xref: /linux/fs/proc/inode.c (revision 64f0962c33d52524deb32d7c34ab8b2c271ee1a3) 
1 /*
2  *  linux/fs/proc/inode.c
3  *
4  *  Copyright (C) 1991, 1992  Linus Torvalds
5  */
6 
7 #include <linux/time.h>
8 #include <linux/proc_fs.h>
9 #include <linux/kernel.h>
10 #include <linux/pid_namespace.h>
11 #include <linux/mm.h>
12 #include <linux/string.h>
13 #include <linux/stat.h>
14 #include <linux/completion.h>
15 #include <linux/poll.h>
16 #include <linux/printk.h>
17 #include <linux/file.h>
18 #include <linux/limits.h>
19 #include <linux/init.h>
20 #include <linux/module.h>
21 #include <linux/sysctl.h>
22 #include <linux/seq_file.h>
23 #include <linux/slab.h>
24 #include <linux/mount.h>
25 
26 #include <asm/uaccess.h>
27 
28 #include "internal.h"
29 
30 static void proc_evict_inode(struct inode *inode)
31 {
32 	struct proc_dir_entry *de;
33 	struct ctl_table_header *head;
34 	const struct proc_ns_operations *ns_ops;
35 	void *ns;
36 
37 	truncate_inode_pages(&inode->i_data, 0);
38 	clear_inode(inode);
39 
40 	/* Stop tracking associated processes */
41 	put_pid(PROC_I(inode)->pid);
42 
43 	/* Let go of any associated proc directory entry */
44 	de = PROC_I(inode)->pde;
45 	if (de)
46 		pde_put(de);
47 	head = PROC_I(inode)->sysctl;
48 	if (head) {
49 		rcu_assign_pointer(PROC_I(inode)->sysctl, NULL);
50 		sysctl_head_put(head);
51 	}
52 	/* Release any associated namespace */
53 	ns_ops = PROC_I(inode)->ns_ops;
54 	ns = PROC_I(inode)->ns;
55 	if (ns_ops && ns)
56 		ns_ops->put(ns);
57 }
58 
59 static struct kmem_cache * proc_inode_cachep;
60 
61 static struct inode *proc_alloc_inode(struct super_block *sb)
62 {
63 	struct proc_inode *ei;
64 	struct inode *inode;
65 
66 	ei = (struct proc_inode *)kmem_cache_alloc(proc_inode_cachep, GFP_KERNEL);
67 	if (!ei)
68 		return NULL;
69 	ei->pid = NULL;
70 	ei->fd = 0;
71 	ei->op.proc_get_link = NULL;
72 	ei->pde = NULL;
73 	ei->sysctl = NULL;
74 	ei->sysctl_entry = NULL;
75 	ei->ns = NULL;
76 	ei->ns_ops = NULL;
77 	inode = &ei->vfs_inode;
78 	inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
79 	return inode;
80 }
81 
82 static void proc_i_callback(struct rcu_head *head)
83 {
84 	struct inode *inode = container_of(head, struct inode, i_rcu);
85 	kmem_cache_free(proc_inode_cachep, PROC_I(inode));
86 }
87 
88 static void proc_destroy_inode(struct inode *inode)
89 {
90 	call_rcu(&inode->i_rcu, proc_i_callback);
91 }
92 
93 static void init_once(void *foo)
94 {
95 	struct proc_inode *ei = (struct proc_inode *) foo;
96 
97 	inode_init_once(&ei->vfs_inode);
98 }
99 
100 void __init proc_init_inodecache(void)
101 {
102 	proc_inode_cachep = kmem_cache_create("proc_inode_cache",
103 					     sizeof(struct proc_inode),
104 					     0, (SLAB_RECLAIM_ACCOUNT|
105 						SLAB_MEM_SPREAD|SLAB_PANIC),
106 					     init_once);
107 }
108 
109 static int proc_show_options(struct seq_file *seq, struct dentry *root)
110 {
111 	struct super_block *sb = root->d_sb;
112 	struct pid_namespace *pid = sb->s_fs_info;
113 
114 	if (!gid_eq(pid->pid_gid, GLOBAL_ROOT_GID))
115 		seq_printf(seq, ",gid=%u", from_kgid_munged(&init_user_ns, pid->pid_gid));
116 	if (pid->hide_pid != 0)
117 		seq_printf(seq, ",hidepid=%u", pid->hide_pid);
118 
119 	return 0;
120 }
121 
122 static const struct super_operations proc_sops = {
123 	.alloc_inode	= proc_alloc_inode,
124 	.destroy_inode	= proc_destroy_inode,
125 	.drop_inode	= generic_delete_inode,
126 	.evict_inode	= proc_evict_inode,
127 	.statfs		= simple_statfs,
128 	.remount_fs	= proc_remount,
129 	.show_options	= proc_show_options,
130 };
131 
132 enum {BIAS = -1U<<31};
133 
134 static inline int use_pde(struct proc_dir_entry *pde)
135 {
136 	return atomic_inc_unless_negative(&pde->in_use);
137 }
138 
139 static void unuse_pde(struct proc_dir_entry *pde)
140 {
141 	if (atomic_dec_return(&pde->in_use) == BIAS)
142 		complete(pde->pde_unload_completion);
143 }
144 
145 /* pde is locked */
146 static void close_pdeo(struct proc_dir_entry *pde, struct pde_opener *pdeo)
147 {
148 	if (pdeo->closing) {
149 		/* somebody else is doing that, just wait */
150 		DECLARE_COMPLETION_ONSTACK(c);
151 		pdeo->c = &c;
152 		spin_unlock(&pde->pde_unload_lock);
153 		wait_for_completion(&c);
154 		spin_lock(&pde->pde_unload_lock);
155 	} else {
156 		struct file *file;
157 		pdeo->closing = 1;
158 		spin_unlock(&pde->pde_unload_lock);
159 		file = pdeo->file;
160 		pde->proc_fops->release(file_inode(file), file);
161 		spin_lock(&pde->pde_unload_lock);
162 		list_del_init(&pdeo->lh);
163 		if (pdeo->c)
164 			complete(pdeo->c);
165 		kfree(pdeo);
166 	}
167 }
168 
169 void proc_entry_rundown(struct proc_dir_entry *de)
170 {
171 	DECLARE_COMPLETION_ONSTACK(c);
172 	/* Wait until all existing callers into module are done. */
173 	de->pde_unload_completion = &c;
174 	if (atomic_add_return(BIAS, &de->in_use) != BIAS)
175 		wait_for_completion(&c);
176 
177 	spin_lock(&de->pde_unload_lock);
178 	while (!list_empty(&de->pde_openers)) {
179 		struct pde_opener *pdeo;
180 		pdeo = list_first_entry(&de->pde_openers, struct pde_opener, lh);
181 		close_pdeo(de, pdeo);
182 	}
183 	spin_unlock(&de->pde_unload_lock);
184 }
185 
186 /* ->read_proc() users - legacy crap */
187 static ssize_t
188 proc_file_read(struct file *file, char __user *buf, size_t nbytes,
189 	       loff_t *ppos)
190 {
191 	struct proc_dir_entry *pde = PDE(file_inode(file));
192 	ssize_t rv = -EIO;
193 	if (use_pde(pde)) {
194 		rv = __proc_file_read(file, buf, nbytes, ppos);
195 		unuse_pde(pde);
196 	}
197 	return rv;
198 }
199 
200 static loff_t
201 proc_file_lseek(struct file *file, loff_t offset, int orig)
202 {
203 	loff_t retval = -EINVAL;
204 	switch (orig) {
205 	case 1:
206 		offset += file->f_pos;
207 	/* fallthrough */
208 	case 0:
209 		if (offset < 0 || offset > MAX_NON_LFS)
210 			break;
211 		file->f_pos = retval = offset;
212 	}
213 	return retval;
214 }
215 
216 const struct file_operations proc_file_operations = {
217 	.llseek		= proc_file_lseek,
218 	.read		= proc_file_read,
219 };
220 
221 static loff_t proc_reg_llseek(struct file *file, loff_t offset, int whence)
222 {
223 	struct proc_dir_entry *pde = PDE(file_inode(file));
224 	loff_t rv = -EINVAL;
225 	if (use_pde(pde)) {
226 		loff_t (*llseek)(struct file *, loff_t, int);
227 		llseek = pde->proc_fops->llseek;
228 		if (!llseek)
229 			llseek = default_llseek;
230 		rv = llseek(file, offset, whence);
231 		unuse_pde(pde);
232 	}
233 	return rv;
234 }
235 
236 static ssize_t proc_reg_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
237 {
238 	ssize_t (*read)(struct file *, char __user *, size_t, loff_t *);
239 	struct proc_dir_entry *pde = PDE(file_inode(file));
240 	ssize_t rv = -EIO;
241 	if (use_pde(pde)) {
242 		read = pde->proc_fops->read;
243 		if (read)
244 			rv = read(file, buf, count, ppos);
245 		unuse_pde(pde);
246 	}
247 	return rv;
248 }
249 
250 static ssize_t proc_reg_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
251 {
252 	ssize_t (*write)(struct file *, const char __user *, size_t, loff_t *);
253 	struct proc_dir_entry *pde = PDE(file_inode(file));
254 	ssize_t rv = -EIO;
255 	if (use_pde(pde)) {
256 		write = pde->proc_fops->write;
257 		if (write)
258 			rv = write(file, buf, count, ppos);
259 		unuse_pde(pde);
260 	}
261 	return rv;
262 }
263 
264 static unsigned int proc_reg_poll(struct file *file, struct poll_table_struct *pts)
265 {
266 	struct proc_dir_entry *pde = PDE(file_inode(file));
267 	unsigned int rv = DEFAULT_POLLMASK;
268 	unsigned int (*poll)(struct file *, struct poll_table_struct *);
269 	if (use_pde(pde)) {
270 		poll = pde->proc_fops->poll;
271 		if (poll)
272 			rv = poll(file, pts);
273 		unuse_pde(pde);
274 	}
275 	return rv;
276 }
277 
278 static long proc_reg_unlocked_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
279 {
280 	struct proc_dir_entry *pde = PDE(file_inode(file));
281 	long rv = -ENOTTY;
282 	long (*ioctl)(struct file *, unsigned int, unsigned long);
283 	if (use_pde(pde)) {
284 		ioctl = pde->proc_fops->unlocked_ioctl;
285 		if (ioctl)
286 			rv = ioctl(file, cmd, arg);
287 		unuse_pde(pde);
288 	}
289 	return rv;
290 }
291 
292 #ifdef CONFIG_COMPAT
293 static long proc_reg_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
294 {
295 	struct proc_dir_entry *pde = PDE(file_inode(file));
296 	long rv = -ENOTTY;
297 	long (*compat_ioctl)(struct file *, unsigned int, unsigned long);
298 	if (use_pde(pde)) {
299 		compat_ioctl = pde->proc_fops->compat_ioctl;
300 		if (compat_ioctl)
301 			rv = compat_ioctl(file, cmd, arg);
302 		unuse_pde(pde);
303 	}
304 	return rv;
305 }
306 #endif
307 
308 static int proc_reg_mmap(struct file *file, struct vm_area_struct *vma)
309 {
310 	struct proc_dir_entry *pde = PDE(file_inode(file));
311 	int rv = -EIO;
312 	int (*mmap)(struct file *, struct vm_area_struct *);
313 	if (use_pde(pde)) {
314 		mmap = pde->proc_fops->mmap;
315 		if (mmap)
316 			rv = mmap(file, vma);
317 		unuse_pde(pde);
318 	}
319 	return rv;
320 }
321 
322 static int proc_reg_open(struct inode *inode, struct file *file)
323 {
324 	struct proc_dir_entry *pde = PDE(inode);
325 	int rv = 0;
326 	int (*open)(struct inode *, struct file *);
327 	int (*release)(struct inode *, struct file *);
328 	struct pde_opener *pdeo;
329 
330 	/*
331 	 * What for, you ask? Well, we can have open, rmmod, remove_proc_entry
332 	 * sequence. ->release won't be called because ->proc_fops will be
333 	 * cleared. Depending on complexity of ->release, consequences vary.
334 	 *
335 	 * We can't wait for mercy when close will be done for real, it's
336 	 * deadlockable: rmmod foo </proc/foo . So, we're going to do ->release
337 	 * by hand in remove_proc_entry(). For this, save opener's credentials
338 	 * for later.
339 	 */
340 	pdeo = kzalloc(sizeof(struct pde_opener), GFP_KERNEL);
341 	if (!pdeo)
342 		return -ENOMEM;
343 
344 	if (!use_pde(pde)) {
345 		kfree(pdeo);
346 		return -ENOENT;
347 	}
348 	open = pde->proc_fops->open;
349 	release = pde->proc_fops->release;
350 
351 	if (open)
352 		rv = open(inode, file);
353 
354 	if (rv == 0 && release) {
355 		/* To know what to release. */
356 		pdeo->file = file;
357 		/* Strictly for "too late" ->release in proc_reg_release(). */
358 		spin_lock(&pde->pde_unload_lock);
359 		list_add(&pdeo->lh, &pde->pde_openers);
360 		spin_unlock(&pde->pde_unload_lock);
361 	} else
362 		kfree(pdeo);
363 
364 	unuse_pde(pde);
365 	return rv;
366 }
367 
368 static int proc_reg_release(struct inode *inode, struct file *file)
369 {
370 	struct proc_dir_entry *pde = PDE(inode);
371 	struct pde_opener *pdeo;
372 	spin_lock(&pde->pde_unload_lock);
373 	list_for_each_entry(pdeo, &pde->pde_openers, lh) {
374 		if (pdeo->file == file) {
375 			close_pdeo(pde, pdeo);
376 			break;
377 		}
378 	}
379 	spin_unlock(&pde->pde_unload_lock);
380 	return 0;
381 }
382 
383 static const struct file_operations proc_reg_file_ops = {
384 	.llseek		= proc_reg_llseek,
385 	.read		= proc_reg_read,
386 	.write		= proc_reg_write,
387 	.poll		= proc_reg_poll,
388 	.unlocked_ioctl	= proc_reg_unlocked_ioctl,
389 #ifdef CONFIG_COMPAT
390 	.compat_ioctl	= proc_reg_compat_ioctl,
391 #endif
392 	.mmap		= proc_reg_mmap,
393 	.open		= proc_reg_open,
394 	.release	= proc_reg_release,
395 };
396 
397 #ifdef CONFIG_COMPAT
398 static const struct file_operations proc_reg_file_ops_no_compat = {
399 	.llseek		= proc_reg_llseek,
400 	.read		= proc_reg_read,
401 	.write		= proc_reg_write,
402 	.poll		= proc_reg_poll,
403 	.unlocked_ioctl	= proc_reg_unlocked_ioctl,
404 	.mmap		= proc_reg_mmap,
405 	.open		= proc_reg_open,
406 	.release	= proc_reg_release,
407 };
408 #endif
409 
410 struct inode *proc_get_inode(struct super_block *sb, struct proc_dir_entry *de)
411 {
412 	struct inode *inode = new_inode_pseudo(sb);
413 
414 	if (inode) {
415 		inode->i_ino = de->low_ino;
416 		inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
417 		PROC_I(inode)->pde = de;
418 
419 		if (de->mode) {
420 			inode->i_mode = de->mode;
421 			inode->i_uid = de->uid;
422 			inode->i_gid = de->gid;
423 		}
424 		if (de->size)
425 			inode->i_size = de->size;
426 		if (de->nlink)
427 			set_nlink(inode, de->nlink);
428 		WARN_ON(!de->proc_iops);
429 		inode->i_op = de->proc_iops;
430 		if (de->proc_fops) {
431 			if (S_ISREG(inode->i_mode)) {
432 #ifdef CONFIG_COMPAT
433 				if (!de->proc_fops->compat_ioctl)
434 					inode->i_fop =
435 						&proc_reg_file_ops_no_compat;
436 				else
437 #endif
438 					inode->i_fop = &proc_reg_file_ops;
439 			} else {
440 				inode->i_fop = de->proc_fops;
441 			}
442 		}
443 	} else
444 	       pde_put(de);
445 	return inode;
446 }
447 
448 int proc_fill_super(struct super_block *s)
449 {
450 	struct inode *root_inode;
451 
452 	s->s_flags |= MS_NODIRATIME | MS_NOSUID | MS_NOEXEC;
453 	s->s_blocksize = 1024;
454 	s->s_blocksize_bits = 10;
455 	s->s_magic = PROC_SUPER_MAGIC;
456 	s->s_op = &proc_sops;
457 	s->s_time_gran = 1;
458 
459 	pde_get(&proc_root);
460 	root_inode = proc_get_inode(s, &proc_root);
461 	if (!root_inode) {
462 		pr_err("proc_fill_super: get root inode failed\n");
463 		return -ENOMEM;
464 	}
465 
466 	s->s_root = d_make_root(root_inode);
467 	if (!s->s_root) {
468 		pr_err("proc_fill_super: allocate dentry failed\n");
469 		return -ENOMEM;
470 	}
471 
472 	return proc_setup_self(s);
473 }
474