Merge git://git.kernel.org/pub/scm/linux/kernel/git/mason/btrfs-unstable
[~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 const char *btrfs_decode_error(struct btrfs_fs_info *fs_info, int errno,
58                                       char nbuf[16])
59 {
60         char *errstr = NULL;
61
62         switch (errno) {
63         case -EIO:
64                 errstr = "IO failure";
65                 break;
66         case -ENOMEM:
67                 errstr = "Out of memory";
68                 break;
69         case -EROFS:
70                 errstr = "Readonly filesystem";
71                 break;
72         default:
73                 if (nbuf) {
74                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
75                                 errstr = nbuf;
76                 }
77                 break;
78         }
79
80         return errstr;
81 }
82
83 static void __save_error_info(struct btrfs_fs_info *fs_info)
84 {
85         /*
86          * today we only save the error info into ram.  Long term we'll
87          * also send it down to the disk
88          */
89         fs_info->fs_state = BTRFS_SUPER_FLAG_ERROR;
90 }
91
92 /* NOTE:
93  *      We move write_super stuff at umount in order to avoid deadlock
94  *      for umount hold all lock.
95  */
96 static void save_error_info(struct btrfs_fs_info *fs_info)
97 {
98         __save_error_info(fs_info);
99 }
100
101 /* btrfs handle error by forcing the filesystem readonly */
102 static void btrfs_handle_error(struct btrfs_fs_info *fs_info)
103 {
104         struct super_block *sb = fs_info->sb;
105
106         if (sb->s_flags & MS_RDONLY)
107                 return;
108
109         if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
110                 sb->s_flags |= MS_RDONLY;
111                 printk(KERN_INFO "btrfs is forced readonly\n");
112         }
113 }
114
115 /*
116  * __btrfs_std_error decodes expected errors from the caller and
117  * invokes the approciate error response.
118  */
119 void __btrfs_std_error(struct btrfs_fs_info *fs_info, const char *function,
120                      unsigned int line, int errno)
121 {
122         struct super_block *sb = fs_info->sb;
123         char nbuf[16];
124         const char *errstr;
125
126         /*
127          * Special case: if the error is EROFS, and we're already
128          * under MS_RDONLY, then it is safe here.
129          */
130         if (errno == -EROFS && (sb->s_flags & MS_RDONLY))
131                 return;
132
133         errstr = btrfs_decode_error(fs_info, errno, nbuf);
134         printk(KERN_CRIT "BTRFS error (device %s) in %s:%d: %s\n",
135                 sb->s_id, function, line, errstr);
136         save_error_info(fs_info);
137
138         btrfs_handle_error(fs_info);
139 }
140
141 static void btrfs_put_super(struct super_block *sb)
142 {
143         struct btrfs_root *root = btrfs_sb(sb);
144         int ret;
145
146         ret = close_ctree(root);
147         sb->s_fs_info = NULL;
148
149         (void)ret; /* FIXME: need to fix VFS to return error? */
150 }
151
152 enum {
153         Opt_degraded, Opt_subvol, Opt_subvolid, Opt_device, Opt_nodatasum,
154         Opt_nodatacow, Opt_max_inline, Opt_alloc_start, Opt_nobarrier, Opt_ssd,
155         Opt_nossd, Opt_ssd_spread, Opt_thread_pool, Opt_noacl, Opt_compress,
156         Opt_compress_type, Opt_compress_force, Opt_compress_force_type,
157         Opt_notreelog, Opt_ratio, Opt_flushoncommit, Opt_discard,
158         Opt_space_cache, Opt_clear_cache, Opt_user_subvol_rm_allowed, Opt_err,
159 };
160
161 static match_table_t tokens = {
162         {Opt_degraded, "degraded"},
163         {Opt_subvol, "subvol=%s"},
164         {Opt_subvolid, "subvolid=%d"},
165         {Opt_device, "device=%s"},
166         {Opt_nodatasum, "nodatasum"},
167         {Opt_nodatacow, "nodatacow"},
168         {Opt_nobarrier, "nobarrier"},
169         {Opt_max_inline, "max_inline=%s"},
170         {Opt_alloc_start, "alloc_start=%s"},
171         {Opt_thread_pool, "thread_pool=%d"},
172         {Opt_compress, "compress"},
173         {Opt_compress_type, "compress=%s"},
174         {Opt_compress_force, "compress-force"},
175         {Opt_compress_force_type, "compress-force=%s"},
176         {Opt_ssd, "ssd"},
177         {Opt_ssd_spread, "ssd_spread"},
178         {Opt_nossd, "nossd"},
179         {Opt_noacl, "noacl"},
180         {Opt_notreelog, "notreelog"},
181         {Opt_flushoncommit, "flushoncommit"},
182         {Opt_ratio, "metadata_ratio=%d"},
183         {Opt_discard, "discard"},
184         {Opt_space_cache, "space_cache"},
185         {Opt_clear_cache, "clear_cache"},
186         {Opt_user_subvol_rm_allowed, "user_subvol_rm_allowed"},
187         {Opt_err, NULL},
188 };
189
190 /*
191  * Regular mount options parser.  Everything that is needed only when
192  * reading in a new superblock is parsed here.
193  */
194 int btrfs_parse_options(struct btrfs_root *root, char *options)
195 {
196         struct btrfs_fs_info *info = root->fs_info;
197         substring_t args[MAX_OPT_ARGS];
198         char *p, *num, *orig;
199         int intarg;
200         int ret = 0;
201         char *compress_type;
202         bool compress_force = false;
203
204         if (!options)
205                 return 0;
206
207         /*
208          * strsep changes the string, duplicate it because parse_options
209          * gets called twice
210          */
211         options = kstrdup(options, GFP_NOFS);
212         if (!options)
213                 return -ENOMEM;
214
215         orig = options;
216
217         while ((p = strsep(&options, ",")) != NULL) {
218                 int token;
219                 if (!*p)
220                         continue;
221
222                 token = match_token(p, tokens, args);
223                 switch (token) {
224                 case Opt_degraded:
225                         printk(KERN_INFO "btrfs: allowing degraded mounts\n");
226                         btrfs_set_opt(info->mount_opt, DEGRADED);
227                         break;
228                 case Opt_subvol:
229                 case Opt_subvolid:
230                 case Opt_device:
231                         /*
232                          * These are parsed by btrfs_parse_early_options
233                          * and can be happily ignored here.
234                          */
235                         break;
236                 case Opt_nodatasum:
237                         printk(KERN_INFO "btrfs: setting nodatasum\n");
238                         btrfs_set_opt(info->mount_opt, NODATASUM);
239                         break;
240                 case Opt_nodatacow:
241                         printk(KERN_INFO "btrfs: setting nodatacow\n");
242                         btrfs_set_opt(info->mount_opt, NODATACOW);
243                         btrfs_set_opt(info->mount_opt, NODATASUM);
244                         break;
245                 case Opt_compress_force:
246                 case Opt_compress_force_type:
247                         compress_force = true;
248                 case Opt_compress:
249                 case Opt_compress_type:
250                         if (token == Opt_compress ||
251                             token == Opt_compress_force ||
252                             strcmp(args[0].from, "zlib") == 0) {
253                                 compress_type = "zlib";
254                                 info->compress_type = BTRFS_COMPRESS_ZLIB;
255                         } else if (strcmp(args[0].from, "lzo") == 0) {
256                                 compress_type = "lzo";
257                                 info->compress_type = BTRFS_COMPRESS_LZO;
258                         } else {
259                                 ret = -EINVAL;
260                                 goto out;
261                         }
262
263                         btrfs_set_opt(info->mount_opt, COMPRESS);
264                         if (compress_force) {
265                                 btrfs_set_opt(info->mount_opt, FORCE_COMPRESS);
266                                 pr_info("btrfs: force %s compression\n",
267                                         compress_type);
268                         } else
269                                 pr_info("btrfs: use %s compression\n",
270                                         compress_type);
271                         break;
272                 case Opt_ssd:
273                         printk(KERN_INFO "btrfs: use ssd allocation scheme\n");
274                         btrfs_set_opt(info->mount_opt, SSD);
275                         break;
276                 case Opt_ssd_spread:
277                         printk(KERN_INFO "btrfs: use spread ssd "
278                                "allocation scheme\n");
279                         btrfs_set_opt(info->mount_opt, SSD);
280                         btrfs_set_opt(info->mount_opt, SSD_SPREAD);
281                         break;
282                 case Opt_nossd:
283                         printk(KERN_INFO "btrfs: not using ssd allocation "
284                                "scheme\n");
285                         btrfs_set_opt(info->mount_opt, NOSSD);
286                         btrfs_clear_opt(info->mount_opt, SSD);
287                         btrfs_clear_opt(info->mount_opt, SSD_SPREAD);
288                         break;
289                 case Opt_nobarrier:
290                         printk(KERN_INFO "btrfs: turning off barriers\n");
291                         btrfs_set_opt(info->mount_opt, NOBARRIER);
292                         break;
293                 case Opt_thread_pool:
294                         intarg = 0;
295                         match_int(&args[0], &intarg);
296                         if (intarg) {
297                                 info->thread_pool_size = intarg;
298                                 printk(KERN_INFO "btrfs: thread pool %d\n",
299                                        info->thread_pool_size);
300                         }
301                         break;
302                 case Opt_max_inline:
303                         num = match_strdup(&args[0]);
304                         if (num) {
305                                 info->max_inline = memparse(num, NULL);
306                                 kfree(num);
307
308                                 if (info->max_inline) {
309                                         info->max_inline = max_t(u64,
310                                                 info->max_inline,
311                                                 root->sectorsize);
312                                 }
313                                 printk(KERN_INFO "btrfs: max_inline at %llu\n",
314                                         (unsigned long long)info->max_inline);
315                         }
316                         break;
317                 case Opt_alloc_start:
318                         num = match_strdup(&args[0]);
319                         if (num) {
320                                 info->alloc_start = memparse(num, NULL);
321                                 kfree(num);
322                                 printk(KERN_INFO
323                                         "btrfs: allocations start at %llu\n",
324                                         (unsigned long long)info->alloc_start);
325                         }
326                         break;
327                 case Opt_noacl:
328                         root->fs_info->sb->s_flags &= ~MS_POSIXACL;
329                         break;
330                 case Opt_notreelog:
331                         printk(KERN_INFO "btrfs: disabling tree log\n");
332                         btrfs_set_opt(info->mount_opt, NOTREELOG);
333                         break;
334                 case Opt_flushoncommit:
335                         printk(KERN_INFO "btrfs: turning on flush-on-commit\n");
336                         btrfs_set_opt(info->mount_opt, FLUSHONCOMMIT);
337                         break;
338                 case Opt_ratio:
339                         intarg = 0;
340                         match_int(&args[0], &intarg);
341                         if (intarg) {
342                                 info->metadata_ratio = intarg;
343                                 printk(KERN_INFO "btrfs: metadata ratio %d\n",
344                                        info->metadata_ratio);
345                         }
346                         break;
347                 case Opt_discard:
348                         btrfs_set_opt(info->mount_opt, DISCARD);
349                         break;
350                 case Opt_space_cache:
351                         printk(KERN_INFO "btrfs: enabling disk space caching\n");
352                         btrfs_set_opt(info->mount_opt, SPACE_CACHE);
353                         break;
354                 case Opt_clear_cache:
355                         printk(KERN_INFO "btrfs: force clearing of disk cache\n");
356                         btrfs_set_opt(info->mount_opt, CLEAR_CACHE);
357                         break;
358                 case Opt_user_subvol_rm_allowed:
359                         btrfs_set_opt(info->mount_opt, USER_SUBVOL_RM_ALLOWED);
360                         break;
361                 case Opt_err:
362                         printk(KERN_INFO "btrfs: unrecognized mount option "
363                                "'%s'\n", p);
364                         ret = -EINVAL;
365                         goto out;
366                 default:
367                         break;
368                 }
369         }
370 out:
371         kfree(orig);
372         return ret;
373 }
374
375 /*
376  * Parse mount options that are required early in the mount process.
377  *
378  * All other options will be parsed on much later in the mount process and
379  * only when we need to allocate a new super block.
380  */
381 static int btrfs_parse_early_options(const char *options, fmode_t flags,
382                 void *holder, char **subvol_name, u64 *subvol_objectid,
383                 struct btrfs_fs_devices **fs_devices)
384 {
385         substring_t args[MAX_OPT_ARGS];
386         char *opts, *orig, *p;
387         int error = 0;
388         int intarg;
389
390         if (!options)
391                 goto out;
392
393         /*
394          * strsep changes the string, duplicate it because parse_options
395          * gets called twice
396          */
397         opts = kstrdup(options, GFP_KERNEL);
398         if (!opts)
399                 return -ENOMEM;
400         orig = opts;
401
402         while ((p = strsep(&opts, ",")) != NULL) {
403                 int token;
404                 if (!*p)
405                         continue;
406
407                 token = match_token(p, tokens, args);
408                 switch (token) {
409                 case Opt_subvol:
410                         *subvol_name = match_strdup(&args[0]);
411                         break;
412                 case Opt_subvolid:
413                         intarg = 0;
414                         error = match_int(&args[0], &intarg);
415                         if (!error) {
416                                 /* we want the original fs_tree */
417                                 if (!intarg)
418                                         *subvol_objectid =
419                                                 BTRFS_FS_TREE_OBJECTID;
420                                 else
421                                         *subvol_objectid = intarg;
422                         }
423                         break;
424                 case Opt_device:
425                         error = btrfs_scan_one_device(match_strdup(&args[0]),
426                                         flags, holder, fs_devices);
427                         if (error)
428                                 goto out_free_opts;
429                         break;
430                 default:
431                         break;
432                 }
433         }
434
435  out_free_opts:
436         kfree(orig);
437  out:
438         /*
439          * If no subvolume name is specified we use the default one.  Allocate
440          * a copy of the string "." here so that code later in the
441          * mount path doesn't care if it's the default volume or another one.
442          */
443         if (!*subvol_name) {
444                 *subvol_name = kstrdup(".", GFP_KERNEL);
445                 if (!*subvol_name)
446                         return -ENOMEM;
447         }
448         return error;
449 }
450
451 static struct dentry *get_default_root(struct super_block *sb,
452                                        u64 subvol_objectid)
453 {
454         struct btrfs_root *root = sb->s_fs_info;
455         struct btrfs_root *new_root;
456         struct btrfs_dir_item *di;
457         struct btrfs_path *path;
458         struct btrfs_key location;
459         struct inode *inode;
460         struct dentry *dentry;
461         u64 dir_id;
462         int new = 0;
463
464         /*
465          * We have a specific subvol we want to mount, just setup location and
466          * go look up the root.
467          */
468         if (subvol_objectid) {
469                 location.objectid = subvol_objectid;
470                 location.type = BTRFS_ROOT_ITEM_KEY;
471                 location.offset = (u64)-1;
472                 goto find_root;
473         }
474
475         path = btrfs_alloc_path();
476         if (!path)
477                 return ERR_PTR(-ENOMEM);
478         path->leave_spinning = 1;
479
480         /*
481          * Find the "default" dir item which points to the root item that we
482          * will mount by default if we haven't been given a specific subvolume
483          * to mount.
484          */
485         dir_id = btrfs_super_root_dir(&root->fs_info->super_copy);
486         di = btrfs_lookup_dir_item(NULL, root, path, dir_id, "default", 7, 0);
487         if (IS_ERR(di))
488                 return ERR_CAST(di);
489         if (!di) {
490                 /*
491                  * Ok the default dir item isn't there.  This is weird since
492                  * it's always been there, but don't freak out, just try and
493                  * mount to root most subvolume.
494                  */
495                 btrfs_free_path(path);
496                 dir_id = BTRFS_FIRST_FREE_OBJECTID;
497                 new_root = root->fs_info->fs_root;
498                 goto setup_root;
499         }
500
501         btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
502         btrfs_free_path(path);
503
504 find_root:
505         new_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
506         if (IS_ERR(new_root))
507                 return ERR_CAST(new_root);
508
509         if (btrfs_root_refs(&new_root->root_item) == 0)
510                 return ERR_PTR(-ENOENT);
511
512         dir_id = btrfs_root_dirid(&new_root->root_item);
513 setup_root:
514         location.objectid = dir_id;
515         location.type = BTRFS_INODE_ITEM_KEY;
516         location.offset = 0;
517
518         inode = btrfs_iget(sb, &location, new_root, &new);
519         if (IS_ERR(inode))
520                 return ERR_CAST(inode);
521
522         /*
523          * If we're just mounting the root most subvol put the inode and return
524          * a reference to the dentry.  We will have already gotten a reference
525          * to the inode in btrfs_fill_super so we're good to go.
526          */
527         if (!new && sb->s_root->d_inode == inode) {
528                 iput(inode);
529                 return dget(sb->s_root);
530         }
531
532         if (new) {
533                 const struct qstr name = { .name = "/", .len = 1 };
534
535                 /*
536                  * New inode, we need to make the dentry a sibling of s_root so
537                  * everything gets cleaned up properly on unmount.
538                  */
539                 dentry = d_alloc(sb->s_root, &name);
540                 if (!dentry) {
541                         iput(inode);
542                         return ERR_PTR(-ENOMEM);
543                 }
544                 d_splice_alias(inode, dentry);
545         } else {
546                 /*
547                  * We found the inode in cache, just find a dentry for it and
548                  * put the reference to the inode we just got.
549                  */
550                 dentry = d_find_alias(inode);
551                 iput(inode);
552         }
553
554         return dentry;
555 }
556
557 static int btrfs_fill_super(struct super_block *sb,
558                             struct btrfs_fs_devices *fs_devices,
559                             void *data, int silent)
560 {
561         struct inode *inode;
562         struct dentry *root_dentry;
563         struct btrfs_root *tree_root;
564         struct btrfs_key key;
565         int err;
566
567         sb->s_maxbytes = MAX_LFS_FILESIZE;
568         sb->s_magic = BTRFS_SUPER_MAGIC;
569         sb->s_op = &btrfs_super_ops;
570         sb->s_d_op = &btrfs_dentry_operations;
571         sb->s_export_op = &btrfs_export_ops;
572         sb->s_xattr = btrfs_xattr_handlers;
573         sb->s_time_gran = 1;
574 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
575         sb->s_flags |= MS_POSIXACL;
576 #endif
577
578         tree_root = open_ctree(sb, fs_devices, (char *)data);
579
580         if (IS_ERR(tree_root)) {
581                 printk("btrfs: open_ctree failed\n");
582                 return PTR_ERR(tree_root);
583         }
584         sb->s_fs_info = tree_root;
585
586         key.objectid = BTRFS_FIRST_FREE_OBJECTID;
587         key.type = BTRFS_INODE_ITEM_KEY;
588         key.offset = 0;
589         inode = btrfs_iget(sb, &key, tree_root->fs_info->fs_root, NULL);
590         if (IS_ERR(inode)) {
591                 err = PTR_ERR(inode);
592                 goto fail_close;
593         }
594
595         root_dentry = d_alloc_root(inode);
596         if (!root_dentry) {
597                 iput(inode);
598                 err = -ENOMEM;
599                 goto fail_close;
600         }
601
602         sb->s_root = root_dentry;
603
604         save_mount_options(sb, data);
605         return 0;
606
607 fail_close:
608         close_ctree(tree_root);
609         return err;
610 }
611
612 int btrfs_sync_fs(struct super_block *sb, int wait)
613 {
614         struct btrfs_trans_handle *trans;
615         struct btrfs_root *root = btrfs_sb(sb);
616         int ret;
617
618         if (!wait) {
619                 filemap_flush(root->fs_info->btree_inode->i_mapping);
620                 return 0;
621         }
622
623         btrfs_start_delalloc_inodes(root, 0);
624         btrfs_wait_ordered_extents(root, 0, 0);
625
626         trans = btrfs_start_transaction(root, 0);
627         if (IS_ERR(trans))
628                 return PTR_ERR(trans);
629         ret = btrfs_commit_transaction(trans, root);
630         return ret;
631 }
632
633 static int btrfs_show_options(struct seq_file *seq, struct vfsmount *vfs)
634 {
635         struct btrfs_root *root = btrfs_sb(vfs->mnt_sb);
636         struct btrfs_fs_info *info = root->fs_info;
637
638         if (btrfs_test_opt(root, DEGRADED))
639                 seq_puts(seq, ",degraded");
640         if (btrfs_test_opt(root, NODATASUM))
641                 seq_puts(seq, ",nodatasum");
642         if (btrfs_test_opt(root, NODATACOW))
643                 seq_puts(seq, ",nodatacow");
644         if (btrfs_test_opt(root, NOBARRIER))
645                 seq_puts(seq, ",nobarrier");
646         if (info->max_inline != 8192 * 1024)
647                 seq_printf(seq, ",max_inline=%llu",
648                            (unsigned long long)info->max_inline);
649         if (info->alloc_start != 0)
650                 seq_printf(seq, ",alloc_start=%llu",
651                            (unsigned long long)info->alloc_start);
652         if (info->thread_pool_size !=  min_t(unsigned long,
653                                              num_online_cpus() + 2, 8))
654                 seq_printf(seq, ",thread_pool=%d", info->thread_pool_size);
655         if (btrfs_test_opt(root, COMPRESS))
656                 seq_puts(seq, ",compress");
657         if (btrfs_test_opt(root, NOSSD))
658                 seq_puts(seq, ",nossd");
659         if (btrfs_test_opt(root, SSD_SPREAD))
660                 seq_puts(seq, ",ssd_spread");
661         else if (btrfs_test_opt(root, SSD))
662                 seq_puts(seq, ",ssd");
663         if (btrfs_test_opt(root, NOTREELOG))
664                 seq_puts(seq, ",notreelog");
665         if (btrfs_test_opt(root, FLUSHONCOMMIT))
666                 seq_puts(seq, ",flushoncommit");
667         if (btrfs_test_opt(root, DISCARD))
668                 seq_puts(seq, ",discard");
669         if (!(root->fs_info->sb->s_flags & MS_POSIXACL))
670                 seq_puts(seq, ",noacl");
671         return 0;
672 }
673
674 static int btrfs_test_super(struct super_block *s, void *data)
675 {
676         struct btrfs_root *test_root = data;
677         struct btrfs_root *root = btrfs_sb(s);
678
679         /*
680          * If this super block is going away, return false as it
681          * can't match as an existing super block.
682          */
683         if (!atomic_read(&s->s_active))
684                 return 0;
685         return root->fs_info->fs_devices == test_root->fs_info->fs_devices;
686 }
687
688 static int btrfs_set_super(struct super_block *s, void *data)
689 {
690         s->s_fs_info = data;
691
692         return set_anon_super(s, data);
693 }
694
695
696 /*
697  * Find a superblock for the given device / mount point.
698  *
699  * Note:  This is based on get_sb_bdev from fs/super.c with a few additions
700  *        for multiple device setup.  Make sure to keep it in sync.
701  */
702 static struct dentry *btrfs_mount(struct file_system_type *fs_type, int flags,
703                 const char *dev_name, void *data)
704 {
705         struct block_device *bdev = NULL;
706         struct super_block *s;
707         struct dentry *root;
708         struct btrfs_fs_devices *fs_devices = NULL;
709         struct btrfs_root *tree_root = NULL;
710         struct btrfs_fs_info *fs_info = NULL;
711         fmode_t mode = FMODE_READ;
712         char *subvol_name = NULL;
713         u64 subvol_objectid = 0;
714         int error = 0;
715
716         if (!(flags & MS_RDONLY))
717                 mode |= FMODE_WRITE;
718
719         error = btrfs_parse_early_options(data, mode, fs_type,
720                                           &subvol_name, &subvol_objectid,
721                                           &fs_devices);
722         if (error)
723                 return ERR_PTR(error);
724
725         error = btrfs_scan_one_device(dev_name, mode, fs_type, &fs_devices);
726         if (error)
727                 goto error_free_subvol_name;
728
729         error = btrfs_open_devices(fs_devices, mode, fs_type);
730         if (error)
731                 goto error_free_subvol_name;
732
733         if (!(flags & MS_RDONLY) && fs_devices->rw_devices == 0) {
734                 error = -EACCES;
735                 goto error_close_devices;
736         }
737
738         /*
739          * Setup a dummy root and fs_info for test/set super.  This is because
740          * we don't actually fill this stuff out until open_ctree, but we need
741          * it for searching for existing supers, so this lets us do that and
742          * then open_ctree will properly initialize everything later.
743          */
744         fs_info = kzalloc(sizeof(struct btrfs_fs_info), GFP_NOFS);
745         tree_root = kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
746         if (!fs_info || !tree_root) {
747                 error = -ENOMEM;
748                 goto error_close_devices;
749         }
750         fs_info->tree_root = tree_root;
751         fs_info->fs_devices = fs_devices;
752         tree_root->fs_info = fs_info;
753
754         bdev = fs_devices->latest_bdev;
755         s = sget(fs_type, btrfs_test_super, btrfs_set_super, tree_root);
756         if (IS_ERR(s))
757                 goto error_s;
758
759         if (s->s_root) {
760                 if ((flags ^ s->s_flags) & MS_RDONLY) {
761                         deactivate_locked_super(s);
762                         error = -EBUSY;
763                         goto error_close_devices;
764                 }
765
766                 btrfs_close_devices(fs_devices);
767                 kfree(fs_info);
768                 kfree(tree_root);
769         } else {
770                 char b[BDEVNAME_SIZE];
771
772                 s->s_flags = flags;
773                 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
774                 error = btrfs_fill_super(s, fs_devices, data,
775                                          flags & MS_SILENT ? 1 : 0);
776                 if (error) {
777                         deactivate_locked_super(s);
778                         goto error_free_subvol_name;
779                 }
780
781                 btrfs_sb(s)->fs_info->bdev_holder = fs_type;
782                 s->s_flags |= MS_ACTIVE;
783         }
784
785         root = get_default_root(s, subvol_objectid);
786         if (IS_ERR(root)) {
787                 error = PTR_ERR(root);
788                 deactivate_locked_super(s);
789                 goto error_free_subvol_name;
790         }
791         /* if they gave us a subvolume name bind mount into that */
792         if (strcmp(subvol_name, ".")) {
793                 struct dentry *new_root;
794                 mutex_lock(&root->d_inode->i_mutex);
795                 new_root = lookup_one_len(subvol_name, root,
796                                       strlen(subvol_name));
797                 mutex_unlock(&root->d_inode->i_mutex);
798
799                 if (IS_ERR(new_root)) {
800                         dput(root);
801                         deactivate_locked_super(s);
802                         error = PTR_ERR(new_root);
803                         goto error_free_subvol_name;
804                 }
805                 if (!new_root->d_inode) {
806                         dput(root);
807                         dput(new_root);
808                         deactivate_locked_super(s);
809                         error = -ENXIO;
810                         goto error_free_subvol_name;
811                 }
812                 dput(root);
813                 root = new_root;
814         }
815
816         kfree(subvol_name);
817         return root;
818
819 error_s:
820         error = PTR_ERR(s);
821 error_close_devices:
822         btrfs_close_devices(fs_devices);
823         kfree(fs_info);
824         kfree(tree_root);
825 error_free_subvol_name:
826         kfree(subvol_name);
827         return ERR_PTR(error);
828 }
829
830 static int btrfs_remount(struct super_block *sb, int *flags, char *data)
831 {
832         struct btrfs_root *root = btrfs_sb(sb);
833         int ret;
834
835         ret = btrfs_parse_options(root, data);
836         if (ret)
837                 return -EINVAL;
838
839         if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
840                 return 0;
841
842         if (*flags & MS_RDONLY) {
843                 sb->s_flags |= MS_RDONLY;
844
845                 ret =  btrfs_commit_super(root);
846                 WARN_ON(ret);
847         } else {
848                 if (root->fs_info->fs_devices->rw_devices == 0)
849                         return -EACCES;
850
851                 if (btrfs_super_log_root(&root->fs_info->super_copy) != 0)
852                         return -EINVAL;
853
854                 ret = btrfs_cleanup_fs_roots(root->fs_info);
855                 WARN_ON(ret);
856
857                 /* recover relocation */
858                 ret = btrfs_recover_relocation(root);
859                 WARN_ON(ret);
860
861                 sb->s_flags &= ~MS_RDONLY;
862         }
863
864         return 0;
865 }
866
867 /*
868  * The helper to calc the free space on the devices that can be used to store
869  * file data.
870  */
871 static int btrfs_calc_avail_data_space(struct btrfs_root *root, u64 *free_bytes)
872 {
873         struct btrfs_fs_info *fs_info = root->fs_info;
874         struct btrfs_device_info *devices_info;
875         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
876         struct btrfs_device *device;
877         u64 skip_space;
878         u64 type;
879         u64 avail_space;
880         u64 used_space;
881         u64 min_stripe_size;
882         int min_stripes = 1;
883         int i = 0, nr_devices;
884         int ret;
885
886         nr_devices = fs_info->fs_devices->rw_devices;
887         BUG_ON(!nr_devices);
888
889         devices_info = kmalloc(sizeof(*devices_info) * nr_devices,
890                                GFP_NOFS);
891         if (!devices_info)
892                 return -ENOMEM;
893
894         /* calc min stripe number for data space alloction */
895         type = btrfs_get_alloc_profile(root, 1);
896         if (type & BTRFS_BLOCK_GROUP_RAID0)
897                 min_stripes = 2;
898         else if (type & BTRFS_BLOCK_GROUP_RAID1)
899                 min_stripes = 2;
900         else if (type & BTRFS_BLOCK_GROUP_RAID10)
901                 min_stripes = 4;
902
903         if (type & BTRFS_BLOCK_GROUP_DUP)
904                 min_stripe_size = 2 * BTRFS_STRIPE_LEN;
905         else
906                 min_stripe_size = BTRFS_STRIPE_LEN;
907
908         list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
909                 if (!device->in_fs_metadata)
910                         continue;
911
912                 avail_space = device->total_bytes - device->bytes_used;
913
914                 /* align with stripe_len */
915                 do_div(avail_space, BTRFS_STRIPE_LEN);
916                 avail_space *= BTRFS_STRIPE_LEN;
917
918                 /*
919                  * In order to avoid overwritting the superblock on the drive,
920                  * btrfs starts at an offset of at least 1MB when doing chunk
921                  * allocation.
922                  */
923                 skip_space = 1024 * 1024;
924
925                 /* user can set the offset in fs_info->alloc_start. */
926                 if (fs_info->alloc_start + BTRFS_STRIPE_LEN <=
927                     device->total_bytes)
928                         skip_space = max(fs_info->alloc_start, skip_space);
929
930                 /*
931                  * btrfs can not use the free space in [0, skip_space - 1],
932                  * we must subtract it from the total. In order to implement
933                  * it, we account the used space in this range first.
934                  */
935                 ret = btrfs_account_dev_extents_size(device, 0, skip_space - 1,
936                                                      &used_space);
937                 if (ret) {
938                         kfree(devices_info);
939                         return ret;
940                 }
941
942                 /* calc the free space in [0, skip_space - 1] */
943                 skip_space -= used_space;
944
945                 /*
946                  * we can use the free space in [0, skip_space - 1], subtract
947                  * it from the total.
948                  */
949                 if (avail_space && avail_space >= skip_space)
950                         avail_space -= skip_space;
951                 else
952                         avail_space = 0;
953
954                 if (avail_space < min_stripe_size)
955                         continue;
956
957                 devices_info[i].dev = device;
958                 devices_info[i].max_avail = avail_space;
959
960                 i++;
961         }
962
963         nr_devices = i;
964
965         btrfs_descending_sort_devices(devices_info, nr_devices);
966
967         i = nr_devices - 1;
968         avail_space = 0;
969         while (nr_devices >= min_stripes) {
970                 if (devices_info[i].max_avail >= min_stripe_size) {
971                         int j;
972                         u64 alloc_size;
973
974                         avail_space += devices_info[i].max_avail * min_stripes;
975                         alloc_size = devices_info[i].max_avail;
976                         for (j = i + 1 - min_stripes; j <= i; j++)
977                                 devices_info[j].max_avail -= alloc_size;
978                 }
979                 i--;
980                 nr_devices--;
981         }
982
983         kfree(devices_info);
984         *free_bytes = avail_space;
985         return 0;
986 }
987
988 static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
989 {
990         struct btrfs_root *root = btrfs_sb(dentry->d_sb);
991         struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
992         struct list_head *head = &root->fs_info->space_info;
993         struct btrfs_space_info *found;
994         u64 total_used = 0;
995         u64 total_free_data = 0;
996         int bits = dentry->d_sb->s_blocksize_bits;
997         __be32 *fsid = (__be32 *)root->fs_info->fsid;
998         int ret;
999
1000         /* holding chunk_muext to avoid allocating new chunks */
1001         mutex_lock(&root->fs_info->chunk_mutex);
1002         rcu_read_lock();
1003         list_for_each_entry_rcu(found, head, list) {
1004                 if (found->flags & BTRFS_BLOCK_GROUP_DATA) {
1005                         total_free_data += found->disk_total - found->disk_used;
1006                         total_free_data -=
1007                                 btrfs_account_ro_block_groups_free_space(found);
1008                 }
1009
1010                 total_used += found->disk_used;
1011         }
1012         rcu_read_unlock();
1013
1014         buf->f_namelen = BTRFS_NAME_LEN;
1015         buf->f_blocks = btrfs_super_total_bytes(disk_super) >> bits;
1016         buf->f_bfree = buf->f_blocks - (total_used >> bits);
1017         buf->f_bsize = dentry->d_sb->s_blocksize;
1018         buf->f_type = BTRFS_SUPER_MAGIC;
1019         buf->f_bavail = total_free_data;
1020         ret = btrfs_calc_avail_data_space(root, &total_free_data);
1021         if (ret) {
1022                 mutex_unlock(&root->fs_info->chunk_mutex);
1023                 return ret;
1024         }
1025         buf->f_bavail += total_free_data;
1026         buf->f_bavail = buf->f_bavail >> bits;
1027         mutex_unlock(&root->fs_info->chunk_mutex);
1028
1029         /* We treat it as constant endianness (it doesn't matter _which_)
1030            because we want the fsid to come out the same whether mounted
1031            on a big-endian or little-endian host */
1032         buf->f_fsid.val[0] = be32_to_cpu(fsid[0]) ^ be32_to_cpu(fsid[2]);
1033         buf->f_fsid.val[1] = be32_to_cpu(fsid[1]) ^ be32_to_cpu(fsid[3]);
1034         /* Mask in the root object ID too, to disambiguate subvols */
1035         buf->f_fsid.val[0] ^= BTRFS_I(dentry->d_inode)->root->objectid >> 32;
1036         buf->f_fsid.val[1] ^= BTRFS_I(dentry->d_inode)->root->objectid;
1037
1038         return 0;
1039 }
1040
1041 static struct file_system_type btrfs_fs_type = {
1042         .owner          = THIS_MODULE,
1043         .name           = "btrfs",
1044         .mount          = btrfs_mount,
1045         .kill_sb        = kill_anon_super,
1046         .fs_flags       = FS_REQUIRES_DEV,
1047 };
1048
1049 /*
1050  * used by btrfsctl to scan devices when no FS is mounted
1051  */
1052 static long btrfs_control_ioctl(struct file *file, unsigned int cmd,
1053                                 unsigned long arg)
1054 {
1055         struct btrfs_ioctl_vol_args *vol;
1056         struct btrfs_fs_devices *fs_devices;
1057         int ret = -ENOTTY;
1058
1059         if (!capable(CAP_SYS_ADMIN))
1060                 return -EPERM;
1061
1062         vol = memdup_user((void __user *)arg, sizeof(*vol));
1063         if (IS_ERR(vol))
1064                 return PTR_ERR(vol);
1065
1066         switch (cmd) {
1067         case BTRFS_IOC_SCAN_DEV:
1068                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
1069                                             &btrfs_fs_type, &fs_devices);
1070                 break;
1071         }
1072
1073         kfree(vol);
1074         return ret;
1075 }
1076
1077 static int btrfs_freeze(struct super_block *sb)
1078 {
1079         struct btrfs_root *root = btrfs_sb(sb);
1080         mutex_lock(&root->fs_info->transaction_kthread_mutex);
1081         mutex_lock(&root->fs_info->cleaner_mutex);
1082         return 0;
1083 }
1084
1085 static int btrfs_unfreeze(struct super_block *sb)
1086 {
1087         struct btrfs_root *root = btrfs_sb(sb);
1088         mutex_unlock(&root->fs_info->cleaner_mutex);
1089         mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1090         return 0;
1091 }
1092
1093 static const struct super_operations btrfs_super_ops = {
1094         .drop_inode     = btrfs_drop_inode,
1095         .evict_inode    = btrfs_evict_inode,
1096         .put_super      = btrfs_put_super,
1097         .sync_fs        = btrfs_sync_fs,
1098         .show_options   = btrfs_show_options,
1099         .write_inode    = btrfs_write_inode,
1100         .dirty_inode    = btrfs_dirty_inode,
1101         .alloc_inode    = btrfs_alloc_inode,
1102         .destroy_inode  = btrfs_destroy_inode,
1103         .statfs         = btrfs_statfs,
1104         .remount_fs     = btrfs_remount,
1105         .freeze_fs      = btrfs_freeze,
1106         .unfreeze_fs    = btrfs_unfreeze,
1107 };
1108
1109 static const struct file_operations btrfs_ctl_fops = {
1110         .unlocked_ioctl  = btrfs_control_ioctl,
1111         .compat_ioctl = btrfs_control_ioctl,
1112         .owner   = THIS_MODULE,
1113         .llseek = noop_llseek,
1114 };
1115
1116 static struct miscdevice btrfs_misc = {
1117         .minor          = BTRFS_MINOR,
1118         .name           = "btrfs-control",
1119         .fops           = &btrfs_ctl_fops
1120 };
1121
1122 MODULE_ALIAS_MISCDEV(BTRFS_MINOR);
1123 MODULE_ALIAS("devname:btrfs-control");
1124
1125 static int btrfs_interface_init(void)
1126 {
1127         return misc_register(&btrfs_misc);
1128 }
1129
1130 static void btrfs_interface_exit(void)
1131 {
1132         if (misc_deregister(&btrfs_misc) < 0)
1133                 printk(KERN_INFO "misc_deregister failed for control device");
1134 }
1135
1136 static int __init init_btrfs_fs(void)
1137 {
1138         int err;
1139
1140         err = btrfs_init_sysfs();
1141         if (err)
1142                 return err;
1143
1144         err = btrfs_init_compress();
1145         if (err)
1146                 goto free_sysfs;
1147
1148         err = btrfs_init_cachep();
1149         if (err)
1150                 goto free_compress;
1151
1152         err = extent_io_init();
1153         if (err)
1154                 goto free_cachep;
1155
1156         err = extent_map_init();
1157         if (err)
1158                 goto free_extent_io;
1159
1160         err = btrfs_interface_init();
1161         if (err)
1162                 goto free_extent_map;
1163
1164         err = register_filesystem(&btrfs_fs_type);
1165         if (err)
1166                 goto unregister_ioctl;
1167
1168         printk(KERN_INFO "%s loaded\n", BTRFS_BUILD_VERSION);
1169         return 0;
1170
1171 unregister_ioctl:
1172         btrfs_interface_exit();
1173 free_extent_map:
1174         extent_map_exit();
1175 free_extent_io:
1176         extent_io_exit();
1177 free_cachep:
1178         btrfs_destroy_cachep();
1179 free_compress:
1180         btrfs_exit_compress();
1181 free_sysfs:
1182         btrfs_exit_sysfs();
1183         return err;
1184 }
1185
1186 static void __exit exit_btrfs_fs(void)
1187 {
1188         btrfs_destroy_cachep();
1189         extent_map_exit();
1190         extent_io_exit();
1191         btrfs_interface_exit();
1192         unregister_filesystem(&btrfs_fs_type);
1193         btrfs_exit_sysfs();
1194         btrfs_cleanup_fs_uuids();
1195         btrfs_exit_compress();
1196 }
1197
1198 module_init(init_btrfs_fs)
1199 module_exit(exit_btrfs_fs)
1200
1201 MODULE_LICENSE("GPL");