Merge branch 'next' of git://git.kernel.org/pub/scm/linux/kernel/git/rzhang/linux
[cascardo/linux.git] / fs / omfs / inode.c
1 /*
2  * Optimized MPEG FS - inode and super operations.
3  * Copyright (C) 2006 Bob Copeland <me@bobcopeland.com>
4  * Released under GPL v2.
5  */
6 #include <linux/module.h>
7 #include <linux/sched.h>
8 #include <linux/slab.h>
9 #include <linux/fs.h>
10 #include <linux/vfs.h>
11 #include <linux/parser.h>
12 #include <linux/buffer_head.h>
13 #include <linux/vmalloc.h>
14 #include <linux/writeback.h>
15 #include <linux/crc-itu-t.h>
16 #include "omfs.h"
17
18 MODULE_AUTHOR("Bob Copeland <me@bobcopeland.com>");
19 MODULE_DESCRIPTION("OMFS (ReplayTV/Karma) Filesystem for Linux");
20 MODULE_LICENSE("GPL");
21
22 struct buffer_head *omfs_bread(struct super_block *sb, sector_t block)
23 {
24         struct omfs_sb_info *sbi = OMFS_SB(sb);
25         if (block >= sbi->s_num_blocks)
26                 return NULL;
27
28         return sb_bread(sb, clus_to_blk(sbi, block));
29 }
30
31 struct inode *omfs_new_inode(struct inode *dir, umode_t mode)
32 {
33         struct inode *inode;
34         u64 new_block;
35         int err;
36         int len;
37         struct omfs_sb_info *sbi = OMFS_SB(dir->i_sb);
38
39         inode = new_inode(dir->i_sb);
40         if (!inode)
41                 return ERR_PTR(-ENOMEM);
42
43         err = omfs_allocate_range(dir->i_sb, sbi->s_mirrors, sbi->s_mirrors,
44                         &new_block, &len);
45         if (err)
46                 goto fail;
47
48         inode->i_ino = new_block;
49         inode_init_owner(inode, NULL, mode);
50         inode->i_mapping->a_ops = &omfs_aops;
51
52         inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
53         switch (mode & S_IFMT) {
54         case S_IFDIR:
55                 inode->i_op = &omfs_dir_inops;
56                 inode->i_fop = &omfs_dir_operations;
57                 inode->i_size = sbi->s_sys_blocksize;
58                 inc_nlink(inode);
59                 break;
60         case S_IFREG:
61                 inode->i_op = &omfs_file_inops;
62                 inode->i_fop = &omfs_file_operations;
63                 inode->i_size = 0;
64                 break;
65         }
66
67         insert_inode_hash(inode);
68         mark_inode_dirty(inode);
69         return inode;
70 fail:
71         make_bad_inode(inode);
72         iput(inode);
73         return ERR_PTR(err);
74 }
75
76 /*
77  * Update the header checksums for a dirty inode based on its contents.
78  * Caller is expected to hold the buffer head underlying oi and mark it
79  * dirty.
80  */
81 static void omfs_update_checksums(struct omfs_inode *oi)
82 {
83         int xor, i, ofs = 0, count;
84         u16 crc = 0;
85         unsigned char *ptr = (unsigned char *) oi;
86
87         count = be32_to_cpu(oi->i_head.h_body_size);
88         ofs = sizeof(struct omfs_header);
89
90         crc = crc_itu_t(crc, ptr + ofs, count);
91         oi->i_head.h_crc = cpu_to_be16(crc);
92
93         xor = ptr[0];
94         for (i = 1; i < OMFS_XOR_COUNT; i++)
95                 xor ^= ptr[i];
96
97         oi->i_head.h_check_xor = xor;
98 }
99
100 static int __omfs_write_inode(struct inode *inode, int wait)
101 {
102         struct omfs_inode *oi;
103         struct omfs_sb_info *sbi = OMFS_SB(inode->i_sb);
104         struct buffer_head *bh, *bh2;
105         u64 ctime;
106         int i;
107         int ret = -EIO;
108         int sync_failed = 0;
109
110         /* get current inode since we may have written sibling ptrs etc. */
111         bh = omfs_bread(inode->i_sb, inode->i_ino);
112         if (!bh)
113                 goto out;
114
115         oi = (struct omfs_inode *) bh->b_data;
116
117         oi->i_head.h_self = cpu_to_be64(inode->i_ino);
118         if (S_ISDIR(inode->i_mode))
119                 oi->i_type = OMFS_DIR;
120         else if (S_ISREG(inode->i_mode))
121                 oi->i_type = OMFS_FILE;
122         else {
123                 printk(KERN_WARNING "omfs: unknown file type: %d\n",
124                         inode->i_mode);
125                 goto out_brelse;
126         }
127
128         oi->i_head.h_body_size = cpu_to_be32(sbi->s_sys_blocksize -
129                 sizeof(struct omfs_header));
130         oi->i_head.h_version = 1;
131         oi->i_head.h_type = OMFS_INODE_NORMAL;
132         oi->i_head.h_magic = OMFS_IMAGIC;
133         oi->i_size = cpu_to_be64(inode->i_size);
134
135         ctime = inode->i_ctime.tv_sec * 1000LL +
136                 ((inode->i_ctime.tv_nsec + 999)/1000);
137         oi->i_ctime = cpu_to_be64(ctime);
138
139         omfs_update_checksums(oi);
140
141         mark_buffer_dirty(bh);
142         if (wait) {
143                 sync_dirty_buffer(bh);
144                 if (buffer_req(bh) && !buffer_uptodate(bh))
145                         sync_failed = 1;
146         }
147
148         /* if mirroring writes, copy to next fsblock */
149         for (i = 1; i < sbi->s_mirrors; i++) {
150                 bh2 = omfs_bread(inode->i_sb, inode->i_ino + i);
151                 if (!bh2)
152                         goto out_brelse;
153
154                 memcpy(bh2->b_data, bh->b_data, bh->b_size);
155                 mark_buffer_dirty(bh2);
156                 if (wait) {
157                         sync_dirty_buffer(bh2);
158                         if (buffer_req(bh2) && !buffer_uptodate(bh2))
159                                 sync_failed = 1;
160                 }
161                 brelse(bh2);
162         }
163         ret = (sync_failed) ? -EIO : 0;
164 out_brelse:
165         brelse(bh);
166 out:
167         return ret;
168 }
169
170 static int omfs_write_inode(struct inode *inode, struct writeback_control *wbc)
171 {
172         return __omfs_write_inode(inode, wbc->sync_mode == WB_SYNC_ALL);
173 }
174
175 int omfs_sync_inode(struct inode *inode)
176 {
177         return __omfs_write_inode(inode, 1);
178 }
179
180 /*
181  * called when an entry is deleted, need to clear the bits in the
182  * bitmaps.
183  */
184 static void omfs_evict_inode(struct inode *inode)
185 {
186         truncate_inode_pages_final(&inode->i_data);
187         clear_inode(inode);
188
189         if (inode->i_nlink)
190                 return;
191
192         if (S_ISREG(inode->i_mode)) {
193                 inode->i_size = 0;
194                 omfs_shrink_inode(inode);
195         }
196
197         omfs_clear_range(inode->i_sb, inode->i_ino, 2);
198 }
199
200 struct inode *omfs_iget(struct super_block *sb, ino_t ino)
201 {
202         struct omfs_sb_info *sbi = OMFS_SB(sb);
203         struct omfs_inode *oi;
204         struct buffer_head *bh;
205         u64 ctime;
206         unsigned long nsecs;
207         struct inode *inode;
208
209         inode = iget_locked(sb, ino);
210         if (!inode)
211                 return ERR_PTR(-ENOMEM);
212         if (!(inode->i_state & I_NEW))
213                 return inode;
214
215         bh = omfs_bread(inode->i_sb, ino);
216         if (!bh)
217                 goto iget_failed;
218
219         oi = (struct omfs_inode *)bh->b_data;
220
221         /* check self */
222         if (ino != be64_to_cpu(oi->i_head.h_self))
223                 goto fail_bh;
224
225         inode->i_uid = sbi->s_uid;
226         inode->i_gid = sbi->s_gid;
227
228         ctime = be64_to_cpu(oi->i_ctime);
229         nsecs = do_div(ctime, 1000) * 1000L;
230
231         inode->i_atime.tv_sec = ctime;
232         inode->i_mtime.tv_sec = ctime;
233         inode->i_ctime.tv_sec = ctime;
234         inode->i_atime.tv_nsec = nsecs;
235         inode->i_mtime.tv_nsec = nsecs;
236         inode->i_ctime.tv_nsec = nsecs;
237
238         inode->i_mapping->a_ops = &omfs_aops;
239
240         switch (oi->i_type) {
241         case OMFS_DIR:
242                 inode->i_mode = S_IFDIR | (S_IRWXUGO & ~sbi->s_dmask);
243                 inode->i_op = &omfs_dir_inops;
244                 inode->i_fop = &omfs_dir_operations;
245                 inode->i_size = sbi->s_sys_blocksize;
246                 inc_nlink(inode);
247                 break;
248         case OMFS_FILE:
249                 inode->i_mode = S_IFREG | (S_IRWXUGO & ~sbi->s_fmask);
250                 inode->i_fop = &omfs_file_operations;
251                 inode->i_size = be64_to_cpu(oi->i_size);
252                 break;
253         }
254         brelse(bh);
255         unlock_new_inode(inode);
256         return inode;
257 fail_bh:
258         brelse(bh);
259 iget_failed:
260         iget_failed(inode);
261         return ERR_PTR(-EIO);
262 }
263
264 static void omfs_put_super(struct super_block *sb)
265 {
266         struct omfs_sb_info *sbi = OMFS_SB(sb);
267         kfree(sbi->s_imap);
268         kfree(sbi);
269         sb->s_fs_info = NULL;
270 }
271
272 static int omfs_statfs(struct dentry *dentry, struct kstatfs *buf)
273 {
274         struct super_block *s = dentry->d_sb;
275         struct omfs_sb_info *sbi = OMFS_SB(s);
276         u64 id = huge_encode_dev(s->s_bdev->bd_dev);
277
278         buf->f_type = OMFS_MAGIC;
279         buf->f_bsize = sbi->s_blocksize;
280         buf->f_blocks = sbi->s_num_blocks;
281         buf->f_files = sbi->s_num_blocks;
282         buf->f_namelen = OMFS_NAMELEN;
283         buf->f_fsid.val[0] = (u32)id;
284         buf->f_fsid.val[1] = (u32)(id >> 32);
285
286         buf->f_bfree = buf->f_bavail = buf->f_ffree =
287                 omfs_count_free(s);
288
289         return 0;
290 }
291
292 static const struct super_operations omfs_sops = {
293         .write_inode    = omfs_write_inode,
294         .evict_inode    = omfs_evict_inode,
295         .put_super      = omfs_put_super,
296         .statfs         = omfs_statfs,
297         .show_options   = generic_show_options,
298 };
299
300 /*
301  * For Rio Karma, there is an on-disk free bitmap whose location is
302  * stored in the root block.  For ReplayTV, there is no such free bitmap
303  * so we have to walk the tree.  Both inodes and file data are allocated
304  * from the same map.  This array can be big (300k) so we allocate
305  * in units of the blocksize.
306  */
307 static int omfs_get_imap(struct super_block *sb)
308 {
309         unsigned int bitmap_size, count, array_size;
310         struct omfs_sb_info *sbi = OMFS_SB(sb);
311         struct buffer_head *bh;
312         unsigned long **ptr;
313         sector_t block;
314
315         bitmap_size = DIV_ROUND_UP(sbi->s_num_blocks, 8);
316         array_size = DIV_ROUND_UP(bitmap_size, sb->s_blocksize);
317
318         if (sbi->s_bitmap_ino == ~0ULL)
319                 goto out;
320
321         sbi->s_imap_size = array_size;
322         sbi->s_imap = kcalloc(array_size, sizeof(unsigned long *), GFP_KERNEL);
323         if (!sbi->s_imap)
324                 goto nomem;
325
326         block = clus_to_blk(sbi, sbi->s_bitmap_ino);
327         if (block >= sbi->s_num_blocks)
328                 goto nomem;
329
330         ptr = sbi->s_imap;
331         for (count = bitmap_size; count > 0; count -= sb->s_blocksize) {
332                 bh = sb_bread(sb, block++);
333                 if (!bh)
334                         goto nomem_free;
335                 *ptr = kmalloc(sb->s_blocksize, GFP_KERNEL);
336                 if (!*ptr) {
337                         brelse(bh);
338                         goto nomem_free;
339                 }
340                 memcpy(*ptr, bh->b_data, sb->s_blocksize);
341                 if (count < sb->s_blocksize)
342                         memset((void *)*ptr + count, 0xff,
343                                 sb->s_blocksize - count);
344                 brelse(bh);
345                 ptr++;
346         }
347 out:
348         return 0;
349
350 nomem_free:
351         for (count = 0; count < array_size; count++)
352                 kfree(sbi->s_imap[count]);
353
354         kfree(sbi->s_imap);
355 nomem:
356         sbi->s_imap = NULL;
357         sbi->s_imap_size = 0;
358         return -ENOMEM;
359 }
360
361 enum {
362         Opt_uid, Opt_gid, Opt_umask, Opt_dmask, Opt_fmask
363 };
364
365 static const match_table_t tokens = {
366         {Opt_uid, "uid=%u"},
367         {Opt_gid, "gid=%u"},
368         {Opt_umask, "umask=%o"},
369         {Opt_dmask, "dmask=%o"},
370         {Opt_fmask, "fmask=%o"},
371 };
372
373 static int parse_options(char *options, struct omfs_sb_info *sbi)
374 {
375         char *p;
376         substring_t args[MAX_OPT_ARGS];
377         int option;
378
379         if (!options)
380                 return 1;
381
382         while ((p = strsep(&options, ",")) != NULL) {
383                 int token;
384                 if (!*p)
385                         continue;
386
387                 token = match_token(p, tokens, args);
388                 switch (token) {
389                 case Opt_uid:
390                         if (match_int(&args[0], &option))
391                                 return 0;
392                         sbi->s_uid = make_kuid(current_user_ns(), option);
393                         if (!uid_valid(sbi->s_uid))
394                                 return 0;
395                         break;
396                 case Opt_gid:
397                         if (match_int(&args[0], &option))
398                                 return 0;
399                         sbi->s_gid = make_kgid(current_user_ns(), option);
400                         if (!gid_valid(sbi->s_gid))
401                                 return 0;
402                         break;
403                 case Opt_umask:
404                         if (match_octal(&args[0], &option))
405                                 return 0;
406                         sbi->s_fmask = sbi->s_dmask = option;
407                         break;
408                 case Opt_dmask:
409                         if (match_octal(&args[0], &option))
410                                 return 0;
411                         sbi->s_dmask = option;
412                         break;
413                 case Opt_fmask:
414                         if (match_octal(&args[0], &option))
415                                 return 0;
416                         sbi->s_fmask = option;
417                         break;
418                 default:
419                         return 0;
420                 }
421         }
422         return 1;
423 }
424
425 static int omfs_fill_super(struct super_block *sb, void *data, int silent)
426 {
427         struct buffer_head *bh, *bh2;
428         struct omfs_super_block *omfs_sb;
429         struct omfs_root_block *omfs_rb;
430         struct omfs_sb_info *sbi;
431         struct inode *root;
432         int ret = -EINVAL;
433
434         save_mount_options(sb, (char *) data);
435
436         sbi = kzalloc(sizeof(struct omfs_sb_info), GFP_KERNEL);
437         if (!sbi)
438                 return -ENOMEM;
439
440         sb->s_fs_info = sbi;
441
442         sbi->s_uid = current_uid();
443         sbi->s_gid = current_gid();
444         sbi->s_dmask = sbi->s_fmask = current_umask();
445
446         if (!parse_options((char *) data, sbi))
447                 goto end;
448
449         sb->s_maxbytes = 0xffffffff;
450
451         sb_set_blocksize(sb, 0x200);
452
453         bh = sb_bread(sb, 0);
454         if (!bh)
455                 goto end;
456
457         omfs_sb = (struct omfs_super_block *)bh->b_data;
458
459         if (omfs_sb->s_magic != cpu_to_be32(OMFS_MAGIC)) {
460                 if (!silent)
461                         printk(KERN_ERR "omfs: Invalid superblock (%x)\n",
462                                    omfs_sb->s_magic);
463                 goto out_brelse_bh;
464         }
465         sb->s_magic = OMFS_MAGIC;
466
467         sbi->s_num_blocks = be64_to_cpu(omfs_sb->s_num_blocks);
468         sbi->s_blocksize = be32_to_cpu(omfs_sb->s_blocksize);
469         sbi->s_mirrors = be32_to_cpu(omfs_sb->s_mirrors);
470         sbi->s_root_ino = be64_to_cpu(omfs_sb->s_root_block);
471         sbi->s_sys_blocksize = be32_to_cpu(omfs_sb->s_sys_blocksize);
472         mutex_init(&sbi->s_bitmap_lock);
473
474         if (sbi->s_num_blocks > OMFS_MAX_BLOCKS) {
475                 printk(KERN_ERR "omfs: sysblock number (%llx) is out of range\n",
476                        (unsigned long long)sbi->s_num_blocks);
477                 goto out_brelse_bh;
478         }
479
480         if (sbi->s_sys_blocksize > PAGE_SIZE) {
481                 printk(KERN_ERR "omfs: sysblock size (%d) is out of range\n",
482                         sbi->s_sys_blocksize);
483                 goto out_brelse_bh;
484         }
485
486         if (sbi->s_blocksize < sbi->s_sys_blocksize ||
487             sbi->s_blocksize > OMFS_MAX_BLOCK_SIZE) {
488                 printk(KERN_ERR "omfs: block size (%d) is out of range\n",
489                         sbi->s_blocksize);
490                 goto out_brelse_bh;
491         }
492
493         /*
494          * Use sys_blocksize as the fs block since it is smaller than a
495          * page while the fs blocksize can be larger.
496          */
497         sb_set_blocksize(sb, sbi->s_sys_blocksize);
498
499         /*
500          * ...and the difference goes into a shift.  sys_blocksize is always
501          * a power of two factor of blocksize.
502          */
503         sbi->s_block_shift = get_bitmask_order(sbi->s_blocksize) -
504                 get_bitmask_order(sbi->s_sys_blocksize);
505
506         bh2 = omfs_bread(sb, be64_to_cpu(omfs_sb->s_root_block));
507         if (!bh2)
508                 goto out_brelse_bh;
509
510         omfs_rb = (struct omfs_root_block *)bh2->b_data;
511
512         sbi->s_bitmap_ino = be64_to_cpu(omfs_rb->r_bitmap);
513         sbi->s_clustersize = be32_to_cpu(omfs_rb->r_clustersize);
514
515         if (sbi->s_num_blocks != be64_to_cpu(omfs_rb->r_num_blocks)) {
516                 printk(KERN_ERR "omfs: block count discrepancy between "
517                         "super and root blocks (%llx, %llx)\n",
518                         (unsigned long long)sbi->s_num_blocks,
519                         (unsigned long long)be64_to_cpu(omfs_rb->r_num_blocks));
520                 goto out_brelse_bh2;
521         }
522
523         if (sbi->s_bitmap_ino != ~0ULL &&
524             sbi->s_bitmap_ino > sbi->s_num_blocks) {
525                 printk(KERN_ERR "omfs: free space bitmap location is corrupt "
526                         "(%llx, total blocks %llx)\n",
527                         (unsigned long long) sbi->s_bitmap_ino,
528                         (unsigned long long) sbi->s_num_blocks);
529                 goto out_brelse_bh2;
530         }
531         if (sbi->s_clustersize < 1 ||
532             sbi->s_clustersize > OMFS_MAX_CLUSTER_SIZE) {
533                 printk(KERN_ERR "omfs: cluster size out of range (%d)",
534                         sbi->s_clustersize);
535                 goto out_brelse_bh2;
536         }
537
538         ret = omfs_get_imap(sb);
539         if (ret)
540                 goto out_brelse_bh2;
541
542         sb->s_op = &omfs_sops;
543
544         root = omfs_iget(sb, be64_to_cpu(omfs_rb->r_root_dir));
545         if (IS_ERR(root)) {
546                 ret = PTR_ERR(root);
547                 goto out_brelse_bh2;
548         }
549
550         sb->s_root = d_make_root(root);
551         if (!sb->s_root)
552                 goto out_brelse_bh2;
553         printk(KERN_DEBUG "omfs: Mounted volume %s\n", omfs_rb->r_name);
554
555         ret = 0;
556 out_brelse_bh2:
557         brelse(bh2);
558 out_brelse_bh:
559         brelse(bh);
560 end:
561         if (ret)
562                 kfree(sbi);
563         return ret;
564 }
565
566 static struct dentry *omfs_mount(struct file_system_type *fs_type,
567                         int flags, const char *dev_name, void *data)
568 {
569         return mount_bdev(fs_type, flags, dev_name, data, omfs_fill_super);
570 }
571
572 static struct file_system_type omfs_fs_type = {
573         .owner = THIS_MODULE,
574         .name = "omfs",
575         .mount = omfs_mount,
576         .kill_sb = kill_block_super,
577         .fs_flags = FS_REQUIRES_DEV,
578 };
579 MODULE_ALIAS_FS("omfs");
580
581 static int __init init_omfs_fs(void)
582 {
583         return register_filesystem(&omfs_fs_type);
584 }
585
586 static void __exit exit_omfs_fs(void)
587 {
588         unregister_filesystem(&omfs_fs_type);
589 }
590
591 module_init(init_omfs_fs);
592 module_exit(exit_omfs_fs);