xref: /linux/fs/proc/inode.c (revision 9d14070f656addddce3d63fd483de46930b51850)
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/mm.h>
11 #include <linux/string.h>
12 #include <linux/stat.h>
13 #include <linux/completion.h>
14 #include <linux/poll.h>
15 #include <linux/file.h>
16 #include <linux/limits.h>
17 #include <linux/init.h>
18 #include <linux/module.h>
19 #include <linux/sysctl.h>
20 #include <linux/slab.h>
21 
22 #include <asm/system.h>
23 #include <asm/uaccess.h>
24 
25 #include "internal.h"
26 
27 static void proc_evict_inode(struct inode *inode)
28 {
29 	struct proc_dir_entry *de;
30 	struct ctl_table_header *head;
31 	const struct proc_ns_operations *ns_ops;
32 
33 	truncate_inode_pages(&inode->i_data, 0);
34 	end_writeback(inode);
35 
36 	/* Stop tracking associated processes */
37 	put_pid(PROC_I(inode)->pid);
38 
39 	/* Let go of any associated proc directory entry */
40 	de = PROC_I(inode)->pde;
41 	if (de)
42 		pde_put(de);
43 	head = PROC_I(inode)->sysctl;
44 	if (head) {
45 		rcu_assign_pointer(PROC_I(inode)->sysctl, NULL);
46 		sysctl_head_put(head);
47 	}
48 	/* Release any associated namespace */
49 	ns_ops = PROC_I(inode)->ns_ops;
50 	if (ns_ops && ns_ops->put)
51 		ns_ops->put(PROC_I(inode)->ns);
52 }
53 
54 static struct kmem_cache * proc_inode_cachep;
55 
56 static struct inode *proc_alloc_inode(struct super_block *sb)
57 {
58 	struct proc_inode *ei;
59 	struct inode *inode;
60 
61 	ei = (struct proc_inode *)kmem_cache_alloc(proc_inode_cachep, GFP_KERNEL);
62 	if (!ei)
63 		return NULL;
64 	ei->pid = NULL;
65 	ei->fd = 0;
66 	ei->op.proc_get_link = NULL;
67 	ei->pde = NULL;
68 	ei->sysctl = NULL;
69 	ei->sysctl_entry = NULL;
70 	ei->ns = NULL;
71 	ei->ns_ops = NULL;
72 	inode = &ei->vfs_inode;
73 	inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
74 	return inode;
75 }
76 
77 static void proc_i_callback(struct rcu_head *head)
78 {
79 	struct inode *inode = container_of(head, struct inode, i_rcu);
80 	kmem_cache_free(proc_inode_cachep, PROC_I(inode));
81 }
82 
83 static void proc_destroy_inode(struct inode *inode)
84 {
85 	call_rcu(&inode->i_rcu, proc_i_callback);
86 }
87 
88 static void init_once(void *foo)
89 {
90 	struct proc_inode *ei = (struct proc_inode *) foo;
91 
92 	inode_init_once(&ei->vfs_inode);
93 }
94 
95 void __init proc_init_inodecache(void)
96 {
97 	proc_inode_cachep = kmem_cache_create("proc_inode_cache",
98 					     sizeof(struct proc_inode),
99 					     0, (SLAB_RECLAIM_ACCOUNT|
100 						SLAB_MEM_SPREAD|SLAB_PANIC),
101 					     init_once);
102 }
103 
104 static const struct super_operations proc_sops = {
105 	.alloc_inode	= proc_alloc_inode,
106 	.destroy_inode	= proc_destroy_inode,
107 	.drop_inode	= generic_delete_inode,
108 	.evict_inode	= proc_evict_inode,
109 	.statfs		= simple_statfs,
110 };
111 
112 static void __pde_users_dec(struct proc_dir_entry *pde)
113 {
114 	pde->pde_users--;
115 	if (pde->pde_unload_completion && pde->pde_users == 0)
116 		complete(pde->pde_unload_completion);
117 }
118 
119 void pde_users_dec(struct proc_dir_entry *pde)
120 {
121 	spin_lock(&pde->pde_unload_lock);
122 	__pde_users_dec(pde);
123 	spin_unlock(&pde->pde_unload_lock);
124 }
125 
126 static loff_t proc_reg_llseek(struct file *file, loff_t offset, int whence)
127 {
128 	struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
129 	loff_t rv = -EINVAL;
130 	loff_t (*llseek)(struct file *, loff_t, int);
131 
132 	spin_lock(&pde->pde_unload_lock);
133 	/*
134 	 * remove_proc_entry() is going to delete PDE (as part of module
135 	 * cleanup sequence). No new callers into module allowed.
136 	 */
137 	if (!pde->proc_fops) {
138 		spin_unlock(&pde->pde_unload_lock);
139 		return rv;
140 	}
141 	/*
142 	 * Bump refcount so that remove_proc_entry will wail for ->llseek to
143 	 * complete.
144 	 */
145 	pde->pde_users++;
146 	/*
147 	 * Save function pointer under lock, to protect against ->proc_fops
148 	 * NULL'ifying right after ->pde_unload_lock is dropped.
149 	 */
150 	llseek = pde->proc_fops->llseek;
151 	spin_unlock(&pde->pde_unload_lock);
152 
153 	if (!llseek)
154 		llseek = default_llseek;
155 	rv = llseek(file, offset, whence);
156 
157 	pde_users_dec(pde);
158 	return rv;
159 }
160 
161 static ssize_t proc_reg_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
162 {
163 	struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
164 	ssize_t rv = -EIO;
165 	ssize_t (*read)(struct file *, char __user *, size_t, loff_t *);
166 
167 	spin_lock(&pde->pde_unload_lock);
168 	if (!pde->proc_fops) {
169 		spin_unlock(&pde->pde_unload_lock);
170 		return rv;
171 	}
172 	pde->pde_users++;
173 	read = pde->proc_fops->read;
174 	spin_unlock(&pde->pde_unload_lock);
175 
176 	if (read)
177 		rv = read(file, buf, count, ppos);
178 
179 	pde_users_dec(pde);
180 	return rv;
181 }
182 
183 static ssize_t proc_reg_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
184 {
185 	struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
186 	ssize_t rv = -EIO;
187 	ssize_t (*write)(struct file *, const char __user *, size_t, loff_t *);
188 
189 	spin_lock(&pde->pde_unload_lock);
190 	if (!pde->proc_fops) {
191 		spin_unlock(&pde->pde_unload_lock);
192 		return rv;
193 	}
194 	pde->pde_users++;
195 	write = pde->proc_fops->write;
196 	spin_unlock(&pde->pde_unload_lock);
197 
198 	if (write)
199 		rv = write(file, buf, count, ppos);
200 
201 	pde_users_dec(pde);
202 	return rv;
203 }
204 
205 static unsigned int proc_reg_poll(struct file *file, struct poll_table_struct *pts)
206 {
207 	struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
208 	unsigned int rv = DEFAULT_POLLMASK;
209 	unsigned int (*poll)(struct file *, struct poll_table_struct *);
210 
211 	spin_lock(&pde->pde_unload_lock);
212 	if (!pde->proc_fops) {
213 		spin_unlock(&pde->pde_unload_lock);
214 		return rv;
215 	}
216 	pde->pde_users++;
217 	poll = pde->proc_fops->poll;
218 	spin_unlock(&pde->pde_unload_lock);
219 
220 	if (poll)
221 		rv = poll(file, pts);
222 
223 	pde_users_dec(pde);
224 	return rv;
225 }
226 
227 static long proc_reg_unlocked_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
228 {
229 	struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
230 	long rv = -ENOTTY;
231 	long (*ioctl)(struct file *, unsigned int, unsigned long);
232 
233 	spin_lock(&pde->pde_unload_lock);
234 	if (!pde->proc_fops) {
235 		spin_unlock(&pde->pde_unload_lock);
236 		return rv;
237 	}
238 	pde->pde_users++;
239 	ioctl = pde->proc_fops->unlocked_ioctl;
240 	spin_unlock(&pde->pde_unload_lock);
241 
242 	if (ioctl)
243 		rv = ioctl(file, cmd, arg);
244 
245 	pde_users_dec(pde);
246 	return rv;
247 }
248 
249 #ifdef CONFIG_COMPAT
250 static long proc_reg_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
251 {
252 	struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
253 	long rv = -ENOTTY;
254 	long (*compat_ioctl)(struct file *, unsigned int, unsigned long);
255 
256 	spin_lock(&pde->pde_unload_lock);
257 	if (!pde->proc_fops) {
258 		spin_unlock(&pde->pde_unload_lock);
259 		return rv;
260 	}
261 	pde->pde_users++;
262 	compat_ioctl = pde->proc_fops->compat_ioctl;
263 	spin_unlock(&pde->pde_unload_lock);
264 
265 	if (compat_ioctl)
266 		rv = compat_ioctl(file, cmd, arg);
267 
268 	pde_users_dec(pde);
269 	return rv;
270 }
271 #endif
272 
273 static int proc_reg_mmap(struct file *file, struct vm_area_struct *vma)
274 {
275 	struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
276 	int rv = -EIO;
277 	int (*mmap)(struct file *, struct vm_area_struct *);
278 
279 	spin_lock(&pde->pde_unload_lock);
280 	if (!pde->proc_fops) {
281 		spin_unlock(&pde->pde_unload_lock);
282 		return rv;
283 	}
284 	pde->pde_users++;
285 	mmap = pde->proc_fops->mmap;
286 	spin_unlock(&pde->pde_unload_lock);
287 
288 	if (mmap)
289 		rv = mmap(file, vma);
290 
291 	pde_users_dec(pde);
292 	return rv;
293 }
294 
295 static int proc_reg_open(struct inode *inode, struct file *file)
296 {
297 	struct proc_dir_entry *pde = PDE(inode);
298 	int rv = 0;
299 	int (*open)(struct inode *, struct file *);
300 	int (*release)(struct inode *, struct file *);
301 	struct pde_opener *pdeo;
302 
303 	/*
304 	 * What for, you ask? Well, we can have open, rmmod, remove_proc_entry
305 	 * sequence. ->release won't be called because ->proc_fops will be
306 	 * cleared. Depending on complexity of ->release, consequences vary.
307 	 *
308 	 * We can't wait for mercy when close will be done for real, it's
309 	 * deadlockable: rmmod foo </proc/foo . So, we're going to do ->release
310 	 * by hand in remove_proc_entry(). For this, save opener's credentials
311 	 * for later.
312 	 */
313 	pdeo = kmalloc(sizeof(struct pde_opener), GFP_KERNEL);
314 	if (!pdeo)
315 		return -ENOMEM;
316 
317 	spin_lock(&pde->pde_unload_lock);
318 	if (!pde->proc_fops) {
319 		spin_unlock(&pde->pde_unload_lock);
320 		kfree(pdeo);
321 		return -ENOENT;
322 	}
323 	pde->pde_users++;
324 	open = pde->proc_fops->open;
325 	release = pde->proc_fops->release;
326 	spin_unlock(&pde->pde_unload_lock);
327 
328 	if (open)
329 		rv = open(inode, file);
330 
331 	spin_lock(&pde->pde_unload_lock);
332 	if (rv == 0 && release) {
333 		/* To know what to release. */
334 		pdeo->inode = inode;
335 		pdeo->file = file;
336 		/* Strictly for "too late" ->release in proc_reg_release(). */
337 		pdeo->release = release;
338 		list_add(&pdeo->lh, &pde->pde_openers);
339 	} else
340 		kfree(pdeo);
341 	__pde_users_dec(pde);
342 	spin_unlock(&pde->pde_unload_lock);
343 	return rv;
344 }
345 
346 static struct pde_opener *find_pde_opener(struct proc_dir_entry *pde,
347 					struct inode *inode, struct file *file)
348 {
349 	struct pde_opener *pdeo;
350 
351 	list_for_each_entry(pdeo, &pde->pde_openers, lh) {
352 		if (pdeo->inode == inode && pdeo->file == file)
353 			return pdeo;
354 	}
355 	return NULL;
356 }
357 
358 static int proc_reg_release(struct inode *inode, struct file *file)
359 {
360 	struct proc_dir_entry *pde = PDE(inode);
361 	int rv = 0;
362 	int (*release)(struct inode *, struct file *);
363 	struct pde_opener *pdeo;
364 
365 	spin_lock(&pde->pde_unload_lock);
366 	pdeo = find_pde_opener(pde, inode, file);
367 	if (!pde->proc_fops) {
368 		/*
369 		 * Can't simply exit, __fput() will think that everything is OK,
370 		 * and move on to freeing struct file. remove_proc_entry() will
371 		 * find slacker in opener's list and will try to do non-trivial
372 		 * things with struct file. Therefore, remove opener from list.
373 		 *
374 		 * But if opener is removed from list, who will ->release it?
375 		 */
376 		if (pdeo) {
377 			list_del(&pdeo->lh);
378 			spin_unlock(&pde->pde_unload_lock);
379 			rv = pdeo->release(inode, file);
380 			kfree(pdeo);
381 		} else
382 			spin_unlock(&pde->pde_unload_lock);
383 		return rv;
384 	}
385 	pde->pde_users++;
386 	release = pde->proc_fops->release;
387 	if (pdeo) {
388 		list_del(&pdeo->lh);
389 		kfree(pdeo);
390 	}
391 	spin_unlock(&pde->pde_unload_lock);
392 
393 	if (release)
394 		rv = release(inode, file);
395 
396 	pde_users_dec(pde);
397 	return rv;
398 }
399 
400 static const struct file_operations proc_reg_file_ops = {
401 	.llseek		= proc_reg_llseek,
402 	.read		= proc_reg_read,
403 	.write		= proc_reg_write,
404 	.poll		= proc_reg_poll,
405 	.unlocked_ioctl	= proc_reg_unlocked_ioctl,
406 #ifdef CONFIG_COMPAT
407 	.compat_ioctl	= proc_reg_compat_ioctl,
408 #endif
409 	.mmap		= proc_reg_mmap,
410 	.open		= proc_reg_open,
411 	.release	= proc_reg_release,
412 };
413 
414 #ifdef CONFIG_COMPAT
415 static const struct file_operations proc_reg_file_ops_no_compat = {
416 	.llseek		= proc_reg_llseek,
417 	.read		= proc_reg_read,
418 	.write		= proc_reg_write,
419 	.poll		= proc_reg_poll,
420 	.unlocked_ioctl	= proc_reg_unlocked_ioctl,
421 	.mmap		= proc_reg_mmap,
422 	.open		= proc_reg_open,
423 	.release	= proc_reg_release,
424 };
425 #endif
426 
427 struct inode *proc_get_inode(struct super_block *sb, struct proc_dir_entry *de)
428 {
429 	struct inode * inode;
430 
431 	inode = iget_locked(sb, de->low_ino);
432 	if (!inode)
433 		return NULL;
434 	if (inode->i_state & I_NEW) {
435 		inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
436 		PROC_I(inode)->fd = 0;
437 		PROC_I(inode)->pde = de;
438 
439 		if (de->mode) {
440 			inode->i_mode = de->mode;
441 			inode->i_uid = de->uid;
442 			inode->i_gid = de->gid;
443 		}
444 		if (de->size)
445 			inode->i_size = de->size;
446 		if (de->nlink)
447 			set_nlink(inode, de->nlink);
448 		if (de->proc_iops)
449 			inode->i_op = de->proc_iops;
450 		if (de->proc_fops) {
451 			if (S_ISREG(inode->i_mode)) {
452 #ifdef CONFIG_COMPAT
453 				if (!de->proc_fops->compat_ioctl)
454 					inode->i_fop =
455 						&proc_reg_file_ops_no_compat;
456 				else
457 #endif
458 					inode->i_fop = &proc_reg_file_ops;
459 			} else {
460 				inode->i_fop = de->proc_fops;
461 			}
462 		}
463 		unlock_new_inode(inode);
464 	} else
465 	       pde_put(de);
466 	return inode;
467 }
468 
469 int proc_fill_super(struct super_block *s)
470 {
471 	struct inode * root_inode;
472 
473 	s->s_flags |= MS_NODIRATIME | MS_NOSUID | MS_NOEXEC;
474 	s->s_blocksize = 1024;
475 	s->s_blocksize_bits = 10;
476 	s->s_magic = PROC_SUPER_MAGIC;
477 	s->s_op = &proc_sops;
478 	s->s_time_gran = 1;
479 
480 	pde_get(&proc_root);
481 	root_inode = proc_get_inode(s, &proc_root);
482 	if (!root_inode)
483 		goto out_no_root;
484 	root_inode->i_uid = 0;
485 	root_inode->i_gid = 0;
486 	s->s_root = d_alloc_root(root_inode);
487 	if (!s->s_root)
488 		goto out_no_root;
489 	return 0;
490 
491 out_no_root:
492 	printk("proc_read_super: get root inode failed\n");
493 	iput(root_inode);
494 	pde_put(&proc_root);
495 	return -ENOMEM;
496 }
497