xref: /openbmc/linux/fs/file_table.c (revision d87c25e8) 
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 static 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 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
79 
80 /*
81  * Handle nr_files sysctl
82  */
83 static int proc_nr_files(struct ctl_table *table, int write, void *buffer,
84 			 size_t *lenp, loff_t *ppos)
85 {
86 	files_stat.nr_files = get_nr_files();
87 	return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
88 }
89 
90 static struct ctl_table fs_stat_sysctls[] = {
91 	{
92 		.procname	= "file-nr",
93 		.data		= &files_stat,
94 		.maxlen		= sizeof(files_stat),
95 		.mode		= 0444,
96 		.proc_handler	= proc_nr_files,
97 	},
98 	{
99 		.procname	= "file-max",
100 		.data		= &files_stat.max_files,
101 		.maxlen		= sizeof(files_stat.max_files),
102 		.mode		= 0644,
103 		.proc_handler	= proc_doulongvec_minmax,
104 		.extra1		= SYSCTL_LONG_ZERO,
105 		.extra2		= SYSCTL_LONG_MAX,
106 	},
107 	{
108 		.procname	= "nr_open",
109 		.data		= &sysctl_nr_open,
110 		.maxlen		= sizeof(unsigned int),
111 		.mode		= 0644,
112 		.proc_handler	= proc_dointvec_minmax,
113 		.extra1		= &sysctl_nr_open_min,
114 		.extra2		= &sysctl_nr_open_max,
115 	},
116 	{ }
117 };
118 
119 static int __init init_fs_stat_sysctls(void)
120 {
121 	register_sysctl_init("fs", fs_stat_sysctls);
122 	if (IS_ENABLED(CONFIG_BINFMT_MISC))
123 		register_sysctl_mount_point("fs/binfmt_misc");
124 	return 0;
125 }
126 fs_initcall(init_fs_stat_sysctls);
127 #endif
128 
129 static struct file *__alloc_file(int flags, const struct cred *cred)
130 {
131 	struct file *f;
132 	int error;
133 
134 	f = kmem_cache_zalloc(filp_cachep, GFP_KERNEL);
135 	if (unlikely(!f))
136 		return ERR_PTR(-ENOMEM);
137 
138 	f->f_cred = get_cred(cred);
139 	error = security_file_alloc(f);
140 	if (unlikely(error)) {
141 		file_free_rcu(&f->f_u.fu_rcuhead);
142 		return ERR_PTR(error);
143 	}
144 
145 	atomic_long_set(&f->f_count, 1);
146 	rwlock_init(&f->f_owner.lock);
147 	spin_lock_init(&f->f_lock);
148 	mutex_init(&f->f_pos_lock);
149 	f->f_flags = flags;
150 	f->f_mode = OPEN_FMODE(flags);
151 	/* f->f_version: 0 */
152 
153 	return f;
154 }
155 
156 /* Find an unused file structure and return a pointer to it.
157  * Returns an error pointer if some error happend e.g. we over file
158  * structures limit, run out of memory or operation is not permitted.
159  *
160  * Be very careful using this.  You are responsible for
161  * getting write access to any mount that you might assign
162  * to this filp, if it is opened for write.  If this is not
163  * done, you will imbalance int the mount's writer count
164  * and a warning at __fput() time.
165  */
166 struct file *alloc_empty_file(int flags, const struct cred *cred)
167 {
168 	static long old_max;
169 	struct file *f;
170 
171 	/*
172 	 * Privileged users can go above max_files
173 	 */
174 	if (get_nr_files() >= files_stat.max_files && !capable(CAP_SYS_ADMIN)) {
175 		/*
176 		 * percpu_counters are inaccurate.  Do an expensive check before
177 		 * we go and fail.
178 		 */
179 		if (percpu_counter_sum_positive(&nr_files) >= files_stat.max_files)
180 			goto over;
181 	}
182 
183 	f = __alloc_file(flags, cred);
184 	if (!IS_ERR(f))
185 		percpu_counter_inc(&nr_files);
186 
187 	return f;
188 
189 over:
190 	/* Ran out of filps - report that */
191 	if (get_nr_files() > old_max) {
192 		pr_info("VFS: file-max limit %lu reached\n", get_max_files());
193 		old_max = get_nr_files();
194 	}
195 	return ERR_PTR(-ENFILE);
196 }
197 
198 /*
199  * Variant of alloc_empty_file() that doesn't check and modify nr_files.
200  *
201  * Should not be used unless there's a very good reason to do so.
202  */
203 struct file *alloc_empty_file_noaccount(int flags, const struct cred *cred)
204 {
205 	struct file *f = __alloc_file(flags, cred);
206 
207 	if (!IS_ERR(f))
208 		f->f_mode |= FMODE_NOACCOUNT;
209 
210 	return f;
211 }
212 
213 /**
214  * alloc_file - allocate and initialize a 'struct file'
215  *
216  * @path: the (dentry, vfsmount) pair for the new file
217  * @flags: O_... flags with which the new file will be opened
218  * @fop: the 'struct file_operations' for the new file
219  */
220 static struct file *alloc_file(const struct path *path, int flags,
221 		const struct file_operations *fop)
222 {
223 	struct file *file;
224 
225 	file = alloc_empty_file(flags, current_cred());
226 	if (IS_ERR(file))
227 		return file;
228 
229 	file->f_path = *path;
230 	file->f_inode = path->dentry->d_inode;
231 	file->f_mapping = path->dentry->d_inode->i_mapping;
232 	file->f_wb_err = filemap_sample_wb_err(file->f_mapping);
233 	file->f_sb_err = file_sample_sb_err(file);
234 	if ((file->f_mode & FMODE_READ) &&
235 	     likely(fop->read || fop->read_iter))
236 		file->f_mode |= FMODE_CAN_READ;
237 	if ((file->f_mode & FMODE_WRITE) &&
238 	     likely(fop->write || fop->write_iter))
239 		file->f_mode |= FMODE_CAN_WRITE;
240 	file->f_mode |= FMODE_OPENED;
241 	file->f_op = fop;
242 	if ((file->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
243 		i_readcount_inc(path->dentry->d_inode);
244 	return file;
245 }
246 
247 struct file *alloc_file_pseudo(struct inode *inode, struct vfsmount *mnt,
248 				const char *name, int flags,
249 				const struct file_operations *fops)
250 {
251 	static const struct dentry_operations anon_ops = {
252 		.d_dname = simple_dname
253 	};
254 	struct qstr this = QSTR_INIT(name, strlen(name));
255 	struct path path;
256 	struct file *file;
257 
258 	path.dentry = d_alloc_pseudo(mnt->mnt_sb, &this);
259 	if (!path.dentry)
260 		return ERR_PTR(-ENOMEM);
261 	if (!mnt->mnt_sb->s_d_op)
262 		d_set_d_op(path.dentry, &anon_ops);
263 	path.mnt = mntget(mnt);
264 	d_instantiate(path.dentry, inode);
265 	file = alloc_file(&path, flags, fops);
266 	if (IS_ERR(file)) {
267 		ihold(inode);
268 		path_put(&path);
269 	}
270 	return file;
271 }
272 EXPORT_SYMBOL(alloc_file_pseudo);
273 
274 struct file *alloc_file_clone(struct file *base, int flags,
275 				const struct file_operations *fops)
276 {
277 	struct file *f = alloc_file(&base->f_path, flags, fops);
278 	if (!IS_ERR(f)) {
279 		path_get(&f->f_path);
280 		f->f_mapping = base->f_mapping;
281 	}
282 	return f;
283 }
284 
285 /* the real guts of fput() - releasing the last reference to file
286  */
287 static void __fput(struct file *file)
288 {
289 	struct dentry *dentry = file->f_path.dentry;
290 	struct vfsmount *mnt = file->f_path.mnt;
291 	struct inode *inode = file->f_inode;
292 	fmode_t mode = file->f_mode;
293 
294 	if (unlikely(!(file->f_mode & FMODE_OPENED)))
295 		goto out;
296 
297 	might_sleep();
298 
299 	fsnotify_close(file);
300 	/*
301 	 * The function eventpoll_release() should be the first called
302 	 * in the file cleanup chain.
303 	 */
304 	eventpoll_release(file);
305 	locks_remove_file(file);
306 
307 	ima_file_free(file);
308 	if (unlikely(file->f_flags & FASYNC)) {
309 		if (file->f_op->fasync)
310 			file->f_op->fasync(-1, file, 0);
311 	}
312 	if (file->f_op->release)
313 		file->f_op->release(inode, file);
314 	if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL &&
315 		     !(mode & FMODE_PATH))) {
316 		cdev_put(inode->i_cdev);
317 	}
318 	fops_put(file->f_op);
319 	put_pid(file->f_owner.pid);
320 	if ((mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
321 		i_readcount_dec(inode);
322 	if (mode & FMODE_WRITER) {
323 		put_write_access(inode);
324 		__mnt_drop_write(mnt);
325 	}
326 	dput(dentry);
327 	if (unlikely(mode & FMODE_NEED_UNMOUNT))
328 		dissolve_on_fput(mnt);
329 	mntput(mnt);
330 out:
331 	file_free(file);
332 }
333 
334 static LLIST_HEAD(delayed_fput_list);
335 static void delayed_fput(struct work_struct *unused)
336 {
337 	struct llist_node *node = llist_del_all(&delayed_fput_list);
338 	struct file *f, *t;
339 
340 	llist_for_each_entry_safe(f, t, node, f_u.fu_llist)
341 		__fput(f);
342 }
343 
344 static void ____fput(struct callback_head *work)
345 {
346 	__fput(container_of(work, struct file, f_u.fu_rcuhead));
347 }
348 
349 /*
350  * If kernel thread really needs to have the final fput() it has done
351  * to complete, call this.  The only user right now is the boot - we
352  * *do* need to make sure our writes to binaries on initramfs has
353  * not left us with opened struct file waiting for __fput() - execve()
354  * won't work without that.  Please, don't add more callers without
355  * very good reasons; in particular, never call that with locks
356  * held and never call that from a thread that might need to do
357  * some work on any kind of umount.
358  */
359 void flush_delayed_fput(void)
360 {
361 	delayed_fput(NULL);
362 }
363 EXPORT_SYMBOL_GPL(flush_delayed_fput);
364 
365 static DECLARE_DELAYED_WORK(delayed_fput_work, delayed_fput);
366 
367 void fput_many(struct file *file, unsigned int refs)
368 {
369 	if (atomic_long_sub_and_test(refs, &file->f_count)) {
370 		struct task_struct *task = current;
371 
372 		if (likely(!in_interrupt() && !(task->flags & PF_KTHREAD))) {
373 			init_task_work(&file->f_u.fu_rcuhead, ____fput);
374 			if (!task_work_add(task, &file->f_u.fu_rcuhead, TWA_RESUME))
375 				return;
376 			/*
377 			 * After this task has run exit_task_work(),
378 			 * task_work_add() will fail.  Fall through to delayed
379 			 * fput to avoid leaking *file.
380 			 */
381 		}
382 
383 		if (llist_add(&file->f_u.fu_llist, &delayed_fput_list))
384 			schedule_delayed_work(&delayed_fput_work, 1);
385 	}
386 }
387 
388 void fput(struct file *file)
389 {
390 	fput_many(file, 1);
391 }
392 
393 /*
394  * synchronous analog of fput(); for kernel threads that might be needed
395  * in some umount() (and thus can't use flush_delayed_fput() without
396  * risking deadlocks), need to wait for completion of __fput() and know
397  * for this specific struct file it won't involve anything that would
398  * need them.  Use only if you really need it - at the very least,
399  * don't blindly convert fput() by kernel thread to that.
400  */
401 void __fput_sync(struct file *file)
402 {
403 	if (atomic_long_dec_and_test(&file->f_count)) {
404 		struct task_struct *task = current;
405 		BUG_ON(!(task->flags & PF_KTHREAD));
406 		__fput(file);
407 	}
408 }
409 
410 EXPORT_SYMBOL(fput);
411 
412 void __init files_init(void)
413 {
414 	filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
415 			SLAB_HWCACHE_ALIGN | SLAB_PANIC | SLAB_ACCOUNT, NULL);
416 	percpu_counter_init(&nr_files, 0, GFP_KERNEL);
417 }
418 
419 /*
420  * One file with associated inode and dcache is very roughly 1K. Per default
421  * do not use more than 10% of our memory for files.
422  */
423 void __init files_maxfiles_init(void)
424 {
425 	unsigned long n;
426 	unsigned long nr_pages = totalram_pages();
427 	unsigned long memreserve = (nr_pages - nr_free_pages()) * 3/2;
428 
429 	memreserve = min(memreserve, nr_pages - 1);
430 	n = ((nr_pages - memreserve) * (PAGE_SIZE / 1024)) / 10;
431 
432 	files_stat.max_files = max_t(unsigned long, n, NR_FILE);
433 }
434