Merge branch 'allocator' of git://git.kernel.org/pub/scm/linux/kernel/git/arne/btrfs...
[sfrench/cifs-2.6.git] / fs / btrfs / super.c
1 /*
2  * Copyright (C) 2007 Oracle.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18
19 #include <linux/blkdev.h>
20 #include <linux/module.h>
21 #include <linux/buffer_head.h>
22 #include <linux/fs.h>
23 #include <linux/pagemap.h>
24 #include <linux/highmem.h>
25 #include <linux/time.h>
26 #include <linux/init.h>
27 #include <linux/seq_file.h>
28 #include <linux/string.h>
29 #include <linux/backing-dev.h>
30 #include <linux/mount.h>
31 #include <linux/mpage.h>
32 #include <linux/swap.h>
33 #include <linux/writeback.h>
34 #include <linux/statfs.h>
35 #include <linux/compat.h>
36 #include <linux/parser.h>
37 #include <linux/ctype.h>
38 #include <linux/namei.h>
39 #include <linux/miscdevice.h>
40 #include <linux/magic.h>
41 #include <linux/slab.h>
42 #include "compat.h"
43 #include "delayed-inode.h"
44 #include "ctree.h"
45 #include "disk-io.h"
46 #include "transaction.h"
47 #include "btrfs_inode.h"
48 #include "ioctl.h"
49 #include "print-tree.h"
50 #include "xattr.h"
51 #include "volumes.h"
52 #include "version.h"
53 #include "export.h"
54 #include "compression.h"
55
56 #define CREATE_TRACE_POINTS
57 #include <trace/events/btrfs.h>
58
59 static const struct super_operations btrfs_super_ops;
60
61 static const char *btrfs_decode_error(struct btrfs_fs_info *fs_info, int errno,
62                                       char nbuf[16])
63 {
64         char *errstr = NULL;
65
66         switch (errno) {
67         case -EIO:
68                 errstr = "IO failure";
69                 break;
70         case -ENOMEM:
71                 errstr = "Out of memory";
72                 break;
73         case -EROFS:
74                 errstr = "Readonly filesystem";
75                 break;
76         default:
77                 if (nbuf) {
78                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
79                                 errstr = nbuf;
80                 }
81                 break;
82         }
83
84         return errstr;
85 }
86
87 static void __save_error_info(struct btrfs_fs_info *fs_info)
88 {
89         /*
90          * today we only save the error info into ram.  Long term we'll
91          * also send it down to the disk
92          */
93         fs_info->fs_state = BTRFS_SUPER_FLAG_ERROR;
94 }
95
96 /* NOTE:
97  *      We move write_super stuff at umount in order to avoid deadlock
98  *      for umount hold all lock.
99  */
100 static void save_error_info(struct btrfs_fs_info *fs_info)
101 {
102         __save_error_info(fs_info);
103 }
104
105 /* btrfs handle error by forcing the filesystem readonly */
106 static void btrfs_handle_error(struct btrfs_fs_info *fs_info)
107 {
108         struct super_block *sb = fs_info->sb;
109
110         if (sb->s_flags & MS_RDONLY)
111                 return;
112
113         if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
114                 sb->s_flags |= MS_RDONLY;
115                 printk(KERN_INFO "btrfs is forced readonly\n");
116         }
117 }
118
119 /*
120  * __btrfs_std_error decodes expected errors from the caller and
121  * invokes the approciate error response.
122  */
123 void __btrfs_std_error(struct btrfs_fs_info *fs_info, const char *function,
124                      unsigned int line, int errno)
125 {
126         struct super_block *sb = fs_info->sb;
127         char nbuf[16];
128         const char *errstr;
129
130         /*
131          * Special case: if the error is EROFS, and we're already
132          * under MS_RDONLY, then it is safe here.
133          */
134         if (errno == -EROFS && (sb->s_flags & MS_RDONLY))
135                 return;
136
137         errstr = btrfs_decode_error(fs_info, errno, nbuf);
138         printk(KERN_CRIT "BTRFS error (device %s) in %s:%d: %s\n",
139                 sb->s_id, function, line, errstr);
140         save_error_info(fs_info);
141
142         btrfs_handle_error(fs_info);
143 }
144
145 static void btrfs_put_super(struct super_block *sb)
146 {
147         struct btrfs_root *root = btrfs_sb(sb);
148         int ret;
149
150         ret = close_ctree(root);
151         sb->s_fs_info = NULL;
152
153         (void)ret; /* FIXME: need to fix VFS to return error? */
154 }
155
156 enum {
157         Opt_degraded, Opt_subvol, Opt_subvolid, Opt_device, Opt_nodatasum,
158         Opt_nodatacow, Opt_max_inline, Opt_alloc_start, Opt_nobarrier, Opt_ssd,
159         Opt_nossd, Opt_ssd_spread, Opt_thread_pool, Opt_noacl, Opt_compress,
160         Opt_compress_type, Opt_compress_force, Opt_compress_force_type,
161         Opt_notreelog, Opt_ratio, Opt_flushoncommit, Opt_discard,
162         Opt_space_cache, Opt_clear_cache, Opt_user_subvol_rm_allowed,
163         Opt_enospc_debug, Opt_subvolrootid, Opt_err,
164 };
165
166 static match_table_t tokens = {
167         {Opt_degraded, "degraded"},
168         {Opt_subvol, "subvol=%s"},
169         {Opt_subvolid, "subvolid=%d"},
170         {Opt_device, "device=%s"},
171         {Opt_nodatasum, "nodatasum"},
172         {Opt_nodatacow, "nodatacow"},
173         {Opt_nobarrier, "nobarrier"},
174         {Opt_max_inline, "max_inline=%s"},
175         {Opt_alloc_start, "alloc_start=%s"},
176         {Opt_thread_pool, "thread_pool=%d"},
177         {Opt_compress, "compress"},
178         {Opt_compress_type, "compress=%s"},
179         {Opt_compress_force, "compress-force"},
180         {Opt_compress_force_type, "compress-force=%s"},
181         {Opt_ssd, "ssd"},
182         {Opt_ssd_spread, "ssd_spread"},
183         {Opt_nossd, "nossd"},
184         {Opt_noacl, "noacl"},
185         {Opt_notreelog, "notreelog"},
186         {Opt_flushoncommit, "flushoncommit"},
187         {Opt_ratio, "metadata_ratio=%d"},
188         {Opt_discard, "discard"},
189         {Opt_space_cache, "space_cache"},
190         {Opt_clear_cache, "clear_cache"},
191         {Opt_user_subvol_rm_allowed, "user_subvol_rm_allowed"},
192         {Opt_enospc_debug, "enospc_debug"},
193         {Opt_subvolrootid, "subvolrootid=%d"},
194         {Opt_err, NULL},
195 };
196
197 /*
198  * Regular mount options parser.  Everything that is needed only when
199  * reading in a new superblock is parsed here.
200  */
201 int btrfs_parse_options(struct btrfs_root *root, char *options)
202 {
203         struct btrfs_fs_info *info = root->fs_info;
204         substring_t args[MAX_OPT_ARGS];
205         char *p, *num, *orig;
206         int intarg;
207         int ret = 0;
208         char *compress_type;
209         bool compress_force = false;
210
211         if (!options)
212                 return 0;
213
214         /*
215          * strsep changes the string, duplicate it because parse_options
216          * gets called twice
217          */
218         options = kstrdup(options, GFP_NOFS);
219         if (!options)
220                 return -ENOMEM;
221
222         orig = options;
223
224         while ((p = strsep(&options, ",")) != NULL) {
225                 int token;
226                 if (!*p)
227                         continue;
228
229                 token = match_token(p, tokens, args);
230                 switch (token) {
231                 case Opt_degraded:
232                         printk(KERN_INFO "btrfs: allowing degraded mounts\n");
233                         btrfs_set_opt(info->mount_opt, DEGRADED);
234                         break;
235                 case Opt_subvol:
236                 case Opt_subvolid:
237                 case Opt_subvolrootid:
238                 case Opt_device:
239                         /*
240                          * These are parsed by btrfs_parse_early_options
241                          * and can be happily ignored here.
242                          */
243                         break;
244                 case Opt_nodatasum:
245                         printk(KERN_INFO "btrfs: setting nodatasum\n");
246                         btrfs_set_opt(info->mount_opt, NODATASUM);
247                         break;
248                 case Opt_nodatacow:
249                         printk(KERN_INFO "btrfs: setting nodatacow\n");
250                         btrfs_set_opt(info->mount_opt, NODATACOW);
251                         btrfs_set_opt(info->mount_opt, NODATASUM);
252                         break;
253                 case Opt_compress_force:
254                 case Opt_compress_force_type:
255                         compress_force = true;
256                 case Opt_compress:
257                 case Opt_compress_type:
258                         if (token == Opt_compress ||
259                             token == Opt_compress_force ||
260                             strcmp(args[0].from, "zlib") == 0) {
261                                 compress_type = "zlib";
262                                 info->compress_type = BTRFS_COMPRESS_ZLIB;
263                         } else if (strcmp(args[0].from, "lzo") == 0) {
264                                 compress_type = "lzo";
265                                 info->compress_type = BTRFS_COMPRESS_LZO;
266                         } else {
267                                 ret = -EINVAL;
268                                 goto out;
269                         }
270
271                         btrfs_set_opt(info->mount_opt, COMPRESS);
272                         if (compress_force) {
273                                 btrfs_set_opt(info->mount_opt, FORCE_COMPRESS);
274                                 pr_info("btrfs: force %s compression\n",
275                                         compress_type);
276                         } else
277                                 pr_info("btrfs: use %s compression\n",
278                                         compress_type);
279                         break;
280                 case Opt_ssd:
281                         printk(KERN_INFO "btrfs: use ssd allocation scheme\n");
282                         btrfs_set_opt(info->mount_opt, SSD);
283                         break;
284                 case Opt_ssd_spread:
285                         printk(KERN_INFO "btrfs: use spread ssd "
286                                "allocation scheme\n");
287                         btrfs_set_opt(info->mount_opt, SSD);
288                         btrfs_set_opt(info->mount_opt, SSD_SPREAD);
289                         break;
290                 case Opt_nossd:
291                         printk(KERN_INFO "btrfs: not using ssd allocation "
292                                "scheme\n");
293                         btrfs_set_opt(info->mount_opt, NOSSD);
294                         btrfs_clear_opt(info->mount_opt, SSD);
295                         btrfs_clear_opt(info->mount_opt, SSD_SPREAD);
296                         break;
297                 case Opt_nobarrier:
298                         printk(KERN_INFO "btrfs: turning off barriers\n");
299                         btrfs_set_opt(info->mount_opt, NOBARRIER);
300                         break;
301                 case Opt_thread_pool:
302                         intarg = 0;
303                         match_int(&args[0], &intarg);
304                         if (intarg) {
305                                 info->thread_pool_size = intarg;
306                                 printk(KERN_INFO "btrfs: thread pool %d\n",
307                                        info->thread_pool_size);
308                         }
309                         break;
310                 case Opt_max_inline:
311                         num = match_strdup(&args[0]);
312                         if (num) {
313                                 info->max_inline = memparse(num, NULL);
314                                 kfree(num);
315
316                                 if (info->max_inline) {
317                                         info->max_inline = max_t(u64,
318                                                 info->max_inline,
319                                                 root->sectorsize);
320                                 }
321                                 printk(KERN_INFO "btrfs: max_inline at %llu\n",
322                                         (unsigned long long)info->max_inline);
323                         }
324                         break;
325                 case Opt_alloc_start:
326                         num = match_strdup(&args[0]);
327                         if (num) {
328                                 info->alloc_start = memparse(num, NULL);
329                                 kfree(num);
330                                 printk(KERN_INFO
331                                         "btrfs: allocations start at %llu\n",
332                                         (unsigned long long)info->alloc_start);
333                         }
334                         break;
335                 case Opt_noacl:
336                         root->fs_info->sb->s_flags &= ~MS_POSIXACL;
337                         break;
338                 case Opt_notreelog:
339                         printk(KERN_INFO "btrfs: disabling tree log\n");
340                         btrfs_set_opt(info->mount_opt, NOTREELOG);
341                         break;
342                 case Opt_flushoncommit:
343                         printk(KERN_INFO "btrfs: turning on flush-on-commit\n");
344                         btrfs_set_opt(info->mount_opt, FLUSHONCOMMIT);
345                         break;
346                 case Opt_ratio:
347                         intarg = 0;
348                         match_int(&args[0], &intarg);
349                         if (intarg) {
350                                 info->metadata_ratio = intarg;
351                                 printk(KERN_INFO "btrfs: metadata ratio %d\n",
352                                        info->metadata_ratio);
353                         }
354                         break;
355                 case Opt_discard:
356                         btrfs_set_opt(info->mount_opt, DISCARD);
357                         break;
358                 case Opt_space_cache:
359                         printk(KERN_INFO "btrfs: enabling disk space caching\n");
360                         btrfs_set_opt(info->mount_opt, SPACE_CACHE);
361                         break;
362                 case Opt_clear_cache:
363                         printk(KERN_INFO "btrfs: force clearing of disk cache\n");
364                         btrfs_set_opt(info->mount_opt, CLEAR_CACHE);
365                         break;
366                 case Opt_user_subvol_rm_allowed:
367                         btrfs_set_opt(info->mount_opt, USER_SUBVOL_RM_ALLOWED);
368                         break;
369                 case Opt_enospc_debug:
370                         btrfs_set_opt(info->mount_opt, ENOSPC_DEBUG);
371                         break;
372                 case Opt_err:
373                         printk(KERN_INFO "btrfs: unrecognized mount option "
374                                "'%s'\n", p);
375                         ret = -EINVAL;
376                         goto out;
377                 default:
378                         break;
379                 }
380         }
381 out:
382         kfree(orig);
383         return ret;
384 }
385
386 /*
387  * Parse mount options that are required early in the mount process.
388  *
389  * All other options will be parsed on much later in the mount process and
390  * only when we need to allocate a new super block.
391  */
392 static int btrfs_parse_early_options(const char *options, fmode_t flags,
393                 void *holder, char **subvol_name, u64 *subvol_objectid,
394                 u64 *subvol_rootid, struct btrfs_fs_devices **fs_devices)
395 {
396         substring_t args[MAX_OPT_ARGS];
397         char *opts, *orig, *p;
398         int error = 0;
399         int intarg;
400
401         if (!options)
402                 goto out;
403
404         /*
405          * strsep changes the string, duplicate it because parse_options
406          * gets called twice
407          */
408         opts = kstrdup(options, GFP_KERNEL);
409         if (!opts)
410                 return -ENOMEM;
411         orig = opts;
412
413         while ((p = strsep(&opts, ",")) != NULL) {
414                 int token;
415                 if (!*p)
416                         continue;
417
418                 token = match_token(p, tokens, args);
419                 switch (token) {
420                 case Opt_subvol:
421                         *subvol_name = match_strdup(&args[0]);
422                         break;
423                 case Opt_subvolid:
424                         intarg = 0;
425                         error = match_int(&args[0], &intarg);
426                         if (!error) {
427                                 /* we want the original fs_tree */
428                                 if (!intarg)
429                                         *subvol_objectid =
430                                                 BTRFS_FS_TREE_OBJECTID;
431                                 else
432                                         *subvol_objectid = intarg;
433                         }
434                         break;
435                 case Opt_subvolrootid:
436                         intarg = 0;
437                         error = match_int(&args[0], &intarg);
438                         if (!error) {
439                                 /* we want the original fs_tree */
440                                 if (!intarg)
441                                         *subvol_rootid =
442                                                 BTRFS_FS_TREE_OBJECTID;
443                                 else
444                                         *subvol_rootid = intarg;
445                         }
446                         break;
447                 case Opt_device:
448                         error = btrfs_scan_one_device(match_strdup(&args[0]),
449                                         flags, holder, fs_devices);
450                         if (error)
451                                 goto out_free_opts;
452                         break;
453                 default:
454                         break;
455                 }
456         }
457
458  out_free_opts:
459         kfree(orig);
460  out:
461         /*
462          * If no subvolume name is specified we use the default one.  Allocate
463          * a copy of the string "." here so that code later in the
464          * mount path doesn't care if it's the default volume or another one.
465          */
466         if (!*subvol_name) {
467                 *subvol_name = kstrdup(".", GFP_KERNEL);
468                 if (!*subvol_name)
469                         return -ENOMEM;
470         }
471         return error;
472 }
473
474 static struct dentry *get_default_root(struct super_block *sb,
475                                        u64 subvol_objectid)
476 {
477         struct btrfs_root *root = sb->s_fs_info;
478         struct btrfs_root *new_root;
479         struct btrfs_dir_item *di;
480         struct btrfs_path *path;
481         struct btrfs_key location;
482         struct inode *inode;
483         struct dentry *dentry;
484         u64 dir_id;
485         int new = 0;
486
487         /*
488          * We have a specific subvol we want to mount, just setup location and
489          * go look up the root.
490          */
491         if (subvol_objectid) {
492                 location.objectid = subvol_objectid;
493                 location.type = BTRFS_ROOT_ITEM_KEY;
494                 location.offset = (u64)-1;
495                 goto find_root;
496         }
497
498         path = btrfs_alloc_path();
499         if (!path)
500                 return ERR_PTR(-ENOMEM);
501         path->leave_spinning = 1;
502
503         /*
504          * Find the "default" dir item which points to the root item that we
505          * will mount by default if we haven't been given a specific subvolume
506          * to mount.
507          */
508         dir_id = btrfs_super_root_dir(&root->fs_info->super_copy);
509         di = btrfs_lookup_dir_item(NULL, root, path, dir_id, "default", 7, 0);
510         if (IS_ERR(di))
511                 return ERR_CAST(di);
512         if (!di) {
513                 /*
514                  * Ok the default dir item isn't there.  This is weird since
515                  * it's always been there, but don't freak out, just try and
516                  * mount to root most subvolume.
517                  */
518                 btrfs_free_path(path);
519                 dir_id = BTRFS_FIRST_FREE_OBJECTID;
520                 new_root = root->fs_info->fs_root;
521                 goto setup_root;
522         }
523
524         btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
525         btrfs_free_path(path);
526
527 find_root:
528         new_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
529         if (IS_ERR(new_root))
530                 return ERR_CAST(new_root);
531
532         if (btrfs_root_refs(&new_root->root_item) == 0)
533                 return ERR_PTR(-ENOENT);
534
535         dir_id = btrfs_root_dirid(&new_root->root_item);
536 setup_root:
537         location.objectid = dir_id;
538         location.type = BTRFS_INODE_ITEM_KEY;
539         location.offset = 0;
540
541         inode = btrfs_iget(sb, &location, new_root, &new);
542         if (IS_ERR(inode))
543                 return ERR_CAST(inode);
544
545         /*
546          * If we're just mounting the root most subvol put the inode and return
547          * a reference to the dentry.  We will have already gotten a reference
548          * to the inode in btrfs_fill_super so we're good to go.
549          */
550         if (!new && sb->s_root->d_inode == inode) {
551                 iput(inode);
552                 return dget(sb->s_root);
553         }
554
555         if (new) {
556                 const struct qstr name = { .name = "/", .len = 1 };
557
558                 /*
559                  * New inode, we need to make the dentry a sibling of s_root so
560                  * everything gets cleaned up properly on unmount.
561                  */
562                 dentry = d_alloc(sb->s_root, &name);
563                 if (!dentry) {
564                         iput(inode);
565                         return ERR_PTR(-ENOMEM);
566                 }
567                 d_splice_alias(inode, dentry);
568         } else {
569                 /*
570                  * We found the inode in cache, just find a dentry for it and
571                  * put the reference to the inode we just got.
572                  */
573                 dentry = d_find_alias(inode);
574                 iput(inode);
575         }
576
577         return dentry;
578 }
579
580 static int btrfs_fill_super(struct super_block *sb,
581                             struct btrfs_fs_devices *fs_devices,
582                             void *data, int silent)
583 {
584         struct inode *inode;
585         struct dentry *root_dentry;
586         struct btrfs_root *tree_root;
587         struct btrfs_key key;
588         int err;
589
590         sb->s_maxbytes = MAX_LFS_FILESIZE;
591         sb->s_magic = BTRFS_SUPER_MAGIC;
592         sb->s_op = &btrfs_super_ops;
593         sb->s_d_op = &btrfs_dentry_operations;
594         sb->s_export_op = &btrfs_export_ops;
595         sb->s_xattr = btrfs_xattr_handlers;
596         sb->s_time_gran = 1;
597 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
598         sb->s_flags |= MS_POSIXACL;
599 #endif
600
601         tree_root = open_ctree(sb, fs_devices, (char *)data);
602
603         if (IS_ERR(tree_root)) {
604                 printk("btrfs: open_ctree failed\n");
605                 return PTR_ERR(tree_root);
606         }
607         sb->s_fs_info = tree_root;
608
609         key.objectid = BTRFS_FIRST_FREE_OBJECTID;
610         key.type = BTRFS_INODE_ITEM_KEY;
611         key.offset = 0;
612         inode = btrfs_iget(sb, &key, tree_root->fs_info->fs_root, NULL);
613         if (IS_ERR(inode)) {
614                 err = PTR_ERR(inode);
615                 goto fail_close;
616         }
617
618         root_dentry = d_alloc_root(inode);
619         if (!root_dentry) {
620                 iput(inode);
621                 err = -ENOMEM;
622                 goto fail_close;
623         }
624
625         sb->s_root = root_dentry;
626
627         save_mount_options(sb, data);
628         return 0;
629
630 fail_close:
631         close_ctree(tree_root);
632         return err;
633 }
634
635 int btrfs_sync_fs(struct super_block *sb, int wait)
636 {
637         struct btrfs_trans_handle *trans;
638         struct btrfs_root *root = btrfs_sb(sb);
639         int ret;
640
641         trace_btrfs_sync_fs(wait);
642
643         if (!wait) {
644                 filemap_flush(root->fs_info->btree_inode->i_mapping);
645                 return 0;
646         }
647
648         btrfs_start_delalloc_inodes(root, 0);
649         btrfs_wait_ordered_extents(root, 0, 0);
650
651         trans = btrfs_start_transaction(root, 0);
652         if (IS_ERR(trans))
653                 return PTR_ERR(trans);
654         ret = btrfs_commit_transaction(trans, root);
655         return ret;
656 }
657
658 static int btrfs_show_options(struct seq_file *seq, struct vfsmount *vfs)
659 {
660         struct btrfs_root *root = btrfs_sb(vfs->mnt_sb);
661         struct btrfs_fs_info *info = root->fs_info;
662         char *compress_type;
663
664         if (btrfs_test_opt(root, DEGRADED))
665                 seq_puts(seq, ",degraded");
666         if (btrfs_test_opt(root, NODATASUM))
667                 seq_puts(seq, ",nodatasum");
668         if (btrfs_test_opt(root, NODATACOW))
669                 seq_puts(seq, ",nodatacow");
670         if (btrfs_test_opt(root, NOBARRIER))
671                 seq_puts(seq, ",nobarrier");
672         if (info->max_inline != 8192 * 1024)
673                 seq_printf(seq, ",max_inline=%llu",
674                            (unsigned long long)info->max_inline);
675         if (info->alloc_start != 0)
676                 seq_printf(seq, ",alloc_start=%llu",
677                            (unsigned long long)info->alloc_start);
678         if (info->thread_pool_size !=  min_t(unsigned long,
679                                              num_online_cpus() + 2, 8))
680                 seq_printf(seq, ",thread_pool=%d", info->thread_pool_size);
681         if (btrfs_test_opt(root, COMPRESS)) {
682                 if (info->compress_type == BTRFS_COMPRESS_ZLIB)
683                         compress_type = "zlib";
684                 else
685                         compress_type = "lzo";
686                 if (btrfs_test_opt(root, FORCE_COMPRESS))
687                         seq_printf(seq, ",compress-force=%s", compress_type);
688                 else
689                         seq_printf(seq, ",compress=%s", compress_type);
690         }
691         if (btrfs_test_opt(root, NOSSD))
692                 seq_puts(seq, ",nossd");
693         if (btrfs_test_opt(root, SSD_SPREAD))
694                 seq_puts(seq, ",ssd_spread");
695         else if (btrfs_test_opt(root, SSD))
696                 seq_puts(seq, ",ssd");
697         if (btrfs_test_opt(root, NOTREELOG))
698                 seq_puts(seq, ",notreelog");
699         if (btrfs_test_opt(root, FLUSHONCOMMIT))
700                 seq_puts(seq, ",flushoncommit");
701         if (btrfs_test_opt(root, DISCARD))
702                 seq_puts(seq, ",discard");
703         if (!(root->fs_info->sb->s_flags & MS_POSIXACL))
704                 seq_puts(seq, ",noacl");
705         if (btrfs_test_opt(root, SPACE_CACHE))
706                 seq_puts(seq, ",space_cache");
707         if (btrfs_test_opt(root, CLEAR_CACHE))
708                 seq_puts(seq, ",clear_cache");
709         if (btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
710                 seq_puts(seq, ",user_subvol_rm_allowed");
711         return 0;
712 }
713
714 static int btrfs_test_super(struct super_block *s, void *data)
715 {
716         struct btrfs_root *test_root = data;
717         struct btrfs_root *root = btrfs_sb(s);
718
719         /*
720          * If this super block is going away, return false as it
721          * can't match as an existing super block.
722          */
723         if (!atomic_read(&s->s_active))
724                 return 0;
725         return root->fs_info->fs_devices == test_root->fs_info->fs_devices;
726 }
727
728 static int btrfs_set_super(struct super_block *s, void *data)
729 {
730         s->s_fs_info = data;
731
732         return set_anon_super(s, data);
733 }
734
735
736 /*
737  * Find a superblock for the given device / mount point.
738  *
739  * Note:  This is based on get_sb_bdev from fs/super.c with a few additions
740  *        for multiple device setup.  Make sure to keep it in sync.
741  */
742 static struct dentry *btrfs_mount(struct file_system_type *fs_type, int flags,
743                 const char *device_name, void *data)
744 {
745         struct block_device *bdev = NULL;
746         struct super_block *s;
747         struct dentry *root;
748         struct btrfs_fs_devices *fs_devices = NULL;
749         struct btrfs_root *tree_root = NULL;
750         struct btrfs_fs_info *fs_info = NULL;
751         fmode_t mode = FMODE_READ;
752         char *subvol_name = NULL;
753         u64 subvol_objectid = 0;
754         u64 subvol_rootid = 0;
755         int error = 0;
756
757         if (!(flags & MS_RDONLY))
758                 mode |= FMODE_WRITE;
759
760         error = btrfs_parse_early_options(data, mode, fs_type,
761                                           &subvol_name, &subvol_objectid,
762                                           &subvol_rootid, &fs_devices);
763         if (error)
764                 return ERR_PTR(error);
765
766         error = btrfs_scan_one_device(device_name, mode, fs_type, &fs_devices);
767         if (error)
768                 goto error_free_subvol_name;
769
770         error = btrfs_open_devices(fs_devices, mode, fs_type);
771         if (error)
772                 goto error_free_subvol_name;
773
774         if (!(flags & MS_RDONLY) && fs_devices->rw_devices == 0) {
775                 error = -EACCES;
776                 goto error_close_devices;
777         }
778
779         /*
780          * Setup a dummy root and fs_info for test/set super.  This is because
781          * we don't actually fill this stuff out until open_ctree, but we need
782          * it for searching for existing supers, so this lets us do that and
783          * then open_ctree will properly initialize everything later.
784          */
785         fs_info = kzalloc(sizeof(struct btrfs_fs_info), GFP_NOFS);
786         tree_root = kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
787         if (!fs_info || !tree_root) {
788                 error = -ENOMEM;
789                 goto error_close_devices;
790         }
791         fs_info->tree_root = tree_root;
792         fs_info->fs_devices = fs_devices;
793         tree_root->fs_info = fs_info;
794
795         bdev = fs_devices->latest_bdev;
796         s = sget(fs_type, btrfs_test_super, btrfs_set_super, tree_root);
797         if (IS_ERR(s))
798                 goto error_s;
799
800         if (s->s_root) {
801                 if ((flags ^ s->s_flags) & MS_RDONLY) {
802                         deactivate_locked_super(s);
803                         error = -EBUSY;
804                         goto error_close_devices;
805                 }
806
807                 btrfs_close_devices(fs_devices);
808                 kfree(fs_info);
809                 kfree(tree_root);
810         } else {
811                 char b[BDEVNAME_SIZE];
812
813                 s->s_flags = flags;
814                 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
815                 error = btrfs_fill_super(s, fs_devices, data,
816                                          flags & MS_SILENT ? 1 : 0);
817                 if (error) {
818                         deactivate_locked_super(s);
819                         goto error_free_subvol_name;
820                 }
821
822                 btrfs_sb(s)->fs_info->bdev_holder = fs_type;
823                 s->s_flags |= MS_ACTIVE;
824         }
825
826         /* if they gave us a subvolume name bind mount into that */
827         if (strcmp(subvol_name, ".")) {
828                 struct dentry *new_root;
829
830                 root = get_default_root(s, subvol_rootid);
831                 if (IS_ERR(root)) {
832                         error = PTR_ERR(root);
833                         deactivate_locked_super(s);
834                         goto error_free_subvol_name;
835                 }
836
837                 mutex_lock(&root->d_inode->i_mutex);
838                 new_root = lookup_one_len(subvol_name, root,
839                                       strlen(subvol_name));
840                 mutex_unlock(&root->d_inode->i_mutex);
841
842                 if (IS_ERR(new_root)) {
843                         dput(root);
844                         deactivate_locked_super(s);
845                         error = PTR_ERR(new_root);
846                         goto error_free_subvol_name;
847                 }
848                 if (!new_root->d_inode) {
849                         dput(root);
850                         dput(new_root);
851                         deactivate_locked_super(s);
852                         error = -ENXIO;
853                         goto error_free_subvol_name;
854                 }
855                 dput(root);
856                 root = new_root;
857         } else {
858                 root = get_default_root(s, subvol_objectid);
859                 if (IS_ERR(root)) {
860                         error = PTR_ERR(root);
861                         deactivate_locked_super(s);
862                         goto error_free_subvol_name;
863                 }
864         }
865
866         kfree(subvol_name);
867         return root;
868
869 error_s:
870         error = PTR_ERR(s);
871 error_close_devices:
872         btrfs_close_devices(fs_devices);
873         kfree(fs_info);
874         kfree(tree_root);
875 error_free_subvol_name:
876         kfree(subvol_name);
877         return ERR_PTR(error);
878 }
879
880 static int btrfs_remount(struct super_block *sb, int *flags, char *data)
881 {
882         struct btrfs_root *root = btrfs_sb(sb);
883         int ret;
884
885         ret = btrfs_parse_options(root, data);
886         if (ret)
887                 return -EINVAL;
888
889         if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
890                 return 0;
891
892         if (*flags & MS_RDONLY) {
893                 sb->s_flags |= MS_RDONLY;
894
895                 ret =  btrfs_commit_super(root);
896                 WARN_ON(ret);
897         } else {
898                 if (root->fs_info->fs_devices->rw_devices == 0)
899                         return -EACCES;
900
901                 if (btrfs_super_log_root(&root->fs_info->super_copy) != 0)
902                         return -EINVAL;
903
904                 ret = btrfs_cleanup_fs_roots(root->fs_info);
905                 WARN_ON(ret);
906
907                 /* recover relocation */
908                 ret = btrfs_recover_relocation(root);
909                 WARN_ON(ret);
910
911                 sb->s_flags &= ~MS_RDONLY;
912         }
913
914         return 0;
915 }
916
917 /* Used to sort the devices by max_avail(descending sort) */
918 static int btrfs_cmp_device_free_bytes(const void *dev_info1,
919                                        const void *dev_info2)
920 {
921         if (((struct btrfs_device_info *)dev_info1)->max_avail >
922             ((struct btrfs_device_info *)dev_info2)->max_avail)
923                 return -1;
924         else if (((struct btrfs_device_info *)dev_info1)->max_avail <
925                  ((struct btrfs_device_info *)dev_info2)->max_avail)
926                 return 1;
927         else
928         return 0;
929 }
930
931 /*
932  * sort the devices by max_avail, in which max free extent size of each device
933  * is stored.(Descending Sort)
934  */
935 static inline void btrfs_descending_sort_devices(
936                                         struct btrfs_device_info *devices,
937                                         size_t nr_devices)
938 {
939         sort(devices, nr_devices, sizeof(struct btrfs_device_info),
940              btrfs_cmp_device_free_bytes, NULL);
941 }
942
943 /*
944  * The helper to calc the free space on the devices that can be used to store
945  * file data.
946  */
947 static int btrfs_calc_avail_data_space(struct btrfs_root *root, u64 *free_bytes)
948 {
949         struct btrfs_fs_info *fs_info = root->fs_info;
950         struct btrfs_device_info *devices_info;
951         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
952         struct btrfs_device *device;
953         u64 skip_space;
954         u64 type;
955         u64 avail_space;
956         u64 used_space;
957         u64 min_stripe_size;
958         int min_stripes = 1;
959         int i = 0, nr_devices;
960         int ret;
961
962         nr_devices = fs_info->fs_devices->rw_devices;
963         BUG_ON(!nr_devices);
964
965         devices_info = kmalloc(sizeof(*devices_info) * nr_devices,
966                                GFP_NOFS);
967         if (!devices_info)
968                 return -ENOMEM;
969
970         /* calc min stripe number for data space alloction */
971         type = btrfs_get_alloc_profile(root, 1);
972         if (type & BTRFS_BLOCK_GROUP_RAID0)
973                 min_stripes = 2;
974         else if (type & BTRFS_BLOCK_GROUP_RAID1)
975                 min_stripes = 2;
976         else if (type & BTRFS_BLOCK_GROUP_RAID10)
977                 min_stripes = 4;
978
979         if (type & BTRFS_BLOCK_GROUP_DUP)
980                 min_stripe_size = 2 * BTRFS_STRIPE_LEN;
981         else
982                 min_stripe_size = BTRFS_STRIPE_LEN;
983
984         list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
985                 if (!device->in_fs_metadata)
986                         continue;
987
988                 avail_space = device->total_bytes - device->bytes_used;
989
990                 /* align with stripe_len */
991                 do_div(avail_space, BTRFS_STRIPE_LEN);
992                 avail_space *= BTRFS_STRIPE_LEN;
993
994                 /*
995                  * In order to avoid overwritting the superblock on the drive,
996                  * btrfs starts at an offset of at least 1MB when doing chunk
997                  * allocation.
998                  */
999                 skip_space = 1024 * 1024;
1000
1001                 /* user can set the offset in fs_info->alloc_start. */
1002                 if (fs_info->alloc_start + BTRFS_STRIPE_LEN <=
1003                     device->total_bytes)
1004                         skip_space = max(fs_info->alloc_start, skip_space);
1005
1006                 /*
1007                  * btrfs can not use the free space in [0, skip_space - 1],
1008                  * we must subtract it from the total. In order to implement
1009                  * it, we account the used space in this range first.
1010                  */
1011                 ret = btrfs_account_dev_extents_size(device, 0, skip_space - 1,
1012                                                      &used_space);
1013                 if (ret) {
1014                         kfree(devices_info);
1015                         return ret;
1016                 }
1017
1018                 /* calc the free space in [0, skip_space - 1] */
1019                 skip_space -= used_space;
1020
1021                 /*
1022                  * we can use the free space in [0, skip_space - 1], subtract
1023                  * it from the total.
1024                  */
1025                 if (avail_space && avail_space >= skip_space)
1026                         avail_space -= skip_space;
1027                 else
1028                         avail_space = 0;
1029
1030                 if (avail_space < min_stripe_size)
1031                         continue;
1032
1033                 devices_info[i].dev = device;
1034                 devices_info[i].max_avail = avail_space;
1035
1036                 i++;
1037         }
1038
1039         nr_devices = i;
1040
1041         btrfs_descending_sort_devices(devices_info, nr_devices);
1042
1043         i = nr_devices - 1;
1044         avail_space = 0;
1045         while (nr_devices >= min_stripes) {
1046                 if (devices_info[i].max_avail >= min_stripe_size) {
1047                         int j;
1048                         u64 alloc_size;
1049
1050                         avail_space += devices_info[i].max_avail * min_stripes;
1051                         alloc_size = devices_info[i].max_avail;
1052                         for (j = i + 1 - min_stripes; j <= i; j++)
1053                                 devices_info[j].max_avail -= alloc_size;
1054                 }
1055                 i--;
1056                 nr_devices--;
1057         }
1058
1059         kfree(devices_info);
1060         *free_bytes = avail_space;
1061         return 0;
1062 }
1063
1064 static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
1065 {
1066         struct btrfs_root *root = btrfs_sb(dentry->d_sb);
1067         struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
1068         struct list_head *head = &root->fs_info->space_info;
1069         struct btrfs_space_info *found;
1070         u64 total_used = 0;
1071         u64 total_free_data = 0;
1072         int bits = dentry->d_sb->s_blocksize_bits;
1073         __be32 *fsid = (__be32 *)root->fs_info->fsid;
1074         int ret;
1075
1076         /* holding chunk_muext to avoid allocating new chunks */
1077         mutex_lock(&root->fs_info->chunk_mutex);
1078         rcu_read_lock();
1079         list_for_each_entry_rcu(found, head, list) {
1080                 if (found->flags & BTRFS_BLOCK_GROUP_DATA) {
1081                         total_free_data += found->disk_total - found->disk_used;
1082                         total_free_data -=
1083                                 btrfs_account_ro_block_groups_free_space(found);
1084                 }
1085
1086                 total_used += found->disk_used;
1087         }
1088         rcu_read_unlock();
1089
1090         buf->f_namelen = BTRFS_NAME_LEN;
1091         buf->f_blocks = btrfs_super_total_bytes(disk_super) >> bits;
1092         buf->f_bfree = buf->f_blocks - (total_used >> bits);
1093         buf->f_bsize = dentry->d_sb->s_blocksize;
1094         buf->f_type = BTRFS_SUPER_MAGIC;
1095         buf->f_bavail = total_free_data;
1096         ret = btrfs_calc_avail_data_space(root, &total_free_data);
1097         if (ret) {
1098                 mutex_unlock(&root->fs_info->chunk_mutex);
1099                 return ret;
1100         }
1101         buf->f_bavail += total_free_data;
1102         buf->f_bavail = buf->f_bavail >> bits;
1103         mutex_unlock(&root->fs_info->chunk_mutex);
1104
1105         /* We treat it as constant endianness (it doesn't matter _which_)
1106            because we want the fsid to come out the same whether mounted
1107            on a big-endian or little-endian host */
1108         buf->f_fsid.val[0] = be32_to_cpu(fsid[0]) ^ be32_to_cpu(fsid[2]);
1109         buf->f_fsid.val[1] = be32_to_cpu(fsid[1]) ^ be32_to_cpu(fsid[3]);
1110         /* Mask in the root object ID too, to disambiguate subvols */
1111         buf->f_fsid.val[0] ^= BTRFS_I(dentry->d_inode)->root->objectid >> 32;
1112         buf->f_fsid.val[1] ^= BTRFS_I(dentry->d_inode)->root->objectid;
1113
1114         return 0;
1115 }
1116
1117 static struct file_system_type btrfs_fs_type = {
1118         .owner          = THIS_MODULE,
1119         .name           = "btrfs",
1120         .mount          = btrfs_mount,
1121         .kill_sb        = kill_anon_super,
1122         .fs_flags       = FS_REQUIRES_DEV,
1123 };
1124
1125 /*
1126  * used by btrfsctl to scan devices when no FS is mounted
1127  */
1128 static long btrfs_control_ioctl(struct file *file, unsigned int cmd,
1129                                 unsigned long arg)
1130 {
1131         struct btrfs_ioctl_vol_args *vol;
1132         struct btrfs_fs_devices *fs_devices;
1133         int ret = -ENOTTY;
1134
1135         if (!capable(CAP_SYS_ADMIN))
1136                 return -EPERM;
1137
1138         vol = memdup_user((void __user *)arg, sizeof(*vol));
1139         if (IS_ERR(vol))
1140                 return PTR_ERR(vol);
1141
1142         switch (cmd) {
1143         case BTRFS_IOC_SCAN_DEV:
1144                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
1145                                             &btrfs_fs_type, &fs_devices);
1146                 break;
1147         }
1148
1149         kfree(vol);
1150         return ret;
1151 }
1152
1153 static int btrfs_freeze(struct super_block *sb)
1154 {
1155         struct btrfs_root *root = btrfs_sb(sb);
1156         mutex_lock(&root->fs_info->transaction_kthread_mutex);
1157         mutex_lock(&root->fs_info->cleaner_mutex);
1158         return 0;
1159 }
1160
1161 static int btrfs_unfreeze(struct super_block *sb)
1162 {
1163         struct btrfs_root *root = btrfs_sb(sb);
1164         mutex_unlock(&root->fs_info->cleaner_mutex);
1165         mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1166         return 0;
1167 }
1168
1169 static const struct super_operations btrfs_super_ops = {
1170         .drop_inode     = btrfs_drop_inode,
1171         .evict_inode    = btrfs_evict_inode,
1172         .put_super      = btrfs_put_super,
1173         .sync_fs        = btrfs_sync_fs,
1174         .show_options   = btrfs_show_options,
1175         .write_inode    = btrfs_write_inode,
1176         .dirty_inode    = btrfs_dirty_inode,
1177         .alloc_inode    = btrfs_alloc_inode,
1178         .destroy_inode  = btrfs_destroy_inode,
1179         .statfs         = btrfs_statfs,
1180         .remount_fs     = btrfs_remount,
1181         .freeze_fs      = btrfs_freeze,
1182         .unfreeze_fs    = btrfs_unfreeze,
1183 };
1184
1185 static const struct file_operations btrfs_ctl_fops = {
1186         .unlocked_ioctl  = btrfs_control_ioctl,
1187         .compat_ioctl = btrfs_control_ioctl,
1188         .owner   = THIS_MODULE,
1189         .llseek = noop_llseek,
1190 };
1191
1192 static struct miscdevice btrfs_misc = {
1193         .minor          = BTRFS_MINOR,
1194         .name           = "btrfs-control",
1195         .fops           = &btrfs_ctl_fops
1196 };
1197
1198 MODULE_ALIAS_MISCDEV(BTRFS_MINOR);
1199 MODULE_ALIAS("devname:btrfs-control");
1200
1201 static int btrfs_interface_init(void)
1202 {
1203         return misc_register(&btrfs_misc);
1204 }
1205
1206 static void btrfs_interface_exit(void)
1207 {
1208         if (misc_deregister(&btrfs_misc) < 0)
1209                 printk(KERN_INFO "misc_deregister failed for control device");
1210 }
1211
1212 static int __init init_btrfs_fs(void)
1213 {
1214         int err;
1215
1216         err = btrfs_init_sysfs();
1217         if (err)
1218                 return err;
1219
1220         err = btrfs_init_compress();
1221         if (err)
1222                 goto free_sysfs;
1223
1224         err = btrfs_init_cachep();
1225         if (err)
1226                 goto free_compress;
1227
1228         err = extent_io_init();
1229         if (err)
1230                 goto free_cachep;
1231
1232         err = extent_map_init();
1233         if (err)
1234                 goto free_extent_io;
1235
1236         err = btrfs_delayed_inode_init();
1237         if (err)
1238                 goto free_extent_map;
1239
1240         err = btrfs_interface_init();
1241         if (err)
1242                 goto free_delayed_inode;
1243
1244         err = register_filesystem(&btrfs_fs_type);
1245         if (err)
1246                 goto unregister_ioctl;
1247
1248         printk(KERN_INFO "%s loaded\n", BTRFS_BUILD_VERSION);
1249         return 0;
1250
1251 unregister_ioctl:
1252         btrfs_interface_exit();
1253 free_delayed_inode:
1254         btrfs_delayed_inode_exit();
1255 free_extent_map:
1256         extent_map_exit();
1257 free_extent_io:
1258         extent_io_exit();
1259 free_cachep:
1260         btrfs_destroy_cachep();
1261 free_compress:
1262         btrfs_exit_compress();
1263 free_sysfs:
1264         btrfs_exit_sysfs();
1265         return err;
1266 }
1267
1268 static void __exit exit_btrfs_fs(void)
1269 {
1270         btrfs_destroy_cachep();
1271         btrfs_delayed_inode_exit();
1272         extent_map_exit();
1273         extent_io_exit();
1274         btrfs_interface_exit();
1275         unregister_filesystem(&btrfs_fs_type);
1276         btrfs_exit_sysfs();
1277         btrfs_cleanup_fs_uuids();
1278         btrfs_exit_compress();
1279 }
1280
1281 module_init(init_btrfs_fs)
1282 module_exit(exit_btrfs_fs)
1283
1284 MODULE_LICENSE("GPL");