file_table.c

  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/init.h>
 13#include <linux/module.h>
 14#include <linux/fs.h>
 15#include <linux/filelock.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/swap.h>
 29#include <linux/kmemleak.h>
 30
 31#include <linux/atomic.h>
 32
 33#include "internal.h"
 34
 35/* sysctl tunables... */
 36static struct files_stat_struct files_stat = {
 37	.max_files = NR_FILE
 38};
 39
 40/* SLAB cache for file structures */
 41static struct kmem_cache *filp_cachep __ro_after_init;
 42static struct kmem_cache *bfilp_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	union {
 50		struct path user_path;
 51		freeptr_t bf_freeptr;
 52	};
 53};
 54
 55#define backing_file(f) container_of(f, struct backing_file, file)
 56
 57const struct path *backing_file_user_path(const struct file *f)
 58{
 59	return &backing_file(f)->user_path;
 60}
 61EXPORT_SYMBOL_GPL(backing_file_user_path);
 62
 63void backing_file_set_user_path(struct file *f, const struct path *path)
 64{
 65	backing_file(f)->user_path = *path;
 66}
 67EXPORT_SYMBOL_GPL(backing_file_set_user_path);
 68
 69static inline void file_free(struct file *f)
 70{
 71	security_file_free(f);
 72	if (likely(!(f->f_mode & FMODE_NOACCOUNT)))
 73		percpu_counter_dec(&nr_files);
 74	put_cred(f->f_cred);
 75	if (unlikely(f->f_mode & FMODE_BACKING)) {
 76		path_put(backing_file_user_path(f));
 77		kmem_cache_free(bfilp_cachep, backing_file(f));
 78	} else {
 79		kmem_cache_free(filp_cachep, f);
 80	}
 81}
 82
 83/*
 84 * Return the total number of open files in the system
 85 */
 86static long get_nr_files(void)
 87{
 88	return percpu_counter_read_positive(&nr_files);
 89}
 90
 91/*
 92 * Return the maximum number of open files in the system
 93 */
 94unsigned long get_max_files(void)
 95{
 96	return files_stat.max_files;
 97}
 98EXPORT_SYMBOL_GPL(get_max_files);
 99
100#if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
101
102/*
103 * Handle nr_files sysctl
104 */
105static int proc_nr_files(const struct ctl_table *table, int write, void *buffer,
106			 size_t *lenp, loff_t *ppos)
107{
108	files_stat.nr_files = percpu_counter_sum_positive(&nr_files);
109	return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
110}
111
112static const struct ctl_table fs_stat_sysctls[] = {
113	{
114		.procname	= "file-nr",
115		.data		= &files_stat,
116		.maxlen		= sizeof(files_stat),
117		.mode		= 0444,
118		.proc_handler	= proc_nr_files,
119	},
120	{
121		.procname	= "file-max",
122		.data		= &files_stat.max_files,
123		.maxlen		= sizeof(files_stat.max_files),
124		.mode		= 0644,
125		.proc_handler	= proc_doulongvec_minmax,
126		.extra1		= SYSCTL_LONG_ZERO,
127		.extra2		= SYSCTL_LONG_MAX,
128	},
129	{
130		.procname	= "nr_open",
131		.data		= &sysctl_nr_open,
132		.maxlen		= sizeof(unsigned int),
133		.mode		= 0644,
134		.proc_handler	= proc_douintvec_minmax,
135		.extra1		= &sysctl_nr_open_min,
136		.extra2		= &sysctl_nr_open_max,
137	},
138};
139
140static int __init init_fs_stat_sysctls(void)
141{
142	register_sysctl_init("fs", fs_stat_sysctls);
143	if (IS_ENABLED(CONFIG_BINFMT_MISC)) {
144		struct ctl_table_header *hdr;
145
146		hdr = register_sysctl_mount_point("fs/binfmt_misc");
147		kmemleak_not_leak(hdr);
148	}
149	return 0;
150}
151fs_initcall(init_fs_stat_sysctls);
152#endif
153
154static int init_file(struct file *f, int flags, const struct cred *cred)
155{
156	int error;
157
158	f->f_cred = get_cred(cred);
159	error = security_file_alloc(f);
160	if (unlikely(error)) {
161		put_cred(f->f_cred);
162		return error;
163	}
164
165	spin_lock_init(&f->f_lock);
166	/*
167	 * Note that f_pos_lock is only used for files raising
168	 * FMODE_ATOMIC_POS and directories. Other files such as pipes
169	 * don't need it and since f_pos_lock is in a union may reuse
170	 * the space for other purposes. They are expected to initialize
171	 * the respective member when opening the file.
172	 */
173	mutex_init(&f->f_pos_lock);
174	memset(&f->__f_path, 0, sizeof(f->f_path));
175	memset(&f->f_ra, 0, sizeof(f->f_ra));
176
177	f->f_flags	= flags;
178	f->f_mode	= OPEN_FMODE(flags);
179	/*
180	 * Disable permission and pre-content events for all files by default.
181	 * They may be enabled later by fsnotify_open_perm_and_set_mode().
182	 */
183	file_set_fsnotify_mode(f, FMODE_NONOTIFY_PERM);
184
185	f->f_op		= NULL;
186	f->f_mapping	= NULL;
187	f->private_data = NULL;
188	f->f_inode	= NULL;
189	f->f_owner	= NULL;
190#ifdef CONFIG_EPOLL
191	f->f_ep		= NULL;
192#endif
193
194	f->f_iocb_flags = 0;
195	f->f_pos	= 0;
196	f->f_wb_err	= 0;
197	f->f_sb_err	= 0;
198
199	/*
200	 * We're SLAB_TYPESAFE_BY_RCU so initialize f_ref last. While
201	 * fget-rcu pattern users need to be able to handle spurious
202	 * refcount bumps we should reinitialize the reused file first.
203	 */
204	file_ref_init(&f->f_ref, 1);
205	return 0;
206}
207
208/* Find an unused file structure and return a pointer to it.
209 * Returns an error pointer if some error happend e.g. we over file
210 * structures limit, run out of memory or operation is not permitted.
211 *
212 * Be very careful using this.  You are responsible for
213 * getting write access to any mount that you might assign
214 * to this filp, if it is opened for write.  If this is not
215 * done, you will imbalance int the mount's writer count
216 * and a warning at __fput() time.
217 */
218struct file *alloc_empty_file(int flags, const struct cred *cred)
219{
220	static long old_max;
221	struct file *f;
222	int error;
223
224	/*
225	 * Privileged users can go above max_files
226	 */
227	if (unlikely(get_nr_files() >= files_stat.max_files) &&
228	    !capable(CAP_SYS_ADMIN)) {
229		/*
230		 * percpu_counters are inaccurate.  Do an expensive check before
231		 * we go and fail.
232		 */
233		if (percpu_counter_sum_positive(&nr_files) >= files_stat.max_files)
234			goto over;
235	}
236
237	f = kmem_cache_alloc(filp_cachep, GFP_KERNEL);
238	if (unlikely(!f))
239		return ERR_PTR(-ENOMEM);
240
241	error = init_file(f, flags, cred);
242	if (unlikely(error)) {
243		kmem_cache_free(filp_cachep, f);
244		return ERR_PTR(error);
245	}
246
247	percpu_counter_inc(&nr_files);
248
249	return f;
250
251over:
252	/* Ran out of filps - report that */
253	if (get_nr_files() > old_max) {
254		pr_info("VFS: file-max limit %lu reached\n", get_max_files());
255		old_max = get_nr_files();
256	}
257	return ERR_PTR(-ENFILE);
258}
259
260/*
261 * Variant of alloc_empty_file() that doesn't check and modify nr_files.
262 *
263 * This is only for kernel internal use, and the allocate file must not be
264 * installed into file tables or such.
265 */
266struct file *alloc_empty_file_noaccount(int flags, const struct cred *cred)
267{
268	struct file *f;
269	int error;
270
271	f = kmem_cache_alloc(filp_cachep, GFP_KERNEL);
272	if (unlikely(!f))
273		return ERR_PTR(-ENOMEM);
274
275	error = init_file(f, flags, cred);
276	if (unlikely(error)) {
277		kmem_cache_free(filp_cachep, f);
278		return ERR_PTR(error);
279	}
280
281	f->f_mode |= FMODE_NOACCOUNT;
282
283	return f;
284}
285
286/*
287 * Variant of alloc_empty_file() that allocates a backing_file container
288 * and doesn't check and modify nr_files.
289 *
290 * This is only for kernel internal use, and the allocate file must not be
291 * installed into file tables or such.
292 */
293struct file *alloc_empty_backing_file(int flags, const struct cred *cred)
294{
295	struct backing_file *ff;
296	int error;
297
298	ff = kmem_cache_alloc(bfilp_cachep, GFP_KERNEL);
299	if (unlikely(!ff))
300		return ERR_PTR(-ENOMEM);
301
302	error = init_file(&ff->file, flags, cred);
303	if (unlikely(error)) {
304		kmem_cache_free(bfilp_cachep, ff);
305		return ERR_PTR(error);
306	}
307
308	ff->file.f_mode |= FMODE_BACKING | FMODE_NOACCOUNT;
309	return &ff->file;
310}
311
312/**
313 * file_init_path - initialize a 'struct file' based on path
314 *
315 * @file: the file to set up
316 * @path: the (dentry, vfsmount) pair for the new file
317 * @fop: the 'struct file_operations' for the new file
318 */
319static void file_init_path(struct file *file, const struct path *path,
320			   const struct file_operations *fop)
321{
322	file->__f_path = *path;
323	file->f_inode = path->dentry->d_inode;
324	file->f_mapping = path->dentry->d_inode->i_mapping;
325	file->f_wb_err = filemap_sample_wb_err(file->f_mapping);
326	file->f_sb_err = file_sample_sb_err(file);
327	if (fop->llseek)
328		file->f_mode |= FMODE_LSEEK;
329	if ((file->f_mode & FMODE_READ) &&
330	     likely(fop->read || fop->read_iter))
331		file->f_mode |= FMODE_CAN_READ;
332	if ((file->f_mode & FMODE_WRITE) &&
333	     likely(fop->write || fop->write_iter))
334		file->f_mode |= FMODE_CAN_WRITE;
335	file->f_iocb_flags = iocb_flags(file);
336	file->f_mode |= FMODE_OPENED;
337	file->f_op = fop;
338	if ((file->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
339		i_readcount_inc(path->dentry->d_inode);
340}
341
342/**
343 * alloc_file - allocate and initialize a 'struct file'
344 *
345 * @path: the (dentry, vfsmount) pair for the new file
346 * @flags: O_... flags with which the new file will be opened
347 * @fop: the 'struct file_operations' for the new file
348 */
349static struct file *alloc_file(const struct path *path, int flags,
350		const struct file_operations *fop)
351{
352	struct file *file;
353
354	file = alloc_empty_file(flags, current_cred());
355	if (!IS_ERR(file))
356		file_init_path(file, path, fop);
357	return file;
358}
359
360static inline int alloc_path_pseudo(const char *name, struct inode *inode,
361				    struct vfsmount *mnt, struct path *path)
362{
363	path->dentry = d_alloc_pseudo(mnt->mnt_sb, &QSTR(name));
364	if (!path->dentry)
365		return -ENOMEM;
366	path->mnt = mntget(mnt);
367	d_instantiate(path->dentry, inode);
368	return 0;
369}
370
371struct file *alloc_file_pseudo(struct inode *inode, struct vfsmount *mnt,
372			       const char *name, int flags,
373			       const struct file_operations *fops)
374{
375	int ret;
376	struct path path;
377	struct file *file;
378
379	ret = alloc_path_pseudo(name, inode, mnt, &path);
380	if (ret)
381		return ERR_PTR(ret);
382
383	file = alloc_file(&path, flags, fops);
384	if (IS_ERR(file)) {
385		ihold(inode);
386		path_put(&path);
387		return file;
388	}
389	/*
390	 * Disable all fsnotify events for pseudo files by default.
391	 * They may be enabled by caller with file_set_fsnotify_mode().
392	 */
393	file_set_fsnotify_mode(file, FMODE_NONOTIFY);
394	return file;
395}
396EXPORT_SYMBOL(alloc_file_pseudo);
397
398struct file *alloc_file_pseudo_noaccount(struct inode *inode,
399					 struct vfsmount *mnt, const char *name,
400					 int flags,
401					 const struct file_operations *fops)
402{
403	int ret;
404	struct path path;
405	struct file *file;
406
407	ret = alloc_path_pseudo(name, inode, mnt, &path);
408	if (ret)
409		return ERR_PTR(ret);
410
411	file = alloc_empty_file_noaccount(flags, current_cred());
412	if (IS_ERR(file)) {
413		ihold(inode);
414		path_put(&path);
415		return file;
416	}
417	file_init_path(file, &path, fops);
418	/*
419	 * Disable all fsnotify events for pseudo files by default.
420	 * They may be enabled by caller with file_set_fsnotify_mode().
421	 */
422	file_set_fsnotify_mode(file, FMODE_NONOTIFY);
423	return file;
424}
425EXPORT_SYMBOL_GPL(alloc_file_pseudo_noaccount);
426
427struct file *alloc_file_clone(struct file *base, int flags,
428				const struct file_operations *fops)
429{
430	struct file *f;
431
432	f = alloc_file(&base->f_path, flags, fops);
433	if (!IS_ERR(f)) {
434		path_get(&f->f_path);
435		f->f_mapping = base->f_mapping;
436	}
437	return f;
438}
439
440/* the real guts of fput() - releasing the last reference to file
441 */
442static void __fput(struct file *file)
443{
444	struct dentry *dentry = file->f_path.dentry;
445	struct vfsmount *mnt = file->f_path.mnt;
446	struct inode *inode = file->f_inode;
447	fmode_t mode = file->f_mode;
448
449	if (unlikely(!(file->f_mode & FMODE_OPENED)))
450		goto out;
451
452	might_sleep();
453
454	fsnotify_close(file);
455	/*
456	 * The function eventpoll_release() should be the first called
457	 * in the file cleanup chain.
458	 */
459	eventpoll_release(file);
460	locks_remove_file(file);
461
462	security_file_release(file);
463	if (unlikely(file->f_flags & FASYNC)) {
464		if (file->f_op->fasync)
465			file->f_op->fasync(-1, file, 0);
466	}
467	if (file->f_op->release)
468		file->f_op->release(inode, file);
469	if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL &&
470		     !(mode & FMODE_PATH))) {
471		cdev_put(inode->i_cdev);
472	}
473	fops_put(file->f_op);
474	file_f_owner_release(file);
475	put_file_access(file);
476	dput(dentry);
477	if (unlikely(mode & FMODE_NEED_UNMOUNT))
478		dissolve_on_fput(mnt);
479	mntput(mnt);
480out:
481	file_free(file);
482}
483
484static LLIST_HEAD(delayed_fput_list);
485static void delayed_fput(struct work_struct *unused)
486{
487	struct llist_node *node = llist_del_all(&delayed_fput_list);
488	struct file *f, *t;
489
490	llist_for_each_entry_safe(f, t, node, f_llist)
491		__fput(f);
492}
493
494static void ____fput(struct callback_head *work)
495{
496	__fput(container_of(work, struct file, f_task_work));
497}
498
499static DECLARE_DELAYED_WORK(delayed_fput_work, delayed_fput);
500
501/*
502 * If kernel thread really needs to have the final fput() it has done
503 * to complete, call this.  The only user right now is the boot - we
504 * *do* need to make sure our writes to binaries on initramfs has
505 * not left us with opened struct file waiting for __fput() - execve()
506 * won't work without that.  Please, don't add more callers without
507 * very good reasons; in particular, never call that with locks
508 * held and never call that from a thread that might need to do
509 * some work on any kind of umount.
510 */
511void flush_delayed_fput(void)
512{
513	delayed_fput(NULL);
514	flush_delayed_work(&delayed_fput_work);
515}
516EXPORT_SYMBOL_GPL(flush_delayed_fput);
517
518static void __fput_deferred(struct file *file)
519{
520	struct task_struct *task = current;
521
522	if (unlikely(!(file->f_mode & (FMODE_BACKING | FMODE_OPENED)))) {
523		file_free(file);
524		return;
525	}
526
527	if (likely(!in_interrupt() && !(task->flags & PF_KTHREAD))) {
528		init_task_work(&file->f_task_work, ____fput);
529		if (!task_work_add(task, &file->f_task_work, TWA_RESUME))
530			return;
531		/*
532		 * After this task has run exit_task_work(),
533		 * task_work_add() will fail.  Fall through to delayed
534		 * fput to avoid leaking *file.
535		 */
536	}
537
538	if (llist_add(&file->f_llist, &delayed_fput_list))
539		schedule_delayed_work(&delayed_fput_work, 1);
540}
541
542void fput(struct file *file)
543{
544	if (unlikely(file_ref_put(&file->f_ref)))
545		__fput_deferred(file);
546}
547EXPORT_SYMBOL(fput);
548
549/*
550 * synchronous analog of fput(); for kernel threads that might be needed
551 * in some umount() (and thus can't use flush_delayed_fput() without
552 * risking deadlocks), need to wait for completion of __fput() and know
553 * for this specific struct file it won't involve anything that would
554 * need them.  Use only if you really need it - at the very least,
555 * don't blindly convert fput() by kernel thread to that.
556 */
557void __fput_sync(struct file *file)
558{
559	if (file_ref_put(&file->f_ref))
560		__fput(file);
561}
562EXPORT_SYMBOL(__fput_sync);
563
564/*
565 * Equivalent to __fput_sync(), but optimized for being called with the last
566 * reference.
567 *
568 * See file_ref_put_close() for details.
569 */
570void fput_close_sync(struct file *file)
571{
572	if (likely(file_ref_put_close(&file->f_ref)))
573		__fput(file);
574}
575
576/*
577 * Equivalent to fput(), but optimized for being called with the last
578 * reference.
579 *
580 * See file_ref_put_close() for details.
581 */
582void fput_close(struct file *file)
583{
584	if (file_ref_put_close(&file->f_ref))
585		__fput_deferred(file);
586}
587
588void __init files_init(void)
589{
590	struct kmem_cache_args args = {
591		.use_freeptr_offset = true,
592		.freeptr_offset = offsetof(struct file, f_freeptr),
593	};
594
595	filp_cachep = kmem_cache_create("filp", sizeof(struct file), &args,
596				SLAB_HWCACHE_ALIGN | SLAB_PANIC |
597				SLAB_ACCOUNT | SLAB_TYPESAFE_BY_RCU);
598
599	args.freeptr_offset = offsetof(struct backing_file, bf_freeptr);
600	bfilp_cachep = kmem_cache_create("bfilp", sizeof(struct backing_file),
601				&args, SLAB_HWCACHE_ALIGN | SLAB_PANIC |
602				SLAB_ACCOUNT | SLAB_TYPESAFE_BY_RCU);
603	percpu_counter_init(&nr_files, 0, GFP_KERNEL);
604}
605
606/*
607 * One file with associated inode and dcache is very roughly 1K. Per default
608 * do not use more than 10% of our memory for files.
609 */
610void __init files_maxfiles_init(void)
611{
612	unsigned long n;
613	unsigned long nr_pages = totalram_pages();
614	unsigned long memreserve = (nr_pages - nr_free_pages()) * 3/2;
615
616	memreserve = min(memreserve, nr_pages - 1);
617	n = ((nr_pages - memreserve) * (PAGE_SIZE / 1024)) / 10;
618
619	files_stat.max_files = max_t(unsigned long, n, NR_FILE);
620}