blob: 7a0a9b8722513faae34fc0f145c06f241f2630f7 [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);
Dave Hansenad775f52008-02-15 14:38:01 -080045 file_check_state(f);
Eric Dumazet2f512012005-10-30 15:02:16 -080046 call_rcu(&f->f_u.fu_rcuhead, file_free_rcu);
Linus Torvalds1da177e2005-04-16 15:20:36 -070047}
48
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080049/*
50 * Return the total number of open files in the system
51 */
52static int get_nr_files(void)
53{
54 return percpu_counter_read_positive(&nr_files);
55}
56
57/*
58 * Return the maximum number of open files in the system
59 */
60int get_max_files(void)
61{
62 return files_stat.max_files;
63}
64EXPORT_SYMBOL_GPL(get_max_files);
65
66/*
67 * Handle nr_files sysctl
68 */
69#if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
70int proc_nr_files(ctl_table *table, int write, struct file *filp,
71 void __user *buffer, size_t *lenp, loff_t *ppos)
72{
73 files_stat.nr_files = get_nr_files();
74 return proc_dointvec(table, write, filp, buffer, lenp, ppos);
75}
76#else
77int proc_nr_files(ctl_table *table, int write, struct file *filp,
78 void __user *buffer, size_t *lenp, loff_t *ppos)
79{
80 return -ENOSYS;
81}
82#endif
83
Linus Torvalds1da177e2005-04-16 15:20:36 -070084/* Find an unused file structure and return a pointer to it.
85 * Returns NULL, if there are no more free file structures or
86 * we run out of memory.
Dave Hansen430e2852008-02-15 14:37:26 -080087 *
88 * Be very careful using this. You are responsible for
89 * getting write access to any mount that you might assign
90 * to this filp, if it is opened for write. If this is not
91 * done, you will imbalance int the mount's writer count
92 * and a warning at __fput() time.
Linus Torvalds1da177e2005-04-16 15:20:36 -070093 */
94struct file *get_empty_filp(void)
95{
Benjamin LaHaise5a6b7952006-03-23 03:01:03 -080096 struct task_struct *tsk;
Kirill Korotaevaf4d2ec2005-06-23 00:09:50 -070097 static int old_max;
Linus Torvalds1da177e2005-04-16 15:20:36 -070098 struct file * f;
99
100 /*
101 * Privileged users can go above max_files
102 */
Dipankar Sarma529bf6b2006-03-07 21:55:35 -0800103 if (get_nr_files() >= files_stat.max_files && !capable(CAP_SYS_ADMIN)) {
104 /*
105 * percpu_counters are inaccurate. Do an expensive check before
106 * we go and fail.
107 */
Peter Zijlstra52d9f3b2007-10-16 23:25:44 -0700108 if (percpu_counter_sum_positive(&nr_files) >= files_stat.max_files)
Dipankar Sarma529bf6b2006-03-07 21:55:35 -0800109 goto over;
110 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111
Denis Cheng4975e452007-10-16 23:26:19 -0700112 f = kmem_cache_zalloc(filp_cachep, GFP_KERNEL);
Kirill Korotaevaf4d2ec2005-06-23 00:09:50 -0700113 if (f == NULL)
114 goto fail;
115
Dipankar Sarma529bf6b2006-03-07 21:55:35 -0800116 percpu_counter_inc(&nr_files);
Kirill Korotaevaf4d2ec2005-06-23 00:09:50 -0700117 if (security_file_alloc(f))
118 goto fail_sec;
119
Benjamin LaHaise5a6b7952006-03-23 03:01:03 -0800120 tsk = current;
Eric Dumazet2f512012005-10-30 15:02:16 -0800121 INIT_LIST_HEAD(&f->f_u.fu_list);
Benjamin LaHaise5a6b7952006-03-23 03:01:03 -0800122 atomic_set(&f->f_count, 1);
123 rwlock_init(&f->f_owner.lock);
124 f->f_uid = tsk->fsuid;
125 f->f_gid = tsk->fsgid;
126 eventpoll_init_file(f);
127 /* f->f_version: 0 */
Kirill Korotaevaf4d2ec2005-06-23 00:09:50 -0700128 return f;
129
130over:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131 /* Ran out of filps - report that */
Dipankar Sarma529bf6b2006-03-07 21:55:35 -0800132 if (get_nr_files() > old_max) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133 printk(KERN_INFO "VFS: file-max limit %d reached\n",
Dipankar Sarma529bf6b2006-03-07 21:55:35 -0800134 get_max_files());
135 old_max = get_nr_files();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136 }
Kirill Korotaevaf4d2ec2005-06-23 00:09:50 -0700137 goto fail;
138
139fail_sec:
140 file_free(f);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141fail:
142 return NULL;
143}
144
145EXPORT_SYMBOL(get_empty_filp);
146
Dave Hansence8d2cd2007-10-16 23:31:13 -0700147/**
148 * alloc_file - allocate and initialize a 'struct file'
149 * @mnt: the vfsmount on which the file will reside
150 * @dentry: the dentry representing the new file
151 * @mode: the mode with which the new file will be opened
152 * @fop: the 'struct file_operations' for the new file
153 *
154 * Use this instead of get_empty_filp() to get a new
155 * 'struct file'. Do so because of the same initialization
156 * pitfalls reasons listed for init_file(). This is a
157 * preferred interface to using init_file().
158 *
159 * If all the callers of init_file() are eliminated, its
160 * code should be moved into this function.
161 */
162struct file *alloc_file(struct vfsmount *mnt, struct dentry *dentry,
163 mode_t mode, const struct file_operations *fop)
164{
165 struct file *file;
166 struct path;
167
168 file = get_empty_filp();
169 if (!file)
170 return NULL;
171
172 init_file(file, mnt, dentry, mode, fop);
173 return file;
174}
175EXPORT_SYMBOL(alloc_file);
176
177/**
178 * init_file - initialize a 'struct file'
179 * @file: the already allocated 'struct file' to initialized
180 * @mnt: the vfsmount on which the file resides
181 * @dentry: the dentry representing this file
182 * @mode: the mode the file is opened with
183 * @fop: the 'struct file_operations' for this file
184 *
185 * Use this instead of setting the members directly. Doing so
186 * avoids making mistakes like forgetting the mntget() or
187 * forgetting to take a write on the mnt.
188 *
189 * Note: This is a crappy interface. It is here to make
190 * merging with the existing users of get_empty_filp()
191 * who have complex failure logic easier. All users
192 * of this should be moving to alloc_file().
193 */
194int init_file(struct file *file, struct vfsmount *mnt, struct dentry *dentry,
195 mode_t mode, const struct file_operations *fop)
196{
197 int error = 0;
198 file->f_path.dentry = dentry;
199 file->f_path.mnt = mntget(mnt);
200 file->f_mapping = dentry->d_inode->i_mapping;
201 file->f_mode = mode;
202 file->f_op = fop;
Dave Hansen4a3fd212008-02-15 14:37:48 -0800203
204 /*
205 * These mounts don't really matter in practice
206 * for r/o bind mounts. They aren't userspace-
207 * visible. We do this for consistency, and so
208 * that we can do debugging checks at __fput()
209 */
210 if ((mode & FMODE_WRITE) && !special_file(dentry->d_inode->i_mode)) {
Dave Hansenad775f52008-02-15 14:38:01 -0800211 file_take_write(file);
Dave Hansen4a3fd212008-02-15 14:37:48 -0800212 error = mnt_want_write(mnt);
213 WARN_ON(error);
214 }
Dave Hansence8d2cd2007-10-16 23:31:13 -0700215 return error;
216}
217EXPORT_SYMBOL(init_file);
218
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -0800219void fput(struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220{
Nick Piggin095975d2006-01-08 01:02:19 -0800221 if (atomic_dec_and_test(&file->f_count))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222 __fput(file);
223}
224
225EXPORT_SYMBOL(fput);
226
Dave Hansenaceaf782008-02-15 14:37:31 -0800227/**
228 * drop_file_write_access - give up ability to write to a file
229 * @file: the file to which we will stop writing
230 *
231 * This is a central place which will give up the ability
232 * to write to @file, along with access to write through
233 * its vfsmount.
234 */
235void drop_file_write_access(struct file *file)
236{
Dave Hansen4a3fd212008-02-15 14:37:48 -0800237 struct vfsmount *mnt = file->f_path.mnt;
Dave Hansenaceaf782008-02-15 14:37:31 -0800238 struct dentry *dentry = file->f_path.dentry;
239 struct inode *inode = dentry->d_inode;
240
241 put_write_access(inode);
Dave Hansenad775f52008-02-15 14:38:01 -0800242
243 if (special_file(inode->i_mode))
244 return;
245 if (file_check_writeable(file) != 0)
246 return;
247 mnt_drop_write(mnt);
248 file_release_write(file);
Dave Hansenaceaf782008-02-15 14:37:31 -0800249}
250EXPORT_SYMBOL_GPL(drop_file_write_access);
251
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252/* __fput is called from task context when aio completion releases the last
253 * last use of a struct file *. Do not use otherwise.
254 */
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -0800255void __fput(struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800257 struct dentry *dentry = file->f_path.dentry;
258 struct vfsmount *mnt = file->f_path.mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259 struct inode *inode = dentry->d_inode;
260
261 might_sleep();
Robert Love0eeca282005-07-12 17:06:03 -0400262
263 fsnotify_close(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264 /*
265 * The function eventpoll_release() should be the first called
266 * in the file cleanup chain.
267 */
268 eventpoll_release(file);
269 locks_remove_flock(file);
270
271 if (file->f_op && file->f_op->release)
272 file->f_op->release(inode, file);
273 security_file_free(file);
Theodore Ts'o577c4eb2006-09-27 01:50:49 -0700274 if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275 cdev_put(inode->i_cdev);
276 fops_put(file->f_op);
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700277 put_pid(file->f_owner.pid);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278 file_kill(file);
Dave Hansenaceaf782008-02-15 14:37:31 -0800279 if (file->f_mode & FMODE_WRITE)
280 drop_file_write_access(file);
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800281 file->f_path.dentry = NULL;
282 file->f_path.mnt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283 file_free(file);
284 dput(dentry);
285 mntput(mnt);
286}
287
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -0800288struct file *fget(unsigned int fd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289{
290 struct file *file;
291 struct files_struct *files = current->files;
292
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700293 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294 file = fcheck_files(files, fd);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700295 if (file) {
Nick Piggin095975d2006-01-08 01:02:19 -0800296 if (!atomic_inc_not_zero(&file->f_count)) {
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700297 /* File object ref couldn't be taken */
298 rcu_read_unlock();
299 return NULL;
300 }
301 }
302 rcu_read_unlock();
303
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304 return file;
305}
306
307EXPORT_SYMBOL(fget);
308
309/*
310 * Lightweight file lookup - no refcnt increment if fd table isn't shared.
311 * You can use this only if it is guranteed that the current task already
312 * holds a refcnt to that file. That check has to be done at fget() only
313 * and a flag is returned to be passed to the corresponding fput_light().
314 * There must not be a cloning between an fget_light/fput_light pair.
315 */
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -0800316struct file *fget_light(unsigned int fd, int *fput_needed)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317{
318 struct file *file;
319 struct files_struct *files = current->files;
320
321 *fput_needed = 0;
322 if (likely((atomic_read(&files->count) == 1))) {
323 file = fcheck_files(files, fd);
324 } else {
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700325 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326 file = fcheck_files(files, fd);
327 if (file) {
Nick Piggin095975d2006-01-08 01:02:19 -0800328 if (atomic_inc_not_zero(&file->f_count))
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700329 *fput_needed = 1;
330 else
331 /* Didn't get the reference, someone's freed */
332 file = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333 }
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700334 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335 }
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700336
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337 return file;
338}
339
340
341void put_filp(struct file *file)
342{
Nick Piggin095975d2006-01-08 01:02:19 -0800343 if (atomic_dec_and_test(&file->f_count)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344 security_file_free(file);
345 file_kill(file);
346 file_free(file);
347 }
348}
349
350void file_move(struct file *file, struct list_head *list)
351{
352 if (!list)
353 return;
354 file_list_lock();
Eric Dumazet2f512012005-10-30 15:02:16 -0800355 list_move(&file->f_u.fu_list, list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356 file_list_unlock();
357}
358
359void file_kill(struct file *file)
360{
Eric Dumazet2f512012005-10-30 15:02:16 -0800361 if (!list_empty(&file->f_u.fu_list)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362 file_list_lock();
Eric Dumazet2f512012005-10-30 15:02:16 -0800363 list_del_init(&file->f_u.fu_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364 file_list_unlock();
365 }
366}
367
368int fs_may_remount_ro(struct super_block *sb)
369{
Matthias Kaehlckecfdaf9e2007-10-18 23:39:56 -0700370 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371
372 /* Check that no files are currently opened for writing. */
373 file_list_lock();
Matthias Kaehlckecfdaf9e2007-10-18 23:39:56 -0700374 list_for_each_entry(file, &sb->s_files, f_u.fu_list) {
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800375 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376
377 /* File with pending delete? */
378 if (inode->i_nlink == 0)
379 goto too_bad;
380
381 /* Writeable file? */
382 if (S_ISREG(inode->i_mode) && (file->f_mode & FMODE_WRITE))
383 goto too_bad;
384 }
385 file_list_unlock();
386 return 1; /* Tis' cool bro. */
387too_bad:
388 file_list_unlock();
389 return 0;
390}
391
392void __init files_init(unsigned long mempages)
393{
394 int n;
395 /* One file with associated inode and dcache is very roughly 1K.
396 * Per default don't use more than 10% of our memory for files.
397 */
398
399 n = (mempages * (PAGE_SIZE / 1024)) / 10;
400 files_stat.max_files = n;
401 if (files_stat.max_files < NR_FILE)
402 files_stat.max_files = NR_FILE;
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700403 files_defer_init();
Mingming Cao0216bfc2006-06-23 02:05:41 -0700404 percpu_counter_init(&nr_files, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405}