fs-writeback.c: unify some common code
[~shefty/rdma-dev.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 (IS_ERR(di))
364                 return ERR_CAST(di);
365         if (!di) {
366                 /*
367                  * Ok the default dir item isn't there.  This is weird since
368                  * it's always been there, but don't freak out, just try and
369                  * mount to root most subvolume.
370                  */
371                 btrfs_free_path(path);
372                 dir_id = BTRFS_FIRST_FREE_OBJECTID;
373                 new_root = root->fs_info->fs_root;
374                 goto setup_root;
375         }
376
377         btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
378         btrfs_free_path(path);
379
380 find_root:
381         new_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
382         if (IS_ERR(new_root))
383                 return ERR_PTR(PTR_ERR(new_root));
384
385         if (btrfs_root_refs(&new_root->root_item) == 0)
386                 return ERR_PTR(-ENOENT);
387
388         dir_id = btrfs_root_dirid(&new_root->root_item);
389 setup_root:
390         location.objectid = dir_id;
391         location.type = BTRFS_INODE_ITEM_KEY;
392         location.offset = 0;
393
394         inode = btrfs_iget(sb, &location, new_root, &new);
395         if (IS_ERR(inode))
396                 return ERR_CAST(inode);
397
398         /*
399          * If we're just mounting the root most subvol put the inode and return
400          * a reference to the dentry.  We will have already gotten a reference
401          * to the inode in btrfs_fill_super so we're good to go.
402          */
403         if (!new && sb->s_root->d_inode == inode) {
404                 iput(inode);
405                 return dget(sb->s_root);
406         }
407
408         if (new) {
409                 const struct qstr name = { .name = "/", .len = 1 };
410
411                 /*
412                  * New inode, we need to make the dentry a sibling of s_root so
413                  * everything gets cleaned up properly on unmount.
414                  */
415                 dentry = d_alloc(sb->s_root, &name);
416                 if (!dentry) {
417                         iput(inode);
418                         return ERR_PTR(-ENOMEM);
419                 }
420                 d_splice_alias(inode, dentry);
421         } else {
422                 /*
423                  * We found the inode in cache, just find a dentry for it and
424                  * put the reference to the inode we just got.
425                  */
426                 dentry = d_find_alias(inode);
427                 iput(inode);
428         }
429
430         return dentry;
431 }
432
433 static int btrfs_fill_super(struct super_block *sb,
434                             struct btrfs_fs_devices *fs_devices,
435                             void *data, int silent)
436 {
437         struct inode *inode;
438         struct dentry *root_dentry;
439         struct btrfs_super_block *disk_super;
440         struct btrfs_root *tree_root;
441         struct btrfs_key key;
442         int err;
443
444         sb->s_maxbytes = MAX_LFS_FILESIZE;
445         sb->s_magic = BTRFS_SUPER_MAGIC;
446         sb->s_op = &btrfs_super_ops;
447         sb->s_export_op = &btrfs_export_ops;
448         sb->s_xattr = btrfs_xattr_handlers;
449         sb->s_time_gran = 1;
450 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
451         sb->s_flags |= MS_POSIXACL;
452 #endif
453
454         tree_root = open_ctree(sb, fs_devices, (char *)data);
455
456         if (IS_ERR(tree_root)) {
457                 printk("btrfs: open_ctree failed\n");
458                 return PTR_ERR(tree_root);
459         }
460         sb->s_fs_info = tree_root;
461         disk_super = &tree_root->fs_info->super_copy;
462
463         key.objectid = BTRFS_FIRST_FREE_OBJECTID;
464         key.type = BTRFS_INODE_ITEM_KEY;
465         key.offset = 0;
466         inode = btrfs_iget(sb, &key, tree_root->fs_info->fs_root, NULL);
467         if (IS_ERR(inode)) {
468                 err = PTR_ERR(inode);
469                 goto fail_close;
470         }
471
472         root_dentry = d_alloc_root(inode);
473         if (!root_dentry) {
474                 iput(inode);
475                 err = -ENOMEM;
476                 goto fail_close;
477         }
478
479         sb->s_root = root_dentry;
480
481         save_mount_options(sb, data);
482         return 0;
483
484 fail_close:
485         close_ctree(tree_root);
486         return err;
487 }
488
489 int btrfs_sync_fs(struct super_block *sb, int wait)
490 {
491         struct btrfs_trans_handle *trans;
492         struct btrfs_root *root = btrfs_sb(sb);
493         int ret;
494
495         if (!wait) {
496                 filemap_flush(root->fs_info->btree_inode->i_mapping);
497                 return 0;
498         }
499
500         btrfs_start_delalloc_inodes(root, 0);
501         btrfs_wait_ordered_extents(root, 0, 0);
502
503         trans = btrfs_start_transaction(root, 0);
504         ret = btrfs_commit_transaction(trans, root);
505         return ret;
506 }
507
508 static int btrfs_show_options(struct seq_file *seq, struct vfsmount *vfs)
509 {
510         struct btrfs_root *root = btrfs_sb(vfs->mnt_sb);
511         struct btrfs_fs_info *info = root->fs_info;
512
513         if (btrfs_test_opt(root, DEGRADED))
514                 seq_puts(seq, ",degraded");
515         if (btrfs_test_opt(root, NODATASUM))
516                 seq_puts(seq, ",nodatasum");
517         if (btrfs_test_opt(root, NODATACOW))
518                 seq_puts(seq, ",nodatacow");
519         if (btrfs_test_opt(root, NOBARRIER))
520                 seq_puts(seq, ",nobarrier");
521         if (info->max_inline != 8192 * 1024)
522                 seq_printf(seq, ",max_inline=%llu",
523                            (unsigned long long)info->max_inline);
524         if (info->alloc_start != 0)
525                 seq_printf(seq, ",alloc_start=%llu",
526                            (unsigned long long)info->alloc_start);
527         if (info->thread_pool_size !=  min_t(unsigned long,
528                                              num_online_cpus() + 2, 8))
529                 seq_printf(seq, ",thread_pool=%d", info->thread_pool_size);
530         if (btrfs_test_opt(root, COMPRESS))
531                 seq_puts(seq, ",compress");
532         if (btrfs_test_opt(root, NOSSD))
533                 seq_puts(seq, ",nossd");
534         if (btrfs_test_opt(root, SSD_SPREAD))
535                 seq_puts(seq, ",ssd_spread");
536         else if (btrfs_test_opt(root, SSD))
537                 seq_puts(seq, ",ssd");
538         if (btrfs_test_opt(root, NOTREELOG))
539                 seq_puts(seq, ",notreelog");
540         if (btrfs_test_opt(root, FLUSHONCOMMIT))
541                 seq_puts(seq, ",flushoncommit");
542         if (btrfs_test_opt(root, DISCARD))
543                 seq_puts(seq, ",discard");
544         if (!(root->fs_info->sb->s_flags & MS_POSIXACL))
545                 seq_puts(seq, ",noacl");
546         return 0;
547 }
548
549 static int btrfs_test_super(struct super_block *s, void *data)
550 {
551         struct btrfs_fs_devices *test_fs_devices = data;
552         struct btrfs_root *root = btrfs_sb(s);
553
554         return root->fs_info->fs_devices == test_fs_devices;
555 }
556
557 /*
558  * Find a superblock for the given device / mount point.
559  *
560  * Note:  This is based on get_sb_bdev from fs/super.c with a few additions
561  *        for multiple device setup.  Make sure to keep it in sync.
562  */
563 static struct dentry *btrfs_mount(struct file_system_type *fs_type, int flags,
564                 const char *dev_name, void *data)
565 {
566         struct block_device *bdev = NULL;
567         struct super_block *s;
568         struct dentry *root;
569         struct btrfs_fs_devices *fs_devices = NULL;
570         fmode_t mode = FMODE_READ;
571         char *subvol_name = NULL;
572         u64 subvol_objectid = 0;
573         int error = 0;
574         int found = 0;
575
576         if (!(flags & MS_RDONLY))
577                 mode |= FMODE_WRITE;
578
579         error = btrfs_parse_early_options(data, mode, fs_type,
580                                           &subvol_name, &subvol_objectid,
581                                           &fs_devices);
582         if (error)
583                 return ERR_PTR(error);
584
585         error = btrfs_scan_one_device(dev_name, mode, fs_type, &fs_devices);
586         if (error)
587                 goto error_free_subvol_name;
588
589         error = btrfs_open_devices(fs_devices, mode, fs_type);
590         if (error)
591                 goto error_free_subvol_name;
592
593         if (!(flags & MS_RDONLY) && fs_devices->rw_devices == 0) {
594                 error = -EACCES;
595                 goto error_close_devices;
596         }
597
598         bdev = fs_devices->latest_bdev;
599         s = sget(fs_type, btrfs_test_super, set_anon_super, fs_devices);
600         if (IS_ERR(s))
601                 goto error_s;
602
603         if (s->s_root) {
604                 if ((flags ^ s->s_flags) & MS_RDONLY) {
605                         deactivate_locked_super(s);
606                         error = -EBUSY;
607                         goto error_close_devices;
608                 }
609
610                 found = 1;
611                 btrfs_close_devices(fs_devices);
612         } else {
613                 char b[BDEVNAME_SIZE];
614
615                 s->s_flags = flags;
616                 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
617                 error = btrfs_fill_super(s, fs_devices, data,
618                                          flags & MS_SILENT ? 1 : 0);
619                 if (error) {
620                         deactivate_locked_super(s);
621                         goto error_free_subvol_name;
622                 }
623
624                 btrfs_sb(s)->fs_info->bdev_holder = fs_type;
625                 s->s_flags |= MS_ACTIVE;
626         }
627
628         root = get_default_root(s, subvol_objectid);
629         if (IS_ERR(root)) {
630                 error = PTR_ERR(root);
631                 deactivate_locked_super(s);
632                 goto error;
633         }
634         /* if they gave us a subvolume name bind mount into that */
635         if (strcmp(subvol_name, ".")) {
636                 struct dentry *new_root;
637                 mutex_lock(&root->d_inode->i_mutex);
638                 new_root = lookup_one_len(subvol_name, root,
639                                       strlen(subvol_name));
640                 mutex_unlock(&root->d_inode->i_mutex);
641
642                 if (IS_ERR(new_root)) {
643                         deactivate_locked_super(s);
644                         error = PTR_ERR(new_root);
645                         dput(root);
646                         goto error_close_devices;
647                 }
648                 if (!new_root->d_inode) {
649                         dput(root);
650                         dput(new_root);
651                         deactivate_locked_super(s);
652                         error = -ENXIO;
653                         goto error_close_devices;
654                 }
655                 dput(root);
656                 root = new_root;
657         }
658
659         kfree(subvol_name);
660         return root;
661
662 error_s:
663         error = PTR_ERR(s);
664 error_close_devices:
665         btrfs_close_devices(fs_devices);
666 error_free_subvol_name:
667         kfree(subvol_name);
668 error:
669         return ERR_PTR(error);
670 }
671
672 static int btrfs_remount(struct super_block *sb, int *flags, char *data)
673 {
674         struct btrfs_root *root = btrfs_sb(sb);
675         int ret;
676
677         ret = btrfs_parse_options(root, data);
678         if (ret)
679                 return -EINVAL;
680
681         if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
682                 return 0;
683
684         if (*flags & MS_RDONLY) {
685                 sb->s_flags |= MS_RDONLY;
686
687                 ret =  btrfs_commit_super(root);
688                 WARN_ON(ret);
689         } else {
690                 if (root->fs_info->fs_devices->rw_devices == 0)
691                         return -EACCES;
692
693                 if (btrfs_super_log_root(&root->fs_info->super_copy) != 0)
694                         return -EINVAL;
695
696                 ret = btrfs_cleanup_fs_roots(root->fs_info);
697                 WARN_ON(ret);
698
699                 /* recover relocation */
700                 ret = btrfs_recover_relocation(root);
701                 WARN_ON(ret);
702
703                 sb->s_flags &= ~MS_RDONLY;
704         }
705
706         return 0;
707 }
708
709 static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
710 {
711         struct btrfs_root *root = btrfs_sb(dentry->d_sb);
712         struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
713         struct list_head *head = &root->fs_info->space_info;
714         struct btrfs_space_info *found;
715         u64 total_used = 0;
716         int bits = dentry->d_sb->s_blocksize_bits;
717         __be32 *fsid = (__be32 *)root->fs_info->fsid;
718
719         rcu_read_lock();
720         list_for_each_entry_rcu(found, head, list)
721                 total_used += found->disk_used;
722         rcu_read_unlock();
723
724         buf->f_namelen = BTRFS_NAME_LEN;
725         buf->f_blocks = btrfs_super_total_bytes(disk_super) >> bits;
726         buf->f_bfree = buf->f_blocks - (total_used >> bits);
727         buf->f_bavail = buf->f_bfree;
728         buf->f_bsize = dentry->d_sb->s_blocksize;
729         buf->f_type = BTRFS_SUPER_MAGIC;
730
731         /* We treat it as constant endianness (it doesn't matter _which_)
732            because we want the fsid to come out the same whether mounted
733            on a big-endian or little-endian host */
734         buf->f_fsid.val[0] = be32_to_cpu(fsid[0]) ^ be32_to_cpu(fsid[2]);
735         buf->f_fsid.val[1] = be32_to_cpu(fsid[1]) ^ be32_to_cpu(fsid[3]);
736         /* Mask in the root object ID too, to disambiguate subvols */
737         buf->f_fsid.val[0] ^= BTRFS_I(dentry->d_inode)->root->objectid >> 32;
738         buf->f_fsid.val[1] ^= BTRFS_I(dentry->d_inode)->root->objectid;
739
740         return 0;
741 }
742
743 static struct file_system_type btrfs_fs_type = {
744         .owner          = THIS_MODULE,
745         .name           = "btrfs",
746         .mount          = btrfs_mount,
747         .kill_sb        = kill_anon_super,
748         .fs_flags       = FS_REQUIRES_DEV,
749 };
750
751 /*
752  * used by btrfsctl to scan devices when no FS is mounted
753  */
754 static long btrfs_control_ioctl(struct file *file, unsigned int cmd,
755                                 unsigned long arg)
756 {
757         struct btrfs_ioctl_vol_args *vol;
758         struct btrfs_fs_devices *fs_devices;
759         int ret = -ENOTTY;
760
761         if (!capable(CAP_SYS_ADMIN))
762                 return -EPERM;
763
764         vol = memdup_user((void __user *)arg, sizeof(*vol));
765         if (IS_ERR(vol))
766                 return PTR_ERR(vol);
767
768         switch (cmd) {
769         case BTRFS_IOC_SCAN_DEV:
770                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
771                                             &btrfs_fs_type, &fs_devices);
772                 break;
773         }
774
775         kfree(vol);
776         return ret;
777 }
778
779 static int btrfs_freeze(struct super_block *sb)
780 {
781         struct btrfs_root *root = btrfs_sb(sb);
782         mutex_lock(&root->fs_info->transaction_kthread_mutex);
783         mutex_lock(&root->fs_info->cleaner_mutex);
784         return 0;
785 }
786
787 static int btrfs_unfreeze(struct super_block *sb)
788 {
789         struct btrfs_root *root = btrfs_sb(sb);
790         mutex_unlock(&root->fs_info->cleaner_mutex);
791         mutex_unlock(&root->fs_info->transaction_kthread_mutex);
792         return 0;
793 }
794
795 static const struct super_operations btrfs_super_ops = {
796         .drop_inode     = btrfs_drop_inode,
797         .evict_inode    = btrfs_evict_inode,
798         .put_super      = btrfs_put_super,
799         .sync_fs        = btrfs_sync_fs,
800         .show_options   = btrfs_show_options,
801         .write_inode    = btrfs_write_inode,
802         .dirty_inode    = btrfs_dirty_inode,
803         .alloc_inode    = btrfs_alloc_inode,
804         .destroy_inode  = btrfs_destroy_inode,
805         .statfs         = btrfs_statfs,
806         .remount_fs     = btrfs_remount,
807         .freeze_fs      = btrfs_freeze,
808         .unfreeze_fs    = btrfs_unfreeze,
809 };
810
811 static const struct file_operations btrfs_ctl_fops = {
812         .unlocked_ioctl  = btrfs_control_ioctl,
813         .compat_ioctl = btrfs_control_ioctl,
814         .owner   = THIS_MODULE,
815         .llseek = noop_llseek,
816 };
817
818 static struct miscdevice btrfs_misc = {
819         .minor          = BTRFS_MINOR,
820         .name           = "btrfs-control",
821         .fops           = &btrfs_ctl_fops
822 };
823
824 MODULE_ALIAS_MISCDEV(BTRFS_MINOR);
825 MODULE_ALIAS("devname:btrfs-control");
826
827 static int btrfs_interface_init(void)
828 {
829         return misc_register(&btrfs_misc);
830 }
831
832 static void btrfs_interface_exit(void)
833 {
834         if (misc_deregister(&btrfs_misc) < 0)
835                 printk(KERN_INFO "misc_deregister failed for control device");
836 }
837
838 static int __init init_btrfs_fs(void)
839 {
840         int err;
841
842         err = btrfs_init_sysfs();
843         if (err)
844                 return err;
845
846         err = btrfs_init_cachep();
847         if (err)
848                 goto free_sysfs;
849
850         err = extent_io_init();
851         if (err)
852                 goto free_cachep;
853
854         err = extent_map_init();
855         if (err)
856                 goto free_extent_io;
857
858         err = btrfs_interface_init();
859         if (err)
860                 goto free_extent_map;
861
862         err = register_filesystem(&btrfs_fs_type);
863         if (err)
864                 goto unregister_ioctl;
865
866         printk(KERN_INFO "%s loaded\n", BTRFS_BUILD_VERSION);
867         return 0;
868
869 unregister_ioctl:
870         btrfs_interface_exit();
871 free_extent_map:
872         extent_map_exit();
873 free_extent_io:
874         extent_io_exit();
875 free_cachep:
876         btrfs_destroy_cachep();
877 free_sysfs:
878         btrfs_exit_sysfs();
879         return err;
880 }
881
882 static void __exit exit_btrfs_fs(void)
883 {
884         btrfs_destroy_cachep();
885         extent_map_exit();
886         extent_io_exit();
887         btrfs_interface_exit();
888         unregister_filesystem(&btrfs_fs_type);
889         btrfs_exit_sysfs();
890         btrfs_cleanup_fs_uuids();
891         btrfs_zlib_exit();
892 }
893
894 module_init(init_btrfs_fs)
895 module_exit(exit_btrfs_fs)
896
897 MODULE_LICENSE("GPL");