xref: /linux/fs/file_table.c (revision 7025bec9125b0a02edcaf22c2dce753bf2c95480)
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/percpu_counter.h>
24 #include <linux/ima.h>
25 
26 #include <asm/atomic.h>
27 
28 #include "internal.h"
29 
30 /* sysctl tunables... */
31 struct files_stat_struct files_stat = {
32 	.max_files = NR_FILE
33 };
34 
35 /* public. Not pretty! */
36 __cacheline_aligned_in_smp DEFINE_SPINLOCK(files_lock);
37 
38 /* SLAB cache for file structures */
39 static struct kmem_cache *filp_cachep __read_mostly;
40 
41 static struct percpu_counter nr_files __cacheline_aligned_in_smp;
42 
43 static inline void file_free_rcu(struct rcu_head *head)
44 {
45 	struct file *f = container_of(head, struct file, f_u.fu_rcuhead);
46 
47 	put_cred(f->f_cred);
48 	kmem_cache_free(filp_cachep, f);
49 }
50 
51 static inline void file_free(struct file *f)
52 {
53 	percpu_counter_dec(&nr_files);
54 	file_check_state(f);
55 	call_rcu(&f->f_u.fu_rcuhead, file_free_rcu);
56 }
57 
58 /*
59  * Return the total number of open files in the system
60  */
61 static 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  */
69 int get_max_files(void)
70 {
71 	return files_stat.max_files;
72 }
73 EXPORT_SYMBOL_GPL(get_max_files);
74 
75 /*
76  * Handle nr_files sysctl
77  */
78 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
79 int proc_nr_files(ctl_table *table, int write,
80                      void __user *buffer, size_t *lenp, loff_t *ppos)
81 {
82 	files_stat.nr_files = get_nr_files();
83 	return proc_dointvec(table, write, buffer, lenp, ppos);
84 }
85 #else
86 int proc_nr_files(ctl_table *table, int write,
87                      void __user *buffer, size_t *lenp, loff_t *ppos)
88 {
89 	return -ENOSYS;
90 }
91 #endif
92 
93 /* 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.
96  *
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.
102  */
103 struct file *get_empty_filp(void)
104 {
105 	const struct cred *cred = current_cred();
106 	static int old_max;
107 	struct file * f;
108 
109 	/*
110 	 * Privileged users can go above max_files
111 	 */
112 	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 		 */
117 		if (percpu_counter_sum_positive(&nr_files) >= files_stat.max_files)
118 			goto over;
119 	}
120 
121 	f = kmem_cache_zalloc(filp_cachep, GFP_KERNEL);
122 	if (f == NULL)
123 		goto fail;
124 
125 	percpu_counter_inc(&nr_files);
126 	if (security_file_alloc(f))
127 		goto fail_sec;
128 
129 	INIT_LIST_HEAD(&f->f_u.fu_list);
130 	atomic_long_set(&f->f_count, 1);
131 	rwlock_init(&f->f_owner.lock);
132 	f->f_cred = get_cred(cred);
133 	spin_lock_init(&f->f_lock);
134 	eventpoll_init_file(f);
135 	/* f->f_version: 0 */
136 	return f;
137 
138 over:
139 	/* Ran out of filps - report that */
140 	if (get_nr_files() > old_max) {
141 		printk(KERN_INFO "VFS: file-max limit %d reached\n",
142 					get_max_files());
143 		old_max = get_nr_files();
144 	}
145 	goto fail;
146 
147 fail_sec:
148 	file_free(f);
149 fail:
150 	return NULL;
151 }
152 
153 /**
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  */
168 struct file *alloc_file(struct path *path, fmode_t mode,
169 		const struct file_operations *fop)
170 {
171 	struct file *file;
172 
173 	file = get_empty_filp();
174 	if (!file)
175 		return NULL;
176 
177 	file->f_path = *path;
178 	file->f_mapping = path->dentry->d_inode->i_mapping;
179 	file->f_mode = mode;
180 	file->f_op = fop;
181 
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 	 */
188 	if ((mode & FMODE_WRITE) && !special_file(path->dentry->d_inode->i_mode)) {
189 		int error = 0;
190 		file_take_write(file);
191 		error = mnt_clone_write(path->mnt);
192 		WARN_ON(error);
193 	}
194 	ima_counts_get(file);
195 	return file;
196 }
197 EXPORT_SYMBOL(alloc_file);
198 
199 void fput(struct file *file)
200 {
201 	if (atomic_long_dec_and_test(&file->f_count))
202 		__fput(file);
203 }
204 
205 EXPORT_SYMBOL(fput);
206 
207 /**
208  * drop_file_write_access - give up ability to write to a file
209  * @file: the file to which we will stop writing
210  *
211  * This is a central place which will give up the ability
212  * to write to @file, along with access to write through
213  * its vfsmount.
214  */
215 void drop_file_write_access(struct file *file)
216 {
217 	struct vfsmount *mnt = file->f_path.mnt;
218 	struct dentry *dentry = file->f_path.dentry;
219 	struct inode *inode = dentry->d_inode;
220 
221 	put_write_access(inode);
222 
223 	if (special_file(inode->i_mode))
224 		return;
225 	if (file_check_writeable(file) != 0)
226 		return;
227 	mnt_drop_write(mnt);
228 	file_release_write(file);
229 }
230 EXPORT_SYMBOL_GPL(drop_file_write_access);
231 
232 /* __fput is called from task context when aio completion releases the last
233  * last use of a struct file *.  Do not use otherwise.
234  */
235 void __fput(struct file *file)
236 {
237 	struct dentry *dentry = file->f_path.dentry;
238 	struct vfsmount *mnt = file->f_path.mnt;
239 	struct inode *inode = dentry->d_inode;
240 
241 	might_sleep();
242 
243 	fsnotify_close(file);
244 	/*
245 	 * The function eventpoll_release() should be the first called
246 	 * in the file cleanup chain.
247 	 */
248 	eventpoll_release(file);
249 	locks_remove_flock(file);
250 
251 	if (unlikely(file->f_flags & FASYNC)) {
252 		if (file->f_op && file->f_op->fasync)
253 			file->f_op->fasync(-1, file, 0);
254 	}
255 	if (file->f_op && file->f_op->release)
256 		file->f_op->release(inode, file);
257 	security_file_free(file);
258 	if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL))
259 		cdev_put(inode->i_cdev);
260 	fops_put(file->f_op);
261 	put_pid(file->f_owner.pid);
262 	file_kill(file);
263 	if (file->f_mode & FMODE_WRITE)
264 		drop_file_write_access(file);
265 	file->f_path.dentry = NULL;
266 	file->f_path.mnt = NULL;
267 	file_free(file);
268 	dput(dentry);
269 	mntput(mnt);
270 }
271 
272 struct 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 
291 EXPORT_SYMBOL(fget);
292 
293 /*
294  * Lightweight file lookup - no refcnt increment if fd table isn't shared.
295  * You can use this only if it is guranteed that the current task already
296  * holds a refcnt to that file. That check has to be done at fget() only
297  * and a flag is returned to be passed to the corresponding fput_light().
298  * There must not be a cloning between an fget_light/fput_light pair.
299  */
300 struct file *fget_light(unsigned int fd, int *fput_needed)
301 {
302 	struct file *file;
303 	struct files_struct *files = current->files;
304 
305 	*fput_needed = 0;
306 	if (likely((atomic_read(&files->count) == 1))) {
307 		file = fcheck_files(files, fd);
308 	} else {
309 		rcu_read_lock();
310 		file = fcheck_files(files, fd);
311 		if (file) {
312 			if (atomic_long_inc_not_zero(&file->f_count))
313 				*fput_needed = 1;
314 			else
315 				/* Didn't get the reference, someone's freed */
316 				file = NULL;
317 		}
318 		rcu_read_unlock();
319 	}
320 
321 	return file;
322 }
323 
324 
325 void put_filp(struct file *file)
326 {
327 	if (atomic_long_dec_and_test(&file->f_count)) {
328 		security_file_free(file);
329 		file_kill(file);
330 		file_free(file);
331 	}
332 }
333 
334 void file_move(struct file *file, struct list_head *list)
335 {
336 	if (!list)
337 		return;
338 	file_list_lock();
339 	list_move(&file->f_u.fu_list, list);
340 	file_list_unlock();
341 }
342 
343 void file_kill(struct file *file)
344 {
345 	if (!list_empty(&file->f_u.fu_list)) {
346 		file_list_lock();
347 		list_del_init(&file->f_u.fu_list);
348 		file_list_unlock();
349 	}
350 }
351 
352 int fs_may_remount_ro(struct super_block *sb)
353 {
354 	struct file *file;
355 
356 	/* Check that no files are currently opened for writing. */
357 	file_list_lock();
358 	list_for_each_entry(file, &sb->s_files, f_u.fu_list) {
359 		struct inode *inode = file->f_path.dentry->d_inode;
360 
361 		/* File with pending delete? */
362 		if (inode->i_nlink == 0)
363 			goto too_bad;
364 
365 		/* Writeable file? */
366 		if (S_ISREG(inode->i_mode) && (file->f_mode & FMODE_WRITE))
367 			goto too_bad;
368 	}
369 	file_list_unlock();
370 	return 1; /* Tis' cool bro. */
371 too_bad:
372 	file_list_unlock();
373 	return 0;
374 }
375 
376 /**
377  *	mark_files_ro - mark all files read-only
378  *	@sb: superblock in question
379  *
380  *	All files are marked read-only.  We don't care about pending
381  *	delete files so this should be used in 'force' mode only.
382  */
383 void mark_files_ro(struct super_block *sb)
384 {
385 	struct file *f;
386 
387 retry:
388 	file_list_lock();
389 	list_for_each_entry(f, &sb->s_files, f_u.fu_list) {
390 		struct vfsmount *mnt;
391 		if (!S_ISREG(f->f_path.dentry->d_inode->i_mode))
392 		       continue;
393 		if (!file_count(f))
394 			continue;
395 		if (!(f->f_mode & FMODE_WRITE))
396 			continue;
397 		f->f_mode &= ~FMODE_WRITE;
398 		if (file_check_writeable(f) != 0)
399 			continue;
400 		file_release_write(f);
401 		mnt = mntget(f->f_path.mnt);
402 		file_list_unlock();
403 		/*
404 		 * This can sleep, so we can't hold
405 		 * the file_list_lock() spinlock.
406 		 */
407 		mnt_drop_write(mnt);
408 		mntput(mnt);
409 		goto retry;
410 	}
411 	file_list_unlock();
412 }
413 
414 void __init files_init(unsigned long mempages)
415 {
416 	int n;
417 
418 	filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
419 			SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
420 
421 	/*
422 	 * One file with associated inode and dcache is very roughly 1K.
423 	 * Per default don't use more than 10% of our memory for files.
424 	 */
425 
426 	n = (mempages * (PAGE_SIZE / 1024)) / 10;
427 	files_stat.max_files = n;
428 	if (files_stat.max_files < NR_FILE)
429 		files_stat.max_files = NR_FILE;
430 	files_defer_init();
431 	percpu_counter_init(&nr_files, 0);
432 }
433