blob: 4bef4c01ec6ff4b0ef7ce00816f3c124517031c0 [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>
24
25#include <asm/atomic.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070026
27/* sysctl tunables... */
28struct files_stat_struct files_stat = {
29 .max_files = NR_FILE
30};
31
Linus Torvalds1da177e2005-04-16 15:20:36 -070032/* public. Not pretty! */
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080033__cacheline_aligned_in_smp DEFINE_SPINLOCK(files_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070034
Eric Dumazetb6b3fde2008-12-10 09:35:45 -080035/* SLAB cache for file structures */
36static struct kmem_cache *filp_cachep __read_mostly;
37
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080038static struct percpu_counter nr_files __cacheline_aligned_in_smp;
Linus Torvalds1da177e2005-04-16 15:20:36 -070039
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070040static inline void file_free_rcu(struct rcu_head *head)
41{
David Howellsd76b0d92008-11-14 10:39:25 +110042 struct file *f = container_of(head, struct file, f_u.fu_rcuhead);
43
44 put_cred(f->f_cred);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070045 kmem_cache_free(filp_cachep, f);
46}
47
Linus Torvalds1da177e2005-04-16 15:20:36 -070048static inline void file_free(struct file *f)
49{
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080050 percpu_counter_dec(&nr_files);
Dave Hansenad775f52008-02-15 14:38:01 -080051 file_check_state(f);
Eric Dumazet2f512012005-10-30 15:02:16 -080052 call_rcu(&f->f_u.fu_rcuhead, file_free_rcu);
Linus Torvalds1da177e2005-04-16 15:20:36 -070053}
54
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080055/*
56 * Return the total number of open files in the system
57 */
58static int get_nr_files(void)
59{
60 return percpu_counter_read_positive(&nr_files);
61}
62
63/*
64 * Return the maximum number of open files in the system
65 */
66int get_max_files(void)
67{
68 return files_stat.max_files;
69}
70EXPORT_SYMBOL_GPL(get_max_files);
71
72/*
73 * Handle nr_files sysctl
74 */
75#if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
Alexey Dobriyan8d65af72009-09-23 15:57:19 -070076int proc_nr_files(ctl_table *table, int write,
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080077 void __user *buffer, size_t *lenp, loff_t *ppos)
78{
79 files_stat.nr_files = get_nr_files();
Alexey Dobriyan8d65af72009-09-23 15:57:19 -070080 return proc_dointvec(table, write, buffer, lenp, ppos);
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080081}
82#else
Alexey Dobriyan8d65af72009-09-23 15:57:19 -070083int proc_nr_files(ctl_table *table, int write,
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080084 void __user *buffer, size_t *lenp, loff_t *ppos)
85{
86 return -ENOSYS;
87}
88#endif
89
Linus Torvalds1da177e2005-04-16 15:20:36 -070090/* Find an unused file structure and return a pointer to it.
91 * Returns NULL, if there are no more free file structures or
92 * we run out of memory.
Dave Hansen430e2852008-02-15 14:37:26 -080093 *
94 * Be very careful using this. You are responsible for
95 * getting write access to any mount that you might assign
96 * to this filp, if it is opened for write. If this is not
97 * done, you will imbalance int the mount's writer count
98 * and a warning at __fput() time.
Linus Torvalds1da177e2005-04-16 15:20:36 -070099 */
100struct file *get_empty_filp(void)
101{
David Howells86a264a2008-11-14 10:39:18 +1100102 const struct cred *cred = current_cred();
Kirill Korotaevaf4d2ec2005-06-23 00:09:50 -0700103 static int old_max;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104 struct file * f;
105
106 /*
107 * Privileged users can go above max_files
108 */
Dipankar Sarma529bf6b2006-03-07 21:55:35 -0800109 if (get_nr_files() >= files_stat.max_files && !capable(CAP_SYS_ADMIN)) {
110 /*
111 * percpu_counters are inaccurate. Do an expensive check before
112 * we go and fail.
113 */
Peter Zijlstra52d9f3b2007-10-16 23:25:44 -0700114 if (percpu_counter_sum_positive(&nr_files) >= files_stat.max_files)
Dipankar Sarma529bf6b2006-03-07 21:55:35 -0800115 goto over;
116 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117
Denis Cheng4975e452007-10-16 23:26:19 -0700118 f = kmem_cache_zalloc(filp_cachep, GFP_KERNEL);
Kirill Korotaevaf4d2ec2005-06-23 00:09:50 -0700119 if (f == NULL)
120 goto fail;
121
Dipankar Sarma529bf6b2006-03-07 21:55:35 -0800122 percpu_counter_inc(&nr_files);
Kirill Korotaevaf4d2ec2005-06-23 00:09:50 -0700123 if (security_file_alloc(f))
124 goto fail_sec;
125
Eric Dumazet2f512012005-10-30 15:02:16 -0800126 INIT_LIST_HEAD(&f->f_u.fu_list);
Al Viro516e0cc2008-07-26 00:39:17 -0400127 atomic_long_set(&f->f_count, 1);
Benjamin LaHaise5a6b7952006-03-23 03:01:03 -0800128 rwlock_init(&f->f_owner.lock);
David Howellsd76b0d92008-11-14 10:39:25 +1100129 f->f_cred = get_cred(cred);
Jonathan Corbet68499912009-02-06 13:52:43 -0700130 spin_lock_init(&f->f_lock);
Benjamin LaHaise5a6b7952006-03-23 03:01:03 -0800131 eventpoll_init_file(f);
132 /* f->f_version: 0 */
Kirill Korotaevaf4d2ec2005-06-23 00:09:50 -0700133 return f;
134
135over:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136 /* Ran out of filps - report that */
Dipankar Sarma529bf6b2006-03-07 21:55:35 -0800137 if (get_nr_files() > old_max) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138 printk(KERN_INFO "VFS: file-max limit %d reached\n",
Dipankar Sarma529bf6b2006-03-07 21:55:35 -0800139 get_max_files());
140 old_max = get_nr_files();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141 }
Kirill Korotaevaf4d2ec2005-06-23 00:09:50 -0700142 goto fail;
143
144fail_sec:
145 file_free(f);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146fail:
147 return NULL;
148}
149
150EXPORT_SYMBOL(get_empty_filp);
151
Dave Hansence8d2cd2007-10-16 23:31:13 -0700152/**
153 * alloc_file - allocate and initialize a 'struct file'
154 * @mnt: the vfsmount on which the file will reside
155 * @dentry: the dentry representing the new file
156 * @mode: the mode with which the new file will be opened
157 * @fop: the 'struct file_operations' for the new file
158 *
159 * Use this instead of get_empty_filp() to get a new
160 * 'struct file'. Do so because of the same initialization
161 * pitfalls reasons listed for init_file(). This is a
162 * preferred interface to using init_file().
163 *
164 * If all the callers of init_file() are eliminated, its
165 * code should be moved into this function.
166 */
167struct file *alloc_file(struct vfsmount *mnt, struct dentry *dentry,
Al Viroaeb5d722008-09-02 15:28:45 -0400168 fmode_t mode, const struct file_operations *fop)
Dave Hansence8d2cd2007-10-16 23:31:13 -0700169{
170 struct file *file;
Dave Hansence8d2cd2007-10-16 23:31:13 -0700171
172 file = get_empty_filp();
173 if (!file)
174 return NULL;
175
176 init_file(file, mnt, dentry, mode, fop);
177 return file;
178}
179EXPORT_SYMBOL(alloc_file);
180
181/**
182 * init_file - initialize a 'struct file'
183 * @file: the already allocated 'struct file' to initialized
184 * @mnt: the vfsmount on which the file resides
185 * @dentry: the dentry representing this file
186 * @mode: the mode the file is opened with
187 * @fop: the 'struct file_operations' for this file
188 *
189 * Use this instead of setting the members directly. Doing so
190 * avoids making mistakes like forgetting the mntget() or
191 * forgetting to take a write on the mnt.
192 *
193 * Note: This is a crappy interface. It is here to make
194 * merging with the existing users of get_empty_filp()
195 * who have complex failure logic easier. All users
196 * of this should be moving to alloc_file().
197 */
198int init_file(struct file *file, struct vfsmount *mnt, struct dentry *dentry,
Al Viroaeb5d722008-09-02 15:28:45 -0400199 fmode_t mode, const struct file_operations *fop)
Dave Hansence8d2cd2007-10-16 23:31:13 -0700200{
201 int error = 0;
202 file->f_path.dentry = dentry;
203 file->f_path.mnt = mntget(mnt);
204 file->f_mapping = dentry->d_inode->i_mapping;
205 file->f_mode = mode;
206 file->f_op = fop;
Dave Hansen4a3fd212008-02-15 14:37:48 -0800207
208 /*
209 * These mounts don't really matter in practice
210 * for r/o bind mounts. They aren't userspace-
211 * visible. We do this for consistency, and so
212 * that we can do debugging checks at __fput()
213 */
214 if ((mode & FMODE_WRITE) && !special_file(dentry->d_inode->i_mode)) {
Dave Hansenad775f52008-02-15 14:38:01 -0800215 file_take_write(file);
npiggin@suse.de96029c42009-04-26 20:25:55 +1000216 error = mnt_clone_write(mnt);
Dave Hansen4a3fd212008-02-15 14:37:48 -0800217 WARN_ON(error);
218 }
Dave Hansence8d2cd2007-10-16 23:31:13 -0700219 return error;
220}
221EXPORT_SYMBOL(init_file);
222
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -0800223void fput(struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224{
Al Viro516e0cc2008-07-26 00:39:17 -0400225 if (atomic_long_dec_and_test(&file->f_count))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226 __fput(file);
227}
228
229EXPORT_SYMBOL(fput);
230
Dave Hansenaceaf782008-02-15 14:37:31 -0800231/**
232 * drop_file_write_access - give up ability to write to a file
233 * @file: the file to which we will stop writing
234 *
235 * This is a central place which will give up the ability
236 * to write to @file, along with access to write through
237 * its vfsmount.
238 */
239void drop_file_write_access(struct file *file)
240{
Dave Hansen4a3fd212008-02-15 14:37:48 -0800241 struct vfsmount *mnt = file->f_path.mnt;
Dave Hansenaceaf782008-02-15 14:37:31 -0800242 struct dentry *dentry = file->f_path.dentry;
243 struct inode *inode = dentry->d_inode;
244
245 put_write_access(inode);
Dave Hansenad775f52008-02-15 14:38:01 -0800246
247 if (special_file(inode->i_mode))
248 return;
249 if (file_check_writeable(file) != 0)
250 return;
251 mnt_drop_write(mnt);
252 file_release_write(file);
Dave Hansenaceaf782008-02-15 14:37:31 -0800253}
254EXPORT_SYMBOL_GPL(drop_file_write_access);
255
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256/* __fput is called from task context when aio completion releases the last
257 * last use of a struct file *. Do not use otherwise.
258 */
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -0800259void __fput(struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800261 struct dentry *dentry = file->f_path.dentry;
262 struct vfsmount *mnt = file->f_path.mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 struct inode *inode = dentry->d_inode;
264
265 might_sleep();
Robert Love0eeca282005-07-12 17:06:03 -0400266
267 fsnotify_close(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268 /*
269 * The function eventpoll_release() should be the first called
270 * in the file cleanup chain.
271 */
272 eventpoll_release(file);
273 locks_remove_flock(file);
274
Al Viro233e70f2008-10-31 23:28:30 +0000275 if (unlikely(file->f_flags & FASYNC)) {
276 if (file->f_op && file->f_op->fasync)
277 file->f_op->fasync(-1, file, 0);
278 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279 if (file->f_op && file->f_op->release)
280 file->f_op->release(inode, file);
281 security_file_free(file);
Theodore Ts'o577c4eb2006-09-27 01:50:49 -0700282 if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283 cdev_put(inode->i_cdev);
284 fops_put(file->f_op);
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700285 put_pid(file->f_owner.pid);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286 file_kill(file);
Dave Hansenaceaf782008-02-15 14:37:31 -0800287 if (file->f_mode & FMODE_WRITE)
288 drop_file_write_access(file);
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800289 file->f_path.dentry = NULL;
290 file->f_path.mnt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291 file_free(file);
292 dput(dentry);
293 mntput(mnt);
294}
295
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -0800296struct file *fget(unsigned int fd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297{
298 struct file *file;
299 struct files_struct *files = current->files;
300
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700301 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302 file = fcheck_files(files, fd);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700303 if (file) {
Al Viro516e0cc2008-07-26 00:39:17 -0400304 if (!atomic_long_inc_not_zero(&file->f_count)) {
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700305 /* File object ref couldn't be taken */
306 rcu_read_unlock();
307 return NULL;
308 }
309 }
310 rcu_read_unlock();
311
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312 return file;
313}
314
315EXPORT_SYMBOL(fget);
316
317/*
318 * Lightweight file lookup - no refcnt increment if fd table isn't shared.
319 * You can use this only if it is guranteed that the current task already
320 * holds a refcnt to that file. That check has to be done at fget() only
321 * and a flag is returned to be passed to the corresponding fput_light().
322 * There must not be a cloning between an fget_light/fput_light pair.
323 */
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -0800324struct file *fget_light(unsigned int fd, int *fput_needed)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325{
326 struct file *file;
327 struct files_struct *files = current->files;
328
329 *fput_needed = 0;
330 if (likely((atomic_read(&files->count) == 1))) {
331 file = fcheck_files(files, fd);
332 } else {
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700333 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334 file = fcheck_files(files, fd);
335 if (file) {
Al Viro516e0cc2008-07-26 00:39:17 -0400336 if (atomic_long_inc_not_zero(&file->f_count))
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700337 *fput_needed = 1;
338 else
339 /* Didn't get the reference, someone's freed */
340 file = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341 }
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700342 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343 }
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700344
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345 return file;
346}
347
348
349void put_filp(struct file *file)
350{
Al Viro516e0cc2008-07-26 00:39:17 -0400351 if (atomic_long_dec_and_test(&file->f_count)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352 security_file_free(file);
353 file_kill(file);
354 file_free(file);
355 }
356}
357
358void file_move(struct file *file, struct list_head *list)
359{
360 if (!list)
361 return;
362 file_list_lock();
Eric Dumazet2f512012005-10-30 15:02:16 -0800363 list_move(&file->f_u.fu_list, list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364 file_list_unlock();
365}
366
367void file_kill(struct file *file)
368{
Eric Dumazet2f512012005-10-30 15:02:16 -0800369 if (!list_empty(&file->f_u.fu_list)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370 file_list_lock();
Eric Dumazet2f512012005-10-30 15:02:16 -0800371 list_del_init(&file->f_u.fu_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372 file_list_unlock();
373 }
374}
375
376int fs_may_remount_ro(struct super_block *sb)
377{
Matthias Kaehlckecfdaf9e2007-10-18 23:39:56 -0700378 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379
380 /* Check that no files are currently opened for writing. */
381 file_list_lock();
Matthias Kaehlckecfdaf9e2007-10-18 23:39:56 -0700382 list_for_each_entry(file, &sb->s_files, f_u.fu_list) {
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800383 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384
385 /* File with pending delete? */
386 if (inode->i_nlink == 0)
387 goto too_bad;
388
389 /* Writeable file? */
390 if (S_ISREG(inode->i_mode) && (file->f_mode & FMODE_WRITE))
391 goto too_bad;
392 }
393 file_list_unlock();
394 return 1; /* Tis' cool bro. */
395too_bad:
396 file_list_unlock();
397 return 0;
398}
399
npiggin@suse.de864d7c42009-04-26 20:25:56 +1000400/**
401 * mark_files_ro - mark all files read-only
402 * @sb: superblock in question
403 *
404 * All files are marked read-only. We don't care about pending
405 * delete files so this should be used in 'force' mode only.
406 */
407void mark_files_ro(struct super_block *sb)
408{
409 struct file *f;
410
411retry:
412 file_list_lock();
413 list_for_each_entry(f, &sb->s_files, f_u.fu_list) {
414 struct vfsmount *mnt;
415 if (!S_ISREG(f->f_path.dentry->d_inode->i_mode))
416 continue;
417 if (!file_count(f))
418 continue;
419 if (!(f->f_mode & FMODE_WRITE))
420 continue;
421 f->f_mode &= ~FMODE_WRITE;
422 if (file_check_writeable(f) != 0)
423 continue;
424 file_release_write(f);
425 mnt = mntget(f->f_path.mnt);
426 file_list_unlock();
427 /*
428 * This can sleep, so we can't hold
429 * the file_list_lock() spinlock.
430 */
431 mnt_drop_write(mnt);
432 mntput(mnt);
433 goto retry;
434 }
435 file_list_unlock();
436}
437
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438void __init files_init(unsigned long mempages)
439{
440 int n;
Eric Dumazetb6b3fde2008-12-10 09:35:45 -0800441
442 filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
443 SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
444
445 /*
446 * One file with associated inode and dcache is very roughly 1K.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447 * Per default don't use more than 10% of our memory for files.
448 */
449
450 n = (mempages * (PAGE_SIZE / 1024)) / 10;
451 files_stat.max_files = n;
452 if (files_stat.max_files < NR_FILE)
453 files_stat.max_files = NR_FILE;
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700454 files_defer_init();
Mingming Cao0216bfc2006-06-23 02:05:41 -0700455 percpu_counter_init(&nr_files, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456}