Btrfs: setup blank root and fs_info for mount time
[linux-3.10.git] / fs / btrfs / super.c
1 /*
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
19 #include <linux/blkdev.h>
20 #include <linux/module.h>
21 #include <linux/buffer_head.h>
22 #include <linux/fs.h>
23 #include <linux/pagemap.h>
24 #include <linux/highmem.h>
25 #include <linux/time.h>
26 #include <linux/init.h>
27 #include <linux/seq_file.h>
28 #include <linux/string.h>
29 #include <linux/backing-dev.h>
30 #include <linux/mount.h>
31 #include <linux/mpage.h>
32 #include <linux/swap.h>
33 #include <linux/writeback.h>
34 #include <linux/statfs.h>
35 #include <linux/compat.h>
36 #include <linux/parser.h>
37 #include <linux/ctype.h>
38 #include <linux/namei.h>
39 #include <linux/miscdevice.h>
40 #include <linux/magic.h>
41 #include <linux/slab.h>
42 #include "compat.h"
43 #include "ctree.h"
44 #include "disk-io.h"
45 #include "transaction.h"
46 #include "btrfs_inode.h"
47 #include "ioctl.h"
48 #include "print-tree.h"
49 #include "xattr.h"
50 #include "volumes.h"
51 #include "version.h"
52 #include "export.h"
53 #include "compression.h"
54
55 static const struct super_operations btrfs_super_ops;
56
57 static void btrfs_put_super(struct super_block *sb)
58 {
59         struct btrfs_root *root = btrfs_sb(sb);
60         int ret;
61
62         ret = close_ctree(root);
63         sb->s_fs_info = NULL;
64
65         (void)ret; /* FIXME: need to fix VFS to return error? */
66 }
67
68 enum {
69         Opt_degraded, Opt_subvol, Opt_subvolid, Opt_device, Opt_nodatasum,
70         Opt_nodatacow, Opt_max_inline, Opt_alloc_start, Opt_nobarrier, Opt_ssd,
71         Opt_nossd, Opt_ssd_spread, Opt_thread_pool, Opt_noacl, Opt_compress,
72         Opt_compress_force, Opt_notreelog, Opt_ratio, Opt_flushoncommit,
73         Opt_discard, Opt_space_cache, Opt_clear_cache, Opt_err,
74         Opt_user_subvol_rm_allowed,
75 };
76
77 static match_table_t tokens = {
78         {Opt_degraded, "degraded"},
79         {Opt_subvol, "subvol=%s"},
80         {Opt_subvolid, "subvolid=%d"},
81         {Opt_device, "device=%s"},
82         {Opt_nodatasum, "nodatasum"},
83         {Opt_nodatacow, "nodatacow"},
84         {Opt_nobarrier, "nobarrier"},
85         {Opt_max_inline, "max_inline=%s"},
86         {Opt_alloc_start, "alloc_start=%s"},
87         {Opt_thread_pool, "thread_pool=%d"},
88         {Opt_compress, "compress"},
89         {Opt_compress_force, "compress-force"},
90         {Opt_ssd, "ssd"},
91         {Opt_ssd_spread, "ssd_spread"},
92         {Opt_nossd, "nossd"},
93         {Opt_noacl, "noacl"},
94         {Opt_notreelog, "notreelog"},
95         {Opt_flushoncommit, "flushoncommit"},
96         {Opt_ratio, "metadata_ratio=%d"},
97         {Opt_discard, "discard"},
98         {Opt_space_cache, "space_cache"},
99         {Opt_clear_cache, "clear_cache"},
100         {Opt_user_subvol_rm_allowed, "user_subvol_rm_allowed"},
101         {Opt_err, NULL},
102 };
103
104 /*
105  * Regular mount options parser.  Everything that is needed only when
106  * reading in a new superblock is parsed here.
107  */
108 int btrfs_parse_options(struct btrfs_root *root, char *options)
109 {
110         struct btrfs_fs_info *info = root->fs_info;
111         substring_t args[MAX_OPT_ARGS];
112         char *p, *num, *orig;
113         int intarg;
114         int ret = 0;
115
116         if (!options)
117                 return 0;
118
119         /*
120          * strsep changes the string, duplicate it because parse_options
121          * gets called twice
122          */
123         options = kstrdup(options, GFP_NOFS);
124         if (!options)
125                 return -ENOMEM;
126
127         orig = options;
128
129         while ((p = strsep(&options, ",")) != NULL) {
130                 int token;
131                 if (!*p)
132                         continue;
133
134                 token = match_token(p, tokens, args);
135                 switch (token) {
136                 case Opt_degraded:
137                         printk(KERN_INFO "btrfs: allowing degraded mounts\n");
138                         btrfs_set_opt(info->mount_opt, DEGRADED);
139                         break;
140                 case Opt_subvol:
141                 case Opt_subvolid:
142                 case Opt_device:
143                         /*
144                          * These are parsed by btrfs_parse_early_options
145                          * and can be happily ignored here.
146                          */
147                         break;
148                 case Opt_nodatasum:
149                         printk(KERN_INFO "btrfs: setting nodatasum\n");
150                         btrfs_set_opt(info->mount_opt, NODATASUM);
151                         break;
152                 case Opt_nodatacow:
153                         printk(KERN_INFO "btrfs: setting nodatacow\n");
154                         btrfs_set_opt(info->mount_opt, NODATACOW);
155                         btrfs_set_opt(info->mount_opt, NODATASUM);
156                         break;
157                 case Opt_compress:
158                         printk(KERN_INFO "btrfs: use compression\n");
159                         btrfs_set_opt(info->mount_opt, COMPRESS);
160                         break;
161                 case Opt_compress_force:
162                         printk(KERN_INFO "btrfs: forcing compression\n");
163                         btrfs_set_opt(info->mount_opt, FORCE_COMPRESS);
164                         btrfs_set_opt(info->mount_opt, COMPRESS);
165                         break;
166                 case Opt_ssd:
167                         printk(KERN_INFO "btrfs: use ssd allocation scheme\n");
168                         btrfs_set_opt(info->mount_opt, SSD);
169                         break;
170                 case Opt_ssd_spread:
171                         printk(KERN_INFO "btrfs: use spread ssd "
172                                "allocation scheme\n");
173                         btrfs_set_opt(info->mount_opt, SSD);
174                         btrfs_set_opt(info->mount_opt, SSD_SPREAD);
175                         break;
176                 case Opt_nossd:
177                         printk(KERN_INFO "btrfs: not using ssd allocation "
178                                "scheme\n");
179                         btrfs_set_opt(info->mount_opt, NOSSD);
180                         btrfs_clear_opt(info->mount_opt, SSD);
181                         btrfs_clear_opt(info->mount_opt, SSD_SPREAD);
182                         break;
183                 case Opt_nobarrier:
184                         printk(KERN_INFO "btrfs: turning off barriers\n");
185                         btrfs_set_opt(info->mount_opt, NOBARRIER);
186                         break;
187                 case Opt_thread_pool:
188                         intarg = 0;
189                         match_int(&args[0], &intarg);
190                         if (intarg) {
191                                 info->thread_pool_size = intarg;
192                                 printk(KERN_INFO "btrfs: thread pool %d\n",
193                                        info->thread_pool_size);
194                         }
195                         break;
196                 case Opt_max_inline:
197                         num = match_strdup(&args[0]);
198                         if (num) {
199                                 info->max_inline = memparse(num, NULL);
200                                 kfree(num);
201
202                                 if (info->max_inline) {
203                                         info->max_inline = max_t(u64,
204                                                 info->max_inline,
205                                                 root->sectorsize);
206                                 }
207                                 printk(KERN_INFO "btrfs: max_inline at %llu\n",
208                                         (unsigned long long)info->max_inline);
209                         }
210                         break;
211                 case Opt_alloc_start:
212                         num = match_strdup(&args[0]);
213                         if (num) {
214                                 info->alloc_start = memparse(num, NULL);
215                                 kfree(num);
216                                 printk(KERN_INFO
217                                         "btrfs: allocations start at %llu\n",
218                                         (unsigned long long)info->alloc_start);
219                         }
220                         break;
221                 case Opt_noacl:
222                         root->fs_info->sb->s_flags &= ~MS_POSIXACL;
223                         break;
224                 case Opt_notreelog:
225                         printk(KERN_INFO "btrfs: disabling tree log\n");
226                         btrfs_set_opt(info->mount_opt, NOTREELOG);
227                         break;
228                 case Opt_flushoncommit:
229                         printk(KERN_INFO "btrfs: turning on flush-on-commit\n");
230                         btrfs_set_opt(info->mount_opt, FLUSHONCOMMIT);
231                         break;
232                 case Opt_ratio:
233                         intarg = 0;
234                         match_int(&args[0], &intarg);
235                         if (intarg) {
236                                 info->metadata_ratio = intarg;
237                                 printk(KERN_INFO "btrfs: metadata ratio %d\n",
238                                        info->metadata_ratio);
239                         }
240                         break;
241                 case Opt_discard:
242                         btrfs_set_opt(info->mount_opt, DISCARD);
243                         break;
244                 case Opt_space_cache:
245                         printk(KERN_INFO "btrfs: enabling disk space caching\n");
246                         btrfs_set_opt(info->mount_opt, SPACE_CACHE);
247                         break;
248                 case Opt_clear_cache:
249                         printk(KERN_INFO "btrfs: force clearing of disk cache\n");
250                         btrfs_set_opt(info->mount_opt, CLEAR_CACHE);
251                         break;
252                 case Opt_user_subvol_rm_allowed:
253                         btrfs_set_opt(info->mount_opt, USER_SUBVOL_RM_ALLOWED);
254                         break;
255                 case Opt_err:
256                         printk(KERN_INFO "btrfs: unrecognized mount option "
257                                "'%s'\n", p);
258                         ret = -EINVAL;
259                         goto out;
260                 default:
261                         break;
262                 }
263         }
264 out:
265         kfree(orig);
266         return ret;
267 }
268
269 /*
270  * Parse mount options that are required early in the mount process.
271  *
272  * All other options will be parsed on much later in the mount process and
273  * only when we need to allocate a new super block.
274  */
275 static int btrfs_parse_early_options(const char *options, fmode_t flags,
276                 void *holder, char **subvol_name, u64 *subvol_objectid,
277                 struct btrfs_fs_devices **fs_devices)
278 {
279         substring_t args[MAX_OPT_ARGS];
280         char *opts, *p;
281         int error = 0;
282         int intarg;
283
284         if (!options)
285                 goto out;
286
287         /*
288          * strsep changes the string, duplicate it because parse_options
289          * gets called twice
290          */
291         opts = kstrdup(options, GFP_KERNEL);
292         if (!opts)
293                 return -ENOMEM;
294
295         while ((p = strsep(&opts, ",")) != NULL) {
296                 int token;
297                 if (!*p)
298                         continue;
299
300                 token = match_token(p, tokens, args);
301                 switch (token) {
302                 case Opt_subvol:
303                         *subvol_name = match_strdup(&args[0]);
304                         break;
305                 case Opt_subvolid:
306                         intarg = 0;
307                         error = match_int(&args[0], &intarg);
308                         if (!error) {
309                                 /* we want the original fs_tree */
310                                 if (!intarg)
311                                         *subvol_objectid =
312                                                 BTRFS_FS_TREE_OBJECTID;
313                                 else
314                                         *subvol_objectid = intarg;
315                         }
316                         break;
317                 case Opt_device:
318                         error = btrfs_scan_one_device(match_strdup(&args[0]),
319                                         flags, holder, fs_devices);
320                         if (error)
321                                 goto out_free_opts;
322                         break;
323                 default:
324                         break;
325                 }
326         }
327
328  out_free_opts:
329         kfree(opts);
330  out:
331         /*
332          * If no subvolume name is specified we use the default one.  Allocate
333          * a copy of the string "." here so that code later in the
334          * mount path doesn't care if it's the default volume or another one.
335          */
336         if (!*subvol_name) {
337                 *subvol_name = kstrdup(".", GFP_KERNEL);
338                 if (!*subvol_name)
339                         return -ENOMEM;
340         }
341         return error;
342 }
343
344 static struct dentry *get_default_root(struct super_block *sb,
345                                        u64 subvol_objectid)
346 {
347         struct btrfs_root *root = sb->s_fs_info;
348         struct btrfs_root *new_root;
349         struct btrfs_dir_item *di;
350         struct btrfs_path *path;
351         struct btrfs_key location;
352         struct inode *inode;
353         struct dentry *dentry;
354         u64 dir_id;
355         int new = 0;
356
357         /*
358          * We have a specific subvol we want to mount, just setup location and
359          * go look up the root.
360          */
361         if (subvol_objectid) {
362                 location.objectid = subvol_objectid;
363                 location.type = BTRFS_ROOT_ITEM_KEY;
364                 location.offset = (u64)-1;
365                 goto find_root;
366         }
367
368         path = btrfs_alloc_path();
369         if (!path)
370                 return ERR_PTR(-ENOMEM);
371         path->leave_spinning = 1;
372
373         /*
374          * Find the "default" dir item which points to the root item that we
375          * will mount by default if we haven't been given a specific subvolume
376          * to mount.
377          */
378         dir_id = btrfs_super_root_dir(&root->fs_info->super_copy);
379         di = btrfs_lookup_dir_item(NULL, root, path, dir_id, "default", 7, 0);
380         if (IS_ERR(di))
381                 return ERR_CAST(di);
382         if (!di) {
383                 /*
384                  * Ok the default dir item isn't there.  This is weird since
385                  * it's always been there, but don't freak out, just try and
386                  * mount to root most subvolume.
387                  */
388                 btrfs_free_path(path);
389                 dir_id = BTRFS_FIRST_FREE_OBJECTID;
390                 new_root = root->fs_info->fs_root;
391                 goto setup_root;
392         }
393
394         btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
395         btrfs_free_path(path);
396
397 find_root:
398         new_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
399         if (IS_ERR(new_root))
400                 return ERR_CAST(new_root);
401
402         if (btrfs_root_refs(&new_root->root_item) == 0)
403                 return ERR_PTR(-ENOENT);
404
405         dir_id = btrfs_root_dirid(&new_root->root_item);
406 setup_root:
407         location.objectid = dir_id;
408         location.type = BTRFS_INODE_ITEM_KEY;
409         location.offset = 0;
410
411         inode = btrfs_iget(sb, &location, new_root, &new);
412         if (IS_ERR(inode))
413                 return ERR_CAST(inode);
414
415         /*
416          * If we're just mounting the root most subvol put the inode and return
417          * a reference to the dentry.  We will have already gotten a reference
418          * to the inode in btrfs_fill_super so we're good to go.
419          */
420         if (!new && sb->s_root->d_inode == inode) {
421                 iput(inode);
422                 return dget(sb->s_root);
423         }
424
425         if (new) {
426                 const struct qstr name = { .name = "/", .len = 1 };
427
428                 /*
429                  * New inode, we need to make the dentry a sibling of s_root so
430                  * everything gets cleaned up properly on unmount.
431                  */
432                 dentry = d_alloc(sb->s_root, &name);
433                 if (!dentry) {
434                         iput(inode);
435                         return ERR_PTR(-ENOMEM);
436                 }
437                 d_splice_alias(inode, dentry);
438         } else {
439                 /*
440                  * We found the inode in cache, just find a dentry for it and
441                  * put the reference to the inode we just got.
442                  */
443                 dentry = d_find_alias(inode);
444                 iput(inode);
445         }
446
447         return dentry;
448 }
449
450 static int btrfs_fill_super(struct super_block *sb,
451                             struct btrfs_fs_devices *fs_devices,
452                             void *data, int silent)
453 {
454         struct inode *inode;
455         struct dentry *root_dentry;
456         struct btrfs_root *tree_root;
457         struct btrfs_key key;
458         int err;
459
460         sb->s_maxbytes = MAX_LFS_FILESIZE;
461         sb->s_magic = BTRFS_SUPER_MAGIC;
462         sb->s_op = &btrfs_super_ops;
463         sb->s_export_op = &btrfs_export_ops;
464         sb->s_xattr = btrfs_xattr_handlers;
465         sb->s_time_gran = 1;
466 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
467         sb->s_flags |= MS_POSIXACL;
468 #endif
469
470         tree_root = open_ctree(sb, fs_devices, (char *)data);
471
472         if (IS_ERR(tree_root)) {
473                 printk("btrfs: open_ctree failed\n");
474                 return PTR_ERR(tree_root);
475         }
476         sb->s_fs_info = tree_root;
477
478         key.objectid = BTRFS_FIRST_FREE_OBJECTID;
479         key.type = BTRFS_INODE_ITEM_KEY;
480         key.offset = 0;
481         inode = btrfs_iget(sb, &key, tree_root->fs_info->fs_root, NULL);
482         if (IS_ERR(inode)) {
483                 err = PTR_ERR(inode);
484                 goto fail_close;
485         }
486
487         root_dentry = d_alloc_root(inode);
488         if (!root_dentry) {
489                 iput(inode);
490                 err = -ENOMEM;
491                 goto fail_close;
492         }
493
494         sb->s_root = root_dentry;
495
496         save_mount_options(sb, data);
497         return 0;
498
499 fail_close:
500         close_ctree(tree_root);
501         return err;
502 }
503
504 int btrfs_sync_fs(struct super_block *sb, int wait)
505 {
506         struct btrfs_trans_handle *trans;
507         struct btrfs_root *root = btrfs_sb(sb);
508         int ret;
509
510         if (!wait) {
511                 filemap_flush(root->fs_info->btree_inode->i_mapping);
512                 return 0;
513         }
514
515         btrfs_start_delalloc_inodes(root, 0);
516         btrfs_wait_ordered_extents(root, 0, 0);
517
518         trans = btrfs_start_transaction(root, 0);
519         ret = btrfs_commit_transaction(trans, root);
520         return ret;
521 }
522
523 static int btrfs_show_options(struct seq_file *seq, struct vfsmount *vfs)
524 {
525         struct btrfs_root *root = btrfs_sb(vfs->mnt_sb);
526         struct btrfs_fs_info *info = root->fs_info;
527
528         if (btrfs_test_opt(root, DEGRADED))
529                 seq_puts(seq, ",degraded");
530         if (btrfs_test_opt(root, NODATASUM))
531                 seq_puts(seq, ",nodatasum");
532         if (btrfs_test_opt(root, NODATACOW))
533                 seq_puts(seq, ",nodatacow");
534         if (btrfs_test_opt(root, NOBARRIER))
535                 seq_puts(seq, ",nobarrier");
536         if (info->max_inline != 8192 * 1024)
537                 seq_printf(seq, ",max_inline=%llu",
538                            (unsigned long long)info->max_inline);
539         if (info->alloc_start != 0)
540                 seq_printf(seq, ",alloc_start=%llu",
541                            (unsigned long long)info->alloc_start);
542         if (info->thread_pool_size !=  min_t(unsigned long,
543                                              num_online_cpus() + 2, 8))
544                 seq_printf(seq, ",thread_pool=%d", info->thread_pool_size);
545         if (btrfs_test_opt(root, COMPRESS))
546                 seq_puts(seq, ",compress");
547         if (btrfs_test_opt(root, NOSSD))
548                 seq_puts(seq, ",nossd");
549         if (btrfs_test_opt(root, SSD_SPREAD))
550                 seq_puts(seq, ",ssd_spread");
551         else if (btrfs_test_opt(root, SSD))
552                 seq_puts(seq, ",ssd");
553         if (btrfs_test_opt(root, NOTREELOG))
554                 seq_puts(seq, ",notreelog");
555         if (btrfs_test_opt(root, FLUSHONCOMMIT))
556                 seq_puts(seq, ",flushoncommit");
557         if (btrfs_test_opt(root, DISCARD))
558                 seq_puts(seq, ",discard");
559         if (!(root->fs_info->sb->s_flags & MS_POSIXACL))
560                 seq_puts(seq, ",noacl");
561         return 0;
562 }
563
564 static int btrfs_test_super(struct super_block *s, void *data)
565 {
566         struct btrfs_root *test_root = data;
567         struct btrfs_root *root = btrfs_sb(s);
568
569         /*
570          * If this super block is going away, return false as it
571          * can't match as an existing super block.
572          */
573         if (!atomic_read(&s->s_active))
574                 return 0;
575         return root->fs_info->fs_devices == test_root->fs_info->fs_devices;
576 }
577
578 static int btrfs_set_super(struct super_block *s, void *data)
579 {
580         s->s_fs_info = data;
581
582         return set_anon_super(s, data);
583 }
584
585
586 /*
587  * Find a superblock for the given device / mount point.
588  *
589  * Note:  This is based on get_sb_bdev from fs/super.c with a few additions
590  *        for multiple device setup.  Make sure to keep it in sync.
591  */
592 static int btrfs_get_sb(struct file_system_type *fs_type, int flags,
593                 const char *dev_name, void *data, struct vfsmount *mnt)
594 {
595         struct block_device *bdev = NULL;
596         struct super_block *s;
597         struct dentry *root;
598         struct btrfs_fs_devices *fs_devices = NULL;
599         struct btrfs_root *tree_root = NULL;
600         struct btrfs_fs_info *fs_info = NULL;
601         fmode_t mode = FMODE_READ;
602         char *subvol_name = NULL;
603         u64 subvol_objectid = 0;
604         int error = 0;
605
606         if (!(flags & MS_RDONLY))
607                 mode |= FMODE_WRITE;
608
609         error = btrfs_parse_early_options(data, mode, fs_type,
610                                           &subvol_name, &subvol_objectid,
611                                           &fs_devices);
612         if (error)
613                 return error;
614
615         error = btrfs_scan_one_device(dev_name, mode, fs_type, &fs_devices);
616         if (error)
617                 goto error_free_subvol_name;
618
619         error = btrfs_open_devices(fs_devices, mode, fs_type);
620         if (error)
621                 goto error_free_subvol_name;
622
623         if (!(flags & MS_RDONLY) && fs_devices->rw_devices == 0) {
624                 error = -EACCES;
625                 goto error_close_devices;
626         }
627
628         /*
629          * Setup a dummy root and fs_info for test/set super.  This is because
630          * we don't actually fill this stuff out until open_ctree, but we need
631          * it for searching for existing supers, so this lets us do that and
632          * then open_ctree will properly initialize everything later.
633          */
634         fs_info = kzalloc(sizeof(struct btrfs_fs_info), GFP_NOFS);
635         tree_root = kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
636         if (!fs_info || !tree_root) {
637                 error = -ENOMEM;
638                 goto error_close_devices;
639         }
640         fs_info->tree_root = tree_root;
641         fs_info->fs_devices = fs_devices;
642         tree_root->fs_info = fs_info;
643
644         bdev = fs_devices->latest_bdev;
645         s = sget(fs_type, btrfs_test_super, btrfs_set_super, tree_root);
646         if (IS_ERR(s))
647                 goto error_s;
648
649         if (s->s_root) {
650                 if ((flags ^ s->s_flags) & MS_RDONLY) {
651                         deactivate_locked_super(s);
652                         error = -EBUSY;
653                         goto error_close_devices;
654                 }
655
656                 btrfs_close_devices(fs_devices);
657         } else {
658                 char b[BDEVNAME_SIZE];
659
660                 s->s_flags = flags;
661                 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
662                 error = btrfs_fill_super(s, fs_devices, data,
663                                          flags & MS_SILENT ? 1 : 0);
664                 if (error) {
665                         deactivate_locked_super(s);
666                         goto error_free_subvol_name;
667                 }
668
669                 btrfs_sb(s)->fs_info->bdev_holder = fs_type;
670                 s->s_flags |= MS_ACTIVE;
671         }
672
673         root = get_default_root(s, subvol_objectid);
674         if (IS_ERR(root)) {
675                 error = PTR_ERR(root);
676                 deactivate_locked_super(s);
677                 goto error_free_subvol_name;
678         }
679         /* if they gave us a subvolume name bind mount into that */
680         if (strcmp(subvol_name, ".")) {
681                 struct dentry *new_root;
682                 mutex_lock(&root->d_inode->i_mutex);
683                 new_root = lookup_one_len(subvol_name, root,
684                                       strlen(subvol_name));
685                 mutex_unlock(&root->d_inode->i_mutex);
686
687                 if (IS_ERR(new_root)) {
688                         deactivate_locked_super(s);
689                         error = PTR_ERR(new_root);
690                         dput(root);
691                         goto error_free_subvol_name;
692                 }
693                 if (!new_root->d_inode) {
694                         dput(root);
695                         dput(new_root);
696                         deactivate_locked_super(s);
697                         error = -ENXIO;
698                         goto error_free_subvol_name;
699                 }
700                 dput(root);
701                 root = new_root;
702         }
703
704         mnt->mnt_sb = s;
705         mnt->mnt_root = root;
706
707         kfree(subvol_name);
708         return 0;
709
710 error_s:
711         error = PTR_ERR(s);
712 error_close_devices:
713         btrfs_close_devices(fs_devices);
714         kfree(fs_info);
715         kfree(tree_root);
716 error_free_subvol_name:
717         kfree(subvol_name);
718         return error;
719 }
720
721 static int btrfs_remount(struct super_block *sb, int *flags, char *data)
722 {
723         struct btrfs_root *root = btrfs_sb(sb);
724         int ret;
725
726         ret = btrfs_parse_options(root, data);
727         if (ret)
728                 return -EINVAL;
729
730         if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
731                 return 0;
732
733         if (*flags & MS_RDONLY) {
734                 sb->s_flags |= MS_RDONLY;
735
736                 ret =  btrfs_commit_super(root);
737                 WARN_ON(ret);
738         } else {
739                 if (root->fs_info->fs_devices->rw_devices == 0)
740                         return -EACCES;
741
742                 if (btrfs_super_log_root(&root->fs_info->super_copy) != 0)
743                         return -EINVAL;
744
745                 ret = btrfs_cleanup_fs_roots(root->fs_info);
746                 WARN_ON(ret);
747
748                 /* recover relocation */
749                 ret = btrfs_recover_relocation(root);
750                 WARN_ON(ret);
751
752                 sb->s_flags &= ~MS_RDONLY;
753         }
754
755         return 0;
756 }
757
758 static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
759 {
760         struct btrfs_root *root = btrfs_sb(dentry->d_sb);
761         struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
762         struct list_head *head = &root->fs_info->space_info;
763         struct btrfs_space_info *found;
764         u64 total_used = 0;
765         u64 total_used_data = 0;
766         int bits = dentry->d_sb->s_blocksize_bits;
767         __be32 *fsid = (__be32 *)root->fs_info->fsid;
768
769         rcu_read_lock();
770         list_for_each_entry_rcu(found, head, list) {
771                 if (found->flags & (BTRFS_BLOCK_GROUP_METADATA |
772                                     BTRFS_BLOCK_GROUP_SYSTEM))
773                         total_used_data += found->disk_total;
774                 else
775                         total_used_data += found->disk_used;
776                 total_used += found->disk_used;
777         }
778         rcu_read_unlock();
779
780         buf->f_namelen = BTRFS_NAME_LEN;
781         buf->f_blocks = btrfs_super_total_bytes(disk_super) >> bits;
782         buf->f_bfree = buf->f_blocks - (total_used >> bits);
783         buf->f_bavail = buf->f_blocks - (total_used_data >> bits);
784         buf->f_bsize = dentry->d_sb->s_blocksize;
785         buf->f_type = BTRFS_SUPER_MAGIC;
786
787         /* We treat it as constant endianness (it doesn't matter _which_)
788            because we want the fsid to come out the same whether mounted
789            on a big-endian or little-endian host */
790         buf->f_fsid.val[0] = be32_to_cpu(fsid[0]) ^ be32_to_cpu(fsid[2]);
791         buf->f_fsid.val[1] = be32_to_cpu(fsid[1]) ^ be32_to_cpu(fsid[3]);
792         /* Mask in the root object ID too, to disambiguate subvols */
793         buf->f_fsid.val[0] ^= BTRFS_I(dentry->d_inode)->root->objectid >> 32;
794         buf->f_fsid.val[1] ^= BTRFS_I(dentry->d_inode)->root->objectid;
795
796         return 0;
797 }
798
799 static struct file_system_type btrfs_fs_type = {
800         .owner          = THIS_MODULE,
801         .name           = "btrfs",
802         .get_sb         = btrfs_get_sb,
803         .kill_sb        = kill_anon_super,
804         .fs_flags       = FS_REQUIRES_DEV,
805 };
806
807 /*
808  * used by btrfsctl to scan devices when no FS is mounted
809  */
810 static long btrfs_control_ioctl(struct file *file, unsigned int cmd,
811                                 unsigned long arg)
812 {
813         struct btrfs_ioctl_vol_args *vol;
814         struct btrfs_fs_devices *fs_devices;
815         int ret = -ENOTTY;
816
817         if (!capable(CAP_SYS_ADMIN))
818                 return -EPERM;
819
820         vol = memdup_user((void __user *)arg, sizeof(*vol));
821         if (IS_ERR(vol))
822                 return PTR_ERR(vol);
823
824         switch (cmd) {
825         case BTRFS_IOC_SCAN_DEV:
826                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
827                                             &btrfs_fs_type, &fs_devices);
828                 break;
829         }
830
831         kfree(vol);
832         return ret;
833 }
834
835 static int btrfs_freeze(struct super_block *sb)
836 {
837         struct btrfs_root *root = btrfs_sb(sb);
838         mutex_lock(&root->fs_info->transaction_kthread_mutex);
839         mutex_lock(&root->fs_info->cleaner_mutex);
840         return 0;
841 }
842
843 static int btrfs_unfreeze(struct super_block *sb)
844 {
845         struct btrfs_root *root = btrfs_sb(sb);
846         mutex_unlock(&root->fs_info->cleaner_mutex);
847         mutex_unlock(&root->fs_info->transaction_kthread_mutex);
848         return 0;
849 }
850
851 static const struct super_operations btrfs_super_ops = {
852         .drop_inode     = btrfs_drop_inode,
853         .evict_inode    = btrfs_evict_inode,
854         .put_super      = btrfs_put_super,
855         .sync_fs        = btrfs_sync_fs,
856         .show_options   = btrfs_show_options,
857         .write_inode    = btrfs_write_inode,
858         .dirty_inode    = btrfs_dirty_inode,
859         .alloc_inode    = btrfs_alloc_inode,
860         .destroy_inode  = btrfs_destroy_inode,
861         .statfs         = btrfs_statfs,
862         .remount_fs     = btrfs_remount,
863         .freeze_fs      = btrfs_freeze,
864         .unfreeze_fs    = btrfs_unfreeze,
865 };
866
867 static const struct file_operations btrfs_ctl_fops = {
868         .unlocked_ioctl  = btrfs_control_ioctl,
869         .compat_ioctl = btrfs_control_ioctl,
870         .owner   = THIS_MODULE,
871 };
872
873 static struct miscdevice btrfs_misc = {
874         .minor          = BTRFS_MINOR,
875         .name           = "btrfs-control",
876         .fops           = &btrfs_ctl_fops
877 };
878
879 MODULE_ALIAS_MISCDEV(BTRFS_MINOR);
880 MODULE_ALIAS("devname:btrfs-control");
881
882 static int btrfs_interface_init(void)
883 {
884         return misc_register(&btrfs_misc);
885 }
886
887 static void btrfs_interface_exit(void)
888 {
889         if (misc_deregister(&btrfs_misc) < 0)
890                 printk(KERN_INFO "misc_deregister failed for control device");
891 }
892
893 static int __init init_btrfs_fs(void)
894 {
895         int err;
896
897         err = btrfs_init_sysfs();
898         if (err)
899                 return err;
900
901         err = btrfs_init_cachep();
902         if (err)
903                 goto free_sysfs;
904
905         err = extent_io_init();
906         if (err)
907                 goto free_cachep;
908
909         err = extent_map_init();
910         if (err)
911                 goto free_extent_io;
912
913         err = btrfs_interface_init();
914         if (err)
915                 goto free_extent_map;
916
917         err = register_filesystem(&btrfs_fs_type);
918         if (err)
919                 goto unregister_ioctl;
920
921         printk(KERN_INFO "%s loaded\n", BTRFS_BUILD_VERSION);
922         return 0;
923
924 unregister_ioctl:
925         btrfs_interface_exit();
926 free_extent_map:
927         extent_map_exit();
928 free_extent_io:
929         extent_io_exit();
930 free_cachep:
931         btrfs_destroy_cachep();
932 free_sysfs:
933         btrfs_exit_sysfs();
934         return err;
935 }
936
937 static void __exit exit_btrfs_fs(void)
938 {
939         btrfs_destroy_cachep();
940         extent_map_exit();
941         extent_io_exit();
942         btrfs_interface_exit();
943         unregister_filesystem(&btrfs_fs_type);
944         btrfs_exit_sysfs();
945         btrfs_cleanup_fs_uuids();
946         btrfs_zlib_exit();
947 }
948
949 module_init(init_btrfs_fs)
950 module_exit(exit_btrfs_fs)
951
952 MODULE_LICENSE("GPL");