blob: 340c14715091a94c5b1d0e4358d9ec74571cccde [file] [log] [blame]
Chris Mason6cbd5572007-06-12 09:07:21 -04001/*
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 */
Zach Brownec6b9102007-07-11 10:00:37 -040018#include <linux/sched.h>
Chris Masonedbd8d42007-12-21 16:27:24 -050019#include <linux/pagemap.h>
Chris Masonec44a352008-04-28 15:29:52 -040020#include <linux/writeback.h>
David Woodhouse21af8042008-08-12 14:13:26 +010021#include <linux/blkdev.h>
Chris Masonb7a9f292009-02-04 09:23:45 -050022#include <linux/sort.h>
Chris Mason4184ea72009-03-10 12:39:20 -040023#include <linux/rcupdate.h>
Josef Bacik817d52f2009-07-13 21:29:25 -040024#include <linux/kthread.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090025#include <linux/slab.h>
Chris Mason4b4e25f2008-11-20 10:22:27 -050026#include "compat.h"
Chris Mason74493f72007-12-11 09:25:06 -050027#include "hash.h"
Chris Masonfec577f2007-02-26 10:40:21 -050028#include "ctree.h"
29#include "disk-io.h"
30#include "print-tree.h"
Chris Masone089f052007-03-16 16:20:31 -040031#include "transaction.h"
Chris Mason0b86a832008-03-24 15:01:56 -040032#include "volumes.h"
Chris Mason925baed2008-06-25 16:01:30 -040033#include "locking.h"
Chris Masonfa9c0d72009-04-03 09:47:43 -040034#include "free-space-cache.h"
Chris Masonfec577f2007-02-26 10:40:21 -050035
Chris Mason0e4f8f82011-04-15 16:05:44 -040036/* control flags for do_chunk_alloc's force field
37 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
38 * if we really need one.
39 *
40 * CHUNK_ALLOC_FORCE means it must try to allocate one
41 *
42 * CHUNK_ALLOC_LIMITED means to only try and allocate one
43 * if we have very few chunks already allocated. This is
44 * used as part of the clustering code to help make sure
45 * we have a good pool of storage to cluster in, without
46 * filling the FS with empty chunks
47 *
48 */
49enum {
50 CHUNK_ALLOC_NO_FORCE = 0,
51 CHUNK_ALLOC_FORCE = 1,
52 CHUNK_ALLOC_LIMITED = 2,
53};
54
Josef Bacikf3465ca2008-11-12 14:19:50 -050055static int update_block_group(struct btrfs_trans_handle *trans,
56 struct btrfs_root *root,
Yan, Zhengf0486c62010-05-16 10:46:25 -040057 u64 bytenr, u64 num_bytes, int alloc);
Yan Zheng5d4f98a2009-06-10 10:45:14 -040058static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
59 struct btrfs_root *root,
60 u64 bytenr, u64 num_bytes, u64 parent,
61 u64 root_objectid, u64 owner_objectid,
62 u64 owner_offset, int refs_to_drop,
63 struct btrfs_delayed_extent_op *extra_op);
64static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
65 struct extent_buffer *leaf,
66 struct btrfs_extent_item *ei);
67static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
68 struct btrfs_root *root,
69 u64 parent, u64 root_objectid,
70 u64 flags, u64 owner, u64 offset,
71 struct btrfs_key *ins, int ref_mod);
72static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
73 struct btrfs_root *root,
74 u64 parent, u64 root_objectid,
75 u64 flags, struct btrfs_disk_key *key,
76 int level, struct btrfs_key *ins);
Josef Bacik6a632092009-02-20 11:00:09 -050077static int do_chunk_alloc(struct btrfs_trans_handle *trans,
78 struct btrfs_root *extent_root, u64 alloc_bytes,
79 u64 flags, int force);
Yan Zheng11833d62009-09-11 16:11:19 -040080static int find_next_key(struct btrfs_path *path, int level,
81 struct btrfs_key *key);
Josef Bacik9ed74f22009-09-11 16:12:44 -040082static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
83 int dump_block_groups);
Josef Bacik6a632092009-02-20 11:00:09 -050084
Josef Bacik817d52f2009-07-13 21:29:25 -040085static noinline int
86block_group_cache_done(struct btrfs_block_group_cache *cache)
87{
88 smp_mb();
89 return cache->cached == BTRFS_CACHE_FINISHED;
90}
91
Josef Bacik0f9dd462008-09-23 13:14:11 -040092static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
93{
94 return (cache->flags & bits) == bits;
95}
96
David Sterba62a45b62011-04-20 15:52:26 +020097static void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
Josef Bacik11dfe352009-11-13 20:12:59 +000098{
99 atomic_inc(&cache->count);
100}
101
102void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
103{
Yan, Zhengf0486c62010-05-16 10:46:25 -0400104 if (atomic_dec_and_test(&cache->count)) {
105 WARN_ON(cache->pinned > 0);
106 WARN_ON(cache->reserved > 0);
107 WARN_ON(cache->reserved_pinned > 0);
Li Zefan34d52cb2011-03-29 13:46:06 +0800108 kfree(cache->free_space_ctl);
Josef Bacik11dfe352009-11-13 20:12:59 +0000109 kfree(cache);
Yan, Zhengf0486c62010-05-16 10:46:25 -0400110 }
Josef Bacik11dfe352009-11-13 20:12:59 +0000111}
112
Josef Bacik0f9dd462008-09-23 13:14:11 -0400113/*
114 * this adds the block group to the fs_info rb tree for the block group
115 * cache
116 */
Christoph Hellwigb2950862008-12-02 09:54:17 -0500117static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
Josef Bacik0f9dd462008-09-23 13:14:11 -0400118 struct btrfs_block_group_cache *block_group)
119{
120 struct rb_node **p;
121 struct rb_node *parent = NULL;
122 struct btrfs_block_group_cache *cache;
123
124 spin_lock(&info->block_group_cache_lock);
125 p = &info->block_group_cache_tree.rb_node;
126
127 while (*p) {
128 parent = *p;
129 cache = rb_entry(parent, struct btrfs_block_group_cache,
130 cache_node);
131 if (block_group->key.objectid < cache->key.objectid) {
132 p = &(*p)->rb_left;
133 } else if (block_group->key.objectid > cache->key.objectid) {
134 p = &(*p)->rb_right;
135 } else {
136 spin_unlock(&info->block_group_cache_lock);
137 return -EEXIST;
138 }
139 }
140
141 rb_link_node(&block_group->cache_node, parent, p);
142 rb_insert_color(&block_group->cache_node,
143 &info->block_group_cache_tree);
144 spin_unlock(&info->block_group_cache_lock);
145
146 return 0;
147}
148
149/*
150 * This will return the block group at or after bytenr if contains is 0, else
151 * it will return the block group that contains the bytenr
152 */
153static struct btrfs_block_group_cache *
154block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
155 int contains)
156{
157 struct btrfs_block_group_cache *cache, *ret = NULL;
158 struct rb_node *n;
159 u64 end, start;
160
161 spin_lock(&info->block_group_cache_lock);
162 n = info->block_group_cache_tree.rb_node;
163
164 while (n) {
165 cache = rb_entry(n, struct btrfs_block_group_cache,
166 cache_node);
167 end = cache->key.objectid + cache->key.offset - 1;
168 start = cache->key.objectid;
169
170 if (bytenr < start) {
171 if (!contains && (!ret || start < ret->key.objectid))
172 ret = cache;
173 n = n->rb_left;
174 } else if (bytenr > start) {
175 if (contains && bytenr <= end) {
176 ret = cache;
177 break;
178 }
179 n = n->rb_right;
180 } else {
181 ret = cache;
182 break;
183 }
184 }
Yan Zhengd2fb3432008-12-11 16:30:39 -0500185 if (ret)
Josef Bacik11dfe352009-11-13 20:12:59 +0000186 btrfs_get_block_group(ret);
Josef Bacik0f9dd462008-09-23 13:14:11 -0400187 spin_unlock(&info->block_group_cache_lock);
188
189 return ret;
190}
191
Yan Zheng11833d62009-09-11 16:11:19 -0400192static int add_excluded_extent(struct btrfs_root *root,
193 u64 start, u64 num_bytes)
Josef Bacik817d52f2009-07-13 21:29:25 -0400194{
Yan Zheng11833d62009-09-11 16:11:19 -0400195 u64 end = start + num_bytes - 1;
196 set_extent_bits(&root->fs_info->freed_extents[0],
197 start, end, EXTENT_UPTODATE, GFP_NOFS);
198 set_extent_bits(&root->fs_info->freed_extents[1],
199 start, end, EXTENT_UPTODATE, GFP_NOFS);
200 return 0;
Josef Bacik817d52f2009-07-13 21:29:25 -0400201}
202
Yan Zheng11833d62009-09-11 16:11:19 -0400203static void free_excluded_extents(struct btrfs_root *root,
204 struct btrfs_block_group_cache *cache)
Josef Bacik817d52f2009-07-13 21:29:25 -0400205{
Yan Zheng11833d62009-09-11 16:11:19 -0400206 u64 start, end;
207
208 start = cache->key.objectid;
209 end = start + cache->key.offset - 1;
210
211 clear_extent_bits(&root->fs_info->freed_extents[0],
212 start, end, EXTENT_UPTODATE, GFP_NOFS);
213 clear_extent_bits(&root->fs_info->freed_extents[1],
214 start, end, EXTENT_UPTODATE, GFP_NOFS);
215}
216
217static int exclude_super_stripes(struct btrfs_root *root,
218 struct btrfs_block_group_cache *cache)
219{
Josef Bacik817d52f2009-07-13 21:29:25 -0400220 u64 bytenr;
221 u64 *logical;
222 int stripe_len;
223 int i, nr, ret;
224
Yan, Zheng06b23312009-11-26 09:31:11 +0000225 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
226 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
227 cache->bytes_super += stripe_len;
228 ret = add_excluded_extent(root, cache->key.objectid,
229 stripe_len);
230 BUG_ON(ret);
231 }
232
Josef Bacik817d52f2009-07-13 21:29:25 -0400233 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
234 bytenr = btrfs_sb_offset(i);
235 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
236 cache->key.objectid, bytenr,
237 0, &logical, &nr, &stripe_len);
238 BUG_ON(ret);
Yan Zheng11833d62009-09-11 16:11:19 -0400239
Josef Bacik817d52f2009-07-13 21:29:25 -0400240 while (nr--) {
Josef Bacik1b2da372009-09-11 16:11:20 -0400241 cache->bytes_super += stripe_len;
Yan Zheng11833d62009-09-11 16:11:19 -0400242 ret = add_excluded_extent(root, logical[nr],
243 stripe_len);
244 BUG_ON(ret);
Josef Bacik817d52f2009-07-13 21:29:25 -0400245 }
Yan Zheng11833d62009-09-11 16:11:19 -0400246
Josef Bacik817d52f2009-07-13 21:29:25 -0400247 kfree(logical);
248 }
Josef Bacik817d52f2009-07-13 21:29:25 -0400249 return 0;
250}
251
Yan Zheng11833d62009-09-11 16:11:19 -0400252static struct btrfs_caching_control *
253get_caching_control(struct btrfs_block_group_cache *cache)
254{
255 struct btrfs_caching_control *ctl;
256
257 spin_lock(&cache->lock);
258 if (cache->cached != BTRFS_CACHE_STARTED) {
259 spin_unlock(&cache->lock);
260 return NULL;
261 }
262
Josef Bacikdde5abe2010-09-16 16:17:03 -0400263 /* We're loading it the fast way, so we don't have a caching_ctl. */
264 if (!cache->caching_ctl) {
265 spin_unlock(&cache->lock);
266 return NULL;
267 }
268
Yan Zheng11833d62009-09-11 16:11:19 -0400269 ctl = cache->caching_ctl;
270 atomic_inc(&ctl->count);
271 spin_unlock(&cache->lock);
272 return ctl;
273}
274
275static void put_caching_control(struct btrfs_caching_control *ctl)
276{
277 if (atomic_dec_and_test(&ctl->count))
278 kfree(ctl);
279}
280
Josef Bacik0f9dd462008-09-23 13:14:11 -0400281/*
282 * this is only called by cache_block_group, since we could have freed extents
283 * we need to check the pinned_extents for any extents that can't be used yet
284 * since their free space will be released as soon as the transaction commits.
285 */
Josef Bacik817d52f2009-07-13 21:29:25 -0400286static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
Josef Bacik0f9dd462008-09-23 13:14:11 -0400287 struct btrfs_fs_info *info, u64 start, u64 end)
288{
Josef Bacik817d52f2009-07-13 21:29:25 -0400289 u64 extent_start, extent_end, size, total_added = 0;
Josef Bacik0f9dd462008-09-23 13:14:11 -0400290 int ret;
291
292 while (start < end) {
Yan Zheng11833d62009-09-11 16:11:19 -0400293 ret = find_first_extent_bit(info->pinned_extents, start,
Josef Bacik0f9dd462008-09-23 13:14:11 -0400294 &extent_start, &extent_end,
Yan Zheng11833d62009-09-11 16:11:19 -0400295 EXTENT_DIRTY | EXTENT_UPTODATE);
Josef Bacik0f9dd462008-09-23 13:14:11 -0400296 if (ret)
297 break;
298
Yan, Zheng06b23312009-11-26 09:31:11 +0000299 if (extent_start <= start) {
Josef Bacik0f9dd462008-09-23 13:14:11 -0400300 start = extent_end + 1;
301 } else if (extent_start > start && extent_start < end) {
302 size = extent_start - start;
Josef Bacik817d52f2009-07-13 21:29:25 -0400303 total_added += size;
Josef Bacikea6a4782008-11-20 12:16:16 -0500304 ret = btrfs_add_free_space(block_group, start,
305 size);
Josef Bacik0f9dd462008-09-23 13:14:11 -0400306 BUG_ON(ret);
307 start = extent_end + 1;
308 } else {
309 break;
310 }
311 }
312
313 if (start < end) {
314 size = end - start;
Josef Bacik817d52f2009-07-13 21:29:25 -0400315 total_added += size;
Josef Bacikea6a4782008-11-20 12:16:16 -0500316 ret = btrfs_add_free_space(block_group, start, size);
Josef Bacik0f9dd462008-09-23 13:14:11 -0400317 BUG_ON(ret);
318 }
319
Josef Bacik817d52f2009-07-13 21:29:25 -0400320 return total_added;
Josef Bacik0f9dd462008-09-23 13:14:11 -0400321}
322
Josef Bacik817d52f2009-07-13 21:29:25 -0400323static int caching_kthread(void *data)
Chris Masone37c9e62007-05-09 20:13:14 -0400324{
Josef Bacik817d52f2009-07-13 21:29:25 -0400325 struct btrfs_block_group_cache *block_group = data;
326 struct btrfs_fs_info *fs_info = block_group->fs_info;
Yan Zheng11833d62009-09-11 16:11:19 -0400327 struct btrfs_caching_control *caching_ctl = block_group->caching_ctl;
328 struct btrfs_root *extent_root = fs_info->extent_root;
Chris Masone37c9e62007-05-09 20:13:14 -0400329 struct btrfs_path *path;
Chris Mason5f39d392007-10-15 16:14:19 -0400330 struct extent_buffer *leaf;
Yan Zheng11833d62009-09-11 16:11:19 -0400331 struct btrfs_key key;
Josef Bacik817d52f2009-07-13 21:29:25 -0400332 u64 total_found = 0;
Yan Zheng11833d62009-09-11 16:11:19 -0400333 u64 last = 0;
334 u32 nritems;
335 int ret = 0;
Chris Masonf510cfe2007-10-15 16:14:48 -0400336
Chris Masone37c9e62007-05-09 20:13:14 -0400337 path = btrfs_alloc_path();
338 if (!path)
339 return -ENOMEM;
Yan7d7d6062007-09-14 16:15:28 -0400340
Josef Bacik817d52f2009-07-13 21:29:25 -0400341 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
Yan Zheng11833d62009-09-11 16:11:19 -0400342
Chris Mason5cd57b22008-06-25 16:01:30 -0400343 /*
Josef Bacik817d52f2009-07-13 21:29:25 -0400344 * We don't want to deadlock with somebody trying to allocate a new
345 * extent for the extent root while also trying to search the extent
346 * root to add free space. So we skip locking and search the commit
347 * root, since its read-only
Chris Mason5cd57b22008-06-25 16:01:30 -0400348 */
349 path->skip_locking = 1;
Josef Bacik817d52f2009-07-13 21:29:25 -0400350 path->search_commit_root = 1;
Josef Bacik026fd312011-05-13 10:32:11 -0400351 path->reada = 1;
Josef Bacik817d52f2009-07-13 21:29:25 -0400352
Yan Zhenge4404d62008-12-12 10:03:26 -0500353 key.objectid = last;
Chris Masone37c9e62007-05-09 20:13:14 -0400354 key.offset = 0;
Yan Zheng11833d62009-09-11 16:11:19 -0400355 key.type = BTRFS_EXTENT_ITEM_KEY;
Chris Mason013f1b12009-07-31 14:57:55 -0400356again:
Yan Zheng11833d62009-09-11 16:11:19 -0400357 mutex_lock(&caching_ctl->mutex);
Chris Mason013f1b12009-07-31 14:57:55 -0400358 /* need to make sure the commit_root doesn't disappear */
359 down_read(&fs_info->extent_commit_sem);
360
Yan Zheng11833d62009-09-11 16:11:19 -0400361 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
Chris Masone37c9e62007-05-09 20:13:14 -0400362 if (ret < 0)
Josef Bacikef8bbdf2008-09-23 13:14:11 -0400363 goto err;
Yan Zhenga512bbf2008-12-08 16:46:26 -0500364
Yan Zheng11833d62009-09-11 16:11:19 -0400365 leaf = path->nodes[0];
366 nritems = btrfs_header_nritems(leaf);
367
Chris Masond3977122009-01-05 21:25:51 -0500368 while (1) {
David Sterba7841cb22011-05-31 18:07:27 +0200369 if (btrfs_fs_closing(fs_info) > 1) {
Yan Zhengf25784b2009-07-28 08:41:57 -0400370 last = (u64)-1;
Josef Bacik817d52f2009-07-13 21:29:25 -0400371 break;
Yan Zhengf25784b2009-07-28 08:41:57 -0400372 }
Josef Bacik817d52f2009-07-13 21:29:25 -0400373
Yan Zheng11833d62009-09-11 16:11:19 -0400374 if (path->slots[0] < nritems) {
375 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
376 } else {
377 ret = find_next_key(path, 0, &key);
378 if (ret)
Chris Masone37c9e62007-05-09 20:13:14 -0400379 break;
Josef Bacik817d52f2009-07-13 21:29:25 -0400380
Josef Bacik589d8ad2011-05-11 17:30:53 -0400381 if (need_resched() ||
382 btrfs_next_leaf(extent_root, path)) {
383 caching_ctl->progress = last;
Chris Masonff5714c2011-05-28 07:00:39 -0400384 btrfs_release_path(path);
Josef Bacik589d8ad2011-05-11 17:30:53 -0400385 up_read(&fs_info->extent_commit_sem);
386 mutex_unlock(&caching_ctl->mutex);
Yan Zheng11833d62009-09-11 16:11:19 -0400387 cond_resched();
Josef Bacik589d8ad2011-05-11 17:30:53 -0400388 goto again;
389 }
390 leaf = path->nodes[0];
391 nritems = btrfs_header_nritems(leaf);
392 continue;
Yan Zheng11833d62009-09-11 16:11:19 -0400393 }
Josef Bacik817d52f2009-07-13 21:29:25 -0400394
Yan Zheng11833d62009-09-11 16:11:19 -0400395 if (key.objectid < block_group->key.objectid) {
396 path->slots[0]++;
Josef Bacik817d52f2009-07-13 21:29:25 -0400397 continue;
Chris Masone37c9e62007-05-09 20:13:14 -0400398 }
Josef Bacik0f9dd462008-09-23 13:14:11 -0400399
Chris Masone37c9e62007-05-09 20:13:14 -0400400 if (key.objectid >= block_group->key.objectid +
Josef Bacik0f9dd462008-09-23 13:14:11 -0400401 block_group->key.offset)
Yan7d7d6062007-09-14 16:15:28 -0400402 break;
Yan7d7d6062007-09-14 16:15:28 -0400403
Yan Zheng11833d62009-09-11 16:11:19 -0400404 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
Josef Bacik817d52f2009-07-13 21:29:25 -0400405 total_found += add_new_free_space(block_group,
406 fs_info, last,
407 key.objectid);
Yan7d7d6062007-09-14 16:15:28 -0400408 last = key.objectid + key.offset;
Josef Bacik817d52f2009-07-13 21:29:25 -0400409
Yan Zheng11833d62009-09-11 16:11:19 -0400410 if (total_found > (1024 * 1024 * 2)) {
411 total_found = 0;
412 wake_up(&caching_ctl->wait);
413 }
Josef Bacik817d52f2009-07-13 21:29:25 -0400414 }
Chris Masone37c9e62007-05-09 20:13:14 -0400415 path->slots[0]++;
416 }
Josef Bacikef8bbdf2008-09-23 13:14:11 -0400417 ret = 0;
Josef Bacik817d52f2009-07-13 21:29:25 -0400418
419 total_found += add_new_free_space(block_group, fs_info, last,
420 block_group->key.objectid +
421 block_group->key.offset);
Yan Zheng11833d62009-09-11 16:11:19 -0400422 caching_ctl->progress = (u64)-1;
Josef Bacik817d52f2009-07-13 21:29:25 -0400423
424 spin_lock(&block_group->lock);
Yan Zheng11833d62009-09-11 16:11:19 -0400425 block_group->caching_ctl = NULL;
Josef Bacik817d52f2009-07-13 21:29:25 -0400426 block_group->cached = BTRFS_CACHE_FINISHED;
427 spin_unlock(&block_group->lock);
428
Chris Mason54aa1f42007-06-22 14:16:25 -0400429err:
Chris Masone37c9e62007-05-09 20:13:14 -0400430 btrfs_free_path(path);
Yan Zheng276e6802009-07-30 09:40:40 -0400431 up_read(&fs_info->extent_commit_sem);
Josef Bacik817d52f2009-07-13 21:29:25 -0400432
Yan Zheng11833d62009-09-11 16:11:19 -0400433 free_excluded_extents(extent_root, block_group);
434
435 mutex_unlock(&caching_ctl->mutex);
436 wake_up(&caching_ctl->wait);
437
438 put_caching_control(caching_ctl);
439 atomic_dec(&block_group->space_info->caching_threads);
Josef Bacik11dfe352009-11-13 20:12:59 +0000440 btrfs_put_block_group(block_group);
441
Josef Bacik817d52f2009-07-13 21:29:25 -0400442 return 0;
443}
444
Josef Bacik9d66e232010-08-25 16:54:15 -0400445static int cache_block_group(struct btrfs_block_group_cache *cache,
446 struct btrfs_trans_handle *trans,
Josef Bacikb8399de2010-12-08 09:15:11 -0500447 struct btrfs_root *root,
Josef Bacik9d66e232010-08-25 16:54:15 -0400448 int load_cache_only)
Josef Bacik817d52f2009-07-13 21:29:25 -0400449{
Yan Zheng11833d62009-09-11 16:11:19 -0400450 struct btrfs_fs_info *fs_info = cache->fs_info;
451 struct btrfs_caching_control *caching_ctl;
Josef Bacik817d52f2009-07-13 21:29:25 -0400452 struct task_struct *tsk;
453 int ret = 0;
454
Yan Zheng11833d62009-09-11 16:11:19 -0400455 smp_mb();
456 if (cache->cached != BTRFS_CACHE_NO)
457 return 0;
458
Josef Bacik9d66e232010-08-25 16:54:15 -0400459 /*
460 * We can't do the read from on-disk cache during a commit since we need
Josef Bacikb8399de2010-12-08 09:15:11 -0500461 * to have the normal tree locking. Also if we are currently trying to
462 * allocate blocks for the tree root we can't do the fast caching since
463 * we likely hold important locks.
Josef Bacik9d66e232010-08-25 16:54:15 -0400464 */
Li Dongyangf7039b12011-03-24 10:24:28 +0000465 if (trans && (!trans->transaction->in_commit) &&
Josef Bacikb8399de2010-12-08 09:15:11 -0500466 (root && root != root->fs_info->tree_root)) {
Josef Bacik9d66e232010-08-25 16:54:15 -0400467 spin_lock(&cache->lock);
468 if (cache->cached != BTRFS_CACHE_NO) {
469 spin_unlock(&cache->lock);
470 return 0;
471 }
472 cache->cached = BTRFS_CACHE_STARTED;
473 spin_unlock(&cache->lock);
474
475 ret = load_free_space_cache(fs_info, cache);
476
477 spin_lock(&cache->lock);
478 if (ret == 1) {
479 cache->cached = BTRFS_CACHE_FINISHED;
480 cache->last_byte_to_unpin = (u64)-1;
481 } else {
482 cache->cached = BTRFS_CACHE_NO;
483 }
484 spin_unlock(&cache->lock);
Josef Bacik3c148742011-02-02 15:53:47 +0000485 if (ret == 1) {
486 free_excluded_extents(fs_info->extent_root, cache);
Josef Bacik9d66e232010-08-25 16:54:15 -0400487 return 0;
Josef Bacik3c148742011-02-02 15:53:47 +0000488 }
Josef Bacik9d66e232010-08-25 16:54:15 -0400489 }
490
491 if (load_cache_only)
492 return 0;
493
Miao Xiefc0e4a32011-03-24 11:41:21 +0000494 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
Yan Zheng11833d62009-09-11 16:11:19 -0400495 BUG_ON(!caching_ctl);
496
497 INIT_LIST_HEAD(&caching_ctl->list);
498 mutex_init(&caching_ctl->mutex);
499 init_waitqueue_head(&caching_ctl->wait);
500 caching_ctl->block_group = cache;
501 caching_ctl->progress = cache->key.objectid;
502 /* one for caching kthread, one for caching block group list */
503 atomic_set(&caching_ctl->count, 2);
504
Josef Bacik817d52f2009-07-13 21:29:25 -0400505 spin_lock(&cache->lock);
506 if (cache->cached != BTRFS_CACHE_NO) {
507 spin_unlock(&cache->lock);
Yan Zheng11833d62009-09-11 16:11:19 -0400508 kfree(caching_ctl);
509 return 0;
Josef Bacik817d52f2009-07-13 21:29:25 -0400510 }
Yan Zheng11833d62009-09-11 16:11:19 -0400511 cache->caching_ctl = caching_ctl;
Josef Bacik817d52f2009-07-13 21:29:25 -0400512 cache->cached = BTRFS_CACHE_STARTED;
513 spin_unlock(&cache->lock);
514
Yan Zheng11833d62009-09-11 16:11:19 -0400515 down_write(&fs_info->extent_commit_sem);
516 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
517 up_write(&fs_info->extent_commit_sem);
518
519 atomic_inc(&cache->space_info->caching_threads);
Josef Bacik11dfe352009-11-13 20:12:59 +0000520 btrfs_get_block_group(cache);
Yan Zheng11833d62009-09-11 16:11:19 -0400521
Josef Bacik817d52f2009-07-13 21:29:25 -0400522 tsk = kthread_run(caching_kthread, cache, "btrfs-cache-%llu\n",
523 cache->key.objectid);
524 if (IS_ERR(tsk)) {
525 ret = PTR_ERR(tsk);
526 printk(KERN_ERR "error running thread %d\n", ret);
527 BUG();
528 }
529
Josef Bacikef8bbdf2008-09-23 13:14:11 -0400530 return ret;
Chris Masone37c9e62007-05-09 20:13:14 -0400531}
532
Josef Bacik0f9dd462008-09-23 13:14:11 -0400533/*
534 * return the block group that starts at or after bytenr
535 */
Chris Masond3977122009-01-05 21:25:51 -0500536static struct btrfs_block_group_cache *
537btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
Chris Mason0ef3e662008-05-24 14:04:53 -0400538{
Josef Bacik0f9dd462008-09-23 13:14:11 -0400539 struct btrfs_block_group_cache *cache;
Chris Mason0ef3e662008-05-24 14:04:53 -0400540
Josef Bacik0f9dd462008-09-23 13:14:11 -0400541 cache = block_group_cache_tree_search(info, bytenr, 0);
Chris Mason0ef3e662008-05-24 14:04:53 -0400542
Josef Bacik0f9dd462008-09-23 13:14:11 -0400543 return cache;
Chris Mason0ef3e662008-05-24 14:04:53 -0400544}
545
Josef Bacik0f9dd462008-09-23 13:14:11 -0400546/*
Sankar P9f556842009-05-14 13:52:22 -0400547 * return the block group that contains the given bytenr
Josef Bacik0f9dd462008-09-23 13:14:11 -0400548 */
Chris Masond3977122009-01-05 21:25:51 -0500549struct btrfs_block_group_cache *btrfs_lookup_block_group(
550 struct btrfs_fs_info *info,
551 u64 bytenr)
Chris Masonbe744172007-05-06 10:15:01 -0400552{
Josef Bacik0f9dd462008-09-23 13:14:11 -0400553 struct btrfs_block_group_cache *cache;
Chris Masonbe744172007-05-06 10:15:01 -0400554
Josef Bacik0f9dd462008-09-23 13:14:11 -0400555 cache = block_group_cache_tree_search(info, bytenr, 1);
Chris Mason96b51792007-10-15 16:15:19 -0400556
Josef Bacik0f9dd462008-09-23 13:14:11 -0400557 return cache;
Chris Masonbe744172007-05-06 10:15:01 -0400558}
Chris Mason0b86a832008-03-24 15:01:56 -0400559
Josef Bacik0f9dd462008-09-23 13:14:11 -0400560static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
561 u64 flags)
Chris Mason6324fbf2008-03-24 15:01:59 -0400562{
Josef Bacik0f9dd462008-09-23 13:14:11 -0400563 struct list_head *head = &info->space_info;
Josef Bacik0f9dd462008-09-23 13:14:11 -0400564 struct btrfs_space_info *found;
Chris Mason4184ea72009-03-10 12:39:20 -0400565
Yan, Zhengb742bb82010-05-16 10:46:24 -0400566 flags &= BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_SYSTEM |
567 BTRFS_BLOCK_GROUP_METADATA;
568
Chris Mason4184ea72009-03-10 12:39:20 -0400569 rcu_read_lock();
570 list_for_each_entry_rcu(found, head, list) {
Josef Bacik67377732010-09-16 16:19:09 -0400571 if (found->flags & flags) {
Chris Mason4184ea72009-03-10 12:39:20 -0400572 rcu_read_unlock();
Josef Bacik0f9dd462008-09-23 13:14:11 -0400573 return found;
Chris Mason4184ea72009-03-10 12:39:20 -0400574 }
Josef Bacik0f9dd462008-09-23 13:14:11 -0400575 }
Chris Mason4184ea72009-03-10 12:39:20 -0400576 rcu_read_unlock();
Josef Bacik0f9dd462008-09-23 13:14:11 -0400577 return NULL;
Chris Mason6324fbf2008-03-24 15:01:59 -0400578}
579
Chris Mason4184ea72009-03-10 12:39:20 -0400580/*
581 * after adding space to the filesystem, we need to clear the full flags
582 * on all the space infos.
583 */
584void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
585{
586 struct list_head *head = &info->space_info;
587 struct btrfs_space_info *found;
588
589 rcu_read_lock();
590 list_for_each_entry_rcu(found, head, list)
591 found->full = 0;
592 rcu_read_unlock();
593}
594
Josef Bacik80eb2342008-10-29 14:49:05 -0400595static u64 div_factor(u64 num, int factor)
596{
597 if (factor == 10)
598 return num;
599 num *= factor;
600 do_div(num, 10);
601 return num;
602}
603
Chris Masone5bc2452010-10-26 13:37:56 -0400604static u64 div_factor_fine(u64 num, int factor)
605{
606 if (factor == 100)
607 return num;
608 num *= factor;
609 do_div(num, 100);
610 return num;
611}
612
Yan Zhengd2fb3432008-12-11 16:30:39 -0500613u64 btrfs_find_block_group(struct btrfs_root *root,
614 u64 search_start, u64 search_hint, int owner)
Chris Masoncd1bc462007-04-27 10:08:34 -0400615{
Chris Mason96b51792007-10-15 16:15:19 -0400616 struct btrfs_block_group_cache *cache;
Chris Masoncd1bc462007-04-27 10:08:34 -0400617 u64 used;
Yan Zhengd2fb3432008-12-11 16:30:39 -0500618 u64 last = max(search_hint, search_start);
619 u64 group_start = 0;
Chris Mason31f3c992007-04-30 15:25:45 -0400620 int full_search = 0;
Yan Zhengd2fb3432008-12-11 16:30:39 -0500621 int factor = 9;
Chris Mason0ef3e662008-05-24 14:04:53 -0400622 int wrapped = 0;
Chris Mason31f3c992007-04-30 15:25:45 -0400623again:
Zheng Yane8569812008-09-26 10:05:48 -0400624 while (1) {
625 cache = btrfs_lookup_first_block_group(root->fs_info, last);
Josef Bacik0f9dd462008-09-23 13:14:11 -0400626 if (!cache)
Chris Masoncd1bc462007-04-27 10:08:34 -0400627 break;
Chris Mason96b51792007-10-15 16:15:19 -0400628
Chris Masonc286ac42008-07-22 23:06:41 -0400629 spin_lock(&cache->lock);
Chris Mason96b51792007-10-15 16:15:19 -0400630 last = cache->key.objectid + cache->key.offset;
631 used = btrfs_block_group_used(&cache->item);
632
Yan Zhengd2fb3432008-12-11 16:30:39 -0500633 if ((full_search || !cache->ro) &&
634 block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
Zheng Yane8569812008-09-26 10:05:48 -0400635 if (used + cache->pinned + cache->reserved <
Yan Zhengd2fb3432008-12-11 16:30:39 -0500636 div_factor(cache->key.offset, factor)) {
637 group_start = cache->key.objectid;
Chris Masonc286ac42008-07-22 23:06:41 -0400638 spin_unlock(&cache->lock);
Chris Masonfa9c0d72009-04-03 09:47:43 -0400639 btrfs_put_block_group(cache);
Chris Mason8790d502008-04-03 16:29:03 -0400640 goto found;
641 }
Chris Masoncd1bc462007-04-27 10:08:34 -0400642 }
Chris Masonc286ac42008-07-22 23:06:41 -0400643 spin_unlock(&cache->lock);
Chris Masonfa9c0d72009-04-03 09:47:43 -0400644 btrfs_put_block_group(cache);
Chris Masonde428b62007-05-18 13:28:27 -0400645 cond_resched();
Chris Masoncd1bc462007-04-27 10:08:34 -0400646 }
Chris Mason0ef3e662008-05-24 14:04:53 -0400647 if (!wrapped) {
648 last = search_start;
649 wrapped = 1;
650 goto again;
651 }
652 if (!full_search && factor < 10) {
Chris Masonbe744172007-05-06 10:15:01 -0400653 last = search_start;
Chris Mason31f3c992007-04-30 15:25:45 -0400654 full_search = 1;
Chris Mason0ef3e662008-05-24 14:04:53 -0400655 factor = 10;
Chris Mason31f3c992007-04-30 15:25:45 -0400656 goto again;
657 }
Chris Masonbe744172007-05-06 10:15:01 -0400658found:
Yan Zhengd2fb3432008-12-11 16:30:39 -0500659 return group_start;
Chris Mason925baed2008-06-25 16:01:30 -0400660}
Josef Bacik0f9dd462008-09-23 13:14:11 -0400661
Chris Masone02119d2008-09-05 16:13:11 -0400662/* simple helper to search for an existing extent at a given offset */
Zheng Yan31840ae2008-09-23 13:14:14 -0400663int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
Chris Masone02119d2008-09-05 16:13:11 -0400664{
665 int ret;
666 struct btrfs_key key;
Zheng Yan31840ae2008-09-23 13:14:14 -0400667 struct btrfs_path *path;
Chris Masone02119d2008-09-05 16:13:11 -0400668
Zheng Yan31840ae2008-09-23 13:14:14 -0400669 path = btrfs_alloc_path();
670 BUG_ON(!path);
Chris Masone02119d2008-09-05 16:13:11 -0400671 key.objectid = start;
672 key.offset = len;
673 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
674 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
675 0, 0);
Zheng Yan31840ae2008-09-23 13:14:14 -0400676 btrfs_free_path(path);
Chris Mason7bb86312007-12-11 09:25:06 -0500677 return ret;
678}
679
Chris Masond8d5f3e2007-12-11 12:42:00 -0500680/*
Yan, Zhenga22285a2010-05-16 10:48:46 -0400681 * helper function to lookup reference count and flags of extent.
682 *
683 * the head node for delayed ref is used to store the sum of all the
684 * reference count modifications queued up in the rbtree. the head
685 * node may also store the extent flags to set. This way you can check
686 * to see what the reference count and extent flags would be if all of
687 * the delayed refs are not processed.
688 */
689int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
690 struct btrfs_root *root, u64 bytenr,
691 u64 num_bytes, u64 *refs, u64 *flags)
692{
693 struct btrfs_delayed_ref_head *head;
694 struct btrfs_delayed_ref_root *delayed_refs;
695 struct btrfs_path *path;
696 struct btrfs_extent_item *ei;
697 struct extent_buffer *leaf;
698 struct btrfs_key key;
699 u32 item_size;
700 u64 num_refs;
701 u64 extent_flags;
702 int ret;
703
704 path = btrfs_alloc_path();
705 if (!path)
706 return -ENOMEM;
707
708 key.objectid = bytenr;
709 key.type = BTRFS_EXTENT_ITEM_KEY;
710 key.offset = num_bytes;
711 if (!trans) {
712 path->skip_locking = 1;
713 path->search_commit_root = 1;
714 }
715again:
716 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
717 &key, path, 0, 0);
718 if (ret < 0)
719 goto out_free;
720
721 if (ret == 0) {
722 leaf = path->nodes[0];
723 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
724 if (item_size >= sizeof(*ei)) {
725 ei = btrfs_item_ptr(leaf, path->slots[0],
726 struct btrfs_extent_item);
727 num_refs = btrfs_extent_refs(leaf, ei);
728 extent_flags = btrfs_extent_flags(leaf, ei);
729 } else {
730#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
731 struct btrfs_extent_item_v0 *ei0;
732 BUG_ON(item_size != sizeof(*ei0));
733 ei0 = btrfs_item_ptr(leaf, path->slots[0],
734 struct btrfs_extent_item_v0);
735 num_refs = btrfs_extent_refs_v0(leaf, ei0);
736 /* FIXME: this isn't correct for data */
737 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
738#else
739 BUG();
740#endif
741 }
742 BUG_ON(num_refs == 0);
743 } else {
744 num_refs = 0;
745 extent_flags = 0;
746 ret = 0;
747 }
748
749 if (!trans)
750 goto out;
751
752 delayed_refs = &trans->transaction->delayed_refs;
753 spin_lock(&delayed_refs->lock);
754 head = btrfs_find_delayed_ref_head(trans, bytenr);
755 if (head) {
756 if (!mutex_trylock(&head->mutex)) {
757 atomic_inc(&head->node.refs);
758 spin_unlock(&delayed_refs->lock);
759
David Sterbab3b4aa72011-04-21 01:20:15 +0200760 btrfs_release_path(path);
Yan, Zhenga22285a2010-05-16 10:48:46 -0400761
David Sterba8cc33e52011-05-02 15:29:25 +0200762 /*
763 * Mutex was contended, block until it's released and try
764 * again
765 */
Yan, Zhenga22285a2010-05-16 10:48:46 -0400766 mutex_lock(&head->mutex);
767 mutex_unlock(&head->mutex);
768 btrfs_put_delayed_ref(&head->node);
769 goto again;
770 }
771 if (head->extent_op && head->extent_op->update_flags)
772 extent_flags |= head->extent_op->flags_to_set;
773 else
774 BUG_ON(num_refs == 0);
775
776 num_refs += head->node.ref_mod;
777 mutex_unlock(&head->mutex);
778 }
779 spin_unlock(&delayed_refs->lock);
780out:
781 WARN_ON(num_refs == 0);
782 if (refs)
783 *refs = num_refs;
784 if (flags)
785 *flags = extent_flags;
786out_free:
787 btrfs_free_path(path);
788 return ret;
789}
790
791/*
Chris Masond8d5f3e2007-12-11 12:42:00 -0500792 * Back reference rules. Back refs have three main goals:
793 *
794 * 1) differentiate between all holders of references to an extent so that
795 * when a reference is dropped we can make sure it was a valid reference
796 * before freeing the extent.
797 *
798 * 2) Provide enough information to quickly find the holders of an extent
799 * if we notice a given block is corrupted or bad.
800 *
801 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
802 * maintenance. This is actually the same as #2, but with a slightly
803 * different use case.
804 *
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400805 * There are two kinds of back refs. The implicit back refs is optimized
806 * for pointers in non-shared tree blocks. For a given pointer in a block,
807 * back refs of this kind provide information about the block's owner tree
808 * and the pointer's key. These information allow us to find the block by
809 * b-tree searching. The full back refs is for pointers in tree blocks not
810 * referenced by their owner trees. The location of tree block is recorded
811 * in the back refs. Actually the full back refs is generic, and can be
812 * used in all cases the implicit back refs is used. The major shortcoming
813 * of the full back refs is its overhead. Every time a tree block gets
814 * COWed, we have to update back refs entry for all pointers in it.
815 *
816 * For a newly allocated tree block, we use implicit back refs for
817 * pointers in it. This means most tree related operations only involve
818 * implicit back refs. For a tree block created in old transaction, the
819 * only way to drop a reference to it is COW it. So we can detect the
820 * event that tree block loses its owner tree's reference and do the
821 * back refs conversion.
822 *
823 * When a tree block is COW'd through a tree, there are four cases:
824 *
825 * The reference count of the block is one and the tree is the block's
826 * owner tree. Nothing to do in this case.
827 *
828 * The reference count of the block is one and the tree is not the
829 * block's owner tree. In this case, full back refs is used for pointers
830 * in the block. Remove these full back refs, add implicit back refs for
831 * every pointers in the new block.
832 *
833 * The reference count of the block is greater than one and the tree is
834 * the block's owner tree. In this case, implicit back refs is used for
835 * pointers in the block. Add full back refs for every pointers in the
836 * block, increase lower level extents' reference counts. The original
837 * implicit back refs are entailed to the new block.
838 *
839 * The reference count of the block is greater than one and the tree is
840 * not the block's owner tree. Add implicit back refs for every pointer in
841 * the new block, increase lower level extents' reference count.
842 *
843 * Back Reference Key composing:
844 *
845 * The key objectid corresponds to the first byte in the extent,
846 * The key type is used to differentiate between types of back refs.
847 * There are different meanings of the key offset for different types
848 * of back refs.
849 *
Chris Masond8d5f3e2007-12-11 12:42:00 -0500850 * File extents can be referenced by:
851 *
852 * - multiple snapshots, subvolumes, or different generations in one subvol
Zheng Yan31840ae2008-09-23 13:14:14 -0400853 * - different files inside a single subvolume
Chris Masond8d5f3e2007-12-11 12:42:00 -0500854 * - different offsets inside a file (bookend extents in file.c)
855 *
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400856 * The extent ref structure for the implicit back refs has fields for:
Chris Masond8d5f3e2007-12-11 12:42:00 -0500857 *
858 * - Objectid of the subvolume root
Chris Masond8d5f3e2007-12-11 12:42:00 -0500859 * - objectid of the file holding the reference
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400860 * - original offset in the file
861 * - how many bookend extents
Zheng Yan31840ae2008-09-23 13:14:14 -0400862 *
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400863 * The key offset for the implicit back refs is hash of the first
864 * three fields.
Chris Masond8d5f3e2007-12-11 12:42:00 -0500865 *
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400866 * The extent ref structure for the full back refs has field for:
Chris Masond8d5f3e2007-12-11 12:42:00 -0500867 *
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400868 * - number of pointers in the tree leaf
Chris Masond8d5f3e2007-12-11 12:42:00 -0500869 *
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400870 * The key offset for the implicit back refs is the first byte of
871 * the tree leaf
Chris Masond8d5f3e2007-12-11 12:42:00 -0500872 *
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400873 * When a file extent is allocated, The implicit back refs is used.
874 * the fields are filled in:
Chris Masond8d5f3e2007-12-11 12:42:00 -0500875 *
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400876 * (root_key.objectid, inode objectid, offset in file, 1)
877 *
878 * When a file extent is removed file truncation, we find the
879 * corresponding implicit back refs and check the following fields:
880 *
881 * (btrfs_header_owner(leaf), inode objectid, offset in file)
Chris Masond8d5f3e2007-12-11 12:42:00 -0500882 *
883 * Btree extents can be referenced by:
884 *
885 * - Different subvolumes
Chris Masond8d5f3e2007-12-11 12:42:00 -0500886 *
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400887 * Both the implicit back refs and the full back refs for tree blocks
888 * only consist of key. The key offset for the implicit back refs is
889 * objectid of block's owner tree. The key offset for the full back refs
890 * is the first byte of parent block.
Chris Masond8d5f3e2007-12-11 12:42:00 -0500891 *
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400892 * When implicit back refs is used, information about the lowest key and
893 * level of the tree block are required. These information are stored in
894 * tree block info structure.
Chris Masond8d5f3e2007-12-11 12:42:00 -0500895 */
Zheng Yan31840ae2008-09-23 13:14:14 -0400896
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400897#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
898static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
899 struct btrfs_root *root,
900 struct btrfs_path *path,
901 u64 owner, u32 extra_size)
Chris Mason74493f72007-12-11 09:25:06 -0500902{
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400903 struct btrfs_extent_item *item;
904 struct btrfs_extent_item_v0 *ei0;
905 struct btrfs_extent_ref_v0 *ref0;
906 struct btrfs_tree_block_info *bi;
Zheng Yan31840ae2008-09-23 13:14:14 -0400907 struct extent_buffer *leaf;
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400908 struct btrfs_key key;
909 struct btrfs_key found_key;
910 u32 new_size = sizeof(*item);
911 u64 refs;
Chris Mason74493f72007-12-11 09:25:06 -0500912 int ret;
913
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400914 leaf = path->nodes[0];
915 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
Chris Mason74493f72007-12-11 09:25:06 -0500916
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400917 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
918 ei0 = btrfs_item_ptr(leaf, path->slots[0],
919 struct btrfs_extent_item_v0);
920 refs = btrfs_extent_refs_v0(leaf, ei0);
921
922 if (owner == (u64)-1) {
923 while (1) {
924 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
925 ret = btrfs_next_leaf(root, path);
926 if (ret < 0)
927 return ret;
928 BUG_ON(ret > 0);
929 leaf = path->nodes[0];
930 }
931 btrfs_item_key_to_cpu(leaf, &found_key,
932 path->slots[0]);
933 BUG_ON(key.objectid != found_key.objectid);
934 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
935 path->slots[0]++;
936 continue;
937 }
938 ref0 = btrfs_item_ptr(leaf, path->slots[0],
939 struct btrfs_extent_ref_v0);
940 owner = btrfs_ref_objectid_v0(leaf, ref0);
941 break;
942 }
943 }
David Sterbab3b4aa72011-04-21 01:20:15 +0200944 btrfs_release_path(path);
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400945
946 if (owner < BTRFS_FIRST_FREE_OBJECTID)
947 new_size += sizeof(*bi);
948
949 new_size -= sizeof(*ei0);
950 ret = btrfs_search_slot(trans, root, &key, path,
951 new_size + extra_size, 1);
Zheng Yan31840ae2008-09-23 13:14:14 -0400952 if (ret < 0)
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400953 return ret;
954 BUG_ON(ret);
955
956 ret = btrfs_extend_item(trans, root, path, new_size);
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400957
958 leaf = path->nodes[0];
959 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
960 btrfs_set_extent_refs(leaf, item, refs);
961 /* FIXME: get real generation */
962 btrfs_set_extent_generation(leaf, item, 0);
963 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
964 btrfs_set_extent_flags(leaf, item,
965 BTRFS_EXTENT_FLAG_TREE_BLOCK |
966 BTRFS_BLOCK_FLAG_FULL_BACKREF);
967 bi = (struct btrfs_tree_block_info *)(item + 1);
968 /* FIXME: get first key of the block */
969 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
970 btrfs_set_tree_block_level(leaf, bi, (int)owner);
971 } else {
972 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
973 }
974 btrfs_mark_buffer_dirty(leaf);
975 return 0;
976}
977#endif
978
979static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
980{
981 u32 high_crc = ~(u32)0;
982 u32 low_crc = ~(u32)0;
983 __le64 lenum;
984
985 lenum = cpu_to_le64(root_objectid);
David Woodhouse163e7832009-04-19 13:02:41 +0100986 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400987 lenum = cpu_to_le64(owner);
David Woodhouse163e7832009-04-19 13:02:41 +0100988 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400989 lenum = cpu_to_le64(offset);
David Woodhouse163e7832009-04-19 13:02:41 +0100990 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400991
992 return ((u64)high_crc << 31) ^ (u64)low_crc;
993}
994
995static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
996 struct btrfs_extent_data_ref *ref)
997{
998 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
999 btrfs_extent_data_ref_objectid(leaf, ref),
1000 btrfs_extent_data_ref_offset(leaf, ref));
1001}
1002
1003static int match_extent_data_ref(struct extent_buffer *leaf,
1004 struct btrfs_extent_data_ref *ref,
1005 u64 root_objectid, u64 owner, u64 offset)
1006{
1007 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1008 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1009 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1010 return 0;
1011 return 1;
1012}
1013
1014static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1015 struct btrfs_root *root,
1016 struct btrfs_path *path,
1017 u64 bytenr, u64 parent,
1018 u64 root_objectid,
1019 u64 owner, u64 offset)
1020{
1021 struct btrfs_key key;
1022 struct btrfs_extent_data_ref *ref;
1023 struct extent_buffer *leaf;
1024 u32 nritems;
1025 int ret;
1026 int recow;
1027 int err = -ENOENT;
1028
1029 key.objectid = bytenr;
1030 if (parent) {
1031 key.type = BTRFS_SHARED_DATA_REF_KEY;
1032 key.offset = parent;
1033 } else {
1034 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1035 key.offset = hash_extent_data_ref(root_objectid,
1036 owner, offset);
1037 }
1038again:
1039 recow = 0;
1040 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1041 if (ret < 0) {
1042 err = ret;
1043 goto fail;
1044 }
1045
1046 if (parent) {
1047 if (!ret)
1048 return 0;
1049#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1050 key.type = BTRFS_EXTENT_REF_V0_KEY;
David Sterbab3b4aa72011-04-21 01:20:15 +02001051 btrfs_release_path(path);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001052 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1053 if (ret < 0) {
1054 err = ret;
1055 goto fail;
1056 }
1057 if (!ret)
1058 return 0;
1059#endif
1060 goto fail;
Zheng Yan31840ae2008-09-23 13:14:14 -04001061 }
1062
1063 leaf = path->nodes[0];
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001064 nritems = btrfs_header_nritems(leaf);
1065 while (1) {
1066 if (path->slots[0] >= nritems) {
1067 ret = btrfs_next_leaf(root, path);
1068 if (ret < 0)
1069 err = ret;
1070 if (ret)
1071 goto fail;
1072
1073 leaf = path->nodes[0];
1074 nritems = btrfs_header_nritems(leaf);
1075 recow = 1;
1076 }
1077
1078 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1079 if (key.objectid != bytenr ||
1080 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1081 goto fail;
1082
1083 ref = btrfs_item_ptr(leaf, path->slots[0],
1084 struct btrfs_extent_data_ref);
1085
1086 if (match_extent_data_ref(leaf, ref, root_objectid,
1087 owner, offset)) {
1088 if (recow) {
David Sterbab3b4aa72011-04-21 01:20:15 +02001089 btrfs_release_path(path);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001090 goto again;
1091 }
1092 err = 0;
1093 break;
1094 }
1095 path->slots[0]++;
Zheng Yan31840ae2008-09-23 13:14:14 -04001096 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001097fail:
1098 return err;
Zheng Yan31840ae2008-09-23 13:14:14 -04001099}
1100
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001101static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1102 struct btrfs_root *root,
1103 struct btrfs_path *path,
1104 u64 bytenr, u64 parent,
1105 u64 root_objectid, u64 owner,
1106 u64 offset, int refs_to_add)
Zheng Yan31840ae2008-09-23 13:14:14 -04001107{
1108 struct btrfs_key key;
1109 struct extent_buffer *leaf;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001110 u32 size;
Zheng Yan31840ae2008-09-23 13:14:14 -04001111 u32 num_refs;
1112 int ret;
1113
1114 key.objectid = bytenr;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001115 if (parent) {
1116 key.type = BTRFS_SHARED_DATA_REF_KEY;
1117 key.offset = parent;
1118 size = sizeof(struct btrfs_shared_data_ref);
Zheng Yan31840ae2008-09-23 13:14:14 -04001119 } else {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001120 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1121 key.offset = hash_extent_data_ref(root_objectid,
1122 owner, offset);
1123 size = sizeof(struct btrfs_extent_data_ref);
Zheng Yan31840ae2008-09-23 13:14:14 -04001124 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001125
1126 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1127 if (ret && ret != -EEXIST)
1128 goto fail;
1129
1130 leaf = path->nodes[0];
1131 if (parent) {
1132 struct btrfs_shared_data_ref *ref;
1133 ref = btrfs_item_ptr(leaf, path->slots[0],
1134 struct btrfs_shared_data_ref);
1135 if (ret == 0) {
1136 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1137 } else {
1138 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1139 num_refs += refs_to_add;
1140 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1141 }
1142 } else {
1143 struct btrfs_extent_data_ref *ref;
1144 while (ret == -EEXIST) {
1145 ref = btrfs_item_ptr(leaf, path->slots[0],
1146 struct btrfs_extent_data_ref);
1147 if (match_extent_data_ref(leaf, ref, root_objectid,
1148 owner, offset))
1149 break;
David Sterbab3b4aa72011-04-21 01:20:15 +02001150 btrfs_release_path(path);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001151 key.offset++;
1152 ret = btrfs_insert_empty_item(trans, root, path, &key,
1153 size);
1154 if (ret && ret != -EEXIST)
1155 goto fail;
1156
1157 leaf = path->nodes[0];
1158 }
1159 ref = btrfs_item_ptr(leaf, path->slots[0],
1160 struct btrfs_extent_data_ref);
1161 if (ret == 0) {
1162 btrfs_set_extent_data_ref_root(leaf, ref,
1163 root_objectid);
1164 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1165 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1166 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1167 } else {
1168 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1169 num_refs += refs_to_add;
1170 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1171 }
1172 }
1173 btrfs_mark_buffer_dirty(leaf);
1174 ret = 0;
1175fail:
David Sterbab3b4aa72011-04-21 01:20:15 +02001176 btrfs_release_path(path);
Chris Mason7bb86312007-12-11 09:25:06 -05001177 return ret;
Chris Mason74493f72007-12-11 09:25:06 -05001178}
1179
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001180static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1181 struct btrfs_root *root,
1182 struct btrfs_path *path,
1183 int refs_to_drop)
Zheng Yan31840ae2008-09-23 13:14:14 -04001184{
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001185 struct btrfs_key key;
1186 struct btrfs_extent_data_ref *ref1 = NULL;
1187 struct btrfs_shared_data_ref *ref2 = NULL;
Zheng Yan31840ae2008-09-23 13:14:14 -04001188 struct extent_buffer *leaf;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001189 u32 num_refs = 0;
Zheng Yan31840ae2008-09-23 13:14:14 -04001190 int ret = 0;
1191
1192 leaf = path->nodes[0];
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001193 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1194
1195 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1196 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1197 struct btrfs_extent_data_ref);
1198 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1199 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1200 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1201 struct btrfs_shared_data_ref);
1202 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1203#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1204 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1205 struct btrfs_extent_ref_v0 *ref0;
1206 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1207 struct btrfs_extent_ref_v0);
1208 num_refs = btrfs_ref_count_v0(leaf, ref0);
1209#endif
1210 } else {
1211 BUG();
1212 }
1213
Chris Mason56bec292009-03-13 10:10:06 -04001214 BUG_ON(num_refs < refs_to_drop);
1215 num_refs -= refs_to_drop;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001216
Zheng Yan31840ae2008-09-23 13:14:14 -04001217 if (num_refs == 0) {
1218 ret = btrfs_del_item(trans, root, path);
1219 } else {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001220 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1221 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1222 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1223 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1224#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1225 else {
1226 struct btrfs_extent_ref_v0 *ref0;
1227 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1228 struct btrfs_extent_ref_v0);
1229 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1230 }
1231#endif
Zheng Yan31840ae2008-09-23 13:14:14 -04001232 btrfs_mark_buffer_dirty(leaf);
1233 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001234 return ret;
1235}
1236
1237static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1238 struct btrfs_path *path,
1239 struct btrfs_extent_inline_ref *iref)
1240{
1241 struct btrfs_key key;
1242 struct extent_buffer *leaf;
1243 struct btrfs_extent_data_ref *ref1;
1244 struct btrfs_shared_data_ref *ref2;
1245 u32 num_refs = 0;
1246
1247 leaf = path->nodes[0];
1248 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1249 if (iref) {
1250 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1251 BTRFS_EXTENT_DATA_REF_KEY) {
1252 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1253 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1254 } else {
1255 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1256 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1257 }
1258 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1259 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1260 struct btrfs_extent_data_ref);
1261 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1262 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1263 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1264 struct btrfs_shared_data_ref);
1265 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1266#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1267 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1268 struct btrfs_extent_ref_v0 *ref0;
1269 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1270 struct btrfs_extent_ref_v0);
1271 num_refs = btrfs_ref_count_v0(leaf, ref0);
1272#endif
1273 } else {
1274 WARN_ON(1);
1275 }
1276 return num_refs;
1277}
1278
1279static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1280 struct btrfs_root *root,
1281 struct btrfs_path *path,
1282 u64 bytenr, u64 parent,
1283 u64 root_objectid)
1284{
1285 struct btrfs_key key;
1286 int ret;
1287
1288 key.objectid = bytenr;
1289 if (parent) {
1290 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1291 key.offset = parent;
1292 } else {
1293 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1294 key.offset = root_objectid;
1295 }
1296
1297 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1298 if (ret > 0)
1299 ret = -ENOENT;
1300#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1301 if (ret == -ENOENT && parent) {
David Sterbab3b4aa72011-04-21 01:20:15 +02001302 btrfs_release_path(path);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001303 key.type = BTRFS_EXTENT_REF_V0_KEY;
1304 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1305 if (ret > 0)
1306 ret = -ENOENT;
1307 }
1308#endif
1309 return ret;
1310}
1311
1312static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1313 struct btrfs_root *root,
1314 struct btrfs_path *path,
1315 u64 bytenr, u64 parent,
1316 u64 root_objectid)
1317{
1318 struct btrfs_key key;
1319 int ret;
1320
1321 key.objectid = bytenr;
1322 if (parent) {
1323 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1324 key.offset = parent;
1325 } else {
1326 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1327 key.offset = root_objectid;
1328 }
1329
1330 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
David Sterbab3b4aa72011-04-21 01:20:15 +02001331 btrfs_release_path(path);
Zheng Yan31840ae2008-09-23 13:14:14 -04001332 return ret;
1333}
1334
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001335static inline int extent_ref_type(u64 parent, u64 owner)
1336{
1337 int type;
1338 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1339 if (parent > 0)
1340 type = BTRFS_SHARED_BLOCK_REF_KEY;
1341 else
1342 type = BTRFS_TREE_BLOCK_REF_KEY;
1343 } else {
1344 if (parent > 0)
1345 type = BTRFS_SHARED_DATA_REF_KEY;
1346 else
1347 type = BTRFS_EXTENT_DATA_REF_KEY;
1348 }
1349 return type;
1350}
1351
Yan Zheng2c47e6052009-06-27 21:07:35 -04001352static int find_next_key(struct btrfs_path *path, int level,
1353 struct btrfs_key *key)
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001354
1355{
Yan Zheng2c47e6052009-06-27 21:07:35 -04001356 for (; level < BTRFS_MAX_LEVEL; level++) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001357 if (!path->nodes[level])
1358 break;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001359 if (path->slots[level] + 1 >=
1360 btrfs_header_nritems(path->nodes[level]))
1361 continue;
1362 if (level == 0)
1363 btrfs_item_key_to_cpu(path->nodes[level], key,
1364 path->slots[level] + 1);
1365 else
1366 btrfs_node_key_to_cpu(path->nodes[level], key,
1367 path->slots[level] + 1);
1368 return 0;
1369 }
1370 return 1;
1371}
1372
1373/*
1374 * look for inline back ref. if back ref is found, *ref_ret is set
1375 * to the address of inline back ref, and 0 is returned.
1376 *
1377 * if back ref isn't found, *ref_ret is set to the address where it
1378 * should be inserted, and -ENOENT is returned.
1379 *
1380 * if insert is true and there are too many inline back refs, the path
1381 * points to the extent item, and -EAGAIN is returned.
1382 *
1383 * NOTE: inline back refs are ordered in the same way that back ref
1384 * items in the tree are ordered.
1385 */
1386static noinline_for_stack
1387int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1388 struct btrfs_root *root,
1389 struct btrfs_path *path,
1390 struct btrfs_extent_inline_ref **ref_ret,
1391 u64 bytenr, u64 num_bytes,
1392 u64 parent, u64 root_objectid,
1393 u64 owner, u64 offset, int insert)
1394{
1395 struct btrfs_key key;
1396 struct extent_buffer *leaf;
1397 struct btrfs_extent_item *ei;
1398 struct btrfs_extent_inline_ref *iref;
1399 u64 flags;
1400 u64 item_size;
1401 unsigned long ptr;
1402 unsigned long end;
1403 int extra_size;
1404 int type;
1405 int want;
1406 int ret;
1407 int err = 0;
1408
1409 key.objectid = bytenr;
1410 key.type = BTRFS_EXTENT_ITEM_KEY;
1411 key.offset = num_bytes;
1412
1413 want = extent_ref_type(parent, owner);
1414 if (insert) {
1415 extra_size = btrfs_extent_inline_ref_size(want);
Yan Zheng85d41982009-06-11 08:51:10 -04001416 path->keep_locks = 1;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001417 } else
1418 extra_size = -1;
1419 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1420 if (ret < 0) {
1421 err = ret;
1422 goto out;
1423 }
1424 BUG_ON(ret);
1425
1426 leaf = path->nodes[0];
1427 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1428#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1429 if (item_size < sizeof(*ei)) {
1430 if (!insert) {
1431 err = -ENOENT;
1432 goto out;
1433 }
1434 ret = convert_extent_item_v0(trans, root, path, owner,
1435 extra_size);
1436 if (ret < 0) {
1437 err = ret;
1438 goto out;
1439 }
1440 leaf = path->nodes[0];
1441 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1442 }
1443#endif
1444 BUG_ON(item_size < sizeof(*ei));
1445
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001446 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1447 flags = btrfs_extent_flags(leaf, ei);
1448
1449 ptr = (unsigned long)(ei + 1);
1450 end = (unsigned long)ei + item_size;
1451
1452 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1453 ptr += sizeof(struct btrfs_tree_block_info);
1454 BUG_ON(ptr > end);
1455 } else {
1456 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1457 }
1458
1459 err = -ENOENT;
1460 while (1) {
1461 if (ptr >= end) {
1462 WARN_ON(ptr > end);
1463 break;
1464 }
1465 iref = (struct btrfs_extent_inline_ref *)ptr;
1466 type = btrfs_extent_inline_ref_type(leaf, iref);
1467 if (want < type)
1468 break;
1469 if (want > type) {
1470 ptr += btrfs_extent_inline_ref_size(type);
1471 continue;
1472 }
1473
1474 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1475 struct btrfs_extent_data_ref *dref;
1476 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1477 if (match_extent_data_ref(leaf, dref, root_objectid,
1478 owner, offset)) {
1479 err = 0;
1480 break;
1481 }
1482 if (hash_extent_data_ref_item(leaf, dref) <
1483 hash_extent_data_ref(root_objectid, owner, offset))
1484 break;
1485 } else {
1486 u64 ref_offset;
1487 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1488 if (parent > 0) {
1489 if (parent == ref_offset) {
1490 err = 0;
1491 break;
1492 }
1493 if (ref_offset < parent)
1494 break;
1495 } else {
1496 if (root_objectid == ref_offset) {
1497 err = 0;
1498 break;
1499 }
1500 if (ref_offset < root_objectid)
1501 break;
1502 }
1503 }
1504 ptr += btrfs_extent_inline_ref_size(type);
1505 }
1506 if (err == -ENOENT && insert) {
1507 if (item_size + extra_size >=
1508 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1509 err = -EAGAIN;
1510 goto out;
1511 }
1512 /*
1513 * To add new inline back ref, we have to make sure
1514 * there is no corresponding back ref item.
1515 * For simplicity, we just do not add new inline back
1516 * ref if there is any kind of item for this block
1517 */
Yan Zheng2c47e6052009-06-27 21:07:35 -04001518 if (find_next_key(path, 0, &key) == 0 &&
1519 key.objectid == bytenr &&
Yan Zheng85d41982009-06-11 08:51:10 -04001520 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001521 err = -EAGAIN;
1522 goto out;
1523 }
1524 }
1525 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1526out:
Yan Zheng85d41982009-06-11 08:51:10 -04001527 if (insert) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001528 path->keep_locks = 0;
1529 btrfs_unlock_up_safe(path, 1);
1530 }
1531 return err;
1532}
1533
1534/*
1535 * helper to add new inline back ref
1536 */
1537static noinline_for_stack
1538int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1539 struct btrfs_root *root,
1540 struct btrfs_path *path,
1541 struct btrfs_extent_inline_ref *iref,
1542 u64 parent, u64 root_objectid,
1543 u64 owner, u64 offset, int refs_to_add,
1544 struct btrfs_delayed_extent_op *extent_op)
1545{
1546 struct extent_buffer *leaf;
1547 struct btrfs_extent_item *ei;
1548 unsigned long ptr;
1549 unsigned long end;
1550 unsigned long item_offset;
1551 u64 refs;
1552 int size;
1553 int type;
1554 int ret;
1555
1556 leaf = path->nodes[0];
1557 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1558 item_offset = (unsigned long)iref - (unsigned long)ei;
1559
1560 type = extent_ref_type(parent, owner);
1561 size = btrfs_extent_inline_ref_size(type);
1562
1563 ret = btrfs_extend_item(trans, root, path, size);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001564
1565 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1566 refs = btrfs_extent_refs(leaf, ei);
1567 refs += refs_to_add;
1568 btrfs_set_extent_refs(leaf, ei, refs);
1569 if (extent_op)
1570 __run_delayed_extent_op(extent_op, leaf, ei);
1571
1572 ptr = (unsigned long)ei + item_offset;
1573 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1574 if (ptr < end - size)
1575 memmove_extent_buffer(leaf, ptr + size, ptr,
1576 end - size - ptr);
1577
1578 iref = (struct btrfs_extent_inline_ref *)ptr;
1579 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1580 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1581 struct btrfs_extent_data_ref *dref;
1582 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1583 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1584 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1585 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1586 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1587 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1588 struct btrfs_shared_data_ref *sref;
1589 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1590 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1591 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1592 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1593 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1594 } else {
1595 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1596 }
1597 btrfs_mark_buffer_dirty(leaf);
1598 return 0;
1599}
1600
1601static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1602 struct btrfs_root *root,
1603 struct btrfs_path *path,
1604 struct btrfs_extent_inline_ref **ref_ret,
1605 u64 bytenr, u64 num_bytes, u64 parent,
1606 u64 root_objectid, u64 owner, u64 offset)
1607{
1608 int ret;
1609
1610 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1611 bytenr, num_bytes, parent,
1612 root_objectid, owner, offset, 0);
1613 if (ret != -ENOENT)
1614 return ret;
1615
David Sterbab3b4aa72011-04-21 01:20:15 +02001616 btrfs_release_path(path);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001617 *ref_ret = NULL;
1618
1619 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1620 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1621 root_objectid);
1622 } else {
1623 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1624 root_objectid, owner, offset);
1625 }
1626 return ret;
1627}
1628
1629/*
1630 * helper to update/remove inline back ref
1631 */
1632static noinline_for_stack
1633int update_inline_extent_backref(struct btrfs_trans_handle *trans,
1634 struct btrfs_root *root,
1635 struct btrfs_path *path,
1636 struct btrfs_extent_inline_ref *iref,
1637 int refs_to_mod,
1638 struct btrfs_delayed_extent_op *extent_op)
1639{
1640 struct extent_buffer *leaf;
1641 struct btrfs_extent_item *ei;
1642 struct btrfs_extent_data_ref *dref = NULL;
1643 struct btrfs_shared_data_ref *sref = NULL;
1644 unsigned long ptr;
1645 unsigned long end;
1646 u32 item_size;
1647 int size;
1648 int type;
1649 int ret;
1650 u64 refs;
1651
1652 leaf = path->nodes[0];
1653 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1654 refs = btrfs_extent_refs(leaf, ei);
1655 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1656 refs += refs_to_mod;
1657 btrfs_set_extent_refs(leaf, ei, refs);
1658 if (extent_op)
1659 __run_delayed_extent_op(extent_op, leaf, ei);
1660
1661 type = btrfs_extent_inline_ref_type(leaf, iref);
1662
1663 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1664 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1665 refs = btrfs_extent_data_ref_count(leaf, dref);
1666 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1667 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1668 refs = btrfs_shared_data_ref_count(leaf, sref);
1669 } else {
1670 refs = 1;
1671 BUG_ON(refs_to_mod != -1);
1672 }
1673
1674 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1675 refs += refs_to_mod;
1676
1677 if (refs > 0) {
1678 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1679 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1680 else
1681 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1682 } else {
1683 size = btrfs_extent_inline_ref_size(type);
1684 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1685 ptr = (unsigned long)iref;
1686 end = (unsigned long)ei + item_size;
1687 if (ptr + size < end)
1688 memmove_extent_buffer(leaf, ptr, ptr + size,
1689 end - ptr - size);
1690 item_size -= size;
1691 ret = btrfs_truncate_item(trans, root, path, item_size, 1);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001692 }
1693 btrfs_mark_buffer_dirty(leaf);
1694 return 0;
1695}
1696
1697static noinline_for_stack
1698int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1699 struct btrfs_root *root,
1700 struct btrfs_path *path,
1701 u64 bytenr, u64 num_bytes, u64 parent,
1702 u64 root_objectid, u64 owner,
1703 u64 offset, int refs_to_add,
1704 struct btrfs_delayed_extent_op *extent_op)
1705{
1706 struct btrfs_extent_inline_ref *iref;
1707 int ret;
1708
1709 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1710 bytenr, num_bytes, parent,
1711 root_objectid, owner, offset, 1);
1712 if (ret == 0) {
1713 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1714 ret = update_inline_extent_backref(trans, root, path, iref,
1715 refs_to_add, extent_op);
1716 } else if (ret == -ENOENT) {
1717 ret = setup_inline_extent_backref(trans, root, path, iref,
1718 parent, root_objectid,
1719 owner, offset, refs_to_add,
1720 extent_op);
1721 }
1722 return ret;
1723}
1724
1725static int insert_extent_backref(struct btrfs_trans_handle *trans,
1726 struct btrfs_root *root,
1727 struct btrfs_path *path,
1728 u64 bytenr, u64 parent, u64 root_objectid,
1729 u64 owner, u64 offset, int refs_to_add)
1730{
1731 int ret;
1732 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1733 BUG_ON(refs_to_add != 1);
1734 ret = insert_tree_block_ref(trans, root, path, bytenr,
1735 parent, root_objectid);
1736 } else {
1737 ret = insert_extent_data_ref(trans, root, path, bytenr,
1738 parent, root_objectid,
1739 owner, offset, refs_to_add);
1740 }
1741 return ret;
1742}
1743
1744static int remove_extent_backref(struct btrfs_trans_handle *trans,
1745 struct btrfs_root *root,
1746 struct btrfs_path *path,
1747 struct btrfs_extent_inline_ref *iref,
1748 int refs_to_drop, int is_data)
1749{
1750 int ret;
1751
1752 BUG_ON(!is_data && refs_to_drop != 1);
1753 if (iref) {
1754 ret = update_inline_extent_backref(trans, root, path, iref,
1755 -refs_to_drop, NULL);
1756 } else if (is_data) {
1757 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1758 } else {
1759 ret = btrfs_del_item(trans, root, path);
1760 }
1761 return ret;
1762}
1763
Li Dongyang5378e602011-03-24 10:24:27 +00001764static int btrfs_issue_discard(struct block_device *bdev,
Chris Mason15916de2008-11-19 21:17:22 -05001765 u64 start, u64 len)
1766{
Li Dongyang5378e602011-03-24 10:24:27 +00001767 return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
Chris Mason15916de2008-11-19 21:17:22 -05001768}
Chris Mason15916de2008-11-19 21:17:22 -05001769
Liu Hui1f3c79a2009-01-05 15:57:51 -05001770static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
Li Dongyang5378e602011-03-24 10:24:27 +00001771 u64 num_bytes, u64 *actual_bytes)
Liu Hui1f3c79a2009-01-05 15:57:51 -05001772{
Liu Hui1f3c79a2009-01-05 15:57:51 -05001773 int ret;
Li Dongyang5378e602011-03-24 10:24:27 +00001774 u64 discarded_bytes = 0;
Liu Hui1f3c79a2009-01-05 15:57:51 -05001775 struct btrfs_multi_bio *multi = NULL;
1776
Christoph Hellwige244a0a2009-10-14 09:24:59 -04001777
Liu Hui1f3c79a2009-01-05 15:57:51 -05001778 /* Tell the block device(s) that the sectors can be discarded */
Li Dongyang5378e602011-03-24 10:24:27 +00001779 ret = btrfs_map_block(&root->fs_info->mapping_tree, REQ_DISCARD,
1780 bytenr, &num_bytes, &multi, 0);
Liu Hui1f3c79a2009-01-05 15:57:51 -05001781 if (!ret) {
1782 struct btrfs_bio_stripe *stripe = multi->stripes;
1783 int i;
1784
Liu Hui1f3c79a2009-01-05 15:57:51 -05001785
1786 for (i = 0; i < multi->num_stripes; i++, stripe++) {
Li Dongyang5378e602011-03-24 10:24:27 +00001787 ret = btrfs_issue_discard(stripe->dev->bdev,
1788 stripe->physical,
1789 stripe->length);
1790 if (!ret)
1791 discarded_bytes += stripe->length;
1792 else if (ret != -EOPNOTSUPP)
1793 break;
Liu Hui1f3c79a2009-01-05 15:57:51 -05001794 }
1795 kfree(multi);
1796 }
Li Dongyang5378e602011-03-24 10:24:27 +00001797 if (discarded_bytes && ret == -EOPNOTSUPP)
1798 ret = 0;
1799
1800 if (actual_bytes)
1801 *actual_bytes = discarded_bytes;
1802
Liu Hui1f3c79a2009-01-05 15:57:51 -05001803
1804 return ret;
Liu Hui1f3c79a2009-01-05 15:57:51 -05001805}
1806
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001807int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1808 struct btrfs_root *root,
1809 u64 bytenr, u64 num_bytes, u64 parent,
1810 u64 root_objectid, u64 owner, u64 offset)
Zheng Yan31840ae2008-09-23 13:14:14 -04001811{
1812 int ret;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001813 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1814 root_objectid == BTRFS_TREE_LOG_OBJECTID);
Zheng Yan31840ae2008-09-23 13:14:14 -04001815
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001816 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1817 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
1818 parent, root_objectid, (int)owner,
1819 BTRFS_ADD_DELAYED_REF, NULL);
1820 } else {
1821 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
1822 parent, root_objectid, owner, offset,
1823 BTRFS_ADD_DELAYED_REF, NULL);
1824 }
Zheng Yan31840ae2008-09-23 13:14:14 -04001825 return ret;
1826}
1827
Chris Mason925baed2008-06-25 16:01:30 -04001828static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001829 struct btrfs_root *root,
1830 u64 bytenr, u64 num_bytes,
1831 u64 parent, u64 root_objectid,
1832 u64 owner, u64 offset, int refs_to_add,
1833 struct btrfs_delayed_extent_op *extent_op)
Chris Mason56bec292009-03-13 10:10:06 -04001834{
Chris Mason5caf2a02007-04-02 11:20:42 -04001835 struct btrfs_path *path;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001836 struct extent_buffer *leaf;
Chris Mason234b63a2007-03-13 10:46:10 -04001837 struct btrfs_extent_item *item;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001838 u64 refs;
1839 int ret;
1840 int err = 0;
Chris Mason037e6392007-03-07 11:50:24 -05001841
Chris Mason5caf2a02007-04-02 11:20:42 -04001842 path = btrfs_alloc_path();
Chris Mason54aa1f42007-06-22 14:16:25 -04001843 if (!path)
1844 return -ENOMEM;
Chris Mason26b80032007-08-08 20:17:12 -04001845
Chris Mason3c12ac72008-04-21 12:01:38 -04001846 path->reada = 1;
Chris Masonb9473432009-03-13 11:00:37 -04001847 path->leave_spinning = 1;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001848 /* this will setup the path even if it fails to insert the back ref */
1849 ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1850 path, bytenr, num_bytes, parent,
1851 root_objectid, owner, offset,
1852 refs_to_add, extent_op);
1853 if (ret == 0)
1854 goto out;
Zheng Yan31840ae2008-09-23 13:14:14 -04001855
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001856 if (ret != -EAGAIN) {
1857 err = ret;
1858 goto out;
Chris Masonb9473432009-03-13 11:00:37 -04001859 }
Zheng Yan31840ae2008-09-23 13:14:14 -04001860
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001861 leaf = path->nodes[0];
1862 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1863 refs = btrfs_extent_refs(leaf, item);
1864 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1865 if (extent_op)
1866 __run_delayed_extent_op(extent_op, leaf, item);
Zheng Yan31840ae2008-09-23 13:14:14 -04001867
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001868 btrfs_mark_buffer_dirty(leaf);
David Sterbab3b4aa72011-04-21 01:20:15 +02001869 btrfs_release_path(path);
Chris Mason7bb86312007-12-11 09:25:06 -05001870
Chris Mason3c12ac72008-04-21 12:01:38 -04001871 path->reada = 1;
Chris Masonb9473432009-03-13 11:00:37 -04001872 path->leave_spinning = 1;
1873
Chris Mason56bec292009-03-13 10:10:06 -04001874 /* now insert the actual backref */
Zheng Yan31840ae2008-09-23 13:14:14 -04001875 ret = insert_extent_backref(trans, root->fs_info->extent_root,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001876 path, bytenr, parent, root_objectid,
1877 owner, offset, refs_to_add);
Chris Mason7bb86312007-12-11 09:25:06 -05001878 BUG_ON(ret);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001879out:
Chris Mason74493f72007-12-11 09:25:06 -05001880 btrfs_free_path(path);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001881 return err;
Chris Mason02217ed2007-03-02 16:08:05 -05001882}
1883
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001884static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1885 struct btrfs_root *root,
1886 struct btrfs_delayed_ref_node *node,
1887 struct btrfs_delayed_extent_op *extent_op,
1888 int insert_reserved)
Chris Masone9d0b132007-08-10 14:06:19 -04001889{
Chris Mason56bec292009-03-13 10:10:06 -04001890 int ret = 0;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001891 struct btrfs_delayed_data_ref *ref;
1892 struct btrfs_key ins;
1893 u64 parent = 0;
1894 u64 ref_root = 0;
1895 u64 flags = 0;
Chris Mason56bec292009-03-13 10:10:06 -04001896
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001897 ins.objectid = node->bytenr;
1898 ins.offset = node->num_bytes;
1899 ins.type = BTRFS_EXTENT_ITEM_KEY;
Chris Mason56bec292009-03-13 10:10:06 -04001900
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001901 ref = btrfs_delayed_node_to_data_ref(node);
1902 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1903 parent = ref->parent;
1904 else
1905 ref_root = ref->root;
1906
1907 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1908 if (extent_op) {
1909 BUG_ON(extent_op->update_key);
1910 flags |= extent_op->flags_to_set;
1911 }
1912 ret = alloc_reserved_file_extent(trans, root,
1913 parent, ref_root, flags,
1914 ref->objectid, ref->offset,
1915 &ins, node->ref_mod);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001916 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1917 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1918 node->num_bytes, parent,
1919 ref_root, ref->objectid,
1920 ref->offset, node->ref_mod,
1921 extent_op);
1922 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1923 ret = __btrfs_free_extent(trans, root, node->bytenr,
1924 node->num_bytes, parent,
1925 ref_root, ref->objectid,
1926 ref->offset, node->ref_mod,
1927 extent_op);
1928 } else {
1929 BUG();
1930 }
Chris Mason56bec292009-03-13 10:10:06 -04001931 return ret;
1932}
1933
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001934static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1935 struct extent_buffer *leaf,
1936 struct btrfs_extent_item *ei)
1937{
1938 u64 flags = btrfs_extent_flags(leaf, ei);
1939 if (extent_op->update_flags) {
1940 flags |= extent_op->flags_to_set;
1941 btrfs_set_extent_flags(leaf, ei, flags);
1942 }
1943
1944 if (extent_op->update_key) {
1945 struct btrfs_tree_block_info *bi;
1946 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
1947 bi = (struct btrfs_tree_block_info *)(ei + 1);
1948 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
1949 }
1950}
1951
1952static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
1953 struct btrfs_root *root,
1954 struct btrfs_delayed_ref_node *node,
1955 struct btrfs_delayed_extent_op *extent_op)
1956{
1957 struct btrfs_key key;
1958 struct btrfs_path *path;
1959 struct btrfs_extent_item *ei;
1960 struct extent_buffer *leaf;
1961 u32 item_size;
1962 int ret;
1963 int err = 0;
1964
1965 path = btrfs_alloc_path();
1966 if (!path)
1967 return -ENOMEM;
1968
1969 key.objectid = node->bytenr;
1970 key.type = BTRFS_EXTENT_ITEM_KEY;
1971 key.offset = node->num_bytes;
1972
1973 path->reada = 1;
1974 path->leave_spinning = 1;
1975 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
1976 path, 0, 1);
1977 if (ret < 0) {
1978 err = ret;
1979 goto out;
1980 }
1981 if (ret > 0) {
1982 err = -EIO;
1983 goto out;
1984 }
1985
1986 leaf = path->nodes[0];
1987 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1988#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1989 if (item_size < sizeof(*ei)) {
1990 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
1991 path, (u64)-1, 0);
1992 if (ret < 0) {
1993 err = ret;
1994 goto out;
1995 }
1996 leaf = path->nodes[0];
1997 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1998 }
1999#endif
2000 BUG_ON(item_size < sizeof(*ei));
2001 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2002 __run_delayed_extent_op(extent_op, leaf, ei);
2003
2004 btrfs_mark_buffer_dirty(leaf);
2005out:
2006 btrfs_free_path(path);
2007 return err;
2008}
2009
2010static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2011 struct btrfs_root *root,
2012 struct btrfs_delayed_ref_node *node,
2013 struct btrfs_delayed_extent_op *extent_op,
2014 int insert_reserved)
2015{
2016 int ret = 0;
2017 struct btrfs_delayed_tree_ref *ref;
2018 struct btrfs_key ins;
2019 u64 parent = 0;
2020 u64 ref_root = 0;
2021
2022 ins.objectid = node->bytenr;
2023 ins.offset = node->num_bytes;
2024 ins.type = BTRFS_EXTENT_ITEM_KEY;
2025
2026 ref = btrfs_delayed_node_to_tree_ref(node);
2027 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2028 parent = ref->parent;
2029 else
2030 ref_root = ref->root;
2031
2032 BUG_ON(node->ref_mod != 1);
2033 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2034 BUG_ON(!extent_op || !extent_op->update_flags ||
2035 !extent_op->update_key);
2036 ret = alloc_reserved_tree_block(trans, root,
2037 parent, ref_root,
2038 extent_op->flags_to_set,
2039 &extent_op->key,
2040 ref->level, &ins);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002041 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2042 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2043 node->num_bytes, parent, ref_root,
2044 ref->level, 0, 1, extent_op);
2045 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2046 ret = __btrfs_free_extent(trans, root, node->bytenr,
2047 node->num_bytes, parent, ref_root,
2048 ref->level, 0, 1, extent_op);
2049 } else {
2050 BUG();
2051 }
2052 return ret;
2053}
2054
Chris Mason56bec292009-03-13 10:10:06 -04002055/* helper function to actually process a single delayed ref entry */
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002056static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2057 struct btrfs_root *root,
2058 struct btrfs_delayed_ref_node *node,
2059 struct btrfs_delayed_extent_op *extent_op,
2060 int insert_reserved)
Chris Mason56bec292009-03-13 10:10:06 -04002061{
Josef Bacikeb099672009-02-12 09:27:38 -05002062 int ret;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002063 if (btrfs_delayed_ref_is_head(node)) {
Chris Mason56bec292009-03-13 10:10:06 -04002064 struct btrfs_delayed_ref_head *head;
2065 /*
2066 * we've hit the end of the chain and we were supposed
2067 * to insert this extent into the tree. But, it got
2068 * deleted before we ever needed to insert it, so all
2069 * we have to do is clean up the accounting
2070 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002071 BUG_ON(extent_op);
2072 head = btrfs_delayed_node_to_head(node);
Chris Mason56bec292009-03-13 10:10:06 -04002073 if (insert_reserved) {
Yan, Zhengf0486c62010-05-16 10:46:25 -04002074 btrfs_pin_extent(root, node->bytenr,
2075 node->num_bytes, 1);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002076 if (head->is_data) {
2077 ret = btrfs_del_csums(trans, root,
2078 node->bytenr,
2079 node->num_bytes);
2080 BUG_ON(ret);
2081 }
Chris Mason56bec292009-03-13 10:10:06 -04002082 }
Chris Mason56bec292009-03-13 10:10:06 -04002083 mutex_unlock(&head->mutex);
2084 return 0;
2085 }
Josef Bacikeb099672009-02-12 09:27:38 -05002086
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002087 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2088 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2089 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2090 insert_reserved);
2091 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2092 node->type == BTRFS_SHARED_DATA_REF_KEY)
2093 ret = run_delayed_data_ref(trans, root, node, extent_op,
2094 insert_reserved);
2095 else
2096 BUG();
2097 return ret;
Chris Masone9d0b132007-08-10 14:06:19 -04002098}
2099
Chris Mason56bec292009-03-13 10:10:06 -04002100static noinline struct btrfs_delayed_ref_node *
2101select_delayed_ref(struct btrfs_delayed_ref_head *head)
Chris Masona28ec192007-03-06 20:08:01 -05002102{
Chris Mason56bec292009-03-13 10:10:06 -04002103 struct rb_node *node;
2104 struct btrfs_delayed_ref_node *ref;
2105 int action = BTRFS_ADD_DELAYED_REF;
2106again:
2107 /*
2108 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2109 * this prevents ref count from going down to zero when
2110 * there still are pending delayed ref.
2111 */
2112 node = rb_prev(&head->node.rb_node);
2113 while (1) {
2114 if (!node)
2115 break;
2116 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2117 rb_node);
2118 if (ref->bytenr != head->node.bytenr)
2119 break;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002120 if (ref->action == action)
Chris Mason56bec292009-03-13 10:10:06 -04002121 return ref;
2122 node = rb_prev(node);
Chris Mason5f39d392007-10-15 16:14:19 -04002123 }
Chris Mason56bec292009-03-13 10:10:06 -04002124 if (action == BTRFS_ADD_DELAYED_REF) {
2125 action = BTRFS_DROP_DELAYED_REF;
2126 goto again;
2127 }
2128 return NULL;
2129}
2130
Chris Masonc3e69d52009-03-13 10:17:05 -04002131static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2132 struct btrfs_root *root,
2133 struct list_head *cluster)
Chris Mason56bec292009-03-13 10:10:06 -04002134{
Chris Mason56bec292009-03-13 10:10:06 -04002135 struct btrfs_delayed_ref_root *delayed_refs;
2136 struct btrfs_delayed_ref_node *ref;
2137 struct btrfs_delayed_ref_head *locked_ref = NULL;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002138 struct btrfs_delayed_extent_op *extent_op;
Chris Mason56bec292009-03-13 10:10:06 -04002139 int ret;
Chris Masonc3e69d52009-03-13 10:17:05 -04002140 int count = 0;
Chris Mason56bec292009-03-13 10:10:06 -04002141 int must_insert_reserved = 0;
Chris Mason56bec292009-03-13 10:10:06 -04002142
2143 delayed_refs = &trans->transaction->delayed_refs;
Chris Mason56bec292009-03-13 10:10:06 -04002144 while (1) {
2145 if (!locked_ref) {
Chris Masonc3e69d52009-03-13 10:17:05 -04002146 /* pick a new head ref from the cluster list */
2147 if (list_empty(cluster))
Chris Mason56bec292009-03-13 10:10:06 -04002148 break;
Chris Mason56bec292009-03-13 10:10:06 -04002149
Chris Masonc3e69d52009-03-13 10:17:05 -04002150 locked_ref = list_entry(cluster->next,
2151 struct btrfs_delayed_ref_head, cluster);
2152
2153 /* grab the lock that says we are going to process
2154 * all the refs for this head */
2155 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2156
2157 /*
2158 * we may have dropped the spin lock to get the head
2159 * mutex lock, and that might have given someone else
2160 * time to free the head. If that's true, it has been
2161 * removed from our list and we can move on.
2162 */
2163 if (ret == -EAGAIN) {
2164 locked_ref = NULL;
2165 count++;
2166 continue;
Chris Mason56bec292009-03-13 10:10:06 -04002167 }
2168 }
2169
2170 /*
2171 * record the must insert reserved flag before we
2172 * drop the spin lock.
2173 */
2174 must_insert_reserved = locked_ref->must_insert_reserved;
2175 locked_ref->must_insert_reserved = 0;
2176
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002177 extent_op = locked_ref->extent_op;
2178 locked_ref->extent_op = NULL;
2179
Chris Mason56bec292009-03-13 10:10:06 -04002180 /*
2181 * locked_ref is the head node, so we have to go one
2182 * node back for any delayed ref updates
2183 */
Chris Mason56bec292009-03-13 10:10:06 -04002184 ref = select_delayed_ref(locked_ref);
2185 if (!ref) {
2186 /* All delayed refs have been processed, Go ahead
2187 * and send the head node to run_one_delayed_ref,
2188 * so that any accounting fixes can happen
2189 */
2190 ref = &locked_ref->node;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002191
2192 if (extent_op && must_insert_reserved) {
2193 kfree(extent_op);
2194 extent_op = NULL;
2195 }
2196
2197 if (extent_op) {
2198 spin_unlock(&delayed_refs->lock);
2199
2200 ret = run_delayed_extent_op(trans, root,
2201 ref, extent_op);
2202 BUG_ON(ret);
2203 kfree(extent_op);
2204
2205 cond_resched();
2206 spin_lock(&delayed_refs->lock);
2207 continue;
2208 }
2209
Chris Masonc3e69d52009-03-13 10:17:05 -04002210 list_del_init(&locked_ref->cluster);
Chris Mason56bec292009-03-13 10:10:06 -04002211 locked_ref = NULL;
2212 }
2213
2214 ref->in_tree = 0;
2215 rb_erase(&ref->rb_node, &delayed_refs->root);
2216 delayed_refs->num_entries--;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002217
Chris Mason56bec292009-03-13 10:10:06 -04002218 spin_unlock(&delayed_refs->lock);
2219
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002220 ret = run_one_delayed_ref(trans, root, ref, extent_op,
Chris Mason56bec292009-03-13 10:10:06 -04002221 must_insert_reserved);
2222 BUG_ON(ret);
Chris Mason56bec292009-03-13 10:10:06 -04002223
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002224 btrfs_put_delayed_ref(ref);
2225 kfree(extent_op);
Chris Masonc3e69d52009-03-13 10:17:05 -04002226 count++;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002227
Chris Mason1887be62009-03-13 10:11:24 -04002228 cond_resched();
Chris Mason56bec292009-03-13 10:10:06 -04002229 spin_lock(&delayed_refs->lock);
2230 }
Chris Masonc3e69d52009-03-13 10:17:05 -04002231 return count;
2232}
2233
2234/*
2235 * this starts processing the delayed reference count updates and
2236 * extent insertions we have queued up so far. count can be
2237 * 0, which means to process everything in the tree at the start
2238 * of the run (but not newly added entries), or it can be some target
2239 * number you'd like to process.
2240 */
2241int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2242 struct btrfs_root *root, unsigned long count)
2243{
2244 struct rb_node *node;
2245 struct btrfs_delayed_ref_root *delayed_refs;
2246 struct btrfs_delayed_ref_node *ref;
2247 struct list_head cluster;
2248 int ret;
2249 int run_all = count == (unsigned long)-1;
2250 int run_most = 0;
2251
2252 if (root == root->fs_info->extent_root)
2253 root = root->fs_info->tree_root;
2254
2255 delayed_refs = &trans->transaction->delayed_refs;
2256 INIT_LIST_HEAD(&cluster);
2257again:
2258 spin_lock(&delayed_refs->lock);
2259 if (count == 0) {
2260 count = delayed_refs->num_entries * 2;
2261 run_most = 1;
2262 }
2263 while (1) {
2264 if (!(run_all || run_most) &&
2265 delayed_refs->num_heads_ready < 64)
2266 break;
2267
2268 /*
2269 * go find something we can process in the rbtree. We start at
2270 * the beginning of the tree, and then build a cluster
2271 * of refs to process starting at the first one we are able to
2272 * lock
2273 */
2274 ret = btrfs_find_ref_cluster(trans, &cluster,
2275 delayed_refs->run_delayed_start);
2276 if (ret)
2277 break;
2278
2279 ret = run_clustered_refs(trans, root, &cluster);
2280 BUG_ON(ret < 0);
2281
2282 count -= min_t(unsigned long, ret, count);
2283
2284 if (count == 0)
2285 break;
2286 }
2287
Chris Mason56bec292009-03-13 10:10:06 -04002288 if (run_all) {
Chris Mason56bec292009-03-13 10:10:06 -04002289 node = rb_first(&delayed_refs->root);
Chris Masonc3e69d52009-03-13 10:17:05 -04002290 if (!node)
Chris Mason56bec292009-03-13 10:10:06 -04002291 goto out;
Chris Masonc3e69d52009-03-13 10:17:05 -04002292 count = (unsigned long)-1;
Chris Mason56bec292009-03-13 10:10:06 -04002293
2294 while (node) {
2295 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2296 rb_node);
2297 if (btrfs_delayed_ref_is_head(ref)) {
2298 struct btrfs_delayed_ref_head *head;
2299
2300 head = btrfs_delayed_node_to_head(ref);
2301 atomic_inc(&ref->refs);
2302
2303 spin_unlock(&delayed_refs->lock);
David Sterba8cc33e52011-05-02 15:29:25 +02002304 /*
2305 * Mutex was contended, block until it's
2306 * released and try again
2307 */
Chris Mason56bec292009-03-13 10:10:06 -04002308 mutex_lock(&head->mutex);
2309 mutex_unlock(&head->mutex);
2310
2311 btrfs_put_delayed_ref(ref);
Chris Mason1887be62009-03-13 10:11:24 -04002312 cond_resched();
Chris Mason56bec292009-03-13 10:10:06 -04002313 goto again;
2314 }
2315 node = rb_next(node);
2316 }
2317 spin_unlock(&delayed_refs->lock);
Chris Mason56bec292009-03-13 10:10:06 -04002318 schedule_timeout(1);
2319 goto again;
2320 }
Chris Mason54aa1f42007-06-22 14:16:25 -04002321out:
Chris Masonc3e69d52009-03-13 10:17:05 -04002322 spin_unlock(&delayed_refs->lock);
Chris Masona28ec192007-03-06 20:08:01 -05002323 return 0;
2324}
2325
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002326int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2327 struct btrfs_root *root,
2328 u64 bytenr, u64 num_bytes, u64 flags,
2329 int is_data)
2330{
2331 struct btrfs_delayed_extent_op *extent_op;
2332 int ret;
2333
2334 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2335 if (!extent_op)
2336 return -ENOMEM;
2337
2338 extent_op->flags_to_set = flags;
2339 extent_op->update_flags = 1;
2340 extent_op->update_key = 0;
2341 extent_op->is_data = is_data ? 1 : 0;
2342
2343 ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op);
2344 if (ret)
2345 kfree(extent_op);
2346 return ret;
2347}
2348
2349static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2350 struct btrfs_root *root,
2351 struct btrfs_path *path,
2352 u64 objectid, u64 offset, u64 bytenr)
2353{
2354 struct btrfs_delayed_ref_head *head;
2355 struct btrfs_delayed_ref_node *ref;
2356 struct btrfs_delayed_data_ref *data_ref;
2357 struct btrfs_delayed_ref_root *delayed_refs;
2358 struct rb_node *node;
2359 int ret = 0;
2360
2361 ret = -ENOENT;
2362 delayed_refs = &trans->transaction->delayed_refs;
2363 spin_lock(&delayed_refs->lock);
2364 head = btrfs_find_delayed_ref_head(trans, bytenr);
2365 if (!head)
2366 goto out;
2367
2368 if (!mutex_trylock(&head->mutex)) {
2369 atomic_inc(&head->node.refs);
2370 spin_unlock(&delayed_refs->lock);
2371
David Sterbab3b4aa72011-04-21 01:20:15 +02002372 btrfs_release_path(path);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002373
David Sterba8cc33e52011-05-02 15:29:25 +02002374 /*
2375 * Mutex was contended, block until it's released and let
2376 * caller try again
2377 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002378 mutex_lock(&head->mutex);
2379 mutex_unlock(&head->mutex);
2380 btrfs_put_delayed_ref(&head->node);
2381 return -EAGAIN;
2382 }
2383
2384 node = rb_prev(&head->node.rb_node);
2385 if (!node)
2386 goto out_unlock;
2387
2388 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2389
2390 if (ref->bytenr != bytenr)
2391 goto out_unlock;
2392
2393 ret = 1;
2394 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2395 goto out_unlock;
2396
2397 data_ref = btrfs_delayed_node_to_data_ref(ref);
2398
2399 node = rb_prev(node);
2400 if (node) {
2401 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2402 if (ref->bytenr == bytenr)
2403 goto out_unlock;
2404 }
2405
2406 if (data_ref->root != root->root_key.objectid ||
2407 data_ref->objectid != objectid || data_ref->offset != offset)
2408 goto out_unlock;
2409
2410 ret = 0;
2411out_unlock:
2412 mutex_unlock(&head->mutex);
2413out:
2414 spin_unlock(&delayed_refs->lock);
2415 return ret;
2416}
2417
2418static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2419 struct btrfs_root *root,
2420 struct btrfs_path *path,
2421 u64 objectid, u64 offset, u64 bytenr)
Chris Masonbe20aa92007-12-17 20:14:01 -05002422{
2423 struct btrfs_root *extent_root = root->fs_info->extent_root;
Yan Zhengf321e492008-07-30 09:26:11 -04002424 struct extent_buffer *leaf;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002425 struct btrfs_extent_data_ref *ref;
2426 struct btrfs_extent_inline_ref *iref;
2427 struct btrfs_extent_item *ei;
Chris Masonbe20aa92007-12-17 20:14:01 -05002428 struct btrfs_key key;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002429 u32 item_size;
Yan Zhengf321e492008-07-30 09:26:11 -04002430 int ret;
Chris Masonbe20aa92007-12-17 20:14:01 -05002431
Chris Masonbe20aa92007-12-17 20:14:01 -05002432 key.objectid = bytenr;
Zheng Yan31840ae2008-09-23 13:14:14 -04002433 key.offset = (u64)-1;
Yan Zhengf321e492008-07-30 09:26:11 -04002434 key.type = BTRFS_EXTENT_ITEM_KEY;
Chris Masonbe20aa92007-12-17 20:14:01 -05002435
Chris Masonbe20aa92007-12-17 20:14:01 -05002436 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2437 if (ret < 0)
2438 goto out;
2439 BUG_ON(ret == 0);
Yan Zheng80ff3852008-10-30 14:20:02 -04002440
2441 ret = -ENOENT;
2442 if (path->slots[0] == 0)
Zheng Yan31840ae2008-09-23 13:14:14 -04002443 goto out;
Chris Masonbe20aa92007-12-17 20:14:01 -05002444
Zheng Yan31840ae2008-09-23 13:14:14 -04002445 path->slots[0]--;
Yan Zhengf321e492008-07-30 09:26:11 -04002446 leaf = path->nodes[0];
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002447 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
Chris Masonbe20aa92007-12-17 20:14:01 -05002448
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002449 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
Chris Masonbe20aa92007-12-17 20:14:01 -05002450 goto out;
Chris Masonbe20aa92007-12-17 20:14:01 -05002451
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002452 ret = 1;
2453 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2454#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2455 if (item_size < sizeof(*ei)) {
2456 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2457 goto out;
Chris Masonbe20aa92007-12-17 20:14:01 -05002458 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002459#endif
2460 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2461
2462 if (item_size != sizeof(*ei) +
2463 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2464 goto out;
2465
2466 if (btrfs_extent_generation(leaf, ei) <=
2467 btrfs_root_last_snapshot(&root->root_item))
2468 goto out;
2469
2470 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2471 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2472 BTRFS_EXTENT_DATA_REF_KEY)
2473 goto out;
2474
2475 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2476 if (btrfs_extent_refs(leaf, ei) !=
2477 btrfs_extent_data_ref_count(leaf, ref) ||
2478 btrfs_extent_data_ref_root(leaf, ref) !=
2479 root->root_key.objectid ||
2480 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2481 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2482 goto out;
2483
Yan Zhengf321e492008-07-30 09:26:11 -04002484 ret = 0;
Chris Masonbe20aa92007-12-17 20:14:01 -05002485out:
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002486 return ret;
2487}
2488
2489int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2490 struct btrfs_root *root,
2491 u64 objectid, u64 offset, u64 bytenr)
2492{
2493 struct btrfs_path *path;
2494 int ret;
2495 int ret2;
2496
2497 path = btrfs_alloc_path();
2498 if (!path)
2499 return -ENOENT;
2500
2501 do {
2502 ret = check_committed_ref(trans, root, path, objectid,
2503 offset, bytenr);
2504 if (ret && ret != -ENOENT)
2505 goto out;
2506
2507 ret2 = check_delayed_ref(trans, root, path, objectid,
2508 offset, bytenr);
2509 } while (ret2 == -EAGAIN);
2510
2511 if (ret2 && ret2 != -ENOENT) {
2512 ret = ret2;
2513 goto out;
2514 }
2515
2516 if (ret != -ENOENT || ret2 != -ENOENT)
2517 ret = 0;
2518out:
Yan Zhengf321e492008-07-30 09:26:11 -04002519 btrfs_free_path(path);
Yan, Zhengf0486c62010-05-16 10:46:25 -04002520 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2521 WARN_ON(ret > 0);
Yan Zhengf321e492008-07-30 09:26:11 -04002522 return ret;
2523}
2524
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002525static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
Chris Masonb7a9f292009-02-04 09:23:45 -05002526 struct btrfs_root *root,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002527 struct extent_buffer *buf,
2528 int full_backref, int inc)
Zheng Yan31840ae2008-09-23 13:14:14 -04002529{
2530 u64 bytenr;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002531 u64 num_bytes;
2532 u64 parent;
Zheng Yan31840ae2008-09-23 13:14:14 -04002533 u64 ref_root;
Zheng Yan31840ae2008-09-23 13:14:14 -04002534 u32 nritems;
Zheng Yan31840ae2008-09-23 13:14:14 -04002535 struct btrfs_key key;
2536 struct btrfs_file_extent_item *fi;
2537 int i;
2538 int level;
2539 int ret = 0;
Zheng Yan31840ae2008-09-23 13:14:14 -04002540 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002541 u64, u64, u64, u64, u64, u64);
Zheng Yan31840ae2008-09-23 13:14:14 -04002542
2543 ref_root = btrfs_header_owner(buf);
Zheng Yan31840ae2008-09-23 13:14:14 -04002544 nritems = btrfs_header_nritems(buf);
2545 level = btrfs_header_level(buf);
2546
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002547 if (!root->ref_cows && level == 0)
2548 return 0;
Chris Masonb7a9f292009-02-04 09:23:45 -05002549
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002550 if (inc)
2551 process_func = btrfs_inc_extent_ref;
2552 else
2553 process_func = btrfs_free_extent;
Zheng Yan31840ae2008-09-23 13:14:14 -04002554
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002555 if (full_backref)
2556 parent = buf->start;
2557 else
2558 parent = 0;
2559
Zheng Yan31840ae2008-09-23 13:14:14 -04002560 for (i = 0; i < nritems; i++) {
Chris Masondb945352007-10-15 16:15:53 -04002561 if (level == 0) {
Chris Mason5f39d392007-10-15 16:14:19 -04002562 btrfs_item_key_to_cpu(buf, &key, i);
2563 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
Chris Mason54aa1f42007-06-22 14:16:25 -04002564 continue;
Chris Mason5f39d392007-10-15 16:14:19 -04002565 fi = btrfs_item_ptr(buf, i,
Chris Mason54aa1f42007-06-22 14:16:25 -04002566 struct btrfs_file_extent_item);
Chris Mason5f39d392007-10-15 16:14:19 -04002567 if (btrfs_file_extent_type(buf, fi) ==
Chris Mason54aa1f42007-06-22 14:16:25 -04002568 BTRFS_FILE_EXTENT_INLINE)
2569 continue;
Zheng Yan31840ae2008-09-23 13:14:14 -04002570 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2571 if (bytenr == 0)
Chris Mason54aa1f42007-06-22 14:16:25 -04002572 continue;
Zheng Yan31840ae2008-09-23 13:14:14 -04002573
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002574 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2575 key.offset -= btrfs_file_extent_offset(buf, fi);
2576 ret = process_func(trans, root, bytenr, num_bytes,
2577 parent, ref_root, key.objectid,
2578 key.offset);
2579 if (ret)
2580 goto fail;
Chris Masonb7a9f292009-02-04 09:23:45 -05002581 } else {
2582 bytenr = btrfs_node_blockptr(buf, i);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002583 num_bytes = btrfs_level_size(root, level - 1);
2584 ret = process_func(trans, root, bytenr, num_bytes,
2585 parent, ref_root, level - 1, 0);
2586 if (ret)
Zheng Yan31840ae2008-09-23 13:14:14 -04002587 goto fail;
Chris Mason54aa1f42007-06-22 14:16:25 -04002588 }
2589 }
Zheng Yan31840ae2008-09-23 13:14:14 -04002590 return 0;
2591fail:
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002592 BUG();
Chris Mason54aa1f42007-06-22 14:16:25 -04002593 return ret;
Chris Mason02217ed2007-03-02 16:08:05 -05002594}
2595
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002596int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2597 struct extent_buffer *buf, int full_backref)
Zheng Yan31840ae2008-09-23 13:14:14 -04002598{
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002599 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
2600}
Zheng Yan31840ae2008-09-23 13:14:14 -04002601
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002602int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2603 struct extent_buffer *buf, int full_backref)
2604{
2605 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
Zheng Yan31840ae2008-09-23 13:14:14 -04002606}
2607
Chris Mason9078a3e2007-04-26 16:46:15 -04002608static int write_one_cache_group(struct btrfs_trans_handle *trans,
2609 struct btrfs_root *root,
2610 struct btrfs_path *path,
2611 struct btrfs_block_group_cache *cache)
2612{
2613 int ret;
Chris Mason9078a3e2007-04-26 16:46:15 -04002614 struct btrfs_root *extent_root = root->fs_info->extent_root;
Chris Mason5f39d392007-10-15 16:14:19 -04002615 unsigned long bi;
2616 struct extent_buffer *leaf;
Chris Mason9078a3e2007-04-26 16:46:15 -04002617
Chris Mason9078a3e2007-04-26 16:46:15 -04002618 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
Chris Mason54aa1f42007-06-22 14:16:25 -04002619 if (ret < 0)
2620 goto fail;
Chris Mason9078a3e2007-04-26 16:46:15 -04002621 BUG_ON(ret);
Chris Mason5f39d392007-10-15 16:14:19 -04002622
2623 leaf = path->nodes[0];
2624 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2625 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2626 btrfs_mark_buffer_dirty(leaf);
David Sterbab3b4aa72011-04-21 01:20:15 +02002627 btrfs_release_path(path);
Chris Mason54aa1f42007-06-22 14:16:25 -04002628fail:
Chris Mason9078a3e2007-04-26 16:46:15 -04002629 if (ret)
2630 return ret;
Chris Mason9078a3e2007-04-26 16:46:15 -04002631 return 0;
2632
2633}
2634
Yan Zheng4a8c9a62009-07-22 10:07:05 -04002635static struct btrfs_block_group_cache *
2636next_block_group(struct btrfs_root *root,
2637 struct btrfs_block_group_cache *cache)
2638{
2639 struct rb_node *node;
2640 spin_lock(&root->fs_info->block_group_cache_lock);
2641 node = rb_next(&cache->cache_node);
2642 btrfs_put_block_group(cache);
2643 if (node) {
2644 cache = rb_entry(node, struct btrfs_block_group_cache,
2645 cache_node);
Josef Bacik11dfe352009-11-13 20:12:59 +00002646 btrfs_get_block_group(cache);
Yan Zheng4a8c9a62009-07-22 10:07:05 -04002647 } else
2648 cache = NULL;
2649 spin_unlock(&root->fs_info->block_group_cache_lock);
2650 return cache;
2651}
2652
Josef Bacik0af3d002010-06-21 14:48:16 -04002653static int cache_save_setup(struct btrfs_block_group_cache *block_group,
2654 struct btrfs_trans_handle *trans,
2655 struct btrfs_path *path)
2656{
2657 struct btrfs_root *root = block_group->fs_info->tree_root;
2658 struct inode *inode = NULL;
2659 u64 alloc_hint = 0;
Josef Bacik2b209822010-12-03 13:17:53 -05002660 int dcs = BTRFS_DC_ERROR;
Josef Bacik0af3d002010-06-21 14:48:16 -04002661 int num_pages = 0;
2662 int retries = 0;
2663 int ret = 0;
2664
2665 /*
2666 * If this block group is smaller than 100 megs don't bother caching the
2667 * block group.
2668 */
2669 if (block_group->key.offset < (100 * 1024 * 1024)) {
2670 spin_lock(&block_group->lock);
2671 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
2672 spin_unlock(&block_group->lock);
2673 return 0;
2674 }
2675
2676again:
2677 inode = lookup_free_space_inode(root, block_group, path);
2678 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
2679 ret = PTR_ERR(inode);
David Sterbab3b4aa72011-04-21 01:20:15 +02002680 btrfs_release_path(path);
Josef Bacik0af3d002010-06-21 14:48:16 -04002681 goto out;
2682 }
2683
2684 if (IS_ERR(inode)) {
2685 BUG_ON(retries);
2686 retries++;
2687
2688 if (block_group->ro)
2689 goto out_free;
2690
2691 ret = create_free_space_inode(root, trans, block_group, path);
2692 if (ret)
2693 goto out_free;
2694 goto again;
2695 }
2696
2697 /*
2698 * We want to set the generation to 0, that way if anything goes wrong
2699 * from here on out we know not to trust this cache when we load up next
2700 * time.
2701 */
2702 BTRFS_I(inode)->generation = 0;
2703 ret = btrfs_update_inode(trans, root, inode);
2704 WARN_ON(ret);
2705
2706 if (i_size_read(inode) > 0) {
2707 ret = btrfs_truncate_free_space_cache(root, trans, path,
2708 inode);
2709 if (ret)
2710 goto out_put;
2711 }
2712
2713 spin_lock(&block_group->lock);
2714 if (block_group->cached != BTRFS_CACHE_FINISHED) {
Josef Bacik2b209822010-12-03 13:17:53 -05002715 /* We're not cached, don't bother trying to write stuff out */
2716 dcs = BTRFS_DC_WRITTEN;
Josef Bacik0af3d002010-06-21 14:48:16 -04002717 spin_unlock(&block_group->lock);
2718 goto out_put;
2719 }
2720 spin_unlock(&block_group->lock);
2721
2722 num_pages = (int)div64_u64(block_group->key.offset, 1024 * 1024 * 1024);
2723 if (!num_pages)
2724 num_pages = 1;
2725
2726 /*
2727 * Just to make absolutely sure we have enough space, we're going to
2728 * preallocate 12 pages worth of space for each block group. In
2729 * practice we ought to use at most 8, but we need extra space so we can
2730 * add our header and have a terminator between the extents and the
2731 * bitmaps.
2732 */
2733 num_pages *= 16;
2734 num_pages *= PAGE_CACHE_SIZE;
2735
2736 ret = btrfs_check_data_free_space(inode, num_pages);
2737 if (ret)
2738 goto out_put;
2739
2740 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
2741 num_pages, num_pages,
2742 &alloc_hint);
Josef Bacik2b209822010-12-03 13:17:53 -05002743 if (!ret)
2744 dcs = BTRFS_DC_SETUP;
Josef Bacik0af3d002010-06-21 14:48:16 -04002745 btrfs_free_reserved_data_space(inode, num_pages);
2746out_put:
2747 iput(inode);
2748out_free:
David Sterbab3b4aa72011-04-21 01:20:15 +02002749 btrfs_release_path(path);
Josef Bacik0af3d002010-06-21 14:48:16 -04002750out:
2751 spin_lock(&block_group->lock);
Josef Bacik2b209822010-12-03 13:17:53 -05002752 block_group->disk_cache_state = dcs;
Josef Bacik0af3d002010-06-21 14:48:16 -04002753 spin_unlock(&block_group->lock);
2754
2755 return ret;
2756}
2757
Chris Mason96b51792007-10-15 16:15:19 -04002758int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
2759 struct btrfs_root *root)
Chris Mason9078a3e2007-04-26 16:46:15 -04002760{
Yan Zheng4a8c9a62009-07-22 10:07:05 -04002761 struct btrfs_block_group_cache *cache;
Chris Mason9078a3e2007-04-26 16:46:15 -04002762 int err = 0;
Chris Mason9078a3e2007-04-26 16:46:15 -04002763 struct btrfs_path *path;
Chris Mason96b51792007-10-15 16:15:19 -04002764 u64 last = 0;
Chris Mason9078a3e2007-04-26 16:46:15 -04002765
2766 path = btrfs_alloc_path();
2767 if (!path)
2768 return -ENOMEM;
2769
Josef Bacik0af3d002010-06-21 14:48:16 -04002770again:
2771 while (1) {
2772 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2773 while (cache) {
2774 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
2775 break;
2776 cache = next_block_group(root, cache);
2777 }
2778 if (!cache) {
2779 if (last == 0)
2780 break;
2781 last = 0;
2782 continue;
2783 }
2784 err = cache_save_setup(cache, trans, path);
2785 last = cache->key.objectid + cache->key.offset;
2786 btrfs_put_block_group(cache);
2787 }
2788
Chris Masond3977122009-01-05 21:25:51 -05002789 while (1) {
Yan Zheng4a8c9a62009-07-22 10:07:05 -04002790 if (last == 0) {
2791 err = btrfs_run_delayed_refs(trans, root,
2792 (unsigned long)-1);
2793 BUG_ON(err);
Josef Bacik0f9dd462008-09-23 13:14:11 -04002794 }
Josef Bacik0f9dd462008-09-23 13:14:11 -04002795
Yan Zheng4a8c9a62009-07-22 10:07:05 -04002796 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2797 while (cache) {
Josef Bacik0af3d002010-06-21 14:48:16 -04002798 if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
2799 btrfs_put_block_group(cache);
2800 goto again;
2801 }
2802
Yan Zheng4a8c9a62009-07-22 10:07:05 -04002803 if (cache->dirty)
2804 break;
2805 cache = next_block_group(root, cache);
2806 }
2807 if (!cache) {
2808 if (last == 0)
2809 break;
2810 last = 0;
Chris Mason96b51792007-10-15 16:15:19 -04002811 continue;
Chris Mason9078a3e2007-04-26 16:46:15 -04002812 }
Yan Zheng4a8c9a62009-07-22 10:07:05 -04002813
Josef Bacik0cb59c92010-07-02 12:14:14 -04002814 if (cache->disk_cache_state == BTRFS_DC_SETUP)
2815 cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
Yan Zheng4a8c9a62009-07-22 10:07:05 -04002816 cache->dirty = 0;
2817 last = cache->key.objectid + cache->key.offset;
2818
2819 err = write_one_cache_group(trans, root, path, cache);
2820 BUG_ON(err);
2821 btrfs_put_block_group(cache);
Chris Mason9078a3e2007-04-26 16:46:15 -04002822 }
Yan Zheng4a8c9a62009-07-22 10:07:05 -04002823
Josef Bacik0cb59c92010-07-02 12:14:14 -04002824 while (1) {
2825 /*
2826 * I don't think this is needed since we're just marking our
2827 * preallocated extent as written, but just in case it can't
2828 * hurt.
2829 */
2830 if (last == 0) {
2831 err = btrfs_run_delayed_refs(trans, root,
2832 (unsigned long)-1);
2833 BUG_ON(err);
2834 }
2835
2836 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2837 while (cache) {
2838 /*
2839 * Really this shouldn't happen, but it could if we
2840 * couldn't write the entire preallocated extent and
2841 * splitting the extent resulted in a new block.
2842 */
2843 if (cache->dirty) {
2844 btrfs_put_block_group(cache);
2845 goto again;
2846 }
2847 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2848 break;
2849 cache = next_block_group(root, cache);
2850 }
2851 if (!cache) {
2852 if (last == 0)
2853 break;
2854 last = 0;
2855 continue;
2856 }
2857
2858 btrfs_write_out_cache(root, trans, cache, path);
2859
2860 /*
2861 * If we didn't have an error then the cache state is still
2862 * NEED_WRITE, so we can set it to WRITTEN.
2863 */
2864 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2865 cache->disk_cache_state = BTRFS_DC_WRITTEN;
2866 last = cache->key.objectid + cache->key.offset;
2867 btrfs_put_block_group(cache);
2868 }
2869
Chris Mason9078a3e2007-04-26 16:46:15 -04002870 btrfs_free_path(path);
Yan Zheng4a8c9a62009-07-22 10:07:05 -04002871 return 0;
Chris Mason9078a3e2007-04-26 16:46:15 -04002872}
2873
Yan Zhengd2fb3432008-12-11 16:30:39 -05002874int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
2875{
2876 struct btrfs_block_group_cache *block_group;
2877 int readonly = 0;
2878
2879 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
2880 if (!block_group || block_group->ro)
2881 readonly = 1;
2882 if (block_group)
Chris Masonfa9c0d72009-04-03 09:47:43 -04002883 btrfs_put_block_group(block_group);
Yan Zhengd2fb3432008-12-11 16:30:39 -05002884 return readonly;
2885}
2886
Chris Mason593060d2008-03-25 16:50:33 -04002887static int update_space_info(struct btrfs_fs_info *info, u64 flags,
2888 u64 total_bytes, u64 bytes_used,
2889 struct btrfs_space_info **space_info)
2890{
2891 struct btrfs_space_info *found;
Yan, Zhengb742bb82010-05-16 10:46:24 -04002892 int i;
2893 int factor;
2894
2895 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
2896 BTRFS_BLOCK_GROUP_RAID10))
2897 factor = 2;
2898 else
2899 factor = 1;
Chris Mason593060d2008-03-25 16:50:33 -04002900
2901 found = __find_space_info(info, flags);
2902 if (found) {
Josef Bacik25179202008-10-29 14:49:05 -04002903 spin_lock(&found->lock);
Chris Mason593060d2008-03-25 16:50:33 -04002904 found->total_bytes += total_bytes;
Josef Bacik89a55892010-10-14 14:52:27 -04002905 found->disk_total += total_bytes * factor;
Chris Mason593060d2008-03-25 16:50:33 -04002906 found->bytes_used += bytes_used;
Yan, Zhengb742bb82010-05-16 10:46:24 -04002907 found->disk_used += bytes_used * factor;
Chris Mason8f18cf12008-04-25 16:53:30 -04002908 found->full = 0;
Josef Bacik25179202008-10-29 14:49:05 -04002909 spin_unlock(&found->lock);
Chris Mason593060d2008-03-25 16:50:33 -04002910 *space_info = found;
2911 return 0;
2912 }
Yan Zhengc146afa2008-11-12 14:34:12 -05002913 found = kzalloc(sizeof(*found), GFP_NOFS);
Chris Mason593060d2008-03-25 16:50:33 -04002914 if (!found)
2915 return -ENOMEM;
2916
Yan, Zhengb742bb82010-05-16 10:46:24 -04002917 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
2918 INIT_LIST_HEAD(&found->block_groups[i]);
Josef Bacik80eb2342008-10-29 14:49:05 -04002919 init_rwsem(&found->groups_sem);
Josef Bacik0f9dd462008-09-23 13:14:11 -04002920 spin_lock_init(&found->lock);
Yan, Zhengb742bb82010-05-16 10:46:24 -04002921 found->flags = flags & (BTRFS_BLOCK_GROUP_DATA |
2922 BTRFS_BLOCK_GROUP_SYSTEM |
2923 BTRFS_BLOCK_GROUP_METADATA);
Chris Mason593060d2008-03-25 16:50:33 -04002924 found->total_bytes = total_bytes;
Josef Bacik89a55892010-10-14 14:52:27 -04002925 found->disk_total = total_bytes * factor;
Chris Mason593060d2008-03-25 16:50:33 -04002926 found->bytes_used = bytes_used;
Yan, Zhengb742bb82010-05-16 10:46:24 -04002927 found->disk_used = bytes_used * factor;
Chris Mason593060d2008-03-25 16:50:33 -04002928 found->bytes_pinned = 0;
Zheng Yane8569812008-09-26 10:05:48 -04002929 found->bytes_reserved = 0;
Yan Zhengc146afa2008-11-12 14:34:12 -05002930 found->bytes_readonly = 0;
Yan, Zhengf0486c62010-05-16 10:46:25 -04002931 found->bytes_may_use = 0;
Chris Mason593060d2008-03-25 16:50:33 -04002932 found->full = 0;
Chris Mason0e4f8f82011-04-15 16:05:44 -04002933 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
Josef Bacik6d741192011-04-11 20:20:11 -04002934 found->chunk_alloc = 0;
Josef Bacikfdb5eff2011-06-07 16:07:44 -04002935 found->flush = 0;
2936 init_waitqueue_head(&found->wait);
Chris Mason593060d2008-03-25 16:50:33 -04002937 *space_info = found;
Chris Mason4184ea72009-03-10 12:39:20 -04002938 list_add_rcu(&found->list, &info->space_info);
Josef Bacik817d52f2009-07-13 21:29:25 -04002939 atomic_set(&found->caching_threads, 0);
Chris Mason593060d2008-03-25 16:50:33 -04002940 return 0;
2941}
2942
Chris Mason8790d502008-04-03 16:29:03 -04002943static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
2944{
2945 u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
Chris Mason611f0e02008-04-03 16:29:03 -04002946 BTRFS_BLOCK_GROUP_RAID1 |
Chris Mason321aecc2008-04-16 10:49:51 -04002947 BTRFS_BLOCK_GROUP_RAID10 |
Chris Mason611f0e02008-04-03 16:29:03 -04002948 BTRFS_BLOCK_GROUP_DUP);
Chris Mason8790d502008-04-03 16:29:03 -04002949 if (extra_flags) {
2950 if (flags & BTRFS_BLOCK_GROUP_DATA)
2951 fs_info->avail_data_alloc_bits |= extra_flags;
2952 if (flags & BTRFS_BLOCK_GROUP_METADATA)
2953 fs_info->avail_metadata_alloc_bits |= extra_flags;
2954 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
2955 fs_info->avail_system_alloc_bits |= extra_flags;
2956 }
2957}
Chris Mason593060d2008-03-25 16:50:33 -04002958
Yan Zheng2b820322008-11-17 21:11:30 -05002959u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
Chris Masonec44a352008-04-28 15:29:52 -04002960{
Chris Masoncd02dca2010-12-13 14:56:23 -05002961 /*
2962 * we add in the count of missing devices because we want
2963 * to make sure that any RAID levels on a degraded FS
2964 * continue to be honored.
2965 */
2966 u64 num_devices = root->fs_info->fs_devices->rw_devices +
2967 root->fs_info->fs_devices->missing_devices;
Chris Masona061fc82008-05-07 11:43:44 -04002968
2969 if (num_devices == 1)
2970 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
2971 if (num_devices < 4)
2972 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
2973
Chris Masonec44a352008-04-28 15:29:52 -04002974 if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
2975 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
Chris Masona061fc82008-05-07 11:43:44 -04002976 BTRFS_BLOCK_GROUP_RAID10))) {
Chris Masonec44a352008-04-28 15:29:52 -04002977 flags &= ~BTRFS_BLOCK_GROUP_DUP;
Chris Masona061fc82008-05-07 11:43:44 -04002978 }
Chris Masonec44a352008-04-28 15:29:52 -04002979
2980 if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
Chris Masona061fc82008-05-07 11:43:44 -04002981 (flags & BTRFS_BLOCK_GROUP_RAID10)) {
Chris Masonec44a352008-04-28 15:29:52 -04002982 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
Chris Masona061fc82008-05-07 11:43:44 -04002983 }
Chris Masonec44a352008-04-28 15:29:52 -04002984
2985 if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
2986 ((flags & BTRFS_BLOCK_GROUP_RAID1) |
2987 (flags & BTRFS_BLOCK_GROUP_RAID10) |
2988 (flags & BTRFS_BLOCK_GROUP_DUP)))
2989 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
2990 return flags;
2991}
2992
Yan, Zhengb742bb82010-05-16 10:46:24 -04002993static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
Josef Bacik6a632092009-02-20 11:00:09 -05002994{
Yan, Zhengb742bb82010-05-16 10:46:24 -04002995 if (flags & BTRFS_BLOCK_GROUP_DATA)
2996 flags |= root->fs_info->avail_data_alloc_bits &
2997 root->fs_info->data_alloc_profile;
2998 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
2999 flags |= root->fs_info->avail_system_alloc_bits &
3000 root->fs_info->system_alloc_profile;
3001 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3002 flags |= root->fs_info->avail_metadata_alloc_bits &
3003 root->fs_info->metadata_alloc_profile;
3004 return btrfs_reduce_alloc_profile(root, flags);
3005}
Josef Bacik6a632092009-02-20 11:00:09 -05003006
Miao Xie6d07bce2011-01-05 10:07:31 +00003007u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
Yan, Zhengb742bb82010-05-16 10:46:24 -04003008{
3009 u64 flags;
Josef Bacik6a632092009-02-20 11:00:09 -05003010
Yan, Zhengb742bb82010-05-16 10:46:24 -04003011 if (data)
3012 flags = BTRFS_BLOCK_GROUP_DATA;
3013 else if (root == root->fs_info->chunk_root)
3014 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3015 else
3016 flags = BTRFS_BLOCK_GROUP_METADATA;
3017
3018 return get_alloc_profile(root, flags);
Josef Bacik6a632092009-02-20 11:00:09 -05003019}
3020
3021void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
3022{
Josef Bacik6a632092009-02-20 11:00:09 -05003023 BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
Yan, Zhengf0486c62010-05-16 10:46:25 -04003024 BTRFS_BLOCK_GROUP_DATA);
Josef Bacik6a632092009-02-20 11:00:09 -05003025}
3026
3027/*
3028 * This will check the space that the inode allocates from to make sure we have
3029 * enough space for bytes.
3030 */
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04003031int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
Josef Bacik6a632092009-02-20 11:00:09 -05003032{
3033 struct btrfs_space_info *data_sinfo;
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04003034 struct btrfs_root *root = BTRFS_I(inode)->root;
Josef Bacikab6e24102010-03-19 14:38:13 +00003035 u64 used;
Josef Bacik0af3d002010-06-21 14:48:16 -04003036 int ret = 0, committed = 0, alloc_chunk = 1;
Josef Bacik6a632092009-02-20 11:00:09 -05003037
3038 /* make sure bytes are sectorsize aligned */
3039 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3040
Li Zefan82d59022011-04-20 10:33:24 +08003041 if (root == root->fs_info->tree_root ||
3042 BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) {
Josef Bacik0af3d002010-06-21 14:48:16 -04003043 alloc_chunk = 0;
3044 committed = 1;
3045 }
3046
Josef Bacik6a632092009-02-20 11:00:09 -05003047 data_sinfo = BTRFS_I(inode)->space_info;
Chris Mason33b4d472009-09-22 14:45:50 -04003048 if (!data_sinfo)
3049 goto alloc;
3050
Josef Bacik6a632092009-02-20 11:00:09 -05003051again:
3052 /* make sure we have enough space to handle the data first */
3053 spin_lock(&data_sinfo->lock);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04003054 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3055 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3056 data_sinfo->bytes_may_use;
Josef Bacikab6e24102010-03-19 14:38:13 +00003057
3058 if (used + bytes > data_sinfo->total_bytes) {
Josef Bacik4e06bdd2009-02-20 10:59:53 -05003059 struct btrfs_trans_handle *trans;
3060
Josef Bacik6a632092009-02-20 11:00:09 -05003061 /*
3062 * if we don't have enough free bytes in this space then we need
3063 * to alloc a new chunk.
3064 */
Josef Bacik0af3d002010-06-21 14:48:16 -04003065 if (!data_sinfo->full && alloc_chunk) {
Josef Bacik6a632092009-02-20 11:00:09 -05003066 u64 alloc_target;
Josef Bacik6a632092009-02-20 11:00:09 -05003067
Chris Mason0e4f8f82011-04-15 16:05:44 -04003068 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
Josef Bacik6a632092009-02-20 11:00:09 -05003069 spin_unlock(&data_sinfo->lock);
Chris Mason33b4d472009-09-22 14:45:50 -04003070alloc:
Josef Bacik6a632092009-02-20 11:00:09 -05003071 alloc_target = btrfs_get_alloc_profile(root, 1);
Josef Bacik7a7eaa42011-04-13 12:54:33 -04003072 trans = btrfs_join_transaction(root);
Yan, Zhenga22285a2010-05-16 10:48:46 -04003073 if (IS_ERR(trans))
3074 return PTR_ERR(trans);
Josef Bacik6a632092009-02-20 11:00:09 -05003075
3076 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3077 bytes + 2 * 1024 * 1024,
Chris Mason0e4f8f82011-04-15 16:05:44 -04003078 alloc_target,
3079 CHUNK_ALLOC_NO_FORCE);
Josef Bacik6a632092009-02-20 11:00:09 -05003080 btrfs_end_transaction(trans, root);
Miao Xied52a5b52011-01-05 10:07:18 +00003081 if (ret < 0) {
3082 if (ret != -ENOSPC)
3083 return ret;
3084 else
3085 goto commit_trans;
3086 }
Chris Mason33b4d472009-09-22 14:45:50 -04003087
3088 if (!data_sinfo) {
3089 btrfs_set_inode_space_info(root, inode);
3090 data_sinfo = BTRFS_I(inode)->space_info;
3091 }
Josef Bacik6a632092009-02-20 11:00:09 -05003092 goto again;
3093 }
Josef Bacikf2bb8f52011-05-25 13:10:16 -04003094
3095 /*
3096 * If we have less pinned bytes than we want to allocate then
3097 * don't bother committing the transaction, it won't help us.
3098 */
3099 if (data_sinfo->bytes_pinned < bytes)
3100 committed = 1;
Josef Bacik6a632092009-02-20 11:00:09 -05003101 spin_unlock(&data_sinfo->lock);
Josef Bacik4e06bdd2009-02-20 10:59:53 -05003102
3103 /* commit the current transaction and try again */
Miao Xied52a5b52011-01-05 10:07:18 +00003104commit_trans:
Josef Bacika4abeea2011-04-11 17:25:13 -04003105 if (!committed &&
3106 !atomic_read(&root->fs_info->open_ioctl_trans)) {
Josef Bacik4e06bdd2009-02-20 10:59:53 -05003107 committed = 1;
Josef Bacik7a7eaa42011-04-13 12:54:33 -04003108 trans = btrfs_join_transaction(root);
Yan, Zhenga22285a2010-05-16 10:48:46 -04003109 if (IS_ERR(trans))
3110 return PTR_ERR(trans);
Josef Bacik4e06bdd2009-02-20 10:59:53 -05003111 ret = btrfs_commit_transaction(trans, root);
3112 if (ret)
3113 return ret;
3114 goto again;
3115 }
3116
Josef Bacik6a632092009-02-20 11:00:09 -05003117 return -ENOSPC;
3118 }
3119 data_sinfo->bytes_may_use += bytes;
3120 BTRFS_I(inode)->reserved_bytes += bytes;
3121 spin_unlock(&data_sinfo->lock);
3122
Josef Bacik9ed74f22009-09-11 16:12:44 -04003123 return 0;
Josef Bacik6a632092009-02-20 11:00:09 -05003124}
3125
3126/*
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04003127 * called when we are clearing an delalloc extent from the
3128 * inode's io_tree or there was an error for whatever reason
3129 * after calling btrfs_check_data_free_space
Josef Bacik6a632092009-02-20 11:00:09 -05003130 */
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04003131void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
Josef Bacik6a632092009-02-20 11:00:09 -05003132{
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04003133 struct btrfs_root *root = BTRFS_I(inode)->root;
Josef Bacik6a632092009-02-20 11:00:09 -05003134 struct btrfs_space_info *data_sinfo;
3135
3136 /* make sure bytes are sectorsize aligned */
3137 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3138
3139 data_sinfo = BTRFS_I(inode)->space_info;
3140 spin_lock(&data_sinfo->lock);
3141 data_sinfo->bytes_may_use -= bytes;
3142 BTRFS_I(inode)->reserved_bytes -= bytes;
3143 spin_unlock(&data_sinfo->lock);
3144}
3145
Josef Bacik97e728d2009-04-21 17:40:57 -04003146static void force_metadata_allocation(struct btrfs_fs_info *info)
3147{
3148 struct list_head *head = &info->space_info;
3149 struct btrfs_space_info *found;
3150
3151 rcu_read_lock();
3152 list_for_each_entry_rcu(found, head, list) {
3153 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
Chris Mason0e4f8f82011-04-15 16:05:44 -04003154 found->force_alloc = CHUNK_ALLOC_FORCE;
Josef Bacik97e728d2009-04-21 17:40:57 -04003155 }
3156 rcu_read_unlock();
3157}
3158
Chris Masone5bc2452010-10-26 13:37:56 -04003159static int should_alloc_chunk(struct btrfs_root *root,
Chris Mason0e4f8f82011-04-15 16:05:44 -04003160 struct btrfs_space_info *sinfo, u64 alloc_bytes,
3161 int force)
Yan, Zheng424499d2010-05-16 10:46:25 -04003162{
3163 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
Chris Mason0e4f8f82011-04-15 16:05:44 -04003164 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
Chris Masone5bc2452010-10-26 13:37:56 -04003165 u64 thresh;
Yan, Zheng424499d2010-05-16 10:46:25 -04003166
Chris Mason0e4f8f82011-04-15 16:05:44 -04003167 if (force == CHUNK_ALLOC_FORCE)
3168 return 1;
3169
3170 /*
3171 * in limited mode, we want to have some free space up to
3172 * about 1% of the FS size.
3173 */
3174 if (force == CHUNK_ALLOC_LIMITED) {
3175 thresh = btrfs_super_total_bytes(&root->fs_info->super_copy);
3176 thresh = max_t(u64, 64 * 1024 * 1024,
3177 div_factor_fine(thresh, 1));
3178
3179 if (num_bytes - num_allocated < thresh)
3180 return 1;
3181 }
3182
3183 /*
3184 * we have two similar checks here, one based on percentage
3185 * and once based on a hard number of 256MB. The idea
3186 * is that if we have a good amount of free
3187 * room, don't allocate a chunk. A good mount is
3188 * less than 80% utilized of the chunks we have allocated,
3189 * or more than 256MB free
3190 */
3191 if (num_allocated + alloc_bytes + 256 * 1024 * 1024 < num_bytes)
Yan, Zheng424499d2010-05-16 10:46:25 -04003192 return 0;
3193
Chris Mason0e4f8f82011-04-15 16:05:44 -04003194 if (num_allocated + alloc_bytes < div_factor(num_bytes, 8))
Yan, Zheng424499d2010-05-16 10:46:25 -04003195 return 0;
3196
Chris Masone5bc2452010-10-26 13:37:56 -04003197 thresh = btrfs_super_total_bytes(&root->fs_info->super_copy);
Chris Mason0e4f8f82011-04-15 16:05:44 -04003198
3199 /* 256MB or 5% of the FS */
Chris Masone5bc2452010-10-26 13:37:56 -04003200 thresh = max_t(u64, 256 * 1024 * 1024, div_factor_fine(thresh, 5));
3201
3202 if (num_bytes > thresh && sinfo->bytes_used < div_factor(num_bytes, 3))
Josef Bacik14ed0ca2010-10-15 15:23:48 -04003203 return 0;
Yan, Zheng424499d2010-05-16 10:46:25 -04003204 return 1;
3205}
3206
Chris Mason6324fbf2008-03-24 15:01:59 -04003207static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3208 struct btrfs_root *extent_root, u64 alloc_bytes,
Chris Mason0ef3e662008-05-24 14:04:53 -04003209 u64 flags, int force)
Chris Mason6324fbf2008-03-24 15:01:59 -04003210{
3211 struct btrfs_space_info *space_info;
Josef Bacik97e728d2009-04-21 17:40:57 -04003212 struct btrfs_fs_info *fs_info = extent_root->fs_info;
Josef Bacik6d741192011-04-11 20:20:11 -04003213 int wait_for_alloc = 0;
Yan Zhengc146afa2008-11-12 14:34:12 -05003214 int ret = 0;
3215
Yan Zheng2b820322008-11-17 21:11:30 -05003216 flags = btrfs_reduce_alloc_profile(extent_root, flags);
Chris Masonec44a352008-04-28 15:29:52 -04003217
Chris Mason6324fbf2008-03-24 15:01:59 -04003218 space_info = __find_space_info(extent_root->fs_info, flags);
Chris Mason593060d2008-03-25 16:50:33 -04003219 if (!space_info) {
3220 ret = update_space_info(extent_root->fs_info, flags,
3221 0, 0, &space_info);
3222 BUG_ON(ret);
3223 }
Chris Mason6324fbf2008-03-24 15:01:59 -04003224 BUG_ON(!space_info);
3225
Josef Bacik6d741192011-04-11 20:20:11 -04003226again:
Josef Bacik25179202008-10-29 14:49:05 -04003227 spin_lock(&space_info->lock);
Josef Bacik9ed74f22009-09-11 16:12:44 -04003228 if (space_info->force_alloc)
Chris Mason0e4f8f82011-04-15 16:05:44 -04003229 force = space_info->force_alloc;
Josef Bacik25179202008-10-29 14:49:05 -04003230 if (space_info->full) {
3231 spin_unlock(&space_info->lock);
Josef Bacik6d741192011-04-11 20:20:11 -04003232 return 0;
Josef Bacik25179202008-10-29 14:49:05 -04003233 }
Chris Mason6324fbf2008-03-24 15:01:59 -04003234
Chris Mason0e4f8f82011-04-15 16:05:44 -04003235 if (!should_alloc_chunk(extent_root, space_info, alloc_bytes, force)) {
Josef Bacik25179202008-10-29 14:49:05 -04003236 spin_unlock(&space_info->lock);
Josef Bacik6d741192011-04-11 20:20:11 -04003237 return 0;
3238 } else if (space_info->chunk_alloc) {
3239 wait_for_alloc = 1;
3240 } else {
3241 space_info->chunk_alloc = 1;
Josef Bacik25179202008-10-29 14:49:05 -04003242 }
Chris Mason0e4f8f82011-04-15 16:05:44 -04003243
Josef Bacik25179202008-10-29 14:49:05 -04003244 spin_unlock(&space_info->lock);
3245
Josef Bacik6d741192011-04-11 20:20:11 -04003246 mutex_lock(&fs_info->chunk_mutex);
3247
3248 /*
3249 * The chunk_mutex is held throughout the entirety of a chunk
3250 * allocation, so once we've acquired the chunk_mutex we know that the
3251 * other guy is done and we need to recheck and see if we should
3252 * allocate.
3253 */
3254 if (wait_for_alloc) {
3255 mutex_unlock(&fs_info->chunk_mutex);
3256 wait_for_alloc = 0;
3257 goto again;
3258 }
3259
Josef Bacik97e728d2009-04-21 17:40:57 -04003260 /*
Josef Bacik67377732010-09-16 16:19:09 -04003261 * If we have mixed data/metadata chunks we want to make sure we keep
3262 * allocating mixed chunks instead of individual chunks.
3263 */
3264 if (btrfs_mixed_space_info(space_info))
3265 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3266
3267 /*
Josef Bacik97e728d2009-04-21 17:40:57 -04003268 * if we're doing a data chunk, go ahead and make sure that
3269 * we keep a reasonable number of metadata chunks allocated in the
3270 * FS as well.
3271 */
Josef Bacik9ed74f22009-09-11 16:12:44 -04003272 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
Josef Bacik97e728d2009-04-21 17:40:57 -04003273 fs_info->data_chunk_allocations++;
3274 if (!(fs_info->data_chunk_allocations %
3275 fs_info->metadata_ratio))
3276 force_metadata_allocation(fs_info);
3277 }
3278
Yan Zheng2b820322008-11-17 21:11:30 -05003279 ret = btrfs_alloc_chunk(trans, extent_root, flags);
Josef Bacik9ed74f22009-09-11 16:12:44 -04003280 spin_lock(&space_info->lock);
Chris Masond3977122009-01-05 21:25:51 -05003281 if (ret)
Chris Mason6324fbf2008-03-24 15:01:59 -04003282 space_info->full = 1;
Yan, Zheng424499d2010-05-16 10:46:25 -04003283 else
3284 ret = 1;
Josef Bacik6d741192011-04-11 20:20:11 -04003285
Chris Mason0e4f8f82011-04-15 16:05:44 -04003286 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
Josef Bacik6d741192011-04-11 20:20:11 -04003287 space_info->chunk_alloc = 0;
Josef Bacik9ed74f22009-09-11 16:12:44 -04003288 spin_unlock(&space_info->lock);
Yan Zhengc146afa2008-11-12 14:34:12 -05003289 mutex_unlock(&extent_root->fs_info->chunk_mutex);
Josef Bacik0f9dd462008-09-23 13:14:11 -04003290 return ret;
Chris Mason6324fbf2008-03-24 15:01:59 -04003291}
3292
Yan, Zheng5da9d012010-05-16 10:46:25 -04003293/*
3294 * shrink metadata reservation for delalloc
3295 */
3296static int shrink_delalloc(struct btrfs_trans_handle *trans,
Josef Bacik0019f102010-10-15 15:18:40 -04003297 struct btrfs_root *root, u64 to_reclaim, int sync)
Yan, Zheng5da9d012010-05-16 10:46:25 -04003298{
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04003299 struct btrfs_block_rsv *block_rsv;
Josef Bacik0019f102010-10-15 15:18:40 -04003300 struct btrfs_space_info *space_info;
Yan, Zheng5da9d012010-05-16 10:46:25 -04003301 u64 reserved;
3302 u64 max_reclaim;
3303 u64 reclaimed = 0;
Josef Bacikb1953bc2011-01-21 21:10:01 +00003304 long time_left;
Chris Masonbf9022e2010-10-26 13:40:45 -04003305 int nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
Josef Bacikb1953bc2011-01-21 21:10:01 +00003306 int loops = 0;
Chris Mason36e39c42011-03-12 07:08:42 -05003307 unsigned long progress;
Yan, Zheng5da9d012010-05-16 10:46:25 -04003308
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04003309 block_rsv = &root->fs_info->delalloc_block_rsv;
Josef Bacik0019f102010-10-15 15:18:40 -04003310 space_info = block_rsv->space_info;
Chris Masonbf9022e2010-10-26 13:40:45 -04003311
3312 smp_mb();
Josef Bacik0019f102010-10-15 15:18:40 -04003313 reserved = space_info->bytes_reserved;
Chris Mason36e39c42011-03-12 07:08:42 -05003314 progress = space_info->reservation_progress;
Yan, Zheng5da9d012010-05-16 10:46:25 -04003315
3316 if (reserved == 0)
3317 return 0;
3318
Josef Bacikfdb5eff2011-06-07 16:07:44 -04003319 smp_mb();
3320 if (root->fs_info->delalloc_bytes == 0) {
3321 if (trans)
3322 return 0;
3323 btrfs_wait_ordered_extents(root, 0, 0);
3324 return 0;
3325 }
3326
Yan, Zheng5da9d012010-05-16 10:46:25 -04003327 max_reclaim = min(reserved, to_reclaim);
3328
Josef Bacikb1953bc2011-01-21 21:10:01 +00003329 while (loops < 1024) {
Chris Masonbf9022e2010-10-26 13:40:45 -04003330 /* have the flusher threads jump in and do some IO */
3331 smp_mb();
3332 nr_pages = min_t(unsigned long, nr_pages,
3333 root->fs_info->delalloc_bytes >> PAGE_CACHE_SHIFT);
3334 writeback_inodes_sb_nr_if_idle(root->fs_info->sb, nr_pages);
Yan, Zheng5da9d012010-05-16 10:46:25 -04003335
Josef Bacik0019f102010-10-15 15:18:40 -04003336 spin_lock(&space_info->lock);
Chris Mason36e39c42011-03-12 07:08:42 -05003337 if (reserved > space_info->bytes_reserved)
Josef Bacik0019f102010-10-15 15:18:40 -04003338 reclaimed += reserved - space_info->bytes_reserved;
3339 reserved = space_info->bytes_reserved;
3340 spin_unlock(&space_info->lock);
Yan, Zheng5da9d012010-05-16 10:46:25 -04003341
Chris Mason36e39c42011-03-12 07:08:42 -05003342 loops++;
3343
Yan, Zheng5da9d012010-05-16 10:46:25 -04003344 if (reserved == 0 || reclaimed >= max_reclaim)
3345 break;
3346
3347 if (trans && trans->transaction->blocked)
3348 return -EAGAIN;
Chris Masonbf9022e2010-10-26 13:40:45 -04003349
Chris Mason36e39c42011-03-12 07:08:42 -05003350 time_left = schedule_timeout_interruptible(1);
Josef Bacikb1953bc2011-01-21 21:10:01 +00003351
3352 /* We were interrupted, exit */
3353 if (time_left)
3354 break;
3355
Chris Mason36e39c42011-03-12 07:08:42 -05003356 /* we've kicked the IO a few times, if anything has been freed,
3357 * exit. There is no sense in looping here for a long time
3358 * when we really need to commit the transaction, or there are
3359 * just too many writers without enough free space
3360 */
3361
3362 if (loops > 3) {
3363 smp_mb();
3364 if (progress != space_info->reservation_progress)
3365 break;
3366 }
Chris Masonbf9022e2010-10-26 13:40:45 -04003367
Yan, Zheng5da9d012010-05-16 10:46:25 -04003368 }
Josef Bacikfdb5eff2011-06-07 16:07:44 -04003369 if (reclaimed >= to_reclaim && !trans)
3370 btrfs_wait_ordered_extents(root, 0, 0);
Yan, Zheng5da9d012010-05-16 10:46:25 -04003371 return reclaimed >= to_reclaim;
3372}
3373
Josef Bacik8bb8ab22010-10-15 16:52:49 -04003374/*
3375 * Retries tells us how many times we've called reserve_metadata_bytes. The
3376 * idea is if this is the first call (retries == 0) then we will add to our
3377 * reserved count if we can't make the allocation in order to hold our place
3378 * while we go and try and free up space. That way for retries > 1 we don't try
3379 * and add space, we just check to see if the amount of unused space is >= the
3380 * total space, meaning that our reservation is valid.
3381 *
3382 * However if we don't intend to retry this reservation, pass -1 as retries so
3383 * that it short circuits this logic.
3384 */
3385static int reserve_metadata_bytes(struct btrfs_trans_handle *trans,
3386 struct btrfs_root *root,
3387 struct btrfs_block_rsv *block_rsv,
3388 u64 orig_bytes, int flush)
Yan, Zhengf0486c62010-05-16 10:46:25 -04003389{
3390 struct btrfs_space_info *space_info = block_rsv->space_info;
3391 u64 unused;
Josef Bacik8bb8ab22010-10-15 16:52:49 -04003392 u64 num_bytes = orig_bytes;
3393 int retries = 0;
3394 int ret = 0;
Josef Bacik38227932010-10-26 12:52:53 -04003395 bool committed = false;
Josef Bacikfdb5eff2011-06-07 16:07:44 -04003396 bool flushing = false;
Josef Bacik8bb8ab22010-10-15 16:52:49 -04003397
3398again:
Josef Bacikfdb5eff2011-06-07 16:07:44 -04003399 ret = 0;
Yan, Zhengf0486c62010-05-16 10:46:25 -04003400 spin_lock(&space_info->lock);
Josef Bacikfdb5eff2011-06-07 16:07:44 -04003401 /*
3402 * We only want to wait if somebody other than us is flushing and we are
3403 * actually alloed to flush.
3404 */
3405 while (flush && !flushing && space_info->flush) {
3406 spin_unlock(&space_info->lock);
3407 /*
3408 * If we have a trans handle we can't wait because the flusher
3409 * may have to commit the transaction, which would mean we would
3410 * deadlock since we are waiting for the flusher to finish, but
3411 * hold the current transaction open.
3412 */
3413 if (trans)
3414 return -EAGAIN;
3415 ret = wait_event_interruptible(space_info->wait,
3416 !space_info->flush);
3417 /* Must have been interrupted, return */
3418 if (ret)
3419 return -EINTR;
3420
3421 spin_lock(&space_info->lock);
3422 }
3423
3424 ret = -ENOSPC;
Yan, Zhengf0486c62010-05-16 10:46:25 -04003425 unused = space_info->bytes_used + space_info->bytes_reserved +
Josef Bacik6d487552010-10-15 15:13:32 -04003426 space_info->bytes_pinned + space_info->bytes_readonly +
3427 space_info->bytes_may_use;
Yan, Zhengf0486c62010-05-16 10:46:25 -04003428
Josef Bacik8bb8ab22010-10-15 16:52:49 -04003429 /*
3430 * The idea here is that we've not already over-reserved the block group
3431 * then we can go ahead and save our reservation first and then start
3432 * flushing if we need to. Otherwise if we've already overcommitted
3433 * lets start flushing stuff first and then come back and try to make
3434 * our reservation.
3435 */
3436 if (unused <= space_info->total_bytes) {
Arne Jansen6f334342010-11-12 23:17:56 +00003437 unused = space_info->total_bytes - unused;
Josef Bacik8bb8ab22010-10-15 16:52:49 -04003438 if (unused >= num_bytes) {
Josef Bacikfdb5eff2011-06-07 16:07:44 -04003439 space_info->bytes_reserved += orig_bytes;
Yan, Zhengf0486c62010-05-16 10:46:25 -04003440 ret = 0;
3441 } else {
Josef Bacik8bb8ab22010-10-15 16:52:49 -04003442 /*
3443 * Ok set num_bytes to orig_bytes since we aren't
3444 * overocmmitted, this way we only try and reclaim what
3445 * we need.
3446 */
3447 num_bytes = orig_bytes;
Yan, Zhengf0486c62010-05-16 10:46:25 -04003448 }
Josef Bacik8bb8ab22010-10-15 16:52:49 -04003449 } else {
3450 /*
3451 * Ok we're over committed, set num_bytes to the overcommitted
3452 * amount plus the amount of bytes that we need for this
3453 * reservation.
3454 */
3455 num_bytes = unused - space_info->total_bytes +
3456 (orig_bytes * (retries + 1));
Yan, Zhengf0486c62010-05-16 10:46:25 -04003457 }
Josef Bacik8bb8ab22010-10-15 16:52:49 -04003458
3459 /*
3460 * Couldn't make our reservation, save our place so while we're trying
3461 * to reclaim space we can actually use it instead of somebody else
3462 * stealing it from us.
3463 */
Josef Bacikfdb5eff2011-06-07 16:07:44 -04003464 if (ret && flush) {
3465 flushing = true;
3466 space_info->flush = 1;
Josef Bacik8bb8ab22010-10-15 16:52:49 -04003467 }
3468
Yan, Zhengf0486c62010-05-16 10:46:25 -04003469 spin_unlock(&space_info->lock);
3470
Josef Bacikfdb5eff2011-06-07 16:07:44 -04003471 if (!ret || !flush)
Josef Bacik8bb8ab22010-10-15 16:52:49 -04003472 goto out;
3473
3474 /*
3475 * We do synchronous shrinking since we don't actually unreserve
3476 * metadata until after the IO is completed.
3477 */
3478 ret = shrink_delalloc(trans, root, num_bytes, 1);
Josef Bacikfdb5eff2011-06-07 16:07:44 -04003479 if (ret < 0)
Josef Bacik8bb8ab22010-10-15 16:52:49 -04003480 goto out;
3481
3482 /*
3483 * So if we were overcommitted it's possible that somebody else flushed
3484 * out enough space and we simply didn't have enough space to reclaim,
3485 * so go back around and try again.
3486 */
3487 if (retries < 2) {
3488 retries++;
3489 goto again;
3490 }
3491
Josef Bacik8bb8ab22010-10-15 16:52:49 -04003492 /*
3493 * Not enough space to be reclaimed, don't bother committing the
3494 * transaction.
3495 */
Josef Bacikfdb5eff2011-06-07 16:07:44 -04003496 spin_lock(&space_info->lock);
Josef Bacik8bb8ab22010-10-15 16:52:49 -04003497 if (space_info->bytes_pinned < orig_bytes)
3498 ret = -ENOSPC;
3499 spin_unlock(&space_info->lock);
3500 if (ret)
3501 goto out;
3502
3503 ret = -EAGAIN;
Josef Bacik38227932010-10-26 12:52:53 -04003504 if (trans || committed)
Josef Bacik8bb8ab22010-10-15 16:52:49 -04003505 goto out;
3506
Josef Bacik8bb8ab22010-10-15 16:52:49 -04003507 ret = -ENOSPC;
Josef Bacik7a7eaa42011-04-13 12:54:33 -04003508 trans = btrfs_join_transaction(root);
Josef Bacik8bb8ab22010-10-15 16:52:49 -04003509 if (IS_ERR(trans))
3510 goto out;
3511 ret = btrfs_commit_transaction(trans, root);
Josef Bacik38227932010-10-26 12:52:53 -04003512 if (!ret) {
3513 trans = NULL;
3514 committed = true;
Josef Bacik8bb8ab22010-10-15 16:52:49 -04003515 goto again;
Josef Bacik38227932010-10-26 12:52:53 -04003516 }
Josef Bacik8bb8ab22010-10-15 16:52:49 -04003517
3518out:
Josef Bacikfdb5eff2011-06-07 16:07:44 -04003519 if (flushing) {
Josef Bacik8bb8ab22010-10-15 16:52:49 -04003520 spin_lock(&space_info->lock);
Josef Bacikfdb5eff2011-06-07 16:07:44 -04003521 space_info->flush = 0;
3522 wake_up_all(&space_info->wait);
Josef Bacik8bb8ab22010-10-15 16:52:49 -04003523 spin_unlock(&space_info->lock);
3524 }
Yan, Zhengf0486c62010-05-16 10:46:25 -04003525 return ret;
3526}
3527
3528static struct btrfs_block_rsv *get_block_rsv(struct btrfs_trans_handle *trans,
3529 struct btrfs_root *root)
3530{
3531 struct btrfs_block_rsv *block_rsv;
3532 if (root->ref_cows)
3533 block_rsv = trans->block_rsv;
3534 else
3535 block_rsv = root->block_rsv;
3536
3537 if (!block_rsv)
3538 block_rsv = &root->fs_info->empty_block_rsv;
3539
3540 return block_rsv;
3541}
3542
3543static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
3544 u64 num_bytes)
3545{
3546 int ret = -ENOSPC;
3547 spin_lock(&block_rsv->lock);
3548 if (block_rsv->reserved >= num_bytes) {
3549 block_rsv->reserved -= num_bytes;
3550 if (block_rsv->reserved < block_rsv->size)
3551 block_rsv->full = 0;
3552 ret = 0;
3553 }
3554 spin_unlock(&block_rsv->lock);
3555 return ret;
3556}
3557
3558static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
3559 u64 num_bytes, int update_size)
3560{
3561 spin_lock(&block_rsv->lock);
3562 block_rsv->reserved += num_bytes;
3563 if (update_size)
3564 block_rsv->size += num_bytes;
3565 else if (block_rsv->reserved >= block_rsv->size)
3566 block_rsv->full = 1;
3567 spin_unlock(&block_rsv->lock);
3568}
3569
David Sterba62a45b62011-04-20 15:52:26 +02003570static void block_rsv_release_bytes(struct btrfs_block_rsv *block_rsv,
3571 struct btrfs_block_rsv *dest, u64 num_bytes)
Yan, Zhengf0486c62010-05-16 10:46:25 -04003572{
3573 struct btrfs_space_info *space_info = block_rsv->space_info;
3574
3575 spin_lock(&block_rsv->lock);
3576 if (num_bytes == (u64)-1)
3577 num_bytes = block_rsv->size;
3578 block_rsv->size -= num_bytes;
3579 if (block_rsv->reserved >= block_rsv->size) {
3580 num_bytes = block_rsv->reserved - block_rsv->size;
3581 block_rsv->reserved = block_rsv->size;
3582 block_rsv->full = 1;
3583 } else {
3584 num_bytes = 0;
3585 }
3586 spin_unlock(&block_rsv->lock);
3587
3588 if (num_bytes > 0) {
3589 if (dest) {
Josef Bacike9e22892011-01-24 21:43:19 +00003590 spin_lock(&dest->lock);
3591 if (!dest->full) {
3592 u64 bytes_to_add;
3593
3594 bytes_to_add = dest->size - dest->reserved;
3595 bytes_to_add = min(num_bytes, bytes_to_add);
3596 dest->reserved += bytes_to_add;
3597 if (dest->reserved >= dest->size)
3598 dest->full = 1;
3599 num_bytes -= bytes_to_add;
3600 }
3601 spin_unlock(&dest->lock);
3602 }
3603 if (num_bytes) {
Yan, Zhengf0486c62010-05-16 10:46:25 -04003604 spin_lock(&space_info->lock);
3605 space_info->bytes_reserved -= num_bytes;
Chris Mason36e39c42011-03-12 07:08:42 -05003606 space_info->reservation_progress++;
Yan, Zhengf0486c62010-05-16 10:46:25 -04003607 spin_unlock(&space_info->lock);
3608 }
3609 }
3610}
3611
3612static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
3613 struct btrfs_block_rsv *dst, u64 num_bytes)
3614{
3615 int ret;
3616
3617 ret = block_rsv_use_bytes(src, num_bytes);
3618 if (ret)
3619 return ret;
3620
3621 block_rsv_add_bytes(dst, num_bytes, 1);
3622 return 0;
3623}
3624
3625void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv)
3626{
3627 memset(rsv, 0, sizeof(*rsv));
3628 spin_lock_init(&rsv->lock);
3629 atomic_set(&rsv->usage, 1);
3630 rsv->priority = 6;
3631 INIT_LIST_HEAD(&rsv->list);
3632}
3633
3634struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root)
3635{
3636 struct btrfs_block_rsv *block_rsv;
3637 struct btrfs_fs_info *fs_info = root->fs_info;
Yan, Zhengf0486c62010-05-16 10:46:25 -04003638
3639 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
3640 if (!block_rsv)
3641 return NULL;
3642
3643 btrfs_init_block_rsv(block_rsv);
Yan, Zhengf0486c62010-05-16 10:46:25 -04003644 block_rsv->space_info = __find_space_info(fs_info,
3645 BTRFS_BLOCK_GROUP_METADATA);
Yan, Zhengf0486c62010-05-16 10:46:25 -04003646 return block_rsv;
3647}
3648
3649void btrfs_free_block_rsv(struct btrfs_root *root,
3650 struct btrfs_block_rsv *rsv)
3651{
3652 if (rsv && atomic_dec_and_test(&rsv->usage)) {
3653 btrfs_block_rsv_release(root, rsv, (u64)-1);
3654 if (!rsv->durable)
3655 kfree(rsv);
3656 }
3657}
3658
3659/*
3660 * make the block_rsv struct be able to capture freed space.
3661 * the captured space will re-add to the the block_rsv struct
3662 * after transaction commit
3663 */
3664void btrfs_add_durable_block_rsv(struct btrfs_fs_info *fs_info,
3665 struct btrfs_block_rsv *block_rsv)
3666{
3667 block_rsv->durable = 1;
3668 mutex_lock(&fs_info->durable_block_rsv_mutex);
3669 list_add_tail(&block_rsv->list, &fs_info->durable_block_rsv_list);
3670 mutex_unlock(&fs_info->durable_block_rsv_mutex);
3671}
3672
3673int btrfs_block_rsv_add(struct btrfs_trans_handle *trans,
3674 struct btrfs_root *root,
3675 struct btrfs_block_rsv *block_rsv,
Josef Bacik8bb8ab22010-10-15 16:52:49 -04003676 u64 num_bytes)
Yan, Zhengf0486c62010-05-16 10:46:25 -04003677{
3678 int ret;
3679
3680 if (num_bytes == 0)
3681 return 0;
Josef Bacik8bb8ab22010-10-15 16:52:49 -04003682
3683 ret = reserve_metadata_bytes(trans, root, block_rsv, num_bytes, 1);
Yan, Zhengf0486c62010-05-16 10:46:25 -04003684 if (!ret) {
3685 block_rsv_add_bytes(block_rsv, num_bytes, 1);
3686 return 0;
3687 }
3688
Yan, Zhengf0486c62010-05-16 10:46:25 -04003689 return ret;
3690}
3691
3692int btrfs_block_rsv_check(struct btrfs_trans_handle *trans,
3693 struct btrfs_root *root,
3694 struct btrfs_block_rsv *block_rsv,
3695 u64 min_reserved, int min_factor)
3696{
3697 u64 num_bytes = 0;
3698 int commit_trans = 0;
3699 int ret = -ENOSPC;
3700
3701 if (!block_rsv)
3702 return 0;
3703
3704 spin_lock(&block_rsv->lock);
3705 if (min_factor > 0)
3706 num_bytes = div_factor(block_rsv->size, min_factor);
3707 if (min_reserved > num_bytes)
3708 num_bytes = min_reserved;
3709
3710 if (block_rsv->reserved >= num_bytes) {
3711 ret = 0;
3712 } else {
3713 num_bytes -= block_rsv->reserved;
3714 if (block_rsv->durable &&
3715 block_rsv->freed[0] + block_rsv->freed[1] >= num_bytes)
3716 commit_trans = 1;
3717 }
3718 spin_unlock(&block_rsv->lock);
3719 if (!ret)
3720 return 0;
3721
3722 if (block_rsv->refill_used) {
Josef Bacik8bb8ab22010-10-15 16:52:49 -04003723 ret = reserve_metadata_bytes(trans, root, block_rsv,
3724 num_bytes, 0);
Yan, Zhengf0486c62010-05-16 10:46:25 -04003725 if (!ret) {
3726 block_rsv_add_bytes(block_rsv, num_bytes, 0);
3727 return 0;
3728 }
3729 }
3730
3731 if (commit_trans) {
3732 if (trans)
3733 return -EAGAIN;
3734
Josef Bacik7a7eaa42011-04-13 12:54:33 -04003735 trans = btrfs_join_transaction(root);
Yan, Zhengf0486c62010-05-16 10:46:25 -04003736 BUG_ON(IS_ERR(trans));
3737 ret = btrfs_commit_transaction(trans, root);
3738 return 0;
3739 }
3740
Yan, Zhengf0486c62010-05-16 10:46:25 -04003741 return -ENOSPC;
3742}
3743
3744int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
3745 struct btrfs_block_rsv *dst_rsv,
3746 u64 num_bytes)
3747{
3748 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3749}
3750
3751void btrfs_block_rsv_release(struct btrfs_root *root,
3752 struct btrfs_block_rsv *block_rsv,
3753 u64 num_bytes)
3754{
3755 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3756 if (global_rsv->full || global_rsv == block_rsv ||
3757 block_rsv->space_info != global_rsv->space_info)
3758 global_rsv = NULL;
3759 block_rsv_release_bytes(block_rsv, global_rsv, num_bytes);
3760}
3761
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04003762/*
3763 * helper to calculate size of global block reservation.
3764 * the desired value is sum of space used by extent tree,
3765 * checksum tree and root tree
3766 */
3767static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
3768{
3769 struct btrfs_space_info *sinfo;
3770 u64 num_bytes;
3771 u64 meta_used;
3772 u64 data_used;
3773 int csum_size = btrfs_super_csum_size(&fs_info->super_copy);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04003774
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04003775 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
3776 spin_lock(&sinfo->lock);
3777 data_used = sinfo->bytes_used;
3778 spin_unlock(&sinfo->lock);
3779
3780 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3781 spin_lock(&sinfo->lock);
Josef Bacik6d487552010-10-15 15:13:32 -04003782 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
3783 data_used = 0;
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04003784 meta_used = sinfo->bytes_used;
3785 spin_unlock(&sinfo->lock);
3786
3787 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
3788 csum_size * 2;
3789 num_bytes += div64_u64(data_used + meta_used, 50);
3790
3791 if (num_bytes * 3 > meta_used)
3792 num_bytes = div64_u64(meta_used, 3);
3793
3794 return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
3795}
3796
3797static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
3798{
3799 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
3800 struct btrfs_space_info *sinfo = block_rsv->space_info;
3801 u64 num_bytes;
3802
3803 num_bytes = calc_global_metadata_size(fs_info);
3804
3805 spin_lock(&block_rsv->lock);
3806 spin_lock(&sinfo->lock);
3807
3808 block_rsv->size = num_bytes;
3809
3810 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
Josef Bacik6d487552010-10-15 15:13:32 -04003811 sinfo->bytes_reserved + sinfo->bytes_readonly +
3812 sinfo->bytes_may_use;
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04003813
3814 if (sinfo->total_bytes > num_bytes) {
3815 num_bytes = sinfo->total_bytes - num_bytes;
3816 block_rsv->reserved += num_bytes;
3817 sinfo->bytes_reserved += num_bytes;
3818 }
3819
3820 if (block_rsv->reserved >= block_rsv->size) {
3821 num_bytes = block_rsv->reserved - block_rsv->size;
3822 sinfo->bytes_reserved -= num_bytes;
Chris Mason36e39c42011-03-12 07:08:42 -05003823 sinfo->reservation_progress++;
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04003824 block_rsv->reserved = block_rsv->size;
3825 block_rsv->full = 1;
3826 }
David Sterba182608c2011-05-05 13:13:16 +02003827
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04003828 spin_unlock(&sinfo->lock);
3829 spin_unlock(&block_rsv->lock);
3830}
3831
Yan, Zhengf0486c62010-05-16 10:46:25 -04003832static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
3833{
3834 struct btrfs_space_info *space_info;
3835
3836 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3837 fs_info->chunk_block_rsv.space_info = space_info;
3838 fs_info->chunk_block_rsv.priority = 10;
3839
3840 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04003841 fs_info->global_block_rsv.space_info = space_info;
3842 fs_info->global_block_rsv.priority = 10;
3843 fs_info->global_block_rsv.refill_used = 1;
3844 fs_info->delalloc_block_rsv.space_info = space_info;
Yan, Zhengf0486c62010-05-16 10:46:25 -04003845 fs_info->trans_block_rsv.space_info = space_info;
3846 fs_info->empty_block_rsv.space_info = space_info;
3847 fs_info->empty_block_rsv.priority = 10;
3848
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04003849 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
3850 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
3851 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
3852 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
Yan, Zhengf0486c62010-05-16 10:46:25 -04003853 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04003854
3855 btrfs_add_durable_block_rsv(fs_info, &fs_info->global_block_rsv);
3856
3857 btrfs_add_durable_block_rsv(fs_info, &fs_info->delalloc_block_rsv);
3858
3859 update_global_block_rsv(fs_info);
3860}
3861
3862static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
3863{
3864 block_rsv_release_bytes(&fs_info->global_block_rsv, NULL, (u64)-1);
3865 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
3866 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
3867 WARN_ON(fs_info->trans_block_rsv.size > 0);
3868 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
3869 WARN_ON(fs_info->chunk_block_rsv.size > 0);
3870 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
Yan, Zhengf0486c62010-05-16 10:46:25 -04003871}
3872
Josef Bacikfcb80c22011-05-03 10:40:22 -04003873int btrfs_truncate_reserve_metadata(struct btrfs_trans_handle *trans,
3874 struct btrfs_root *root,
3875 struct btrfs_block_rsv *rsv)
3876{
3877 struct btrfs_block_rsv *trans_rsv = &root->fs_info->trans_block_rsv;
3878 u64 num_bytes;
3879 int ret;
3880
3881 /*
3882 * Truncate should be freeing data, but give us 2 items just in case it
3883 * needs to use some space. We may want to be smarter about this in the
3884 * future.
3885 */
Chris Masonff5714c2011-05-28 07:00:39 -04003886 num_bytes = btrfs_calc_trans_metadata_size(root, 2);
Josef Bacikfcb80c22011-05-03 10:40:22 -04003887
3888 /* We already have enough bytes, just return */
3889 if (rsv->reserved >= num_bytes)
3890 return 0;
3891
3892 num_bytes -= rsv->reserved;
3893
3894 /*
3895 * You should have reserved enough space before hand to do this, so this
3896 * should not fail.
3897 */
3898 ret = block_rsv_migrate_bytes(trans_rsv, rsv, num_bytes);
3899 BUG_ON(ret);
3900
3901 return 0;
3902}
3903
Yan, Zhenga22285a2010-05-16 10:48:46 -04003904void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
3905 struct btrfs_root *root)
3906{
3907 if (!trans->bytes_reserved)
3908 return;
3909
3910 BUG_ON(trans->block_rsv != &root->fs_info->trans_block_rsv);
3911 btrfs_block_rsv_release(root, trans->block_rsv,
3912 trans->bytes_reserved);
3913 trans->bytes_reserved = 0;
3914}
3915
Yan, Zhengd68fc572010-05-16 10:49:58 -04003916int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
3917 struct inode *inode)
3918{
3919 struct btrfs_root *root = BTRFS_I(inode)->root;
3920 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
3921 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
3922
3923 /*
Josef Bacikfcb80c22011-05-03 10:40:22 -04003924 * We need to hold space in order to delete our orphan item once we've
3925 * added it, so this takes the reservation so we can release it later
3926 * when we are truly done with the orphan item.
Yan, Zhengd68fc572010-05-16 10:49:58 -04003927 */
Chris Masonff5714c2011-05-28 07:00:39 -04003928 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
Yan, Zhengd68fc572010-05-16 10:49:58 -04003929 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3930}
3931
3932void btrfs_orphan_release_metadata(struct inode *inode)
3933{
3934 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Masonff5714c2011-05-28 07:00:39 -04003935 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
Yan, Zhengd68fc572010-05-16 10:49:58 -04003936 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
3937}
3938
Yan, Zhenga22285a2010-05-16 10:48:46 -04003939int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
3940 struct btrfs_pending_snapshot *pending)
3941{
3942 struct btrfs_root *root = pending->root;
3943 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
3944 struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
3945 /*
3946 * two for root back/forward refs, two for directory entries
3947 * and one for root of the snapshot.
3948 */
Miao Xie16cdcec2011-04-22 18:12:22 +08003949 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
Yan, Zhenga22285a2010-05-16 10:48:46 -04003950 dst_rsv->space_info = src_rsv->space_info;
3951 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3952}
3953
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04003954static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes)
3955{
3956 return num_bytes >>= 3;
3957}
3958
3959int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
3960{
3961 struct btrfs_root *root = BTRFS_I(inode)->root;
3962 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
3963 u64 to_reserve;
3964 int nr_extents;
Josef Bacik57a45ced2011-01-25 16:30:38 -05003965 int reserved_extents;
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04003966 int ret;
3967
3968 if (btrfs_transaction_in_commit(root->fs_info))
3969 schedule_timeout(1);
3970
3971 num_bytes = ALIGN(num_bytes, root->sectorsize);
Josef Bacik8bb8ab22010-10-15 16:52:49 -04003972
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04003973 nr_extents = atomic_read(&BTRFS_I(inode)->outstanding_extents) + 1;
Josef Bacik57a45ced2011-01-25 16:30:38 -05003974 reserved_extents = atomic_read(&BTRFS_I(inode)->reserved_extents);
3975
3976 if (nr_extents > reserved_extents) {
3977 nr_extents -= reserved_extents;
Miao Xie16cdcec2011-04-22 18:12:22 +08003978 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04003979 } else {
3980 nr_extents = 0;
3981 to_reserve = 0;
3982 }
Josef Bacik57a45ced2011-01-25 16:30:38 -05003983
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04003984 to_reserve += calc_csum_metadata_size(inode, num_bytes);
Josef Bacik8bb8ab22010-10-15 16:52:49 -04003985 ret = reserve_metadata_bytes(NULL, root, block_rsv, to_reserve, 1);
3986 if (ret)
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04003987 return ret;
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04003988
Josef Bacik57a45ced2011-01-25 16:30:38 -05003989 atomic_add(nr_extents, &BTRFS_I(inode)->reserved_extents);
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04003990 atomic_inc(&BTRFS_I(inode)->outstanding_extents);
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04003991
3992 block_rsv_add_bytes(block_rsv, to_reserve, 1);
3993
3994 if (block_rsv->size > 512 * 1024 * 1024)
Josef Bacik0019f102010-10-15 15:18:40 -04003995 shrink_delalloc(NULL, root, to_reserve, 0);
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04003996
3997 return 0;
3998}
3999
4000void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
4001{
4002 struct btrfs_root *root = BTRFS_I(inode)->root;
4003 u64 to_free;
4004 int nr_extents;
Josef Bacik57a45ced2011-01-25 16:30:38 -05004005 int reserved_extents;
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04004006
4007 num_bytes = ALIGN(num_bytes, root->sectorsize);
4008 atomic_dec(&BTRFS_I(inode)->outstanding_extents);
Josef Bacik3c148742011-02-02 15:53:47 +00004009 WARN_ON(atomic_read(&BTRFS_I(inode)->outstanding_extents) < 0);
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04004010
Josef Bacik57a45ced2011-01-25 16:30:38 -05004011 reserved_extents = atomic_read(&BTRFS_I(inode)->reserved_extents);
4012 do {
4013 int old, new;
4014
4015 nr_extents = atomic_read(&BTRFS_I(inode)->outstanding_extents);
4016 if (nr_extents >= reserved_extents) {
4017 nr_extents = 0;
4018 break;
4019 }
4020 old = reserved_extents;
4021 nr_extents = reserved_extents - nr_extents;
4022 new = reserved_extents - nr_extents;
4023 old = atomic_cmpxchg(&BTRFS_I(inode)->reserved_extents,
4024 reserved_extents, new);
4025 if (likely(old == reserved_extents))
4026 break;
4027 reserved_extents = old;
4028 } while (1);
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04004029
4030 to_free = calc_csum_metadata_size(inode, num_bytes);
4031 if (nr_extents > 0)
Miao Xie16cdcec2011-04-22 18:12:22 +08004032 to_free += btrfs_calc_trans_metadata_size(root, nr_extents);
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04004033
4034 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
4035 to_free);
4036}
4037
4038int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
4039{
4040 int ret;
4041
4042 ret = btrfs_check_data_free_space(inode, num_bytes);
4043 if (ret)
4044 return ret;
4045
4046 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
4047 if (ret) {
4048 btrfs_free_reserved_data_space(inode, num_bytes);
4049 return ret;
4050 }
4051
4052 return 0;
4053}
4054
4055void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
4056{
4057 btrfs_delalloc_release_metadata(inode, num_bytes);
4058 btrfs_free_reserved_data_space(inode, num_bytes);
4059}
4060
Chris Mason9078a3e2007-04-26 16:46:15 -04004061static int update_block_group(struct btrfs_trans_handle *trans,
4062 struct btrfs_root *root,
Yan, Zhengf0486c62010-05-16 10:46:25 -04004063 u64 bytenr, u64 num_bytes, int alloc)
Chris Mason9078a3e2007-04-26 16:46:15 -04004064{
Josef Bacik0af3d002010-06-21 14:48:16 -04004065 struct btrfs_block_group_cache *cache = NULL;
Chris Mason9078a3e2007-04-26 16:46:15 -04004066 struct btrfs_fs_info *info = root->fs_info;
Chris Masondb945352007-10-15 16:15:53 -04004067 u64 total = num_bytes;
Chris Mason9078a3e2007-04-26 16:46:15 -04004068 u64 old_val;
Chris Masondb945352007-10-15 16:15:53 -04004069 u64 byte_in_group;
Josef Bacik0af3d002010-06-21 14:48:16 -04004070 int factor;
Chris Mason3e1ad542007-05-07 20:03:49 -04004071
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004072 /* block accounting for super block */
4073 spin_lock(&info->delalloc_lock);
4074 old_val = btrfs_super_bytes_used(&info->super_copy);
4075 if (alloc)
4076 old_val += num_bytes;
4077 else
4078 old_val -= num_bytes;
4079 btrfs_set_super_bytes_used(&info->super_copy, old_val);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004080 spin_unlock(&info->delalloc_lock);
4081
Chris Masond3977122009-01-05 21:25:51 -05004082 while (total) {
Chris Masondb945352007-10-15 16:15:53 -04004083 cache = btrfs_lookup_block_group(info, bytenr);
Josef Bacikf3465ca2008-11-12 14:19:50 -05004084 if (!cache)
Chris Mason9078a3e2007-04-26 16:46:15 -04004085 return -1;
Yan, Zhengb742bb82010-05-16 10:46:24 -04004086 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
4087 BTRFS_BLOCK_GROUP_RAID1 |
4088 BTRFS_BLOCK_GROUP_RAID10))
4089 factor = 2;
4090 else
4091 factor = 1;
Josef Bacik9d66e232010-08-25 16:54:15 -04004092 /*
4093 * If this block group has free space cache written out, we
4094 * need to make sure to load it if we are removing space. This
4095 * is because we need the unpinning stage to actually add the
4096 * space back to the block group, otherwise we will leak space.
4097 */
4098 if (!alloc && cache->cached == BTRFS_CACHE_NO)
Josef Bacikb8399de2010-12-08 09:15:11 -05004099 cache_block_group(cache, trans, NULL, 1);
Josef Bacik0af3d002010-06-21 14:48:16 -04004100
Chris Masondb945352007-10-15 16:15:53 -04004101 byte_in_group = bytenr - cache->key.objectid;
4102 WARN_ON(byte_in_group > cache->key.offset);
Chris Mason9078a3e2007-04-26 16:46:15 -04004103
Josef Bacik25179202008-10-29 14:49:05 -04004104 spin_lock(&cache->space_info->lock);
Chris Masonc286ac42008-07-22 23:06:41 -04004105 spin_lock(&cache->lock);
Josef Bacik0af3d002010-06-21 14:48:16 -04004106
4107 if (btrfs_super_cache_generation(&info->super_copy) != 0 &&
4108 cache->disk_cache_state < BTRFS_DC_CLEAR)
4109 cache->disk_cache_state = BTRFS_DC_CLEAR;
4110
Josef Bacik0f9dd462008-09-23 13:14:11 -04004111 cache->dirty = 1;
Chris Mason9078a3e2007-04-26 16:46:15 -04004112 old_val = btrfs_block_group_used(&cache->item);
Chris Masondb945352007-10-15 16:15:53 -04004113 num_bytes = min(total, cache->key.offset - byte_in_group);
Chris Masoncd1bc462007-04-27 10:08:34 -04004114 if (alloc) {
Chris Masondb945352007-10-15 16:15:53 -04004115 old_val += num_bytes;
Yan Zheng11833d62009-09-11 16:11:19 -04004116 btrfs_set_block_group_used(&cache->item, old_val);
4117 cache->reserved -= num_bytes;
Yan Zheng11833d62009-09-11 16:11:19 -04004118 cache->space_info->bytes_reserved -= num_bytes;
Chris Mason36e39c42011-03-12 07:08:42 -05004119 cache->space_info->reservation_progress++;
Yan, Zhengb742bb82010-05-16 10:46:24 -04004120 cache->space_info->bytes_used += num_bytes;
4121 cache->space_info->disk_used += num_bytes * factor;
Chris Masonc286ac42008-07-22 23:06:41 -04004122 spin_unlock(&cache->lock);
Josef Bacik25179202008-10-29 14:49:05 -04004123 spin_unlock(&cache->space_info->lock);
Chris Masoncd1bc462007-04-27 10:08:34 -04004124 } else {
Chris Masondb945352007-10-15 16:15:53 -04004125 old_val -= num_bytes;
Chris Masonc286ac42008-07-22 23:06:41 -04004126 btrfs_set_block_group_used(&cache->item, old_val);
Yan, Zhengf0486c62010-05-16 10:46:25 -04004127 cache->pinned += num_bytes;
4128 cache->space_info->bytes_pinned += num_bytes;
Chris Mason9078a3e2007-04-26 16:46:15 -04004129 cache->space_info->bytes_used -= num_bytes;
Yan, Zhengb742bb82010-05-16 10:46:24 -04004130 cache->space_info->disk_used -= num_bytes * factor;
Chris Masonc286ac42008-07-22 23:06:41 -04004131 spin_unlock(&cache->lock);
Josef Bacik25179202008-10-29 14:49:05 -04004132 spin_unlock(&cache->space_info->lock);
Liu Hui1f3c79a2009-01-05 15:57:51 -05004133
Yan, Zhengf0486c62010-05-16 10:46:25 -04004134 set_extent_dirty(info->pinned_extents,
4135 bytenr, bytenr + num_bytes - 1,
4136 GFP_NOFS | __GFP_NOFAIL);
Chris Masoncd1bc462007-04-27 10:08:34 -04004137 }
Chris Masonfa9c0d72009-04-03 09:47:43 -04004138 btrfs_put_block_group(cache);
Chris Masondb945352007-10-15 16:15:53 -04004139 total -= num_bytes;
4140 bytenr += num_bytes;
Chris Mason9078a3e2007-04-26 16:46:15 -04004141 }
4142 return 0;
4143}
Chris Mason6324fbf2008-03-24 15:01:59 -04004144
Chris Masona061fc82008-05-07 11:43:44 -04004145static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
4146{
Josef Bacik0f9dd462008-09-23 13:14:11 -04004147 struct btrfs_block_group_cache *cache;
Yan Zhengd2fb3432008-12-11 16:30:39 -05004148 u64 bytenr;
Josef Bacik0f9dd462008-09-23 13:14:11 -04004149
4150 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
4151 if (!cache)
Chris Masona061fc82008-05-07 11:43:44 -04004152 return 0;
Josef Bacik0f9dd462008-09-23 13:14:11 -04004153
Yan Zhengd2fb3432008-12-11 16:30:39 -05004154 bytenr = cache->key.objectid;
Chris Masonfa9c0d72009-04-03 09:47:43 -04004155 btrfs_put_block_group(cache);
Yan Zhengd2fb3432008-12-11 16:30:39 -05004156
4157 return bytenr;
Chris Masona061fc82008-05-07 11:43:44 -04004158}
4159
Yan, Zhengf0486c62010-05-16 10:46:25 -04004160static int pin_down_extent(struct btrfs_root *root,
4161 struct btrfs_block_group_cache *cache,
4162 u64 bytenr, u64 num_bytes, int reserved)
Yan324ae4d2007-11-16 14:57:08 -05004163{
Yan Zheng11833d62009-09-11 16:11:19 -04004164 spin_lock(&cache->space_info->lock);
4165 spin_lock(&cache->lock);
4166 cache->pinned += num_bytes;
4167 cache->space_info->bytes_pinned += num_bytes;
4168 if (reserved) {
4169 cache->reserved -= num_bytes;
4170 cache->space_info->bytes_reserved -= num_bytes;
Chris Mason36e39c42011-03-12 07:08:42 -05004171 cache->space_info->reservation_progress++;
Yan324ae4d2007-11-16 14:57:08 -05004172 }
Yan Zheng11833d62009-09-11 16:11:19 -04004173 spin_unlock(&cache->lock);
4174 spin_unlock(&cache->space_info->lock);
4175
Yan, Zhengf0486c62010-05-16 10:46:25 -04004176 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
4177 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
Yan324ae4d2007-11-16 14:57:08 -05004178 return 0;
4179}
Chris Mason9078a3e2007-04-26 16:46:15 -04004180
Yan, Zhengf0486c62010-05-16 10:46:25 -04004181/*
4182 * this function must be called within transaction
4183 */
4184int btrfs_pin_extent(struct btrfs_root *root,
4185 u64 bytenr, u64 num_bytes, int reserved)
Zheng Yane8569812008-09-26 10:05:48 -04004186{
Yan, Zhengf0486c62010-05-16 10:46:25 -04004187 struct btrfs_block_group_cache *cache;
4188
4189 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4190 BUG_ON(!cache);
4191
4192 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
4193
4194 btrfs_put_block_group(cache);
Yan Zheng11833d62009-09-11 16:11:19 -04004195 return 0;
4196}
Zheng Yane8569812008-09-26 10:05:48 -04004197
Yan, Zhengf0486c62010-05-16 10:46:25 -04004198/*
4199 * update size of reserved extents. this function may return -EAGAIN
4200 * if 'reserve' is true or 'sinfo' is false.
4201 */
Li Dongyangb4d00d52011-03-24 10:24:25 +00004202int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
4203 u64 num_bytes, int reserve, int sinfo)
Yan, Zhengf0486c62010-05-16 10:46:25 -04004204{
4205 int ret = 0;
4206 if (sinfo) {
4207 struct btrfs_space_info *space_info = cache->space_info;
4208 spin_lock(&space_info->lock);
4209 spin_lock(&cache->lock);
4210 if (reserve) {
4211 if (cache->ro) {
4212 ret = -EAGAIN;
4213 } else {
4214 cache->reserved += num_bytes;
4215 space_info->bytes_reserved += num_bytes;
4216 }
4217 } else {
4218 if (cache->ro)
4219 space_info->bytes_readonly += num_bytes;
4220 cache->reserved -= num_bytes;
4221 space_info->bytes_reserved -= num_bytes;
Chris Mason36e39c42011-03-12 07:08:42 -05004222 space_info->reservation_progress++;
Yan, Zhengf0486c62010-05-16 10:46:25 -04004223 }
4224 spin_unlock(&cache->lock);
4225 spin_unlock(&space_info->lock);
4226 } else {
4227 spin_lock(&cache->lock);
4228 if (cache->ro) {
4229 ret = -EAGAIN;
4230 } else {
4231 if (reserve)
4232 cache->reserved += num_bytes;
4233 else
4234 cache->reserved -= num_bytes;
4235 }
4236 spin_unlock(&cache->lock);
4237 }
4238 return ret;
4239}
4240
Yan Zheng11833d62009-09-11 16:11:19 -04004241int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
4242 struct btrfs_root *root)
4243{
4244 struct btrfs_fs_info *fs_info = root->fs_info;
4245 struct btrfs_caching_control *next;
4246 struct btrfs_caching_control *caching_ctl;
4247 struct btrfs_block_group_cache *cache;
4248
4249 down_write(&fs_info->extent_commit_sem);
4250
4251 list_for_each_entry_safe(caching_ctl, next,
4252 &fs_info->caching_block_groups, list) {
4253 cache = caching_ctl->block_group;
4254 if (block_group_cache_done(cache)) {
4255 cache->last_byte_to_unpin = (u64)-1;
4256 list_del_init(&caching_ctl->list);
4257 put_caching_control(caching_ctl);
4258 } else {
4259 cache->last_byte_to_unpin = caching_ctl->progress;
4260 }
4261 }
4262
4263 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4264 fs_info->pinned_extents = &fs_info->freed_extents[1];
4265 else
4266 fs_info->pinned_extents = &fs_info->freed_extents[0];
4267
4268 up_write(&fs_info->extent_commit_sem);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04004269
4270 update_global_block_rsv(fs_info);
Yan Zheng11833d62009-09-11 16:11:19 -04004271 return 0;
4272}
4273
4274static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
4275{
4276 struct btrfs_fs_info *fs_info = root->fs_info;
4277 struct btrfs_block_group_cache *cache = NULL;
4278 u64 len;
4279
4280 while (start <= end) {
4281 if (!cache ||
4282 start >= cache->key.objectid + cache->key.offset) {
4283 if (cache)
4284 btrfs_put_block_group(cache);
4285 cache = btrfs_lookup_block_group(fs_info, start);
4286 BUG_ON(!cache);
4287 }
4288
4289 len = cache->key.objectid + cache->key.offset - start;
4290 len = min(len, end + 1 - start);
4291
4292 if (start < cache->last_byte_to_unpin) {
4293 len = min(len, cache->last_byte_to_unpin - start);
4294 btrfs_add_free_space(cache, start, len);
4295 }
Josef Bacik25179202008-10-29 14:49:05 -04004296
Yan, Zhengf0486c62010-05-16 10:46:25 -04004297 start += len;
4298
Josef Bacik25179202008-10-29 14:49:05 -04004299 spin_lock(&cache->space_info->lock);
4300 spin_lock(&cache->lock);
Yan Zheng11833d62009-09-11 16:11:19 -04004301 cache->pinned -= len;
4302 cache->space_info->bytes_pinned -= len;
Yan, Zhengf0486c62010-05-16 10:46:25 -04004303 if (cache->ro) {
4304 cache->space_info->bytes_readonly += len;
4305 } else if (cache->reserved_pinned > 0) {
4306 len = min(len, cache->reserved_pinned);
4307 cache->reserved_pinned -= len;
4308 cache->space_info->bytes_reserved += len;
4309 }
Josef Bacik25179202008-10-29 14:49:05 -04004310 spin_unlock(&cache->lock);
4311 spin_unlock(&cache->space_info->lock);
Yan Zheng11833d62009-09-11 16:11:19 -04004312 }
4313
4314 if (cache)
Chris Masonfa9c0d72009-04-03 09:47:43 -04004315 btrfs_put_block_group(cache);
Chris Masonccd467d2007-06-28 15:57:36 -04004316 return 0;
4317}
4318
4319int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
Yan Zheng11833d62009-09-11 16:11:19 -04004320 struct btrfs_root *root)
Chris Masona28ec192007-03-06 20:08:01 -05004321{
Yan Zheng11833d62009-09-11 16:11:19 -04004322 struct btrfs_fs_info *fs_info = root->fs_info;
4323 struct extent_io_tree *unpin;
Yan, Zhengf0486c62010-05-16 10:46:25 -04004324 struct btrfs_block_rsv *block_rsv;
4325 struct btrfs_block_rsv *next_rsv;
Chris Mason1a5bc162007-10-15 16:15:26 -04004326 u64 start;
4327 u64 end;
Yan, Zhengf0486c62010-05-16 10:46:25 -04004328 int idx;
Chris Masona28ec192007-03-06 20:08:01 -05004329 int ret;
Chris Masona28ec192007-03-06 20:08:01 -05004330
Yan Zheng11833d62009-09-11 16:11:19 -04004331 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4332 unpin = &fs_info->freed_extents[1];
4333 else
4334 unpin = &fs_info->freed_extents[0];
4335
Chris Masond3977122009-01-05 21:25:51 -05004336 while (1) {
Chris Mason1a5bc162007-10-15 16:15:26 -04004337 ret = find_first_extent_bit(unpin, 0, &start, &end,
4338 EXTENT_DIRTY);
4339 if (ret)
Chris Masona28ec192007-03-06 20:08:01 -05004340 break;
Liu Hui1f3c79a2009-01-05 15:57:51 -05004341
Li Dongyang5378e602011-03-24 10:24:27 +00004342 if (btrfs_test_opt(root, DISCARD))
4343 ret = btrfs_discard_extent(root, start,
4344 end + 1 - start, NULL);
Liu Hui1f3c79a2009-01-05 15:57:51 -05004345
Chris Mason1a5bc162007-10-15 16:15:26 -04004346 clear_extent_dirty(unpin, start, end, GFP_NOFS);
Yan Zheng11833d62009-09-11 16:11:19 -04004347 unpin_extent_range(root, start, end);
Chris Masonb9473432009-03-13 11:00:37 -04004348 cond_resched();
Chris Masona28ec192007-03-06 20:08:01 -05004349 }
Josef Bacik817d52f2009-07-13 21:29:25 -04004350
Yan, Zhengf0486c62010-05-16 10:46:25 -04004351 mutex_lock(&fs_info->durable_block_rsv_mutex);
4352 list_for_each_entry_safe(block_rsv, next_rsv,
4353 &fs_info->durable_block_rsv_list, list) {
Chris Masona28ec192007-03-06 20:08:01 -05004354
Yan, Zhengf0486c62010-05-16 10:46:25 -04004355 idx = trans->transid & 0x1;
4356 if (block_rsv->freed[idx] > 0) {
4357 block_rsv_add_bytes(block_rsv,
4358 block_rsv->freed[idx], 0);
4359 block_rsv->freed[idx] = 0;
Chris Mason8ef97622007-03-26 10:15:30 -04004360 }
Yan, Zhengf0486c62010-05-16 10:46:25 -04004361 if (atomic_read(&block_rsv->usage) == 0) {
4362 btrfs_block_rsv_release(root, block_rsv, (u64)-1);
Zheng Yan31840ae2008-09-23 13:14:14 -04004363
Yan, Zhengf0486c62010-05-16 10:46:25 -04004364 if (block_rsv->freed[0] == 0 &&
4365 block_rsv->freed[1] == 0) {
4366 list_del_init(&block_rsv->list);
4367 kfree(block_rsv);
4368 }
4369 } else {
4370 btrfs_block_rsv_release(root, block_rsv, 0);
4371 }
4372 }
4373 mutex_unlock(&fs_info->durable_block_rsv_mutex);
4374
Chris Masone20d96d2007-03-22 12:13:20 -04004375 return 0;
4376}
4377
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004378static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
4379 struct btrfs_root *root,
4380 u64 bytenr, u64 num_bytes, u64 parent,
4381 u64 root_objectid, u64 owner_objectid,
4382 u64 owner_offset, int refs_to_drop,
4383 struct btrfs_delayed_extent_op *extent_op)
Chris Masona28ec192007-03-06 20:08:01 -05004384{
Chris Masone2fa7222007-03-12 16:22:34 -04004385 struct btrfs_key key;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004386 struct btrfs_path *path;
Chris Mason1261ec42007-03-20 20:35:03 -04004387 struct btrfs_fs_info *info = root->fs_info;
4388 struct btrfs_root *extent_root = info->extent_root;
Chris Mason5f39d392007-10-15 16:14:19 -04004389 struct extent_buffer *leaf;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004390 struct btrfs_extent_item *ei;
4391 struct btrfs_extent_inline_ref *iref;
Chris Masona28ec192007-03-06 20:08:01 -05004392 int ret;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004393 int is_data;
Chris Mason952fcca2008-02-18 16:33:44 -05004394 int extent_slot = 0;
4395 int found_extent = 0;
4396 int num_to_del = 1;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004397 u32 item_size;
4398 u64 refs;
Chris Mason037e6392007-03-07 11:50:24 -05004399
Chris Mason5caf2a02007-04-02 11:20:42 -04004400 path = btrfs_alloc_path();
Chris Mason54aa1f42007-06-22 14:16:25 -04004401 if (!path)
4402 return -ENOMEM;
4403
Chris Mason3c12ac72008-04-21 12:01:38 -04004404 path->reada = 1;
Chris Masonb9473432009-03-13 11:00:37 -04004405 path->leave_spinning = 1;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004406
4407 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
4408 BUG_ON(!is_data && refs_to_drop != 1);
4409
4410 ret = lookup_extent_backref(trans, extent_root, path, &iref,
4411 bytenr, num_bytes, parent,
4412 root_objectid, owner_objectid,
4413 owner_offset);
Chris Mason7bb86312007-12-11 09:25:06 -05004414 if (ret == 0) {
Chris Mason952fcca2008-02-18 16:33:44 -05004415 extent_slot = path->slots[0];
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004416 while (extent_slot >= 0) {
4417 btrfs_item_key_to_cpu(path->nodes[0], &key,
Chris Mason952fcca2008-02-18 16:33:44 -05004418 extent_slot);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004419 if (key.objectid != bytenr)
Chris Mason952fcca2008-02-18 16:33:44 -05004420 break;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004421 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
4422 key.offset == num_bytes) {
Chris Mason952fcca2008-02-18 16:33:44 -05004423 found_extent = 1;
4424 break;
4425 }
4426 if (path->slots[0] - extent_slot > 5)
4427 break;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004428 extent_slot--;
Chris Mason952fcca2008-02-18 16:33:44 -05004429 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004430#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4431 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
4432 if (found_extent && item_size < sizeof(*ei))
4433 found_extent = 0;
4434#endif
Zheng Yan31840ae2008-09-23 13:14:14 -04004435 if (!found_extent) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004436 BUG_ON(iref);
Chris Mason56bec292009-03-13 10:10:06 -04004437 ret = remove_extent_backref(trans, extent_root, path,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004438 NULL, refs_to_drop,
4439 is_data);
Zheng Yan31840ae2008-09-23 13:14:14 -04004440 BUG_ON(ret);
David Sterbab3b4aa72011-04-21 01:20:15 +02004441 btrfs_release_path(path);
Chris Masonb9473432009-03-13 11:00:37 -04004442 path->leave_spinning = 1;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004443
4444 key.objectid = bytenr;
4445 key.type = BTRFS_EXTENT_ITEM_KEY;
4446 key.offset = num_bytes;
4447
Zheng Yan31840ae2008-09-23 13:14:14 -04004448 ret = btrfs_search_slot(trans, extent_root,
4449 &key, path, -1, 1);
Josef Bacikf3465ca2008-11-12 14:19:50 -05004450 if (ret) {
4451 printk(KERN_ERR "umm, got %d back from search"
Chris Masond3977122009-01-05 21:25:51 -05004452 ", was looking for %llu\n", ret,
4453 (unsigned long long)bytenr);
Josef Bacikf3465ca2008-11-12 14:19:50 -05004454 btrfs_print_leaf(extent_root, path->nodes[0]);
4455 }
Zheng Yan31840ae2008-09-23 13:14:14 -04004456 BUG_ON(ret);
4457 extent_slot = path->slots[0];
4458 }
Chris Mason7bb86312007-12-11 09:25:06 -05004459 } else {
4460 btrfs_print_leaf(extent_root, path->nodes[0]);
4461 WARN_ON(1);
Chris Masond3977122009-01-05 21:25:51 -05004462 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004463 "parent %llu root %llu owner %llu offset %llu\n",
Chris Masond3977122009-01-05 21:25:51 -05004464 (unsigned long long)bytenr,
Chris Mason56bec292009-03-13 10:10:06 -04004465 (unsigned long long)parent,
Chris Masond3977122009-01-05 21:25:51 -05004466 (unsigned long long)root_objectid,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004467 (unsigned long long)owner_objectid,
4468 (unsigned long long)owner_offset);
Chris Mason7bb86312007-12-11 09:25:06 -05004469 }
Chris Mason5f39d392007-10-15 16:14:19 -04004470
4471 leaf = path->nodes[0];
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004472 item_size = btrfs_item_size_nr(leaf, extent_slot);
4473#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4474 if (item_size < sizeof(*ei)) {
4475 BUG_ON(found_extent || extent_slot != path->slots[0]);
4476 ret = convert_extent_item_v0(trans, extent_root, path,
4477 owner_objectid, 0);
4478 BUG_ON(ret < 0);
4479
David Sterbab3b4aa72011-04-21 01:20:15 +02004480 btrfs_release_path(path);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004481 path->leave_spinning = 1;
4482
4483 key.objectid = bytenr;
4484 key.type = BTRFS_EXTENT_ITEM_KEY;
4485 key.offset = num_bytes;
4486
4487 ret = btrfs_search_slot(trans, extent_root, &key, path,
4488 -1, 1);
4489 if (ret) {
4490 printk(KERN_ERR "umm, got %d back from search"
4491 ", was looking for %llu\n", ret,
4492 (unsigned long long)bytenr);
4493 btrfs_print_leaf(extent_root, path->nodes[0]);
4494 }
4495 BUG_ON(ret);
4496 extent_slot = path->slots[0];
4497 leaf = path->nodes[0];
4498 item_size = btrfs_item_size_nr(leaf, extent_slot);
4499 }
4500#endif
4501 BUG_ON(item_size < sizeof(*ei));
Chris Mason952fcca2008-02-18 16:33:44 -05004502 ei = btrfs_item_ptr(leaf, extent_slot,
Chris Mason123abc82007-03-14 14:14:43 -04004503 struct btrfs_extent_item);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004504 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4505 struct btrfs_tree_block_info *bi;
4506 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
4507 bi = (struct btrfs_tree_block_info *)(ei + 1);
4508 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
Chris Mason952fcca2008-02-18 16:33:44 -05004509 }
4510
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004511 refs = btrfs_extent_refs(leaf, ei);
4512 BUG_ON(refs < refs_to_drop);
4513 refs -= refs_to_drop;
4514
4515 if (refs > 0) {
4516 if (extent_op)
4517 __run_delayed_extent_op(extent_op, leaf, ei);
4518 /*
4519 * In the case of inline back ref, reference count will
4520 * be updated by remove_extent_backref
4521 */
4522 if (iref) {
4523 BUG_ON(!found_extent);
4524 } else {
4525 btrfs_set_extent_refs(leaf, ei, refs);
4526 btrfs_mark_buffer_dirty(leaf);
4527 }
4528 if (found_extent) {
4529 ret = remove_extent_backref(trans, extent_root, path,
4530 iref, refs_to_drop,
4531 is_data);
4532 BUG_ON(ret);
4533 }
4534 } else {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004535 if (found_extent) {
4536 BUG_ON(is_data && refs_to_drop !=
4537 extent_data_ref_count(root, path, iref));
4538 if (iref) {
4539 BUG_ON(path->slots[0] != extent_slot);
4540 } else {
4541 BUG_ON(path->slots[0] != extent_slot + 1);
4542 path->slots[0] = extent_slot;
4543 num_to_del = 2;
4544 }
Chris Mason78fae272007-03-25 11:35:08 -04004545 }
Chris Masonb9473432009-03-13 11:00:37 -04004546
Chris Mason952fcca2008-02-18 16:33:44 -05004547 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
4548 num_to_del);
Zheng Yan31840ae2008-09-23 13:14:14 -04004549 BUG_ON(ret);
David Sterbab3b4aa72011-04-21 01:20:15 +02004550 btrfs_release_path(path);
David Woodhouse21af8042008-08-12 14:13:26 +01004551
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004552 if (is_data) {
Chris Mason459931e2008-12-10 09:10:46 -05004553 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
4554 BUG_ON(ret);
Chris Masond57e62b2009-03-31 13:47:50 -04004555 } else {
4556 invalidate_mapping_pages(info->btree_inode->i_mapping,
4557 bytenr >> PAGE_CACHE_SHIFT,
4558 (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
Chris Mason459931e2008-12-10 09:10:46 -05004559 }
4560
Yan, Zhengf0486c62010-05-16 10:46:25 -04004561 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
Chris Masondcbdd4d2008-12-16 13:51:01 -05004562 BUG_ON(ret);
Chris Masona28ec192007-03-06 20:08:01 -05004563 }
Chris Mason5caf2a02007-04-02 11:20:42 -04004564 btrfs_free_path(path);
Chris Masona28ec192007-03-06 20:08:01 -05004565 return ret;
4566}
4567
4568/*
Yan, Zhengf0486c62010-05-16 10:46:25 -04004569 * when we free an block, it is possible (and likely) that we free the last
Chris Mason1887be62009-03-13 10:11:24 -04004570 * delayed ref for that extent as well. This searches the delayed ref tree for
4571 * a given extent, and if there are no other delayed refs to be processed, it
4572 * removes it from the tree.
4573 */
4574static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
4575 struct btrfs_root *root, u64 bytenr)
4576{
4577 struct btrfs_delayed_ref_head *head;
4578 struct btrfs_delayed_ref_root *delayed_refs;
4579 struct btrfs_delayed_ref_node *ref;
4580 struct rb_node *node;
Yan, Zhengf0486c62010-05-16 10:46:25 -04004581 int ret = 0;
Chris Mason1887be62009-03-13 10:11:24 -04004582
4583 delayed_refs = &trans->transaction->delayed_refs;
4584 spin_lock(&delayed_refs->lock);
4585 head = btrfs_find_delayed_ref_head(trans, bytenr);
4586 if (!head)
4587 goto out;
4588
4589 node = rb_prev(&head->node.rb_node);
4590 if (!node)
4591 goto out;
4592
4593 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
4594
4595 /* there are still entries for this ref, we can't drop it */
4596 if (ref->bytenr == bytenr)
4597 goto out;
4598
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004599 if (head->extent_op) {
4600 if (!head->must_insert_reserved)
4601 goto out;
4602 kfree(head->extent_op);
4603 head->extent_op = NULL;
4604 }
4605
Chris Mason1887be62009-03-13 10:11:24 -04004606 /*
4607 * waiting for the lock here would deadlock. If someone else has it
4608 * locked they are already in the process of dropping it anyway
4609 */
4610 if (!mutex_trylock(&head->mutex))
4611 goto out;
4612
4613 /*
4614 * at this point we have a head with no other entries. Go
4615 * ahead and process it.
4616 */
4617 head->node.in_tree = 0;
4618 rb_erase(&head->node.rb_node, &delayed_refs->root);
Chris Masonc3e69d52009-03-13 10:17:05 -04004619
Chris Mason1887be62009-03-13 10:11:24 -04004620 delayed_refs->num_entries--;
4621
4622 /*
4623 * we don't take a ref on the node because we're removing it from the
4624 * tree, so we just steal the ref the tree was holding.
4625 */
Chris Masonc3e69d52009-03-13 10:17:05 -04004626 delayed_refs->num_heads--;
4627 if (list_empty(&head->cluster))
4628 delayed_refs->num_heads_ready--;
4629
4630 list_del_init(&head->cluster);
Chris Mason1887be62009-03-13 10:11:24 -04004631 spin_unlock(&delayed_refs->lock);
4632
Yan, Zhengf0486c62010-05-16 10:46:25 -04004633 BUG_ON(head->extent_op);
4634 if (head->must_insert_reserved)
4635 ret = 1;
4636
4637 mutex_unlock(&head->mutex);
Chris Mason1887be62009-03-13 10:11:24 -04004638 btrfs_put_delayed_ref(&head->node);
Yan, Zhengf0486c62010-05-16 10:46:25 -04004639 return ret;
Chris Mason1887be62009-03-13 10:11:24 -04004640out:
4641 spin_unlock(&delayed_refs->lock);
4642 return 0;
4643}
4644
Yan, Zhengf0486c62010-05-16 10:46:25 -04004645void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
4646 struct btrfs_root *root,
4647 struct extent_buffer *buf,
4648 u64 parent, int last_ref)
4649{
4650 struct btrfs_block_rsv *block_rsv;
4651 struct btrfs_block_group_cache *cache = NULL;
4652 int ret;
4653
4654 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4655 ret = btrfs_add_delayed_tree_ref(trans, buf->start, buf->len,
4656 parent, root->root_key.objectid,
4657 btrfs_header_level(buf),
4658 BTRFS_DROP_DELAYED_REF, NULL);
4659 BUG_ON(ret);
4660 }
4661
4662 if (!last_ref)
4663 return;
4664
4665 block_rsv = get_block_rsv(trans, root);
4666 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
Yan, Zheng3bf84a52010-05-31 09:04:46 +00004667 if (block_rsv->space_info != cache->space_info)
4668 goto out;
Yan, Zhengf0486c62010-05-16 10:46:25 -04004669
4670 if (btrfs_header_generation(buf) == trans->transid) {
4671 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4672 ret = check_ref_cleanup(trans, root, buf->start);
4673 if (!ret)
4674 goto pin;
4675 }
4676
4677 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
4678 pin_down_extent(root, cache, buf->start, buf->len, 1);
4679 goto pin;
4680 }
4681
4682 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
4683
4684 btrfs_add_free_space(cache, buf->start, buf->len);
Li Dongyangb4d00d52011-03-24 10:24:25 +00004685 ret = btrfs_update_reserved_bytes(cache, buf->len, 0, 0);
Yan, Zhengf0486c62010-05-16 10:46:25 -04004686 if (ret == -EAGAIN) {
4687 /* block group became read-only */
Li Dongyangb4d00d52011-03-24 10:24:25 +00004688 btrfs_update_reserved_bytes(cache, buf->len, 0, 1);
Yan, Zhengf0486c62010-05-16 10:46:25 -04004689 goto out;
4690 }
4691
4692 ret = 1;
4693 spin_lock(&block_rsv->lock);
4694 if (block_rsv->reserved < block_rsv->size) {
4695 block_rsv->reserved += buf->len;
4696 ret = 0;
4697 }
4698 spin_unlock(&block_rsv->lock);
4699
4700 if (ret) {
4701 spin_lock(&cache->space_info->lock);
4702 cache->space_info->bytes_reserved -= buf->len;
Chris Mason36e39c42011-03-12 07:08:42 -05004703 cache->space_info->reservation_progress++;
Yan, Zhengf0486c62010-05-16 10:46:25 -04004704 spin_unlock(&cache->space_info->lock);
4705 }
4706 goto out;
4707 }
4708pin:
4709 if (block_rsv->durable && !cache->ro) {
4710 ret = 0;
4711 spin_lock(&cache->lock);
4712 if (!cache->ro) {
4713 cache->reserved_pinned += buf->len;
4714 ret = 1;
4715 }
4716 spin_unlock(&cache->lock);
4717
4718 if (ret) {
4719 spin_lock(&block_rsv->lock);
4720 block_rsv->freed[trans->transid & 0x1] += buf->len;
4721 spin_unlock(&block_rsv->lock);
4722 }
4723 }
4724out:
Josef Bacika826d6d2011-03-16 13:42:43 -04004725 /*
4726 * Deleting the buffer, clear the corrupt flag since it doesn't matter
4727 * anymore.
4728 */
4729 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
Yan, Zhengf0486c62010-05-16 10:46:25 -04004730 btrfs_put_block_group(cache);
4731}
4732
Chris Mason925baed2008-06-25 16:01:30 -04004733int btrfs_free_extent(struct btrfs_trans_handle *trans,
Zheng Yan31840ae2008-09-23 13:14:14 -04004734 struct btrfs_root *root,
4735 u64 bytenr, u64 num_bytes, u64 parent,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004736 u64 root_objectid, u64 owner, u64 offset)
Chris Mason925baed2008-06-25 16:01:30 -04004737{
4738 int ret;
4739
Chris Mason56bec292009-03-13 10:10:06 -04004740 /*
4741 * tree log blocks never actually go into the extent allocation
4742 * tree, just update pinning info and exit early.
Chris Mason56bec292009-03-13 10:10:06 -04004743 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004744 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
4745 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
Chris Masonb9473432009-03-13 11:00:37 -04004746 /* unlocks the pinned mutex */
Yan Zheng11833d62009-09-11 16:11:19 -04004747 btrfs_pin_extent(root, bytenr, num_bytes, 1);
Chris Mason56bec292009-03-13 10:10:06 -04004748 ret = 0;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004749 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
4750 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
4751 parent, root_objectid, (int)owner,
4752 BTRFS_DROP_DELAYED_REF, NULL);
Chris Mason1887be62009-03-13 10:11:24 -04004753 BUG_ON(ret);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004754 } else {
4755 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
4756 parent, root_objectid, owner,
4757 offset, BTRFS_DROP_DELAYED_REF, NULL);
4758 BUG_ON(ret);
Chris Mason56bec292009-03-13 10:10:06 -04004759 }
Chris Mason925baed2008-06-25 16:01:30 -04004760 return ret;
4761}
4762
Chris Mason87ee04e2007-11-30 11:30:34 -05004763static u64 stripe_align(struct btrfs_root *root, u64 val)
4764{
4765 u64 mask = ((u64)root->stripesize - 1);
4766 u64 ret = (val + mask) & ~mask;
4767 return ret;
4768}
4769
Chris Masonfec577f2007-02-26 10:40:21 -05004770/*
Josef Bacik817d52f2009-07-13 21:29:25 -04004771 * when we wait for progress in the block group caching, its because
4772 * our allocation attempt failed at least once. So, we must sleep
4773 * and let some progress happen before we try again.
4774 *
4775 * This function will sleep at least once waiting for new free space to
4776 * show up, and then it will check the block group free space numbers
4777 * for our min num_bytes. Another option is to have it go ahead
4778 * and look in the rbtree for a free extent of a given size, but this
4779 * is a good start.
4780 */
4781static noinline int
4782wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
4783 u64 num_bytes)
4784{
Yan Zheng11833d62009-09-11 16:11:19 -04004785 struct btrfs_caching_control *caching_ctl;
Josef Bacik817d52f2009-07-13 21:29:25 -04004786 DEFINE_WAIT(wait);
4787
Yan Zheng11833d62009-09-11 16:11:19 -04004788 caching_ctl = get_caching_control(cache);
4789 if (!caching_ctl)
Josef Bacik817d52f2009-07-13 21:29:25 -04004790 return 0;
Josef Bacik817d52f2009-07-13 21:29:25 -04004791
Yan Zheng11833d62009-09-11 16:11:19 -04004792 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
Li Zefan34d52cb2011-03-29 13:46:06 +08004793 (cache->free_space_ctl->free_space >= num_bytes));
Yan Zheng11833d62009-09-11 16:11:19 -04004794
4795 put_caching_control(caching_ctl);
4796 return 0;
4797}
4798
4799static noinline int
4800wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
4801{
4802 struct btrfs_caching_control *caching_ctl;
4803 DEFINE_WAIT(wait);
4804
4805 caching_ctl = get_caching_control(cache);
4806 if (!caching_ctl)
4807 return 0;
4808
4809 wait_event(caching_ctl->wait, block_group_cache_done(cache));
4810
4811 put_caching_control(caching_ctl);
Josef Bacik817d52f2009-07-13 21:29:25 -04004812 return 0;
4813}
4814
Yan, Zhengb742bb82010-05-16 10:46:24 -04004815static int get_block_group_index(struct btrfs_block_group_cache *cache)
4816{
4817 int index;
4818 if (cache->flags & BTRFS_BLOCK_GROUP_RAID10)
4819 index = 0;
4820 else if (cache->flags & BTRFS_BLOCK_GROUP_RAID1)
4821 index = 1;
4822 else if (cache->flags & BTRFS_BLOCK_GROUP_DUP)
4823 index = 2;
4824 else if (cache->flags & BTRFS_BLOCK_GROUP_RAID0)
4825 index = 3;
4826 else
4827 index = 4;
4828 return index;
4829}
4830
Josef Bacik817d52f2009-07-13 21:29:25 -04004831enum btrfs_loop_type {
Josef Bacikccf0e722009-11-10 21:23:48 -05004832 LOOP_FIND_IDEAL = 0,
Josef Bacik817d52f2009-07-13 21:29:25 -04004833 LOOP_CACHING_NOWAIT = 1,
4834 LOOP_CACHING_WAIT = 2,
4835 LOOP_ALLOC_CHUNK = 3,
4836 LOOP_NO_EMPTY_SIZE = 4,
4837};
4838
4839/*
Chris Masonfec577f2007-02-26 10:40:21 -05004840 * walks the btree of allocated extents and find a hole of a given size.
4841 * The key ins is changed to record the hole:
4842 * ins->objectid == block start
Chris Mason62e27492007-03-15 12:56:47 -04004843 * ins->flags = BTRFS_EXTENT_ITEM_KEY
Chris Masonfec577f2007-02-26 10:40:21 -05004844 * ins->offset == number of blocks
4845 * Any available blocks before search_start are skipped.
4846 */
Chris Masond3977122009-01-05 21:25:51 -05004847static noinline int find_free_extent(struct btrfs_trans_handle *trans,
Chris Mason98ed5172008-01-03 10:01:48 -05004848 struct btrfs_root *orig_root,
4849 u64 num_bytes, u64 empty_size,
4850 u64 search_start, u64 search_end,
4851 u64 hint_byte, struct btrfs_key *ins,
Ilya Dryomove0f54062011-06-18 20:26:38 +00004852 u64 data)
Chris Masonfec577f2007-02-26 10:40:21 -05004853{
Josef Bacik80eb2342008-10-29 14:49:05 -04004854 int ret = 0;
Chris Masond3977122009-01-05 21:25:51 -05004855 struct btrfs_root *root = orig_root->fs_info->extent_root;
Chris Masonfa9c0d72009-04-03 09:47:43 -04004856 struct btrfs_free_cluster *last_ptr = NULL;
Josef Bacik80eb2342008-10-29 14:49:05 -04004857 struct btrfs_block_group_cache *block_group = NULL;
Chris Mason239b14b2008-03-24 15:02:07 -04004858 int empty_cluster = 2 * 1024 * 1024;
Chris Mason0ef3e662008-05-24 14:04:53 -04004859 int allowed_chunk_alloc = 0;
Josef Bacikccf0e722009-11-10 21:23:48 -05004860 int done_chunk_alloc = 0;
Josef Bacik80eb2342008-10-29 14:49:05 -04004861 struct btrfs_space_info *space_info;
Chris Masonfa9c0d72009-04-03 09:47:43 -04004862 int last_ptr_loop = 0;
4863 int loop = 0;
Yan, Zhengf0486c62010-05-16 10:46:25 -04004864 int index = 0;
Josef Bacik817d52f2009-07-13 21:29:25 -04004865 bool found_uncached_bg = false;
Josef Bacik0a243252009-09-11 16:11:20 -04004866 bool failed_cluster_refill = false;
Josef Bacik1cdda9b2009-10-06 10:04:28 -04004867 bool failed_alloc = false;
Josef Bacik67377732010-09-16 16:19:09 -04004868 bool use_cluster = true;
Josef Bacikccf0e722009-11-10 21:23:48 -05004869 u64 ideal_cache_percent = 0;
4870 u64 ideal_cache_offset = 0;
Chris Masonfec577f2007-02-26 10:40:21 -05004871
Chris Masondb945352007-10-15 16:15:53 -04004872 WARN_ON(num_bytes < root->sectorsize);
Chris Masonb1a4d962007-04-04 15:27:52 -04004873 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
Josef Bacik80eb2342008-10-29 14:49:05 -04004874 ins->objectid = 0;
4875 ins->offset = 0;
Chris Masonb1a4d962007-04-04 15:27:52 -04004876
Josef Bacik2552d172009-04-03 10:14:19 -04004877 space_info = __find_space_info(root->fs_info, data);
Josef Bacik1b1d1f62010-03-19 20:49:55 +00004878 if (!space_info) {
Ilya Dryomove0f54062011-06-18 20:26:38 +00004879 printk(KERN_ERR "No space info for %llu\n", data);
Josef Bacik1b1d1f62010-03-19 20:49:55 +00004880 return -ENOSPC;
4881 }
Josef Bacik2552d172009-04-03 10:14:19 -04004882
Josef Bacik67377732010-09-16 16:19:09 -04004883 /*
4884 * If the space info is for both data and metadata it means we have a
4885 * small filesystem and we can't use the clustering stuff.
4886 */
4887 if (btrfs_mixed_space_info(space_info))
4888 use_cluster = false;
4889
Chris Mason0ef3e662008-05-24 14:04:53 -04004890 if (orig_root->ref_cows || empty_size)
4891 allowed_chunk_alloc = 1;
4892
Josef Bacik67377732010-09-16 16:19:09 -04004893 if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
Chris Masonfa9c0d72009-04-03 09:47:43 -04004894 last_ptr = &root->fs_info->meta_alloc_cluster;
Chris Mason536ac8a2009-02-12 09:41:38 -05004895 if (!btrfs_test_opt(root, SSD))
4896 empty_cluster = 64 * 1024;
Chris Mason239b14b2008-03-24 15:02:07 -04004897 }
4898
Josef Bacik67377732010-09-16 16:19:09 -04004899 if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
4900 btrfs_test_opt(root, SSD)) {
Chris Masonfa9c0d72009-04-03 09:47:43 -04004901 last_ptr = &root->fs_info->data_alloc_cluster;
4902 }
Josef Bacik0f9dd462008-09-23 13:14:11 -04004903
Chris Mason239b14b2008-03-24 15:02:07 -04004904 if (last_ptr) {
Chris Masonfa9c0d72009-04-03 09:47:43 -04004905 spin_lock(&last_ptr->lock);
4906 if (last_ptr->block_group)
4907 hint_byte = last_ptr->window_start;
4908 spin_unlock(&last_ptr->lock);
Chris Mason239b14b2008-03-24 15:02:07 -04004909 }
Chris Masonfa9c0d72009-04-03 09:47:43 -04004910
Chris Masona061fc82008-05-07 11:43:44 -04004911 search_start = max(search_start, first_logical_byte(root, 0));
Chris Mason239b14b2008-03-24 15:02:07 -04004912 search_start = max(search_start, hint_byte);
Chris Mason0b86a832008-03-24 15:01:56 -04004913
Josef Bacik817d52f2009-07-13 21:29:25 -04004914 if (!last_ptr)
Chris Masonfa9c0d72009-04-03 09:47:43 -04004915 empty_cluster = 0;
Chris Masonfa9c0d72009-04-03 09:47:43 -04004916
Josef Bacik2552d172009-04-03 10:14:19 -04004917 if (search_start == hint_byte) {
Josef Bacikccf0e722009-11-10 21:23:48 -05004918ideal_cache:
Josef Bacik2552d172009-04-03 10:14:19 -04004919 block_group = btrfs_lookup_block_group(root->fs_info,
4920 search_start);
Josef Bacik817d52f2009-07-13 21:29:25 -04004921 /*
4922 * we don't want to use the block group if it doesn't match our
4923 * allocation bits, or if its not cached.
Josef Bacikccf0e722009-11-10 21:23:48 -05004924 *
4925 * However if we are re-searching with an ideal block group
4926 * picked out then we don't care that the block group is cached.
Josef Bacik817d52f2009-07-13 21:29:25 -04004927 */
4928 if (block_group && block_group_bits(block_group, data) &&
Josef Bacikccf0e722009-11-10 21:23:48 -05004929 (block_group->cached != BTRFS_CACHE_NO ||
4930 search_start == ideal_cache_offset)) {
Josef Bacik2552d172009-04-03 10:14:19 -04004931 down_read(&space_info->groups_sem);
Chris Mason44fb5512009-06-04 15:34:51 -04004932 if (list_empty(&block_group->list) ||
4933 block_group->ro) {
4934 /*
4935 * someone is removing this block group,
4936 * we can't jump into the have_block_group
4937 * target because our list pointers are not
4938 * valid
4939 */
4940 btrfs_put_block_group(block_group);
4941 up_read(&space_info->groups_sem);
Josef Bacikccf0e722009-11-10 21:23:48 -05004942 } else {
Yan, Zhengb742bb82010-05-16 10:46:24 -04004943 index = get_block_group_index(block_group);
Chris Mason44fb5512009-06-04 15:34:51 -04004944 goto have_block_group;
Josef Bacikccf0e722009-11-10 21:23:48 -05004945 }
Josef Bacik2552d172009-04-03 10:14:19 -04004946 } else if (block_group) {
Chris Masonfa9c0d72009-04-03 09:47:43 -04004947 btrfs_put_block_group(block_group);
Josef Bacik2552d172009-04-03 10:14:19 -04004948 }
Chris Mason42e70e72008-11-07 18:17:11 -05004949 }
Josef Bacik2552d172009-04-03 10:14:19 -04004950search:
Josef Bacik80eb2342008-10-29 14:49:05 -04004951 down_read(&space_info->groups_sem);
Yan, Zhengb742bb82010-05-16 10:46:24 -04004952 list_for_each_entry(block_group, &space_info->block_groups[index],
4953 list) {
Josef Bacik6226cb02009-04-03 10:14:18 -04004954 u64 offset;
Josef Bacik817d52f2009-07-13 21:29:25 -04004955 int cached;
Chris Mason8a1413a2008-11-10 16:13:54 -05004956
Josef Bacik11dfe352009-11-13 20:12:59 +00004957 btrfs_get_block_group(block_group);
Josef Bacik2552d172009-04-03 10:14:19 -04004958 search_start = block_group->key.objectid;
Chris Mason42e70e72008-11-07 18:17:11 -05004959
Chris Mason83a50de2010-12-13 15:06:46 -05004960 /*
4961 * this can happen if we end up cycling through all the
4962 * raid types, but we want to make sure we only allocate
4963 * for the proper type.
4964 */
4965 if (!block_group_bits(block_group, data)) {
4966 u64 extra = BTRFS_BLOCK_GROUP_DUP |
4967 BTRFS_BLOCK_GROUP_RAID1 |
4968 BTRFS_BLOCK_GROUP_RAID10;
4969
4970 /*
4971 * if they asked for extra copies and this block group
4972 * doesn't provide them, bail. This does allow us to
4973 * fill raid0 from raid1.
4974 */
4975 if ((data & extra) && !(block_group->flags & extra))
4976 goto loop;
4977 }
4978
Josef Bacik2552d172009-04-03 10:14:19 -04004979have_block_group:
Josef Bacik817d52f2009-07-13 21:29:25 -04004980 if (unlikely(block_group->cached == BTRFS_CACHE_NO)) {
Josef Bacikccf0e722009-11-10 21:23:48 -05004981 u64 free_percent;
4982
Josef Bacikb8399de2010-12-08 09:15:11 -05004983 ret = cache_block_group(block_group, trans,
4984 orig_root, 1);
Josef Bacik9d66e232010-08-25 16:54:15 -04004985 if (block_group->cached == BTRFS_CACHE_FINISHED)
4986 goto have_block_group;
4987
Josef Bacikccf0e722009-11-10 21:23:48 -05004988 free_percent = btrfs_block_group_used(&block_group->item);
4989 free_percent *= 100;
4990 free_percent = div64_u64(free_percent,
4991 block_group->key.offset);
4992 free_percent = 100 - free_percent;
4993 if (free_percent > ideal_cache_percent &&
4994 likely(!block_group->ro)) {
4995 ideal_cache_offset = block_group->key.objectid;
4996 ideal_cache_percent = free_percent;
4997 }
4998
Josef Bacik817d52f2009-07-13 21:29:25 -04004999 /*
Josef Bacikccf0e722009-11-10 21:23:48 -05005000 * We only want to start kthread caching if we are at
5001 * the point where we will wait for caching to make
5002 * progress, or if our ideal search is over and we've
5003 * found somebody to start caching.
Josef Bacik817d52f2009-07-13 21:29:25 -04005004 */
5005 if (loop > LOOP_CACHING_NOWAIT ||
Josef Bacikccf0e722009-11-10 21:23:48 -05005006 (loop > LOOP_FIND_IDEAL &&
5007 atomic_read(&space_info->caching_threads) < 2)) {
Josef Bacikb8399de2010-12-08 09:15:11 -05005008 ret = cache_block_group(block_group, trans,
5009 orig_root, 0);
Josef Bacik817d52f2009-07-13 21:29:25 -04005010 BUG_ON(ret);
Josef Bacik2552d172009-04-03 10:14:19 -04005011 }
Josef Bacikccf0e722009-11-10 21:23:48 -05005012 found_uncached_bg = true;
5013
5014 /*
5015 * If loop is set for cached only, try the next block
5016 * group.
5017 */
5018 if (loop == LOOP_FIND_IDEAL)
5019 goto loop;
Josef Bacikea6a4782008-11-20 12:16:16 -05005020 }
5021
Josef Bacik817d52f2009-07-13 21:29:25 -04005022 cached = block_group_cache_done(block_group);
Josef Bacikccf0e722009-11-10 21:23:48 -05005023 if (unlikely(!cached))
Josef Bacik817d52f2009-07-13 21:29:25 -04005024 found_uncached_bg = true;
5025
Josef Bacikea6a4782008-11-20 12:16:16 -05005026 if (unlikely(block_group->ro))
Josef Bacik2552d172009-04-03 10:14:19 -04005027 goto loop;
Josef Bacik0f9dd462008-09-23 13:14:11 -04005028
Chris Masonff5714c2011-05-28 07:00:39 -04005029 spin_lock(&block_group->free_space_ctl->tree_lock);
Josef Bacikcca1c812011-05-13 11:07:12 -04005030 if (cached &&
Chris Masonff5714c2011-05-28 07:00:39 -04005031 block_group->free_space_ctl->free_space <
5032 num_bytes + empty_size) {
5033 spin_unlock(&block_group->free_space_ctl->tree_lock);
Josef Bacikcca1c812011-05-13 11:07:12 -04005034 goto loop;
5035 }
Chris Masonff5714c2011-05-28 07:00:39 -04005036 spin_unlock(&block_group->free_space_ctl->tree_lock);
Josef Bacikcca1c812011-05-13 11:07:12 -04005037
Josef Bacik0a243252009-09-11 16:11:20 -04005038 /*
5039 * Ok we want to try and use the cluster allocator, so lets look
5040 * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will
5041 * have tried the cluster allocator plenty of times at this
5042 * point and not have found anything, so we are likely way too
5043 * fragmented for the clustering stuff to find anything, so lets
5044 * just skip it and let the allocator find whatever block it can
5045 * find
5046 */
5047 if (last_ptr && loop < LOOP_NO_EMPTY_SIZE) {
Chris Masonfa9c0d72009-04-03 09:47:43 -04005048 /*
5049 * the refill lock keeps out other
5050 * people trying to start a new cluster
5051 */
5052 spin_lock(&last_ptr->refill_lock);
Chris Mason44fb5512009-06-04 15:34:51 -04005053 if (last_ptr->block_group &&
5054 (last_ptr->block_group->ro ||
5055 !block_group_bits(last_ptr->block_group, data))) {
5056 offset = 0;
5057 goto refill_cluster;
5058 }
5059
Chris Masonfa9c0d72009-04-03 09:47:43 -04005060 offset = btrfs_alloc_from_cluster(block_group, last_ptr,
5061 num_bytes, search_start);
5062 if (offset) {
5063 /* we have a block, we're done */
5064 spin_unlock(&last_ptr->refill_lock);
5065 goto checks;
5066 }
5067
5068 spin_lock(&last_ptr->lock);
5069 /*
5070 * whoops, this cluster doesn't actually point to
5071 * this block group. Get a ref on the block
5072 * group is does point to and try again
5073 */
5074 if (!last_ptr_loop && last_ptr->block_group &&
5075 last_ptr->block_group != block_group) {
5076
5077 btrfs_put_block_group(block_group);
5078 block_group = last_ptr->block_group;
Josef Bacik11dfe352009-11-13 20:12:59 +00005079 btrfs_get_block_group(block_group);
Chris Masonfa9c0d72009-04-03 09:47:43 -04005080 spin_unlock(&last_ptr->lock);
5081 spin_unlock(&last_ptr->refill_lock);
5082
5083 last_ptr_loop = 1;
5084 search_start = block_group->key.objectid;
Chris Mason44fb5512009-06-04 15:34:51 -04005085 /*
5086 * we know this block group is properly
5087 * in the list because
5088 * btrfs_remove_block_group, drops the
5089 * cluster before it removes the block
5090 * group from the list
5091 */
Chris Masonfa9c0d72009-04-03 09:47:43 -04005092 goto have_block_group;
5093 }
5094 spin_unlock(&last_ptr->lock);
Chris Mason44fb5512009-06-04 15:34:51 -04005095refill_cluster:
Chris Masonfa9c0d72009-04-03 09:47:43 -04005096 /*
5097 * this cluster didn't work out, free it and
5098 * start over
5099 */
5100 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5101
5102 last_ptr_loop = 0;
5103
5104 /* allocate a cluster in this block group */
Chris Mason451d7582009-06-09 20:28:34 -04005105 ret = btrfs_find_space_cluster(trans, root,
Chris Masonfa9c0d72009-04-03 09:47:43 -04005106 block_group, last_ptr,
5107 offset, num_bytes,
5108 empty_cluster + empty_size);
5109 if (ret == 0) {
5110 /*
5111 * now pull our allocation out of this
5112 * cluster
5113 */
5114 offset = btrfs_alloc_from_cluster(block_group,
5115 last_ptr, num_bytes,
5116 search_start);
5117 if (offset) {
5118 /* we found one, proceed */
5119 spin_unlock(&last_ptr->refill_lock);
5120 goto checks;
5121 }
Josef Bacik0a243252009-09-11 16:11:20 -04005122 } else if (!cached && loop > LOOP_CACHING_NOWAIT
5123 && !failed_cluster_refill) {
Josef Bacik817d52f2009-07-13 21:29:25 -04005124 spin_unlock(&last_ptr->refill_lock);
5125
Josef Bacik0a243252009-09-11 16:11:20 -04005126 failed_cluster_refill = true;
Josef Bacik817d52f2009-07-13 21:29:25 -04005127 wait_block_group_cache_progress(block_group,
5128 num_bytes + empty_cluster + empty_size);
5129 goto have_block_group;
Chris Masonfa9c0d72009-04-03 09:47:43 -04005130 }
Josef Bacik817d52f2009-07-13 21:29:25 -04005131
Chris Masonfa9c0d72009-04-03 09:47:43 -04005132 /*
5133 * at this point we either didn't find a cluster
5134 * or we weren't able to allocate a block from our
5135 * cluster. Free the cluster we've been trying
5136 * to use, and go to the next block group
5137 */
Josef Bacik0a243252009-09-11 16:11:20 -04005138 btrfs_return_cluster_to_free_space(NULL, last_ptr);
Chris Masonfa9c0d72009-04-03 09:47:43 -04005139 spin_unlock(&last_ptr->refill_lock);
Josef Bacik0a243252009-09-11 16:11:20 -04005140 goto loop;
Chris Masonfa9c0d72009-04-03 09:47:43 -04005141 }
5142
Josef Bacik6226cb02009-04-03 10:14:18 -04005143 offset = btrfs_find_space_for_alloc(block_group, search_start,
5144 num_bytes, empty_size);
Josef Bacik1cdda9b2009-10-06 10:04:28 -04005145 /*
5146 * If we didn't find a chunk, and we haven't failed on this
5147 * block group before, and this block group is in the middle of
5148 * caching and we are ok with waiting, then go ahead and wait
5149 * for progress to be made, and set failed_alloc to true.
5150 *
5151 * If failed_alloc is true then we've already waited on this
5152 * block group once and should move on to the next block group.
5153 */
5154 if (!offset && !failed_alloc && !cached &&
5155 loop > LOOP_CACHING_NOWAIT) {
Josef Bacik817d52f2009-07-13 21:29:25 -04005156 wait_block_group_cache_progress(block_group,
Josef Bacik1cdda9b2009-10-06 10:04:28 -04005157 num_bytes + empty_size);
5158 failed_alloc = true;
Josef Bacik817d52f2009-07-13 21:29:25 -04005159 goto have_block_group;
Josef Bacik1cdda9b2009-10-06 10:04:28 -04005160 } else if (!offset) {
5161 goto loop;
Josef Bacik817d52f2009-07-13 21:29:25 -04005162 }
Chris Masonfa9c0d72009-04-03 09:47:43 -04005163checks:
Josef Bacik6226cb02009-04-03 10:14:18 -04005164 search_start = stripe_align(root, offset);
Josef Bacik2552d172009-04-03 10:14:19 -04005165 /* move on to the next group */
Josef Bacik6226cb02009-04-03 10:14:18 -04005166 if (search_start + num_bytes >= search_end) {
5167 btrfs_add_free_space(block_group, offset, num_bytes);
Josef Bacik2552d172009-04-03 10:14:19 -04005168 goto loop;
Josef Bacik6226cb02009-04-03 10:14:18 -04005169 }
Chris Masone37c9e62007-05-09 20:13:14 -04005170
Josef Bacik2552d172009-04-03 10:14:19 -04005171 /* move on to the next group */
5172 if (search_start + num_bytes >
Josef Bacik6226cb02009-04-03 10:14:18 -04005173 block_group->key.objectid + block_group->key.offset) {
5174 btrfs_add_free_space(block_group, offset, num_bytes);
Josef Bacik2552d172009-04-03 10:14:19 -04005175 goto loop;
Josef Bacik6226cb02009-04-03 10:14:18 -04005176 }
Josef Bacik80eb2342008-10-29 14:49:05 -04005177
Josef Bacik2552d172009-04-03 10:14:19 -04005178 ins->objectid = search_start;
5179 ins->offset = num_bytes;
5180
Josef Bacik6226cb02009-04-03 10:14:18 -04005181 if (offset < search_start)
5182 btrfs_add_free_space(block_group, offset,
5183 search_start - offset);
5184 BUG_ON(offset > search_start);
5185
Li Dongyangb4d00d52011-03-24 10:24:25 +00005186 ret = btrfs_update_reserved_bytes(block_group, num_bytes, 1,
Yan, Zhengf0486c62010-05-16 10:46:25 -04005187 (data & BTRFS_BLOCK_GROUP_DATA));
5188 if (ret == -EAGAIN) {
5189 btrfs_add_free_space(block_group, offset, num_bytes);
5190 goto loop;
5191 }
Yan Zheng11833d62009-09-11 16:11:19 -04005192
Josef Bacik2552d172009-04-03 10:14:19 -04005193 /* we are all good, lets return */
Yan, Zhengf0486c62010-05-16 10:46:25 -04005194 ins->objectid = search_start;
5195 ins->offset = num_bytes;
5196
5197 if (offset < search_start)
5198 btrfs_add_free_space(block_group, offset,
5199 search_start - offset);
5200 BUG_ON(offset > search_start);
Josef Bacikd82a6f12011-05-11 15:26:06 -04005201 btrfs_put_block_group(block_group);
Josef Bacik2552d172009-04-03 10:14:19 -04005202 break;
5203loop:
Josef Bacik0a243252009-09-11 16:11:20 -04005204 failed_cluster_refill = false;
Josef Bacik1cdda9b2009-10-06 10:04:28 -04005205 failed_alloc = false;
Yan, Zhengb742bb82010-05-16 10:46:24 -04005206 BUG_ON(index != get_block_group_index(block_group));
Chris Masonfa9c0d72009-04-03 09:47:43 -04005207 btrfs_put_block_group(block_group);
Josef Bacik2552d172009-04-03 10:14:19 -04005208 }
5209 up_read(&space_info->groups_sem);
Chris Masonf5a31e12008-11-10 11:47:09 -05005210
Yan, Zhengb742bb82010-05-16 10:46:24 -04005211 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
5212 goto search;
5213
Josef Bacikccf0e722009-11-10 21:23:48 -05005214 /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
5215 * for them to make caching progress. Also
5216 * determine the best possible bg to cache
5217 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5218 * caching kthreads as we move along
Josef Bacik817d52f2009-07-13 21:29:25 -04005219 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5220 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5221 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5222 * again
Chris Masonfa9c0d72009-04-03 09:47:43 -04005223 */
Josef Bacik723bda22011-05-27 16:11:38 -04005224 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
Yan, Zhengb742bb82010-05-16 10:46:24 -04005225 index = 0;
Josef Bacikccf0e722009-11-10 21:23:48 -05005226 if (loop == LOOP_FIND_IDEAL && found_uncached_bg) {
Josef Bacik817d52f2009-07-13 21:29:25 -04005227 found_uncached_bg = false;
Josef Bacikccf0e722009-11-10 21:23:48 -05005228 loop++;
5229 if (!ideal_cache_percent &&
5230 atomic_read(&space_info->caching_threads))
Josef Bacik817d52f2009-07-13 21:29:25 -04005231 goto search;
Josef Bacikccf0e722009-11-10 21:23:48 -05005232
5233 /*
5234 * 1 of the following 2 things have happened so far
5235 *
5236 * 1) We found an ideal block group for caching that
5237 * is mostly full and will cache quickly, so we might
5238 * as well wait for it.
5239 *
5240 * 2) We searched for cached only and we didn't find
5241 * anything, and we didn't start any caching kthreads
5242 * either, so chances are we will loop through and
5243 * start a couple caching kthreads, and then come back
5244 * around and just wait for them. This will be slower
5245 * because we will have 2 caching kthreads reading at
5246 * the same time when we could have just started one
5247 * and waited for it to get far enough to give us an
5248 * allocation, so go ahead and go to the wait caching
5249 * loop.
5250 */
5251 loop = LOOP_CACHING_WAIT;
5252 search_start = ideal_cache_offset;
5253 ideal_cache_percent = 0;
5254 goto ideal_cache;
5255 } else if (loop == LOOP_FIND_IDEAL) {
5256 /*
5257 * Didn't find a uncached bg, wait on anything we find
5258 * next.
5259 */
5260 loop = LOOP_CACHING_WAIT;
5261 goto search;
5262 }
5263
Josef Bacik723bda22011-05-27 16:11:38 -04005264 loop++;
Josef Bacik817d52f2009-07-13 21:29:25 -04005265
5266 if (loop == LOOP_ALLOC_CHUNK) {
Josef Bacik723bda22011-05-27 16:11:38 -04005267 if (allowed_chunk_alloc) {
5268 ret = do_chunk_alloc(trans, root, num_bytes +
5269 2 * 1024 * 1024, data,
5270 CHUNK_ALLOC_LIMITED);
5271 allowed_chunk_alloc = 0;
5272 if (ret == 1)
5273 done_chunk_alloc = 1;
5274 } else if (!done_chunk_alloc &&
5275 space_info->force_alloc ==
5276 CHUNK_ALLOC_NO_FORCE) {
5277 space_info->force_alloc = CHUNK_ALLOC_LIMITED;
5278 }
5279
5280 /*
5281 * We didn't allocate a chunk, go ahead and drop the
5282 * empty size and loop again.
5283 */
5284 if (!done_chunk_alloc)
5285 loop = LOOP_NO_EMPTY_SIZE;
5286 }
5287
5288 if (loop == LOOP_NO_EMPTY_SIZE) {
Chris Masonfa9c0d72009-04-03 09:47:43 -04005289 empty_size = 0;
5290 empty_cluster = 0;
5291 }
Chris Mason42e70e72008-11-07 18:17:11 -05005292
Josef Bacik723bda22011-05-27 16:11:38 -04005293 goto search;
Josef Bacik2552d172009-04-03 10:14:19 -04005294 } else if (!ins->objectid) {
5295 ret = -ENOSPC;
Josef Bacikd82a6f12011-05-11 15:26:06 -04005296 } else if (ins->objectid) {
Josef Bacik2552d172009-04-03 10:14:19 -04005297 ret = 0;
5298 }
Chris Mason0b86a832008-03-24 15:01:56 -04005299
Chris Mason0f70abe2007-02-28 16:46:22 -05005300 return ret;
Chris Masonfec577f2007-02-26 10:40:21 -05005301}
Chris Masonec44a352008-04-28 15:29:52 -04005302
Josef Bacik9ed74f22009-09-11 16:12:44 -04005303static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
5304 int dump_block_groups)
Josef Bacik0f9dd462008-09-23 13:14:11 -04005305{
5306 struct btrfs_block_group_cache *cache;
Yan, Zhengb742bb82010-05-16 10:46:24 -04005307 int index = 0;
Josef Bacik0f9dd462008-09-23 13:14:11 -04005308
Josef Bacik9ed74f22009-09-11 16:12:44 -04005309 spin_lock(&info->lock);
Chris Masond3977122009-01-05 21:25:51 -05005310 printk(KERN_INFO "space_info has %llu free, is %sfull\n",
5311 (unsigned long long)(info->total_bytes - info->bytes_used -
Josef Bacik9ed74f22009-09-11 16:12:44 -04005312 info->bytes_pinned - info->bytes_reserved -
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04005313 info->bytes_readonly),
Chris Masond3977122009-01-05 21:25:51 -05005314 (info->full) ? "" : "not ");
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04005315 printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
5316 "reserved=%llu, may_use=%llu, readonly=%llu\n",
Joel Becker21380932009-04-21 12:38:29 -07005317 (unsigned long long)info->total_bytes,
Josef Bacik9ed74f22009-09-11 16:12:44 -04005318 (unsigned long long)info->bytes_used,
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04005319 (unsigned long long)info->bytes_pinned,
5320 (unsigned long long)info->bytes_reserved,
5321 (unsigned long long)info->bytes_may_use,
5322 (unsigned long long)info->bytes_readonly);
Josef Bacik9ed74f22009-09-11 16:12:44 -04005323 spin_unlock(&info->lock);
5324
5325 if (!dump_block_groups)
5326 return;
Josef Bacik0f9dd462008-09-23 13:14:11 -04005327
Josef Bacik80eb2342008-10-29 14:49:05 -04005328 down_read(&info->groups_sem);
Yan, Zhengb742bb82010-05-16 10:46:24 -04005329again:
5330 list_for_each_entry(cache, &info->block_groups[index], list) {
Josef Bacik0f9dd462008-09-23 13:14:11 -04005331 spin_lock(&cache->lock);
Chris Masond3977122009-01-05 21:25:51 -05005332 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
5333 "%llu pinned %llu reserved\n",
5334 (unsigned long long)cache->key.objectid,
5335 (unsigned long long)cache->key.offset,
5336 (unsigned long long)btrfs_block_group_used(&cache->item),
5337 (unsigned long long)cache->pinned,
5338 (unsigned long long)cache->reserved);
Josef Bacik0f9dd462008-09-23 13:14:11 -04005339 btrfs_dump_free_space(cache, bytes);
5340 spin_unlock(&cache->lock);
5341 }
Yan, Zhengb742bb82010-05-16 10:46:24 -04005342 if (++index < BTRFS_NR_RAID_TYPES)
5343 goto again;
Josef Bacik80eb2342008-10-29 14:49:05 -04005344 up_read(&info->groups_sem);
Josef Bacik0f9dd462008-09-23 13:14:11 -04005345}
Zheng Yane8569812008-09-26 10:05:48 -04005346
Yan Zheng11833d62009-09-11 16:11:19 -04005347int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
5348 struct btrfs_root *root,
5349 u64 num_bytes, u64 min_alloc_size,
5350 u64 empty_size, u64 hint_byte,
5351 u64 search_end, struct btrfs_key *ins,
5352 u64 data)
Chris Masonfec577f2007-02-26 10:40:21 -05005353{
5354 int ret;
Chris Masonfbdc7622007-05-30 10:22:12 -04005355 u64 search_start = 0;
Chris Mason925baed2008-06-25 16:01:30 -04005356
Josef Bacik6a632092009-02-20 11:00:09 -05005357 data = btrfs_get_alloc_profile(root, data);
Chris Mason98d20f62008-04-14 09:46:10 -04005358again:
Chris Mason0ef3e662008-05-24 14:04:53 -04005359 /*
5360 * the only place that sets empty_size is btrfs_realloc_node, which
5361 * is not called recursively on allocations
5362 */
Josef Bacik83d3c962009-12-07 21:45:59 +00005363 if (empty_size || root->ref_cows)
Chris Mason6324fbf2008-03-24 15:01:59 -04005364 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
Chris Mason0e4f8f82011-04-15 16:05:44 -04005365 num_bytes + 2 * 1024 * 1024, data,
5366 CHUNK_ALLOC_NO_FORCE);
Chris Mason0b86a832008-03-24 15:01:56 -04005367
Chris Masondb945352007-10-15 16:15:53 -04005368 WARN_ON(num_bytes < root->sectorsize);
5369 ret = find_free_extent(trans, root, num_bytes, empty_size,
Yan, Zhengf0486c62010-05-16 10:46:25 -04005370 search_start, search_end, hint_byte,
5371 ins, data);
Chris Mason3b951512008-04-17 11:29:12 -04005372
Chris Mason98d20f62008-04-14 09:46:10 -04005373 if (ret == -ENOSPC && num_bytes > min_alloc_size) {
5374 num_bytes = num_bytes >> 1;
Josef Bacik0f9dd462008-09-23 13:14:11 -04005375 num_bytes = num_bytes & ~(root->sectorsize - 1);
Chris Mason98d20f62008-04-14 09:46:10 -04005376 num_bytes = max(num_bytes, min_alloc_size);
Chris Mason0ef3e662008-05-24 14:04:53 -04005377 do_chunk_alloc(trans, root->fs_info->extent_root,
Chris Mason0e4f8f82011-04-15 16:05:44 -04005378 num_bytes, data, CHUNK_ALLOC_FORCE);
Chris Mason98d20f62008-04-14 09:46:10 -04005379 goto again;
5380 }
Chris Mason91435652011-02-16 13:10:41 -05005381 if (ret == -ENOSPC && btrfs_test_opt(root, ENOSPC_DEBUG)) {
Josef Bacik0f9dd462008-09-23 13:14:11 -04005382 struct btrfs_space_info *sinfo;
5383
5384 sinfo = __find_space_info(root->fs_info, data);
Chris Masond3977122009-01-05 21:25:51 -05005385 printk(KERN_ERR "btrfs allocation failed flags %llu, "
5386 "wanted %llu\n", (unsigned long long)data,
5387 (unsigned long long)num_bytes);
Josef Bacik9ed74f22009-09-11 16:12:44 -04005388 dump_space_info(sinfo, num_bytes, 1);
Chris Mason925baed2008-06-25 16:01:30 -04005389 }
Josef Bacik0f9dd462008-09-23 13:14:11 -04005390
liubo1abe9b82011-03-24 11:18:59 +00005391 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
5392
Josef Bacik0f9dd462008-09-23 13:14:11 -04005393 return ret;
Chris Masone6dcd2d2008-07-17 12:53:50 -04005394}
5395
Chris Mason65b51a02008-08-01 15:11:20 -04005396int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
5397{
Josef Bacik0f9dd462008-09-23 13:14:11 -04005398 struct btrfs_block_group_cache *cache;
Liu Hui1f3c79a2009-01-05 15:57:51 -05005399 int ret = 0;
Josef Bacik0f9dd462008-09-23 13:14:11 -04005400
Josef Bacik0f9dd462008-09-23 13:14:11 -04005401 cache = btrfs_lookup_block_group(root->fs_info, start);
5402 if (!cache) {
Chris Masond3977122009-01-05 21:25:51 -05005403 printk(KERN_ERR "Unable to find block group for %llu\n",
5404 (unsigned long long)start);
Josef Bacik0f9dd462008-09-23 13:14:11 -04005405 return -ENOSPC;
5406 }
Liu Hui1f3c79a2009-01-05 15:57:51 -05005407
Li Dongyang5378e602011-03-24 10:24:27 +00005408 if (btrfs_test_opt(root, DISCARD))
5409 ret = btrfs_discard_extent(root, start, len, NULL);
Liu Hui1f3c79a2009-01-05 15:57:51 -05005410
Josef Bacik0f9dd462008-09-23 13:14:11 -04005411 btrfs_add_free_space(cache, start, len);
Li Dongyangb4d00d52011-03-24 10:24:25 +00005412 btrfs_update_reserved_bytes(cache, len, 0, 1);
Chris Masonfa9c0d72009-04-03 09:47:43 -04005413 btrfs_put_block_group(cache);
Josef Bacik817d52f2009-07-13 21:29:25 -04005414
liubo1abe9b82011-03-24 11:18:59 +00005415 trace_btrfs_reserved_extent_free(root, start, len);
5416
Chris Masone6dcd2d2008-07-17 12:53:50 -04005417 return ret;
5418}
5419
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005420static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5421 struct btrfs_root *root,
5422 u64 parent, u64 root_objectid,
5423 u64 flags, u64 owner, u64 offset,
5424 struct btrfs_key *ins, int ref_mod)
Chris Masone6dcd2d2008-07-17 12:53:50 -04005425{
5426 int ret;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005427 struct btrfs_fs_info *fs_info = root->fs_info;
Chris Masone6dcd2d2008-07-17 12:53:50 -04005428 struct btrfs_extent_item *extent_item;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005429 struct btrfs_extent_inline_ref *iref;
Chris Masone6dcd2d2008-07-17 12:53:50 -04005430 struct btrfs_path *path;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005431 struct extent_buffer *leaf;
5432 int type;
5433 u32 size;
Chris Masonf2654de2007-06-26 12:20:46 -04005434
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005435 if (parent > 0)
5436 type = BTRFS_SHARED_DATA_REF_KEY;
5437 else
5438 type = BTRFS_EXTENT_DATA_REF_KEY;
Zheng Yan31840ae2008-09-23 13:14:14 -04005439
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005440 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
Chris Mason7bb86312007-12-11 09:25:06 -05005441
5442 path = btrfs_alloc_path();
Tsutomu Itohdb5b4932011-03-23 08:14:16 +00005443 if (!path)
5444 return -ENOMEM;
Chris Mason47e4bb92008-02-01 14:51:59 -05005445
Chris Masonb9473432009-03-13 11:00:37 -04005446 path->leave_spinning = 1;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005447 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5448 ins, size);
Chris Masonccd467d2007-06-28 15:57:36 -04005449 BUG_ON(ret);
Josef Bacik0f9dd462008-09-23 13:14:11 -04005450
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005451 leaf = path->nodes[0];
5452 extent_item = btrfs_item_ptr(leaf, path->slots[0],
Chris Mason47e4bb92008-02-01 14:51:59 -05005453 struct btrfs_extent_item);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005454 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
5455 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5456 btrfs_set_extent_flags(leaf, extent_item,
5457 flags | BTRFS_EXTENT_FLAG_DATA);
Chris Mason47e4bb92008-02-01 14:51:59 -05005458
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005459 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
5460 btrfs_set_extent_inline_ref_type(leaf, iref, type);
5461 if (parent > 0) {
5462 struct btrfs_shared_data_ref *ref;
5463 ref = (struct btrfs_shared_data_ref *)(iref + 1);
5464 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5465 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
5466 } else {
5467 struct btrfs_extent_data_ref *ref;
5468 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
5469 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
5470 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
5471 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
5472 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
5473 }
Chris Mason47e4bb92008-02-01 14:51:59 -05005474
5475 btrfs_mark_buffer_dirty(path->nodes[0]);
Chris Mason7bb86312007-12-11 09:25:06 -05005476 btrfs_free_path(path);
Chris Masonf510cfe2007-10-15 16:14:48 -04005477
Yan, Zhengf0486c62010-05-16 10:46:25 -04005478 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
Chris Masonf5947062008-02-04 10:10:13 -05005479 if (ret) {
Chris Masond3977122009-01-05 21:25:51 -05005480 printk(KERN_ERR "btrfs update block group failed for %llu "
5481 "%llu\n", (unsigned long long)ins->objectid,
5482 (unsigned long long)ins->offset);
Chris Masonf5947062008-02-04 10:10:13 -05005483 BUG();
5484 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04005485 return ret;
5486}
5487
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005488static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
5489 struct btrfs_root *root,
5490 u64 parent, u64 root_objectid,
5491 u64 flags, struct btrfs_disk_key *key,
5492 int level, struct btrfs_key *ins)
5493{
5494 int ret;
5495 struct btrfs_fs_info *fs_info = root->fs_info;
5496 struct btrfs_extent_item *extent_item;
5497 struct btrfs_tree_block_info *block_info;
5498 struct btrfs_extent_inline_ref *iref;
5499 struct btrfs_path *path;
5500 struct extent_buffer *leaf;
5501 u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
5502
5503 path = btrfs_alloc_path();
5504 BUG_ON(!path);
5505
5506 path->leave_spinning = 1;
5507 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5508 ins, size);
5509 BUG_ON(ret);
5510
5511 leaf = path->nodes[0];
5512 extent_item = btrfs_item_ptr(leaf, path->slots[0],
5513 struct btrfs_extent_item);
5514 btrfs_set_extent_refs(leaf, extent_item, 1);
5515 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5516 btrfs_set_extent_flags(leaf, extent_item,
5517 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
5518 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
5519
5520 btrfs_set_tree_block_key(leaf, block_info, key);
5521 btrfs_set_tree_block_level(leaf, block_info, level);
5522
5523 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
5524 if (parent > 0) {
5525 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
5526 btrfs_set_extent_inline_ref_type(leaf, iref,
5527 BTRFS_SHARED_BLOCK_REF_KEY);
5528 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5529 } else {
5530 btrfs_set_extent_inline_ref_type(leaf, iref,
5531 BTRFS_TREE_BLOCK_REF_KEY);
5532 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
5533 }
5534
5535 btrfs_mark_buffer_dirty(leaf);
5536 btrfs_free_path(path);
5537
Yan, Zhengf0486c62010-05-16 10:46:25 -04005538 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005539 if (ret) {
5540 printk(KERN_ERR "btrfs update block group failed for %llu "
5541 "%llu\n", (unsigned long long)ins->objectid,
5542 (unsigned long long)ins->offset);
5543 BUG();
5544 }
5545 return ret;
5546}
5547
5548int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5549 struct btrfs_root *root,
5550 u64 root_objectid, u64 owner,
5551 u64 offset, struct btrfs_key *ins)
Chris Masone6dcd2d2008-07-17 12:53:50 -04005552{
5553 int ret;
Chris Mason1c2308f2008-09-23 13:14:13 -04005554
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005555 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
Chris Mason56bec292009-03-13 10:10:06 -04005556
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005557 ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset,
5558 0, root_objectid, owner, offset,
5559 BTRFS_ADD_DELAYED_EXTENT, NULL);
Chris Masone6dcd2d2008-07-17 12:53:50 -04005560 return ret;
5561}
Chris Masone02119d2008-09-05 16:13:11 -04005562
5563/*
5564 * this is used by the tree logging recovery code. It records that
5565 * an extent has been allocated and makes sure to clear the free
5566 * space cache bits as well
5567 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005568int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
5569 struct btrfs_root *root,
5570 u64 root_objectid, u64 owner, u64 offset,
5571 struct btrfs_key *ins)
Chris Masone02119d2008-09-05 16:13:11 -04005572{
5573 int ret;
5574 struct btrfs_block_group_cache *block_group;
Yan Zheng11833d62009-09-11 16:11:19 -04005575 struct btrfs_caching_control *caching_ctl;
5576 u64 start = ins->objectid;
5577 u64 num_bytes = ins->offset;
Chris Masone02119d2008-09-05 16:13:11 -04005578
Chris Masone02119d2008-09-05 16:13:11 -04005579 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
Josef Bacikb8399de2010-12-08 09:15:11 -05005580 cache_block_group(block_group, trans, NULL, 0);
Yan Zheng11833d62009-09-11 16:11:19 -04005581 caching_ctl = get_caching_control(block_group);
Chris Masone02119d2008-09-05 16:13:11 -04005582
Yan Zheng11833d62009-09-11 16:11:19 -04005583 if (!caching_ctl) {
5584 BUG_ON(!block_group_cache_done(block_group));
5585 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5586 BUG_ON(ret);
5587 } else {
5588 mutex_lock(&caching_ctl->mutex);
5589
5590 if (start >= caching_ctl->progress) {
5591 ret = add_excluded_extent(root, start, num_bytes);
5592 BUG_ON(ret);
5593 } else if (start + num_bytes <= caching_ctl->progress) {
5594 ret = btrfs_remove_free_space(block_group,
5595 start, num_bytes);
5596 BUG_ON(ret);
5597 } else {
5598 num_bytes = caching_ctl->progress - start;
5599 ret = btrfs_remove_free_space(block_group,
5600 start, num_bytes);
5601 BUG_ON(ret);
5602
5603 start = caching_ctl->progress;
5604 num_bytes = ins->objectid + ins->offset -
5605 caching_ctl->progress;
5606 ret = add_excluded_extent(root, start, num_bytes);
5607 BUG_ON(ret);
5608 }
5609
5610 mutex_unlock(&caching_ctl->mutex);
5611 put_caching_control(caching_ctl);
5612 }
5613
Li Dongyangb4d00d52011-03-24 10:24:25 +00005614 ret = btrfs_update_reserved_bytes(block_group, ins->offset, 1, 1);
Yan, Zhengf0486c62010-05-16 10:46:25 -04005615 BUG_ON(ret);
Chris Masonfa9c0d72009-04-03 09:47:43 -04005616 btrfs_put_block_group(block_group);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005617 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
5618 0, owner, offset, ins, 1);
Chris Masone02119d2008-09-05 16:13:11 -04005619 return ret;
5620}
5621
Chris Mason65b51a02008-08-01 15:11:20 -04005622struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
5623 struct btrfs_root *root,
Chris Mason4008c042009-02-12 14:09:45 -05005624 u64 bytenr, u32 blocksize,
5625 int level)
Chris Mason65b51a02008-08-01 15:11:20 -04005626{
5627 struct extent_buffer *buf;
5628
5629 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5630 if (!buf)
5631 return ERR_PTR(-ENOMEM);
5632 btrfs_set_header_generation(buf, trans->transid);
Chris Mason4008c042009-02-12 14:09:45 -05005633 btrfs_set_buffer_lockdep_class(buf, level);
Chris Mason65b51a02008-08-01 15:11:20 -04005634 btrfs_tree_lock(buf);
5635 clean_tree_block(trans, root, buf);
Chris Masonb4ce94d2009-02-04 09:25:08 -05005636
5637 btrfs_set_lock_blocking(buf);
Chris Mason65b51a02008-08-01 15:11:20 -04005638 btrfs_set_buffer_uptodate(buf);
Chris Masonb4ce94d2009-02-04 09:25:08 -05005639
Chris Masond0c803c2008-09-11 16:17:57 -04005640 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
Yan, Zheng8cef4e12009-11-12 09:33:26 +00005641 /*
5642 * we allow two log transactions at a time, use different
5643 * EXENT bit to differentiate dirty pages.
5644 */
5645 if (root->log_transid % 2 == 0)
5646 set_extent_dirty(&root->dirty_log_pages, buf->start,
5647 buf->start + buf->len - 1, GFP_NOFS);
5648 else
5649 set_extent_new(&root->dirty_log_pages, buf->start,
5650 buf->start + buf->len - 1, GFP_NOFS);
Chris Masond0c803c2008-09-11 16:17:57 -04005651 } else {
5652 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
5653 buf->start + buf->len - 1, GFP_NOFS);
5654 }
Chris Mason65b51a02008-08-01 15:11:20 -04005655 trans->blocks_used++;
Chris Masonb4ce94d2009-02-04 09:25:08 -05005656 /* this returns a buffer locked for blocking */
Chris Mason65b51a02008-08-01 15:11:20 -04005657 return buf;
5658}
5659
Yan, Zhengf0486c62010-05-16 10:46:25 -04005660static struct btrfs_block_rsv *
5661use_block_rsv(struct btrfs_trans_handle *trans,
5662 struct btrfs_root *root, u32 blocksize)
5663{
5664 struct btrfs_block_rsv *block_rsv;
Josef Bacik68a82272011-01-24 21:43:20 +00005665 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
Yan, Zhengf0486c62010-05-16 10:46:25 -04005666 int ret;
5667
5668 block_rsv = get_block_rsv(trans, root);
5669
5670 if (block_rsv->size == 0) {
Josef Bacik8bb8ab22010-10-15 16:52:49 -04005671 ret = reserve_metadata_bytes(trans, root, block_rsv,
5672 blocksize, 0);
Josef Bacik68a82272011-01-24 21:43:20 +00005673 /*
5674 * If we couldn't reserve metadata bytes try and use some from
5675 * the global reserve.
5676 */
5677 if (ret && block_rsv != global_rsv) {
5678 ret = block_rsv_use_bytes(global_rsv, blocksize);
5679 if (!ret)
5680 return global_rsv;
Yan, Zhengf0486c62010-05-16 10:46:25 -04005681 return ERR_PTR(ret);
Josef Bacik68a82272011-01-24 21:43:20 +00005682 } else if (ret) {
5683 return ERR_PTR(ret);
5684 }
Yan, Zhengf0486c62010-05-16 10:46:25 -04005685 return block_rsv;
5686 }
5687
5688 ret = block_rsv_use_bytes(block_rsv, blocksize);
5689 if (!ret)
5690 return block_rsv;
Josef Bacik68a82272011-01-24 21:43:20 +00005691 if (ret) {
5692 WARN_ON(1);
5693 ret = reserve_metadata_bytes(trans, root, block_rsv, blocksize,
5694 0);
5695 if (!ret) {
5696 spin_lock(&block_rsv->lock);
5697 block_rsv->size += blocksize;
5698 spin_unlock(&block_rsv->lock);
5699 return block_rsv;
5700 } else if (ret && block_rsv != global_rsv) {
5701 ret = block_rsv_use_bytes(global_rsv, blocksize);
5702 if (!ret)
5703 return global_rsv;
5704 }
5705 }
Yan, Zhengf0486c62010-05-16 10:46:25 -04005706
Yan, Zhengf0486c62010-05-16 10:46:25 -04005707 return ERR_PTR(-ENOSPC);
5708}
5709
5710static void unuse_block_rsv(struct btrfs_block_rsv *block_rsv, u32 blocksize)
5711{
5712 block_rsv_add_bytes(block_rsv, blocksize, 0);
5713 block_rsv_release_bytes(block_rsv, NULL, 0);
5714}
5715
Chris Masonfec577f2007-02-26 10:40:21 -05005716/*
Yan, Zhengf0486c62010-05-16 10:46:25 -04005717 * finds a free extent and does all the dirty work required for allocation
5718 * returns the key for the extent through ins, and a tree buffer for
5719 * the first block of the extent through buf.
5720 *
Chris Masonfec577f2007-02-26 10:40:21 -05005721 * returns the tree buffer or NULL.
5722 */
Chris Mason5f39d392007-10-15 16:14:19 -04005723struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005724 struct btrfs_root *root, u32 blocksize,
5725 u64 parent, u64 root_objectid,
5726 struct btrfs_disk_key *key, int level,
5727 u64 hint, u64 empty_size)
Chris Masonfec577f2007-02-26 10:40:21 -05005728{
Chris Masone2fa7222007-03-12 16:22:34 -04005729 struct btrfs_key ins;
Yan, Zhengf0486c62010-05-16 10:46:25 -04005730 struct btrfs_block_rsv *block_rsv;
Chris Mason5f39d392007-10-15 16:14:19 -04005731 struct extent_buffer *buf;
Yan, Zhengf0486c62010-05-16 10:46:25 -04005732 u64 flags = 0;
5733 int ret;
Chris Masonfec577f2007-02-26 10:40:21 -05005734
Yan, Zhengf0486c62010-05-16 10:46:25 -04005735
5736 block_rsv = use_block_rsv(trans, root, blocksize);
5737 if (IS_ERR(block_rsv))
5738 return ERR_CAST(block_rsv);
5739
5740 ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
5741 empty_size, hint, (u64)-1, &ins, 0);
Chris Masonfec577f2007-02-26 10:40:21 -05005742 if (ret) {
Yan, Zhengf0486c62010-05-16 10:46:25 -04005743 unuse_block_rsv(block_rsv, blocksize);
Chris Mason54aa1f42007-06-22 14:16:25 -04005744 return ERR_PTR(ret);
Chris Masonfec577f2007-02-26 10:40:21 -05005745 }
Chris Mason55c69072008-01-09 15:55:33 -05005746
Chris Mason4008c042009-02-12 14:09:45 -05005747 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
5748 blocksize, level);
Yan, Zhengf0486c62010-05-16 10:46:25 -04005749 BUG_ON(IS_ERR(buf));
5750
5751 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
5752 if (parent == 0)
5753 parent = ins.objectid;
5754 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
5755 } else
5756 BUG_ON(parent > 0);
5757
5758 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
5759 struct btrfs_delayed_extent_op *extent_op;
5760 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
5761 BUG_ON(!extent_op);
5762 if (key)
5763 memcpy(&extent_op->key, key, sizeof(extent_op->key));
5764 else
5765 memset(&extent_op->key, 0, sizeof(extent_op->key));
5766 extent_op->flags_to_set = flags;
5767 extent_op->update_key = 1;
5768 extent_op->update_flags = 1;
5769 extent_op->is_data = 0;
5770
5771 ret = btrfs_add_delayed_tree_ref(trans, ins.objectid,
5772 ins.offset, parent, root_objectid,
5773 level, BTRFS_ADD_DELAYED_EXTENT,
5774 extent_op);
5775 BUG_ON(ret);
5776 }
Chris Masonfec577f2007-02-26 10:40:21 -05005777 return buf;
5778}
Chris Masona28ec192007-03-06 20:08:01 -05005779
Yan Zheng2c47e6052009-06-27 21:07:35 -04005780struct walk_control {
5781 u64 refs[BTRFS_MAX_LEVEL];
5782 u64 flags[BTRFS_MAX_LEVEL];
5783 struct btrfs_key update_progress;
5784 int stage;
5785 int level;
5786 int shared_level;
5787 int update_ref;
5788 int keep_locks;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04005789 int reada_slot;
5790 int reada_count;
Yan Zheng2c47e6052009-06-27 21:07:35 -04005791};
5792
5793#define DROP_REFERENCE 1
5794#define UPDATE_BACKREF 2
5795
Yan, Zheng1c4850e2009-09-21 15:55:59 -04005796static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
5797 struct btrfs_root *root,
5798 struct walk_control *wc,
5799 struct btrfs_path *path)
5800{
5801 u64 bytenr;
5802 u64 generation;
5803 u64 refs;
Yan, Zheng94fcca92009-10-09 09:25:16 -04005804 u64 flags;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04005805 u32 nritems;
5806 u32 blocksize;
5807 struct btrfs_key key;
5808 struct extent_buffer *eb;
5809 int ret;
5810 int slot;
5811 int nread = 0;
5812
5813 if (path->slots[wc->level] < wc->reada_slot) {
5814 wc->reada_count = wc->reada_count * 2 / 3;
5815 wc->reada_count = max(wc->reada_count, 2);
5816 } else {
5817 wc->reada_count = wc->reada_count * 3 / 2;
5818 wc->reada_count = min_t(int, wc->reada_count,
5819 BTRFS_NODEPTRS_PER_BLOCK(root));
5820 }
5821
5822 eb = path->nodes[wc->level];
5823 nritems = btrfs_header_nritems(eb);
5824 blocksize = btrfs_level_size(root, wc->level - 1);
5825
5826 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
5827 if (nread >= wc->reada_count)
5828 break;
5829
5830 cond_resched();
5831 bytenr = btrfs_node_blockptr(eb, slot);
5832 generation = btrfs_node_ptr_generation(eb, slot);
5833
5834 if (slot == path->slots[wc->level])
5835 goto reada;
5836
5837 if (wc->stage == UPDATE_BACKREF &&
5838 generation <= root->root_key.offset)
5839 continue;
5840
Yan, Zheng94fcca92009-10-09 09:25:16 -04005841 /* We don't lock the tree block, it's OK to be racy here */
5842 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
5843 &refs, &flags);
5844 BUG_ON(ret);
5845 BUG_ON(refs == 0);
5846
Yan, Zheng1c4850e2009-09-21 15:55:59 -04005847 if (wc->stage == DROP_REFERENCE) {
Yan, Zheng1c4850e2009-09-21 15:55:59 -04005848 if (refs == 1)
5849 goto reada;
5850
Yan, Zheng94fcca92009-10-09 09:25:16 -04005851 if (wc->level == 1 &&
5852 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5853 continue;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04005854 if (!wc->update_ref ||
5855 generation <= root->root_key.offset)
5856 continue;
5857 btrfs_node_key_to_cpu(eb, &key, slot);
5858 ret = btrfs_comp_cpu_keys(&key,
5859 &wc->update_progress);
5860 if (ret < 0)
5861 continue;
Yan, Zheng94fcca92009-10-09 09:25:16 -04005862 } else {
5863 if (wc->level == 1 &&
5864 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5865 continue;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04005866 }
5867reada:
5868 ret = readahead_tree_block(root, bytenr, blocksize,
5869 generation);
5870 if (ret)
5871 break;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04005872 nread++;
5873 }
5874 wc->reada_slot = slot;
5875}
5876
Chris Mason9aca1d52007-03-13 11:09:37 -04005877/*
Yan Zheng2c47e6052009-06-27 21:07:35 -04005878 * hepler to process tree block while walking down the tree.
5879 *
Yan Zheng2c47e6052009-06-27 21:07:35 -04005880 * when wc->stage == UPDATE_BACKREF, this function updates
5881 * back refs for pointers in the block.
5882 *
5883 * NOTE: return value 1 means we should stop walking down.
Yan Zhengf82d02d2008-10-29 14:49:05 -04005884 */
Yan Zheng2c47e6052009-06-27 21:07:35 -04005885static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
5886 struct btrfs_root *root,
5887 struct btrfs_path *path,
Yan, Zheng94fcca92009-10-09 09:25:16 -04005888 struct walk_control *wc, int lookup_info)
Yan Zheng2c47e6052009-06-27 21:07:35 -04005889{
5890 int level = wc->level;
5891 struct extent_buffer *eb = path->nodes[level];
Yan Zheng2c47e6052009-06-27 21:07:35 -04005892 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5893 int ret;
5894
5895 if (wc->stage == UPDATE_BACKREF &&
5896 btrfs_header_owner(eb) != root->root_key.objectid)
5897 return 1;
5898
5899 /*
5900 * when reference count of tree block is 1, it won't increase
5901 * again. once full backref flag is set, we never clear it.
5902 */
Yan, Zheng94fcca92009-10-09 09:25:16 -04005903 if (lookup_info &&
5904 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
5905 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
Yan Zheng2c47e6052009-06-27 21:07:35 -04005906 BUG_ON(!path->locks[level]);
5907 ret = btrfs_lookup_extent_info(trans, root,
5908 eb->start, eb->len,
5909 &wc->refs[level],
5910 &wc->flags[level]);
5911 BUG_ON(ret);
5912 BUG_ON(wc->refs[level] == 0);
5913 }
5914
Yan Zheng2c47e6052009-06-27 21:07:35 -04005915 if (wc->stage == DROP_REFERENCE) {
5916 if (wc->refs[level] > 1)
5917 return 1;
5918
5919 if (path->locks[level] && !wc->keep_locks) {
5920 btrfs_tree_unlock(eb);
5921 path->locks[level] = 0;
5922 }
5923 return 0;
5924 }
5925
5926 /* wc->stage == UPDATE_BACKREF */
5927 if (!(wc->flags[level] & flag)) {
5928 BUG_ON(!path->locks[level]);
5929 ret = btrfs_inc_ref(trans, root, eb, 1);
5930 BUG_ON(ret);
5931 ret = btrfs_dec_ref(trans, root, eb, 0);
5932 BUG_ON(ret);
5933 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
5934 eb->len, flag, 0);
5935 BUG_ON(ret);
5936 wc->flags[level] |= flag;
5937 }
5938
5939 /*
5940 * the block is shared by multiple trees, so it's not good to
5941 * keep the tree lock
5942 */
5943 if (path->locks[level] && level > 0) {
5944 btrfs_tree_unlock(eb);
5945 path->locks[level] = 0;
5946 }
5947 return 0;
5948}
5949
5950/*
Yan, Zheng1c4850e2009-09-21 15:55:59 -04005951 * hepler to process tree block pointer.
5952 *
5953 * when wc->stage == DROP_REFERENCE, this function checks
5954 * reference count of the block pointed to. if the block
5955 * is shared and we need update back refs for the subtree
5956 * rooted at the block, this function changes wc->stage to
5957 * UPDATE_BACKREF. if the block is shared and there is no
5958 * need to update back, this function drops the reference
5959 * to the block.
5960 *
5961 * NOTE: return value 1 means we should stop walking down.
5962 */
5963static noinline int do_walk_down(struct btrfs_trans_handle *trans,
5964 struct btrfs_root *root,
5965 struct btrfs_path *path,
Yan, Zheng94fcca92009-10-09 09:25:16 -04005966 struct walk_control *wc, int *lookup_info)
Yan, Zheng1c4850e2009-09-21 15:55:59 -04005967{
5968 u64 bytenr;
5969 u64 generation;
5970 u64 parent;
5971 u32 blocksize;
5972 struct btrfs_key key;
5973 struct extent_buffer *next;
5974 int level = wc->level;
5975 int reada = 0;
5976 int ret = 0;
5977
5978 generation = btrfs_node_ptr_generation(path->nodes[level],
5979 path->slots[level]);
5980 /*
5981 * if the lower level block was created before the snapshot
5982 * was created, we know there is no need to update back refs
5983 * for the subtree
5984 */
5985 if (wc->stage == UPDATE_BACKREF &&
Yan, Zheng94fcca92009-10-09 09:25:16 -04005986 generation <= root->root_key.offset) {
5987 *lookup_info = 1;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04005988 return 1;
Yan, Zheng94fcca92009-10-09 09:25:16 -04005989 }
Yan, Zheng1c4850e2009-09-21 15:55:59 -04005990
5991 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
5992 blocksize = btrfs_level_size(root, level - 1);
5993
5994 next = btrfs_find_tree_block(root, bytenr, blocksize);
5995 if (!next) {
5996 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
Miao Xie90d2c512010-03-25 12:37:12 +00005997 if (!next)
5998 return -ENOMEM;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04005999 reada = 1;
6000 }
6001 btrfs_tree_lock(next);
6002 btrfs_set_lock_blocking(next);
6003
Yan, Zheng94fcca92009-10-09 09:25:16 -04006004 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6005 &wc->refs[level - 1],
6006 &wc->flags[level - 1]);
6007 BUG_ON(ret);
6008 BUG_ON(wc->refs[level - 1] == 0);
6009 *lookup_info = 0;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04006010
Yan, Zheng94fcca92009-10-09 09:25:16 -04006011 if (wc->stage == DROP_REFERENCE) {
Yan, Zheng1c4850e2009-09-21 15:55:59 -04006012 if (wc->refs[level - 1] > 1) {
Yan, Zheng94fcca92009-10-09 09:25:16 -04006013 if (level == 1 &&
6014 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6015 goto skip;
6016
Yan, Zheng1c4850e2009-09-21 15:55:59 -04006017 if (!wc->update_ref ||
6018 generation <= root->root_key.offset)
6019 goto skip;
6020
6021 btrfs_node_key_to_cpu(path->nodes[level], &key,
6022 path->slots[level]);
6023 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
6024 if (ret < 0)
6025 goto skip;
6026
6027 wc->stage = UPDATE_BACKREF;
6028 wc->shared_level = level - 1;
6029 }
Yan, Zheng94fcca92009-10-09 09:25:16 -04006030 } else {
6031 if (level == 1 &&
6032 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6033 goto skip;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04006034 }
6035
6036 if (!btrfs_buffer_uptodate(next, generation)) {
6037 btrfs_tree_unlock(next);
6038 free_extent_buffer(next);
6039 next = NULL;
Yan, Zheng94fcca92009-10-09 09:25:16 -04006040 *lookup_info = 1;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04006041 }
6042
6043 if (!next) {
6044 if (reada && level == 1)
6045 reada_walk_down(trans, root, wc, path);
6046 next = read_tree_block(root, bytenr, blocksize, generation);
Tsutomu Itoh97d9a8a2011-03-24 06:33:21 +00006047 if (!next)
6048 return -EIO;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04006049 btrfs_tree_lock(next);
6050 btrfs_set_lock_blocking(next);
6051 }
6052
6053 level--;
6054 BUG_ON(level != btrfs_header_level(next));
6055 path->nodes[level] = next;
6056 path->slots[level] = 0;
6057 path->locks[level] = 1;
6058 wc->level = level;
6059 if (wc->level == 1)
6060 wc->reada_slot = 0;
6061 return 0;
6062skip:
6063 wc->refs[level - 1] = 0;
6064 wc->flags[level - 1] = 0;
Yan, Zheng94fcca92009-10-09 09:25:16 -04006065 if (wc->stage == DROP_REFERENCE) {
6066 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
6067 parent = path->nodes[level]->start;
6068 } else {
6069 BUG_ON(root->root_key.objectid !=
6070 btrfs_header_owner(path->nodes[level]));
6071 parent = 0;
6072 }
Yan, Zheng1c4850e2009-09-21 15:55:59 -04006073
Yan, Zheng94fcca92009-10-09 09:25:16 -04006074 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
6075 root->root_key.objectid, level - 1, 0);
6076 BUG_ON(ret);
Yan, Zheng1c4850e2009-09-21 15:55:59 -04006077 }
Yan, Zheng1c4850e2009-09-21 15:55:59 -04006078 btrfs_tree_unlock(next);
6079 free_extent_buffer(next);
Yan, Zheng94fcca92009-10-09 09:25:16 -04006080 *lookup_info = 1;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04006081 return 1;
6082}
6083
6084/*
Yan Zheng2c47e6052009-06-27 21:07:35 -04006085 * hepler to process tree block while walking up the tree.
6086 *
6087 * when wc->stage == DROP_REFERENCE, this function drops
6088 * reference count on the block.
6089 *
6090 * when wc->stage == UPDATE_BACKREF, this function changes
6091 * wc->stage back to DROP_REFERENCE if we changed wc->stage
6092 * to UPDATE_BACKREF previously while processing the block.
6093 *
6094 * NOTE: return value 1 means we should stop walking up.
6095 */
6096static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
6097 struct btrfs_root *root,
6098 struct btrfs_path *path,
6099 struct walk_control *wc)
6100{
Yan, Zhengf0486c62010-05-16 10:46:25 -04006101 int ret;
Yan Zheng2c47e6052009-06-27 21:07:35 -04006102 int level = wc->level;
6103 struct extent_buffer *eb = path->nodes[level];
6104 u64 parent = 0;
6105
6106 if (wc->stage == UPDATE_BACKREF) {
6107 BUG_ON(wc->shared_level < level);
6108 if (level < wc->shared_level)
6109 goto out;
6110
Yan Zheng2c47e6052009-06-27 21:07:35 -04006111 ret = find_next_key(path, level + 1, &wc->update_progress);
6112 if (ret > 0)
6113 wc->update_ref = 0;
6114
6115 wc->stage = DROP_REFERENCE;
6116 wc->shared_level = -1;
6117 path->slots[level] = 0;
6118
6119 /*
6120 * check reference count again if the block isn't locked.
6121 * we should start walking down the tree again if reference
6122 * count is one.
6123 */
6124 if (!path->locks[level]) {
6125 BUG_ON(level == 0);
6126 btrfs_tree_lock(eb);
6127 btrfs_set_lock_blocking(eb);
6128 path->locks[level] = 1;
6129
6130 ret = btrfs_lookup_extent_info(trans, root,
6131 eb->start, eb->len,
6132 &wc->refs[level],
6133 &wc->flags[level]);
6134 BUG_ON(ret);
6135 BUG_ON(wc->refs[level] == 0);
6136 if (wc->refs[level] == 1) {
6137 btrfs_tree_unlock(eb);
6138 path->locks[level] = 0;
6139 return 1;
6140 }
Yan Zheng2c47e6052009-06-27 21:07:35 -04006141 }
6142 }
6143
6144 /* wc->stage == DROP_REFERENCE */
6145 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
6146
6147 if (wc->refs[level] == 1) {
6148 if (level == 0) {
6149 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6150 ret = btrfs_dec_ref(trans, root, eb, 1);
6151 else
6152 ret = btrfs_dec_ref(trans, root, eb, 0);
6153 BUG_ON(ret);
6154 }
6155 /* make block locked assertion in clean_tree_block happy */
6156 if (!path->locks[level] &&
6157 btrfs_header_generation(eb) == trans->transid) {
6158 btrfs_tree_lock(eb);
6159 btrfs_set_lock_blocking(eb);
6160 path->locks[level] = 1;
6161 }
6162 clean_tree_block(trans, root, eb);
6163 }
6164
6165 if (eb == root->node) {
6166 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6167 parent = eb->start;
6168 else
6169 BUG_ON(root->root_key.objectid !=
6170 btrfs_header_owner(eb));
6171 } else {
6172 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6173 parent = path->nodes[level + 1]->start;
6174 else
6175 BUG_ON(root->root_key.objectid !=
6176 btrfs_header_owner(path->nodes[level + 1]));
6177 }
6178
Yan, Zhengf0486c62010-05-16 10:46:25 -04006179 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
Yan Zheng2c47e6052009-06-27 21:07:35 -04006180out:
6181 wc->refs[level] = 0;
6182 wc->flags[level] = 0;
Yan, Zhengf0486c62010-05-16 10:46:25 -04006183 return 0;
Yan Zheng2c47e6052009-06-27 21:07:35 -04006184}
6185
Yan Zheng5d4f98a2009-06-10 10:45:14 -04006186static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
6187 struct btrfs_root *root,
Yan Zheng2c47e6052009-06-27 21:07:35 -04006188 struct btrfs_path *path,
6189 struct walk_control *wc)
Yan Zhengf82d02d2008-10-29 14:49:05 -04006190{
Yan Zheng2c47e6052009-06-27 21:07:35 -04006191 int level = wc->level;
Yan, Zheng94fcca92009-10-09 09:25:16 -04006192 int lookup_info = 1;
Yan Zhengf82d02d2008-10-29 14:49:05 -04006193 int ret;
6194
Yan Zheng2c47e6052009-06-27 21:07:35 -04006195 while (level >= 0) {
Yan, Zheng94fcca92009-10-09 09:25:16 -04006196 ret = walk_down_proc(trans, root, path, wc, lookup_info);
Yan Zheng2c47e6052009-06-27 21:07:35 -04006197 if (ret > 0)
Yan Zhengf82d02d2008-10-29 14:49:05 -04006198 break;
Yan Zhengf82d02d2008-10-29 14:49:05 -04006199
Yan Zheng2c47e6052009-06-27 21:07:35 -04006200 if (level == 0)
6201 break;
6202
Yan, Zheng7a7965f2010-02-01 02:41:17 +00006203 if (path->slots[level] >=
6204 btrfs_header_nritems(path->nodes[level]))
6205 break;
6206
Yan, Zheng94fcca92009-10-09 09:25:16 -04006207 ret = do_walk_down(trans, root, path, wc, &lookup_info);
Yan, Zheng1c4850e2009-09-21 15:55:59 -04006208 if (ret > 0) {
6209 path->slots[level]++;
6210 continue;
Miao Xie90d2c512010-03-25 12:37:12 +00006211 } else if (ret < 0)
6212 return ret;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04006213 level = wc->level;
Yan Zhengf82d02d2008-10-29 14:49:05 -04006214 }
Yan Zhengf82d02d2008-10-29 14:49:05 -04006215 return 0;
6216}
6217
Chris Masond3977122009-01-05 21:25:51 -05006218static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
Chris Mason98ed5172008-01-03 10:01:48 -05006219 struct btrfs_root *root,
Yan Zhengf82d02d2008-10-29 14:49:05 -04006220 struct btrfs_path *path,
Yan Zheng2c47e6052009-06-27 21:07:35 -04006221 struct walk_control *wc, int max_level)
Chris Mason20524f02007-03-10 06:35:47 -05006222{
Yan Zheng2c47e6052009-06-27 21:07:35 -04006223 int level = wc->level;
Chris Mason20524f02007-03-10 06:35:47 -05006224 int ret;
Chris Mason9f3a7422007-08-07 15:52:19 -04006225
Yan Zheng2c47e6052009-06-27 21:07:35 -04006226 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
6227 while (level < max_level && path->nodes[level]) {
6228 wc->level = level;
6229 if (path->slots[level] + 1 <
6230 btrfs_header_nritems(path->nodes[level])) {
6231 path->slots[level]++;
Chris Mason20524f02007-03-10 06:35:47 -05006232 return 0;
6233 } else {
Yan Zheng2c47e6052009-06-27 21:07:35 -04006234 ret = walk_up_proc(trans, root, path, wc);
6235 if (ret > 0)
6236 return 0;
Chris Masonbd56b302009-02-04 09:27:02 -05006237
Yan Zheng2c47e6052009-06-27 21:07:35 -04006238 if (path->locks[level]) {
6239 btrfs_tree_unlock(path->nodes[level]);
6240 path->locks[level] = 0;
Yan Zhengf82d02d2008-10-29 14:49:05 -04006241 }
Yan Zheng2c47e6052009-06-27 21:07:35 -04006242 free_extent_buffer(path->nodes[level]);
6243 path->nodes[level] = NULL;
6244 level++;
Chris Mason20524f02007-03-10 06:35:47 -05006245 }
6246 }
6247 return 1;
6248}
6249
Chris Mason9aca1d52007-03-13 11:09:37 -04006250/*
Yan Zheng2c47e6052009-06-27 21:07:35 -04006251 * drop a subvolume tree.
6252 *
6253 * this function traverses the tree freeing any blocks that only
6254 * referenced by the tree.
6255 *
6256 * when a shared tree block is found. this function decreases its
6257 * reference count by one. if update_ref is true, this function
6258 * also make sure backrefs for the shared block and all lower level
6259 * blocks are properly updated.
Chris Mason9aca1d52007-03-13 11:09:37 -04006260 */
Yan, Zheng3fd0a552010-05-16 10:49:59 -04006261int btrfs_drop_snapshot(struct btrfs_root *root,
6262 struct btrfs_block_rsv *block_rsv, int update_ref)
Chris Mason20524f02007-03-10 06:35:47 -05006263{
Chris Mason5caf2a02007-04-02 11:20:42 -04006264 struct btrfs_path *path;
Yan Zheng2c47e6052009-06-27 21:07:35 -04006265 struct btrfs_trans_handle *trans;
6266 struct btrfs_root *tree_root = root->fs_info->tree_root;
Chris Mason9f3a7422007-08-07 15:52:19 -04006267 struct btrfs_root_item *root_item = &root->root_item;
Yan Zheng2c47e6052009-06-27 21:07:35 -04006268 struct walk_control *wc;
6269 struct btrfs_key key;
6270 int err = 0;
6271 int ret;
6272 int level;
Chris Mason20524f02007-03-10 06:35:47 -05006273
Chris Mason5caf2a02007-04-02 11:20:42 -04006274 path = btrfs_alloc_path();
6275 BUG_ON(!path);
Chris Mason20524f02007-03-10 06:35:47 -05006276
Yan Zheng2c47e6052009-06-27 21:07:35 -04006277 wc = kzalloc(sizeof(*wc), GFP_NOFS);
6278 BUG_ON(!wc);
6279
Yan, Zhenga22285a2010-05-16 10:48:46 -04006280 trans = btrfs_start_transaction(tree_root, 0);
Tsutomu Itoh98d5dc12011-01-20 06:19:37 +00006281 BUG_ON(IS_ERR(trans));
6282
Yan, Zheng3fd0a552010-05-16 10:49:59 -04006283 if (block_rsv)
6284 trans->block_rsv = block_rsv;
Yan Zheng2c47e6052009-06-27 21:07:35 -04006285
Chris Mason9f3a7422007-08-07 15:52:19 -04006286 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
Yan Zheng2c47e6052009-06-27 21:07:35 -04006287 level = btrfs_header_level(root->node);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04006288 path->nodes[level] = btrfs_lock_root_node(root);
6289 btrfs_set_lock_blocking(path->nodes[level]);
Chris Mason9f3a7422007-08-07 15:52:19 -04006290 path->slots[level] = 0;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04006291 path->locks[level] = 1;
Yan Zheng2c47e6052009-06-27 21:07:35 -04006292 memset(&wc->update_progress, 0,
6293 sizeof(wc->update_progress));
Chris Mason9f3a7422007-08-07 15:52:19 -04006294 } else {
Chris Mason9f3a7422007-08-07 15:52:19 -04006295 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
Yan Zheng2c47e6052009-06-27 21:07:35 -04006296 memcpy(&wc->update_progress, &key,
6297 sizeof(wc->update_progress));
6298
Chris Mason6702ed42007-08-07 16:15:09 -04006299 level = root_item->drop_level;
Yan Zheng2c47e6052009-06-27 21:07:35 -04006300 BUG_ON(level == 0);
Chris Mason6702ed42007-08-07 16:15:09 -04006301 path->lowest_level = level;
Yan Zheng2c47e6052009-06-27 21:07:35 -04006302 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6303 path->lowest_level = 0;
6304 if (ret < 0) {
6305 err = ret;
Chris Mason9f3a7422007-08-07 15:52:19 -04006306 goto out;
6307 }
Yan, Zheng1c4850e2009-09-21 15:55:59 -04006308 WARN_ON(ret > 0);
Yan Zheng2c47e6052009-06-27 21:07:35 -04006309
Chris Mason7d9eb122008-07-08 14:19:17 -04006310 /*
6311 * unlock our path, this is safe because only this
6312 * function is allowed to delete this snapshot
6313 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -04006314 btrfs_unlock_up_safe(path, 0);
Chris Mason9aca1d52007-03-13 11:09:37 -04006315
Yan Zheng2c47e6052009-06-27 21:07:35 -04006316 level = btrfs_header_level(root->node);
6317 while (1) {
6318 btrfs_tree_lock(path->nodes[level]);
6319 btrfs_set_lock_blocking(path->nodes[level]);
6320
6321 ret = btrfs_lookup_extent_info(trans, root,
6322 path->nodes[level]->start,
6323 path->nodes[level]->len,
6324 &wc->refs[level],
6325 &wc->flags[level]);
6326 BUG_ON(ret);
6327 BUG_ON(wc->refs[level] == 0);
6328
6329 if (level == root_item->drop_level)
6330 break;
6331
6332 btrfs_tree_unlock(path->nodes[level]);
6333 WARN_ON(wc->refs[level] != 1);
6334 level--;
6335 }
6336 }
6337
6338 wc->level = level;
6339 wc->shared_level = -1;
6340 wc->stage = DROP_REFERENCE;
6341 wc->update_ref = update_ref;
6342 wc->keep_locks = 0;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04006343 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
Yan Zheng2c47e6052009-06-27 21:07:35 -04006344
6345 while (1) {
6346 ret = walk_down_tree(trans, root, path, wc);
6347 if (ret < 0) {
6348 err = ret;
Chris Masone7a84562008-06-25 16:01:31 -04006349 break;
6350 }
Yan Zheng2c47e6052009-06-27 21:07:35 -04006351
6352 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
6353 if (ret < 0) {
6354 err = ret;
6355 break;
6356 }
6357
6358 if (ret > 0) {
6359 BUG_ON(wc->stage != DROP_REFERENCE);
6360 break;
6361 }
6362
6363 if (wc->stage == DROP_REFERENCE) {
6364 level = wc->level;
6365 btrfs_node_key(path->nodes[level],
6366 &root_item->drop_progress,
6367 path->slots[level]);
6368 root_item->drop_level = level;
6369 }
6370
6371 BUG_ON(wc->level == 0);
Yan, Zheng3fd0a552010-05-16 10:49:59 -04006372 if (btrfs_should_end_transaction(trans, tree_root)) {
Yan Zheng2c47e6052009-06-27 21:07:35 -04006373 ret = btrfs_update_root(trans, tree_root,
6374 &root->root_key,
6375 root_item);
6376 BUG_ON(ret);
6377
Yan, Zheng3fd0a552010-05-16 10:49:59 -04006378 btrfs_end_transaction_throttle(trans, tree_root);
Yan, Zhenga22285a2010-05-16 10:48:46 -04006379 trans = btrfs_start_transaction(tree_root, 0);
Tsutomu Itoh98d5dc12011-01-20 06:19:37 +00006380 BUG_ON(IS_ERR(trans));
Yan, Zheng3fd0a552010-05-16 10:49:59 -04006381 if (block_rsv)
6382 trans->block_rsv = block_rsv;
Chris Masonc3e69d52009-03-13 10:17:05 -04006383 }
Chris Mason20524f02007-03-10 06:35:47 -05006384 }
David Sterbab3b4aa72011-04-21 01:20:15 +02006385 btrfs_release_path(path);
Yan Zheng2c47e6052009-06-27 21:07:35 -04006386 BUG_ON(err);
6387
6388 ret = btrfs_del_root(trans, tree_root, &root->root_key);
6389 BUG_ON(ret);
6390
Yan, Zheng76dda932009-09-21 16:00:26 -04006391 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
6392 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
6393 NULL, NULL);
6394 BUG_ON(ret < 0);
6395 if (ret > 0) {
Josef Bacik84cd9482010-12-08 12:24:01 -05006396 /* if we fail to delete the orphan item this time
6397 * around, it'll get picked up the next time.
6398 *
6399 * The most common failure here is just -ENOENT.
6400 */
6401 btrfs_del_orphan_item(trans, tree_root,
6402 root->root_key.objectid);
Yan, Zheng76dda932009-09-21 16:00:26 -04006403 }
6404 }
6405
6406 if (root->in_radix) {
6407 btrfs_free_fs_root(tree_root->fs_info, root);
6408 } else {
6409 free_extent_buffer(root->node);
6410 free_extent_buffer(root->commit_root);
6411 kfree(root);
6412 }
Chris Mason9f3a7422007-08-07 15:52:19 -04006413out:
Yan, Zheng3fd0a552010-05-16 10:49:59 -04006414 btrfs_end_transaction_throttle(trans, tree_root);
Yan Zheng2c47e6052009-06-27 21:07:35 -04006415 kfree(wc);
Chris Mason5caf2a02007-04-02 11:20:42 -04006416 btrfs_free_path(path);
Yan Zheng2c47e6052009-06-27 21:07:35 -04006417 return err;
Chris Mason20524f02007-03-10 06:35:47 -05006418}
Chris Mason9078a3e2007-04-26 16:46:15 -04006419
Yan Zheng2c47e6052009-06-27 21:07:35 -04006420/*
6421 * drop subtree rooted at tree block 'node'.
6422 *
6423 * NOTE: this function will unlock and release tree block 'node'
6424 */
Yan Zhengf82d02d2008-10-29 14:49:05 -04006425int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
6426 struct btrfs_root *root,
6427 struct extent_buffer *node,
6428 struct extent_buffer *parent)
6429{
6430 struct btrfs_path *path;
Yan Zheng2c47e6052009-06-27 21:07:35 -04006431 struct walk_control *wc;
Yan Zhengf82d02d2008-10-29 14:49:05 -04006432 int level;
6433 int parent_level;
6434 int ret = 0;
6435 int wret;
6436
Yan Zheng2c47e6052009-06-27 21:07:35 -04006437 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6438
Yan Zhengf82d02d2008-10-29 14:49:05 -04006439 path = btrfs_alloc_path();
Tsutomu Itohdb5b4932011-03-23 08:14:16 +00006440 if (!path)
6441 return -ENOMEM;
Yan Zhengf82d02d2008-10-29 14:49:05 -04006442
Yan Zheng2c47e6052009-06-27 21:07:35 -04006443 wc = kzalloc(sizeof(*wc), GFP_NOFS);
Tsutomu Itohdb5b4932011-03-23 08:14:16 +00006444 if (!wc) {
6445 btrfs_free_path(path);
6446 return -ENOMEM;
6447 }
Yan Zheng2c47e6052009-06-27 21:07:35 -04006448
Chris Masonb9447ef2009-03-09 11:45:38 -04006449 btrfs_assert_tree_locked(parent);
Yan Zhengf82d02d2008-10-29 14:49:05 -04006450 parent_level = btrfs_header_level(parent);
6451 extent_buffer_get(parent);
6452 path->nodes[parent_level] = parent;
6453 path->slots[parent_level] = btrfs_header_nritems(parent);
6454
Chris Masonb9447ef2009-03-09 11:45:38 -04006455 btrfs_assert_tree_locked(node);
Yan Zhengf82d02d2008-10-29 14:49:05 -04006456 level = btrfs_header_level(node);
Yan Zhengf82d02d2008-10-29 14:49:05 -04006457 path->nodes[level] = node;
6458 path->slots[level] = 0;
Yan Zheng2c47e6052009-06-27 21:07:35 -04006459 path->locks[level] = 1;
6460
6461 wc->refs[parent_level] = 1;
6462 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6463 wc->level = level;
6464 wc->shared_level = -1;
6465 wc->stage = DROP_REFERENCE;
6466 wc->update_ref = 0;
6467 wc->keep_locks = 1;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04006468 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
Yan Zhengf82d02d2008-10-29 14:49:05 -04006469
6470 while (1) {
Yan Zheng2c47e6052009-06-27 21:07:35 -04006471 wret = walk_down_tree(trans, root, path, wc);
6472 if (wret < 0) {
Yan Zhengf82d02d2008-10-29 14:49:05 -04006473 ret = wret;
Yan Zhengf82d02d2008-10-29 14:49:05 -04006474 break;
Yan Zheng2c47e6052009-06-27 21:07:35 -04006475 }
Yan Zhengf82d02d2008-10-29 14:49:05 -04006476
Yan Zheng2c47e6052009-06-27 21:07:35 -04006477 wret = walk_up_tree(trans, root, path, wc, parent_level);
Yan Zhengf82d02d2008-10-29 14:49:05 -04006478 if (wret < 0)
6479 ret = wret;
6480 if (wret != 0)
6481 break;
6482 }
6483
Yan Zheng2c47e6052009-06-27 21:07:35 -04006484 kfree(wc);
Yan Zhengf82d02d2008-10-29 14:49:05 -04006485 btrfs_free_path(path);
6486 return ret;
6487}
6488
Chris Masonec44a352008-04-28 15:29:52 -04006489static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
6490{
6491 u64 num_devices;
6492 u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
6493 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
6494
Chris Masoncd02dca2010-12-13 14:56:23 -05006495 /*
6496 * we add in the count of missing devices because we want
6497 * to make sure that any RAID levels on a degraded FS
6498 * continue to be honored.
6499 */
6500 num_devices = root->fs_info->fs_devices->rw_devices +
6501 root->fs_info->fs_devices->missing_devices;
6502
Chris Masonec44a352008-04-28 15:29:52 -04006503 if (num_devices == 1) {
6504 stripped |= BTRFS_BLOCK_GROUP_DUP;
6505 stripped = flags & ~stripped;
6506
6507 /* turn raid0 into single device chunks */
6508 if (flags & BTRFS_BLOCK_GROUP_RAID0)
6509 return stripped;
6510
6511 /* turn mirroring into duplication */
6512 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
6513 BTRFS_BLOCK_GROUP_RAID10))
6514 return stripped | BTRFS_BLOCK_GROUP_DUP;
6515 return flags;
6516 } else {
6517 /* they already had raid on here, just return */
Chris Masonec44a352008-04-28 15:29:52 -04006518 if (flags & stripped)
6519 return flags;
6520
6521 stripped |= BTRFS_BLOCK_GROUP_DUP;
6522 stripped = flags & ~stripped;
6523
6524 /* switch duplicated blocks with raid1 */
6525 if (flags & BTRFS_BLOCK_GROUP_DUP)
6526 return stripped | BTRFS_BLOCK_GROUP_RAID1;
6527
6528 /* turn single device chunks into raid0 */
6529 return stripped | BTRFS_BLOCK_GROUP_RAID0;
6530 }
6531 return flags;
6532}
6533
Yan, Zhengf0486c62010-05-16 10:46:25 -04006534static int set_block_group_ro(struct btrfs_block_group_cache *cache)
Chris Mason0ef3e662008-05-24 14:04:53 -04006535{
Yan, Zhengf0486c62010-05-16 10:46:25 -04006536 struct btrfs_space_info *sinfo = cache->space_info;
6537 u64 num_bytes;
6538 int ret = -ENOSPC;
Chris Mason0ef3e662008-05-24 14:04:53 -04006539
Yan, Zhengf0486c62010-05-16 10:46:25 -04006540 if (cache->ro)
6541 return 0;
Chris Masonc286ac42008-07-22 23:06:41 -04006542
Yan, Zhengf0486c62010-05-16 10:46:25 -04006543 spin_lock(&sinfo->lock);
6544 spin_lock(&cache->lock);
6545 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
6546 cache->bytes_super - btrfs_block_group_used(&cache->item);
Chris Mason7d9eb122008-07-08 14:19:17 -04006547
Yan, Zhengf0486c62010-05-16 10:46:25 -04006548 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
6549 sinfo->bytes_may_use + sinfo->bytes_readonly +
Chris Mason65e53412010-12-24 06:41:52 -05006550 cache->reserved_pinned + num_bytes <= sinfo->total_bytes) {
Yan, Zhengf0486c62010-05-16 10:46:25 -04006551 sinfo->bytes_readonly += num_bytes;
6552 sinfo->bytes_reserved += cache->reserved_pinned;
6553 cache->reserved_pinned = 0;
6554 cache->ro = 1;
6555 ret = 0;
6556 }
Chris Mason65e53412010-12-24 06:41:52 -05006557
Yan, Zhengf0486c62010-05-16 10:46:25 -04006558 spin_unlock(&cache->lock);
6559 spin_unlock(&sinfo->lock);
6560 return ret;
Chris Mason0ef3e662008-05-24 14:04:53 -04006561}
6562
Yan, Zhengf0486c62010-05-16 10:46:25 -04006563int btrfs_set_block_group_ro(struct btrfs_root *root,
6564 struct btrfs_block_group_cache *cache)
Yan Zheng5d4f98a2009-06-10 10:45:14 -04006565
6566{
Yan, Zhengf0486c62010-05-16 10:46:25 -04006567 struct btrfs_trans_handle *trans;
6568 u64 alloc_flags;
6569 int ret;
6570
6571 BUG_ON(cache->ro);
6572
Josef Bacik7a7eaa42011-04-13 12:54:33 -04006573 trans = btrfs_join_transaction(root);
Yan, Zhengf0486c62010-05-16 10:46:25 -04006574 BUG_ON(IS_ERR(trans));
6575
6576 alloc_flags = update_block_group_flags(root, cache->flags);
6577 if (alloc_flags != cache->flags)
Chris Mason0e4f8f82011-04-15 16:05:44 -04006578 do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
6579 CHUNK_ALLOC_FORCE);
Yan, Zhengf0486c62010-05-16 10:46:25 -04006580
6581 ret = set_block_group_ro(cache);
6582 if (!ret)
6583 goto out;
6584 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
Chris Mason0e4f8f82011-04-15 16:05:44 -04006585 ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
6586 CHUNK_ALLOC_FORCE);
Yan, Zhengf0486c62010-05-16 10:46:25 -04006587 if (ret < 0)
6588 goto out;
6589 ret = set_block_group_ro(cache);
6590out:
6591 btrfs_end_transaction(trans, root);
6592 return ret;
6593}
6594
Chris Masonc87f08c2011-02-16 13:57:04 -05006595int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
6596 struct btrfs_root *root, u64 type)
6597{
6598 u64 alloc_flags = get_alloc_profile(root, type);
Chris Mason0e4f8f82011-04-15 16:05:44 -04006599 return do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
6600 CHUNK_ALLOC_FORCE);
Chris Masonc87f08c2011-02-16 13:57:04 -05006601}
6602
Miao Xie6d07bce2011-01-05 10:07:31 +00006603/*
6604 * helper to account the unused space of all the readonly block group in the
6605 * list. takes mirrors into account.
6606 */
6607static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
6608{
6609 struct btrfs_block_group_cache *block_group;
6610 u64 free_bytes = 0;
6611 int factor;
6612
6613 list_for_each_entry(block_group, groups_list, list) {
6614 spin_lock(&block_group->lock);
6615
6616 if (!block_group->ro) {
6617 spin_unlock(&block_group->lock);
6618 continue;
6619 }
6620
6621 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
6622 BTRFS_BLOCK_GROUP_RAID10 |
6623 BTRFS_BLOCK_GROUP_DUP))
6624 factor = 2;
6625 else
6626 factor = 1;
6627
6628 free_bytes += (block_group->key.offset -
6629 btrfs_block_group_used(&block_group->item)) *
6630 factor;
6631
6632 spin_unlock(&block_group->lock);
6633 }
6634
6635 return free_bytes;
6636}
6637
6638/*
6639 * helper to account the unused space of all the readonly block group in the
6640 * space_info. takes mirrors into account.
6641 */
6642u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
6643{
6644 int i;
6645 u64 free_bytes = 0;
6646
6647 spin_lock(&sinfo->lock);
6648
6649 for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
6650 if (!list_empty(&sinfo->block_groups[i]))
6651 free_bytes += __btrfs_get_ro_block_group_free_space(
6652 &sinfo->block_groups[i]);
6653
6654 spin_unlock(&sinfo->lock);
6655
6656 return free_bytes;
6657}
6658
Yan, Zhengf0486c62010-05-16 10:46:25 -04006659int btrfs_set_block_group_rw(struct btrfs_root *root,
6660 struct btrfs_block_group_cache *cache)
6661{
6662 struct btrfs_space_info *sinfo = cache->space_info;
6663 u64 num_bytes;
6664
6665 BUG_ON(!cache->ro);
6666
6667 spin_lock(&sinfo->lock);
6668 spin_lock(&cache->lock);
6669 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
6670 cache->bytes_super - btrfs_block_group_used(&cache->item);
6671 sinfo->bytes_readonly -= num_bytes;
6672 cache->ro = 0;
6673 spin_unlock(&cache->lock);
6674 spin_unlock(&sinfo->lock);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04006675 return 0;
6676}
6677
Josef Bacikba1bf482009-09-11 16:11:19 -04006678/*
6679 * checks to see if its even possible to relocate this block group.
6680 *
6681 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
6682 * ok to go ahead and try.
6683 */
6684int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
Zheng Yan1a40e232008-09-26 10:09:34 -04006685{
Zheng Yan1a40e232008-09-26 10:09:34 -04006686 struct btrfs_block_group_cache *block_group;
Josef Bacikba1bf482009-09-11 16:11:19 -04006687 struct btrfs_space_info *space_info;
6688 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6689 struct btrfs_device *device;
6690 int full = 0;
6691 int ret = 0;
Chris Masonedbd8d42007-12-21 16:27:24 -05006692
Josef Bacikba1bf482009-09-11 16:11:19 -04006693 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
Zheng Yan1a40e232008-09-26 10:09:34 -04006694
Josef Bacikba1bf482009-09-11 16:11:19 -04006695 /* odd, couldn't find the block group, leave it alone */
6696 if (!block_group)
6697 return -1;
Chris Masonedbd8d42007-12-21 16:27:24 -05006698
Josef Bacikba1bf482009-09-11 16:11:19 -04006699 /* no bytes used, we're good */
6700 if (!btrfs_block_group_used(&block_group->item))
6701 goto out;
Chris Mason323da792008-05-09 11:46:48 -04006702
Josef Bacikba1bf482009-09-11 16:11:19 -04006703 space_info = block_group->space_info;
6704 spin_lock(&space_info->lock);
Chris Mason323da792008-05-09 11:46:48 -04006705
Josef Bacikba1bf482009-09-11 16:11:19 -04006706 full = space_info->full;
Zheng Yan1a40e232008-09-26 10:09:34 -04006707
Josef Bacikba1bf482009-09-11 16:11:19 -04006708 /*
6709 * if this is the last block group we have in this space, we can't
Chris Mason7ce618d2009-09-22 14:48:44 -04006710 * relocate it unless we're able to allocate a new chunk below.
6711 *
6712 * Otherwise, we need to make sure we have room in the space to handle
6713 * all of the extents from this block group. If we can, we're good
Josef Bacikba1bf482009-09-11 16:11:19 -04006714 */
Chris Mason7ce618d2009-09-22 14:48:44 -04006715 if ((space_info->total_bytes != block_group->key.offset) &&
6716 (space_info->bytes_used + space_info->bytes_reserved +
Josef Bacikba1bf482009-09-11 16:11:19 -04006717 space_info->bytes_pinned + space_info->bytes_readonly +
6718 btrfs_block_group_used(&block_group->item) <
Chris Mason7ce618d2009-09-22 14:48:44 -04006719 space_info->total_bytes)) {
Josef Bacikba1bf482009-09-11 16:11:19 -04006720 spin_unlock(&space_info->lock);
6721 goto out;
6722 }
6723 spin_unlock(&space_info->lock);
Zheng Yan1a40e232008-09-26 10:09:34 -04006724
Josef Bacikba1bf482009-09-11 16:11:19 -04006725 /*
6726 * ok we don't have enough space, but maybe we have free space on our
6727 * devices to allocate new chunks for relocation, so loop through our
6728 * alloc devices and guess if we have enough space. However, if we
6729 * were marked as full, then we know there aren't enough chunks, and we
6730 * can just return.
6731 */
6732 ret = -1;
6733 if (full)
6734 goto out;
Chris Mason4313b392008-01-03 09:08:48 -05006735
Josef Bacikba1bf482009-09-11 16:11:19 -04006736 mutex_lock(&root->fs_info->chunk_mutex);
6737 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
6738 u64 min_free = btrfs_block_group_used(&block_group->item);
Miao Xie7bfc8372011-01-05 10:07:26 +00006739 u64 dev_offset;
Chris Masonea8c2812008-08-04 23:17:27 -04006740
Josef Bacikba1bf482009-09-11 16:11:19 -04006741 /*
6742 * check to make sure we can actually find a chunk with enough
6743 * space to fit our block group in.
6744 */
6745 if (device->total_bytes > device->bytes_used + min_free) {
6746 ret = find_free_dev_extent(NULL, device, min_free,
Miao Xie7bfc8372011-01-05 10:07:26 +00006747 &dev_offset, NULL);
Josef Bacikba1bf482009-09-11 16:11:19 -04006748 if (!ret)
Yan73e48b22008-01-03 14:14:39 -05006749 break;
Josef Bacikba1bf482009-09-11 16:11:19 -04006750 ret = -1;
Yan73e48b22008-01-03 14:14:39 -05006751 }
Chris Masonedbd8d42007-12-21 16:27:24 -05006752 }
Josef Bacikba1bf482009-09-11 16:11:19 -04006753 mutex_unlock(&root->fs_info->chunk_mutex);
Chris Masonedbd8d42007-12-21 16:27:24 -05006754out:
Josef Bacikba1bf482009-09-11 16:11:19 -04006755 btrfs_put_block_group(block_group);
Chris Masonedbd8d42007-12-21 16:27:24 -05006756 return ret;
6757}
6758
Christoph Hellwigb2950862008-12-02 09:54:17 -05006759static int find_first_block_group(struct btrfs_root *root,
6760 struct btrfs_path *path, struct btrfs_key *key)
Chris Mason0b86a832008-03-24 15:01:56 -04006761{
Chris Mason925baed2008-06-25 16:01:30 -04006762 int ret = 0;
Chris Mason0b86a832008-03-24 15:01:56 -04006763 struct btrfs_key found_key;
6764 struct extent_buffer *leaf;
6765 int slot;
6766
6767 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
6768 if (ret < 0)
Chris Mason925baed2008-06-25 16:01:30 -04006769 goto out;
6770
Chris Masond3977122009-01-05 21:25:51 -05006771 while (1) {
Chris Mason0b86a832008-03-24 15:01:56 -04006772 slot = path->slots[0];
6773 leaf = path->nodes[0];
6774 if (slot >= btrfs_header_nritems(leaf)) {
6775 ret = btrfs_next_leaf(root, path);
6776 if (ret == 0)
6777 continue;
6778 if (ret < 0)
Chris Mason925baed2008-06-25 16:01:30 -04006779 goto out;
Chris Mason0b86a832008-03-24 15:01:56 -04006780 break;
6781 }
6782 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6783
6784 if (found_key.objectid >= key->objectid &&
Chris Mason925baed2008-06-25 16:01:30 -04006785 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
6786 ret = 0;
6787 goto out;
6788 }
Chris Mason0b86a832008-03-24 15:01:56 -04006789 path->slots[0]++;
6790 }
Chris Mason925baed2008-06-25 16:01:30 -04006791out:
Chris Mason0b86a832008-03-24 15:01:56 -04006792 return ret;
6793}
6794
Josef Bacik0af3d002010-06-21 14:48:16 -04006795void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
6796{
6797 struct btrfs_block_group_cache *block_group;
6798 u64 last = 0;
6799
6800 while (1) {
6801 struct inode *inode;
6802
6803 block_group = btrfs_lookup_first_block_group(info, last);
6804 while (block_group) {
6805 spin_lock(&block_group->lock);
6806 if (block_group->iref)
6807 break;
6808 spin_unlock(&block_group->lock);
6809 block_group = next_block_group(info->tree_root,
6810 block_group);
6811 }
6812 if (!block_group) {
6813 if (last == 0)
6814 break;
6815 last = 0;
6816 continue;
6817 }
6818
6819 inode = block_group->inode;
6820 block_group->iref = 0;
6821 block_group->inode = NULL;
6822 spin_unlock(&block_group->lock);
6823 iput(inode);
6824 last = block_group->key.objectid + block_group->key.offset;
6825 btrfs_put_block_group(block_group);
6826 }
6827}
6828
Zheng Yan1a40e232008-09-26 10:09:34 -04006829int btrfs_free_block_groups(struct btrfs_fs_info *info)
6830{
6831 struct btrfs_block_group_cache *block_group;
Chris Mason4184ea72009-03-10 12:39:20 -04006832 struct btrfs_space_info *space_info;
Yan Zheng11833d62009-09-11 16:11:19 -04006833 struct btrfs_caching_control *caching_ctl;
Zheng Yan1a40e232008-09-26 10:09:34 -04006834 struct rb_node *n;
6835
Yan Zheng11833d62009-09-11 16:11:19 -04006836 down_write(&info->extent_commit_sem);
6837 while (!list_empty(&info->caching_block_groups)) {
6838 caching_ctl = list_entry(info->caching_block_groups.next,
6839 struct btrfs_caching_control, list);
6840 list_del(&caching_ctl->list);
6841 put_caching_control(caching_ctl);
6842 }
6843 up_write(&info->extent_commit_sem);
6844
Zheng Yan1a40e232008-09-26 10:09:34 -04006845 spin_lock(&info->block_group_cache_lock);
6846 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
6847 block_group = rb_entry(n, struct btrfs_block_group_cache,
6848 cache_node);
Zheng Yan1a40e232008-09-26 10:09:34 -04006849 rb_erase(&block_group->cache_node,
6850 &info->block_group_cache_tree);
Yan Zhengd899e052008-10-30 14:25:28 -04006851 spin_unlock(&info->block_group_cache_lock);
6852
Josef Bacik80eb2342008-10-29 14:49:05 -04006853 down_write(&block_group->space_info->groups_sem);
Zheng Yan1a40e232008-09-26 10:09:34 -04006854 list_del(&block_group->list);
Josef Bacik80eb2342008-10-29 14:49:05 -04006855 up_write(&block_group->space_info->groups_sem);
Yan Zhengd2fb3432008-12-11 16:30:39 -05006856
Josef Bacik817d52f2009-07-13 21:29:25 -04006857 if (block_group->cached == BTRFS_CACHE_STARTED)
Yan Zheng11833d62009-09-11 16:11:19 -04006858 wait_block_group_cache_done(block_group);
Josef Bacik817d52f2009-07-13 21:29:25 -04006859
Josef Bacik3c148742011-02-02 15:53:47 +00006860 /*
6861 * We haven't cached this block group, which means we could
6862 * possibly have excluded extents on this block group.
6863 */
6864 if (block_group->cached == BTRFS_CACHE_NO)
6865 free_excluded_extents(info->extent_root, block_group);
6866
Josef Bacik817d52f2009-07-13 21:29:25 -04006867 btrfs_remove_free_space_cache(block_group);
Josef Bacik11dfe352009-11-13 20:12:59 +00006868 btrfs_put_block_group(block_group);
Yan Zhengd899e052008-10-30 14:25:28 -04006869
6870 spin_lock(&info->block_group_cache_lock);
Zheng Yan1a40e232008-09-26 10:09:34 -04006871 }
6872 spin_unlock(&info->block_group_cache_lock);
Chris Mason4184ea72009-03-10 12:39:20 -04006873
6874 /* now that all the block groups are freed, go through and
6875 * free all the space_info structs. This is only called during
6876 * the final stages of unmount, and so we know nobody is
6877 * using them. We call synchronize_rcu() once before we start,
6878 * just to be on the safe side.
6879 */
6880 synchronize_rcu();
6881
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04006882 release_global_block_rsv(info);
6883
Chris Mason4184ea72009-03-10 12:39:20 -04006884 while(!list_empty(&info->space_info)) {
6885 space_info = list_entry(info->space_info.next,
6886 struct btrfs_space_info,
6887 list);
Yan, Zhengf0486c62010-05-16 10:46:25 -04006888 if (space_info->bytes_pinned > 0 ||
6889 space_info->bytes_reserved > 0) {
6890 WARN_ON(1);
6891 dump_space_info(space_info, 0, 0);
6892 }
Chris Mason4184ea72009-03-10 12:39:20 -04006893 list_del(&space_info->list);
6894 kfree(space_info);
6895 }
Zheng Yan1a40e232008-09-26 10:09:34 -04006896 return 0;
6897}
6898
Yan, Zhengb742bb82010-05-16 10:46:24 -04006899static void __link_block_group(struct btrfs_space_info *space_info,
6900 struct btrfs_block_group_cache *cache)
6901{
6902 int index = get_block_group_index(cache);
6903
6904 down_write(&space_info->groups_sem);
6905 list_add_tail(&cache->list, &space_info->block_groups[index]);
6906 up_write(&space_info->groups_sem);
6907}
6908
Chris Mason9078a3e2007-04-26 16:46:15 -04006909int btrfs_read_block_groups(struct btrfs_root *root)
6910{
6911 struct btrfs_path *path;
6912 int ret;
Chris Mason9078a3e2007-04-26 16:46:15 -04006913 struct btrfs_block_group_cache *cache;
Chris Masonbe744172007-05-06 10:15:01 -04006914 struct btrfs_fs_info *info = root->fs_info;
Chris Mason6324fbf2008-03-24 15:01:59 -04006915 struct btrfs_space_info *space_info;
Chris Mason9078a3e2007-04-26 16:46:15 -04006916 struct btrfs_key key;
6917 struct btrfs_key found_key;
Chris Mason5f39d392007-10-15 16:14:19 -04006918 struct extent_buffer *leaf;
Josef Bacik0af3d002010-06-21 14:48:16 -04006919 int need_clear = 0;
6920 u64 cache_gen;
Chris Mason96b51792007-10-15 16:15:19 -04006921
Chris Masonbe744172007-05-06 10:15:01 -04006922 root = info->extent_root;
Chris Mason9078a3e2007-04-26 16:46:15 -04006923 key.objectid = 0;
Chris Mason0b86a832008-03-24 15:01:56 -04006924 key.offset = 0;
Chris Mason9078a3e2007-04-26 16:46:15 -04006925 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
Chris Mason9078a3e2007-04-26 16:46:15 -04006926 path = btrfs_alloc_path();
6927 if (!path)
6928 return -ENOMEM;
Josef Bacik026fd312011-05-13 10:32:11 -04006929 path->reada = 1;
Chris Mason9078a3e2007-04-26 16:46:15 -04006930
Josef Bacik0af3d002010-06-21 14:48:16 -04006931 cache_gen = btrfs_super_cache_generation(&root->fs_info->super_copy);
6932 if (cache_gen != 0 &&
6933 btrfs_super_generation(&root->fs_info->super_copy) != cache_gen)
6934 need_clear = 1;
Josef Bacik88c2ba32010-09-21 14:21:34 -04006935 if (btrfs_test_opt(root, CLEAR_CACHE))
6936 need_clear = 1;
Josef Bacik8216ef82010-10-28 16:55:47 -04006937 if (!btrfs_test_opt(root, SPACE_CACHE) && cache_gen)
6938 printk(KERN_INFO "btrfs: disk space caching is enabled\n");
Josef Bacik0af3d002010-06-21 14:48:16 -04006939
Chris Masond3977122009-01-05 21:25:51 -05006940 while (1) {
Chris Mason0b86a832008-03-24 15:01:56 -04006941 ret = find_first_block_group(root, path, &key);
Yan, Zhengb742bb82010-05-16 10:46:24 -04006942 if (ret > 0)
6943 break;
Chris Mason0b86a832008-03-24 15:01:56 -04006944 if (ret != 0)
6945 goto error;
Chris Mason5f39d392007-10-15 16:14:19 -04006946 leaf = path->nodes[0];
6947 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
Chris Mason8f18cf12008-04-25 16:53:30 -04006948 cache = kzalloc(sizeof(*cache), GFP_NOFS);
Chris Mason9078a3e2007-04-26 16:46:15 -04006949 if (!cache) {
Chris Mason0b86a832008-03-24 15:01:56 -04006950 ret = -ENOMEM;
Yan, Zhengf0486c62010-05-16 10:46:25 -04006951 goto error;
Chris Mason9078a3e2007-04-26 16:46:15 -04006952 }
Li Zefan34d52cb2011-03-29 13:46:06 +08006953 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
6954 GFP_NOFS);
6955 if (!cache->free_space_ctl) {
6956 kfree(cache);
6957 ret = -ENOMEM;
6958 goto error;
6959 }
Chris Mason3e1ad542007-05-07 20:03:49 -04006960
Yan Zhengd2fb3432008-12-11 16:30:39 -05006961 atomic_set(&cache->count, 1);
Chris Masonc286ac42008-07-22 23:06:41 -04006962 spin_lock_init(&cache->lock);
Josef Bacik817d52f2009-07-13 21:29:25 -04006963 cache->fs_info = info;
Josef Bacik0f9dd462008-09-23 13:14:11 -04006964 INIT_LIST_HEAD(&cache->list);
Chris Masonfa9c0d72009-04-03 09:47:43 -04006965 INIT_LIST_HEAD(&cache->cluster_list);
Josef Bacik96303082009-07-13 21:29:25 -04006966
Josef Bacik0af3d002010-06-21 14:48:16 -04006967 if (need_clear)
6968 cache->disk_cache_state = BTRFS_DC_CLEAR;
6969
Chris Mason5f39d392007-10-15 16:14:19 -04006970 read_extent_buffer(leaf, &cache->item,
6971 btrfs_item_ptr_offset(leaf, path->slots[0]),
6972 sizeof(cache->item));
Chris Mason9078a3e2007-04-26 16:46:15 -04006973 memcpy(&cache->key, &found_key, sizeof(found_key));
Chris Mason0b86a832008-03-24 15:01:56 -04006974
Chris Mason9078a3e2007-04-26 16:46:15 -04006975 key.objectid = found_key.objectid + found_key.offset;
David Sterbab3b4aa72011-04-21 01:20:15 +02006976 btrfs_release_path(path);
Chris Mason0b86a832008-03-24 15:01:56 -04006977 cache->flags = btrfs_block_group_flags(&cache->item);
Josef Bacik817d52f2009-07-13 21:29:25 -04006978 cache->sectorsize = root->sectorsize;
6979
Li Zefan34d52cb2011-03-29 13:46:06 +08006980 btrfs_init_free_space_ctl(cache);
6981
Josef Bacik817d52f2009-07-13 21:29:25 -04006982 /*
Josef Bacik3c148742011-02-02 15:53:47 +00006983 * We need to exclude the super stripes now so that the space
6984 * info has super bytes accounted for, otherwise we'll think
6985 * we have more space than we actually do.
6986 */
6987 exclude_super_stripes(root, cache);
6988
6989 /*
Josef Bacik817d52f2009-07-13 21:29:25 -04006990 * check for two cases, either we are full, and therefore
6991 * don't need to bother with the caching work since we won't
6992 * find any space, or we are empty, and we can just add all
6993 * the space in and be done with it. This saves us _alot_ of
6994 * time, particularly in the full case.
6995 */
6996 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
Yan Zheng11833d62009-09-11 16:11:19 -04006997 cache->last_byte_to_unpin = (u64)-1;
Josef Bacik817d52f2009-07-13 21:29:25 -04006998 cache->cached = BTRFS_CACHE_FINISHED;
Josef Bacik1b2da372009-09-11 16:11:20 -04006999 free_excluded_extents(root, cache);
Josef Bacik817d52f2009-07-13 21:29:25 -04007000 } else if (btrfs_block_group_used(&cache->item) == 0) {
Yan Zheng11833d62009-09-11 16:11:19 -04007001 cache->last_byte_to_unpin = (u64)-1;
Josef Bacik817d52f2009-07-13 21:29:25 -04007002 cache->cached = BTRFS_CACHE_FINISHED;
7003 add_new_free_space(cache, root->fs_info,
7004 found_key.objectid,
7005 found_key.objectid +
7006 found_key.offset);
Yan Zheng11833d62009-09-11 16:11:19 -04007007 free_excluded_extents(root, cache);
Josef Bacik817d52f2009-07-13 21:29:25 -04007008 }
Chris Mason96b51792007-10-15 16:15:19 -04007009
Chris Mason6324fbf2008-03-24 15:01:59 -04007010 ret = update_space_info(info, cache->flags, found_key.offset,
7011 btrfs_block_group_used(&cache->item),
7012 &space_info);
7013 BUG_ON(ret);
7014 cache->space_info = space_info;
Josef Bacik1b2da372009-09-11 16:11:20 -04007015 spin_lock(&cache->space_info->lock);
Yan, Zhengf0486c62010-05-16 10:46:25 -04007016 cache->space_info->bytes_readonly += cache->bytes_super;
Josef Bacik1b2da372009-09-11 16:11:20 -04007017 spin_unlock(&cache->space_info->lock);
7018
Yan, Zhengb742bb82010-05-16 10:46:24 -04007019 __link_block_group(space_info, cache);
Chris Mason6324fbf2008-03-24 15:01:59 -04007020
Josef Bacik0f9dd462008-09-23 13:14:11 -04007021 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7022 BUG_ON(ret);
Chris Mason75ccf472008-09-30 19:24:06 -04007023
7024 set_avail_alloc_bits(root->fs_info, cache->flags);
Yan Zheng2b820322008-11-17 21:11:30 -05007025 if (btrfs_chunk_readonly(root, cache->key.objectid))
Yan, Zhengf0486c62010-05-16 10:46:25 -04007026 set_block_group_ro(cache);
Chris Mason9078a3e2007-04-26 16:46:15 -04007027 }
Yan, Zhengb742bb82010-05-16 10:46:24 -04007028
7029 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
7030 if (!(get_alloc_profile(root, space_info->flags) &
7031 (BTRFS_BLOCK_GROUP_RAID10 |
7032 BTRFS_BLOCK_GROUP_RAID1 |
7033 BTRFS_BLOCK_GROUP_DUP)))
7034 continue;
7035 /*
7036 * avoid allocating from un-mirrored block group if there are
7037 * mirrored block groups.
7038 */
7039 list_for_each_entry(cache, &space_info->block_groups[3], list)
Yan, Zhengf0486c62010-05-16 10:46:25 -04007040 set_block_group_ro(cache);
Yan, Zhengb742bb82010-05-16 10:46:24 -04007041 list_for_each_entry(cache, &space_info->block_groups[4], list)
Yan, Zhengf0486c62010-05-16 10:46:25 -04007042 set_block_group_ro(cache);
Yan, Zhengb742bb82010-05-16 10:46:24 -04007043 }
Yan, Zhengf0486c62010-05-16 10:46:25 -04007044
7045 init_global_block_rsv(info);
Chris Mason0b86a832008-03-24 15:01:56 -04007046 ret = 0;
7047error:
Chris Mason9078a3e2007-04-26 16:46:15 -04007048 btrfs_free_path(path);
Chris Mason0b86a832008-03-24 15:01:56 -04007049 return ret;
Chris Mason9078a3e2007-04-26 16:46:15 -04007050}
Chris Mason6324fbf2008-03-24 15:01:59 -04007051
7052int btrfs_make_block_group(struct btrfs_trans_handle *trans,
7053 struct btrfs_root *root, u64 bytes_used,
Chris Masone17cade2008-04-15 15:41:47 -04007054 u64 type, u64 chunk_objectid, u64 chunk_offset,
Chris Mason6324fbf2008-03-24 15:01:59 -04007055 u64 size)
7056{
7057 int ret;
Chris Mason6324fbf2008-03-24 15:01:59 -04007058 struct btrfs_root *extent_root;
7059 struct btrfs_block_group_cache *cache;
Chris Mason6324fbf2008-03-24 15:01:59 -04007060
7061 extent_root = root->fs_info->extent_root;
Chris Mason6324fbf2008-03-24 15:01:59 -04007062
Chris Mason12fcfd22009-03-24 10:24:20 -04007063 root->fs_info->last_trans_log_full_commit = trans->transid;
Chris Masone02119d2008-09-05 16:13:11 -04007064
Chris Mason8f18cf12008-04-25 16:53:30 -04007065 cache = kzalloc(sizeof(*cache), GFP_NOFS);
Josef Bacik0f9dd462008-09-23 13:14:11 -04007066 if (!cache)
7067 return -ENOMEM;
Li Zefan34d52cb2011-03-29 13:46:06 +08007068 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7069 GFP_NOFS);
7070 if (!cache->free_space_ctl) {
7071 kfree(cache);
7072 return -ENOMEM;
7073 }
Josef Bacik0f9dd462008-09-23 13:14:11 -04007074
Chris Masone17cade2008-04-15 15:41:47 -04007075 cache->key.objectid = chunk_offset;
Chris Mason6324fbf2008-03-24 15:01:59 -04007076 cache->key.offset = size;
Yan Zhengd2fb3432008-12-11 16:30:39 -05007077 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
Josef Bacik96303082009-07-13 21:29:25 -04007078 cache->sectorsize = root->sectorsize;
Josef Bacik0af3d002010-06-21 14:48:16 -04007079 cache->fs_info = root->fs_info;
Josef Bacik96303082009-07-13 21:29:25 -04007080
Yan Zhengd2fb3432008-12-11 16:30:39 -05007081 atomic_set(&cache->count, 1);
Chris Masonc286ac42008-07-22 23:06:41 -04007082 spin_lock_init(&cache->lock);
Josef Bacik0f9dd462008-09-23 13:14:11 -04007083 INIT_LIST_HEAD(&cache->list);
Chris Masonfa9c0d72009-04-03 09:47:43 -04007084 INIT_LIST_HEAD(&cache->cluster_list);
Chris Mason0ef3e662008-05-24 14:04:53 -04007085
Li Zefan34d52cb2011-03-29 13:46:06 +08007086 btrfs_init_free_space_ctl(cache);
7087
Chris Mason6324fbf2008-03-24 15:01:59 -04007088 btrfs_set_block_group_used(&cache->item, bytes_used);
Chris Mason6324fbf2008-03-24 15:01:59 -04007089 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
7090 cache->flags = type;
7091 btrfs_set_block_group_flags(&cache->item, type);
7092
Yan Zheng11833d62009-09-11 16:11:19 -04007093 cache->last_byte_to_unpin = (u64)-1;
Josef Bacik817d52f2009-07-13 21:29:25 -04007094 cache->cached = BTRFS_CACHE_FINISHED;
Yan Zheng11833d62009-09-11 16:11:19 -04007095 exclude_super_stripes(root, cache);
Josef Bacik96303082009-07-13 21:29:25 -04007096
Josef Bacik817d52f2009-07-13 21:29:25 -04007097 add_new_free_space(cache, root->fs_info, chunk_offset,
7098 chunk_offset + size);
7099
Yan Zheng11833d62009-09-11 16:11:19 -04007100 free_excluded_extents(root, cache);
7101
Chris Mason6324fbf2008-03-24 15:01:59 -04007102 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
7103 &cache->space_info);
7104 BUG_ON(ret);
Josef Bacik1b2da372009-09-11 16:11:20 -04007105
7106 spin_lock(&cache->space_info->lock);
Yan, Zhengf0486c62010-05-16 10:46:25 -04007107 cache->space_info->bytes_readonly += cache->bytes_super;
Josef Bacik1b2da372009-09-11 16:11:20 -04007108 spin_unlock(&cache->space_info->lock);
7109
Yan, Zhengb742bb82010-05-16 10:46:24 -04007110 __link_block_group(cache->space_info, cache);
Chris Mason6324fbf2008-03-24 15:01:59 -04007111
Josef Bacik0f9dd462008-09-23 13:14:11 -04007112 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7113 BUG_ON(ret);
Chris Masonc286ac42008-07-22 23:06:41 -04007114
Chris Mason6324fbf2008-03-24 15:01:59 -04007115 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
7116 sizeof(cache->item));
7117 BUG_ON(ret);
7118
Chris Masond18a2c42008-04-04 15:40:00 -04007119 set_avail_alloc_bits(extent_root->fs_info, type);
Chris Mason925baed2008-06-25 16:01:30 -04007120
Chris Mason6324fbf2008-03-24 15:01:59 -04007121 return 0;
7122}
Zheng Yan1a40e232008-09-26 10:09:34 -04007123
7124int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
7125 struct btrfs_root *root, u64 group_start)
7126{
7127 struct btrfs_path *path;
7128 struct btrfs_block_group_cache *block_group;
Chris Mason44fb5512009-06-04 15:34:51 -04007129 struct btrfs_free_cluster *cluster;
Josef Bacik0af3d002010-06-21 14:48:16 -04007130 struct btrfs_root *tree_root = root->fs_info->tree_root;
Zheng Yan1a40e232008-09-26 10:09:34 -04007131 struct btrfs_key key;
Josef Bacik0af3d002010-06-21 14:48:16 -04007132 struct inode *inode;
Zheng Yan1a40e232008-09-26 10:09:34 -04007133 int ret;
Josef Bacik89a55892010-10-14 14:52:27 -04007134 int factor;
Zheng Yan1a40e232008-09-26 10:09:34 -04007135
Zheng Yan1a40e232008-09-26 10:09:34 -04007136 root = root->fs_info->extent_root;
7137
7138 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
7139 BUG_ON(!block_group);
Yan Zhengc146afa2008-11-12 14:34:12 -05007140 BUG_ON(!block_group->ro);
Zheng Yan1a40e232008-09-26 10:09:34 -04007141
liubo9f7c43c2011-03-07 02:13:33 +00007142 /*
7143 * Free the reserved super bytes from this block group before
7144 * remove it.
7145 */
7146 free_excluded_extents(root, block_group);
7147
Zheng Yan1a40e232008-09-26 10:09:34 -04007148 memcpy(&key, &block_group->key, sizeof(key));
Josef Bacik89a55892010-10-14 14:52:27 -04007149 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
7150 BTRFS_BLOCK_GROUP_RAID1 |
7151 BTRFS_BLOCK_GROUP_RAID10))
7152 factor = 2;
7153 else
7154 factor = 1;
Zheng Yan1a40e232008-09-26 10:09:34 -04007155
Chris Mason44fb5512009-06-04 15:34:51 -04007156 /* make sure this block group isn't part of an allocation cluster */
7157 cluster = &root->fs_info->data_alloc_cluster;
7158 spin_lock(&cluster->refill_lock);
7159 btrfs_return_cluster_to_free_space(block_group, cluster);
7160 spin_unlock(&cluster->refill_lock);
7161
7162 /*
7163 * make sure this block group isn't part of a metadata
7164 * allocation cluster
7165 */
7166 cluster = &root->fs_info->meta_alloc_cluster;
7167 spin_lock(&cluster->refill_lock);
7168 btrfs_return_cluster_to_free_space(block_group, cluster);
7169 spin_unlock(&cluster->refill_lock);
7170
Zheng Yan1a40e232008-09-26 10:09:34 -04007171 path = btrfs_alloc_path();
7172 BUG_ON(!path);
7173
Josef Bacik0af3d002010-06-21 14:48:16 -04007174 inode = lookup_free_space_inode(root, block_group, path);
7175 if (!IS_ERR(inode)) {
7176 btrfs_orphan_add(trans, inode);
7177 clear_nlink(inode);
7178 /* One for the block groups ref */
7179 spin_lock(&block_group->lock);
7180 if (block_group->iref) {
7181 block_group->iref = 0;
7182 block_group->inode = NULL;
7183 spin_unlock(&block_group->lock);
7184 iput(inode);
7185 } else {
7186 spin_unlock(&block_group->lock);
7187 }
7188 /* One for our lookup ref */
7189 iput(inode);
7190 }
7191
7192 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
7193 key.offset = block_group->key.objectid;
7194 key.type = 0;
7195
7196 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
7197 if (ret < 0)
7198 goto out;
7199 if (ret > 0)
David Sterbab3b4aa72011-04-21 01:20:15 +02007200 btrfs_release_path(path);
Josef Bacik0af3d002010-06-21 14:48:16 -04007201 if (ret == 0) {
7202 ret = btrfs_del_item(trans, tree_root, path);
7203 if (ret)
7204 goto out;
David Sterbab3b4aa72011-04-21 01:20:15 +02007205 btrfs_release_path(path);
Josef Bacik0af3d002010-06-21 14:48:16 -04007206 }
7207
Yan Zheng3dfdb932009-01-21 10:49:16 -05007208 spin_lock(&root->fs_info->block_group_cache_lock);
Zheng Yan1a40e232008-09-26 10:09:34 -04007209 rb_erase(&block_group->cache_node,
7210 &root->fs_info->block_group_cache_tree);
Yan Zheng3dfdb932009-01-21 10:49:16 -05007211 spin_unlock(&root->fs_info->block_group_cache_lock);
Josef Bacik817d52f2009-07-13 21:29:25 -04007212
Josef Bacik80eb2342008-10-29 14:49:05 -04007213 down_write(&block_group->space_info->groups_sem);
Chris Mason44fb5512009-06-04 15:34:51 -04007214 /*
7215 * we must use list_del_init so people can check to see if they
7216 * are still on the list after taking the semaphore
7217 */
7218 list_del_init(&block_group->list);
Josef Bacik80eb2342008-10-29 14:49:05 -04007219 up_write(&block_group->space_info->groups_sem);
Zheng Yan1a40e232008-09-26 10:09:34 -04007220
Josef Bacik817d52f2009-07-13 21:29:25 -04007221 if (block_group->cached == BTRFS_CACHE_STARTED)
Yan Zheng11833d62009-09-11 16:11:19 -04007222 wait_block_group_cache_done(block_group);
Josef Bacik817d52f2009-07-13 21:29:25 -04007223
7224 btrfs_remove_free_space_cache(block_group);
7225
Yan Zhengc146afa2008-11-12 14:34:12 -05007226 spin_lock(&block_group->space_info->lock);
7227 block_group->space_info->total_bytes -= block_group->key.offset;
7228 block_group->space_info->bytes_readonly -= block_group->key.offset;
Josef Bacik89a55892010-10-14 14:52:27 -04007229 block_group->space_info->disk_total -= block_group->key.offset * factor;
Yan Zhengc146afa2008-11-12 14:34:12 -05007230 spin_unlock(&block_group->space_info->lock);
Chris Mason283bb192009-07-24 16:30:55 -04007231
Josef Bacik0af3d002010-06-21 14:48:16 -04007232 memcpy(&key, &block_group->key, sizeof(key));
7233
Chris Mason283bb192009-07-24 16:30:55 -04007234 btrfs_clear_space_info_full(root->fs_info);
Yan Zhengc146afa2008-11-12 14:34:12 -05007235
Chris Masonfa9c0d72009-04-03 09:47:43 -04007236 btrfs_put_block_group(block_group);
7237 btrfs_put_block_group(block_group);
Zheng Yan1a40e232008-09-26 10:09:34 -04007238
7239 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
7240 if (ret > 0)
7241 ret = -EIO;
7242 if (ret < 0)
7243 goto out;
7244
7245 ret = btrfs_del_item(trans, root, path);
7246out:
7247 btrfs_free_path(path);
7248 return ret;
7249}
liuboacce9522011-01-06 19:30:25 +08007250
liuboc59021f2011-03-07 02:13:14 +00007251int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
7252{
7253 struct btrfs_space_info *space_info;
liubo1aba86d2011-04-08 08:44:37 +00007254 struct btrfs_super_block *disk_super;
7255 u64 features;
7256 u64 flags;
7257 int mixed = 0;
liuboc59021f2011-03-07 02:13:14 +00007258 int ret;
7259
liubo1aba86d2011-04-08 08:44:37 +00007260 disk_super = &fs_info->super_copy;
7261 if (!btrfs_super_root(disk_super))
7262 return 1;
liuboc59021f2011-03-07 02:13:14 +00007263
liubo1aba86d2011-04-08 08:44:37 +00007264 features = btrfs_super_incompat_flags(disk_super);
7265 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
7266 mixed = 1;
liuboc59021f2011-03-07 02:13:14 +00007267
liubo1aba86d2011-04-08 08:44:37 +00007268 flags = BTRFS_BLOCK_GROUP_SYSTEM;
7269 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
liuboc59021f2011-03-07 02:13:14 +00007270 if (ret)
liubo1aba86d2011-04-08 08:44:37 +00007271 goto out;
liuboc59021f2011-03-07 02:13:14 +00007272
liubo1aba86d2011-04-08 08:44:37 +00007273 if (mixed) {
7274 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
7275 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7276 } else {
7277 flags = BTRFS_BLOCK_GROUP_METADATA;
7278 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7279 if (ret)
7280 goto out;
7281
7282 flags = BTRFS_BLOCK_GROUP_DATA;
7283 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7284 }
7285out:
liuboc59021f2011-03-07 02:13:14 +00007286 return ret;
7287}
7288
liuboacce9522011-01-06 19:30:25 +08007289int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
7290{
7291 return unpin_extent_range(root, start, end);
7292}
7293
7294int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
Li Dongyang5378e602011-03-24 10:24:27 +00007295 u64 num_bytes, u64 *actual_bytes)
liuboacce9522011-01-06 19:30:25 +08007296{
Li Dongyang5378e602011-03-24 10:24:27 +00007297 return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
liuboacce9522011-01-06 19:30:25 +08007298}
Li Dongyangf7039b12011-03-24 10:24:28 +00007299
7300int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
7301{
7302 struct btrfs_fs_info *fs_info = root->fs_info;
7303 struct btrfs_block_group_cache *cache = NULL;
7304 u64 group_trimmed;
7305 u64 start;
7306 u64 end;
7307 u64 trimmed = 0;
7308 int ret = 0;
7309
7310 cache = btrfs_lookup_block_group(fs_info, range->start);
7311
7312 while (cache) {
7313 if (cache->key.objectid >= (range->start + range->len)) {
7314 btrfs_put_block_group(cache);
7315 break;
7316 }
7317
7318 start = max(range->start, cache->key.objectid);
7319 end = min(range->start + range->len,
7320 cache->key.objectid + cache->key.offset);
7321
7322 if (end - start >= range->minlen) {
7323 if (!block_group_cache_done(cache)) {
7324 ret = cache_block_group(cache, NULL, root, 0);
7325 if (!ret)
7326 wait_block_group_cache_done(cache);
7327 }
7328 ret = btrfs_trim_block_group(cache,
7329 &group_trimmed,
7330 start,
7331 end,
7332 range->minlen);
7333
7334 trimmed += group_trimmed;
7335 if (ret) {
7336 btrfs_put_block_group(cache);
7337 break;
7338 }
7339 }
7340
7341 cache = next_block_group(fs_info->tree_root, cache);
7342 }
7343
7344 range->len = trimmed;
7345 return ret;
7346}