blob: 701985e4ccda4fc5afc05977f511bd6ca11ba3af [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
Nick Piggin6416ccb2010-08-18 04:37:38 +100039DEFINE_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.
97 * Returns NULL, if there are no more free file structures or
98 * we run out of memory.
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;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110 struct file * f;
111
112 /*
113 * Privileged users can go above max_files
114 */
Dipankar Sarma529bf6b2006-03-07 21:55:35 -0800115 if (get_nr_files() >= files_stat.max_files && !capable(CAP_SYS_ADMIN)) {
116 /*
117 * percpu_counters are inaccurate. Do an expensive check before
118 * we go and fail.
119 */
Peter Zijlstra52d9f3b2007-10-16 23:25:44 -0700120 if (percpu_counter_sum_positive(&nr_files) >= files_stat.max_files)
Dipankar Sarma529bf6b2006-03-07 21:55:35 -0800121 goto over;
122 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123
Denis Cheng4975e452007-10-16 23:26:19 -0700124 f = kmem_cache_zalloc(filp_cachep, GFP_KERNEL);
Kirill Korotaevaf4d2ec2005-06-23 00:09:50 -0700125 if (f == NULL)
126 goto fail;
127
Dipankar Sarma529bf6b2006-03-07 21:55:35 -0800128 percpu_counter_inc(&nr_files);
Tetsuo Handa78d29782011-02-04 18:13:24 +0000129 f->f_cred = get_cred(cred);
Kirill Korotaevaf4d2ec2005-06-23 00:09:50 -0700130 if (security_file_alloc(f))
131 goto fail_sec;
132
Eric Dumazet2f512012005-10-30 15:02:16 -0800133 INIT_LIST_HEAD(&f->f_u.fu_list);
Al Viro516e0cc2008-07-26 00:39:17 -0400134 atomic_long_set(&f->f_count, 1);
Benjamin LaHaise5a6b79512006-03-23 03:01:03 -0800135 rwlock_init(&f->f_owner.lock);
Jonathan Corbet68499912009-02-06 13:52:43 -0700136 spin_lock_init(&f->f_lock);
Benjamin LaHaise5a6b79512006-03-23 03:01:03 -0800137 eventpoll_init_file(f);
138 /* f->f_version: 0 */
Kirill Korotaevaf4d2ec2005-06-23 00:09:50 -0700139 return f;
140
141over:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142 /* Ran out of filps - report that */
Dipankar Sarma529bf6b2006-03-07 21:55:35 -0800143 if (get_nr_files() > old_max) {
Eric Dumazet518de9b2010-10-26 14:22:44 -0700144 pr_info("VFS: file-max limit %lu reached\n", get_max_files());
Dipankar Sarma529bf6b2006-03-07 21:55:35 -0800145 old_max = get_nr_files();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146 }
Kirill Korotaevaf4d2ec2005-06-23 00:09:50 -0700147 goto fail;
148
149fail_sec:
150 file_free(f);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151fail:
152 return NULL;
153}
154
Dave Hansence8d2cd2007-10-16 23:31:13 -0700155/**
156 * alloc_file - allocate and initialize a 'struct file'
157 * @mnt: the vfsmount on which the file will reside
158 * @dentry: the dentry representing the new file
159 * @mode: the mode with which the new file will be opened
160 * @fop: the 'struct file_operations' for the new file
161 *
162 * Use this instead of get_empty_filp() to get a new
163 * 'struct file'. Do so because of the same initialization
164 * pitfalls reasons listed for init_file(). This is a
165 * preferred interface to using init_file().
166 *
167 * If all the callers of init_file() are eliminated, its
168 * code should be moved into this function.
169 */
Al Viro2c48b9c2009-08-09 00:52:35 +0400170struct file *alloc_file(struct path *path, fmode_t mode,
171 const struct file_operations *fop)
Dave Hansence8d2cd2007-10-16 23:31:13 -0700172{
173 struct file *file;
Dave Hansence8d2cd2007-10-16 23:31:13 -0700174
175 file = get_empty_filp();
176 if (!file)
177 return NULL;
178
Al Viro2c48b9c2009-08-09 00:52:35 +0400179 file->f_path = *path;
180 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;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230 struct inode *inode = dentry->d_inode;
231
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 }
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);
Al Viro89068c52010-02-07 03:07:29 -0500249 ima_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;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262 file_free(file);
263 dput(dentry);
264 mntput(mnt);
265}
266
Al Viro4a9d4b02012-06-24 09:56:45 +0400267static DEFINE_SPINLOCK(delayed_fput_lock);
268static LIST_HEAD(delayed_fput_list);
269static void delayed_fput(struct work_struct *unused)
270{
271 LIST_HEAD(head);
272 spin_lock_irq(&delayed_fput_lock);
273 list_splice_init(&delayed_fput_list, &head);
274 spin_unlock_irq(&delayed_fput_lock);
275 while (!list_empty(&head)) {
276 struct file *f = list_first_entry(&head, struct file, f_u.fu_list);
277 list_del_init(&f->f_u.fu_list);
278 __fput(f);
279 }
280}
281
282static void ____fput(struct callback_head *work)
283{
284 __fput(container_of(work, struct file, f_u.fu_rcuhead));
285}
286
287/*
288 * If kernel thread really needs to have the final fput() it has done
289 * to complete, call this. The only user right now is the boot - we
290 * *do* need to make sure our writes to binaries on initramfs has
291 * not left us with opened struct file waiting for __fput() - execve()
292 * won't work without that. Please, don't add more callers without
293 * very good reasons; in particular, never call that with locks
294 * held and never call that from a thread that might need to do
295 * some work on any kind of umount.
296 */
297void flush_delayed_fput(void)
298{
299 delayed_fput(NULL);
300}
301
302static DECLARE_WORK(delayed_fput_work, delayed_fput);
303
Al Virod7065da2010-05-26 15:13:55 -0400304void fput(struct file *file)
305{
Al Viro4a9d4b02012-06-24 09:56:45 +0400306 if (atomic_long_dec_and_test(&file->f_count)) {
307 struct task_struct *task = current;
308 file_sb_list_del(file);
309 if (unlikely(in_interrupt() || task->flags & PF_KTHREAD)) {
310 unsigned long flags;
311 spin_lock_irqsave(&delayed_fput_lock, flags);
312 list_add(&file->f_u.fu_list, &delayed_fput_list);
313 schedule_work(&delayed_fput_work);
314 spin_unlock_irqrestore(&delayed_fput_lock, flags);
315 return;
316 }
317 init_task_work(&file->f_u.fu_rcuhead, ____fput);
318 task_work_add(task, &file->f_u.fu_rcuhead, true);
319 }
320}
321
322/*
323 * synchronous analog of fput(); for kernel threads that might be needed
324 * in some umount() (and thus can't use flush_delayed_fput() without
325 * risking deadlocks), need to wait for completion of __fput() and know
326 * for this specific struct file it won't involve anything that would
327 * need them. Use only if you really need it - at the very least,
328 * don't blindly convert fput() by kernel thread to that.
329 */
330void __fput_sync(struct file *file)
331{
332 if (atomic_long_dec_and_test(&file->f_count)) {
333 struct task_struct *task = current;
334 file_sb_list_del(file);
335 BUG_ON(!(task->flags & PF_KTHREAD));
Al Virod7065da2010-05-26 15:13:55 -0400336 __fput(file);
Al Viro4a9d4b02012-06-24 09:56:45 +0400337 }
Al Virod7065da2010-05-26 15:13:55 -0400338}
339
340EXPORT_SYMBOL(fput);
341
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -0800342struct file *fget(unsigned int fd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343{
344 struct file *file;
345 struct files_struct *files = current->files;
346
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700347 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348 file = fcheck_files(files, fd);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700349 if (file) {
Al Viro1abf0c72011-03-13 03:51:11 -0400350 /* File object ref couldn't be taken */
351 if (file->f_mode & FMODE_PATH ||
352 !atomic_long_inc_not_zero(&file->f_count))
353 file = NULL;
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700354 }
355 rcu_read_unlock();
356
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357 return file;
358}
359
360EXPORT_SYMBOL(fget);
361
Al Viro1abf0c72011-03-13 03:51:11 -0400362struct file *fget_raw(unsigned int fd)
363{
364 struct file *file;
365 struct files_struct *files = current->files;
366
367 rcu_read_lock();
368 file = fcheck_files(files, fd);
369 if (file) {
370 /* File object ref couldn't be taken */
371 if (!atomic_long_inc_not_zero(&file->f_count))
372 file = NULL;
373 }
374 rcu_read_unlock();
375
376 return file;
377}
378
Al Viro326be7b2011-03-13 17:08:22 -0400379EXPORT_SYMBOL(fget_raw);
380
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381/*
Tony Battersby58939472010-08-10 18:01:29 -0700382 * Lightweight file lookup - no refcnt increment if fd table isn't shared.
383 *
384 * You can use this instead of fget if you satisfy all of the following
385 * conditions:
386 * 1) You must call fput_light before exiting the syscall and returning control
387 * to userspace (i.e. you cannot remember the returned struct file * after
388 * returning to userspace).
389 * 2) You must not call filp_close on the returned struct file * in between
390 * calls to fget_light and fput_light.
391 * 3) You must not clone the current task in between the calls to fget_light
392 * and fput_light.
393 *
394 * The fput_needed flag returned by fget_light should be passed to the
395 * corresponding fput_light.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396 */
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -0800397struct file *fget_light(unsigned int fd, int *fput_needed)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398{
399 struct file *file;
400 struct files_struct *files = current->files;
401
402 *fput_needed = 0;
Steven Rostedt3bc0ba42010-12-13 19:38:09 -0500403 if (atomic_read(&files->count) == 1) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404 file = fcheck_files(files, fd);
Al Viro1abf0c72011-03-13 03:51:11 -0400405 if (file && (file->f_mode & FMODE_PATH))
406 file = NULL;
407 } else {
408 rcu_read_lock();
409 file = fcheck_files(files, fd);
410 if (file) {
411 if (!(file->f_mode & FMODE_PATH) &&
412 atomic_long_inc_not_zero(&file->f_count))
413 *fput_needed = 1;
414 else
415 /* Didn't get the reference, someone's freed */
416 file = NULL;
417 }
418 rcu_read_unlock();
419 }
420
421 return file;
422}
423
424struct file *fget_raw_light(unsigned int fd, int *fput_needed)
425{
426 struct file *file;
427 struct files_struct *files = current->files;
428
429 *fput_needed = 0;
430 if (atomic_read(&files->count) == 1) {
431 file = fcheck_files(files, fd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432 } else {
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700433 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434 file = fcheck_files(files, fd);
435 if (file) {
Al Viro516e0cc2008-07-26 00:39:17 -0400436 if (atomic_long_inc_not_zero(&file->f_count))
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700437 *fput_needed = 1;
438 else
439 /* Didn't get the reference, someone's freed */
440 file = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 }
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700442 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443 }
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700444
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445 return file;
446}
447
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448void put_filp(struct file *file)
449{
Al Viro516e0cc2008-07-26 00:39:17 -0400450 if (atomic_long_dec_and_test(&file->f_count)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451 security_file_free(file);
Nick Pigginee2ffa02010-08-18 04:37:35 +1000452 file_sb_list_del(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453 file_free(file);
454 }
455}
456
Nick Piggin6416ccb2010-08-18 04:37:38 +1000457static inline int file_list_cpu(struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458{
Nick Piggin6416ccb2010-08-18 04:37:38 +1000459#ifdef CONFIG_SMP
460 return file->f_sb_list_cpu;
461#else
462 return smp_processor_id();
463#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464}
465
Nick Piggin6416ccb2010-08-18 04:37:38 +1000466/* helper for file_sb_list_add to reduce ifdefs */
467static inline void __file_sb_list_add(struct file *file, struct super_block *sb)
468{
469 struct list_head *list;
470#ifdef CONFIG_SMP
471 int cpu;
472 cpu = smp_processor_id();
473 file->f_sb_list_cpu = cpu;
474 list = per_cpu_ptr(sb->s_files, cpu);
475#else
476 list = &sb->s_files;
477#endif
478 list_add(&file->f_u.fu_list, list);
479}
480
481/**
482 * file_sb_list_add - add a file to the sb's file list
483 * @file: file to add
484 * @sb: sb to add it to
485 *
486 * Use this function to associate a file with the superblock of the inode it
487 * refers to.
488 */
489void file_sb_list_add(struct file *file, struct super_block *sb)
490{
Andi Kleen962830d2012-05-08 13:32:02 +0930491 lg_local_lock(&files_lglock);
Nick Piggin6416ccb2010-08-18 04:37:38 +1000492 __file_sb_list_add(file, sb);
Andi Kleen962830d2012-05-08 13:32:02 +0930493 lg_local_unlock(&files_lglock);
Nick Piggin6416ccb2010-08-18 04:37:38 +1000494}
495
496/**
497 * file_sb_list_del - remove a file from the sb's file list
498 * @file: file to remove
499 * @sb: sb to remove it from
500 *
501 * Use this function to remove a file from its superblock.
502 */
Nick Pigginee2ffa02010-08-18 04:37:35 +1000503void file_sb_list_del(struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504{
Eric Dumazet2f512012005-10-30 15:02:16 -0800505 if (!list_empty(&file->f_u.fu_list)) {
Andi Kleen962830d2012-05-08 13:32:02 +0930506 lg_local_lock_cpu(&files_lglock, file_list_cpu(file));
Eric Dumazet2f512012005-10-30 15:02:16 -0800507 list_del_init(&file->f_u.fu_list);
Andi Kleen962830d2012-05-08 13:32:02 +0930508 lg_local_unlock_cpu(&files_lglock, file_list_cpu(file));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509 }
510}
511
Nick Piggin6416ccb2010-08-18 04:37:38 +1000512#ifdef CONFIG_SMP
513
514/*
515 * These macros iterate all files on all CPUs for a given superblock.
516 * files_lglock must be held globally.
517 */
518#define do_file_list_for_each_entry(__sb, __file) \
519{ \
520 int i; \
521 for_each_possible_cpu(i) { \
522 struct list_head *list; \
523 list = per_cpu_ptr((__sb)->s_files, i); \
524 list_for_each_entry((__file), list, f_u.fu_list)
525
526#define while_file_list_for_each_entry \
527 } \
528}
529
530#else
531
532#define do_file_list_for_each_entry(__sb, __file) \
533{ \
534 struct list_head *list; \
535 list = &(sb)->s_files; \
536 list_for_each_entry((__file), list, f_u.fu_list)
537
538#define while_file_list_for_each_entry \
539}
540
541#endif
542
npiggin@suse.de864d7c42009-04-26 20:25:56 +1000543/**
544 * mark_files_ro - mark all files read-only
545 * @sb: superblock in question
546 *
547 * All files are marked read-only. We don't care about pending
548 * delete files so this should be used in 'force' mode only.
549 */
550void mark_files_ro(struct super_block *sb)
551{
552 struct file *f;
553
Andi Kleen962830d2012-05-08 13:32:02 +0930554 lg_global_lock(&files_lglock);
Nick Piggin6416ccb2010-08-18 04:37:38 +1000555 do_file_list_for_each_entry(sb, f) {
npiggin@suse.de864d7c42009-04-26 20:25:56 +1000556 if (!S_ISREG(f->f_path.dentry->d_inode->i_mode))
557 continue;
558 if (!file_count(f))
559 continue;
560 if (!(f->f_mode & FMODE_WRITE))
561 continue;
Wu Fengguang42e49602010-03-05 13:42:01 -0800562 spin_lock(&f->f_lock);
npiggin@suse.de864d7c42009-04-26 20:25:56 +1000563 f->f_mode &= ~FMODE_WRITE;
Wu Fengguang42e49602010-03-05 13:42:01 -0800564 spin_unlock(&f->f_lock);
npiggin@suse.de864d7c42009-04-26 20:25:56 +1000565 if (file_check_writeable(f) != 0)
566 continue;
567 file_release_write(f);
Al Viro85d7d612012-06-23 22:41:54 +0400568 mnt_drop_write_file(f);
Nick Piggin6416ccb2010-08-18 04:37:38 +1000569 } while_file_list_for_each_entry;
Andi Kleen962830d2012-05-08 13:32:02 +0930570 lg_global_unlock(&files_lglock);
npiggin@suse.de864d7c42009-04-26 20:25:56 +1000571}
572
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573void __init files_init(unsigned long mempages)
574{
Eric Dumazet518de9b2010-10-26 14:22:44 -0700575 unsigned long n;
Eric Dumazetb6b3fde2008-12-10 09:35:45 -0800576
577 filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
578 SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
579
580 /*
581 * One file with associated inode and dcache is very roughly 1K.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 * Per default don't use more than 10% of our memory for files.
583 */
584
585 n = (mempages * (PAGE_SIZE / 1024)) / 10;
Eric Dumazet518de9b2010-10-26 14:22:44 -0700586 files_stat.max_files = max_t(unsigned long, n, NR_FILE);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700587 files_defer_init();
Andi Kleen962830d2012-05-08 13:32:02 +0930588 lg_lock_init(&files_lglock, "files_lglock");
Mingming Cao0216bfc2006-06-23 02:05:41 -0700589 percpu_counter_init(&nr_files, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590}