blob: abe3a66bd3ba6d31f9fb072c6d4d6c6be42c174e [file] [log] [blame]
Josef Bacik0f9dd462008-09-23 13:14:11 -04001/*
2 * Copyright (C) 2008 Red Hat. 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 */
18
Josef Bacik96303082009-07-13 21:29:25 -040019#include <linux/pagemap.h>
Josef Bacik0f9dd462008-09-23 13:14:11 -040020#include <linux/sched.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090021#include <linux/slab.h>
Josef Bacik96303082009-07-13 21:29:25 -040022#include <linux/math64.h>
Josef Bacik6ab60602011-08-08 08:24:46 -040023#include <linux/ratelimit.h>
Josef Bacik0f9dd462008-09-23 13:14:11 -040024#include "ctree.h"
Chris Masonfa9c0d792009-04-03 09:47:43 -040025#include "free-space-cache.h"
26#include "transaction.h"
Josef Bacik0af3d002010-06-21 14:48:16 -040027#include "disk-io.h"
Josef Bacik43be2142011-04-01 14:55:00 +000028#include "extent_io.h"
Li Zefan581bb052011-04-20 10:06:11 +080029#include "inode-map.h"
Filipe Manana04216822014-11-27 21:14:15 +000030#include "volumes.h"
Chris Masonfa9c0d792009-04-03 09:47:43 -040031
Josef Bacik96303082009-07-13 21:29:25 -040032#define BITS_PER_BITMAP (PAGE_CACHE_SIZE * 8)
33#define MAX_CACHE_BYTES_PER_GIG (32 * 1024)
34
Filipe Manana55507ce2014-12-01 17:04:09 +000035struct btrfs_trim_range {
36 u64 start;
37 u64 bytes;
38 struct list_head list;
39};
40
Li Zefan34d52cb2011-03-29 13:46:06 +080041static int link_free_space(struct btrfs_free_space_ctl *ctl,
Josef Bacik0cb59c92010-07-02 12:14:14 -040042 struct btrfs_free_space *info);
Josef Bacikcd023e72012-05-14 10:06:40 -040043static void unlink_free_space(struct btrfs_free_space_ctl *ctl,
44 struct btrfs_free_space *info);
Josef Bacik0cb59c92010-07-02 12:14:14 -040045
Li Zefan0414efa2011-04-20 10:20:14 +080046static struct inode *__lookup_free_space_inode(struct btrfs_root *root,
47 struct btrfs_path *path,
48 u64 offset)
Josef Bacik0af3d002010-06-21 14:48:16 -040049{
50 struct btrfs_key key;
51 struct btrfs_key location;
52 struct btrfs_disk_key disk_key;
53 struct btrfs_free_space_header *header;
54 struct extent_buffer *leaf;
55 struct inode *inode = NULL;
56 int ret;
57
Josef Bacik0af3d002010-06-21 14:48:16 -040058 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
Li Zefan0414efa2011-04-20 10:20:14 +080059 key.offset = offset;
Josef Bacik0af3d002010-06-21 14:48:16 -040060 key.type = 0;
61
62 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
63 if (ret < 0)
64 return ERR_PTR(ret);
65 if (ret > 0) {
David Sterbab3b4aa72011-04-21 01:20:15 +020066 btrfs_release_path(path);
Josef Bacik0af3d002010-06-21 14:48:16 -040067 return ERR_PTR(-ENOENT);
68 }
69
70 leaf = path->nodes[0];
71 header = btrfs_item_ptr(leaf, path->slots[0],
72 struct btrfs_free_space_header);
73 btrfs_free_space_key(leaf, header, &disk_key);
74 btrfs_disk_key_to_cpu(&location, &disk_key);
David Sterbab3b4aa72011-04-21 01:20:15 +020075 btrfs_release_path(path);
Josef Bacik0af3d002010-06-21 14:48:16 -040076
77 inode = btrfs_iget(root->fs_info->sb, &location, root, NULL);
78 if (!inode)
79 return ERR_PTR(-ENOENT);
80 if (IS_ERR(inode))
81 return inode;
82 if (is_bad_inode(inode)) {
83 iput(inode);
84 return ERR_PTR(-ENOENT);
85 }
86
Al Viro528c0322012-04-13 11:03:55 -040087 mapping_set_gfp_mask(inode->i_mapping,
Chris Mason2b108262015-04-06 07:48:20 -070088 mapping_gfp_mask(inode->i_mapping) &
Filipe Manana1d3c61c2015-05-06 11:17:01 +010089 ~(__GFP_FS | __GFP_HIGHMEM));
Miao Xieadae52b2011-03-31 09:43:23 +000090
Li Zefan0414efa2011-04-20 10:20:14 +080091 return inode;
92}
93
94struct inode *lookup_free_space_inode(struct btrfs_root *root,
95 struct btrfs_block_group_cache
96 *block_group, struct btrfs_path *path)
97{
98 struct inode *inode = NULL;
Josef Bacik5b0e95b2011-10-06 08:58:24 -040099 u32 flags = BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW;
Li Zefan0414efa2011-04-20 10:20:14 +0800100
101 spin_lock(&block_group->lock);
102 if (block_group->inode)
103 inode = igrab(block_group->inode);
104 spin_unlock(&block_group->lock);
105 if (inode)
106 return inode;
107
108 inode = __lookup_free_space_inode(root, path,
109 block_group->key.objectid);
110 if (IS_ERR(inode))
111 return inode;
112
Josef Bacik0af3d002010-06-21 14:48:16 -0400113 spin_lock(&block_group->lock);
Josef Bacik5b0e95b2011-10-06 08:58:24 -0400114 if (!((BTRFS_I(inode)->flags & flags) == flags)) {
Simon Kirbyc2cf52e2013-03-19 22:41:23 +0000115 btrfs_info(root->fs_info,
116 "Old style space inode found, converting.");
Josef Bacik5b0e95b2011-10-06 08:58:24 -0400117 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM |
118 BTRFS_INODE_NODATACOW;
Josef Bacik2f356122011-06-10 15:31:13 -0400119 block_group->disk_cache_state = BTRFS_DC_CLEAR;
120 }
121
Josef Bacik300e4f82011-08-29 14:06:00 -0400122 if (!block_group->iref) {
Josef Bacik0af3d002010-06-21 14:48:16 -0400123 block_group->inode = igrab(inode);
124 block_group->iref = 1;
125 }
126 spin_unlock(&block_group->lock);
127
128 return inode;
129}
130
Eric Sandeen48a3b632013-04-25 20:41:01 +0000131static int __create_free_space_inode(struct btrfs_root *root,
132 struct btrfs_trans_handle *trans,
133 struct btrfs_path *path,
134 u64 ino, u64 offset)
Josef Bacik0af3d002010-06-21 14:48:16 -0400135{
136 struct btrfs_key key;
137 struct btrfs_disk_key disk_key;
138 struct btrfs_free_space_header *header;
139 struct btrfs_inode_item *inode_item;
140 struct extent_buffer *leaf;
Josef Bacik5b0e95b2011-10-06 08:58:24 -0400141 u64 flags = BTRFS_INODE_NOCOMPRESS | BTRFS_INODE_PREALLOC;
Josef Bacik0af3d002010-06-21 14:48:16 -0400142 int ret;
143
Li Zefan0414efa2011-04-20 10:20:14 +0800144 ret = btrfs_insert_empty_inode(trans, root, path, ino);
Josef Bacik0af3d002010-06-21 14:48:16 -0400145 if (ret)
146 return ret;
147
Josef Bacik5b0e95b2011-10-06 08:58:24 -0400148 /* We inline crc's for the free disk space cache */
149 if (ino != BTRFS_FREE_INO_OBJECTID)
150 flags |= BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW;
151
Josef Bacik0af3d002010-06-21 14:48:16 -0400152 leaf = path->nodes[0];
153 inode_item = btrfs_item_ptr(leaf, path->slots[0],
154 struct btrfs_inode_item);
155 btrfs_item_key(leaf, &disk_key, path->slots[0]);
156 memset_extent_buffer(leaf, 0, (unsigned long)inode_item,
157 sizeof(*inode_item));
158 btrfs_set_inode_generation(leaf, inode_item, trans->transid);
159 btrfs_set_inode_size(leaf, inode_item, 0);
160 btrfs_set_inode_nbytes(leaf, inode_item, 0);
161 btrfs_set_inode_uid(leaf, inode_item, 0);
162 btrfs_set_inode_gid(leaf, inode_item, 0);
163 btrfs_set_inode_mode(leaf, inode_item, S_IFREG | 0600);
Josef Bacik5b0e95b2011-10-06 08:58:24 -0400164 btrfs_set_inode_flags(leaf, inode_item, flags);
Josef Bacik0af3d002010-06-21 14:48:16 -0400165 btrfs_set_inode_nlink(leaf, inode_item, 1);
166 btrfs_set_inode_transid(leaf, inode_item, trans->transid);
Li Zefan0414efa2011-04-20 10:20:14 +0800167 btrfs_set_inode_block_group(leaf, inode_item, offset);
Josef Bacik0af3d002010-06-21 14:48:16 -0400168 btrfs_mark_buffer_dirty(leaf);
David Sterbab3b4aa72011-04-21 01:20:15 +0200169 btrfs_release_path(path);
Josef Bacik0af3d002010-06-21 14:48:16 -0400170
171 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
Li Zefan0414efa2011-04-20 10:20:14 +0800172 key.offset = offset;
Josef Bacik0af3d002010-06-21 14:48:16 -0400173 key.type = 0;
Josef Bacik0af3d002010-06-21 14:48:16 -0400174 ret = btrfs_insert_empty_item(trans, root, path, &key,
175 sizeof(struct btrfs_free_space_header));
176 if (ret < 0) {
David Sterbab3b4aa72011-04-21 01:20:15 +0200177 btrfs_release_path(path);
Josef Bacik0af3d002010-06-21 14:48:16 -0400178 return ret;
179 }
Chris Masonc9dc4c62015-04-04 17:14:42 -0700180
Josef Bacik0af3d002010-06-21 14:48:16 -0400181 leaf = path->nodes[0];
182 header = btrfs_item_ptr(leaf, path->slots[0],
183 struct btrfs_free_space_header);
184 memset_extent_buffer(leaf, 0, (unsigned long)header, sizeof(*header));
185 btrfs_set_free_space_key(leaf, header, &disk_key);
186 btrfs_mark_buffer_dirty(leaf);
David Sterbab3b4aa72011-04-21 01:20:15 +0200187 btrfs_release_path(path);
Josef Bacik0af3d002010-06-21 14:48:16 -0400188
189 return 0;
190}
191
Li Zefan0414efa2011-04-20 10:20:14 +0800192int create_free_space_inode(struct btrfs_root *root,
193 struct btrfs_trans_handle *trans,
194 struct btrfs_block_group_cache *block_group,
195 struct btrfs_path *path)
196{
197 int ret;
198 u64 ino;
199
200 ret = btrfs_find_free_objectid(root, &ino);
201 if (ret < 0)
202 return ret;
203
204 return __create_free_space_inode(root, trans, path, ino,
205 block_group->key.objectid);
206}
207
Miao Xie7b61cd92013-05-13 13:55:09 +0000208int btrfs_check_trunc_cache_free_space(struct btrfs_root *root,
209 struct btrfs_block_rsv *rsv)
Josef Bacik0af3d002010-06-21 14:48:16 -0400210{
Josef Bacikc8174312011-11-02 09:29:35 -0400211 u64 needed_bytes;
Miao Xie7b61cd92013-05-13 13:55:09 +0000212 int ret;
Josef Bacikc8174312011-11-02 09:29:35 -0400213
214 /* 1 for slack space, 1 for updating the inode */
215 needed_bytes = btrfs_calc_trunc_metadata_size(root, 1) +
216 btrfs_calc_trans_metadata_size(root, 1);
217
Miao Xie7b61cd92013-05-13 13:55:09 +0000218 spin_lock(&rsv->lock);
219 if (rsv->reserved < needed_bytes)
220 ret = -ENOSPC;
221 else
222 ret = 0;
223 spin_unlock(&rsv->lock);
Wei Yongjun4b286cd2013-05-21 02:39:21 +0000224 return ret;
Miao Xie7b61cd92013-05-13 13:55:09 +0000225}
226
227int btrfs_truncate_free_space_cache(struct btrfs_root *root,
228 struct btrfs_trans_handle *trans,
Chris Mason1bbc6212015-04-06 12:46:08 -0700229 struct btrfs_block_group_cache *block_group,
Miao Xie7b61cd92013-05-13 13:55:09 +0000230 struct inode *inode)
231{
Miao Xie7b61cd92013-05-13 13:55:09 +0000232 int ret = 0;
Chris Mason1bbc6212015-04-06 12:46:08 -0700233 struct btrfs_path *path = btrfs_alloc_path();
Filipe Manana35c76642015-04-30 17:47:05 +0100234 bool locked = false;
Chris Mason1bbc6212015-04-06 12:46:08 -0700235
236 if (!path) {
237 ret = -ENOMEM;
238 goto fail;
239 }
240
241 if (block_group) {
Filipe Manana35c76642015-04-30 17:47:05 +0100242 locked = true;
Chris Mason1bbc6212015-04-06 12:46:08 -0700243 mutex_lock(&trans->transaction->cache_write_mutex);
244 if (!list_empty(&block_group->io_list)) {
245 list_del_init(&block_group->io_list);
246
247 btrfs_wait_cache_io(root, trans, block_group,
248 &block_group->io_ctl, path,
249 block_group->key.objectid);
250 btrfs_put_block_group(block_group);
251 }
252
253 /*
254 * now that we've truncated the cache away, its no longer
255 * setup or written
256 */
257 spin_lock(&block_group->lock);
258 block_group->disk_cache_state = BTRFS_DC_CLEAR;
259 spin_unlock(&block_group->lock);
260 }
261 btrfs_free_path(path);
Josef Bacik0af3d002010-06-21 14:48:16 -0400262
Josef Bacik0af3d002010-06-21 14:48:16 -0400263 btrfs_i_size_write(inode, 0);
Kirill A. Shutemov7caef262013-09-12 15:13:56 -0700264 truncate_pagecache(inode, 0);
Josef Bacik0af3d002010-06-21 14:48:16 -0400265
266 /*
267 * We don't need an orphan item because truncating the free space cache
268 * will never be split across transactions.
Chris Mason28ed1342014-12-17 09:41:04 -0800269 * We don't need to check for -EAGAIN because we're a free space
270 * cache inode
Josef Bacik0af3d002010-06-21 14:48:16 -0400271 */
272 ret = btrfs_truncate_inode_items(trans, root, inode,
273 0, BTRFS_EXTENT_DATA_KEY);
Filipe Manana35c76642015-04-30 17:47:05 +0100274 if (ret)
275 goto fail;
Josef Bacik0af3d002010-06-21 14:48:16 -0400276
Li Zefan82d59022011-04-20 10:33:24 +0800277 ret = btrfs_update_inode(trans, root, inode);
Chris Mason1bbc6212015-04-06 12:46:08 -0700278
Chris Mason1bbc6212015-04-06 12:46:08 -0700279fail:
Filipe Manana35c76642015-04-30 17:47:05 +0100280 if (locked)
281 mutex_unlock(&trans->transaction->cache_write_mutex);
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100282 if (ret)
283 btrfs_abort_transaction(trans, root, ret);
Josef Bacikc8174312011-11-02 09:29:35 -0400284
Li Zefan82d59022011-04-20 10:33:24 +0800285 return ret;
Josef Bacik0af3d002010-06-21 14:48:16 -0400286}
287
Josef Bacik9d66e232010-08-25 16:54:15 -0400288static int readahead_cache(struct inode *inode)
289{
290 struct file_ra_state *ra;
291 unsigned long last_index;
292
293 ra = kzalloc(sizeof(*ra), GFP_NOFS);
294 if (!ra)
295 return -ENOMEM;
296
297 file_ra_state_init(ra, inode->i_mapping);
298 last_index = (i_size_read(inode) - 1) >> PAGE_CACHE_SHIFT;
299
300 page_cache_sync_readahead(inode->i_mapping, ra, NULL, 0, last_index);
301
302 kfree(ra);
303
304 return 0;
305}
306
Chris Mason4c6d1d82015-04-06 13:17:20 -0700307static int io_ctl_init(struct btrfs_io_ctl *io_ctl, struct inode *inode,
Miao Xie5349d6c2014-06-19 10:42:49 +0800308 struct btrfs_root *root, int write)
Josef Bacika67509c2011-10-05 15:18:58 -0400309{
Miao Xie5349d6c2014-06-19 10:42:49 +0800310 int num_pages;
311 int check_crcs = 0;
312
David Sterbaed6078f2014-06-05 01:59:57 +0200313 num_pages = DIV_ROUND_UP(i_size_read(inode), PAGE_CACHE_SIZE);
Miao Xie5349d6c2014-06-19 10:42:49 +0800314
315 if (btrfs_ino(inode) != BTRFS_FREE_INO_OBJECTID)
316 check_crcs = 1;
317
318 /* Make sure we can fit our crcs into the first page */
319 if (write && check_crcs &&
320 (num_pages * sizeof(u32)) >= PAGE_CACHE_SIZE)
321 return -ENOSPC;
322
Chris Mason4c6d1d82015-04-06 13:17:20 -0700323 memset(io_ctl, 0, sizeof(struct btrfs_io_ctl));
Miao Xie5349d6c2014-06-19 10:42:49 +0800324
David Sterba31e818f2015-02-20 18:00:26 +0100325 io_ctl->pages = kcalloc(num_pages, sizeof(struct page *), GFP_NOFS);
Josef Bacika67509c2011-10-05 15:18:58 -0400326 if (!io_ctl->pages)
327 return -ENOMEM;
Miao Xie5349d6c2014-06-19 10:42:49 +0800328
329 io_ctl->num_pages = num_pages;
Josef Bacika67509c2011-10-05 15:18:58 -0400330 io_ctl->root = root;
Miao Xie5349d6c2014-06-19 10:42:49 +0800331 io_ctl->check_crcs = check_crcs;
Chris Masonc9dc4c62015-04-04 17:14:42 -0700332 io_ctl->inode = inode;
Miao Xie5349d6c2014-06-19 10:42:49 +0800333
Josef Bacika67509c2011-10-05 15:18:58 -0400334 return 0;
335}
336
Chris Mason4c6d1d82015-04-06 13:17:20 -0700337static void io_ctl_free(struct btrfs_io_ctl *io_ctl)
Josef Bacika67509c2011-10-05 15:18:58 -0400338{
339 kfree(io_ctl->pages);
Chris Masonc9dc4c62015-04-04 17:14:42 -0700340 io_ctl->pages = NULL;
Josef Bacika67509c2011-10-05 15:18:58 -0400341}
342
Chris Mason4c6d1d82015-04-06 13:17:20 -0700343static void io_ctl_unmap_page(struct btrfs_io_ctl *io_ctl)
Josef Bacika67509c2011-10-05 15:18:58 -0400344{
345 if (io_ctl->cur) {
Josef Bacika67509c2011-10-05 15:18:58 -0400346 io_ctl->cur = NULL;
347 io_ctl->orig = NULL;
348 }
349}
350
Chris Mason4c6d1d82015-04-06 13:17:20 -0700351static void io_ctl_map_page(struct btrfs_io_ctl *io_ctl, int clear)
Josef Bacika67509c2011-10-05 15:18:58 -0400352{
Josef Bacikb12d6862013-08-26 17:14:08 -0400353 ASSERT(io_ctl->index < io_ctl->num_pages);
Josef Bacika67509c2011-10-05 15:18:58 -0400354 io_ctl->page = io_ctl->pages[io_ctl->index++];
Chris Mason2b108262015-04-06 07:48:20 -0700355 io_ctl->cur = page_address(io_ctl->page);
Josef Bacika67509c2011-10-05 15:18:58 -0400356 io_ctl->orig = io_ctl->cur;
357 io_ctl->size = PAGE_CACHE_SIZE;
358 if (clear)
359 memset(io_ctl->cur, 0, PAGE_CACHE_SIZE);
360}
361
Chris Mason4c6d1d82015-04-06 13:17:20 -0700362static void io_ctl_drop_pages(struct btrfs_io_ctl *io_ctl)
Josef Bacika67509c2011-10-05 15:18:58 -0400363{
364 int i;
365
366 io_ctl_unmap_page(io_ctl);
367
368 for (i = 0; i < io_ctl->num_pages; i++) {
Li Zefana1ee5a42012-01-09 14:27:42 +0800369 if (io_ctl->pages[i]) {
370 ClearPageChecked(io_ctl->pages[i]);
371 unlock_page(io_ctl->pages[i]);
372 page_cache_release(io_ctl->pages[i]);
373 }
Josef Bacika67509c2011-10-05 15:18:58 -0400374 }
375}
376
Chris Mason4c6d1d82015-04-06 13:17:20 -0700377static int io_ctl_prepare_pages(struct btrfs_io_ctl *io_ctl, struct inode *inode,
Josef Bacika67509c2011-10-05 15:18:58 -0400378 int uptodate)
379{
380 struct page *page;
381 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
382 int i;
383
384 for (i = 0; i < io_ctl->num_pages; i++) {
385 page = find_or_create_page(inode->i_mapping, i, mask);
386 if (!page) {
387 io_ctl_drop_pages(io_ctl);
388 return -ENOMEM;
389 }
390 io_ctl->pages[i] = page;
391 if (uptodate && !PageUptodate(page)) {
392 btrfs_readpage(NULL, page);
393 lock_page(page);
394 if (!PageUptodate(page)) {
Frank Holtonefe120a2013-12-20 11:37:06 -0500395 btrfs_err(BTRFS_I(inode)->root->fs_info,
396 "error reading free space cache");
Josef Bacika67509c2011-10-05 15:18:58 -0400397 io_ctl_drop_pages(io_ctl);
398 return -EIO;
399 }
400 }
401 }
402
Josef Bacikf7d61dc2011-11-15 09:31:24 -0500403 for (i = 0; i < io_ctl->num_pages; i++) {
404 clear_page_dirty_for_io(io_ctl->pages[i]);
405 set_page_extent_mapped(io_ctl->pages[i]);
406 }
407
Josef Bacika67509c2011-10-05 15:18:58 -0400408 return 0;
409}
410
Chris Mason4c6d1d82015-04-06 13:17:20 -0700411static void io_ctl_set_generation(struct btrfs_io_ctl *io_ctl, u64 generation)
Josef Bacika67509c2011-10-05 15:18:58 -0400412{
Al Viro528c0322012-04-13 11:03:55 -0400413 __le64 *val;
Josef Bacika67509c2011-10-05 15:18:58 -0400414
415 io_ctl_map_page(io_ctl, 1);
416
417 /*
Josef Bacik5b0e95b2011-10-06 08:58:24 -0400418 * Skip the csum areas. If we don't check crcs then we just have a
419 * 64bit chunk at the front of the first page.
Josef Bacika67509c2011-10-05 15:18:58 -0400420 */
Josef Bacik5b0e95b2011-10-06 08:58:24 -0400421 if (io_ctl->check_crcs) {
422 io_ctl->cur += (sizeof(u32) * io_ctl->num_pages);
423 io_ctl->size -= sizeof(u64) + (sizeof(u32) * io_ctl->num_pages);
424 } else {
425 io_ctl->cur += sizeof(u64);
426 io_ctl->size -= sizeof(u64) * 2;
427 }
Josef Bacika67509c2011-10-05 15:18:58 -0400428
429 val = io_ctl->cur;
430 *val = cpu_to_le64(generation);
431 io_ctl->cur += sizeof(u64);
Josef Bacika67509c2011-10-05 15:18:58 -0400432}
433
Chris Mason4c6d1d82015-04-06 13:17:20 -0700434static int io_ctl_check_generation(struct btrfs_io_ctl *io_ctl, u64 generation)
Josef Bacika67509c2011-10-05 15:18:58 -0400435{
Al Viro528c0322012-04-13 11:03:55 -0400436 __le64 *gen;
Josef Bacika67509c2011-10-05 15:18:58 -0400437
Josef Bacik5b0e95b2011-10-06 08:58:24 -0400438 /*
439 * Skip the crc area. If we don't check crcs then we just have a 64bit
440 * chunk at the front of the first page.
441 */
442 if (io_ctl->check_crcs) {
443 io_ctl->cur += sizeof(u32) * io_ctl->num_pages;
444 io_ctl->size -= sizeof(u64) +
445 (sizeof(u32) * io_ctl->num_pages);
446 } else {
447 io_ctl->cur += sizeof(u64);
448 io_ctl->size -= sizeof(u64) * 2;
449 }
Josef Bacika67509c2011-10-05 15:18:58 -0400450
Josef Bacika67509c2011-10-05 15:18:58 -0400451 gen = io_ctl->cur;
452 if (le64_to_cpu(*gen) != generation) {
Frank Holtonefe120a2013-12-20 11:37:06 -0500453 printk_ratelimited(KERN_ERR "BTRFS: space cache generation "
Josef Bacika67509c2011-10-05 15:18:58 -0400454 "(%Lu) does not match inode (%Lu)\n", *gen,
455 generation);
456 io_ctl_unmap_page(io_ctl);
457 return -EIO;
458 }
459 io_ctl->cur += sizeof(u64);
Josef Bacik5b0e95b2011-10-06 08:58:24 -0400460 return 0;
461}
462
Chris Mason4c6d1d82015-04-06 13:17:20 -0700463static void io_ctl_set_crc(struct btrfs_io_ctl *io_ctl, int index)
Josef Bacik5b0e95b2011-10-06 08:58:24 -0400464{
465 u32 *tmp;
466 u32 crc = ~(u32)0;
467 unsigned offset = 0;
468
469 if (!io_ctl->check_crcs) {
470 io_ctl_unmap_page(io_ctl);
471 return;
472 }
473
474 if (index == 0)
Justin P. Mattockcb54f252011-11-21 08:43:28 -0800475 offset = sizeof(u32) * io_ctl->num_pages;
Josef Bacik5b0e95b2011-10-06 08:58:24 -0400476
Liu Bob0496682013-03-14 14:57:45 +0000477 crc = btrfs_csum_data(io_ctl->orig + offset, crc,
Josef Bacik5b0e95b2011-10-06 08:58:24 -0400478 PAGE_CACHE_SIZE - offset);
479 btrfs_csum_final(crc, (char *)&crc);
480 io_ctl_unmap_page(io_ctl);
Chris Mason2b108262015-04-06 07:48:20 -0700481 tmp = page_address(io_ctl->pages[0]);
Josef Bacik5b0e95b2011-10-06 08:58:24 -0400482 tmp += index;
483 *tmp = crc;
Josef Bacik5b0e95b2011-10-06 08:58:24 -0400484}
485
Chris Mason4c6d1d82015-04-06 13:17:20 -0700486static int io_ctl_check_crc(struct btrfs_io_ctl *io_ctl, int index)
Josef Bacik5b0e95b2011-10-06 08:58:24 -0400487{
488 u32 *tmp, val;
489 u32 crc = ~(u32)0;
490 unsigned offset = 0;
491
492 if (!io_ctl->check_crcs) {
493 io_ctl_map_page(io_ctl, 0);
494 return 0;
495 }
496
497 if (index == 0)
498 offset = sizeof(u32) * io_ctl->num_pages;
499
Chris Mason2b108262015-04-06 07:48:20 -0700500 tmp = page_address(io_ctl->pages[0]);
Josef Bacik5b0e95b2011-10-06 08:58:24 -0400501 tmp += index;
502 val = *tmp;
Josef Bacik5b0e95b2011-10-06 08:58:24 -0400503
504 io_ctl_map_page(io_ctl, 0);
Liu Bob0496682013-03-14 14:57:45 +0000505 crc = btrfs_csum_data(io_ctl->orig + offset, crc,
Josef Bacik5b0e95b2011-10-06 08:58:24 -0400506 PAGE_CACHE_SIZE - offset);
507 btrfs_csum_final(crc, (char *)&crc);
508 if (val != crc) {
Frank Holtonefe120a2013-12-20 11:37:06 -0500509 printk_ratelimited(KERN_ERR "BTRFS: csum mismatch on free "
Josef Bacik5b0e95b2011-10-06 08:58:24 -0400510 "space cache\n");
511 io_ctl_unmap_page(io_ctl);
512 return -EIO;
513 }
514
Josef Bacika67509c2011-10-05 15:18:58 -0400515 return 0;
516}
517
Chris Mason4c6d1d82015-04-06 13:17:20 -0700518static int io_ctl_add_entry(struct btrfs_io_ctl *io_ctl, u64 offset, u64 bytes,
Josef Bacika67509c2011-10-05 15:18:58 -0400519 void *bitmap)
520{
521 struct btrfs_free_space_entry *entry;
522
523 if (!io_ctl->cur)
524 return -ENOSPC;
525
526 entry = io_ctl->cur;
527 entry->offset = cpu_to_le64(offset);
528 entry->bytes = cpu_to_le64(bytes);
529 entry->type = (bitmap) ? BTRFS_FREE_SPACE_BITMAP :
530 BTRFS_FREE_SPACE_EXTENT;
531 io_ctl->cur += sizeof(struct btrfs_free_space_entry);
532 io_ctl->size -= sizeof(struct btrfs_free_space_entry);
533
534 if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
535 return 0;
536
Josef Bacik5b0e95b2011-10-06 08:58:24 -0400537 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
Josef Bacika67509c2011-10-05 15:18:58 -0400538
539 /* No more pages to map */
540 if (io_ctl->index >= io_ctl->num_pages)
541 return 0;
542
543 /* map the next page */
544 io_ctl_map_page(io_ctl, 1);
545 return 0;
546}
547
Chris Mason4c6d1d82015-04-06 13:17:20 -0700548static int io_ctl_add_bitmap(struct btrfs_io_ctl *io_ctl, void *bitmap)
Josef Bacika67509c2011-10-05 15:18:58 -0400549{
550 if (!io_ctl->cur)
551 return -ENOSPC;
552
553 /*
554 * If we aren't at the start of the current page, unmap this one and
555 * map the next one if there is any left.
556 */
557 if (io_ctl->cur != io_ctl->orig) {
Josef Bacik5b0e95b2011-10-06 08:58:24 -0400558 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
Josef Bacika67509c2011-10-05 15:18:58 -0400559 if (io_ctl->index >= io_ctl->num_pages)
560 return -ENOSPC;
561 io_ctl_map_page(io_ctl, 0);
562 }
563
564 memcpy(io_ctl->cur, bitmap, PAGE_CACHE_SIZE);
Josef Bacik5b0e95b2011-10-06 08:58:24 -0400565 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
Josef Bacika67509c2011-10-05 15:18:58 -0400566 if (io_ctl->index < io_ctl->num_pages)
567 io_ctl_map_page(io_ctl, 0);
568 return 0;
569}
570
Chris Mason4c6d1d82015-04-06 13:17:20 -0700571static void io_ctl_zero_remaining_pages(struct btrfs_io_ctl *io_ctl)
Josef Bacika67509c2011-10-05 15:18:58 -0400572{
Josef Bacik5b0e95b2011-10-06 08:58:24 -0400573 /*
574 * If we're not on the boundary we know we've modified the page and we
575 * need to crc the page.
576 */
577 if (io_ctl->cur != io_ctl->orig)
578 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
579 else
580 io_ctl_unmap_page(io_ctl);
Josef Bacika67509c2011-10-05 15:18:58 -0400581
582 while (io_ctl->index < io_ctl->num_pages) {
583 io_ctl_map_page(io_ctl, 1);
Josef Bacik5b0e95b2011-10-06 08:58:24 -0400584 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
Josef Bacika67509c2011-10-05 15:18:58 -0400585 }
586}
587
Chris Mason4c6d1d82015-04-06 13:17:20 -0700588static int io_ctl_read_entry(struct btrfs_io_ctl *io_ctl,
Josef Bacik5b0e95b2011-10-06 08:58:24 -0400589 struct btrfs_free_space *entry, u8 *type)
Josef Bacika67509c2011-10-05 15:18:58 -0400590{
591 struct btrfs_free_space_entry *e;
Josef Bacik2f120c02011-11-10 20:45:05 -0500592 int ret;
593
594 if (!io_ctl->cur) {
595 ret = io_ctl_check_crc(io_ctl, io_ctl->index);
596 if (ret)
597 return ret;
598 }
Josef Bacika67509c2011-10-05 15:18:58 -0400599
600 e = io_ctl->cur;
601 entry->offset = le64_to_cpu(e->offset);
602 entry->bytes = le64_to_cpu(e->bytes);
Josef Bacik5b0e95b2011-10-06 08:58:24 -0400603 *type = e->type;
Josef Bacika67509c2011-10-05 15:18:58 -0400604 io_ctl->cur += sizeof(struct btrfs_free_space_entry);
605 io_ctl->size -= sizeof(struct btrfs_free_space_entry);
606
607 if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
Josef Bacik5b0e95b2011-10-06 08:58:24 -0400608 return 0;
Josef Bacika67509c2011-10-05 15:18:58 -0400609
610 io_ctl_unmap_page(io_ctl);
611
Josef Bacik2f120c02011-11-10 20:45:05 -0500612 return 0;
Josef Bacika67509c2011-10-05 15:18:58 -0400613}
614
Chris Mason4c6d1d82015-04-06 13:17:20 -0700615static int io_ctl_read_bitmap(struct btrfs_io_ctl *io_ctl,
Josef Bacik5b0e95b2011-10-06 08:58:24 -0400616 struct btrfs_free_space *entry)
Josef Bacika67509c2011-10-05 15:18:58 -0400617{
Josef Bacik5b0e95b2011-10-06 08:58:24 -0400618 int ret;
619
Josef Bacik5b0e95b2011-10-06 08:58:24 -0400620 ret = io_ctl_check_crc(io_ctl, io_ctl->index);
621 if (ret)
622 return ret;
623
Josef Bacika67509c2011-10-05 15:18:58 -0400624 memcpy(entry->bitmap, io_ctl->cur, PAGE_CACHE_SIZE);
625 io_ctl_unmap_page(io_ctl);
Josef Bacik5b0e95b2011-10-06 08:58:24 -0400626
627 return 0;
Josef Bacika67509c2011-10-05 15:18:58 -0400628}
629
Josef Bacikcd023e72012-05-14 10:06:40 -0400630/*
631 * Since we attach pinned extents after the fact we can have contiguous sections
632 * of free space that are split up in entries. This poses a problem with the
633 * tree logging stuff since it could have allocated across what appears to be 2
634 * entries since we would have merged the entries when adding the pinned extents
635 * back to the free space cache. So run through the space cache that we just
636 * loaded and merge contiguous entries. This will make the log replay stuff not
637 * blow up and it will make for nicer allocator behavior.
638 */
639static void merge_space_tree(struct btrfs_free_space_ctl *ctl)
640{
641 struct btrfs_free_space *e, *prev = NULL;
642 struct rb_node *n;
643
644again:
645 spin_lock(&ctl->tree_lock);
646 for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) {
647 e = rb_entry(n, struct btrfs_free_space, offset_index);
648 if (!prev)
649 goto next;
650 if (e->bitmap || prev->bitmap)
651 goto next;
652 if (prev->offset + prev->bytes == e->offset) {
653 unlink_free_space(ctl, prev);
654 unlink_free_space(ctl, e);
655 prev->bytes += e->bytes;
656 kmem_cache_free(btrfs_free_space_cachep, e);
657 link_free_space(ctl, prev);
658 prev = NULL;
659 spin_unlock(&ctl->tree_lock);
660 goto again;
661 }
662next:
663 prev = e;
664 }
665 spin_unlock(&ctl->tree_lock);
666}
667
Eric Sandeen48a3b632013-04-25 20:41:01 +0000668static int __load_free_space_cache(struct btrfs_root *root, struct inode *inode,
669 struct btrfs_free_space_ctl *ctl,
670 struct btrfs_path *path, u64 offset)
Josef Bacik9d66e232010-08-25 16:54:15 -0400671{
Josef Bacik9d66e232010-08-25 16:54:15 -0400672 struct btrfs_free_space_header *header;
673 struct extent_buffer *leaf;
Chris Mason4c6d1d82015-04-06 13:17:20 -0700674 struct btrfs_io_ctl io_ctl;
Josef Bacik9d66e232010-08-25 16:54:15 -0400675 struct btrfs_key key;
Josef Bacika67509c2011-10-05 15:18:58 -0400676 struct btrfs_free_space *e, *n;
Gui Hechengb76808f2014-12-31 09:51:35 +0800677 LIST_HEAD(bitmaps);
Josef Bacik9d66e232010-08-25 16:54:15 -0400678 u64 num_entries;
679 u64 num_bitmaps;
680 u64 generation;
Josef Bacika67509c2011-10-05 15:18:58 -0400681 u8 type;
Josef Bacikf6a39822011-06-06 10:50:35 -0400682 int ret = 0;
Josef Bacik9d66e232010-08-25 16:54:15 -0400683
Josef Bacik9d66e232010-08-25 16:54:15 -0400684 /* Nothing in the space cache, goodbye */
Li Zefan0414efa2011-04-20 10:20:14 +0800685 if (!i_size_read(inode))
Josef Bacika67509c2011-10-05 15:18:58 -0400686 return 0;
Josef Bacik9d66e232010-08-25 16:54:15 -0400687
688 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
Li Zefan0414efa2011-04-20 10:20:14 +0800689 key.offset = offset;
Josef Bacik9d66e232010-08-25 16:54:15 -0400690 key.type = 0;
691
692 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
Li Zefan0414efa2011-04-20 10:20:14 +0800693 if (ret < 0)
Josef Bacika67509c2011-10-05 15:18:58 -0400694 return 0;
Li Zefan0414efa2011-04-20 10:20:14 +0800695 else if (ret > 0) {
Chris Mason945d8962011-05-22 12:33:42 -0400696 btrfs_release_path(path);
Josef Bacika67509c2011-10-05 15:18:58 -0400697 return 0;
Josef Bacik9d66e232010-08-25 16:54:15 -0400698 }
699
Li Zefan0414efa2011-04-20 10:20:14 +0800700 ret = -1;
701
Josef Bacik9d66e232010-08-25 16:54:15 -0400702 leaf = path->nodes[0];
703 header = btrfs_item_ptr(leaf, path->slots[0],
704 struct btrfs_free_space_header);
705 num_entries = btrfs_free_space_entries(leaf, header);
706 num_bitmaps = btrfs_free_space_bitmaps(leaf, header);
707 generation = btrfs_free_space_generation(leaf, header);
Chris Mason945d8962011-05-22 12:33:42 -0400708 btrfs_release_path(path);
Josef Bacik9d66e232010-08-25 16:54:15 -0400709
Miao Xiee570fd22014-06-19 10:42:50 +0800710 if (!BTRFS_I(inode)->generation) {
711 btrfs_info(root->fs_info,
712 "The free space cache file (%llu) is invalid. skip it\n",
713 offset);
714 return 0;
715 }
716
Josef Bacik9d66e232010-08-25 16:54:15 -0400717 if (BTRFS_I(inode)->generation != generation) {
Simon Kirbyc2cf52e2013-03-19 22:41:23 +0000718 btrfs_err(root->fs_info,
719 "free space inode generation (%llu) "
720 "did not match free space cache generation (%llu)",
Geert Uytterhoevenc1c9ff72013-08-20 13:20:07 +0200721 BTRFS_I(inode)->generation, generation);
Josef Bacika67509c2011-10-05 15:18:58 -0400722 return 0;
Josef Bacik9d66e232010-08-25 16:54:15 -0400723 }
724
725 if (!num_entries)
Josef Bacika67509c2011-10-05 15:18:58 -0400726 return 0;
Josef Bacik9d66e232010-08-25 16:54:15 -0400727
Miao Xie5349d6c2014-06-19 10:42:49 +0800728 ret = io_ctl_init(&io_ctl, inode, root, 0);
Li Zefan706efc62012-01-09 14:36:28 +0800729 if (ret)
730 return ret;
731
Josef Bacik9d66e232010-08-25 16:54:15 -0400732 ret = readahead_cache(inode);
Li Zefan0414efa2011-04-20 10:20:14 +0800733 if (ret)
Josef Bacik9d66e232010-08-25 16:54:15 -0400734 goto out;
Josef Bacik9d66e232010-08-25 16:54:15 -0400735
Josef Bacika67509c2011-10-05 15:18:58 -0400736 ret = io_ctl_prepare_pages(&io_ctl, inode, 1);
737 if (ret)
738 goto out;
Josef Bacik9d66e232010-08-25 16:54:15 -0400739
Josef Bacik5b0e95b2011-10-06 08:58:24 -0400740 ret = io_ctl_check_crc(&io_ctl, 0);
741 if (ret)
742 goto free_cache;
743
Josef Bacika67509c2011-10-05 15:18:58 -0400744 ret = io_ctl_check_generation(&io_ctl, generation);
745 if (ret)
746 goto free_cache;
Josef Bacik9d66e232010-08-25 16:54:15 -0400747
Josef Bacika67509c2011-10-05 15:18:58 -0400748 while (num_entries) {
749 e = kmem_cache_zalloc(btrfs_free_space_cachep,
750 GFP_NOFS);
751 if (!e)
Josef Bacik9d66e232010-08-25 16:54:15 -0400752 goto free_cache;
Josef Bacik9d66e232010-08-25 16:54:15 -0400753
Josef Bacik5b0e95b2011-10-06 08:58:24 -0400754 ret = io_ctl_read_entry(&io_ctl, e, &type);
755 if (ret) {
756 kmem_cache_free(btrfs_free_space_cachep, e);
757 goto free_cache;
758 }
759
Josef Bacika67509c2011-10-05 15:18:58 -0400760 if (!e->bytes) {
761 kmem_cache_free(btrfs_free_space_cachep, e);
762 goto free_cache;
Josef Bacik9d66e232010-08-25 16:54:15 -0400763 }
Josef Bacik9d66e232010-08-25 16:54:15 -0400764
Josef Bacika67509c2011-10-05 15:18:58 -0400765 if (type == BTRFS_FREE_SPACE_EXTENT) {
766 spin_lock(&ctl->tree_lock);
767 ret = link_free_space(ctl, e);
768 spin_unlock(&ctl->tree_lock);
769 if (ret) {
Simon Kirbyc2cf52e2013-03-19 22:41:23 +0000770 btrfs_err(root->fs_info,
771 "Duplicate entries in free space cache, dumping");
Josef Bacikdc89e982011-01-28 17:05:48 -0500772 kmem_cache_free(btrfs_free_space_cachep, e);
Josef Bacik9d66e232010-08-25 16:54:15 -0400773 goto free_cache;
774 }
Josef Bacika67509c2011-10-05 15:18:58 -0400775 } else {
Josef Bacikb12d6862013-08-26 17:14:08 -0400776 ASSERT(num_bitmaps);
Josef Bacika67509c2011-10-05 15:18:58 -0400777 num_bitmaps--;
778 e->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
779 if (!e->bitmap) {
780 kmem_cache_free(
781 btrfs_free_space_cachep, e);
782 goto free_cache;
Josef Bacik9d66e232010-08-25 16:54:15 -0400783 }
Josef Bacika67509c2011-10-05 15:18:58 -0400784 spin_lock(&ctl->tree_lock);
785 ret = link_free_space(ctl, e);
786 ctl->total_bitmaps++;
787 ctl->op->recalc_thresholds(ctl);
788 spin_unlock(&ctl->tree_lock);
789 if (ret) {
Simon Kirbyc2cf52e2013-03-19 22:41:23 +0000790 btrfs_err(root->fs_info,
791 "Duplicate entries in free space cache, dumping");
Josef Bacika67509c2011-10-05 15:18:58 -0400792 kmem_cache_free(btrfs_free_space_cachep, e);
793 goto free_cache;
794 }
795 list_add_tail(&e->list, &bitmaps);
Josef Bacik9d66e232010-08-25 16:54:15 -0400796 }
797
Josef Bacika67509c2011-10-05 15:18:58 -0400798 num_entries--;
Josef Bacik9d66e232010-08-25 16:54:15 -0400799 }
800
Josef Bacik2f120c02011-11-10 20:45:05 -0500801 io_ctl_unmap_page(&io_ctl);
802
Josef Bacika67509c2011-10-05 15:18:58 -0400803 /*
804 * We add the bitmaps at the end of the entries in order that
805 * the bitmap entries are added to the cache.
806 */
807 list_for_each_entry_safe(e, n, &bitmaps, list) {
808 list_del_init(&e->list);
Josef Bacik5b0e95b2011-10-06 08:58:24 -0400809 ret = io_ctl_read_bitmap(&io_ctl, e);
810 if (ret)
811 goto free_cache;
Josef Bacika67509c2011-10-05 15:18:58 -0400812 }
813
814 io_ctl_drop_pages(&io_ctl);
Josef Bacikcd023e72012-05-14 10:06:40 -0400815 merge_space_tree(ctl);
Josef Bacik9d66e232010-08-25 16:54:15 -0400816 ret = 1;
817out:
Josef Bacika67509c2011-10-05 15:18:58 -0400818 io_ctl_free(&io_ctl);
Josef Bacik9d66e232010-08-25 16:54:15 -0400819 return ret;
Josef Bacik9d66e232010-08-25 16:54:15 -0400820free_cache:
Josef Bacika67509c2011-10-05 15:18:58 -0400821 io_ctl_drop_pages(&io_ctl);
Li Zefan0414efa2011-04-20 10:20:14 +0800822 __btrfs_remove_free_space_cache(ctl);
Josef Bacik9d66e232010-08-25 16:54:15 -0400823 goto out;
824}
825
Li Zefan0414efa2011-04-20 10:20:14 +0800826int load_free_space_cache(struct btrfs_fs_info *fs_info,
827 struct btrfs_block_group_cache *block_group)
Josef Bacik0cb59c92010-07-02 12:14:14 -0400828{
Li Zefan34d52cb2011-03-29 13:46:06 +0800829 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
Li Zefan0414efa2011-04-20 10:20:14 +0800830 struct btrfs_root *root = fs_info->tree_root;
831 struct inode *inode;
832 struct btrfs_path *path;
Josef Bacik5b0e95b2011-10-06 08:58:24 -0400833 int ret = 0;
Li Zefan0414efa2011-04-20 10:20:14 +0800834 bool matched;
835 u64 used = btrfs_block_group_used(&block_group->item);
836
837 /*
Li Zefan0414efa2011-04-20 10:20:14 +0800838 * If this block group has been marked to be cleared for one reason or
839 * another then we can't trust the on disk cache, so just return.
840 */
841 spin_lock(&block_group->lock);
842 if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
843 spin_unlock(&block_group->lock);
844 return 0;
845 }
846 spin_unlock(&block_group->lock);
847
848 path = btrfs_alloc_path();
849 if (!path)
850 return 0;
Josef Bacikd53ba472012-04-12 16:03:57 -0400851 path->search_commit_root = 1;
852 path->skip_locking = 1;
Li Zefan0414efa2011-04-20 10:20:14 +0800853
854 inode = lookup_free_space_inode(root, block_group, path);
855 if (IS_ERR(inode)) {
856 btrfs_free_path(path);
857 return 0;
858 }
859
Josef Bacik5b0e95b2011-10-06 08:58:24 -0400860 /* We may have converted the inode and made the cache invalid. */
861 spin_lock(&block_group->lock);
862 if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
863 spin_unlock(&block_group->lock);
Tsutomu Itoha7e221e2012-02-14 17:12:23 +0900864 btrfs_free_path(path);
Josef Bacik5b0e95b2011-10-06 08:58:24 -0400865 goto out;
866 }
867 spin_unlock(&block_group->lock);
868
Li Zefan0414efa2011-04-20 10:20:14 +0800869 ret = __load_free_space_cache(fs_info->tree_root, inode, ctl,
870 path, block_group->key.objectid);
871 btrfs_free_path(path);
872 if (ret <= 0)
873 goto out;
874
875 spin_lock(&ctl->tree_lock);
876 matched = (ctl->free_space == (block_group->key.offset - used -
877 block_group->bytes_super));
878 spin_unlock(&ctl->tree_lock);
879
880 if (!matched) {
881 __btrfs_remove_free_space_cache(ctl);
Miao Xie32d6b472014-04-24 13:31:55 +0800882 btrfs_warn(fs_info, "block group %llu has wrong amount of free space",
Simon Kirbyc2cf52e2013-03-19 22:41:23 +0000883 block_group->key.objectid);
Li Zefan0414efa2011-04-20 10:20:14 +0800884 ret = -1;
885 }
886out:
887 if (ret < 0) {
888 /* This cache is bogus, make sure it gets cleared */
889 spin_lock(&block_group->lock);
890 block_group->disk_cache_state = BTRFS_DC_CLEAR;
891 spin_unlock(&block_group->lock);
Li Zefan82d59022011-04-20 10:33:24 +0800892 ret = 0;
Li Zefan0414efa2011-04-20 10:20:14 +0800893
Miao Xie32d6b472014-04-24 13:31:55 +0800894 btrfs_warn(fs_info, "failed to load free space cache for block group %llu, rebuild it now",
Simon Kirbyc2cf52e2013-03-19 22:41:23 +0000895 block_group->key.objectid);
Li Zefan0414efa2011-04-20 10:20:14 +0800896 }
897
898 iput(inode);
899 return ret;
900}
901
Chris Masond4452bc2014-05-19 20:47:56 -0700902static noinline_for_stack
Chris Mason4c6d1d82015-04-06 13:17:20 -0700903int write_cache_extent_entries(struct btrfs_io_ctl *io_ctl,
Chris Masond4452bc2014-05-19 20:47:56 -0700904 struct btrfs_free_space_ctl *ctl,
905 struct btrfs_block_group_cache *block_group,
906 int *entries, int *bitmaps,
907 struct list_head *bitmap_list)
Josef Bacik0cb59c92010-07-02 12:14:14 -0400908{
Josef Bacikc09544e2011-08-30 10:19:10 -0400909 int ret;
Chris Masond4452bc2014-05-19 20:47:56 -0700910 struct btrfs_free_cluster *cluster = NULL;
Chris Mason1bbc6212015-04-06 12:46:08 -0700911 struct btrfs_free_cluster *cluster_locked = NULL;
Chris Masond4452bc2014-05-19 20:47:56 -0700912 struct rb_node *node = rb_first(&ctl->free_space_offset);
Filipe Manana55507ce2014-12-01 17:04:09 +0000913 struct btrfs_trim_range *trim_entry;
Josef Bacikbe1a12a2011-04-06 13:05:22 -0400914
Josef Bacik43be2142011-04-01 14:55:00 +0000915 /* Get the cluster for this block_group if it exists */
Chris Masond4452bc2014-05-19 20:47:56 -0700916 if (block_group && !list_empty(&block_group->cluster_list)) {
Josef Bacik43be2142011-04-01 14:55:00 +0000917 cluster = list_entry(block_group->cluster_list.next,
918 struct btrfs_free_cluster,
919 block_group_list);
Chris Masond4452bc2014-05-19 20:47:56 -0700920 }
Josef Bacik43be2142011-04-01 14:55:00 +0000921
Josef Bacikf75b1302011-10-05 10:00:18 -0400922 if (!node && cluster) {
Chris Mason1bbc6212015-04-06 12:46:08 -0700923 cluster_locked = cluster;
924 spin_lock(&cluster_locked->lock);
Josef Bacikf75b1302011-10-05 10:00:18 -0400925 node = rb_first(&cluster->root);
926 cluster = NULL;
927 }
928
Josef Bacik0cb59c92010-07-02 12:14:14 -0400929 /* Write out the extent entries */
Josef Bacika67509c2011-10-05 15:18:58 -0400930 while (node) {
931 struct btrfs_free_space *e;
Josef Bacik0cb59c92010-07-02 12:14:14 -0400932
Josef Bacika67509c2011-10-05 15:18:58 -0400933 e = rb_entry(node, struct btrfs_free_space, offset_index);
Chris Masond4452bc2014-05-19 20:47:56 -0700934 *entries += 1;
Josef Bacik43be2142011-04-01 14:55:00 +0000935
Chris Masond4452bc2014-05-19 20:47:56 -0700936 ret = io_ctl_add_entry(io_ctl, e->offset, e->bytes,
Josef Bacika67509c2011-10-05 15:18:58 -0400937 e->bitmap);
938 if (ret)
Chris Masond4452bc2014-05-19 20:47:56 -0700939 goto fail;
Josef Bacika67509c2011-10-05 15:18:58 -0400940
941 if (e->bitmap) {
Chris Masond4452bc2014-05-19 20:47:56 -0700942 list_add_tail(&e->list, bitmap_list);
943 *bitmaps += 1;
Josef Bacika67509c2011-10-05 15:18:58 -0400944 }
945 node = rb_next(node);
946 if (!node && cluster) {
947 node = rb_first(&cluster->root);
Chris Mason1bbc6212015-04-06 12:46:08 -0700948 cluster_locked = cluster;
949 spin_lock(&cluster_locked->lock);
Josef Bacika67509c2011-10-05 15:18:58 -0400950 cluster = NULL;
951 }
952 }
Chris Mason1bbc6212015-04-06 12:46:08 -0700953 if (cluster_locked) {
954 spin_unlock(&cluster_locked->lock);
955 cluster_locked = NULL;
956 }
Filipe Manana55507ce2014-12-01 17:04:09 +0000957
958 /*
959 * Make sure we don't miss any range that was removed from our rbtree
960 * because trimming is running. Otherwise after a umount+mount (or crash
961 * after committing the transaction) we would leak free space and get
962 * an inconsistent free space cache report from fsck.
963 */
964 list_for_each_entry(trim_entry, &ctl->trimming_ranges, list) {
965 ret = io_ctl_add_entry(io_ctl, trim_entry->start,
966 trim_entry->bytes, NULL);
967 if (ret)
968 goto fail;
969 *entries += 1;
970 }
971
Chris Masond4452bc2014-05-19 20:47:56 -0700972 return 0;
973fail:
Chris Mason1bbc6212015-04-06 12:46:08 -0700974 if (cluster_locked)
975 spin_unlock(&cluster_locked->lock);
Chris Masond4452bc2014-05-19 20:47:56 -0700976 return -ENOSPC;
977}
978
979static noinline_for_stack int
980update_cache_item(struct btrfs_trans_handle *trans,
981 struct btrfs_root *root,
982 struct inode *inode,
983 struct btrfs_path *path, u64 offset,
984 int entries, int bitmaps)
985{
986 struct btrfs_key key;
987 struct btrfs_free_space_header *header;
988 struct extent_buffer *leaf;
989 int ret;
990
991 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
992 key.offset = offset;
993 key.type = 0;
994
995 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
996 if (ret < 0) {
997 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1,
998 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, NULL,
999 GFP_NOFS);
1000 goto fail;
1001 }
1002 leaf = path->nodes[0];
1003 if (ret > 0) {
1004 struct btrfs_key found_key;
1005 ASSERT(path->slots[0]);
1006 path->slots[0]--;
1007 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1008 if (found_key.objectid != BTRFS_FREE_SPACE_OBJECTID ||
1009 found_key.offset != offset) {
1010 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0,
1011 inode->i_size - 1,
1012 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0,
1013 NULL, GFP_NOFS);
1014 btrfs_release_path(path);
1015 goto fail;
1016 }
1017 }
1018
1019 BTRFS_I(inode)->generation = trans->transid;
1020 header = btrfs_item_ptr(leaf, path->slots[0],
1021 struct btrfs_free_space_header);
1022 btrfs_set_free_space_entries(leaf, header, entries);
1023 btrfs_set_free_space_bitmaps(leaf, header, bitmaps);
1024 btrfs_set_free_space_generation(leaf, header, trans->transid);
1025 btrfs_mark_buffer_dirty(leaf);
1026 btrfs_release_path(path);
1027
1028 return 0;
1029
1030fail:
1031 return -1;
1032}
1033
1034static noinline_for_stack int
Miao Xie5349d6c2014-06-19 10:42:49 +08001035write_pinned_extent_entries(struct btrfs_root *root,
1036 struct btrfs_block_group_cache *block_group,
Chris Mason4c6d1d82015-04-06 13:17:20 -07001037 struct btrfs_io_ctl *io_ctl,
Miao Xie5349d6c2014-06-19 10:42:49 +08001038 int *entries)
Chris Masond4452bc2014-05-19 20:47:56 -07001039{
1040 u64 start, extent_start, extent_end, len;
Chris Masond4452bc2014-05-19 20:47:56 -07001041 struct extent_io_tree *unpin = NULL;
1042 int ret;
Josef Bacika67509c2011-10-05 15:18:58 -04001043
Miao Xie5349d6c2014-06-19 10:42:49 +08001044 if (!block_group)
1045 return 0;
1046
Josef Bacika67509c2011-10-05 15:18:58 -04001047 /*
1048 * We want to add any pinned extents to our free space cache
1049 * so we don't leak the space
Chris Masond4452bc2014-05-19 20:47:56 -07001050 *
Li Zefandb804f22012-01-10 16:41:01 +08001051 * We shouldn't have switched the pinned extents yet so this is the
1052 * right one
1053 */
1054 unpin = root->fs_info->pinned_extents;
1055
Miao Xie5349d6c2014-06-19 10:42:49 +08001056 start = block_group->key.objectid;
Li Zefandb804f22012-01-10 16:41:01 +08001057
Miao Xie5349d6c2014-06-19 10:42:49 +08001058 while (start < block_group->key.objectid + block_group->key.offset) {
Li Zefandb804f22012-01-10 16:41:01 +08001059 ret = find_first_extent_bit(unpin, start,
1060 &extent_start, &extent_end,
Josef Bacike6138872012-09-27 17:07:30 -04001061 EXTENT_DIRTY, NULL);
Miao Xie5349d6c2014-06-19 10:42:49 +08001062 if (ret)
1063 return 0;
Josef Bacikbe1a12a2011-04-06 13:05:22 -04001064
Josef Bacika67509c2011-10-05 15:18:58 -04001065 /* This pinned extent is out of our range */
Li Zefandb804f22012-01-10 16:41:01 +08001066 if (extent_start >= block_group->key.objectid +
Josef Bacika67509c2011-10-05 15:18:58 -04001067 block_group->key.offset)
Miao Xie5349d6c2014-06-19 10:42:49 +08001068 return 0;
Josef Bacik0cb59c92010-07-02 12:14:14 -04001069
Li Zefandb804f22012-01-10 16:41:01 +08001070 extent_start = max(extent_start, start);
1071 extent_end = min(block_group->key.objectid +
1072 block_group->key.offset, extent_end + 1);
1073 len = extent_end - extent_start;
Josef Bacik0cb59c92010-07-02 12:14:14 -04001074
Chris Masond4452bc2014-05-19 20:47:56 -07001075 *entries += 1;
1076 ret = io_ctl_add_entry(io_ctl, extent_start, len, NULL);
Josef Bacika67509c2011-10-05 15:18:58 -04001077 if (ret)
Miao Xie5349d6c2014-06-19 10:42:49 +08001078 return -ENOSPC;
Josef Bacik2f356122011-06-10 15:31:13 -04001079
Li Zefandb804f22012-01-10 16:41:01 +08001080 start = extent_end;
Josef Bacika67509c2011-10-05 15:18:58 -04001081 }
Josef Bacik0cb59c92010-07-02 12:14:14 -04001082
Miao Xie5349d6c2014-06-19 10:42:49 +08001083 return 0;
1084}
1085
1086static noinline_for_stack int
Chris Mason4c6d1d82015-04-06 13:17:20 -07001087write_bitmap_entries(struct btrfs_io_ctl *io_ctl, struct list_head *bitmap_list)
Miao Xie5349d6c2014-06-19 10:42:49 +08001088{
1089 struct list_head *pos, *n;
1090 int ret;
1091
Josef Bacik0cb59c92010-07-02 12:14:14 -04001092 /* Write out the bitmaps */
Chris Masond4452bc2014-05-19 20:47:56 -07001093 list_for_each_safe(pos, n, bitmap_list) {
Josef Bacik0cb59c92010-07-02 12:14:14 -04001094 struct btrfs_free_space *entry =
1095 list_entry(pos, struct btrfs_free_space, list);
1096
Chris Masond4452bc2014-05-19 20:47:56 -07001097 ret = io_ctl_add_bitmap(io_ctl, entry->bitmap);
Josef Bacika67509c2011-10-05 15:18:58 -04001098 if (ret)
Miao Xie5349d6c2014-06-19 10:42:49 +08001099 return -ENOSPC;
Josef Bacik0cb59c92010-07-02 12:14:14 -04001100 list_del_init(&entry->list);
Josef Bacikbe1a12a2011-04-06 13:05:22 -04001101 }
1102
Miao Xie5349d6c2014-06-19 10:42:49 +08001103 return 0;
1104}
Josef Bacik0cb59c92010-07-02 12:14:14 -04001105
Miao Xie5349d6c2014-06-19 10:42:49 +08001106static int flush_dirty_cache(struct inode *inode)
1107{
1108 int ret;
Josef Bacikbe1a12a2011-04-06 13:05:22 -04001109
Josef Bacik0ef8b722013-10-25 16:13:35 -04001110 ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
Miao Xie5349d6c2014-06-19 10:42:49 +08001111 if (ret)
Josef Bacik0ef8b722013-10-25 16:13:35 -04001112 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1,
1113 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, NULL,
1114 GFP_NOFS);
Chris Masond4452bc2014-05-19 20:47:56 -07001115
Miao Xie5349d6c2014-06-19 10:42:49 +08001116 return ret;
Chris Masond4452bc2014-05-19 20:47:56 -07001117}
1118
1119static void noinline_for_stack
Chris Masona3bdccc2015-04-24 11:00:00 -07001120cleanup_bitmap_list(struct list_head *bitmap_list)
Chris Masond4452bc2014-05-19 20:47:56 -07001121{
1122 struct list_head *pos, *n;
Miao Xie5349d6c2014-06-19 10:42:49 +08001123
Chris Masond4452bc2014-05-19 20:47:56 -07001124 list_for_each_safe(pos, n, bitmap_list) {
1125 struct btrfs_free_space *entry =
1126 list_entry(pos, struct btrfs_free_space, list);
1127 list_del_init(&entry->list);
1128 }
Chris Masona3bdccc2015-04-24 11:00:00 -07001129}
1130
1131static void noinline_for_stack
1132cleanup_write_cache_enospc(struct inode *inode,
1133 struct btrfs_io_ctl *io_ctl,
1134 struct extent_state **cached_state,
1135 struct list_head *bitmap_list)
1136{
Chris Masond4452bc2014-05-19 20:47:56 -07001137 io_ctl_drop_pages(io_ctl);
1138 unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
1139 i_size_read(inode) - 1, cached_state,
1140 GFP_NOFS);
1141}
1142
Chris Masonc9dc4c62015-04-04 17:14:42 -07001143int btrfs_wait_cache_io(struct btrfs_root *root,
1144 struct btrfs_trans_handle *trans,
1145 struct btrfs_block_group_cache *block_group,
1146 struct btrfs_io_ctl *io_ctl,
1147 struct btrfs_path *path, u64 offset)
1148{
1149 int ret;
1150 struct inode *inode = io_ctl->inode;
1151
Chris Mason1bbc6212015-04-06 12:46:08 -07001152 if (!inode)
1153 return 0;
1154
Chris Mason85db36c2015-04-23 08:02:49 -07001155 if (block_group)
1156 root = root->fs_info->tree_root;
Chris Masonc9dc4c62015-04-04 17:14:42 -07001157
1158 /* Flush the dirty pages in the cache file. */
1159 ret = flush_dirty_cache(inode);
1160 if (ret)
1161 goto out;
1162
1163 /* Update the cache item to tell everyone this cache file is valid. */
1164 ret = update_cache_item(trans, root, inode, path, offset,
1165 io_ctl->entries, io_ctl->bitmaps);
1166out:
1167 io_ctl_free(io_ctl);
1168 if (ret) {
1169 invalidate_inode_pages2(inode->i_mapping);
1170 BTRFS_I(inode)->generation = 0;
1171 if (block_group) {
1172#ifdef DEBUG
1173 btrfs_err(root->fs_info,
1174 "failed to write free space cache for block group %llu",
1175 block_group->key.objectid);
1176#endif
1177 }
1178 }
1179 btrfs_update_inode(trans, root, inode);
1180
1181 if (block_group) {
Chris Mason1bbc6212015-04-06 12:46:08 -07001182 /* the dirty list is protected by the dirty_bgs_lock */
1183 spin_lock(&trans->transaction->dirty_bgs_lock);
1184
1185 /* the disk_cache_state is protected by the block group lock */
Chris Masonc9dc4c62015-04-04 17:14:42 -07001186 spin_lock(&block_group->lock);
1187
1188 /*
1189 * only mark this as written if we didn't get put back on
Chris Mason1bbc6212015-04-06 12:46:08 -07001190 * the dirty list while waiting for IO. Otherwise our
1191 * cache state won't be right, and we won't get written again
Chris Masonc9dc4c62015-04-04 17:14:42 -07001192 */
1193 if (!ret && list_empty(&block_group->dirty_list))
1194 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
1195 else if (ret)
1196 block_group->disk_cache_state = BTRFS_DC_ERROR;
1197
1198 spin_unlock(&block_group->lock);
Chris Mason1bbc6212015-04-06 12:46:08 -07001199 spin_unlock(&trans->transaction->dirty_bgs_lock);
Chris Masonc9dc4c62015-04-04 17:14:42 -07001200 io_ctl->inode = NULL;
1201 iput(inode);
1202 }
1203
1204 return ret;
1205
1206}
1207
Chris Masond4452bc2014-05-19 20:47:56 -07001208/**
1209 * __btrfs_write_out_cache - write out cached info to an inode
1210 * @root - the root the inode belongs to
1211 * @ctl - the free space cache we are going to write out
1212 * @block_group - the block_group for this cache if it belongs to a block_group
1213 * @trans - the trans handle
1214 * @path - the path to use
1215 * @offset - the offset for the key we'll insert
1216 *
1217 * This function writes out a free space cache struct to disk for quick recovery
1218 * on mount. This will return 0 if it was successfull in writing the cache out,
Omar Sandovalb8605452015-02-24 02:47:06 -08001219 * or an errno if it was not.
Chris Masond4452bc2014-05-19 20:47:56 -07001220 */
1221static int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode,
1222 struct btrfs_free_space_ctl *ctl,
1223 struct btrfs_block_group_cache *block_group,
Chris Masonc9dc4c62015-04-04 17:14:42 -07001224 struct btrfs_io_ctl *io_ctl,
Chris Masond4452bc2014-05-19 20:47:56 -07001225 struct btrfs_trans_handle *trans,
1226 struct btrfs_path *path, u64 offset)
1227{
1228 struct extent_state *cached_state = NULL;
Miao Xie5349d6c2014-06-19 10:42:49 +08001229 LIST_HEAD(bitmap_list);
Chris Masond4452bc2014-05-19 20:47:56 -07001230 int entries = 0;
1231 int bitmaps = 0;
1232 int ret;
Chris Masonc9dc4c62015-04-04 17:14:42 -07001233 int must_iput = 0;
Chris Masond4452bc2014-05-19 20:47:56 -07001234
1235 if (!i_size_read(inode))
Omar Sandovalb8605452015-02-24 02:47:06 -08001236 return -EIO;
Chris Masond4452bc2014-05-19 20:47:56 -07001237
Chris Masonc9dc4c62015-04-04 17:14:42 -07001238 WARN_ON(io_ctl->pages);
1239 ret = io_ctl_init(io_ctl, inode, root, 1);
Chris Masond4452bc2014-05-19 20:47:56 -07001240 if (ret)
Omar Sandovalb8605452015-02-24 02:47:06 -08001241 return ret;
Chris Masond4452bc2014-05-19 20:47:56 -07001242
Miao Xiee570fd22014-06-19 10:42:50 +08001243 if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA)) {
1244 down_write(&block_group->data_rwsem);
1245 spin_lock(&block_group->lock);
1246 if (block_group->delalloc_bytes) {
1247 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
1248 spin_unlock(&block_group->lock);
1249 up_write(&block_group->data_rwsem);
1250 BTRFS_I(inode)->generation = 0;
1251 ret = 0;
Chris Masonc9dc4c62015-04-04 17:14:42 -07001252 must_iput = 1;
Miao Xiee570fd22014-06-19 10:42:50 +08001253 goto out;
1254 }
1255 spin_unlock(&block_group->lock);
1256 }
1257
Chris Masond4452bc2014-05-19 20:47:56 -07001258 /* Lock all pages first so we can lock the extent safely. */
Omar Sandovalb8605452015-02-24 02:47:06 -08001259 ret = io_ctl_prepare_pages(io_ctl, inode, 0);
1260 if (ret)
1261 goto out;
Chris Masond4452bc2014-05-19 20:47:56 -07001262
1263 lock_extent_bits(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1,
1264 0, &cached_state);
1265
Chris Masonc9dc4c62015-04-04 17:14:42 -07001266 io_ctl_set_generation(io_ctl, trans->transid);
Chris Masond4452bc2014-05-19 20:47:56 -07001267
Filipe Manana55507ce2014-12-01 17:04:09 +00001268 mutex_lock(&ctl->cache_writeout_mutex);
Miao Xie5349d6c2014-06-19 10:42:49 +08001269 /* Write out the extent entries in the free space cache */
Chris Mason1bbc6212015-04-06 12:46:08 -07001270 spin_lock(&ctl->tree_lock);
Chris Masonc9dc4c62015-04-04 17:14:42 -07001271 ret = write_cache_extent_entries(io_ctl, ctl,
Chris Masond4452bc2014-05-19 20:47:56 -07001272 block_group, &entries, &bitmaps,
1273 &bitmap_list);
Chris Masona3bdccc2015-04-24 11:00:00 -07001274 if (ret)
1275 goto out_nospc_locked;
Chris Masond4452bc2014-05-19 20:47:56 -07001276
Miao Xie5349d6c2014-06-19 10:42:49 +08001277 /*
1278 * Some spaces that are freed in the current transaction are pinned,
1279 * they will be added into free space cache after the transaction is
1280 * committed, we shouldn't lose them.
Chris Mason1bbc6212015-04-06 12:46:08 -07001281 *
1282 * If this changes while we are working we'll get added back to
1283 * the dirty list and redo it. No locking needed
Miao Xie5349d6c2014-06-19 10:42:49 +08001284 */
Chris Masonc9dc4c62015-04-04 17:14:42 -07001285 ret = write_pinned_extent_entries(root, block_group, io_ctl, &entries);
Chris Masona3bdccc2015-04-24 11:00:00 -07001286 if (ret)
1287 goto out_nospc_locked;
Miao Xie5349d6c2014-06-19 10:42:49 +08001288
Filipe Manana55507ce2014-12-01 17:04:09 +00001289 /*
1290 * At last, we write out all the bitmaps and keep cache_writeout_mutex
1291 * locked while doing it because a concurrent trim can be manipulating
1292 * or freeing the bitmap.
1293 */
Chris Masonc9dc4c62015-04-04 17:14:42 -07001294 ret = write_bitmap_entries(io_ctl, &bitmap_list);
Chris Mason1bbc6212015-04-06 12:46:08 -07001295 spin_unlock(&ctl->tree_lock);
Filipe Manana55507ce2014-12-01 17:04:09 +00001296 mutex_unlock(&ctl->cache_writeout_mutex);
Miao Xie5349d6c2014-06-19 10:42:49 +08001297 if (ret)
1298 goto out_nospc;
1299
1300 /* Zero out the rest of the pages just to make sure */
Chris Masonc9dc4c62015-04-04 17:14:42 -07001301 io_ctl_zero_remaining_pages(io_ctl);
Miao Xie5349d6c2014-06-19 10:42:49 +08001302
1303 /* Everything is written out, now we dirty the pages in the file. */
Chris Masonc9dc4c62015-04-04 17:14:42 -07001304 ret = btrfs_dirty_pages(root, inode, io_ctl->pages, io_ctl->num_pages,
Miao Xie5349d6c2014-06-19 10:42:49 +08001305 0, i_size_read(inode), &cached_state);
1306 if (ret)
1307 goto out_nospc;
1308
Miao Xiee570fd22014-06-19 10:42:50 +08001309 if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA))
1310 up_write(&block_group->data_rwsem);
Miao Xie5349d6c2014-06-19 10:42:49 +08001311 /*
1312 * Release the pages and unlock the extent, we will flush
1313 * them out later
1314 */
Chris Masonc9dc4c62015-04-04 17:14:42 -07001315 io_ctl_drop_pages(io_ctl);
Miao Xie5349d6c2014-06-19 10:42:49 +08001316
1317 unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
1318 i_size_read(inode) - 1, &cached_state, GFP_NOFS);
1319
Chris Masonc9dc4c62015-04-04 17:14:42 -07001320 /*
1321 * at this point the pages are under IO and we're happy,
1322 * The caller is responsible for waiting on them and updating the
1323 * the cache and the inode
1324 */
1325 io_ctl->entries = entries;
1326 io_ctl->bitmaps = bitmaps;
1327
1328 ret = btrfs_fdatawrite_range(inode, 0, (u64)-1);
Miao Xie5349d6c2014-06-19 10:42:49 +08001329 if (ret)
Josef Bacik0ef8b722013-10-25 16:13:35 -04001330 goto out;
Josef Bacik0cb59c92010-07-02 12:14:14 -04001331
Chris Masonc9dc4c62015-04-04 17:14:42 -07001332 return 0;
1333
Josef Bacik2f356122011-06-10 15:31:13 -04001334out:
Chris Masonc9dc4c62015-04-04 17:14:42 -07001335 io_ctl->inode = NULL;
1336 io_ctl_free(io_ctl);
Miao Xie5349d6c2014-06-19 10:42:49 +08001337 if (ret) {
Josef Bacika67509c2011-10-05 15:18:58 -04001338 invalidate_inode_pages2(inode->i_mapping);
Josef Bacik0cb59c92010-07-02 12:14:14 -04001339 BTRFS_I(inode)->generation = 0;
1340 }
Josef Bacik0cb59c92010-07-02 12:14:14 -04001341 btrfs_update_inode(trans, root, inode);
Chris Masonc9dc4c62015-04-04 17:14:42 -07001342 if (must_iput)
1343 iput(inode);
Miao Xie5349d6c2014-06-19 10:42:49 +08001344 return ret;
Josef Bacika67509c2011-10-05 15:18:58 -04001345
Chris Masona3bdccc2015-04-24 11:00:00 -07001346out_nospc_locked:
1347 cleanup_bitmap_list(&bitmap_list);
1348 spin_unlock(&ctl->tree_lock);
1349 mutex_unlock(&ctl->cache_writeout_mutex);
1350
Josef Bacika67509c2011-10-05 15:18:58 -04001351out_nospc:
Chris Masonc9dc4c62015-04-04 17:14:42 -07001352 cleanup_write_cache_enospc(inode, io_ctl, &cached_state, &bitmap_list);
Miao Xiee570fd22014-06-19 10:42:50 +08001353
1354 if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA))
1355 up_write(&block_group->data_rwsem);
1356
Josef Bacika67509c2011-10-05 15:18:58 -04001357 goto out;
Li Zefan0414efa2011-04-20 10:20:14 +08001358}
1359
1360int btrfs_write_out_cache(struct btrfs_root *root,
1361 struct btrfs_trans_handle *trans,
1362 struct btrfs_block_group_cache *block_group,
1363 struct btrfs_path *path)
1364{
1365 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1366 struct inode *inode;
1367 int ret = 0;
1368
1369 root = root->fs_info->tree_root;
1370
1371 spin_lock(&block_group->lock);
1372 if (block_group->disk_cache_state < BTRFS_DC_SETUP) {
1373 spin_unlock(&block_group->lock);
1374 return 0;
1375 }
1376 spin_unlock(&block_group->lock);
1377
1378 inode = lookup_free_space_inode(root, block_group, path);
1379 if (IS_ERR(inode))
1380 return 0;
1381
Chris Masonc9dc4c62015-04-04 17:14:42 -07001382 ret = __btrfs_write_out_cache(root, inode, ctl, block_group,
1383 &block_group->io_ctl, trans,
Li Zefan0414efa2011-04-20 10:20:14 +08001384 path, block_group->key.objectid);
Josef Bacikc09544e2011-08-30 10:19:10 -04001385 if (ret) {
Josef Bacikc09544e2011-08-30 10:19:10 -04001386#ifdef DEBUG
Simon Kirbyc2cf52e2013-03-19 22:41:23 +00001387 btrfs_err(root->fs_info,
1388 "failed to write free space cache for block group %llu",
1389 block_group->key.objectid);
Josef Bacikc09544e2011-08-30 10:19:10 -04001390#endif
Chris Masonc9dc4c62015-04-04 17:14:42 -07001391 spin_lock(&block_group->lock);
1392 block_group->disk_cache_state = BTRFS_DC_ERROR;
1393 spin_unlock(&block_group->lock);
1394
1395 block_group->io_ctl.inode = NULL;
1396 iput(inode);
Li Zefan0414efa2011-04-20 10:20:14 +08001397 }
1398
Chris Masonc9dc4c62015-04-04 17:14:42 -07001399 /*
1400 * if ret == 0 the caller is expected to call btrfs_wait_cache_io
1401 * to wait for IO and put the inode
1402 */
1403
Josef Bacik0cb59c92010-07-02 12:14:14 -04001404 return ret;
1405}
1406
Li Zefan34d52cb2011-03-29 13:46:06 +08001407static inline unsigned long offset_to_bit(u64 bitmap_start, u32 unit,
Josef Bacik96303082009-07-13 21:29:25 -04001408 u64 offset)
1409{
Josef Bacikb12d6862013-08-26 17:14:08 -04001410 ASSERT(offset >= bitmap_start);
Josef Bacik96303082009-07-13 21:29:25 -04001411 offset -= bitmap_start;
Li Zefan34d52cb2011-03-29 13:46:06 +08001412 return (unsigned long)(div_u64(offset, unit));
Josef Bacik96303082009-07-13 21:29:25 -04001413}
1414
Li Zefan34d52cb2011-03-29 13:46:06 +08001415static inline unsigned long bytes_to_bits(u64 bytes, u32 unit)
Josef Bacik96303082009-07-13 21:29:25 -04001416{
Li Zefan34d52cb2011-03-29 13:46:06 +08001417 return (unsigned long)(div_u64(bytes, unit));
Josef Bacik96303082009-07-13 21:29:25 -04001418}
1419
Li Zefan34d52cb2011-03-29 13:46:06 +08001420static inline u64 offset_to_bitmap(struct btrfs_free_space_ctl *ctl,
Josef Bacik96303082009-07-13 21:29:25 -04001421 u64 offset)
1422{
1423 u64 bitmap_start;
David Sterbab8b93ad2015-01-16 17:26:13 +01001424 u32 bytes_per_bitmap;
Josef Bacik96303082009-07-13 21:29:25 -04001425
Li Zefan34d52cb2011-03-29 13:46:06 +08001426 bytes_per_bitmap = BITS_PER_BITMAP * ctl->unit;
1427 bitmap_start = offset - ctl->start;
David Sterbab8b93ad2015-01-16 17:26:13 +01001428 bitmap_start = div_u64(bitmap_start, bytes_per_bitmap);
Josef Bacik96303082009-07-13 21:29:25 -04001429 bitmap_start *= bytes_per_bitmap;
Li Zefan34d52cb2011-03-29 13:46:06 +08001430 bitmap_start += ctl->start;
Josef Bacik96303082009-07-13 21:29:25 -04001431
1432 return bitmap_start;
1433}
Josef Bacik0f9dd462008-09-23 13:14:11 -04001434
1435static int tree_insert_offset(struct rb_root *root, u64 offset,
Josef Bacik96303082009-07-13 21:29:25 -04001436 struct rb_node *node, int bitmap)
Josef Bacik0f9dd462008-09-23 13:14:11 -04001437{
1438 struct rb_node **p = &root->rb_node;
1439 struct rb_node *parent = NULL;
1440 struct btrfs_free_space *info;
1441
1442 while (*p) {
1443 parent = *p;
1444 info = rb_entry(parent, struct btrfs_free_space, offset_index);
1445
Josef Bacik96303082009-07-13 21:29:25 -04001446 if (offset < info->offset) {
Josef Bacik0f9dd462008-09-23 13:14:11 -04001447 p = &(*p)->rb_left;
Josef Bacik96303082009-07-13 21:29:25 -04001448 } else if (offset > info->offset) {
Josef Bacik0f9dd462008-09-23 13:14:11 -04001449 p = &(*p)->rb_right;
Josef Bacik96303082009-07-13 21:29:25 -04001450 } else {
1451 /*
1452 * we could have a bitmap entry and an extent entry
1453 * share the same offset. If this is the case, we want
1454 * the extent entry to always be found first if we do a
1455 * linear search through the tree, since we want to have
1456 * the quickest allocation time, and allocating from an
1457 * extent is faster than allocating from a bitmap. So
1458 * if we're inserting a bitmap and we find an entry at
1459 * this offset, we want to go right, or after this entry
1460 * logically. If we are inserting an extent and we've
1461 * found a bitmap, we want to go left, or before
1462 * logically.
1463 */
1464 if (bitmap) {
Josef Bacik207dde82011-05-13 14:49:23 -04001465 if (info->bitmap) {
1466 WARN_ON_ONCE(1);
1467 return -EEXIST;
1468 }
Josef Bacik96303082009-07-13 21:29:25 -04001469 p = &(*p)->rb_right;
1470 } else {
Josef Bacik207dde82011-05-13 14:49:23 -04001471 if (!info->bitmap) {
1472 WARN_ON_ONCE(1);
1473 return -EEXIST;
1474 }
Josef Bacik96303082009-07-13 21:29:25 -04001475 p = &(*p)->rb_left;
1476 }
1477 }
Josef Bacik0f9dd462008-09-23 13:14:11 -04001478 }
1479
1480 rb_link_node(node, parent, p);
1481 rb_insert_color(node, root);
1482
1483 return 0;
1484}
1485
1486/*
Josef Bacik70cb0742009-04-03 10:14:19 -04001487 * searches the tree for the given offset.
1488 *
Josef Bacik96303082009-07-13 21:29:25 -04001489 * fuzzy - If this is set, then we are trying to make an allocation, and we just
1490 * want a section that has at least bytes size and comes at or after the given
1491 * offset.
Josef Bacik0f9dd462008-09-23 13:14:11 -04001492 */
Josef Bacik96303082009-07-13 21:29:25 -04001493static struct btrfs_free_space *
Li Zefan34d52cb2011-03-29 13:46:06 +08001494tree_search_offset(struct btrfs_free_space_ctl *ctl,
Josef Bacik96303082009-07-13 21:29:25 -04001495 u64 offset, int bitmap_only, int fuzzy)
Josef Bacik0f9dd462008-09-23 13:14:11 -04001496{
Li Zefan34d52cb2011-03-29 13:46:06 +08001497 struct rb_node *n = ctl->free_space_offset.rb_node;
Josef Bacik96303082009-07-13 21:29:25 -04001498 struct btrfs_free_space *entry, *prev = NULL;
Josef Bacik0f9dd462008-09-23 13:14:11 -04001499
Josef Bacik96303082009-07-13 21:29:25 -04001500 /* find entry that is closest to the 'offset' */
1501 while (1) {
1502 if (!n) {
1503 entry = NULL;
Josef Bacik0f9dd462008-09-23 13:14:11 -04001504 break;
1505 }
Josef Bacik96303082009-07-13 21:29:25 -04001506
1507 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1508 prev = entry;
1509
1510 if (offset < entry->offset)
1511 n = n->rb_left;
1512 else if (offset > entry->offset)
1513 n = n->rb_right;
1514 else
1515 break;
Josef Bacik0f9dd462008-09-23 13:14:11 -04001516 }
1517
Josef Bacik96303082009-07-13 21:29:25 -04001518 if (bitmap_only) {
1519 if (!entry)
1520 return NULL;
1521 if (entry->bitmap)
1522 return entry;
Josef Bacik0f9dd462008-09-23 13:14:11 -04001523
Josef Bacik96303082009-07-13 21:29:25 -04001524 /*
1525 * bitmap entry and extent entry may share same offset,
1526 * in that case, bitmap entry comes after extent entry.
1527 */
1528 n = rb_next(n);
1529 if (!n)
1530 return NULL;
1531 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1532 if (entry->offset != offset)
1533 return NULL;
Josef Bacik0f9dd462008-09-23 13:14:11 -04001534
Josef Bacik96303082009-07-13 21:29:25 -04001535 WARN_ON(!entry->bitmap);
1536 return entry;
1537 } else if (entry) {
1538 if (entry->bitmap) {
Josef Bacik0f9dd462008-09-23 13:14:11 -04001539 /*
Josef Bacik96303082009-07-13 21:29:25 -04001540 * if previous extent entry covers the offset,
1541 * we should return it instead of the bitmap entry
Josef Bacik0f9dd462008-09-23 13:14:11 -04001542 */
Miao Xiede6c4112012-10-18 08:18:01 +00001543 n = rb_prev(&entry->offset_index);
1544 if (n) {
Josef Bacik96303082009-07-13 21:29:25 -04001545 prev = rb_entry(n, struct btrfs_free_space,
1546 offset_index);
Miao Xiede6c4112012-10-18 08:18:01 +00001547 if (!prev->bitmap &&
1548 prev->offset + prev->bytes > offset)
1549 entry = prev;
Josef Bacik0f9dd462008-09-23 13:14:11 -04001550 }
Josef Bacik96303082009-07-13 21:29:25 -04001551 }
1552 return entry;
1553 }
1554
1555 if (!prev)
1556 return NULL;
1557
1558 /* find last entry before the 'offset' */
1559 entry = prev;
1560 if (entry->offset > offset) {
1561 n = rb_prev(&entry->offset_index);
1562 if (n) {
1563 entry = rb_entry(n, struct btrfs_free_space,
1564 offset_index);
Josef Bacikb12d6862013-08-26 17:14:08 -04001565 ASSERT(entry->offset <= offset);
Josef Bacik0f9dd462008-09-23 13:14:11 -04001566 } else {
Josef Bacik96303082009-07-13 21:29:25 -04001567 if (fuzzy)
1568 return entry;
1569 else
1570 return NULL;
Josef Bacik0f9dd462008-09-23 13:14:11 -04001571 }
1572 }
1573
Josef Bacik96303082009-07-13 21:29:25 -04001574 if (entry->bitmap) {
Miao Xiede6c4112012-10-18 08:18:01 +00001575 n = rb_prev(&entry->offset_index);
1576 if (n) {
Josef Bacik96303082009-07-13 21:29:25 -04001577 prev = rb_entry(n, struct btrfs_free_space,
1578 offset_index);
Miao Xiede6c4112012-10-18 08:18:01 +00001579 if (!prev->bitmap &&
1580 prev->offset + prev->bytes > offset)
1581 return prev;
Josef Bacik96303082009-07-13 21:29:25 -04001582 }
Li Zefan34d52cb2011-03-29 13:46:06 +08001583 if (entry->offset + BITS_PER_BITMAP * ctl->unit > offset)
Josef Bacik96303082009-07-13 21:29:25 -04001584 return entry;
1585 } else if (entry->offset + entry->bytes > offset)
1586 return entry;
1587
1588 if (!fuzzy)
1589 return NULL;
1590
1591 while (1) {
1592 if (entry->bitmap) {
1593 if (entry->offset + BITS_PER_BITMAP *
Li Zefan34d52cb2011-03-29 13:46:06 +08001594 ctl->unit > offset)
Josef Bacik96303082009-07-13 21:29:25 -04001595 break;
1596 } else {
1597 if (entry->offset + entry->bytes > offset)
1598 break;
1599 }
1600
1601 n = rb_next(&entry->offset_index);
1602 if (!n)
1603 return NULL;
1604 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1605 }
1606 return entry;
Josef Bacik0f9dd462008-09-23 13:14:11 -04001607}
1608
Li Zefanf333adb2010-11-09 14:57:39 +08001609static inline void
Li Zefan34d52cb2011-03-29 13:46:06 +08001610__unlink_free_space(struct btrfs_free_space_ctl *ctl,
Li Zefanf333adb2010-11-09 14:57:39 +08001611 struct btrfs_free_space *info)
Josef Bacik0f9dd462008-09-23 13:14:11 -04001612{
Li Zefan34d52cb2011-03-29 13:46:06 +08001613 rb_erase(&info->offset_index, &ctl->free_space_offset);
1614 ctl->free_extents--;
Li Zefanf333adb2010-11-09 14:57:39 +08001615}
1616
Li Zefan34d52cb2011-03-29 13:46:06 +08001617static void unlink_free_space(struct btrfs_free_space_ctl *ctl,
Li Zefanf333adb2010-11-09 14:57:39 +08001618 struct btrfs_free_space *info)
1619{
Li Zefan34d52cb2011-03-29 13:46:06 +08001620 __unlink_free_space(ctl, info);
1621 ctl->free_space -= info->bytes;
Josef Bacik0f9dd462008-09-23 13:14:11 -04001622}
1623
Li Zefan34d52cb2011-03-29 13:46:06 +08001624static int link_free_space(struct btrfs_free_space_ctl *ctl,
Josef Bacik0f9dd462008-09-23 13:14:11 -04001625 struct btrfs_free_space *info)
1626{
1627 int ret = 0;
1628
Josef Bacikb12d6862013-08-26 17:14:08 -04001629 ASSERT(info->bytes || info->bitmap);
Li Zefan34d52cb2011-03-29 13:46:06 +08001630 ret = tree_insert_offset(&ctl->free_space_offset, info->offset,
Josef Bacik96303082009-07-13 21:29:25 -04001631 &info->offset_index, (info->bitmap != NULL));
Josef Bacik0f9dd462008-09-23 13:14:11 -04001632 if (ret)
1633 return ret;
1634
Li Zefan34d52cb2011-03-29 13:46:06 +08001635 ctl->free_space += info->bytes;
1636 ctl->free_extents++;
Josef Bacik96303082009-07-13 21:29:25 -04001637 return ret;
1638}
1639
Li Zefan34d52cb2011-03-29 13:46:06 +08001640static void recalculate_thresholds(struct btrfs_free_space_ctl *ctl)
Josef Bacik96303082009-07-13 21:29:25 -04001641{
Li Zefan34d52cb2011-03-29 13:46:06 +08001642 struct btrfs_block_group_cache *block_group = ctl->private;
Josef Bacik25891f72009-09-11 16:11:20 -04001643 u64 max_bytes;
1644 u64 bitmap_bytes;
1645 u64 extent_bytes;
Li Zefan8eb2d822010-11-09 14:48:01 +08001646 u64 size = block_group->key.offset;
David Sterbab8b93ad2015-01-16 17:26:13 +01001647 u32 bytes_per_bg = BITS_PER_BITMAP * ctl->unit;
1648 u32 max_bitmaps = div_u64(size + bytes_per_bg - 1, bytes_per_bg);
Li Zefan34d52cb2011-03-29 13:46:06 +08001649
David Sterbab8b93ad2015-01-16 17:26:13 +01001650 max_bitmaps = max_t(u32, max_bitmaps, 1);
Josef Bacikdde57402013-02-12 14:07:51 -05001651
Josef Bacikb12d6862013-08-26 17:14:08 -04001652 ASSERT(ctl->total_bitmaps <= max_bitmaps);
Josef Bacik96303082009-07-13 21:29:25 -04001653
1654 /*
1655 * The goal is to keep the total amount of memory used per 1gb of space
1656 * at or below 32k, so we need to adjust how much memory we allow to be
1657 * used by extent based free space tracking
1658 */
Li Zefan8eb2d822010-11-09 14:48:01 +08001659 if (size < 1024 * 1024 * 1024)
1660 max_bytes = MAX_CACHE_BYTES_PER_GIG;
1661 else
1662 max_bytes = MAX_CACHE_BYTES_PER_GIG *
David Sterbaf8c269d2015-01-16 17:21:12 +01001663 div_u64(size, 1024 * 1024 * 1024);
Josef Bacik96303082009-07-13 21:29:25 -04001664
Josef Bacik25891f72009-09-11 16:11:20 -04001665 /*
1666 * we want to account for 1 more bitmap than what we have so we can make
1667 * sure we don't go over our overall goal of MAX_CACHE_BYTES_PER_GIG as
1668 * we add more bitmaps.
1669 */
Li Zefan34d52cb2011-03-29 13:46:06 +08001670 bitmap_bytes = (ctl->total_bitmaps + 1) * PAGE_CACHE_SIZE;
Josef Bacik96303082009-07-13 21:29:25 -04001671
Josef Bacik25891f72009-09-11 16:11:20 -04001672 if (bitmap_bytes >= max_bytes) {
Li Zefan34d52cb2011-03-29 13:46:06 +08001673 ctl->extents_thresh = 0;
Josef Bacik25891f72009-09-11 16:11:20 -04001674 return;
Josef Bacik96303082009-07-13 21:29:25 -04001675 }
Josef Bacik25891f72009-09-11 16:11:20 -04001676
1677 /*
David Sterbaf8c269d2015-01-16 17:21:12 +01001678 * we want the extent entry threshold to always be at most 1/2 the max
Josef Bacik25891f72009-09-11 16:11:20 -04001679 * bytes we can have, or whatever is less than that.
1680 */
1681 extent_bytes = max_bytes - bitmap_bytes;
David Sterbaf8c269d2015-01-16 17:21:12 +01001682 extent_bytes = min_t(u64, extent_bytes, max_bytes >> 1);
Josef Bacik25891f72009-09-11 16:11:20 -04001683
Li Zefan34d52cb2011-03-29 13:46:06 +08001684 ctl->extents_thresh =
David Sterbaf8c269d2015-01-16 17:21:12 +01001685 div_u64(extent_bytes, sizeof(struct btrfs_free_space));
Josef Bacik96303082009-07-13 21:29:25 -04001686}
1687
Miao Xiebb3ac5a2011-08-05 09:32:35 +00001688static inline void __bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
1689 struct btrfs_free_space *info,
1690 u64 offset, u64 bytes)
Josef Bacik96303082009-07-13 21:29:25 -04001691{
Li Zefanf38b6e72011-03-14 13:40:51 +08001692 unsigned long start, count;
Josef Bacik96303082009-07-13 21:29:25 -04001693
Li Zefan34d52cb2011-03-29 13:46:06 +08001694 start = offset_to_bit(info->offset, ctl->unit, offset);
1695 count = bytes_to_bits(bytes, ctl->unit);
Josef Bacikb12d6862013-08-26 17:14:08 -04001696 ASSERT(start + count <= BITS_PER_BITMAP);
Josef Bacik96303082009-07-13 21:29:25 -04001697
Li Zefanf38b6e72011-03-14 13:40:51 +08001698 bitmap_clear(info->bitmap, start, count);
Josef Bacik96303082009-07-13 21:29:25 -04001699
1700 info->bytes -= bytes;
Miao Xiebb3ac5a2011-08-05 09:32:35 +00001701}
1702
1703static void bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
1704 struct btrfs_free_space *info, u64 offset,
1705 u64 bytes)
1706{
1707 __bitmap_clear_bits(ctl, info, offset, bytes);
Li Zefan34d52cb2011-03-29 13:46:06 +08001708 ctl->free_space -= bytes;
Josef Bacik96303082009-07-13 21:29:25 -04001709}
1710
Li Zefan34d52cb2011-03-29 13:46:06 +08001711static void bitmap_set_bits(struct btrfs_free_space_ctl *ctl,
Josef Bacik817d52f2009-07-13 21:29:25 -04001712 struct btrfs_free_space *info, u64 offset,
1713 u64 bytes)
Josef Bacik96303082009-07-13 21:29:25 -04001714{
Li Zefanf38b6e72011-03-14 13:40:51 +08001715 unsigned long start, count;
Josef Bacik96303082009-07-13 21:29:25 -04001716
Li Zefan34d52cb2011-03-29 13:46:06 +08001717 start = offset_to_bit(info->offset, ctl->unit, offset);
1718 count = bytes_to_bits(bytes, ctl->unit);
Josef Bacikb12d6862013-08-26 17:14:08 -04001719 ASSERT(start + count <= BITS_PER_BITMAP);
Josef Bacik96303082009-07-13 21:29:25 -04001720
Li Zefanf38b6e72011-03-14 13:40:51 +08001721 bitmap_set(info->bitmap, start, count);
Josef Bacik96303082009-07-13 21:29:25 -04001722
1723 info->bytes += bytes;
Li Zefan34d52cb2011-03-29 13:46:06 +08001724 ctl->free_space += bytes;
Josef Bacik96303082009-07-13 21:29:25 -04001725}
1726
Miao Xiea4820392013-09-09 13:19:42 +08001727/*
1728 * If we can not find suitable extent, we will use bytes to record
1729 * the size of the max extent.
1730 */
Li Zefan34d52cb2011-03-29 13:46:06 +08001731static int search_bitmap(struct btrfs_free_space_ctl *ctl,
Josef Bacik96303082009-07-13 21:29:25 -04001732 struct btrfs_free_space *bitmap_info, u64 *offset,
1733 u64 *bytes)
1734{
1735 unsigned long found_bits = 0;
Miao Xiea4820392013-09-09 13:19:42 +08001736 unsigned long max_bits = 0;
Josef Bacik96303082009-07-13 21:29:25 -04001737 unsigned long bits, i;
1738 unsigned long next_zero;
Miao Xiea4820392013-09-09 13:19:42 +08001739 unsigned long extent_bits;
Josef Bacik96303082009-07-13 21:29:25 -04001740
Li Zefan34d52cb2011-03-29 13:46:06 +08001741 i = offset_to_bit(bitmap_info->offset, ctl->unit,
Josef Bacik96303082009-07-13 21:29:25 -04001742 max_t(u64, *offset, bitmap_info->offset));
Li Zefan34d52cb2011-03-29 13:46:06 +08001743 bits = bytes_to_bits(*bytes, ctl->unit);
Josef Bacik96303082009-07-13 21:29:25 -04001744
Wei Yongjunebb3dad2012-09-13 20:29:02 -06001745 for_each_set_bit_from(i, bitmap_info->bitmap, BITS_PER_BITMAP) {
Josef Bacik96303082009-07-13 21:29:25 -04001746 next_zero = find_next_zero_bit(bitmap_info->bitmap,
1747 BITS_PER_BITMAP, i);
Miao Xiea4820392013-09-09 13:19:42 +08001748 extent_bits = next_zero - i;
1749 if (extent_bits >= bits) {
1750 found_bits = extent_bits;
Josef Bacik96303082009-07-13 21:29:25 -04001751 break;
Miao Xiea4820392013-09-09 13:19:42 +08001752 } else if (extent_bits > max_bits) {
1753 max_bits = extent_bits;
Josef Bacik96303082009-07-13 21:29:25 -04001754 }
1755 i = next_zero;
1756 }
1757
1758 if (found_bits) {
Li Zefan34d52cb2011-03-29 13:46:06 +08001759 *offset = (u64)(i * ctl->unit) + bitmap_info->offset;
1760 *bytes = (u64)(found_bits) * ctl->unit;
Josef Bacik96303082009-07-13 21:29:25 -04001761 return 0;
1762 }
1763
Miao Xiea4820392013-09-09 13:19:42 +08001764 *bytes = (u64)(max_bits) * ctl->unit;
Josef Bacik96303082009-07-13 21:29:25 -04001765 return -1;
1766}
1767
Miao Xiea4820392013-09-09 13:19:42 +08001768/* Cache the size of the max extent in bytes */
Li Zefan34d52cb2011-03-29 13:46:06 +08001769static struct btrfs_free_space *
David Woodhouse53b381b2013-01-29 18:40:14 -05001770find_free_space(struct btrfs_free_space_ctl *ctl, u64 *offset, u64 *bytes,
Miao Xiea4820392013-09-09 13:19:42 +08001771 unsigned long align, u64 *max_extent_size)
Josef Bacik96303082009-07-13 21:29:25 -04001772{
1773 struct btrfs_free_space *entry;
1774 struct rb_node *node;
David Woodhouse53b381b2013-01-29 18:40:14 -05001775 u64 tmp;
1776 u64 align_off;
Josef Bacik96303082009-07-13 21:29:25 -04001777 int ret;
1778
Li Zefan34d52cb2011-03-29 13:46:06 +08001779 if (!ctl->free_space_offset.rb_node)
Miao Xiea4820392013-09-09 13:19:42 +08001780 goto out;
Josef Bacik96303082009-07-13 21:29:25 -04001781
Li Zefan34d52cb2011-03-29 13:46:06 +08001782 entry = tree_search_offset(ctl, offset_to_bitmap(ctl, *offset), 0, 1);
Josef Bacik96303082009-07-13 21:29:25 -04001783 if (!entry)
Miao Xiea4820392013-09-09 13:19:42 +08001784 goto out;
Josef Bacik96303082009-07-13 21:29:25 -04001785
1786 for (node = &entry->offset_index; node; node = rb_next(node)) {
1787 entry = rb_entry(node, struct btrfs_free_space, offset_index);
Miao Xiea4820392013-09-09 13:19:42 +08001788 if (entry->bytes < *bytes) {
1789 if (entry->bytes > *max_extent_size)
1790 *max_extent_size = entry->bytes;
Josef Bacik96303082009-07-13 21:29:25 -04001791 continue;
Miao Xiea4820392013-09-09 13:19:42 +08001792 }
Josef Bacik96303082009-07-13 21:29:25 -04001793
David Woodhouse53b381b2013-01-29 18:40:14 -05001794 /* make sure the space returned is big enough
1795 * to match our requested alignment
1796 */
1797 if (*bytes >= align) {
Miao Xiea4820392013-09-09 13:19:42 +08001798 tmp = entry->offset - ctl->start + align - 1;
David Sterba47c57132015-02-20 18:43:47 +01001799 tmp = div64_u64(tmp, align);
David Woodhouse53b381b2013-01-29 18:40:14 -05001800 tmp = tmp * align + ctl->start;
1801 align_off = tmp - entry->offset;
1802 } else {
1803 align_off = 0;
1804 tmp = entry->offset;
1805 }
1806
Miao Xiea4820392013-09-09 13:19:42 +08001807 if (entry->bytes < *bytes + align_off) {
1808 if (entry->bytes > *max_extent_size)
1809 *max_extent_size = entry->bytes;
David Woodhouse53b381b2013-01-29 18:40:14 -05001810 continue;
Miao Xiea4820392013-09-09 13:19:42 +08001811 }
David Woodhouse53b381b2013-01-29 18:40:14 -05001812
Josef Bacik96303082009-07-13 21:29:25 -04001813 if (entry->bitmap) {
Miao Xiea4820392013-09-09 13:19:42 +08001814 u64 size = *bytes;
1815
1816 ret = search_bitmap(ctl, entry, &tmp, &size);
David Woodhouse53b381b2013-01-29 18:40:14 -05001817 if (!ret) {
1818 *offset = tmp;
Miao Xiea4820392013-09-09 13:19:42 +08001819 *bytes = size;
Josef Bacik96303082009-07-13 21:29:25 -04001820 return entry;
Miao Xiea4820392013-09-09 13:19:42 +08001821 } else if (size > *max_extent_size) {
1822 *max_extent_size = size;
David Woodhouse53b381b2013-01-29 18:40:14 -05001823 }
Josef Bacik96303082009-07-13 21:29:25 -04001824 continue;
1825 }
1826
David Woodhouse53b381b2013-01-29 18:40:14 -05001827 *offset = tmp;
1828 *bytes = entry->bytes - align_off;
Josef Bacik96303082009-07-13 21:29:25 -04001829 return entry;
1830 }
Miao Xiea4820392013-09-09 13:19:42 +08001831out:
Josef Bacik96303082009-07-13 21:29:25 -04001832 return NULL;
1833}
1834
Li Zefan34d52cb2011-03-29 13:46:06 +08001835static void add_new_bitmap(struct btrfs_free_space_ctl *ctl,
Josef Bacik96303082009-07-13 21:29:25 -04001836 struct btrfs_free_space *info, u64 offset)
1837{
Li Zefan34d52cb2011-03-29 13:46:06 +08001838 info->offset = offset_to_bitmap(ctl, offset);
Josef Bacikf019f422009-09-11 16:11:20 -04001839 info->bytes = 0;
Alexandre Olivaf2d0f672011-11-28 12:04:43 -02001840 INIT_LIST_HEAD(&info->list);
Li Zefan34d52cb2011-03-29 13:46:06 +08001841 link_free_space(ctl, info);
1842 ctl->total_bitmaps++;
Josef Bacik96303082009-07-13 21:29:25 -04001843
Li Zefan34d52cb2011-03-29 13:46:06 +08001844 ctl->op->recalc_thresholds(ctl);
Josef Bacik96303082009-07-13 21:29:25 -04001845}
1846
Li Zefan34d52cb2011-03-29 13:46:06 +08001847static void free_bitmap(struct btrfs_free_space_ctl *ctl,
Li Zefanedf6e2d2010-11-09 14:50:07 +08001848 struct btrfs_free_space *bitmap_info)
1849{
Li Zefan34d52cb2011-03-29 13:46:06 +08001850 unlink_free_space(ctl, bitmap_info);
Li Zefanedf6e2d2010-11-09 14:50:07 +08001851 kfree(bitmap_info->bitmap);
Josef Bacikdc89e982011-01-28 17:05:48 -05001852 kmem_cache_free(btrfs_free_space_cachep, bitmap_info);
Li Zefan34d52cb2011-03-29 13:46:06 +08001853 ctl->total_bitmaps--;
1854 ctl->op->recalc_thresholds(ctl);
Li Zefanedf6e2d2010-11-09 14:50:07 +08001855}
1856
Li Zefan34d52cb2011-03-29 13:46:06 +08001857static noinline int remove_from_bitmap(struct btrfs_free_space_ctl *ctl,
Josef Bacik96303082009-07-13 21:29:25 -04001858 struct btrfs_free_space *bitmap_info,
1859 u64 *offset, u64 *bytes)
1860{
1861 u64 end;
Josef Bacik6606bb92009-07-31 11:03:58 -04001862 u64 search_start, search_bytes;
1863 int ret;
Josef Bacik96303082009-07-13 21:29:25 -04001864
1865again:
Li Zefan34d52cb2011-03-29 13:46:06 +08001866 end = bitmap_info->offset + (u64)(BITS_PER_BITMAP * ctl->unit) - 1;
Josef Bacik96303082009-07-13 21:29:25 -04001867
Josef Bacik6606bb92009-07-31 11:03:58 -04001868 /*
Josef Bacikbdb7d302012-06-27 15:10:56 -04001869 * We need to search for bits in this bitmap. We could only cover some
1870 * of the extent in this bitmap thanks to how we add space, so we need
1871 * to search for as much as it as we can and clear that amount, and then
1872 * go searching for the next bit.
Josef Bacik6606bb92009-07-31 11:03:58 -04001873 */
1874 search_start = *offset;
Josef Bacikbdb7d302012-06-27 15:10:56 -04001875 search_bytes = ctl->unit;
Josef Bacik13dbc082011-02-03 02:39:52 +00001876 search_bytes = min(search_bytes, end - search_start + 1);
Li Zefan34d52cb2011-03-29 13:46:06 +08001877 ret = search_bitmap(ctl, bitmap_info, &search_start, &search_bytes);
Josef Bacikb50c6e22013-04-25 15:55:30 -04001878 if (ret < 0 || search_start != *offset)
1879 return -EINVAL;
Josef Bacik6606bb92009-07-31 11:03:58 -04001880
Josef Bacikbdb7d302012-06-27 15:10:56 -04001881 /* We may have found more bits than what we need */
1882 search_bytes = min(search_bytes, *bytes);
1883
1884 /* Cannot clear past the end of the bitmap */
1885 search_bytes = min(search_bytes, end - search_start + 1);
1886
1887 bitmap_clear_bits(ctl, bitmap_info, search_start, search_bytes);
1888 *offset += search_bytes;
1889 *bytes -= search_bytes;
Josef Bacik96303082009-07-13 21:29:25 -04001890
1891 if (*bytes) {
Josef Bacik6606bb92009-07-31 11:03:58 -04001892 struct rb_node *next = rb_next(&bitmap_info->offset_index);
Li Zefanedf6e2d2010-11-09 14:50:07 +08001893 if (!bitmap_info->bytes)
Li Zefan34d52cb2011-03-29 13:46:06 +08001894 free_bitmap(ctl, bitmap_info);
Josef Bacik96303082009-07-13 21:29:25 -04001895
Josef Bacik6606bb92009-07-31 11:03:58 -04001896 /*
1897 * no entry after this bitmap, but we still have bytes to
1898 * remove, so something has gone wrong.
1899 */
1900 if (!next)
Josef Bacik96303082009-07-13 21:29:25 -04001901 return -EINVAL;
1902
Josef Bacik6606bb92009-07-31 11:03:58 -04001903 bitmap_info = rb_entry(next, struct btrfs_free_space,
1904 offset_index);
1905
1906 /*
1907 * if the next entry isn't a bitmap we need to return to let the
1908 * extent stuff do its work.
1909 */
Josef Bacik96303082009-07-13 21:29:25 -04001910 if (!bitmap_info->bitmap)
1911 return -EAGAIN;
1912
Josef Bacik6606bb92009-07-31 11:03:58 -04001913 /*
1914 * Ok the next item is a bitmap, but it may not actually hold
1915 * the information for the rest of this free space stuff, so
1916 * look for it, and if we don't find it return so we can try
1917 * everything over again.
1918 */
1919 search_start = *offset;
Josef Bacikbdb7d302012-06-27 15:10:56 -04001920 search_bytes = ctl->unit;
Li Zefan34d52cb2011-03-29 13:46:06 +08001921 ret = search_bitmap(ctl, bitmap_info, &search_start,
Josef Bacik6606bb92009-07-31 11:03:58 -04001922 &search_bytes);
1923 if (ret < 0 || search_start != *offset)
1924 return -EAGAIN;
1925
Josef Bacik96303082009-07-13 21:29:25 -04001926 goto again;
Li Zefanedf6e2d2010-11-09 14:50:07 +08001927 } else if (!bitmap_info->bytes)
Li Zefan34d52cb2011-03-29 13:46:06 +08001928 free_bitmap(ctl, bitmap_info);
Josef Bacik96303082009-07-13 21:29:25 -04001929
1930 return 0;
1931}
1932
Josef Bacik2cdc3422011-05-27 14:07:49 -04001933static u64 add_bytes_to_bitmap(struct btrfs_free_space_ctl *ctl,
1934 struct btrfs_free_space *info, u64 offset,
1935 u64 bytes)
1936{
1937 u64 bytes_to_set = 0;
1938 u64 end;
1939
1940 end = info->offset + (u64)(BITS_PER_BITMAP * ctl->unit);
1941
1942 bytes_to_set = min(end - offset, bytes);
1943
1944 bitmap_set_bits(ctl, info, offset, bytes_to_set);
1945
1946 return bytes_to_set;
1947
1948}
1949
Li Zefan34d52cb2011-03-29 13:46:06 +08001950static bool use_bitmap(struct btrfs_free_space_ctl *ctl,
1951 struct btrfs_free_space *info)
Josef Bacik96303082009-07-13 21:29:25 -04001952{
Li Zefan34d52cb2011-03-29 13:46:06 +08001953 struct btrfs_block_group_cache *block_group = ctl->private;
Josef Bacik96303082009-07-13 21:29:25 -04001954
1955 /*
1956 * If we are below the extents threshold then we can add this as an
1957 * extent, and don't have to deal with the bitmap
1958 */
Li Zefan34d52cb2011-03-29 13:46:06 +08001959 if (ctl->free_extents < ctl->extents_thresh) {
Josef Bacik32cb0842011-03-18 16:16:21 -04001960 /*
1961 * If this block group has some small extents we don't want to
1962 * use up all of our free slots in the cache with them, we want
1963 * to reserve them to larger extents, however if we have plent
1964 * of cache left then go ahead an dadd them, no sense in adding
1965 * the overhead of a bitmap if we don't have to.
1966 */
1967 if (info->bytes <= block_group->sectorsize * 4) {
Li Zefan34d52cb2011-03-29 13:46:06 +08001968 if (ctl->free_extents * 2 <= ctl->extents_thresh)
1969 return false;
Josef Bacik32cb0842011-03-18 16:16:21 -04001970 } else {
Li Zefan34d52cb2011-03-29 13:46:06 +08001971 return false;
Josef Bacik32cb0842011-03-18 16:16:21 -04001972 }
1973 }
Josef Bacik96303082009-07-13 21:29:25 -04001974
1975 /*
Josef Bacikdde57402013-02-12 14:07:51 -05001976 * The original block groups from mkfs can be really small, like 8
1977 * megabytes, so don't bother with a bitmap for those entries. However
1978 * some block groups can be smaller than what a bitmap would cover but
1979 * are still large enough that they could overflow the 32k memory limit,
1980 * so allow those block groups to still be allowed to have a bitmap
1981 * entry.
Josef Bacik96303082009-07-13 21:29:25 -04001982 */
Josef Bacikdde57402013-02-12 14:07:51 -05001983 if (((BITS_PER_BITMAP * ctl->unit) >> 1) > block_group->key.offset)
Li Zefan34d52cb2011-03-29 13:46:06 +08001984 return false;
1985
1986 return true;
1987}
1988
Josef Bacik2cdc3422011-05-27 14:07:49 -04001989static struct btrfs_free_space_op free_space_op = {
1990 .recalc_thresholds = recalculate_thresholds,
1991 .use_bitmap = use_bitmap,
1992};
1993
Li Zefan34d52cb2011-03-29 13:46:06 +08001994static int insert_into_bitmap(struct btrfs_free_space_ctl *ctl,
1995 struct btrfs_free_space *info)
1996{
1997 struct btrfs_free_space *bitmap_info;
Josef Bacik2cdc3422011-05-27 14:07:49 -04001998 struct btrfs_block_group_cache *block_group = NULL;
Li Zefan34d52cb2011-03-29 13:46:06 +08001999 int added = 0;
Josef Bacik2cdc3422011-05-27 14:07:49 -04002000 u64 bytes, offset, bytes_added;
Li Zefan34d52cb2011-03-29 13:46:06 +08002001 int ret;
Josef Bacik96303082009-07-13 21:29:25 -04002002
2003 bytes = info->bytes;
2004 offset = info->offset;
2005
Li Zefan34d52cb2011-03-29 13:46:06 +08002006 if (!ctl->op->use_bitmap(ctl, info))
2007 return 0;
2008
Josef Bacik2cdc3422011-05-27 14:07:49 -04002009 if (ctl->op == &free_space_op)
2010 block_group = ctl->private;
Chris Mason38e87882011-06-10 16:36:57 -04002011again:
Josef Bacik2cdc3422011-05-27 14:07:49 -04002012 /*
2013 * Since we link bitmaps right into the cluster we need to see if we
2014 * have a cluster here, and if so and it has our bitmap we need to add
2015 * the free space to that bitmap.
2016 */
2017 if (block_group && !list_empty(&block_group->cluster_list)) {
2018 struct btrfs_free_cluster *cluster;
2019 struct rb_node *node;
2020 struct btrfs_free_space *entry;
2021
2022 cluster = list_entry(block_group->cluster_list.next,
2023 struct btrfs_free_cluster,
2024 block_group_list);
2025 spin_lock(&cluster->lock);
2026 node = rb_first(&cluster->root);
2027 if (!node) {
2028 spin_unlock(&cluster->lock);
Chris Mason38e87882011-06-10 16:36:57 -04002029 goto no_cluster_bitmap;
Josef Bacik2cdc3422011-05-27 14:07:49 -04002030 }
2031
2032 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2033 if (!entry->bitmap) {
2034 spin_unlock(&cluster->lock);
Chris Mason38e87882011-06-10 16:36:57 -04002035 goto no_cluster_bitmap;
Josef Bacik2cdc3422011-05-27 14:07:49 -04002036 }
2037
2038 if (entry->offset == offset_to_bitmap(ctl, offset)) {
2039 bytes_added = add_bytes_to_bitmap(ctl, entry,
2040 offset, bytes);
2041 bytes -= bytes_added;
2042 offset += bytes_added;
2043 }
2044 spin_unlock(&cluster->lock);
2045 if (!bytes) {
2046 ret = 1;
2047 goto out;
2048 }
2049 }
Chris Mason38e87882011-06-10 16:36:57 -04002050
2051no_cluster_bitmap:
Li Zefan34d52cb2011-03-29 13:46:06 +08002052 bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
Josef Bacik96303082009-07-13 21:29:25 -04002053 1, 0);
2054 if (!bitmap_info) {
Josef Bacikb12d6862013-08-26 17:14:08 -04002055 ASSERT(added == 0);
Josef Bacik96303082009-07-13 21:29:25 -04002056 goto new_bitmap;
2057 }
2058
Josef Bacik2cdc3422011-05-27 14:07:49 -04002059 bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes);
2060 bytes -= bytes_added;
2061 offset += bytes_added;
2062 added = 0;
Josef Bacik96303082009-07-13 21:29:25 -04002063
2064 if (!bytes) {
2065 ret = 1;
2066 goto out;
2067 } else
2068 goto again;
2069
2070new_bitmap:
2071 if (info && info->bitmap) {
Li Zefan34d52cb2011-03-29 13:46:06 +08002072 add_new_bitmap(ctl, info, offset);
Josef Bacik96303082009-07-13 21:29:25 -04002073 added = 1;
2074 info = NULL;
2075 goto again;
2076 } else {
Li Zefan34d52cb2011-03-29 13:46:06 +08002077 spin_unlock(&ctl->tree_lock);
Josef Bacik96303082009-07-13 21:29:25 -04002078
2079 /* no pre-allocated info, allocate a new one */
2080 if (!info) {
Josef Bacikdc89e982011-01-28 17:05:48 -05002081 info = kmem_cache_zalloc(btrfs_free_space_cachep,
2082 GFP_NOFS);
Josef Bacik96303082009-07-13 21:29:25 -04002083 if (!info) {
Li Zefan34d52cb2011-03-29 13:46:06 +08002084 spin_lock(&ctl->tree_lock);
Josef Bacik96303082009-07-13 21:29:25 -04002085 ret = -ENOMEM;
2086 goto out;
2087 }
2088 }
2089
2090 /* allocate the bitmap */
2091 info->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
Li Zefan34d52cb2011-03-29 13:46:06 +08002092 spin_lock(&ctl->tree_lock);
Josef Bacik96303082009-07-13 21:29:25 -04002093 if (!info->bitmap) {
2094 ret = -ENOMEM;
2095 goto out;
2096 }
2097 goto again;
2098 }
2099
2100out:
2101 if (info) {
2102 if (info->bitmap)
2103 kfree(info->bitmap);
Josef Bacikdc89e982011-01-28 17:05:48 -05002104 kmem_cache_free(btrfs_free_space_cachep, info);
Josef Bacik96303082009-07-13 21:29:25 -04002105 }
Josef Bacik0f9dd462008-09-23 13:14:11 -04002106
2107 return ret;
2108}
2109
Chris Mason945d8962011-05-22 12:33:42 -04002110static bool try_merge_free_space(struct btrfs_free_space_ctl *ctl,
Li Zefanf333adb2010-11-09 14:57:39 +08002111 struct btrfs_free_space *info, bool update_stat)
Josef Bacik0f9dd462008-09-23 13:14:11 -04002112{
Li Zefan120d66e2010-11-09 14:56:50 +08002113 struct btrfs_free_space *left_info;
2114 struct btrfs_free_space *right_info;
2115 bool merged = false;
2116 u64 offset = info->offset;
2117 u64 bytes = info->bytes;
Josef Bacik6226cb0a2009-04-03 10:14:18 -04002118
Josef Bacik0f9dd462008-09-23 13:14:11 -04002119 /*
2120 * first we want to see if there is free space adjacent to the range we
2121 * are adding, if there is remove that struct and add a new one to
2122 * cover the entire range
2123 */
Li Zefan34d52cb2011-03-29 13:46:06 +08002124 right_info = tree_search_offset(ctl, offset + bytes, 0, 0);
Josef Bacik96303082009-07-13 21:29:25 -04002125 if (right_info && rb_prev(&right_info->offset_index))
2126 left_info = rb_entry(rb_prev(&right_info->offset_index),
2127 struct btrfs_free_space, offset_index);
2128 else
Li Zefan34d52cb2011-03-29 13:46:06 +08002129 left_info = tree_search_offset(ctl, offset - 1, 0, 0);
Josef Bacik0f9dd462008-09-23 13:14:11 -04002130
Josef Bacik96303082009-07-13 21:29:25 -04002131 if (right_info && !right_info->bitmap) {
Li Zefanf333adb2010-11-09 14:57:39 +08002132 if (update_stat)
Li Zefan34d52cb2011-03-29 13:46:06 +08002133 unlink_free_space(ctl, right_info);
Li Zefanf333adb2010-11-09 14:57:39 +08002134 else
Li Zefan34d52cb2011-03-29 13:46:06 +08002135 __unlink_free_space(ctl, right_info);
Josef Bacik6226cb0a2009-04-03 10:14:18 -04002136 info->bytes += right_info->bytes;
Josef Bacikdc89e982011-01-28 17:05:48 -05002137 kmem_cache_free(btrfs_free_space_cachep, right_info);
Li Zefan120d66e2010-11-09 14:56:50 +08002138 merged = true;
Josef Bacik0f9dd462008-09-23 13:14:11 -04002139 }
2140
Josef Bacik96303082009-07-13 21:29:25 -04002141 if (left_info && !left_info->bitmap &&
2142 left_info->offset + left_info->bytes == offset) {
Li Zefanf333adb2010-11-09 14:57:39 +08002143 if (update_stat)
Li Zefan34d52cb2011-03-29 13:46:06 +08002144 unlink_free_space(ctl, left_info);
Li Zefanf333adb2010-11-09 14:57:39 +08002145 else
Li Zefan34d52cb2011-03-29 13:46:06 +08002146 __unlink_free_space(ctl, left_info);
Josef Bacik6226cb0a2009-04-03 10:14:18 -04002147 info->offset = left_info->offset;
2148 info->bytes += left_info->bytes;
Josef Bacikdc89e982011-01-28 17:05:48 -05002149 kmem_cache_free(btrfs_free_space_cachep, left_info);
Li Zefan120d66e2010-11-09 14:56:50 +08002150 merged = true;
Josef Bacik0f9dd462008-09-23 13:14:11 -04002151 }
2152
Li Zefan120d66e2010-11-09 14:56:50 +08002153 return merged;
2154}
2155
Filipe Manana20005522014-08-29 13:35:13 +01002156static bool steal_from_bitmap_to_end(struct btrfs_free_space_ctl *ctl,
2157 struct btrfs_free_space *info,
2158 bool update_stat)
2159{
2160 struct btrfs_free_space *bitmap;
2161 unsigned long i;
2162 unsigned long j;
2163 const u64 end = info->offset + info->bytes;
2164 const u64 bitmap_offset = offset_to_bitmap(ctl, end);
2165 u64 bytes;
2166
2167 bitmap = tree_search_offset(ctl, bitmap_offset, 1, 0);
2168 if (!bitmap)
2169 return false;
2170
2171 i = offset_to_bit(bitmap->offset, ctl->unit, end);
2172 j = find_next_zero_bit(bitmap->bitmap, BITS_PER_BITMAP, i);
2173 if (j == i)
2174 return false;
2175 bytes = (j - i) * ctl->unit;
2176 info->bytes += bytes;
2177
2178 if (update_stat)
2179 bitmap_clear_bits(ctl, bitmap, end, bytes);
2180 else
2181 __bitmap_clear_bits(ctl, bitmap, end, bytes);
2182
2183 if (!bitmap->bytes)
2184 free_bitmap(ctl, bitmap);
2185
2186 return true;
2187}
2188
2189static bool steal_from_bitmap_to_front(struct btrfs_free_space_ctl *ctl,
2190 struct btrfs_free_space *info,
2191 bool update_stat)
2192{
2193 struct btrfs_free_space *bitmap;
2194 u64 bitmap_offset;
2195 unsigned long i;
2196 unsigned long j;
2197 unsigned long prev_j;
2198 u64 bytes;
2199
2200 bitmap_offset = offset_to_bitmap(ctl, info->offset);
2201 /* If we're on a boundary, try the previous logical bitmap. */
2202 if (bitmap_offset == info->offset) {
2203 if (info->offset == 0)
2204 return false;
2205 bitmap_offset = offset_to_bitmap(ctl, info->offset - 1);
2206 }
2207
2208 bitmap = tree_search_offset(ctl, bitmap_offset, 1, 0);
2209 if (!bitmap)
2210 return false;
2211
2212 i = offset_to_bit(bitmap->offset, ctl->unit, info->offset) - 1;
2213 j = 0;
2214 prev_j = (unsigned long)-1;
2215 for_each_clear_bit_from(j, bitmap->bitmap, BITS_PER_BITMAP) {
2216 if (j > i)
2217 break;
2218 prev_j = j;
2219 }
2220 if (prev_j == i)
2221 return false;
2222
2223 if (prev_j == (unsigned long)-1)
2224 bytes = (i + 1) * ctl->unit;
2225 else
2226 bytes = (i - prev_j) * ctl->unit;
2227
2228 info->offset -= bytes;
2229 info->bytes += bytes;
2230
2231 if (update_stat)
2232 bitmap_clear_bits(ctl, bitmap, info->offset, bytes);
2233 else
2234 __bitmap_clear_bits(ctl, bitmap, info->offset, bytes);
2235
2236 if (!bitmap->bytes)
2237 free_bitmap(ctl, bitmap);
2238
2239 return true;
2240}
2241
2242/*
2243 * We prefer always to allocate from extent entries, both for clustered and
2244 * non-clustered allocation requests. So when attempting to add a new extent
2245 * entry, try to see if there's adjacent free space in bitmap entries, and if
2246 * there is, migrate that space from the bitmaps to the extent.
2247 * Like this we get better chances of satisfying space allocation requests
2248 * because we attempt to satisfy them based on a single cache entry, and never
2249 * on 2 or more entries - even if the entries represent a contiguous free space
2250 * region (e.g. 1 extent entry + 1 bitmap entry starting where the extent entry
2251 * ends).
2252 */
2253static void steal_from_bitmap(struct btrfs_free_space_ctl *ctl,
2254 struct btrfs_free_space *info,
2255 bool update_stat)
2256{
2257 /*
2258 * Only work with disconnected entries, as we can change their offset,
2259 * and must be extent entries.
2260 */
2261 ASSERT(!info->bitmap);
2262 ASSERT(RB_EMPTY_NODE(&info->offset_index));
2263
2264 if (ctl->total_bitmaps > 0) {
2265 bool stole_end;
2266 bool stole_front = false;
2267
2268 stole_end = steal_from_bitmap_to_end(ctl, info, update_stat);
2269 if (ctl->total_bitmaps > 0)
2270 stole_front = steal_from_bitmap_to_front(ctl, info,
2271 update_stat);
2272
2273 if (stole_end || stole_front)
2274 try_merge_free_space(ctl, info, update_stat);
2275 }
2276}
2277
Li Zefan581bb052011-04-20 10:06:11 +08002278int __btrfs_add_free_space(struct btrfs_free_space_ctl *ctl,
2279 u64 offset, u64 bytes)
Li Zefan120d66e2010-11-09 14:56:50 +08002280{
2281 struct btrfs_free_space *info;
2282 int ret = 0;
2283
Josef Bacikdc89e982011-01-28 17:05:48 -05002284 info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
Li Zefan120d66e2010-11-09 14:56:50 +08002285 if (!info)
2286 return -ENOMEM;
2287
2288 info->offset = offset;
2289 info->bytes = bytes;
Filipe Manana20005522014-08-29 13:35:13 +01002290 RB_CLEAR_NODE(&info->offset_index);
Li Zefan120d66e2010-11-09 14:56:50 +08002291
Li Zefan34d52cb2011-03-29 13:46:06 +08002292 spin_lock(&ctl->tree_lock);
Li Zefan120d66e2010-11-09 14:56:50 +08002293
Li Zefan34d52cb2011-03-29 13:46:06 +08002294 if (try_merge_free_space(ctl, info, true))
Li Zefan120d66e2010-11-09 14:56:50 +08002295 goto link;
2296
2297 /*
2298 * There was no extent directly to the left or right of this new
2299 * extent then we know we're going to have to allocate a new extent, so
2300 * before we do that see if we need to drop this into a bitmap
2301 */
Li Zefan34d52cb2011-03-29 13:46:06 +08002302 ret = insert_into_bitmap(ctl, info);
Li Zefan120d66e2010-11-09 14:56:50 +08002303 if (ret < 0) {
2304 goto out;
2305 } else if (ret) {
2306 ret = 0;
2307 goto out;
2308 }
2309link:
Filipe Manana20005522014-08-29 13:35:13 +01002310 /*
2311 * Only steal free space from adjacent bitmaps if we're sure we're not
2312 * going to add the new free space to existing bitmap entries - because
2313 * that would mean unnecessary work that would be reverted. Therefore
2314 * attempt to steal space from bitmaps if we're adding an extent entry.
2315 */
2316 steal_from_bitmap(ctl, info, true);
2317
Li Zefan34d52cb2011-03-29 13:46:06 +08002318 ret = link_free_space(ctl, info);
Josef Bacik0f9dd462008-09-23 13:14:11 -04002319 if (ret)
Josef Bacikdc89e982011-01-28 17:05:48 -05002320 kmem_cache_free(btrfs_free_space_cachep, info);
Josef Bacik96303082009-07-13 21:29:25 -04002321out:
Li Zefan34d52cb2011-03-29 13:46:06 +08002322 spin_unlock(&ctl->tree_lock);
Josef Bacik6226cb0a2009-04-03 10:14:18 -04002323
Josef Bacik0f9dd462008-09-23 13:14:11 -04002324 if (ret) {
Frank Holtonefe120a2013-12-20 11:37:06 -05002325 printk(KERN_CRIT "BTRFS: unable to add free space :%d\n", ret);
Josef Bacikb12d6862013-08-26 17:14:08 -04002326 ASSERT(ret != -EEXIST);
Josef Bacik0f9dd462008-09-23 13:14:11 -04002327 }
2328
Josef Bacik0f9dd462008-09-23 13:14:11 -04002329 return ret;
2330}
2331
Josef Bacik6226cb0a2009-04-03 10:14:18 -04002332int btrfs_remove_free_space(struct btrfs_block_group_cache *block_group,
2333 u64 offset, u64 bytes)
Josef Bacik0f9dd462008-09-23 13:14:11 -04002334{
Li Zefan34d52cb2011-03-29 13:46:06 +08002335 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
Josef Bacik0f9dd462008-09-23 13:14:11 -04002336 struct btrfs_free_space *info;
Josef Bacikb0175112012-12-18 11:39:19 -05002337 int ret;
2338 bool re_search = false;
Josef Bacik0f9dd462008-09-23 13:14:11 -04002339
Li Zefan34d52cb2011-03-29 13:46:06 +08002340 spin_lock(&ctl->tree_lock);
Josef Bacik6226cb0a2009-04-03 10:14:18 -04002341
Josef Bacik96303082009-07-13 21:29:25 -04002342again:
Josef Bacikb0175112012-12-18 11:39:19 -05002343 ret = 0;
Josef Bacikbdb7d302012-06-27 15:10:56 -04002344 if (!bytes)
2345 goto out_lock;
2346
Li Zefan34d52cb2011-03-29 13:46:06 +08002347 info = tree_search_offset(ctl, offset, 0, 0);
Josef Bacik96303082009-07-13 21:29:25 -04002348 if (!info) {
Josef Bacik6606bb92009-07-31 11:03:58 -04002349 /*
2350 * oops didn't find an extent that matched the space we wanted
2351 * to remove, look for a bitmap instead
2352 */
Li Zefan34d52cb2011-03-29 13:46:06 +08002353 info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
Josef Bacik6606bb92009-07-31 11:03:58 -04002354 1, 0);
2355 if (!info) {
Josef Bacikb0175112012-12-18 11:39:19 -05002356 /*
2357 * If we found a partial bit of our free space in a
2358 * bitmap but then couldn't find the other part this may
2359 * be a problem, so WARN about it.
Chris Mason24a70312011-11-21 09:39:11 -05002360 */
Josef Bacikb0175112012-12-18 11:39:19 -05002361 WARN_ON(re_search);
Josef Bacik6606bb92009-07-31 11:03:58 -04002362 goto out_lock;
2363 }
Josef Bacik96303082009-07-13 21:29:25 -04002364 }
2365
Josef Bacikb0175112012-12-18 11:39:19 -05002366 re_search = false;
Josef Bacikbdb7d302012-06-27 15:10:56 -04002367 if (!info->bitmap) {
Li Zefan34d52cb2011-03-29 13:46:06 +08002368 unlink_free_space(ctl, info);
Josef Bacikbdb7d302012-06-27 15:10:56 -04002369 if (offset == info->offset) {
2370 u64 to_free = min(bytes, info->bytes);
Josef Bacik0f9dd462008-09-23 13:14:11 -04002371
Josef Bacikbdb7d302012-06-27 15:10:56 -04002372 info->bytes -= to_free;
2373 info->offset += to_free;
2374 if (info->bytes) {
2375 ret = link_free_space(ctl, info);
2376 WARN_ON(ret);
2377 } else {
2378 kmem_cache_free(btrfs_free_space_cachep, info);
2379 }
Josef Bacik0f9dd462008-09-23 13:14:11 -04002380
Josef Bacikbdb7d302012-06-27 15:10:56 -04002381 offset += to_free;
2382 bytes -= to_free;
2383 goto again;
2384 } else {
2385 u64 old_end = info->bytes + info->offset;
Chris Mason9b49c9b2008-09-24 11:23:25 -04002386
Josef Bacikbdb7d302012-06-27 15:10:56 -04002387 info->bytes = offset - info->offset;
Li Zefan34d52cb2011-03-29 13:46:06 +08002388 ret = link_free_space(ctl, info);
Josef Bacik96303082009-07-13 21:29:25 -04002389 WARN_ON(ret);
2390 if (ret)
2391 goto out_lock;
Josef Bacik96303082009-07-13 21:29:25 -04002392
Josef Bacikbdb7d302012-06-27 15:10:56 -04002393 /* Not enough bytes in this entry to satisfy us */
2394 if (old_end < offset + bytes) {
2395 bytes -= old_end - offset;
2396 offset = old_end;
2397 goto again;
2398 } else if (old_end == offset + bytes) {
2399 /* all done */
2400 goto out_lock;
2401 }
2402 spin_unlock(&ctl->tree_lock);
2403
2404 ret = btrfs_add_free_space(block_group, offset + bytes,
2405 old_end - (offset + bytes));
2406 WARN_ON(ret);
2407 goto out;
2408 }
Josef Bacik0f9dd462008-09-23 13:14:11 -04002409 }
Josef Bacik96303082009-07-13 21:29:25 -04002410
Li Zefan34d52cb2011-03-29 13:46:06 +08002411 ret = remove_from_bitmap(ctl, info, &offset, &bytes);
Josef Bacikb0175112012-12-18 11:39:19 -05002412 if (ret == -EAGAIN) {
2413 re_search = true;
Josef Bacik96303082009-07-13 21:29:25 -04002414 goto again;
Josef Bacikb0175112012-12-18 11:39:19 -05002415 }
Josef Bacik96303082009-07-13 21:29:25 -04002416out_lock:
Li Zefan34d52cb2011-03-29 13:46:06 +08002417 spin_unlock(&ctl->tree_lock);
Josef Bacik0f9dd462008-09-23 13:14:11 -04002418out:
Josef Bacik25179202008-10-29 14:49:05 -04002419 return ret;
2420}
2421
Josef Bacik0f9dd462008-09-23 13:14:11 -04002422void btrfs_dump_free_space(struct btrfs_block_group_cache *block_group,
2423 u64 bytes)
2424{
Li Zefan34d52cb2011-03-29 13:46:06 +08002425 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
Josef Bacik0f9dd462008-09-23 13:14:11 -04002426 struct btrfs_free_space *info;
2427 struct rb_node *n;
2428 int count = 0;
2429
Li Zefan34d52cb2011-03-29 13:46:06 +08002430 for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) {
Josef Bacik0f9dd462008-09-23 13:14:11 -04002431 info = rb_entry(n, struct btrfs_free_space, offset_index);
Liu Bof6175ef2012-07-06 03:31:36 -06002432 if (info->bytes >= bytes && !block_group->ro)
Josef Bacik0f9dd462008-09-23 13:14:11 -04002433 count++;
Frank Holtonefe120a2013-12-20 11:37:06 -05002434 btrfs_crit(block_group->fs_info,
2435 "entry offset %llu, bytes %llu, bitmap %s",
2436 info->offset, info->bytes,
Josef Bacik96303082009-07-13 21:29:25 -04002437 (info->bitmap) ? "yes" : "no");
Josef Bacik0f9dd462008-09-23 13:14:11 -04002438 }
Frank Holtonefe120a2013-12-20 11:37:06 -05002439 btrfs_info(block_group->fs_info, "block group has cluster?: %s",
Josef Bacik96303082009-07-13 21:29:25 -04002440 list_empty(&block_group->cluster_list) ? "no" : "yes");
Frank Holtonefe120a2013-12-20 11:37:06 -05002441 btrfs_info(block_group->fs_info,
2442 "%d blocks of free space at or bigger than bytes is", count);
Josef Bacik0f9dd462008-09-23 13:14:11 -04002443}
2444
Li Zefan34d52cb2011-03-29 13:46:06 +08002445void btrfs_init_free_space_ctl(struct btrfs_block_group_cache *block_group)
Josef Bacik0f9dd462008-09-23 13:14:11 -04002446{
Li Zefan34d52cb2011-03-29 13:46:06 +08002447 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
Josef Bacik0f9dd462008-09-23 13:14:11 -04002448
Li Zefan34d52cb2011-03-29 13:46:06 +08002449 spin_lock_init(&ctl->tree_lock);
2450 ctl->unit = block_group->sectorsize;
2451 ctl->start = block_group->key.objectid;
2452 ctl->private = block_group;
2453 ctl->op = &free_space_op;
Filipe Manana55507ce2014-12-01 17:04:09 +00002454 INIT_LIST_HEAD(&ctl->trimming_ranges);
2455 mutex_init(&ctl->cache_writeout_mutex);
Josef Bacik0f9dd462008-09-23 13:14:11 -04002456
Li Zefan34d52cb2011-03-29 13:46:06 +08002457 /*
2458 * we only want to have 32k of ram per block group for keeping
2459 * track of free space, and if we pass 1/2 of that we want to
2460 * start converting things over to using bitmaps
2461 */
2462 ctl->extents_thresh = ((1024 * 32) / 2) /
2463 sizeof(struct btrfs_free_space);
Josef Bacik0f9dd462008-09-23 13:14:11 -04002464}
2465
Chris Masonfa9c0d792009-04-03 09:47:43 -04002466/*
2467 * for a given cluster, put all of its extents back into the free
2468 * space cache. If the block group passed doesn't match the block group
2469 * pointed to by the cluster, someone else raced in and freed the
2470 * cluster already. In that case, we just return without changing anything
2471 */
2472static int
2473__btrfs_return_cluster_to_free_space(
2474 struct btrfs_block_group_cache *block_group,
2475 struct btrfs_free_cluster *cluster)
2476{
Li Zefan34d52cb2011-03-29 13:46:06 +08002477 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
Chris Masonfa9c0d792009-04-03 09:47:43 -04002478 struct btrfs_free_space *entry;
2479 struct rb_node *node;
2480
2481 spin_lock(&cluster->lock);
2482 if (cluster->block_group != block_group)
2483 goto out;
2484
Josef Bacik96303082009-07-13 21:29:25 -04002485 cluster->block_group = NULL;
Chris Masonfa9c0d792009-04-03 09:47:43 -04002486 cluster->window_start = 0;
Josef Bacik96303082009-07-13 21:29:25 -04002487 list_del_init(&cluster->block_group_list);
Josef Bacik96303082009-07-13 21:29:25 -04002488
Chris Masonfa9c0d792009-04-03 09:47:43 -04002489 node = rb_first(&cluster->root);
Josef Bacik96303082009-07-13 21:29:25 -04002490 while (node) {
Josef Bacik4e69b592011-03-21 10:11:24 -04002491 bool bitmap;
2492
Chris Masonfa9c0d792009-04-03 09:47:43 -04002493 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2494 node = rb_next(&entry->offset_index);
2495 rb_erase(&entry->offset_index, &cluster->root);
Filipe Manana20005522014-08-29 13:35:13 +01002496 RB_CLEAR_NODE(&entry->offset_index);
Josef Bacik4e69b592011-03-21 10:11:24 -04002497
2498 bitmap = (entry->bitmap != NULL);
Filipe Manana20005522014-08-29 13:35:13 +01002499 if (!bitmap) {
Li Zefan34d52cb2011-03-29 13:46:06 +08002500 try_merge_free_space(ctl, entry, false);
Filipe Manana20005522014-08-29 13:35:13 +01002501 steal_from_bitmap(ctl, entry, false);
2502 }
Li Zefan34d52cb2011-03-29 13:46:06 +08002503 tree_insert_offset(&ctl->free_space_offset,
Josef Bacik4e69b592011-03-21 10:11:24 -04002504 entry->offset, &entry->offset_index, bitmap);
Chris Masonfa9c0d792009-04-03 09:47:43 -04002505 }
Eric Paris6bef4d32010-02-23 19:43:04 +00002506 cluster->root = RB_ROOT;
Josef Bacik96303082009-07-13 21:29:25 -04002507
Chris Masonfa9c0d792009-04-03 09:47:43 -04002508out:
2509 spin_unlock(&cluster->lock);
Josef Bacik96303082009-07-13 21:29:25 -04002510 btrfs_put_block_group(block_group);
Chris Masonfa9c0d792009-04-03 09:47:43 -04002511 return 0;
2512}
2513
Eric Sandeen48a3b632013-04-25 20:41:01 +00002514static void __btrfs_remove_free_space_cache_locked(
2515 struct btrfs_free_space_ctl *ctl)
Josef Bacik0f9dd462008-09-23 13:14:11 -04002516{
2517 struct btrfs_free_space *info;
2518 struct rb_node *node;
Li Zefan581bb052011-04-20 10:06:11 +08002519
Li Zefan581bb052011-04-20 10:06:11 +08002520 while ((node = rb_last(&ctl->free_space_offset)) != NULL) {
2521 info = rb_entry(node, struct btrfs_free_space, offset_index);
Josef Bacik9b90f512011-06-24 16:02:51 +00002522 if (!info->bitmap) {
2523 unlink_free_space(ctl, info);
2524 kmem_cache_free(btrfs_free_space_cachep, info);
2525 } else {
2526 free_bitmap(ctl, info);
2527 }
David Sterba351810c2015-01-08 15:20:54 +01002528
2529 cond_resched_lock(&ctl->tree_lock);
Li Zefan581bb052011-04-20 10:06:11 +08002530 }
Chris Mason09655372011-05-21 09:27:38 -04002531}
2532
2533void __btrfs_remove_free_space_cache(struct btrfs_free_space_ctl *ctl)
2534{
2535 spin_lock(&ctl->tree_lock);
2536 __btrfs_remove_free_space_cache_locked(ctl);
Li Zefan581bb052011-04-20 10:06:11 +08002537 spin_unlock(&ctl->tree_lock);
2538}
2539
Josef Bacik0f9dd462008-09-23 13:14:11 -04002540void btrfs_remove_free_space_cache(struct btrfs_block_group_cache *block_group)
2541{
Li Zefan34d52cb2011-03-29 13:46:06 +08002542 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
Chris Masonfa9c0d792009-04-03 09:47:43 -04002543 struct btrfs_free_cluster *cluster;
Josef Bacik96303082009-07-13 21:29:25 -04002544 struct list_head *head;
Josef Bacik0f9dd462008-09-23 13:14:11 -04002545
Li Zefan34d52cb2011-03-29 13:46:06 +08002546 spin_lock(&ctl->tree_lock);
Josef Bacik96303082009-07-13 21:29:25 -04002547 while ((head = block_group->cluster_list.next) !=
2548 &block_group->cluster_list) {
2549 cluster = list_entry(head, struct btrfs_free_cluster,
2550 block_group_list);
Chris Masonfa9c0d792009-04-03 09:47:43 -04002551
2552 WARN_ON(cluster->block_group != block_group);
2553 __btrfs_return_cluster_to_free_space(block_group, cluster);
David Sterba351810c2015-01-08 15:20:54 +01002554
2555 cond_resched_lock(&ctl->tree_lock);
Chris Masonfa9c0d792009-04-03 09:47:43 -04002556 }
Chris Mason09655372011-05-21 09:27:38 -04002557 __btrfs_remove_free_space_cache_locked(ctl);
Li Zefan34d52cb2011-03-29 13:46:06 +08002558 spin_unlock(&ctl->tree_lock);
Chris Masonfa9c0d792009-04-03 09:47:43 -04002559
Josef Bacik0f9dd462008-09-23 13:14:11 -04002560}
2561
Josef Bacik6226cb0a2009-04-03 10:14:18 -04002562u64 btrfs_find_space_for_alloc(struct btrfs_block_group_cache *block_group,
Miao Xiea4820392013-09-09 13:19:42 +08002563 u64 offset, u64 bytes, u64 empty_size,
2564 u64 *max_extent_size)
Josef Bacik0f9dd462008-09-23 13:14:11 -04002565{
Li Zefan34d52cb2011-03-29 13:46:06 +08002566 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
Josef Bacik6226cb0a2009-04-03 10:14:18 -04002567 struct btrfs_free_space *entry = NULL;
Josef Bacik96303082009-07-13 21:29:25 -04002568 u64 bytes_search = bytes + empty_size;
Josef Bacik6226cb0a2009-04-03 10:14:18 -04002569 u64 ret = 0;
David Woodhouse53b381b2013-01-29 18:40:14 -05002570 u64 align_gap = 0;
2571 u64 align_gap_len = 0;
Josef Bacik0f9dd462008-09-23 13:14:11 -04002572
Li Zefan34d52cb2011-03-29 13:46:06 +08002573 spin_lock(&ctl->tree_lock);
David Woodhouse53b381b2013-01-29 18:40:14 -05002574 entry = find_free_space(ctl, &offset, &bytes_search,
Miao Xiea4820392013-09-09 13:19:42 +08002575 block_group->full_stripe_len, max_extent_size);
Josef Bacik6226cb0a2009-04-03 10:14:18 -04002576 if (!entry)
Josef Bacik96303082009-07-13 21:29:25 -04002577 goto out;
2578
2579 ret = offset;
2580 if (entry->bitmap) {
Li Zefan34d52cb2011-03-29 13:46:06 +08002581 bitmap_clear_bits(ctl, entry, offset, bytes);
Li Zefanedf6e2d2010-11-09 14:50:07 +08002582 if (!entry->bytes)
Li Zefan34d52cb2011-03-29 13:46:06 +08002583 free_bitmap(ctl, entry);
Josef Bacik96303082009-07-13 21:29:25 -04002584 } else {
Li Zefan34d52cb2011-03-29 13:46:06 +08002585 unlink_free_space(ctl, entry);
David Woodhouse53b381b2013-01-29 18:40:14 -05002586 align_gap_len = offset - entry->offset;
2587 align_gap = entry->offset;
2588
2589 entry->offset = offset + bytes;
2590 WARN_ON(entry->bytes < bytes + align_gap_len);
2591
2592 entry->bytes -= bytes + align_gap_len;
Josef Bacik6226cb0a2009-04-03 10:14:18 -04002593 if (!entry->bytes)
Josef Bacikdc89e982011-01-28 17:05:48 -05002594 kmem_cache_free(btrfs_free_space_cachep, entry);
Josef Bacik6226cb0a2009-04-03 10:14:18 -04002595 else
Li Zefan34d52cb2011-03-29 13:46:06 +08002596 link_free_space(ctl, entry);
Josef Bacik6226cb0a2009-04-03 10:14:18 -04002597 }
Josef Bacik96303082009-07-13 21:29:25 -04002598out:
Li Zefan34d52cb2011-03-29 13:46:06 +08002599 spin_unlock(&ctl->tree_lock);
Josef Bacik817d52f2009-07-13 21:29:25 -04002600
David Woodhouse53b381b2013-01-29 18:40:14 -05002601 if (align_gap_len)
2602 __btrfs_add_free_space(ctl, align_gap, align_gap_len);
Josef Bacik0f9dd462008-09-23 13:14:11 -04002603 return ret;
2604}
Chris Masonfa9c0d792009-04-03 09:47:43 -04002605
2606/*
2607 * given a cluster, put all of its extents back into the free space
2608 * cache. If a block group is passed, this function will only free
2609 * a cluster that belongs to the passed block group.
2610 *
2611 * Otherwise, it'll get a reference on the block group pointed to by the
2612 * cluster and remove the cluster from it.
2613 */
2614int btrfs_return_cluster_to_free_space(
2615 struct btrfs_block_group_cache *block_group,
2616 struct btrfs_free_cluster *cluster)
2617{
Li Zefan34d52cb2011-03-29 13:46:06 +08002618 struct btrfs_free_space_ctl *ctl;
Chris Masonfa9c0d792009-04-03 09:47:43 -04002619 int ret;
2620
2621 /* first, get a safe pointer to the block group */
2622 spin_lock(&cluster->lock);
2623 if (!block_group) {
2624 block_group = cluster->block_group;
2625 if (!block_group) {
2626 spin_unlock(&cluster->lock);
2627 return 0;
2628 }
2629 } else if (cluster->block_group != block_group) {
2630 /* someone else has already freed it don't redo their work */
2631 spin_unlock(&cluster->lock);
2632 return 0;
2633 }
2634 atomic_inc(&block_group->count);
2635 spin_unlock(&cluster->lock);
2636
Li Zefan34d52cb2011-03-29 13:46:06 +08002637 ctl = block_group->free_space_ctl;
2638
Chris Masonfa9c0d792009-04-03 09:47:43 -04002639 /* now return any extents the cluster had on it */
Li Zefan34d52cb2011-03-29 13:46:06 +08002640 spin_lock(&ctl->tree_lock);
Chris Masonfa9c0d792009-04-03 09:47:43 -04002641 ret = __btrfs_return_cluster_to_free_space(block_group, cluster);
Li Zefan34d52cb2011-03-29 13:46:06 +08002642 spin_unlock(&ctl->tree_lock);
Chris Masonfa9c0d792009-04-03 09:47:43 -04002643
2644 /* finally drop our ref */
2645 btrfs_put_block_group(block_group);
2646 return ret;
2647}
2648
Josef Bacik96303082009-07-13 21:29:25 -04002649static u64 btrfs_alloc_from_bitmap(struct btrfs_block_group_cache *block_group,
2650 struct btrfs_free_cluster *cluster,
Josef Bacik4e69b592011-03-21 10:11:24 -04002651 struct btrfs_free_space *entry,
Miao Xiea4820392013-09-09 13:19:42 +08002652 u64 bytes, u64 min_start,
2653 u64 *max_extent_size)
Josef Bacik96303082009-07-13 21:29:25 -04002654{
Li Zefan34d52cb2011-03-29 13:46:06 +08002655 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
Josef Bacik96303082009-07-13 21:29:25 -04002656 int err;
2657 u64 search_start = cluster->window_start;
2658 u64 search_bytes = bytes;
2659 u64 ret = 0;
2660
Josef Bacik96303082009-07-13 21:29:25 -04002661 search_start = min_start;
2662 search_bytes = bytes;
2663
Li Zefan34d52cb2011-03-29 13:46:06 +08002664 err = search_bitmap(ctl, entry, &search_start, &search_bytes);
Miao Xiea4820392013-09-09 13:19:42 +08002665 if (err) {
2666 if (search_bytes > *max_extent_size)
2667 *max_extent_size = search_bytes;
Josef Bacik4e69b592011-03-21 10:11:24 -04002668 return 0;
Miao Xiea4820392013-09-09 13:19:42 +08002669 }
Josef Bacik96303082009-07-13 21:29:25 -04002670
2671 ret = search_start;
Miao Xiebb3ac5a2011-08-05 09:32:35 +00002672 __bitmap_clear_bits(ctl, entry, ret, bytes);
Josef Bacik96303082009-07-13 21:29:25 -04002673
2674 return ret;
2675}
2676
Chris Masonfa9c0d792009-04-03 09:47:43 -04002677/*
2678 * given a cluster, try to allocate 'bytes' from it, returns 0
2679 * if it couldn't find anything suitably large, or a logical disk offset
2680 * if things worked out
2681 */
2682u64 btrfs_alloc_from_cluster(struct btrfs_block_group_cache *block_group,
2683 struct btrfs_free_cluster *cluster, u64 bytes,
Miao Xiea4820392013-09-09 13:19:42 +08002684 u64 min_start, u64 *max_extent_size)
Chris Masonfa9c0d792009-04-03 09:47:43 -04002685{
Li Zefan34d52cb2011-03-29 13:46:06 +08002686 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
Chris Masonfa9c0d792009-04-03 09:47:43 -04002687 struct btrfs_free_space *entry = NULL;
2688 struct rb_node *node;
2689 u64 ret = 0;
2690
2691 spin_lock(&cluster->lock);
2692 if (bytes > cluster->max_size)
2693 goto out;
2694
2695 if (cluster->block_group != block_group)
2696 goto out;
2697
2698 node = rb_first(&cluster->root);
2699 if (!node)
2700 goto out;
2701
2702 entry = rb_entry(node, struct btrfs_free_space, offset_index);
Dulshani Gunawardhana67871252013-10-31 10:33:04 +05302703 while (1) {
Miao Xiea4820392013-09-09 13:19:42 +08002704 if (entry->bytes < bytes && entry->bytes > *max_extent_size)
2705 *max_extent_size = entry->bytes;
2706
Josef Bacik4e69b592011-03-21 10:11:24 -04002707 if (entry->bytes < bytes ||
2708 (!entry->bitmap && entry->offset < min_start)) {
Chris Masonfa9c0d792009-04-03 09:47:43 -04002709 node = rb_next(&entry->offset_index);
2710 if (!node)
2711 break;
2712 entry = rb_entry(node, struct btrfs_free_space,
2713 offset_index);
2714 continue;
2715 }
Chris Masonfa9c0d792009-04-03 09:47:43 -04002716
Josef Bacik4e69b592011-03-21 10:11:24 -04002717 if (entry->bitmap) {
2718 ret = btrfs_alloc_from_bitmap(block_group,
2719 cluster, entry, bytes,
Miao Xiea4820392013-09-09 13:19:42 +08002720 cluster->window_start,
2721 max_extent_size);
Josef Bacik4e69b592011-03-21 10:11:24 -04002722 if (ret == 0) {
Josef Bacik4e69b592011-03-21 10:11:24 -04002723 node = rb_next(&entry->offset_index);
2724 if (!node)
2725 break;
2726 entry = rb_entry(node, struct btrfs_free_space,
2727 offset_index);
2728 continue;
2729 }
Josef Bacik9b230622012-01-26 15:01:12 -05002730 cluster->window_start += bytes;
Josef Bacik4e69b592011-03-21 10:11:24 -04002731 } else {
Josef Bacik4e69b592011-03-21 10:11:24 -04002732 ret = entry->offset;
2733
2734 entry->offset += bytes;
2735 entry->bytes -= bytes;
2736 }
Chris Masonfa9c0d792009-04-03 09:47:43 -04002737
Li Zefan5e71b5d2010-11-09 14:55:34 +08002738 if (entry->bytes == 0)
Chris Masonfa9c0d792009-04-03 09:47:43 -04002739 rb_erase(&entry->offset_index, &cluster->root);
Chris Masonfa9c0d792009-04-03 09:47:43 -04002740 break;
2741 }
2742out:
2743 spin_unlock(&cluster->lock);
Josef Bacik96303082009-07-13 21:29:25 -04002744
Li Zefan5e71b5d2010-11-09 14:55:34 +08002745 if (!ret)
2746 return 0;
2747
Li Zefan34d52cb2011-03-29 13:46:06 +08002748 spin_lock(&ctl->tree_lock);
Li Zefan5e71b5d2010-11-09 14:55:34 +08002749
Li Zefan34d52cb2011-03-29 13:46:06 +08002750 ctl->free_space -= bytes;
Li Zefan5e71b5d2010-11-09 14:55:34 +08002751 if (entry->bytes == 0) {
Li Zefan34d52cb2011-03-29 13:46:06 +08002752 ctl->free_extents--;
Josef Bacik4e69b592011-03-21 10:11:24 -04002753 if (entry->bitmap) {
2754 kfree(entry->bitmap);
Li Zefan34d52cb2011-03-29 13:46:06 +08002755 ctl->total_bitmaps--;
2756 ctl->op->recalc_thresholds(ctl);
Josef Bacik4e69b592011-03-21 10:11:24 -04002757 }
Josef Bacikdc89e982011-01-28 17:05:48 -05002758 kmem_cache_free(btrfs_free_space_cachep, entry);
Li Zefan5e71b5d2010-11-09 14:55:34 +08002759 }
2760
Li Zefan34d52cb2011-03-29 13:46:06 +08002761 spin_unlock(&ctl->tree_lock);
Li Zefan5e71b5d2010-11-09 14:55:34 +08002762
Chris Masonfa9c0d792009-04-03 09:47:43 -04002763 return ret;
2764}
2765
Josef Bacik96303082009-07-13 21:29:25 -04002766static int btrfs_bitmap_cluster(struct btrfs_block_group_cache *block_group,
2767 struct btrfs_free_space *entry,
2768 struct btrfs_free_cluster *cluster,
Alexandre Oliva1bb91902011-10-14 12:10:36 -03002769 u64 offset, u64 bytes,
2770 u64 cont1_bytes, u64 min_bytes)
Josef Bacik96303082009-07-13 21:29:25 -04002771{
Li Zefan34d52cb2011-03-29 13:46:06 +08002772 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
Josef Bacik96303082009-07-13 21:29:25 -04002773 unsigned long next_zero;
2774 unsigned long i;
Alexandre Oliva1bb91902011-10-14 12:10:36 -03002775 unsigned long want_bits;
2776 unsigned long min_bits;
Josef Bacik96303082009-07-13 21:29:25 -04002777 unsigned long found_bits;
2778 unsigned long start = 0;
2779 unsigned long total_found = 0;
Josef Bacik4e69b592011-03-21 10:11:24 -04002780 int ret;
Josef Bacik96303082009-07-13 21:29:25 -04002781
Wang Sheng-Hui96009762012-11-30 06:30:14 +00002782 i = offset_to_bit(entry->offset, ctl->unit,
Josef Bacik96303082009-07-13 21:29:25 -04002783 max_t(u64, offset, entry->offset));
Wang Sheng-Hui96009762012-11-30 06:30:14 +00002784 want_bits = bytes_to_bits(bytes, ctl->unit);
2785 min_bits = bytes_to_bits(min_bytes, ctl->unit);
Josef Bacik96303082009-07-13 21:29:25 -04002786
2787again:
2788 found_bits = 0;
Wei Yongjunebb3dad2012-09-13 20:29:02 -06002789 for_each_set_bit_from(i, entry->bitmap, BITS_PER_BITMAP) {
Josef Bacik96303082009-07-13 21:29:25 -04002790 next_zero = find_next_zero_bit(entry->bitmap,
2791 BITS_PER_BITMAP, i);
Alexandre Oliva1bb91902011-10-14 12:10:36 -03002792 if (next_zero - i >= min_bits) {
Josef Bacik96303082009-07-13 21:29:25 -04002793 found_bits = next_zero - i;
2794 break;
2795 }
2796 i = next_zero;
2797 }
2798
2799 if (!found_bits)
Josef Bacik4e69b592011-03-21 10:11:24 -04002800 return -ENOSPC;
Josef Bacik96303082009-07-13 21:29:25 -04002801
Alexandre Oliva1bb91902011-10-14 12:10:36 -03002802 if (!total_found) {
Josef Bacik96303082009-07-13 21:29:25 -04002803 start = i;
Alexandre Olivab78d09b2011-11-30 13:43:00 -05002804 cluster->max_size = 0;
Josef Bacik96303082009-07-13 21:29:25 -04002805 }
2806
2807 total_found += found_bits;
2808
Wang Sheng-Hui96009762012-11-30 06:30:14 +00002809 if (cluster->max_size < found_bits * ctl->unit)
2810 cluster->max_size = found_bits * ctl->unit;
Josef Bacik96303082009-07-13 21:29:25 -04002811
Alexandre Oliva1bb91902011-10-14 12:10:36 -03002812 if (total_found < want_bits || cluster->max_size < cont1_bytes) {
2813 i = next_zero + 1;
Josef Bacik96303082009-07-13 21:29:25 -04002814 goto again;
2815 }
2816
Wang Sheng-Hui96009762012-11-30 06:30:14 +00002817 cluster->window_start = start * ctl->unit + entry->offset;
Li Zefan34d52cb2011-03-29 13:46:06 +08002818 rb_erase(&entry->offset_index, &ctl->free_space_offset);
Josef Bacik4e69b592011-03-21 10:11:24 -04002819 ret = tree_insert_offset(&cluster->root, entry->offset,
2820 &entry->offset_index, 1);
Josef Bacikb12d6862013-08-26 17:14:08 -04002821 ASSERT(!ret); /* -EEXIST; Logic error */
Josef Bacik96303082009-07-13 21:29:25 -04002822
Josef Bacik3f7de032011-11-10 08:29:20 -05002823 trace_btrfs_setup_cluster(block_group, cluster,
Wang Sheng-Hui96009762012-11-30 06:30:14 +00002824 total_found * ctl->unit, 1);
Josef Bacik96303082009-07-13 21:29:25 -04002825 return 0;
2826}
2827
Chris Masonfa9c0d792009-04-03 09:47:43 -04002828/*
Josef Bacik4e69b592011-03-21 10:11:24 -04002829 * This searches the block group for just extents to fill the cluster with.
Alexandre Oliva1bb91902011-10-14 12:10:36 -03002830 * Try to find a cluster with at least bytes total bytes, at least one
2831 * extent of cont1_bytes, and other clusters of at least min_bytes.
Josef Bacik4e69b592011-03-21 10:11:24 -04002832 */
Josef Bacik3de85bb2011-05-25 13:07:37 -04002833static noinline int
2834setup_cluster_no_bitmap(struct btrfs_block_group_cache *block_group,
2835 struct btrfs_free_cluster *cluster,
2836 struct list_head *bitmaps, u64 offset, u64 bytes,
Alexandre Oliva1bb91902011-10-14 12:10:36 -03002837 u64 cont1_bytes, u64 min_bytes)
Josef Bacik4e69b592011-03-21 10:11:24 -04002838{
Li Zefan34d52cb2011-03-29 13:46:06 +08002839 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
Josef Bacik4e69b592011-03-21 10:11:24 -04002840 struct btrfs_free_space *first = NULL;
2841 struct btrfs_free_space *entry = NULL;
Josef Bacik4e69b592011-03-21 10:11:24 -04002842 struct btrfs_free_space *last;
2843 struct rb_node *node;
Josef Bacik4e69b592011-03-21 10:11:24 -04002844 u64 window_free;
2845 u64 max_extent;
Josef Bacik3f7de032011-11-10 08:29:20 -05002846 u64 total_size = 0;
Josef Bacik4e69b592011-03-21 10:11:24 -04002847
Li Zefan34d52cb2011-03-29 13:46:06 +08002848 entry = tree_search_offset(ctl, offset, 0, 1);
Josef Bacik4e69b592011-03-21 10:11:24 -04002849 if (!entry)
2850 return -ENOSPC;
2851
2852 /*
2853 * We don't want bitmaps, so just move along until we find a normal
2854 * extent entry.
2855 */
Alexandre Oliva1bb91902011-10-14 12:10:36 -03002856 while (entry->bitmap || entry->bytes < min_bytes) {
2857 if (entry->bitmap && list_empty(&entry->list))
Josef Bacik86d4a772011-05-25 13:03:16 -04002858 list_add_tail(&entry->list, bitmaps);
Josef Bacik4e69b592011-03-21 10:11:24 -04002859 node = rb_next(&entry->offset_index);
2860 if (!node)
2861 return -ENOSPC;
2862 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2863 }
2864
Josef Bacik4e69b592011-03-21 10:11:24 -04002865 window_free = entry->bytes;
2866 max_extent = entry->bytes;
2867 first = entry;
2868 last = entry;
Josef Bacik4e69b592011-03-21 10:11:24 -04002869
Alexandre Oliva1bb91902011-10-14 12:10:36 -03002870 for (node = rb_next(&entry->offset_index); node;
2871 node = rb_next(&entry->offset_index)) {
Josef Bacik4e69b592011-03-21 10:11:24 -04002872 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2873
Josef Bacik86d4a772011-05-25 13:03:16 -04002874 if (entry->bitmap) {
2875 if (list_empty(&entry->list))
2876 list_add_tail(&entry->list, bitmaps);
Josef Bacik4e69b592011-03-21 10:11:24 -04002877 continue;
Josef Bacik86d4a772011-05-25 13:03:16 -04002878 }
2879
Alexandre Oliva1bb91902011-10-14 12:10:36 -03002880 if (entry->bytes < min_bytes)
2881 continue;
2882
2883 last = entry;
2884 window_free += entry->bytes;
2885 if (entry->bytes > max_extent)
Josef Bacik4e69b592011-03-21 10:11:24 -04002886 max_extent = entry->bytes;
Josef Bacik4e69b592011-03-21 10:11:24 -04002887 }
2888
Alexandre Oliva1bb91902011-10-14 12:10:36 -03002889 if (window_free < bytes || max_extent < cont1_bytes)
2890 return -ENOSPC;
2891
Josef Bacik4e69b592011-03-21 10:11:24 -04002892 cluster->window_start = first->offset;
2893
2894 node = &first->offset_index;
2895
2896 /*
2897 * now we've found our entries, pull them out of the free space
2898 * cache and put them into the cluster rbtree
2899 */
2900 do {
2901 int ret;
2902
2903 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2904 node = rb_next(&entry->offset_index);
Alexandre Oliva1bb91902011-10-14 12:10:36 -03002905 if (entry->bitmap || entry->bytes < min_bytes)
Josef Bacik4e69b592011-03-21 10:11:24 -04002906 continue;
2907
Li Zefan34d52cb2011-03-29 13:46:06 +08002908 rb_erase(&entry->offset_index, &ctl->free_space_offset);
Josef Bacik4e69b592011-03-21 10:11:24 -04002909 ret = tree_insert_offset(&cluster->root, entry->offset,
2910 &entry->offset_index, 0);
Josef Bacik3f7de032011-11-10 08:29:20 -05002911 total_size += entry->bytes;
Josef Bacikb12d6862013-08-26 17:14:08 -04002912 ASSERT(!ret); /* -EEXIST; Logic error */
Josef Bacik4e69b592011-03-21 10:11:24 -04002913 } while (node && entry != last);
2914
2915 cluster->max_size = max_extent;
Josef Bacik3f7de032011-11-10 08:29:20 -05002916 trace_btrfs_setup_cluster(block_group, cluster, total_size, 0);
Josef Bacik4e69b592011-03-21 10:11:24 -04002917 return 0;
2918}
2919
2920/*
2921 * This specifically looks for bitmaps that may work in the cluster, we assume
2922 * that we have already failed to find extents that will work.
2923 */
Josef Bacik3de85bb2011-05-25 13:07:37 -04002924static noinline int
2925setup_cluster_bitmap(struct btrfs_block_group_cache *block_group,
2926 struct btrfs_free_cluster *cluster,
2927 struct list_head *bitmaps, u64 offset, u64 bytes,
Alexandre Oliva1bb91902011-10-14 12:10:36 -03002928 u64 cont1_bytes, u64 min_bytes)
Josef Bacik4e69b592011-03-21 10:11:24 -04002929{
Li Zefan34d52cb2011-03-29 13:46:06 +08002930 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
Josef Bacik4e69b592011-03-21 10:11:24 -04002931 struct btrfs_free_space *entry;
Josef Bacik4e69b592011-03-21 10:11:24 -04002932 int ret = -ENOSPC;
Li Zefan0f0fbf12011-11-20 07:33:38 -05002933 u64 bitmap_offset = offset_to_bitmap(ctl, offset);
Josef Bacik4e69b592011-03-21 10:11:24 -04002934
Li Zefan34d52cb2011-03-29 13:46:06 +08002935 if (ctl->total_bitmaps == 0)
Josef Bacik4e69b592011-03-21 10:11:24 -04002936 return -ENOSPC;
2937
Josef Bacik86d4a772011-05-25 13:03:16 -04002938 /*
Li Zefan0f0fbf12011-11-20 07:33:38 -05002939 * The bitmap that covers offset won't be in the list unless offset
2940 * is just its start offset.
2941 */
2942 entry = list_first_entry(bitmaps, struct btrfs_free_space, list);
2943 if (entry->offset != bitmap_offset) {
2944 entry = tree_search_offset(ctl, bitmap_offset, 1, 0);
2945 if (entry && list_empty(&entry->list))
2946 list_add(&entry->list, bitmaps);
2947 }
2948
Josef Bacik86d4a772011-05-25 13:03:16 -04002949 list_for_each_entry(entry, bitmaps, list) {
Josef Bacik357b9782012-01-26 15:01:11 -05002950 if (entry->bytes < bytes)
Josef Bacik86d4a772011-05-25 13:03:16 -04002951 continue;
2952 ret = btrfs_bitmap_cluster(block_group, entry, cluster, offset,
Alexandre Oliva1bb91902011-10-14 12:10:36 -03002953 bytes, cont1_bytes, min_bytes);
Josef Bacik86d4a772011-05-25 13:03:16 -04002954 if (!ret)
2955 return 0;
2956 }
2957
2958 /*
Li Zefan52621cb2011-11-20 07:33:38 -05002959 * The bitmaps list has all the bitmaps that record free space
2960 * starting after offset, so no more search is required.
Josef Bacik86d4a772011-05-25 13:03:16 -04002961 */
Li Zefan52621cb2011-11-20 07:33:38 -05002962 return -ENOSPC;
Josef Bacik4e69b592011-03-21 10:11:24 -04002963}
2964
2965/*
Chris Masonfa9c0d792009-04-03 09:47:43 -04002966 * here we try to find a cluster of blocks in a block group. The goal
Alexandre Oliva1bb91902011-10-14 12:10:36 -03002967 * is to find at least bytes+empty_size.
Chris Masonfa9c0d792009-04-03 09:47:43 -04002968 * We might not find them all in one contiguous area.
2969 *
2970 * returns zero and sets up cluster if things worked out, otherwise
2971 * it returns -enospc
2972 */
Josef Bacik00361582013-08-14 14:02:47 -04002973int btrfs_find_space_cluster(struct btrfs_root *root,
Chris Masonfa9c0d792009-04-03 09:47:43 -04002974 struct btrfs_block_group_cache *block_group,
2975 struct btrfs_free_cluster *cluster,
2976 u64 offset, u64 bytes, u64 empty_size)
2977{
Li Zefan34d52cb2011-03-29 13:46:06 +08002978 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
Josef Bacik86d4a772011-05-25 13:03:16 -04002979 struct btrfs_free_space *entry, *tmp;
Li Zefan52621cb2011-11-20 07:33:38 -05002980 LIST_HEAD(bitmaps);
Chris Masonfa9c0d792009-04-03 09:47:43 -04002981 u64 min_bytes;
Alexandre Oliva1bb91902011-10-14 12:10:36 -03002982 u64 cont1_bytes;
Chris Masonfa9c0d792009-04-03 09:47:43 -04002983 int ret;
2984
Alexandre Oliva1bb91902011-10-14 12:10:36 -03002985 /*
2986 * Choose the minimum extent size we'll require for this
2987 * cluster. For SSD_SPREAD, don't allow any fragmentation.
2988 * For metadata, allow allocates with smaller extents. For
2989 * data, keep it dense.
2990 */
Chris Mason451d7582009-06-09 20:28:34 -04002991 if (btrfs_test_opt(root, SSD_SPREAD)) {
Alexandre Oliva1bb91902011-10-14 12:10:36 -03002992 cont1_bytes = min_bytes = bytes + empty_size;
Chris Mason451d7582009-06-09 20:28:34 -04002993 } else if (block_group->flags & BTRFS_BLOCK_GROUP_METADATA) {
Alexandre Oliva1bb91902011-10-14 12:10:36 -03002994 cont1_bytes = bytes;
2995 min_bytes = block_group->sectorsize;
2996 } else {
2997 cont1_bytes = max(bytes, (bytes + empty_size) >> 2);
2998 min_bytes = block_group->sectorsize;
2999 }
Chris Masonfa9c0d792009-04-03 09:47:43 -04003000
Li Zefan34d52cb2011-03-29 13:46:06 +08003001 spin_lock(&ctl->tree_lock);
Josef Bacik7d0d2e82011-03-18 15:13:42 -04003002
3003 /*
3004 * If we know we don't have enough space to make a cluster don't even
3005 * bother doing all the work to try and find one.
3006 */
Alexandre Oliva1bb91902011-10-14 12:10:36 -03003007 if (ctl->free_space < bytes) {
Li Zefan34d52cb2011-03-29 13:46:06 +08003008 spin_unlock(&ctl->tree_lock);
Josef Bacik7d0d2e82011-03-18 15:13:42 -04003009 return -ENOSPC;
3010 }
3011
Chris Masonfa9c0d792009-04-03 09:47:43 -04003012 spin_lock(&cluster->lock);
3013
3014 /* someone already found a cluster, hooray */
3015 if (cluster->block_group) {
3016 ret = 0;
3017 goto out;
3018 }
Josef Bacik4e69b592011-03-21 10:11:24 -04003019
Josef Bacik3f7de032011-11-10 08:29:20 -05003020 trace_btrfs_find_cluster(block_group, offset, bytes, empty_size,
3021 min_bytes);
3022
Josef Bacik86d4a772011-05-25 13:03:16 -04003023 ret = setup_cluster_no_bitmap(block_group, cluster, &bitmaps, offset,
Alexandre Oliva1bb91902011-10-14 12:10:36 -03003024 bytes + empty_size,
3025 cont1_bytes, min_bytes);
Josef Bacik4e69b592011-03-21 10:11:24 -04003026 if (ret)
Josef Bacik86d4a772011-05-25 13:03:16 -04003027 ret = setup_cluster_bitmap(block_group, cluster, &bitmaps,
Alexandre Oliva1bb91902011-10-14 12:10:36 -03003028 offset, bytes + empty_size,
3029 cont1_bytes, min_bytes);
Josef Bacik86d4a772011-05-25 13:03:16 -04003030
3031 /* Clear our temporary list */
3032 list_for_each_entry_safe(entry, tmp, &bitmaps, list)
3033 list_del_init(&entry->list);
Josef Bacik4e69b592011-03-21 10:11:24 -04003034
3035 if (!ret) {
3036 atomic_inc(&block_group->count);
3037 list_add_tail(&cluster->block_group_list,
3038 &block_group->cluster_list);
3039 cluster->block_group = block_group;
Josef Bacik3f7de032011-11-10 08:29:20 -05003040 } else {
3041 trace_btrfs_failed_cluster_setup(block_group);
Chris Masonfa9c0d792009-04-03 09:47:43 -04003042 }
Chris Masonfa9c0d792009-04-03 09:47:43 -04003043out:
3044 spin_unlock(&cluster->lock);
Li Zefan34d52cb2011-03-29 13:46:06 +08003045 spin_unlock(&ctl->tree_lock);
Chris Masonfa9c0d792009-04-03 09:47:43 -04003046
3047 return ret;
3048}
3049
3050/*
3051 * simple code to zero out a cluster
3052 */
3053void btrfs_init_free_cluster(struct btrfs_free_cluster *cluster)
3054{
3055 spin_lock_init(&cluster->lock);
3056 spin_lock_init(&cluster->refill_lock);
Eric Paris6bef4d32010-02-23 19:43:04 +00003057 cluster->root = RB_ROOT;
Chris Masonfa9c0d792009-04-03 09:47:43 -04003058 cluster->max_size = 0;
3059 INIT_LIST_HEAD(&cluster->block_group_list);
3060 cluster->block_group = NULL;
3061}
3062
Li Zefan7fe1e642011-12-29 14:47:27 +08003063static int do_trimming(struct btrfs_block_group_cache *block_group,
3064 u64 *total_trimmed, u64 start, u64 bytes,
Filipe Manana55507ce2014-12-01 17:04:09 +00003065 u64 reserved_start, u64 reserved_bytes,
3066 struct btrfs_trim_range *trim_entry)
Li Zefan7fe1e642011-12-29 14:47:27 +08003067{
3068 struct btrfs_space_info *space_info = block_group->space_info;
3069 struct btrfs_fs_info *fs_info = block_group->fs_info;
Filipe Manana55507ce2014-12-01 17:04:09 +00003070 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
Li Zefan7fe1e642011-12-29 14:47:27 +08003071 int ret;
3072 int update = 0;
3073 u64 trimmed = 0;
3074
3075 spin_lock(&space_info->lock);
3076 spin_lock(&block_group->lock);
3077 if (!block_group->ro) {
3078 block_group->reserved += reserved_bytes;
3079 space_info->bytes_reserved += reserved_bytes;
3080 update = 1;
3081 }
3082 spin_unlock(&block_group->lock);
3083 spin_unlock(&space_info->lock);
3084
Filipe Manana1edb647b2014-12-08 14:01:12 +00003085 ret = btrfs_discard_extent(fs_info->extent_root,
3086 start, bytes, &trimmed);
Li Zefan7fe1e642011-12-29 14:47:27 +08003087 if (!ret)
3088 *total_trimmed += trimmed;
3089
Filipe Manana55507ce2014-12-01 17:04:09 +00003090 mutex_lock(&ctl->cache_writeout_mutex);
Li Zefan7fe1e642011-12-29 14:47:27 +08003091 btrfs_add_free_space(block_group, reserved_start, reserved_bytes);
Filipe Manana55507ce2014-12-01 17:04:09 +00003092 list_del(&trim_entry->list);
3093 mutex_unlock(&ctl->cache_writeout_mutex);
Li Zefan7fe1e642011-12-29 14:47:27 +08003094
3095 if (update) {
3096 spin_lock(&space_info->lock);
3097 spin_lock(&block_group->lock);
3098 if (block_group->ro)
3099 space_info->bytes_readonly += reserved_bytes;
3100 block_group->reserved -= reserved_bytes;
3101 space_info->bytes_reserved -= reserved_bytes;
3102 spin_unlock(&space_info->lock);
3103 spin_unlock(&block_group->lock);
3104 }
3105
3106 return ret;
3107}
3108
3109static int trim_no_bitmap(struct btrfs_block_group_cache *block_group,
3110 u64 *total_trimmed, u64 start, u64 end, u64 minlen)
Li Dongyangf7039b12011-03-24 10:24:28 +00003111{
Li Zefan34d52cb2011-03-29 13:46:06 +08003112 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
Li Zefan7fe1e642011-12-29 14:47:27 +08003113 struct btrfs_free_space *entry;
3114 struct rb_node *node;
Li Dongyangf7039b12011-03-24 10:24:28 +00003115 int ret = 0;
Li Zefan7fe1e642011-12-29 14:47:27 +08003116 u64 extent_start;
3117 u64 extent_bytes;
3118 u64 bytes;
Li Dongyangf7039b12011-03-24 10:24:28 +00003119
3120 while (start < end) {
Filipe Manana55507ce2014-12-01 17:04:09 +00003121 struct btrfs_trim_range trim_entry;
3122
3123 mutex_lock(&ctl->cache_writeout_mutex);
Li Zefan34d52cb2011-03-29 13:46:06 +08003124 spin_lock(&ctl->tree_lock);
Li Dongyangf7039b12011-03-24 10:24:28 +00003125
Li Zefan34d52cb2011-03-29 13:46:06 +08003126 if (ctl->free_space < minlen) {
3127 spin_unlock(&ctl->tree_lock);
Filipe Manana55507ce2014-12-01 17:04:09 +00003128 mutex_unlock(&ctl->cache_writeout_mutex);
Li Dongyangf7039b12011-03-24 10:24:28 +00003129 break;
3130 }
3131
Li Zefan34d52cb2011-03-29 13:46:06 +08003132 entry = tree_search_offset(ctl, start, 0, 1);
Li Zefan7fe1e642011-12-29 14:47:27 +08003133 if (!entry) {
Li Zefan34d52cb2011-03-29 13:46:06 +08003134 spin_unlock(&ctl->tree_lock);
Filipe Manana55507ce2014-12-01 17:04:09 +00003135 mutex_unlock(&ctl->cache_writeout_mutex);
Li Dongyangf7039b12011-03-24 10:24:28 +00003136 break;
3137 }
3138
Li Zefan7fe1e642011-12-29 14:47:27 +08003139 /* skip bitmaps */
3140 while (entry->bitmap) {
3141 node = rb_next(&entry->offset_index);
3142 if (!node) {
Li Zefan34d52cb2011-03-29 13:46:06 +08003143 spin_unlock(&ctl->tree_lock);
Filipe Manana55507ce2014-12-01 17:04:09 +00003144 mutex_unlock(&ctl->cache_writeout_mutex);
Li Zefan7fe1e642011-12-29 14:47:27 +08003145 goto out;
Li Dongyangf7039b12011-03-24 10:24:28 +00003146 }
Li Zefan7fe1e642011-12-29 14:47:27 +08003147 entry = rb_entry(node, struct btrfs_free_space,
3148 offset_index);
Li Dongyangf7039b12011-03-24 10:24:28 +00003149 }
3150
Li Zefan7fe1e642011-12-29 14:47:27 +08003151 if (entry->offset >= end) {
3152 spin_unlock(&ctl->tree_lock);
Filipe Manana55507ce2014-12-01 17:04:09 +00003153 mutex_unlock(&ctl->cache_writeout_mutex);
Li Zefan7fe1e642011-12-29 14:47:27 +08003154 break;
3155 }
3156
3157 extent_start = entry->offset;
3158 extent_bytes = entry->bytes;
3159 start = max(start, extent_start);
3160 bytes = min(extent_start + extent_bytes, end) - start;
3161 if (bytes < minlen) {
3162 spin_unlock(&ctl->tree_lock);
Filipe Manana55507ce2014-12-01 17:04:09 +00003163 mutex_unlock(&ctl->cache_writeout_mutex);
Li Zefan7fe1e642011-12-29 14:47:27 +08003164 goto next;
3165 }
3166
3167 unlink_free_space(ctl, entry);
3168 kmem_cache_free(btrfs_free_space_cachep, entry);
3169
Li Zefan34d52cb2011-03-29 13:46:06 +08003170 spin_unlock(&ctl->tree_lock);
Filipe Manana55507ce2014-12-01 17:04:09 +00003171 trim_entry.start = extent_start;
3172 trim_entry.bytes = extent_bytes;
3173 list_add_tail(&trim_entry.list, &ctl->trimming_ranges);
3174 mutex_unlock(&ctl->cache_writeout_mutex);
Li Dongyangf7039b12011-03-24 10:24:28 +00003175
Li Zefan7fe1e642011-12-29 14:47:27 +08003176 ret = do_trimming(block_group, total_trimmed, start, bytes,
Filipe Manana55507ce2014-12-01 17:04:09 +00003177 extent_start, extent_bytes, &trim_entry);
Li Zefan7fe1e642011-12-29 14:47:27 +08003178 if (ret)
3179 break;
3180next:
Li Dongyangf7039b12011-03-24 10:24:28 +00003181 start += bytes;
Li Dongyangf7039b12011-03-24 10:24:28 +00003182
3183 if (fatal_signal_pending(current)) {
3184 ret = -ERESTARTSYS;
3185 break;
3186 }
3187
3188 cond_resched();
3189 }
Li Zefan7fe1e642011-12-29 14:47:27 +08003190out:
3191 return ret;
3192}
3193
3194static int trim_bitmaps(struct btrfs_block_group_cache *block_group,
3195 u64 *total_trimmed, u64 start, u64 end, u64 minlen)
3196{
3197 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3198 struct btrfs_free_space *entry;
3199 int ret = 0;
3200 int ret2;
3201 u64 bytes;
3202 u64 offset = offset_to_bitmap(ctl, start);
3203
3204 while (offset < end) {
3205 bool next_bitmap = false;
Filipe Manana55507ce2014-12-01 17:04:09 +00003206 struct btrfs_trim_range trim_entry;
Li Zefan7fe1e642011-12-29 14:47:27 +08003207
Filipe Manana55507ce2014-12-01 17:04:09 +00003208 mutex_lock(&ctl->cache_writeout_mutex);
Li Zefan7fe1e642011-12-29 14:47:27 +08003209 spin_lock(&ctl->tree_lock);
3210
3211 if (ctl->free_space < minlen) {
3212 spin_unlock(&ctl->tree_lock);
Filipe Manana55507ce2014-12-01 17:04:09 +00003213 mutex_unlock(&ctl->cache_writeout_mutex);
Li Zefan7fe1e642011-12-29 14:47:27 +08003214 break;
3215 }
3216
3217 entry = tree_search_offset(ctl, offset, 1, 0);
3218 if (!entry) {
3219 spin_unlock(&ctl->tree_lock);
Filipe Manana55507ce2014-12-01 17:04:09 +00003220 mutex_unlock(&ctl->cache_writeout_mutex);
Li Zefan7fe1e642011-12-29 14:47:27 +08003221 next_bitmap = true;
3222 goto next;
3223 }
3224
3225 bytes = minlen;
3226 ret2 = search_bitmap(ctl, entry, &start, &bytes);
3227 if (ret2 || start >= end) {
3228 spin_unlock(&ctl->tree_lock);
Filipe Manana55507ce2014-12-01 17:04:09 +00003229 mutex_unlock(&ctl->cache_writeout_mutex);
Li Zefan7fe1e642011-12-29 14:47:27 +08003230 next_bitmap = true;
3231 goto next;
3232 }
3233
3234 bytes = min(bytes, end - start);
3235 if (bytes < minlen) {
3236 spin_unlock(&ctl->tree_lock);
Filipe Manana55507ce2014-12-01 17:04:09 +00003237 mutex_unlock(&ctl->cache_writeout_mutex);
Li Zefan7fe1e642011-12-29 14:47:27 +08003238 goto next;
3239 }
3240
3241 bitmap_clear_bits(ctl, entry, start, bytes);
3242 if (entry->bytes == 0)
3243 free_bitmap(ctl, entry);
3244
3245 spin_unlock(&ctl->tree_lock);
Filipe Manana55507ce2014-12-01 17:04:09 +00003246 trim_entry.start = start;
3247 trim_entry.bytes = bytes;
3248 list_add_tail(&trim_entry.list, &ctl->trimming_ranges);
3249 mutex_unlock(&ctl->cache_writeout_mutex);
Li Zefan7fe1e642011-12-29 14:47:27 +08003250
3251 ret = do_trimming(block_group, total_trimmed, start, bytes,
Filipe Manana55507ce2014-12-01 17:04:09 +00003252 start, bytes, &trim_entry);
Li Zefan7fe1e642011-12-29 14:47:27 +08003253 if (ret)
3254 break;
3255next:
3256 if (next_bitmap) {
3257 offset += BITS_PER_BITMAP * ctl->unit;
3258 } else {
3259 start += bytes;
3260 if (start >= offset + BITS_PER_BITMAP * ctl->unit)
3261 offset += BITS_PER_BITMAP * ctl->unit;
3262 }
3263
3264 if (fatal_signal_pending(current)) {
3265 ret = -ERESTARTSYS;
3266 break;
3267 }
3268
3269 cond_resched();
3270 }
3271
3272 return ret;
3273}
3274
Jeff Mahoneye33e17e2015-06-15 09:41:19 -04003275void btrfs_get_block_group_trimming(struct btrfs_block_group_cache *cache)
Li Zefan7fe1e642011-12-29 14:47:27 +08003276{
Jeff Mahoneye33e17e2015-06-15 09:41:19 -04003277 atomic_inc(&cache->trimming);
3278}
Li Zefan7fe1e642011-12-29 14:47:27 +08003279
Jeff Mahoneye33e17e2015-06-15 09:41:19 -04003280void btrfs_put_block_group_trimming(struct btrfs_block_group_cache *block_group)
3281{
3282 struct extent_map_tree *em_tree;
3283 struct extent_map *em;
3284 bool cleanup;
Li Zefan7fe1e642011-12-29 14:47:27 +08003285
Filipe Manana04216822014-11-27 21:14:15 +00003286 spin_lock(&block_group->lock);
Jeff Mahoneye33e17e2015-06-15 09:41:19 -04003287 cleanup = (atomic_dec_and_test(&block_group->trimming) &&
3288 block_group->removed);
Filipe Manana04216822014-11-27 21:14:15 +00003289 spin_unlock(&block_group->lock);
3290
Jeff Mahoneye33e17e2015-06-15 09:41:19 -04003291 if (cleanup) {
Filipe Mananaa1e7e162014-12-04 15:31:01 +00003292 lock_chunks(block_group->fs_info->chunk_root);
Filipe Manana04216822014-11-27 21:14:15 +00003293 em_tree = &block_group->fs_info->mapping_tree.map_tree;
3294 write_lock(&em_tree->lock);
3295 em = lookup_extent_mapping(em_tree, block_group->key.objectid,
3296 1);
3297 BUG_ON(!em); /* logic error, can't happen */
Filipe Mananaa1e7e162014-12-04 15:31:01 +00003298 /*
3299 * remove_extent_mapping() will delete us from the pinned_chunks
3300 * list, which is protected by the chunk mutex.
3301 */
Filipe Manana04216822014-11-27 21:14:15 +00003302 remove_extent_mapping(em_tree, em);
3303 write_unlock(&em_tree->lock);
Filipe Manana04216822014-11-27 21:14:15 +00003304 unlock_chunks(block_group->fs_info->chunk_root);
3305
3306 /* once for us and once for the tree */
3307 free_extent_map(em);
3308 free_extent_map(em);
Filipe Manana946ddbe2014-12-01 17:04:40 +00003309
3310 /*
3311 * We've left one free space entry and other tasks trimming
3312 * this block group have left 1 entry each one. Free them.
3313 */
3314 __btrfs_remove_free_space_cache(block_group->free_space_ctl);
Filipe Manana04216822014-11-27 21:14:15 +00003315 }
Jeff Mahoneye33e17e2015-06-15 09:41:19 -04003316}
Li Dongyangf7039b12011-03-24 10:24:28 +00003317
Jeff Mahoneye33e17e2015-06-15 09:41:19 -04003318int btrfs_trim_block_group(struct btrfs_block_group_cache *block_group,
3319 u64 *trimmed, u64 start, u64 end, u64 minlen)
3320{
3321 int ret;
3322
3323 *trimmed = 0;
3324
3325 spin_lock(&block_group->lock);
3326 if (block_group->removed) {
3327 spin_unlock(&block_group->lock);
3328 return 0;
3329 }
3330 btrfs_get_block_group_trimming(block_group);
3331 spin_unlock(&block_group->lock);
3332
3333 ret = trim_no_bitmap(block_group, trimmed, start, end, minlen);
3334 if (ret)
3335 goto out;
3336
3337 ret = trim_bitmaps(block_group, trimmed, start, end, minlen);
3338out:
3339 btrfs_put_block_group_trimming(block_group);
Li Dongyangf7039b12011-03-24 10:24:28 +00003340 return ret;
3341}
Li Zefan581bb052011-04-20 10:06:11 +08003342
3343/*
3344 * Find the left-most item in the cache tree, and then return the
3345 * smallest inode number in the item.
3346 *
3347 * Note: the returned inode number may not be the smallest one in
3348 * the tree, if the left-most item is a bitmap.
3349 */
3350u64 btrfs_find_ino_for_alloc(struct btrfs_root *fs_root)
3351{
3352 struct btrfs_free_space_ctl *ctl = fs_root->free_ino_ctl;
3353 struct btrfs_free_space *entry = NULL;
3354 u64 ino = 0;
3355
3356 spin_lock(&ctl->tree_lock);
3357
3358 if (RB_EMPTY_ROOT(&ctl->free_space_offset))
3359 goto out;
3360
3361 entry = rb_entry(rb_first(&ctl->free_space_offset),
3362 struct btrfs_free_space, offset_index);
3363
3364 if (!entry->bitmap) {
3365 ino = entry->offset;
3366
3367 unlink_free_space(ctl, entry);
3368 entry->offset++;
3369 entry->bytes--;
3370 if (!entry->bytes)
3371 kmem_cache_free(btrfs_free_space_cachep, entry);
3372 else
3373 link_free_space(ctl, entry);
3374 } else {
3375 u64 offset = 0;
3376 u64 count = 1;
3377 int ret;
3378
3379 ret = search_bitmap(ctl, entry, &offset, &count);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003380 /* Logic error; Should be empty if it can't find anything */
Josef Bacikb12d6862013-08-26 17:14:08 -04003381 ASSERT(!ret);
Li Zefan581bb052011-04-20 10:06:11 +08003382
3383 ino = offset;
3384 bitmap_clear_bits(ctl, entry, offset, 1);
3385 if (entry->bytes == 0)
3386 free_bitmap(ctl, entry);
3387 }
3388out:
3389 spin_unlock(&ctl->tree_lock);
3390
3391 return ino;
3392}
Li Zefan82d59022011-04-20 10:33:24 +08003393
3394struct inode *lookup_free_ino_inode(struct btrfs_root *root,
3395 struct btrfs_path *path)
3396{
3397 struct inode *inode = NULL;
3398
David Sterba57cdc8d2014-02-05 02:37:48 +01003399 spin_lock(&root->ino_cache_lock);
3400 if (root->ino_cache_inode)
3401 inode = igrab(root->ino_cache_inode);
3402 spin_unlock(&root->ino_cache_lock);
Li Zefan82d59022011-04-20 10:33:24 +08003403 if (inode)
3404 return inode;
3405
3406 inode = __lookup_free_space_inode(root, path, 0);
3407 if (IS_ERR(inode))
3408 return inode;
3409
David Sterba57cdc8d2014-02-05 02:37:48 +01003410 spin_lock(&root->ino_cache_lock);
David Sterba7841cb22011-05-31 18:07:27 +02003411 if (!btrfs_fs_closing(root->fs_info))
David Sterba57cdc8d2014-02-05 02:37:48 +01003412 root->ino_cache_inode = igrab(inode);
3413 spin_unlock(&root->ino_cache_lock);
Li Zefan82d59022011-04-20 10:33:24 +08003414
3415 return inode;
3416}
3417
3418int create_free_ino_inode(struct btrfs_root *root,
3419 struct btrfs_trans_handle *trans,
3420 struct btrfs_path *path)
3421{
3422 return __create_free_space_inode(root, trans, path,
3423 BTRFS_FREE_INO_OBJECTID, 0);
3424}
3425
3426int load_free_ino_cache(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
3427{
3428 struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
3429 struct btrfs_path *path;
3430 struct inode *inode;
3431 int ret = 0;
3432 u64 root_gen = btrfs_root_generation(&root->root_item);
3433
Chris Mason4b9465c2011-06-03 09:36:29 -04003434 if (!btrfs_test_opt(root, INODE_MAP_CACHE))
3435 return 0;
3436
Li Zefan82d59022011-04-20 10:33:24 +08003437 /*
3438 * If we're unmounting then just return, since this does a search on the
3439 * normal root and not the commit root and we could deadlock.
3440 */
David Sterba7841cb22011-05-31 18:07:27 +02003441 if (btrfs_fs_closing(fs_info))
Li Zefan82d59022011-04-20 10:33:24 +08003442 return 0;
3443
3444 path = btrfs_alloc_path();
3445 if (!path)
3446 return 0;
3447
3448 inode = lookup_free_ino_inode(root, path);
3449 if (IS_ERR(inode))
3450 goto out;
3451
3452 if (root_gen != BTRFS_I(inode)->generation)
3453 goto out_put;
3454
3455 ret = __load_free_space_cache(root, inode, ctl, path, 0);
3456
3457 if (ret < 0)
Simon Kirbyc2cf52e2013-03-19 22:41:23 +00003458 btrfs_err(fs_info,
3459 "failed to load free ino cache for root %llu",
3460 root->root_key.objectid);
Li Zefan82d59022011-04-20 10:33:24 +08003461out_put:
3462 iput(inode);
3463out:
3464 btrfs_free_path(path);
3465 return ret;
3466}
3467
3468int btrfs_write_out_ino_cache(struct btrfs_root *root,
3469 struct btrfs_trans_handle *trans,
Filipe David Borba Manana53645a92013-09-20 14:43:28 +01003470 struct btrfs_path *path,
3471 struct inode *inode)
Li Zefan82d59022011-04-20 10:33:24 +08003472{
3473 struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
Li Zefan82d59022011-04-20 10:33:24 +08003474 int ret;
Chris Masonc9dc4c62015-04-04 17:14:42 -07003475 struct btrfs_io_ctl io_ctl;
Filipe Mananae43699d2015-05-05 15:21:27 +01003476 bool release_metadata = true;
Li Zefan82d59022011-04-20 10:33:24 +08003477
Chris Mason4b9465c2011-06-03 09:36:29 -04003478 if (!btrfs_test_opt(root, INODE_MAP_CACHE))
3479 return 0;
3480
Chris Mason85db36c2015-04-23 08:02:49 -07003481 memset(&io_ctl, 0, sizeof(io_ctl));
Chris Masonc9dc4c62015-04-04 17:14:42 -07003482 ret = __btrfs_write_out_cache(root, inode, ctl, NULL, &io_ctl,
Chris Mason85db36c2015-04-23 08:02:49 -07003483 trans, path, 0);
Filipe Mananae43699d2015-05-05 15:21:27 +01003484 if (!ret) {
3485 /*
3486 * At this point writepages() didn't error out, so our metadata
3487 * reservation is released when the writeback finishes, at
3488 * inode.c:btrfs_finish_ordered_io(), regardless of it finishing
3489 * with or without an error.
3490 */
3491 release_metadata = false;
Chris Mason85db36c2015-04-23 08:02:49 -07003492 ret = btrfs_wait_cache_io(root, trans, NULL, &io_ctl, path, 0);
Filipe Mananae43699d2015-05-05 15:21:27 +01003493 }
Chris Mason85db36c2015-04-23 08:02:49 -07003494
Josef Bacikc09544e2011-08-30 10:19:10 -04003495 if (ret) {
Filipe Mananae43699d2015-05-05 15:21:27 +01003496 if (release_metadata)
3497 btrfs_delalloc_release_metadata(inode, inode->i_size);
Josef Bacikc09544e2011-08-30 10:19:10 -04003498#ifdef DEBUG
Simon Kirbyc2cf52e2013-03-19 22:41:23 +00003499 btrfs_err(root->fs_info,
3500 "failed to write free ino cache for root %llu",
3501 root->root_key.objectid);
Josef Bacikc09544e2011-08-30 10:19:10 -04003502#endif
3503 }
Li Zefan82d59022011-04-20 10:33:24 +08003504
Li Zefan82d59022011-04-20 10:33:24 +08003505 return ret;
3506}
Josef Bacik74255aa2013-03-15 09:47:08 -04003507
3508#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
Josef Bacikdc11dd5d2013-08-14 15:05:12 -04003509/*
3510 * Use this if you need to make a bitmap or extent entry specifically, it
3511 * doesn't do any of the merging that add_free_space does, this acts a lot like
3512 * how the free space cache loading stuff works, so you can get really weird
3513 * configurations.
3514 */
3515int test_add_free_space_entry(struct btrfs_block_group_cache *cache,
3516 u64 offset, u64 bytes, bool bitmap)
Josef Bacik74255aa2013-03-15 09:47:08 -04003517{
Josef Bacikdc11dd5d2013-08-14 15:05:12 -04003518 struct btrfs_free_space_ctl *ctl = cache->free_space_ctl;
3519 struct btrfs_free_space *info = NULL, *bitmap_info;
3520 void *map = NULL;
3521 u64 bytes_added;
3522 int ret;
Josef Bacik74255aa2013-03-15 09:47:08 -04003523
Josef Bacikdc11dd5d2013-08-14 15:05:12 -04003524again:
3525 if (!info) {
3526 info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
3527 if (!info)
3528 return -ENOMEM;
Josef Bacik74255aa2013-03-15 09:47:08 -04003529 }
3530
Josef Bacikdc11dd5d2013-08-14 15:05:12 -04003531 if (!bitmap) {
3532 spin_lock(&ctl->tree_lock);
3533 info->offset = offset;
3534 info->bytes = bytes;
3535 ret = link_free_space(ctl, info);
3536 spin_unlock(&ctl->tree_lock);
3537 if (ret)
3538 kmem_cache_free(btrfs_free_space_cachep, info);
3539 return ret;
3540 }
Josef Bacik74255aa2013-03-15 09:47:08 -04003541
Josef Bacikdc11dd5d2013-08-14 15:05:12 -04003542 if (!map) {
3543 map = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
3544 if (!map) {
3545 kmem_cache_free(btrfs_free_space_cachep, info);
3546 return -ENOMEM;
3547 }
3548 }
Josef Bacik74255aa2013-03-15 09:47:08 -04003549
Josef Bacikdc11dd5d2013-08-14 15:05:12 -04003550 spin_lock(&ctl->tree_lock);
3551 bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
3552 1, 0);
3553 if (!bitmap_info) {
3554 info->bitmap = map;
3555 map = NULL;
3556 add_new_bitmap(ctl, info, offset);
3557 bitmap_info = info;
Filipe Manana20005522014-08-29 13:35:13 +01003558 info = NULL;
Josef Bacikdc11dd5d2013-08-14 15:05:12 -04003559 }
Josef Bacik74255aa2013-03-15 09:47:08 -04003560
Josef Bacikdc11dd5d2013-08-14 15:05:12 -04003561 bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes);
3562 bytes -= bytes_added;
3563 offset += bytes_added;
3564 spin_unlock(&ctl->tree_lock);
3565
3566 if (bytes)
3567 goto again;
3568
Filipe Manana20005522014-08-29 13:35:13 +01003569 if (info)
3570 kmem_cache_free(btrfs_free_space_cachep, info);
Josef Bacikdc11dd5d2013-08-14 15:05:12 -04003571 if (map)
3572 kfree(map);
3573 return 0;
Josef Bacik74255aa2013-03-15 09:47:08 -04003574}
3575
3576/*
3577 * Checks to see if the given range is in the free space cache. This is really
3578 * just used to check the absence of space, so if there is free space in the
3579 * range at all we will return 1.
3580 */
Josef Bacikdc11dd5d2013-08-14 15:05:12 -04003581int test_check_exists(struct btrfs_block_group_cache *cache,
3582 u64 offset, u64 bytes)
Josef Bacik74255aa2013-03-15 09:47:08 -04003583{
3584 struct btrfs_free_space_ctl *ctl = cache->free_space_ctl;
3585 struct btrfs_free_space *info;
3586 int ret = 0;
3587
3588 spin_lock(&ctl->tree_lock);
3589 info = tree_search_offset(ctl, offset, 0, 0);
3590 if (!info) {
3591 info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
3592 1, 0);
3593 if (!info)
3594 goto out;
3595 }
3596
3597have_info:
3598 if (info->bitmap) {
3599 u64 bit_off, bit_bytes;
3600 struct rb_node *n;
3601 struct btrfs_free_space *tmp;
3602
3603 bit_off = offset;
3604 bit_bytes = ctl->unit;
3605 ret = search_bitmap(ctl, info, &bit_off, &bit_bytes);
3606 if (!ret) {
3607 if (bit_off == offset) {
3608 ret = 1;
3609 goto out;
3610 } else if (bit_off > offset &&
3611 offset + bytes > bit_off) {
3612 ret = 1;
3613 goto out;
3614 }
3615 }
3616
3617 n = rb_prev(&info->offset_index);
3618 while (n) {
3619 tmp = rb_entry(n, struct btrfs_free_space,
3620 offset_index);
3621 if (tmp->offset + tmp->bytes < offset)
3622 break;
3623 if (offset + bytes < tmp->offset) {
3624 n = rb_prev(&info->offset_index);
3625 continue;
3626 }
3627 info = tmp;
3628 goto have_info;
3629 }
3630
3631 n = rb_next(&info->offset_index);
3632 while (n) {
3633 tmp = rb_entry(n, struct btrfs_free_space,
3634 offset_index);
3635 if (offset + bytes < tmp->offset)
3636 break;
3637 if (tmp->offset + tmp->bytes < offset) {
3638 n = rb_next(&info->offset_index);
3639 continue;
3640 }
3641 info = tmp;
3642 goto have_info;
3643 }
3644
Filipe Manana20005522014-08-29 13:35:13 +01003645 ret = 0;
Josef Bacik74255aa2013-03-15 09:47:08 -04003646 goto out;
3647 }
3648
3649 if (info->offset == offset) {
3650 ret = 1;
3651 goto out;
3652 }
3653
3654 if (offset > info->offset && offset < info->offset + info->bytes)
3655 ret = 1;
3656out:
3657 spin_unlock(&ctl->tree_lock);
3658 return ret;
3659}
Josef Bacikdc11dd5d2013-08-14 15:05:12 -04003660#endif /* CONFIG_BTRFS_FS_RUN_SANITY_TESTS */