Merge git://git.kernel.org/pub/scm/linux/kernel/git/mason/btrfs-unstable
[linux-2.6.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
66 enum {
67         Opt_degraded, Opt_subvol, Opt_subvolid, Opt_device, Opt_nodatasum,
68         Opt_nodatacow, Opt_max_inline, Opt_alloc_start, Opt_nobarrier, Opt_ssd,
69         Opt_nossd, Opt_ssd_spread, Opt_thread_pool, Opt_noacl, Opt_compress,
70         Opt_compress_force, Opt_notreelog, Opt_ratio, Opt_flushoncommit,
71         Opt_discard, Opt_err,
72 };
73
74 static match_table_t tokens = {
75         {Opt_degraded, "degraded"},
76         {Opt_subvol, "subvol=%s"},
77         {Opt_subvolid, "subvolid=%d"},
78         {Opt_device, "device=%s"},
79         {Opt_nodatasum, "nodatasum"},
80         {Opt_nodatacow, "nodatacow"},
81         {Opt_nobarrier, "nobarrier"},
82         {Opt_max_inline, "max_inline=%s"},
83         {Opt_alloc_start, "alloc_start=%s"},
84         {Opt_thread_pool, "thread_pool=%d"},
85         {Opt_compress, "compress"},
86         {Opt_compress_force, "compress-force"},
87         {Opt_ssd, "ssd"},
88         {Opt_ssd_spread, "ssd_spread"},
89         {Opt_nossd, "nossd"},
90         {Opt_noacl, "noacl"},
91         {Opt_notreelog, "notreelog"},
92         {Opt_flushoncommit, "flushoncommit"},
93         {Opt_ratio, "metadata_ratio=%d"},
94         {Opt_discard, "discard"},
95         {Opt_err, NULL},
96 };
97
98 /*
99  * Regular mount options parser.  Everything that is needed only when
100  * reading in a new superblock is parsed here.
101  */
102 int btrfs_parse_options(struct btrfs_root *root, char *options)
103 {
104         struct btrfs_fs_info *info = root->fs_info;
105         substring_t args[MAX_OPT_ARGS];
106         char *p, *num, *orig;
107         int intarg;
108         int ret = 0;
109
110         if (!options)
111                 return 0;
112
113         /*
114          * strsep changes the string, duplicate it because parse_options
115          * gets called twice
116          */
117         options = kstrdup(options, GFP_NOFS);
118         if (!options)
119                 return -ENOMEM;
120
121         orig = options;
122
123         while ((p = strsep(&options, ",")) != NULL) {
124                 int token;
125                 if (!*p)
126                         continue;
127
128                 token = match_token(p, tokens, args);
129                 switch (token) {
130                 case Opt_degraded:
131                         printk(KERN_INFO "btrfs: allowing degraded mounts\n");
132                         btrfs_set_opt(info->mount_opt, DEGRADED);
133                         break;
134                 case Opt_subvol:
135                 case Opt_subvolid:
136                 case Opt_device:
137                         /*
138                          * These are parsed by btrfs_parse_early_options
139                          * and can be happily ignored here.
140                          */
141                         break;
142                 case Opt_nodatasum:
143                         printk(KERN_INFO "btrfs: setting nodatasum\n");
144                         btrfs_set_opt(info->mount_opt, NODATASUM);
145                         break;
146                 case Opt_nodatacow:
147                         printk(KERN_INFO "btrfs: setting nodatacow\n");
148                         btrfs_set_opt(info->mount_opt, NODATACOW);
149                         btrfs_set_opt(info->mount_opt, NODATASUM);
150                         break;
151                 case Opt_compress:
152                         printk(KERN_INFO "btrfs: use compression\n");
153                         btrfs_set_opt(info->mount_opt, COMPRESS);
154                         break;
155                 case Opt_compress_force:
156                         printk(KERN_INFO "btrfs: forcing compression\n");
157                         btrfs_set_opt(info->mount_opt, FORCE_COMPRESS);
158                         btrfs_set_opt(info->mount_opt, COMPRESS);
159                         break;
160                 case Opt_ssd:
161                         printk(KERN_INFO "btrfs: use ssd allocation scheme\n");
162                         btrfs_set_opt(info->mount_opt, SSD);
163                         break;
164                 case Opt_ssd_spread:
165                         printk(KERN_INFO "btrfs: use spread ssd "
166                                "allocation scheme\n");
167                         btrfs_set_opt(info->mount_opt, SSD);
168                         btrfs_set_opt(info->mount_opt, SSD_SPREAD);
169                         break;
170                 case Opt_nossd:
171                         printk(KERN_INFO "btrfs: not using ssd allocation "
172                                "scheme\n");
173                         btrfs_set_opt(info->mount_opt, NOSSD);
174                         btrfs_clear_opt(info->mount_opt, SSD);
175                         btrfs_clear_opt(info->mount_opt, SSD_SPREAD);
176                         break;
177                 case Opt_nobarrier:
178                         printk(KERN_INFO "btrfs: turning off barriers\n");
179                         btrfs_set_opt(info->mount_opt, NOBARRIER);
180                         break;
181                 case Opt_thread_pool:
182                         intarg = 0;
183                         match_int(&args[0], &intarg);
184                         if (intarg) {
185                                 info->thread_pool_size = intarg;
186                                 printk(KERN_INFO "btrfs: thread pool %d\n",
187                                        info->thread_pool_size);
188                         }
189                         break;
190                 case Opt_max_inline:
191                         num = match_strdup(&args[0]);
192                         if (num) {
193                                 info->max_inline = memparse(num, NULL);
194                                 kfree(num);
195
196                                 if (info->max_inline) {
197                                         info->max_inline = max_t(u64,
198                                                 info->max_inline,
199                                                 root->sectorsize);
200                                 }
201                                 printk(KERN_INFO "btrfs: max_inline at %llu\n",
202                                         (unsigned long long)info->max_inline);
203                         }
204                         break;
205                 case Opt_alloc_start:
206                         num = match_strdup(&args[0]);
207                         if (num) {
208                                 info->alloc_start = memparse(num, NULL);
209                                 kfree(num);
210                                 printk(KERN_INFO
211                                         "btrfs: allocations start at %llu\n",
212                                         (unsigned long long)info->alloc_start);
213                         }
214                         break;
215                 case Opt_noacl:
216                         root->fs_info->sb->s_flags &= ~MS_POSIXACL;
217                         break;
218                 case Opt_notreelog:
219                         printk(KERN_INFO "btrfs: disabling tree log\n");
220                         btrfs_set_opt(info->mount_opt, NOTREELOG);
221                         break;
222                 case Opt_flushoncommit:
223                         printk(KERN_INFO "btrfs: turning on flush-on-commit\n");
224                         btrfs_set_opt(info->mount_opt, FLUSHONCOMMIT);
225                         break;
226                 case Opt_ratio:
227                         intarg = 0;
228                         match_int(&args[0], &intarg);
229                         if (intarg) {
230                                 info->metadata_ratio = intarg;
231                                 printk(KERN_INFO "btrfs: metadata ratio %d\n",
232                                        info->metadata_ratio);
233                         }
234                         break;
235                 case Opt_discard:
236                         btrfs_set_opt(info->mount_opt, DISCARD);
237                         break;
238                 case Opt_err:
239                         printk(KERN_INFO "btrfs: unrecognized mount option "
240                                "'%s'\n", p);
241                         ret = -EINVAL;
242                         goto out;
243                 default:
244                         break;
245                 }
246         }
247 out:
248         kfree(orig);
249         return ret;
250 }
251
252 /*
253  * Parse mount options that are required early in the mount process.
254  *
255  * All other options will be parsed on much later in the mount process and
256  * only when we need to allocate a new super block.
257  */
258 static int btrfs_parse_early_options(const char *options, fmode_t flags,
259                 void *holder, char **subvol_name, u64 *subvol_objectid,
260                 struct btrfs_fs_devices **fs_devices)
261 {
262         substring_t args[MAX_OPT_ARGS];
263         char *opts, *p;
264         int error = 0;
265         int intarg;
266
267         if (!options)
268                 goto out;
269
270         /*
271          * strsep changes the string, duplicate it because parse_options
272          * gets called twice
273          */
274         opts = kstrdup(options, GFP_KERNEL);
275         if (!opts)
276                 return -ENOMEM;
277
278         while ((p = strsep(&opts, ",")) != NULL) {
279                 int token;
280                 if (!*p)
281                         continue;
282
283                 token = match_token(p, tokens, args);
284                 switch (token) {
285                 case Opt_subvol:
286                         *subvol_name = match_strdup(&args[0]);
287                         break;
288                 case Opt_subvolid:
289                         intarg = 0;
290                         error = match_int(&args[0], &intarg);
291                         if (!error) {
292                                 /* we want the original fs_tree */
293                                 if (!intarg)
294                                         *subvol_objectid =
295                                                 BTRFS_FS_TREE_OBJECTID;
296                                 else
297                                         *subvol_objectid = intarg;
298                         }
299                         break;
300                 case Opt_device:
301                         error = btrfs_scan_one_device(match_strdup(&args[0]),
302                                         flags, holder, fs_devices);
303                         if (error)
304                                 goto out_free_opts;
305                         break;
306                 default:
307                         break;
308                 }
309         }
310
311  out_free_opts:
312         kfree(opts);
313  out:
314         /*
315          * If no subvolume name is specified we use the default one.  Allocate
316          * a copy of the string "." here so that code later in the
317          * mount path doesn't care if it's the default volume or another one.
318          */
319         if (!*subvol_name) {
320                 *subvol_name = kstrdup(".", GFP_KERNEL);
321                 if (!*subvol_name)
322                         return -ENOMEM;
323         }
324         return error;
325 }
326
327 static struct dentry *get_default_root(struct super_block *sb,
328                                        u64 subvol_objectid)
329 {
330         struct btrfs_root *root = sb->s_fs_info;
331         struct btrfs_root *new_root;
332         struct btrfs_dir_item *di;
333         struct btrfs_path *path;
334         struct btrfs_key location;
335         struct inode *inode;
336         struct dentry *dentry;
337         u64 dir_id;
338         int new = 0;
339
340         /*
341          * We have a specific subvol we want to mount, just setup location and
342          * go look up the root.
343          */
344         if (subvol_objectid) {
345                 location.objectid = subvol_objectid;
346                 location.type = BTRFS_ROOT_ITEM_KEY;
347                 location.offset = (u64)-1;
348                 goto find_root;
349         }
350
351         path = btrfs_alloc_path();
352         if (!path)
353                 return ERR_PTR(-ENOMEM);
354         path->leave_spinning = 1;
355
356         /*
357          * Find the "default" dir item which points to the root item that we
358          * will mount by default if we haven't been given a specific subvolume
359          * to mount.
360          */
361         dir_id = btrfs_super_root_dir(&root->fs_info->super_copy);
362         di = btrfs_lookup_dir_item(NULL, root, path, dir_id, "default", 7, 0);
363         if (!di) {
364                 /*
365                  * Ok the default dir item isn't there.  This is weird since
366                  * it's always been there, but don't freak out, just try and
367                  * mount to root most subvolume.
368                  */
369                 btrfs_free_path(path);
370                 dir_id = BTRFS_FIRST_FREE_OBJECTID;
371                 new_root = root->fs_info->fs_root;
372                 goto setup_root;
373         }
374
375         btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
376         btrfs_free_path(path);
377
378 find_root:
379         new_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
380         if (IS_ERR(new_root))
381                 return ERR_PTR(PTR_ERR(new_root));
382
383         if (btrfs_root_refs(&new_root->root_item) == 0)
384                 return ERR_PTR(-ENOENT);
385
386         dir_id = btrfs_root_dirid(&new_root->root_item);
387 setup_root:
388         location.objectid = dir_id;
389         location.type = BTRFS_INODE_ITEM_KEY;
390         location.offset = 0;
391
392         inode = btrfs_iget(sb, &location, new_root, &new);
393         if (!inode)
394                 return ERR_PTR(-ENOMEM);
395
396         /*
397          * If we're just mounting the root most subvol put the inode and return
398          * a reference to the dentry.  We will have already gotten a reference
399          * to the inode in btrfs_fill_super so we're good to go.
400          */
401         if (!new && sb->s_root->d_inode == inode) {
402                 iput(inode);
403                 return dget(sb->s_root);
404         }
405
406         if (new) {
407                 const struct qstr name = { .name = "/", .len = 1 };
408
409                 /*
410                  * New inode, we need to make the dentry a sibling of s_root so
411                  * everything gets cleaned up properly on unmount.
412                  */
413                 dentry = d_alloc(sb->s_root, &name);
414                 if (!dentry) {
415                         iput(inode);
416                         return ERR_PTR(-ENOMEM);
417                 }
418                 d_splice_alias(inode, dentry);
419         } else {
420                 /*
421                  * We found the inode in cache, just find a dentry for it and
422                  * put the reference to the inode we just got.
423                  */
424                 dentry = d_find_alias(inode);
425                 iput(inode);
426         }
427
428         return dentry;
429 }
430
431 static int btrfs_fill_super(struct super_block *sb,
432                             struct btrfs_fs_devices *fs_devices,
433                             void *data, int silent)
434 {
435         struct inode *inode;
436         struct dentry *root_dentry;
437         struct btrfs_super_block *disk_super;
438         struct btrfs_root *tree_root;
439         struct btrfs_key key;
440         int err;
441
442         sb->s_maxbytes = MAX_LFS_FILESIZE;
443         sb->s_magic = BTRFS_SUPER_MAGIC;
444         sb->s_op = &btrfs_super_ops;
445         sb->s_export_op = &btrfs_export_ops;
446         sb->s_xattr = btrfs_xattr_handlers;
447         sb->s_time_gran = 1;
448 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
449         sb->s_flags |= MS_POSIXACL;
450 #endif
451
452         tree_root = open_ctree(sb, fs_devices, (char *)data);
453
454         if (IS_ERR(tree_root)) {
455                 printk("btrfs: open_ctree failed\n");
456                 return PTR_ERR(tree_root);
457         }
458         sb->s_fs_info = tree_root;
459         disk_super = &tree_root->fs_info->super_copy;
460
461         key.objectid = BTRFS_FIRST_FREE_OBJECTID;
462         key.type = BTRFS_INODE_ITEM_KEY;
463         key.offset = 0;
464         inode = btrfs_iget(sb, &key, tree_root->fs_info->fs_root, NULL);
465         if (IS_ERR(inode)) {
466                 err = PTR_ERR(inode);
467                 goto fail_close;
468         }
469
470         root_dentry = d_alloc_root(inode);
471         if (!root_dentry) {
472                 iput(inode);
473                 err = -ENOMEM;
474                 goto fail_close;
475         }
476
477         sb->s_root = root_dentry;
478
479         save_mount_options(sb, data);
480         return 0;
481
482 fail_close:
483         close_ctree(tree_root);
484         return err;
485 }
486
487 int btrfs_sync_fs(struct super_block *sb, int wait)
488 {
489         struct btrfs_trans_handle *trans;
490         struct btrfs_root *root = btrfs_sb(sb);
491         int ret;
492
493         if (!wait) {
494                 filemap_flush(root->fs_info->btree_inode->i_mapping);
495                 return 0;
496         }
497
498         btrfs_start_delalloc_inodes(root, 0);
499         btrfs_wait_ordered_extents(root, 0, 0);
500
501         trans = btrfs_start_transaction(root, 1);
502         ret = btrfs_commit_transaction(trans, root);
503         return ret;
504 }
505
506 static int btrfs_show_options(struct seq_file *seq, struct vfsmount *vfs)
507 {
508         struct btrfs_root *root = btrfs_sb(vfs->mnt_sb);
509         struct btrfs_fs_info *info = root->fs_info;
510
511         if (btrfs_test_opt(root, DEGRADED))
512                 seq_puts(seq, ",degraded");
513         if (btrfs_test_opt(root, NODATASUM))
514                 seq_puts(seq, ",nodatasum");
515         if (btrfs_test_opt(root, NODATACOW))
516                 seq_puts(seq, ",nodatacow");
517         if (btrfs_test_opt(root, NOBARRIER))
518                 seq_puts(seq, ",nobarrier");
519         if (info->max_inline != 8192 * 1024)
520                 seq_printf(seq, ",max_inline=%llu",
521                            (unsigned long long)info->max_inline);
522         if (info->alloc_start != 0)
523                 seq_printf(seq, ",alloc_start=%llu",
524                            (unsigned long long)info->alloc_start);
525         if (info->thread_pool_size !=  min_t(unsigned long,
526                                              num_online_cpus() + 2, 8))
527                 seq_printf(seq, ",thread_pool=%d", info->thread_pool_size);
528         if (btrfs_test_opt(root, COMPRESS))
529                 seq_puts(seq, ",compress");
530         if (btrfs_test_opt(root, NOSSD))
531                 seq_puts(seq, ",nossd");
532         if (btrfs_test_opt(root, SSD_SPREAD))
533                 seq_puts(seq, ",ssd_spread");
534         else if (btrfs_test_opt(root, SSD))
535                 seq_puts(seq, ",ssd");
536         if (btrfs_test_opt(root, NOTREELOG))
537                 seq_puts(seq, ",notreelog");
538         if (btrfs_test_opt(root, FLUSHONCOMMIT))
539                 seq_puts(seq, ",flushoncommit");
540         if (btrfs_test_opt(root, DISCARD))
541                 seq_puts(seq, ",discard");
542         if (!(root->fs_info->sb->s_flags & MS_POSIXACL))
543                 seq_puts(seq, ",noacl");
544         return 0;
545 }
546
547 static int btrfs_test_super(struct super_block *s, void *data)
548 {
549         struct btrfs_fs_devices *test_fs_devices = data;
550         struct btrfs_root *root = btrfs_sb(s);
551
552         return root->fs_info->fs_devices == test_fs_devices;
553 }
554
555 /*
556  * Find a superblock for the given device / mount point.
557  *
558  * Note:  This is based on get_sb_bdev from fs/super.c with a few additions
559  *        for multiple device setup.  Make sure to keep it in sync.
560  */
561 static int btrfs_get_sb(struct file_system_type *fs_type, int flags,
562                 const char *dev_name, void *data, struct vfsmount *mnt)
563 {
564         struct block_device *bdev = NULL;
565         struct super_block *s;
566         struct dentry *root;
567         struct btrfs_fs_devices *fs_devices = NULL;
568         fmode_t mode = FMODE_READ;
569         char *subvol_name = NULL;
570         u64 subvol_objectid = 0;
571         int error = 0;
572         int found = 0;
573
574         if (!(flags & MS_RDONLY))
575                 mode |= FMODE_WRITE;
576
577         error = btrfs_parse_early_options(data, mode, fs_type,
578                                           &subvol_name, &subvol_objectid,
579                                           &fs_devices);
580         if (error)
581                 return error;
582
583         error = btrfs_scan_one_device(dev_name, mode, fs_type, &fs_devices);
584         if (error)
585                 goto error_free_subvol_name;
586
587         error = btrfs_open_devices(fs_devices, mode, fs_type);
588         if (error)
589                 goto error_free_subvol_name;
590
591         if (!(flags & MS_RDONLY) && fs_devices->rw_devices == 0) {
592                 error = -EACCES;
593                 goto error_close_devices;
594         }
595
596         bdev = fs_devices->latest_bdev;
597         s = sget(fs_type, btrfs_test_super, set_anon_super, fs_devices);
598         if (IS_ERR(s))
599                 goto error_s;
600
601         if (s->s_root) {
602                 if ((flags ^ s->s_flags) & MS_RDONLY) {
603                         deactivate_locked_super(s);
604                         error = -EBUSY;
605                         goto error_close_devices;
606                 }
607
608                 found = 1;
609                 btrfs_close_devices(fs_devices);
610         } else {
611                 char b[BDEVNAME_SIZE];
612
613                 s->s_flags = flags;
614                 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
615                 error = btrfs_fill_super(s, fs_devices, data,
616                                          flags & MS_SILENT ? 1 : 0);
617                 if (error) {
618                         deactivate_locked_super(s);
619                         goto error_free_subvol_name;
620                 }
621
622                 btrfs_sb(s)->fs_info->bdev_holder = fs_type;
623                 s->s_flags |= MS_ACTIVE;
624         }
625
626         root = get_default_root(s, subvol_objectid);
627         if (IS_ERR(root)) {
628                 error = PTR_ERR(root);
629                 deactivate_locked_super(s);
630                 goto error;
631         }
632         /* if they gave us a subvolume name bind mount into that */
633         if (strcmp(subvol_name, ".")) {
634                 struct dentry *new_root;
635                 mutex_lock(&root->d_inode->i_mutex);
636                 new_root = lookup_one_len(subvol_name, root,
637                                       strlen(subvol_name));
638                 mutex_unlock(&root->d_inode->i_mutex);
639
640                 if (IS_ERR(new_root)) {
641                         deactivate_locked_super(s);
642                         error = PTR_ERR(new_root);
643                         dput(root);
644                         goto error_close_devices;
645                 }
646                 if (!new_root->d_inode) {
647                         dput(root);
648                         dput(new_root);
649                         deactivate_locked_super(s);
650                         error = -ENXIO;
651                         goto error_close_devices;
652                 }
653                 dput(root);
654                 root = new_root;
655         }
656
657         mnt->mnt_sb = s;
658         mnt->mnt_root = root;
659
660         kfree(subvol_name);
661         return 0;
662
663 error_s:
664         error = PTR_ERR(s);
665 error_close_devices:
666         btrfs_close_devices(fs_devices);
667 error_free_subvol_name:
668         kfree(subvol_name);
669 error:
670         return error;
671 }
672
673 static int btrfs_remount(struct super_block *sb, int *flags, char *data)
674 {
675         struct btrfs_root *root = btrfs_sb(sb);
676         int ret;
677
678         ret = btrfs_parse_options(root, data);
679         if (ret)
680                 return -EINVAL;
681
682         if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
683                 return 0;
684
685         if (*flags & MS_RDONLY) {
686                 sb->s_flags |= MS_RDONLY;
687
688                 ret =  btrfs_commit_super(root);
689                 WARN_ON(ret);
690         } else {
691                 if (root->fs_info->fs_devices->rw_devices == 0)
692                         return -EACCES;
693
694                 if (btrfs_super_log_root(&root->fs_info->super_copy) != 0)
695                         return -EINVAL;
696
697                 /* recover relocation */
698                 ret = btrfs_recover_relocation(root);
699                 WARN_ON(ret);
700
701                 ret = btrfs_cleanup_fs_roots(root->fs_info);
702                 WARN_ON(ret);
703
704                 sb->s_flags &= ~MS_RDONLY;
705         }
706
707         return 0;
708 }
709
710 static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
711 {
712         struct btrfs_root *root = btrfs_sb(dentry->d_sb);
713         struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
714         struct list_head *head = &root->fs_info->space_info;
715         struct btrfs_space_info *found;
716         u64 total_used = 0;
717         u64 data_used = 0;
718         int bits = dentry->d_sb->s_blocksize_bits;
719         __be32 *fsid = (__be32 *)root->fs_info->fsid;
720
721         rcu_read_lock();
722         list_for_each_entry_rcu(found, head, list) {
723                 if (found->flags & (BTRFS_BLOCK_GROUP_DUP|
724                                     BTRFS_BLOCK_GROUP_RAID10|
725                                     BTRFS_BLOCK_GROUP_RAID1)) {
726                         total_used += found->bytes_used;
727                         if (found->flags & BTRFS_BLOCK_GROUP_DATA)
728                                 data_used += found->bytes_used;
729                         else
730                                 data_used += found->total_bytes;
731                 }
732
733                 total_used += found->bytes_used;
734                 if (found->flags & BTRFS_BLOCK_GROUP_DATA)
735                         data_used += found->bytes_used;
736                 else
737                         data_used += found->total_bytes;
738         }
739         rcu_read_unlock();
740
741         buf->f_namelen = BTRFS_NAME_LEN;
742         buf->f_blocks = btrfs_super_total_bytes(disk_super) >> bits;
743         buf->f_bfree = buf->f_blocks - (total_used >> bits);
744         buf->f_bavail = buf->f_blocks - (data_used >> bits);
745         buf->f_bsize = dentry->d_sb->s_blocksize;
746         buf->f_type = BTRFS_SUPER_MAGIC;
747
748         /* We treat it as constant endianness (it doesn't matter _which_)
749            because we want the fsid to come out the same whether mounted
750            on a big-endian or little-endian host */
751         buf->f_fsid.val[0] = be32_to_cpu(fsid[0]) ^ be32_to_cpu(fsid[2]);
752         buf->f_fsid.val[1] = be32_to_cpu(fsid[1]) ^ be32_to_cpu(fsid[3]);
753         /* Mask in the root object ID too, to disambiguate subvols */
754         buf->f_fsid.val[0] ^= BTRFS_I(dentry->d_inode)->root->objectid >> 32;
755         buf->f_fsid.val[1] ^= BTRFS_I(dentry->d_inode)->root->objectid;
756
757         return 0;
758 }
759
760 static struct file_system_type btrfs_fs_type = {
761         .owner          = THIS_MODULE,
762         .name           = "btrfs",
763         .get_sb         = btrfs_get_sb,
764         .kill_sb        = kill_anon_super,
765         .fs_flags       = FS_REQUIRES_DEV,
766 };
767
768 /*
769  * used by btrfsctl to scan devices when no FS is mounted
770  */
771 static long btrfs_control_ioctl(struct file *file, unsigned int cmd,
772                                 unsigned long arg)
773 {
774         struct btrfs_ioctl_vol_args *vol;
775         struct btrfs_fs_devices *fs_devices;
776         int ret = -ENOTTY;
777
778         if (!capable(CAP_SYS_ADMIN))
779                 return -EPERM;
780
781         vol = memdup_user((void __user *)arg, sizeof(*vol));
782         if (IS_ERR(vol))
783                 return PTR_ERR(vol);
784
785         switch (cmd) {
786         case BTRFS_IOC_SCAN_DEV:
787                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
788                                             &btrfs_fs_type, &fs_devices);
789                 break;
790         }
791
792         kfree(vol);
793         return ret;
794 }
795
796 static int btrfs_freeze(struct super_block *sb)
797 {
798         struct btrfs_root *root = btrfs_sb(sb);
799         mutex_lock(&root->fs_info->transaction_kthread_mutex);
800         mutex_lock(&root->fs_info->cleaner_mutex);
801         return 0;
802 }
803
804 static int btrfs_unfreeze(struct super_block *sb)
805 {
806         struct btrfs_root *root = btrfs_sb(sb);
807         mutex_unlock(&root->fs_info->cleaner_mutex);
808         mutex_unlock(&root->fs_info->transaction_kthread_mutex);
809         return 0;
810 }
811
812 static const struct super_operations btrfs_super_ops = {
813         .drop_inode     = btrfs_drop_inode,
814         .delete_inode   = btrfs_delete_inode,
815         .put_super      = btrfs_put_super,
816         .sync_fs        = btrfs_sync_fs,
817         .show_options   = btrfs_show_options,
818         .write_inode    = btrfs_write_inode,
819         .dirty_inode    = btrfs_dirty_inode,
820         .alloc_inode    = btrfs_alloc_inode,
821         .destroy_inode  = btrfs_destroy_inode,
822         .statfs         = btrfs_statfs,
823         .remount_fs     = btrfs_remount,
824         .freeze_fs      = btrfs_freeze,
825         .unfreeze_fs    = btrfs_unfreeze,
826 };
827
828 static const struct file_operations btrfs_ctl_fops = {
829         .unlocked_ioctl  = btrfs_control_ioctl,
830         .compat_ioctl = btrfs_control_ioctl,
831         .owner   = THIS_MODULE,
832 };
833
834 static struct miscdevice btrfs_misc = {
835         .minor          = MISC_DYNAMIC_MINOR,
836         .name           = "btrfs-control",
837         .fops           = &btrfs_ctl_fops
838 };
839
840 static int btrfs_interface_init(void)
841 {
842         return misc_register(&btrfs_misc);
843 }
844
845 static void btrfs_interface_exit(void)
846 {
847         if (misc_deregister(&btrfs_misc) < 0)
848                 printk(KERN_INFO "misc_deregister failed for control device");
849 }
850
851 static int __init init_btrfs_fs(void)
852 {
853         int err;
854
855         err = btrfs_init_sysfs();
856         if (err)
857                 return err;
858
859         err = btrfs_init_cachep();
860         if (err)
861                 goto free_sysfs;
862
863         err = extent_io_init();
864         if (err)
865                 goto free_cachep;
866
867         err = extent_map_init();
868         if (err)
869                 goto free_extent_io;
870
871         err = btrfs_interface_init();
872         if (err)
873                 goto free_extent_map;
874
875         err = register_filesystem(&btrfs_fs_type);
876         if (err)
877                 goto unregister_ioctl;
878
879         printk(KERN_INFO "%s loaded\n", BTRFS_BUILD_VERSION);
880         return 0;
881
882 unregister_ioctl:
883         btrfs_interface_exit();
884 free_extent_map:
885         extent_map_exit();
886 free_extent_io:
887         extent_io_exit();
888 free_cachep:
889         btrfs_destroy_cachep();
890 free_sysfs:
891         btrfs_exit_sysfs();
892         return err;
893 }
894
895 static void __exit exit_btrfs_fs(void)
896 {
897         btrfs_destroy_cachep();
898         extent_map_exit();
899         extent_io_exit();
900         btrfs_interface_exit();
901         unregister_filesystem(&btrfs_fs_type);
902         btrfs_exit_sysfs();
903         btrfs_cleanup_fs_uuids();
904         btrfs_zlib_exit();
905 }
906
907 module_init(init_btrfs_fs)
908 module_exit(exit_btrfs_fs)
909
910 MODULE_LICENSE("GPL");