Btrfs: stop using write_one_page
[pandora-kernel.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 <linux/mnt_namespace.h>
44 #include "compat.h"
45 #include "delayed-inode.h"
46 #include "ctree.h"
47 #include "disk-io.h"
48 #include "transaction.h"
49 #include "btrfs_inode.h"
50 #include "ioctl.h"
51 #include "print-tree.h"
52 #include "xattr.h"
53 #include "volumes.h"
54 #include "version.h"
55 #include "export.h"
56 #include "compression.h"
57
58 #define CREATE_TRACE_POINTS
59 #include <trace/events/btrfs.h>
60
61 static const struct super_operations btrfs_super_ops;
62 static struct file_system_type btrfs_fs_type;
63
64 static const char *btrfs_decode_error(struct btrfs_fs_info *fs_info, int errno,
65                                       char nbuf[16])
66 {
67         char *errstr = NULL;
68
69         switch (errno) {
70         case -EIO:
71                 errstr = "IO failure";
72                 break;
73         case -ENOMEM:
74                 errstr = "Out of memory";
75                 break;
76         case -EROFS:
77                 errstr = "Readonly filesystem";
78                 break;
79         default:
80                 if (nbuf) {
81                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
82                                 errstr = nbuf;
83                 }
84                 break;
85         }
86
87         return errstr;
88 }
89
90 static void __save_error_info(struct btrfs_fs_info *fs_info)
91 {
92         /*
93          * today we only save the error info into ram.  Long term we'll
94          * also send it down to the disk
95          */
96         fs_info->fs_state = BTRFS_SUPER_FLAG_ERROR;
97 }
98
99 /* NOTE:
100  *      We move write_super stuff at umount in order to avoid deadlock
101  *      for umount hold all lock.
102  */
103 static void save_error_info(struct btrfs_fs_info *fs_info)
104 {
105         __save_error_info(fs_info);
106 }
107
108 /* btrfs handle error by forcing the filesystem readonly */
109 static void btrfs_handle_error(struct btrfs_fs_info *fs_info)
110 {
111         struct super_block *sb = fs_info->sb;
112
113         if (sb->s_flags & MS_RDONLY)
114                 return;
115
116         if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
117                 sb->s_flags |= MS_RDONLY;
118                 printk(KERN_INFO "btrfs is forced readonly\n");
119         }
120 }
121
122 /*
123  * __btrfs_std_error decodes expected errors from the caller and
124  * invokes the approciate error response.
125  */
126 void __btrfs_std_error(struct btrfs_fs_info *fs_info, const char *function,
127                      unsigned int line, int errno)
128 {
129         struct super_block *sb = fs_info->sb;
130         char nbuf[16];
131         const char *errstr;
132
133         /*
134          * Special case: if the error is EROFS, and we're already
135          * under MS_RDONLY, then it is safe here.
136          */
137         if (errno == -EROFS && (sb->s_flags & MS_RDONLY))
138                 return;
139
140         errstr = btrfs_decode_error(fs_info, errno, nbuf);
141         printk(KERN_CRIT "BTRFS error (device %s) in %s:%d: %s\n",
142                 sb->s_id, function, line, errstr);
143         save_error_info(fs_info);
144
145         btrfs_handle_error(fs_info);
146 }
147
148 static void btrfs_put_super(struct super_block *sb)
149 {
150         struct btrfs_root *root = btrfs_sb(sb);
151         int ret;
152
153         ret = close_ctree(root);
154         sb->s_fs_info = NULL;
155
156         (void)ret; /* FIXME: need to fix VFS to return error? */
157 }
158
159 enum {
160         Opt_degraded, Opt_subvol, Opt_subvolid, Opt_device, Opt_nodatasum,
161         Opt_nodatacow, Opt_max_inline, Opt_alloc_start, Opt_nobarrier, Opt_ssd,
162         Opt_nossd, Opt_ssd_spread, Opt_thread_pool, Opt_noacl, Opt_compress,
163         Opt_compress_type, Opt_compress_force, Opt_compress_force_type,
164         Opt_notreelog, Opt_ratio, Opt_flushoncommit, Opt_discard,
165         Opt_space_cache, Opt_clear_cache, Opt_user_subvol_rm_allowed,
166         Opt_enospc_debug, Opt_subvolrootid, Opt_defrag,
167         Opt_inode_cache, Opt_err,
168 };
169
170 static match_table_t tokens = {
171         {Opt_degraded, "degraded"},
172         {Opt_subvol, "subvol=%s"},
173         {Opt_subvolid, "subvolid=%d"},
174         {Opt_device, "device=%s"},
175         {Opt_nodatasum, "nodatasum"},
176         {Opt_nodatacow, "nodatacow"},
177         {Opt_nobarrier, "nobarrier"},
178         {Opt_max_inline, "max_inline=%s"},
179         {Opt_alloc_start, "alloc_start=%s"},
180         {Opt_thread_pool, "thread_pool=%d"},
181         {Opt_compress, "compress"},
182         {Opt_compress_type, "compress=%s"},
183         {Opt_compress_force, "compress-force"},
184         {Opt_compress_force_type, "compress-force=%s"},
185         {Opt_ssd, "ssd"},
186         {Opt_ssd_spread, "ssd_spread"},
187         {Opt_nossd, "nossd"},
188         {Opt_noacl, "noacl"},
189         {Opt_notreelog, "notreelog"},
190         {Opt_flushoncommit, "flushoncommit"},
191         {Opt_ratio, "metadata_ratio=%d"},
192         {Opt_discard, "discard"},
193         {Opt_space_cache, "space_cache"},
194         {Opt_clear_cache, "clear_cache"},
195         {Opt_user_subvol_rm_allowed, "user_subvol_rm_allowed"},
196         {Opt_enospc_debug, "enospc_debug"},
197         {Opt_subvolrootid, "subvolrootid=%d"},
198         {Opt_defrag, "autodefrag"},
199         {Opt_inode_cache, "inode_cache"},
200         {Opt_err, NULL},
201 };
202
203 /*
204  * Regular mount options parser.  Everything that is needed only when
205  * reading in a new superblock is parsed here.
206  */
207 int btrfs_parse_options(struct btrfs_root *root, char *options)
208 {
209         struct btrfs_fs_info *info = root->fs_info;
210         substring_t args[MAX_OPT_ARGS];
211         char *p, *num, *orig;
212         int intarg;
213         int ret = 0;
214         char *compress_type;
215         bool compress_force = false;
216
217         if (!options)
218                 return 0;
219
220         /*
221          * strsep changes the string, duplicate it because parse_options
222          * gets called twice
223          */
224         options = kstrdup(options, GFP_NOFS);
225         if (!options)
226                 return -ENOMEM;
227
228         orig = options;
229
230         while ((p = strsep(&options, ",")) != NULL) {
231                 int token;
232                 if (!*p)
233                         continue;
234
235                 token = match_token(p, tokens, args);
236                 switch (token) {
237                 case Opt_degraded:
238                         printk(KERN_INFO "btrfs: allowing degraded mounts\n");
239                         btrfs_set_opt(info->mount_opt, DEGRADED);
240                         break;
241                 case Opt_subvol:
242                 case Opt_subvolid:
243                 case Opt_subvolrootid:
244                 case Opt_device:
245                         /*
246                          * These are parsed by btrfs_parse_early_options
247                          * and can be happily ignored here.
248                          */
249                         break;
250                 case Opt_nodatasum:
251                         printk(KERN_INFO "btrfs: setting nodatasum\n");
252                         btrfs_set_opt(info->mount_opt, NODATASUM);
253                         break;
254                 case Opt_nodatacow:
255                         printk(KERN_INFO "btrfs: setting nodatacow\n");
256                         btrfs_set_opt(info->mount_opt, NODATACOW);
257                         btrfs_set_opt(info->mount_opt, NODATASUM);
258                         break;
259                 case Opt_compress_force:
260                 case Opt_compress_force_type:
261                         compress_force = true;
262                 case Opt_compress:
263                 case Opt_compress_type:
264                         if (token == Opt_compress ||
265                             token == Opt_compress_force ||
266                             strcmp(args[0].from, "zlib") == 0) {
267                                 compress_type = "zlib";
268                                 info->compress_type = BTRFS_COMPRESS_ZLIB;
269                         } else if (strcmp(args[0].from, "lzo") == 0) {
270                                 compress_type = "lzo";
271                                 info->compress_type = BTRFS_COMPRESS_LZO;
272                         } else {
273                                 ret = -EINVAL;
274                                 goto out;
275                         }
276
277                         btrfs_set_opt(info->mount_opt, COMPRESS);
278                         if (compress_force) {
279                                 btrfs_set_opt(info->mount_opt, FORCE_COMPRESS);
280                                 pr_info("btrfs: force %s compression\n",
281                                         compress_type);
282                         } else
283                                 pr_info("btrfs: use %s compression\n",
284                                         compress_type);
285                         break;
286                 case Opt_ssd:
287                         printk(KERN_INFO "btrfs: use ssd allocation scheme\n");
288                         btrfs_set_opt(info->mount_opt, SSD);
289                         break;
290                 case Opt_ssd_spread:
291                         printk(KERN_INFO "btrfs: use spread ssd "
292                                "allocation scheme\n");
293                         btrfs_set_opt(info->mount_opt, SSD);
294                         btrfs_set_opt(info->mount_opt, SSD_SPREAD);
295                         break;
296                 case Opt_nossd:
297                         printk(KERN_INFO "btrfs: not using ssd allocation "
298                                "scheme\n");
299                         btrfs_set_opt(info->mount_opt, NOSSD);
300                         btrfs_clear_opt(info->mount_opt, SSD);
301                         btrfs_clear_opt(info->mount_opt, SSD_SPREAD);
302                         break;
303                 case Opt_nobarrier:
304                         printk(KERN_INFO "btrfs: turning off barriers\n");
305                         btrfs_set_opt(info->mount_opt, NOBARRIER);
306                         break;
307                 case Opt_thread_pool:
308                         intarg = 0;
309                         match_int(&args[0], &intarg);
310                         if (intarg) {
311                                 info->thread_pool_size = intarg;
312                                 printk(KERN_INFO "btrfs: thread pool %d\n",
313                                        info->thread_pool_size);
314                         }
315                         break;
316                 case Opt_max_inline:
317                         num = match_strdup(&args[0]);
318                         if (num) {
319                                 info->max_inline = memparse(num, NULL);
320                                 kfree(num);
321
322                                 if (info->max_inline) {
323                                         info->max_inline = max_t(u64,
324                                                 info->max_inline,
325                                                 root->sectorsize);
326                                 }
327                                 printk(KERN_INFO "btrfs: max_inline at %llu\n",
328                                         (unsigned long long)info->max_inline);
329                         }
330                         break;
331                 case Opt_alloc_start:
332                         num = match_strdup(&args[0]);
333                         if (num) {
334                                 info->alloc_start = memparse(num, NULL);
335                                 kfree(num);
336                                 printk(KERN_INFO
337                                         "btrfs: allocations start at %llu\n",
338                                         (unsigned long long)info->alloc_start);
339                         }
340                         break;
341                 case Opt_noacl:
342                         root->fs_info->sb->s_flags &= ~MS_POSIXACL;
343                         break;
344                 case Opt_notreelog:
345                         printk(KERN_INFO "btrfs: disabling tree log\n");
346                         btrfs_set_opt(info->mount_opt, NOTREELOG);
347                         break;
348                 case Opt_flushoncommit:
349                         printk(KERN_INFO "btrfs: turning on flush-on-commit\n");
350                         btrfs_set_opt(info->mount_opt, FLUSHONCOMMIT);
351                         break;
352                 case Opt_ratio:
353                         intarg = 0;
354                         match_int(&args[0], &intarg);
355                         if (intarg) {
356                                 info->metadata_ratio = intarg;
357                                 printk(KERN_INFO "btrfs: metadata ratio %d\n",
358                                        info->metadata_ratio);
359                         }
360                         break;
361                 case Opt_discard:
362                         btrfs_set_opt(info->mount_opt, DISCARD);
363                         break;
364                 case Opt_space_cache:
365                         printk(KERN_INFO "btrfs: enabling disk space caching\n");
366                         btrfs_set_opt(info->mount_opt, SPACE_CACHE);
367                         break;
368                 case Opt_inode_cache:
369                         printk(KERN_INFO "btrfs: enabling inode map caching\n");
370                         btrfs_set_opt(info->mount_opt, INODE_MAP_CACHE);
371                         break;
372                 case Opt_clear_cache:
373                         printk(KERN_INFO "btrfs: force clearing of disk cache\n");
374                         btrfs_set_opt(info->mount_opt, CLEAR_CACHE);
375                         break;
376                 case Opt_user_subvol_rm_allowed:
377                         btrfs_set_opt(info->mount_opt, USER_SUBVOL_RM_ALLOWED);
378                         break;
379                 case Opt_enospc_debug:
380                         btrfs_set_opt(info->mount_opt, ENOSPC_DEBUG);
381                         break;
382                 case Opt_defrag:
383                         printk(KERN_INFO "btrfs: enabling auto defrag");
384                         btrfs_set_opt(info->mount_opt, AUTO_DEFRAG);
385                         break;
386                 case Opt_err:
387                         printk(KERN_INFO "btrfs: unrecognized mount option "
388                                "'%s'\n", p);
389                         ret = -EINVAL;
390                         goto out;
391                 default:
392                         break;
393                 }
394         }
395 out:
396         kfree(orig);
397         return ret;
398 }
399
400 /*
401  * Parse mount options that are required early in the mount process.
402  *
403  * All other options will be parsed on much later in the mount process and
404  * only when we need to allocate a new super block.
405  */
406 static int btrfs_parse_early_options(const char *options, fmode_t flags,
407                 void *holder, char **subvol_name, u64 *subvol_objectid,
408                 u64 *subvol_rootid, struct btrfs_fs_devices **fs_devices)
409 {
410         substring_t args[MAX_OPT_ARGS];
411         char *opts, *orig, *p;
412         int error = 0;
413         int intarg;
414
415         if (!options)
416                 return 0;
417
418         /*
419          * strsep changes the string, duplicate it because parse_options
420          * gets called twice
421          */
422         opts = kstrdup(options, GFP_KERNEL);
423         if (!opts)
424                 return -ENOMEM;
425         orig = opts;
426
427         while ((p = strsep(&opts, ",")) != NULL) {
428                 int token;
429                 if (!*p)
430                         continue;
431
432                 token = match_token(p, tokens, args);
433                 switch (token) {
434                 case Opt_subvol:
435                         *subvol_name = match_strdup(&args[0]);
436                         break;
437                 case Opt_subvolid:
438                         intarg = 0;
439                         error = match_int(&args[0], &intarg);
440                         if (!error) {
441                                 /* we want the original fs_tree */
442                                 if (!intarg)
443                                         *subvol_objectid =
444                                                 BTRFS_FS_TREE_OBJECTID;
445                                 else
446                                         *subvol_objectid = intarg;
447                         }
448                         break;
449                 case Opt_subvolrootid:
450                         intarg = 0;
451                         error = match_int(&args[0], &intarg);
452                         if (!error) {
453                                 /* we want the original fs_tree */
454                                 if (!intarg)
455                                         *subvol_rootid =
456                                                 BTRFS_FS_TREE_OBJECTID;
457                                 else
458                                         *subvol_rootid = intarg;
459                         }
460                         break;
461                 case Opt_device:
462                         error = btrfs_scan_one_device(match_strdup(&args[0]),
463                                         flags, holder, fs_devices);
464                         if (error)
465                                 goto out;
466                         break;
467                 default:
468                         break;
469                 }
470         }
471
472 out:
473         kfree(orig);
474         return error;
475 }
476
477 static struct dentry *get_default_root(struct super_block *sb,
478                                        u64 subvol_objectid)
479 {
480         struct btrfs_root *root = sb->s_fs_info;
481         struct btrfs_root *new_root;
482         struct btrfs_dir_item *di;
483         struct btrfs_path *path;
484         struct btrfs_key location;
485         struct inode *inode;
486         u64 dir_id;
487         int new = 0;
488
489         /*
490          * We have a specific subvol we want to mount, just setup location and
491          * go look up the root.
492          */
493         if (subvol_objectid) {
494                 location.objectid = subvol_objectid;
495                 location.type = BTRFS_ROOT_ITEM_KEY;
496                 location.offset = (u64)-1;
497                 goto find_root;
498         }
499
500         path = btrfs_alloc_path();
501         if (!path)
502                 return ERR_PTR(-ENOMEM);
503         path->leave_spinning = 1;
504
505         /*
506          * Find the "default" dir item which points to the root item that we
507          * will mount by default if we haven't been given a specific subvolume
508          * to mount.
509          */
510         dir_id = btrfs_super_root_dir(&root->fs_info->super_copy);
511         di = btrfs_lookup_dir_item(NULL, root, path, dir_id, "default", 7, 0);
512         if (IS_ERR(di)) {
513                 btrfs_free_path(path);
514                 return ERR_CAST(di);
515         }
516         if (!di) {
517                 /*
518                  * Ok the default dir item isn't there.  This is weird since
519                  * it's always been there, but don't freak out, just try and
520                  * mount to root most subvolume.
521                  */
522                 btrfs_free_path(path);
523                 dir_id = BTRFS_FIRST_FREE_OBJECTID;
524                 new_root = root->fs_info->fs_root;
525                 goto setup_root;
526         }
527
528         btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
529         btrfs_free_path(path);
530
531 find_root:
532         new_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
533         if (IS_ERR(new_root))
534                 return ERR_CAST(new_root);
535
536         if (btrfs_root_refs(&new_root->root_item) == 0)
537                 return ERR_PTR(-ENOENT);
538
539         dir_id = btrfs_root_dirid(&new_root->root_item);
540 setup_root:
541         location.objectid = dir_id;
542         location.type = BTRFS_INODE_ITEM_KEY;
543         location.offset = 0;
544
545         inode = btrfs_iget(sb, &location, new_root, &new);
546         if (IS_ERR(inode))
547                 return ERR_CAST(inode);
548
549         /*
550          * If we're just mounting the root most subvol put the inode and return
551          * a reference to the dentry.  We will have already gotten a reference
552          * to the inode in btrfs_fill_super so we're good to go.
553          */
554         if (!new && sb->s_root->d_inode == inode) {
555                 iput(inode);
556                 return dget(sb->s_root);
557         }
558
559         return d_obtain_alias(inode);
560 }
561
562 static int btrfs_fill_super(struct super_block *sb,
563                             struct btrfs_fs_devices *fs_devices,
564                             void *data, int silent)
565 {
566         struct inode *inode;
567         struct dentry *root_dentry;
568         struct btrfs_root *tree_root;
569         struct btrfs_key key;
570         int err;
571
572         sb->s_maxbytes = MAX_LFS_FILESIZE;
573         sb->s_magic = BTRFS_SUPER_MAGIC;
574         sb->s_op = &btrfs_super_ops;
575         sb->s_d_op = &btrfs_dentry_operations;
576         sb->s_export_op = &btrfs_export_ops;
577         sb->s_xattr = btrfs_xattr_handlers;
578         sb->s_time_gran = 1;
579 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
580         sb->s_flags |= MS_POSIXACL;
581 #endif
582
583         tree_root = open_ctree(sb, fs_devices, (char *)data);
584
585         if (IS_ERR(tree_root)) {
586                 printk("btrfs: open_ctree failed\n");
587                 return PTR_ERR(tree_root);
588         }
589         sb->s_fs_info = tree_root;
590
591         key.objectid = BTRFS_FIRST_FREE_OBJECTID;
592         key.type = BTRFS_INODE_ITEM_KEY;
593         key.offset = 0;
594         inode = btrfs_iget(sb, &key, tree_root->fs_info->fs_root, NULL);
595         if (IS_ERR(inode)) {
596                 err = PTR_ERR(inode);
597                 goto fail_close;
598         }
599
600         root_dentry = d_alloc_root(inode);
601         if (!root_dentry) {
602                 iput(inode);
603                 err = -ENOMEM;
604                 goto fail_close;
605         }
606
607         sb->s_root = root_dentry;
608
609         save_mount_options(sb, data);
610         cleancache_init_fs(sb);
611         return 0;
612
613 fail_close:
614         close_ctree(tree_root);
615         return err;
616 }
617
618 int btrfs_sync_fs(struct super_block *sb, int wait)
619 {
620         struct btrfs_trans_handle *trans;
621         struct btrfs_root *root = btrfs_sb(sb);
622         int ret;
623
624         trace_btrfs_sync_fs(wait);
625
626         if (!wait) {
627                 filemap_flush(root->fs_info->btree_inode->i_mapping);
628                 return 0;
629         }
630
631         btrfs_start_delalloc_inodes(root, 0);
632         btrfs_wait_ordered_extents(root, 0, 0);
633
634         trans = btrfs_start_transaction(root, 0);
635         if (IS_ERR(trans))
636                 return PTR_ERR(trans);
637         ret = btrfs_commit_transaction(trans, root);
638         return ret;
639 }
640
641 static int btrfs_show_options(struct seq_file *seq, struct vfsmount *vfs)
642 {
643         struct btrfs_root *root = btrfs_sb(vfs->mnt_sb);
644         struct btrfs_fs_info *info = root->fs_info;
645         char *compress_type;
646
647         if (btrfs_test_opt(root, DEGRADED))
648                 seq_puts(seq, ",degraded");
649         if (btrfs_test_opt(root, NODATASUM))
650                 seq_puts(seq, ",nodatasum");
651         if (btrfs_test_opt(root, NODATACOW))
652                 seq_puts(seq, ",nodatacow");
653         if (btrfs_test_opt(root, NOBARRIER))
654                 seq_puts(seq, ",nobarrier");
655         if (info->max_inline != 8192 * 1024)
656                 seq_printf(seq, ",max_inline=%llu",
657                            (unsigned long long)info->max_inline);
658         if (info->alloc_start != 0)
659                 seq_printf(seq, ",alloc_start=%llu",
660                            (unsigned long long)info->alloc_start);
661         if (info->thread_pool_size !=  min_t(unsigned long,
662                                              num_online_cpus() + 2, 8))
663                 seq_printf(seq, ",thread_pool=%d", info->thread_pool_size);
664         if (btrfs_test_opt(root, COMPRESS)) {
665                 if (info->compress_type == BTRFS_COMPRESS_ZLIB)
666                         compress_type = "zlib";
667                 else
668                         compress_type = "lzo";
669                 if (btrfs_test_opt(root, FORCE_COMPRESS))
670                         seq_printf(seq, ",compress-force=%s", compress_type);
671                 else
672                         seq_printf(seq, ",compress=%s", compress_type);
673         }
674         if (btrfs_test_opt(root, NOSSD))
675                 seq_puts(seq, ",nossd");
676         if (btrfs_test_opt(root, SSD_SPREAD))
677                 seq_puts(seq, ",ssd_spread");
678         else if (btrfs_test_opt(root, SSD))
679                 seq_puts(seq, ",ssd");
680         if (btrfs_test_opt(root, NOTREELOG))
681                 seq_puts(seq, ",notreelog");
682         if (btrfs_test_opt(root, FLUSHONCOMMIT))
683                 seq_puts(seq, ",flushoncommit");
684         if (btrfs_test_opt(root, DISCARD))
685                 seq_puts(seq, ",discard");
686         if (!(root->fs_info->sb->s_flags & MS_POSIXACL))
687                 seq_puts(seq, ",noacl");
688         if (btrfs_test_opt(root, SPACE_CACHE))
689                 seq_puts(seq, ",space_cache");
690         if (btrfs_test_opt(root, CLEAR_CACHE))
691                 seq_puts(seq, ",clear_cache");
692         if (btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
693                 seq_puts(seq, ",user_subvol_rm_allowed");
694         if (btrfs_test_opt(root, ENOSPC_DEBUG))
695                 seq_puts(seq, ",enospc_debug");
696         if (btrfs_test_opt(root, AUTO_DEFRAG))
697                 seq_puts(seq, ",autodefrag");
698         if (btrfs_test_opt(root, INODE_MAP_CACHE))
699                 seq_puts(seq, ",inode_cache");
700         return 0;
701 }
702
703 static int btrfs_test_super(struct super_block *s, void *data)
704 {
705         struct btrfs_root *test_root = data;
706         struct btrfs_root *root = btrfs_sb(s);
707
708         /*
709          * If this super block is going away, return false as it
710          * can't match as an existing super block.
711          */
712         if (!atomic_read(&s->s_active))
713                 return 0;
714         return root->fs_info->fs_devices == test_root->fs_info->fs_devices;
715 }
716
717 static int btrfs_set_super(struct super_block *s, void *data)
718 {
719         s->s_fs_info = data;
720
721         return set_anon_super(s, data);
722 }
723
724 /*
725  * This will strip out the subvol=%s argument for an argument string and add
726  * subvolid=0 to make sure we get the actual tree root for path walking to the
727  * subvol we want.
728  */
729 static char *setup_root_args(char *args)
730 {
731         unsigned copied = 0;
732         unsigned len = strlen(args) + 2;
733         char *pos;
734         char *ret;
735
736         /*
737          * We need the same args as before, but minus
738          *
739          * subvol=a
740          *
741          * and add
742          *
743          * subvolid=0
744          *
745          * which is a difference of 2 characters, so we allocate strlen(args) +
746          * 2 characters.
747          */
748         ret = kzalloc(len * sizeof(char), GFP_NOFS);
749         if (!ret)
750                 return NULL;
751         pos = strstr(args, "subvol=");
752
753         /* This shouldn't happen, but just in case.. */
754         if (!pos) {
755                 kfree(ret);
756                 return NULL;
757         }
758
759         /*
760          * The subvol=<> arg is not at the front of the string, copy everybody
761          * up to that into ret.
762          */
763         if (pos != args) {
764                 *pos = '\0';
765                 strcpy(ret, args);
766                 copied += strlen(args);
767                 pos++;
768         }
769
770         strncpy(ret + copied, "subvolid=0", len - copied);
771
772         /* Length of subvolid=0 */
773         copied += 10;
774
775         /*
776          * If there is no , after the subvol= option then we know there's no
777          * other options and we can just return.
778          */
779         pos = strchr(pos, ',');
780         if (!pos)
781                 return ret;
782
783         /* Copy the rest of the arguments into our buffer */
784         strncpy(ret + copied, pos, len - copied);
785         copied += strlen(pos);
786
787         return ret;
788 }
789
790 static struct dentry *mount_subvol(const char *subvol_name, int flags,
791                                    const char *device_name, char *data)
792 {
793         struct super_block *s;
794         struct dentry *root;
795         struct vfsmount *mnt;
796         struct mnt_namespace *ns_private;
797         char *newargs;
798         struct path path;
799         int error;
800
801         newargs = setup_root_args(data);
802         if (!newargs)
803                 return ERR_PTR(-ENOMEM);
804         mnt = vfs_kern_mount(&btrfs_fs_type, flags, device_name,
805                              newargs);
806         kfree(newargs);
807         if (IS_ERR(mnt))
808                 return ERR_CAST(mnt);
809
810         ns_private = create_mnt_ns(mnt);
811         if (IS_ERR(ns_private)) {
812                 mntput(mnt);
813                 return ERR_CAST(ns_private);
814         }
815
816         /*
817          * This will trigger the automount of the subvol so we can just
818          * drop the mnt we have here and return the dentry that we
819          * found.
820          */
821         error = vfs_path_lookup(mnt->mnt_root, mnt, subvol_name,
822                                 LOOKUP_FOLLOW, &path);
823         put_mnt_ns(ns_private);
824         if (error)
825                 return ERR_PTR(error);
826
827         /* Get a ref to the sb and the dentry we found and return it */
828         s = path.mnt->mnt_sb;
829         atomic_inc(&s->s_active);
830         root = dget(path.dentry);
831         path_put(&path);
832         down_write(&s->s_umount);
833
834         return root;
835 }
836
837 /*
838  * Find a superblock for the given device / mount point.
839  *
840  * Note:  This is based on get_sb_bdev from fs/super.c with a few additions
841  *        for multiple device setup.  Make sure to keep it in sync.
842  */
843 static struct dentry *btrfs_mount(struct file_system_type *fs_type, int flags,
844                 const char *device_name, void *data)
845 {
846         struct block_device *bdev = NULL;
847         struct super_block *s;
848         struct dentry *root;
849         struct btrfs_fs_devices *fs_devices = NULL;
850         struct btrfs_root *tree_root = NULL;
851         struct btrfs_fs_info *fs_info = NULL;
852         fmode_t mode = FMODE_READ;
853         char *subvol_name = NULL;
854         u64 subvol_objectid = 0;
855         u64 subvol_rootid = 0;
856         int error = 0;
857
858         if (!(flags & MS_RDONLY))
859                 mode |= FMODE_WRITE;
860
861         error = btrfs_parse_early_options(data, mode, fs_type,
862                                           &subvol_name, &subvol_objectid,
863                                           &subvol_rootid, &fs_devices);
864         if (error)
865                 return ERR_PTR(error);
866
867         if (subvol_name) {
868                 root = mount_subvol(subvol_name, flags, device_name, data);
869                 kfree(subvol_name);
870                 return root;
871         }
872
873         error = btrfs_scan_one_device(device_name, mode, fs_type, &fs_devices);
874         if (error)
875                 return ERR_PTR(error);
876
877         error = btrfs_open_devices(fs_devices, mode, fs_type);
878         if (error)
879                 return ERR_PTR(error);
880
881         if (!(flags & MS_RDONLY) && fs_devices->rw_devices == 0) {
882                 error = -EACCES;
883                 goto error_close_devices;
884         }
885
886         /*
887          * Setup a dummy root and fs_info for test/set super.  This is because
888          * we don't actually fill this stuff out until open_ctree, but we need
889          * it for searching for existing supers, so this lets us do that and
890          * then open_ctree will properly initialize everything later.
891          */
892         fs_info = kzalloc(sizeof(struct btrfs_fs_info), GFP_NOFS);
893         tree_root = kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
894         if (!fs_info || !tree_root) {
895                 error = -ENOMEM;
896                 goto error_close_devices;
897         }
898         fs_info->tree_root = tree_root;
899         fs_info->fs_devices = fs_devices;
900         tree_root->fs_info = fs_info;
901
902         bdev = fs_devices->latest_bdev;
903         s = sget(fs_type, btrfs_test_super, btrfs_set_super, tree_root);
904         if (IS_ERR(s)) {
905                 error = PTR_ERR(s);
906                 goto error_close_devices;
907         }
908
909         if (s->s_root) {
910                 if ((flags ^ s->s_flags) & MS_RDONLY) {
911                         deactivate_locked_super(s);
912                         return ERR_PTR(-EBUSY);
913                 }
914
915                 btrfs_close_devices(fs_devices);
916                 kfree(fs_info);
917                 kfree(tree_root);
918         } else {
919                 char b[BDEVNAME_SIZE];
920
921                 s->s_flags = flags | MS_NOSEC;
922                 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
923                 error = btrfs_fill_super(s, fs_devices, data,
924                                          flags & MS_SILENT ? 1 : 0);
925                 if (error) {
926                         deactivate_locked_super(s);
927                         return ERR_PTR(error);
928                 }
929
930                 btrfs_sb(s)->fs_info->bdev_holder = fs_type;
931                 s->s_flags |= MS_ACTIVE;
932         }
933
934         root = get_default_root(s, subvol_objectid);
935         if (IS_ERR(root)) {
936                 deactivate_locked_super(s);
937                 return root;
938         }
939
940         return root;
941
942 error_close_devices:
943         btrfs_close_devices(fs_devices);
944         kfree(fs_info);
945         kfree(tree_root);
946         return ERR_PTR(error);
947 }
948
949 static int btrfs_remount(struct super_block *sb, int *flags, char *data)
950 {
951         struct btrfs_root *root = btrfs_sb(sb);
952         int ret;
953
954         ret = btrfs_parse_options(root, data);
955         if (ret)
956                 return -EINVAL;
957
958         if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
959                 return 0;
960
961         if (*flags & MS_RDONLY) {
962                 sb->s_flags |= MS_RDONLY;
963
964                 ret =  btrfs_commit_super(root);
965                 WARN_ON(ret);
966         } else {
967                 if (root->fs_info->fs_devices->rw_devices == 0)
968                         return -EACCES;
969
970                 if (btrfs_super_log_root(&root->fs_info->super_copy) != 0)
971                         return -EINVAL;
972
973                 ret = btrfs_cleanup_fs_roots(root->fs_info);
974                 WARN_ON(ret);
975
976                 /* recover relocation */
977                 ret = btrfs_recover_relocation(root);
978                 WARN_ON(ret);
979
980                 sb->s_flags &= ~MS_RDONLY;
981         }
982
983         return 0;
984 }
985
986 /* Used to sort the devices by max_avail(descending sort) */
987 static int btrfs_cmp_device_free_bytes(const void *dev_info1,
988                                        const void *dev_info2)
989 {
990         if (((struct btrfs_device_info *)dev_info1)->max_avail >
991             ((struct btrfs_device_info *)dev_info2)->max_avail)
992                 return -1;
993         else if (((struct btrfs_device_info *)dev_info1)->max_avail <
994                  ((struct btrfs_device_info *)dev_info2)->max_avail)
995                 return 1;
996         else
997         return 0;
998 }
999
1000 /*
1001  * sort the devices by max_avail, in which max free extent size of each device
1002  * is stored.(Descending Sort)
1003  */
1004 static inline void btrfs_descending_sort_devices(
1005                                         struct btrfs_device_info *devices,
1006                                         size_t nr_devices)
1007 {
1008         sort(devices, nr_devices, sizeof(struct btrfs_device_info),
1009              btrfs_cmp_device_free_bytes, NULL);
1010 }
1011
1012 /*
1013  * The helper to calc the free space on the devices that can be used to store
1014  * file data.
1015  */
1016 static int btrfs_calc_avail_data_space(struct btrfs_root *root, u64 *free_bytes)
1017 {
1018         struct btrfs_fs_info *fs_info = root->fs_info;
1019         struct btrfs_device_info *devices_info;
1020         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
1021         struct btrfs_device *device;
1022         u64 skip_space;
1023         u64 type;
1024         u64 avail_space;
1025         u64 used_space;
1026         u64 min_stripe_size;
1027         int min_stripes = 1;
1028         int i = 0, nr_devices;
1029         int ret;
1030
1031         nr_devices = fs_info->fs_devices->rw_devices;
1032         BUG_ON(!nr_devices);
1033
1034         devices_info = kmalloc(sizeof(*devices_info) * nr_devices,
1035                                GFP_NOFS);
1036         if (!devices_info)
1037                 return -ENOMEM;
1038
1039         /* calc min stripe number for data space alloction */
1040         type = btrfs_get_alloc_profile(root, 1);
1041         if (type & BTRFS_BLOCK_GROUP_RAID0)
1042                 min_stripes = 2;
1043         else if (type & BTRFS_BLOCK_GROUP_RAID1)
1044                 min_stripes = 2;
1045         else if (type & BTRFS_BLOCK_GROUP_RAID10)
1046                 min_stripes = 4;
1047
1048         if (type & BTRFS_BLOCK_GROUP_DUP)
1049                 min_stripe_size = 2 * BTRFS_STRIPE_LEN;
1050         else
1051                 min_stripe_size = BTRFS_STRIPE_LEN;
1052
1053         list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
1054                 if (!device->in_fs_metadata)
1055                         continue;
1056
1057                 avail_space = device->total_bytes - device->bytes_used;
1058
1059                 /* align with stripe_len */
1060                 do_div(avail_space, BTRFS_STRIPE_LEN);
1061                 avail_space *= BTRFS_STRIPE_LEN;
1062
1063                 /*
1064                  * In order to avoid overwritting the superblock on the drive,
1065                  * btrfs starts at an offset of at least 1MB when doing chunk
1066                  * allocation.
1067                  */
1068                 skip_space = 1024 * 1024;
1069
1070                 /* user can set the offset in fs_info->alloc_start. */
1071                 if (fs_info->alloc_start + BTRFS_STRIPE_LEN <=
1072                     device->total_bytes)
1073                         skip_space = max(fs_info->alloc_start, skip_space);
1074
1075                 /*
1076                  * btrfs can not use the free space in [0, skip_space - 1],
1077                  * we must subtract it from the total. In order to implement
1078                  * it, we account the used space in this range first.
1079                  */
1080                 ret = btrfs_account_dev_extents_size(device, 0, skip_space - 1,
1081                                                      &used_space);
1082                 if (ret) {
1083                         kfree(devices_info);
1084                         return ret;
1085                 }
1086
1087                 /* calc the free space in [0, skip_space - 1] */
1088                 skip_space -= used_space;
1089
1090                 /*
1091                  * we can use the free space in [0, skip_space - 1], subtract
1092                  * it from the total.
1093                  */
1094                 if (avail_space && avail_space >= skip_space)
1095                         avail_space -= skip_space;
1096                 else
1097                         avail_space = 0;
1098
1099                 if (avail_space < min_stripe_size)
1100                         continue;
1101
1102                 devices_info[i].dev = device;
1103                 devices_info[i].max_avail = avail_space;
1104
1105                 i++;
1106         }
1107
1108         nr_devices = i;
1109
1110         btrfs_descending_sort_devices(devices_info, nr_devices);
1111
1112         i = nr_devices - 1;
1113         avail_space = 0;
1114         while (nr_devices >= min_stripes) {
1115                 if (devices_info[i].max_avail >= min_stripe_size) {
1116                         int j;
1117                         u64 alloc_size;
1118
1119                         avail_space += devices_info[i].max_avail * min_stripes;
1120                         alloc_size = devices_info[i].max_avail;
1121                         for (j = i + 1 - min_stripes; j <= i; j++)
1122                                 devices_info[j].max_avail -= alloc_size;
1123                 }
1124                 i--;
1125                 nr_devices--;
1126         }
1127
1128         kfree(devices_info);
1129         *free_bytes = avail_space;
1130         return 0;
1131 }
1132
1133 static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
1134 {
1135         struct btrfs_root *root = btrfs_sb(dentry->d_sb);
1136         struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
1137         struct list_head *head = &root->fs_info->space_info;
1138         struct btrfs_space_info *found;
1139         u64 total_used = 0;
1140         u64 total_free_data = 0;
1141         int bits = dentry->d_sb->s_blocksize_bits;
1142         __be32 *fsid = (__be32 *)root->fs_info->fsid;
1143         int ret;
1144
1145         /* holding chunk_muext to avoid allocating new chunks */
1146         mutex_lock(&root->fs_info->chunk_mutex);
1147         rcu_read_lock();
1148         list_for_each_entry_rcu(found, head, list) {
1149                 if (found->flags & BTRFS_BLOCK_GROUP_DATA) {
1150                         total_free_data += found->disk_total - found->disk_used;
1151                         total_free_data -=
1152                                 btrfs_account_ro_block_groups_free_space(found);
1153                 }
1154
1155                 total_used += found->disk_used;
1156         }
1157         rcu_read_unlock();
1158
1159         buf->f_namelen = BTRFS_NAME_LEN;
1160         buf->f_blocks = btrfs_super_total_bytes(disk_super) >> bits;
1161         buf->f_bfree = buf->f_blocks - (total_used >> bits);
1162         buf->f_bsize = dentry->d_sb->s_blocksize;
1163         buf->f_type = BTRFS_SUPER_MAGIC;
1164         buf->f_bavail = total_free_data;
1165         ret = btrfs_calc_avail_data_space(root, &total_free_data);
1166         if (ret) {
1167                 mutex_unlock(&root->fs_info->chunk_mutex);
1168                 return ret;
1169         }
1170         buf->f_bavail += total_free_data;
1171         buf->f_bavail = buf->f_bavail >> bits;
1172         mutex_unlock(&root->fs_info->chunk_mutex);
1173
1174         /* We treat it as constant endianness (it doesn't matter _which_)
1175            because we want the fsid to come out the same whether mounted
1176            on a big-endian or little-endian host */
1177         buf->f_fsid.val[0] = be32_to_cpu(fsid[0]) ^ be32_to_cpu(fsid[2]);
1178         buf->f_fsid.val[1] = be32_to_cpu(fsid[1]) ^ be32_to_cpu(fsid[3]);
1179         /* Mask in the root object ID too, to disambiguate subvols */
1180         buf->f_fsid.val[0] ^= BTRFS_I(dentry->d_inode)->root->objectid >> 32;
1181         buf->f_fsid.val[1] ^= BTRFS_I(dentry->d_inode)->root->objectid;
1182
1183         return 0;
1184 }
1185
1186 static struct file_system_type btrfs_fs_type = {
1187         .owner          = THIS_MODULE,
1188         .name           = "btrfs",
1189         .mount          = btrfs_mount,
1190         .kill_sb        = kill_anon_super,
1191         .fs_flags       = FS_REQUIRES_DEV,
1192 };
1193
1194 /*
1195  * used by btrfsctl to scan devices when no FS is mounted
1196  */
1197 static long btrfs_control_ioctl(struct file *file, unsigned int cmd,
1198                                 unsigned long arg)
1199 {
1200         struct btrfs_ioctl_vol_args *vol;
1201         struct btrfs_fs_devices *fs_devices;
1202         int ret = -ENOTTY;
1203
1204         if (!capable(CAP_SYS_ADMIN))
1205                 return -EPERM;
1206
1207         vol = memdup_user((void __user *)arg, sizeof(*vol));
1208         if (IS_ERR(vol))
1209                 return PTR_ERR(vol);
1210
1211         switch (cmd) {
1212         case BTRFS_IOC_SCAN_DEV:
1213                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
1214                                             &btrfs_fs_type, &fs_devices);
1215                 break;
1216         }
1217
1218         kfree(vol);
1219         return ret;
1220 }
1221
1222 static int btrfs_freeze(struct super_block *sb)
1223 {
1224         struct btrfs_root *root = btrfs_sb(sb);
1225         mutex_lock(&root->fs_info->transaction_kthread_mutex);
1226         mutex_lock(&root->fs_info->cleaner_mutex);
1227         return 0;
1228 }
1229
1230 static int btrfs_unfreeze(struct super_block *sb)
1231 {
1232         struct btrfs_root *root = btrfs_sb(sb);
1233         mutex_unlock(&root->fs_info->cleaner_mutex);
1234         mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1235         return 0;
1236 }
1237
1238 static const struct super_operations btrfs_super_ops = {
1239         .drop_inode     = btrfs_drop_inode,
1240         .evict_inode    = btrfs_evict_inode,
1241         .put_super      = btrfs_put_super,
1242         .sync_fs        = btrfs_sync_fs,
1243         .show_options   = btrfs_show_options,
1244         .write_inode    = btrfs_write_inode,
1245         .dirty_inode    = btrfs_dirty_inode,
1246         .alloc_inode    = btrfs_alloc_inode,
1247         .destroy_inode  = btrfs_destroy_inode,
1248         .statfs         = btrfs_statfs,
1249         .remount_fs     = btrfs_remount,
1250         .freeze_fs      = btrfs_freeze,
1251         .unfreeze_fs    = btrfs_unfreeze,
1252 };
1253
1254 static const struct file_operations btrfs_ctl_fops = {
1255         .unlocked_ioctl  = btrfs_control_ioctl,
1256         .compat_ioctl = btrfs_control_ioctl,
1257         .owner   = THIS_MODULE,
1258         .llseek = noop_llseek,
1259 };
1260
1261 static struct miscdevice btrfs_misc = {
1262         .minor          = BTRFS_MINOR,
1263         .name           = "btrfs-control",
1264         .fops           = &btrfs_ctl_fops
1265 };
1266
1267 MODULE_ALIAS_MISCDEV(BTRFS_MINOR);
1268 MODULE_ALIAS("devname:btrfs-control");
1269
1270 static int btrfs_interface_init(void)
1271 {
1272         return misc_register(&btrfs_misc);
1273 }
1274
1275 static void btrfs_interface_exit(void)
1276 {
1277         if (misc_deregister(&btrfs_misc) < 0)
1278                 printk(KERN_INFO "misc_deregister failed for control device");
1279 }
1280
1281 static int __init init_btrfs_fs(void)
1282 {
1283         int err;
1284
1285         err = btrfs_init_sysfs();
1286         if (err)
1287                 return err;
1288
1289         err = btrfs_init_compress();
1290         if (err)
1291                 goto free_sysfs;
1292
1293         err = btrfs_init_cachep();
1294         if (err)
1295                 goto free_compress;
1296
1297         err = extent_io_init();
1298         if (err)
1299                 goto free_cachep;
1300
1301         err = extent_map_init();
1302         if (err)
1303                 goto free_extent_io;
1304
1305         err = btrfs_delayed_inode_init();
1306         if (err)
1307                 goto free_extent_map;
1308
1309         err = btrfs_interface_init();
1310         if (err)
1311                 goto free_delayed_inode;
1312
1313         err = register_filesystem(&btrfs_fs_type);
1314         if (err)
1315                 goto unregister_ioctl;
1316
1317         printk(KERN_INFO "%s loaded\n", BTRFS_BUILD_VERSION);
1318         return 0;
1319
1320 unregister_ioctl:
1321         btrfs_interface_exit();
1322 free_delayed_inode:
1323         btrfs_delayed_inode_exit();
1324 free_extent_map:
1325         extent_map_exit();
1326 free_extent_io:
1327         extent_io_exit();
1328 free_cachep:
1329         btrfs_destroy_cachep();
1330 free_compress:
1331         btrfs_exit_compress();
1332 free_sysfs:
1333         btrfs_exit_sysfs();
1334         return err;
1335 }
1336
1337 static void __exit exit_btrfs_fs(void)
1338 {
1339         btrfs_destroy_cachep();
1340         btrfs_delayed_inode_exit();
1341         extent_map_exit();
1342         extent_io_exit();
1343         btrfs_interface_exit();
1344         unregister_filesystem(&btrfs_fs_type);
1345         btrfs_exit_sysfs();
1346         btrfs_cleanup_fs_uuids();
1347         btrfs_exit_compress();
1348 }
1349
1350 module_init(init_btrfs_fs)
1351 module_exit(exit_btrfs_fs)
1352
1353 MODULE_LICENSE("GPL");