1 // SPDX-License-Identifier: GPL-2.0-only
3 * linux/fs/file_table.c
5 * Copyright (C) 1991, 1992 Linus Torvalds
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/filelock.h>
17 #include <linux/security.h>
18 #include <linux/cred.h>
19 #include <linux/eventpoll.h>
20 #include <linux/rcupdate.h>
21 #include <linux/mount.h>
22 #include <linux/capability.h>
23 #include <linux/cdev.h>
24 #include <linux/fsnotify.h>
25 #include <linux/sysctl.h>
26 #include <linux/percpu_counter.h>
27 #include <linux/percpu.h>
28 #include <linux/task_work.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 __ro_after_init;
44 static struct percpu_counter nr_files __cacheline_aligned_in_smp;
46 /* Container for backing file with optional user path */
49 struct path user_path;
52 static inline struct backing_file *backing_file(struct file *f)
54 return container_of(f, struct backing_file, file);
57 struct path *backing_file_user_path(struct file *f)
59 return &backing_file(f)->user_path;
61 EXPORT_SYMBOL_GPL(backing_file_user_path);
63 static inline void file_free(struct file *f)
65 security_file_free(f);
66 if (likely(!(f->f_mode & FMODE_NOACCOUNT)))
67 percpu_counter_dec(&nr_files);
69 if (unlikely(f->f_mode & FMODE_BACKING)) {
70 path_put(backing_file_user_path(f));
71 kfree(backing_file(f));
73 kmem_cache_free(filp_cachep, f);
78 * Return the total number of open files in the system
80 static long get_nr_files(void)
82 return percpu_counter_read_positive(&nr_files);
86 * Return the maximum number of open files in the system
88 unsigned long get_max_files(void)
90 return files_stat.max_files;
92 EXPORT_SYMBOL_GPL(get_max_files);
94 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
97 * Handle nr_files sysctl
99 static int proc_nr_files(const struct ctl_table *table, int write, void *buffer,
100 size_t *lenp, loff_t *ppos)
102 files_stat.nr_files = get_nr_files();
103 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
106 static struct ctl_table fs_stat_sysctls[] = {
108 .procname = "file-nr",
110 .maxlen = sizeof(files_stat),
112 .proc_handler = proc_nr_files,
115 .procname = "file-max",
116 .data = &files_stat.max_files,
117 .maxlen = sizeof(files_stat.max_files),
119 .proc_handler = proc_doulongvec_minmax,
120 .extra1 = SYSCTL_LONG_ZERO,
121 .extra2 = SYSCTL_LONG_MAX,
124 .procname = "nr_open",
125 .data = &sysctl_nr_open,
126 .maxlen = sizeof(unsigned int),
128 .proc_handler = proc_dointvec_minmax,
129 .extra1 = &sysctl_nr_open_min,
130 .extra2 = &sysctl_nr_open_max,
134 static int __init init_fs_stat_sysctls(void)
136 register_sysctl_init("fs", fs_stat_sysctls);
137 if (IS_ENABLED(CONFIG_BINFMT_MISC)) {
138 struct ctl_table_header *hdr;
140 hdr = register_sysctl_mount_point("fs/binfmt_misc");
141 kmemleak_not_leak(hdr);
145 fs_initcall(init_fs_stat_sysctls);
148 static int init_file(struct file *f, int flags, const struct cred *cred)
152 f->f_cred = get_cred(cred);
153 error = security_file_alloc(f);
154 if (unlikely(error)) {
159 spin_lock_init(&f->f_lock);
161 * Note that f_pos_lock is only used for files raising
162 * FMODE_ATOMIC_POS and directories. Other files such as pipes
163 * don't need it and since f_pos_lock is in a union may reuse
164 * the space for other purposes. They are expected to initialize
165 * the respective member when opening the file.
167 mutex_init(&f->f_pos_lock);
169 f->f_mode = OPEN_FMODE(flags);
170 /* f->f_version: 0 */
173 * We're SLAB_TYPESAFE_BY_RCU so initialize f_count last. While
174 * fget-rcu pattern users need to be able to handle spurious
175 * refcount bumps we should reinitialize the reused file first.
177 atomic_long_set(&f->f_count, 1);
181 /* Find an unused file structure and return a pointer to it.
182 * Returns an error pointer if some error happend e.g. we over file
183 * structures limit, run out of memory or operation is not permitted.
185 * Be very careful using this. You are responsible for
186 * getting write access to any mount that you might assign
187 * to this filp, if it is opened for write. If this is not
188 * done, you will imbalance int the mount's writer count
189 * and a warning at __fput() time.
191 struct file *alloc_empty_file(int flags, const struct cred *cred)
198 * Privileged users can go above max_files
200 if (get_nr_files() >= files_stat.max_files && !capable(CAP_SYS_ADMIN)) {
202 * percpu_counters are inaccurate. Do an expensive check before
205 if (percpu_counter_sum_positive(&nr_files) >= files_stat.max_files)
209 f = kmem_cache_zalloc(filp_cachep, GFP_KERNEL);
211 return ERR_PTR(-ENOMEM);
213 error = init_file(f, flags, cred);
214 if (unlikely(error)) {
215 kmem_cache_free(filp_cachep, f);
216 return ERR_PTR(error);
219 percpu_counter_inc(&nr_files);
224 /* Ran out of filps - report that */
225 if (get_nr_files() > old_max) {
226 pr_info("VFS: file-max limit %lu reached\n", get_max_files());
227 old_max = get_nr_files();
229 return ERR_PTR(-ENFILE);
233 * Variant of alloc_empty_file() that doesn't check and modify nr_files.
235 * This is only for kernel internal use, and the allocate file must not be
236 * installed into file tables or such.
238 struct file *alloc_empty_file_noaccount(int flags, const struct cred *cred)
243 f = kmem_cache_zalloc(filp_cachep, GFP_KERNEL);
245 return ERR_PTR(-ENOMEM);
247 error = init_file(f, flags, cred);
248 if (unlikely(error)) {
249 kmem_cache_free(filp_cachep, f);
250 return ERR_PTR(error);
253 f->f_mode |= FMODE_NOACCOUNT;
259 * Variant of alloc_empty_file() that allocates a backing_file container
260 * and doesn't check and modify nr_files.
262 * This is only for kernel internal use, and the allocate file must not be
263 * installed into file tables or such.
265 struct file *alloc_empty_backing_file(int flags, const struct cred *cred)
267 struct backing_file *ff;
270 ff = kzalloc(sizeof(struct backing_file), GFP_KERNEL);
272 return ERR_PTR(-ENOMEM);
274 error = init_file(&ff->file, flags, cred);
275 if (unlikely(error)) {
277 return ERR_PTR(error);
280 ff->file.f_mode |= FMODE_BACKING | FMODE_NOACCOUNT;
285 * file_init_path - initialize a 'struct file' based on path
287 * @file: the file to set up
288 * @path: the (dentry, vfsmount) pair for the new file
289 * @fop: the 'struct file_operations' for the new file
291 static void file_init_path(struct file *file, const struct path *path,
292 const struct file_operations *fop)
294 file->f_path = *path;
295 file->f_inode = path->dentry->d_inode;
296 file->f_mapping = path->dentry->d_inode->i_mapping;
297 file->f_wb_err = filemap_sample_wb_err(file->f_mapping);
298 file->f_sb_err = file_sample_sb_err(file);
300 file->f_mode |= FMODE_LSEEK;
301 if ((file->f_mode & FMODE_READ) &&
302 likely(fop->read || fop->read_iter))
303 file->f_mode |= FMODE_CAN_READ;
304 if ((file->f_mode & FMODE_WRITE) &&
305 likely(fop->write || fop->write_iter))
306 file->f_mode |= FMODE_CAN_WRITE;
307 file->f_iocb_flags = iocb_flags(file);
308 file->f_mode |= FMODE_OPENED;
310 if ((file->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
311 i_readcount_inc(path->dentry->d_inode);
315 * alloc_file - allocate and initialize a 'struct file'
317 * @path: the (dentry, vfsmount) pair for the new file
318 * @flags: O_... flags with which the new file will be opened
319 * @fop: the 'struct file_operations' for the new file
321 static struct file *alloc_file(const struct path *path, int flags,
322 const struct file_operations *fop)
326 file = alloc_empty_file(flags, current_cred());
328 file_init_path(file, path, fop);
332 static inline int alloc_path_pseudo(const char *name, struct inode *inode,
333 struct vfsmount *mnt, struct path *path)
335 struct qstr this = QSTR_INIT(name, strlen(name));
337 path->dentry = d_alloc_pseudo(mnt->mnt_sb, &this);
340 path->mnt = mntget(mnt);
341 d_instantiate(path->dentry, inode);
345 struct file *alloc_file_pseudo(struct inode *inode, struct vfsmount *mnt,
346 const char *name, int flags,
347 const struct file_operations *fops)
353 ret = alloc_path_pseudo(name, inode, mnt, &path);
357 file = alloc_file(&path, flags, fops);
364 EXPORT_SYMBOL(alloc_file_pseudo);
366 struct file *alloc_file_pseudo_noaccount(struct inode *inode,
367 struct vfsmount *mnt, const char *name,
369 const struct file_operations *fops)
375 ret = alloc_path_pseudo(name, inode, mnt, &path);
379 file = alloc_empty_file_noaccount(flags, current_cred());
385 file_init_path(file, &path, fops);
388 EXPORT_SYMBOL_GPL(alloc_file_pseudo_noaccount);
390 struct file *alloc_file_clone(struct file *base, int flags,
391 const struct file_operations *fops)
395 f = alloc_file(&base->f_path, flags, fops);
397 path_get(&f->f_path);
398 f->f_mapping = base->f_mapping;
403 /* the real guts of fput() - releasing the last reference to file
405 static void __fput(struct file *file)
407 struct dentry *dentry = file->f_path.dentry;
408 struct vfsmount *mnt = file->f_path.mnt;
409 struct inode *inode = file->f_inode;
410 fmode_t mode = file->f_mode;
412 if (unlikely(!(file->f_mode & FMODE_OPENED)))
417 fsnotify_close(file);
419 * The function eventpoll_release() should be the first called
420 * in the file cleanup chain.
422 eventpoll_release(file);
423 locks_remove_file(file);
425 security_file_release(file);
426 if (unlikely(file->f_flags & FASYNC)) {
427 if (file->f_op->fasync)
428 file->f_op->fasync(-1, file, 0);
430 if (file->f_op->release)
431 file->f_op->release(inode, file);
432 if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL &&
433 !(mode & FMODE_PATH))) {
434 cdev_put(inode->i_cdev);
436 fops_put(file->f_op);
437 file_f_owner_release(file);
438 put_file_access(file);
440 if (unlikely(mode & FMODE_NEED_UNMOUNT))
441 dissolve_on_fput(mnt);
447 static LLIST_HEAD(delayed_fput_list);
448 static void delayed_fput(struct work_struct *unused)
450 struct llist_node *node = llist_del_all(&delayed_fput_list);
453 llist_for_each_entry_safe(f, t, node, f_llist)
457 static void ____fput(struct callback_head *work)
459 __fput(container_of(work, struct file, f_task_work));
463 * If kernel thread really needs to have the final fput() it has done
464 * to complete, call this. The only user right now is the boot - we
465 * *do* need to make sure our writes to binaries on initramfs has
466 * not left us with opened struct file waiting for __fput() - execve()
467 * won't work without that. Please, don't add more callers without
468 * very good reasons; in particular, never call that with locks
469 * held and never call that from a thread that might need to do
470 * some work on any kind of umount.
472 void flush_delayed_fput(void)
476 EXPORT_SYMBOL_GPL(flush_delayed_fput);
478 static DECLARE_DELAYED_WORK(delayed_fput_work, delayed_fput);
480 void fput(struct file *file)
482 if (atomic_long_dec_and_test(&file->f_count)) {
483 struct task_struct *task = current;
485 if (unlikely(!(file->f_mode & (FMODE_BACKING | FMODE_OPENED)))) {
489 if (likely(!in_interrupt() && !(task->flags & PF_KTHREAD))) {
490 init_task_work(&file->f_task_work, ____fput);
491 if (!task_work_add(task, &file->f_task_work, TWA_RESUME))
494 * After this task has run exit_task_work(),
495 * task_work_add() will fail. Fall through to delayed
496 * fput to avoid leaking *file.
500 if (llist_add(&file->f_llist, &delayed_fput_list))
501 schedule_delayed_work(&delayed_fput_work, 1);
506 * synchronous analog of fput(); for kernel threads that might be needed
507 * in some umount() (and thus can't use flush_delayed_fput() without
508 * risking deadlocks), need to wait for completion of __fput() and know
509 * for this specific struct file it won't involve anything that would
510 * need them. Use only if you really need it - at the very least,
511 * don't blindly convert fput() by kernel thread to that.
513 void __fput_sync(struct file *file)
515 if (atomic_long_dec_and_test(&file->f_count))
520 EXPORT_SYMBOL(__fput_sync);
522 void __init files_init(void)
524 struct kmem_cache_args args = {
525 .use_freeptr_offset = true,
526 .freeptr_offset = offsetof(struct file, f_freeptr),
529 filp_cachep = kmem_cache_create("filp", sizeof(struct file), &args,
530 SLAB_HWCACHE_ALIGN | SLAB_PANIC |
531 SLAB_ACCOUNT | SLAB_TYPESAFE_BY_RCU);
532 percpu_counter_init(&nr_files, 0, GFP_KERNEL);
536 * One file with associated inode and dcache is very roughly 1K. Per default
537 * do not use more than 10% of our memory for files.
539 void __init files_maxfiles_init(void)
542 unsigned long nr_pages = totalram_pages();
543 unsigned long memreserve = (nr_pages - nr_free_pages()) * 3/2;
545 memreserve = min(memreserve, nr_pages - 1);
546 n = ((nr_pages - memreserve) * (PAGE_SIZE / 1024)) / 10;
548 files_stat.max_files = max_t(unsigned long, n, NR_FILE);