xref: /openbmc/linux/fs/file_table.c (revision 33fb626b) 
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 	eventpoll_init_file(f);
117 	f->f_flags = flags;
118 	f->f_mode = OPEN_FMODE(flags);
119 	/* f->f_version: 0 */
120 
121 	return f;
122 }
123 
124 /* Find an unused file structure and return a pointer to it.
125  * Returns an error pointer if some error happend e.g. we over file
126  * structures limit, run out of memory or operation is not permitted.
127  *
128  * Be very careful using this.  You are responsible for
129  * getting write access to any mount that you might assign
130  * to this filp, if it is opened for write.  If this is not
131  * done, you will imbalance int the mount's writer count
132  * and a warning at __fput() time.
133  */
134 struct file *alloc_empty_file(int flags, const struct cred *cred)
135 {
136 	static long old_max;
137 	struct file *f;
138 
139 	/*
140 	 * Privileged users can go above max_files
141 	 */
142 	if (get_nr_files() >= files_stat.max_files && !capable(CAP_SYS_ADMIN)) {
143 		/*
144 		 * percpu_counters are inaccurate.  Do an expensive check before
145 		 * we go and fail.
146 		 */
147 		if (percpu_counter_sum_positive(&nr_files) >= files_stat.max_files)
148 			goto over;
149 	}
150 
151 	f = __alloc_file(flags, cred);
152 	if (!IS_ERR(f))
153 		percpu_counter_inc(&nr_files);
154 
155 	return f;
156 
157 over:
158 	/* Ran out of filps - report that */
159 	if (get_nr_files() > old_max) {
160 		pr_info("VFS: file-max limit %lu reached\n", get_max_files());
161 		old_max = get_nr_files();
162 	}
163 	return ERR_PTR(-ENFILE);
164 }
165 
166 /*
167  * Variant of alloc_empty_file() that doesn't check and modify nr_files.
168  *
169  * Should not be used unless there's a very good reason to do so.
170  */
171 struct file *alloc_empty_file_noaccount(int flags, const struct cred *cred)
172 {
173 	struct file *f = __alloc_file(flags, cred);
174 
175 	if (!IS_ERR(f))
176 		f->f_mode |= FMODE_NOACCOUNT;
177 
178 	return f;
179 }
180 
181 /**
182  * alloc_file - allocate and initialize a 'struct file'
183  *
184  * @path: the (dentry, vfsmount) pair for the new file
185  * @flags: O_... flags with which the new file will be opened
186  * @fop: the 'struct file_operations' for the new file
187  */
188 static struct file *alloc_file(const struct path *path, int flags,
189 		const struct file_operations *fop)
190 {
191 	struct file *file;
192 
193 	file = alloc_empty_file(flags, current_cred());
194 	if (IS_ERR(file))
195 		return file;
196 
197 	file->f_path = *path;
198 	file->f_inode = path->dentry->d_inode;
199 	file->f_mapping = path->dentry->d_inode->i_mapping;
200 	file->f_wb_err = filemap_sample_wb_err(file->f_mapping);
201 	file->f_sb_err = file_sample_sb_err(file);
202 	if ((file->f_mode & FMODE_READ) &&
203 	     likely(fop->read || fop->read_iter))
204 		file->f_mode |= FMODE_CAN_READ;
205 	if ((file->f_mode & FMODE_WRITE) &&
206 	     likely(fop->write || fop->write_iter))
207 		file->f_mode |= FMODE_CAN_WRITE;
208 	file->f_mode |= FMODE_OPENED;
209 	file->f_op = fop;
210 	if ((file->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
211 		i_readcount_inc(path->dentry->d_inode);
212 	return file;
213 }
214 
215 struct file *alloc_file_pseudo(struct inode *inode, struct vfsmount *mnt,
216 				const char *name, int flags,
217 				const struct file_operations *fops)
218 {
219 	static const struct dentry_operations anon_ops = {
220 		.d_dname = simple_dname
221 	};
222 	struct qstr this = QSTR_INIT(name, strlen(name));
223 	struct path path;
224 	struct file *file;
225 
226 	path.dentry = d_alloc_pseudo(mnt->mnt_sb, &this);
227 	if (!path.dentry)
228 		return ERR_PTR(-ENOMEM);
229 	if (!mnt->mnt_sb->s_d_op)
230 		d_set_d_op(path.dentry, &anon_ops);
231 	path.mnt = mntget(mnt);
232 	d_instantiate(path.dentry, inode);
233 	file = alloc_file(&path, flags, fops);
234 	if (IS_ERR(file)) {
235 		ihold(inode);
236 		path_put(&path);
237 	}
238 	return file;
239 }
240 EXPORT_SYMBOL(alloc_file_pseudo);
241 
242 struct file *alloc_file_clone(struct file *base, int flags,
243 				const struct file_operations *fops)
244 {
245 	struct file *f = alloc_file(&base->f_path, flags, fops);
246 	if (!IS_ERR(f)) {
247 		path_get(&f->f_path);
248 		f->f_mapping = base->f_mapping;
249 	}
250 	return f;
251 }
252 
253 /* the real guts of fput() - releasing the last reference to file
254  */
255 static void __fput(struct file *file)
256 {
257 	struct dentry *dentry = file->f_path.dentry;
258 	struct vfsmount *mnt = file->f_path.mnt;
259 	struct inode *inode = file->f_inode;
260 	fmode_t mode = file->f_mode;
261 
262 	if (unlikely(!(file->f_mode & FMODE_OPENED)))
263 		goto out;
264 
265 	might_sleep();
266 
267 	fsnotify_close(file);
268 	/*
269 	 * The function eventpoll_release() should be the first called
270 	 * in the file cleanup chain.
271 	 */
272 	eventpoll_release(file);
273 	locks_remove_file(file);
274 
275 	ima_file_free(file);
276 	if (unlikely(file->f_flags & FASYNC)) {
277 		if (file->f_op->fasync)
278 			file->f_op->fasync(-1, file, 0);
279 	}
280 	if (file->f_op->release)
281 		file->f_op->release(inode, file);
282 	if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL &&
283 		     !(mode & FMODE_PATH))) {
284 		cdev_put(inode->i_cdev);
285 	}
286 	fops_put(file->f_op);
287 	put_pid(file->f_owner.pid);
288 	if ((mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
289 		i_readcount_dec(inode);
290 	if (mode & FMODE_WRITER) {
291 		put_write_access(inode);
292 		__mnt_drop_write(mnt);
293 	}
294 	dput(dentry);
295 	if (unlikely(mode & FMODE_NEED_UNMOUNT))
296 		dissolve_on_fput(mnt);
297 	mntput(mnt);
298 out:
299 	file_free(file);
300 }
301 
302 static LLIST_HEAD(delayed_fput_list);
303 static void delayed_fput(struct work_struct *unused)
304 {
305 	struct llist_node *node = llist_del_all(&delayed_fput_list);
306 	struct file *f, *t;
307 
308 	llist_for_each_entry_safe(f, t, node, f_u.fu_llist)
309 		__fput(f);
310 }
311 
312 static void ____fput(struct callback_head *work)
313 {
314 	__fput(container_of(work, struct file, f_u.fu_rcuhead));
315 }
316 
317 /*
318  * If kernel thread really needs to have the final fput() it has done
319  * to complete, call this.  The only user right now is the boot - we
320  * *do* need to make sure our writes to binaries on initramfs has
321  * not left us with opened struct file waiting for __fput() - execve()
322  * won't work without that.  Please, don't add more callers without
323  * very good reasons; in particular, never call that with locks
324  * held and never call that from a thread that might need to do
325  * some work on any kind of umount.
326  */
327 void flush_delayed_fput(void)
328 {
329 	delayed_fput(NULL);
330 }
331 EXPORT_SYMBOL_GPL(flush_delayed_fput);
332 
333 static DECLARE_DELAYED_WORK(delayed_fput_work, delayed_fput);
334 
335 void fput_many(struct file *file, unsigned int refs)
336 {
337 	if (atomic_long_sub_and_test(refs, &file->f_count)) {
338 		struct task_struct *task = current;
339 
340 		if (likely(!in_interrupt() && !(task->flags & PF_KTHREAD))) {
341 			init_task_work(&file->f_u.fu_rcuhead, ____fput);
342 			if (!task_work_add(task, &file->f_u.fu_rcuhead, TWA_RESUME))
343 				return;
344 			/*
345 			 * After this task has run exit_task_work(),
346 			 * task_work_add() will fail.  Fall through to delayed
347 			 * fput to avoid leaking *file.
348 			 */
349 		}
350 
351 		if (llist_add(&file->f_u.fu_llist, &delayed_fput_list))
352 			schedule_delayed_work(&delayed_fput_work, 1);
353 	}
354 }
355 
356 void fput(struct file *file)
357 {
358 	fput_many(file, 1);
359 }
360 
361 /*
362  * synchronous analog of fput(); for kernel threads that might be needed
363  * in some umount() (and thus can't use flush_delayed_fput() without
364  * risking deadlocks), need to wait for completion of __fput() and know
365  * for this specific struct file it won't involve anything that would
366  * need them.  Use only if you really need it - at the very least,
367  * don't blindly convert fput() by kernel thread to that.
368  */
369 void __fput_sync(struct file *file)
370 {
371 	if (atomic_long_dec_and_test(&file->f_count)) {
372 		struct task_struct *task = current;
373 		BUG_ON(!(task->flags & PF_KTHREAD));
374 		__fput(file);
375 	}
376 }
377 
378 EXPORT_SYMBOL(fput);
379 
380 void __init files_init(void)
381 {
382 	filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
383 			SLAB_HWCACHE_ALIGN | SLAB_PANIC | SLAB_ACCOUNT, NULL);
384 	percpu_counter_init(&nr_files, 0, GFP_KERNEL);
385 }
386 
387 /*
388  * One file with associated inode and dcache is very roughly 1K. Per default
389  * do not use more than 10% of our memory for files.
390  */
391 void __init files_maxfiles_init(void)
392 {
393 	unsigned long n;
394 	unsigned long nr_pages = totalram_pages();
395 	unsigned long memreserve = (nr_pages - nr_free_pages()) * 3/2;
396 
397 	memreserve = min(memreserve, nr_pages - 1);
398 	n = ((nr_pages - memreserve) * (PAGE_SIZE / 1024)) / 10;
399 
400 	files_stat.max_files = max_t(unsigned long, n, NR_FILE);
401 }
402