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/eventpoll.h>
17 #include <linux/rcupdate.h>
18 #include <linux/mount.h>
19 #include <linux/capability.h>
20 #include <linux/cdev.h>
21 #include <linux/fsnotify.h>
22 #include <linux/sysctl.h>
23 #include <linux/percpu_counter.h>
25 #include <asm/atomic.h>
27 /* sysctl tunables... */
28 struct files_stat_struct files_stat
= {
32 /* public. Not pretty! */
33 __cacheline_aligned_in_smp
DEFINE_SPINLOCK(files_lock
);
35 /* SLAB cache for file structures */
36 static struct kmem_cache
*filp_cachep __read_mostly
;
38 static struct percpu_counter nr_files __cacheline_aligned_in_smp
;
40 static inline void file_free_rcu(struct rcu_head
*head
)
42 struct file
*f
= container_of(head
, struct file
, f_u
.fu_rcuhead
);
45 kmem_cache_free(filp_cachep
, f
);
48 static inline void file_free(struct file
*f
)
50 percpu_counter_dec(&nr_files
);
52 call_rcu(&f
->f_u
.fu_rcuhead
, file_free_rcu
);
56 * Return the total number of open files in the system
58 static int get_nr_files(void)
60 return percpu_counter_read_positive(&nr_files
);
64 * Return the maximum number of open files in the system
66 int get_max_files(void)
68 return files_stat
.max_files
;
70 EXPORT_SYMBOL_GPL(get_max_files
);
73 * Handle nr_files sysctl
75 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
76 int proc_nr_files(ctl_table
*table
, int write
,
77 void __user
*buffer
, size_t *lenp
, loff_t
*ppos
)
79 files_stat
.nr_files
= get_nr_files();
80 return proc_dointvec(table
, write
, buffer
, lenp
, ppos
);
83 int proc_nr_files(ctl_table
*table
, int write
,
84 void __user
*buffer
, size_t *lenp
, loff_t
*ppos
)
90 /* 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.
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.
100 struct file
*get_empty_filp(void)
102 const struct cred
*cred
= current_cred();
107 * Privileged users can go above max_files
109 if (get_nr_files() >= files_stat
.max_files
&& !capable(CAP_SYS_ADMIN
)) {
111 * percpu_counters are inaccurate. Do an expensive check before
114 if (percpu_counter_sum_positive(&nr_files
) >= files_stat
.max_files
)
118 f
= kmem_cache_zalloc(filp_cachep
, GFP_KERNEL
);
122 percpu_counter_inc(&nr_files
);
123 if (security_file_alloc(f
))
126 INIT_LIST_HEAD(&f
->f_u
.fu_list
);
127 atomic_long_set(&f
->f_count
, 1);
128 rwlock_init(&f
->f_owner
.lock
);
129 f
->f_cred
= get_cred(cred
);
130 spin_lock_init(&f
->f_lock
);
131 eventpoll_init_file(f
);
132 /* f->f_version: 0 */
136 /* Ran out of filps - report that */
137 if (get_nr_files() > old_max
) {
138 printk(KERN_INFO
"VFS: file-max limit %d reached\n",
140 old_max
= get_nr_files();
151 * alloc_file - allocate and initialize a 'struct file'
152 * @mnt: the vfsmount on which the file will reside
153 * @dentry: the dentry representing the new file
154 * @mode: the mode with which the new file will be opened
155 * @fop: the 'struct file_operations' for the new file
157 * Use this instead of get_empty_filp() to get a new
158 * 'struct file'. Do so because of the same initialization
159 * pitfalls reasons listed for init_file(). This is a
160 * preferred interface to using init_file().
162 * If all the callers of init_file() are eliminated, its
163 * code should be moved into this function.
165 struct file
*alloc_file(struct vfsmount
*mnt
, struct dentry
*dentry
,
166 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_clone_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
);
280 if (unlikely(S_ISCHR(inode
->i_mode
) && inode
->i_cdev
!= NULL
))
281 cdev_put(inode
->i_cdev
);
282 fops_put(file
->f_op
);
283 put_pid(file
->f_owner
.pid
);
285 if (file
->f_mode
& FMODE_WRITE
)
286 drop_file_write_access(file
);
287 file
->f_path
.dentry
= NULL
;
288 file
->f_path
.mnt
= NULL
;
294 struct file
*fget(unsigned int fd
)
297 struct files_struct
*files
= current
->files
;
300 file
= fcheck_files(files
, fd
);
302 if (!atomic_long_inc_not_zero(&file
->f_count
)) {
303 /* File object ref couldn't be taken */
316 * Lightweight file lookup - no refcnt increment if fd table isn't shared.
317 * You can use this only if it is guranteed that the current task already
318 * holds a refcnt to that file. That check has to be done at fget() only
319 * and a flag is returned to be passed to the corresponding fput_light().
320 * There must not be a cloning between an fget_light/fput_light pair.
322 struct file
*fget_light(unsigned int fd
, int *fput_needed
)
325 struct files_struct
*files
= current
->files
;
328 if (likely((atomic_read(&files
->count
) == 1))) {
329 file
= fcheck_files(files
, fd
);
332 file
= fcheck_files(files
, fd
);
334 if (atomic_long_inc_not_zero(&file
->f_count
))
337 /* Didn't get the reference, someone's freed */
347 void put_filp(struct file
*file
)
349 if (atomic_long_dec_and_test(&file
->f_count
)) {
350 security_file_free(file
);
356 void file_move(struct file
*file
, struct list_head
*list
)
361 list_move(&file
->f_u
.fu_list
, list
);
365 void file_kill(struct file
*file
)
367 if (!list_empty(&file
->f_u
.fu_list
)) {
369 list_del_init(&file
->f_u
.fu_list
);
374 int fs_may_remount_ro(struct super_block
*sb
)
378 /* Check that no files are currently opened for writing. */
380 list_for_each_entry(file
, &sb
->s_files
, f_u
.fu_list
) {
381 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
383 /* File with pending delete? */
384 if (inode
->i_nlink
== 0)
387 /* Writeable file? */
388 if (S_ISREG(inode
->i_mode
) && (file
->f_mode
& FMODE_WRITE
))
392 return 1; /* Tis' cool bro. */
399 * mark_files_ro - mark all files read-only
400 * @sb: superblock in question
402 * All files are marked read-only. We don't care about pending
403 * delete files so this should be used in 'force' mode only.
405 void mark_files_ro(struct super_block
*sb
)
411 list_for_each_entry(f
, &sb
->s_files
, f_u
.fu_list
) {
412 struct vfsmount
*mnt
;
413 if (!S_ISREG(f
->f_path
.dentry
->d_inode
->i_mode
))
417 if (!(f
->f_mode
& FMODE_WRITE
))
419 f
->f_mode
&= ~FMODE_WRITE
;
420 if (file_check_writeable(f
) != 0)
422 file_release_write(f
);
423 mnt
= mntget(f
->f_path
.mnt
);
426 * This can sleep, so we can't hold
427 * the file_list_lock() spinlock.
436 void __init
files_init(unsigned long mempages
)
440 filp_cachep
= kmem_cache_create("filp", sizeof(struct file
), 0,
441 SLAB_HWCACHE_ALIGN
| SLAB_PANIC
, NULL
);
444 * One file with associated inode and dcache is very roughly 1K.
445 * Per default don't use more than 10% of our memory for files.
448 n
= (mempages
* (PAGE_SIZE
/ 1024)) / 10;
449 files_stat
.max_files
= n
;
450 if (files_stat
.max_files
< NR_FILE
)
451 files_stat
.max_files
= NR_FILE
;
453 percpu_counter_init(&nr_files
, 0);