xref: /linux/fs/file_table.c (revision 04c71976500352d02f60616d2b960267d8c5fe24)
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/init.h>
12 #include <linux/module.h>
13 #include <linux/fs.h>
14 #include <linux/security.h>
15 #include <linux/eventpoll.h>
16 #include <linux/rcupdate.h>
17 #include <linux/mount.h>
18 #include <linux/capability.h>
19 #include <linux/cdev.h>
20 #include <linux/fsnotify.h>
21 #include <linux/sysctl.h>
22 #include <linux/percpu_counter.h>
23 
24 #include <asm/atomic.h>
25 
26 /* sysctl tunables... */
27 struct files_stat_struct files_stat = {
28 	.max_files = NR_FILE
29 };
30 
31 /* public. Not pretty! */
32 __cacheline_aligned_in_smp DEFINE_SPINLOCK(files_lock);
33 
34 static struct percpu_counter nr_files __cacheline_aligned_in_smp;
35 
36 static inline void file_free_rcu(struct rcu_head *head)
37 {
38 	struct file *f =  container_of(head, struct file, f_u.fu_rcuhead);
39 	kmem_cache_free(filp_cachep, f);
40 }
41 
42 static inline void file_free(struct file *f)
43 {
44 	percpu_counter_dec(&nr_files);
45 	call_rcu(&f->f_u.fu_rcuhead, file_free_rcu);
46 }
47 
48 /*
49  * Return the total number of open files in the system
50  */
51 static int get_nr_files(void)
52 {
53 	return percpu_counter_read_positive(&nr_files);
54 }
55 
56 /*
57  * Return the maximum number of open files in the system
58  */
59 int get_max_files(void)
60 {
61 	return files_stat.max_files;
62 }
63 EXPORT_SYMBOL_GPL(get_max_files);
64 
65 /*
66  * Handle nr_files sysctl
67  */
68 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
69 int proc_nr_files(ctl_table *table, int write, struct file *filp,
70                      void __user *buffer, size_t *lenp, loff_t *ppos)
71 {
72 	files_stat.nr_files = get_nr_files();
73 	return proc_dointvec(table, write, filp, buffer, lenp, ppos);
74 }
75 #else
76 int proc_nr_files(ctl_table *table, int write, struct file *filp,
77                      void __user *buffer, size_t *lenp, loff_t *ppos)
78 {
79 	return -ENOSYS;
80 }
81 #endif
82 
83 /* Find an unused file structure and return a pointer to it.
84  * Returns NULL, if there are no more free file structures or
85  * we run out of memory.
86  */
87 struct file *get_empty_filp(void)
88 {
89 	struct task_struct *tsk;
90 	static int old_max;
91 	struct file * f;
92 
93 	/*
94 	 * Privileged users can go above max_files
95 	 */
96 	if (get_nr_files() >= files_stat.max_files && !capable(CAP_SYS_ADMIN)) {
97 		/*
98 		 * percpu_counters are inaccurate.  Do an expensive check before
99 		 * we go and fail.
100 		 */
101 		if (percpu_counter_sum_positive(&nr_files) >= files_stat.max_files)
102 			goto over;
103 	}
104 
105 	f = kmem_cache_zalloc(filp_cachep, GFP_KERNEL);
106 	if (f == NULL)
107 		goto fail;
108 
109 	percpu_counter_inc(&nr_files);
110 	if (security_file_alloc(f))
111 		goto fail_sec;
112 
113 	tsk = current;
114 	INIT_LIST_HEAD(&f->f_u.fu_list);
115 	atomic_set(&f->f_count, 1);
116 	rwlock_init(&f->f_owner.lock);
117 	f->f_uid = tsk->fsuid;
118 	f->f_gid = tsk->fsgid;
119 	eventpoll_init_file(f);
120 	/* f->f_version: 0 */
121 	return f;
122 
123 over:
124 	/* Ran out of filps - report that */
125 	if (get_nr_files() > old_max) {
126 		printk(KERN_INFO "VFS: file-max limit %d reached\n",
127 					get_max_files());
128 		old_max = get_nr_files();
129 	}
130 	goto fail;
131 
132 fail_sec:
133 	file_free(f);
134 fail:
135 	return NULL;
136 }
137 
138 EXPORT_SYMBOL(get_empty_filp);
139 
140 void fastcall fput(struct file *file)
141 {
142 	if (atomic_dec_and_test(&file->f_count))
143 		__fput(file);
144 }
145 
146 EXPORT_SYMBOL(fput);
147 
148 /* __fput is called from task context when aio completion releases the last
149  * last use of a struct file *.  Do not use otherwise.
150  */
151 void fastcall __fput(struct file *file)
152 {
153 	struct dentry *dentry = file->f_path.dentry;
154 	struct vfsmount *mnt = file->f_path.mnt;
155 	struct inode *inode = dentry->d_inode;
156 
157 	might_sleep();
158 
159 	fsnotify_close(file);
160 	/*
161 	 * The function eventpoll_release() should be the first called
162 	 * in the file cleanup chain.
163 	 */
164 	eventpoll_release(file);
165 	locks_remove_flock(file);
166 
167 	if (file->f_op && file->f_op->release)
168 		file->f_op->release(inode, file);
169 	security_file_free(file);
170 	if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL))
171 		cdev_put(inode->i_cdev);
172 	fops_put(file->f_op);
173 	if (file->f_mode & FMODE_WRITE)
174 		put_write_access(inode);
175 	put_pid(file->f_owner.pid);
176 	file_kill(file);
177 	file->f_path.dentry = NULL;
178 	file->f_path.mnt = NULL;
179 	file_free(file);
180 	dput(dentry);
181 	mntput(mnt);
182 }
183 
184 struct file fastcall *fget(unsigned int fd)
185 {
186 	struct file *file;
187 	struct files_struct *files = current->files;
188 
189 	rcu_read_lock();
190 	file = fcheck_files(files, fd);
191 	if (file) {
192 		if (!atomic_inc_not_zero(&file->f_count)) {
193 			/* File object ref couldn't be taken */
194 			rcu_read_unlock();
195 			return NULL;
196 		}
197 	}
198 	rcu_read_unlock();
199 
200 	return file;
201 }
202 
203 EXPORT_SYMBOL(fget);
204 
205 /*
206  * Lightweight file lookup - no refcnt increment if fd table isn't shared.
207  * You can use this only if it is guranteed that the current task already
208  * holds a refcnt to that file. That check has to be done at fget() only
209  * and a flag is returned to be passed to the corresponding fput_light().
210  * There must not be a cloning between an fget_light/fput_light pair.
211  */
212 struct file fastcall *fget_light(unsigned int fd, int *fput_needed)
213 {
214 	struct file *file;
215 	struct files_struct *files = current->files;
216 
217 	*fput_needed = 0;
218 	if (likely((atomic_read(&files->count) == 1))) {
219 		file = fcheck_files(files, fd);
220 	} else {
221 		rcu_read_lock();
222 		file = fcheck_files(files, fd);
223 		if (file) {
224 			if (atomic_inc_not_zero(&file->f_count))
225 				*fput_needed = 1;
226 			else
227 				/* Didn't get the reference, someone's freed */
228 				file = NULL;
229 		}
230 		rcu_read_unlock();
231 	}
232 
233 	return file;
234 }
235 
236 
237 void put_filp(struct file *file)
238 {
239 	if (atomic_dec_and_test(&file->f_count)) {
240 		security_file_free(file);
241 		file_kill(file);
242 		file_free(file);
243 	}
244 }
245 
246 void file_move(struct file *file, struct list_head *list)
247 {
248 	if (!list)
249 		return;
250 	file_list_lock();
251 	list_move(&file->f_u.fu_list, list);
252 	file_list_unlock();
253 }
254 
255 void file_kill(struct file *file)
256 {
257 	if (!list_empty(&file->f_u.fu_list)) {
258 		file_list_lock();
259 		list_del_init(&file->f_u.fu_list);
260 		file_list_unlock();
261 	}
262 }
263 
264 int fs_may_remount_ro(struct super_block *sb)
265 {
266 	struct list_head *p;
267 
268 	/* Check that no files are currently opened for writing. */
269 	file_list_lock();
270 	list_for_each(p, &sb->s_files) {
271 		struct file *file = list_entry(p, struct file, f_u.fu_list);
272 		struct inode *inode = file->f_path.dentry->d_inode;
273 
274 		/* File with pending delete? */
275 		if (inode->i_nlink == 0)
276 			goto too_bad;
277 
278 		/* Writeable file? */
279 		if (S_ISREG(inode->i_mode) && (file->f_mode & FMODE_WRITE))
280 			goto too_bad;
281 	}
282 	file_list_unlock();
283 	return 1; /* Tis' cool bro. */
284 too_bad:
285 	file_list_unlock();
286 	return 0;
287 }
288 
289 void __init files_init(unsigned long mempages)
290 {
291 	int n;
292 	/* One file with associated inode and dcache is very roughly 1K.
293 	 * Per default don't use more than 10% of our memory for files.
294 	 */
295 
296 	n = (mempages * (PAGE_SIZE / 1024)) / 10;
297 	files_stat.max_files = n;
298 	if (files_stat.max_files < NR_FILE)
299 		files_stat.max_files = NR_FILE;
300 	files_defer_init();
301 	percpu_counter_init(&nr_files, 0);
302 }
303