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/seq_file.h> 17 #include <linux/crc-itu-t.h> 18 #include "omfs.h" 19 20 MODULE_AUTHOR("Bob Copeland <me@bobcopeland.com>"); 21 MODULE_DESCRIPTION("OMFS (ReplayTV/Karma) Filesystem for Linux"); 22 MODULE_LICENSE("GPL"); 23 24 struct buffer_head *omfs_bread(struct super_block *sb, sector_t block) 25 { 26 struct omfs_sb_info *sbi = OMFS_SB(sb); 27 if (block >= sbi->s_num_blocks) 28 return NULL; 29 30 return sb_bread(sb, clus_to_blk(sbi, block)); 31 } 32 33 struct inode *omfs_new_inode(struct inode *dir, umode_t mode) 34 { 35 struct inode *inode; 36 u64 new_block; 37 int err; 38 int len; 39 struct omfs_sb_info *sbi = OMFS_SB(dir->i_sb); 40 41 inode = new_inode(dir->i_sb); 42 if (!inode) 43 return ERR_PTR(-ENOMEM); 44 45 err = omfs_allocate_range(dir->i_sb, sbi->s_mirrors, sbi->s_mirrors, 46 &new_block, &len); 47 if (err) 48 goto fail; 49 50 inode->i_ino = new_block; 51 inode_init_owner(inode, NULL, mode); 52 inode->i_mapping->a_ops = &omfs_aops; 53 54 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode); 55 switch (mode & S_IFMT) { 56 case S_IFDIR: 57 inode->i_op = &omfs_dir_inops; 58 inode->i_fop = &omfs_dir_operations; 59 inode->i_size = sbi->s_sys_blocksize; 60 inc_nlink(inode); 61 break; 62 case S_IFREG: 63 inode->i_op = &omfs_file_inops; 64 inode->i_fop = &omfs_file_operations; 65 inode->i_size = 0; 66 break; 67 } 68 69 insert_inode_hash(inode); 70 mark_inode_dirty(inode); 71 return inode; 72 fail: 73 make_bad_inode(inode); 74 iput(inode); 75 return ERR_PTR(err); 76 } 77 78 /* 79 * Update the header checksums for a dirty inode based on its contents. 80 * Caller is expected to hold the buffer head underlying oi and mark it 81 * dirty. 82 */ 83 static void omfs_update_checksums(struct omfs_inode *oi) 84 { 85 int xor, i, ofs = 0, count; 86 u16 crc = 0; 87 unsigned char *ptr = (unsigned char *) oi; 88 89 count = be32_to_cpu(oi->i_head.h_body_size); 90 ofs = sizeof(struct omfs_header); 91 92 crc = crc_itu_t(crc, ptr + ofs, count); 93 oi->i_head.h_crc = cpu_to_be16(crc); 94 95 xor = ptr[0]; 96 for (i = 1; i < OMFS_XOR_COUNT; i++) 97 xor ^= ptr[i]; 98 99 oi->i_head.h_check_xor = xor; 100 } 101 102 static int __omfs_write_inode(struct inode *inode, int wait) 103 { 104 struct omfs_inode *oi; 105 struct omfs_sb_info *sbi = OMFS_SB(inode->i_sb); 106 struct buffer_head *bh, *bh2; 107 u64 ctime; 108 int i; 109 int ret = -EIO; 110 int sync_failed = 0; 111 112 /* get current inode since we may have written sibling ptrs etc. */ 113 bh = omfs_bread(inode->i_sb, inode->i_ino); 114 if (!bh) 115 goto out; 116 117 oi = (struct omfs_inode *) bh->b_data; 118 119 oi->i_head.h_self = cpu_to_be64(inode->i_ino); 120 if (S_ISDIR(inode->i_mode)) 121 oi->i_type = OMFS_DIR; 122 else if (S_ISREG(inode->i_mode)) 123 oi->i_type = OMFS_FILE; 124 else { 125 printk(KERN_WARNING "omfs: unknown file type: %d\n", 126 inode->i_mode); 127 goto out_brelse; 128 } 129 130 oi->i_head.h_body_size = cpu_to_be32(sbi->s_sys_blocksize - 131 sizeof(struct omfs_header)); 132 oi->i_head.h_version = 1; 133 oi->i_head.h_type = OMFS_INODE_NORMAL; 134 oi->i_head.h_magic = OMFS_IMAGIC; 135 oi->i_size = cpu_to_be64(inode->i_size); 136 137 ctime = inode->i_ctime.tv_sec * 1000LL + 138 ((inode->i_ctime.tv_nsec + 999)/1000); 139 oi->i_ctime = cpu_to_be64(ctime); 140 141 omfs_update_checksums(oi); 142 143 mark_buffer_dirty(bh); 144 if (wait) { 145 sync_dirty_buffer(bh); 146 if (buffer_req(bh) && !buffer_uptodate(bh)) 147 sync_failed = 1; 148 } 149 150 /* if mirroring writes, copy to next fsblock */ 151 for (i = 1; i < sbi->s_mirrors; i++) { 152 bh2 = omfs_bread(inode->i_sb, inode->i_ino + i); 153 if (!bh2) 154 goto out_brelse; 155 156 memcpy(bh2->b_data, bh->b_data, bh->b_size); 157 mark_buffer_dirty(bh2); 158 if (wait) { 159 sync_dirty_buffer(bh2); 160 if (buffer_req(bh2) && !buffer_uptodate(bh2)) 161 sync_failed = 1; 162 } 163 brelse(bh2); 164 } 165 ret = (sync_failed) ? -EIO : 0; 166 out_brelse: 167 brelse(bh); 168 out: 169 return ret; 170 } 171 172 static int omfs_write_inode(struct inode *inode, struct writeback_control *wbc) 173 { 174 return __omfs_write_inode(inode, wbc->sync_mode == WB_SYNC_ALL); 175 } 176 177 int omfs_sync_inode(struct inode *inode) 178 { 179 return __omfs_write_inode(inode, 1); 180 } 181 182 /* 183 * called when an entry is deleted, need to clear the bits in the 184 * bitmaps. 185 */ 186 static void omfs_evict_inode(struct inode *inode) 187 { 188 truncate_inode_pages_final(&inode->i_data); 189 clear_inode(inode); 190 191 if (inode->i_nlink) 192 return; 193 194 if (S_ISREG(inode->i_mode)) { 195 inode->i_size = 0; 196 omfs_shrink_inode(inode); 197 } 198 199 omfs_clear_range(inode->i_sb, inode->i_ino, 2); 200 } 201 202 struct inode *omfs_iget(struct super_block *sb, ino_t ino) 203 { 204 struct omfs_sb_info *sbi = OMFS_SB(sb); 205 struct omfs_inode *oi; 206 struct buffer_head *bh; 207 u64 ctime; 208 unsigned long nsecs; 209 struct inode *inode; 210 211 inode = iget_locked(sb, ino); 212 if (!inode) 213 return ERR_PTR(-ENOMEM); 214 if (!(inode->i_state & I_NEW)) 215 return inode; 216 217 bh = omfs_bread(inode->i_sb, ino); 218 if (!bh) 219 goto iget_failed; 220 221 oi = (struct omfs_inode *)bh->b_data; 222 223 /* check self */ 224 if (ino != be64_to_cpu(oi->i_head.h_self)) 225 goto fail_bh; 226 227 inode->i_uid = sbi->s_uid; 228 inode->i_gid = sbi->s_gid; 229 230 ctime = be64_to_cpu(oi->i_ctime); 231 nsecs = do_div(ctime, 1000) * 1000L; 232 233 inode->i_atime.tv_sec = ctime; 234 inode->i_mtime.tv_sec = ctime; 235 inode->i_ctime.tv_sec = ctime; 236 inode->i_atime.tv_nsec = nsecs; 237 inode->i_mtime.tv_nsec = nsecs; 238 inode->i_ctime.tv_nsec = nsecs; 239 240 inode->i_mapping->a_ops = &omfs_aops; 241 242 switch (oi->i_type) { 243 case OMFS_DIR: 244 inode->i_mode = S_IFDIR | (S_IRWXUGO & ~sbi->s_dmask); 245 inode->i_op = &omfs_dir_inops; 246 inode->i_fop = &omfs_dir_operations; 247 inode->i_size = sbi->s_sys_blocksize; 248 inc_nlink(inode); 249 break; 250 case OMFS_FILE: 251 inode->i_mode = S_IFREG | (S_IRWXUGO & ~sbi->s_fmask); 252 inode->i_fop = &omfs_file_operations; 253 inode->i_size = be64_to_cpu(oi->i_size); 254 break; 255 } 256 brelse(bh); 257 unlock_new_inode(inode); 258 return inode; 259 fail_bh: 260 brelse(bh); 261 iget_failed: 262 iget_failed(inode); 263 return ERR_PTR(-EIO); 264 } 265 266 static void omfs_put_super(struct super_block *sb) 267 { 268 struct omfs_sb_info *sbi = OMFS_SB(sb); 269 kfree(sbi->s_imap); 270 kfree(sbi); 271 sb->s_fs_info = NULL; 272 } 273 274 static int omfs_statfs(struct dentry *dentry, struct kstatfs *buf) 275 { 276 struct super_block *s = dentry->d_sb; 277 struct omfs_sb_info *sbi = OMFS_SB(s); 278 u64 id = huge_encode_dev(s->s_bdev->bd_dev); 279 280 buf->f_type = OMFS_MAGIC; 281 buf->f_bsize = sbi->s_blocksize; 282 buf->f_blocks = sbi->s_num_blocks; 283 buf->f_files = sbi->s_num_blocks; 284 buf->f_namelen = OMFS_NAMELEN; 285 buf->f_fsid.val[0] = (u32)id; 286 buf->f_fsid.val[1] = (u32)(id >> 32); 287 288 buf->f_bfree = buf->f_bavail = buf->f_ffree = 289 omfs_count_free(s); 290 291 return 0; 292 } 293 294 /* 295 * Display the mount options in /proc/mounts. 296 */ 297 static int omfs_show_options(struct seq_file *m, struct dentry *root) 298 { 299 struct omfs_sb_info *sbi = OMFS_SB(root->d_sb); 300 umode_t cur_umask = current_umask(); 301 302 if (!uid_eq(sbi->s_uid, current_uid())) 303 seq_printf(m, ",uid=%u", 304 from_kuid_munged(&init_user_ns, sbi->s_uid)); 305 if (!gid_eq(sbi->s_gid, current_gid())) 306 seq_printf(m, ",gid=%u", 307 from_kgid_munged(&init_user_ns, sbi->s_gid)); 308 309 if (sbi->s_dmask == sbi->s_fmask) { 310 if (sbi->s_fmask != cur_umask) 311 seq_printf(m, ",umask=%o", sbi->s_fmask); 312 } else { 313 if (sbi->s_dmask != cur_umask) 314 seq_printf(m, ",dmask=%o", sbi->s_dmask); 315 if (sbi->s_fmask != cur_umask) 316 seq_printf(m, ",fmask=%o", sbi->s_fmask); 317 } 318 319 return 0; 320 } 321 322 static const struct super_operations omfs_sops = { 323 .write_inode = omfs_write_inode, 324 .evict_inode = omfs_evict_inode, 325 .put_super = omfs_put_super, 326 .statfs = omfs_statfs, 327 .show_options = omfs_show_options, 328 }; 329 330 /* 331 * For Rio Karma, there is an on-disk free bitmap whose location is 332 * stored in the root block. For ReplayTV, there is no such free bitmap 333 * so we have to walk the tree. Both inodes and file data are allocated 334 * from the same map. This array can be big (300k) so we allocate 335 * in units of the blocksize. 336 */ 337 static int omfs_get_imap(struct super_block *sb) 338 { 339 unsigned int bitmap_size, array_size; 340 int count; 341 struct omfs_sb_info *sbi = OMFS_SB(sb); 342 struct buffer_head *bh; 343 unsigned long **ptr; 344 sector_t block; 345 346 bitmap_size = DIV_ROUND_UP(sbi->s_num_blocks, 8); 347 array_size = DIV_ROUND_UP(bitmap_size, sb->s_blocksize); 348 349 if (sbi->s_bitmap_ino == ~0ULL) 350 goto out; 351 352 sbi->s_imap_size = array_size; 353 sbi->s_imap = kcalloc(array_size, sizeof(unsigned long *), GFP_KERNEL); 354 if (!sbi->s_imap) 355 goto nomem; 356 357 block = clus_to_blk(sbi, sbi->s_bitmap_ino); 358 if (block >= sbi->s_num_blocks) 359 goto nomem; 360 361 ptr = sbi->s_imap; 362 for (count = bitmap_size; count > 0; count -= sb->s_blocksize) { 363 bh = sb_bread(sb, block++); 364 if (!bh) 365 goto nomem_free; 366 *ptr = kmalloc(sb->s_blocksize, GFP_KERNEL); 367 if (!*ptr) { 368 brelse(bh); 369 goto nomem_free; 370 } 371 memcpy(*ptr, bh->b_data, sb->s_blocksize); 372 if (count < sb->s_blocksize) 373 memset((void *)*ptr + count, 0xff, 374 sb->s_blocksize - count); 375 brelse(bh); 376 ptr++; 377 } 378 out: 379 return 0; 380 381 nomem_free: 382 for (count = 0; count < array_size; count++) 383 kfree(sbi->s_imap[count]); 384 385 kfree(sbi->s_imap); 386 nomem: 387 sbi->s_imap = NULL; 388 sbi->s_imap_size = 0; 389 return -ENOMEM; 390 } 391 392 enum { 393 Opt_uid, Opt_gid, Opt_umask, Opt_dmask, Opt_fmask, Opt_err 394 }; 395 396 static const match_table_t tokens = { 397 {Opt_uid, "uid=%u"}, 398 {Opt_gid, "gid=%u"}, 399 {Opt_umask, "umask=%o"}, 400 {Opt_dmask, "dmask=%o"}, 401 {Opt_fmask, "fmask=%o"}, 402 {Opt_err, NULL}, 403 }; 404 405 static int parse_options(char *options, struct omfs_sb_info *sbi) 406 { 407 char *p; 408 substring_t args[MAX_OPT_ARGS]; 409 int option; 410 411 if (!options) 412 return 1; 413 414 while ((p = strsep(&options, ",")) != NULL) { 415 int token; 416 if (!*p) 417 continue; 418 419 token = match_token(p, tokens, args); 420 switch (token) { 421 case Opt_uid: 422 if (match_int(&args[0], &option)) 423 return 0; 424 sbi->s_uid = make_kuid(current_user_ns(), option); 425 if (!uid_valid(sbi->s_uid)) 426 return 0; 427 break; 428 case Opt_gid: 429 if (match_int(&args[0], &option)) 430 return 0; 431 sbi->s_gid = make_kgid(current_user_ns(), option); 432 if (!gid_valid(sbi->s_gid)) 433 return 0; 434 break; 435 case Opt_umask: 436 if (match_octal(&args[0], &option)) 437 return 0; 438 sbi->s_fmask = sbi->s_dmask = option; 439 break; 440 case Opt_dmask: 441 if (match_octal(&args[0], &option)) 442 return 0; 443 sbi->s_dmask = option; 444 break; 445 case Opt_fmask: 446 if (match_octal(&args[0], &option)) 447 return 0; 448 sbi->s_fmask = option; 449 break; 450 default: 451 return 0; 452 } 453 } 454 return 1; 455 } 456 457 static int omfs_fill_super(struct super_block *sb, void *data, int silent) 458 { 459 struct buffer_head *bh, *bh2; 460 struct omfs_super_block *omfs_sb; 461 struct omfs_root_block *omfs_rb; 462 struct omfs_sb_info *sbi; 463 struct inode *root; 464 int ret = -EINVAL; 465 466 sbi = kzalloc(sizeof(struct omfs_sb_info), GFP_KERNEL); 467 if (!sbi) 468 return -ENOMEM; 469 470 sb->s_fs_info = sbi; 471 472 sbi->s_uid = current_uid(); 473 sbi->s_gid = current_gid(); 474 sbi->s_dmask = sbi->s_fmask = current_umask(); 475 476 if (!parse_options((char *) data, sbi)) 477 goto end; 478 479 sb->s_maxbytes = 0xffffffff; 480 481 sb_set_blocksize(sb, 0x200); 482 483 bh = sb_bread(sb, 0); 484 if (!bh) 485 goto end; 486 487 omfs_sb = (struct omfs_super_block *)bh->b_data; 488 489 if (omfs_sb->s_magic != cpu_to_be32(OMFS_MAGIC)) { 490 if (!silent) 491 printk(KERN_ERR "omfs: Invalid superblock (%x)\n", 492 omfs_sb->s_magic); 493 goto out_brelse_bh; 494 } 495 sb->s_magic = OMFS_MAGIC; 496 497 sbi->s_num_blocks = be64_to_cpu(omfs_sb->s_num_blocks); 498 sbi->s_blocksize = be32_to_cpu(omfs_sb->s_blocksize); 499 sbi->s_mirrors = be32_to_cpu(omfs_sb->s_mirrors); 500 sbi->s_root_ino = be64_to_cpu(omfs_sb->s_root_block); 501 sbi->s_sys_blocksize = be32_to_cpu(omfs_sb->s_sys_blocksize); 502 mutex_init(&sbi->s_bitmap_lock); 503 504 if (sbi->s_num_blocks > OMFS_MAX_BLOCKS) { 505 printk(KERN_ERR "omfs: sysblock number (%llx) is out of range\n", 506 (unsigned long long)sbi->s_num_blocks); 507 goto out_brelse_bh; 508 } 509 510 if (sbi->s_sys_blocksize > PAGE_SIZE) { 511 printk(KERN_ERR "omfs: sysblock size (%d) is out of range\n", 512 sbi->s_sys_blocksize); 513 goto out_brelse_bh; 514 } 515 516 if (sbi->s_blocksize < sbi->s_sys_blocksize || 517 sbi->s_blocksize > OMFS_MAX_BLOCK_SIZE) { 518 printk(KERN_ERR "omfs: block size (%d) is out of range\n", 519 sbi->s_blocksize); 520 goto out_brelse_bh; 521 } 522 523 /* 524 * Use sys_blocksize as the fs block since it is smaller than a 525 * page while the fs blocksize can be larger. 526 */ 527 sb_set_blocksize(sb, sbi->s_sys_blocksize); 528 529 /* 530 * ...and the difference goes into a shift. sys_blocksize is always 531 * a power of two factor of blocksize. 532 */ 533 sbi->s_block_shift = get_bitmask_order(sbi->s_blocksize) - 534 get_bitmask_order(sbi->s_sys_blocksize); 535 536 bh2 = omfs_bread(sb, be64_to_cpu(omfs_sb->s_root_block)); 537 if (!bh2) 538 goto out_brelse_bh; 539 540 omfs_rb = (struct omfs_root_block *)bh2->b_data; 541 542 sbi->s_bitmap_ino = be64_to_cpu(omfs_rb->r_bitmap); 543 sbi->s_clustersize = be32_to_cpu(omfs_rb->r_clustersize); 544 545 if (sbi->s_num_blocks != be64_to_cpu(omfs_rb->r_num_blocks)) { 546 printk(KERN_ERR "omfs: block count discrepancy between " 547 "super and root blocks (%llx, %llx)\n", 548 (unsigned long long)sbi->s_num_blocks, 549 (unsigned long long)be64_to_cpu(omfs_rb->r_num_blocks)); 550 goto out_brelse_bh2; 551 } 552 553 if (sbi->s_bitmap_ino != ~0ULL && 554 sbi->s_bitmap_ino > sbi->s_num_blocks) { 555 printk(KERN_ERR "omfs: free space bitmap location is corrupt " 556 "(%llx, total blocks %llx)\n", 557 (unsigned long long) sbi->s_bitmap_ino, 558 (unsigned long long) sbi->s_num_blocks); 559 goto out_brelse_bh2; 560 } 561 if (sbi->s_clustersize < 1 || 562 sbi->s_clustersize > OMFS_MAX_CLUSTER_SIZE) { 563 printk(KERN_ERR "omfs: cluster size out of range (%d)", 564 sbi->s_clustersize); 565 goto out_brelse_bh2; 566 } 567 568 ret = omfs_get_imap(sb); 569 if (ret) 570 goto out_brelse_bh2; 571 572 sb->s_op = &omfs_sops; 573 574 root = omfs_iget(sb, be64_to_cpu(omfs_rb->r_root_dir)); 575 if (IS_ERR(root)) { 576 ret = PTR_ERR(root); 577 goto out_brelse_bh2; 578 } 579 580 sb->s_root = d_make_root(root); 581 if (!sb->s_root) { 582 ret = -ENOMEM; 583 goto out_brelse_bh2; 584 } 585 printk(KERN_DEBUG "omfs: Mounted volume %s\n", omfs_rb->r_name); 586 587 ret = 0; 588 out_brelse_bh2: 589 brelse(bh2); 590 out_brelse_bh: 591 brelse(bh); 592 end: 593 if (ret) 594 kfree(sbi); 595 return ret; 596 } 597 598 static struct dentry *omfs_mount(struct file_system_type *fs_type, 599 int flags, const char *dev_name, void *data) 600 { 601 return mount_bdev(fs_type, flags, dev_name, data, omfs_fill_super); 602 } 603 604 static struct file_system_type omfs_fs_type = { 605 .owner = THIS_MODULE, 606 .name = "omfs", 607 .mount = omfs_mount, 608 .kill_sb = kill_block_super, 609 .fs_flags = FS_REQUIRES_DEV, 610 }; 611 MODULE_ALIAS_FS("omfs"); 612 613 static int __init init_omfs_fs(void) 614 { 615 return register_filesystem(&omfs_fs_type); 616 } 617 618 static void __exit exit_omfs_fs(void) 619 { 620 unregister_filesystem(&omfs_fs_type); 621 } 622 623 module_init(init_omfs_fs); 624 module_exit(exit_omfs_fs); 625