inode.c 14 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597
  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. MODULE_AUTHOR("Bob Copeland <me@bobcopeland.com>");
  19. MODULE_DESCRIPTION("OMFS (ReplayTV/Karma) Filesystem for Linux");
  20. MODULE_LICENSE("GPL");
  21. struct buffer_head *omfs_bread(struct super_block *sb, sector_t block)
  22. {
  23. struct omfs_sb_info *sbi = OMFS_SB(sb);
  24. if (block >= sbi->s_num_blocks)
  25. return NULL;
  26. return sb_bread(sb, clus_to_blk(sbi, block));
  27. }
  28. struct inode *omfs_new_inode(struct inode *dir, umode_t mode)
  29. {
  30. struct inode *inode;
  31. u64 new_block;
  32. int err;
  33. int len;
  34. struct omfs_sb_info *sbi = OMFS_SB(dir->i_sb);
  35. inode = new_inode(dir->i_sb);
  36. if (!inode)
  37. return ERR_PTR(-ENOMEM);
  38. err = omfs_allocate_range(dir->i_sb, sbi->s_mirrors, sbi->s_mirrors,
  39. &new_block, &len);
  40. if (err)
  41. goto fail;
  42. inode->i_ino = new_block;
  43. inode_init_owner(inode, NULL, mode);
  44. inode->i_mapping->a_ops = &omfs_aops;
  45. inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
  46. switch (mode & S_IFMT) {
  47. case S_IFDIR:
  48. inode->i_op = &omfs_dir_inops;
  49. inode->i_fop = &omfs_dir_operations;
  50. inode->i_size = sbi->s_sys_blocksize;
  51. inc_nlink(inode);
  52. break;
  53. case S_IFREG:
  54. inode->i_op = &omfs_file_inops;
  55. inode->i_fop = &omfs_file_operations;
  56. inode->i_size = 0;
  57. break;
  58. }
  59. insert_inode_hash(inode);
  60. mark_inode_dirty(inode);
  61. return inode;
  62. fail:
  63. make_bad_inode(inode);
  64. iput(inode);
  65. return ERR_PTR(err);
  66. }
  67. /*
  68. * Update the header checksums for a dirty inode based on its contents.
  69. * Caller is expected to hold the buffer head underlying oi and mark it
  70. * dirty.
  71. */
  72. static void omfs_update_checksums(struct omfs_inode *oi)
  73. {
  74. int xor, i, ofs = 0, count;
  75. u16 crc = 0;
  76. unsigned char *ptr = (unsigned char *) oi;
  77. count = be32_to_cpu(oi->i_head.h_body_size);
  78. ofs = sizeof(struct omfs_header);
  79. crc = crc_itu_t(crc, ptr + ofs, count);
  80. oi->i_head.h_crc = cpu_to_be16(crc);
  81. xor = ptr[0];
  82. for (i = 1; i < OMFS_XOR_COUNT; i++)
  83. xor ^= ptr[i];
  84. oi->i_head.h_check_xor = xor;
  85. }
  86. static int __omfs_write_inode(struct inode *inode, int wait)
  87. {
  88. struct omfs_inode *oi;
  89. struct omfs_sb_info *sbi = OMFS_SB(inode->i_sb);
  90. struct buffer_head *bh, *bh2;
  91. u64 ctime;
  92. int i;
  93. int ret = -EIO;
  94. int sync_failed = 0;
  95. /* get current inode since we may have written sibling ptrs etc. */
  96. bh = omfs_bread(inode->i_sb, inode->i_ino);
  97. if (!bh)
  98. goto out;
  99. oi = (struct omfs_inode *) bh->b_data;
  100. oi->i_head.h_self = cpu_to_be64(inode->i_ino);
  101. if (S_ISDIR(inode->i_mode))
  102. oi->i_type = OMFS_DIR;
  103. else if (S_ISREG(inode->i_mode))
  104. oi->i_type = OMFS_FILE;
  105. else {
  106. printk(KERN_WARNING "omfs: unknown file type: %d\n",
  107. inode->i_mode);
  108. goto out_brelse;
  109. }
  110. oi->i_head.h_body_size = cpu_to_be32(sbi->s_sys_blocksize -
  111. sizeof(struct omfs_header));
  112. oi->i_head.h_version = 1;
  113. oi->i_head.h_type = OMFS_INODE_NORMAL;
  114. oi->i_head.h_magic = OMFS_IMAGIC;
  115. oi->i_size = cpu_to_be64(inode->i_size);
  116. ctime = inode->i_ctime.tv_sec * 1000LL +
  117. ((inode->i_ctime.tv_nsec + 999)/1000);
  118. oi->i_ctime = cpu_to_be64(ctime);
  119. omfs_update_checksums(oi);
  120. mark_buffer_dirty(bh);
  121. if (wait) {
  122. sync_dirty_buffer(bh);
  123. if (buffer_req(bh) && !buffer_uptodate(bh))
  124. sync_failed = 1;
  125. }
  126. /* if mirroring writes, copy to next fsblock */
  127. for (i = 1; i < sbi->s_mirrors; i++) {
  128. bh2 = omfs_bread(inode->i_sb, inode->i_ino + i);
  129. if (!bh2)
  130. goto out_brelse;
  131. memcpy(bh2->b_data, bh->b_data, bh->b_size);
  132. mark_buffer_dirty(bh2);
  133. if (wait) {
  134. sync_dirty_buffer(bh2);
  135. if (buffer_req(bh2) && !buffer_uptodate(bh2))
  136. sync_failed = 1;
  137. }
  138. brelse(bh2);
  139. }
  140. ret = (sync_failed) ? -EIO : 0;
  141. out_brelse:
  142. brelse(bh);
  143. out:
  144. return ret;
  145. }
  146. static int omfs_write_inode(struct inode *inode, struct writeback_control *wbc)
  147. {
  148. return __omfs_write_inode(inode, wbc->sync_mode == WB_SYNC_ALL);
  149. }
  150. int omfs_sync_inode(struct inode *inode)
  151. {
  152. return __omfs_write_inode(inode, 1);
  153. }
  154. /*
  155. * called when an entry is deleted, need to clear the bits in the
  156. * bitmaps.
  157. */
  158. static void omfs_evict_inode(struct inode *inode)
  159. {
  160. truncate_inode_pages_final(&inode->i_data);
  161. clear_inode(inode);
  162. if (inode->i_nlink)
  163. return;
  164. if (S_ISREG(inode->i_mode)) {
  165. inode->i_size = 0;
  166. omfs_shrink_inode(inode);
  167. }
  168. omfs_clear_range(inode->i_sb, inode->i_ino, 2);
  169. }
  170. struct inode *omfs_iget(struct super_block *sb, ino_t ino)
  171. {
  172. struct omfs_sb_info *sbi = OMFS_SB(sb);
  173. struct omfs_inode *oi;
  174. struct buffer_head *bh;
  175. u64 ctime;
  176. unsigned long nsecs;
  177. struct inode *inode;
  178. inode = iget_locked(sb, ino);
  179. if (!inode)
  180. return ERR_PTR(-ENOMEM);
  181. if (!(inode->i_state & I_NEW))
  182. return inode;
  183. bh = omfs_bread(inode->i_sb, ino);
  184. if (!bh)
  185. goto iget_failed;
  186. oi = (struct omfs_inode *)bh->b_data;
  187. /* check self */
  188. if (ino != be64_to_cpu(oi->i_head.h_self))
  189. goto fail_bh;
  190. inode->i_uid = sbi->s_uid;
  191. inode->i_gid = sbi->s_gid;
  192. ctime = be64_to_cpu(oi->i_ctime);
  193. nsecs = do_div(ctime, 1000) * 1000L;
  194. inode->i_atime.tv_sec = ctime;
  195. inode->i_mtime.tv_sec = ctime;
  196. inode->i_ctime.tv_sec = ctime;
  197. inode->i_atime.tv_nsec = nsecs;
  198. inode->i_mtime.tv_nsec = nsecs;
  199. inode->i_ctime.tv_nsec = nsecs;
  200. inode->i_mapping->a_ops = &omfs_aops;
  201. switch (oi->i_type) {
  202. case OMFS_DIR:
  203. inode->i_mode = S_IFDIR | (S_IRWXUGO & ~sbi->s_dmask);
  204. inode->i_op = &omfs_dir_inops;
  205. inode->i_fop = &omfs_dir_operations;
  206. inode->i_size = sbi->s_sys_blocksize;
  207. inc_nlink(inode);
  208. break;
  209. case OMFS_FILE:
  210. inode->i_mode = S_IFREG | (S_IRWXUGO & ~sbi->s_fmask);
  211. inode->i_fop = &omfs_file_operations;
  212. inode->i_size = be64_to_cpu(oi->i_size);
  213. break;
  214. }
  215. brelse(bh);
  216. unlock_new_inode(inode);
  217. return inode;
  218. fail_bh:
  219. brelse(bh);
  220. iget_failed:
  221. iget_failed(inode);
  222. return ERR_PTR(-EIO);
  223. }
  224. static void omfs_put_super(struct super_block *sb)
  225. {
  226. struct omfs_sb_info *sbi = OMFS_SB(sb);
  227. kfree(sbi->s_imap);
  228. kfree(sbi);
  229. sb->s_fs_info = NULL;
  230. }
  231. static int omfs_statfs(struct dentry *dentry, struct kstatfs *buf)
  232. {
  233. struct super_block *s = dentry->d_sb;
  234. struct omfs_sb_info *sbi = OMFS_SB(s);
  235. u64 id = huge_encode_dev(s->s_bdev->bd_dev);
  236. buf->f_type = OMFS_MAGIC;
  237. buf->f_bsize = sbi->s_blocksize;
  238. buf->f_blocks = sbi->s_num_blocks;
  239. buf->f_files = sbi->s_num_blocks;
  240. buf->f_namelen = OMFS_NAMELEN;
  241. buf->f_fsid.val[0] = (u32)id;
  242. buf->f_fsid.val[1] = (u32)(id >> 32);
  243. buf->f_bfree = buf->f_bavail = buf->f_ffree =
  244. omfs_count_free(s);
  245. return 0;
  246. }
  247. static const struct super_operations omfs_sops = {
  248. .write_inode = omfs_write_inode,
  249. .evict_inode = omfs_evict_inode,
  250. .put_super = omfs_put_super,
  251. .statfs = omfs_statfs,
  252. .show_options = generic_show_options,
  253. };
  254. /*
  255. * For Rio Karma, there is an on-disk free bitmap whose location is
  256. * stored in the root block. For ReplayTV, there is no such free bitmap
  257. * so we have to walk the tree. Both inodes and file data are allocated
  258. * from the same map. This array can be big (300k) so we allocate
  259. * in units of the blocksize.
  260. */
  261. static int omfs_get_imap(struct super_block *sb)
  262. {
  263. unsigned int bitmap_size, array_size;
  264. int count;
  265. struct omfs_sb_info *sbi = OMFS_SB(sb);
  266. struct buffer_head *bh;
  267. unsigned long **ptr;
  268. sector_t block;
  269. bitmap_size = DIV_ROUND_UP(sbi->s_num_blocks, 8);
  270. array_size = DIV_ROUND_UP(bitmap_size, sb->s_blocksize);
  271. if (sbi->s_bitmap_ino == ~0ULL)
  272. goto out;
  273. sbi->s_imap_size = array_size;
  274. sbi->s_imap = kcalloc(array_size, sizeof(unsigned long *), GFP_KERNEL);
  275. if (!sbi->s_imap)
  276. goto nomem;
  277. block = clus_to_blk(sbi, sbi->s_bitmap_ino);
  278. if (block >= sbi->s_num_blocks)
  279. goto nomem;
  280. ptr = sbi->s_imap;
  281. for (count = bitmap_size; count > 0; count -= sb->s_blocksize) {
  282. bh = sb_bread(sb, block++);
  283. if (!bh)
  284. goto nomem_free;
  285. *ptr = kmalloc(sb->s_blocksize, GFP_KERNEL);
  286. if (!*ptr) {
  287. brelse(bh);
  288. goto nomem_free;
  289. }
  290. memcpy(*ptr, bh->b_data, sb->s_blocksize);
  291. if (count < sb->s_blocksize)
  292. memset((void *)*ptr + count, 0xff,
  293. sb->s_blocksize - count);
  294. brelse(bh);
  295. ptr++;
  296. }
  297. out:
  298. return 0;
  299. nomem_free:
  300. for (count = 0; count < array_size; count++)
  301. kfree(sbi->s_imap[count]);
  302. kfree(sbi->s_imap);
  303. nomem:
  304. sbi->s_imap = NULL;
  305. sbi->s_imap_size = 0;
  306. return -ENOMEM;
  307. }
  308. enum {
  309. Opt_uid, Opt_gid, Opt_umask, Opt_dmask, Opt_fmask, Opt_err
  310. };
  311. static const match_table_t tokens = {
  312. {Opt_uid, "uid=%u"},
  313. {Opt_gid, "gid=%u"},
  314. {Opt_umask, "umask=%o"},
  315. {Opt_dmask, "dmask=%o"},
  316. {Opt_fmask, "fmask=%o"},
  317. {Opt_err, NULL},
  318. };
  319. static int parse_options(char *options, struct omfs_sb_info *sbi)
  320. {
  321. char *p;
  322. substring_t args[MAX_OPT_ARGS];
  323. int option;
  324. if (!options)
  325. return 1;
  326. while ((p = strsep(&options, ",")) != NULL) {
  327. int token;
  328. if (!*p)
  329. continue;
  330. token = match_token(p, tokens, args);
  331. switch (token) {
  332. case Opt_uid:
  333. if (match_int(&args[0], &option))
  334. return 0;
  335. sbi->s_uid = make_kuid(current_user_ns(), option);
  336. if (!uid_valid(sbi->s_uid))
  337. return 0;
  338. break;
  339. case Opt_gid:
  340. if (match_int(&args[0], &option))
  341. return 0;
  342. sbi->s_gid = make_kgid(current_user_ns(), option);
  343. if (!gid_valid(sbi->s_gid))
  344. return 0;
  345. break;
  346. case Opt_umask:
  347. if (match_octal(&args[0], &option))
  348. return 0;
  349. sbi->s_fmask = sbi->s_dmask = option;
  350. break;
  351. case Opt_dmask:
  352. if (match_octal(&args[0], &option))
  353. return 0;
  354. sbi->s_dmask = option;
  355. break;
  356. case Opt_fmask:
  357. if (match_octal(&args[0], &option))
  358. return 0;
  359. sbi->s_fmask = option;
  360. break;
  361. default:
  362. return 0;
  363. }
  364. }
  365. return 1;
  366. }
  367. static int omfs_fill_super(struct super_block *sb, void *data, int silent)
  368. {
  369. struct buffer_head *bh, *bh2;
  370. struct omfs_super_block *omfs_sb;
  371. struct omfs_root_block *omfs_rb;
  372. struct omfs_sb_info *sbi;
  373. struct inode *root;
  374. int ret = -EINVAL;
  375. save_mount_options(sb, (char *) data);
  376. sbi = kzalloc(sizeof(struct omfs_sb_info), GFP_KERNEL);
  377. if (!sbi)
  378. return -ENOMEM;
  379. sb->s_fs_info = sbi;
  380. sbi->s_uid = current_uid();
  381. sbi->s_gid = current_gid();
  382. sbi->s_dmask = sbi->s_fmask = current_umask();
  383. if (!parse_options((char *) data, sbi))
  384. goto end;
  385. sb->s_maxbytes = 0xffffffff;
  386. sb_set_blocksize(sb, 0x200);
  387. bh = sb_bread(sb, 0);
  388. if (!bh)
  389. goto end;
  390. omfs_sb = (struct omfs_super_block *)bh->b_data;
  391. if (omfs_sb->s_magic != cpu_to_be32(OMFS_MAGIC)) {
  392. if (!silent)
  393. printk(KERN_ERR "omfs: Invalid superblock (%x)\n",
  394. omfs_sb->s_magic);
  395. goto out_brelse_bh;
  396. }
  397. sb->s_magic = OMFS_MAGIC;
  398. sbi->s_num_blocks = be64_to_cpu(omfs_sb->s_num_blocks);
  399. sbi->s_blocksize = be32_to_cpu(omfs_sb->s_blocksize);
  400. sbi->s_mirrors = be32_to_cpu(omfs_sb->s_mirrors);
  401. sbi->s_root_ino = be64_to_cpu(omfs_sb->s_root_block);
  402. sbi->s_sys_blocksize = be32_to_cpu(omfs_sb->s_sys_blocksize);
  403. mutex_init(&sbi->s_bitmap_lock);
  404. if (sbi->s_num_blocks > OMFS_MAX_BLOCKS) {
  405. printk(KERN_ERR "omfs: sysblock number (%llx) is out of range\n",
  406. (unsigned long long)sbi->s_num_blocks);
  407. goto out_brelse_bh;
  408. }
  409. if (sbi->s_sys_blocksize > PAGE_SIZE) {
  410. printk(KERN_ERR "omfs: sysblock size (%d) is out of range\n",
  411. sbi->s_sys_blocksize);
  412. goto out_brelse_bh;
  413. }
  414. if (sbi->s_blocksize < sbi->s_sys_blocksize ||
  415. sbi->s_blocksize > OMFS_MAX_BLOCK_SIZE) {
  416. printk(KERN_ERR "omfs: block size (%d) is out of range\n",
  417. sbi->s_blocksize);
  418. goto out_brelse_bh;
  419. }
  420. /*
  421. * Use sys_blocksize as the fs block since it is smaller than a
  422. * page while the fs blocksize can be larger.
  423. */
  424. sb_set_blocksize(sb, sbi->s_sys_blocksize);
  425. /*
  426. * ...and the difference goes into a shift. sys_blocksize is always
  427. * a power of two factor of blocksize.
  428. */
  429. sbi->s_block_shift = get_bitmask_order(sbi->s_blocksize) -
  430. get_bitmask_order(sbi->s_sys_blocksize);
  431. bh2 = omfs_bread(sb, be64_to_cpu(omfs_sb->s_root_block));
  432. if (!bh2)
  433. goto out_brelse_bh;
  434. omfs_rb = (struct omfs_root_block *)bh2->b_data;
  435. sbi->s_bitmap_ino = be64_to_cpu(omfs_rb->r_bitmap);
  436. sbi->s_clustersize = be32_to_cpu(omfs_rb->r_clustersize);
  437. if (sbi->s_num_blocks != be64_to_cpu(omfs_rb->r_num_blocks)) {
  438. printk(KERN_ERR "omfs: block count discrepancy between "
  439. "super and root blocks (%llx, %llx)\n",
  440. (unsigned long long)sbi->s_num_blocks,
  441. (unsigned long long)be64_to_cpu(omfs_rb->r_num_blocks));
  442. goto out_brelse_bh2;
  443. }
  444. if (sbi->s_bitmap_ino != ~0ULL &&
  445. sbi->s_bitmap_ino > sbi->s_num_blocks) {
  446. printk(KERN_ERR "omfs: free space bitmap location is corrupt "
  447. "(%llx, total blocks %llx)\n",
  448. (unsigned long long) sbi->s_bitmap_ino,
  449. (unsigned long long) sbi->s_num_blocks);
  450. goto out_brelse_bh2;
  451. }
  452. if (sbi->s_clustersize < 1 ||
  453. sbi->s_clustersize > OMFS_MAX_CLUSTER_SIZE) {
  454. printk(KERN_ERR "omfs: cluster size out of range (%d)",
  455. sbi->s_clustersize);
  456. goto out_brelse_bh2;
  457. }
  458. ret = omfs_get_imap(sb);
  459. if (ret)
  460. goto out_brelse_bh2;
  461. sb->s_op = &omfs_sops;
  462. root = omfs_iget(sb, be64_to_cpu(omfs_rb->r_root_dir));
  463. if (IS_ERR(root)) {
  464. ret = PTR_ERR(root);
  465. goto out_brelse_bh2;
  466. }
  467. sb->s_root = d_make_root(root);
  468. if (!sb->s_root) {
  469. ret = -ENOMEM;
  470. goto out_brelse_bh2;
  471. }
  472. printk(KERN_DEBUG "omfs: Mounted volume %s\n", omfs_rb->r_name);
  473. ret = 0;
  474. out_brelse_bh2:
  475. brelse(bh2);
  476. out_brelse_bh:
  477. brelse(bh);
  478. end:
  479. if (ret)
  480. kfree(sbi);
  481. return ret;
  482. }
  483. static struct dentry *omfs_mount(struct file_system_type *fs_type,
  484. int flags, const char *dev_name, void *data)
  485. {
  486. return mount_bdev(fs_type, flags, dev_name, data, omfs_fill_super);
  487. }
  488. static struct file_system_type omfs_fs_type = {
  489. .owner = THIS_MODULE,
  490. .name = "omfs",
  491. .mount = omfs_mount,
  492. .kill_sb = kill_block_super,
  493. .fs_flags = FS_REQUIRES_DEV,
  494. };
  495. MODULE_ALIAS_FS("omfs");
  496. static int __init init_omfs_fs(void)
  497. {
  498. return register_filesystem(&omfs_fs_type);
  499. }
  500. static void __exit exit_omfs_fs(void)
  501. {
  502. unregister_filesystem(&omfs_fs_type);
  503. }
  504. module_init(init_omfs_fs);
  505. module_exit(exit_omfs_fs);