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 file->f_path.dentry = dentry;
175 file->f_path.mnt = mntget(mnt);
176 file->f_mapping = dentry->d_inode->i_mapping;
181 * These mounts don't really matter in practice
182 * for r/o bind mounts. They aren't userspace-
183 * visible. We do this for consistency, and so
184 * that we can do debugging checks at __fput()
186 if ((mode & FMODE_WRITE) && !special_file(dentry->d_inode->i_mode)) {
188 file_take_write(file);
189 error = mnt_clone_write(mnt);
195 void fput(struct file *file)
197 if (atomic_long_dec_and_test(&file->f_count))
204 * drop_file_write_access - give up ability to write to a file
205 * @file: the file to which we will stop writing
207 * This is a central place which will give up the ability
208 * to write to @file, along with access to write through
211 void drop_file_write_access(struct file *file)
213 struct vfsmount *mnt = file->f_path.mnt;
214 struct dentry *dentry = file->f_path.dentry;
215 struct inode *inode = dentry->d_inode;
217 put_write_access(inode);
219 if (special_file(inode->i_mode))
221 if (file_check_writeable(file) != 0)
224 file_release_write(file);
226 EXPORT_SYMBOL_GPL(drop_file_write_access);
228 /* __fput is called from task context when aio completion releases the last
229 * last use of a struct file *. Do not use otherwise.
231 void __fput(struct file *file)
233 struct dentry *dentry = file->f_path.dentry;
234 struct vfsmount *mnt = file->f_path.mnt;
235 struct inode *inode = dentry->d_inode;
239 fsnotify_close(file);
241 * The function eventpoll_release() should be the first called
242 * in the file cleanup chain.
244 eventpoll_release(file);
245 locks_remove_flock(file);
247 if (unlikely(file->f_flags & FASYNC)) {
248 if (file->f_op && file->f_op->fasync)
249 file->f_op->fasync(-1, file, 0);
251 if (file->f_op && file->f_op->release)
252 file->f_op->release(inode, file);
253 security_file_free(file);
254 if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL))
255 cdev_put(inode->i_cdev);
256 fops_put(file->f_op);
257 put_pid(file->f_owner.pid);
259 if (file->f_mode & FMODE_WRITE)
260 drop_file_write_access(file);
261 file->f_path.dentry = NULL;
262 file->f_path.mnt = NULL;
268 struct file *fget(unsigned int fd)
271 struct files_struct *files = current->files;
274 file = fcheck_files(files, fd);
276 if (!atomic_long_inc_not_zero(&file->f_count)) {
277 /* File object ref couldn't be taken */
290 * Lightweight file lookup - no refcnt increment if fd table isn't shared.
291 * You can use this only if it is guranteed that the current task already
292 * holds a refcnt to that file. That check has to be done at fget() only
293 * and a flag is returned to be passed to the corresponding fput_light().
294 * There must not be a cloning between an fget_light/fput_light pair.
296 struct file *fget_light(unsigned int fd, int *fput_needed)
299 struct files_struct *files = current->files;
302 if (likely((atomic_read(&files->count) == 1))) {
303 file = fcheck_files(files, fd);
306 file = fcheck_files(files, fd);
308 if (atomic_long_inc_not_zero(&file->f_count))
311 /* Didn't get the reference, someone's freed */
321 void put_filp(struct file *file)
323 if (atomic_long_dec_and_test(&file->f_count)) {
324 security_file_free(file);
330 void file_move(struct file *file, struct list_head *list)
335 list_move(&file->f_u.fu_list, list);
339 void file_kill(struct file *file)
341 if (!list_empty(&file->f_u.fu_list)) {
343 list_del_init(&file->f_u.fu_list);
348 int fs_may_remount_ro(struct super_block *sb)
352 /* Check that no files are currently opened for writing. */
354 list_for_each_entry(file, &sb->s_files, f_u.fu_list) {
355 struct inode *inode = file->f_path.dentry->d_inode;
357 /* File with pending delete? */
358 if (inode->i_nlink == 0)
361 /* Writeable file? */
362 if (S_ISREG(inode->i_mode) && (file->f_mode & FMODE_WRITE))
366 return 1; /* Tis' cool bro. */
373 * mark_files_ro - mark all files read-only
374 * @sb: superblock in question
376 * All files are marked read-only. We don't care about pending
377 * delete files so this should be used in 'force' mode only.
379 void mark_files_ro(struct super_block *sb)
385 list_for_each_entry(f, &sb->s_files, f_u.fu_list) {
386 struct vfsmount *mnt;
387 if (!S_ISREG(f->f_path.dentry->d_inode->i_mode))
391 if (!(f->f_mode & FMODE_WRITE))
393 f->f_mode &= ~FMODE_WRITE;
394 if (file_check_writeable(f) != 0)
396 file_release_write(f);
397 mnt = mntget(f->f_path.mnt);
400 * This can sleep, so we can't hold
401 * the file_list_lock() spinlock.
410 void __init files_init(unsigned long mempages)
414 filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
415 SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
418 * One file with associated inode and dcache is very roughly 1K.
419 * Per default don't use more than 10% of our memory for files.
422 n = (mempages * (PAGE_SIZE / 1024)) / 10;
423 files_stat.max_files = n;
424 if (files_stat.max_files < NR_FILE)
425 files_stat.max_files = NR_FILE;
427 percpu_counter_init(&nr_files, 0);