vfs: Deduplicate code shared by xattr system calls operating on paths
[deliverable/linux.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
DS
54 percpu_counter_dec(&nr_files);
55 call_rcu(&f->f_u.fu_rcuhead, file_free_rcu);
1da177e4
LT
56}
57
529bf6be
DS
58/*
59 * Return the total number of open files in the system
60 */
518de9b3 61static long get_nr_files(void)
1da177e4 62{
529bf6be 63 return percpu_counter_read_positive(&nr_files);
1da177e4
LT
64}
65
529bf6be
DS
66/*
67 * Return the maximum number of open files in the system
68 */
518de9b3 69unsigned long get_max_files(void)
ab2af1f5 70{
529bf6be 71 return files_stat.max_files;
ab2af1f5 72}
529bf6be
DS
73EXPORT_SYMBOL_GPL(get_max_files);
74
75/*
76 * Handle nr_files sysctl
77 */
78#if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
1f7e0616 79int proc_nr_files(struct ctl_table *table, int write,
529bf6be
DS
80 void __user *buffer, size_t *lenp, loff_t *ppos)
81{
82 files_stat.nr_files = get_nr_files();
518de9b3 83 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
529bf6be
DS
84}
85#else
1f7e0616 86int proc_nr_files(struct ctl_table *table, int write,
529bf6be
DS
87 void __user *buffer, size_t *lenp, loff_t *ppos)
88{
89 return -ENOSYS;
90}
91#endif
ab2af1f5 92
1da177e4 93/* Find an unused file structure and return a pointer to it.
1afc99be
AV
94 * Returns an error pointer if some error happend e.g. we over file
95 * structures limit, run out of memory or operation is not permitted.
430e285e
DH
96 *
97 * Be very careful using this. You are responsible for
98 * getting write access to any mount that you might assign
99 * to this filp, if it is opened for write. If this is not
100 * done, you will imbalance int the mount's writer count
101 * and a warning at __fput() time.
1da177e4
LT
102 */
103struct file *get_empty_filp(void)
104{
86a264ab 105 const struct cred *cred = current_cred();
518de9b3 106 static long old_max;
1afc99be
AV
107 struct file *f;
108 int error;
1da177e4
LT
109
110 /*
111 * Privileged users can go above max_files
112 */
529bf6be
DS
113 if (get_nr_files() >= files_stat.max_files && !capable(CAP_SYS_ADMIN)) {
114 /*
115 * percpu_counters are inaccurate. Do an expensive check before
116 * we go and fail.
117 */
52d9f3b4 118 if (percpu_counter_sum_positive(&nr_files) >= files_stat.max_files)
529bf6be
DS
119 goto over;
120 }
af4d2ecb 121
4975e45f 122 f = kmem_cache_zalloc(filp_cachep, GFP_KERNEL);
1afc99be
AV
123 if (unlikely(!f))
124 return ERR_PTR(-ENOMEM);
af4d2ecb 125
529bf6be 126 percpu_counter_inc(&nr_files);
78d29788 127 f->f_cred = get_cred(cred);
1afc99be
AV
128 error = security_file_alloc(f);
129 if (unlikely(error)) {
130 file_free(f);
131 return ERR_PTR(error);
132 }
1da177e4 133
516e0cc5 134 atomic_long_set(&f->f_count, 1);
af4d2ecb 135 rwlock_init(&f->f_owner.lock);
68499914 136 spin_lock_init(&f->f_lock);
9c225f26 137 mutex_init(&f->f_pos_lock);
5a6b7951 138 eventpoll_init_file(f);
af4d2ecb 139 /* f->f_version: 0 */
af4d2ecb
KK
140 return f;
141
142over:
1da177e4 143 /* Ran out of filps - report that */
529bf6be 144 if (get_nr_files() > old_max) {
518de9b3 145 pr_info("VFS: file-max limit %lu reached\n", get_max_files());
529bf6be 146 old_max = get_nr_files();
1da177e4 147 }
1afc99be 148 return ERR_PTR(-ENFILE);
1da177e4
LT
149}
150
ce8d2cdf
DH
151/**
152 * alloc_file - allocate and initialize a 'struct file'
153 * @mnt: the vfsmount on which the file will reside
154 * @dentry: the dentry representing the new file
155 * @mode: the mode with which the new file will be opened
156 * @fop: the 'struct file_operations' for the new file
157 *
158 * Use this instead of get_empty_filp() to get a new
159 * 'struct file'. Do so because of the same initialization
160 * pitfalls reasons listed for init_file(). This is a
161 * preferred interface to using init_file().
162 *
163 * If all the callers of init_file() are eliminated, its
164 * code should be moved into this function.
165 */
2c48b9c4
AV
166struct file *alloc_file(struct path *path, fmode_t mode,
167 const struct file_operations *fop)
ce8d2cdf
DH
168{
169 struct file *file;
ce8d2cdf
DH
170
171 file = get_empty_filp();
1afc99be 172 if (IS_ERR(file))
39b65252 173 return file;
ce8d2cdf 174
2c48b9c4 175 file->f_path = *path;
dd37978c 176 file->f_inode = path->dentry->d_inode;
2c48b9c4 177 file->f_mapping = path->dentry->d_inode->i_mapping;
293bc982
AV
178 if ((mode & FMODE_READ) &&
179 likely(fop->read || fop->aio_read || fop->read_iter))
7f7f25e8 180 mode |= FMODE_CAN_READ;
293bc982
AV
181 if ((mode & FMODE_WRITE) &&
182 likely(fop->write || fop->aio_write || fop->write_iter))
7f7f25e8 183 mode |= FMODE_CAN_WRITE;
ce8d2cdf
DH
184 file->f_mode = mode;
185 file->f_op = fop;
890275b5
MZ
186 if ((mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
187 i_readcount_inc(path->dentry->d_inode);
3d1e4631 188 return file;
ce8d2cdf 189}
73efc468 190EXPORT_SYMBOL(alloc_file);
ce8d2cdf 191
d7065da0 192/* the real guts of fput() - releasing the last reference to file
1da177e4 193 */
d7065da0 194static void __fput(struct file *file)
1da177e4 195{
0f7fc9e4
JJS
196 struct dentry *dentry = file->f_path.dentry;
197 struct vfsmount *mnt = file->f_path.mnt;
c77cecee 198 struct inode *inode = file->f_inode;
1da177e4
LT
199
200 might_sleep();
0eeca283
RL
201
202 fsnotify_close(file);
1da177e4
LT
203 /*
204 * The function eventpoll_release() should be the first called
205 * in the file cleanup chain.
206 */
207 eventpoll_release(file);
78ed8a13 208 locks_remove_file(file);
1da177e4 209
233e70f4 210 if (unlikely(file->f_flags & FASYNC)) {
72c2d531 211 if (file->f_op->fasync)
233e70f4
AV
212 file->f_op->fasync(-1, file, 0);
213 }
4199d35c 214 ima_file_free(file);
72c2d531 215 if (file->f_op->release)
1da177e4
LT
216 file->f_op->release(inode, file);
217 security_file_free(file);
60ed8cf7
MS
218 if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL &&
219 !(file->f_mode & FMODE_PATH))) {
1da177e4 220 cdev_put(inode->i_cdev);
60ed8cf7 221 }
1da177e4 222 fops_put(file->f_op);
609d7fa9 223 put_pid(file->f_owner.pid);
890275b5
MZ
224 if ((file->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
225 i_readcount_dec(inode);
83f936c7
AV
226 if (file->f_mode & FMODE_WRITER) {
227 put_write_access(inode);
228 __mnt_drop_write(mnt);
229 }
0f7fc9e4
JJS
230 file->f_path.dentry = NULL;
231 file->f_path.mnt = NULL;
dd37978c 232 file->f_inode = NULL;
1da177e4
LT
233 file_free(file);
234 dput(dentry);
235 mntput(mnt);
236}
237
4f5e65a1 238static LLIST_HEAD(delayed_fput_list);
4a9d4b02
AV
239static void delayed_fput(struct work_struct *unused)
240{
4f5e65a1
ON
241 struct llist_node *node = llist_del_all(&delayed_fput_list);
242 struct llist_node *next;
243
244 for (; node; node = next) {
245 next = llist_next(node);
246 __fput(llist_entry(node, struct file, f_u.fu_llist));
4a9d4b02
AV
247 }
248}
249
250static void ____fput(struct callback_head *work)
251{
252 __fput(container_of(work, struct file, f_u.fu_rcuhead));
253}
254
255/*
256 * If kernel thread really needs to have the final fput() it has done
257 * to complete, call this. The only user right now is the boot - we
258 * *do* need to make sure our writes to binaries on initramfs has
259 * not left us with opened struct file waiting for __fput() - execve()
260 * won't work without that. Please, don't add more callers without
261 * very good reasons; in particular, never call that with locks
262 * held and never call that from a thread that might need to do
263 * some work on any kind of umount.
264 */
265void flush_delayed_fput(void)
266{
267 delayed_fput(NULL);
268}
269
c7314d74 270static DECLARE_DELAYED_WORK(delayed_fput_work, delayed_fput);
4a9d4b02 271
d7065da0
AV
272void fput(struct file *file)
273{
4a9d4b02
AV
274 if (atomic_long_dec_and_test(&file->f_count)) {
275 struct task_struct *task = current;
e7b2c406 276
e7b2c406
ON
277 if (likely(!in_interrupt() && !(task->flags & PF_KTHREAD))) {
278 init_task_work(&file->f_u.fu_rcuhead, ____fput);
279 if (!task_work_add(task, &file->f_u.fu_rcuhead, true))
280 return;
64372501
AM
281 /*
282 * After this task has run exit_task_work(),
be49b30a 283 * task_work_add() will fail. Fall through to delayed
64372501
AM
284 * fput to avoid leaking *file.
285 */
4a9d4b02 286 }
4f5e65a1
ON
287
288 if (llist_add(&file->f_u.fu_llist, &delayed_fput_list))
c7314d74 289 schedule_delayed_work(&delayed_fput_work, 1);
4a9d4b02
AV
290 }
291}
292
293/*
294 * synchronous analog of fput(); for kernel threads that might be needed
295 * in some umount() (and thus can't use flush_delayed_fput() without
296 * risking deadlocks), need to wait for completion of __fput() and know
297 * for this specific struct file it won't involve anything that would
298 * need them. Use only if you really need it - at the very least,
299 * don't blindly convert fput() by kernel thread to that.
300 */
301void __fput_sync(struct file *file)
302{
303 if (atomic_long_dec_and_test(&file->f_count)) {
304 struct task_struct *task = current;
4a9d4b02 305 BUG_ON(!(task->flags & PF_KTHREAD));
d7065da0 306 __fput(file);
4a9d4b02 307 }
d7065da0
AV
308}
309
310EXPORT_SYMBOL(fput);
311
1da177e4
LT
312void put_filp(struct file *file)
313{
516e0cc5 314 if (atomic_long_dec_and_test(&file->f_count)) {
1da177e4 315 security_file_free(file);
1da177e4
LT
316 file_free(file);
317 }
318}
319
1da177e4
LT
320void __init files_init(unsigned long mempages)
321{
518de9b3 322 unsigned long n;
b6b3fdea
ED
323
324 filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
325 SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
326
327 /*
328 * One file with associated inode and dcache is very roughly 1K.
1da177e4
LT
329 * Per default don't use more than 10% of our memory for files.
330 */
331
332 n = (mempages * (PAGE_SIZE / 1024)) / 10;
518de9b3 333 files_stat.max_files = max_t(unsigned long, n, NR_FILE);
0216bfcf 334 percpu_counter_init(&nr_files, 0);
1da177e4 335}
This page took 0.883114 seconds and 5 git commands to generate.