inode.c 12 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521
  1. /*
  2. * Hypervisor filesystem for Linux on s390.
  3. *
  4. * Copyright IBM Corp. 2006, 2008
  5. * Author(s): Michael Holzheu <holzheu@de.ibm.com>
  6. */
  7. #define KMSG_COMPONENT "hypfs"
  8. #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
  9. #include <linux/types.h>
  10. #include <linux/errno.h>
  11. #include <linux/fs.h>
  12. #include <linux/namei.h>
  13. #include <linux/vfs.h>
  14. #include <linux/slab.h>
  15. #include <linux/pagemap.h>
  16. #include <linux/time.h>
  17. #include <linux/parser.h>
  18. #include <linux/sysfs.h>
  19. #include <linux/module.h>
  20. #include <linux/seq_file.h>
  21. #include <linux/mount.h>
  22. #include <linux/aio.h>
  23. #include <asm/ebcdic.h>
  24. #include "hypfs.h"
  25. #define HYPFS_MAGIC 0x687970 /* ASCII 'hyp' */
  26. #define TMP_SIZE 64 /* size of temporary buffers */
  27. static struct dentry *hypfs_create_update_file(struct super_block *sb,
  28. struct dentry *dir);
  29. struct hypfs_sb_info {
  30. kuid_t uid; /* uid used for files and dirs */
  31. kgid_t gid; /* gid used for files and dirs */
  32. struct dentry *update_file; /* file to trigger update */
  33. time_t last_update; /* last update time in secs since 1970 */
  34. struct mutex lock; /* lock to protect update process */
  35. };
  36. static const struct file_operations hypfs_file_ops;
  37. static struct file_system_type hypfs_type;
  38. static const struct super_operations hypfs_s_ops;
  39. /* start of list of all dentries, which have to be deleted on update */
  40. static struct dentry *hypfs_last_dentry;
  41. static void hypfs_update_update(struct super_block *sb)
  42. {
  43. struct hypfs_sb_info *sb_info = sb->s_fs_info;
  44. struct inode *inode = sb_info->update_file->d_inode;
  45. sb_info->last_update = get_seconds();
  46. inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
  47. }
  48. /* directory tree removal functions */
  49. static void hypfs_add_dentry(struct dentry *dentry)
  50. {
  51. dentry->d_fsdata = hypfs_last_dentry;
  52. hypfs_last_dentry = dentry;
  53. }
  54. static inline int hypfs_positive(struct dentry *dentry)
  55. {
  56. return dentry->d_inode && !d_unhashed(dentry);
  57. }
  58. static void hypfs_remove(struct dentry *dentry)
  59. {
  60. struct dentry *parent;
  61. parent = dentry->d_parent;
  62. mutex_lock(&parent->d_inode->i_mutex);
  63. if (hypfs_positive(dentry)) {
  64. if (S_ISDIR(dentry->d_inode->i_mode))
  65. simple_rmdir(parent->d_inode, dentry);
  66. else
  67. simple_unlink(parent->d_inode, dentry);
  68. }
  69. d_delete(dentry);
  70. dput(dentry);
  71. mutex_unlock(&parent->d_inode->i_mutex);
  72. }
  73. static void hypfs_delete_tree(struct dentry *root)
  74. {
  75. while (hypfs_last_dentry) {
  76. struct dentry *next_dentry;
  77. next_dentry = hypfs_last_dentry->d_fsdata;
  78. hypfs_remove(hypfs_last_dentry);
  79. hypfs_last_dentry = next_dentry;
  80. }
  81. }
  82. static struct inode *hypfs_make_inode(struct super_block *sb, umode_t mode)
  83. {
  84. struct inode *ret = new_inode(sb);
  85. if (ret) {
  86. struct hypfs_sb_info *hypfs_info = sb->s_fs_info;
  87. ret->i_ino = get_next_ino();
  88. ret->i_mode = mode;
  89. ret->i_uid = hypfs_info->uid;
  90. ret->i_gid = hypfs_info->gid;
  91. ret->i_atime = ret->i_mtime = ret->i_ctime = CURRENT_TIME;
  92. if (S_ISDIR(mode))
  93. set_nlink(ret, 2);
  94. }
  95. return ret;
  96. }
  97. static void hypfs_evict_inode(struct inode *inode)
  98. {
  99. clear_inode(inode);
  100. kfree(inode->i_private);
  101. }
  102. static int hypfs_open(struct inode *inode, struct file *filp)
  103. {
  104. char *data = file_inode(filp)->i_private;
  105. struct hypfs_sb_info *fs_info;
  106. if (filp->f_mode & FMODE_WRITE) {
  107. if (!(inode->i_mode & S_IWUGO))
  108. return -EACCES;
  109. }
  110. if (filp->f_mode & FMODE_READ) {
  111. if (!(inode->i_mode & S_IRUGO))
  112. return -EACCES;
  113. }
  114. fs_info = inode->i_sb->s_fs_info;
  115. if(data) {
  116. mutex_lock(&fs_info->lock);
  117. filp->private_data = kstrdup(data, GFP_KERNEL);
  118. if (!filp->private_data) {
  119. mutex_unlock(&fs_info->lock);
  120. return -ENOMEM;
  121. }
  122. mutex_unlock(&fs_info->lock);
  123. }
  124. return nonseekable_open(inode, filp);
  125. }
  126. static ssize_t hypfs_aio_read(struct kiocb *iocb, const struct iovec *iov,
  127. unsigned long nr_segs, loff_t offset)
  128. {
  129. char *data;
  130. ssize_t ret;
  131. struct file *filp = iocb->ki_filp;
  132. /* XXX: temporary */
  133. char __user *buf = iov[0].iov_base;
  134. size_t count = iov[0].iov_len;
  135. if (nr_segs != 1)
  136. return -EINVAL;
  137. data = filp->private_data;
  138. ret = simple_read_from_buffer(buf, count, &offset, data, strlen(data));
  139. if (ret <= 0)
  140. return ret;
  141. iocb->ki_pos += ret;
  142. file_accessed(filp);
  143. return ret;
  144. }
  145. static ssize_t hypfs_aio_write(struct kiocb *iocb, const struct iovec *iov,
  146. unsigned long nr_segs, loff_t offset)
  147. {
  148. int rc;
  149. struct super_block *sb = file_inode(iocb->ki_filp)->i_sb;
  150. struct hypfs_sb_info *fs_info = sb->s_fs_info;
  151. size_t count = iov_length(iov, nr_segs);
  152. /*
  153. * Currently we only allow one update per second for two reasons:
  154. * 1. diag 204 is VERY expensive
  155. * 2. If several processes do updates in parallel and then read the
  156. * hypfs data, the likelihood of collisions is reduced, if we restrict
  157. * the minimum update interval. A collision occurs, if during the
  158. * data gathering of one process another process triggers an update
  159. * If the first process wants to ensure consistent data, it has
  160. * to restart data collection in this case.
  161. */
  162. mutex_lock(&fs_info->lock);
  163. if (fs_info->last_update == get_seconds()) {
  164. rc = -EBUSY;
  165. goto out;
  166. }
  167. hypfs_delete_tree(sb->s_root);
  168. if (MACHINE_IS_VM)
  169. rc = hypfs_vm_create_files(sb, sb->s_root);
  170. else
  171. rc = hypfs_diag_create_files(sb, sb->s_root);
  172. if (rc) {
  173. pr_err("Updating the hypfs tree failed\n");
  174. hypfs_delete_tree(sb->s_root);
  175. goto out;
  176. }
  177. hypfs_update_update(sb);
  178. rc = count;
  179. out:
  180. mutex_unlock(&fs_info->lock);
  181. return rc;
  182. }
  183. static int hypfs_release(struct inode *inode, struct file *filp)
  184. {
  185. kfree(filp->private_data);
  186. return 0;
  187. }
  188. enum { opt_uid, opt_gid, opt_err };
  189. static const match_table_t hypfs_tokens = {
  190. {opt_uid, "uid=%u"},
  191. {opt_gid, "gid=%u"},
  192. {opt_err, NULL}
  193. };
  194. static int hypfs_parse_options(char *options, struct super_block *sb)
  195. {
  196. char *str;
  197. substring_t args[MAX_OPT_ARGS];
  198. kuid_t uid;
  199. kgid_t gid;
  200. if (!options)
  201. return 0;
  202. while ((str = strsep(&options, ",")) != NULL) {
  203. int token, option;
  204. struct hypfs_sb_info *hypfs_info = sb->s_fs_info;
  205. if (!*str)
  206. continue;
  207. token = match_token(str, hypfs_tokens, args);
  208. switch (token) {
  209. case opt_uid:
  210. if (match_int(&args[0], &option))
  211. return -EINVAL;
  212. uid = make_kuid(current_user_ns(), option);
  213. if (!uid_valid(uid))
  214. return -EINVAL;
  215. hypfs_info->uid = uid;
  216. break;
  217. case opt_gid:
  218. if (match_int(&args[0], &option))
  219. return -EINVAL;
  220. gid = make_kgid(current_user_ns(), option);
  221. if (!gid_valid(gid))
  222. return -EINVAL;
  223. hypfs_info->gid = gid;
  224. break;
  225. case opt_err:
  226. default:
  227. pr_err("%s is not a valid mount option\n", str);
  228. return -EINVAL;
  229. }
  230. }
  231. return 0;
  232. }
  233. static int hypfs_show_options(struct seq_file *s, struct dentry *root)
  234. {
  235. struct hypfs_sb_info *hypfs_info = root->d_sb->s_fs_info;
  236. seq_printf(s, ",uid=%u", from_kuid_munged(&init_user_ns, hypfs_info->uid));
  237. seq_printf(s, ",gid=%u", from_kgid_munged(&init_user_ns, hypfs_info->gid));
  238. return 0;
  239. }
  240. static int hypfs_fill_super(struct super_block *sb, void *data, int silent)
  241. {
  242. struct inode *root_inode;
  243. struct dentry *root_dentry;
  244. int rc = 0;
  245. struct hypfs_sb_info *sbi;
  246. sbi = kzalloc(sizeof(struct hypfs_sb_info), GFP_KERNEL);
  247. if (!sbi)
  248. return -ENOMEM;
  249. mutex_init(&sbi->lock);
  250. sbi->uid = current_uid();
  251. sbi->gid = current_gid();
  252. sb->s_fs_info = sbi;
  253. sb->s_blocksize = PAGE_CACHE_SIZE;
  254. sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
  255. sb->s_magic = HYPFS_MAGIC;
  256. sb->s_op = &hypfs_s_ops;
  257. if (hypfs_parse_options(data, sb))
  258. return -EINVAL;
  259. root_inode = hypfs_make_inode(sb, S_IFDIR | 0755);
  260. if (!root_inode)
  261. return -ENOMEM;
  262. root_inode->i_op = &simple_dir_inode_operations;
  263. root_inode->i_fop = &simple_dir_operations;
  264. sb->s_root = root_dentry = d_make_root(root_inode);
  265. if (!root_dentry)
  266. return -ENOMEM;
  267. if (MACHINE_IS_VM)
  268. rc = hypfs_vm_create_files(sb, root_dentry);
  269. else
  270. rc = hypfs_diag_create_files(sb, root_dentry);
  271. if (rc)
  272. return rc;
  273. sbi->update_file = hypfs_create_update_file(sb, root_dentry);
  274. if (IS_ERR(sbi->update_file))
  275. return PTR_ERR(sbi->update_file);
  276. hypfs_update_update(sb);
  277. pr_info("Hypervisor filesystem mounted\n");
  278. return 0;
  279. }
  280. static struct dentry *hypfs_mount(struct file_system_type *fst, int flags,
  281. const char *devname, void *data)
  282. {
  283. return mount_single(fst, flags, data, hypfs_fill_super);
  284. }
  285. static void hypfs_kill_super(struct super_block *sb)
  286. {
  287. struct hypfs_sb_info *sb_info = sb->s_fs_info;
  288. if (sb->s_root)
  289. hypfs_delete_tree(sb->s_root);
  290. if (sb_info->update_file)
  291. hypfs_remove(sb_info->update_file);
  292. kfree(sb->s_fs_info);
  293. sb->s_fs_info = NULL;
  294. kill_litter_super(sb);
  295. }
  296. static struct dentry *hypfs_create_file(struct super_block *sb,
  297. struct dentry *parent, const char *name,
  298. char *data, umode_t mode)
  299. {
  300. struct dentry *dentry;
  301. struct inode *inode;
  302. mutex_lock(&parent->d_inode->i_mutex);
  303. dentry = lookup_one_len(name, parent, strlen(name));
  304. if (IS_ERR(dentry)) {
  305. dentry = ERR_PTR(-ENOMEM);
  306. goto fail;
  307. }
  308. inode = hypfs_make_inode(sb, mode);
  309. if (!inode) {
  310. dput(dentry);
  311. dentry = ERR_PTR(-ENOMEM);
  312. goto fail;
  313. }
  314. if (S_ISREG(mode)) {
  315. inode->i_fop = &hypfs_file_ops;
  316. if (data)
  317. inode->i_size = strlen(data);
  318. else
  319. inode->i_size = 0;
  320. } else if (S_ISDIR(mode)) {
  321. inode->i_op = &simple_dir_inode_operations;
  322. inode->i_fop = &simple_dir_operations;
  323. inc_nlink(parent->d_inode);
  324. } else
  325. BUG();
  326. inode->i_private = data;
  327. d_instantiate(dentry, inode);
  328. dget(dentry);
  329. fail:
  330. mutex_unlock(&parent->d_inode->i_mutex);
  331. return dentry;
  332. }
  333. struct dentry *hypfs_mkdir(struct super_block *sb, struct dentry *parent,
  334. const char *name)
  335. {
  336. struct dentry *dentry;
  337. dentry = hypfs_create_file(sb, parent, name, NULL, S_IFDIR | DIR_MODE);
  338. if (IS_ERR(dentry))
  339. return dentry;
  340. hypfs_add_dentry(dentry);
  341. return dentry;
  342. }
  343. static struct dentry *hypfs_create_update_file(struct super_block *sb,
  344. struct dentry *dir)
  345. {
  346. struct dentry *dentry;
  347. dentry = hypfs_create_file(sb, dir, "update", NULL,
  348. S_IFREG | UPDATE_FILE_MODE);
  349. /*
  350. * We do not put the update file on the 'delete' list with
  351. * hypfs_add_dentry(), since it should not be removed when the tree
  352. * is updated.
  353. */
  354. return dentry;
  355. }
  356. struct dentry *hypfs_create_u64(struct super_block *sb, struct dentry *dir,
  357. const char *name, __u64 value)
  358. {
  359. char *buffer;
  360. char tmp[TMP_SIZE];
  361. struct dentry *dentry;
  362. snprintf(tmp, TMP_SIZE, "%llu\n", (unsigned long long int)value);
  363. buffer = kstrdup(tmp, GFP_KERNEL);
  364. if (!buffer)
  365. return ERR_PTR(-ENOMEM);
  366. dentry =
  367. hypfs_create_file(sb, dir, name, buffer, S_IFREG | REG_FILE_MODE);
  368. if (IS_ERR(dentry)) {
  369. kfree(buffer);
  370. return ERR_PTR(-ENOMEM);
  371. }
  372. hypfs_add_dentry(dentry);
  373. return dentry;
  374. }
  375. struct dentry *hypfs_create_str(struct super_block *sb, struct dentry *dir,
  376. const char *name, char *string)
  377. {
  378. char *buffer;
  379. struct dentry *dentry;
  380. buffer = kmalloc(strlen(string) + 2, GFP_KERNEL);
  381. if (!buffer)
  382. return ERR_PTR(-ENOMEM);
  383. sprintf(buffer, "%s\n", string);
  384. dentry =
  385. hypfs_create_file(sb, dir, name, buffer, S_IFREG | REG_FILE_MODE);
  386. if (IS_ERR(dentry)) {
  387. kfree(buffer);
  388. return ERR_PTR(-ENOMEM);
  389. }
  390. hypfs_add_dentry(dentry);
  391. return dentry;
  392. }
  393. static const struct file_operations hypfs_file_ops = {
  394. .open = hypfs_open,
  395. .release = hypfs_release,
  396. .read = do_sync_read,
  397. .write = do_sync_write,
  398. .aio_read = hypfs_aio_read,
  399. .aio_write = hypfs_aio_write,
  400. .llseek = no_llseek,
  401. };
  402. static struct file_system_type hypfs_type = {
  403. .owner = THIS_MODULE,
  404. .name = "s390_hypfs",
  405. .mount = hypfs_mount,
  406. .kill_sb = hypfs_kill_super
  407. };
  408. MODULE_ALIAS_FS("s390_hypfs");
  409. static const struct super_operations hypfs_s_ops = {
  410. .statfs = simple_statfs,
  411. .evict_inode = hypfs_evict_inode,
  412. .show_options = hypfs_show_options,
  413. };
  414. static struct kobject *s390_kobj;
  415. static int __init hypfs_init(void)
  416. {
  417. int rc;
  418. rc = hypfs_dbfs_init();
  419. if (rc)
  420. return rc;
  421. if (hypfs_diag_init()) {
  422. rc = -ENODATA;
  423. goto fail_dbfs_exit;
  424. }
  425. if (hypfs_vm_init()) {
  426. rc = -ENODATA;
  427. goto fail_hypfs_diag_exit;
  428. }
  429. s390_kobj = kobject_create_and_add("s390", hypervisor_kobj);
  430. if (!s390_kobj) {
  431. rc = -ENOMEM;
  432. goto fail_hypfs_vm_exit;
  433. }
  434. rc = register_filesystem(&hypfs_type);
  435. if (rc)
  436. goto fail_filesystem;
  437. return 0;
  438. fail_filesystem:
  439. kobject_put(s390_kobj);
  440. fail_hypfs_vm_exit:
  441. hypfs_vm_exit();
  442. fail_hypfs_diag_exit:
  443. hypfs_diag_exit();
  444. fail_dbfs_exit:
  445. hypfs_dbfs_exit();
  446. pr_err("Initialization of hypfs failed with rc=%i\n", rc);
  447. return rc;
  448. }
  449. static void __exit hypfs_exit(void)
  450. {
  451. hypfs_diag_exit();
  452. hypfs_vm_exit();
  453. hypfs_dbfs_exit();
  454. unregister_filesystem(&hypfs_type);
  455. kobject_put(s390_kobj);
  456. }
  457. module_init(hypfs_init)
  458. module_exit(hypfs_exit)
  459. MODULE_LICENSE("GPL");
  460. MODULE_AUTHOR("Michael Holzheu <holzheu@de.ibm.com>");
  461. MODULE_DESCRIPTION("s390 Hypervisor Filesystem");