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>
24 #include <linux/ima.h>
26 #include <asm/atomic.h>
30 /* sysctl tunables... */
31 struct files_stat_struct files_stat = {
35 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(files_lock);
37 /* SLAB cache for file structures */
38 static struct kmem_cache *filp_cachep __read_mostly;
40 static struct percpu_counter nr_files __cacheline_aligned_in_smp;
42 static inline void file_free_rcu(struct rcu_head *head)
44 struct file *f = container_of(head, struct file, f_u.fu_rcuhead);
47 kmem_cache_free(filp_cachep, f);
50 static inline void file_free(struct file *f)
52 percpu_counter_dec(&nr_files);
54 call_rcu(&f->f_u.fu_rcuhead, file_free_rcu);
58 * Return the total number of open files in the system
60 static int get_nr_files(void)
62 return percpu_counter_read_positive(&nr_files);
66 * Return the maximum number of open files in the system
68 int get_max_files(void)
70 return files_stat.max_files;
72 EXPORT_SYMBOL_GPL(get_max_files);
75 * Handle nr_files sysctl
77 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
78 int proc_nr_files(ctl_table *table, int write,
79 void __user *buffer, size_t *lenp, loff_t *ppos)
81 files_stat.nr_files = get_nr_files();
82 return proc_dointvec(table, write, buffer, lenp, ppos);
85 int proc_nr_files(ctl_table *table, int write,
86 void __user *buffer, size_t *lenp, loff_t *ppos)
92 /* Find an unused file structure and return a pointer to it.
93 * Returns NULL, if there are no more free file structures or
94 * we run out of memory.
96 * Be very careful using this. You are responsible for
97 * getting write access to any mount that you might assign
98 * to this filp, if it is opened for write. If this is not
99 * done, you will imbalance int the mount's writer count
100 * and a warning at __fput() time.
102 struct file *get_empty_filp(void)
104 const struct cred *cred = current_cred();
109 * Privileged users can go above max_files
111 if (get_nr_files() >= files_stat.max_files && !capable(CAP_SYS_ADMIN)) {
113 * percpu_counters are inaccurate. Do an expensive check before
116 if (percpu_counter_sum_positive(&nr_files) >= files_stat.max_files)
120 f = kmem_cache_zalloc(filp_cachep, GFP_KERNEL);
124 percpu_counter_inc(&nr_files);
125 if (security_file_alloc(f))
128 INIT_LIST_HEAD(&f->f_u.fu_list);
129 atomic_long_set(&f->f_count, 1);
130 rwlock_init(&f->f_owner.lock);
131 f->f_cred = get_cred(cred);
132 spin_lock_init(&f->f_lock);
133 eventpoll_init_file(f);
134 /* f->f_version: 0 */
138 /* Ran out of filps - report that */
139 if (get_nr_files() > old_max) {
140 printk(KERN_INFO "VFS: file-max limit %d reached\n",
142 old_max = get_nr_files();
153 * alloc_file - allocate and initialize a 'struct file'
154 * @mnt: the vfsmount on which the file will reside
155 * @dentry: the dentry representing the new file
156 * @mode: the mode with which the new file will be opened
157 * @fop: the 'struct file_operations' for the new file
159 * Use this instead of get_empty_filp() to get a new
160 * 'struct file'. Do so because of the same initialization
161 * pitfalls reasons listed for init_file(). This is a
162 * preferred interface to using init_file().
164 * If all the callers of init_file() are eliminated, its
165 * code should be moved into this function.
167 struct file *alloc_file(struct path *path, fmode_t mode,
168 const struct file_operations *fop)
172 file = get_empty_filp();
176 file->f_path = *path;
177 file->f_mapping = path->dentry->d_inode->i_mapping;
182 * These mounts don't really matter in practice
183 * for r/o bind mounts. They aren't userspace-
184 * visible. We do this for consistency, and so
185 * that we can do debugging checks at __fput()
187 if ((mode & FMODE_WRITE) && !special_file(path->dentry->d_inode->i_mode)) {
188 file_take_write(file);
189 WARN_ON(mnt_clone_write(path->mnt));
191 ima_counts_get(file);
194 EXPORT_SYMBOL(alloc_file);
197 * drop_file_write_access - give up ability to write to a file
198 * @file: the file to which we will stop writing
200 * This is a central place which will give up the ability
201 * to write to @file, along with access to write through
204 void drop_file_write_access(struct file *file)
206 struct vfsmount *mnt = file->f_path.mnt;
207 struct dentry *dentry = file->f_path.dentry;
208 struct inode *inode = dentry->d_inode;
210 put_write_access(inode);
212 if (special_file(inode->i_mode))
214 if (file_check_writeable(file) != 0)
217 file_release_write(file);
219 EXPORT_SYMBOL_GPL(drop_file_write_access);
221 /* the real guts of fput() - releasing the last reference to file
223 static void __fput(struct file *file)
225 struct dentry *dentry = file->f_path.dentry;
226 struct vfsmount *mnt = file->f_path.mnt;
227 struct inode *inode = dentry->d_inode;
231 fsnotify_close(file);
233 * The function eventpoll_release() should be the first called
234 * in the file cleanup chain.
236 eventpoll_release(file);
237 locks_remove_flock(file);
239 if (unlikely(file->f_flags & FASYNC)) {
240 if (file->f_op && file->f_op->fasync)
241 file->f_op->fasync(-1, file, 0);
243 if (file->f_op && file->f_op->release)
244 file->f_op->release(inode, file);
245 security_file_free(file);
247 if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL))
248 cdev_put(inode->i_cdev);
249 fops_put(file->f_op);
250 put_pid(file->f_owner.pid);
251 file_sb_list_del(file);
252 if (file->f_mode & FMODE_WRITE)
253 drop_file_write_access(file);
254 file->f_path.dentry = NULL;
255 file->f_path.mnt = NULL;
261 void fput(struct file *file)
263 if (atomic_long_dec_and_test(&file->f_count))
269 struct file *fget(unsigned int fd)
272 struct files_struct *files = current->files;
275 file = fcheck_files(files, fd);
277 if (!atomic_long_inc_not_zero(&file->f_count)) {
278 /* File object ref couldn't be taken */
291 * Lightweight file lookup - no refcnt increment if fd table isn't shared.
293 * You can use this instead of fget if you satisfy all of the following
295 * 1) You must call fput_light before exiting the syscall and returning control
296 * to userspace (i.e. you cannot remember the returned struct file * after
297 * returning to userspace).
298 * 2) You must not call filp_close on the returned struct file * in between
299 * calls to fget_light and fput_light.
300 * 3) You must not clone the current task in between the calls to fget_light
303 * The fput_needed flag returned by fget_light should be passed to the
304 * corresponding fput_light.
306 struct file *fget_light(unsigned int fd, int *fput_needed)
309 struct files_struct *files = current->files;
312 if (likely((atomic_read(&files->count) == 1))) {
313 file = fcheck_files(files, fd);
316 file = fcheck_files(files, fd);
318 if (atomic_long_inc_not_zero(&file->f_count))
321 /* Didn't get the reference, someone's freed */
330 void put_filp(struct file *file)
332 if (atomic_long_dec_and_test(&file->f_count)) {
333 security_file_free(file);
334 file_sb_list_del(file);
339 void file_sb_list_add(struct file *file, struct super_block *sb)
341 spin_lock(&files_lock);
342 BUG_ON(!list_empty(&file->f_u.fu_list));
343 list_add(&file->f_u.fu_list, &sb->s_files);
344 spin_unlock(&files_lock);
347 void file_sb_list_del(struct file *file)
349 if (!list_empty(&file->f_u.fu_list)) {
350 spin_lock(&files_lock);
351 list_del_init(&file->f_u.fu_list);
352 spin_unlock(&files_lock);
356 int fs_may_remount_ro(struct super_block *sb)
360 /* Check that no files are currently opened for writing. */
361 spin_lock(&files_lock);
362 list_for_each_entry(file, &sb->s_files, f_u.fu_list) {
363 struct inode *inode = file->f_path.dentry->d_inode;
365 /* File with pending delete? */
366 if (inode->i_nlink == 0)
369 /* Writeable file? */
370 if (S_ISREG(inode->i_mode) && (file->f_mode & FMODE_WRITE))
373 spin_unlock(&files_lock);
374 return 1; /* Tis' cool bro. */
376 spin_unlock(&files_lock);
381 * mark_files_ro - mark all files read-only
382 * @sb: superblock in question
384 * All files are marked read-only. We don't care about pending
385 * delete files so this should be used in 'force' mode only.
387 void mark_files_ro(struct super_block *sb)
392 spin_lock(&files_lock);
393 list_for_each_entry(f, &sb->s_files, f_u.fu_list) {
394 struct vfsmount *mnt;
395 if (!S_ISREG(f->f_path.dentry->d_inode->i_mode))
399 if (!(f->f_mode & FMODE_WRITE))
401 spin_lock(&f->f_lock);
402 f->f_mode &= ~FMODE_WRITE;
403 spin_unlock(&f->f_lock);
404 if (file_check_writeable(f) != 0)
406 file_release_write(f);
407 mnt = mntget(f->f_path.mnt);
408 /* This can sleep, so we can't hold the spinlock. */
409 spin_unlock(&files_lock);
414 spin_unlock(&files_lock);
417 void __init files_init(unsigned long mempages)
421 filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
422 SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
425 * One file with associated inode and dcache is very roughly 1K.
426 * Per default don't use more than 10% of our memory for files.
429 n = (mempages * (PAGE_SIZE / 1024)) / 10;
430 files_stat.max_files = n;
431 if (files_stat.max_files < NR_FILE)
432 files_stat.max_files = NR_FILE;
434 percpu_counter_init(&nr_files, 0);