projects/linux·blob master · refs

668 lines · 16.8 KiB

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 ([email protected])
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 <asm/runtime-const.h>
34
35#include "internal.h"
36
37/* sysctl tunables... */
38static struct files_stat_struct files_stat = {
39 .max_files = NR_FILE
40};
41
42/* SLAB cache for file structures */
43static struct kmem_cache *__filp_cache __ro_after_init;
44#define filp_cache runtime_const_ptr(__filp_cache)
45static struct kmem_cache *__bfilp_cache __ro_after_init;
46#define bfilp_cache runtime_const_ptr(__bfilp_cache)
47
48static struct percpu_counter nr_files __cacheline_aligned_in_smp;
49
50/* Container for backing file with optional user path */
51struct backing_file {
52 struct file file;
53 union {
54 struct path user_path;
55 freeptr_t bf_freeptr;
56 };
57#ifdef CONFIG_SECURITY
58 void *security;
59#endif
60};
61
62#define backing_file(f) container_of(f, struct backing_file, file)
63
64const struct path *backing_file_user_path(const struct file *f)
65{
66 return &backing_file(f)->user_path;
67}
68EXPORT_SYMBOL_GPL(backing_file_user_path);
69
70void backing_file_set_user_path(struct file *f, const struct path *path)
71{
72 backing_file(f)->user_path = *path;
73}
74EXPORT_SYMBOL_GPL(backing_file_set_user_path);
75
76#ifdef CONFIG_SECURITY
77void *backing_file_security(const struct file *f)
78{
79 return backing_file(f)->security;
80}
81
82void backing_file_set_security(struct file *f, void *security)
83{
84 backing_file(f)->security = security;
85}
86#endif /* CONFIG_SECURITY */
87
88static inline void backing_file_free(struct backing_file *ff)
89{
90 security_backing_file_free(&ff->file);
91 path_put(&ff->user_path);
92 kmem_cache_free(bfilp_cache, ff);
93}
94
95static inline void file_free(struct file *f)
96{
97 security_file_free(f);
98 if (likely(!(f->f_mode & FMODE_NOACCOUNT)))
99 percpu_counter_dec(&nr_files);
100 put_cred(f->f_cred);
101 if (unlikely(f->f_mode & FMODE_BACKING)) {
102 backing_file_free(backing_file(f));
103 } else {
104 kmem_cache_free(filp_cache, f);
105 }
106}
107
108/*
109 * Return the total number of open files in the system
110 */
111static long get_nr_files(void)
112{
113 return percpu_counter_read_positive(&nr_files);
114}
115
116/*
117 * Return the maximum number of open files in the system
118 */
119unsigned long get_max_files(void)
120{
121 return files_stat.max_files;
122}
123EXPORT_SYMBOL_GPL(get_max_files);
124
125#if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
126
127/*
128 * Handle nr_files sysctl
129 */
130static int proc_nr_files(const struct ctl_table *table, int write, void *buffer,
131 size_t *lenp, loff_t *ppos)
132{
133 files_stat.nr_files = percpu_counter_sum_positive(&nr_files);
134 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
135}
136
137static const struct ctl_table fs_stat_sysctls[] = {
138 {
139 .procname = "file-nr",
140 .data = &files_stat,
141 .maxlen = sizeof(files_stat),
142 .mode = 0444,
143 .proc_handler = proc_nr_files,
144 },
145 {
146 .procname = "file-max",
147 .data = &files_stat.max_files,
148 .maxlen = sizeof(files_stat.max_files),
149 .mode = 0644,
150 .proc_handler = proc_doulongvec_minmax,
151 .extra1 = SYSCTL_LONG_ZERO,
152 .extra2 = SYSCTL_LONG_MAX,
153 },
154 {
155 .procname = "nr_open",
156 .data = &sysctl_nr_open,
157 .maxlen = sizeof(unsigned int),
158 .mode = 0644,
159 .proc_handler = proc_douintvec_minmax,
160 .extra1 = &sysctl_nr_open_min,
161 .extra2 = &sysctl_nr_open_max,
162 },
163};
164
165static int __init init_fs_stat_sysctls(void)
166{
167 register_sysctl_init("fs", fs_stat_sysctls);
168 if (IS_ENABLED(CONFIG_BINFMT_MISC)) {
169 struct ctl_table_header *hdr;
170
171 hdr = register_sysctl_mount_point("fs/binfmt_misc");
172 kmemleak_not_leak(hdr);
173 }
174 return 0;
175}
176fs_initcall(init_fs_stat_sysctls);
177#endif
178
179static int init_file(struct file *f, int flags, const struct cred *cred)
180{
181 int error;
182
183 f->f_cred = get_cred(cred);
184 error = security_file_alloc(f);
185 if (unlikely(error)) {
186 put_cred(f->f_cred);
187 return error;
188 }
189
190 spin_lock_init(&f->f_lock);
191 /*
192 * Note that f_pos_lock is only used for files raising
193 * FMODE_ATOMIC_POS and directories. Other files such as pipes
194 * don't need it and since f_pos_lock is in a union may reuse
195 * the space for other purposes. They are expected to initialize
196 * the respective member when opening the file.
197 */
198 mutex_init(&f->f_pos_lock);
199 memset(&f->__f_path, 0, sizeof(f->f_path));
200 memset(&f->f_ra, 0, sizeof(f->f_ra));
201
202 f->f_flags = flags;
203 f->f_mode = OPEN_FMODE(flags);
204 /*
205 * Disable permission and pre-content events for all files by default.
206 * They may be enabled later by fsnotify_open_perm_and_set_mode().
207 */
208 file_set_fsnotify_mode(f, FMODE_NONOTIFY_PERM);
209
210 f->f_op = NULL;
211 f->f_mapping = NULL;
212 f->private_data = NULL;
213 f->f_inode = NULL;
214 f->f_owner = NULL;
215#ifdef CONFIG_EPOLL
216 f->f_ep = NULL;
217#endif
218
219 f->f_iocb_flags = 0;
220 f->f_pos = 0;
221 f->f_wb_err = 0;
222 f->f_sb_err = 0;
223
224 /*
225 * We're SLAB_TYPESAFE_BY_RCU so initialize f_ref last. While
226 * fget-rcu pattern users need to be able to handle spurious
227 * refcount bumps we should reinitialize the reused file first.
228 */
229 file_ref_init(&f->f_ref, 1);
230 return 0;
231}
232
233/* Find an unused file structure and return a pointer to it.
234 * Returns an error pointer if some error happened, e.g., we exceed the file
235 * structures limit, run out of memory or operation is not permitted.
236 *
237 * Be very careful using this. You are responsible for
238 * getting write access to any mount that you might assign
239 * to this filp, if it is opened for write. If this is not
240 * done, the mount's writer count will be wrong
241 * and a warning at __fput() time.
242 */
243struct file *alloc_empty_file(int flags, const struct cred *cred)
244{
245 static long old_max;
246 struct file *f;
247 int error;
248
249 /*
250 * Privileged users can go above max_files
251 */
252 if (unlikely(get_nr_files() >= files_stat.max_files) &&
253 !capable(CAP_SYS_ADMIN)) {
254 /*
255 * percpu_counters are inaccurate. Do an expensive check before
256 * we go and fail.
257 */
258 if (percpu_counter_sum_positive(&nr_files) >= files_stat.max_files)
259 goto over;
260 }
261
262 f = kmem_cache_alloc(filp_cache, GFP_KERNEL);
263 if (unlikely(!f))
264 return ERR_PTR(-ENOMEM);
265
266 error = init_file(f, flags, cred);
267 if (unlikely(error)) {
268 kmem_cache_free(filp_cache, f);
269 return ERR_PTR(error);
270 }
271
272 percpu_counter_inc(&nr_files);
273
274 return f;
275
276over:
277 /* Ran out of filps - report that */
278 if (get_nr_files() > old_max) {
279 pr_info("VFS: file-max limit %lu reached\n", get_max_files());
280 old_max = get_nr_files();
281 }
282 return ERR_PTR(-ENFILE);
283}
284
285/*
286 * Variant of alloc_empty_file() that doesn't check and modify nr_files.
287 *
288 * This is only for kernel internal use, and the allocate file must not be
289 * installed into file tables or such.
290 */
291struct file *alloc_empty_file_noaccount(int flags, const struct cred *cred)
292{
293 struct file *f;
294 int error;
295
296 f = kmem_cache_alloc(filp_cache, GFP_KERNEL);
297 if (unlikely(!f))
298 return ERR_PTR(-ENOMEM);
299
300 error = init_file(f, flags, cred);
301 if (unlikely(error)) {
302 kmem_cache_free(filp_cache, f);
303 return ERR_PTR(error);
304 }
305
306 f->f_mode |= FMODE_NOACCOUNT;
307
308 return f;
309}
310
311static int init_backing_file(struct backing_file *ff,
312 const struct file *user_file)
313{
314 memset(&ff->user_path, 0, sizeof(ff->user_path));
315 backing_file_set_security(&ff->file, NULL);
316 return security_backing_file_alloc(&ff->file, user_file);
317}
318
319/*
320 * Variant of alloc_empty_file() that allocates a backing_file container
321 * and doesn't check and modify nr_files.
322 *
323 * This is only for kernel internal use, and the allocate file must not be
324 * installed into file tables or such.
325 */
326struct file *alloc_empty_backing_file(int flags, const struct cred *cred,
327 const struct file *user_file)
328{
329 struct backing_file *ff;
330 int error;
331
332 ff = kmem_cache_alloc(bfilp_cache, GFP_KERNEL);
333 if (unlikely(!ff))
334 return ERR_PTR(-ENOMEM);
335
336 error = init_file(&ff->file, flags, cred);
337 if (unlikely(error)) {
338 kmem_cache_free(bfilp_cache, ff);
339 return ERR_PTR(error);
340 }
341
342 /* The f_mode flags must be set before fput(). */
343 ff->file.f_mode |= FMODE_BACKING | FMODE_NOACCOUNT;
344 error = init_backing_file(ff, user_file);
345 if (unlikely(error)) {
346 fput(&ff->file);
347 return ERR_PTR(error);
348 }
349
350 return &ff->file;
351}
352EXPORT_SYMBOL_GPL(alloc_empty_backing_file);
353
354/**
355 * file_init_path - initialize a 'struct file' based on path
356 *
357 * @file: the file to set up
358 * @path: the (dentry, vfsmount) pair for the new file
359 * @fop: the 'struct file_operations' for the new file
360 */
361static void file_init_path(struct file *file, const struct path *path,
362 const struct file_operations *fop)
363{
364 file->__f_path = *path;
365 file->f_inode = path->dentry->d_inode;
366 file->f_mapping = path->dentry->d_inode->i_mapping;
367 file->f_wb_err = filemap_sample_wb_err(file->f_mapping);
368 file->f_sb_err = file_sample_sb_err(file);
369 if (fop->llseek)
370 file->f_mode |= FMODE_LSEEK;
371 if ((file->f_mode & FMODE_READ) &&
372 likely(fop->read || fop->read_iter))
373 file->f_mode |= FMODE_CAN_READ;
374 if ((file->f_mode & FMODE_WRITE) &&
375 likely(fop->write || fop->write_iter))
376 file->f_mode |= FMODE_CAN_WRITE;
377 file->f_iocb_flags = iocb_flags(file);
378 file->f_mode |= FMODE_OPENED;
379 file->f_op = fop;
380 if ((file->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
381 i_readcount_inc(path->dentry->d_inode);
382}
383
384/**
385 * alloc_file - allocate and initialize a 'struct file'
386 *
387 * @path: the (dentry, vfsmount) pair for the new file
388 * @flags: O_... flags with which the new file will be opened
389 * @fop: the 'struct file_operations' for the new file
390 */
391static struct file *alloc_file(const struct path *path, int flags,
392 const struct file_operations *fop)
393{
394 struct file *file;
395
396 file = alloc_empty_file(flags, current_cred());
397 if (!IS_ERR(file))
398 file_init_path(file, path, fop);
399 return file;
400}
401
402static inline int alloc_path_pseudo(const char *name, struct inode *inode,
403 struct vfsmount *mnt, struct path *path)
404{
405 if (WARN_ON_ONCE(S_ISDIR(inode->i_mode)))
406 return -EINVAL;
407 path->dentry = d_alloc_pseudo(mnt->mnt_sb, &QSTR(name));
408 if (!path->dentry)
409 return -ENOMEM;
410 path->mnt = mntget(mnt);
411 d_instantiate(path->dentry, inode);
412 return 0;
413}
414
415struct file *alloc_file_pseudo(struct inode *inode, struct vfsmount *mnt,
416 const char *name, int flags,
417 const struct file_operations *fops)
418{
419 int ret;
420 struct path path;
421 struct file *file;
422
423 ret = alloc_path_pseudo(name, inode, mnt, &path);
424 if (ret)
425 return ERR_PTR(ret);
426
427 file = alloc_file(&path, flags, fops);
428 if (IS_ERR(file)) {
429 ihold(inode);
430 path_put(&path);
431 return file;
432 }
433 /*
434 * Disable all fsnotify events for pseudo files by default.
435 * They may be enabled by caller with file_set_fsnotify_mode().
436 */
437 file_set_fsnotify_mode(file, FMODE_NONOTIFY);
438 return file;
439}
440EXPORT_SYMBOL(alloc_file_pseudo);
441
442struct file *alloc_file_pseudo_noaccount(struct inode *inode,
443 struct vfsmount *mnt, const char *name,
444 int flags,
445 const struct file_operations *fops)
446{
447 int ret;
448 struct path path;
449 struct file *file;
450
451 ret = alloc_path_pseudo(name, inode, mnt, &path);
452 if (ret)
453 return ERR_PTR(ret);
454
455 file = alloc_empty_file_noaccount(flags, current_cred());
456 if (IS_ERR(file)) {
457 ihold(inode);
458 path_put(&path);
459 return file;
460 }
461 file_init_path(file, &path, fops);
462 /*
463 * Disable all fsnotify events for pseudo files by default.
464 * They may be enabled by caller with file_set_fsnotify_mode().
465 */
466 file_set_fsnotify_mode(file, FMODE_NONOTIFY);
467 return file;
468}
469EXPORT_SYMBOL_GPL(alloc_file_pseudo_noaccount);
470
471struct file *alloc_file_clone(struct file *base, int flags,
472 const struct file_operations *fops)
473{
474 struct file *f;
475
476 f = alloc_file(&base->f_path, flags, fops);
477 if (!IS_ERR(f)) {
478 path_get(&f->f_path);
479 f->f_mapping = base->f_mapping;
480 }
481 return f;
482}
483
484/* the real guts of fput() - releasing the last reference to file
485 */
486static void __fput(struct file *file)
487{
488 struct dentry *dentry = file->f_path.dentry;
489 struct vfsmount *mnt = file->f_path.mnt;
490 struct inode *inode = file->f_inode;
491 fmode_t mode = file->f_mode;
492
493 if (unlikely(!(file->f_mode & FMODE_OPENED)))
494 goto out;
495
496 might_sleep();
497
498 fsnotify_close(file);
499 /*
500 * The function eventpoll_release() should be the first called
501 * in the file cleanup chain.
502 */
503 eventpoll_release(file);
504 locks_remove_file(file);
505
506 security_file_release(file);
507 if (unlikely(file->f_flags & FASYNC)) {
508 if (file->f_op->fasync)
509 file->f_op->fasync(-1, file, 0);
510 }
511 if (file->f_op->release)
512 file->f_op->release(inode, file);
513 if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL &&
514 !(mode & FMODE_PATH))) {
515 cdev_put(inode->i_cdev);
516 }
517 fops_put(file->f_op);
518 file_f_owner_release(file);
519 put_file_access(file);
520 dput(dentry);
521 if (unlikely(mode & FMODE_NEED_UNMOUNT))
522 dissolve_on_fput(mnt);
523 else
524 mntput(mnt);
525out:
526 file_free(file);
527}
528
529static LLIST_HEAD(delayed_fput_list);
530static void delayed_fput(struct work_struct *unused)
531{
532 struct llist_node *node = llist_del_all(&delayed_fput_list);
533 struct file *f, *t;
534
535 llist_for_each_entry_safe(f, t, node, f_llist)
536 __fput(f);
537}
538
539static void ____fput(struct callback_head *work)
540{
541 __fput(container_of(work, struct file, f_task_work));
542}
543
544static DECLARE_DELAYED_WORK(delayed_fput_work, delayed_fput);
545
546/*
547 * If kernel thread really needs to have the final fput() it has done
548 * to complete, call this. The only user right now is the boot - we
549 * *do* need to make sure our writes to binaries on initramfs has
550 * not left us with opened struct file waiting for __fput() - execve()
551 * won't work without that. Please, don't add more callers without
552 * very good reasons; in particular, never call that with locks
553 * held and never call that from a thread that might need to do
554 * some work on any kind of umount.
555 */
556void flush_delayed_fput(void)
557{
558 delayed_fput(NULL);
559 flush_delayed_work(&delayed_fput_work);
560}
561EXPORT_SYMBOL_GPL(flush_delayed_fput);
562
563static void __fput_deferred(struct file *file)
564{
565 struct task_struct *task = current;
566
567 if (unlikely(!(file->f_mode & (FMODE_BACKING | FMODE_OPENED)))) {
568 file_free(file);
569 return;
570 }
571
572 if (likely(!in_interrupt() && !(task->flags & PF_KTHREAD))) {
573 init_task_work(&file->f_task_work, ____fput);
574 if (!task_work_add(task, &file->f_task_work, TWA_RESUME))
575 return;
576 /*
577 * After this task has run exit_task_work(),
578 * task_work_add() will fail. Fall through to delayed
579 * fput to avoid leaking *file.
580 */
581 }
582
583 if (llist_add(&file->f_llist, &delayed_fput_list))
584 schedule_delayed_work(&delayed_fput_work, 1);
585}
586
587void fput(struct file *file)
588{
589 if (unlikely(file_ref_put(&file->f_ref)))
590 __fput_deferred(file);
591}
592EXPORT_SYMBOL(fput);
593
594/*
595 * synchronous analog of fput(); for kernel threads that might be needed
596 * in some umount() (and thus can't use flush_delayed_fput() without
597 * risking deadlocks), need to wait for completion of __fput() and know
598 * for this specific struct file it won't involve anything that would
599 * need them. Use only if you really need it - at the very least,
600 * don't blindly convert fput() by kernel thread to that.
601 */
602void __fput_sync(struct file *file)
603{
604 if (file_ref_put(&file->f_ref))
605 __fput(file);
606}
607EXPORT_SYMBOL(__fput_sync);
608
609/*
610 * Equivalent to __fput_sync(), but optimized for being called with the last
611 * reference.
612 *
613 * See file_ref_put_close() for details.
614 */
615void fput_close_sync(struct file *file)
616{
617 if (likely(file_ref_put_close(&file->f_ref)))
618 __fput(file);
619}
620
621/*
622 * Equivalent to fput(), but optimized for being called with the last
623 * reference.
624 *
625 * See file_ref_put_close() for details.
626 */
627void fput_close(struct file *file)
628{
629 if (file_ref_put_close(&file->f_ref))
630 __fput_deferred(file);
631}
632
633void __init files_init(void)
634{
635 struct kmem_cache_args args = {
636 .use_freeptr_offset = true,
637 .freeptr_offset = offsetof(struct file, f_freeptr),
638 };
639
640 __filp_cache = kmem_cache_create("filp", sizeof(struct file), &args,
641 SLAB_HWCACHE_ALIGN | SLAB_PANIC |
642 SLAB_ACCOUNT | SLAB_TYPESAFE_BY_RCU);
643 runtime_const_init(ptr, __filp_cache);
644
645 args.freeptr_offset = offsetof(struct backing_file, bf_freeptr);
646 __bfilp_cache = kmem_cache_create("bfilp", sizeof(struct backing_file),
647 &args, SLAB_HWCACHE_ALIGN | SLAB_PANIC |
648 SLAB_ACCOUNT | SLAB_TYPESAFE_BY_RCU);
649 runtime_const_init(ptr, __bfilp_cache);
650
651 percpu_counter_init(&nr_files, 0, GFP_KERNEL);
652}
653
654/*
655 * One file with associated inode and dcache is very roughly 1K. Per default
656 * do not use more than 10% of our memory for files.
657 */
658void __init files_maxfiles_init(void)
659{
660 unsigned long n;
661 unsigned long nr_pages = totalram_pages();
662 unsigned long memreserve = (nr_pages - nr_free_pages()) * 3/2;
663
664 memreserve = min(memreserve, nr_pages - 1);
665 n = ((nr_pages - memreserve) * (PAGE_SIZE / 1024)) / 10;
666
667 files_stat.max_files = max_t(unsigned long, n, NR_FILE);
668}