]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blame - fs/file_table.c
UBUNTU: Ubuntu-4.13.0-45.50
[mirror_ubuntu-artful-kernel.git] / fs / file_table.c
CommitLineData
1da177e4
LT
1/*
2 * linux/fs/file_table.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 * Copyright (C) 1997 David S. Miller (davem@caip.rutgers.edu)
6 */
7
8#include <linux/string.h>
9#include <linux/slab.h>
10#include <linux/file.h>
9f3acc31 11#include <linux/fdtable.h>
1da177e4
LT
12#include <linux/init.h>
13#include <linux/module.h>
1da177e4
LT
14#include <linux/fs.h>
15#include <linux/security.h>
5b825c3a 16#include <linux/cred.h>
1da177e4 17#include <linux/eventpoll.h>
ab2af1f5 18#include <linux/rcupdate.h>
1da177e4 19#include <linux/mount.h>
16f7e0fe 20#include <linux/capability.h>
1da177e4 21#include <linux/cdev.h>
0eeca283 22#include <linux/fsnotify.h>
529bf6be
DS
23#include <linux/sysctl.h>
24#include <linux/percpu_counter.h>
6416ccb7 25#include <linux/percpu.h>
4a9d4b02
AV
26#include <linux/hardirq.h>
27#include <linux/task_work.h>
0552f879 28#include <linux/ima.h>
4248b0da 29#include <linux/swap.h>
529bf6be 30
60063497 31#include <linux/atomic.h>
1da177e4 32
e81e3f4d
EP
33#include "internal.h"
34
1da177e4
LT
35/* sysctl tunables... */
36struct files_stat_struct files_stat = {
37 .max_files = NR_FILE
38};
39
b6b3fdea
ED
40/* SLAB cache for file structures */
41static struct kmem_cache *filp_cachep __read_mostly;
42
529bf6be 43static struct percpu_counter nr_files __cacheline_aligned_in_smp;
1da177e4 44
5c33b183 45static void file_free_rcu(struct rcu_head *head)
1da177e4 46{
d76b0d9b
DH
47 struct file *f = container_of(head, struct file, f_u.fu_rcuhead);
48
49 put_cred(f->f_cred);
529bf6be 50 kmem_cache_free(filp_cachep, f);
1da177e4
LT
51}
52
529bf6be 53static inline void file_free(struct file *f)
1da177e4 54{
529bf6be
DS
55 percpu_counter_dec(&nr_files);
56 call_rcu(&f->f_u.fu_rcuhead, file_free_rcu);
1da177e4
LT
57}
58
529bf6be
DS
59/*
60 * Return the total number of open files in the system
61 */
518de9b3 62static long get_nr_files(void)
1da177e4 63{
529bf6be 64 return percpu_counter_read_positive(&nr_files);
1da177e4
LT
65}
66
529bf6be
DS
67/*
68 * Return the maximum number of open files in the system
69 */
518de9b3 70unsigned long get_max_files(void)
ab2af1f5 71{
529bf6be 72 return files_stat.max_files;
ab2af1f5 73}
529bf6be
DS
74EXPORT_SYMBOL_GPL(get_max_files);
75
76/*
77 * Handle nr_files sysctl
78 */
79#if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
1f7e0616 80int proc_nr_files(struct ctl_table *table, int write,
529bf6be
DS
81 void __user *buffer, size_t *lenp, loff_t *ppos)
82{
83 files_stat.nr_files = get_nr_files();
518de9b3 84 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
529bf6be
DS
85}
86#else
1f7e0616 87int proc_nr_files(struct ctl_table *table, int write,
529bf6be
DS
88 void __user *buffer, size_t *lenp, loff_t *ppos)
89{
90 return -ENOSYS;
91}
92#endif
ab2af1f5 93
1da177e4 94/* Find an unused file structure and return a pointer to it.
1afc99be
AV
95 * Returns an error pointer if some error happend e.g. we over file
96 * structures limit, run out of memory or operation is not permitted.
430e285e
DH
97 *
98 * Be very careful using this. You are responsible for
99 * getting write access to any mount that you might assign
100 * to this filp, if it is opened for write. If this is not
101 * done, you will imbalance int the mount's writer count
102 * and a warning at __fput() time.
1da177e4
LT
103 */
104struct file *get_empty_filp(void)
105{
86a264ab 106 const struct cred *cred = current_cred();
518de9b3 107 static long old_max;
1afc99be
AV
108 struct file *f;
109 int error;
1da177e4
LT
110
111 /*
112 * Privileged users can go above max_files
113 */
529bf6be
DS
114 if (get_nr_files() >= files_stat.max_files && !capable(CAP_SYS_ADMIN)) {
115 /*
116 * percpu_counters are inaccurate. Do an expensive check before
117 * we go and fail.
118 */
52d9f3b4 119 if (percpu_counter_sum_positive(&nr_files) >= files_stat.max_files)
529bf6be
DS
120 goto over;
121 }
af4d2ecb 122
4975e45f 123 f = kmem_cache_zalloc(filp_cachep, GFP_KERNEL);
1afc99be
AV
124 if (unlikely(!f))
125 return ERR_PTR(-ENOMEM);
af4d2ecb 126
529bf6be 127 percpu_counter_inc(&nr_files);
78d29788 128 f->f_cred = get_cred(cred);
1afc99be
AV
129 error = security_file_alloc(f);
130 if (unlikely(error)) {
131 file_free(f);
132 return ERR_PTR(error);
133 }
1da177e4 134
516e0cc5 135 atomic_long_set(&f->f_count, 1);
af4d2ecb 136 rwlock_init(&f->f_owner.lock);
68499914 137 spin_lock_init(&f->f_lock);
9c225f26 138 mutex_init(&f->f_pos_lock);
5a6b7951 139 eventpoll_init_file(f);
af4d2ecb 140 /* f->f_version: 0 */
af4d2ecb
KK
141 return f;
142
143over:
1da177e4 144 /* Ran out of filps - report that */
529bf6be 145 if (get_nr_files() > old_max) {
518de9b3 146 pr_info("VFS: file-max limit %lu reached\n", get_max_files());
529bf6be 147 old_max = get_nr_files();
1da177e4 148 }
1afc99be 149 return ERR_PTR(-ENFILE);
1da177e4 150}
b6450630 151EXPORT_SYMBOL_GPL(get_empty_filp);
1da177e4 152
ce8d2cdf
DH
153/**
154 * alloc_file - allocate and initialize a 'struct file'
a457606a
EB
155 *
156 * @path: the (dentry, vfsmount) pair for the new file
ce8d2cdf
DH
157 * @mode: the mode with which the new file will be opened
158 * @fop: the 'struct file_operations' for the new file
ce8d2cdf 159 */
a4141d7c 160struct file *alloc_file(const struct path *path, fmode_t mode,
2c48b9c4 161 const struct file_operations *fop)
ce8d2cdf
DH
162{
163 struct file *file;
ce8d2cdf
DH
164
165 file = get_empty_filp();
1afc99be 166 if (IS_ERR(file))
39b65252 167 return file;
ce8d2cdf 168
2c48b9c4 169 file->f_path = *path;
dd37978c 170 file->f_inode = path->dentry->d_inode;
2c48b9c4 171 file->f_mapping = path->dentry->d_inode->i_mapping;
5660e13d 172 file->f_wb_err = filemap_sample_wb_err(file->f_mapping);
293bc982 173 if ((mode & FMODE_READ) &&
84363182 174 likely(fop->read || fop->read_iter))
7f7f25e8 175 mode |= FMODE_CAN_READ;
293bc982 176 if ((mode & FMODE_WRITE) &&
84363182 177 likely(fop->write || fop->write_iter))
7f7f25e8 178 mode |= FMODE_CAN_WRITE;
ce8d2cdf
DH
179 file->f_mode = mode;
180 file->f_op = fop;
890275b5
MZ
181 if ((mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
182 i_readcount_inc(path->dentry->d_inode);
3d1e4631 183 return file;
ce8d2cdf 184}
73efc468 185EXPORT_SYMBOL(alloc_file);
ce8d2cdf 186
d7065da0 187/* the real guts of fput() - releasing the last reference to file
1da177e4 188 */
d7065da0 189static void __fput(struct file *file)
1da177e4 190{
0f7fc9e4
JJS
191 struct dentry *dentry = file->f_path.dentry;
192 struct vfsmount *mnt = file->f_path.mnt;
c77cecee 193 struct inode *inode = file->f_inode;
1da177e4
LT
194
195 might_sleep();
0eeca283
RL
196
197 fsnotify_close(file);
1da177e4
LT
198 /*
199 * The function eventpoll_release() should be the first called
200 * in the file cleanup chain.
201 */
202 eventpoll_release(file);
78ed8a13 203 locks_remove_file(file);
1da177e4 204
233e70f4 205 if (unlikely(file->f_flags & FASYNC)) {
72c2d531 206 if (file->f_op->fasync)
233e70f4
AV
207 file->f_op->fasync(-1, file, 0);
208 }
4199d35c 209 ima_file_free(file);
72c2d531 210 if (file->f_op->release)
1da177e4
LT
211 file->f_op->release(inode, file);
212 security_file_free(file);
60ed8cf7
MS
213 if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL &&
214 !(file->f_mode & FMODE_PATH))) {
1da177e4 215 cdev_put(inode->i_cdev);
60ed8cf7 216 }
1da177e4 217 fops_put(file->f_op);
609d7fa9 218 put_pid(file->f_owner.pid);
890275b5
MZ
219 if ((file->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
220 i_readcount_dec(inode);
83f936c7
AV
221 if (file->f_mode & FMODE_WRITER) {
222 put_write_access(inode);
223 __mnt_drop_write(mnt);
224 }
0f7fc9e4
JJS
225 file->f_path.dentry = NULL;
226 file->f_path.mnt = NULL;
dd37978c 227 file->f_inode = NULL;
1da177e4
LT
228 file_free(file);
229 dput(dentry);
230 mntput(mnt);
231}
232
4f5e65a1 233static LLIST_HEAD(delayed_fput_list);
4a9d4b02
AV
234static void delayed_fput(struct work_struct *unused)
235{
4f5e65a1
ON
236 struct llist_node *node = llist_del_all(&delayed_fput_list);
237 struct llist_node *next;
238
239 for (; node; node = next) {
240 next = llist_next(node);
241 __fput(llist_entry(node, struct file, f_u.fu_llist));
4a9d4b02
AV
242 }
243}
244
245static void ____fput(struct callback_head *work)
246{
247 __fput(container_of(work, struct file, f_u.fu_rcuhead));
248}
249
250/*
251 * If kernel thread really needs to have the final fput() it has done
252 * to complete, call this. The only user right now is the boot - we
253 * *do* need to make sure our writes to binaries on initramfs has
254 * not left us with opened struct file waiting for __fput() - execve()
255 * won't work without that. Please, don't add more callers without
256 * very good reasons; in particular, never call that with locks
257 * held and never call that from a thread that might need to do
258 * some work on any kind of umount.
259 */
260void flush_delayed_fput(void)
261{
262 delayed_fput(NULL);
263}
b6450630 264EXPORT_SYMBOL_GPL(flush_delayed_fput);
4a9d4b02 265
c7314d74 266static DECLARE_DELAYED_WORK(delayed_fput_work, delayed_fput);
4a9d4b02 267
d7065da0
AV
268void fput(struct file *file)
269{
4a9d4b02
AV
270 if (atomic_long_dec_and_test(&file->f_count)) {
271 struct task_struct *task = current;
e7b2c406 272
e7b2c406
ON
273 if (likely(!in_interrupt() && !(task->flags & PF_KTHREAD))) {
274 init_task_work(&file->f_u.fu_rcuhead, ____fput);
275 if (!task_work_add(task, &file->f_u.fu_rcuhead, true))
276 return;
64372501
AM
277 /*
278 * After this task has run exit_task_work(),
be49b30a 279 * task_work_add() will fail. Fall through to delayed
64372501
AM
280 * fput to avoid leaking *file.
281 */
4a9d4b02 282 }
4f5e65a1
ON
283
284 if (llist_add(&file->f_u.fu_llist, &delayed_fput_list))
c7314d74 285 schedule_delayed_work(&delayed_fput_work, 1);
4a9d4b02
AV
286 }
287}
288
289/*
290 * synchronous analog of fput(); for kernel threads that might be needed
291 * in some umount() (and thus can't use flush_delayed_fput() without
292 * risking deadlocks), need to wait for completion of __fput() and know
293 * for this specific struct file it won't involve anything that would
294 * need them. Use only if you really need it - at the very least,
295 * don't blindly convert fput() by kernel thread to that.
296 */
297void __fput_sync(struct file *file)
298{
299 if (atomic_long_dec_and_test(&file->f_count)) {
300 struct task_struct *task = current;
4a9d4b02 301 BUG_ON(!(task->flags & PF_KTHREAD));
d7065da0 302 __fput(file);
4a9d4b02 303 }
d7065da0
AV
304}
305
306EXPORT_SYMBOL(fput);
b6450630 307EXPORT_SYMBOL_GPL(__fput_sync);
d7065da0 308
1da177e4
LT
309void put_filp(struct file *file)
310{
516e0cc5 311 if (atomic_long_dec_and_test(&file->f_count)) {
1da177e4 312 security_file_free(file);
1da177e4
LT
313 file_free(file);
314 }
315}
b6450630 316EXPORT_SYMBOL_GPL(put_filp);
1da177e4 317
4248b0da 318void __init files_init(void)
1da177e4 319{
b6b3fdea
ED
320 filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
321 SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
4248b0da
MG
322 percpu_counter_init(&nr_files, 0, GFP_KERNEL);
323}
b6b3fdea 324
4248b0da
MG
325/*
326 * One file with associated inode and dcache is very roughly 1K. Per default
327 * do not use more than 10% of our memory for files.
328 */
329void __init files_maxfiles_init(void)
330{
331 unsigned long n;
332 unsigned long memreserve = (totalram_pages - nr_free_pages()) * 3/2;
333
334 memreserve = min(memreserve, totalram_pages - 1);
335 n = ((totalram_pages - memreserve) * (PAGE_SIZE / 1024)) / 10;
1da177e4 336
518de9b3 337 files_stat.max_files = max_t(unsigned long, n, NR_FILE);
1da177e4 338}