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/init.h>
12 #include <linux/module.h>
13 #include <linux/smp_lock.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);
40 kmem_cache_free(filp_cachep, f);
43 static inline void file_free(struct file *f)
45 percpu_counter_dec(&nr_files);
46 call_rcu(&f->f_u.fu_rcuhead, file_free_rcu);
50 * Return the total number of open files in the system
52 static int get_nr_files(void)
54 return percpu_counter_read_positive(&nr_files);
58 * Return the maximum number of open files in the system
60 int get_max_files(void)
62 return files_stat.max_files;
64 EXPORT_SYMBOL_GPL(get_max_files);
67 * Handle nr_files sysctl
69 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
70 int proc_nr_files(ctl_table *table, int write, struct file *filp,
71 void __user *buffer, size_t *lenp, loff_t *ppos)
73 files_stat.nr_files = get_nr_files();
74 return proc_dointvec(table, write, filp, buffer, lenp, ppos);
77 int proc_nr_files(ctl_table *table, int write, struct file *filp,
78 void __user *buffer, size_t *lenp, loff_t *ppos)
84 /* Find an unused file structure and return a pointer to it.
85 * Returns NULL, if there are no more free file structures or
86 * we run out of memory.
88 struct file *get_empty_filp(void)
90 struct task_struct *tsk;
95 * Privileged users can go above max_files
97 if (get_nr_files() >= files_stat.max_files && !capable(CAP_SYS_ADMIN)) {
99 * percpu_counters are inaccurate. Do an expensive check before
102 if (percpu_counter_sum(&nr_files) >= files_stat.max_files)
106 f = kmem_cache_alloc(filp_cachep, GFP_KERNEL);
110 percpu_counter_inc(&nr_files);
111 memset(f, 0, sizeof(*f));
112 if (security_file_alloc(f))
116 INIT_LIST_HEAD(&f->f_u.fu_list);
117 atomic_set(&f->f_count, 1);
118 rwlock_init(&f->f_owner.lock);
119 f->f_uid = tsk->fsuid;
120 f->f_gid = tsk->fsgid;
121 eventpoll_init_file(f);
122 /* f->f_version: 0 */
126 /* Ran out of filps - report that */
127 if (get_nr_files() > old_max) {
128 printk(KERN_INFO "VFS: file-max limit %d reached\n",
130 old_max = get_nr_files();
140 EXPORT_SYMBOL(get_empty_filp);
142 void fastcall fput(struct file *file)
144 if (atomic_dec_and_test(&file->f_count))
150 /* __fput is called from task context when aio completion releases the last
151 * last use of a struct file *. Do not use otherwise.
153 void fastcall __fput(struct file *file)
155 struct dentry *dentry = file->f_dentry;
156 struct vfsmount *mnt = file->f_vfsmnt;
157 struct inode *inode = dentry->d_inode;
161 fsnotify_close(file);
163 * The function eventpoll_release() should be the first called
164 * in the file cleanup chain.
166 eventpoll_release(file);
167 locks_remove_flock(file);
169 if (file->f_op && file->f_op->release)
170 file->f_op->release(inode, file);
171 security_file_free(file);
172 if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL))
173 cdev_put(inode->i_cdev);
174 fops_put(file->f_op);
175 if (file->f_mode & FMODE_WRITE)
176 put_write_access(inode);
177 put_pid(file->f_owner.pid);
179 file->f_dentry = NULL;
180 file->f_vfsmnt = NULL;
186 struct file fastcall *fget(unsigned int fd)
189 struct files_struct *files = current->files;
192 file = fcheck_files(files, fd);
194 if (!atomic_inc_not_zero(&file->f_count)) {
195 /* File object ref couldn't be taken */
208 * Lightweight file lookup - no refcnt increment if fd table isn't shared.
209 * You can use this only if it is guranteed that the current task already
210 * holds a refcnt to that file. That check has to be done at fget() only
211 * and a flag is returned to be passed to the corresponding fput_light().
212 * There must not be a cloning between an fget_light/fput_light pair.
214 struct file fastcall *fget_light(unsigned int fd, int *fput_needed)
217 struct files_struct *files = current->files;
220 if (likely((atomic_read(&files->count) == 1))) {
221 file = fcheck_files(files, fd);
224 file = fcheck_files(files, fd);
226 if (atomic_inc_not_zero(&file->f_count))
229 /* Didn't get the reference, someone's freed */
239 void put_filp(struct file *file)
241 if (atomic_dec_and_test(&file->f_count)) {
242 security_file_free(file);
248 void file_move(struct file *file, struct list_head *list)
253 list_move(&file->f_u.fu_list, list);
257 void file_kill(struct file *file)
259 if (!list_empty(&file->f_u.fu_list)) {
261 list_del_init(&file->f_u.fu_list);
266 int fs_may_remount_ro(struct super_block *sb)
270 /* Check that no files are currently opened for writing. */
272 list_for_each(p, &sb->s_files) {
273 struct file *file = list_entry(p, struct file, f_u.fu_list);
274 struct inode *inode = file->f_dentry->d_inode;
276 /* File with pending delete? */
277 if (inode->i_nlink == 0)
280 /* Writeable file? */
281 if (S_ISREG(inode->i_mode) && (file->f_mode & FMODE_WRITE))
285 return 1; /* Tis' cool bro. */
291 void __init files_init(unsigned long mempages)
294 /* One file with associated inode and dcache is very roughly 1K.
295 * Per default don't use more than 10% of our memory for files.
298 n = (mempages * (PAGE_SIZE / 1024)) / 10;
299 files_stat.max_files = n;
300 if (files_stat.max_files < NR_FILE)
301 files_stat.max_files = NR_FILE;
303 percpu_counter_init(&nr_files, 0);