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/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/ima.h>
30 #include <linux/swap.h>
31 #include <linux/kmemleak.h>
33 #include <linux/atomic.h>
37 /* sysctl tunables... */
38 static struct files_stat_struct files_stat
= {
42 /* SLAB cache for file structures */
43 static struct kmem_cache
*filp_cachep __ro_after_init
;
45 static struct percpu_counter nr_files __cacheline_aligned_in_smp
;
47 /* Container for backing file with optional user path */
50 struct path user_path
;
53 static inline struct backing_file
*backing_file(struct file
*f
)
55 return container_of(f
, struct backing_file
, file
);
58 struct path
*backing_file_user_path(struct file
*f
)
60 return &backing_file(f
)->user_path
;
62 EXPORT_SYMBOL_GPL(backing_file_user_path
);
64 static inline void file_free(struct file
*f
)
66 security_file_free(f
);
67 if (likely(!(f
->f_mode
& FMODE_NOACCOUNT
)))
68 percpu_counter_dec(&nr_files
);
70 if (unlikely(f
->f_mode
& FMODE_BACKING
)) {
71 path_put(backing_file_user_path(f
));
72 kfree(backing_file(f
));
74 kmem_cache_free(filp_cachep
, f
);
78 void release_empty_file(struct file
*f
)
80 WARN_ON_ONCE(f
->f_mode
& (FMODE_BACKING
| FMODE_OPENED
));
81 if (atomic_long_dec_and_test(&f
->f_count
)) {
82 security_file_free(f
);
84 if (likely(!(f
->f_mode
& FMODE_NOACCOUNT
)))
85 percpu_counter_dec(&nr_files
);
86 kmem_cache_free(filp_cachep
, f
);
91 * Return the total number of open files in the system
93 static long get_nr_files(void)
95 return percpu_counter_read_positive(&nr_files
);
99 * Return the maximum number of open files in the system
101 unsigned long get_max_files(void)
103 return files_stat
.max_files
;
105 EXPORT_SYMBOL_GPL(get_max_files
);
107 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
110 * Handle nr_files sysctl
112 static int proc_nr_files(struct ctl_table
*table
, int write
, void *buffer
,
113 size_t *lenp
, loff_t
*ppos
)
115 files_stat
.nr_files
= get_nr_files();
116 return proc_doulongvec_minmax(table
, write
, buffer
, lenp
, ppos
);
119 static struct ctl_table fs_stat_sysctls
[] = {
121 .procname
= "file-nr",
123 .maxlen
= sizeof(files_stat
),
125 .proc_handler
= proc_nr_files
,
128 .procname
= "file-max",
129 .data
= &files_stat
.max_files
,
130 .maxlen
= sizeof(files_stat
.max_files
),
132 .proc_handler
= proc_doulongvec_minmax
,
133 .extra1
= SYSCTL_LONG_ZERO
,
134 .extra2
= SYSCTL_LONG_MAX
,
137 .procname
= "nr_open",
138 .data
= &sysctl_nr_open
,
139 .maxlen
= sizeof(unsigned int),
141 .proc_handler
= proc_dointvec_minmax
,
142 .extra1
= &sysctl_nr_open_min
,
143 .extra2
= &sysctl_nr_open_max
,
148 static int __init
init_fs_stat_sysctls(void)
150 register_sysctl_init("fs", fs_stat_sysctls
);
151 if (IS_ENABLED(CONFIG_BINFMT_MISC
)) {
152 struct ctl_table_header
*hdr
;
153 hdr
= register_sysctl_mount_point("fs/binfmt_misc");
154 kmemleak_not_leak(hdr
);
158 fs_initcall(init_fs_stat_sysctls
);
161 static int init_file(struct file
*f
, int flags
, const struct cred
*cred
)
165 f
->f_cred
= get_cred(cred
);
166 error
= security_file_alloc(f
);
167 if (unlikely(error
)) {
172 rwlock_init(&f
->f_owner
.lock
);
173 spin_lock_init(&f
->f_lock
);
174 mutex_init(&f
->f_pos_lock
);
176 f
->f_mode
= OPEN_FMODE(flags
);
177 /* f->f_version: 0 */
180 * We're SLAB_TYPESAFE_BY_RCU so initialize f_count last. While
181 * fget-rcu pattern users need to be able to handle spurious
182 * refcount bumps we should reinitialize the reused file first.
184 atomic_long_set(&f
->f_count
, 1);
188 /* Find an unused file structure and return a pointer to it.
189 * Returns an error pointer if some error happend e.g. we over file
190 * structures limit, run out of memory or operation is not permitted.
192 * Be very careful using this. You are responsible for
193 * getting write access to any mount that you might assign
194 * to this filp, if it is opened for write. If this is not
195 * done, you will imbalance int the mount's writer count
196 * and a warning at __fput() time.
198 struct file
*alloc_empty_file(int flags
, const struct cred
*cred
)
205 * Privileged users can go above max_files
207 if (get_nr_files() >= files_stat
.max_files
&& !capable(CAP_SYS_ADMIN
)) {
209 * percpu_counters are inaccurate. Do an expensive check before
212 if (percpu_counter_sum_positive(&nr_files
) >= files_stat
.max_files
)
216 f
= kmem_cache_zalloc(filp_cachep
, GFP_KERNEL
);
218 return ERR_PTR(-ENOMEM
);
220 error
= init_file(f
, flags
, cred
);
221 if (unlikely(error
)) {
222 kmem_cache_free(filp_cachep
, f
);
223 return ERR_PTR(error
);
226 percpu_counter_inc(&nr_files
);
231 /* Ran out of filps - report that */
232 if (get_nr_files() > old_max
) {
233 pr_info("VFS: file-max limit %lu reached\n", get_max_files());
234 old_max
= get_nr_files();
236 return ERR_PTR(-ENFILE
);
240 * Variant of alloc_empty_file() that doesn't check and modify nr_files.
242 * This is only for kernel internal use, and the allocate file must not be
243 * installed into file tables or such.
245 struct file
*alloc_empty_file_noaccount(int flags
, const struct cred
*cred
)
250 f
= kmem_cache_zalloc(filp_cachep
, GFP_KERNEL
);
252 return ERR_PTR(-ENOMEM
);
254 error
= init_file(f
, flags
, cred
);
255 if (unlikely(error
)) {
256 kmem_cache_free(filp_cachep
, f
);
257 return ERR_PTR(error
);
260 f
->f_mode
|= FMODE_NOACCOUNT
;
266 * Variant of alloc_empty_file() that allocates a backing_file container
267 * and doesn't check and modify nr_files.
269 * This is only for kernel internal use, and the allocate file must not be
270 * installed into file tables or such.
272 struct file
*alloc_empty_backing_file(int flags
, const struct cred
*cred
)
274 struct backing_file
*ff
;
277 ff
= kzalloc(sizeof(struct backing_file
), GFP_KERNEL
);
279 return ERR_PTR(-ENOMEM
);
281 error
= init_file(&ff
->file
, flags
, cred
);
282 if (unlikely(error
)) {
284 return ERR_PTR(error
);
287 ff
->file
.f_mode
|= FMODE_BACKING
| FMODE_NOACCOUNT
;
292 * alloc_file - allocate and initialize a 'struct file'
294 * @path: the (dentry, vfsmount) pair for the new file
295 * @flags: O_... flags with which the new file will be opened
296 * @fop: the 'struct file_operations' for the new file
298 static struct file
*alloc_file(const struct path
*path
, int flags
,
299 const struct file_operations
*fop
)
303 file
= alloc_empty_file(flags
, current_cred());
307 file
->f_path
= *path
;
308 file
->f_inode
= path
->dentry
->d_inode
;
309 file
->f_mapping
= path
->dentry
->d_inode
->i_mapping
;
310 file
->f_wb_err
= filemap_sample_wb_err(file
->f_mapping
);
311 file
->f_sb_err
= file_sample_sb_err(file
);
313 file
->f_mode
|= FMODE_LSEEK
;
314 if ((file
->f_mode
& FMODE_READ
) &&
315 likely(fop
->read
|| fop
->read_iter
))
316 file
->f_mode
|= FMODE_CAN_READ
;
317 if ((file
->f_mode
& FMODE_WRITE
) &&
318 likely(fop
->write
|| fop
->write_iter
))
319 file
->f_mode
|= FMODE_CAN_WRITE
;
320 file
->f_iocb_flags
= iocb_flags(file
);
321 file
->f_mode
|= FMODE_OPENED
;
323 if ((file
->f_mode
& (FMODE_READ
| FMODE_WRITE
)) == FMODE_READ
)
324 i_readcount_inc(path
->dentry
->d_inode
);
328 struct file
*alloc_file_pseudo(struct inode
*inode
, struct vfsmount
*mnt
,
329 const char *name
, int flags
,
330 const struct file_operations
*fops
)
332 static const struct dentry_operations anon_ops
= {
333 .d_dname
= simple_dname
335 struct qstr
this = QSTR_INIT(name
, strlen(name
));
339 path
.dentry
= d_alloc_pseudo(mnt
->mnt_sb
, &this);
341 return ERR_PTR(-ENOMEM
);
342 if (!mnt
->mnt_sb
->s_d_op
)
343 d_set_d_op(path
.dentry
, &anon_ops
);
344 path
.mnt
= mntget(mnt
);
345 d_instantiate(path
.dentry
, inode
);
346 file
= alloc_file(&path
, flags
, fops
);
353 EXPORT_SYMBOL(alloc_file_pseudo
);
355 struct file
*alloc_file_clone(struct file
*base
, int flags
,
356 const struct file_operations
*fops
)
358 struct file
*f
= alloc_file(&base
->f_path
, flags
, fops
);
360 path_get(&f
->f_path
);
361 f
->f_mapping
= base
->f_mapping
;
366 /* the real guts of fput() - releasing the last reference to file
368 static void __fput(struct file
*file
)
370 struct dentry
*dentry
= file
->f_path
.dentry
;
371 struct vfsmount
*mnt
= file
->f_path
.mnt
;
372 struct inode
*inode
= file
->f_inode
;
373 fmode_t mode
= file
->f_mode
;
375 if (unlikely(!(file
->f_mode
& FMODE_OPENED
)))
380 fsnotify_close(file
);
382 * The function eventpoll_release() should be the first called
383 * in the file cleanup chain.
385 eventpoll_release(file
);
386 locks_remove_file(file
);
389 if (unlikely(file
->f_flags
& FASYNC
)) {
390 if (file
->f_op
->fasync
)
391 file
->f_op
->fasync(-1, file
, 0);
393 if (file
->f_op
->release
)
394 file
->f_op
->release(inode
, file
);
395 if (unlikely(S_ISCHR(inode
->i_mode
) && inode
->i_cdev
!= NULL
&&
396 !(mode
& FMODE_PATH
))) {
397 cdev_put(inode
->i_cdev
);
399 fops_put(file
->f_op
);
400 put_pid(file
->f_owner
.pid
);
401 put_file_access(file
);
403 if (unlikely(mode
& FMODE_NEED_UNMOUNT
))
404 dissolve_on_fput(mnt
);
410 static LLIST_HEAD(delayed_fput_list
);
411 static void delayed_fput(struct work_struct
*unused
)
413 struct llist_node
*node
= llist_del_all(&delayed_fput_list
);
416 llist_for_each_entry_safe(f
, t
, node
, f_llist
)
420 static void ____fput(struct callback_head
*work
)
422 __fput(container_of(work
, struct file
, f_rcuhead
));
426 * If kernel thread really needs to have the final fput() it has done
427 * to complete, call this. The only user right now is the boot - we
428 * *do* need to make sure our writes to binaries on initramfs has
429 * not left us with opened struct file waiting for __fput() - execve()
430 * won't work without that. Please, don't add more callers without
431 * very good reasons; in particular, never call that with locks
432 * held and never call that from a thread that might need to do
433 * some work on any kind of umount.
435 void flush_delayed_fput(void)
439 EXPORT_SYMBOL_GPL(flush_delayed_fput
);
441 static DECLARE_DELAYED_WORK(delayed_fput_work
, delayed_fput
);
443 void fput(struct file
*file
)
445 if (atomic_long_dec_and_test(&file
->f_count
)) {
446 struct task_struct
*task
= current
;
448 if (likely(!in_interrupt() && !(task
->flags
& PF_KTHREAD
))) {
449 init_task_work(&file
->f_rcuhead
, ____fput
);
450 if (!task_work_add(task
, &file
->f_rcuhead
, TWA_RESUME
))
453 * After this task has run exit_task_work(),
454 * task_work_add() will fail. Fall through to delayed
455 * fput to avoid leaking *file.
459 if (llist_add(&file
->f_llist
, &delayed_fput_list
))
460 schedule_delayed_work(&delayed_fput_work
, 1);
465 * synchronous analog of fput(); for kernel threads that might be needed
466 * in some umount() (and thus can't use flush_delayed_fput() without
467 * risking deadlocks), need to wait for completion of __fput() and know
468 * for this specific struct file it won't involve anything that would
469 * need them. Use only if you really need it - at the very least,
470 * don't blindly convert fput() by kernel thread to that.
472 void __fput_sync(struct file
*file
)
474 if (atomic_long_dec_and_test(&file
->f_count
))
479 EXPORT_SYMBOL(__fput_sync
);
481 void __init
files_init(void)
483 filp_cachep
= kmem_cache_create("filp", sizeof(struct file
), 0,
484 SLAB_TYPESAFE_BY_RCU
| SLAB_HWCACHE_ALIGN
|
485 SLAB_PANIC
| SLAB_ACCOUNT
, NULL
);
486 percpu_counter_init(&nr_files
, 0, GFP_KERNEL
);
490 * One file with associated inode and dcache is very roughly 1K. Per default
491 * do not use more than 10% of our memory for files.
493 void __init
files_maxfiles_init(void)
496 unsigned long nr_pages
= totalram_pages();
497 unsigned long memreserve
= (nr_pages
- nr_free_pages()) * 3/2;
499 memreserve
= min(memreserve
, nr_pages
- 1);
500 n
= ((nr_pages
- memreserve
) * (PAGE_SIZE
/ 1024)) / 10;
502 files_stat
.max_files
= max_t(unsigned long, n
, NR_FILE
);