Btrfs: Run igrab on data=ordered inodes to prevent deadlocks during writeout
[safe/jmp/linux-2.6] / 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/string.h>
28 #include <linux/smp_lock.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 "ctree.h"
40 #include "disk-io.h"
41 #include "transaction.h"
42 #include "btrfs_inode.h"
43 #include "ioctl.h"
44 #include "print-tree.h"
45 #include "xattr.h"
46
47 #define BTRFS_SUPER_MAGIC 0x9123683E
48
49 static struct super_operations btrfs_super_ops;
50
51 static void btrfs_put_super (struct super_block * sb)
52 {
53         struct btrfs_root *root = btrfs_sb(sb);
54         struct btrfs_fs_info *fs = root->fs_info;
55         int ret;
56
57         ret = close_ctree(root);
58         if (ret) {
59                 printk("close ctree returns %d\n", ret);
60         }
61         btrfs_sysfs_del_super(fs);
62         sb->s_fs_info = NULL;
63 }
64
65 enum {
66         Opt_subvol, Opt_nodatasum, Opt_nodatacow, Opt_max_extent,
67         Opt_alloc_start, Opt_nobarrier, Opt_err,
68 };
69
70 static match_table_t tokens = {
71         {Opt_subvol, "subvol=%s"},
72         {Opt_nodatasum, "nodatasum"},
73         {Opt_nodatacow, "nodatacow"},
74         {Opt_nobarrier, "nobarrier"},
75         {Opt_max_extent, "max_extent=%s"},
76         {Opt_alloc_start, "alloc_start=%s"},
77         {Opt_err, NULL}
78 };
79
80 u64 btrfs_parse_size(char *str)
81 {
82         u64 res;
83         int mult = 1;
84         char *end;
85         char last;
86
87         res = simple_strtoul(str, &end, 10);
88
89         last = end[0];
90         if (isalpha(last)) {
91                 last = tolower(last);
92                 switch (last) {
93                 case 'g':
94                         mult *= 1024;
95                 case 'm':
96                         mult *= 1024;
97                 case 'k':
98                         mult *= 1024;
99                 }
100                 res = res * mult;
101         }
102         return res;
103 }
104
105 static int parse_options (char * options,
106                           struct btrfs_root *root,
107                           char **subvol_name)
108 {
109         char * p;
110         struct btrfs_fs_info *info = NULL;
111         substring_t args[MAX_OPT_ARGS];
112
113         if (!options)
114                 return 1;
115
116         /*
117          * strsep changes the string, duplicate it because parse_options
118          * gets called twice
119          */
120         options = kstrdup(options, GFP_NOFS);
121         if (!options)
122                 return -ENOMEM;
123
124         if (root)
125                 info = root->fs_info;
126
127         while ((p = strsep (&options, ",")) != NULL) {
128                 int token;
129                 if (!*p)
130                         continue;
131
132                 token = match_token(p, tokens, args);
133                 switch (token) {
134                 case Opt_subvol:
135                         if (subvol_name) {
136                                 *subvol_name = match_strdup(&args[0]);
137                         }
138                         break;
139                 case Opt_nodatasum:
140                         if (info) {
141                                 printk("btrfs: setting nodatacsum\n");
142                                 btrfs_set_opt(info->mount_opt, NODATASUM);
143                         }
144                         break;
145                 case Opt_nodatacow:
146                         if (info) {
147                                 printk("btrfs: setting nodatacow\n");
148                                 btrfs_set_opt(info->mount_opt, NODATACOW);
149                                 btrfs_set_opt(info->mount_opt, NODATASUM);
150                         }
151                         break;
152                 case Opt_nobarrier:
153                         if (info) {
154                                 printk("btrfs: turning off barriers\n");
155                                 btrfs_set_opt(info->mount_opt, NOBARRIER);
156                         }
157                         break;
158                 case Opt_max_extent:
159                         if (info) {
160                                 char *num = match_strdup(&args[0]);
161                                 if (num) {
162                                         info->max_extent =
163                                                 btrfs_parse_size(num);
164                                         kfree(num);
165
166                                         info->max_extent = max_t(u64,
167                                                          info->max_extent,
168                                                          root->sectorsize);
169                                         printk("btrfs: max_extent at %Lu\n",
170                                                info->max_extent);
171                                 }
172                         }
173                         break;
174                 case Opt_alloc_start:
175                         if (info) {
176                                 char *num = match_strdup(&args[0]);
177                                 if (num) {
178                                         info->alloc_start =
179                                                 btrfs_parse_size(num);
180                                         kfree(num);
181                                         printk("btrfs: allocations start at "
182                                                "%Lu\n", info->alloc_start);
183                                 }
184                         }
185                         break;
186                 default:
187                         break;
188                 }
189         }
190         kfree(options);
191         return 1;
192 }
193
194 static int btrfs_fill_super(struct super_block * sb, void * data, int silent)
195 {
196         struct inode * inode;
197         struct dentry * root_dentry;
198         struct btrfs_super_block *disk_super;
199         struct btrfs_root *tree_root;
200         struct btrfs_inode *bi;
201         int err;
202
203         sb->s_maxbytes = MAX_LFS_FILESIZE;
204         sb->s_magic = BTRFS_SUPER_MAGIC;
205         sb->s_op = &btrfs_super_ops;
206         sb->s_xattr = btrfs_xattr_handlers;
207         sb->s_time_gran = 1;
208
209         tree_root = open_ctree(sb);
210
211         if (!tree_root || IS_ERR(tree_root)) {
212                 printk("btrfs: open_ctree failed\n");
213                 return -EIO;
214         }
215         sb->s_fs_info = tree_root;
216         disk_super = &tree_root->fs_info->super_copy;
217         inode = btrfs_iget_locked(sb, btrfs_super_root_dir(disk_super),
218                                   tree_root);
219         bi = BTRFS_I(inode);
220         bi->location.objectid = inode->i_ino;
221         bi->location.offset = 0;
222         bi->root = tree_root;
223
224         btrfs_set_key_type(&bi->location, BTRFS_INODE_ITEM_KEY);
225
226         if (!inode) {
227                 err = -ENOMEM;
228                 goto fail_close;
229         }
230         if (inode->i_state & I_NEW) {
231                 btrfs_read_locked_inode(inode);
232                 unlock_new_inode(inode);
233         }
234
235         root_dentry = d_alloc_root(inode);
236         if (!root_dentry) {
237                 iput(inode);
238                 err = -ENOMEM;
239                 goto fail_close;
240         }
241
242         parse_options((char *)data, tree_root, NULL);
243
244         /* this does the super kobj at the same time */
245         err = btrfs_sysfs_add_super(tree_root->fs_info);
246         if (err)
247                 goto fail_close;
248
249         sb->s_root = root_dentry;
250         btrfs_transaction_queue_work(tree_root, HZ * 30);
251         return 0;
252
253 fail_close:
254         close_ctree(tree_root);
255         return err;
256 }
257
258 static int btrfs_sync_fs(struct super_block *sb, int wait)
259 {
260         struct btrfs_trans_handle *trans;
261         struct btrfs_root *root;
262         int ret;
263         root = btrfs_sb(sb);
264
265         sb->s_dirt = 0;
266         if (!wait) {
267                 filemap_flush(root->fs_info->btree_inode->i_mapping);
268                 return 0;
269         }
270         btrfs_clean_old_snapshots(root);
271         mutex_lock(&root->fs_info->fs_mutex);
272         btrfs_defrag_dirty_roots(root->fs_info);
273         trans = btrfs_start_transaction(root, 1);
274         ret = btrfs_commit_transaction(trans, root);
275         sb->s_dirt = 0;
276         mutex_unlock(&root->fs_info->fs_mutex);
277         return ret;
278 }
279
280 static void btrfs_write_super(struct super_block *sb)
281 {
282         sb->s_dirt = 0;
283 }
284
285 /*
286  * This is almost a copy of get_sb_bdev in fs/super.c.
287  * We need the local copy to allow direct mounting of
288  * subvolumes, but this could be easily integrated back
289  * into the generic version.  --hch
290  */
291
292 /* start copy & paste */
293 static int set_bdev_super(struct super_block *s, void *data)
294 {
295         s->s_bdev = data;
296         s->s_dev = s->s_bdev->bd_dev;
297         return 0;
298 }
299
300 static int test_bdev_super(struct super_block *s, void *data)
301 {
302         return (void *)s->s_bdev == data;
303 }
304
305 int btrfs_get_sb_bdev(struct file_system_type *fs_type,
306         int flags, const char *dev_name, void *data,
307         int (*fill_super)(struct super_block *, void *, int),
308         struct vfsmount *mnt, const char *subvol)
309 {
310         struct block_device *bdev = NULL;
311         struct super_block *s;
312         struct dentry *root;
313         int error = 0;
314
315         bdev = open_bdev_excl(dev_name, flags, fs_type);
316         if (IS_ERR(bdev))
317                 return PTR_ERR(bdev);
318
319         /*
320          * once the super is inserted into the list by sget, s_umount
321          * will protect the lockfs code from trying to start a snapshot
322          * while we are mounting
323          */
324         down(&bdev->bd_mount_sem);
325         s = sget(fs_type, test_bdev_super, set_bdev_super, bdev);
326         up(&bdev->bd_mount_sem);
327         if (IS_ERR(s))
328                 goto error_s;
329
330         if (s->s_root) {
331                 if ((flags ^ s->s_flags) & MS_RDONLY) {
332                         up_write(&s->s_umount);
333                         deactivate_super(s);
334                         error = -EBUSY;
335                         goto error_bdev;
336                 }
337
338                 close_bdev_excl(bdev);
339         } else {
340                 char b[BDEVNAME_SIZE];
341
342                 s->s_flags = flags;
343                 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
344                 sb_set_blocksize(s, block_size(bdev));
345                 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
346                 if (error) {
347                         up_write(&s->s_umount);
348                         deactivate_super(s);
349                         goto error;
350                 }
351
352                 s->s_flags |= MS_ACTIVE;
353         }
354
355         if (subvol) {
356                 root = lookup_one_len(subvol, s->s_root, strlen(subvol));
357                 if (IS_ERR(root)) {
358                         up_write(&s->s_umount);
359                         deactivate_super(s);
360                         error = PTR_ERR(root);
361                         goto error;
362                 }
363                 if (!root->d_inode) {
364                         dput(root);
365                         up_write(&s->s_umount);
366                         deactivate_super(s);
367                         error = -ENXIO;
368                         goto error;
369                 }
370         } else {
371                 root = dget(s->s_root);
372         }
373
374         mnt->mnt_sb = s;
375         mnt->mnt_root = root;
376         return 0;
377
378 error_s:
379         error = PTR_ERR(s);
380 error_bdev:
381         close_bdev_excl(bdev);
382 error:
383         return error;
384 }
385 /* end copy & paste */
386
387 static int btrfs_get_sb(struct file_system_type *fs_type,
388         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
389 {
390         int ret;
391         char *subvol_name = NULL;
392
393         parse_options((char *)data, NULL, &subvol_name);
394         ret = btrfs_get_sb_bdev(fs_type, flags, dev_name, data,
395                         btrfs_fill_super, mnt,
396                         subvol_name ? subvol_name : "default");
397         if (subvol_name)
398                 kfree(subvol_name);
399         return ret;
400 }
401
402 static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
403 {
404         struct btrfs_root *root = btrfs_sb(dentry->d_sb);
405         struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
406         int bits = dentry->d_sb->s_blocksize_bits;
407
408         buf->f_namelen = BTRFS_NAME_LEN;
409         buf->f_blocks = btrfs_super_total_bytes(disk_super) >> bits;
410         buf->f_bfree = buf->f_blocks -
411                 (btrfs_super_bytes_used(disk_super) >> bits);
412         buf->f_bavail = buf->f_bfree;
413         buf->f_bsize = dentry->d_sb->s_blocksize;
414         buf->f_type = BTRFS_SUPER_MAGIC;
415         return 0;
416 }
417
418 static struct file_system_type btrfs_fs_type = {
419         .owner          = THIS_MODULE,
420         .name           = "btrfs",
421         .get_sb         = btrfs_get_sb,
422         .kill_sb        = kill_block_super,
423         .fs_flags       = FS_REQUIRES_DEV,
424 };
425
426 static struct super_operations btrfs_super_ops = {
427         .delete_inode   = btrfs_delete_inode,
428         .put_inode      = btrfs_put_inode,
429         .put_super      = btrfs_put_super,
430         .read_inode     = btrfs_read_locked_inode,
431         .write_super    = btrfs_write_super,
432         .sync_fs        = btrfs_sync_fs,
433         .write_inode    = btrfs_write_inode,
434         .dirty_inode    = btrfs_dirty_inode,
435         .alloc_inode    = btrfs_alloc_inode,
436         .destroy_inode  = btrfs_destroy_inode,
437         .statfs         = btrfs_statfs,
438 };
439
440 static int __init init_btrfs_fs(void)
441 {
442         int err;
443
444         err = btrfs_init_sysfs();
445         if (err)
446                 return err;
447
448         btrfs_init_transaction_sys();
449         err = btrfs_init_cachep();
450         if (err)
451                 goto free_transaction_sys;
452         err = extent_map_init();
453         if (err)
454                 goto free_cachep;
455
456         err = register_filesystem(&btrfs_fs_type);
457         if (err)
458                 goto free_extent_map;
459         return 0;
460
461 free_extent_map:
462         extent_map_exit();
463 free_cachep:
464         btrfs_destroy_cachep();
465 free_transaction_sys:
466         btrfs_exit_transaction_sys();
467         btrfs_exit_sysfs();
468         return err;
469 }
470
471 static void __exit exit_btrfs_fs(void)
472 {
473         btrfs_exit_transaction_sys();
474         btrfs_destroy_cachep();
475         extent_map_exit();
476         unregister_filesystem(&btrfs_fs_type);
477         btrfs_exit_sysfs();
478 }
479
480 module_init(init_btrfs_fs)
481 module_exit(exit_btrfs_fs)
482
483 MODULE_LICENSE("GPL");