xref: /linux/fs/file_table.c (revision f3d9478b2ce468c3115b02ecae7e975990697f15)
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/config.h>
9 #include <linux/string.h>
10 #include <linux/slab.h>
11 #include <linux/file.h>
12 #include <linux/init.h>
13 #include <linux/module.h>
14 #include <linux/smp_lock.h>
15 #include <linux/fs.h>
16 #include <linux/security.h>
17 #include <linux/eventpoll.h>
18 #include <linux/rcupdate.h>
19 #include <linux/mount.h>
20 #include <linux/capability.h>
21 #include <linux/cdev.h>
22 #include <linux/fsnotify.h>
23 #include <linux/sysctl.h>
24 #include <linux/percpu_counter.h>
25 
26 #include <asm/atomic.h>
27 
28 /* sysctl tunables... */
29 struct files_stat_struct files_stat = {
30 	.max_files = NR_FILE
31 };
32 
33 /* public. Not pretty! */
34 __cacheline_aligned_in_smp DEFINE_SPINLOCK(files_lock);
35 
36 static struct percpu_counter nr_files __cacheline_aligned_in_smp;
37 
38 static inline void file_free_rcu(struct rcu_head *head)
39 {
40 	struct file *f =  container_of(head, struct file, f_u.fu_rcuhead);
41 	kmem_cache_free(filp_cachep, f);
42 }
43 
44 static inline void file_free(struct file *f)
45 {
46 	percpu_counter_dec(&nr_files);
47 	call_rcu(&f->f_u.fu_rcuhead, file_free_rcu);
48 }
49 
50 /*
51  * Return the total number of open files in the system
52  */
53 static int get_nr_files(void)
54 {
55 	return percpu_counter_read_positive(&nr_files);
56 }
57 
58 /*
59  * Return the maximum number of open files in the system
60  */
61 int get_max_files(void)
62 {
63 	return files_stat.max_files;
64 }
65 EXPORT_SYMBOL_GPL(get_max_files);
66 
67 /*
68  * Handle nr_files sysctl
69  */
70 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
71 int proc_nr_files(ctl_table *table, int write, struct file *filp,
72                      void __user *buffer, size_t *lenp, loff_t *ppos)
73 {
74 	files_stat.nr_files = get_nr_files();
75 	return proc_dointvec(table, write, filp, buffer, lenp, ppos);
76 }
77 #else
78 int proc_nr_files(ctl_table *table, int write, struct file *filp,
79                      void __user *buffer, size_t *lenp, loff_t *ppos)
80 {
81 	return -ENOSYS;
82 }
83 #endif
84 
85 /* Find an unused file structure and return a pointer to it.
86  * Returns NULL, if there are no more free file structures or
87  * we run out of memory.
88  */
89 struct file *get_empty_filp(void)
90 {
91 	struct task_struct *tsk;
92 	static int old_max;
93 	struct file * f;
94 
95 	/*
96 	 * Privileged users can go above max_files
97 	 */
98 	if (get_nr_files() >= files_stat.max_files && !capable(CAP_SYS_ADMIN)) {
99 		/*
100 		 * percpu_counters are inaccurate.  Do an expensive check before
101 		 * we go and fail.
102 		 */
103 		if (percpu_counter_sum(&nr_files) >= files_stat.max_files)
104 			goto over;
105 	}
106 
107 	f = kmem_cache_alloc(filp_cachep, GFP_KERNEL);
108 	if (f == NULL)
109 		goto fail;
110 
111 	percpu_counter_inc(&nr_files);
112 	memset(f, 0, sizeof(*f));
113 	if (security_file_alloc(f))
114 		goto fail_sec;
115 
116 	tsk = current;
117 	INIT_LIST_HEAD(&f->f_u.fu_list);
118 	atomic_set(&f->f_count, 1);
119 	rwlock_init(&f->f_owner.lock);
120 	f->f_uid = tsk->fsuid;
121 	f->f_gid = tsk->fsgid;
122 	eventpoll_init_file(f);
123 	/* f->f_version: 0 */
124 	return f;
125 
126 over:
127 	/* Ran out of filps - report that */
128 	if (get_nr_files() > old_max) {
129 		printk(KERN_INFO "VFS: file-max limit %d reached\n",
130 					get_max_files());
131 		old_max = get_nr_files();
132 	}
133 	goto fail;
134 
135 fail_sec:
136 	file_free(f);
137 fail:
138 	return NULL;
139 }
140 
141 EXPORT_SYMBOL(get_empty_filp);
142 
143 void fastcall fput(struct file *file)
144 {
145 	if (atomic_dec_and_test(&file->f_count))
146 		__fput(file);
147 }
148 
149 EXPORT_SYMBOL(fput);
150 
151 /* __fput is called from task context when aio completion releases the last
152  * last use of a struct file *.  Do not use otherwise.
153  */
154 void fastcall __fput(struct file *file)
155 {
156 	struct dentry *dentry = file->f_dentry;
157 	struct vfsmount *mnt = file->f_vfsmnt;
158 	struct inode *inode = dentry->d_inode;
159 
160 	might_sleep();
161 
162 	fsnotify_close(file);
163 	/*
164 	 * The function eventpoll_release() should be the first called
165 	 * in the file cleanup chain.
166 	 */
167 	eventpoll_release(file);
168 	locks_remove_flock(file);
169 
170 	if (file->f_op && file->f_op->release)
171 		file->f_op->release(inode, file);
172 	security_file_free(file);
173 	if (unlikely(inode->i_cdev != NULL))
174 		cdev_put(inode->i_cdev);
175 	fops_put(file->f_op);
176 	if (file->f_mode & FMODE_WRITE)
177 		put_write_access(inode);
178 	file_kill(file);
179 	file->f_dentry = NULL;
180 	file->f_vfsmnt = NULL;
181 	file_free(file);
182 	dput(dentry);
183 	mntput(mnt);
184 }
185 
186 struct file fastcall *fget(unsigned int fd)
187 {
188 	struct file *file;
189 	struct files_struct *files = current->files;
190 
191 	rcu_read_lock();
192 	file = fcheck_files(files, fd);
193 	if (file) {
194 		if (!atomic_inc_not_zero(&file->f_count)) {
195 			/* File object ref couldn't be taken */
196 			rcu_read_unlock();
197 			return NULL;
198 		}
199 	}
200 	rcu_read_unlock();
201 
202 	return file;
203 }
204 
205 EXPORT_SYMBOL(fget);
206 
207 /*
208  * Lightweight file lookup - no refcnt increment if fd table isn't shared.
209  * You can use this only if it is guranteed that the current task already
210  * holds a refcnt to that file. That check has to be done at fget() only
211  * and a flag is returned to be passed to the corresponding fput_light().
212  * There must not be a cloning between an fget_light/fput_light pair.
213  */
214 struct file fastcall *fget_light(unsigned int fd, int *fput_needed)
215 {
216 	struct file *file;
217 	struct files_struct *files = current->files;
218 
219 	*fput_needed = 0;
220 	if (likely((atomic_read(&files->count) == 1))) {
221 		file = fcheck_files(files, fd);
222 	} else {
223 		rcu_read_lock();
224 		file = fcheck_files(files, fd);
225 		if (file) {
226 			if (atomic_inc_not_zero(&file->f_count))
227 				*fput_needed = 1;
228 			else
229 				/* Didn't get the reference, someone's freed */
230 				file = NULL;
231 		}
232 		rcu_read_unlock();
233 	}
234 
235 	return file;
236 }
237 
238 
239 void put_filp(struct file *file)
240 {
241 	if (atomic_dec_and_test(&file->f_count)) {
242 		security_file_free(file);
243 		file_kill(file);
244 		file_free(file);
245 	}
246 }
247 
248 void file_move(struct file *file, struct list_head *list)
249 {
250 	if (!list)
251 		return;
252 	file_list_lock();
253 	list_move(&file->f_u.fu_list, list);
254 	file_list_unlock();
255 }
256 
257 void file_kill(struct file *file)
258 {
259 	if (!list_empty(&file->f_u.fu_list)) {
260 		file_list_lock();
261 		list_del_init(&file->f_u.fu_list);
262 		file_list_unlock();
263 	}
264 }
265 
266 int fs_may_remount_ro(struct super_block *sb)
267 {
268 	struct list_head *p;
269 
270 	/* Check that no files are currently opened for writing. */
271 	file_list_lock();
272 	list_for_each(p, &sb->s_files) {
273 		struct file *file = list_entry(p, struct file, f_u.fu_list);
274 		struct inode *inode = file->f_dentry->d_inode;
275 
276 		/* File with pending delete? */
277 		if (inode->i_nlink == 0)
278 			goto too_bad;
279 
280 		/* Writeable file? */
281 		if (S_ISREG(inode->i_mode) && (file->f_mode & FMODE_WRITE))
282 			goto too_bad;
283 	}
284 	file_list_unlock();
285 	return 1; /* Tis' cool bro. */
286 too_bad:
287 	file_list_unlock();
288 	return 0;
289 }
290 
291 void __init files_init(unsigned long mempages)
292 {
293 	int n;
294 	/* One file with associated inode and dcache is very roughly 1K.
295 	 * Per default don't use more than 10% of our memory for files.
296 	 */
297 
298 	n = (mempages * (PAGE_SIZE / 1024)) / 10;
299 	files_stat.max_files = n;
300 	if (files_stat.max_files < NR_FILE)
301 		files_stat.max_files = NR_FILE;
302 	files_defer_init();
303 	percpu_counter_init(&nr_files);
304 }
305