projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Btrfs: Fix bookend extent race v2
[linux-2.6-block.git] / fs / btrfs / extent-tree.c
CommitLineData
6cbd5570
CM		 1/*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
ec6b910f 18#include <linux/sched.h>
edbd8d4e 19#include <linux/pagemap.h>
ec44a35c 20#include <linux/writeback.h>
21af804c 21#include <linux/blkdev.h>
74493f7a 22#include "hash.h"
a5eb62e3 23#include "crc32c.h"
fec577fb
CM		 24#include "ctree.h"
25#include "disk-io.h"
26#include "print-tree.h"
e089f05c 27#include "transaction.h"
0b86a832 28#include "volumes.h"
925baedd 29#include "locking.h"
31153d81 30#include "ref-cache.h"
fec577fb 31
31840ae1
ZY		 32#define PENDING_EXTENT_INSERT 0
33#define PENDING_EXTENT_DELETE 1
34#define PENDING_BACKREF_UPDATE 2
35
36struct pending_extent_op {
37 int type;
38 u64 bytenr;
39 u64 num_bytes;
40 u64 parent;
41 u64 orig_parent;
42 u64 generation;
43 u64 orig_generation;
44 int level;
45};
46
e089f05c 47static int finish_current_insert(struct btrfs_trans_handle *trans, struct
87ef2bb4 48 btrfs_root *extent_root, int all);
e20d96d6 49static int del_pending_extents(struct btrfs_trans_handle *trans, struct
87ef2bb4 50 btrfs_root *extent_root, int all);
925baedd
CM		 51static struct btrfs_block_group_cache *
52__btrfs_find_block_group(struct btrfs_root *root,
53 struct btrfs_block_group_cache *hint,
54 u64 search_start, int data, int owner);
d548ee51 55
0f9dd46c
JB		 56static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
57{
58 return (cache->flags & bits) == bits;
59}
60
61/*
62 * this adds the block group to the fs_info rb tree for the block group
63 * cache
64 */
65int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
66 struct btrfs_block_group_cache *block_group)
67{
68 struct rb_node **p;
69 struct rb_node *parent = NULL;
70 struct btrfs_block_group_cache *cache;
71
72 spin_lock(&info->block_group_cache_lock);
73 p = &info->block_group_cache_tree.rb_node;
74
75 while (*p) {
76 parent = *p;
77 cache = rb_entry(parent, struct btrfs_block_group_cache,
78 cache_node);
79 if (block_group->key.objectid < cache->key.objectid) {
80 p = &(*p)->rb_left;
81 } else if (block_group->key.objectid > cache->key.objectid) {
82 p = &(*p)->rb_right;
83 } else {
84 spin_unlock(&info->block_group_cache_lock);
85 return -EEXIST;
86 }
87 }
88
89 rb_link_node(&block_group->cache_node, parent, p);
90 rb_insert_color(&block_group->cache_node,
91 &info->block_group_cache_tree);
92 spin_unlock(&info->block_group_cache_lock);
93
94 return 0;
95}
96
97/*
98 * This will return the block group at or after bytenr if contains is 0, else
99 * it will return the block group that contains the bytenr
100 */
101static struct btrfs_block_group_cache *
102block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
103 int contains)
104{
105 struct btrfs_block_group_cache *cache, *ret = NULL;
106 struct rb_node *n;
107 u64 end, start;
108
109 spin_lock(&info->block_group_cache_lock);
110 n = info->block_group_cache_tree.rb_node;
111
112 while (n) {
113 cache = rb_entry(n, struct btrfs_block_group_cache,
114 cache_node);
115 end = cache->key.objectid + cache->key.offset - 1;
116 start = cache->key.objectid;
117
118 if (bytenr < start) {
119 if (!contains && (!ret || start < ret->key.objectid))
120 ret = cache;
121 n = n->rb_left;
122 } else if (bytenr > start) {
123 if (contains && bytenr <= end) {
124 ret = cache;
125 break;
126 }
127 n = n->rb_right;
128 } else {
129 ret = cache;
130 break;
131 }
132 }
133 spin_unlock(&info->block_group_cache_lock);
134
135 return ret;
136}
137
138/*
139 * this is only called by cache_block_group, since we could have freed extents
140 * we need to check the pinned_extents for any extents that can't be used yet
141 * since their free space will be released as soon as the transaction commits.
142 */
143static int add_new_free_space(struct btrfs_block_group_cache *block_group,
144 struct btrfs_fs_info *info, u64 start, u64 end)
145{
146 u64 extent_start, extent_end, size;
147 int ret;
148
25179201 149 mutex_lock(&info->pinned_mutex);
0f9dd46c
JB		 150 while (start < end) {
151 ret = find_first_extent_bit(&info->pinned_extents, start,
152 &extent_start, &extent_end,
153 EXTENT_DIRTY);
154 if (ret)
155 break;
156
157 if (extent_start == start) {
158 start = extent_end + 1;
159 } else if (extent_start > start && extent_start < end) {
160 size = extent_start - start;
25179201
JB		 161 ret = btrfs_add_free_space_lock(block_group, start,
162 size);
0f9dd46c
JB		 163 BUG_ON(ret);
164 start = extent_end + 1;
165 } else {
166 break;
167 }
168 }
169
170 if (start < end) {
171 size = end - start;
25179201 172 ret = btrfs_add_free_space_lock(block_group, start, size);
0f9dd46c
JB		 173 BUG_ON(ret);
174 }
25179201 175 mutex_unlock(&info->pinned_mutex);
0f9dd46c
JB		 176
177 return 0;
178}
179
e37c9e69
CM		 180static int cache_block_group(struct btrfs_root *root,
181 struct btrfs_block_group_cache *block_group)
182{
183 struct btrfs_path *path;
ef8bbdfe 184 int ret = 0;
e37c9e69 185 struct btrfs_key key;
5f39d397 186 struct extent_buffer *leaf;
e37c9e69 187 int slot;
e37c9e69 188 u64 last = 0;
7d7d6068 189 u64 first_free;
e37c9e69
CM		 190 int found = 0;
191
00f5c795
CM		 192 if (!block_group)
193 return 0;
194
e37c9e69 195 root = root->fs_info->extent_root;
e37c9e69
CM		 196
197 if (block_group->cached)
198 return 0;
f510cfec 199
e37c9e69
CM		 200 path = btrfs_alloc_path();
201 if (!path)
202 return -ENOMEM;
7d7d6068 203
2cc58cf2 204 path->reada = 2;
5cd57b2c
CM		 205 /*
206 * we get into deadlocks with paths held by callers of this function.
207 * since the alloc_mutex is protecting things right now, just
208 * skip the locking here
209 */
210 path->skip_locking = 1;
0f9dd46c
JB		 211 first_free = max_t(u64, block_group->key.objectid,
212 BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE);
e37c9e69 213 key.objectid = block_group->key.objectid;
e37c9e69
CM		 214 key.offset = 0;
215 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
216 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
217 if (ret < 0)
ef8bbdfe 218 goto err;
0b86a832 219 ret = btrfs_previous_item(root, path, 0, BTRFS_EXTENT_ITEM_KEY);
d548ee51 220 if (ret < 0)
ef8bbdfe 221 goto err;
d548ee51
Y		 222 if (ret == 0) {
223 leaf = path->nodes[0];
224 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
225 if (key.objectid + key.offset > first_free)
226 first_free = key.objectid + key.offset;
227 }
e37c9e69 228 while(1) {
5f39d397 229 leaf = path->nodes[0];
e37c9e69 230 slot = path->slots[0];
5f39d397 231 if (slot >= btrfs_header_nritems(leaf)) {
e37c9e69 232 ret = btrfs_next_leaf(root, path);
54aa1f4d
CM		 233 if (ret < 0)
234 goto err;
0f9dd46c 235 if (ret == 0)
e37c9e69 236 continue;
0f9dd46c 237 else
e37c9e69 238 break;
e37c9e69 239 }
5f39d397 240 btrfs_item_key_to_cpu(leaf, &key, slot);
0f9dd46c 241 if (key.objectid < block_group->key.objectid)
7d7d6068 242 goto next;
0f9dd46c 243
e37c9e69 244 if (key.objectid >= block_group->key.objectid +
0f9dd46c 245 block_group->key.offset)
e37c9e69 246 break;
7d7d6068 247
e37c9e69
CM		 248 if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
249 if (!found) {
7d7d6068 250 last = first_free;
e37c9e69 251 found = 1;
e37c9e69 252 }
0f9dd46c
JB		 253
254 add_new_free_space(block_group, root->fs_info, last,
255 key.objectid);
256
7d7d6068 257 last = key.objectid + key.offset;
e37c9e69 258 }
7d7d6068 259next:
e37c9e69
CM		 260 path->slots[0]++;
261 }
262
7d7d6068
Y		 263 if (!found)
264 last = first_free;
0f9dd46c
JB		 265
266 add_new_free_space(block_group, root->fs_info, last,
267 block_group->key.objectid +
268 block_group->key.offset);
269
e37c9e69 270 block_group->cached = 1;
ef8bbdfe 271 ret = 0;
54aa1f4d 272err:
e37c9e69 273 btrfs_free_path(path);
ef8bbdfe 274 return ret;
e37c9e69
CM		 275}
276
0f9dd46c
JB		 277/*
278 * return the block group that starts at or after bytenr
279 */
0ef3e66b
CM		 280struct btrfs_block_group_cache *btrfs_lookup_first_block_group(struct
281 btrfs_fs_info *info,
282 u64 bytenr)
283{
0f9dd46c 284 struct btrfs_block_group_cache *cache;
0ef3e66b 285
0f9dd46c 286 cache = block_group_cache_tree_search(info, bytenr, 0);
0ef3e66b 287
0f9dd46c 288 return cache;
0ef3e66b
CM		 289}
290
0f9dd46c
JB		 291/*
292 * return the block group that contains teh given bytenr
293 */
5276aeda
CM		 294struct btrfs_block_group_cache *btrfs_lookup_block_group(struct
295 btrfs_fs_info *info,
db94535d 296 u64 bytenr)
be744175 297{
0f9dd46c 298 struct btrfs_block_group_cache *cache;
be744175 299
0f9dd46c 300 cache = block_group_cache_tree_search(info, bytenr, 1);
96b5179d 301
0f9dd46c 302 return cache;
be744175 303}
0b86a832 304
0f9dd46c
JB		 305static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
306 u64 flags)
6324fbf3 307{
0f9dd46c
JB		 308 struct list_head *head = &info->space_info;
309 struct list_head *cur;
310 struct btrfs_space_info *found;
311 list_for_each(cur, head) {
312 found = list_entry(cur, struct btrfs_space_info, list);
313 if (found->flags == flags)
314 return found;
315 }
316 return NULL;
6324fbf3
CM		 317}
318
80eb234a
JB		 319static u64 div_factor(u64 num, int factor)
320{
321 if (factor == 10)
322 return num;
323 num *= factor;
324 do_div(num, 10);
325 return num;
326}
327
925baedd
CM		 328static struct btrfs_block_group_cache *
329__btrfs_find_block_group(struct btrfs_root *root,
330 struct btrfs_block_group_cache *hint,
331 u64 search_start, int data, int owner)
cd1bc465 332{
96b5179d 333 struct btrfs_block_group_cache *cache;
31f3c99b 334 struct btrfs_block_group_cache *found_group = NULL;
cd1bc465
CM		 335 struct btrfs_fs_info *info = root->fs_info;
336 u64 used;
31f3c99b 337 u64 last = 0;
96b5179d 338 u64 free_check;
31f3c99b 339 int full_search = 0;
bce4eae9 340 int factor = 10;
0ef3e66b 341 int wrapped = 0;
de428b63 342
a236aed1
CM		 343 if (data & BTRFS_BLOCK_GROUP_METADATA)
344 factor = 9;
be744175 345
0ef3e66b 346 if (search_start) {
be744175 347 struct btrfs_block_group_cache *shint;
0ef3e66b 348 shint = btrfs_lookup_first_block_group(info, search_start);
8f18cf13 349 if (shint && block_group_bits(shint, data) && !shint->ro) {
c286ac48 350 spin_lock(&shint->lock);
be744175 351 used = btrfs_block_group_used(&shint->item);
e8569813 352 if (used + shint->pinned + shint->reserved <
324ae4df 353 div_factor(shint->key.offset, factor)) {
c286ac48 354 spin_unlock(&shint->lock);
be744175
CM		 355 return shint;
356 }
c286ac48 357 spin_unlock(&shint->lock);
be744175
CM		 358 }
359 }
0ef3e66b 360 if (hint && !hint->ro && block_group_bits(hint, data)) {
c286ac48 361 spin_lock(&hint->lock);
31f3c99b 362 used = btrfs_block_group_used(&hint->item);
e8569813 363 if (used + hint->pinned + hint->reserved <
324ae4df 364 div_factor(hint->key.offset, factor)) {
c286ac48 365 spin_unlock(&hint->lock);
31f3c99b
CM		 366 return hint;
367 }
c286ac48 368 spin_unlock(&hint->lock);
e19caa5f 369 last = hint->key.objectid + hint->key.offset;
31f3c99b 370 } else {
e37c9e69 371 if (hint)
0ef3e66b 372 last = max(hint->key.objectid, search_start);
e37c9e69 373 else
0ef3e66b 374 last = search_start;
31f3c99b 375 }
31f3c99b 376again:
e8569813
ZY		 377 while (1) {
378 cache = btrfs_lookup_first_block_group(root->fs_info, last);
0f9dd46c
JB		 379 if (!cache)
380 break;
96b5179d 381
c286ac48 382 spin_lock(&cache->lock);
96b5179d
CM		 383 last = cache->key.objectid + cache->key.offset;
384 used = btrfs_block_group_used(&cache->item);
385
8f18cf13 386 if (!cache->ro && block_group_bits(cache, data)) {
0ef3e66b 387 free_check = div_factor(cache->key.offset, factor);
e8569813
ZY		 388 if (used + cache->pinned + cache->reserved <
389 free_check) {
8790d502 390 found_group = cache;
c286ac48 391 spin_unlock(&cache->lock);
8790d502
CM		 392 goto found;
393 }
6324fbf3 394 }
c286ac48 395 spin_unlock(&cache->lock);
de428b63 396 cond_resched();
cd1bc465 397 }
0ef3e66b
CM		 398 if (!wrapped) {
399 last = search_start;
400 wrapped = 1;
401 goto again;
402 }
403 if (!full_search && factor < 10) {
be744175 404 last = search_start;
31f3c99b 405 full_search = 1;
0ef3e66b 406 factor = 10;
31f3c99b
CM		 407 goto again;
408 }
be744175 409found:
31f3c99b 410 return found_group;
cd1bc465
CM		 411}
412
925baedd
CM		 413struct btrfs_block_group_cache *btrfs_find_block_group(struct btrfs_root *root,
414 struct btrfs_block_group_cache
415 *hint, u64 search_start,
416 int data, int owner)
417{
418
419 struct btrfs_block_group_cache *ret;
925baedd 420 ret = __btrfs_find_block_group(root, hint, search_start, data, owner);
925baedd
CM		 421 return ret;
422}
0f9dd46c 423
e02119d5 424/* simple helper to search for an existing extent at a given offset */
31840ae1 425int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
e02119d5
CM		 426{
427 int ret;
428 struct btrfs_key key;
31840ae1 429 struct btrfs_path *path;
e02119d5 430
31840ae1
ZY		 431 path = btrfs_alloc_path();
432 BUG_ON(!path);
e02119d5
CM		 433 key.objectid = start;
434 key.offset = len;
435 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
436 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
437 0, 0);
31840ae1 438 btrfs_free_path(path);
7bb86316
CM		 439 return ret;
440}
441
d8d5f3e1
CM		 442/*
443 * Back reference rules. Back refs have three main goals:
444 *
445 * 1) differentiate between all holders of references to an extent so that
446 * when a reference is dropped we can make sure it was a valid reference
447 * before freeing the extent.
448 *
449 * 2) Provide enough information to quickly find the holders of an extent
450 * if we notice a given block is corrupted or bad.
451 *
452 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
453 * maintenance. This is actually the same as #2, but with a slightly
454 * different use case.
455 *
456 * File extents can be referenced by:
457 *
458 * - multiple snapshots, subvolumes, or different generations in one subvol
31840ae1 459 * - different files inside a single subvolume
d8d5f3e1
CM		 460 * - different offsets inside a file (bookend extents in file.c)
461 *
462 * The extent ref structure has fields for:
463 *
464 * - Objectid of the subvolume root
465 * - Generation number of the tree holding the reference
466 * - objectid of the file holding the reference
31840ae1
ZY		 467 * - number of references holding by parent node (alway 1 for tree blocks)
468 *
469 * Btree leaf may hold multiple references to a file extent. In most cases,
470 * these references are from same file and the corresponding offsets inside
3bb1a1bc 471 * the file are close together.
d8d5f3e1
CM		 472 *
473 * When a file extent is allocated the fields are filled in:
3bb1a1bc 474 * (root_key.objectid, trans->transid, inode objectid, 1)
d8d5f3e1
CM		 475 *
476 * When a leaf is cow'd new references are added for every file extent found
31840ae1
ZY		 477 * in the leaf. It looks similar to the create case, but trans->transid will
478 * be different when the block is cow'd.
d8d5f3e1 479 *
3bb1a1bc 480 * (root_key.objectid, trans->transid, inode objectid,
31840ae1 481 * number of references in the leaf)
d8d5f3e1 482 *
3bb1a1bc
YZ		 483 * When a file extent is removed either during snapshot deletion or
484 * file truncation, we find the corresponding back reference and check
485 * the following fields:
d8d5f3e1 486 *
3bb1a1bc
YZ		 487 * (btrfs_header_owner(leaf), btrfs_header_generation(leaf),
488 * inode objectid)
d8d5f3e1
CM		 489 *
490 * Btree extents can be referenced by:
491 *
492 * - Different subvolumes
493 * - Different generations of the same subvolume
494 *
d8d5f3e1
CM		 495 * When a tree block is created, back references are inserted:
496 *
3bb1a1bc 497 * (root->root_key.objectid, trans->transid, level, 1)
d8d5f3e1 498 *
31840ae1
ZY		 499 * When a tree block is cow'd, new back references are added for all the
500 * blocks it points to. If the tree block isn't in reference counted root,
501 * the old back references are removed. These new back references are of
502 * the form (trans->transid will have increased since creation):
d8d5f3e1 503 *
3bb1a1bc 504 * (root->root_key.objectid, trans->transid, level, 1)
d8d5f3e1 505 *
31840ae1 506 * When a backref is in deleting, the following fields are checked:
d8d5f3e1
CM		 507 *
508 * if backref was for a tree root:
3bb1a1bc 509 * (btrfs_header_owner(itself), btrfs_header_generation(itself), level)
d8d5f3e1 510 * else
3bb1a1bc 511 * (btrfs_header_owner(parent), btrfs_header_generation(parent), level)
d8d5f3e1 512 *
31840ae1 513 * Back Reference Key composing:
d8d5f3e1 514 *
31840ae1
ZY		 515 * The key objectid corresponds to the first byte in the extent, the key
516 * type is set to BTRFS_EXTENT_REF_KEY, and the key offset is the first
517 * byte of parent extent. If a extent is tree root, the key offset is set
518 * to the key objectid.
d8d5f3e1 519 */
31840ae1
ZY		 520
521static int noinline lookup_extent_backref(struct btrfs_trans_handle *trans,
522 struct btrfs_root *root,
3bb1a1bc
YZ		 523 struct btrfs_path *path,
524 u64 bytenr, u64 parent,
525 u64 ref_root, u64 ref_generation,
526 u64 owner_objectid, int del)
7bb86316 527{
7bb86316 528 struct btrfs_key key;
31840ae1
ZY		 529 struct btrfs_extent_ref *ref;
530 struct extent_buffer *leaf;
3bb1a1bc 531 u64 ref_objectid;
74493f7a
CM		 532 int ret;
533
31840ae1
ZY		 534 key.objectid = bytenr;
535 key.type = BTRFS_EXTENT_REF_KEY;
536 key.offset = parent;
537
538 ret = btrfs_search_slot(trans, root, &key, path, del ? -1 : 0, 1);
539 if (ret < 0)
540 goto out;
541 if (ret > 0) {
542 ret = -ENOENT;
543 goto out;
544 }
545
546 leaf = path->nodes[0];
547 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
3bb1a1bc 548 ref_objectid = btrfs_ref_objectid(leaf, ref);
31840ae1 549 if (btrfs_ref_root(leaf, ref) != ref_root ||
3bb1a1bc
YZ		 550 btrfs_ref_generation(leaf, ref) != ref_generation ||
551 (ref_objectid != owner_objectid &&
552 ref_objectid != BTRFS_MULTIPLE_OBJECTIDS)) {
31840ae1
ZY		 553 ret = -EIO;
554 WARN_ON(1);
555 goto out;
556 }
557 ret = 0;
558out:
559 return ret;
560}
561
562static int noinline insert_extent_backref(struct btrfs_trans_handle *trans,
563 struct btrfs_root *root,
564 struct btrfs_path *path,
565 u64 bytenr, u64 parent,
566 u64 ref_root, u64 ref_generation,
3bb1a1bc 567 u64 owner_objectid)
31840ae1
ZY		 568{
569 struct btrfs_key key;
570 struct extent_buffer *leaf;
571 struct btrfs_extent_ref *ref;
572 u32 num_refs;
573 int ret;
74493f7a 574
74493f7a
CM		 575 key.objectid = bytenr;
576 key.type = BTRFS_EXTENT_REF_KEY;
31840ae1 577 key.offset = parent;
74493f7a 578
31840ae1
ZY		 579 ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*ref));
580 if (ret == 0) {
581 leaf = path->nodes[0];
582 ref = btrfs_item_ptr(leaf, path->slots[0],
583 struct btrfs_extent_ref);
584 btrfs_set_ref_root(leaf, ref, ref_root);
585 btrfs_set_ref_generation(leaf, ref, ref_generation);
586 btrfs_set_ref_objectid(leaf, ref, owner_objectid);
31840ae1
ZY		 587 btrfs_set_ref_num_refs(leaf, ref, 1);
588 } else if (ret == -EEXIST) {
589 u64 existing_owner;
590 BUG_ON(owner_objectid < BTRFS_FIRST_FREE_OBJECTID);
591 leaf = path->nodes[0];
592 ref = btrfs_item_ptr(leaf, path->slots[0],
593 struct btrfs_extent_ref);
594 if (btrfs_ref_root(leaf, ref) != ref_root ||
595 btrfs_ref_generation(leaf, ref) != ref_generation) {
596 ret = -EIO;
597 WARN_ON(1);
7bb86316 598 goto out;
31840ae1
ZY		 599 }
600
601 num_refs = btrfs_ref_num_refs(leaf, ref);
602 BUG_ON(num_refs == 0);
603 btrfs_set_ref_num_refs(leaf, ref, num_refs + 1);
604
605 existing_owner = btrfs_ref_objectid(leaf, ref);
3bb1a1bc
YZ		 606 if (existing_owner != owner_objectid &&
607 existing_owner != BTRFS_MULTIPLE_OBJECTIDS) {
31840ae1
ZY		 608 btrfs_set_ref_objectid(leaf, ref,
609 BTRFS_MULTIPLE_OBJECTIDS);
31840ae1
ZY		 610 }
611 ret = 0;
612 } else {
7bb86316 613 goto out;
31840ae1 614 }
7bb86316
CM		 615 btrfs_mark_buffer_dirty(path->nodes[0]);
616out:
617 btrfs_release_path(root, path);
618 return ret;
74493f7a
CM		 619}
620
31840ae1
ZY		 621static int noinline remove_extent_backref(struct btrfs_trans_handle *trans,
622 struct btrfs_root *root,
623 struct btrfs_path *path)
624{
625 struct extent_buffer *leaf;
626 struct btrfs_extent_ref *ref;
627 u32 num_refs;
628 int ret = 0;
629
630 leaf = path->nodes[0];
631 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
632 num_refs = btrfs_ref_num_refs(leaf, ref);
633 BUG_ON(num_refs == 0);
634 num_refs -= 1;
635 if (num_refs == 0) {
636 ret = btrfs_del_item(trans, root, path);
637 } else {
638 btrfs_set_ref_num_refs(leaf, ref, num_refs);
639 btrfs_mark_buffer_dirty(leaf);
640 }
641 btrfs_release_path(root, path);
642 return ret;
643}
644
645static int __btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
646 struct btrfs_root *root, u64 bytenr,
647 u64 orig_parent, u64 parent,
648 u64 orig_root, u64 ref_root,
649 u64 orig_generation, u64 ref_generation,
3bb1a1bc 650 u64 owner_objectid)
31840ae1
ZY		 651{
652 int ret;
653 struct btrfs_root *extent_root = root->fs_info->extent_root;
654 struct btrfs_path *path;
655
656 if (root == root->fs_info->extent_root) {
657 struct pending_extent_op *extent_op;
658 u64 num_bytes;
659
660 BUG_ON(owner_objectid >= BTRFS_MAX_LEVEL);
661 num_bytes = btrfs_level_size(root, (int)owner_objectid);
25179201 662 mutex_lock(&root->fs_info->extent_ins_mutex);
31840ae1 663 if (test_range_bit(&root->fs_info->extent_ins, bytenr,
25179201 664 bytenr + num_bytes - 1, EXTENT_WRITEBACK, 0)) {
31840ae1
ZY		 665 u64 priv;
666 ret = get_state_private(&root->fs_info->extent_ins,
667 bytenr, &priv);
668 BUG_ON(ret);
669 extent_op = (struct pending_extent_op *)
670 (unsigned long)priv;
671 BUG_ON(extent_op->parent != orig_parent);
672 BUG_ON(extent_op->generation != orig_generation);
25179201 673
31840ae1
ZY		 674 extent_op->parent = parent;
675 extent_op->generation = ref_generation;
676 } else {
677 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
678 BUG_ON(!extent_op);
679
680 extent_op->type = PENDING_BACKREF_UPDATE;
681 extent_op->bytenr = bytenr;
682 extent_op->num_bytes = num_bytes;
683 extent_op->parent = parent;
684 extent_op->orig_parent = orig_parent;
685 extent_op->generation = ref_generation;
686 extent_op->orig_generation = orig_generation;
687 extent_op->level = (int)owner_objectid;
688
689 set_extent_bits(&root->fs_info->extent_ins,
690 bytenr, bytenr + num_bytes - 1,
25179201 691 EXTENT_WRITEBACK, GFP_NOFS);
31840ae1
ZY		 692 set_state_private(&root->fs_info->extent_ins,
693 bytenr, (unsigned long)extent_op);
694 }
25179201 695 mutex_unlock(&root->fs_info->extent_ins_mutex);
31840ae1
ZY		 696 return 0;
697 }
698
699 path = btrfs_alloc_path();
700 if (!path)
701 return -ENOMEM;
702 ret = lookup_extent_backref(trans, extent_root, path,
703 bytenr, orig_parent, orig_root,
3bb1a1bc 704 orig_generation, owner_objectid, 1);
31840ae1
ZY		 705 if (ret)
706 goto out;
707 ret = remove_extent_backref(trans, extent_root, path);
708 if (ret)
709 goto out;
710 ret = insert_extent_backref(trans, extent_root, path, bytenr,
711 parent, ref_root, ref_generation,
3bb1a1bc 712 owner_objectid);
31840ae1 713 BUG_ON(ret);
87ef2bb4
CM		 714 finish_current_insert(trans, extent_root, 0);
715 del_pending_extents(trans, extent_root, 0);
31840ae1
ZY		 716out:
717 btrfs_free_path(path);
718 return ret;
719}
720
721int btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
722 struct btrfs_root *root, u64 bytenr,
723 u64 orig_parent, u64 parent,
724 u64 ref_root, u64 ref_generation,
3bb1a1bc 725 u64 owner_objectid)
31840ae1
ZY		 726{
727 int ret;
728 if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
729 owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
730 return 0;
31840ae1
ZY		 731 ret = __btrfs_update_extent_ref(trans, root, bytenr, orig_parent,
732 parent, ref_root, ref_root,
733 ref_generation, ref_generation,
3bb1a1bc 734 owner_objectid);
31840ae1
ZY		 735 return ret;
736}
737
925baedd 738static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
31840ae1
ZY		 739 struct btrfs_root *root, u64 bytenr,
740 u64 orig_parent, u64 parent,
741 u64 orig_root, u64 ref_root,
742 u64 orig_generation, u64 ref_generation,
3bb1a1bc 743 u64 owner_objectid)
02217ed2 744{
5caf2a00 745 struct btrfs_path *path;
02217ed2 746 int ret;
e2fa7227 747 struct btrfs_key key;
5f39d397 748 struct extent_buffer *l;
234b63a0 749 struct btrfs_extent_item *item;
cf27e1ee 750 u32 refs;
037e6390 751
5caf2a00 752 path = btrfs_alloc_path();
54aa1f4d
CM		 753 if (!path)
754 return -ENOMEM;
26b8003f 755
3c12ac72 756 path->reada = 1;
db94535d 757 key.objectid = bytenr;
31840ae1
ZY		 758 key.type = BTRFS_EXTENT_ITEM_KEY;
759 key.offset = (u64)-1;
760
5caf2a00 761 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
9f5fae2f 762 0, 1);
54aa1f4d
CM		 763 if (ret < 0)
764 return ret;
31840ae1
ZY		 765 BUG_ON(ret == 0 || path->slots[0] == 0);
766
767 path->slots[0]--;
5f39d397 768 l = path->nodes[0];
31840ae1
ZY		 769
770 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
771 BUG_ON(key.objectid != bytenr);
772 BUG_ON(key.type != BTRFS_EXTENT_ITEM_KEY);
773
5caf2a00 774 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
5f39d397
CM		 775 refs = btrfs_extent_refs(l, item);
776 btrfs_set_extent_refs(l, item, refs + 1);
5caf2a00 777 btrfs_mark_buffer_dirty(path->nodes[0]);
a28ec197 778
5caf2a00 779 btrfs_release_path(root->fs_info->extent_root, path);
7bb86316 780
3c12ac72 781 path->reada = 1;
31840ae1
ZY		 782 ret = insert_extent_backref(trans, root->fs_info->extent_root,
783 path, bytenr, parent,
784 ref_root, ref_generation,
3bb1a1bc 785 owner_objectid);
7bb86316 786 BUG_ON(ret);
87ef2bb4
CM		 787 finish_current_insert(trans, root->fs_info->extent_root, 0);
788 del_pending_extents(trans, root->fs_info->extent_root, 0);
74493f7a
CM		 789
790 btrfs_free_path(path);
02217ed2
CM		 791 return 0;
792}
793
925baedd 794int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
31840ae1
ZY		 795 struct btrfs_root *root,
796 u64 bytenr, u64 num_bytes, u64 parent,
797 u64 ref_root, u64 ref_generation,
3bb1a1bc 798 u64 owner_objectid)
925baedd
CM		 799{
800 int ret;
31840ae1
ZY		 801 if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
802 owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
803 return 0;
31840ae1
ZY		 804 ret = __btrfs_inc_extent_ref(trans, root, bytenr, 0, parent,
805 0, ref_root, 0, ref_generation,
3bb1a1bc 806 owner_objectid);
925baedd
CM		 807 return ret;
808}
809
e9d0b13b
CM		 810int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
811 struct btrfs_root *root)
812{
87ef2bb4
CM		 813 finish_current_insert(trans, root->fs_info->extent_root, 1);
814 del_pending_extents(trans, root->fs_info->extent_root, 1);
e9d0b13b
CM		 815 return 0;
816}
817
31840ae1
ZY		 818int btrfs_lookup_extent_ref(struct btrfs_trans_handle *trans,
819 struct btrfs_root *root, u64 bytenr,
820 u64 num_bytes, u32 *refs)
a28ec197 821{
5caf2a00 822 struct btrfs_path *path;
a28ec197 823 int ret;
e2fa7227 824 struct btrfs_key key;
5f39d397 825 struct extent_buffer *l;
234b63a0 826 struct btrfs_extent_item *item;
5caf2a00 827
db94535d 828 WARN_ON(num_bytes < root->sectorsize);
5caf2a00 829 path = btrfs_alloc_path();
3c12ac72 830 path->reada = 1;
db94535d
CM		 831 key.objectid = bytenr;
832 key.offset = num_bytes;
62e2749e 833 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
5caf2a00 834 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
9f5fae2f 835 0, 0);
54aa1f4d
CM		 836 if (ret < 0)
837 goto out;
5f39d397
CM		 838 if (ret != 0) {
839 btrfs_print_leaf(root, path->nodes[0]);
db94535d 840 printk("failed to find block number %Lu\n", bytenr);
a28ec197 841 BUG();
5f39d397
CM		 842 }
843 l = path->nodes[0];
5caf2a00 844 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
5f39d397 845 *refs = btrfs_extent_refs(l, item);
54aa1f4d 846out:
5caf2a00 847 btrfs_free_path(path);
a28ec197
CM		 848 return 0;
849}
850
f321e491
YZ		 851static int get_reference_status(struct btrfs_root *root, u64 bytenr,
852 u64 parent_gen, u64 ref_objectid,
853 u64 *min_generation, u32 *ref_count)
be20aa9d
CM		 854{
855 struct btrfs_root *extent_root = root->fs_info->extent_root;
856 struct btrfs_path *path;
f321e491
YZ		 857 struct extent_buffer *leaf;
858 struct btrfs_extent_ref *ref_item;
859 struct btrfs_key key;
860 struct btrfs_key found_key;
56b453c9 861 u64 root_objectid = root->root_key.objectid;
f321e491 862 u64 ref_generation;
be20aa9d
CM		 863 u32 nritems;
864 int ret;
925baedd 865
be20aa9d 866 key.objectid = bytenr;
31840ae1 867 key.offset = (u64)-1;
f321e491 868 key.type = BTRFS_EXTENT_ITEM_KEY;
be20aa9d 869
f321e491 870 path = btrfs_alloc_path();
be20aa9d
CM		 871 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
872 if (ret < 0)
873 goto out;
874 BUG_ON(ret == 0);
31840ae1
ZY		 875 if (ret < 0 || path->slots[0] == 0)
876 goto out;
be20aa9d 877
31840ae1 878 path->slots[0]--;
f321e491
YZ		 879 leaf = path->nodes[0];
880 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
be20aa9d
CM		 881
882 if (found_key.objectid != bytenr ||
883 found_key.type != BTRFS_EXTENT_ITEM_KEY) {
f321e491 884 ret = 1;
be20aa9d
CM		 885 goto out;
886 }
887
f321e491
YZ		 888 *ref_count = 0;
889 *min_generation = (u64)-1;
890
be20aa9d 891 while (1) {
f321e491
YZ		 892 leaf = path->nodes[0];
893 nritems = btrfs_header_nritems(leaf);
be20aa9d
CM		 894 if (path->slots[0] >= nritems) {
895 ret = btrfs_next_leaf(extent_root, path);
f321e491
YZ		 896 if (ret < 0)
897 goto out;
be20aa9d
CM		 898 if (ret == 0)
899 continue;
900 break;
901 }
f321e491 902 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
be20aa9d
CM		 903 if (found_key.objectid != bytenr)
904 break;
bd09835d 905
be20aa9d
CM		 906 if (found_key.type != BTRFS_EXTENT_REF_KEY) {
907 path->slots[0]++;
908 continue;
909 }
910
f321e491 911 ref_item = btrfs_item_ptr(leaf, path->slots[0],
be20aa9d 912 struct btrfs_extent_ref);
f321e491
YZ		 913 ref_generation = btrfs_ref_generation(leaf, ref_item);
914 /*
31840ae1 915 * For (parent_gen > 0 && parent_gen > ref_generation):
f321e491 916 *
bcc63abb
Y		 917 * we reach here through the oldest root, therefore
918 * all other reference from same snapshot should have
f321e491
YZ		 919 * a larger generation.
920 */
921 if ((root_objectid != btrfs_ref_root(leaf, ref_item)) ||
922 (parent_gen > 0 && parent_gen > ref_generation) ||
923 (ref_objectid >= BTRFS_FIRST_FREE_OBJECTID &&
924 ref_objectid != btrfs_ref_objectid(leaf, ref_item))) {
31840ae1 925 *ref_count = 2;
f321e491 926 break;
a68d5933 927 }
f321e491
YZ		 928
929 *ref_count = 1;
930 if (*min_generation > ref_generation)
931 *min_generation = ref_generation;
932
be20aa9d
CM		 933 path->slots[0]++;
934 }
f321e491
YZ		 935 ret = 0;
936out:
f321e491
YZ		 937 btrfs_free_path(path);
938 return ret;
939}
940
7ea394f1
YZ		 941int btrfs_cross_ref_exists(struct btrfs_trans_handle *trans,
942 struct btrfs_root *root,
f321e491
YZ		 943 struct btrfs_key *key, u64 bytenr)
944{
f321e491
YZ		 945 struct btrfs_root *old_root;
946 struct btrfs_path *path = NULL;
947 struct extent_buffer *eb;
948 struct btrfs_file_extent_item *item;
949 u64 ref_generation;
950 u64 min_generation;
951 u64 extent_start;
952 u32 ref_count;
953 int level;
954 int ret;
955
7ea394f1 956 BUG_ON(trans == NULL);
f321e491
YZ		 957 BUG_ON(key->type != BTRFS_EXTENT_DATA_KEY);
958 ret = get_reference_status(root, bytenr, 0, key->objectid,
959 &min_generation, &ref_count);
960 if (ret)
961 return ret;
962
963 if (ref_count != 1)
964 return 1;
965
f321e491
YZ		 966 old_root = root->dirty_root->root;
967 ref_generation = old_root->root_key.offset;
968
969 /* all references are created in running transaction */
970 if (min_generation > ref_generation) {
971 ret = 0;
bbaf549e
CM		 972 goto out;
973 }
f321e491
YZ		 974
975 path = btrfs_alloc_path();
976 if (!path) {
977 ret = -ENOMEM;
be20aa9d
CM		 978 goto out;
979 }
f321e491
YZ		 980
981 path->skip_locking = 1;
982 /* if no item found, the extent is referenced by other snapshot */
983 ret = btrfs_search_slot(NULL, old_root, key, path, 0, 0);
984 if (ret)
be20aa9d 985 goto out;
be20aa9d 986
f321e491
YZ		 987 eb = path->nodes[0];
988 item = btrfs_item_ptr(eb, path->slots[0],
989 struct btrfs_file_extent_item);
990 if (btrfs_file_extent_type(eb, item) != BTRFS_FILE_EXTENT_REG ||
991 btrfs_file_extent_disk_bytenr(eb, item) != bytenr) {
992 ret = 1;
993 goto out;
994 }
995
996 for (level = BTRFS_MAX_LEVEL - 1; level >= -1; level--) {
997 if (level >= 0) {
998 eb = path->nodes[level];
999 if (!eb)
1000 continue;
1001 extent_start = eb->start;
bcc63abb 1002 } else
f321e491
YZ		 1003 extent_start = bytenr;
1004
1005 ret = get_reference_status(root, extent_start, ref_generation,
1006 0, &min_generation, &ref_count);
1007 if (ret)
1008 goto out;
1009
1010 if (ref_count != 1) {
1011 ret = 1;
1012 goto out;
1013 }
1014 if (level >= 0)
1015 ref_generation = btrfs_header_generation(eb);
1016 }
1017 ret = 0;
be20aa9d 1018out:
f321e491
YZ		 1019 if (path)
1020 btrfs_free_path(path);
f321e491 1021 return ret;
be20aa9d 1022}
c5739bba 1023
31840ae1
ZY		 1024int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1025 struct extent_buffer *buf, u32 nr_extents)
02217ed2 1026{
5f39d397 1027 struct btrfs_key key;
6407bf6d 1028 struct btrfs_file_extent_item *fi;
e4657689
ZY		 1029 u64 root_gen;
1030 u32 nritems;
02217ed2 1031 int i;
db94535d 1032 int level;
31840ae1 1033 int ret = 0;
e4657689 1034 int shared = 0;
a28ec197 1035
3768f368 1036 if (!root->ref_cows)
a28ec197 1037 return 0;
5f39d397 1038
e4657689
ZY		 1039 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
1040 shared = 0;
1041 root_gen = root->root_key.offset;
1042 } else {
1043 shared = 1;
1044 root_gen = trans->transid - 1;
1045 }
1046
db94535d 1047 level = btrfs_header_level(buf);
5f39d397 1048 nritems = btrfs_header_nritems(buf);
4a096752 1049
31840ae1 1050 if (level == 0) {
31153d81
YZ		 1051 struct btrfs_leaf_ref *ref;
1052 struct btrfs_extent_info *info;
1053
31840ae1 1054 ref = btrfs_alloc_leaf_ref(root, nr_extents);
31153d81 1055 if (!ref) {
31840ae1 1056 ret = -ENOMEM;
31153d81
YZ		 1057 goto out;
1058 }
1059
e4657689 1060 ref->root_gen = root_gen;
31153d81
YZ		 1061 ref->bytenr = buf->start;
1062 ref->owner = btrfs_header_owner(buf);
1063 ref->generation = btrfs_header_generation(buf);
31840ae1 1064 ref->nritems = nr_extents;
31153d81 1065 info = ref->extents;
bcc63abb 1066
31840ae1 1067 for (i = 0; nr_extents > 0 && i < nritems; i++) {
31153d81
YZ		 1068 u64 disk_bytenr;
1069 btrfs_item_key_to_cpu(buf, &key, i);
1070 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1071 continue;
1072 fi = btrfs_item_ptr(buf, i,
1073 struct btrfs_file_extent_item);
1074 if (btrfs_file_extent_type(buf, fi) ==
1075 BTRFS_FILE_EXTENT_INLINE)
1076 continue;
1077 disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1078 if (disk_bytenr == 0)
1079 continue;
1080
1081 info->bytenr = disk_bytenr;
1082 info->num_bytes =
1083 btrfs_file_extent_disk_num_bytes(buf, fi);
1084 info->objectid = key.objectid;
1085 info->offset = key.offset;
1086 info++;
1087 }
1088
e4657689 1089 ret = btrfs_add_leaf_ref(root, ref, shared);
5b84e8d6
YZ		 1090 if (ret == -EEXIST && shared) {
1091 struct btrfs_leaf_ref *old;
1092 old = btrfs_lookup_leaf_ref(root, ref->bytenr);
1093 BUG_ON(!old);
1094 btrfs_remove_leaf_ref(root, old);
1095 btrfs_free_leaf_ref(root, old);
1096 ret = btrfs_add_leaf_ref(root, ref, shared);
1097 }
31153d81 1098 WARN_ON(ret);
bcc63abb 1099 btrfs_free_leaf_ref(root, ref);
31153d81
YZ		 1100 }
1101out:
31840ae1
ZY		 1102 return ret;
1103}
1104
1105int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1106 struct extent_buffer *orig_buf, struct extent_buffer *buf,
1107 u32 *nr_extents)
1108{
1109 u64 bytenr;
1110 u64 ref_root;
1111 u64 orig_root;
1112 u64 ref_generation;
1113 u64 orig_generation;
1114 u32 nritems;
1115 u32 nr_file_extents = 0;
1116 struct btrfs_key key;
1117 struct btrfs_file_extent_item *fi;
1118 int i;
1119 int level;
1120 int ret = 0;
1121 int faili = 0;
1122 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
3bb1a1bc 1123 u64, u64, u64, u64, u64, u64, u64, u64);
31840ae1
ZY		 1124
1125 ref_root = btrfs_header_owner(buf);
1126 ref_generation = btrfs_header_generation(buf);
1127 orig_root = btrfs_header_owner(orig_buf);
1128 orig_generation = btrfs_header_generation(orig_buf);
1129
1130 nritems = btrfs_header_nritems(buf);
1131 level = btrfs_header_level(buf);
1132
1133 if (root->ref_cows) {
1134 process_func = __btrfs_inc_extent_ref;
1135 } else {
1136 if (level == 0 &&
1137 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
1138 goto out;
1139 if (level != 0 &&
1140 root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
1141 goto out;
1142 process_func = __btrfs_update_extent_ref;
1143 }
1144
1145 for (i = 0; i < nritems; i++) {
1146 cond_resched();
db94535d 1147 if (level == 0) {
5f39d397
CM		 1148 btrfs_item_key_to_cpu(buf, &key, i);
1149 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
54aa1f4d 1150 continue;
5f39d397 1151 fi = btrfs_item_ptr(buf, i,
54aa1f4d 1152 struct btrfs_file_extent_item);
5f39d397 1153 if (btrfs_file_extent_type(buf, fi) ==
54aa1f4d
CM		 1154 BTRFS_FILE_EXTENT_INLINE)
1155 continue;
31840ae1
ZY		 1156 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1157 if (bytenr == 0)
54aa1f4d 1158 continue;
31840ae1
ZY		 1159
1160 nr_file_extents++;
1161
31840ae1
ZY		 1162 ret = process_func(trans, root, bytenr,
1163 orig_buf->start, buf->start,
1164 orig_root, ref_root,
1165 orig_generation, ref_generation,
3bb1a1bc 1166 key.objectid);
31840ae1
ZY		 1167
1168 if (ret) {
1169 faili = i;
1170 WARN_ON(1);
1171 goto fail;
1172 }
54aa1f4d 1173 } else {
db94535d 1174 bytenr = btrfs_node_blockptr(buf, i);
31840ae1
ZY		 1175 ret = process_func(trans, root, bytenr,
1176 orig_buf->start, buf->start,
1177 orig_root, ref_root,
1178 orig_generation, ref_generation,
3bb1a1bc 1179 level - 1);
31840ae1
ZY		 1180 if (ret) {
1181 faili = i;
1182 WARN_ON(1);
1183 goto fail;
1184 }
54aa1f4d
CM		 1185 }
1186 }
31840ae1
ZY		 1187out:
1188 if (nr_extents) {
1189 if (level == 0)
1190 *nr_extents = nr_file_extents;
1191 else
1192 *nr_extents = nritems;
1193 }
1194 return 0;
1195fail:
1196 WARN_ON(1);
54aa1f4d 1197 return ret;
02217ed2
CM		 1198}
1199
31840ae1
ZY		 1200int btrfs_update_ref(struct btrfs_trans_handle *trans,
1201 struct btrfs_root *root, struct extent_buffer *orig_buf,
1202 struct extent_buffer *buf, int start_slot, int nr)
1203
1204{
1205 u64 bytenr;
1206 u64 ref_root;
1207 u64 orig_root;
1208 u64 ref_generation;
1209 u64 orig_generation;
1210 struct btrfs_key key;
1211 struct btrfs_file_extent_item *fi;
1212 int i;
1213 int ret;
1214 int slot;
1215 int level;
1216
1217 BUG_ON(start_slot < 0);
1218 BUG_ON(start_slot + nr > btrfs_header_nritems(buf));
1219
1220 ref_root = btrfs_header_owner(buf);
1221 ref_generation = btrfs_header_generation(buf);
1222 orig_root = btrfs_header_owner(orig_buf);
1223 orig_generation = btrfs_header_generation(orig_buf);
1224 level = btrfs_header_level(buf);
1225
1226 if (!root->ref_cows) {
1227 if (level == 0 &&
1228 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
1229 return 0;
1230 if (level != 0 &&
1231 root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
1232 return 0;
1233 }
1234
1235 for (i = 0, slot = start_slot; i < nr; i++, slot++) {
1236 cond_resched();
1237 if (level == 0) {
1238 btrfs_item_key_to_cpu(buf, &key, slot);
1239 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1240 continue;
1241 fi = btrfs_item_ptr(buf, slot,
1242 struct btrfs_file_extent_item);
1243 if (btrfs_file_extent_type(buf, fi) ==
1244 BTRFS_FILE_EXTENT_INLINE)
1245 continue;
1246 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1247 if (bytenr == 0)
1248 continue;
31840ae1
ZY		 1249 ret = __btrfs_update_extent_ref(trans, root, bytenr,
1250 orig_buf->start, buf->start,
1251 orig_root, ref_root,
1252 orig_generation, ref_generation,
3bb1a1bc 1253 key.objectid);
31840ae1
ZY		 1254 if (ret)
1255 goto fail;
1256 } else {
1257 bytenr = btrfs_node_blockptr(buf, slot);
31840ae1
ZY		 1258 ret = __btrfs_update_extent_ref(trans, root, bytenr,
1259 orig_buf->start, buf->start,
1260 orig_root, ref_root,
1261 orig_generation, ref_generation,
3bb1a1bc 1262 level - 1);
31840ae1
ZY		 1263 if (ret)
1264 goto fail;
1265 }
1266 }
1267 return 0;
1268fail:
1269 WARN_ON(1);
1270 return -1;
1271}
1272
9078a3e1
CM		 1273static int write_one_cache_group(struct btrfs_trans_handle *trans,
1274 struct btrfs_root *root,
1275 struct btrfs_path *path,
1276 struct btrfs_block_group_cache *cache)
1277{
1278 int ret;
1279 int pending_ret;
1280 struct btrfs_root *extent_root = root->fs_info->extent_root;
5f39d397
CM		 1281 unsigned long bi;
1282 struct extent_buffer *leaf;
9078a3e1 1283
9078a3e1 1284 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
54aa1f4d
CM		 1285 if (ret < 0)
1286 goto fail;
9078a3e1 1287 BUG_ON(ret);
5f39d397
CM		 1288
1289 leaf = path->nodes[0];
1290 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
1291 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
1292 btrfs_mark_buffer_dirty(leaf);
9078a3e1 1293 btrfs_release_path(extent_root, path);
54aa1f4d 1294fail:
87ef2bb4
CM		 1295 finish_current_insert(trans, extent_root, 0);
1296 pending_ret = del_pending_extents(trans, extent_root, 0);
9078a3e1
CM		 1297 if (ret)
1298 return ret;
1299 if (pending_ret)
1300 return pending_ret;
1301 return 0;
1302
1303}
1304
96b5179d
CM		 1305int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
1306 struct btrfs_root *root)
9078a3e1 1307{
0f9dd46c
JB		 1308 struct btrfs_block_group_cache *cache, *entry;
1309 struct rb_node *n;
9078a3e1
CM		 1310 int err = 0;
1311 int werr = 0;
9078a3e1 1312 struct btrfs_path *path;
96b5179d 1313 u64 last = 0;
9078a3e1
CM		 1314
1315 path = btrfs_alloc_path();
1316 if (!path)
1317 return -ENOMEM;
1318
1319 while(1) {
0f9dd46c
JB		 1320 cache = NULL;
1321 spin_lock(&root->fs_info->block_group_cache_lock);
1322 for (n = rb_first(&root->fs_info->block_group_cache_tree);
1323 n; n = rb_next(n)) {
1324 entry = rb_entry(n, struct btrfs_block_group_cache,
1325 cache_node);
1326 if (entry->dirty) {
1327 cache = entry;
1328 break;
1329 }
1330 }
1331 spin_unlock(&root->fs_info->block_group_cache_lock);
54aa1f4d 1332
0f9dd46c 1333 if (!cache)
96b5179d 1334 break;
0f9dd46c 1335
e8569813 1336 cache->dirty = 0;
0f9dd46c
JB		 1337 last += cache->key.offset;
1338
96b5179d
CM		 1339 err = write_one_cache_group(trans, root,
1340 path, cache);
1341 /*
1342 * if we fail to write the cache group, we want
1343 * to keep it marked dirty in hopes that a later
1344 * write will work
1345 */
1346 if (err) {
1347 werr = err;
1348 continue;
9078a3e1
CM		 1349 }
1350 }
1351 btrfs_free_path(path);
1352 return werr;
1353}
1354
593060d7
CM		 1355static int update_space_info(struct btrfs_fs_info *info, u64 flags,
1356 u64 total_bytes, u64 bytes_used,
1357 struct btrfs_space_info **space_info)
1358{
1359 struct btrfs_space_info *found;
1360
1361 found = __find_space_info(info, flags);
1362 if (found) {
25179201 1363 spin_lock(&found->lock);
593060d7
CM		 1364 found->total_bytes += total_bytes;
1365 found->bytes_used += bytes_used;
8f18cf13 1366 found->full = 0;
25179201 1367 spin_unlock(&found->lock);
593060d7
CM		 1368 *space_info = found;
1369 return 0;
1370 }
1371 found = kmalloc(sizeof(*found), GFP_NOFS);
1372 if (!found)
1373 return -ENOMEM;
1374
1375 list_add(&found->list, &info->space_info);
0f9dd46c 1376 INIT_LIST_HEAD(&found->block_groups);
80eb234a 1377 init_rwsem(&found->groups_sem);
0f9dd46c 1378 spin_lock_init(&found->lock);
593060d7
CM		 1379 found->flags = flags;
1380 found->total_bytes = total_bytes;
1381 found->bytes_used = bytes_used;
1382 found->bytes_pinned = 0;
e8569813 1383 found->bytes_reserved = 0;
593060d7 1384 found->full = 0;
0ef3e66b 1385 found->force_alloc = 0;
593060d7
CM		 1386 *space_info = found;
1387 return 0;
1388}
1389
8790d502
CM		 1390static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
1391{
1392 u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
611f0e00 1393 BTRFS_BLOCK_GROUP_RAID1 |
321aecc6 1394 BTRFS_BLOCK_GROUP_RAID10 |
611f0e00 1395 BTRFS_BLOCK_GROUP_DUP);
8790d502
CM		 1396 if (extra_flags) {
1397 if (flags & BTRFS_BLOCK_GROUP_DATA)
1398 fs_info->avail_data_alloc_bits |= extra_flags;
1399 if (flags & BTRFS_BLOCK_GROUP_METADATA)
1400 fs_info->avail_metadata_alloc_bits |= extra_flags;
1401 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
1402 fs_info->avail_system_alloc_bits |= extra_flags;
1403 }
1404}
593060d7 1405
a061fc8d 1406static u64 reduce_alloc_profile(struct btrfs_root *root, u64 flags)
ec44a35c 1407{
a061fc8d
CM		 1408 u64 num_devices = root->fs_info->fs_devices->num_devices;
1409
1410 if (num_devices == 1)
1411 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
1412 if (num_devices < 4)
1413 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
1414
ec44a35c
CM		 1415 if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
1416 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
a061fc8d 1417 BTRFS_BLOCK_GROUP_RAID10))) {
ec44a35c 1418 flags &= ~BTRFS_BLOCK_GROUP_DUP;
a061fc8d 1419 }
ec44a35c
CM		 1420
1421 if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
a061fc8d 1422 (flags & BTRFS_BLOCK_GROUP_RAID10)) {
ec44a35c 1423 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
a061fc8d 1424 }
ec44a35c
CM		 1425
1426 if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
1427 ((flags & BTRFS_BLOCK_GROUP_RAID1) |
1428 (flags & BTRFS_BLOCK_GROUP_RAID10) |
1429 (flags & BTRFS_BLOCK_GROUP_DUP)))
1430 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
1431 return flags;
1432}
1433
6324fbf3
CM		 1434static int do_chunk_alloc(struct btrfs_trans_handle *trans,
1435 struct btrfs_root *extent_root, u64 alloc_bytes,
0ef3e66b 1436 u64 flags, int force)
6324fbf3
CM		 1437{
1438 struct btrfs_space_info *space_info;
1439 u64 thresh;
1440 u64 start;
1441 u64 num_bytes;
cf749823 1442 int ret = 0, waited = 0;
6324fbf3 1443
a061fc8d 1444 flags = reduce_alloc_profile(extent_root, flags);
ec44a35c 1445
6324fbf3 1446 space_info = __find_space_info(extent_root->fs_info, flags);
593060d7
CM		 1447 if (!space_info) {
1448 ret = update_space_info(extent_root->fs_info, flags,
1449 0, 0, &space_info);
1450 BUG_ON(ret);
1451 }
6324fbf3
CM		 1452 BUG_ON(!space_info);
1453
25179201 1454 spin_lock(&space_info->lock);
0ef3e66b
CM		 1455 if (space_info->force_alloc) {
1456 force = 1;
1457 space_info->force_alloc = 0;
1458 }
25179201
JB		 1459 if (space_info->full) {
1460 spin_unlock(&space_info->lock);
925baedd 1461 goto out;
25179201 1462 }
6324fbf3 1463
8790d502 1464 thresh = div_factor(space_info->total_bytes, 6);
0ef3e66b 1465 if (!force &&
e8569813 1466 (space_info->bytes_used + space_info->bytes_pinned +
25179201
JB		 1467 space_info->bytes_reserved + alloc_bytes) < thresh) {
1468 spin_unlock(&space_info->lock);
925baedd 1469 goto out;
25179201 1470 }
6324fbf3 1471
25179201
JB		 1472 spin_unlock(&space_info->lock);
1473
1474 ret = mutex_trylock(&extent_root->fs_info->chunk_mutex);
1475 if (!ret && !force) {
1476 goto out;
1477 } else if (!ret) {
1478 mutex_lock(&extent_root->fs_info->chunk_mutex);
cf749823
JB		 1479 waited = 1;
1480 }
1481
25179201
JB		 1482 if (waited) {
1483 spin_lock(&space_info->lock);
1484 if (space_info->full) {
1485 spin_unlock(&space_info->lock);
1486 goto out_unlock;
1487 }
1488 spin_unlock(&space_info->lock);
1489 }
cf749823 1490
6324fbf3 1491 ret = btrfs_alloc_chunk(trans, extent_root, &start, &num_bytes, flags);
25179201 1492 if (ret) {
6324fbf3
CM		 1493printk("space info full %Lu\n", flags);
1494 space_info->full = 1;
a74a4b97 1495 goto out_unlock;
6324fbf3 1496 }
6324fbf3
CM		 1497
1498 ret = btrfs_make_block_group(trans, extent_root, 0, flags,
e17cade2 1499 BTRFS_FIRST_CHUNK_TREE_OBJECTID, start, num_bytes);
6324fbf3 1500 BUG_ON(ret);
a74a4b97 1501out_unlock:
333db94c 1502 mutex_unlock(&extent_root->fs_info->chunk_mutex);
a74a4b97 1503out:
0f9dd46c 1504 return ret;
6324fbf3
CM		 1505}
1506
9078a3e1
CM		 1507static int update_block_group(struct btrfs_trans_handle *trans,
1508 struct btrfs_root *root,
db94535d 1509 u64 bytenr, u64 num_bytes, int alloc,
0b86a832 1510 int mark_free)
9078a3e1
CM		 1511{
1512 struct btrfs_block_group_cache *cache;
1513 struct btrfs_fs_info *info = root->fs_info;
db94535d 1514 u64 total = num_bytes;
9078a3e1 1515 u64 old_val;
db94535d 1516 u64 byte_in_group;
3e1ad54f 1517
9078a3e1 1518 while(total) {
db94535d 1519 cache = btrfs_lookup_block_group(info, bytenr);
3e1ad54f 1520 if (!cache) {
9078a3e1 1521 return -1;
cd1bc465 1522 }
db94535d
CM		 1523 byte_in_group = bytenr - cache->key.objectid;
1524 WARN_ON(byte_in_group > cache->key.offset);
9078a3e1 1525
25179201 1526 spin_lock(&cache->space_info->lock);
c286ac48 1527 spin_lock(&cache->lock);
0f9dd46c 1528 cache->dirty = 1;
9078a3e1 1529 old_val = btrfs_block_group_used(&cache->item);
db94535d 1530 num_bytes = min(total, cache->key.offset - byte_in_group);
cd1bc465 1531 if (alloc) {
db94535d 1532 old_val += num_bytes;
6324fbf3 1533 cache->space_info->bytes_used += num_bytes;
c286ac48
CM		 1534 btrfs_set_block_group_used(&cache->item, old_val);
1535 spin_unlock(&cache->lock);
25179201 1536 spin_unlock(&cache->space_info->lock);
cd1bc465 1537 } else {
db94535d 1538 old_val -= num_bytes;
6324fbf3 1539 cache->space_info->bytes_used -= num_bytes;
c286ac48
CM		 1540 btrfs_set_block_group_used(&cache->item, old_val);
1541 spin_unlock(&cache->lock);
25179201 1542 spin_unlock(&cache->space_info->lock);
f510cfec 1543 if (mark_free) {
0f9dd46c
JB		 1544 int ret;
1545 ret = btrfs_add_free_space(cache, bytenr,
1546 num_bytes);
1547 if (ret)
1548 return -1;
e37c9e69 1549 }
cd1bc465 1550 }
db94535d
CM		 1551 total -= num_bytes;
1552 bytenr += num_bytes;
9078a3e1
CM		 1553 }
1554 return 0;
1555}
6324fbf3 1556
a061fc8d
CM		 1557static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
1558{
0f9dd46c
JB		 1559 struct btrfs_block_group_cache *cache;
1560
1561 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
1562 if (!cache)
a061fc8d 1563 return 0;
0f9dd46c
JB		 1564
1565 return cache->key.objectid;
a061fc8d
CM		 1566}
1567
e02119d5 1568int btrfs_update_pinned_extents(struct btrfs_root *root,
324ae4df
Y		 1569 u64 bytenr, u64 num, int pin)
1570{
1571 u64 len;
1572 struct btrfs_block_group_cache *cache;
1573 struct btrfs_fs_info *fs_info = root->fs_info;
1574
25179201 1575 WARN_ON(!mutex_is_locked(&root->fs_info->pinned_mutex));
324ae4df
Y		 1576 if (pin) {
1577 set_extent_dirty(&fs_info->pinned_extents,
1578 bytenr, bytenr + num - 1, GFP_NOFS);
1579 } else {
1580 clear_extent_dirty(&fs_info->pinned_extents,
1581 bytenr, bytenr + num - 1, GFP_NOFS);
1582 }
1583 while (num > 0) {
1584 cache = btrfs_lookup_block_group(fs_info, bytenr);
e8569813
ZY		 1585 BUG_ON(!cache);
1586 len = min(num, cache->key.offset -
1587 (bytenr - cache->key.objectid));
324ae4df 1588 if (pin) {
25179201 1589 spin_lock(&cache->space_info->lock);
e8569813
ZY		 1590 spin_lock(&cache->lock);
1591 cache->pinned += len;
1592 cache->space_info->bytes_pinned += len;
1593 spin_unlock(&cache->lock);
25179201 1594 spin_unlock(&cache->space_info->lock);
324ae4df
Y		 1595 fs_info->total_pinned += len;
1596 } else {
25179201 1597 spin_lock(&cache->space_info->lock);
e8569813
ZY		 1598 spin_lock(&cache->lock);
1599 cache->pinned -= len;
1600 cache->space_info->bytes_pinned -= len;
1601 spin_unlock(&cache->lock);
25179201 1602 spin_unlock(&cache->space_info->lock);
324ae4df
Y		 1603 fs_info->total_pinned -= len;
1604 }
1605 bytenr += len;
1606 num -= len;
1607 }
1608 return 0;
1609}
9078a3e1 1610
e8569813
ZY		 1611static int update_reserved_extents(struct btrfs_root *root,
1612 u64 bytenr, u64 num, int reserve)
1613{
1614 u64 len;
1615 struct btrfs_block_group_cache *cache;
1616 struct btrfs_fs_info *fs_info = root->fs_info;
1617
e8569813
ZY		 1618 while (num > 0) {
1619 cache = btrfs_lookup_block_group(fs_info, bytenr);
1620 BUG_ON(!cache);
1621 len = min(num, cache->key.offset -
1622 (bytenr - cache->key.objectid));
25179201
JB		 1623
1624 spin_lock(&cache->space_info->lock);
1625 spin_lock(&cache->lock);
e8569813 1626 if (reserve) {
e8569813
ZY		 1627 cache->reserved += len;
1628 cache->space_info->bytes_reserved += len;
e8569813 1629 } else {
e8569813
ZY		 1630 cache->reserved -= len;
1631 cache->space_info->bytes_reserved -= len;
e8569813 1632 }
25179201
JB		 1633 spin_unlock(&cache->lock);
1634 spin_unlock(&cache->space_info->lock);
e8569813
ZY		 1635 bytenr += len;
1636 num -= len;
1637 }
1638 return 0;
1639}
1640
d1310b2e 1641int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
ccd467d6 1642{
ccd467d6 1643 u64 last = 0;
1a5bc167
CM		 1644 u64 start;
1645 u64 end;
d1310b2e 1646 struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
ccd467d6 1647 int ret;
ccd467d6 1648
25179201 1649 mutex_lock(&root->fs_info->pinned_mutex);
ccd467d6 1650 while(1) {
1a5bc167
CM		 1651 ret = find_first_extent_bit(pinned_extents, last,
1652 &start, &end, EXTENT_DIRTY);
1653 if (ret)
ccd467d6 1654 break;
1a5bc167
CM		 1655 set_extent_dirty(copy, start, end, GFP_NOFS);
1656 last = end + 1;
ccd467d6 1657 }
25179201 1658 mutex_unlock(&root->fs_info->pinned_mutex);
ccd467d6
CM		 1659 return 0;
1660}
1661
1662int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
1663 struct btrfs_root *root,
d1310b2e 1664 struct extent_io_tree *unpin)
a28ec197 1665{
1a5bc167
CM		 1666 u64 start;
1667 u64 end;
a28ec197 1668 int ret;
0f9dd46c 1669 struct btrfs_block_group_cache *cache;
a28ec197 1670
25179201 1671 mutex_lock(&root->fs_info->pinned_mutex);
a28ec197 1672 while(1) {
1a5bc167
CM		 1673 ret = find_first_extent_bit(unpin, 0, &start, &end,
1674 EXTENT_DIRTY);
1675 if (ret)
a28ec197 1676 break;
e02119d5 1677 btrfs_update_pinned_extents(root, start, end + 1 - start, 0);
1a5bc167 1678 clear_extent_dirty(unpin, start, end, GFP_NOFS);
0f9dd46c
JB		 1679 cache = btrfs_lookup_block_group(root->fs_info, start);
1680 if (cache->cached)
1681 btrfs_add_free_space(cache, start, end - start + 1);
c286ac48 1682 if (need_resched()) {
25179201 1683 mutex_unlock(&root->fs_info->pinned_mutex);
c286ac48 1684 cond_resched();
25179201 1685 mutex_lock(&root->fs_info->pinned_mutex);
c286ac48 1686 }
a28ec197 1687 }
25179201 1688 mutex_unlock(&root->fs_info->pinned_mutex);
a28ec197
CM		 1689 return 0;
1690}
1691
98ed5174 1692static int finish_current_insert(struct btrfs_trans_handle *trans,
87ef2bb4 1693 struct btrfs_root *extent_root, int all)
037e6390 1694{
7bb86316
CM		 1695 u64 start;
1696 u64 end;
31840ae1 1697 u64 priv;
25179201 1698 u64 search = 0;
7bb86316
CM		 1699 struct btrfs_fs_info *info = extent_root->fs_info;
1700 struct btrfs_path *path;
31840ae1
ZY		 1701 struct btrfs_extent_ref *ref;
1702 struct pending_extent_op *extent_op;
1703 struct btrfs_key key;
234b63a0 1704 struct btrfs_extent_item extent_item;
037e6390 1705 int ret;
1a5bc167 1706 int err = 0;
037e6390 1707
5f39d397 1708 btrfs_set_stack_extent_refs(&extent_item, 1);
7bb86316 1709 path = btrfs_alloc_path();
037e6390 1710
26b8003f 1711 while(1) {
25179201
JB		 1712 mutex_lock(&info->extent_ins_mutex);
1713 ret = find_first_extent_bit(&info->extent_ins, search, &start,
1714 &end, EXTENT_WRITEBACK);
1715 if (ret) {
1716 mutex_unlock(&info->extent_ins_mutex);
87ef2bb4 1717 if (search && all) {
25179201
JB		 1718 search = 0;
1719 continue;
1720 }
26b8003f 1721 break;
25179201
JB		 1722 }
1723
1724 ret = try_lock_extent(&info->extent_ins, start, end, GFP_NOFS);
1725 if (!ret) {
87ef2bb4 1726 search = end + 1;
25179201
JB		 1727 mutex_unlock(&info->extent_ins_mutex);
1728 cond_resched();
1729 continue;
1730 }
1731 BUG_ON(ret < 0);
26b8003f 1732
31840ae1
ZY		 1733 ret = get_state_private(&info->extent_ins, start, &priv);
1734 BUG_ON(ret);
1735 extent_op = (struct pending_extent_op *)(unsigned long)priv;
1736
25179201
JB		 1737 mutex_unlock(&info->extent_ins_mutex);
1738
31840ae1
ZY		 1739 if (extent_op->type == PENDING_EXTENT_INSERT) {
1740 key.objectid = start;
1741 key.offset = end + 1 - start;
1742 key.type = BTRFS_EXTENT_ITEM_KEY;
1743 err = btrfs_insert_item(trans, extent_root, &key,
1a5bc167 1744 &extent_item, sizeof(extent_item));
31840ae1 1745 BUG_ON(err);
c286ac48 1746
25179201 1747 mutex_lock(&info->extent_ins_mutex);
31840ae1 1748 clear_extent_bits(&info->extent_ins, start, end,
25179201
JB		 1749 EXTENT_WRITEBACK, GFP_NOFS);
1750 mutex_unlock(&info->extent_ins_mutex);
c286ac48 1751
31840ae1
ZY		 1752 err = insert_extent_backref(trans, extent_root, path,
1753 start, extent_op->parent,
1754 extent_root->root_key.objectid,
1755 extent_op->generation,
3bb1a1bc 1756 extent_op->level);
31840ae1
ZY		 1757 BUG_ON(err);
1758 } else if (extent_op->type == PENDING_BACKREF_UPDATE) {
1759 err = lookup_extent_backref(trans, extent_root, path,
1760 start, extent_op->orig_parent,
1761 extent_root->root_key.objectid,
3bb1a1bc
YZ		 1762 extent_op->orig_generation,
1763 extent_op->level, 0);
31840ae1 1764 BUG_ON(err);
c286ac48 1765
25179201 1766 mutex_lock(&info->extent_ins_mutex);
31840ae1 1767 clear_extent_bits(&info->extent_ins, start, end,
25179201
JB		 1768 EXTENT_WRITEBACK, GFP_NOFS);
1769 mutex_unlock(&info->extent_ins_mutex);
31840ae1
ZY		 1770
1771 key.objectid = start;
1772 key.offset = extent_op->parent;
1773 key.type = BTRFS_EXTENT_REF_KEY;
1774 err = btrfs_set_item_key_safe(trans, extent_root, path,
1775 &key);
1776 BUG_ON(err);
1777 ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1778 struct btrfs_extent_ref);
1779 btrfs_set_ref_generation(path->nodes[0], ref,
1780 extent_op->generation);
1781 btrfs_mark_buffer_dirty(path->nodes[0]);
1782 btrfs_release_path(extent_root, path);
d8d5f3e1 1783 } else {
31840ae1 1784 BUG_ON(1);
d8d5f3e1 1785 }
31840ae1 1786 kfree(extent_op);
25179201 1787 unlock_extent(&info->extent_ins, start, end, GFP_NOFS);
87ef2bb4
CM		 1788 if (all)
1789 search = 0;
1790 else
1791 search = end + 1;
31840ae1 1792
25179201 1793 cond_resched();
037e6390 1794 }
7bb86316 1795 btrfs_free_path(path);
037e6390
CM		 1796 return 0;
1797}
1798
31840ae1
ZY		 1799static int pin_down_bytes(struct btrfs_trans_handle *trans,
1800 struct btrfs_root *root,
1801 u64 bytenr, u64 num_bytes, int is_data)
e20d96d6 1802{
1a5bc167 1803 int err = 0;
31840ae1 1804 struct extent_buffer *buf;
8ef97622 1805
31840ae1
ZY		 1806 if (is_data)
1807 goto pinit;
1808
1809 buf = btrfs_find_tree_block(root, bytenr, num_bytes);
1810 if (!buf)
1811 goto pinit;
1812
1813 /* we can reuse a block if it hasn't been written
1814 * and it is from this transaction. We can't
1815 * reuse anything from the tree log root because
1816 * it has tiny sub-transactions.
1817 */
1818 if (btrfs_buffer_uptodate(buf, 0) &&
1819 btrfs_try_tree_lock(buf)) {
1820 u64 header_owner = btrfs_header_owner(buf);
1821 u64 header_transid = btrfs_header_generation(buf);
1822 if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
1a40e23b 1823 header_owner != BTRFS_TREE_RELOC_OBJECTID &&
31840ae1
ZY		 1824 header_transid == trans->transid &&
1825 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
1826 clean_tree_block(NULL, root, buf);
1827 btrfs_tree_unlock(buf);
5f39d397 1828 free_extent_buffer(buf);
31840ae1 1829 return 1;
8ef97622 1830 }
31840ae1 1831 btrfs_tree_unlock(buf);
f4b9aa8d 1832 }
31840ae1
ZY		 1833 free_extent_buffer(buf);
1834pinit:
1835 btrfs_update_pinned_extents(root, bytenr, num_bytes, 1);
1836
be744175 1837 BUG_ON(err < 0);
e20d96d6
CM		 1838 return 0;
1839}
1840
fec577fb 1841/*
a28ec197 1842 * remove an extent from the root, returns 0 on success
fec577fb 1843 */
31840ae1
ZY		 1844static int __free_extent(struct btrfs_trans_handle *trans,
1845 struct btrfs_root *root,
1846 u64 bytenr, u64 num_bytes, u64 parent,
7bb86316 1847 u64 root_objectid, u64 ref_generation,
3bb1a1bc 1848 u64 owner_objectid, int pin, int mark_free)
a28ec197 1849{
5caf2a00 1850 struct btrfs_path *path;
e2fa7227 1851 struct btrfs_key key;
1261ec42
CM		 1852 struct btrfs_fs_info *info = root->fs_info;
1853 struct btrfs_root *extent_root = info->extent_root;
5f39d397 1854 struct extent_buffer *leaf;
a28ec197 1855 int ret;
952fccac
CM		 1856 int extent_slot = 0;
1857 int found_extent = 0;
1858 int num_to_del = 1;
234b63a0 1859 struct btrfs_extent_item *ei;
cf27e1ee 1860 u32 refs;
037e6390 1861
db94535d 1862 key.objectid = bytenr;
62e2749e 1863 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
db94535d 1864 key.offset = num_bytes;
5caf2a00 1865 path = btrfs_alloc_path();
54aa1f4d
CM		 1866 if (!path)
1867 return -ENOMEM;
5f26f772 1868
3c12ac72 1869 path->reada = 1;
3bb1a1bc
YZ		 1870 ret = lookup_extent_backref(trans, extent_root, path,
1871 bytenr, parent, root_objectid,
1872 ref_generation, owner_objectid, 1);
7bb86316 1873 if (ret == 0) {
952fccac
CM		 1874 struct btrfs_key found_key;
1875 extent_slot = path->slots[0];
1876 while(extent_slot > 0) {
1877 extent_slot--;
1878 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1879 extent_slot);
1880 if (found_key.objectid != bytenr)
1881 break;
1882 if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
1883 found_key.offset == num_bytes) {
1884 found_extent = 1;
1885 break;
1886 }
1887 if (path->slots[0] - extent_slot > 5)
1888 break;
1889 }
31840ae1
ZY		 1890 if (!found_extent) {
1891 ret = remove_extent_backref(trans, extent_root, path);
1892 BUG_ON(ret);
1893 btrfs_release_path(extent_root, path);
1894 ret = btrfs_search_slot(trans, extent_root,
1895 &key, path, -1, 1);
1896 BUG_ON(ret);
1897 extent_slot = path->slots[0];
1898 }
7bb86316
CM		 1899 } else {
1900 btrfs_print_leaf(extent_root, path->nodes[0]);
1901 WARN_ON(1);
1902 printk("Unable to find ref byte nr %Lu root %Lu "
3bb1a1bc
YZ		 1903 "gen %Lu owner %Lu\n", bytenr,
1904 root_objectid, ref_generation, owner_objectid);
7bb86316 1905 }
5f39d397
CM		 1906
1907 leaf = path->nodes[0];
952fccac 1908 ei = btrfs_item_ptr(leaf, extent_slot,
123abc88 1909 struct btrfs_extent_item);
5f39d397
CM		 1910 refs = btrfs_extent_refs(leaf, ei);
1911 BUG_ON(refs == 0);
1912 refs -= 1;
1913 btrfs_set_extent_refs(leaf, ei, refs);
952fccac 1914
5f39d397
CM		 1915 btrfs_mark_buffer_dirty(leaf);
1916
952fccac 1917 if (refs == 0 && found_extent && path->slots[0] == extent_slot + 1) {
31840ae1
ZY		 1918 struct btrfs_extent_ref *ref;
1919 ref = btrfs_item_ptr(leaf, path->slots[0],
1920 struct btrfs_extent_ref);
1921 BUG_ON(btrfs_ref_num_refs(leaf, ref) != 1);
952fccac
CM		 1922 /* if the back ref and the extent are next to each other
1923 * they get deleted below in one shot
1924 */
1925 path->slots[0] = extent_slot;
1926 num_to_del = 2;
1927 } else if (found_extent) {
1928 /* otherwise delete the extent back ref */
31840ae1 1929 ret = remove_extent_backref(trans, extent_root, path);
952fccac
CM		 1930 BUG_ON(ret);
1931 /* if refs are 0, we need to setup the path for deletion */
1932 if (refs == 0) {
1933 btrfs_release_path(extent_root, path);
1934 ret = btrfs_search_slot(trans, extent_root, &key, path,
1935 -1, 1);
952fccac
CM		 1936 BUG_ON(ret);
1937 }
1938 }
1939
cf27e1ee 1940 if (refs == 0) {
db94535d
CM		 1941 u64 super_used;
1942 u64 root_used;
21af804c
DW		 1943#ifdef BIO_RW_DISCARD
1944 u64 map_length = num_bytes;
1945 struct btrfs_multi_bio *multi = NULL;
1946#endif
78fae27e
CM		 1947
1948 if (pin) {
25179201 1949 mutex_lock(&root->fs_info->pinned_mutex);
31840ae1
ZY		 1950 ret = pin_down_bytes(trans, root, bytenr, num_bytes,
1951 owner_objectid >= BTRFS_FIRST_FREE_OBJECTID);
25179201 1952 mutex_unlock(&root->fs_info->pinned_mutex);
c549228f
Y		 1953 if (ret > 0)
1954 mark_free = 1;
1955 BUG_ON(ret < 0);
78fae27e
CM		 1956 }
1957
58176a96 1958 /* block accounting for super block */
a2135011 1959 spin_lock_irq(&info->delalloc_lock);
db94535d
CM		 1960 super_used = btrfs_super_bytes_used(&info->super_copy);
1961 btrfs_set_super_bytes_used(&info->super_copy,
1962 super_used - num_bytes);
a2135011 1963 spin_unlock_irq(&info->delalloc_lock);
58176a96
JB		 1964
1965 /* block accounting for root item */
db94535d 1966 root_used = btrfs_root_used(&root->root_item);
5f39d397 1967 btrfs_set_root_used(&root->root_item,
db94535d 1968 root_used - num_bytes);
952fccac
CM		 1969 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
1970 num_to_del);
31840ae1 1971 BUG_ON(ret);
25179201 1972 btrfs_release_path(extent_root, path);
db94535d 1973 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
0b86a832 1974 mark_free);
9078a3e1 1975 BUG_ON(ret);
21af804c
DW		 1976
1977#ifdef BIO_RW_DISCARD
1978 /* Tell the block device(s) that the sectors can be discarded */
1979 ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
1980 bytenr, &map_length, &multi, 0);
1981 if (!ret) {
1982 struct btrfs_bio_stripe *stripe = multi->stripes;
1983 int i;
1984
1985 if (map_length > num_bytes)
1986 map_length = num_bytes;
1987
1988 for (i = 0; i < multi->num_stripes; i++, stripe++) {
1989 blkdev_issue_discard(stripe->dev->bdev,
1990 stripe->physical >> 9,
1991 map_length >> 9);
1992 }
1993 kfree(multi);
1994 }
1995#endif
a28ec197 1996 }
5caf2a00 1997 btrfs_free_path(path);
87ef2bb4 1998 finish_current_insert(trans, extent_root, 0);
a28ec197
CM		 1999 return ret;
2000}
2001
a28ec197
CM		 2002/*
2003 * find all the blocks marked as pending in the radix tree and remove
2004 * them from the extent map
2005 */
e089f05c 2006static int del_pending_extents(struct btrfs_trans_handle *trans, struct
87ef2bb4 2007 btrfs_root *extent_root, int all)
a28ec197
CM		 2008{
2009 int ret;
e20d96d6 2010 int err = 0;
1a5bc167
CM		 2011 u64 start;
2012 u64 end;
31840ae1 2013 u64 priv;
25179201 2014 u64 search = 0;
d1310b2e 2015 struct extent_io_tree *pending_del;
31840ae1
ZY		 2016 struct extent_io_tree *extent_ins;
2017 struct pending_extent_op *extent_op;
25179201 2018 struct btrfs_fs_info *info = extent_root->fs_info;
8ef97622 2019
31840ae1 2020 extent_ins = &extent_root->fs_info->extent_ins;
1a5bc167 2021 pending_del = &extent_root->fs_info->pending_del;
a28ec197
CM		 2022
2023 while(1) {
25179201
JB		 2024 mutex_lock(&info->extent_ins_mutex);
2025 ret = find_first_extent_bit(pending_del, search, &start, &end,
2026 EXTENT_WRITEBACK);
2027 if (ret) {
2028 mutex_unlock(&info->extent_ins_mutex);
87ef2bb4 2029 if (all && search) {
25179201
JB		 2030 search = 0;
2031 continue;
2032 }
a28ec197 2033 break;
25179201
JB		 2034 }
2035
2036 ret = try_lock_extent(extent_ins, start, end, GFP_NOFS);
2037 if (!ret) {
2038 search = end+1;
2039 mutex_unlock(&info->extent_ins_mutex);
2040 cond_resched();
2041 continue;
2042 }
2043 BUG_ON(ret < 0);
31840ae1
ZY		 2044
2045 ret = get_state_private(pending_del, start, &priv);
2046 BUG_ON(ret);
2047 extent_op = (struct pending_extent_op *)(unsigned long)priv;
2048
25179201 2049 clear_extent_bits(pending_del, start, end, EXTENT_WRITEBACK,
1a5bc167 2050 GFP_NOFS);
31840ae1 2051 if (!test_range_bit(extent_ins, start, end,
25179201
JB		 2052 EXTENT_WRITEBACK, 0)) {
2053 mutex_unlock(&info->extent_ins_mutex);
31840ae1 2054free_extent:
c286ac48 2055 ret = __free_extent(trans, extent_root,
31840ae1
ZY		 2056 start, end + 1 - start,
2057 extent_op->orig_parent,
2058 extent_root->root_key.objectid,
2059 extent_op->orig_generation,
25179201 2060 extent_op->level, 1, 0);
31840ae1 2061 kfree(extent_op);
c286ac48 2062 } else {
31840ae1 2063 kfree(extent_op);
25179201
JB		 2064
2065 ret = get_state_private(&info->extent_ins, start,
2066 &priv);
31840ae1
ZY		 2067 BUG_ON(ret);
2068 extent_op = (struct pending_extent_op *)
25179201
JB		 2069 (unsigned long)priv;
2070
2071 clear_extent_bits(&info->extent_ins, start, end,
2072 EXTENT_WRITEBACK, GFP_NOFS);
31840ae1 2073
25179201 2074 mutex_unlock(&info->extent_ins_mutex);
31840ae1
ZY		 2075
2076 if (extent_op->type == PENDING_BACKREF_UPDATE)
2077 goto free_extent;
2078
25179201
JB		 2079 mutex_lock(&extent_root->fs_info->pinned_mutex);
2080 ret = pin_down_bytes(trans, extent_root, start,
2081 end + 1 - start, 0);
2082 mutex_unlock(&extent_root->fs_info->pinned_mutex);
2083
31840ae1 2084 ret = update_block_group(trans, extent_root, start,
25179201
JB		 2085 end + 1 - start, 0, ret > 0);
2086
31840ae1
ZY		 2087 BUG_ON(ret);
2088 kfree(extent_op);
c286ac48 2089 }
1a5bc167
CM		 2090 if (ret)
2091 err = ret;
25179201 2092 unlock_extent(extent_ins, start, end, GFP_NOFS);
c286ac48 2093
87ef2bb4
CM		 2094 if (all)
2095 search = 0;
2096 else
2097 search = end + 1;
25179201 2098 cond_resched();
fec577fb 2099 }
e20d96d6 2100 return err;
fec577fb
CM		 2101}
2102
2103/*
2104 * remove an extent from the root, returns 0 on success
2105 */
925baedd 2106static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
31840ae1
ZY		 2107 struct btrfs_root *root,
2108 u64 bytenr, u64 num_bytes, u64 parent,
2109 u64 root_objectid, u64 ref_generation,
3bb1a1bc 2110 u64 owner_objectid, int pin)
fec577fb 2111{
9f5fae2f 2112 struct btrfs_root *extent_root = root->fs_info->extent_root;
fec577fb
CM		 2113 int pending_ret;
2114 int ret;
a28ec197 2115
db94535d 2116 WARN_ON(num_bytes < root->sectorsize);
fec577fb 2117 if (root == extent_root) {
31840ae1
ZY		 2118 struct pending_extent_op *extent_op;
2119
2120 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2121 BUG_ON(!extent_op);
2122
2123 extent_op->type = PENDING_EXTENT_DELETE;
2124 extent_op->bytenr = bytenr;
2125 extent_op->num_bytes = num_bytes;
2126 extent_op->parent = parent;
2127 extent_op->orig_parent = parent;
2128 extent_op->generation = ref_generation;
2129 extent_op->orig_generation = ref_generation;
2130 extent_op->level = (int)owner_objectid;
2131
25179201 2132 mutex_lock(&root->fs_info->extent_ins_mutex);
31840ae1
ZY		 2133 set_extent_bits(&root->fs_info->pending_del,
2134 bytenr, bytenr + num_bytes - 1,
25179201 2135 EXTENT_WRITEBACK, GFP_NOFS);
31840ae1
ZY		 2136 set_state_private(&root->fs_info->pending_del,
2137 bytenr, (unsigned long)extent_op);
25179201 2138 mutex_unlock(&root->fs_info->extent_ins_mutex);
fec577fb
CM		 2139 return 0;
2140 }
4bef0848 2141 /* if metadata always pin */
d00aff00
CM		 2142 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
2143 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
0f9dd46c
JB		 2144 struct btrfs_block_group_cache *cache;
2145
d00aff00 2146 /* btrfs_free_reserved_extent */
0f9dd46c
JB		 2147 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
2148 BUG_ON(!cache);
2149 btrfs_add_free_space(cache, bytenr, num_bytes);
e8569813 2150 update_reserved_extents(root, bytenr, num_bytes, 0);
d00aff00
CM		 2151 return 0;
2152 }
4bef0848 2153 pin = 1;
d00aff00 2154 }
4bef0848
CM		 2155
2156 /* if data pin when any transaction has committed this */
2157 if (ref_generation != trans->transid)
2158 pin = 1;
2159
31840ae1 2160 ret = __free_extent(trans, root, bytenr, num_bytes, parent,
3bb1a1bc
YZ		 2161 root_objectid, ref_generation,
2162 owner_objectid, pin, pin == 0);
ee6e6504 2163
87ef2bb4
CM		 2164 finish_current_insert(trans, root->fs_info->extent_root, 0);
2165 pending_ret = del_pending_extents(trans, root->fs_info->extent_root, 0);
fec577fb
CM		 2166 return ret ? ret : pending_ret;
2167}
2168
925baedd 2169int btrfs_free_extent(struct btrfs_trans_handle *trans,
31840ae1
ZY		 2170 struct btrfs_root *root,
2171 u64 bytenr, u64 num_bytes, u64 parent,
2172 u64 root_objectid, u64 ref_generation,
3bb1a1bc 2173 u64 owner_objectid, int pin)
925baedd
CM		 2174{
2175 int ret;
2176
31840ae1 2177 ret = __btrfs_free_extent(trans, root, bytenr, num_bytes, parent,
925baedd 2178 root_objectid, ref_generation,
3bb1a1bc 2179 owner_objectid, pin);
925baedd
CM		 2180 return ret;
2181}
2182
87ee04eb
CM		 2183static u64 stripe_align(struct btrfs_root *root, u64 val)
2184{
2185 u64 mask = ((u64)root->stripesize - 1);
2186 u64 ret = (val + mask) & ~mask;
2187 return ret;
2188}
2189
fec577fb
CM		 2190/*
2191 * walks the btree of allocated extents and find a hole of a given size.
2192 * The key ins is changed to record the hole:
2193 * ins->objectid == block start
62e2749e 2194 * ins->flags = BTRFS_EXTENT_ITEM_KEY
fec577fb
CM		 2195 * ins->offset == number of blocks
2196 * Any available blocks before search_start are skipped.
2197 */
98ed5174
CM		 2198static int noinline find_free_extent(struct btrfs_trans_handle *trans,
2199 struct btrfs_root *orig_root,
2200 u64 num_bytes, u64 empty_size,
2201 u64 search_start, u64 search_end,
2202 u64 hint_byte, struct btrfs_key *ins,
2203 u64 exclude_start, u64 exclude_nr,
2204 int data)
fec577fb 2205{
80eb234a 2206 int ret = 0;
9f5fae2f 2207 struct btrfs_root * root = orig_root->fs_info->extent_root;
db94535d 2208 u64 total_needed = num_bytes;
239b14b3 2209 u64 *last_ptr = NULL;
80eb234a 2210 struct btrfs_block_group_cache *block_group = NULL;
0ef3e66b 2211 int chunk_alloc_done = 0;
239b14b3 2212 int empty_cluster = 2 * 1024 * 1024;
0ef3e66b 2213 int allowed_chunk_alloc = 0;
80eb234a
JB		 2214 struct list_head *head = NULL, *cur = NULL;
2215 int loop = 0;
2216 struct btrfs_space_info *space_info;
fec577fb 2217
db94535d 2218 WARN_ON(num_bytes < root->sectorsize);
b1a4d965 2219 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
80eb234a
JB		 2220 ins->objectid = 0;
2221 ins->offset = 0;
b1a4d965 2222
0ef3e66b
CM		 2223 if (orig_root->ref_cows || empty_size)
2224 allowed_chunk_alloc = 1;
2225
239b14b3
CM		 2226 if (data & BTRFS_BLOCK_GROUP_METADATA) {
2227 last_ptr = &root->fs_info->last_alloc;
8790d502 2228 empty_cluster = 256 * 1024;
239b14b3
CM		 2229 }
2230
0f9dd46c 2231 if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD))
239b14b3 2232 last_ptr = &root->fs_info->last_data_alloc;
0f9dd46c 2233
239b14b3
CM		 2234 if (last_ptr) {
2235 if (*last_ptr)
2236 hint_byte = *last_ptr;
0f9dd46c 2237 else
239b14b3 2238 empty_size += empty_cluster;
239b14b3 2239 }
a061fc8d 2240 search_start = max(search_start, first_logical_byte(root, 0));
239b14b3 2241 search_start = max(search_start, hint_byte);
6702ed49 2242 total_needed += empty_size;
0b86a832 2243
80eb234a
JB		 2244 block_group = btrfs_lookup_block_group(root->fs_info, search_start);
2245 space_info = __find_space_info(root->fs_info, data);
0f9dd46c 2246
80eb234a
JB		 2247 down_read(&space_info->groups_sem);
2248 while (1) {
2249 struct btrfs_free_space *free_space;
0f9dd46c 2250 /*
80eb234a
JB		 2251 * the only way this happens if our hint points to a block
2252 * group thats not of the proper type, while looping this
2253 * should never happen
0f9dd46c 2254 */
25179201
JB		 2255 WARN_ON(!block_group);
2256 mutex_lock(&block_group->alloc_mutex);
80eb234a 2257 if (unlikely(!block_group_bits(block_group, data)))
0f9dd46c 2258 goto new_group;
0f9dd46c 2259
80eb234a 2260 ret = cache_block_group(root, block_group);
25179201
JB		 2261 if (ret) {
2262 mutex_unlock(&block_group->alloc_mutex);
80eb234a 2263 break;
25179201 2264 }
0f9dd46c 2265
80eb234a
JB		 2266 if (block_group->ro)
2267 goto new_group;
0f9dd46c 2268
80eb234a
JB		 2269 free_space = btrfs_find_free_space(block_group, search_start,
2270 total_needed);
2271 if (free_space) {
2272 u64 start = block_group->key.objectid;
2273 u64 end = block_group->key.objectid +
0f9dd46c 2274 block_group->key.offset;
239b14b3 2275
80eb234a 2276 search_start = stripe_align(root, free_space->offset);
0f9dd46c 2277
80eb234a
JB		 2278 /* move on to the next group */
2279 if (search_start + num_bytes >= search_end)
2280 goto new_group;
e37c9e69 2281
80eb234a
JB		 2282 /* move on to the next group */
2283 if (search_start + num_bytes > end)
2284 goto new_group;
2285
2286 if (exclude_nr > 0 &&
2287 (search_start + num_bytes > exclude_start &&
2288 search_start < exclude_start + exclude_nr)) {
2289 search_start = exclude_start + exclude_nr;
2290 /*
2291 * if search_start is still in this block group
2292 * then we just re-search this block group
2293 */
2294 if (search_start >= start &&
25179201
JB		 2295 search_start < end) {
2296 mutex_unlock(&block_group->alloc_mutex);
80eb234a 2297 continue;
25179201 2298 }
80eb234a
JB		 2299
2300 /* else we go to the next block group */
2301 goto new_group;
2302 }
2303
2304 ins->objectid = search_start;
2305 ins->offset = num_bytes;
25179201
JB		 2306
2307 btrfs_remove_free_space_lock(block_group, search_start,
2308 num_bytes);
80eb234a 2309 /* we are all good, lets return */
25179201 2310 mutex_unlock(&block_group->alloc_mutex);
80eb234a 2311 break;
0f9dd46c 2312 }
80eb234a 2313new_group:
25179201 2314 mutex_unlock(&block_group->alloc_mutex);
80eb234a
JB		 2315 /*
2316 * Here's how this works.
2317 * loop == 0: we were searching a block group via a hint
2318 * and didn't find anything, so we start at
2319 * the head of the block groups and keep searching
2320 * loop == 1: we're searching through all of the block groups
2321 * if we hit the head again we have searched
2322 * all of the block groups for this space and we
2323 * need to try and allocate, if we cant error out.
2324 * loop == 2: we allocated more space and are looping through
2325 * all of the block groups again.
2326 */
2327 if (loop == 0) {
2328 head = &space_info->block_groups;
2329 cur = head->next;
0b86a832 2330
80eb234a
JB		 2331 if (last_ptr && *last_ptr) {
2332 total_needed += empty_cluster;
2333 *last_ptr = 0;
2334 }
2335 loop++;
2336 } else if (loop == 1 && cur == head) {
2337 if (allowed_chunk_alloc && !chunk_alloc_done) {
2338 up_read(&space_info->groups_sem);
2339 ret = do_chunk_alloc(trans, root, num_bytes +
2340 2 * 1024 * 1024, data, 1);
2341 if (ret < 0)
2342 break;
2343 down_read(&space_info->groups_sem);
2344 loop++;
2345 head = &space_info->block_groups;
2346 cur = head->next;
2347 chunk_alloc_done = 1;
2348 } else if (!allowed_chunk_alloc) {
2349 space_info->force_alloc = 1;
2350 break;
2351 } else {
2352 break;
2353 }
2354 } else if (cur == head) {
2355 break;
0f9dd46c 2356 }
0b86a832 2357
80eb234a
JB		 2358 block_group = list_entry(cur, struct btrfs_block_group_cache,
2359 list);
2360 search_start = block_group->key.objectid;
2361 cur = cur->next;
f2654de4 2362 }
0b86a832 2363
80eb234a
JB		 2364 /* we found what we needed */
2365 if (ins->objectid) {
2366 if (!(data & BTRFS_BLOCK_GROUP_DATA))
2367 trans->block_group = block_group;
0f9dd46c 2368
80eb234a
JB		 2369 if (last_ptr)
2370 *last_ptr = ins->objectid + ins->offset;
2371 ret = 0;
2372 } else if (!ret) {
2373 ret = -ENOSPC;
be744175 2374 }
be744175 2375
80eb234a 2376 up_read(&space_info->groups_sem);
0f70abe2 2377 return ret;
fec577fb 2378}
ec44a35c 2379
0f9dd46c
JB		 2380static void dump_space_info(struct btrfs_space_info *info, u64 bytes)
2381{
2382 struct btrfs_block_group_cache *cache;
2383 struct list_head *l;
2384
2385 printk(KERN_INFO "space_info has %Lu free, is %sfull\n",
e8569813
ZY		 2386 info->total_bytes - info->bytes_used - info->bytes_pinned -
2387 info->bytes_reserved, (info->full) ? "" : "not ");
0f9dd46c 2388
80eb234a 2389 down_read(&info->groups_sem);
0f9dd46c
JB		 2390 list_for_each(l, &info->block_groups) {
2391 cache = list_entry(l, struct btrfs_block_group_cache, list);
2392 spin_lock(&cache->lock);
2393 printk(KERN_INFO "block group %Lu has %Lu bytes, %Lu used "
e8569813 2394 "%Lu pinned %Lu reserved\n",
0f9dd46c 2395 cache->key.objectid, cache->key.offset,
e8569813
ZY		 2396 btrfs_block_group_used(&cache->item),
2397 cache->pinned, cache->reserved);
0f9dd46c
JB		 2398 btrfs_dump_free_space(cache, bytes);
2399 spin_unlock(&cache->lock);
2400 }
80eb234a 2401 up_read(&info->groups_sem);
0f9dd46c 2402}
e8569813 2403
e6dcd2dc
CM		 2404static int __btrfs_reserve_extent(struct btrfs_trans_handle *trans,
2405 struct btrfs_root *root,
2406 u64 num_bytes, u64 min_alloc_size,
2407 u64 empty_size, u64 hint_byte,
2408 u64 search_end, struct btrfs_key *ins,
2409 u64 data)
fec577fb
CM		 2410{
2411 int ret;
fbdc762b 2412 u64 search_start = 0;
8790d502 2413 u64 alloc_profile;
1261ec42 2414 struct btrfs_fs_info *info = root->fs_info;
925baedd 2415
6324fbf3 2416 if (data) {
8790d502
CM		 2417 alloc_profile = info->avail_data_alloc_bits &
2418 info->data_alloc_profile;
2419 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
6324fbf3 2420 } else if (root == root->fs_info->chunk_root) {
8790d502
CM		 2421 alloc_profile = info->avail_system_alloc_bits &
2422 info->system_alloc_profile;
2423 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
6324fbf3 2424 } else {
8790d502
CM		 2425 alloc_profile = info->avail_metadata_alloc_bits &
2426 info->metadata_alloc_profile;
2427 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
6324fbf3 2428 }
98d20f67 2429again:
a061fc8d 2430 data = reduce_alloc_profile(root, data);
0ef3e66b
CM		 2431 /*
2432 * the only place that sets empty_size is btrfs_realloc_node, which
2433 * is not called recursively on allocations
2434 */
2435 if (empty_size || root->ref_cows) {
593060d7 2436 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
6324fbf3 2437 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
0ef3e66b
CM		 2438 2 * 1024 * 1024,
2439 BTRFS_BLOCK_GROUP_METADATA |
2440 (info->metadata_alloc_profile &
2441 info->avail_metadata_alloc_bits), 0);
6324fbf3
CM		 2442 }
2443 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
0ef3e66b 2444 num_bytes + 2 * 1024 * 1024, data, 0);
6324fbf3 2445 }
0b86a832 2446
db94535d
CM		 2447 WARN_ON(num_bytes < root->sectorsize);
2448 ret = find_free_extent(trans, root, num_bytes, empty_size,
2449 search_start, search_end, hint_byte, ins,
26b8003f
CM		 2450 trans->alloc_exclude_start,
2451 trans->alloc_exclude_nr, data);
3b951516 2452
98d20f67
CM		 2453 if (ret == -ENOSPC && num_bytes > min_alloc_size) {
2454 num_bytes = num_bytes >> 1;
0f9dd46c 2455 num_bytes = num_bytes & ~(root->sectorsize - 1);
98d20f67 2456 num_bytes = max(num_bytes, min_alloc_size);
0ef3e66b
CM		 2457 do_chunk_alloc(trans, root->fs_info->extent_root,
2458 num_bytes, data, 1);
98d20f67
CM		 2459 goto again;
2460 }
ec44a35c 2461 if (ret) {
0f9dd46c
JB		 2462 struct btrfs_space_info *sinfo;
2463
2464 sinfo = __find_space_info(root->fs_info, data);
2465 printk("allocation failed flags %Lu, wanted %Lu\n",
2466 data, num_bytes);
2467 dump_space_info(sinfo, num_bytes);
925baedd 2468 BUG();
925baedd 2469 }
0f9dd46c
JB		 2470
2471 return ret;
e6dcd2dc
CM		 2472}
2473
65b51a00
CM		 2474int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
2475{
0f9dd46c
JB		 2476 struct btrfs_block_group_cache *cache;
2477
0f9dd46c
JB		 2478 cache = btrfs_lookup_block_group(root->fs_info, start);
2479 if (!cache) {
2480 printk(KERN_ERR "Unable to find block group for %Lu\n", start);
0f9dd46c
JB		 2481 return -ENOSPC;
2482 }
2483 btrfs_add_free_space(cache, start, len);
1a40e23b 2484 update_reserved_extents(root, start, len, 0);
65b51a00
CM		 2485 return 0;
2486}
2487
e6dcd2dc
CM		 2488int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
2489 struct btrfs_root *root,
2490 u64 num_bytes, u64 min_alloc_size,
2491 u64 empty_size, u64 hint_byte,
2492 u64 search_end, struct btrfs_key *ins,
2493 u64 data)
2494{
2495 int ret;
e6dcd2dc
CM		 2496 ret = __btrfs_reserve_extent(trans, root, num_bytes, min_alloc_size,
2497 empty_size, hint_byte, search_end, ins,
2498 data);
e8569813 2499 update_reserved_extents(root, ins->objectid, ins->offset, 1);
e6dcd2dc
CM		 2500 return ret;
2501}
2502
2503static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
31840ae1 2504 struct btrfs_root *root, u64 parent,
e6dcd2dc 2505 u64 root_objectid, u64 ref_generation,
3bb1a1bc 2506 u64 owner, struct btrfs_key *ins)
e6dcd2dc
CM		 2507{
2508 int ret;
2509 int pending_ret;
2510 u64 super_used;
2511 u64 root_used;
2512 u64 num_bytes = ins->offset;
2513 u32 sizes[2];
2514 struct btrfs_fs_info *info = root->fs_info;
2515 struct btrfs_root *extent_root = info->extent_root;
2516 struct btrfs_extent_item *extent_item;
2517 struct btrfs_extent_ref *ref;
2518 struct btrfs_path *path;
2519 struct btrfs_key keys[2];
fec577fb 2520
31840ae1
ZY		 2521 if (parent == 0)
2522 parent = ins->objectid;
2523
58176a96 2524 /* block accounting for super block */
a2135011 2525 spin_lock_irq(&info->delalloc_lock);
db94535d
CM		 2526 super_used = btrfs_super_bytes_used(&info->super_copy);
2527 btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes);
a2135011 2528 spin_unlock_irq(&info->delalloc_lock);
26b8003f 2529
58176a96 2530 /* block accounting for root item */
db94535d
CM		 2531 root_used = btrfs_root_used(&root->root_item);
2532 btrfs_set_root_used(&root->root_item, root_used + num_bytes);
58176a96 2533
26b8003f 2534 if (root == extent_root) {
31840ae1
ZY		 2535 struct pending_extent_op *extent_op;
2536
2537 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2538 BUG_ON(!extent_op);
2539
2540 extent_op->type = PENDING_EXTENT_INSERT;
2541 extent_op->bytenr = ins->objectid;
2542 extent_op->num_bytes = ins->offset;
2543 extent_op->parent = parent;
2544 extent_op->orig_parent = 0;
2545 extent_op->generation = ref_generation;
2546 extent_op->orig_generation = 0;
2547 extent_op->level = (int)owner;
2548
25179201 2549 mutex_lock(&root->fs_info->extent_ins_mutex);
1a5bc167
CM		 2550 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
2551 ins->objectid + ins->offset - 1,
25179201 2552 EXTENT_WRITEBACK, GFP_NOFS);
31840ae1
ZY		 2553 set_state_private(&root->fs_info->extent_ins,
2554 ins->objectid, (unsigned long)extent_op);
25179201 2555 mutex_unlock(&root->fs_info->extent_ins_mutex);
26b8003f
CM		 2556 goto update_block;
2557 }
2558
47e4bb98 2559 memcpy(&keys[0], ins, sizeof(*ins));
47e4bb98
CM		 2560 keys[1].objectid = ins->objectid;
2561 keys[1].type = BTRFS_EXTENT_REF_KEY;
31840ae1 2562 keys[1].offset = parent;
47e4bb98
CM		 2563 sizes[0] = sizeof(*extent_item);
2564 sizes[1] = sizeof(*ref);
7bb86316
CM		 2565
2566 path = btrfs_alloc_path();
2567 BUG_ON(!path);
47e4bb98
CM		 2568
2569 ret = btrfs_insert_empty_items(trans, extent_root, path, keys,
2570 sizes, 2);
ccd467d6 2571 BUG_ON(ret);
0f9dd46c 2572
47e4bb98
CM		 2573 extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2574 struct btrfs_extent_item);
2575 btrfs_set_extent_refs(path->nodes[0], extent_item, 1);
2576 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
2577 struct btrfs_extent_ref);
2578
2579 btrfs_set_ref_root(path->nodes[0], ref, root_objectid);
2580 btrfs_set_ref_generation(path->nodes[0], ref, ref_generation);
2581 btrfs_set_ref_objectid(path->nodes[0], ref, owner);
31840ae1 2582 btrfs_set_ref_num_refs(path->nodes[0], ref, 1);
47e4bb98
CM		 2583
2584 btrfs_mark_buffer_dirty(path->nodes[0]);
2585
2586 trans->alloc_exclude_start = 0;
2587 trans->alloc_exclude_nr = 0;
7bb86316 2588 btrfs_free_path(path);
87ef2bb4
CM		 2589 finish_current_insert(trans, extent_root, 0);
2590 pending_ret = del_pending_extents(trans, extent_root, 0);
f510cfec 2591
925baedd
CM		 2592 if (ret)
2593 goto out;
e37c9e69 2594 if (pending_ret) {
925baedd
CM		 2595 ret = pending_ret;
2596 goto out;
e37c9e69 2597 }
26b8003f
CM		 2598
2599update_block:
0b86a832 2600 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1, 0);
f5947066
CM		 2601 if (ret) {
2602 printk("update block group failed for %Lu %Lu\n",
2603 ins->objectid, ins->offset);
2604 BUG();
2605 }
925baedd 2606out:
e6dcd2dc
CM		 2607 return ret;
2608}
2609
2610int btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
31840ae1 2611 struct btrfs_root *root, u64 parent,
e6dcd2dc 2612 u64 root_objectid, u64 ref_generation,
3bb1a1bc 2613 u64 owner, struct btrfs_key *ins)
e6dcd2dc
CM		 2614{
2615 int ret;
1c2308f8
CM		 2616
2617 if (root_objectid == BTRFS_TREE_LOG_OBJECTID)
2618 return 0;
3bb1a1bc
YZ		 2619 ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
2620 ref_generation, owner, ins);
e8569813 2621 update_reserved_extents(root, ins->objectid, ins->offset, 0);
e6dcd2dc
CM		 2622 return ret;
2623}
e02119d5
CM		 2624
2625/*
2626 * this is used by the tree logging recovery code. It records that
2627 * an extent has been allocated and makes sure to clear the free
2628 * space cache bits as well
2629 */
2630int btrfs_alloc_logged_extent(struct btrfs_trans_handle *trans,
31840ae1 2631 struct btrfs_root *root, u64 parent,
e02119d5 2632 u64 root_objectid, u64 ref_generation,
3bb1a1bc 2633 u64 owner, struct btrfs_key *ins)
e02119d5
CM		 2634{
2635 int ret;
2636 struct btrfs_block_group_cache *block_group;
2637
e02119d5 2638 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
25179201 2639 mutex_lock(&block_group->alloc_mutex);
e02119d5
CM		 2640 cache_block_group(root, block_group);
2641
25179201
JB		 2642 ret = btrfs_remove_free_space_lock(block_group, ins->objectid,
2643 ins->offset);
2644 mutex_unlock(&block_group->alloc_mutex);
0f9dd46c 2645 BUG_ON(ret);
3bb1a1bc
YZ		 2646 ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
2647 ref_generation, owner, ins);
e02119d5
CM		 2648 return ret;
2649}
2650
e6dcd2dc
CM		 2651/*
2652 * finds a free extent and does all the dirty work required for allocation
2653 * returns the key for the extent through ins, and a tree buffer for
2654 * the first block of the extent through buf.
2655 *
2656 * returns 0 if everything worked, non-zero otherwise.
2657 */
2658int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
2659 struct btrfs_root *root,
31840ae1 2660 u64 num_bytes, u64 parent, u64 min_alloc_size,
e6dcd2dc 2661 u64 root_objectid, u64 ref_generation,
3bb1a1bc 2662 u64 owner_objectid, u64 empty_size, u64 hint_byte,
e6dcd2dc
CM		 2663 u64 search_end, struct btrfs_key *ins, u64 data)
2664{
2665 int ret;
2666
e6dcd2dc
CM		 2667 ret = __btrfs_reserve_extent(trans, root, num_bytes,
2668 min_alloc_size, empty_size, hint_byte,
2669 search_end, ins, data);
2670 BUG_ON(ret);
d00aff00 2671 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
31840ae1
ZY		 2672 ret = __btrfs_alloc_reserved_extent(trans, root, parent,
2673 root_objectid, ref_generation,
3bb1a1bc 2674 owner_objectid, ins);
d00aff00 2675 BUG_ON(ret);
e6dcd2dc 2676
e8569813
ZY		 2677 } else {
2678 update_reserved_extents(root, ins->objectid, ins->offset, 1);
d00aff00 2679 }
925baedd 2680 return ret;
fec577fb 2681}
65b51a00
CM		 2682
2683struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
2684 struct btrfs_root *root,
2685 u64 bytenr, u32 blocksize)
2686{
2687 struct extent_buffer *buf;
2688
2689 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
2690 if (!buf)
2691 return ERR_PTR(-ENOMEM);
2692 btrfs_set_header_generation(buf, trans->transid);
2693 btrfs_tree_lock(buf);
2694 clean_tree_block(trans, root, buf);
2695 btrfs_set_buffer_uptodate(buf);
d0c803c4
CM		 2696 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
2697 set_extent_dirty(&root->dirty_log_pages, buf->start,
2698 buf->start + buf->len - 1, GFP_NOFS);
2699 } else {
2700 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
65b51a00 2701 buf->start + buf->len - 1, GFP_NOFS);
d0c803c4 2702 }
65b51a00
CM		 2703 trans->blocks_used++;
2704 return buf;
2705}
2706
fec577fb
CM		 2707/*
2708 * helper function to allocate a block for a given tree
2709 * returns the tree buffer or NULL.
2710 */
5f39d397 2711struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
7bb86316 2712 struct btrfs_root *root,
31840ae1 2713 u32 blocksize, u64 parent,
7bb86316
CM		 2714 u64 root_objectid,
2715 u64 ref_generation,
7bb86316
CM		 2716 int level,
2717 u64 hint,
5f39d397 2718 u64 empty_size)
fec577fb 2719{
e2fa7227 2720 struct btrfs_key ins;
fec577fb 2721 int ret;
5f39d397 2722 struct extent_buffer *buf;
fec577fb 2723
31840ae1 2724 ret = btrfs_alloc_extent(trans, root, blocksize, parent, blocksize,
3bb1a1bc 2725 root_objectid, ref_generation, level,
31840ae1 2726 empty_size, hint, (u64)-1, &ins, 0);
fec577fb 2727 if (ret) {
54aa1f4d
CM		 2728 BUG_ON(ret > 0);
2729 return ERR_PTR(ret);
fec577fb 2730 }
55c69072 2731
65b51a00 2732 buf = btrfs_init_new_buffer(trans, root, ins.objectid, blocksize);
fec577fb
CM		 2733 return buf;
2734}
a28ec197 2735
e02119d5
CM		 2736int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans,
2737 struct btrfs_root *root, struct extent_buffer *leaf)
6407bf6d 2738{
7bb86316
CM		 2739 u64 leaf_owner;
2740 u64 leaf_generation;
5f39d397 2741 struct btrfs_key key;
6407bf6d
CM		 2742 struct btrfs_file_extent_item *fi;
2743 int i;
2744 int nritems;
2745 int ret;
2746
5f39d397
CM		 2747 BUG_ON(!btrfs_is_leaf(leaf));
2748 nritems = btrfs_header_nritems(leaf);
7bb86316
CM		 2749 leaf_owner = btrfs_header_owner(leaf);
2750 leaf_generation = btrfs_header_generation(leaf);
2751
6407bf6d 2752 for (i = 0; i < nritems; i++) {
db94535d 2753 u64 disk_bytenr;
e34a5b4f 2754 cond_resched();
5f39d397
CM		 2755
2756 btrfs_item_key_to_cpu(leaf, &key, i);
2757 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
6407bf6d
CM		 2758 continue;
2759 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
5f39d397
CM		 2760 if (btrfs_file_extent_type(leaf, fi) ==
2761 BTRFS_FILE_EXTENT_INLINE)
236454df 2762 continue;
6407bf6d
CM		 2763 /*
2764 * FIXME make sure to insert a trans record that
2765 * repeats the snapshot del on crash
2766 */
db94535d
CM		 2767 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
2768 if (disk_bytenr == 0)
3a686375 2769 continue;
4a096752 2770
925baedd 2771 ret = __btrfs_free_extent(trans, root, disk_bytenr,
7bb86316 2772 btrfs_file_extent_disk_num_bytes(leaf, fi),
31840ae1 2773 leaf->start, leaf_owner, leaf_generation,
3bb1a1bc 2774 key.objectid, 0);
31840ae1 2775 BUG_ON(ret);
2dd3e67b
CM		 2776
2777 atomic_inc(&root->fs_info->throttle_gen);
2778 wake_up(&root->fs_info->transaction_throttle);
2779 cond_resched();
6407bf6d
CM		 2780 }
2781 return 0;
2782}
2783
e02119d5
CM		 2784static int noinline cache_drop_leaf_ref(struct btrfs_trans_handle *trans,
2785 struct btrfs_root *root,
2786 struct btrfs_leaf_ref *ref)
31153d81
YZ		 2787{
2788 int i;
2789 int ret;
2790 struct btrfs_extent_info *info = ref->extents;
2791
31153d81 2792 for (i = 0; i < ref->nritems; i++) {
31840ae1
ZY		 2793 ret = __btrfs_free_extent(trans, root, info->bytenr,
2794 info->num_bytes, ref->bytenr,
2795 ref->owner, ref->generation,
3bb1a1bc 2796 info->objectid, 0);
2dd3e67b
CM		 2797
2798 atomic_inc(&root->fs_info->throttle_gen);
2799 wake_up(&root->fs_info->transaction_throttle);
2800 cond_resched();
2801
31153d81
YZ		 2802 BUG_ON(ret);
2803 info++;
2804 }
31153d81
YZ		 2805
2806 return 0;
2807}
2808
333db94c
CM		 2809int drop_snap_lookup_refcount(struct btrfs_root *root, u64 start, u64 len,
2810 u32 *refs)
2811{
017e5369 2812 int ret;
f87f057b 2813
31840ae1 2814 ret = btrfs_lookup_extent_ref(NULL, root, start, len, refs);
f87f057b
CM		 2815 BUG_ON(ret);
2816
2817#if 0 // some debugging code in case we see problems here
2818 /* if the refs count is one, it won't get increased again. But
2819 * if the ref count is > 1, someone may be decreasing it at
2820 * the same time we are.
2821 */
2822 if (*refs != 1) {
2823 struct extent_buffer *eb = NULL;
2824 eb = btrfs_find_create_tree_block(root, start, len);
2825 if (eb)
2826 btrfs_tree_lock(eb);
2827
2828 mutex_lock(&root->fs_info->alloc_mutex);
2829 ret = lookup_extent_ref(NULL, root, start, len, refs);
2830 BUG_ON(ret);
2831 mutex_unlock(&root->fs_info->alloc_mutex);
2832
2833 if (eb) {
2834 btrfs_tree_unlock(eb);
2835 free_extent_buffer(eb);
2836 }
2837 if (*refs == 1) {
2838 printk("block %llu went down to one during drop_snap\n",
2839 (unsigned long long)start);
2840 }
2841
2842 }
2843#endif
2844
e7a84565 2845 cond_resched();
017e5369 2846 return ret;
333db94c
CM		 2847}
2848
9aca1d51
CM		 2849/*
2850 * helper function for drop_snapshot, this walks down the tree dropping ref
2851 * counts as it goes.
2852 */
98ed5174
CM		 2853static int noinline walk_down_tree(struct btrfs_trans_handle *trans,
2854 struct btrfs_root *root,
2855 struct btrfs_path *path, int *level)
20524f02 2856{
7bb86316
CM		 2857 u64 root_owner;
2858 u64 root_gen;
2859 u64 bytenr;
ca7a79ad 2860 u64 ptr_gen;
5f39d397
CM		 2861 struct extent_buffer *next;
2862 struct extent_buffer *cur;
7bb86316 2863 struct extent_buffer *parent;
31153d81 2864 struct btrfs_leaf_ref *ref;
db94535d 2865 u32 blocksize;
20524f02
CM		 2866 int ret;
2867 u32 refs;
2868
5caf2a00
CM		 2869 WARN_ON(*level < 0);
2870 WARN_ON(*level >= BTRFS_MAX_LEVEL);
333db94c 2871 ret = drop_snap_lookup_refcount(root, path->nodes[*level]->start,
db94535d 2872 path->nodes[*level]->len, &refs);
20524f02
CM		 2873 BUG_ON(ret);
2874 if (refs > 1)
2875 goto out;
e011599b 2876
9aca1d51
CM		 2877 /*
2878 * walk down to the last node level and free all the leaves
2879 */
6407bf6d 2880 while(*level >= 0) {
5caf2a00
CM		 2881 WARN_ON(*level < 0);
2882 WARN_ON(*level >= BTRFS_MAX_LEVEL);
20524f02 2883 cur = path->nodes[*level];
e011599b 2884
5f39d397 2885 if (btrfs_header_level(cur) != *level)
2c90e5d6 2886 WARN_ON(1);
e011599b 2887
7518a238 2888 if (path->slots[*level] >=
5f39d397 2889 btrfs_header_nritems(cur))
20524f02 2890 break;
6407bf6d 2891 if (*level == 0) {
e02119d5 2892 ret = btrfs_drop_leaf_ref(trans, root, cur);
6407bf6d
CM		 2893 BUG_ON(ret);
2894 break;
2895 }
db94535d 2896 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
ca7a79ad 2897 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
db94535d 2898 blocksize = btrfs_level_size(root, *level - 1);
925baedd 2899
333db94c 2900 ret = drop_snap_lookup_refcount(root, bytenr, blocksize, &refs);
6407bf6d
CM		 2901 BUG_ON(ret);
2902 if (refs != 1) {
7bb86316
CM		 2903 parent = path->nodes[*level];
2904 root_owner = btrfs_header_owner(parent);
2905 root_gen = btrfs_header_generation(parent);
20524f02 2906 path->slots[*level]++;
f87f057b 2907
925baedd 2908 ret = __btrfs_free_extent(trans, root, bytenr,
31840ae1 2909 blocksize, parent->start,
3bb1a1bc
YZ		 2910 root_owner, root_gen,
2911 *level - 1, 1);
20524f02 2912 BUG_ON(ret);
18e35e0a
CM		 2913
2914 atomic_inc(&root->fs_info->throttle_gen);
2915 wake_up(&root->fs_info->transaction_throttle);
2dd3e67b 2916 cond_resched();
18e35e0a 2917
20524f02
CM		 2918 continue;
2919 }
f87f057b
CM		 2920 /*
2921 * at this point, we have a single ref, and since the
2922 * only place referencing this extent is a dead root
2923 * the reference count should never go higher.
2924 * So, we don't need to check it again
2925 */
31153d81 2926 if (*level == 1) {
017e5369 2927 ref = btrfs_lookup_leaf_ref(root, bytenr);
1a40e23b
ZY		 2928 if (ref && ref->generation != ptr_gen) {
2929 btrfs_free_leaf_ref(root, ref);
2930 ref = NULL;
2931 }
31153d81 2932 if (ref) {
e02119d5 2933 ret = cache_drop_leaf_ref(trans, root, ref);
31153d81
YZ		 2934 BUG_ON(ret);
2935 btrfs_remove_leaf_ref(root, ref);
bcc63abb 2936 btrfs_free_leaf_ref(root, ref);
31153d81
YZ		 2937 *level = 0;
2938 break;
2939 }
3bb1a1bc 2940 if (printk_ratelimit()) {
37d1aeee
CM		 2941 printk("leaf ref miss for bytenr %llu\n",
2942 (unsigned long long)bytenr);
3bb1a1bc 2943 }
31153d81 2944 }
db94535d 2945 next = btrfs_find_tree_block(root, bytenr, blocksize);
1259ab75 2946 if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
5f39d397 2947 free_extent_buffer(next);
333db94c 2948
ca7a79ad
CM		 2949 next = read_tree_block(root, bytenr, blocksize,
2950 ptr_gen);
e7a84565 2951 cond_resched();
f87f057b
CM		 2952#if 0
2953 /*
2954 * this is a debugging check and can go away
2955 * the ref should never go all the way down to 1
2956 * at this point
2957 */
e6dcd2dc
CM		 2958 ret = lookup_extent_ref(NULL, root, bytenr, blocksize,
2959 &refs);
e9d0b13b 2960 BUG_ON(ret);
f87f057b
CM		 2961 WARN_ON(refs != 1);
2962#endif
e9d0b13b 2963 }
5caf2a00 2964 WARN_ON(*level <= 0);
83e15a28 2965 if (path->nodes[*level-1])
5f39d397 2966 free_extent_buffer(path->nodes[*level-1]);
20524f02 2967 path->nodes[*level-1] = next;
5f39d397 2968 *level = btrfs_header_level(next);
20524f02 2969 path->slots[*level] = 0;
2dd3e67b 2970 cond_resched();
20524f02
CM		 2971 }
2972out:
5caf2a00
CM		 2973 WARN_ON(*level < 0);
2974 WARN_ON(*level >= BTRFS_MAX_LEVEL);
7bb86316
CM		 2975
2976 if (path->nodes[*level] == root->node) {
7bb86316 2977 parent = path->nodes[*level];
31153d81 2978 bytenr = path->nodes[*level]->start;
7bb86316
CM		 2979 } else {
2980 parent = path->nodes[*level + 1];
31153d81 2981 bytenr = btrfs_node_blockptr(parent, path->slots[*level + 1]);
7bb86316
CM		 2982 }
2983
31153d81
YZ		 2984 blocksize = btrfs_level_size(root, *level);
2985 root_owner = btrfs_header_owner(parent);
7bb86316 2986 root_gen = btrfs_header_generation(parent);
31153d81
YZ		 2987
2988 ret = __btrfs_free_extent(trans, root, bytenr, blocksize,
31840ae1 2989 parent->start, root_owner, root_gen,
3bb1a1bc 2990 *level, 1);
5f39d397 2991 free_extent_buffer(path->nodes[*level]);
20524f02
CM		 2992 path->nodes[*level] = NULL;
2993 *level += 1;
2994 BUG_ON(ret);
f87f057b 2995
e7a84565 2996 cond_resched();
20524f02
CM		 2997 return 0;
2998}
2999
f82d02d9
YZ		 3000/*
3001 * helper function for drop_subtree, this function is similar to
3002 * walk_down_tree. The main difference is that it checks reference
3003 * counts while tree blocks are locked.
3004 */
3005static int noinline walk_down_subtree(struct btrfs_trans_handle *trans,
3006 struct btrfs_root *root,
3007 struct btrfs_path *path, int *level)
3008{
3009 struct extent_buffer *next;
3010 struct extent_buffer *cur;
3011 struct extent_buffer *parent;
3012 u64 bytenr;
3013 u64 ptr_gen;
3014 u32 blocksize;
3015 u32 refs;
3016 int ret;
3017
3018 cur = path->nodes[*level];
3019 ret = btrfs_lookup_extent_ref(trans, root, cur->start, cur->len,
3020 &refs);
3021 BUG_ON(ret);
3022 if (refs > 1)
3023 goto out;
3024
3025 while (*level >= 0) {
3026 cur = path->nodes[*level];
3027 if (*level == 0) {
3028 ret = btrfs_drop_leaf_ref(trans, root, cur);
3029 BUG_ON(ret);
3030 clean_tree_block(trans, root, cur);
3031 break;
3032 }
3033 if (path->slots[*level] >= btrfs_header_nritems(cur)) {
3034 clean_tree_block(trans, root, cur);
3035 break;
3036 }
3037
3038 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
3039 blocksize = btrfs_level_size(root, *level - 1);
3040 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
3041
3042 next = read_tree_block(root, bytenr, blocksize, ptr_gen);
3043 btrfs_tree_lock(next);
3044
3045 ret = btrfs_lookup_extent_ref(trans, root, bytenr, blocksize,
3046 &refs);
3047 BUG_ON(ret);
3048 if (refs > 1) {
3049 parent = path->nodes[*level];
3050 ret = btrfs_free_extent(trans, root, bytenr,
3051 blocksize, parent->start,
3052 btrfs_header_owner(parent),
3053 btrfs_header_generation(parent),
3054 *level - 1, 1);
3055 BUG_ON(ret);
3056 path->slots[*level]++;
3057 btrfs_tree_unlock(next);
3058 free_extent_buffer(next);
3059 continue;
3060 }
3061
3062 *level = btrfs_header_level(next);
3063 path->nodes[*level] = next;
3064 path->slots[*level] = 0;
3065 path->locks[*level] = 1;
3066 cond_resched();
3067 }
3068out:
3069 parent = path->nodes[*level + 1];
3070 bytenr = path->nodes[*level]->start;
3071 blocksize = path->nodes[*level]->len;
3072
3073 ret = btrfs_free_extent(trans, root, bytenr, blocksize,
3074 parent->start, btrfs_header_owner(parent),
3075 btrfs_header_generation(parent), *level, 1);
3076 BUG_ON(ret);
3077
3078 if (path->locks[*level]) {
3079 btrfs_tree_unlock(path->nodes[*level]);
3080 path->locks[*level] = 0;
3081 }
3082 free_extent_buffer(path->nodes[*level]);
3083 path->nodes[*level] = NULL;
3084 *level += 1;
3085 cond_resched();
3086 return 0;
3087}
3088
9aca1d51
CM		 3089/*
3090 * helper for dropping snapshots. This walks back up the tree in the path
3091 * to find the first node higher up where we haven't yet gone through
3092 * all the slots
3093 */
98ed5174
CM		 3094static int noinline walk_up_tree(struct btrfs_trans_handle *trans,
3095 struct btrfs_root *root,
f82d02d9
YZ		 3096 struct btrfs_path *path,
3097 int *level, int max_level)
20524f02 3098{
7bb86316
CM		 3099 u64 root_owner;
3100 u64 root_gen;
3101 struct btrfs_root_item *root_item = &root->root_item;
20524f02
CM		 3102 int i;
3103 int slot;
3104 int ret;
9f3a7427 3105
f82d02d9 3106 for (i = *level; i < max_level && path->nodes[i]; i++) {
20524f02 3107 slot = path->slots[i];
5f39d397
CM		 3108 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
3109 struct extent_buffer *node;
3110 struct btrfs_disk_key disk_key;
3111 node = path->nodes[i];
20524f02
CM		 3112 path->slots[i]++;
3113 *level = i;
9f3a7427 3114 WARN_ON(*level == 0);
5f39d397 3115 btrfs_node_key(node, &disk_key, path->slots[i]);
9f3a7427 3116 memcpy(&root_item->drop_progress,
5f39d397 3117 &disk_key, sizeof(disk_key));
9f3a7427 3118 root_item->drop_level = i;
20524f02
CM		 3119 return 0;
3120 } else {
31840ae1
ZY		 3121 struct extent_buffer *parent;
3122 if (path->nodes[*level] == root->node)
3123 parent = path->nodes[*level];
3124 else
3125 parent = path->nodes[*level + 1];
3126
3127 root_owner = btrfs_header_owner(parent);
3128 root_gen = btrfs_header_generation(parent);
f82d02d9
YZ		 3129
3130 clean_tree_block(trans, root, path->nodes[*level]);
e089f05c 3131 ret = btrfs_free_extent(trans, root,
db94535d 3132 path->nodes[*level]->start,
7bb86316 3133 path->nodes[*level]->len,
3bb1a1bc
YZ		 3134 parent->start, root_owner,
3135 root_gen, *level, 1);
6407bf6d 3136 BUG_ON(ret);
f82d02d9
YZ		 3137 if (path->locks[*level]) {
3138 btrfs_tree_unlock(path->nodes[*level]);
3139 path->locks[*level] = 0;
3140 }
5f39d397 3141 free_extent_buffer(path->nodes[*level]);
83e15a28 3142 path->nodes[*level] = NULL;
20524f02 3143 *level = i + 1;
20524f02
CM		 3144 }
3145 }
3146 return 1;
3147}
3148
9aca1d51
CM		 3149/*
3150 * drop the reference count on the tree rooted at 'snap'. This traverses
3151 * the tree freeing any blocks that have a ref count of zero after being
3152 * decremented.
3153 */
e089f05c 3154int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
9f3a7427 3155 *root)
20524f02 3156{
3768f368 3157 int ret = 0;
9aca1d51 3158 int wret;
20524f02 3159 int level;
5caf2a00 3160 struct btrfs_path *path;
20524f02
CM		 3161 int i;
3162 int orig_level;
9f3a7427 3163 struct btrfs_root_item *root_item = &root->root_item;
20524f02 3164
a2135011 3165 WARN_ON(!mutex_is_locked(&root->fs_info->drop_mutex));
5caf2a00
CM		 3166 path = btrfs_alloc_path();
3167 BUG_ON(!path);
20524f02 3168
5f39d397 3169 level = btrfs_header_level(root->node);
20524f02 3170 orig_level = level;
9f3a7427
CM		 3171 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
3172 path->nodes[level] = root->node;
f510cfec 3173 extent_buffer_get(root->node);
9f3a7427
CM		 3174 path->slots[level] = 0;
3175 } else {
3176 struct btrfs_key key;
5f39d397
CM		 3177 struct btrfs_disk_key found_key;
3178 struct extent_buffer *node;
6702ed49 3179
9f3a7427 3180 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
6702ed49
CM		 3181 level = root_item->drop_level;
3182 path->lowest_level = level;
9f3a7427 3183 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6702ed49 3184 if (wret < 0) {
9f3a7427
CM		 3185 ret = wret;
3186 goto out;
3187 }
5f39d397
CM		 3188 node = path->nodes[level];
3189 btrfs_node_key(node, &found_key, path->slots[level]);
3190 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
3191 sizeof(found_key)));
7d9eb12c
CM		 3192 /*
3193 * unlock our path, this is safe because only this
3194 * function is allowed to delete this snapshot
3195 */
925baedd
CM		 3196 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
3197 if (path->nodes[i] && path->locks[i]) {
3198 path->locks[i] = 0;
3199 btrfs_tree_unlock(path->nodes[i]);
3200 }
3201 }
9f3a7427 3202 }
20524f02 3203 while(1) {
5caf2a00 3204 wret = walk_down_tree(trans, root, path, &level);
9aca1d51 3205 if (wret > 0)
20524f02 3206 break;
9aca1d51
CM		 3207 if (wret < 0)
3208 ret = wret;
3209
f82d02d9
YZ		 3210 wret = walk_up_tree(trans, root, path, &level,
3211 BTRFS_MAX_LEVEL);
9aca1d51 3212 if (wret > 0)
20524f02 3213 break;
9aca1d51
CM		 3214 if (wret < 0)
3215 ret = wret;
e7a84565
CM		 3216 if (trans->transaction->in_commit) {
3217 ret = -EAGAIN;
3218 break;
3219 }
18e35e0a 3220 atomic_inc(&root->fs_info->throttle_gen);
017e5369 3221 wake_up(&root->fs_info->transaction_throttle);
20524f02 3222 }
83e15a28 3223 for (i = 0; i <= orig_level; i++) {
5caf2a00 3224 if (path->nodes[i]) {
5f39d397 3225 free_extent_buffer(path->nodes[i]);
0f82731f 3226 path->nodes[i] = NULL;
83e15a28 3227 }
20524f02 3228 }
9f3a7427 3229out:
5caf2a00 3230 btrfs_free_path(path);
9aca1d51 3231 return ret;
20524f02 3232}
9078a3e1 3233
f82d02d9
YZ		 3234int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
3235 struct btrfs_root *root,
3236 struct extent_buffer *node,
3237 struct extent_buffer *parent)
3238{
3239 struct btrfs_path *path;
3240 int level;
3241 int parent_level;
3242 int ret = 0;
3243 int wret;
3244
3245 path = btrfs_alloc_path();
3246 BUG_ON(!path);
3247
3248 BUG_ON(!btrfs_tree_locked(parent));
3249 parent_level = btrfs_header_level(parent);
3250 extent_buffer_get(parent);
3251 path->nodes[parent_level] = parent;
3252 path->slots[parent_level] = btrfs_header_nritems(parent);
3253
3254 BUG_ON(!btrfs_tree_locked(node));
3255 level = btrfs_header_level(node);
3256 extent_buffer_get(node);
3257 path->nodes[level] = node;
3258 path->slots[level] = 0;
3259
3260 while (1) {
3261 wret = walk_down_subtree(trans, root, path, &level);
3262 if (wret < 0)
3263 ret = wret;
3264 if (wret != 0)
3265 break;
3266
3267 wret = walk_up_tree(trans, root, path, &level, parent_level);
3268 if (wret < 0)
3269 ret = wret;
3270 if (wret != 0)
3271 break;
3272 }
3273
3274 btrfs_free_path(path);
3275 return ret;
3276}
3277
8e7bf94f
CM		 3278static unsigned long calc_ra(unsigned long start, unsigned long last,
3279 unsigned long nr)
3280{
3281 return min(last, start + nr - 1);
3282}
3283
98ed5174
CM		 3284static int noinline relocate_inode_pages(struct inode *inode, u64 start,
3285 u64 len)
edbd8d4e
CM		 3286{
3287 u64 page_start;
3288 u64 page_end;
1a40e23b 3289 unsigned long first_index;
edbd8d4e 3290 unsigned long last_index;
edbd8d4e
CM		 3291 unsigned long i;
3292 struct page *page;
d1310b2e 3293 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4313b399 3294 struct file_ra_state *ra;
3eaa2885 3295 struct btrfs_ordered_extent *ordered;
1a40e23b
ZY		 3296 unsigned int total_read = 0;
3297 unsigned int total_dirty = 0;
3298 int ret = 0;
4313b399
CM		 3299
3300 ra = kzalloc(sizeof(*ra), GFP_NOFS);
edbd8d4e
CM		 3301
3302 mutex_lock(&inode->i_mutex);
1a40e23b 3303 first_index = start >> PAGE_CACHE_SHIFT;
edbd8d4e
CM		 3304 last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
3305
1a40e23b
ZY		 3306 /* make sure the dirty trick played by the caller work */
3307 ret = invalidate_inode_pages2_range(inode->i_mapping,
3308 first_index, last_index);
3309 if (ret)
3310 goto out_unlock;
8e7bf94f 3311
4313b399 3312 file_ra_state_init(ra, inode->i_mapping);
edbd8d4e 3313
1a40e23b
ZY		 3314 for (i = first_index ; i <= last_index; i++) {
3315 if (total_read % ra->ra_pages == 0) {
8e7bf94f 3316 btrfs_force_ra(inode->i_mapping, ra, NULL, i,
1a40e23b 3317 calc_ra(i, last_index, ra->ra_pages));
8e7bf94f
CM		 3318 }
3319 total_read++;
3eaa2885
CM		 3320again:
3321 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
1a40e23b 3322 BUG_ON(1);
edbd8d4e 3323 page = grab_cache_page(inode->i_mapping, i);
a061fc8d 3324 if (!page) {
1a40e23b 3325 ret = -ENOMEM;
edbd8d4e 3326 goto out_unlock;
a061fc8d 3327 }
edbd8d4e
CM		 3328 if (!PageUptodate(page)) {
3329 btrfs_readpage(NULL, page);
3330 lock_page(page);
3331 if (!PageUptodate(page)) {
3332 unlock_page(page);
3333 page_cache_release(page);
1a40e23b 3334 ret = -EIO;
edbd8d4e
CM		 3335 goto out_unlock;
3336 }
3337 }
ec44a35c 3338 wait_on_page_writeback(page);
3eaa2885 3339
edbd8d4e
CM		 3340 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
3341 page_end = page_start + PAGE_CACHE_SIZE - 1;
d1310b2e 3342 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
edbd8d4e 3343
3eaa2885
CM		 3344 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3345 if (ordered) {
3346 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3347 unlock_page(page);
3348 page_cache_release(page);
3349 btrfs_start_ordered_extent(inode, ordered, 1);
3350 btrfs_put_ordered_extent(ordered);
3351 goto again;
3352 }
3353 set_page_extent_mapped(page);
3354
ea8c2819 3355 btrfs_set_extent_delalloc(inode, page_start, page_end);
1a40e23b
ZY		 3356 if (i == first_index)
3357 set_extent_bits(io_tree, page_start, page_end,
3358 EXTENT_BOUNDARY, GFP_NOFS);
3359
a061fc8d 3360 set_page_dirty(page);
1a40e23b 3361 total_dirty++;
edbd8d4e 3362
d1310b2e 3363 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
edbd8d4e
CM		 3364 unlock_page(page);
3365 page_cache_release(page);
3366 }
3367
3368out_unlock:
ec44a35c 3369 kfree(ra);
edbd8d4e 3370 mutex_unlock(&inode->i_mutex);
1a40e23b
ZY		 3371 balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
3372 return ret;
edbd8d4e
CM		 3373}
3374
1a40e23b
ZY		 3375static int noinline relocate_data_extent(struct inode *reloc_inode,
3376 struct btrfs_key *extent_key,
3377 u64 offset)
3378{
3379 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
3380 struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
3381 struct extent_map *em;
6643558d
YZ		 3382 u64 start = extent_key->objectid - offset;
3383 u64 end = start + extent_key->offset - 1;
bf4ef679 3384
1a40e23b
ZY		 3385 em = alloc_extent_map(GFP_NOFS);
3386 BUG_ON(!em || IS_ERR(em));
bf4ef679 3387
6643558d 3388 em->start = start;
1a40e23b 3389 em->len = extent_key->offset;
c8b97818 3390 em->block_len = extent_key->offset;
1a40e23b
ZY		 3391 em->block_start = extent_key->objectid;
3392 em->bdev = root->fs_info->fs_devices->latest_bdev;
3393 set_bit(EXTENT_FLAG_PINNED, &em->flags);
3394
3395 /* setup extent map to cheat btrfs_readpage */
6643558d 3396 lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
1a40e23b
ZY		 3397 while (1) {
3398 int ret;
3399 spin_lock(&em_tree->lock);
3400 ret = add_extent_mapping(em_tree, em);
3401 spin_unlock(&em_tree->lock);
3402 if (ret != -EEXIST) {
3403 free_extent_map(em);
bf4ef679
CM		 3404 break;
3405 }
6643558d 3406 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
bf4ef679 3407 }
6643558d 3408 unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
bf4ef679 3409
6643558d 3410 return relocate_inode_pages(reloc_inode, start, extent_key->offset);
1a40e23b 3411}
edbd8d4e 3412
1a40e23b
ZY		 3413struct btrfs_ref_path {
3414 u64 extent_start;
3415 u64 nodes[BTRFS_MAX_LEVEL];
3416 u64 root_objectid;
3417 u64 root_generation;
3418 u64 owner_objectid;
1a40e23b
ZY		 3419 u32 num_refs;
3420 int lowest_level;
3421 int current_level;
f82d02d9
YZ		 3422 int shared_level;
3423
3424 struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
3425 u64 new_nodes[BTRFS_MAX_LEVEL];
1a40e23b 3426};
7d9eb12c 3427
1a40e23b 3428struct disk_extent {
c8b97818 3429 u64 ram_bytes;
1a40e23b
ZY		 3430 u64 disk_bytenr;
3431 u64 disk_num_bytes;
3432 u64 offset;
3433 u64 num_bytes;
c8b97818
CM		 3434 u8 compression;
3435 u8 encryption;
3436 u16 other_encoding;
1a40e23b 3437};
4313b399 3438
1a40e23b
ZY		 3439static int is_cowonly_root(u64 root_objectid)
3440{
3441 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
3442 root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
3443 root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
3444 root_objectid == BTRFS_DEV_TREE_OBJECTID ||
3445 root_objectid == BTRFS_TREE_LOG_OBJECTID)
3446 return 1;
3447 return 0;
3448}
edbd8d4e 3449
1a40e23b
ZY		 3450static int noinline __next_ref_path(struct btrfs_trans_handle *trans,
3451 struct btrfs_root *extent_root,
3452 struct btrfs_ref_path *ref_path,
3453 int first_time)
3454{
3455 struct extent_buffer *leaf;
3456 struct btrfs_path *path;
3457 struct btrfs_extent_ref *ref;
3458 struct btrfs_key key;
3459 struct btrfs_key found_key;
3460 u64 bytenr;
3461 u32 nritems;
3462 int level;
3463 int ret = 1;
edbd8d4e 3464
1a40e23b
ZY		 3465 path = btrfs_alloc_path();
3466 if (!path)
3467 return -ENOMEM;
bf4ef679 3468
1a40e23b
ZY		 3469 if (first_time) {
3470 ref_path->lowest_level = -1;
3471 ref_path->current_level = -1;
f82d02d9 3472 ref_path->shared_level = -1;
1a40e23b
ZY		 3473 goto walk_up;
3474 }
3475walk_down:
3476 level = ref_path->current_level - 1;
3477 while (level >= -1) {
3478 u64 parent;
3479 if (level < ref_path->lowest_level)
3480 break;
bf4ef679 3481
1a40e23b
ZY		 3482 if (level >= 0) {
3483 bytenr = ref_path->nodes[level];
3484 } else {
3485 bytenr = ref_path->extent_start;
3486 }
3487 BUG_ON(bytenr == 0);
bf4ef679 3488
1a40e23b
ZY		 3489 parent = ref_path->nodes[level + 1];
3490 ref_path->nodes[level + 1] = 0;
3491 ref_path->current_level = level;
3492 BUG_ON(parent == 0);
0ef3e66b 3493
1a40e23b
ZY		 3494 key.objectid = bytenr;
3495 key.offset = parent + 1;
3496 key.type = BTRFS_EXTENT_REF_KEY;
edbd8d4e 3497
1a40e23b
ZY		 3498 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
3499 if (ret < 0)
edbd8d4e 3500 goto out;
1a40e23b 3501 BUG_ON(ret == 0);
7d9eb12c 3502
1a40e23b
ZY		 3503 leaf = path->nodes[0];
3504 nritems = btrfs_header_nritems(leaf);
3505 if (path->slots[0] >= nritems) {
3506 ret = btrfs_next_leaf(extent_root, path);
3507 if (ret < 0)
3508 goto out;
3509 if (ret > 0)
3510 goto next;
3511 leaf = path->nodes[0];
3512 }
0ef3e66b 3513
1a40e23b
ZY		 3514 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3515 if (found_key.objectid == bytenr &&
f82d02d9
YZ		 3516 found_key.type == BTRFS_EXTENT_REF_KEY) {
3517 if (level < ref_path->shared_level)
3518 ref_path->shared_level = level;
1a40e23b 3519 goto found;
f82d02d9 3520 }
1a40e23b
ZY		 3521next:
3522 level--;
3523 btrfs_release_path(extent_root, path);
3524 if (need_resched()) {
1a40e23b 3525 cond_resched();
1a40e23b
ZY		 3526 }
3527 }
3528 /* reached lowest level */
3529 ret = 1;
3530 goto out;
3531walk_up:
3532 level = ref_path->current_level;
3533 while (level < BTRFS_MAX_LEVEL - 1) {
3534 u64 ref_objectid;
3535 if (level >= 0) {
3536 bytenr = ref_path->nodes[level];
3537 } else {
3538 bytenr = ref_path->extent_start;
3539 }
3540 BUG_ON(bytenr == 0);
edbd8d4e 3541
1a40e23b
ZY		 3542 key.objectid = bytenr;
3543 key.offset = 0;
3544 key.type = BTRFS_EXTENT_REF_KEY;
edbd8d4e 3545
1a40e23b
ZY		 3546 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
3547 if (ret < 0)
3548 goto out;
edbd8d4e 3549
1a40e23b
ZY		 3550 leaf = path->nodes[0];
3551 nritems = btrfs_header_nritems(leaf);
3552 if (path->slots[0] >= nritems) {
3553 ret = btrfs_next_leaf(extent_root, path);
3554 if (ret < 0)
3555 goto out;
3556 if (ret > 0) {
3557 /* the extent was freed by someone */
3558 if (ref_path->lowest_level == level)
3559 goto out;
3560 btrfs_release_path(extent_root, path);
3561 goto walk_down;
3562 }
3563 leaf = path->nodes[0];
3564 }
edbd8d4e 3565
1a40e23b
ZY		 3566 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3567 if (found_key.objectid != bytenr ||
3568 found_key.type != BTRFS_EXTENT_REF_KEY) {
3569 /* the extent was freed by someone */
3570 if (ref_path->lowest_level == level) {
3571 ret = 1;
3572 goto out;
3573 }
3574 btrfs_release_path(extent_root, path);
3575 goto walk_down;
3576 }
3577found:
3578 ref = btrfs_item_ptr(leaf, path->slots[0],
3579 struct btrfs_extent_ref);
3580 ref_objectid = btrfs_ref_objectid(leaf, ref);
3581 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
3582 if (first_time) {
3583 level = (int)ref_objectid;
3584 BUG_ON(level >= BTRFS_MAX_LEVEL);
3585 ref_path->lowest_level = level;
3586 ref_path->current_level = level;
3587 ref_path->nodes[level] = bytenr;
3588 } else {
3589 WARN_ON(ref_objectid != level);
3590 }
3591 } else {
3592 WARN_ON(level != -1);
3593 }
3594 first_time = 0;
bf4ef679 3595
1a40e23b
ZY		 3596 if (ref_path->lowest_level == level) {
3597 ref_path->owner_objectid = ref_objectid;
1a40e23b
ZY		 3598 ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
3599 }
bf4ef679 3600
7d9eb12c 3601 /*
1a40e23b
ZY		 3602 * the block is tree root or the block isn't in reference
3603 * counted tree.
7d9eb12c 3604 */
1a40e23b
ZY		 3605 if (found_key.objectid == found_key.offset ||
3606 is_cowonly_root(btrfs_ref_root(leaf, ref))) {
3607 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
3608 ref_path->root_generation =
3609 btrfs_ref_generation(leaf, ref);
3610 if (level < 0) {
3611 /* special reference from the tree log */
3612 ref_path->nodes[0] = found_key.offset;
3613 ref_path->current_level = 0;
3614 }
3615 ret = 0;
3616 goto out;
3617 }
7d9eb12c 3618
1a40e23b
ZY		 3619 level++;
3620 BUG_ON(ref_path->nodes[level] != 0);
3621 ref_path->nodes[level] = found_key.offset;
3622 ref_path->current_level = level;
bf4ef679 3623
1a40e23b
ZY		 3624 /*
3625 * the reference was created in the running transaction,
3626 * no need to continue walking up.
3627 */
3628 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
3629 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
3630 ref_path->root_generation =
3631 btrfs_ref_generation(leaf, ref);
3632 ret = 0;
3633 goto out;
7d9eb12c
CM		 3634 }
3635
1a40e23b
ZY		 3636 btrfs_release_path(extent_root, path);
3637 if (need_resched()) {
1a40e23b 3638 cond_resched();
1a40e23b 3639 }
7d9eb12c 3640 }
1a40e23b
ZY		 3641 /* reached max tree level, but no tree root found. */
3642 BUG();
edbd8d4e 3643out:
1a40e23b
ZY		 3644 btrfs_free_path(path);
3645 return ret;
edbd8d4e
CM		 3646}
3647
1a40e23b
ZY		 3648static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
3649 struct btrfs_root *extent_root,
3650 struct btrfs_ref_path *ref_path,
3651 u64 extent_start)
a061fc8d 3652{
1a40e23b
ZY		 3653 memset(ref_path, 0, sizeof(*ref_path));
3654 ref_path->extent_start = extent_start;
a061fc8d 3655
1a40e23b 3656 return __next_ref_path(trans, extent_root, ref_path, 1);
a061fc8d
CM		 3657}
3658
1a40e23b
ZY		 3659static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
3660 struct btrfs_root *extent_root,
3661 struct btrfs_ref_path *ref_path)
edbd8d4e 3662{
1a40e23b
ZY		 3663 return __next_ref_path(trans, extent_root, ref_path, 0);
3664}
3665
3666static int noinline get_new_locations(struct inode *reloc_inode,
3667 struct btrfs_key *extent_key,
3668 u64 offset, int no_fragment,
3669 struct disk_extent **extents,
3670 int *nr_extents)
3671{
3672 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
3673 struct btrfs_path *path;
3674 struct btrfs_file_extent_item *fi;
edbd8d4e 3675 struct extent_buffer *leaf;
1a40e23b
ZY		 3676 struct disk_extent *exts = *extents;
3677 struct btrfs_key found_key;
3678 u64 cur_pos;
3679 u64 last_byte;
edbd8d4e 3680 u32 nritems;
1a40e23b
ZY		 3681 int nr = 0;
3682 int max = *nr_extents;
3683 int ret;
edbd8d4e 3684
1a40e23b
ZY		 3685 WARN_ON(!no_fragment && *extents);
3686 if (!exts) {
3687 max = 1;
3688 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
3689 if (!exts)
3690 return -ENOMEM;
a061fc8d 3691 }
edbd8d4e 3692
1a40e23b
ZY		 3693 path = btrfs_alloc_path();
3694 BUG_ON(!path);
edbd8d4e 3695
1a40e23b
ZY		 3696 cur_pos = extent_key->objectid - offset;
3697 last_byte = extent_key->objectid + extent_key->offset;
3698 ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
3699 cur_pos, 0);
3700 if (ret < 0)
3701 goto out;
3702 if (ret > 0) {
3703 ret = -ENOENT;
3704 goto out;
3705 }
edbd8d4e 3706
1a40e23b 3707 while (1) {
edbd8d4e
CM		 3708 leaf = path->nodes[0];
3709 nritems = btrfs_header_nritems(leaf);
1a40e23b
ZY		 3710 if (path->slots[0] >= nritems) {
3711 ret = btrfs_next_leaf(root, path);
a061fc8d
CM		 3712 if (ret < 0)
3713 goto out;
1a40e23b
ZY		 3714 if (ret > 0)
3715 break;
bf4ef679 3716 leaf = path->nodes[0];
a061fc8d 3717 }
edbd8d4e
CM		 3718
3719 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1a40e23b
ZY		 3720 if (found_key.offset != cur_pos ||
3721 found_key.type != BTRFS_EXTENT_DATA_KEY ||
3722 found_key.objectid != reloc_inode->i_ino)
edbd8d4e
CM		 3723 break;
3724
1a40e23b
ZY		 3725 fi = btrfs_item_ptr(leaf, path->slots[0],
3726 struct btrfs_file_extent_item);
3727 if (btrfs_file_extent_type(leaf, fi) !=
3728 BTRFS_FILE_EXTENT_REG ||
3729 btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
edbd8d4e 3730 break;
1a40e23b
ZY		 3731
3732 if (nr == max) {
3733 struct disk_extent *old = exts;
3734 max *= 2;
3735 exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
3736 memcpy(exts, old, sizeof(*exts) * nr);
3737 if (old != *extents)
3738 kfree(old);
a061fc8d 3739 }
edbd8d4e 3740
1a40e23b
ZY		 3741 exts[nr].disk_bytenr =
3742 btrfs_file_extent_disk_bytenr(leaf, fi);
3743 exts[nr].disk_num_bytes =
3744 btrfs_file_extent_disk_num_bytes(leaf, fi);
3745 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
3746 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
c8b97818
CM		 3747 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
3748 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
3749 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
3750 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
3751 fi);
1a40e23b
ZY		 3752 WARN_ON(exts[nr].offset > 0);
3753 WARN_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
edbd8d4e 3754
1a40e23b
ZY		 3755 cur_pos += exts[nr].num_bytes;
3756 nr++;
3757
3758 if (cur_pos + offset >= last_byte)
3759 break;
3760
3761 if (no_fragment) {
3762 ret = 1;
edbd8d4e 3763 goto out;
1a40e23b
ZY		 3764 }
3765 path->slots[0]++;
3766 }
3767
3768 WARN_ON(cur_pos + offset > last_byte);
3769 if (cur_pos + offset < last_byte) {
3770 ret = -ENOENT;
3771 goto out;
edbd8d4e
CM		 3772 }
3773 ret = 0;
3774out:
1a40e23b
ZY		 3775 btrfs_free_path(path);
3776 if (ret) {
3777 if (exts != *extents)
3778 kfree(exts);
3779 } else {
3780 *extents = exts;
3781 *nr_extents = nr;
3782 }
3783 return ret;
3784}
3785
3786static int noinline replace_one_extent(struct btrfs_trans_handle *trans,
3787 struct btrfs_root *root,
3788 struct btrfs_path *path,
3789 struct btrfs_key *extent_key,
3790 struct btrfs_key *leaf_key,
3791 struct btrfs_ref_path *ref_path,
3792 struct disk_extent *new_extents,
3793 int nr_extents)
3794{
3795 struct extent_buffer *leaf;
3796 struct btrfs_file_extent_item *fi;
3797 struct inode *inode = NULL;
3798 struct btrfs_key key;
3799 u64 lock_start = 0;
3800 u64 lock_end = 0;
3801 u64 num_bytes;
3802 u64 ext_offset;
3803 u64 first_pos;
3804 u32 nritems;
3bb1a1bc 3805 int nr_scaned = 0;
1a40e23b
ZY		 3806 int extent_locked = 0;
3807 int ret;
3808
3bb1a1bc
YZ		 3809 memcpy(&key, leaf_key, sizeof(key));
3810 first_pos = INT_LIMIT(loff_t) - extent_key->offset;
1a40e23b 3811 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
3bb1a1bc
YZ		 3812 if (key.objectid < ref_path->owner_objectid ||
3813 (key.objectid == ref_path->owner_objectid &&
3814 key.type < BTRFS_EXTENT_DATA_KEY)) {
3815 key.objectid = ref_path->owner_objectid;
3816 key.type = BTRFS_EXTENT_DATA_KEY;
3817 key.offset = 0;
3818 }
1a40e23b
ZY		 3819 }
3820
3821 while (1) {
3822 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
3823 if (ret < 0)
3824 goto out;
3825
3826 leaf = path->nodes[0];
3827 nritems = btrfs_header_nritems(leaf);
3828next:
3829 if (extent_locked && ret > 0) {
3830 /*
3831 * the file extent item was modified by someone
3832 * before the extent got locked.
3833 */
1a40e23b
ZY		 3834 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
3835 lock_end, GFP_NOFS);
3836 extent_locked = 0;
3837 }
3838
3839 if (path->slots[0] >= nritems) {
3bb1a1bc 3840 if (++nr_scaned > 2)
1a40e23b
ZY		 3841 break;
3842
3843 BUG_ON(extent_locked);
3844 ret = btrfs_next_leaf(root, path);
3845 if (ret < 0)
3846 goto out;
3847 if (ret > 0)
3848 break;
3849 leaf = path->nodes[0];
3850 nritems = btrfs_header_nritems(leaf);
3851 }
3852
3853 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3854
3855 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
3856 if ((key.objectid > ref_path->owner_objectid) ||
3857 (key.objectid == ref_path->owner_objectid &&
3858 key.type > BTRFS_EXTENT_DATA_KEY) ||
3859 (key.offset >= first_pos + extent_key->offset))
3860 break;
3861 }
3862
3863 if (inode && key.objectid != inode->i_ino) {
3864 BUG_ON(extent_locked);
3865 btrfs_release_path(root, path);
3866 mutex_unlock(&inode->i_mutex);
3867 iput(inode);
3868 inode = NULL;
3869 continue;
3870 }
3871
3872 if (key.type != BTRFS_EXTENT_DATA_KEY) {
3873 path->slots[0]++;
3874 ret = 1;
3875 goto next;
3876 }
3877 fi = btrfs_item_ptr(leaf, path->slots[0],
3878 struct btrfs_file_extent_item);
3879 if ((btrfs_file_extent_type(leaf, fi) !=
3880 BTRFS_FILE_EXTENT_REG) ||
3881 (btrfs_file_extent_disk_bytenr(leaf, fi) !=
3882 extent_key->objectid)) {
3883 path->slots[0]++;
3884 ret = 1;
3885 goto next;
3886 }
3887
3888 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
3889 ext_offset = btrfs_file_extent_offset(leaf, fi);
3890
3891 if (first_pos > key.offset - ext_offset)
3892 first_pos = key.offset - ext_offset;
3893
3894 if (!extent_locked) {
3895 lock_start = key.offset;
3896 lock_end = lock_start + num_bytes - 1;
3897 } else {
6643558d
YZ		 3898 if (lock_start > key.offset ||
3899 lock_end + 1 < key.offset + num_bytes) {
3900 unlock_extent(&BTRFS_I(inode)->io_tree,
3901 lock_start, lock_end, GFP_NOFS);
3902 extent_locked = 0;
3903 }
1a40e23b
ZY		 3904 }
3905
3906 if (!inode) {
3907 btrfs_release_path(root, path);
3908
3909 inode = btrfs_iget_locked(root->fs_info->sb,
3910 key.objectid, root);
3911 if (inode->i_state & I_NEW) {
3912 BTRFS_I(inode)->root = root;
3913 BTRFS_I(inode)->location.objectid =
3914 key.objectid;
3915 BTRFS_I(inode)->location.type =
3916 BTRFS_INODE_ITEM_KEY;
3917 BTRFS_I(inode)->location.offset = 0;
3918 btrfs_read_locked_inode(inode);
3919 unlock_new_inode(inode);
3920 }
3921 /*
3922 * some code call btrfs_commit_transaction while
3923 * holding the i_mutex, so we can't use mutex_lock
3924 * here.
3925 */
3926 if (is_bad_inode(inode) ||
3927 !mutex_trylock(&inode->i_mutex)) {
3928 iput(inode);
3929 inode = NULL;
3930 key.offset = (u64)-1;
3931 goto skip;
3932 }
3933 }
3934
3935 if (!extent_locked) {
3936 struct btrfs_ordered_extent *ordered;
3937
3938 btrfs_release_path(root, path);
3939
3940 lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
3941 lock_end, GFP_NOFS);
3942 ordered = btrfs_lookup_first_ordered_extent(inode,
3943 lock_end);
3944 if (ordered &&
3945 ordered->file_offset <= lock_end &&
3946 ordered->file_offset + ordered->len > lock_start) {
3947 unlock_extent(&BTRFS_I(inode)->io_tree,
3948 lock_start, lock_end, GFP_NOFS);
3949 btrfs_start_ordered_extent(inode, ordered, 1);
3950 btrfs_put_ordered_extent(ordered);
3951 key.offset += num_bytes;
3952 goto skip;
3953 }
3954 if (ordered)
3955 btrfs_put_ordered_extent(ordered);
3956
1a40e23b
ZY		 3957 extent_locked = 1;
3958 continue;
3959 }
3960
3961 if (nr_extents == 1) {
3962 /* update extent pointer in place */
3963 btrfs_set_file_extent_generation(leaf, fi,
3964 trans->transid);
3965 btrfs_set_file_extent_disk_bytenr(leaf, fi,
3966 new_extents[0].disk_bytenr);
3967 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
3968 new_extents[0].disk_num_bytes);
c8b97818
CM		 3969 btrfs_set_file_extent_ram_bytes(leaf, fi,
3970 new_extents[0].ram_bytes);
1a40e23b
ZY		 3971 ext_offset += new_extents[0].offset;
3972 btrfs_set_file_extent_offset(leaf, fi, ext_offset);
3973 btrfs_mark_buffer_dirty(leaf);
3974
3975 btrfs_drop_extent_cache(inode, key.offset,
3976 key.offset + num_bytes - 1, 0);
3977
3978 ret = btrfs_inc_extent_ref(trans, root,
3979 new_extents[0].disk_bytenr,
3980 new_extents[0].disk_num_bytes,
3981 leaf->start,
3982 root->root_key.objectid,
3983 trans->transid,
3bb1a1bc 3984 key.objectid);
1a40e23b
ZY		 3985 BUG_ON(ret);
3986
3987 ret = btrfs_free_extent(trans, root,
3988 extent_key->objectid,
3989 extent_key->offset,
3990 leaf->start,
3991 btrfs_header_owner(leaf),
3992 btrfs_header_generation(leaf),
3bb1a1bc 3993 key.objectid, 0);
1a40e23b
ZY		 3994 BUG_ON(ret);
3995
3996 btrfs_release_path(root, path);
3997 key.offset += num_bytes;
3998 } else {
3999 u64 alloc_hint;
4000 u64 extent_len;
4001 int i;
4002 /*
4003 * drop old extent pointer at first, then insert the
4004 * new pointers one bye one
4005 */
4006 btrfs_release_path(root, path);
4007 ret = btrfs_drop_extents(trans, root, inode, key.offset,
4008 key.offset + num_bytes,
4009 key.offset, &alloc_hint);
4010 BUG_ON(ret);
4011
4012 for (i = 0; i < nr_extents; i++) {
4013 if (ext_offset >= new_extents[i].num_bytes) {
4014 ext_offset -= new_extents[i].num_bytes;
4015 continue;
4016 }
4017 extent_len = min(new_extents[i].num_bytes -
4018 ext_offset, num_bytes);
4019
4020 ret = btrfs_insert_empty_item(trans, root,
4021 path, &key,
4022 sizeof(*fi));
4023 BUG_ON(ret);
4024
4025 leaf = path->nodes[0];
4026 fi = btrfs_item_ptr(leaf, path->slots[0],
4027 struct btrfs_file_extent_item);
4028 btrfs_set_file_extent_generation(leaf, fi,
4029 trans->transid);
4030 btrfs_set_file_extent_type(leaf, fi,
4031 BTRFS_FILE_EXTENT_REG);
4032 btrfs_set_file_extent_disk_bytenr(leaf, fi,
4033 new_extents[i].disk_bytenr);
4034 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4035 new_extents[i].disk_num_bytes);
c8b97818
CM		 4036 btrfs_set_file_extent_ram_bytes(leaf, fi,
4037 new_extents[i].ram_bytes);
4038
4039 btrfs_set_file_extent_compression(leaf, fi,
4040 new_extents[i].compression);
4041 btrfs_set_file_extent_encryption(leaf, fi,
4042 new_extents[i].encryption);
4043 btrfs_set_file_extent_other_encoding(leaf, fi,
4044 new_extents[i].other_encoding);
4045
1a40e23b
ZY		 4046 btrfs_set_file_extent_num_bytes(leaf, fi,
4047 extent_len);
4048 ext_offset += new_extents[i].offset;
4049 btrfs_set_file_extent_offset(leaf, fi,
4050 ext_offset);
4051 btrfs_mark_buffer_dirty(leaf);
4052
4053 btrfs_drop_extent_cache(inode, key.offset,
4054 key.offset + extent_len - 1, 0);
4055
4056 ret = btrfs_inc_extent_ref(trans, root,
4057 new_extents[i].disk_bytenr,
4058 new_extents[i].disk_num_bytes,
4059 leaf->start,
4060 root->root_key.objectid,
3bb1a1bc 4061 trans->transid, key.objectid);
1a40e23b
ZY		 4062 BUG_ON(ret);
4063 btrfs_release_path(root, path);
4064
a76a3cd4 4065 inode_add_bytes(inode, extent_len);
1a40e23b
ZY		 4066
4067 ext_offset = 0;
4068 num_bytes -= extent_len;
4069 key.offset += extent_len;
4070
4071 if (num_bytes == 0)
4072 break;
4073 }
4074 BUG_ON(i >= nr_extents);
4075 }
4076
4077 if (extent_locked) {
1a40e23b
ZY		 4078 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4079 lock_end, GFP_NOFS);
4080 extent_locked = 0;
4081 }
4082skip:
4083 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
4084 key.offset >= first_pos + extent_key->offset)
4085 break;
4086
4087 cond_resched();
4088 }
4089 ret = 0;
4090out:
4091 btrfs_release_path(root, path);
4092 if (inode) {
4093 mutex_unlock(&inode->i_mutex);
4094 if (extent_locked) {
1a40e23b
ZY		 4095 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4096 lock_end, GFP_NOFS);
4097 }
4098 iput(inode);
4099 }
4100 return ret;
4101}
4102
1a40e23b
ZY		 4103int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
4104 struct btrfs_root *root,
4105 struct extent_buffer *buf, u64 orig_start)
4106{
4107 int level;
4108 int ret;
4109
4110 BUG_ON(btrfs_header_generation(buf) != trans->transid);
4111 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
4112
4113 level = btrfs_header_level(buf);
4114 if (level == 0) {
4115 struct btrfs_leaf_ref *ref;
4116 struct btrfs_leaf_ref *orig_ref;
4117
4118 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
4119 if (!orig_ref)
4120 return -ENOENT;
4121
4122 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
4123 if (!ref) {
4124 btrfs_free_leaf_ref(root, orig_ref);
4125 return -ENOMEM;
4126 }
4127
4128 ref->nritems = orig_ref->nritems;
4129 memcpy(ref->extents, orig_ref->extents,
4130 sizeof(ref->extents[0]) * ref->nritems);
4131
4132 btrfs_free_leaf_ref(root, orig_ref);
4133
4134 ref->root_gen = trans->transid;
4135 ref->bytenr = buf->start;
4136 ref->owner = btrfs_header_owner(buf);
4137 ref->generation = btrfs_header_generation(buf);
4138 ret = btrfs_add_leaf_ref(root, ref, 0);
4139 WARN_ON(ret);
4140 btrfs_free_leaf_ref(root, ref);
4141 }
4142 return 0;
4143}
4144
4145static int noinline invalidate_extent_cache(struct btrfs_root *root,
4146 struct extent_buffer *leaf,
4147 struct btrfs_block_group_cache *group,
4148 struct btrfs_root *target_root)
4149{
4150 struct btrfs_key key;
4151 struct inode *inode = NULL;
4152 struct btrfs_file_extent_item *fi;
4153 u64 num_bytes;
4154 u64 skip_objectid = 0;
4155 u32 nritems;
4156 u32 i;
4157
4158 nritems = btrfs_header_nritems(leaf);
4159 for (i = 0; i < nritems; i++) {
4160 btrfs_item_key_to_cpu(leaf, &key, i);
4161 if (key.objectid == skip_objectid ||
4162 key.type != BTRFS_EXTENT_DATA_KEY)
4163 continue;
4164 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4165 if (btrfs_file_extent_type(leaf, fi) ==
4166 BTRFS_FILE_EXTENT_INLINE)
4167 continue;
4168 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
4169 continue;
4170 if (!inode || inode->i_ino != key.objectid) {
4171 iput(inode);
4172 inode = btrfs_ilookup(target_root->fs_info->sb,
4173 key.objectid, target_root, 1);
4174 }
4175 if (!inode) {
4176 skip_objectid = key.objectid;
4177 continue;
4178 }
4179 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4180
4181 lock_extent(&BTRFS_I(inode)->io_tree, key.offset,
4182 key.offset + num_bytes - 1, GFP_NOFS);
1a40e23b
ZY		 4183 btrfs_drop_extent_cache(inode, key.offset,
4184 key.offset + num_bytes - 1, 1);
1a40e23b
ZY		 4185 unlock_extent(&BTRFS_I(inode)->io_tree, key.offset,
4186 key.offset + num_bytes - 1, GFP_NOFS);
4187 cond_resched();
4188 }
4189 iput(inode);
4190 return 0;
4191}
4192
4193static int noinline replace_extents_in_leaf(struct btrfs_trans_handle *trans,
4194 struct btrfs_root *root,
4195 struct extent_buffer *leaf,
4196 struct btrfs_block_group_cache *group,
4197 struct inode *reloc_inode)
4198{
4199 struct btrfs_key key;
4200 struct btrfs_key extent_key;
4201 struct btrfs_file_extent_item *fi;
4202 struct btrfs_leaf_ref *ref;
4203 struct disk_extent *new_extent;
4204 u64 bytenr;
4205 u64 num_bytes;
4206 u32 nritems;
4207 u32 i;
4208 int ext_index;
4209 int nr_extent;
4210 int ret;
4211
4212 new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
4213 BUG_ON(!new_extent);
4214
4215 ref = btrfs_lookup_leaf_ref(root, leaf->start);
4216 BUG_ON(!ref);
4217
4218 ext_index = -1;
4219 nritems = btrfs_header_nritems(leaf);
4220 for (i = 0; i < nritems; i++) {
4221 btrfs_item_key_to_cpu(leaf, &key, i);
4222 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
4223 continue;
4224 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4225 if (btrfs_file_extent_type(leaf, fi) ==
4226 BTRFS_FILE_EXTENT_INLINE)
4227 continue;
4228 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
4229 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
4230 if (bytenr == 0)
4231 continue;
4232
4233 ext_index++;
4234 if (bytenr >= group->key.objectid + group->key.offset ||
4235 bytenr + num_bytes <= group->key.objectid)
4236 continue;
4237
4238 extent_key.objectid = bytenr;
4239 extent_key.offset = num_bytes;
4240 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
4241 nr_extent = 1;
4242 ret = get_new_locations(reloc_inode, &extent_key,
4243 group->key.objectid, 1,
4244 &new_extent, &nr_extent);
4245 if (ret > 0)
4246 continue;
4247 BUG_ON(ret < 0);
4248
4249 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
4250 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
4251 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
4252 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
4253
4254 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
c8b97818
CM		 4255 btrfs_set_file_extent_ram_bytes(leaf, fi,
4256 new_extent->ram_bytes);
1a40e23b
ZY		 4257 btrfs_set_file_extent_disk_bytenr(leaf, fi,
4258 new_extent->disk_bytenr);
4259 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4260 new_extent->disk_num_bytes);
4261 new_extent->offset += btrfs_file_extent_offset(leaf, fi);
4262 btrfs_set_file_extent_offset(leaf, fi, new_extent->offset);
4263 btrfs_mark_buffer_dirty(leaf);
4264
4265 ret = btrfs_inc_extent_ref(trans, root,
4266 new_extent->disk_bytenr,
4267 new_extent->disk_num_bytes,
4268 leaf->start,
4269 root->root_key.objectid,
3bb1a1bc 4270 trans->transid, key.objectid);
1a40e23b
ZY		 4271 BUG_ON(ret);
4272 ret = btrfs_free_extent(trans, root,
4273 bytenr, num_bytes, leaf->start,
4274 btrfs_header_owner(leaf),
4275 btrfs_header_generation(leaf),
3bb1a1bc 4276 key.objectid, 0);
1a40e23b
ZY		 4277 BUG_ON(ret);
4278 cond_resched();
4279 }
4280 kfree(new_extent);
4281 BUG_ON(ext_index + 1 != ref->nritems);
4282 btrfs_free_leaf_ref(root, ref);
4283 return 0;
4284}
4285
f82d02d9
YZ		 4286int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
4287 struct btrfs_root *root)
1a40e23b
ZY		 4288{
4289 struct btrfs_root *reloc_root;
f82d02d9 4290 int ret;
1a40e23b
ZY		 4291
4292 if (root->reloc_root) {
4293 reloc_root = root->reloc_root;
4294 root->reloc_root = NULL;
4295 list_add(&reloc_root->dead_list,
4296 &root->fs_info->dead_reloc_roots);
f82d02d9
YZ		 4297
4298 btrfs_set_root_bytenr(&reloc_root->root_item,
4299 reloc_root->node->start);
4300 btrfs_set_root_level(&root->root_item,
4301 btrfs_header_level(reloc_root->node));
4302 memset(&reloc_root->root_item.drop_progress, 0,
4303 sizeof(struct btrfs_disk_key));
4304 reloc_root->root_item.drop_level = 0;
4305
4306 ret = btrfs_update_root(trans, root->fs_info->tree_root,
4307 &reloc_root->root_key,
4308 &reloc_root->root_item);
4309 BUG_ON(ret);
1a40e23b
ZY		 4310 }
4311 return 0;
4312}
4313
4314int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
4315{
4316 struct btrfs_trans_handle *trans;
4317 struct btrfs_root *reloc_root;
4318 struct btrfs_root *prev_root = NULL;
4319 struct list_head dead_roots;
4320 int ret;
4321 unsigned long nr;
4322
4323 INIT_LIST_HEAD(&dead_roots);
4324 list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
4325
4326 while (!list_empty(&dead_roots)) {
4327 reloc_root = list_entry(dead_roots.prev,
4328 struct btrfs_root, dead_list);
4329 list_del_init(&reloc_root->dead_list);
4330
4331 BUG_ON(reloc_root->commit_root != NULL);
4332 while (1) {
4333 trans = btrfs_join_transaction(root, 1);
4334 BUG_ON(!trans);
4335
4336 mutex_lock(&root->fs_info->drop_mutex);
4337 ret = btrfs_drop_snapshot(trans, reloc_root);
4338 if (ret != -EAGAIN)
4339 break;
4340 mutex_unlock(&root->fs_info->drop_mutex);
4341
4342 nr = trans->blocks_used;
4343 ret = btrfs_end_transaction(trans, root);
4344 BUG_ON(ret);
4345 btrfs_btree_balance_dirty(root, nr);
4346 }
4347
4348 free_extent_buffer(reloc_root->node);
4349
4350 ret = btrfs_del_root(trans, root->fs_info->tree_root,
4351 &reloc_root->root_key);
4352 BUG_ON(ret);
4353 mutex_unlock(&root->fs_info->drop_mutex);
4354
4355 nr = trans->blocks_used;
4356 ret = btrfs_end_transaction(trans, root);
4357 BUG_ON(ret);
4358 btrfs_btree_balance_dirty(root, nr);
4359
4360 kfree(prev_root);
4361 prev_root = reloc_root;
4362 }
4363 if (prev_root) {
4364 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
4365 kfree(prev_root);
4366 }
4367 return 0;
4368}
4369
4370int btrfs_add_dead_reloc_root(struct btrfs_root *root)
4371{
4372 list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
4373 return 0;
4374}
4375
4376int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
4377{
4378 struct btrfs_root *reloc_root;
4379 struct btrfs_trans_handle *trans;
4380 struct btrfs_key location;
4381 int found;
4382 int ret;
4383
4384 mutex_lock(&root->fs_info->tree_reloc_mutex);
4385 ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
4386 BUG_ON(ret);
4387 found = !list_empty(&root->fs_info->dead_reloc_roots);
4388 mutex_unlock(&root->fs_info->tree_reloc_mutex);
4389
4390 if (found) {
4391 trans = btrfs_start_transaction(root, 1);
4392 BUG_ON(!trans);
4393 ret = btrfs_commit_transaction(trans, root);
4394 BUG_ON(ret);
4395 }
4396
4397 location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
4398 location.offset = (u64)-1;
4399 location.type = BTRFS_ROOT_ITEM_KEY;
4400
4401 reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
4402 BUG_ON(!reloc_root);
4403 btrfs_orphan_cleanup(reloc_root);
4404 return 0;
4405}
4406
4407static int noinline init_reloc_tree(struct btrfs_trans_handle *trans,
4408 struct btrfs_root *root)
4409{
4410 struct btrfs_root *reloc_root;
4411 struct extent_buffer *eb;
4412 struct btrfs_root_item *root_item;
4413 struct btrfs_key root_key;
4414 int ret;
4415
4416 BUG_ON(!root->ref_cows);
4417 if (root->reloc_root)
4418 return 0;
4419
4420 root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
4421 BUG_ON(!root_item);
4422
4423 ret = btrfs_copy_root(trans, root, root->commit_root,
4424 &eb, BTRFS_TREE_RELOC_OBJECTID);
4425 BUG_ON(ret);
4426
4427 root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
4428 root_key.offset = root->root_key.objectid;
4429 root_key.type = BTRFS_ROOT_ITEM_KEY;
4430
4431 memcpy(root_item, &root->root_item, sizeof(root_item));
4432 btrfs_set_root_refs(root_item, 0);
4433 btrfs_set_root_bytenr(root_item, eb->start);
4434 btrfs_set_root_level(root_item, btrfs_header_level(eb));
84234f3a 4435 btrfs_set_root_generation(root_item, trans->transid);
1a40e23b
ZY		 4436
4437 btrfs_tree_unlock(eb);
4438 free_extent_buffer(eb);
4439
4440 ret = btrfs_insert_root(trans, root->fs_info->tree_root,
4441 &root_key, root_item);
4442 BUG_ON(ret);
4443 kfree(root_item);
4444
4445 reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
4446 &root_key);
4447 BUG_ON(!reloc_root);
4448 reloc_root->last_trans = trans->transid;
4449 reloc_root->commit_root = NULL;
4450 reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
4451
4452 root->reloc_root = reloc_root;
4453 return 0;
4454}
4455
4456/*
4457 * Core function of space balance.
4458 *
4459 * The idea is using reloc trees to relocate tree blocks in reference
f82d02d9
YZ		 4460 * counted roots. There is one reloc tree for each subvol, and all
4461 * reloc trees share same root key objectid. Reloc trees are snapshots
4462 * of the latest committed roots of subvols (root->commit_root).
4463 *
4464 * To relocate a tree block referenced by a subvol, there are two steps.
4465 * COW the block through subvol's reloc tree, then update block pointer
4466 * in the subvol to point to the new block. Since all reloc trees share
4467 * same root key objectid, doing special handing for tree blocks owned
4468 * by them is easy. Once a tree block has been COWed in one reloc tree,
4469 * we can use the resulting new block directly when the same block is
4470 * required to COW again through other reloc trees. By this way, relocated
4471 * tree blocks are shared between reloc trees, so they are also shared
4472 * between subvols.
1a40e23b
ZY		 4473 */
4474static int noinline relocate_one_path(struct btrfs_trans_handle *trans,
4475 struct btrfs_root *root,
4476 struct btrfs_path *path,
4477 struct btrfs_key *first_key,
4478 struct btrfs_ref_path *ref_path,
4479 struct btrfs_block_group_cache *group,
4480 struct inode *reloc_inode)
4481{
4482 struct btrfs_root *reloc_root;
4483 struct extent_buffer *eb = NULL;
4484 struct btrfs_key *keys;
4485 u64 *nodes;
4486 int level;
f82d02d9 4487 int shared_level;
1a40e23b 4488 int lowest_level = 0;
1a40e23b
ZY		 4489 int ret;
4490
4491 if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
4492 lowest_level = ref_path->owner_objectid;
4493
f82d02d9 4494 if (!root->ref_cows) {
1a40e23b
ZY		 4495 path->lowest_level = lowest_level;
4496 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
4497 BUG_ON(ret < 0);
4498 path->lowest_level = 0;
4499 btrfs_release_path(root, path);
4500 return 0;
4501 }
4502
1a40e23b
ZY		 4503 mutex_lock(&root->fs_info->tree_reloc_mutex);
4504 ret = init_reloc_tree(trans, root);
4505 BUG_ON(ret);
4506 reloc_root = root->reloc_root;
4507
f82d02d9
YZ		 4508 shared_level = ref_path->shared_level;
4509 ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
1a40e23b 4510
f82d02d9
YZ		 4511 keys = ref_path->node_keys;
4512 nodes = ref_path->new_nodes;
4513 memset(&keys[shared_level + 1], 0,
4514 sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
4515 memset(&nodes[shared_level + 1], 0,
4516 sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
1a40e23b 4517
f82d02d9
YZ		 4518 if (nodes[lowest_level] == 0) {
4519 path->lowest_level = lowest_level;
4520 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
4521 0, 1);
4522 BUG_ON(ret);
4523 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
4524 eb = path->nodes[level];
4525 if (!eb || eb == reloc_root->node)
4526 break;
4527 nodes[level] = eb->start;
4528 if (level == 0)
4529 btrfs_item_key_to_cpu(eb, &keys[level], 0);
4530 else
4531 btrfs_node_key_to_cpu(eb, &keys[level], 0);
4532 }
4533 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
4534 eb = path->nodes[0];
4535 ret = replace_extents_in_leaf(trans, reloc_root, eb,
4536 group, reloc_inode);
4537 BUG_ON(ret);
4538 }
4539 btrfs_release_path(reloc_root, path);
4540 } else {
1a40e23b 4541 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
f82d02d9 4542 lowest_level);
1a40e23b
ZY		 4543 BUG_ON(ret);
4544 }
4545
1a40e23b
ZY		 4546 /*
4547 * replace tree blocks in the fs tree with tree blocks in
4548 * the reloc tree.
4549 */
4550 ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
4551 BUG_ON(ret < 0);
4552
4553 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
f82d02d9
YZ		 4554 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
4555 0, 0);
4556 BUG_ON(ret);
4557 extent_buffer_get(path->nodes[0]);
4558 eb = path->nodes[0];
4559 btrfs_release_path(reloc_root, path);
1a40e23b
ZY		 4560 ret = invalidate_extent_cache(reloc_root, eb, group, root);
4561 BUG_ON(ret);
4562 free_extent_buffer(eb);
4563 }
1a40e23b 4564
f82d02d9 4565 mutex_unlock(&root->fs_info->tree_reloc_mutex);
1a40e23b 4566 path->lowest_level = 0;
1a40e23b
ZY		 4567 return 0;
4568}
4569
4570static int noinline relocate_tree_block(struct btrfs_trans_handle *trans,
4571 struct btrfs_root *root,
4572 struct btrfs_path *path,
4573 struct btrfs_key *first_key,
4574 struct btrfs_ref_path *ref_path)
4575{
4576 int ret;
1a40e23b
ZY		 4577
4578 ret = relocate_one_path(trans, root, path, first_key,
4579 ref_path, NULL, NULL);
4580 BUG_ON(ret);
4581
4582 if (root == root->fs_info->extent_root)
4583 btrfs_extent_post_op(trans, root);
1a40e23b
ZY		 4584
4585 return 0;
4586}
4587
4588static int noinline del_extent_zero(struct btrfs_trans_handle *trans,
4589 struct btrfs_root *extent_root,
4590 struct btrfs_path *path,
4591 struct btrfs_key *extent_key)
4592{
4593 int ret;
4594
1a40e23b
ZY		 4595 ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
4596 if (ret)
4597 goto out;
4598 ret = btrfs_del_item(trans, extent_root, path);
4599out:
4600 btrfs_release_path(extent_root, path);
1a40e23b
ZY		 4601 return ret;
4602}
4603
4604static struct btrfs_root noinline *read_ref_root(struct btrfs_fs_info *fs_info,
4605 struct btrfs_ref_path *ref_path)
4606{
4607 struct btrfs_key root_key;
4608
4609 root_key.objectid = ref_path->root_objectid;
4610 root_key.type = BTRFS_ROOT_ITEM_KEY;
4611 if (is_cowonly_root(ref_path->root_objectid))
4612 root_key.offset = 0;
4613 else
4614 root_key.offset = (u64)-1;
4615
4616 return btrfs_read_fs_root_no_name(fs_info, &root_key);
4617}
4618
4619static int noinline relocate_one_extent(struct btrfs_root *extent_root,
4620 struct btrfs_path *path,
4621 struct btrfs_key *extent_key,
4622 struct btrfs_block_group_cache *group,
4623 struct inode *reloc_inode, int pass)
4624{
4625 struct btrfs_trans_handle *trans;
4626 struct btrfs_root *found_root;
4627 struct btrfs_ref_path *ref_path = NULL;
4628 struct disk_extent *new_extents = NULL;
4629 int nr_extents = 0;
4630 int loops;
4631 int ret;
4632 int level;
4633 struct btrfs_key first_key;
4634 u64 prev_block = 0;
4635
1a40e23b
ZY		 4636
4637 trans = btrfs_start_transaction(extent_root, 1);
4638 BUG_ON(!trans);
4639
4640 if (extent_key->objectid == 0) {
4641 ret = del_extent_zero(trans, extent_root, path, extent_key);
4642 goto out;
4643 }
4644
4645 ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
4646 if (!ref_path) {
4647 ret = -ENOMEM;
4648 goto out;
4649 }
4650
4651 for (loops = 0; ; loops++) {
4652 if (loops == 0) {
4653 ret = btrfs_first_ref_path(trans, extent_root, ref_path,
4654 extent_key->objectid);
4655 } else {
4656 ret = btrfs_next_ref_path(trans, extent_root, ref_path);
4657 }
4658 if (ret < 0)
4659 goto out;
4660 if (ret > 0)
4661 break;
4662
4663 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
4664 ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
4665 continue;
4666
4667 found_root = read_ref_root(extent_root->fs_info, ref_path);
4668 BUG_ON(!found_root);
4669 /*
4670 * for reference counted tree, only process reference paths
4671 * rooted at the latest committed root.
4672 */
4673 if (found_root->ref_cows &&
4674 ref_path->root_generation != found_root->root_key.offset)
4675 continue;
4676
4677 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
4678 if (pass == 0) {
4679 /*
4680 * copy data extents to new locations
4681 */
4682 u64 group_start = group->key.objectid;
4683 ret = relocate_data_extent(reloc_inode,
4684 extent_key,
4685 group_start);
4686 if (ret < 0)
4687 goto out;
4688 break;
4689 }
4690 level = 0;
4691 } else {
4692 level = ref_path->owner_objectid;
4693 }
4694
4695 if (prev_block != ref_path->nodes[level]) {
4696 struct extent_buffer *eb;
4697 u64 block_start = ref_path->nodes[level];
4698 u64 block_size = btrfs_level_size(found_root, level);
4699
4700 eb = read_tree_block(found_root, block_start,
4701 block_size, 0);
4702 btrfs_tree_lock(eb);
4703 BUG_ON(level != btrfs_header_level(eb));
4704
4705 if (level == 0)
4706 btrfs_item_key_to_cpu(eb, &first_key, 0);
4707 else
4708 btrfs_node_key_to_cpu(eb, &first_key, 0);
4709
4710 btrfs_tree_unlock(eb);
4711 free_extent_buffer(eb);
4712 prev_block = block_start;
4713 }
4714
4715 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID &&
4716 pass >= 2) {
4717 /*
4718 * use fallback method to process the remaining
4719 * references.
4720 */
4721 if (!new_extents) {
4722 u64 group_start = group->key.objectid;
4723 ret = get_new_locations(reloc_inode,
4724 extent_key,
4725 group_start, 0,
4726 &new_extents,
4727 &nr_extents);
4728 if (ret < 0)
4729 goto out;
4730 }
4731 btrfs_record_root_in_trans(found_root);
4732 ret = replace_one_extent(trans, found_root,
4733 path, extent_key,
4734 &first_key, ref_path,
4735 new_extents, nr_extents);
4736 if (ret < 0)
4737 goto out;
4738 continue;
4739 }
4740
4741 btrfs_record_root_in_trans(found_root);
4742 if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4743 ret = relocate_tree_block(trans, found_root, path,
4744 &first_key, ref_path);
4745 } else {
4746 /*
4747 * try to update data extent references while
4748 * keeping metadata shared between snapshots.
4749 */
4750 ret = relocate_one_path(trans, found_root, path,
4751 &first_key, ref_path,
4752 group, reloc_inode);
4753 }
4754 if (ret < 0)
4755 goto out;
4756 }
4757 ret = 0;
4758out:
4759 btrfs_end_transaction(trans, extent_root);
4760 kfree(new_extents);
4761 kfree(ref_path);
1a40e23b
ZY		 4762 return ret;
4763}
4764
ec44a35c
CM		 4765static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
4766{
4767 u64 num_devices;
4768 u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
4769 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
4770
a061fc8d 4771 num_devices = root->fs_info->fs_devices->num_devices;
ec44a35c
CM		 4772 if (num_devices == 1) {
4773 stripped |= BTRFS_BLOCK_GROUP_DUP;
4774 stripped = flags & ~stripped;
4775
4776 /* turn raid0 into single device chunks */
4777 if (flags & BTRFS_BLOCK_GROUP_RAID0)
4778 return stripped;
4779
4780 /* turn mirroring into duplication */
4781 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
4782 BTRFS_BLOCK_GROUP_RAID10))
4783 return stripped | BTRFS_BLOCK_GROUP_DUP;
4784 return flags;
4785 } else {
4786 /* they already had raid on here, just return */
ec44a35c
CM		 4787 if (flags & stripped)
4788 return flags;
4789
4790 stripped |= BTRFS_BLOCK_GROUP_DUP;
4791 stripped = flags & ~stripped;
4792
4793 /* switch duplicated blocks with raid1 */
4794 if (flags & BTRFS_BLOCK_GROUP_DUP)
4795 return stripped | BTRFS_BLOCK_GROUP_RAID1;
4796
4797 /* turn single device chunks into raid0 */
4798 return stripped | BTRFS_BLOCK_GROUP_RAID0;
4799 }
4800 return flags;
4801}
4802
0ef3e66b
CM		 4803int __alloc_chunk_for_shrink(struct btrfs_root *root,
4804 struct btrfs_block_group_cache *shrink_block_group,
4805 int force)
4806{
4807 struct btrfs_trans_handle *trans;
4808 u64 new_alloc_flags;
4809 u64 calc;
4810
c286ac48 4811 spin_lock(&shrink_block_group->lock);
0ef3e66b 4812 if (btrfs_block_group_used(&shrink_block_group->item) > 0) {
c286ac48 4813 spin_unlock(&shrink_block_group->lock);
c286ac48 4814
0ef3e66b 4815 trans = btrfs_start_transaction(root, 1);
c286ac48 4816 spin_lock(&shrink_block_group->lock);
7d9eb12c 4817
0ef3e66b
CM		 4818 new_alloc_flags = update_block_group_flags(root,
4819 shrink_block_group->flags);
4820 if (new_alloc_flags != shrink_block_group->flags) {
4821 calc =
4822 btrfs_block_group_used(&shrink_block_group->item);
4823 } else {
4824 calc = shrink_block_group->key.offset;
4825 }
c286ac48
CM		 4826 spin_unlock(&shrink_block_group->lock);
4827
0ef3e66b
CM		 4828 do_chunk_alloc(trans, root->fs_info->extent_root,
4829 calc + 2 * 1024 * 1024, new_alloc_flags, force);
7d9eb12c 4830
0ef3e66b 4831 btrfs_end_transaction(trans, root);
c286ac48
CM		 4832 } else
4833 spin_unlock(&shrink_block_group->lock);
0ef3e66b
CM		 4834 return 0;
4835}
4836
1a40e23b
ZY		 4837static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
4838 struct btrfs_root *root,
4839 u64 objectid, u64 size)
4840{
4841 struct btrfs_path *path;
4842 struct btrfs_inode_item *item;
4843 struct extent_buffer *leaf;
4844 int ret;
4845
4846 path = btrfs_alloc_path();
4847 if (!path)
4848 return -ENOMEM;
4849
4850 ret = btrfs_insert_empty_inode(trans, root, path, objectid);
4851 if (ret)
4852 goto out;
4853
4854 leaf = path->nodes[0];
4855 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
4856 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
4857 btrfs_set_inode_generation(leaf, item, 1);
4858 btrfs_set_inode_size(leaf, item, size);
4859 btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
4860 btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NODATASUM);
4861 btrfs_mark_buffer_dirty(leaf);
4862 btrfs_release_path(root, path);
4863out:
4864 btrfs_free_path(path);
4865 return ret;
4866}
4867
4868static struct inode noinline *create_reloc_inode(struct btrfs_fs_info *fs_info,
4869 struct btrfs_block_group_cache *group)
4870{
4871 struct inode *inode = NULL;
4872 struct btrfs_trans_handle *trans;
4873 struct btrfs_root *root;
4874 struct btrfs_key root_key;
4875 u64 objectid = BTRFS_FIRST_FREE_OBJECTID;
4876 int err = 0;
4877
4878 root_key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
4879 root_key.type = BTRFS_ROOT_ITEM_KEY;
4880 root_key.offset = (u64)-1;
4881 root = btrfs_read_fs_root_no_name(fs_info, &root_key);
4882 if (IS_ERR(root))
4883 return ERR_CAST(root);
4884
4885 trans = btrfs_start_transaction(root, 1);
4886 BUG_ON(!trans);
4887
4888 err = btrfs_find_free_objectid(trans, root, objectid, &objectid);
4889 if (err)
4890 goto out;
4891
4892 err = __insert_orphan_inode(trans, root, objectid, group->key.offset);
4893 BUG_ON(err);
4894
4895 err = btrfs_insert_file_extent(trans, root, objectid, 0, 0, 0,
c8b97818
CM		 4896 group->key.offset, 0, group->key.offset,
4897 0, 0, 0);
1a40e23b
ZY		 4898 BUG_ON(err);
4899
4900 inode = btrfs_iget_locked(root->fs_info->sb, objectid, root);
4901 if (inode->i_state & I_NEW) {
4902 BTRFS_I(inode)->root = root;
4903 BTRFS_I(inode)->location.objectid = objectid;
4904 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
4905 BTRFS_I(inode)->location.offset = 0;
4906 btrfs_read_locked_inode(inode);
4907 unlock_new_inode(inode);
4908 BUG_ON(is_bad_inode(inode));
4909 } else {
4910 BUG_ON(1);
4911 }
4912
4913 err = btrfs_orphan_add(trans, inode);
4914out:
4915 btrfs_end_transaction(trans, root);
4916 if (err) {
4917 if (inode)
4918 iput(inode);
4919 inode = ERR_PTR(err);
4920 }
4921 return inode;
4922}
4923
4924int btrfs_relocate_block_group(struct btrfs_root *root, u64 group_start)
edbd8d4e
CM		 4925{
4926 struct btrfs_trans_handle *trans;
edbd8d4e 4927 struct btrfs_path *path;
1a40e23b
ZY		 4928 struct btrfs_fs_info *info = root->fs_info;
4929 struct extent_buffer *leaf;
4930 struct inode *reloc_inode;
4931 struct btrfs_block_group_cache *block_group;
4932 struct btrfs_key key;
edbd8d4e
CM		 4933 u64 cur_byte;
4934 u64 total_found;
edbd8d4e
CM		 4935 u32 nritems;
4936 int ret;
a061fc8d 4937 int progress;
1a40e23b 4938 int pass = 0;
edbd8d4e 4939
1a40e23b
ZY		 4940 root = root->fs_info->extent_root;
4941
4942 block_group = btrfs_lookup_block_group(info, group_start);
4943 BUG_ON(!block_group);
8f18cf13 4944
1a40e23b
ZY		 4945 printk("btrfs relocating block group %llu flags %llu\n",
4946 (unsigned long long)block_group->key.objectid,
4947 (unsigned long long)block_group->flags);
8f18cf13 4948
edbd8d4e 4949 path = btrfs_alloc_path();
1a40e23b 4950 BUG_ON(!path);
edbd8d4e 4951
1a40e23b
ZY		 4952 reloc_inode = create_reloc_inode(info, block_group);
4953 BUG_ON(IS_ERR(reloc_inode));
323da79c 4954
1a40e23b
ZY		 4955 __alloc_chunk_for_shrink(root, block_group, 1);
4956 block_group->ro = 1;
4957 block_group->space_info->total_bytes -= block_group->key.offset;
323da79c 4958
1a40e23b
ZY		 4959 btrfs_start_delalloc_inodes(info->tree_root);
4960 btrfs_wait_ordered_extents(info->tree_root, 0);
4961again:
edbd8d4e 4962 total_found = 0;
a061fc8d 4963 progress = 0;
1a40e23b 4964 key.objectid = block_group->key.objectid;
edbd8d4e
CM		 4965 key.offset = 0;
4966 key.type = 0;
73e48b27 4967 cur_byte = key.objectid;
4313b399 4968
1a40e23b
ZY		 4969 trans = btrfs_start_transaction(info->tree_root, 1);
4970 btrfs_commit_transaction(trans, info->tree_root);
ea8c2819 4971
1a40e23b
ZY		 4972 mutex_lock(&root->fs_info->cleaner_mutex);
4973 btrfs_clean_old_snapshots(info->tree_root);
4974 btrfs_remove_leaf_refs(info->tree_root, (u64)-1, 1);
4975 mutex_unlock(&root->fs_info->cleaner_mutex);
ea8c2819 4976
73e48b27 4977 while(1) {
edbd8d4e
CM		 4978 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4979 if (ret < 0)
4980 goto out;
7d9eb12c 4981next:
edbd8d4e 4982 leaf = path->nodes[0];
73e48b27 4983 nritems = btrfs_header_nritems(leaf);
73e48b27
Y		 4984 if (path->slots[0] >= nritems) {
4985 ret = btrfs_next_leaf(root, path);
4986 if (ret < 0)
4987 goto out;
4988 if (ret == 1) {
4989 ret = 0;
4990 break;
edbd8d4e 4991 }
73e48b27
Y		 4992 leaf = path->nodes[0];
4993 nritems = btrfs_header_nritems(leaf);
edbd8d4e 4994 }
73e48b27 4995
1a40e23b 4996 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
725c8463 4997
1a40e23b
ZY		 4998 if (key.objectid >= block_group->key.objectid +
4999 block_group->key.offset)
8f18cf13
CM		 5000 break;
5001
725c8463 5002 if (progress && need_resched()) {
725c8463 5003 btrfs_release_path(root, path);
1a40e23b 5004 cond_resched();
725c8463 5005 progress = 0;
1a40e23b 5006 continue;
725c8463
CM		 5007 }
5008 progress = 1;
5009
1a40e23b
ZY		 5010 if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY ||
5011 key.objectid + key.offset <= cur_byte) {
edbd8d4e 5012 path->slots[0]++;
edbd8d4e
CM		 5013 goto next;
5014 }
73e48b27 5015
edbd8d4e 5016 total_found++;
1a40e23b 5017 cur_byte = key.objectid + key.offset;
edbd8d4e 5018 btrfs_release_path(root, path);
edbd8d4e 5019
1a40e23b
ZY		 5020 __alloc_chunk_for_shrink(root, block_group, 0);
5021 ret = relocate_one_extent(root, path, &key, block_group,
5022 reloc_inode, pass);
5023 BUG_ON(ret < 0);
edbd8d4e 5024
1a40e23b
ZY		 5025 key.objectid = cur_byte;
5026 key.type = 0;
5027 key.offset = 0;
edbd8d4e
CM		 5028 }
5029
1a40e23b 5030 btrfs_release_path(root, path);
7d9eb12c 5031
1a40e23b
ZY		 5032 if (pass == 0) {
5033 btrfs_wait_ordered_range(reloc_inode, 0, (u64)-1);
5034 invalidate_mapping_pages(reloc_inode->i_mapping, 0, -1);
5035 WARN_ON(reloc_inode->i_mapping->nrpages);
8e8a1e31 5036 }
73e48b27 5037
1a40e23b
ZY		 5038 if (total_found > 0) {
5039 printk("btrfs found %llu extents in pass %d\n",
5040 (unsigned long long)total_found, pass);
5041 pass++;
5042 goto again;
0f9dd46c 5043 }
0ef3e66b 5044
1a40e23b
ZY		 5045 /* delete reloc_inode */
5046 iput(reloc_inode);
5047
5048 /* unpin extents in this range */
5049 trans = btrfs_start_transaction(info->tree_root, 1);
5050 btrfs_commit_transaction(trans, info->tree_root);
0ef3e66b 5051
1a40e23b
ZY		 5052 spin_lock(&block_group->lock);
5053 WARN_ON(block_group->pinned > 0);
5054 WARN_ON(block_group->reserved > 0);
5055 WARN_ON(btrfs_block_group_used(&block_group->item) > 0);
5056 spin_unlock(&block_group->lock);
5057 ret = 0;
edbd8d4e 5058out:
1a40e23b 5059 btrfs_free_path(path);
edbd8d4e
CM		 5060 return ret;
5061}
5062
0b86a832
CM		 5063int find_first_block_group(struct btrfs_root *root, struct btrfs_path *path,
5064 struct btrfs_key *key)
5065{
925baedd 5066 int ret = 0;
0b86a832
CM		 5067 struct btrfs_key found_key;
5068 struct extent_buffer *leaf;
5069 int slot;
edbd8d4e 5070
0b86a832
CM		 5071 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
5072 if (ret < 0)
925baedd
CM		 5073 goto out;
5074
0b86a832
CM		 5075 while(1) {
5076 slot = path->slots[0];
edbd8d4e 5077 leaf = path->nodes[0];
0b86a832
CM		 5078 if (slot >= btrfs_header_nritems(leaf)) {
5079 ret = btrfs_next_leaf(root, path);
5080 if (ret == 0)
5081 continue;
5082 if (ret < 0)
925baedd 5083 goto out;
0b86a832 5084 break;
edbd8d4e 5085 }
0b86a832 5086 btrfs_item_key_to_cpu(leaf, &found_key, slot);
edbd8d4e 5087
0b86a832 5088 if (found_key.objectid >= key->objectid &&
925baedd
CM		 5089 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
5090 ret = 0;
5091 goto out;
5092 }
0b86a832 5093 path->slots[0]++;
edbd8d4e 5094 }
0b86a832 5095 ret = -ENOENT;
925baedd 5096out:
0b86a832 5097 return ret;
edbd8d4e
CM		 5098}
5099
1a40e23b
ZY		 5100int btrfs_free_block_groups(struct btrfs_fs_info *info)
5101{
5102 struct btrfs_block_group_cache *block_group;
5103 struct rb_node *n;
5104
1a40e23b
ZY		 5105 spin_lock(&info->block_group_cache_lock);
5106 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
5107 block_group = rb_entry(n, struct btrfs_block_group_cache,
5108 cache_node);
5109
5110 spin_unlock(&info->block_group_cache_lock);
5111 btrfs_remove_free_space_cache(block_group);
5112 spin_lock(&info->block_group_cache_lock);
5113
5114 rb_erase(&block_group->cache_node,
5115 &info->block_group_cache_tree);
80eb234a 5116 down_write(&block_group->space_info->groups_sem);
1a40e23b 5117 list_del(&block_group->list);
80eb234a 5118 up_write(&block_group->space_info->groups_sem);
1a40e23b
ZY		 5119 kfree(block_group);
5120 }
5121 spin_unlock(&info->block_group_cache_lock);
1a40e23b
ZY		 5122 return 0;
5123}
5124
9078a3e1
CM		 5125int btrfs_read_block_groups(struct btrfs_root *root)
5126{
5127 struct btrfs_path *path;
5128 int ret;
9078a3e1 5129 struct btrfs_block_group_cache *cache;
be744175 5130 struct btrfs_fs_info *info = root->fs_info;
6324fbf3 5131 struct btrfs_space_info *space_info;
9078a3e1
CM		 5132 struct btrfs_key key;
5133 struct btrfs_key found_key;
5f39d397 5134 struct extent_buffer *leaf;
96b5179d 5135
be744175 5136 root = info->extent_root;
9078a3e1 5137 key.objectid = 0;
0b86a832 5138 key.offset = 0;
9078a3e1 5139 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
9078a3e1
CM		 5140 path = btrfs_alloc_path();
5141 if (!path)
5142 return -ENOMEM;
5143
5144 while(1) {
0b86a832
CM		 5145 ret = find_first_block_group(root, path, &key);
5146 if (ret > 0) {
5147 ret = 0;
5148 goto error;
9078a3e1 5149 }
0b86a832
CM		 5150 if (ret != 0)
5151 goto error;
5152
5f39d397
CM		 5153 leaf = path->nodes[0];
5154 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
8f18cf13 5155 cache = kzalloc(sizeof(*cache), GFP_NOFS);
9078a3e1 5156 if (!cache) {
0b86a832 5157 ret = -ENOMEM;
9078a3e1
CM		 5158 break;
5159 }
3e1ad54f 5160
c286ac48 5161 spin_lock_init(&cache->lock);
25179201 5162 mutex_init(&cache->alloc_mutex);
0f9dd46c 5163 INIT_LIST_HEAD(&cache->list);
5f39d397
CM		 5164 read_extent_buffer(leaf, &cache->item,
5165 btrfs_item_ptr_offset(leaf, path->slots[0]),
5166 sizeof(cache->item));
9078a3e1 5167 memcpy(&cache->key, &found_key, sizeof(found_key));
0b86a832 5168
9078a3e1
CM		 5169 key.objectid = found_key.objectid + found_key.offset;
5170 btrfs_release_path(root, path);
0b86a832 5171 cache->flags = btrfs_block_group_flags(&cache->item);
96b5179d 5172
6324fbf3
CM		 5173 ret = update_space_info(info, cache->flags, found_key.offset,
5174 btrfs_block_group_used(&cache->item),
5175 &space_info);
5176 BUG_ON(ret);
5177 cache->space_info = space_info;
80eb234a
JB		 5178 down_write(&space_info->groups_sem);
5179 list_add_tail(&cache->list, &space_info->block_groups);
5180 up_write(&space_info->groups_sem);
0f9dd46c
JB		 5181
5182 ret = btrfs_add_block_group_cache(root->fs_info, cache);
5183 BUG_ON(ret);
75ccf47d
CM		 5184
5185 set_avail_alloc_bits(root->fs_info, cache->flags);
9078a3e1 5186 }
0b86a832
CM		 5187 ret = 0;
5188error:
9078a3e1 5189 btrfs_free_path(path);
0b86a832 5190 return ret;
9078a3e1 5191}
6324fbf3
CM		 5192
5193int btrfs_make_block_group(struct btrfs_trans_handle *trans,
5194 struct btrfs_root *root, u64 bytes_used,
e17cade2 5195 u64 type, u64 chunk_objectid, u64 chunk_offset,
6324fbf3
CM		 5196 u64 size)
5197{
5198 int ret;
6324fbf3
CM		 5199 struct btrfs_root *extent_root;
5200 struct btrfs_block_group_cache *cache;
6324fbf3
CM		 5201
5202 extent_root = root->fs_info->extent_root;
6324fbf3 5203
e02119d5
CM		 5204 root->fs_info->last_trans_new_blockgroup = trans->transid;
5205
8f18cf13 5206 cache = kzalloc(sizeof(*cache), GFP_NOFS);
0f9dd46c
JB		 5207 if (!cache)
5208 return -ENOMEM;
5209
e17cade2 5210 cache->key.objectid = chunk_offset;
6324fbf3 5211 cache->key.offset = size;
c286ac48 5212 spin_lock_init(&cache->lock);
25179201 5213 mutex_init(&cache->alloc_mutex);
0f9dd46c 5214 INIT_LIST_HEAD(&cache->list);
6324fbf3 5215 btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY);
0ef3e66b 5216
6324fbf3 5217 btrfs_set_block_group_used(&cache->item, bytes_used);
6324fbf3
CM		 5218 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
5219 cache->flags = type;
5220 btrfs_set_block_group_flags(&cache->item, type);
5221
5222 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
5223 &cache->space_info);
5224 BUG_ON(ret);
80eb234a
JB		 5225 down_write(&cache->space_info->groups_sem);
5226 list_add_tail(&cache->list, &cache->space_info->block_groups);
5227 up_write(&cache->space_info->groups_sem);
6324fbf3 5228
0f9dd46c
JB		 5229 ret = btrfs_add_block_group_cache(root->fs_info, cache);
5230 BUG_ON(ret);
c286ac48 5231
6324fbf3
CM		 5232 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
5233 sizeof(cache->item));
5234 BUG_ON(ret);
5235
87ef2bb4
CM		 5236 finish_current_insert(trans, extent_root, 0);
5237 ret = del_pending_extents(trans, extent_root, 0);
6324fbf3 5238 BUG_ON(ret);
d18a2c44 5239 set_avail_alloc_bits(extent_root->fs_info, type);
925baedd 5240
6324fbf3
CM		 5241 return 0;
5242}
1a40e23b
ZY		 5243
5244int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
5245 struct btrfs_root *root, u64 group_start)
5246{
5247 struct btrfs_path *path;
5248 struct btrfs_block_group_cache *block_group;
5249 struct btrfs_key key;
5250 int ret;
5251
1a40e23b
ZY		 5252 root = root->fs_info->extent_root;
5253
5254 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
5255 BUG_ON(!block_group);
5256
5257 memcpy(&key, &block_group->key, sizeof(key));
5258
5259 path = btrfs_alloc_path();
5260 BUG_ON(!path);
5261
5262 btrfs_remove_free_space_cache(block_group);
5263 rb_erase(&block_group->cache_node,
5264 &root->fs_info->block_group_cache_tree);
80eb234a 5265 down_write(&block_group->space_info->groups_sem);
1a40e23b 5266 list_del(&block_group->list);
80eb234a 5267 up_write(&block_group->space_info->groups_sem);
1a40e23b
ZY		 5268
5269 /*
5270 memset(shrink_block_group, 0, sizeof(*shrink_block_group));
5271 kfree(shrink_block_group);
5272 */
5273
5274 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
5275 if (ret > 0)
5276 ret = -EIO;
5277 if (ret < 0)
5278 goto out;
5279
5280 ret = btrfs_del_item(trans, root, path);
5281out:
5282 btrfs_free_path(path);
5283 return ret;
5284}
Linux 6.x block layer and io_uring tree(s)