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 path
*path
, fmode_t mode
,
166 const struct file_operations
*fop
)
170 file
= get_empty_filp();
174 file
->f_path
= *path
;
175 file
->f_mapping
= path
->dentry
->d_inode
->i_mapping
;
180 * These mounts don't really matter in practice
181 * for r/o bind mounts. They aren't userspace-
182 * visible. We do this for consistency, and so
183 * that we can do debugging checks at __fput()
185 if ((mode
& FMODE_WRITE
) && !special_file(path
->dentry
->d_inode
->i_mode
)) {
187 file_take_write(file
);
188 error
= mnt_clone_write(path
->mnt
);
194 void fput(struct file
*file
)
196 if (atomic_long_dec_and_test(&file
->f_count
))
203 * drop_file_write_access - give up ability to write to a file
204 * @file: the file to which we will stop writing
206 * This is a central place which will give up the ability
207 * to write to @file, along with access to write through
210 void drop_file_write_access(struct file
*file
)
212 struct vfsmount
*mnt
= file
->f_path
.mnt
;
213 struct dentry
*dentry
= file
->f_path
.dentry
;
214 struct inode
*inode
= dentry
->d_inode
;
216 put_write_access(inode
);
218 if (special_file(inode
->i_mode
))
220 if (file_check_writeable(file
) != 0)
223 file_release_write(file
);
225 EXPORT_SYMBOL_GPL(drop_file_write_access
);
227 /* __fput is called from task context when aio completion releases the last
228 * last use of a struct file *. Do not use otherwise.
230 void __fput(struct file
*file
)
232 struct dentry
*dentry
= file
->f_path
.dentry
;
233 struct vfsmount
*mnt
= file
->f_path
.mnt
;
234 struct inode
*inode
= dentry
->d_inode
;
238 fsnotify_close(file
);
240 * The function eventpoll_release() should be the first called
241 * in the file cleanup chain.
243 eventpoll_release(file
);
244 locks_remove_flock(file
);
246 if (unlikely(file
->f_flags
& FASYNC
)) {
247 if (file
->f_op
&& file
->f_op
->fasync
)
248 file
->f_op
->fasync(-1, file
, 0);
250 if (file
->f_op
&& file
->f_op
->release
)
251 file
->f_op
->release(inode
, file
);
252 security_file_free(file
);
253 if (unlikely(S_ISCHR(inode
->i_mode
) && inode
->i_cdev
!= NULL
))
254 cdev_put(inode
->i_cdev
);
255 fops_put(file
->f_op
);
256 put_pid(file
->f_owner
.pid
);
258 if (file
->f_mode
& FMODE_WRITE
)
259 drop_file_write_access(file
);
260 file
->f_path
.dentry
= NULL
;
261 file
->f_path
.mnt
= NULL
;
267 struct file
*fget(unsigned int fd
)
270 struct files_struct
*files
= current
->files
;
273 file
= fcheck_files(files
, fd
);
275 if (!atomic_long_inc_not_zero(&file
->f_count
)) {
276 /* File object ref couldn't be taken */
289 * Lightweight file lookup - no refcnt increment if fd table isn't shared.
290 * You can use this only if it is guranteed that the current task already
291 * holds a refcnt to that file. That check has to be done at fget() only
292 * and a flag is returned to be passed to the corresponding fput_light().
293 * There must not be a cloning between an fget_light/fput_light pair.
295 struct file
*fget_light(unsigned int fd
, int *fput_needed
)
298 struct files_struct
*files
= current
->files
;
301 if (likely((atomic_read(&files
->count
) == 1))) {
302 file
= fcheck_files(files
, fd
);
305 file
= fcheck_files(files
, fd
);
307 if (atomic_long_inc_not_zero(&file
->f_count
))
310 /* Didn't get the reference, someone's freed */
320 void put_filp(struct file
*file
)
322 if (atomic_long_dec_and_test(&file
->f_count
)) {
323 security_file_free(file
);
329 void file_move(struct file
*file
, struct list_head
*list
)
334 list_move(&file
->f_u
.fu_list
, list
);
338 void file_kill(struct file
*file
)
340 if (!list_empty(&file
->f_u
.fu_list
)) {
342 list_del_init(&file
->f_u
.fu_list
);
347 int fs_may_remount_ro(struct super_block
*sb
)
351 /* Check that no files are currently opened for writing. */
353 list_for_each_entry(file
, &sb
->s_files
, f_u
.fu_list
) {
354 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
356 /* File with pending delete? */
357 if (inode
->i_nlink
== 0)
360 /* Writeable file? */
361 if (S_ISREG(inode
->i_mode
) && (file
->f_mode
& FMODE_WRITE
))
365 return 1; /* Tis' cool bro. */
372 * mark_files_ro - mark all files read-only
373 * @sb: superblock in question
375 * All files are marked read-only. We don't care about pending
376 * delete files so this should be used in 'force' mode only.
378 void mark_files_ro(struct super_block
*sb
)
384 list_for_each_entry(f
, &sb
->s_files
, f_u
.fu_list
) {
385 struct vfsmount
*mnt
;
386 if (!S_ISREG(f
->f_path
.dentry
->d_inode
->i_mode
))
390 if (!(f
->f_mode
& FMODE_WRITE
))
392 f
->f_mode
&= ~FMODE_WRITE
;
393 if (file_check_writeable(f
) != 0)
395 file_release_write(f
);
396 mnt
= mntget(f
->f_path
.mnt
);
399 * This can sleep, so we can't hold
400 * the file_list_lock() spinlock.
409 void __init
files_init(unsigned long mempages
)
413 filp_cachep
= kmem_cache_create("filp", sizeof(struct file
), 0,
414 SLAB_HWCACHE_ALIGN
| SLAB_PANIC
, NULL
);
417 * One file with associated inode and dcache is very roughly 1K.
418 * Per default don't use more than 10% of our memory for files.
421 n
= (mempages
* (PAGE_SIZE
/ 1024)) / 10;
422 files_stat
.max_files
= n
;
423 if (files_stat
.max_files
< NR_FILE
)
424 files_stat
.max_files
= NR_FILE
;
426 percpu_counter_init(&nr_files
, 0);