xref: /linux/fs/file_table.c (revision 23f50b5ac331c8c27c421a7116618355508e8427) 
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *  linux/fs/file_table.c
4  *
5  *  Copyright (C) 1991, 1992  Linus Torvalds
6  *  Copyright (C) 1997 David S. Miller (davem@caip.rutgers.edu)
7  */
8 
9 #include <linux/string.h>
10 #include <linux/slab.h>
11 #include <linux/file.h>
12 #include <linux/fdtable.h>
13 #include <linux/init.h>
14 #include <linux/module.h>
15 #include <linux/fs.h>
16 #include <linux/security.h>
17 #include <linux/cred.h>
18 #include <linux/eventpoll.h>
19 #include <linux/rcupdate.h>
20 #include <linux/mount.h>
21 #include <linux/capability.h>
22 #include <linux/cdev.h>
23 #include <linux/fsnotify.h>
24 #include <linux/sysctl.h>
25 #include <linux/percpu_counter.h>
26 #include <linux/percpu.h>
27 #include <linux/task_work.h>
28 #include <linux/ima.h>
29 #include <linux/swap.h>
30 
31 #include <linux/atomic.h>
32 
33 #include "internal.h"
34 
35 /* sysctl tunables... */
36 struct files_stat_struct files_stat = {
37 	.max_files = NR_FILE
38 };
39 
40 /* SLAB cache for file structures */
41 static struct kmem_cache *filp_cachep __read_mostly;
42 
43 static struct percpu_counter nr_files __cacheline_aligned_in_smp;
44 
45 static void file_free_rcu(struct rcu_head *head)
46 {
47 	struct file *f = container_of(head, struct file, f_u.fu_rcuhead);
48 
49 	put_cred(f->f_cred);
50 	kmem_cache_free(filp_cachep, f);
51 }
52 
53 static inline void file_free(struct file *f)
54 {
55 	security_file_free(f);
56 	if (!(f->f_mode & FMODE_NOACCOUNT))
57 		percpu_counter_dec(&nr_files);
58 	call_rcu(&f->f_u.fu_rcuhead, file_free_rcu);
59 }
60 
61 /*
62  * Return the total number of open files in the system
63  */
64 static long get_nr_files(void)
65 {
66 	return percpu_counter_read_positive(&nr_files);
67 }
68 
69 /*
70  * Return the maximum number of open files in the system
71  */
72 unsigned long get_max_files(void)
73 {
74 	return files_stat.max_files;
75 }
76 EXPORT_SYMBOL_GPL(get_max_files);
77 
78 /*
79  * Handle nr_files sysctl
80  */
81 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
82 int proc_nr_files(struct ctl_table *table, int write,
83                      void *buffer, size_t *lenp, loff_t *ppos)
84 {
85 	files_stat.nr_files = get_nr_files();
86 	return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
87 }
88 #else
89 int proc_nr_files(struct ctl_table *table, int write,
90                      void *buffer, size_t *lenp, loff_t *ppos)
91 {
92 	return -ENOSYS;
93 }
94 #endif
95 
96 static struct file *__alloc_file(int flags, const struct cred *cred)
97 {
98 	struct file *f;
99 	int error;
100 
101 	f = kmem_cache_zalloc(filp_cachep, GFP_KERNEL);
102 	if (unlikely(!f))
103 		return ERR_PTR(-ENOMEM);
104 
105 	f->f_cred = get_cred(cred);
106 	error = security_file_alloc(f);
107 	if (unlikely(error)) {
108 		file_free_rcu(&f->f_u.fu_rcuhead);
109 		return ERR_PTR(error);
110 	}
111 
112 	atomic_long_set(&f->f_count, 1);
113 	rwlock_init(&f->f_owner.lock);
114 	spin_lock_init(&f->f_lock);
115 	mutex_init(&f->f_pos_lock);
116 	f->f_flags = flags;
117 	f->f_mode = OPEN_FMODE(flags);
118 	/* f->f_version: 0 */
119 
120 	return f;
121 }
122 
123 /* Find an unused file structure and return a pointer to it.
124  * Returns an error pointer if some error happend e.g. we over file
125  * structures limit, run out of memory or operation is not permitted.
126  *
127  * Be very careful using this.  You are responsible for
128  * getting write access to any mount that you might assign
129  * to this filp, if it is opened for write.  If this is not
130  * done, you will imbalance int the mount's writer count
131  * and a warning at __fput() time.
132  */
133 struct file *alloc_empty_file(int flags, const struct cred *cred)
134 {
135 	static long old_max;
136 	struct file *f;
137 
138 	/*
139 	 * Privileged users can go above max_files
140 	 */
141 	if (get_nr_files() >= files_stat.max_files && !capable(CAP_SYS_ADMIN)) {
142 		/*
143 		 * percpu_counters are inaccurate.  Do an expensive check before
144 		 * we go and fail.
145 		 */
146 		if (percpu_counter_sum_positive(&nr_files) >= files_stat.max_files)
147 			goto over;
148 	}
149 
150 	f = __alloc_file(flags, cred);
151 	if (!IS_ERR(f))
152 		percpu_counter_inc(&nr_files);
153 
154 	return f;
155 
156 over:
157 	/* Ran out of filps - report that */
158 	if (get_nr_files() > old_max) {
159 		pr_info("VFS: file-max limit %lu reached\n", get_max_files());
160 		old_max = get_nr_files();
161 	}
162 	return ERR_PTR(-ENFILE);
163 }
164 
165 /*
166  * Variant of alloc_empty_file() that doesn't check and modify nr_files.
167  *
168  * Should not be used unless there's a very good reason to do so.
169  */
170 struct file *alloc_empty_file_noaccount(int flags, const struct cred *cred)
171 {
172 	struct file *f = __alloc_file(flags, cred);
173 
174 	if (!IS_ERR(f))
175 		f->f_mode |= FMODE_NOACCOUNT;
176 
177 	return f;
178 }
179 
180 /**
181  * alloc_file - allocate and initialize a 'struct file'
182  *
183  * @path: the (dentry, vfsmount) pair for the new file
184  * @flags: O_... flags with which the new file will be opened
185  * @fop: the 'struct file_operations' for the new file
186  */
187 static struct file *alloc_file(const struct path *path, int flags,
188 		const struct file_operations *fop)
189 {
190 	struct file *file;
191 
192 	file = alloc_empty_file(flags, current_cred());
193 	if (IS_ERR(file))
194 		return file;
195 
196 	file->f_path = *path;
197 	file->f_inode = path->dentry->d_inode;
198 	file->f_mapping = path->dentry->d_inode->i_mapping;
199 	file->f_wb_err = filemap_sample_wb_err(file->f_mapping);
200 	file->f_sb_err = file_sample_sb_err(file);
201 	if ((file->f_mode & FMODE_READ) &&
202 	     likely(fop->read || fop->read_iter))
203 		file->f_mode |= FMODE_CAN_READ;
204 	if ((file->f_mode & FMODE_WRITE) &&
205 	     likely(fop->write || fop->write_iter))
206 		file->f_mode |= FMODE_CAN_WRITE;
207 	file->f_mode |= FMODE_OPENED;
208 	file->f_op = fop;
209 	if ((file->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
210 		i_readcount_inc(path->dentry->d_inode);
211 	return file;
212 }
213 
214 struct file *alloc_file_pseudo(struct inode *inode, struct vfsmount *mnt,
215 				const char *name, int flags,
216 				const struct file_operations *fops)
217 {
218 	static const struct dentry_operations anon_ops = {
219 		.d_dname = simple_dname
220 	};
221 	struct qstr this = QSTR_INIT(name, strlen(name));
222 	struct path path;
223 	struct file *file;
224 
225 	path.dentry = d_alloc_pseudo(mnt->mnt_sb, &this);
226 	if (!path.dentry)
227 		return ERR_PTR(-ENOMEM);
228 	if (!mnt->mnt_sb->s_d_op)
229 		d_set_d_op(path.dentry, &anon_ops);
230 	path.mnt = mntget(mnt);
231 	d_instantiate(path.dentry, inode);
232 	file = alloc_file(&path, flags, fops);
233 	if (IS_ERR(file)) {
234 		ihold(inode);
235 		path_put(&path);
236 	}
237 	return file;
238 }
239 EXPORT_SYMBOL(alloc_file_pseudo);
240 
241 struct file *alloc_file_clone(struct file *base, int flags,
242 				const struct file_operations *fops)
243 {
244 	struct file *f = alloc_file(&base->f_path, flags, fops);
245 	if (!IS_ERR(f)) {
246 		path_get(&f->f_path);
247 		f->f_mapping = base->f_mapping;
248 	}
249 	return f;
250 }
251 
252 /* the real guts of fput() - releasing the last reference to file
253  */
254 static void __fput(struct file *file)
255 {
256 	struct dentry *dentry = file->f_path.dentry;
257 	struct vfsmount *mnt = file->f_path.mnt;
258 	struct inode *inode = file->f_inode;
259 	fmode_t mode = file->f_mode;
260 
261 	if (unlikely(!(file->f_mode & FMODE_OPENED)))
262 		goto out;
263 
264 	might_sleep();
265 
266 	fsnotify_close(file);
267 	/*
268 	 * The function eventpoll_release() should be the first called
269 	 * in the file cleanup chain.
270 	 */
271 	eventpoll_release(file);
272 	locks_remove_file(file);
273 
274 	ima_file_free(file);
275 	if (unlikely(file->f_flags & FASYNC)) {
276 		if (file->f_op->fasync)
277 			file->f_op->fasync(-1, file, 0);
278 	}
279 	if (file->f_op->release)
280 		file->f_op->release(inode, file);
281 	if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL &&
282 		     !(mode & FMODE_PATH))) {
283 		cdev_put(inode->i_cdev);
284 	}
285 	fops_put(file->f_op);
286 	put_pid(file->f_owner.pid);
287 	if ((mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
288 		i_readcount_dec(inode);
289 	if (mode & FMODE_WRITER) {
290 		put_write_access(inode);
291 		__mnt_drop_write(mnt);
292 	}
293 	dput(dentry);
294 	if (unlikely(mode & FMODE_NEED_UNMOUNT))
295 		dissolve_on_fput(mnt);
296 	mntput(mnt);
297 out:
298 	file_free(file);
299 }
300 
301 static LLIST_HEAD(delayed_fput_list);
302 static void delayed_fput(struct work_struct *unused)
303 {
304 	struct llist_node *node = llist_del_all(&delayed_fput_list);
305 	struct file *f, *t;
306 
307 	llist_for_each_entry_safe(f, t, node, f_u.fu_llist)
308 		__fput(f);
309 }
310 
311 static void ____fput(struct callback_head *work)
312 {
313 	__fput(container_of(work, struct file, f_u.fu_rcuhead));
314 }
315 
316 /*
317  * If kernel thread really needs to have the final fput() it has done
318  * to complete, call this.  The only user right now is the boot - we
319  * *do* need to make sure our writes to binaries on initramfs has
320  * not left us with opened struct file waiting for __fput() - execve()
321  * won't work without that.  Please, don't add more callers without
322  * very good reasons; in particular, never call that with locks
323  * held and never call that from a thread that might need to do
324  * some work on any kind of umount.
325  */
326 void flush_delayed_fput(void)
327 {
328 	delayed_fput(NULL);
329 }
330 EXPORT_SYMBOL_GPL(flush_delayed_fput);
331 
332 static DECLARE_DELAYED_WORK(delayed_fput_work, delayed_fput);
333 
334 void fput_many(struct file *file, unsigned int refs)
335 {
336 	if (atomic_long_sub_and_test(refs, &file->f_count)) {
337 		struct task_struct *task = current;
338 
339 		if (likely(!in_interrupt() && !(task->flags & PF_KTHREAD))) {
340 			init_task_work(&file->f_u.fu_rcuhead, ____fput);
341 			if (!task_work_add(task, &file->f_u.fu_rcuhead, TWA_RESUME))
342 				return;
343 			/*
344 			 * After this task has run exit_task_work(),
345 			 * task_work_add() will fail.  Fall through to delayed
346 			 * fput to avoid leaking *file.
347 			 */
348 		}
349 
350 		if (llist_add(&file->f_u.fu_llist, &delayed_fput_list))
351 			schedule_delayed_work(&delayed_fput_work, 1);
352 	}
353 }
354 
355 void fput(struct file *file)
356 {
357 	fput_many(file, 1);
358 }
359 
360 /*
361  * synchronous analog of fput(); for kernel threads that might be needed
362  * in some umount() (and thus can't use flush_delayed_fput() without
363  * risking deadlocks), need to wait for completion of __fput() and know
364  * for this specific struct file it won't involve anything that would
365  * need them.  Use only if you really need it - at the very least,
366  * don't blindly convert fput() by kernel thread to that.
367  */
368 void __fput_sync(struct file *file)
369 {
370 	if (atomic_long_dec_and_test(&file->f_count)) {
371 		struct task_struct *task = current;
372 		BUG_ON(!(task->flags & PF_KTHREAD));
373 		__fput(file);
374 	}
375 }
376 
377 EXPORT_SYMBOL(fput);
378 
379 void __init files_init(void)
380 {
381 	filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
382 			SLAB_HWCACHE_ALIGN | SLAB_PANIC | SLAB_ACCOUNT, NULL);
383 	percpu_counter_init(&nr_files, 0, GFP_KERNEL);
384 }
385 
386 /*
387  * One file with associated inode and dcache is very roughly 1K. Per default
388  * do not use more than 10% of our memory for files.
389  */
390 void __init files_maxfiles_init(void)
391 {
392 	unsigned long n;
393 	unsigned long nr_pages = totalram_pages();
394 	unsigned long memreserve = (nr_pages - nr_free_pages()) * 3/2;
395 
396 	memreserve = min(memreserve, nr_pages - 1);
397 	n = ((nr_pages - memreserve) * (PAGE_SIZE / 1024)) / 10;
398 
399 	files_stat.max_files = max_t(unsigned long, n, NR_FILE);
400 }
401