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