btrfs: move btrfs_cmp_device_free_bytes to super.c
[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 #define CREATE_TRACE_POINTS
56 #include <trace/events/btrfs.h>
57
58 static const struct super_operations btrfs_super_ops;
59
60 static const char *btrfs_decode_error(struct btrfs_fs_info *fs_info, int errno,
61                                       char nbuf[16])
62 {
63         char *errstr = NULL;
64
65         switch (errno) {
66         case -EIO:
67                 errstr = "IO failure";
68                 break;
69         case -ENOMEM:
70                 errstr = "Out of memory";
71                 break;
72         case -EROFS:
73                 errstr = "Readonly filesystem";
74                 break;
75         default:
76                 if (nbuf) {
77                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
78                                 errstr = nbuf;
79                 }
80                 break;
81         }
82
83         return errstr;
84 }
85
86 static void __save_error_info(struct btrfs_fs_info *fs_info)
87 {
88         /*
89          * today we only save the error info into ram.  Long term we'll
90          * also send it down to the disk
91          */
92         fs_info->fs_state = BTRFS_SUPER_FLAG_ERROR;
93 }
94
95 /* NOTE:
96  *      We move write_super stuff at umount in order to avoid deadlock
97  *      for umount hold all lock.
98  */
99 static void save_error_info(struct btrfs_fs_info *fs_info)
100 {
101         __save_error_info(fs_info);
102 }
103
104 /* btrfs handle error by forcing the filesystem readonly */
105 static void btrfs_handle_error(struct btrfs_fs_info *fs_info)
106 {
107         struct super_block *sb = fs_info->sb;
108
109         if (sb->s_flags & MS_RDONLY)
110                 return;
111
112         if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
113                 sb->s_flags |= MS_RDONLY;
114                 printk(KERN_INFO "btrfs is forced readonly\n");
115         }
116 }
117
118 /*
119  * __btrfs_std_error decodes expected errors from the caller and
120  * invokes the approciate error response.
121  */
122 void __btrfs_std_error(struct btrfs_fs_info *fs_info, const char *function,
123                      unsigned int line, int errno)
124 {
125         struct super_block *sb = fs_info->sb;
126         char nbuf[16];
127         const char *errstr;
128
129         /*
130          * Special case: if the error is EROFS, and we're already
131          * under MS_RDONLY, then it is safe here.
132          */
133         if (errno == -EROFS && (sb->s_flags & MS_RDONLY))
134                 return;
135
136         errstr = btrfs_decode_error(fs_info, errno, nbuf);
137         printk(KERN_CRIT "BTRFS error (device %s) in %s:%d: %s\n",
138                 sb->s_id, function, line, errstr);
139         save_error_info(fs_info);
140
141         btrfs_handle_error(fs_info);
142 }
143
144 static void btrfs_put_super(struct super_block *sb)
145 {
146         struct btrfs_root *root = btrfs_sb(sb);
147         int ret;
148
149         ret = close_ctree(root);
150         sb->s_fs_info = NULL;
151
152         (void)ret; /* FIXME: need to fix VFS to return error? */
153 }
154
155 enum {
156         Opt_degraded, Opt_subvol, Opt_subvolid, Opt_device, Opt_nodatasum,
157         Opt_nodatacow, Opt_max_inline, Opt_alloc_start, Opt_nobarrier, Opt_ssd,
158         Opt_nossd, Opt_ssd_spread, Opt_thread_pool, Opt_noacl, Opt_compress,
159         Opt_compress_type, Opt_compress_force, Opt_compress_force_type,
160         Opt_notreelog, Opt_ratio, Opt_flushoncommit, Opt_discard,
161         Opt_space_cache, Opt_clear_cache, Opt_user_subvol_rm_allowed,
162         Opt_enospc_debug, Opt_subvolrootid, Opt_err,
163 };
164
165 static match_table_t tokens = {
166         {Opt_degraded, "degraded"},
167         {Opt_subvol, "subvol=%s"},
168         {Opt_subvolid, "subvolid=%d"},
169         {Opt_device, "device=%s"},
170         {Opt_nodatasum, "nodatasum"},
171         {Opt_nodatacow, "nodatacow"},
172         {Opt_nobarrier, "nobarrier"},
173         {Opt_max_inline, "max_inline=%s"},
174         {Opt_alloc_start, "alloc_start=%s"},
175         {Opt_thread_pool, "thread_pool=%d"},
176         {Opt_compress, "compress"},
177         {Opt_compress_type, "compress=%s"},
178         {Opt_compress_force, "compress-force"},
179         {Opt_compress_force_type, "compress-force=%s"},
180         {Opt_ssd, "ssd"},
181         {Opt_ssd_spread, "ssd_spread"},
182         {Opt_nossd, "nossd"},
183         {Opt_noacl, "noacl"},
184         {Opt_notreelog, "notreelog"},
185         {Opt_flushoncommit, "flushoncommit"},
186         {Opt_ratio, "metadata_ratio=%d"},
187         {Opt_discard, "discard"},
188         {Opt_space_cache, "space_cache"},
189         {Opt_clear_cache, "clear_cache"},
190         {Opt_user_subvol_rm_allowed, "user_subvol_rm_allowed"},
191         {Opt_enospc_debug, "enospc_debug"},
192         {Opt_subvolrootid, "subvolrootid=%d"},
193         {Opt_err, NULL},
194 };
195
196 /*
197  * Regular mount options parser.  Everything that is needed only when
198  * reading in a new superblock is parsed here.
199  */
200 int btrfs_parse_options(struct btrfs_root *root, char *options)
201 {
202         struct btrfs_fs_info *info = root->fs_info;
203         substring_t args[MAX_OPT_ARGS];
204         char *p, *num, *orig;
205         int intarg;
206         int ret = 0;
207         char *compress_type;
208         bool compress_force = false;
209
210         if (!options)
211                 return 0;
212
213         /*
214          * strsep changes the string, duplicate it because parse_options
215          * gets called twice
216          */
217         options = kstrdup(options, GFP_NOFS);
218         if (!options)
219                 return -ENOMEM;
220
221         orig = options;
222
223         while ((p = strsep(&options, ",")) != NULL) {
224                 int token;
225                 if (!*p)
226                         continue;
227
228                 token = match_token(p, tokens, args);
229                 switch (token) {
230                 case Opt_degraded:
231                         printk(KERN_INFO "btrfs: allowing degraded mounts\n");
232                         btrfs_set_opt(info->mount_opt, DEGRADED);
233                         break;
234                 case Opt_subvol:
235                 case Opt_subvolid:
236                 case Opt_subvolrootid:
237                 case Opt_device:
238                         /*
239                          * These are parsed by btrfs_parse_early_options
240                          * and can be happily ignored here.
241                          */
242                         break;
243                 case Opt_nodatasum:
244                         printk(KERN_INFO "btrfs: setting nodatasum\n");
245                         btrfs_set_opt(info->mount_opt, NODATASUM);
246                         break;
247                 case Opt_nodatacow:
248                         printk(KERN_INFO "btrfs: setting nodatacow\n");
249                         btrfs_set_opt(info->mount_opt, NODATACOW);
250                         btrfs_set_opt(info->mount_opt, NODATASUM);
251                         break;
252                 case Opt_compress_force:
253                 case Opt_compress_force_type:
254                         compress_force = true;
255                 case Opt_compress:
256                 case Opt_compress_type:
257                         if (token == Opt_compress ||
258                             token == Opt_compress_force ||
259                             strcmp(args[0].from, "zlib") == 0) {
260                                 compress_type = "zlib";
261                                 info->compress_type = BTRFS_COMPRESS_ZLIB;
262                         } else if (strcmp(args[0].from, "lzo") == 0) {
263                                 compress_type = "lzo";
264                                 info->compress_type = BTRFS_COMPRESS_LZO;
265                         } else {
266                                 ret = -EINVAL;
267                                 goto out;
268                         }
269
270                         btrfs_set_opt(info->mount_opt, COMPRESS);
271                         if (compress_force) {
272                                 btrfs_set_opt(info->mount_opt, FORCE_COMPRESS);
273                                 pr_info("btrfs: force %s compression\n",
274                                         compress_type);
275                         } else
276                                 pr_info("btrfs: use %s compression\n",
277                                         compress_type);
278                         break;
279                 case Opt_ssd:
280                         printk(KERN_INFO "btrfs: use ssd allocation scheme\n");
281                         btrfs_set_opt(info->mount_opt, SSD);
282                         break;
283                 case Opt_ssd_spread:
284                         printk(KERN_INFO "btrfs: use spread ssd "
285                                "allocation scheme\n");
286                         btrfs_set_opt(info->mount_opt, SSD);
287                         btrfs_set_opt(info->mount_opt, SSD_SPREAD);
288                         break;
289                 case Opt_nossd:
290                         printk(KERN_INFO "btrfs: not using ssd allocation "
291                                "scheme\n");
292                         btrfs_set_opt(info->mount_opt, NOSSD);
293                         btrfs_clear_opt(info->mount_opt, SSD);
294                         btrfs_clear_opt(info->mount_opt, SSD_SPREAD);
295                         break;
296                 case Opt_nobarrier:
297                         printk(KERN_INFO "btrfs: turning off barriers\n");
298                         btrfs_set_opt(info->mount_opt, NOBARRIER);
299                         break;
300                 case Opt_thread_pool:
301                         intarg = 0;
302                         match_int(&args[0], &intarg);
303                         if (intarg) {
304                                 info->thread_pool_size = intarg;
305                                 printk(KERN_INFO "btrfs: thread pool %d\n",
306                                        info->thread_pool_size);
307                         }
308                         break;
309                 case Opt_max_inline:
310                         num = match_strdup(&args[0]);
311                         if (num) {
312                                 info->max_inline = memparse(num, NULL);
313                                 kfree(num);
314
315                                 if (info->max_inline) {
316                                         info->max_inline = max_t(u64,
317                                                 info->max_inline,
318                                                 root->sectorsize);
319                                 }
320                                 printk(KERN_INFO "btrfs: max_inline at %llu\n",
321                                         (unsigned long long)info->max_inline);
322                         }
323                         break;
324                 case Opt_alloc_start:
325                         num = match_strdup(&args[0]);
326                         if (num) {
327                                 info->alloc_start = memparse(num, NULL);
328                                 kfree(num);
329                                 printk(KERN_INFO
330                                         "btrfs: allocations start at %llu\n",
331                                         (unsigned long long)info->alloc_start);
332                         }
333                         break;
334                 case Opt_noacl:
335                         root->fs_info->sb->s_flags &= ~MS_POSIXACL;
336                         break;
337                 case Opt_notreelog:
338                         printk(KERN_INFO "btrfs: disabling tree log\n");
339                         btrfs_set_opt(info->mount_opt, NOTREELOG);
340                         break;
341                 case Opt_flushoncommit:
342                         printk(KERN_INFO "btrfs: turning on flush-on-commit\n");
343                         btrfs_set_opt(info->mount_opt, FLUSHONCOMMIT);
344                         break;
345                 case Opt_ratio:
346                         intarg = 0;
347                         match_int(&args[0], &intarg);
348                         if (intarg) {
349                                 info->metadata_ratio = intarg;
350                                 printk(KERN_INFO "btrfs: metadata ratio %d\n",
351                                        info->metadata_ratio);
352                         }
353                         break;
354                 case Opt_discard:
355                         btrfs_set_opt(info->mount_opt, DISCARD);
356                         break;
357                 case Opt_space_cache:
358                         printk(KERN_INFO "btrfs: enabling disk space caching\n");
359                         btrfs_set_opt(info->mount_opt, SPACE_CACHE);
360                         break;
361                 case Opt_clear_cache:
362                         printk(KERN_INFO "btrfs: force clearing of disk cache\n");
363                         btrfs_set_opt(info->mount_opt, CLEAR_CACHE);
364                         break;
365                 case Opt_user_subvol_rm_allowed:
366                         btrfs_set_opt(info->mount_opt, USER_SUBVOL_RM_ALLOWED);
367                         break;
368                 case Opt_enospc_debug:
369                         btrfs_set_opt(info->mount_opt, ENOSPC_DEBUG);
370                         break;
371                 case Opt_err:
372                         printk(KERN_INFO "btrfs: unrecognized mount option "
373                                "'%s'\n", p);
374                         ret = -EINVAL;
375                         goto out;
376                 default:
377                         break;
378                 }
379         }
380 out:
381         kfree(orig);
382         return ret;
383 }
384
385 /*
386  * Parse mount options that are required early in the mount process.
387  *
388  * All other options will be parsed on much later in the mount process and
389  * only when we need to allocate a new super block.
390  */
391 static int btrfs_parse_early_options(const char *options, fmode_t flags,
392                 void *holder, char **subvol_name, u64 *subvol_objectid,
393                 u64 *subvol_rootid, struct btrfs_fs_devices **fs_devices)
394 {
395         substring_t args[MAX_OPT_ARGS];
396         char *opts, *orig, *p;
397         int error = 0;
398         int intarg;
399
400         if (!options)
401                 goto out;
402
403         /*
404          * strsep changes the string, duplicate it because parse_options
405          * gets called twice
406          */
407         opts = kstrdup(options, GFP_KERNEL);
408         if (!opts)
409                 return -ENOMEM;
410         orig = opts;
411
412         while ((p = strsep(&opts, ",")) != NULL) {
413                 int token;
414                 if (!*p)
415                         continue;
416
417                 token = match_token(p, tokens, args);
418                 switch (token) {
419                 case Opt_subvol:
420                         *subvol_name = match_strdup(&args[0]);
421                         break;
422                 case Opt_subvolid:
423                         intarg = 0;
424                         error = match_int(&args[0], &intarg);
425                         if (!error) {
426                                 /* we want the original fs_tree */
427                                 if (!intarg)
428                                         *subvol_objectid =
429                                                 BTRFS_FS_TREE_OBJECTID;
430                                 else
431                                         *subvol_objectid = intarg;
432                         }
433                         break;
434                 case Opt_subvolrootid:
435                         intarg = 0;
436                         error = match_int(&args[0], &intarg);
437                         if (!error) {
438                                 /* we want the original fs_tree */
439                                 if (!intarg)
440                                         *subvol_rootid =
441                                                 BTRFS_FS_TREE_OBJECTID;
442                                 else
443                                         *subvol_rootid = intarg;
444                         }
445                         break;
446                 case Opt_device:
447                         error = btrfs_scan_one_device(match_strdup(&args[0]),
448                                         flags, holder, fs_devices);
449                         if (error)
450                                 goto out_free_opts;
451                         break;
452                 default:
453                         break;
454                 }
455         }
456
457  out_free_opts:
458         kfree(orig);
459  out:
460         /*
461          * If no subvolume name is specified we use the default one.  Allocate
462          * a copy of the string "." here so that code later in the
463          * mount path doesn't care if it's the default volume or another one.
464          */
465         if (!*subvol_name) {
466                 *subvol_name = kstrdup(".", GFP_KERNEL);
467                 if (!*subvol_name)
468                         return -ENOMEM;
469         }
470         return error;
471 }
472
473 static struct dentry *get_default_root(struct super_block *sb,
474                                        u64 subvol_objectid)
475 {
476         struct btrfs_root *root = sb->s_fs_info;
477         struct btrfs_root *new_root;
478         struct btrfs_dir_item *di;
479         struct btrfs_path *path;
480         struct btrfs_key location;
481         struct inode *inode;
482         struct dentry *dentry;
483         u64 dir_id;
484         int new = 0;
485
486         /*
487          * We have a specific subvol we want to mount, just setup location and
488          * go look up the root.
489          */
490         if (subvol_objectid) {
491                 location.objectid = subvol_objectid;
492                 location.type = BTRFS_ROOT_ITEM_KEY;
493                 location.offset = (u64)-1;
494                 goto find_root;
495         }
496
497         path = btrfs_alloc_path();
498         if (!path)
499                 return ERR_PTR(-ENOMEM);
500         path->leave_spinning = 1;
501
502         /*
503          * Find the "default" dir item which points to the root item that we
504          * will mount by default if we haven't been given a specific subvolume
505          * to mount.
506          */
507         dir_id = btrfs_super_root_dir(&root->fs_info->super_copy);
508         di = btrfs_lookup_dir_item(NULL, root, path, dir_id, "default", 7, 0);
509         if (IS_ERR(di))
510                 return ERR_CAST(di);
511         if (!di) {
512                 /*
513                  * Ok the default dir item isn't there.  This is weird since
514                  * it's always been there, but don't freak out, just try and
515                  * mount to root most subvolume.
516                  */
517                 btrfs_free_path(path);
518                 dir_id = BTRFS_FIRST_FREE_OBJECTID;
519                 new_root = root->fs_info->fs_root;
520                 goto setup_root;
521         }
522
523         btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
524         btrfs_free_path(path);
525
526 find_root:
527         new_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
528         if (IS_ERR(new_root))
529                 return ERR_CAST(new_root);
530
531         if (btrfs_root_refs(&new_root->root_item) == 0)
532                 return ERR_PTR(-ENOENT);
533
534         dir_id = btrfs_root_dirid(&new_root->root_item);
535 setup_root:
536         location.objectid = dir_id;
537         location.type = BTRFS_INODE_ITEM_KEY;
538         location.offset = 0;
539
540         inode = btrfs_iget(sb, &location, new_root, &new);
541         if (IS_ERR(inode))
542                 return ERR_CAST(inode);
543
544         /*
545          * If we're just mounting the root most subvol put the inode and return
546          * a reference to the dentry.  We will have already gotten a reference
547          * to the inode in btrfs_fill_super so we're good to go.
548          */
549         if (!new && sb->s_root->d_inode == inode) {
550                 iput(inode);
551                 return dget(sb->s_root);
552         }
553
554         if (new) {
555                 const struct qstr name = { .name = "/", .len = 1 };
556
557                 /*
558                  * New inode, we need to make the dentry a sibling of s_root so
559                  * everything gets cleaned up properly on unmount.
560                  */
561                 dentry = d_alloc(sb->s_root, &name);
562                 if (!dentry) {
563                         iput(inode);
564                         return ERR_PTR(-ENOMEM);
565                 }
566                 d_splice_alias(inode, dentry);
567         } else {
568                 /*
569                  * We found the inode in cache, just find a dentry for it and
570                  * put the reference to the inode we just got.
571                  */
572                 dentry = d_find_alias(inode);
573                 iput(inode);
574         }
575
576         return dentry;
577 }
578
579 static int btrfs_fill_super(struct super_block *sb,
580                             struct btrfs_fs_devices *fs_devices,
581                             void *data, int silent)
582 {
583         struct inode *inode;
584         struct dentry *root_dentry;
585         struct btrfs_root *tree_root;
586         struct btrfs_key key;
587         int err;
588
589         sb->s_maxbytes = MAX_LFS_FILESIZE;
590         sb->s_magic = BTRFS_SUPER_MAGIC;
591         sb->s_op = &btrfs_super_ops;
592         sb->s_d_op = &btrfs_dentry_operations;
593         sb->s_export_op = &btrfs_export_ops;
594         sb->s_xattr = btrfs_xattr_handlers;
595         sb->s_time_gran = 1;
596 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
597         sb->s_flags |= MS_POSIXACL;
598 #endif
599
600         tree_root = open_ctree(sb, fs_devices, (char *)data);
601
602         if (IS_ERR(tree_root)) {
603                 printk("btrfs: open_ctree failed\n");
604                 return PTR_ERR(tree_root);
605         }
606         sb->s_fs_info = tree_root;
607
608         key.objectid = BTRFS_FIRST_FREE_OBJECTID;
609         key.type = BTRFS_INODE_ITEM_KEY;
610         key.offset = 0;
611         inode = btrfs_iget(sb, &key, tree_root->fs_info->fs_root, NULL);
612         if (IS_ERR(inode)) {
613                 err = PTR_ERR(inode);
614                 goto fail_close;
615         }
616
617         root_dentry = d_alloc_root(inode);
618         if (!root_dentry) {
619                 iput(inode);
620                 err = -ENOMEM;
621                 goto fail_close;
622         }
623
624         sb->s_root = root_dentry;
625
626         save_mount_options(sb, data);
627         return 0;
628
629 fail_close:
630         close_ctree(tree_root);
631         return err;
632 }
633
634 int btrfs_sync_fs(struct super_block *sb, int wait)
635 {
636         struct btrfs_trans_handle *trans;
637         struct btrfs_root *root = btrfs_sb(sb);
638         int ret;
639
640         trace_btrfs_sync_fs(wait);
641
642         if (!wait) {
643                 filemap_flush(root->fs_info->btree_inode->i_mapping);
644                 return 0;
645         }
646
647         btrfs_start_delalloc_inodes(root, 0);
648         btrfs_wait_ordered_extents(root, 0, 0);
649
650         trans = btrfs_start_transaction(root, 0);
651         if (IS_ERR(trans))
652                 return PTR_ERR(trans);
653         ret = btrfs_commit_transaction(trans, root);
654         return ret;
655 }
656
657 static int btrfs_show_options(struct seq_file *seq, struct vfsmount *vfs)
658 {
659         struct btrfs_root *root = btrfs_sb(vfs->mnt_sb);
660         struct btrfs_fs_info *info = root->fs_info;
661         char *compress_type;
662
663         if (btrfs_test_opt(root, DEGRADED))
664                 seq_puts(seq, ",degraded");
665         if (btrfs_test_opt(root, NODATASUM))
666                 seq_puts(seq, ",nodatasum");
667         if (btrfs_test_opt(root, NODATACOW))
668                 seq_puts(seq, ",nodatacow");
669         if (btrfs_test_opt(root, NOBARRIER))
670                 seq_puts(seq, ",nobarrier");
671         if (info->max_inline != 8192 * 1024)
672                 seq_printf(seq, ",max_inline=%llu",
673                            (unsigned long long)info->max_inline);
674         if (info->alloc_start != 0)
675                 seq_printf(seq, ",alloc_start=%llu",
676                            (unsigned long long)info->alloc_start);
677         if (info->thread_pool_size !=  min_t(unsigned long,
678                                              num_online_cpus() + 2, 8))
679                 seq_printf(seq, ",thread_pool=%d", info->thread_pool_size);
680         if (btrfs_test_opt(root, COMPRESS)) {
681                 if (info->compress_type == BTRFS_COMPRESS_ZLIB)
682                         compress_type = "zlib";
683                 else
684                         compress_type = "lzo";
685                 if (btrfs_test_opt(root, FORCE_COMPRESS))
686                         seq_printf(seq, ",compress-force=%s", compress_type);
687                 else
688                         seq_printf(seq, ",compress=%s", compress_type);
689         }
690         if (btrfs_test_opt(root, NOSSD))
691                 seq_puts(seq, ",nossd");
692         if (btrfs_test_opt(root, SSD_SPREAD))
693                 seq_puts(seq, ",ssd_spread");
694         else if (btrfs_test_opt(root, SSD))
695                 seq_puts(seq, ",ssd");
696         if (btrfs_test_opt(root, NOTREELOG))
697                 seq_puts(seq, ",notreelog");
698         if (btrfs_test_opt(root, FLUSHONCOMMIT))
699                 seq_puts(seq, ",flushoncommit");
700         if (btrfs_test_opt(root, DISCARD))
701                 seq_puts(seq, ",discard");
702         if (!(root->fs_info->sb->s_flags & MS_POSIXACL))
703                 seq_puts(seq, ",noacl");
704         if (btrfs_test_opt(root, SPACE_CACHE))
705                 seq_puts(seq, ",space_cache");
706         if (btrfs_test_opt(root, CLEAR_CACHE))
707                 seq_puts(seq, ",clear_cache");
708         if (btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
709                 seq_puts(seq, ",user_subvol_rm_allowed");
710         return 0;
711 }
712
713 static int btrfs_test_super(struct super_block *s, void *data)
714 {
715         struct btrfs_root *test_root = data;
716         struct btrfs_root *root = btrfs_sb(s);
717
718         /*
719          * If this super block is going away, return false as it
720          * can't match as an existing super block.
721          */
722         if (!atomic_read(&s->s_active))
723                 return 0;
724         return root->fs_info->fs_devices == test_root->fs_info->fs_devices;
725 }
726
727 static int btrfs_set_super(struct super_block *s, void *data)
728 {
729         s->s_fs_info = data;
730
731         return set_anon_super(s, data);
732 }
733
734
735 /*
736  * Find a superblock for the given device / mount point.
737  *
738  * Note:  This is based on get_sb_bdev from fs/super.c with a few additions
739  *        for multiple device setup.  Make sure to keep it in sync.
740  */
741 static struct dentry *btrfs_mount(struct file_system_type *fs_type, int flags,
742                 const char *dev_name, void *data)
743 {
744         struct block_device *bdev = NULL;
745         struct super_block *s;
746         struct dentry *root;
747         struct btrfs_fs_devices *fs_devices = NULL;
748         struct btrfs_root *tree_root = NULL;
749         struct btrfs_fs_info *fs_info = NULL;
750         fmode_t mode = FMODE_READ;
751         char *subvol_name = NULL;
752         u64 subvol_objectid = 0;
753         u64 subvol_rootid = 0;
754         int error = 0;
755
756         if (!(flags & MS_RDONLY))
757                 mode |= FMODE_WRITE;
758
759         error = btrfs_parse_early_options(data, mode, fs_type,
760                                           &subvol_name, &subvol_objectid,
761                                           &subvol_rootid, &fs_devices);
762         if (error)
763                 return ERR_PTR(error);
764
765         error = btrfs_scan_one_device(dev_name, mode, fs_type, &fs_devices);
766         if (error)
767                 goto error_free_subvol_name;
768
769         error = btrfs_open_devices(fs_devices, mode, fs_type);
770         if (error)
771                 goto error_free_subvol_name;
772
773         if (!(flags & MS_RDONLY) && fs_devices->rw_devices == 0) {
774                 error = -EACCES;
775                 goto error_close_devices;
776         }
777
778         /*
779          * Setup a dummy root and fs_info for test/set super.  This is because
780          * we don't actually fill this stuff out until open_ctree, but we need
781          * it for searching for existing supers, so this lets us do that and
782          * then open_ctree will properly initialize everything later.
783          */
784         fs_info = kzalloc(sizeof(struct btrfs_fs_info), GFP_NOFS);
785         tree_root = kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
786         if (!fs_info || !tree_root) {
787                 error = -ENOMEM;
788                 goto error_close_devices;
789         }
790         fs_info->tree_root = tree_root;
791         fs_info->fs_devices = fs_devices;
792         tree_root->fs_info = fs_info;
793
794         bdev = fs_devices->latest_bdev;
795         s = sget(fs_type, btrfs_test_super, btrfs_set_super, tree_root);
796         if (IS_ERR(s))
797                 goto error_s;
798
799         if (s->s_root) {
800                 if ((flags ^ s->s_flags) & MS_RDONLY) {
801                         deactivate_locked_super(s);
802                         error = -EBUSY;
803                         goto error_close_devices;
804                 }
805
806                 btrfs_close_devices(fs_devices);
807                 kfree(fs_info);
808                 kfree(tree_root);
809         } else {
810                 char b[BDEVNAME_SIZE];
811
812                 s->s_flags = flags;
813                 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
814                 error = btrfs_fill_super(s, fs_devices, data,
815                                          flags & MS_SILENT ? 1 : 0);
816                 if (error) {
817                         deactivate_locked_super(s);
818                         goto error_free_subvol_name;
819                 }
820
821                 btrfs_sb(s)->fs_info->bdev_holder = fs_type;
822                 s->s_flags |= MS_ACTIVE;
823         }
824
825         /* if they gave us a subvolume name bind mount into that */
826         if (strcmp(subvol_name, ".")) {
827                 struct dentry *new_root;
828
829                 root = get_default_root(s, subvol_rootid);
830                 if (IS_ERR(root)) {
831                         error = PTR_ERR(root);
832                         deactivate_locked_super(s);
833                         goto error_free_subvol_name;
834                 }
835
836                 mutex_lock(&root->d_inode->i_mutex);
837                 new_root = lookup_one_len(subvol_name, root,
838                                       strlen(subvol_name));
839                 mutex_unlock(&root->d_inode->i_mutex);
840
841                 if (IS_ERR(new_root)) {
842                         dput(root);
843                         deactivate_locked_super(s);
844                         error = PTR_ERR(new_root);
845                         goto error_free_subvol_name;
846                 }
847                 if (!new_root->d_inode) {
848                         dput(root);
849                         dput(new_root);
850                         deactivate_locked_super(s);
851                         error = -ENXIO;
852                         goto error_free_subvol_name;
853                 }
854                 dput(root);
855                 root = new_root;
856         } else {
857                 root = get_default_root(s, subvol_objectid);
858                 if (IS_ERR(root)) {
859                         error = PTR_ERR(root);
860                         deactivate_locked_super(s);
861                         goto error_free_subvol_name;
862                 }
863         }
864
865         kfree(subvol_name);
866         return root;
867
868 error_s:
869         error = PTR_ERR(s);
870 error_close_devices:
871         btrfs_close_devices(fs_devices);
872         kfree(fs_info);
873         kfree(tree_root);
874 error_free_subvol_name:
875         kfree(subvol_name);
876         return ERR_PTR(error);
877 }
878
879 static int btrfs_remount(struct super_block *sb, int *flags, char *data)
880 {
881         struct btrfs_root *root = btrfs_sb(sb);
882         int ret;
883
884         ret = btrfs_parse_options(root, data);
885         if (ret)
886                 return -EINVAL;
887
888         if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
889                 return 0;
890
891         if (*flags & MS_RDONLY) {
892                 sb->s_flags |= MS_RDONLY;
893
894                 ret =  btrfs_commit_super(root);
895                 WARN_ON(ret);
896         } else {
897                 if (root->fs_info->fs_devices->rw_devices == 0)
898                         return -EACCES;
899
900                 if (btrfs_super_log_root(&root->fs_info->super_copy) != 0)
901                         return -EINVAL;
902
903                 ret = btrfs_cleanup_fs_roots(root->fs_info);
904                 WARN_ON(ret);
905
906                 /* recover relocation */
907                 ret = btrfs_recover_relocation(root);
908                 WARN_ON(ret);
909
910                 sb->s_flags &= ~MS_RDONLY;
911         }
912
913         return 0;
914 }
915
916 /* Used to sort the devices by max_avail(descending sort) */
917 static int btrfs_cmp_device_free_bytes(const void *dev_info1,
918                                        const void *dev_info2)
919 {
920         if (((struct btrfs_device_info *)dev_info1)->max_avail >
921             ((struct btrfs_device_info *)dev_info2)->max_avail)
922                 return -1;
923         else if (((struct btrfs_device_info *)dev_info1)->max_avail <
924                  ((struct btrfs_device_info *)dev_info2)->max_avail)
925                 return 1;
926         else
927         return 0;
928 }
929
930 /*
931  * sort the devices by max_avail, in which max free extent size of each device
932  * is stored.(Descending Sort)
933  */
934 static inline void btrfs_descending_sort_devices(
935                                         struct btrfs_device_info *devices,
936                                         size_t nr_devices)
937 {
938         sort(devices, nr_devices, sizeof(struct btrfs_device_info),
939              btrfs_cmp_device_free_bytes, NULL);
940 }
941
942 /*
943  * The helper to calc the free space on the devices that can be used to store
944  * file data.
945  */
946 static int btrfs_calc_avail_data_space(struct btrfs_root *root, u64 *free_bytes)
947 {
948         struct btrfs_fs_info *fs_info = root->fs_info;
949         struct btrfs_device_info *devices_info;
950         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
951         struct btrfs_device *device;
952         u64 skip_space;
953         u64 type;
954         u64 avail_space;
955         u64 used_space;
956         u64 min_stripe_size;
957         int min_stripes = 1;
958         int i = 0, nr_devices;
959         int ret;
960
961         nr_devices = fs_info->fs_devices->rw_devices;
962         BUG_ON(!nr_devices);
963
964         devices_info = kmalloc(sizeof(*devices_info) * nr_devices,
965                                GFP_NOFS);
966         if (!devices_info)
967                 return -ENOMEM;
968
969         /* calc min stripe number for data space alloction */
970         type = btrfs_get_alloc_profile(root, 1);
971         if (type & BTRFS_BLOCK_GROUP_RAID0)
972                 min_stripes = 2;
973         else if (type & BTRFS_BLOCK_GROUP_RAID1)
974                 min_stripes = 2;
975         else if (type & BTRFS_BLOCK_GROUP_RAID10)
976                 min_stripes = 4;
977
978         if (type & BTRFS_BLOCK_GROUP_DUP)
979                 min_stripe_size = 2 * BTRFS_STRIPE_LEN;
980         else
981                 min_stripe_size = BTRFS_STRIPE_LEN;
982
983         list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
984                 if (!device->in_fs_metadata)
985                         continue;
986
987                 avail_space = device->total_bytes - device->bytes_used;
988
989                 /* align with stripe_len */
990                 do_div(avail_space, BTRFS_STRIPE_LEN);
991                 avail_space *= BTRFS_STRIPE_LEN;
992
993                 /*
994                  * In order to avoid overwritting the superblock on the drive,
995                  * btrfs starts at an offset of at least 1MB when doing chunk
996                  * allocation.
997                  */
998                 skip_space = 1024 * 1024;
999
1000                 /* user can set the offset in fs_info->alloc_start. */
1001                 if (fs_info->alloc_start + BTRFS_STRIPE_LEN <=
1002                     device->total_bytes)
1003                         skip_space = max(fs_info->alloc_start, skip_space);
1004
1005                 /*
1006                  * btrfs can not use the free space in [0, skip_space - 1],
1007                  * we must subtract it from the total. In order to implement
1008                  * it, we account the used space in this range first.
1009                  */
1010                 ret = btrfs_account_dev_extents_size(device, 0, skip_space - 1,
1011                                                      &used_space);
1012                 if (ret) {
1013                         kfree(devices_info);
1014                         return ret;
1015                 }
1016
1017                 /* calc the free space in [0, skip_space - 1] */
1018                 skip_space -= used_space;
1019
1020                 /*
1021                  * we can use the free space in [0, skip_space - 1], subtract
1022                  * it from the total.
1023                  */
1024                 if (avail_space && avail_space >= skip_space)
1025                         avail_space -= skip_space;
1026                 else
1027                         avail_space = 0;
1028
1029                 if (avail_space < min_stripe_size)
1030                         continue;
1031
1032                 devices_info[i].dev = device;
1033                 devices_info[i].max_avail = avail_space;
1034
1035                 i++;
1036         }
1037
1038         nr_devices = i;
1039
1040         btrfs_descending_sort_devices(devices_info, nr_devices);
1041
1042         i = nr_devices - 1;
1043         avail_space = 0;
1044         while (nr_devices >= min_stripes) {
1045                 if (devices_info[i].max_avail >= min_stripe_size) {
1046                         int j;
1047                         u64 alloc_size;
1048
1049                         avail_space += devices_info[i].max_avail * min_stripes;
1050                         alloc_size = devices_info[i].max_avail;
1051                         for (j = i + 1 - min_stripes; j <= i; j++)
1052                                 devices_info[j].max_avail -= alloc_size;
1053                 }
1054                 i--;
1055                 nr_devices--;
1056         }
1057
1058         kfree(devices_info);
1059         *free_bytes = avail_space;
1060         return 0;
1061 }
1062
1063 static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
1064 {
1065         struct btrfs_root *root = btrfs_sb(dentry->d_sb);
1066         struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
1067         struct list_head *head = &root->fs_info->space_info;
1068         struct btrfs_space_info *found;
1069         u64 total_used = 0;
1070         u64 total_free_data = 0;
1071         int bits = dentry->d_sb->s_blocksize_bits;
1072         __be32 *fsid = (__be32 *)root->fs_info->fsid;
1073         int ret;
1074
1075         /* holding chunk_muext to avoid allocating new chunks */
1076         mutex_lock(&root->fs_info->chunk_mutex);
1077         rcu_read_lock();
1078         list_for_each_entry_rcu(found, head, list) {
1079                 if (found->flags & BTRFS_BLOCK_GROUP_DATA) {
1080                         total_free_data += found->disk_total - found->disk_used;
1081                         total_free_data -=
1082                                 btrfs_account_ro_block_groups_free_space(found);
1083                 }
1084
1085                 total_used += found->disk_used;
1086         }
1087         rcu_read_unlock();
1088
1089         buf->f_namelen = BTRFS_NAME_LEN;
1090         buf->f_blocks = btrfs_super_total_bytes(disk_super) >> bits;
1091         buf->f_bfree = buf->f_blocks - (total_used >> bits);
1092         buf->f_bsize = dentry->d_sb->s_blocksize;
1093         buf->f_type = BTRFS_SUPER_MAGIC;
1094         buf->f_bavail = total_free_data;
1095         ret = btrfs_calc_avail_data_space(root, &total_free_data);
1096         if (ret) {
1097                 mutex_unlock(&root->fs_info->chunk_mutex);
1098                 return ret;
1099         }
1100         buf->f_bavail += total_free_data;
1101         buf->f_bavail = buf->f_bavail >> bits;
1102         mutex_unlock(&root->fs_info->chunk_mutex);
1103
1104         /* We treat it as constant endianness (it doesn't matter _which_)
1105            because we want the fsid to come out the same whether mounted
1106            on a big-endian or little-endian host */
1107         buf->f_fsid.val[0] = be32_to_cpu(fsid[0]) ^ be32_to_cpu(fsid[2]);
1108         buf->f_fsid.val[1] = be32_to_cpu(fsid[1]) ^ be32_to_cpu(fsid[3]);
1109         /* Mask in the root object ID too, to disambiguate subvols */
1110         buf->f_fsid.val[0] ^= BTRFS_I(dentry->d_inode)->root->objectid >> 32;
1111         buf->f_fsid.val[1] ^= BTRFS_I(dentry->d_inode)->root->objectid;
1112
1113         return 0;
1114 }
1115
1116 static struct file_system_type btrfs_fs_type = {
1117         .owner          = THIS_MODULE,
1118         .name           = "btrfs",
1119         .mount          = btrfs_mount,
1120         .kill_sb        = kill_anon_super,
1121         .fs_flags       = FS_REQUIRES_DEV,
1122 };
1123
1124 /*
1125  * used by btrfsctl to scan devices when no FS is mounted
1126  */
1127 static long btrfs_control_ioctl(struct file *file, unsigned int cmd,
1128                                 unsigned long arg)
1129 {
1130         struct btrfs_ioctl_vol_args *vol;
1131         struct btrfs_fs_devices *fs_devices;
1132         int ret = -ENOTTY;
1133
1134         if (!capable(CAP_SYS_ADMIN))
1135                 return -EPERM;
1136
1137         vol = memdup_user((void __user *)arg, sizeof(*vol));
1138         if (IS_ERR(vol))
1139                 return PTR_ERR(vol);
1140
1141         switch (cmd) {
1142         case BTRFS_IOC_SCAN_DEV:
1143                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
1144                                             &btrfs_fs_type, &fs_devices);
1145                 break;
1146         }
1147
1148         kfree(vol);
1149         return ret;
1150 }
1151
1152 static int btrfs_freeze(struct super_block *sb)
1153 {
1154         struct btrfs_root *root = btrfs_sb(sb);
1155         mutex_lock(&root->fs_info->transaction_kthread_mutex);
1156         mutex_lock(&root->fs_info->cleaner_mutex);
1157         return 0;
1158 }
1159
1160 static int btrfs_unfreeze(struct super_block *sb)
1161 {
1162         struct btrfs_root *root = btrfs_sb(sb);
1163         mutex_unlock(&root->fs_info->cleaner_mutex);
1164         mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1165         return 0;
1166 }
1167
1168 static const struct super_operations btrfs_super_ops = {
1169         .drop_inode     = btrfs_drop_inode,
1170         .evict_inode    = btrfs_evict_inode,
1171         .put_super      = btrfs_put_super,
1172         .sync_fs        = btrfs_sync_fs,
1173         .show_options   = btrfs_show_options,
1174         .write_inode    = btrfs_write_inode,
1175         .dirty_inode    = btrfs_dirty_inode,
1176         .alloc_inode    = btrfs_alloc_inode,
1177         .destroy_inode  = btrfs_destroy_inode,
1178         .statfs         = btrfs_statfs,
1179         .remount_fs     = btrfs_remount,
1180         .freeze_fs      = btrfs_freeze,
1181         .unfreeze_fs    = btrfs_unfreeze,
1182 };
1183
1184 static const struct file_operations btrfs_ctl_fops = {
1185         .unlocked_ioctl  = btrfs_control_ioctl,
1186         .compat_ioctl = btrfs_control_ioctl,
1187         .owner   = THIS_MODULE,
1188         .llseek = noop_llseek,
1189 };
1190
1191 static struct miscdevice btrfs_misc = {
1192         .minor          = BTRFS_MINOR,
1193         .name           = "btrfs-control",
1194         .fops           = &btrfs_ctl_fops
1195 };
1196
1197 MODULE_ALIAS_MISCDEV(BTRFS_MINOR);
1198 MODULE_ALIAS("devname:btrfs-control");
1199
1200 static int btrfs_interface_init(void)
1201 {
1202         return misc_register(&btrfs_misc);
1203 }
1204
1205 static void btrfs_interface_exit(void)
1206 {
1207         if (misc_deregister(&btrfs_misc) < 0)
1208                 printk(KERN_INFO "misc_deregister failed for control device");
1209 }
1210
1211 static int __init init_btrfs_fs(void)
1212 {
1213         int err;
1214
1215         err = btrfs_init_sysfs();
1216         if (err)
1217                 return err;
1218
1219         err = btrfs_init_compress();
1220         if (err)
1221                 goto free_sysfs;
1222
1223         err = btrfs_init_cachep();
1224         if (err)
1225                 goto free_compress;
1226
1227         err = extent_io_init();
1228         if (err)
1229                 goto free_cachep;
1230
1231         err = extent_map_init();
1232         if (err)
1233                 goto free_extent_io;
1234
1235         err = btrfs_interface_init();
1236         if (err)
1237                 goto free_extent_map;
1238
1239         err = register_filesystem(&btrfs_fs_type);
1240         if (err)
1241                 goto unregister_ioctl;
1242
1243         printk(KERN_INFO "%s loaded\n", BTRFS_BUILD_VERSION);
1244         return 0;
1245
1246 unregister_ioctl:
1247         btrfs_interface_exit();
1248 free_extent_map:
1249         extent_map_exit();
1250 free_extent_io:
1251         extent_io_exit();
1252 free_cachep:
1253         btrfs_destroy_cachep();
1254 free_compress:
1255         btrfs_exit_compress();
1256 free_sysfs:
1257         btrfs_exit_sysfs();
1258         return err;
1259 }
1260
1261 static void __exit exit_btrfs_fs(void)
1262 {
1263         btrfs_destroy_cachep();
1264         extent_map_exit();
1265         extent_io_exit();
1266         btrfs_interface_exit();
1267         unregister_filesystem(&btrfs_fs_type);
1268         btrfs_exit_sysfs();
1269         btrfs_cleanup_fs_uuids();
1270         btrfs_exit_compress();
1271 }
1272
1273 module_init(init_btrfs_fs)
1274 module_exit(exit_btrfs_fs)
1275
1276 MODULE_LICENSE("GPL");