1 // SPDX-License-Identifier: GPL-2.0-only
3 * linux/fs/file_table.c
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 * Copyright (C) 1997 David S. Miller (davem@caip.rutgers.edu)
9 #include <linux/string.h>
10 #include <linux/slab.h>
11 #include <linux/file.h>
12 #include <linux/fdtable.h>
13 #include <linux/init.h>
14 #include <linux/module.h>
16 #include <linux/security.h>
17 #include <linux/cred.h>
18 #include <linux/eventpoll.h>
19 #include <linux/rcupdate.h>
20 #include <linux/mount.h>
21 #include <linux/capability.h>
22 #include <linux/cdev.h>
23 #include <linux/fsnotify.h>
24 #include <linux/sysctl.h>
25 #include <linux/percpu_counter.h>
26 #include <linux/percpu.h>
27 #include <linux/task_work.h>
28 #include <linux/ima.h>
29 #include <linux/swap.h>
31 #include <linux/atomic.h>
35 /* sysctl tunables... */
36 struct files_stat_struct files_stat = {
40 /* SLAB cache for file structures */
41 static struct kmem_cache *filp_cachep __read_mostly;
43 static struct percpu_counter nr_files __cacheline_aligned_in_smp;
45 static void file_free_rcu(struct rcu_head *head)
47 struct file *f = container_of(head, struct file, f_u.fu_rcuhead);
50 kmem_cache_free(filp_cachep, f);
53 static inline void file_free(struct file *f)
55 security_file_free(f);
56 if (!(f->f_mode & FMODE_NOACCOUNT))
57 percpu_counter_dec(&nr_files);
58 call_rcu(&f->f_u.fu_rcuhead, file_free_rcu);
62 * Return the total number of open files in the system
64 static long get_nr_files(void)
66 return percpu_counter_read_positive(&nr_files);
70 * Return the maximum number of open files in the system
72 unsigned long get_max_files(void)
74 return files_stat.max_files;
76 EXPORT_SYMBOL_GPL(get_max_files);
79 * Handle nr_files sysctl
81 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
82 int proc_nr_files(struct ctl_table *table, int write,
83 void __user *buffer, size_t *lenp, loff_t *ppos)
85 files_stat.nr_files = get_nr_files();
86 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
89 int proc_nr_files(struct ctl_table *table, int write,
90 void __user *buffer, size_t *lenp, loff_t *ppos)
96 static struct file *__alloc_file(int flags, const struct cred *cred)
101 f = kmem_cache_zalloc(filp_cachep, GFP_KERNEL);
103 return ERR_PTR(-ENOMEM);
105 f->f_cred = get_cred(cred);
106 error = security_file_alloc(f);
107 if (unlikely(error)) {
108 file_free_rcu(&f->f_u.fu_rcuhead);
109 return ERR_PTR(error);
112 atomic_long_set(&f->f_count, 1);
113 rwlock_init(&f->f_owner.lock);
114 spin_lock_init(&f->f_lock);
115 mutex_init(&f->f_pos_lock);
116 eventpoll_init_file(f);
118 f->f_mode = OPEN_FMODE(flags);
119 /* f->f_version: 0 */
124 /* Find an unused file structure and return a pointer to it.
125 * Returns an error pointer if some error happend e.g. we over file
126 * structures limit, run out of memory or operation is not permitted.
128 * Be very careful using this. You are responsible for
129 * getting write access to any mount that you might assign
130 * to this filp, if it is opened for write. If this is not
131 * done, you will imbalance int the mount's writer count
132 * and a warning at __fput() time.
134 struct file *alloc_empty_file(int flags, const struct cred *cred)
140 * Privileged users can go above max_files
142 if (get_nr_files() >= files_stat.max_files && !capable(CAP_SYS_ADMIN)) {
144 * percpu_counters are inaccurate. Do an expensive check before
147 if (percpu_counter_sum_positive(&nr_files) >= files_stat.max_files)
151 f = __alloc_file(flags, cred);
153 percpu_counter_inc(&nr_files);
158 /* Ran out of filps - report that */
159 if (get_nr_files() > old_max) {
160 pr_info("VFS: file-max limit %lu reached\n", get_max_files());
161 old_max = get_nr_files();
163 return ERR_PTR(-ENFILE);
167 * Variant of alloc_empty_file() that doesn't check and modify nr_files.
169 * Should not be used unless there's a very good reason to do so.
171 struct file *alloc_empty_file_noaccount(int flags, const struct cred *cred)
173 struct file *f = __alloc_file(flags, cred);
176 f->f_mode |= FMODE_NOACCOUNT;
182 * alloc_file - allocate and initialize a 'struct file'
184 * @path: the (dentry, vfsmount) pair for the new file
185 * @flags: O_... flags with which the new file will be opened
186 * @fop: the 'struct file_operations' for the new file
188 static struct file *alloc_file(const struct path *path, int flags,
189 const struct file_operations *fop)
193 file = alloc_empty_file(flags, current_cred());
197 file->f_path = *path;
198 file->f_inode = path->dentry->d_inode;
199 file->f_mapping = path->dentry->d_inode->i_mapping;
200 file->f_wb_err = filemap_sample_wb_err(file->f_mapping);
201 if ((file->f_mode & FMODE_READ) &&
202 likely(fop->read || fop->read_iter))
203 file->f_mode |= FMODE_CAN_READ;
204 if ((file->f_mode & FMODE_WRITE) &&
205 likely(fop->write || fop->write_iter))
206 file->f_mode |= FMODE_CAN_WRITE;
207 file->f_mode |= FMODE_OPENED;
209 if ((file->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
210 i_readcount_inc(path->dentry->d_inode);
214 struct file *alloc_file_pseudo(struct inode *inode, struct vfsmount *mnt,
215 const char *name, int flags,
216 const struct file_operations *fops)
218 static const struct dentry_operations anon_ops = {
219 .d_dname = simple_dname
221 struct qstr this = QSTR_INIT(name, strlen(name));
225 path.dentry = d_alloc_pseudo(mnt->mnt_sb, &this);
227 return ERR_PTR(-ENOMEM);
228 if (!mnt->mnt_sb->s_d_op)
229 d_set_d_op(path.dentry, &anon_ops);
230 path.mnt = mntget(mnt);
231 d_instantiate(path.dentry, inode);
232 file = alloc_file(&path, flags, fops);
239 EXPORT_SYMBOL(alloc_file_pseudo);
241 struct file *alloc_file_clone(struct file *base, int flags,
242 const struct file_operations *fops)
244 struct file *f = alloc_file(&base->f_path, flags, fops);
246 path_get(&f->f_path);
247 f->f_mapping = base->f_mapping;
252 /* the real guts of fput() - releasing the last reference to file
254 static void __fput(struct file *file)
256 struct dentry *dentry = file->f_path.dentry;
257 struct vfsmount *mnt = file->f_path.mnt;
258 struct inode *inode = file->f_inode;
259 fmode_t mode = file->f_mode;
261 if (unlikely(!(file->f_mode & FMODE_OPENED)))
266 fsnotify_close(file);
268 * The function eventpoll_release() should be the first called
269 * in the file cleanup chain.
271 eventpoll_release(file);
272 locks_remove_file(file);
275 if (unlikely(file->f_flags & FASYNC)) {
276 if (file->f_op->fasync)
277 file->f_op->fasync(-1, file, 0);
279 if (file->f_op->release)
280 file->f_op->release(inode, file);
281 if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL &&
282 !(mode & FMODE_PATH))) {
283 cdev_put(inode->i_cdev);
285 fops_put(file->f_op);
286 put_pid(file->f_owner.pid);
287 if ((mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
288 i_readcount_dec(inode);
289 if (mode & FMODE_WRITER) {
290 put_write_access(inode);
291 __mnt_drop_write(mnt);
294 if (unlikely(mode & FMODE_NEED_UNMOUNT))
295 dissolve_on_fput(mnt);
301 static LLIST_HEAD(delayed_fput_list);
302 static void delayed_fput(struct work_struct *unused)
304 struct llist_node *node = llist_del_all(&delayed_fput_list);
307 llist_for_each_entry_safe(f, t, node, f_u.fu_llist)
311 static void ____fput(struct callback_head *work)
313 __fput(container_of(work, struct file, f_u.fu_rcuhead));
317 * If kernel thread really needs to have the final fput() it has done
318 * to complete, call this. The only user right now is the boot - we
319 * *do* need to make sure our writes to binaries on initramfs has
320 * not left us with opened struct file waiting for __fput() - execve()
321 * won't work without that. Please, don't add more callers without
322 * very good reasons; in particular, never call that with locks
323 * held and never call that from a thread that might need to do
324 * some work on any kind of umount.
326 void flush_delayed_fput(void)
331 static DECLARE_DELAYED_WORK(delayed_fput_work, delayed_fput);
333 void fput_many(struct file *file, unsigned int refs)
335 if (atomic_long_sub_and_test(refs, &file->f_count)) {
336 struct task_struct *task = current;
338 if (likely(!in_interrupt() && !(task->flags & PF_KTHREAD))) {
339 init_task_work(&file->f_u.fu_rcuhead, ____fput);
340 if (!task_work_add(task, &file->f_u.fu_rcuhead, true))
343 * After this task has run exit_task_work(),
344 * task_work_add() will fail. Fall through to delayed
345 * fput to avoid leaking *file.
349 if (llist_add(&file->f_u.fu_llist, &delayed_fput_list))
350 schedule_delayed_work(&delayed_fput_work, 1);
354 void fput(struct file *file)
360 * synchronous analog of fput(); for kernel threads that might be needed
361 * in some umount() (and thus can't use flush_delayed_fput() without
362 * risking deadlocks), need to wait for completion of __fput() and know
363 * for this specific struct file it won't involve anything that would
364 * need them. Use only if you really need it - at the very least,
365 * don't blindly convert fput() by kernel thread to that.
367 void __fput_sync(struct file *file)
369 if (atomic_long_dec_and_test(&file->f_count)) {
370 struct task_struct *task = current;
371 BUG_ON(!(task->flags & PF_KTHREAD));
378 void __init files_init(void)
380 filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
381 SLAB_HWCACHE_ALIGN | SLAB_PANIC | SLAB_ACCOUNT, NULL);
382 percpu_counter_init(&nr_files, 0, GFP_KERNEL);
386 * One file with associated inode and dcache is very roughly 1K. Per default
387 * do not use more than 10% of our memory for files.
389 void __init files_maxfiles_init(void)
392 unsigned long nr_pages = totalram_pages();
393 unsigned long memreserve = (nr_pages - nr_free_pages()) * 3/2;
395 memreserve = min(memreserve, nr_pages - 1);
396 n = ((nr_pages - memreserve) * (PAGE_SIZE / 1024)) / 10;
398 files_stat.max_files = max_t(unsigned long, n, NR_FILE);