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/lglock.h>
24 #include <linux/percpu_counter.h>
25 #include <linux/percpu.h>
26 #include <linux/hardirq.h>
27 #include <linux/task_work.h>
28 #include <linux/ima.h>
30 #include <linux/atomic.h>
34 /* sysctl tunables... */
35 struct files_stat_struct files_stat = {
39 /* SLAB cache for file structures */
40 static struct kmem_cache *filp_cachep __read_mostly;
42 static struct percpu_counter nr_files __cacheline_aligned_in_smp;
44 static void file_free_rcu(struct rcu_head *head)
46 struct file *f = container_of(head, struct file, f_u.fu_rcuhead);
49 kmem_cache_free(filp_cachep, f);
52 static inline void file_free(struct file *f)
54 percpu_counter_dec(&nr_files);
56 call_rcu(&f->f_u.fu_rcuhead, file_free_rcu);
60 * Return the total number of open files in the system
62 static long get_nr_files(void)
64 return percpu_counter_read_positive(&nr_files);
68 * Return the maximum number of open files in the system
70 unsigned long get_max_files(void)
72 return files_stat.max_files;
74 EXPORT_SYMBOL_GPL(get_max_files);
77 * Handle nr_files sysctl
79 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
80 int proc_nr_files(ctl_table *table, int write,
81 void __user *buffer, size_t *lenp, loff_t *ppos)
83 files_stat.nr_files = get_nr_files();
84 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
87 int proc_nr_files(ctl_table *table, int write,
88 void __user *buffer, size_t *lenp, loff_t *ppos)
94 /* Find an unused file structure and return a pointer to it.
95 * Returns an error pointer if some error happend e.g. we over file
96 * structures limit, run out of memory or operation is not permitted.
98 * Be very careful using this. You are responsible for
99 * getting write access to any mount that you might assign
100 * to this filp, if it is opened for write. If this is not
101 * done, you will imbalance int the mount's writer count
102 * and a warning at __fput() time.
104 struct file *get_empty_filp(void)
106 const struct cred *cred = current_cred();
112 * Privileged users can go above max_files
114 if (get_nr_files() >= files_stat.max_files && !capable(CAP_SYS_ADMIN)) {
116 * percpu_counters are inaccurate. Do an expensive check before
119 if (percpu_counter_sum_positive(&nr_files) >= files_stat.max_files)
123 f = kmem_cache_zalloc(filp_cachep, GFP_KERNEL);
125 return ERR_PTR(-ENOMEM);
127 percpu_counter_inc(&nr_files);
128 f->f_cred = get_cred(cred);
129 error = security_file_alloc(f);
130 if (unlikely(error)) {
132 return ERR_PTR(error);
135 atomic_long_set(&f->f_count, 1);
136 rwlock_init(&f->f_owner.lock);
137 spin_lock_init(&f->f_lock);
138 mutex_init(&f->f_pos_lock);
139 eventpoll_init_file(f);
140 /* f->f_version: 0 */
144 /* Ran out of filps - report that */
145 if (get_nr_files() > old_max) {
146 pr_info("VFS: file-max limit %lu reached\n", get_max_files());
147 old_max = get_nr_files();
149 return ERR_PTR(-ENFILE);
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_inode = path->dentry->d_inode;
178 file->f_mapping = path->dentry->d_inode->i_mapping;
183 * These mounts don't really matter in practice
184 * for r/o bind mounts. They aren't userspace-
185 * visible. We do this for consistency, and so
186 * that we can do debugging checks at __fput()
188 if ((mode & FMODE_WRITE) && !special_file(path->dentry->d_inode->i_mode)) {
189 file_take_write(file);
190 WARN_ON(mnt_clone_write(path->mnt));
192 if ((mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
193 i_readcount_inc(path->dentry->d_inode);
196 EXPORT_SYMBOL(alloc_file);
199 * drop_file_write_access - give up ability to write to a file
200 * @file: the file to which we will stop writing
202 * This is a central place which will give up the ability
203 * to write to @file, along with access to write through
206 static void drop_file_write_access(struct file *file)
208 struct vfsmount *mnt = file->f_path.mnt;
209 struct dentry *dentry = file->f_path.dentry;
210 struct inode *inode = dentry->d_inode;
212 put_write_access(inode);
214 if (special_file(inode->i_mode))
216 if (file_check_writeable(file) != 0)
218 __mnt_drop_write(mnt);
219 file_release_write(file);
222 /* the real guts of fput() - releasing the last reference to file
224 static void __fput(struct file *file)
226 struct dentry *dentry = file->f_path.dentry;
227 struct vfsmount *mnt = file->f_path.mnt;
228 struct inode *inode = file->f_inode;
232 fsnotify_close(file);
234 * The function eventpoll_release() should be the first called
235 * in the file cleanup chain.
237 eventpoll_release(file);
238 locks_remove_flock(file);
240 if (unlikely(file->f_flags & FASYNC)) {
241 if (file->f_op->fasync)
242 file->f_op->fasync(-1, file, 0);
245 if (file->f_op->release)
246 file->f_op->release(inode, file);
247 security_file_free(file);
248 if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL &&
249 !(file->f_mode & FMODE_PATH))) {
250 cdev_put(inode->i_cdev);
252 fops_put(file->f_op);
253 put_pid(file->f_owner.pid);
254 if ((file->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
255 i_readcount_dec(inode);
256 if (file->f_mode & FMODE_WRITE)
257 drop_file_write_access(file);
258 file->f_path.dentry = NULL;
259 file->f_path.mnt = NULL;
260 file->f_inode = NULL;
266 static LLIST_HEAD(delayed_fput_list);
267 static void delayed_fput(struct work_struct *unused)
269 struct llist_node *node = llist_del_all(&delayed_fput_list);
270 struct llist_node *next;
272 for (; node; node = next) {
273 next = llist_next(node);
274 __fput(llist_entry(node, struct file, f_u.fu_llist));
278 static void ____fput(struct callback_head *work)
280 __fput(container_of(work, struct file, f_u.fu_rcuhead));
284 * If kernel thread really needs to have the final fput() it has done
285 * to complete, call this. The only user right now is the boot - we
286 * *do* need to make sure our writes to binaries on initramfs has
287 * not left us with opened struct file waiting for __fput() - execve()
288 * won't work without that. Please, don't add more callers without
289 * very good reasons; in particular, never call that with locks
290 * held and never call that from a thread that might need to do
291 * some work on any kind of umount.
293 void flush_delayed_fput(void)
298 static DECLARE_DELAYED_WORK(delayed_fput_work, delayed_fput);
300 void fput(struct file *file)
302 if (atomic_long_dec_and_test(&file->f_count)) {
303 struct task_struct *task = current;
305 if (likely(!in_interrupt() && !(task->flags & PF_KTHREAD))) {
306 init_task_work(&file->f_u.fu_rcuhead, ____fput);
307 if (!task_work_add(task, &file->f_u.fu_rcuhead, true))
310 * After this task has run exit_task_work(),
311 * task_work_add() will fail. Fall through to delayed
312 * fput to avoid leaking *file.
316 if (llist_add(&file->f_u.fu_llist, &delayed_fput_list))
317 schedule_delayed_work(&delayed_fput_work, 1);
322 * synchronous analog of fput(); for kernel threads that might be needed
323 * in some umount() (and thus can't use flush_delayed_fput() without
324 * risking deadlocks), need to wait for completion of __fput() and know
325 * for this specific struct file it won't involve anything that would
326 * need them. Use only if you really need it - at the very least,
327 * don't blindly convert fput() by kernel thread to that.
329 void __fput_sync(struct file *file)
331 if (atomic_long_dec_and_test(&file->f_count)) {
332 struct task_struct *task = current;
333 BUG_ON(!(task->flags & PF_KTHREAD));
340 void put_filp(struct file *file)
342 if (atomic_long_dec_and_test(&file->f_count)) {
343 security_file_free(file);
348 void __init files_init(unsigned long mempages)
352 filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
353 SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
356 * One file with associated inode and dcache is very roughly 1K.
357 * Per default don't use more than 10% of our memory for files.
360 n = (mempages * (PAGE_SIZE / 1024)) / 10;
361 files_stat.max_files = max_t(unsigned long, n, NR_FILE);
363 percpu_counter_init(&nr_files, 0);