blob: 664e3f2309b89811b028603da0f17a6e266ee9fd [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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>
11#include <linux/init.h>
12#include <linux/module.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070013#include <linux/fs.h>
14#include <linux/security.h>
15#include <linux/eventpoll.h>
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070016#include <linux/rcupdate.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/mount.h>
Randy Dunlap16f7e0f2006-01-11 12:17:46 -080018#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070019#include <linux/cdev.h>
Robert Love0eeca282005-07-12 17:06:03 -040020#include <linux/fsnotify.h>
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080021#include <linux/sysctl.h>
22#include <linux/percpu_counter.h>
23
24#include <asm/atomic.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025
26/* sysctl tunables... */
27struct files_stat_struct files_stat = {
28 .max_files = NR_FILE
29};
30
Linus Torvalds1da177e2005-04-16 15:20:36 -070031/* public. Not pretty! */
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080032__cacheline_aligned_in_smp DEFINE_SPINLOCK(files_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070033
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080034static struct percpu_counter nr_files __cacheline_aligned_in_smp;
Linus Torvalds1da177e2005-04-16 15:20:36 -070035
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070036static inline void file_free_rcu(struct rcu_head *head)
37{
Eric Dumazet2f512012005-10-30 15:02:16 -080038 struct file *f = container_of(head, struct file, f_u.fu_rcuhead);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070039 kmem_cache_free(filp_cachep, f);
40}
41
Linus Torvalds1da177e2005-04-16 15:20:36 -070042static inline void file_free(struct file *f)
43{
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080044 percpu_counter_dec(&nr_files);
Eric Dumazet2f512012005-10-30 15:02:16 -080045 call_rcu(&f->f_u.fu_rcuhead, file_free_rcu);
Linus Torvalds1da177e2005-04-16 15:20:36 -070046}
47
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080048/*
49 * Return the total number of open files in the system
50 */
51static int get_nr_files(void)
52{
53 return percpu_counter_read_positive(&nr_files);
54}
55
56/*
57 * Return the maximum number of open files in the system
58 */
59int get_max_files(void)
60{
61 return files_stat.max_files;
62}
63EXPORT_SYMBOL_GPL(get_max_files);
64
65/*
66 * Handle nr_files sysctl
67 */
68#if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
69int proc_nr_files(ctl_table *table, int write, struct file *filp,
70 void __user *buffer, size_t *lenp, loff_t *ppos)
71{
72 files_stat.nr_files = get_nr_files();
73 return proc_dointvec(table, write, filp, buffer, lenp, ppos);
74}
75#else
76int proc_nr_files(ctl_table *table, int write, struct file *filp,
77 void __user *buffer, size_t *lenp, loff_t *ppos)
78{
79 return -ENOSYS;
80}
81#endif
82
Linus Torvalds1da177e2005-04-16 15:20:36 -070083/* Find an unused file structure and return a pointer to it.
84 * Returns NULL, if there are no more free file structures or
85 * we run out of memory.
86 */
87struct file *get_empty_filp(void)
88{
Benjamin LaHaise5a6b7952006-03-23 03:01:03 -080089 struct task_struct *tsk;
Kirill Korotaevaf4d2ec2005-06-23 00:09:50 -070090 static int old_max;
Linus Torvalds1da177e2005-04-16 15:20:36 -070091 struct file * f;
92
93 /*
94 * Privileged users can go above max_files
95 */
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080096 if (get_nr_files() >= files_stat.max_files && !capable(CAP_SYS_ADMIN)) {
97 /*
98 * percpu_counters are inaccurate. Do an expensive check before
99 * we go and fail.
100 */
Peter Zijlstra52d9f3b2007-10-16 23:25:44 -0700101 if (percpu_counter_sum_positive(&nr_files) >= files_stat.max_files)
Dipankar Sarma529bf6b2006-03-07 21:55:35 -0800102 goto over;
103 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104
Denis Cheng4975e452007-10-16 23:26:19 -0700105 f = kmem_cache_zalloc(filp_cachep, GFP_KERNEL);
Kirill Korotaevaf4d2ec2005-06-23 00:09:50 -0700106 if (f == NULL)
107 goto fail;
108
Dipankar Sarma529bf6b2006-03-07 21:55:35 -0800109 percpu_counter_inc(&nr_files);
Kirill Korotaevaf4d2ec2005-06-23 00:09:50 -0700110 if (security_file_alloc(f))
111 goto fail_sec;
112
Benjamin LaHaise5a6b7952006-03-23 03:01:03 -0800113 tsk = current;
Eric Dumazet2f512012005-10-30 15:02:16 -0800114 INIT_LIST_HEAD(&f->f_u.fu_list);
Benjamin LaHaise5a6b7952006-03-23 03:01:03 -0800115 atomic_set(&f->f_count, 1);
116 rwlock_init(&f->f_owner.lock);
117 f->f_uid = tsk->fsuid;
118 f->f_gid = tsk->fsgid;
119 eventpoll_init_file(f);
120 /* f->f_version: 0 */
Kirill Korotaevaf4d2ec2005-06-23 00:09:50 -0700121 return f;
122
123over:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124 /* Ran out of filps - report that */
Dipankar Sarma529bf6b2006-03-07 21:55:35 -0800125 if (get_nr_files() > old_max) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126 printk(KERN_INFO "VFS: file-max limit %d reached\n",
Dipankar Sarma529bf6b2006-03-07 21:55:35 -0800127 get_max_files());
128 old_max = get_nr_files();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129 }
Kirill Korotaevaf4d2ec2005-06-23 00:09:50 -0700130 goto fail;
131
132fail_sec:
133 file_free(f);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134fail:
135 return NULL;
136}
137
138EXPORT_SYMBOL(get_empty_filp);
139
Dave Hansence8d2cd2007-10-16 23:31:13 -0700140/**
141 * alloc_file - allocate and initialize a 'struct file'
142 * @mnt: the vfsmount on which the file will reside
143 * @dentry: the dentry representing the new file
144 * @mode: the mode with which the new file will be opened
145 * @fop: the 'struct file_operations' for the new file
146 *
147 * Use this instead of get_empty_filp() to get a new
148 * 'struct file'. Do so because of the same initialization
149 * pitfalls reasons listed for init_file(). This is a
150 * preferred interface to using init_file().
151 *
152 * If all the callers of init_file() are eliminated, its
153 * code should be moved into this function.
154 */
155struct file *alloc_file(struct vfsmount *mnt, struct dentry *dentry,
156 mode_t mode, const struct file_operations *fop)
157{
158 struct file *file;
159 struct path;
160
161 file = get_empty_filp();
162 if (!file)
163 return NULL;
164
165 init_file(file, mnt, dentry, mode, fop);
166 return file;
167}
168EXPORT_SYMBOL(alloc_file);
169
170/**
171 * init_file - initialize a 'struct file'
172 * @file: the already allocated 'struct file' to initialized
173 * @mnt: the vfsmount on which the file resides
174 * @dentry: the dentry representing this file
175 * @mode: the mode the file is opened with
176 * @fop: the 'struct file_operations' for this file
177 *
178 * Use this instead of setting the members directly. Doing so
179 * avoids making mistakes like forgetting the mntget() or
180 * forgetting to take a write on the mnt.
181 *
182 * Note: This is a crappy interface. It is here to make
183 * merging with the existing users of get_empty_filp()
184 * who have complex failure logic easier. All users
185 * of this should be moving to alloc_file().
186 */
187int init_file(struct file *file, struct vfsmount *mnt, struct dentry *dentry,
188 mode_t mode, const struct file_operations *fop)
189{
190 int error = 0;
191 file->f_path.dentry = dentry;
192 file->f_path.mnt = mntget(mnt);
193 file->f_mapping = dentry->d_inode->i_mapping;
194 file->f_mode = mode;
195 file->f_op = fop;
196 return error;
197}
198EXPORT_SYMBOL(init_file);
199
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200void fastcall fput(struct file *file)
201{
Nick Piggin095975d2006-01-08 01:02:19 -0800202 if (atomic_dec_and_test(&file->f_count))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203 __fput(file);
204}
205
206EXPORT_SYMBOL(fput);
207
208/* __fput is called from task context when aio completion releases the last
209 * last use of a struct file *. Do not use otherwise.
210 */
211void fastcall __fput(struct file *file)
212{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800213 struct dentry *dentry = file->f_path.dentry;
214 struct vfsmount *mnt = file->f_path.mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215 struct inode *inode = dentry->d_inode;
216
217 might_sleep();
Robert Love0eeca282005-07-12 17:06:03 -0400218
219 fsnotify_close(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220 /*
221 * The function eventpoll_release() should be the first called
222 * in the file cleanup chain.
223 */
224 eventpoll_release(file);
225 locks_remove_flock(file);
226
227 if (file->f_op && file->f_op->release)
228 file->f_op->release(inode, file);
229 security_file_free(file);
Theodore Ts'o577c4eb2006-09-27 01:50:49 -0700230 if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231 cdev_put(inode->i_cdev);
232 fops_put(file->f_op);
233 if (file->f_mode & FMODE_WRITE)
234 put_write_access(inode);
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700235 put_pid(file->f_owner.pid);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236 file_kill(file);
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800237 file->f_path.dentry = NULL;
238 file->f_path.mnt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239 file_free(file);
240 dput(dentry);
241 mntput(mnt);
242}
243
244struct file fastcall *fget(unsigned int fd)
245{
246 struct file *file;
247 struct files_struct *files = current->files;
248
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700249 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250 file = fcheck_files(files, fd);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700251 if (file) {
Nick Piggin095975d2006-01-08 01:02:19 -0800252 if (!atomic_inc_not_zero(&file->f_count)) {
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700253 /* File object ref couldn't be taken */
254 rcu_read_unlock();
255 return NULL;
256 }
257 }
258 rcu_read_unlock();
259
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260 return file;
261}
262
263EXPORT_SYMBOL(fget);
264
265/*
266 * Lightweight file lookup - no refcnt increment if fd table isn't shared.
267 * You can use this only if it is guranteed that the current task already
268 * holds a refcnt to that file. That check has to be done at fget() only
269 * and a flag is returned to be passed to the corresponding fput_light().
270 * There must not be a cloning between an fget_light/fput_light pair.
271 */
272struct file fastcall *fget_light(unsigned int fd, int *fput_needed)
273{
274 struct file *file;
275 struct files_struct *files = current->files;
276
277 *fput_needed = 0;
278 if (likely((atomic_read(&files->count) == 1))) {
279 file = fcheck_files(files, fd);
280 } else {
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700281 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282 file = fcheck_files(files, fd);
283 if (file) {
Nick Piggin095975d2006-01-08 01:02:19 -0800284 if (atomic_inc_not_zero(&file->f_count))
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700285 *fput_needed = 1;
286 else
287 /* Didn't get the reference, someone's freed */
288 file = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289 }
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700290 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291 }
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700292
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293 return file;
294}
295
296
297void put_filp(struct file *file)
298{
Nick Piggin095975d2006-01-08 01:02:19 -0800299 if (atomic_dec_and_test(&file->f_count)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300 security_file_free(file);
301 file_kill(file);
302 file_free(file);
303 }
304}
305
306void file_move(struct file *file, struct list_head *list)
307{
308 if (!list)
309 return;
310 file_list_lock();
Eric Dumazet2f512012005-10-30 15:02:16 -0800311 list_move(&file->f_u.fu_list, list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312 file_list_unlock();
313}
314
315void file_kill(struct file *file)
316{
Eric Dumazet2f512012005-10-30 15:02:16 -0800317 if (!list_empty(&file->f_u.fu_list)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318 file_list_lock();
Eric Dumazet2f512012005-10-30 15:02:16 -0800319 list_del_init(&file->f_u.fu_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320 file_list_unlock();
321 }
322}
323
324int fs_may_remount_ro(struct super_block *sb)
325{
Matthias Kaehlckecfdaf9e2007-10-18 23:39:56 -0700326 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327
328 /* Check that no files are currently opened for writing. */
329 file_list_lock();
Matthias Kaehlckecfdaf9e2007-10-18 23:39:56 -0700330 list_for_each_entry(file, &sb->s_files, f_u.fu_list) {
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800331 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332
333 /* File with pending delete? */
334 if (inode->i_nlink == 0)
335 goto too_bad;
336
337 /* Writeable file? */
338 if (S_ISREG(inode->i_mode) && (file->f_mode & FMODE_WRITE))
339 goto too_bad;
340 }
341 file_list_unlock();
342 return 1; /* Tis' cool bro. */
343too_bad:
344 file_list_unlock();
345 return 0;
346}
347
348void __init files_init(unsigned long mempages)
349{
350 int n;
351 /* One file with associated inode and dcache is very roughly 1K.
352 * Per default don't use more than 10% of our memory for files.
353 */
354
355 n = (mempages * (PAGE_SIZE / 1024)) / 10;
356 files_stat.max_files = n;
357 if (files_stat.max_files < NR_FILE)
358 files_stat.max_files = NR_FILE;
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700359 files_defer_init();
Mingming Cao0216bfc2006-06-23 02:05:41 -0700360 percpu_counter_init(&nr_files, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361}