]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - fs/file_table.c
vfs: atomic f_pos accesses as per POSIX
[mirror_ubuntu-bionic-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>
16#include <linux/eventpoll.h>
ab2af1f5 17#include <linux/rcupdate.h>
1da177e4 18#include <linux/mount.h>
16f7e0fe 19#include <linux/capability.h>
1da177e4 20#include <linux/cdev.h>
0eeca283 21#include <linux/fsnotify.h>
529bf6be 22#include <linux/sysctl.h>
6416ccb7 23#include <linux/lglock.h>
529bf6be 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>
529bf6be 29
60063497 30#include <linux/atomic.h>
1da177e4 31
e81e3f4d
EP
32#include "internal.h"
33
1da177e4
LT
34/* sysctl tunables... */
35struct files_stat_struct files_stat = {
36 .max_files = NR_FILE
37};
38
b6b3fdea
ED
39/* SLAB cache for file structures */
40static struct kmem_cache *filp_cachep __read_mostly;
41
529bf6be 42static struct percpu_counter nr_files __cacheline_aligned_in_smp;
1da177e4 43
5c33b183 44static void file_free_rcu(struct rcu_head *head)
1da177e4 45{
d76b0d9b
DH
46 struct file *f = container_of(head, struct file, f_u.fu_rcuhead);
47
48 put_cred(f->f_cred);
529bf6be 49 kmem_cache_free(filp_cachep, f);
1da177e4
LT
50}
51
529bf6be 52static inline void file_free(struct file *f)
1da177e4 53{
529bf6be 54 percpu_counter_dec(&nr_files);
ad775f5a 55 file_check_state(f);
529bf6be 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)
8d65af78 80int proc_nr_files(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
8d65af78 87int proc_nr_files(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
LT
150}
151
ce8d2cdf
DH
152/**
153 * alloc_file - allocate and initialize a 'struct file'
154 * @mnt: the vfsmount on which the file will reside
155 * @dentry: the dentry representing the new file
156 * @mode: the mode with which the new file will be opened
157 * @fop: the 'struct file_operations' for the new file
158 *
159 * Use this instead of get_empty_filp() to get a new
160 * 'struct file'. Do so because of the same initialization
161 * pitfalls reasons listed for init_file(). This is a
162 * preferred interface to using init_file().
163 *
164 * If all the callers of init_file() are eliminated, its
165 * code should be moved into this function.
166 */
2c48b9c4
AV
167struct file *alloc_file(struct path *path, fmode_t mode,
168 const struct file_operations *fop)
ce8d2cdf
DH
169{
170 struct file *file;
ce8d2cdf
DH
171
172 file = get_empty_filp();
1afc99be 173 if (IS_ERR(file))
39b65252 174 return file;
ce8d2cdf 175
2c48b9c4 176 file->f_path = *path;
dd37978c 177 file->f_inode = path->dentry->d_inode;
2c48b9c4 178 file->f_mapping = path->dentry->d_inode->i_mapping;
ce8d2cdf
DH
179 file->f_mode = mode;
180 file->f_op = fop;
4a3fd211
DH
181
182 /*
183 * These mounts don't really matter in practice
184 * for r/o bind mounts. They aren't userspace-
185 * visible. We do this for consistency, and so
186 * that we can do debugging checks at __fput()
187 */
2c48b9c4 188 if ((mode & FMODE_WRITE) && !special_file(path->dentry->d_inode->i_mode)) {
ad775f5a 189 file_take_write(file);
385e3ed4 190 WARN_ON(mnt_clone_write(path->mnt));
4a3fd211 191 }
890275b5
MZ
192 if ((mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
193 i_readcount_inc(path->dentry->d_inode);
3d1e4631 194 return file;
ce8d2cdf 195}
73efc468 196EXPORT_SYMBOL(alloc_file);
ce8d2cdf 197
aceaf78d
DH
198/**
199 * drop_file_write_access - give up ability to write to a file
200 * @file: the file to which we will stop writing
201 *
202 * This is a central place which will give up the ability
203 * to write to @file, along with access to write through
204 * its vfsmount.
205 */
b57ce969 206static void drop_file_write_access(struct file *file)
aceaf78d 207{
4a3fd211 208 struct vfsmount *mnt = file->f_path.mnt;
aceaf78d
DH
209 struct dentry *dentry = file->f_path.dentry;
210 struct inode *inode = dentry->d_inode;
211
212 put_write_access(inode);
ad775f5a
DH
213
214 if (special_file(inode->i_mode))
215 return;
216 if (file_check_writeable(file) != 0)
217 return;
eb04c282 218 __mnt_drop_write(mnt);
ad775f5a 219 file_release_write(file);
aceaf78d 220}
aceaf78d 221
d7065da0 222/* the real guts of fput() - releasing the last reference to file
1da177e4 223 */
d7065da0 224static void __fput(struct file *file)
1da177e4 225{
0f7fc9e4
JJS
226 struct dentry *dentry = file->f_path.dentry;
227 struct vfsmount *mnt = file->f_path.mnt;
c77cecee 228 struct inode *inode = file->f_inode;
1da177e4
LT
229
230 might_sleep();
0eeca283
RL
231
232 fsnotify_close(file);
1da177e4
LT
233 /*
234 * The function eventpoll_release() should be the first called
235 * in the file cleanup chain.
236 */
237 eventpoll_release(file);
238 locks_remove_flock(file);
239
233e70f4 240 if (unlikely(file->f_flags & FASYNC)) {
72c2d531 241 if (file->f_op->fasync)
233e70f4
AV
242 file->f_op->fasync(-1, file, 0);
243 }
4199d35c 244 ima_file_free(file);
72c2d531 245 if (file->f_op->release)
1da177e4
LT
246 file->f_op->release(inode, file);
247 security_file_free(file);
60ed8cf7
MS
248 if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL &&
249 !(file->f_mode & FMODE_PATH))) {
1da177e4 250 cdev_put(inode->i_cdev);
60ed8cf7 251 }
1da177e4 252 fops_put(file->f_op);
609d7fa9 253 put_pid(file->f_owner.pid);
890275b5
MZ
254 if ((file->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
255 i_readcount_dec(inode);
aceaf78d
DH
256 if (file->f_mode & FMODE_WRITE)
257 drop_file_write_access(file);
0f7fc9e4
JJS
258 file->f_path.dentry = NULL;
259 file->f_path.mnt = NULL;
dd37978c 260 file->f_inode = NULL;
1da177e4
LT
261 file_free(file);
262 dput(dentry);
263 mntput(mnt);
264}
265
4f5e65a1 266static LLIST_HEAD(delayed_fput_list);
4a9d4b02
AV
267static void delayed_fput(struct work_struct *unused)
268{
4f5e65a1
ON
269 struct llist_node *node = llist_del_all(&delayed_fput_list);
270 struct llist_node *next;
271
272 for (; node; node = next) {
273 next = llist_next(node);
274 __fput(llist_entry(node, struct file, f_u.fu_llist));
4a9d4b02
AV
275 }
276}
277
278static void ____fput(struct callback_head *work)
279{
280 __fput(container_of(work, struct file, f_u.fu_rcuhead));
281}
282
283/*
284 * If kernel thread really needs to have the final fput() it has done
285 * to complete, call this. The only user right now is the boot - we
286 * *do* need to make sure our writes to binaries on initramfs has
287 * not left us with opened struct file waiting for __fput() - execve()
288 * won't work without that. Please, don't add more callers without
289 * very good reasons; in particular, never call that with locks
290 * held and never call that from a thread that might need to do
291 * some work on any kind of umount.
292 */
293void flush_delayed_fput(void)
294{
295 delayed_fput(NULL);
296}
297
c7314d74 298static DECLARE_DELAYED_WORK(delayed_fput_work, delayed_fput);
4a9d4b02 299
d7065da0
AV
300void fput(struct file *file)
301{
4a9d4b02
AV
302 if (atomic_long_dec_and_test(&file->f_count)) {
303 struct task_struct *task = current;
e7b2c406 304
e7b2c406
ON
305 if (likely(!in_interrupt() && !(task->flags & PF_KTHREAD))) {
306 init_task_work(&file->f_u.fu_rcuhead, ____fput);
307 if (!task_work_add(task, &file->f_u.fu_rcuhead, true))
308 return;
64372501
AM
309 /*
310 * After this task has run exit_task_work(),
be49b30a 311 * task_work_add() will fail. Fall through to delayed
64372501
AM
312 * fput to avoid leaking *file.
313 */
4a9d4b02 314 }
4f5e65a1
ON
315
316 if (llist_add(&file->f_u.fu_llist, &delayed_fput_list))
c7314d74 317 schedule_delayed_work(&delayed_fput_work, 1);
4a9d4b02
AV
318 }
319}
320
321/*
322 * synchronous analog of fput(); for kernel threads that might be needed
323 * in some umount() (and thus can't use flush_delayed_fput() without
324 * risking deadlocks), need to wait for completion of __fput() and know
325 * for this specific struct file it won't involve anything that would
326 * need them. Use only if you really need it - at the very least,
327 * don't blindly convert fput() by kernel thread to that.
328 */
329void __fput_sync(struct file *file)
330{
331 if (atomic_long_dec_and_test(&file->f_count)) {
332 struct task_struct *task = current;
4a9d4b02 333 BUG_ON(!(task->flags & PF_KTHREAD));
d7065da0 334 __fput(file);
4a9d4b02 335 }
d7065da0
AV
336}
337
338EXPORT_SYMBOL(fput);
339
1da177e4
LT
340void put_filp(struct file *file)
341{
516e0cc5 342 if (atomic_long_dec_and_test(&file->f_count)) {
1da177e4 343 security_file_free(file);
1da177e4
LT
344 file_free(file);
345 }
346}
347
1da177e4
LT
348void __init files_init(unsigned long mempages)
349{
518de9b3 350 unsigned long n;
b6b3fdea
ED
351
352 filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
353 SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
354
355 /*
356 * One file with associated inode and dcache is very roughly 1K.
1da177e4
LT
357 * Per default don't use more than 10% of our memory for files.
358 */
359
360 n = (mempages * (PAGE_SIZE / 1024)) / 10;
518de9b3 361 files_stat.max_files = max_t(unsigned long, n, NR_FILE);
ab2af1f5 362 files_defer_init();
0216bfcf 363 percpu_counter_init(&nr_files, 0);
1da177e4 364}