xref: /openbmc/linux/fs/file_table.c (revision 9b68f30b) 
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/filelock.h>
17 #include <linux/security.h>
18 #include <linux/cred.h>
19 #include <linux/eventpoll.h>
20 #include <linux/rcupdate.h>
21 #include <linux/mount.h>
22 #include <linux/capability.h>
23 #include <linux/cdev.h>
24 #include <linux/fsnotify.h>
25 #include <linux/sysctl.h>
26 #include <linux/percpu_counter.h>
27 #include <linux/percpu.h>
28 #include <linux/task_work.h>
29 #include <linux/ima.h>
30 #include <linux/swap.h>
31 #include <linux/kmemleak.h>
32 
33 #include <linux/atomic.h>
34 
35 #include "internal.h"
36 
37 /* sysctl tunables... */
38 static struct files_stat_struct files_stat = {
39 	.max_files = NR_FILE
40 };
41 
42 /* SLAB cache for file structures */
43 static struct kmem_cache *filp_cachep __read_mostly;
44 
45 static struct percpu_counter nr_files __cacheline_aligned_in_smp;
46 
47 static void file_free_rcu(struct rcu_head *head)
48 {
49 	struct file *f = container_of(head, struct file, f_rcuhead);
50 
51 	put_cred(f->f_cred);
52 	kmem_cache_free(filp_cachep, f);
53 }
54 
55 static inline void file_free(struct file *f)
56 {
57 	security_file_free(f);
58 	if (!(f->f_mode & FMODE_NOACCOUNT))
59 		percpu_counter_dec(&nr_files);
60 	call_rcu(&f->f_rcuhead, file_free_rcu);
61 }
62 
63 /*
64  * Return the total number of open files in the system
65  */
66 static long get_nr_files(void)
67 {
68 	return percpu_counter_read_positive(&nr_files);
69 }
70 
71 /*
72  * Return the maximum number of open files in the system
73  */
74 unsigned long get_max_files(void)
75 {
76 	return files_stat.max_files;
77 }
78 EXPORT_SYMBOL_GPL(get_max_files);
79 
80 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
81 
82 /*
83  * Handle nr_files sysctl
84  */
85 static int proc_nr_files(struct ctl_table *table, int write, void *buffer,
86 			 size_t *lenp, loff_t *ppos)
87 {
88 	files_stat.nr_files = get_nr_files();
89 	return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
90 }
91 
92 static struct ctl_table fs_stat_sysctls[] = {
93 	{
94 		.procname	= "file-nr",
95 		.data		= &files_stat,
96 		.maxlen		= sizeof(files_stat),
97 		.mode		= 0444,
98 		.proc_handler	= proc_nr_files,
99 	},
100 	{
101 		.procname	= "file-max",
102 		.data		= &files_stat.max_files,
103 		.maxlen		= sizeof(files_stat.max_files),
104 		.mode		= 0644,
105 		.proc_handler	= proc_doulongvec_minmax,
106 		.extra1		= SYSCTL_LONG_ZERO,
107 		.extra2		= SYSCTL_LONG_MAX,
108 	},
109 	{
110 		.procname	= "nr_open",
111 		.data		= &sysctl_nr_open,
112 		.maxlen		= sizeof(unsigned int),
113 		.mode		= 0644,
114 		.proc_handler	= proc_dointvec_minmax,
115 		.extra1		= &sysctl_nr_open_min,
116 		.extra2		= &sysctl_nr_open_max,
117 	},
118 	{ }
119 };
120 
121 static int __init init_fs_stat_sysctls(void)
122 {
123 	register_sysctl_init("fs", fs_stat_sysctls);
124 	if (IS_ENABLED(CONFIG_BINFMT_MISC)) {
125 		struct ctl_table_header *hdr;
126 		hdr = register_sysctl_mount_point("fs/binfmt_misc");
127 		kmemleak_not_leak(hdr);
128 	}
129 	return 0;
130 }
131 fs_initcall(init_fs_stat_sysctls);
132 #endif
133 
134 static struct file *__alloc_file(int flags, const struct cred *cred)
135 {
136 	struct file *f;
137 	int error;
138 
139 	f = kmem_cache_zalloc(filp_cachep, GFP_KERNEL);
140 	if (unlikely(!f))
141 		return ERR_PTR(-ENOMEM);
142 
143 	f->f_cred = get_cred(cred);
144 	error = security_file_alloc(f);
145 	if (unlikely(error)) {
146 		file_free_rcu(&f->f_rcuhead);
147 		return ERR_PTR(error);
148 	}
149 
150 	atomic_long_set(&f->f_count, 1);
151 	rwlock_init(&f->f_owner.lock);
152 	spin_lock_init(&f->f_lock);
153 	mutex_init(&f->f_pos_lock);
154 	f->f_flags = flags;
155 	f->f_mode = OPEN_FMODE(flags);
156 	/* f->f_version: 0 */
157 
158 	return f;
159 }
160 
161 /* Find an unused file structure and return a pointer to it.
162  * Returns an error pointer if some error happend e.g. we over file
163  * structures limit, run out of memory or operation is not permitted.
164  *
165  * Be very careful using this.  You are responsible for
166  * getting write access to any mount that you might assign
167  * to this filp, if it is opened for write.  If this is not
168  * done, you will imbalance int the mount's writer count
169  * and a warning at __fput() time.
170  */
171 struct file *alloc_empty_file(int flags, const struct cred *cred)
172 {
173 	static long old_max;
174 	struct file *f;
175 
176 	/*
177 	 * Privileged users can go above max_files
178 	 */
179 	if (get_nr_files() >= files_stat.max_files && !capable(CAP_SYS_ADMIN)) {
180 		/*
181 		 * percpu_counters are inaccurate.  Do an expensive check before
182 		 * we go and fail.
183 		 */
184 		if (percpu_counter_sum_positive(&nr_files) >= files_stat.max_files)
185 			goto over;
186 	}
187 
188 	f = __alloc_file(flags, cred);
189 	if (!IS_ERR(f))
190 		percpu_counter_inc(&nr_files);
191 
192 	return f;
193 
194 over:
195 	/* Ran out of filps - report that */
196 	if (get_nr_files() > old_max) {
197 		pr_info("VFS: file-max limit %lu reached\n", get_max_files());
198 		old_max = get_nr_files();
199 	}
200 	return ERR_PTR(-ENFILE);
201 }
202 
203 /*
204  * Variant of alloc_empty_file() that doesn't check and modify nr_files.
205  *
206  * Should not be used unless there's a very good reason to do so.
207  */
208 struct file *alloc_empty_file_noaccount(int flags, const struct cred *cred)
209 {
210 	struct file *f = __alloc_file(flags, cred);
211 
212 	if (!IS_ERR(f))
213 		f->f_mode |= FMODE_NOACCOUNT;
214 
215 	return f;
216 }
217 
218 /**
219  * alloc_file - allocate and initialize a 'struct file'
220  *
221  * @path: the (dentry, vfsmount) pair for the new file
222  * @flags: O_... flags with which the new file will be opened
223  * @fop: the 'struct file_operations' for the new file
224  */
225 static struct file *alloc_file(const struct path *path, int flags,
226 		const struct file_operations *fop)
227 {
228 	struct file *file;
229 
230 	file = alloc_empty_file(flags, current_cred());
231 	if (IS_ERR(file))
232 		return file;
233 
234 	file->f_path = *path;
235 	file->f_inode = path->dentry->d_inode;
236 	file->f_mapping = path->dentry->d_inode->i_mapping;
237 	file->f_wb_err = filemap_sample_wb_err(file->f_mapping);
238 	file->f_sb_err = file_sample_sb_err(file);
239 	if (fop->llseek)
240 		file->f_mode |= FMODE_LSEEK;
241 	if ((file->f_mode & FMODE_READ) &&
242 	     likely(fop->read || fop->read_iter))
243 		file->f_mode |= FMODE_CAN_READ;
244 	if ((file->f_mode & FMODE_WRITE) &&
245 	     likely(fop->write || fop->write_iter))
246 		file->f_mode |= FMODE_CAN_WRITE;
247 	file->f_iocb_flags = iocb_flags(file);
248 	file->f_mode |= FMODE_OPENED;
249 	file->f_op = fop;
250 	if ((file->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
251 		i_readcount_inc(path->dentry->d_inode);
252 	return file;
253 }
254 
255 struct file *alloc_file_pseudo(struct inode *inode, struct vfsmount *mnt,
256 				const char *name, int flags,
257 				const struct file_operations *fops)
258 {
259 	static const struct dentry_operations anon_ops = {
260 		.d_dname = simple_dname
261 	};
262 	struct qstr this = QSTR_INIT(name, strlen(name));
263 	struct path path;
264 	struct file *file;
265 
266 	path.dentry = d_alloc_pseudo(mnt->mnt_sb, &this);
267 	if (!path.dentry)
268 		return ERR_PTR(-ENOMEM);
269 	if (!mnt->mnt_sb->s_d_op)
270 		d_set_d_op(path.dentry, &anon_ops);
271 	path.mnt = mntget(mnt);
272 	d_instantiate(path.dentry, inode);
273 	file = alloc_file(&path, flags, fops);
274 	if (IS_ERR(file)) {
275 		ihold(inode);
276 		path_put(&path);
277 	}
278 	return file;
279 }
280 EXPORT_SYMBOL(alloc_file_pseudo);
281 
282 struct file *alloc_file_clone(struct file *base, int flags,
283 				const struct file_operations *fops)
284 {
285 	struct file *f = alloc_file(&base->f_path, flags, fops);
286 	if (!IS_ERR(f)) {
287 		path_get(&f->f_path);
288 		f->f_mapping = base->f_mapping;
289 	}
290 	return f;
291 }
292 
293 /* the real guts of fput() - releasing the last reference to file
294  */
295 static void __fput(struct file *file)
296 {
297 	struct dentry *dentry = file->f_path.dentry;
298 	struct vfsmount *mnt = file->f_path.mnt;
299 	struct inode *inode = file->f_inode;
300 	fmode_t mode = file->f_mode;
301 
302 	if (unlikely(!(file->f_mode & FMODE_OPENED)))
303 		goto out;
304 
305 	might_sleep();
306 
307 	fsnotify_close(file);
308 	/*
309 	 * The function eventpoll_release() should be the first called
310 	 * in the file cleanup chain.
311 	 */
312 	eventpoll_release(file);
313 	locks_remove_file(file);
314 
315 	ima_file_free(file);
316 	if (unlikely(file->f_flags & FASYNC)) {
317 		if (file->f_op->fasync)
318 			file->f_op->fasync(-1, file, 0);
319 	}
320 	if (file->f_op->release)
321 		file->f_op->release(inode, file);
322 	if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL &&
323 		     !(mode & FMODE_PATH))) {
324 		cdev_put(inode->i_cdev);
325 	}
326 	fops_put(file->f_op);
327 	put_pid(file->f_owner.pid);
328 	put_file_access(file);
329 	dput(dentry);
330 	if (unlikely(mode & FMODE_NEED_UNMOUNT))
331 		dissolve_on_fput(mnt);
332 	mntput(mnt);
333 out:
334 	file_free(file);
335 }
336 
337 static LLIST_HEAD(delayed_fput_list);
338 static void delayed_fput(struct work_struct *unused)
339 {
340 	struct llist_node *node = llist_del_all(&delayed_fput_list);
341 	struct file *f, *t;
342 
343 	llist_for_each_entry_safe(f, t, node, f_llist)
344 		__fput(f);
345 }
346 
347 static void ____fput(struct callback_head *work)
348 {
349 	__fput(container_of(work, struct file, f_rcuhead));
350 }
351 
352 /*
353  * If kernel thread really needs to have the final fput() it has done
354  * to complete, call this.  The only user right now is the boot - we
355  * *do* need to make sure our writes to binaries on initramfs has
356  * not left us with opened struct file waiting for __fput() - execve()
357  * won't work without that.  Please, don't add more callers without
358  * very good reasons; in particular, never call that with locks
359  * held and never call that from a thread that might need to do
360  * some work on any kind of umount.
361  */
362 void flush_delayed_fput(void)
363 {
364 	delayed_fput(NULL);
365 }
366 EXPORT_SYMBOL_GPL(flush_delayed_fput);
367 
368 static DECLARE_DELAYED_WORK(delayed_fput_work, delayed_fput);
369 
370 void fput(struct file *file)
371 {
372 	if (atomic_long_dec_and_test(&file->f_count)) {
373 		struct task_struct *task = current;
374 
375 		if (likely(!in_interrupt() && !(task->flags & PF_KTHREAD))) {
376 			init_task_work(&file->f_rcuhead, ____fput);
377 			if (!task_work_add(task, &file->f_rcuhead, TWA_RESUME))
378 				return;
379 			/*
380 			 * After this task has run exit_task_work(),
381 			 * task_work_add() will fail.  Fall through to delayed
382 			 * fput to avoid leaking *file.
383 			 */
384 		}
385 
386 		if (llist_add(&file->f_llist, &delayed_fput_list))
387 			schedule_delayed_work(&delayed_fput_work, 1);
388 	}
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 EXPORT_SYMBOL(__fput_sync);
410 
411 void __init files_init(void)
412 {
413 	filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
414 			SLAB_HWCACHE_ALIGN | SLAB_PANIC | SLAB_ACCOUNT, NULL);
415 	percpu_counter_init(&nr_files, 0, GFP_KERNEL);
416 }
417 
418 /*
419  * One file with associated inode and dcache is very roughly 1K. Per default
420  * do not use more than 10% of our memory for files.
421  */
422 void __init files_maxfiles_init(void)
423 {
424 	unsigned long n;
425 	unsigned long nr_pages = totalram_pages();
426 	unsigned long memreserve = (nr_pages - nr_free_pages()) * 3/2;
427 
428 	memreserve = min(memreserve, nr_pages - 1);
429 	n = ((nr_pages - memreserve) * (PAGE_SIZE / 1024)) / 10;
430 
431 	files_stat.max_files = max_t(unsigned long, n, NR_FILE);
432 }
433