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>
30 #include <linux/kmemleak.h>
32 #include <linux/atomic.h>
36 /* sysctl tunables... */
37 static struct files_stat_struct files_stat = {
41 /* SLAB cache for file structures */
42 static struct kmem_cache *filp_cachep __read_mostly;
44 static struct percpu_counter nr_files __cacheline_aligned_in_smp;
46 static void file_free_rcu(struct rcu_head *head)
48 struct file *f = container_of(head, struct file, f_u.fu_rcuhead);
51 kmem_cache_free(filp_cachep, f);
54 static inline void file_free(struct file *f)
56 security_file_free(f);
57 if (!(f->f_mode & FMODE_NOACCOUNT))
58 percpu_counter_dec(&nr_files);
59 call_rcu(&f->f_u.fu_rcuhead, file_free_rcu);
63 * Return the total number of open files in the system
65 static long get_nr_files(void)
67 return percpu_counter_read_positive(&nr_files);
71 * Return the maximum number of open files in the system
73 unsigned long get_max_files(void)
75 return files_stat.max_files;
77 EXPORT_SYMBOL_GPL(get_max_files);
79 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
82 * Handle nr_files sysctl
84 static int proc_nr_files(struct ctl_table *table, int write, void *buffer,
85 size_t *lenp, loff_t *ppos)
87 files_stat.nr_files = get_nr_files();
88 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
91 static struct ctl_table fs_stat_sysctls[] = {
93 .procname = "file-nr",
95 .maxlen = sizeof(files_stat),
97 .proc_handler = proc_nr_files,
100 .procname = "file-max",
101 .data = &files_stat.max_files,
102 .maxlen = sizeof(files_stat.max_files),
104 .proc_handler = proc_doulongvec_minmax,
105 .extra1 = SYSCTL_LONG_ZERO,
106 .extra2 = SYSCTL_LONG_MAX,
109 .procname = "nr_open",
110 .data = &sysctl_nr_open,
111 .maxlen = sizeof(unsigned int),
113 .proc_handler = proc_dointvec_minmax,
114 .extra1 = &sysctl_nr_open_min,
115 .extra2 = &sysctl_nr_open_max,
120 static int __init init_fs_stat_sysctls(void)
122 register_sysctl_init("fs", fs_stat_sysctls);
123 if (IS_ENABLED(CONFIG_BINFMT_MISC)) {
124 struct ctl_table_header *hdr;
125 hdr = register_sysctl_mount_point("fs/binfmt_misc");
126 kmemleak_not_leak(hdr);
130 fs_initcall(init_fs_stat_sysctls);
133 static struct file *__alloc_file(int flags, const struct cred *cred)
138 f = kmem_cache_zalloc(filp_cachep, GFP_KERNEL);
140 return ERR_PTR(-ENOMEM);
142 f->f_cred = get_cred(cred);
143 error = security_file_alloc(f);
144 if (unlikely(error)) {
145 file_free_rcu(&f->f_u.fu_rcuhead);
146 return ERR_PTR(error);
149 atomic_long_set(&f->f_count, 1);
150 rwlock_init(&f->f_owner.lock);
151 spin_lock_init(&f->f_lock);
152 mutex_init(&f->f_pos_lock);
154 f->f_mode = OPEN_FMODE(flags);
155 /* f->f_version: 0 */
160 /* Find an unused file structure and return a pointer to it.
161 * Returns an error pointer if some error happend e.g. we over file
162 * structures limit, run out of memory or operation is not permitted.
164 * Be very careful using this. You are responsible for
165 * getting write access to any mount that you might assign
166 * to this filp, if it is opened for write. If this is not
167 * done, you will imbalance int the mount's writer count
168 * and a warning at __fput() time.
170 struct file *alloc_empty_file(int flags, const struct cred *cred)
176 * Privileged users can go above max_files
178 if (get_nr_files() >= files_stat.max_files && !capable(CAP_SYS_ADMIN)) {
180 * percpu_counters are inaccurate. Do an expensive check before
183 if (percpu_counter_sum_positive(&nr_files) >= files_stat.max_files)
187 f = __alloc_file(flags, cred);
189 percpu_counter_inc(&nr_files);
194 /* Ran out of filps - report that */
195 if (get_nr_files() > old_max) {
196 pr_info("VFS: file-max limit %lu reached\n", get_max_files());
197 old_max = get_nr_files();
199 return ERR_PTR(-ENFILE);
203 * Variant of alloc_empty_file() that doesn't check and modify nr_files.
205 * Should not be used unless there's a very good reason to do so.
207 struct file *alloc_empty_file_noaccount(int flags, const struct cred *cred)
209 struct file *f = __alloc_file(flags, cred);
212 f->f_mode |= FMODE_NOACCOUNT;
218 * alloc_file - allocate and initialize a 'struct file'
220 * @path: the (dentry, vfsmount) pair for the new file
221 * @flags: O_... flags with which the new file will be opened
222 * @fop: the 'struct file_operations' for the new file
224 static struct file *alloc_file(const struct path *path, int flags,
225 const struct file_operations *fop)
229 file = alloc_empty_file(flags, current_cred());
233 file->f_path = *path;
234 file->f_inode = path->dentry->d_inode;
235 file->f_mapping = path->dentry->d_inode->i_mapping;
236 file->f_wb_err = filemap_sample_wb_err(file->f_mapping);
237 file->f_sb_err = file_sample_sb_err(file);
238 if ((file->f_mode & FMODE_READ) &&
239 likely(fop->read || fop->read_iter))
240 file->f_mode |= FMODE_CAN_READ;
241 if ((file->f_mode & FMODE_WRITE) &&
242 likely(fop->write || fop->write_iter))
243 file->f_mode |= FMODE_CAN_WRITE;
244 file->f_mode |= FMODE_OPENED;
246 if ((file->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
247 i_readcount_inc(path->dentry->d_inode);
251 struct file *alloc_file_pseudo(struct inode *inode, struct vfsmount *mnt,
252 const char *name, int flags,
253 const struct file_operations *fops)
255 static const struct dentry_operations anon_ops = {
256 .d_dname = simple_dname
258 struct qstr this = QSTR_INIT(name, strlen(name));
262 path.dentry = d_alloc_pseudo(mnt->mnt_sb, &this);
264 return ERR_PTR(-ENOMEM);
265 if (!mnt->mnt_sb->s_d_op)
266 d_set_d_op(path.dentry, &anon_ops);
267 path.mnt = mntget(mnt);
268 d_instantiate(path.dentry, inode);
269 file = alloc_file(&path, flags, fops);
276 EXPORT_SYMBOL(alloc_file_pseudo);
278 struct file *alloc_file_clone(struct file *base, int flags,
279 const struct file_operations *fops)
281 struct file *f = alloc_file(&base->f_path, flags, fops);
283 path_get(&f->f_path);
284 f->f_mapping = base->f_mapping;
289 /* the real guts of fput() - releasing the last reference to file
291 static void __fput(struct file *file)
293 struct dentry *dentry = file->f_path.dentry;
294 struct vfsmount *mnt = file->f_path.mnt;
295 struct inode *inode = file->f_inode;
296 fmode_t mode = file->f_mode;
298 if (unlikely(!(file->f_mode & FMODE_OPENED)))
303 fsnotify_close(file);
305 * The function eventpoll_release() should be the first called
306 * in the file cleanup chain.
308 eventpoll_release(file);
309 locks_remove_file(file);
312 if (unlikely(file->f_flags & FASYNC)) {
313 if (file->f_op->fasync)
314 file->f_op->fasync(-1, file, 0);
316 if (file->f_op->release)
317 file->f_op->release(inode, file);
318 if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL &&
319 !(mode & FMODE_PATH))) {
320 cdev_put(inode->i_cdev);
322 fops_put(file->f_op);
323 put_pid(file->f_owner.pid);
324 if ((mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
325 i_readcount_dec(inode);
326 if (mode & FMODE_WRITER) {
327 put_write_access(inode);
328 __mnt_drop_write(mnt);
331 if (unlikely(mode & FMODE_NEED_UNMOUNT))
332 dissolve_on_fput(mnt);
338 static LLIST_HEAD(delayed_fput_list);
339 static void delayed_fput(struct work_struct *unused)
341 struct llist_node *node = llist_del_all(&delayed_fput_list);
344 llist_for_each_entry_safe(f, t, node, f_u.fu_llist)
348 static void ____fput(struct callback_head *work)
350 __fput(container_of(work, struct file, f_u.fu_rcuhead));
354 * If kernel thread really needs to have the final fput() it has done
355 * to complete, call this. The only user right now is the boot - we
356 * *do* need to make sure our writes to binaries on initramfs has
357 * not left us with opened struct file waiting for __fput() - execve()
358 * won't work without that. Please, don't add more callers without
359 * very good reasons; in particular, never call that with locks
360 * held and never call that from a thread that might need to do
361 * some work on any kind of umount.
363 void flush_delayed_fput(void)
367 EXPORT_SYMBOL_GPL(flush_delayed_fput);
369 static DECLARE_DELAYED_WORK(delayed_fput_work, delayed_fput);
371 void fput_many(struct file *file, unsigned int refs)
373 if (atomic_long_sub_and_test(refs, &file->f_count)) {
374 struct task_struct *task = current;
376 if (likely(!in_interrupt() && !(task->flags & PF_KTHREAD))) {
377 init_task_work(&file->f_u.fu_rcuhead, ____fput);
378 if (!task_work_add(task, &file->f_u.fu_rcuhead, TWA_RESUME))
381 * After this task has run exit_task_work(),
382 * task_work_add() will fail. Fall through to delayed
383 * fput to avoid leaking *file.
387 if (llist_add(&file->f_u.fu_llist, &delayed_fput_list))
388 schedule_delayed_work(&delayed_fput_work, 1);
392 void fput(struct file *file)
398 * synchronous analog of fput(); for kernel threads that might be needed
399 * in some umount() (and thus can't use flush_delayed_fput() without
400 * risking deadlocks), need to wait for completion of __fput() and know
401 * for this specific struct file it won't involve anything that would
402 * need them. Use only if you really need it - at the very least,
403 * don't blindly convert fput() by kernel thread to that.
405 void __fput_sync(struct file *file)
407 if (atomic_long_dec_and_test(&file->f_count)) {
408 struct task_struct *task = current;
409 BUG_ON(!(task->flags & PF_KTHREAD));
415 EXPORT_SYMBOL(__fput_sync);
417 void __init files_init(void)
419 filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
420 SLAB_HWCACHE_ALIGN | SLAB_PANIC | SLAB_ACCOUNT, NULL);
421 percpu_counter_init(&nr_files, 0, GFP_KERNEL);
425 * One file with associated inode and dcache is very roughly 1K. Per default
426 * do not use more than 10% of our memory for files.
428 void __init files_maxfiles_init(void)
431 unsigned long nr_pages = totalram_pages();
432 unsigned long memreserve = (nr_pages - nr_free_pages()) * 3/2;
434 memreserve = min(memreserve, nr_pages - 1);
435 n = ((nr_pages - memreserve) * (PAGE_SIZE / 1024)) / 10;
437 files_stat.max_files = max_t(unsigned long, n, NR_FILE);