more conservative S_NOSEC handling
[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 <linux/cleancache.h>
43 #include "compat.h"
44 #include "delayed-inode.h"
45 #include "ctree.h"
46 #include "disk-io.h"
47 #include "transaction.h"
48 #include "btrfs_inode.h"
49 #include "ioctl.h"
50 #include "print-tree.h"
51 #include "xattr.h"
52 #include "volumes.h"
53 #include "version.h"
54 #include "export.h"
55 #include "compression.h"
56
57 #define CREATE_TRACE_POINTS
58 #include <trace/events/btrfs.h>
59
60 static const struct super_operations btrfs_super_ops;
61
62 static const char *btrfs_decode_error(struct btrfs_fs_info *fs_info, int errno,
63                                       char nbuf[16])
64 {
65         char *errstr = NULL;
66
67         switch (errno) {
68         case -EIO:
69                 errstr = "IO failure";
70                 break;
71         case -ENOMEM:
72                 errstr = "Out of memory";
73                 break;
74         case -EROFS:
75                 errstr = "Readonly filesystem";
76                 break;
77         default:
78                 if (nbuf) {
79                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
80                                 errstr = nbuf;
81                 }
82                 break;
83         }
84
85         return errstr;
86 }
87
88 static void __save_error_info(struct btrfs_fs_info *fs_info)
89 {
90         /*
91          * today we only save the error info into ram.  Long term we'll
92          * also send it down to the disk
93          */
94         fs_info->fs_state = BTRFS_SUPER_FLAG_ERROR;
95 }
96
97 /* NOTE:
98  *      We move write_super stuff at umount in order to avoid deadlock
99  *      for umount hold all lock.
100  */
101 static void save_error_info(struct btrfs_fs_info *fs_info)
102 {
103         __save_error_info(fs_info);
104 }
105
106 /* btrfs handle error by forcing the filesystem readonly */
107 static void btrfs_handle_error(struct btrfs_fs_info *fs_info)
108 {
109         struct super_block *sb = fs_info->sb;
110
111         if (sb->s_flags & MS_RDONLY)
112                 return;
113
114         if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
115                 sb->s_flags |= MS_RDONLY;
116                 printk(KERN_INFO "btrfs is forced readonly\n");
117         }
118 }
119
120 /*
121  * __btrfs_std_error decodes expected errors from the caller and
122  * invokes the approciate error response.
123  */
124 void __btrfs_std_error(struct btrfs_fs_info *fs_info, const char *function,
125                      unsigned int line, int errno)
126 {
127         struct super_block *sb = fs_info->sb;
128         char nbuf[16];
129         const char *errstr;
130
131         /*
132          * Special case: if the error is EROFS, and we're already
133          * under MS_RDONLY, then it is safe here.
134          */
135         if (errno == -EROFS && (sb->s_flags & MS_RDONLY))
136                 return;
137
138         errstr = btrfs_decode_error(fs_info, errno, nbuf);
139         printk(KERN_CRIT "BTRFS error (device %s) in %s:%d: %s\n",
140                 sb->s_id, function, line, errstr);
141         save_error_info(fs_info);
142
143         btrfs_handle_error(fs_info);
144 }
145
146 static void btrfs_put_super(struct super_block *sb)
147 {
148         struct btrfs_root *root = btrfs_sb(sb);
149         int ret;
150
151         ret = close_ctree(root);
152         sb->s_fs_info = NULL;
153
154         (void)ret; /* FIXME: need to fix VFS to return error? */
155 }
156
157 enum {
158         Opt_degraded, Opt_subvol, Opt_subvolid, Opt_device, Opt_nodatasum,
159         Opt_nodatacow, Opt_max_inline, Opt_alloc_start, Opt_nobarrier, Opt_ssd,
160         Opt_nossd, Opt_ssd_spread, Opt_thread_pool, Opt_noacl, Opt_compress,
161         Opt_compress_type, Opt_compress_force, Opt_compress_force_type,
162         Opt_notreelog, Opt_ratio, Opt_flushoncommit, Opt_discard,
163         Opt_space_cache, Opt_clear_cache, Opt_user_subvol_rm_allowed,
164         Opt_enospc_debug, Opt_subvolrootid, Opt_defrag, Opt_err,
165 };
166
167 static match_table_t tokens = {
168         {Opt_degraded, "degraded"},
169         {Opt_subvol, "subvol=%s"},
170         {Opt_subvolid, "subvolid=%d"},
171         {Opt_device, "device=%s"},
172         {Opt_nodatasum, "nodatasum"},
173         {Opt_nodatacow, "nodatacow"},
174         {Opt_nobarrier, "nobarrier"},
175         {Opt_max_inline, "max_inline=%s"},
176         {Opt_alloc_start, "alloc_start=%s"},
177         {Opt_thread_pool, "thread_pool=%d"},
178         {Opt_compress, "compress"},
179         {Opt_compress_type, "compress=%s"},
180         {Opt_compress_force, "compress-force"},
181         {Opt_compress_force_type, "compress-force=%s"},
182         {Opt_ssd, "ssd"},
183         {Opt_ssd_spread, "ssd_spread"},
184         {Opt_nossd, "nossd"},
185         {Opt_noacl, "noacl"},
186         {Opt_notreelog, "notreelog"},
187         {Opt_flushoncommit, "flushoncommit"},
188         {Opt_ratio, "metadata_ratio=%d"},
189         {Opt_discard, "discard"},
190         {Opt_space_cache, "space_cache"},
191         {Opt_clear_cache, "clear_cache"},
192         {Opt_user_subvol_rm_allowed, "user_subvol_rm_allowed"},
193         {Opt_enospc_debug, "enospc_debug"},
194         {Opt_subvolrootid, "subvolrootid=%d"},
195         {Opt_defrag, "autodefrag"},
196         {Opt_err, NULL},
197 };
198
199 /*
200  * Regular mount options parser.  Everything that is needed only when
201  * reading in a new superblock is parsed here.
202  */
203 int btrfs_parse_options(struct btrfs_root *root, char *options)
204 {
205         struct btrfs_fs_info *info = root->fs_info;
206         substring_t args[MAX_OPT_ARGS];
207         char *p, *num, *orig;
208         int intarg;
209         int ret = 0;
210         char *compress_type;
211         bool compress_force = false;
212
213         if (!options)
214                 return 0;
215
216         /*
217          * strsep changes the string, duplicate it because parse_options
218          * gets called twice
219          */
220         options = kstrdup(options, GFP_NOFS);
221         if (!options)
222                 return -ENOMEM;
223
224         orig = options;
225
226         while ((p = strsep(&options, ",")) != NULL) {
227                 int token;
228                 if (!*p)
229                         continue;
230
231                 token = match_token(p, tokens, args);
232                 switch (token) {
233                 case Opt_degraded:
234                         printk(KERN_INFO "btrfs: allowing degraded mounts\n");
235                         btrfs_set_opt(info->mount_opt, DEGRADED);
236                         break;
237                 case Opt_subvol:
238                 case Opt_subvolid:
239                 case Opt_subvolrootid:
240                 case Opt_device:
241                         /*
242                          * These are parsed by btrfs_parse_early_options
243                          * and can be happily ignored here.
244                          */
245                         break;
246                 case Opt_nodatasum:
247                         printk(KERN_INFO "btrfs: setting nodatasum\n");
248                         btrfs_set_opt(info->mount_opt, NODATASUM);
249                         break;
250                 case Opt_nodatacow:
251                         printk(KERN_INFO "btrfs: setting nodatacow\n");
252                         btrfs_set_opt(info->mount_opt, NODATACOW);
253                         btrfs_set_opt(info->mount_opt, NODATASUM);
254                         break;
255                 case Opt_compress_force:
256                 case Opt_compress_force_type:
257                         compress_force = true;
258                 case Opt_compress:
259                 case Opt_compress_type:
260                         if (token == Opt_compress ||
261                             token == Opt_compress_force ||
262                             strcmp(args[0].from, "zlib") == 0) {
263                                 compress_type = "zlib";
264                                 info->compress_type = BTRFS_COMPRESS_ZLIB;
265                         } else if (strcmp(args[0].from, "lzo") == 0) {
266                                 compress_type = "lzo";
267                                 info->compress_type = BTRFS_COMPRESS_LZO;
268                         } else {
269                                 ret = -EINVAL;
270                                 goto out;
271                         }
272
273                         btrfs_set_opt(info->mount_opt, COMPRESS);
274                         if (compress_force) {
275                                 btrfs_set_opt(info->mount_opt, FORCE_COMPRESS);
276                                 pr_info("btrfs: force %s compression\n",
277                                         compress_type);
278                         } else
279                                 pr_info("btrfs: use %s compression\n",
280                                         compress_type);
281                         break;
282                 case Opt_ssd:
283                         printk(KERN_INFO "btrfs: use ssd allocation scheme\n");
284                         btrfs_set_opt(info->mount_opt, SSD);
285                         break;
286                 case Opt_ssd_spread:
287                         printk(KERN_INFO "btrfs: use spread ssd "
288                                "allocation scheme\n");
289                         btrfs_set_opt(info->mount_opt, SSD);
290                         btrfs_set_opt(info->mount_opt, SSD_SPREAD);
291                         break;
292                 case Opt_nossd:
293                         printk(KERN_INFO "btrfs: not using ssd allocation "
294                                "scheme\n");
295                         btrfs_set_opt(info->mount_opt, NOSSD);
296                         btrfs_clear_opt(info->mount_opt, SSD);
297                         btrfs_clear_opt(info->mount_opt, SSD_SPREAD);
298                         break;
299                 case Opt_nobarrier:
300                         printk(KERN_INFO "btrfs: turning off barriers\n");
301                         btrfs_set_opt(info->mount_opt, NOBARRIER);
302                         break;
303                 case Opt_thread_pool:
304                         intarg = 0;
305                         match_int(&args[0], &intarg);
306                         if (intarg) {
307                                 info->thread_pool_size = intarg;
308                                 printk(KERN_INFO "btrfs: thread pool %d\n",
309                                        info->thread_pool_size);
310                         }
311                         break;
312                 case Opt_max_inline:
313                         num = match_strdup(&args[0]);
314                         if (num) {
315                                 info->max_inline = memparse(num, NULL);
316                                 kfree(num);
317
318                                 if (info->max_inline) {
319                                         info->max_inline = max_t(u64,
320                                                 info->max_inline,
321                                                 root->sectorsize);
322                                 }
323                                 printk(KERN_INFO "btrfs: max_inline at %llu\n",
324                                         (unsigned long long)info->max_inline);
325                         }
326                         break;
327                 case Opt_alloc_start:
328                         num = match_strdup(&args[0]);
329                         if (num) {
330                                 info->alloc_start = memparse(num, NULL);
331                                 kfree(num);
332                                 printk(KERN_INFO
333                                         "btrfs: allocations start at %llu\n",
334                                         (unsigned long long)info->alloc_start);
335                         }
336                         break;
337                 case Opt_noacl:
338                         root->fs_info->sb->s_flags &= ~MS_POSIXACL;
339                         break;
340                 case Opt_notreelog:
341                         printk(KERN_INFO "btrfs: disabling tree log\n");
342                         btrfs_set_opt(info->mount_opt, NOTREELOG);
343                         break;
344                 case Opt_flushoncommit:
345                         printk(KERN_INFO "btrfs: turning on flush-on-commit\n");
346                         btrfs_set_opt(info->mount_opt, FLUSHONCOMMIT);
347                         break;
348                 case Opt_ratio:
349                         intarg = 0;
350                         match_int(&args[0], &intarg);
351                         if (intarg) {
352                                 info->metadata_ratio = intarg;
353                                 printk(KERN_INFO "btrfs: metadata ratio %d\n",
354                                        info->metadata_ratio);
355                         }
356                         break;
357                 case Opt_discard:
358                         btrfs_set_opt(info->mount_opt, DISCARD);
359                         break;
360                 case Opt_space_cache:
361                         printk(KERN_INFO "btrfs: enabling disk space caching\n");
362                         btrfs_set_opt(info->mount_opt, SPACE_CACHE);
363                         break;
364                 case Opt_clear_cache:
365                         printk(KERN_INFO "btrfs: force clearing of disk cache\n");
366                         btrfs_set_opt(info->mount_opt, CLEAR_CACHE);
367                         break;
368                 case Opt_user_subvol_rm_allowed:
369                         btrfs_set_opt(info->mount_opt, USER_SUBVOL_RM_ALLOWED);
370                         break;
371                 case Opt_enospc_debug:
372                         btrfs_set_opt(info->mount_opt, ENOSPC_DEBUG);
373                         break;
374                 case Opt_defrag:
375                         printk(KERN_INFO "btrfs: enabling auto defrag");
376                         btrfs_set_opt(info->mount_opt, AUTO_DEFRAG);
377                         break;
378                 case Opt_err:
379                         printk(KERN_INFO "btrfs: unrecognized mount option "
380                                "'%s'\n", p);
381                         ret = -EINVAL;
382                         goto out;
383                 default:
384                         break;
385                 }
386         }
387 out:
388         kfree(orig);
389         return ret;
390 }
391
392 /*
393  * Parse mount options that are required early in the mount process.
394  *
395  * All other options will be parsed on much later in the mount process and
396  * only when we need to allocate a new super block.
397  */
398 static int btrfs_parse_early_options(const char *options, fmode_t flags,
399                 void *holder, char **subvol_name, u64 *subvol_objectid,
400                 u64 *subvol_rootid, struct btrfs_fs_devices **fs_devices)
401 {
402         substring_t args[MAX_OPT_ARGS];
403         char *opts, *orig, *p;
404         int error = 0;
405         int intarg;
406
407         if (!options)
408                 goto out;
409
410         /*
411          * strsep changes the string, duplicate it because parse_options
412          * gets called twice
413          */
414         opts = kstrdup(options, GFP_KERNEL);
415         if (!opts)
416                 return -ENOMEM;
417         orig = opts;
418
419         while ((p = strsep(&opts, ",")) != NULL) {
420                 int token;
421                 if (!*p)
422                         continue;
423
424                 token = match_token(p, tokens, args);
425                 switch (token) {
426                 case Opt_subvol:
427                         *subvol_name = match_strdup(&args[0]);
428                         break;
429                 case Opt_subvolid:
430                         intarg = 0;
431                         error = match_int(&args[0], &intarg);
432                         if (!error) {
433                                 /* we want the original fs_tree */
434                                 if (!intarg)
435                                         *subvol_objectid =
436                                                 BTRFS_FS_TREE_OBJECTID;
437                                 else
438                                         *subvol_objectid = intarg;
439                         }
440                         break;
441                 case Opt_subvolrootid:
442                         intarg = 0;
443                         error = match_int(&args[0], &intarg);
444                         if (!error) {
445                                 /* we want the original fs_tree */
446                                 if (!intarg)
447                                         *subvol_rootid =
448                                                 BTRFS_FS_TREE_OBJECTID;
449                                 else
450                                         *subvol_rootid = intarg;
451                         }
452                         break;
453                 case Opt_device:
454                         error = btrfs_scan_one_device(match_strdup(&args[0]),
455                                         flags, holder, fs_devices);
456                         if (error)
457                                 goto out_free_opts;
458                         break;
459                 default:
460                         break;
461                 }
462         }
463
464  out_free_opts:
465         kfree(orig);
466  out:
467         /*
468          * If no subvolume name is specified we use the default one.  Allocate
469          * a copy of the string "." here so that code later in the
470          * mount path doesn't care if it's the default volume or another one.
471          */
472         if (!*subvol_name) {
473                 *subvol_name = kstrdup(".", GFP_KERNEL);
474                 if (!*subvol_name)
475                         return -ENOMEM;
476         }
477         return error;
478 }
479
480 static struct dentry *get_default_root(struct super_block *sb,
481                                        u64 subvol_objectid)
482 {
483         struct btrfs_root *root = sb->s_fs_info;
484         struct btrfs_root *new_root;
485         struct btrfs_dir_item *di;
486         struct btrfs_path *path;
487         struct btrfs_key location;
488         struct inode *inode;
489         struct dentry *dentry;
490         u64 dir_id;
491         int new = 0;
492
493         /*
494          * We have a specific subvol we want to mount, just setup location and
495          * go look up the root.
496          */
497         if (subvol_objectid) {
498                 location.objectid = subvol_objectid;
499                 location.type = BTRFS_ROOT_ITEM_KEY;
500                 location.offset = (u64)-1;
501                 goto find_root;
502         }
503
504         path = btrfs_alloc_path();
505         if (!path)
506                 return ERR_PTR(-ENOMEM);
507         path->leave_spinning = 1;
508
509         /*
510          * Find the "default" dir item which points to the root item that we
511          * will mount by default if we haven't been given a specific subvolume
512          * to mount.
513          */
514         dir_id = btrfs_super_root_dir(&root->fs_info->super_copy);
515         di = btrfs_lookup_dir_item(NULL, root, path, dir_id, "default", 7, 0);
516         if (IS_ERR(di)) {
517                 btrfs_free_path(path);
518                 return ERR_CAST(di);
519         }
520         if (!di) {
521                 /*
522                  * Ok the default dir item isn't there.  This is weird since
523                  * it's always been there, but don't freak out, just try and
524                  * mount to root most subvolume.
525                  */
526                 btrfs_free_path(path);
527                 dir_id = BTRFS_FIRST_FREE_OBJECTID;
528                 new_root = root->fs_info->fs_root;
529                 goto setup_root;
530         }
531
532         btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
533         btrfs_free_path(path);
534
535 find_root:
536         new_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
537         if (IS_ERR(new_root))
538                 return ERR_CAST(new_root);
539
540         if (btrfs_root_refs(&new_root->root_item) == 0)
541                 return ERR_PTR(-ENOENT);
542
543         dir_id = btrfs_root_dirid(&new_root->root_item);
544 setup_root:
545         location.objectid = dir_id;
546         location.type = BTRFS_INODE_ITEM_KEY;
547         location.offset = 0;
548
549         inode = btrfs_iget(sb, &location, new_root, &new);
550         if (IS_ERR(inode))
551                 return ERR_CAST(inode);
552
553         /*
554          * If we're just mounting the root most subvol put the inode and return
555          * a reference to the dentry.  We will have already gotten a reference
556          * to the inode in btrfs_fill_super so we're good to go.
557          */
558         if (!new && sb->s_root->d_inode == inode) {
559                 iput(inode);
560                 return dget(sb->s_root);
561         }
562
563         if (new) {
564                 const struct qstr name = { .name = "/", .len = 1 };
565
566                 /*
567                  * New inode, we need to make the dentry a sibling of s_root so
568                  * everything gets cleaned up properly on unmount.
569                  */
570                 dentry = d_alloc(sb->s_root, &name);
571                 if (!dentry) {
572                         iput(inode);
573                         return ERR_PTR(-ENOMEM);
574                 }
575                 d_splice_alias(inode, dentry);
576         } else {
577                 /*
578                  * We found the inode in cache, just find a dentry for it and
579                  * put the reference to the inode we just got.
580                  */
581                 dentry = d_find_alias(inode);
582                 iput(inode);
583         }
584
585         return dentry;
586 }
587
588 static int btrfs_fill_super(struct super_block *sb,
589                             struct btrfs_fs_devices *fs_devices,
590                             void *data, int silent)
591 {
592         struct inode *inode;
593         struct dentry *root_dentry;
594         struct btrfs_root *tree_root;
595         struct btrfs_key key;
596         int err;
597
598         sb->s_maxbytes = MAX_LFS_FILESIZE;
599         sb->s_magic = BTRFS_SUPER_MAGIC;
600         sb->s_op = &btrfs_super_ops;
601         sb->s_d_op = &btrfs_dentry_operations;
602         sb->s_export_op = &btrfs_export_ops;
603         sb->s_xattr = btrfs_xattr_handlers;
604         sb->s_time_gran = 1;
605 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
606         sb->s_flags |= MS_POSIXACL;
607 #endif
608
609         tree_root = open_ctree(sb, fs_devices, (char *)data);
610
611         if (IS_ERR(tree_root)) {
612                 printk("btrfs: open_ctree failed\n");
613                 return PTR_ERR(tree_root);
614         }
615         sb->s_fs_info = tree_root;
616
617         key.objectid = BTRFS_FIRST_FREE_OBJECTID;
618         key.type = BTRFS_INODE_ITEM_KEY;
619         key.offset = 0;
620         inode = btrfs_iget(sb, &key, tree_root->fs_info->fs_root, NULL);
621         if (IS_ERR(inode)) {
622                 err = PTR_ERR(inode);
623                 goto fail_close;
624         }
625
626         root_dentry = d_alloc_root(inode);
627         if (!root_dentry) {
628                 iput(inode);
629                 err = -ENOMEM;
630                 goto fail_close;
631         }
632
633         sb->s_root = root_dentry;
634
635         save_mount_options(sb, data);
636         cleancache_init_fs(sb);
637         return 0;
638
639 fail_close:
640         close_ctree(tree_root);
641         return err;
642 }
643
644 int btrfs_sync_fs(struct super_block *sb, int wait)
645 {
646         struct btrfs_trans_handle *trans;
647         struct btrfs_root *root = btrfs_sb(sb);
648         int ret;
649
650         trace_btrfs_sync_fs(wait);
651
652         if (!wait) {
653                 filemap_flush(root->fs_info->btree_inode->i_mapping);
654                 return 0;
655         }
656
657         btrfs_start_delalloc_inodes(root, 0);
658         btrfs_wait_ordered_extents(root, 0, 0);
659
660         trans = btrfs_start_transaction(root, 0);
661         if (IS_ERR(trans))
662                 return PTR_ERR(trans);
663         ret = btrfs_commit_transaction(trans, root);
664         return ret;
665 }
666
667 static int btrfs_show_options(struct seq_file *seq, struct vfsmount *vfs)
668 {
669         struct btrfs_root *root = btrfs_sb(vfs->mnt_sb);
670         struct btrfs_fs_info *info = root->fs_info;
671         char *compress_type;
672
673         if (btrfs_test_opt(root, DEGRADED))
674                 seq_puts(seq, ",degraded");
675         if (btrfs_test_opt(root, NODATASUM))
676                 seq_puts(seq, ",nodatasum");
677         if (btrfs_test_opt(root, NODATACOW))
678                 seq_puts(seq, ",nodatacow");
679         if (btrfs_test_opt(root, NOBARRIER))
680                 seq_puts(seq, ",nobarrier");
681         if (info->max_inline != 8192 * 1024)
682                 seq_printf(seq, ",max_inline=%llu",
683                            (unsigned long long)info->max_inline);
684         if (info->alloc_start != 0)
685                 seq_printf(seq, ",alloc_start=%llu",
686                            (unsigned long long)info->alloc_start);
687         if (info->thread_pool_size !=  min_t(unsigned long,
688                                              num_online_cpus() + 2, 8))
689                 seq_printf(seq, ",thread_pool=%d", info->thread_pool_size);
690         if (btrfs_test_opt(root, COMPRESS)) {
691                 if (info->compress_type == BTRFS_COMPRESS_ZLIB)
692                         compress_type = "zlib";
693                 else
694                         compress_type = "lzo";
695                 if (btrfs_test_opt(root, FORCE_COMPRESS))
696                         seq_printf(seq, ",compress-force=%s", compress_type);
697                 else
698                         seq_printf(seq, ",compress=%s", compress_type);
699         }
700         if (btrfs_test_opt(root, NOSSD))
701                 seq_puts(seq, ",nossd");
702         if (btrfs_test_opt(root, SSD_SPREAD))
703                 seq_puts(seq, ",ssd_spread");
704         else if (btrfs_test_opt(root, SSD))
705                 seq_puts(seq, ",ssd");
706         if (btrfs_test_opt(root, NOTREELOG))
707                 seq_puts(seq, ",notreelog");
708         if (btrfs_test_opt(root, FLUSHONCOMMIT))
709                 seq_puts(seq, ",flushoncommit");
710         if (btrfs_test_opt(root, DISCARD))
711                 seq_puts(seq, ",discard");
712         if (!(root->fs_info->sb->s_flags & MS_POSIXACL))
713                 seq_puts(seq, ",noacl");
714         if (btrfs_test_opt(root, SPACE_CACHE))
715                 seq_puts(seq, ",space_cache");
716         if (btrfs_test_opt(root, CLEAR_CACHE))
717                 seq_puts(seq, ",clear_cache");
718         if (btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
719                 seq_puts(seq, ",user_subvol_rm_allowed");
720         return 0;
721 }
722
723 static int btrfs_test_super(struct super_block *s, void *data)
724 {
725         struct btrfs_root *test_root = data;
726         struct btrfs_root *root = btrfs_sb(s);
727
728         /*
729          * If this super block is going away, return false as it
730          * can't match as an existing super block.
731          */
732         if (!atomic_read(&s->s_active))
733                 return 0;
734         return root->fs_info->fs_devices == test_root->fs_info->fs_devices;
735 }
736
737 static int btrfs_set_super(struct super_block *s, void *data)
738 {
739         s->s_fs_info = data;
740
741         return set_anon_super(s, data);
742 }
743
744
745 /*
746  * Find a superblock for the given device / mount point.
747  *
748  * Note:  This is based on get_sb_bdev from fs/super.c with a few additions
749  *        for multiple device setup.  Make sure to keep it in sync.
750  */
751 static struct dentry *btrfs_mount(struct file_system_type *fs_type, int flags,
752                 const char *device_name, void *data)
753 {
754         struct block_device *bdev = NULL;
755         struct super_block *s;
756         struct dentry *root;
757         struct btrfs_fs_devices *fs_devices = NULL;
758         struct btrfs_root *tree_root = NULL;
759         struct btrfs_fs_info *fs_info = NULL;
760         fmode_t mode = FMODE_READ;
761         char *subvol_name = NULL;
762         u64 subvol_objectid = 0;
763         u64 subvol_rootid = 0;
764         int error = 0;
765
766         if (!(flags & MS_RDONLY))
767                 mode |= FMODE_WRITE;
768
769         error = btrfs_parse_early_options(data, mode, fs_type,
770                                           &subvol_name, &subvol_objectid,
771                                           &subvol_rootid, &fs_devices);
772         if (error)
773                 return ERR_PTR(error);
774
775         error = btrfs_scan_one_device(device_name, mode, fs_type, &fs_devices);
776         if (error)
777                 goto error_free_subvol_name;
778
779         error = btrfs_open_devices(fs_devices, mode, fs_type);
780         if (error)
781                 goto error_free_subvol_name;
782
783         if (!(flags & MS_RDONLY) && fs_devices->rw_devices == 0) {
784                 error = -EACCES;
785                 goto error_close_devices;
786         }
787
788         /*
789          * Setup a dummy root and fs_info for test/set super.  This is because
790          * we don't actually fill this stuff out until open_ctree, but we need
791          * it for searching for existing supers, so this lets us do that and
792          * then open_ctree will properly initialize everything later.
793          */
794         fs_info = kzalloc(sizeof(struct btrfs_fs_info), GFP_NOFS);
795         tree_root = kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
796         if (!fs_info || !tree_root) {
797                 error = -ENOMEM;
798                 goto error_close_devices;
799         }
800         fs_info->tree_root = tree_root;
801         fs_info->fs_devices = fs_devices;
802         tree_root->fs_info = fs_info;
803
804         bdev = fs_devices->latest_bdev;
805         s = sget(fs_type, btrfs_test_super, btrfs_set_super, tree_root);
806         if (IS_ERR(s))
807                 goto error_s;
808
809         if (s->s_root) {
810                 if ((flags ^ s->s_flags) & MS_RDONLY) {
811                         deactivate_locked_super(s);
812                         error = -EBUSY;
813                         goto error_close_devices;
814                 }
815
816                 btrfs_close_devices(fs_devices);
817                 kfree(fs_info);
818                 kfree(tree_root);
819         } else {
820                 char b[BDEVNAME_SIZE];
821
822                 s->s_flags = flags | MS_NOSEC;
823                 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
824                 error = btrfs_fill_super(s, fs_devices, data,
825                                          flags & MS_SILENT ? 1 : 0);
826                 if (error) {
827                         deactivate_locked_super(s);
828                         goto error_free_subvol_name;
829                 }
830
831                 btrfs_sb(s)->fs_info->bdev_holder = fs_type;
832                 s->s_flags |= MS_ACTIVE;
833         }
834
835         /* if they gave us a subvolume name bind mount into that */
836         if (strcmp(subvol_name, ".")) {
837                 struct dentry *new_root;
838
839                 root = get_default_root(s, subvol_rootid);
840                 if (IS_ERR(root)) {
841                         error = PTR_ERR(root);
842                         deactivate_locked_super(s);
843                         goto error_free_subvol_name;
844                 }
845
846                 mutex_lock(&root->d_inode->i_mutex);
847                 new_root = lookup_one_len(subvol_name, root,
848                                       strlen(subvol_name));
849                 mutex_unlock(&root->d_inode->i_mutex);
850
851                 if (IS_ERR(new_root)) {
852                         dput(root);
853                         deactivate_locked_super(s);
854                         error = PTR_ERR(new_root);
855                         goto error_free_subvol_name;
856                 }
857                 if (!new_root->d_inode) {
858                         dput(root);
859                         dput(new_root);
860                         deactivate_locked_super(s);
861                         error = -ENXIO;
862                         goto error_free_subvol_name;
863                 }
864                 dput(root);
865                 root = new_root;
866         } else {
867                 root = get_default_root(s, subvol_objectid);
868                 if (IS_ERR(root)) {
869                         error = PTR_ERR(root);
870                         deactivate_locked_super(s);
871                         goto error_free_subvol_name;
872                 }
873         }
874
875         kfree(subvol_name);
876         return root;
877
878 error_s:
879         error = PTR_ERR(s);
880 error_close_devices:
881         btrfs_close_devices(fs_devices);
882         kfree(fs_info);
883         kfree(tree_root);
884 error_free_subvol_name:
885         kfree(subvol_name);
886         return ERR_PTR(error);
887 }
888
889 static int btrfs_remount(struct super_block *sb, int *flags, char *data)
890 {
891         struct btrfs_root *root = btrfs_sb(sb);
892         int ret;
893
894         ret = btrfs_parse_options(root, data);
895         if (ret)
896                 return -EINVAL;
897
898         if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
899                 return 0;
900
901         if (*flags & MS_RDONLY) {
902                 sb->s_flags |= MS_RDONLY;
903
904                 ret =  btrfs_commit_super(root);
905                 WARN_ON(ret);
906         } else {
907                 if (root->fs_info->fs_devices->rw_devices == 0)
908                         return -EACCES;
909
910                 if (btrfs_super_log_root(&root->fs_info->super_copy) != 0)
911                         return -EINVAL;
912
913                 ret = btrfs_cleanup_fs_roots(root->fs_info);
914                 WARN_ON(ret);
915
916                 /* recover relocation */
917                 ret = btrfs_recover_relocation(root);
918                 WARN_ON(ret);
919
920                 sb->s_flags &= ~MS_RDONLY;
921         }
922
923         return 0;
924 }
925
926 /* Used to sort the devices by max_avail(descending sort) */
927 static int btrfs_cmp_device_free_bytes(const void *dev_info1,
928                                        const void *dev_info2)
929 {
930         if (((struct btrfs_device_info *)dev_info1)->max_avail >
931             ((struct btrfs_device_info *)dev_info2)->max_avail)
932                 return -1;
933         else if (((struct btrfs_device_info *)dev_info1)->max_avail <
934                  ((struct btrfs_device_info *)dev_info2)->max_avail)
935                 return 1;
936         else
937         return 0;
938 }
939
940 /*
941  * sort the devices by max_avail, in which max free extent size of each device
942  * is stored.(Descending Sort)
943  */
944 static inline void btrfs_descending_sort_devices(
945                                         struct btrfs_device_info *devices,
946                                         size_t nr_devices)
947 {
948         sort(devices, nr_devices, sizeof(struct btrfs_device_info),
949              btrfs_cmp_device_free_bytes, NULL);
950 }
951
952 /*
953  * The helper to calc the free space on the devices that can be used to store
954  * file data.
955  */
956 static int btrfs_calc_avail_data_space(struct btrfs_root *root, u64 *free_bytes)
957 {
958         struct btrfs_fs_info *fs_info = root->fs_info;
959         struct btrfs_device_info *devices_info;
960         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
961         struct btrfs_device *device;
962         u64 skip_space;
963         u64 type;
964         u64 avail_space;
965         u64 used_space;
966         u64 min_stripe_size;
967         int min_stripes = 1;
968         int i = 0, nr_devices;
969         int ret;
970
971         nr_devices = fs_info->fs_devices->rw_devices;
972         BUG_ON(!nr_devices);
973
974         devices_info = kmalloc(sizeof(*devices_info) * nr_devices,
975                                GFP_NOFS);
976         if (!devices_info)
977                 return -ENOMEM;
978
979         /* calc min stripe number for data space alloction */
980         type = btrfs_get_alloc_profile(root, 1);
981         if (type & BTRFS_BLOCK_GROUP_RAID0)
982                 min_stripes = 2;
983         else if (type & BTRFS_BLOCK_GROUP_RAID1)
984                 min_stripes = 2;
985         else if (type & BTRFS_BLOCK_GROUP_RAID10)
986                 min_stripes = 4;
987
988         if (type & BTRFS_BLOCK_GROUP_DUP)
989                 min_stripe_size = 2 * BTRFS_STRIPE_LEN;
990         else
991                 min_stripe_size = BTRFS_STRIPE_LEN;
992
993         list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
994                 if (!device->in_fs_metadata)
995                         continue;
996
997                 avail_space = device->total_bytes - device->bytes_used;
998
999                 /* align with stripe_len */
1000                 do_div(avail_space, BTRFS_STRIPE_LEN);
1001                 avail_space *= BTRFS_STRIPE_LEN;
1002
1003                 /*
1004                  * In order to avoid overwritting the superblock on the drive,
1005                  * btrfs starts at an offset of at least 1MB when doing chunk
1006                  * allocation.
1007                  */
1008                 skip_space = 1024 * 1024;
1009
1010                 /* user can set the offset in fs_info->alloc_start. */
1011                 if (fs_info->alloc_start + BTRFS_STRIPE_LEN <=
1012                     device->total_bytes)
1013                         skip_space = max(fs_info->alloc_start, skip_space);
1014
1015                 /*
1016                  * btrfs can not use the free space in [0, skip_space - 1],
1017                  * we must subtract it from the total. In order to implement
1018                  * it, we account the used space in this range first.
1019                  */
1020                 ret = btrfs_account_dev_extents_size(device, 0, skip_space - 1,
1021                                                      &used_space);
1022                 if (ret) {
1023                         kfree(devices_info);
1024                         return ret;
1025                 }
1026
1027                 /* calc the free space in [0, skip_space - 1] */
1028                 skip_space -= used_space;
1029
1030                 /*
1031                  * we can use the free space in [0, skip_space - 1], subtract
1032                  * it from the total.
1033                  */
1034                 if (avail_space && avail_space >= skip_space)
1035                         avail_space -= skip_space;
1036                 else
1037                         avail_space = 0;
1038
1039                 if (avail_space < min_stripe_size)
1040                         continue;
1041
1042                 devices_info[i].dev = device;
1043                 devices_info[i].max_avail = avail_space;
1044
1045                 i++;
1046         }
1047
1048         nr_devices = i;
1049
1050         btrfs_descending_sort_devices(devices_info, nr_devices);
1051
1052         i = nr_devices - 1;
1053         avail_space = 0;
1054         while (nr_devices >= min_stripes) {
1055                 if (devices_info[i].max_avail >= min_stripe_size) {
1056                         int j;
1057                         u64 alloc_size;
1058
1059                         avail_space += devices_info[i].max_avail * min_stripes;
1060                         alloc_size = devices_info[i].max_avail;
1061                         for (j = i + 1 - min_stripes; j <= i; j++)
1062                                 devices_info[j].max_avail -= alloc_size;
1063                 }
1064                 i--;
1065                 nr_devices--;
1066         }
1067
1068         kfree(devices_info);
1069         *free_bytes = avail_space;
1070         return 0;
1071 }
1072
1073 static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
1074 {
1075         struct btrfs_root *root = btrfs_sb(dentry->d_sb);
1076         struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
1077         struct list_head *head = &root->fs_info->space_info;
1078         struct btrfs_space_info *found;
1079         u64 total_used = 0;
1080         u64 total_free_data = 0;
1081         int bits = dentry->d_sb->s_blocksize_bits;
1082         __be32 *fsid = (__be32 *)root->fs_info->fsid;
1083         int ret;
1084
1085         /* holding chunk_muext to avoid allocating new chunks */
1086         mutex_lock(&root->fs_info->chunk_mutex);
1087         rcu_read_lock();
1088         list_for_each_entry_rcu(found, head, list) {
1089                 if (found->flags & BTRFS_BLOCK_GROUP_DATA) {
1090                         total_free_data += found->disk_total - found->disk_used;
1091                         total_free_data -=
1092                                 btrfs_account_ro_block_groups_free_space(found);
1093                 }
1094
1095                 total_used += found->disk_used;
1096         }
1097         rcu_read_unlock();
1098
1099         buf->f_namelen = BTRFS_NAME_LEN;
1100         buf->f_blocks = btrfs_super_total_bytes(disk_super) >> bits;
1101         buf->f_bfree = buf->f_blocks - (total_used >> bits);
1102         buf->f_bsize = dentry->d_sb->s_blocksize;
1103         buf->f_type = BTRFS_SUPER_MAGIC;
1104         buf->f_bavail = total_free_data;
1105         ret = btrfs_calc_avail_data_space(root, &total_free_data);
1106         if (ret) {
1107                 mutex_unlock(&root->fs_info->chunk_mutex);
1108                 return ret;
1109         }
1110         buf->f_bavail += total_free_data;
1111         buf->f_bavail = buf->f_bavail >> bits;
1112         mutex_unlock(&root->fs_info->chunk_mutex);
1113
1114         /* We treat it as constant endianness (it doesn't matter _which_)
1115            because we want the fsid to come out the same whether mounted
1116            on a big-endian or little-endian host */
1117         buf->f_fsid.val[0] = be32_to_cpu(fsid[0]) ^ be32_to_cpu(fsid[2]);
1118         buf->f_fsid.val[1] = be32_to_cpu(fsid[1]) ^ be32_to_cpu(fsid[3]);
1119         /* Mask in the root object ID too, to disambiguate subvols */
1120         buf->f_fsid.val[0] ^= BTRFS_I(dentry->d_inode)->root->objectid >> 32;
1121         buf->f_fsid.val[1] ^= BTRFS_I(dentry->d_inode)->root->objectid;
1122
1123         return 0;
1124 }
1125
1126 static struct file_system_type btrfs_fs_type = {
1127         .owner          = THIS_MODULE,
1128         .name           = "btrfs",
1129         .mount          = btrfs_mount,
1130         .kill_sb        = kill_anon_super,
1131         .fs_flags       = FS_REQUIRES_DEV,
1132 };
1133
1134 /*
1135  * used by btrfsctl to scan devices when no FS is mounted
1136  */
1137 static long btrfs_control_ioctl(struct file *file, unsigned int cmd,
1138                                 unsigned long arg)
1139 {
1140         struct btrfs_ioctl_vol_args *vol;
1141         struct btrfs_fs_devices *fs_devices;
1142         int ret = -ENOTTY;
1143
1144         if (!capable(CAP_SYS_ADMIN))
1145                 return -EPERM;
1146
1147         vol = memdup_user((void __user *)arg, sizeof(*vol));
1148         if (IS_ERR(vol))
1149                 return PTR_ERR(vol);
1150
1151         switch (cmd) {
1152         case BTRFS_IOC_SCAN_DEV:
1153                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
1154                                             &btrfs_fs_type, &fs_devices);
1155                 break;
1156         }
1157
1158         kfree(vol);
1159         return ret;
1160 }
1161
1162 static int btrfs_freeze(struct super_block *sb)
1163 {
1164         struct btrfs_root *root = btrfs_sb(sb);
1165         mutex_lock(&root->fs_info->transaction_kthread_mutex);
1166         mutex_lock(&root->fs_info->cleaner_mutex);
1167         return 0;
1168 }
1169
1170 static int btrfs_unfreeze(struct super_block *sb)
1171 {
1172         struct btrfs_root *root = btrfs_sb(sb);
1173         mutex_unlock(&root->fs_info->cleaner_mutex);
1174         mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1175         return 0;
1176 }
1177
1178 static const struct super_operations btrfs_super_ops = {
1179         .drop_inode     = btrfs_drop_inode,
1180         .evict_inode    = btrfs_evict_inode,
1181         .put_super      = btrfs_put_super,
1182         .sync_fs        = btrfs_sync_fs,
1183         .show_options   = btrfs_show_options,
1184         .write_inode    = btrfs_write_inode,
1185         .dirty_inode    = btrfs_dirty_inode,
1186         .alloc_inode    = btrfs_alloc_inode,
1187         .destroy_inode  = btrfs_destroy_inode,
1188         .statfs         = btrfs_statfs,
1189         .remount_fs     = btrfs_remount,
1190         .freeze_fs      = btrfs_freeze,
1191         .unfreeze_fs    = btrfs_unfreeze,
1192 };
1193
1194 static const struct file_operations btrfs_ctl_fops = {
1195         .unlocked_ioctl  = btrfs_control_ioctl,
1196         .compat_ioctl = btrfs_control_ioctl,
1197         .owner   = THIS_MODULE,
1198         .llseek = noop_llseek,
1199 };
1200
1201 static struct miscdevice btrfs_misc = {
1202         .minor          = BTRFS_MINOR,
1203         .name           = "btrfs-control",
1204         .fops           = &btrfs_ctl_fops
1205 };
1206
1207 MODULE_ALIAS_MISCDEV(BTRFS_MINOR);
1208 MODULE_ALIAS("devname:btrfs-control");
1209
1210 static int btrfs_interface_init(void)
1211 {
1212         return misc_register(&btrfs_misc);
1213 }
1214
1215 static void btrfs_interface_exit(void)
1216 {
1217         if (misc_deregister(&btrfs_misc) < 0)
1218                 printk(KERN_INFO "misc_deregister failed for control device");
1219 }
1220
1221 static int __init init_btrfs_fs(void)
1222 {
1223         int err;
1224
1225         err = btrfs_init_sysfs();
1226         if (err)
1227                 return err;
1228
1229         err = btrfs_init_compress();
1230         if (err)
1231                 goto free_sysfs;
1232
1233         err = btrfs_init_cachep();
1234         if (err)
1235                 goto free_compress;
1236
1237         err = extent_io_init();
1238         if (err)
1239                 goto free_cachep;
1240
1241         err = extent_map_init();
1242         if (err)
1243                 goto free_extent_io;
1244
1245         err = btrfs_delayed_inode_init();
1246         if (err)
1247                 goto free_extent_map;
1248
1249         err = btrfs_interface_init();
1250         if (err)
1251                 goto free_delayed_inode;
1252
1253         err = register_filesystem(&btrfs_fs_type);
1254         if (err)
1255                 goto unregister_ioctl;
1256
1257         printk(KERN_INFO "%s loaded\n", BTRFS_BUILD_VERSION);
1258         return 0;
1259
1260 unregister_ioctl:
1261         btrfs_interface_exit();
1262 free_delayed_inode:
1263         btrfs_delayed_inode_exit();
1264 free_extent_map:
1265         extent_map_exit();
1266 free_extent_io:
1267         extent_io_exit();
1268 free_cachep:
1269         btrfs_destroy_cachep();
1270 free_compress:
1271         btrfs_exit_compress();
1272 free_sysfs:
1273         btrfs_exit_sysfs();
1274         return err;
1275 }
1276
1277 static void __exit exit_btrfs_fs(void)
1278 {
1279         btrfs_destroy_cachep();
1280         btrfs_delayed_inode_exit();
1281         extent_map_exit();
1282         extent_io_exit();
1283         btrfs_interface_exit();
1284         unregister_filesystem(&btrfs_fs_type);
1285         btrfs_exit_sysfs();
1286         btrfs_cleanup_fs_uuids();
1287         btrfs_exit_compress();
1288 }
1289
1290 module_init(init_btrfs_fs)
1291 module_exit(exit_btrfs_fs)
1292
1293 MODULE_LICENSE("GPL");