blob: b2130c46fdb5a8d1f368b4494a0982a1597b7235 [file] [log] [blame]
Chris Mason6cbd5572007-06-12 09:07:21 -04001/*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
Yan4b82d6e2007-08-29 09:11:44 -040019#include <linux/blkdev.h>
Chris Mason2e635a22007-03-21 11:12:56 -040020#include <linux/module.h>
Chris Masone20d96d2007-03-22 12:13:20 -040021#include <linux/buffer_head.h>
Chris Mason2e635a22007-03-21 11:12:56 -040022#include <linux/fs.h>
23#include <linux/pagemap.h>
24#include <linux/highmem.h>
25#include <linux/time.h>
26#include <linux/init.h>
Eric Parisa9572a12009-04-02 16:46:06 -040027#include <linux/seq_file.h>
Chris Mason2e635a22007-03-21 11:12:56 -040028#include <linux/string.h>
Chris Mason2e635a22007-03-21 11:12:56 -040029#include <linux/backing-dev.h>
Yan4b82d6e2007-08-29 09:11:44 -040030#include <linux/mount.h>
Chris Masondee26a92007-03-26 16:00:06 -040031#include <linux/mpage.h>
Chris Mason75dfe392007-03-29 11:56:46 -040032#include <linux/swap.h>
33#include <linux/writeback.h>
Chris Mason8fd17792007-04-19 21:01:03 -040034#include <linux/statfs.h>
Chris Mason08607c12007-06-08 15:33:54 -040035#include <linux/compat.h>
Chris Mason95e05282007-08-29 09:11:44 -040036#include <linux/parser.h>
Chris Masonc59f8952007-12-17 20:14:04 -050037#include <linux/ctype.h>
Chris Mason6da6aba2007-12-18 16:15:09 -050038#include <linux/namei.h>
Chris Masona9218f62008-03-24 15:02:04 -040039#include <linux/miscdevice.h>
Qinghuang Feng1bcbf312009-01-15 13:51:03 -080040#include <linux/magic.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090041#include <linux/slab.h>
Chris Mason4b4e25f2008-11-20 10:22:27 -050042#include "compat.h"
Chris Mason2e635a22007-03-21 11:12:56 -040043#include "ctree.h"
Chris Masone20d96d2007-03-22 12:13:20 -040044#include "disk-io.h"
Chris Masond5719762007-03-23 10:01:08 -040045#include "transaction.h"
Chris Mason2c90e5d2007-04-02 10:50:19 -040046#include "btrfs_inode.h"
Chris Masonc5739bb2007-04-10 09:27:04 -040047#include "ioctl.h"
Chris Mason3a686372007-05-24 13:35:57 -040048#include "print-tree.h"
Josef Bacik5103e942007-11-16 11:45:54 -050049#include "xattr.h"
Chris Mason8a4b83c2008-03-24 15:02:07 -040050#include "volumes.h"
Chris Masonb3c3da72008-07-23 12:12:13 -040051#include "version.h"
Balaji Raobe6e8dc2008-07-21 02:01:56 +053052#include "export.h"
Chris Masonc8b97812008-10-29 14:49:59 -040053#include "compression.h"
Chris Mason2e635a22007-03-21 11:12:56 -040054
Alexey Dobriyanb87221d2009-09-21 17:01:09 -070055static const struct super_operations btrfs_super_ops;
Chris Masone20d96d2007-03-22 12:13:20 -040056
liuboacce9522011-01-06 19:30:25 +080057static const char *btrfs_decode_error(struct btrfs_fs_info *fs_info, int errno,
58 char nbuf[16])
59{
60 char *errstr = NULL;
61
62 switch (errno) {
63 case -EIO:
64 errstr = "IO failure";
65 break;
66 case -ENOMEM:
67 errstr = "Out of memory";
68 break;
69 case -EROFS:
70 errstr = "Readonly filesystem";
71 break;
72 default:
73 if (nbuf) {
74 if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
75 errstr = nbuf;
76 }
77 break;
78 }
79
80 return errstr;
81}
82
83static void __save_error_info(struct btrfs_fs_info *fs_info)
84{
85 /*
86 * today we only save the error info into ram. Long term we'll
87 * also send it down to the disk
88 */
89 fs_info->fs_state = BTRFS_SUPER_FLAG_ERROR;
90}
91
92/* NOTE:
93 * We move write_super stuff at umount in order to avoid deadlock
94 * for umount hold all lock.
95 */
96static void save_error_info(struct btrfs_fs_info *fs_info)
97{
98 __save_error_info(fs_info);
99}
100
101/* btrfs handle error by forcing the filesystem readonly */
102static void btrfs_handle_error(struct btrfs_fs_info *fs_info)
103{
104 struct super_block *sb = fs_info->sb;
105
106 if (sb->s_flags & MS_RDONLY)
107 return;
108
109 if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
110 sb->s_flags |= MS_RDONLY;
111 printk(KERN_INFO "btrfs is forced readonly\n");
112 }
113}
114
115/*
116 * __btrfs_std_error decodes expected errors from the caller and
117 * invokes the approciate error response.
118 */
119void __btrfs_std_error(struct btrfs_fs_info *fs_info, const char *function,
120 unsigned int line, int errno)
121{
122 struct super_block *sb = fs_info->sb;
123 char nbuf[16];
124 const char *errstr;
125
126 /*
127 * Special case: if the error is EROFS, and we're already
128 * under MS_RDONLY, then it is safe here.
129 */
130 if (errno == -EROFS && (sb->s_flags & MS_RDONLY))
131 return;
132
133 errstr = btrfs_decode_error(fs_info, errno, nbuf);
134 printk(KERN_CRIT "BTRFS error (device %s) in %s:%d: %s\n",
135 sb->s_id, function, line, errstr);
136 save_error_info(fs_info);
137
138 btrfs_handle_error(fs_info);
139}
140
Chris Masond3977122009-01-05 21:25:51 -0500141static void btrfs_put_super(struct super_block *sb)
Chris Masone20d96d2007-03-22 12:13:20 -0400142{
143 struct btrfs_root *root = btrfs_sb(sb);
144 int ret;
145
146 ret = close_ctree(root);
Chris Masone20d96d2007-03-22 12:13:20 -0400147 sb->s_fs_info = NULL;
Andi Kleen559af822010-10-29 15:14:37 -0400148
149 (void)ret; /* FIXME: need to fix VFS to return error? */
Chris Masone20d96d2007-03-22 12:13:20 -0400150}
Chris Mason2e635a22007-03-21 11:12:56 -0400151
Chris Mason95e05282007-08-29 09:11:44 -0400152enum {
Josef Bacik73f73412009-12-04 17:38:27 +0000153 Opt_degraded, Opt_subvol, Opt_subvolid, Opt_device, Opt_nodatasum,
Josef Bacik287a0ab2010-03-19 18:07:23 +0000154 Opt_nodatacow, Opt_max_inline, Opt_alloc_start, Opt_nobarrier, Opt_ssd,
155 Opt_nossd, Opt_ssd_spread, Opt_thread_pool, Opt_noacl, Opt_compress,
Li Zefan261507a02010-12-17 14:21:50 +0800156 Opt_compress_type, Opt_compress_force, Opt_compress_force_type,
157 Opt_notreelog, Opt_ratio, Opt_flushoncommit, Opt_discard,
158 Opt_space_cache, Opt_clear_cache, Opt_user_subvol_rm_allowed, Opt_err,
Chris Mason95e05282007-08-29 09:11:44 -0400159};
160
161static match_table_t tokens = {
Chris Masondfe25022008-05-13 13:46:40 -0400162 {Opt_degraded, "degraded"},
Chris Mason95e05282007-08-29 09:11:44 -0400163 {Opt_subvol, "subvol=%s"},
Josef Bacik73f73412009-12-04 17:38:27 +0000164 {Opt_subvolid, "subvolid=%d"},
Christoph Hellwig43e570b2008-06-10 10:40:46 -0400165 {Opt_device, "device=%s"},
Chris Masonb6cda9b2007-12-14 15:30:32 -0500166 {Opt_nodatasum, "nodatasum"},
Chris Masonbe20aa92007-12-17 20:14:01 -0500167 {Opt_nodatacow, "nodatacow"},
Chris Mason21ad10c2008-01-09 09:23:21 -0500168 {Opt_nobarrier, "nobarrier"},
Chris Mason6f568d32008-01-29 16:03:38 -0500169 {Opt_max_inline, "max_inline=%s"},
Chris Mason8f662a72008-01-02 10:01:11 -0500170 {Opt_alloc_start, "alloc_start=%s"},
Chris Mason4543df72008-06-11 21:47:56 -0400171 {Opt_thread_pool, "thread_pool=%d"},
Chris Masonc8b97812008-10-29 14:49:59 -0400172 {Opt_compress, "compress"},
Li Zefan261507a02010-12-17 14:21:50 +0800173 {Opt_compress_type, "compress=%s"},
Chris Masona555f812010-01-28 16:18:15 -0500174 {Opt_compress_force, "compress-force"},
Li Zefan261507a02010-12-17 14:21:50 +0800175 {Opt_compress_force_type, "compress-force=%s"},
Chris Masone18e4802008-01-18 10:54:22 -0500176 {Opt_ssd, "ssd"},
Chris Mason451d7582009-06-09 20:28:34 -0400177 {Opt_ssd_spread, "ssd_spread"},
Chris Mason3b30c222009-06-09 16:42:22 -0400178 {Opt_nossd, "nossd"},
Josef Bacik33268ea2008-07-24 12:16:36 -0400179 {Opt_noacl, "noacl"},
Sage Weil3a5e1402009-04-02 16:49:40 -0400180 {Opt_notreelog, "notreelog"},
Sage Weildccae992009-04-02 16:59:01 -0400181 {Opt_flushoncommit, "flushoncommit"},
Josef Bacik97e728d2009-04-21 17:40:57 -0400182 {Opt_ratio, "metadata_ratio=%d"},
Christoph Hellwige244a0a2009-10-14 09:24:59 -0400183 {Opt_discard, "discard"},
Josef Bacik0af3d002010-06-21 14:48:16 -0400184 {Opt_space_cache, "space_cache"},
Josef Bacik88c2ba32010-09-21 14:21:34 -0400185 {Opt_clear_cache, "clear_cache"},
Sage Weil4260f7c2010-10-29 15:46:43 -0400186 {Opt_user_subvol_rm_allowed, "user_subvol_rm_allowed"},
Josef Bacik33268ea2008-07-24 12:16:36 -0400187 {Opt_err, NULL},
Chris Mason95e05282007-08-29 09:11:44 -0400188};
189
Christoph Hellwigedf24ab2008-06-10 10:40:29 -0400190/*
191 * Regular mount options parser. Everything that is needed only when
192 * reading in a new superblock is parsed here.
193 */
194int btrfs_parse_options(struct btrfs_root *root, char *options)
Chris Mason95e05282007-08-29 09:11:44 -0400195{
Christoph Hellwigedf24ab2008-06-10 10:40:29 -0400196 struct btrfs_fs_info *info = root->fs_info;
Chris Mason95e05282007-08-29 09:11:44 -0400197 substring_t args[MAX_OPT_ARGS];
Josef Bacikda495ec2010-02-25 20:38:35 +0000198 char *p, *num, *orig;
Chris Mason4543df72008-06-11 21:47:56 -0400199 int intarg;
Sage Weila7a3f7c2009-11-07 06:19:16 +0000200 int ret = 0;
Li Zefan261507a02010-12-17 14:21:50 +0800201 char *compress_type;
202 bool compress_force = false;
Chris Masonb6cda9b2007-12-14 15:30:32 -0500203
Chris Mason95e05282007-08-29 09:11:44 -0400204 if (!options)
Christoph Hellwigedf24ab2008-06-10 10:40:29 -0400205 return 0;
Chris Mason95e05282007-08-29 09:11:44 -0400206
Chris Masonbe20aa92007-12-17 20:14:01 -0500207 /*
208 * strsep changes the string, duplicate it because parse_options
209 * gets called twice
210 */
211 options = kstrdup(options, GFP_NOFS);
212 if (!options)
213 return -ENOMEM;
214
Josef Bacikda495ec2010-02-25 20:38:35 +0000215 orig = options;
Chris Masonbe20aa92007-12-17 20:14:01 -0500216
Christoph Hellwigedf24ab2008-06-10 10:40:29 -0400217 while ((p = strsep(&options, ",")) != NULL) {
Chris Mason95e05282007-08-29 09:11:44 -0400218 int token;
219 if (!*p)
220 continue;
221
222 token = match_token(p, tokens, args);
223 switch (token) {
Chris Masondfe25022008-05-13 13:46:40 -0400224 case Opt_degraded:
Christoph Hellwigedf24ab2008-06-10 10:40:29 -0400225 printk(KERN_INFO "btrfs: allowing degraded mounts\n");
226 btrfs_set_opt(info->mount_opt, DEGRADED);
Chris Masondfe25022008-05-13 13:46:40 -0400227 break;
Chris Mason95e05282007-08-29 09:11:44 -0400228 case Opt_subvol:
Josef Bacik73f73412009-12-04 17:38:27 +0000229 case Opt_subvolid:
Christoph Hellwig43e570b2008-06-10 10:40:46 -0400230 case Opt_device:
Christoph Hellwigedf24ab2008-06-10 10:40:29 -0400231 /*
Christoph Hellwig43e570b2008-06-10 10:40:46 -0400232 * These are parsed by btrfs_parse_early_options
Christoph Hellwigedf24ab2008-06-10 10:40:29 -0400233 * and can be happily ignored here.
234 */
Chris Masonb6cda9b2007-12-14 15:30:32 -0500235 break;
236 case Opt_nodatasum:
Chris Mason067c28a2009-06-11 09:30:13 -0400237 printk(KERN_INFO "btrfs: setting nodatasum\n");
Christoph Hellwigedf24ab2008-06-10 10:40:29 -0400238 btrfs_set_opt(info->mount_opt, NODATASUM);
Chris Masonbe20aa92007-12-17 20:14:01 -0500239 break;
240 case Opt_nodatacow:
Christoph Hellwigedf24ab2008-06-10 10:40:29 -0400241 printk(KERN_INFO "btrfs: setting nodatacow\n");
242 btrfs_set_opt(info->mount_opt, NODATACOW);
243 btrfs_set_opt(info->mount_opt, NODATASUM);
Chris Mason95e05282007-08-29 09:11:44 -0400244 break;
Chris Masona555f812010-01-28 16:18:15 -0500245 case Opt_compress_force:
Li Zefan261507a02010-12-17 14:21:50 +0800246 case Opt_compress_force_type:
247 compress_force = true;
248 case Opt_compress:
249 case Opt_compress_type:
250 if (token == Opt_compress ||
251 token == Opt_compress_force ||
252 strcmp(args[0].from, "zlib") == 0) {
253 compress_type = "zlib";
254 info->compress_type = BTRFS_COMPRESS_ZLIB;
Li Zefana6fa6fa2010-10-25 15:12:26 +0800255 } else if (strcmp(args[0].from, "lzo") == 0) {
256 compress_type = "lzo";
257 info->compress_type = BTRFS_COMPRESS_LZO;
Li Zefan261507a02010-12-17 14:21:50 +0800258 } else {
259 ret = -EINVAL;
260 goto out;
261 }
262
Chris Masona555f812010-01-28 16:18:15 -0500263 btrfs_set_opt(info->mount_opt, COMPRESS);
Li Zefan261507a02010-12-17 14:21:50 +0800264 if (compress_force) {
265 btrfs_set_opt(info->mount_opt, FORCE_COMPRESS);
266 pr_info("btrfs: force %s compression\n",
267 compress_type);
268 } else
269 pr_info("btrfs: use %s compression\n",
270 compress_type);
Chris Masona555f812010-01-28 16:18:15 -0500271 break;
Chris Masone18e4802008-01-18 10:54:22 -0500272 case Opt_ssd:
Christoph Hellwigedf24ab2008-06-10 10:40:29 -0400273 printk(KERN_INFO "btrfs: use ssd allocation scheme\n");
274 btrfs_set_opt(info->mount_opt, SSD);
Chris Masone18e4802008-01-18 10:54:22 -0500275 break;
Chris Mason451d7582009-06-09 20:28:34 -0400276 case Opt_ssd_spread:
277 printk(KERN_INFO "btrfs: use spread ssd "
278 "allocation scheme\n");
279 btrfs_set_opt(info->mount_opt, SSD);
280 btrfs_set_opt(info->mount_opt, SSD_SPREAD);
281 break;
Chris Mason3b30c222009-06-09 16:42:22 -0400282 case Opt_nossd:
Chris Mason451d7582009-06-09 20:28:34 -0400283 printk(KERN_INFO "btrfs: not using ssd allocation "
284 "scheme\n");
Chris Masonc2898112009-06-10 09:51:32 -0400285 btrfs_set_opt(info->mount_opt, NOSSD);
Chris Mason3b30c222009-06-09 16:42:22 -0400286 btrfs_clear_opt(info->mount_opt, SSD);
Chris Mason451d7582009-06-09 20:28:34 -0400287 btrfs_clear_opt(info->mount_opt, SSD_SPREAD);
Chris Mason3b30c222009-06-09 16:42:22 -0400288 break;
Chris Mason21ad10c2008-01-09 09:23:21 -0500289 case Opt_nobarrier:
Christoph Hellwigedf24ab2008-06-10 10:40:29 -0400290 printk(KERN_INFO "btrfs: turning off barriers\n");
291 btrfs_set_opt(info->mount_opt, NOBARRIER);
Chris Mason21ad10c2008-01-09 09:23:21 -0500292 break;
Chris Mason4543df72008-06-11 21:47:56 -0400293 case Opt_thread_pool:
294 intarg = 0;
295 match_int(&args[0], &intarg);
296 if (intarg) {
297 info->thread_pool_size = intarg;
298 printk(KERN_INFO "btrfs: thread pool %d\n",
299 info->thread_pool_size);
300 }
301 break;
Chris Mason6f568d32008-01-29 16:03:38 -0500302 case Opt_max_inline:
Christoph Hellwigedf24ab2008-06-10 10:40:29 -0400303 num = match_strdup(&args[0]);
304 if (num) {
Akinobu Mita91748462010-02-28 10:59:11 +0000305 info->max_inline = memparse(num, NULL);
Christoph Hellwigedf24ab2008-06-10 10:40:29 -0400306 kfree(num);
Chris Mason6f568d32008-01-29 16:03:38 -0500307
Chris Mason15ada042008-06-11 16:51:38 -0400308 if (info->max_inline) {
309 info->max_inline = max_t(u64,
310 info->max_inline,
311 root->sectorsize);
312 }
Christoph Hellwigedf24ab2008-06-10 10:40:29 -0400313 printk(KERN_INFO "btrfs: max_inline at %llu\n",
Joel Becker21380932009-04-21 12:38:29 -0700314 (unsigned long long)info->max_inline);
Chris Mason6f568d32008-01-29 16:03:38 -0500315 }
316 break;
Chris Mason8f662a72008-01-02 10:01:11 -0500317 case Opt_alloc_start:
Christoph Hellwigedf24ab2008-06-10 10:40:29 -0400318 num = match_strdup(&args[0]);
319 if (num) {
Akinobu Mita91748462010-02-28 10:59:11 +0000320 info->alloc_start = memparse(num, NULL);
Christoph Hellwigedf24ab2008-06-10 10:40:29 -0400321 kfree(num);
322 printk(KERN_INFO
323 "btrfs: allocations start at %llu\n",
Joel Becker21380932009-04-21 12:38:29 -0700324 (unsigned long long)info->alloc_start);
Chris Mason8f662a72008-01-02 10:01:11 -0500325 }
326 break;
Josef Bacik33268ea2008-07-24 12:16:36 -0400327 case Opt_noacl:
328 root->fs_info->sb->s_flags &= ~MS_POSIXACL;
329 break;
Sage Weil3a5e1402009-04-02 16:49:40 -0400330 case Opt_notreelog:
331 printk(KERN_INFO "btrfs: disabling tree log\n");
332 btrfs_set_opt(info->mount_opt, NOTREELOG);
333 break;
Sage Weildccae992009-04-02 16:59:01 -0400334 case Opt_flushoncommit:
335 printk(KERN_INFO "btrfs: turning on flush-on-commit\n");
336 btrfs_set_opt(info->mount_opt, FLUSHONCOMMIT);
337 break;
Josef Bacik97e728d2009-04-21 17:40:57 -0400338 case Opt_ratio:
339 intarg = 0;
340 match_int(&args[0], &intarg);
341 if (intarg) {
342 info->metadata_ratio = intarg;
343 printk(KERN_INFO "btrfs: metadata ratio %d\n",
344 info->metadata_ratio);
345 }
346 break;
Christoph Hellwige244a0a2009-10-14 09:24:59 -0400347 case Opt_discard:
348 btrfs_set_opt(info->mount_opt, DISCARD);
349 break;
Josef Bacik0af3d002010-06-21 14:48:16 -0400350 case Opt_space_cache:
351 printk(KERN_INFO "btrfs: enabling disk space caching\n");
352 btrfs_set_opt(info->mount_opt, SPACE_CACHE);
Josef Bacik0de90872010-11-19 13:40:41 +0000353 break;
Josef Bacik88c2ba32010-09-21 14:21:34 -0400354 case Opt_clear_cache:
355 printk(KERN_INFO "btrfs: force clearing of disk cache\n");
356 btrfs_set_opt(info->mount_opt, CLEAR_CACHE);
Josef Bacik0af3d002010-06-21 14:48:16 -0400357 break;
Sage Weil4260f7c2010-10-29 15:46:43 -0400358 case Opt_user_subvol_rm_allowed:
359 btrfs_set_opt(info->mount_opt, USER_SUBVOL_RM_ALLOWED);
360 break;
Sage Weila7a3f7c2009-11-07 06:19:16 +0000361 case Opt_err:
362 printk(KERN_INFO "btrfs: unrecognized mount option "
363 "'%s'\n", p);
364 ret = -EINVAL;
365 goto out;
Chris Mason95e05282007-08-29 09:11:44 -0400366 default:
Chris Masonbe20aa92007-12-17 20:14:01 -0500367 break;
Chris Mason95e05282007-08-29 09:11:44 -0400368 }
369 }
Sage Weila7a3f7c2009-11-07 06:19:16 +0000370out:
Josef Bacikda495ec2010-02-25 20:38:35 +0000371 kfree(orig);
Sage Weila7a3f7c2009-11-07 06:19:16 +0000372 return ret;
Christoph Hellwigedf24ab2008-06-10 10:40:29 -0400373}
374
375/*
376 * Parse mount options that are required early in the mount process.
377 *
378 * All other options will be parsed on much later in the mount process and
379 * only when we need to allocate a new super block.
380 */
Christoph Hellwig97288f22008-12-02 06:36:09 -0500381static int btrfs_parse_early_options(const char *options, fmode_t flags,
Josef Bacik73f73412009-12-04 17:38:27 +0000382 void *holder, char **subvol_name, u64 *subvol_objectid,
Christoph Hellwig43e570b2008-06-10 10:40:46 -0400383 struct btrfs_fs_devices **fs_devices)
Christoph Hellwigedf24ab2008-06-10 10:40:29 -0400384{
385 substring_t args[MAX_OPT_ARGS];
386 char *opts, *p;
387 int error = 0;
Josef Bacik73f73412009-12-04 17:38:27 +0000388 int intarg;
Christoph Hellwigedf24ab2008-06-10 10:40:29 -0400389
390 if (!options)
391 goto out;
392
393 /*
394 * strsep changes the string, duplicate it because parse_options
395 * gets called twice
396 */
397 opts = kstrdup(options, GFP_KERNEL);
398 if (!opts)
399 return -ENOMEM;
400
401 while ((p = strsep(&opts, ",")) != NULL) {
402 int token;
403 if (!*p)
404 continue;
405
406 token = match_token(p, tokens, args);
407 switch (token) {
408 case Opt_subvol:
409 *subvol_name = match_strdup(&args[0]);
410 break;
Josef Bacik73f73412009-12-04 17:38:27 +0000411 case Opt_subvolid:
412 intarg = 0;
Josef Bacik4849f012009-12-14 19:18:38 +0000413 error = match_int(&args[0], &intarg);
414 if (!error) {
415 /* we want the original fs_tree */
416 if (!intarg)
417 *subvol_objectid =
418 BTRFS_FS_TREE_OBJECTID;
419 else
420 *subvol_objectid = intarg;
421 }
Josef Bacik73f73412009-12-04 17:38:27 +0000422 break;
Christoph Hellwig43e570b2008-06-10 10:40:46 -0400423 case Opt_device:
424 error = btrfs_scan_one_device(match_strdup(&args[0]),
425 flags, holder, fs_devices);
426 if (error)
427 goto out_free_opts;
428 break;
Christoph Hellwigedf24ab2008-06-10 10:40:29 -0400429 default:
430 break;
431 }
432 }
433
Christoph Hellwig43e570b2008-06-10 10:40:46 -0400434 out_free_opts:
Christoph Hellwigedf24ab2008-06-10 10:40:29 -0400435 kfree(opts);
436 out:
437 /*
438 * If no subvolume name is specified we use the default one. Allocate
Chris Mason3de45862008-11-17 21:02:50 -0500439 * a copy of the string "." here so that code later in the
Christoph Hellwigedf24ab2008-06-10 10:40:29 -0400440 * mount path doesn't care if it's the default volume or another one.
441 */
442 if (!*subvol_name) {
Chris Mason3de45862008-11-17 21:02:50 -0500443 *subvol_name = kstrdup(".", GFP_KERNEL);
Christoph Hellwigedf24ab2008-06-10 10:40:29 -0400444 if (!*subvol_name)
445 return -ENOMEM;
446 }
447 return error;
Chris Mason95e05282007-08-29 09:11:44 -0400448}
449
Josef Bacik73f73412009-12-04 17:38:27 +0000450static struct dentry *get_default_root(struct super_block *sb,
451 u64 subvol_objectid)
452{
453 struct btrfs_root *root = sb->s_fs_info;
454 struct btrfs_root *new_root;
455 struct btrfs_dir_item *di;
456 struct btrfs_path *path;
457 struct btrfs_key location;
458 struct inode *inode;
459 struct dentry *dentry;
460 u64 dir_id;
461 int new = 0;
462
463 /*
464 * We have a specific subvol we want to mount, just setup location and
465 * go look up the root.
466 */
467 if (subvol_objectid) {
468 location.objectid = subvol_objectid;
469 location.type = BTRFS_ROOT_ITEM_KEY;
470 location.offset = (u64)-1;
471 goto find_root;
472 }
473
474 path = btrfs_alloc_path();
475 if (!path)
476 return ERR_PTR(-ENOMEM);
477 path->leave_spinning = 1;
478
479 /*
480 * Find the "default" dir item which points to the root item that we
481 * will mount by default if we haven't been given a specific subvolume
482 * to mount.
483 */
484 dir_id = btrfs_super_root_dir(&root->fs_info->super_copy);
485 di = btrfs_lookup_dir_item(NULL, root, path, dir_id, "default", 7, 0);
Dan Carpenterfb4f6f92010-05-29 09:40:57 +0000486 if (IS_ERR(di))
487 return ERR_CAST(di);
Josef Bacik73f73412009-12-04 17:38:27 +0000488 if (!di) {
489 /*
490 * Ok the default dir item isn't there. This is weird since
491 * it's always been there, but don't freak out, just try and
492 * mount to root most subvolume.
493 */
494 btrfs_free_path(path);
495 dir_id = BTRFS_FIRST_FREE_OBJECTID;
496 new_root = root->fs_info->fs_root;
497 goto setup_root;
498 }
499
500 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
501 btrfs_free_path(path);
502
503find_root:
504 new_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
505 if (IS_ERR(new_root))
Julia Lawalld0b678c2010-10-29 15:14:23 -0400506 return ERR_CAST(new_root);
Josef Bacik73f73412009-12-04 17:38:27 +0000507
508 if (btrfs_root_refs(&new_root->root_item) == 0)
509 return ERR_PTR(-ENOENT);
510
511 dir_id = btrfs_root_dirid(&new_root->root_item);
512setup_root:
513 location.objectid = dir_id;
514 location.type = BTRFS_INODE_ITEM_KEY;
515 location.offset = 0;
516
517 inode = btrfs_iget(sb, &location, new_root, &new);
Dan Carpenter4cbd1142010-05-29 09:42:19 +0000518 if (IS_ERR(inode))
519 return ERR_CAST(inode);
Josef Bacik73f73412009-12-04 17:38:27 +0000520
521 /*
522 * If we're just mounting the root most subvol put the inode and return
523 * a reference to the dentry. We will have already gotten a reference
524 * to the inode in btrfs_fill_super so we're good to go.
525 */
526 if (!new && sb->s_root->d_inode == inode) {
527 iput(inode);
528 return dget(sb->s_root);
529 }
530
531 if (new) {
532 const struct qstr name = { .name = "/", .len = 1 };
533
534 /*
535 * New inode, we need to make the dentry a sibling of s_root so
536 * everything gets cleaned up properly on unmount.
537 */
538 dentry = d_alloc(sb->s_root, &name);
539 if (!dentry) {
540 iput(inode);
541 return ERR_PTR(-ENOMEM);
542 }
543 d_splice_alias(inode, dentry);
544 } else {
545 /*
546 * We found the inode in cache, just find a dentry for it and
547 * put the reference to the inode we just got.
548 */
549 dentry = d_find_alias(inode);
550 iput(inode);
551 }
552
553 return dentry;
554}
555
Chris Masond3977122009-01-05 21:25:51 -0500556static int btrfs_fill_super(struct super_block *sb,
Chris Mason8a4b83c2008-03-24 15:02:07 -0400557 struct btrfs_fs_devices *fs_devices,
Chris Masond3977122009-01-05 21:25:51 -0500558 void *data, int silent)
Chris Mason2e635a22007-03-21 11:12:56 -0400559{
Chris Masond3977122009-01-05 21:25:51 -0500560 struct inode *inode;
561 struct dentry *root_dentry;
Chris Mason0f7d52f2007-04-09 10:42:37 -0400562 struct btrfs_root *tree_root;
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400563 struct btrfs_key key;
Chris Mason39279cc2007-06-12 06:35:45 -0400564 int err;
Chris Mason2e635a22007-03-21 11:12:56 -0400565
566 sb->s_maxbytes = MAX_LFS_FILESIZE;
Chris Mason2e635a22007-03-21 11:12:56 -0400567 sb->s_magic = BTRFS_SUPER_MAGIC;
Chris Masone20d96d2007-03-22 12:13:20 -0400568 sb->s_op = &btrfs_super_ops;
Al Viroaf53d292010-12-20 10:56:06 -0500569 sb->s_d_op = &btrfs_dentry_operations;
Balaji Raobe6e8dc2008-07-21 02:01:56 +0530570 sb->s_export_op = &btrfs_export_ops;
Josef Bacik5103e942007-11-16 11:45:54 -0500571 sb->s_xattr = btrfs_xattr_handlers;
Chris Mason2e635a22007-03-21 11:12:56 -0400572 sb->s_time_gran = 1;
Chris Mason0eda2942009-10-13 13:50:18 -0400573#ifdef CONFIG_BTRFS_FS_POSIX_ACL
Josef Bacik33268ea2008-07-24 12:16:36 -0400574 sb->s_flags |= MS_POSIXACL;
Chris Ball49cf6f42009-09-29 13:51:04 -0400575#endif
Chris Masone20d96d2007-03-22 12:13:20 -0400576
Chris Masondfe25022008-05-13 13:46:40 -0400577 tree_root = open_ctree(sb, fs_devices, (char *)data);
Chris Masond98237b2007-03-28 13:57:48 -0400578
Yane58ca022008-04-01 11:21:34 -0400579 if (IS_ERR(tree_root)) {
Chris Masone20d96d2007-03-22 12:13:20 -0400580 printk("btrfs: open_ctree failed\n");
Yane58ca022008-04-01 11:21:34 -0400581 return PTR_ERR(tree_root);
Chris Masone20d96d2007-03-22 12:13:20 -0400582 }
Chris Mason0f7d52f2007-04-09 10:42:37 -0400583 sb->s_fs_info = tree_root;
Chris Masonb888db22007-08-27 16:49:44 -0400584
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400585 key.objectid = BTRFS_FIRST_FREE_OBJECTID;
586 key.type = BTRFS_INODE_ITEM_KEY;
587 key.offset = 0;
Josef Bacik73f73412009-12-04 17:38:27 +0000588 inode = btrfs_iget(sb, &key, tree_root->fs_info->fs_root, NULL);
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400589 if (IS_ERR(inode)) {
590 err = PTR_ERR(inode);
Chris Mason39279cc2007-06-12 06:35:45 -0400591 goto fail_close;
592 }
Chris Mason2e635a22007-03-21 11:12:56 -0400593
Chris Masone20d96d2007-03-22 12:13:20 -0400594 root_dentry = d_alloc_root(inode);
595 if (!root_dentry) {
Chris Mason2e635a22007-03-21 11:12:56 -0400596 iput(inode);
Chris Mason39279cc2007-06-12 06:35:45 -0400597 err = -ENOMEM;
598 goto fail_close;
Chris Mason2e635a22007-03-21 11:12:56 -0400599 }
Josef Bacik58176a92007-08-29 15:47:34 -0400600
Chris Masone20d96d2007-03-22 12:13:20 -0400601 sb->s_root = root_dentry;
Chris Mason6885f302008-02-20 16:11:05 -0500602
Chris Mason6885f302008-02-20 16:11:05 -0500603 save_mount_options(sb, data);
Chris Mason2e635a22007-03-21 11:12:56 -0400604 return 0;
Chris Mason2e635a22007-03-21 11:12:56 -0400605
Chris Mason39279cc2007-06-12 06:35:45 -0400606fail_close:
607 close_ctree(tree_root);
Chris Masond5719762007-03-23 10:01:08 -0400608 return err;
609}
610
Sage Weil6bf13c02008-06-10 10:07:39 -0400611int btrfs_sync_fs(struct super_block *sb, int wait)
Chris Masond5719762007-03-23 10:01:08 -0400612{
613 struct btrfs_trans_handle *trans;
Sage Weildccae992009-04-02 16:59:01 -0400614 struct btrfs_root *root = btrfs_sb(sb);
Chris Masond5719762007-03-23 10:01:08 -0400615 int ret;
Chris Masondf2ce342007-03-23 11:00:45 -0400616
Chris Masond561c022007-03-23 19:47:49 -0400617 if (!wait) {
Chris Mason7cfcc172007-04-02 14:53:59 -0400618 filemap_flush(root->fs_info->btree_inode->i_mapping);
Chris Masond561c022007-03-23 19:47:49 -0400619 return 0;
620 }
Chris Mason771ed682008-11-06 22:02:51 -0500621
Yan, Zheng24bbcf02009-11-12 09:36:34 +0000622 btrfs_start_delalloc_inodes(root, 0);
623 btrfs_wait_ordered_extents(root, 0, 0);
Chris Mason771ed682008-11-06 22:02:51 -0500624
Yan, Zhenga22285a2010-05-16 10:48:46 -0400625 trans = btrfs_start_transaction(root, 0);
Chris Masond5719762007-03-23 10:01:08 -0400626 ret = btrfs_commit_transaction(trans, root);
Chris Mason54aa1f42007-06-22 14:16:25 -0400627 return ret;
Chris Masond5719762007-03-23 10:01:08 -0400628}
629
Eric Parisa9572a12009-04-02 16:46:06 -0400630static int btrfs_show_options(struct seq_file *seq, struct vfsmount *vfs)
631{
632 struct btrfs_root *root = btrfs_sb(vfs->mnt_sb);
633 struct btrfs_fs_info *info = root->fs_info;
634
635 if (btrfs_test_opt(root, DEGRADED))
636 seq_puts(seq, ",degraded");
637 if (btrfs_test_opt(root, NODATASUM))
638 seq_puts(seq, ",nodatasum");
639 if (btrfs_test_opt(root, NODATACOW))
640 seq_puts(seq, ",nodatacow");
641 if (btrfs_test_opt(root, NOBARRIER))
642 seq_puts(seq, ",nobarrier");
Eric Parisa9572a12009-04-02 16:46:06 -0400643 if (info->max_inline != 8192 * 1024)
Joel Becker21380932009-04-21 12:38:29 -0700644 seq_printf(seq, ",max_inline=%llu",
645 (unsigned long long)info->max_inline);
Eric Parisa9572a12009-04-02 16:46:06 -0400646 if (info->alloc_start != 0)
Joel Becker21380932009-04-21 12:38:29 -0700647 seq_printf(seq, ",alloc_start=%llu",
648 (unsigned long long)info->alloc_start);
Eric Parisa9572a12009-04-02 16:46:06 -0400649 if (info->thread_pool_size != min_t(unsigned long,
650 num_online_cpus() + 2, 8))
651 seq_printf(seq, ",thread_pool=%d", info->thread_pool_size);
652 if (btrfs_test_opt(root, COMPRESS))
653 seq_puts(seq, ",compress");
Chris Masonc2898112009-06-10 09:51:32 -0400654 if (btrfs_test_opt(root, NOSSD))
655 seq_puts(seq, ",nossd");
Chris Mason451d7582009-06-09 20:28:34 -0400656 if (btrfs_test_opt(root, SSD_SPREAD))
657 seq_puts(seq, ",ssd_spread");
658 else if (btrfs_test_opt(root, SSD))
Eric Parisa9572a12009-04-02 16:46:06 -0400659 seq_puts(seq, ",ssd");
Sage Weil3a5e1402009-04-02 16:49:40 -0400660 if (btrfs_test_opt(root, NOTREELOG))
Sage Weil6b65c5c2009-05-14 13:52:21 -0400661 seq_puts(seq, ",notreelog");
Sage Weildccae992009-04-02 16:59:01 -0400662 if (btrfs_test_opt(root, FLUSHONCOMMIT))
Sage Weil6b65c5c2009-05-14 13:52:21 -0400663 seq_puts(seq, ",flushoncommit");
Matthew Wilcox20a52392009-12-14 22:01:12 +0000664 if (btrfs_test_opt(root, DISCARD))
665 seq_puts(seq, ",discard");
Eric Parisa9572a12009-04-02 16:46:06 -0400666 if (!(root->fs_info->sb->s_flags & MS_POSIXACL))
667 seq_puts(seq, ",noacl");
668 return 0;
669}
670
Chris Masona061fc82008-05-07 11:43:44 -0400671static int btrfs_test_super(struct super_block *s, void *data)
Chris Mason2e635a22007-03-21 11:12:56 -0400672{
Josef Bacik450ba0e2010-11-19 14:59:15 -0500673 struct btrfs_root *test_root = data;
Chris Masona061fc82008-05-07 11:43:44 -0400674 struct btrfs_root *root = btrfs_sb(s);
Yan4b82d6e2007-08-29 09:11:44 -0400675
Ian Kent619c8c72010-11-22 02:21:38 +0000676 /*
677 * If this super block is going away, return false as it
678 * can't match as an existing super block.
679 */
680 if (!atomic_read(&s->s_active))
681 return 0;
Josef Bacik450ba0e2010-11-19 14:59:15 -0500682 return root->fs_info->fs_devices == test_root->fs_info->fs_devices;
Yan4b82d6e2007-08-29 09:11:44 -0400683}
684
Josef Bacik450ba0e2010-11-19 14:59:15 -0500685static int btrfs_set_super(struct super_block *s, void *data)
686{
687 s->s_fs_info = data;
688
689 return set_anon_super(s, data);
690}
691
692
Christoph Hellwigedf24ab2008-06-10 10:40:29 -0400693/*
694 * Find a superblock for the given device / mount point.
695 *
696 * Note: This is based on get_sb_bdev from fs/super.c with a few additions
697 * for multiple device setup. Make sure to keep it in sync.
698 */
Al Viro061dbc62010-07-26 16:21:33 +0400699static struct dentry *btrfs_mount(struct file_system_type *fs_type, int flags,
700 const char *dev_name, void *data)
Yan4b82d6e2007-08-29 09:11:44 -0400701{
702 struct block_device *bdev = NULL;
703 struct super_block *s;
704 struct dentry *root;
Chris Mason8a4b83c2008-03-24 15:02:07 -0400705 struct btrfs_fs_devices *fs_devices = NULL;
Josef Bacik450ba0e2010-11-19 14:59:15 -0500706 struct btrfs_root *tree_root = NULL;
707 struct btrfs_fs_info *fs_info = NULL;
Christoph Hellwig97288f22008-12-02 06:36:09 -0500708 fmode_t mode = FMODE_READ;
Josef Bacik73f73412009-12-04 17:38:27 +0000709 char *subvol_name = NULL;
710 u64 subvol_objectid = 0;
Yan4b82d6e2007-08-29 09:11:44 -0400711 int error = 0;
712
Christoph Hellwig97288f22008-12-02 06:36:09 -0500713 if (!(flags & MS_RDONLY))
714 mode |= FMODE_WRITE;
715
716 error = btrfs_parse_early_options(data, mode, fs_type,
Josef Bacik73f73412009-12-04 17:38:27 +0000717 &subvol_name, &subvol_objectid,
718 &fs_devices);
Christoph Hellwigedf24ab2008-06-10 10:40:29 -0400719 if (error)
Al Viro061dbc62010-07-26 16:21:33 +0400720 return ERR_PTR(error);
Christoph Hellwigedf24ab2008-06-10 10:40:29 -0400721
Christoph Hellwig97288f22008-12-02 06:36:09 -0500722 error = btrfs_scan_one_device(dev_name, mode, fs_type, &fs_devices);
Chris Mason8a4b83c2008-03-24 15:02:07 -0400723 if (error)
Christoph Hellwigedf24ab2008-06-10 10:40:29 -0400724 goto error_free_subvol_name;
Yan4b82d6e2007-08-29 09:11:44 -0400725
Christoph Hellwig97288f22008-12-02 06:36:09 -0500726 error = btrfs_open_devices(fs_devices, mode, fs_type);
Chris Mason8a4b83c2008-03-24 15:02:07 -0400727 if (error)
Christoph Hellwigedf24ab2008-06-10 10:40:29 -0400728 goto error_free_subvol_name;
Chris Mason8a4b83c2008-03-24 15:02:07 -0400729
Yan Zheng2b820322008-11-17 21:11:30 -0500730 if (!(flags & MS_RDONLY) && fs_devices->rw_devices == 0) {
731 error = -EACCES;
732 goto error_close_devices;
733 }
734
Josef Bacik450ba0e2010-11-19 14:59:15 -0500735 /*
736 * Setup a dummy root and fs_info for test/set super. This is because
737 * we don't actually fill this stuff out until open_ctree, but we need
738 * it for searching for existing supers, so this lets us do that and
739 * then open_ctree will properly initialize everything later.
740 */
741 fs_info = kzalloc(sizeof(struct btrfs_fs_info), GFP_NOFS);
742 tree_root = kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
743 if (!fs_info || !tree_root) {
744 error = -ENOMEM;
745 goto error_close_devices;
746 }
747 fs_info->tree_root = tree_root;
748 fs_info->fs_devices = fs_devices;
749 tree_root->fs_info = fs_info;
750
Chris Masondfe25022008-05-13 13:46:40 -0400751 bdev = fs_devices->latest_bdev;
Josef Bacik450ba0e2010-11-19 14:59:15 -0500752 s = sget(fs_type, btrfs_test_super, btrfs_set_super, tree_root);
Yan4b82d6e2007-08-29 09:11:44 -0400753 if (IS_ERR(s))
754 goto error_s;
755
756 if (s->s_root) {
757 if ((flags ^ s->s_flags) & MS_RDONLY) {
Al Viro6f5bbff2009-05-06 01:34:22 -0400758 deactivate_locked_super(s);
Yan4b82d6e2007-08-29 09:11:44 -0400759 error = -EBUSY;
Yan Zhengc146afa2008-11-12 14:34:12 -0500760 goto error_close_devices;
Yan4b82d6e2007-08-29 09:11:44 -0400761 }
762
Yan Zheng2b820322008-11-17 21:11:30 -0500763 btrfs_close_devices(fs_devices);
Yan4b82d6e2007-08-29 09:11:44 -0400764 } else {
765 char b[BDEVNAME_SIZE];
766
767 s->s_flags = flags;
768 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
Chris Mason8a4b83c2008-03-24 15:02:07 -0400769 error = btrfs_fill_super(s, fs_devices, data,
770 flags & MS_SILENT ? 1 : 0);
Yan4b82d6e2007-08-29 09:11:44 -0400771 if (error) {
Al Viro6f5bbff2009-05-06 01:34:22 -0400772 deactivate_locked_super(s);
Shen Feng1f483662009-01-05 15:43:42 -0500773 goto error_free_subvol_name;
Yan4b82d6e2007-08-29 09:11:44 -0400774 }
775
Chris Mason788f20e2008-04-28 15:29:42 -0400776 btrfs_sb(s)->fs_info->bdev_holder = fs_type;
Yan4b82d6e2007-08-29 09:11:44 -0400777 s->s_flags |= MS_ACTIVE;
778 }
779
Josef Bacik73f73412009-12-04 17:38:27 +0000780 root = get_default_root(s, subvol_objectid);
781 if (IS_ERR(root)) {
782 error = PTR_ERR(root);
783 deactivate_locked_super(s);
Josef Bacik0e783402010-10-22 15:26:53 -0400784 goto error_free_subvol_name;
Josef Bacik73f73412009-12-04 17:38:27 +0000785 }
786 /* if they gave us a subvolume name bind mount into that */
787 if (strcmp(subvol_name, ".")) {
788 struct dentry *new_root;
789 mutex_lock(&root->d_inode->i_mutex);
790 new_root = lookup_one_len(subvol_name, root,
Chris Masond3977122009-01-05 21:25:51 -0500791 strlen(subvol_name));
Josef Bacik73f73412009-12-04 17:38:27 +0000792 mutex_unlock(&root->d_inode->i_mutex);
Chris Masond3977122009-01-05 21:25:51 -0500793
Josef Bacik73f73412009-12-04 17:38:27 +0000794 if (IS_ERR(new_root)) {
Li Zefanf106e822010-12-07 01:51:26 +0000795 dput(root);
Al Viro6f5bbff2009-05-06 01:34:22 -0400796 deactivate_locked_super(s);
Josef Bacik73f73412009-12-04 17:38:27 +0000797 error = PTR_ERR(new_root);
Josef Bacik0e783402010-10-22 15:26:53 -0400798 goto error_free_subvol_name;
Josef Bacik73f73412009-12-04 17:38:27 +0000799 }
800 if (!new_root->d_inode) {
801 dput(root);
802 dput(new_root);
Al Viro6f5bbff2009-05-06 01:34:22 -0400803 deactivate_locked_super(s);
David Woodhouse76fcef12008-08-19 16:49:35 +0100804 error = -ENXIO;
Josef Bacik0e783402010-10-22 15:26:53 -0400805 goto error_free_subvol_name;
David Woodhouse76fcef12008-08-19 16:49:35 +0100806 }
Josef Bacik73f73412009-12-04 17:38:27 +0000807 dput(root);
808 root = new_root;
Yan4b82d6e2007-08-29 09:11:44 -0400809 }
810
Christoph Hellwigedf24ab2008-06-10 10:40:29 -0400811 kfree(subvol_name);
Al Viro061dbc62010-07-26 16:21:33 +0400812 return root;
Yan4b82d6e2007-08-29 09:11:44 -0400813
814error_s:
815 error = PTR_ERR(s);
Yan Zhengc146afa2008-11-12 14:34:12 -0500816error_close_devices:
Chris Mason8a4b83c2008-03-24 15:02:07 -0400817 btrfs_close_devices(fs_devices);
Josef Bacik450ba0e2010-11-19 14:59:15 -0500818 kfree(fs_info);
819 kfree(tree_root);
Christoph Hellwigedf24ab2008-06-10 10:40:29 -0400820error_free_subvol_name:
821 kfree(subvol_name);
Al Viro061dbc62010-07-26 16:21:33 +0400822 return ERR_PTR(error);
Yan4b82d6e2007-08-29 09:11:44 -0400823}
Chris Mason2e635a22007-03-21 11:12:56 -0400824
Yan Zhengc146afa2008-11-12 14:34:12 -0500825static int btrfs_remount(struct super_block *sb, int *flags, char *data)
826{
827 struct btrfs_root *root = btrfs_sb(sb);
828 int ret;
829
Chris Masonb2880522009-02-12 09:37:35 -0500830 ret = btrfs_parse_options(root, data);
831 if (ret)
832 return -EINVAL;
833
Yan Zhengc146afa2008-11-12 14:34:12 -0500834 if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
835 return 0;
836
837 if (*flags & MS_RDONLY) {
838 sb->s_flags |= MS_RDONLY;
839
840 ret = btrfs_commit_super(root);
841 WARN_ON(ret);
842 } else {
Yan Zheng2b820322008-11-17 21:11:30 -0500843 if (root->fs_info->fs_devices->rw_devices == 0)
844 return -EACCES;
845
Yan Zhengc146afa2008-11-12 14:34:12 -0500846 if (btrfs_super_log_root(&root->fs_info->super_copy) != 0)
847 return -EINVAL;
848
Yan, Zhengd68fc572010-05-16 10:49:58 -0400849 ret = btrfs_cleanup_fs_roots(root->fs_info);
Yan Zhengc146afa2008-11-12 14:34:12 -0500850 WARN_ON(ret);
851
Yan, Zhengd68fc572010-05-16 10:49:58 -0400852 /* recover relocation */
853 ret = btrfs_recover_relocation(root);
Yan Zhengc146afa2008-11-12 14:34:12 -0500854 WARN_ON(ret);
855
856 sb->s_flags &= ~MS_RDONLY;
857 }
858
859 return 0;
860}
861
Miao Xie6d07bce2011-01-05 10:07:31 +0000862/*
863 * The helper to calc the free space on the devices that can be used to store
864 * file data.
865 */
866static int btrfs_calc_avail_data_space(struct btrfs_root *root, u64 *free_bytes)
867{
868 struct btrfs_fs_info *fs_info = root->fs_info;
869 struct btrfs_device_info *devices_info;
870 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
871 struct btrfs_device *device;
872 u64 skip_space;
873 u64 type;
874 u64 avail_space;
875 u64 used_space;
876 u64 min_stripe_size;
877 int min_stripes = 1;
878 int i = 0, nr_devices;
879 int ret;
880
881 nr_devices = fs_info->fs_devices->rw_devices;
882 BUG_ON(!nr_devices);
883
884 devices_info = kmalloc(sizeof(*devices_info) * nr_devices,
885 GFP_NOFS);
886 if (!devices_info)
887 return -ENOMEM;
888
889 /* calc min stripe number for data space alloction */
890 type = btrfs_get_alloc_profile(root, 1);
891 if (type & BTRFS_BLOCK_GROUP_RAID0)
892 min_stripes = 2;
893 else if (type & BTRFS_BLOCK_GROUP_RAID1)
894 min_stripes = 2;
895 else if (type & BTRFS_BLOCK_GROUP_RAID10)
896 min_stripes = 4;
897
898 if (type & BTRFS_BLOCK_GROUP_DUP)
899 min_stripe_size = 2 * BTRFS_STRIPE_LEN;
900 else
901 min_stripe_size = BTRFS_STRIPE_LEN;
902
903 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
904 if (!device->in_fs_metadata)
905 continue;
906
907 avail_space = device->total_bytes - device->bytes_used;
908
909 /* align with stripe_len */
910 do_div(avail_space, BTRFS_STRIPE_LEN);
911 avail_space *= BTRFS_STRIPE_LEN;
912
913 /*
914 * In order to avoid overwritting the superblock on the drive,
915 * btrfs starts at an offset of at least 1MB when doing chunk
916 * allocation.
917 */
918 skip_space = 1024 * 1024;
919
920 /* user can set the offset in fs_info->alloc_start. */
921 if (fs_info->alloc_start + BTRFS_STRIPE_LEN <=
922 device->total_bytes)
923 skip_space = max(fs_info->alloc_start, skip_space);
924
925 /*
926 * btrfs can not use the free space in [0, skip_space - 1],
927 * we must subtract it from the total. In order to implement
928 * it, we account the used space in this range first.
929 */
930 ret = btrfs_account_dev_extents_size(device, 0, skip_space - 1,
931 &used_space);
932 if (ret) {
933 kfree(devices_info);
934 return ret;
935 }
936
937 /* calc the free space in [0, skip_space - 1] */
938 skip_space -= used_space;
939
940 /*
941 * we can use the free space in [0, skip_space - 1], subtract
942 * it from the total.
943 */
944 if (avail_space && avail_space >= skip_space)
945 avail_space -= skip_space;
946 else
947 avail_space = 0;
948
949 if (avail_space < min_stripe_size)
950 continue;
951
952 devices_info[i].dev = device;
953 devices_info[i].max_avail = avail_space;
954
955 i++;
956 }
957
958 nr_devices = i;
959
960 btrfs_descending_sort_devices(devices_info, nr_devices);
961
962 i = nr_devices - 1;
963 avail_space = 0;
964 while (nr_devices >= min_stripes) {
965 if (devices_info[i].max_avail >= min_stripe_size) {
966 int j;
967 u64 alloc_size;
968
969 avail_space += devices_info[i].max_avail * min_stripes;
970 alloc_size = devices_info[i].max_avail;
971 for (j = i + 1 - min_stripes; j <= i; j++)
972 devices_info[j].max_avail -= alloc_size;
973 }
974 i--;
975 nr_devices--;
976 }
977
978 kfree(devices_info);
979 *free_bytes = avail_space;
980 return 0;
981}
982
Chris Mason8fd17792007-04-19 21:01:03 -0400983static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
984{
985 struct btrfs_root *root = btrfs_sb(dentry->d_sb);
Chris Mason4b52dff2007-06-26 10:06:50 -0400986 struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
Josef Bacikbd4d1082010-03-05 21:59:21 +0000987 struct list_head *head = &root->fs_info->space_info;
988 struct btrfs_space_info *found;
989 u64 total_used = 0;
Miao Xie6d07bce2011-01-05 10:07:31 +0000990 u64 total_free_data = 0;
Chris Masondb945352007-10-15 16:15:53 -0400991 int bits = dentry->d_sb->s_blocksize_bits;
David Woodhouse9d036322008-08-18 12:01:52 +0100992 __be32 *fsid = (__be32 *)root->fs_info->fsid;
Miao Xie6d07bce2011-01-05 10:07:31 +0000993 int ret;
Chris Mason8fd17792007-04-19 21:01:03 -0400994
Miao Xie6d07bce2011-01-05 10:07:31 +0000995 /* holding chunk_muext to avoid allocating new chunks */
996 mutex_lock(&root->fs_info->chunk_mutex);
Josef Bacikbd4d1082010-03-05 21:59:21 +0000997 rcu_read_lock();
Josef Bacik89a55892010-10-14 14:52:27 -0400998 list_for_each_entry_rcu(found, head, list) {
Miao Xie6d07bce2011-01-05 10:07:31 +0000999 if (found->flags & BTRFS_BLOCK_GROUP_DATA) {
1000 total_free_data += found->disk_total - found->disk_used;
1001 total_free_data -=
1002 btrfs_account_ro_block_groups_free_space(found);
1003 }
1004
Yan, Zhengb742bb82010-05-16 10:46:24 -04001005 total_used += found->disk_used;
Josef Bacik89a55892010-10-14 14:52:27 -04001006 }
Josef Bacikbd4d1082010-03-05 21:59:21 +00001007 rcu_read_unlock();
1008
Chris Mason8fd17792007-04-19 21:01:03 -04001009 buf->f_namelen = BTRFS_NAME_LEN;
Chris Masondb945352007-10-15 16:15:53 -04001010 buf->f_blocks = btrfs_super_total_bytes(disk_super) >> bits;
Josef Bacikbd4d1082010-03-05 21:59:21 +00001011 buf->f_bfree = buf->f_blocks - (total_used >> bits);
Chris Mason8fd17792007-04-19 21:01:03 -04001012 buf->f_bsize = dentry->d_sb->s_blocksize;
1013 buf->f_type = BTRFS_SUPER_MAGIC;
Miao Xie6d07bce2011-01-05 10:07:31 +00001014 buf->f_bavail = total_free_data;
1015 ret = btrfs_calc_avail_data_space(root, &total_free_data);
1016 if (ret) {
1017 mutex_unlock(&root->fs_info->chunk_mutex);
1018 return ret;
1019 }
1020 buf->f_bavail += total_free_data;
1021 buf->f_bavail = buf->f_bavail >> bits;
1022 mutex_unlock(&root->fs_info->chunk_mutex);
Chris Masond3977122009-01-05 21:25:51 -05001023
David Woodhouse9d036322008-08-18 12:01:52 +01001024 /* We treat it as constant endianness (it doesn't matter _which_)
Chris Masond3977122009-01-05 21:25:51 -05001025 because we want the fsid to come out the same whether mounted
David Woodhouse9d036322008-08-18 12:01:52 +01001026 on a big-endian or little-endian host */
1027 buf->f_fsid.val[0] = be32_to_cpu(fsid[0]) ^ be32_to_cpu(fsid[2]);
1028 buf->f_fsid.val[1] = be32_to_cpu(fsid[1]) ^ be32_to_cpu(fsid[3]);
David Woodhouse32d48fa2008-08-18 13:10:20 +01001029 /* Mask in the root object ID too, to disambiguate subvols */
1030 buf->f_fsid.val[0] ^= BTRFS_I(dentry->d_inode)->root->objectid >> 32;
1031 buf->f_fsid.val[1] ^= BTRFS_I(dentry->d_inode)->root->objectid;
1032
Chris Mason8fd17792007-04-19 21:01:03 -04001033 return 0;
1034}
Chris Masonb5133862007-04-24 11:52:22 -04001035
Chris Mason2e635a22007-03-21 11:12:56 -04001036static struct file_system_type btrfs_fs_type = {
1037 .owner = THIS_MODULE,
1038 .name = "btrfs",
Al Viro061dbc62010-07-26 16:21:33 +04001039 .mount = btrfs_mount,
Chris Masona061fc82008-05-07 11:43:44 -04001040 .kill_sb = kill_anon_super,
Chris Mason2e635a22007-03-21 11:12:56 -04001041 .fs_flags = FS_REQUIRES_DEV,
1042};
Chris Masona9218f62008-03-24 15:02:04 -04001043
Chris Masond352ac62008-09-29 15:18:18 -04001044/*
1045 * used by btrfsctl to scan devices when no FS is mounted
1046 */
Chris Mason8a4b83c2008-03-24 15:02:07 -04001047static long btrfs_control_ioctl(struct file *file, unsigned int cmd,
1048 unsigned long arg)
1049{
1050 struct btrfs_ioctl_vol_args *vol;
1051 struct btrfs_fs_devices *fs_devices;
Chris Masonc071fcf2009-01-16 11:59:08 -05001052 int ret = -ENOTTY;
Chris Mason8a4b83c2008-03-24 15:02:07 -04001053
Chris Masone441d542009-01-05 16:57:23 -05001054 if (!capable(CAP_SYS_ADMIN))
1055 return -EPERM;
1056
Li Zefandae7b662009-04-08 15:06:54 +08001057 vol = memdup_user((void __user *)arg, sizeof(*vol));
1058 if (IS_ERR(vol))
1059 return PTR_ERR(vol);
Chris Masonc071fcf2009-01-16 11:59:08 -05001060
Chris Mason8a4b83c2008-03-24 15:02:07 -04001061 switch (cmd) {
1062 case BTRFS_IOC_SCAN_DEV:
Christoph Hellwig97288f22008-12-02 06:36:09 -05001063 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
Chris Mason8a4b83c2008-03-24 15:02:07 -04001064 &btrfs_fs_type, &fs_devices);
1065 break;
1066 }
Li Zefandae7b662009-04-08 15:06:54 +08001067
Chris Mason8a4b83c2008-03-24 15:02:07 -04001068 kfree(vol);
Linda Knippersf819d832008-06-09 22:17:11 -04001069 return ret;
Chris Mason8a4b83c2008-03-24 15:02:07 -04001070}
1071
Linus Torvalds01762602009-01-10 06:09:52 -08001072static int btrfs_freeze(struct super_block *sb)
Yaned0dab62008-01-22 12:46:56 -05001073{
1074 struct btrfs_root *root = btrfs_sb(sb);
Chris Masona74a4b92008-06-25 16:01:31 -04001075 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1076 mutex_lock(&root->fs_info->cleaner_mutex);
Linus Torvalds01762602009-01-10 06:09:52 -08001077 return 0;
Yaned0dab62008-01-22 12:46:56 -05001078}
1079
Linus Torvalds01762602009-01-10 06:09:52 -08001080static int btrfs_unfreeze(struct super_block *sb)
Yaned0dab62008-01-22 12:46:56 -05001081{
1082 struct btrfs_root *root = btrfs_sb(sb);
Chris Masona74a4b92008-06-25 16:01:31 -04001083 mutex_unlock(&root->fs_info->cleaner_mutex);
1084 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
Linus Torvalds01762602009-01-10 06:09:52 -08001085 return 0;
Yaned0dab62008-01-22 12:46:56 -05001086}
Chris Mason2e635a22007-03-21 11:12:56 -04001087
Alexey Dobriyanb87221d2009-09-21 17:01:09 -07001088static const struct super_operations btrfs_super_ops = {
Yan, Zheng76dda932009-09-21 16:00:26 -04001089 .drop_inode = btrfs_drop_inode,
Al Virobd555972010-06-07 11:35:40 -04001090 .evict_inode = btrfs_evict_inode,
Chris Masone20d96d2007-03-22 12:13:20 -04001091 .put_super = btrfs_put_super,
Chris Masond5719762007-03-23 10:01:08 -04001092 .sync_fs = btrfs_sync_fs,
Eric Parisa9572a12009-04-02 16:46:06 -04001093 .show_options = btrfs_show_options,
Chris Mason4730a4b2007-03-26 12:00:39 -04001094 .write_inode = btrfs_write_inode,
Chris Masonb5133862007-04-24 11:52:22 -04001095 .dirty_inode = btrfs_dirty_inode,
Chris Mason2c90e5d2007-04-02 10:50:19 -04001096 .alloc_inode = btrfs_alloc_inode,
1097 .destroy_inode = btrfs_destroy_inode,
Chris Mason8fd17792007-04-19 21:01:03 -04001098 .statfs = btrfs_statfs,
Yan Zhengc146afa2008-11-12 14:34:12 -05001099 .remount_fs = btrfs_remount,
Linus Torvalds01762602009-01-10 06:09:52 -08001100 .freeze_fs = btrfs_freeze,
1101 .unfreeze_fs = btrfs_unfreeze,
Chris Masone20d96d2007-03-22 12:13:20 -04001102};
Chris Masona9218f62008-03-24 15:02:04 -04001103
1104static const struct file_operations btrfs_ctl_fops = {
1105 .unlocked_ioctl = btrfs_control_ioctl,
1106 .compat_ioctl = btrfs_control_ioctl,
1107 .owner = THIS_MODULE,
Arnd Bergmann6038f372010-08-15 18:52:59 +02001108 .llseek = noop_llseek,
Chris Masona9218f62008-03-24 15:02:04 -04001109};
1110
1111static struct miscdevice btrfs_misc = {
Kay Sievers578454f2010-05-20 18:07:20 +02001112 .minor = BTRFS_MINOR,
Chris Masona9218f62008-03-24 15:02:04 -04001113 .name = "btrfs-control",
1114 .fops = &btrfs_ctl_fops
1115};
1116
Kay Sievers578454f2010-05-20 18:07:20 +02001117MODULE_ALIAS_MISCDEV(BTRFS_MINOR);
1118MODULE_ALIAS("devname:btrfs-control");
1119
Chris Masona9218f62008-03-24 15:02:04 -04001120static int btrfs_interface_init(void)
1121{
1122 return misc_register(&btrfs_misc);
1123}
1124
Christoph Hellwigb2950862008-12-02 09:54:17 -05001125static void btrfs_interface_exit(void)
Chris Masona9218f62008-03-24 15:02:04 -04001126{
1127 if (misc_deregister(&btrfs_misc) < 0)
Chris Masond3977122009-01-05 21:25:51 -05001128 printk(KERN_INFO "misc_deregister failed for control device");
Chris Masona9218f62008-03-24 15:02:04 -04001129}
1130
Chris Mason2e635a22007-03-21 11:12:56 -04001131static int __init init_btrfs_fs(void)
1132{
Chris Mason2c90e5d2007-04-02 10:50:19 -04001133 int err;
Josef Bacik58176a92007-08-29 15:47:34 -04001134
1135 err = btrfs_init_sysfs();
1136 if (err)
1137 return err;
1138
Li Zefan261507a02010-12-17 14:21:50 +08001139 err = btrfs_init_compress();
Chris Mason2c90e5d2007-04-02 10:50:19 -04001140 if (err)
Chris Masona74a4b92008-06-25 16:01:31 -04001141 goto free_sysfs;
Chris Masond1310b22008-01-24 16:13:08 -05001142
Li Zefan261507a02010-12-17 14:21:50 +08001143 err = btrfs_init_cachep();
1144 if (err)
1145 goto free_compress;
1146
Chris Masond1310b22008-01-24 16:13:08 -05001147 err = extent_io_init();
Wyatt Banks2f4cbe62007-11-19 10:22:33 -05001148 if (err)
1149 goto free_cachep;
1150
Chris Masond1310b22008-01-24 16:13:08 -05001151 err = extent_map_init();
1152 if (err)
1153 goto free_extent_io;
1154
Chris Masona9218f62008-03-24 15:02:04 -04001155 err = btrfs_interface_init();
Wyatt Banks2f4cbe62007-11-19 10:22:33 -05001156 if (err)
1157 goto free_extent_map;
Chris Masonc8b97812008-10-29 14:49:59 -04001158
Chris Masona9218f62008-03-24 15:02:04 -04001159 err = register_filesystem(&btrfs_fs_type);
1160 if (err)
1161 goto unregister_ioctl;
Chris Masonb3c3da72008-07-23 12:12:13 -04001162
1163 printk(KERN_INFO "%s loaded\n", BTRFS_BUILD_VERSION);
Wyatt Banks2f4cbe62007-11-19 10:22:33 -05001164 return 0;
1165
Chris Masona9218f62008-03-24 15:02:04 -04001166unregister_ioctl:
1167 btrfs_interface_exit();
Wyatt Banks2f4cbe62007-11-19 10:22:33 -05001168free_extent_map:
1169 extent_map_exit();
Chris Masond1310b22008-01-24 16:13:08 -05001170free_extent_io:
1171 extent_io_exit();
Wyatt Banks2f4cbe62007-11-19 10:22:33 -05001172free_cachep:
1173 btrfs_destroy_cachep();
Li Zefan261507a02010-12-17 14:21:50 +08001174free_compress:
1175 btrfs_exit_compress();
Chris Masona74a4b92008-06-25 16:01:31 -04001176free_sysfs:
Wyatt Banks2f4cbe62007-11-19 10:22:33 -05001177 btrfs_exit_sysfs();
1178 return err;
Chris Mason2e635a22007-03-21 11:12:56 -04001179}
1180
1181static void __exit exit_btrfs_fs(void)
1182{
Chris Mason39279cc2007-06-12 06:35:45 -04001183 btrfs_destroy_cachep();
Chris Masona52d9a82007-08-27 16:49:44 -04001184 extent_map_exit();
Chris Masond1310b22008-01-24 16:13:08 -05001185 extent_io_exit();
Chris Masona9218f62008-03-24 15:02:04 -04001186 btrfs_interface_exit();
Chris Mason2e635a22007-03-21 11:12:56 -04001187 unregister_filesystem(&btrfs_fs_type);
Josef Bacik58176a92007-08-29 15:47:34 -04001188 btrfs_exit_sysfs();
Chris Mason8a4b83c2008-03-24 15:02:07 -04001189 btrfs_cleanup_fs_uuids();
Li Zefan261507a02010-12-17 14:21:50 +08001190 btrfs_exit_compress();
Chris Mason2e635a22007-03-21 11:12:56 -04001191}
1192
1193module_init(init_btrfs_fs)
1194module_exit(exit_btrfs_fs)
1195
1196MODULE_LICENSE("GPL");