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>
14 #include <linux/security.h>
15 #include <linux/eventpoll.h>
16 #include <linux/rcupdate.h>
17 #include <linux/mount.h>
18 #include <linux/capability.h>
19 #include <linux/cdev.h>
20 #include <linux/fsnotify.h>
21 #include <linux/sysctl.h>
22 #include <linux/percpu_counter.h>
24 #include <asm/atomic.h>
26 /* sysctl tunables... */
27 struct files_stat_struct files_stat = {
31 /* public. Not pretty! */
32 __cacheline_aligned_in_smp DEFINE_SPINLOCK(files_lock);
34 static struct percpu_counter nr_files __cacheline_aligned_in_smp;
36 static inline void file_free_rcu(struct rcu_head *head)
38 struct file *f = container_of(head, struct file, f_u.fu_rcuhead);
39 kmem_cache_free(filp_cachep, f);
42 static inline void file_free(struct file *f)
44 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 * Be very careful using this. You are responsible for
89 * getting write access to any mount that you might assign
90 * to this filp, if it is opened for write. If this is not
91 * done, you will imbalance int the mount's writer count
92 * and a warning at __fput() time.
94 struct file *get_empty_filp(void)
96 struct task_struct *tsk;
101 * Privileged users can go above max_files
103 if (get_nr_files() >= files_stat.max_files && !capable(CAP_SYS_ADMIN)) {
105 * percpu_counters are inaccurate. Do an expensive check before
108 if (percpu_counter_sum_positive(&nr_files) >= files_stat.max_files)
112 f = kmem_cache_zalloc(filp_cachep, GFP_KERNEL);
116 percpu_counter_inc(&nr_files);
117 if (security_file_alloc(f))
121 INIT_LIST_HEAD(&f->f_u.fu_list);
122 atomic_set(&f->f_count, 1);
123 rwlock_init(&f->f_owner.lock);
124 f->f_uid = tsk->fsuid;
125 f->f_gid = tsk->fsgid;
126 eventpoll_init_file(f);
127 /* f->f_version: 0 */
131 /* Ran out of filps - report that */
132 if (get_nr_files() > old_max) {
133 printk(KERN_INFO "VFS: file-max limit %d reached\n",
135 old_max = get_nr_files();
145 EXPORT_SYMBOL(get_empty_filp);
148 * alloc_file - allocate and initialize a 'struct file'
149 * @mnt: the vfsmount on which the file will reside
150 * @dentry: the dentry representing the new file
151 * @mode: the mode with which the new file will be opened
152 * @fop: the 'struct file_operations' for the new file
154 * Use this instead of get_empty_filp() to get a new
155 * 'struct file'. Do so because of the same initialization
156 * pitfalls reasons listed for init_file(). This is a
157 * preferred interface to using init_file().
159 * If all the callers of init_file() are eliminated, its
160 * code should be moved into this function.
162 struct file *alloc_file(struct vfsmount *mnt, struct dentry *dentry,
163 mode_t mode, const struct file_operations *fop)
168 file = get_empty_filp();
172 init_file(file, mnt, dentry, mode, fop);
175 EXPORT_SYMBOL(alloc_file);
178 * init_file - initialize a 'struct file'
179 * @file: the already allocated 'struct file' to initialized
180 * @mnt: the vfsmount on which the file resides
181 * @dentry: the dentry representing this file
182 * @mode: the mode the file is opened with
183 * @fop: the 'struct file_operations' for this file
185 * Use this instead of setting the members directly. Doing so
186 * avoids making mistakes like forgetting the mntget() or
187 * forgetting to take a write on the mnt.
189 * Note: This is a crappy interface. It is here to make
190 * merging with the existing users of get_empty_filp()
191 * who have complex failure logic easier. All users
192 * of this should be moving to alloc_file().
194 int init_file(struct file *file, struct vfsmount *mnt, struct dentry *dentry,
195 mode_t mode, const struct file_operations *fop)
198 file->f_path.dentry = dentry;
199 file->f_path.mnt = mntget(mnt);
200 file->f_mapping = dentry->d_inode->i_mapping;
205 * These mounts don't really matter in practice
206 * for r/o bind mounts. They aren't userspace-
207 * visible. We do this for consistency, and so
208 * that we can do debugging checks at __fput()
210 if ((mode & FMODE_WRITE) && !special_file(dentry->d_inode->i_mode)) {
211 file_take_write(file);
212 error = mnt_want_write(mnt);
217 EXPORT_SYMBOL(init_file);
219 void fput(struct file *file)
221 if (atomic_dec_and_test(&file->f_count))
228 * drop_file_write_access - give up ability to write to a file
229 * @file: the file to which we will stop writing
231 * This is a central place which will give up the ability
232 * to write to @file, along with access to write through
235 void drop_file_write_access(struct file *file)
237 struct vfsmount *mnt = file->f_path.mnt;
238 struct dentry *dentry = file->f_path.dentry;
239 struct inode *inode = dentry->d_inode;
241 put_write_access(inode);
243 if (special_file(inode->i_mode))
245 if (file_check_writeable(file) != 0)
248 file_release_write(file);
250 EXPORT_SYMBOL_GPL(drop_file_write_access);
252 /* __fput is called from task context when aio completion releases the last
253 * last use of a struct file *. Do not use otherwise.
255 void __fput(struct file *file)
257 struct dentry *dentry = file->f_path.dentry;
258 struct vfsmount *mnt = file->f_path.mnt;
259 struct inode *inode = dentry->d_inode;
263 fsnotify_close(file);
265 * The function eventpoll_release() should be the first called
266 * in the file cleanup chain.
268 eventpoll_release(file);
269 locks_remove_flock(file);
271 if (file->f_op && file->f_op->release)
272 file->f_op->release(inode, file);
273 security_file_free(file);
274 if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL))
275 cdev_put(inode->i_cdev);
276 fops_put(file->f_op);
277 put_pid(file->f_owner.pid);
279 if (file->f_mode & FMODE_WRITE)
280 drop_file_write_access(file);
281 file->f_path.dentry = NULL;
282 file->f_path.mnt = NULL;
288 struct file *fget(unsigned int fd)
291 struct files_struct *files = current->files;
294 file = fcheck_files(files, fd);
296 if (!atomic_inc_not_zero(&file->f_count)) {
297 /* File object ref couldn't be taken */
310 * Lightweight file lookup - no refcnt increment if fd table isn't shared.
311 * You can use this only if it is guranteed that the current task already
312 * holds a refcnt to that file. That check has to be done at fget() only
313 * and a flag is returned to be passed to the corresponding fput_light().
314 * There must not be a cloning between an fget_light/fput_light pair.
316 struct file *fget_light(unsigned int fd, int *fput_needed)
319 struct files_struct *files = current->files;
322 if (likely((atomic_read(&files->count) == 1))) {
323 file = fcheck_files(files, fd);
326 file = fcheck_files(files, fd);
328 if (atomic_inc_not_zero(&file->f_count))
331 /* Didn't get the reference, someone's freed */
341 void put_filp(struct file *file)
343 if (atomic_dec_and_test(&file->f_count)) {
344 security_file_free(file);
350 void file_move(struct file *file, struct list_head *list)
355 list_move(&file->f_u.fu_list, list);
359 void file_kill(struct file *file)
361 if (!list_empty(&file->f_u.fu_list)) {
363 list_del_init(&file->f_u.fu_list);
368 int fs_may_remount_ro(struct super_block *sb)
372 /* Check that no files are currently opened for writing. */
374 list_for_each_entry(file, &sb->s_files, f_u.fu_list) {
375 struct inode *inode = file->f_path.dentry->d_inode;
377 /* File with pending delete? */
378 if (inode->i_nlink == 0)
381 /* Writeable file? */
382 if (S_ISREG(inode->i_mode) && (file->f_mode & FMODE_WRITE))
386 return 1; /* Tis' cool bro. */
392 void __init files_init(unsigned long mempages)
395 /* One file with associated inode and dcache is very roughly 1K.
396 * Per default don't use more than 10% of our memory for files.
399 n = (mempages * (PAGE_SIZE / 1024)) / 10;
400 files_stat.max_files = n;
401 if (files_stat.max_files < NR_FILE)
402 files_stat.max_files = NR_FILE;
404 percpu_counter_init(&nr_files, 0);