blob: aa07d3684a2e8e13bdf4ae36eb5dc0ba98641223 [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>
Al Viro9f3acc32008-04-24 07:44:08 -040011#include <linux/fdtable.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070012#include <linux/init.h>
13#include <linux/module.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070014#include <linux/fs.h>
15#include <linux/security.h>
16#include <linux/eventpoll.h>
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070017#include <linux/rcupdate.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include <linux/mount.h>
Randy Dunlap16f7e0f2006-01-11 12:17:46 -080019#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include <linux/cdev.h>
Robert Love0eeca282005-07-12 17:06:03 -040021#include <linux/fsnotify.h>
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080022#include <linux/sysctl.h>
Nick Piggin6416ccb2010-08-18 04:37:38 +100023#include <linux/lglock.h>
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080024#include <linux/percpu_counter.h>
Nick Piggin6416ccb2010-08-18 04:37:38 +100025#include <linux/percpu.h>
Al Viro4a9d4b02012-06-24 09:56:45 +040026#include <linux/hardirq.h>
27#include <linux/task_work.h>
Al Viro0552f872009-12-16 04:53:03 -050028#include <linux/ima.h>
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080029
Arun Sharma600634972011-07-26 16:09:06 -070030#include <linux/atomic.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070031
Eric Parise81e3f42009-12-04 15:47:36 -050032#include "internal.h"
33
Linus Torvalds1da177e2005-04-16 15:20:36 -070034/* sysctl tunables... */
35struct files_stat_struct files_stat = {
36 .max_files = NR_FILE
37};
38
Lai Jiangshan4b2c551f2012-10-09 14:49:54 -070039DEFINE_STATIC_LGLOCK(files_lglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070040
Eric Dumazetb6b3fde2008-12-10 09:35:45 -080041/* SLAB cache for file structures */
42static struct kmem_cache *filp_cachep __read_mostly;
43
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080044static struct percpu_counter nr_files __cacheline_aligned_in_smp;
Linus Torvalds1da177e2005-04-16 15:20:36 -070045
Al Viro5c33b182012-07-20 23:05:59 +040046static void file_free_rcu(struct rcu_head *head)
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070047{
David Howellsd76b0d92008-11-14 10:39:25 +110048 struct file *f = container_of(head, struct file, f_u.fu_rcuhead);
49
50 put_cred(f->f_cred);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070051 kmem_cache_free(filp_cachep, f);
52}
53
Linus Torvalds1da177e2005-04-16 15:20:36 -070054static inline void file_free(struct file *f)
55{
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080056 percpu_counter_dec(&nr_files);
Dave Hansenad775f52008-02-15 14:38:01 -080057 file_check_state(f);
Eric Dumazet2f512012005-10-30 15:02:16 -080058 call_rcu(&f->f_u.fu_rcuhead, file_free_rcu);
Linus Torvalds1da177e2005-04-16 15:20:36 -070059}
60
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080061/*
62 * Return the total number of open files in the system
63 */
Eric Dumazet518de9b2010-10-26 14:22:44 -070064static long get_nr_files(void)
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080065{
66 return percpu_counter_read_positive(&nr_files);
67}
68
69/*
70 * Return the maximum number of open files in the system
71 */
Eric Dumazet518de9b2010-10-26 14:22:44 -070072unsigned long get_max_files(void)
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080073{
74 return files_stat.max_files;
75}
76EXPORT_SYMBOL_GPL(get_max_files);
77
78/*
79 * Handle nr_files sysctl
80 */
81#if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
Alexey Dobriyan8d65af72009-09-23 15:57:19 -070082int proc_nr_files(ctl_table *table, int write,
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080083 void __user *buffer, size_t *lenp, loff_t *ppos)
84{
85 files_stat.nr_files = get_nr_files();
Eric Dumazet518de9b2010-10-26 14:22:44 -070086 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080087}
88#else
Alexey Dobriyan8d65af72009-09-23 15:57:19 -070089int proc_nr_files(ctl_table *table, int write,
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080090 void __user *buffer, size_t *lenp, loff_t *ppos)
91{
92 return -ENOSYS;
93}
94#endif
95
Linus Torvalds1da177e2005-04-16 15:20:36 -070096/* Find an unused file structure and return a pointer to it.
Al Viro1afc99b2013-02-14 20:41:04 -050097 * Returns an error pointer if some error happend e.g. we over file
98 * structures limit, run out of memory or operation is not permitted.
Dave Hansen430e2852008-02-15 14:37:26 -080099 *
100 * Be very careful using this. You are responsible for
101 * getting write access to any mount that you might assign
102 * to this filp, if it is opened for write. If this is not
103 * done, you will imbalance int the mount's writer count
104 * and a warning at __fput() time.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105 */
106struct file *get_empty_filp(void)
107{
David Howells86a264a2008-11-14 10:39:18 +1100108 const struct cred *cred = current_cred();
Eric Dumazet518de9b2010-10-26 14:22:44 -0700109 static long old_max;
Al Viro1afc99b2013-02-14 20:41:04 -0500110 struct file *f;
111 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112
113 /*
114 * Privileged users can go above max_files
115 */
Dipankar Sarma529bf6b2006-03-07 21:55:35 -0800116 if (get_nr_files() >= files_stat.max_files && !capable(CAP_SYS_ADMIN)) {
117 /*
118 * percpu_counters are inaccurate. Do an expensive check before
119 * we go and fail.
120 */
Peter Zijlstra52d9f3b2007-10-16 23:25:44 -0700121 if (percpu_counter_sum_positive(&nr_files) >= files_stat.max_files)
Dipankar Sarma529bf6b2006-03-07 21:55:35 -0800122 goto over;
123 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124
Denis Cheng4975e452007-10-16 23:26:19 -0700125 f = kmem_cache_zalloc(filp_cachep, GFP_KERNEL);
Al Viro1afc99b2013-02-14 20:41:04 -0500126 if (unlikely(!f))
127 return ERR_PTR(-ENOMEM);
Kirill Korotaevaf4d2ec2005-06-23 00:09:50 -0700128
Dipankar Sarma529bf6b2006-03-07 21:55:35 -0800129 percpu_counter_inc(&nr_files);
Tetsuo Handa78d29782011-02-04 18:13:24 +0000130 f->f_cred = get_cred(cred);
Al Viro1afc99b2013-02-14 20:41:04 -0500131 error = security_file_alloc(f);
132 if (unlikely(error)) {
133 file_free(f);
134 return ERR_PTR(error);
135 }
Kirill Korotaevaf4d2ec2005-06-23 00:09:50 -0700136
Eric Dumazet2f512012005-10-30 15:02:16 -0800137 INIT_LIST_HEAD(&f->f_u.fu_list);
Al Viro516e0cc2008-07-26 00:39:17 -0400138 atomic_long_set(&f->f_count, 1);
Benjamin LaHaise5a6b79512006-03-23 03:01:03 -0800139 rwlock_init(&f->f_owner.lock);
Jonathan Corbet68499912009-02-06 13:52:43 -0700140 spin_lock_init(&f->f_lock);
Benjamin LaHaise5a6b79512006-03-23 03:01:03 -0800141 eventpoll_init_file(f);
142 /* f->f_version: 0 */
Kirill Korotaevaf4d2ec2005-06-23 00:09:50 -0700143 return f;
144
145over:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146 /* Ran out of filps - report that */
Dipankar Sarma529bf6b2006-03-07 21:55:35 -0800147 if (get_nr_files() > old_max) {
Eric Dumazet518de9b2010-10-26 14:22:44 -0700148 pr_info("VFS: file-max limit %lu reached\n", get_max_files());
Dipankar Sarma529bf6b2006-03-07 21:55:35 -0800149 old_max = get_nr_files();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150 }
Al Viro1afc99b2013-02-14 20:41:04 -0500151 return ERR_PTR(-ENFILE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152}
153
Dave Hansence8d2cd2007-10-16 23:31:13 -0700154/**
155 * alloc_file - allocate and initialize a 'struct file'
156 * @mnt: the vfsmount on which the file will reside
157 * @dentry: the dentry representing the new file
158 * @mode: the mode with which the new file will be opened
159 * @fop: the 'struct file_operations' for the new file
160 *
161 * Use this instead of get_empty_filp() to get a new
162 * 'struct file'. Do so because of the same initialization
163 * pitfalls reasons listed for init_file(). This is a
164 * preferred interface to using init_file().
165 *
166 * If all the callers of init_file() are eliminated, its
167 * code should be moved into this function.
168 */
Al Viro2c48b9c2009-08-09 00:52:35 +0400169struct file *alloc_file(struct path *path, fmode_t mode,
170 const struct file_operations *fop)
Dave Hansence8d2cd2007-10-16 23:31:13 -0700171{
172 struct file *file;
Dave Hansence8d2cd2007-10-16 23:31:13 -0700173
174 file = get_empty_filp();
Al Viro1afc99b2013-02-14 20:41:04 -0500175 if (IS_ERR(file))
Anatol Pomozov39b65252012-09-12 20:11:55 -0700176 return file;
Dave Hansence8d2cd2007-10-16 23:31:13 -0700177
Al Viro2c48b9c2009-08-09 00:52:35 +0400178 file->f_path = *path;
179 file->f_mapping = path->dentry->d_inode->i_mapping;
Dave Hansence8d2cd2007-10-16 23:31:13 -0700180 file->f_mode = mode;
181 file->f_op = fop;
Dave Hansen4a3fd212008-02-15 14:37:48 -0800182
183 /*
184 * These mounts don't really matter in practice
185 * for r/o bind mounts. They aren't userspace-
186 * visible. We do this for consistency, and so
187 * that we can do debugging checks at __fput()
188 */
Al Viro2c48b9c2009-08-09 00:52:35 +0400189 if ((mode & FMODE_WRITE) && !special_file(path->dentry->d_inode->i_mode)) {
Dave Hansenad775f52008-02-15 14:38:01 -0800190 file_take_write(file);
Roland Dreier385e3ed2009-12-16 12:48:44 -0800191 WARN_ON(mnt_clone_write(path->mnt));
Dave Hansen4a3fd212008-02-15 14:37:48 -0800192 }
Mimi Zohar890275b52010-11-02 10:13:07 -0400193 if ((mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
194 i_readcount_inc(path->dentry->d_inode);
Al Viro3d1e4632009-08-08 23:56:29 +0400195 return file;
Dave Hansence8d2cd2007-10-16 23:31:13 -0700196}
Roland Dreier73efc462009-12-16 12:43:11 -0800197EXPORT_SYMBOL(alloc_file);
Dave Hansence8d2cd2007-10-16 23:31:13 -0700198
Dave Hansenaceaf782008-02-15 14:37:31 -0800199/**
200 * drop_file_write_access - give up ability to write to a file
201 * @file: the file to which we will stop writing
202 *
203 * This is a central place which will give up the ability
204 * to write to @file, along with access to write through
205 * its vfsmount.
206 */
Al Virob57ce962012-02-12 02:38:16 -0500207static void drop_file_write_access(struct file *file)
Dave Hansenaceaf782008-02-15 14:37:31 -0800208{
Dave Hansen4a3fd212008-02-15 14:37:48 -0800209 struct vfsmount *mnt = file->f_path.mnt;
Dave Hansenaceaf782008-02-15 14:37:31 -0800210 struct dentry *dentry = file->f_path.dentry;
211 struct inode *inode = dentry->d_inode;
212
213 put_write_access(inode);
Dave Hansenad775f52008-02-15 14:38:01 -0800214
215 if (special_file(inode->i_mode))
216 return;
217 if (file_check_writeable(file) != 0)
218 return;
Jan Karaeb04c282012-06-12 16:20:35 +0200219 __mnt_drop_write(mnt);
Dave Hansenad775f52008-02-15 14:38:01 -0800220 file_release_write(file);
Dave Hansenaceaf782008-02-15 14:37:31 -0800221}
Dave Hansenaceaf782008-02-15 14:37:31 -0800222
Al Virod7065da2010-05-26 15:13:55 -0400223/* the real guts of fput() - releasing the last reference to file
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224 */
Al Virod7065da2010-05-26 15:13:55 -0400225static void __fput(struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800227 struct dentry *dentry = file->f_path.dentry;
228 struct vfsmount *mnt = file->f_path.mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229 struct inode *inode = dentry->d_inode;
230
231 might_sleep();
Robert Love0eeca282005-07-12 17:06:03 -0400232
233 fsnotify_close(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234 /*
235 * The function eventpoll_release() should be the first called
236 * in the file cleanup chain.
237 */
238 eventpoll_release(file);
239 locks_remove_flock(file);
240
Al Viro233e70f2008-10-31 23:28:30 +0000241 if (unlikely(file->f_flags & FASYNC)) {
242 if (file->f_op && file->f_op->fasync)
243 file->f_op->fasync(-1, file, 0);
244 }
Mimi Zohar4199d352011-03-16 22:48:43 -0400245 ima_file_free(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246 if (file->f_op && file->f_op->release)
247 file->f_op->release(inode, file);
248 security_file_free(file);
Miklos Szeredi60ed8cf2011-03-16 18:17:54 +0100249 if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL &&
250 !(file->f_mode & FMODE_PATH))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251 cdev_put(inode->i_cdev);
Miklos Szeredi60ed8cf2011-03-16 18:17:54 +0100252 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253 fops_put(file->f_op);
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700254 put_pid(file->f_owner.pid);
Mimi Zohar890275b52010-11-02 10:13:07 -0400255 if ((file->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
256 i_readcount_dec(inode);
Dave Hansenaceaf782008-02-15 14:37:31 -0800257 if (file->f_mode & FMODE_WRITE)
258 drop_file_write_access(file);
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800259 file->f_path.dentry = NULL;
260 file->f_path.mnt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 file_free(file);
262 dput(dentry);
263 mntput(mnt);
264}
265
Al Viro4a9d4b02012-06-24 09:56:45 +0400266static DEFINE_SPINLOCK(delayed_fput_lock);
267static LIST_HEAD(delayed_fput_list);
268static void delayed_fput(struct work_struct *unused)
269{
270 LIST_HEAD(head);
271 spin_lock_irq(&delayed_fput_lock);
272 list_splice_init(&delayed_fput_list, &head);
273 spin_unlock_irq(&delayed_fput_lock);
274 while (!list_empty(&head)) {
275 struct file *f = list_first_entry(&head, struct file, f_u.fu_list);
276 list_del_init(&f->f_u.fu_list);
277 __fput(f);
278 }
279}
280
281static void ____fput(struct callback_head *work)
282{
283 __fput(container_of(work, struct file, f_u.fu_rcuhead));
284}
285
286/*
287 * If kernel thread really needs to have the final fput() it has done
288 * to complete, call this. The only user right now is the boot - we
289 * *do* need to make sure our writes to binaries on initramfs has
290 * not left us with opened struct file waiting for __fput() - execve()
291 * won't work without that. Please, don't add more callers without
292 * very good reasons; in particular, never call that with locks
293 * held and never call that from a thread that might need to do
294 * some work on any kind of umount.
295 */
296void flush_delayed_fput(void)
297{
298 delayed_fput(NULL);
299}
300
301static DECLARE_WORK(delayed_fput_work, delayed_fput);
302
Al Virod7065da2010-05-26 15:13:55 -0400303void fput(struct file *file)
304{
Al Viro4a9d4b02012-06-24 09:56:45 +0400305 if (atomic_long_dec_and_test(&file->f_count)) {
306 struct task_struct *task = current;
307 file_sb_list_del(file);
308 if (unlikely(in_interrupt() || task->flags & PF_KTHREAD)) {
309 unsigned long flags;
310 spin_lock_irqsave(&delayed_fput_lock, flags);
311 list_add(&file->f_u.fu_list, &delayed_fput_list);
312 schedule_work(&delayed_fput_work);
313 spin_unlock_irqrestore(&delayed_fput_lock, flags);
314 return;
315 }
316 init_task_work(&file->f_u.fu_rcuhead, ____fput);
317 task_work_add(task, &file->f_u.fu_rcuhead, true);
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;
333 file_sb_list_del(file);
334 BUG_ON(!(task->flags & PF_KTHREAD));
Al Virod7065da2010-05-26 15:13:55 -0400335 __fput(file);
Al Viro4a9d4b02012-06-24 09:56:45 +0400336 }
Al Virod7065da2010-05-26 15:13:55 -0400337}
338
339EXPORT_SYMBOL(fput);
340
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341void put_filp(struct file *file)
342{
Al Viro516e0cc2008-07-26 00:39:17 -0400343 if (atomic_long_dec_and_test(&file->f_count)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344 security_file_free(file);
Nick Pigginee2ffa02010-08-18 04:37:35 +1000345 file_sb_list_del(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346 file_free(file);
347 }
348}
349
Nick Piggin6416ccb2010-08-18 04:37:38 +1000350static inline int file_list_cpu(struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351{
Nick Piggin6416ccb2010-08-18 04:37:38 +1000352#ifdef CONFIG_SMP
353 return file->f_sb_list_cpu;
354#else
355 return smp_processor_id();
356#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357}
358
Nick Piggin6416ccb2010-08-18 04:37:38 +1000359/* helper for file_sb_list_add to reduce ifdefs */
360static inline void __file_sb_list_add(struct file *file, struct super_block *sb)
361{
362 struct list_head *list;
363#ifdef CONFIG_SMP
364 int cpu;
365 cpu = smp_processor_id();
366 file->f_sb_list_cpu = cpu;
367 list = per_cpu_ptr(sb->s_files, cpu);
368#else
369 list = &sb->s_files;
370#endif
371 list_add(&file->f_u.fu_list, list);
372}
373
374/**
375 * file_sb_list_add - add a file to the sb's file list
376 * @file: file to add
377 * @sb: sb to add it to
378 *
379 * Use this function to associate a file with the superblock of the inode it
380 * refers to.
381 */
382void file_sb_list_add(struct file *file, struct super_block *sb)
383{
Andi Kleen962830d2012-05-08 13:32:02 +0930384 lg_local_lock(&files_lglock);
Nick Piggin6416ccb2010-08-18 04:37:38 +1000385 __file_sb_list_add(file, sb);
Andi Kleen962830d2012-05-08 13:32:02 +0930386 lg_local_unlock(&files_lglock);
Nick Piggin6416ccb2010-08-18 04:37:38 +1000387}
388
389/**
390 * file_sb_list_del - remove a file from the sb's file list
391 * @file: file to remove
392 * @sb: sb to remove it from
393 *
394 * Use this function to remove a file from its superblock.
395 */
Nick Pigginee2ffa02010-08-18 04:37:35 +1000396void file_sb_list_del(struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397{
Eric Dumazet2f512012005-10-30 15:02:16 -0800398 if (!list_empty(&file->f_u.fu_list)) {
Andi Kleen962830d2012-05-08 13:32:02 +0930399 lg_local_lock_cpu(&files_lglock, file_list_cpu(file));
Eric Dumazet2f512012005-10-30 15:02:16 -0800400 list_del_init(&file->f_u.fu_list);
Andi Kleen962830d2012-05-08 13:32:02 +0930401 lg_local_unlock_cpu(&files_lglock, file_list_cpu(file));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402 }
403}
404
Nick Piggin6416ccb2010-08-18 04:37:38 +1000405#ifdef CONFIG_SMP
406
407/*
408 * These macros iterate all files on all CPUs for a given superblock.
409 * files_lglock must be held globally.
410 */
411#define do_file_list_for_each_entry(__sb, __file) \
412{ \
413 int i; \
414 for_each_possible_cpu(i) { \
415 struct list_head *list; \
416 list = per_cpu_ptr((__sb)->s_files, i); \
417 list_for_each_entry((__file), list, f_u.fu_list)
418
419#define while_file_list_for_each_entry \
420 } \
421}
422
423#else
424
425#define do_file_list_for_each_entry(__sb, __file) \
426{ \
427 struct list_head *list; \
428 list = &(sb)->s_files; \
429 list_for_each_entry((__file), list, f_u.fu_list)
430
431#define while_file_list_for_each_entry \
432}
433
434#endif
435
npiggin@suse.de864d7c42009-04-26 20:25:56 +1000436/**
437 * mark_files_ro - mark all files read-only
438 * @sb: superblock in question
439 *
440 * All files are marked read-only. We don't care about pending
441 * delete files so this should be used in 'force' mode only.
442 */
443void mark_files_ro(struct super_block *sb)
444{
445 struct file *f;
446
Andi Kleen962830d2012-05-08 13:32:02 +0930447 lg_global_lock(&files_lglock);
Nick Piggin6416ccb2010-08-18 04:37:38 +1000448 do_file_list_for_each_entry(sb, f) {
Al Viro496ad9a2013-01-23 17:07:38 -0500449 if (!S_ISREG(file_inode(f)->i_mode))
npiggin@suse.de864d7c42009-04-26 20:25:56 +1000450 continue;
451 if (!file_count(f))
452 continue;
453 if (!(f->f_mode & FMODE_WRITE))
454 continue;
Wu Fengguang42e49602010-03-05 13:42:01 -0800455 spin_lock(&f->f_lock);
npiggin@suse.de864d7c42009-04-26 20:25:56 +1000456 f->f_mode &= ~FMODE_WRITE;
Wu Fengguang42e49602010-03-05 13:42:01 -0800457 spin_unlock(&f->f_lock);
npiggin@suse.de864d7c42009-04-26 20:25:56 +1000458 if (file_check_writeable(f) != 0)
459 continue;
Jan Kara72651ca2012-12-05 14:40:14 +0100460 __mnt_drop_write(f->f_path.mnt);
npiggin@suse.de864d7c42009-04-26 20:25:56 +1000461 file_release_write(f);
Nick Piggin6416ccb2010-08-18 04:37:38 +1000462 } while_file_list_for_each_entry;
Andi Kleen962830d2012-05-08 13:32:02 +0930463 lg_global_unlock(&files_lglock);
npiggin@suse.de864d7c42009-04-26 20:25:56 +1000464}
465
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466void __init files_init(unsigned long mempages)
467{
Eric Dumazet518de9b2010-10-26 14:22:44 -0700468 unsigned long n;
Eric Dumazetb6b3fde2008-12-10 09:35:45 -0800469
470 filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
471 SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
472
473 /*
474 * One file with associated inode and dcache is very roughly 1K.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475 * Per default don't use more than 10% of our memory for files.
476 */
477
478 n = (mempages * (PAGE_SIZE / 1024)) / 10;
Eric Dumazet518de9b2010-10-26 14:22:44 -0700479 files_stat.max_files = max_t(unsigned long, n, NR_FILE);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700480 files_defer_init();
Andi Kleen962830d2012-05-08 13:32:02 +0930481 lg_lock_init(&files_lglock, "files_lglock");
Mingming Cao0216bfc2006-06-23 02:05:41 -0700482 percpu_counter_init(&nr_files, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483}