blob: 07f167a1d27141873a962566d66c796c45d42be5 [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>
David Sterbadff51cd2011-06-14 12:52:17 +020026#include <linux/ratelimit.h>
Chris Mason4b4e25f2008-11-20 10:22:27 -050027#include "compat.h"
Chris Mason74493f72007-12-11 09:25:06 -050028#include "hash.h"
Chris Masonfec577f2007-02-26 10:40:21 -050029#include "ctree.h"
30#include "disk-io.h"
31#include "print-tree.h"
Chris Masone089f052007-03-16 16:20:31 -040032#include "transaction.h"
Chris Mason0b86a832008-03-24 15:01:56 -040033#include "volumes.h"
David Woodhouse53b381b2013-01-29 18:40:14 -050034#include "raid56.h"
Chris Mason925baed2008-06-25 16:01:30 -040035#include "locking.h"
Chris Masonfa9c0d792009-04-03 09:47:43 -040036#include "free-space-cache.h"
Miao Xie3fed40c2012-09-13 04:51:36 -060037#include "math.h"
Chris Masonfec577f2007-02-26 10:40:21 -050038
Arne Jansen709c0482011-09-12 12:22:57 +020039#undef SCRAMBLE_DELAYED_REFS
40
Miao Xie9e622d62012-01-26 15:01:12 -050041/*
42 * control flags for do_chunk_alloc's force field
Chris Mason0e4f8f82011-04-15 16:05:44 -040043 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
44 * if we really need one.
45 *
Chris Mason0e4f8f82011-04-15 16:05:44 -040046 * CHUNK_ALLOC_LIMITED means to only try and allocate one
47 * if we have very few chunks already allocated. This is
48 * used as part of the clustering code to help make sure
49 * we have a good pool of storage to cluster in, without
50 * filling the FS with empty chunks
51 *
Miao Xie9e622d62012-01-26 15:01:12 -050052 * CHUNK_ALLOC_FORCE means it must try to allocate one
53 *
Chris Mason0e4f8f82011-04-15 16:05:44 -040054 */
55enum {
56 CHUNK_ALLOC_NO_FORCE = 0,
Miao Xie9e622d62012-01-26 15:01:12 -050057 CHUNK_ALLOC_LIMITED = 1,
58 CHUNK_ALLOC_FORCE = 2,
Chris Mason0e4f8f82011-04-15 16:05:44 -040059};
60
Josef Bacikfb25e912011-07-26 17:00:46 -040061/*
62 * Control how reservations are dealt with.
63 *
64 * RESERVE_FREE - freeing a reservation.
65 * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
66 * ENOSPC accounting
67 * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
68 * bytes_may_use as the ENOSPC accounting is done elsewhere
69 */
70enum {
71 RESERVE_FREE = 0,
72 RESERVE_ALLOC = 1,
73 RESERVE_ALLOC_NO_ACCOUNT = 2,
74};
75
Liu Boc53d6132012-12-27 09:01:19 +000076static int update_block_group(struct btrfs_root *root,
Yan, Zhengf0486c62010-05-16 10:46:25 -040077 u64 bytenr, u64 num_bytes, int alloc);
Yan Zheng5d4f98a2009-06-10 10:45:14 -040078static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
79 struct btrfs_root *root,
80 u64 bytenr, u64 num_bytes, u64 parent,
81 u64 root_objectid, u64 owner_objectid,
82 u64 owner_offset, int refs_to_drop,
83 struct btrfs_delayed_extent_op *extra_op);
84static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
85 struct extent_buffer *leaf,
86 struct btrfs_extent_item *ei);
87static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
88 struct btrfs_root *root,
89 u64 parent, u64 root_objectid,
90 u64 flags, u64 owner, u64 offset,
91 struct btrfs_key *ins, int ref_mod);
92static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
93 struct btrfs_root *root,
94 u64 parent, u64 root_objectid,
95 u64 flags, struct btrfs_disk_key *key,
96 int level, struct btrfs_key *ins);
Josef Bacik6a632092009-02-20 11:00:09 -050097static int do_chunk_alloc(struct btrfs_trans_handle *trans,
Josef Bacik698d0082012-09-12 14:08:47 -040098 struct btrfs_root *extent_root, u64 flags,
99 int force);
Yan Zheng11833d62009-09-11 16:11:19 -0400100static int find_next_key(struct btrfs_path *path, int level,
101 struct btrfs_key *key);
Josef Bacik9ed74f22009-09-11 16:12:44 -0400102static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
103 int dump_block_groups);
Josef Bacikfb25e912011-07-26 17:00:46 -0400104static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
105 u64 num_bytes, int reserve);
Josef Bacik5d803662013-02-07 16:06:02 -0500106static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
107 u64 num_bytes);
Eric Sandeen48a3b632013-04-25 20:41:01 +0000108int btrfs_pin_extent(struct btrfs_root *root,
109 u64 bytenr, u64 num_bytes, int reserved);
Josef Bacik6a632092009-02-20 11:00:09 -0500110
Josef Bacik817d52f2009-07-13 21:29:25 -0400111static noinline int
112block_group_cache_done(struct btrfs_block_group_cache *cache)
113{
114 smp_mb();
115 return cache->cached == BTRFS_CACHE_FINISHED;
116}
117
Josef Bacik0f9dd462008-09-23 13:14:11 -0400118static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
119{
120 return (cache->flags & bits) == bits;
121}
122
David Sterba62a45b62011-04-20 15:52:26 +0200123static void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
Josef Bacik11dfe352009-11-13 20:12:59 +0000124{
125 atomic_inc(&cache->count);
126}
127
128void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
129{
Yan, Zhengf0486c62010-05-16 10:46:25 -0400130 if (atomic_dec_and_test(&cache->count)) {
131 WARN_ON(cache->pinned > 0);
132 WARN_ON(cache->reserved > 0);
Li Zefan34d52cb2011-03-29 13:46:06 +0800133 kfree(cache->free_space_ctl);
Josef Bacik11dfe352009-11-13 20:12:59 +0000134 kfree(cache);
Yan, Zhengf0486c62010-05-16 10:46:25 -0400135 }
Josef Bacik11dfe352009-11-13 20:12:59 +0000136}
137
Josef Bacik0f9dd462008-09-23 13:14:11 -0400138/*
139 * this adds the block group to the fs_info rb tree for the block group
140 * cache
141 */
Christoph Hellwigb2950862008-12-02 09:54:17 -0500142static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
Josef Bacik0f9dd462008-09-23 13:14:11 -0400143 struct btrfs_block_group_cache *block_group)
144{
145 struct rb_node **p;
146 struct rb_node *parent = NULL;
147 struct btrfs_block_group_cache *cache;
148
149 spin_lock(&info->block_group_cache_lock);
150 p = &info->block_group_cache_tree.rb_node;
151
152 while (*p) {
153 parent = *p;
154 cache = rb_entry(parent, struct btrfs_block_group_cache,
155 cache_node);
156 if (block_group->key.objectid < cache->key.objectid) {
157 p = &(*p)->rb_left;
158 } else if (block_group->key.objectid > cache->key.objectid) {
159 p = &(*p)->rb_right;
160 } else {
161 spin_unlock(&info->block_group_cache_lock);
162 return -EEXIST;
163 }
164 }
165
166 rb_link_node(&block_group->cache_node, parent, p);
167 rb_insert_color(&block_group->cache_node,
168 &info->block_group_cache_tree);
Liu Boa1897fd2012-12-27 09:01:23 +0000169
170 if (info->first_logical_byte > block_group->key.objectid)
171 info->first_logical_byte = block_group->key.objectid;
172
Josef Bacik0f9dd462008-09-23 13:14:11 -0400173 spin_unlock(&info->block_group_cache_lock);
174
175 return 0;
176}
177
178/*
179 * This will return the block group at or after bytenr if contains is 0, else
180 * it will return the block group that contains the bytenr
181 */
182static struct btrfs_block_group_cache *
183block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
184 int contains)
185{
186 struct btrfs_block_group_cache *cache, *ret = NULL;
187 struct rb_node *n;
188 u64 end, start;
189
190 spin_lock(&info->block_group_cache_lock);
191 n = info->block_group_cache_tree.rb_node;
192
193 while (n) {
194 cache = rb_entry(n, struct btrfs_block_group_cache,
195 cache_node);
196 end = cache->key.objectid + cache->key.offset - 1;
197 start = cache->key.objectid;
198
199 if (bytenr < start) {
200 if (!contains && (!ret || start < ret->key.objectid))
201 ret = cache;
202 n = n->rb_left;
203 } else if (bytenr > start) {
204 if (contains && bytenr <= end) {
205 ret = cache;
206 break;
207 }
208 n = n->rb_right;
209 } else {
210 ret = cache;
211 break;
212 }
213 }
Liu Boa1897fd2012-12-27 09:01:23 +0000214 if (ret) {
Josef Bacik11dfe352009-11-13 20:12:59 +0000215 btrfs_get_block_group(ret);
Liu Boa1897fd2012-12-27 09:01:23 +0000216 if (bytenr == 0 && info->first_logical_byte > ret->key.objectid)
217 info->first_logical_byte = ret->key.objectid;
218 }
Josef Bacik0f9dd462008-09-23 13:14:11 -0400219 spin_unlock(&info->block_group_cache_lock);
220
221 return ret;
222}
223
Yan Zheng11833d62009-09-11 16:11:19 -0400224static int add_excluded_extent(struct btrfs_root *root,
225 u64 start, u64 num_bytes)
Josef Bacik817d52f2009-07-13 21:29:25 -0400226{
Yan Zheng11833d62009-09-11 16:11:19 -0400227 u64 end = start + num_bytes - 1;
228 set_extent_bits(&root->fs_info->freed_extents[0],
229 start, end, EXTENT_UPTODATE, GFP_NOFS);
230 set_extent_bits(&root->fs_info->freed_extents[1],
231 start, end, EXTENT_UPTODATE, GFP_NOFS);
232 return 0;
Josef Bacik817d52f2009-07-13 21:29:25 -0400233}
234
Yan Zheng11833d62009-09-11 16:11:19 -0400235static void free_excluded_extents(struct btrfs_root *root,
236 struct btrfs_block_group_cache *cache)
Josef Bacik817d52f2009-07-13 21:29:25 -0400237{
Yan Zheng11833d62009-09-11 16:11:19 -0400238 u64 start, end;
239
240 start = cache->key.objectid;
241 end = start + cache->key.offset - 1;
242
243 clear_extent_bits(&root->fs_info->freed_extents[0],
244 start, end, EXTENT_UPTODATE, GFP_NOFS);
245 clear_extent_bits(&root->fs_info->freed_extents[1],
246 start, end, EXTENT_UPTODATE, GFP_NOFS);
247}
248
249static int exclude_super_stripes(struct btrfs_root *root,
250 struct btrfs_block_group_cache *cache)
251{
Josef Bacik817d52f2009-07-13 21:29:25 -0400252 u64 bytenr;
253 u64 *logical;
254 int stripe_len;
255 int i, nr, ret;
256
Yan, Zheng06b23312009-11-26 09:31:11 +0000257 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
258 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
259 cache->bytes_super += stripe_len;
260 ret = add_excluded_extent(root, cache->key.objectid,
261 stripe_len);
Josef Bacik835d9742013-03-19 12:13:25 -0400262 if (ret)
263 return ret;
Yan, Zheng06b23312009-11-26 09:31:11 +0000264 }
265
Josef Bacik817d52f2009-07-13 21:29:25 -0400266 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
267 bytenr = btrfs_sb_offset(i);
268 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
269 cache->key.objectid, bytenr,
270 0, &logical, &nr, &stripe_len);
Josef Bacik835d9742013-03-19 12:13:25 -0400271 if (ret)
272 return ret;
Yan Zheng11833d62009-09-11 16:11:19 -0400273
Josef Bacik817d52f2009-07-13 21:29:25 -0400274 while (nr--) {
Josef Bacik51bf5f02013-04-23 12:55:21 -0400275 u64 start, len;
276
277 if (logical[nr] > cache->key.objectid +
278 cache->key.offset)
279 continue;
280
281 if (logical[nr] + stripe_len <= cache->key.objectid)
282 continue;
283
284 start = logical[nr];
285 if (start < cache->key.objectid) {
286 start = cache->key.objectid;
287 len = (logical[nr] + stripe_len) - start;
288 } else {
289 len = min_t(u64, stripe_len,
290 cache->key.objectid +
291 cache->key.offset - start);
292 }
293
294 cache->bytes_super += len;
295 ret = add_excluded_extent(root, start, len);
Josef Bacik835d9742013-03-19 12:13:25 -0400296 if (ret) {
297 kfree(logical);
298 return ret;
299 }
Josef Bacik817d52f2009-07-13 21:29:25 -0400300 }
Yan Zheng11833d62009-09-11 16:11:19 -0400301
Josef Bacik817d52f2009-07-13 21:29:25 -0400302 kfree(logical);
303 }
Josef Bacik817d52f2009-07-13 21:29:25 -0400304 return 0;
305}
306
Yan Zheng11833d62009-09-11 16:11:19 -0400307static struct btrfs_caching_control *
308get_caching_control(struct btrfs_block_group_cache *cache)
309{
310 struct btrfs_caching_control *ctl;
311
312 spin_lock(&cache->lock);
313 if (cache->cached != BTRFS_CACHE_STARTED) {
314 spin_unlock(&cache->lock);
315 return NULL;
316 }
317
Josef Bacikdde5abe2010-09-16 16:17:03 -0400318 /* We're loading it the fast way, so we don't have a caching_ctl. */
319 if (!cache->caching_ctl) {
320 spin_unlock(&cache->lock);
321 return NULL;
322 }
323
Yan Zheng11833d62009-09-11 16:11:19 -0400324 ctl = cache->caching_ctl;
325 atomic_inc(&ctl->count);
326 spin_unlock(&cache->lock);
327 return ctl;
328}
329
330static void put_caching_control(struct btrfs_caching_control *ctl)
331{
332 if (atomic_dec_and_test(&ctl->count))
333 kfree(ctl);
334}
335
Josef Bacik0f9dd462008-09-23 13:14:11 -0400336/*
337 * this is only called by cache_block_group, since we could have freed extents
338 * we need to check the pinned_extents for any extents that can't be used yet
339 * since their free space will be released as soon as the transaction commits.
340 */
Josef Bacik817d52f2009-07-13 21:29:25 -0400341static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
Josef Bacik0f9dd462008-09-23 13:14:11 -0400342 struct btrfs_fs_info *info, u64 start, u64 end)
343{
Josef Bacik817d52f2009-07-13 21:29:25 -0400344 u64 extent_start, extent_end, size, total_added = 0;
Josef Bacik0f9dd462008-09-23 13:14:11 -0400345 int ret;
346
347 while (start < end) {
Yan Zheng11833d62009-09-11 16:11:19 -0400348 ret = find_first_extent_bit(info->pinned_extents, start,
Josef Bacik0f9dd462008-09-23 13:14:11 -0400349 &extent_start, &extent_end,
Josef Bacike6138872012-09-27 17:07:30 -0400350 EXTENT_DIRTY | EXTENT_UPTODATE,
351 NULL);
Josef Bacik0f9dd462008-09-23 13:14:11 -0400352 if (ret)
353 break;
354
Yan, Zheng06b23312009-11-26 09:31:11 +0000355 if (extent_start <= start) {
Josef Bacik0f9dd462008-09-23 13:14:11 -0400356 start = extent_end + 1;
357 } else if (extent_start > start && extent_start < end) {
358 size = extent_start - start;
Josef Bacik817d52f2009-07-13 21:29:25 -0400359 total_added += size;
Josef Bacikea6a4782008-11-20 12:16:16 -0500360 ret = btrfs_add_free_space(block_group, start,
361 size);
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100362 BUG_ON(ret); /* -ENOMEM or logic error */
Josef Bacik0f9dd462008-09-23 13:14:11 -0400363 start = extent_end + 1;
364 } else {
365 break;
366 }
367 }
368
369 if (start < end) {
370 size = end - start;
Josef Bacik817d52f2009-07-13 21:29:25 -0400371 total_added += size;
Josef Bacikea6a4782008-11-20 12:16:16 -0500372 ret = btrfs_add_free_space(block_group, start, size);
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100373 BUG_ON(ret); /* -ENOMEM or logic error */
Josef Bacik0f9dd462008-09-23 13:14:11 -0400374 }
375
Josef Bacik817d52f2009-07-13 21:29:25 -0400376 return total_added;
Josef Bacik0f9dd462008-09-23 13:14:11 -0400377}
378
Josef Bacikbab39bf2011-06-30 14:42:28 -0400379static noinline void caching_thread(struct btrfs_work *work)
Chris Masone37c9e62007-05-09 20:13:14 -0400380{
Josef Bacikbab39bf2011-06-30 14:42:28 -0400381 struct btrfs_block_group_cache *block_group;
382 struct btrfs_fs_info *fs_info;
383 struct btrfs_caching_control *caching_ctl;
384 struct btrfs_root *extent_root;
Chris Masone37c9e62007-05-09 20:13:14 -0400385 struct btrfs_path *path;
Chris Mason5f39d392007-10-15 16:14:19 -0400386 struct extent_buffer *leaf;
Yan Zheng11833d62009-09-11 16:11:19 -0400387 struct btrfs_key key;
Josef Bacik817d52f2009-07-13 21:29:25 -0400388 u64 total_found = 0;
Yan Zheng11833d62009-09-11 16:11:19 -0400389 u64 last = 0;
390 u32 nritems;
391 int ret = 0;
Chris Masonf510cfe2007-10-15 16:14:48 -0400392
Josef Bacikbab39bf2011-06-30 14:42:28 -0400393 caching_ctl = container_of(work, struct btrfs_caching_control, work);
394 block_group = caching_ctl->block_group;
395 fs_info = block_group->fs_info;
396 extent_root = fs_info->extent_root;
397
Chris Masone37c9e62007-05-09 20:13:14 -0400398 path = btrfs_alloc_path();
399 if (!path)
Josef Bacikbab39bf2011-06-30 14:42:28 -0400400 goto out;
Yan7d7d6062007-09-14 16:15:28 -0400401
Josef Bacik817d52f2009-07-13 21:29:25 -0400402 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
Yan Zheng11833d62009-09-11 16:11:19 -0400403
Chris Mason5cd57b22008-06-25 16:01:30 -0400404 /*
Josef Bacik817d52f2009-07-13 21:29:25 -0400405 * We don't want to deadlock with somebody trying to allocate a new
406 * extent for the extent root while also trying to search the extent
407 * root to add free space. So we skip locking and search the commit
408 * root, since its read-only
Chris Mason5cd57b22008-06-25 16:01:30 -0400409 */
410 path->skip_locking = 1;
Josef Bacik817d52f2009-07-13 21:29:25 -0400411 path->search_commit_root = 1;
Josef Bacik026fd312011-05-13 10:32:11 -0400412 path->reada = 1;
Josef Bacik817d52f2009-07-13 21:29:25 -0400413
Yan Zhenge4404d62008-12-12 10:03:26 -0500414 key.objectid = last;
Chris Masone37c9e62007-05-09 20:13:14 -0400415 key.offset = 0;
Yan Zheng11833d62009-09-11 16:11:19 -0400416 key.type = BTRFS_EXTENT_ITEM_KEY;
Chris Mason013f1b12009-07-31 14:57:55 -0400417again:
Yan Zheng11833d62009-09-11 16:11:19 -0400418 mutex_lock(&caching_ctl->mutex);
Chris Mason013f1b12009-07-31 14:57:55 -0400419 /* need to make sure the commit_root doesn't disappear */
420 down_read(&fs_info->extent_commit_sem);
421
Yan Zheng11833d62009-09-11 16:11:19 -0400422 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
Chris Masone37c9e62007-05-09 20:13:14 -0400423 if (ret < 0)
Josef Bacikef8bbdf2008-09-23 13:14:11 -0400424 goto err;
Yan Zhenga512bbf2008-12-08 16:46:26 -0500425
Yan Zheng11833d62009-09-11 16:11:19 -0400426 leaf = path->nodes[0];
427 nritems = btrfs_header_nritems(leaf);
428
Chris Masond3977122009-01-05 21:25:51 -0500429 while (1) {
David Sterba7841cb22011-05-31 18:07:27 +0200430 if (btrfs_fs_closing(fs_info) > 1) {
Yan Zhengf25784b2009-07-28 08:41:57 -0400431 last = (u64)-1;
Josef Bacik817d52f2009-07-13 21:29:25 -0400432 break;
Yan Zhengf25784b2009-07-28 08:41:57 -0400433 }
Josef Bacik817d52f2009-07-13 21:29:25 -0400434
Yan Zheng11833d62009-09-11 16:11:19 -0400435 if (path->slots[0] < nritems) {
436 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
437 } else {
438 ret = find_next_key(path, 0, &key);
439 if (ret)
Chris Masone37c9e62007-05-09 20:13:14 -0400440 break;
Josef Bacik817d52f2009-07-13 21:29:25 -0400441
Josef Bacik0a3896d2013-04-19 14:37:26 -0400442 if (need_resched()) {
Josef Bacik589d8ad2011-05-11 17:30:53 -0400443 caching_ctl->progress = last;
Chris Masonff5714c2011-05-28 07:00:39 -0400444 btrfs_release_path(path);
Josef Bacik589d8ad2011-05-11 17:30:53 -0400445 up_read(&fs_info->extent_commit_sem);
446 mutex_unlock(&caching_ctl->mutex);
Yan Zheng11833d62009-09-11 16:11:19 -0400447 cond_resched();
Josef Bacik589d8ad2011-05-11 17:30:53 -0400448 goto again;
449 }
Josef Bacik0a3896d2013-04-19 14:37:26 -0400450
451 ret = btrfs_next_leaf(extent_root, path);
452 if (ret < 0)
453 goto err;
454 if (ret)
455 break;
Josef Bacik589d8ad2011-05-11 17:30:53 -0400456 leaf = path->nodes[0];
457 nritems = btrfs_header_nritems(leaf);
458 continue;
Yan Zheng11833d62009-09-11 16:11:19 -0400459 }
Josef Bacik817d52f2009-07-13 21:29:25 -0400460
Yan Zheng11833d62009-09-11 16:11:19 -0400461 if (key.objectid < block_group->key.objectid) {
462 path->slots[0]++;
Josef Bacik817d52f2009-07-13 21:29:25 -0400463 continue;
Chris Masone37c9e62007-05-09 20:13:14 -0400464 }
Josef Bacik0f9dd462008-09-23 13:14:11 -0400465
Chris Masone37c9e62007-05-09 20:13:14 -0400466 if (key.objectid >= block_group->key.objectid +
Josef Bacik0f9dd462008-09-23 13:14:11 -0400467 block_group->key.offset)
Yan7d7d6062007-09-14 16:15:28 -0400468 break;
Yan7d7d6062007-09-14 16:15:28 -0400469
Josef Bacik3173a182013-03-07 14:22:04 -0500470 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
471 key.type == BTRFS_METADATA_ITEM_KEY) {
Josef Bacik817d52f2009-07-13 21:29:25 -0400472 total_found += add_new_free_space(block_group,
473 fs_info, last,
474 key.objectid);
Josef Bacik3173a182013-03-07 14:22:04 -0500475 if (key.type == BTRFS_METADATA_ITEM_KEY)
476 last = key.objectid +
477 fs_info->tree_root->leafsize;
478 else
479 last = key.objectid + key.offset;
Josef Bacik817d52f2009-07-13 21:29:25 -0400480
Yan Zheng11833d62009-09-11 16:11:19 -0400481 if (total_found > (1024 * 1024 * 2)) {
482 total_found = 0;
483 wake_up(&caching_ctl->wait);
484 }
Josef Bacik817d52f2009-07-13 21:29:25 -0400485 }
Chris Masone37c9e62007-05-09 20:13:14 -0400486 path->slots[0]++;
487 }
Josef Bacikef8bbdf2008-09-23 13:14:11 -0400488 ret = 0;
Josef Bacik817d52f2009-07-13 21:29:25 -0400489
490 total_found += add_new_free_space(block_group, fs_info, last,
491 block_group->key.objectid +
492 block_group->key.offset);
Yan Zheng11833d62009-09-11 16:11:19 -0400493 caching_ctl->progress = (u64)-1;
Josef Bacik817d52f2009-07-13 21:29:25 -0400494
495 spin_lock(&block_group->lock);
Yan Zheng11833d62009-09-11 16:11:19 -0400496 block_group->caching_ctl = NULL;
Josef Bacik817d52f2009-07-13 21:29:25 -0400497 block_group->cached = BTRFS_CACHE_FINISHED;
498 spin_unlock(&block_group->lock);
499
Chris Mason54aa1f42007-06-22 14:16:25 -0400500err:
Chris Masone37c9e62007-05-09 20:13:14 -0400501 btrfs_free_path(path);
Yan Zheng276e6802009-07-30 09:40:40 -0400502 up_read(&fs_info->extent_commit_sem);
Josef Bacik817d52f2009-07-13 21:29:25 -0400503
Yan Zheng11833d62009-09-11 16:11:19 -0400504 free_excluded_extents(extent_root, block_group);
505
506 mutex_unlock(&caching_ctl->mutex);
Josef Bacikbab39bf2011-06-30 14:42:28 -0400507out:
Yan Zheng11833d62009-09-11 16:11:19 -0400508 wake_up(&caching_ctl->wait);
509
510 put_caching_control(caching_ctl);
Josef Bacik11dfe352009-11-13 20:12:59 +0000511 btrfs_put_block_group(block_group);
Josef Bacik817d52f2009-07-13 21:29:25 -0400512}
513
Josef Bacik9d66e232010-08-25 16:54:15 -0400514static int cache_block_group(struct btrfs_block_group_cache *cache,
Josef Bacik9d66e232010-08-25 16:54:15 -0400515 int load_cache_only)
Josef Bacik817d52f2009-07-13 21:29:25 -0400516{
Josef Bacik291c7d22011-11-14 13:52:14 -0500517 DEFINE_WAIT(wait);
Yan Zheng11833d62009-09-11 16:11:19 -0400518 struct btrfs_fs_info *fs_info = cache->fs_info;
519 struct btrfs_caching_control *caching_ctl;
Josef Bacik817d52f2009-07-13 21:29:25 -0400520 int ret = 0;
521
Josef Bacik291c7d22011-11-14 13:52:14 -0500522 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100523 if (!caching_ctl)
524 return -ENOMEM;
Josef Bacik291c7d22011-11-14 13:52:14 -0500525
526 INIT_LIST_HEAD(&caching_ctl->list);
527 mutex_init(&caching_ctl->mutex);
528 init_waitqueue_head(&caching_ctl->wait);
529 caching_ctl->block_group = cache;
530 caching_ctl->progress = cache->key.objectid;
531 atomic_set(&caching_ctl->count, 1);
532 caching_ctl->work.func = caching_thread;
533
534 spin_lock(&cache->lock);
535 /*
536 * This should be a rare occasion, but this could happen I think in the
537 * case where one thread starts to load the space cache info, and then
538 * some other thread starts a transaction commit which tries to do an
539 * allocation while the other thread is still loading the space cache
540 * info. The previous loop should have kept us from choosing this block
541 * group, but if we've moved to the state where we will wait on caching
542 * block groups we need to first check if we're doing a fast load here,
543 * so we can wait for it to finish, otherwise we could end up allocating
544 * from a block group who's cache gets evicted for one reason or
545 * another.
546 */
547 while (cache->cached == BTRFS_CACHE_FAST) {
548 struct btrfs_caching_control *ctl;
549
550 ctl = cache->caching_ctl;
551 atomic_inc(&ctl->count);
552 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
553 spin_unlock(&cache->lock);
554
555 schedule();
556
557 finish_wait(&ctl->wait, &wait);
558 put_caching_control(ctl);
559 spin_lock(&cache->lock);
560 }
561
562 if (cache->cached != BTRFS_CACHE_NO) {
563 spin_unlock(&cache->lock);
564 kfree(caching_ctl);
Yan Zheng11833d62009-09-11 16:11:19 -0400565 return 0;
Josef Bacik291c7d22011-11-14 13:52:14 -0500566 }
567 WARN_ON(cache->caching_ctl);
568 cache->caching_ctl = caching_ctl;
569 cache->cached = BTRFS_CACHE_FAST;
570 spin_unlock(&cache->lock);
Yan Zheng11833d62009-09-11 16:11:19 -0400571
Josef Bacikd53ba472012-04-12 16:03:57 -0400572 if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) {
Josef Bacik9d66e232010-08-25 16:54:15 -0400573 ret = load_free_space_cache(fs_info, cache);
574
575 spin_lock(&cache->lock);
576 if (ret == 1) {
Josef Bacik291c7d22011-11-14 13:52:14 -0500577 cache->caching_ctl = NULL;
Josef Bacik9d66e232010-08-25 16:54:15 -0400578 cache->cached = BTRFS_CACHE_FINISHED;
579 cache->last_byte_to_unpin = (u64)-1;
580 } else {
Josef Bacik291c7d22011-11-14 13:52:14 -0500581 if (load_cache_only) {
582 cache->caching_ctl = NULL;
583 cache->cached = BTRFS_CACHE_NO;
584 } else {
585 cache->cached = BTRFS_CACHE_STARTED;
586 }
Josef Bacik9d66e232010-08-25 16:54:15 -0400587 }
588 spin_unlock(&cache->lock);
Josef Bacik291c7d22011-11-14 13:52:14 -0500589 wake_up(&caching_ctl->wait);
Josef Bacik3c148742011-02-02 15:53:47 +0000590 if (ret == 1) {
Josef Bacik291c7d22011-11-14 13:52:14 -0500591 put_caching_control(caching_ctl);
Josef Bacik3c148742011-02-02 15:53:47 +0000592 free_excluded_extents(fs_info->extent_root, cache);
Josef Bacik9d66e232010-08-25 16:54:15 -0400593 return 0;
Josef Bacik3c148742011-02-02 15:53:47 +0000594 }
Josef Bacik291c7d22011-11-14 13:52:14 -0500595 } else {
596 /*
597 * We are not going to do the fast caching, set cached to the
598 * appropriate value and wakeup any waiters.
599 */
600 spin_lock(&cache->lock);
601 if (load_cache_only) {
602 cache->caching_ctl = NULL;
603 cache->cached = BTRFS_CACHE_NO;
604 } else {
605 cache->cached = BTRFS_CACHE_STARTED;
606 }
Josef Bacik817d52f2009-07-13 21:29:25 -0400607 spin_unlock(&cache->lock);
Josef Bacik291c7d22011-11-14 13:52:14 -0500608 wake_up(&caching_ctl->wait);
609 }
610
611 if (load_cache_only) {
612 put_caching_control(caching_ctl);
Yan Zheng11833d62009-09-11 16:11:19 -0400613 return 0;
Josef Bacik817d52f2009-07-13 21:29:25 -0400614 }
Josef Bacik817d52f2009-07-13 21:29:25 -0400615
Yan Zheng11833d62009-09-11 16:11:19 -0400616 down_write(&fs_info->extent_commit_sem);
Josef Bacik291c7d22011-11-14 13:52:14 -0500617 atomic_inc(&caching_ctl->count);
Yan Zheng11833d62009-09-11 16:11:19 -0400618 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
619 up_write(&fs_info->extent_commit_sem);
620
Josef Bacik11dfe352009-11-13 20:12:59 +0000621 btrfs_get_block_group(cache);
Yan Zheng11833d62009-09-11 16:11:19 -0400622
Josef Bacikbab39bf2011-06-30 14:42:28 -0400623 btrfs_queue_worker(&fs_info->caching_workers, &caching_ctl->work);
Josef Bacik817d52f2009-07-13 21:29:25 -0400624
Josef Bacikef8bbdf2008-09-23 13:14:11 -0400625 return ret;
Chris Masone37c9e62007-05-09 20:13:14 -0400626}
627
Josef Bacik0f9dd462008-09-23 13:14:11 -0400628/*
629 * return the block group that starts at or after bytenr
630 */
Chris Masond3977122009-01-05 21:25:51 -0500631static struct btrfs_block_group_cache *
632btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
Chris Mason0ef3e662008-05-24 14:04:53 -0400633{
Josef Bacik0f9dd462008-09-23 13:14:11 -0400634 struct btrfs_block_group_cache *cache;
Chris Mason0ef3e662008-05-24 14:04:53 -0400635
Josef Bacik0f9dd462008-09-23 13:14:11 -0400636 cache = block_group_cache_tree_search(info, bytenr, 0);
Chris Mason0ef3e662008-05-24 14:04:53 -0400637
Josef Bacik0f9dd462008-09-23 13:14:11 -0400638 return cache;
Chris Mason0ef3e662008-05-24 14:04:53 -0400639}
640
Josef Bacik0f9dd462008-09-23 13:14:11 -0400641/*
Sankar P9f556842009-05-14 13:52:22 -0400642 * return the block group that contains the given bytenr
Josef Bacik0f9dd462008-09-23 13:14:11 -0400643 */
Chris Masond3977122009-01-05 21:25:51 -0500644struct btrfs_block_group_cache *btrfs_lookup_block_group(
645 struct btrfs_fs_info *info,
646 u64 bytenr)
Chris Masonbe744172007-05-06 10:15:01 -0400647{
Josef Bacik0f9dd462008-09-23 13:14:11 -0400648 struct btrfs_block_group_cache *cache;
Chris Masonbe744172007-05-06 10:15:01 -0400649
Josef Bacik0f9dd462008-09-23 13:14:11 -0400650 cache = block_group_cache_tree_search(info, bytenr, 1);
Chris Mason96b51792007-10-15 16:15:19 -0400651
Josef Bacik0f9dd462008-09-23 13:14:11 -0400652 return cache;
Chris Masonbe744172007-05-06 10:15:01 -0400653}
Chris Mason0b86a832008-03-24 15:01:56 -0400654
Josef Bacik0f9dd462008-09-23 13:14:11 -0400655static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
656 u64 flags)
Chris Mason6324fbf2008-03-24 15:01:59 -0400657{
Josef Bacik0f9dd462008-09-23 13:14:11 -0400658 struct list_head *head = &info->space_info;
Josef Bacik0f9dd462008-09-23 13:14:11 -0400659 struct btrfs_space_info *found;
Chris Mason4184ea72009-03-10 12:39:20 -0400660
Ilya Dryomov52ba6922012-01-16 22:04:47 +0200661 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
Yan, Zhengb742bb82010-05-16 10:46:24 -0400662
Chris Mason4184ea72009-03-10 12:39:20 -0400663 rcu_read_lock();
664 list_for_each_entry_rcu(found, head, list) {
Josef Bacik67377732010-09-16 16:19:09 -0400665 if (found->flags & flags) {
Chris Mason4184ea72009-03-10 12:39:20 -0400666 rcu_read_unlock();
Josef Bacik0f9dd462008-09-23 13:14:11 -0400667 return found;
Chris Mason4184ea72009-03-10 12:39:20 -0400668 }
Josef Bacik0f9dd462008-09-23 13:14:11 -0400669 }
Chris Mason4184ea72009-03-10 12:39:20 -0400670 rcu_read_unlock();
Josef Bacik0f9dd462008-09-23 13:14:11 -0400671 return NULL;
Chris Mason6324fbf2008-03-24 15:01:59 -0400672}
673
Chris Mason4184ea72009-03-10 12:39:20 -0400674/*
675 * after adding space to the filesystem, we need to clear the full flags
676 * on all the space infos.
677 */
678void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
679{
680 struct list_head *head = &info->space_info;
681 struct btrfs_space_info *found;
682
683 rcu_read_lock();
684 list_for_each_entry_rcu(found, head, list)
685 found->full = 0;
686 rcu_read_unlock();
687}
688
Chris Masone02119d2008-09-05 16:13:11 -0400689/* simple helper to search for an existing extent at a given offset */
Zheng Yan31840ae2008-09-23 13:14:14 -0400690int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
Chris Masone02119d2008-09-05 16:13:11 -0400691{
692 int ret;
693 struct btrfs_key key;
Zheng Yan31840ae2008-09-23 13:14:14 -0400694 struct btrfs_path *path;
Chris Masone02119d2008-09-05 16:13:11 -0400695
Zheng Yan31840ae2008-09-23 13:14:14 -0400696 path = btrfs_alloc_path();
Mark Fashehd8926bb2011-07-13 10:38:47 -0700697 if (!path)
698 return -ENOMEM;
699
Chris Masone02119d2008-09-05 16:13:11 -0400700 key.objectid = start;
701 key.offset = len;
Josef Bacik3173a182013-03-07 14:22:04 -0500702 key.type = BTRFS_EXTENT_ITEM_KEY;
Chris Masone02119d2008-09-05 16:13:11 -0400703 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
704 0, 0);
Josef Bacik3173a182013-03-07 14:22:04 -0500705 if (ret > 0) {
706 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
707 if (key.objectid == start &&
708 key.type == BTRFS_METADATA_ITEM_KEY)
709 ret = 0;
710 }
Zheng Yan31840ae2008-09-23 13:14:14 -0400711 btrfs_free_path(path);
Chris Mason7bb86312007-12-11 09:25:06 -0500712 return ret;
713}
714
Chris Masond8d5f3e2007-12-11 12:42:00 -0500715/*
Josef Bacik3173a182013-03-07 14:22:04 -0500716 * helper function to lookup reference count and flags of a tree block.
Yan, Zhenga22285a2010-05-16 10:48:46 -0400717 *
718 * the head node for delayed ref is used to store the sum of all the
719 * reference count modifications queued up in the rbtree. the head
720 * node may also store the extent flags to set. This way you can check
721 * to see what the reference count and extent flags would be if all of
722 * the delayed refs are not processed.
723 */
724int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
725 struct btrfs_root *root, u64 bytenr,
Josef Bacik3173a182013-03-07 14:22:04 -0500726 u64 offset, int metadata, u64 *refs, u64 *flags)
Yan, Zhenga22285a2010-05-16 10:48:46 -0400727{
728 struct btrfs_delayed_ref_head *head;
729 struct btrfs_delayed_ref_root *delayed_refs;
730 struct btrfs_path *path;
731 struct btrfs_extent_item *ei;
732 struct extent_buffer *leaf;
733 struct btrfs_key key;
734 u32 item_size;
735 u64 num_refs;
736 u64 extent_flags;
737 int ret;
738
Josef Bacik3173a182013-03-07 14:22:04 -0500739 /*
740 * If we don't have skinny metadata, don't bother doing anything
741 * different
742 */
743 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) {
744 offset = root->leafsize;
745 metadata = 0;
746 }
747
Yan, Zhenga22285a2010-05-16 10:48:46 -0400748 path = btrfs_alloc_path();
749 if (!path)
750 return -ENOMEM;
751
Josef Bacik3173a182013-03-07 14:22:04 -0500752 if (metadata) {
753 key.objectid = bytenr;
754 key.type = BTRFS_METADATA_ITEM_KEY;
755 key.offset = offset;
756 } else {
757 key.objectid = bytenr;
758 key.type = BTRFS_EXTENT_ITEM_KEY;
759 key.offset = offset;
760 }
761
Yan, Zhenga22285a2010-05-16 10:48:46 -0400762 if (!trans) {
763 path->skip_locking = 1;
764 path->search_commit_root = 1;
765 }
766again:
767 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
768 &key, path, 0, 0);
769 if (ret < 0)
770 goto out_free;
771
Josef Bacik3173a182013-03-07 14:22:04 -0500772 if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
773 key.type = BTRFS_EXTENT_ITEM_KEY;
774 key.offset = root->leafsize;
775 btrfs_release_path(path);
776 goto again;
777 }
778
Yan, Zhenga22285a2010-05-16 10:48:46 -0400779 if (ret == 0) {
780 leaf = path->nodes[0];
781 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
782 if (item_size >= sizeof(*ei)) {
783 ei = btrfs_item_ptr(leaf, path->slots[0],
784 struct btrfs_extent_item);
785 num_refs = btrfs_extent_refs(leaf, ei);
786 extent_flags = btrfs_extent_flags(leaf, ei);
787 } else {
788#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
789 struct btrfs_extent_item_v0 *ei0;
790 BUG_ON(item_size != sizeof(*ei0));
791 ei0 = btrfs_item_ptr(leaf, path->slots[0],
792 struct btrfs_extent_item_v0);
793 num_refs = btrfs_extent_refs_v0(leaf, ei0);
794 /* FIXME: this isn't correct for data */
795 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
796#else
797 BUG();
798#endif
799 }
800 BUG_ON(num_refs == 0);
801 } else {
802 num_refs = 0;
803 extent_flags = 0;
804 ret = 0;
805 }
806
807 if (!trans)
808 goto out;
809
810 delayed_refs = &trans->transaction->delayed_refs;
811 spin_lock(&delayed_refs->lock);
812 head = btrfs_find_delayed_ref_head(trans, bytenr);
813 if (head) {
814 if (!mutex_trylock(&head->mutex)) {
815 atomic_inc(&head->node.refs);
816 spin_unlock(&delayed_refs->lock);
817
David Sterbab3b4aa72011-04-21 01:20:15 +0200818 btrfs_release_path(path);
Yan, Zhenga22285a2010-05-16 10:48:46 -0400819
David Sterba8cc33e52011-05-02 15:29:25 +0200820 /*
821 * Mutex was contended, block until it's released and try
822 * again
823 */
Yan, Zhenga22285a2010-05-16 10:48:46 -0400824 mutex_lock(&head->mutex);
825 mutex_unlock(&head->mutex);
826 btrfs_put_delayed_ref(&head->node);
827 goto again;
828 }
829 if (head->extent_op && head->extent_op->update_flags)
830 extent_flags |= head->extent_op->flags_to_set;
831 else
832 BUG_ON(num_refs == 0);
833
834 num_refs += head->node.ref_mod;
835 mutex_unlock(&head->mutex);
836 }
837 spin_unlock(&delayed_refs->lock);
838out:
839 WARN_ON(num_refs == 0);
840 if (refs)
841 *refs = num_refs;
842 if (flags)
843 *flags = extent_flags;
844out_free:
845 btrfs_free_path(path);
846 return ret;
847}
848
849/*
Chris Masond8d5f3e2007-12-11 12:42:00 -0500850 * Back reference rules. Back refs have three main goals:
851 *
852 * 1) differentiate between all holders of references to an extent so that
853 * when a reference is dropped we can make sure it was a valid reference
854 * before freeing the extent.
855 *
856 * 2) Provide enough information to quickly find the holders of an extent
857 * if we notice a given block is corrupted or bad.
858 *
859 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
860 * maintenance. This is actually the same as #2, but with a slightly
861 * different use case.
862 *
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400863 * There are two kinds of back refs. The implicit back refs is optimized
864 * for pointers in non-shared tree blocks. For a given pointer in a block,
865 * back refs of this kind provide information about the block's owner tree
866 * and the pointer's key. These information allow us to find the block by
867 * b-tree searching. The full back refs is for pointers in tree blocks not
868 * referenced by their owner trees. The location of tree block is recorded
869 * in the back refs. Actually the full back refs is generic, and can be
870 * used in all cases the implicit back refs is used. The major shortcoming
871 * of the full back refs is its overhead. Every time a tree block gets
872 * COWed, we have to update back refs entry for all pointers in it.
873 *
874 * For a newly allocated tree block, we use implicit back refs for
875 * pointers in it. This means most tree related operations only involve
876 * implicit back refs. For a tree block created in old transaction, the
877 * only way to drop a reference to it is COW it. So we can detect the
878 * event that tree block loses its owner tree's reference and do the
879 * back refs conversion.
880 *
881 * When a tree block is COW'd through a tree, there are four cases:
882 *
883 * The reference count of the block is one and the tree is the block's
884 * owner tree. Nothing to do in this case.
885 *
886 * The reference count of the block is one and the tree is not the
887 * block's owner tree. In this case, full back refs is used for pointers
888 * in the block. Remove these full back refs, add implicit back refs for
889 * every pointers in the new block.
890 *
891 * The reference count of the block is greater than one and the tree is
892 * the block's owner tree. In this case, implicit back refs is used for
893 * pointers in the block. Add full back refs for every pointers in the
894 * block, increase lower level extents' reference counts. The original
895 * implicit back refs are entailed to the new block.
896 *
897 * The reference count of the block is greater than one and the tree is
898 * not the block's owner tree. Add implicit back refs for every pointer in
899 * the new block, increase lower level extents' reference count.
900 *
901 * Back Reference Key composing:
902 *
903 * The key objectid corresponds to the first byte in the extent,
904 * The key type is used to differentiate between types of back refs.
905 * There are different meanings of the key offset for different types
906 * of back refs.
907 *
Chris Masond8d5f3e2007-12-11 12:42:00 -0500908 * File extents can be referenced by:
909 *
910 * - multiple snapshots, subvolumes, or different generations in one subvol
Zheng Yan31840ae2008-09-23 13:14:14 -0400911 * - different files inside a single subvolume
Chris Masond8d5f3e2007-12-11 12:42:00 -0500912 * - different offsets inside a file (bookend extents in file.c)
913 *
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400914 * The extent ref structure for the implicit back refs has fields for:
Chris Masond8d5f3e2007-12-11 12:42:00 -0500915 *
916 * - Objectid of the subvolume root
Chris Masond8d5f3e2007-12-11 12:42:00 -0500917 * - objectid of the file holding the reference
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400918 * - original offset in the file
919 * - how many bookend extents
Zheng Yan31840ae2008-09-23 13:14:14 -0400920 *
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400921 * The key offset for the implicit back refs is hash of the first
922 * three fields.
Chris Masond8d5f3e2007-12-11 12:42:00 -0500923 *
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400924 * The extent ref structure for the full back refs has field for:
Chris Masond8d5f3e2007-12-11 12:42:00 -0500925 *
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400926 * - number of pointers in the tree leaf
Chris Masond8d5f3e2007-12-11 12:42:00 -0500927 *
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400928 * The key offset for the implicit back refs is the first byte of
929 * the tree leaf
Chris Masond8d5f3e2007-12-11 12:42:00 -0500930 *
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400931 * When a file extent is allocated, The implicit back refs is used.
932 * the fields are filled in:
Chris Masond8d5f3e2007-12-11 12:42:00 -0500933 *
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400934 * (root_key.objectid, inode objectid, offset in file, 1)
935 *
936 * When a file extent is removed file truncation, we find the
937 * corresponding implicit back refs and check the following fields:
938 *
939 * (btrfs_header_owner(leaf), inode objectid, offset in file)
Chris Masond8d5f3e2007-12-11 12:42:00 -0500940 *
941 * Btree extents can be referenced by:
942 *
943 * - Different subvolumes
Chris Masond8d5f3e2007-12-11 12:42:00 -0500944 *
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400945 * Both the implicit back refs and the full back refs for tree blocks
946 * only consist of key. The key offset for the implicit back refs is
947 * objectid of block's owner tree. The key offset for the full back refs
948 * is the first byte of parent block.
Chris Masond8d5f3e2007-12-11 12:42:00 -0500949 *
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400950 * When implicit back refs is used, information about the lowest key and
951 * level of the tree block are required. These information are stored in
952 * tree block info structure.
Chris Masond8d5f3e2007-12-11 12:42:00 -0500953 */
Zheng Yan31840ae2008-09-23 13:14:14 -0400954
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400955#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
956static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
957 struct btrfs_root *root,
958 struct btrfs_path *path,
959 u64 owner, u32 extra_size)
Chris Mason74493f72007-12-11 09:25:06 -0500960{
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400961 struct btrfs_extent_item *item;
962 struct btrfs_extent_item_v0 *ei0;
963 struct btrfs_extent_ref_v0 *ref0;
964 struct btrfs_tree_block_info *bi;
Zheng Yan31840ae2008-09-23 13:14:14 -0400965 struct extent_buffer *leaf;
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400966 struct btrfs_key key;
967 struct btrfs_key found_key;
968 u32 new_size = sizeof(*item);
969 u64 refs;
Chris Mason74493f72007-12-11 09:25:06 -0500970 int ret;
971
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400972 leaf = path->nodes[0];
973 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
Chris Mason74493f72007-12-11 09:25:06 -0500974
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400975 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
976 ei0 = btrfs_item_ptr(leaf, path->slots[0],
977 struct btrfs_extent_item_v0);
978 refs = btrfs_extent_refs_v0(leaf, ei0);
979
980 if (owner == (u64)-1) {
981 while (1) {
982 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
983 ret = btrfs_next_leaf(root, path);
984 if (ret < 0)
985 return ret;
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100986 BUG_ON(ret > 0); /* Corruption */
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400987 leaf = path->nodes[0];
988 }
989 btrfs_item_key_to_cpu(leaf, &found_key,
990 path->slots[0]);
991 BUG_ON(key.objectid != found_key.objectid);
992 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
993 path->slots[0]++;
994 continue;
995 }
996 ref0 = btrfs_item_ptr(leaf, path->slots[0],
997 struct btrfs_extent_ref_v0);
998 owner = btrfs_ref_objectid_v0(leaf, ref0);
999 break;
1000 }
1001 }
David Sterbab3b4aa72011-04-21 01:20:15 +02001002 btrfs_release_path(path);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001003
1004 if (owner < BTRFS_FIRST_FREE_OBJECTID)
1005 new_size += sizeof(*bi);
1006
1007 new_size -= sizeof(*ei0);
1008 ret = btrfs_search_slot(trans, root, &key, path,
1009 new_size + extra_size, 1);
Zheng Yan31840ae2008-09-23 13:14:14 -04001010 if (ret < 0)
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001011 return ret;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001012 BUG_ON(ret); /* Corruption */
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001013
Tsutomu Itoh4b90c682013-04-16 05:18:49 +00001014 btrfs_extend_item(root, path, new_size);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001015
1016 leaf = path->nodes[0];
1017 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1018 btrfs_set_extent_refs(leaf, item, refs);
1019 /* FIXME: get real generation */
1020 btrfs_set_extent_generation(leaf, item, 0);
1021 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1022 btrfs_set_extent_flags(leaf, item,
1023 BTRFS_EXTENT_FLAG_TREE_BLOCK |
1024 BTRFS_BLOCK_FLAG_FULL_BACKREF);
1025 bi = (struct btrfs_tree_block_info *)(item + 1);
1026 /* FIXME: get first key of the block */
1027 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
1028 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1029 } else {
1030 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1031 }
1032 btrfs_mark_buffer_dirty(leaf);
1033 return 0;
1034}
1035#endif
1036
1037static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1038{
1039 u32 high_crc = ~(u32)0;
1040 u32 low_crc = ~(u32)0;
1041 __le64 lenum;
1042
1043 lenum = cpu_to_le64(root_objectid);
David Woodhouse163e7832009-04-19 13:02:41 +01001044 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001045 lenum = cpu_to_le64(owner);
David Woodhouse163e7832009-04-19 13:02:41 +01001046 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001047 lenum = cpu_to_le64(offset);
David Woodhouse163e7832009-04-19 13:02:41 +01001048 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001049
1050 return ((u64)high_crc << 31) ^ (u64)low_crc;
1051}
1052
1053static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1054 struct btrfs_extent_data_ref *ref)
1055{
1056 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1057 btrfs_extent_data_ref_objectid(leaf, ref),
1058 btrfs_extent_data_ref_offset(leaf, ref));
1059}
1060
1061static int match_extent_data_ref(struct extent_buffer *leaf,
1062 struct btrfs_extent_data_ref *ref,
1063 u64 root_objectid, u64 owner, u64 offset)
1064{
1065 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1066 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1067 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1068 return 0;
1069 return 1;
1070}
1071
1072static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1073 struct btrfs_root *root,
1074 struct btrfs_path *path,
1075 u64 bytenr, u64 parent,
1076 u64 root_objectid,
1077 u64 owner, u64 offset)
1078{
1079 struct btrfs_key key;
1080 struct btrfs_extent_data_ref *ref;
1081 struct extent_buffer *leaf;
1082 u32 nritems;
1083 int ret;
1084 int recow;
1085 int err = -ENOENT;
1086
1087 key.objectid = bytenr;
1088 if (parent) {
1089 key.type = BTRFS_SHARED_DATA_REF_KEY;
1090 key.offset = parent;
1091 } else {
1092 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1093 key.offset = hash_extent_data_ref(root_objectid,
1094 owner, offset);
1095 }
1096again:
1097 recow = 0;
1098 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1099 if (ret < 0) {
1100 err = ret;
1101 goto fail;
1102 }
1103
1104 if (parent) {
1105 if (!ret)
1106 return 0;
1107#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1108 key.type = BTRFS_EXTENT_REF_V0_KEY;
David Sterbab3b4aa72011-04-21 01:20:15 +02001109 btrfs_release_path(path);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001110 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1111 if (ret < 0) {
1112 err = ret;
1113 goto fail;
1114 }
1115 if (!ret)
1116 return 0;
1117#endif
1118 goto fail;
Zheng Yan31840ae2008-09-23 13:14:14 -04001119 }
1120
1121 leaf = path->nodes[0];
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001122 nritems = btrfs_header_nritems(leaf);
1123 while (1) {
1124 if (path->slots[0] >= nritems) {
1125 ret = btrfs_next_leaf(root, path);
1126 if (ret < 0)
1127 err = ret;
1128 if (ret)
1129 goto fail;
1130
1131 leaf = path->nodes[0];
1132 nritems = btrfs_header_nritems(leaf);
1133 recow = 1;
1134 }
1135
1136 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1137 if (key.objectid != bytenr ||
1138 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1139 goto fail;
1140
1141 ref = btrfs_item_ptr(leaf, path->slots[0],
1142 struct btrfs_extent_data_ref);
1143
1144 if (match_extent_data_ref(leaf, ref, root_objectid,
1145 owner, offset)) {
1146 if (recow) {
David Sterbab3b4aa72011-04-21 01:20:15 +02001147 btrfs_release_path(path);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001148 goto again;
1149 }
1150 err = 0;
1151 break;
1152 }
1153 path->slots[0]++;
Zheng Yan31840ae2008-09-23 13:14:14 -04001154 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001155fail:
1156 return err;
Zheng Yan31840ae2008-09-23 13:14:14 -04001157}
1158
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001159static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1160 struct btrfs_root *root,
1161 struct btrfs_path *path,
1162 u64 bytenr, u64 parent,
1163 u64 root_objectid, u64 owner,
1164 u64 offset, int refs_to_add)
Zheng Yan31840ae2008-09-23 13:14:14 -04001165{
1166 struct btrfs_key key;
1167 struct extent_buffer *leaf;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001168 u32 size;
Zheng Yan31840ae2008-09-23 13:14:14 -04001169 u32 num_refs;
1170 int ret;
1171
1172 key.objectid = bytenr;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001173 if (parent) {
1174 key.type = BTRFS_SHARED_DATA_REF_KEY;
1175 key.offset = parent;
1176 size = sizeof(struct btrfs_shared_data_ref);
Zheng Yan31840ae2008-09-23 13:14:14 -04001177 } else {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001178 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1179 key.offset = hash_extent_data_ref(root_objectid,
1180 owner, offset);
1181 size = sizeof(struct btrfs_extent_data_ref);
Zheng Yan31840ae2008-09-23 13:14:14 -04001182 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001183
1184 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1185 if (ret && ret != -EEXIST)
1186 goto fail;
1187
1188 leaf = path->nodes[0];
1189 if (parent) {
1190 struct btrfs_shared_data_ref *ref;
1191 ref = btrfs_item_ptr(leaf, path->slots[0],
1192 struct btrfs_shared_data_ref);
1193 if (ret == 0) {
1194 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1195 } else {
1196 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1197 num_refs += refs_to_add;
1198 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1199 }
1200 } else {
1201 struct btrfs_extent_data_ref *ref;
1202 while (ret == -EEXIST) {
1203 ref = btrfs_item_ptr(leaf, path->slots[0],
1204 struct btrfs_extent_data_ref);
1205 if (match_extent_data_ref(leaf, ref, root_objectid,
1206 owner, offset))
1207 break;
David Sterbab3b4aa72011-04-21 01:20:15 +02001208 btrfs_release_path(path);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001209 key.offset++;
1210 ret = btrfs_insert_empty_item(trans, root, path, &key,
1211 size);
1212 if (ret && ret != -EEXIST)
1213 goto fail;
1214
1215 leaf = path->nodes[0];
1216 }
1217 ref = btrfs_item_ptr(leaf, path->slots[0],
1218 struct btrfs_extent_data_ref);
1219 if (ret == 0) {
1220 btrfs_set_extent_data_ref_root(leaf, ref,
1221 root_objectid);
1222 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1223 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1224 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1225 } else {
1226 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1227 num_refs += refs_to_add;
1228 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1229 }
1230 }
1231 btrfs_mark_buffer_dirty(leaf);
1232 ret = 0;
1233fail:
David Sterbab3b4aa72011-04-21 01:20:15 +02001234 btrfs_release_path(path);
Chris Mason7bb86312007-12-11 09:25:06 -05001235 return ret;
Chris Mason74493f72007-12-11 09:25:06 -05001236}
1237
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001238static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1239 struct btrfs_root *root,
1240 struct btrfs_path *path,
1241 int refs_to_drop)
Zheng Yan31840ae2008-09-23 13:14:14 -04001242{
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001243 struct btrfs_key key;
1244 struct btrfs_extent_data_ref *ref1 = NULL;
1245 struct btrfs_shared_data_ref *ref2 = NULL;
Zheng Yan31840ae2008-09-23 13:14:14 -04001246 struct extent_buffer *leaf;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001247 u32 num_refs = 0;
Zheng Yan31840ae2008-09-23 13:14:14 -04001248 int ret = 0;
1249
1250 leaf = path->nodes[0];
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001251 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1252
1253 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1254 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1255 struct btrfs_extent_data_ref);
1256 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1257 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1258 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1259 struct btrfs_shared_data_ref);
1260 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1261#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1262 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1263 struct btrfs_extent_ref_v0 *ref0;
1264 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1265 struct btrfs_extent_ref_v0);
1266 num_refs = btrfs_ref_count_v0(leaf, ref0);
1267#endif
1268 } else {
1269 BUG();
1270 }
1271
Chris Mason56bec292009-03-13 10:10:06 -04001272 BUG_ON(num_refs < refs_to_drop);
1273 num_refs -= refs_to_drop;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001274
Zheng Yan31840ae2008-09-23 13:14:14 -04001275 if (num_refs == 0) {
1276 ret = btrfs_del_item(trans, root, path);
1277 } else {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001278 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1279 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1280 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1281 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1282#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1283 else {
1284 struct btrfs_extent_ref_v0 *ref0;
1285 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1286 struct btrfs_extent_ref_v0);
1287 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1288 }
1289#endif
Zheng Yan31840ae2008-09-23 13:14:14 -04001290 btrfs_mark_buffer_dirty(leaf);
1291 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001292 return ret;
1293}
1294
1295static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1296 struct btrfs_path *path,
1297 struct btrfs_extent_inline_ref *iref)
1298{
1299 struct btrfs_key key;
1300 struct extent_buffer *leaf;
1301 struct btrfs_extent_data_ref *ref1;
1302 struct btrfs_shared_data_ref *ref2;
1303 u32 num_refs = 0;
1304
1305 leaf = path->nodes[0];
1306 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1307 if (iref) {
1308 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1309 BTRFS_EXTENT_DATA_REF_KEY) {
1310 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1311 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1312 } else {
1313 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1314 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1315 }
1316 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1317 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1318 struct btrfs_extent_data_ref);
1319 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1320 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1321 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1322 struct btrfs_shared_data_ref);
1323 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1324#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1325 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1326 struct btrfs_extent_ref_v0 *ref0;
1327 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1328 struct btrfs_extent_ref_v0);
1329 num_refs = btrfs_ref_count_v0(leaf, ref0);
1330#endif
1331 } else {
1332 WARN_ON(1);
1333 }
1334 return num_refs;
1335}
1336
1337static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1338 struct btrfs_root *root,
1339 struct btrfs_path *path,
1340 u64 bytenr, u64 parent,
1341 u64 root_objectid)
1342{
1343 struct btrfs_key key;
1344 int ret;
1345
1346 key.objectid = bytenr;
1347 if (parent) {
1348 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1349 key.offset = parent;
1350 } else {
1351 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1352 key.offset = root_objectid;
1353 }
1354
1355 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1356 if (ret > 0)
1357 ret = -ENOENT;
1358#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1359 if (ret == -ENOENT && parent) {
David Sterbab3b4aa72011-04-21 01:20:15 +02001360 btrfs_release_path(path);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001361 key.type = BTRFS_EXTENT_REF_V0_KEY;
1362 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1363 if (ret > 0)
1364 ret = -ENOENT;
1365 }
1366#endif
1367 return ret;
1368}
1369
1370static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1371 struct btrfs_root *root,
1372 struct btrfs_path *path,
1373 u64 bytenr, u64 parent,
1374 u64 root_objectid)
1375{
1376 struct btrfs_key key;
1377 int ret;
1378
1379 key.objectid = bytenr;
1380 if (parent) {
1381 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1382 key.offset = parent;
1383 } else {
1384 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1385 key.offset = root_objectid;
1386 }
1387
1388 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
David Sterbab3b4aa72011-04-21 01:20:15 +02001389 btrfs_release_path(path);
Zheng Yan31840ae2008-09-23 13:14:14 -04001390 return ret;
1391}
1392
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001393static inline int extent_ref_type(u64 parent, u64 owner)
1394{
1395 int type;
1396 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1397 if (parent > 0)
1398 type = BTRFS_SHARED_BLOCK_REF_KEY;
1399 else
1400 type = BTRFS_TREE_BLOCK_REF_KEY;
1401 } else {
1402 if (parent > 0)
1403 type = BTRFS_SHARED_DATA_REF_KEY;
1404 else
1405 type = BTRFS_EXTENT_DATA_REF_KEY;
1406 }
1407 return type;
1408}
1409
Yan Zheng2c47e6052009-06-27 21:07:35 -04001410static int find_next_key(struct btrfs_path *path, int level,
1411 struct btrfs_key *key)
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001412
1413{
Yan Zheng2c47e6052009-06-27 21:07:35 -04001414 for (; level < BTRFS_MAX_LEVEL; level++) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001415 if (!path->nodes[level])
1416 break;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001417 if (path->slots[level] + 1 >=
1418 btrfs_header_nritems(path->nodes[level]))
1419 continue;
1420 if (level == 0)
1421 btrfs_item_key_to_cpu(path->nodes[level], key,
1422 path->slots[level] + 1);
1423 else
1424 btrfs_node_key_to_cpu(path->nodes[level], key,
1425 path->slots[level] + 1);
1426 return 0;
1427 }
1428 return 1;
1429}
1430
1431/*
1432 * look for inline back ref. if back ref is found, *ref_ret is set
1433 * to the address of inline back ref, and 0 is returned.
1434 *
1435 * if back ref isn't found, *ref_ret is set to the address where it
1436 * should be inserted, and -ENOENT is returned.
1437 *
1438 * if insert is true and there are too many inline back refs, the path
1439 * points to the extent item, and -EAGAIN is returned.
1440 *
1441 * NOTE: inline back refs are ordered in the same way that back ref
1442 * items in the tree are ordered.
1443 */
1444static noinline_for_stack
1445int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1446 struct btrfs_root *root,
1447 struct btrfs_path *path,
1448 struct btrfs_extent_inline_ref **ref_ret,
1449 u64 bytenr, u64 num_bytes,
1450 u64 parent, u64 root_objectid,
1451 u64 owner, u64 offset, int insert)
1452{
1453 struct btrfs_key key;
1454 struct extent_buffer *leaf;
1455 struct btrfs_extent_item *ei;
1456 struct btrfs_extent_inline_ref *iref;
1457 u64 flags;
1458 u64 item_size;
1459 unsigned long ptr;
1460 unsigned long end;
1461 int extra_size;
1462 int type;
1463 int want;
1464 int ret;
1465 int err = 0;
Josef Bacik3173a182013-03-07 14:22:04 -05001466 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
1467 SKINNY_METADATA);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001468
1469 key.objectid = bytenr;
1470 key.type = BTRFS_EXTENT_ITEM_KEY;
1471 key.offset = num_bytes;
1472
1473 want = extent_ref_type(parent, owner);
1474 if (insert) {
1475 extra_size = btrfs_extent_inline_ref_size(want);
Yan Zheng85d41982009-06-11 08:51:10 -04001476 path->keep_locks = 1;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001477 } else
1478 extra_size = -1;
Josef Bacik3173a182013-03-07 14:22:04 -05001479
1480 /*
1481 * Owner is our parent level, so we can just add one to get the level
1482 * for the block we are interested in.
1483 */
1484 if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
1485 key.type = BTRFS_METADATA_ITEM_KEY;
1486 key.offset = owner;
1487 }
1488
1489again:
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001490 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1491 if (ret < 0) {
1492 err = ret;
1493 goto out;
1494 }
Josef Bacik3173a182013-03-07 14:22:04 -05001495
1496 /*
1497 * We may be a newly converted file system which still has the old fat
1498 * extent entries for metadata, so try and see if we have one of those.
1499 */
1500 if (ret > 0 && skinny_metadata) {
1501 skinny_metadata = false;
1502 if (path->slots[0]) {
1503 path->slots[0]--;
1504 btrfs_item_key_to_cpu(path->nodes[0], &key,
1505 path->slots[0]);
1506 if (key.objectid == bytenr &&
1507 key.type == BTRFS_EXTENT_ITEM_KEY &&
1508 key.offset == num_bytes)
1509 ret = 0;
1510 }
1511 if (ret) {
1512 key.type = BTRFS_EXTENT_ITEM_KEY;
1513 key.offset = num_bytes;
1514 btrfs_release_path(path);
1515 goto again;
1516 }
1517 }
1518
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001519 if (ret && !insert) {
1520 err = -ENOENT;
1521 goto out;
Josef Bacik492104c2013-03-08 15:41:02 -05001522 } else if (ret) {
1523 err = -EIO;
1524 WARN_ON(1);
1525 goto out;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001526 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001527
1528 leaf = path->nodes[0];
1529 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1530#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1531 if (item_size < sizeof(*ei)) {
1532 if (!insert) {
1533 err = -ENOENT;
1534 goto out;
1535 }
1536 ret = convert_extent_item_v0(trans, root, path, owner,
1537 extra_size);
1538 if (ret < 0) {
1539 err = ret;
1540 goto out;
1541 }
1542 leaf = path->nodes[0];
1543 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1544 }
1545#endif
1546 BUG_ON(item_size < sizeof(*ei));
1547
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001548 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1549 flags = btrfs_extent_flags(leaf, ei);
1550
1551 ptr = (unsigned long)(ei + 1);
1552 end = (unsigned long)ei + item_size;
1553
Josef Bacik3173a182013-03-07 14:22:04 -05001554 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001555 ptr += sizeof(struct btrfs_tree_block_info);
1556 BUG_ON(ptr > end);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001557 }
1558
1559 err = -ENOENT;
1560 while (1) {
1561 if (ptr >= end) {
1562 WARN_ON(ptr > end);
1563 break;
1564 }
1565 iref = (struct btrfs_extent_inline_ref *)ptr;
1566 type = btrfs_extent_inline_ref_type(leaf, iref);
1567 if (want < type)
1568 break;
1569 if (want > type) {
1570 ptr += btrfs_extent_inline_ref_size(type);
1571 continue;
1572 }
1573
1574 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1575 struct btrfs_extent_data_ref *dref;
1576 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1577 if (match_extent_data_ref(leaf, dref, root_objectid,
1578 owner, offset)) {
1579 err = 0;
1580 break;
1581 }
1582 if (hash_extent_data_ref_item(leaf, dref) <
1583 hash_extent_data_ref(root_objectid, owner, offset))
1584 break;
1585 } else {
1586 u64 ref_offset;
1587 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1588 if (parent > 0) {
1589 if (parent == ref_offset) {
1590 err = 0;
1591 break;
1592 }
1593 if (ref_offset < parent)
1594 break;
1595 } else {
1596 if (root_objectid == ref_offset) {
1597 err = 0;
1598 break;
1599 }
1600 if (ref_offset < root_objectid)
1601 break;
1602 }
1603 }
1604 ptr += btrfs_extent_inline_ref_size(type);
1605 }
1606 if (err == -ENOENT && insert) {
1607 if (item_size + extra_size >=
1608 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1609 err = -EAGAIN;
1610 goto out;
1611 }
1612 /*
1613 * To add new inline back ref, we have to make sure
1614 * there is no corresponding back ref item.
1615 * For simplicity, we just do not add new inline back
1616 * ref if there is any kind of item for this block
1617 */
Yan Zheng2c47e6052009-06-27 21:07:35 -04001618 if (find_next_key(path, 0, &key) == 0 &&
1619 key.objectid == bytenr &&
Yan Zheng85d41982009-06-11 08:51:10 -04001620 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001621 err = -EAGAIN;
1622 goto out;
1623 }
1624 }
1625 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1626out:
Yan Zheng85d41982009-06-11 08:51:10 -04001627 if (insert) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001628 path->keep_locks = 0;
1629 btrfs_unlock_up_safe(path, 1);
1630 }
1631 return err;
1632}
1633
1634/*
1635 * helper to add new inline back ref
1636 */
1637static noinline_for_stack
Tsutomu Itohfd279fa2013-04-16 05:19:11 +00001638void setup_inline_extent_backref(struct btrfs_root *root,
Jeff Mahoney143bede2012-03-01 14:56:26 +01001639 struct btrfs_path *path,
1640 struct btrfs_extent_inline_ref *iref,
1641 u64 parent, u64 root_objectid,
1642 u64 owner, u64 offset, int refs_to_add,
1643 struct btrfs_delayed_extent_op *extent_op)
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001644{
1645 struct extent_buffer *leaf;
1646 struct btrfs_extent_item *ei;
1647 unsigned long ptr;
1648 unsigned long end;
1649 unsigned long item_offset;
1650 u64 refs;
1651 int size;
1652 int type;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001653
1654 leaf = path->nodes[0];
1655 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1656 item_offset = (unsigned long)iref - (unsigned long)ei;
1657
1658 type = extent_ref_type(parent, owner);
1659 size = btrfs_extent_inline_ref_size(type);
1660
Tsutomu Itoh4b90c682013-04-16 05:18:49 +00001661 btrfs_extend_item(root, path, size);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001662
1663 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1664 refs = btrfs_extent_refs(leaf, ei);
1665 refs += refs_to_add;
1666 btrfs_set_extent_refs(leaf, ei, refs);
1667 if (extent_op)
1668 __run_delayed_extent_op(extent_op, leaf, ei);
1669
1670 ptr = (unsigned long)ei + item_offset;
1671 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1672 if (ptr < end - size)
1673 memmove_extent_buffer(leaf, ptr + size, ptr,
1674 end - size - ptr);
1675
1676 iref = (struct btrfs_extent_inline_ref *)ptr;
1677 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1678 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1679 struct btrfs_extent_data_ref *dref;
1680 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1681 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1682 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1683 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1684 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1685 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1686 struct btrfs_shared_data_ref *sref;
1687 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1688 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1689 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1690 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1691 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1692 } else {
1693 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1694 }
1695 btrfs_mark_buffer_dirty(leaf);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001696}
1697
1698static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1699 struct btrfs_root *root,
1700 struct btrfs_path *path,
1701 struct btrfs_extent_inline_ref **ref_ret,
1702 u64 bytenr, u64 num_bytes, u64 parent,
1703 u64 root_objectid, u64 owner, u64 offset)
1704{
1705 int ret;
1706
1707 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1708 bytenr, num_bytes, parent,
1709 root_objectid, owner, offset, 0);
1710 if (ret != -ENOENT)
1711 return ret;
1712
David Sterbab3b4aa72011-04-21 01:20:15 +02001713 btrfs_release_path(path);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001714 *ref_ret = NULL;
1715
1716 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1717 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1718 root_objectid);
1719 } else {
1720 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1721 root_objectid, owner, offset);
1722 }
1723 return ret;
1724}
1725
1726/*
1727 * helper to update/remove inline back ref
1728 */
1729static noinline_for_stack
Tsutomu Itohafe5fea2013-04-16 05:18:22 +00001730void update_inline_extent_backref(struct btrfs_root *root,
Jeff Mahoney143bede2012-03-01 14:56:26 +01001731 struct btrfs_path *path,
1732 struct btrfs_extent_inline_ref *iref,
1733 int refs_to_mod,
1734 struct btrfs_delayed_extent_op *extent_op)
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001735{
1736 struct extent_buffer *leaf;
1737 struct btrfs_extent_item *ei;
1738 struct btrfs_extent_data_ref *dref = NULL;
1739 struct btrfs_shared_data_ref *sref = NULL;
1740 unsigned long ptr;
1741 unsigned long end;
1742 u32 item_size;
1743 int size;
1744 int type;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001745 u64 refs;
1746
1747 leaf = path->nodes[0];
1748 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1749 refs = btrfs_extent_refs(leaf, ei);
1750 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1751 refs += refs_to_mod;
1752 btrfs_set_extent_refs(leaf, ei, refs);
1753 if (extent_op)
1754 __run_delayed_extent_op(extent_op, leaf, ei);
1755
1756 type = btrfs_extent_inline_ref_type(leaf, iref);
1757
1758 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1759 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1760 refs = btrfs_extent_data_ref_count(leaf, dref);
1761 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1762 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1763 refs = btrfs_shared_data_ref_count(leaf, sref);
1764 } else {
1765 refs = 1;
1766 BUG_ON(refs_to_mod != -1);
1767 }
1768
1769 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1770 refs += refs_to_mod;
1771
1772 if (refs > 0) {
1773 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1774 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1775 else
1776 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1777 } else {
1778 size = btrfs_extent_inline_ref_size(type);
1779 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1780 ptr = (unsigned long)iref;
1781 end = (unsigned long)ei + item_size;
1782 if (ptr + size < end)
1783 memmove_extent_buffer(leaf, ptr, ptr + size,
1784 end - ptr - size);
1785 item_size -= size;
Tsutomu Itohafe5fea2013-04-16 05:18:22 +00001786 btrfs_truncate_item(root, path, item_size, 1);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001787 }
1788 btrfs_mark_buffer_dirty(leaf);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001789}
1790
1791static noinline_for_stack
1792int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1793 struct btrfs_root *root,
1794 struct btrfs_path *path,
1795 u64 bytenr, u64 num_bytes, u64 parent,
1796 u64 root_objectid, u64 owner,
1797 u64 offset, int refs_to_add,
1798 struct btrfs_delayed_extent_op *extent_op)
1799{
1800 struct btrfs_extent_inline_ref *iref;
1801 int ret;
1802
1803 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1804 bytenr, num_bytes, parent,
1805 root_objectid, owner, offset, 1);
1806 if (ret == 0) {
1807 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
Tsutomu Itohafe5fea2013-04-16 05:18:22 +00001808 update_inline_extent_backref(root, path, iref,
Jeff Mahoney143bede2012-03-01 14:56:26 +01001809 refs_to_add, extent_op);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001810 } else if (ret == -ENOENT) {
Tsutomu Itohfd279fa2013-04-16 05:19:11 +00001811 setup_inline_extent_backref(root, path, iref, parent,
Jeff Mahoney143bede2012-03-01 14:56:26 +01001812 root_objectid, owner, offset,
1813 refs_to_add, extent_op);
1814 ret = 0;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001815 }
1816 return ret;
1817}
1818
1819static int insert_extent_backref(struct btrfs_trans_handle *trans,
1820 struct btrfs_root *root,
1821 struct btrfs_path *path,
1822 u64 bytenr, u64 parent, u64 root_objectid,
1823 u64 owner, u64 offset, int refs_to_add)
1824{
1825 int ret;
1826 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1827 BUG_ON(refs_to_add != 1);
1828 ret = insert_tree_block_ref(trans, root, path, bytenr,
1829 parent, root_objectid);
1830 } else {
1831 ret = insert_extent_data_ref(trans, root, path, bytenr,
1832 parent, root_objectid,
1833 owner, offset, refs_to_add);
1834 }
1835 return ret;
1836}
1837
1838static int remove_extent_backref(struct btrfs_trans_handle *trans,
1839 struct btrfs_root *root,
1840 struct btrfs_path *path,
1841 struct btrfs_extent_inline_ref *iref,
1842 int refs_to_drop, int is_data)
1843{
Jeff Mahoney143bede2012-03-01 14:56:26 +01001844 int ret = 0;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001845
1846 BUG_ON(!is_data && refs_to_drop != 1);
1847 if (iref) {
Tsutomu Itohafe5fea2013-04-16 05:18:22 +00001848 update_inline_extent_backref(root, path, iref,
Jeff Mahoney143bede2012-03-01 14:56:26 +01001849 -refs_to_drop, NULL);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001850 } else if (is_data) {
1851 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1852 } else {
1853 ret = btrfs_del_item(trans, root, path);
1854 }
1855 return ret;
1856}
1857
Li Dongyang5378e602011-03-24 10:24:27 +00001858static int btrfs_issue_discard(struct block_device *bdev,
Chris Mason15916de2008-11-19 21:17:22 -05001859 u64 start, u64 len)
1860{
Li Dongyang5378e602011-03-24 10:24:27 +00001861 return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
Chris Mason15916de2008-11-19 21:17:22 -05001862}
Chris Mason15916de2008-11-19 21:17:22 -05001863
Liu Hui1f3c79a2009-01-05 15:57:51 -05001864static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
Li Dongyang5378e602011-03-24 10:24:27 +00001865 u64 num_bytes, u64 *actual_bytes)
Liu Hui1f3c79a2009-01-05 15:57:51 -05001866{
Liu Hui1f3c79a2009-01-05 15:57:51 -05001867 int ret;
Li Dongyang5378e602011-03-24 10:24:27 +00001868 u64 discarded_bytes = 0;
Jan Schmidta1d3c472011-08-04 17:15:33 +02001869 struct btrfs_bio *bbio = NULL;
Liu Hui1f3c79a2009-01-05 15:57:51 -05001870
Christoph Hellwige244a0a2009-10-14 09:24:59 -04001871
Liu Hui1f3c79a2009-01-05 15:57:51 -05001872 /* Tell the block device(s) that the sectors can be discarded */
Stefan Behrens3ec706c2012-11-05 15:46:42 +01001873 ret = btrfs_map_block(root->fs_info, REQ_DISCARD,
Jan Schmidta1d3c472011-08-04 17:15:33 +02001874 bytenr, &num_bytes, &bbio, 0);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001875 /* Error condition is -ENOMEM */
Liu Hui1f3c79a2009-01-05 15:57:51 -05001876 if (!ret) {
Jan Schmidta1d3c472011-08-04 17:15:33 +02001877 struct btrfs_bio_stripe *stripe = bbio->stripes;
Liu Hui1f3c79a2009-01-05 15:57:51 -05001878 int i;
1879
Liu Hui1f3c79a2009-01-05 15:57:51 -05001880
Jan Schmidta1d3c472011-08-04 17:15:33 +02001881 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
Josef Bacikd5e20032011-08-04 14:52:27 +00001882 if (!stripe->dev->can_discard)
1883 continue;
1884
Li Dongyang5378e602011-03-24 10:24:27 +00001885 ret = btrfs_issue_discard(stripe->dev->bdev,
1886 stripe->physical,
1887 stripe->length);
1888 if (!ret)
1889 discarded_bytes += stripe->length;
1890 else if (ret != -EOPNOTSUPP)
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001891 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
Josef Bacikd5e20032011-08-04 14:52:27 +00001892
1893 /*
1894 * Just in case we get back EOPNOTSUPP for some reason,
1895 * just ignore the return value so we don't screw up
1896 * people calling discard_extent.
1897 */
1898 ret = 0;
Liu Hui1f3c79a2009-01-05 15:57:51 -05001899 }
Jan Schmidta1d3c472011-08-04 17:15:33 +02001900 kfree(bbio);
Liu Hui1f3c79a2009-01-05 15:57:51 -05001901 }
Li Dongyang5378e602011-03-24 10:24:27 +00001902
1903 if (actual_bytes)
1904 *actual_bytes = discarded_bytes;
1905
Liu Hui1f3c79a2009-01-05 15:57:51 -05001906
David Woodhouse53b381b2013-01-29 18:40:14 -05001907 if (ret == -EOPNOTSUPP)
1908 ret = 0;
Liu Hui1f3c79a2009-01-05 15:57:51 -05001909 return ret;
Liu Hui1f3c79a2009-01-05 15:57:51 -05001910}
1911
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001912/* Can return -ENOMEM */
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001913int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1914 struct btrfs_root *root,
1915 u64 bytenr, u64 num_bytes, u64 parent,
Arne Jansen66d7e7f2011-09-12 15:26:38 +02001916 u64 root_objectid, u64 owner, u64 offset, int for_cow)
Zheng Yan31840ae2008-09-23 13:14:14 -04001917{
1918 int ret;
Arne Jansen66d7e7f2011-09-12 15:26:38 +02001919 struct btrfs_fs_info *fs_info = root->fs_info;
1920
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001921 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1922 root_objectid == BTRFS_TREE_LOG_OBJECTID);
Zheng Yan31840ae2008-09-23 13:14:14 -04001923
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001924 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
Arne Jansen66d7e7f2011-09-12 15:26:38 +02001925 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
1926 num_bytes,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001927 parent, root_objectid, (int)owner,
Arne Jansen66d7e7f2011-09-12 15:26:38 +02001928 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001929 } else {
Arne Jansen66d7e7f2011-09-12 15:26:38 +02001930 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
1931 num_bytes,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001932 parent, root_objectid, owner, offset,
Arne Jansen66d7e7f2011-09-12 15:26:38 +02001933 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001934 }
Zheng Yan31840ae2008-09-23 13:14:14 -04001935 return ret;
1936}
1937
Chris Mason925baed2008-06-25 16:01:30 -04001938static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001939 struct btrfs_root *root,
1940 u64 bytenr, u64 num_bytes,
1941 u64 parent, u64 root_objectid,
1942 u64 owner, u64 offset, int refs_to_add,
1943 struct btrfs_delayed_extent_op *extent_op)
Chris Mason56bec292009-03-13 10:10:06 -04001944{
Chris Mason5caf2a02007-04-02 11:20:42 -04001945 struct btrfs_path *path;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001946 struct extent_buffer *leaf;
Chris Mason234b63a2007-03-13 10:46:10 -04001947 struct btrfs_extent_item *item;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001948 u64 refs;
1949 int ret;
1950 int err = 0;
Chris Mason037e6392007-03-07 11:50:24 -05001951
Chris Mason5caf2a02007-04-02 11:20:42 -04001952 path = btrfs_alloc_path();
Chris Mason54aa1f42007-06-22 14:16:25 -04001953 if (!path)
1954 return -ENOMEM;
Chris Mason26b80032007-08-08 20:17:12 -04001955
Chris Mason3c12ac72008-04-21 12:01:38 -04001956 path->reada = 1;
Chris Masonb9473432009-03-13 11:00:37 -04001957 path->leave_spinning = 1;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001958 /* this will setup the path even if it fails to insert the back ref */
1959 ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1960 path, bytenr, num_bytes, parent,
1961 root_objectid, owner, offset,
1962 refs_to_add, extent_op);
1963 if (ret == 0)
1964 goto out;
Zheng Yan31840ae2008-09-23 13:14:14 -04001965
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001966 if (ret != -EAGAIN) {
1967 err = ret;
1968 goto out;
Chris Masonb9473432009-03-13 11:00:37 -04001969 }
Zheng Yan31840ae2008-09-23 13:14:14 -04001970
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001971 leaf = path->nodes[0];
1972 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1973 refs = btrfs_extent_refs(leaf, item);
1974 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1975 if (extent_op)
1976 __run_delayed_extent_op(extent_op, leaf, item);
Zheng Yan31840ae2008-09-23 13:14:14 -04001977
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001978 btrfs_mark_buffer_dirty(leaf);
David Sterbab3b4aa72011-04-21 01:20:15 +02001979 btrfs_release_path(path);
Chris Mason7bb86312007-12-11 09:25:06 -05001980
Chris Mason3c12ac72008-04-21 12:01:38 -04001981 path->reada = 1;
Chris Masonb9473432009-03-13 11:00:37 -04001982 path->leave_spinning = 1;
1983
Chris Mason56bec292009-03-13 10:10:06 -04001984 /* now insert the actual backref */
Zheng Yan31840ae2008-09-23 13:14:14 -04001985 ret = insert_extent_backref(trans, root->fs_info->extent_root,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001986 path, bytenr, parent, root_objectid,
1987 owner, offset, refs_to_add);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001988 if (ret)
1989 btrfs_abort_transaction(trans, root, ret);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001990out:
Chris Mason74493f72007-12-11 09:25:06 -05001991 btrfs_free_path(path);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001992 return err;
Chris Mason02217ed2007-03-02 16:08:05 -05001993}
1994
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001995static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1996 struct btrfs_root *root,
1997 struct btrfs_delayed_ref_node *node,
1998 struct btrfs_delayed_extent_op *extent_op,
1999 int insert_reserved)
Chris Masone9d0b132007-08-10 14:06:19 -04002000{
Chris Mason56bec292009-03-13 10:10:06 -04002001 int ret = 0;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002002 struct btrfs_delayed_data_ref *ref;
2003 struct btrfs_key ins;
2004 u64 parent = 0;
2005 u64 ref_root = 0;
2006 u64 flags = 0;
Chris Mason56bec292009-03-13 10:10:06 -04002007
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002008 ins.objectid = node->bytenr;
2009 ins.offset = node->num_bytes;
2010 ins.type = BTRFS_EXTENT_ITEM_KEY;
Chris Mason56bec292009-03-13 10:10:06 -04002011
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002012 ref = btrfs_delayed_node_to_data_ref(node);
2013 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2014 parent = ref->parent;
2015 else
2016 ref_root = ref->root;
2017
2018 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
Josef Bacik3173a182013-03-07 14:22:04 -05002019 if (extent_op)
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002020 flags |= extent_op->flags_to_set;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002021 ret = alloc_reserved_file_extent(trans, root,
2022 parent, ref_root, flags,
2023 ref->objectid, ref->offset,
2024 &ins, node->ref_mod);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002025 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2026 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2027 node->num_bytes, parent,
2028 ref_root, ref->objectid,
2029 ref->offset, node->ref_mod,
2030 extent_op);
2031 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2032 ret = __btrfs_free_extent(trans, root, node->bytenr,
2033 node->num_bytes, parent,
2034 ref_root, ref->objectid,
2035 ref->offset, node->ref_mod,
2036 extent_op);
2037 } else {
2038 BUG();
2039 }
Chris Mason56bec292009-03-13 10:10:06 -04002040 return ret;
2041}
2042
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002043static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2044 struct extent_buffer *leaf,
2045 struct btrfs_extent_item *ei)
2046{
2047 u64 flags = btrfs_extent_flags(leaf, ei);
2048 if (extent_op->update_flags) {
2049 flags |= extent_op->flags_to_set;
2050 btrfs_set_extent_flags(leaf, ei, flags);
2051 }
2052
2053 if (extent_op->update_key) {
2054 struct btrfs_tree_block_info *bi;
2055 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2056 bi = (struct btrfs_tree_block_info *)(ei + 1);
2057 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2058 }
2059}
2060
2061static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2062 struct btrfs_root *root,
2063 struct btrfs_delayed_ref_node *node,
2064 struct btrfs_delayed_extent_op *extent_op)
2065{
2066 struct btrfs_key key;
2067 struct btrfs_path *path;
2068 struct btrfs_extent_item *ei;
2069 struct extent_buffer *leaf;
2070 u32 item_size;
2071 int ret;
2072 int err = 0;
Josef Bacikb1c79e02013-05-09 13:49:30 -04002073 int metadata = !extent_op->is_data;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002074
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002075 if (trans->aborted)
2076 return 0;
2077
Josef Bacik3173a182013-03-07 14:22:04 -05002078 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2079 metadata = 0;
2080
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002081 path = btrfs_alloc_path();
2082 if (!path)
2083 return -ENOMEM;
2084
2085 key.objectid = node->bytenr;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002086
Josef Bacik3173a182013-03-07 14:22:04 -05002087 if (metadata) {
Josef Bacik3173a182013-03-07 14:22:04 -05002088 key.type = BTRFS_METADATA_ITEM_KEY;
Josef Bacikb1c79e02013-05-09 13:49:30 -04002089 key.offset = extent_op->level;
Josef Bacik3173a182013-03-07 14:22:04 -05002090 } else {
2091 key.type = BTRFS_EXTENT_ITEM_KEY;
2092 key.offset = node->num_bytes;
2093 }
2094
2095again:
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002096 path->reada = 1;
2097 path->leave_spinning = 1;
2098 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2099 path, 0, 1);
2100 if (ret < 0) {
2101 err = ret;
2102 goto out;
2103 }
2104 if (ret > 0) {
Josef Bacik3173a182013-03-07 14:22:04 -05002105 if (metadata) {
2106 btrfs_release_path(path);
2107 metadata = 0;
2108
2109 key.offset = node->num_bytes;
2110 key.type = BTRFS_EXTENT_ITEM_KEY;
2111 goto again;
2112 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002113 err = -EIO;
2114 goto out;
2115 }
2116
2117 leaf = path->nodes[0];
2118 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2119#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2120 if (item_size < sizeof(*ei)) {
2121 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2122 path, (u64)-1, 0);
2123 if (ret < 0) {
2124 err = ret;
2125 goto out;
2126 }
2127 leaf = path->nodes[0];
2128 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2129 }
2130#endif
2131 BUG_ON(item_size < sizeof(*ei));
2132 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2133 __run_delayed_extent_op(extent_op, leaf, ei);
2134
2135 btrfs_mark_buffer_dirty(leaf);
2136out:
2137 btrfs_free_path(path);
2138 return err;
2139}
2140
2141static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2142 struct btrfs_root *root,
2143 struct btrfs_delayed_ref_node *node,
2144 struct btrfs_delayed_extent_op *extent_op,
2145 int insert_reserved)
2146{
2147 int ret = 0;
2148 struct btrfs_delayed_tree_ref *ref;
2149 struct btrfs_key ins;
2150 u64 parent = 0;
2151 u64 ref_root = 0;
Josef Bacik3173a182013-03-07 14:22:04 -05002152 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
2153 SKINNY_METADATA);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002154
2155 ref = btrfs_delayed_node_to_tree_ref(node);
2156 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2157 parent = ref->parent;
2158 else
2159 ref_root = ref->root;
2160
Josef Bacik3173a182013-03-07 14:22:04 -05002161 ins.objectid = node->bytenr;
2162 if (skinny_metadata) {
2163 ins.offset = ref->level;
2164 ins.type = BTRFS_METADATA_ITEM_KEY;
2165 } else {
2166 ins.offset = node->num_bytes;
2167 ins.type = BTRFS_EXTENT_ITEM_KEY;
2168 }
2169
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002170 BUG_ON(node->ref_mod != 1);
2171 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
Josef Bacik3173a182013-03-07 14:22:04 -05002172 BUG_ON(!extent_op || !extent_op->update_flags);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002173 ret = alloc_reserved_tree_block(trans, root,
2174 parent, ref_root,
2175 extent_op->flags_to_set,
2176 &extent_op->key,
2177 ref->level, &ins);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002178 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2179 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2180 node->num_bytes, parent, ref_root,
2181 ref->level, 0, 1, extent_op);
2182 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2183 ret = __btrfs_free_extent(trans, root, node->bytenr,
2184 node->num_bytes, parent, ref_root,
2185 ref->level, 0, 1, extent_op);
2186 } else {
2187 BUG();
2188 }
2189 return ret;
2190}
2191
Chris Mason56bec292009-03-13 10:10:06 -04002192/* helper function to actually process a single delayed ref entry */
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002193static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2194 struct btrfs_root *root,
2195 struct btrfs_delayed_ref_node *node,
2196 struct btrfs_delayed_extent_op *extent_op,
2197 int insert_reserved)
Chris Mason56bec292009-03-13 10:10:06 -04002198{
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002199 int ret = 0;
2200
2201 if (trans->aborted)
2202 return 0;
2203
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002204 if (btrfs_delayed_ref_is_head(node)) {
Chris Mason56bec292009-03-13 10:10:06 -04002205 struct btrfs_delayed_ref_head *head;
2206 /*
2207 * we've hit the end of the chain and we were supposed
2208 * to insert this extent into the tree. But, it got
2209 * deleted before we ever needed to insert it, so all
2210 * we have to do is clean up the accounting
2211 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002212 BUG_ON(extent_op);
2213 head = btrfs_delayed_node_to_head(node);
Chris Mason56bec292009-03-13 10:10:06 -04002214 if (insert_reserved) {
Yan, Zhengf0486c62010-05-16 10:46:25 -04002215 btrfs_pin_extent(root, node->bytenr,
2216 node->num_bytes, 1);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002217 if (head->is_data) {
2218 ret = btrfs_del_csums(trans, root,
2219 node->bytenr,
2220 node->num_bytes);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002221 }
Chris Mason56bec292009-03-13 10:10:06 -04002222 }
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002223 return ret;
Chris Mason56bec292009-03-13 10:10:06 -04002224 }
Josef Bacikeb099672009-02-12 09:27:38 -05002225
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002226 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2227 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2228 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2229 insert_reserved);
2230 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2231 node->type == BTRFS_SHARED_DATA_REF_KEY)
2232 ret = run_delayed_data_ref(trans, root, node, extent_op,
2233 insert_reserved);
2234 else
2235 BUG();
2236 return ret;
Chris Masone9d0b132007-08-10 14:06:19 -04002237}
2238
Chris Mason56bec292009-03-13 10:10:06 -04002239static noinline struct btrfs_delayed_ref_node *
2240select_delayed_ref(struct btrfs_delayed_ref_head *head)
Chris Masona28ec192007-03-06 20:08:01 -05002241{
Chris Mason56bec292009-03-13 10:10:06 -04002242 struct rb_node *node;
2243 struct btrfs_delayed_ref_node *ref;
2244 int action = BTRFS_ADD_DELAYED_REF;
2245again:
2246 /*
2247 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2248 * this prevents ref count from going down to zero when
2249 * there still are pending delayed ref.
2250 */
2251 node = rb_prev(&head->node.rb_node);
2252 while (1) {
2253 if (!node)
2254 break;
2255 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2256 rb_node);
2257 if (ref->bytenr != head->node.bytenr)
2258 break;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002259 if (ref->action == action)
Chris Mason56bec292009-03-13 10:10:06 -04002260 return ref;
2261 node = rb_prev(node);
Chris Mason5f39d392007-10-15 16:14:19 -04002262 }
Chris Mason56bec292009-03-13 10:10:06 -04002263 if (action == BTRFS_ADD_DELAYED_REF) {
2264 action = BTRFS_DROP_DELAYED_REF;
2265 goto again;
2266 }
2267 return NULL;
2268}
2269
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002270/*
2271 * Returns 0 on success or if called with an already aborted transaction.
2272 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2273 */
Chris Masonc3e69d52009-03-13 10:17:05 -04002274static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2275 struct btrfs_root *root,
2276 struct list_head *cluster)
Chris Mason56bec292009-03-13 10:10:06 -04002277{
Chris Mason56bec292009-03-13 10:10:06 -04002278 struct btrfs_delayed_ref_root *delayed_refs;
2279 struct btrfs_delayed_ref_node *ref;
2280 struct btrfs_delayed_ref_head *locked_ref = NULL;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002281 struct btrfs_delayed_extent_op *extent_op;
Jan Schmidt097b8a72012-06-21 11:08:04 +02002282 struct btrfs_fs_info *fs_info = root->fs_info;
Chris Mason56bec292009-03-13 10:10:06 -04002283 int ret;
Chris Masonc3e69d52009-03-13 10:17:05 -04002284 int count = 0;
Chris Mason56bec292009-03-13 10:10:06 -04002285 int must_insert_reserved = 0;
Chris Mason56bec292009-03-13 10:10:06 -04002286
2287 delayed_refs = &trans->transaction->delayed_refs;
Chris Mason56bec292009-03-13 10:10:06 -04002288 while (1) {
2289 if (!locked_ref) {
Chris Masonc3e69d52009-03-13 10:17:05 -04002290 /* pick a new head ref from the cluster list */
2291 if (list_empty(cluster))
Chris Mason56bec292009-03-13 10:10:06 -04002292 break;
Chris Mason56bec292009-03-13 10:10:06 -04002293
Chris Masonc3e69d52009-03-13 10:17:05 -04002294 locked_ref = list_entry(cluster->next,
2295 struct btrfs_delayed_ref_head, cluster);
2296
2297 /* grab the lock that says we are going to process
2298 * all the refs for this head */
2299 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2300
2301 /*
2302 * we may have dropped the spin lock to get the head
2303 * mutex lock, and that might have given someone else
2304 * time to free the head. If that's true, it has been
2305 * removed from our list and we can move on.
2306 */
2307 if (ret == -EAGAIN) {
2308 locked_ref = NULL;
2309 count++;
2310 continue;
Chris Mason56bec292009-03-13 10:10:06 -04002311 }
2312 }
2313
2314 /*
Josef Bacikae1e2062012-08-07 16:00:32 -04002315 * We need to try and merge add/drops of the same ref since we
2316 * can run into issues with relocate dropping the implicit ref
2317 * and then it being added back again before the drop can
2318 * finish. If we merged anything we need to re-loop so we can
2319 * get a good ref.
2320 */
2321 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2322 locked_ref);
2323
2324 /*
Arne Jansend1270cd2011-09-13 15:16:43 +02002325 * locked_ref is the head node, so we have to go one
2326 * node back for any delayed ref updates
2327 */
2328 ref = select_delayed_ref(locked_ref);
2329
2330 if (ref && ref->seq &&
Jan Schmidt097b8a72012-06-21 11:08:04 +02002331 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
Arne Jansend1270cd2011-09-13 15:16:43 +02002332 /*
2333 * there are still refs with lower seq numbers in the
2334 * process of being added. Don't run this ref yet.
2335 */
2336 list_del_init(&locked_ref->cluster);
Miao Xie093486c2012-12-19 08:10:10 +00002337 btrfs_delayed_ref_unlock(locked_ref);
Arne Jansend1270cd2011-09-13 15:16:43 +02002338 locked_ref = NULL;
2339 delayed_refs->num_heads_ready++;
2340 spin_unlock(&delayed_refs->lock);
2341 cond_resched();
2342 spin_lock(&delayed_refs->lock);
2343 continue;
2344 }
2345
2346 /*
Chris Mason56bec292009-03-13 10:10:06 -04002347 * record the must insert reserved flag before we
2348 * drop the spin lock.
2349 */
2350 must_insert_reserved = locked_ref->must_insert_reserved;
2351 locked_ref->must_insert_reserved = 0;
2352
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002353 extent_op = locked_ref->extent_op;
2354 locked_ref->extent_op = NULL;
2355
Chris Mason56bec292009-03-13 10:10:06 -04002356 if (!ref) {
2357 /* All delayed refs have been processed, Go ahead
2358 * and send the head node to run_one_delayed_ref,
2359 * so that any accounting fixes can happen
2360 */
2361 ref = &locked_ref->node;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002362
2363 if (extent_op && must_insert_reserved) {
Miao Xie78a61842012-11-21 02:21:28 +00002364 btrfs_free_delayed_extent_op(extent_op);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002365 extent_op = NULL;
2366 }
2367
2368 if (extent_op) {
2369 spin_unlock(&delayed_refs->lock);
2370
2371 ret = run_delayed_extent_op(trans, root,
2372 ref, extent_op);
Miao Xie78a61842012-11-21 02:21:28 +00002373 btrfs_free_delayed_extent_op(extent_op);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002374
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002375 if (ret) {
Simon Kirbyc2cf52e2013-03-19 22:41:23 +00002376 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
Dan Carpenter253beeb2012-04-18 09:59:03 +03002377 spin_lock(&delayed_refs->lock);
Miao Xie093486c2012-12-19 08:10:10 +00002378 btrfs_delayed_ref_unlock(locked_ref);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002379 return ret;
2380 }
2381
Chris Mason203bf282012-01-06 15:23:57 -05002382 goto next;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002383 }
Chris Mason56bec292009-03-13 10:10:06 -04002384 }
2385
2386 ref->in_tree = 0;
2387 rb_erase(&ref->rb_node, &delayed_refs->root);
2388 delayed_refs->num_entries--;
Miao Xie093486c2012-12-19 08:10:10 +00002389 if (!btrfs_delayed_ref_is_head(ref)) {
Arne Jansen22cd2e72012-08-09 00:16:53 -06002390 /*
2391 * when we play the delayed ref, also correct the
2392 * ref_mod on head
2393 */
2394 switch (ref->action) {
2395 case BTRFS_ADD_DELAYED_REF:
2396 case BTRFS_ADD_DELAYED_EXTENT:
2397 locked_ref->node.ref_mod -= ref->ref_mod;
2398 break;
2399 case BTRFS_DROP_DELAYED_REF:
2400 locked_ref->node.ref_mod += ref->ref_mod;
2401 break;
2402 default:
2403 WARN_ON(1);
2404 }
Josef Bacik322d9a92013-08-22 17:03:29 -04002405 } else {
2406 list_del_init(&locked_ref->cluster);
Arne Jansen22cd2e72012-08-09 00:16:53 -06002407 }
Chris Mason56bec292009-03-13 10:10:06 -04002408 spin_unlock(&delayed_refs->lock);
2409
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002410 ret = run_one_delayed_ref(trans, root, ref, extent_op,
Chris Mason56bec292009-03-13 10:10:06 -04002411 must_insert_reserved);
Chris Mason56bec292009-03-13 10:10:06 -04002412
Miao Xie78a61842012-11-21 02:21:28 +00002413 btrfs_free_delayed_extent_op(extent_op);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002414 if (ret) {
Miao Xie093486c2012-12-19 08:10:10 +00002415 btrfs_delayed_ref_unlock(locked_ref);
2416 btrfs_put_delayed_ref(ref);
Simon Kirbyc2cf52e2013-03-19 22:41:23 +00002417 btrfs_debug(fs_info, "run_one_delayed_ref returned %d", ret);
Dan Carpenter253beeb2012-04-18 09:59:03 +03002418 spin_lock(&delayed_refs->lock);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002419 return ret;
2420 }
2421
Miao Xie093486c2012-12-19 08:10:10 +00002422 /*
2423 * If this node is a head, that means all the refs in this head
2424 * have been dealt with, and we will pick the next head to deal
2425 * with, so we must unlock the head and drop it from the cluster
2426 * list before we release it.
2427 */
2428 if (btrfs_delayed_ref_is_head(ref)) {
Miao Xie093486c2012-12-19 08:10:10 +00002429 btrfs_delayed_ref_unlock(locked_ref);
2430 locked_ref = NULL;
2431 }
2432 btrfs_put_delayed_ref(ref);
2433 count++;
Chris Mason203bf282012-01-06 15:23:57 -05002434next:
Chris Mason1887be62009-03-13 10:11:24 -04002435 cond_resched();
Chris Mason56bec292009-03-13 10:10:06 -04002436 spin_lock(&delayed_refs->lock);
2437 }
Chris Masonc3e69d52009-03-13 10:17:05 -04002438 return count;
2439}
2440
Arne Jansen709c0482011-09-12 12:22:57 +02002441#ifdef SCRAMBLE_DELAYED_REFS
2442/*
2443 * Normally delayed refs get processed in ascending bytenr order. This
2444 * correlates in most cases to the order added. To expose dependencies on this
2445 * order, we start to process the tree in the middle instead of the beginning
2446 */
2447static u64 find_middle(struct rb_root *root)
2448{
2449 struct rb_node *n = root->rb_node;
2450 struct btrfs_delayed_ref_node *entry;
2451 int alt = 1;
2452 u64 middle;
2453 u64 first = 0, last = 0;
2454
2455 n = rb_first(root);
2456 if (n) {
2457 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2458 first = entry->bytenr;
2459 }
2460 n = rb_last(root);
2461 if (n) {
2462 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2463 last = entry->bytenr;
2464 }
2465 n = root->rb_node;
2466
2467 while (n) {
2468 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2469 WARN_ON(!entry->in_tree);
2470
2471 middle = entry->bytenr;
2472
2473 if (alt)
2474 n = n->rb_left;
2475 else
2476 n = n->rb_right;
2477
2478 alt = 1 - alt;
2479 }
2480 return middle;
2481}
2482#endif
2483
Arne Jansenbed92ea2012-06-28 18:03:02 +02002484int btrfs_delayed_refs_qgroup_accounting(struct btrfs_trans_handle *trans,
2485 struct btrfs_fs_info *fs_info)
2486{
2487 struct qgroup_update *qgroup_update;
2488 int ret = 0;
2489
2490 if (list_empty(&trans->qgroup_ref_list) !=
2491 !trans->delayed_ref_elem.seq) {
2492 /* list without seq or seq without list */
Simon Kirbyc2cf52e2013-03-19 22:41:23 +00002493 btrfs_err(fs_info,
Jan Schmidtfc36ed7e2013-04-24 16:57:33 +00002494 "qgroup accounting update error, list is%s empty, seq is %#x.%x",
Arne Jansenbed92ea2012-06-28 18:03:02 +02002495 list_empty(&trans->qgroup_ref_list) ? "" : " not",
Jan Schmidtfc36ed7e2013-04-24 16:57:33 +00002496 (u32)(trans->delayed_ref_elem.seq >> 32),
2497 (u32)trans->delayed_ref_elem.seq);
Arne Jansenbed92ea2012-06-28 18:03:02 +02002498 BUG();
2499 }
2500
2501 if (!trans->delayed_ref_elem.seq)
2502 return 0;
2503
2504 while (!list_empty(&trans->qgroup_ref_list)) {
2505 qgroup_update = list_first_entry(&trans->qgroup_ref_list,
2506 struct qgroup_update, list);
2507 list_del(&qgroup_update->list);
2508 if (!ret)
2509 ret = btrfs_qgroup_account_ref(
2510 trans, fs_info, qgroup_update->node,
2511 qgroup_update->extent_op);
2512 kfree(qgroup_update);
2513 }
2514
2515 btrfs_put_tree_mod_seq(fs_info, &trans->delayed_ref_elem);
2516
2517 return ret;
2518}
2519
Chris Masonbb721702013-01-29 18:44:12 -05002520static int refs_newer(struct btrfs_delayed_ref_root *delayed_refs, int seq,
2521 int count)
2522{
2523 int val = atomic_read(&delayed_refs->ref_seq);
2524
2525 if (val < seq || val >= seq + count)
2526 return 1;
2527 return 0;
2528}
2529
Chris Masonc3e69d52009-03-13 10:17:05 -04002530/*
2531 * this starts processing the delayed reference count updates and
2532 * extent insertions we have queued up so far. count can be
2533 * 0, which means to process everything in the tree at the start
2534 * of the run (but not newly added entries), or it can be some target
2535 * number you'd like to process.
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002536 *
2537 * Returns 0 on success or if called with an aborted transaction
2538 * Returns <0 on error and aborts the transaction
Chris Masonc3e69d52009-03-13 10:17:05 -04002539 */
2540int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2541 struct btrfs_root *root, unsigned long count)
2542{
2543 struct rb_node *node;
2544 struct btrfs_delayed_ref_root *delayed_refs;
2545 struct btrfs_delayed_ref_node *ref;
2546 struct list_head cluster;
2547 int ret;
Jan Schmidta1686502011-12-12 16:10:07 +01002548 u64 delayed_start;
Chris Masonc3e69d52009-03-13 10:17:05 -04002549 int run_all = count == (unsigned long)-1;
2550 int run_most = 0;
Arne Jansen1fa11e22012-08-06 14:18:51 -06002551 int loops;
Chris Masonc3e69d52009-03-13 10:17:05 -04002552
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002553 /* We'll clean this up in btrfs_cleanup_transaction */
2554 if (trans->aborted)
2555 return 0;
2556
Chris Masonc3e69d52009-03-13 10:17:05 -04002557 if (root == root->fs_info->extent_root)
2558 root = root->fs_info->tree_root;
2559
Jan Schmidtedf39272012-06-28 18:04:55 +02002560 btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info);
2561
Chris Masonc3e69d52009-03-13 10:17:05 -04002562 delayed_refs = &trans->transaction->delayed_refs;
2563 INIT_LIST_HEAD(&cluster);
Chris Masonbb721702013-01-29 18:44:12 -05002564 if (count == 0) {
2565 count = delayed_refs->num_entries * 2;
2566 run_most = 1;
2567 }
2568
2569 if (!run_all && !run_most) {
2570 int old;
2571 int seq = atomic_read(&delayed_refs->ref_seq);
2572
2573progress:
2574 old = atomic_cmpxchg(&delayed_refs->procs_running_refs, 0, 1);
2575 if (old) {
2576 DEFINE_WAIT(__wait);
2577 if (delayed_refs->num_entries < 16348)
2578 return 0;
2579
2580 prepare_to_wait(&delayed_refs->wait, &__wait,
2581 TASK_UNINTERRUPTIBLE);
2582
2583 old = atomic_cmpxchg(&delayed_refs->procs_running_refs, 0, 1);
2584 if (old) {
2585 schedule();
2586 finish_wait(&delayed_refs->wait, &__wait);
2587
2588 if (!refs_newer(delayed_refs, seq, 256))
2589 goto progress;
2590 else
2591 return 0;
2592 } else {
2593 finish_wait(&delayed_refs->wait, &__wait);
2594 goto again;
2595 }
2596 }
2597
2598 } else {
2599 atomic_inc(&delayed_refs->procs_running_refs);
2600 }
2601
Chris Masonc3e69d52009-03-13 10:17:05 -04002602again:
Arne Jansen1fa11e22012-08-06 14:18:51 -06002603 loops = 0;
Chris Masonc3e69d52009-03-13 10:17:05 -04002604 spin_lock(&delayed_refs->lock);
Jan Schmidt097b8a72012-06-21 11:08:04 +02002605
Arne Jansen709c0482011-09-12 12:22:57 +02002606#ifdef SCRAMBLE_DELAYED_REFS
2607 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2608#endif
2609
Chris Masonc3e69d52009-03-13 10:17:05 -04002610 while (1) {
2611 if (!(run_all || run_most) &&
2612 delayed_refs->num_heads_ready < 64)
2613 break;
2614
2615 /*
2616 * go find something we can process in the rbtree. We start at
2617 * the beginning of the tree, and then build a cluster
2618 * of refs to process starting at the first one we are able to
2619 * lock
2620 */
Jan Schmidta1686502011-12-12 16:10:07 +01002621 delayed_start = delayed_refs->run_delayed_start;
Chris Masonc3e69d52009-03-13 10:17:05 -04002622 ret = btrfs_find_ref_cluster(trans, &cluster,
2623 delayed_refs->run_delayed_start);
2624 if (ret)
2625 break;
2626
2627 ret = run_clustered_refs(trans, root, &cluster);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002628 if (ret < 0) {
Miao Xie093486c2012-12-19 08:10:10 +00002629 btrfs_release_ref_cluster(&cluster);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002630 spin_unlock(&delayed_refs->lock);
2631 btrfs_abort_transaction(trans, root, ret);
Chris Masonbb721702013-01-29 18:44:12 -05002632 atomic_dec(&delayed_refs->procs_running_refs);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002633 return ret;
2634 }
Chris Masonc3e69d52009-03-13 10:17:05 -04002635
Chris Masonbb721702013-01-29 18:44:12 -05002636 atomic_add(ret, &delayed_refs->ref_seq);
2637
Chris Masonc3e69d52009-03-13 10:17:05 -04002638 count -= min_t(unsigned long, ret, count);
2639
2640 if (count == 0)
2641 break;
Jan Schmidta1686502011-12-12 16:10:07 +01002642
Arne Jansen1fa11e22012-08-06 14:18:51 -06002643 if (delayed_start >= delayed_refs->run_delayed_start) {
2644 if (loops == 0) {
2645 /*
2646 * btrfs_find_ref_cluster looped. let's do one
2647 * more cycle. if we don't run any delayed ref
2648 * during that cycle (because we can't because
2649 * all of them are blocked), bail out.
2650 */
2651 loops = 1;
2652 } else {
2653 /*
2654 * no runnable refs left, stop trying
2655 */
2656 BUG_ON(run_all);
2657 break;
2658 }
2659 }
2660 if (ret) {
Jan Schmidta1686502011-12-12 16:10:07 +01002661 /* refs were run, let's reset staleness detection */
Arne Jansen1fa11e22012-08-06 14:18:51 -06002662 loops = 0;
Jan Schmidta1686502011-12-12 16:10:07 +01002663 }
Chris Masonc3e69d52009-03-13 10:17:05 -04002664 }
2665
Chris Mason56bec292009-03-13 10:10:06 -04002666 if (run_all) {
Josef Bacikea658ba2012-09-11 16:57:25 -04002667 if (!list_empty(&trans->new_bgs)) {
2668 spin_unlock(&delayed_refs->lock);
2669 btrfs_create_pending_block_groups(trans, root);
2670 spin_lock(&delayed_refs->lock);
2671 }
2672
Chris Mason56bec292009-03-13 10:10:06 -04002673 node = rb_first(&delayed_refs->root);
Chris Masonc3e69d52009-03-13 10:17:05 -04002674 if (!node)
Chris Mason56bec292009-03-13 10:10:06 -04002675 goto out;
Chris Masonc3e69d52009-03-13 10:17:05 -04002676 count = (unsigned long)-1;
Chris Mason56bec292009-03-13 10:10:06 -04002677
2678 while (node) {
2679 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2680 rb_node);
2681 if (btrfs_delayed_ref_is_head(ref)) {
2682 struct btrfs_delayed_ref_head *head;
2683
2684 head = btrfs_delayed_node_to_head(ref);
2685 atomic_inc(&ref->refs);
2686
2687 spin_unlock(&delayed_refs->lock);
David Sterba8cc33e52011-05-02 15:29:25 +02002688 /*
2689 * Mutex was contended, block until it's
2690 * released and try again
2691 */
Chris Mason56bec292009-03-13 10:10:06 -04002692 mutex_lock(&head->mutex);
2693 mutex_unlock(&head->mutex);
2694
2695 btrfs_put_delayed_ref(ref);
Chris Mason1887be62009-03-13 10:11:24 -04002696 cond_resched();
Chris Mason56bec292009-03-13 10:10:06 -04002697 goto again;
2698 }
2699 node = rb_next(node);
2700 }
2701 spin_unlock(&delayed_refs->lock);
Chris Mason56bec292009-03-13 10:10:06 -04002702 schedule_timeout(1);
2703 goto again;
2704 }
Chris Mason54aa1f42007-06-22 14:16:25 -04002705out:
Chris Masonbb721702013-01-29 18:44:12 -05002706 atomic_dec(&delayed_refs->procs_running_refs);
2707 smp_mb();
2708 if (waitqueue_active(&delayed_refs->wait))
2709 wake_up(&delayed_refs->wait);
2710
Chris Masonc3e69d52009-03-13 10:17:05 -04002711 spin_unlock(&delayed_refs->lock);
Jan Schmidtedf39272012-06-28 18:04:55 +02002712 assert_qgroups_uptodate(trans);
Chris Masona28ec192007-03-06 20:08:01 -05002713 return 0;
2714}
2715
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002716int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2717 struct btrfs_root *root,
2718 u64 bytenr, u64 num_bytes, u64 flags,
Josef Bacikb1c79e02013-05-09 13:49:30 -04002719 int level, int is_data)
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002720{
2721 struct btrfs_delayed_extent_op *extent_op;
2722 int ret;
2723
Miao Xie78a61842012-11-21 02:21:28 +00002724 extent_op = btrfs_alloc_delayed_extent_op();
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002725 if (!extent_op)
2726 return -ENOMEM;
2727
2728 extent_op->flags_to_set = flags;
2729 extent_op->update_flags = 1;
2730 extent_op->update_key = 0;
2731 extent_op->is_data = is_data ? 1 : 0;
Josef Bacikb1c79e02013-05-09 13:49:30 -04002732 extent_op->level = level;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002733
Arne Jansen66d7e7f2011-09-12 15:26:38 +02002734 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2735 num_bytes, extent_op);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002736 if (ret)
Miao Xie78a61842012-11-21 02:21:28 +00002737 btrfs_free_delayed_extent_op(extent_op);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002738 return ret;
2739}
2740
2741static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2742 struct btrfs_root *root,
2743 struct btrfs_path *path,
2744 u64 objectid, u64 offset, u64 bytenr)
2745{
2746 struct btrfs_delayed_ref_head *head;
2747 struct btrfs_delayed_ref_node *ref;
2748 struct btrfs_delayed_data_ref *data_ref;
2749 struct btrfs_delayed_ref_root *delayed_refs;
2750 struct rb_node *node;
2751 int ret = 0;
2752
2753 ret = -ENOENT;
2754 delayed_refs = &trans->transaction->delayed_refs;
2755 spin_lock(&delayed_refs->lock);
2756 head = btrfs_find_delayed_ref_head(trans, bytenr);
2757 if (!head)
2758 goto out;
2759
2760 if (!mutex_trylock(&head->mutex)) {
2761 atomic_inc(&head->node.refs);
2762 spin_unlock(&delayed_refs->lock);
2763
David Sterbab3b4aa72011-04-21 01:20:15 +02002764 btrfs_release_path(path);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002765
David Sterba8cc33e52011-05-02 15:29:25 +02002766 /*
2767 * Mutex was contended, block until it's released and let
2768 * caller try again
2769 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002770 mutex_lock(&head->mutex);
2771 mutex_unlock(&head->mutex);
2772 btrfs_put_delayed_ref(&head->node);
2773 return -EAGAIN;
2774 }
2775
2776 node = rb_prev(&head->node.rb_node);
2777 if (!node)
2778 goto out_unlock;
2779
2780 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2781
2782 if (ref->bytenr != bytenr)
2783 goto out_unlock;
2784
2785 ret = 1;
2786 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2787 goto out_unlock;
2788
2789 data_ref = btrfs_delayed_node_to_data_ref(ref);
2790
2791 node = rb_prev(node);
2792 if (node) {
Liu Bodf57dbe2012-07-23 05:50:03 -06002793 int seq = ref->seq;
2794
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002795 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
Liu Bodf57dbe2012-07-23 05:50:03 -06002796 if (ref->bytenr == bytenr && ref->seq == seq)
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002797 goto out_unlock;
2798 }
2799
2800 if (data_ref->root != root->root_key.objectid ||
2801 data_ref->objectid != objectid || data_ref->offset != offset)
2802 goto out_unlock;
2803
2804 ret = 0;
2805out_unlock:
2806 mutex_unlock(&head->mutex);
2807out:
2808 spin_unlock(&delayed_refs->lock);
2809 return ret;
2810}
2811
2812static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2813 struct btrfs_root *root,
2814 struct btrfs_path *path,
2815 u64 objectid, u64 offset, u64 bytenr)
Chris Masonbe20aa92007-12-17 20:14:01 -05002816{
2817 struct btrfs_root *extent_root = root->fs_info->extent_root;
Yan Zhengf321e492008-07-30 09:26:11 -04002818 struct extent_buffer *leaf;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002819 struct btrfs_extent_data_ref *ref;
2820 struct btrfs_extent_inline_ref *iref;
2821 struct btrfs_extent_item *ei;
Chris Masonbe20aa92007-12-17 20:14:01 -05002822 struct btrfs_key key;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002823 u32 item_size;
Yan Zhengf321e492008-07-30 09:26:11 -04002824 int ret;
Chris Masonbe20aa92007-12-17 20:14:01 -05002825
Chris Masonbe20aa92007-12-17 20:14:01 -05002826 key.objectid = bytenr;
Zheng Yan31840ae2008-09-23 13:14:14 -04002827 key.offset = (u64)-1;
Yan Zhengf321e492008-07-30 09:26:11 -04002828 key.type = BTRFS_EXTENT_ITEM_KEY;
Chris Masonbe20aa92007-12-17 20:14:01 -05002829
Chris Masonbe20aa92007-12-17 20:14:01 -05002830 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2831 if (ret < 0)
2832 goto out;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002833 BUG_ON(ret == 0); /* Corruption */
Yan Zheng80ff3852008-10-30 14:20:02 -04002834
2835 ret = -ENOENT;
2836 if (path->slots[0] == 0)
Zheng Yan31840ae2008-09-23 13:14:14 -04002837 goto out;
Chris Masonbe20aa92007-12-17 20:14:01 -05002838
Zheng Yan31840ae2008-09-23 13:14:14 -04002839 path->slots[0]--;
Yan Zhengf321e492008-07-30 09:26:11 -04002840 leaf = path->nodes[0];
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002841 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
Chris Masonbe20aa92007-12-17 20:14:01 -05002842
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002843 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
Chris Masonbe20aa92007-12-17 20:14:01 -05002844 goto out;
Chris Masonbe20aa92007-12-17 20:14:01 -05002845
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002846 ret = 1;
2847 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2848#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2849 if (item_size < sizeof(*ei)) {
2850 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2851 goto out;
Chris Masonbe20aa92007-12-17 20:14:01 -05002852 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002853#endif
2854 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2855
2856 if (item_size != sizeof(*ei) +
2857 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2858 goto out;
2859
2860 if (btrfs_extent_generation(leaf, ei) <=
2861 btrfs_root_last_snapshot(&root->root_item))
2862 goto out;
2863
2864 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2865 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2866 BTRFS_EXTENT_DATA_REF_KEY)
2867 goto out;
2868
2869 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2870 if (btrfs_extent_refs(leaf, ei) !=
2871 btrfs_extent_data_ref_count(leaf, ref) ||
2872 btrfs_extent_data_ref_root(leaf, ref) !=
2873 root->root_key.objectid ||
2874 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2875 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2876 goto out;
2877
Yan Zhengf321e492008-07-30 09:26:11 -04002878 ret = 0;
Chris Masonbe20aa92007-12-17 20:14:01 -05002879out:
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002880 return ret;
2881}
2882
2883int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2884 struct btrfs_root *root,
2885 u64 objectid, u64 offset, u64 bytenr)
2886{
2887 struct btrfs_path *path;
2888 int ret;
2889 int ret2;
2890
2891 path = btrfs_alloc_path();
2892 if (!path)
2893 return -ENOENT;
2894
2895 do {
2896 ret = check_committed_ref(trans, root, path, objectid,
2897 offset, bytenr);
2898 if (ret && ret != -ENOENT)
2899 goto out;
2900
2901 ret2 = check_delayed_ref(trans, root, path, objectid,
2902 offset, bytenr);
2903 } while (ret2 == -EAGAIN);
2904
2905 if (ret2 && ret2 != -ENOENT) {
2906 ret = ret2;
2907 goto out;
2908 }
2909
2910 if (ret != -ENOENT || ret2 != -ENOENT)
2911 ret = 0;
2912out:
Yan Zhengf321e492008-07-30 09:26:11 -04002913 btrfs_free_path(path);
Yan, Zhengf0486c62010-05-16 10:46:25 -04002914 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2915 WARN_ON(ret > 0);
Yan Zhengf321e492008-07-30 09:26:11 -04002916 return ret;
2917}
2918
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002919static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
Chris Masonb7a9f292009-02-04 09:23:45 -05002920 struct btrfs_root *root,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002921 struct extent_buffer *buf,
Arne Jansen66d7e7f2011-09-12 15:26:38 +02002922 int full_backref, int inc, int for_cow)
Zheng Yan31840ae2008-09-23 13:14:14 -04002923{
2924 u64 bytenr;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002925 u64 num_bytes;
2926 u64 parent;
Zheng Yan31840ae2008-09-23 13:14:14 -04002927 u64 ref_root;
Zheng Yan31840ae2008-09-23 13:14:14 -04002928 u32 nritems;
Zheng Yan31840ae2008-09-23 13:14:14 -04002929 struct btrfs_key key;
2930 struct btrfs_file_extent_item *fi;
2931 int i;
2932 int level;
2933 int ret = 0;
Zheng Yan31840ae2008-09-23 13:14:14 -04002934 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
Arne Jansen66d7e7f2011-09-12 15:26:38 +02002935 u64, u64, u64, u64, u64, u64, int);
Zheng Yan31840ae2008-09-23 13:14:14 -04002936
2937 ref_root = btrfs_header_owner(buf);
Zheng Yan31840ae2008-09-23 13:14:14 -04002938 nritems = btrfs_header_nritems(buf);
2939 level = btrfs_header_level(buf);
2940
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002941 if (!root->ref_cows && level == 0)
2942 return 0;
Chris Masonb7a9f292009-02-04 09:23:45 -05002943
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002944 if (inc)
2945 process_func = btrfs_inc_extent_ref;
2946 else
2947 process_func = btrfs_free_extent;
Zheng Yan31840ae2008-09-23 13:14:14 -04002948
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002949 if (full_backref)
2950 parent = buf->start;
2951 else
2952 parent = 0;
2953
Zheng Yan31840ae2008-09-23 13:14:14 -04002954 for (i = 0; i < nritems; i++) {
Chris Masondb945352007-10-15 16:15:53 -04002955 if (level == 0) {
Chris Mason5f39d392007-10-15 16:14:19 -04002956 btrfs_item_key_to_cpu(buf, &key, i);
2957 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
Chris Mason54aa1f42007-06-22 14:16:25 -04002958 continue;
Chris Mason5f39d392007-10-15 16:14:19 -04002959 fi = btrfs_item_ptr(buf, i,
Chris Mason54aa1f42007-06-22 14:16:25 -04002960 struct btrfs_file_extent_item);
Chris Mason5f39d392007-10-15 16:14:19 -04002961 if (btrfs_file_extent_type(buf, fi) ==
Chris Mason54aa1f42007-06-22 14:16:25 -04002962 BTRFS_FILE_EXTENT_INLINE)
2963 continue;
Zheng Yan31840ae2008-09-23 13:14:14 -04002964 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2965 if (bytenr == 0)
Chris Mason54aa1f42007-06-22 14:16:25 -04002966 continue;
Zheng Yan31840ae2008-09-23 13:14:14 -04002967
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002968 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2969 key.offset -= btrfs_file_extent_offset(buf, fi);
2970 ret = process_func(trans, root, bytenr, num_bytes,
2971 parent, ref_root, key.objectid,
Arne Jansen66d7e7f2011-09-12 15:26:38 +02002972 key.offset, for_cow);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002973 if (ret)
2974 goto fail;
Chris Masonb7a9f292009-02-04 09:23:45 -05002975 } else {
2976 bytenr = btrfs_node_blockptr(buf, i);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002977 num_bytes = btrfs_level_size(root, level - 1);
2978 ret = process_func(trans, root, bytenr, num_bytes,
Arne Jansen66d7e7f2011-09-12 15:26:38 +02002979 parent, ref_root, level - 1, 0,
2980 for_cow);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002981 if (ret)
Zheng Yan31840ae2008-09-23 13:14:14 -04002982 goto fail;
Chris Mason54aa1f42007-06-22 14:16:25 -04002983 }
2984 }
Zheng Yan31840ae2008-09-23 13:14:14 -04002985 return 0;
2986fail:
Chris Mason54aa1f42007-06-22 14:16:25 -04002987 return ret;
Chris Mason02217ed2007-03-02 16:08:05 -05002988}
2989
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002990int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
Arne Jansen66d7e7f2011-09-12 15:26:38 +02002991 struct extent_buffer *buf, int full_backref, int for_cow)
Zheng Yan31840ae2008-09-23 13:14:14 -04002992{
Arne Jansen66d7e7f2011-09-12 15:26:38 +02002993 return __btrfs_mod_ref(trans, root, buf, full_backref, 1, for_cow);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002994}
Zheng Yan31840ae2008-09-23 13:14:14 -04002995
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002996int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
Arne Jansen66d7e7f2011-09-12 15:26:38 +02002997 struct extent_buffer *buf, int full_backref, int for_cow)
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002998{
Arne Jansen66d7e7f2011-09-12 15:26:38 +02002999 return __btrfs_mod_ref(trans, root, buf, full_backref, 0, for_cow);
Zheng Yan31840ae2008-09-23 13:14:14 -04003000}
3001
Chris Mason9078a3e2007-04-26 16:46:15 -04003002static int write_one_cache_group(struct btrfs_trans_handle *trans,
3003 struct btrfs_root *root,
3004 struct btrfs_path *path,
3005 struct btrfs_block_group_cache *cache)
3006{
3007 int ret;
Chris Mason9078a3e2007-04-26 16:46:15 -04003008 struct btrfs_root *extent_root = root->fs_info->extent_root;
Chris Mason5f39d392007-10-15 16:14:19 -04003009 unsigned long bi;
3010 struct extent_buffer *leaf;
Chris Mason9078a3e2007-04-26 16:46:15 -04003011
Chris Mason9078a3e2007-04-26 16:46:15 -04003012 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
Chris Mason54aa1f42007-06-22 14:16:25 -04003013 if (ret < 0)
3014 goto fail;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003015 BUG_ON(ret); /* Corruption */
Chris Mason5f39d392007-10-15 16:14:19 -04003016
3017 leaf = path->nodes[0];
3018 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3019 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3020 btrfs_mark_buffer_dirty(leaf);
David Sterbab3b4aa72011-04-21 01:20:15 +02003021 btrfs_release_path(path);
Chris Mason54aa1f42007-06-22 14:16:25 -04003022fail:
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003023 if (ret) {
3024 btrfs_abort_transaction(trans, root, ret);
Chris Mason9078a3e2007-04-26 16:46:15 -04003025 return ret;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003026 }
Chris Mason9078a3e2007-04-26 16:46:15 -04003027 return 0;
3028
3029}
3030
Yan Zheng4a8c9a62009-07-22 10:07:05 -04003031static struct btrfs_block_group_cache *
3032next_block_group(struct btrfs_root *root,
3033 struct btrfs_block_group_cache *cache)
3034{
3035 struct rb_node *node;
3036 spin_lock(&root->fs_info->block_group_cache_lock);
3037 node = rb_next(&cache->cache_node);
3038 btrfs_put_block_group(cache);
3039 if (node) {
3040 cache = rb_entry(node, struct btrfs_block_group_cache,
3041 cache_node);
Josef Bacik11dfe352009-11-13 20:12:59 +00003042 btrfs_get_block_group(cache);
Yan Zheng4a8c9a62009-07-22 10:07:05 -04003043 } else
3044 cache = NULL;
3045 spin_unlock(&root->fs_info->block_group_cache_lock);
3046 return cache;
3047}
3048
Josef Bacik0af3d002010-06-21 14:48:16 -04003049static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3050 struct btrfs_trans_handle *trans,
3051 struct btrfs_path *path)
3052{
3053 struct btrfs_root *root = block_group->fs_info->tree_root;
3054 struct inode *inode = NULL;
3055 u64 alloc_hint = 0;
Josef Bacik2b209822010-12-03 13:17:53 -05003056 int dcs = BTRFS_DC_ERROR;
Josef Bacik0af3d002010-06-21 14:48:16 -04003057 int num_pages = 0;
3058 int retries = 0;
3059 int ret = 0;
3060
3061 /*
3062 * If this block group is smaller than 100 megs don't bother caching the
3063 * block group.
3064 */
3065 if (block_group->key.offset < (100 * 1024 * 1024)) {
3066 spin_lock(&block_group->lock);
3067 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3068 spin_unlock(&block_group->lock);
3069 return 0;
3070 }
3071
3072again:
3073 inode = lookup_free_space_inode(root, block_group, path);
3074 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3075 ret = PTR_ERR(inode);
David Sterbab3b4aa72011-04-21 01:20:15 +02003076 btrfs_release_path(path);
Josef Bacik0af3d002010-06-21 14:48:16 -04003077 goto out;
3078 }
3079
3080 if (IS_ERR(inode)) {
3081 BUG_ON(retries);
3082 retries++;
3083
3084 if (block_group->ro)
3085 goto out_free;
3086
3087 ret = create_free_space_inode(root, trans, block_group, path);
3088 if (ret)
3089 goto out_free;
3090 goto again;
3091 }
3092
Josef Bacik5b0e95b2011-10-06 08:58:24 -04003093 /* We've already setup this transaction, go ahead and exit */
3094 if (block_group->cache_generation == trans->transid &&
3095 i_size_read(inode)) {
3096 dcs = BTRFS_DC_SETUP;
3097 goto out_put;
3098 }
3099
Josef Bacik0af3d002010-06-21 14:48:16 -04003100 /*
3101 * We want to set the generation to 0, that way if anything goes wrong
3102 * from here on out we know not to trust this cache when we load up next
3103 * time.
3104 */
3105 BTRFS_I(inode)->generation = 0;
3106 ret = btrfs_update_inode(trans, root, inode);
3107 WARN_ON(ret);
3108
3109 if (i_size_read(inode) > 0) {
Miao Xie7b61cd92013-05-13 13:55:09 +00003110 ret = btrfs_check_trunc_cache_free_space(root,
3111 &root->fs_info->global_block_rsv);
3112 if (ret)
3113 goto out_put;
3114
Josef Bacik0af3d002010-06-21 14:48:16 -04003115 ret = btrfs_truncate_free_space_cache(root, trans, path,
3116 inode);
3117 if (ret)
3118 goto out_put;
3119 }
3120
3121 spin_lock(&block_group->lock);
Liu Bocf7c1ef2012-07-06 03:31:34 -06003122 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3123 !btrfs_test_opt(root, SPACE_CACHE)) {
3124 /*
3125 * don't bother trying to write stuff out _if_
3126 * a) we're not cached,
3127 * b) we're with nospace_cache mount option.
3128 */
Josef Bacik2b209822010-12-03 13:17:53 -05003129 dcs = BTRFS_DC_WRITTEN;
Josef Bacik0af3d002010-06-21 14:48:16 -04003130 spin_unlock(&block_group->lock);
3131 goto out_put;
3132 }
3133 spin_unlock(&block_group->lock);
3134
Josef Bacik6fc823b2012-08-06 13:46:38 -06003135 /*
3136 * Try to preallocate enough space based on how big the block group is.
3137 * Keep in mind this has to include any pinned space which could end up
3138 * taking up quite a bit since it's not folded into the other space
3139 * cache.
3140 */
3141 num_pages = (int)div64_u64(block_group->key.offset, 256 * 1024 * 1024);
Josef Bacik0af3d002010-06-21 14:48:16 -04003142 if (!num_pages)
3143 num_pages = 1;
3144
Josef Bacik0af3d002010-06-21 14:48:16 -04003145 num_pages *= 16;
3146 num_pages *= PAGE_CACHE_SIZE;
3147
3148 ret = btrfs_check_data_free_space(inode, num_pages);
3149 if (ret)
3150 goto out_put;
3151
3152 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3153 num_pages, num_pages,
3154 &alloc_hint);
Josef Bacik2b209822010-12-03 13:17:53 -05003155 if (!ret)
3156 dcs = BTRFS_DC_SETUP;
Josef Bacik0af3d002010-06-21 14:48:16 -04003157 btrfs_free_reserved_data_space(inode, num_pages);
Josef Bacikc09544e2011-08-30 10:19:10 -04003158
Josef Bacik0af3d002010-06-21 14:48:16 -04003159out_put:
3160 iput(inode);
3161out_free:
David Sterbab3b4aa72011-04-21 01:20:15 +02003162 btrfs_release_path(path);
Josef Bacik0af3d002010-06-21 14:48:16 -04003163out:
3164 spin_lock(&block_group->lock);
Josef Bacike65cbb92011-12-13 16:04:54 -05003165 if (!ret && dcs == BTRFS_DC_SETUP)
Josef Bacik5b0e95b2011-10-06 08:58:24 -04003166 block_group->cache_generation = trans->transid;
Josef Bacik2b209822010-12-03 13:17:53 -05003167 block_group->disk_cache_state = dcs;
Josef Bacik0af3d002010-06-21 14:48:16 -04003168 spin_unlock(&block_group->lock);
3169
3170 return ret;
3171}
3172
Chris Mason96b51792007-10-15 16:15:19 -04003173int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3174 struct btrfs_root *root)
Chris Mason9078a3e2007-04-26 16:46:15 -04003175{
Yan Zheng4a8c9a62009-07-22 10:07:05 -04003176 struct btrfs_block_group_cache *cache;
Chris Mason9078a3e2007-04-26 16:46:15 -04003177 int err = 0;
Chris Mason9078a3e2007-04-26 16:46:15 -04003178 struct btrfs_path *path;
Chris Mason96b51792007-10-15 16:15:19 -04003179 u64 last = 0;
Chris Mason9078a3e2007-04-26 16:46:15 -04003180
3181 path = btrfs_alloc_path();
3182 if (!path)
3183 return -ENOMEM;
3184
Josef Bacik0af3d002010-06-21 14:48:16 -04003185again:
3186 while (1) {
3187 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3188 while (cache) {
3189 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3190 break;
3191 cache = next_block_group(root, cache);
3192 }
3193 if (!cache) {
3194 if (last == 0)
3195 break;
3196 last = 0;
3197 continue;
3198 }
3199 err = cache_save_setup(cache, trans, path);
3200 last = cache->key.objectid + cache->key.offset;
3201 btrfs_put_block_group(cache);
3202 }
3203
Chris Masond3977122009-01-05 21:25:51 -05003204 while (1) {
Yan Zheng4a8c9a62009-07-22 10:07:05 -04003205 if (last == 0) {
3206 err = btrfs_run_delayed_refs(trans, root,
3207 (unsigned long)-1);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003208 if (err) /* File system offline */
3209 goto out;
Josef Bacik0f9dd462008-09-23 13:14:11 -04003210 }
Josef Bacik0f9dd462008-09-23 13:14:11 -04003211
Yan Zheng4a8c9a62009-07-22 10:07:05 -04003212 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3213 while (cache) {
Josef Bacik0af3d002010-06-21 14:48:16 -04003214 if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
3215 btrfs_put_block_group(cache);
3216 goto again;
3217 }
3218
Yan Zheng4a8c9a62009-07-22 10:07:05 -04003219 if (cache->dirty)
3220 break;
3221 cache = next_block_group(root, cache);
3222 }
3223 if (!cache) {
3224 if (last == 0)
3225 break;
3226 last = 0;
Chris Mason96b51792007-10-15 16:15:19 -04003227 continue;
Chris Mason9078a3e2007-04-26 16:46:15 -04003228 }
Yan Zheng4a8c9a62009-07-22 10:07:05 -04003229
Josef Bacik0cb59c92010-07-02 12:14:14 -04003230 if (cache->disk_cache_state == BTRFS_DC_SETUP)
3231 cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
Yan Zheng4a8c9a62009-07-22 10:07:05 -04003232 cache->dirty = 0;
3233 last = cache->key.objectid + cache->key.offset;
3234
3235 err = write_one_cache_group(trans, root, path, cache);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003236 if (err) /* File system offline */
3237 goto out;
3238
Yan Zheng4a8c9a62009-07-22 10:07:05 -04003239 btrfs_put_block_group(cache);
Chris Mason9078a3e2007-04-26 16:46:15 -04003240 }
Yan Zheng4a8c9a62009-07-22 10:07:05 -04003241
Josef Bacik0cb59c92010-07-02 12:14:14 -04003242 while (1) {
3243 /*
3244 * I don't think this is needed since we're just marking our
3245 * preallocated extent as written, but just in case it can't
3246 * hurt.
3247 */
3248 if (last == 0) {
3249 err = btrfs_run_delayed_refs(trans, root,
3250 (unsigned long)-1);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003251 if (err) /* File system offline */
3252 goto out;
Josef Bacik0cb59c92010-07-02 12:14:14 -04003253 }
3254
3255 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3256 while (cache) {
3257 /*
3258 * Really this shouldn't happen, but it could if we
3259 * couldn't write the entire preallocated extent and
3260 * splitting the extent resulted in a new block.
3261 */
3262 if (cache->dirty) {
3263 btrfs_put_block_group(cache);
3264 goto again;
3265 }
3266 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3267 break;
3268 cache = next_block_group(root, cache);
3269 }
3270 if (!cache) {
3271 if (last == 0)
3272 break;
3273 last = 0;
3274 continue;
3275 }
3276
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003277 err = btrfs_write_out_cache(root, trans, cache, path);
Josef Bacik0cb59c92010-07-02 12:14:14 -04003278
3279 /*
3280 * If we didn't have an error then the cache state is still
3281 * NEED_WRITE, so we can set it to WRITTEN.
3282 */
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003283 if (!err && cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
Josef Bacik0cb59c92010-07-02 12:14:14 -04003284 cache->disk_cache_state = BTRFS_DC_WRITTEN;
3285 last = cache->key.objectid + cache->key.offset;
3286 btrfs_put_block_group(cache);
3287 }
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003288out:
Josef Bacik0cb59c92010-07-02 12:14:14 -04003289
Chris Mason9078a3e2007-04-26 16:46:15 -04003290 btrfs_free_path(path);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003291 return err;
Chris Mason9078a3e2007-04-26 16:46:15 -04003292}
3293
Yan Zhengd2fb3432008-12-11 16:30:39 -05003294int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3295{
3296 struct btrfs_block_group_cache *block_group;
3297 int readonly = 0;
3298
3299 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3300 if (!block_group || block_group->ro)
3301 readonly = 1;
3302 if (block_group)
Chris Masonfa9c0d792009-04-03 09:47:43 -04003303 btrfs_put_block_group(block_group);
Yan Zhengd2fb3432008-12-11 16:30:39 -05003304 return readonly;
3305}
3306
Chris Mason593060d2008-03-25 16:50:33 -04003307static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3308 u64 total_bytes, u64 bytes_used,
3309 struct btrfs_space_info **space_info)
3310{
3311 struct btrfs_space_info *found;
Yan, Zhengb742bb82010-05-16 10:46:24 -04003312 int i;
3313 int factor;
3314
3315 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3316 BTRFS_BLOCK_GROUP_RAID10))
3317 factor = 2;
3318 else
3319 factor = 1;
Chris Mason593060d2008-03-25 16:50:33 -04003320
3321 found = __find_space_info(info, flags);
3322 if (found) {
Josef Bacik25179202008-10-29 14:49:05 -04003323 spin_lock(&found->lock);
Chris Mason593060d2008-03-25 16:50:33 -04003324 found->total_bytes += total_bytes;
Josef Bacik89a55892010-10-14 14:52:27 -04003325 found->disk_total += total_bytes * factor;
Chris Mason593060d2008-03-25 16:50:33 -04003326 found->bytes_used += bytes_used;
Yan, Zhengb742bb82010-05-16 10:46:24 -04003327 found->disk_used += bytes_used * factor;
Chris Mason8f18cf12008-04-25 16:53:30 -04003328 found->full = 0;
Josef Bacik25179202008-10-29 14:49:05 -04003329 spin_unlock(&found->lock);
Chris Mason593060d2008-03-25 16:50:33 -04003330 *space_info = found;
3331 return 0;
3332 }
Yan Zhengc146afa2008-11-12 14:34:12 -05003333 found = kzalloc(sizeof(*found), GFP_NOFS);
Chris Mason593060d2008-03-25 16:50:33 -04003334 if (!found)
3335 return -ENOMEM;
3336
Yan, Zhengb742bb82010-05-16 10:46:24 -04003337 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3338 INIT_LIST_HEAD(&found->block_groups[i]);
Josef Bacik80eb2342008-10-29 14:49:05 -04003339 init_rwsem(&found->groups_sem);
Josef Bacik0f9dd462008-09-23 13:14:11 -04003340 spin_lock_init(&found->lock);
Ilya Dryomov52ba6922012-01-16 22:04:47 +02003341 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
Chris Mason593060d2008-03-25 16:50:33 -04003342 found->total_bytes = total_bytes;
Josef Bacik89a55892010-10-14 14:52:27 -04003343 found->disk_total = total_bytes * factor;
Chris Mason593060d2008-03-25 16:50:33 -04003344 found->bytes_used = bytes_used;
Yan, Zhengb742bb82010-05-16 10:46:24 -04003345 found->disk_used = bytes_used * factor;
Chris Mason593060d2008-03-25 16:50:33 -04003346 found->bytes_pinned = 0;
Zheng Yane8569812008-09-26 10:05:48 -04003347 found->bytes_reserved = 0;
Yan Zhengc146afa2008-11-12 14:34:12 -05003348 found->bytes_readonly = 0;
Yan, Zhengf0486c62010-05-16 10:46:25 -04003349 found->bytes_may_use = 0;
Chris Mason593060d2008-03-25 16:50:33 -04003350 found->full = 0;
Chris Mason0e4f8f82011-04-15 16:05:44 -04003351 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
Josef Bacik6d741192011-04-11 20:20:11 -04003352 found->chunk_alloc = 0;
Josef Bacikfdb5eff2011-06-07 16:07:44 -04003353 found->flush = 0;
3354 init_waitqueue_head(&found->wait);
Chris Mason593060d2008-03-25 16:50:33 -04003355 *space_info = found;
Chris Mason4184ea72009-03-10 12:39:20 -04003356 list_add_rcu(&found->list, &info->space_info);
Li Zefanb4d7c3c2012-07-09 20:21:07 -06003357 if (flags & BTRFS_BLOCK_GROUP_DATA)
3358 info->data_sinfo = found;
Chris Mason593060d2008-03-25 16:50:33 -04003359 return 0;
3360}
3361
Chris Mason8790d502008-04-03 16:29:03 -04003362static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3363{
Ilya Dryomov899c81e2012-03-27 17:09:16 +03003364 u64 extra_flags = chunk_to_extended(flags) &
3365 BTRFS_EXTENDED_PROFILE_MASK;
Ilya Dryomova46d11a2012-01-16 22:04:47 +02003366
Miao Xiede98ced2013-01-29 10:13:12 +00003367 write_seqlock(&fs_info->profiles_lock);
Ilya Dryomova46d11a2012-01-16 22:04:47 +02003368 if (flags & BTRFS_BLOCK_GROUP_DATA)
3369 fs_info->avail_data_alloc_bits |= extra_flags;
3370 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3371 fs_info->avail_metadata_alloc_bits |= extra_flags;
3372 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3373 fs_info->avail_system_alloc_bits |= extra_flags;
Miao Xiede98ced2013-01-29 10:13:12 +00003374 write_sequnlock(&fs_info->profiles_lock);
Chris Mason8790d502008-04-03 16:29:03 -04003375}
Chris Mason593060d2008-03-25 16:50:33 -04003376
Ilya Dryomova46d11a2012-01-16 22:04:47 +02003377/*
Ilya Dryomovfc67c452012-03-27 17:09:17 +03003378 * returns target flags in extended format or 0 if restripe for this
3379 * chunk_type is not in progress
Ilya Dryomovc6664b42012-04-12 16:03:56 -04003380 *
3381 * should be called with either volume_mutex or balance_lock held
Ilya Dryomovfc67c452012-03-27 17:09:17 +03003382 */
3383static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3384{
3385 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3386 u64 target = 0;
3387
Ilya Dryomovfc67c452012-03-27 17:09:17 +03003388 if (!bctl)
3389 return 0;
3390
3391 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3392 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3393 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3394 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3395 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3396 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3397 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3398 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3399 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3400 }
3401
3402 return target;
3403}
3404
3405/*
Ilya Dryomova46d11a2012-01-16 22:04:47 +02003406 * @flags: available profiles in extended format (see ctree.h)
3407 *
Ilya Dryomove4d8ec02012-01-16 22:04:48 +02003408 * Returns reduced profile in chunk format. If profile changing is in
3409 * progress (either running or paused) picks the target profile (if it's
3410 * already available), otherwise falls back to plain reducing.
Ilya Dryomova46d11a2012-01-16 22:04:47 +02003411 */
Eric Sandeen48a3b632013-04-25 20:41:01 +00003412static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
Chris Masonec44a352008-04-28 15:29:52 -04003413{
Chris Masoncd02dca2010-12-13 14:56:23 -05003414 /*
3415 * we add in the count of missing devices because we want
3416 * to make sure that any RAID levels on a degraded FS
3417 * continue to be honored.
3418 */
3419 u64 num_devices = root->fs_info->fs_devices->rw_devices +
3420 root->fs_info->fs_devices->missing_devices;
Ilya Dryomovfc67c452012-03-27 17:09:17 +03003421 u64 target;
David Woodhouse53b381b2013-01-29 18:40:14 -05003422 u64 tmp;
Chris Masona061fc82008-05-07 11:43:44 -04003423
Ilya Dryomovfc67c452012-03-27 17:09:17 +03003424 /*
3425 * see if restripe for this chunk_type is in progress, if so
3426 * try to reduce to the target profile
3427 */
Ilya Dryomove4d8ec02012-01-16 22:04:48 +02003428 spin_lock(&root->fs_info->balance_lock);
Ilya Dryomovfc67c452012-03-27 17:09:17 +03003429 target = get_restripe_target(root->fs_info, flags);
3430 if (target) {
3431 /* pick target profile only if it's already available */
3432 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
Ilya Dryomove4d8ec02012-01-16 22:04:48 +02003433 spin_unlock(&root->fs_info->balance_lock);
Ilya Dryomovfc67c452012-03-27 17:09:17 +03003434 return extended_to_chunk(target);
Ilya Dryomove4d8ec02012-01-16 22:04:48 +02003435 }
3436 }
3437 spin_unlock(&root->fs_info->balance_lock);
3438
David Woodhouse53b381b2013-01-29 18:40:14 -05003439 /* First, mask out the RAID levels which aren't possible */
Chris Masona061fc82008-05-07 11:43:44 -04003440 if (num_devices == 1)
David Woodhouse53b381b2013-01-29 18:40:14 -05003441 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0 |
3442 BTRFS_BLOCK_GROUP_RAID5);
3443 if (num_devices < 3)
3444 flags &= ~BTRFS_BLOCK_GROUP_RAID6;
Chris Masona061fc82008-05-07 11:43:44 -04003445 if (num_devices < 4)
3446 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3447
David Woodhouse53b381b2013-01-29 18:40:14 -05003448 tmp = flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3449 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID5 |
3450 BTRFS_BLOCK_GROUP_RAID6 | BTRFS_BLOCK_GROUP_RAID10);
3451 flags &= ~tmp;
Chris Masonec44a352008-04-28 15:29:52 -04003452
David Woodhouse53b381b2013-01-29 18:40:14 -05003453 if (tmp & BTRFS_BLOCK_GROUP_RAID6)
3454 tmp = BTRFS_BLOCK_GROUP_RAID6;
3455 else if (tmp & BTRFS_BLOCK_GROUP_RAID5)
3456 tmp = BTRFS_BLOCK_GROUP_RAID5;
3457 else if (tmp & BTRFS_BLOCK_GROUP_RAID10)
3458 tmp = BTRFS_BLOCK_GROUP_RAID10;
3459 else if (tmp & BTRFS_BLOCK_GROUP_RAID1)
3460 tmp = BTRFS_BLOCK_GROUP_RAID1;
3461 else if (tmp & BTRFS_BLOCK_GROUP_RAID0)
3462 tmp = BTRFS_BLOCK_GROUP_RAID0;
Chris Masonec44a352008-04-28 15:29:52 -04003463
David Woodhouse53b381b2013-01-29 18:40:14 -05003464 return extended_to_chunk(flags | tmp);
Chris Masonec44a352008-04-28 15:29:52 -04003465}
3466
Yan, Zhengb742bb82010-05-16 10:46:24 -04003467static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
Josef Bacik6a632092009-02-20 11:00:09 -05003468{
Miao Xiede98ced2013-01-29 10:13:12 +00003469 unsigned seq;
3470
3471 do {
3472 seq = read_seqbegin(&root->fs_info->profiles_lock);
3473
3474 if (flags & BTRFS_BLOCK_GROUP_DATA)
3475 flags |= root->fs_info->avail_data_alloc_bits;
3476 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3477 flags |= root->fs_info->avail_system_alloc_bits;
3478 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3479 flags |= root->fs_info->avail_metadata_alloc_bits;
3480 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
Ilya Dryomov6fef8df2012-01-16 22:04:47 +02003481
Yan, Zhengb742bb82010-05-16 10:46:24 -04003482 return btrfs_reduce_alloc_profile(root, flags);
3483}
Josef Bacik6a632092009-02-20 11:00:09 -05003484
Miao Xie6d07bce2011-01-05 10:07:31 +00003485u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
Yan, Zhengb742bb82010-05-16 10:46:24 -04003486{
3487 u64 flags;
David Woodhouse53b381b2013-01-29 18:40:14 -05003488 u64 ret;
Josef Bacik6a632092009-02-20 11:00:09 -05003489
Yan, Zhengb742bb82010-05-16 10:46:24 -04003490 if (data)
3491 flags = BTRFS_BLOCK_GROUP_DATA;
3492 else if (root == root->fs_info->chunk_root)
3493 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3494 else
3495 flags = BTRFS_BLOCK_GROUP_METADATA;
3496
David Woodhouse53b381b2013-01-29 18:40:14 -05003497 ret = get_alloc_profile(root, flags);
3498 return ret;
Josef Bacik6a632092009-02-20 11:00:09 -05003499}
3500
Josef Bacik6a632092009-02-20 11:00:09 -05003501/*
3502 * This will check the space that the inode allocates from to make sure we have
3503 * enough space for bytes.
3504 */
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04003505int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
Josef Bacik6a632092009-02-20 11:00:09 -05003506{
3507 struct btrfs_space_info *data_sinfo;
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04003508 struct btrfs_root *root = BTRFS_I(inode)->root;
Li Zefanb4d7c3c2012-07-09 20:21:07 -06003509 struct btrfs_fs_info *fs_info = root->fs_info;
Josef Bacikab6e24102010-03-19 14:38:13 +00003510 u64 used;
Josef Bacik0af3d002010-06-21 14:48:16 -04003511 int ret = 0, committed = 0, alloc_chunk = 1;
Josef Bacik6a632092009-02-20 11:00:09 -05003512
3513 /* make sure bytes are sectorsize aligned */
Qu Wenruofda28322013-02-26 08:10:22 +00003514 bytes = ALIGN(bytes, root->sectorsize);
Josef Bacik6a632092009-02-20 11:00:09 -05003515
Li Zefan82d59022011-04-20 10:33:24 +08003516 if (root == root->fs_info->tree_root ||
3517 BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) {
Josef Bacik0af3d002010-06-21 14:48:16 -04003518 alloc_chunk = 0;
3519 committed = 1;
3520 }
3521
Li Zefanb4d7c3c2012-07-09 20:21:07 -06003522 data_sinfo = fs_info->data_sinfo;
Chris Mason33b4d472009-09-22 14:45:50 -04003523 if (!data_sinfo)
3524 goto alloc;
3525
Josef Bacik6a632092009-02-20 11:00:09 -05003526again:
3527 /* make sure we have enough space to handle the data first */
3528 spin_lock(&data_sinfo->lock);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04003529 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3530 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3531 data_sinfo->bytes_may_use;
Josef Bacikab6e24102010-03-19 14:38:13 +00003532
3533 if (used + bytes > data_sinfo->total_bytes) {
Josef Bacik4e06bdd2009-02-20 10:59:53 -05003534 struct btrfs_trans_handle *trans;
3535
Josef Bacik6a632092009-02-20 11:00:09 -05003536 /*
3537 * if we don't have enough free bytes in this space then we need
3538 * to alloc a new chunk.
3539 */
Josef Bacik0af3d002010-06-21 14:48:16 -04003540 if (!data_sinfo->full && alloc_chunk) {
Josef Bacik6a632092009-02-20 11:00:09 -05003541 u64 alloc_target;
Josef Bacik6a632092009-02-20 11:00:09 -05003542
Chris Mason0e4f8f82011-04-15 16:05:44 -04003543 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
Josef Bacik6a632092009-02-20 11:00:09 -05003544 spin_unlock(&data_sinfo->lock);
Chris Mason33b4d472009-09-22 14:45:50 -04003545alloc:
Josef Bacik6a632092009-02-20 11:00:09 -05003546 alloc_target = btrfs_get_alloc_profile(root, 1);
Josef Bacik7a7eaa42011-04-13 12:54:33 -04003547 trans = btrfs_join_transaction(root);
Yan, Zhenga22285a2010-05-16 10:48:46 -04003548 if (IS_ERR(trans))
3549 return PTR_ERR(trans);
Josef Bacik6a632092009-02-20 11:00:09 -05003550
3551 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
Chris Mason0e4f8f82011-04-15 16:05:44 -04003552 alloc_target,
3553 CHUNK_ALLOC_NO_FORCE);
Josef Bacik6a632092009-02-20 11:00:09 -05003554 btrfs_end_transaction(trans, root);
Miao Xied52a5b52011-01-05 10:07:18 +00003555 if (ret < 0) {
3556 if (ret != -ENOSPC)
3557 return ret;
3558 else
3559 goto commit_trans;
3560 }
Chris Mason33b4d472009-09-22 14:45:50 -04003561
Li Zefanb4d7c3c2012-07-09 20:21:07 -06003562 if (!data_sinfo)
3563 data_sinfo = fs_info->data_sinfo;
3564
Josef Bacik6a632092009-02-20 11:00:09 -05003565 goto again;
3566 }
Josef Bacikf2bb8f52011-05-25 13:10:16 -04003567
3568 /*
3569 * If we have less pinned bytes than we want to allocate then
3570 * don't bother committing the transaction, it won't help us.
3571 */
3572 if (data_sinfo->bytes_pinned < bytes)
3573 committed = 1;
Josef Bacik6a632092009-02-20 11:00:09 -05003574 spin_unlock(&data_sinfo->lock);
Josef Bacik4e06bdd2009-02-20 10:59:53 -05003575
3576 /* commit the current transaction and try again */
Miao Xied52a5b52011-01-05 10:07:18 +00003577commit_trans:
Josef Bacika4abeea2011-04-11 17:25:13 -04003578 if (!committed &&
3579 !atomic_read(&root->fs_info->open_ioctl_trans)) {
Josef Bacik4e06bdd2009-02-20 10:59:53 -05003580 committed = 1;
Josef Bacik7a7eaa42011-04-13 12:54:33 -04003581 trans = btrfs_join_transaction(root);
Yan, Zhenga22285a2010-05-16 10:48:46 -04003582 if (IS_ERR(trans))
3583 return PTR_ERR(trans);
Josef Bacik4e06bdd2009-02-20 10:59:53 -05003584 ret = btrfs_commit_transaction(trans, root);
3585 if (ret)
3586 return ret;
3587 goto again;
3588 }
3589
Josef Bacik6a632092009-02-20 11:00:09 -05003590 return -ENOSPC;
3591 }
3592 data_sinfo->bytes_may_use += bytes;
Josef Bacik8c2a3ca2012-01-10 10:31:31 -05003593 trace_btrfs_space_reservation(root->fs_info, "space_info",
Liu Bo2bcc0322012-03-29 09:57:44 -04003594 data_sinfo->flags, bytes, 1);
Josef Bacik6a632092009-02-20 11:00:09 -05003595 spin_unlock(&data_sinfo->lock);
3596
Josef Bacik9ed74f22009-09-11 16:12:44 -04003597 return 0;
Josef Bacik6a632092009-02-20 11:00:09 -05003598}
3599
3600/*
Josef Bacikfb25e912011-07-26 17:00:46 -04003601 * Called if we need to clear a data reservation for this inode.
Josef Bacik6a632092009-02-20 11:00:09 -05003602 */
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04003603void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
Josef Bacik6a632092009-02-20 11:00:09 -05003604{
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04003605 struct btrfs_root *root = BTRFS_I(inode)->root;
Josef Bacik6a632092009-02-20 11:00:09 -05003606 struct btrfs_space_info *data_sinfo;
3607
3608 /* make sure bytes are sectorsize aligned */
Qu Wenruofda28322013-02-26 08:10:22 +00003609 bytes = ALIGN(bytes, root->sectorsize);
Josef Bacik6a632092009-02-20 11:00:09 -05003610
Li Zefanb4d7c3c2012-07-09 20:21:07 -06003611 data_sinfo = root->fs_info->data_sinfo;
Josef Bacik6a632092009-02-20 11:00:09 -05003612 spin_lock(&data_sinfo->lock);
3613 data_sinfo->bytes_may_use -= bytes;
Josef Bacik8c2a3ca2012-01-10 10:31:31 -05003614 trace_btrfs_space_reservation(root->fs_info, "space_info",
Liu Bo2bcc0322012-03-29 09:57:44 -04003615 data_sinfo->flags, bytes, 0);
Josef Bacik6a632092009-02-20 11:00:09 -05003616 spin_unlock(&data_sinfo->lock);
3617}
3618
Josef Bacik97e728d2009-04-21 17:40:57 -04003619static void force_metadata_allocation(struct btrfs_fs_info *info)
3620{
3621 struct list_head *head = &info->space_info;
3622 struct btrfs_space_info *found;
3623
3624 rcu_read_lock();
3625 list_for_each_entry_rcu(found, head, list) {
3626 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
Chris Mason0e4f8f82011-04-15 16:05:44 -04003627 found->force_alloc = CHUNK_ALLOC_FORCE;
Josef Bacik97e728d2009-04-21 17:40:57 -04003628 }
3629 rcu_read_unlock();
3630}
3631
Miao Xie3c76cd82013-04-25 10:12:38 +00003632static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
3633{
3634 return (global->size << 1);
3635}
3636
Chris Masone5bc2452010-10-26 13:37:56 -04003637static int should_alloc_chunk(struct btrfs_root *root,
Josef Bacik698d0082012-09-12 14:08:47 -04003638 struct btrfs_space_info *sinfo, int force)
Yan, Zheng424499d2010-05-16 10:46:25 -04003639{
Josef Bacikfb25e912011-07-26 17:00:46 -04003640 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
Yan, Zheng424499d2010-05-16 10:46:25 -04003641 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
Chris Mason0e4f8f82011-04-15 16:05:44 -04003642 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
Chris Masone5bc2452010-10-26 13:37:56 -04003643 u64 thresh;
Yan, Zheng424499d2010-05-16 10:46:25 -04003644
Chris Mason0e4f8f82011-04-15 16:05:44 -04003645 if (force == CHUNK_ALLOC_FORCE)
3646 return 1;
3647
3648 /*
Josef Bacikfb25e912011-07-26 17:00:46 -04003649 * We need to take into account the global rsv because for all intents
3650 * and purposes it's used space. Don't worry about locking the
3651 * global_rsv, it doesn't change except when the transaction commits.
3652 */
Josef Bacik54338b52012-08-14 16:20:52 -04003653 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
Miao Xie3c76cd82013-04-25 10:12:38 +00003654 num_allocated += calc_global_rsv_need_space(global_rsv);
Josef Bacikfb25e912011-07-26 17:00:46 -04003655
3656 /*
Chris Mason0e4f8f82011-04-15 16:05:44 -04003657 * in limited mode, we want to have some free space up to
3658 * about 1% of the FS size.
3659 */
3660 if (force == CHUNK_ALLOC_LIMITED) {
David Sterba6c417612011-04-13 15:41:04 +02003661 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
Chris Mason0e4f8f82011-04-15 16:05:44 -04003662 thresh = max_t(u64, 64 * 1024 * 1024,
3663 div_factor_fine(thresh, 1));
3664
3665 if (num_bytes - num_allocated < thresh)
3666 return 1;
3667 }
Chris Mason0e4f8f82011-04-15 16:05:44 -04003668
Josef Bacik698d0082012-09-12 14:08:47 -04003669 if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8))
Josef Bacik14ed0ca2010-10-15 15:23:48 -04003670 return 0;
Yan, Zheng424499d2010-05-16 10:46:25 -04003671 return 1;
3672}
3673
Liu Bo15d1ff82012-03-29 09:57:44 -04003674static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type)
3675{
3676 u64 num_dev;
3677
David Woodhouse53b381b2013-01-29 18:40:14 -05003678 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
3679 BTRFS_BLOCK_GROUP_RAID0 |
3680 BTRFS_BLOCK_GROUP_RAID5 |
3681 BTRFS_BLOCK_GROUP_RAID6))
Liu Bo15d1ff82012-03-29 09:57:44 -04003682 num_dev = root->fs_info->fs_devices->rw_devices;
3683 else if (type & BTRFS_BLOCK_GROUP_RAID1)
3684 num_dev = 2;
3685 else
3686 num_dev = 1; /* DUP or single */
3687
3688 /* metadata for updaing devices and chunk tree */
3689 return btrfs_calc_trans_metadata_size(root, num_dev + 1);
3690}
3691
3692static void check_system_chunk(struct btrfs_trans_handle *trans,
3693 struct btrfs_root *root, u64 type)
3694{
3695 struct btrfs_space_info *info;
3696 u64 left;
3697 u64 thresh;
3698
3699 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3700 spin_lock(&info->lock);
3701 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
3702 info->bytes_reserved - info->bytes_readonly;
3703 spin_unlock(&info->lock);
3704
3705 thresh = get_system_chunk_thresh(root, type);
3706 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
Simon Kirbyc2cf52e2013-03-19 22:41:23 +00003707 btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu",
3708 left, thresh, type);
Liu Bo15d1ff82012-03-29 09:57:44 -04003709 dump_space_info(info, 0, 0);
3710 }
3711
3712 if (left < thresh) {
3713 u64 flags;
3714
3715 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
3716 btrfs_alloc_chunk(trans, root, flags);
3717 }
3718}
3719
Chris Mason6324fbf2008-03-24 15:01:59 -04003720static int do_chunk_alloc(struct btrfs_trans_handle *trans,
Josef Bacik698d0082012-09-12 14:08:47 -04003721 struct btrfs_root *extent_root, u64 flags, int force)
Chris Mason6324fbf2008-03-24 15:01:59 -04003722{
3723 struct btrfs_space_info *space_info;
Josef Bacik97e728d2009-04-21 17:40:57 -04003724 struct btrfs_fs_info *fs_info = extent_root->fs_info;
Josef Bacik6d741192011-04-11 20:20:11 -04003725 int wait_for_alloc = 0;
Yan Zhengc146afa2008-11-12 14:34:12 -05003726 int ret = 0;
3727
Josef Bacikc6b305a2012-12-18 09:16:16 -05003728 /* Don't re-enter if we're already allocating a chunk */
3729 if (trans->allocating_chunk)
3730 return -ENOSPC;
3731
Chris Mason6324fbf2008-03-24 15:01:59 -04003732 space_info = __find_space_info(extent_root->fs_info, flags);
Chris Mason593060d2008-03-25 16:50:33 -04003733 if (!space_info) {
3734 ret = update_space_info(extent_root->fs_info, flags,
3735 0, 0, &space_info);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003736 BUG_ON(ret); /* -ENOMEM */
Chris Mason593060d2008-03-25 16:50:33 -04003737 }
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003738 BUG_ON(!space_info); /* Logic error */
Chris Mason6324fbf2008-03-24 15:01:59 -04003739
Josef Bacik6d741192011-04-11 20:20:11 -04003740again:
Josef Bacik25179202008-10-29 14:49:05 -04003741 spin_lock(&space_info->lock);
Miao Xie9e622d62012-01-26 15:01:12 -05003742 if (force < space_info->force_alloc)
Chris Mason0e4f8f82011-04-15 16:05:44 -04003743 force = space_info->force_alloc;
Josef Bacik25179202008-10-29 14:49:05 -04003744 if (space_info->full) {
3745 spin_unlock(&space_info->lock);
Josef Bacik6d741192011-04-11 20:20:11 -04003746 return 0;
Josef Bacik25179202008-10-29 14:49:05 -04003747 }
Chris Mason6324fbf2008-03-24 15:01:59 -04003748
Josef Bacik698d0082012-09-12 14:08:47 -04003749 if (!should_alloc_chunk(extent_root, space_info, force)) {
Josef Bacik25179202008-10-29 14:49:05 -04003750 spin_unlock(&space_info->lock);
Josef Bacik6d741192011-04-11 20:20:11 -04003751 return 0;
3752 } else if (space_info->chunk_alloc) {
3753 wait_for_alloc = 1;
3754 } else {
3755 space_info->chunk_alloc = 1;
Josef Bacik25179202008-10-29 14:49:05 -04003756 }
Chris Mason0e4f8f82011-04-15 16:05:44 -04003757
Josef Bacik25179202008-10-29 14:49:05 -04003758 spin_unlock(&space_info->lock);
3759
Josef Bacik6d741192011-04-11 20:20:11 -04003760 mutex_lock(&fs_info->chunk_mutex);
3761
3762 /*
3763 * The chunk_mutex is held throughout the entirety of a chunk
3764 * allocation, so once we've acquired the chunk_mutex we know that the
3765 * other guy is done and we need to recheck and see if we should
3766 * allocate.
3767 */
3768 if (wait_for_alloc) {
3769 mutex_unlock(&fs_info->chunk_mutex);
3770 wait_for_alloc = 0;
3771 goto again;
3772 }
3773
Josef Bacikc6b305a2012-12-18 09:16:16 -05003774 trans->allocating_chunk = true;
3775
Josef Bacik97e728d2009-04-21 17:40:57 -04003776 /*
Josef Bacik67377732010-09-16 16:19:09 -04003777 * If we have mixed data/metadata chunks we want to make sure we keep
3778 * allocating mixed chunks instead of individual chunks.
3779 */
3780 if (btrfs_mixed_space_info(space_info))
3781 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3782
3783 /*
Josef Bacik97e728d2009-04-21 17:40:57 -04003784 * if we're doing a data chunk, go ahead and make sure that
3785 * we keep a reasonable number of metadata chunks allocated in the
3786 * FS as well.
3787 */
Josef Bacik9ed74f22009-09-11 16:12:44 -04003788 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
Josef Bacik97e728d2009-04-21 17:40:57 -04003789 fs_info->data_chunk_allocations++;
3790 if (!(fs_info->data_chunk_allocations %
3791 fs_info->metadata_ratio))
3792 force_metadata_allocation(fs_info);
3793 }
3794
Liu Bo15d1ff82012-03-29 09:57:44 -04003795 /*
3796 * Check if we have enough space in SYSTEM chunk because we may need
3797 * to update devices.
3798 */
3799 check_system_chunk(trans, extent_root, flags);
3800
Yan Zheng2b820322008-11-17 21:11:30 -05003801 ret = btrfs_alloc_chunk(trans, extent_root, flags);
Josef Bacikc6b305a2012-12-18 09:16:16 -05003802 trans->allocating_chunk = false;
Mark Fasheh92b8e8972011-07-12 10:57:59 -07003803
Josef Bacik9ed74f22009-09-11 16:12:44 -04003804 spin_lock(&space_info->lock);
Alexandre Olivaa81cb9a2013-02-21 21:15:14 +00003805 if (ret < 0 && ret != -ENOSPC)
3806 goto out;
Chris Masond3977122009-01-05 21:25:51 -05003807 if (ret)
Chris Mason6324fbf2008-03-24 15:01:59 -04003808 space_info->full = 1;
Yan, Zheng424499d2010-05-16 10:46:25 -04003809 else
3810 ret = 1;
Josef Bacik6d741192011-04-11 20:20:11 -04003811
Chris Mason0e4f8f82011-04-15 16:05:44 -04003812 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
Alexandre Olivaa81cb9a2013-02-21 21:15:14 +00003813out:
Josef Bacik6d741192011-04-11 20:20:11 -04003814 space_info->chunk_alloc = 0;
Josef Bacik9ed74f22009-09-11 16:12:44 -04003815 spin_unlock(&space_info->lock);
Dan Carpentera25c75d2012-04-18 09:59:29 +03003816 mutex_unlock(&fs_info->chunk_mutex);
Josef Bacik0f9dd462008-09-23 13:14:11 -04003817 return ret;
Chris Mason6324fbf2008-03-24 15:01:59 -04003818}
3819
Josef Bacika80c8dc2012-09-06 16:59:33 -04003820static int can_overcommit(struct btrfs_root *root,
3821 struct btrfs_space_info *space_info, u64 bytes,
Miao Xie08e007d2012-10-16 11:33:38 +00003822 enum btrfs_reserve_flush_enum flush)
Josef Bacika80c8dc2012-09-06 16:59:33 -04003823{
Josef Bacik96f1bb52013-01-30 17:02:51 -05003824 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
Josef Bacika80c8dc2012-09-06 16:59:33 -04003825 u64 profile = btrfs_get_alloc_profile(root, 0);
Miao Xie3c76cd82013-04-25 10:12:38 +00003826 u64 space_size;
Josef Bacika80c8dc2012-09-06 16:59:33 -04003827 u64 avail;
3828 u64 used;
Josef Bacik70afa392013-02-06 13:53:19 -05003829 u64 to_add;
Josef Bacika80c8dc2012-09-06 16:59:33 -04003830
3831 used = space_info->bytes_used + space_info->bytes_reserved +
Josef Bacik96f1bb52013-01-30 17:02:51 -05003832 space_info->bytes_pinned + space_info->bytes_readonly;
3833
Josef Bacik96f1bb52013-01-30 17:02:51 -05003834 /*
3835 * We only want to allow over committing if we have lots of actual space
3836 * free, but if we don't have enough space to handle the global reserve
3837 * space then we could end up having a real enospc problem when trying
3838 * to allocate a chunk or some other such important allocation.
3839 */
Miao Xie3c76cd82013-04-25 10:12:38 +00003840 spin_lock(&global_rsv->lock);
3841 space_size = calc_global_rsv_need_space(global_rsv);
3842 spin_unlock(&global_rsv->lock);
3843 if (used + space_size >= space_info->total_bytes)
Josef Bacik96f1bb52013-01-30 17:02:51 -05003844 return 0;
3845
3846 used += space_info->bytes_may_use;
Josef Bacika80c8dc2012-09-06 16:59:33 -04003847
3848 spin_lock(&root->fs_info->free_chunk_lock);
3849 avail = root->fs_info->free_chunk_space;
3850 spin_unlock(&root->fs_info->free_chunk_lock);
3851
3852 /*
3853 * If we have dup, raid1 or raid10 then only half of the free
David Woodhouse53b381b2013-01-29 18:40:14 -05003854 * space is actually useable. For raid56, the space info used
3855 * doesn't include the parity drive, so we don't have to
3856 * change the math
Josef Bacika80c8dc2012-09-06 16:59:33 -04003857 */
3858 if (profile & (BTRFS_BLOCK_GROUP_DUP |
3859 BTRFS_BLOCK_GROUP_RAID1 |
3860 BTRFS_BLOCK_GROUP_RAID10))
3861 avail >>= 1;
3862
Josef Bacik70afa392013-02-06 13:53:19 -05003863 to_add = space_info->total_bytes;
3864
Josef Bacika80c8dc2012-09-06 16:59:33 -04003865 /*
Miao Xie561c2942012-10-16 11:32:18 +00003866 * If we aren't flushing all things, let us overcommit up to
3867 * 1/2th of the space. If we can flush, don't let us overcommit
3868 * too much, let it overcommit up to 1/8 of the space.
Josef Bacika80c8dc2012-09-06 16:59:33 -04003869 */
Miao Xie08e007d2012-10-16 11:33:38 +00003870 if (flush == BTRFS_RESERVE_FLUSH_ALL)
Josef Bacik70afa392013-02-06 13:53:19 -05003871 to_add >>= 3;
Josef Bacika80c8dc2012-09-06 16:59:33 -04003872 else
Josef Bacik70afa392013-02-06 13:53:19 -05003873 to_add >>= 1;
Josef Bacika80c8dc2012-09-06 16:59:33 -04003874
Josef Bacik70afa392013-02-06 13:53:19 -05003875 /*
3876 * Limit the overcommit to the amount of free space we could possibly
3877 * allocate for chunks.
3878 */
3879 to_add = min(avail, to_add);
3880
3881 if (used + bytes < space_info->total_bytes + to_add)
Josef Bacika80c8dc2012-09-06 16:59:33 -04003882 return 1;
3883 return 0;
3884}
3885
Eric Sandeen48a3b632013-04-25 20:41:01 +00003886static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
3887 unsigned long nr_pages)
Miao Xieda633a42012-12-20 11:19:09 +00003888{
3889 struct super_block *sb = root->fs_info->sb;
3890 int started;
3891
3892 /* If we can not start writeback, just sync all the delalloc file. */
Linus Torvaldsb6951882013-03-02 16:41:54 -08003893 started = try_to_writeback_inodes_sb_nr(sb, nr_pages,
Miao Xieda633a42012-12-20 11:19:09 +00003894 WB_REASON_FS_FREE_SPACE);
3895 if (!started) {
3896 /*
3897 * We needn't worry the filesystem going from r/w to r/o though
3898 * we don't acquire ->s_umount mutex, because the filesystem
3899 * should guarantee the delalloc inodes list be empty after
3900 * the filesystem is readonly(all dirty pages are written to
3901 * the disk).
3902 */
3903 btrfs_start_delalloc_inodes(root, 0);
Josef Bacik98ad69c2013-04-04 11:55:49 -04003904 if (!current->journal_info)
3905 btrfs_wait_ordered_extents(root, 0);
Miao Xieda633a42012-12-20 11:19:09 +00003906 }
3907}
3908
Yan, Zheng5da9d012010-05-16 10:46:25 -04003909/*
3910 * shrink metadata reservation for delalloc
3911 */
Josef Bacikf4c738c2012-07-02 17:10:51 -04003912static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
3913 bool wait_ordered)
Yan, Zheng5da9d012010-05-16 10:46:25 -04003914{
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04003915 struct btrfs_block_rsv *block_rsv;
Josef Bacik0019f102010-10-15 15:18:40 -04003916 struct btrfs_space_info *space_info;
Josef Bacik663350a2011-11-03 22:54:25 -04003917 struct btrfs_trans_handle *trans;
Josef Bacikf4c738c2012-07-02 17:10:51 -04003918 u64 delalloc_bytes;
Yan, Zheng5da9d012010-05-16 10:46:25 -04003919 u64 max_reclaim;
Josef Bacikb1953bc2011-01-21 21:10:01 +00003920 long time_left;
Josef Bacik877da172011-10-14 14:02:10 -04003921 unsigned long nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
Josef Bacikb1953bc2011-01-21 21:10:01 +00003922 int loops = 0;
Miao Xie08e007d2012-10-16 11:33:38 +00003923 enum btrfs_reserve_flush_enum flush;
Yan, Zheng5da9d012010-05-16 10:46:25 -04003924
Josef Bacik663350a2011-11-03 22:54:25 -04003925 trans = (struct btrfs_trans_handle *)current->journal_info;
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04003926 block_rsv = &root->fs_info->delalloc_block_rsv;
Josef Bacik0019f102010-10-15 15:18:40 -04003927 space_info = block_rsv->space_info;
Chris Masonbf9022e2010-10-26 13:40:45 -04003928
3929 smp_mb();
Miao Xie963d6782013-01-29 10:10:51 +00003930 delalloc_bytes = percpu_counter_sum_positive(
3931 &root->fs_info->delalloc_bytes);
Josef Bacikf4c738c2012-07-02 17:10:51 -04003932 if (delalloc_bytes == 0) {
Josef Bacikfdb5eff2011-06-07 16:07:44 -04003933 if (trans)
Josef Bacikf4c738c2012-07-02 17:10:51 -04003934 return;
Liu Bo6bbe3a92012-09-14 02:58:07 -06003935 btrfs_wait_ordered_extents(root, 0);
Josef Bacikf4c738c2012-07-02 17:10:51 -04003936 return;
Josef Bacikfdb5eff2011-06-07 16:07:44 -04003937 }
3938
Josef Bacikf4c738c2012-07-02 17:10:51 -04003939 while (delalloc_bytes && loops < 3) {
3940 max_reclaim = min(delalloc_bytes, to_reclaim);
3941 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
Miao Xieda633a42012-12-20 11:19:09 +00003942 btrfs_writeback_inodes_sb_nr(root, nr_pages);
Josef Bacikdea31f52012-09-06 16:47:00 -04003943 /*
3944 * We need to wait for the async pages to actually start before
3945 * we do anything.
3946 */
3947 wait_event(root->fs_info->async_submit_wait,
3948 !atomic_read(&root->fs_info->async_delalloc_pages));
3949
Miao Xie08e007d2012-10-16 11:33:38 +00003950 if (!trans)
3951 flush = BTRFS_RESERVE_FLUSH_ALL;
3952 else
3953 flush = BTRFS_RESERVE_NO_FLUSH;
Josef Bacik0019f102010-10-15 15:18:40 -04003954 spin_lock(&space_info->lock);
Miao Xie08e007d2012-10-16 11:33:38 +00003955 if (can_overcommit(root, space_info, orig, flush)) {
Josef Bacikf4c738c2012-07-02 17:10:51 -04003956 spin_unlock(&space_info->lock);
3957 break;
3958 }
Josef Bacik0019f102010-10-15 15:18:40 -04003959 spin_unlock(&space_info->lock);
Yan, Zheng5da9d012010-05-16 10:46:25 -04003960
Chris Mason36e39c42011-03-12 07:08:42 -05003961 loops++;
Josef Bacikf104d042011-10-14 13:56:58 -04003962 if (wait_ordered && !trans) {
Liu Bo6bbe3a92012-09-14 02:58:07 -06003963 btrfs_wait_ordered_extents(root, 0);
Josef Bacikf104d042011-10-14 13:56:58 -04003964 } else {
Josef Bacikf4c738c2012-07-02 17:10:51 -04003965 time_left = schedule_timeout_killable(1);
Josef Bacikf104d042011-10-14 13:56:58 -04003966 if (time_left)
3967 break;
3968 }
Josef Bacikf4c738c2012-07-02 17:10:51 -04003969 smp_mb();
Miao Xie963d6782013-01-29 10:10:51 +00003970 delalloc_bytes = percpu_counter_sum_positive(
3971 &root->fs_info->delalloc_bytes);
Yan, Zheng5da9d012010-05-16 10:46:25 -04003972 }
Yan, Zheng5da9d012010-05-16 10:46:25 -04003973}
3974
Josef Bacik4a92b1b2011-08-30 12:34:28 -04003975/**
Josef Bacik663350a2011-11-03 22:54:25 -04003976 * maybe_commit_transaction - possibly commit the transaction if its ok to
3977 * @root - the root we're allocating for
3978 * @bytes - the number of bytes we want to reserve
3979 * @force - force the commit
Josef Bacik8bb8ab22010-10-15 16:52:49 -04003980 *
Josef Bacik663350a2011-11-03 22:54:25 -04003981 * This will check to make sure that committing the transaction will actually
3982 * get us somewhere and then commit the transaction if it does. Otherwise it
3983 * will return -ENOSPC.
Josef Bacik8bb8ab22010-10-15 16:52:49 -04003984 */
Josef Bacik663350a2011-11-03 22:54:25 -04003985static int may_commit_transaction(struct btrfs_root *root,
3986 struct btrfs_space_info *space_info,
3987 u64 bytes, int force)
3988{
3989 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
3990 struct btrfs_trans_handle *trans;
3991
3992 trans = (struct btrfs_trans_handle *)current->journal_info;
3993 if (trans)
3994 return -EAGAIN;
3995
3996 if (force)
3997 goto commit;
3998
3999 /* See if there is enough pinned space to make this reservation */
4000 spin_lock(&space_info->lock);
4001 if (space_info->bytes_pinned >= bytes) {
4002 spin_unlock(&space_info->lock);
4003 goto commit;
4004 }
4005 spin_unlock(&space_info->lock);
4006
4007 /*
4008 * See if there is some space in the delayed insertion reservation for
4009 * this reservation.
4010 */
4011 if (space_info != delayed_rsv->space_info)
4012 return -ENOSPC;
4013
Liu Bod9b02182012-02-16 18:34:39 +08004014 spin_lock(&space_info->lock);
Josef Bacik663350a2011-11-03 22:54:25 -04004015 spin_lock(&delayed_rsv->lock);
Liu Bod9b02182012-02-16 18:34:39 +08004016 if (space_info->bytes_pinned + delayed_rsv->size < bytes) {
Josef Bacik663350a2011-11-03 22:54:25 -04004017 spin_unlock(&delayed_rsv->lock);
Liu Bod9b02182012-02-16 18:34:39 +08004018 spin_unlock(&space_info->lock);
Josef Bacik663350a2011-11-03 22:54:25 -04004019 return -ENOSPC;
4020 }
4021 spin_unlock(&delayed_rsv->lock);
Liu Bod9b02182012-02-16 18:34:39 +08004022 spin_unlock(&space_info->lock);
Josef Bacik663350a2011-11-03 22:54:25 -04004023
4024commit:
4025 trans = btrfs_join_transaction(root);
4026 if (IS_ERR(trans))
4027 return -ENOSPC;
4028
4029 return btrfs_commit_transaction(trans, root);
4030}
4031
Josef Bacik96c3f432012-06-21 14:05:49 -04004032enum flush_state {
Josef Bacik67b0fd62012-09-24 13:42:00 -04004033 FLUSH_DELAYED_ITEMS_NR = 1,
4034 FLUSH_DELAYED_ITEMS = 2,
4035 FLUSH_DELALLOC = 3,
4036 FLUSH_DELALLOC_WAIT = 4,
Josef Bacikea658ba2012-09-11 16:57:25 -04004037 ALLOC_CHUNK = 5,
4038 COMMIT_TRANS = 6,
Josef Bacik96c3f432012-06-21 14:05:49 -04004039};
4040
4041static int flush_space(struct btrfs_root *root,
4042 struct btrfs_space_info *space_info, u64 num_bytes,
4043 u64 orig_bytes, int state)
4044{
4045 struct btrfs_trans_handle *trans;
4046 int nr;
Josef Bacikf4c738c2012-07-02 17:10:51 -04004047 int ret = 0;
Josef Bacik96c3f432012-06-21 14:05:49 -04004048
4049 switch (state) {
Josef Bacik96c3f432012-06-21 14:05:49 -04004050 case FLUSH_DELAYED_ITEMS_NR:
4051 case FLUSH_DELAYED_ITEMS:
4052 if (state == FLUSH_DELAYED_ITEMS_NR) {
4053 u64 bytes = btrfs_calc_trans_metadata_size(root, 1);
4054
4055 nr = (int)div64_u64(num_bytes, bytes);
4056 if (!nr)
4057 nr = 1;
4058 nr *= 2;
4059 } else {
4060 nr = -1;
4061 }
4062 trans = btrfs_join_transaction(root);
4063 if (IS_ERR(trans)) {
4064 ret = PTR_ERR(trans);
4065 break;
4066 }
4067 ret = btrfs_run_delayed_items_nr(trans, root, nr);
4068 btrfs_end_transaction(trans, root);
4069 break;
Josef Bacik67b0fd62012-09-24 13:42:00 -04004070 case FLUSH_DELALLOC:
4071 case FLUSH_DELALLOC_WAIT:
4072 shrink_delalloc(root, num_bytes, orig_bytes,
4073 state == FLUSH_DELALLOC_WAIT);
4074 break;
Josef Bacikea658ba2012-09-11 16:57:25 -04004075 case ALLOC_CHUNK:
4076 trans = btrfs_join_transaction(root);
4077 if (IS_ERR(trans)) {
4078 ret = PTR_ERR(trans);
4079 break;
4080 }
4081 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
Josef Bacikea658ba2012-09-11 16:57:25 -04004082 btrfs_get_alloc_profile(root, 0),
4083 CHUNK_ALLOC_NO_FORCE);
4084 btrfs_end_transaction(trans, root);
4085 if (ret == -ENOSPC)
4086 ret = 0;
4087 break;
Josef Bacik96c3f432012-06-21 14:05:49 -04004088 case COMMIT_TRANS:
4089 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
4090 break;
4091 default:
4092 ret = -ENOSPC;
4093 break;
4094 }
4095
4096 return ret;
4097}
Josef Bacik663350a2011-11-03 22:54:25 -04004098/**
Josef Bacik4a92b1b2011-08-30 12:34:28 -04004099 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4100 * @root - the root we're allocating for
4101 * @block_rsv - the block_rsv we're allocating for
4102 * @orig_bytes - the number of bytes we want
Adam Buchbinder48fc7f72012-09-19 21:48:00 -04004103 * @flush - whether or not we can flush to make our reservation
Josef Bacik8bb8ab22010-10-15 16:52:49 -04004104 *
Josef Bacik4a92b1b2011-08-30 12:34:28 -04004105 * This will reserve orgi_bytes number of bytes from the space info associated
4106 * with the block_rsv. If there is not enough space it will make an attempt to
4107 * flush out space to make room. It will do this by flushing delalloc if
4108 * possible or committing the transaction. If flush is 0 then no attempts to
4109 * regain reservations will be made and this will fail if there is not enough
4110 * space already.
Josef Bacik8bb8ab22010-10-15 16:52:49 -04004111 */
Josef Bacik4a92b1b2011-08-30 12:34:28 -04004112static int reserve_metadata_bytes(struct btrfs_root *root,
Josef Bacik8bb8ab22010-10-15 16:52:49 -04004113 struct btrfs_block_rsv *block_rsv,
Miao Xie08e007d2012-10-16 11:33:38 +00004114 u64 orig_bytes,
4115 enum btrfs_reserve_flush_enum flush)
Yan, Zhengf0486c62010-05-16 10:46:25 -04004116{
4117 struct btrfs_space_info *space_info = block_rsv->space_info;
Josef Bacik2bf64752011-09-26 17:12:22 -04004118 u64 used;
Josef Bacik8bb8ab22010-10-15 16:52:49 -04004119 u64 num_bytes = orig_bytes;
Josef Bacik67b0fd62012-09-24 13:42:00 -04004120 int flush_state = FLUSH_DELAYED_ITEMS_NR;
Josef Bacik8bb8ab22010-10-15 16:52:49 -04004121 int ret = 0;
Josef Bacikfdb5eff2011-06-07 16:07:44 -04004122 bool flushing = false;
Josef Bacik8bb8ab22010-10-15 16:52:49 -04004123
4124again:
Josef Bacikfdb5eff2011-06-07 16:07:44 -04004125 ret = 0;
Yan, Zhengf0486c62010-05-16 10:46:25 -04004126 spin_lock(&space_info->lock);
Josef Bacikfdb5eff2011-06-07 16:07:44 -04004127 /*
Miao Xie08e007d2012-10-16 11:33:38 +00004128 * We only want to wait if somebody other than us is flushing and we
4129 * are actually allowed to flush all things.
Josef Bacikfdb5eff2011-06-07 16:07:44 -04004130 */
Miao Xie08e007d2012-10-16 11:33:38 +00004131 while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
4132 space_info->flush) {
Josef Bacikfdb5eff2011-06-07 16:07:44 -04004133 spin_unlock(&space_info->lock);
4134 /*
4135 * If we have a trans handle we can't wait because the flusher
4136 * may have to commit the transaction, which would mean we would
4137 * deadlock since we are waiting for the flusher to finish, but
4138 * hold the current transaction open.
4139 */
Josef Bacik663350a2011-11-03 22:54:25 -04004140 if (current->journal_info)
Josef Bacikfdb5eff2011-06-07 16:07:44 -04004141 return -EAGAIN;
Arne Jansenb9688bb2012-04-18 10:27:16 +02004142 ret = wait_event_killable(space_info->wait, !space_info->flush);
4143 /* Must have been killed, return */
4144 if (ret)
Josef Bacikfdb5eff2011-06-07 16:07:44 -04004145 return -EINTR;
4146
4147 spin_lock(&space_info->lock);
4148 }
4149
4150 ret = -ENOSPC;
Josef Bacik2bf64752011-09-26 17:12:22 -04004151 used = space_info->bytes_used + space_info->bytes_reserved +
4152 space_info->bytes_pinned + space_info->bytes_readonly +
4153 space_info->bytes_may_use;
Yan, Zhengf0486c62010-05-16 10:46:25 -04004154
Josef Bacik8bb8ab22010-10-15 16:52:49 -04004155 /*
4156 * The idea here is that we've not already over-reserved the block group
4157 * then we can go ahead and save our reservation first and then start
4158 * flushing if we need to. Otherwise if we've already overcommitted
4159 * lets start flushing stuff first and then come back and try to make
4160 * our reservation.
4161 */
Josef Bacik2bf64752011-09-26 17:12:22 -04004162 if (used <= space_info->total_bytes) {
4163 if (used + orig_bytes <= space_info->total_bytes) {
Josef Bacikfb25e912011-07-26 17:00:46 -04004164 space_info->bytes_may_use += orig_bytes;
Josef Bacik8c2a3ca2012-01-10 10:31:31 -05004165 trace_btrfs_space_reservation(root->fs_info,
Liu Bo2bcc0322012-03-29 09:57:44 -04004166 "space_info", space_info->flags, orig_bytes, 1);
Yan, Zhengf0486c62010-05-16 10:46:25 -04004167 ret = 0;
4168 } else {
Josef Bacik8bb8ab22010-10-15 16:52:49 -04004169 /*
4170 * Ok set num_bytes to orig_bytes since we aren't
4171 * overocmmitted, this way we only try and reclaim what
4172 * we need.
4173 */
4174 num_bytes = orig_bytes;
Yan, Zhengf0486c62010-05-16 10:46:25 -04004175 }
Josef Bacik8bb8ab22010-10-15 16:52:49 -04004176 } else {
4177 /*
4178 * Ok we're over committed, set num_bytes to the overcommitted
4179 * amount plus the amount of bytes that we need for this
4180 * reservation.
4181 */
Josef Bacik2bf64752011-09-26 17:12:22 -04004182 num_bytes = used - space_info->total_bytes +
Josef Bacik96c3f432012-06-21 14:05:49 -04004183 (orig_bytes * 2);
Yan, Zhengf0486c62010-05-16 10:46:25 -04004184 }
Josef Bacik8bb8ab22010-10-15 16:52:49 -04004185
Josef Bacik44734ed2012-09-28 16:04:19 -04004186 if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
4187 space_info->bytes_may_use += orig_bytes;
4188 trace_btrfs_space_reservation(root->fs_info, "space_info",
4189 space_info->flags, orig_bytes,
4190 1);
4191 ret = 0;
Josef Bacik2bf64752011-09-26 17:12:22 -04004192 }
4193
Josef Bacik8bb8ab22010-10-15 16:52:49 -04004194 /*
4195 * Couldn't make our reservation, save our place so while we're trying
4196 * to reclaim space we can actually use it instead of somebody else
4197 * stealing it from us.
Miao Xie08e007d2012-10-16 11:33:38 +00004198 *
4199 * We make the other tasks wait for the flush only when we can flush
4200 * all things.
Josef Bacik8bb8ab22010-10-15 16:52:49 -04004201 */
Josef Bacik72bcd992012-12-18 15:16:34 -05004202 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
Josef Bacikfdb5eff2011-06-07 16:07:44 -04004203 flushing = true;
4204 space_info->flush = 1;
Josef Bacik8bb8ab22010-10-15 16:52:49 -04004205 }
4206
Yan, Zhengf0486c62010-05-16 10:46:25 -04004207 spin_unlock(&space_info->lock);
4208
Miao Xie08e007d2012-10-16 11:33:38 +00004209 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
Josef Bacik8bb8ab22010-10-15 16:52:49 -04004210 goto out;
4211
Josef Bacik96c3f432012-06-21 14:05:49 -04004212 ret = flush_space(root, space_info, num_bytes, orig_bytes,
4213 flush_state);
4214 flush_state++;
Miao Xie08e007d2012-10-16 11:33:38 +00004215
4216 /*
4217 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4218 * would happen. So skip delalloc flush.
4219 */
4220 if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4221 (flush_state == FLUSH_DELALLOC ||
4222 flush_state == FLUSH_DELALLOC_WAIT))
4223 flush_state = ALLOC_CHUNK;
4224
Josef Bacik96c3f432012-06-21 14:05:49 -04004225 if (!ret)
Josef Bacik8bb8ab22010-10-15 16:52:49 -04004226 goto again;
Miao Xie08e007d2012-10-16 11:33:38 +00004227 else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4228 flush_state < COMMIT_TRANS)
4229 goto again;
4230 else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
4231 flush_state <= COMMIT_TRANS)
Josef Bacik8bb8ab22010-10-15 16:52:49 -04004232 goto again;
4233
4234out:
Josef Bacik5d803662013-02-07 16:06:02 -05004235 if (ret == -ENOSPC &&
4236 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
4237 struct btrfs_block_rsv *global_rsv =
4238 &root->fs_info->global_block_rsv;
4239
4240 if (block_rsv != global_rsv &&
4241 !block_rsv_use_bytes(global_rsv, orig_bytes))
4242 ret = 0;
4243 }
Josef Bacikfdb5eff2011-06-07 16:07:44 -04004244 if (flushing) {
Josef Bacik8bb8ab22010-10-15 16:52:49 -04004245 spin_lock(&space_info->lock);
Josef Bacikfdb5eff2011-06-07 16:07:44 -04004246 space_info->flush = 0;
4247 wake_up_all(&space_info->wait);
Josef Bacik8bb8ab22010-10-15 16:52:49 -04004248 spin_unlock(&space_info->lock);
4249 }
Yan, Zhengf0486c62010-05-16 10:46:25 -04004250 return ret;
4251}
4252
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004253static struct btrfs_block_rsv *get_block_rsv(
4254 const struct btrfs_trans_handle *trans,
4255 const struct btrfs_root *root)
Yan, Zhengf0486c62010-05-16 10:46:25 -04004256{
Josef Bacik4c13d752011-08-30 11:31:29 -04004257 struct btrfs_block_rsv *block_rsv = NULL;
4258
Josef Bacik0e721102012-06-26 16:13:18 -04004259 if (root->ref_cows)
4260 block_rsv = trans->block_rsv;
4261
4262 if (root == root->fs_info->csum_root && trans->adding_csums)
Yan, Zhengf0486c62010-05-16 10:46:25 -04004263 block_rsv = trans->block_rsv;
Josef Bacik4c13d752011-08-30 11:31:29 -04004264
4265 if (!block_rsv)
Yan, Zhengf0486c62010-05-16 10:46:25 -04004266 block_rsv = root->block_rsv;
4267
4268 if (!block_rsv)
4269 block_rsv = &root->fs_info->empty_block_rsv;
4270
4271 return block_rsv;
4272}
4273
4274static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
4275 u64 num_bytes)
4276{
4277 int ret = -ENOSPC;
4278 spin_lock(&block_rsv->lock);
4279 if (block_rsv->reserved >= num_bytes) {
4280 block_rsv->reserved -= num_bytes;
4281 if (block_rsv->reserved < block_rsv->size)
4282 block_rsv->full = 0;
4283 ret = 0;
4284 }
4285 spin_unlock(&block_rsv->lock);
4286 return ret;
4287}
4288
4289static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
4290 u64 num_bytes, int update_size)
4291{
4292 spin_lock(&block_rsv->lock);
4293 block_rsv->reserved += num_bytes;
4294 if (update_size)
4295 block_rsv->size += num_bytes;
4296 else if (block_rsv->reserved >= block_rsv->size)
4297 block_rsv->full = 1;
4298 spin_unlock(&block_rsv->lock);
4299}
4300
Josef Bacik8c2a3ca2012-01-10 10:31:31 -05004301static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
4302 struct btrfs_block_rsv *block_rsv,
David Sterba62a45b62011-04-20 15:52:26 +02004303 struct btrfs_block_rsv *dest, u64 num_bytes)
Yan, Zhengf0486c62010-05-16 10:46:25 -04004304{
4305 struct btrfs_space_info *space_info = block_rsv->space_info;
4306
4307 spin_lock(&block_rsv->lock);
4308 if (num_bytes == (u64)-1)
4309 num_bytes = block_rsv->size;
4310 block_rsv->size -= num_bytes;
4311 if (block_rsv->reserved >= block_rsv->size) {
4312 num_bytes = block_rsv->reserved - block_rsv->size;
4313 block_rsv->reserved = block_rsv->size;
4314 block_rsv->full = 1;
4315 } else {
4316 num_bytes = 0;
4317 }
4318 spin_unlock(&block_rsv->lock);
4319
4320 if (num_bytes > 0) {
4321 if (dest) {
Josef Bacike9e22892011-01-24 21:43:19 +00004322 spin_lock(&dest->lock);
4323 if (!dest->full) {
4324 u64 bytes_to_add;
4325
4326 bytes_to_add = dest->size - dest->reserved;
4327 bytes_to_add = min(num_bytes, bytes_to_add);
4328 dest->reserved += bytes_to_add;
4329 if (dest->reserved >= dest->size)
4330 dest->full = 1;
4331 num_bytes -= bytes_to_add;
4332 }
4333 spin_unlock(&dest->lock);
4334 }
4335 if (num_bytes) {
Yan, Zhengf0486c62010-05-16 10:46:25 -04004336 spin_lock(&space_info->lock);
Josef Bacikfb25e912011-07-26 17:00:46 -04004337 space_info->bytes_may_use -= num_bytes;
Josef Bacik8c2a3ca2012-01-10 10:31:31 -05004338 trace_btrfs_space_reservation(fs_info, "space_info",
Liu Bo2bcc0322012-03-29 09:57:44 -04004339 space_info->flags, num_bytes, 0);
Chris Mason36e39c42011-03-12 07:08:42 -05004340 space_info->reservation_progress++;
Yan, Zhengf0486c62010-05-16 10:46:25 -04004341 spin_unlock(&space_info->lock);
4342 }
4343 }
4344}
4345
4346static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
4347 struct btrfs_block_rsv *dst, u64 num_bytes)
4348{
4349 int ret;
4350
4351 ret = block_rsv_use_bytes(src, num_bytes);
4352 if (ret)
4353 return ret;
4354
4355 block_rsv_add_bytes(dst, num_bytes, 1);
4356 return 0;
4357}
4358
Miao Xie66d8f3d2012-09-06 04:02:28 -06004359void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
Yan, Zhengf0486c62010-05-16 10:46:25 -04004360{
4361 memset(rsv, 0, sizeof(*rsv));
4362 spin_lock_init(&rsv->lock);
Miao Xie66d8f3d2012-09-06 04:02:28 -06004363 rsv->type = type;
Yan, Zhengf0486c62010-05-16 10:46:25 -04004364}
4365
Miao Xie66d8f3d2012-09-06 04:02:28 -06004366struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
4367 unsigned short type)
Yan, Zhengf0486c62010-05-16 10:46:25 -04004368{
4369 struct btrfs_block_rsv *block_rsv;
4370 struct btrfs_fs_info *fs_info = root->fs_info;
Yan, Zhengf0486c62010-05-16 10:46:25 -04004371
4372 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
4373 if (!block_rsv)
4374 return NULL;
4375
Miao Xie66d8f3d2012-09-06 04:02:28 -06004376 btrfs_init_block_rsv(block_rsv, type);
Yan, Zhengf0486c62010-05-16 10:46:25 -04004377 block_rsv->space_info = __find_space_info(fs_info,
4378 BTRFS_BLOCK_GROUP_METADATA);
Yan, Zhengf0486c62010-05-16 10:46:25 -04004379 return block_rsv;
4380}
4381
4382void btrfs_free_block_rsv(struct btrfs_root *root,
4383 struct btrfs_block_rsv *rsv)
4384{
Josef Bacik2aaa6652012-08-29 14:27:18 -04004385 if (!rsv)
4386 return;
Josef Bacikdabdb642011-08-08 12:50:18 -04004387 btrfs_block_rsv_release(root, rsv, (u64)-1);
4388 kfree(rsv);
Yan, Zhengf0486c62010-05-16 10:46:25 -04004389}
4390
Miao Xie08e007d2012-10-16 11:33:38 +00004391int btrfs_block_rsv_add(struct btrfs_root *root,
4392 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
4393 enum btrfs_reserve_flush_enum flush)
Yan, Zhengf0486c62010-05-16 10:46:25 -04004394{
4395 int ret;
4396
4397 if (num_bytes == 0)
4398 return 0;
Josef Bacik8bb8ab22010-10-15 16:52:49 -04004399
Miao Xie61b520a2011-11-10 20:45:05 -05004400 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
Yan, Zhengf0486c62010-05-16 10:46:25 -04004401 if (!ret) {
4402 block_rsv_add_bytes(block_rsv, num_bytes, 1);
4403 return 0;
4404 }
4405
Yan, Zhengf0486c62010-05-16 10:46:25 -04004406 return ret;
4407}
4408
Josef Bacik4a92b1b2011-08-30 12:34:28 -04004409int btrfs_block_rsv_check(struct btrfs_root *root,
Josef Bacik36ba0222011-10-18 12:15:48 -04004410 struct btrfs_block_rsv *block_rsv, int min_factor)
Yan, Zhengf0486c62010-05-16 10:46:25 -04004411{
4412 u64 num_bytes = 0;
Yan, Zhengf0486c62010-05-16 10:46:25 -04004413 int ret = -ENOSPC;
4414
4415 if (!block_rsv)
4416 return 0;
4417
4418 spin_lock(&block_rsv->lock);
Josef Bacik36ba0222011-10-18 12:15:48 -04004419 num_bytes = div_factor(block_rsv->size, min_factor);
4420 if (block_rsv->reserved >= num_bytes)
Yan, Zhengf0486c62010-05-16 10:46:25 -04004421 ret = 0;
Yan, Zhengf0486c62010-05-16 10:46:25 -04004422 spin_unlock(&block_rsv->lock);
Yan, Zhengf0486c62010-05-16 10:46:25 -04004423
Josef Bacik36ba0222011-10-18 12:15:48 -04004424 return ret;
4425}
4426
Miao Xie08e007d2012-10-16 11:33:38 +00004427int btrfs_block_rsv_refill(struct btrfs_root *root,
4428 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
4429 enum btrfs_reserve_flush_enum flush)
Josef Bacik36ba0222011-10-18 12:15:48 -04004430{
4431 u64 num_bytes = 0;
4432 int ret = -ENOSPC;
4433
4434 if (!block_rsv)
4435 return 0;
4436
4437 spin_lock(&block_rsv->lock);
4438 num_bytes = min_reserved;
Josef Bacik13553e52011-08-08 13:33:21 -04004439 if (block_rsv->reserved >= num_bytes)
Yan, Zhengf0486c62010-05-16 10:46:25 -04004440 ret = 0;
Josef Bacik13553e52011-08-08 13:33:21 -04004441 else
Yan, Zhengf0486c62010-05-16 10:46:25 -04004442 num_bytes -= block_rsv->reserved;
Yan, Zhengf0486c62010-05-16 10:46:25 -04004443 spin_unlock(&block_rsv->lock);
Josef Bacik13553e52011-08-08 13:33:21 -04004444
Yan, Zhengf0486c62010-05-16 10:46:25 -04004445 if (!ret)
4446 return 0;
4447
Miao Xieaa38a712011-11-18 17:43:00 +08004448 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
Josef Bacikdabdb642011-08-08 12:50:18 -04004449 if (!ret) {
4450 block_rsv_add_bytes(block_rsv, num_bytes, 0);
Yan, Zhengf0486c62010-05-16 10:46:25 -04004451 return 0;
4452 }
4453
Josef Bacik13553e52011-08-08 13:33:21 -04004454 return ret;
Yan, Zhengf0486c62010-05-16 10:46:25 -04004455}
4456
4457int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
4458 struct btrfs_block_rsv *dst_rsv,
4459 u64 num_bytes)
4460{
4461 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4462}
4463
4464void btrfs_block_rsv_release(struct btrfs_root *root,
4465 struct btrfs_block_rsv *block_rsv,
4466 u64 num_bytes)
4467{
4468 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4469 if (global_rsv->full || global_rsv == block_rsv ||
4470 block_rsv->space_info != global_rsv->space_info)
4471 global_rsv = NULL;
Josef Bacik8c2a3ca2012-01-10 10:31:31 -05004472 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
4473 num_bytes);
Yan, Zhengf0486c62010-05-16 10:46:25 -04004474}
4475
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04004476/*
4477 * helper to calculate size of global block reservation.
4478 * the desired value is sum of space used by extent tree,
4479 * checksum tree and root tree
4480 */
4481static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
4482{
4483 struct btrfs_space_info *sinfo;
4484 u64 num_bytes;
4485 u64 meta_used;
4486 u64 data_used;
David Sterba6c417612011-04-13 15:41:04 +02004487 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04004488
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04004489 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
4490 spin_lock(&sinfo->lock);
4491 data_used = sinfo->bytes_used;
4492 spin_unlock(&sinfo->lock);
4493
4494 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4495 spin_lock(&sinfo->lock);
Josef Bacik6d487552010-10-15 15:13:32 -04004496 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
4497 data_used = 0;
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04004498 meta_used = sinfo->bytes_used;
4499 spin_unlock(&sinfo->lock);
4500
4501 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
4502 csum_size * 2;
4503 num_bytes += div64_u64(data_used + meta_used, 50);
4504
4505 if (num_bytes * 3 > meta_used)
Chris Mason8e62c2d2012-04-12 13:46:48 -04004506 num_bytes = div64_u64(meta_used, 3);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04004507
4508 return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
4509}
4510
4511static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
4512{
4513 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4514 struct btrfs_space_info *sinfo = block_rsv->space_info;
4515 u64 num_bytes;
4516
4517 num_bytes = calc_global_metadata_size(fs_info);
4518
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04004519 spin_lock(&sinfo->lock);
Stefan Behrens1f699d32012-04-27 12:41:46 -04004520 spin_lock(&block_rsv->lock);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04004521
Josef Bacikfdf30d12013-03-26 15:31:45 -04004522 block_rsv->size = min_t(u64, num_bytes, 512 * 1024 * 1024);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04004523
4524 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
Josef Bacik6d487552010-10-15 15:13:32 -04004525 sinfo->bytes_reserved + sinfo->bytes_readonly +
4526 sinfo->bytes_may_use;
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04004527
4528 if (sinfo->total_bytes > num_bytes) {
4529 num_bytes = sinfo->total_bytes - num_bytes;
4530 block_rsv->reserved += num_bytes;
Josef Bacikfb25e912011-07-26 17:00:46 -04004531 sinfo->bytes_may_use += num_bytes;
Josef Bacik8c2a3ca2012-01-10 10:31:31 -05004532 trace_btrfs_space_reservation(fs_info, "space_info",
Liu Bo2bcc0322012-03-29 09:57:44 -04004533 sinfo->flags, num_bytes, 1);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04004534 }
4535
4536 if (block_rsv->reserved >= block_rsv->size) {
4537 num_bytes = block_rsv->reserved - block_rsv->size;
Josef Bacikfb25e912011-07-26 17:00:46 -04004538 sinfo->bytes_may_use -= num_bytes;
Josef Bacik8c2a3ca2012-01-10 10:31:31 -05004539 trace_btrfs_space_reservation(fs_info, "space_info",
Liu Bo2bcc0322012-03-29 09:57:44 -04004540 sinfo->flags, num_bytes, 0);
Chris Mason36e39c42011-03-12 07:08:42 -05004541 sinfo->reservation_progress++;
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04004542 block_rsv->reserved = block_rsv->size;
4543 block_rsv->full = 1;
4544 }
David Sterba182608c2011-05-05 13:13:16 +02004545
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04004546 spin_unlock(&block_rsv->lock);
Stefan Behrens1f699d32012-04-27 12:41:46 -04004547 spin_unlock(&sinfo->lock);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04004548}
4549
Yan, Zhengf0486c62010-05-16 10:46:25 -04004550static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
4551{
4552 struct btrfs_space_info *space_info;
4553
4554 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4555 fs_info->chunk_block_rsv.space_info = space_info;
Yan, Zhengf0486c62010-05-16 10:46:25 -04004556
4557 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04004558 fs_info->global_block_rsv.space_info = space_info;
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04004559 fs_info->delalloc_block_rsv.space_info = space_info;
Yan, Zhengf0486c62010-05-16 10:46:25 -04004560 fs_info->trans_block_rsv.space_info = space_info;
4561 fs_info->empty_block_rsv.space_info = space_info;
Josef Bacik6d668dd2011-11-03 22:54:25 -04004562 fs_info->delayed_block_rsv.space_info = space_info;
Yan, Zhengf0486c62010-05-16 10:46:25 -04004563
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04004564 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
4565 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
4566 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
4567 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
Stefan Behrens3a6cad92013-05-16 14:48:19 +00004568 if (fs_info->quota_root)
4569 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
Yan, Zhengf0486c62010-05-16 10:46:25 -04004570 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04004571
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04004572 update_global_block_rsv(fs_info);
4573}
4574
4575static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
4576{
Josef Bacik8c2a3ca2012-01-10 10:31:31 -05004577 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
4578 (u64)-1);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04004579 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
4580 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
4581 WARN_ON(fs_info->trans_block_rsv.size > 0);
4582 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
4583 WARN_ON(fs_info->chunk_block_rsv.size > 0);
4584 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
Josef Bacik6d668dd2011-11-03 22:54:25 -04004585 WARN_ON(fs_info->delayed_block_rsv.size > 0);
4586 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
Josef Bacikfcb80c22011-05-03 10:40:22 -04004587}
4588
Yan, Zhenga22285a2010-05-16 10:48:46 -04004589void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4590 struct btrfs_root *root)
4591{
Josef Bacik0e721102012-06-26 16:13:18 -04004592 if (!trans->block_rsv)
4593 return;
4594
Yan, Zhenga22285a2010-05-16 10:48:46 -04004595 if (!trans->bytes_reserved)
4596 return;
4597
Chris Masone77266e2012-02-24 10:39:05 -05004598 trace_btrfs_space_reservation(root->fs_info, "transaction",
Liu Bo2bcc0322012-03-29 09:57:44 -04004599 trans->transid, trans->bytes_reserved, 0);
Josef Bacikb24e03d2011-10-14 14:40:17 -04004600 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
Yan, Zhenga22285a2010-05-16 10:48:46 -04004601 trans->bytes_reserved = 0;
4602}
4603
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004604/* Can only return 0 or -ENOSPC */
Yan, Zhengd68fc572010-05-16 10:49:58 -04004605int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4606 struct inode *inode)
4607{
4608 struct btrfs_root *root = BTRFS_I(inode)->root;
4609 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4610 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4611
4612 /*
Josef Bacikfcb80c22011-05-03 10:40:22 -04004613 * We need to hold space in order to delete our orphan item once we've
4614 * added it, so this takes the reservation so we can release it later
4615 * when we are truly done with the orphan item.
Yan, Zhengd68fc572010-05-16 10:49:58 -04004616 */
Chris Masonff5714c2011-05-28 07:00:39 -04004617 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
Josef Bacik8c2a3ca2012-01-10 10:31:31 -05004618 trace_btrfs_space_reservation(root->fs_info, "orphan",
4619 btrfs_ino(inode), num_bytes, 1);
Yan, Zhengd68fc572010-05-16 10:49:58 -04004620 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4621}
4622
4623void btrfs_orphan_release_metadata(struct inode *inode)
4624{
4625 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Masonff5714c2011-05-28 07:00:39 -04004626 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
Josef Bacik8c2a3ca2012-01-10 10:31:31 -05004627 trace_btrfs_space_reservation(root->fs_info, "orphan",
4628 btrfs_ino(inode), num_bytes, 0);
Yan, Zhengd68fc572010-05-16 10:49:58 -04004629 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4630}
4631
Miao Xied5c12072013-02-28 10:04:33 +00004632/*
4633 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
4634 * root: the root of the parent directory
4635 * rsv: block reservation
4636 * items: the number of items that we need do reservation
4637 * qgroup_reserved: used to return the reserved size in qgroup
4638 *
4639 * This function is used to reserve the space for snapshot/subvolume
4640 * creation and deletion. Those operations are different with the
4641 * common file/directory operations, they change two fs/file trees
4642 * and root tree, the number of items that the qgroup reserves is
4643 * different with the free space reservation. So we can not use
4644 * the space reseravtion mechanism in start_transaction().
4645 */
4646int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
4647 struct btrfs_block_rsv *rsv,
4648 int items,
4649 u64 *qgroup_reserved)
Yan, Zhenga22285a2010-05-16 10:48:46 -04004650{
Miao Xied5c12072013-02-28 10:04:33 +00004651 u64 num_bytes;
4652 int ret;
4653
4654 if (root->fs_info->quota_enabled) {
4655 /* One for parent inode, two for dir entries */
4656 num_bytes = 3 * root->leafsize;
4657 ret = btrfs_qgroup_reserve(root, num_bytes);
4658 if (ret)
4659 return ret;
4660 } else {
4661 num_bytes = 0;
4662 }
4663
4664 *qgroup_reserved = num_bytes;
4665
4666 num_bytes = btrfs_calc_trans_metadata_size(root, items);
4667 rsv->space_info = __find_space_info(root->fs_info,
4668 BTRFS_BLOCK_GROUP_METADATA);
4669 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
4670 BTRFS_RESERVE_FLUSH_ALL);
4671 if (ret) {
4672 if (*qgroup_reserved)
4673 btrfs_qgroup_free(root, *qgroup_reserved);
4674 }
4675
4676 return ret;
4677}
4678
4679void btrfs_subvolume_release_metadata(struct btrfs_root *root,
4680 struct btrfs_block_rsv *rsv,
4681 u64 qgroup_reserved)
4682{
4683 btrfs_block_rsv_release(root, rsv, (u64)-1);
4684 if (qgroup_reserved)
4685 btrfs_qgroup_free(root, qgroup_reserved);
Yan, Zhenga22285a2010-05-16 10:48:46 -04004686}
4687
Josef Bacik7709cde2011-08-04 10:25:02 -04004688/**
4689 * drop_outstanding_extent - drop an outstanding extent
4690 * @inode: the inode we're dropping the extent for
4691 *
4692 * This is called when we are freeing up an outstanding extent, either called
4693 * after an error or after an extent is written. This will return the number of
4694 * reserved extents that need to be freed. This must be called with
4695 * BTRFS_I(inode)->lock held.
4696 */
Josef Bacik9e0baf62011-07-15 15:16:44 +00004697static unsigned drop_outstanding_extent(struct inode *inode)
4698{
Josef Bacik7fd2ae22011-11-08 15:47:34 -05004699 unsigned drop_inode_space = 0;
Josef Bacik9e0baf62011-07-15 15:16:44 +00004700 unsigned dropped_extents = 0;
4701
Josef Bacik9e0baf62011-07-15 15:16:44 +00004702 BUG_ON(!BTRFS_I(inode)->outstanding_extents);
4703 BTRFS_I(inode)->outstanding_extents--;
4704
Josef Bacik7fd2ae22011-11-08 15:47:34 -05004705 if (BTRFS_I(inode)->outstanding_extents == 0 &&
Josef Bacik72ac3c02012-05-23 14:13:11 -04004706 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4707 &BTRFS_I(inode)->runtime_flags))
Josef Bacik7fd2ae22011-11-08 15:47:34 -05004708 drop_inode_space = 1;
Josef Bacik7fd2ae22011-11-08 15:47:34 -05004709
Josef Bacik9e0baf62011-07-15 15:16:44 +00004710 /*
4711 * If we have more or the same amount of outsanding extents than we have
4712 * reserved then we need to leave the reserved extents count alone.
4713 */
4714 if (BTRFS_I(inode)->outstanding_extents >=
4715 BTRFS_I(inode)->reserved_extents)
Josef Bacik7fd2ae22011-11-08 15:47:34 -05004716 return drop_inode_space;
Josef Bacik9e0baf62011-07-15 15:16:44 +00004717
4718 dropped_extents = BTRFS_I(inode)->reserved_extents -
4719 BTRFS_I(inode)->outstanding_extents;
4720 BTRFS_I(inode)->reserved_extents -= dropped_extents;
Josef Bacik7fd2ae22011-11-08 15:47:34 -05004721 return dropped_extents + drop_inode_space;
Josef Bacik9e0baf62011-07-15 15:16:44 +00004722}
4723
Josef Bacik7709cde2011-08-04 10:25:02 -04004724/**
4725 * calc_csum_metadata_size - return the amount of metada space that must be
4726 * reserved/free'd for the given bytes.
4727 * @inode: the inode we're manipulating
4728 * @num_bytes: the number of bytes in question
4729 * @reserve: 1 if we are reserving space, 0 if we are freeing space
4730 *
4731 * This adjusts the number of csum_bytes in the inode and then returns the
4732 * correct amount of metadata that must either be reserved or freed. We
4733 * calculate how many checksums we can fit into one leaf and then divide the
4734 * number of bytes that will need to be checksumed by this value to figure out
4735 * how many checksums will be required. If we are adding bytes then the number
4736 * may go up and we will return the number of additional bytes that must be
4737 * reserved. If it is going down we will return the number of bytes that must
4738 * be freed.
4739 *
4740 * This must be called with BTRFS_I(inode)->lock held.
4741 */
4742static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
4743 int reserve)
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04004744{
Josef Bacik7709cde2011-08-04 10:25:02 -04004745 struct btrfs_root *root = BTRFS_I(inode)->root;
4746 u64 csum_size;
4747 int num_csums_per_leaf;
4748 int num_csums;
4749 int old_csums;
4750
4751 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
4752 BTRFS_I(inode)->csum_bytes == 0)
4753 return 0;
4754
4755 old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4756 if (reserve)
4757 BTRFS_I(inode)->csum_bytes += num_bytes;
4758 else
4759 BTRFS_I(inode)->csum_bytes -= num_bytes;
4760 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
4761 num_csums_per_leaf = (int)div64_u64(csum_size,
4762 sizeof(struct btrfs_csum_item) +
4763 sizeof(struct btrfs_disk_key));
4764 num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4765 num_csums = num_csums + num_csums_per_leaf - 1;
4766 num_csums = num_csums / num_csums_per_leaf;
4767
4768 old_csums = old_csums + num_csums_per_leaf - 1;
4769 old_csums = old_csums / num_csums_per_leaf;
4770
4771 /* No change, no need to reserve more */
4772 if (old_csums == num_csums)
4773 return 0;
4774
4775 if (reserve)
4776 return btrfs_calc_trans_metadata_size(root,
4777 num_csums - old_csums);
4778
4779 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04004780}
4781
4782int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4783{
4784 struct btrfs_root *root = BTRFS_I(inode)->root;
4785 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
Josef Bacik9e0baf62011-07-15 15:16:44 +00004786 u64 to_reserve = 0;
Josef Bacik660d3f62011-12-09 11:18:51 -05004787 u64 csum_bytes;
Josef Bacik9e0baf62011-07-15 15:16:44 +00004788 unsigned nr_extents = 0;
Josef Bacik660d3f62011-12-09 11:18:51 -05004789 int extra_reserve = 0;
Miao Xie08e007d2012-10-16 11:33:38 +00004790 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
Jan Schmidteb6b88d2013-01-27 23:26:00 -07004791 int ret = 0;
Josef Bacikc64c2bd2012-12-14 13:48:14 -05004792 bool delalloc_lock = true;
Wang Shilong88e081bf2013-03-01 11:36:01 +00004793 u64 to_free = 0;
4794 unsigned dropped;
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04004795
Josef Bacikc64c2bd2012-12-14 13:48:14 -05004796 /* If we are a free space inode we need to not flush since we will be in
4797 * the middle of a transaction commit. We also don't need the delalloc
4798 * mutex since we won't race with anybody. We need this mostly to make
4799 * lockdep shut its filthy mouth.
4800 */
4801 if (btrfs_is_free_space_inode(inode)) {
Miao Xie08e007d2012-10-16 11:33:38 +00004802 flush = BTRFS_RESERVE_NO_FLUSH;
Josef Bacikc64c2bd2012-12-14 13:48:14 -05004803 delalloc_lock = false;
4804 }
Josef Bacikc09544e2011-08-30 10:19:10 -04004805
Miao Xie08e007d2012-10-16 11:33:38 +00004806 if (flush != BTRFS_RESERVE_NO_FLUSH &&
4807 btrfs_transaction_in_commit(root->fs_info))
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04004808 schedule_timeout(1);
4809
Josef Bacikc64c2bd2012-12-14 13:48:14 -05004810 if (delalloc_lock)
4811 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
4812
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04004813 num_bytes = ALIGN(num_bytes, root->sectorsize);
Josef Bacik8bb8ab22010-10-15 16:52:49 -04004814
Josef Bacik9e0baf62011-07-15 15:16:44 +00004815 spin_lock(&BTRFS_I(inode)->lock);
4816 BTRFS_I(inode)->outstanding_extents++;
Josef Bacik57a45ced2011-01-25 16:30:38 -05004817
Josef Bacik9e0baf62011-07-15 15:16:44 +00004818 if (BTRFS_I(inode)->outstanding_extents >
Josef Bacik660d3f62011-12-09 11:18:51 -05004819 BTRFS_I(inode)->reserved_extents)
Josef Bacik9e0baf62011-07-15 15:16:44 +00004820 nr_extents = BTRFS_I(inode)->outstanding_extents -
4821 BTRFS_I(inode)->reserved_extents;
Josef Bacik7fd2ae22011-11-08 15:47:34 -05004822
4823 /*
4824 * Add an item to reserve for updating the inode when we complete the
4825 * delalloc io.
4826 */
Josef Bacik72ac3c02012-05-23 14:13:11 -04004827 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4828 &BTRFS_I(inode)->runtime_flags)) {
Josef Bacik7fd2ae22011-11-08 15:47:34 -05004829 nr_extents++;
Josef Bacik660d3f62011-12-09 11:18:51 -05004830 extra_reserve = 1;
Josef Bacik7fd2ae22011-11-08 15:47:34 -05004831 }
4832
4833 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
Josef Bacik7709cde2011-08-04 10:25:02 -04004834 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
Josef Bacik660d3f62011-12-09 11:18:51 -05004835 csum_bytes = BTRFS_I(inode)->csum_bytes;
Josef Bacik9e0baf62011-07-15 15:16:44 +00004836 spin_unlock(&BTRFS_I(inode)->lock);
Josef Bacik57a45ced2011-01-25 16:30:38 -05004837
Wang Shilong88e081bf2013-03-01 11:36:01 +00004838 if (root->fs_info->quota_enabled) {
Arne Jansenc5567232011-09-14 15:44:05 +02004839 ret = btrfs_qgroup_reserve(root, num_bytes +
4840 nr_extents * root->leafsize);
Wang Shilong88e081bf2013-03-01 11:36:01 +00004841 if (ret)
4842 goto out_fail;
Wang Shilonga9870c02013-03-01 11:33:01 +00004843 }
Arne Jansenc5567232011-09-14 15:44:05 +02004844
Wang Shilong88e081bf2013-03-01 11:36:01 +00004845 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
4846 if (unlikely(ret)) {
4847 if (root->fs_info->quota_enabled)
Miao Xie4b5829a2012-12-05 10:53:25 +00004848 btrfs_qgroup_free(root, num_bytes +
4849 nr_extents * root->leafsize);
Wang Shilong88e081bf2013-03-01 11:36:01 +00004850 goto out_fail;
Josef Bacik9e0baf62011-07-15 15:16:44 +00004851 }
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04004852
Josef Bacik660d3f62011-12-09 11:18:51 -05004853 spin_lock(&BTRFS_I(inode)->lock);
4854 if (extra_reserve) {
Josef Bacik72ac3c02012-05-23 14:13:11 -04004855 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4856 &BTRFS_I(inode)->runtime_flags);
Josef Bacik660d3f62011-12-09 11:18:51 -05004857 nr_extents--;
4858 }
4859 BTRFS_I(inode)->reserved_extents += nr_extents;
4860 spin_unlock(&BTRFS_I(inode)->lock);
Josef Bacikc64c2bd2012-12-14 13:48:14 -05004861
4862 if (delalloc_lock)
4863 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
Josef Bacik660d3f62011-12-09 11:18:51 -05004864
Josef Bacik8c2a3ca2012-01-10 10:31:31 -05004865 if (to_reserve)
4866 trace_btrfs_space_reservation(root->fs_info,"delalloc",
4867 btrfs_ino(inode), to_reserve, 1);
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04004868 block_rsv_add_bytes(block_rsv, to_reserve, 1);
4869
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04004870 return 0;
Wang Shilong88e081bf2013-03-01 11:36:01 +00004871
4872out_fail:
4873 spin_lock(&BTRFS_I(inode)->lock);
4874 dropped = drop_outstanding_extent(inode);
4875 /*
4876 * If the inodes csum_bytes is the same as the original
4877 * csum_bytes then we know we haven't raced with any free()ers
4878 * so we can just reduce our inodes csum bytes and carry on.
Wang Shilong88e081bf2013-03-01 11:36:01 +00004879 */
Josef Bacikf4881bc2013-03-25 16:03:35 -04004880 if (BTRFS_I(inode)->csum_bytes == csum_bytes) {
Wang Shilong88e081bf2013-03-01 11:36:01 +00004881 calc_csum_metadata_size(inode, num_bytes, 0);
Josef Bacikf4881bc2013-03-25 16:03:35 -04004882 } else {
4883 u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes;
4884 u64 bytes;
4885
4886 /*
4887 * This is tricky, but first we need to figure out how much we
4888 * free'd from any free-ers that occured during this
4889 * reservation, so we reset ->csum_bytes to the csum_bytes
4890 * before we dropped our lock, and then call the free for the
4891 * number of bytes that were freed while we were trying our
4892 * reservation.
4893 */
4894 bytes = csum_bytes - BTRFS_I(inode)->csum_bytes;
4895 BTRFS_I(inode)->csum_bytes = csum_bytes;
4896 to_free = calc_csum_metadata_size(inode, bytes, 0);
4897
4898
4899 /*
4900 * Now we need to see how much we would have freed had we not
4901 * been making this reservation and our ->csum_bytes were not
4902 * artificially inflated.
4903 */
4904 BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes;
4905 bytes = csum_bytes - orig_csum_bytes;
4906 bytes = calc_csum_metadata_size(inode, bytes, 0);
4907
4908 /*
4909 * Now reset ->csum_bytes to what it should be. If bytes is
4910 * more than to_free then we would have free'd more space had we
4911 * not had an artificially high ->csum_bytes, so we need to free
4912 * the remainder. If bytes is the same or less then we don't
4913 * need to do anything, the other free-ers did the correct
4914 * thing.
4915 */
4916 BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes;
4917 if (bytes > to_free)
4918 to_free = bytes - to_free;
4919 else
4920 to_free = 0;
4921 }
Wang Shilong88e081bf2013-03-01 11:36:01 +00004922 spin_unlock(&BTRFS_I(inode)->lock);
4923 if (dropped)
4924 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4925
4926 if (to_free) {
4927 btrfs_block_rsv_release(root, block_rsv, to_free);
4928 trace_btrfs_space_reservation(root->fs_info, "delalloc",
4929 btrfs_ino(inode), to_free, 0);
4930 }
4931 if (delalloc_lock)
4932 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4933 return ret;
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04004934}
4935
Josef Bacik7709cde2011-08-04 10:25:02 -04004936/**
4937 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
4938 * @inode: the inode to release the reservation for
4939 * @num_bytes: the number of bytes we're releasing
4940 *
4941 * This will release the metadata reservation for an inode. This can be called
4942 * once we complete IO for a given set of bytes to release their metadata
4943 * reservations.
4944 */
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04004945void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
4946{
4947 struct btrfs_root *root = BTRFS_I(inode)->root;
Josef Bacik9e0baf62011-07-15 15:16:44 +00004948 u64 to_free = 0;
4949 unsigned dropped;
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04004950
4951 num_bytes = ALIGN(num_bytes, root->sectorsize);
Josef Bacik7709cde2011-08-04 10:25:02 -04004952 spin_lock(&BTRFS_I(inode)->lock);
Josef Bacik9e0baf62011-07-15 15:16:44 +00004953 dropped = drop_outstanding_extent(inode);
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04004954
Miao Xie09348562013-02-07 10:12:07 +00004955 if (num_bytes)
4956 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
Josef Bacik7709cde2011-08-04 10:25:02 -04004957 spin_unlock(&BTRFS_I(inode)->lock);
Josef Bacik9e0baf62011-07-15 15:16:44 +00004958 if (dropped > 0)
4959 to_free += btrfs_calc_trans_metadata_size(root, dropped);
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04004960
Josef Bacik8c2a3ca2012-01-10 10:31:31 -05004961 trace_btrfs_space_reservation(root->fs_info, "delalloc",
4962 btrfs_ino(inode), to_free, 0);
Arne Jansenc5567232011-09-14 15:44:05 +02004963 if (root->fs_info->quota_enabled) {
4964 btrfs_qgroup_free(root, num_bytes +
4965 dropped * root->leafsize);
4966 }
4967
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04004968 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
4969 to_free);
4970}
4971
Josef Bacik7709cde2011-08-04 10:25:02 -04004972/**
4973 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
4974 * @inode: inode we're writing to
4975 * @num_bytes: the number of bytes we want to allocate
4976 *
4977 * This will do the following things
4978 *
4979 * o reserve space in the data space info for num_bytes
4980 * o reserve space in the metadata space info based on number of outstanding
4981 * extents and how much csums will be needed
4982 * o add to the inodes ->delalloc_bytes
4983 * o add it to the fs_info's delalloc inodes list.
4984 *
4985 * This will return 0 for success and -ENOSPC if there is no space left.
4986 */
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04004987int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
4988{
4989 int ret;
4990
4991 ret = btrfs_check_data_free_space(inode, num_bytes);
4992 if (ret)
4993 return ret;
4994
4995 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
4996 if (ret) {
4997 btrfs_free_reserved_data_space(inode, num_bytes);
4998 return ret;
4999 }
5000
5001 return 0;
5002}
5003
Josef Bacik7709cde2011-08-04 10:25:02 -04005004/**
5005 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5006 * @inode: inode we're releasing space for
5007 * @num_bytes: the number of bytes we want to free up
5008 *
5009 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5010 * called in the case that we don't need the metadata AND data reservations
5011 * anymore. So if there is an error or we insert an inline extent.
5012 *
5013 * This function will release the metadata space that was not used and will
5014 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5015 * list if there are no delalloc bytes left.
5016 */
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04005017void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
5018{
5019 btrfs_delalloc_release_metadata(inode, num_bytes);
5020 btrfs_free_reserved_data_space(inode, num_bytes);
5021}
5022
Liu Boc53d6132012-12-27 09:01:19 +00005023static int update_block_group(struct btrfs_root *root,
Yan, Zhengf0486c62010-05-16 10:46:25 -04005024 u64 bytenr, u64 num_bytes, int alloc)
Chris Mason9078a3e2007-04-26 16:46:15 -04005025{
Josef Bacik0af3d002010-06-21 14:48:16 -04005026 struct btrfs_block_group_cache *cache = NULL;
Chris Mason9078a3e2007-04-26 16:46:15 -04005027 struct btrfs_fs_info *info = root->fs_info;
Chris Masondb945352007-10-15 16:15:53 -04005028 u64 total = num_bytes;
Chris Mason9078a3e2007-04-26 16:46:15 -04005029 u64 old_val;
Chris Masondb945352007-10-15 16:15:53 -04005030 u64 byte_in_group;
Josef Bacik0af3d002010-06-21 14:48:16 -04005031 int factor;
Chris Mason3e1ad542007-05-07 20:03:49 -04005032
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005033 /* block accounting for super block */
5034 spin_lock(&info->delalloc_lock);
David Sterba6c417612011-04-13 15:41:04 +02005035 old_val = btrfs_super_bytes_used(info->super_copy);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005036 if (alloc)
5037 old_val += num_bytes;
5038 else
5039 old_val -= num_bytes;
David Sterba6c417612011-04-13 15:41:04 +02005040 btrfs_set_super_bytes_used(info->super_copy, old_val);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005041 spin_unlock(&info->delalloc_lock);
5042
Chris Masond3977122009-01-05 21:25:51 -05005043 while (total) {
Chris Masondb945352007-10-15 16:15:53 -04005044 cache = btrfs_lookup_block_group(info, bytenr);
Josef Bacikf3465ca2008-11-12 14:19:50 -05005045 if (!cache)
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005046 return -ENOENT;
Yan, Zhengb742bb82010-05-16 10:46:24 -04005047 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
5048 BTRFS_BLOCK_GROUP_RAID1 |
5049 BTRFS_BLOCK_GROUP_RAID10))
5050 factor = 2;
5051 else
5052 factor = 1;
Josef Bacik9d66e232010-08-25 16:54:15 -04005053 /*
5054 * If this block group has free space cache written out, we
5055 * need to make sure to load it if we are removing space. This
5056 * is because we need the unpinning stage to actually add the
5057 * space back to the block group, otherwise we will leak space.
5058 */
5059 if (!alloc && cache->cached == BTRFS_CACHE_NO)
Liu Bof6373bf2012-12-27 09:01:18 +00005060 cache_block_group(cache, 1);
Josef Bacik0af3d002010-06-21 14:48:16 -04005061
Chris Masondb945352007-10-15 16:15:53 -04005062 byte_in_group = bytenr - cache->key.objectid;
5063 WARN_ON(byte_in_group > cache->key.offset);
Chris Mason9078a3e2007-04-26 16:46:15 -04005064
Josef Bacik25179202008-10-29 14:49:05 -04005065 spin_lock(&cache->space_info->lock);
Chris Masonc286ac42008-07-22 23:06:41 -04005066 spin_lock(&cache->lock);
Josef Bacik0af3d002010-06-21 14:48:16 -04005067
Josef Bacik73bc1872011-10-03 14:07:49 -04005068 if (btrfs_test_opt(root, SPACE_CACHE) &&
Josef Bacik0af3d002010-06-21 14:48:16 -04005069 cache->disk_cache_state < BTRFS_DC_CLEAR)
5070 cache->disk_cache_state = BTRFS_DC_CLEAR;
5071
Josef Bacik0f9dd462008-09-23 13:14:11 -04005072 cache->dirty = 1;
Chris Mason9078a3e2007-04-26 16:46:15 -04005073 old_val = btrfs_block_group_used(&cache->item);
Chris Masondb945352007-10-15 16:15:53 -04005074 num_bytes = min(total, cache->key.offset - byte_in_group);
Chris Masoncd1bc462007-04-27 10:08:34 -04005075 if (alloc) {
Chris Masondb945352007-10-15 16:15:53 -04005076 old_val += num_bytes;
Yan Zheng11833d62009-09-11 16:11:19 -04005077 btrfs_set_block_group_used(&cache->item, old_val);
5078 cache->reserved -= num_bytes;
Yan Zheng11833d62009-09-11 16:11:19 -04005079 cache->space_info->bytes_reserved -= num_bytes;
Yan, Zhengb742bb82010-05-16 10:46:24 -04005080 cache->space_info->bytes_used += num_bytes;
5081 cache->space_info->disk_used += num_bytes * factor;
Chris Masonc286ac42008-07-22 23:06:41 -04005082 spin_unlock(&cache->lock);
Josef Bacik25179202008-10-29 14:49:05 -04005083 spin_unlock(&cache->space_info->lock);
Chris Masoncd1bc462007-04-27 10:08:34 -04005084 } else {
Chris Masondb945352007-10-15 16:15:53 -04005085 old_val -= num_bytes;
Chris Masonc286ac42008-07-22 23:06:41 -04005086 btrfs_set_block_group_used(&cache->item, old_val);
Yan, Zhengf0486c62010-05-16 10:46:25 -04005087 cache->pinned += num_bytes;
5088 cache->space_info->bytes_pinned += num_bytes;
Chris Mason9078a3e2007-04-26 16:46:15 -04005089 cache->space_info->bytes_used -= num_bytes;
Yan, Zhengb742bb82010-05-16 10:46:24 -04005090 cache->space_info->disk_used -= num_bytes * factor;
Chris Masonc286ac42008-07-22 23:06:41 -04005091 spin_unlock(&cache->lock);
Josef Bacik25179202008-10-29 14:49:05 -04005092 spin_unlock(&cache->space_info->lock);
Liu Hui1f3c79a2009-01-05 15:57:51 -05005093
Yan, Zhengf0486c62010-05-16 10:46:25 -04005094 set_extent_dirty(info->pinned_extents,
5095 bytenr, bytenr + num_bytes - 1,
5096 GFP_NOFS | __GFP_NOFAIL);
Chris Masoncd1bc462007-04-27 10:08:34 -04005097 }
Chris Masonfa9c0d792009-04-03 09:47:43 -04005098 btrfs_put_block_group(cache);
Chris Masondb945352007-10-15 16:15:53 -04005099 total -= num_bytes;
5100 bytenr += num_bytes;
Chris Mason9078a3e2007-04-26 16:46:15 -04005101 }
5102 return 0;
5103}
Chris Mason6324fbf2008-03-24 15:01:59 -04005104
Chris Masona061fc82008-05-07 11:43:44 -04005105static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
5106{
Josef Bacik0f9dd462008-09-23 13:14:11 -04005107 struct btrfs_block_group_cache *cache;
Yan Zhengd2fb3432008-12-11 16:30:39 -05005108 u64 bytenr;
Josef Bacik0f9dd462008-09-23 13:14:11 -04005109
Liu Boa1897fd2012-12-27 09:01:23 +00005110 spin_lock(&root->fs_info->block_group_cache_lock);
5111 bytenr = root->fs_info->first_logical_byte;
5112 spin_unlock(&root->fs_info->block_group_cache_lock);
5113
5114 if (bytenr < (u64)-1)
5115 return bytenr;
5116
Josef Bacik0f9dd462008-09-23 13:14:11 -04005117 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
5118 if (!cache)
Chris Masona061fc82008-05-07 11:43:44 -04005119 return 0;
Josef Bacik0f9dd462008-09-23 13:14:11 -04005120
Yan Zhengd2fb3432008-12-11 16:30:39 -05005121 bytenr = cache->key.objectid;
Chris Masonfa9c0d792009-04-03 09:47:43 -04005122 btrfs_put_block_group(cache);
Yan Zhengd2fb3432008-12-11 16:30:39 -05005123
5124 return bytenr;
Chris Masona061fc82008-05-07 11:43:44 -04005125}
5126
Yan, Zhengf0486c62010-05-16 10:46:25 -04005127static int pin_down_extent(struct btrfs_root *root,
5128 struct btrfs_block_group_cache *cache,
5129 u64 bytenr, u64 num_bytes, int reserved)
Yan324ae4d2007-11-16 14:57:08 -05005130{
Yan Zheng11833d62009-09-11 16:11:19 -04005131 spin_lock(&cache->space_info->lock);
5132 spin_lock(&cache->lock);
5133 cache->pinned += num_bytes;
5134 cache->space_info->bytes_pinned += num_bytes;
5135 if (reserved) {
5136 cache->reserved -= num_bytes;
5137 cache->space_info->bytes_reserved -= num_bytes;
Yan324ae4d2007-11-16 14:57:08 -05005138 }
Yan Zheng11833d62009-09-11 16:11:19 -04005139 spin_unlock(&cache->lock);
5140 spin_unlock(&cache->space_info->lock);
5141
Yan, Zhengf0486c62010-05-16 10:46:25 -04005142 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
5143 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
Yan324ae4d2007-11-16 14:57:08 -05005144 return 0;
5145}
Chris Mason9078a3e2007-04-26 16:46:15 -04005146
Yan, Zhengf0486c62010-05-16 10:46:25 -04005147/*
5148 * this function must be called within transaction
5149 */
5150int btrfs_pin_extent(struct btrfs_root *root,
5151 u64 bytenr, u64 num_bytes, int reserved)
Zheng Yane8569812008-09-26 10:05:48 -04005152{
Yan, Zhengf0486c62010-05-16 10:46:25 -04005153 struct btrfs_block_group_cache *cache;
5154
5155 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005156 BUG_ON(!cache); /* Logic error */
Yan, Zhengf0486c62010-05-16 10:46:25 -04005157
5158 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
5159
5160 btrfs_put_block_group(cache);
Yan Zheng11833d62009-09-11 16:11:19 -04005161 return 0;
5162}
Zheng Yane8569812008-09-26 10:05:48 -04005163
Yan, Zhengf0486c62010-05-16 10:46:25 -04005164/*
Chris Masone688b7252011-10-31 20:52:39 -04005165 * this function must be called within transaction
Yan, Zhengf0486c62010-05-16 10:46:25 -04005166 */
Liu Bodcfac412012-12-27 09:01:20 +00005167int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
Chris Masone688b7252011-10-31 20:52:39 -04005168 u64 bytenr, u64 num_bytes)
Yan, Zhengf0486c62010-05-16 10:46:25 -04005169{
Chris Masone688b7252011-10-31 20:52:39 -04005170 struct btrfs_block_group_cache *cache;
Josef Bacikb50c6e22013-04-25 15:55:30 -04005171 int ret;
Chris Masone688b7252011-10-31 20:52:39 -04005172
5173 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
Josef Bacikb50c6e22013-04-25 15:55:30 -04005174 if (!cache)
5175 return -EINVAL;
Chris Masone688b7252011-10-31 20:52:39 -04005176
5177 /*
5178 * pull in the free space cache (if any) so that our pin
5179 * removes the free space from the cache. We have load_only set
5180 * to one because the slow code to read in the free extents does check
5181 * the pinned extents.
5182 */
Liu Bof6373bf2012-12-27 09:01:18 +00005183 cache_block_group(cache, 1);
Chris Masone688b7252011-10-31 20:52:39 -04005184
5185 pin_down_extent(root, cache, bytenr, num_bytes, 0);
5186
5187 /* remove us from the free space cache (if we're there at all) */
Josef Bacikb50c6e22013-04-25 15:55:30 -04005188 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
Chris Masone688b7252011-10-31 20:52:39 -04005189 btrfs_put_block_group(cache);
Josef Bacikb50c6e22013-04-25 15:55:30 -04005190 return ret;
Chris Masone688b7252011-10-31 20:52:39 -04005191}
5192
Josef Bacikfb25e912011-07-26 17:00:46 -04005193/**
5194 * btrfs_update_reserved_bytes - update the block_group and space info counters
5195 * @cache: The cache we are manipulating
5196 * @num_bytes: The number of bytes in question
5197 * @reserve: One of the reservation enums
5198 *
5199 * This is called by the allocator when it reserves space, or by somebody who is
5200 * freeing space that was never actually used on disk. For example if you
5201 * reserve some space for a new leaf in transaction A and before transaction A
5202 * commits you free that leaf, you call this with reserve set to 0 in order to
5203 * clear the reservation.
5204 *
5205 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
5206 * ENOSPC accounting. For data we handle the reservation through clearing the
5207 * delalloc bits in the io_tree. We have to do this since we could end up
5208 * allocating less disk space for the amount of data we have reserved in the
5209 * case of compression.
5210 *
5211 * If this is a reservation and the block group has become read only we cannot
5212 * make the reservation and return -EAGAIN, otherwise this function always
5213 * succeeds.
Yan, Zhengf0486c62010-05-16 10:46:25 -04005214 */
Josef Bacikfb25e912011-07-26 17:00:46 -04005215static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
5216 u64 num_bytes, int reserve)
Yan, Zhengf0486c62010-05-16 10:46:25 -04005217{
Josef Bacikfb25e912011-07-26 17:00:46 -04005218 struct btrfs_space_info *space_info = cache->space_info;
Yan, Zhengf0486c62010-05-16 10:46:25 -04005219 int ret = 0;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005220
Josef Bacikfb25e912011-07-26 17:00:46 -04005221 spin_lock(&space_info->lock);
5222 spin_lock(&cache->lock);
5223 if (reserve != RESERVE_FREE) {
Yan, Zhengf0486c62010-05-16 10:46:25 -04005224 if (cache->ro) {
5225 ret = -EAGAIN;
5226 } else {
Josef Bacikfb25e912011-07-26 17:00:46 -04005227 cache->reserved += num_bytes;
5228 space_info->bytes_reserved += num_bytes;
5229 if (reserve == RESERVE_ALLOC) {
Josef Bacik8c2a3ca2012-01-10 10:31:31 -05005230 trace_btrfs_space_reservation(cache->fs_info,
Liu Bo2bcc0322012-03-29 09:57:44 -04005231 "space_info", space_info->flags,
5232 num_bytes, 0);
Josef Bacikfb25e912011-07-26 17:00:46 -04005233 space_info->bytes_may_use -= num_bytes;
5234 }
Yan, Zhengf0486c62010-05-16 10:46:25 -04005235 }
Josef Bacikfb25e912011-07-26 17:00:46 -04005236 } else {
5237 if (cache->ro)
5238 space_info->bytes_readonly += num_bytes;
5239 cache->reserved -= num_bytes;
5240 space_info->bytes_reserved -= num_bytes;
5241 space_info->reservation_progress++;
Yan, Zhengf0486c62010-05-16 10:46:25 -04005242 }
Josef Bacikfb25e912011-07-26 17:00:46 -04005243 spin_unlock(&cache->lock);
5244 spin_unlock(&space_info->lock);
Yan, Zhengf0486c62010-05-16 10:46:25 -04005245 return ret;
5246}
5247
Jeff Mahoney143bede2012-03-01 14:56:26 +01005248void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
Yan Zheng11833d62009-09-11 16:11:19 -04005249 struct btrfs_root *root)
5250{
5251 struct btrfs_fs_info *fs_info = root->fs_info;
5252 struct btrfs_caching_control *next;
5253 struct btrfs_caching_control *caching_ctl;
5254 struct btrfs_block_group_cache *cache;
5255
5256 down_write(&fs_info->extent_commit_sem);
5257
5258 list_for_each_entry_safe(caching_ctl, next,
5259 &fs_info->caching_block_groups, list) {
5260 cache = caching_ctl->block_group;
5261 if (block_group_cache_done(cache)) {
5262 cache->last_byte_to_unpin = (u64)-1;
5263 list_del_init(&caching_ctl->list);
5264 put_caching_control(caching_ctl);
5265 } else {
5266 cache->last_byte_to_unpin = caching_ctl->progress;
5267 }
5268 }
5269
5270 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5271 fs_info->pinned_extents = &fs_info->freed_extents[1];
5272 else
5273 fs_info->pinned_extents = &fs_info->freed_extents[0];
5274
5275 up_write(&fs_info->extent_commit_sem);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04005276
5277 update_global_block_rsv(fs_info);
Yan Zheng11833d62009-09-11 16:11:19 -04005278}
5279
Filipe Manana5f55aee2014-12-07 21:31:47 +00005280static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end,
5281 const bool return_free_space)
Yan Zheng11833d62009-09-11 16:11:19 -04005282{
5283 struct btrfs_fs_info *fs_info = root->fs_info;
5284 struct btrfs_block_group_cache *cache = NULL;
Josef Bacik7b398f82012-10-22 15:52:28 -04005285 struct btrfs_space_info *space_info;
5286 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
Yan Zheng11833d62009-09-11 16:11:19 -04005287 u64 len;
Josef Bacik7b398f82012-10-22 15:52:28 -04005288 bool readonly;
Yan Zheng11833d62009-09-11 16:11:19 -04005289
5290 while (start <= end) {
Josef Bacik7b398f82012-10-22 15:52:28 -04005291 readonly = false;
Yan Zheng11833d62009-09-11 16:11:19 -04005292 if (!cache ||
5293 start >= cache->key.objectid + cache->key.offset) {
5294 if (cache)
5295 btrfs_put_block_group(cache);
5296 cache = btrfs_lookup_block_group(fs_info, start);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005297 BUG_ON(!cache); /* Logic error */
Yan Zheng11833d62009-09-11 16:11:19 -04005298 }
5299
5300 len = cache->key.objectid + cache->key.offset - start;
5301 len = min(len, end + 1 - start);
5302
5303 if (start < cache->last_byte_to_unpin) {
5304 len = min(len, cache->last_byte_to_unpin - start);
Filipe Manana5f55aee2014-12-07 21:31:47 +00005305 if (return_free_space)
5306 btrfs_add_free_space(cache, start, len);
Yan Zheng11833d62009-09-11 16:11:19 -04005307 }
Josef Bacik25179202008-10-29 14:49:05 -04005308
Yan, Zhengf0486c62010-05-16 10:46:25 -04005309 start += len;
Josef Bacik7b398f82012-10-22 15:52:28 -04005310 space_info = cache->space_info;
Yan, Zhengf0486c62010-05-16 10:46:25 -04005311
Josef Bacik7b398f82012-10-22 15:52:28 -04005312 spin_lock(&space_info->lock);
Josef Bacik25179202008-10-29 14:49:05 -04005313 spin_lock(&cache->lock);
Yan Zheng11833d62009-09-11 16:11:19 -04005314 cache->pinned -= len;
Josef Bacik7b398f82012-10-22 15:52:28 -04005315 space_info->bytes_pinned -= len;
5316 if (cache->ro) {
5317 space_info->bytes_readonly += len;
5318 readonly = true;
5319 }
Josef Bacik25179202008-10-29 14:49:05 -04005320 spin_unlock(&cache->lock);
Josef Bacik7b398f82012-10-22 15:52:28 -04005321 if (!readonly && global_rsv->space_info == space_info) {
5322 spin_lock(&global_rsv->lock);
5323 if (!global_rsv->full) {
5324 len = min(len, global_rsv->size -
5325 global_rsv->reserved);
5326 global_rsv->reserved += len;
5327 space_info->bytes_may_use += len;
5328 if (global_rsv->reserved >= global_rsv->size)
5329 global_rsv->full = 1;
5330 }
5331 spin_unlock(&global_rsv->lock);
5332 }
5333 spin_unlock(&space_info->lock);
Yan Zheng11833d62009-09-11 16:11:19 -04005334 }
5335
5336 if (cache)
Chris Masonfa9c0d792009-04-03 09:47:43 -04005337 btrfs_put_block_group(cache);
Chris Masonccd467d2007-06-28 15:57:36 -04005338 return 0;
5339}
5340
5341int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
Yan Zheng11833d62009-09-11 16:11:19 -04005342 struct btrfs_root *root)
Chris Masona28ec192007-03-06 20:08:01 -05005343{
Yan Zheng11833d62009-09-11 16:11:19 -04005344 struct btrfs_fs_info *fs_info = root->fs_info;
5345 struct extent_io_tree *unpin;
Chris Mason1a5bc162007-10-15 16:15:26 -04005346 u64 start;
5347 u64 end;
Chris Masona28ec192007-03-06 20:08:01 -05005348 int ret;
Chris Masona28ec192007-03-06 20:08:01 -05005349
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005350 if (trans->aborted)
5351 return 0;
5352
Yan Zheng11833d62009-09-11 16:11:19 -04005353 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5354 unpin = &fs_info->freed_extents[1];
5355 else
5356 unpin = &fs_info->freed_extents[0];
5357
Chris Masond3977122009-01-05 21:25:51 -05005358 while (1) {
Chris Mason1a5bc162007-10-15 16:15:26 -04005359 ret = find_first_extent_bit(unpin, 0, &start, &end,
Josef Bacike6138872012-09-27 17:07:30 -04005360 EXTENT_DIRTY, NULL);
Chris Mason1a5bc162007-10-15 16:15:26 -04005361 if (ret)
Chris Masona28ec192007-03-06 20:08:01 -05005362 break;
Liu Hui1f3c79a2009-01-05 15:57:51 -05005363
Li Dongyang5378e602011-03-24 10:24:27 +00005364 if (btrfs_test_opt(root, DISCARD))
5365 ret = btrfs_discard_extent(root, start,
5366 end + 1 - start, NULL);
Liu Hui1f3c79a2009-01-05 15:57:51 -05005367
Chris Mason1a5bc162007-10-15 16:15:26 -04005368 clear_extent_dirty(unpin, start, end, GFP_NOFS);
Filipe Manana5f55aee2014-12-07 21:31:47 +00005369 unpin_extent_range(root, start, end, true);
Chris Masonb9473432009-03-13 11:00:37 -04005370 cond_resched();
Chris Masona28ec192007-03-06 20:08:01 -05005371 }
Josef Bacik817d52f2009-07-13 21:29:25 -04005372
Chris Masone20d96d2007-03-22 12:13:20 -04005373 return 0;
5374}
5375
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005376static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
5377 struct btrfs_root *root,
5378 u64 bytenr, u64 num_bytes, u64 parent,
5379 u64 root_objectid, u64 owner_objectid,
5380 u64 owner_offset, int refs_to_drop,
5381 struct btrfs_delayed_extent_op *extent_op)
Chris Masona28ec192007-03-06 20:08:01 -05005382{
Chris Masone2fa7222007-03-12 16:22:34 -04005383 struct btrfs_key key;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005384 struct btrfs_path *path;
Chris Mason1261ec42007-03-20 20:35:03 -04005385 struct btrfs_fs_info *info = root->fs_info;
5386 struct btrfs_root *extent_root = info->extent_root;
Chris Mason5f39d392007-10-15 16:14:19 -04005387 struct extent_buffer *leaf;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005388 struct btrfs_extent_item *ei;
5389 struct btrfs_extent_inline_ref *iref;
Chris Masona28ec192007-03-06 20:08:01 -05005390 int ret;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005391 int is_data;
Chris Mason952fcca2008-02-18 16:33:44 -05005392 int extent_slot = 0;
5393 int found_extent = 0;
5394 int num_to_del = 1;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005395 u32 item_size;
5396 u64 refs;
Josef Bacik3173a182013-03-07 14:22:04 -05005397 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
5398 SKINNY_METADATA);
Chris Mason037e6392007-03-07 11:50:24 -05005399
Chris Mason5caf2a02007-04-02 11:20:42 -04005400 path = btrfs_alloc_path();
Chris Mason54aa1f42007-06-22 14:16:25 -04005401 if (!path)
5402 return -ENOMEM;
5403
Chris Mason3c12ac72008-04-21 12:01:38 -04005404 path->reada = 1;
Chris Masonb9473432009-03-13 11:00:37 -04005405 path->leave_spinning = 1;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005406
5407 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
5408 BUG_ON(!is_data && refs_to_drop != 1);
5409
Josef Bacik3173a182013-03-07 14:22:04 -05005410 if (is_data)
5411 skinny_metadata = 0;
5412
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005413 ret = lookup_extent_backref(trans, extent_root, path, &iref,
5414 bytenr, num_bytes, parent,
5415 root_objectid, owner_objectid,
5416 owner_offset);
Chris Mason7bb86312007-12-11 09:25:06 -05005417 if (ret == 0) {
Chris Mason952fcca2008-02-18 16:33:44 -05005418 extent_slot = path->slots[0];
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005419 while (extent_slot >= 0) {
5420 btrfs_item_key_to_cpu(path->nodes[0], &key,
Chris Mason952fcca2008-02-18 16:33:44 -05005421 extent_slot);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005422 if (key.objectid != bytenr)
Chris Mason952fcca2008-02-18 16:33:44 -05005423 break;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005424 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
5425 key.offset == num_bytes) {
Chris Mason952fcca2008-02-18 16:33:44 -05005426 found_extent = 1;
5427 break;
5428 }
Josef Bacik3173a182013-03-07 14:22:04 -05005429 if (key.type == BTRFS_METADATA_ITEM_KEY &&
5430 key.offset == owner_objectid) {
5431 found_extent = 1;
5432 break;
5433 }
Chris Mason952fcca2008-02-18 16:33:44 -05005434 if (path->slots[0] - extent_slot > 5)
5435 break;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005436 extent_slot--;
Chris Mason952fcca2008-02-18 16:33:44 -05005437 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005438#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5439 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
5440 if (found_extent && item_size < sizeof(*ei))
5441 found_extent = 0;
5442#endif
Zheng Yan31840ae2008-09-23 13:14:14 -04005443 if (!found_extent) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005444 BUG_ON(iref);
Chris Mason56bec292009-03-13 10:10:06 -04005445 ret = remove_extent_backref(trans, extent_root, path,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005446 NULL, refs_to_drop,
5447 is_data);
David Sterba005d6422012-09-18 07:52:32 -06005448 if (ret) {
5449 btrfs_abort_transaction(trans, extent_root, ret);
5450 goto out;
5451 }
David Sterbab3b4aa72011-04-21 01:20:15 +02005452 btrfs_release_path(path);
Chris Masonb9473432009-03-13 11:00:37 -04005453 path->leave_spinning = 1;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005454
5455 key.objectid = bytenr;
5456 key.type = BTRFS_EXTENT_ITEM_KEY;
5457 key.offset = num_bytes;
5458
Josef Bacik3173a182013-03-07 14:22:04 -05005459 if (!is_data && skinny_metadata) {
5460 key.type = BTRFS_METADATA_ITEM_KEY;
5461 key.offset = owner_objectid;
5462 }
5463
Zheng Yan31840ae2008-09-23 13:14:14 -04005464 ret = btrfs_search_slot(trans, extent_root,
5465 &key, path, -1, 1);
Josef Bacik3173a182013-03-07 14:22:04 -05005466 if (ret > 0 && skinny_metadata && path->slots[0]) {
5467 /*
5468 * Couldn't find our skinny metadata item,
5469 * see if we have ye olde extent item.
5470 */
5471 path->slots[0]--;
5472 btrfs_item_key_to_cpu(path->nodes[0], &key,
5473 path->slots[0]);
5474 if (key.objectid == bytenr &&
5475 key.type == BTRFS_EXTENT_ITEM_KEY &&
5476 key.offset == num_bytes)
5477 ret = 0;
5478 }
5479
5480 if (ret > 0 && skinny_metadata) {
5481 skinny_metadata = false;
5482 key.type = BTRFS_EXTENT_ITEM_KEY;
5483 key.offset = num_bytes;
5484 btrfs_release_path(path);
5485 ret = btrfs_search_slot(trans, extent_root,
5486 &key, path, -1, 1);
5487 }
5488
Josef Bacikf3465ca2008-11-12 14:19:50 -05005489 if (ret) {
Simon Kirbyc2cf52e2013-03-19 22:41:23 +00005490 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
5491 ret, (unsigned long long)bytenr);
Josef Bacikb783e622011-07-13 15:03:50 +00005492 if (ret > 0)
5493 btrfs_print_leaf(extent_root,
5494 path->nodes[0]);
Josef Bacikf3465ca2008-11-12 14:19:50 -05005495 }
David Sterba005d6422012-09-18 07:52:32 -06005496 if (ret < 0) {
5497 btrfs_abort_transaction(trans, extent_root, ret);
5498 goto out;
5499 }
Zheng Yan31840ae2008-09-23 13:14:14 -04005500 extent_slot = path->slots[0];
5501 }
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005502 } else if (ret == -ENOENT) {
Chris Mason7bb86312007-12-11 09:25:06 -05005503 btrfs_print_leaf(extent_root, path->nodes[0]);
5504 WARN_ON(1);
Simon Kirbyc2cf52e2013-03-19 22:41:23 +00005505 btrfs_err(info,
5506 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
5507 (unsigned long long)bytenr,
5508 (unsigned long long)parent,
5509 (unsigned long long)root_objectid,
5510 (unsigned long long)owner_objectid,
5511 (unsigned long long)owner_offset);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005512 } else {
David Sterba005d6422012-09-18 07:52:32 -06005513 btrfs_abort_transaction(trans, extent_root, ret);
5514 goto out;
Chris Mason7bb86312007-12-11 09:25:06 -05005515 }
Chris Mason5f39d392007-10-15 16:14:19 -04005516
5517 leaf = path->nodes[0];
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005518 item_size = btrfs_item_size_nr(leaf, extent_slot);
5519#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5520 if (item_size < sizeof(*ei)) {
5521 BUG_ON(found_extent || extent_slot != path->slots[0]);
5522 ret = convert_extent_item_v0(trans, extent_root, path,
5523 owner_objectid, 0);
David Sterba005d6422012-09-18 07:52:32 -06005524 if (ret < 0) {
5525 btrfs_abort_transaction(trans, extent_root, ret);
5526 goto out;
5527 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005528
David Sterbab3b4aa72011-04-21 01:20:15 +02005529 btrfs_release_path(path);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005530 path->leave_spinning = 1;
5531
5532 key.objectid = bytenr;
5533 key.type = BTRFS_EXTENT_ITEM_KEY;
5534 key.offset = num_bytes;
5535
5536 ret = btrfs_search_slot(trans, extent_root, &key, path,
5537 -1, 1);
5538 if (ret) {
Simon Kirbyc2cf52e2013-03-19 22:41:23 +00005539 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
5540 ret, (unsigned long long)bytenr);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005541 btrfs_print_leaf(extent_root, path->nodes[0]);
5542 }
David Sterba005d6422012-09-18 07:52:32 -06005543 if (ret < 0) {
5544 btrfs_abort_transaction(trans, extent_root, ret);
5545 goto out;
5546 }
5547
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005548 extent_slot = path->slots[0];
5549 leaf = path->nodes[0];
5550 item_size = btrfs_item_size_nr(leaf, extent_slot);
5551 }
5552#endif
5553 BUG_ON(item_size < sizeof(*ei));
Chris Mason952fcca2008-02-18 16:33:44 -05005554 ei = btrfs_item_ptr(leaf, extent_slot,
Chris Mason123abc82007-03-14 14:14:43 -04005555 struct btrfs_extent_item);
Josef Bacik3173a182013-03-07 14:22:04 -05005556 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
5557 key.type == BTRFS_EXTENT_ITEM_KEY) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005558 struct btrfs_tree_block_info *bi;
5559 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
5560 bi = (struct btrfs_tree_block_info *)(ei + 1);
5561 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
Chris Mason952fcca2008-02-18 16:33:44 -05005562 }
5563
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005564 refs = btrfs_extent_refs(leaf, ei);
Josef Bacik32b02532013-04-24 16:38:50 -04005565 if (refs < refs_to_drop) {
5566 btrfs_err(info, "trying to drop %d refs but we only have %Lu "
5567 "for bytenr %Lu\n", refs_to_drop, refs, bytenr);
5568 ret = -EINVAL;
5569 btrfs_abort_transaction(trans, extent_root, ret);
5570 goto out;
5571 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005572 refs -= refs_to_drop;
5573
5574 if (refs > 0) {
5575 if (extent_op)
5576 __run_delayed_extent_op(extent_op, leaf, ei);
5577 /*
5578 * In the case of inline back ref, reference count will
5579 * be updated by remove_extent_backref
5580 */
5581 if (iref) {
5582 BUG_ON(!found_extent);
5583 } else {
5584 btrfs_set_extent_refs(leaf, ei, refs);
5585 btrfs_mark_buffer_dirty(leaf);
5586 }
5587 if (found_extent) {
5588 ret = remove_extent_backref(trans, extent_root, path,
5589 iref, refs_to_drop,
5590 is_data);
David Sterba005d6422012-09-18 07:52:32 -06005591 if (ret) {
5592 btrfs_abort_transaction(trans, extent_root, ret);
5593 goto out;
5594 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005595 }
5596 } else {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005597 if (found_extent) {
5598 BUG_ON(is_data && refs_to_drop !=
5599 extent_data_ref_count(root, path, iref));
5600 if (iref) {
5601 BUG_ON(path->slots[0] != extent_slot);
5602 } else {
5603 BUG_ON(path->slots[0] != extent_slot + 1);
5604 path->slots[0] = extent_slot;
5605 num_to_del = 2;
5606 }
Chris Mason78fae272007-03-25 11:35:08 -04005607 }
Chris Masonb9473432009-03-13 11:00:37 -04005608
Chris Mason952fcca2008-02-18 16:33:44 -05005609 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
5610 num_to_del);
David Sterba005d6422012-09-18 07:52:32 -06005611 if (ret) {
5612 btrfs_abort_transaction(trans, extent_root, ret);
5613 goto out;
5614 }
David Sterbab3b4aa72011-04-21 01:20:15 +02005615 btrfs_release_path(path);
David Woodhouse21af8042008-08-12 14:13:26 +01005616
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005617 if (is_data) {
Chris Mason459931e2008-12-10 09:10:46 -05005618 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
David Sterba005d6422012-09-18 07:52:32 -06005619 if (ret) {
5620 btrfs_abort_transaction(trans, extent_root, ret);
5621 goto out;
5622 }
Chris Mason459931e2008-12-10 09:10:46 -05005623 }
5624
Liu Boc53d6132012-12-27 09:01:19 +00005625 ret = update_block_group(root, bytenr, num_bytes, 0);
David Sterba005d6422012-09-18 07:52:32 -06005626 if (ret) {
5627 btrfs_abort_transaction(trans, extent_root, ret);
5628 goto out;
5629 }
Chris Masona28ec192007-03-06 20:08:01 -05005630 }
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005631out:
Chris Mason5caf2a02007-04-02 11:20:42 -04005632 btrfs_free_path(path);
Chris Masona28ec192007-03-06 20:08:01 -05005633 return ret;
5634}
5635
5636/*
Yan, Zhengf0486c62010-05-16 10:46:25 -04005637 * when we free an block, it is possible (and likely) that we free the last
Chris Mason1887be62009-03-13 10:11:24 -04005638 * delayed ref for that extent as well. This searches the delayed ref tree for
5639 * a given extent, and if there are no other delayed refs to be processed, it
5640 * removes it from the tree.
5641 */
5642static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
5643 struct btrfs_root *root, u64 bytenr)
5644{
5645 struct btrfs_delayed_ref_head *head;
5646 struct btrfs_delayed_ref_root *delayed_refs;
5647 struct btrfs_delayed_ref_node *ref;
5648 struct rb_node *node;
Yan, Zhengf0486c62010-05-16 10:46:25 -04005649 int ret = 0;
Chris Mason1887be62009-03-13 10:11:24 -04005650
5651 delayed_refs = &trans->transaction->delayed_refs;
5652 spin_lock(&delayed_refs->lock);
5653 head = btrfs_find_delayed_ref_head(trans, bytenr);
5654 if (!head)
5655 goto out;
5656
5657 node = rb_prev(&head->node.rb_node);
5658 if (!node)
5659 goto out;
5660
5661 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
5662
5663 /* there are still entries for this ref, we can't drop it */
5664 if (ref->bytenr == bytenr)
5665 goto out;
5666
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005667 if (head->extent_op) {
5668 if (!head->must_insert_reserved)
5669 goto out;
Miao Xie78a61842012-11-21 02:21:28 +00005670 btrfs_free_delayed_extent_op(head->extent_op);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005671 head->extent_op = NULL;
5672 }
5673
Chris Mason1887be62009-03-13 10:11:24 -04005674 /*
5675 * waiting for the lock here would deadlock. If someone else has it
5676 * locked they are already in the process of dropping it anyway
5677 */
5678 if (!mutex_trylock(&head->mutex))
5679 goto out;
5680
5681 /*
5682 * at this point we have a head with no other entries. Go
5683 * ahead and process it.
5684 */
5685 head->node.in_tree = 0;
5686 rb_erase(&head->node.rb_node, &delayed_refs->root);
Chris Masonc3e69d52009-03-13 10:17:05 -04005687
Chris Mason1887be62009-03-13 10:11:24 -04005688 delayed_refs->num_entries--;
5689
5690 /*
5691 * we don't take a ref on the node because we're removing it from the
5692 * tree, so we just steal the ref the tree was holding.
5693 */
Chris Masonc3e69d52009-03-13 10:17:05 -04005694 delayed_refs->num_heads--;
5695 if (list_empty(&head->cluster))
5696 delayed_refs->num_heads_ready--;
5697
5698 list_del_init(&head->cluster);
Chris Mason1887be62009-03-13 10:11:24 -04005699 spin_unlock(&delayed_refs->lock);
5700
Yan, Zhengf0486c62010-05-16 10:46:25 -04005701 BUG_ON(head->extent_op);
5702 if (head->must_insert_reserved)
5703 ret = 1;
5704
5705 mutex_unlock(&head->mutex);
Chris Mason1887be62009-03-13 10:11:24 -04005706 btrfs_put_delayed_ref(&head->node);
Yan, Zhengf0486c62010-05-16 10:46:25 -04005707 return ret;
Chris Mason1887be62009-03-13 10:11:24 -04005708out:
5709 spin_unlock(&delayed_refs->lock);
5710 return 0;
5711}
5712
Yan, Zhengf0486c62010-05-16 10:46:25 -04005713void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
5714 struct btrfs_root *root,
5715 struct extent_buffer *buf,
Jan Schmidt5581a512012-05-16 17:04:52 +02005716 u64 parent, int last_ref)
Yan, Zhengf0486c62010-05-16 10:46:25 -04005717{
Yan, Zhengf0486c62010-05-16 10:46:25 -04005718 struct btrfs_block_group_cache *cache = NULL;
5719 int ret;
5720
5721 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
Arne Jansen66d7e7f2011-09-12 15:26:38 +02005722 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
5723 buf->start, buf->len,
5724 parent, root->root_key.objectid,
5725 btrfs_header_level(buf),
Jan Schmidt5581a512012-05-16 17:04:52 +02005726 BTRFS_DROP_DELAYED_REF, NULL, 0);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005727 BUG_ON(ret); /* -ENOMEM */
Yan, Zhengf0486c62010-05-16 10:46:25 -04005728 }
5729
5730 if (!last_ref)
5731 return;
5732
Yan, Zhengf0486c62010-05-16 10:46:25 -04005733 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
Yan, Zhengf0486c62010-05-16 10:46:25 -04005734
5735 if (btrfs_header_generation(buf) == trans->transid) {
5736 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5737 ret = check_ref_cleanup(trans, root, buf->start);
5738 if (!ret)
Josef Bacik37be25b2011-08-05 10:25:38 -04005739 goto out;
Yan, Zhengf0486c62010-05-16 10:46:25 -04005740 }
5741
5742 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
5743 pin_down_extent(root, cache, buf->start, buf->len, 1);
Josef Bacik37be25b2011-08-05 10:25:38 -04005744 goto out;
Yan, Zhengf0486c62010-05-16 10:46:25 -04005745 }
5746
5747 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
5748
5749 btrfs_add_free_space(cache, buf->start, buf->len);
Josef Bacikfb25e912011-07-26 17:00:46 -04005750 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE);
Yan, Zhengf0486c62010-05-16 10:46:25 -04005751 }
5752out:
Josef Bacika826d6d2011-03-16 13:42:43 -04005753 /*
5754 * Deleting the buffer, clear the corrupt flag since it doesn't matter
5755 * anymore.
5756 */
5757 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
Yan, Zhengf0486c62010-05-16 10:46:25 -04005758 btrfs_put_block_group(cache);
5759}
5760
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005761/* Can return -ENOMEM */
Arne Jansen66d7e7f2011-09-12 15:26:38 +02005762int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
5763 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
5764 u64 owner, u64 offset, int for_cow)
Chris Mason925baed2008-06-25 16:01:30 -04005765{
5766 int ret;
Arne Jansen66d7e7f2011-09-12 15:26:38 +02005767 struct btrfs_fs_info *fs_info = root->fs_info;
Chris Mason925baed2008-06-25 16:01:30 -04005768
Chris Mason56bec292009-03-13 10:10:06 -04005769 /*
5770 * tree log blocks never actually go into the extent allocation
5771 * tree, just update pinning info and exit early.
Chris Mason56bec292009-03-13 10:10:06 -04005772 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005773 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
5774 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
Chris Masonb9473432009-03-13 11:00:37 -04005775 /* unlocks the pinned mutex */
Yan Zheng11833d62009-09-11 16:11:19 -04005776 btrfs_pin_extent(root, bytenr, num_bytes, 1);
Chris Mason56bec292009-03-13 10:10:06 -04005777 ret = 0;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005778 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
Arne Jansen66d7e7f2011-09-12 15:26:38 +02005779 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
5780 num_bytes,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005781 parent, root_objectid, (int)owner,
Arne Jansen66d7e7f2011-09-12 15:26:38 +02005782 BTRFS_DROP_DELAYED_REF, NULL, for_cow);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005783 } else {
Arne Jansen66d7e7f2011-09-12 15:26:38 +02005784 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
5785 num_bytes,
5786 parent, root_objectid, owner,
5787 offset, BTRFS_DROP_DELAYED_REF,
5788 NULL, for_cow);
Chris Mason56bec292009-03-13 10:10:06 -04005789 }
Chris Mason925baed2008-06-25 16:01:30 -04005790 return ret;
5791}
5792
David Woodhouse53b381b2013-01-29 18:40:14 -05005793static u64 stripe_align(struct btrfs_root *root,
5794 struct btrfs_block_group_cache *cache,
5795 u64 val, u64 num_bytes)
Chris Mason87ee04e2007-11-30 11:30:34 -05005796{
Qu Wenruofda28322013-02-26 08:10:22 +00005797 u64 ret = ALIGN(val, root->stripesize);
Chris Mason87ee04e2007-11-30 11:30:34 -05005798 return ret;
5799}
5800
Chris Masonfec577f2007-02-26 10:40:21 -05005801/*
Josef Bacik817d52f2009-07-13 21:29:25 -04005802 * when we wait for progress in the block group caching, its because
5803 * our allocation attempt failed at least once. So, we must sleep
5804 * and let some progress happen before we try again.
5805 *
5806 * This function will sleep at least once waiting for new free space to
5807 * show up, and then it will check the block group free space numbers
5808 * for our min num_bytes. Another option is to have it go ahead
5809 * and look in the rbtree for a free extent of a given size, but this
5810 * is a good start.
5811 */
5812static noinline int
5813wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
5814 u64 num_bytes)
5815{
Yan Zheng11833d62009-09-11 16:11:19 -04005816 struct btrfs_caching_control *caching_ctl;
Josef Bacik817d52f2009-07-13 21:29:25 -04005817
Yan Zheng11833d62009-09-11 16:11:19 -04005818 caching_ctl = get_caching_control(cache);
5819 if (!caching_ctl)
Josef Bacik817d52f2009-07-13 21:29:25 -04005820 return 0;
Josef Bacik817d52f2009-07-13 21:29:25 -04005821
Yan Zheng11833d62009-09-11 16:11:19 -04005822 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
Li Zefan34d52cb2011-03-29 13:46:06 +08005823 (cache->free_space_ctl->free_space >= num_bytes));
Yan Zheng11833d62009-09-11 16:11:19 -04005824
5825 put_caching_control(caching_ctl);
5826 return 0;
5827}
5828
5829static noinline int
5830wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
5831{
5832 struct btrfs_caching_control *caching_ctl;
Yan Zheng11833d62009-09-11 16:11:19 -04005833
5834 caching_ctl = get_caching_control(cache);
5835 if (!caching_ctl)
5836 return 0;
5837
5838 wait_event(caching_ctl->wait, block_group_cache_done(cache));
5839
5840 put_caching_control(caching_ctl);
Josef Bacik817d52f2009-07-13 21:29:25 -04005841 return 0;
5842}
5843
Liu Bo31e50222012-11-21 14:18:10 +00005844int __get_raid_index(u64 flags)
Yan, Zhengb742bb82010-05-16 10:46:24 -04005845{
Ilya Dryomov7738a532012-03-27 17:09:17 +03005846 if (flags & BTRFS_BLOCK_GROUP_RAID10)
Miao Xiee6ec7162013-01-17 05:38:51 +00005847 return BTRFS_RAID_RAID10;
Ilya Dryomov7738a532012-03-27 17:09:17 +03005848 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
Miao Xiee6ec7162013-01-17 05:38:51 +00005849 return BTRFS_RAID_RAID1;
Ilya Dryomov7738a532012-03-27 17:09:17 +03005850 else if (flags & BTRFS_BLOCK_GROUP_DUP)
Miao Xiee6ec7162013-01-17 05:38:51 +00005851 return BTRFS_RAID_DUP;
Ilya Dryomov7738a532012-03-27 17:09:17 +03005852 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
Miao Xiee6ec7162013-01-17 05:38:51 +00005853 return BTRFS_RAID_RAID0;
David Woodhouse53b381b2013-01-29 18:40:14 -05005854 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
Chris Masone942f882013-02-20 14:06:05 -05005855 return BTRFS_RAID_RAID5;
David Woodhouse53b381b2013-01-29 18:40:14 -05005856 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
Chris Masone942f882013-02-20 14:06:05 -05005857 return BTRFS_RAID_RAID6;
Ilya Dryomov7738a532012-03-27 17:09:17 +03005858
Chris Masone942f882013-02-20 14:06:05 -05005859 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
Yan, Zhengb742bb82010-05-16 10:46:24 -04005860}
5861
Ilya Dryomov7738a532012-03-27 17:09:17 +03005862static int get_block_group_index(struct btrfs_block_group_cache *cache)
5863{
Liu Bo31e50222012-11-21 14:18:10 +00005864 return __get_raid_index(cache->flags);
Ilya Dryomov7738a532012-03-27 17:09:17 +03005865}
5866
Josef Bacik817d52f2009-07-13 21:29:25 -04005867enum btrfs_loop_type {
Josef Bacik285ff5a2012-01-13 15:27:45 -05005868 LOOP_CACHING_NOWAIT = 0,
5869 LOOP_CACHING_WAIT = 1,
5870 LOOP_ALLOC_CHUNK = 2,
5871 LOOP_NO_EMPTY_SIZE = 3,
Josef Bacik817d52f2009-07-13 21:29:25 -04005872};
5873
5874/*
Chris Masonfec577f2007-02-26 10:40:21 -05005875 * walks the btree of allocated extents and find a hole of a given size.
5876 * The key ins is changed to record the hole:
5877 * ins->objectid == block start
Chris Mason62e27492007-03-15 12:56:47 -04005878 * ins->flags = BTRFS_EXTENT_ITEM_KEY
Chris Masonfec577f2007-02-26 10:40:21 -05005879 * ins->offset == number of blocks
5880 * Any available blocks before search_start are skipped.
5881 */
Chris Masond3977122009-01-05 21:25:51 -05005882static noinline int find_free_extent(struct btrfs_trans_handle *trans,
Chris Mason98ed5172008-01-03 10:01:48 -05005883 struct btrfs_root *orig_root,
5884 u64 num_bytes, u64 empty_size,
Chris Mason98ed5172008-01-03 10:01:48 -05005885 u64 hint_byte, struct btrfs_key *ins,
David Sterbab6919a52013-04-29 13:39:40 +00005886 u64 flags)
Chris Masonfec577f2007-02-26 10:40:21 -05005887{
Josef Bacik80eb2342008-10-29 14:49:05 -04005888 int ret = 0;
Chris Masond3977122009-01-05 21:25:51 -05005889 struct btrfs_root *root = orig_root->fs_info->extent_root;
Chris Masonfa9c0d792009-04-03 09:47:43 -04005890 struct btrfs_free_cluster *last_ptr = NULL;
Josef Bacik80eb2342008-10-29 14:49:05 -04005891 struct btrfs_block_group_cache *block_group = NULL;
Alexandre Oliva274bd4f2011-12-07 20:08:40 -05005892 struct btrfs_block_group_cache *used_block_group;
Josef Bacik81c9ad22012-01-18 10:56:06 -05005893 u64 search_start = 0;
Chris Mason239b14b2008-03-24 15:02:07 -04005894 int empty_cluster = 2 * 1024 * 1024;
Josef Bacik80eb2342008-10-29 14:49:05 -04005895 struct btrfs_space_info *space_info;
Chris Masonfa9c0d792009-04-03 09:47:43 -04005896 int loop = 0;
David Sterbab6919a52013-04-29 13:39:40 +00005897 int index = __get_raid_index(flags);
5898 int alloc_type = (flags & BTRFS_BLOCK_GROUP_DATA) ?
Josef Bacikfb25e912011-07-26 17:00:46 -04005899 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
Josef Bacik817d52f2009-07-13 21:29:25 -04005900 bool found_uncached_bg = false;
Josef Bacik0a243252009-09-11 16:11:20 -04005901 bool failed_cluster_refill = false;
Josef Bacik1cdda9b2009-10-06 10:04:28 -04005902 bool failed_alloc = false;
Josef Bacik67377732010-09-16 16:19:09 -04005903 bool use_cluster = true;
Miao Xie60d2adb2011-09-09 17:34:35 +08005904 bool have_caching_bg = false;
Chris Masonfec577f2007-02-26 10:40:21 -05005905
Chris Masondb945352007-10-15 16:15:53 -04005906 WARN_ON(num_bytes < root->sectorsize);
Chris Masonb1a4d962007-04-04 15:27:52 -04005907 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
Josef Bacik80eb2342008-10-29 14:49:05 -04005908 ins->objectid = 0;
5909 ins->offset = 0;
Chris Masonb1a4d962007-04-04 15:27:52 -04005910
David Sterbab6919a52013-04-29 13:39:40 +00005911 trace_find_free_extent(orig_root, num_bytes, empty_size, flags);
Josef Bacik3f7de032011-11-10 08:29:20 -05005912
David Sterbab6919a52013-04-29 13:39:40 +00005913 space_info = __find_space_info(root->fs_info, flags);
Josef Bacik1b1d1f62010-03-19 20:49:55 +00005914 if (!space_info) {
David Sterbab6919a52013-04-29 13:39:40 +00005915 btrfs_err(root->fs_info, "No space info for %llu", flags);
Josef Bacik1b1d1f62010-03-19 20:49:55 +00005916 return -ENOSPC;
5917 }
Josef Bacik2552d172009-04-03 10:14:19 -04005918
Josef Bacik67377732010-09-16 16:19:09 -04005919 /*
5920 * If the space info is for both data and metadata it means we have a
5921 * small filesystem and we can't use the clustering stuff.
5922 */
5923 if (btrfs_mixed_space_info(space_info))
5924 use_cluster = false;
5925
David Sterbab6919a52013-04-29 13:39:40 +00005926 if (flags & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
Chris Masonfa9c0d792009-04-03 09:47:43 -04005927 last_ptr = &root->fs_info->meta_alloc_cluster;
Chris Mason536ac8a2009-02-12 09:41:38 -05005928 if (!btrfs_test_opt(root, SSD))
5929 empty_cluster = 64 * 1024;
Chris Mason239b14b2008-03-24 15:02:07 -04005930 }
5931
David Sterbab6919a52013-04-29 13:39:40 +00005932 if ((flags & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
Josef Bacik67377732010-09-16 16:19:09 -04005933 btrfs_test_opt(root, SSD)) {
Chris Masonfa9c0d792009-04-03 09:47:43 -04005934 last_ptr = &root->fs_info->data_alloc_cluster;
5935 }
Josef Bacik0f9dd462008-09-23 13:14:11 -04005936
Chris Mason239b14b2008-03-24 15:02:07 -04005937 if (last_ptr) {
Chris Masonfa9c0d792009-04-03 09:47:43 -04005938 spin_lock(&last_ptr->lock);
5939 if (last_ptr->block_group)
5940 hint_byte = last_ptr->window_start;
5941 spin_unlock(&last_ptr->lock);
Chris Mason239b14b2008-03-24 15:02:07 -04005942 }
Chris Masonfa9c0d792009-04-03 09:47:43 -04005943
Chris Masona061fc82008-05-07 11:43:44 -04005944 search_start = max(search_start, first_logical_byte(root, 0));
Chris Mason239b14b2008-03-24 15:02:07 -04005945 search_start = max(search_start, hint_byte);
Chris Mason0b86a832008-03-24 15:01:56 -04005946
Josef Bacik817d52f2009-07-13 21:29:25 -04005947 if (!last_ptr)
Chris Masonfa9c0d792009-04-03 09:47:43 -04005948 empty_cluster = 0;
Chris Masonfa9c0d792009-04-03 09:47:43 -04005949
Josef Bacik2552d172009-04-03 10:14:19 -04005950 if (search_start == hint_byte) {
Josef Bacik2552d172009-04-03 10:14:19 -04005951 block_group = btrfs_lookup_block_group(root->fs_info,
5952 search_start);
Alexandre Oliva274bd4f2011-12-07 20:08:40 -05005953 used_block_group = block_group;
Josef Bacik817d52f2009-07-13 21:29:25 -04005954 /*
5955 * we don't want to use the block group if it doesn't match our
5956 * allocation bits, or if its not cached.
Josef Bacikccf0e722009-11-10 21:23:48 -05005957 *
5958 * However if we are re-searching with an ideal block group
5959 * picked out then we don't care that the block group is cached.
Josef Bacik817d52f2009-07-13 21:29:25 -04005960 */
David Sterbab6919a52013-04-29 13:39:40 +00005961 if (block_group && block_group_bits(block_group, flags) &&
Josef Bacik285ff5a2012-01-13 15:27:45 -05005962 block_group->cached != BTRFS_CACHE_NO) {
Josef Bacik2552d172009-04-03 10:14:19 -04005963 down_read(&space_info->groups_sem);
Chris Mason44fb5512009-06-04 15:34:51 -04005964 if (list_empty(&block_group->list) ||
5965 block_group->ro) {
5966 /*
5967 * someone is removing this block group,
5968 * we can't jump into the have_block_group
5969 * target because our list pointers are not
5970 * valid
5971 */
5972 btrfs_put_block_group(block_group);
5973 up_read(&space_info->groups_sem);
Josef Bacikccf0e722009-11-10 21:23:48 -05005974 } else {
Yan, Zhengb742bb82010-05-16 10:46:24 -04005975 index = get_block_group_index(block_group);
Chris Mason44fb5512009-06-04 15:34:51 -04005976 goto have_block_group;
Josef Bacikccf0e722009-11-10 21:23:48 -05005977 }
Josef Bacik2552d172009-04-03 10:14:19 -04005978 } else if (block_group) {
Chris Masonfa9c0d792009-04-03 09:47:43 -04005979 btrfs_put_block_group(block_group);
Josef Bacik2552d172009-04-03 10:14:19 -04005980 }
Chris Mason42e70e72008-11-07 18:17:11 -05005981 }
Josef Bacik2552d172009-04-03 10:14:19 -04005982search:
Miao Xie60d2adb2011-09-09 17:34:35 +08005983 have_caching_bg = false;
Josef Bacik80eb2342008-10-29 14:49:05 -04005984 down_read(&space_info->groups_sem);
Yan, Zhengb742bb82010-05-16 10:46:24 -04005985 list_for_each_entry(block_group, &space_info->block_groups[index],
5986 list) {
Josef Bacik6226cb0a2009-04-03 10:14:18 -04005987 u64 offset;
Josef Bacik817d52f2009-07-13 21:29:25 -04005988 int cached;
Chris Mason8a1413a2008-11-10 16:13:54 -05005989
Alexandre Oliva274bd4f2011-12-07 20:08:40 -05005990 used_block_group = block_group;
Josef Bacik11dfe352009-11-13 20:12:59 +00005991 btrfs_get_block_group(block_group);
Josef Bacik2552d172009-04-03 10:14:19 -04005992 search_start = block_group->key.objectid;
Chris Mason42e70e72008-11-07 18:17:11 -05005993
Chris Mason83a50de2010-12-13 15:06:46 -05005994 /*
5995 * this can happen if we end up cycling through all the
5996 * raid types, but we want to make sure we only allocate
5997 * for the proper type.
5998 */
David Sterbab6919a52013-04-29 13:39:40 +00005999 if (!block_group_bits(block_group, flags)) {
Chris Mason83a50de2010-12-13 15:06:46 -05006000 u64 extra = BTRFS_BLOCK_GROUP_DUP |
6001 BTRFS_BLOCK_GROUP_RAID1 |
David Woodhouse53b381b2013-01-29 18:40:14 -05006002 BTRFS_BLOCK_GROUP_RAID5 |
6003 BTRFS_BLOCK_GROUP_RAID6 |
Chris Mason83a50de2010-12-13 15:06:46 -05006004 BTRFS_BLOCK_GROUP_RAID10;
6005
6006 /*
6007 * if they asked for extra copies and this block group
6008 * doesn't provide them, bail. This does allow us to
6009 * fill raid0 from raid1.
6010 */
David Sterbab6919a52013-04-29 13:39:40 +00006011 if ((flags & extra) && !(block_group->flags & extra))
Chris Mason83a50de2010-12-13 15:06:46 -05006012 goto loop;
6013 }
6014
Josef Bacik2552d172009-04-03 10:14:19 -04006015have_block_group:
Josef Bacik291c7d22011-11-14 13:52:14 -05006016 cached = block_group_cache_done(block_group);
6017 if (unlikely(!cached)) {
Josef Bacik291c7d22011-11-14 13:52:14 -05006018 found_uncached_bg = true;
Liu Bof6373bf2012-12-27 09:01:18 +00006019 ret = cache_block_group(block_group, 0);
Chris Mason1d4284b2012-03-28 20:31:37 -04006020 BUG_ON(ret < 0);
6021 ret = 0;
Josef Bacikea6a4782008-11-20 12:16:16 -05006022 }
6023
Josef Bacikea6a4782008-11-20 12:16:16 -05006024 if (unlikely(block_group->ro))
Josef Bacik2552d172009-04-03 10:14:19 -04006025 goto loop;
Josef Bacik0f9dd462008-09-23 13:14:11 -04006026
Josef Bacik0a243252009-09-11 16:11:20 -04006027 /*
Alexandre Oliva062c05c2011-12-07 19:50:42 -05006028 * Ok we want to try and use the cluster allocator, so
6029 * lets look there
Josef Bacik0a243252009-09-11 16:11:20 -04006030 */
Alexandre Oliva062c05c2011-12-07 19:50:42 -05006031 if (last_ptr) {
Chris Mason8de972b2013-01-04 15:39:43 -05006032 unsigned long aligned_cluster;
Chris Masonfa9c0d792009-04-03 09:47:43 -04006033 /*
6034 * the refill lock keeps out other
6035 * people trying to start a new cluster
6036 */
6037 spin_lock(&last_ptr->refill_lock);
Alexandre Oliva274bd4f2011-12-07 20:08:40 -05006038 used_block_group = last_ptr->block_group;
6039 if (used_block_group != block_group &&
6040 (!used_block_group ||
6041 used_block_group->ro ||
David Sterbab6919a52013-04-29 13:39:40 +00006042 !block_group_bits(used_block_group, flags))) {
Alexandre Oliva274bd4f2011-12-07 20:08:40 -05006043 used_block_group = block_group;
Chris Mason44fb5512009-06-04 15:34:51 -04006044 goto refill_cluster;
Alexandre Oliva274bd4f2011-12-07 20:08:40 -05006045 }
Chris Mason44fb5512009-06-04 15:34:51 -04006046
Alexandre Oliva274bd4f2011-12-07 20:08:40 -05006047 if (used_block_group != block_group)
6048 btrfs_get_block_group(used_block_group);
6049
6050 offset = btrfs_alloc_from_cluster(used_block_group,
6051 last_ptr, num_bytes, used_block_group->key.objectid);
Chris Masonfa9c0d792009-04-03 09:47:43 -04006052 if (offset) {
6053 /* we have a block, we're done */
6054 spin_unlock(&last_ptr->refill_lock);
Josef Bacik3f7de032011-11-10 08:29:20 -05006055 trace_btrfs_reserve_extent_cluster(root,
6056 block_group, search_start, num_bytes);
Chris Masonfa9c0d792009-04-03 09:47:43 -04006057 goto checks;
6058 }
6059
Alexandre Oliva274bd4f2011-12-07 20:08:40 -05006060 WARN_ON(last_ptr->block_group != used_block_group);
6061 if (used_block_group != block_group) {
6062 btrfs_put_block_group(used_block_group);
6063 used_block_group = block_group;
Chris Masonfa9c0d792009-04-03 09:47:43 -04006064 }
Chris Mason44fb5512009-06-04 15:34:51 -04006065refill_cluster:
Alexandre Oliva274bd4f2011-12-07 20:08:40 -05006066 BUG_ON(used_block_group != block_group);
Alexandre Oliva062c05c2011-12-07 19:50:42 -05006067 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
6068 * set up a new clusters, so lets just skip it
6069 * and let the allocator find whatever block
6070 * it can find. If we reach this point, we
6071 * will have tried the cluster allocator
6072 * plenty of times and not have found
6073 * anything, so we are likely way too
6074 * fragmented for the clustering stuff to find
Alexandre Olivaa5f6f712011-12-12 04:48:19 -02006075 * anything.
6076 *
6077 * However, if the cluster is taken from the
6078 * current block group, release the cluster
6079 * first, so that we stand a better chance of
6080 * succeeding in the unclustered
6081 * allocation. */
6082 if (loop >= LOOP_NO_EMPTY_SIZE &&
6083 last_ptr->block_group != block_group) {
Alexandre Oliva062c05c2011-12-07 19:50:42 -05006084 spin_unlock(&last_ptr->refill_lock);
6085 goto unclustered_alloc;
6086 }
6087
Chris Masonfa9c0d792009-04-03 09:47:43 -04006088 /*
6089 * this cluster didn't work out, free it and
6090 * start over
6091 */
6092 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6093
Alexandre Olivaa5f6f712011-12-12 04:48:19 -02006094 if (loop >= LOOP_NO_EMPTY_SIZE) {
6095 spin_unlock(&last_ptr->refill_lock);
6096 goto unclustered_alloc;
6097 }
6098
Chris Mason8de972b2013-01-04 15:39:43 -05006099 aligned_cluster = max_t(unsigned long,
6100 empty_cluster + empty_size,
6101 block_group->full_stripe_len);
6102
Chris Masonfa9c0d792009-04-03 09:47:43 -04006103 /* allocate a cluster in this block group */
Chris Mason451d7582009-06-09 20:28:34 -04006104 ret = btrfs_find_space_cluster(trans, root,
Chris Masonfa9c0d792009-04-03 09:47:43 -04006105 block_group, last_ptr,
Alexandre Oliva1b22bad2011-11-30 13:43:00 -05006106 search_start, num_bytes,
Chris Mason8de972b2013-01-04 15:39:43 -05006107 aligned_cluster);
Chris Masonfa9c0d792009-04-03 09:47:43 -04006108 if (ret == 0) {
6109 /*
6110 * now pull our allocation out of this
6111 * cluster
6112 */
6113 offset = btrfs_alloc_from_cluster(block_group,
6114 last_ptr, num_bytes,
6115 search_start);
6116 if (offset) {
6117 /* we found one, proceed */
6118 spin_unlock(&last_ptr->refill_lock);
Josef Bacik3f7de032011-11-10 08:29:20 -05006119 trace_btrfs_reserve_extent_cluster(root,
6120 block_group, search_start,
6121 num_bytes);
Chris Masonfa9c0d792009-04-03 09:47:43 -04006122 goto checks;
6123 }
Josef Bacik0a243252009-09-11 16:11:20 -04006124 } else if (!cached && loop > LOOP_CACHING_NOWAIT
6125 && !failed_cluster_refill) {
Josef Bacik817d52f2009-07-13 21:29:25 -04006126 spin_unlock(&last_ptr->refill_lock);
6127
Josef Bacik0a243252009-09-11 16:11:20 -04006128 failed_cluster_refill = true;
Josef Bacik817d52f2009-07-13 21:29:25 -04006129 wait_block_group_cache_progress(block_group,
6130 num_bytes + empty_cluster + empty_size);
6131 goto have_block_group;
Chris Masonfa9c0d792009-04-03 09:47:43 -04006132 }
Josef Bacik817d52f2009-07-13 21:29:25 -04006133
Chris Masonfa9c0d792009-04-03 09:47:43 -04006134 /*
6135 * at this point we either didn't find a cluster
6136 * or we weren't able to allocate a block from our
6137 * cluster. Free the cluster we've been trying
6138 * to use, and go to the next block group
6139 */
Josef Bacik0a243252009-09-11 16:11:20 -04006140 btrfs_return_cluster_to_free_space(NULL, last_ptr);
Chris Masonfa9c0d792009-04-03 09:47:43 -04006141 spin_unlock(&last_ptr->refill_lock);
Josef Bacik0a243252009-09-11 16:11:20 -04006142 goto loop;
Chris Masonfa9c0d792009-04-03 09:47:43 -04006143 }
6144
Alexandre Oliva062c05c2011-12-07 19:50:42 -05006145unclustered_alloc:
Alexandre Olivaa5f6f712011-12-12 04:48:19 -02006146 spin_lock(&block_group->free_space_ctl->tree_lock);
6147 if (cached &&
6148 block_group->free_space_ctl->free_space <
6149 num_bytes + empty_cluster + empty_size) {
6150 spin_unlock(&block_group->free_space_ctl->tree_lock);
6151 goto loop;
6152 }
6153 spin_unlock(&block_group->free_space_ctl->tree_lock);
6154
Josef Bacik6226cb0a2009-04-03 10:14:18 -04006155 offset = btrfs_find_space_for_alloc(block_group, search_start,
6156 num_bytes, empty_size);
Josef Bacik1cdda9b2009-10-06 10:04:28 -04006157 /*
6158 * If we didn't find a chunk, and we haven't failed on this
6159 * block group before, and this block group is in the middle of
6160 * caching and we are ok with waiting, then go ahead and wait
6161 * for progress to be made, and set failed_alloc to true.
6162 *
6163 * If failed_alloc is true then we've already waited on this
6164 * block group once and should move on to the next block group.
6165 */
6166 if (!offset && !failed_alloc && !cached &&
6167 loop > LOOP_CACHING_NOWAIT) {
Josef Bacik817d52f2009-07-13 21:29:25 -04006168 wait_block_group_cache_progress(block_group,
Josef Bacik1cdda9b2009-10-06 10:04:28 -04006169 num_bytes + empty_size);
6170 failed_alloc = true;
Josef Bacik817d52f2009-07-13 21:29:25 -04006171 goto have_block_group;
Josef Bacik1cdda9b2009-10-06 10:04:28 -04006172 } else if (!offset) {
Miao Xie60d2adb2011-09-09 17:34:35 +08006173 if (!cached)
6174 have_caching_bg = true;
Josef Bacik1cdda9b2009-10-06 10:04:28 -04006175 goto loop;
Josef Bacik817d52f2009-07-13 21:29:25 -04006176 }
Chris Masonfa9c0d792009-04-03 09:47:43 -04006177checks:
David Woodhouse53b381b2013-01-29 18:40:14 -05006178 search_start = stripe_align(root, used_block_group,
6179 offset, num_bytes);
Chris Masone37c9e62007-05-09 20:13:14 -04006180
Josef Bacik2552d172009-04-03 10:14:19 -04006181 /* move on to the next group */
6182 if (search_start + num_bytes >
Alexandre Oliva274bd4f2011-12-07 20:08:40 -05006183 used_block_group->key.objectid + used_block_group->key.offset) {
6184 btrfs_add_free_space(used_block_group, offset, num_bytes);
Josef Bacik2552d172009-04-03 10:14:19 -04006185 goto loop;
Josef Bacik6226cb0a2009-04-03 10:14:18 -04006186 }
Josef Bacik80eb2342008-10-29 14:49:05 -04006187
Josef Bacik6226cb0a2009-04-03 10:14:18 -04006188 if (offset < search_start)
Alexandre Oliva274bd4f2011-12-07 20:08:40 -05006189 btrfs_add_free_space(used_block_group, offset,
Josef Bacik6226cb0a2009-04-03 10:14:18 -04006190 search_start - offset);
6191 BUG_ON(offset > search_start);
6192
Alexandre Oliva274bd4f2011-12-07 20:08:40 -05006193 ret = btrfs_update_reserved_bytes(used_block_group, num_bytes,
Josef Bacikfb25e912011-07-26 17:00:46 -04006194 alloc_type);
Yan, Zhengf0486c62010-05-16 10:46:25 -04006195 if (ret == -EAGAIN) {
Alexandre Oliva274bd4f2011-12-07 20:08:40 -05006196 btrfs_add_free_space(used_block_group, offset, num_bytes);
Yan, Zhengf0486c62010-05-16 10:46:25 -04006197 goto loop;
6198 }
Yan Zheng11833d62009-09-11 16:11:19 -04006199
Josef Bacik2552d172009-04-03 10:14:19 -04006200 /* we are all good, lets return */
Yan, Zhengf0486c62010-05-16 10:46:25 -04006201 ins->objectid = search_start;
6202 ins->offset = num_bytes;
6203
Josef Bacik3f7de032011-11-10 08:29:20 -05006204 trace_btrfs_reserve_extent(orig_root, block_group,
6205 search_start, num_bytes);
Alexandre Oliva274bd4f2011-12-07 20:08:40 -05006206 if (used_block_group != block_group)
6207 btrfs_put_block_group(used_block_group);
Josef Bacikd82a6f12011-05-11 15:26:06 -04006208 btrfs_put_block_group(block_group);
Josef Bacik2552d172009-04-03 10:14:19 -04006209 break;
6210loop:
Josef Bacik0a243252009-09-11 16:11:20 -04006211 failed_cluster_refill = false;
Josef Bacik1cdda9b2009-10-06 10:04:28 -04006212 failed_alloc = false;
Yan, Zhengb742bb82010-05-16 10:46:24 -04006213 BUG_ON(index != get_block_group_index(block_group));
Alexandre Oliva274bd4f2011-12-07 20:08:40 -05006214 if (used_block_group != block_group)
6215 btrfs_put_block_group(used_block_group);
Chris Masonfa9c0d792009-04-03 09:47:43 -04006216 btrfs_put_block_group(block_group);
Josef Bacik2552d172009-04-03 10:14:19 -04006217 }
6218 up_read(&space_info->groups_sem);
Chris Masonf5a31e12008-11-10 11:47:09 -05006219
Miao Xie60d2adb2011-09-09 17:34:35 +08006220 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
6221 goto search;
6222
Yan, Zhengb742bb82010-05-16 10:46:24 -04006223 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
6224 goto search;
6225
Josef Bacik285ff5a2012-01-13 15:27:45 -05006226 /*
Josef Bacikccf0e722009-11-10 21:23:48 -05006227 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
6228 * caching kthreads as we move along
Josef Bacik817d52f2009-07-13 21:29:25 -04006229 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
6230 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
6231 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
6232 * again
Chris Masonfa9c0d792009-04-03 09:47:43 -04006233 */
Josef Bacik723bda22011-05-27 16:11:38 -04006234 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
Yan, Zhengb742bb82010-05-16 10:46:24 -04006235 index = 0;
Josef Bacik723bda22011-05-27 16:11:38 -04006236 loop++;
Josef Bacik817d52f2009-07-13 21:29:25 -04006237 if (loop == LOOP_ALLOC_CHUNK) {
David Sterbab6919a52013-04-29 13:39:40 +00006238 ret = do_chunk_alloc(trans, root, flags,
Josef Bacikea658ba2012-09-11 16:57:25 -04006239 CHUNK_ALLOC_FORCE);
6240 /*
6241 * Do not bail out on ENOSPC since we
6242 * can do more things.
6243 */
6244 if (ret < 0 && ret != -ENOSPC) {
6245 btrfs_abort_transaction(trans,
6246 root, ret);
6247 goto out;
Josef Bacik723bda22011-05-27 16:11:38 -04006248 }
Josef Bacik723bda22011-05-27 16:11:38 -04006249 }
6250
6251 if (loop == LOOP_NO_EMPTY_SIZE) {
Chris Masonfa9c0d792009-04-03 09:47:43 -04006252 empty_size = 0;
6253 empty_cluster = 0;
6254 }
Chris Mason42e70e72008-11-07 18:17:11 -05006255
Josef Bacik723bda22011-05-27 16:11:38 -04006256 goto search;
Josef Bacik2552d172009-04-03 10:14:19 -04006257 } else if (!ins->objectid) {
6258 ret = -ENOSPC;
Josef Bacikd82a6f12011-05-11 15:26:06 -04006259 } else if (ins->objectid) {
Josef Bacik2552d172009-04-03 10:14:19 -04006260 ret = 0;
6261 }
Jeff Mahoney79787ea2012-03-12 16:03:00 +01006262out:
Chris Mason0b86a832008-03-24 15:01:56 -04006263
Chris Mason0f70abe2007-02-28 16:46:22 -05006264 return ret;
Chris Masonfec577f2007-02-26 10:40:21 -05006265}
Chris Masonec44a352008-04-28 15:29:52 -04006266
Josef Bacik9ed74f22009-09-11 16:12:44 -04006267static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
6268 int dump_block_groups)
Josef Bacik0f9dd462008-09-23 13:14:11 -04006269{
6270 struct btrfs_block_group_cache *cache;
Yan, Zhengb742bb82010-05-16 10:46:24 -04006271 int index = 0;
Josef Bacik0f9dd462008-09-23 13:14:11 -04006272
Josef Bacik9ed74f22009-09-11 16:12:44 -04006273 spin_lock(&info->lock);
Josef Bacikfb25e912011-07-26 17:00:46 -04006274 printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n",
6275 (unsigned long long)info->flags,
Chris Masond3977122009-01-05 21:25:51 -05006276 (unsigned long long)(info->total_bytes - info->bytes_used -
Josef Bacik9ed74f22009-09-11 16:12:44 -04006277 info->bytes_pinned - info->bytes_reserved -
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04006278 info->bytes_readonly),
Chris Masond3977122009-01-05 21:25:51 -05006279 (info->full) ? "" : "not ");
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04006280 printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
6281 "reserved=%llu, may_use=%llu, readonly=%llu\n",
Joel Becker21380932009-04-21 12:38:29 -07006282 (unsigned long long)info->total_bytes,
Josef Bacik9ed74f22009-09-11 16:12:44 -04006283 (unsigned long long)info->bytes_used,
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04006284 (unsigned long long)info->bytes_pinned,
6285 (unsigned long long)info->bytes_reserved,
6286 (unsigned long long)info->bytes_may_use,
6287 (unsigned long long)info->bytes_readonly);
Josef Bacik9ed74f22009-09-11 16:12:44 -04006288 spin_unlock(&info->lock);
6289
6290 if (!dump_block_groups)
6291 return;
Josef Bacik0f9dd462008-09-23 13:14:11 -04006292
Josef Bacik80eb2342008-10-29 14:49:05 -04006293 down_read(&info->groups_sem);
Yan, Zhengb742bb82010-05-16 10:46:24 -04006294again:
6295 list_for_each_entry(cache, &info->block_groups[index], list) {
Josef Bacik0f9dd462008-09-23 13:14:11 -04006296 spin_lock(&cache->lock);
Liu Bo799ffc32012-07-06 03:31:35 -06006297 printk(KERN_INFO "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s\n",
Chris Masond3977122009-01-05 21:25:51 -05006298 (unsigned long long)cache->key.objectid,
6299 (unsigned long long)cache->key.offset,
6300 (unsigned long long)btrfs_block_group_used(&cache->item),
6301 (unsigned long long)cache->pinned,
Liu Bo799ffc32012-07-06 03:31:35 -06006302 (unsigned long long)cache->reserved,
6303 cache->ro ? "[readonly]" : "");
Josef Bacik0f9dd462008-09-23 13:14:11 -04006304 btrfs_dump_free_space(cache, bytes);
6305 spin_unlock(&cache->lock);
6306 }
Yan, Zhengb742bb82010-05-16 10:46:24 -04006307 if (++index < BTRFS_NR_RAID_TYPES)
6308 goto again;
Josef Bacik80eb2342008-10-29 14:49:05 -04006309 up_read(&info->groups_sem);
Josef Bacik0f9dd462008-09-23 13:14:11 -04006310}
Zheng Yane8569812008-09-26 10:05:48 -04006311
Yan Zheng11833d62009-09-11 16:11:19 -04006312int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
6313 struct btrfs_root *root,
6314 u64 num_bytes, u64 min_alloc_size,
6315 u64 empty_size, u64 hint_byte,
David Sterbab6919a52013-04-29 13:39:40 +00006316 struct btrfs_key *ins, int is_data)
Chris Masonfec577f2007-02-26 10:40:21 -05006317{
Miao Xie9e622d62012-01-26 15:01:12 -05006318 bool final_tried = false;
David Sterbab6919a52013-04-29 13:39:40 +00006319 u64 flags;
Chris Masonfec577f2007-02-26 10:40:21 -05006320 int ret;
Chris Mason925baed2008-06-25 16:01:30 -04006321
David Sterbab6919a52013-04-29 13:39:40 +00006322 flags = btrfs_get_alloc_profile(root, is_data);
Chris Mason98d20f62008-04-14 09:46:10 -04006323again:
Chris Masondb945352007-10-15 16:15:53 -04006324 WARN_ON(num_bytes < root->sectorsize);
6325 ret = find_free_extent(trans, root, num_bytes, empty_size,
David Sterbab6919a52013-04-29 13:39:40 +00006326 hint_byte, ins, flags);
Chris Mason3b951512008-04-17 11:29:12 -04006327
Miao Xie9e622d62012-01-26 15:01:12 -05006328 if (ret == -ENOSPC) {
6329 if (!final_tried) {
6330 num_bytes = num_bytes >> 1;
Zach Brown24542bf2012-11-16 00:04:43 +00006331 num_bytes = round_down(num_bytes, root->sectorsize);
Miao Xie9e622d62012-01-26 15:01:12 -05006332 num_bytes = max(num_bytes, min_alloc_size);
Miao Xie9e622d62012-01-26 15:01:12 -05006333 if (num_bytes == min_alloc_size)
6334 final_tried = true;
6335 goto again;
6336 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
6337 struct btrfs_space_info *sinfo;
Josef Bacik0f9dd462008-09-23 13:14:11 -04006338
David Sterbab6919a52013-04-29 13:39:40 +00006339 sinfo = __find_space_info(root->fs_info, flags);
Simon Kirbyc2cf52e2013-03-19 22:41:23 +00006340 btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu",
David Sterbab6919a52013-04-29 13:39:40 +00006341 (unsigned long long)flags,
Simon Kirbyc2cf52e2013-03-19 22:41:23 +00006342 (unsigned long long)num_bytes);
Jeff Mahoney53804282012-03-01 14:56:28 +01006343 if (sinfo)
6344 dump_space_info(sinfo, num_bytes, 1);
Miao Xie9e622d62012-01-26 15:01:12 -05006345 }
Chris Mason925baed2008-06-25 16:01:30 -04006346 }
Josef Bacik0f9dd462008-09-23 13:14:11 -04006347
liubo1abe9b82011-03-24 11:18:59 +00006348 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
6349
Josef Bacik0f9dd462008-09-23 13:14:11 -04006350 return ret;
Chris Masone6dcd2d2008-07-17 12:53:50 -04006351}
6352
Chris Masone688b7252011-10-31 20:52:39 -04006353static int __btrfs_free_reserved_extent(struct btrfs_root *root,
6354 u64 start, u64 len, int pin)
Chris Mason65b51a02008-08-01 15:11:20 -04006355{
Josef Bacik0f9dd462008-09-23 13:14:11 -04006356 struct btrfs_block_group_cache *cache;
Liu Hui1f3c79a2009-01-05 15:57:51 -05006357 int ret = 0;
Josef Bacik0f9dd462008-09-23 13:14:11 -04006358
Josef Bacik0f9dd462008-09-23 13:14:11 -04006359 cache = btrfs_lookup_block_group(root->fs_info, start);
6360 if (!cache) {
Simon Kirbyc2cf52e2013-03-19 22:41:23 +00006361 btrfs_err(root->fs_info, "Unable to find block group for %llu",
6362 (unsigned long long)start);
Josef Bacik0f9dd462008-09-23 13:14:11 -04006363 return -ENOSPC;
6364 }
Liu Hui1f3c79a2009-01-05 15:57:51 -05006365
Chris Masone688b7252011-10-31 20:52:39 -04006366 if (pin)
6367 pin_down_extent(root, cache, start, len, 1);
6368 else {
Filipe Mananabf339142015-03-23 14:07:40 +00006369 if (btrfs_test_opt(root, DISCARD))
6370 ret = btrfs_discard_extent(root, start, len, NULL);
Chris Masone688b7252011-10-31 20:52:39 -04006371 btrfs_add_free_space(cache, start, len);
6372 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE);
6373 }
Chris Masonfa9c0d792009-04-03 09:47:43 -04006374 btrfs_put_block_group(cache);
Josef Bacik817d52f2009-07-13 21:29:25 -04006375
liubo1abe9b82011-03-24 11:18:59 +00006376 trace_btrfs_reserved_extent_free(root, start, len);
6377
Chris Masone6dcd2d2008-07-17 12:53:50 -04006378 return ret;
6379}
6380
Chris Masone688b7252011-10-31 20:52:39 -04006381int btrfs_free_reserved_extent(struct btrfs_root *root,
6382 u64 start, u64 len)
6383{
6384 return __btrfs_free_reserved_extent(root, start, len, 0);
6385}
6386
6387int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
6388 u64 start, u64 len)
6389{
6390 return __btrfs_free_reserved_extent(root, start, len, 1);
6391}
6392
Yan Zheng5d4f98a2009-06-10 10:45:14 -04006393static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6394 struct btrfs_root *root,
6395 u64 parent, u64 root_objectid,
6396 u64 flags, u64 owner, u64 offset,
6397 struct btrfs_key *ins, int ref_mod)
Chris Masone6dcd2d2008-07-17 12:53:50 -04006398{
6399 int ret;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04006400 struct btrfs_fs_info *fs_info = root->fs_info;
Chris Masone6dcd2d2008-07-17 12:53:50 -04006401 struct btrfs_extent_item *extent_item;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04006402 struct btrfs_extent_inline_ref *iref;
Chris Masone6dcd2d2008-07-17 12:53:50 -04006403 struct btrfs_path *path;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04006404 struct extent_buffer *leaf;
6405 int type;
6406 u32 size;
Chris Masonf2654de2007-06-26 12:20:46 -04006407
Yan Zheng5d4f98a2009-06-10 10:45:14 -04006408 if (parent > 0)
6409 type = BTRFS_SHARED_DATA_REF_KEY;
6410 else
6411 type = BTRFS_EXTENT_DATA_REF_KEY;
Zheng Yan31840ae2008-09-23 13:14:14 -04006412
Yan Zheng5d4f98a2009-06-10 10:45:14 -04006413 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
Chris Mason7bb86312007-12-11 09:25:06 -05006414
6415 path = btrfs_alloc_path();
Tsutomu Itohdb5b4932011-03-23 08:14:16 +00006416 if (!path)
6417 return -ENOMEM;
Chris Mason47e4bb92008-02-01 14:51:59 -05006418
Chris Masonb9473432009-03-13 11:00:37 -04006419 path->leave_spinning = 1;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04006420 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6421 ins, size);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01006422 if (ret) {
6423 btrfs_free_path(path);
6424 return ret;
6425 }
Josef Bacik0f9dd462008-09-23 13:14:11 -04006426
Yan Zheng5d4f98a2009-06-10 10:45:14 -04006427 leaf = path->nodes[0];
6428 extent_item = btrfs_item_ptr(leaf, path->slots[0],
Chris Mason47e4bb92008-02-01 14:51:59 -05006429 struct btrfs_extent_item);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04006430 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
6431 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6432 btrfs_set_extent_flags(leaf, extent_item,
6433 flags | BTRFS_EXTENT_FLAG_DATA);
Chris Mason47e4bb92008-02-01 14:51:59 -05006434
Yan Zheng5d4f98a2009-06-10 10:45:14 -04006435 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6436 btrfs_set_extent_inline_ref_type(leaf, iref, type);
6437 if (parent > 0) {
6438 struct btrfs_shared_data_ref *ref;
6439 ref = (struct btrfs_shared_data_ref *)(iref + 1);
6440 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6441 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
6442 } else {
6443 struct btrfs_extent_data_ref *ref;
6444 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
6445 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
6446 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
6447 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
6448 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
6449 }
Chris Mason47e4bb92008-02-01 14:51:59 -05006450
6451 btrfs_mark_buffer_dirty(path->nodes[0]);
Chris Mason7bb86312007-12-11 09:25:06 -05006452 btrfs_free_path(path);
Chris Masonf510cfe2007-10-15 16:14:48 -04006453
Liu Boc53d6132012-12-27 09:01:19 +00006454 ret = update_block_group(root, ins->objectid, ins->offset, 1);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01006455 if (ret) { /* -ENOENT, logic error */
Simon Kirbyc2cf52e2013-03-19 22:41:23 +00006456 btrfs_err(fs_info, "update block group failed for %llu %llu",
6457 (unsigned long long)ins->objectid,
6458 (unsigned long long)ins->offset);
Chris Masonf5947062008-02-04 10:10:13 -05006459 BUG();
6460 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04006461 return ret;
6462}
6463
Yan Zheng5d4f98a2009-06-10 10:45:14 -04006464static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
6465 struct btrfs_root *root,
6466 u64 parent, u64 root_objectid,
6467 u64 flags, struct btrfs_disk_key *key,
6468 int level, struct btrfs_key *ins)
6469{
6470 int ret;
6471 struct btrfs_fs_info *fs_info = root->fs_info;
6472 struct btrfs_extent_item *extent_item;
6473 struct btrfs_tree_block_info *block_info;
6474 struct btrfs_extent_inline_ref *iref;
6475 struct btrfs_path *path;
6476 struct extent_buffer *leaf;
Josef Bacik3173a182013-03-07 14:22:04 -05006477 u32 size = sizeof(*extent_item) + sizeof(*iref);
6478 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6479 SKINNY_METADATA);
6480
6481 if (!skinny_metadata)
6482 size += sizeof(*block_info);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04006483
6484 path = btrfs_alloc_path();
Mark Fashehd8926bb2011-07-13 10:38:47 -07006485 if (!path)
6486 return -ENOMEM;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04006487
6488 path->leave_spinning = 1;
6489 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6490 ins, size);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01006491 if (ret) {
6492 btrfs_free_path(path);
6493 return ret;
6494 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04006495
6496 leaf = path->nodes[0];
6497 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6498 struct btrfs_extent_item);
6499 btrfs_set_extent_refs(leaf, extent_item, 1);
6500 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6501 btrfs_set_extent_flags(leaf, extent_item,
6502 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04006503
Josef Bacik3173a182013-03-07 14:22:04 -05006504 if (skinny_metadata) {
6505 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6506 } else {
6507 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
6508 btrfs_set_tree_block_key(leaf, block_info, key);
6509 btrfs_set_tree_block_level(leaf, block_info, level);
6510 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
6511 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04006512
Yan Zheng5d4f98a2009-06-10 10:45:14 -04006513 if (parent > 0) {
6514 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
6515 btrfs_set_extent_inline_ref_type(leaf, iref,
6516 BTRFS_SHARED_BLOCK_REF_KEY);
6517 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6518 } else {
6519 btrfs_set_extent_inline_ref_type(leaf, iref,
6520 BTRFS_TREE_BLOCK_REF_KEY);
6521 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
6522 }
6523
6524 btrfs_mark_buffer_dirty(leaf);
6525 btrfs_free_path(path);
6526
Josef Bacik3173a182013-03-07 14:22:04 -05006527 ret = update_block_group(root, ins->objectid, root->leafsize, 1);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01006528 if (ret) { /* -ENOENT, logic error */
Simon Kirbyc2cf52e2013-03-19 22:41:23 +00006529 btrfs_err(fs_info, "update block group failed for %llu %llu",
6530 (unsigned long long)ins->objectid,
6531 (unsigned long long)ins->offset);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04006532 BUG();
6533 }
6534 return ret;
6535}
6536
6537int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6538 struct btrfs_root *root,
6539 u64 root_objectid, u64 owner,
6540 u64 offset, struct btrfs_key *ins)
Chris Masone6dcd2d2008-07-17 12:53:50 -04006541{
6542 int ret;
Chris Mason1c2308f2008-09-23 13:14:13 -04006543
Yan Zheng5d4f98a2009-06-10 10:45:14 -04006544 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
Chris Mason56bec292009-03-13 10:10:06 -04006545
Arne Jansen66d7e7f2011-09-12 15:26:38 +02006546 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
6547 ins->offset, 0,
6548 root_objectid, owner, offset,
6549 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
Chris Masone6dcd2d2008-07-17 12:53:50 -04006550 return ret;
6551}
Chris Masone02119d2008-09-05 16:13:11 -04006552
6553/*
6554 * this is used by the tree logging recovery code. It records that
6555 * an extent has been allocated and makes sure to clear the free
6556 * space cache bits as well
6557 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -04006558int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
6559 struct btrfs_root *root,
6560 u64 root_objectid, u64 owner, u64 offset,
6561 struct btrfs_key *ins)
Chris Masone02119d2008-09-05 16:13:11 -04006562{
6563 int ret;
6564 struct btrfs_block_group_cache *block_group;
Yan Zheng11833d62009-09-11 16:11:19 -04006565 struct btrfs_caching_control *caching_ctl;
6566 u64 start = ins->objectid;
6567 u64 num_bytes = ins->offset;
Chris Masone02119d2008-09-05 16:13:11 -04006568
Chris Masone02119d2008-09-05 16:13:11 -04006569 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
Liu Bof6373bf2012-12-27 09:01:18 +00006570 cache_block_group(block_group, 0);
Yan Zheng11833d62009-09-11 16:11:19 -04006571 caching_ctl = get_caching_control(block_group);
Chris Masone02119d2008-09-05 16:13:11 -04006572
Yan Zheng11833d62009-09-11 16:11:19 -04006573 if (!caching_ctl) {
6574 BUG_ON(!block_group_cache_done(block_group));
6575 ret = btrfs_remove_free_space(block_group, start, num_bytes);
Josef Bacikb50c6e22013-04-25 15:55:30 -04006576 if (ret)
6577 goto out;
Yan Zheng11833d62009-09-11 16:11:19 -04006578 } else {
6579 mutex_lock(&caching_ctl->mutex);
6580
6581 if (start >= caching_ctl->progress) {
6582 ret = add_excluded_extent(root, start, num_bytes);
Yan Zheng11833d62009-09-11 16:11:19 -04006583 } else if (start + num_bytes <= caching_ctl->progress) {
6584 ret = btrfs_remove_free_space(block_group,
6585 start, num_bytes);
Yan Zheng11833d62009-09-11 16:11:19 -04006586 } else {
6587 num_bytes = caching_ctl->progress - start;
6588 ret = btrfs_remove_free_space(block_group,
6589 start, num_bytes);
Josef Bacikb50c6e22013-04-25 15:55:30 -04006590 if (ret)
6591 goto out_lock;
Yan Zheng11833d62009-09-11 16:11:19 -04006592
6593 start = caching_ctl->progress;
6594 num_bytes = ins->objectid + ins->offset -
6595 caching_ctl->progress;
6596 ret = add_excluded_extent(root, start, num_bytes);
Yan Zheng11833d62009-09-11 16:11:19 -04006597 }
Josef Bacikb50c6e22013-04-25 15:55:30 -04006598out_lock:
Yan Zheng11833d62009-09-11 16:11:19 -04006599 mutex_unlock(&caching_ctl->mutex);
6600 put_caching_control(caching_ctl);
Josef Bacikb50c6e22013-04-25 15:55:30 -04006601 if (ret)
6602 goto out;
Yan Zheng11833d62009-09-11 16:11:19 -04006603 }
6604
Josef Bacikfb25e912011-07-26 17:00:46 -04006605 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
6606 RESERVE_ALLOC_NO_ACCOUNT);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01006607 BUG_ON(ret); /* logic error */
Yan Zheng5d4f98a2009-06-10 10:45:14 -04006608 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
6609 0, owner, offset, ins, 1);
Josef Bacikb50c6e22013-04-25 15:55:30 -04006610out:
6611 btrfs_put_block_group(block_group);
Chris Masone02119d2008-09-05 16:13:11 -04006612 return ret;
6613}
6614
Eric Sandeen48a3b632013-04-25 20:41:01 +00006615static struct extent_buffer *
6616btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
6617 u64 bytenr, u32 blocksize, int level)
Chris Mason65b51a02008-08-01 15:11:20 -04006618{
6619 struct extent_buffer *buf;
6620
6621 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
6622 if (!buf)
6623 return ERR_PTR(-ENOMEM);
6624 btrfs_set_header_generation(buf, trans->transid);
Chris Mason85d4e462011-07-26 16:11:19 -04006625 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
Chris Mason65b51a02008-08-01 15:11:20 -04006626 btrfs_tree_lock(buf);
6627 clean_tree_block(trans, root, buf);
Josef Bacik3083ee22012-03-09 16:01:49 -05006628 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
Chris Masonb4ce94d2009-02-04 09:25:08 -05006629
6630 btrfs_set_lock_blocking(buf);
Chris Mason65b51a02008-08-01 15:11:20 -04006631 btrfs_set_buffer_uptodate(buf);
Chris Masonb4ce94d2009-02-04 09:25:08 -05006632
Chris Masond0c803c2008-09-11 16:17:57 -04006633 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
Yan, Zheng8cef4e12009-11-12 09:33:26 +00006634 /*
6635 * we allow two log transactions at a time, use different
6636 * EXENT bit to differentiate dirty pages.
6637 */
6638 if (root->log_transid % 2 == 0)
6639 set_extent_dirty(&root->dirty_log_pages, buf->start,
6640 buf->start + buf->len - 1, GFP_NOFS);
6641 else
6642 set_extent_new(&root->dirty_log_pages, buf->start,
6643 buf->start + buf->len - 1, GFP_NOFS);
Chris Masond0c803c2008-09-11 16:17:57 -04006644 } else {
6645 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
6646 buf->start + buf->len - 1, GFP_NOFS);
6647 }
Chris Mason65b51a02008-08-01 15:11:20 -04006648 trans->blocks_used++;
Chris Masonb4ce94d2009-02-04 09:25:08 -05006649 /* this returns a buffer locked for blocking */
Chris Mason65b51a02008-08-01 15:11:20 -04006650 return buf;
6651}
6652
Yan, Zhengf0486c62010-05-16 10:46:25 -04006653static struct btrfs_block_rsv *
6654use_block_rsv(struct btrfs_trans_handle *trans,
6655 struct btrfs_root *root, u32 blocksize)
6656{
6657 struct btrfs_block_rsv *block_rsv;
Josef Bacik68a82272011-01-24 21:43:20 +00006658 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
Yan, Zhengf0486c62010-05-16 10:46:25 -04006659 int ret;
Miao Xied88033d2013-05-13 13:55:12 +00006660 bool global_updated = false;
Yan, Zhengf0486c62010-05-16 10:46:25 -04006661
6662 block_rsv = get_block_rsv(trans, root);
6663
Miao Xieb586b322013-05-13 13:55:10 +00006664 if (unlikely(block_rsv->size == 0))
6665 goto try_reserve;
Miao Xied88033d2013-05-13 13:55:12 +00006666again:
Yan, Zhengf0486c62010-05-16 10:46:25 -04006667 ret = block_rsv_use_bytes(block_rsv, blocksize);
6668 if (!ret)
6669 return block_rsv;
6670
Miao Xieb586b322013-05-13 13:55:10 +00006671 if (block_rsv->failfast)
6672 return ERR_PTR(ret);
6673
Miao Xied88033d2013-05-13 13:55:12 +00006674 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
6675 global_updated = true;
6676 update_global_block_rsv(root->fs_info);
6677 goto again;
6678 }
6679
Miao Xieb586b322013-05-13 13:55:10 +00006680 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
6681 static DEFINE_RATELIMIT_STATE(_rs,
6682 DEFAULT_RATELIMIT_INTERVAL * 10,
6683 /*DEFAULT_RATELIMIT_BURST*/ 1);
6684 if (__ratelimit(&_rs))
6685 WARN(1, KERN_DEBUG
6686 "btrfs: block rsv returned %d\n", ret);
6687 }
6688try_reserve:
6689 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
6690 BTRFS_RESERVE_NO_FLUSH);
6691 if (!ret)
6692 return block_rsv;
6693 /*
6694 * If we couldn't reserve metadata bytes try and use some from
Miao Xie5881cfc2013-05-13 13:55:11 +00006695 * the global reserve if its space type is the same as the global
6696 * reservation.
Miao Xieb586b322013-05-13 13:55:10 +00006697 */
Miao Xie5881cfc2013-05-13 13:55:11 +00006698 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
6699 block_rsv->space_info == global_rsv->space_info) {
Miao Xieb586b322013-05-13 13:55:10 +00006700 ret = block_rsv_use_bytes(global_rsv, blocksize);
6701 if (!ret)
6702 return global_rsv;
6703 }
6704 return ERR_PTR(ret);
Yan, Zhengf0486c62010-05-16 10:46:25 -04006705}
6706
Josef Bacik8c2a3ca2012-01-10 10:31:31 -05006707static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
6708 struct btrfs_block_rsv *block_rsv, u32 blocksize)
Yan, Zhengf0486c62010-05-16 10:46:25 -04006709{
6710 block_rsv_add_bytes(block_rsv, blocksize, 0);
Josef Bacik8c2a3ca2012-01-10 10:31:31 -05006711 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
Yan, Zhengf0486c62010-05-16 10:46:25 -04006712}
6713
Chris Masonfec577f2007-02-26 10:40:21 -05006714/*
Yan, Zhengf0486c62010-05-16 10:46:25 -04006715 * finds a free extent and does all the dirty work required for allocation
6716 * returns the key for the extent through ins, and a tree buffer for
6717 * the first block of the extent through buf.
6718 *
Chris Masonfec577f2007-02-26 10:40:21 -05006719 * returns the tree buffer or NULL.
6720 */
Chris Mason5f39d392007-10-15 16:14:19 -04006721struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04006722 struct btrfs_root *root, u32 blocksize,
6723 u64 parent, u64 root_objectid,
6724 struct btrfs_disk_key *key, int level,
Jan Schmidt5581a512012-05-16 17:04:52 +02006725 u64 hint, u64 empty_size)
Chris Masonfec577f2007-02-26 10:40:21 -05006726{
Chris Masone2fa7222007-03-12 16:22:34 -04006727 struct btrfs_key ins;
Yan, Zhengf0486c62010-05-16 10:46:25 -04006728 struct btrfs_block_rsv *block_rsv;
Chris Mason5f39d392007-10-15 16:14:19 -04006729 struct extent_buffer *buf;
Yan, Zhengf0486c62010-05-16 10:46:25 -04006730 u64 flags = 0;
6731 int ret;
Josef Bacik3173a182013-03-07 14:22:04 -05006732 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6733 SKINNY_METADATA);
Yan, Zhengf0486c62010-05-16 10:46:25 -04006734
6735 block_rsv = use_block_rsv(trans, root, blocksize);
6736 if (IS_ERR(block_rsv))
6737 return ERR_CAST(block_rsv);
6738
6739 ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
Josef Bacik81c9ad22012-01-18 10:56:06 -05006740 empty_size, hint, &ins, 0);
Chris Masonfec577f2007-02-26 10:40:21 -05006741 if (ret) {
Josef Bacik8c2a3ca2012-01-10 10:31:31 -05006742 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
Chris Mason54aa1f42007-06-22 14:16:25 -04006743 return ERR_PTR(ret);
Chris Masonfec577f2007-02-26 10:40:21 -05006744 }
Chris Mason55c69072008-01-09 15:55:33 -05006745
Chris Mason4008c042009-02-12 14:09:45 -05006746 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
6747 blocksize, level);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01006748 BUG_ON(IS_ERR(buf)); /* -ENOMEM */
Yan, Zhengf0486c62010-05-16 10:46:25 -04006749
6750 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
6751 if (parent == 0)
6752 parent = ins.objectid;
6753 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
6754 } else
6755 BUG_ON(parent > 0);
6756
6757 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
6758 struct btrfs_delayed_extent_op *extent_op;
Miao Xie78a61842012-11-21 02:21:28 +00006759 extent_op = btrfs_alloc_delayed_extent_op();
Jeff Mahoney79787ea2012-03-12 16:03:00 +01006760 BUG_ON(!extent_op); /* -ENOMEM */
Yan, Zhengf0486c62010-05-16 10:46:25 -04006761 if (key)
6762 memcpy(&extent_op->key, key, sizeof(extent_op->key));
6763 else
6764 memset(&extent_op->key, 0, sizeof(extent_op->key));
6765 extent_op->flags_to_set = flags;
Josef Bacik3173a182013-03-07 14:22:04 -05006766 if (skinny_metadata)
6767 extent_op->update_key = 0;
6768 else
6769 extent_op->update_key = 1;
Yan, Zhengf0486c62010-05-16 10:46:25 -04006770 extent_op->update_flags = 1;
6771 extent_op->is_data = 0;
Josef Bacikb1c79e02013-05-09 13:49:30 -04006772 extent_op->level = level;
Yan, Zhengf0486c62010-05-16 10:46:25 -04006773
Arne Jansen66d7e7f2011-09-12 15:26:38 +02006774 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6775 ins.objectid,
Yan, Zhengf0486c62010-05-16 10:46:25 -04006776 ins.offset, parent, root_objectid,
6777 level, BTRFS_ADD_DELAYED_EXTENT,
Jan Schmidt5581a512012-05-16 17:04:52 +02006778 extent_op, 0);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01006779 BUG_ON(ret); /* -ENOMEM */
Yan, Zhengf0486c62010-05-16 10:46:25 -04006780 }
Chris Masonfec577f2007-02-26 10:40:21 -05006781 return buf;
6782}
Chris Masona28ec192007-03-06 20:08:01 -05006783
Yan Zheng2c47e6052009-06-27 21:07:35 -04006784struct walk_control {
6785 u64 refs[BTRFS_MAX_LEVEL];
6786 u64 flags[BTRFS_MAX_LEVEL];
6787 struct btrfs_key update_progress;
6788 int stage;
6789 int level;
6790 int shared_level;
6791 int update_ref;
6792 int keep_locks;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04006793 int reada_slot;
6794 int reada_count;
Arne Jansen66d7e7f2011-09-12 15:26:38 +02006795 int for_reloc;
Yan Zheng2c47e6052009-06-27 21:07:35 -04006796};
6797
6798#define DROP_REFERENCE 1
6799#define UPDATE_BACKREF 2
6800
Yan, Zheng1c4850e2009-09-21 15:55:59 -04006801static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
6802 struct btrfs_root *root,
6803 struct walk_control *wc,
6804 struct btrfs_path *path)
6805{
6806 u64 bytenr;
6807 u64 generation;
6808 u64 refs;
Yan, Zheng94fcca92009-10-09 09:25:16 -04006809 u64 flags;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04006810 u32 nritems;
6811 u32 blocksize;
6812 struct btrfs_key key;
6813 struct extent_buffer *eb;
6814 int ret;
6815 int slot;
6816 int nread = 0;
6817
6818 if (path->slots[wc->level] < wc->reada_slot) {
6819 wc->reada_count = wc->reada_count * 2 / 3;
6820 wc->reada_count = max(wc->reada_count, 2);
6821 } else {
6822 wc->reada_count = wc->reada_count * 3 / 2;
6823 wc->reada_count = min_t(int, wc->reada_count,
6824 BTRFS_NODEPTRS_PER_BLOCK(root));
6825 }
6826
6827 eb = path->nodes[wc->level];
6828 nritems = btrfs_header_nritems(eb);
6829 blocksize = btrfs_level_size(root, wc->level - 1);
6830
6831 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
6832 if (nread >= wc->reada_count)
6833 break;
6834
6835 cond_resched();
6836 bytenr = btrfs_node_blockptr(eb, slot);
6837 generation = btrfs_node_ptr_generation(eb, slot);
6838
6839 if (slot == path->slots[wc->level])
6840 goto reada;
6841
6842 if (wc->stage == UPDATE_BACKREF &&
6843 generation <= root->root_key.offset)
6844 continue;
6845
Yan, Zheng94fcca92009-10-09 09:25:16 -04006846 /* We don't lock the tree block, it's OK to be racy here */
Josef Bacik3173a182013-03-07 14:22:04 -05006847 ret = btrfs_lookup_extent_info(trans, root, bytenr,
6848 wc->level - 1, 1, &refs,
6849 &flags);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01006850 /* We don't care about errors in readahead. */
6851 if (ret < 0)
6852 continue;
Yan, Zheng94fcca92009-10-09 09:25:16 -04006853 BUG_ON(refs == 0);
6854
Yan, Zheng1c4850e2009-09-21 15:55:59 -04006855 if (wc->stage == DROP_REFERENCE) {
Yan, Zheng1c4850e2009-09-21 15:55:59 -04006856 if (refs == 1)
6857 goto reada;
6858
Yan, Zheng94fcca92009-10-09 09:25:16 -04006859 if (wc->level == 1 &&
6860 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6861 continue;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04006862 if (!wc->update_ref ||
6863 generation <= root->root_key.offset)
6864 continue;
6865 btrfs_node_key_to_cpu(eb, &key, slot);
6866 ret = btrfs_comp_cpu_keys(&key,
6867 &wc->update_progress);
6868 if (ret < 0)
6869 continue;
Yan, Zheng94fcca92009-10-09 09:25:16 -04006870 } else {
6871 if (wc->level == 1 &&
6872 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6873 continue;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04006874 }
6875reada:
6876 ret = readahead_tree_block(root, bytenr, blocksize,
6877 generation);
6878 if (ret)
6879 break;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04006880 nread++;
6881 }
6882 wc->reada_slot = slot;
6883}
6884
Chris Mason9aca1d52007-03-13 11:09:37 -04006885/*
Liu Bo2c016dc2012-12-26 15:32:17 +08006886 * helper to process tree block while walking down the tree.
Yan Zheng2c47e6052009-06-27 21:07:35 -04006887 *
Yan Zheng2c47e6052009-06-27 21:07:35 -04006888 * when wc->stage == UPDATE_BACKREF, this function updates
6889 * back refs for pointers in the block.
6890 *
6891 * NOTE: return value 1 means we should stop walking down.
Yan Zhengf82d02d2008-10-29 14:49:05 -04006892 */
Yan Zheng2c47e6052009-06-27 21:07:35 -04006893static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
6894 struct btrfs_root *root,
6895 struct btrfs_path *path,
Yan, Zheng94fcca92009-10-09 09:25:16 -04006896 struct walk_control *wc, int lookup_info)
Yan Zheng2c47e6052009-06-27 21:07:35 -04006897{
6898 int level = wc->level;
6899 struct extent_buffer *eb = path->nodes[level];
Yan Zheng2c47e6052009-06-27 21:07:35 -04006900 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6901 int ret;
6902
6903 if (wc->stage == UPDATE_BACKREF &&
6904 btrfs_header_owner(eb) != root->root_key.objectid)
6905 return 1;
6906
6907 /*
6908 * when reference count of tree block is 1, it won't increase
6909 * again. once full backref flag is set, we never clear it.
6910 */
Yan, Zheng94fcca92009-10-09 09:25:16 -04006911 if (lookup_info &&
6912 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
6913 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
Yan Zheng2c47e6052009-06-27 21:07:35 -04006914 BUG_ON(!path->locks[level]);
6915 ret = btrfs_lookup_extent_info(trans, root,
Josef Bacik3173a182013-03-07 14:22:04 -05006916 eb->start, level, 1,
Yan Zheng2c47e6052009-06-27 21:07:35 -04006917 &wc->refs[level],
6918 &wc->flags[level]);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01006919 BUG_ON(ret == -ENOMEM);
6920 if (ret)
6921 return ret;
Yan Zheng2c47e6052009-06-27 21:07:35 -04006922 BUG_ON(wc->refs[level] == 0);
6923 }
6924
Yan Zheng2c47e6052009-06-27 21:07:35 -04006925 if (wc->stage == DROP_REFERENCE) {
6926 if (wc->refs[level] > 1)
6927 return 1;
6928
6929 if (path->locks[level] && !wc->keep_locks) {
Chris Masonbd681512011-07-16 15:23:14 -04006930 btrfs_tree_unlock_rw(eb, path->locks[level]);
Yan Zheng2c47e6052009-06-27 21:07:35 -04006931 path->locks[level] = 0;
6932 }
6933 return 0;
6934 }
6935
6936 /* wc->stage == UPDATE_BACKREF */
6937 if (!(wc->flags[level] & flag)) {
6938 BUG_ON(!path->locks[level]);
Arne Jansen66d7e7f2011-09-12 15:26:38 +02006939 ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01006940 BUG_ON(ret); /* -ENOMEM */
Arne Jansen66d7e7f2011-09-12 15:26:38 +02006941 ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01006942 BUG_ON(ret); /* -ENOMEM */
Yan Zheng2c47e6052009-06-27 21:07:35 -04006943 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
Josef Bacikb1c79e02013-05-09 13:49:30 -04006944 eb->len, flag,
6945 btrfs_header_level(eb), 0);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01006946 BUG_ON(ret); /* -ENOMEM */
Yan Zheng2c47e6052009-06-27 21:07:35 -04006947 wc->flags[level] |= flag;
6948 }
6949
6950 /*
6951 * the block is shared by multiple trees, so it's not good to
6952 * keep the tree lock
6953 */
6954 if (path->locks[level] && level > 0) {
Chris Masonbd681512011-07-16 15:23:14 -04006955 btrfs_tree_unlock_rw(eb, path->locks[level]);
Yan Zheng2c47e6052009-06-27 21:07:35 -04006956 path->locks[level] = 0;
6957 }
6958 return 0;
6959}
6960
6961/*
Liu Bo2c016dc2012-12-26 15:32:17 +08006962 * helper to process tree block pointer.
Yan, Zheng1c4850e2009-09-21 15:55:59 -04006963 *
6964 * when wc->stage == DROP_REFERENCE, this function checks
6965 * reference count of the block pointed to. if the block
6966 * is shared and we need update back refs for the subtree
6967 * rooted at the block, this function changes wc->stage to
6968 * UPDATE_BACKREF. if the block is shared and there is no
6969 * need to update back, this function drops the reference
6970 * to the block.
6971 *
6972 * NOTE: return value 1 means we should stop walking down.
6973 */
6974static noinline int do_walk_down(struct btrfs_trans_handle *trans,
6975 struct btrfs_root *root,
6976 struct btrfs_path *path,
Yan, Zheng94fcca92009-10-09 09:25:16 -04006977 struct walk_control *wc, int *lookup_info)
Yan, Zheng1c4850e2009-09-21 15:55:59 -04006978{
6979 u64 bytenr;
6980 u64 generation;
6981 u64 parent;
6982 u32 blocksize;
6983 struct btrfs_key key;
6984 struct extent_buffer *next;
6985 int level = wc->level;
6986 int reada = 0;
6987 int ret = 0;
6988
6989 generation = btrfs_node_ptr_generation(path->nodes[level],
6990 path->slots[level]);
6991 /*
6992 * if the lower level block was created before the snapshot
6993 * was created, we know there is no need to update back refs
6994 * for the subtree
6995 */
6996 if (wc->stage == UPDATE_BACKREF &&
Yan, Zheng94fcca92009-10-09 09:25:16 -04006997 generation <= root->root_key.offset) {
6998 *lookup_info = 1;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04006999 return 1;
Yan, Zheng94fcca92009-10-09 09:25:16 -04007000 }
Yan, Zheng1c4850e2009-09-21 15:55:59 -04007001
7002 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
7003 blocksize = btrfs_level_size(root, level - 1);
7004
7005 next = btrfs_find_tree_block(root, bytenr, blocksize);
7006 if (!next) {
7007 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
Miao Xie90d2c51d2010-03-25 12:37:12 +00007008 if (!next)
7009 return -ENOMEM;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04007010 reada = 1;
7011 }
7012 btrfs_tree_lock(next);
7013 btrfs_set_lock_blocking(next);
7014
Josef Bacik3173a182013-03-07 14:22:04 -05007015 ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1,
Yan, Zheng94fcca92009-10-09 09:25:16 -04007016 &wc->refs[level - 1],
7017 &wc->flags[level - 1]);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01007018 if (ret < 0) {
7019 btrfs_tree_unlock(next);
7020 return ret;
7021 }
7022
Simon Kirbyc2cf52e2013-03-19 22:41:23 +00007023 if (unlikely(wc->refs[level - 1] == 0)) {
7024 btrfs_err(root->fs_info, "Missing references.");
7025 BUG();
7026 }
Yan, Zheng94fcca92009-10-09 09:25:16 -04007027 *lookup_info = 0;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04007028
Yan, Zheng94fcca92009-10-09 09:25:16 -04007029 if (wc->stage == DROP_REFERENCE) {
Yan, Zheng1c4850e2009-09-21 15:55:59 -04007030 if (wc->refs[level - 1] > 1) {
Yan, Zheng94fcca92009-10-09 09:25:16 -04007031 if (level == 1 &&
7032 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7033 goto skip;
7034
Yan, Zheng1c4850e2009-09-21 15:55:59 -04007035 if (!wc->update_ref ||
7036 generation <= root->root_key.offset)
7037 goto skip;
7038
7039 btrfs_node_key_to_cpu(path->nodes[level], &key,
7040 path->slots[level]);
7041 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
7042 if (ret < 0)
7043 goto skip;
7044
7045 wc->stage = UPDATE_BACKREF;
7046 wc->shared_level = level - 1;
7047 }
Yan, Zheng94fcca92009-10-09 09:25:16 -04007048 } else {
7049 if (level == 1 &&
7050 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7051 goto skip;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04007052 }
7053
Chris Masonb9fab912012-05-06 07:23:47 -04007054 if (!btrfs_buffer_uptodate(next, generation, 0)) {
Yan, Zheng1c4850e2009-09-21 15:55:59 -04007055 btrfs_tree_unlock(next);
7056 free_extent_buffer(next);
7057 next = NULL;
Yan, Zheng94fcca92009-10-09 09:25:16 -04007058 *lookup_info = 1;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04007059 }
7060
7061 if (!next) {
7062 if (reada && level == 1)
7063 reada_walk_down(trans, root, wc, path);
7064 next = read_tree_block(root, bytenr, blocksize, generation);
Josef Bacik416bc652013-04-23 14:17:42 -04007065 if (!next || !extent_buffer_uptodate(next)) {
7066 free_extent_buffer(next);
Tsutomu Itoh97d9a8a2011-03-24 06:33:21 +00007067 return -EIO;
Josef Bacik416bc652013-04-23 14:17:42 -04007068 }
Yan, Zheng1c4850e2009-09-21 15:55:59 -04007069 btrfs_tree_lock(next);
7070 btrfs_set_lock_blocking(next);
7071 }
7072
7073 level--;
7074 BUG_ON(level != btrfs_header_level(next));
7075 path->nodes[level] = next;
7076 path->slots[level] = 0;
Chris Masonbd681512011-07-16 15:23:14 -04007077 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04007078 wc->level = level;
7079 if (wc->level == 1)
7080 wc->reada_slot = 0;
7081 return 0;
7082skip:
7083 wc->refs[level - 1] = 0;
7084 wc->flags[level - 1] = 0;
Yan, Zheng94fcca92009-10-09 09:25:16 -04007085 if (wc->stage == DROP_REFERENCE) {
7086 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
7087 parent = path->nodes[level]->start;
7088 } else {
7089 BUG_ON(root->root_key.objectid !=
7090 btrfs_header_owner(path->nodes[level]));
7091 parent = 0;
7092 }
Yan, Zheng1c4850e2009-09-21 15:55:59 -04007093
Yan, Zheng94fcca92009-10-09 09:25:16 -04007094 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
Arne Jansen66d7e7f2011-09-12 15:26:38 +02007095 root->root_key.objectid, level - 1, 0, 0);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01007096 BUG_ON(ret); /* -ENOMEM */
Yan, Zheng1c4850e2009-09-21 15:55:59 -04007097 }
Yan, Zheng1c4850e2009-09-21 15:55:59 -04007098 btrfs_tree_unlock(next);
7099 free_extent_buffer(next);
Yan, Zheng94fcca92009-10-09 09:25:16 -04007100 *lookup_info = 1;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04007101 return 1;
7102}
7103
7104/*
Liu Bo2c016dc2012-12-26 15:32:17 +08007105 * helper to process tree block while walking up the tree.
Yan Zheng2c47e6052009-06-27 21:07:35 -04007106 *
7107 * when wc->stage == DROP_REFERENCE, this function drops
7108 * reference count on the block.
7109 *
7110 * when wc->stage == UPDATE_BACKREF, this function changes
7111 * wc->stage back to DROP_REFERENCE if we changed wc->stage
7112 * to UPDATE_BACKREF previously while processing the block.
7113 *
7114 * NOTE: return value 1 means we should stop walking up.
7115 */
7116static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
7117 struct btrfs_root *root,
7118 struct btrfs_path *path,
7119 struct walk_control *wc)
7120{
Yan, Zhengf0486c62010-05-16 10:46:25 -04007121 int ret;
Yan Zheng2c47e6052009-06-27 21:07:35 -04007122 int level = wc->level;
7123 struct extent_buffer *eb = path->nodes[level];
7124 u64 parent = 0;
7125
7126 if (wc->stage == UPDATE_BACKREF) {
7127 BUG_ON(wc->shared_level < level);
7128 if (level < wc->shared_level)
7129 goto out;
7130
Yan Zheng2c47e6052009-06-27 21:07:35 -04007131 ret = find_next_key(path, level + 1, &wc->update_progress);
7132 if (ret > 0)
7133 wc->update_ref = 0;
7134
7135 wc->stage = DROP_REFERENCE;
7136 wc->shared_level = -1;
7137 path->slots[level] = 0;
7138
7139 /*
7140 * check reference count again if the block isn't locked.
7141 * we should start walking down the tree again if reference
7142 * count is one.
7143 */
7144 if (!path->locks[level]) {
7145 BUG_ON(level == 0);
7146 btrfs_tree_lock(eb);
7147 btrfs_set_lock_blocking(eb);
Chris Masonbd681512011-07-16 15:23:14 -04007148 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
Yan Zheng2c47e6052009-06-27 21:07:35 -04007149
7150 ret = btrfs_lookup_extent_info(trans, root,
Josef Bacik3173a182013-03-07 14:22:04 -05007151 eb->start, level, 1,
Yan Zheng2c47e6052009-06-27 21:07:35 -04007152 &wc->refs[level],
7153 &wc->flags[level]);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01007154 if (ret < 0) {
7155 btrfs_tree_unlock_rw(eb, path->locks[level]);
Liu Bo3268a242012-12-28 09:33:19 +00007156 path->locks[level] = 0;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01007157 return ret;
7158 }
Yan Zheng2c47e6052009-06-27 21:07:35 -04007159 BUG_ON(wc->refs[level] == 0);
7160 if (wc->refs[level] == 1) {
Chris Masonbd681512011-07-16 15:23:14 -04007161 btrfs_tree_unlock_rw(eb, path->locks[level]);
Liu Bo3268a242012-12-28 09:33:19 +00007162 path->locks[level] = 0;
Yan Zheng2c47e6052009-06-27 21:07:35 -04007163 return 1;
7164 }
Yan Zheng2c47e6052009-06-27 21:07:35 -04007165 }
7166 }
7167
7168 /* wc->stage == DROP_REFERENCE */
7169 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
7170
7171 if (wc->refs[level] == 1) {
7172 if (level == 0) {
7173 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
Arne Jansen66d7e7f2011-09-12 15:26:38 +02007174 ret = btrfs_dec_ref(trans, root, eb, 1,
7175 wc->for_reloc);
Yan Zheng2c47e6052009-06-27 21:07:35 -04007176 else
Arne Jansen66d7e7f2011-09-12 15:26:38 +02007177 ret = btrfs_dec_ref(trans, root, eb, 0,
7178 wc->for_reloc);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01007179 BUG_ON(ret); /* -ENOMEM */
Yan Zheng2c47e6052009-06-27 21:07:35 -04007180 }
7181 /* make block locked assertion in clean_tree_block happy */
7182 if (!path->locks[level] &&
7183 btrfs_header_generation(eb) == trans->transid) {
7184 btrfs_tree_lock(eb);
7185 btrfs_set_lock_blocking(eb);
Chris Masonbd681512011-07-16 15:23:14 -04007186 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
Yan Zheng2c47e6052009-06-27 21:07:35 -04007187 }
7188 clean_tree_block(trans, root, eb);
7189 }
7190
7191 if (eb == root->node) {
7192 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7193 parent = eb->start;
7194 else
7195 BUG_ON(root->root_key.objectid !=
7196 btrfs_header_owner(eb));
7197 } else {
7198 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7199 parent = path->nodes[level + 1]->start;
7200 else
7201 BUG_ON(root->root_key.objectid !=
7202 btrfs_header_owner(path->nodes[level + 1]));
7203 }
7204
Jan Schmidt5581a512012-05-16 17:04:52 +02007205 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
Yan Zheng2c47e6052009-06-27 21:07:35 -04007206out:
7207 wc->refs[level] = 0;
7208 wc->flags[level] = 0;
Yan, Zhengf0486c62010-05-16 10:46:25 -04007209 return 0;
Yan Zheng2c47e6052009-06-27 21:07:35 -04007210}
7211
Yan Zheng5d4f98a2009-06-10 10:45:14 -04007212static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
7213 struct btrfs_root *root,
Yan Zheng2c47e6052009-06-27 21:07:35 -04007214 struct btrfs_path *path,
7215 struct walk_control *wc)
Yan Zhengf82d02d2008-10-29 14:49:05 -04007216{
Yan Zheng2c47e6052009-06-27 21:07:35 -04007217 int level = wc->level;
Yan, Zheng94fcca92009-10-09 09:25:16 -04007218 int lookup_info = 1;
Yan Zhengf82d02d2008-10-29 14:49:05 -04007219 int ret;
7220
Yan Zheng2c47e6052009-06-27 21:07:35 -04007221 while (level >= 0) {
Yan, Zheng94fcca92009-10-09 09:25:16 -04007222 ret = walk_down_proc(trans, root, path, wc, lookup_info);
Yan Zheng2c47e6052009-06-27 21:07:35 -04007223 if (ret > 0)
Yan Zhengf82d02d2008-10-29 14:49:05 -04007224 break;
Yan Zhengf82d02d2008-10-29 14:49:05 -04007225
Yan Zheng2c47e6052009-06-27 21:07:35 -04007226 if (level == 0)
7227 break;
7228
Yan, Zheng7a7965f2010-02-01 02:41:17 +00007229 if (path->slots[level] >=
7230 btrfs_header_nritems(path->nodes[level]))
7231 break;
7232
Yan, Zheng94fcca92009-10-09 09:25:16 -04007233 ret = do_walk_down(trans, root, path, wc, &lookup_info);
Yan, Zheng1c4850e2009-09-21 15:55:59 -04007234 if (ret > 0) {
7235 path->slots[level]++;
7236 continue;
Miao Xie90d2c51d2010-03-25 12:37:12 +00007237 } else if (ret < 0)
7238 return ret;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04007239 level = wc->level;
Yan Zhengf82d02d2008-10-29 14:49:05 -04007240 }
Yan Zhengf82d02d2008-10-29 14:49:05 -04007241 return 0;
7242}
7243
Chris Masond3977122009-01-05 21:25:51 -05007244static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
Chris Mason98ed5172008-01-03 10:01:48 -05007245 struct btrfs_root *root,
Yan Zhengf82d02d2008-10-29 14:49:05 -04007246 struct btrfs_path *path,
Yan Zheng2c47e6052009-06-27 21:07:35 -04007247 struct walk_control *wc, int max_level)
Chris Mason20524f02007-03-10 06:35:47 -05007248{
Yan Zheng2c47e6052009-06-27 21:07:35 -04007249 int level = wc->level;
Chris Mason20524f02007-03-10 06:35:47 -05007250 int ret;
Chris Mason9f3a7422007-08-07 15:52:19 -04007251
Yan Zheng2c47e6052009-06-27 21:07:35 -04007252 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
7253 while (level < max_level && path->nodes[level]) {
7254 wc->level = level;
7255 if (path->slots[level] + 1 <
7256 btrfs_header_nritems(path->nodes[level])) {
7257 path->slots[level]++;
Chris Mason20524f02007-03-10 06:35:47 -05007258 return 0;
7259 } else {
Yan Zheng2c47e6052009-06-27 21:07:35 -04007260 ret = walk_up_proc(trans, root, path, wc);
7261 if (ret > 0)
7262 return 0;
Chris Masonbd56b302009-02-04 09:27:02 -05007263
Yan Zheng2c47e6052009-06-27 21:07:35 -04007264 if (path->locks[level]) {
Chris Masonbd681512011-07-16 15:23:14 -04007265 btrfs_tree_unlock_rw(path->nodes[level],
7266 path->locks[level]);
Yan Zheng2c47e6052009-06-27 21:07:35 -04007267 path->locks[level] = 0;
Yan Zhengf82d02d2008-10-29 14:49:05 -04007268 }
Yan Zheng2c47e6052009-06-27 21:07:35 -04007269 free_extent_buffer(path->nodes[level]);
7270 path->nodes[level] = NULL;
7271 level++;
Chris Mason20524f02007-03-10 06:35:47 -05007272 }
7273 }
7274 return 1;
7275}
7276
Chris Mason9aca1d52007-03-13 11:09:37 -04007277/*
Yan Zheng2c47e6052009-06-27 21:07:35 -04007278 * drop a subvolume tree.
7279 *
7280 * this function traverses the tree freeing any blocks that only
7281 * referenced by the tree.
7282 *
7283 * when a shared tree block is found. this function decreases its
7284 * reference count by one. if update_ref is true, this function
7285 * also make sure backrefs for the shared block and all lower level
7286 * blocks are properly updated.
David Sterba9d1a2a32013-03-12 15:13:28 +00007287 *
7288 * If called with for_reloc == 0, may exit early with -EAGAIN
Chris Mason9aca1d52007-03-13 11:09:37 -04007289 */
Jeff Mahoney2c536792011-10-03 23:22:41 -04007290int btrfs_drop_snapshot(struct btrfs_root *root,
Arne Jansen66d7e7f2011-09-12 15:26:38 +02007291 struct btrfs_block_rsv *block_rsv, int update_ref,
7292 int for_reloc)
Chris Mason20524f02007-03-10 06:35:47 -05007293{
Chris Mason5caf2a02007-04-02 11:20:42 -04007294 struct btrfs_path *path;
Yan Zheng2c47e6052009-06-27 21:07:35 -04007295 struct btrfs_trans_handle *trans;
7296 struct btrfs_root *tree_root = root->fs_info->tree_root;
Chris Mason9f3a7422007-08-07 15:52:19 -04007297 struct btrfs_root_item *root_item = &root->root_item;
Yan Zheng2c47e6052009-06-27 21:07:35 -04007298 struct walk_control *wc;
7299 struct btrfs_key key;
7300 int err = 0;
7301 int ret;
7302 int level;
Josef Bacik0183a8e2013-07-17 19:30:20 -04007303 bool root_dropped = false;
Chris Mason20524f02007-03-10 06:35:47 -05007304
Chris Mason5caf2a02007-04-02 11:20:42 -04007305 path = btrfs_alloc_path();
Tsutomu Itohcb1b69f2011-08-09 07:11:13 +00007306 if (!path) {
7307 err = -ENOMEM;
7308 goto out;
7309 }
Chris Mason20524f02007-03-10 06:35:47 -05007310
Yan Zheng2c47e6052009-06-27 21:07:35 -04007311 wc = kzalloc(sizeof(*wc), GFP_NOFS);
Mark Fasheh38a1a912011-07-13 10:59:59 -07007312 if (!wc) {
7313 btrfs_free_path(path);
Tsutomu Itohcb1b69f2011-08-09 07:11:13 +00007314 err = -ENOMEM;
7315 goto out;
Mark Fasheh38a1a912011-07-13 10:59:59 -07007316 }
Yan Zheng2c47e6052009-06-27 21:07:35 -04007317
Yan, Zhenga22285a2010-05-16 10:48:46 -04007318 trans = btrfs_start_transaction(tree_root, 0);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01007319 if (IS_ERR(trans)) {
7320 err = PTR_ERR(trans);
7321 goto out_free;
7322 }
Tsutomu Itoh98d5dc12011-01-20 06:19:37 +00007323
Yan, Zheng3fd0a552010-05-16 10:49:59 -04007324 if (block_rsv)
7325 trans->block_rsv = block_rsv;
Yan Zheng2c47e6052009-06-27 21:07:35 -04007326
Chris Mason9f3a7422007-08-07 15:52:19 -04007327 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
Yan Zheng2c47e6052009-06-27 21:07:35 -04007328 level = btrfs_header_level(root->node);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04007329 path->nodes[level] = btrfs_lock_root_node(root);
7330 btrfs_set_lock_blocking(path->nodes[level]);
Chris Mason9f3a7422007-08-07 15:52:19 -04007331 path->slots[level] = 0;
Chris Masonbd681512011-07-16 15:23:14 -04007332 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
Yan Zheng2c47e6052009-06-27 21:07:35 -04007333 memset(&wc->update_progress, 0,
7334 sizeof(wc->update_progress));
Chris Mason9f3a7422007-08-07 15:52:19 -04007335 } else {
Chris Mason9f3a7422007-08-07 15:52:19 -04007336 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
Yan Zheng2c47e6052009-06-27 21:07:35 -04007337 memcpy(&wc->update_progress, &key,
7338 sizeof(wc->update_progress));
7339
Chris Mason6702ed42007-08-07 16:15:09 -04007340 level = root_item->drop_level;
Yan Zheng2c47e6052009-06-27 21:07:35 -04007341 BUG_ON(level == 0);
Chris Mason6702ed42007-08-07 16:15:09 -04007342 path->lowest_level = level;
Yan Zheng2c47e6052009-06-27 21:07:35 -04007343 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7344 path->lowest_level = 0;
7345 if (ret < 0) {
7346 err = ret;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01007347 goto out_end_trans;
Chris Mason9f3a7422007-08-07 15:52:19 -04007348 }
Yan, Zheng1c4850e2009-09-21 15:55:59 -04007349 WARN_ON(ret > 0);
Yan Zheng2c47e6052009-06-27 21:07:35 -04007350
Chris Mason7d9eb122008-07-08 14:19:17 -04007351 /*
7352 * unlock our path, this is safe because only this
7353 * function is allowed to delete this snapshot
7354 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -04007355 btrfs_unlock_up_safe(path, 0);
Chris Mason9aca1d52007-03-13 11:09:37 -04007356
Yan Zheng2c47e6052009-06-27 21:07:35 -04007357 level = btrfs_header_level(root->node);
7358 while (1) {
7359 btrfs_tree_lock(path->nodes[level]);
7360 btrfs_set_lock_blocking(path->nodes[level]);
Josef Bacik14e86562013-07-15 12:41:42 -04007361 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
Yan Zheng2c47e6052009-06-27 21:07:35 -04007362
7363 ret = btrfs_lookup_extent_info(trans, root,
7364 path->nodes[level]->start,
Josef Bacik3173a182013-03-07 14:22:04 -05007365 level, 1, &wc->refs[level],
Yan Zheng2c47e6052009-06-27 21:07:35 -04007366 &wc->flags[level]);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01007367 if (ret < 0) {
7368 err = ret;
7369 goto out_end_trans;
7370 }
Yan Zheng2c47e6052009-06-27 21:07:35 -04007371 BUG_ON(wc->refs[level] == 0);
7372
7373 if (level == root_item->drop_level)
7374 break;
7375
7376 btrfs_tree_unlock(path->nodes[level]);
Josef Bacik14e86562013-07-15 12:41:42 -04007377 path->locks[level] = 0;
Yan Zheng2c47e6052009-06-27 21:07:35 -04007378 WARN_ON(wc->refs[level] != 1);
7379 level--;
7380 }
7381 }
7382
7383 wc->level = level;
7384 wc->shared_level = -1;
7385 wc->stage = DROP_REFERENCE;
7386 wc->update_ref = update_ref;
7387 wc->keep_locks = 0;
Arne Jansen66d7e7f2011-09-12 15:26:38 +02007388 wc->for_reloc = for_reloc;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04007389 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
Yan Zheng2c47e6052009-06-27 21:07:35 -04007390
7391 while (1) {
David Sterba9d1a2a32013-03-12 15:13:28 +00007392 if (!for_reloc && btrfs_fs_closing(root->fs_info)) {
7393 pr_debug("btrfs: drop snapshot early exit\n");
7394 err = -EAGAIN;
7395 goto out_end_trans;
7396 }
7397
Yan Zheng2c47e6052009-06-27 21:07:35 -04007398 ret = walk_down_tree(trans, root, path, wc);
7399 if (ret < 0) {
7400 err = ret;
Chris Masone7a84562008-06-25 16:01:31 -04007401 break;
7402 }
Yan Zheng2c47e6052009-06-27 21:07:35 -04007403
7404 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
7405 if (ret < 0) {
7406 err = ret;
7407 break;
7408 }
7409
7410 if (ret > 0) {
7411 BUG_ON(wc->stage != DROP_REFERENCE);
7412 break;
7413 }
7414
7415 if (wc->stage == DROP_REFERENCE) {
7416 level = wc->level;
7417 btrfs_node_key(path->nodes[level],
7418 &root_item->drop_progress,
7419 path->slots[level]);
7420 root_item->drop_level = level;
7421 }
7422
7423 BUG_ON(wc->level == 0);
Yan, Zheng3fd0a552010-05-16 10:49:59 -04007424 if (btrfs_should_end_transaction(trans, tree_root)) {
Yan Zheng2c47e6052009-06-27 21:07:35 -04007425 ret = btrfs_update_root(trans, tree_root,
7426 &root->root_key,
7427 root_item);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01007428 if (ret) {
7429 btrfs_abort_transaction(trans, tree_root, ret);
7430 err = ret;
7431 goto out_end_trans;
7432 }
Yan Zheng2c47e6052009-06-27 21:07:35 -04007433
Yan, Zheng3fd0a552010-05-16 10:49:59 -04007434 btrfs_end_transaction_throttle(trans, tree_root);
Yan, Zhenga22285a2010-05-16 10:48:46 -04007435 trans = btrfs_start_transaction(tree_root, 0);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01007436 if (IS_ERR(trans)) {
7437 err = PTR_ERR(trans);
7438 goto out_free;
7439 }
Yan, Zheng3fd0a552010-05-16 10:49:59 -04007440 if (block_rsv)
7441 trans->block_rsv = block_rsv;
Chris Masonc3e69d52009-03-13 10:17:05 -04007442 }
Chris Mason20524f02007-03-10 06:35:47 -05007443 }
David Sterbab3b4aa72011-04-21 01:20:15 +02007444 btrfs_release_path(path);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01007445 if (err)
7446 goto out_end_trans;
Yan Zheng2c47e6052009-06-27 21:07:35 -04007447
7448 ret = btrfs_del_root(trans, tree_root, &root->root_key);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01007449 if (ret) {
7450 btrfs_abort_transaction(trans, tree_root, ret);
7451 goto out_end_trans;
7452 }
Yan Zheng2c47e6052009-06-27 21:07:35 -04007453
Yan, Zheng76dda932009-09-21 16:00:26 -04007454 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
7455 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
7456 NULL, NULL);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01007457 if (ret < 0) {
7458 btrfs_abort_transaction(trans, tree_root, ret);
7459 err = ret;
7460 goto out_end_trans;
7461 } else if (ret > 0) {
Josef Bacik84cd9482010-12-08 12:24:01 -05007462 /* if we fail to delete the orphan item this time
7463 * around, it'll get picked up the next time.
7464 *
7465 * The most common failure here is just -ENOENT.
7466 */
7467 btrfs_del_orphan_item(trans, tree_root,
7468 root->root_key.objectid);
Yan, Zheng76dda932009-09-21 16:00:26 -04007469 }
7470 }
7471
7472 if (root->in_radix) {
7473 btrfs_free_fs_root(tree_root->fs_info, root);
7474 } else {
7475 free_extent_buffer(root->node);
7476 free_extent_buffer(root->commit_root);
7477 kfree(root);
7478 }
Josef Bacik0183a8e2013-07-17 19:30:20 -04007479 root_dropped = true;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01007480out_end_trans:
Yan, Zheng3fd0a552010-05-16 10:49:59 -04007481 btrfs_end_transaction_throttle(trans, tree_root);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01007482out_free:
Yan Zheng2c47e6052009-06-27 21:07:35 -04007483 kfree(wc);
Chris Mason5caf2a02007-04-02 11:20:42 -04007484 btrfs_free_path(path);
Tsutomu Itohcb1b69f2011-08-09 07:11:13 +00007485out:
Josef Bacik0183a8e2013-07-17 19:30:20 -04007486 /*
7487 * So if we need to stop dropping the snapshot for whatever reason we
7488 * need to make sure to add it back to the dead root list so that we
7489 * keep trying to do the work later. This also cleans up roots if we
7490 * don't have it in the radix (like when we recover after a power fail
7491 * or unmount) so we don't leak memory.
7492 */
7493 if (root_dropped == false)
7494 btrfs_add_dead_root(root);
Wang Shilong5c61a3d2014-01-07 17:26:58 +08007495 if (err && err != -EAGAIN)
Tsutomu Itohcb1b69f2011-08-09 07:11:13 +00007496 btrfs_std_error(root->fs_info, err);
Jeff Mahoney2c536792011-10-03 23:22:41 -04007497 return err;
Chris Mason20524f02007-03-10 06:35:47 -05007498}
Chris Mason9078a3e2007-04-26 16:46:15 -04007499
Yan Zheng2c47e6052009-06-27 21:07:35 -04007500/*
7501 * drop subtree rooted at tree block 'node'.
7502 *
7503 * NOTE: this function will unlock and release tree block 'node'
Arne Jansen66d7e7f2011-09-12 15:26:38 +02007504 * only used by relocation code
Yan Zheng2c47e6052009-06-27 21:07:35 -04007505 */
Yan Zhengf82d02d2008-10-29 14:49:05 -04007506int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
7507 struct btrfs_root *root,
7508 struct extent_buffer *node,
7509 struct extent_buffer *parent)
7510{
7511 struct btrfs_path *path;
Yan Zheng2c47e6052009-06-27 21:07:35 -04007512 struct walk_control *wc;
Yan Zhengf82d02d2008-10-29 14:49:05 -04007513 int level;
7514 int parent_level;
7515 int ret = 0;
7516 int wret;
7517
Yan Zheng2c47e6052009-06-27 21:07:35 -04007518 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
7519
Yan Zhengf82d02d2008-10-29 14:49:05 -04007520 path = btrfs_alloc_path();
Tsutomu Itohdb5b4932011-03-23 08:14:16 +00007521 if (!path)
7522 return -ENOMEM;
Yan Zhengf82d02d2008-10-29 14:49:05 -04007523
Yan Zheng2c47e6052009-06-27 21:07:35 -04007524 wc = kzalloc(sizeof(*wc), GFP_NOFS);
Tsutomu Itohdb5b4932011-03-23 08:14:16 +00007525 if (!wc) {
7526 btrfs_free_path(path);
7527 return -ENOMEM;
7528 }
Yan Zheng2c47e6052009-06-27 21:07:35 -04007529
Chris Masonb9447ef2009-03-09 11:45:38 -04007530 btrfs_assert_tree_locked(parent);
Yan Zhengf82d02d2008-10-29 14:49:05 -04007531 parent_level = btrfs_header_level(parent);
7532 extent_buffer_get(parent);
7533 path->nodes[parent_level] = parent;
7534 path->slots[parent_level] = btrfs_header_nritems(parent);
7535
Chris Masonb9447ef2009-03-09 11:45:38 -04007536 btrfs_assert_tree_locked(node);
Yan Zhengf82d02d2008-10-29 14:49:05 -04007537 level = btrfs_header_level(node);
Yan Zhengf82d02d2008-10-29 14:49:05 -04007538 path->nodes[level] = node;
7539 path->slots[level] = 0;
Chris Masonbd681512011-07-16 15:23:14 -04007540 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
Yan Zheng2c47e6052009-06-27 21:07:35 -04007541
7542 wc->refs[parent_level] = 1;
7543 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7544 wc->level = level;
7545 wc->shared_level = -1;
7546 wc->stage = DROP_REFERENCE;
7547 wc->update_ref = 0;
7548 wc->keep_locks = 1;
Arne Jansen66d7e7f2011-09-12 15:26:38 +02007549 wc->for_reloc = 1;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04007550 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
Yan Zhengf82d02d2008-10-29 14:49:05 -04007551
7552 while (1) {
Yan Zheng2c47e6052009-06-27 21:07:35 -04007553 wret = walk_down_tree(trans, root, path, wc);
7554 if (wret < 0) {
Yan Zhengf82d02d2008-10-29 14:49:05 -04007555 ret = wret;
Yan Zhengf82d02d2008-10-29 14:49:05 -04007556 break;
Yan Zheng2c47e6052009-06-27 21:07:35 -04007557 }
Yan Zhengf82d02d2008-10-29 14:49:05 -04007558
Yan Zheng2c47e6052009-06-27 21:07:35 -04007559 wret = walk_up_tree(trans, root, path, wc, parent_level);
Yan Zhengf82d02d2008-10-29 14:49:05 -04007560 if (wret < 0)
7561 ret = wret;
7562 if (wret != 0)
7563 break;
7564 }
7565
Yan Zheng2c47e6052009-06-27 21:07:35 -04007566 kfree(wc);
Yan Zhengf82d02d2008-10-29 14:49:05 -04007567 btrfs_free_path(path);
7568 return ret;
7569}
7570
Chris Masonec44a352008-04-28 15:29:52 -04007571static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7572{
7573 u64 num_devices;
Ilya Dryomovfc67c452012-03-27 17:09:17 +03007574 u64 stripped;
Chris Masonec44a352008-04-28 15:29:52 -04007575
Ilya Dryomovfc67c452012-03-27 17:09:17 +03007576 /*
7577 * if restripe for this chunk_type is on pick target profile and
7578 * return, otherwise do the usual balance
7579 */
7580 stripped = get_restripe_target(root->fs_info, flags);
7581 if (stripped)
7582 return extended_to_chunk(stripped);
Ilya Dryomove4d8ec02012-01-16 22:04:48 +02007583
Chris Masoncd02dca2010-12-13 14:56:23 -05007584 /*
7585 * we add in the count of missing devices because we want
7586 * to make sure that any RAID levels on a degraded FS
7587 * continue to be honored.
7588 */
7589 num_devices = root->fs_info->fs_devices->rw_devices +
7590 root->fs_info->fs_devices->missing_devices;
7591
Ilya Dryomovfc67c452012-03-27 17:09:17 +03007592 stripped = BTRFS_BLOCK_GROUP_RAID0 |
David Woodhouse53b381b2013-01-29 18:40:14 -05007593 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
Ilya Dryomovfc67c452012-03-27 17:09:17 +03007594 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7595
Chris Masonec44a352008-04-28 15:29:52 -04007596 if (num_devices == 1) {
7597 stripped |= BTRFS_BLOCK_GROUP_DUP;
7598 stripped = flags & ~stripped;
7599
7600 /* turn raid0 into single device chunks */
7601 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7602 return stripped;
7603
7604 /* turn mirroring into duplication */
7605 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7606 BTRFS_BLOCK_GROUP_RAID10))
7607 return stripped | BTRFS_BLOCK_GROUP_DUP;
Chris Masonec44a352008-04-28 15:29:52 -04007608 } else {
7609 /* they already had raid on here, just return */
Chris Masonec44a352008-04-28 15:29:52 -04007610 if (flags & stripped)
7611 return flags;
7612
7613 stripped |= BTRFS_BLOCK_GROUP_DUP;
7614 stripped = flags & ~stripped;
7615
7616 /* switch duplicated blocks with raid1 */
7617 if (flags & BTRFS_BLOCK_GROUP_DUP)
7618 return stripped | BTRFS_BLOCK_GROUP_RAID1;
7619
Ilya Dryomove3176ca2012-03-27 17:09:16 +03007620 /* this is drive concat, leave it alone */
Chris Masonec44a352008-04-28 15:29:52 -04007621 }
Ilya Dryomove3176ca2012-03-27 17:09:16 +03007622
Chris Masonec44a352008-04-28 15:29:52 -04007623 return flags;
7624}
7625
Miao Xie199c36e2011-07-15 10:34:36 +00007626static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
Chris Mason0ef3e662008-05-24 14:04:53 -04007627{
Yan, Zhengf0486c62010-05-16 10:46:25 -04007628 struct btrfs_space_info *sinfo = cache->space_info;
7629 u64 num_bytes;
Miao Xie199c36e2011-07-15 10:34:36 +00007630 u64 min_allocable_bytes;
Yan, Zhengf0486c62010-05-16 10:46:25 -04007631 int ret = -ENOSPC;
Chris Mason0ef3e662008-05-24 14:04:53 -04007632
Chris Masonc286ac42008-07-22 23:06:41 -04007633
Miao Xie199c36e2011-07-15 10:34:36 +00007634 /*
7635 * We need some metadata space and system metadata space for
7636 * allocating chunks in some corner cases until we force to set
7637 * it to be readonly.
7638 */
7639 if ((sinfo->flags &
7640 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
7641 !force)
7642 min_allocable_bytes = 1 * 1024 * 1024;
7643 else
7644 min_allocable_bytes = 0;
7645
Yan, Zhengf0486c62010-05-16 10:46:25 -04007646 spin_lock(&sinfo->lock);
7647 spin_lock(&cache->lock);
WuBo61cfea92011-07-26 03:30:11 +00007648
7649 if (cache->ro) {
7650 ret = 0;
7651 goto out;
7652 }
7653
Yan, Zhengf0486c62010-05-16 10:46:25 -04007654 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7655 cache->bytes_super - btrfs_block_group_used(&cache->item);
Chris Mason7d9eb122008-07-08 14:19:17 -04007656
Yan, Zhengf0486c62010-05-16 10:46:25 -04007657 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
Josef Bacik37be25b2011-08-05 10:25:38 -04007658 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
7659 min_allocable_bytes <= sinfo->total_bytes) {
Yan, Zhengf0486c62010-05-16 10:46:25 -04007660 sinfo->bytes_readonly += num_bytes;
Yan, Zhengf0486c62010-05-16 10:46:25 -04007661 cache->ro = 1;
7662 ret = 0;
7663 }
WuBo61cfea92011-07-26 03:30:11 +00007664out:
Yan, Zhengf0486c62010-05-16 10:46:25 -04007665 spin_unlock(&cache->lock);
7666 spin_unlock(&sinfo->lock);
7667 return ret;
Chris Mason0ef3e662008-05-24 14:04:53 -04007668}
7669
Yan, Zhengf0486c62010-05-16 10:46:25 -04007670int btrfs_set_block_group_ro(struct btrfs_root *root,
7671 struct btrfs_block_group_cache *cache)
Yan Zheng5d4f98a2009-06-10 10:45:14 -04007672
7673{
Yan, Zhengf0486c62010-05-16 10:46:25 -04007674 struct btrfs_trans_handle *trans;
7675 u64 alloc_flags;
7676 int ret;
7677
7678 BUG_ON(cache->ro);
7679
Josef Bacik7a7eaa42011-04-13 12:54:33 -04007680 trans = btrfs_join_transaction(root);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01007681 if (IS_ERR(trans))
7682 return PTR_ERR(trans);
Yan, Zhengf0486c62010-05-16 10:46:25 -04007683
7684 alloc_flags = update_block_group_flags(root, cache->flags);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01007685 if (alloc_flags != cache->flags) {
Josef Bacik698d0082012-09-12 14:08:47 -04007686 ret = do_chunk_alloc(trans, root, alloc_flags,
Jeff Mahoney79787ea2012-03-12 16:03:00 +01007687 CHUNK_ALLOC_FORCE);
7688 if (ret < 0)
7689 goto out;
7690 }
Yan, Zhengf0486c62010-05-16 10:46:25 -04007691
Miao Xie199c36e2011-07-15 10:34:36 +00007692 ret = set_block_group_ro(cache, 0);
Yan, Zhengf0486c62010-05-16 10:46:25 -04007693 if (!ret)
7694 goto out;
7695 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
Josef Bacik698d0082012-09-12 14:08:47 -04007696 ret = do_chunk_alloc(trans, root, alloc_flags,
Chris Mason0e4f8f82011-04-15 16:05:44 -04007697 CHUNK_ALLOC_FORCE);
Yan, Zhengf0486c62010-05-16 10:46:25 -04007698 if (ret < 0)
7699 goto out;
Miao Xie199c36e2011-07-15 10:34:36 +00007700 ret = set_block_group_ro(cache, 0);
Yan, Zhengf0486c62010-05-16 10:46:25 -04007701out:
7702 btrfs_end_transaction(trans, root);
7703 return ret;
7704}
7705
Chris Masonc87f08c2011-02-16 13:57:04 -05007706int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
7707 struct btrfs_root *root, u64 type)
7708{
7709 u64 alloc_flags = get_alloc_profile(root, type);
Josef Bacik698d0082012-09-12 14:08:47 -04007710 return do_chunk_alloc(trans, root, alloc_flags,
Chris Mason0e4f8f82011-04-15 16:05:44 -04007711 CHUNK_ALLOC_FORCE);
Chris Masonc87f08c2011-02-16 13:57:04 -05007712}
7713
Miao Xie6d07bce2011-01-05 10:07:31 +00007714/*
7715 * helper to account the unused space of all the readonly block group in the
7716 * list. takes mirrors into account.
7717 */
7718static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
7719{
7720 struct btrfs_block_group_cache *block_group;
7721 u64 free_bytes = 0;
7722 int factor;
7723
7724 list_for_each_entry(block_group, groups_list, list) {
7725 spin_lock(&block_group->lock);
7726
7727 if (!block_group->ro) {
7728 spin_unlock(&block_group->lock);
7729 continue;
7730 }
7731
7732 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
7733 BTRFS_BLOCK_GROUP_RAID10 |
7734 BTRFS_BLOCK_GROUP_DUP))
7735 factor = 2;
7736 else
7737 factor = 1;
7738
7739 free_bytes += (block_group->key.offset -
7740 btrfs_block_group_used(&block_group->item)) *
7741 factor;
7742
7743 spin_unlock(&block_group->lock);
7744 }
7745
7746 return free_bytes;
7747}
7748
7749/*
7750 * helper to account the unused space of all the readonly block group in the
7751 * space_info. takes mirrors into account.
7752 */
7753u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
7754{
7755 int i;
7756 u64 free_bytes = 0;
7757
7758 spin_lock(&sinfo->lock);
7759
7760 for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
7761 if (!list_empty(&sinfo->block_groups[i]))
7762 free_bytes += __btrfs_get_ro_block_group_free_space(
7763 &sinfo->block_groups[i]);
7764
7765 spin_unlock(&sinfo->lock);
7766
7767 return free_bytes;
7768}
7769
Jeff Mahoney143bede2012-03-01 14:56:26 +01007770void btrfs_set_block_group_rw(struct btrfs_root *root,
Yan, Zhengf0486c62010-05-16 10:46:25 -04007771 struct btrfs_block_group_cache *cache)
7772{
7773 struct btrfs_space_info *sinfo = cache->space_info;
7774 u64 num_bytes;
7775
7776 BUG_ON(!cache->ro);
7777
7778 spin_lock(&sinfo->lock);
7779 spin_lock(&cache->lock);
7780 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7781 cache->bytes_super - btrfs_block_group_used(&cache->item);
7782 sinfo->bytes_readonly -= num_bytes;
7783 cache->ro = 0;
7784 spin_unlock(&cache->lock);
7785 spin_unlock(&sinfo->lock);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04007786}
7787
Josef Bacikba1bf482009-09-11 16:11:19 -04007788/*
7789 * checks to see if its even possible to relocate this block group.
7790 *
7791 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
7792 * ok to go ahead and try.
7793 */
7794int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
Zheng Yan1a40e232008-09-26 10:09:34 -04007795{
Zheng Yan1a40e232008-09-26 10:09:34 -04007796 struct btrfs_block_group_cache *block_group;
Josef Bacikba1bf482009-09-11 16:11:19 -04007797 struct btrfs_space_info *space_info;
7798 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
7799 struct btrfs_device *device;
liubocdcb7252011-08-03 10:15:25 +00007800 u64 min_free;
Josef Bacik6719db62011-08-20 08:29:51 -04007801 u64 dev_min = 1;
7802 u64 dev_nr = 0;
Ilya Dryomov4a5e98f2012-03-27 17:09:17 +03007803 u64 target;
liubocdcb7252011-08-03 10:15:25 +00007804 int index;
Josef Bacikba1bf482009-09-11 16:11:19 -04007805 int full = 0;
7806 int ret = 0;
Chris Masonedbd8d42007-12-21 16:27:24 -05007807
Josef Bacikba1bf482009-09-11 16:11:19 -04007808 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
Zheng Yan1a40e232008-09-26 10:09:34 -04007809
Josef Bacikba1bf482009-09-11 16:11:19 -04007810 /* odd, couldn't find the block group, leave it alone */
7811 if (!block_group)
7812 return -1;
Chris Masonedbd8d42007-12-21 16:27:24 -05007813
liubocdcb7252011-08-03 10:15:25 +00007814 min_free = btrfs_block_group_used(&block_group->item);
7815
Josef Bacikba1bf482009-09-11 16:11:19 -04007816 /* no bytes used, we're good */
liubocdcb7252011-08-03 10:15:25 +00007817 if (!min_free)
Josef Bacikba1bf482009-09-11 16:11:19 -04007818 goto out;
Chris Mason323da792008-05-09 11:46:48 -04007819
Josef Bacikba1bf482009-09-11 16:11:19 -04007820 space_info = block_group->space_info;
7821 spin_lock(&space_info->lock);
Chris Mason323da792008-05-09 11:46:48 -04007822
Josef Bacikba1bf482009-09-11 16:11:19 -04007823 full = space_info->full;
Zheng Yan1a40e232008-09-26 10:09:34 -04007824
Josef Bacikba1bf482009-09-11 16:11:19 -04007825 /*
7826 * if this is the last block group we have in this space, we can't
Chris Mason7ce618d2009-09-22 14:48:44 -04007827 * relocate it unless we're able to allocate a new chunk below.
7828 *
7829 * Otherwise, we need to make sure we have room in the space to handle
7830 * all of the extents from this block group. If we can, we're good
Josef Bacikba1bf482009-09-11 16:11:19 -04007831 */
Chris Mason7ce618d2009-09-22 14:48:44 -04007832 if ((space_info->total_bytes != block_group->key.offset) &&
liubocdcb7252011-08-03 10:15:25 +00007833 (space_info->bytes_used + space_info->bytes_reserved +
7834 space_info->bytes_pinned + space_info->bytes_readonly +
7835 min_free < space_info->total_bytes)) {
Josef Bacikba1bf482009-09-11 16:11:19 -04007836 spin_unlock(&space_info->lock);
7837 goto out;
7838 }
7839 spin_unlock(&space_info->lock);
Zheng Yan1a40e232008-09-26 10:09:34 -04007840
Josef Bacikba1bf482009-09-11 16:11:19 -04007841 /*
7842 * ok we don't have enough space, but maybe we have free space on our
7843 * devices to allocate new chunks for relocation, so loop through our
Ilya Dryomov4a5e98f2012-03-27 17:09:17 +03007844 * alloc devices and guess if we have enough space. if this block
7845 * group is going to be restriped, run checks against the target
7846 * profile instead of the current one.
Josef Bacikba1bf482009-09-11 16:11:19 -04007847 */
7848 ret = -1;
Chris Mason4313b392008-01-03 09:08:48 -05007849
liubocdcb7252011-08-03 10:15:25 +00007850 /*
7851 * index:
7852 * 0: raid10
7853 * 1: raid1
7854 * 2: dup
7855 * 3: raid0
7856 * 4: single
7857 */
Ilya Dryomov4a5e98f2012-03-27 17:09:17 +03007858 target = get_restripe_target(root->fs_info, block_group->flags);
7859 if (target) {
Liu Bo31e50222012-11-21 14:18:10 +00007860 index = __get_raid_index(extended_to_chunk(target));
Ilya Dryomov4a5e98f2012-03-27 17:09:17 +03007861 } else {
7862 /*
7863 * this is just a balance, so if we were marked as full
7864 * we know there is no space for a new chunk
7865 */
7866 if (full)
7867 goto out;
7868
7869 index = get_block_group_index(block_group);
7870 }
7871
Miao Xiee6ec7162013-01-17 05:38:51 +00007872 if (index == BTRFS_RAID_RAID10) {
liubocdcb7252011-08-03 10:15:25 +00007873 dev_min = 4;
Josef Bacik6719db62011-08-20 08:29:51 -04007874 /* Divide by 2 */
7875 min_free >>= 1;
Miao Xiee6ec7162013-01-17 05:38:51 +00007876 } else if (index == BTRFS_RAID_RAID1) {
liubocdcb7252011-08-03 10:15:25 +00007877 dev_min = 2;
Miao Xiee6ec7162013-01-17 05:38:51 +00007878 } else if (index == BTRFS_RAID_DUP) {
Josef Bacik6719db62011-08-20 08:29:51 -04007879 /* Multiply by 2 */
7880 min_free <<= 1;
Miao Xiee6ec7162013-01-17 05:38:51 +00007881 } else if (index == BTRFS_RAID_RAID0) {
liubocdcb7252011-08-03 10:15:25 +00007882 dev_min = fs_devices->rw_devices;
Josef Bacik6719db62011-08-20 08:29:51 -04007883 do_div(min_free, dev_min);
liubocdcb7252011-08-03 10:15:25 +00007884 }
7885
Josef Bacikba1bf482009-09-11 16:11:19 -04007886 mutex_lock(&root->fs_info->chunk_mutex);
7887 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
Miao Xie7bfc8372011-01-05 10:07:26 +00007888 u64 dev_offset;
Chris Masonea8c2812008-08-04 23:17:27 -04007889
Josef Bacikba1bf482009-09-11 16:11:19 -04007890 /*
7891 * check to make sure we can actually find a chunk with enough
7892 * space to fit our block group in.
7893 */
Stefan Behrens63a212a2012-11-05 18:29:28 +01007894 if (device->total_bytes > device->bytes_used + min_free &&
7895 !device->is_tgtdev_for_dev_replace) {
Li Zefan125ccb02011-12-08 15:07:24 +08007896 ret = find_free_dev_extent(device, min_free,
Miao Xie7bfc8372011-01-05 10:07:26 +00007897 &dev_offset, NULL);
Josef Bacikba1bf482009-09-11 16:11:19 -04007898 if (!ret)
liubocdcb7252011-08-03 10:15:25 +00007899 dev_nr++;
7900
7901 if (dev_nr >= dev_min)
Yan73e48b22008-01-03 14:14:39 -05007902 break;
liubocdcb7252011-08-03 10:15:25 +00007903
Josef Bacikba1bf482009-09-11 16:11:19 -04007904 ret = -1;
Yan73e48b22008-01-03 14:14:39 -05007905 }
Chris Masonedbd8d42007-12-21 16:27:24 -05007906 }
Josef Bacikba1bf482009-09-11 16:11:19 -04007907 mutex_unlock(&root->fs_info->chunk_mutex);
Chris Masonedbd8d42007-12-21 16:27:24 -05007908out:
Josef Bacikba1bf482009-09-11 16:11:19 -04007909 btrfs_put_block_group(block_group);
Chris Masonedbd8d42007-12-21 16:27:24 -05007910 return ret;
7911}
7912
Christoph Hellwigb2950862008-12-02 09:54:17 -05007913static int find_first_block_group(struct btrfs_root *root,
7914 struct btrfs_path *path, struct btrfs_key *key)
Chris Mason0b86a832008-03-24 15:01:56 -04007915{
Chris Mason925baed2008-06-25 16:01:30 -04007916 int ret = 0;
Chris Mason0b86a832008-03-24 15:01:56 -04007917 struct btrfs_key found_key;
7918 struct extent_buffer *leaf;
7919 int slot;
7920
7921 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
7922 if (ret < 0)
Chris Mason925baed2008-06-25 16:01:30 -04007923 goto out;
7924
Chris Masond3977122009-01-05 21:25:51 -05007925 while (1) {
Chris Mason0b86a832008-03-24 15:01:56 -04007926 slot = path->slots[0];
7927 leaf = path->nodes[0];
7928 if (slot >= btrfs_header_nritems(leaf)) {
7929 ret = btrfs_next_leaf(root, path);
7930 if (ret == 0)
7931 continue;
7932 if (ret < 0)
Chris Mason925baed2008-06-25 16:01:30 -04007933 goto out;
Chris Mason0b86a832008-03-24 15:01:56 -04007934 break;
7935 }
7936 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7937
7938 if (found_key.objectid >= key->objectid &&
Chris Mason925baed2008-06-25 16:01:30 -04007939 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
7940 ret = 0;
7941 goto out;
7942 }
Chris Mason0b86a832008-03-24 15:01:56 -04007943 path->slots[0]++;
7944 }
Chris Mason925baed2008-06-25 16:01:30 -04007945out:
Chris Mason0b86a832008-03-24 15:01:56 -04007946 return ret;
7947}
7948
Josef Bacik0af3d002010-06-21 14:48:16 -04007949void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
7950{
7951 struct btrfs_block_group_cache *block_group;
7952 u64 last = 0;
7953
7954 while (1) {
7955 struct inode *inode;
7956
7957 block_group = btrfs_lookup_first_block_group(info, last);
7958 while (block_group) {
7959 spin_lock(&block_group->lock);
7960 if (block_group->iref)
7961 break;
7962 spin_unlock(&block_group->lock);
7963 block_group = next_block_group(info->tree_root,
7964 block_group);
7965 }
7966 if (!block_group) {
7967 if (last == 0)
7968 break;
7969 last = 0;
7970 continue;
7971 }
7972
7973 inode = block_group->inode;
7974 block_group->iref = 0;
7975 block_group->inode = NULL;
7976 spin_unlock(&block_group->lock);
7977 iput(inode);
7978 last = block_group->key.objectid + block_group->key.offset;
7979 btrfs_put_block_group(block_group);
7980 }
7981}
7982
Zheng Yan1a40e232008-09-26 10:09:34 -04007983int btrfs_free_block_groups(struct btrfs_fs_info *info)
7984{
7985 struct btrfs_block_group_cache *block_group;
Chris Mason4184ea72009-03-10 12:39:20 -04007986 struct btrfs_space_info *space_info;
Yan Zheng11833d62009-09-11 16:11:19 -04007987 struct btrfs_caching_control *caching_ctl;
Zheng Yan1a40e232008-09-26 10:09:34 -04007988 struct rb_node *n;
7989
Yan Zheng11833d62009-09-11 16:11:19 -04007990 down_write(&info->extent_commit_sem);
7991 while (!list_empty(&info->caching_block_groups)) {
7992 caching_ctl = list_entry(info->caching_block_groups.next,
7993 struct btrfs_caching_control, list);
7994 list_del(&caching_ctl->list);
7995 put_caching_control(caching_ctl);
7996 }
7997 up_write(&info->extent_commit_sem);
7998
Zheng Yan1a40e232008-09-26 10:09:34 -04007999 spin_lock(&info->block_group_cache_lock);
8000 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
8001 block_group = rb_entry(n, struct btrfs_block_group_cache,
8002 cache_node);
Zheng Yan1a40e232008-09-26 10:09:34 -04008003 rb_erase(&block_group->cache_node,
8004 &info->block_group_cache_tree);
Yan Zhengd899e052008-10-30 14:25:28 -04008005 spin_unlock(&info->block_group_cache_lock);
8006
Josef Bacik80eb2342008-10-29 14:49:05 -04008007 down_write(&block_group->space_info->groups_sem);
Zheng Yan1a40e232008-09-26 10:09:34 -04008008 list_del(&block_group->list);
Josef Bacik80eb2342008-10-29 14:49:05 -04008009 up_write(&block_group->space_info->groups_sem);
Yan Zhengd2fb3432008-12-11 16:30:39 -05008010
Josef Bacik817d52f2009-07-13 21:29:25 -04008011 if (block_group->cached == BTRFS_CACHE_STARTED)
Yan Zheng11833d62009-09-11 16:11:19 -04008012 wait_block_group_cache_done(block_group);
Josef Bacik817d52f2009-07-13 21:29:25 -04008013
Josef Bacik3c148742011-02-02 15:53:47 +00008014 /*
8015 * We haven't cached this block group, which means we could
8016 * possibly have excluded extents on this block group.
8017 */
8018 if (block_group->cached == BTRFS_CACHE_NO)
8019 free_excluded_extents(info->extent_root, block_group);
8020
Josef Bacik817d52f2009-07-13 21:29:25 -04008021 btrfs_remove_free_space_cache(block_group);
Josef Bacik11dfe352009-11-13 20:12:59 +00008022 btrfs_put_block_group(block_group);
Yan Zhengd899e052008-10-30 14:25:28 -04008023
8024 spin_lock(&info->block_group_cache_lock);
Zheng Yan1a40e232008-09-26 10:09:34 -04008025 }
8026 spin_unlock(&info->block_group_cache_lock);
Chris Mason4184ea72009-03-10 12:39:20 -04008027
8028 /* now that all the block groups are freed, go through and
8029 * free all the space_info structs. This is only called during
8030 * the final stages of unmount, and so we know nobody is
8031 * using them. We call synchronize_rcu() once before we start,
8032 * just to be on the safe side.
8033 */
8034 synchronize_rcu();
8035
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04008036 release_global_block_rsv(info);
8037
Chris Mason4184ea72009-03-10 12:39:20 -04008038 while(!list_empty(&info->space_info)) {
8039 space_info = list_entry(info->space_info.next,
8040 struct btrfs_space_info,
8041 list);
David Sterbab069e0c2013-02-08 21:28:17 +00008042 if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) {
8043 if (space_info->bytes_pinned > 0 ||
8044 space_info->bytes_reserved > 0 ||
8045 space_info->bytes_may_use > 0) {
8046 WARN_ON(1);
8047 dump_space_info(space_info, 0, 0);
8048 }
Yan, Zhengf0486c62010-05-16 10:46:25 -04008049 }
Chris Mason4184ea72009-03-10 12:39:20 -04008050 list_del(&space_info->list);
8051 kfree(space_info);
8052 }
Zheng Yan1a40e232008-09-26 10:09:34 -04008053 return 0;
8054}
8055
Yan, Zhengb742bb82010-05-16 10:46:24 -04008056static void __link_block_group(struct btrfs_space_info *space_info,
8057 struct btrfs_block_group_cache *cache)
8058{
8059 int index = get_block_group_index(cache);
8060
8061 down_write(&space_info->groups_sem);
8062 list_add_tail(&cache->list, &space_info->block_groups[index]);
8063 up_write(&space_info->groups_sem);
8064}
8065
Chris Mason9078a3e2007-04-26 16:46:15 -04008066int btrfs_read_block_groups(struct btrfs_root *root)
8067{
8068 struct btrfs_path *path;
8069 int ret;
Chris Mason9078a3e2007-04-26 16:46:15 -04008070 struct btrfs_block_group_cache *cache;
Chris Masonbe744172007-05-06 10:15:01 -04008071 struct btrfs_fs_info *info = root->fs_info;
Chris Mason6324fbf2008-03-24 15:01:59 -04008072 struct btrfs_space_info *space_info;
Chris Mason9078a3e2007-04-26 16:46:15 -04008073 struct btrfs_key key;
8074 struct btrfs_key found_key;
Chris Mason5f39d392007-10-15 16:14:19 -04008075 struct extent_buffer *leaf;
Josef Bacik0af3d002010-06-21 14:48:16 -04008076 int need_clear = 0;
8077 u64 cache_gen;
Chris Mason96b51792007-10-15 16:15:19 -04008078
Chris Masonbe744172007-05-06 10:15:01 -04008079 root = info->extent_root;
Chris Mason9078a3e2007-04-26 16:46:15 -04008080 key.objectid = 0;
Chris Mason0b86a832008-03-24 15:01:56 -04008081 key.offset = 0;
Chris Mason9078a3e2007-04-26 16:46:15 -04008082 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
Chris Mason9078a3e2007-04-26 16:46:15 -04008083 path = btrfs_alloc_path();
8084 if (!path)
8085 return -ENOMEM;
Josef Bacik026fd312011-05-13 10:32:11 -04008086 path->reada = 1;
Chris Mason9078a3e2007-04-26 16:46:15 -04008087
David Sterba6c417612011-04-13 15:41:04 +02008088 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
Josef Bacik73bc1872011-10-03 14:07:49 -04008089 if (btrfs_test_opt(root, SPACE_CACHE) &&
David Sterba6c417612011-04-13 15:41:04 +02008090 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
Josef Bacik0af3d002010-06-21 14:48:16 -04008091 need_clear = 1;
Josef Bacik88c2ba32010-09-21 14:21:34 -04008092 if (btrfs_test_opt(root, CLEAR_CACHE))
8093 need_clear = 1;
Josef Bacik0af3d002010-06-21 14:48:16 -04008094
Chris Masond3977122009-01-05 21:25:51 -05008095 while (1) {
Chris Mason0b86a832008-03-24 15:01:56 -04008096 ret = find_first_block_group(root, path, &key);
Yan, Zhengb742bb82010-05-16 10:46:24 -04008097 if (ret > 0)
8098 break;
Chris Mason0b86a832008-03-24 15:01:56 -04008099 if (ret != 0)
8100 goto error;
Chris Mason5f39d392007-10-15 16:14:19 -04008101 leaf = path->nodes[0];
8102 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
Chris Mason8f18cf12008-04-25 16:53:30 -04008103 cache = kzalloc(sizeof(*cache), GFP_NOFS);
Chris Mason9078a3e2007-04-26 16:46:15 -04008104 if (!cache) {
Chris Mason0b86a832008-03-24 15:01:56 -04008105 ret = -ENOMEM;
Yan, Zhengf0486c62010-05-16 10:46:25 -04008106 goto error;
Chris Mason9078a3e2007-04-26 16:46:15 -04008107 }
Li Zefan34d52cb2011-03-29 13:46:06 +08008108 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
8109 GFP_NOFS);
8110 if (!cache->free_space_ctl) {
8111 kfree(cache);
8112 ret = -ENOMEM;
8113 goto error;
8114 }
Chris Mason3e1ad542007-05-07 20:03:49 -04008115
Yan Zhengd2fb3432008-12-11 16:30:39 -05008116 atomic_set(&cache->count, 1);
Chris Masonc286ac42008-07-22 23:06:41 -04008117 spin_lock_init(&cache->lock);
Josef Bacik817d52f2009-07-13 21:29:25 -04008118 cache->fs_info = info;
Josef Bacik0f9dd462008-09-23 13:14:11 -04008119 INIT_LIST_HEAD(&cache->list);
Chris Masonfa9c0d792009-04-03 09:47:43 -04008120 INIT_LIST_HEAD(&cache->cluster_list);
Josef Bacik96303082009-07-13 21:29:25 -04008121
Liu Bocf7c1ef2012-07-06 03:31:34 -06008122 if (need_clear) {
8123 /*
8124 * When we mount with old space cache, we need to
8125 * set BTRFS_DC_CLEAR and set dirty flag.
8126 *
8127 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
8128 * truncate the old free space cache inode and
8129 * setup a new one.
8130 * b) Setting 'dirty flag' makes sure that we flush
8131 * the new space cache info onto disk.
8132 */
Josef Bacik0af3d002010-06-21 14:48:16 -04008133 cache->disk_cache_state = BTRFS_DC_CLEAR;
Liu Bocf7c1ef2012-07-06 03:31:34 -06008134 if (btrfs_test_opt(root, SPACE_CACHE))
8135 cache->dirty = 1;
8136 }
Josef Bacik0af3d002010-06-21 14:48:16 -04008137
Chris Mason5f39d392007-10-15 16:14:19 -04008138 read_extent_buffer(leaf, &cache->item,
8139 btrfs_item_ptr_offset(leaf, path->slots[0]),
8140 sizeof(cache->item));
Chris Mason9078a3e2007-04-26 16:46:15 -04008141 memcpy(&cache->key, &found_key, sizeof(found_key));
Chris Mason0b86a832008-03-24 15:01:56 -04008142
Chris Mason9078a3e2007-04-26 16:46:15 -04008143 key.objectid = found_key.objectid + found_key.offset;
David Sterbab3b4aa72011-04-21 01:20:15 +02008144 btrfs_release_path(path);
Chris Mason0b86a832008-03-24 15:01:56 -04008145 cache->flags = btrfs_block_group_flags(&cache->item);
Josef Bacik817d52f2009-07-13 21:29:25 -04008146 cache->sectorsize = root->sectorsize;
David Woodhouse53b381b2013-01-29 18:40:14 -05008147 cache->full_stripe_len = btrfs_full_stripe_len(root,
8148 &root->fs_info->mapping_tree,
8149 found_key.objectid);
Li Zefan34d52cb2011-03-29 13:46:06 +08008150 btrfs_init_free_space_ctl(cache);
8151
Josef Bacik817d52f2009-07-13 21:29:25 -04008152 /*
Josef Bacik3c148742011-02-02 15:53:47 +00008153 * We need to exclude the super stripes now so that the space
8154 * info has super bytes accounted for, otherwise we'll think
8155 * we have more space than we actually do.
8156 */
Josef Bacik835d9742013-03-19 12:13:25 -04008157 ret = exclude_super_stripes(root, cache);
8158 if (ret) {
8159 /*
8160 * We may have excluded something, so call this just in
8161 * case.
8162 */
8163 free_excluded_extents(root, cache);
8164 kfree(cache->free_space_ctl);
8165 kfree(cache);
8166 goto error;
8167 }
Josef Bacik3c148742011-02-02 15:53:47 +00008168
8169 /*
Josef Bacik817d52f2009-07-13 21:29:25 -04008170 * check for two cases, either we are full, and therefore
8171 * don't need to bother with the caching work since we won't
8172 * find any space, or we are empty, and we can just add all
8173 * the space in and be done with it. This saves us _alot_ of
8174 * time, particularly in the full case.
8175 */
8176 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
Yan Zheng11833d62009-09-11 16:11:19 -04008177 cache->last_byte_to_unpin = (u64)-1;
Josef Bacik817d52f2009-07-13 21:29:25 -04008178 cache->cached = BTRFS_CACHE_FINISHED;
Josef Bacik1b2da372009-09-11 16:11:20 -04008179 free_excluded_extents(root, cache);
Josef Bacik817d52f2009-07-13 21:29:25 -04008180 } else if (btrfs_block_group_used(&cache->item) == 0) {
Yan Zheng11833d62009-09-11 16:11:19 -04008181 cache->last_byte_to_unpin = (u64)-1;
Josef Bacik817d52f2009-07-13 21:29:25 -04008182 cache->cached = BTRFS_CACHE_FINISHED;
8183 add_new_free_space(cache, root->fs_info,
8184 found_key.objectid,
8185 found_key.objectid +
8186 found_key.offset);
Yan Zheng11833d62009-09-11 16:11:19 -04008187 free_excluded_extents(root, cache);
Josef Bacik817d52f2009-07-13 21:29:25 -04008188 }
Chris Mason96b51792007-10-15 16:15:19 -04008189
Josef Bacik8c579fe2013-04-02 12:40:42 -04008190 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8191 if (ret) {
8192 btrfs_remove_free_space_cache(cache);
8193 btrfs_put_block_group(cache);
8194 goto error;
8195 }
8196
Chris Mason6324fbf2008-03-24 15:01:59 -04008197 ret = update_space_info(info, cache->flags, found_key.offset,
8198 btrfs_block_group_used(&cache->item),
8199 &space_info);
Josef Bacik8c579fe2013-04-02 12:40:42 -04008200 if (ret) {
8201 btrfs_remove_free_space_cache(cache);
8202 spin_lock(&info->block_group_cache_lock);
8203 rb_erase(&cache->cache_node,
8204 &info->block_group_cache_tree);
8205 spin_unlock(&info->block_group_cache_lock);
8206 btrfs_put_block_group(cache);
8207 goto error;
8208 }
8209
Chris Mason6324fbf2008-03-24 15:01:59 -04008210 cache->space_info = space_info;
Josef Bacik1b2da372009-09-11 16:11:20 -04008211 spin_lock(&cache->space_info->lock);
Yan, Zhengf0486c62010-05-16 10:46:25 -04008212 cache->space_info->bytes_readonly += cache->bytes_super;
Josef Bacik1b2da372009-09-11 16:11:20 -04008213 spin_unlock(&cache->space_info->lock);
8214
Yan, Zhengb742bb82010-05-16 10:46:24 -04008215 __link_block_group(space_info, cache);
Chris Mason6324fbf2008-03-24 15:01:59 -04008216
Chris Mason75ccf472008-09-30 19:24:06 -04008217 set_avail_alloc_bits(root->fs_info, cache->flags);
Yan Zheng2b820322008-11-17 21:11:30 -05008218 if (btrfs_chunk_readonly(root, cache->key.objectid))
Miao Xie199c36e2011-07-15 10:34:36 +00008219 set_block_group_ro(cache, 1);
Chris Mason9078a3e2007-04-26 16:46:15 -04008220 }
Yan, Zhengb742bb82010-05-16 10:46:24 -04008221
8222 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
8223 if (!(get_alloc_profile(root, space_info->flags) &
8224 (BTRFS_BLOCK_GROUP_RAID10 |
8225 BTRFS_BLOCK_GROUP_RAID1 |
David Woodhouse53b381b2013-01-29 18:40:14 -05008226 BTRFS_BLOCK_GROUP_RAID5 |
8227 BTRFS_BLOCK_GROUP_RAID6 |
Yan, Zhengb742bb82010-05-16 10:46:24 -04008228 BTRFS_BLOCK_GROUP_DUP)))
8229 continue;
8230 /*
8231 * avoid allocating from un-mirrored block group if there are
8232 * mirrored block groups.
8233 */
8234 list_for_each_entry(cache, &space_info->block_groups[3], list)
Miao Xie199c36e2011-07-15 10:34:36 +00008235 set_block_group_ro(cache, 1);
Yan, Zhengb742bb82010-05-16 10:46:24 -04008236 list_for_each_entry(cache, &space_info->block_groups[4], list)
Miao Xie199c36e2011-07-15 10:34:36 +00008237 set_block_group_ro(cache, 1);
Yan, Zhengb742bb82010-05-16 10:46:24 -04008238 }
Yan, Zhengf0486c62010-05-16 10:46:25 -04008239
8240 init_global_block_rsv(info);
Chris Mason0b86a832008-03-24 15:01:56 -04008241 ret = 0;
8242error:
Chris Mason9078a3e2007-04-26 16:46:15 -04008243 btrfs_free_path(path);
Chris Mason0b86a832008-03-24 15:01:56 -04008244 return ret;
Chris Mason9078a3e2007-04-26 16:46:15 -04008245}
Chris Mason6324fbf2008-03-24 15:01:59 -04008246
Josef Bacikea658ba2012-09-11 16:57:25 -04008247void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
8248 struct btrfs_root *root)
8249{
8250 struct btrfs_block_group_cache *block_group, *tmp;
8251 struct btrfs_root *extent_root = root->fs_info->extent_root;
8252 struct btrfs_block_group_item item;
8253 struct btrfs_key key;
8254 int ret = 0;
8255
8256 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs,
8257 new_bg_list) {
8258 list_del_init(&block_group->new_bg_list);
8259
8260 if (ret)
8261 continue;
8262
8263 spin_lock(&block_group->lock);
8264 memcpy(&item, &block_group->item, sizeof(item));
8265 memcpy(&key, &block_group->key, sizeof(key));
8266 spin_unlock(&block_group->lock);
8267
8268 ret = btrfs_insert_item(trans, extent_root, &key, &item,
8269 sizeof(item));
8270 if (ret)
8271 btrfs_abort_transaction(trans, extent_root, ret);
8272 }
8273}
8274
Chris Mason6324fbf2008-03-24 15:01:59 -04008275int btrfs_make_block_group(struct btrfs_trans_handle *trans,
8276 struct btrfs_root *root, u64 bytes_used,
Chris Masone17cade2008-04-15 15:41:47 -04008277 u64 type, u64 chunk_objectid, u64 chunk_offset,
Chris Mason6324fbf2008-03-24 15:01:59 -04008278 u64 size)
8279{
8280 int ret;
Chris Mason6324fbf2008-03-24 15:01:59 -04008281 struct btrfs_root *extent_root;
8282 struct btrfs_block_group_cache *cache;
Chris Mason6324fbf2008-03-24 15:01:59 -04008283
8284 extent_root = root->fs_info->extent_root;
Chris Mason6324fbf2008-03-24 15:01:59 -04008285
Chris Mason12fcfd22009-03-24 10:24:20 -04008286 root->fs_info->last_trans_log_full_commit = trans->transid;
Chris Masone02119d2008-09-05 16:13:11 -04008287
Chris Mason8f18cf12008-04-25 16:53:30 -04008288 cache = kzalloc(sizeof(*cache), GFP_NOFS);
Josef Bacik0f9dd462008-09-23 13:14:11 -04008289 if (!cache)
8290 return -ENOMEM;
Li Zefan34d52cb2011-03-29 13:46:06 +08008291 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
8292 GFP_NOFS);
8293 if (!cache->free_space_ctl) {
8294 kfree(cache);
8295 return -ENOMEM;
8296 }
Josef Bacik0f9dd462008-09-23 13:14:11 -04008297
Chris Masone17cade2008-04-15 15:41:47 -04008298 cache->key.objectid = chunk_offset;
Chris Mason6324fbf2008-03-24 15:01:59 -04008299 cache->key.offset = size;
Yan Zhengd2fb3432008-12-11 16:30:39 -05008300 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
Josef Bacik96303082009-07-13 21:29:25 -04008301 cache->sectorsize = root->sectorsize;
Josef Bacik0af3d002010-06-21 14:48:16 -04008302 cache->fs_info = root->fs_info;
David Woodhouse53b381b2013-01-29 18:40:14 -05008303 cache->full_stripe_len = btrfs_full_stripe_len(root,
8304 &root->fs_info->mapping_tree,
8305 chunk_offset);
Josef Bacik96303082009-07-13 21:29:25 -04008306
Yan Zhengd2fb3432008-12-11 16:30:39 -05008307 atomic_set(&cache->count, 1);
Chris Masonc286ac42008-07-22 23:06:41 -04008308 spin_lock_init(&cache->lock);
Josef Bacik0f9dd462008-09-23 13:14:11 -04008309 INIT_LIST_HEAD(&cache->list);
Chris Masonfa9c0d792009-04-03 09:47:43 -04008310 INIT_LIST_HEAD(&cache->cluster_list);
Josef Bacikea658ba2012-09-11 16:57:25 -04008311 INIT_LIST_HEAD(&cache->new_bg_list);
Chris Mason0ef3e662008-05-24 14:04:53 -04008312
Li Zefan34d52cb2011-03-29 13:46:06 +08008313 btrfs_init_free_space_ctl(cache);
8314
Chris Mason6324fbf2008-03-24 15:01:59 -04008315 btrfs_set_block_group_used(&cache->item, bytes_used);
Chris Mason6324fbf2008-03-24 15:01:59 -04008316 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
8317 cache->flags = type;
8318 btrfs_set_block_group_flags(&cache->item, type);
8319
Yan Zheng11833d62009-09-11 16:11:19 -04008320 cache->last_byte_to_unpin = (u64)-1;
Josef Bacik817d52f2009-07-13 21:29:25 -04008321 cache->cached = BTRFS_CACHE_FINISHED;
Josef Bacik835d9742013-03-19 12:13:25 -04008322 ret = exclude_super_stripes(root, cache);
8323 if (ret) {
8324 /*
8325 * We may have excluded something, so call this just in
8326 * case.
8327 */
8328 free_excluded_extents(root, cache);
8329 kfree(cache->free_space_ctl);
8330 kfree(cache);
8331 return ret;
8332 }
Josef Bacik96303082009-07-13 21:29:25 -04008333
Josef Bacik817d52f2009-07-13 21:29:25 -04008334 add_new_free_space(cache, root->fs_info, chunk_offset,
8335 chunk_offset + size);
8336
Yan Zheng11833d62009-09-11 16:11:19 -04008337 free_excluded_extents(root, cache);
8338
Josef Bacik8c579fe2013-04-02 12:40:42 -04008339 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8340 if (ret) {
8341 btrfs_remove_free_space_cache(cache);
8342 btrfs_put_block_group(cache);
8343 return ret;
8344 }
8345
Chris Mason6324fbf2008-03-24 15:01:59 -04008346 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
8347 &cache->space_info);
Josef Bacik8c579fe2013-04-02 12:40:42 -04008348 if (ret) {
8349 btrfs_remove_free_space_cache(cache);
8350 spin_lock(&root->fs_info->block_group_cache_lock);
8351 rb_erase(&cache->cache_node,
8352 &root->fs_info->block_group_cache_tree);
8353 spin_unlock(&root->fs_info->block_group_cache_lock);
8354 btrfs_put_block_group(cache);
8355 return ret;
8356 }
Li Zefanc7c144d2011-12-07 10:39:22 +08008357 update_global_block_rsv(root->fs_info);
Josef Bacik1b2da372009-09-11 16:11:20 -04008358
8359 spin_lock(&cache->space_info->lock);
Yan, Zhengf0486c62010-05-16 10:46:25 -04008360 cache->space_info->bytes_readonly += cache->bytes_super;
Josef Bacik1b2da372009-09-11 16:11:20 -04008361 spin_unlock(&cache->space_info->lock);
8362
Yan, Zhengb742bb82010-05-16 10:46:24 -04008363 __link_block_group(cache->space_info, cache);
Chris Mason6324fbf2008-03-24 15:01:59 -04008364
Josef Bacikea658ba2012-09-11 16:57:25 -04008365 list_add_tail(&cache->new_bg_list, &trans->new_bgs);
Chris Mason6324fbf2008-03-24 15:01:59 -04008366
Chris Masond18a2c42008-04-04 15:40:00 -04008367 set_avail_alloc_bits(extent_root->fs_info, type);
Chris Mason925baed2008-06-25 16:01:30 -04008368
Chris Mason6324fbf2008-03-24 15:01:59 -04008369 return 0;
8370}
Zheng Yan1a40e232008-09-26 10:09:34 -04008371
Ilya Dryomov10ea00f2012-01-16 22:04:47 +02008372static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
8373{
Ilya Dryomov899c81e2012-03-27 17:09:16 +03008374 u64 extra_flags = chunk_to_extended(flags) &
8375 BTRFS_EXTENDED_PROFILE_MASK;
Ilya Dryomov10ea00f2012-01-16 22:04:47 +02008376
Miao Xiede98ced2013-01-29 10:13:12 +00008377 write_seqlock(&fs_info->profiles_lock);
Ilya Dryomov10ea00f2012-01-16 22:04:47 +02008378 if (flags & BTRFS_BLOCK_GROUP_DATA)
8379 fs_info->avail_data_alloc_bits &= ~extra_flags;
8380 if (flags & BTRFS_BLOCK_GROUP_METADATA)
8381 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
8382 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
8383 fs_info->avail_system_alloc_bits &= ~extra_flags;
Miao Xiede98ced2013-01-29 10:13:12 +00008384 write_sequnlock(&fs_info->profiles_lock);
Ilya Dryomov10ea00f2012-01-16 22:04:47 +02008385}
8386
Zheng Yan1a40e232008-09-26 10:09:34 -04008387int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
8388 struct btrfs_root *root, u64 group_start)
8389{
8390 struct btrfs_path *path;
8391 struct btrfs_block_group_cache *block_group;
Chris Mason44fb5512009-06-04 15:34:51 -04008392 struct btrfs_free_cluster *cluster;
Josef Bacik0af3d002010-06-21 14:48:16 -04008393 struct btrfs_root *tree_root = root->fs_info->tree_root;
Zheng Yan1a40e232008-09-26 10:09:34 -04008394 struct btrfs_key key;
Josef Bacik0af3d002010-06-21 14:48:16 -04008395 struct inode *inode;
Zheng Yan1a40e232008-09-26 10:09:34 -04008396 int ret;
Ilya Dryomov10ea00f2012-01-16 22:04:47 +02008397 int index;
Josef Bacik89a55892010-10-14 14:52:27 -04008398 int factor;
Zheng Yan1a40e232008-09-26 10:09:34 -04008399
Zheng Yan1a40e232008-09-26 10:09:34 -04008400 root = root->fs_info->extent_root;
8401
8402 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
8403 BUG_ON(!block_group);
Yan Zhengc146afa2008-11-12 14:34:12 -05008404 BUG_ON(!block_group->ro);
Zheng Yan1a40e232008-09-26 10:09:34 -04008405
liubo9f7c43c2011-03-07 02:13:33 +00008406 /*
8407 * Free the reserved super bytes from this block group before
8408 * remove it.
8409 */
8410 free_excluded_extents(root, block_group);
8411
Zheng Yan1a40e232008-09-26 10:09:34 -04008412 memcpy(&key, &block_group->key, sizeof(key));
Ilya Dryomov10ea00f2012-01-16 22:04:47 +02008413 index = get_block_group_index(block_group);
Josef Bacik89a55892010-10-14 14:52:27 -04008414 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
8415 BTRFS_BLOCK_GROUP_RAID1 |
8416 BTRFS_BLOCK_GROUP_RAID10))
8417 factor = 2;
8418 else
8419 factor = 1;
Zheng Yan1a40e232008-09-26 10:09:34 -04008420
Chris Mason44fb5512009-06-04 15:34:51 -04008421 /* make sure this block group isn't part of an allocation cluster */
8422 cluster = &root->fs_info->data_alloc_cluster;
8423 spin_lock(&cluster->refill_lock);
8424 btrfs_return_cluster_to_free_space(block_group, cluster);
8425 spin_unlock(&cluster->refill_lock);
8426
8427 /*
8428 * make sure this block group isn't part of a metadata
8429 * allocation cluster
8430 */
8431 cluster = &root->fs_info->meta_alloc_cluster;
8432 spin_lock(&cluster->refill_lock);
8433 btrfs_return_cluster_to_free_space(block_group, cluster);
8434 spin_unlock(&cluster->refill_lock);
8435
Zheng Yan1a40e232008-09-26 10:09:34 -04008436 path = btrfs_alloc_path();
Mark Fashehd8926bb2011-07-13 10:38:47 -07008437 if (!path) {
8438 ret = -ENOMEM;
8439 goto out;
8440 }
Zheng Yan1a40e232008-09-26 10:09:34 -04008441
Ilya Dryomov10b2f342011-10-02 13:56:53 +03008442 inode = lookup_free_space_inode(tree_root, block_group, path);
Josef Bacik0af3d002010-06-21 14:48:16 -04008443 if (!IS_ERR(inode)) {
Tsutomu Itohb5324022011-07-19 07:27:20 +00008444 ret = btrfs_orphan_add(trans, inode);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01008445 if (ret) {
8446 btrfs_add_delayed_iput(inode);
8447 goto out;
8448 }
Josef Bacik0af3d002010-06-21 14:48:16 -04008449 clear_nlink(inode);
8450 /* One for the block groups ref */
8451 spin_lock(&block_group->lock);
8452 if (block_group->iref) {
8453 block_group->iref = 0;
8454 block_group->inode = NULL;
8455 spin_unlock(&block_group->lock);
8456 iput(inode);
8457 } else {
8458 spin_unlock(&block_group->lock);
8459 }
8460 /* One for our lookup ref */
Josef Bacik455757c2011-09-19 12:26:24 -04008461 btrfs_add_delayed_iput(inode);
Josef Bacik0af3d002010-06-21 14:48:16 -04008462 }
8463
8464 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
8465 key.offset = block_group->key.objectid;
8466 key.type = 0;
8467
8468 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
8469 if (ret < 0)
8470 goto out;
8471 if (ret > 0)
David Sterbab3b4aa72011-04-21 01:20:15 +02008472 btrfs_release_path(path);
Josef Bacik0af3d002010-06-21 14:48:16 -04008473 if (ret == 0) {
8474 ret = btrfs_del_item(trans, tree_root, path);
8475 if (ret)
8476 goto out;
David Sterbab3b4aa72011-04-21 01:20:15 +02008477 btrfs_release_path(path);
Josef Bacik0af3d002010-06-21 14:48:16 -04008478 }
8479
Yan Zheng3dfdb932009-01-21 10:49:16 -05008480 spin_lock(&root->fs_info->block_group_cache_lock);
Zheng Yan1a40e232008-09-26 10:09:34 -04008481 rb_erase(&block_group->cache_node,
8482 &root->fs_info->block_group_cache_tree);
Liu Boa1897fd2012-12-27 09:01:23 +00008483
8484 if (root->fs_info->first_logical_byte == block_group->key.objectid)
8485 root->fs_info->first_logical_byte = (u64)-1;
Yan Zheng3dfdb932009-01-21 10:49:16 -05008486 spin_unlock(&root->fs_info->block_group_cache_lock);
Josef Bacik817d52f2009-07-13 21:29:25 -04008487
Josef Bacik80eb2342008-10-29 14:49:05 -04008488 down_write(&block_group->space_info->groups_sem);
Chris Mason44fb5512009-06-04 15:34:51 -04008489 /*
8490 * we must use list_del_init so people can check to see if they
8491 * are still on the list after taking the semaphore
8492 */
8493 list_del_init(&block_group->list);
Ilya Dryomov10ea00f2012-01-16 22:04:47 +02008494 if (list_empty(&block_group->space_info->block_groups[index]))
8495 clear_avail_alloc_bits(root->fs_info, block_group->flags);
Josef Bacik80eb2342008-10-29 14:49:05 -04008496 up_write(&block_group->space_info->groups_sem);
Zheng Yan1a40e232008-09-26 10:09:34 -04008497
Josef Bacik817d52f2009-07-13 21:29:25 -04008498 if (block_group->cached == BTRFS_CACHE_STARTED)
Yan Zheng11833d62009-09-11 16:11:19 -04008499 wait_block_group_cache_done(block_group);
Josef Bacik817d52f2009-07-13 21:29:25 -04008500
8501 btrfs_remove_free_space_cache(block_group);
8502
Yan Zhengc146afa2008-11-12 14:34:12 -05008503 spin_lock(&block_group->space_info->lock);
8504 block_group->space_info->total_bytes -= block_group->key.offset;
8505 block_group->space_info->bytes_readonly -= block_group->key.offset;
Josef Bacik89a55892010-10-14 14:52:27 -04008506 block_group->space_info->disk_total -= block_group->key.offset * factor;
Yan Zhengc146afa2008-11-12 14:34:12 -05008507 spin_unlock(&block_group->space_info->lock);
Chris Mason283bb192009-07-24 16:30:55 -04008508
Josef Bacik0af3d002010-06-21 14:48:16 -04008509 memcpy(&key, &block_group->key, sizeof(key));
8510
Chris Mason283bb192009-07-24 16:30:55 -04008511 btrfs_clear_space_info_full(root->fs_info);
Yan Zhengc146afa2008-11-12 14:34:12 -05008512
Chris Masonfa9c0d792009-04-03 09:47:43 -04008513 btrfs_put_block_group(block_group);
8514 btrfs_put_block_group(block_group);
Zheng Yan1a40e232008-09-26 10:09:34 -04008515
8516 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8517 if (ret > 0)
8518 ret = -EIO;
8519 if (ret < 0)
8520 goto out;
8521
8522 ret = btrfs_del_item(trans, root, path);
8523out:
8524 btrfs_free_path(path);
8525 return ret;
8526}
liuboacce9522011-01-06 19:30:25 +08008527
liuboc59021f2011-03-07 02:13:14 +00008528int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
8529{
8530 struct btrfs_space_info *space_info;
liubo1aba86d2011-04-08 08:44:37 +00008531 struct btrfs_super_block *disk_super;
8532 u64 features;
8533 u64 flags;
8534 int mixed = 0;
liuboc59021f2011-03-07 02:13:14 +00008535 int ret;
8536
David Sterba6c417612011-04-13 15:41:04 +02008537 disk_super = fs_info->super_copy;
liubo1aba86d2011-04-08 08:44:37 +00008538 if (!btrfs_super_root(disk_super))
8539 return 1;
liuboc59021f2011-03-07 02:13:14 +00008540
liubo1aba86d2011-04-08 08:44:37 +00008541 features = btrfs_super_incompat_flags(disk_super);
8542 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
8543 mixed = 1;
liuboc59021f2011-03-07 02:13:14 +00008544
liubo1aba86d2011-04-08 08:44:37 +00008545 flags = BTRFS_BLOCK_GROUP_SYSTEM;
8546 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
liuboc59021f2011-03-07 02:13:14 +00008547 if (ret)
liubo1aba86d2011-04-08 08:44:37 +00008548 goto out;
liuboc59021f2011-03-07 02:13:14 +00008549
liubo1aba86d2011-04-08 08:44:37 +00008550 if (mixed) {
8551 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
8552 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8553 } else {
8554 flags = BTRFS_BLOCK_GROUP_METADATA;
8555 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8556 if (ret)
8557 goto out;
8558
8559 flags = BTRFS_BLOCK_GROUP_DATA;
8560 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8561 }
8562out:
liuboc59021f2011-03-07 02:13:14 +00008563 return ret;
8564}
8565
liuboacce9522011-01-06 19:30:25 +08008566int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
8567{
Filipe Manana5f55aee2014-12-07 21:31:47 +00008568 return unpin_extent_range(root, start, end, false);
liuboacce9522011-01-06 19:30:25 +08008569}
8570
8571int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
Li Dongyang5378e602011-03-24 10:24:27 +00008572 u64 num_bytes, u64 *actual_bytes)
liuboacce9522011-01-06 19:30:25 +08008573{
Li Dongyang5378e602011-03-24 10:24:27 +00008574 return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
liuboacce9522011-01-06 19:30:25 +08008575}
Li Dongyangf7039b12011-03-24 10:24:28 +00008576
8577int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
8578{
8579 struct btrfs_fs_info *fs_info = root->fs_info;
8580 struct btrfs_block_group_cache *cache = NULL;
8581 u64 group_trimmed;
8582 u64 start;
8583 u64 end;
8584 u64 trimmed = 0;
Liu Bo2cac13e2012-02-09 18:17:41 +08008585 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
Li Dongyangf7039b12011-03-24 10:24:28 +00008586 int ret = 0;
8587
Liu Bo2cac13e2012-02-09 18:17:41 +08008588 /*
8589 * try to trim all FS space, our block group may start from non-zero.
8590 */
8591 if (range->len == total_bytes)
8592 cache = btrfs_lookup_first_block_group(fs_info, range->start);
8593 else
8594 cache = btrfs_lookup_block_group(fs_info, range->start);
Li Dongyangf7039b12011-03-24 10:24:28 +00008595
8596 while (cache) {
8597 if (cache->key.objectid >= (range->start + range->len)) {
8598 btrfs_put_block_group(cache);
8599 break;
8600 }
8601
8602 start = max(range->start, cache->key.objectid);
8603 end = min(range->start + range->len,
8604 cache->key.objectid + cache->key.offset);
8605
8606 if (end - start >= range->minlen) {
8607 if (!block_group_cache_done(cache)) {
Liu Bof6373bf2012-12-27 09:01:18 +00008608 ret = cache_block_group(cache, 0);
Li Dongyangf7039b12011-03-24 10:24:28 +00008609 if (!ret)
8610 wait_block_group_cache_done(cache);
8611 }
8612 ret = btrfs_trim_block_group(cache,
8613 &group_trimmed,
8614 start,
8615 end,
8616 range->minlen);
8617
8618 trimmed += group_trimmed;
8619 if (ret) {
8620 btrfs_put_block_group(cache);
8621 break;
8622 }
8623 }
8624
8625 cache = next_block_group(fs_info->tree_root, cache);
8626 }
8627
8628 range->len = trimmed;
8629 return ret;
8630}