Btrfs: use the commit_root for reading free_space_inode crcs
[pandora-kernel.git] / fs / btrfs / super.c
1 /*
2  * Copyright (C) 2007 Oracle.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18
19 #include <linux/blkdev.h>
20 #include <linux/module.h>
21 #include <linux/buffer_head.h>
22 #include <linux/fs.h>
23 #include <linux/pagemap.h>
24 #include <linux/highmem.h>
25 #include <linux/time.h>
26 #include <linux/init.h>
27 #include <linux/seq_file.h>
28 #include <linux/string.h>
29 #include <linux/backing-dev.h>
30 #include <linux/mount.h>
31 #include <linux/mpage.h>
32 #include <linux/swap.h>
33 #include <linux/writeback.h>
34 #include <linux/statfs.h>
35 #include <linux/compat.h>
36 #include <linux/parser.h>
37 #include <linux/ctype.h>
38 #include <linux/namei.h>
39 #include <linux/miscdevice.h>
40 #include <linux/magic.h>
41 #include <linux/slab.h>
42 #include "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_defrag,
164         Opt_inode_cache, Opt_err,
165 };
166
167 static match_table_t tokens = {
168         {Opt_degraded, "degraded"},
169         {Opt_subvol, "subvol=%s"},
170         {Opt_subvolid, "subvolid=%d"},
171         {Opt_device, "device=%s"},
172         {Opt_nodatasum, "nodatasum"},
173         {Opt_nodatacow, "nodatacow"},
174         {Opt_nobarrier, "nobarrier"},
175         {Opt_max_inline, "max_inline=%s"},
176         {Opt_alloc_start, "alloc_start=%s"},
177         {Opt_thread_pool, "thread_pool=%d"},
178         {Opt_compress, "compress"},
179         {Opt_compress_type, "compress=%s"},
180         {Opt_compress_force, "compress-force"},
181         {Opt_compress_force_type, "compress-force=%s"},
182         {Opt_ssd, "ssd"},
183         {Opt_ssd_spread, "ssd_spread"},
184         {Opt_nossd, "nossd"},
185         {Opt_noacl, "noacl"},
186         {Opt_notreelog, "notreelog"},
187         {Opt_flushoncommit, "flushoncommit"},
188         {Opt_ratio, "metadata_ratio=%d"},
189         {Opt_discard, "discard"},
190         {Opt_space_cache, "space_cache"},
191         {Opt_clear_cache, "clear_cache"},
192         {Opt_user_subvol_rm_allowed, "user_subvol_rm_allowed"},
193         {Opt_enospc_debug, "enospc_debug"},
194         {Opt_subvolrootid, "subvolrootid=%d"},
195         {Opt_defrag, "autodefrag"},
196         {Opt_inode_cache, "inode_cache"},
197         {Opt_err, NULL},
198 };
199
200 /*
201  * Regular mount options parser.  Everything that is needed only when
202  * reading in a new superblock is parsed here.
203  */
204 int btrfs_parse_options(struct btrfs_root *root, char *options)
205 {
206         struct btrfs_fs_info *info = root->fs_info;
207         substring_t args[MAX_OPT_ARGS];
208         char *p, *num, *orig;
209         int intarg;
210         int ret = 0;
211         char *compress_type;
212         bool compress_force = false;
213
214         if (!options)
215                 return 0;
216
217         /*
218          * strsep changes the string, duplicate it because parse_options
219          * gets called twice
220          */
221         options = kstrdup(options, GFP_NOFS);
222         if (!options)
223                 return -ENOMEM;
224
225         orig = options;
226
227         while ((p = strsep(&options, ",")) != NULL) {
228                 int token;
229                 if (!*p)
230                         continue;
231
232                 token = match_token(p, tokens, args);
233                 switch (token) {
234                 case Opt_degraded:
235                         printk(KERN_INFO "btrfs: allowing degraded mounts\n");
236                         btrfs_set_opt(info->mount_opt, DEGRADED);
237                         break;
238                 case Opt_subvol:
239                 case Opt_subvolid:
240                 case Opt_subvolrootid:
241                 case Opt_device:
242                         /*
243                          * These are parsed by btrfs_parse_early_options
244                          * and can be happily ignored here.
245                          */
246                         break;
247                 case Opt_nodatasum:
248                         printk(KERN_INFO "btrfs: setting nodatasum\n");
249                         btrfs_set_opt(info->mount_opt, NODATASUM);
250                         break;
251                 case Opt_nodatacow:
252                         printk(KERN_INFO "btrfs: setting nodatacow\n");
253                         btrfs_set_opt(info->mount_opt, NODATACOW);
254                         btrfs_set_opt(info->mount_opt, NODATASUM);
255                         break;
256                 case Opt_compress_force:
257                 case Opt_compress_force_type:
258                         compress_force = true;
259                 case Opt_compress:
260                 case Opt_compress_type:
261                         if (token == Opt_compress ||
262                             token == Opt_compress_force ||
263                             strcmp(args[0].from, "zlib") == 0) {
264                                 compress_type = "zlib";
265                                 info->compress_type = BTRFS_COMPRESS_ZLIB;
266                         } else if (strcmp(args[0].from, "lzo") == 0) {
267                                 compress_type = "lzo";
268                                 info->compress_type = BTRFS_COMPRESS_LZO;
269                         } else {
270                                 ret = -EINVAL;
271                                 goto out;
272                         }
273
274                         btrfs_set_opt(info->mount_opt, COMPRESS);
275                         if (compress_force) {
276                                 btrfs_set_opt(info->mount_opt, FORCE_COMPRESS);
277                                 pr_info("btrfs: force %s compression\n",
278                                         compress_type);
279                         } else
280                                 pr_info("btrfs: use %s compression\n",
281                                         compress_type);
282                         break;
283                 case Opt_ssd:
284                         printk(KERN_INFO "btrfs: use ssd allocation scheme\n");
285                         btrfs_set_opt(info->mount_opt, SSD);
286                         break;
287                 case Opt_ssd_spread:
288                         printk(KERN_INFO "btrfs: use spread ssd "
289                                "allocation scheme\n");
290                         btrfs_set_opt(info->mount_opt, SSD);
291                         btrfs_set_opt(info->mount_opt, SSD_SPREAD);
292                         break;
293                 case Opt_nossd:
294                         printk(KERN_INFO "btrfs: not using ssd allocation "
295                                "scheme\n");
296                         btrfs_set_opt(info->mount_opt, NOSSD);
297                         btrfs_clear_opt(info->mount_opt, SSD);
298                         btrfs_clear_opt(info->mount_opt, SSD_SPREAD);
299                         break;
300                 case Opt_nobarrier:
301                         printk(KERN_INFO "btrfs: turning off barriers\n");
302                         btrfs_set_opt(info->mount_opt, NOBARRIER);
303                         break;
304                 case Opt_thread_pool:
305                         intarg = 0;
306                         match_int(&args[0], &intarg);
307                         if (intarg) {
308                                 info->thread_pool_size = intarg;
309                                 printk(KERN_INFO "btrfs: thread pool %d\n",
310                                        info->thread_pool_size);
311                         }
312                         break;
313                 case Opt_max_inline:
314                         num = match_strdup(&args[0]);
315                         if (num) {
316                                 info->max_inline = memparse(num, NULL);
317                                 kfree(num);
318
319                                 if (info->max_inline) {
320                                         info->max_inline = max_t(u64,
321                                                 info->max_inline,
322                                                 root->sectorsize);
323                                 }
324                                 printk(KERN_INFO "btrfs: max_inline at %llu\n",
325                                         (unsigned long long)info->max_inline);
326                         }
327                         break;
328                 case Opt_alloc_start:
329                         num = match_strdup(&args[0]);
330                         if (num) {
331                                 info->alloc_start = memparse(num, NULL);
332                                 kfree(num);
333                                 printk(KERN_INFO
334                                         "btrfs: allocations start at %llu\n",
335                                         (unsigned long long)info->alloc_start);
336                         }
337                         break;
338                 case Opt_noacl:
339                         root->fs_info->sb->s_flags &= ~MS_POSIXACL;
340                         break;
341                 case Opt_notreelog:
342                         printk(KERN_INFO "btrfs: disabling tree log\n");
343                         btrfs_set_opt(info->mount_opt, NOTREELOG);
344                         break;
345                 case Opt_flushoncommit:
346                         printk(KERN_INFO "btrfs: turning on flush-on-commit\n");
347                         btrfs_set_opt(info->mount_opt, FLUSHONCOMMIT);
348                         break;
349                 case Opt_ratio:
350                         intarg = 0;
351                         match_int(&args[0], &intarg);
352                         if (intarg) {
353                                 info->metadata_ratio = intarg;
354                                 printk(KERN_INFO "btrfs: metadata ratio %d\n",
355                                        info->metadata_ratio);
356                         }
357                         break;
358                 case Opt_discard:
359                         btrfs_set_opt(info->mount_opt, DISCARD);
360                         break;
361                 case Opt_space_cache:
362                         printk(KERN_INFO "btrfs: enabling disk space caching\n");
363                         btrfs_set_opt(info->mount_opt, SPACE_CACHE);
364                         break;
365                 case Opt_inode_cache:
366                         printk(KERN_INFO "btrfs: enabling inode map caching\n");
367                         btrfs_set_opt(info->mount_opt, INODE_MAP_CACHE);
368                         break;
369                 case Opt_clear_cache:
370                         printk(KERN_INFO "btrfs: force clearing of disk cache\n");
371                         btrfs_set_opt(info->mount_opt, CLEAR_CACHE);
372                         break;
373                 case Opt_user_subvol_rm_allowed:
374                         btrfs_set_opt(info->mount_opt, USER_SUBVOL_RM_ALLOWED);
375                         break;
376                 case Opt_enospc_debug:
377                         btrfs_set_opt(info->mount_opt, ENOSPC_DEBUG);
378                         break;
379                 case Opt_defrag:
380                         printk(KERN_INFO "btrfs: enabling auto defrag");
381                         btrfs_set_opt(info->mount_opt, AUTO_DEFRAG);
382                         break;
383                 case Opt_err:
384                         printk(KERN_INFO "btrfs: unrecognized mount option "
385                                "'%s'\n", p);
386                         ret = -EINVAL;
387                         goto out;
388                 default:
389                         break;
390                 }
391         }
392 out:
393         kfree(orig);
394         return ret;
395 }
396
397 /*
398  * Parse mount options that are required early in the mount process.
399  *
400  * All other options will be parsed on much later in the mount process and
401  * only when we need to allocate a new super block.
402  */
403 static int btrfs_parse_early_options(const char *options, fmode_t flags,
404                 void *holder, char **subvol_name, u64 *subvol_objectid,
405                 u64 *subvol_rootid, struct btrfs_fs_devices **fs_devices)
406 {
407         substring_t args[MAX_OPT_ARGS];
408         char *opts, *orig, *p;
409         int error = 0;
410         int intarg;
411
412         if (!options)
413                 goto out;
414
415         /*
416          * strsep changes the string, duplicate it because parse_options
417          * gets called twice
418          */
419         opts = kstrdup(options, GFP_KERNEL);
420         if (!opts)
421                 return -ENOMEM;
422         orig = opts;
423
424         while ((p = strsep(&opts, ",")) != NULL) {
425                 int token;
426                 if (!*p)
427                         continue;
428
429                 token = match_token(p, tokens, args);
430                 switch (token) {
431                 case Opt_subvol:
432                         *subvol_name = match_strdup(&args[0]);
433                         break;
434                 case Opt_subvolid:
435                         intarg = 0;
436                         error = match_int(&args[0], &intarg);
437                         if (!error) {
438                                 /* we want the original fs_tree */
439                                 if (!intarg)
440                                         *subvol_objectid =
441                                                 BTRFS_FS_TREE_OBJECTID;
442                                 else
443                                         *subvol_objectid = intarg;
444                         }
445                         break;
446                 case Opt_subvolrootid:
447                         intarg = 0;
448                         error = match_int(&args[0], &intarg);
449                         if (!error) {
450                                 /* we want the original fs_tree */
451                                 if (!intarg)
452                                         *subvol_rootid =
453                                                 BTRFS_FS_TREE_OBJECTID;
454                                 else
455                                         *subvol_rootid = intarg;
456                         }
457                         break;
458                 case Opt_device:
459                         error = btrfs_scan_one_device(match_strdup(&args[0]),
460                                         flags, holder, fs_devices);
461                         if (error)
462                                 goto out_free_opts;
463                         break;
464                 default:
465                         break;
466                 }
467         }
468
469  out_free_opts:
470         kfree(orig);
471  out:
472         /*
473          * If no subvolume name is specified we use the default one.  Allocate
474          * a copy of the string "." here so that code later in the
475          * mount path doesn't care if it's the default volume or another one.
476          */
477         if (!*subvol_name) {
478                 *subvol_name = kstrdup(".", GFP_KERNEL);
479                 if (!*subvol_name)
480                         return -ENOMEM;
481         }
482         return error;
483 }
484
485 static struct dentry *get_default_root(struct super_block *sb,
486                                        u64 subvol_objectid)
487 {
488         struct btrfs_root *root = sb->s_fs_info;
489         struct btrfs_root *new_root;
490         struct btrfs_dir_item *di;
491         struct btrfs_path *path;
492         struct btrfs_key location;
493         struct inode *inode;
494         struct dentry *dentry;
495         u64 dir_id;
496         int new = 0;
497
498         /*
499          * We have a specific subvol we want to mount, just setup location and
500          * go look up the root.
501          */
502         if (subvol_objectid) {
503                 location.objectid = subvol_objectid;
504                 location.type = BTRFS_ROOT_ITEM_KEY;
505                 location.offset = (u64)-1;
506                 goto find_root;
507         }
508
509         path = btrfs_alloc_path();
510         if (!path)
511                 return ERR_PTR(-ENOMEM);
512         path->leave_spinning = 1;
513
514         /*
515          * Find the "default" dir item which points to the root item that we
516          * will mount by default if we haven't been given a specific subvolume
517          * to mount.
518          */
519         dir_id = btrfs_super_root_dir(&root->fs_info->super_copy);
520         di = btrfs_lookup_dir_item(NULL, root, path, dir_id, "default", 7, 0);
521         if (IS_ERR(di)) {
522                 btrfs_free_path(path);
523                 return ERR_CAST(di);
524         }
525         if (!di) {
526                 /*
527                  * Ok the default dir item isn't there.  This is weird since
528                  * it's always been there, but don't freak out, just try and
529                  * mount to root most subvolume.
530                  */
531                 btrfs_free_path(path);
532                 dir_id = BTRFS_FIRST_FREE_OBJECTID;
533                 new_root = root->fs_info->fs_root;
534                 goto setup_root;
535         }
536
537         btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
538         btrfs_free_path(path);
539
540 find_root:
541         new_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
542         if (IS_ERR(new_root))
543                 return ERR_CAST(new_root);
544
545         if (btrfs_root_refs(&new_root->root_item) == 0)
546                 return ERR_PTR(-ENOENT);
547
548         dir_id = btrfs_root_dirid(&new_root->root_item);
549 setup_root:
550         location.objectid = dir_id;
551         location.type = BTRFS_INODE_ITEM_KEY;
552         location.offset = 0;
553
554         inode = btrfs_iget(sb, &location, new_root, &new);
555         if (IS_ERR(inode))
556                 return ERR_CAST(inode);
557
558         /*
559          * If we're just mounting the root most subvol put the inode and return
560          * a reference to the dentry.  We will have already gotten a reference
561          * to the inode in btrfs_fill_super so we're good to go.
562          */
563         if (!new && sb->s_root->d_inode == inode) {
564                 iput(inode);
565                 return dget(sb->s_root);
566         }
567
568         if (new) {
569                 const struct qstr name = { .name = "/", .len = 1 };
570
571                 /*
572                  * New inode, we need to make the dentry a sibling of s_root so
573                  * everything gets cleaned up properly on unmount.
574                  */
575                 dentry = d_alloc(sb->s_root, &name);
576                 if (!dentry) {
577                         iput(inode);
578                         return ERR_PTR(-ENOMEM);
579                 }
580                 d_splice_alias(inode, dentry);
581         } else {
582                 /*
583                  * We found the inode in cache, just find a dentry for it and
584                  * put the reference to the inode we just got.
585                  */
586                 dentry = d_find_alias(inode);
587                 iput(inode);
588         }
589
590         return dentry;
591 }
592
593 static int btrfs_fill_super(struct super_block *sb,
594                             struct btrfs_fs_devices *fs_devices,
595                             void *data, int silent)
596 {
597         struct inode *inode;
598         struct dentry *root_dentry;
599         struct btrfs_root *tree_root;
600         struct btrfs_key key;
601         int err;
602
603         sb->s_maxbytes = MAX_LFS_FILESIZE;
604         sb->s_magic = BTRFS_SUPER_MAGIC;
605         sb->s_op = &btrfs_super_ops;
606         sb->s_d_op = &btrfs_dentry_operations;
607         sb->s_export_op = &btrfs_export_ops;
608         sb->s_xattr = btrfs_xattr_handlers;
609         sb->s_time_gran = 1;
610 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
611         sb->s_flags |= MS_POSIXACL;
612 #endif
613
614         tree_root = open_ctree(sb, fs_devices, (char *)data);
615
616         if (IS_ERR(tree_root)) {
617                 printk("btrfs: open_ctree failed\n");
618                 return PTR_ERR(tree_root);
619         }
620         sb->s_fs_info = tree_root;
621
622         key.objectid = BTRFS_FIRST_FREE_OBJECTID;
623         key.type = BTRFS_INODE_ITEM_KEY;
624         key.offset = 0;
625         inode = btrfs_iget(sb, &key, tree_root->fs_info->fs_root, NULL);
626         if (IS_ERR(inode)) {
627                 err = PTR_ERR(inode);
628                 goto fail_close;
629         }
630
631         root_dentry = d_alloc_root(inode);
632         if (!root_dentry) {
633                 iput(inode);
634                 err = -ENOMEM;
635                 goto fail_close;
636         }
637
638         sb->s_root = root_dentry;
639
640         save_mount_options(sb, data);
641         return 0;
642
643 fail_close:
644         close_ctree(tree_root);
645         return err;
646 }
647
648 int btrfs_sync_fs(struct super_block *sb, int wait)
649 {
650         struct btrfs_trans_handle *trans;
651         struct btrfs_root *root = btrfs_sb(sb);
652         int ret;
653
654         trace_btrfs_sync_fs(wait);
655
656         if (!wait) {
657                 filemap_flush(root->fs_info->btree_inode->i_mapping);
658                 return 0;
659         }
660
661         btrfs_start_delalloc_inodes(root, 0);
662         btrfs_wait_ordered_extents(root, 0, 0);
663
664         trans = btrfs_start_transaction(root, 0);
665         if (IS_ERR(trans))
666                 return PTR_ERR(trans);
667         ret = btrfs_commit_transaction(trans, root);
668         return ret;
669 }
670
671 static int btrfs_show_options(struct seq_file *seq, struct vfsmount *vfs)
672 {
673         struct btrfs_root *root = btrfs_sb(vfs->mnt_sb);
674         struct btrfs_fs_info *info = root->fs_info;
675         char *compress_type;
676
677         if (btrfs_test_opt(root, DEGRADED))
678                 seq_puts(seq, ",degraded");
679         if (btrfs_test_opt(root, NODATASUM))
680                 seq_puts(seq, ",nodatasum");
681         if (btrfs_test_opt(root, NODATACOW))
682                 seq_puts(seq, ",nodatacow");
683         if (btrfs_test_opt(root, NOBARRIER))
684                 seq_puts(seq, ",nobarrier");
685         if (info->max_inline != 8192 * 1024)
686                 seq_printf(seq, ",max_inline=%llu",
687                            (unsigned long long)info->max_inline);
688         if (info->alloc_start != 0)
689                 seq_printf(seq, ",alloc_start=%llu",
690                            (unsigned long long)info->alloc_start);
691         if (info->thread_pool_size !=  min_t(unsigned long,
692                                              num_online_cpus() + 2, 8))
693                 seq_printf(seq, ",thread_pool=%d", info->thread_pool_size);
694         if (btrfs_test_opt(root, COMPRESS)) {
695                 if (info->compress_type == BTRFS_COMPRESS_ZLIB)
696                         compress_type = "zlib";
697                 else
698                         compress_type = "lzo";
699                 if (btrfs_test_opt(root, FORCE_COMPRESS))
700                         seq_printf(seq, ",compress-force=%s", compress_type);
701                 else
702                         seq_printf(seq, ",compress=%s", compress_type);
703         }
704         if (btrfs_test_opt(root, NOSSD))
705                 seq_puts(seq, ",nossd");
706         if (btrfs_test_opt(root, SSD_SPREAD))
707                 seq_puts(seq, ",ssd_spread");
708         else if (btrfs_test_opt(root, SSD))
709                 seq_puts(seq, ",ssd");
710         if (btrfs_test_opt(root, NOTREELOG))
711                 seq_puts(seq, ",notreelog");
712         if (btrfs_test_opt(root, FLUSHONCOMMIT))
713                 seq_puts(seq, ",flushoncommit");
714         if (btrfs_test_opt(root, DISCARD))
715                 seq_puts(seq, ",discard");
716         if (!(root->fs_info->sb->s_flags & MS_POSIXACL))
717                 seq_puts(seq, ",noacl");
718         if (btrfs_test_opt(root, SPACE_CACHE))
719                 seq_puts(seq, ",space_cache");
720         if (btrfs_test_opt(root, CLEAR_CACHE))
721                 seq_puts(seq, ",clear_cache");
722         if (btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
723                 seq_puts(seq, ",user_subvol_rm_allowed");
724         if (btrfs_test_opt(root, ENOSPC_DEBUG))
725                 seq_puts(seq, ",enospc_debug");
726         if (btrfs_test_opt(root, AUTO_DEFRAG))
727                 seq_puts(seq, ",autodefrag");
728         if (btrfs_test_opt(root, INODE_MAP_CACHE))
729                 seq_puts(seq, ",inode_cache");
730         return 0;
731 }
732
733 static int btrfs_test_super(struct super_block *s, void *data)
734 {
735         struct btrfs_root *test_root = data;
736         struct btrfs_root *root = btrfs_sb(s);
737
738         /*
739          * If this super block is going away, return false as it
740          * can't match as an existing super block.
741          */
742         if (!atomic_read(&s->s_active))
743                 return 0;
744         return root->fs_info->fs_devices == test_root->fs_info->fs_devices;
745 }
746
747 static int btrfs_set_super(struct super_block *s, void *data)
748 {
749         s->s_fs_info = data;
750
751         return set_anon_super(s, data);
752 }
753
754
755 /*
756  * Find a superblock for the given device / mount point.
757  *
758  * Note:  This is based on get_sb_bdev from fs/super.c with a few additions
759  *        for multiple device setup.  Make sure to keep it in sync.
760  */
761 static struct dentry *btrfs_mount(struct file_system_type *fs_type, int flags,
762                 const char *device_name, void *data)
763 {
764         struct block_device *bdev = NULL;
765         struct super_block *s;
766         struct dentry *root;
767         struct btrfs_fs_devices *fs_devices = NULL;
768         struct btrfs_root *tree_root = NULL;
769         struct btrfs_fs_info *fs_info = NULL;
770         fmode_t mode = FMODE_READ;
771         char *subvol_name = NULL;
772         u64 subvol_objectid = 0;
773         u64 subvol_rootid = 0;
774         int error = 0;
775
776         if (!(flags & MS_RDONLY))
777                 mode |= FMODE_WRITE;
778
779         error = btrfs_parse_early_options(data, mode, fs_type,
780                                           &subvol_name, &subvol_objectid,
781                                           &subvol_rootid, &fs_devices);
782         if (error)
783                 return ERR_PTR(error);
784
785         error = btrfs_scan_one_device(device_name, mode, fs_type, &fs_devices);
786         if (error)
787                 goto error_free_subvol_name;
788
789         error = btrfs_open_devices(fs_devices, mode, fs_type);
790         if (error)
791                 goto error_free_subvol_name;
792
793         if (!(flags & MS_RDONLY) && fs_devices->rw_devices == 0) {
794                 error = -EACCES;
795                 goto error_close_devices;
796         }
797
798         /*
799          * Setup a dummy root and fs_info for test/set super.  This is because
800          * we don't actually fill this stuff out until open_ctree, but we need
801          * it for searching for existing supers, so this lets us do that and
802          * then open_ctree will properly initialize everything later.
803          */
804         fs_info = kzalloc(sizeof(struct btrfs_fs_info), GFP_NOFS);
805         tree_root = kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
806         if (!fs_info || !tree_root) {
807                 error = -ENOMEM;
808                 goto error_close_devices;
809         }
810         fs_info->tree_root = tree_root;
811         fs_info->fs_devices = fs_devices;
812         tree_root->fs_info = fs_info;
813
814         bdev = fs_devices->latest_bdev;
815         s = sget(fs_type, btrfs_test_super, btrfs_set_super, tree_root);
816         if (IS_ERR(s))
817                 goto error_s;
818
819         if (s->s_root) {
820                 if ((flags ^ s->s_flags) & MS_RDONLY) {
821                         deactivate_locked_super(s);
822                         error = -EBUSY;
823                         goto error_close_devices;
824                 }
825
826                 btrfs_close_devices(fs_devices);
827                 kfree(fs_info);
828                 kfree(tree_root);
829         } else {
830                 char b[BDEVNAME_SIZE];
831
832                 s->s_flags = flags;
833                 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
834                 error = btrfs_fill_super(s, fs_devices, data,
835                                          flags & MS_SILENT ? 1 : 0);
836                 if (error) {
837                         deactivate_locked_super(s);
838                         goto error_free_subvol_name;
839                 }
840
841                 btrfs_sb(s)->fs_info->bdev_holder = fs_type;
842                 s->s_flags |= MS_ACTIVE;
843         }
844
845         /* if they gave us a subvolume name bind mount into that */
846         if (strcmp(subvol_name, ".")) {
847                 struct dentry *new_root;
848
849                 root = get_default_root(s, subvol_rootid);
850                 if (IS_ERR(root)) {
851                         error = PTR_ERR(root);
852                         deactivate_locked_super(s);
853                         goto error_free_subvol_name;
854                 }
855
856                 mutex_lock(&root->d_inode->i_mutex);
857                 new_root = lookup_one_len(subvol_name, root,
858                                       strlen(subvol_name));
859                 mutex_unlock(&root->d_inode->i_mutex);
860
861                 if (IS_ERR(new_root)) {
862                         dput(root);
863                         deactivate_locked_super(s);
864                         error = PTR_ERR(new_root);
865                         goto error_free_subvol_name;
866                 }
867                 if (!new_root->d_inode) {
868                         dput(root);
869                         dput(new_root);
870                         deactivate_locked_super(s);
871                         error = -ENXIO;
872                         goto error_free_subvol_name;
873                 }
874                 dput(root);
875                 root = new_root;
876         } else {
877                 root = get_default_root(s, subvol_objectid);
878                 if (IS_ERR(root)) {
879                         error = PTR_ERR(root);
880                         deactivate_locked_super(s);
881                         goto error_free_subvol_name;
882                 }
883         }
884
885         kfree(subvol_name);
886         return root;
887
888 error_s:
889         error = PTR_ERR(s);
890 error_close_devices:
891         btrfs_close_devices(fs_devices);
892         kfree(fs_info);
893         kfree(tree_root);
894 error_free_subvol_name:
895         kfree(subvol_name);
896         return ERR_PTR(error);
897 }
898
899 static int btrfs_remount(struct super_block *sb, int *flags, char *data)
900 {
901         struct btrfs_root *root = btrfs_sb(sb);
902         int ret;
903
904         ret = btrfs_parse_options(root, data);
905         if (ret)
906                 return -EINVAL;
907
908         if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
909                 return 0;
910
911         if (*flags & MS_RDONLY) {
912                 sb->s_flags |= MS_RDONLY;
913
914                 ret =  btrfs_commit_super(root);
915                 WARN_ON(ret);
916         } else {
917                 if (root->fs_info->fs_devices->rw_devices == 0)
918                         return -EACCES;
919
920                 if (btrfs_super_log_root(&root->fs_info->super_copy) != 0)
921                         return -EINVAL;
922
923                 ret = btrfs_cleanup_fs_roots(root->fs_info);
924                 WARN_ON(ret);
925
926                 /* recover relocation */
927                 ret = btrfs_recover_relocation(root);
928                 WARN_ON(ret);
929
930                 sb->s_flags &= ~MS_RDONLY;
931         }
932
933         return 0;
934 }
935
936 /* Used to sort the devices by max_avail(descending sort) */
937 static int btrfs_cmp_device_free_bytes(const void *dev_info1,
938                                        const void *dev_info2)
939 {
940         if (((struct btrfs_device_info *)dev_info1)->max_avail >
941             ((struct btrfs_device_info *)dev_info2)->max_avail)
942                 return -1;
943         else if (((struct btrfs_device_info *)dev_info1)->max_avail <
944                  ((struct btrfs_device_info *)dev_info2)->max_avail)
945                 return 1;
946         else
947         return 0;
948 }
949
950 /*
951  * sort the devices by max_avail, in which max free extent size of each device
952  * is stored.(Descending Sort)
953  */
954 static inline void btrfs_descending_sort_devices(
955                                         struct btrfs_device_info *devices,
956                                         size_t nr_devices)
957 {
958         sort(devices, nr_devices, sizeof(struct btrfs_device_info),
959              btrfs_cmp_device_free_bytes, NULL);
960 }
961
962 /*
963  * The helper to calc the free space on the devices that can be used to store
964  * file data.
965  */
966 static int btrfs_calc_avail_data_space(struct btrfs_root *root, u64 *free_bytes)
967 {
968         struct btrfs_fs_info *fs_info = root->fs_info;
969         struct btrfs_device_info *devices_info;
970         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
971         struct btrfs_device *device;
972         u64 skip_space;
973         u64 type;
974         u64 avail_space;
975         u64 used_space;
976         u64 min_stripe_size;
977         int min_stripes = 1;
978         int i = 0, nr_devices;
979         int ret;
980
981         nr_devices = fs_info->fs_devices->rw_devices;
982         BUG_ON(!nr_devices);
983
984         devices_info = kmalloc(sizeof(*devices_info) * nr_devices,
985                                GFP_NOFS);
986         if (!devices_info)
987                 return -ENOMEM;
988
989         /* calc min stripe number for data space alloction */
990         type = btrfs_get_alloc_profile(root, 1);
991         if (type & BTRFS_BLOCK_GROUP_RAID0)
992                 min_stripes = 2;
993         else if (type & BTRFS_BLOCK_GROUP_RAID1)
994                 min_stripes = 2;
995         else if (type & BTRFS_BLOCK_GROUP_RAID10)
996                 min_stripes = 4;
997
998         if (type & BTRFS_BLOCK_GROUP_DUP)
999                 min_stripe_size = 2 * BTRFS_STRIPE_LEN;
1000         else
1001                 min_stripe_size = BTRFS_STRIPE_LEN;
1002
1003         list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
1004                 if (!device->in_fs_metadata)
1005                         continue;
1006
1007                 avail_space = device->total_bytes - device->bytes_used;
1008
1009                 /* align with stripe_len */
1010                 do_div(avail_space, BTRFS_STRIPE_LEN);
1011                 avail_space *= BTRFS_STRIPE_LEN;
1012
1013                 /*
1014                  * In order to avoid overwritting the superblock on the drive,
1015                  * btrfs starts at an offset of at least 1MB when doing chunk
1016                  * allocation.
1017                  */
1018                 skip_space = 1024 * 1024;
1019
1020                 /* user can set the offset in fs_info->alloc_start. */
1021                 if (fs_info->alloc_start + BTRFS_STRIPE_LEN <=
1022                     device->total_bytes)
1023                         skip_space = max(fs_info->alloc_start, skip_space);
1024
1025                 /*
1026                  * btrfs can not use the free space in [0, skip_space - 1],
1027                  * we must subtract it from the total. In order to implement
1028                  * it, we account the used space in this range first.
1029                  */
1030                 ret = btrfs_account_dev_extents_size(device, 0, skip_space - 1,
1031                                                      &used_space);
1032                 if (ret) {
1033                         kfree(devices_info);
1034                         return ret;
1035                 }
1036
1037                 /* calc the free space in [0, skip_space - 1] */
1038                 skip_space -= used_space;
1039
1040                 /*
1041                  * we can use the free space in [0, skip_space - 1], subtract
1042                  * it from the total.
1043                  */
1044                 if (avail_space && avail_space >= skip_space)
1045                         avail_space -= skip_space;
1046                 else
1047                         avail_space = 0;
1048
1049                 if (avail_space < min_stripe_size)
1050                         continue;
1051
1052                 devices_info[i].dev = device;
1053                 devices_info[i].max_avail = avail_space;
1054
1055                 i++;
1056         }
1057
1058         nr_devices = i;
1059
1060         btrfs_descending_sort_devices(devices_info, nr_devices);
1061
1062         i = nr_devices - 1;
1063         avail_space = 0;
1064         while (nr_devices >= min_stripes) {
1065                 if (devices_info[i].max_avail >= min_stripe_size) {
1066                         int j;
1067                         u64 alloc_size;
1068
1069                         avail_space += devices_info[i].max_avail * min_stripes;
1070                         alloc_size = devices_info[i].max_avail;
1071                         for (j = i + 1 - min_stripes; j <= i; j++)
1072                                 devices_info[j].max_avail -= alloc_size;
1073                 }
1074                 i--;
1075                 nr_devices--;
1076         }
1077
1078         kfree(devices_info);
1079         *free_bytes = avail_space;
1080         return 0;
1081 }
1082
1083 static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
1084 {
1085         struct btrfs_root *root = btrfs_sb(dentry->d_sb);
1086         struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
1087         struct list_head *head = &root->fs_info->space_info;
1088         struct btrfs_space_info *found;
1089         u64 total_used = 0;
1090         u64 total_free_data = 0;
1091         int bits = dentry->d_sb->s_blocksize_bits;
1092         __be32 *fsid = (__be32 *)root->fs_info->fsid;
1093         int ret;
1094
1095         /* holding chunk_muext to avoid allocating new chunks */
1096         mutex_lock(&root->fs_info->chunk_mutex);
1097         rcu_read_lock();
1098         list_for_each_entry_rcu(found, head, list) {
1099                 if (found->flags & BTRFS_BLOCK_GROUP_DATA) {
1100                         total_free_data += found->disk_total - found->disk_used;
1101                         total_free_data -=
1102                                 btrfs_account_ro_block_groups_free_space(found);
1103                 }
1104
1105                 total_used += found->disk_used;
1106         }
1107         rcu_read_unlock();
1108
1109         buf->f_namelen = BTRFS_NAME_LEN;
1110         buf->f_blocks = btrfs_super_total_bytes(disk_super) >> bits;
1111         buf->f_bfree = buf->f_blocks - (total_used >> bits);
1112         buf->f_bsize = dentry->d_sb->s_blocksize;
1113         buf->f_type = BTRFS_SUPER_MAGIC;
1114         buf->f_bavail = total_free_data;
1115         ret = btrfs_calc_avail_data_space(root, &total_free_data);
1116         if (ret) {
1117                 mutex_unlock(&root->fs_info->chunk_mutex);
1118                 return ret;
1119         }
1120         buf->f_bavail += total_free_data;
1121         buf->f_bavail = buf->f_bavail >> bits;
1122         mutex_unlock(&root->fs_info->chunk_mutex);
1123
1124         /* We treat it as constant endianness (it doesn't matter _which_)
1125            because we want the fsid to come out the same whether mounted
1126            on a big-endian or little-endian host */
1127         buf->f_fsid.val[0] = be32_to_cpu(fsid[0]) ^ be32_to_cpu(fsid[2]);
1128         buf->f_fsid.val[1] = be32_to_cpu(fsid[1]) ^ be32_to_cpu(fsid[3]);
1129         /* Mask in the root object ID too, to disambiguate subvols */
1130         buf->f_fsid.val[0] ^= BTRFS_I(dentry->d_inode)->root->objectid >> 32;
1131         buf->f_fsid.val[1] ^= BTRFS_I(dentry->d_inode)->root->objectid;
1132
1133         return 0;
1134 }
1135
1136 static struct file_system_type btrfs_fs_type = {
1137         .owner          = THIS_MODULE,
1138         .name           = "btrfs",
1139         .mount          = btrfs_mount,
1140         .kill_sb        = kill_anon_super,
1141         .fs_flags       = FS_REQUIRES_DEV,
1142 };
1143
1144 /*
1145  * used by btrfsctl to scan devices when no FS is mounted
1146  */
1147 static long btrfs_control_ioctl(struct file *file, unsigned int cmd,
1148                                 unsigned long arg)
1149 {
1150         struct btrfs_ioctl_vol_args *vol;
1151         struct btrfs_fs_devices *fs_devices;
1152         int ret = -ENOTTY;
1153
1154         if (!capable(CAP_SYS_ADMIN))
1155                 return -EPERM;
1156
1157         vol = memdup_user((void __user *)arg, sizeof(*vol));
1158         if (IS_ERR(vol))
1159                 return PTR_ERR(vol);
1160
1161         switch (cmd) {
1162         case BTRFS_IOC_SCAN_DEV:
1163                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
1164                                             &btrfs_fs_type, &fs_devices);
1165                 break;
1166         }
1167
1168         kfree(vol);
1169         return ret;
1170 }
1171
1172 static int btrfs_freeze(struct super_block *sb)
1173 {
1174         struct btrfs_root *root = btrfs_sb(sb);
1175         mutex_lock(&root->fs_info->transaction_kthread_mutex);
1176         mutex_lock(&root->fs_info->cleaner_mutex);
1177         return 0;
1178 }
1179
1180 static int btrfs_unfreeze(struct super_block *sb)
1181 {
1182         struct btrfs_root *root = btrfs_sb(sb);
1183         mutex_unlock(&root->fs_info->cleaner_mutex);
1184         mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1185         return 0;
1186 }
1187
1188 static const struct super_operations btrfs_super_ops = {
1189         .drop_inode     = btrfs_drop_inode,
1190         .evict_inode    = btrfs_evict_inode,
1191         .put_super      = btrfs_put_super,
1192         .sync_fs        = btrfs_sync_fs,
1193         .show_options   = btrfs_show_options,
1194         .write_inode    = btrfs_write_inode,
1195         .dirty_inode    = btrfs_dirty_inode,
1196         .alloc_inode    = btrfs_alloc_inode,
1197         .destroy_inode  = btrfs_destroy_inode,
1198         .statfs         = btrfs_statfs,
1199         .remount_fs     = btrfs_remount,
1200         .freeze_fs      = btrfs_freeze,
1201         .unfreeze_fs    = btrfs_unfreeze,
1202 };
1203
1204 static const struct file_operations btrfs_ctl_fops = {
1205         .unlocked_ioctl  = btrfs_control_ioctl,
1206         .compat_ioctl = btrfs_control_ioctl,
1207         .owner   = THIS_MODULE,
1208         .llseek = noop_llseek,
1209 };
1210
1211 static struct miscdevice btrfs_misc = {
1212         .minor          = BTRFS_MINOR,
1213         .name           = "btrfs-control",
1214         .fops           = &btrfs_ctl_fops
1215 };
1216
1217 MODULE_ALIAS_MISCDEV(BTRFS_MINOR);
1218 MODULE_ALIAS("devname:btrfs-control");
1219
1220 static int btrfs_interface_init(void)
1221 {
1222         return misc_register(&btrfs_misc);
1223 }
1224
1225 static void btrfs_interface_exit(void)
1226 {
1227         if (misc_deregister(&btrfs_misc) < 0)
1228                 printk(KERN_INFO "misc_deregister failed for control device");
1229 }
1230
1231 static int __init init_btrfs_fs(void)
1232 {
1233         int err;
1234
1235         err = btrfs_init_sysfs();
1236         if (err)
1237                 return err;
1238
1239         err = btrfs_init_compress();
1240         if (err)
1241                 goto free_sysfs;
1242
1243         err = btrfs_init_cachep();
1244         if (err)
1245                 goto free_compress;
1246
1247         err = extent_io_init();
1248         if (err)
1249                 goto free_cachep;
1250
1251         err = extent_map_init();
1252         if (err)
1253                 goto free_extent_io;
1254
1255         err = btrfs_delayed_inode_init();
1256         if (err)
1257                 goto free_extent_map;
1258
1259         err = btrfs_interface_init();
1260         if (err)
1261                 goto free_delayed_inode;
1262
1263         err = register_filesystem(&btrfs_fs_type);
1264         if (err)
1265                 goto unregister_ioctl;
1266
1267         printk(KERN_INFO "%s loaded\n", BTRFS_BUILD_VERSION);
1268         return 0;
1269
1270 unregister_ioctl:
1271         btrfs_interface_exit();
1272 free_delayed_inode:
1273         btrfs_delayed_inode_exit();
1274 free_extent_map:
1275         extent_map_exit();
1276 free_extent_io:
1277         extent_io_exit();
1278 free_cachep:
1279         btrfs_destroy_cachep();
1280 free_compress:
1281         btrfs_exit_compress();
1282 free_sysfs:
1283         btrfs_exit_sysfs();
1284         return err;
1285 }
1286
1287 static void __exit exit_btrfs_fs(void)
1288 {
1289         btrfs_destroy_cachep();
1290         btrfs_delayed_inode_exit();
1291         extent_map_exit();
1292         extent_io_exit();
1293         btrfs_interface_exit();
1294         unregister_filesystem(&btrfs_fs_type);
1295         btrfs_exit_sysfs();
1296         btrfs_cleanup_fs_uuids();
1297         btrfs_exit_compress();
1298 }
1299
1300 module_init(init_btrfs_fs)
1301 module_exit(exit_btrfs_fs)
1302
1303 MODULE_LICENSE("GPL");