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/cred.h>
17 #include <linux/eventpoll.h>
18 #include <linux/rcupdate.h>
19 #include <linux/mount.h>
20 #include <linux/capability.h>
21 #include <linux/cdev.h>
22 #include <linux/fsnotify.h>
23 #include <linux/sysctl.h>
24 #include <linux/percpu_counter.h>
25 #include <linux/percpu.h>
26 #include <linux/task_work.h>
27 #include <linux/ima.h>
28 #include <linux/swap.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 security_file_free(f);
55 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(struct 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(struct 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 *alloc_empty_file(int flags, const struct cred *cred)
111 * Privileged users can go above max_files
113 if (get_nr_files() >= files_stat.max_files && !capable(CAP_SYS_ADMIN)) {
115 * percpu_counters are inaccurate. Do an expensive check before
118 if (percpu_counter_sum_positive(&nr_files) >= files_stat.max_files)
122 f = kmem_cache_zalloc(filp_cachep, GFP_KERNEL);
124 return ERR_PTR(-ENOMEM);
126 f->f_cred = get_cred(cred);
127 error = security_file_alloc(f);
128 if (unlikely(error)) {
129 file_free_rcu(&f->f_u.fu_rcuhead);
130 return ERR_PTR(error);
133 atomic_long_set(&f->f_count, 1);
134 rwlock_init(&f->f_owner.lock);
135 spin_lock_init(&f->f_lock);
136 mutex_init(&f->f_pos_lock);
137 eventpoll_init_file(f);
139 f->f_mode = OPEN_FMODE(flags);
140 /* f->f_version: 0 */
141 percpu_counter_inc(&nr_files);
145 /* Ran out of filps - report that */
146 if (get_nr_files() > old_max) {
147 pr_info("VFS: file-max limit %lu reached\n", get_max_files());
148 old_max = get_nr_files();
150 return ERR_PTR(-ENFILE);
154 * alloc_file - allocate and initialize a 'struct file'
156 * @path: the (dentry, vfsmount) pair for the new file
157 * @flags: O_... flags with which the new file will be opened
158 * @fop: the 'struct file_operations' for the new file
160 struct file *alloc_file(const struct path *path, int flags,
161 const struct file_operations *fop)
165 file = alloc_empty_file(flags, current_cred());
169 file->f_path = *path;
170 file->f_inode = path->dentry->d_inode;
171 file->f_mapping = path->dentry->d_inode->i_mapping;
172 file->f_wb_err = filemap_sample_wb_err(file->f_mapping);
173 if ((file->f_mode & FMODE_READ) &&
174 likely(fop->read || fop->read_iter))
175 file->f_mode |= FMODE_CAN_READ;
176 if ((file->f_mode & FMODE_WRITE) &&
177 likely(fop->write || fop->write_iter))
178 file->f_mode |= FMODE_CAN_WRITE;
179 file->f_mode |= FMODE_OPENED;
181 if ((file->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
182 i_readcount_inc(path->dentry->d_inode);
185 EXPORT_SYMBOL(alloc_file);
187 struct file *alloc_file_pseudo(struct inode *inode, struct vfsmount *mnt,
188 const char *name, int flags,
189 const struct file_operations *fops)
191 static const struct dentry_operations anon_ops = {
192 .d_dname = simple_dname
194 struct qstr this = QSTR_INIT(name, strlen(name));
198 path.dentry = d_alloc_pseudo(mnt->mnt_sb, &this);
200 return ERR_PTR(-ENOMEM);
201 if (!mnt->mnt_sb->s_d_op)
202 d_set_d_op(path.dentry, &anon_ops);
203 path.mnt = mntget(mnt);
204 d_instantiate(path.dentry, inode);
205 file = alloc_file(&path, flags, fops);
212 EXPORT_SYMBOL(alloc_file_pseudo);
214 struct file *alloc_file_clone(struct file *base, int flags,
215 const struct file_operations *fops)
217 struct file *f = alloc_file(&base->f_path, flags, fops);
219 path_get(&f->f_path);
220 f->f_mapping = base->f_mapping;
225 /* the real guts of fput() - releasing the last reference to file
227 static void __fput(struct file *file)
229 struct dentry *dentry = file->f_path.dentry;
230 struct vfsmount *mnt = file->f_path.mnt;
231 struct inode *inode = file->f_inode;
233 if (unlikely(!(file->f_mode & FMODE_OPENED)))
238 fsnotify_close(file);
240 * The function eventpoll_release() should be the first called
241 * in the file cleanup chain.
243 eventpoll_release(file);
244 locks_remove_file(file);
247 if (unlikely(file->f_flags & FASYNC)) {
248 if (file->f_op->fasync)
249 file->f_op->fasync(-1, file, 0);
251 if (file->f_op->release)
252 file->f_op->release(inode, file);
253 if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL &&
254 !(file->f_mode & FMODE_PATH))) {
255 cdev_put(inode->i_cdev);
257 fops_put(file->f_op);
258 put_pid(file->f_owner.pid);
259 if ((file->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
260 i_readcount_dec(inode);
261 if (file->f_mode & FMODE_WRITER) {
262 put_write_access(inode);
263 __mnt_drop_write(mnt);
271 static LLIST_HEAD(delayed_fput_list);
272 static void delayed_fput(struct work_struct *unused)
274 struct llist_node *node = llist_del_all(&delayed_fput_list);
277 llist_for_each_entry_safe(f, t, node, f_u.fu_llist)
281 static void ____fput(struct callback_head *work)
283 __fput(container_of(work, struct file, f_u.fu_rcuhead));
287 * If kernel thread really needs to have the final fput() it has done
288 * to complete, call this. The only user right now is the boot - we
289 * *do* need to make sure our writes to binaries on initramfs has
290 * not left us with opened struct file waiting for __fput() - execve()
291 * won't work without that. Please, don't add more callers without
292 * very good reasons; in particular, never call that with locks
293 * held and never call that from a thread that might need to do
294 * some work on any kind of umount.
296 void flush_delayed_fput(void)
301 static DECLARE_DELAYED_WORK(delayed_fput_work, delayed_fput);
303 void fput(struct file *file)
305 if (atomic_long_dec_and_test(&file->f_count)) {
306 struct task_struct *task = current;
308 if (likely(!in_interrupt() && !(task->flags & PF_KTHREAD))) {
309 init_task_work(&file->f_u.fu_rcuhead, ____fput);
310 if (!task_work_add(task, &file->f_u.fu_rcuhead, true))
313 * After this task has run exit_task_work(),
314 * task_work_add() will fail. Fall through to delayed
315 * fput to avoid leaking *file.
319 if (llist_add(&file->f_u.fu_llist, &delayed_fput_list))
320 schedule_delayed_work(&delayed_fput_work, 1);
325 * synchronous analog of fput(); for kernel threads that might be needed
326 * in some umount() (and thus can't use flush_delayed_fput() without
327 * risking deadlocks), need to wait for completion of __fput() and know
328 * for this specific struct file it won't involve anything that would
329 * need them. Use only if you really need it - at the very least,
330 * don't blindly convert fput() by kernel thread to that.
332 void __fput_sync(struct file *file)
334 if (atomic_long_dec_and_test(&file->f_count)) {
335 struct task_struct *task = current;
336 BUG_ON(!(task->flags & PF_KTHREAD));
343 void __init files_init(void)
345 filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
346 SLAB_HWCACHE_ALIGN | SLAB_PANIC | SLAB_ACCOUNT, NULL);
347 percpu_counter_init(&nr_files, 0, GFP_KERNEL);
351 * One file with associated inode and dcache is very roughly 1K. Per default
352 * do not use more than 10% of our memory for files.
354 void __init files_maxfiles_init(void)
357 unsigned long memreserve = (totalram_pages - nr_free_pages()) * 3/2;
359 memreserve = min(memreserve, totalram_pages - 1);
360 n = ((totalram_pages - memreserve) * (PAGE_SIZE / 1024)) / 10;
362 files_stat.max_files = max_t(unsigned long, n, NR_FILE);