file_table.c 9.4 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363
  1. /*
  2. * linux/fs/file_table.c
  3. *
  4. * Copyright (C) 1991, 1992 Linus Torvalds
  5. * Copyright (C) 1997 David S. Miller (davem@caip.rutgers.edu)
  6. */
  7. #include <linux/string.h>
  8. #include <linux/slab.h>
  9. #include <linux/file.h>
  10. #include <linux/fdtable.h>
  11. #include <linux/init.h>
  12. #include <linux/module.h>
  13. #include <linux/fs.h>
  14. #include <linux/security.h>
  15. #include <linux/eventpoll.h>
  16. #include <linux/rcupdate.h>
  17. #include <linux/mount.h>
  18. #include <linux/capability.h>
  19. #include <linux/cdev.h>
  20. #include <linux/fsnotify.h>
  21. #include <linux/sysctl.h>
  22. #include <linux/lglock.h>
  23. #include <linux/percpu_counter.h>
  24. #include <linux/percpu.h>
  25. #include <linux/hardirq.h>
  26. #include <linux/task_work.h>
  27. #include <linux/ima.h>
  28. #include <linux/atomic.h>
  29. #include "internal.h"
  30. /* sysctl tunables... */
  31. struct files_stat_struct files_stat = {
  32. .max_files = NR_FILE
  33. };
  34. /* SLAB cache for file structures */
  35. static struct kmem_cache *filp_cachep __read_mostly;
  36. static struct percpu_counter nr_files __cacheline_aligned_in_smp;
  37. static void file_free_rcu(struct rcu_head *head)
  38. {
  39. struct file *f = container_of(head, struct file, f_u.fu_rcuhead);
  40. put_cred(f->f_cred);
  41. kmem_cache_free(filp_cachep, f);
  42. }
  43. static inline void file_free(struct file *f)
  44. {
  45. percpu_counter_dec(&nr_files);
  46. file_check_state(f);
  47. call_rcu(&f->f_u.fu_rcuhead, file_free_rcu);
  48. }
  49. /*
  50. * Return the total number of open files in the system
  51. */
  52. static long get_nr_files(void)
  53. {
  54. return percpu_counter_read_positive(&nr_files);
  55. }
  56. /*
  57. * Return the maximum number of open files in the system
  58. */
  59. unsigned long get_max_files(void)
  60. {
  61. return files_stat.max_files;
  62. }
  63. EXPORT_SYMBOL_GPL(get_max_files);
  64. /*
  65. * Handle nr_files sysctl
  66. */
  67. #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
  68. int proc_nr_files(ctl_table *table, int write,
  69. void __user *buffer, size_t *lenp, loff_t *ppos)
  70. {
  71. files_stat.nr_files = get_nr_files();
  72. return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
  73. }
  74. #else
  75. int proc_nr_files(ctl_table *table, int write,
  76. void __user *buffer, size_t *lenp, loff_t *ppos)
  77. {
  78. return -ENOSYS;
  79. }
  80. #endif
  81. /* Find an unused file structure and return a pointer to it.
  82. * Returns an error pointer if some error happend e.g. we over file
  83. * structures limit, run out of memory or operation is not permitted.
  84. *
  85. * Be very careful using this. You are responsible for
  86. * getting write access to any mount that you might assign
  87. * to this filp, if it is opened for write. If this is not
  88. * done, you will imbalance int the mount's writer count
  89. * and a warning at __fput() time.
  90. */
  91. struct file *get_empty_filp(void)
  92. {
  93. const struct cred *cred = current_cred();
  94. static long old_max;
  95. struct file *f;
  96. int error;
  97. /*
  98. * Privileged users can go above max_files
  99. */
  100. if (get_nr_files() >= files_stat.max_files && !capable(CAP_SYS_ADMIN)) {
  101. /*
  102. * percpu_counters are inaccurate. Do an expensive check before
  103. * we go and fail.
  104. */
  105. if (percpu_counter_sum_positive(&nr_files) >= files_stat.max_files)
  106. goto over;
  107. }
  108. f = kmem_cache_zalloc(filp_cachep, GFP_KERNEL);
  109. if (unlikely(!f))
  110. return ERR_PTR(-ENOMEM);
  111. percpu_counter_inc(&nr_files);
  112. f->f_cred = get_cred(cred);
  113. error = security_file_alloc(f);
  114. if (unlikely(error)) {
  115. file_free(f);
  116. return ERR_PTR(error);
  117. }
  118. atomic_long_set(&f->f_count, 1);
  119. rwlock_init(&f->f_owner.lock);
  120. spin_lock_init(&f->f_lock);
  121. eventpoll_init_file(f);
  122. /* f->f_version: 0 */
  123. return f;
  124. over:
  125. /* Ran out of filps - report that */
  126. if (get_nr_files() > old_max) {
  127. pr_info("VFS: file-max limit %lu reached\n", get_max_files());
  128. old_max = get_nr_files();
  129. }
  130. return ERR_PTR(-ENFILE);
  131. }
  132. /**
  133. * alloc_file - allocate and initialize a 'struct file'
  134. * @mnt: the vfsmount on which the file will reside
  135. * @dentry: the dentry representing the new file
  136. * @mode: the mode with which the new file will be opened
  137. * @fop: the 'struct file_operations' for the new file
  138. *
  139. * Use this instead of get_empty_filp() to get a new
  140. * 'struct file'. Do so because of the same initialization
  141. * pitfalls reasons listed for init_file(). This is a
  142. * preferred interface to using init_file().
  143. *
  144. * If all the callers of init_file() are eliminated, its
  145. * code should be moved into this function.
  146. */
  147. struct file *alloc_file(struct path *path, fmode_t mode,
  148. const struct file_operations *fop)
  149. {
  150. struct file *file;
  151. file = get_empty_filp();
  152. if (IS_ERR(file))
  153. return file;
  154. file->f_path = *path;
  155. file->f_inode = path->dentry->d_inode;
  156. file->f_mapping = path->dentry->d_inode->i_mapping;
  157. file->f_mode = mode;
  158. file->f_op = fop;
  159. /*
  160. * These mounts don't really matter in practice
  161. * for r/o bind mounts. They aren't userspace-
  162. * visible. We do this for consistency, and so
  163. * that we can do debugging checks at __fput()
  164. */
  165. if ((mode & FMODE_WRITE) && !special_file(path->dentry->d_inode->i_mode)) {
  166. file_take_write(file);
  167. WARN_ON(mnt_clone_write(path->mnt));
  168. }
  169. if ((mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
  170. i_readcount_inc(path->dentry->d_inode);
  171. return file;
  172. }
  173. EXPORT_SYMBOL(alloc_file);
  174. /**
  175. * drop_file_write_access - give up ability to write to a file
  176. * @file: the file to which we will stop writing
  177. *
  178. * This is a central place which will give up the ability
  179. * to write to @file, along with access to write through
  180. * its vfsmount.
  181. */
  182. static void drop_file_write_access(struct file *file)
  183. {
  184. struct vfsmount *mnt = file->f_path.mnt;
  185. struct dentry *dentry = file->f_path.dentry;
  186. struct inode *inode = dentry->d_inode;
  187. put_write_access(inode);
  188. if (special_file(inode->i_mode))
  189. return;
  190. if (file_check_writeable(file) != 0)
  191. return;
  192. __mnt_drop_write(mnt);
  193. file_release_write(file);
  194. }
  195. /* the real guts of fput() - releasing the last reference to file
  196. */
  197. static void __fput(struct file *file)
  198. {
  199. struct dentry *dentry = file->f_path.dentry;
  200. struct vfsmount *mnt = file->f_path.mnt;
  201. struct inode *inode = file->f_inode;
  202. might_sleep();
  203. fsnotify_close(file);
  204. /*
  205. * The function eventpoll_release() should be the first called
  206. * in the file cleanup chain.
  207. */
  208. eventpoll_release(file);
  209. locks_remove_flock(file);
  210. if (unlikely(file->f_flags & FASYNC)) {
  211. if (file->f_op->fasync)
  212. file->f_op->fasync(-1, file, 0);
  213. }
  214. ima_file_free(file);
  215. if (file->f_op->release)
  216. file->f_op->release(inode, file);
  217. security_file_free(file);
  218. if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL &&
  219. !(file->f_mode & FMODE_PATH))) {
  220. cdev_put(inode->i_cdev);
  221. }
  222. fops_put(file->f_op);
  223. put_pid(file->f_owner.pid);
  224. if ((file->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
  225. i_readcount_dec(inode);
  226. if (file->f_mode & FMODE_WRITE)
  227. drop_file_write_access(file);
  228. file->f_path.dentry = NULL;
  229. file->f_path.mnt = NULL;
  230. file->f_inode = NULL;
  231. file_free(file);
  232. dput(dentry);
  233. mntput(mnt);
  234. }
  235. static LLIST_HEAD(delayed_fput_list);
  236. static void delayed_fput(struct work_struct *unused)
  237. {
  238. struct llist_node *node = llist_del_all(&delayed_fput_list);
  239. struct llist_node *next;
  240. for (; node; node = next) {
  241. next = llist_next(node);
  242. __fput(llist_entry(node, struct file, f_u.fu_llist));
  243. }
  244. }
  245. static void ____fput(struct callback_head *work)
  246. {
  247. __fput(container_of(work, struct file, f_u.fu_rcuhead));
  248. }
  249. /*
  250. * If kernel thread really needs to have the final fput() it has done
  251. * to complete, call this. The only user right now is the boot - we
  252. * *do* need to make sure our writes to binaries on initramfs has
  253. * not left us with opened struct file waiting for __fput() - execve()
  254. * won't work without that. Please, don't add more callers without
  255. * very good reasons; in particular, never call that with locks
  256. * held and never call that from a thread that might need to do
  257. * some work on any kind of umount.
  258. */
  259. void flush_delayed_fput(void)
  260. {
  261. delayed_fput(NULL);
  262. }
  263. static DECLARE_DELAYED_WORK(delayed_fput_work, delayed_fput);
  264. void fput(struct file *file)
  265. {
  266. if (atomic_long_dec_and_test(&file->f_count)) {
  267. struct task_struct *task = current;
  268. if (likely(!in_interrupt() && !(task->flags & PF_KTHREAD))) {
  269. init_task_work(&file->f_u.fu_rcuhead, ____fput);
  270. if (!task_work_add(task, &file->f_u.fu_rcuhead, true))
  271. return;
  272. /*
  273. * After this task has run exit_task_work(),
  274. * task_work_add() will fail. Fall through to delayed
  275. * fput to avoid leaking *file.
  276. */
  277. }
  278. if (llist_add(&file->f_u.fu_llist, &delayed_fput_list))
  279. schedule_delayed_work(&delayed_fput_work, 1);
  280. }
  281. }
  282. /*
  283. * synchronous analog of fput(); for kernel threads that might be needed
  284. * in some umount() (and thus can't use flush_delayed_fput() without
  285. * risking deadlocks), need to wait for completion of __fput() and know
  286. * for this specific struct file it won't involve anything that would
  287. * need them. Use only if you really need it - at the very least,
  288. * don't blindly convert fput() by kernel thread to that.
  289. */
  290. void __fput_sync(struct file *file)
  291. {
  292. if (atomic_long_dec_and_test(&file->f_count)) {
  293. struct task_struct *task = current;
  294. BUG_ON(!(task->flags & PF_KTHREAD));
  295. __fput(file);
  296. }
  297. }
  298. EXPORT_SYMBOL(fput);
  299. void put_filp(struct file *file)
  300. {
  301. if (atomic_long_dec_and_test(&file->f_count)) {
  302. security_file_free(file);
  303. file_free(file);
  304. }
  305. }
  306. void __init files_init(unsigned long mempages)
  307. {
  308. unsigned long n;
  309. filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
  310. SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
  311. /*
  312. * One file with associated inode and dcache is very roughly 1K.
  313. * Per default don't use more than 10% of our memory for files.
  314. */
  315. n = (mempages * (PAGE_SIZE / 1024)) / 10;
  316. files_stat.max_files = max_t(unsigned long, n, NR_FILE);
  317. files_defer_init();
  318. percpu_counter_init(&nr_files, 0);
  319. }