blob: 322cd37626cbc6de1e75ec5bdd6e4010d851e730 [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;
Al Virodd379782013-03-01 19:48:30 -0500179 file->f_inode = path->dentry->d_inode;
Al Viro2c48b9c2009-08-09 00:52:35 +0400180 file->f_mapping = path->dentry->d_inode->i_mapping;
Dave Hansence8d2cd2007-10-16 23:31:13 -0700181 file->f_mode = mode;
182 file->f_op = fop;
Dave Hansen4a3fd212008-02-15 14:37:48 -0800183
184 /*
185 * These mounts don't really matter in practice
186 * for r/o bind mounts. They aren't userspace-
187 * visible. We do this for consistency, and so
188 * that we can do debugging checks at __fput()
189 */
Al Viro2c48b9c2009-08-09 00:52:35 +0400190 if ((mode & FMODE_WRITE) && !special_file(path->dentry->d_inode->i_mode)) {
Dave Hansenad775f52008-02-15 14:38:01 -0800191 file_take_write(file);
Roland Dreier385e3ed2009-12-16 12:48:44 -0800192 WARN_ON(mnt_clone_write(path->mnt));
Dave Hansen4a3fd212008-02-15 14:37:48 -0800193 }
Mimi Zohar890275b52010-11-02 10:13:07 -0400194 if ((mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
195 i_readcount_inc(path->dentry->d_inode);
Al Viro3d1e4632009-08-08 23:56:29 +0400196 return file;
Dave Hansence8d2cd2007-10-16 23:31:13 -0700197}
Roland Dreier73efc462009-12-16 12:43:11 -0800198EXPORT_SYMBOL(alloc_file);
Dave Hansence8d2cd2007-10-16 23:31:13 -0700199
Dave Hansenaceaf782008-02-15 14:37:31 -0800200/**
201 * drop_file_write_access - give up ability to write to a file
202 * @file: the file to which we will stop writing
203 *
204 * This is a central place which will give up the ability
205 * to write to @file, along with access to write through
206 * its vfsmount.
207 */
Al Virob57ce962012-02-12 02:38:16 -0500208static void drop_file_write_access(struct file *file)
Dave Hansenaceaf782008-02-15 14:37:31 -0800209{
Dave Hansen4a3fd212008-02-15 14:37:48 -0800210 struct vfsmount *mnt = file->f_path.mnt;
Dave Hansenaceaf782008-02-15 14:37:31 -0800211 struct dentry *dentry = file->f_path.dentry;
212 struct inode *inode = dentry->d_inode;
213
214 put_write_access(inode);
Dave Hansenad775f52008-02-15 14:38:01 -0800215
216 if (special_file(inode->i_mode))
217 return;
218 if (file_check_writeable(file) != 0)
219 return;
Jan Karaeb04c282012-06-12 16:20:35 +0200220 __mnt_drop_write(mnt);
Dave Hansenad775f52008-02-15 14:38:01 -0800221 file_release_write(file);
Dave Hansenaceaf782008-02-15 14:37:31 -0800222}
Dave Hansenaceaf782008-02-15 14:37:31 -0800223
Al Virod7065da2010-05-26 15:13:55 -0400224/* the real guts of fput() - releasing the last reference to file
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225 */
Al Virod7065da2010-05-26 15:13:55 -0400226static void __fput(struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800228 struct dentry *dentry = file->f_path.dentry;
229 struct vfsmount *mnt = file->f_path.mnt;
David Howellsc77cece2013-06-13 23:37:49 +0100230 struct inode *inode = file->f_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231
232 might_sleep();
Robert Love0eeca282005-07-12 17:06:03 -0400233
234 fsnotify_close(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235 /*
236 * The function eventpoll_release() should be the first called
237 * in the file cleanup chain.
238 */
239 eventpoll_release(file);
240 locks_remove_flock(file);
241
Al Viro233e70f2008-10-31 23:28:30 +0000242 if (unlikely(file->f_flags & FASYNC)) {
243 if (file->f_op && file->f_op->fasync)
244 file->f_op->fasync(-1, file, 0);
245 }
Mimi Zohar4199d352011-03-16 22:48:43 -0400246 ima_file_free(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247 if (file->f_op && file->f_op->release)
248 file->f_op->release(inode, file);
249 security_file_free(file);
Miklos Szeredi60ed8cf2011-03-16 18:17:54 +0100250 if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL &&
251 !(file->f_mode & FMODE_PATH))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252 cdev_put(inode->i_cdev);
Miklos Szeredi60ed8cf2011-03-16 18:17:54 +0100253 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254 fops_put(file->f_op);
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700255 put_pid(file->f_owner.pid);
Mimi Zohar890275b52010-11-02 10:13:07 -0400256 if ((file->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
257 i_readcount_dec(inode);
Dave Hansenaceaf782008-02-15 14:37:31 -0800258 if (file->f_mode & FMODE_WRITE)
259 drop_file_write_access(file);
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800260 file->f_path.dentry = NULL;
261 file->f_path.mnt = NULL;
Al Virodd379782013-03-01 19:48:30 -0500262 file->f_inode = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 file_free(file);
264 dput(dentry);
265 mntput(mnt);
266}
267
Oleg Nesterov4f5e65a2013-07-08 14:24:16 -0700268static LLIST_HEAD(delayed_fput_list);
Al Viro4a9d4b02012-06-24 09:56:45 +0400269static void delayed_fput(struct work_struct *unused)
270{
Oleg Nesterov4f5e65a2013-07-08 14:24:16 -0700271 struct llist_node *node = llist_del_all(&delayed_fput_list);
272 struct llist_node *next;
273
274 for (; node; node = next) {
275 next = llist_next(node);
276 __fput(llist_entry(node, struct file, f_u.fu_llist));
Al Viro4a9d4b02012-06-24 09:56:45 +0400277 }
278}
279
280static void ____fput(struct callback_head *work)
281{
282 __fput(container_of(work, struct file, f_u.fu_rcuhead));
283}
284
285/*
286 * If kernel thread really needs to have the final fput() it has done
287 * to complete, call this. The only user right now is the boot - we
288 * *do* need to make sure our writes to binaries on initramfs has
289 * not left us with opened struct file waiting for __fput() - execve()
290 * won't work without that. Please, don't add more callers without
291 * very good reasons; in particular, never call that with locks
292 * held and never call that from a thread that might need to do
293 * some work on any kind of umount.
294 */
295void flush_delayed_fput(void)
296{
297 delayed_fput(NULL);
298}
299
300static DECLARE_WORK(delayed_fput_work, delayed_fput);
301
Al Virod7065da2010-05-26 15:13:55 -0400302void fput(struct file *file)
303{
Al Viro4a9d4b02012-06-24 09:56:45 +0400304 if (atomic_long_dec_and_test(&file->f_count)) {
305 struct task_struct *task = current;
Oleg Nesterove7b2c402013-06-14 21:09:47 +0200306
Al Viro4a9d4b02012-06-24 09:56:45 +0400307 file_sb_list_del(file);
Oleg Nesterove7b2c402013-06-14 21:09:47 +0200308 if (likely(!in_interrupt() && !(task->flags & PF_KTHREAD))) {
309 init_task_work(&file->f_u.fu_rcuhead, ____fput);
310 if (!task_work_add(task, &file->f_u.fu_rcuhead, true))
311 return;
Andrew Morton64372502013-07-08 14:24:15 -0700312 /*
313 * After this task has run exit_task_work(),
314 * task_work_add() will fail. free_ipc_ns()->
315 * shm_destroy() can do this. Fall through to delayed
316 * fput to avoid leaking *file.
317 */
Al Viro4a9d4b02012-06-24 09:56:45 +0400318 }
Oleg Nesterov4f5e65a2013-07-08 14:24:16 -0700319
320 if (llist_add(&file->f_u.fu_llist, &delayed_fput_list))
321 schedule_work(&delayed_fput_work);
Al Viro4a9d4b02012-06-24 09:56:45 +0400322 }
323}
324
325/*
326 * synchronous analog of fput(); for kernel threads that might be needed
327 * in some umount() (and thus can't use flush_delayed_fput() without
328 * risking deadlocks), need to wait for completion of __fput() and know
329 * for this specific struct file it won't involve anything that would
330 * need them. Use only if you really need it - at the very least,
331 * don't blindly convert fput() by kernel thread to that.
332 */
333void __fput_sync(struct file *file)
334{
335 if (atomic_long_dec_and_test(&file->f_count)) {
336 struct task_struct *task = current;
337 file_sb_list_del(file);
338 BUG_ON(!(task->flags & PF_KTHREAD));
Al Virod7065da2010-05-26 15:13:55 -0400339 __fput(file);
Al Viro4a9d4b02012-06-24 09:56:45 +0400340 }
Al Virod7065da2010-05-26 15:13:55 -0400341}
342
343EXPORT_SYMBOL(fput);
344
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345void put_filp(struct file *file)
346{
Al Viro516e0cc2008-07-26 00:39:17 -0400347 if (atomic_long_dec_and_test(&file->f_count)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348 security_file_free(file);
Nick Pigginee2ffa02010-08-18 04:37:35 +1000349 file_sb_list_del(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350 file_free(file);
351 }
352}
353
Nick Piggin6416ccb2010-08-18 04:37:38 +1000354static inline int file_list_cpu(struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355{
Nick Piggin6416ccb2010-08-18 04:37:38 +1000356#ifdef CONFIG_SMP
357 return file->f_sb_list_cpu;
358#else
359 return smp_processor_id();
360#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361}
362
Nick Piggin6416ccb2010-08-18 04:37:38 +1000363/* helper for file_sb_list_add to reduce ifdefs */
364static inline void __file_sb_list_add(struct file *file, struct super_block *sb)
365{
366 struct list_head *list;
367#ifdef CONFIG_SMP
368 int cpu;
369 cpu = smp_processor_id();
370 file->f_sb_list_cpu = cpu;
371 list = per_cpu_ptr(sb->s_files, cpu);
372#else
373 list = &sb->s_files;
374#endif
375 list_add(&file->f_u.fu_list, list);
376}
377
378/**
379 * file_sb_list_add - add a file to the sb's file list
380 * @file: file to add
381 * @sb: sb to add it to
382 *
383 * Use this function to associate a file with the superblock of the inode it
384 * refers to.
385 */
386void file_sb_list_add(struct file *file, struct super_block *sb)
387{
Al Viro184caca2013-08-30 15:46:46 -0400388 if (likely(!(file->f_mode & FMODE_WRITE)))
389 return;
390 if (!S_ISREG(file_inode(file)->i_mode))
391 return;
Andi Kleen962830d2012-05-08 13:32:02 +0930392 lg_local_lock(&files_lglock);
Nick Piggin6416ccb2010-08-18 04:37:38 +1000393 __file_sb_list_add(file, sb);
Andi Kleen962830d2012-05-08 13:32:02 +0930394 lg_local_unlock(&files_lglock);
Nick Piggin6416ccb2010-08-18 04:37:38 +1000395}
396
397/**
398 * file_sb_list_del - remove a file from the sb's file list
399 * @file: file to remove
400 * @sb: sb to remove it from
401 *
402 * Use this function to remove a file from its superblock.
403 */
Nick Pigginee2ffa02010-08-18 04:37:35 +1000404void file_sb_list_del(struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405{
Eric Dumazet2f512012005-10-30 15:02:16 -0800406 if (!list_empty(&file->f_u.fu_list)) {
Andi Kleen962830d2012-05-08 13:32:02 +0930407 lg_local_lock_cpu(&files_lglock, file_list_cpu(file));
Eric Dumazet2f512012005-10-30 15:02:16 -0800408 list_del_init(&file->f_u.fu_list);
Andi Kleen962830d2012-05-08 13:32:02 +0930409 lg_local_unlock_cpu(&files_lglock, file_list_cpu(file));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410 }
411}
412
Nick Piggin6416ccb2010-08-18 04:37:38 +1000413#ifdef CONFIG_SMP
414
415/*
416 * These macros iterate all files on all CPUs for a given superblock.
417 * files_lglock must be held globally.
418 */
419#define do_file_list_for_each_entry(__sb, __file) \
420{ \
421 int i; \
422 for_each_possible_cpu(i) { \
423 struct list_head *list; \
424 list = per_cpu_ptr((__sb)->s_files, i); \
425 list_for_each_entry((__file), list, f_u.fu_list)
426
427#define while_file_list_for_each_entry \
428 } \
429}
430
431#else
432
433#define do_file_list_for_each_entry(__sb, __file) \
434{ \
435 struct list_head *list; \
436 list = &(sb)->s_files; \
437 list_for_each_entry((__file), list, f_u.fu_list)
438
439#define while_file_list_for_each_entry \
440}
441
442#endif
443
npiggin@suse.de864d7c42009-04-26 20:25:56 +1000444/**
445 * mark_files_ro - mark all files read-only
446 * @sb: superblock in question
447 *
448 * All files are marked read-only. We don't care about pending
449 * delete files so this should be used in 'force' mode only.
450 */
451void mark_files_ro(struct super_block *sb)
452{
453 struct file *f;
454
Andi Kleen962830d2012-05-08 13:32:02 +0930455 lg_global_lock(&files_lglock);
Nick Piggin6416ccb2010-08-18 04:37:38 +1000456 do_file_list_for_each_entry(sb, f) {
npiggin@suse.de864d7c42009-04-26 20:25:56 +1000457 if (!file_count(f))
458 continue;
459 if (!(f->f_mode & FMODE_WRITE))
460 continue;
Wu Fengguang42e49602010-03-05 13:42:01 -0800461 spin_lock(&f->f_lock);
npiggin@suse.de864d7c42009-04-26 20:25:56 +1000462 f->f_mode &= ~FMODE_WRITE;
Wu Fengguang42e49602010-03-05 13:42:01 -0800463 spin_unlock(&f->f_lock);
npiggin@suse.de864d7c42009-04-26 20:25:56 +1000464 if (file_check_writeable(f) != 0)
465 continue;
Jan Kara72651ca2012-12-05 14:40:14 +0100466 __mnt_drop_write(f->f_path.mnt);
npiggin@suse.de864d7c42009-04-26 20:25:56 +1000467 file_release_write(f);
Nick Piggin6416ccb2010-08-18 04:37:38 +1000468 } while_file_list_for_each_entry;
Andi Kleen962830d2012-05-08 13:32:02 +0930469 lg_global_unlock(&files_lglock);
npiggin@suse.de864d7c42009-04-26 20:25:56 +1000470}
471
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472void __init files_init(unsigned long mempages)
473{
Eric Dumazet518de9b2010-10-26 14:22:44 -0700474 unsigned long n;
Eric Dumazetb6b3fde2008-12-10 09:35:45 -0800475
476 filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
477 SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
478
479 /*
480 * One file with associated inode and dcache is very roughly 1K.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481 * Per default don't use more than 10% of our memory for files.
482 */
483
484 n = (mempages * (PAGE_SIZE / 1024)) / 10;
Eric Dumazet518de9b2010-10-26 14:22:44 -0700485 files_stat.max_files = max_t(unsigned long, n, NR_FILE);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700486 files_defer_init();
Andi Kleen962830d2012-05-08 13:32:02 +0930487 lg_lock_init(&files_lglock, "files_lglock");
Mingming Cao0216bfc2006-06-23 02:05:41 -0700488 percpu_counter_init(&nr_files, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489}