blob: 72e861a35a7f7a05d185b496da5615c74a02552f [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>
Ingo Molnar5b825c32017-02-02 17:54:15 +010016#include <linux/cred.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/eventpoll.h>
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070018#include <linux/rcupdate.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070019#include <linux/mount.h>
Randy Dunlap16f7e0f2006-01-11 12:17:46 -080020#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070021#include <linux/cdev.h>
Robert Love0eeca282005-07-12 17:06:03 -040022#include <linux/fsnotify.h>
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080023#include <linux/sysctl.h>
24#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>
Mel Gorman4248b0d2015-08-06 15:46:20 -070029#include <linux/swap.h>
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080030
Arun Sharma600634972011-07-26 16:09:06 -070031#include <linux/atomic.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070032
Eric Parise81e3f42009-12-04 15:47:36 -050033#include "internal.h"
34
Linus Torvalds1da177e2005-04-16 15:20:36 -070035/* sysctl tunables... */
36struct files_stat_struct files_stat = {
37 .max_files = NR_FILE
38};
39
Eric Dumazetb6b3fde2008-12-10 09:35:45 -080040/* SLAB cache for file structures */
41static struct kmem_cache *filp_cachep __read_mostly;
42
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080043static struct percpu_counter nr_files __cacheline_aligned_in_smp;
Linus Torvalds1da177e2005-04-16 15:20:36 -070044
Al Viro5c33b182012-07-20 23:05:59 +040045static void file_free_rcu(struct rcu_head *head)
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070046{
David Howellsd76b0d92008-11-14 10:39:25 +110047 struct file *f = container_of(head, struct file, f_u.fu_rcuhead);
48
49 put_cred(f->f_cred);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070050 kmem_cache_free(filp_cachep, f);
51}
52
Linus Torvalds1da177e2005-04-16 15:20:36 -070053static inline void file_free(struct file *f)
54{
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080055 percpu_counter_dec(&nr_files);
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)
Joe Perches1f7e0612014-06-06 14:38:05 -070080int proc_nr_files(struct 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
Joe Perches1f7e0612014-06-06 14:38:05 -070087int proc_nr_files(struct 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);
Linus Torvalds9c225f22014-03-03 09:36:58 -0800138 mutex_init(&f->f_pos_lock);
Benjamin LaHaise5a6b7952006-03-23 03:01:03 -0800139 eventpoll_init_file(f);
140 /* f->f_version: 0 */
Kirill Korotaevaf4d2ec2005-06-23 00:09:50 -0700141 return f;
142
143over:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144 /* Ran out of filps - report that */
Dipankar Sarma529bf6b2006-03-07 21:55:35 -0800145 if (get_nr_files() > old_max) {
Eric Dumazet518de9b2010-10-26 14:22:44 -0700146 pr_info("VFS: file-max limit %lu reached\n", get_max_files());
Dipankar Sarma529bf6b2006-03-07 21:55:35 -0800147 old_max = get_nr_files();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148 }
Al Viro1afc99b2013-02-14 20:41:04 -0500149 return ERR_PTR(-ENFILE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150}
151
Dave Hansence8d2cd2007-10-16 23:31:13 -0700152/**
153 * alloc_file - allocate and initialize a 'struct file'
Eric Biggersa4576062014-10-12 14:29:29 -0500154 *
155 * @path: the (dentry, vfsmount) pair for the new file
Dave Hansence8d2cd2007-10-16 23:31:13 -0700156 * @mode: the mode with which the new file will be opened
157 * @fop: the 'struct file_operations' for the new file
Dave Hansence8d2cd2007-10-16 23:31:13 -0700158 */
Al Viroa4141d72016-11-20 20:28:12 -0500159struct file *alloc_file(const struct path *path, fmode_t mode,
Al Viro2c48b9c2009-08-09 00:52:35 +0400160 const struct file_operations *fop)
Dave Hansence8d2cd2007-10-16 23:31:13 -0700161{
162 struct file *file;
Dave Hansence8d2cd2007-10-16 23:31:13 -0700163
164 file = get_empty_filp();
Al Viro1afc99b2013-02-14 20:41:04 -0500165 if (IS_ERR(file))
Anatol Pomozov39b65252012-09-12 20:11:55 -0700166 return file;
Dave Hansence8d2cd2007-10-16 23:31:13 -0700167
Al Viro2c48b9c2009-08-09 00:52:35 +0400168 file->f_path = *path;
Al Virodd379782013-03-01 19:48:30 -0500169 file->f_inode = path->dentry->d_inode;
Al Viro2c48b9c2009-08-09 00:52:35 +0400170 file->f_mapping = path->dentry->d_inode->i_mapping;
Jeff Layton5660e132017-07-06 07:02:25 -0400171 file->f_wb_err = filemap_sample_wb_err(file->f_mapping);
Al Viro293bc982014-02-11 18:37:41 -0500172 if ((mode & FMODE_READ) &&
Al Viro84363182015-04-04 01:14:53 -0400173 likely(fop->read || fop->read_iter))
Al Viro7f7f25e2014-02-11 17:49:24 -0500174 mode |= FMODE_CAN_READ;
Al Viro293bc982014-02-11 18:37:41 -0500175 if ((mode & FMODE_WRITE) &&
Al Viro84363182015-04-04 01:14:53 -0400176 likely(fop->write || fop->write_iter))
Al Viro7f7f25e2014-02-11 17:49:24 -0500177 mode |= FMODE_CAN_WRITE;
Dave Hansence8d2cd2007-10-16 23:31:13 -0700178 file->f_mode = mode;
179 file->f_op = fop;
Mimi Zohar890275b52010-11-02 10:13:07 -0400180 if ((mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
181 i_readcount_inc(path->dentry->d_inode);
Al Viro3d1e4632009-08-08 23:56:29 +0400182 return file;
Dave Hansence8d2cd2007-10-16 23:31:13 -0700183}
Roland Dreier73efc462009-12-16 12:43:11 -0800184EXPORT_SYMBOL(alloc_file);
Dave Hansence8d2cd2007-10-16 23:31:13 -0700185
Al Virod7065da2010-05-26 15:13:55 -0400186/* the real guts of fput() - releasing the last reference to file
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187 */
Al Virod7065da2010-05-26 15:13:55 -0400188static void __fput(struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800190 struct dentry *dentry = file->f_path.dentry;
191 struct vfsmount *mnt = file->f_path.mnt;
David Howellsc77cece2013-06-13 23:37:49 +0100192 struct inode *inode = file->f_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193
194 might_sleep();
Robert Love0eeca282005-07-12 17:06:03 -0400195
196 fsnotify_close(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197 /*
198 * The function eventpoll_release() should be the first called
199 * in the file cleanup chain.
200 */
201 eventpoll_release(file);
Jeff Layton78ed8a12014-02-03 12:13:08 -0500202 locks_remove_file(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203
Al Viro233e70f2008-10-31 23:28:30 +0000204 if (unlikely(file->f_flags & FASYNC)) {
Al Viro72c2d532013-09-22 16:27:52 -0400205 if (file->f_op->fasync)
Al Viro233e70f2008-10-31 23:28:30 +0000206 file->f_op->fasync(-1, file, 0);
207 }
Mimi Zohar4199d352011-03-16 22:48:43 -0400208 ima_file_free(file);
Al Viro72c2d532013-09-22 16:27:52 -0400209 if (file->f_op->release)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210 file->f_op->release(inode, file);
211 security_file_free(file);
Miklos Szeredi60ed8cf2011-03-16 18:17:54 +0100212 if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL &&
213 !(file->f_mode & FMODE_PATH))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214 cdev_put(inode->i_cdev);
Miklos Szeredi60ed8cf2011-03-16 18:17:54 +0100215 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216 fops_put(file->f_op);
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700217 put_pid(file->f_owner.pid);
Mimi Zohar890275b52010-11-02 10:13:07 -0400218 if ((file->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
219 i_readcount_dec(inode);
Al Viro83f936c2014-03-14 12:02:47 -0400220 if (file->f_mode & FMODE_WRITER) {
221 put_write_access(inode);
222 __mnt_drop_write(mnt);
223 }
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800224 file->f_path.dentry = NULL;
225 file->f_path.mnt = NULL;
Al Virodd379782013-03-01 19:48:30 -0500226 file->f_inode = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227 file_free(file);
228 dput(dentry);
229 mntput(mnt);
230}
231
Oleg Nesterov4f5e65a2013-07-08 14:24:16 -0700232static LLIST_HEAD(delayed_fput_list);
Al Viro4a9d4b02012-06-24 09:56:45 +0400233static void delayed_fput(struct work_struct *unused)
234{
Oleg Nesterov4f5e65a2013-07-08 14:24:16 -0700235 struct llist_node *node = llist_del_all(&delayed_fput_list);
236 struct llist_node *next;
237
238 for (; node; node = next) {
239 next = llist_next(node);
240 __fput(llist_entry(node, struct file, f_u.fu_llist));
Al Viro4a9d4b02012-06-24 09:56:45 +0400241 }
242}
243
244static void ____fput(struct callback_head *work)
245{
246 __fput(container_of(work, struct file, f_u.fu_rcuhead));
247}
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 */
259void flush_delayed_fput(void)
260{
261 delayed_fput(NULL);
262}
263
Al Viroc7314d72013-10-20 08:44:39 -0400264static DECLARE_DELAYED_WORK(delayed_fput_work, delayed_fput);
Al Viro4a9d4b02012-06-24 09:56:45 +0400265
Al Virod7065da2010-05-26 15:13:55 -0400266void fput(struct file *file)
267{
Al Viro4a9d4b02012-06-24 09:56:45 +0400268 if (atomic_long_dec_and_test(&file->f_count)) {
269 struct task_struct *task = current;
Oleg Nesterove7b2c402013-06-14 21:09:47 +0200270
Oleg Nesterove7b2c402013-06-14 21:09:47 +0200271 if (likely(!in_interrupt() && !(task->flags & PF_KTHREAD))) {
272 init_task_work(&file->f_u.fu_rcuhead, ____fput);
273 if (!task_work_add(task, &file->f_u.fu_rcuhead, true))
274 return;
Andrew Morton64372502013-07-08 14:24:15 -0700275 /*
276 * After this task has run exit_task_work(),
Andrew Mortonbe49b302013-09-11 14:24:34 -0700277 * task_work_add() will fail. Fall through to delayed
Andrew Morton64372502013-07-08 14:24:15 -0700278 * fput to avoid leaking *file.
279 */
Al Viro4a9d4b02012-06-24 09:56:45 +0400280 }
Oleg Nesterov4f5e65a2013-07-08 14:24:16 -0700281
282 if (llist_add(&file->f_u.fu_llist, &delayed_fput_list))
Al Viroc7314d72013-10-20 08:44:39 -0400283 schedule_delayed_work(&delayed_fput_work, 1);
Al Viro4a9d4b02012-06-24 09:56:45 +0400284 }
285}
286
287/*
288 * synchronous analog of fput(); for kernel threads that might be needed
289 * in some umount() (and thus can't use flush_delayed_fput() without
290 * risking deadlocks), need to wait for completion of __fput() and know
291 * for this specific struct file it won't involve anything that would
292 * need them. Use only if you really need it - at the very least,
293 * don't blindly convert fput() by kernel thread to that.
294 */
295void __fput_sync(struct file *file)
296{
297 if (atomic_long_dec_and_test(&file->f_count)) {
298 struct task_struct *task = current;
Al Viro4a9d4b02012-06-24 09:56:45 +0400299 BUG_ON(!(task->flags & PF_KTHREAD));
Al Virod7065da2010-05-26 15:13:55 -0400300 __fput(file);
Al Viro4a9d4b02012-06-24 09:56:45 +0400301 }
Al Virod7065da2010-05-26 15:13:55 -0400302}
303
304EXPORT_SYMBOL(fput);
305
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306void put_filp(struct file *file)
307{
Al Viro516e0cc2008-07-26 00:39:17 -0400308 if (atomic_long_dec_and_test(&file->f_count)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309 security_file_free(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310 file_free(file);
311 }
312}
313
Mel Gorman4248b0d2015-08-06 15:46:20 -0700314void __init files_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315{
Eric Dumazetb6b3fde2008-12-10 09:35:45 -0800316 filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
317 SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
Tejun Heo908c7f12014-09-08 09:51:29 +0900318 percpu_counter_init(&nr_files, 0, GFP_KERNEL);
Mel Gorman4248b0d2015-08-06 15:46:20 -0700319}
320
321/*
322 * One file with associated inode and dcache is very roughly 1K. Per default
323 * do not use more than 10% of our memory for files.
324 */
325void __init files_maxfiles_init(void)
326{
327 unsigned long n;
328 unsigned long memreserve = (totalram_pages - nr_free_pages()) * 3/2;
329
330 memreserve = min(memreserve, totalram_pages - 1);
331 n = ((totalram_pages - memreserve) * (PAGE_SIZE / 1024)) / 10;
332
333 files_stat.max_files = max_t(unsigned long, n, NR_FILE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334}