xref: /openbmc/linux/fs/file_table.c (revision 98ddec80) 
1 /*
2  *  linux/fs/file_table.c
3  *
4  *  Copyright (C) 1991, 1992  Linus Torvalds
5  *  Copyright (C) 1997 David S. Miller (davem@caip.rutgers.edu)
6  */
7 
8 #include <linux/string.h>
9 #include <linux/slab.h>
10 #include <linux/file.h>
11 #include <linux/fdtable.h>
12 #include <linux/init.h>
13 #include <linux/module.h>
14 #include <linux/fs.h>
15 #include <linux/security.h>
16 #include <linux/cred.h>
17 #include <linux/eventpoll.h>
18 #include <linux/rcupdate.h>
19 #include <linux/mount.h>
20 #include <linux/capability.h>
21 #include <linux/cdev.h>
22 #include <linux/fsnotify.h>
23 #include <linux/sysctl.h>
24 #include <linux/percpu_counter.h>
25 #include <linux/percpu.h>
26 #include <linux/task_work.h>
27 #include <linux/ima.h>
28 #include <linux/swap.h>
29 
30 #include <linux/atomic.h>
31 
32 #include "internal.h"
33 
34 /* sysctl tunables... */
35 struct files_stat_struct files_stat = {
36 	.max_files = NR_FILE
37 };
38 
39 /* SLAB cache for file structures */
40 static struct kmem_cache *filp_cachep __read_mostly;
41 
42 static struct percpu_counter nr_files __cacheline_aligned_in_smp;
43 
44 static void file_free_rcu(struct rcu_head *head)
45 {
46 	struct file *f = container_of(head, struct file, f_u.fu_rcuhead);
47 
48 	put_cred(f->f_cred);
49 	kmem_cache_free(filp_cachep, f);
50 }
51 
52 static inline void file_free(struct file *f)
53 {
54 	percpu_counter_dec(&nr_files);
55 	call_rcu(&f->f_u.fu_rcuhead, file_free_rcu);
56 }
57 
58 /*
59  * Return the total number of open files in the system
60  */
61 static long get_nr_files(void)
62 {
63 	return percpu_counter_read_positive(&nr_files);
64 }
65 
66 /*
67  * Return the maximum number of open files in the system
68  */
69 unsigned long get_max_files(void)
70 {
71 	return files_stat.max_files;
72 }
73 EXPORT_SYMBOL_GPL(get_max_files);
74 
75 /*
76  * Handle nr_files sysctl
77  */
78 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
79 int proc_nr_files(struct ctl_table *table, int write,
80                      void __user *buffer, size_t *lenp, loff_t *ppos)
81 {
82 	files_stat.nr_files = get_nr_files();
83 	return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
84 }
85 #else
86 int proc_nr_files(struct ctl_table *table, int write,
87                      void __user *buffer, size_t *lenp, loff_t *ppos)
88 {
89 	return -ENOSYS;
90 }
91 #endif
92 
93 /* Find an unused file structure and return a pointer to it.
94  * Returns an error pointer if some error happend e.g. we over file
95  * structures limit, run out of memory or operation is not permitted.
96  *
97  * Be very careful using this.  You are responsible for
98  * getting write access to any mount that you might assign
99  * to this filp, if it is opened for write.  If this is not
100  * done, you will imbalance int the mount's writer count
101  * and a warning at __fput() time.
102  */
103 struct file *get_empty_filp(void)
104 {
105 	const struct cred *cred = current_cred();
106 	static long old_max;
107 	struct file *f;
108 	int error;
109 
110 	/*
111 	 * Privileged users can go above max_files
112 	 */
113 	if (get_nr_files() >= files_stat.max_files && !capable(CAP_SYS_ADMIN)) {
114 		/*
115 		 * percpu_counters are inaccurate.  Do an expensive check before
116 		 * we go and fail.
117 		 */
118 		if (percpu_counter_sum_positive(&nr_files) >= files_stat.max_files)
119 			goto over;
120 	}
121 
122 	f = kmem_cache_zalloc(filp_cachep, GFP_KERNEL);
123 	if (unlikely(!f))
124 		return ERR_PTR(-ENOMEM);
125 
126 	percpu_counter_inc(&nr_files);
127 	f->f_cred = get_cred(cred);
128 	error = security_file_alloc(f);
129 	if (unlikely(error)) {
130 		file_free(f);
131 		return ERR_PTR(error);
132 	}
133 
134 	atomic_long_set(&f->f_count, 1);
135 	rwlock_init(&f->f_owner.lock);
136 	spin_lock_init(&f->f_lock);
137 	mutex_init(&f->f_pos_lock);
138 	eventpoll_init_file(f);
139 	/* f->f_version: 0 */
140 	return f;
141 
142 over:
143 	/* Ran out of filps - report that */
144 	if (get_nr_files() > old_max) {
145 		pr_info("VFS: file-max limit %lu reached\n", get_max_files());
146 		old_max = get_nr_files();
147 	}
148 	return ERR_PTR(-ENFILE);
149 }
150 
151 /**
152  * alloc_file - allocate and initialize a 'struct file'
153  *
154  * @path: the (dentry, vfsmount) pair for the new file
155  * @mode: the mode with which the new file will be opened
156  * @fop: the 'struct file_operations' for the new file
157  */
158 struct file *alloc_file(const struct path *path, fmode_t mode,
159 		const struct file_operations *fop)
160 {
161 	struct file *file;
162 
163 	file = get_empty_filp();
164 	if (IS_ERR(file))
165 		return file;
166 
167 	file->f_path = *path;
168 	file->f_inode = path->dentry->d_inode;
169 	file->f_mapping = path->dentry->d_inode->i_mapping;
170 	file->f_wb_err = filemap_sample_wb_err(file->f_mapping);
171 	if ((mode & FMODE_READ) &&
172 	     likely(fop->read || fop->read_iter))
173 		mode |= FMODE_CAN_READ;
174 	if ((mode & FMODE_WRITE) &&
175 	     likely(fop->write || fop->write_iter))
176 		mode |= FMODE_CAN_WRITE;
177 	file->f_mode = mode;
178 	file->f_op = fop;
179 	if ((mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
180 		i_readcount_inc(path->dentry->d_inode);
181 	return file;
182 }
183 EXPORT_SYMBOL(alloc_file);
184 
185 /* the real guts of fput() - releasing the last reference to file
186  */
187 static void __fput(struct file *file)
188 {
189 	struct dentry *dentry = file->f_path.dentry;
190 	struct vfsmount *mnt = file->f_path.mnt;
191 	struct inode *inode = file->f_inode;
192 
193 	might_sleep();
194 
195 	fsnotify_close(file);
196 	/*
197 	 * The function eventpoll_release() should be the first called
198 	 * in the file cleanup chain.
199 	 */
200 	eventpoll_release(file);
201 	locks_remove_file(file);
202 
203 	ima_file_free(file);
204 	if (unlikely(file->f_flags & FASYNC)) {
205 		if (file->f_op->fasync)
206 			file->f_op->fasync(-1, file, 0);
207 	}
208 	if (file->f_op->release)
209 		file->f_op->release(inode, file);
210 	security_file_free(file);
211 	if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL &&
212 		     !(file->f_mode & FMODE_PATH))) {
213 		cdev_put(inode->i_cdev);
214 	}
215 	fops_put(file->f_op);
216 	put_pid(file->f_owner.pid);
217 	if ((file->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
218 		i_readcount_dec(inode);
219 	if (file->f_mode & FMODE_WRITER) {
220 		put_write_access(inode);
221 		__mnt_drop_write(mnt);
222 	}
223 	file->f_path.dentry = NULL;
224 	file->f_path.mnt = NULL;
225 	file->f_inode = NULL;
226 	file_free(file);
227 	dput(dentry);
228 	mntput(mnt);
229 }
230 
231 static LLIST_HEAD(delayed_fput_list);
232 static void delayed_fput(struct work_struct *unused)
233 {
234 	struct llist_node *node = llist_del_all(&delayed_fput_list);
235 	struct file *f, *t;
236 
237 	llist_for_each_entry_safe(f, t, node, f_u.fu_llist)
238 		__fput(f);
239 }
240 
241 static void ____fput(struct callback_head *work)
242 {
243 	__fput(container_of(work, struct file, f_u.fu_rcuhead));
244 }
245 
246 /*
247  * If kernel thread really needs to have the final fput() it has done
248  * to complete, call this.  The only user right now is the boot - we
249  * *do* need to make sure our writes to binaries on initramfs has
250  * not left us with opened struct file waiting for __fput() - execve()
251  * won't work without that.  Please, don't add more callers without
252  * very good reasons; in particular, never call that with locks
253  * held and never call that from a thread that might need to do
254  * some work on any kind of umount.
255  */
256 void flush_delayed_fput(void)
257 {
258 	delayed_fput(NULL);
259 }
260 
261 static DECLARE_DELAYED_WORK(delayed_fput_work, delayed_fput);
262 
263 void fput(struct file *file)
264 {
265 	if (atomic_long_dec_and_test(&file->f_count)) {
266 		struct task_struct *task = current;
267 
268 		if (likely(!in_interrupt() && !(task->flags & PF_KTHREAD))) {
269 			init_task_work(&file->f_u.fu_rcuhead, ____fput);
270 			if (!task_work_add(task, &file->f_u.fu_rcuhead, true))
271 				return;
272 			/*
273 			 * After this task has run exit_task_work(),
274 			 * task_work_add() will fail.  Fall through to delayed
275 			 * fput to avoid leaking *file.
276 			 */
277 		}
278 
279 		if (llist_add(&file->f_u.fu_llist, &delayed_fput_list))
280 			schedule_delayed_work(&delayed_fput_work, 1);
281 	}
282 }
283 
284 /*
285  * synchronous analog of fput(); for kernel threads that might be needed
286  * in some umount() (and thus can't use flush_delayed_fput() without
287  * risking deadlocks), need to wait for completion of __fput() and know
288  * for this specific struct file it won't involve anything that would
289  * need them.  Use only if you really need it - at the very least,
290  * don't blindly convert fput() by kernel thread to that.
291  */
292 void __fput_sync(struct file *file)
293 {
294 	if (atomic_long_dec_and_test(&file->f_count)) {
295 		struct task_struct *task = current;
296 		BUG_ON(!(task->flags & PF_KTHREAD));
297 		__fput(file);
298 	}
299 }
300 
301 EXPORT_SYMBOL(fput);
302 
303 void put_filp(struct file *file)
304 {
305 	if (atomic_long_dec_and_test(&file->f_count)) {
306 		security_file_free(file);
307 		file_free(file);
308 	}
309 }
310 
311 void __init files_init(void)
312 {
313 	filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
314 			SLAB_HWCACHE_ALIGN | SLAB_PANIC | SLAB_ACCOUNT, NULL);
315 	percpu_counter_init(&nr_files, 0, GFP_KERNEL);
316 }
317 
318 /*
319  * One file with associated inode and dcache is very roughly 1K. Per default
320  * do not use more than 10% of our memory for files.
321  */
322 void __init files_maxfiles_init(void)
323 {
324 	unsigned long n;
325 	unsigned long memreserve = (totalram_pages - nr_free_pages()) * 3/2;
326 
327 	memreserve = min(memreserve, totalram_pages - 1);
328 	n = ((totalram_pages - memreserve) * (PAGE_SIZE / 1024)) / 10;
329 
330 	files_stat.max_files = max_t(unsigned long, n, NR_FILE);
331 }
332