xref: /linux/fs/file_table.c (revision 827634added7f38b7d724cab1dccdb2b004c13c3)
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/hardirq.h>
27 #include <linux/task_work.h>
28 #include <linux/ima.h>
29 
30 #include <linux/atomic.h>
31 
32 #include "internal.h"
33 
34 /* sysctl tunables... */
35 struct files_stat_struct files_stat = {
36 	.max_files = NR_FILE
37 };
38 
39 /* SLAB cache for file structures */
40 static struct kmem_cache *filp_cachep __read_mostly;
41 
42 static struct percpu_counter nr_files __cacheline_aligned_in_smp;
43 
44 static void file_free_rcu(struct rcu_head *head)
45 {
46 	struct file *f = container_of(head, struct file, f_u.fu_rcuhead);
47 
48 	put_cred(f->f_cred);
49 	kmem_cache_free(filp_cachep, f);
50 }
51 
52 static inline void file_free(struct file *f)
53 {
54 	percpu_counter_dec(&nr_files);
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 long 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 unsigned long 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(struct 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_doulongvec_minmax(table, write, buffer, lenp, ppos);
84 }
85 #else
86 int proc_nr_files(struct 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 an error pointer if some error happend e.g. we over file
95  * structures limit, run out of memory or operation is not permitted.
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 long old_max;
107 	struct file *f;
108 	int error;
109 
110 	/*
111 	 * Privileged users can go above max_files
112 	 */
113 	if (get_nr_files() >= files_stat.max_files && !capable(CAP_SYS_ADMIN)) {
114 		/*
115 		 * percpu_counters are inaccurate.  Do an expensive check before
116 		 * we go and fail.
117 		 */
118 		if (percpu_counter_sum_positive(&nr_files) >= files_stat.max_files)
119 			goto over;
120 	}
121 
122 	f = kmem_cache_zalloc(filp_cachep, GFP_KERNEL);
123 	if (unlikely(!f))
124 		return ERR_PTR(-ENOMEM);
125 
126 	percpu_counter_inc(&nr_files);
127 	f->f_cred = get_cred(cred);
128 	error = security_file_alloc(f);
129 	if (unlikely(error)) {
130 		file_free(f);
131 		return ERR_PTR(error);
132 	}
133 
134 	atomic_long_set(&f->f_count, 1);
135 	rwlock_init(&f->f_owner.lock);
136 	spin_lock_init(&f->f_lock);
137 	mutex_init(&f->f_pos_lock);
138 	eventpoll_init_file(f);
139 	/* f->f_version: 0 */
140 	return f;
141 
142 over:
143 	/* Ran out of filps - report that */
144 	if (get_nr_files() > old_max) {
145 		pr_info("VFS: file-max limit %lu reached\n", get_max_files());
146 		old_max = get_nr_files();
147 	}
148 	return ERR_PTR(-ENFILE);
149 }
150 
151 /**
152  * alloc_file - allocate and initialize a 'struct file'
153  *
154  * @path: the (dentry, vfsmount) pair for the new file
155  * @mode: the mode with which the new file will be opened
156  * @fop: the 'struct file_operations' for the new file
157  */
158 struct file *alloc_file(struct path *path, fmode_t mode,
159 		const struct file_operations *fop)
160 {
161 	struct file *file;
162 
163 	file = get_empty_filp();
164 	if (IS_ERR(file))
165 		return file;
166 
167 	file->f_path = *path;
168 	file->f_inode = path->dentry->d_inode;
169 	file->f_mapping = path->dentry->d_inode->i_mapping;
170 	if ((mode & FMODE_READ) &&
171 	     likely(fop->read || fop->read_iter))
172 		mode |= FMODE_CAN_READ;
173 	if ((mode & FMODE_WRITE) &&
174 	     likely(fop->write || fop->write_iter))
175 		mode |= FMODE_CAN_WRITE;
176 	file->f_mode = mode;
177 	file->f_op = fop;
178 	if ((mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
179 		i_readcount_inc(path->dentry->d_inode);
180 	return file;
181 }
182 EXPORT_SYMBOL(alloc_file);
183 
184 /* the real guts of fput() - releasing the last reference to file
185  */
186 static void __fput(struct file *file)
187 {
188 	struct dentry *dentry = file->f_path.dentry;
189 	struct vfsmount *mnt = file->f_path.mnt;
190 	struct inode *inode = file->f_inode;
191 
192 	might_sleep();
193 
194 	fsnotify_close(file);
195 	/*
196 	 * The function eventpoll_release() should be the first called
197 	 * in the file cleanup chain.
198 	 */
199 	eventpoll_release(file);
200 	locks_remove_file(file);
201 
202 	if (unlikely(file->f_flags & FASYNC)) {
203 		if (file->f_op->fasync)
204 			file->f_op->fasync(-1, file, 0);
205 	}
206 	ima_file_free(file);
207 	if (file->f_op->release)
208 		file->f_op->release(inode, file);
209 	security_file_free(file);
210 	if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL &&
211 		     !(file->f_mode & FMODE_PATH))) {
212 		cdev_put(inode->i_cdev);
213 	}
214 	fops_put(file->f_op);
215 	put_pid(file->f_owner.pid);
216 	if ((file->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
217 		i_readcount_dec(inode);
218 	if (file->f_mode & FMODE_WRITER) {
219 		put_write_access(inode);
220 		__mnt_drop_write(mnt);
221 	}
222 	file->f_path.dentry = NULL;
223 	file->f_path.mnt = NULL;
224 	file->f_inode = NULL;
225 	file_free(file);
226 	dput(dentry);
227 	mntput(mnt);
228 }
229 
230 static LLIST_HEAD(delayed_fput_list);
231 static void delayed_fput(struct work_struct *unused)
232 {
233 	struct llist_node *node = llist_del_all(&delayed_fput_list);
234 	struct llist_node *next;
235 
236 	for (; node; node = next) {
237 		next = llist_next(node);
238 		__fput(llist_entry(node, struct file, f_u.fu_llist));
239 	}
240 }
241 
242 static void ____fput(struct callback_head *work)
243 {
244 	__fput(container_of(work, struct file, f_u.fu_rcuhead));
245 }
246 
247 /*
248  * If kernel thread really needs to have the final fput() it has done
249  * to complete, call this.  The only user right now is the boot - we
250  * *do* need to make sure our writes to binaries on initramfs has
251  * not left us with opened struct file waiting for __fput() - execve()
252  * won't work without that.  Please, don't add more callers without
253  * very good reasons; in particular, never call that with locks
254  * held and never call that from a thread that might need to do
255  * some work on any kind of umount.
256  */
257 void flush_delayed_fput(void)
258 {
259 	delayed_fput(NULL);
260 }
261 
262 static DECLARE_DELAYED_WORK(delayed_fput_work, delayed_fput);
263 
264 void fput(struct file *file)
265 {
266 	if (atomic_long_dec_and_test(&file->f_count)) {
267 		struct task_struct *task = current;
268 
269 		if (likely(!in_interrupt() && !(task->flags & PF_KTHREAD))) {
270 			init_task_work(&file->f_u.fu_rcuhead, ____fput);
271 			if (!task_work_add(task, &file->f_u.fu_rcuhead, true))
272 				return;
273 			/*
274 			 * After this task has run exit_task_work(),
275 			 * task_work_add() will fail.  Fall through to delayed
276 			 * fput to avoid leaking *file.
277 			 */
278 		}
279 
280 		if (llist_add(&file->f_u.fu_llist, &delayed_fput_list))
281 			schedule_delayed_work(&delayed_fput_work, 1);
282 	}
283 }
284 
285 /*
286  * synchronous analog of fput(); for kernel threads that might be needed
287  * in some umount() (and thus can't use flush_delayed_fput() without
288  * risking deadlocks), need to wait for completion of __fput() and know
289  * for this specific struct file it won't involve anything that would
290  * need them.  Use only if you really need it - at the very least,
291  * don't blindly convert fput() by kernel thread to that.
292  */
293 void __fput_sync(struct file *file)
294 {
295 	if (atomic_long_dec_and_test(&file->f_count)) {
296 		struct task_struct *task = current;
297 		BUG_ON(!(task->flags & PF_KTHREAD));
298 		__fput(file);
299 	}
300 }
301 
302 EXPORT_SYMBOL(fput);
303 
304 void put_filp(struct file *file)
305 {
306 	if (atomic_long_dec_and_test(&file->f_count)) {
307 		security_file_free(file);
308 		file_free(file);
309 	}
310 }
311 
312 void __init files_init(unsigned long mempages)
313 {
314 	unsigned long n;
315 
316 	filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
317 			SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
318 
319 	/*
320 	 * One file with associated inode and dcache is very roughly 1K.
321 	 * Per default don't use more than 10% of our memory for files.
322 	 */
323 
324 	n = (mempages * (PAGE_SIZE / 1024)) / 10;
325 	files_stat.max_files = max_t(unsigned long, n, NR_FILE);
326 	percpu_counter_init(&nr_files, 0, GFP_KERNEL);
327 }
328