blob: f45a4493f9e71b5afe52640f453a33bce0603c70 [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
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080035static struct percpu_counter nr_files __cacheline_aligned_in_smp;
Linus Torvalds1da177e2005-04-16 15:20:36 -070036
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070037static inline void file_free_rcu(struct rcu_head *head)
38{
Eric Dumazet2f512012005-10-30 15:02:16 -080039 struct file *f = container_of(head, struct file, f_u.fu_rcuhead);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070040 kmem_cache_free(filp_cachep, f);
41}
42
Linus Torvalds1da177e2005-04-16 15:20:36 -070043static inline void file_free(struct file *f)
44{
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080045 percpu_counter_dec(&nr_files);
Dave Hansenad775f52008-02-15 14:38:01 -080046 file_check_state(f);
Eric Dumazet2f512012005-10-30 15:02:16 -080047 call_rcu(&f->f_u.fu_rcuhead, file_free_rcu);
Linus Torvalds1da177e2005-04-16 15:20:36 -070048}
49
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080050/*
51 * Return the total number of open files in the system
52 */
53static int get_nr_files(void)
54{
55 return percpu_counter_read_positive(&nr_files);
56}
57
58/*
59 * Return the maximum number of open files in the system
60 */
61int get_max_files(void)
62{
63 return files_stat.max_files;
64}
65EXPORT_SYMBOL_GPL(get_max_files);
66
67/*
68 * Handle nr_files sysctl
69 */
70#if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
71int proc_nr_files(ctl_table *table, int write, struct file *filp,
72 void __user *buffer, size_t *lenp, loff_t *ppos)
73{
74 files_stat.nr_files = get_nr_files();
75 return proc_dointvec(table, write, filp, buffer, lenp, ppos);
76}
77#else
78int proc_nr_files(ctl_table *table, int write, struct file *filp,
79 void __user *buffer, size_t *lenp, loff_t *ppos)
80{
81 return -ENOSYS;
82}
83#endif
84
Linus Torvalds1da177e2005-04-16 15:20:36 -070085/* Find an unused file structure and return a pointer to it.
86 * Returns NULL, if there are no more free file structures or
87 * we run out of memory.
Dave Hansen430e2852008-02-15 14:37:26 -080088 *
89 * Be very careful using this. You are responsible for
90 * getting write access to any mount that you might assign
91 * to this filp, if it is opened for write. If this is not
92 * done, you will imbalance int the mount's writer count
93 * and a warning at __fput() time.
Linus Torvalds1da177e2005-04-16 15:20:36 -070094 */
95struct file *get_empty_filp(void)
96{
Benjamin LaHaise5a6b79512006-03-23 03:01:03 -080097 struct task_struct *tsk;
Kirill Korotaevaf4d2ec2005-06-23 00:09:50 -070098 static int old_max;
Linus Torvalds1da177e2005-04-16 15:20:36 -070099 struct file * f;
100
101 /*
102 * Privileged users can go above max_files
103 */
Dipankar Sarma529bf6b2006-03-07 21:55:35 -0800104 if (get_nr_files() >= files_stat.max_files && !capable(CAP_SYS_ADMIN)) {
105 /*
106 * percpu_counters are inaccurate. Do an expensive check before
107 * we go and fail.
108 */
Peter Zijlstra52d9f3b2007-10-16 23:25:44 -0700109 if (percpu_counter_sum_positive(&nr_files) >= files_stat.max_files)
Dipankar Sarma529bf6b2006-03-07 21:55:35 -0800110 goto over;
111 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112
Denis Cheng4975e452007-10-16 23:26:19 -0700113 f = kmem_cache_zalloc(filp_cachep, GFP_KERNEL);
Kirill Korotaevaf4d2ec2005-06-23 00:09:50 -0700114 if (f == NULL)
115 goto fail;
116
Dipankar Sarma529bf6b2006-03-07 21:55:35 -0800117 percpu_counter_inc(&nr_files);
Kirill Korotaevaf4d2ec2005-06-23 00:09:50 -0700118 if (security_file_alloc(f))
119 goto fail_sec;
120
Benjamin LaHaise5a6b79512006-03-23 03:01:03 -0800121 tsk = current;
Eric Dumazet2f512012005-10-30 15:02:16 -0800122 INIT_LIST_HEAD(&f->f_u.fu_list);
Al Viro516e0cc2008-07-26 00:39:17 -0400123 atomic_long_set(&f->f_count, 1);
Benjamin LaHaise5a6b79512006-03-23 03:01:03 -0800124 rwlock_init(&f->f_owner.lock);
125 f->f_uid = tsk->fsuid;
126 f->f_gid = tsk->fsgid;
127 eventpoll_init_file(f);
128 /* f->f_version: 0 */
Kirill Korotaevaf4d2ec2005-06-23 00:09:50 -0700129 return f;
130
131over:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132 /* Ran out of filps - report that */
Dipankar Sarma529bf6b2006-03-07 21:55:35 -0800133 if (get_nr_files() > old_max) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134 printk(KERN_INFO "VFS: file-max limit %d reached\n",
Dipankar Sarma529bf6b2006-03-07 21:55:35 -0800135 get_max_files());
136 old_max = get_nr_files();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137 }
Kirill Korotaevaf4d2ec2005-06-23 00:09:50 -0700138 goto fail;
139
140fail_sec:
141 file_free(f);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142fail:
143 return NULL;
144}
145
146EXPORT_SYMBOL(get_empty_filp);
147
Dave Hansence8d2cd2007-10-16 23:31:13 -0700148/**
149 * alloc_file - allocate and initialize a 'struct file'
150 * @mnt: the vfsmount on which the file will reside
151 * @dentry: the dentry representing the new file
152 * @mode: the mode with which the new file will be opened
153 * @fop: the 'struct file_operations' for the new file
154 *
155 * Use this instead of get_empty_filp() to get a new
156 * 'struct file'. Do so because of the same initialization
157 * pitfalls reasons listed for init_file(). This is a
158 * preferred interface to using init_file().
159 *
160 * If all the callers of init_file() are eliminated, its
161 * code should be moved into this function.
162 */
163struct file *alloc_file(struct vfsmount *mnt, struct dentry *dentry,
164 mode_t mode, const struct file_operations *fop)
165{
166 struct file *file;
167 struct path;
168
169 file = get_empty_filp();
170 if (!file)
171 return NULL;
172
173 init_file(file, mnt, dentry, mode, fop);
174 return file;
175}
176EXPORT_SYMBOL(alloc_file);
177
178/**
179 * init_file - initialize a 'struct file'
180 * @file: the already allocated 'struct file' to initialized
181 * @mnt: the vfsmount on which the file resides
182 * @dentry: the dentry representing this file
183 * @mode: the mode the file is opened with
184 * @fop: the 'struct file_operations' for this file
185 *
186 * Use this instead of setting the members directly. Doing so
187 * avoids making mistakes like forgetting the mntget() or
188 * forgetting to take a write on the mnt.
189 *
190 * Note: This is a crappy interface. It is here to make
191 * merging with the existing users of get_empty_filp()
192 * who have complex failure logic easier. All users
193 * of this should be moving to alloc_file().
194 */
195int init_file(struct file *file, struct vfsmount *mnt, struct dentry *dentry,
196 mode_t mode, const struct file_operations *fop)
197{
198 int error = 0;
199 file->f_path.dentry = dentry;
200 file->f_path.mnt = mntget(mnt);
201 file->f_mapping = dentry->d_inode->i_mapping;
202 file->f_mode = mode;
203 file->f_op = fop;
Dave Hansen4a3fd212008-02-15 14:37:48 -0800204
205 /*
206 * These mounts don't really matter in practice
207 * for r/o bind mounts. They aren't userspace-
208 * visible. We do this for consistency, and so
209 * that we can do debugging checks at __fput()
210 */
211 if ((mode & FMODE_WRITE) && !special_file(dentry->d_inode->i_mode)) {
Dave Hansenad775f52008-02-15 14:38:01 -0800212 file_take_write(file);
Dave Hansen4a3fd212008-02-15 14:37:48 -0800213 error = mnt_want_write(mnt);
214 WARN_ON(error);
215 }
Dave Hansence8d2cd2007-10-16 23:31:13 -0700216 return error;
217}
218EXPORT_SYMBOL(init_file);
219
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -0800220void fput(struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221{
Al Viro516e0cc2008-07-26 00:39:17 -0400222 if (atomic_long_dec_and_test(&file->f_count))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223 __fput(file);
224}
225
226EXPORT_SYMBOL(fput);
227
Dave Hansenaceaf782008-02-15 14:37:31 -0800228/**
229 * drop_file_write_access - give up ability to write to a file
230 * @file: the file to which we will stop writing
231 *
232 * This is a central place which will give up the ability
233 * to write to @file, along with access to write through
234 * its vfsmount.
235 */
236void drop_file_write_access(struct file *file)
237{
Dave Hansen4a3fd212008-02-15 14:37:48 -0800238 struct vfsmount *mnt = file->f_path.mnt;
Dave Hansenaceaf782008-02-15 14:37:31 -0800239 struct dentry *dentry = file->f_path.dentry;
240 struct inode *inode = dentry->d_inode;
241
242 put_write_access(inode);
Dave Hansenad775f52008-02-15 14:38:01 -0800243
244 if (special_file(inode->i_mode))
245 return;
246 if (file_check_writeable(file) != 0)
247 return;
248 mnt_drop_write(mnt);
249 file_release_write(file);
Dave Hansenaceaf782008-02-15 14:37:31 -0800250}
251EXPORT_SYMBOL_GPL(drop_file_write_access);
252
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253/* __fput is called from task context when aio completion releases the last
254 * last use of a struct file *. Do not use otherwise.
255 */
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -0800256void __fput(struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800258 struct dentry *dentry = file->f_path.dentry;
259 struct vfsmount *mnt = file->f_path.mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260 struct inode *inode = dentry->d_inode;
261
262 might_sleep();
Robert Love0eeca282005-07-12 17:06:03 -0400263
264 fsnotify_close(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265 /*
266 * The function eventpoll_release() should be the first called
267 * in the file cleanup chain.
268 */
269 eventpoll_release(file);
270 locks_remove_flock(file);
271
272 if (file->f_op && file->f_op->release)
273 file->f_op->release(inode, file);
274 security_file_free(file);
Theodore Ts'o577c4eb2006-09-27 01:50:49 -0700275 if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276 cdev_put(inode->i_cdev);
277 fops_put(file->f_op);
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700278 put_pid(file->f_owner.pid);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279 file_kill(file);
Dave Hansenaceaf782008-02-15 14:37:31 -0800280 if (file->f_mode & FMODE_WRITE)
281 drop_file_write_access(file);
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800282 file->f_path.dentry = NULL;
283 file->f_path.mnt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284 file_free(file);
285 dput(dentry);
286 mntput(mnt);
287}
288
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -0800289struct file *fget(unsigned int fd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290{
291 struct file *file;
292 struct files_struct *files = current->files;
293
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700294 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295 file = fcheck_files(files, fd);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700296 if (file) {
Al Viro516e0cc2008-07-26 00:39:17 -0400297 if (!atomic_long_inc_not_zero(&file->f_count)) {
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700298 /* File object ref couldn't be taken */
299 rcu_read_unlock();
300 return NULL;
301 }
302 }
303 rcu_read_unlock();
304
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305 return file;
306}
307
308EXPORT_SYMBOL(fget);
309
310/*
311 * Lightweight file lookup - no refcnt increment if fd table isn't shared.
312 * You can use this only if it is guranteed that the current task already
313 * holds a refcnt to that file. That check has to be done at fget() only
314 * and a flag is returned to be passed to the corresponding fput_light().
315 * There must not be a cloning between an fget_light/fput_light pair.
316 */
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -0800317struct file *fget_light(unsigned int fd, int *fput_needed)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318{
319 struct file *file;
320 struct files_struct *files = current->files;
321
322 *fput_needed = 0;
323 if (likely((atomic_read(&files->count) == 1))) {
324 file = fcheck_files(files, fd);
325 } else {
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700326 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327 file = fcheck_files(files, fd);
328 if (file) {
Al Viro516e0cc2008-07-26 00:39:17 -0400329 if (atomic_long_inc_not_zero(&file->f_count))
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700330 *fput_needed = 1;
331 else
332 /* Didn't get the reference, someone's freed */
333 file = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334 }
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700335 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336 }
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700337
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338 return file;
339}
340
341
342void put_filp(struct file *file)
343{
Al Viro516e0cc2008-07-26 00:39:17 -0400344 if (atomic_long_dec_and_test(&file->f_count)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345 security_file_free(file);
346 file_kill(file);
347 file_free(file);
348 }
349}
350
351void file_move(struct file *file, struct list_head *list)
352{
353 if (!list)
354 return;
355 file_list_lock();
Eric Dumazet2f512012005-10-30 15:02:16 -0800356 list_move(&file->f_u.fu_list, list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357 file_list_unlock();
358}
359
360void file_kill(struct file *file)
361{
Eric Dumazet2f512012005-10-30 15:02:16 -0800362 if (!list_empty(&file->f_u.fu_list)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363 file_list_lock();
Eric Dumazet2f512012005-10-30 15:02:16 -0800364 list_del_init(&file->f_u.fu_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365 file_list_unlock();
366 }
367}
368
369int fs_may_remount_ro(struct super_block *sb)
370{
Matthias Kaehlckecfdaf9e2007-10-18 23:39:56 -0700371 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372
373 /* Check that no files are currently opened for writing. */
374 file_list_lock();
Matthias Kaehlckecfdaf9e2007-10-18 23:39:56 -0700375 list_for_each_entry(file, &sb->s_files, f_u.fu_list) {
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800376 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377
378 /* File with pending delete? */
379 if (inode->i_nlink == 0)
380 goto too_bad;
381
382 /* Writeable file? */
383 if (S_ISREG(inode->i_mode) && (file->f_mode & FMODE_WRITE))
384 goto too_bad;
385 }
386 file_list_unlock();
387 return 1; /* Tis' cool bro. */
388too_bad:
389 file_list_unlock();
390 return 0;
391}
392
393void __init files_init(unsigned long mempages)
394{
395 int n;
396 /* One file with associated inode and dcache is very roughly 1K.
397 * Per default don't use more than 10% of our memory for files.
398 */
399
400 n = (mempages * (PAGE_SIZE / 1024)) / 10;
401 files_stat.max_files = n;
402 if (files_stat.max_files < NR_FILE)
403 files_stat.max_files = NR_FILE;
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700404 files_defer_init();
Mingming Cao0216bfc2006-06-23 02:05:41 -0700405 percpu_counter_init(&nr_files, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406}