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
);
79 * Return the total number of open files in the system
81 static long get_nr_files(void)
83 return percpu_counter_read_positive(&nr_files
);
87 * Return the maximum number of open files in the system
89 unsigned long get_max_files(void)
91 return files_stat
.max_files
;
93 EXPORT_SYMBOL_GPL(get_max_files
);
95 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
98 * Handle nr_files sysctl
100 static int proc_nr_files(struct ctl_table
*table
, int write
, void *buffer
,
101 size_t *lenp
, loff_t
*ppos
)
103 files_stat
.nr_files
= get_nr_files();
104 return proc_doulongvec_minmax(table
, write
, buffer
, lenp
, ppos
);
107 static struct ctl_table fs_stat_sysctls
[] = {
109 .procname
= "file-nr",
111 .maxlen
= sizeof(files_stat
),
113 .proc_handler
= proc_nr_files
,
116 .procname
= "file-max",
117 .data
= &files_stat
.max_files
,
118 .maxlen
= sizeof(files_stat
.max_files
),
120 .proc_handler
= proc_doulongvec_minmax
,
121 .extra1
= SYSCTL_LONG_ZERO
,
122 .extra2
= SYSCTL_LONG_MAX
,
125 .procname
= "nr_open",
126 .data
= &sysctl_nr_open
,
127 .maxlen
= sizeof(unsigned int),
129 .proc_handler
= proc_dointvec_minmax
,
130 .extra1
= &sysctl_nr_open_min
,
131 .extra2
= &sysctl_nr_open_max
,
135 static int __init
init_fs_stat_sysctls(void)
137 register_sysctl_init("fs", fs_stat_sysctls
);
138 if (IS_ENABLED(CONFIG_BINFMT_MISC
)) {
139 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 rwlock_init(&f
->f_owner
.lock
);
160 spin_lock_init(&f
->f_lock
);
161 mutex_init(&f
->f_pos_lock
);
163 f
->f_mode
= OPEN_FMODE(flags
);
164 /* f->f_version: 0 */
167 * We're SLAB_TYPESAFE_BY_RCU so initialize f_count last. While
168 * fget-rcu pattern users need to be able to handle spurious
169 * refcount bumps we should reinitialize the reused file first.
171 atomic_long_set(&f
->f_count
, 1);
175 /* Find an unused file structure and return a pointer to it.
176 * Returns an error pointer if some error happend e.g. we over file
177 * structures limit, run out of memory or operation is not permitted.
179 * Be very careful using this. You are responsible for
180 * getting write access to any mount that you might assign
181 * to this filp, if it is opened for write. If this is not
182 * done, you will imbalance int the mount's writer count
183 * and a warning at __fput() time.
185 struct file
*alloc_empty_file(int flags
, const struct cred
*cred
)
192 * Privileged users can go above max_files
194 if (get_nr_files() >= files_stat
.max_files
&& !capable(CAP_SYS_ADMIN
)) {
196 * percpu_counters are inaccurate. Do an expensive check before
199 if (percpu_counter_sum_positive(&nr_files
) >= files_stat
.max_files
)
203 f
= kmem_cache_zalloc(filp_cachep
, GFP_KERNEL
);
205 return ERR_PTR(-ENOMEM
);
207 error
= init_file(f
, flags
, cred
);
208 if (unlikely(error
)) {
209 kmem_cache_free(filp_cachep
, f
);
210 return ERR_PTR(error
);
213 percpu_counter_inc(&nr_files
);
218 /* Ran out of filps - report that */
219 if (get_nr_files() > old_max
) {
220 pr_info("VFS: file-max limit %lu reached\n", get_max_files());
221 old_max
= get_nr_files();
223 return ERR_PTR(-ENFILE
);
227 * Variant of alloc_empty_file() that doesn't check and modify nr_files.
229 * This is only for kernel internal use, and the allocate file must not be
230 * installed into file tables or such.
232 struct file
*alloc_empty_file_noaccount(int flags
, const struct cred
*cred
)
237 f
= kmem_cache_zalloc(filp_cachep
, GFP_KERNEL
);
239 return ERR_PTR(-ENOMEM
);
241 error
= init_file(f
, flags
, cred
);
242 if (unlikely(error
)) {
243 kmem_cache_free(filp_cachep
, f
);
244 return ERR_PTR(error
);
247 f
->f_mode
|= FMODE_NOACCOUNT
;
253 * Variant of alloc_empty_file() that allocates a backing_file container
254 * and doesn't check and modify nr_files.
256 * This is only for kernel internal use, and the allocate file must not be
257 * installed into file tables or such.
259 struct file
*alloc_empty_backing_file(int flags
, const struct cred
*cred
)
261 struct backing_file
*ff
;
264 ff
= kzalloc(sizeof(struct backing_file
), GFP_KERNEL
);
266 return ERR_PTR(-ENOMEM
);
268 error
= init_file(&ff
->file
, flags
, cred
);
269 if (unlikely(error
)) {
271 return ERR_PTR(error
);
274 ff
->file
.f_mode
|= FMODE_BACKING
| FMODE_NOACCOUNT
;
279 * alloc_file - allocate and initialize a 'struct file'
281 * @path: the (dentry, vfsmount) pair for the new file
282 * @flags: O_... flags with which the new file will be opened
283 * @fop: the 'struct file_operations' for the new file
285 static struct file
*alloc_file(const struct path
*path
, int flags
,
286 const struct file_operations
*fop
)
290 file
= alloc_empty_file(flags
, current_cred());
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 struct file
*alloc_file_pseudo(struct inode
*inode
, struct vfsmount
*mnt
,
316 const char *name
, int flags
,
317 const struct file_operations
*fops
)
319 struct qstr
this = QSTR_INIT(name
, strlen(name
));
323 path
.dentry
= d_alloc_pseudo(mnt
->mnt_sb
, &this);
325 return ERR_PTR(-ENOMEM
);
326 path
.mnt
= mntget(mnt
);
327 d_instantiate(path
.dentry
, inode
);
328 file
= alloc_file(&path
, flags
, fops
);
335 EXPORT_SYMBOL(alloc_file_pseudo
);
337 struct file
*alloc_file_clone(struct file
*base
, int flags
,
338 const struct file_operations
*fops
)
340 struct file
*f
= alloc_file(&base
->f_path
, flags
, fops
);
342 path_get(&f
->f_path
);
343 f
->f_mapping
= base
->f_mapping
;
348 /* the real guts of fput() - releasing the last reference to file
350 static void __fput(struct file
*file
)
352 struct dentry
*dentry
= file
->f_path
.dentry
;
353 struct vfsmount
*mnt
= file
->f_path
.mnt
;
354 struct inode
*inode
= file
->f_inode
;
355 fmode_t mode
= file
->f_mode
;
357 if (unlikely(!(file
->f_mode
& FMODE_OPENED
)))
362 fsnotify_close(file
);
364 * The function eventpoll_release() should be the first called
365 * in the file cleanup chain.
367 eventpoll_release(file
);
368 locks_remove_file(file
);
371 if (unlikely(file
->f_flags
& FASYNC
)) {
372 if (file
->f_op
->fasync
)
373 file
->f_op
->fasync(-1, file
, 0);
375 if (file
->f_op
->release
)
376 file
->f_op
->release(inode
, file
);
377 if (unlikely(S_ISCHR(inode
->i_mode
) && inode
->i_cdev
!= NULL
&&
378 !(mode
& FMODE_PATH
))) {
379 cdev_put(inode
->i_cdev
);
381 fops_put(file
->f_op
);
382 put_pid(file
->f_owner
.pid
);
383 put_file_access(file
);
385 if (unlikely(mode
& FMODE_NEED_UNMOUNT
))
386 dissolve_on_fput(mnt
);
392 static LLIST_HEAD(delayed_fput_list
);
393 static void delayed_fput(struct work_struct
*unused
)
395 struct llist_node
*node
= llist_del_all(&delayed_fput_list
);
398 llist_for_each_entry_safe(f
, t
, node
, f_llist
)
402 static void ____fput(struct callback_head
*work
)
404 __fput(container_of(work
, struct file
, f_task_work
));
408 * If kernel thread really needs to have the final fput() it has done
409 * to complete, call this. The only user right now is the boot - we
410 * *do* need to make sure our writes to binaries on initramfs has
411 * not left us with opened struct file waiting for __fput() - execve()
412 * won't work without that. Please, don't add more callers without
413 * very good reasons; in particular, never call that with locks
414 * held and never call that from a thread that might need to do
415 * some work on any kind of umount.
417 void flush_delayed_fput(void)
421 EXPORT_SYMBOL_GPL(flush_delayed_fput
);
423 static DECLARE_DELAYED_WORK(delayed_fput_work
, delayed_fput
);
425 void fput(struct file
*file
)
427 if (atomic_long_dec_and_test(&file
->f_count
)) {
428 struct task_struct
*task
= current
;
430 if (unlikely(!(file
->f_mode
& (FMODE_BACKING
| FMODE_OPENED
)))) {
434 if (likely(!in_interrupt() && !(task
->flags
& PF_KTHREAD
))) {
435 init_task_work(&file
->f_task_work
, ____fput
);
436 if (!task_work_add(task
, &file
->f_task_work
, TWA_RESUME
))
439 * After this task has run exit_task_work(),
440 * task_work_add() will fail. Fall through to delayed
441 * fput to avoid leaking *file.
445 if (llist_add(&file
->f_llist
, &delayed_fput_list
))
446 schedule_delayed_work(&delayed_fput_work
, 1);
451 * synchronous analog of fput(); for kernel threads that might be needed
452 * in some umount() (and thus can't use flush_delayed_fput() without
453 * risking deadlocks), need to wait for completion of __fput() and know
454 * for this specific struct file it won't involve anything that would
455 * need them. Use only if you really need it - at the very least,
456 * don't blindly convert fput() by kernel thread to that.
458 void __fput_sync(struct file
*file
)
460 if (atomic_long_dec_and_test(&file
->f_count
))
465 EXPORT_SYMBOL(__fput_sync
);
467 void __init
files_init(void)
469 filp_cachep
= kmem_cache_create("filp", sizeof(struct file
), 0,
470 SLAB_TYPESAFE_BY_RCU
| SLAB_HWCACHE_ALIGN
|
471 SLAB_PANIC
| SLAB_ACCOUNT
, NULL
);
472 percpu_counter_init(&nr_files
, 0, GFP_KERNEL
);
476 * One file with associated inode and dcache is very roughly 1K. Per default
477 * do not use more than 10% of our memory for files.
479 void __init
files_maxfiles_init(void)
482 unsigned long nr_pages
= totalram_pages();
483 unsigned long memreserve
= (nr_pages
- nr_free_pages()) * 3/2;
485 memreserve
= min(memreserve
, nr_pages
- 1);
486 n
= ((nr_pages
- memreserve
) * (PAGE_SIZE
/ 1024)) / 10;
488 files_stat
.max_files
= max_t(unsigned long, n
, NR_FILE
);