blob: a305d9e2d1b2aac05dcd456bdd23885652272439 [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 +100037DEFINE_LGLOCK(files_lglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070038
Eric Dumazetb6b3fde2008-12-10 09:35:45 -080039/* SLAB cache for file structures */
40static struct kmem_cache *filp_cachep __read_mostly;
41
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080042static struct percpu_counter nr_files __cacheline_aligned_in_smp;
Linus Torvalds1da177e2005-04-16 15:20:36 -070043
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070044static inline void file_free_rcu(struct rcu_head *head)
45{
David Howellsd76b0d92008-11-14 10:39:25 +110046 struct file *f = container_of(head, struct file, f_u.fu_rcuhead);
47
48 put_cred(f->f_cred);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070049 kmem_cache_free(filp_cachep, f);
50}
51
Linus Torvalds1da177e2005-04-16 15:20:36 -070052static inline void file_free(struct file *f)
53{
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080054 percpu_counter_dec(&nr_files);
Dave Hansenad775f52008-02-15 14:38:01 -080055 file_check_state(f);
Eric Dumazet2f512012005-10-30 15:02:16 -080056 call_rcu(&f->f_u.fu_rcuhead, file_free_rcu);
Linus Torvalds1da177e2005-04-16 15:20:36 -070057}
58
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080059/*
60 * Return the total number of open files in the system
61 */
Eric Dumazet518de9b2010-10-26 14:22:44 -070062static long get_nr_files(void)
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080063{
64 return percpu_counter_read_positive(&nr_files);
65}
66
67/*
68 * Return the maximum number of open files in the system
69 */
Eric Dumazet518de9b2010-10-26 14:22:44 -070070unsigned long get_max_files(void)
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080071{
72 return files_stat.max_files;
73}
74EXPORT_SYMBOL_GPL(get_max_files);
75
76/*
77 * Handle nr_files sysctl
78 */
79#if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
Alexey Dobriyan8d65af72009-09-23 15:57:19 -070080int proc_nr_files(ctl_table *table, int write,
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080081 void __user *buffer, size_t *lenp, loff_t *ppos)
82{
83 files_stat.nr_files = get_nr_files();
Eric Dumazet518de9b2010-10-26 14:22:44 -070084 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080085}
86#else
Alexey Dobriyan8d65af72009-09-23 15:57:19 -070087int proc_nr_files(ctl_table *table, int write,
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080088 void __user *buffer, size_t *lenp, loff_t *ppos)
89{
90 return -ENOSYS;
91}
92#endif
93
Linus Torvalds1da177e2005-04-16 15:20:36 -070094/* Find an unused file structure and return a pointer to it.
95 * Returns NULL, if there are no more free file structures or
96 * we run out of memory.
Dave Hansen430e2852008-02-15 14:37:26 -080097 *
98 * Be very careful using this. You are responsible for
99 * getting write access to any mount that you might assign
100 * to this filp, if it is opened for write. If this is not
101 * done, you will imbalance int the mount's writer count
102 * and a warning at __fput() time.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103 */
104struct file *get_empty_filp(void)
105{
David Howells86a264a2008-11-14 10:39:18 +1100106 const struct cred *cred = current_cred();
Eric Dumazet518de9b2010-10-26 14:22:44 -0700107 static long old_max;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108 struct file * f;
109
110 /*
111 * Privileged users can go above max_files
112 */
Dipankar Sarma529bf6b2006-03-07 21:55:35 -0800113 if (get_nr_files() >= files_stat.max_files && !capable(CAP_SYS_ADMIN)) {
114 /*
115 * percpu_counters are inaccurate. Do an expensive check before
116 * we go and fail.
117 */
Peter Zijlstra52d9f3b2007-10-16 23:25:44 -0700118 if (percpu_counter_sum_positive(&nr_files) >= files_stat.max_files)
Dipankar Sarma529bf6b2006-03-07 21:55:35 -0800119 goto over;
120 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121
Denis Cheng4975e452007-10-16 23:26:19 -0700122 f = kmem_cache_zalloc(filp_cachep, GFP_KERNEL);
Kirill Korotaevaf4d2ec2005-06-23 00:09:50 -0700123 if (f == NULL)
124 goto fail;
125
Dipankar Sarma529bf6b2006-03-07 21:55:35 -0800126 percpu_counter_inc(&nr_files);
Tetsuo Handa78d29782011-02-04 18:13:24 +0000127 f->f_cred = get_cred(cred);
Kirill Korotaevaf4d2ec2005-06-23 00:09:50 -0700128 if (security_file_alloc(f))
129 goto fail_sec;
130
Eric Dumazet2f512012005-10-30 15:02:16 -0800131 INIT_LIST_HEAD(&f->f_u.fu_list);
Al Viro516e0cc2008-07-26 00:39:17 -0400132 atomic_long_set(&f->f_count, 1);
Benjamin LaHaise5a6b79512006-03-23 03:01:03 -0800133 rwlock_init(&f->f_owner.lock);
Jonathan Corbet68499912009-02-06 13:52:43 -0700134 spin_lock_init(&f->f_lock);
Benjamin LaHaise5a6b79512006-03-23 03:01:03 -0800135 eventpoll_init_file(f);
136 /* f->f_version: 0 */
Kirill Korotaevaf4d2ec2005-06-23 00:09:50 -0700137 return f;
138
139over:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140 /* Ran out of filps - report that */
Dipankar Sarma529bf6b2006-03-07 21:55:35 -0800141 if (get_nr_files() > old_max) {
Eric Dumazet518de9b2010-10-26 14:22:44 -0700142 pr_info("VFS: file-max limit %lu reached\n", get_max_files());
Dipankar Sarma529bf6b2006-03-07 21:55:35 -0800143 old_max = get_nr_files();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144 }
Kirill Korotaevaf4d2ec2005-06-23 00:09:50 -0700145 goto fail;
146
147fail_sec:
148 file_free(f);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149fail:
150 return NULL;
151}
152
Dave Hansence8d2cd2007-10-16 23:31:13 -0700153/**
154 * alloc_file - allocate and initialize a 'struct file'
155 * @mnt: the vfsmount on which the file will reside
156 * @dentry: the dentry representing the new file
157 * @mode: the mode with which the new file will be opened
158 * @fop: the 'struct file_operations' for the new file
159 *
160 * Use this instead of get_empty_filp() to get a new
161 * 'struct file'. Do so because of the same initialization
162 * pitfalls reasons listed for init_file(). This is a
163 * preferred interface to using init_file().
164 *
165 * If all the callers of init_file() are eliminated, its
166 * code should be moved into this function.
167 */
Al Viro2c48b9c2009-08-09 00:52:35 +0400168struct file *alloc_file(struct path *path, fmode_t mode,
169 const struct file_operations *fop)
Dave Hansence8d2cd2007-10-16 23:31:13 -0700170{
171 struct file *file;
Dave Hansence8d2cd2007-10-16 23:31:13 -0700172
173 file = get_empty_filp();
174 if (!file)
175 return NULL;
176
Al Viro2c48b9c2009-08-09 00:52:35 +0400177 file->f_path = *path;
178 file->f_mapping = path->dentry->d_inode->i_mapping;
Dave Hansence8d2cd2007-10-16 23:31:13 -0700179 file->f_mode = mode;
180 file->f_op = fop;
Dave Hansen4a3fd212008-02-15 14:37:48 -0800181
182 /*
183 * These mounts don't really matter in practice
184 * for r/o bind mounts. They aren't userspace-
185 * visible. We do this for consistency, and so
186 * that we can do debugging checks at __fput()
187 */
Al Viro2c48b9c2009-08-09 00:52:35 +0400188 if ((mode & FMODE_WRITE) && !special_file(path->dentry->d_inode->i_mode)) {
Dave Hansenad775f52008-02-15 14:38:01 -0800189 file_take_write(file);
Roland Dreier385e3ed2009-12-16 12:48:44 -0800190 WARN_ON(mnt_clone_write(path->mnt));
Dave Hansen4a3fd212008-02-15 14:37:48 -0800191 }
Mimi Zohar890275b52010-11-02 10:13:07 -0400192 if ((mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
193 i_readcount_inc(path->dentry->d_inode);
Al Viro3d1e4632009-08-08 23:56:29 +0400194 return file;
Dave Hansence8d2cd2007-10-16 23:31:13 -0700195}
Roland Dreier73efc462009-12-16 12:43:11 -0800196EXPORT_SYMBOL(alloc_file);
Dave Hansence8d2cd2007-10-16 23:31:13 -0700197
Dave Hansenaceaf782008-02-15 14:37:31 -0800198/**
199 * drop_file_write_access - give up ability to write to a file
200 * @file: the file to which we will stop writing
201 *
202 * This is a central place which will give up the ability
203 * to write to @file, along with access to write through
204 * its vfsmount.
205 */
Al Virob57ce962012-02-12 02:38:16 -0500206static void drop_file_write_access(struct file *file)
Dave Hansenaceaf782008-02-15 14:37:31 -0800207{
Dave Hansen4a3fd212008-02-15 14:37:48 -0800208 struct vfsmount *mnt = file->f_path.mnt;
Dave Hansenaceaf782008-02-15 14:37:31 -0800209 struct dentry *dentry = file->f_path.dentry;
210 struct inode *inode = dentry->d_inode;
211
212 put_write_access(inode);
Dave Hansenad775f52008-02-15 14:38:01 -0800213
214 if (special_file(inode->i_mode))
215 return;
216 if (file_check_writeable(file) != 0)
217 return;
218 mnt_drop_write(mnt);
219 file_release_write(file);
Dave Hansenaceaf782008-02-15 14:37:31 -0800220}
Dave Hansenaceaf782008-02-15 14:37:31 -0800221
Al Virod7065da2010-05-26 15:13:55 -0400222/* the real guts of fput() - releasing the last reference to file
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223 */
Al Virod7065da2010-05-26 15:13:55 -0400224static void __fput(struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800226 struct dentry *dentry = file->f_path.dentry;
227 struct vfsmount *mnt = file->f_path.mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228 struct inode *inode = dentry->d_inode;
229
230 might_sleep();
Robert Love0eeca282005-07-12 17:06:03 -0400231
232 fsnotify_close(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233 /*
234 * The function eventpoll_release() should be the first called
235 * in the file cleanup chain.
236 */
237 eventpoll_release(file);
238 locks_remove_flock(file);
239
Al Viro233e70f2008-10-31 23:28:30 +0000240 if (unlikely(file->f_flags & FASYNC)) {
241 if (file->f_op && file->f_op->fasync)
242 file->f_op->fasync(-1, file, 0);
243 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244 if (file->f_op && file->f_op->release)
245 file->f_op->release(inode, file);
246 security_file_free(file);
Al Viro89068c52010-02-07 03:07:29 -0500247 ima_file_free(file);
Miklos Szeredi60ed8cf2011-03-16 18:17:54 +0100248 if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL &&
249 !(file->f_mode & FMODE_PATH))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250 cdev_put(inode->i_cdev);
Miklos Szeredi60ed8cf2011-03-16 18:17:54 +0100251 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252 fops_put(file->f_op);
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700253 put_pid(file->f_owner.pid);
Nick Pigginee2ffa02010-08-18 04:37:35 +1000254 file_sb_list_del(file);
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 Virod7065da2010-05-26 15:13:55 -0400266void fput(struct file *file)
267{
268 if (atomic_long_dec_and_test(&file->f_count))
269 __fput(file);
270}
271
272EXPORT_SYMBOL(fput);
273
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -0800274struct file *fget(unsigned int fd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275{
276 struct file *file;
277 struct files_struct *files = current->files;
278
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700279 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280 file = fcheck_files(files, fd);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700281 if (file) {
Al Viro1abf0c72011-03-13 03:51:11 -0400282 /* File object ref couldn't be taken */
283 if (file->f_mode & FMODE_PATH ||
284 !atomic_long_inc_not_zero(&file->f_count))
285 file = NULL;
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700286 }
287 rcu_read_unlock();
288
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289 return file;
290}
291
292EXPORT_SYMBOL(fget);
293
Al Viro1abf0c72011-03-13 03:51:11 -0400294struct file *fget_raw(unsigned int fd)
295{
296 struct file *file;
297 struct files_struct *files = current->files;
298
299 rcu_read_lock();
300 file = fcheck_files(files, fd);
301 if (file) {
302 /* File object ref couldn't be taken */
303 if (!atomic_long_inc_not_zero(&file->f_count))
304 file = NULL;
305 }
306 rcu_read_unlock();
307
308 return file;
309}
310
Al Viro326be7b2011-03-13 17:08:22 -0400311EXPORT_SYMBOL(fget_raw);
312
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313/*
Tony Battersby58939472010-08-10 18:01:29 -0700314 * Lightweight file lookup - no refcnt increment if fd table isn't shared.
315 *
316 * You can use this instead of fget if you satisfy all of the following
317 * conditions:
318 * 1) You must call fput_light before exiting the syscall and returning control
319 * to userspace (i.e. you cannot remember the returned struct file * after
320 * returning to userspace).
321 * 2) You must not call filp_close on the returned struct file * in between
322 * calls to fget_light and fput_light.
323 * 3) You must not clone the current task in between the calls to fget_light
324 * and fput_light.
325 *
326 * The fput_needed flag returned by fget_light should be passed to the
327 * corresponding fput_light.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328 */
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -0800329struct file *fget_light(unsigned int fd, int *fput_needed)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330{
331 struct file *file;
332 struct files_struct *files = current->files;
333
334 *fput_needed = 0;
Steven Rostedt3bc0ba42010-12-13 19:38:09 -0500335 if (atomic_read(&files->count) == 1) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336 file = fcheck_files(files, fd);
Al Viro1abf0c72011-03-13 03:51:11 -0400337 if (file && (file->f_mode & FMODE_PATH))
338 file = NULL;
339 } else {
340 rcu_read_lock();
341 file = fcheck_files(files, fd);
342 if (file) {
343 if (!(file->f_mode & FMODE_PATH) &&
344 atomic_long_inc_not_zero(&file->f_count))
345 *fput_needed = 1;
346 else
347 /* Didn't get the reference, someone's freed */
348 file = NULL;
349 }
350 rcu_read_unlock();
351 }
352
353 return file;
354}
355
356struct file *fget_raw_light(unsigned int fd, int *fput_needed)
357{
358 struct file *file;
359 struct files_struct *files = current->files;
360
361 *fput_needed = 0;
362 if (atomic_read(&files->count) == 1) {
363 file = fcheck_files(files, fd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364 } else {
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700365 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366 file = fcheck_files(files, fd);
367 if (file) {
Al Viro516e0cc2008-07-26 00:39:17 -0400368 if (atomic_long_inc_not_zero(&file->f_count))
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700369 *fput_needed = 1;
370 else
371 /* Didn't get the reference, someone's freed */
372 file = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373 }
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700374 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375 }
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700376
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377 return file;
378}
379
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380void put_filp(struct file *file)
381{
Al Viro516e0cc2008-07-26 00:39:17 -0400382 if (atomic_long_dec_and_test(&file->f_count)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383 security_file_free(file);
Nick Pigginee2ffa02010-08-18 04:37:35 +1000384 file_sb_list_del(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385 file_free(file);
386 }
387}
388
Nick Piggin6416ccb2010-08-18 04:37:38 +1000389static inline int file_list_cpu(struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390{
Nick Piggin6416ccb2010-08-18 04:37:38 +1000391#ifdef CONFIG_SMP
392 return file->f_sb_list_cpu;
393#else
394 return smp_processor_id();
395#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396}
397
Nick Piggin6416ccb2010-08-18 04:37:38 +1000398/* helper for file_sb_list_add to reduce ifdefs */
399static inline void __file_sb_list_add(struct file *file, struct super_block *sb)
400{
401 struct list_head *list;
402#ifdef CONFIG_SMP
403 int cpu;
404 cpu = smp_processor_id();
405 file->f_sb_list_cpu = cpu;
406 list = per_cpu_ptr(sb->s_files, cpu);
407#else
408 list = &sb->s_files;
409#endif
410 list_add(&file->f_u.fu_list, list);
411}
412
413/**
414 * file_sb_list_add - add a file to the sb's file list
415 * @file: file to add
416 * @sb: sb to add it to
417 *
418 * Use this function to associate a file with the superblock of the inode it
419 * refers to.
420 */
421void file_sb_list_add(struct file *file, struct super_block *sb)
422{
Andi Kleen962830d2012-05-08 13:32:02 +0930423 lg_local_lock(&files_lglock);
Nick Piggin6416ccb2010-08-18 04:37:38 +1000424 __file_sb_list_add(file, sb);
Andi Kleen962830d2012-05-08 13:32:02 +0930425 lg_local_unlock(&files_lglock);
Nick Piggin6416ccb2010-08-18 04:37:38 +1000426}
427
428/**
429 * file_sb_list_del - remove a file from the sb's file list
430 * @file: file to remove
431 * @sb: sb to remove it from
432 *
433 * Use this function to remove a file from its superblock.
434 */
Nick Pigginee2ffa02010-08-18 04:37:35 +1000435void file_sb_list_del(struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436{
Eric Dumazet2f512012005-10-30 15:02:16 -0800437 if (!list_empty(&file->f_u.fu_list)) {
Andi Kleen962830d2012-05-08 13:32:02 +0930438 lg_local_lock_cpu(&files_lglock, file_list_cpu(file));
Eric Dumazet2f512012005-10-30 15:02:16 -0800439 list_del_init(&file->f_u.fu_list);
Andi Kleen962830d2012-05-08 13:32:02 +0930440 lg_local_unlock_cpu(&files_lglock, file_list_cpu(file));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 }
442}
443
Nick Piggin6416ccb2010-08-18 04:37:38 +1000444#ifdef CONFIG_SMP
445
446/*
447 * These macros iterate all files on all CPUs for a given superblock.
448 * files_lglock must be held globally.
449 */
450#define do_file_list_for_each_entry(__sb, __file) \
451{ \
452 int i; \
453 for_each_possible_cpu(i) { \
454 struct list_head *list; \
455 list = per_cpu_ptr((__sb)->s_files, i); \
456 list_for_each_entry((__file), list, f_u.fu_list)
457
458#define while_file_list_for_each_entry \
459 } \
460}
461
462#else
463
464#define do_file_list_for_each_entry(__sb, __file) \
465{ \
466 struct list_head *list; \
467 list = &(sb)->s_files; \
468 list_for_each_entry((__file), list, f_u.fu_list)
469
470#define while_file_list_for_each_entry \
471}
472
473#endif
474
npiggin@suse.de864d7c42009-04-26 20:25:56 +1000475/**
476 * mark_files_ro - mark all files read-only
477 * @sb: superblock in question
478 *
479 * All files are marked read-only. We don't care about pending
480 * delete files so this should be used in 'force' mode only.
481 */
482void mark_files_ro(struct super_block *sb)
483{
484 struct file *f;
485
486retry:
Andi Kleen962830d2012-05-08 13:32:02 +0930487 lg_global_lock(&files_lglock);
Nick Piggin6416ccb2010-08-18 04:37:38 +1000488 do_file_list_for_each_entry(sb, f) {
npiggin@suse.de864d7c42009-04-26 20:25:56 +1000489 struct vfsmount *mnt;
490 if (!S_ISREG(f->f_path.dentry->d_inode->i_mode))
491 continue;
492 if (!file_count(f))
493 continue;
494 if (!(f->f_mode & FMODE_WRITE))
495 continue;
Wu Fengguang42e49602010-03-05 13:42:01 -0800496 spin_lock(&f->f_lock);
npiggin@suse.de864d7c42009-04-26 20:25:56 +1000497 f->f_mode &= ~FMODE_WRITE;
Wu Fengguang42e49602010-03-05 13:42:01 -0800498 spin_unlock(&f->f_lock);
npiggin@suse.de864d7c42009-04-26 20:25:56 +1000499 if (file_check_writeable(f) != 0)
500 continue;
501 file_release_write(f);
502 mnt = mntget(f->f_path.mnt);
Nick Pigginee2ffa02010-08-18 04:37:35 +1000503 /* This can sleep, so we can't hold the spinlock. */
Andi Kleen962830d2012-05-08 13:32:02 +0930504 lg_global_unlock(&files_lglock);
npiggin@suse.de864d7c42009-04-26 20:25:56 +1000505 mnt_drop_write(mnt);
506 mntput(mnt);
507 goto retry;
Nick Piggin6416ccb2010-08-18 04:37:38 +1000508 } while_file_list_for_each_entry;
Andi Kleen962830d2012-05-08 13:32:02 +0930509 lg_global_unlock(&files_lglock);
npiggin@suse.de864d7c42009-04-26 20:25:56 +1000510}
511
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512void __init files_init(unsigned long mempages)
513{
Eric Dumazet518de9b2010-10-26 14:22:44 -0700514 unsigned long n;
Eric Dumazetb6b3fde2008-12-10 09:35:45 -0800515
516 filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
517 SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
518
519 /*
520 * One file with associated inode and dcache is very roughly 1K.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521 * Per default don't use more than 10% of our memory for files.
522 */
523
524 n = (mempages * (PAGE_SIZE / 1024)) / 10;
Eric Dumazet518de9b2010-10-26 14:22:44 -0700525 files_stat.max_files = max_t(unsigned long, n, NR_FILE);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700526 files_defer_init();
Andi Kleen962830d2012-05-08 13:32:02 +0930527 lg_lock_init(&files_lglock, "files_lglock");
Mingming Cao0216bfc2006-06-23 02:05:41 -0700528 percpu_counter_init(&nr_files, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529}