inode.c 12 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517
  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 dentry *dir);
  28. struct hypfs_sb_info {
  29. kuid_t uid; /* uid used for files and dirs */
  30. kgid_t gid; /* gid used for files and dirs */
  31. struct dentry *update_file; /* file to trigger update */
  32. time_t last_update; /* last update time in secs since 1970 */
  33. struct mutex lock; /* lock to protect update process */
  34. };
  35. static const struct file_operations hypfs_file_ops;
  36. static struct file_system_type hypfs_type;
  37. static const struct super_operations hypfs_s_ops;
  38. /* start of list of all dentries, which have to be deleted on update */
  39. static struct dentry *hypfs_last_dentry;
  40. static void hypfs_update_update(struct super_block *sb)
  41. {
  42. struct hypfs_sb_info *sb_info = sb->s_fs_info;
  43. struct inode *inode = sb_info->update_file->d_inode;
  44. sb_info->last_update = get_seconds();
  45. inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
  46. }
  47. /* directory tree removal functions */
  48. static void hypfs_add_dentry(struct dentry *dentry)
  49. {
  50. dentry->d_fsdata = hypfs_last_dentry;
  51. hypfs_last_dentry = dentry;
  52. }
  53. static inline int hypfs_positive(struct dentry *dentry)
  54. {
  55. return dentry->d_inode && !d_unhashed(dentry);
  56. }
  57. static void hypfs_remove(struct dentry *dentry)
  58. {
  59. struct dentry *parent;
  60. parent = dentry->d_parent;
  61. mutex_lock(&parent->d_inode->i_mutex);
  62. if (hypfs_positive(dentry)) {
  63. if (S_ISDIR(dentry->d_inode->i_mode))
  64. simple_rmdir(parent->d_inode, dentry);
  65. else
  66. simple_unlink(parent->d_inode, dentry);
  67. }
  68. d_delete(dentry);
  69. dput(dentry);
  70. mutex_unlock(&parent->d_inode->i_mutex);
  71. }
  72. static void hypfs_delete_tree(struct dentry *root)
  73. {
  74. while (hypfs_last_dentry) {
  75. struct dentry *next_dentry;
  76. next_dentry = hypfs_last_dentry->d_fsdata;
  77. hypfs_remove(hypfs_last_dentry);
  78. hypfs_last_dentry = next_dentry;
  79. }
  80. }
  81. static struct inode *hypfs_make_inode(struct super_block *sb, umode_t mode)
  82. {
  83. struct inode *ret = new_inode(sb);
  84. if (ret) {
  85. struct hypfs_sb_info *hypfs_info = sb->s_fs_info;
  86. ret->i_ino = get_next_ino();
  87. ret->i_mode = mode;
  88. ret->i_uid = hypfs_info->uid;
  89. ret->i_gid = hypfs_info->gid;
  90. ret->i_atime = ret->i_mtime = ret->i_ctime = CURRENT_TIME;
  91. if (S_ISDIR(mode))
  92. set_nlink(ret, 2);
  93. }
  94. return ret;
  95. }
  96. static void hypfs_evict_inode(struct inode *inode)
  97. {
  98. clear_inode(inode);
  99. kfree(inode->i_private);
  100. }
  101. static int hypfs_open(struct inode *inode, struct file *filp)
  102. {
  103. char *data = file_inode(filp)->i_private;
  104. struct hypfs_sb_info *fs_info;
  105. if (filp->f_mode & FMODE_WRITE) {
  106. if (!(inode->i_mode & S_IWUGO))
  107. return -EACCES;
  108. }
  109. if (filp->f_mode & FMODE_READ) {
  110. if (!(inode->i_mode & S_IRUGO))
  111. return -EACCES;
  112. }
  113. fs_info = inode->i_sb->s_fs_info;
  114. if(data) {
  115. mutex_lock(&fs_info->lock);
  116. filp->private_data = kstrdup(data, GFP_KERNEL);
  117. if (!filp->private_data) {
  118. mutex_unlock(&fs_info->lock);
  119. return -ENOMEM;
  120. }
  121. mutex_unlock(&fs_info->lock);
  122. }
  123. return nonseekable_open(inode, filp);
  124. }
  125. static ssize_t hypfs_aio_read(struct kiocb *iocb, const struct iovec *iov,
  126. unsigned long nr_segs, loff_t offset)
  127. {
  128. char *data;
  129. ssize_t ret;
  130. struct file *filp = iocb->ki_filp;
  131. /* XXX: temporary */
  132. char __user *buf = iov[0].iov_base;
  133. size_t count = iov[0].iov_len;
  134. if (nr_segs != 1)
  135. return -EINVAL;
  136. data = filp->private_data;
  137. ret = simple_read_from_buffer(buf, count, &offset, data, strlen(data));
  138. if (ret <= 0)
  139. return ret;
  140. iocb->ki_pos += ret;
  141. file_accessed(filp);
  142. return ret;
  143. }
  144. static ssize_t hypfs_aio_write(struct kiocb *iocb, const struct iovec *iov,
  145. unsigned long nr_segs, loff_t offset)
  146. {
  147. int rc;
  148. struct super_block *sb = file_inode(iocb->ki_filp)->i_sb;
  149. struct hypfs_sb_info *fs_info = sb->s_fs_info;
  150. size_t count = iov_length(iov, nr_segs);
  151. /*
  152. * Currently we only allow one update per second for two reasons:
  153. * 1. diag 204 is VERY expensive
  154. * 2. If several processes do updates in parallel and then read the
  155. * hypfs data, the likelihood of collisions is reduced, if we restrict
  156. * the minimum update interval. A collision occurs, if during the
  157. * data gathering of one process another process triggers an update
  158. * If the first process wants to ensure consistent data, it has
  159. * to restart data collection in this case.
  160. */
  161. mutex_lock(&fs_info->lock);
  162. if (fs_info->last_update == get_seconds()) {
  163. rc = -EBUSY;
  164. goto out;
  165. }
  166. hypfs_delete_tree(sb->s_root);
  167. if (MACHINE_IS_VM)
  168. rc = hypfs_vm_create_files(sb->s_root);
  169. else
  170. rc = hypfs_diag_create_files(sb->s_root);
  171. if (rc) {
  172. pr_err("Updating the hypfs tree failed\n");
  173. hypfs_delete_tree(sb->s_root);
  174. goto out;
  175. }
  176. hypfs_update_update(sb);
  177. rc = count;
  178. out:
  179. mutex_unlock(&fs_info->lock);
  180. return rc;
  181. }
  182. static int hypfs_release(struct inode *inode, struct file *filp)
  183. {
  184. kfree(filp->private_data);
  185. return 0;
  186. }
  187. enum { opt_uid, opt_gid, opt_err };
  188. static const match_table_t hypfs_tokens = {
  189. {opt_uid, "uid=%u"},
  190. {opt_gid, "gid=%u"},
  191. {opt_err, NULL}
  192. };
  193. static int hypfs_parse_options(char *options, struct super_block *sb)
  194. {
  195. char *str;
  196. substring_t args[MAX_OPT_ARGS];
  197. kuid_t uid;
  198. kgid_t gid;
  199. if (!options)
  200. return 0;
  201. while ((str = strsep(&options, ",")) != NULL) {
  202. int token, option;
  203. struct hypfs_sb_info *hypfs_info = sb->s_fs_info;
  204. if (!*str)
  205. continue;
  206. token = match_token(str, hypfs_tokens, args);
  207. switch (token) {
  208. case opt_uid:
  209. if (match_int(&args[0], &option))
  210. return -EINVAL;
  211. uid = make_kuid(current_user_ns(), option);
  212. if (!uid_valid(uid))
  213. return -EINVAL;
  214. hypfs_info->uid = uid;
  215. break;
  216. case opt_gid:
  217. if (match_int(&args[0], &option))
  218. return -EINVAL;
  219. gid = make_kgid(current_user_ns(), option);
  220. if (!gid_valid(gid))
  221. return -EINVAL;
  222. hypfs_info->gid = gid;
  223. break;
  224. case opt_err:
  225. default:
  226. pr_err("%s is not a valid mount option\n", str);
  227. return -EINVAL;
  228. }
  229. }
  230. return 0;
  231. }
  232. static int hypfs_show_options(struct seq_file *s, struct dentry *root)
  233. {
  234. struct hypfs_sb_info *hypfs_info = root->d_sb->s_fs_info;
  235. seq_printf(s, ",uid=%u", from_kuid_munged(&init_user_ns, hypfs_info->uid));
  236. seq_printf(s, ",gid=%u", from_kgid_munged(&init_user_ns, hypfs_info->gid));
  237. return 0;
  238. }
  239. static int hypfs_fill_super(struct super_block *sb, void *data, int silent)
  240. {
  241. struct inode *root_inode;
  242. struct dentry *root_dentry;
  243. int rc = 0;
  244. struct hypfs_sb_info *sbi;
  245. sbi = kzalloc(sizeof(struct hypfs_sb_info), GFP_KERNEL);
  246. if (!sbi)
  247. return -ENOMEM;
  248. mutex_init(&sbi->lock);
  249. sbi->uid = current_uid();
  250. sbi->gid = current_gid();
  251. sb->s_fs_info = sbi;
  252. sb->s_blocksize = PAGE_CACHE_SIZE;
  253. sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
  254. sb->s_magic = HYPFS_MAGIC;
  255. sb->s_op = &hypfs_s_ops;
  256. if (hypfs_parse_options(data, sb))
  257. return -EINVAL;
  258. root_inode = hypfs_make_inode(sb, S_IFDIR | 0755);
  259. if (!root_inode)
  260. return -ENOMEM;
  261. root_inode->i_op = &simple_dir_inode_operations;
  262. root_inode->i_fop = &simple_dir_operations;
  263. sb->s_root = root_dentry = d_make_root(root_inode);
  264. if (!root_dentry)
  265. return -ENOMEM;
  266. if (MACHINE_IS_VM)
  267. rc = hypfs_vm_create_files(root_dentry);
  268. else
  269. rc = hypfs_diag_create_files(root_dentry);
  270. if (rc)
  271. return rc;
  272. sbi->update_file = hypfs_create_update_file(root_dentry);
  273. if (IS_ERR(sbi->update_file))
  274. return PTR_ERR(sbi->update_file);
  275. hypfs_update_update(sb);
  276. pr_info("Hypervisor filesystem mounted\n");
  277. return 0;
  278. }
  279. static struct dentry *hypfs_mount(struct file_system_type *fst, int flags,
  280. const char *devname, void *data)
  281. {
  282. return mount_single(fst, flags, data, hypfs_fill_super);
  283. }
  284. static void hypfs_kill_super(struct super_block *sb)
  285. {
  286. struct hypfs_sb_info *sb_info = sb->s_fs_info;
  287. if (sb->s_root)
  288. hypfs_delete_tree(sb->s_root);
  289. if (sb_info->update_file)
  290. hypfs_remove(sb_info->update_file);
  291. kfree(sb->s_fs_info);
  292. sb->s_fs_info = NULL;
  293. kill_litter_super(sb);
  294. }
  295. static struct dentry *hypfs_create_file(struct dentry *parent, const char *name,
  296. char *data, umode_t mode)
  297. {
  298. struct dentry *dentry;
  299. struct inode *inode;
  300. mutex_lock(&parent->d_inode->i_mutex);
  301. dentry = lookup_one_len(name, parent, strlen(name));
  302. if (IS_ERR(dentry)) {
  303. dentry = ERR_PTR(-ENOMEM);
  304. goto fail;
  305. }
  306. inode = hypfs_make_inode(parent->d_sb, mode);
  307. if (!inode) {
  308. dput(dentry);
  309. dentry = ERR_PTR(-ENOMEM);
  310. goto fail;
  311. }
  312. if (S_ISREG(mode)) {
  313. inode->i_fop = &hypfs_file_ops;
  314. if (data)
  315. inode->i_size = strlen(data);
  316. else
  317. inode->i_size = 0;
  318. } else if (S_ISDIR(mode)) {
  319. inode->i_op = &simple_dir_inode_operations;
  320. inode->i_fop = &simple_dir_operations;
  321. inc_nlink(parent->d_inode);
  322. } else
  323. BUG();
  324. inode->i_private = data;
  325. d_instantiate(dentry, inode);
  326. dget(dentry);
  327. fail:
  328. mutex_unlock(&parent->d_inode->i_mutex);
  329. return dentry;
  330. }
  331. struct dentry *hypfs_mkdir(struct dentry *parent, const char *name)
  332. {
  333. struct dentry *dentry;
  334. dentry = hypfs_create_file(parent, name, NULL, S_IFDIR | DIR_MODE);
  335. if (IS_ERR(dentry))
  336. return dentry;
  337. hypfs_add_dentry(dentry);
  338. return dentry;
  339. }
  340. static struct dentry *hypfs_create_update_file(struct dentry *dir)
  341. {
  342. struct dentry *dentry;
  343. dentry = hypfs_create_file(dir, "update", NULL,
  344. S_IFREG | UPDATE_FILE_MODE);
  345. /*
  346. * We do not put the update file on the 'delete' list with
  347. * hypfs_add_dentry(), since it should not be removed when the tree
  348. * is updated.
  349. */
  350. return dentry;
  351. }
  352. struct dentry *hypfs_create_u64(struct dentry *dir,
  353. const char *name, __u64 value)
  354. {
  355. char *buffer;
  356. char tmp[TMP_SIZE];
  357. struct dentry *dentry;
  358. snprintf(tmp, TMP_SIZE, "%llu\n", (unsigned long long int)value);
  359. buffer = kstrdup(tmp, GFP_KERNEL);
  360. if (!buffer)
  361. return ERR_PTR(-ENOMEM);
  362. dentry =
  363. hypfs_create_file(dir, name, buffer, S_IFREG | REG_FILE_MODE);
  364. if (IS_ERR(dentry)) {
  365. kfree(buffer);
  366. return ERR_PTR(-ENOMEM);
  367. }
  368. hypfs_add_dentry(dentry);
  369. return dentry;
  370. }
  371. struct dentry *hypfs_create_str(struct dentry *dir,
  372. const char *name, char *string)
  373. {
  374. char *buffer;
  375. struct dentry *dentry;
  376. buffer = kmalloc(strlen(string) + 2, GFP_KERNEL);
  377. if (!buffer)
  378. return ERR_PTR(-ENOMEM);
  379. sprintf(buffer, "%s\n", string);
  380. dentry =
  381. hypfs_create_file(dir, name, buffer, S_IFREG | REG_FILE_MODE);
  382. if (IS_ERR(dentry)) {
  383. kfree(buffer);
  384. return ERR_PTR(-ENOMEM);
  385. }
  386. hypfs_add_dentry(dentry);
  387. return dentry;
  388. }
  389. static const struct file_operations hypfs_file_ops = {
  390. .open = hypfs_open,
  391. .release = hypfs_release,
  392. .read = do_sync_read,
  393. .write = do_sync_write,
  394. .aio_read = hypfs_aio_read,
  395. .aio_write = hypfs_aio_write,
  396. .llseek = no_llseek,
  397. };
  398. static struct file_system_type hypfs_type = {
  399. .owner = THIS_MODULE,
  400. .name = "s390_hypfs",
  401. .mount = hypfs_mount,
  402. .kill_sb = hypfs_kill_super
  403. };
  404. MODULE_ALIAS_FS("s390_hypfs");
  405. static const struct super_operations hypfs_s_ops = {
  406. .statfs = simple_statfs,
  407. .evict_inode = hypfs_evict_inode,
  408. .show_options = hypfs_show_options,
  409. };
  410. static struct kobject *s390_kobj;
  411. static int __init hypfs_init(void)
  412. {
  413. int rc;
  414. rc = hypfs_dbfs_init();
  415. if (rc)
  416. return rc;
  417. if (hypfs_diag_init()) {
  418. rc = -ENODATA;
  419. goto fail_dbfs_exit;
  420. }
  421. if (hypfs_vm_init()) {
  422. rc = -ENODATA;
  423. goto fail_hypfs_diag_exit;
  424. }
  425. s390_kobj = kobject_create_and_add("s390", hypervisor_kobj);
  426. if (!s390_kobj) {
  427. rc = -ENOMEM;
  428. goto fail_hypfs_vm_exit;
  429. }
  430. rc = register_filesystem(&hypfs_type);
  431. if (rc)
  432. goto fail_filesystem;
  433. return 0;
  434. fail_filesystem:
  435. kobject_put(s390_kobj);
  436. fail_hypfs_vm_exit:
  437. hypfs_vm_exit();
  438. fail_hypfs_diag_exit:
  439. hypfs_diag_exit();
  440. fail_dbfs_exit:
  441. hypfs_dbfs_exit();
  442. pr_err("Initialization of hypfs failed with rc=%i\n", rc);
  443. return rc;
  444. }
  445. static void __exit hypfs_exit(void)
  446. {
  447. hypfs_diag_exit();
  448. hypfs_vm_exit();
  449. hypfs_dbfs_exit();
  450. unregister_filesystem(&hypfs_type);
  451. kobject_put(s390_kobj);
  452. }
  453. module_init(hypfs_init)
  454. module_exit(hypfs_exit)
  455. MODULE_LICENSE("GPL");
  456. MODULE_AUTHOR("Michael Holzheu <holzheu@de.ibm.com>");
  457. MODULE_DESCRIPTION("s390 Hypervisor Filesystem");