blob: 32d12b78bac8e2c05a80dc53386b355bdc10ef73 [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>
23#include <linux/percpu_counter.h>
Al Viro0552f872009-12-16 04:53:03 -050024#include <linux/ima.h>
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080025
26#include <asm/atomic.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070027
Eric Parise81e3f42009-12-04 15:47:36 -050028#include "internal.h"
29
Linus Torvalds1da177e2005-04-16 15:20:36 -070030/* sysctl tunables... */
31struct files_stat_struct files_stat = {
32 .max_files = NR_FILE
33};
34
Linus Torvalds1da177e2005-04-16 15:20:36 -070035/* public. Not pretty! */
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080036__cacheline_aligned_in_smp DEFINE_SPINLOCK(files_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070037
Eric Dumazetb6b3fde2008-12-10 09:35:45 -080038/* SLAB cache for file structures */
39static struct kmem_cache *filp_cachep __read_mostly;
40
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080041static struct percpu_counter nr_files __cacheline_aligned_in_smp;
Linus Torvalds1da177e2005-04-16 15:20:36 -070042
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070043static inline void file_free_rcu(struct rcu_head *head)
44{
David Howellsd76b0d92008-11-14 10:39:25 +110045 struct file *f = container_of(head, struct file, f_u.fu_rcuhead);
46
47 put_cred(f->f_cred);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070048 kmem_cache_free(filp_cachep, f);
49}
50
Linus Torvalds1da177e2005-04-16 15:20:36 -070051static inline void file_free(struct file *f)
52{
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080053 percpu_counter_dec(&nr_files);
Dave Hansenad775f52008-02-15 14:38:01 -080054 file_check_state(f);
Eric Dumazet2f512012005-10-30 15:02:16 -080055 call_rcu(&f->f_u.fu_rcuhead, file_free_rcu);
Linus Torvalds1da177e2005-04-16 15:20:36 -070056}
57
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080058/*
59 * Return the total number of open files in the system
60 */
61static int get_nr_files(void)
62{
63 return percpu_counter_read_positive(&nr_files);
64}
65
66/*
67 * Return the maximum number of open files in the system
68 */
69int get_max_files(void)
70{
71 return files_stat.max_files;
72}
73EXPORT_SYMBOL_GPL(get_max_files);
74
75/*
76 * Handle nr_files sysctl
77 */
78#if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
Alexey Dobriyan8d65af72009-09-23 15:57:19 -070079int proc_nr_files(ctl_table *table, int write,
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080080 void __user *buffer, size_t *lenp, loff_t *ppos)
81{
82 files_stat.nr_files = get_nr_files();
Alexey Dobriyan8d65af72009-09-23 15:57:19 -070083 return proc_dointvec(table, write, buffer, lenp, ppos);
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080084}
85#else
Alexey Dobriyan8d65af72009-09-23 15:57:19 -070086int proc_nr_files(ctl_table *table, int write,
Dipankar Sarma529bf6b2006-03-07 21:55:35 -080087 void __user *buffer, size_t *lenp, loff_t *ppos)
88{
89 return -ENOSYS;
90}
91#endif
92
Linus Torvalds1da177e2005-04-16 15:20:36 -070093/* Find an unused file structure and return a pointer to it.
94 * Returns NULL, if there are no more free file structures or
95 * we run out of memory.
Dave Hansen430e2852008-02-15 14:37:26 -080096 *
97 * Be very careful using this. You are responsible for
98 * getting write access to any mount that you might assign
99 * to this filp, if it is opened for write. If this is not
100 * done, you will imbalance int the mount's writer count
101 * and a warning at __fput() time.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102 */
103struct file *get_empty_filp(void)
104{
David Howells86a264a2008-11-14 10:39:18 +1100105 const struct cred *cred = current_cred();
Kirill Korotaevaf4d2ec2005-06-23 00:09:50 -0700106 static int old_max;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107 struct file * f;
108
109 /*
110 * Privileged users can go above max_files
111 */
Dipankar Sarma529bf6b2006-03-07 21:55:35 -0800112 if (get_nr_files() >= files_stat.max_files && !capable(CAP_SYS_ADMIN)) {
113 /*
114 * percpu_counters are inaccurate. Do an expensive check before
115 * we go and fail.
116 */
Peter Zijlstra52d9f3b2007-10-16 23:25:44 -0700117 if (percpu_counter_sum_positive(&nr_files) >= files_stat.max_files)
Dipankar Sarma529bf6b2006-03-07 21:55:35 -0800118 goto over;
119 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120
Denis Cheng4975e452007-10-16 23:26:19 -0700121 f = kmem_cache_zalloc(filp_cachep, GFP_KERNEL);
Kirill Korotaevaf4d2ec2005-06-23 00:09:50 -0700122 if (f == NULL)
123 goto fail;
124
Dipankar Sarma529bf6b2006-03-07 21:55:35 -0800125 percpu_counter_inc(&nr_files);
Kirill Korotaevaf4d2ec2005-06-23 00:09:50 -0700126 if (security_file_alloc(f))
127 goto fail_sec;
128
Eric Dumazet2f512012005-10-30 15:02:16 -0800129 INIT_LIST_HEAD(&f->f_u.fu_list);
Al Viro516e0cc2008-07-26 00:39:17 -0400130 atomic_long_set(&f->f_count, 1);
Benjamin LaHaise5a6b7952006-03-23 03:01:03 -0800131 rwlock_init(&f->f_owner.lock);
David Howellsd76b0d92008-11-14 10:39:25 +1100132 f->f_cred = get_cred(cred);
Jonathan Corbet68499912009-02-06 13:52:43 -0700133 spin_lock_init(&f->f_lock);
Benjamin LaHaise5a6b7952006-03-23 03:01:03 -0800134 eventpoll_init_file(f);
135 /* f->f_version: 0 */
Kirill Korotaevaf4d2ec2005-06-23 00:09:50 -0700136 return f;
137
138over:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139 /* Ran out of filps - report that */
Dipankar Sarma529bf6b2006-03-07 21:55:35 -0800140 if (get_nr_files() > old_max) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141 printk(KERN_INFO "VFS: file-max limit %d reached\n",
Dipankar Sarma529bf6b2006-03-07 21:55:35 -0800142 get_max_files());
143 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 }
Al Viro0552f872009-12-16 04:53:03 -0500192 ima_counts_get(file);
Al Viro3d1e4632009-08-08 23:56:29 +0400193 return file;
Dave Hansence8d2cd2007-10-16 23:31:13 -0700194}
Roland Dreier73efc462009-12-16 12:43:11 -0800195EXPORT_SYMBOL(alloc_file);
Dave Hansence8d2cd2007-10-16 23:31:13 -0700196
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -0800197void fput(struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198{
Al Viro516e0cc2008-07-26 00:39:17 -0400199 if (atomic_long_dec_and_test(&file->f_count))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200 __fput(file);
201}
202
203EXPORT_SYMBOL(fput);
204
Dave Hansenaceaf782008-02-15 14:37:31 -0800205/**
206 * drop_file_write_access - give up ability to write to a file
207 * @file: the file to which we will stop writing
208 *
209 * This is a central place which will give up the ability
210 * to write to @file, along with access to write through
211 * its vfsmount.
212 */
213void drop_file_write_access(struct file *file)
214{
Dave Hansen4a3fd212008-02-15 14:37:48 -0800215 struct vfsmount *mnt = file->f_path.mnt;
Dave Hansenaceaf782008-02-15 14:37:31 -0800216 struct dentry *dentry = file->f_path.dentry;
217 struct inode *inode = dentry->d_inode;
218
219 put_write_access(inode);
Dave Hansenad775f52008-02-15 14:38:01 -0800220
221 if (special_file(inode->i_mode))
222 return;
223 if (file_check_writeable(file) != 0)
224 return;
225 mnt_drop_write(mnt);
226 file_release_write(file);
Dave Hansenaceaf782008-02-15 14:37:31 -0800227}
228EXPORT_SYMBOL_GPL(drop_file_write_access);
229
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230/* __fput is called from task context when aio completion releases the last
231 * last use of a struct file *. Do not use otherwise.
232 */
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -0800233void __fput(struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800235 struct dentry *dentry = file->f_path.dentry;
236 struct vfsmount *mnt = file->f_path.mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237 struct inode *inode = dentry->d_inode;
238
239 might_sleep();
Robert Love0eeca282005-07-12 17:06:03 -0400240
241 fsnotify_close(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242 /*
243 * The function eventpoll_release() should be the first called
244 * in the file cleanup chain.
245 */
246 eventpoll_release(file);
247 locks_remove_flock(file);
248
Al Viro233e70f2008-10-31 23:28:30 +0000249 if (unlikely(file->f_flags & FASYNC)) {
250 if (file->f_op && file->f_op->fasync)
251 file->f_op->fasync(-1, file, 0);
252 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253 if (file->f_op && file->f_op->release)
254 file->f_op->release(inode, file);
255 security_file_free(file);
Al Viro89068c52010-02-07 03:07:29 -0500256 ima_file_free(file);
Theodore Ts'o577c4eb2006-09-27 01:50:49 -0700257 if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258 cdev_put(inode->i_cdev);
259 fops_put(file->f_op);
Eric W. Biederman609d7fa2006-10-02 02:17:15 -0700260 put_pid(file->f_owner.pid);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 file_kill(file);
Dave Hansenaceaf782008-02-15 14:37:31 -0800262 if (file->f_mode & FMODE_WRITE)
263 drop_file_write_access(file);
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800264 file->f_path.dentry = NULL;
265 file->f_path.mnt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266 file_free(file);
267 dput(dentry);
268 mntput(mnt);
269}
270
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -0800271struct file *fget(unsigned int fd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272{
273 struct file *file;
274 struct files_struct *files = current->files;
275
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700276 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277 file = fcheck_files(files, fd);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700278 if (file) {
Al Viro516e0cc2008-07-26 00:39:17 -0400279 if (!atomic_long_inc_not_zero(&file->f_count)) {
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700280 /* File object ref couldn't be taken */
281 rcu_read_unlock();
282 return NULL;
283 }
284 }
285 rcu_read_unlock();
286
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287 return file;
288}
289
290EXPORT_SYMBOL(fget);
291
292/*
293 * Lightweight file lookup - no refcnt increment if fd table isn't shared.
294 * You can use this only if it is guranteed that the current task already
295 * holds a refcnt to that file. That check has to be done at fget() only
296 * and a flag is returned to be passed to the corresponding fput_light().
297 * There must not be a cloning between an fget_light/fput_light pair.
298 */
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -0800299struct file *fget_light(unsigned int fd, int *fput_needed)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300{
301 struct file *file;
302 struct files_struct *files = current->files;
303
304 *fput_needed = 0;
305 if (likely((atomic_read(&files->count) == 1))) {
306 file = fcheck_files(files, fd);
307 } else {
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700308 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309 file = fcheck_files(files, fd);
310 if (file) {
Al Viro516e0cc2008-07-26 00:39:17 -0400311 if (atomic_long_inc_not_zero(&file->f_count))
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700312 *fput_needed = 1;
313 else
314 /* Didn't get the reference, someone's freed */
315 file = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316 }
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700317 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318 }
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700319
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320 return file;
321}
322
323
324void put_filp(struct file *file)
325{
Al Viro516e0cc2008-07-26 00:39:17 -0400326 if (atomic_long_dec_and_test(&file->f_count)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327 security_file_free(file);
328 file_kill(file);
329 file_free(file);
330 }
331}
332
333void file_move(struct file *file, struct list_head *list)
334{
335 if (!list)
336 return;
337 file_list_lock();
Eric Dumazet2f512012005-10-30 15:02:16 -0800338 list_move(&file->f_u.fu_list, list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339 file_list_unlock();
340}
341
342void file_kill(struct file *file)
343{
Eric Dumazet2f512012005-10-30 15:02:16 -0800344 if (!list_empty(&file->f_u.fu_list)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345 file_list_lock();
Eric Dumazet2f512012005-10-30 15:02:16 -0800346 list_del_init(&file->f_u.fu_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347 file_list_unlock();
348 }
349}
350
351int fs_may_remount_ro(struct super_block *sb)
352{
Matthias Kaehlckecfdaf9e2007-10-18 23:39:56 -0700353 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354
355 /* Check that no files are currently opened for writing. */
356 file_list_lock();
Matthias Kaehlckecfdaf9e2007-10-18 23:39:56 -0700357 list_for_each_entry(file, &sb->s_files, f_u.fu_list) {
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800358 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359
360 /* File with pending delete? */
361 if (inode->i_nlink == 0)
362 goto too_bad;
363
364 /* Writeable file? */
365 if (S_ISREG(inode->i_mode) && (file->f_mode & FMODE_WRITE))
366 goto too_bad;
367 }
368 file_list_unlock();
369 return 1; /* Tis' cool bro. */
370too_bad:
371 file_list_unlock();
372 return 0;
373}
374
npiggin@suse.de864d7c42009-04-26 20:25:56 +1000375/**
376 * mark_files_ro - mark all files read-only
377 * @sb: superblock in question
378 *
379 * All files are marked read-only. We don't care about pending
380 * delete files so this should be used in 'force' mode only.
381 */
382void mark_files_ro(struct super_block *sb)
383{
384 struct file *f;
385
386retry:
387 file_list_lock();
388 list_for_each_entry(f, &sb->s_files, f_u.fu_list) {
389 struct vfsmount *mnt;
390 if (!S_ISREG(f->f_path.dentry->d_inode->i_mode))
391 continue;
392 if (!file_count(f))
393 continue;
394 if (!(f->f_mode & FMODE_WRITE))
395 continue;
Wu Fengguang42e49602010-03-05 13:42:01 -0800396 spin_lock(&f->f_lock);
npiggin@suse.de864d7c42009-04-26 20:25:56 +1000397 f->f_mode &= ~FMODE_WRITE;
Wu Fengguang42e49602010-03-05 13:42:01 -0800398 spin_unlock(&f->f_lock);
npiggin@suse.de864d7c42009-04-26 20:25:56 +1000399 if (file_check_writeable(f) != 0)
400 continue;
401 file_release_write(f);
402 mnt = mntget(f->f_path.mnt);
403 file_list_unlock();
404 /*
405 * This can sleep, so we can't hold
406 * the file_list_lock() spinlock.
407 */
408 mnt_drop_write(mnt);
409 mntput(mnt);
410 goto retry;
411 }
412 file_list_unlock();
413}
414
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415void __init files_init(unsigned long mempages)
416{
417 int n;
Eric Dumazetb6b3fde2008-12-10 09:35:45 -0800418
419 filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
420 SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
421
422 /*
423 * One file with associated inode and dcache is very roughly 1K.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424 * Per default don't use more than 10% of our memory for files.
425 */
426
427 n = (mempages * (PAGE_SIZE / 1024)) / 10;
428 files_stat.max_files = n;
429 if (files_stat.max_files < NR_FILE)
430 files_stat.max_files = NR_FILE;
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700431 files_defer_init();
Mingming Cao0216bfc2006-06-23 02:05:41 -0700432 percpu_counter_init(&nr_files, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433}