blob: 202008ec367d4c4c2cfcf73f7289692dd910b25c [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 Masonfa9c0d72009-04-03 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"
Chris Masonfa9c0d72009-04-03 09:47:43 -040030
Josef Bacik96303082009-07-13 21:29:25 -040031#define BITS_PER_BITMAP (PAGE_CACHE_SIZE * 8)
32#define MAX_CACHE_BYTES_PER_GIG (32 * 1024)
33
Li Zefan34d52cb2011-03-29 13:46:06 +080034static int link_free_space(struct btrfs_free_space_ctl *ctl,
Josef Bacik0cb59c92010-07-02 12:14:14 -040035 struct btrfs_free_space *info);
36
Li Zefan0414efa2011-04-20 10:20:14 +080037static struct inode *__lookup_free_space_inode(struct btrfs_root *root,
38 struct btrfs_path *path,
39 u64 offset)
Josef Bacik0af3d002010-06-21 14:48:16 -040040{
41 struct btrfs_key key;
42 struct btrfs_key location;
43 struct btrfs_disk_key disk_key;
44 struct btrfs_free_space_header *header;
45 struct extent_buffer *leaf;
46 struct inode *inode = NULL;
47 int ret;
48
Josef Bacik0af3d002010-06-21 14:48:16 -040049 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
Li Zefan0414efa2011-04-20 10:20:14 +080050 key.offset = offset;
Josef Bacik0af3d002010-06-21 14:48:16 -040051 key.type = 0;
52
53 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
54 if (ret < 0)
55 return ERR_PTR(ret);
56 if (ret > 0) {
David Sterbab3b4aa72011-04-21 01:20:15 +020057 btrfs_release_path(path);
Josef Bacik0af3d002010-06-21 14:48:16 -040058 return ERR_PTR(-ENOENT);
59 }
60
61 leaf = path->nodes[0];
62 header = btrfs_item_ptr(leaf, path->slots[0],
63 struct btrfs_free_space_header);
64 btrfs_free_space_key(leaf, header, &disk_key);
65 btrfs_disk_key_to_cpu(&location, &disk_key);
David Sterbab3b4aa72011-04-21 01:20:15 +020066 btrfs_release_path(path);
Josef Bacik0af3d002010-06-21 14:48:16 -040067
68 inode = btrfs_iget(root->fs_info->sb, &location, root, NULL);
69 if (!inode)
70 return ERR_PTR(-ENOENT);
71 if (IS_ERR(inode))
72 return inode;
73 if (is_bad_inode(inode)) {
74 iput(inode);
75 return ERR_PTR(-ENOENT);
76 }
77
Miao Xieadae52b2011-03-31 09:43:23 +000078 inode->i_mapping->flags &= ~__GFP_FS;
79
Li Zefan0414efa2011-04-20 10:20:14 +080080 return inode;
81}
82
83struct inode *lookup_free_space_inode(struct btrfs_root *root,
84 struct btrfs_block_group_cache
85 *block_group, struct btrfs_path *path)
86{
87 struct inode *inode = NULL;
Josef Bacik5b0e95b2011-10-06 08:58:24 -040088 u32 flags = BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW;
Li Zefan0414efa2011-04-20 10:20:14 +080089
90 spin_lock(&block_group->lock);
91 if (block_group->inode)
92 inode = igrab(block_group->inode);
93 spin_unlock(&block_group->lock);
94 if (inode)
95 return inode;
96
97 inode = __lookup_free_space_inode(root, path,
98 block_group->key.objectid);
99 if (IS_ERR(inode))
100 return inode;
101
Josef Bacik0af3d002010-06-21 14:48:16 -0400102 spin_lock(&block_group->lock);
Josef Bacik5b0e95b2011-10-06 08:58:24 -0400103 if (!((BTRFS_I(inode)->flags & flags) == flags)) {
Josef Bacik2f356122011-06-10 15:31:13 -0400104 printk(KERN_INFO "Old style space inode found, converting.\n");
Josef Bacik5b0e95b2011-10-06 08:58:24 -0400105 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM |
106 BTRFS_INODE_NODATACOW;
Josef Bacik2f356122011-06-10 15:31:13 -0400107 block_group->disk_cache_state = BTRFS_DC_CLEAR;
108 }
109
Josef Bacik300e4f82011-08-29 14:06:00 -0400110 if (!block_group->iref) {
Josef Bacik0af3d002010-06-21 14:48:16 -0400111 block_group->inode = igrab(inode);
112 block_group->iref = 1;
113 }
114 spin_unlock(&block_group->lock);
115
116 return inode;
117}
118
Li Zefan0414efa2011-04-20 10:20:14 +0800119int __create_free_space_inode(struct btrfs_root *root,
120 struct btrfs_trans_handle *trans,
121 struct btrfs_path *path, u64 ino, u64 offset)
Josef Bacik0af3d002010-06-21 14:48:16 -0400122{
123 struct btrfs_key key;
124 struct btrfs_disk_key disk_key;
125 struct btrfs_free_space_header *header;
126 struct btrfs_inode_item *inode_item;
127 struct extent_buffer *leaf;
Josef Bacik5b0e95b2011-10-06 08:58:24 -0400128 u64 flags = BTRFS_INODE_NOCOMPRESS | BTRFS_INODE_PREALLOC;
Josef Bacik0af3d002010-06-21 14:48:16 -0400129 int ret;
130
Li Zefan0414efa2011-04-20 10:20:14 +0800131 ret = btrfs_insert_empty_inode(trans, root, path, ino);
Josef Bacik0af3d002010-06-21 14:48:16 -0400132 if (ret)
133 return ret;
134
Josef Bacik5b0e95b2011-10-06 08:58:24 -0400135 /* We inline crc's for the free disk space cache */
136 if (ino != BTRFS_FREE_INO_OBJECTID)
137 flags |= BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW;
138
Josef Bacik0af3d002010-06-21 14:48:16 -0400139 leaf = path->nodes[0];
140 inode_item = btrfs_item_ptr(leaf, path->slots[0],
141 struct btrfs_inode_item);
142 btrfs_item_key(leaf, &disk_key, path->slots[0]);
143 memset_extent_buffer(leaf, 0, (unsigned long)inode_item,
144 sizeof(*inode_item));
145 btrfs_set_inode_generation(leaf, inode_item, trans->transid);
146 btrfs_set_inode_size(leaf, inode_item, 0);
147 btrfs_set_inode_nbytes(leaf, inode_item, 0);
148 btrfs_set_inode_uid(leaf, inode_item, 0);
149 btrfs_set_inode_gid(leaf, inode_item, 0);
150 btrfs_set_inode_mode(leaf, inode_item, S_IFREG | 0600);
Josef Bacik5b0e95b2011-10-06 08:58:24 -0400151 btrfs_set_inode_flags(leaf, inode_item, flags);
Josef Bacik0af3d002010-06-21 14:48:16 -0400152 btrfs_set_inode_nlink(leaf, inode_item, 1);
153 btrfs_set_inode_transid(leaf, inode_item, trans->transid);
Li Zefan0414efa2011-04-20 10:20:14 +0800154 btrfs_set_inode_block_group(leaf, inode_item, offset);
Josef Bacik0af3d002010-06-21 14:48:16 -0400155 btrfs_mark_buffer_dirty(leaf);
David Sterbab3b4aa72011-04-21 01:20:15 +0200156 btrfs_release_path(path);
Josef Bacik0af3d002010-06-21 14:48:16 -0400157
158 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
Li Zefan0414efa2011-04-20 10:20:14 +0800159 key.offset = offset;
Josef Bacik0af3d002010-06-21 14:48:16 -0400160 key.type = 0;
161
162 ret = btrfs_insert_empty_item(trans, root, path, &key,
163 sizeof(struct btrfs_free_space_header));
164 if (ret < 0) {
David Sterbab3b4aa72011-04-21 01:20:15 +0200165 btrfs_release_path(path);
Josef Bacik0af3d002010-06-21 14:48:16 -0400166 return ret;
167 }
168 leaf = path->nodes[0];
169 header = btrfs_item_ptr(leaf, path->slots[0],
170 struct btrfs_free_space_header);
171 memset_extent_buffer(leaf, 0, (unsigned long)header, sizeof(*header));
172 btrfs_set_free_space_key(leaf, header, &disk_key);
173 btrfs_mark_buffer_dirty(leaf);
David Sterbab3b4aa72011-04-21 01:20:15 +0200174 btrfs_release_path(path);
Josef Bacik0af3d002010-06-21 14:48:16 -0400175
176 return 0;
177}
178
Li Zefan0414efa2011-04-20 10:20:14 +0800179int create_free_space_inode(struct btrfs_root *root,
180 struct btrfs_trans_handle *trans,
181 struct btrfs_block_group_cache *block_group,
182 struct btrfs_path *path)
183{
184 int ret;
185 u64 ino;
186
187 ret = btrfs_find_free_objectid(root, &ino);
188 if (ret < 0)
189 return ret;
190
191 return __create_free_space_inode(root, trans, path, ino,
192 block_group->key.objectid);
193}
194
Josef Bacik0af3d002010-06-21 14:48:16 -0400195int btrfs_truncate_free_space_cache(struct btrfs_root *root,
196 struct btrfs_trans_handle *trans,
197 struct btrfs_path *path,
198 struct inode *inode)
199{
Liu Bo65450aa2011-09-11 10:52:24 -0400200 struct btrfs_block_rsv *rsv;
Josef Bacikc8174312011-11-02 09:29:35 -0400201 u64 needed_bytes;
Josef Bacik0af3d002010-06-21 14:48:16 -0400202 loff_t oldsize;
203 int ret = 0;
204
Liu Bo65450aa2011-09-11 10:52:24 -0400205 rsv = trans->block_rsv;
Josef Bacikc8174312011-11-02 09:29:35 -0400206 trans->block_rsv = &root->fs_info->global_block_rsv;
207
208 /* 1 for slack space, 1 for updating the inode */
209 needed_bytes = btrfs_calc_trunc_metadata_size(root, 1) +
210 btrfs_calc_trans_metadata_size(root, 1);
211
212 spin_lock(&trans->block_rsv->lock);
213 if (trans->block_rsv->reserved < needed_bytes) {
214 spin_unlock(&trans->block_rsv->lock);
215 trans->block_rsv = rsv;
216 return -ENOSPC;
217 }
218 spin_unlock(&trans->block_rsv->lock);
Josef Bacik0af3d002010-06-21 14:48:16 -0400219
220 oldsize = i_size_read(inode);
221 btrfs_i_size_write(inode, 0);
222 truncate_pagecache(inode, oldsize, 0);
223
224 /*
225 * We don't need an orphan item because truncating the free space cache
226 * will never be split across transactions.
227 */
228 ret = btrfs_truncate_inode_items(trans, root, inode,
229 0, BTRFS_EXTENT_DATA_KEY);
Liu Bo65450aa2011-09-11 10:52:24 -0400230
Josef Bacik0af3d002010-06-21 14:48:16 -0400231 if (ret) {
Josef Bacikc8174312011-11-02 09:29:35 -0400232 trans->block_rsv = rsv;
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100233 btrfs_abort_transaction(trans, root, ret);
Josef Bacik0af3d002010-06-21 14:48:16 -0400234 return ret;
235 }
236
Li Zefan82d59022011-04-20 10:33:24 +0800237 ret = btrfs_update_inode(trans, root, inode);
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100238 if (ret)
239 btrfs_abort_transaction(trans, root, ret);
Josef Bacikc8174312011-11-02 09:29:35 -0400240 trans->block_rsv = rsv;
241
Li Zefan82d59022011-04-20 10:33:24 +0800242 return ret;
Josef Bacik0af3d002010-06-21 14:48:16 -0400243}
244
Josef Bacik9d66e232010-08-25 16:54:15 -0400245static int readahead_cache(struct inode *inode)
246{
247 struct file_ra_state *ra;
248 unsigned long last_index;
249
250 ra = kzalloc(sizeof(*ra), GFP_NOFS);
251 if (!ra)
252 return -ENOMEM;
253
254 file_ra_state_init(ra, inode->i_mapping);
255 last_index = (i_size_read(inode) - 1) >> PAGE_CACHE_SHIFT;
256
257 page_cache_sync_readahead(inode->i_mapping, ra, NULL, 0, last_index);
258
259 kfree(ra);
260
261 return 0;
262}
263
Josef Bacika67509c2011-10-05 15:18:58 -0400264struct io_ctl {
265 void *cur, *orig;
266 struct page *page;
267 struct page **pages;
268 struct btrfs_root *root;
269 unsigned long size;
270 int index;
271 int num_pages;
Josef Bacik5b0e95b2011-10-06 08:58:24 -0400272 unsigned check_crcs:1;
Josef Bacika67509c2011-10-05 15:18:58 -0400273};
274
275static int io_ctl_init(struct io_ctl *io_ctl, struct inode *inode,
276 struct btrfs_root *root)
277{
278 memset(io_ctl, 0, sizeof(struct io_ctl));
279 io_ctl->num_pages = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >>
280 PAGE_CACHE_SHIFT;
281 io_ctl->pages = kzalloc(sizeof(struct page *) * io_ctl->num_pages,
282 GFP_NOFS);
283 if (!io_ctl->pages)
284 return -ENOMEM;
285 io_ctl->root = root;
Josef Bacik5b0e95b2011-10-06 08:58:24 -0400286 if (btrfs_ino(inode) != BTRFS_FREE_INO_OBJECTID)
287 io_ctl->check_crcs = 1;
Josef Bacika67509c2011-10-05 15:18:58 -0400288 return 0;
289}
290
291static void io_ctl_free(struct io_ctl *io_ctl)
292{
293 kfree(io_ctl->pages);
294}
295
296static void io_ctl_unmap_page(struct io_ctl *io_ctl)
297{
298 if (io_ctl->cur) {
299 kunmap(io_ctl->page);
300 io_ctl->cur = NULL;
301 io_ctl->orig = NULL;
302 }
303}
304
305static void io_ctl_map_page(struct io_ctl *io_ctl, int clear)
306{
307 WARN_ON(io_ctl->cur);
308 BUG_ON(io_ctl->index >= io_ctl->num_pages);
309 io_ctl->page = io_ctl->pages[io_ctl->index++];
310 io_ctl->cur = kmap(io_ctl->page);
311 io_ctl->orig = io_ctl->cur;
312 io_ctl->size = PAGE_CACHE_SIZE;
313 if (clear)
314 memset(io_ctl->cur, 0, PAGE_CACHE_SIZE);
315}
316
317static void io_ctl_drop_pages(struct io_ctl *io_ctl)
318{
319 int i;
320
321 io_ctl_unmap_page(io_ctl);
322
323 for (i = 0; i < io_ctl->num_pages; i++) {
Li Zefana1ee5a42012-01-09 14:27:42 +0800324 if (io_ctl->pages[i]) {
325 ClearPageChecked(io_ctl->pages[i]);
326 unlock_page(io_ctl->pages[i]);
327 page_cache_release(io_ctl->pages[i]);
328 }
Josef Bacika67509c2011-10-05 15:18:58 -0400329 }
330}
331
332static int io_ctl_prepare_pages(struct io_ctl *io_ctl, struct inode *inode,
333 int uptodate)
334{
335 struct page *page;
336 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
337 int i;
338
339 for (i = 0; i < io_ctl->num_pages; i++) {
340 page = find_or_create_page(inode->i_mapping, i, mask);
341 if (!page) {
342 io_ctl_drop_pages(io_ctl);
343 return -ENOMEM;
344 }
345 io_ctl->pages[i] = page;
346 if (uptodate && !PageUptodate(page)) {
347 btrfs_readpage(NULL, page);
348 lock_page(page);
349 if (!PageUptodate(page)) {
350 printk(KERN_ERR "btrfs: error reading free "
351 "space cache\n");
352 io_ctl_drop_pages(io_ctl);
353 return -EIO;
354 }
355 }
356 }
357
Josef Bacikf7d61dc2011-11-15 09:31:24 -0500358 for (i = 0; i < io_ctl->num_pages; i++) {
359 clear_page_dirty_for_io(io_ctl->pages[i]);
360 set_page_extent_mapped(io_ctl->pages[i]);
361 }
362
Josef Bacika67509c2011-10-05 15:18:58 -0400363 return 0;
364}
365
366static void io_ctl_set_generation(struct io_ctl *io_ctl, u64 generation)
367{
368 u64 *val;
369
370 io_ctl_map_page(io_ctl, 1);
371
372 /*
Josef Bacik5b0e95b2011-10-06 08:58:24 -0400373 * Skip the csum areas. If we don't check crcs then we just have a
374 * 64bit chunk at the front of the first page.
Josef Bacika67509c2011-10-05 15:18:58 -0400375 */
Josef Bacik5b0e95b2011-10-06 08:58:24 -0400376 if (io_ctl->check_crcs) {
377 io_ctl->cur += (sizeof(u32) * io_ctl->num_pages);
378 io_ctl->size -= sizeof(u64) + (sizeof(u32) * io_ctl->num_pages);
379 } else {
380 io_ctl->cur += sizeof(u64);
381 io_ctl->size -= sizeof(u64) * 2;
382 }
Josef Bacika67509c2011-10-05 15:18:58 -0400383
384 val = io_ctl->cur;
385 *val = cpu_to_le64(generation);
386 io_ctl->cur += sizeof(u64);
Josef Bacika67509c2011-10-05 15:18:58 -0400387}
388
389static int io_ctl_check_generation(struct io_ctl *io_ctl, u64 generation)
390{
391 u64 *gen;
392
Josef Bacik5b0e95b2011-10-06 08:58:24 -0400393 /*
394 * Skip the crc area. If we don't check crcs then we just have a 64bit
395 * chunk at the front of the first page.
396 */
397 if (io_ctl->check_crcs) {
398 io_ctl->cur += sizeof(u32) * io_ctl->num_pages;
399 io_ctl->size -= sizeof(u64) +
400 (sizeof(u32) * io_ctl->num_pages);
401 } else {
402 io_ctl->cur += sizeof(u64);
403 io_ctl->size -= sizeof(u64) * 2;
404 }
Josef Bacika67509c2011-10-05 15:18:58 -0400405
Josef Bacika67509c2011-10-05 15:18:58 -0400406 gen = io_ctl->cur;
407 if (le64_to_cpu(*gen) != generation) {
408 printk_ratelimited(KERN_ERR "btrfs: space cache generation "
409 "(%Lu) does not match inode (%Lu)\n", *gen,
410 generation);
411 io_ctl_unmap_page(io_ctl);
412 return -EIO;
413 }
414 io_ctl->cur += sizeof(u64);
Josef Bacik5b0e95b2011-10-06 08:58:24 -0400415 return 0;
416}
417
418static void io_ctl_set_crc(struct io_ctl *io_ctl, int index)
419{
420 u32 *tmp;
421 u32 crc = ~(u32)0;
422 unsigned offset = 0;
423
424 if (!io_ctl->check_crcs) {
425 io_ctl_unmap_page(io_ctl);
426 return;
427 }
428
429 if (index == 0)
Justin P. Mattockcb54f252011-11-21 08:43:28 -0800430 offset = sizeof(u32) * io_ctl->num_pages;
Josef Bacik5b0e95b2011-10-06 08:58:24 -0400431
432 crc = btrfs_csum_data(io_ctl->root, io_ctl->orig + offset, crc,
433 PAGE_CACHE_SIZE - offset);
434 btrfs_csum_final(crc, (char *)&crc);
435 io_ctl_unmap_page(io_ctl);
436 tmp = kmap(io_ctl->pages[0]);
437 tmp += index;
438 *tmp = crc;
439 kunmap(io_ctl->pages[0]);
440}
441
442static int io_ctl_check_crc(struct io_ctl *io_ctl, int index)
443{
444 u32 *tmp, val;
445 u32 crc = ~(u32)0;
446 unsigned offset = 0;
447
448 if (!io_ctl->check_crcs) {
449 io_ctl_map_page(io_ctl, 0);
450 return 0;
451 }
452
453 if (index == 0)
454 offset = sizeof(u32) * io_ctl->num_pages;
455
456 tmp = kmap(io_ctl->pages[0]);
457 tmp += index;
458 val = *tmp;
459 kunmap(io_ctl->pages[0]);
460
461 io_ctl_map_page(io_ctl, 0);
462 crc = btrfs_csum_data(io_ctl->root, io_ctl->orig + offset, crc,
463 PAGE_CACHE_SIZE - offset);
464 btrfs_csum_final(crc, (char *)&crc);
465 if (val != crc) {
466 printk_ratelimited(KERN_ERR "btrfs: csum mismatch on free "
467 "space cache\n");
468 io_ctl_unmap_page(io_ctl);
469 return -EIO;
470 }
471
Josef Bacika67509c2011-10-05 15:18:58 -0400472 return 0;
473}
474
475static int io_ctl_add_entry(struct io_ctl *io_ctl, u64 offset, u64 bytes,
476 void *bitmap)
477{
478 struct btrfs_free_space_entry *entry;
479
480 if (!io_ctl->cur)
481 return -ENOSPC;
482
483 entry = io_ctl->cur;
484 entry->offset = cpu_to_le64(offset);
485 entry->bytes = cpu_to_le64(bytes);
486 entry->type = (bitmap) ? BTRFS_FREE_SPACE_BITMAP :
487 BTRFS_FREE_SPACE_EXTENT;
488 io_ctl->cur += sizeof(struct btrfs_free_space_entry);
489 io_ctl->size -= sizeof(struct btrfs_free_space_entry);
490
491 if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
492 return 0;
493
Josef Bacik5b0e95b2011-10-06 08:58:24 -0400494 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
Josef Bacika67509c2011-10-05 15:18:58 -0400495
496 /* No more pages to map */
497 if (io_ctl->index >= io_ctl->num_pages)
498 return 0;
499
500 /* map the next page */
501 io_ctl_map_page(io_ctl, 1);
502 return 0;
503}
504
505static int io_ctl_add_bitmap(struct io_ctl *io_ctl, void *bitmap)
506{
507 if (!io_ctl->cur)
508 return -ENOSPC;
509
510 /*
511 * If we aren't at the start of the current page, unmap this one and
512 * map the next one if there is any left.
513 */
514 if (io_ctl->cur != io_ctl->orig) {
Josef Bacik5b0e95b2011-10-06 08:58:24 -0400515 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
Josef Bacika67509c2011-10-05 15:18:58 -0400516 if (io_ctl->index >= io_ctl->num_pages)
517 return -ENOSPC;
518 io_ctl_map_page(io_ctl, 0);
519 }
520
521 memcpy(io_ctl->cur, bitmap, PAGE_CACHE_SIZE);
Josef Bacik5b0e95b2011-10-06 08:58:24 -0400522 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
Josef Bacika67509c2011-10-05 15:18:58 -0400523 if (io_ctl->index < io_ctl->num_pages)
524 io_ctl_map_page(io_ctl, 0);
525 return 0;
526}
527
528static void io_ctl_zero_remaining_pages(struct io_ctl *io_ctl)
529{
Josef Bacik5b0e95b2011-10-06 08:58:24 -0400530 /*
531 * If we're not on the boundary we know we've modified the page and we
532 * need to crc the page.
533 */
534 if (io_ctl->cur != io_ctl->orig)
535 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
536 else
537 io_ctl_unmap_page(io_ctl);
Josef Bacika67509c2011-10-05 15:18:58 -0400538
539 while (io_ctl->index < io_ctl->num_pages) {
540 io_ctl_map_page(io_ctl, 1);
Josef Bacik5b0e95b2011-10-06 08:58:24 -0400541 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
Josef Bacika67509c2011-10-05 15:18:58 -0400542 }
543}
544
Josef Bacik5b0e95b2011-10-06 08:58:24 -0400545static int io_ctl_read_entry(struct io_ctl *io_ctl,
546 struct btrfs_free_space *entry, u8 *type)
Josef Bacika67509c2011-10-05 15:18:58 -0400547{
548 struct btrfs_free_space_entry *e;
Josef Bacik2f120c02011-11-10 20:45:05 -0500549 int ret;
550
551 if (!io_ctl->cur) {
552 ret = io_ctl_check_crc(io_ctl, io_ctl->index);
553 if (ret)
554 return ret;
555 }
Josef Bacika67509c2011-10-05 15:18:58 -0400556
557 e = io_ctl->cur;
558 entry->offset = le64_to_cpu(e->offset);
559 entry->bytes = le64_to_cpu(e->bytes);
Josef Bacik5b0e95b2011-10-06 08:58:24 -0400560 *type = e->type;
Josef Bacika67509c2011-10-05 15:18:58 -0400561 io_ctl->cur += sizeof(struct btrfs_free_space_entry);
562 io_ctl->size -= sizeof(struct btrfs_free_space_entry);
563
564 if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
Josef Bacik5b0e95b2011-10-06 08:58:24 -0400565 return 0;
Josef Bacika67509c2011-10-05 15:18:58 -0400566
567 io_ctl_unmap_page(io_ctl);
568
Josef Bacik2f120c02011-11-10 20:45:05 -0500569 return 0;
Josef Bacika67509c2011-10-05 15:18:58 -0400570}
571
Josef Bacik5b0e95b2011-10-06 08:58:24 -0400572static int io_ctl_read_bitmap(struct io_ctl *io_ctl,
573 struct btrfs_free_space *entry)
Josef Bacika67509c2011-10-05 15:18:58 -0400574{
Josef Bacik5b0e95b2011-10-06 08:58:24 -0400575 int ret;
576
Josef Bacik5b0e95b2011-10-06 08:58:24 -0400577 ret = io_ctl_check_crc(io_ctl, io_ctl->index);
578 if (ret)
579 return ret;
580
Josef Bacika67509c2011-10-05 15:18:58 -0400581 memcpy(entry->bitmap, io_ctl->cur, PAGE_CACHE_SIZE);
582 io_ctl_unmap_page(io_ctl);
Josef Bacik5b0e95b2011-10-06 08:58:24 -0400583
584 return 0;
Josef Bacika67509c2011-10-05 15:18:58 -0400585}
586
Li Zefan0414efa2011-04-20 10:20:14 +0800587int __load_free_space_cache(struct btrfs_root *root, struct inode *inode,
588 struct btrfs_free_space_ctl *ctl,
589 struct btrfs_path *path, u64 offset)
Josef Bacik9d66e232010-08-25 16:54:15 -0400590{
Josef Bacik9d66e232010-08-25 16:54:15 -0400591 struct btrfs_free_space_header *header;
592 struct extent_buffer *leaf;
Josef Bacika67509c2011-10-05 15:18:58 -0400593 struct io_ctl io_ctl;
Josef Bacik9d66e232010-08-25 16:54:15 -0400594 struct btrfs_key key;
Josef Bacika67509c2011-10-05 15:18:58 -0400595 struct btrfs_free_space *e, *n;
Josef Bacik9d66e232010-08-25 16:54:15 -0400596 struct list_head bitmaps;
597 u64 num_entries;
598 u64 num_bitmaps;
599 u64 generation;
Josef Bacika67509c2011-10-05 15:18:58 -0400600 u8 type;
Josef Bacikf6a39822011-06-06 10:50:35 -0400601 int ret = 0;
Josef Bacik9d66e232010-08-25 16:54:15 -0400602
603 INIT_LIST_HEAD(&bitmaps);
604
Josef Bacik9d66e232010-08-25 16:54:15 -0400605 /* Nothing in the space cache, goodbye */
Li Zefan0414efa2011-04-20 10:20:14 +0800606 if (!i_size_read(inode))
Josef Bacika67509c2011-10-05 15:18:58 -0400607 return 0;
Josef Bacik9d66e232010-08-25 16:54:15 -0400608
609 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
Li Zefan0414efa2011-04-20 10:20:14 +0800610 key.offset = offset;
Josef Bacik9d66e232010-08-25 16:54:15 -0400611 key.type = 0;
612
613 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
Li Zefan0414efa2011-04-20 10:20:14 +0800614 if (ret < 0)
Josef Bacika67509c2011-10-05 15:18:58 -0400615 return 0;
Li Zefan0414efa2011-04-20 10:20:14 +0800616 else if (ret > 0) {
Chris Mason945d8962011-05-22 12:33:42 -0400617 btrfs_release_path(path);
Josef Bacika67509c2011-10-05 15:18:58 -0400618 return 0;
Josef Bacik9d66e232010-08-25 16:54:15 -0400619 }
620
Li Zefan0414efa2011-04-20 10:20:14 +0800621 ret = -1;
622
Josef Bacik9d66e232010-08-25 16:54:15 -0400623 leaf = path->nodes[0];
624 header = btrfs_item_ptr(leaf, path->slots[0],
625 struct btrfs_free_space_header);
626 num_entries = btrfs_free_space_entries(leaf, header);
627 num_bitmaps = btrfs_free_space_bitmaps(leaf, header);
628 generation = btrfs_free_space_generation(leaf, header);
Chris Mason945d8962011-05-22 12:33:42 -0400629 btrfs_release_path(path);
Josef Bacik9d66e232010-08-25 16:54:15 -0400630
631 if (BTRFS_I(inode)->generation != generation) {
632 printk(KERN_ERR "btrfs: free space inode generation (%llu) did"
Li Zefan0414efa2011-04-20 10:20:14 +0800633 " not match free space cache generation (%llu)\n",
Josef Bacik9d66e232010-08-25 16:54:15 -0400634 (unsigned long long)BTRFS_I(inode)->generation,
Li Zefan0414efa2011-04-20 10:20:14 +0800635 (unsigned long long)generation);
Josef Bacika67509c2011-10-05 15:18:58 -0400636 return 0;
Josef Bacik9d66e232010-08-25 16:54:15 -0400637 }
638
639 if (!num_entries)
Josef Bacika67509c2011-10-05 15:18:58 -0400640 return 0;
Josef Bacik9d66e232010-08-25 16:54:15 -0400641
Li Zefan706efc62012-01-09 14:36:28 +0800642 ret = io_ctl_init(&io_ctl, inode, root);
643 if (ret)
644 return ret;
645
Josef Bacik9d66e232010-08-25 16:54:15 -0400646 ret = readahead_cache(inode);
Li Zefan0414efa2011-04-20 10:20:14 +0800647 if (ret)
Josef Bacik9d66e232010-08-25 16:54:15 -0400648 goto out;
Josef Bacik9d66e232010-08-25 16:54:15 -0400649
Josef Bacika67509c2011-10-05 15:18:58 -0400650 ret = io_ctl_prepare_pages(&io_ctl, inode, 1);
651 if (ret)
652 goto out;
Josef Bacik9d66e232010-08-25 16:54:15 -0400653
Josef Bacik5b0e95b2011-10-06 08:58:24 -0400654 ret = io_ctl_check_crc(&io_ctl, 0);
655 if (ret)
656 goto free_cache;
657
Josef Bacika67509c2011-10-05 15:18:58 -0400658 ret = io_ctl_check_generation(&io_ctl, generation);
659 if (ret)
660 goto free_cache;
Josef Bacik9d66e232010-08-25 16:54:15 -0400661
Josef Bacika67509c2011-10-05 15:18:58 -0400662 while (num_entries) {
663 e = kmem_cache_zalloc(btrfs_free_space_cachep,
664 GFP_NOFS);
665 if (!e)
Josef Bacik9d66e232010-08-25 16:54:15 -0400666 goto free_cache;
Josef Bacik9d66e232010-08-25 16:54:15 -0400667
Josef Bacik5b0e95b2011-10-06 08:58:24 -0400668 ret = io_ctl_read_entry(&io_ctl, e, &type);
669 if (ret) {
670 kmem_cache_free(btrfs_free_space_cachep, e);
671 goto free_cache;
672 }
673
Josef Bacika67509c2011-10-05 15:18:58 -0400674 if (!e->bytes) {
675 kmem_cache_free(btrfs_free_space_cachep, e);
676 goto free_cache;
Josef Bacik9d66e232010-08-25 16:54:15 -0400677 }
Josef Bacik9d66e232010-08-25 16:54:15 -0400678
Josef Bacika67509c2011-10-05 15:18:58 -0400679 if (type == BTRFS_FREE_SPACE_EXTENT) {
680 spin_lock(&ctl->tree_lock);
681 ret = link_free_space(ctl, e);
682 spin_unlock(&ctl->tree_lock);
683 if (ret) {
684 printk(KERN_ERR "Duplicate entries in "
685 "free space cache, dumping\n");
Josef Bacikdc89e982011-01-28 17:05:48 -0500686 kmem_cache_free(btrfs_free_space_cachep, e);
Josef Bacik9d66e232010-08-25 16:54:15 -0400687 goto free_cache;
688 }
Josef Bacika67509c2011-10-05 15:18:58 -0400689 } else {
690 BUG_ON(!num_bitmaps);
691 num_bitmaps--;
692 e->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
693 if (!e->bitmap) {
694 kmem_cache_free(
695 btrfs_free_space_cachep, e);
696 goto free_cache;
Josef Bacik9d66e232010-08-25 16:54:15 -0400697 }
Josef Bacika67509c2011-10-05 15:18:58 -0400698 spin_lock(&ctl->tree_lock);
699 ret = link_free_space(ctl, e);
700 ctl->total_bitmaps++;
701 ctl->op->recalc_thresholds(ctl);
702 spin_unlock(&ctl->tree_lock);
703 if (ret) {
704 printk(KERN_ERR "Duplicate entries in "
705 "free space cache, dumping\n");
706 kmem_cache_free(btrfs_free_space_cachep, e);
707 goto free_cache;
708 }
709 list_add_tail(&e->list, &bitmaps);
Josef Bacik9d66e232010-08-25 16:54:15 -0400710 }
711
Josef Bacika67509c2011-10-05 15:18:58 -0400712 num_entries--;
Josef Bacik9d66e232010-08-25 16:54:15 -0400713 }
714
Josef Bacik2f120c02011-11-10 20:45:05 -0500715 io_ctl_unmap_page(&io_ctl);
716
Josef Bacika67509c2011-10-05 15:18:58 -0400717 /*
718 * We add the bitmaps at the end of the entries in order that
719 * the bitmap entries are added to the cache.
720 */
721 list_for_each_entry_safe(e, n, &bitmaps, list) {
722 list_del_init(&e->list);
Josef Bacik5b0e95b2011-10-06 08:58:24 -0400723 ret = io_ctl_read_bitmap(&io_ctl, e);
724 if (ret)
725 goto free_cache;
Josef Bacika67509c2011-10-05 15:18:58 -0400726 }
727
728 io_ctl_drop_pages(&io_ctl);
Josef Bacik9d66e232010-08-25 16:54:15 -0400729 ret = 1;
730out:
Josef Bacika67509c2011-10-05 15:18:58 -0400731 io_ctl_free(&io_ctl);
Josef Bacik9d66e232010-08-25 16:54:15 -0400732 return ret;
Josef Bacik9d66e232010-08-25 16:54:15 -0400733free_cache:
Josef Bacika67509c2011-10-05 15:18:58 -0400734 io_ctl_drop_pages(&io_ctl);
Li Zefan0414efa2011-04-20 10:20:14 +0800735 __btrfs_remove_free_space_cache(ctl);
Josef Bacik9d66e232010-08-25 16:54:15 -0400736 goto out;
737}
738
Li Zefan0414efa2011-04-20 10:20:14 +0800739int load_free_space_cache(struct btrfs_fs_info *fs_info,
740 struct btrfs_block_group_cache *block_group)
Josef Bacik0cb59c92010-07-02 12:14:14 -0400741{
Li Zefan34d52cb2011-03-29 13:46:06 +0800742 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
Li Zefan0414efa2011-04-20 10:20:14 +0800743 struct btrfs_root *root = fs_info->tree_root;
744 struct inode *inode;
745 struct btrfs_path *path;
Josef Bacik5b0e95b2011-10-06 08:58:24 -0400746 int ret = 0;
Li Zefan0414efa2011-04-20 10:20:14 +0800747 bool matched;
748 u64 used = btrfs_block_group_used(&block_group->item);
749
750 /*
Li Zefan0414efa2011-04-20 10:20:14 +0800751 * If this block group has been marked to be cleared for one reason or
752 * another then we can't trust the on disk cache, so just return.
753 */
754 spin_lock(&block_group->lock);
755 if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
756 spin_unlock(&block_group->lock);
757 return 0;
758 }
759 spin_unlock(&block_group->lock);
760
761 path = btrfs_alloc_path();
762 if (!path)
763 return 0;
Josef Bacikd53ba472012-04-12 16:03:57 -0400764 path->search_commit_root = 1;
765 path->skip_locking = 1;
Li Zefan0414efa2011-04-20 10:20:14 +0800766
767 inode = lookup_free_space_inode(root, block_group, path);
768 if (IS_ERR(inode)) {
769 btrfs_free_path(path);
770 return 0;
771 }
772
Josef Bacik5b0e95b2011-10-06 08:58:24 -0400773 /* We may have converted the inode and made the cache invalid. */
774 spin_lock(&block_group->lock);
775 if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
776 spin_unlock(&block_group->lock);
Tsutomu Itoha7e221e2012-02-14 17:12:23 +0900777 btrfs_free_path(path);
Josef Bacik5b0e95b2011-10-06 08:58:24 -0400778 goto out;
779 }
780 spin_unlock(&block_group->lock);
781
Li Zefan0414efa2011-04-20 10:20:14 +0800782 ret = __load_free_space_cache(fs_info->tree_root, inode, ctl,
783 path, block_group->key.objectid);
784 btrfs_free_path(path);
785 if (ret <= 0)
786 goto out;
787
788 spin_lock(&ctl->tree_lock);
789 matched = (ctl->free_space == (block_group->key.offset - used -
790 block_group->bytes_super));
791 spin_unlock(&ctl->tree_lock);
792
793 if (!matched) {
794 __btrfs_remove_free_space_cache(ctl);
795 printk(KERN_ERR "block group %llu has an wrong amount of free "
796 "space\n", block_group->key.objectid);
797 ret = -1;
798 }
799out:
800 if (ret < 0) {
801 /* This cache is bogus, make sure it gets cleared */
802 spin_lock(&block_group->lock);
803 block_group->disk_cache_state = BTRFS_DC_CLEAR;
804 spin_unlock(&block_group->lock);
Li Zefan82d59022011-04-20 10:33:24 +0800805 ret = 0;
Li Zefan0414efa2011-04-20 10:20:14 +0800806
807 printk(KERN_ERR "btrfs: failed to load free space cache "
808 "for block group %llu\n", block_group->key.objectid);
809 }
810
811 iput(inode);
812 return ret;
813}
814
Josef Bacikc09544e2011-08-30 10:19:10 -0400815/**
816 * __btrfs_write_out_cache - write out cached info to an inode
817 * @root - the root the inode belongs to
818 * @ctl - the free space cache we are going to write out
819 * @block_group - the block_group for this cache if it belongs to a block_group
820 * @trans - the trans handle
821 * @path - the path to use
822 * @offset - the offset for the key we'll insert
823 *
824 * This function writes out a free space cache struct to disk for quick recovery
825 * on mount. This will return 0 if it was successfull in writing the cache out,
826 * and -1 if it was not.
827 */
Li Zefan0414efa2011-04-20 10:20:14 +0800828int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode,
829 struct btrfs_free_space_ctl *ctl,
830 struct btrfs_block_group_cache *block_group,
831 struct btrfs_trans_handle *trans,
832 struct btrfs_path *path, u64 offset)
Josef Bacik0cb59c92010-07-02 12:14:14 -0400833{
834 struct btrfs_free_space_header *header;
835 struct extent_buffer *leaf;
Josef Bacik0cb59c92010-07-02 12:14:14 -0400836 struct rb_node *node;
837 struct list_head *pos, *n;
Josef Bacik0cb59c92010-07-02 12:14:14 -0400838 struct extent_state *cached_state = NULL;
Josef Bacik43be2142011-04-01 14:55:00 +0000839 struct btrfs_free_cluster *cluster = NULL;
840 struct extent_io_tree *unpin = NULL;
Josef Bacika67509c2011-10-05 15:18:58 -0400841 struct io_ctl io_ctl;
Josef Bacik0cb59c92010-07-02 12:14:14 -0400842 struct list_head bitmap_list;
843 struct btrfs_key key;
Li Zefandb804f22012-01-10 16:41:01 +0800844 u64 start, extent_start, extent_end, len;
Josef Bacik0cb59c92010-07-02 12:14:14 -0400845 int entries = 0;
846 int bitmaps = 0;
Josef Bacikc09544e2011-08-30 10:19:10 -0400847 int ret;
848 int err = -1;
Josef Bacik0cb59c92010-07-02 12:14:14 -0400849
Josef Bacik0cb59c92010-07-02 12:14:14 -0400850 INIT_LIST_HEAD(&bitmap_list);
851
Li Zefan0414efa2011-04-20 10:20:14 +0800852 if (!i_size_read(inode))
853 return -1;
Josef Bacik2b209822010-12-03 13:17:53 -0500854
Li Zefan706efc62012-01-09 14:36:28 +0800855 ret = io_ctl_init(&io_ctl, inode, root);
856 if (ret)
857 return -1;
Josef Bacikbe1a12a2011-04-06 13:05:22 -0400858
Josef Bacik43be2142011-04-01 14:55:00 +0000859 /* Get the cluster for this block_group if it exists */
Li Zefan0414efa2011-04-20 10:20:14 +0800860 if (block_group && !list_empty(&block_group->cluster_list))
Josef Bacik43be2142011-04-01 14:55:00 +0000861 cluster = list_entry(block_group->cluster_list.next,
862 struct btrfs_free_cluster,
863 block_group_list);
864
Josef Bacika67509c2011-10-05 15:18:58 -0400865 /* Lock all pages first so we can lock the extent safely. */
866 io_ctl_prepare_pages(&io_ctl, inode, 0);
Josef Bacik0cb59c92010-07-02 12:14:14 -0400867
Josef Bacik0cb59c92010-07-02 12:14:14 -0400868 lock_extent_bits(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1,
Jeff Mahoneyd0082372012-03-01 14:57:19 +0100869 0, &cached_state);
Josef Bacik0cb59c92010-07-02 12:14:14 -0400870
Josef Bacikf75b1302011-10-05 10:00:18 -0400871 node = rb_first(&ctl->free_space_offset);
872 if (!node && cluster) {
873 node = rb_first(&cluster->root);
874 cluster = NULL;
875 }
876
Josef Bacik5b0e95b2011-10-06 08:58:24 -0400877 /* Make sure we can fit our crcs into the first page */
878 if (io_ctl.check_crcs &&
879 (io_ctl.num_pages * sizeof(u32)) >= PAGE_CACHE_SIZE) {
880 WARN_ON(1);
881 goto out_nospc;
882 }
883
Josef Bacika67509c2011-10-05 15:18:58 -0400884 io_ctl_set_generation(&io_ctl, trans->transid);
885
Josef Bacik0cb59c92010-07-02 12:14:14 -0400886 /* Write out the extent entries */
Josef Bacika67509c2011-10-05 15:18:58 -0400887 while (node) {
888 struct btrfs_free_space *e;
Josef Bacik0cb59c92010-07-02 12:14:14 -0400889
Josef Bacika67509c2011-10-05 15:18:58 -0400890 e = rb_entry(node, struct btrfs_free_space, offset_index);
891 entries++;
Josef Bacik43be2142011-04-01 14:55:00 +0000892
Josef Bacika67509c2011-10-05 15:18:58 -0400893 ret = io_ctl_add_entry(&io_ctl, e->offset, e->bytes,
894 e->bitmap);
895 if (ret)
896 goto out_nospc;
897
898 if (e->bitmap) {
899 list_add_tail(&e->list, &bitmap_list);
900 bitmaps++;
901 }
902 node = rb_next(node);
903 if (!node && cluster) {
904 node = rb_first(&cluster->root);
905 cluster = NULL;
906 }
907 }
908
909 /*
910 * We want to add any pinned extents to our free space cache
911 * so we don't leak the space
912 */
Li Zefandb804f22012-01-10 16:41:01 +0800913
914 /*
915 * We shouldn't have switched the pinned extents yet so this is the
916 * right one
917 */
918 unpin = root->fs_info->pinned_extents;
919
920 if (block_group)
921 start = block_group->key.objectid;
922
Josef Bacika67509c2011-10-05 15:18:58 -0400923 while (block_group && (start < block_group->key.objectid +
924 block_group->key.offset)) {
Li Zefandb804f22012-01-10 16:41:01 +0800925 ret = find_first_extent_bit(unpin, start,
926 &extent_start, &extent_end,
Josef Bacika67509c2011-10-05 15:18:58 -0400927 EXTENT_DIRTY);
928 if (ret) {
929 ret = 0;
Josef Bacikbe1a12a2011-04-06 13:05:22 -0400930 break;
931 }
932
Josef Bacika67509c2011-10-05 15:18:58 -0400933 /* This pinned extent is out of our range */
Li Zefandb804f22012-01-10 16:41:01 +0800934 if (extent_start >= block_group->key.objectid +
Josef Bacika67509c2011-10-05 15:18:58 -0400935 block_group->key.offset)
936 break;
Josef Bacik0cb59c92010-07-02 12:14:14 -0400937
Li Zefandb804f22012-01-10 16:41:01 +0800938 extent_start = max(extent_start, start);
939 extent_end = min(block_group->key.objectid +
940 block_group->key.offset, extent_end + 1);
941 len = extent_end - extent_start;
Josef Bacik0cb59c92010-07-02 12:14:14 -0400942
Josef Bacika67509c2011-10-05 15:18:58 -0400943 entries++;
Li Zefandb804f22012-01-10 16:41:01 +0800944 ret = io_ctl_add_entry(&io_ctl, extent_start, len, NULL);
Josef Bacika67509c2011-10-05 15:18:58 -0400945 if (ret)
946 goto out_nospc;
Josef Bacik2f356122011-06-10 15:31:13 -0400947
Li Zefandb804f22012-01-10 16:41:01 +0800948 start = extent_end;
Josef Bacika67509c2011-10-05 15:18:58 -0400949 }
Josef Bacik0cb59c92010-07-02 12:14:14 -0400950
951 /* Write out the bitmaps */
952 list_for_each_safe(pos, n, &bitmap_list) {
Josef Bacik0cb59c92010-07-02 12:14:14 -0400953 struct btrfs_free_space *entry =
954 list_entry(pos, struct btrfs_free_space, list);
955
Josef Bacika67509c2011-10-05 15:18:58 -0400956 ret = io_ctl_add_bitmap(&io_ctl, entry->bitmap);
957 if (ret)
958 goto out_nospc;
Josef Bacik0cb59c92010-07-02 12:14:14 -0400959 list_del_init(&entry->list);
Josef Bacikbe1a12a2011-04-06 13:05:22 -0400960 }
961
Josef Bacik0cb59c92010-07-02 12:14:14 -0400962 /* Zero out the rest of the pages just to make sure */
Josef Bacika67509c2011-10-05 15:18:58 -0400963 io_ctl_zero_remaining_pages(&io_ctl);
Josef Bacik0cb59c92010-07-02 12:14:14 -0400964
Josef Bacika67509c2011-10-05 15:18:58 -0400965 ret = btrfs_dirty_pages(root, inode, io_ctl.pages, io_ctl.num_pages,
966 0, i_size_read(inode), &cached_state);
967 io_ctl_drop_pages(&io_ctl);
Josef Bacik0cb59c92010-07-02 12:14:14 -0400968 unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
969 i_size_read(inode) - 1, &cached_state, GFP_NOFS);
970
Josef Bacikc09544e2011-08-30 10:19:10 -0400971 if (ret)
Josef Bacik2f356122011-06-10 15:31:13 -0400972 goto out;
Josef Bacikbe1a12a2011-04-06 13:05:22 -0400973
Josef Bacikbe1a12a2011-04-06 13:05:22 -0400974
Josef Bacik549b4fd2011-10-05 16:33:53 -0400975 ret = filemap_write_and_wait(inode->i_mapping);
976 if (ret)
977 goto out;
Josef Bacik0cb59c92010-07-02 12:14:14 -0400978
979 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
Li Zefan0414efa2011-04-20 10:20:14 +0800980 key.offset = offset;
Josef Bacik0cb59c92010-07-02 12:14:14 -0400981 key.type = 0;
982
Josef Bacika9b5fcd2011-08-19 12:06:12 -0400983 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
Josef Bacik0cb59c92010-07-02 12:14:14 -0400984 if (ret < 0) {
Josef Bacika67509c2011-10-05 15:18:58 -0400985 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1,
Josef Bacik5b0e95b2011-10-06 08:58:24 -0400986 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, NULL,
987 GFP_NOFS);
Josef Bacik2f356122011-06-10 15:31:13 -0400988 goto out;
Josef Bacik0cb59c92010-07-02 12:14:14 -0400989 }
990 leaf = path->nodes[0];
991 if (ret > 0) {
992 struct btrfs_key found_key;
993 BUG_ON(!path->slots[0]);
994 path->slots[0]--;
995 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
996 if (found_key.objectid != BTRFS_FREE_SPACE_OBJECTID ||
Li Zefan0414efa2011-04-20 10:20:14 +0800997 found_key.offset != offset) {
Josef Bacika67509c2011-10-05 15:18:58 -0400998 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0,
999 inode->i_size - 1,
Josef Bacik5b0e95b2011-10-06 08:58:24 -04001000 EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0,
1001 NULL, GFP_NOFS);
David Sterbab3b4aa72011-04-21 01:20:15 +02001002 btrfs_release_path(path);
Josef Bacik2f356122011-06-10 15:31:13 -04001003 goto out;
Josef Bacik0cb59c92010-07-02 12:14:14 -04001004 }
1005 }
Josef Bacik549b4fd2011-10-05 16:33:53 -04001006
1007 BTRFS_I(inode)->generation = trans->transid;
Josef Bacik0cb59c92010-07-02 12:14:14 -04001008 header = btrfs_item_ptr(leaf, path->slots[0],
1009 struct btrfs_free_space_header);
1010 btrfs_set_free_space_entries(leaf, header, entries);
1011 btrfs_set_free_space_bitmaps(leaf, header, bitmaps);
1012 btrfs_set_free_space_generation(leaf, header, trans->transid);
1013 btrfs_mark_buffer_dirty(leaf);
David Sterbab3b4aa72011-04-21 01:20:15 +02001014 btrfs_release_path(path);
Josef Bacik0cb59c92010-07-02 12:14:14 -04001015
Josef Bacikc09544e2011-08-30 10:19:10 -04001016 err = 0;
Josef Bacik2f356122011-06-10 15:31:13 -04001017out:
Josef Bacika67509c2011-10-05 15:18:58 -04001018 io_ctl_free(&io_ctl);
Josef Bacikc09544e2011-08-30 10:19:10 -04001019 if (err) {
Josef Bacika67509c2011-10-05 15:18:58 -04001020 invalidate_inode_pages2(inode->i_mapping);
Josef Bacik0cb59c92010-07-02 12:14:14 -04001021 BTRFS_I(inode)->generation = 0;
1022 }
Josef Bacik0cb59c92010-07-02 12:14:14 -04001023 btrfs_update_inode(trans, root, inode);
Josef Bacikc09544e2011-08-30 10:19:10 -04001024 return err;
Josef Bacika67509c2011-10-05 15:18:58 -04001025
1026out_nospc:
1027 list_for_each_safe(pos, n, &bitmap_list) {
1028 struct btrfs_free_space *entry =
1029 list_entry(pos, struct btrfs_free_space, list);
1030 list_del_init(&entry->list);
1031 }
1032 io_ctl_drop_pages(&io_ctl);
1033 unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
1034 i_size_read(inode) - 1, &cached_state, GFP_NOFS);
1035 goto out;
Li Zefan0414efa2011-04-20 10:20:14 +08001036}
1037
1038int btrfs_write_out_cache(struct btrfs_root *root,
1039 struct btrfs_trans_handle *trans,
1040 struct btrfs_block_group_cache *block_group,
1041 struct btrfs_path *path)
1042{
1043 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1044 struct inode *inode;
1045 int ret = 0;
1046
1047 root = root->fs_info->tree_root;
1048
1049 spin_lock(&block_group->lock);
1050 if (block_group->disk_cache_state < BTRFS_DC_SETUP) {
1051 spin_unlock(&block_group->lock);
1052 return 0;
1053 }
1054 spin_unlock(&block_group->lock);
1055
1056 inode = lookup_free_space_inode(root, block_group, path);
1057 if (IS_ERR(inode))
1058 return 0;
1059
1060 ret = __btrfs_write_out_cache(root, inode, ctl, block_group, trans,
1061 path, block_group->key.objectid);
Josef Bacikc09544e2011-08-30 10:19:10 -04001062 if (ret) {
Li Zefan0414efa2011-04-20 10:20:14 +08001063 spin_lock(&block_group->lock);
1064 block_group->disk_cache_state = BTRFS_DC_ERROR;
1065 spin_unlock(&block_group->lock);
Li Zefan82d59022011-04-20 10:33:24 +08001066 ret = 0;
Josef Bacikc09544e2011-08-30 10:19:10 -04001067#ifdef DEBUG
Masanari Iida934e7d42012-02-07 22:21:45 +09001068 printk(KERN_ERR "btrfs: failed to write free space cache "
Li Zefan0414efa2011-04-20 10:20:14 +08001069 "for block group %llu\n", block_group->key.objectid);
Josef Bacikc09544e2011-08-30 10:19:10 -04001070#endif
Li Zefan0414efa2011-04-20 10:20:14 +08001071 }
1072
Josef Bacik0cb59c92010-07-02 12:14:14 -04001073 iput(inode);
1074 return ret;
1075}
1076
Li Zefan34d52cb2011-03-29 13:46:06 +08001077static inline unsigned long offset_to_bit(u64 bitmap_start, u32 unit,
Josef Bacik96303082009-07-13 21:29:25 -04001078 u64 offset)
1079{
1080 BUG_ON(offset < bitmap_start);
1081 offset -= bitmap_start;
Li Zefan34d52cb2011-03-29 13:46:06 +08001082 return (unsigned long)(div_u64(offset, unit));
Josef Bacik96303082009-07-13 21:29:25 -04001083}
1084
Li Zefan34d52cb2011-03-29 13:46:06 +08001085static inline unsigned long bytes_to_bits(u64 bytes, u32 unit)
Josef Bacik96303082009-07-13 21:29:25 -04001086{
Li Zefan34d52cb2011-03-29 13:46:06 +08001087 return (unsigned long)(div_u64(bytes, unit));
Josef Bacik96303082009-07-13 21:29:25 -04001088}
1089
Li Zefan34d52cb2011-03-29 13:46:06 +08001090static inline u64 offset_to_bitmap(struct btrfs_free_space_ctl *ctl,
Josef Bacik96303082009-07-13 21:29:25 -04001091 u64 offset)
1092{
1093 u64 bitmap_start;
1094 u64 bytes_per_bitmap;
1095
Li Zefan34d52cb2011-03-29 13:46:06 +08001096 bytes_per_bitmap = BITS_PER_BITMAP * ctl->unit;
1097 bitmap_start = offset - ctl->start;
Josef Bacik96303082009-07-13 21:29:25 -04001098 bitmap_start = div64_u64(bitmap_start, bytes_per_bitmap);
1099 bitmap_start *= bytes_per_bitmap;
Li Zefan34d52cb2011-03-29 13:46:06 +08001100 bitmap_start += ctl->start;
Josef Bacik96303082009-07-13 21:29:25 -04001101
1102 return bitmap_start;
1103}
Josef Bacik0f9dd462008-09-23 13:14:11 -04001104
1105static int tree_insert_offset(struct rb_root *root, u64 offset,
Josef Bacik96303082009-07-13 21:29:25 -04001106 struct rb_node *node, int bitmap)
Josef Bacik0f9dd462008-09-23 13:14:11 -04001107{
1108 struct rb_node **p = &root->rb_node;
1109 struct rb_node *parent = NULL;
1110 struct btrfs_free_space *info;
1111
1112 while (*p) {
1113 parent = *p;
1114 info = rb_entry(parent, struct btrfs_free_space, offset_index);
1115
Josef Bacik96303082009-07-13 21:29:25 -04001116 if (offset < info->offset) {
Josef Bacik0f9dd462008-09-23 13:14:11 -04001117 p = &(*p)->rb_left;
Josef Bacik96303082009-07-13 21:29:25 -04001118 } else if (offset > info->offset) {
Josef Bacik0f9dd462008-09-23 13:14:11 -04001119 p = &(*p)->rb_right;
Josef Bacik96303082009-07-13 21:29:25 -04001120 } else {
1121 /*
1122 * we could have a bitmap entry and an extent entry
1123 * share the same offset. If this is the case, we want
1124 * the extent entry to always be found first if we do a
1125 * linear search through the tree, since we want to have
1126 * the quickest allocation time, and allocating from an
1127 * extent is faster than allocating from a bitmap. So
1128 * if we're inserting a bitmap and we find an entry at
1129 * this offset, we want to go right, or after this entry
1130 * logically. If we are inserting an extent and we've
1131 * found a bitmap, we want to go left, or before
1132 * logically.
1133 */
1134 if (bitmap) {
Josef Bacik207dde82011-05-13 14:49:23 -04001135 if (info->bitmap) {
1136 WARN_ON_ONCE(1);
1137 return -EEXIST;
1138 }
Josef Bacik96303082009-07-13 21:29:25 -04001139 p = &(*p)->rb_right;
1140 } else {
Josef Bacik207dde82011-05-13 14:49:23 -04001141 if (!info->bitmap) {
1142 WARN_ON_ONCE(1);
1143 return -EEXIST;
1144 }
Josef Bacik96303082009-07-13 21:29:25 -04001145 p = &(*p)->rb_left;
1146 }
1147 }
Josef Bacik0f9dd462008-09-23 13:14:11 -04001148 }
1149
1150 rb_link_node(node, parent, p);
1151 rb_insert_color(node, root);
1152
1153 return 0;
1154}
1155
1156/*
Josef Bacik70cb0742009-04-03 10:14:19 -04001157 * searches the tree for the given offset.
1158 *
Josef Bacik96303082009-07-13 21:29:25 -04001159 * fuzzy - If this is set, then we are trying to make an allocation, and we just
1160 * want a section that has at least bytes size and comes at or after the given
1161 * offset.
Josef Bacik0f9dd462008-09-23 13:14:11 -04001162 */
Josef Bacik96303082009-07-13 21:29:25 -04001163static struct btrfs_free_space *
Li Zefan34d52cb2011-03-29 13:46:06 +08001164tree_search_offset(struct btrfs_free_space_ctl *ctl,
Josef Bacik96303082009-07-13 21:29:25 -04001165 u64 offset, int bitmap_only, int fuzzy)
Josef Bacik0f9dd462008-09-23 13:14:11 -04001166{
Li Zefan34d52cb2011-03-29 13:46:06 +08001167 struct rb_node *n = ctl->free_space_offset.rb_node;
Josef Bacik96303082009-07-13 21:29:25 -04001168 struct btrfs_free_space *entry, *prev = NULL;
Josef Bacik0f9dd462008-09-23 13:14:11 -04001169
Josef Bacik96303082009-07-13 21:29:25 -04001170 /* find entry that is closest to the 'offset' */
1171 while (1) {
1172 if (!n) {
1173 entry = NULL;
Josef Bacik0f9dd462008-09-23 13:14:11 -04001174 break;
1175 }
Josef Bacik96303082009-07-13 21:29:25 -04001176
1177 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1178 prev = entry;
1179
1180 if (offset < entry->offset)
1181 n = n->rb_left;
1182 else if (offset > entry->offset)
1183 n = n->rb_right;
1184 else
1185 break;
Josef Bacik0f9dd462008-09-23 13:14:11 -04001186 }
1187
Josef Bacik96303082009-07-13 21:29:25 -04001188 if (bitmap_only) {
1189 if (!entry)
1190 return NULL;
1191 if (entry->bitmap)
1192 return entry;
Josef Bacik0f9dd462008-09-23 13:14:11 -04001193
Josef Bacik96303082009-07-13 21:29:25 -04001194 /*
1195 * bitmap entry and extent entry may share same offset,
1196 * in that case, bitmap entry comes after extent entry.
1197 */
1198 n = rb_next(n);
1199 if (!n)
1200 return NULL;
1201 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1202 if (entry->offset != offset)
1203 return NULL;
Josef Bacik0f9dd462008-09-23 13:14:11 -04001204
Josef Bacik96303082009-07-13 21:29:25 -04001205 WARN_ON(!entry->bitmap);
1206 return entry;
1207 } else if (entry) {
1208 if (entry->bitmap) {
Josef Bacik0f9dd462008-09-23 13:14:11 -04001209 /*
Josef Bacik96303082009-07-13 21:29:25 -04001210 * if previous extent entry covers the offset,
1211 * we should return it instead of the bitmap entry
Josef Bacik0f9dd462008-09-23 13:14:11 -04001212 */
Josef Bacik96303082009-07-13 21:29:25 -04001213 n = &entry->offset_index;
1214 while (1) {
1215 n = rb_prev(n);
1216 if (!n)
1217 break;
1218 prev = rb_entry(n, struct btrfs_free_space,
1219 offset_index);
1220 if (!prev->bitmap) {
1221 if (prev->offset + prev->bytes > offset)
1222 entry = prev;
1223 break;
1224 }
Josef Bacik0f9dd462008-09-23 13:14:11 -04001225 }
Josef Bacik96303082009-07-13 21:29:25 -04001226 }
1227 return entry;
1228 }
1229
1230 if (!prev)
1231 return NULL;
1232
1233 /* find last entry before the 'offset' */
1234 entry = prev;
1235 if (entry->offset > offset) {
1236 n = rb_prev(&entry->offset_index);
1237 if (n) {
1238 entry = rb_entry(n, struct btrfs_free_space,
1239 offset_index);
1240 BUG_ON(entry->offset > offset);
Josef Bacik0f9dd462008-09-23 13:14:11 -04001241 } else {
Josef Bacik96303082009-07-13 21:29:25 -04001242 if (fuzzy)
1243 return entry;
1244 else
1245 return NULL;
Josef Bacik0f9dd462008-09-23 13:14:11 -04001246 }
1247 }
1248
Josef Bacik96303082009-07-13 21:29:25 -04001249 if (entry->bitmap) {
1250 n = &entry->offset_index;
1251 while (1) {
1252 n = rb_prev(n);
1253 if (!n)
1254 break;
1255 prev = rb_entry(n, struct btrfs_free_space,
1256 offset_index);
1257 if (!prev->bitmap) {
1258 if (prev->offset + prev->bytes > offset)
1259 return prev;
1260 break;
1261 }
1262 }
Li Zefan34d52cb2011-03-29 13:46:06 +08001263 if (entry->offset + BITS_PER_BITMAP * ctl->unit > offset)
Josef Bacik96303082009-07-13 21:29:25 -04001264 return entry;
1265 } else if (entry->offset + entry->bytes > offset)
1266 return entry;
1267
1268 if (!fuzzy)
1269 return NULL;
1270
1271 while (1) {
1272 if (entry->bitmap) {
1273 if (entry->offset + BITS_PER_BITMAP *
Li Zefan34d52cb2011-03-29 13:46:06 +08001274 ctl->unit > offset)
Josef Bacik96303082009-07-13 21:29:25 -04001275 break;
1276 } else {
1277 if (entry->offset + entry->bytes > offset)
1278 break;
1279 }
1280
1281 n = rb_next(&entry->offset_index);
1282 if (!n)
1283 return NULL;
1284 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1285 }
1286 return entry;
Josef Bacik0f9dd462008-09-23 13:14:11 -04001287}
1288
Li Zefanf333adb2010-11-09 14:57:39 +08001289static inline void
Li Zefan34d52cb2011-03-29 13:46:06 +08001290__unlink_free_space(struct btrfs_free_space_ctl *ctl,
Li Zefanf333adb2010-11-09 14:57:39 +08001291 struct btrfs_free_space *info)
Josef Bacik0f9dd462008-09-23 13:14:11 -04001292{
Li Zefan34d52cb2011-03-29 13:46:06 +08001293 rb_erase(&info->offset_index, &ctl->free_space_offset);
1294 ctl->free_extents--;
Li Zefanf333adb2010-11-09 14:57:39 +08001295}
1296
Li Zefan34d52cb2011-03-29 13:46:06 +08001297static void unlink_free_space(struct btrfs_free_space_ctl *ctl,
Li Zefanf333adb2010-11-09 14:57:39 +08001298 struct btrfs_free_space *info)
1299{
Li Zefan34d52cb2011-03-29 13:46:06 +08001300 __unlink_free_space(ctl, info);
1301 ctl->free_space -= info->bytes;
Josef Bacik0f9dd462008-09-23 13:14:11 -04001302}
1303
Li Zefan34d52cb2011-03-29 13:46:06 +08001304static int link_free_space(struct btrfs_free_space_ctl *ctl,
Josef Bacik0f9dd462008-09-23 13:14:11 -04001305 struct btrfs_free_space *info)
1306{
1307 int ret = 0;
1308
Josef Bacik96303082009-07-13 21:29:25 -04001309 BUG_ON(!info->bitmap && !info->bytes);
Li Zefan34d52cb2011-03-29 13:46:06 +08001310 ret = tree_insert_offset(&ctl->free_space_offset, info->offset,
Josef Bacik96303082009-07-13 21:29:25 -04001311 &info->offset_index, (info->bitmap != NULL));
Josef Bacik0f9dd462008-09-23 13:14:11 -04001312 if (ret)
1313 return ret;
1314
Li Zefan34d52cb2011-03-29 13:46:06 +08001315 ctl->free_space += info->bytes;
1316 ctl->free_extents++;
Josef Bacik96303082009-07-13 21:29:25 -04001317 return ret;
1318}
1319
Li Zefan34d52cb2011-03-29 13:46:06 +08001320static void recalculate_thresholds(struct btrfs_free_space_ctl *ctl)
Josef Bacik96303082009-07-13 21:29:25 -04001321{
Li Zefan34d52cb2011-03-29 13:46:06 +08001322 struct btrfs_block_group_cache *block_group = ctl->private;
Josef Bacik25891f72009-09-11 16:11:20 -04001323 u64 max_bytes;
1324 u64 bitmap_bytes;
1325 u64 extent_bytes;
Li Zefan8eb2d822010-11-09 14:48:01 +08001326 u64 size = block_group->key.offset;
Li Zefan34d52cb2011-03-29 13:46:06 +08001327 u64 bytes_per_bg = BITS_PER_BITMAP * block_group->sectorsize;
1328 int max_bitmaps = div64_u64(size + bytes_per_bg - 1, bytes_per_bg);
1329
1330 BUG_ON(ctl->total_bitmaps > max_bitmaps);
Josef Bacik96303082009-07-13 21:29:25 -04001331
1332 /*
1333 * The goal is to keep the total amount of memory used per 1gb of space
1334 * at or below 32k, so we need to adjust how much memory we allow to be
1335 * used by extent based free space tracking
1336 */
Li Zefan8eb2d822010-11-09 14:48:01 +08001337 if (size < 1024 * 1024 * 1024)
1338 max_bytes = MAX_CACHE_BYTES_PER_GIG;
1339 else
1340 max_bytes = MAX_CACHE_BYTES_PER_GIG *
1341 div64_u64(size, 1024 * 1024 * 1024);
Josef Bacik96303082009-07-13 21:29:25 -04001342
Josef Bacik25891f72009-09-11 16:11:20 -04001343 /*
1344 * we want to account for 1 more bitmap than what we have so we can make
1345 * sure we don't go over our overall goal of MAX_CACHE_BYTES_PER_GIG as
1346 * we add more bitmaps.
1347 */
Li Zefan34d52cb2011-03-29 13:46:06 +08001348 bitmap_bytes = (ctl->total_bitmaps + 1) * PAGE_CACHE_SIZE;
Josef Bacik96303082009-07-13 21:29:25 -04001349
Josef Bacik25891f72009-09-11 16:11:20 -04001350 if (bitmap_bytes >= max_bytes) {
Li Zefan34d52cb2011-03-29 13:46:06 +08001351 ctl->extents_thresh = 0;
Josef Bacik25891f72009-09-11 16:11:20 -04001352 return;
Josef Bacik96303082009-07-13 21:29:25 -04001353 }
Josef Bacik25891f72009-09-11 16:11:20 -04001354
1355 /*
1356 * we want the extent entry threshold to always be at most 1/2 the maxw
1357 * bytes we can have, or whatever is less than that.
1358 */
1359 extent_bytes = max_bytes - bitmap_bytes;
1360 extent_bytes = min_t(u64, extent_bytes, div64_u64(max_bytes, 2));
1361
Li Zefan34d52cb2011-03-29 13:46:06 +08001362 ctl->extents_thresh =
Josef Bacik25891f72009-09-11 16:11:20 -04001363 div64_u64(extent_bytes, (sizeof(struct btrfs_free_space)));
Josef Bacik96303082009-07-13 21:29:25 -04001364}
1365
Miao Xiebb3ac5a2011-08-05 09:32:35 +00001366static inline void __bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
1367 struct btrfs_free_space *info,
1368 u64 offset, u64 bytes)
Josef Bacik96303082009-07-13 21:29:25 -04001369{
Li Zefanf38b6e72011-03-14 13:40:51 +08001370 unsigned long start, count;
Josef Bacik96303082009-07-13 21:29:25 -04001371
Li Zefan34d52cb2011-03-29 13:46:06 +08001372 start = offset_to_bit(info->offset, ctl->unit, offset);
1373 count = bytes_to_bits(bytes, ctl->unit);
Li Zefanf38b6e72011-03-14 13:40:51 +08001374 BUG_ON(start + count > BITS_PER_BITMAP);
Josef Bacik96303082009-07-13 21:29:25 -04001375
Li Zefanf38b6e72011-03-14 13:40:51 +08001376 bitmap_clear(info->bitmap, start, count);
Josef Bacik96303082009-07-13 21:29:25 -04001377
1378 info->bytes -= bytes;
Miao Xiebb3ac5a2011-08-05 09:32:35 +00001379}
1380
1381static void bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
1382 struct btrfs_free_space *info, u64 offset,
1383 u64 bytes)
1384{
1385 __bitmap_clear_bits(ctl, info, offset, bytes);
Li Zefan34d52cb2011-03-29 13:46:06 +08001386 ctl->free_space -= bytes;
Josef Bacik96303082009-07-13 21:29:25 -04001387}
1388
Li Zefan34d52cb2011-03-29 13:46:06 +08001389static void bitmap_set_bits(struct btrfs_free_space_ctl *ctl,
Josef Bacik817d52f2009-07-13 21:29:25 -04001390 struct btrfs_free_space *info, u64 offset,
1391 u64 bytes)
Josef Bacik96303082009-07-13 21:29:25 -04001392{
Li Zefanf38b6e72011-03-14 13:40:51 +08001393 unsigned long start, count;
Josef Bacik96303082009-07-13 21:29:25 -04001394
Li Zefan34d52cb2011-03-29 13:46:06 +08001395 start = offset_to_bit(info->offset, ctl->unit, offset);
1396 count = bytes_to_bits(bytes, ctl->unit);
Li Zefanf38b6e72011-03-14 13:40:51 +08001397 BUG_ON(start + count > BITS_PER_BITMAP);
Josef Bacik96303082009-07-13 21:29:25 -04001398
Li Zefanf38b6e72011-03-14 13:40:51 +08001399 bitmap_set(info->bitmap, start, count);
Josef Bacik96303082009-07-13 21:29:25 -04001400
1401 info->bytes += bytes;
Li Zefan34d52cb2011-03-29 13:46:06 +08001402 ctl->free_space += bytes;
Josef Bacik96303082009-07-13 21:29:25 -04001403}
1404
Li Zefan34d52cb2011-03-29 13:46:06 +08001405static int search_bitmap(struct btrfs_free_space_ctl *ctl,
Josef Bacik96303082009-07-13 21:29:25 -04001406 struct btrfs_free_space *bitmap_info, u64 *offset,
1407 u64 *bytes)
1408{
1409 unsigned long found_bits = 0;
1410 unsigned long bits, i;
1411 unsigned long next_zero;
1412
Li Zefan34d52cb2011-03-29 13:46:06 +08001413 i = offset_to_bit(bitmap_info->offset, ctl->unit,
Josef Bacik96303082009-07-13 21:29:25 -04001414 max_t(u64, *offset, bitmap_info->offset));
Li Zefan34d52cb2011-03-29 13:46:06 +08001415 bits = bytes_to_bits(*bytes, ctl->unit);
Josef Bacik96303082009-07-13 21:29:25 -04001416
1417 for (i = find_next_bit(bitmap_info->bitmap, BITS_PER_BITMAP, i);
1418 i < BITS_PER_BITMAP;
1419 i = find_next_bit(bitmap_info->bitmap, BITS_PER_BITMAP, i + 1)) {
1420 next_zero = find_next_zero_bit(bitmap_info->bitmap,
1421 BITS_PER_BITMAP, i);
1422 if ((next_zero - i) >= bits) {
1423 found_bits = next_zero - i;
1424 break;
1425 }
1426 i = next_zero;
1427 }
1428
1429 if (found_bits) {
Li Zefan34d52cb2011-03-29 13:46:06 +08001430 *offset = (u64)(i * ctl->unit) + bitmap_info->offset;
1431 *bytes = (u64)(found_bits) * ctl->unit;
Josef Bacik96303082009-07-13 21:29:25 -04001432 return 0;
1433 }
1434
1435 return -1;
1436}
1437
Li Zefan34d52cb2011-03-29 13:46:06 +08001438static struct btrfs_free_space *
1439find_free_space(struct btrfs_free_space_ctl *ctl, u64 *offset, u64 *bytes)
Josef Bacik96303082009-07-13 21:29:25 -04001440{
1441 struct btrfs_free_space *entry;
1442 struct rb_node *node;
1443 int ret;
1444
Li Zefan34d52cb2011-03-29 13:46:06 +08001445 if (!ctl->free_space_offset.rb_node)
Josef Bacik96303082009-07-13 21:29:25 -04001446 return NULL;
1447
Li Zefan34d52cb2011-03-29 13:46:06 +08001448 entry = tree_search_offset(ctl, offset_to_bitmap(ctl, *offset), 0, 1);
Josef Bacik96303082009-07-13 21:29:25 -04001449 if (!entry)
1450 return NULL;
1451
1452 for (node = &entry->offset_index; node; node = rb_next(node)) {
1453 entry = rb_entry(node, struct btrfs_free_space, offset_index);
1454 if (entry->bytes < *bytes)
1455 continue;
1456
1457 if (entry->bitmap) {
Li Zefan34d52cb2011-03-29 13:46:06 +08001458 ret = search_bitmap(ctl, entry, offset, bytes);
Josef Bacik96303082009-07-13 21:29:25 -04001459 if (!ret)
1460 return entry;
1461 continue;
1462 }
1463
1464 *offset = entry->offset;
1465 *bytes = entry->bytes;
1466 return entry;
1467 }
1468
1469 return NULL;
1470}
1471
Li Zefan34d52cb2011-03-29 13:46:06 +08001472static void add_new_bitmap(struct btrfs_free_space_ctl *ctl,
Josef Bacik96303082009-07-13 21:29:25 -04001473 struct btrfs_free_space *info, u64 offset)
1474{
Li Zefan34d52cb2011-03-29 13:46:06 +08001475 info->offset = offset_to_bitmap(ctl, offset);
Josef Bacikf019f422009-09-11 16:11:20 -04001476 info->bytes = 0;
Alexandre Olivaf2d0f672011-11-28 12:04:43 -02001477 INIT_LIST_HEAD(&info->list);
Li Zefan34d52cb2011-03-29 13:46:06 +08001478 link_free_space(ctl, info);
1479 ctl->total_bitmaps++;
Josef Bacik96303082009-07-13 21:29:25 -04001480
Li Zefan34d52cb2011-03-29 13:46:06 +08001481 ctl->op->recalc_thresholds(ctl);
Josef Bacik96303082009-07-13 21:29:25 -04001482}
1483
Li Zefan34d52cb2011-03-29 13:46:06 +08001484static void free_bitmap(struct btrfs_free_space_ctl *ctl,
Li Zefanedf6e2d2010-11-09 14:50:07 +08001485 struct btrfs_free_space *bitmap_info)
1486{
Li Zefan34d52cb2011-03-29 13:46:06 +08001487 unlink_free_space(ctl, bitmap_info);
Li Zefanedf6e2d2010-11-09 14:50:07 +08001488 kfree(bitmap_info->bitmap);
Josef Bacikdc89e982011-01-28 17:05:48 -05001489 kmem_cache_free(btrfs_free_space_cachep, bitmap_info);
Li Zefan34d52cb2011-03-29 13:46:06 +08001490 ctl->total_bitmaps--;
1491 ctl->op->recalc_thresholds(ctl);
Li Zefanedf6e2d2010-11-09 14:50:07 +08001492}
1493
Li Zefan34d52cb2011-03-29 13:46:06 +08001494static noinline int remove_from_bitmap(struct btrfs_free_space_ctl *ctl,
Josef Bacik96303082009-07-13 21:29:25 -04001495 struct btrfs_free_space *bitmap_info,
1496 u64 *offset, u64 *bytes)
1497{
1498 u64 end;
Josef Bacik6606bb92009-07-31 11:03:58 -04001499 u64 search_start, search_bytes;
1500 int ret;
Josef Bacik96303082009-07-13 21:29:25 -04001501
1502again:
Li Zefan34d52cb2011-03-29 13:46:06 +08001503 end = bitmap_info->offset + (u64)(BITS_PER_BITMAP * ctl->unit) - 1;
Josef Bacik96303082009-07-13 21:29:25 -04001504
Josef Bacik6606bb92009-07-31 11:03:58 -04001505 /*
1506 * XXX - this can go away after a few releases.
1507 *
1508 * since the only user of btrfs_remove_free_space is the tree logging
1509 * stuff, and the only way to test that is under crash conditions, we
1510 * want to have this debug stuff here just in case somethings not
1511 * working. Search the bitmap for the space we are trying to use to
1512 * make sure its actually there. If its not there then we need to stop
1513 * because something has gone wrong.
1514 */
1515 search_start = *offset;
1516 search_bytes = *bytes;
Josef Bacik13dbc082011-02-03 02:39:52 +00001517 search_bytes = min(search_bytes, end - search_start + 1);
Li Zefan34d52cb2011-03-29 13:46:06 +08001518 ret = search_bitmap(ctl, bitmap_info, &search_start, &search_bytes);
Josef Bacik6606bb92009-07-31 11:03:58 -04001519 BUG_ON(ret < 0 || search_start != *offset);
1520
Josef Bacik96303082009-07-13 21:29:25 -04001521 if (*offset > bitmap_info->offset && *offset + *bytes > end) {
Li Zefan34d52cb2011-03-29 13:46:06 +08001522 bitmap_clear_bits(ctl, bitmap_info, *offset, end - *offset + 1);
Josef Bacik96303082009-07-13 21:29:25 -04001523 *bytes -= end - *offset + 1;
1524 *offset = end + 1;
1525 } else if (*offset >= bitmap_info->offset && *offset + *bytes <= end) {
Li Zefan34d52cb2011-03-29 13:46:06 +08001526 bitmap_clear_bits(ctl, bitmap_info, *offset, *bytes);
Josef Bacik96303082009-07-13 21:29:25 -04001527 *bytes = 0;
1528 }
1529
1530 if (*bytes) {
Josef Bacik6606bb92009-07-31 11:03:58 -04001531 struct rb_node *next = rb_next(&bitmap_info->offset_index);
Li Zefanedf6e2d2010-11-09 14:50:07 +08001532 if (!bitmap_info->bytes)
Li Zefan34d52cb2011-03-29 13:46:06 +08001533 free_bitmap(ctl, bitmap_info);
Josef Bacik96303082009-07-13 21:29:25 -04001534
Josef Bacik6606bb92009-07-31 11:03:58 -04001535 /*
1536 * no entry after this bitmap, but we still have bytes to
1537 * remove, so something has gone wrong.
1538 */
1539 if (!next)
Josef Bacik96303082009-07-13 21:29:25 -04001540 return -EINVAL;
1541
Josef Bacik6606bb92009-07-31 11:03:58 -04001542 bitmap_info = rb_entry(next, struct btrfs_free_space,
1543 offset_index);
1544
1545 /*
1546 * if the next entry isn't a bitmap we need to return to let the
1547 * extent stuff do its work.
1548 */
Josef Bacik96303082009-07-13 21:29:25 -04001549 if (!bitmap_info->bitmap)
1550 return -EAGAIN;
1551
Josef Bacik6606bb92009-07-31 11:03:58 -04001552 /*
1553 * Ok the next item is a bitmap, but it may not actually hold
1554 * the information for the rest of this free space stuff, so
1555 * look for it, and if we don't find it return so we can try
1556 * everything over again.
1557 */
1558 search_start = *offset;
1559 search_bytes = *bytes;
Li Zefan34d52cb2011-03-29 13:46:06 +08001560 ret = search_bitmap(ctl, bitmap_info, &search_start,
Josef Bacik6606bb92009-07-31 11:03:58 -04001561 &search_bytes);
1562 if (ret < 0 || search_start != *offset)
1563 return -EAGAIN;
1564
Josef Bacik96303082009-07-13 21:29:25 -04001565 goto again;
Li Zefanedf6e2d2010-11-09 14:50:07 +08001566 } else if (!bitmap_info->bytes)
Li Zefan34d52cb2011-03-29 13:46:06 +08001567 free_bitmap(ctl, bitmap_info);
Josef Bacik96303082009-07-13 21:29:25 -04001568
1569 return 0;
1570}
1571
Josef Bacik2cdc3422011-05-27 14:07:49 -04001572static u64 add_bytes_to_bitmap(struct btrfs_free_space_ctl *ctl,
1573 struct btrfs_free_space *info, u64 offset,
1574 u64 bytes)
1575{
1576 u64 bytes_to_set = 0;
1577 u64 end;
1578
1579 end = info->offset + (u64)(BITS_PER_BITMAP * ctl->unit);
1580
1581 bytes_to_set = min(end - offset, bytes);
1582
1583 bitmap_set_bits(ctl, info, offset, bytes_to_set);
1584
1585 return bytes_to_set;
1586
1587}
1588
Li Zefan34d52cb2011-03-29 13:46:06 +08001589static bool use_bitmap(struct btrfs_free_space_ctl *ctl,
1590 struct btrfs_free_space *info)
Josef Bacik96303082009-07-13 21:29:25 -04001591{
Li Zefan34d52cb2011-03-29 13:46:06 +08001592 struct btrfs_block_group_cache *block_group = ctl->private;
Josef Bacik96303082009-07-13 21:29:25 -04001593
1594 /*
1595 * If we are below the extents threshold then we can add this as an
1596 * extent, and don't have to deal with the bitmap
1597 */
Li Zefan34d52cb2011-03-29 13:46:06 +08001598 if (ctl->free_extents < ctl->extents_thresh) {
Josef Bacik32cb0842011-03-18 16:16:21 -04001599 /*
1600 * If this block group has some small extents we don't want to
1601 * use up all of our free slots in the cache with them, we want
1602 * to reserve them to larger extents, however if we have plent
1603 * of cache left then go ahead an dadd them, no sense in adding
1604 * the overhead of a bitmap if we don't have to.
1605 */
1606 if (info->bytes <= block_group->sectorsize * 4) {
Li Zefan34d52cb2011-03-29 13:46:06 +08001607 if (ctl->free_extents * 2 <= ctl->extents_thresh)
1608 return false;
Josef Bacik32cb0842011-03-18 16:16:21 -04001609 } else {
Li Zefan34d52cb2011-03-29 13:46:06 +08001610 return false;
Josef Bacik32cb0842011-03-18 16:16:21 -04001611 }
1612 }
Josef Bacik96303082009-07-13 21:29:25 -04001613
1614 /*
1615 * some block groups are so tiny they can't be enveloped by a bitmap, so
1616 * don't even bother to create a bitmap for this
1617 */
1618 if (BITS_PER_BITMAP * block_group->sectorsize >
1619 block_group->key.offset)
Li Zefan34d52cb2011-03-29 13:46:06 +08001620 return false;
1621
1622 return true;
1623}
1624
Josef Bacik2cdc3422011-05-27 14:07:49 -04001625static struct btrfs_free_space_op free_space_op = {
1626 .recalc_thresholds = recalculate_thresholds,
1627 .use_bitmap = use_bitmap,
1628};
1629
Li Zefan34d52cb2011-03-29 13:46:06 +08001630static int insert_into_bitmap(struct btrfs_free_space_ctl *ctl,
1631 struct btrfs_free_space *info)
1632{
1633 struct btrfs_free_space *bitmap_info;
Josef Bacik2cdc3422011-05-27 14:07:49 -04001634 struct btrfs_block_group_cache *block_group = NULL;
Li Zefan34d52cb2011-03-29 13:46:06 +08001635 int added = 0;
Josef Bacik2cdc3422011-05-27 14:07:49 -04001636 u64 bytes, offset, bytes_added;
Li Zefan34d52cb2011-03-29 13:46:06 +08001637 int ret;
Josef Bacik96303082009-07-13 21:29:25 -04001638
1639 bytes = info->bytes;
1640 offset = info->offset;
1641
Li Zefan34d52cb2011-03-29 13:46:06 +08001642 if (!ctl->op->use_bitmap(ctl, info))
1643 return 0;
1644
Josef Bacik2cdc3422011-05-27 14:07:49 -04001645 if (ctl->op == &free_space_op)
1646 block_group = ctl->private;
Chris Mason38e87882011-06-10 16:36:57 -04001647again:
Josef Bacik2cdc3422011-05-27 14:07:49 -04001648 /*
1649 * Since we link bitmaps right into the cluster we need to see if we
1650 * have a cluster here, and if so and it has our bitmap we need to add
1651 * the free space to that bitmap.
1652 */
1653 if (block_group && !list_empty(&block_group->cluster_list)) {
1654 struct btrfs_free_cluster *cluster;
1655 struct rb_node *node;
1656 struct btrfs_free_space *entry;
1657
1658 cluster = list_entry(block_group->cluster_list.next,
1659 struct btrfs_free_cluster,
1660 block_group_list);
1661 spin_lock(&cluster->lock);
1662 node = rb_first(&cluster->root);
1663 if (!node) {
1664 spin_unlock(&cluster->lock);
Chris Mason38e87882011-06-10 16:36:57 -04001665 goto no_cluster_bitmap;
Josef Bacik2cdc3422011-05-27 14:07:49 -04001666 }
1667
1668 entry = rb_entry(node, struct btrfs_free_space, offset_index);
1669 if (!entry->bitmap) {
1670 spin_unlock(&cluster->lock);
Chris Mason38e87882011-06-10 16:36:57 -04001671 goto no_cluster_bitmap;
Josef Bacik2cdc3422011-05-27 14:07:49 -04001672 }
1673
1674 if (entry->offset == offset_to_bitmap(ctl, offset)) {
1675 bytes_added = add_bytes_to_bitmap(ctl, entry,
1676 offset, bytes);
1677 bytes -= bytes_added;
1678 offset += bytes_added;
1679 }
1680 spin_unlock(&cluster->lock);
1681 if (!bytes) {
1682 ret = 1;
1683 goto out;
1684 }
1685 }
Chris Mason38e87882011-06-10 16:36:57 -04001686
1687no_cluster_bitmap:
Li Zefan34d52cb2011-03-29 13:46:06 +08001688 bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
Josef Bacik96303082009-07-13 21:29:25 -04001689 1, 0);
1690 if (!bitmap_info) {
1691 BUG_ON(added);
1692 goto new_bitmap;
1693 }
1694
Josef Bacik2cdc3422011-05-27 14:07:49 -04001695 bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes);
1696 bytes -= bytes_added;
1697 offset += bytes_added;
1698 added = 0;
Josef Bacik96303082009-07-13 21:29:25 -04001699
1700 if (!bytes) {
1701 ret = 1;
1702 goto out;
1703 } else
1704 goto again;
1705
1706new_bitmap:
1707 if (info && info->bitmap) {
Li Zefan34d52cb2011-03-29 13:46:06 +08001708 add_new_bitmap(ctl, info, offset);
Josef Bacik96303082009-07-13 21:29:25 -04001709 added = 1;
1710 info = NULL;
1711 goto again;
1712 } else {
Li Zefan34d52cb2011-03-29 13:46:06 +08001713 spin_unlock(&ctl->tree_lock);
Josef Bacik96303082009-07-13 21:29:25 -04001714
1715 /* no pre-allocated info, allocate a new one */
1716 if (!info) {
Josef Bacikdc89e982011-01-28 17:05:48 -05001717 info = kmem_cache_zalloc(btrfs_free_space_cachep,
1718 GFP_NOFS);
Josef Bacik96303082009-07-13 21:29:25 -04001719 if (!info) {
Li Zefan34d52cb2011-03-29 13:46:06 +08001720 spin_lock(&ctl->tree_lock);
Josef Bacik96303082009-07-13 21:29:25 -04001721 ret = -ENOMEM;
1722 goto out;
1723 }
1724 }
1725
1726 /* allocate the bitmap */
1727 info->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
Li Zefan34d52cb2011-03-29 13:46:06 +08001728 spin_lock(&ctl->tree_lock);
Josef Bacik96303082009-07-13 21:29:25 -04001729 if (!info->bitmap) {
1730 ret = -ENOMEM;
1731 goto out;
1732 }
1733 goto again;
1734 }
1735
1736out:
1737 if (info) {
1738 if (info->bitmap)
1739 kfree(info->bitmap);
Josef Bacikdc89e982011-01-28 17:05:48 -05001740 kmem_cache_free(btrfs_free_space_cachep, info);
Josef Bacik96303082009-07-13 21:29:25 -04001741 }
Josef Bacik0f9dd462008-09-23 13:14:11 -04001742
1743 return ret;
1744}
1745
Chris Mason945d8962011-05-22 12:33:42 -04001746static bool try_merge_free_space(struct btrfs_free_space_ctl *ctl,
Li Zefanf333adb2010-11-09 14:57:39 +08001747 struct btrfs_free_space *info, bool update_stat)
Josef Bacik0f9dd462008-09-23 13:14:11 -04001748{
Li Zefan120d66e2010-11-09 14:56:50 +08001749 struct btrfs_free_space *left_info;
1750 struct btrfs_free_space *right_info;
1751 bool merged = false;
1752 u64 offset = info->offset;
1753 u64 bytes = info->bytes;
Josef Bacik6226cb02009-04-03 10:14:18 -04001754
Josef Bacik0f9dd462008-09-23 13:14:11 -04001755 /*
1756 * first we want to see if there is free space adjacent to the range we
1757 * are adding, if there is remove that struct and add a new one to
1758 * cover the entire range
1759 */
Li Zefan34d52cb2011-03-29 13:46:06 +08001760 right_info = tree_search_offset(ctl, offset + bytes, 0, 0);
Josef Bacik96303082009-07-13 21:29:25 -04001761 if (right_info && rb_prev(&right_info->offset_index))
1762 left_info = rb_entry(rb_prev(&right_info->offset_index),
1763 struct btrfs_free_space, offset_index);
1764 else
Li Zefan34d52cb2011-03-29 13:46:06 +08001765 left_info = tree_search_offset(ctl, offset - 1, 0, 0);
Josef Bacik0f9dd462008-09-23 13:14:11 -04001766
Josef Bacik96303082009-07-13 21:29:25 -04001767 if (right_info && !right_info->bitmap) {
Li Zefanf333adb2010-11-09 14:57:39 +08001768 if (update_stat)
Li Zefan34d52cb2011-03-29 13:46:06 +08001769 unlink_free_space(ctl, right_info);
Li Zefanf333adb2010-11-09 14:57:39 +08001770 else
Li Zefan34d52cb2011-03-29 13:46:06 +08001771 __unlink_free_space(ctl, right_info);
Josef Bacik6226cb02009-04-03 10:14:18 -04001772 info->bytes += right_info->bytes;
Josef Bacikdc89e982011-01-28 17:05:48 -05001773 kmem_cache_free(btrfs_free_space_cachep, right_info);
Li Zefan120d66e2010-11-09 14:56:50 +08001774 merged = true;
Josef Bacik0f9dd462008-09-23 13:14:11 -04001775 }
1776
Josef Bacik96303082009-07-13 21:29:25 -04001777 if (left_info && !left_info->bitmap &&
1778 left_info->offset + left_info->bytes == offset) {
Li Zefanf333adb2010-11-09 14:57:39 +08001779 if (update_stat)
Li Zefan34d52cb2011-03-29 13:46:06 +08001780 unlink_free_space(ctl, left_info);
Li Zefanf333adb2010-11-09 14:57:39 +08001781 else
Li Zefan34d52cb2011-03-29 13:46:06 +08001782 __unlink_free_space(ctl, left_info);
Josef Bacik6226cb02009-04-03 10:14:18 -04001783 info->offset = left_info->offset;
1784 info->bytes += left_info->bytes;
Josef Bacikdc89e982011-01-28 17:05:48 -05001785 kmem_cache_free(btrfs_free_space_cachep, left_info);
Li Zefan120d66e2010-11-09 14:56:50 +08001786 merged = true;
Josef Bacik0f9dd462008-09-23 13:14:11 -04001787 }
1788
Li Zefan120d66e2010-11-09 14:56:50 +08001789 return merged;
1790}
1791
Li Zefan581bb052011-04-20 10:06:11 +08001792int __btrfs_add_free_space(struct btrfs_free_space_ctl *ctl,
1793 u64 offset, u64 bytes)
Li Zefan120d66e2010-11-09 14:56:50 +08001794{
1795 struct btrfs_free_space *info;
1796 int ret = 0;
1797
Josef Bacikdc89e982011-01-28 17:05:48 -05001798 info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
Li Zefan120d66e2010-11-09 14:56:50 +08001799 if (!info)
1800 return -ENOMEM;
1801
1802 info->offset = offset;
1803 info->bytes = bytes;
1804
Li Zefan34d52cb2011-03-29 13:46:06 +08001805 spin_lock(&ctl->tree_lock);
Li Zefan120d66e2010-11-09 14:56:50 +08001806
Li Zefan34d52cb2011-03-29 13:46:06 +08001807 if (try_merge_free_space(ctl, info, true))
Li Zefan120d66e2010-11-09 14:56:50 +08001808 goto link;
1809
1810 /*
1811 * There was no extent directly to the left or right of this new
1812 * extent then we know we're going to have to allocate a new extent, so
1813 * before we do that see if we need to drop this into a bitmap
1814 */
Li Zefan34d52cb2011-03-29 13:46:06 +08001815 ret = insert_into_bitmap(ctl, info);
Li Zefan120d66e2010-11-09 14:56:50 +08001816 if (ret < 0) {
1817 goto out;
1818 } else if (ret) {
1819 ret = 0;
1820 goto out;
1821 }
1822link:
Li Zefan34d52cb2011-03-29 13:46:06 +08001823 ret = link_free_space(ctl, info);
Josef Bacik0f9dd462008-09-23 13:14:11 -04001824 if (ret)
Josef Bacikdc89e982011-01-28 17:05:48 -05001825 kmem_cache_free(btrfs_free_space_cachep, info);
Josef Bacik96303082009-07-13 21:29:25 -04001826out:
Li Zefan34d52cb2011-03-29 13:46:06 +08001827 spin_unlock(&ctl->tree_lock);
Josef Bacik6226cb02009-04-03 10:14:18 -04001828
Josef Bacik0f9dd462008-09-23 13:14:11 -04001829 if (ret) {
Josef Bacik96303082009-07-13 21:29:25 -04001830 printk(KERN_CRIT "btrfs: unable to add free space :%d\n", ret);
Stoyan Gaydarovc2934982009-04-02 17:05:11 -04001831 BUG_ON(ret == -EEXIST);
Josef Bacik0f9dd462008-09-23 13:14:11 -04001832 }
1833
Josef Bacik0f9dd462008-09-23 13:14:11 -04001834 return ret;
1835}
1836
Josef Bacik6226cb02009-04-03 10:14:18 -04001837int btrfs_remove_free_space(struct btrfs_block_group_cache *block_group,
1838 u64 offset, u64 bytes)
Josef Bacik0f9dd462008-09-23 13:14:11 -04001839{
Li Zefan34d52cb2011-03-29 13:46:06 +08001840 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
Josef Bacik0f9dd462008-09-23 13:14:11 -04001841 struct btrfs_free_space *info;
Josef Bacik96303082009-07-13 21:29:25 -04001842 struct btrfs_free_space *next_info = NULL;
Josef Bacik0f9dd462008-09-23 13:14:11 -04001843 int ret = 0;
1844
Li Zefan34d52cb2011-03-29 13:46:06 +08001845 spin_lock(&ctl->tree_lock);
Josef Bacik6226cb02009-04-03 10:14:18 -04001846
Josef Bacik96303082009-07-13 21:29:25 -04001847again:
Li Zefan34d52cb2011-03-29 13:46:06 +08001848 info = tree_search_offset(ctl, offset, 0, 0);
Josef Bacik96303082009-07-13 21:29:25 -04001849 if (!info) {
Josef Bacik6606bb92009-07-31 11:03:58 -04001850 /*
1851 * oops didn't find an extent that matched the space we wanted
1852 * to remove, look for a bitmap instead
1853 */
Li Zefan34d52cb2011-03-29 13:46:06 +08001854 info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
Josef Bacik6606bb92009-07-31 11:03:58 -04001855 1, 0);
1856 if (!info) {
Chris Mason24a70312011-11-21 09:39:11 -05001857 /* the tree logging code might be calling us before we
1858 * have fully loaded the free space rbtree for this
1859 * block group. So it is possible the entry won't
1860 * be in the rbtree yet at all. The caching code
1861 * will make sure not to put it in the rbtree if
1862 * the logging code has pinned it.
1863 */
Josef Bacik6606bb92009-07-31 11:03:58 -04001864 goto out_lock;
1865 }
Josef Bacik96303082009-07-13 21:29:25 -04001866 }
1867
1868 if (info->bytes < bytes && rb_next(&info->offset_index)) {
1869 u64 end;
1870 next_info = rb_entry(rb_next(&info->offset_index),
1871 struct btrfs_free_space,
1872 offset_index);
1873
1874 if (next_info->bitmap)
Li Zefan34d52cb2011-03-29 13:46:06 +08001875 end = next_info->offset +
1876 BITS_PER_BITMAP * ctl->unit - 1;
Josef Bacik96303082009-07-13 21:29:25 -04001877 else
1878 end = next_info->offset + next_info->bytes;
1879
1880 if (next_info->bytes < bytes ||
1881 next_info->offset > offset || offset > end) {
1882 printk(KERN_CRIT "Found free space at %llu, size %llu,"
1883 " trying to use %llu\n",
1884 (unsigned long long)info->offset,
1885 (unsigned long long)info->bytes,
1886 (unsigned long long)bytes);
Josef Bacik0f9dd462008-09-23 13:14:11 -04001887 WARN_ON(1);
1888 ret = -EINVAL;
Josef Bacik96303082009-07-13 21:29:25 -04001889 goto out_lock;
Josef Bacik0f9dd462008-09-23 13:14:11 -04001890 }
Josef Bacik96303082009-07-13 21:29:25 -04001891
1892 info = next_info;
1893 }
1894
1895 if (info->bytes == bytes) {
Li Zefan34d52cb2011-03-29 13:46:06 +08001896 unlink_free_space(ctl, info);
Josef Bacik96303082009-07-13 21:29:25 -04001897 if (info->bitmap) {
1898 kfree(info->bitmap);
Li Zefan34d52cb2011-03-29 13:46:06 +08001899 ctl->total_bitmaps--;
Josef Bacik0f9dd462008-09-23 13:14:11 -04001900 }
Josef Bacikdc89e982011-01-28 17:05:48 -05001901 kmem_cache_free(btrfs_free_space_cachep, info);
Chris Mason1eae31e2011-10-14 06:31:20 -04001902 ret = 0;
Josef Bacik96303082009-07-13 21:29:25 -04001903 goto out_lock;
1904 }
Josef Bacik0f9dd462008-09-23 13:14:11 -04001905
Josef Bacik96303082009-07-13 21:29:25 -04001906 if (!info->bitmap && info->offset == offset) {
Li Zefan34d52cb2011-03-29 13:46:06 +08001907 unlink_free_space(ctl, info);
Josef Bacik0f9dd462008-09-23 13:14:11 -04001908 info->offset += bytes;
1909 info->bytes -= bytes;
Chris Mason1eae31e2011-10-14 06:31:20 -04001910 ret = link_free_space(ctl, info);
1911 WARN_ON(ret);
Josef Bacik96303082009-07-13 21:29:25 -04001912 goto out_lock;
1913 }
Josef Bacik0f9dd462008-09-23 13:14:11 -04001914
Josef Bacik96303082009-07-13 21:29:25 -04001915 if (!info->bitmap && info->offset <= offset &&
1916 info->offset + info->bytes >= offset + bytes) {
Chris Mason9b49c9b2008-09-24 11:23:25 -04001917 u64 old_start = info->offset;
1918 /*
1919 * we're freeing space in the middle of the info,
1920 * this can happen during tree log replay
1921 *
1922 * first unlink the old info and then
1923 * insert it again after the hole we're creating
1924 */
Li Zefan34d52cb2011-03-29 13:46:06 +08001925 unlink_free_space(ctl, info);
Chris Mason9b49c9b2008-09-24 11:23:25 -04001926 if (offset + bytes < info->offset + info->bytes) {
1927 u64 old_end = info->offset + info->bytes;
1928
1929 info->offset = offset + bytes;
1930 info->bytes = old_end - info->offset;
Li Zefan34d52cb2011-03-29 13:46:06 +08001931 ret = link_free_space(ctl, info);
Josef Bacik96303082009-07-13 21:29:25 -04001932 WARN_ON(ret);
1933 if (ret)
1934 goto out_lock;
Chris Mason9b49c9b2008-09-24 11:23:25 -04001935 } else {
1936 /* the hole we're creating ends at the end
1937 * of the info struct, just free the info
1938 */
Josef Bacikdc89e982011-01-28 17:05:48 -05001939 kmem_cache_free(btrfs_free_space_cachep, info);
Chris Mason9b49c9b2008-09-24 11:23:25 -04001940 }
Li Zefan34d52cb2011-03-29 13:46:06 +08001941 spin_unlock(&ctl->tree_lock);
Josef Bacik96303082009-07-13 21:29:25 -04001942
1943 /* step two, insert a new info struct to cover
1944 * anything before the hole
Chris Mason9b49c9b2008-09-24 11:23:25 -04001945 */
Josef Bacik6226cb02009-04-03 10:14:18 -04001946 ret = btrfs_add_free_space(block_group, old_start,
1947 offset - old_start);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001948 WARN_ON(ret); /* -ENOMEM */
Josef Bacik96303082009-07-13 21:29:25 -04001949 goto out;
Josef Bacik0f9dd462008-09-23 13:14:11 -04001950 }
Josef Bacik96303082009-07-13 21:29:25 -04001951
Li Zefan34d52cb2011-03-29 13:46:06 +08001952 ret = remove_from_bitmap(ctl, info, &offset, &bytes);
Josef Bacik96303082009-07-13 21:29:25 -04001953 if (ret == -EAGAIN)
1954 goto again;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001955 BUG_ON(ret); /* logic error */
Josef Bacik96303082009-07-13 21:29:25 -04001956out_lock:
Li Zefan34d52cb2011-03-29 13:46:06 +08001957 spin_unlock(&ctl->tree_lock);
Josef Bacik0f9dd462008-09-23 13:14:11 -04001958out:
Josef Bacik25179202008-10-29 14:49:05 -04001959 return ret;
1960}
1961
Josef Bacik0f9dd462008-09-23 13:14:11 -04001962void btrfs_dump_free_space(struct btrfs_block_group_cache *block_group,
1963 u64 bytes)
1964{
Li Zefan34d52cb2011-03-29 13:46:06 +08001965 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
Josef Bacik0f9dd462008-09-23 13:14:11 -04001966 struct btrfs_free_space *info;
1967 struct rb_node *n;
1968 int count = 0;
1969
Li Zefan34d52cb2011-03-29 13:46:06 +08001970 for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) {
Josef Bacik0f9dd462008-09-23 13:14:11 -04001971 info = rb_entry(n, struct btrfs_free_space, offset_index);
1972 if (info->bytes >= bytes)
1973 count++;
Josef Bacik96303082009-07-13 21:29:25 -04001974 printk(KERN_CRIT "entry offset %llu, bytes %llu, bitmap %s\n",
Joel Becker21380932009-04-21 12:38:29 -07001975 (unsigned long long)info->offset,
Josef Bacik96303082009-07-13 21:29:25 -04001976 (unsigned long long)info->bytes,
1977 (info->bitmap) ? "yes" : "no");
Josef Bacik0f9dd462008-09-23 13:14:11 -04001978 }
Josef Bacik96303082009-07-13 21:29:25 -04001979 printk(KERN_INFO "block group has cluster?: %s\n",
1980 list_empty(&block_group->cluster_list) ? "no" : "yes");
Josef Bacik0f9dd462008-09-23 13:14:11 -04001981 printk(KERN_INFO "%d blocks of free space at or bigger than bytes is"
1982 "\n", count);
1983}
1984
Li Zefan34d52cb2011-03-29 13:46:06 +08001985void btrfs_init_free_space_ctl(struct btrfs_block_group_cache *block_group)
Josef Bacik0f9dd462008-09-23 13:14:11 -04001986{
Li Zefan34d52cb2011-03-29 13:46:06 +08001987 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
Josef Bacik0f9dd462008-09-23 13:14:11 -04001988
Li Zefan34d52cb2011-03-29 13:46:06 +08001989 spin_lock_init(&ctl->tree_lock);
1990 ctl->unit = block_group->sectorsize;
1991 ctl->start = block_group->key.objectid;
1992 ctl->private = block_group;
1993 ctl->op = &free_space_op;
Josef Bacik0f9dd462008-09-23 13:14:11 -04001994
Li Zefan34d52cb2011-03-29 13:46:06 +08001995 /*
1996 * we only want to have 32k of ram per block group for keeping
1997 * track of free space, and if we pass 1/2 of that we want to
1998 * start converting things over to using bitmaps
1999 */
2000 ctl->extents_thresh = ((1024 * 32) / 2) /
2001 sizeof(struct btrfs_free_space);
Josef Bacik0f9dd462008-09-23 13:14:11 -04002002}
2003
Chris Masonfa9c0d72009-04-03 09:47:43 -04002004/*
2005 * for a given cluster, put all of its extents back into the free
2006 * space cache. If the block group passed doesn't match the block group
2007 * pointed to by the cluster, someone else raced in and freed the
2008 * cluster already. In that case, we just return without changing anything
2009 */
2010static int
2011__btrfs_return_cluster_to_free_space(
2012 struct btrfs_block_group_cache *block_group,
2013 struct btrfs_free_cluster *cluster)
2014{
Li Zefan34d52cb2011-03-29 13:46:06 +08002015 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
Chris Masonfa9c0d72009-04-03 09:47:43 -04002016 struct btrfs_free_space *entry;
2017 struct rb_node *node;
2018
2019 spin_lock(&cluster->lock);
2020 if (cluster->block_group != block_group)
2021 goto out;
2022
Josef Bacik96303082009-07-13 21:29:25 -04002023 cluster->block_group = NULL;
Chris Masonfa9c0d72009-04-03 09:47:43 -04002024 cluster->window_start = 0;
Josef Bacik96303082009-07-13 21:29:25 -04002025 list_del_init(&cluster->block_group_list);
Josef Bacik96303082009-07-13 21:29:25 -04002026
Chris Masonfa9c0d72009-04-03 09:47:43 -04002027 node = rb_first(&cluster->root);
Josef Bacik96303082009-07-13 21:29:25 -04002028 while (node) {
Josef Bacik4e69b592011-03-21 10:11:24 -04002029 bool bitmap;
2030
Chris Masonfa9c0d72009-04-03 09:47:43 -04002031 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2032 node = rb_next(&entry->offset_index);
2033 rb_erase(&entry->offset_index, &cluster->root);
Josef Bacik4e69b592011-03-21 10:11:24 -04002034
2035 bitmap = (entry->bitmap != NULL);
2036 if (!bitmap)
Li Zefan34d52cb2011-03-29 13:46:06 +08002037 try_merge_free_space(ctl, entry, false);
2038 tree_insert_offset(&ctl->free_space_offset,
Josef Bacik4e69b592011-03-21 10:11:24 -04002039 entry->offset, &entry->offset_index, bitmap);
Chris Masonfa9c0d72009-04-03 09:47:43 -04002040 }
Eric Paris6bef4d32010-02-23 19:43:04 +00002041 cluster->root = RB_ROOT;
Josef Bacik96303082009-07-13 21:29:25 -04002042
Chris Masonfa9c0d72009-04-03 09:47:43 -04002043out:
2044 spin_unlock(&cluster->lock);
Josef Bacik96303082009-07-13 21:29:25 -04002045 btrfs_put_block_group(block_group);
Chris Masonfa9c0d72009-04-03 09:47:43 -04002046 return 0;
2047}
2048
Chris Mason09655372011-05-21 09:27:38 -04002049void __btrfs_remove_free_space_cache_locked(struct btrfs_free_space_ctl *ctl)
Josef Bacik0f9dd462008-09-23 13:14:11 -04002050{
2051 struct btrfs_free_space *info;
2052 struct rb_node *node;
Li Zefan581bb052011-04-20 10:06:11 +08002053
Li Zefan581bb052011-04-20 10:06:11 +08002054 while ((node = rb_last(&ctl->free_space_offset)) != NULL) {
2055 info = rb_entry(node, struct btrfs_free_space, offset_index);
Josef Bacik9b90f512011-06-24 16:02:51 +00002056 if (!info->bitmap) {
2057 unlink_free_space(ctl, info);
2058 kmem_cache_free(btrfs_free_space_cachep, info);
2059 } else {
2060 free_bitmap(ctl, info);
2061 }
Li Zefan581bb052011-04-20 10:06:11 +08002062 if (need_resched()) {
2063 spin_unlock(&ctl->tree_lock);
2064 cond_resched();
2065 spin_lock(&ctl->tree_lock);
2066 }
2067 }
Chris Mason09655372011-05-21 09:27:38 -04002068}
2069
2070void __btrfs_remove_free_space_cache(struct btrfs_free_space_ctl *ctl)
2071{
2072 spin_lock(&ctl->tree_lock);
2073 __btrfs_remove_free_space_cache_locked(ctl);
Li Zefan581bb052011-04-20 10:06:11 +08002074 spin_unlock(&ctl->tree_lock);
2075}
2076
Josef Bacik0f9dd462008-09-23 13:14:11 -04002077void btrfs_remove_free_space_cache(struct btrfs_block_group_cache *block_group)
2078{
Li Zefan34d52cb2011-03-29 13:46:06 +08002079 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
Chris Masonfa9c0d72009-04-03 09:47:43 -04002080 struct btrfs_free_cluster *cluster;
Josef Bacik96303082009-07-13 21:29:25 -04002081 struct list_head *head;
Josef Bacik0f9dd462008-09-23 13:14:11 -04002082
Li Zefan34d52cb2011-03-29 13:46:06 +08002083 spin_lock(&ctl->tree_lock);
Josef Bacik96303082009-07-13 21:29:25 -04002084 while ((head = block_group->cluster_list.next) !=
2085 &block_group->cluster_list) {
2086 cluster = list_entry(head, struct btrfs_free_cluster,
2087 block_group_list);
Chris Masonfa9c0d72009-04-03 09:47:43 -04002088
2089 WARN_ON(cluster->block_group != block_group);
2090 __btrfs_return_cluster_to_free_space(block_group, cluster);
Josef Bacik96303082009-07-13 21:29:25 -04002091 if (need_resched()) {
Li Zefan34d52cb2011-03-29 13:46:06 +08002092 spin_unlock(&ctl->tree_lock);
Josef Bacik96303082009-07-13 21:29:25 -04002093 cond_resched();
Li Zefan34d52cb2011-03-29 13:46:06 +08002094 spin_lock(&ctl->tree_lock);
Josef Bacik96303082009-07-13 21:29:25 -04002095 }
Chris Masonfa9c0d72009-04-03 09:47:43 -04002096 }
Chris Mason09655372011-05-21 09:27:38 -04002097 __btrfs_remove_free_space_cache_locked(ctl);
Li Zefan34d52cb2011-03-29 13:46:06 +08002098 spin_unlock(&ctl->tree_lock);
Chris Masonfa9c0d72009-04-03 09:47:43 -04002099
Josef Bacik0f9dd462008-09-23 13:14:11 -04002100}
2101
Josef Bacik6226cb02009-04-03 10:14:18 -04002102u64 btrfs_find_space_for_alloc(struct btrfs_block_group_cache *block_group,
2103 u64 offset, u64 bytes, u64 empty_size)
Josef Bacik0f9dd462008-09-23 13:14:11 -04002104{
Li Zefan34d52cb2011-03-29 13:46:06 +08002105 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
Josef Bacik6226cb02009-04-03 10:14:18 -04002106 struct btrfs_free_space *entry = NULL;
Josef Bacik96303082009-07-13 21:29:25 -04002107 u64 bytes_search = bytes + empty_size;
Josef Bacik6226cb02009-04-03 10:14:18 -04002108 u64 ret = 0;
Josef Bacik0f9dd462008-09-23 13:14:11 -04002109
Li Zefan34d52cb2011-03-29 13:46:06 +08002110 spin_lock(&ctl->tree_lock);
2111 entry = find_free_space(ctl, &offset, &bytes_search);
Josef Bacik6226cb02009-04-03 10:14:18 -04002112 if (!entry)
Josef Bacik96303082009-07-13 21:29:25 -04002113 goto out;
2114
2115 ret = offset;
2116 if (entry->bitmap) {
Li Zefan34d52cb2011-03-29 13:46:06 +08002117 bitmap_clear_bits(ctl, entry, offset, bytes);
Li Zefanedf6e2d2010-11-09 14:50:07 +08002118 if (!entry->bytes)
Li Zefan34d52cb2011-03-29 13:46:06 +08002119 free_bitmap(ctl, entry);
Josef Bacik96303082009-07-13 21:29:25 -04002120 } else {
Li Zefan34d52cb2011-03-29 13:46:06 +08002121 unlink_free_space(ctl, entry);
Josef Bacik6226cb02009-04-03 10:14:18 -04002122 entry->offset += bytes;
2123 entry->bytes -= bytes;
Josef Bacik6226cb02009-04-03 10:14:18 -04002124 if (!entry->bytes)
Josef Bacikdc89e982011-01-28 17:05:48 -05002125 kmem_cache_free(btrfs_free_space_cachep, entry);
Josef Bacik6226cb02009-04-03 10:14:18 -04002126 else
Li Zefan34d52cb2011-03-29 13:46:06 +08002127 link_free_space(ctl, entry);
Josef Bacik6226cb02009-04-03 10:14:18 -04002128 }
Josef Bacik0f9dd462008-09-23 13:14:11 -04002129
Josef Bacik96303082009-07-13 21:29:25 -04002130out:
Li Zefan34d52cb2011-03-29 13:46:06 +08002131 spin_unlock(&ctl->tree_lock);
Josef Bacik817d52f2009-07-13 21:29:25 -04002132
Josef Bacik0f9dd462008-09-23 13:14:11 -04002133 return ret;
2134}
Chris Masonfa9c0d72009-04-03 09:47:43 -04002135
2136/*
2137 * given a cluster, put all of its extents back into the free space
2138 * cache. If a block group is passed, this function will only free
2139 * a cluster that belongs to the passed block group.
2140 *
2141 * Otherwise, it'll get a reference on the block group pointed to by the
2142 * cluster and remove the cluster from it.
2143 */
2144int btrfs_return_cluster_to_free_space(
2145 struct btrfs_block_group_cache *block_group,
2146 struct btrfs_free_cluster *cluster)
2147{
Li Zefan34d52cb2011-03-29 13:46:06 +08002148 struct btrfs_free_space_ctl *ctl;
Chris Masonfa9c0d72009-04-03 09:47:43 -04002149 int ret;
2150
2151 /* first, get a safe pointer to the block group */
2152 spin_lock(&cluster->lock);
2153 if (!block_group) {
2154 block_group = cluster->block_group;
2155 if (!block_group) {
2156 spin_unlock(&cluster->lock);
2157 return 0;
2158 }
2159 } else if (cluster->block_group != block_group) {
2160 /* someone else has already freed it don't redo their work */
2161 spin_unlock(&cluster->lock);
2162 return 0;
2163 }
2164 atomic_inc(&block_group->count);
2165 spin_unlock(&cluster->lock);
2166
Li Zefan34d52cb2011-03-29 13:46:06 +08002167 ctl = block_group->free_space_ctl;
2168
Chris Masonfa9c0d72009-04-03 09:47:43 -04002169 /* now return any extents the cluster had on it */
Li Zefan34d52cb2011-03-29 13:46:06 +08002170 spin_lock(&ctl->tree_lock);
Chris Masonfa9c0d72009-04-03 09:47:43 -04002171 ret = __btrfs_return_cluster_to_free_space(block_group, cluster);
Li Zefan34d52cb2011-03-29 13:46:06 +08002172 spin_unlock(&ctl->tree_lock);
Chris Masonfa9c0d72009-04-03 09:47:43 -04002173
2174 /* finally drop our ref */
2175 btrfs_put_block_group(block_group);
2176 return ret;
2177}
2178
Josef Bacik96303082009-07-13 21:29:25 -04002179static u64 btrfs_alloc_from_bitmap(struct btrfs_block_group_cache *block_group,
2180 struct btrfs_free_cluster *cluster,
Josef Bacik4e69b592011-03-21 10:11:24 -04002181 struct btrfs_free_space *entry,
Josef Bacik96303082009-07-13 21:29:25 -04002182 u64 bytes, u64 min_start)
2183{
Li Zefan34d52cb2011-03-29 13:46:06 +08002184 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
Josef Bacik96303082009-07-13 21:29:25 -04002185 int err;
2186 u64 search_start = cluster->window_start;
2187 u64 search_bytes = bytes;
2188 u64 ret = 0;
2189
Josef Bacik96303082009-07-13 21:29:25 -04002190 search_start = min_start;
2191 search_bytes = bytes;
2192
Li Zefan34d52cb2011-03-29 13:46:06 +08002193 err = search_bitmap(ctl, entry, &search_start, &search_bytes);
Josef Bacik96303082009-07-13 21:29:25 -04002194 if (err)
Josef Bacik4e69b592011-03-21 10:11:24 -04002195 return 0;
Josef Bacik96303082009-07-13 21:29:25 -04002196
2197 ret = search_start;
Miao Xiebb3ac5a2011-08-05 09:32:35 +00002198 __bitmap_clear_bits(ctl, entry, ret, bytes);
Josef Bacik96303082009-07-13 21:29:25 -04002199
2200 return ret;
2201}
2202
Chris Masonfa9c0d72009-04-03 09:47:43 -04002203/*
2204 * given a cluster, try to allocate 'bytes' from it, returns 0
2205 * if it couldn't find anything suitably large, or a logical disk offset
2206 * if things worked out
2207 */
2208u64 btrfs_alloc_from_cluster(struct btrfs_block_group_cache *block_group,
2209 struct btrfs_free_cluster *cluster, u64 bytes,
2210 u64 min_start)
2211{
Li Zefan34d52cb2011-03-29 13:46:06 +08002212 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
Chris Masonfa9c0d72009-04-03 09:47:43 -04002213 struct btrfs_free_space *entry = NULL;
2214 struct rb_node *node;
2215 u64 ret = 0;
2216
2217 spin_lock(&cluster->lock);
2218 if (bytes > cluster->max_size)
2219 goto out;
2220
2221 if (cluster->block_group != block_group)
2222 goto out;
2223
2224 node = rb_first(&cluster->root);
2225 if (!node)
2226 goto out;
2227
2228 entry = rb_entry(node, struct btrfs_free_space, offset_index);
Chris Masonfa9c0d72009-04-03 09:47:43 -04002229 while(1) {
Josef Bacik4e69b592011-03-21 10:11:24 -04002230 if (entry->bytes < bytes ||
2231 (!entry->bitmap && entry->offset < min_start)) {
Chris Masonfa9c0d72009-04-03 09:47:43 -04002232 node = rb_next(&entry->offset_index);
2233 if (!node)
2234 break;
2235 entry = rb_entry(node, struct btrfs_free_space,
2236 offset_index);
2237 continue;
2238 }
Chris Masonfa9c0d72009-04-03 09:47:43 -04002239
Josef Bacik4e69b592011-03-21 10:11:24 -04002240 if (entry->bitmap) {
2241 ret = btrfs_alloc_from_bitmap(block_group,
2242 cluster, entry, bytes,
Josef Bacik0b4a9d22012-01-26 15:01:11 -05002243 cluster->window_start);
Josef Bacik4e69b592011-03-21 10:11:24 -04002244 if (ret == 0) {
Josef Bacik4e69b592011-03-21 10:11:24 -04002245 node = rb_next(&entry->offset_index);
2246 if (!node)
2247 break;
2248 entry = rb_entry(node, struct btrfs_free_space,
2249 offset_index);
2250 continue;
2251 }
Josef Bacik9b230622012-01-26 15:01:12 -05002252 cluster->window_start += bytes;
Josef Bacik4e69b592011-03-21 10:11:24 -04002253 } else {
Josef Bacik4e69b592011-03-21 10:11:24 -04002254 ret = entry->offset;
2255
2256 entry->offset += bytes;
2257 entry->bytes -= bytes;
2258 }
Chris Masonfa9c0d72009-04-03 09:47:43 -04002259
Li Zefan5e71b5d2010-11-09 14:55:34 +08002260 if (entry->bytes == 0)
Chris Masonfa9c0d72009-04-03 09:47:43 -04002261 rb_erase(&entry->offset_index, &cluster->root);
Chris Masonfa9c0d72009-04-03 09:47:43 -04002262 break;
2263 }
2264out:
2265 spin_unlock(&cluster->lock);
Josef Bacik96303082009-07-13 21:29:25 -04002266
Li Zefan5e71b5d2010-11-09 14:55:34 +08002267 if (!ret)
2268 return 0;
2269
Li Zefan34d52cb2011-03-29 13:46:06 +08002270 spin_lock(&ctl->tree_lock);
Li Zefan5e71b5d2010-11-09 14:55:34 +08002271
Li Zefan34d52cb2011-03-29 13:46:06 +08002272 ctl->free_space -= bytes;
Li Zefan5e71b5d2010-11-09 14:55:34 +08002273 if (entry->bytes == 0) {
Li Zefan34d52cb2011-03-29 13:46:06 +08002274 ctl->free_extents--;
Josef Bacik4e69b592011-03-21 10:11:24 -04002275 if (entry->bitmap) {
2276 kfree(entry->bitmap);
Li Zefan34d52cb2011-03-29 13:46:06 +08002277 ctl->total_bitmaps--;
2278 ctl->op->recalc_thresholds(ctl);
Josef Bacik4e69b592011-03-21 10:11:24 -04002279 }
Josef Bacikdc89e982011-01-28 17:05:48 -05002280 kmem_cache_free(btrfs_free_space_cachep, entry);
Li Zefan5e71b5d2010-11-09 14:55:34 +08002281 }
2282
Li Zefan34d52cb2011-03-29 13:46:06 +08002283 spin_unlock(&ctl->tree_lock);
Li Zefan5e71b5d2010-11-09 14:55:34 +08002284
Chris Masonfa9c0d72009-04-03 09:47:43 -04002285 return ret;
2286}
2287
Josef Bacik96303082009-07-13 21:29:25 -04002288static int btrfs_bitmap_cluster(struct btrfs_block_group_cache *block_group,
2289 struct btrfs_free_space *entry,
2290 struct btrfs_free_cluster *cluster,
Alexandre Oliva1bb91902011-10-14 12:10:36 -03002291 u64 offset, u64 bytes,
2292 u64 cont1_bytes, u64 min_bytes)
Josef Bacik96303082009-07-13 21:29:25 -04002293{
Li Zefan34d52cb2011-03-29 13:46:06 +08002294 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
Josef Bacik96303082009-07-13 21:29:25 -04002295 unsigned long next_zero;
2296 unsigned long i;
Alexandre Oliva1bb91902011-10-14 12:10:36 -03002297 unsigned long want_bits;
2298 unsigned long min_bits;
Josef Bacik96303082009-07-13 21:29:25 -04002299 unsigned long found_bits;
2300 unsigned long start = 0;
2301 unsigned long total_found = 0;
Josef Bacik4e69b592011-03-21 10:11:24 -04002302 int ret;
Josef Bacik96303082009-07-13 21:29:25 -04002303
2304 i = offset_to_bit(entry->offset, block_group->sectorsize,
2305 max_t(u64, offset, entry->offset));
Alexandre Oliva1bb91902011-10-14 12:10:36 -03002306 want_bits = bytes_to_bits(bytes, block_group->sectorsize);
2307 min_bits = bytes_to_bits(min_bytes, block_group->sectorsize);
Josef Bacik96303082009-07-13 21:29:25 -04002308
2309again:
2310 found_bits = 0;
2311 for (i = find_next_bit(entry->bitmap, BITS_PER_BITMAP, i);
2312 i < BITS_PER_BITMAP;
2313 i = find_next_bit(entry->bitmap, BITS_PER_BITMAP, i + 1)) {
2314 next_zero = find_next_zero_bit(entry->bitmap,
2315 BITS_PER_BITMAP, i);
Alexandre Oliva1bb91902011-10-14 12:10:36 -03002316 if (next_zero - i >= min_bits) {
Josef Bacik96303082009-07-13 21:29:25 -04002317 found_bits = next_zero - i;
2318 break;
2319 }
2320 i = next_zero;
2321 }
2322
2323 if (!found_bits)
Josef Bacik4e69b592011-03-21 10:11:24 -04002324 return -ENOSPC;
Josef Bacik96303082009-07-13 21:29:25 -04002325
Alexandre Oliva1bb91902011-10-14 12:10:36 -03002326 if (!total_found) {
Josef Bacik96303082009-07-13 21:29:25 -04002327 start = i;
Alexandre Olivab78d09b2011-11-30 13:43:00 -05002328 cluster->max_size = 0;
Josef Bacik96303082009-07-13 21:29:25 -04002329 }
2330
2331 total_found += found_bits;
2332
2333 if (cluster->max_size < found_bits * block_group->sectorsize)
2334 cluster->max_size = found_bits * block_group->sectorsize;
2335
Alexandre Oliva1bb91902011-10-14 12:10:36 -03002336 if (total_found < want_bits || cluster->max_size < cont1_bytes) {
2337 i = next_zero + 1;
Josef Bacik96303082009-07-13 21:29:25 -04002338 goto again;
2339 }
2340
2341 cluster->window_start = start * block_group->sectorsize +
2342 entry->offset;
Li Zefan34d52cb2011-03-29 13:46:06 +08002343 rb_erase(&entry->offset_index, &ctl->free_space_offset);
Josef Bacik4e69b592011-03-21 10:11:24 -04002344 ret = tree_insert_offset(&cluster->root, entry->offset,
2345 &entry->offset_index, 1);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002346 BUG_ON(ret); /* -EEXIST; Logic error */
Josef Bacik96303082009-07-13 21:29:25 -04002347
Josef Bacik3f7de032011-11-10 08:29:20 -05002348 trace_btrfs_setup_cluster(block_group, cluster,
2349 total_found * block_group->sectorsize, 1);
Josef Bacik96303082009-07-13 21:29:25 -04002350 return 0;
2351}
2352
Chris Masonfa9c0d72009-04-03 09:47:43 -04002353/*
Josef Bacik4e69b592011-03-21 10:11:24 -04002354 * This searches the block group for just extents to fill the cluster with.
Alexandre Oliva1bb91902011-10-14 12:10:36 -03002355 * Try to find a cluster with at least bytes total bytes, at least one
2356 * extent of cont1_bytes, and other clusters of at least min_bytes.
Josef Bacik4e69b592011-03-21 10:11:24 -04002357 */
Josef Bacik3de85bb2011-05-25 13:07:37 -04002358static noinline int
2359setup_cluster_no_bitmap(struct btrfs_block_group_cache *block_group,
2360 struct btrfs_free_cluster *cluster,
2361 struct list_head *bitmaps, u64 offset, u64 bytes,
Alexandre Oliva1bb91902011-10-14 12:10:36 -03002362 u64 cont1_bytes, u64 min_bytes)
Josef Bacik4e69b592011-03-21 10:11:24 -04002363{
Li Zefan34d52cb2011-03-29 13:46:06 +08002364 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
Josef Bacik4e69b592011-03-21 10:11:24 -04002365 struct btrfs_free_space *first = NULL;
2366 struct btrfs_free_space *entry = NULL;
Josef Bacik4e69b592011-03-21 10:11:24 -04002367 struct btrfs_free_space *last;
2368 struct rb_node *node;
2369 u64 window_start;
2370 u64 window_free;
2371 u64 max_extent;
Josef Bacik3f7de032011-11-10 08:29:20 -05002372 u64 total_size = 0;
Josef Bacik4e69b592011-03-21 10:11:24 -04002373
Li Zefan34d52cb2011-03-29 13:46:06 +08002374 entry = tree_search_offset(ctl, offset, 0, 1);
Josef Bacik4e69b592011-03-21 10:11:24 -04002375 if (!entry)
2376 return -ENOSPC;
2377
2378 /*
2379 * We don't want bitmaps, so just move along until we find a normal
2380 * extent entry.
2381 */
Alexandre Oliva1bb91902011-10-14 12:10:36 -03002382 while (entry->bitmap || entry->bytes < min_bytes) {
2383 if (entry->bitmap && list_empty(&entry->list))
Josef Bacik86d4a772011-05-25 13:03:16 -04002384 list_add_tail(&entry->list, bitmaps);
Josef Bacik4e69b592011-03-21 10:11:24 -04002385 node = rb_next(&entry->offset_index);
2386 if (!node)
2387 return -ENOSPC;
2388 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2389 }
2390
2391 window_start = entry->offset;
2392 window_free = entry->bytes;
2393 max_extent = entry->bytes;
2394 first = entry;
2395 last = entry;
Josef Bacik4e69b592011-03-21 10:11:24 -04002396
Alexandre Oliva1bb91902011-10-14 12:10:36 -03002397 for (node = rb_next(&entry->offset_index); node;
2398 node = rb_next(&entry->offset_index)) {
Josef Bacik4e69b592011-03-21 10:11:24 -04002399 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2400
Josef Bacik86d4a772011-05-25 13:03:16 -04002401 if (entry->bitmap) {
2402 if (list_empty(&entry->list))
2403 list_add_tail(&entry->list, bitmaps);
Josef Bacik4e69b592011-03-21 10:11:24 -04002404 continue;
Josef Bacik86d4a772011-05-25 13:03:16 -04002405 }
2406
Alexandre Oliva1bb91902011-10-14 12:10:36 -03002407 if (entry->bytes < min_bytes)
2408 continue;
2409
2410 last = entry;
2411 window_free += entry->bytes;
2412 if (entry->bytes > max_extent)
Josef Bacik4e69b592011-03-21 10:11:24 -04002413 max_extent = entry->bytes;
Josef Bacik4e69b592011-03-21 10:11:24 -04002414 }
2415
Alexandre Oliva1bb91902011-10-14 12:10:36 -03002416 if (window_free < bytes || max_extent < cont1_bytes)
2417 return -ENOSPC;
2418
Josef Bacik4e69b592011-03-21 10:11:24 -04002419 cluster->window_start = first->offset;
2420
2421 node = &first->offset_index;
2422
2423 /*
2424 * now we've found our entries, pull them out of the free space
2425 * cache and put them into the cluster rbtree
2426 */
2427 do {
2428 int ret;
2429
2430 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2431 node = rb_next(&entry->offset_index);
Alexandre Oliva1bb91902011-10-14 12:10:36 -03002432 if (entry->bitmap || entry->bytes < min_bytes)
Josef Bacik4e69b592011-03-21 10:11:24 -04002433 continue;
2434
Li Zefan34d52cb2011-03-29 13:46:06 +08002435 rb_erase(&entry->offset_index, &ctl->free_space_offset);
Josef Bacik4e69b592011-03-21 10:11:24 -04002436 ret = tree_insert_offset(&cluster->root, entry->offset,
2437 &entry->offset_index, 0);
Josef Bacik3f7de032011-11-10 08:29:20 -05002438 total_size += entry->bytes;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002439 BUG_ON(ret); /* -EEXIST; Logic error */
Josef Bacik4e69b592011-03-21 10:11:24 -04002440 } while (node && entry != last);
2441
2442 cluster->max_size = max_extent;
Josef Bacik3f7de032011-11-10 08:29:20 -05002443 trace_btrfs_setup_cluster(block_group, cluster, total_size, 0);
Josef Bacik4e69b592011-03-21 10:11:24 -04002444 return 0;
2445}
2446
2447/*
2448 * This specifically looks for bitmaps that may work in the cluster, we assume
2449 * that we have already failed to find extents that will work.
2450 */
Josef Bacik3de85bb2011-05-25 13:07:37 -04002451static noinline int
2452setup_cluster_bitmap(struct btrfs_block_group_cache *block_group,
2453 struct btrfs_free_cluster *cluster,
2454 struct list_head *bitmaps, u64 offset, u64 bytes,
Alexandre Oliva1bb91902011-10-14 12:10:36 -03002455 u64 cont1_bytes, u64 min_bytes)
Josef Bacik4e69b592011-03-21 10:11:24 -04002456{
Li Zefan34d52cb2011-03-29 13:46:06 +08002457 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
Josef Bacik4e69b592011-03-21 10:11:24 -04002458 struct btrfs_free_space *entry;
Josef Bacik4e69b592011-03-21 10:11:24 -04002459 int ret = -ENOSPC;
Li Zefan0f0fbf12011-11-20 07:33:38 -05002460 u64 bitmap_offset = offset_to_bitmap(ctl, offset);
Josef Bacik4e69b592011-03-21 10:11:24 -04002461
Li Zefan34d52cb2011-03-29 13:46:06 +08002462 if (ctl->total_bitmaps == 0)
Josef Bacik4e69b592011-03-21 10:11:24 -04002463 return -ENOSPC;
2464
Josef Bacik86d4a772011-05-25 13:03:16 -04002465 /*
Li Zefan0f0fbf12011-11-20 07:33:38 -05002466 * The bitmap that covers offset won't be in the list unless offset
2467 * is just its start offset.
2468 */
2469 entry = list_first_entry(bitmaps, struct btrfs_free_space, list);
2470 if (entry->offset != bitmap_offset) {
2471 entry = tree_search_offset(ctl, bitmap_offset, 1, 0);
2472 if (entry && list_empty(&entry->list))
2473 list_add(&entry->list, bitmaps);
2474 }
2475
Josef Bacik86d4a772011-05-25 13:03:16 -04002476 list_for_each_entry(entry, bitmaps, list) {
Josef Bacik357b9782012-01-26 15:01:11 -05002477 if (entry->bytes < bytes)
Josef Bacik86d4a772011-05-25 13:03:16 -04002478 continue;
2479 ret = btrfs_bitmap_cluster(block_group, entry, cluster, offset,
Alexandre Oliva1bb91902011-10-14 12:10:36 -03002480 bytes, cont1_bytes, min_bytes);
Josef Bacik86d4a772011-05-25 13:03:16 -04002481 if (!ret)
2482 return 0;
2483 }
2484
2485 /*
Li Zefan52621cb2011-11-20 07:33:38 -05002486 * The bitmaps list has all the bitmaps that record free space
2487 * starting after offset, so no more search is required.
Josef Bacik86d4a772011-05-25 13:03:16 -04002488 */
Li Zefan52621cb2011-11-20 07:33:38 -05002489 return -ENOSPC;
Josef Bacik4e69b592011-03-21 10:11:24 -04002490}
2491
2492/*
Chris Masonfa9c0d72009-04-03 09:47:43 -04002493 * here we try to find a cluster of blocks in a block group. The goal
Alexandre Oliva1bb91902011-10-14 12:10:36 -03002494 * is to find at least bytes+empty_size.
Chris Masonfa9c0d72009-04-03 09:47:43 -04002495 * We might not find them all in one contiguous area.
2496 *
2497 * returns zero and sets up cluster if things worked out, otherwise
2498 * it returns -enospc
2499 */
2500int btrfs_find_space_cluster(struct btrfs_trans_handle *trans,
Chris Mason451d7582009-06-09 20:28:34 -04002501 struct btrfs_root *root,
Chris Masonfa9c0d72009-04-03 09:47:43 -04002502 struct btrfs_block_group_cache *block_group,
2503 struct btrfs_free_cluster *cluster,
2504 u64 offset, u64 bytes, u64 empty_size)
2505{
Li Zefan34d52cb2011-03-29 13:46:06 +08002506 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
Josef Bacik86d4a772011-05-25 13:03:16 -04002507 struct btrfs_free_space *entry, *tmp;
Li Zefan52621cb2011-11-20 07:33:38 -05002508 LIST_HEAD(bitmaps);
Chris Masonfa9c0d72009-04-03 09:47:43 -04002509 u64 min_bytes;
Alexandre Oliva1bb91902011-10-14 12:10:36 -03002510 u64 cont1_bytes;
Chris Masonfa9c0d72009-04-03 09:47:43 -04002511 int ret;
2512
Alexandre Oliva1bb91902011-10-14 12:10:36 -03002513 /*
2514 * Choose the minimum extent size we'll require for this
2515 * cluster. For SSD_SPREAD, don't allow any fragmentation.
2516 * For metadata, allow allocates with smaller extents. For
2517 * data, keep it dense.
2518 */
Chris Mason451d7582009-06-09 20:28:34 -04002519 if (btrfs_test_opt(root, SSD_SPREAD)) {
Alexandre Oliva1bb91902011-10-14 12:10:36 -03002520 cont1_bytes = min_bytes = bytes + empty_size;
Chris Mason451d7582009-06-09 20:28:34 -04002521 } else if (block_group->flags & BTRFS_BLOCK_GROUP_METADATA) {
Alexandre Oliva1bb91902011-10-14 12:10:36 -03002522 cont1_bytes = bytes;
2523 min_bytes = block_group->sectorsize;
2524 } else {
2525 cont1_bytes = max(bytes, (bytes + empty_size) >> 2);
2526 min_bytes = block_group->sectorsize;
2527 }
Chris Masonfa9c0d72009-04-03 09:47:43 -04002528
Li Zefan34d52cb2011-03-29 13:46:06 +08002529 spin_lock(&ctl->tree_lock);
Josef Bacik7d0d2e82011-03-18 15:13:42 -04002530
2531 /*
2532 * If we know we don't have enough space to make a cluster don't even
2533 * bother doing all the work to try and find one.
2534 */
Alexandre Oliva1bb91902011-10-14 12:10:36 -03002535 if (ctl->free_space < bytes) {
Li Zefan34d52cb2011-03-29 13:46:06 +08002536 spin_unlock(&ctl->tree_lock);
Josef Bacik7d0d2e82011-03-18 15:13:42 -04002537 return -ENOSPC;
2538 }
2539
Chris Masonfa9c0d72009-04-03 09:47:43 -04002540 spin_lock(&cluster->lock);
2541
2542 /* someone already found a cluster, hooray */
2543 if (cluster->block_group) {
2544 ret = 0;
2545 goto out;
2546 }
Josef Bacik4e69b592011-03-21 10:11:24 -04002547
Josef Bacik3f7de032011-11-10 08:29:20 -05002548 trace_btrfs_find_cluster(block_group, offset, bytes, empty_size,
2549 min_bytes);
2550
2551 INIT_LIST_HEAD(&bitmaps);
Josef Bacik86d4a772011-05-25 13:03:16 -04002552 ret = setup_cluster_no_bitmap(block_group, cluster, &bitmaps, offset,
Alexandre Oliva1bb91902011-10-14 12:10:36 -03002553 bytes + empty_size,
2554 cont1_bytes, min_bytes);
Josef Bacik4e69b592011-03-21 10:11:24 -04002555 if (ret)
Josef Bacik86d4a772011-05-25 13:03:16 -04002556 ret = setup_cluster_bitmap(block_group, cluster, &bitmaps,
Alexandre Oliva1bb91902011-10-14 12:10:36 -03002557 offset, bytes + empty_size,
2558 cont1_bytes, min_bytes);
Josef Bacik86d4a772011-05-25 13:03:16 -04002559
2560 /* Clear our temporary list */
2561 list_for_each_entry_safe(entry, tmp, &bitmaps, list)
2562 list_del_init(&entry->list);
Josef Bacik4e69b592011-03-21 10:11:24 -04002563
2564 if (!ret) {
2565 atomic_inc(&block_group->count);
2566 list_add_tail(&cluster->block_group_list,
2567 &block_group->cluster_list);
2568 cluster->block_group = block_group;
Josef Bacik3f7de032011-11-10 08:29:20 -05002569 } else {
2570 trace_btrfs_failed_cluster_setup(block_group);
Chris Masonfa9c0d72009-04-03 09:47:43 -04002571 }
Chris Masonfa9c0d72009-04-03 09:47:43 -04002572out:
2573 spin_unlock(&cluster->lock);
Li Zefan34d52cb2011-03-29 13:46:06 +08002574 spin_unlock(&ctl->tree_lock);
Chris Masonfa9c0d72009-04-03 09:47:43 -04002575
2576 return ret;
2577}
2578
2579/*
2580 * simple code to zero out a cluster
2581 */
2582void btrfs_init_free_cluster(struct btrfs_free_cluster *cluster)
2583{
2584 spin_lock_init(&cluster->lock);
2585 spin_lock_init(&cluster->refill_lock);
Eric Paris6bef4d32010-02-23 19:43:04 +00002586 cluster->root = RB_ROOT;
Chris Masonfa9c0d72009-04-03 09:47:43 -04002587 cluster->max_size = 0;
2588 INIT_LIST_HEAD(&cluster->block_group_list);
2589 cluster->block_group = NULL;
2590}
2591
Li Zefan7fe1e642011-12-29 14:47:27 +08002592static int do_trimming(struct btrfs_block_group_cache *block_group,
2593 u64 *total_trimmed, u64 start, u64 bytes,
2594 u64 reserved_start, u64 reserved_bytes)
2595{
2596 struct btrfs_space_info *space_info = block_group->space_info;
2597 struct btrfs_fs_info *fs_info = block_group->fs_info;
2598 int ret;
2599 int update = 0;
2600 u64 trimmed = 0;
2601
2602 spin_lock(&space_info->lock);
2603 spin_lock(&block_group->lock);
2604 if (!block_group->ro) {
2605 block_group->reserved += reserved_bytes;
2606 space_info->bytes_reserved += reserved_bytes;
2607 update = 1;
2608 }
2609 spin_unlock(&block_group->lock);
2610 spin_unlock(&space_info->lock);
2611
2612 ret = btrfs_error_discard_extent(fs_info->extent_root,
2613 start, bytes, &trimmed);
2614 if (!ret)
2615 *total_trimmed += trimmed;
2616
2617 btrfs_add_free_space(block_group, reserved_start, reserved_bytes);
2618
2619 if (update) {
2620 spin_lock(&space_info->lock);
2621 spin_lock(&block_group->lock);
2622 if (block_group->ro)
2623 space_info->bytes_readonly += reserved_bytes;
2624 block_group->reserved -= reserved_bytes;
2625 space_info->bytes_reserved -= reserved_bytes;
2626 spin_unlock(&space_info->lock);
2627 spin_unlock(&block_group->lock);
2628 }
2629
2630 return ret;
2631}
2632
2633static int trim_no_bitmap(struct btrfs_block_group_cache *block_group,
2634 u64 *total_trimmed, u64 start, u64 end, u64 minlen)
Li Dongyangf7039b12011-03-24 10:24:28 +00002635{
Li Zefan34d52cb2011-03-29 13:46:06 +08002636 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
Li Zefan7fe1e642011-12-29 14:47:27 +08002637 struct btrfs_free_space *entry;
2638 struct rb_node *node;
Li Dongyangf7039b12011-03-24 10:24:28 +00002639 int ret = 0;
Li Zefan7fe1e642011-12-29 14:47:27 +08002640 u64 extent_start;
2641 u64 extent_bytes;
2642 u64 bytes;
Li Dongyangf7039b12011-03-24 10:24:28 +00002643
2644 while (start < end) {
Li Zefan34d52cb2011-03-29 13:46:06 +08002645 spin_lock(&ctl->tree_lock);
Li Dongyangf7039b12011-03-24 10:24:28 +00002646
Li Zefan34d52cb2011-03-29 13:46:06 +08002647 if (ctl->free_space < minlen) {
2648 spin_unlock(&ctl->tree_lock);
Li Dongyangf7039b12011-03-24 10:24:28 +00002649 break;
2650 }
2651
Li Zefan34d52cb2011-03-29 13:46:06 +08002652 entry = tree_search_offset(ctl, start, 0, 1);
Li Zefan7fe1e642011-12-29 14:47:27 +08002653 if (!entry) {
Li Zefan34d52cb2011-03-29 13:46:06 +08002654 spin_unlock(&ctl->tree_lock);
Li Dongyangf7039b12011-03-24 10:24:28 +00002655 break;
2656 }
2657
Li Zefan7fe1e642011-12-29 14:47:27 +08002658 /* skip bitmaps */
2659 while (entry->bitmap) {
2660 node = rb_next(&entry->offset_index);
2661 if (!node) {
Li Zefan34d52cb2011-03-29 13:46:06 +08002662 spin_unlock(&ctl->tree_lock);
Li Zefan7fe1e642011-12-29 14:47:27 +08002663 goto out;
Li Dongyangf7039b12011-03-24 10:24:28 +00002664 }
Li Zefan7fe1e642011-12-29 14:47:27 +08002665 entry = rb_entry(node, struct btrfs_free_space,
2666 offset_index);
Li Dongyangf7039b12011-03-24 10:24:28 +00002667 }
2668
Li Zefan7fe1e642011-12-29 14:47:27 +08002669 if (entry->offset >= end) {
2670 spin_unlock(&ctl->tree_lock);
2671 break;
2672 }
2673
2674 extent_start = entry->offset;
2675 extent_bytes = entry->bytes;
2676 start = max(start, extent_start);
2677 bytes = min(extent_start + extent_bytes, end) - start;
2678 if (bytes < minlen) {
2679 spin_unlock(&ctl->tree_lock);
2680 goto next;
2681 }
2682
2683 unlink_free_space(ctl, entry);
2684 kmem_cache_free(btrfs_free_space_cachep, entry);
2685
Li Zefan34d52cb2011-03-29 13:46:06 +08002686 spin_unlock(&ctl->tree_lock);
Li Dongyangf7039b12011-03-24 10:24:28 +00002687
Li Zefan7fe1e642011-12-29 14:47:27 +08002688 ret = do_trimming(block_group, total_trimmed, start, bytes,
2689 extent_start, extent_bytes);
2690 if (ret)
2691 break;
2692next:
Li Dongyangf7039b12011-03-24 10:24:28 +00002693 start += bytes;
Li Dongyangf7039b12011-03-24 10:24:28 +00002694
2695 if (fatal_signal_pending(current)) {
2696 ret = -ERESTARTSYS;
2697 break;
2698 }
2699
2700 cond_resched();
2701 }
Li Zefan7fe1e642011-12-29 14:47:27 +08002702out:
2703 return ret;
2704}
2705
2706static int trim_bitmaps(struct btrfs_block_group_cache *block_group,
2707 u64 *total_trimmed, u64 start, u64 end, u64 minlen)
2708{
2709 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2710 struct btrfs_free_space *entry;
2711 int ret = 0;
2712 int ret2;
2713 u64 bytes;
2714 u64 offset = offset_to_bitmap(ctl, start);
2715
2716 while (offset < end) {
2717 bool next_bitmap = false;
2718
2719 spin_lock(&ctl->tree_lock);
2720
2721 if (ctl->free_space < minlen) {
2722 spin_unlock(&ctl->tree_lock);
2723 break;
2724 }
2725
2726 entry = tree_search_offset(ctl, offset, 1, 0);
2727 if (!entry) {
2728 spin_unlock(&ctl->tree_lock);
2729 next_bitmap = true;
2730 goto next;
2731 }
2732
2733 bytes = minlen;
2734 ret2 = search_bitmap(ctl, entry, &start, &bytes);
2735 if (ret2 || start >= end) {
2736 spin_unlock(&ctl->tree_lock);
2737 next_bitmap = true;
2738 goto next;
2739 }
2740
2741 bytes = min(bytes, end - start);
2742 if (bytes < minlen) {
2743 spin_unlock(&ctl->tree_lock);
2744 goto next;
2745 }
2746
2747 bitmap_clear_bits(ctl, entry, start, bytes);
2748 if (entry->bytes == 0)
2749 free_bitmap(ctl, entry);
2750
2751 spin_unlock(&ctl->tree_lock);
2752
2753 ret = do_trimming(block_group, total_trimmed, start, bytes,
2754 start, bytes);
2755 if (ret)
2756 break;
2757next:
2758 if (next_bitmap) {
2759 offset += BITS_PER_BITMAP * ctl->unit;
2760 } else {
2761 start += bytes;
2762 if (start >= offset + BITS_PER_BITMAP * ctl->unit)
2763 offset += BITS_PER_BITMAP * ctl->unit;
2764 }
2765
2766 if (fatal_signal_pending(current)) {
2767 ret = -ERESTARTSYS;
2768 break;
2769 }
2770
2771 cond_resched();
2772 }
2773
2774 return ret;
2775}
2776
2777int btrfs_trim_block_group(struct btrfs_block_group_cache *block_group,
2778 u64 *trimmed, u64 start, u64 end, u64 minlen)
2779{
2780 int ret;
2781
2782 *trimmed = 0;
2783
2784 ret = trim_no_bitmap(block_group, trimmed, start, end, minlen);
2785 if (ret)
2786 return ret;
2787
2788 ret = trim_bitmaps(block_group, trimmed, start, end, minlen);
Li Dongyangf7039b12011-03-24 10:24:28 +00002789
2790 return ret;
2791}
Li Zefan581bb052011-04-20 10:06:11 +08002792
2793/*
2794 * Find the left-most item in the cache tree, and then return the
2795 * smallest inode number in the item.
2796 *
2797 * Note: the returned inode number may not be the smallest one in
2798 * the tree, if the left-most item is a bitmap.
2799 */
2800u64 btrfs_find_ino_for_alloc(struct btrfs_root *fs_root)
2801{
2802 struct btrfs_free_space_ctl *ctl = fs_root->free_ino_ctl;
2803 struct btrfs_free_space *entry = NULL;
2804 u64 ino = 0;
2805
2806 spin_lock(&ctl->tree_lock);
2807
2808 if (RB_EMPTY_ROOT(&ctl->free_space_offset))
2809 goto out;
2810
2811 entry = rb_entry(rb_first(&ctl->free_space_offset),
2812 struct btrfs_free_space, offset_index);
2813
2814 if (!entry->bitmap) {
2815 ino = entry->offset;
2816
2817 unlink_free_space(ctl, entry);
2818 entry->offset++;
2819 entry->bytes--;
2820 if (!entry->bytes)
2821 kmem_cache_free(btrfs_free_space_cachep, entry);
2822 else
2823 link_free_space(ctl, entry);
2824 } else {
2825 u64 offset = 0;
2826 u64 count = 1;
2827 int ret;
2828
2829 ret = search_bitmap(ctl, entry, &offset, &count);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002830 /* Logic error; Should be empty if it can't find anything */
Li Zefan581bb052011-04-20 10:06:11 +08002831 BUG_ON(ret);
2832
2833 ino = offset;
2834 bitmap_clear_bits(ctl, entry, offset, 1);
2835 if (entry->bytes == 0)
2836 free_bitmap(ctl, entry);
2837 }
2838out:
2839 spin_unlock(&ctl->tree_lock);
2840
2841 return ino;
2842}
Li Zefan82d59022011-04-20 10:33:24 +08002843
2844struct inode *lookup_free_ino_inode(struct btrfs_root *root,
2845 struct btrfs_path *path)
2846{
2847 struct inode *inode = NULL;
2848
2849 spin_lock(&root->cache_lock);
2850 if (root->cache_inode)
2851 inode = igrab(root->cache_inode);
2852 spin_unlock(&root->cache_lock);
2853 if (inode)
2854 return inode;
2855
2856 inode = __lookup_free_space_inode(root, path, 0);
2857 if (IS_ERR(inode))
2858 return inode;
2859
2860 spin_lock(&root->cache_lock);
David Sterba7841cb22011-05-31 18:07:27 +02002861 if (!btrfs_fs_closing(root->fs_info))
Li Zefan82d59022011-04-20 10:33:24 +08002862 root->cache_inode = igrab(inode);
2863 spin_unlock(&root->cache_lock);
2864
2865 return inode;
2866}
2867
2868int create_free_ino_inode(struct btrfs_root *root,
2869 struct btrfs_trans_handle *trans,
2870 struct btrfs_path *path)
2871{
2872 return __create_free_space_inode(root, trans, path,
2873 BTRFS_FREE_INO_OBJECTID, 0);
2874}
2875
2876int load_free_ino_cache(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2877{
2878 struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
2879 struct btrfs_path *path;
2880 struct inode *inode;
2881 int ret = 0;
2882 u64 root_gen = btrfs_root_generation(&root->root_item);
2883
Chris Mason4b9465c2011-06-03 09:36:29 -04002884 if (!btrfs_test_opt(root, INODE_MAP_CACHE))
2885 return 0;
2886
Li Zefan82d59022011-04-20 10:33:24 +08002887 /*
2888 * If we're unmounting then just return, since this does a search on the
2889 * normal root and not the commit root and we could deadlock.
2890 */
David Sterba7841cb22011-05-31 18:07:27 +02002891 if (btrfs_fs_closing(fs_info))
Li Zefan82d59022011-04-20 10:33:24 +08002892 return 0;
2893
2894 path = btrfs_alloc_path();
2895 if (!path)
2896 return 0;
2897
2898 inode = lookup_free_ino_inode(root, path);
2899 if (IS_ERR(inode))
2900 goto out;
2901
2902 if (root_gen != BTRFS_I(inode)->generation)
2903 goto out_put;
2904
2905 ret = __load_free_space_cache(root, inode, ctl, path, 0);
2906
2907 if (ret < 0)
2908 printk(KERN_ERR "btrfs: failed to load free ino cache for "
2909 "root %llu\n", root->root_key.objectid);
2910out_put:
2911 iput(inode);
2912out:
2913 btrfs_free_path(path);
2914 return ret;
2915}
2916
2917int btrfs_write_out_ino_cache(struct btrfs_root *root,
2918 struct btrfs_trans_handle *trans,
2919 struct btrfs_path *path)
2920{
2921 struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
2922 struct inode *inode;
2923 int ret;
2924
Chris Mason4b9465c2011-06-03 09:36:29 -04002925 if (!btrfs_test_opt(root, INODE_MAP_CACHE))
2926 return 0;
2927
Li Zefan82d59022011-04-20 10:33:24 +08002928 inode = lookup_free_ino_inode(root, path);
2929 if (IS_ERR(inode))
2930 return 0;
2931
2932 ret = __btrfs_write_out_cache(root, inode, ctl, NULL, trans, path, 0);
Josef Bacikc09544e2011-08-30 10:19:10 -04002933 if (ret) {
2934 btrfs_delalloc_release_metadata(inode, inode->i_size);
2935#ifdef DEBUG
Li Zefan82d59022011-04-20 10:33:24 +08002936 printk(KERN_ERR "btrfs: failed to write free ino cache "
2937 "for root %llu\n", root->root_key.objectid);
Josef Bacikc09544e2011-08-30 10:19:10 -04002938#endif
2939 }
Li Zefan82d59022011-04-20 10:33:24 +08002940
2941 iput(inode);
2942 return ret;
2943}