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 static struct percpu_counter nr_files __cacheline_aligned_in_smp;
37 static inline void file_free_rcu(struct rcu_head *head)
39 struct file *f = container_of(head, struct file, f_u.fu_rcuhead);
42 kmem_cache_free(filp_cachep, f);
45 static inline void file_free(struct file *f)
47 percpu_counter_dec(&nr_files);
49 call_rcu(&f->f_u.fu_rcuhead, file_free_rcu);
53 * Return the total number of open files in the system
55 static int get_nr_files(void)
57 return percpu_counter_read_positive(&nr_files);
61 * Return the maximum number of open files in the system
63 int get_max_files(void)
65 return files_stat.max_files;
67 EXPORT_SYMBOL_GPL(get_max_files);
70 * Handle nr_files sysctl
72 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
73 int proc_nr_files(ctl_table *table, int write, struct file *filp,
74 void __user *buffer, size_t *lenp, loff_t *ppos)
76 files_stat.nr_files = get_nr_files();
77 return proc_dointvec(table, write, filp, buffer, lenp, ppos);
80 int proc_nr_files(ctl_table *table, int write, struct file *filp,
81 void __user *buffer, size_t *lenp, loff_t *ppos)
87 /* Find an unused file structure and return a pointer to it.
88 * Returns NULL, if there are no more free file structures or
89 * we run out of memory.
91 * Be very careful using this. You are responsible for
92 * getting write access to any mount that you might assign
93 * to this filp, if it is opened for write. If this is not
94 * done, you will imbalance int the mount's writer count
95 * and a warning at __fput() time.
97 struct file *get_empty_filp(void)
99 const struct cred *cred = current_cred();
104 * Privileged users can go above max_files
106 if (get_nr_files() >= files_stat.max_files && !capable(CAP_SYS_ADMIN)) {
108 * percpu_counters are inaccurate. Do an expensive check before
111 if (percpu_counter_sum_positive(&nr_files) >= files_stat.max_files)
115 f = kmem_cache_zalloc(filp_cachep, GFP_KERNEL);
119 percpu_counter_inc(&nr_files);
120 if (security_file_alloc(f))
123 INIT_LIST_HEAD(&f->f_u.fu_list);
124 atomic_long_set(&f->f_count, 1);
125 rwlock_init(&f->f_owner.lock);
126 f->f_cred = get_cred(cred);
127 eventpoll_init_file(f);
128 /* f->f_version: 0 */
132 /* Ran out of filps - report that */
133 if (get_nr_files() > old_max) {
134 printk(KERN_INFO "VFS: file-max limit %d reached\n",
136 old_max = get_nr_files();
146 EXPORT_SYMBOL(get_empty_filp);
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
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().
160 * If all the callers of init_file() are eliminated, its
161 * code should be moved into this function.
163 struct file *alloc_file(struct vfsmount *mnt, struct dentry *dentry,
164 fmode_t mode, const struct file_operations *fop)
169 file = get_empty_filp();
173 init_file(file, mnt, dentry, mode, fop);
176 EXPORT_SYMBOL(alloc_file);
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
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.
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().
195 int init_file(struct file *file, struct vfsmount *mnt, struct dentry *dentry,
196 fmode_t mode, const struct file_operations *fop)
199 file->f_path.dentry = dentry;
200 file->f_path.mnt = mntget(mnt);
201 file->f_mapping = dentry->d_inode->i_mapping;
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()
211 if ((mode & FMODE_WRITE) && !special_file(dentry->d_inode->i_mode)) {
212 file_take_write(file);
213 error = mnt_want_write(mnt);
218 EXPORT_SYMBOL(init_file);
220 void fput(struct file *file)
222 if (atomic_long_dec_and_test(&file->f_count))
229 * drop_file_write_access - give up ability to write to a file
230 * @file: the file to which we will stop writing
232 * This is a central place which will give up the ability
233 * to write to @file, along with access to write through
236 void drop_file_write_access(struct file *file)
238 struct vfsmount *mnt = file->f_path.mnt;
239 struct dentry *dentry = file->f_path.dentry;
240 struct inode *inode = dentry->d_inode;
242 put_write_access(inode);
244 if (special_file(inode->i_mode))
246 if (file_check_writeable(file) != 0)
249 file_release_write(file);
251 EXPORT_SYMBOL_GPL(drop_file_write_access);
253 /* __fput is called from task context when aio completion releases the last
254 * last use of a struct file *. Do not use otherwise.
256 void __fput(struct file *file)
258 struct dentry *dentry = file->f_path.dentry;
259 struct vfsmount *mnt = file->f_path.mnt;
260 struct inode *inode = dentry->d_inode;
264 fsnotify_close(file);
266 * The function eventpoll_release() should be the first called
267 * in the file cleanup chain.
269 eventpoll_release(file);
270 locks_remove_flock(file);
272 if (unlikely(file->f_flags & FASYNC)) {
273 if (file->f_op && file->f_op->fasync)
274 file->f_op->fasync(-1, file, 0);
276 if (file->f_op && file->f_op->release)
277 file->f_op->release(inode, file);
278 security_file_free(file);
279 if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL))
280 cdev_put(inode->i_cdev);
281 fops_put(file->f_op);
282 put_pid(file->f_owner.pid);
284 if (file->f_mode & FMODE_WRITE)
285 drop_file_write_access(file);
286 file->f_path.dentry = NULL;
287 file->f_path.mnt = NULL;
293 struct file *fget(unsigned int fd)
296 struct files_struct *files = current->files;
299 file = fcheck_files(files, fd);
301 if (!atomic_long_inc_not_zero(&file->f_count)) {
302 /* File object ref couldn't be taken */
315 * Lightweight file lookup - no refcnt increment if fd table isn't shared.
316 * You can use this only if it is guranteed that the current task already
317 * holds a refcnt to that file. That check has to be done at fget() only
318 * and a flag is returned to be passed to the corresponding fput_light().
319 * There must not be a cloning between an fget_light/fput_light pair.
321 struct file *fget_light(unsigned int fd, int *fput_needed)
324 struct files_struct *files = current->files;
327 if (likely((atomic_read(&files->count) == 1))) {
328 file = fcheck_files(files, fd);
331 file = fcheck_files(files, fd);
333 if (atomic_long_inc_not_zero(&file->f_count))
336 /* Didn't get the reference, someone's freed */
346 void put_filp(struct file *file)
348 if (atomic_long_dec_and_test(&file->f_count)) {
349 security_file_free(file);
355 void file_move(struct file *file, struct list_head *list)
360 list_move(&file->f_u.fu_list, list);
364 void file_kill(struct file *file)
366 if (!list_empty(&file->f_u.fu_list)) {
368 list_del_init(&file->f_u.fu_list);
373 int fs_may_remount_ro(struct super_block *sb)
377 /* Check that no files are currently opened for writing. */
379 list_for_each_entry(file, &sb->s_files, f_u.fu_list) {
380 struct inode *inode = file->f_path.dentry->d_inode;
382 /* File with pending delete? */
383 if (inode->i_nlink == 0)
386 /* Writeable file? */
387 if (S_ISREG(inode->i_mode) && (file->f_mode & FMODE_WRITE))
391 return 1; /* Tis' cool bro. */
397 void __init files_init(unsigned long mempages)
400 /* One file with associated inode and dcache is very roughly 1K.
401 * Per default don't use more than 10% of our memory for files.
404 n = (mempages * (PAGE_SIZE / 1024)) / 10;
405 files_stat.max_files = n;
406 if (files_stat.max_files < NR_FILE)
407 files_stat.max_files = NR_FILE;
409 percpu_counter_init(&nr_files, 0);