xref: /openbmc/linux/fs/file_table.c (revision e044e3e6) 
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 	return 0;
123 }
124 fs_initcall(init_fs_stat_sysctls);
125 #endif
126 
127 static struct file *__alloc_file(int flags, const struct cred *cred)
128 {
129 	struct file *f;
130 	int error;
131 
132 	f = kmem_cache_zalloc(filp_cachep, GFP_KERNEL);
133 	if (unlikely(!f))
134 		return ERR_PTR(-ENOMEM);
135 
136 	f->f_cred = get_cred(cred);
137 	error = security_file_alloc(f);
138 	if (unlikely(error)) {
139 		file_free_rcu(&f->f_u.fu_rcuhead);
140 		return ERR_PTR(error);
141 	}
142 
143 	atomic_long_set(&f->f_count, 1);
144 	rwlock_init(&f->f_owner.lock);
145 	spin_lock_init(&f->f_lock);
146 	mutex_init(&f->f_pos_lock);
147 	f->f_flags = flags;
148 	f->f_mode = OPEN_FMODE(flags);
149 	/* f->f_version: 0 */
150 
151 	return f;
152 }
153 
154 /* Find an unused file structure and return a pointer to it.
155  * Returns an error pointer if some error happend e.g. we over file
156  * structures limit, run out of memory or operation is not permitted.
157  *
158  * Be very careful using this.  You are responsible for
159  * getting write access to any mount that you might assign
160  * to this filp, if it is opened for write.  If this is not
161  * done, you will imbalance int the mount's writer count
162  * and a warning at __fput() time.
163  */
164 struct file *alloc_empty_file(int flags, const struct cred *cred)
165 {
166 	static long old_max;
167 	struct file *f;
168 
169 	/*
170 	 * Privileged users can go above max_files
171 	 */
172 	if (get_nr_files() >= files_stat.max_files && !capable(CAP_SYS_ADMIN)) {
173 		/*
174 		 * percpu_counters are inaccurate.  Do an expensive check before
175 		 * we go and fail.
176 		 */
177 		if (percpu_counter_sum_positive(&nr_files) >= files_stat.max_files)
178 			goto over;
179 	}
180 
181 	f = __alloc_file(flags, cred);
182 	if (!IS_ERR(f))
183 		percpu_counter_inc(&nr_files);
184 
185 	return f;
186 
187 over:
188 	/* Ran out of filps - report that */
189 	if (get_nr_files() > old_max) {
190 		pr_info("VFS: file-max limit %lu reached\n", get_max_files());
191 		old_max = get_nr_files();
192 	}
193 	return ERR_PTR(-ENFILE);
194 }
195 
196 /*
197  * Variant of alloc_empty_file() that doesn't check and modify nr_files.
198  *
199  * Should not be used unless there's a very good reason to do so.
200  */
201 struct file *alloc_empty_file_noaccount(int flags, const struct cred *cred)
202 {
203 	struct file *f = __alloc_file(flags, cred);
204 
205 	if (!IS_ERR(f))
206 		f->f_mode |= FMODE_NOACCOUNT;
207 
208 	return f;
209 }
210 
211 /**
212  * alloc_file - allocate and initialize a 'struct file'
213  *
214  * @path: the (dentry, vfsmount) pair for the new file
215  * @flags: O_... flags with which the new file will be opened
216  * @fop: the 'struct file_operations' for the new file
217  */
218 static struct file *alloc_file(const struct path *path, int flags,
219 		const struct file_operations *fop)
220 {
221 	struct file *file;
222 
223 	file = alloc_empty_file(flags, current_cred());
224 	if (IS_ERR(file))
225 		return file;
226 
227 	file->f_path = *path;
228 	file->f_inode = path->dentry->d_inode;
229 	file->f_mapping = path->dentry->d_inode->i_mapping;
230 	file->f_wb_err = filemap_sample_wb_err(file->f_mapping);
231 	file->f_sb_err = file_sample_sb_err(file);
232 	if ((file->f_mode & FMODE_READ) &&
233 	     likely(fop->read || fop->read_iter))
234 		file->f_mode |= FMODE_CAN_READ;
235 	if ((file->f_mode & FMODE_WRITE) &&
236 	     likely(fop->write || fop->write_iter))
237 		file->f_mode |= FMODE_CAN_WRITE;
238 	file->f_mode |= FMODE_OPENED;
239 	file->f_op = fop;
240 	if ((file->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
241 		i_readcount_inc(path->dentry->d_inode);
242 	return file;
243 }
244 
245 struct file *alloc_file_pseudo(struct inode *inode, struct vfsmount *mnt,
246 				const char *name, int flags,
247 				const struct file_operations *fops)
248 {
249 	static const struct dentry_operations anon_ops = {
250 		.d_dname = simple_dname
251 	};
252 	struct qstr this = QSTR_INIT(name, strlen(name));
253 	struct path path;
254 	struct file *file;
255 
256 	path.dentry = d_alloc_pseudo(mnt->mnt_sb, &this);
257 	if (!path.dentry)
258 		return ERR_PTR(-ENOMEM);
259 	if (!mnt->mnt_sb->s_d_op)
260 		d_set_d_op(path.dentry, &anon_ops);
261 	path.mnt = mntget(mnt);
262 	d_instantiate(path.dentry, inode);
263 	file = alloc_file(&path, flags, fops);
264 	if (IS_ERR(file)) {
265 		ihold(inode);
266 		path_put(&path);
267 	}
268 	return file;
269 }
270 EXPORT_SYMBOL(alloc_file_pseudo);
271 
272 struct file *alloc_file_clone(struct file *base, int flags,
273 				const struct file_operations *fops)
274 {
275 	struct file *f = alloc_file(&base->f_path, flags, fops);
276 	if (!IS_ERR(f)) {
277 		path_get(&f->f_path);
278 		f->f_mapping = base->f_mapping;
279 	}
280 	return f;
281 }
282 
283 /* the real guts of fput() - releasing the last reference to file
284  */
285 static void __fput(struct file *file)
286 {
287 	struct dentry *dentry = file->f_path.dentry;
288 	struct vfsmount *mnt = file->f_path.mnt;
289 	struct inode *inode = file->f_inode;
290 	fmode_t mode = file->f_mode;
291 
292 	if (unlikely(!(file->f_mode & FMODE_OPENED)))
293 		goto out;
294 
295 	might_sleep();
296 
297 	fsnotify_close(file);
298 	/*
299 	 * The function eventpoll_release() should be the first called
300 	 * in the file cleanup chain.
301 	 */
302 	eventpoll_release(file);
303 	locks_remove_file(file);
304 
305 	ima_file_free(file);
306 	if (unlikely(file->f_flags & FASYNC)) {
307 		if (file->f_op->fasync)
308 			file->f_op->fasync(-1, file, 0);
309 	}
310 	if (file->f_op->release)
311 		file->f_op->release(inode, file);
312 	if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL &&
313 		     !(mode & FMODE_PATH))) {
314 		cdev_put(inode->i_cdev);
315 	}
316 	fops_put(file->f_op);
317 	put_pid(file->f_owner.pid);
318 	if ((mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
319 		i_readcount_dec(inode);
320 	if (mode & FMODE_WRITER) {
321 		put_write_access(inode);
322 		__mnt_drop_write(mnt);
323 	}
324 	dput(dentry);
325 	if (unlikely(mode & FMODE_NEED_UNMOUNT))
326 		dissolve_on_fput(mnt);
327 	mntput(mnt);
328 out:
329 	file_free(file);
330 }
331 
332 static LLIST_HEAD(delayed_fput_list);
333 static void delayed_fput(struct work_struct *unused)
334 {
335 	struct llist_node *node = llist_del_all(&delayed_fput_list);
336 	struct file *f, *t;
337 
338 	llist_for_each_entry_safe(f, t, node, f_u.fu_llist)
339 		__fput(f);
340 }
341 
342 static void ____fput(struct callback_head *work)
343 {
344 	__fput(container_of(work, struct file, f_u.fu_rcuhead));
345 }
346 
347 /*
348  * If kernel thread really needs to have the final fput() it has done
349  * to complete, call this.  The only user right now is the boot - we
350  * *do* need to make sure our writes to binaries on initramfs has
351  * not left us with opened struct file waiting for __fput() - execve()
352  * won't work without that.  Please, don't add more callers without
353  * very good reasons; in particular, never call that with locks
354  * held and never call that from a thread that might need to do
355  * some work on any kind of umount.
356  */
357 void flush_delayed_fput(void)
358 {
359 	delayed_fput(NULL);
360 }
361 EXPORT_SYMBOL_GPL(flush_delayed_fput);
362 
363 static DECLARE_DELAYED_WORK(delayed_fput_work, delayed_fput);
364 
365 void fput_many(struct file *file, unsigned int refs)
366 {
367 	if (atomic_long_sub_and_test(refs, &file->f_count)) {
368 		struct task_struct *task = current;
369 
370 		if (likely(!in_interrupt() && !(task->flags & PF_KTHREAD))) {
371 			init_task_work(&file->f_u.fu_rcuhead, ____fput);
372 			if (!task_work_add(task, &file->f_u.fu_rcuhead, TWA_RESUME))
373 				return;
374 			/*
375 			 * After this task has run exit_task_work(),
376 			 * task_work_add() will fail.  Fall through to delayed
377 			 * fput to avoid leaking *file.
378 			 */
379 		}
380 
381 		if (llist_add(&file->f_u.fu_llist, &delayed_fput_list))
382 			schedule_delayed_work(&delayed_fput_work, 1);
383 	}
384 }
385 
386 void fput(struct file *file)
387 {
388 	fput_many(file, 1);
389 }
390 
391 /*
392  * synchronous analog of fput(); for kernel threads that might be needed
393  * in some umount() (and thus can't use flush_delayed_fput() without
394  * risking deadlocks), need to wait for completion of __fput() and know
395  * for this specific struct file it won't involve anything that would
396  * need them.  Use only if you really need it - at the very least,
397  * don't blindly convert fput() by kernel thread to that.
398  */
399 void __fput_sync(struct file *file)
400 {
401 	if (atomic_long_dec_and_test(&file->f_count)) {
402 		struct task_struct *task = current;
403 		BUG_ON(!(task->flags & PF_KTHREAD));
404 		__fput(file);
405 	}
406 }
407 
408 EXPORT_SYMBOL(fput);
409 
410 void __init files_init(void)
411 {
412 	filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
413 			SLAB_HWCACHE_ALIGN | SLAB_PANIC | SLAB_ACCOUNT, NULL);
414 	percpu_counter_init(&nr_files, 0, GFP_KERNEL);
415 }
416 
417 /*
418  * One file with associated inode and dcache is very roughly 1K. Per default
419  * do not use more than 10% of our memory for files.
420  */
421 void __init files_maxfiles_init(void)
422 {
423 	unsigned long n;
424 	unsigned long nr_pages = totalram_pages();
425 	unsigned long memreserve = (nr_pages - nr_free_pages()) * 3/2;
426 
427 	memreserve = min(memreserve, nr_pages - 1);
428 	n = ((nr_pages - memreserve) * (PAGE_SIZE / 1024)) / 10;
429 
430 	files_stat.max_files = max_t(unsigned long, n, NR_FILE);
431 }
432