1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 *  linux/fs/file_table.c
4 *
5 *  Copyright (C) 1991, 1992  Linus Torvalds
6 *  Copyright (C) 1997 David S. Miller (davem@caip.rutgers.edu)
7 */
8
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>
15#include <linux/fs.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>
31
32#include <linux/atomic.h>
33
34#include "internal.h"
35
36/* sysctl tunables... */
37static struct files_stat_struct files_stat = {
38	.max_files = NR_FILE
39};
40
41/* SLAB cache for file structures */
42static struct kmem_cache *filp_cachep __ro_after_init;
43
44static struct percpu_counter nr_files __cacheline_aligned_in_smp;
45
46/* Container for backing file with optional user path */
47struct backing_file {
48	struct file file;
49	struct path user_path;
50};
51
52static inline struct backing_file *backing_file(struct file *f)
53{
54	return container_of(f, struct backing_file, file);
55}
56
57struct path *backing_file_user_path(struct file *f)
58{
59	return &backing_file(f)->user_path;
60}
61EXPORT_SYMBOL_GPL(backing_file_user_path);
62
63static inline void file_free(struct file *f)
64{
65	security_file_free(f);
66	if (likely(!(f->f_mode & FMODE_NOACCOUNT)))
67		percpu_counter_dec(&nr_files);
68	put_cred(f->f_cred);
69	if (unlikely(f->f_mode & FMODE_BACKING)) {
70		path_put(backing_file_user_path(f));
71		kfree(backing_file(f));
72	} else {
73		kmem_cache_free(filp_cachep, f);
74	}
75}
76
77/*
78 * Return the total number of open files in the system
79 */
80static long get_nr_files(void)
81{
82	return percpu_counter_read_positive(&nr_files);
83}
84
85/*
86 * Return the maximum number of open files in the system
87 */
88unsigned long get_max_files(void)
89{
90	return files_stat.max_files;
91}
92EXPORT_SYMBOL_GPL(get_max_files);
93
94#if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
95
96/*
97 * Handle nr_files sysctl
98 */
99static int proc_nr_files(struct ctl_table *table, int write, void *buffer,
100			 size_t *lenp, loff_t *ppos)
101{
102	files_stat.nr_files = get_nr_files();
103	return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
104}
105
106static struct ctl_table fs_stat_sysctls[] = {
107	{
108		.procname	= "file-nr",
109		.data		= &files_stat,
110		.maxlen		= sizeof(files_stat),
111		.mode		= 0444,
112		.proc_handler	= proc_nr_files,
113	},
114	{
115		.procname	= "file-max",
116		.data		= &files_stat.max_files,
117		.maxlen		= sizeof(files_stat.max_files),
118		.mode		= 0644,
119		.proc_handler	= proc_doulongvec_minmax,
120		.extra1		= SYSCTL_LONG_ZERO,
121		.extra2		= SYSCTL_LONG_MAX,
122	},
123	{
124		.procname	= "nr_open",
125		.data		= &sysctl_nr_open,
126		.maxlen		= sizeof(unsigned int),
127		.mode		= 0644,
128		.proc_handler	= proc_dointvec_minmax,
129		.extra1		= &sysctl_nr_open_min,
130		.extra2		= &sysctl_nr_open_max,
131	},
132};
133
134static int __init init_fs_stat_sysctls(void)
135{
136	register_sysctl_init("fs", fs_stat_sysctls);
137	if (IS_ENABLED(CONFIG_BINFMT_MISC)) {
138		struct ctl_table_header *hdr;
139		hdr = register_sysctl_mount_point("fs/binfmt_misc");
140		kmemleak_not_leak(hdr);
141	}
142	return 0;
143}
144fs_initcall(init_fs_stat_sysctls);
145#endif
146
147static int init_file(struct file *f, int flags, const struct cred *cred)
148{
149	int error;
150
151	f->f_cred = get_cred(cred);
152	error = security_file_alloc(f);
153	if (unlikely(error)) {
154		put_cred(f->f_cred);
155		return error;
156	}
157
158	rwlock_init(&f->f_owner.lock);
159	spin_lock_init(&f->f_lock);
160	mutex_init(&f->f_pos_lock);
161	f->f_flags = flags;
162	f->f_mode = OPEN_FMODE(flags);
163	/* f->f_version: 0 */
164
165	/*
166	 * We're SLAB_TYPESAFE_BY_RCU so initialize f_count last. While
167	 * fget-rcu pattern users need to be able to handle spurious
168	 * refcount bumps we should reinitialize the reused file first.
169	 */
170	atomic_long_set(&f->f_count, 1);
171	return 0;
172}
173
174/* Find an unused file structure and return a pointer to it.
175 * Returns an error pointer if some error happend e.g. we over file
176 * structures limit, run out of memory or operation is not permitted.
177 *
178 * Be very careful using this.  You are responsible for
179 * getting write access to any mount that you might assign
180 * to this filp, if it is opened for write.  If this is not
181 * done, you will imbalance int the mount's writer count
182 * and a warning at __fput() time.
183 */
184struct file *alloc_empty_file(int flags, const struct cred *cred)
185{
186	static long old_max;
187	struct file *f;
188	int error;
189
190	/*
191	 * Privileged users can go above max_files
192	 */
193	if (get_nr_files() >= files_stat.max_files && !capable(CAP_SYS_ADMIN)) {
194		/*
195		 * percpu_counters are inaccurate.  Do an expensive check before
196		 * we go and fail.
197		 */
198		if (percpu_counter_sum_positive(&nr_files) >= files_stat.max_files)
199			goto over;
200	}
201
202	f = kmem_cache_zalloc(filp_cachep, GFP_KERNEL);
203	if (unlikely(!f))
204		return ERR_PTR(-ENOMEM);
205
206	error = init_file(f, flags, cred);
207	if (unlikely(error)) {
208		kmem_cache_free(filp_cachep, f);
209		return ERR_PTR(error);
210	}
211
212	percpu_counter_inc(&nr_files);
213
214	return f;
215
216over:
217	/* Ran out of filps - report that */
218	if (get_nr_files() > old_max) {
219		pr_info("VFS: file-max limit %lu reached\n", get_max_files());
220		old_max = get_nr_files();
221	}
222	return ERR_PTR(-ENFILE);
223}
224
225/*
226 * Variant of alloc_empty_file() that doesn't check and modify nr_files.
227 *
228 * This is only for kernel internal use, and the allocate file must not be
229 * installed into file tables or such.
230 */
231struct file *alloc_empty_file_noaccount(int flags, const struct cred *cred)
232{
233	struct file *f;
234	int error;
235
236	f = kmem_cache_zalloc(filp_cachep, GFP_KERNEL);
237	if (unlikely(!f))
238		return ERR_PTR(-ENOMEM);
239
240	error = init_file(f, flags, cred);
241	if (unlikely(error)) {
242		kmem_cache_free(filp_cachep, f);
243		return ERR_PTR(error);
244	}
245
246	f->f_mode |= FMODE_NOACCOUNT;
247
248	return f;
249}
250
251/*
252 * Variant of alloc_empty_file() that allocates a backing_file container
253 * and doesn't check and modify nr_files.
254 *
255 * This is only for kernel internal use, and the allocate file must not be
256 * installed into file tables or such.
257 */
258struct file *alloc_empty_backing_file(int flags, const struct cred *cred)
259{
260	struct backing_file *ff;
261	int error;
262
263	ff = kzalloc(sizeof(struct backing_file), GFP_KERNEL);
264	if (unlikely(!ff))
265		return ERR_PTR(-ENOMEM);
266
267	error = init_file(&ff->file, flags, cred);
268	if (unlikely(error)) {
269		kfree(ff);
270		return ERR_PTR(error);
271	}
272
273	ff->file.f_mode |= FMODE_BACKING | FMODE_NOACCOUNT;
274	return &ff->file;
275}
276
277/**
278 * file_init_path - initialize a 'struct file' based on path
279 *
280 * @file: the file to set up
281 * @path: the (dentry, vfsmount) pair for the new file
282 * @fop: the 'struct file_operations' for the new file
283 */
284static void file_init_path(struct file *file, const struct path *path,
285			   const struct file_operations *fop)
286{
287	file->f_path = *path;
288	file->f_inode = path->dentry->d_inode;
289	file->f_mapping = path->dentry->d_inode->i_mapping;
290	file->f_wb_err = filemap_sample_wb_err(file->f_mapping);
291	file->f_sb_err = file_sample_sb_err(file);
292	if (fop->llseek)
293		file->f_mode |= FMODE_LSEEK;
294	if ((file->f_mode & FMODE_READ) &&
295	     likely(fop->read || fop->read_iter))
296		file->f_mode |= FMODE_CAN_READ;
297	if ((file->f_mode & FMODE_WRITE) &&
298	     likely(fop->write || fop->write_iter))
299		file->f_mode |= FMODE_CAN_WRITE;
300	file->f_iocb_flags = iocb_flags(file);
301	file->f_mode |= FMODE_OPENED;
302	file->f_op = fop;
303	if ((file->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
304		i_readcount_inc(path->dentry->d_inode);
305}
306
307/**
308 * alloc_file - allocate and initialize a 'struct file'
309 *
310 * @path: the (dentry, vfsmount) pair for the new file
311 * @flags: O_... flags with which the new file will be opened
312 * @fop: the 'struct file_operations' for the new file
313 */
314static struct file *alloc_file(const struct path *path, int flags,
315		const struct file_operations *fop)
316{
317	struct file *file;
318
319	file = alloc_empty_file(flags, current_cred());
320	if (!IS_ERR(file))
321		file_init_path(file, path, fop);
322	return file;
323}
324
325static inline int alloc_path_pseudo(const char *name, struct inode *inode,
326				    struct vfsmount *mnt, struct path *path)
327{
328	struct qstr this = QSTR_INIT(name, strlen(name));
329
330	path->dentry = d_alloc_pseudo(mnt->mnt_sb, &this);
331	if (!path->dentry)
332		return -ENOMEM;
333	path->mnt = mntget(mnt);
334	d_instantiate(path->dentry, inode);
335	return 0;
336}
337
338struct file *alloc_file_pseudo(struct inode *inode, struct vfsmount *mnt,
339			       const char *name, int flags,
340			       const struct file_operations *fops)
341{
342	int ret;
343	struct path path;
344	struct file *file;
345
346	ret = alloc_path_pseudo(name, inode, mnt, &path);
347	if (ret)
348		return ERR_PTR(ret);
349
350	file = alloc_file(&path, flags, fops);
351	if (IS_ERR(file)) {
352		ihold(inode);
353		path_put(&path);
354	}
355	return file;
356}
357EXPORT_SYMBOL(alloc_file_pseudo);
358
359struct file *alloc_file_pseudo_noaccount(struct inode *inode,
360					 struct vfsmount *mnt, const char *name,
361					 int flags,
362					 const struct file_operations *fops)
363{
364	int ret;
365	struct path path;
366	struct file *file;
367
368	ret = alloc_path_pseudo(name, inode, mnt, &path);
369	if (ret)
370		return ERR_PTR(ret);
371
372	file = alloc_empty_file_noaccount(flags, current_cred());
373	if (IS_ERR(file)) {
374		ihold(inode);
375		path_put(&path);
376		return file;
377	}
378	file_init_path(file, &path, fops);
379	return file;
380}
381EXPORT_SYMBOL_GPL(alloc_file_pseudo_noaccount);
382
383struct file *alloc_file_clone(struct file *base, int flags,
384				const struct file_operations *fops)
385{
386	struct file *f = alloc_file(&base->f_path, flags, fops);
387	if (!IS_ERR(f)) {
388		path_get(&f->f_path);
389		f->f_mapping = base->f_mapping;
390	}
391	return f;
392}
393
394/* the real guts of fput() - releasing the last reference to file
395 */
396static void __fput(struct file *file)
397{
398	struct dentry *dentry = file->f_path.dentry;
399	struct vfsmount *mnt = file->f_path.mnt;
400	struct inode *inode = file->f_inode;
401	fmode_t mode = file->f_mode;
402
403	if (unlikely(!(file->f_mode & FMODE_OPENED)))
404		goto out;
405
406	might_sleep();
407
408	fsnotify_close(file);
409	/*
410	 * The function eventpoll_release() should be the first called
411	 * in the file cleanup chain.
412	 */
413	eventpoll_release(file);
414	locks_remove_file(file);
415
416	security_file_release(file);
417	if (unlikely(file->f_flags & FASYNC)) {
418		if (file->f_op->fasync)
419			file->f_op->fasync(-1, file, 0);
420	}
421	if (file->f_op->release)
422		file->f_op->release(inode, file);
423	if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL &&
424		     !(mode & FMODE_PATH))) {
425		cdev_put(inode->i_cdev);
426	}
427	fops_put(file->f_op);
428	put_pid(file->f_owner.pid);
429	put_file_access(file);
430	dput(dentry);
431	if (unlikely(mode & FMODE_NEED_UNMOUNT))
432		dissolve_on_fput(mnt);
433	mntput(mnt);
434out:
435	file_free(file);
436}
437
438static LLIST_HEAD(delayed_fput_list);
439static void delayed_fput(struct work_struct *unused)
440{
441	struct llist_node *node = llist_del_all(&delayed_fput_list);
442	struct file *f, *t;
443
444	llist_for_each_entry_safe(f, t, node, f_llist)
445		__fput(f);
446}
447
448static void ____fput(struct callback_head *work)
449{
450	__fput(container_of(work, struct file, f_task_work));
451}
452
453/*
454 * If kernel thread really needs to have the final fput() it has done
455 * to complete, call this.  The only user right now is the boot - we
456 * *do* need to make sure our writes to binaries on initramfs has
457 * not left us with opened struct file waiting for __fput() - execve()
458 * won't work without that.  Please, don't add more callers without
459 * very good reasons; in particular, never call that with locks
460 * held and never call that from a thread that might need to do
461 * some work on any kind of umount.
462 */
463void flush_delayed_fput(void)
464{
465	delayed_fput(NULL);
466}
467EXPORT_SYMBOL_GPL(flush_delayed_fput);
468
469static DECLARE_DELAYED_WORK(delayed_fput_work, delayed_fput);
470
471void fput(struct file *file)
472{
473	if (atomic_long_dec_and_test(&file->f_count)) {
474		struct task_struct *task = current;
475
476		if (unlikely(!(file->f_mode & (FMODE_BACKING | FMODE_OPENED)))) {
477			file_free(file);
478			return;
479		}
480		if (likely(!in_interrupt() && !(task->flags & PF_KTHREAD))) {
481			init_task_work(&file->f_task_work, ____fput);
482			if (!task_work_add(task, &file->f_task_work, TWA_RESUME))
483				return;
484			/*
485			 * After this task has run exit_task_work(),
486			 * task_work_add() will fail.  Fall through to delayed
487			 * fput to avoid leaking *file.
488			 */
489		}
490
491		if (llist_add(&file->f_llist, &delayed_fput_list))
492			schedule_delayed_work(&delayed_fput_work, 1);
493	}
494}
495
496/*
497 * synchronous analog of fput(); for kernel threads that might be needed
498 * in some umount() (and thus can't use flush_delayed_fput() without
499 * risking deadlocks), need to wait for completion of __fput() and know
500 * for this specific struct file it won't involve anything that would
501 * need them.  Use only if you really need it - at the very least,
502 * don't blindly convert fput() by kernel thread to that.
503 */
504void __fput_sync(struct file *file)
505{
506	if (atomic_long_dec_and_test(&file->f_count))
507		__fput(file);
508}
509
510EXPORT_SYMBOL(fput);
511EXPORT_SYMBOL(__fput_sync);
512
513void __init files_init(void)
514{
515	filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
516				SLAB_TYPESAFE_BY_RCU | SLAB_HWCACHE_ALIGN |
517				SLAB_PANIC | SLAB_ACCOUNT, NULL);
518	percpu_counter_init(&nr_files, 0, GFP_KERNEL);
519}
520
521/*
522 * One file with associated inode and dcache is very roughly 1K. Per default
523 * do not use more than 10% of our memory for files.
524 */
525void __init files_maxfiles_init(void)
526{
527	unsigned long n;
528	unsigned long nr_pages = totalram_pages();
529	unsigned long memreserve = (nr_pages - nr_free_pages()) * 3/2;
530
531	memreserve = min(memreserve, nr_pages - 1);
532	n = ((nr_pages - memreserve) * (PAGE_SIZE / 1024)) / 10;
533
534	files_stat.max_files = max_t(unsigned long, n, NR_FILE);
535}
536