blob: 70f2a0fd6aec62b28724d46e356dc0ff871f88b8 [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 Viro0552f872009-12-16 04:53:03 -050026#include <linux/ima.h>
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080027
Arun Sharma600634972011-07-26 16:09:06 -070028#include <linux/atomic.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070029
Eric Parise81e3f42009-12-04 15:47:36 -050030#include "internal.h"
31
Linus Torvalds1da177e2005-04-16 15:20:36 -070032/* sysctl tunables... */
33struct files_stat_struct files_stat = {
34 .max_files = NR_FILE
35};
36
Nick Piggin6416ccb2010-08-18 04:37:38 +100037DECLARE_LGLOCK(files_lglock);
38DEFINE_LGLOCK(files_lglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070039
Eric Dumazetb6b3fde2008-12-10 09:35:45 -080040/* SLAB cache for file structures */
41static struct kmem_cache *filp_cachep __read_mostly;
42
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080043static struct percpu_counter nr_files __cacheline_aligned_in_smp;
Linus Torvalds1da177e2005-04-16 15:20:36 -070044
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070045static inline void file_free_rcu(struct rcu_head *head)
46{
David Howellsd76b0d92008-11-14 10:39:25 +110047 struct file *f = container_of(head, struct file, f_u.fu_rcuhead);
48
49 put_cred(f->f_cred);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070050 kmem_cache_free(filp_cachep, f);
51}
52
Linus Torvalds1da177e2005-04-16 15:20:36 -070053static inline void file_free(struct file *f)
54{
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080055 percpu_counter_dec(&nr_files);
Dave Hansenad775f52008-02-15 14:38:01 -080056 file_check_state(f);
Eric Dumazet2f512012005-10-30 15:02:16 -080057 call_rcu(&f->f_u.fu_rcuhead, file_free_rcu);
Linus Torvalds1da177e2005-04-16 15:20:36 -070058}
59
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080060/*
61 * Return the total number of open files in the system
62 */
Eric Dumazet518de9b2010-10-26 14:22:44 -070063static long get_nr_files(void)
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080064{
65 return percpu_counter_read_positive(&nr_files);
66}
67
68/*
69 * Return the maximum number of open files in the system
70 */
Eric Dumazet518de9b2010-10-26 14:22:44 -070071unsigned long get_max_files(void)
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080072{
73 return files_stat.max_files;
74}
75EXPORT_SYMBOL_GPL(get_max_files);
76
77/*
78 * Handle nr_files sysctl
79 */
80#if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
Alexey Dobriyan8d65af72009-09-23 15:57:19 -070081int proc_nr_files(ctl_table *table, int write,
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080082 void __user *buffer, size_t *lenp, loff_t *ppos)
83{
84 files_stat.nr_files = get_nr_files();
Eric Dumazet518de9b2010-10-26 14:22:44 -070085 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080086}
87#else
Alexey Dobriyan8d65af72009-09-23 15:57:19 -070088int proc_nr_files(ctl_table *table, int write,
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080089 void __user *buffer, size_t *lenp, loff_t *ppos)
90{
91 return -ENOSYS;
92}
93#endif
94
Linus Torvalds1da177e2005-04-16 15:20:36 -070095/* Find an unused file structure and return a pointer to it.
96 * Returns NULL, if there are no more free file structures or
97 * we run out of memory.
Dave Hansen430e2852008-02-15 14:37:26 -080098 *
99 * Be very careful using this. You are responsible for
100 * getting write access to any mount that you might assign
101 * to this filp, if it is opened for write. If this is not
102 * done, you will imbalance int the mount's writer count
103 * and a warning at __fput() time.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104 */
105struct file *get_empty_filp(void)
106{
David Howells86a264a2008-11-14 10:39:18 +1100107 const struct cred *cred = current_cred();
Eric Dumazet518de9b2010-10-26 14:22:44 -0700108 static long old_max;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109 struct file * f;
110
111 /*
112 * Privileged users can go above max_files
113 */
Dipankar Sarma529bf6b2006-03-07 21:55:35 -0800114 if (get_nr_files() >= files_stat.max_files && !capable(CAP_SYS_ADMIN)) {
115 /*
116 * percpu_counters are inaccurate. Do an expensive check before
117 * we go and fail.
118 */
Peter Zijlstra52d9f3b2007-10-16 23:25:44 -0700119 if (percpu_counter_sum_positive(&nr_files) >= files_stat.max_files)
Dipankar Sarma529bf6b2006-03-07 21:55:35 -0800120 goto over;
121 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122
Denis Cheng4975e452007-10-16 23:26:19 -0700123 f = kmem_cache_zalloc(filp_cachep, GFP_KERNEL);
Kirill Korotaevaf4d2ec2005-06-23 00:09:50 -0700124 if (f == NULL)
125 goto fail;
126
Dipankar Sarma529bf6b2006-03-07 21:55:35 -0800127 percpu_counter_inc(&nr_files);
Tetsuo Handa78d29782011-02-04 18:13:24 +0000128 f->f_cred = get_cred(cred);
Kirill Korotaevaf4d2ec2005-06-23 00:09:50 -0700129 if (security_file_alloc(f))
130 goto fail_sec;
131
Eric Dumazet2f512012005-10-30 15:02:16 -0800132 INIT_LIST_HEAD(&f->f_u.fu_list);
Al Viro516e0cc2008-07-26 00:39:17 -0400133 atomic_long_set(&f->f_count, 1);
Benjamin LaHaise5a6b79512006-03-23 03:01:03 -0800134 rwlock_init(&f->f_owner.lock);
Jonathan Corbet68499912009-02-06 13:52:43 -0700135 spin_lock_init(&f->f_lock);
Benjamin LaHaise5a6b79512006-03-23 03:01:03 -0800136 eventpoll_init_file(f);
137 /* f->f_version: 0 */
Kirill Korotaevaf4d2ec2005-06-23 00:09:50 -0700138 return f;
139
140over:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141 /* Ran out of filps - report that */
Dipankar Sarma529bf6b2006-03-07 21:55:35 -0800142 if (get_nr_files() > old_max) {
Eric Dumazet518de9b2010-10-26 14:22:44 -0700143 pr_info("VFS: file-max limit %lu reached\n", get_max_files());
Dipankar Sarma529bf6b2006-03-07 21:55:35 -0800144 old_max = get_nr_files();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145 }
Kirill Korotaevaf4d2ec2005-06-23 00:09:50 -0700146 goto fail;
147
148fail_sec:
149 file_free(f);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150fail:
151 return NULL;
152}
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();
175 if (!file)
176 return NULL;
177
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;
219 mnt_drop_write(mnt);
220 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 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245 if (file->f_op && file->f_op->release)
246 file->f_op->release(inode, file);
247 security_file_free(file);
Al Viro89068c52010-02-07 03:07:29 -0500248 ima_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);
Nick Pigginee2ffa02010-08-18 04:37:35 +1000255 file_sb_list_del(file);
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;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262 file_free(file);
263 dput(dentry);
264 mntput(mnt);
265}
266
Al Virod7065da2010-05-26 15:13:55 -0400267void fput(struct file *file)
268{
269 if (atomic_long_dec_and_test(&file->f_count))
270 __fput(file);
271}
272
273EXPORT_SYMBOL(fput);
274
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -0800275struct file *fget(unsigned int fd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276{
277 struct file *file;
278 struct files_struct *files = current->files;
279
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700280 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281 file = fcheck_files(files, fd);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700282 if (file) {
Al Viro1abf0c72011-03-13 03:51:11 -0400283 /* File object ref couldn't be taken */
284 if (file->f_mode & FMODE_PATH ||
285 !atomic_long_inc_not_zero(&file->f_count))
286 file = NULL;
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700287 }
288 rcu_read_unlock();
289
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290 return file;
291}
292
293EXPORT_SYMBOL(fget);
294
Al Viro1abf0c72011-03-13 03:51:11 -0400295struct file *fget_raw(unsigned int fd)
296{
297 struct file *file;
298 struct files_struct *files = current->files;
299
300 rcu_read_lock();
301 file = fcheck_files(files, fd);
302 if (file) {
303 /* File object ref couldn't be taken */
304 if (!atomic_long_inc_not_zero(&file->f_count))
305 file = NULL;
306 }
307 rcu_read_unlock();
308
309 return file;
310}
311
Al Viro326be7b2011-03-13 17:08:22 -0400312EXPORT_SYMBOL(fget_raw);
313
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314/*
Tony Battersby58939472010-08-10 18:01:29 -0700315 * Lightweight file lookup - no refcnt increment if fd table isn't shared.
316 *
317 * You can use this instead of fget if you satisfy all of the following
318 * conditions:
319 * 1) You must call fput_light before exiting the syscall and returning control
320 * to userspace (i.e. you cannot remember the returned struct file * after
321 * returning to userspace).
322 * 2) You must not call filp_close on the returned struct file * in between
323 * calls to fget_light and fput_light.
324 * 3) You must not clone the current task in between the calls to fget_light
325 * and fput_light.
326 *
327 * The fput_needed flag returned by fget_light should be passed to the
328 * corresponding fput_light.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329 */
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -0800330struct file *fget_light(unsigned int fd, int *fput_needed)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331{
332 struct file *file;
333 struct files_struct *files = current->files;
334
335 *fput_needed = 0;
Steven Rostedt3bc0ba42010-12-13 19:38:09 -0500336 if (atomic_read(&files->count) == 1) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337 file = fcheck_files(files, fd);
Al Viro1abf0c72011-03-13 03:51:11 -0400338 if (file && (file->f_mode & FMODE_PATH))
339 file = NULL;
340 } else {
341 rcu_read_lock();
342 file = fcheck_files(files, fd);
343 if (file) {
344 if (!(file->f_mode & FMODE_PATH) &&
345 atomic_long_inc_not_zero(&file->f_count))
346 *fput_needed = 1;
347 else
348 /* Didn't get the reference, someone's freed */
349 file = NULL;
350 }
351 rcu_read_unlock();
352 }
353
354 return file;
355}
356
357struct file *fget_raw_light(unsigned int fd, int *fput_needed)
358{
359 struct file *file;
360 struct files_struct *files = current->files;
361
362 *fput_needed = 0;
363 if (atomic_read(&files->count) == 1) {
364 file = fcheck_files(files, fd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365 } else {
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700366 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367 file = fcheck_files(files, fd);
368 if (file) {
Al Viro516e0cc2008-07-26 00:39:17 -0400369 if (atomic_long_inc_not_zero(&file->f_count))
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700370 *fput_needed = 1;
371 else
372 /* Didn't get the reference, someone's freed */
373 file = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374 }
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700375 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376 }
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700377
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378 return file;
379}
380
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381void put_filp(struct file *file)
382{
Al Viro516e0cc2008-07-26 00:39:17 -0400383 if (atomic_long_dec_and_test(&file->f_count)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384 security_file_free(file);
Nick Pigginee2ffa02010-08-18 04:37:35 +1000385 file_sb_list_del(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386 file_free(file);
387 }
388}
389
Nick Piggin6416ccb2010-08-18 04:37:38 +1000390static inline int file_list_cpu(struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391{
Nick Piggin6416ccb2010-08-18 04:37:38 +1000392#ifdef CONFIG_SMP
393 return file->f_sb_list_cpu;
394#else
395 return smp_processor_id();
396#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397}
398
Nick Piggin6416ccb2010-08-18 04:37:38 +1000399/* helper for file_sb_list_add to reduce ifdefs */
400static inline void __file_sb_list_add(struct file *file, struct super_block *sb)
401{
402 struct list_head *list;
403#ifdef CONFIG_SMP
404 int cpu;
405 cpu = smp_processor_id();
406 file->f_sb_list_cpu = cpu;
407 list = per_cpu_ptr(sb->s_files, cpu);
408#else
409 list = &sb->s_files;
410#endif
411 list_add(&file->f_u.fu_list, list);
412}
413
414/**
415 * file_sb_list_add - add a file to the sb's file list
416 * @file: file to add
417 * @sb: sb to add it to
418 *
419 * Use this function to associate a file with the superblock of the inode it
420 * refers to.
421 */
422void file_sb_list_add(struct file *file, struct super_block *sb)
423{
424 lg_local_lock(files_lglock);
425 __file_sb_list_add(file, sb);
426 lg_local_unlock(files_lglock);
427}
428
429/**
430 * file_sb_list_del - remove a file from the sb's file list
431 * @file: file to remove
432 * @sb: sb to remove it from
433 *
434 * Use this function to remove a file from its superblock.
435 */
Nick Pigginee2ffa02010-08-18 04:37:35 +1000436void file_sb_list_del(struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437{
Eric Dumazet2f512012005-10-30 15:02:16 -0800438 if (!list_empty(&file->f_u.fu_list)) {
Nick Piggin6416ccb2010-08-18 04:37:38 +1000439 lg_local_lock_cpu(files_lglock, file_list_cpu(file));
Eric Dumazet2f512012005-10-30 15:02:16 -0800440 list_del_init(&file->f_u.fu_list);
Nick Piggin6416ccb2010-08-18 04:37:38 +1000441 lg_local_unlock_cpu(files_lglock, file_list_cpu(file));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442 }
443}
444
Nick Piggin6416ccb2010-08-18 04:37:38 +1000445#ifdef CONFIG_SMP
446
447/*
448 * These macros iterate all files on all CPUs for a given superblock.
449 * files_lglock must be held globally.
450 */
451#define do_file_list_for_each_entry(__sb, __file) \
452{ \
453 int i; \
454 for_each_possible_cpu(i) { \
455 struct list_head *list; \
456 list = per_cpu_ptr((__sb)->s_files, i); \
457 list_for_each_entry((__file), list, f_u.fu_list)
458
459#define while_file_list_for_each_entry \
460 } \
461}
462
463#else
464
465#define do_file_list_for_each_entry(__sb, __file) \
466{ \
467 struct list_head *list; \
468 list = &(sb)->s_files; \
469 list_for_each_entry((__file), list, f_u.fu_list)
470
471#define while_file_list_for_each_entry \
472}
473
474#endif
475
npiggin@suse.de864d7c42009-04-26 20:25:56 +1000476/**
477 * mark_files_ro - mark all files read-only
478 * @sb: superblock in question
479 *
480 * All files are marked read-only. We don't care about pending
481 * delete files so this should be used in 'force' mode only.
482 */
483void mark_files_ro(struct super_block *sb)
484{
485 struct file *f;
486
487retry:
Nick Piggin6416ccb2010-08-18 04:37:38 +1000488 lg_global_lock(files_lglock);
489 do_file_list_for_each_entry(sb, f) {
npiggin@suse.de864d7c42009-04-26 20:25:56 +1000490 struct vfsmount *mnt;
491 if (!S_ISREG(f->f_path.dentry->d_inode->i_mode))
492 continue;
493 if (!file_count(f))
494 continue;
495 if (!(f->f_mode & FMODE_WRITE))
496 continue;
Wu Fengguang42e49602010-03-05 13:42:01 -0800497 spin_lock(&f->f_lock);
npiggin@suse.de864d7c42009-04-26 20:25:56 +1000498 f->f_mode &= ~FMODE_WRITE;
Wu Fengguang42e49602010-03-05 13:42:01 -0800499 spin_unlock(&f->f_lock);
npiggin@suse.de864d7c42009-04-26 20:25:56 +1000500 if (file_check_writeable(f) != 0)
501 continue;
502 file_release_write(f);
503 mnt = mntget(f->f_path.mnt);
Nick Pigginee2ffa02010-08-18 04:37:35 +1000504 /* This can sleep, so we can't hold the spinlock. */
Nick Piggin6416ccb2010-08-18 04:37:38 +1000505 lg_global_unlock(files_lglock);
npiggin@suse.de864d7c42009-04-26 20:25:56 +1000506 mnt_drop_write(mnt);
507 mntput(mnt);
508 goto retry;
Nick Piggin6416ccb2010-08-18 04:37:38 +1000509 } while_file_list_for_each_entry;
510 lg_global_unlock(files_lglock);
npiggin@suse.de864d7c42009-04-26 20:25:56 +1000511}
512
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513void __init files_init(unsigned long mempages)
514{
Eric Dumazet518de9b2010-10-26 14:22:44 -0700515 unsigned long n;
Eric Dumazetb6b3fde2008-12-10 09:35:45 -0800516
517 filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
518 SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
519
520 /*
521 * One file with associated inode and dcache is very roughly 1K.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522 * Per default don't use more than 10% of our memory for files.
523 */
524
525 n = (mempages * (PAGE_SIZE / 1024)) / 10;
Eric Dumazet518de9b2010-10-26 14:22:44 -0700526 files_stat.max_files = max_t(unsigned long, n, NR_FILE);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700527 files_defer_init();
Nick Piggin6416ccb2010-08-18 04:37:38 +1000528 lg_lock_init(files_lglock);
Mingming Cao0216bfc2006-06-23 02:05:41 -0700529 percpu_counter_init(&nr_files, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530}