goto out;
3445 if (ret > 0) {
3446 /* the extent was freed by someone */
3447 if (ref_path->lowest_level == level)
3448 goto out;
3449 btrfs_release_path(extent_root, path);
3450 goto walk_down;
3451 }
3452 leaf = path->nodes[0];
3453 }
edbd8d4e 3454
1a40e23b
ZY		 3455 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3456 if (found_key.objectid != bytenr ||
3457 found_key.type != BTRFS_EXTENT_REF_KEY) {
3458 /* the extent was freed by someone */
3459 if (ref_path->lowest_level == level) {
3460 ret = 1;
3461 goto out;
3462 }
3463 btrfs_release_path(extent_root, path);
3464 goto walk_down;
3465 }
3466found:
3467 ref = btrfs_item_ptr(leaf, path->slots[0],
3468 struct btrfs_extent_ref);
3469 ref_objectid = btrfs_ref_objectid(leaf, ref);
3470 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
3471 if (first_time) {
3472 level = (int)ref_objectid;
3473 BUG_ON(level >= BTRFS_MAX_LEVEL);
3474 ref_path->lowest_level = level;
3475 ref_path->current_level = level;
3476 ref_path->nodes[level] = bytenr;
3477 } else {
3478 WARN_ON(ref_objectid != level);
3479 }
3480 } else {
3481 WARN_ON(level != -1);
3482 }
3483 first_time = 0;
bf4ef679 3484
1a40e23b
ZY		 3485 if (ref_path->lowest_level == level) {
3486 ref_path->owner_objectid = ref_objectid;
1a40e23b
ZY		 3487 ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
3488 }
bf4ef679 3489
7d9eb12c 3490 /*
1a40e23b
ZY		 3491 * the block is tree root or the block isn't in reference
3492 * counted tree.
7d9eb12c 3493 */
1a40e23b
ZY		 3494 if (found_key.objectid == found_key.offset ||
3495 is_cowonly_root(btrfs_ref_root(leaf, ref))) {
3496 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
3497 ref_path->root_generation =
3498 btrfs_ref_generation(leaf, ref);
3499 if (level < 0) {
3500 /* special reference from the tree log */
3501 ref_path->nodes[0] = found_key.offset;
3502 ref_path->current_level = 0;
3503 }
3504 ret = 0;
3505 goto out;
3506 }
7d9eb12c 3507
1a40e23b
ZY		 3508 level++;
3509 BUG_ON(ref_path->nodes[level] != 0);
3510 ref_path->nodes[level] = found_key.offset;
3511 ref_path->current_level = level;
bf4ef679 3512
1a40e23b
ZY		 3513 /*
3514 * the reference was created in the running transaction,
3515 * no need to continue walking up.
3516 */
3517 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
3518 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
3519 ref_path->root_generation =
3520 btrfs_ref_generation(leaf, ref);
3521 ret = 0;
3522 goto out;
7d9eb12c
CM		 3523 }
3524
1a40e23b
ZY		 3525 btrfs_release_path(extent_root, path);
3526 if (need_resched()) {
7d9eb12c 3527 mutex_unlock(&extent_root->fs_info->alloc_mutex);
1a40e23b
ZY		 3528 cond_resched();
3529 mutex_lock(&extent_root->fs_info->alloc_mutex);
3530 }
7d9eb12c 3531 }
1a40e23b
ZY		 3532 /* reached max tree level, but no tree root found. */
3533 BUG();
edbd8d4e 3534out:
1a40e23b
ZY		 3535 mutex_unlock(&extent_root->fs_info->alloc_mutex);
3536 btrfs_free_path(path);
3537 return ret;
edbd8d4e
CM		 3538}
3539
1a40e23b
ZY		 3540static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
3541 struct btrfs_root *extent_root,
3542 struct btrfs_ref_path *ref_path,
3543 u64 extent_start)
a061fc8d 3544{
1a40e23b
ZY		 3545 memset(ref_path, 0, sizeof(*ref_path));
3546 ref_path->extent_start = extent_start;
a061fc8d 3547
1a40e23b 3548 return __next_ref_path(trans, extent_root, ref_path, 1);
a061fc8d
CM		 3549}
3550
1a40e23b
ZY		 3551static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
3552 struct btrfs_root *extent_root,
3553 struct btrfs_ref_path *ref_path)
edbd8d4e 3554{
1a40e23b
ZY		 3555 return __next_ref_path(trans, extent_root, ref_path, 0);
3556}
3557
3558static int noinline get_new_locations(struct inode *reloc_inode,
3559 struct btrfs_key *extent_key,
3560 u64 offset, int no_fragment,
3561 struct disk_extent **extents,
3562 int *nr_extents)
3563{
3564 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
3565 struct btrfs_path *path;
3566 struct btrfs_file_extent_item *fi;
edbd8d4e 3567 struct extent_buffer *leaf;
1a40e23b
ZY		 3568 struct disk_extent *exts = *extents;
3569 struct btrfs_key found_key;
3570 u64 cur_pos;
3571 u64 last_byte;
edbd8d4e 3572 u32 nritems;
1a40e23b
ZY		 3573 int nr = 0;
3574 int max = *nr_extents;
3575 int ret;
edbd8d4e 3576
1a40e23b
ZY		 3577 WARN_ON(!no_fragment && *extents);
3578 if (!exts) {
3579 max = 1;
3580 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
3581 if (!exts)
3582 return -ENOMEM;
a061fc8d 3583 }
edbd8d4e 3584
1a40e23b
ZY		 3585 path = btrfs_alloc_path();
3586 BUG_ON(!path);
edbd8d4e 3587
1a40e23b
ZY		 3588 cur_pos = extent_key->objectid - offset;
3589 last_byte = extent_key->objectid + extent_key->offset;
3590 ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
3591 cur_pos, 0);
3592 if (ret < 0)
3593 goto out;
3594 if (ret > 0) {
3595 ret = -ENOENT;
3596 goto out;
3597 }
edbd8d4e 3598
1a40e23b 3599 while (1) {
edbd8d4e
CM		 3600 leaf = path->nodes[0];
3601 nritems = btrfs_header_nritems(leaf);
1a40e23b
ZY		 3602 if (path->slots[0] >= nritems) {
3603 ret = btrfs_next_leaf(root, path);
a061fc8d
CM		 3604 if (ret < 0)
3605 goto out;
1a40e23b
ZY		 3606 if (ret > 0)
3607 break;
bf4ef679 3608 leaf = path->nodes[0];
a061fc8d 3609 }
edbd8d4e
CM		 3610
3611 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1a40e23b
ZY		 3612 if (found_key.offset != cur_pos ||
3613 found_key.type != BTRFS_EXTENT_DATA_KEY ||
3614 found_key.objectid != reloc_inode->i_ino)
edbd8d4e
CM		 3615 break;
3616
1a40e23b
ZY		 3617 fi = btrfs_item_ptr(leaf, path->slots[0],
3618 struct btrfs_file_extent_item);
3619 if (btrfs_file_extent_type(leaf, fi) !=
3620 BTRFS_FILE_EXTENT_REG ||
3621 btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
edbd8d4e 3622 break;
1a40e23b
ZY		 3623
3624 if (nr == max) {
3625 struct disk_extent *old = exts;
3626 max *= 2;
3627 exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
3628 memcpy(exts, old, sizeof(*exts) * nr);
3629 if (old != *extents)
3630 kfree(old);
a061fc8d 3631 }
edbd8d4e 3632
1a40e23b
ZY		 3633 exts[nr].disk_bytenr =
3634 btrfs_file_extent_disk_bytenr(leaf, fi);
3635 exts[nr].disk_num_bytes =
3636 btrfs_file_extent_disk_num_bytes(leaf, fi);
3637 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
3638 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
c8b97818
CM		 3639 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
3640 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
3641 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
3642 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
3643 fi);
1a40e23b
ZY		 3644 WARN_ON(exts[nr].offset > 0);
3645 WARN_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
edbd8d4e 3646
1a40e23b
ZY		 3647 cur_pos += exts[nr].num_bytes;
3648 nr++;
3649
3650 if (cur_pos + offset >= last_byte)
3651 break;
3652
3653 if (no_fragment) {
3654 ret = 1;
edbd8d4e 3655 goto out;
1a40e23b
ZY		 3656 }
3657 path->slots[0]++;
3658 }
3659
3660 WARN_ON(cur_pos + offset > last_byte);
3661 if (cur_pos + offset < last_byte) {
3662 ret = -ENOENT;
3663 goto out;
edbd8d4e
CM		 3664 }
3665 ret = 0;
3666out:
1a40e23b
ZY		 3667 btrfs_free_path(path);
3668 if (ret) {
3669 if (exts != *extents)
3670 kfree(exts);
3671 } else {
3672 *extents = exts;
3673 *nr_extents = nr;
3674 }
3675 return ret;
3676}
3677
3678static int noinline replace_one_extent(struct btrfs_trans_handle *trans,
3679 struct btrfs_root *root,
3680 struct btrfs_path *path,
3681 struct btrfs_key *extent_key,
3682 struct btrfs_key *leaf_key,
3683 struct btrfs_ref_path *ref_path,
3684 struct disk_extent *new_extents,
3685 int nr_extents)
3686{
3687 struct extent_buffer *leaf;
3688 struct btrfs_file_extent_item *fi;
3689 struct inode *inode = NULL;
3690 struct btrfs_key key;
3691 u64 lock_start = 0;
3692 u64 lock_end = 0;
3693 u64 num_bytes;
3694 u64 ext_offset;
3695 u64 first_pos;
3696 u32 nritems;
3bb1a1bc 3697 int nr_scaned = 0;
1a40e23b
ZY		 3698 int extent_locked = 0;
3699 int ret;
3700
3bb1a1bc
YZ		 3701 memcpy(&key, leaf_key, sizeof(key));
3702 first_pos = INT_LIMIT(loff_t) - extent_key->offset;
1a40e23b 3703 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
3bb1a1bc
YZ		 3704 if (key.objectid < ref_path->owner_objectid ||
3705 (key.objectid == ref_path->owner_objectid &&
3706 key.type < BTRFS_EXTENT_DATA_KEY)) {
3707 key.objectid = ref_path->owner_objectid;
3708 key.type = BTRFS_EXTENT_DATA_KEY;
3709 key.offset = 0;
3710 }
1a40e23b
ZY		 3711 }
3712
3713 while (1) {
3714 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
3715 if (ret < 0)
3716 goto out;
3717
3718 leaf = path->nodes[0];
3719 nritems = btrfs_header_nritems(leaf);
3720next:
3721 if (extent_locked && ret > 0) {
3722 /*
3723 * the file extent item was modified by someone
3724 * before the extent got locked.
3725 */
3726 mutex_unlock(&BTRFS_I(inode)->extent_mutex);
3727 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
3728 lock_end, GFP_NOFS);
3729 extent_locked = 0;
3730 }
3731
3732 if (path->slots[0] >= nritems) {
3bb1a1bc 3733 if (++nr_scaned > 2)
1a40e23b
ZY		 3734 break;
3735
3736 BUG_ON(extent_locked);
3737 ret = btrfs_next_leaf(root, path);
3738 if (ret < 0)
3739 goto out;
3740 if (ret > 0)
3741 break;
3742 leaf = path->nodes[0];
3743 nritems = btrfs_header_nritems(leaf);
3744 }
3745
3746 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3747
3748 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
3749 if ((key.objectid > ref_path->owner_objectid) ||
3750 (key.objectid == ref_path->owner_objectid &&
3751 key.type > BTRFS_EXTENT_DATA_KEY) ||
3752 (key.offset >= first_pos + extent_key->offset))
3753 break;
3754 }
3755
3756 if (inode && key.objectid != inode->i_ino) {
3757 BUG_ON(extent_locked);
3758 btrfs_release_path(root, path);
3759 mutex_unlock(&inode->i_mutex);
3760 iput(inode);
3761 inode = NULL;
3762 continue;
3763 }
3764
3765 if (key.type != BTRFS_EXTENT_DATA_KEY) {
3766 path->slots[0]++;
3767 ret = 1;
3768 goto next;
3769 }
3770 fi = btrfs_item_ptr(leaf, path->slots[0],
3771 struct btrfs_file_extent_item);
3772 if ((btrfs_file_extent_type(leaf, fi) !=
3773 BTRFS_FILE_EXTENT_REG) ||
3774 (btrfs_file_extent_disk_bytenr(leaf, fi) !=
3775 extent_key->objectid)) {
3776 path->slots[0]++;
3777 ret = 1;
3778 goto next;
3779 }
3780
3781 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
3782 ext_offset = btrfs_file_extent_offset(leaf, fi);
3783
3784 if (first_pos > key.offset - ext_offset)
3785 first_pos = key.offset - ext_offset;
3786
3787 if (!extent_locked) {
3788 lock_start = key.offset;
3789 lock_end = lock_start + num_bytes - 1;
3790 } else {
3791 BUG_ON(lock_start != key.offset);
3792 BUG_ON(lock_end - lock_start + 1 < num_bytes);
3793 }
3794
3795 if (!inode) {
3796 btrfs_release_path(root, path);
3797
3798 inode = btrfs_iget_locked(root->fs_info->sb,
3799 key.objectid, root);
3800 if (inode->i_state & I_NEW) {
3801 BTRFS_I(inode)->root = root;
3802 BTRFS_I(inode)->location.objectid =
3803 key.objectid;
3804 BTRFS_I(inode)->location.type =
3805 BTRFS_INODE_ITEM_KEY;
3806 BTRFS_I(inode)->location.offset = 0;
3807 btrfs_read_locked_inode(inode);
3808 unlock_new_inode(inode);
3809 }
3810 /*
3811 * some code call btrfs_commit_transaction while
3812 * holding the i_mutex, so we can't use mutex_lock
3813 * here.
3814 */
3815 if (is_bad_inode(inode) ||
3816 !mutex_trylock(&inode->i_mutex)) {
3817 iput(inode);
3818 inode = NULL;
3819 key.offset = (u64)-1;
3820 goto skip;
3821 }
3822 }
3823
3824 if (!extent_locked) {
3825 struct btrfs_ordered_extent *ordered;
3826
3827 btrfs_release_path(root, path);
3828
3829 lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
3830 lock_end, GFP_NOFS);
3831 ordered = btrfs_lookup_first_ordered_extent(inode,
3832 lock_end);
3833 if (ordered &&
3834 ordered->file_offset <= lock_end &&
3835 ordered->file_offset + ordered->len > lock_start) {
3836 unlock_extent(&BTRFS_I(inode)->io_tree,
3837 lock_start, lock_end, GFP_NOFS);
3838 btrfs_start_ordered_extent(inode, ordered, 1);
3839 btrfs_put_ordered_extent(ordered);
3840 key.offset += num_bytes;
3841 goto skip;
3842 }
3843 if (ordered)
3844 btrfs_put_ordered_extent(ordered);
3845
3846 mutex_lock(&BTRFS_I(inode)->extent_mutex);
3847 extent_locked = 1;
3848 continue;
3849 }
3850
3851 if (nr_extents == 1) {
3852 /* update extent pointer in place */
3853 btrfs_set_file_extent_generation(leaf, fi,
3854 trans->transid);
3855 btrfs_set_file_extent_disk_bytenr(leaf, fi,
3856 new_extents[0].disk_bytenr);
3857 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
3858 new_extents[0].disk_num_bytes);
c8b97818
CM		 3859 btrfs_set_file_extent_ram_bytes(leaf, fi,
3860 new_extents[0].ram_bytes);
1a40e23b
ZY		 3861 ext_offset += new_extents[0].offset;
3862 btrfs_set_file_extent_offset(leaf, fi, ext_offset);
3863 btrfs_mark_buffer_dirty(leaf);
3864
3865 btrfs_drop_extent_cache(inode, key.offset,
3866 key.offset + num_bytes - 1, 0);
3867
3868 ret = btrfs_inc_extent_ref(trans, root,
3869 new_extents[0].disk_bytenr,
3870 new_extents[0].disk_num_bytes,
3871 leaf->start,
3872 root->root_key.objectid,
3873 trans->transid,
3bb1a1bc 3874 key.objectid);
1a40e23b
ZY		 3875 BUG_ON(ret);
3876
3877 ret = btrfs_free_extent(trans, root,
3878 extent_key->objectid,
3879 extent_key->offset,
3880 leaf->start,
3881 btrfs_header_owner(leaf),
3882 btrfs_header_generation(leaf),
3bb1a1bc 3883 key.objectid, 0);
1a40e23b
ZY		 3884 BUG_ON(ret);
3885
3886 btrfs_release_path(root, path);
3887 key.offset += num_bytes;
3888 } else {
3889 u64 alloc_hint;
3890 u64 extent_len;
3891 int i;
3892 /*
3893 * drop old extent pointer at first, then insert the
3894 * new pointers one bye one
3895 */
3896 btrfs_release_path(root, path);
3897 ret = btrfs_drop_extents(trans, root, inode, key.offset,
3898 key.offset + num_bytes,
3899 key.offset, &alloc_hint);
3900 BUG_ON(ret);
3901
3902 for (i = 0; i < nr_extents; i++) {
3903 if (ext_offset >= new_extents[i].num_bytes) {
3904 ext_offset -= new_extents[i].num_bytes;
3905 continue;
3906 }
3907 extent_len = min(new_extents[i].num_bytes -
3908 ext_offset, num_bytes);
3909
3910 ret = btrfs_insert_empty_item(trans, root,
3911 path, &key,
3912 sizeof(*fi));
3913 BUG_ON(ret);
3914
3915 leaf = path->nodes[0];
3916 fi = btrfs_item_ptr(leaf, path->slots[0],
3917 struct btrfs_file_extent_item);
3918 btrfs_set_file_extent_generation(leaf, fi,
3919 trans->transid);
3920 btrfs_set_file_extent_type(leaf, fi,
3921 BTRFS_FILE_EXTENT_REG);
3922 btrfs_set_file_extent_disk_bytenr(leaf, fi,
3923 new_extents[i].disk_bytenr);
3924 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
3925 new_extents[i].disk_num_bytes);
c8b97818
CM		 3926 btrfs_set_file_extent_ram_bytes(leaf, fi,
3927 new_extents[i].ram_bytes);
3928
3929 btrfs_set_file_extent_compression(leaf, fi,
3930 new_extents[i].compression);
3931 btrfs_set_file_extent_encryption(leaf, fi,
3932 new_extents[i].encryption);
3933 btrfs_set_file_extent_other_encoding(leaf, fi,
3934 new_extents[i].other_encoding);
3935
1a40e23b
ZY		 3936 btrfs_set_file_extent_num_bytes(leaf, fi,
3937 extent_len);
3938 ext_offset += new_extents[i].offset;
3939 btrfs_set_file_extent_offset(leaf, fi,
3940 ext_offset);
3941 btrfs_mark_buffer_dirty(leaf);
3942
3943 btrfs_drop_extent_cache(inode, key.offset,
3944 key.offset + extent_len - 1, 0);
3945
3946 ret = btrfs_inc_extent_ref(trans, root,
3947 new_extents[i].disk_bytenr,
3948 new_extents[i].disk_num_bytes,
3949 leaf->start,
3950 root->root_key.objectid,
3bb1a1bc 3951 trans->transid, key.objectid);
1a40e23b
ZY		 3952 BUG_ON(ret);
3953 btrfs_release_path(root, path);
3954
a76a3cd4 3955 inode_add_bytes(inode, extent_len);
1a40e23b
ZY		 3956
3957 ext_offset = 0;
3958 num_bytes -= extent_len;
3959 key.offset += extent_len;
3960
3961 if (num_bytes == 0)
3962 break;
3963 }
3964 BUG_ON(i >= nr_extents);
3965 }
3966
3967 if (extent_locked) {
3968 mutex_unlock(&BTRFS_I(inode)->extent_mutex);
3969 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
3970 lock_end, GFP_NOFS);
3971 extent_locked = 0;
3972 }
3973skip:
3974 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
3975 key.offset >= first_pos + extent_key->offset)
3976 break;
3977
3978 cond_resched();
3979 }
3980 ret = 0;
3981out:
3982 btrfs_release_path(root, path);
3983 if (inode) {
3984 mutex_unlock(&inode->i_mutex);
3985 if (extent_locked) {
3986 mutex_unlock(&BTRFS_I(inode)->extent_mutex);
3987 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
3988 lock_end, GFP_NOFS);
3989 }
3990 iput(inode);
3991 }
3992 return ret;
3993}
3994
3995int btrfs_add_reloc_mapping(struct btrfs_root *root, u64 orig_bytenr,
3996 u64 num_bytes, u64 new_bytenr)
3997{
3998 set_extent_bits(&root->fs_info->reloc_mapping_tree,
3999 orig_bytenr, orig_bytenr + num_bytes - 1,
4000 EXTENT_LOCKED, GFP_NOFS);
4001 set_state_private(&root->fs_info->reloc_mapping_tree,
4002 orig_bytenr, new_bytenr);
4003 return 0;
4004}
4005
4006int btrfs_get_reloc_mapping(struct btrfs_root *root, u64 orig_bytenr,
4007 u64 num_bytes, u64 *new_bytenr)
4008{
4009 u64 bytenr;
4010 u64 cur_bytenr = orig_bytenr;
4011 u64 prev_bytenr = orig_bytenr;
4012 int ret;
4013
4014 while (1) {
4015 ret = get_state_private(&root->fs_info->reloc_mapping_tree,
4016 cur_bytenr, &bytenr);
4017 if (ret)
4018 break;
4019 prev_bytenr = cur_bytenr;
4020 cur_bytenr = bytenr;
4021 }
4022
4023 if (orig_bytenr == cur_bytenr)
4024 return -ENOENT;
4025
4026 if (prev_bytenr != orig_bytenr) {
4027 set_state_private(&root->fs_info->reloc_mapping_tree,
4028 orig_bytenr, cur_bytenr);
4029 }
4030 *new_bytenr = cur_bytenr;
4031 return 0;
4032}
4033
4034void btrfs_free_reloc_mappings(struct btrfs_root *root)
4035{
4036 clear_extent_bits(&root->fs_info->reloc_mapping_tree,
4037 0, (u64)-1, -1, GFP_NOFS);
4038}
4039
4040int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
4041 struct btrfs_root *root,
4042 struct extent_buffer *buf, u64 orig_start)
4043{
4044 int level;
4045 int ret;
4046
4047 BUG_ON(btrfs_header_generation(buf) != trans->transid);
4048 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
4049
4050 level = btrfs_header_level(buf);
4051 if (level == 0) {
4052 struct btrfs_leaf_ref *ref;
4053 struct btrfs_leaf_ref *orig_ref;
4054
4055 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
4056 if (!orig_ref)
4057 return -ENOENT;
4058
4059 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
4060 if (!ref) {
4061 btrfs_free_leaf_ref(root, orig_ref);
4062 return -ENOMEM;
4063 }
4064
4065 ref->nritems = orig_ref->nritems;
4066 memcpy(ref->extents, orig_ref->extents,
4067 sizeof(ref->extents[0]) * ref->nritems);
4068
4069 btrfs_free_leaf_ref(root, orig_ref);
4070
4071 ref->root_gen = trans->transid;
4072 ref->bytenr = buf->start;
4073 ref->owner = btrfs_header_owner(buf);
4074 ref->generation = btrfs_header_generation(buf);
4075 ret = btrfs_add_leaf_ref(root, ref, 0);
4076 WARN_ON(ret);
4077 btrfs_free_leaf_ref(root, ref);
4078 }
4079 return 0;
4080}
4081
4082static int noinline invalidate_extent_cache(struct btrfs_root *root,
4083 struct extent_buffer *leaf,
4084 struct btrfs_block_group_cache *group,
4085 struct btrfs_root *target_root)
4086{
4087 struct btrfs_key key;
4088 struct inode *inode = NULL;
4089 struct btrfs_file_extent_item *fi;
4090 u64 num_bytes;
4091 u64 skip_objectid = 0;
4092 u32 nritems;
4093 u32 i;
4094
4095 nritems = btrfs_header_nritems(leaf);
4096 for (i = 0; i < nritems; i++) {
4097 btrfs_item_key_to_cpu(leaf, &key, i);
4098 if (key.objectid == skip_objectid ||
4099 key.type != BTRFS_EXTENT_DATA_KEY)
4100 continue;
4101 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4102 if (btrfs_file_extent_type(leaf, fi) ==
4103 BTRFS_FILE_EXTENT_INLINE)
4104 continue;
4105 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
4106 continue;
4107 if (!inode || inode->i_ino != key.objectid) {
4108 iput(inode);
4109 inode = btrfs_ilookup(target_root->fs_info->sb,
4110 key.objectid, target_root, 1);
4111 }
4112 if (!inode) {
4113 skip_objectid = key.objectid;
4114 continue;
4115 }
4116 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4117
4118 lock_extent(&BTRFS_I(inode)->io_tree, key.offset,
4119 key.offset + num_bytes - 1, GFP_NOFS);
4120 mutex_lock(&BTRFS_I(inode)->extent_mutex);
4121 btrfs_drop_extent_cache(inode, key.offset,
4122 key.offset + num_bytes - 1, 1);
4123 mutex_unlock(&BTRFS_I(inode)->extent_mutex);
4124 unlock_extent(&BTRFS_I(inode)->io_tree, key.offset,
4125 key.offset + num_bytes - 1, GFP_NOFS);
4126 cond_resched();
4127 }
4128 iput(inode);
4129 return 0;
4130}
4131
4132static int noinline replace_extents_in_leaf(struct btrfs_trans_handle *trans,
4133 struct btrfs_root *root,
4134 struct extent_buffer *leaf,
4135 struct btrfs_block_group_cache *group,
4136 struct inode *reloc_inode)
4137{
4138 struct btrfs_key key;
4139 struct btrfs_key extent_key;
4140 struct btrfs_file_extent_item *fi;
4141 struct btrfs_leaf_ref *ref;
4142 struct disk_extent *new_extent;
4143 u64 bytenr;
4144 u64 num_bytes;
4145 u32 nritems;
4146 u32 i;
4147 int ext_index;
4148 int nr_extent;
4149 int ret;
4150
4151 new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
4152 BUG_ON(!new_extent);
4153
4154 ref = btrfs_lookup_leaf_ref(root, leaf->start);
4155 BUG_ON(!ref);
4156
4157 ext_index = -1;
4158 nritems = btrfs_header_nritems(leaf);
4159 for (i = 0; i < nritems; i++) {
4160 btrfs_item_key_to_cpu(leaf, &key, i);
4161 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
4162 continue;
4163 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4164 if (btrfs_file_extent_type(leaf, fi) ==
4165 BTRFS_FILE_EXTENT_INLINE)
4166 continue;
4167 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
4168 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
4169 if (bytenr == 0)
4170 continue;
4171
4172 ext_index++;
4173 if (bytenr >= group->key.objectid + group->key.offset ||
4174 bytenr + num_bytes <= group->key.objectid)
4175 continue;
4176
4177 extent_key.objectid = bytenr;
4178 extent_key.offset = num_bytes;
4179 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
4180 nr_extent = 1;
4181 ret = get_new_locations(reloc_inode, &extent_key,
4182 group->key.objectid, 1,
4183 &new_extent, &nr_extent);
4184 if (ret > 0)
4185 continue;
4186 BUG_ON(ret < 0);
4187
4188 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
4189 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
4190 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
4191 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
4192
4193 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
c8b97818
CM		 4194 btrfs_set_file_extent_ram_bytes(leaf, fi,
4195 new_extent->ram_bytes);
1a40e23b
ZY		 4196 btrfs_set_file_extent_disk_bytenr(leaf, fi,
4197 new_extent->disk_bytenr);
4198 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4199 new_extent->disk_num_bytes);
4200 new_extent->offset += btrfs_file_extent_offset(leaf, fi);
4201 btrfs_set_file_extent_offset(leaf, fi, new_extent->offset);
4202 btrfs_mark_buffer_dirty(leaf);
4203
4204 ret = btrfs_inc_extent_ref(trans, root,
4205 new_extent->disk_bytenr,
4206 new_extent->disk_num_bytes,
4207 leaf->start,
4208 root->root_key.objectid,
3bb1a1bc 4209 trans->transid, key.objectid);
1a40e23b
ZY		 4210 BUG_ON(ret);
4211 ret = btrfs_free_extent(trans, root,
4212 bytenr, num_bytes, leaf->start,
4213 btrfs_header_owner(leaf),
4214 btrfs_header_generation(leaf),
3bb1a1bc 4215 key.objectid, 0);
1a40e23b
ZY		 4216 BUG_ON(ret);
4217 cond_resched();
4218 }
4219 kfree(new_extent);
4220 BUG_ON(ext_index + 1 != ref->nritems);
4221 btrfs_free_leaf_ref(root, ref);
4222 return 0;
4223}
4224
4225int btrfs_free_reloc_root(struct btrfs_root *root)
4226{
4227 struct btrfs_root *reloc_root;
4228
4229 if (root->reloc_root) {
4230 reloc_root = root->reloc_root;
4231 root->reloc_root = NULL;
4232 list_add(&reloc_root->dead_list,
4233 &root->fs_info->dead_reloc_roots);
4234 }
4235 return 0;
4236}
4237
4238int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
4239{
4240 struct btrfs_trans_handle *trans;
4241 struct btrfs_root *reloc_root;
4242 struct btrfs_root *prev_root = NULL;
4243 struct list_head dead_roots;
4244 int ret;
4245 unsigned long nr;
4246
4247 INIT_LIST_HEAD(&dead_roots);
4248 list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
4249
4250 while (!list_empty(&dead_roots)) {
4251 reloc_root = list_entry(dead_roots.prev,
4252 struct btrfs_root, dead_list);
4253 list_del_init(&reloc_root->dead_list);
4254
4255 BUG_ON(reloc_root->commit_root != NULL);
4256 while (1) {
4257 trans = btrfs_join_transaction(root, 1);
4258 BUG_ON(!trans);
4259
4260 mutex_lock(&root->fs_info->drop_mutex);
4261 ret = btrfs_drop_snapshot(trans, reloc_root);
4262 if (ret != -EAGAIN)
4263 break;
4264 mutex_unlock(&root->fs_info->drop_mutex);
4265
4266 nr = trans->blocks_used;
4267 ret = btrfs_end_transaction(trans, root);
4268 BUG_ON(ret);
4269 btrfs_btree_balance_dirty(root, nr);
4270 }
4271
4272 free_extent_buffer(reloc_root->node);
4273
4274 ret = btrfs_del_root(trans, root->fs_info->tree_root,
4275 &reloc_root->root_key);
4276 BUG_ON(ret);
4277 mutex_unlock(&root->fs_info->drop_mutex);
4278
4279 nr = trans->blocks_used;
4280 ret = btrfs_end_transaction(trans, root);
4281 BUG_ON(ret);
4282 btrfs_btree_balance_dirty(root, nr);
4283
4284 kfree(prev_root);
4285 prev_root = reloc_root;
4286 }
4287 if (prev_root) {
4288 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
4289 kfree(prev_root);
4290 }
4291 return 0;
4292}
4293
4294int btrfs_add_dead_reloc_root(struct btrfs_root *root)
4295{
4296 list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
4297 return 0;
4298}
4299
4300int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
4301{
4302 struct btrfs_root *reloc_root;
4303 struct btrfs_trans_handle *trans;
4304 struct btrfs_key location;
4305 int found;
4306 int ret;
4307
4308 mutex_lock(&root->fs_info->tree_reloc_mutex);
4309 ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
4310 BUG_ON(ret);
4311 found = !list_empty(&root->fs_info->dead_reloc_roots);
4312 mutex_unlock(&root->fs_info->tree_reloc_mutex);
4313
4314 if (found) {
4315 trans = btrfs_start_transaction(root, 1);
4316 BUG_ON(!trans);
4317 ret = btrfs_commit_transaction(trans, root);
4318 BUG_ON(ret);
4319 }
4320
4321 location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
4322 location.offset = (u64)-1;
4323 location.type = BTRFS_ROOT_ITEM_KEY;
4324
4325 reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
4326 BUG_ON(!reloc_root);
4327 btrfs_orphan_cleanup(reloc_root);
4328 return 0;
4329}
4330
4331static int noinline init_reloc_tree(struct btrfs_trans_handle *trans,
4332 struct btrfs_root *root)
4333{
4334 struct btrfs_root *reloc_root;
4335 struct extent_buffer *eb;
4336 struct btrfs_root_item *root_item;
4337 struct btrfs_key root_key;
4338 int ret;
4339
4340 BUG_ON(!root->ref_cows);
4341 if (root->reloc_root)
4342 return 0;
4343
4344 root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
4345 BUG_ON(!root_item);
4346
4347 ret = btrfs_copy_root(trans, root, root->commit_root,
4348 &eb, BTRFS_TREE_RELOC_OBJECTID);
4349 BUG_ON(ret);
4350
4351 root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
4352 root_key.offset = root->root_key.objectid;
4353 root_key.type = BTRFS_ROOT_ITEM_KEY;
4354
4355 memcpy(root_item, &root->root_item, sizeof(root_item));
4356 btrfs_set_root_refs(root_item, 0);
4357 btrfs_set_root_bytenr(root_item, eb->start);
4358 btrfs_set_root_level(root_item, btrfs_header_level(eb));
4359 memset(&root_item->drop_progress, 0, sizeof(root_item->drop_progress));
4360 root_item->drop_level = 0;
4361
4362 btrfs_tree_unlock(eb);
4363 free_extent_buffer(eb);
4364
4365 ret = btrfs_insert_root(trans, root->fs_info->tree_root,
4366 &root_key, root_item);
4367 BUG_ON(ret);
4368 kfree(root_item);
4369
4370 reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
4371 &root_key);
4372 BUG_ON(!reloc_root);
4373 reloc_root->last_trans = trans->transid;
4374 reloc_root->commit_root = NULL;
4375 reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
4376
4377 root->reloc_root = reloc_root;
4378 return 0;
4379}
4380
4381/*
4382 * Core function of space balance.
4383 *
4384 * The idea is using reloc trees to relocate tree blocks in reference
4385 * counted roots. There is one reloc tree for each subvol, all reloc
4386 * trees share same key objectid. Reloc trees are snapshots of the
4387 * latest committed roots (subvol root->commit_root). To relocate a tree
4388 * block referenced by a subvol, the code COW the block through the reloc
4389 * tree, then update pointer in the subvol to point to the new block.
4390 * Since all reloc trees share same key objectid, we can easily do special
4391 * handing to share tree blocks between reloc trees. Once a tree block has
4392 * been COWed in one reloc tree, we can use the result when the same block
4393 * is COWed again through other reloc trees.
4394 */
4395static int noinline relocate_one_path(struct btrfs_trans_handle *trans,
4396 struct btrfs_root *root,
4397 struct btrfs_path *path,
4398 struct btrfs_key *first_key,
4399 struct btrfs_ref_path *ref_path,
4400 struct btrfs_block_group_cache *group,
4401 struct inode *reloc_inode)
4402{
4403 struct btrfs_root *reloc_root;
4404 struct extent_buffer *eb = NULL;
4405 struct btrfs_key *keys;
4406 u64 *nodes;
4407 int level;
4408 int lowest_merge;
4409 int lowest_level = 0;
4410 int update_refs;
4411 int ret;
4412
4413 if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
4414 lowest_level = ref_path->owner_objectid;
4415
4416 if (is_cowonly_root(ref_path->root_objectid)) {
4417 path->lowest_level = lowest_level;
4418 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
4419 BUG_ON(ret < 0);
4420 path->lowest_level = 0;
4421 btrfs_release_path(root, path);
4422 return 0;
4423 }
4424
4425 keys = kzalloc(sizeof(*keys) * BTRFS_MAX_LEVEL, GFP_NOFS);
4426 BUG_ON(!keys);
4427 nodes = kzalloc(sizeof(*nodes) * BTRFS_MAX_LEVEL, GFP_NOFS);
4428 BUG_ON(!nodes);
4429
4430 mutex_lock(&root->fs_info->tree_reloc_mutex);
4431 ret = init_reloc_tree(trans, root);
4432 BUG_ON(ret);
4433 reloc_root = root->reloc_root;
4434
4435 path->lowest_level = lowest_level;
4436 ret = btrfs_search_slot(trans, reloc_root, first_key, path, 0, 0);
4437 BUG_ON(ret);
4438 /*
4439 * get relocation mapping for tree blocks in the path
4440 */
4441 lowest_merge = BTRFS_MAX_LEVEL;
4442 for (level = BTRFS_MAX_LEVEL - 1; level >= lowest_level; level--) {
4443 u64 new_bytenr;
4444 eb = path->nodes[level];
4445 if (!eb || eb == reloc_root->node)
4446 continue;
4447 ret = btrfs_get_reloc_mapping(reloc_root, eb->start, eb->len,
4448 &new_bytenr);
4449 if (ret)
4450 continue;
4451 if (level == 0)
4452 btrfs_item_key_to_cpu(eb, &keys[level], 0);
4453 else
4454 btrfs_node_key_to_cpu(eb, &keys[level], 0);
4455 nodes[level] = new_bytenr;
4456 lowest_merge = level;
4457 }
4458
4459 update_refs = 0;
4460 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
4461 eb = path->nodes[0];
4462 if (btrfs_header_generation(eb) < trans->transid)
4463 update_refs = 1;
4464 }
4465
4466 btrfs_release_path(reloc_root, path);
4467 /*
4468 * merge tree blocks that already relocated in other reloc trees
4469 */
4470 if (lowest_merge != BTRFS_MAX_LEVEL) {
4471 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
4472 lowest_merge);
4473 BUG_ON(ret < 0);
4474 }
4475 /*
4476 * cow any tree blocks that still haven't been relocated
4477 */
4478 ret = btrfs_search_slot(trans, reloc_root, first_key, path, 0, 1);
4479 BUG_ON(ret);
4480 /*
4481 * if we are relocating data block group, update extent pointers
4482 * in the newly created tree leaf.
4483 */
4484 eb = path->nodes[0];
4485 if (update_refs && nodes[0] != eb->start) {
4486 ret = replace_extents_in_leaf(trans, reloc_root, eb, group,
4487 reloc_inode);
4488 BUG_ON(ret);
4489 }
4490
4491 memset(keys, 0, sizeof(*keys) * BTRFS_MAX_LEVEL);
4492 memset(nodes, 0, sizeof(*nodes) * BTRFS_MAX_LEVEL);
4493 for (level = BTRFS_MAX_LEVEL - 1; level >= lowest_level; level--) {
4494 eb = path->nodes[level];
4495 if (!eb || eb == reloc_root->node)
4496 continue;
4497 BUG_ON(btrfs_header_owner(eb) != BTRFS_TREE_RELOC_OBJECTID);
4498 nodes[level] = eb->start;
4499 if (level == 0)
4500 btrfs_item_key_to_cpu(eb, &keys[level], 0);
4501 else
4502 btrfs_node_key_to_cpu(eb, &keys[level], 0);
4503 }
4504
4505 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
4506 eb = path->nodes[0];
4507 extent_buffer_get(eb);
4508 }
4509 btrfs_release_path(reloc_root, path);
4510 /*
4511 * replace tree blocks in the fs tree with tree blocks in
4512 * the reloc tree.
4513 */
4514 ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
4515 BUG_ON(ret < 0);
4516
4517 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
4518 ret = invalidate_extent_cache(reloc_root, eb, group, root);
4519 BUG_ON(ret);
4520 free_extent_buffer(eb);
4521 }
4522 mutex_unlock(&root->fs_info->tree_reloc_mutex);
4523
4524 path->lowest_level = 0;
4525 kfree(nodes);
4526 kfree(keys);
4527 return 0;
4528}
4529
4530static int noinline relocate_tree_block(struct btrfs_trans_handle *trans,
4531 struct btrfs_root *root,
4532 struct btrfs_path *path,
4533 struct btrfs_key *first_key,
4534 struct btrfs_ref_path *ref_path)
4535{
4536 int ret;
4537 int needs_lock = 0;
4538
4539 if (root == root->fs_info->extent_root ||
4540 root == root->fs_info->chunk_root ||
4541 root == root->fs_info->dev_root) {
4542 needs_lock = 1;
4543 mutex_lock(&root->fs_info->alloc_mutex);
4544 }
4545
4546 ret = relocate_one_path(trans, root, path, first_key,
4547 ref_path, NULL, NULL);
4548 BUG_ON(ret);
4549
4550 if (root == root->fs_info->extent_root)
4551 btrfs_extent_post_op(trans, root);
4552 if (needs_lock)
4553 mutex_unlock(&root->fs_info->alloc_mutex);
4554
4555 return 0;
4556}
4557
4558static int noinline del_extent_zero(struct btrfs_trans_handle *trans,
4559 struct btrfs_root *extent_root,
4560 struct btrfs_path *path,
4561 struct btrfs_key *extent_key)
4562{
4563 int ret;
4564
4565 mutex_lock(&extent_root->fs_info->alloc_mutex);
4566 ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
4567 if (ret)
4568 goto out;
4569 ret = btrfs_del_item(trans, extent_root, path);
4570out:
4571 btrfs_release_path(extent_root, path);
4572 mutex_unlock(&extent_root->fs_info->alloc_mutex);
4573 return ret;
4574}
4575
4576static struct btrfs_root noinline *read_ref_root(struct btrfs_fs_info *fs_info,
4577 struct btrfs_ref_path *ref_path)
4578{
4579 struct btrfs_key root_key;
4580
4581 root_key.objectid = ref_path->root_objectid;
4582 root_key.type = BTRFS_ROOT_ITEM_KEY;
4583 if (is_cowonly_root(ref_path->root_objectid))
4584 root_key.offset = 0;
4585 else
4586 root_key.offset = (u64)-1;
4587
4588 return btrfs_read_fs_root_no_name(fs_info, &root_key);
4589}
4590
4591static int noinline relocate_one_extent(struct btrfs_root *extent_root,
4592 struct btrfs_path *path,
4593 struct btrfs_key *extent_key,
4594 struct btrfs_block_group_cache *group,
4595 struct inode *reloc_inode, int pass)
4596{
4597 struct btrfs_trans_handle *trans;
4598 struct btrfs_root *found_root;
4599 struct btrfs_ref_path *ref_path = NULL;
4600 struct disk_extent *new_extents = NULL;
4601 int nr_extents = 0;
4602 int loops;
4603 int ret;
4604 int level;
4605 struct btrfs_key first_key;
4606 u64 prev_block = 0;
4607
4608 mutex_unlock(&extent_root->fs_info->alloc_mutex);
4609
4610 trans = btrfs_start_transaction(extent_root, 1);
4611 BUG_ON(!trans);
4612
4613 if (extent_key->objectid == 0) {
4614 ret = del_extent_zero(trans, extent_root, path, extent_key);
4615 goto out;
4616 }
4617
4618 ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
4619 if (!ref_path) {
4620 ret = -ENOMEM;
4621 goto out;
4622 }
4623
4624 for (loops = 0; ; loops++) {
4625 if (loops == 0) {
4626 ret = btrfs_first_ref_path(trans, extent_root, ref_path,
4627 extent_key->objectid);
4628 } else {
4629 ret = btrfs_next_ref_path(trans, extent_root, ref_path);
4630 }
4631 if (ret < 0)
4632 goto out;
4633 if (ret > 0)
4634 break;
4635
4636 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
4637 ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
4638 continue;
4639
4640 found_root = read_ref_root(extent_root->fs_info, ref_path);
4641 BUG_ON(!found_root);
4642 /*
4643 * for reference counted tree, only process reference paths
4644 * rooted at the latest committed root.
4645 */
4646 if (found_root->ref_cows &&
4647 ref_path->root_generation != found_root->root_key.offset)
4648 continue;
4649
4650 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
4651 if (pass == 0) {
4652 /*
4653 * copy data extents to new locations
4654 */
4655 u64 group_start = group->key.objectid;
4656 ret = relocate_data_extent(reloc_inode,
4657 extent_key,
4658 group_start);
4659 if (ret < 0)
4660 goto out;
4661 break;
4662 }
4663 level = 0;
4664 } else {
4665 level = ref_path->owner_objectid;
4666 }
4667
4668 if (prev_block != ref_path->nodes[level]) {
4669 struct extent_buffer *eb;
4670 u64 block_start = ref_path->nodes[level];
4671 u64 block_size = btrfs_level_size(found_root, level);
4672
4673 eb = read_tree_block(found_root, block_start,
4674 block_size, 0);
4675 btrfs_tree_lock(eb);
4676 BUG_ON(level != btrfs_header_level(eb));
4677
4678 if (level == 0)
4679 btrfs_item_key_to_cpu(eb, &first_key, 0);
4680 else
4681 btrfs_node_key_to_cpu(eb, &first_key, 0);
4682
4683 btrfs_tree_unlock(eb);
4684 free_extent_buffer(eb);
4685 prev_block = block_start;
4686 }
4687
4688 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID &&
4689 pass >= 2) {
4690 /*
4691 * use fallback method to process the remaining
4692 * references.
4693 */
4694 if (!new_extents) {
4695 u64 group_start = group->key.objectid;
4696 ret = get_new_locations(reloc_inode,
4697 extent_key,
4698 group_start, 0,
4699 &new_extents,
4700 &nr_extents);
4701 if (ret < 0)
4702 goto out;
4703 }
4704 btrfs_record_root_in_trans(found_root);
4705 ret = replace_one_extent(trans, found_root,
4706 path, extent_key,
4707 &first_key, ref_path,
4708 new_extents, nr_extents);
4709 if (ret < 0)
4710 goto out;
4711 continue;
4712 }
4713
4714 btrfs_record_root_in_trans(found_root);
4715 if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4716 ret = relocate_tree_block(trans, found_root, path,
4717 &first_key, ref_path);
4718 } else {
4719 /*
4720 * try to update data extent references while
4721 * keeping metadata shared between snapshots.
4722 */
4723 ret = relocate_one_path(trans, found_root, path,
4724 &first_key, ref_path,
4725 group, reloc_inode);
4726 }
4727 if (ret < 0)
4728 goto out;
4729 }
4730 ret = 0;
4731out:
4732 btrfs_end_transaction(trans, extent_root);
4733 kfree(new_extents);
4734 kfree(ref_path);
4735 mutex_lock(&extent_root->fs_info->alloc_mutex);
4736 return ret;
4737}
4738
ec44a35c
CM		 4739static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
4740{
4741 u64 num_devices;
4742 u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
4743 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
4744
a061fc8d 4745 num_devices = root->fs_info->fs_devices->num_devices;
ec44a35c
CM		 4746 if (num_devices == 1) {
4747 stripped |= BTRFS_BLOCK_GROUP_DUP;
4748 stripped = flags & ~stripped;
4749
4750 /* turn raid0 into single device chunks */
4751 if (flags & BTRFS_BLOCK_GROUP_RAID0)
4752 return stripped;
4753
4754 /* turn mirroring into duplication */
4755 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
4756 BTRFS_BLOCK_GROUP_RAID10))
4757 return stripped | BTRFS_BLOCK_GROUP_DUP;
4758 return flags;
4759 } else {
4760 /* they already had raid on here, just return */
ec44a35c
CM		 4761 if (flags & stripped)
4762 return flags;
4763
4764 stripped |= BTRFS_BLOCK_GROUP_DUP;
4765 stripped = flags & ~stripped;
4766
4767 /* switch duplicated blocks with raid1 */
4768 if (flags & BTRFS_BLOCK_GROUP_DUP)
4769 return stripped | BTRFS_BLOCK_GROUP_RAID1;
4770
4771 /* turn single device chunks into raid0 */
4772 return stripped | BTRFS_BLOCK_GROUP_RAID0;
4773 }
4774 return flags;
4775}
4776
0ef3e66b
CM		 4777int __alloc_chunk_for_shrink(struct btrfs_root *root,
4778 struct btrfs_block_group_cache *shrink_block_group,
4779 int force)
4780{
4781 struct btrfs_trans_handle *trans;
4782 u64 new_alloc_flags;
4783 u64 calc;
4784
c286ac48 4785 spin_lock(&shrink_block_group->lock);
0ef3e66b 4786 if (btrfs_block_group_used(&shrink_block_group->item) > 0) {
c286ac48 4787 spin_unlock(&shrink_block_group->lock);
7d9eb12c 4788 mutex_unlock(&root->fs_info->alloc_mutex);
c286ac48 4789
0ef3e66b 4790 trans = btrfs_start_transaction(root, 1);
7d9eb12c 4791 mutex_lock(&root->fs_info->alloc_mutex);
c286ac48 4792 spin_lock(&shrink_block_group->lock);
7d9eb12c 4793
0ef3e66b
CM		 4794 new_alloc_flags = update_block_group_flags(root,
4795 shrink_block_group->flags);
4796 if (new_alloc_flags != shrink_block_group->flags) {
4797 calc =
4798 btrfs_block_group_used(&shrink_block_group->item);
4799 } else {
4800 calc = shrink_block_group->key.offset;
4801 }
c286ac48
CM		 4802 spin_unlock(&shrink_block_group->lock);
4803
0ef3e66b
CM		 4804 do_chunk_alloc(trans, root->fs_info->extent_root,
4805 calc + 2 * 1024 * 1024, new_alloc_flags, force);
7d9eb12c
CM		 4806
4807 mutex_unlock(&root->fs_info->alloc_mutex);
0ef3e66b 4808 btrfs_end_transaction(trans, root);
7d9eb12c 4809 mutex_lock(&root->fs_info->alloc_mutex);
c286ac48
CM		 4810 } else
4811 spin_unlock(&shrink_block_group->lock);
0ef3e66b
CM		 4812 return 0;
4813}
4814
1a40e23b
ZY		 4815static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
4816 struct btrfs_root *root,
4817 u64 objectid, u64 size)
4818{
4819 struct btrfs_path *path;
4820 struct btrfs_inode_item *item;
4821 struct extent_buffer *leaf;
4822 int ret;
4823
4824 path = btrfs_alloc_path();
4825 if (!path)
4826 return -ENOMEM;
4827
4828 ret = btrfs_insert_empty_inode(trans, root, path, objectid);
4829 if (ret)
4830 goto out;
4831
4832 leaf = path->nodes[0];
4833 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
4834 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
4835 btrfs_set_inode_generation(leaf, item, 1);
4836 btrfs_set_inode_size(leaf, item, size);
4837 btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
4838 btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NODATASUM);
4839 btrfs_mark_buffer_dirty(leaf);
4840 btrfs_release_path(root, path);
4841out:
4842 btrfs_free_path(path);
4843 return ret;
4844}
4845
4846static struct inode noinline *create_reloc_inode(struct btrfs_fs_info *fs_info,
4847 struct btrfs_block_group_cache *group)
4848{
4849 struct inode *inode = NULL;
4850 struct btrfs_trans_handle *trans;
4851 struct btrfs_root *root;
4852 struct btrfs_key root_key;
4853 u64 objectid = BTRFS_FIRST_FREE_OBJECTID;
4854 int err = 0;
4855
4856 root_key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
4857 root_key.type = BTRFS_ROOT_ITEM_KEY;
4858 root_key.offset = (u64)-1;
4859 root = btrfs_read_fs_root_no_name(fs_info, &root_key);
4860 if (IS_ERR(root))
4861 return ERR_CAST(root);
4862
4863 trans = btrfs_start_transaction(root, 1);
4864 BUG_ON(!trans);
4865
4866 err = btrfs_find_free_objectid(trans, root, objectid, &objectid);
4867 if (err)
4868 goto out;
4869
4870 err = __insert_orphan_inode(trans, root, objectid, group->key.offset);
4871 BUG_ON(err);
4872
4873 err = btrfs_insert_file_extent(trans, root, objectid, 0, 0, 0,
c8b97818
CM		 4874 group->key.offset, 0, group->key.offset,
4875 0, 0, 0);
1a40e23b
ZY		 4876 BUG_ON(err);
4877
4878 inode = btrfs_iget_locked(root->fs_info->sb, objectid, root);
4879 if (inode->i_state & I_NEW) {
4880 BTRFS_I(inode)->root = root;
4881 BTRFS_I(inode)->location.objectid = objectid;
4882 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
4883 BTRFS_I(inode)->location.offset = 0;
4884 btrfs_read_locked_inode(inode);
4885 unlock_new_inode(inode);
4886 BUG_ON(is_bad_inode(inode));
4887 } else {
4888 BUG_ON(1);
4889 }
4890
4891 err = btrfs_orphan_add(trans, inode);
4892out:
4893 btrfs_end_transaction(trans, root);
4894 if (err) {
4895 if (inode)
4896 iput(inode);
4897 inode = ERR_PTR(err);
4898 }
4899 return inode;
4900}
4901
4902int btrfs_relocate_block_group(struct btrfs_root *root, u64 group_start)
edbd8d4e
CM		 4903{
4904 struct btrfs_trans_handle *trans;
edbd8d4e 4905 struct btrfs_path *path;
1a40e23b
ZY		 4906 struct btrfs_fs_info *info = root->fs_info;
4907 struct extent_buffer *leaf;
4908 struct inode *reloc_inode;
4909 struct btrfs_block_group_cache *block_group;
4910 struct btrfs_key key;
edbd8d4e
CM		 4911 u64 cur_byte;
4912 u64 total_found;
edbd8d4e
CM		 4913 u32 nritems;
4914 int ret;
a061fc8d 4915 int progress;
1a40e23b 4916 int pass = 0;
edbd8d4e 4917
1a40e23b
ZY		 4918 root = root->fs_info->extent_root;
4919
4920 block_group = btrfs_lookup_block_group(info, group_start);
4921 BUG_ON(!block_group);
8f18cf13 4922
1a40e23b
ZY		 4923 printk("btrfs relocating block group %llu flags %llu\n",
4924 (unsigned long long)block_group->key.objectid,
4925 (unsigned long long)block_group->flags);
8f18cf13 4926
edbd8d4e 4927 path = btrfs_alloc_path();
1a40e23b 4928 BUG_ON(!path);
edbd8d4e 4929
1a40e23b
ZY		 4930 reloc_inode = create_reloc_inode(info, block_group);
4931 BUG_ON(IS_ERR(reloc_inode));
323da79c 4932
1a40e23b 4933 mutex_lock(&root->fs_info->alloc_mutex);
0ef3e66b 4934
1a40e23b
ZY		 4935 __alloc_chunk_for_shrink(root, block_group, 1);
4936 block_group->ro = 1;
4937 block_group->space_info->total_bytes -= block_group->key.offset;
323da79c 4938
1a40e23b 4939 mutex_unlock(&root->fs_info->alloc_mutex);
8f18cf13 4940
1a40e23b
ZY		 4941 btrfs_start_delalloc_inodes(info->tree_root);
4942 btrfs_wait_ordered_extents(info->tree_root, 0);
4943again:
edbd8d4e 4944 total_found = 0;
a061fc8d 4945 progress = 0;
1a40e23b 4946 key.objectid = block_group->key.objectid;
edbd8d4e
CM		 4947 key.offset = 0;
4948 key.type = 0;
73e48b27 4949 cur_byte = key.objectid;
4313b399 4950
1a40e23b
ZY		 4951 trans = btrfs_start_transaction(info->tree_root, 1);
4952 btrfs_commit_transaction(trans, info->tree_root);
ea8c2819 4953
1a40e23b
ZY		 4954 mutex_lock(&root->fs_info->cleaner_mutex);
4955 btrfs_clean_old_snapshots(info->tree_root);
4956 btrfs_remove_leaf_refs(info->tree_root, (u64)-1, 1);
4957 mutex_unlock(&root->fs_info->cleaner_mutex);
ea8c2819
CM		 4958
4959 mutex_lock(&root->fs_info->alloc_mutex);
4960
73e48b27 4961 while(1) {
edbd8d4e
CM		 4962 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4963 if (ret < 0)
4964 goto out;
7d9eb12c 4965next:
edbd8d4e 4966 leaf = path->nodes[0];
73e48b27 4967 nritems = btrfs_header_nritems(leaf);
73e48b27
Y		 4968 if (path->slots[0] >= nritems) {
4969 ret = btrfs_next_leaf(root, path);
4970 if (ret < 0)
4971 goto out;
4972 if (ret == 1) {
4973 ret = 0;
4974 break;
edbd8d4e 4975 }
73e48b27
Y		 4976 leaf = path->nodes[0];
4977 nritems = btrfs_header_nritems(leaf);
edbd8d4e 4978 }
73e48b27 4979
1a40e23b 4980 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
725c8463 4981
1a40e23b
ZY		 4982 if (key.objectid >= block_group->key.objectid +
4983 block_group->key.offset)
8f18cf13
CM		 4984 break;
4985
725c8463 4986 if (progress && need_resched()) {
725c8463 4987 btrfs_release_path(root, path);
1a40e23b
ZY		 4988 mutex_unlock(&root->fs_info->alloc_mutex);
4989 cond_resched();
4990 mutex_lock(&root->fs_info->alloc_mutex);
725c8463 4991 progress = 0;
1a40e23b 4992 continue;
725c8463
CM		 4993 }
4994 progress = 1;
4995
1a40e23b
ZY		 4996 if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY ||
4997 key.objectid + key.offset <= cur_byte) {
edbd8d4e 4998 path->slots[0]++;
edbd8d4e
CM		 4999 goto next;
5000 }
73e48b27 5001
edbd8d4e 5002 total_found++;
1a40e23b 5003 cur_byte = key.objectid + key.offset;
edbd8d4e 5004 btrfs_release_path(root, path);
edbd8d4e 5005
1a40e23b
ZY		 5006 __alloc_chunk_for_shrink(root, block_group, 0);
5007 ret = relocate_one_extent(root, path, &key, block_group,
5008 reloc_inode, pass);
5009 BUG_ON(ret < 0);
edbd8d4e 5010
1a40e23b
ZY		 5011 key.objectid = cur_byte;
5012 key.type = 0;
5013 key.offset = 0;
edbd8d4e
CM		 5014 }
5015
1a40e23b 5016 btrfs_release_path(root, path);
7d9eb12c
CM		 5017 mutex_unlock(&root->fs_info->alloc_mutex);
5018
1a40e23b
ZY		 5019 if (pass == 0) {
5020 btrfs_wait_ordered_range(reloc_inode, 0, (u64)-1);
5021 invalidate_mapping_pages(reloc_inode->i_mapping, 0, -1);
5022 WARN_ON(reloc_inode->i_mapping->nrpages);
8e8a1e31 5023 }
73e48b27 5024
1a40e23b
ZY		 5025 if (total_found > 0) {
5026 printk("btrfs found %llu extents in pass %d\n",
5027 (unsigned long long)total_found, pass);
5028 pass++;
5029 goto again;
0f9dd46c 5030 }
0ef3e66b 5031
1a40e23b
ZY		 5032 /* delete reloc_inode */
5033 iput(reloc_inode);
5034
5035 /* unpin extents in this range */
5036 trans = btrfs_start_transaction(info->tree_root, 1);
5037 btrfs_commit_transaction(trans, info->tree_root);
0ef3e66b 5038
7d9eb12c
CM		 5039 mutex_lock(&root->fs_info->alloc_mutex);
5040
1a40e23b
ZY		 5041 spin_lock(&block_group->lock);
5042 WARN_ON(block_group->pinned > 0);
5043 WARN_ON(block_group->reserved > 0);
5044 WARN_ON(btrfs_block_group_used(&block_group->item) > 0);
5045 spin_unlock(&block_group->lock);
5046 ret = 0;
edbd8d4e 5047out:
925baedd 5048 mutex_unlock(&root->fs_info->alloc_mutex);
1a40e23b 5049 btrfs_free_path(path);
edbd8d4e
CM		 5050 return ret;
5051}
5052
0b86a832
CM		 5053int find_first_block_group(struct btrfs_root *root, struct btrfs_path *path,
5054 struct btrfs_key *key)
5055{
925baedd 5056 int ret = 0;
0b86a832
CM		 5057 struct btrfs_key found_key;
5058 struct extent_buffer *leaf;
5059 int slot;
edbd8d4e 5060
0b86a832
CM		 5061 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
5062 if (ret < 0)
925baedd
CM		 5063 goto out;
5064
0b86a832
CM		 5065 while(1) {
5066 slot = path->slots[0];
edbd8d4e 5067 leaf = path->nodes[0];
0b86a832
CM		 5068 if (slot >= btrfs_header_nritems(leaf)) {
5069 ret = btrfs_next_leaf(root, path);
5070 if (ret == 0)
5071 continue;
5072 if (ret < 0)
925baedd 5073 goto out;
0b86a832 5074 break;
edbd8d4e 5075 }
0b86a832 5076 btrfs_item_key_to_cpu(leaf, &found_key, slot);
edbd8d4e 5077
0b86a832 5078 if (found_key.objectid >= key->objectid &&
925baedd
CM		 5079 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
5080 ret = 0;
5081 goto out;
5082 }
0b86a832 5083 path->slots[0]++;
edbd8d4e 5084 }
0b86a832 5085 ret = -ENOENT;
925baedd 5086out:
0b86a832 5087 return ret;
edbd8d4e
CM		 5088}
5089
1a40e23b
ZY		 5090int btrfs_free_block_groups(struct btrfs_fs_info *info)
5091{
5092 struct btrfs_block_group_cache *block_group;
5093 struct rb_node *n;
5094
5095 mutex_lock(&info->alloc_mutex);
5096 spin_lock(&info->block_group_cache_lock);
5097 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
5098 block_group = rb_entry(n, struct btrfs_block_group_cache,
5099 cache_node);
5100
5101 spin_unlock(&info->block_group_cache_lock);
5102 btrfs_remove_free_space_cache(block_group);
5103 spin_lock(&info->block_group_cache_lock);
5104
5105 rb_erase(&block_group->cache_node,
5106 &info->block_group_cache_tree);
5107 spin_lock(&block_group->space_info->lock);
5108 list_del(&block_group->list);
5109 spin_unlock(&block_group->space_info->lock);
5110 kfree(block_group);
5111 }
5112 spin_unlock(&info->block_group_cache_lock);
5113 mutex_unlock(&info->alloc_mutex);
5114 return 0;
5115}
5116
9078a3e1
CM		 5117int btrfs_read_block_groups(struct btrfs_root *root)
5118{
5119 struct btrfs_path *path;
5120 int ret;
9078a3e1 5121 struct btrfs_block_group_cache *cache;
be744175 5122 struct btrfs_fs_info *info = root->fs_info;
6324fbf3 5123 struct btrfs_space_info *space_info;
9078a3e1
CM		 5124 struct btrfs_key key;
5125 struct btrfs_key found_key;
5f39d397 5126 struct extent_buffer *leaf;
96b5179d 5127
be744175 5128 root = info->extent_root;
9078a3e1 5129 key.objectid = 0;
0b86a832 5130 key.offset = 0;
9078a3e1 5131 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
9078a3e1
CM		 5132 path = btrfs_alloc_path();
5133 if (!path)
5134 return -ENOMEM;
5135
925baedd 5136 mutex_lock(&root->fs_info->alloc_mutex);
9078a3e1 5137 while(1) {
0b86a832
CM		 5138 ret = find_first_block_group(root, path, &key);
5139 if (ret > 0) {
5140 ret = 0;
5141 goto error;
9078a3e1 5142 }
0b86a832
CM		 5143 if (ret != 0)
5144 goto error;
5145
5f39d397
CM		 5146 leaf = path->nodes[0];
5147 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
8f18cf13 5148 cache = kzalloc(sizeof(*cache), GFP_NOFS);
9078a3e1 5149 if (!cache) {
0b86a832 5150 ret = -ENOMEM;
9078a3e1
CM		 5151 break;
5152 }
3e1ad54f 5153
c286ac48 5154 spin_lock_init(&cache->lock);
0f9dd46c 5155 INIT_LIST_HEAD(&cache->list);
5f39d397
CM		 5156 read_extent_buffer(leaf, &cache->item,
5157 btrfs_item_ptr_offset(leaf, path->slots[0]),
5158 sizeof(cache->item));
9078a3e1 5159 memcpy(&cache->key, &found_key, sizeof(found_key));
0b86a832 5160
9078a3e1
CM		 5161 key.objectid = found_key.objectid + found_key.offset;
5162 btrfs_release_path(root, path);
0b86a832 5163 cache->flags = btrfs_block_group_flags(&cache->item);
96b5179d 5164
6324fbf3
CM		 5165 ret = update_space_info(info, cache->flags, found_key.offset,
5166 btrfs_block_group_used(&cache->item),
5167 &space_info);
5168 BUG_ON(ret);
5169 cache->space_info = space_info;
0f9dd46c
JB		 5170 spin_lock(&space_info->lock);
5171 list_add(&cache->list, &space_info->block_groups);
5172 spin_unlock(&space_info->lock);
5173
5174 ret = btrfs_add_block_group_cache(root->fs_info, cache);
5175 BUG_ON(ret);
75ccf47d
CM		 5176
5177 set_avail_alloc_bits(root->fs_info, cache->flags);
9078a3e1 5178 }
0b86a832
CM		 5179 ret = 0;
5180error:
9078a3e1 5181 btrfs_free_path(path);
925baedd 5182 mutex_unlock(&root->fs_info->alloc_mutex);
0b86a832 5183 return ret;
9078a3e1 5184}
6324fbf3
CM		 5185
5186int btrfs_make_block_group(struct btrfs_trans_handle *trans,
5187 struct btrfs_root *root, u64 bytes_used,
e17cade2 5188 u64 type, u64 chunk_objectid, u64 chunk_offset,
6324fbf3
CM		 5189 u64 size)
5190{
5191 int ret;
6324fbf3
CM		 5192 struct btrfs_root *extent_root;
5193 struct btrfs_block_group_cache *cache;
6324fbf3 5194
7d9eb12c 5195 WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex));
6324fbf3 5196 extent_root = root->fs_info->extent_root;
6324fbf3 5197
e02119d5
CM		 5198 root->fs_info->last_trans_new_blockgroup = trans->transid;
5199
8f18cf13 5200 cache = kzalloc(sizeof(*cache), GFP_NOFS);
0f9dd46c
JB		 5201 if (!cache)
5202 return -ENOMEM;
5203
e17cade2 5204 cache->key.objectid = chunk_offset;
6324fbf3 5205 cache->key.offset = size;
c286ac48 5206 spin_lock_init(&cache->lock);
0f9dd46c 5207 INIT_LIST_HEAD(&cache->list);
6324fbf3 5208 btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY);
0ef3e66b 5209
6324fbf3 5210 btrfs_set_block_group_used(&cache->item, bytes_used);
6324fbf3
CM		 5211 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
5212 cache->flags = type;
5213 btrfs_set_block_group_flags(&cache->item, type);
5214
5215 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
5216 &cache->space_info);
5217 BUG_ON(ret);
0f9dd46c
JB		 5218 spin_lock(&cache->space_info->lock);
5219 list_add(&cache->list, &cache->space_info->block_groups);
5220 spin_unlock(&cache->space_info->lock);
6324fbf3 5221
0f9dd46c
JB		 5222 ret = btrfs_add_block_group_cache(root->fs_info, cache);
5223 BUG_ON(ret);
c286ac48 5224
6324fbf3
CM		 5225 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
5226 sizeof(cache->item));
5227 BUG_ON(ret);
5228
5229 finish_current_insert(trans, extent_root);
5230 ret = del_pending_extents(trans, extent_root);
5231 BUG_ON(ret);
d18a2c44 5232 set_avail_alloc_bits(extent_root->fs_info, type);
925baedd 5233
6324fbf3
CM		 5234 return 0;
5235}
1a40e23b
ZY		 5236
5237int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
5238 struct btrfs_root *root, u64 group_start)
5239{
5240 struct btrfs_path *path;
5241 struct btrfs_block_group_cache *block_group;
5242 struct btrfs_key key;
5243 int ret;
5244
5245 BUG_ON(!mutex_is_locked(&root->fs_info->alloc_mutex));
5246 root = root->fs_info->extent_root;
5247
5248 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
5249 BUG_ON(!block_group);
5250
5251 memcpy(&key, &block_group->key, sizeof(key));
5252
5253 path = btrfs_alloc_path();
5254 BUG_ON(!path);
5255
5256 btrfs_remove_free_space_cache(block_group);
5257 rb_erase(&block_group->cache_node,
5258 &root->fs_info->block_group_cache_tree);
5259 spin_lock(&block_group->space_info->lock);
5260 list_del(&block_group->list);
5261 spin_unlock(&block_group->space_info->lock);
5262
5263 /*
5264 memset(shrink_block_group, 0, sizeof(*shrink_block_group));
5265 kfree(shrink_block_group);
5266 */
5267
5268 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
5269 if (ret > 0)
5270 ret = -EIO;
5271 if (ret < 0)
5272 goto out;
5273
5274 ret = btrfs_del_item(trans, root, path);
5275out:
5276 btrfs_free_path(path);
5277 return ret;
5278}
Linux 6.x block layer and io_uring tree(s)