projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Btrfs: only flush down bios for writeback pages
[linux-2.6-block.git] / fs / btrfs / extent-tree.c
CommitLineData
6cbd5570
CM		 1/*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
ec6b910f 18#include <linux/sched.h>
edbd8d4e 19#include <linux/pagemap.h>
ec44a35c 20#include <linux/writeback.h>
21af804c 21#include <linux/blkdev.h>
4b4e25f2
CM		 22#include <linux/version.h>
23#include "compat.h"
74493f7a 24#include "hash.h"
a5eb62e3 25#include "crc32c.h"
fec577fb
CM		 26#include "ctree.h"
27#include "disk-io.h"
28#include "print-tree.h"
e089f05c 29#include "transaction.h"
0b86a832 30#include "volumes.h"
925baedd 31#include "locking.h"
31153d81 32#include "ref-cache.h"
15916de8 33#include "compat.h"
fec577fb 34
31840ae1
ZY		 35#define PENDING_EXTENT_INSERT 0
36#define PENDING_EXTENT_DELETE 1
37#define PENDING_BACKREF_UPDATE 2
38
39struct pending_extent_op {
40 int type;
41 u64 bytenr;
42 u64 num_bytes;
43 u64 parent;
44 u64 orig_parent;
45 u64 generation;
46 u64 orig_generation;
47 int level;
f3465ca4
JB		 48 struct list_head list;
49 int del;
31840ae1
ZY		 50};
51
e089f05c 52static int finish_current_insert(struct btrfs_trans_handle *trans, struct
87ef2bb4 53 btrfs_root *extent_root, int all);
e20d96d6 54static int del_pending_extents(struct btrfs_trans_handle *trans, struct
87ef2bb4 55 btrfs_root *extent_root, int all);
925baedd
CM		 56static struct btrfs_block_group_cache *
57__btrfs_find_block_group(struct btrfs_root *root,
58 struct btrfs_block_group_cache *hint,
59 u64 search_start, int data, int owner);
f3465ca4
JB		 60static int pin_down_bytes(struct btrfs_trans_handle *trans,
61 struct btrfs_root *root,
62 u64 bytenr, u64 num_bytes, int is_data);
63static int update_block_group(struct btrfs_trans_handle *trans,
64 struct btrfs_root *root,
65 u64 bytenr, u64 num_bytes, int alloc,
66 int mark_free);
d548ee51 67
0f9dd46c
JB		 68static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
69{
70 return (cache->flags & bits) == bits;
71}
72
73/*
74 * this adds the block group to the fs_info rb tree for the block group
75 * cache
76 */
77int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
78 struct btrfs_block_group_cache *block_group)
79{
80 struct rb_node **p;
81 struct rb_node *parent = NULL;
82 struct btrfs_block_group_cache *cache;
83
84 spin_lock(&info->block_group_cache_lock);
85 p = &info->block_group_cache_tree.rb_node;
86
87 while (*p) {
88 parent = *p;
89 cache = rb_entry(parent, struct btrfs_block_group_cache,
90 cache_node);
91 if (block_group->key.objectid < cache->key.objectid) {
92 p = &(*p)->rb_left;
93 } else if (block_group->key.objectid > cache->key.objectid) {
94 p = &(*p)->rb_right;
95 } else {
96 spin_unlock(&info->block_group_cache_lock);
97 return -EEXIST;
98 }
99 }
100
101 rb_link_node(&block_group->cache_node, parent, p);
102 rb_insert_color(&block_group->cache_node,
103 &info->block_group_cache_tree);
104 spin_unlock(&info->block_group_cache_lock);
105
106 return 0;
107}
108
109/*
110 * This will return the block group at or after bytenr if contains is 0, else
111 * it will return the block group that contains the bytenr
112 */
113static struct btrfs_block_group_cache *
114block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
115 int contains)
116{
117 struct btrfs_block_group_cache *cache, *ret = NULL;
118 struct rb_node *n;
119 u64 end, start;
120
121 spin_lock(&info->block_group_cache_lock);
122 n = info->block_group_cache_tree.rb_node;
123
124 while (n) {
125 cache = rb_entry(n, struct btrfs_block_group_cache,
126 cache_node);
127 end = cache->key.objectid + cache->key.offset - 1;
128 start = cache->key.objectid;
129
130 if (bytenr < start) {
131 if (!contains && (!ret || start < ret->key.objectid))
132 ret = cache;
133 n = n->rb_left;
134 } else if (bytenr > start) {
135 if (contains && bytenr <= end) {
136 ret = cache;
137 break;
138 }
139 n = n->rb_right;
140 } else {
141 ret = cache;
142 break;
143 }
144 }
145 spin_unlock(&info->block_group_cache_lock);
146
147 return ret;
148}
149
150/*
151 * this is only called by cache_block_group, since we could have freed extents
152 * we need to check the pinned_extents for any extents that can't be used yet
153 * since their free space will be released as soon as the transaction commits.
154 */
155static int add_new_free_space(struct btrfs_block_group_cache *block_group,
156 struct btrfs_fs_info *info, u64 start, u64 end)
157{
158 u64 extent_start, extent_end, size;
159 int ret;
160
25179201 161 mutex_lock(&info->pinned_mutex);
0f9dd46c
JB		 162 while (start < end) {
163 ret = find_first_extent_bit(&info->pinned_extents, start,
164 &extent_start, &extent_end,
165 EXTENT_DIRTY);
166 if (ret)
167 break;
168
169 if (extent_start == start) {
170 start = extent_end + 1;
171 } else if (extent_start > start && extent_start < end) {
172 size = extent_start - start;
25179201
JB		 173 ret = btrfs_add_free_space_lock(block_group, start,
174 size);
0f9dd46c
JB		 175 BUG_ON(ret);
176 start = extent_end + 1;
177 } else {
178 break;
179 }
180 }
181
182 if (start < end) {
183 size = end - start;
25179201 184 ret = btrfs_add_free_space_lock(block_group, start, size);
0f9dd46c
JB		 185 BUG_ON(ret);
186 }
25179201 187 mutex_unlock(&info->pinned_mutex);
0f9dd46c
JB		 188
189 return 0;
190}
191
e37c9e69
CM		 192static int cache_block_group(struct btrfs_root *root,
193 struct btrfs_block_group_cache *block_group)
194{
195 struct btrfs_path *path;
ef8bbdfe 196 int ret = 0;
e37c9e69 197 struct btrfs_key key;
5f39d397 198 struct extent_buffer *leaf;
e37c9e69 199 int slot;
e37c9e69 200 u64 last = 0;
7d7d6068 201 u64 first_free;
e37c9e69
CM		 202 int found = 0;
203
00f5c795
CM		 204 if (!block_group)
205 return 0;
206
e37c9e69 207 root = root->fs_info->extent_root;
e37c9e69
CM		 208
209 if (block_group->cached)
210 return 0;
f510cfec 211
e37c9e69
CM		 212 path = btrfs_alloc_path();
213 if (!path)
214 return -ENOMEM;
7d7d6068 215
2cc58cf2 216 path->reada = 2;
5cd57b2c
CM		 217 /*
218 * we get into deadlocks with paths held by callers of this function.
219 * since the alloc_mutex is protecting things right now, just
220 * skip the locking here
221 */
222 path->skip_locking = 1;
0f9dd46c
JB		 223 first_free = max_t(u64, block_group->key.objectid,
224 BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE);
e37c9e69 225 key.objectid = block_group->key.objectid;
e37c9e69
CM		 226 key.offset = 0;
227 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
228 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
229 if (ret < 0)
ef8bbdfe 230 goto err;
0b86a832 231 ret = btrfs_previous_item(root, path, 0, BTRFS_EXTENT_ITEM_KEY);
d548ee51 232 if (ret < 0)
ef8bbdfe 233 goto err;
d548ee51
Y		 234 if (ret == 0) {
235 leaf = path->nodes[0];
236 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
237 if (key.objectid + key.offset > first_free)
238 first_free = key.objectid + key.offset;
239 }
e37c9e69 240 while(1) {
5f39d397 241 leaf = path->nodes[0];
e37c9e69 242 slot = path->slots[0];
5f39d397 243 if (slot >= btrfs_header_nritems(leaf)) {
e37c9e69 244 ret = btrfs_next_leaf(root, path);
54aa1f4d
CM		 245 if (ret < 0)
246 goto err;
0f9dd46c 247 if (ret == 0)
e37c9e69 248 continue;
0f9dd46c 249 else
e37c9e69 250 break;
e37c9e69 251 }
5f39d397 252 btrfs_item_key_to_cpu(leaf, &key, slot);
0f9dd46c 253 if (key.objectid < block_group->key.objectid)
7d7d6068 254 goto next;
0f9dd46c 255
e37c9e69 256 if (key.objectid >= block_group->key.objectid +
0f9dd46c 257 block_group->key.offset)
e37c9e69 258 break;
7d7d6068 259
e37c9e69
CM		 260 if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
261 if (!found) {
7d7d6068 262 last = first_free;
e37c9e69 263 found = 1;
e37c9e69 264 }
0f9dd46c
JB		 265
266 add_new_free_space(block_group, root->fs_info, last,
267 key.objectid);
268
7d7d6068 269 last = key.objectid + key.offset;
e37c9e69 270 }
7d7d6068 271next:
e37c9e69
CM		 272 path->slots[0]++;
273 }
274
7d7d6068
Y		 275 if (!found)
276 last = first_free;
0f9dd46c
JB		 277
278 add_new_free_space(block_group, root->fs_info, last,
279 block_group->key.objectid +
280 block_group->key.offset);
281
e37c9e69 282 block_group->cached = 1;
ef8bbdfe 283 ret = 0;
54aa1f4d 284err:
e37c9e69 285 btrfs_free_path(path);
ef8bbdfe 286 return ret;
e37c9e69
CM		 287}
288
0f9dd46c
JB		 289/*
290 * return the block group that starts at or after bytenr
291 */
0ef3e66b
CM		 292struct btrfs_block_group_cache *btrfs_lookup_first_block_group(struct
293 btrfs_fs_info *info,
294 u64 bytenr)
295{
0f9dd46c 296 struct btrfs_block_group_cache *cache;
0ef3e66b 297
0f9dd46c 298 cache = block_group_cache_tree_search(info, bytenr, 0);
0ef3e66b 299
0f9dd46c 300 return cache;
0ef3e66b
CM		 301}
302
0f9dd46c
JB		 303/*
304 * return the block group that contains teh given bytenr
305 */
5276aeda
CM		 306struct btrfs_block_group_cache *btrfs_lookup_block_group(struct
307 btrfs_fs_info *info,
db94535d 308 u64 bytenr)
be744175 309{
0f9dd46c 310 struct btrfs_block_group_cache *cache;
be744175 311
0f9dd46c 312 cache = block_group_cache_tree_search(info, bytenr, 1);
96b5179d 313
0f9dd46c 314 return cache;
be744175 315}
0b86a832 316
0f9dd46c
JB		 317static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
318 u64 flags)
6324fbf3 319{
0f9dd46c
JB		 320 struct list_head *head = &info->space_info;
321 struct list_head *cur;
322 struct btrfs_space_info *found;
323 list_for_each(cur, head) {
324 found = list_entry(cur, struct btrfs_space_info, list);
325 if (found->flags == flags)
326 return found;
327 }
328 return NULL;
6324fbf3
CM		 329}
330
80eb234a
JB		 331static u64 div_factor(u64 num, int factor)
332{
333 if (factor == 10)
334 return num;
335 num *= factor;
336 do_div(num, 10);
337 return num;
338}
339
925baedd
CM		 340static struct btrfs_block_group_cache *
341__btrfs_find_block_group(struct btrfs_root *root,
342 struct btrfs_block_group_cache *hint,
343 u64 search_start, int data, int owner)
cd1bc465 344{
96b5179d 345 struct btrfs_block_group_cache *cache;
31f3c99b 346 struct btrfs_block_group_cache *found_group = NULL;
cd1bc465
CM		 347 struct btrfs_fs_info *info = root->fs_info;
348 u64 used;
31f3c99b 349 u64 last = 0;
96b5179d 350 u64 free_check;
31f3c99b 351 int full_search = 0;
bce4eae9 352 int factor = 10;
0ef3e66b 353 int wrapped = 0;
de428b63 354
a236aed1
CM		 355 if (data & BTRFS_BLOCK_GROUP_METADATA)
356 factor = 9;
be744175 357
0ef3e66b 358 if (search_start) {
be744175 359 struct btrfs_block_group_cache *shint;
0ef3e66b 360 shint = btrfs_lookup_first_block_group(info, search_start);
2b82032c 361 if (shint && block_group_bits(shint, data)) {
c286ac48 362 spin_lock(&shint->lock);
be744175 363 used = btrfs_block_group_used(&shint->item);
e8569813 364 if (used + shint->pinned + shint->reserved <
324ae4df 365 div_factor(shint->key.offset, factor)) {
c286ac48 366 spin_unlock(&shint->lock);
be744175
CM		 367 return shint;
368 }
c286ac48 369 spin_unlock(&shint->lock);
be744175
CM		 370 }
371 }
2b82032c 372 if (hint && block_group_bits(hint, data)) {
c286ac48 373 spin_lock(&hint->lock);
31f3c99b 374 used = btrfs_block_group_used(&hint->item);
e8569813 375 if (used + hint->pinned + hint->reserved <
324ae4df 376 div_factor(hint->key.offset, factor)) {
c286ac48 377 spin_unlock(&hint->lock);
31f3c99b
CM		 378 return hint;
379 }
c286ac48 380 spin_unlock(&hint->lock);
e19caa5f 381 last = hint->key.objectid + hint->key.offset;
31f3c99b 382 } else {
e37c9e69 383 if (hint)
0ef3e66b 384 last = max(hint->key.objectid, search_start);
e37c9e69 385 else
0ef3e66b 386 last = search_start;
31f3c99b 387 }
31f3c99b 388again:
e8569813
ZY		 389 while (1) {
390 cache = btrfs_lookup_first_block_group(root->fs_info, last);
0f9dd46c
JB		 391 if (!cache)
392 break;
96b5179d 393
c286ac48 394 spin_lock(&cache->lock);
96b5179d
CM		 395 last = cache->key.objectid + cache->key.offset;
396 used = btrfs_block_group_used(&cache->item);
397
2b82032c 398 if (block_group_bits(cache, data)) {
0ef3e66b 399 free_check = div_factor(cache->key.offset, factor);
e8569813
ZY		 400 if (used + cache->pinned + cache->reserved <
401 free_check) {
8790d502 402 found_group = cache;
c286ac48 403 spin_unlock(&cache->lock);
8790d502
CM		 404 goto found;
405 }
6324fbf3 406 }
c286ac48 407 spin_unlock(&cache->lock);
de428b63 408 cond_resched();
cd1bc465 409 }
0ef3e66b
CM		 410 if (!wrapped) {
411 last = search_start;
412 wrapped = 1;
413 goto again;
414 }
415 if (!full_search && factor < 10) {
be744175 416 last = search_start;
31f3c99b 417 full_search = 1;
0ef3e66b 418 factor = 10;
31f3c99b
CM		 419 goto again;
420 }
be744175 421found:
31f3c99b 422 return found_group;
cd1bc465
CM		 423}
424
925baedd
CM		 425struct btrfs_block_group_cache *btrfs_find_block_group(struct btrfs_root *root,
426 struct btrfs_block_group_cache
427 *hint, u64 search_start,
428 int data, int owner)
429{
430
431 struct btrfs_block_group_cache *ret;
925baedd 432 ret = __btrfs_find_block_group(root, hint, search_start, data, owner);
925baedd
CM		 433 return ret;
434}
0f9dd46c 435
e02119d5 436/* simple helper to search for an existing extent at a given offset */
31840ae1 437int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
e02119d5
CM		 438{
439 int ret;
440 struct btrfs_key key;
31840ae1 441 struct btrfs_path *path;
e02119d5 442
31840ae1
ZY		 443 path = btrfs_alloc_path();
444 BUG_ON(!path);
e02119d5
CM		 445 key.objectid = start;
446 key.offset = len;
447 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
448 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
449 0, 0);
31840ae1 450 btrfs_free_path(path);
7bb86316
CM		 451 return ret;
452}
453
d8d5f3e1
CM		 454/*
455 * Back reference rules. Back refs have three main goals:
456 *
457 * 1) differentiate between all holders of references to an extent so that
458 * when a reference is dropped we can make sure it was a valid reference
459 * before freeing the extent.
460 *
461 * 2) Provide enough information to quickly find the holders of an extent
462 * if we notice a given block is corrupted or bad.
463 *
464 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
465 * maintenance. This is actually the same as #2, but with a slightly
466 * different use case.
467 *
468 * File extents can be referenced by:
469 *
470 * - multiple snapshots, subvolumes, or different generations in one subvol
31840ae1 471 * - different files inside a single subvolume
d8d5f3e1
CM		 472 * - different offsets inside a file (bookend extents in file.c)
473 *
474 * The extent ref structure has fields for:
475 *
476 * - Objectid of the subvolume root
477 * - Generation number of the tree holding the reference
478 * - objectid of the file holding the reference
31840ae1
ZY		 479 * - number of references holding by parent node (alway 1 for tree blocks)
480 *
481 * Btree leaf may hold multiple references to a file extent. In most cases,
482 * these references are from same file and the corresponding offsets inside
3bb1a1bc 483 * the file are close together.
d8d5f3e1
CM		 484 *
485 * When a file extent is allocated the fields are filled in:
3bb1a1bc 486 * (root_key.objectid, trans->transid, inode objectid, 1)
d8d5f3e1
CM		 487 *
488 * When a leaf is cow'd new references are added for every file extent found
31840ae1
ZY		 489 * in the leaf. It looks similar to the create case, but trans->transid will
490 * be different when the block is cow'd.
d8d5f3e1 491 *
3bb1a1bc 492 * (root_key.objectid, trans->transid, inode objectid,
31840ae1 493 * number of references in the leaf)
d8d5f3e1 494 *
3bb1a1bc
YZ		 495 * When a file extent is removed either during snapshot deletion or
496 * file truncation, we find the corresponding back reference and check
497 * the following fields:
d8d5f3e1 498 *
3bb1a1bc
YZ		 499 * (btrfs_header_owner(leaf), btrfs_header_generation(leaf),
500 * inode objectid)
d8d5f3e1
CM		 501 *
502 * Btree extents can be referenced by:
503 *
504 * - Different subvolumes
505 * - Different generations of the same subvolume
506 *
d8d5f3e1
CM		 507 * When a tree block is created, back references are inserted:
508 *
3bb1a1bc 509 * (root->root_key.objectid, trans->transid, level, 1)
d8d5f3e1 510 *
31840ae1
ZY		 511 * When a tree block is cow'd, new back references are added for all the
512 * blocks it points to. If the tree block isn't in reference counted root,
513 * the old back references are removed. These new back references are of
514 * the form (trans->transid will have increased since creation):
d8d5f3e1 515 *
3bb1a1bc 516 * (root->root_key.objectid, trans->transid, level, 1)
d8d5f3e1 517 *
31840ae1 518 * When a backref is in deleting, the following fields are checked:
d8d5f3e1
CM		 519 *
520 * if backref was for a tree root:
3bb1a1bc 521 * (btrfs_header_owner(itself), btrfs_header_generation(itself), level)
d8d5f3e1 522 * else
3bb1a1bc 523 * (btrfs_header_owner(parent), btrfs_header_generation(parent), level)
d8d5f3e1 524 *
31840ae1 525 * Back Reference Key composing:
d8d5f3e1 526 *
31840ae1
ZY		 527 * The key objectid corresponds to the first byte in the extent, the key
528 * type is set to BTRFS_EXTENT_REF_KEY, and the key offset is the first
529 * byte of parent extent. If a extent is tree root, the key offset is set
530 * to the key objectid.
d8d5f3e1 531 */
31840ae1
ZY		 532
533static int noinline lookup_extent_backref(struct btrfs_trans_handle *trans,
534 struct btrfs_root *root,
3bb1a1bc
YZ		 535 struct btrfs_path *path,
536 u64 bytenr, u64 parent,
537 u64 ref_root, u64 ref_generation,
538 u64 owner_objectid, int del)
7bb86316 539{
7bb86316 540 struct btrfs_key key;
31840ae1
ZY		 541 struct btrfs_extent_ref *ref;
542 struct extent_buffer *leaf;
3bb1a1bc 543 u64 ref_objectid;
74493f7a
CM		 544 int ret;
545
31840ae1
ZY		 546 key.objectid = bytenr;
547 key.type = BTRFS_EXTENT_REF_KEY;
548 key.offset = parent;
549
550 ret = btrfs_search_slot(trans, root, &key, path, del ? -1 : 0, 1);
551 if (ret < 0)
552 goto out;
553 if (ret > 0) {
554 ret = -ENOENT;
555 goto out;
556 }
557
558 leaf = path->nodes[0];
559 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
3bb1a1bc 560 ref_objectid = btrfs_ref_objectid(leaf, ref);
31840ae1 561 if (btrfs_ref_root(leaf, ref) != ref_root ||
3bb1a1bc
YZ		 562 btrfs_ref_generation(leaf, ref) != ref_generation ||
563 (ref_objectid != owner_objectid &&
564 ref_objectid != BTRFS_MULTIPLE_OBJECTIDS)) {
31840ae1
ZY		 565 ret = -EIO;
566 WARN_ON(1);
567 goto out;
568 }
569 ret = 0;
570out:
571 return ret;
572}
573
f3465ca4
JB		 574/*
575 * updates all the backrefs that are pending on update_list for the
576 * extent_root
577 */
578static int noinline update_backrefs(struct btrfs_trans_handle *trans,
579 struct btrfs_root *extent_root,
580 struct btrfs_path *path,
581 struct list_head *update_list)
582{
583 struct btrfs_key key;
584 struct btrfs_extent_ref *ref;
585 struct btrfs_fs_info *info = extent_root->fs_info;
586 struct pending_extent_op *op;
587 struct extent_buffer *leaf;
588 int ret = 0;
589 struct list_head *cur = update_list->next;
590 u64 ref_objectid;
591 u64 ref_root = extent_root->root_key.objectid;
592
593 op = list_entry(cur, struct pending_extent_op, list);
594
595search:
596 key.objectid = op->bytenr;
597 key.type = BTRFS_EXTENT_REF_KEY;
598 key.offset = op->orig_parent;
599
600 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 1);
601 BUG_ON(ret);
602
603 leaf = path->nodes[0];
604
605loop:
606 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
607
608 ref_objectid = btrfs_ref_objectid(leaf, ref);
609
610 if (btrfs_ref_root(leaf, ref) != ref_root ||
611 btrfs_ref_generation(leaf, ref) != op->orig_generation ||
612 (ref_objectid != op->level &&
613 ref_objectid != BTRFS_MULTIPLE_OBJECTIDS)) {
614 printk(KERN_ERR "couldn't find %Lu, parent %Lu, root %Lu, "
615 "owner %u\n", op->bytenr, op->orig_parent,
616 ref_root, op->level);
617 btrfs_print_leaf(extent_root, leaf);
618 BUG();
619 }
620
621 key.objectid = op->bytenr;
622 key.offset = op->parent;
623 key.type = BTRFS_EXTENT_REF_KEY;
624 ret = btrfs_set_item_key_safe(trans, extent_root, path, &key);
625 BUG_ON(ret);
626 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
627 btrfs_set_ref_generation(leaf, ref, op->generation);
628
629 cur = cur->next;
630
631 list_del_init(&op->list);
632 unlock_extent(&info->extent_ins, op->bytenr,
633 op->bytenr + op->num_bytes - 1, GFP_NOFS);
634 kfree(op);
635
636 if (cur == update_list) {
637 btrfs_mark_buffer_dirty(path->nodes[0]);
638 btrfs_release_path(extent_root, path);
639 goto out;
640 }
641
642 op = list_entry(cur, struct pending_extent_op, list);
643
644 path->slots[0]++;
645 while (path->slots[0] < btrfs_header_nritems(leaf)) {
646 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
647 if (key.objectid == op->bytenr &&
648 key.type == BTRFS_EXTENT_REF_KEY)
649 goto loop;
650 path->slots[0]++;
651 }
652
653 btrfs_mark_buffer_dirty(path->nodes[0]);
654 btrfs_release_path(extent_root, path);
655 goto search;
656
657out:
658 return 0;
659}
660
661static int noinline insert_extents(struct btrfs_trans_handle *trans,
662 struct btrfs_root *extent_root,
663 struct btrfs_path *path,
664 struct list_head *insert_list, int nr)
665{
666 struct btrfs_key *keys;
667 u32 *data_size;
668 struct pending_extent_op *op;
669 struct extent_buffer *leaf;
670 struct list_head *cur = insert_list->next;
671 struct btrfs_fs_info *info = extent_root->fs_info;
672 u64 ref_root = extent_root->root_key.objectid;
673 int i = 0, last = 0, ret;
674 int total = nr * 2;
675
676 if (!nr)
677 return 0;
678
679 keys = kzalloc(total * sizeof(struct btrfs_key), GFP_NOFS);
680 if (!keys)
681 return -ENOMEM;
682
683 data_size = kzalloc(total * sizeof(u32), GFP_NOFS);
684 if (!data_size) {
685 kfree(keys);
686 return -ENOMEM;
687 }
688
689 list_for_each_entry(op, insert_list, list) {
690 keys[i].objectid = op->bytenr;
691 keys[i].offset = op->num_bytes;
692 keys[i].type = BTRFS_EXTENT_ITEM_KEY;
693 data_size[i] = sizeof(struct btrfs_extent_item);
694 i++;
695
696 keys[i].objectid = op->bytenr;
697 keys[i].offset = op->parent;
698 keys[i].type = BTRFS_EXTENT_REF_KEY;
699 data_size[i] = sizeof(struct btrfs_extent_ref);
700 i++;
701 }
702
703 op = list_entry(cur, struct pending_extent_op, list);
704 i = 0;
705 while (i < total) {
706 int c;
707 ret = btrfs_insert_some_items(trans, extent_root, path,
708 keys+i, data_size+i, total-i);
709 BUG_ON(ret < 0);
710
711 if (last && ret > 1)
712 BUG();
713
714 leaf = path->nodes[0];
715 for (c = 0; c < ret; c++) {
716 int ref_first = keys[i].type == BTRFS_EXTENT_REF_KEY;
717
718 /*
719 * if the first item we inserted was a backref, then
720 * the EXTENT_ITEM will be the odd c's, else it will
721 * be the even c's
722 */
723 if ((ref_first && (c % 2)) ||
724 (!ref_first && !(c % 2))) {
725 struct btrfs_extent_item *itm;
726
727 itm = btrfs_item_ptr(leaf, path->slots[0] + c,
728 struct btrfs_extent_item);
729 btrfs_set_extent_refs(path->nodes[0], itm, 1);
730 op->del++;
731 } else {
732 struct btrfs_extent_ref *ref;
733
734 ref = btrfs_item_ptr(leaf, path->slots[0] + c,
735 struct btrfs_extent_ref);
736 btrfs_set_ref_root(leaf, ref, ref_root);
737 btrfs_set_ref_generation(leaf, ref,
738 op->generation);
739 btrfs_set_ref_objectid(leaf, ref, op->level);
740 btrfs_set_ref_num_refs(leaf, ref, 1);
741 op->del++;
742 }
743
744 /*
745 * using del to see when its ok to free up the
746 * pending_extent_op. In the case where we insert the
747 * last item on the list in order to help do batching
748 * we need to not free the extent op until we actually
749 * insert the extent_item
750 */
751 if (op->del == 2) {
752 unlock_extent(&info->extent_ins, op->bytenr,
753 op->bytenr + op->num_bytes - 1,
754 GFP_NOFS);
755 cur = cur->next;
756 list_del_init(&op->list);
757 kfree(op);
758 if (cur != insert_list)
759 op = list_entry(cur,
760 struct pending_extent_op,
761 list);
762 }
763 }
764 btrfs_mark_buffer_dirty(leaf);
765 btrfs_release_path(extent_root, path);
766
767 /*
768 * Ok backref's and items usually go right next to eachother,
769 * but if we could only insert 1 item that means that we
770 * inserted on the end of a leaf, and we have no idea what may
771 * be on the next leaf so we just play it safe. In order to
772 * try and help this case we insert the last thing on our
773 * insert list so hopefully it will end up being the last
774 * thing on the leaf and everything else will be before it,
775 * which will let us insert a whole bunch of items at the same
776 * time.
777 */
778 if (ret == 1 && !last && (i + ret < total)) {
779 /*
780 * last: where we will pick up the next time around
781 * i: our current key to insert, will be total - 1
782 * cur: the current op we are screwing with
783 * op: duh
784 */
785 last = i + ret;
786 i = total - 1;
787 cur = insert_list->prev;
788 op = list_entry(cur, struct pending_extent_op, list);
789 } else if (last) {
790 /*
791 * ok we successfully inserted the last item on the
792 * list, lets reset everything
793 *
794 * i: our current key to insert, so where we left off
795 * last time
796 * last: done with this
797 * cur: the op we are messing with
798 * op: duh
799 * total: since we inserted the last key, we need to
800 * decrement total so we dont overflow
801 */
802 i = last;
803 last = 0;
f3465ca4 804 total--;
b4eec2ca
LH		 805 if (i < total) {
806 cur = insert_list->next;
807 op = list_entry(cur, struct pending_extent_op,
808 list);
809 }
f3465ca4
JB		 810 } else {
811 i += ret;
812 }
813
814 cond_resched();
815 }
816 ret = 0;
817 kfree(keys);
818 kfree(data_size);
819 return ret;
820}
821
31840ae1
ZY		 822static int noinline insert_extent_backref(struct btrfs_trans_handle *trans,
823 struct btrfs_root *root,
824 struct btrfs_path *path,
825 u64 bytenr, u64 parent,
826 u64 ref_root, u64 ref_generation,
3bb1a1bc 827 u64 owner_objectid)
31840ae1
ZY		 828{
829 struct btrfs_key key;
830 struct extent_buffer *leaf;
831 struct btrfs_extent_ref *ref;
832 u32 num_refs;
833 int ret;
74493f7a 834
74493f7a
CM		 835 key.objectid = bytenr;
836 key.type = BTRFS_EXTENT_REF_KEY;
31840ae1 837 key.offset = parent;
74493f7a 838
31840ae1
ZY		 839 ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*ref));
840 if (ret == 0) {
841 leaf = path->nodes[0];
842 ref = btrfs_item_ptr(leaf, path->slots[0],
843 struct btrfs_extent_ref);
844 btrfs_set_ref_root(leaf, ref, ref_root);
845 btrfs_set_ref_generation(leaf, ref, ref_generation);
846 btrfs_set_ref_objectid(leaf, ref, owner_objectid);
31840ae1
ZY		 847 btrfs_set_ref_num_refs(leaf, ref, 1);
848 } else if (ret == -EEXIST) {
849 u64 existing_owner;
850 BUG_ON(owner_objectid < BTRFS_FIRST_FREE_OBJECTID);
851 leaf = path->nodes[0];
852 ref = btrfs_item_ptr(leaf, path->slots[0],
853 struct btrfs_extent_ref);
854 if (btrfs_ref_root(leaf, ref) != ref_root ||
855 btrfs_ref_generation(leaf, ref) != ref_generation) {
856 ret = -EIO;
857 WARN_ON(1);
7bb86316 858 goto out;
31840ae1
ZY		 859 }
860
861 num_refs = btrfs_ref_num_refs(leaf, ref);
862 BUG_ON(num_refs == 0);
863 btrfs_set_ref_num_refs(leaf, ref, num_refs + 1);
864
865 existing_owner = btrfs_ref_objectid(leaf, ref);
3bb1a1bc
YZ		 866 if (existing_owner != owner_objectid &&
867 existing_owner != BTRFS_MULTIPLE_OBJECTIDS) {
31840ae1
ZY		 868 btrfs_set_ref_objectid(leaf, ref,
869 BTRFS_MULTIPLE_OBJECTIDS);
31840ae1
ZY		 870 }
871 ret = 0;
872 } else {
7bb86316 873 goto out;
31840ae1 874 }
7bb86316
CM		 875 btrfs_mark_buffer_dirty(path->nodes[0]);
876out:
877 btrfs_release_path(root, path);
878 return ret;
74493f7a
CM		 879}
880
31840ae1
ZY		 881static int noinline remove_extent_backref(struct btrfs_trans_handle *trans,
882 struct btrfs_root *root,
883 struct btrfs_path *path)
884{
885 struct extent_buffer *leaf;
886 struct btrfs_extent_ref *ref;
887 u32 num_refs;
888 int ret = 0;
889
890 leaf = path->nodes[0];
891 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
892 num_refs = btrfs_ref_num_refs(leaf, ref);
893 BUG_ON(num_refs == 0);
894 num_refs -= 1;
895 if (num_refs == 0) {
896 ret = btrfs_del_item(trans, root, path);
897 } else {
898 btrfs_set_ref_num_refs(leaf, ref, num_refs);
899 btrfs_mark_buffer_dirty(leaf);
900 }
901 btrfs_release_path(root, path);
902 return ret;
903}
904
4b4e25f2 905#ifdef BIO_RW_DISCARD
15916de8
CM		 906static void btrfs_issue_discard(struct block_device *bdev,
907 u64 start, u64 len)
908{
909#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,28)
910 blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL);
911#else
912 blkdev_issue_discard(bdev, start >> 9, len >> 9);
913#endif
914}
4b4e25f2 915#endif
15916de8 916
f3465ca4
JB		 917static int noinline free_extents(struct btrfs_trans_handle *trans,
918 struct btrfs_root *extent_root,
919 struct list_head *del_list)
920{
921 struct btrfs_fs_info *info = extent_root->fs_info;
922 struct btrfs_path *path;
923 struct btrfs_key key, found_key;
924 struct extent_buffer *leaf;
925 struct list_head *cur;
926 struct pending_extent_op *op;
927 struct btrfs_extent_item *ei;
928 int ret, num_to_del, extent_slot = 0, found_extent = 0;
929 u32 refs;
930 u64 bytes_freed = 0;
931
932 path = btrfs_alloc_path();
933 if (!path)
934 return -ENOMEM;
935 path->reada = 1;
936
937search:
938 /* search for the backref for the current ref we want to delete */
939 cur = del_list->next;
940 op = list_entry(cur, struct pending_extent_op, list);
941 ret = lookup_extent_backref(trans, extent_root, path, op->bytenr,
942 op->orig_parent,
943 extent_root->root_key.objectid,
944 op->orig_generation, op->level, 1);
945 if (ret) {
946 printk("Unable to find backref byte nr %Lu root %Lu gen %Lu "
947 "owner %u\n", op->bytenr,
948 extent_root->root_key.objectid, op->orig_generation,
949 op->level);
950 btrfs_print_leaf(extent_root, path->nodes[0]);
951 WARN_ON(1);
952 goto out;
953 }
954
955 extent_slot = path->slots[0];
956 num_to_del = 1;
957 found_extent = 0;
958
959 /*
960 * if we aren't the first item on the leaf we can move back one and see
961 * if our ref is right next to our extent item
962 */
963 if (likely(extent_slot)) {
964 extent_slot--;
965 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
966 extent_slot);
967 if (found_key.objectid == op->bytenr &&
968 found_key.type == BTRFS_EXTENT_ITEM_KEY &&
969 found_key.offset == op->num_bytes) {
970 num_to_del++;
971 found_extent = 1;
972 }
973 }
974
975 /*
976 * if we didn't find the extent we need to delete the backref and then
977 * search for the extent item key so we can update its ref count
978 */
979 if (!found_extent) {
980 key.objectid = op->bytenr;
981 key.type = BTRFS_EXTENT_ITEM_KEY;
982 key.offset = op->num_bytes;
983
984 ret = remove_extent_backref(trans, extent_root, path);
985 BUG_ON(ret);
986 btrfs_release_path(extent_root, path);
987 ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1);
988 BUG_ON(ret);
989 extent_slot = path->slots[0];
990 }
991
992 /* this is where we update the ref count for the extent */
993 leaf = path->nodes[0];
994 ei = btrfs_item_ptr(leaf, extent_slot, struct btrfs_extent_item);
995 refs = btrfs_extent_refs(leaf, ei);
996 BUG_ON(refs == 0);
997 refs--;
998 btrfs_set_extent_refs(leaf, ei, refs);
999
1000 btrfs_mark_buffer_dirty(leaf);
1001
1002 /*
1003 * This extent needs deleting. The reason cur_slot is extent_slot +
1004 * num_to_del is because extent_slot points to the slot where the extent
1005 * is, and if the backref was not right next to the extent we will be
1006 * deleting at least 1 item, and will want to start searching at the
1007 * slot directly next to extent_slot. However if we did find the
1008 * backref next to the extent item them we will be deleting at least 2
1009 * items and will want to start searching directly after the ref slot
1010 */
1011 if (!refs) {
1012 struct list_head *pos, *n, *end;
1013 int cur_slot = extent_slot+num_to_del;
1014 u64 super_used;
1015 u64 root_used;
1016
1017 path->slots[0] = extent_slot;
1018 bytes_freed = op->num_bytes;
1019
e3e469f8
JB		 1020 mutex_lock(&info->pinned_mutex);
1021 ret = pin_down_bytes(trans, extent_root, op->bytenr,
1022 op->num_bytes, op->level >=
1023 BTRFS_FIRST_FREE_OBJECTID);
1024 mutex_unlock(&info->pinned_mutex);
1025 BUG_ON(ret < 0);
1026 op->del = ret;
1027
f3465ca4
JB		 1028 /*
1029 * we need to see if we can delete multiple things at once, so
1030 * start looping through the list of extents we are wanting to
1031 * delete and see if their extent/backref's are right next to
1032 * eachother and the extents only have 1 ref
1033 */
1034 for (pos = cur->next; pos != del_list; pos = pos->next) {
1035 struct pending_extent_op *tmp;
1036
1037 tmp = list_entry(pos, struct pending_extent_op, list);
1038
1039 /* we only want to delete extent+ref at this stage */
1040 if (cur_slot >= btrfs_header_nritems(leaf) - 1)
1041 break;
1042
1043 btrfs_item_key_to_cpu(leaf, &found_key, cur_slot);
1044 if (found_key.objectid != tmp->bytenr ||
1045 found_key.type != BTRFS_EXTENT_ITEM_KEY ||
1046 found_key.offset != tmp->num_bytes)
1047 break;
1048
1049 /* check to make sure this extent only has one ref */
1050 ei = btrfs_item_ptr(leaf, cur_slot,
1051 struct btrfs_extent_item);
1052 if (btrfs_extent_refs(leaf, ei) != 1)
1053 break;
1054
1055 btrfs_item_key_to_cpu(leaf, &found_key, cur_slot+1);
1056 if (found_key.objectid != tmp->bytenr ||
1057 found_key.type != BTRFS_EXTENT_REF_KEY ||
1058 found_key.offset != tmp->orig_parent)
1059 break;
1060
1061 /*
1062 * the ref is right next to the extent, we can set the
1063 * ref count to 0 since we will delete them both now
1064 */
1065 btrfs_set_extent_refs(leaf, ei, 0);
1066
1067 /* pin down the bytes for this extent */
1068 mutex_lock(&info->pinned_mutex);
1069 ret = pin_down_bytes(trans, extent_root, tmp->bytenr,
1070 tmp->num_bytes, tmp->level >=
1071 BTRFS_FIRST_FREE_OBJECTID);
1072 mutex_unlock(&info->pinned_mutex);
1073 BUG_ON(ret < 0);
1074
1075 /*
1076 * use the del field to tell if we need to go ahead and
1077 * free up the extent when we delete the item or not.
1078 */
1079 tmp->del = ret;
1080 bytes_freed += tmp->num_bytes;
1081
1082 num_to_del += 2;
1083 cur_slot += 2;
1084 }
1085 end = pos;
1086
1087 /* update the free space counters */
1088 spin_lock_irq(&info->delalloc_lock);
1089 super_used = btrfs_super_bytes_used(&info->super_copy);
1090 btrfs_set_super_bytes_used(&info->super_copy,
1091 super_used - bytes_freed);
1092 spin_unlock_irq(&info->delalloc_lock);
1093
1094 root_used = btrfs_root_used(&extent_root->root_item);
1095 btrfs_set_root_used(&extent_root->root_item,
1096 root_used - bytes_freed);
1097
1098 /* delete the items */
1099 ret = btrfs_del_items(trans, extent_root, path,
1100 path->slots[0], num_to_del);
1101 BUG_ON(ret);
1102
1103 /*
1104 * loop through the extents we deleted and do the cleanup work
1105 * on them
1106 */
1107 for (pos = cur, n = pos->next; pos != end;
1108 pos = n, n = pos->next) {
1109 struct pending_extent_op *tmp;
1110#ifdef BIO_RW_DISCARD
1111 u64 map_length;
1112 struct btrfs_multi_bio *multi = NULL;
1113#endif
1114 tmp = list_entry(pos, struct pending_extent_op, list);
1115
1116 /*
1117 * remember tmp->del tells us wether or not we pinned
1118 * down the extent
1119 */
1120 ret = update_block_group(trans, extent_root,
1121 tmp->bytenr, tmp->num_bytes, 0,
1122 tmp->del);
1123 BUG_ON(ret);
1124
1125#ifdef BIO_RW_DISCARD
15916de8 1126 map_length = tmp->num_bytes;
f3465ca4
JB		 1127 ret = btrfs_map_block(&info->mapping_tree, READ,
1128 tmp->bytenr, &map_length, &multi,
1129 0);
1130 if (!ret) {
1131 struct btrfs_bio_stripe *stripe;
1132 int i;
1133
15916de8 1134 stripe = multi->stripes;
f3465ca4
JB		 1135
1136 if (map_length > tmp->num_bytes)
1137 map_length = tmp->num_bytes;
1138
1139 for (i = 0; i < multi->num_stripes;
1140 i++, stripe++)
15916de8
CM		 1141 btrfs_issue_discard(stripe->dev->bdev,
1142 stripe->physical,
1143 map_length);
f3465ca4
JB		 1144 kfree(multi);
1145 }
1146#endif
1147 list_del_init(&tmp->list);
1148 unlock_extent(&info->extent_ins, tmp->bytenr,
1149 tmp->bytenr + tmp->num_bytes - 1,
1150 GFP_NOFS);
1151 kfree(tmp);
1152 }
1153 } else if (refs && found_extent) {
1154 /*
1155 * the ref and extent were right next to eachother, but the
1156 * extent still has a ref, so just free the backref and keep
1157 * going
1158 */
1159 ret = remove_extent_backref(trans, extent_root, path);
1160 BUG_ON(ret);
1161
1162 list_del_init(&op->list);
1163 unlock_extent(&info->extent_ins, op->bytenr,
1164 op->bytenr + op->num_bytes - 1, GFP_NOFS);
1165 kfree(op);
1166 } else {
1167 /*
1168 * the extent has multiple refs and the backref we were looking
1169 * for was not right next to it, so just unlock and go next,
1170 * we're good to go
1171 */
1172 list_del_init(&op->list);
1173 unlock_extent(&info->extent_ins, op->bytenr,
1174 op->bytenr + op->num_bytes - 1, GFP_NOFS);
1175 kfree(op);
1176 }
1177
1178 btrfs_release_path(extent_root, path);
1179 if (!list_empty(del_list))
1180 goto search;
1181
1182out:
1183 btrfs_free_path(path);
1184 return ret;
1185}
1186
31840ae1
ZY		 1187static int __btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
1188 struct btrfs_root *root, u64 bytenr,
1189 u64 orig_parent, u64 parent,
1190 u64 orig_root, u64 ref_root,
1191 u64 orig_generation, u64 ref_generation,
3bb1a1bc 1192 u64 owner_objectid)
31840ae1
ZY		 1193{
1194 int ret;
1195 struct btrfs_root *extent_root = root->fs_info->extent_root;
1196 struct btrfs_path *path;
1197
1198 if (root == root->fs_info->extent_root) {
1199 struct pending_extent_op *extent_op;
1200 u64 num_bytes;
1201
1202 BUG_ON(owner_objectid >= BTRFS_MAX_LEVEL);
1203 num_bytes = btrfs_level_size(root, (int)owner_objectid);
25179201 1204 mutex_lock(&root->fs_info->extent_ins_mutex);
31840ae1 1205 if (test_range_bit(&root->fs_info->extent_ins, bytenr,
25179201 1206 bytenr + num_bytes - 1, EXTENT_WRITEBACK, 0)) {
31840ae1
ZY		 1207 u64 priv;
1208 ret = get_state_private(&root->fs_info->extent_ins,
1209 bytenr, &priv);
1210 BUG_ON(ret);
1211 extent_op = (struct pending_extent_op *)
1212 (unsigned long)priv;
1213 BUG_ON(extent_op->parent != orig_parent);
1214 BUG_ON(extent_op->generation != orig_generation);
25179201 1215
31840ae1
ZY		 1216 extent_op->parent = parent;
1217 extent_op->generation = ref_generation;
1218 } else {
1219 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
1220 BUG_ON(!extent_op);
1221
1222 extent_op->type = PENDING_BACKREF_UPDATE;
1223 extent_op->bytenr = bytenr;
1224 extent_op->num_bytes = num_bytes;
1225 extent_op->parent = parent;
1226 extent_op->orig_parent = orig_parent;
1227 extent_op->generation = ref_generation;
1228 extent_op->orig_generation = orig_generation;
1229 extent_op->level = (int)owner_objectid;
f3465ca4
JB		 1230 INIT_LIST_HEAD(&extent_op->list);
1231 extent_op->del = 0;
31840ae1
ZY		 1232
1233 set_extent_bits(&root->fs_info->extent_ins,
1234 bytenr, bytenr + num_bytes - 1,
25179201 1235 EXTENT_WRITEBACK, GFP_NOFS);
31840ae1
ZY		 1236 set_state_private(&root->fs_info->extent_ins,
1237 bytenr, (unsigned long)extent_op);
1238 }
25179201 1239 mutex_unlock(&root->fs_info->extent_ins_mutex);
31840ae1
ZY		 1240 return 0;
1241 }
1242
1243 path = btrfs_alloc_path();
1244 if (!path)
1245 return -ENOMEM;
1246 ret = lookup_extent_backref(trans, extent_root, path,
1247 bytenr, orig_parent, orig_root,
3bb1a1bc 1248 orig_generation, owner_objectid, 1);
31840ae1
ZY		 1249 if (ret)
1250 goto out;
1251 ret = remove_extent_backref(trans, extent_root, path);
1252 if (ret)
1253 goto out;
1254 ret = insert_extent_backref(trans, extent_root, path, bytenr,
1255 parent, ref_root, ref_generation,
3bb1a1bc 1256 owner_objectid);
31840ae1 1257 BUG_ON(ret);
87ef2bb4
CM		 1258 finish_current_insert(trans, extent_root, 0);
1259 del_pending_extents(trans, extent_root, 0);
31840ae1
ZY		 1260out:
1261 btrfs_free_path(path);
1262 return ret;
1263}
1264
1265int btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
1266 struct btrfs_root *root, u64 bytenr,
1267 u64 orig_parent, u64 parent,
1268 u64 ref_root, u64 ref_generation,
3bb1a1bc 1269 u64 owner_objectid)
31840ae1
ZY		 1270{
1271 int ret;
1272 if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
1273 owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
1274 return 0;
31840ae1
ZY		 1275 ret = __btrfs_update_extent_ref(trans, root, bytenr, orig_parent,
1276 parent, ref_root, ref_root,
1277 ref_generation, ref_generation,
3bb1a1bc 1278 owner_objectid);
31840ae1
ZY		 1279 return ret;
1280}
1281
925baedd 1282static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
31840ae1
ZY		 1283 struct btrfs_root *root, u64 bytenr,
1284 u64 orig_parent, u64 parent,
1285 u64 orig_root, u64 ref_root,
1286 u64 orig_generation, u64 ref_generation,
3bb1a1bc 1287 u64 owner_objectid)
02217ed2 1288{
5caf2a00 1289 struct btrfs_path *path;
02217ed2 1290 int ret;
e2fa7227 1291 struct btrfs_key key;
5f39d397 1292 struct extent_buffer *l;
234b63a0 1293 struct btrfs_extent_item *item;
cf27e1ee 1294 u32 refs;
037e6390 1295
5caf2a00 1296 path = btrfs_alloc_path();
54aa1f4d
CM		 1297 if (!path)
1298 return -ENOMEM;
26b8003f 1299
3c12ac72 1300 path->reada = 1;
db94535d 1301 key.objectid = bytenr;
31840ae1
ZY		 1302 key.type = BTRFS_EXTENT_ITEM_KEY;
1303 key.offset = (u64)-1;
1304
5caf2a00 1305 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
9f5fae2f 1306 0, 1);
54aa1f4d
CM		 1307 if (ret < 0)
1308 return ret;
31840ae1
ZY		 1309 BUG_ON(ret == 0 || path->slots[0] == 0);
1310
1311 path->slots[0]--;
5f39d397 1312 l = path->nodes[0];
31840ae1
ZY		 1313
1314 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
771ed689
CM		 1315 if (key.objectid != bytenr) {
1316 btrfs_print_leaf(root->fs_info->extent_root, path->nodes[0]);
1317 printk("wanted %Lu found %Lu\n", bytenr, key.objectid);
1318 BUG();
1319 }
31840ae1
ZY		 1320 BUG_ON(key.type != BTRFS_EXTENT_ITEM_KEY);
1321
5caf2a00 1322 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
5f39d397
CM		 1323 refs = btrfs_extent_refs(l, item);
1324 btrfs_set_extent_refs(l, item, refs + 1);
5caf2a00 1325 btrfs_mark_buffer_dirty(path->nodes[0]);
a28ec197 1326
5caf2a00 1327 btrfs_release_path(root->fs_info->extent_root, path);
7bb86316 1328
3c12ac72 1329 path->reada = 1;
31840ae1
ZY		 1330 ret = insert_extent_backref(trans, root->fs_info->extent_root,
1331 path, bytenr, parent,
1332 ref_root, ref_generation,
3bb1a1bc 1333 owner_objectid);
7bb86316 1334 BUG_ON(ret);
87ef2bb4
CM		 1335 finish_current_insert(trans, root->fs_info->extent_root, 0);
1336 del_pending_extents(trans, root->fs_info->extent_root, 0);
74493f7a
CM		 1337
1338 btrfs_free_path(path);
02217ed2
CM		 1339 return 0;
1340}
1341
925baedd 1342int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
31840ae1
ZY		 1343 struct btrfs_root *root,
1344 u64 bytenr, u64 num_bytes, u64 parent,
1345 u64 ref_root, u64 ref_generation,
3bb1a1bc 1346 u64 owner_objectid)
925baedd
CM		 1347{
1348 int ret;
31840ae1
ZY		 1349 if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
1350 owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
1351 return 0;
31840ae1
ZY		 1352 ret = __btrfs_inc_extent_ref(trans, root, bytenr, 0, parent,
1353 0, ref_root, 0, ref_generation,
3bb1a1bc 1354 owner_objectid);
925baedd
CM		 1355 return ret;
1356}
1357
e9d0b13b
CM		 1358int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
1359 struct btrfs_root *root)
1360{
87ef2bb4
CM		 1361 finish_current_insert(trans, root->fs_info->extent_root, 1);
1362 del_pending_extents(trans, root->fs_info->extent_root, 1);
e9d0b13b
CM		 1363 return 0;
1364}
1365
31840ae1
ZY		 1366int btrfs_lookup_extent_ref(struct btrfs_trans_handle *trans,
1367 struct btrfs_root *root, u64 bytenr,
1368 u64 num_bytes, u32 *refs)
a28ec197 1369{
5caf2a00 1370 struct btrfs_path *path;
a28ec197 1371 int ret;
e2fa7227 1372 struct btrfs_key key;
5f39d397 1373 struct extent_buffer *l;
234b63a0 1374 struct btrfs_extent_item *item;
5caf2a00 1375
db94535d 1376 WARN_ON(num_bytes < root->sectorsize);
5caf2a00 1377 path = btrfs_alloc_path();
3c12ac72 1378 path->reada = 1;
db94535d
CM		 1379 key.objectid = bytenr;
1380 key.offset = num_bytes;
62e2749e 1381 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
5caf2a00 1382 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
9f5fae2f 1383 0, 0);
54aa1f4d
CM		 1384 if (ret < 0)
1385 goto out;
5f39d397
CM		 1386 if (ret != 0) {
1387 btrfs_print_leaf(root, path->nodes[0]);
db94535d 1388 printk("failed to find block number %Lu\n", bytenr);
a28ec197 1389 BUG();
5f39d397
CM		 1390 }
1391 l = path->nodes[0];
5caf2a00 1392 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
5f39d397 1393 *refs = btrfs_extent_refs(l, item);
54aa1f4d 1394out:
5caf2a00 1395 btrfs_free_path(path);
a28ec197
CM		 1396 return 0;
1397}
1398
80ff3856
YZ		 1399int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
1400 struct btrfs_root *root, u64 bytenr)
be20aa9d
CM		 1401{
1402 struct btrfs_root *extent_root = root->fs_info->extent_root;
1403 struct btrfs_path *path;
f321e491
YZ		 1404 struct extent_buffer *leaf;
1405 struct btrfs_extent_ref *ref_item;
1406 struct btrfs_key key;
1407 struct btrfs_key found_key;
80ff3856
YZ		 1408 u64 ref_root;
1409 u64 last_snapshot;
be20aa9d
CM		 1410 u32 nritems;
1411 int ret;
925baedd 1412
be20aa9d 1413 key.objectid = bytenr;
31840ae1 1414 key.offset = (u64)-1;
f321e491 1415 key.type = BTRFS_EXTENT_ITEM_KEY;
be20aa9d 1416
f321e491 1417 path = btrfs_alloc_path();
be20aa9d
CM		 1418 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
1419 if (ret < 0)
1420 goto out;
1421 BUG_ON(ret == 0);
80ff3856
YZ		 1422
1423 ret = -ENOENT;
1424 if (path->slots[0] == 0)
31840ae1 1425 goto out;
be20aa9d 1426
31840ae1 1427 path->slots[0]--;
f321e491
YZ		 1428 leaf = path->nodes[0];
1429 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
be20aa9d
CM		 1430
1431 if (found_key.objectid != bytenr ||
80ff3856 1432 found_key.type != BTRFS_EXTENT_ITEM_KEY)
be20aa9d 1433 goto out;
f321e491 1434
80ff3856 1435 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
be20aa9d 1436 while (1) {
f321e491
YZ		 1437 leaf = path->nodes[0];
1438 nritems = btrfs_header_nritems(leaf);
be20aa9d
CM		 1439 if (path->slots[0] >= nritems) {
1440 ret = btrfs_next_leaf(extent_root, path);
f321e491
YZ		 1441 if (ret < 0)
1442 goto out;
be20aa9d
CM		 1443 if (ret == 0)
1444 continue;
1445 break;
1446 }
f321e491 1447 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
be20aa9d
CM		 1448 if (found_key.objectid != bytenr)
1449 break;
bd09835d 1450
be20aa9d
CM		 1451 if (found_key.type != BTRFS_EXTENT_REF_KEY) {
1452 path->slots[0]++;
1453 continue;
1454 }
1455
f321e491 1456 ref_item = btrfs_item_ptr(leaf, path->slots[0],
be20aa9d 1457 struct btrfs_extent_ref);
80ff3856
YZ		 1458 ref_root = btrfs_ref_root(leaf, ref_item);
1459 if (ref_root != root->root_key.objectid &&
1460 ref_root != BTRFS_TREE_LOG_OBJECTID) {
1461 ret = 1;
f321e491 1462 goto out;
80ff3856
YZ		 1463 }
1464 if (btrfs_ref_generation(leaf, ref_item) <= last_snapshot) {
f321e491
YZ		 1465 ret = 1;
1466 goto out;
1467 }
80ff3856
YZ		 1468
1469 path->slots[0]++;
f321e491
YZ		 1470 }
1471 ret = 0;
be20aa9d 1472out:
80ff3856 1473 btrfs_free_path(path);
f321e491 1474 return ret;
be20aa9d 1475}
c5739bba 1476
31840ae1
ZY		 1477int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1478 struct extent_buffer *buf, u32 nr_extents)
02217ed2 1479{
5f39d397 1480 struct btrfs_key key;
6407bf6d 1481 struct btrfs_file_extent_item *fi;
e4657689
ZY		 1482 u64 root_gen;
1483 u32 nritems;
02217ed2 1484 int i;
db94535d 1485 int level;
31840ae1 1486 int ret = 0;
e4657689 1487 int shared = 0;
a28ec197 1488
3768f368 1489 if (!root->ref_cows)
a28ec197 1490 return 0;
5f39d397 1491
e4657689
ZY		 1492 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
1493 shared = 0;
1494 root_gen = root->root_key.offset;
1495 } else {
1496 shared = 1;
1497 root_gen = trans->transid - 1;
1498 }
1499
db94535d 1500 level = btrfs_header_level(buf);
5f39d397 1501 nritems = btrfs_header_nritems(buf);
4a096752 1502
31840ae1 1503 if (level == 0) {
31153d81
YZ		 1504 struct btrfs_leaf_ref *ref;
1505 struct btrfs_extent_info *info;
1506
31840ae1 1507 ref = btrfs_alloc_leaf_ref(root, nr_extents);
31153d81 1508 if (!ref) {
31840ae1 1509 ret = -ENOMEM;
31153d81
YZ		 1510 goto out;
1511 }
1512
e4657689 1513 ref->root_gen = root_gen;
31153d81
YZ		 1514 ref->bytenr = buf->start;
1515 ref->owner = btrfs_header_owner(buf);
1516 ref->generation = btrfs_header_generation(buf);
31840ae1 1517 ref->nritems = nr_extents;
31153d81 1518 info = ref->extents;
bcc63abb 1519
31840ae1 1520 for (i = 0; nr_extents > 0 && i < nritems; i++) {
31153d81
YZ		 1521 u64 disk_bytenr;
1522 btrfs_item_key_to_cpu(buf, &key, i);
1523 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1524 continue;
1525 fi = btrfs_item_ptr(buf, i,
1526 struct btrfs_file_extent_item);
1527 if (btrfs_file_extent_type(buf, fi) ==
1528 BTRFS_FILE_EXTENT_INLINE)
1529 continue;
1530 disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1531 if (disk_bytenr == 0)
1532 continue;
1533
1534 info->bytenr = disk_bytenr;
1535 info->num_bytes =
1536 btrfs_file_extent_disk_num_bytes(buf, fi);
1537 info->objectid = key.objectid;
1538 info->offset = key.offset;
1539 info++;
1540 }
1541
e4657689 1542 ret = btrfs_add_leaf_ref(root, ref, shared);
5b84e8d6
YZ		 1543 if (ret == -EEXIST && shared) {
1544 struct btrfs_leaf_ref *old;
1545 old = btrfs_lookup_leaf_ref(root, ref->bytenr);
1546 BUG_ON(!old);
1547 btrfs_remove_leaf_ref(root, old);
1548 btrfs_free_leaf_ref(root, old);
1549 ret = btrfs_add_leaf_ref(root, ref, shared);
1550 }
31153d81 1551 WARN_ON(ret);
bcc63abb 1552 btrfs_free_leaf_ref(root, ref);
31153d81
YZ		 1553 }
1554out:
31840ae1
ZY		 1555 return ret;
1556}
1557
1558int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1559 struct extent_buffer *orig_buf, struct extent_buffer *buf,
1560 u32 *nr_extents)
1561{
1562 u64 bytenr;
1563 u64 ref_root;
1564 u64 orig_root;
1565 u64 ref_generation;
1566 u64 orig_generation;
1567 u32 nritems;
1568 u32 nr_file_extents = 0;
1569 struct btrfs_key key;
1570 struct btrfs_file_extent_item *fi;
1571 int i;
1572 int level;
1573 int ret = 0;
1574 int faili = 0;
1575 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
3bb1a1bc 1576 u64, u64, u64, u64, u64, u64, u64, u64);
31840ae1
ZY		 1577
1578 ref_root = btrfs_header_owner(buf);
1579 ref_generation = btrfs_header_generation(buf);
1580 orig_root = btrfs_header_owner(orig_buf);
1581 orig_generation = btrfs_header_generation(orig_buf);
1582
1583 nritems = btrfs_header_nritems(buf);
1584 level = btrfs_header_level(buf);
1585
1586 if (root->ref_cows) {
1587 process_func = __btrfs_inc_extent_ref;
1588 } else {
1589 if (level == 0 &&
1590 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
1591 goto out;
1592 if (level != 0 &&
1593 root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
1594 goto out;
1595 process_func = __btrfs_update_extent_ref;
1596 }
1597
1598 for (i = 0; i < nritems; i++) {
1599 cond_resched();
db94535d 1600 if (level == 0) {
5f39d397
CM		 1601 btrfs_item_key_to_cpu(buf, &key, i);
1602 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
54aa1f4d 1603 continue;
5f39d397 1604 fi = btrfs_item_ptr(buf, i,
54aa1f4d 1605 struct btrfs_file_extent_item);
5f39d397 1606 if (btrfs_file_extent_type(buf, fi) ==
54aa1f4d
CM		 1607 BTRFS_FILE_EXTENT_INLINE)
1608 continue;
31840ae1
ZY		 1609 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1610 if (bytenr == 0)
54aa1f4d 1611 continue;
31840ae1
ZY		 1612
1613 nr_file_extents++;
1614
31840ae1
ZY		 1615 ret = process_func(trans, root, bytenr,
1616 orig_buf->start, buf->start,
1617 orig_root, ref_root,
1618 orig_generation, ref_generation,
3bb1a1bc 1619 key.objectid);
31840ae1
ZY		 1620
1621 if (ret) {
1622 faili = i;
1623 WARN_ON(1);
1624 goto fail;
1625 }
54aa1f4d 1626 } else {
db94535d 1627 bytenr = btrfs_node_blockptr(buf, i);
31840ae1
ZY		 1628 ret = process_func(trans, root, bytenr,
1629 orig_buf->start, buf->start,
1630 orig_root, ref_root,
1631 orig_generation, ref_generation,
3bb1a1bc 1632 level - 1);
31840ae1
ZY		 1633 if (ret) {
1634 faili = i;
1635 WARN_ON(1);
1636 goto fail;
1637 }
54aa1f4d
CM		 1638 }
1639 }
31840ae1
ZY		 1640out:
1641 if (nr_extents) {
1642 if (level == 0)
1643 *nr_extents = nr_file_extents;
1644 else
1645 *nr_extents = nritems;
1646 }
1647 return 0;
1648fail:
1649 WARN_ON(1);
54aa1f4d 1650 return ret;
02217ed2
CM		 1651}
1652
31840ae1
ZY		 1653int btrfs_update_ref(struct btrfs_trans_handle *trans,
1654 struct btrfs_root *root, struct extent_buffer *orig_buf,
1655 struct extent_buffer *buf, int start_slot, int nr)
1656
1657{
1658 u64 bytenr;
1659 u64 ref_root;
1660 u64 orig_root;
1661 u64 ref_generation;
1662 u64 orig_generation;
1663 struct btrfs_key key;
1664 struct btrfs_file_extent_item *fi;
1665 int i;
1666 int ret;
1667 int slot;
1668 int level;
1669
1670 BUG_ON(start_slot < 0);
1671 BUG_ON(start_slot + nr > btrfs_header_nritems(buf));
1672
1673 ref_root = btrfs_header_owner(buf);
1674 ref_generation = btrfs_header_generation(buf);
1675 orig_root = btrfs_header_owner(orig_buf);
1676 orig_generation = btrfs_header_generation(orig_buf);
1677 level = btrfs_header_level(buf);
1678
1679 if (!root->ref_cows) {
1680 if (level == 0 &&
1681 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
1682 return 0;
1683 if (level != 0 &&
1684 root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
1685 return 0;
1686 }
1687
1688 for (i = 0, slot = start_slot; i < nr; i++, slot++) {
1689 cond_resched();
1690 if (level == 0) {
1691 btrfs_item_key_to_cpu(buf, &key, slot);
1692 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1693 continue;
1694 fi = btrfs_item_ptr(buf, slot,
1695 struct btrfs_file_extent_item);
1696 if (btrfs_file_extent_type(buf, fi) ==
1697 BTRFS_FILE_EXTENT_INLINE)
1698 continue;
1699 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1700 if (bytenr == 0)
1701 continue;
31840ae1
ZY		 1702 ret = __btrfs_update_extent_ref(trans, root, bytenr,
1703 orig_buf->start, buf->start,
1704 orig_root, ref_root,
1705 orig_generation, ref_generation,
3bb1a1bc 1706 key.objectid);
31840ae1
ZY		 1707 if (ret)
1708 goto fail;
1709 } else {
1710 bytenr = btrfs_node_blockptr(buf, slot);
31840ae1
ZY		 1711 ret = __btrfs_update_extent_ref(trans, root, bytenr,
1712 orig_buf->start, buf->start,
1713 orig_root, ref_root,
1714 orig_generation, ref_generation,
3bb1a1bc 1715 level - 1);
31840ae1
ZY		 1716 if (ret)
1717 goto fail;
1718 }
1719 }
1720 return 0;
1721fail:
1722 WARN_ON(1);
1723 return -1;
1724}
1725
9078a3e1
CM		 1726static int write_one_cache_group(struct btrfs_trans_handle *trans,
1727 struct btrfs_root *root,
1728 struct btrfs_path *path,
1729 struct btrfs_block_group_cache *cache)
1730{
1731 int ret;
1732 int pending_ret;
1733 struct btrfs_root *extent_root = root->fs_info->extent_root;
5f39d397
CM		 1734 unsigned long bi;
1735 struct extent_buffer *leaf;
9078a3e1 1736
9078a3e1 1737 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
54aa1f4d
CM		 1738 if (ret < 0)
1739 goto fail;
9078a3e1 1740 BUG_ON(ret);
5f39d397
CM		 1741
1742 leaf = path->nodes[0];
1743 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
1744 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
1745 btrfs_mark_buffer_dirty(leaf);
9078a3e1 1746 btrfs_release_path(extent_root, path);
54aa1f4d 1747fail:
87ef2bb4
CM		 1748 finish_current_insert(trans, extent_root, 0);
1749 pending_ret = del_pending_extents(trans, extent_root, 0);
9078a3e1
CM		 1750 if (ret)
1751 return ret;
1752 if (pending_ret)
1753 return pending_ret;
1754 return 0;
1755
1756}
1757
96b5179d
CM		 1758int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
1759 struct btrfs_root *root)
9078a3e1 1760{
0f9dd46c
JB		 1761 struct btrfs_block_group_cache *cache, *entry;
1762 struct rb_node *n;
9078a3e1
CM		 1763 int err = 0;
1764 int werr = 0;
9078a3e1 1765 struct btrfs_path *path;
96b5179d 1766 u64 last = 0;
9078a3e1
CM		 1767
1768 path = btrfs_alloc_path();
1769 if (!path)
1770 return -ENOMEM;
1771
1772 while(1) {
0f9dd46c
JB		 1773 cache = NULL;
1774 spin_lock(&root->fs_info->block_group_cache_lock);
1775 for (n = rb_first(&root->fs_info->block_group_cache_tree);
1776 n; n = rb_next(n)) {
1777 entry = rb_entry(n, struct btrfs_block_group_cache,
1778 cache_node);
1779 if (entry->dirty) {
1780 cache = entry;
1781 break;
1782 }
1783 }
1784 spin_unlock(&root->fs_info->block_group_cache_lock);
54aa1f4d 1785
0f9dd46c 1786 if (!cache)
96b5179d 1787 break;
0f9dd46c 1788
e8569813 1789 cache->dirty = 0;
0f9dd46c
JB		 1790 last += cache->key.offset;
1791
96b5179d
CM		 1792 err = write_one_cache_group(trans, root,
1793 path, cache);
1794 /*
1795 * if we fail to write the cache group, we want
1796 * to keep it marked dirty in hopes that a later
1797 * write will work
1798 */
1799 if (err) {
1800 werr = err;
1801 continue;
9078a3e1
CM		 1802 }
1803 }
1804 btrfs_free_path(path);
1805 return werr;
1806}
1807
593060d7
CM		 1808static int update_space_info(struct btrfs_fs_info *info, u64 flags,
1809 u64 total_bytes, u64 bytes_used,
1810 struct btrfs_space_info **space_info)
1811{
1812 struct btrfs_space_info *found;
1813
1814 found = __find_space_info(info, flags);
1815 if (found) {
25179201 1816 spin_lock(&found->lock);
593060d7
CM		 1817 found->total_bytes += total_bytes;
1818 found->bytes_used += bytes_used;
8f18cf13 1819 found->full = 0;
25179201 1820 spin_unlock(&found->lock);
593060d7
CM		 1821 *space_info = found;
1822 return 0;
1823 }
c146afad 1824 found = kzalloc(sizeof(*found), GFP_NOFS);
593060d7
CM		 1825 if (!found)
1826 return -ENOMEM;
1827
1828 list_add(&found->list, &info->space_info);
0f9dd46c 1829 INIT_LIST_HEAD(&found->block_groups);
80eb234a 1830 init_rwsem(&found->groups_sem);
0f9dd46c 1831 spin_lock_init(&found->lock);
593060d7
CM		 1832 found->flags = flags;
1833 found->total_bytes = total_bytes;
1834 found->bytes_used = bytes_used;
1835 found->bytes_pinned = 0;
e8569813 1836 found->bytes_reserved = 0;
c146afad 1837 found->bytes_readonly = 0;
593060d7 1838 found->full = 0;
0ef3e66b 1839 found->force_alloc = 0;
593060d7
CM		 1840 *space_info = found;
1841 return 0;
1842}
1843
8790d502
CM		 1844static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
1845{
1846 u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
611f0e00 1847 BTRFS_BLOCK_GROUP_RAID1 |
321aecc6 1848 BTRFS_BLOCK_GROUP_RAID10 |
611f0e00 1849 BTRFS_BLOCK_GROUP_DUP);
8790d502
CM		 1850 if (extra_flags) {
1851 if (flags & BTRFS_BLOCK_GROUP_DATA)
1852 fs_info->avail_data_alloc_bits |= extra_flags;
1853 if (flags & BTRFS_BLOCK_GROUP_METADATA)
1854 fs_info->avail_metadata_alloc_bits |= extra_flags;
1855 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
1856 fs_info->avail_system_alloc_bits |= extra_flags;
1857 }
1858}
593060d7 1859
c146afad
YZ		 1860static void set_block_group_readonly(struct btrfs_block_group_cache *cache)
1861{
1862 spin_lock(&cache->space_info->lock);
1863 spin_lock(&cache->lock);
1864 if (!cache->ro) {
1865 cache->space_info->bytes_readonly += cache->key.offset -
1866 btrfs_block_group_used(&cache->item);
1867 cache->ro = 1;
1868 }
1869 spin_unlock(&cache->lock);
1870 spin_unlock(&cache->space_info->lock);
1871}
1872
2b82032c 1873u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
ec44a35c 1874{
2b82032c 1875 u64 num_devices = root->fs_info->fs_devices->rw_devices;
a061fc8d
CM		 1876
1877 if (num_devices == 1)
1878 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
1879 if (num_devices < 4)
1880 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
1881
ec44a35c
CM		 1882 if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
1883 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
a061fc8d 1884 BTRFS_BLOCK_GROUP_RAID10))) {
ec44a35c 1885 flags &= ~BTRFS_BLOCK_GROUP_DUP;
a061fc8d 1886 }
ec44a35c
CM		 1887
1888 if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
a061fc8d 1889 (flags & BTRFS_BLOCK_GROUP_RAID10)) {
ec44a35c 1890 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
a061fc8d 1891 }
ec44a35c
CM		 1892
1893 if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
1894 ((flags & BTRFS_BLOCK_GROUP_RAID1) |
1895 (flags & BTRFS_BLOCK_GROUP_RAID10) |
1896 (flags & BTRFS_BLOCK_GROUP_DUP)))
1897 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
1898 return flags;
1899}
1900
6324fbf3
CM		 1901static int do_chunk_alloc(struct btrfs_trans_handle *trans,
1902 struct btrfs_root *extent_root, u64 alloc_bytes,
0ef3e66b 1903 u64 flags, int force)
6324fbf3
CM		 1904{
1905 struct btrfs_space_info *space_info;
1906 u64 thresh;
c146afad
YZ		 1907 int ret = 0;
1908
1909 mutex_lock(&extent_root->fs_info->chunk_mutex);
6324fbf3 1910
2b82032c 1911 flags = btrfs_reduce_alloc_profile(extent_root, flags);
ec44a35c 1912
6324fbf3 1913 space_info = __find_space_info(extent_root->fs_info, flags);
593060d7
CM		 1914 if (!space_info) {
1915 ret = update_space_info(extent_root->fs_info, flags,
1916 0, 0, &space_info);
1917 BUG_ON(ret);
1918 }
6324fbf3
CM		 1919 BUG_ON(!space_info);
1920
25179201 1921 spin_lock(&space_info->lock);
0ef3e66b
CM		 1922 if (space_info->force_alloc) {
1923 force = 1;
1924 space_info->force_alloc = 0;
1925 }
25179201
JB		 1926 if (space_info->full) {
1927 spin_unlock(&space_info->lock);
925baedd 1928 goto out;
25179201 1929 }
6324fbf3 1930
c146afad
YZ		 1931 thresh = space_info->total_bytes - space_info->bytes_readonly;
1932 thresh = div_factor(thresh, 6);
0ef3e66b 1933 if (!force &&
e8569813 1934 (space_info->bytes_used + space_info->bytes_pinned +
25179201
JB		 1935 space_info->bytes_reserved + alloc_bytes) < thresh) {
1936 spin_unlock(&space_info->lock);
925baedd 1937 goto out;
25179201 1938 }
25179201
JB		 1939 spin_unlock(&space_info->lock);
1940
2b82032c 1941 ret = btrfs_alloc_chunk(trans, extent_root, flags);
25179201 1942 if (ret) {
6324fbf3
CM		 1943printk("space info full %Lu\n", flags);
1944 space_info->full = 1;
6324fbf3 1945 }
a74a4b97 1946out:
c146afad 1947 mutex_unlock(&extent_root->fs_info->chunk_mutex);
0f9dd46c 1948 return ret;
6324fbf3
CM		 1949}
1950
9078a3e1
CM		 1951static int update_block_group(struct btrfs_trans_handle *trans,
1952 struct btrfs_root *root,
db94535d 1953 u64 bytenr, u64 num_bytes, int alloc,
0b86a832 1954 int mark_free)
9078a3e1
CM		 1955{
1956 struct btrfs_block_group_cache *cache;
1957 struct btrfs_fs_info *info = root->fs_info;
db94535d 1958 u64 total = num_bytes;
9078a3e1 1959 u64 old_val;
db94535d 1960 u64 byte_in_group;
3e1ad54f 1961
9078a3e1 1962 while(total) {
db94535d 1963 cache = btrfs_lookup_block_group(info, bytenr);
f3465ca4 1964 if (!cache)
9078a3e1 1965 return -1;
db94535d
CM		 1966 byte_in_group = bytenr - cache->key.objectid;
1967 WARN_ON(byte_in_group > cache->key.offset);
9078a3e1 1968
25179201 1969 spin_lock(&cache->space_info->lock);
c286ac48 1970 spin_lock(&cache->lock);
0f9dd46c 1971 cache->dirty = 1;
9078a3e1 1972 old_val = btrfs_block_group_used(&cache->item);
db94535d 1973 num_bytes = min(total, cache->key.offset - byte_in_group);
cd1bc465 1974 if (alloc) {
db94535d 1975 old_val += num_bytes;
6324fbf3 1976 cache->space_info->bytes_used += num_bytes;
c146afad
YZ		 1977 if (cache->ro) {
1978 cache->space_info->bytes_readonly -= num_bytes;
1979 WARN_ON(1);
1980 }
c286ac48
CM		 1981 btrfs_set_block_group_used(&cache->item, old_val);
1982 spin_unlock(&cache->lock);
25179201 1983 spin_unlock(&cache->space_info->lock);
cd1bc465 1984 } else {
db94535d 1985 old_val -= num_bytes;
6324fbf3 1986 cache->space_info->bytes_used -= num_bytes;
c146afad
YZ		 1987 if (cache->ro)
1988 cache->space_info->bytes_readonly += num_bytes;
c286ac48
CM		 1989 btrfs_set_block_group_used(&cache->item, old_val);
1990 spin_unlock(&cache->lock);
25179201 1991 spin_unlock(&cache->space_info->lock);
f510cfec 1992 if (mark_free) {
0f9dd46c
JB		 1993 int ret;
1994 ret = btrfs_add_free_space(cache, bytenr,
1995 num_bytes);
1996 if (ret)
1997 return -1;
e37c9e69 1998 }
cd1bc465 1999 }
db94535d
CM		 2000 total -= num_bytes;
2001 bytenr += num_bytes;
9078a3e1
CM		 2002 }
2003 return 0;
2004}
6324fbf3 2005
a061fc8d
CM		 2006static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
2007{
0f9dd46c
JB		 2008 struct btrfs_block_group_cache *cache;
2009
2010 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
2011 if (!cache)
a061fc8d 2012 return 0;
0f9dd46c
JB		 2013
2014 return cache->key.objectid;
a061fc8d
CM		 2015}
2016
e02119d5 2017int btrfs_update_pinned_extents(struct btrfs_root *root,
324ae4df
Y		 2018 u64 bytenr, u64 num, int pin)
2019{
2020 u64 len;
2021 struct btrfs_block_group_cache *cache;
2022 struct btrfs_fs_info *fs_info = root->fs_info;
2023
25179201 2024 WARN_ON(!mutex_is_locked(&root->fs_info->pinned_mutex));
324ae4df
Y		 2025 if (pin) {
2026 set_extent_dirty(&fs_info->pinned_extents,
2027 bytenr, bytenr + num - 1, GFP_NOFS);
2028 } else {
2029 clear_extent_dirty(&fs_info->pinned_extents,
2030 bytenr, bytenr + num - 1, GFP_NOFS);
2031 }
2032 while (num > 0) {
2033 cache = btrfs_lookup_block_group(fs_info, bytenr);
e8569813
ZY		 2034 BUG_ON(!cache);
2035 len = min(num, cache->key.offset -
2036 (bytenr - cache->key.objectid));
324ae4df 2037 if (pin) {
25179201 2038 spin_lock(&cache->space_info->lock);
e8569813
ZY		 2039 spin_lock(&cache->lock);
2040 cache->pinned += len;
2041 cache->space_info->bytes_pinned += len;
2042 spin_unlock(&cache->lock);
25179201 2043 spin_unlock(&cache->space_info->lock);
324ae4df
Y		 2044 fs_info->total_pinned += len;
2045 } else {
25179201 2046 spin_lock(&cache->space_info->lock);
e8569813
ZY		 2047 spin_lock(&cache->lock);
2048 cache->pinned -= len;
2049 cache->space_info->bytes_pinned -= len;
2050 spin_unlock(&cache->lock);
25179201 2051 spin_unlock(&cache->space_info->lock);
324ae4df 2052 fs_info->total_pinned -= len;
07103a3c
JB		 2053 if (cache->cached)
2054 btrfs_add_free_space(cache, bytenr, len);
324ae4df
Y		 2055 }
2056 bytenr += len;
2057 num -= len;
2058 }
2059 return 0;
2060}
9078a3e1 2061
e8569813
ZY		 2062static int update_reserved_extents(struct btrfs_root *root,
2063 u64 bytenr, u64 num, int reserve)
2064{
2065 u64 len;
2066 struct btrfs_block_group_cache *cache;
2067 struct btrfs_fs_info *fs_info = root->fs_info;
2068
e8569813
ZY		 2069 while (num > 0) {
2070 cache = btrfs_lookup_block_group(fs_info, bytenr);
2071 BUG_ON(!cache);
2072 len = min(num, cache->key.offset -
2073 (bytenr - cache->key.objectid));
25179201
JB		 2074
2075 spin_lock(&cache->space_info->lock);
2076 spin_lock(&cache->lock);
e8569813 2077 if (reserve) {
e8569813
ZY		 2078 cache->reserved += len;
2079 cache->space_info->bytes_reserved += len;
e8569813 2080 } else {
e8569813
ZY		 2081 cache->reserved -= len;
2082 cache->space_info->bytes_reserved -= len;
e8569813 2083 }
25179201
JB		 2084 spin_unlock(&cache->lock);
2085 spin_unlock(&cache->space_info->lock);
e8569813
ZY		 2086 bytenr += len;
2087 num -= len;
2088 }
2089 return 0;
2090}
2091
d1310b2e 2092int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
ccd467d6 2093{
ccd467d6 2094 u64 last = 0;
1a5bc167
CM		 2095 u64 start;
2096 u64 end;
d1310b2e 2097 struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
ccd467d6 2098 int ret;
ccd467d6 2099
25179201 2100 mutex_lock(&root->fs_info->pinned_mutex);
ccd467d6 2101 while(1) {
1a5bc167
CM		 2102 ret = find_first_extent_bit(pinned_extents, last,
2103 &start, &end, EXTENT_DIRTY);
2104 if (ret)
ccd467d6 2105 break;
1a5bc167
CM		 2106 set_extent_dirty(copy, start, end, GFP_NOFS);
2107 last = end + 1;
ccd467d6 2108 }
25179201 2109 mutex_unlock(&root->fs_info->pinned_mutex);
ccd467d6
CM		 2110 return 0;
2111}
2112
2113int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
2114 struct btrfs_root *root,
d1310b2e 2115 struct extent_io_tree *unpin)
a28ec197 2116{
1a5bc167
CM		 2117 u64 start;
2118 u64 end;
a28ec197 2119 int ret;
a28ec197 2120
25179201 2121 mutex_lock(&root->fs_info->pinned_mutex);
a28ec197 2122 while(1) {
1a5bc167
CM		 2123 ret = find_first_extent_bit(unpin, 0, &start, &end,
2124 EXTENT_DIRTY);
2125 if (ret)
a28ec197 2126 break;
e02119d5 2127 btrfs_update_pinned_extents(root, start, end + 1 - start, 0);
1a5bc167 2128 clear_extent_dirty(unpin, start, end, GFP_NOFS);
c286ac48 2129 if (need_resched()) {
25179201 2130 mutex_unlock(&root->fs_info->pinned_mutex);
c286ac48 2131 cond_resched();
25179201 2132 mutex_lock(&root->fs_info->pinned_mutex);
c286ac48 2133 }
a28ec197 2134 }
25179201 2135 mutex_unlock(&root->fs_info->pinned_mutex);
a28ec197
CM		 2136 return 0;
2137}
2138
98ed5174 2139static int finish_current_insert(struct btrfs_trans_handle *trans,
87ef2bb4 2140 struct btrfs_root *extent_root, int all)
037e6390 2141{
7bb86316
CM		 2142 u64 start;
2143 u64 end;
31840ae1 2144 u64 priv;
25179201 2145 u64 search = 0;
f3465ca4 2146 u64 skipped = 0;
7bb86316
CM		 2147 struct btrfs_fs_info *info = extent_root->fs_info;
2148 struct btrfs_path *path;
f3465ca4
JB		 2149 struct pending_extent_op *extent_op, *tmp;
2150 struct list_head insert_list, update_list;
037e6390 2151 int ret;
f3465ca4 2152 int num_inserts = 0, max_inserts;
037e6390 2153
7bb86316 2154 path = btrfs_alloc_path();
f3465ca4
JB		 2155 INIT_LIST_HEAD(&insert_list);
2156 INIT_LIST_HEAD(&update_list);
037e6390 2157
f3465ca4
JB		 2158 max_inserts = extent_root->leafsize /
2159 (2 * sizeof(struct btrfs_key) + 2 * sizeof(struct btrfs_item) +
2160 sizeof(struct btrfs_extent_ref) +
2161 sizeof(struct btrfs_extent_item));
2162again:
2163 mutex_lock(&info->extent_ins_mutex);
2164 while (1) {
25179201
JB		 2165 ret = find_first_extent_bit(&info->extent_ins, search, &start,
2166 &end, EXTENT_WRITEBACK);
2167 if (ret) {
f3465ca4
JB		 2168 if (skipped && all && !num_inserts) {
2169 skipped = 0;
b4eec2ca 2170 search = 0;
25179201
JB		 2171 continue;
2172 }
f3465ca4 2173 mutex_unlock(&info->extent_ins_mutex);
26b8003f 2174 break;
25179201
JB		 2175 }
2176
2177 ret = try_lock_extent(&info->extent_ins, start, end, GFP_NOFS);
2178 if (!ret) {
f3465ca4 2179 skipped = 1;
87ef2bb4 2180 search = end + 1;
f3465ca4
JB		 2181 if (need_resched()) {
2182 mutex_unlock(&info->extent_ins_mutex);
2183 cond_resched();
2184 mutex_lock(&info->extent_ins_mutex);
2185 }
25179201
JB		 2186 continue;
2187 }
26b8003f 2188
31840ae1
ZY		 2189 ret = get_state_private(&info->extent_ins, start, &priv);
2190 BUG_ON(ret);
f3465ca4 2191 extent_op = (struct pending_extent_op *)(unsigned long) priv;
25179201 2192
31840ae1 2193 if (extent_op->type == PENDING_EXTENT_INSERT) {
f3465ca4
JB		 2194 num_inserts++;
2195 list_add_tail(&extent_op->list, &insert_list);
2196 search = end + 1;
2197 if (num_inserts == max_inserts) {
2198 mutex_unlock(&info->extent_ins_mutex);
2199 break;
2200 }
2201 } else if (extent_op->type == PENDING_BACKREF_UPDATE) {
2202 list_add_tail(&extent_op->list, &update_list);
2203 search = end + 1;
2204 } else {
2205 BUG();
2206 }
2207 }
c286ac48 2208
f3465ca4 2209 /*
b4eec2ca 2210 * process the update list, clear the writeback bit for it, and if
f3465ca4
JB		 2211 * somebody marked this thing for deletion then just unlock it and be
2212 * done, the free_extents will handle it
2213 */
2214 mutex_lock(&info->extent_ins_mutex);
2215 list_for_each_entry_safe(extent_op, tmp, &update_list, list) {
2216 clear_extent_bits(&info->extent_ins, extent_op->bytenr,
2217 extent_op->bytenr + extent_op->num_bytes - 1,
2218 EXTENT_WRITEBACK, GFP_NOFS);
2219 if (extent_op->del) {
2220 list_del_init(&extent_op->list);
2221 unlock_extent(&info->extent_ins, extent_op->bytenr,
2222 extent_op->bytenr + extent_op->num_bytes
2223 - 1, GFP_NOFS);
2224 kfree(extent_op);
2225 }
2226 }
2227 mutex_unlock(&info->extent_ins_mutex);
c286ac48 2228
f3465ca4
JB		 2229 /*
2230 * still have things left on the update list, go ahead an update
2231 * everything
2232 */
2233 if (!list_empty(&update_list)) {
2234 ret = update_backrefs(trans, extent_root, path, &update_list);
2235 BUG_ON(ret);
2236 }
c286ac48 2237
f3465ca4
JB		 2238 /*
2239 * if no inserts need to be done, but we skipped some extents and we
2240 * need to make sure everything is cleaned then reset everything and
2241 * go back to the beginning
2242 */
2243 if (!num_inserts && all && skipped) {
2244 search = 0;
2245 skipped = 0;
2246 INIT_LIST_HEAD(&update_list);
2247 INIT_LIST_HEAD(&insert_list);
2248 goto again;
2249 } else if (!num_inserts) {
2250 goto out;
2251 }
31840ae1 2252
f3465ca4
JB		 2253 /*
2254 * process the insert extents list. Again if we are deleting this
2255 * extent, then just unlock it, pin down the bytes if need be, and be
2256 * done with it. Saves us from having to actually insert the extent
2257 * into the tree and then subsequently come along and delete it
2258 */
2259 mutex_lock(&info->extent_ins_mutex);
2260 list_for_each_entry_safe(extent_op, tmp, &insert_list, list) {
2261 clear_extent_bits(&info->extent_ins, extent_op->bytenr,
2262 extent_op->bytenr + extent_op->num_bytes - 1,
2263 EXTENT_WRITEBACK, GFP_NOFS);
2264 if (extent_op->del) {
2265 list_del_init(&extent_op->list);
2266 unlock_extent(&info->extent_ins, extent_op->bytenr,
2267 extent_op->bytenr + extent_op->num_bytes
2268 - 1, GFP_NOFS);
2269
2270 mutex_lock(&extent_root->fs_info->pinned_mutex);
2271 ret = pin_down_bytes(trans, extent_root,
2272 extent_op->bytenr,
2273 extent_op->num_bytes, 0);
2274 mutex_unlock(&extent_root->fs_info->pinned_mutex);
2275
2276 ret = update_block_group(trans, extent_root,
2277 extent_op->bytenr,
2278 extent_op->num_bytes,
2279 0, ret > 0);
2280 BUG_ON(ret);
2281 kfree(extent_op);
2282 num_inserts--;
d8d5f3e1 2283 }
f3465ca4
JB		 2284 }
2285 mutex_unlock(&info->extent_ins_mutex);
31840ae1 2286
f3465ca4
JB		 2287 ret = insert_extents(trans, extent_root, path, &insert_list,
2288 num_inserts);
2289 BUG_ON(ret);
2290
2291 /*
2292 * if we broke out of the loop in order to insert stuff because we hit
2293 * the maximum number of inserts at a time we can handle, then loop
2294 * back and pick up where we left off
2295 */
2296 if (num_inserts == max_inserts) {
2297 INIT_LIST_HEAD(&insert_list);
2298 INIT_LIST_HEAD(&update_list);
2299 num_inserts = 0;
2300 goto again;
2301 }
2302
2303 /*
2304 * again, if we need to make absolutely sure there are no more pending
2305 * extent operations left and we know that we skipped some, go back to
2306 * the beginning and do it all again
2307 */
2308 if (all && skipped) {
2309 INIT_LIST_HEAD(&insert_list);
2310 INIT_LIST_HEAD(&update_list);
2311 search = 0;
2312 skipped = 0;
2313 num_inserts = 0;
2314 goto again;
037e6390 2315 }
f3465ca4 2316out:
7bb86316 2317 btrfs_free_path(path);
037e6390
CM		 2318 return 0;
2319}
2320
31840ae1
ZY		 2321static int pin_down_bytes(struct btrfs_trans_handle *trans,
2322 struct btrfs_root *root,
2323 u64 bytenr, u64 num_bytes, int is_data)
e20d96d6 2324{
1a5bc167 2325 int err = 0;
31840ae1 2326 struct extent_buffer *buf;
8ef97622 2327
31840ae1
ZY		 2328 if (is_data)
2329 goto pinit;
2330
2331 buf = btrfs_find_tree_block(root, bytenr, num_bytes);
2332 if (!buf)
2333 goto pinit;
2334
2335 /* we can reuse a block if it hasn't been written
2336 * and it is from this transaction. We can't
2337 * reuse anything from the tree log root because
2338 * it has tiny sub-transactions.
2339 */
2340 if (btrfs_buffer_uptodate(buf, 0) &&
2341 btrfs_try_tree_lock(buf)) {
2342 u64 header_owner = btrfs_header_owner(buf);
2343 u64 header_transid = btrfs_header_generation(buf);
2344 if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
1a40e23b 2345 header_owner != BTRFS_TREE_RELOC_OBJECTID &&
31840ae1
ZY		 2346 header_transid == trans->transid &&
2347 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
2348 clean_tree_block(NULL, root, buf);
2349 btrfs_tree_unlock(buf);
5f39d397 2350 free_extent_buffer(buf);
31840ae1 2351 return 1;
8ef97622 2352 }
31840ae1 2353 btrfs_tree_unlock(buf);
f4b9aa8d 2354 }
31840ae1
ZY		 2355 free_extent_buffer(buf);
2356pinit:
2357 btrfs_update_pinned_extents(root, bytenr, num_bytes, 1);
2358
be744175 2359 BUG_ON(err < 0);
e20d96d6
CM		 2360 return 0;
2361}
2362
fec577fb 2363/*
a28ec197 2364 * remove an extent from the root, returns 0 on success
fec577fb 2365 */
31840ae1
ZY		 2366static int __free_extent(struct btrfs_trans_handle *trans,
2367 struct btrfs_root *root,
2368 u64 bytenr, u64 num_bytes, u64 parent,
7bb86316 2369 u64 root_objectid, u64 ref_generation,
3bb1a1bc 2370 u64 owner_objectid, int pin, int mark_free)
a28ec197 2371{
5caf2a00 2372 struct btrfs_path *path;
e2fa7227 2373 struct btrfs_key key;
1261ec42
CM		 2374 struct btrfs_fs_info *info = root->fs_info;
2375 struct btrfs_root *extent_root = info->extent_root;
5f39d397 2376 struct extent_buffer *leaf;
a28ec197 2377 int ret;
952fccac
CM		 2378 int extent_slot = 0;
2379 int found_extent = 0;
2380 int num_to_del = 1;
234b63a0 2381 struct btrfs_extent_item *ei;
cf27e1ee 2382 u32 refs;
037e6390 2383
db94535d 2384 key.objectid = bytenr;
62e2749e 2385 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
db94535d 2386 key.offset = num_bytes;
5caf2a00 2387 path = btrfs_alloc_path();
54aa1f4d
CM		 2388 if (!path)
2389 return -ENOMEM;
5f26f772 2390
3c12ac72 2391 path->reada = 1;
3bb1a1bc
YZ		 2392 ret = lookup_extent_backref(trans, extent_root, path,
2393 bytenr, parent, root_objectid,
2394 ref_generation, owner_objectid, 1);
7bb86316 2395 if (ret == 0) {
952fccac
CM		 2396 struct btrfs_key found_key;
2397 extent_slot = path->slots[0];
2398 while(extent_slot > 0) {
2399 extent_slot--;
2400 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2401 extent_slot);
2402 if (found_key.objectid != bytenr)
2403 break;
2404 if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
2405 found_key.offset == num_bytes) {
2406 found_extent = 1;
2407 break;
2408 }
2409 if (path->slots[0] - extent_slot > 5)
2410 break;
2411 }
31840ae1
ZY		 2412 if (!found_extent) {
2413 ret = remove_extent_backref(trans, extent_root, path);
2414 BUG_ON(ret);
2415 btrfs_release_path(extent_root, path);
2416 ret = btrfs_search_slot(trans, extent_root,
2417 &key, path, -1, 1);
f3465ca4
JB		 2418 if (ret) {
2419 printk(KERN_ERR "umm, got %d back from search"
2420 ", was looking for %Lu\n", ret,
2421 bytenr);
2422 btrfs_print_leaf(extent_root, path->nodes[0]);
2423 }
31840ae1
ZY		 2424 BUG_ON(ret);
2425 extent_slot = path->slots[0];
2426 }
7bb86316
CM		 2427 } else {
2428 btrfs_print_leaf(extent_root, path->nodes[0]);
2429 WARN_ON(1);
2430 printk("Unable to find ref byte nr %Lu root %Lu "
3bb1a1bc
YZ		 2431 "gen %Lu owner %Lu\n", bytenr,
2432 root_objectid, ref_generation, owner_objectid);
7bb86316 2433 }
5f39d397
CM		 2434
2435 leaf = path->nodes[0];
952fccac 2436 ei = btrfs_item_ptr(leaf, extent_slot,
123abc88 2437 struct btrfs_extent_item);
5f39d397
CM		 2438 refs = btrfs_extent_refs(leaf, ei);
2439 BUG_ON(refs == 0);
2440 refs -= 1;
2441 btrfs_set_extent_refs(leaf, ei, refs);
952fccac 2442
5f39d397
CM		 2443 btrfs_mark_buffer_dirty(leaf);
2444
952fccac 2445 if (refs == 0 && found_extent && path->slots[0] == extent_slot + 1) {
31840ae1
ZY		 2446 struct btrfs_extent_ref *ref;
2447 ref = btrfs_item_ptr(leaf, path->slots[0],
2448 struct btrfs_extent_ref);
2449 BUG_ON(btrfs_ref_num_refs(leaf, ref) != 1);
952fccac
CM		 2450 /* if the back ref and the extent are next to each other
2451 * they get deleted below in one shot
2452 */
2453 path->slots[0] = extent_slot;
2454 num_to_del = 2;
2455 } else if (found_extent) {
2456 /* otherwise delete the extent back ref */
31840ae1 2457 ret = remove_extent_backref(trans, extent_root, path);
952fccac
CM		 2458 BUG_ON(ret);
2459 /* if refs are 0, we need to setup the path for deletion */
2460 if (refs == 0) {
2461 btrfs_release_path(extent_root, path);
2462 ret = btrfs_search_slot(trans, extent_root, &key, path,
2463 -1, 1);
952fccac
CM		 2464 BUG_ON(ret);
2465 }
2466 }
2467
cf27e1ee 2468 if (refs == 0) {
db94535d
CM		 2469 u64 super_used;
2470 u64 root_used;
21af804c
DW		 2471#ifdef BIO_RW_DISCARD
2472 u64 map_length = num_bytes;
2473 struct btrfs_multi_bio *multi = NULL;
2474#endif
78fae27e
CM		 2475
2476 if (pin) {
25179201 2477 mutex_lock(&root->fs_info->pinned_mutex);
31840ae1
ZY		 2478 ret = pin_down_bytes(trans, root, bytenr, num_bytes,
2479 owner_objectid >= BTRFS_FIRST_FREE_OBJECTID);
25179201 2480 mutex_unlock(&root->fs_info->pinned_mutex);
c549228f
Y		 2481 if (ret > 0)
2482 mark_free = 1;
2483 BUG_ON(ret < 0);
78fae27e
CM		 2484 }
2485
58176a96 2486 /* block accounting for super block */
a2135011 2487 spin_lock_irq(&info->delalloc_lock);
db94535d
CM		 2488 super_used = btrfs_super_bytes_used(&info->super_copy);
2489 btrfs_set_super_bytes_used(&info->super_copy,
2490 super_used - num_bytes);
a2135011 2491 spin_unlock_irq(&info->delalloc_lock);
58176a96
JB		 2492
2493 /* block accounting for root item */
db94535d 2494 root_used = btrfs_root_used(&root->root_item);
5f39d397 2495 btrfs_set_root_used(&root->root_item,
db94535d 2496 root_used - num_bytes);
952fccac
CM		 2497 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
2498 num_to_del);
31840ae1 2499 BUG_ON(ret);
25179201 2500 btrfs_release_path(extent_root, path);
db94535d 2501 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
0b86a832 2502 mark_free);
9078a3e1 2503 BUG_ON(ret);
21af804c
DW		 2504
2505#ifdef BIO_RW_DISCARD
2506 /* Tell the block device(s) that the sectors can be discarded */
2507 ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
2508 bytenr, &map_length, &multi, 0);
2509 if (!ret) {
2510 struct btrfs_bio_stripe *stripe = multi->stripes;
2511 int i;
2512
2513 if (map_length > num_bytes)
2514 map_length = num_bytes;
2515
2516 for (i = 0; i < multi->num_stripes; i++, stripe++) {
15916de8
CM		 2517 btrfs_issue_discard(stripe->dev->bdev,
2518 stripe->physical,
2519 map_length);
21af804c
DW		 2520 }
2521 kfree(multi);
2522 }
2523#endif
a28ec197 2524 }
5caf2a00 2525 btrfs_free_path(path);
87ef2bb4 2526 finish_current_insert(trans, extent_root, 0);
a28ec197
CM		 2527 return ret;
2528}
2529
a28ec197
CM		 2530/*
2531 * find all the blocks marked as pending in the radix tree and remove
2532 * them from the extent map
2533 */
e089f05c 2534static int del_pending_extents(struct btrfs_trans_handle *trans, struct
87ef2bb4 2535 btrfs_root *extent_root, int all)
a28ec197
CM		 2536{
2537 int ret;
e20d96d6 2538 int err = 0;
1a5bc167
CM		 2539 u64 start;
2540 u64 end;
31840ae1 2541 u64 priv;
25179201 2542 u64 search = 0;
f3465ca4 2543 int nr = 0, skipped = 0;
d1310b2e 2544 struct extent_io_tree *pending_del;
31840ae1
ZY		 2545 struct extent_io_tree *extent_ins;
2546 struct pending_extent_op *extent_op;
25179201 2547 struct btrfs_fs_info *info = extent_root->fs_info;
f3465ca4 2548 struct list_head delete_list;
8ef97622 2549
f3465ca4 2550 INIT_LIST_HEAD(&delete_list);
31840ae1 2551 extent_ins = &extent_root->fs_info->extent_ins;
1a5bc167 2552 pending_del = &extent_root->fs_info->pending_del;
a28ec197 2553
f3465ca4
JB		 2554again:
2555 mutex_lock(&info->extent_ins_mutex);
a28ec197 2556 while(1) {
25179201
JB		 2557 ret = find_first_extent_bit(pending_del, search, &start, &end,
2558 EXTENT_WRITEBACK);
2559 if (ret) {
f3465ca4 2560 if (all && skipped && !nr) {
25179201
JB		 2561 search = 0;
2562 continue;
2563 }
f3465ca4 2564 mutex_unlock(&info->extent_ins_mutex);
a28ec197 2565 break;
25179201
JB		 2566 }
2567
2568 ret = try_lock_extent(extent_ins, start, end, GFP_NOFS);
2569 if (!ret) {
2570 search = end+1;
f3465ca4
JB		 2571 skipped = 1;
2572
2573 if (need_resched()) {
2574 mutex_unlock(&info->extent_ins_mutex);
2575 cond_resched();
2576 mutex_lock(&info->extent_ins_mutex);
2577 }
2578
25179201
JB		 2579 continue;
2580 }
2581 BUG_ON(ret < 0);
31840ae1
ZY		 2582
2583 ret = get_state_private(pending_del, start, &priv);
2584 BUG_ON(ret);
2585 extent_op = (struct pending_extent_op *)(unsigned long)priv;
2586
25179201 2587 clear_extent_bits(pending_del, start, end, EXTENT_WRITEBACK,
1a5bc167 2588 GFP_NOFS);
31840ae1 2589 if (!test_range_bit(extent_ins, start, end,
25179201 2590 EXTENT_WRITEBACK, 0)) {
f3465ca4
JB		 2591 list_add_tail(&extent_op->list, &delete_list);
2592 nr++;
c286ac48 2593 } else {
31840ae1 2594 kfree(extent_op);
25179201
JB		 2595
2596 ret = get_state_private(&info->extent_ins, start,
2597 &priv);
31840ae1
ZY		 2598 BUG_ON(ret);
2599 extent_op = (struct pending_extent_op *)
25179201
JB		 2600 (unsigned long)priv;
2601
2602 clear_extent_bits(&info->extent_ins, start, end,
2603 EXTENT_WRITEBACK, GFP_NOFS);
31840ae1 2604
f3465ca4
JB		 2605 if (extent_op->type == PENDING_BACKREF_UPDATE) {
2606 list_add_tail(&extent_op->list, &delete_list);
2607 search = end + 1;
2608 nr++;
2609 continue;
2610 }
31840ae1 2611
25179201
JB		 2612 mutex_lock(&extent_root->fs_info->pinned_mutex);
2613 ret = pin_down_bytes(trans, extent_root, start,
2614 end + 1 - start, 0);
2615 mutex_unlock(&extent_root->fs_info->pinned_mutex);
2616
31840ae1 2617 ret = update_block_group(trans, extent_root, start,
25179201
JB		 2618 end + 1 - start, 0, ret > 0);
2619
f3465ca4 2620 unlock_extent(extent_ins, start, end, GFP_NOFS);
31840ae1
ZY		 2621 BUG_ON(ret);
2622 kfree(extent_op);
c286ac48 2623 }
1a5bc167
CM		 2624 if (ret)
2625 err = ret;
c286ac48 2626
f3465ca4
JB		 2627 search = end + 1;
2628
2629 if (need_resched()) {
2630 mutex_unlock(&info->extent_ins_mutex);
2631 cond_resched();
2632 mutex_lock(&info->extent_ins_mutex);
2633 }
2634 }
2635
2636 if (nr) {
2637 ret = free_extents(trans, extent_root, &delete_list);
2638 BUG_ON(ret);
2639 }
2640
2641 if (all && skipped) {
2642 INIT_LIST_HEAD(&delete_list);
2643 search = 0;
2644 nr = 0;
2645 goto again;
fec577fb 2646 }
f3465ca4 2647
e20d96d6 2648 return err;
fec577fb
CM		 2649}
2650
2651/*
2652 * remove an extent from the root, returns 0 on success
2653 */
925baedd 2654static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
31840ae1
ZY		 2655 struct btrfs_root *root,
2656 u64 bytenr, u64 num_bytes, u64 parent,
2657 u64 root_objectid, u64 ref_generation,
3bb1a1bc 2658 u64 owner_objectid, int pin)
fec577fb 2659{
9f5fae2f 2660 struct btrfs_root *extent_root = root->fs_info->extent_root;
fec577fb
CM		 2661 int pending_ret;
2662 int ret;
a28ec197 2663
db94535d 2664 WARN_ON(num_bytes < root->sectorsize);
fec577fb 2665 if (root == extent_root) {
f3465ca4
JB		 2666 struct pending_extent_op *extent_op = NULL;
2667
2668 mutex_lock(&root->fs_info->extent_ins_mutex);
2669 if (test_range_bit(&root->fs_info->extent_ins, bytenr,
2670 bytenr + num_bytes - 1, EXTENT_WRITEBACK, 0)) {
2671 u64 priv;
2672 ret = get_state_private(&root->fs_info->extent_ins,
2673 bytenr, &priv);
2674 BUG_ON(ret);
2675 extent_op = (struct pending_extent_op *)
2676 (unsigned long)priv;
2677
2678 extent_op->del = 1;
2679 if (extent_op->type == PENDING_EXTENT_INSERT) {
2680 mutex_unlock(&root->fs_info->extent_ins_mutex);
2681 return 0;
2682 }
2683 }
2684
2685 if (extent_op) {
2686 ref_generation = extent_op->orig_generation;
2687 parent = extent_op->orig_parent;
2688 }
31840ae1
ZY		 2689
2690 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2691 BUG_ON(!extent_op);
2692
2693 extent_op->type = PENDING_EXTENT_DELETE;
2694 extent_op->bytenr = bytenr;
2695 extent_op->num_bytes = num_bytes;
2696 extent_op->parent = parent;
2697 extent_op->orig_parent = parent;
2698 extent_op->generation = ref_generation;
2699 extent_op->orig_generation = ref_generation;
2700 extent_op->level = (int)owner_objectid;
f3465ca4
JB		 2701 INIT_LIST_HEAD(&extent_op->list);
2702 extent_op->del = 0;
31840ae1
ZY		 2703
2704 set_extent_bits(&root->fs_info->pending_del,
2705 bytenr, bytenr + num_bytes - 1,
25179201 2706 EXTENT_WRITEBACK, GFP_NOFS);
31840ae1
ZY		 2707 set_state_private(&root->fs_info->pending_del,
2708 bytenr, (unsigned long)extent_op);
25179201 2709 mutex_unlock(&root->fs_info->extent_ins_mutex);
fec577fb
CM		 2710 return 0;
2711 }
4bef0848 2712 /* if metadata always pin */
d00aff00
CM		 2713 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
2714 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
0f9dd46c
JB		 2715 struct btrfs_block_group_cache *cache;
2716
d00aff00 2717 /* btrfs_free_reserved_extent */
0f9dd46c
JB		 2718 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
2719 BUG_ON(!cache);
2720 btrfs_add_free_space(cache, bytenr, num_bytes);
e8569813 2721 update_reserved_extents(root, bytenr, num_bytes, 0);
d00aff00
CM		 2722 return 0;
2723 }
4bef0848 2724 pin = 1;
d00aff00 2725 }
4bef0848
CM		 2726
2727 /* if data pin when any transaction has committed this */
2728 if (ref_generation != trans->transid)
2729 pin = 1;
2730
31840ae1 2731 ret = __free_extent(trans, root, bytenr, num_bytes, parent,
3bb1a1bc
YZ		 2732 root_objectid, ref_generation,
2733 owner_objectid, pin, pin == 0);
ee6e6504 2734
87ef2bb4
CM		 2735 finish_current_insert(trans, root->fs_info->extent_root, 0);
2736 pending_ret = del_pending_extents(trans, root->fs_info->extent_root, 0);
fec577fb
CM		 2737 return ret ? ret : pending_ret;
2738}
2739
925baedd 2740int btrfs_free_extent(struct btrfs_trans_handle *trans,
31840ae1
ZY		 2741 struct btrfs_root *root,
2742 u64 bytenr, u64 num_bytes, u64 parent,
2743 u64 root_objectid, u64 ref_generation,
3bb1a1bc 2744 u64 owner_objectid, int pin)
925baedd
CM		 2745{
2746 int ret;
2747
31840ae1 2748 ret = __btrfs_free_extent(trans, root, bytenr, num_bytes, parent,
925baedd 2749 root_objectid, ref_generation,
3bb1a1bc 2750 owner_objectid, pin);
925baedd
CM		 2751 return ret;
2752}
2753
87ee04eb
CM		 2754static u64 stripe_align(struct btrfs_root *root, u64 val)
2755{
2756 u64 mask = ((u64)root->stripesize - 1);
2757 u64 ret = (val + mask) & ~mask;
2758 return ret;
2759}
2760
fec577fb
CM		 2761/*
2762 * walks the btree of allocated extents and find a hole of a given size.
2763 * The key ins is changed to record the hole:
2764 * ins->objectid == block start
62e2749e 2765 * ins->flags = BTRFS_EXTENT_ITEM_KEY
fec577fb
CM		 2766 * ins->offset == number of blocks
2767 * Any available blocks before search_start are skipped.
2768 */
98ed5174
CM		 2769static int noinline find_free_extent(struct btrfs_trans_handle *trans,
2770 struct btrfs_root *orig_root,
2771 u64 num_bytes, u64 empty_size,
2772 u64 search_start, u64 search_end,
2773 u64 hint_byte, struct btrfs_key *ins,
2774 u64 exclude_start, u64 exclude_nr,
2775 int data)
fec577fb 2776{
80eb234a 2777 int ret = 0;
9f5fae2f 2778 struct btrfs_root * root = orig_root->fs_info->extent_root;
db94535d 2779 u64 total_needed = num_bytes;
239b14b3 2780 u64 *last_ptr = NULL;
4366211c 2781 u64 last_wanted = 0;
80eb234a 2782 struct btrfs_block_group_cache *block_group = NULL;
0ef3e66b 2783 int chunk_alloc_done = 0;
239b14b3 2784 int empty_cluster = 2 * 1024 * 1024;
0ef3e66b 2785 int allowed_chunk_alloc = 0;
80eb234a
JB		 2786 struct list_head *head = NULL, *cur = NULL;
2787 int loop = 0;
f5a31e16 2788 int extra_loop = 0;
80eb234a 2789 struct btrfs_space_info *space_info;
fec577fb 2790
db94535d 2791 WARN_ON(num_bytes < root->sectorsize);
b1a4d965 2792 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
80eb234a
JB		 2793 ins->objectid = 0;
2794 ins->offset = 0;
b1a4d965 2795
0ef3e66b
CM		 2796 if (orig_root->ref_cows || empty_size)
2797 allowed_chunk_alloc = 1;
2798
239b14b3
CM		 2799 if (data & BTRFS_BLOCK_GROUP_METADATA) {
2800 last_ptr = &root->fs_info->last_alloc;
4366211c 2801 empty_cluster = 64 * 1024;
239b14b3
CM		 2802 }
2803
0f9dd46c 2804 if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD))
239b14b3 2805 last_ptr = &root->fs_info->last_data_alloc;
0f9dd46c 2806
239b14b3 2807 if (last_ptr) {
4366211c 2808 if (*last_ptr) {
239b14b3 2809 hint_byte = *last_ptr;
4366211c
CM		 2810 last_wanted = *last_ptr;
2811 } else
239b14b3 2812 empty_size += empty_cluster;
4366211c
CM		 2813 } else {
2814 empty_cluster = 0;
239b14b3 2815 }
a061fc8d 2816 search_start = max(search_start, first_logical_byte(root, 0));
239b14b3 2817 search_start = max(search_start, hint_byte);
0b86a832 2818
42e70e7a 2819 if (last_wanted && search_start != last_wanted) {
4366211c 2820 last_wanted = 0;
42e70e7a
CM		 2821 empty_size += empty_cluster;
2822 }
4366211c 2823
42e70e7a 2824 total_needed += empty_size;
80eb234a 2825 block_group = btrfs_lookup_block_group(root->fs_info, search_start);
d899e052
YZ		 2826 if (!block_group)
2827 block_group = btrfs_lookup_first_block_group(root->fs_info,
2828 search_start);
80eb234a 2829 space_info = __find_space_info(root->fs_info, data);
0f9dd46c 2830
80eb234a
JB		 2831 down_read(&space_info->groups_sem);
2832 while (1) {
2833 struct btrfs_free_space *free_space;
0f9dd46c 2834 /*
80eb234a
JB		 2835 * the only way this happens if our hint points to a block
2836 * group thats not of the proper type, while looping this
2837 * should never happen
0f9dd46c 2838 */
8a1413a2
CM		 2839 if (empty_size)
2840 extra_loop = 1;
2841
42e70e7a
CM		 2842 if (!block_group)
2843 goto new_group_no_lock;
2844
25179201 2845 mutex_lock(&block_group->alloc_mutex);
80eb234a 2846 if (unlikely(!block_group_bits(block_group, data)))
0f9dd46c 2847 goto new_group;
0f9dd46c 2848
80eb234a 2849 ret = cache_block_group(root, block_group);
25179201
JB		 2850 if (ret) {
2851 mutex_unlock(&block_group->alloc_mutex);
80eb234a 2852 break;
25179201 2853 }
0f9dd46c 2854
80eb234a
JB		 2855 if (block_group->ro)
2856 goto new_group;
0f9dd46c 2857
80eb234a
JB		 2858 free_space = btrfs_find_free_space(block_group, search_start,
2859 total_needed);
2860 if (free_space) {
2861 u64 start = block_group->key.objectid;
2862 u64 end = block_group->key.objectid +
0f9dd46c 2863 block_group->key.offset;
239b14b3 2864
80eb234a 2865 search_start = stripe_align(root, free_space->offset);
0f9dd46c 2866
80eb234a
JB		 2867 /* move on to the next group */
2868 if (search_start + num_bytes >= search_end)
2869 goto new_group;
e37c9e69 2870
80eb234a
JB		 2871 /* move on to the next group */
2872 if (search_start + num_bytes > end)
2873 goto new_group;
2874
4366211c 2875 if (last_wanted && search_start != last_wanted) {
3b7885bf 2876 total_needed += empty_cluster;
8a1413a2 2877 empty_size += empty_cluster;
4366211c 2878 last_wanted = 0;
3b7885bf
CM		 2879 /*
2880 * if search_start is still in this block group
2881 * then we just re-search this block group
2882 */
2883 if (search_start >= start &&
2884 search_start < end) {
2885 mutex_unlock(&block_group->alloc_mutex);
2886 continue;
2887 }
2888
2889 /* else we go to the next block group */
2890 goto new_group;
2891 }
2892
80eb234a
JB		 2893 if (exclude_nr > 0 &&
2894 (search_start + num_bytes > exclude_start &&
2895 search_start < exclude_start + exclude_nr)) {
2896 search_start = exclude_start + exclude_nr;
2897 /*
2898 * if search_start is still in this block group
2899 * then we just re-search this block group
2900 */
2901 if (search_start >= start &&
25179201
JB		 2902 search_start < end) {
2903 mutex_unlock(&block_group->alloc_mutex);
4366211c 2904 last_wanted = 0;
80eb234a 2905 continue;
25179201 2906 }
80eb234a
JB		 2907
2908 /* else we go to the next block group */
2909 goto new_group;
2910 }
2911
2912 ins->objectid = search_start;
2913 ins->offset = num_bytes;
25179201
JB		 2914
2915 btrfs_remove_free_space_lock(block_group, search_start,
2916 num_bytes);
80eb234a 2917 /* we are all good, lets return */
25179201 2918 mutex_unlock(&block_group->alloc_mutex);
80eb234a 2919 break;
0f9dd46c 2920 }
80eb234a 2921new_group:
42e70e7a
CM		 2922 mutex_unlock(&block_group->alloc_mutex);
2923new_group_no_lock:
f5a31e16
CM		 2924 /* don't try to compare new allocations against the
2925 * last allocation any more
2926 */
4366211c 2927 last_wanted = 0;
f5a31e16 2928
80eb234a
JB		 2929 /*
2930 * Here's how this works.
2931 * loop == 0: we were searching a block group via a hint
2932 * and didn't find anything, so we start at
2933 * the head of the block groups and keep searching
2934 * loop == 1: we're searching through all of the block groups
2935 * if we hit the head again we have searched
2936 * all of the block groups for this space and we
2937 * need to try and allocate, if we cant error out.
2938 * loop == 2: we allocated more space and are looping through
2939 * all of the block groups again.
2940 */
2941 if (loop == 0) {
2942 head = &space_info->block_groups;
2943 cur = head->next;
80eb234a
JB		 2944 loop++;
2945 } else if (loop == 1 && cur == head) {
f5a31e16
CM		 2946 int keep_going;
2947
2948 /* at this point we give up on the empty_size
2949 * allocations and just try to allocate the min
2950 * space.
2951 *
2952 * The extra_loop field was set if an empty_size
2953 * allocation was attempted above, and if this
2954 * is try we need to try the loop again without
2955 * the additional empty_size.
2956 */
5b7c3fcc
CM		 2957 total_needed -= empty_size;
2958 empty_size = 0;
f5a31e16
CM		 2959 keep_going = extra_loop;
2960 loop++;
42e70e7a 2961
80eb234a
JB		 2962 if (allowed_chunk_alloc && !chunk_alloc_done) {
2963 up_read(&space_info->groups_sem);
2964 ret = do_chunk_alloc(trans, root, num_bytes +
2965 2 * 1024 * 1024, data, 1);
80eb234a 2966 down_read(&space_info->groups_sem);
2ed6d664
CM		 2967 if (ret < 0)
2968 goto loop_check;
80eb234a 2969 head = &space_info->block_groups;
f5a31e16
CM		 2970 /*
2971 * we've allocated a new chunk, keep
2972 * trying
2973 */
2974 keep_going = 1;
80eb234a
JB		 2975 chunk_alloc_done = 1;
2976 } else if (!allowed_chunk_alloc) {
2977 space_info->force_alloc = 1;
f5a31e16 2978 }
2ed6d664 2979loop_check:
f5a31e16
CM		 2980 if (keep_going) {
2981 cur = head->next;
2982 extra_loop = 0;
80eb234a
JB		 2983 } else {
2984 break;
2985 }
2986 } else if (cur == head) {
2987 break;
0f9dd46c 2988 }
0b86a832 2989
80eb234a
JB		 2990 block_group = list_entry(cur, struct btrfs_block_group_cache,
2991 list);
2992 search_start = block_group->key.objectid;
2993 cur = cur->next;
f2654de4 2994 }
0b86a832 2995
80eb234a
JB		 2996 /* we found what we needed */
2997 if (ins->objectid) {
2998 if (!(data & BTRFS_BLOCK_GROUP_DATA))
2999 trans->block_group = block_group;
0f9dd46c 3000
80eb234a
JB		 3001 if (last_ptr)
3002 *last_ptr = ins->objectid + ins->offset;
3003 ret = 0;
3004 } else if (!ret) {
4ce4cb52
JB		 3005 printk(KERN_ERR "we were searching for %Lu bytes, num_bytes %Lu,"
3006 " loop %d, allowed_alloc %d\n", total_needed, num_bytes,
3007 loop, allowed_chunk_alloc);
80eb234a 3008 ret = -ENOSPC;
be744175 3009 }
be744175 3010
80eb234a 3011 up_read(&space_info->groups_sem);
0f70abe2 3012 return ret;
fec577fb 3013}
ec44a35c 3014
0f9dd46c
JB		 3015static void dump_space_info(struct btrfs_space_info *info, u64 bytes)
3016{
3017 struct btrfs_block_group_cache *cache;
3018 struct list_head *l;
3019
3020 printk(KERN_INFO "space_info has %Lu free, is %sfull\n",
e8569813
ZY		 3021 info->total_bytes - info->bytes_used - info->bytes_pinned -
3022 info->bytes_reserved, (info->full) ? "" : "not ");
0f9dd46c 3023
80eb234a 3024 down_read(&info->groups_sem);
0f9dd46c
JB		 3025 list_for_each(l, &info->block_groups) {
3026 cache = list_entry(l, struct btrfs_block_group_cache, list);
3027 spin_lock(&cache->lock);
3028 printk(KERN_INFO "block group %Lu has %Lu bytes, %Lu used "
e8569813 3029 "%Lu pinned %Lu reserved\n",
0f9dd46c 3030 cache->key.objectid, cache->key.offset,
e8569813
ZY		 3031 btrfs_block_group_used(&cache->item),
3032 cache->pinned, cache->reserved);
0f9dd46c
JB		 3033 btrfs_dump_free_space(cache, bytes);
3034 spin_unlock(&cache->lock);
3035 }
80eb234a 3036 up_read(&info->groups_sem);
0f9dd46c 3037}
e8569813 3038
e6dcd2dc
CM		 3039static int __btrfs_reserve_extent(struct btrfs_trans_handle *trans,
3040 struct btrfs_root *root,
3041 u64 num_bytes, u64 min_alloc_size,
3042 u64 empty_size, u64 hint_byte,
3043 u64 search_end, struct btrfs_key *ins,
3044 u64 data)
fec577fb
CM		 3045{
3046 int ret;
fbdc762b 3047 u64 search_start = 0;
8790d502 3048 u64 alloc_profile;
1261ec42 3049 struct btrfs_fs_info *info = root->fs_info;
925baedd 3050
6324fbf3 3051 if (data) {
8790d502
CM		 3052 alloc_profile = info->avail_data_alloc_bits &
3053 info->data_alloc_profile;
3054 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
6324fbf3 3055 } else if (root == root->fs_info->chunk_root) {
8790d502
CM		 3056 alloc_profile = info->avail_system_alloc_bits &
3057 info->system_alloc_profile;
3058 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
6324fbf3 3059 } else {
8790d502
CM		 3060 alloc_profile = info->avail_metadata_alloc_bits &
3061 info->metadata_alloc_profile;
3062 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
6324fbf3 3063 }
98d20f67 3064again:
2b82032c 3065 data = btrfs_reduce_alloc_profile(root, data);
0ef3e66b
CM		 3066 /*
3067 * the only place that sets empty_size is btrfs_realloc_node, which
3068 * is not called recursively on allocations
3069 */
3070 if (empty_size || root->ref_cows) {
593060d7 3071 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
6324fbf3 3072 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
0ef3e66b
CM		 3073 2 * 1024 * 1024,
3074 BTRFS_BLOCK_GROUP_METADATA |
3075 (info->metadata_alloc_profile &
3076 info->avail_metadata_alloc_bits), 0);
6324fbf3
CM		 3077 }
3078 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
0ef3e66b 3079 num_bytes + 2 * 1024 * 1024, data, 0);
6324fbf3 3080 }
0b86a832 3081
db94535d
CM		 3082 WARN_ON(num_bytes < root->sectorsize);
3083 ret = find_free_extent(trans, root, num_bytes, empty_size,
3084 search_start, search_end, hint_byte, ins,
26b8003f
CM		 3085 trans->alloc_exclude_start,
3086 trans->alloc_exclude_nr, data);
3b951516 3087
98d20f67
CM		 3088 if (ret == -ENOSPC && num_bytes > min_alloc_size) {
3089 num_bytes = num_bytes >> 1;
0f9dd46c 3090 num_bytes = num_bytes & ~(root->sectorsize - 1);
98d20f67 3091 num_bytes = max(num_bytes, min_alloc_size);
0ef3e66b
CM		 3092 do_chunk_alloc(trans, root->fs_info->extent_root,
3093 num_bytes, data, 1);
98d20f67
CM		 3094 goto again;
3095 }
ec44a35c 3096 if (ret) {
0f9dd46c
JB		 3097 struct btrfs_space_info *sinfo;
3098
3099 sinfo = __find_space_info(root->fs_info, data);
3100 printk("allocation failed flags %Lu, wanted %Lu\n",
3101 data, num_bytes);
3102 dump_space_info(sinfo, num_bytes);
925baedd 3103 BUG();
925baedd 3104 }
0f9dd46c
JB		 3105
3106 return ret;
e6dcd2dc
CM		 3107}
3108
65b51a00
CM		 3109int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
3110{
0f9dd46c
JB		 3111 struct btrfs_block_group_cache *cache;
3112
0f9dd46c
JB		 3113 cache = btrfs_lookup_block_group(root->fs_info, start);
3114 if (!cache) {
3115 printk(KERN_ERR "Unable to find block group for %Lu\n", start);
0f9dd46c
JB		 3116 return -ENOSPC;
3117 }
3118 btrfs_add_free_space(cache, start, len);
1a40e23b 3119 update_reserved_extents(root, start, len, 0);
65b51a00
CM		 3120 return 0;
3121}
3122
e6dcd2dc
CM		 3123int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
3124 struct btrfs_root *root,
3125 u64 num_bytes, u64 min_alloc_size,
3126 u64 empty_size, u64 hint_byte,
3127 u64 search_end, struct btrfs_key *ins,
3128 u64 data)
3129{
3130 int ret;
e6dcd2dc
CM		 3131 ret = __btrfs_reserve_extent(trans, root, num_bytes, min_alloc_size,
3132 empty_size, hint_byte, search_end, ins,
3133 data);
e8569813 3134 update_reserved_extents(root, ins->objectid, ins->offset, 1);
e6dcd2dc
CM		 3135 return ret;
3136}
3137
3138static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
31840ae1 3139 struct btrfs_root *root, u64 parent,
e6dcd2dc 3140 u64 root_objectid, u64 ref_generation,
3bb1a1bc 3141 u64 owner, struct btrfs_key *ins)
e6dcd2dc
CM		 3142{
3143 int ret;
3144 int pending_ret;
3145 u64 super_used;
3146 u64 root_used;
3147 u64 num_bytes = ins->offset;
3148 u32 sizes[2];
3149 struct btrfs_fs_info *info = root->fs_info;
3150 struct btrfs_root *extent_root = info->extent_root;
3151 struct btrfs_extent_item *extent_item;
3152 struct btrfs_extent_ref *ref;
3153 struct btrfs_path *path;
3154 struct btrfs_key keys[2];
fec577fb 3155
31840ae1
ZY		 3156 if (parent == 0)
3157 parent = ins->objectid;
3158
58176a96 3159 /* block accounting for super block */
a2135011 3160 spin_lock_irq(&info->delalloc_lock);
db94535d
CM		 3161 super_used = btrfs_super_bytes_used(&info->super_copy);
3162 btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes);
a2135011 3163 spin_unlock_irq(&info->delalloc_lock);
26b8003f 3164
58176a96 3165 /* block accounting for root item */
db94535d
CM		 3166 root_used = btrfs_root_used(&root->root_item);
3167 btrfs_set_root_used(&root->root_item, root_used + num_bytes);
58176a96 3168
26b8003f 3169 if (root == extent_root) {
31840ae1
ZY		 3170 struct pending_extent_op *extent_op;
3171
3172 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
3173 BUG_ON(!extent_op);
3174
3175 extent_op->type = PENDING_EXTENT_INSERT;
3176 extent_op->bytenr = ins->objectid;
3177 extent_op->num_bytes = ins->offset;
3178 extent_op->parent = parent;
3179 extent_op->orig_parent = 0;
3180 extent_op->generation = ref_generation;
3181 extent_op->orig_generation = 0;
3182 extent_op->level = (int)owner;
f3465ca4
JB		 3183 INIT_LIST_HEAD(&extent_op->list);
3184 extent_op->del = 0;
31840ae1 3185
25179201 3186 mutex_lock(&root->fs_info->extent_ins_mutex);
1a5bc167
CM		 3187 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
3188 ins->objectid + ins->offset - 1,
25179201 3189 EXTENT_WRITEBACK, GFP_NOFS);
31840ae1
ZY		 3190 set_state_private(&root->fs_info->extent_ins,
3191 ins->objectid, (unsigned long)extent_op);
25179201 3192 mutex_unlock(&root->fs_info->extent_ins_mutex);
26b8003f
CM		 3193 goto update_block;
3194 }
3195
47e4bb98 3196 memcpy(&keys[0], ins, sizeof(*ins));
47e4bb98
CM		 3197 keys[1].objectid = ins->objectid;
3198 keys[1].type = BTRFS_EXTENT_REF_KEY;
31840ae1 3199 keys[1].offset = parent;
47e4bb98
CM		 3200 sizes[0] = sizeof(*extent_item);
3201 sizes[1] = sizeof(*ref);
7bb86316
CM		 3202
3203 path = btrfs_alloc_path();
3204 BUG_ON(!path);
47e4bb98
CM		 3205
3206 ret = btrfs_insert_empty_items(trans, extent_root, path, keys,
3207 sizes, 2);
ccd467d6 3208 BUG_ON(ret);
0f9dd46c 3209
47e4bb98
CM		 3210 extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3211 struct btrfs_extent_item);
3212 btrfs_set_extent_refs(path->nodes[0], extent_item, 1);
3213 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
3214 struct btrfs_extent_ref);
3215
3216 btrfs_set_ref_root(path->nodes[0], ref, root_objectid);
3217 btrfs_set_ref_generation(path->nodes[0], ref, ref_generation);
3218 btrfs_set_ref_objectid(path->nodes[0], ref, owner);
31840ae1 3219 btrfs_set_ref_num_refs(path->nodes[0], ref, 1);
47e4bb98
CM		 3220
3221 btrfs_mark_buffer_dirty(path->nodes[0]);
3222
3223 trans->alloc_exclude_start = 0;
3224 trans->alloc_exclude_nr = 0;
7bb86316 3225 btrfs_free_path(path);
87ef2bb4
CM		 3226 finish_current_insert(trans, extent_root, 0);
3227 pending_ret = del_pending_extents(trans, extent_root, 0);
f510cfec 3228
925baedd
CM		 3229 if (ret)
3230 goto out;
e37c9e69 3231 if (pending_ret) {
925baedd
CM		 3232 ret = pending_ret;
3233 goto out;
e37c9e69 3234 }
26b8003f
CM		 3235
3236update_block:
0b86a832 3237 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1, 0);
f5947066
CM		 3238 if (ret) {
3239 printk("update block group failed for %Lu %Lu\n",
3240 ins->objectid, ins->offset);
3241 BUG();
3242 }
925baedd 3243out:
e6dcd2dc
CM		 3244 return ret;
3245}
3246
3247int btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
31840ae1 3248 struct btrfs_root *root, u64 parent,
e6dcd2dc 3249 u64 root_objectid, u64 ref_generation,
3bb1a1bc 3250 u64 owner, struct btrfs_key *ins)
e6dcd2dc
CM		 3251{
3252 int ret;
1c2308f8
CM		 3253
3254 if (root_objectid == BTRFS_TREE_LOG_OBJECTID)
3255 return 0;
3bb1a1bc
YZ		 3256 ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
3257 ref_generation, owner, ins);
e8569813 3258 update_reserved_extents(root, ins->objectid, ins->offset, 0);
e6dcd2dc
CM		 3259 return ret;
3260}
e02119d5
CM		 3261
3262/*
3263 * this is used by the tree logging recovery code. It records that
3264 * an extent has been allocated and makes sure to clear the free
3265 * space cache bits as well
3266 */
3267int btrfs_alloc_logged_extent(struct btrfs_trans_handle *trans,
31840ae1 3268 struct btrfs_root *root, u64 parent,
e02119d5 3269 u64 root_objectid, u64 ref_generation,
3bb1a1bc 3270 u64 owner, struct btrfs_key *ins)
e02119d5
CM		 3271{
3272 int ret;
3273 struct btrfs_block_group_cache *block_group;
3274
e02119d5 3275 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
25179201 3276 mutex_lock(&block_group->alloc_mutex);
e02119d5
CM		 3277 cache_block_group(root, block_group);
3278
25179201
JB		 3279 ret = btrfs_remove_free_space_lock(block_group, ins->objectid,
3280 ins->offset);
3281 mutex_unlock(&block_group->alloc_mutex);
0f9dd46c 3282 BUG_ON(ret);
3bb1a1bc
YZ		 3283 ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
3284 ref_generation, owner, ins);
e02119d5
CM		 3285 return ret;
3286}
3287
e6dcd2dc
CM		 3288/*
3289 * finds a free extent and does all the dirty work required for allocation
3290 * returns the key for the extent through ins, and a tree buffer for
3291 * the first block of the extent through buf.
3292 *
3293 * returns 0 if everything worked, non-zero otherwise.
3294 */
3295int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
3296 struct btrfs_root *root,
31840ae1 3297 u64 num_bytes, u64 parent, u64 min_alloc_size,
e6dcd2dc 3298 u64 root_objectid, u64 ref_generation,
3bb1a1bc 3299 u64 owner_objectid, u64 empty_size, u64 hint_byte,
e6dcd2dc
CM		 3300 u64 search_end, struct btrfs_key *ins, u64 data)
3301{
3302 int ret;
3303
e6dcd2dc
CM		 3304 ret = __btrfs_reserve_extent(trans, root, num_bytes,
3305 min_alloc_size, empty_size, hint_byte,
3306 search_end, ins, data);
3307 BUG_ON(ret);
d00aff00 3308 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
31840ae1
ZY		 3309 ret = __btrfs_alloc_reserved_extent(trans, root, parent,
3310 root_objectid, ref_generation,
3bb1a1bc 3311 owner_objectid, ins);
d00aff00 3312 BUG_ON(ret);
e6dcd2dc 3313
e8569813
ZY		 3314 } else {
3315 update_reserved_extents(root, ins->objectid, ins->offset, 1);
d00aff00 3316 }
925baedd 3317 return ret;
fec577fb 3318}
65b51a00
CM		 3319
3320struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
3321 struct btrfs_root *root,
3322 u64 bytenr, u32 blocksize)
3323{
3324 struct extent_buffer *buf;
3325
3326 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
3327 if (!buf)
3328 return ERR_PTR(-ENOMEM);
3329 btrfs_set_header_generation(buf, trans->transid);
3330 btrfs_tree_lock(buf);
3331 clean_tree_block(trans, root, buf);
3332 btrfs_set_buffer_uptodate(buf);
d0c803c4
CM		 3333 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
3334 set_extent_dirty(&root->dirty_log_pages, buf->start,
3335 buf->start + buf->len - 1, GFP_NOFS);
3336 } else {
3337 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
65b51a00 3338 buf->start + buf->len - 1, GFP_NOFS);
d0c803c4 3339 }
65b51a00
CM		 3340 trans->blocks_used++;
3341 return buf;
3342}
3343
fec577fb
CM		 3344/*
3345 * helper function to allocate a block for a given tree
3346 * returns the tree buffer or NULL.
3347 */
5f39d397 3348struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
7bb86316 3349 struct btrfs_root *root,
31840ae1 3350 u32 blocksize, u64 parent,
7bb86316
CM		 3351 u64 root_objectid,
3352 u64 ref_generation,
7bb86316
CM		 3353 int level,
3354 u64 hint,
5f39d397 3355 u64 empty_size)
fec577fb 3356{
e2fa7227 3357 struct btrfs_key ins;
fec577fb 3358 int ret;
5f39d397 3359 struct extent_buffer *buf;
fec577fb 3360
31840ae1 3361 ret = btrfs_alloc_extent(trans, root, blocksize, parent, blocksize,
3bb1a1bc 3362 root_objectid, ref_generation, level,
31840ae1 3363 empty_size, hint, (u64)-1, &ins, 0);
fec577fb 3364 if (ret) {
54aa1f4d
CM		 3365 BUG_ON(ret > 0);
3366 return ERR_PTR(ret);
fec577fb 3367 }
55c69072 3368
65b51a00 3369 buf = btrfs_init_new_buffer(trans, root, ins.objectid, blocksize);
fec577fb
CM		 3370 return buf;
3371}
a28ec197 3372
e02119d5
CM		 3373int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans,
3374 struct btrfs_root *root, struct extent_buffer *leaf)
6407bf6d 3375{
7bb86316
CM		 3376 u64 leaf_owner;
3377 u64 leaf_generation;
5f39d397 3378 struct btrfs_key key;
6407bf6d
CM		 3379 struct btrfs_file_extent_item *fi;
3380 int i;
3381 int nritems;
3382 int ret;
3383
5f39d397
CM		 3384 BUG_ON(!btrfs_is_leaf(leaf));
3385 nritems = btrfs_header_nritems(leaf);
7bb86316
CM		 3386 leaf_owner = btrfs_header_owner(leaf);
3387 leaf_generation = btrfs_header_generation(leaf);
3388
6407bf6d 3389 for (i = 0; i < nritems; i++) {
db94535d 3390 u64 disk_bytenr;
e34a5b4f 3391 cond_resched();
5f39d397
CM		 3392
3393 btrfs_item_key_to_cpu(leaf, &key, i);
3394 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
6407bf6d
CM		 3395 continue;
3396 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
5f39d397
CM		 3397 if (btrfs_file_extent_type(leaf, fi) ==
3398 BTRFS_FILE_EXTENT_INLINE)
236454df 3399 continue;
6407bf6d
CM		 3400 /*
3401 * FIXME make sure to insert a trans record that
3402 * repeats the snapshot del on crash
3403 */
db94535d
CM		 3404 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
3405 if (disk_bytenr == 0)
3a686375 3406 continue;
4a096752 3407
925baedd 3408 ret = __btrfs_free_extent(trans, root, disk_bytenr,
7bb86316 3409 btrfs_file_extent_disk_num_bytes(leaf, fi),
31840ae1 3410 leaf->start, leaf_owner, leaf_generation,
3bb1a1bc 3411 key.objectid, 0);
31840ae1 3412 BUG_ON(ret);
2dd3e67b
CM		 3413
3414 atomic_inc(&root->fs_info->throttle_gen);
3415 wake_up(&root->fs_info->transaction_throttle);
3416 cond_resched();
6407bf6d
CM		 3417 }
3418 return 0;
3419}
3420
e02119d5
CM		 3421static int noinline cache_drop_leaf_ref(struct btrfs_trans_handle *trans,
3422 struct btrfs_root *root,
3423 struct btrfs_leaf_ref *ref)
31153d81
YZ		 3424{
3425 int i;
3426 int ret;
3427 struct btrfs_extent_info *info = ref->extents;
3428
31153d81 3429 for (i = 0; i < ref->nritems; i++) {
31840ae1
ZY		 3430 ret = __btrfs_free_extent(trans, root, info->bytenr,
3431 info->num_bytes, ref->bytenr,
3432 ref->owner, ref->generation,
3bb1a1bc 3433 info->objectid, 0);
2dd3e67b
CM		 3434
3435 atomic_inc(&root->fs_info->throttle_gen);
3436 wake_up(&root->fs_info->transaction_throttle);
3437 cond_resched();
3438
31153d81
YZ		 3439 BUG_ON(ret);
3440 info++;
3441 }
31153d81
YZ		 3442
3443 return 0;
3444}
3445
333db94c
CM		 3446int drop_snap_lookup_refcount(struct btrfs_root *root, u64 start, u64 len,
3447 u32 *refs)
3448{
017e5369 3449 int ret;
f87f057b 3450
31840ae1 3451 ret = btrfs_lookup_extent_ref(NULL, root, start, len, refs);
f87f057b
CM		 3452 BUG_ON(ret);
3453
3454#if 0 // some debugging code in case we see problems here
3455 /* if the refs count is one, it won't get increased again. But
3456 * if the ref count is > 1, someone may be decreasing it at
3457 * the same time we are.
3458 */
3459 if (*refs != 1) {
3460 struct extent_buffer *eb = NULL;
3461 eb = btrfs_find_create_tree_block(root, start, len);
3462 if (eb)
3463 btrfs_tree_lock(eb);
3464
3465 mutex_lock(&root->fs_info->alloc_mutex);
3466 ret = lookup_extent_ref(NULL, root, start, len, refs);
3467 BUG_ON(ret);
3468 mutex_unlock(&root->fs_info->alloc_mutex);
3469
3470 if (eb) {
3471 btrfs_tree_unlock(eb);
3472 free_extent_buffer(eb);
3473 }
3474 if (*refs == 1) {
3475 printk("block %llu went down to one during drop_snap\n",
3476 (unsigned long long)start);
3477 }
3478
3479 }
3480#endif
3481
e7a84565 3482 cond_resched();
017e5369 3483 return ret;
333db94c
CM		 3484}
3485
9aca1d51
CM		 3486/*
3487 * helper function for drop_snapshot, this walks down the tree dropping ref
3488 * counts as it goes.
3489 */
98ed5174
CM		 3490static int noinline walk_down_tree(struct btrfs_trans_handle *trans,
3491 struct btrfs_root *root,
3492 struct btrfs_path *path, int *level)
20524f02 3493{
7bb86316
CM		 3494 u64 root_owner;
3495 u64 root_gen;
3496 u64 bytenr;
ca7a79ad 3497 u64 ptr_gen;
5f39d397
CM		 3498 struct extent_buffer *next;
3499 struct extent_buffer *cur;
7bb86316 3500 struct extent_buffer *parent;
31153d81 3501 struct btrfs_leaf_ref *ref;
db94535d 3502 u32 blocksize;
20524f02
CM		 3503 int ret;
3504 u32 refs;
3505
5caf2a00
CM		 3506 WARN_ON(*level < 0);
3507 WARN_ON(*level >= BTRFS_MAX_LEVEL);
333db94c 3508 ret = drop_snap_lookup_refcount(root, path->nodes[*level]->start,
db94535d 3509 path->nodes[*level]->len, &refs);
20524f02
CM		 3510 BUG_ON(ret);
3511 if (refs > 1)
3512 goto out;
e011599b 3513
9aca1d51
CM		 3514 /*
3515 * walk down to the last node level and free all the leaves
3516 */
6407bf6d 3517 while(*level >= 0) {
5caf2a00
CM		 3518 WARN_ON(*level < 0);
3519 WARN_ON(*level >= BTRFS_MAX_LEVEL);
20524f02 3520 cur = path->nodes[*level];
e011599b 3521
5f39d397 3522 if (btrfs_header_level(cur) != *level)
2c90e5d6 3523 WARN_ON(1);
e011599b 3524
7518a238 3525 if (path->slots[*level] >=
5f39d397 3526 btrfs_header_nritems(cur))
20524f02 3527 break;
6407bf6d 3528 if (*level == 0) {
e02119d5 3529 ret = btrfs_drop_leaf_ref(trans, root, cur);
6407bf6d
CM		 3530 BUG_ON(ret);
3531 break;
3532 }
db94535d 3533 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
ca7a79ad 3534 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
db94535d 3535 blocksize = btrfs_level_size(root, *level - 1);
925baedd 3536
333db94c 3537 ret = drop_snap_lookup_refcount(root, bytenr, blocksize, &refs);
6407bf6d
CM		 3538 BUG_ON(ret);
3539 if (refs != 1) {
7bb86316
CM		 3540 parent = path->nodes[*level];
3541 root_owner = btrfs_header_owner(parent);
3542 root_gen = btrfs_header_generation(parent);
20524f02 3543 path->slots[*level]++;
f87f057b 3544
925baedd 3545 ret = __btrfs_free_extent(trans, root, bytenr,
31840ae1 3546 blocksize, parent->start,
3bb1a1bc
YZ		 3547 root_owner, root_gen,
3548 *level - 1, 1);
20524f02 3549 BUG_ON(ret);
18e35e0a
CM		 3550
3551 atomic_inc(&root->fs_info->throttle_gen);
3552 wake_up(&root->fs_info->transaction_throttle);
2dd3e67b 3553 cond_resched();
18e35e0a 3554
20524f02
CM		 3555 continue;
3556 }
f87f057b
CM		 3557 /*
3558 * at this point, we have a single ref, and since the
3559 * only place referencing this extent is a dead root
3560 * the reference count should never go higher.
3561 * So, we don't need to check it again
3562 */
31153d81 3563 if (*level == 1) {
017e5369 3564 ref = btrfs_lookup_leaf_ref(root, bytenr);
1a40e23b
ZY		 3565 if (ref && ref->generation != ptr_gen) {
3566 btrfs_free_leaf_ref(root, ref);
3567 ref = NULL;
3568 }
31153d81 3569 if (ref) {
e02119d5 3570 ret = cache_drop_leaf_ref(trans, root, ref);
31153d81
YZ		 3571 BUG_ON(ret);
3572 btrfs_remove_leaf_ref(root, ref);
bcc63abb 3573 btrfs_free_leaf_ref(root, ref);
31153d81
YZ		 3574 *level = 0;
3575 break;
3576 }
3bb1a1bc 3577 if (printk_ratelimit()) {
37d1aeee
CM		 3578 printk("leaf ref miss for bytenr %llu\n",
3579 (unsigned long long)bytenr);
3bb1a1bc 3580 }
31153d81 3581 }
db94535d 3582 next = btrfs_find_tree_block(root, bytenr, blocksize);
1259ab75 3583 if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
5f39d397 3584 free_extent_buffer(next);
333db94c 3585
ca7a79ad
CM		 3586 next = read_tree_block(root, bytenr, blocksize,
3587 ptr_gen);
e7a84565 3588 cond_resched();
f87f057b
CM		 3589#if 0
3590 /*
3591 * this is a debugging check and can go away
3592 * the ref should never go all the way down to 1
3593 * at this point
3594 */
e6dcd2dc
CM		 3595 ret = lookup_extent_ref(NULL, root, bytenr, blocksize,
3596 &refs);
e9d0b13b 3597 BUG_ON(ret);
f87f057b
CM		 3598 WARN_ON(refs != 1);
3599#endif
e9d0b13b 3600 }
5caf2a00 3601 WARN_ON(*level <= 0);
83e15a28 3602 if (path->nodes[*level-1])
5f39d397 3603 free_extent_buffer(path->nodes[*level-1]);
20524f02 3604 path->nodes[*level-1] = next;
5f39d397 3605 *level = btrfs_header_level(next);
20524f02 3606 path->slots[*level] = 0;
2dd3e67b 3607 cond_resched();
20524f02
CM		 3608 }
3609out:
5caf2a00
CM		 3610 WARN_ON(*level < 0);
3611 WARN_ON(*level >= BTRFS_MAX_LEVEL);
7bb86316
CM		 3612
3613 if (path->nodes[*level] == root->node) {
7bb86316 3614 parent = path->nodes[*level];
31153d81 3615 bytenr = path->nodes[*level]->start;
7bb86316
CM		 3616 } else {
3617 parent = path->nodes[*level + 1];
31153d81 3618 bytenr = btrfs_node_blockptr(parent, path->slots[*level + 1]);
7bb86316
CM		 3619 }
3620
31153d81
YZ		 3621 blocksize = btrfs_level_size(root, *level);
3622 root_owner = btrfs_header_owner(parent);
7bb86316 3623 root_gen = btrfs_header_generation(parent);
31153d81
YZ		 3624
3625 ret = __btrfs_free_extent(trans, root, bytenr, blocksize,
31840ae1 3626 parent->start, root_owner, root_gen,
3bb1a1bc 3627 *level, 1);
5f39d397 3628 free_extent_buffer(path->nodes[*level]);
20524f02
CM		 3629 path->nodes[*level] = NULL;
3630 *level += 1;
3631 BUG_ON(ret);
f87f057b 3632
e7a84565 3633 cond_resched();
20524f02
CM		 3634 return 0;
3635}
3636
f82d02d9
YZ		 3637/*
3638 * helper function for drop_subtree, this function is similar to
3639 * walk_down_tree. The main difference is that it checks reference
3640 * counts while tree blocks are locked.
3641 */
3642static int noinline walk_down_subtree(struct btrfs_trans_handle *trans,
3643 struct btrfs_root *root,
3644 struct btrfs_path *path, int *level)
3645{
3646 struct extent_buffer *next;
3647 struct extent_buffer *cur;
3648 struct extent_buffer *parent;
3649 u64 bytenr;
3650 u64 ptr_gen;
3651 u32 blocksize;
3652 u32 refs;
3653 int ret;
3654
3655 cur = path->nodes[*level];
3656 ret = btrfs_lookup_extent_ref(trans, root, cur->start, cur->len,
3657 &refs);
3658 BUG_ON(ret);
3659 if (refs > 1)
3660 goto out;
3661
3662 while (*level >= 0) {
3663 cur = path->nodes[*level];
3664 if (*level == 0) {
3665 ret = btrfs_drop_leaf_ref(trans, root, cur);
3666 BUG_ON(ret);
3667 clean_tree_block(trans, root, cur);
3668 break;
3669 }
3670 if (path->slots[*level] >= btrfs_header_nritems(cur)) {
3671 clean_tree_block(trans, root, cur);
3672 break;
3673 }
3674
3675 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
3676 blocksize = btrfs_level_size(root, *level - 1);
3677 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
3678
3679 next = read_tree_block(root, bytenr, blocksize, ptr_gen);
3680 btrfs_tree_lock(next);
3681
3682 ret = btrfs_lookup_extent_ref(trans, root, bytenr, blocksize,
3683 &refs);
3684 BUG_ON(ret);
3685 if (refs > 1) {
3686 parent = path->nodes[*level];
3687 ret = btrfs_free_extent(trans, root, bytenr,
3688 blocksize, parent->start,
3689 btrfs_header_owner(parent),
3690 btrfs_header_generation(parent),
3691 *level - 1, 1);
3692 BUG_ON(ret);
3693 path->slots[*level]++;
3694 btrfs_tree_unlock(next);
3695 free_extent_buffer(next);
3696 continue;
3697 }
3698
3699 *level = btrfs_header_level(next);
3700 path->nodes[*level] = next;
3701 path->slots[*level] = 0;
3702 path->locks[*level] = 1;
3703 cond_resched();
3704 }
3705out:
3706 parent = path->nodes[*level + 1];
3707 bytenr = path->nodes[*level]->start;
3708 blocksize = path->nodes[*level]->len;
3709
3710 ret = btrfs_free_extent(trans, root, bytenr, blocksize,
3711 parent->start, btrfs_header_owner(parent),
3712 btrfs_header_generation(parent), *level, 1);
3713 BUG_ON(ret);
3714
3715 if (path->locks[*level]) {
3716 btrfs_tree_unlock(path->nodes[*level]);
3717 path->locks[*level] = 0;
3718 }
3719 free_extent_buffer(path->nodes[*level]);
3720 path->nodes[*level] = NULL;
3721 *level += 1;
3722 cond_resched();
3723 return 0;
3724}
3725
9aca1d51
CM		 3726/*
3727 * helper for dropping snapshots. This walks back up the tree in the path
3728 * to find the first node higher up where we haven't yet gone through
3729 * all the slots
3730 */
98ed5174
CM		 3731static int noinline walk_up_tree(struct btrfs_trans_handle *trans,
3732 struct btrfs_root *root,
f82d02d9
YZ		 3733 struct btrfs_path *path,
3734 int *level, int max_level)
20524f02 3735{
7bb86316
CM		 3736 u64 root_owner;
3737 u64 root_gen;
3738 struct btrfs_root_item *root_item = &root->root_item;
20524f02
CM		 3739 int i;
3740 int slot;
3741 int ret;
9f3a7427 3742
f82d02d9 3743 for (i = *level; i < max_level && path->nodes[i]; i++) {
20524f02 3744 slot = path->slots[i];
5f39d397
CM		 3745 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
3746 struct extent_buffer *node;
3747 struct btrfs_disk_key disk_key;
3748 node = path->nodes[i];
20524f02
CM		 3749 path->slots[i]++;
3750 *level = i;
9f3a7427 3751 WARN_ON(*level == 0);
5f39d397 3752 btrfs_node_key(node, &disk_key, path->slots[i]);
9f3a7427 3753 memcpy(&root_item->drop_progress,
5f39d397 3754 &disk_key, sizeof(disk_key));
9f3a7427 3755 root_item->drop_level = i;
20524f02
CM		 3756 return 0;
3757 } else {
31840ae1
ZY		 3758 struct extent_buffer *parent;
3759 if (path->nodes[*level] == root->node)
3760 parent = path->nodes[*level];
3761 else
3762 parent = path->nodes[*level + 1];
3763
3764 root_owner = btrfs_header_owner(parent);
3765 root_gen = btrfs_header_generation(parent);
f82d02d9
YZ		 3766
3767 clean_tree_block(trans, root, path->nodes[*level]);
e089f05c 3768 ret = btrfs_free_extent(trans, root,
db94535d 3769 path->nodes[*level]->start,
7bb86316 3770 path->nodes[*level]->len,
3bb1a1bc
YZ		 3771 parent->start, root_owner,
3772 root_gen, *level, 1);
6407bf6d 3773 BUG_ON(ret);
f82d02d9
YZ		 3774 if (path->locks[*level]) {
3775 btrfs_tree_unlock(path->nodes[*level]);
3776 path->locks[*level] = 0;
3777 }
5f39d397 3778 free_extent_buffer(path->nodes[*level]);
83e15a28 3779 path->nodes[*level] = NULL;
20524f02 3780 *level = i + 1;
20524f02
CM		 3781 }
3782 }
3783 return 1;
3784}
3785
9aca1d51
CM		 3786/*
3787 * drop the reference count on the tree rooted at 'snap'. This traverses
3788 * the tree freeing any blocks that have a ref count of zero after being
3789 * decremented.
3790 */
e089f05c 3791int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
9f3a7427 3792 *root)
20524f02 3793{
3768f368 3794 int ret = 0;
9aca1d51 3795 int wret;
20524f02 3796 int level;
5caf2a00 3797 struct btrfs_path *path;
20524f02
CM		 3798 int i;
3799 int orig_level;
9f3a7427 3800 struct btrfs_root_item *root_item = &root->root_item;
20524f02 3801
a2135011 3802 WARN_ON(!mutex_is_locked(&root->fs_info->drop_mutex));
5caf2a00
CM		 3803 path = btrfs_alloc_path();
3804 BUG_ON(!path);
20524f02 3805
5f39d397 3806 level = btrfs_header_level(root->node);
20524f02 3807 orig_level = level;
9f3a7427
CM		 3808 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
3809 path->nodes[level] = root->node;
f510cfec 3810 extent_buffer_get(root->node);
9f3a7427
CM		 3811 path->slots[level] = 0;
3812 } else {
3813 struct btrfs_key key;
5f39d397
CM		 3814 struct btrfs_disk_key found_key;
3815 struct extent_buffer *node;
6702ed49 3816
9f3a7427 3817 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
6702ed49
CM		 3818 level = root_item->drop_level;
3819 path->lowest_level = level;
9f3a7427 3820 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6702ed49 3821 if (wret < 0) {
9f3a7427
CM		 3822 ret = wret;
3823 goto out;
3824 }
5f39d397
CM		 3825 node = path->nodes[level];
3826 btrfs_node_key(node, &found_key, path->slots[level]);
3827 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
3828 sizeof(found_key)));
7d9eb12c
CM		 3829 /*
3830 * unlock our path, this is safe because only this
3831 * function is allowed to delete this snapshot
3832 */
925baedd
CM		 3833 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
3834 if (path->nodes[i] && path->locks[i]) {
3835 path->locks[i] = 0;
3836 btrfs_tree_unlock(path->nodes[i]);
3837 }
3838 }
9f3a7427 3839 }
20524f02 3840 while(1) {
5caf2a00 3841 wret = walk_down_tree(trans, root, path, &level);
9aca1d51 3842 if (wret > 0)
20524f02 3843 break;
9aca1d51
CM		 3844 if (wret < 0)
3845 ret = wret;
3846
f82d02d9
YZ		 3847 wret = walk_up_tree(trans, root, path, &level,
3848 BTRFS_MAX_LEVEL);
9aca1d51 3849 if (wret > 0)
20524f02 3850 break;
9aca1d51
CM		 3851 if (wret < 0)
3852 ret = wret;
e7a84565
CM		 3853 if (trans->transaction->in_commit) {
3854 ret = -EAGAIN;
3855 break;
3856 }
18e35e0a 3857 atomic_inc(&root->fs_info->throttle_gen);
017e5369 3858 wake_up(&root->fs_info->transaction_throttle);
20524f02 3859 }
83e15a28 3860 for (i = 0; i <= orig_level; i++) {
5caf2a00 3861 if (path->nodes[i]) {
5f39d397 3862 free_extent_buffer(path->nodes[i]);
0f82731f 3863 path->nodes[i] = NULL;
83e15a28 3864 }
20524f02 3865 }
9f3a7427 3866out:
5caf2a00 3867 btrfs_free_path(path);
9aca1d51 3868 return ret;
20524f02 3869}
9078a3e1 3870
f82d02d9
YZ		 3871int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
3872 struct btrfs_root *root,
3873 struct extent_buffer *node,
3874 struct extent_buffer *parent)
3875{
3876 struct btrfs_path *path;
3877 int level;
3878 int parent_level;
3879 int ret = 0;
3880 int wret;
3881
3882 path = btrfs_alloc_path();
3883 BUG_ON(!path);
3884
3885 BUG_ON(!btrfs_tree_locked(parent));
3886 parent_level = btrfs_header_level(parent);
3887 extent_buffer_get(parent);
3888 path->nodes[parent_level] = parent;
3889 path->slots[parent_level] = btrfs_header_nritems(parent);
3890
3891 BUG_ON(!btrfs_tree_locked(node));
3892 level = btrfs_header_level(node);
3893 extent_buffer_get(node);
3894 path->nodes[level] = node;
3895 path->slots[level] = 0;
3896
3897 while (1) {
3898 wret = walk_down_subtree(trans, root, path, &level);
3899 if (wret < 0)
3900 ret = wret;
3901 if (wret != 0)
3902 break;
3903
3904 wret = walk_up_tree(trans, root, path, &level, parent_level);
3905 if (wret < 0)
3906 ret = wret;
3907 if (wret != 0)
3908 break;
3909 }
3910
3911 btrfs_free_path(path);
3912 return ret;
3913}
3914
8e7bf94f
CM		 3915static unsigned long calc_ra(unsigned long start, unsigned long last,
3916 unsigned long nr)
3917{
3918 return min(last, start + nr - 1);
3919}
3920
98ed5174
CM		 3921static int noinline relocate_inode_pages(struct inode *inode, u64 start,
3922 u64 len)
edbd8d4e
CM		 3923{
3924 u64 page_start;
3925 u64 page_end;
1a40e23b 3926 unsigned long first_index;
edbd8d4e 3927 unsigned long last_index;
edbd8d4e
CM		 3928 unsigned long i;
3929 struct page *page;
d1310b2e 3930 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4313b399 3931 struct file_ra_state *ra;
3eaa2885 3932 struct btrfs_ordered_extent *ordered;
1a40e23b
ZY		 3933 unsigned int total_read = 0;
3934 unsigned int total_dirty = 0;
3935 int ret = 0;
4313b399
CM		 3936
3937 ra = kzalloc(sizeof(*ra), GFP_NOFS);
edbd8d4e
CM		 3938
3939 mutex_lock(&inode->i_mutex);
1a40e23b 3940 first_index = start >> PAGE_CACHE_SHIFT;
edbd8d4e
CM		 3941 last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
3942
1a40e23b
ZY		 3943 /* make sure the dirty trick played by the caller work */
3944 ret = invalidate_inode_pages2_range(inode->i_mapping,
3945 first_index, last_index);
3946 if (ret)
3947 goto out_unlock;
8e7bf94f 3948
4313b399 3949 file_ra_state_init(ra, inode->i_mapping);
edbd8d4e 3950
1a40e23b
ZY		 3951 for (i = first_index ; i <= last_index; i++) {
3952 if (total_read % ra->ra_pages == 0) {
8e7bf94f 3953 btrfs_force_ra(inode->i_mapping, ra, NULL, i,
1a40e23b 3954 calc_ra(i, last_index, ra->ra_pages));
8e7bf94f
CM		 3955 }
3956 total_read++;
3eaa2885
CM		 3957again:
3958 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
1a40e23b 3959 BUG_ON(1);
edbd8d4e 3960 page = grab_cache_page(inode->i_mapping, i);
a061fc8d 3961 if (!page) {
1a40e23b 3962 ret = -ENOMEM;
edbd8d4e 3963 goto out_unlock;
a061fc8d 3964 }
edbd8d4e
CM		 3965 if (!PageUptodate(page)) {
3966 btrfs_readpage(NULL, page);
3967 lock_page(page);
3968 if (!PageUptodate(page)) {
3969 unlock_page(page);
3970 page_cache_release(page);
1a40e23b 3971 ret = -EIO;
edbd8d4e
CM		 3972 goto out_unlock;
3973 }
3974 }
ec44a35c 3975 wait_on_page_writeback(page);
3eaa2885 3976
edbd8d4e
CM		 3977 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
3978 page_end = page_start + PAGE_CACHE_SIZE - 1;
d1310b2e 3979 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
edbd8d4e 3980
3eaa2885
CM		 3981 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3982 if (ordered) {
3983 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3984 unlock_page(page);
3985 page_cache_release(page);
3986 btrfs_start_ordered_extent(inode, ordered, 1);
3987 btrfs_put_ordered_extent(ordered);
3988 goto again;
3989 }
3990 set_page_extent_mapped(page);
3991
ea8c2819 3992 btrfs_set_extent_delalloc(inode, page_start, page_end);
1a40e23b
ZY		 3993 if (i == first_index)
3994 set_extent_bits(io_tree, page_start, page_end,
3995 EXTENT_BOUNDARY, GFP_NOFS);
3996
a061fc8d 3997 set_page_dirty(page);
1a40e23b 3998 total_dirty++;
edbd8d4e 3999
d1310b2e 4000 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
edbd8d4e
CM		 4001 unlock_page(page);
4002 page_cache_release(page);
4003 }
4004
4005out_unlock:
ec44a35c 4006 kfree(ra);
edbd8d4e 4007 mutex_unlock(&inode->i_mutex);
1a40e23b
ZY		 4008 balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
4009 return ret;
edbd8d4e
CM		 4010}
4011
1a40e23b
ZY		 4012static int noinline relocate_data_extent(struct inode *reloc_inode,
4013 struct btrfs_key *extent_key,
4014 u64 offset)
4015{
4016 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
4017 struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
4018 struct extent_map *em;
6643558d
YZ		 4019 u64 start = extent_key->objectid - offset;
4020 u64 end = start + extent_key->offset - 1;
bf4ef679 4021
1a40e23b
ZY		 4022 em = alloc_extent_map(GFP_NOFS);
4023 BUG_ON(!em || IS_ERR(em));
bf4ef679 4024
6643558d 4025 em->start = start;
1a40e23b 4026 em->len = extent_key->offset;
c8b97818 4027 em->block_len = extent_key->offset;
1a40e23b
ZY		 4028 em->block_start = extent_key->objectid;
4029 em->bdev = root->fs_info->fs_devices->latest_bdev;
4030 set_bit(EXTENT_FLAG_PINNED, &em->flags);
4031
4032 /* setup extent map to cheat btrfs_readpage */
6643558d 4033 lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
1a40e23b
ZY		 4034 while (1) {
4035 int ret;
4036 spin_lock(&em_tree->lock);
4037 ret = add_extent_mapping(em_tree, em);
4038 spin_unlock(&em_tree->lock);
4039 if (ret != -EEXIST) {
4040 free_extent_map(em);
bf4ef679
CM		 4041 break;
4042 }
6643558d 4043 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
bf4ef679 4044 }
6643558d 4045 unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
bf4ef679 4046
6643558d 4047 return relocate_inode_pages(reloc_inode, start, extent_key->offset);
1a40e23b 4048}
edbd8d4e 4049
1a40e23b
ZY		 4050struct btrfs_ref_path {
4051 u64 extent_start;
4052 u64 nodes[BTRFS_MAX_LEVEL];
4053 u64 root_objectid;
4054 u64 root_generation;
4055 u64 owner_objectid;
1a40e23b
ZY		 4056 u32 num_refs;
4057 int lowest_level;
4058 int current_level;
f82d02d9
YZ		 4059 int shared_level;
4060
4061 struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
4062 u64 new_nodes[BTRFS_MAX_LEVEL];
1a40e23b 4063};
7d9eb12c 4064
1a40e23b 4065struct disk_extent {
c8b97818 4066 u64 ram_bytes;
1a40e23b
ZY		 4067 u64 disk_bytenr;
4068 u64 disk_num_bytes;
4069 u64 offset;
4070 u64 num_bytes;
c8b97818
CM		 4071 u8 compression;
4072 u8 encryption;
4073 u16 other_encoding;
1a40e23b 4074};
4313b399 4075
1a40e23b
ZY		 4076static int is_cowonly_root(u64 root_objectid)
4077{
4078 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
4079 root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
4080 root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
4081 root_objectid == BTRFS_DEV_TREE_OBJECTID ||
4082 root_objectid == BTRFS_TREE_LOG_OBJECTID)
4083 return 1;
4084 return 0;
4085}
edbd8d4e 4086
1a40e23b
ZY		 4087static int noinline __next_ref_path(struct btrfs_trans_handle *trans,
4088 struct btrfs_root *extent_root,
4089 struct btrfs_ref_path *ref_path,
4090 int first_time)
4091{
4092 struct extent_buffer *leaf;
4093 struct btrfs_path *path;
4094 struct btrfs_extent_ref *ref;
4095 struct btrfs_key key;
4096 struct btrfs_key found_key;
4097 u64 bytenr;
4098 u32 nritems;
4099 int level;
4100 int ret = 1;
edbd8d4e 4101
1a40e23b
ZY		 4102 path = btrfs_alloc_path();
4103 if (!path)
4104 return -ENOMEM;
bf4ef679 4105
1a40e23b
ZY		 4106 if (first_time) {
4107 ref_path->lowest_level = -1;
4108 ref_path->current_level = -1;
f82d02d9 4109 ref_path->shared_level = -1;
1a40e23b
ZY		 4110 goto walk_up;
4111 }
4112walk_down:
4113 level = ref_path->current_level - 1;
4114 while (level >= -1) {
4115 u64 parent;
4116 if (level < ref_path->lowest_level)
4117 break;
bf4ef679 4118
1a40e23b
ZY		 4119 if (level >= 0) {
4120 bytenr = ref_path->nodes[level];
4121 } else {
4122 bytenr = ref_path->extent_start;
4123 }
4124 BUG_ON(bytenr == 0);
bf4ef679 4125
1a40e23b
ZY		 4126 parent = ref_path->nodes[level + 1];
4127 ref_path->nodes[level + 1] = 0;
4128 ref_path->current_level = level;
4129 BUG_ON(parent == 0);
0ef3e66b 4130
1a40e23b
ZY		 4131 key.objectid = bytenr;
4132 key.offset = parent + 1;
4133 key.type = BTRFS_EXTENT_REF_KEY;
edbd8d4e 4134
1a40e23b
ZY		 4135 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
4136 if (ret < 0)
edbd8d4e 4137 goto out;
1a40e23b 4138 BUG_ON(ret == 0);
7d9eb12c 4139
1a40e23b
ZY		 4140 leaf = path->nodes[0];
4141 nritems = btrfs_header_nritems(leaf);
4142 if (path->slots[0] >= nritems) {
4143 ret = btrfs_next_leaf(extent_root, path);
4144 if (ret < 0)
4145 goto out;
4146 if (ret > 0)
4147 goto next;
4148 leaf = path->nodes[0];
4149 }
0ef3e66b 4150
1a40e23b
ZY		 4151 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4152 if (found_key.objectid == bytenr &&
f82d02d9
YZ		 4153 found_key.type == BTRFS_EXTENT_REF_KEY) {
4154 if (level < ref_path->shared_level)
4155 ref_path->shared_level = level;
1a40e23b 4156 goto found;
f82d02d9 4157 }
1a40e23b
ZY		 4158next:
4159 level--;
4160 btrfs_release_path(extent_root, path);
d899e052 4161 cond_resched();
1a40e23b
ZY		 4162 }
4163 /* reached lowest level */
4164 ret = 1;
4165 goto out;
4166walk_up:
4167 level = ref_path->current_level;
4168 while (level < BTRFS_MAX_LEVEL - 1) {
4169 u64 ref_objectid;
4170 if (level >= 0) {
4171 bytenr = ref_path->nodes[level];
4172 } else {
4173 bytenr = ref_path->extent_start;
4174 }
4175 BUG_ON(bytenr == 0);
edbd8d4e 4176
1a40e23b
ZY		 4177 key.objectid = bytenr;
4178 key.offset = 0;
4179 key.type = BTRFS_EXTENT_REF_KEY;
edbd8d4e 4180
1a40e23b
ZY		 4181 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
4182 if (ret < 0)
4183 goto out;
edbd8d4e 4184
1a40e23b
ZY		 4185 leaf = path->nodes[0];
4186 nritems = btrfs_header_nritems(leaf);
4187 if (path->slots[0] >= nritems) {
4188 ret = btrfs_next_leaf(extent_root, path);
4189 if (ret < 0)
4190 goto out;
4191 if (ret > 0) {
4192 /* the extent was freed by someone */
4193 if (ref_path->lowest_level == level)
4194 goto out;
4195 btrfs_release_path(extent_root, path);
4196 goto walk_down;
4197 }
4198 leaf = path->nodes[0];
4199 }
edbd8d4e 4200
1a40e23b
ZY		 4201 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4202 if (found_key.objectid != bytenr ||
4203 found_key.type != BTRFS_EXTENT_REF_KEY) {
4204 /* the extent was freed by someone */
4205 if (ref_path->lowest_level == level) {
4206 ret = 1;
4207 goto out;
4208 }
4209 btrfs_release_path(extent_root, path);
4210 goto walk_down;
4211 }
4212found:
4213 ref = btrfs_item_ptr(leaf, path->slots[0],
4214 struct btrfs_extent_ref);
4215 ref_objectid = btrfs_ref_objectid(leaf, ref);
4216 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4217 if (first_time) {
4218 level = (int)ref_objectid;
4219 BUG_ON(level >= BTRFS_MAX_LEVEL);
4220 ref_path->lowest_level = level;
4221 ref_path->current_level = level;
4222 ref_path->nodes[level] = bytenr;
4223 } else {
4224 WARN_ON(ref_objectid != level);
4225 }
4226 } else {
4227 WARN_ON(level != -1);
4228 }
4229 first_time = 0;
bf4ef679 4230
1a40e23b
ZY		 4231 if (ref_path->lowest_level == level) {
4232 ref_path->owner_objectid = ref_objectid;
1a40e23b
ZY		 4233 ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
4234 }
bf4ef679 4235
7d9eb12c 4236 /*
1a40e23b
ZY		 4237 * the block is tree root or the block isn't in reference
4238 * counted tree.
7d9eb12c 4239 */
1a40e23b
ZY		 4240 if (found_key.objectid == found_key.offset ||
4241 is_cowonly_root(btrfs_ref_root(leaf, ref))) {
4242 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
4243 ref_path->root_generation =
4244 btrfs_ref_generation(leaf, ref);
4245 if (level < 0) {
4246 /* special reference from the tree log */
4247 ref_path->nodes[0] = found_key.offset;
4248 ref_path->current_level = 0;
4249 }
4250 ret = 0;
4251 goto out;
4252 }
7d9eb12c 4253
1a40e23b
ZY		 4254 level++;
4255 BUG_ON(ref_path->nodes[level] != 0);
4256 ref_path->nodes[level] = found_key.offset;
4257 ref_path->current_level = level;
bf4ef679 4258
1a40e23b
ZY		 4259 /*
4260 * the reference was created in the running transaction,
4261 * no need to continue walking up.
4262 */
4263 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
4264 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
4265 ref_path->root_generation =
4266 btrfs_ref_generation(leaf, ref);
4267 ret = 0;
4268 goto out;
7d9eb12c
CM		 4269 }
4270
1a40e23b 4271 btrfs_release_path(extent_root, path);
d899e052 4272 cond_resched();
7d9eb12c 4273 }
1a40e23b
ZY		 4274 /* reached max tree level, but no tree root found. */
4275 BUG();
edbd8d4e 4276out:
1a40e23b
ZY		 4277 btrfs_free_path(path);
4278 return ret;
edbd8d4e
CM		 4279}
4280
1a40e23b
ZY		 4281static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
4282 struct btrfs_root *extent_root,
4283 struct btrfs_ref_path *ref_path,
4284 u64 extent_start)
a061fc8d 4285{
1a40e23b
ZY		 4286 memset(ref_path, 0, sizeof(*ref_path));
4287 ref_path->extent_start = extent_start;
a061fc8d 4288
1a40e23b 4289 return __next_ref_path(trans, extent_root, ref_path, 1);
a061fc8d
CM		 4290}
4291
1a40e23b
ZY		 4292static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
4293 struct btrfs_root *extent_root,
4294 struct btrfs_ref_path *ref_path)
edbd8d4e 4295{
1a40e23b
ZY		 4296 return __next_ref_path(trans, extent_root, ref_path, 0);
4297}
4298
4299static int noinline get_new_locations(struct inode *reloc_inode,
4300 struct btrfs_key *extent_key,
4301 u64 offset, int no_fragment,
4302 struct disk_extent **extents,
4303 int *nr_extents)
4304{
4305 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
4306 struct btrfs_path *path;
4307 struct btrfs_file_extent_item *fi;
edbd8d4e 4308 struct extent_buffer *leaf;
1a40e23b
ZY		 4309 struct disk_extent *exts = *extents;
4310 struct btrfs_key found_key;
4311 u64 cur_pos;
4312 u64 last_byte;
edbd8d4e 4313 u32 nritems;
1a40e23b
ZY		 4314 int nr = 0;
4315 int max = *nr_extents;
4316 int ret;
edbd8d4e 4317
1a40e23b
ZY		 4318 WARN_ON(!no_fragment && *extents);
4319 if (!exts) {
4320 max = 1;
4321 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
4322 if (!exts)
4323 return -ENOMEM;
a061fc8d 4324 }
edbd8d4e 4325
1a40e23b
ZY		 4326 path = btrfs_alloc_path();
4327 BUG_ON(!path);
edbd8d4e 4328
1a40e23b
ZY		 4329 cur_pos = extent_key->objectid - offset;
4330 last_byte = extent_key->objectid + extent_key->offset;
4331 ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
4332 cur_pos, 0);
4333 if (ret < 0)
4334 goto out;
4335 if (ret > 0) {
4336 ret = -ENOENT;
4337 goto out;
4338 }
edbd8d4e 4339
1a40e23b 4340 while (1) {
edbd8d4e
CM		 4341 leaf = path->nodes[0];
4342 nritems = btrfs_header_nritems(leaf);
1a40e23b
ZY		 4343 if (path->slots[0] >= nritems) {
4344 ret = btrfs_next_leaf(root, path);
a061fc8d
CM		 4345 if (ret < 0)
4346 goto out;
1a40e23b
ZY		 4347 if (ret > 0)
4348 break;
bf4ef679 4349 leaf = path->nodes[0];
a061fc8d 4350 }
edbd8d4e
CM		 4351
4352 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1a40e23b
ZY		 4353 if (found_key.offset != cur_pos ||
4354 found_key.type != BTRFS_EXTENT_DATA_KEY ||
4355 found_key.objectid != reloc_inode->i_ino)
edbd8d4e
CM		 4356 break;
4357
1a40e23b
ZY		 4358 fi = btrfs_item_ptr(leaf, path->slots[0],
4359 struct btrfs_file_extent_item);
4360 if (btrfs_file_extent_type(leaf, fi) !=
4361 BTRFS_FILE_EXTENT_REG ||
4362 btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
edbd8d4e 4363 break;
1a40e23b
ZY		 4364
4365 if (nr == max) {
4366 struct disk_extent *old = exts;
4367 max *= 2;
4368 exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
4369 memcpy(exts, old, sizeof(*exts) * nr);
4370 if (old != *extents)
4371 kfree(old);
a061fc8d 4372 }
edbd8d4e 4373
1a40e23b
ZY		 4374 exts[nr].disk_bytenr =
4375 btrfs_file_extent_disk_bytenr(leaf, fi);
4376 exts[nr].disk_num_bytes =
4377 btrfs_file_extent_disk_num_bytes(leaf, fi);
4378 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
4379 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
c8b97818
CM		 4380 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
4381 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
4382 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
4383 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
4384 fi);
d899e052
YZ		 4385 BUG_ON(exts[nr].offset > 0);
4386 BUG_ON(exts[nr].compression || exts[nr].encryption);
4387 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
edbd8d4e 4388
1a40e23b
ZY		 4389 cur_pos += exts[nr].num_bytes;
4390 nr++;
4391
4392 if (cur_pos + offset >= last_byte)
4393 break;
4394
4395 if (no_fragment) {
4396 ret = 1;
edbd8d4e 4397 goto out;
1a40e23b
ZY		 4398 }
4399 path->slots[0]++;
4400 }
4401
4402 WARN_ON(cur_pos + offset > last_byte);
4403 if (cur_pos + offset < last_byte) {
4404 ret = -ENOENT;
4405 goto out;
edbd8d4e
CM		 4406 }
4407 ret = 0;
4408out:
1a40e23b
ZY		 4409 btrfs_free_path(path);
4410 if (ret) {
4411 if (exts != *extents)
4412 kfree(exts);
4413 } else {
4414 *extents = exts;
4415 *nr_extents = nr;
4416 }
4417 return ret;
4418}
4419
4420static int noinline replace_one_extent(struct btrfs_trans_handle *trans,
4421 struct btrfs_root *root,
4422 struct btrfs_path *path,
4423 struct btrfs_key *extent_key,
4424 struct btrfs_key *leaf_key,
4425 struct btrfs_ref_path *ref_path,
4426 struct disk_extent *new_extents,
4427 int nr_extents)
4428{
4429 struct extent_buffer *leaf;
4430 struct btrfs_file_extent_item *fi;
4431 struct inode *inode = NULL;
4432 struct btrfs_key key;
4433 u64 lock_start = 0;
4434 u64 lock_end = 0;
4435 u64 num_bytes;
4436 u64 ext_offset;
4437 u64 first_pos;
4438 u32 nritems;
3bb1a1bc 4439 int nr_scaned = 0;
1a40e23b 4440 int extent_locked = 0;
d899e052 4441 int extent_type;
1a40e23b
ZY		 4442 int ret;
4443
3bb1a1bc
YZ		 4444 memcpy(&key, leaf_key, sizeof(key));
4445 first_pos = INT_LIMIT(loff_t) - extent_key->offset;
1a40e23b 4446 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
3bb1a1bc
YZ		 4447 if (key.objectid < ref_path->owner_objectid ||
4448 (key.objectid == ref_path->owner_objectid &&
4449 key.type < BTRFS_EXTENT_DATA_KEY)) {
4450 key.objectid = ref_path->owner_objectid;
4451 key.type = BTRFS_EXTENT_DATA_KEY;
4452 key.offset = 0;
4453 }
1a40e23b
ZY		 4454 }
4455
4456 while (1) {
4457 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
4458 if (ret < 0)
4459 goto out;
4460
4461 leaf = path->nodes[0];
4462 nritems = btrfs_header_nritems(leaf);
4463next:
4464 if (extent_locked && ret > 0) {
4465 /*
4466 * the file extent item was modified by someone
4467 * before the extent got locked.
4468 */
1a40e23b
ZY		 4469 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4470 lock_end, GFP_NOFS);
4471 extent_locked = 0;
4472 }
4473
4474 if (path->slots[0] >= nritems) {
3bb1a1bc 4475 if (++nr_scaned > 2)
1a40e23b
ZY		 4476 break;
4477
4478 BUG_ON(extent_locked);
4479 ret = btrfs_next_leaf(root, path);
4480 if (ret < 0)
4481 goto out;
4482 if (ret > 0)
4483 break;
4484 leaf = path->nodes[0];
4485 nritems = btrfs_header_nritems(leaf);
4486 }
4487
4488 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4489
4490 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
4491 if ((key.objectid > ref_path->owner_objectid) ||
4492 (key.objectid == ref_path->owner_objectid &&
4493 key.type > BTRFS_EXTENT_DATA_KEY) ||
4494 (key.offset >= first_pos + extent_key->offset))
4495 break;
4496 }
4497
4498 if (inode && key.objectid != inode->i_ino) {
4499 BUG_ON(extent_locked);
4500 btrfs_release_path(root, path);
4501 mutex_unlock(&inode->i_mutex);
4502 iput(inode);
4503 inode = NULL;
4504 continue;
4505 }
4506
4507 if (key.type != BTRFS_EXTENT_DATA_KEY) {
4508 path->slots[0]++;
4509 ret = 1;
4510 goto next;
4511 }
4512 fi = btrfs_item_ptr(leaf, path->slots[0],
4513 struct btrfs_file_extent_item);
d899e052
YZ		 4514 extent_type = btrfs_file_extent_type(leaf, fi);
4515 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
4516 extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
1a40e23b
ZY		 4517 (btrfs_file_extent_disk_bytenr(leaf, fi) !=
4518 extent_key->objectid)) {
4519 path->slots[0]++;
4520 ret = 1;
4521 goto next;
4522 }
4523
4524 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4525 ext_offset = btrfs_file_extent_offset(leaf, fi);
4526
4527 if (first_pos > key.offset - ext_offset)
4528 first_pos = key.offset - ext_offset;
4529
4530 if (!extent_locked) {
4531 lock_start = key.offset;
4532 lock_end = lock_start + num_bytes - 1;
4533 } else {
6643558d
YZ		 4534 if (lock_start > key.offset ||
4535 lock_end + 1 < key.offset + num_bytes) {
4536 unlock_extent(&BTRFS_I(inode)->io_tree,
4537 lock_start, lock_end, GFP_NOFS);
4538 extent_locked = 0;
4539 }
1a40e23b
ZY		 4540 }
4541
4542 if (!inode) {
4543 btrfs_release_path(root, path);
4544
4545 inode = btrfs_iget_locked(root->fs_info->sb,
4546 key.objectid, root);
4547 if (inode->i_state & I_NEW) {
4548 BTRFS_I(inode)->root = root;
4549 BTRFS_I(inode)->location.objectid =
4550 key.objectid;
4551 BTRFS_I(inode)->location.type =
4552 BTRFS_INODE_ITEM_KEY;
4553 BTRFS_I(inode)->location.offset = 0;
4554 btrfs_read_locked_inode(inode);
4555 unlock_new_inode(inode);
4556 }
4557 /*
4558 * some code call btrfs_commit_transaction while
4559 * holding the i_mutex, so we can't use mutex_lock
4560 * here.
4561 */
4562 if (is_bad_inode(inode) ||
4563 !mutex_trylock(&inode->i_mutex)) {
4564 iput(inode);
4565 inode = NULL;
4566 key.offset = (u64)-1;
4567 goto skip;
4568 }
4569 }
4570
4571 if (!extent_locked) {
4572 struct btrfs_ordered_extent *ordered;
4573
4574 btrfs_release_path(root, path);
4575
4576 lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4577 lock_end, GFP_NOFS);
4578 ordered = btrfs_lookup_first_ordered_extent(inode,
4579 lock_end);
4580 if (ordered &&
4581 ordered->file_offset <= lock_end &&
4582 ordered->file_offset + ordered->len > lock_start) {
4583 unlock_extent(&BTRFS_I(inode)->io_tree,
4584 lock_start, lock_end, GFP_NOFS);
4585 btrfs_start_ordered_extent(inode, ordered, 1);
4586 btrfs_put_ordered_extent(ordered);
4587 key.offset += num_bytes;
4588 goto skip;
4589 }
4590 if (ordered)
4591 btrfs_put_ordered_extent(ordered);
4592
1a40e23b
ZY		 4593 extent_locked = 1;
4594 continue;
4595 }
4596
4597 if (nr_extents == 1) {
4598 /* update extent pointer in place */
1a40e23b
ZY		 4599 btrfs_set_file_extent_disk_bytenr(leaf, fi,
4600 new_extents[0].disk_bytenr);
4601 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4602 new_extents[0].disk_num_bytes);
1a40e23b
ZY		 4603 btrfs_mark_buffer_dirty(leaf);
4604
4605 btrfs_drop_extent_cache(inode, key.offset,
4606 key.offset + num_bytes - 1, 0);
4607
4608 ret = btrfs_inc_extent_ref(trans, root,
4609 new_extents[0].disk_bytenr,
4610 new_extents[0].disk_num_bytes,
4611 leaf->start,
4612 root->root_key.objectid,
4613 trans->transid,
3bb1a1bc 4614 key.objectid);
1a40e23b
ZY		 4615 BUG_ON(ret);
4616
4617 ret = btrfs_free_extent(trans, root,
4618 extent_key->objectid,
4619 extent_key->offset,
4620 leaf->start,
4621 btrfs_header_owner(leaf),
4622 btrfs_header_generation(leaf),
3bb1a1bc 4623 key.objectid, 0);
1a40e23b
ZY		 4624 BUG_ON(ret);
4625
4626 btrfs_release_path(root, path);
4627 key.offset += num_bytes;
4628 } else {
d899e052
YZ		 4629 BUG_ON(1);
4630#if 0
1a40e23b
ZY		 4631 u64 alloc_hint;
4632 u64 extent_len;
4633 int i;
4634 /*
4635 * drop old extent pointer at first, then insert the
4636 * new pointers one bye one
4637 */
4638 btrfs_release_path(root, path);
4639 ret = btrfs_drop_extents(trans, root, inode, key.offset,
4640 key.offset + num_bytes,
4641 key.offset, &alloc_hint);
4642 BUG_ON(ret);
4643
4644 for (i = 0; i < nr_extents; i++) {
4645 if (ext_offset >= new_extents[i].num_bytes) {
4646 ext_offset -= new_extents[i].num_bytes;
4647 continue;
4648 }
4649 extent_len = min(new_extents[i].num_bytes -
4650 ext_offset, num_bytes);
4651
4652 ret = btrfs_insert_empty_item(trans, root,
4653 path, &key,
4654 sizeof(*fi));
4655 BUG_ON(ret);
4656
4657 leaf = path->nodes[0];
4658 fi = btrfs_item_ptr(leaf, path->slots[0],
4659 struct btrfs_file_extent_item);
4660 btrfs_set_file_extent_generation(leaf, fi,
4661 trans->transid);
4662 btrfs_set_file_extent_type(leaf, fi,
4663 BTRFS_FILE_EXTENT_REG);
4664 btrfs_set_file_extent_disk_bytenr(leaf, fi,
4665 new_extents[i].disk_bytenr);
4666 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4667 new_extents[i].disk_num_bytes);
c8b97818
CM		 4668 btrfs_set_file_extent_ram_bytes(leaf, fi,
4669 new_extents[i].ram_bytes);
4670
4671 btrfs_set_file_extent_compression(leaf, fi,
4672 new_extents[i].compression);
4673 btrfs_set_file_extent_encryption(leaf, fi,
4674 new_extents[i].encryption);
4675 btrfs_set_file_extent_other_encoding(leaf, fi,
4676 new_extents[i].other_encoding);
4677
1a40e23b
ZY		 4678 btrfs_set_file_extent_num_bytes(leaf, fi,
4679 extent_len);
4680 ext_offset += new_extents[i].offset;
4681 btrfs_set_file_extent_offset(leaf, fi,
4682 ext_offset);
4683 btrfs_mark_buffer_dirty(leaf);
4684
4685 btrfs_drop_extent_cache(inode, key.offset,
4686 key.offset + extent_len - 1, 0);
4687
4688 ret = btrfs_inc_extent_ref(trans, root,
4689 new_extents[i].disk_bytenr,
4690 new_extents[i].disk_num_bytes,
4691 leaf->start,
4692 root->root_key.objectid,
3bb1a1bc 4693 trans->transid, key.objectid);
1a40e23b
ZY		 4694 BUG_ON(ret);
4695 btrfs_release_path(root, path);
4696
a76a3cd4 4697 inode_add_bytes(inode, extent_len);
1a40e23b
ZY		 4698
4699 ext_offset = 0;
4700 num_bytes -= extent_len;
4701 key.offset += extent_len;
4702
4703 if (num_bytes == 0)
4704 break;
4705 }
4706 BUG_ON(i >= nr_extents);
d899e052 4707#endif
1a40e23b
ZY		 4708 }
4709
4710 if (extent_locked) {
1a40e23b
ZY		 4711 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4712 lock_end, GFP_NOFS);
4713 extent_locked = 0;
4714 }
4715skip:
4716 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
4717 key.offset >= first_pos + extent_key->offset)
4718 break;
4719
4720 cond_resched();
4721 }
4722 ret = 0;
4723out:
4724 btrfs_release_path(root, path);
4725 if (inode) {
4726 mutex_unlock(&inode->i_mutex);
4727 if (extent_locked) {
1a40e23b
ZY		 4728 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4729 lock_end, GFP_NOFS);
4730 }
4731 iput(inode);
4732 }
4733 return ret;
4734}
4735
1a40e23b
ZY		 4736int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
4737 struct btrfs_root *root,
4738 struct extent_buffer *buf, u64 orig_start)
4739{
4740 int level;
4741 int ret;
4742
4743 BUG_ON(btrfs_header_generation(buf) != trans->transid);
4744 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
4745
4746 level = btrfs_header_level(buf);
4747 if (level == 0) {
4748 struct btrfs_leaf_ref *ref;
4749 struct btrfs_leaf_ref *orig_ref;
4750
4751 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
4752 if (!orig_ref)
4753 return -ENOENT;
4754
4755 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
4756 if (!ref) {
4757 btrfs_free_leaf_ref(root, orig_ref);
4758 return -ENOMEM;
4759 }
4760
4761 ref->nritems = orig_ref->nritems;
4762 memcpy(ref->extents, orig_ref->extents,
4763 sizeof(ref->extents[0]) * ref->nritems);
4764
4765 btrfs_free_leaf_ref(root, orig_ref);
4766
4767 ref->root_gen = trans->transid;
4768 ref->bytenr = buf->start;
4769 ref->owner = btrfs_header_owner(buf);
4770 ref->generation = btrfs_header_generation(buf);
4771 ret = btrfs_add_leaf_ref(root, ref, 0);
4772 WARN_ON(ret);
4773 btrfs_free_leaf_ref(root, ref);
4774 }
4775 return 0;
4776}
4777
4778static int noinline invalidate_extent_cache(struct btrfs_root *root,
4779 struct extent_buffer *leaf,
4780 struct btrfs_block_group_cache *group,
4781 struct btrfs_root *target_root)
4782{
4783 struct btrfs_key key;
4784 struct inode *inode = NULL;
4785 struct btrfs_file_extent_item *fi;
4786 u64 num_bytes;
4787 u64 skip_objectid = 0;
4788 u32 nritems;
4789 u32 i;
4790
4791 nritems = btrfs_header_nritems(leaf);
4792 for (i = 0; i < nritems; i++) {
4793 btrfs_item_key_to_cpu(leaf, &key, i);
4794 if (key.objectid == skip_objectid ||
4795 key.type != BTRFS_EXTENT_DATA_KEY)
4796 continue;
4797 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4798 if (btrfs_file_extent_type(leaf, fi) ==
4799 BTRFS_FILE_EXTENT_INLINE)
4800 continue;
4801 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
4802 continue;
4803 if (!inode || inode->i_ino != key.objectid) {
4804 iput(inode);
4805 inode = btrfs_ilookup(target_root->fs_info->sb,
4806 key.objectid, target_root, 1);
4807 }
4808 if (!inode) {
4809 skip_objectid = key.objectid;
4810 continue;
4811 }
4812 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4813
4814 lock_extent(&BTRFS_I(inode)->io_tree, key.offset,
4815 key.offset + num_bytes - 1, GFP_NOFS);
1a40e23b
ZY		 4816 btrfs_drop_extent_cache(inode, key.offset,
4817 key.offset + num_bytes - 1, 1);
1a40e23b
ZY		 4818 unlock_extent(&BTRFS_I(inode)->io_tree, key.offset,
4819 key.offset + num_bytes - 1, GFP_NOFS);
4820 cond_resched();
4821 }
4822 iput(inode);
4823 return 0;
4824}
4825
4826static int noinline replace_extents_in_leaf(struct btrfs_trans_handle *trans,
4827 struct btrfs_root *root,
4828 struct extent_buffer *leaf,
4829 struct btrfs_block_group_cache *group,
4830 struct inode *reloc_inode)
4831{
4832 struct btrfs_key key;
4833 struct btrfs_key extent_key;
4834 struct btrfs_file_extent_item *fi;
4835 struct btrfs_leaf_ref *ref;
4836 struct disk_extent *new_extent;
4837 u64 bytenr;
4838 u64 num_bytes;
4839 u32 nritems;
4840 u32 i;
4841 int ext_index;
4842 int nr_extent;
4843 int ret;
4844
4845 new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
4846 BUG_ON(!new_extent);
4847
4848 ref = btrfs_lookup_leaf_ref(root, leaf->start);
4849 BUG_ON(!ref);
4850
4851 ext_index = -1;
4852 nritems = btrfs_header_nritems(leaf);
4853 for (i = 0; i < nritems; i++) {
4854 btrfs_item_key_to_cpu(leaf, &key, i);
4855 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
4856 continue;
4857 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4858 if (btrfs_file_extent_type(leaf, fi) ==
4859 BTRFS_FILE_EXTENT_INLINE)
4860 continue;
4861 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
4862 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
4863 if (bytenr == 0)
4864 continue;
4865
4866 ext_index++;
4867 if (bytenr >= group->key.objectid + group->key.offset ||
4868 bytenr + num_bytes <= group->key.objectid)
4869 continue;
4870
4871 extent_key.objectid = bytenr;
4872 extent_key.offset = num_bytes;
4873 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
4874 nr_extent = 1;
4875 ret = get_new_locations(reloc_inode, &extent_key,
4876 group->key.objectid, 1,
4877 &new_extent, &nr_extent);
4878 if (ret > 0)
4879 continue;
4880 BUG_ON(ret < 0);
4881
4882 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
4883 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
4884 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
4885 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
4886
1a40e23b
ZY		 4887 btrfs_set_file_extent_disk_bytenr(leaf, fi,
4888 new_extent->disk_bytenr);
4889 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4890 new_extent->disk_num_bytes);
1a40e23b
ZY		 4891 btrfs_mark_buffer_dirty(leaf);
4892
4893 ret = btrfs_inc_extent_ref(trans, root,
4894 new_extent->disk_bytenr,
4895 new_extent->disk_num_bytes,
4896 leaf->start,
4897 root->root_key.objectid,
3bb1a1bc 4898 trans->transid, key.objectid);
1a40e23b
ZY		 4899 BUG_ON(ret);
4900 ret = btrfs_free_extent(trans, root,
4901 bytenr, num_bytes, leaf->start,
4902 btrfs_header_owner(leaf),
4903 btrfs_header_generation(leaf),
3bb1a1bc 4904 key.objectid, 0);
1a40e23b
ZY		 4905 BUG_ON(ret);
4906 cond_resched();
4907 }
4908 kfree(new_extent);
4909 BUG_ON(ext_index + 1 != ref->nritems);
4910 btrfs_free_leaf_ref(root, ref);
4911 return 0;
4912}
4913
f82d02d9
YZ		 4914int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
4915 struct btrfs_root *root)
1a40e23b
ZY		 4916{
4917 struct btrfs_root *reloc_root;
f82d02d9 4918 int ret;
1a40e23b
ZY		 4919
4920 if (root->reloc_root) {
4921 reloc_root = root->reloc_root;
4922 root->reloc_root = NULL;
4923 list_add(&reloc_root->dead_list,
4924 &root->fs_info->dead_reloc_roots);
f82d02d9
YZ		 4925
4926 btrfs_set_root_bytenr(&reloc_root->root_item,
4927 reloc_root->node->start);
4928 btrfs_set_root_level(&root->root_item,
4929 btrfs_header_level(reloc_root->node));
4930 memset(&reloc_root->root_item.drop_progress, 0,
4931 sizeof(struct btrfs_disk_key));
4932 reloc_root->root_item.drop_level = 0;
4933
4934 ret = btrfs_update_root(trans, root->fs_info->tree_root,
4935 &reloc_root->root_key,
4936 &reloc_root->root_item);
4937 BUG_ON(ret);
1a40e23b
ZY		 4938 }
4939 return 0;
4940}
4941
4942int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
4943{
4944 struct btrfs_trans_handle *trans;
4945 struct btrfs_root *reloc_root;
4946 struct btrfs_root *prev_root = NULL;
4947 struct list_head dead_roots;
4948 int ret;
4949 unsigned long nr;
4950
4951 INIT_LIST_HEAD(&dead_roots);
4952 list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
4953
4954 while (!list_empty(&dead_roots)) {
4955 reloc_root = list_entry(dead_roots.prev,
4956 struct btrfs_root, dead_list);
4957 list_del_init(&reloc_root->dead_list);
4958
4959 BUG_ON(reloc_root->commit_root != NULL);
4960 while (1) {
4961 trans = btrfs_join_transaction(root, 1);
4962 BUG_ON(!trans);
4963
4964 mutex_lock(&root->fs_info->drop_mutex);
4965 ret = btrfs_drop_snapshot(trans, reloc_root);
4966 if (ret != -EAGAIN)
4967 break;
4968 mutex_unlock(&root->fs_info->drop_mutex);
4969
4970 nr = trans->blocks_used;
4971 ret = btrfs_end_transaction(trans, root);
4972 BUG_ON(ret);
4973 btrfs_btree_balance_dirty(root, nr);
4974 }
4975
4976 free_extent_buffer(reloc_root->node);
4977
4978 ret = btrfs_del_root(trans, root->fs_info->tree_root,
4979 &reloc_root->root_key);
4980 BUG_ON(ret);
4981 mutex_unlock(&root->fs_info->drop_mutex);
4982
4983 nr = trans->blocks_used;
4984 ret = btrfs_end_transaction(trans, root);
4985 BUG_ON(ret);
4986 btrfs_btree_balance_dirty(root, nr);
4987
4988 kfree(prev_root);
4989 prev_root = reloc_root;
4990 }
4991 if (prev_root) {
4992 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
4993 kfree(prev_root);
4994 }
4995 return 0;
4996}
4997
4998int btrfs_add_dead_reloc_root(struct btrfs_root *root)
4999{
5000 list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
5001 return 0;
5002}
5003
5004int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
5005{
5006 struct btrfs_root *reloc_root;
5007 struct btrfs_trans_handle *trans;
5008 struct btrfs_key location;
5009 int found;
5010 int ret;
5011
5012 mutex_lock(&root->fs_info->tree_reloc_mutex);
5013 ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
5014 BUG_ON(ret);
5015 found = !list_empty(&root->fs_info->dead_reloc_roots);
5016 mutex_unlock(&root->fs_info->tree_reloc_mutex);
5017
5018 if (found) {
5019 trans = btrfs_start_transaction(root, 1);
5020 BUG_ON(!trans);
5021 ret = btrfs_commit_transaction(trans, root);
5022 BUG_ON(ret);
5023 }
5024
5025 location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
5026 location.offset = (u64)-1;
5027 location.type = BTRFS_ROOT_ITEM_KEY;
5028
5029 reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
5030 BUG_ON(!reloc_root);
5031 btrfs_orphan_cleanup(reloc_root);
5032 return 0;
5033}
5034
5035static int noinline init_reloc_tree(struct btrfs_trans_handle *trans,
5036 struct btrfs_root *root)
5037{
5038 struct btrfs_root *reloc_root;
5039 struct extent_buffer *eb;
5040 struct btrfs_root_item *root_item;
5041 struct btrfs_key root_key;
5042 int ret;
5043
5044 BUG_ON(!root->ref_cows);
5045 if (root->reloc_root)
5046 return 0;
5047
5048 root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
5049 BUG_ON(!root_item);
5050
5051 ret = btrfs_copy_root(trans, root, root->commit_root,
5052 &eb, BTRFS_TREE_RELOC_OBJECTID);
5053 BUG_ON(ret);
5054
5055 root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
5056 root_key.offset = root->root_key.objectid;
5057 root_key.type = BTRFS_ROOT_ITEM_KEY;
5058
5059 memcpy(root_item, &root->root_item, sizeof(root_item));
5060 btrfs_set_root_refs(root_item, 0);
5061 btrfs_set_root_bytenr(root_item, eb->start);
5062 btrfs_set_root_level(root_item, btrfs_header_level(eb));
84234f3a 5063 btrfs_set_root_generation(root_item, trans->transid);
1a40e23b
ZY		 5064
5065 btrfs_tree_unlock(eb);
5066 free_extent_buffer(eb);
5067
5068 ret = btrfs_insert_root(trans, root->fs_info->tree_root,
5069 &root_key, root_item);
5070 BUG_ON(ret);
5071 kfree(root_item);
5072
5073 reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
5074 &root_key);
5075 BUG_ON(!reloc_root);
5076 reloc_root->last_trans = trans->transid;
5077 reloc_root->commit_root = NULL;
5078 reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
5079
5080 root->reloc_root = reloc_root;
5081 return 0;
5082}
5083
5084/*
5085 * Core function of space balance.
5086 *
5087 * The idea is using reloc trees to relocate tree blocks in reference
f82d02d9
YZ		 5088 * counted roots. There is one reloc tree for each subvol, and all
5089 * reloc trees share same root key objectid. Reloc trees are snapshots
5090 * of the latest committed roots of subvols (root->commit_root).
5091 *
5092 * To relocate a tree block referenced by a subvol, there are two steps.
5093 * COW the block through subvol's reloc tree, then update block pointer
5094 * in the subvol to point to the new block. Since all reloc trees share
5095 * same root key objectid, doing special handing for tree blocks owned
5096 * by them is easy. Once a tree block has been COWed in one reloc tree,
5097 * we can use the resulting new block directly when the same block is
5098 * required to COW again through other reloc trees. By this way, relocated
5099 * tree blocks are shared between reloc trees, so they are also shared
5100 * between subvols.
1a40e23b
ZY		 5101 */
5102static int noinline relocate_one_path(struct btrfs_trans_handle *trans,
5103 struct btrfs_root *root,
5104 struct btrfs_path *path,
5105 struct btrfs_key *first_key,
5106 struct btrfs_ref_path *ref_path,
5107 struct btrfs_block_group_cache *group,
5108 struct inode *reloc_inode)
5109{
5110 struct btrfs_root *reloc_root;
5111 struct extent_buffer *eb = NULL;
5112 struct btrfs_key *keys;
5113 u64 *nodes;
5114 int level;
f82d02d9 5115 int shared_level;
1a40e23b 5116 int lowest_level = 0;
1a40e23b
ZY		 5117 int ret;
5118
5119 if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
5120 lowest_level = ref_path->owner_objectid;
5121
f82d02d9 5122 if (!root->ref_cows) {
1a40e23b
ZY		 5123 path->lowest_level = lowest_level;
5124 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
5125 BUG_ON(ret < 0);
5126 path->lowest_level = 0;
5127 btrfs_release_path(root, path);
5128 return 0;
5129 }
5130
1a40e23b
ZY		 5131 mutex_lock(&root->fs_info->tree_reloc_mutex);
5132 ret = init_reloc_tree(trans, root);
5133 BUG_ON(ret);
5134 reloc_root = root->reloc_root;
5135
f82d02d9
YZ		 5136 shared_level = ref_path->shared_level;
5137 ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
1a40e23b 5138
f82d02d9
YZ		 5139 keys = ref_path->node_keys;
5140 nodes = ref_path->new_nodes;
5141 memset(&keys[shared_level + 1], 0,
5142 sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
5143 memset(&nodes[shared_level + 1], 0,
5144 sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
1a40e23b 5145
f82d02d9
YZ		 5146 if (nodes[lowest_level] == 0) {
5147 path->lowest_level = lowest_level;
5148 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
5149 0, 1);
5150 BUG_ON(ret);
5151 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
5152 eb = path->nodes[level];
5153 if (!eb || eb == reloc_root->node)
5154 break;
5155 nodes[level] = eb->start;
5156 if (level == 0)
5157 btrfs_item_key_to_cpu(eb, &keys[level], 0);
5158 else
5159 btrfs_node_key_to_cpu(eb, &keys[level], 0);
5160 }
2b82032c
YZ		 5161 if (nodes[0] &&
5162 ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
f82d02d9
YZ		 5163 eb = path->nodes[0];
5164 ret = replace_extents_in_leaf(trans, reloc_root, eb,
5165 group, reloc_inode);
5166 BUG_ON(ret);
5167 }
5168 btrfs_release_path(reloc_root, path);
5169 } else {
1a40e23b 5170 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
f82d02d9 5171 lowest_level);
1a40e23b
ZY		 5172 BUG_ON(ret);
5173 }
5174
1a40e23b
ZY		 5175 /*
5176 * replace tree blocks in the fs tree with tree blocks in
5177 * the reloc tree.
5178 */
5179 ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
5180 BUG_ON(ret < 0);
5181
5182 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
f82d02d9
YZ		 5183 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
5184 0, 0);
5185 BUG_ON(ret);
5186 extent_buffer_get(path->nodes[0]);
5187 eb = path->nodes[0];
5188 btrfs_release_path(reloc_root, path);
1a40e23b
ZY		 5189 ret = invalidate_extent_cache(reloc_root, eb, group, root);
5190 BUG_ON(ret);
5191 free_extent_buffer(eb);
5192 }
1a40e23b 5193
f82d02d9 5194 mutex_unlock(&root->fs_info->tree_reloc_mutex);
1a40e23b 5195 path->lowest_level = 0;
1a40e23b
ZY		 5196 return 0;
5197}
5198
5199static int noinline relocate_tree_block(struct btrfs_trans_handle *trans,
5200 struct btrfs_root *root,
5201 struct btrfs_path *path,
5202 struct btrfs_key *first_key,
5203 struct btrfs_ref_path *ref_path)
5204{
5205 int ret;
1a40e23b
ZY		 5206
5207 ret = relocate_one_path(trans, root, path, first_key,
5208 ref_path, NULL, NULL);
5209 BUG_ON(ret);
5210
5211 if (root == root->fs_info->extent_root)
5212 btrfs_extent_post_op(trans, root);
1a40e23b
ZY		 5213
5214 return 0;
5215}
5216
5217static int noinline del_extent_zero(struct btrfs_trans_handle *trans,
5218 struct btrfs_root *extent_root,
5219 struct btrfs_path *path,
5220 struct btrfs_key *extent_key)
5221{
5222 int ret;
5223
1a40e23b
ZY		 5224 ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
5225 if (ret)
5226 goto out;
5227 ret = btrfs_del_item(trans, extent_root, path);
5228out:
5229 btrfs_release_path(extent_root, path);
1a40e23b
ZY		 5230 return ret;
5231}
5232
5233static struct btrfs_root noinline *read_ref_root(struct btrfs_fs_info *fs_info,
5234 struct btrfs_ref_path *ref_path)
5235{
5236 struct btrfs_key root_key;
5237
5238 root_key.objectid = ref_path->root_objectid;
5239 root_key.type = BTRFS_ROOT_ITEM_KEY;
5240 if (is_cowonly_root(ref_path->root_objectid))
5241 root_key.offset = 0;
5242 else
5243 root_key.offset = (u64)-1;
5244
5245 return btrfs_read_fs_root_no_name(fs_info, &root_key);
5246}
5247
5248static int noinline relocate_one_extent(struct btrfs_root *extent_root,
5249 struct btrfs_path *path,
5250 struct btrfs_key *extent_key,
5251 struct btrfs_block_group_cache *group,
5252 struct inode *reloc_inode, int pass)
5253{
5254 struct btrfs_trans_handle *trans;
5255 struct btrfs_root *found_root;
5256 struct btrfs_ref_path *ref_path = NULL;
5257 struct disk_extent *new_extents = NULL;
5258 int nr_extents = 0;
5259 int loops;
5260 int ret;
5261 int level;
5262 struct btrfs_key first_key;
5263 u64 prev_block = 0;
5264
1a40e23b
ZY		 5265
5266 trans = btrfs_start_transaction(extent_root, 1);
5267 BUG_ON(!trans);
5268
5269 if (extent_key->objectid == 0) {
5270 ret = del_extent_zero(trans, extent_root, path, extent_key);
5271 goto out;
5272 }
5273
5274 ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
5275 if (!ref_path) {
5276 ret = -ENOMEM;
5277 goto out;
5278 }
5279
5280 for (loops = 0; ; loops++) {
5281 if (loops == 0) {
5282 ret = btrfs_first_ref_path(trans, extent_root, ref_path,
5283 extent_key->objectid);
5284 } else {
5285 ret = btrfs_next_ref_path(trans, extent_root, ref_path);
5286 }
5287 if (ret < 0)
5288 goto out;
5289 if (ret > 0)
5290 break;
5291
5292 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
5293 ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
5294 continue;
5295
5296 found_root = read_ref_root(extent_root->fs_info, ref_path);
5297 BUG_ON(!found_root);
5298 /*
5299 * for reference counted tree, only process reference paths
5300 * rooted at the latest committed root.
5301 */
5302 if (found_root->ref_cows &&
5303 ref_path->root_generation != found_root->root_key.offset)
5304 continue;
5305
5306 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5307 if (pass == 0) {
5308 /*
5309 * copy data extents to new locations
5310 */
5311 u64 group_start = group->key.objectid;
5312 ret = relocate_data_extent(reloc_inode,
5313 extent_key,
5314 group_start);
5315 if (ret < 0)
5316 goto out;
5317 break;
5318 }
5319 level = 0;
5320 } else {
5321 level = ref_path->owner_objectid;
5322 }
5323
5324 if (prev_block != ref_path->nodes[level]) {
5325 struct extent_buffer *eb;
5326 u64 block_start = ref_path->nodes[level];
5327 u64 block_size = btrfs_level_size(found_root, level);
5328
5329 eb = read_tree_block(found_root, block_start,
5330 block_size, 0);
5331 btrfs_tree_lock(eb);
5332 BUG_ON(level != btrfs_header_level(eb));
5333
5334 if (level == 0)
5335 btrfs_item_key_to_cpu(eb, &first_key, 0);
5336 else
5337 btrfs_node_key_to_cpu(eb, &first_key, 0);
5338
5339 btrfs_tree_unlock(eb);
5340 free_extent_buffer(eb);
5341 prev_block = block_start;
5342 }
5343
5344 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID &&
5345 pass >= 2) {
5346 /*
5347 * use fallback method to process the remaining
5348 * references.
5349 */
5350 if (!new_extents) {
5351 u64 group_start = group->key.objectid;
d899e052
YZ		 5352 new_extents = kmalloc(sizeof(*new_extents),
5353 GFP_NOFS);
5354 nr_extents = 1;
1a40e23b
ZY		 5355 ret = get_new_locations(reloc_inode,
5356 extent_key,
d899e052 5357 group_start, 1,
1a40e23b
ZY		 5358 &new_extents,
5359 &nr_extents);
d899e052 5360 if (ret)
1a40e23b
ZY		 5361 goto out;
5362 }
5363 btrfs_record_root_in_trans(found_root);
5364 ret = replace_one_extent(trans, found_root,
5365 path, extent_key,
5366 &first_key, ref_path,
5367 new_extents, nr_extents);
5368 if (ret < 0)
5369 goto out;
5370 continue;
5371 }
5372
5373 btrfs_record_root_in_trans(found_root);
5374 if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
5375 ret = relocate_tree_block(trans, found_root, path,
5376 &first_key, ref_path);
5377 } else {
5378 /*
5379 * try to update data extent references while
5380 * keeping metadata shared between snapshots.
5381 */
5382 ret = relocate_one_path(trans, found_root, path,
5383 &first_key, ref_path,
5384 group, reloc_inode);
5385 }
5386 if (ret < 0)
5387 goto out;
5388 }
5389 ret = 0;
5390out:
5391 btrfs_end_transaction(trans, extent_root);
5392 kfree(new_extents);
5393 kfree(ref_path);
1a40e23b
ZY		 5394 return ret;
5395}
5396
ec44a35c
CM		 5397static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
5398{
5399 u64 num_devices;
5400 u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
5401 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
5402
2b82032c 5403 num_devices = root->fs_info->fs_devices->rw_devices;
ec44a35c
CM		 5404 if (num_devices == 1) {
5405 stripped |= BTRFS_BLOCK_GROUP_DUP;
5406 stripped = flags & ~stripped;
5407
5408 /* turn raid0 into single device chunks */
5409 if (flags & BTRFS_BLOCK_GROUP_RAID0)
5410 return stripped;
5411
5412 /* turn mirroring into duplication */
5413 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
5414 BTRFS_BLOCK_GROUP_RAID10))
5415 return stripped | BTRFS_BLOCK_GROUP_DUP;
5416 return flags;
5417 } else {
5418 /* they already had raid on here, just return */
ec44a35c
CM		 5419 if (flags & stripped)
5420 return flags;
5421
5422 stripped |= BTRFS_BLOCK_GROUP_DUP;
5423 stripped = flags & ~stripped;
5424
5425 /* switch duplicated blocks with raid1 */
5426 if (flags & BTRFS_BLOCK_GROUP_DUP)
5427 return stripped | BTRFS_BLOCK_GROUP_RAID1;
5428
5429 /* turn single device chunks into raid0 */
5430 return stripped | BTRFS_BLOCK_GROUP_RAID0;
5431 }
5432 return flags;
5433}
5434
0ef3e66b
CM		 5435int __alloc_chunk_for_shrink(struct btrfs_root *root,
5436 struct btrfs_block_group_cache *shrink_block_group,
5437 int force)
5438{
5439 struct btrfs_trans_handle *trans;
5440 u64 new_alloc_flags;
5441 u64 calc;
5442
c286ac48 5443 spin_lock(&shrink_block_group->lock);
0ef3e66b 5444 if (btrfs_block_group_used(&shrink_block_group->item) > 0) {
c286ac48 5445 spin_unlock(&shrink_block_group->lock);
c286ac48 5446
0ef3e66b 5447 trans = btrfs_start_transaction(root, 1);
c286ac48 5448 spin_lock(&shrink_block_group->lock);
7d9eb12c 5449
0ef3e66b
CM		 5450 new_alloc_flags = update_block_group_flags(root,
5451 shrink_block_group->flags);
5452 if (new_alloc_flags != shrink_block_group->flags) {
5453 calc =
5454 btrfs_block_group_used(&shrink_block_group->item);
5455 } else {
5456 calc = shrink_block_group->key.offset;
5457 }
c286ac48
CM		 5458 spin_unlock(&shrink_block_group->lock);
5459
0ef3e66b
CM		 5460 do_chunk_alloc(trans, root->fs_info->extent_root,
5461 calc + 2 * 1024 * 1024, new_alloc_flags, force);
7d9eb12c 5462
0ef3e66b 5463 btrfs_end_transaction(trans, root);
c286ac48
CM		 5464 } else
5465 spin_unlock(&shrink_block_group->lock);
0ef3e66b
CM		 5466 return 0;
5467}
5468
1a40e23b
ZY		 5469static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
5470 struct btrfs_root *root,
5471 u64 objectid, u64 size)
5472{
5473 struct btrfs_path *path;
5474 struct btrfs_inode_item *item;
5475 struct extent_buffer *leaf;
5476 int ret;
5477
5478 path = btrfs_alloc_path();
5479 if (!path)
5480 return -ENOMEM;
5481
5482 ret = btrfs_insert_empty_inode(trans, root, path, objectid);
5483 if (ret)
5484 goto out;
5485
5486 leaf = path->nodes[0];
5487 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
5488 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
5489 btrfs_set_inode_generation(leaf, item, 1);
5490 btrfs_set_inode_size(leaf, item, size);
5491 btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
d899e052
YZ		 5492 btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NODATASUM |
5493 BTRFS_INODE_NOCOMPRESS);
1a40e23b
ZY		 5494 btrfs_mark_buffer_dirty(leaf);
5495 btrfs_release_path(root, path);
5496out:
5497 btrfs_free_path(path);
5498 return ret;
5499}
5500
5501static struct inode noinline *create_reloc_inode(struct btrfs_fs_info *fs_info,
5502 struct btrfs_block_group_cache *group)
5503{
5504 struct inode *inode = NULL;
5505 struct btrfs_trans_handle *trans;
5506 struct btrfs_root *root;
5507 struct btrfs_key root_key;
5508 u64 objectid = BTRFS_FIRST_FREE_OBJECTID;
5509 int err = 0;
5510
5511 root_key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
5512 root_key.type = BTRFS_ROOT_ITEM_KEY;
5513 root_key.offset = (u64)-1;
5514 root = btrfs_read_fs_root_no_name(fs_info, &root_key);
5515 if (IS_ERR(root))
5516 return ERR_CAST(root);
5517
5518 trans = btrfs_start_transaction(root, 1);
5519 BUG_ON(!trans);
5520
5521 err = btrfs_find_free_objectid(trans, root, objectid, &objectid);
5522 if (err)
5523 goto out;
5524
5525 err = __insert_orphan_inode(trans, root, objectid, group->key.offset);
5526 BUG_ON(err);
5527
5528 err = btrfs_insert_file_extent(trans, root, objectid, 0, 0, 0,
c8b97818
CM		 5529 group->key.offset, 0, group->key.offset,
5530 0, 0, 0);
1a40e23b
ZY		 5531 BUG_ON(err);
5532
5533 inode = btrfs_iget_locked(root->fs_info->sb, objectid, root);
5534 if (inode->i_state & I_NEW) {
5535 BTRFS_I(inode)->root = root;
5536 BTRFS_I(inode)->location.objectid = objectid;
5537 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
5538 BTRFS_I(inode)->location.offset = 0;
5539 btrfs_read_locked_inode(inode);
5540 unlock_new_inode(inode);
5541 BUG_ON(is_bad_inode(inode));
5542 } else {
5543 BUG_ON(1);
5544 }
5545
5546 err = btrfs_orphan_add(trans, inode);
5547out:
5548 btrfs_end_transaction(trans, root);
5549 if (err) {
5550 if (inode)
5551 iput(inode);
5552 inode = ERR_PTR(err);
5553 }
5554 return inode;
5555}
5556
5557int btrfs_relocate_block_group(struct btrfs_root *root, u64 group_start)
edbd8d4e
CM		 5558{
5559 struct btrfs_trans_handle *trans;
edbd8d4e 5560 struct btrfs_path *path;
1a40e23b
ZY		 5561 struct btrfs_fs_info *info = root->fs_info;
5562 struct extent_buffer *leaf;
5563 struct inode *reloc_inode;
5564 struct btrfs_block_group_cache *block_group;
5565 struct btrfs_key key;
d899e052 5566 u64 skipped;
edbd8d4e
CM		 5567 u64 cur_byte;
5568 u64 total_found;
edbd8d4e
CM		 5569 u32 nritems;
5570 int ret;
a061fc8d 5571 int progress;
1a40e23b 5572 int pass = 0;
edbd8d4e 5573
1a40e23b
ZY		 5574 root = root->fs_info->extent_root;
5575
5576 block_group = btrfs_lookup_block_group(info, group_start);
5577 BUG_ON(!block_group);
8f18cf13 5578
1a40e23b
ZY		 5579 printk("btrfs relocating block group %llu flags %llu\n",
5580 (unsigned long long)block_group->key.objectid,
5581 (unsigned long long)block_group->flags);
8f18cf13 5582
edbd8d4e 5583 path = btrfs_alloc_path();
1a40e23b 5584 BUG_ON(!path);
edbd8d4e 5585
1a40e23b
ZY		 5586 reloc_inode = create_reloc_inode(info, block_group);
5587 BUG_ON(IS_ERR(reloc_inode));
323da79c 5588
1a40e23b 5589 __alloc_chunk_for_shrink(root, block_group, 1);
c146afad 5590 set_block_group_readonly(block_group);
323da79c 5591
1a40e23b
ZY		 5592 btrfs_start_delalloc_inodes(info->tree_root);
5593 btrfs_wait_ordered_extents(info->tree_root, 0);
5594again:
d899e052 5595 skipped = 0;
edbd8d4e 5596 total_found = 0;
a061fc8d 5597 progress = 0;
1a40e23b 5598 key.objectid = block_group->key.objectid;
edbd8d4e
CM		 5599 key.offset = 0;
5600 key.type = 0;
73e48b27 5601 cur_byte = key.objectid;
4313b399 5602
1a40e23b
ZY		 5603 trans = btrfs_start_transaction(info->tree_root, 1);
5604 btrfs_commit_transaction(trans, info->tree_root);
ea8c2819 5605
1a40e23b
ZY		 5606 mutex_lock(&root->fs_info->cleaner_mutex);
5607 btrfs_clean_old_snapshots(info->tree_root);
5608 btrfs_remove_leaf_refs(info->tree_root, (u64)-1, 1);
5609 mutex_unlock(&root->fs_info->cleaner_mutex);
ea8c2819 5610
73e48b27 5611 while(1) {
edbd8d4e
CM		 5612 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5613 if (ret < 0)
5614 goto out;
7d9eb12c 5615next:
edbd8d4e 5616 leaf = path->nodes[0];
73e48b27 5617 nritems = btrfs_header_nritems(leaf);
73e48b27
Y		 5618 if (path->slots[0] >= nritems) {
5619 ret = btrfs_next_leaf(root, path);
5620 if (ret < 0)
5621 goto out;
5622 if (ret == 1) {
5623 ret = 0;
5624 break;
edbd8d4e 5625 }
73e48b27
Y		 5626 leaf = path->nodes[0];
5627 nritems = btrfs_header_nritems(leaf);
edbd8d4e 5628 }
73e48b27 5629
1a40e23b 5630 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
725c8463 5631
1a40e23b
ZY		 5632 if (key.objectid >= block_group->key.objectid +
5633 block_group->key.offset)
8f18cf13
CM		 5634 break;
5635
725c8463 5636 if (progress && need_resched()) {
725c8463 5637 btrfs_release_path(root, path);
1a40e23b 5638 cond_resched();
725c8463 5639 progress = 0;
1a40e23b 5640 continue;
725c8463
CM		 5641 }
5642 progress = 1;
5643
1a40e23b
ZY		 5644 if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY ||
5645 key.objectid + key.offset <= cur_byte) {
edbd8d4e 5646 path->slots[0]++;
edbd8d4e
CM		 5647 goto next;
5648 }
73e48b27 5649
edbd8d4e 5650 total_found++;
1a40e23b 5651 cur_byte = key.objectid + key.offset;
edbd8d4e 5652 btrfs_release_path(root, path);
edbd8d4e 5653
1a40e23b
ZY		 5654 __alloc_chunk_for_shrink(root, block_group, 0);
5655 ret = relocate_one_extent(root, path, &key, block_group,
5656 reloc_inode, pass);
5657 BUG_ON(ret < 0);
d899e052
YZ		 5658 if (ret > 0)
5659 skipped++;
edbd8d4e 5660
1a40e23b
ZY		 5661 key.objectid = cur_byte;
5662 key.type = 0;
5663 key.offset = 0;
edbd8d4e
CM		 5664 }
5665
1a40e23b 5666 btrfs_release_path(root, path);
7d9eb12c 5667
1a40e23b
ZY		 5668 if (pass == 0) {
5669 btrfs_wait_ordered_range(reloc_inode, 0, (u64)-1);
5670 invalidate_mapping_pages(reloc_inode->i_mapping, 0, -1);
5671 WARN_ON(reloc_inode->i_mapping->nrpages);
8e8a1e31 5672 }
73e48b27 5673
1a40e23b
ZY		 5674 if (total_found > 0) {
5675 printk("btrfs found %llu extents in pass %d\n",
5676 (unsigned long long)total_found, pass);
5677 pass++;
d899e052
YZ		 5678 if (total_found == skipped && pass > 2) {
5679 iput(reloc_inode);
5680 reloc_inode = create_reloc_inode(info, block_group);
5681 pass = 0;
5682 }
1a40e23b 5683 goto again;
0f9dd46c 5684 }
0ef3e66b 5685
1a40e23b
ZY		 5686 /* delete reloc_inode */
5687 iput(reloc_inode);
5688
5689 /* unpin extents in this range */
5690 trans = btrfs_start_transaction(info->tree_root, 1);
5691 btrfs_commit_transaction(trans, info->tree_root);
0ef3e66b 5692
1a40e23b
ZY		 5693 spin_lock(&block_group->lock);
5694 WARN_ON(block_group->pinned > 0);
5695 WARN_ON(block_group->reserved > 0);
5696 WARN_ON(btrfs_block_group_used(&block_group->item) > 0);
5697 spin_unlock(&block_group->lock);
5698 ret = 0;
edbd8d4e 5699out:
1a40e23b 5700 btrfs_free_path(path);
edbd8d4e
CM		 5701 return ret;
5702}
5703
0b86a832
CM		 5704int find_first_block_group(struct btrfs_root *root, struct btrfs_path *path,
5705 struct btrfs_key *key)
5706{
925baedd 5707 int ret = 0;
0b86a832
CM		 5708 struct btrfs_key found_key;
5709 struct extent_buffer *leaf;
5710 int slot;
edbd8d4e 5711
0b86a832
CM		 5712 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
5713 if (ret < 0)
925baedd
CM		 5714 goto out;
5715
0b86a832
CM		 5716 while(1) {
5717 slot = path->slots[0];
edbd8d4e 5718 leaf = path->nodes[0];
0b86a832
CM		 5719 if (slot >= btrfs_header_nritems(leaf)) {
5720 ret = btrfs_next_leaf(root, path);
5721 if (ret == 0)
5722 continue;
5723 if (ret < 0)
925baedd 5724 goto out;
0b86a832 5725 break;
edbd8d4e 5726 }
0b86a832 5727 btrfs_item_key_to_cpu(leaf, &found_key, slot);
edbd8d4e 5728
0b86a832 5729 if (found_key.objectid >= key->objectid &&
925baedd
CM		 5730 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
5731 ret = 0;
5732 goto out;
5733 }
0b86a832 5734 path->slots[0]++;
edbd8d4e 5735 }
0b86a832 5736 ret = -ENOENT;
925baedd 5737out:
0b86a832 5738 return ret;
edbd8d4e
CM		 5739}
5740
1a40e23b
ZY		 5741int btrfs_free_block_groups(struct btrfs_fs_info *info)
5742{
5743 struct btrfs_block_group_cache *block_group;
5744 struct rb_node *n;
5745
1a40e23b
ZY		 5746 spin_lock(&info->block_group_cache_lock);
5747 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
5748 block_group = rb_entry(n, struct btrfs_block_group_cache,
5749 cache_node);
1a40e23b
ZY		 5750 rb_erase(&block_group->cache_node,
5751 &info->block_group_cache_tree);
d899e052
YZ		 5752 spin_unlock(&info->block_group_cache_lock);
5753
5754 btrfs_remove_free_space_cache(block_group);
80eb234a 5755 down_write(&block_group->space_info->groups_sem);
1a40e23b 5756 list_del(&block_group->list);
80eb234a 5757 up_write(&block_group->space_info->groups_sem);
1a40e23b 5758 kfree(block_group);
d899e052
YZ		 5759
5760 spin_lock(&info->block_group_cache_lock);
1a40e23b
ZY		 5761 }
5762 spin_unlock(&info->block_group_cache_lock);
1a40e23b
ZY		 5763 return 0;
5764}
5765
9078a3e1
CM		 5766int btrfs_read_block_groups(struct btrfs_root *root)
5767{
5768 struct btrfs_path *path;
5769 int ret;
9078a3e1 5770 struct btrfs_block_group_cache *cache;
be744175 5771 struct btrfs_fs_info *info = root->fs_info;
6324fbf3 5772 struct btrfs_space_info *space_info;
9078a3e1
CM		 5773 struct btrfs_key key;
5774 struct btrfs_key found_key;
5f39d397 5775 struct extent_buffer *leaf;
96b5179d 5776
be744175 5777 root = info->extent_root;
9078a3e1 5778 key.objectid = 0;
0b86a832 5779 key.offset = 0;
9078a3e1 5780 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
9078a3e1
CM		 5781 path = btrfs_alloc_path();
5782 if (!path)
5783 return -ENOMEM;
5784
5785 while(1) {
0b86a832
CM		 5786 ret = find_first_block_group(root, path, &key);
5787 if (ret > 0) {
5788 ret = 0;
5789 goto error;
9078a3e1 5790 }
0b86a832
CM		 5791 if (ret != 0)
5792 goto error;
5793
5f39d397
CM		 5794 leaf = path->nodes[0];
5795 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
8f18cf13 5796 cache = kzalloc(sizeof(*cache), GFP_NOFS);
9078a3e1 5797 if (!cache) {
0b86a832 5798 ret = -ENOMEM;
9078a3e1
CM		 5799 break;
5800 }
3e1ad54f 5801
c286ac48 5802 spin_lock_init(&cache->lock);
25179201 5803 mutex_init(&cache->alloc_mutex);
0f9dd46c 5804 INIT_LIST_HEAD(&cache->list);
5f39d397
CM		 5805 read_extent_buffer(leaf, &cache->item,
5806 btrfs_item_ptr_offset(leaf, path->slots[0]),
5807 sizeof(cache->item));
9078a3e1 5808 memcpy(&cache->key, &found_key, sizeof(found_key));
0b86a832 5809
9078a3e1
CM		 5810 key.objectid = found_key.objectid + found_key.offset;
5811 btrfs_release_path(root, path);
0b86a832 5812 cache->flags = btrfs_block_group_flags(&cache->item);
96b5179d 5813
6324fbf3
CM		 5814 ret = update_space_info(info, cache->flags, found_key.offset,
5815 btrfs_block_group_used(&cache->item),
5816 &space_info);
5817 BUG_ON(ret);
5818 cache->space_info = space_info;
80eb234a
JB		 5819 down_write(&space_info->groups_sem);
5820 list_add_tail(&cache->list, &space_info->block_groups);
5821 up_write(&space_info->groups_sem);
0f9dd46c
JB		 5822
5823 ret = btrfs_add_block_group_cache(root->fs_info, cache);
5824 BUG_ON(ret);
75ccf47d
CM		 5825
5826 set_avail_alloc_bits(root->fs_info, cache->flags);
2b82032c
YZ		 5827 if (btrfs_chunk_readonly(root, cache->key.objectid))
5828 set_block_group_readonly(cache);
9078a3e1 5829 }
0b86a832
CM		 5830 ret = 0;
5831error:
9078a3e1 5832 btrfs_free_path(path);
0b86a832 5833 return ret;
9078a3e1 5834}
6324fbf3
CM		 5835
5836int btrfs_make_block_group(struct btrfs_trans_handle *trans,
5837 struct btrfs_root *root, u64 bytes_used,
e17cade2 5838 u64 type, u64 chunk_objectid, u64 chunk_offset,
6324fbf3
CM		 5839 u64 size)
5840{
5841 int ret;
6324fbf3
CM		 5842 struct btrfs_root *extent_root;
5843 struct btrfs_block_group_cache *cache;
6324fbf3
CM		 5844
5845 extent_root = root->fs_info->extent_root;
6324fbf3 5846
e02119d5
CM		 5847 root->fs_info->last_trans_new_blockgroup = trans->transid;
5848
8f18cf13 5849 cache = kzalloc(sizeof(*cache), GFP_NOFS);
0f9dd46c
JB		 5850 if (!cache)
5851 return -ENOMEM;
5852
e17cade2 5853 cache->key.objectid = chunk_offset;
6324fbf3 5854 cache->key.offset = size;
c286ac48 5855 spin_lock_init(&cache->lock);
25179201 5856 mutex_init(&cache->alloc_mutex);
0f9dd46c 5857 INIT_LIST_HEAD(&cache->list);
6324fbf3 5858 btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY);
0ef3e66b 5859
6324fbf3 5860 btrfs_set_block_group_used(&cache->item, bytes_used);
6324fbf3
CM		 5861 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
5862 cache->flags = type;
5863 btrfs_set_block_group_flags(&cache->item, type);
5864
5865 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
5866 &cache->space_info);
5867 BUG_ON(ret);
80eb234a
JB		 5868 down_write(&cache->space_info->groups_sem);
5869 list_add_tail(&cache->list, &cache->space_info->block_groups);
5870 up_write(&cache->space_info->groups_sem);
6324fbf3 5871
0f9dd46c
JB		 5872 ret = btrfs_add_block_group_cache(root->fs_info, cache);
5873 BUG_ON(ret);
c286ac48 5874
6324fbf3
CM		 5875 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
5876 sizeof(cache->item));
5877 BUG_ON(ret);
5878
87ef2bb4
CM		 5879 finish_current_insert(trans, extent_root, 0);
5880 ret = del_pending_extents(trans, extent_root, 0);
6324fbf3 5881 BUG_ON(ret);
d18a2c44 5882 set_avail_alloc_bits(extent_root->fs_info, type);
925baedd 5883
6324fbf3
CM		 5884 return 0;
5885}
1a40e23b
ZY		 5886
5887int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
5888 struct btrfs_root *root, u64 group_start)
5889{
5890 struct btrfs_path *path;
5891 struct btrfs_block_group_cache *block_group;
5892 struct btrfs_key key;
5893 int ret;
5894
1a40e23b
ZY		 5895 root = root->fs_info->extent_root;
5896
5897 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
5898 BUG_ON(!block_group);
c146afad 5899 BUG_ON(!block_group->ro);
1a40e23b
ZY		 5900
5901 memcpy(&key, &block_group->key, sizeof(key));
5902
5903 path = btrfs_alloc_path();
5904 BUG_ON(!path);
5905
5906 btrfs_remove_free_space_cache(block_group);
5907 rb_erase(&block_group->cache_node,
5908 &root->fs_info->block_group_cache_tree);
80eb234a 5909 down_write(&block_group->space_info->groups_sem);
1a40e23b 5910 list_del(&block_group->list);
80eb234a 5911 up_write(&block_group->space_info->groups_sem);
1a40e23b 5912
c146afad
YZ		 5913 spin_lock(&block_group->space_info->lock);
5914 block_group->space_info->total_bytes -= block_group->key.offset;
5915 block_group->space_info->bytes_readonly -= block_group->key.offset;
5916 spin_unlock(&block_group->space_info->lock);
2b82032c 5917 block_group->space_info->full = 0;
c146afad 5918
1a40e23b
ZY		 5919 /*
5920 memset(shrink_block_group, 0, sizeof(*shrink_block_group));
5921 kfree(shrink_block_group);
5922 */
5923
5924 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
5925 if (ret > 0)
5926 ret = -EIO;
5927 if (ret < 0)
5928 goto out;
5929
5930 ret = btrfs_del_item(trans, root, path);
5931out:
5932 btrfs_free_path(path);
5933 return ret;
5934}
Linux 6.x block layer and io_uring tree(s)