blob: 6d982b57de9241de853b9ee592ff0564a0bad832 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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
8#include <linux/string.h>
9#include <linux/slab.h>
10#include <linux/file.h>
Al Viro9f3acc32008-04-24 07:44:08 -040011#include <linux/fdtable.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070012#include <linux/init.h>
13#include <linux/module.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070014#include <linux/fs.h>
15#include <linux/security.h>
16#include <linux/eventpoll.h>
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070017#include <linux/rcupdate.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include <linux/mount.h>
Randy Dunlap16f7e0f2006-01-11 12:17:46 -080019#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include <linux/cdev.h>
Robert Love0eeca282005-07-12 17:06:03 -040021#include <linux/fsnotify.h>
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080022#include <linux/sysctl.h>
23#include <linux/percpu_counter.h>
Nick Piggin6416ccb2010-08-18 04:37:38 +100024#include <linux/percpu.h>
Al Viro4a9d4b02012-06-24 09:56:45 +040025#include <linux/hardirq.h>
26#include <linux/task_work.h>
Al Viro0552f872009-12-16 04:53:03 -050027#include <linux/ima.h>
Mel Gorman4248b0d2015-08-06 15:46:20 -070028#include <linux/swap.h>
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080029
Arun Sharma600634972011-07-26 16:09:06 -070030#include <linux/atomic.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070031
Eric Parise81e3f42009-12-04 15:47:36 -050032#include "internal.h"
33
Linus Torvalds1da177e2005-04-16 15:20:36 -070034/* sysctl tunables... */
35struct files_stat_struct files_stat = {
36 .max_files = NR_FILE
37};
38
Eric Dumazetb6b3fde2008-12-10 09:35:45 -080039/* SLAB cache for file structures */
40static struct kmem_cache *filp_cachep __read_mostly;
41
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080042static struct percpu_counter nr_files __cacheline_aligned_in_smp;
Linus Torvalds1da177e2005-04-16 15:20:36 -070043
Al Viro5c33b182012-07-20 23:05:59 +040044static void file_free_rcu(struct rcu_head *head)
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070045{
David Howellsd76b0d92008-11-14 10:39:25 +110046 struct file *f = container_of(head, struct file, f_u.fu_rcuhead);
47
48 put_cred(f->f_cred);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070049 kmem_cache_free(filp_cachep, f);
50}
51
Linus Torvalds1da177e2005-04-16 15:20:36 -070052static inline void file_free(struct file *f)
53{
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080054 percpu_counter_dec(&nr_files);
Eric Dumazet2f512012005-10-30 15:02:16 -080055 call_rcu(&f->f_u.fu_rcuhead, file_free_rcu);
Linus Torvalds1da177e2005-04-16 15:20:36 -070056}
57
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080058/*
59 * Return the total number of open files in the system
60 */
Eric Dumazet518de9b2010-10-26 14:22:44 -070061static long get_nr_files(void)
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080062{
63 return percpu_counter_read_positive(&nr_files);
64}
65
66/*
67 * Return the maximum number of open files in the system
68 */
Eric Dumazet518de9b2010-10-26 14:22:44 -070069unsigned long get_max_files(void)
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080070{
71 return files_stat.max_files;
72}
73EXPORT_SYMBOL_GPL(get_max_files);
74
75/*
76 * Handle nr_files sysctl
77 */
78#if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
Joe Perches1f7e0612014-06-06 14:38:05 -070079int proc_nr_files(struct ctl_table *table, int write,
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080080 void __user *buffer, size_t *lenp, loff_t *ppos)
81{
82 files_stat.nr_files = get_nr_files();
Eric Dumazet518de9b2010-10-26 14:22:44 -070083 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080084}
85#else
Joe Perches1f7e0612014-06-06 14:38:05 -070086int proc_nr_files(struct ctl_table *table, int write,
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080087 void __user *buffer, size_t *lenp, loff_t *ppos)
88{
89 return -ENOSYS;
90}
91#endif
92
Linus Torvalds1da177e2005-04-16 15:20:36 -070093/* Find an unused file structure and return a pointer to it.
Al Viro1afc99b2013-02-14 20:41:04 -050094 * Returns an error pointer if some error happend e.g. we over file
95 * structures limit, run out of memory or operation is not permitted.
Dave Hansen430e2852008-02-15 14:37:26 -080096 *
97 * Be very careful using this. You are responsible for
98 * getting write access to any mount that you might assign
99 * to this filp, if it is opened for write. If this is not
100 * done, you will imbalance int the mount's writer count
101 * and a warning at __fput() time.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102 */
103struct file *get_empty_filp(void)
104{
David Howells86a264a2008-11-14 10:39:18 +1100105 const struct cred *cred = current_cred();
Eric Dumazet518de9b2010-10-26 14:22:44 -0700106 static long old_max;
Al Viro1afc99b2013-02-14 20:41:04 -0500107 struct file *f;
108 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109
110 /*
111 * Privileged users can go above max_files
112 */
Dipankar Sarma529bf6b2006-03-07 21:55:35 -0800113 if (get_nr_files() >= files_stat.max_files && !capable(CAP_SYS_ADMIN)) {
114 /*
115 * percpu_counters are inaccurate. Do an expensive check before
116 * we go and fail.
117 */
Peter Zijlstra52d9f3b2007-10-16 23:25:44 -0700118 if (percpu_counter_sum_positive(&nr_files) >= files_stat.max_files)
Dipankar Sarma529bf6b2006-03-07 21:55:35 -0800119 goto over;
120 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121
Denis Cheng4975e452007-10-16 23:26:19 -0700122 f = kmem_cache_zalloc(filp_cachep, GFP_KERNEL);
Al Viro1afc99b2013-02-14 20:41:04 -0500123 if (unlikely(!f))
124 return ERR_PTR(-ENOMEM);
Kirill Korotaevaf4d2ec2005-06-23 00:09:50 -0700125
Dipankar Sarma529bf6b2006-03-07 21:55:35 -0800126 percpu_counter_inc(&nr_files);
Tetsuo Handa78d29782011-02-04 18:13:24 +0000127 f->f_cred = get_cred(cred);
Al Viro1afc99b2013-02-14 20:41:04 -0500128 error = security_file_alloc(f);
129 if (unlikely(error)) {
130 file_free(f);
131 return ERR_PTR(error);
132 }
Kirill Korotaevaf4d2ec2005-06-23 00:09:50 -0700133
Al Viro516e0cc2008-07-26 00:39:17 -0400134 atomic_long_set(&f->f_count, 1);
Benjamin LaHaise5a6b7952006-03-23 03:01:03 -0800135 rwlock_init(&f->f_owner.lock);
Jonathan Corbet68499912009-02-06 13:52:43 -0700136 spin_lock_init(&f->f_lock);
Linus Torvalds9c225f22014-03-03 09:36:58 -0800137 mutex_init(&f->f_pos_lock);
Benjamin LaHaise5a6b7952006-03-23 03:01:03 -0800138 eventpoll_init_file(f);
139 /* f->f_version: 0 */
Kirill Korotaevaf4d2ec2005-06-23 00:09:50 -0700140 return f;
141
142over:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143 /* Ran out of filps - report that */
Dipankar Sarma529bf6b2006-03-07 21:55:35 -0800144 if (get_nr_files() > old_max) {
Eric Dumazet518de9b2010-10-26 14:22:44 -0700145 pr_info("VFS: file-max limit %lu reached\n", get_max_files());
Dipankar Sarma529bf6b2006-03-07 21:55:35 -0800146 old_max = get_nr_files();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147 }
Al Viro1afc99b2013-02-14 20:41:04 -0500148 return ERR_PTR(-ENFILE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149}
150
Dave Hansence8d2cd2007-10-16 23:31:13 -0700151/**
152 * alloc_file - allocate and initialize a 'struct file'
Eric Biggersa4576062014-10-12 14:29:29 -0500153 *
154 * @path: the (dentry, vfsmount) pair for the new file
Dave Hansence8d2cd2007-10-16 23:31:13 -0700155 * @mode: the mode with which the new file will be opened
156 * @fop: the 'struct file_operations' for the new file
Dave Hansence8d2cd2007-10-16 23:31:13 -0700157 */
Al Viroa4141d72016-11-20 20:28:12 -0500158struct file *alloc_file(const struct path *path, fmode_t mode,
Al Viro2c48b9c2009-08-09 00:52:35 +0400159 const struct file_operations *fop)
Dave Hansence8d2cd2007-10-16 23:31:13 -0700160{
161 struct file *file;
Dave Hansence8d2cd2007-10-16 23:31:13 -0700162
163 file = get_empty_filp();
Al Viro1afc99b2013-02-14 20:41:04 -0500164 if (IS_ERR(file))
Anatol Pomozov39b65252012-09-12 20:11:55 -0700165 return file;
Dave Hansence8d2cd2007-10-16 23:31:13 -0700166
Al Viro2c48b9c2009-08-09 00:52:35 +0400167 file->f_path = *path;
Al Virodd379782013-03-01 19:48:30 -0500168 file->f_inode = path->dentry->d_inode;
Al Viro2c48b9c2009-08-09 00:52:35 +0400169 file->f_mapping = path->dentry->d_inode->i_mapping;
Al Viro293bc982014-02-11 18:37:41 -0500170 if ((mode & FMODE_READ) &&
Al Viro84363182015-04-04 01:14:53 -0400171 likely(fop->read || fop->read_iter))
Al Viro7f7f25e2014-02-11 17:49:24 -0500172 mode |= FMODE_CAN_READ;
Al Viro293bc982014-02-11 18:37:41 -0500173 if ((mode & FMODE_WRITE) &&
Al Viro84363182015-04-04 01:14:53 -0400174 likely(fop->write || fop->write_iter))
Al Viro7f7f25e2014-02-11 17:49:24 -0500175 mode |= FMODE_CAN_WRITE;
Dave Hansence8d2cd2007-10-16 23:31:13 -0700176 file->f_mode = mode;
177 file->f_op = fop;
Mimi Zohar890275b52010-11-02 10:13:07 -0400178 if ((mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
179 i_readcount_inc(path->dentry->d_inode);
Al Viro3d1e4632009-08-08 23:56:29 +0400180 return file;
Dave Hansence8d2cd2007-10-16 23:31:13 -0700181}
Roland Dreier73efc462009-12-16 12:43:11 -0800182EXPORT_SYMBOL(alloc_file);
Dave Hansence8d2cd2007-10-16 23:31:13 -0700183
Al Virod7065da2010-05-26 15:13:55 -0400184/* the real guts of fput() - releasing the last reference to file
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185 */
Al Virod7065da2010-05-26 15:13:55 -0400186static void __fput(struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800188 struct dentry *dentry = file->f_path.dentry;
189 struct vfsmount *mnt = file->f_path.mnt;
David Howellsc77cece2013-06-13 23:37:49 +0100190 struct inode *inode = file->f_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191
192 might_sleep();
Robert Love0eeca282005-07-12 17:06:03 -0400193
194 fsnotify_close(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195 /*
196 * The function eventpoll_release() should be the first called
197 * in the file cleanup chain.
198 */
199 eventpoll_release(file);
Jeff Layton78ed8a12014-02-03 12:13:08 -0500200 locks_remove_file(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201
Al Viro233e70f2008-10-31 23:28:30 +0000202 if (unlikely(file->f_flags & FASYNC)) {
Al Viro72c2d532013-09-22 16:27:52 -0400203 if (file->f_op->fasync)
Al Viro233e70f2008-10-31 23:28:30 +0000204 file->f_op->fasync(-1, file, 0);
205 }
Mimi Zohar4199d352011-03-16 22:48:43 -0400206 ima_file_free(file);
Al Viro72c2d532013-09-22 16:27:52 -0400207 if (file->f_op->release)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208 file->f_op->release(inode, file);
209 security_file_free(file);
Miklos Szeredi60ed8cf2011-03-16 18:17:54 +0100210 if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL &&
211 !(file->f_mode & FMODE_PATH))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212 cdev_put(inode->i_cdev);
Miklos Szeredi60ed8cf2011-03-16 18:17:54 +0100213 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214 fops_put(file->f_op);
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700215 put_pid(file->f_owner.pid);
Mimi Zohar890275b52010-11-02 10:13:07 -0400216 if ((file->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
217 i_readcount_dec(inode);
Al Viro83f936c2014-03-14 12:02:47 -0400218 if (file->f_mode & FMODE_WRITER) {
219 put_write_access(inode);
220 __mnt_drop_write(mnt);
221 }
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800222 file->f_path.dentry = NULL;
223 file->f_path.mnt = NULL;
Al Virodd379782013-03-01 19:48:30 -0500224 file->f_inode = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225 file_free(file);
226 dput(dentry);
227 mntput(mnt);
228}
229
Oleg Nesterov4f5e65a2013-07-08 14:24:16 -0700230static LLIST_HEAD(delayed_fput_list);
Al Viro4a9d4b02012-06-24 09:56:45 +0400231static void delayed_fput(struct work_struct *unused)
232{
Oleg Nesterov4f5e65a2013-07-08 14:24:16 -0700233 struct llist_node *node = llist_del_all(&delayed_fput_list);
234 struct llist_node *next;
235
236 for (; node; node = next) {
237 next = llist_next(node);
238 __fput(llist_entry(node, struct file, f_u.fu_llist));
Al Viro4a9d4b02012-06-24 09:56:45 +0400239 }
240}
241
242static void ____fput(struct callback_head *work)
243{
244 __fput(container_of(work, struct file, f_u.fu_rcuhead));
245}
246
247/*
248 * If kernel thread really needs to have the final fput() it has done
249 * to complete, call this. The only user right now is the boot - we
250 * *do* need to make sure our writes to binaries on initramfs has
251 * not left us with opened struct file waiting for __fput() - execve()
252 * won't work without that. Please, don't add more callers without
253 * very good reasons; in particular, never call that with locks
254 * held and never call that from a thread that might need to do
255 * some work on any kind of umount.
256 */
257void flush_delayed_fput(void)
258{
259 delayed_fput(NULL);
260}
261
Al Viroc7314d72013-10-20 08:44:39 -0400262static DECLARE_DELAYED_WORK(delayed_fput_work, delayed_fput);
Al Viro4a9d4b02012-06-24 09:56:45 +0400263
Al Virod7065da2010-05-26 15:13:55 -0400264void fput(struct file *file)
265{
Al Viro4a9d4b02012-06-24 09:56:45 +0400266 if (atomic_long_dec_and_test(&file->f_count)) {
267 struct task_struct *task = current;
Oleg Nesterove7b2c402013-06-14 21:09:47 +0200268
Oleg Nesterove7b2c402013-06-14 21:09:47 +0200269 if (likely(!in_interrupt() && !(task->flags & PF_KTHREAD))) {
270 init_task_work(&file->f_u.fu_rcuhead, ____fput);
271 if (!task_work_add(task, &file->f_u.fu_rcuhead, true))
272 return;
Andrew Morton64372502013-07-08 14:24:15 -0700273 /*
274 * After this task has run exit_task_work(),
Andrew Mortonbe49b302013-09-11 14:24:34 -0700275 * task_work_add() will fail. Fall through to delayed
Andrew Morton64372502013-07-08 14:24:15 -0700276 * fput to avoid leaking *file.
277 */
Al Viro4a9d4b02012-06-24 09:56:45 +0400278 }
Oleg Nesterov4f5e65a2013-07-08 14:24:16 -0700279
280 if (llist_add(&file->f_u.fu_llist, &delayed_fput_list))
Al Viroc7314d72013-10-20 08:44:39 -0400281 schedule_delayed_work(&delayed_fput_work, 1);
Al Viro4a9d4b02012-06-24 09:56:45 +0400282 }
283}
284
285/*
286 * synchronous analog of fput(); for kernel threads that might be needed
287 * in some umount() (and thus can't use flush_delayed_fput() without
288 * risking deadlocks), need to wait for completion of __fput() and know
289 * for this specific struct file it won't involve anything that would
290 * need them. Use only if you really need it - at the very least,
291 * don't blindly convert fput() by kernel thread to that.
292 */
293void __fput_sync(struct file *file)
294{
295 if (atomic_long_dec_and_test(&file->f_count)) {
296 struct task_struct *task = current;
Al Viro4a9d4b02012-06-24 09:56:45 +0400297 BUG_ON(!(task->flags & PF_KTHREAD));
Al Virod7065da2010-05-26 15:13:55 -0400298 __fput(file);
Al Viro4a9d4b02012-06-24 09:56:45 +0400299 }
Al Virod7065da2010-05-26 15:13:55 -0400300}
301
302EXPORT_SYMBOL(fput);
303
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304void put_filp(struct file *file)
305{
Al Viro516e0cc2008-07-26 00:39:17 -0400306 if (atomic_long_dec_and_test(&file->f_count)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307 security_file_free(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308 file_free(file);
309 }
310}
311
Mel Gorman4248b0d2015-08-06 15:46:20 -0700312void __init files_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313{
Eric Dumazetb6b3fde2008-12-10 09:35:45 -0800314 filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
315 SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
Tejun Heo908c7f12014-09-08 09:51:29 +0900316 percpu_counter_init(&nr_files, 0, GFP_KERNEL);
Mel Gorman4248b0d2015-08-06 15:46:20 -0700317}
318
319/*
320 * One file with associated inode and dcache is very roughly 1K. Per default
321 * do not use more than 10% of our memory for files.
322 */
323void __init files_maxfiles_init(void)
324{
325 unsigned long n;
326 unsigned long memreserve = (totalram_pages - nr_free_pages()) * 3/2;
327
328 memreserve = min(memreserve, totalram_pages - 1);
329 n = ((totalram_pages - memreserve) * (PAGE_SIZE / 1024)) / 10;
330
331 files_stat.max_files = max_t(unsigned long, n, NR_FILE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332}