1/*
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>
11#include <linux/fdtable.h>
12#include <linux/init.h>
13#include <linux/module.h>
14#include <linux/fs.h>
15#include <linux/security.h>
16#include <linux/eventpoll.h>
17#include <linux/rcupdate.h>
18#include <linux/mount.h>
19#include <linux/capability.h>
20#include <linux/cdev.h>
21#include <linux/fsnotify.h>
22#include <linux/sysctl.h>
23#include <linux/lglock.h>
24#include <linux/percpu_counter.h>
25#include <linux/percpu.h>
26#include <linux/ima.h>
27
28#include <asm/atomic.h>
29
30#include "internal.h"
31
32/* sysctl tunables... */
33struct files_stat_struct files_stat = {
34	.max_files = NR_FILE
35};
36
37DECLARE_LGLOCK(files_lglock);
38DEFINE_LGLOCK(files_lglock);
39
40/* SLAB cache for file structures */
41static struct kmem_cache *filp_cachep __read_mostly;
42
43static struct percpu_counter nr_files __cacheline_aligned_in_smp;
44
45static inline void file_free_rcu(struct rcu_head *head)
46{
47	struct file *f = container_of(head, struct file, f_u.fu_rcuhead);
48
49	put_cred(f->f_cred);
50	kmem_cache_free(filp_cachep, f);
51}
52
53static inline void file_free(struct file *f)
54{
55	percpu_counter_dec(&nr_files);
56	file_check_state(f);
57	call_rcu(&f->f_u.fu_rcuhead, file_free_rcu);
58}
59
60/*
61 * Return the total number of open files in the system
62 */
63static int get_nr_files(void)
64{
65	return percpu_counter_read_positive(&nr_files);
66}
67
68/*
69 * Return the maximum number of open files in the system
70 */
71int get_max_files(void)
72{
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)
81int proc_nr_files(ctl_table *table, int write,
82                     void __user *buffer, size_t *lenp, loff_t *ppos)
83{
84	files_stat.nr_files = get_nr_files();
85	return proc_dointvec(table, write, buffer, lenp, ppos);
86}
87#else
88int proc_nr_files(ctl_table *table, int write,
89                     void __user *buffer, size_t *lenp, loff_t *ppos)
90{
91	return -ENOSYS;
92}
93#endif
94
95/* 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.
98 *
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.
104 */
105struct file *get_empty_filp(void)
106{
107	const struct cred *cred = current_cred();
108	static int old_max;
109	struct file * f;
110
111	/*
112	 * Privileged users can go above max_files
113	 */
114	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		 */
119		if (percpu_counter_sum_positive(&nr_files) >= files_stat.max_files)
120			goto over;
121	}
122
123	f = kmem_cache_zalloc(filp_cachep, GFP_KERNEL);
124	if (f == NULL)
125		goto fail;
126
127	percpu_counter_inc(&nr_files);
128	if (security_file_alloc(f))
129		goto fail_sec;
130
131	INIT_LIST_HEAD(&f->f_u.fu_list);
132	atomic_long_set(&f->f_count, 1);
133	rwlock_init(&f->f_owner.lock);
134	f->f_cred = get_cred(cred);
135	spin_lock_init(&f->f_lock);
136	eventpoll_init_file(f);
137	/* f->f_version: 0 */
138	return f;
139
140over:
141	/* Ran out of filps - report that */
142	if (get_nr_files() > old_max) {
143		printk(KERN_INFO "VFS: file-max limit %d reached\n",
144					get_max_files());
145		old_max = get_nr_files();
146	}
147	goto fail;
148
149fail_sec:
150	file_free(f);
151fail:
152	return NULL;
153}
154
155/**
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 */
170struct file *alloc_file(struct path *path, fmode_t mode,
171		const struct file_operations *fop)
172{
173	struct file *file;
174
175	file = get_empty_filp();
176	if (!file)
177		return NULL;
178
179	file->f_path = *path;
180	file->f_mapping = path->dentry->d_inode->i_mapping;
181	file->f_mode = mode;
182	file->f_op = fop;
183
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	 */
190	if ((mode & FMODE_WRITE) && !special_file(path->dentry->d_inode->i_mode)) {
191		file_take_write(file);
192		WARN_ON(mnt_clone_write(path->mnt));
193	}
194	ima_counts_get(file);
195	return file;
196}
197EXPORT_SYMBOL(alloc_file);
198
199/**
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 */
207void drop_file_write_access(struct file *file)
208{
209	struct vfsmount *mnt = file->f_path.mnt;
210	struct dentry *dentry = file->f_path.dentry;
211	struct inode *inode = dentry->d_inode;
212
213	put_write_access(inode);
214
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);
221}
222EXPORT_SYMBOL_GPL(drop_file_write_access);
223
224/* the real guts of fput() - releasing the last reference to file
225 */
226static void __fput(struct file *file)
227{
228	struct dentry *dentry = file->f_path.dentry;
229	struct vfsmount *mnt = file->f_path.mnt;
230	struct inode *inode = dentry->d_inode;
231
232	might_sleep();
233
234	fsnotify_close(file);
235	/*
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
242	if (unlikely(file->f_flags & FASYNC)) {
243		if (file->f_op && file->f_op->fasync)
244			file->f_op->fasync(-1, file, 0);
245	}
246	if (file->f_op && file->f_op->release)
247		file->f_op->release(inode, file);
248	security_file_free(file);
249	ima_file_free(file);
250	if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL))
251		cdev_put(inode->i_cdev);
252	fops_put(file->f_op);
253	put_pid(file->f_owner.pid);
254	file_sb_list_del(file);
255	if (file->f_mode & FMODE_WRITE)
256		drop_file_write_access(file);
257	file->f_path.dentry = NULL;
258	file->f_path.mnt = NULL;
259	file_free(file);
260	dput(dentry);
261	mntput(mnt);
262}
263
264void fput(struct file *file)
265{
266	if (atomic_long_dec_and_test(&file->f_count))
267		__fput(file);
268}
269
270EXPORT_SYMBOL(fput);
271
272struct file *fget(unsigned int fd)
273{
274	struct file *file;
275	struct files_struct *files = current->files;
276
277	rcu_read_lock();
278	file = fcheck_files(files, fd);
279	if (file) {
280		if (!atomic_long_inc_not_zero(&file->f_count)) {
281			/* File object ref couldn't be taken */
282			rcu_read_unlock();
283			return NULL;
284		}
285	}
286	rcu_read_unlock();
287
288	return file;
289}
290
291EXPORT_SYMBOL(fget);
292
293/*
294 * Lightweight file lookup - no refcnt increment if fd table isn't shared.
295 *
296 * You can use this instead of fget if you satisfy all of the following
297 * conditions:
298 * 1) You must call fput_light before exiting the syscall and returning control
299 *    to userspace (i.e. you cannot remember the returned struct file * after
300 *    returning to userspace).
301 * 2) You must not call filp_close on the returned struct file * in between
302 *    calls to fget_light and fput_light.
303 * 3) You must not clone the current task in between the calls to fget_light
304 *    and fput_light.
305 *
306 * The fput_needed flag returned by fget_light should be passed to the
307 * corresponding fput_light.
308 */
309struct file *fget_light(unsigned int fd, int *fput_needed)
310{
311	struct file *file;
312	struct files_struct *files = current->files;
313
314	*fput_needed = 0;
315	if (likely((atomic_read(&files->count) == 1))) {
316		file = fcheck_files(files, fd);
317	} else {
318		rcu_read_lock();
319		file = fcheck_files(files, fd);
320		if (file) {
321			if (atomic_long_inc_not_zero(&file->f_count))
322				*fput_needed = 1;
323			else
324				/* Didn't get the reference, someone's freed */
325				file = NULL;
326		}
327		rcu_read_unlock();
328	}
329
330	return file;
331}
332
333void put_filp(struct file *file)
334{
335	if (atomic_long_dec_and_test(&file->f_count)) {
336		security_file_free(file);
337		file_sb_list_del(file);
338		file_free(file);
339	}
340}
341
342static inline int file_list_cpu(struct file *file)
343{
344#ifdef CONFIG_SMP
345	return file->f_sb_list_cpu;
346#else
347	return smp_processor_id();
348#endif
349}
350
351/* helper for file_sb_list_add to reduce ifdefs */
352static inline void __file_sb_list_add(struct file *file, struct super_block *sb)
353{
354	struct list_head *list;
355#ifdef CONFIG_SMP
356	int cpu;
357	cpu = smp_processor_id();
358	file->f_sb_list_cpu = cpu;
359	list = per_cpu_ptr(sb->s_files, cpu);
360#else
361	list = &sb->s_files;
362#endif
363	list_add(&file->f_u.fu_list, list);
364}
365
366/**
367 * file_sb_list_add - add a file to the sb's file list
368 * @file: file to add
369 * @sb: sb to add it to
370 *
371 * Use this function to associate a file with the superblock of the inode it
372 * refers to.
373 */
374void file_sb_list_add(struct file *file, struct super_block *sb)
375{
376	lg_local_lock(files_lglock);
377	__file_sb_list_add(file, sb);
378	lg_local_unlock(files_lglock);
379}
380
381/**
382 * file_sb_list_del - remove a file from the sb's file list
383 * @file: file to remove
384 * @sb: sb to remove it from
385 *
386 * Use this function to remove a file from its superblock.
387 */
388void file_sb_list_del(struct file *file)
389{
390	if (!list_empty(&file->f_u.fu_list)) {
391		lg_local_lock_cpu(files_lglock, file_list_cpu(file));
392		list_del_init(&file->f_u.fu_list);
393		lg_local_unlock_cpu(files_lglock, file_list_cpu(file));
394	}
395}
396
397#ifdef CONFIG_SMP
398
399/*
400 * These macros iterate all files on all CPUs for a given superblock.
401 * files_lglock must be held globally.
402 */
403#define do_file_list_for_each_entry(__sb, __file)		\
404{								\
405	int i;							\
406	for_each_possible_cpu(i) {				\
407		struct list_head *list;				\
408		list = per_cpu_ptr((__sb)->s_files, i);		\
409		list_for_each_entry((__file), list, f_u.fu_list)
410
411#define while_file_list_for_each_entry				\
412	}							\
413}
414
415#else
416
417#define do_file_list_for_each_entry(__sb, __file)		\
418{								\
419	struct list_head *list;					\
420	list = &(sb)->s_files;					\
421	list_for_each_entry((__file), list, f_u.fu_list)
422
423#define while_file_list_for_each_entry				\
424}
425
426#endif
427
428int fs_may_remount_ro(struct super_block *sb)
429{
430	struct file *file;
431	/* Check that no files are currently opened for writing. */
432	lg_global_lock(files_lglock);
433	do_file_list_for_each_entry(sb, file) {
434		struct inode *inode = file->f_path.dentry->d_inode;
435
436		/* File with pending delete? */
437		if (inode->i_nlink == 0)
438			goto too_bad;
439
440		/* Writeable file? */
441		if (S_ISREG(inode->i_mode) && (file->f_mode & FMODE_WRITE))
442			goto too_bad;
443	} while_file_list_for_each_entry;
444	lg_global_unlock(files_lglock);
445	return 1; /* Tis' cool bro. */
446too_bad:
447	lg_global_unlock(files_lglock);
448	return 0;
449}
450
451/**
452 *	mark_files_ro - mark all files read-only
453 *	@sb: superblock in question
454 *
455 *	All files are marked read-only.  We don't care about pending
456 *	delete files so this should be used in 'force' mode only.
457 */
458void mark_files_ro(struct super_block *sb)
459{
460	struct file *f;
461
462retry:
463	lg_global_lock(files_lglock);
464	do_file_list_for_each_entry(sb, f) {
465		struct vfsmount *mnt;
466		if (!S_ISREG(f->f_path.dentry->d_inode->i_mode))
467		       continue;
468		if (!file_count(f))
469			continue;
470		if (!(f->f_mode & FMODE_WRITE))
471			continue;
472		spin_lock(&f->f_lock);
473		f->f_mode &= ~FMODE_WRITE;
474		spin_unlock(&f->f_lock);
475		if (file_check_writeable(f) != 0)
476			continue;
477		file_release_write(f);
478		mnt = mntget(f->f_path.mnt);
479		/* This can sleep, so we can't hold the spinlock. */
480		lg_global_unlock(files_lglock);
481		mnt_drop_write(mnt);
482		mntput(mnt);
483		goto retry;
484	} while_file_list_for_each_entry;
485	lg_global_unlock(files_lglock);
486}
487
488void __init files_init(unsigned long mempages)
489{
490	int n;
491
492	filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
493			SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
494
495	/*
496	 * One file with associated inode and dcache is very roughly 1K.
497	 * Per default don't use more than 10% of our memory for files.
498	 */
499
500	n = (mempages * (PAGE_SIZE / 1024)) / 10;
501	files_stat.max_files = n;
502	if (files_stat.max_files < NR_FILE)
503		files_stat.max_files = NR_FILE;
504	files_defer_init();
505	lg_lock_init(files_lglock);
506	percpu_counter_init(&nr_files, 0);
507}
508