blob: 5fff9030be34df26aff0c0ed56081aee1306bf02 [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>
Nick Piggin6416ccb2010-08-18 04:37:38 +100023#include <linux/lglock.h>
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080024#include <linux/percpu_counter.h>
Nick Piggin6416ccb2010-08-18 04:37:38 +100025#include <linux/percpu.h>
Al Viro4a9d4b02012-06-24 09:56:45 +040026#include <linux/hardirq.h>
27#include <linux/task_work.h>
Al Viro0552f872009-12-16 04:53:03 -050028#include <linux/ima.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);
Dave Hansenad775f52008-02-15 14:38:01 -080055 file_check_state(f);
Eric Dumazet2f512012005-10-30 15:02:16 -080056 call_rcu(&f->f_u.fu_rcuhead, file_free_rcu);
Linus Torvalds1da177e2005-04-16 15:20:36 -070057}
58
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080059/*
60 * Return the total number of open files in the system
61 */
Eric Dumazet518de9b2010-10-26 14:22:44 -070062static long get_nr_files(void)
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080063{
64 return percpu_counter_read_positive(&nr_files);
65}
66
67/*
68 * Return the maximum number of open files in the system
69 */
Eric Dumazet518de9b2010-10-26 14:22:44 -070070unsigned long get_max_files(void)
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080071{
72 return files_stat.max_files;
73}
74EXPORT_SYMBOL_GPL(get_max_files);
75
76/*
77 * Handle nr_files sysctl
78 */
79#if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
Alexey Dobriyan8d65af72009-09-23 15:57:19 -070080int proc_nr_files(ctl_table *table, int write,
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080081 void __user *buffer, size_t *lenp, loff_t *ppos)
82{
83 files_stat.nr_files = get_nr_files();
Eric Dumazet518de9b2010-10-26 14:22:44 -070084 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080085}
86#else
Alexey Dobriyan8d65af72009-09-23 15:57:19 -070087int proc_nr_files(ctl_table *table, int write,
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080088 void __user *buffer, size_t *lenp, loff_t *ppos)
89{
90 return -ENOSYS;
91}
92#endif
93
Linus Torvalds1da177e2005-04-16 15:20:36 -070094/* Find an unused file structure and return a pointer to it.
Al Viro1afc99b2013-02-14 20:41:04 -050095 * Returns an error pointer if some error happend e.g. we over file
96 * structures limit, run out of memory or operation is not permitted.
Dave Hansen430e2852008-02-15 14:37:26 -080097 *
98 * Be very careful using this. You are responsible for
99 * getting write access to any mount that you might assign
100 * to this filp, if it is opened for write. If this is not
101 * done, you will imbalance int the mount's writer count
102 * and a warning at __fput() time.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103 */
104struct file *get_empty_filp(void)
105{
David Howells86a264a2008-11-14 10:39:18 +1100106 const struct cred *cred = current_cred();
Eric Dumazet518de9b2010-10-26 14:22:44 -0700107 static long old_max;
Al Viro1afc99b2013-02-14 20:41:04 -0500108 struct file *f;
109 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110
111 /*
112 * Privileged users can go above max_files
113 */
Dipankar Sarma529bf6b2006-03-07 21:55:35 -0800114 if (get_nr_files() >= files_stat.max_files && !capable(CAP_SYS_ADMIN)) {
115 /*
116 * percpu_counters are inaccurate. Do an expensive check before
117 * we go and fail.
118 */
Peter Zijlstra52d9f3b2007-10-16 23:25:44 -0700119 if (percpu_counter_sum_positive(&nr_files) >= files_stat.max_files)
Dipankar Sarma529bf6b2006-03-07 21:55:35 -0800120 goto over;
121 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122
Denis Cheng4975e452007-10-16 23:26:19 -0700123 f = kmem_cache_zalloc(filp_cachep, GFP_KERNEL);
Al Viro1afc99b2013-02-14 20:41:04 -0500124 if (unlikely(!f))
125 return ERR_PTR(-ENOMEM);
Kirill Korotaevaf4d2ec2005-06-23 00:09:50 -0700126
Dipankar Sarma529bf6b2006-03-07 21:55:35 -0800127 percpu_counter_inc(&nr_files);
Tetsuo Handa78d29782011-02-04 18:13:24 +0000128 f->f_cred = get_cred(cred);
Al Viro1afc99b2013-02-14 20:41:04 -0500129 error = security_file_alloc(f);
130 if (unlikely(error)) {
131 file_free(f);
132 return ERR_PTR(error);
133 }
Kirill Korotaevaf4d2ec2005-06-23 00:09:50 -0700134
Al Viro516e0cc2008-07-26 00:39:17 -0400135 atomic_long_set(&f->f_count, 1);
Benjamin LaHaise5a6b7952006-03-23 03:01:03 -0800136 rwlock_init(&f->f_owner.lock);
Jonathan Corbet68499912009-02-06 13:52:43 -0700137 spin_lock_init(&f->f_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'
153 * @mnt: the vfsmount on which the file will reside
154 * @dentry: the dentry representing the new file
155 * @mode: the mode with which the new file will be opened
156 * @fop: the 'struct file_operations' for the new file
157 *
158 * Use this instead of get_empty_filp() to get a new
159 * 'struct file'. Do so because of the same initialization
160 * pitfalls reasons listed for init_file(). This is a
161 * preferred interface to using init_file().
162 *
163 * If all the callers of init_file() are eliminated, its
164 * code should be moved into this function.
165 */
Al Viro2c48b9c2009-08-09 00:52:35 +0400166struct file *alloc_file(struct path *path, fmode_t mode,
167 const struct file_operations *fop)
Dave Hansence8d2cd2007-10-16 23:31:13 -0700168{
169 struct file *file;
Dave Hansence8d2cd2007-10-16 23:31:13 -0700170
171 file = get_empty_filp();
Al Viro1afc99b2013-02-14 20:41:04 -0500172 if (IS_ERR(file))
Anatol Pomozov39b65252012-09-12 20:11:55 -0700173 return file;
Dave Hansence8d2cd2007-10-16 23:31:13 -0700174
Al Viro2c48b9c2009-08-09 00:52:35 +0400175 file->f_path = *path;
Al Virodd379782013-03-01 19:48:30 -0500176 file->f_inode = path->dentry->d_inode;
Al Viro2c48b9c2009-08-09 00:52:35 +0400177 file->f_mapping = path->dentry->d_inode->i_mapping;
Dave Hansence8d2cd2007-10-16 23:31:13 -0700178 file->f_mode = mode;
179 file->f_op = fop;
Dave Hansen4a3fd212008-02-15 14:37:48 -0800180
181 /*
182 * These mounts don't really matter in practice
183 * for r/o bind mounts. They aren't userspace-
184 * visible. We do this for consistency, and so
185 * that we can do debugging checks at __fput()
186 */
Al Viro2c48b9c2009-08-09 00:52:35 +0400187 if ((mode & FMODE_WRITE) && !special_file(path->dentry->d_inode->i_mode)) {
Dave Hansenad775f52008-02-15 14:38:01 -0800188 file_take_write(file);
Roland Dreier385e3ed2009-12-16 12:48:44 -0800189 WARN_ON(mnt_clone_write(path->mnt));
Dave Hansen4a3fd212008-02-15 14:37:48 -0800190 }
Mimi Zohar890275b52010-11-02 10:13:07 -0400191 if ((mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
192 i_readcount_inc(path->dentry->d_inode);
Al Viro3d1e4632009-08-08 23:56:29 +0400193 return file;
Dave Hansence8d2cd2007-10-16 23:31:13 -0700194}
Roland Dreier73efc462009-12-16 12:43:11 -0800195EXPORT_SYMBOL(alloc_file);
Dave Hansence8d2cd2007-10-16 23:31:13 -0700196
Dave Hansenaceaf782008-02-15 14:37:31 -0800197/**
198 * drop_file_write_access - give up ability to write to a file
199 * @file: the file to which we will stop writing
200 *
201 * This is a central place which will give up the ability
202 * to write to @file, along with access to write through
203 * its vfsmount.
204 */
Al Virob57ce962012-02-12 02:38:16 -0500205static void drop_file_write_access(struct file *file)
Dave Hansenaceaf782008-02-15 14:37:31 -0800206{
Dave Hansen4a3fd212008-02-15 14:37:48 -0800207 struct vfsmount *mnt = file->f_path.mnt;
Dave Hansenaceaf782008-02-15 14:37:31 -0800208 struct dentry *dentry = file->f_path.dentry;
209 struct inode *inode = dentry->d_inode;
210
211 put_write_access(inode);
Dave Hansenad775f52008-02-15 14:38:01 -0800212
213 if (special_file(inode->i_mode))
214 return;
215 if (file_check_writeable(file) != 0)
216 return;
Jan Karaeb04c282012-06-12 16:20:35 +0200217 __mnt_drop_write(mnt);
Dave Hansenad775f52008-02-15 14:38:01 -0800218 file_release_write(file);
Dave Hansenaceaf782008-02-15 14:37:31 -0800219}
Dave Hansenaceaf782008-02-15 14:37:31 -0800220
Al Virod7065da2010-05-26 15:13:55 -0400221/* the real guts of fput() - releasing the last reference to file
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222 */
Al Virod7065da2010-05-26 15:13:55 -0400223static void __fput(struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800225 struct dentry *dentry = file->f_path.dentry;
226 struct vfsmount *mnt = file->f_path.mnt;
David Howellsc77cece2013-06-13 23:37:49 +0100227 struct inode *inode = file->f_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228
229 might_sleep();
Robert Love0eeca282005-07-12 17:06:03 -0400230
231 fsnotify_close(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232 /*
233 * The function eventpoll_release() should be the first called
234 * in the file cleanup chain.
235 */
236 eventpoll_release(file);
237 locks_remove_flock(file);
238
Al Viro233e70f2008-10-31 23:28:30 +0000239 if (unlikely(file->f_flags & FASYNC)) {
Al Viro72c2d532013-09-22 16:27:52 -0400240 if (file->f_op->fasync)
Al Viro233e70f2008-10-31 23:28:30 +0000241 file->f_op->fasync(-1, file, 0);
242 }
Mimi Zohar4199d352011-03-16 22:48:43 -0400243 ima_file_free(file);
Al Viro72c2d532013-09-22 16:27:52 -0400244 if (file->f_op->release)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245 file->f_op->release(inode, file);
246 security_file_free(file);
Miklos Szeredi60ed8cf2011-03-16 18:17:54 +0100247 if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL &&
248 !(file->f_mode & FMODE_PATH))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249 cdev_put(inode->i_cdev);
Miklos Szeredi60ed8cf2011-03-16 18:17:54 +0100250 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251 fops_put(file->f_op);
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700252 put_pid(file->f_owner.pid);
Mimi Zohar890275b52010-11-02 10:13:07 -0400253 if ((file->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
254 i_readcount_dec(inode);
Dave Hansenaceaf782008-02-15 14:37:31 -0800255 if (file->f_mode & FMODE_WRITE)
256 drop_file_write_access(file);
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800257 file->f_path.dentry = NULL;
258 file->f_path.mnt = NULL;
Al Virodd379782013-03-01 19:48:30 -0500259 file->f_inode = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260 file_free(file);
261 dput(dentry);
262 mntput(mnt);
263}
264
Oleg Nesterov4f5e65a2013-07-08 14:24:16 -0700265static LLIST_HEAD(delayed_fput_list);
Al Viro4a9d4b02012-06-24 09:56:45 +0400266static void delayed_fput(struct work_struct *unused)
267{
Oleg Nesterov4f5e65a2013-07-08 14:24:16 -0700268 struct llist_node *node = llist_del_all(&delayed_fput_list);
269 struct llist_node *next;
270
271 for (; node; node = next) {
272 next = llist_next(node);
273 __fput(llist_entry(node, struct file, f_u.fu_llist));
Al Viro4a9d4b02012-06-24 09:56:45 +0400274 }
275}
276
277static void ____fput(struct callback_head *work)
278{
279 __fput(container_of(work, struct file, f_u.fu_rcuhead));
280}
281
282/*
283 * If kernel thread really needs to have the final fput() it has done
284 * to complete, call this. The only user right now is the boot - we
285 * *do* need to make sure our writes to binaries on initramfs has
286 * not left us with opened struct file waiting for __fput() - execve()
287 * won't work without that. Please, don't add more callers without
288 * very good reasons; in particular, never call that with locks
289 * held and never call that from a thread that might need to do
290 * some work on any kind of umount.
291 */
292void flush_delayed_fput(void)
293{
294 delayed_fput(NULL);
295}
296
Al Viroc7314d72013-10-20 08:44:39 -0400297static DECLARE_DELAYED_WORK(delayed_fput_work, delayed_fput);
Al Viro4a9d4b02012-06-24 09:56:45 +0400298
Al Virod7065da2010-05-26 15:13:55 -0400299void fput(struct file *file)
300{
Al Viro4a9d4b02012-06-24 09:56:45 +0400301 if (atomic_long_dec_and_test(&file->f_count)) {
302 struct task_struct *task = current;
Oleg Nesterove7b2c402013-06-14 21:09:47 +0200303
Oleg Nesterove7b2c402013-06-14 21:09:47 +0200304 if (likely(!in_interrupt() && !(task->flags & PF_KTHREAD))) {
305 init_task_work(&file->f_u.fu_rcuhead, ____fput);
306 if (!task_work_add(task, &file->f_u.fu_rcuhead, true))
307 return;
Andrew Morton64372502013-07-08 14:24:15 -0700308 /*
309 * After this task has run exit_task_work(),
Andrew Mortonbe49b302013-09-11 14:24:34 -0700310 * task_work_add() will fail. Fall through to delayed
Andrew Morton64372502013-07-08 14:24:15 -0700311 * fput to avoid leaking *file.
312 */
Al Viro4a9d4b02012-06-24 09:56:45 +0400313 }
Oleg Nesterov4f5e65a2013-07-08 14:24:16 -0700314
315 if (llist_add(&file->f_u.fu_llist, &delayed_fput_list))
Al Viroc7314d72013-10-20 08:44:39 -0400316 schedule_delayed_work(&delayed_fput_work, 1);
Al Viro4a9d4b02012-06-24 09:56:45 +0400317 }
318}
319
320/*
321 * synchronous analog of fput(); for kernel threads that might be needed
322 * in some umount() (and thus can't use flush_delayed_fput() without
323 * risking deadlocks), need to wait for completion of __fput() and know
324 * for this specific struct file it won't involve anything that would
325 * need them. Use only if you really need it - at the very least,
326 * don't blindly convert fput() by kernel thread to that.
327 */
328void __fput_sync(struct file *file)
329{
330 if (atomic_long_dec_and_test(&file->f_count)) {
331 struct task_struct *task = current;
Al Viro4a9d4b02012-06-24 09:56:45 +0400332 BUG_ON(!(task->flags & PF_KTHREAD));
Al Virod7065da2010-05-26 15:13:55 -0400333 __fput(file);
Al Viro4a9d4b02012-06-24 09:56:45 +0400334 }
Al Virod7065da2010-05-26 15:13:55 -0400335}
336
337EXPORT_SYMBOL(fput);
338
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339void put_filp(struct file *file)
340{
Al Viro516e0cc2008-07-26 00:39:17 -0400341 if (atomic_long_dec_and_test(&file->f_count)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342 security_file_free(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343 file_free(file);
344 }
345}
346
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347void __init files_init(unsigned long mempages)
348{
Eric Dumazet518de9b2010-10-26 14:22:44 -0700349 unsigned long n;
Eric Dumazetb6b3fde2008-12-10 09:35:45 -0800350
351 filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
352 SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
353
354 /*
355 * One file with associated inode and dcache is very roughly 1K.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356 * Per default don't use more than 10% of our memory for files.
357 */
358
359 n = (mempages * (PAGE_SIZE / 1024)) / 10;
Eric Dumazet518de9b2010-10-26 14:22:44 -0700360 files_stat.max_files = max_t(unsigned long, n, NR_FILE);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700361 files_defer_init();
Mingming Cao0216bfc2006-06-23 02:05:41 -0700362 percpu_counter_init(&nr_files, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363}