2 * linux/fs/file_table.c
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 * Copyright (C) 1997 David S. Miller (davem@caip.rutgers.edu)
8 #include <linux/string.h>
9 #include <linux/slab.h>
10 #include <linux/file.h>
11 #include <linux/fdtable.h>
12 #include <linux/init.h>
13 #include <linux/module.h>
15 #include <linux/security.h>
16 #include <linux/ima.h>
17 #include <linux/eventpoll.h>
18 #include <linux/rcupdate.h>
19 #include <linux/mount.h>
20 #include <linux/capability.h>
21 #include <linux/cdev.h>
22 #include <linux/fsnotify.h>
23 #include <linux/sysctl.h>
24 #include <linux/percpu_counter.h>
26 #include <asm/atomic.h>
28 /* sysctl tunables... */
29 struct files_stat_struct files_stat
= {
33 /* public. Not pretty! */
34 __cacheline_aligned_in_smp
DEFINE_SPINLOCK(files_lock
);
36 static struct percpu_counter nr_files __cacheline_aligned_in_smp
;
38 static inline void file_free_rcu(struct rcu_head
*head
)
40 struct file
*f
= container_of(head
, struct file
, f_u
.fu_rcuhead
);
43 kmem_cache_free(filp_cachep
, f
);
46 static inline void file_free(struct file
*f
)
48 percpu_counter_dec(&nr_files
);
50 call_rcu(&f
->f_u
.fu_rcuhead
, file_free_rcu
);
54 * Return the total number of open files in the system
56 static int get_nr_files(void)
58 return percpu_counter_read_positive(&nr_files
);
62 * Return the maximum number of open files in the system
64 int get_max_files(void)
66 return files_stat
.max_files
;
68 EXPORT_SYMBOL_GPL(get_max_files
);
71 * Handle nr_files sysctl
73 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
74 int proc_nr_files(ctl_table
*table
, int write
, struct file
*filp
,
75 void __user
*buffer
, size_t *lenp
, loff_t
*ppos
)
77 files_stat
.nr_files
= get_nr_files();
78 return proc_dointvec(table
, write
, filp
, buffer
, lenp
, ppos
);
81 int proc_nr_files(ctl_table
*table
, int write
, struct file
*filp
,
82 void __user
*buffer
, size_t *lenp
, loff_t
*ppos
)
88 /* Find an unused file structure and return a pointer to it.
89 * Returns NULL, if there are no more free file structures or
90 * we run out of memory.
92 * Be very careful using this. You are responsible for
93 * getting write access to any mount that you might assign
94 * to this filp, if it is opened for write. If this is not
95 * done, you will imbalance int the mount's writer count
96 * and a warning at __fput() time.
98 struct file
*get_empty_filp(void)
100 const struct cred
*cred
= current_cred();
105 * Privileged users can go above max_files
107 if (get_nr_files() >= files_stat
.max_files
&& !capable(CAP_SYS_ADMIN
)) {
109 * percpu_counters are inaccurate. Do an expensive check before
112 if (percpu_counter_sum_positive(&nr_files
) >= files_stat
.max_files
)
116 f
= kmem_cache_zalloc(filp_cachep
, GFP_KERNEL
);
120 percpu_counter_inc(&nr_files
);
121 if (security_file_alloc(f
))
124 INIT_LIST_HEAD(&f
->f_u
.fu_list
);
125 atomic_long_set(&f
->f_count
, 1);
126 rwlock_init(&f
->f_owner
.lock
);
127 f
->f_cred
= get_cred(cred
);
128 eventpoll_init_file(f
);
129 /* f->f_version: 0 */
133 /* Ran out of filps - report that */
134 if (get_nr_files() > old_max
) {
135 printk(KERN_INFO
"VFS: file-max limit %d reached\n",
137 old_max
= get_nr_files();
147 EXPORT_SYMBOL(get_empty_filp
);
150 * alloc_file - allocate and initialize a 'struct file'
151 * @mnt: the vfsmount on which the file will reside
152 * @dentry: the dentry representing the new file
153 * @mode: the mode with which the new file will be opened
154 * @fop: the 'struct file_operations' for the new file
156 * Use this instead of get_empty_filp() to get a new
157 * 'struct file'. Do so because of the same initialization
158 * pitfalls reasons listed for init_file(). This is a
159 * preferred interface to using init_file().
161 * If all the callers of init_file() are eliminated, its
162 * code should be moved into this function.
164 struct file
*alloc_file(struct vfsmount
*mnt
, struct dentry
*dentry
,
165 fmode_t mode
, const struct file_operations
*fop
)
170 file
= get_empty_filp();
174 init_file(file
, mnt
, dentry
, mode
, fop
);
177 EXPORT_SYMBOL(alloc_file
);
180 * init_file - initialize a 'struct file'
181 * @file: the already allocated 'struct file' to initialized
182 * @mnt: the vfsmount on which the file resides
183 * @dentry: the dentry representing this file
184 * @mode: the mode the file is opened with
185 * @fop: the 'struct file_operations' for this file
187 * Use this instead of setting the members directly. Doing so
188 * avoids making mistakes like forgetting the mntget() or
189 * forgetting to take a write on the mnt.
191 * Note: This is a crappy interface. It is here to make
192 * merging with the existing users of get_empty_filp()
193 * who have complex failure logic easier. All users
194 * of this should be moving to alloc_file().
196 int init_file(struct file
*file
, struct vfsmount
*mnt
, struct dentry
*dentry
,
197 fmode_t mode
, const struct file_operations
*fop
)
200 file
->f_path
.dentry
= dentry
;
201 file
->f_path
.mnt
= mntget(mnt
);
202 file
->f_mapping
= dentry
->d_inode
->i_mapping
;
207 * These mounts don't really matter in practice
208 * for r/o bind mounts. They aren't userspace-
209 * visible. We do this for consistency, and so
210 * that we can do debugging checks at __fput()
212 if ((mode
& FMODE_WRITE
) && !special_file(dentry
->d_inode
->i_mode
)) {
213 file_take_write(file
);
214 error
= mnt_want_write(mnt
);
219 EXPORT_SYMBOL(init_file
);
221 void fput(struct file
*file
)
223 if (atomic_long_dec_and_test(&file
->f_count
))
230 * drop_file_write_access - give up ability to write to a file
231 * @file: the file to which we will stop writing
233 * This is a central place which will give up the ability
234 * to write to @file, along with access to write through
237 void drop_file_write_access(struct file
*file
)
239 struct vfsmount
*mnt
= file
->f_path
.mnt
;
240 struct dentry
*dentry
= file
->f_path
.dentry
;
241 struct inode
*inode
= dentry
->d_inode
;
243 put_write_access(inode
);
245 if (special_file(inode
->i_mode
))
247 if (file_check_writeable(file
) != 0)
250 file_release_write(file
);
252 EXPORT_SYMBOL_GPL(drop_file_write_access
);
254 /* __fput is called from task context when aio completion releases the last
255 * last use of a struct file *. Do not use otherwise.
257 void __fput(struct file
*file
)
259 struct dentry
*dentry
= file
->f_path
.dentry
;
260 struct vfsmount
*mnt
= file
->f_path
.mnt
;
261 struct inode
*inode
= dentry
->d_inode
;
265 fsnotify_close(file
);
267 * The function eventpoll_release() should be the first called
268 * in the file cleanup chain.
270 eventpoll_release(file
);
271 locks_remove_flock(file
);
273 if (unlikely(file
->f_flags
& FASYNC
)) {
274 if (file
->f_op
&& file
->f_op
->fasync
)
275 file
->f_op
->fasync(-1, file
, 0);
277 if (file
->f_op
&& file
->f_op
->release
)
278 file
->f_op
->release(inode
, file
);
279 security_file_free(file
);
281 if (unlikely(S_ISCHR(inode
->i_mode
) && inode
->i_cdev
!= NULL
))
282 cdev_put(inode
->i_cdev
);
283 fops_put(file
->f_op
);
284 put_pid(file
->f_owner
.pid
);
286 if (file
->f_mode
& FMODE_WRITE
)
287 drop_file_write_access(file
);
288 file
->f_path
.dentry
= NULL
;
289 file
->f_path
.mnt
= NULL
;
295 struct file
*fget(unsigned int fd
)
298 struct files_struct
*files
= current
->files
;
301 file
= fcheck_files(files
, fd
);
303 if (!atomic_long_inc_not_zero(&file
->f_count
)) {
304 /* File object ref couldn't be taken */
317 * Lightweight file lookup - no refcnt increment if fd table isn't shared.
318 * You can use this only if it is guranteed that the current task already
319 * holds a refcnt to that file. That check has to be done at fget() only
320 * and a flag is returned to be passed to the corresponding fput_light().
321 * There must not be a cloning between an fget_light/fput_light pair.
323 struct file
*fget_light(unsigned int fd
, int *fput_needed
)
326 struct files_struct
*files
= current
->files
;
329 if (likely((atomic_read(&files
->count
) == 1))) {
330 file
= fcheck_files(files
, fd
);
333 file
= fcheck_files(files
, fd
);
335 if (atomic_long_inc_not_zero(&file
->f_count
))
338 /* Didn't get the reference, someone's freed */
348 void put_filp(struct file
*file
)
350 if (atomic_long_dec_and_test(&file
->f_count
)) {
351 security_file_free(file
);
357 void file_move(struct file
*file
, struct list_head
*list
)
362 list_move(&file
->f_u
.fu_list
, list
);
366 void file_kill(struct file
*file
)
368 if (!list_empty(&file
->f_u
.fu_list
)) {
370 list_del_init(&file
->f_u
.fu_list
);
375 int fs_may_remount_ro(struct super_block
*sb
)
379 /* Check that no files are currently opened for writing. */
381 list_for_each_entry(file
, &sb
->s_files
, f_u
.fu_list
) {
382 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
384 /* File with pending delete? */
385 if (inode
->i_nlink
== 0)
388 /* Writeable file? */
389 if (S_ISREG(inode
->i_mode
) && (file
->f_mode
& FMODE_WRITE
))
393 return 1; /* Tis' cool bro. */
399 void __init
files_init(unsigned long mempages
)
402 /* One file with associated inode and dcache is very roughly 1K.
403 * Per default don't use more than 10% of our memory for files.
406 n
= (mempages
* (PAGE_SIZE
/ 1024)) / 10;
407 files_stat
.max_files
= n
;
408 if (files_stat
.max_files
< NR_FILE
)
409 files_stat
.max_files
= NR_FILE
;
411 percpu_counter_init(&nr_files
, 0);