xref: /linux/fs/file_table.c (revision 93df8a1ed6231727c5db94a80b1a6bd5ee67cec3)
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/percpu.h>
25 #include <linux/hardirq.h>
26 #include <linux/task_work.h>
27 #include <linux/ima.h>
28 
29 #include <linux/atomic.h>
30 
31 #include "internal.h"
32 
33 /* sysctl tunables... */
34 struct files_stat_struct files_stat = {
35 	.max_files = NR_FILE
36 };
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 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 	call_rcu(&f->f_u.fu_rcuhead, file_free_rcu);
55 }
56 
57 /*
58  * Return the total number of open files in the system
59  */
60 static long get_nr_files(void)
61 {
62 	return percpu_counter_read_positive(&nr_files);
63 }
64 
65 /*
66  * Return the maximum number of open files in the system
67  */
68 unsigned long get_max_files(void)
69 {
70 	return files_stat.max_files;
71 }
72 EXPORT_SYMBOL_GPL(get_max_files);
73 
74 /*
75  * Handle nr_files sysctl
76  */
77 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
78 int proc_nr_files(struct ctl_table *table, int write,
79                      void __user *buffer, size_t *lenp, loff_t *ppos)
80 {
81 	files_stat.nr_files = get_nr_files();
82 	return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
83 }
84 #else
85 int proc_nr_files(struct ctl_table *table, int write,
86                      void __user *buffer, size_t *lenp, loff_t *ppos)
87 {
88 	return -ENOSYS;
89 }
90 #endif
91 
92 /* Find an unused file structure and return a pointer to it.
93  * Returns an error pointer if some error happend e.g. we over file
94  * structures limit, run out of memory or operation is not permitted.
95  *
96  * Be very careful using this.  You are responsible for
97  * getting write access to any mount that you might assign
98  * to this filp, if it is opened for write.  If this is not
99  * done, you will imbalance int the mount's writer count
100  * and a warning at __fput() time.
101  */
102 struct file *get_empty_filp(void)
103 {
104 	const struct cred *cred = current_cred();
105 	static long old_max;
106 	struct file *f;
107 	int error;
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 (unlikely(!f))
123 		return ERR_PTR(-ENOMEM);
124 
125 	percpu_counter_inc(&nr_files);
126 	f->f_cred = get_cred(cred);
127 	error = security_file_alloc(f);
128 	if (unlikely(error)) {
129 		file_free(f);
130 		return ERR_PTR(error);
131 	}
132 
133 	atomic_long_set(&f->f_count, 1);
134 	rwlock_init(&f->f_owner.lock);
135 	spin_lock_init(&f->f_lock);
136 	mutex_init(&f->f_pos_lock);
137 	eventpoll_init_file(f);
138 	/* f->f_version: 0 */
139 	return f;
140 
141 over:
142 	/* Ran out of filps - report that */
143 	if (get_nr_files() > old_max) {
144 		pr_info("VFS: file-max limit %lu reached\n", get_max_files());
145 		old_max = get_nr_files();
146 	}
147 	return ERR_PTR(-ENFILE);
148 }
149 
150 /**
151  * alloc_file - allocate and initialize a 'struct file'
152  *
153  * @path: the (dentry, vfsmount) pair for the new file
154  * @mode: the mode with which the new file will be opened
155  * @fop: the 'struct file_operations' for the new file
156  */
157 struct file *alloc_file(struct path *path, fmode_t mode,
158 		const struct file_operations *fop)
159 {
160 	struct file *file;
161 
162 	file = get_empty_filp();
163 	if (IS_ERR(file))
164 		return file;
165 
166 	file->f_path = *path;
167 	file->f_inode = path->dentry->d_inode;
168 	file->f_mapping = path->dentry->d_inode->i_mapping;
169 	if ((mode & FMODE_READ) &&
170 	     likely(fop->read || fop->read_iter))
171 		mode |= FMODE_CAN_READ;
172 	if ((mode & FMODE_WRITE) &&
173 	     likely(fop->write || fop->write_iter))
174 		mode |= FMODE_CAN_WRITE;
175 	file->f_mode = mode;
176 	file->f_op = fop;
177 	if ((mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
178 		i_readcount_inc(path->dentry->d_inode);
179 	return file;
180 }
181 EXPORT_SYMBOL(alloc_file);
182 
183 /* the real guts of fput() - releasing the last reference to file
184  */
185 static void __fput(struct file *file)
186 {
187 	struct dentry *dentry = file->f_path.dentry;
188 	struct vfsmount *mnt = file->f_path.mnt;
189 	struct inode *inode = file->f_inode;
190 
191 	might_sleep();
192 
193 	fsnotify_close(file);
194 	/*
195 	 * The function eventpoll_release() should be the first called
196 	 * in the file cleanup chain.
197 	 */
198 	eventpoll_release(file);
199 	locks_remove_file(file);
200 
201 	if (unlikely(file->f_flags & FASYNC)) {
202 		if (file->f_op->fasync)
203 			file->f_op->fasync(-1, file, 0);
204 	}
205 	ima_file_free(file);
206 	if (file->f_op->release)
207 		file->f_op->release(inode, file);
208 	security_file_free(file);
209 	if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL &&
210 		     !(file->f_mode & FMODE_PATH))) {
211 		cdev_put(inode->i_cdev);
212 	}
213 	fops_put(file->f_op);
214 	put_pid(file->f_owner.pid);
215 	if ((file->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
216 		i_readcount_dec(inode);
217 	if (file->f_mode & FMODE_WRITER) {
218 		put_write_access(inode);
219 		__mnt_drop_write(mnt);
220 	}
221 	file->f_path.dentry = NULL;
222 	file->f_path.mnt = NULL;
223 	file->f_inode = NULL;
224 	file_free(file);
225 	dput(dentry);
226 	mntput(mnt);
227 }
228 
229 static LLIST_HEAD(delayed_fput_list);
230 static void delayed_fput(struct work_struct *unused)
231 {
232 	struct llist_node *node = llist_del_all(&delayed_fput_list);
233 	struct llist_node *next;
234 
235 	for (; node; node = next) {
236 		next = llist_next(node);
237 		__fput(llist_entry(node, struct file, f_u.fu_llist));
238 	}
239 }
240 
241 static void ____fput(struct callback_head *work)
242 {
243 	__fput(container_of(work, struct file, f_u.fu_rcuhead));
244 }
245 
246 /*
247  * If kernel thread really needs to have the final fput() it has done
248  * to complete, call this.  The only user right now is the boot - we
249  * *do* need to make sure our writes to binaries on initramfs has
250  * not left us with opened struct file waiting for __fput() - execve()
251  * won't work without that.  Please, don't add more callers without
252  * very good reasons; in particular, never call that with locks
253  * held and never call that from a thread that might need to do
254  * some work on any kind of umount.
255  */
256 void flush_delayed_fput(void)
257 {
258 	delayed_fput(NULL);
259 }
260 
261 static DECLARE_DELAYED_WORK(delayed_fput_work, delayed_fput);
262 
263 void fput(struct file *file)
264 {
265 	if (atomic_long_dec_and_test(&file->f_count)) {
266 		struct task_struct *task = current;
267 
268 		if (likely(!in_interrupt() && !(task->flags & PF_KTHREAD))) {
269 			init_task_work(&file->f_u.fu_rcuhead, ____fput);
270 			if (!task_work_add(task, &file->f_u.fu_rcuhead, true))
271 				return;
272 			/*
273 			 * After this task has run exit_task_work(),
274 			 * task_work_add() will fail.  Fall through to delayed
275 			 * fput to avoid leaking *file.
276 			 */
277 		}
278 
279 		if (llist_add(&file->f_u.fu_llist, &delayed_fput_list))
280 			schedule_delayed_work(&delayed_fput_work, 1);
281 	}
282 }
283 
284 /*
285  * synchronous analog of fput(); for kernel threads that might be needed
286  * in some umount() (and thus can't use flush_delayed_fput() without
287  * risking deadlocks), need to wait for completion of __fput() and know
288  * for this specific struct file it won't involve anything that would
289  * need them.  Use only if you really need it - at the very least,
290  * don't blindly convert fput() by kernel thread to that.
291  */
292 void __fput_sync(struct file *file)
293 {
294 	if (atomic_long_dec_and_test(&file->f_count)) {
295 		struct task_struct *task = current;
296 		BUG_ON(!(task->flags & PF_KTHREAD));
297 		__fput(file);
298 	}
299 }
300 
301 EXPORT_SYMBOL(fput);
302 
303 void put_filp(struct file *file)
304 {
305 	if (atomic_long_dec_and_test(&file->f_count)) {
306 		security_file_free(file);
307 		file_free(file);
308 	}
309 }
310 
311 void __init files_init(unsigned long mempages)
312 {
313 	unsigned long n;
314 
315 	filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
316 			SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
317 
318 	/*
319 	 * One file with associated inode and dcache is very roughly 1K.
320 	 * Per default don't use more than 10% of our memory for files.
321 	 */
322 
323 	n = (mempages * (PAGE_SIZE / 1024)) / 10;
324 	files_stat.max_files = max_t(unsigned long, n, NR_FILE);
325 	percpu_counter_init(&nr_files, 0, GFP_KERNEL);
326 }
327