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/cred.h>
  12 #include <linux/parser.h>
  13 #include <linux/buffer_head.h>
  14 #include <linux/vmalloc.h>
  15 #include <linux/writeback.h>
  16 #include <linux/crc-itu-t.h>
  17 #include "omfs.h"
  18
  19 MODULE_AUTHOR("Bob Copeland <me@bobcopeland.com>");
  20 MODULE_DESCRIPTION("OMFS (ReplayTV/Karma) Filesystem for Linux");
  21 MODULE_LICENSE("GPL");
  22
  23 struct buffer_head *omfs_bread(struct super_block *sb, sector_t block)
  24 {
  25         struct omfs_sb_info *sbi = OMFS_SB(sb);
  26         if (block >= sbi->s_num_blocks)
  27                 return NULL;
  28
  29         return sb_bread(sb, clus_to_blk(sbi, block));
  30 }
  31
  32 struct inode *omfs_new_inode(struct inode *dir, umode_t mode)
  33 {
  34         struct inode *inode;
  35         u64 new_block;
  36         int err;
  37         int len;
  38         struct omfs_sb_info *sbi = OMFS_SB(dir->i_sb);
  39
  40         inode = new_inode(dir->i_sb);
  41         if (!inode)
  42                 return ERR_PTR(-ENOMEM);
  43
  44         err = omfs_allocate_range(dir->i_sb, sbi->s_mirrors, sbi->s_mirrors,
  45                         &new_block, &len);
  46         if (err)
  47                 goto fail;
  48
  49         inode->i_ino = new_block;
  50         inode_init_owner(inode, NULL, mode);
  51         inode->i_mapping->a_ops = &omfs_aops;
  52
  53         inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
  54         switch (mode & S_IFMT) {
  55         case S_IFDIR:
  56                 inode->i_op = &omfs_dir_inops;
  57                 inode->i_fop = &omfs_dir_operations;
  58                 inode->i_size = sbi->s_sys_blocksize;
  59                 inc_nlink(inode);
  60                 break;
  61         case S_IFREG:
  62                 inode->i_op = &omfs_file_inops;
  63                 inode->i_fop = &omfs_file_operations;
  64                 inode->i_size = 0;
  65                 break;
  66         }
  67
  68         insert_inode_hash(inode);
  69         mark_inode_dirty(inode);
  70         return inode;
  71 fail:
  72         make_bad_inode(inode);
  73         iput(inode);
  74         return ERR_PTR(err);
  75 }
  76
  77 /*
  78  * Update the header checksums for a dirty inode based on its contents.
  79  * Caller is expected to hold the buffer head underlying oi and mark it
  80  * dirty.
  81  */
  82 static void omfs_update_checksums(struct omfs_inode *oi)
  83 {
  84         int xor, i, ofs = 0, count;
  85         u16 crc = 0;
  86         unsigned char *ptr = (unsigned char *) oi;
  87
  88         count = be32_to_cpu(oi->i_head.h_body_size);
  89         ofs = sizeof(struct omfs_header);
  90
  91         crc = crc_itu_t(crc, ptr + ofs, count);
  92         oi->i_head.h_crc = cpu_to_be16(crc);
  93
  94         xor = ptr[0];
  95         for (i = 1; i < OMFS_XOR_COUNT; i++)
  96                 xor ^= ptr[i];
  97
  98         oi->i_head.h_check_xor = xor;
  99 }
 100
 101 static int __omfs_write_inode(struct inode *inode, int wait)
 102 {
 103         struct omfs_inode *oi;
 104         struct omfs_sb_info *sbi = OMFS_SB(inode->i_sb);
 105         struct buffer_head *bh, *bh2;
 106         u64 ctime;
 107         int i;
 108         int ret = -EIO;
 109         int sync_failed = 0;
 110
 111         /* get current inode since we may have written sibling ptrs etc. */
 112         bh = omfs_bread(inode->i_sb, inode->i_ino);
 113         if (!bh)
 114                 goto out;
 115
 116         oi = (struct omfs_inode *) bh->b_data;
 117
 118         oi->i_head.h_self = cpu_to_be64(inode->i_ino);
 119         if (S_ISDIR(inode->i_mode))
 120                 oi->i_type = OMFS_DIR;
 121         else if (S_ISREG(inode->i_mode))
 122                 oi->i_type = OMFS_FILE;
 123         else {
 124                 printk(KERN_WARNING "omfs: unknown file type: %d\n",
 125                         inode->i_mode);
 126                 goto out_brelse;
 127         }
 128
 129         oi->i_head.h_body_size = cpu_to_be32(sbi->s_sys_blocksize -
 130                 sizeof(struct omfs_header));
 131         oi->i_head.h_version = 1;
 132         oi->i_head.h_type = OMFS_INODE_NORMAL;
 133         oi->i_head.h_magic = OMFS_IMAGIC;
 134         oi->i_size = cpu_to_be64(inode->i_size);
 135
 136         ctime = inode->i_ctime.tv_sec * 1000LL +
 137                 ((inode->i_ctime.tv_nsec + 999)/1000);
 138         oi->i_ctime = cpu_to_be64(ctime);
 139
 140         omfs_update_checksums(oi);
 141
 142         mark_buffer_dirty(bh);
 143         if (wait) {
 144                 sync_dirty_buffer(bh);
 145                 if (buffer_req(bh) && !buffer_uptodate(bh))
 146                         sync_failed = 1;
 147         }
 148
 149         /* if mirroring writes, copy to next fsblock */
 150         for (i = 1; i < sbi->s_mirrors; i++) {
 151                 bh2 = omfs_bread(inode->i_sb, inode->i_ino + i);
 152                 if (!bh2)
 153                         goto out_brelse;
 154
 155                 memcpy(bh2->b_data, bh->b_data, bh->b_size);
 156                 mark_buffer_dirty(bh2);
 157                 if (wait) {
 158                         sync_dirty_buffer(bh2);
 159                         if (buffer_req(bh2) && !buffer_uptodate(bh2))
 160                                 sync_failed = 1;
 161                 }
 162                 brelse(bh2);
 163         }
 164         ret = (sync_failed) ? -EIO : 0;
 165 out_brelse:
 166         brelse(bh);
 167 out:
 168         return ret;
 169 }
 170
 171 static int omfs_write_inode(struct inode *inode, struct writeback_control *wbc)
 172 {
 173         return __omfs_write_inode(inode, wbc->sync_mode == WB_SYNC_ALL);
 174 }
 175
 176 int omfs_sync_inode(struct inode *inode)
 177 {
 178         return __omfs_write_inode(inode, 1);
 179 }
 180
 181 /*
 182  * called when an entry is deleted, need to clear the bits in the
 183  * bitmaps.
 184  */
 185 static void omfs_evict_inode(struct inode *inode)
 186 {
 187         truncate_inode_pages_final(&inode->i_data);
 188         clear_inode(inode);
 189
 190         if (inode->i_nlink)
 191                 return;
 192
 193         if (S_ISREG(inode->i_mode)) {
 194                 inode->i_size = 0;
 195                 omfs_shrink_inode(inode);
 196         }
 197
 198         omfs_clear_range(inode->i_sb, inode->i_ino, 2);
 199 }
 200
 201 struct inode *omfs_iget(struct super_block *sb, ino_t ino)
 202 {
 203         struct omfs_sb_info *sbi = OMFS_SB(sb);
 204         struct omfs_inode *oi;
 205         struct buffer_head *bh;
 206         u64 ctime;
 207         unsigned long nsecs;
 208         struct inode *inode;
 209
 210         inode = iget_locked(sb, ino);
 211         if (!inode)
 212                 return ERR_PTR(-ENOMEM);
 213         if (!(inode->i_state & I_NEW))
 214                 return inode;
 215
 216         bh = omfs_bread(inode->i_sb, ino);
 217         if (!bh)
 218                 goto iget_failed;
 219
 220         oi = (struct omfs_inode *)bh->b_data;
 221
 222         /* check self */
 223         if (ino != be64_to_cpu(oi->i_head.h_self))
 224                 goto fail_bh;
 225
 226         inode->i_uid = sbi->s_uid;
 227         inode->i_gid = sbi->s_gid;
 228
 229         ctime = be64_to_cpu(oi->i_ctime);
 230         nsecs = do_div(ctime, 1000) * 1000L;
 231
 232         inode->i_atime.tv_sec = ctime;
 233         inode->i_mtime.tv_sec = ctime;
 234         inode->i_ctime.tv_sec = ctime;
 235         inode->i_atime.tv_nsec = nsecs;
 236         inode->i_mtime.tv_nsec = nsecs;
 237         inode->i_ctime.tv_nsec = nsecs;
 238
 239         inode->i_mapping->a_ops = &omfs_aops;
 240
 241         switch (oi->i_type) {
 242         case OMFS_DIR:
 243                 inode->i_mode = S_IFDIR | (S_IRWXUGO & ~sbi->s_dmask);
 244                 inode->i_op = &omfs_dir_inops;
 245                 inode->i_fop = &omfs_dir_operations;
 246                 inode->i_size = sbi->s_sys_blocksize;
 247                 inc_nlink(inode);
 248                 break;
 249         case OMFS_FILE:
 250                 inode->i_mode = S_IFREG | (S_IRWXUGO & ~sbi->s_fmask);
 251                 inode->i_fop = &omfs_file_operations;
 252                 inode->i_size = be64_to_cpu(oi->i_size);
 253                 break;
 254         }
 255         brelse(bh);
 256         unlock_new_inode(inode);
 257         return inode;
 258 fail_bh:
 259         brelse(bh);
 260 iget_failed:
 261         iget_failed(inode);
 262         return ERR_PTR(-EIO);
 263 }
 264
 265 static void omfs_put_super(struct super_block *sb)
 266 {
 267         struct omfs_sb_info *sbi = OMFS_SB(sb);
 268         kfree(sbi->s_imap);
 269         kfree(sbi);
 270         sb->s_fs_info = NULL;
 271 }
 272
 273 static int omfs_statfs(struct dentry *dentry, struct kstatfs *buf)
 274 {
 275         struct super_block *s = dentry->d_sb;
 276         struct omfs_sb_info *sbi = OMFS_SB(s);
 277         u64 id = huge_encode_dev(s->s_bdev->bd_dev);
 278
 279         buf->f_type = OMFS_MAGIC;
 280         buf->f_bsize = sbi->s_blocksize;
 281         buf->f_blocks = sbi->s_num_blocks;
 282         buf->f_files = sbi->s_num_blocks;
 283         buf->f_namelen = OMFS_NAMELEN;
 284         buf->f_fsid.val[0] = (u32)id;
 285         buf->f_fsid.val[1] = (u32)(id >> 32);
 286
 287         buf->f_bfree = buf->f_bavail = buf->f_ffree =
 288                 omfs_count_free(s);
 289
 290         return 0;
 291 }
 292
 293 static const struct super_operations omfs_sops = {
 294         .write_inode    = omfs_write_inode,
 295         .evict_inode    = omfs_evict_inode,
 296         .put_super      = omfs_put_super,
 297         .statfs         = omfs_statfs,
 298         .show_options   = generic_show_options,
 299 };
 300
 301 /*
 302  * For Rio Karma, there is an on-disk free bitmap whose location is
 303  * stored in the root block.  For ReplayTV, there is no such free bitmap
 304  * so we have to walk the tree.  Both inodes and file data are allocated
 305  * from the same map.  This array can be big (300k) so we allocate
 306  * in units of the blocksize.
 307  */
 308 static int omfs_get_imap(struct super_block *sb)
 309 {
 310         unsigned int bitmap_size, array_size;
 311         int count;
 312         struct omfs_sb_info *sbi = OMFS_SB(sb);
 313         struct buffer_head *bh;
 314         unsigned long **ptr;
 315         sector_t block;
 316
 317         bitmap_size = DIV_ROUND_UP(sbi->s_num_blocks, 8);
 318         array_size = DIV_ROUND_UP(bitmap_size, sb->s_blocksize);
 319
 320         if (sbi->s_bitmap_ino == ~0ULL)
 321                 goto out;
 322
 323         sbi->s_imap_size = array_size;
 324         sbi->s_imap = kcalloc(array_size, sizeof(unsigned long *), GFP_KERNEL);
 325         if (!sbi->s_imap)
 326                 goto nomem;
 327
 328         block = clus_to_blk(sbi, sbi->s_bitmap_ino);
 329         if (block >= sbi->s_num_blocks)
 330                 goto nomem;
 331
 332         ptr = sbi->s_imap;
 333         for (count = bitmap_size; count > 0; count -= sb->s_blocksize) {
 334                 bh = sb_bread(sb, block++);
 335                 if (!bh)
 336                         goto nomem_free;
 337                 *ptr = kmalloc(sb->s_blocksize, GFP_KERNEL);
 338                 if (!*ptr) {
 339                         brelse(bh);
 340                         goto nomem_free;
 341                 }
 342                 memcpy(*ptr, bh->b_data, sb->s_blocksize);
 343                 if (count < sb->s_blocksize)
 344                         memset((void *)*ptr + count, 0xff,
 345                                 sb->s_blocksize - count);
 346                 brelse(bh);
 347                 ptr++;
 348         }
 349 out:
 350         return 0;
 351
 352 nomem_free:
 353         for (count = 0; count < array_size; count++)
 354                 kfree(sbi->s_imap[count]);
 355
 356         kfree(sbi->s_imap);
 357 nomem:
 358         sbi->s_imap = NULL;
 359         sbi->s_imap_size = 0;
 360         return -ENOMEM;
 361 }
 362
 363 enum {
 364         Opt_uid, Opt_gid, Opt_umask, Opt_dmask, Opt_fmask, Opt_err
 365 };
 366
 367 static const match_table_t tokens = {
 368         {Opt_uid, "uid=%u"},
 369         {Opt_gid, "gid=%u"},
 370         {Opt_umask, "umask=%o"},
 371         {Opt_dmask, "dmask=%o"},
 372         {Opt_fmask, "fmask=%o"},
 373         {Opt_err, NULL},
 374 };
 375
 376 static int parse_options(char *options, struct omfs_sb_info *sbi)
 377 {
 378         char *p;
 379         substring_t args[MAX_OPT_ARGS];
 380         int option;
 381
 382         if (!options)
 383                 return 1;
 384
 385         while ((p = strsep(&options, ",")) != NULL) {
 386                 int token;
 387                 if (!*p)
 388                         continue;
 389
 390                 token = match_token(p, tokens, args);
 391                 switch (token) {
 392                 case Opt_uid:
 393                         if (match_int(&args[0], &option))
 394                                 return 0;
 395                         sbi->s_uid = make_kuid(current_user_ns(), option);
 396                         if (!uid_valid(sbi->s_uid))
 397                                 return 0;
 398                         break;
 399                 case Opt_gid:
 400                         if (match_int(&args[0], &option))
 401                                 return 0;
 402                         sbi->s_gid = make_kgid(current_user_ns(), option);
 403                         if (!gid_valid(sbi->s_gid))
 404                                 return 0;
 405                         break;
 406                 case Opt_umask:
 407                         if (match_octal(&args[0], &option))
 408                                 return 0;
 409                         sbi->s_fmask = sbi->s_dmask = option;
 410                         break;
 411                 case Opt_dmask:
 412                         if (match_octal(&args[0], &option))
 413                                 return 0;
 414                         sbi->s_dmask = option;
 415                         break;
 416                 case Opt_fmask:
 417                         if (match_octal(&args[0], &option))
 418                                 return 0;
 419                         sbi->s_fmask = option;
 420                         break;
 421                 default:
 422                         return 0;
 423                 }
 424         }
 425         return 1;
 426 }
 427
 428 static int omfs_fill_super(struct super_block *sb, void *data, int silent)
 429 {
 430         struct buffer_head *bh, *bh2;
 431         struct omfs_super_block *omfs_sb;
 432         struct omfs_root_block *omfs_rb;
 433         struct omfs_sb_info *sbi;
 434         struct inode *root;
 435         int ret = -EINVAL;
 436
 437         save_mount_options(sb, (char *) data);
 438
 439         sbi = kzalloc(sizeof(struct omfs_sb_info), GFP_KERNEL);
 440         if (!sbi)
 441                 return -ENOMEM;
 442
 443         sb->s_fs_info = sbi;
 444
 445         sbi->s_uid = current_uid();
 446         sbi->s_gid = current_gid();
 447         sbi->s_dmask = sbi->s_fmask = current_umask();
 448
 449         if (!parse_options((char *) data, sbi))
 450                 goto end;
 451
 452         sb->s_maxbytes = 0xffffffff;
 453
 454         sb_set_blocksize(sb, 0x200);
 455
 456         bh = sb_bread(sb, 0);
 457         if (!bh)
 458                 goto end;
 459
 460         omfs_sb = (struct omfs_super_block *)bh->b_data;
 461
 462         if (omfs_sb->s_magic != cpu_to_be32(OMFS_MAGIC)) {
 463                 if (!silent)
 464                         printk(KERN_ERR "omfs: Invalid superblock (%x)\n",
 465                                    omfs_sb->s_magic);
 466                 goto out_brelse_bh;
 467         }
 468         sb->s_magic = OMFS_MAGIC;
 469
 470         sbi->s_num_blocks = be64_to_cpu(omfs_sb->s_num_blocks);
 471         sbi->s_blocksize = be32_to_cpu(omfs_sb->s_blocksize);
 472         sbi->s_mirrors = be32_to_cpu(omfs_sb->s_mirrors);
 473         sbi->s_root_ino = be64_to_cpu(omfs_sb->s_root_block);
 474         sbi->s_sys_blocksize = be32_to_cpu(omfs_sb->s_sys_blocksize);
 475         mutex_init(&sbi->s_bitmap_lock);
 476
 477         if (sbi->s_num_blocks > OMFS_MAX_BLOCKS) {
 478                 printk(KERN_ERR "omfs: sysblock number (%llx) is out of range\n",
 479                        (unsigned long long)sbi->s_num_blocks);
 480                 goto out_brelse_bh;
 481         }
 482
 483         if (sbi->s_sys_blocksize > PAGE_SIZE) {
 484                 printk(KERN_ERR "omfs: sysblock size (%d) is out of range\n",
 485                         sbi->s_sys_blocksize);
 486                 goto out_brelse_bh;
 487         }
 488
 489         if (sbi->s_blocksize < sbi->s_sys_blocksize ||
 490             sbi->s_blocksize > OMFS_MAX_BLOCK_SIZE) {
 491                 printk(KERN_ERR "omfs: block size (%d) is out of range\n",
 492                         sbi->s_blocksize);
 493                 goto out_brelse_bh;
 494         }
 495
 496         /*
 497          * Use sys_blocksize as the fs block since it is smaller than a
 498          * page while the fs blocksize can be larger.
 499          */
 500         sb_set_blocksize(sb, sbi->s_sys_blocksize);
 501
 502         /*
 503          * ...and the difference goes into a shift.  sys_blocksize is always
 504          * a power of two factor of blocksize.
 505          */
 506         sbi->s_block_shift = get_bitmask_order(sbi->s_blocksize) -
 507                 get_bitmask_order(sbi->s_sys_blocksize);
 508
 509         bh2 = omfs_bread(sb, be64_to_cpu(omfs_sb->s_root_block));
 510         if (!bh2)
 511                 goto out_brelse_bh;
 512
 513         omfs_rb = (struct omfs_root_block *)bh2->b_data;
 514
 515         sbi->s_bitmap_ino = be64_to_cpu(omfs_rb->r_bitmap);
 516         sbi->s_clustersize = be32_to_cpu(omfs_rb->r_clustersize);
 517
 518         if (sbi->s_num_blocks != be64_to_cpu(omfs_rb->r_num_blocks)) {
 519                 printk(KERN_ERR "omfs: block count discrepancy between "
 520                         "super and root blocks (%llx, %llx)\n",
 521                         (unsigned long long)sbi->s_num_blocks,
 522                         (unsigned long long)be64_to_cpu(omfs_rb->r_num_blocks));
 523                 goto out_brelse_bh2;
 524         }
 525
 526         if (sbi->s_bitmap_ino != ~0ULL &&
 527             sbi->s_bitmap_ino > sbi->s_num_blocks) {
 528                 printk(KERN_ERR "omfs: free space bitmap location is corrupt "
 529                         "(%llx, total blocks %llx)\n",
 530                         (unsigned long long) sbi->s_bitmap_ino,
 531                         (unsigned long long) sbi->s_num_blocks);
 532                 goto out_brelse_bh2;
 533         }
 534         if (sbi->s_clustersize < 1 ||
 535             sbi->s_clustersize > OMFS_MAX_CLUSTER_SIZE) {
 536                 printk(KERN_ERR "omfs: cluster size out of range (%d)",
 537                         sbi->s_clustersize);
 538                 goto out_brelse_bh2;
 539         }
 540
 541         ret = omfs_get_imap(sb);
 542         if (ret)
 543                 goto out_brelse_bh2;
 544
 545         sb->s_op = &omfs_sops;
 546
 547         root = omfs_iget(sb, be64_to_cpu(omfs_rb->r_root_dir));
 548         if (IS_ERR(root)) {
 549                 ret = PTR_ERR(root);
 550                 goto out_brelse_bh2;
 551         }
 552
 553         sb->s_root = d_make_root(root);
 554         if (!sb->s_root) {
 555                 ret = -ENOMEM;
 556                 goto out_brelse_bh2;
 557         }
 558         printk(KERN_DEBUG "omfs: Mounted volume %s\n", omfs_rb->r_name);
 559
 560         ret = 0;
 561 out_brelse_bh2:
 562         brelse(bh2);
 563 out_brelse_bh:
 564         brelse(bh);
 565 end:
 566         if (ret)
 567                 kfree(sbi);
 568         return ret;
 569 }
 570
 571 static struct dentry *omfs_mount(struct file_system_type *fs_type,
 572                         int flags, const char *dev_name, void *data)
 573 {
 574         return mount_bdev(fs_type, flags, dev_name, data, omfs_fill_super);
 575 }
 576
 577 static struct file_system_type omfs_fs_type = {
 578         .owner = THIS_MODULE,
 579         .name = "omfs",
 580         .mount = omfs_mount,
 581         .kill_sb = kill_block_super,
 582         .fs_flags = FS_REQUIRES_DEV,
 583 };
 584 MODULE_ALIAS_FS("omfs");
 585
 586 static int __init init_omfs_fs(void)
 587 {
 588         return register_filesystem(&omfs_fs_type);
 589 }
 590
 591 static void __exit exit_omfs_fs(void)
 592 {
 593         unregister_filesystem(&omfs_fs_type);
 594 }
 595
 596 module_init(init_omfs_fs);
 597 module_exit(exit_omfs_fs);