xref: /linux/fs/proc/inode.c (revision 0c93ea4064a209cdc36de8a9a3003d43d08f46f7)
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/smp_lock.h>
20 #include <linux/sysctl.h>
21 
22 #include <asm/system.h>
23 #include <asm/uaccess.h>
24 
25 #include "internal.h"
26 
27 struct proc_dir_entry *de_get(struct proc_dir_entry *de)
28 {
29 	atomic_inc(&de->count);
30 	return de;
31 }
32 
33 /*
34  * Decrements the use count and checks for deferred deletion.
35  */
36 void de_put(struct proc_dir_entry *de)
37 {
38 	if (!atomic_read(&de->count)) {
39 		printk("de_put: entry %s already free!\n", de->name);
40 		return;
41 	}
42 
43 	if (atomic_dec_and_test(&de->count))
44 		free_proc_entry(de);
45 }
46 
47 /*
48  * Decrement the use count of the proc_dir_entry.
49  */
50 static void proc_delete_inode(struct inode *inode)
51 {
52 	struct proc_dir_entry *de;
53 
54 	truncate_inode_pages(&inode->i_data, 0);
55 
56 	/* Stop tracking associated processes */
57 	put_pid(PROC_I(inode)->pid);
58 
59 	/* Let go of any associated proc directory entry */
60 	de = PROC_I(inode)->pde;
61 	if (de) {
62 		if (de->owner)
63 			module_put(de->owner);
64 		de_put(de);
65 	}
66 	if (PROC_I(inode)->sysctl)
67 		sysctl_head_put(PROC_I(inode)->sysctl);
68 	clear_inode(inode);
69 }
70 
71 struct vfsmount *proc_mnt;
72 
73 static struct kmem_cache * proc_inode_cachep;
74 
75 static struct inode *proc_alloc_inode(struct super_block *sb)
76 {
77 	struct proc_inode *ei;
78 	struct inode *inode;
79 
80 	ei = (struct proc_inode *)kmem_cache_alloc(proc_inode_cachep, GFP_KERNEL);
81 	if (!ei)
82 		return NULL;
83 	ei->pid = NULL;
84 	ei->fd = 0;
85 	ei->op.proc_get_link = NULL;
86 	ei->pde = NULL;
87 	ei->sysctl = NULL;
88 	ei->sysctl_entry = NULL;
89 	inode = &ei->vfs_inode;
90 	inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
91 	return inode;
92 }
93 
94 static void proc_destroy_inode(struct inode *inode)
95 {
96 	kmem_cache_free(proc_inode_cachep, PROC_I(inode));
97 }
98 
99 static void init_once(void *foo)
100 {
101 	struct proc_inode *ei = (struct proc_inode *) foo;
102 
103 	inode_init_once(&ei->vfs_inode);
104 }
105 
106 void __init proc_init_inodecache(void)
107 {
108 	proc_inode_cachep = kmem_cache_create("proc_inode_cache",
109 					     sizeof(struct proc_inode),
110 					     0, (SLAB_RECLAIM_ACCOUNT|
111 						SLAB_MEM_SPREAD|SLAB_PANIC),
112 					     init_once);
113 }
114 
115 static const struct super_operations proc_sops = {
116 	.alloc_inode	= proc_alloc_inode,
117 	.destroy_inode	= proc_destroy_inode,
118 	.drop_inode	= generic_delete_inode,
119 	.delete_inode	= proc_delete_inode,
120 	.statfs		= simple_statfs,
121 };
122 
123 static void __pde_users_dec(struct proc_dir_entry *pde)
124 {
125 	pde->pde_users--;
126 	if (pde->pde_unload_completion && pde->pde_users == 0)
127 		complete(pde->pde_unload_completion);
128 }
129 
130 static void pde_users_dec(struct proc_dir_entry *pde)
131 {
132 	spin_lock(&pde->pde_unload_lock);
133 	__pde_users_dec(pde);
134 	spin_unlock(&pde->pde_unload_lock);
135 }
136 
137 static loff_t proc_reg_llseek(struct file *file, loff_t offset, int whence)
138 {
139 	struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
140 	loff_t rv = -EINVAL;
141 	loff_t (*llseek)(struct file *, loff_t, int);
142 
143 	spin_lock(&pde->pde_unload_lock);
144 	/*
145 	 * remove_proc_entry() is going to delete PDE (as part of module
146 	 * cleanup sequence). No new callers into module allowed.
147 	 */
148 	if (!pde->proc_fops) {
149 		spin_unlock(&pde->pde_unload_lock);
150 		return rv;
151 	}
152 	/*
153 	 * Bump refcount so that remove_proc_entry will wail for ->llseek to
154 	 * complete.
155 	 */
156 	pde->pde_users++;
157 	/*
158 	 * Save function pointer under lock, to protect against ->proc_fops
159 	 * NULL'ifying right after ->pde_unload_lock is dropped.
160 	 */
161 	llseek = pde->proc_fops->llseek;
162 	spin_unlock(&pde->pde_unload_lock);
163 
164 	if (!llseek)
165 		llseek = default_llseek;
166 	rv = llseek(file, offset, whence);
167 
168 	pde_users_dec(pde);
169 	return rv;
170 }
171 
172 static ssize_t proc_reg_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
173 {
174 	struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
175 	ssize_t rv = -EIO;
176 	ssize_t (*read)(struct file *, char __user *, size_t, loff_t *);
177 
178 	spin_lock(&pde->pde_unload_lock);
179 	if (!pde->proc_fops) {
180 		spin_unlock(&pde->pde_unload_lock);
181 		return rv;
182 	}
183 	pde->pde_users++;
184 	read = pde->proc_fops->read;
185 	spin_unlock(&pde->pde_unload_lock);
186 
187 	if (read)
188 		rv = read(file, buf, count, ppos);
189 
190 	pde_users_dec(pde);
191 	return rv;
192 }
193 
194 static ssize_t proc_reg_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
195 {
196 	struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
197 	ssize_t rv = -EIO;
198 	ssize_t (*write)(struct file *, const char __user *, size_t, loff_t *);
199 
200 	spin_lock(&pde->pde_unload_lock);
201 	if (!pde->proc_fops) {
202 		spin_unlock(&pde->pde_unload_lock);
203 		return rv;
204 	}
205 	pde->pde_users++;
206 	write = pde->proc_fops->write;
207 	spin_unlock(&pde->pde_unload_lock);
208 
209 	if (write)
210 		rv = write(file, buf, count, ppos);
211 
212 	pde_users_dec(pde);
213 	return rv;
214 }
215 
216 static unsigned int proc_reg_poll(struct file *file, struct poll_table_struct *pts)
217 {
218 	struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
219 	unsigned int rv = DEFAULT_POLLMASK;
220 	unsigned int (*poll)(struct file *, struct poll_table_struct *);
221 
222 	spin_lock(&pde->pde_unload_lock);
223 	if (!pde->proc_fops) {
224 		spin_unlock(&pde->pde_unload_lock);
225 		return rv;
226 	}
227 	pde->pde_users++;
228 	poll = pde->proc_fops->poll;
229 	spin_unlock(&pde->pde_unload_lock);
230 
231 	if (poll)
232 		rv = poll(file, pts);
233 
234 	pde_users_dec(pde);
235 	return rv;
236 }
237 
238 static long proc_reg_unlocked_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
239 {
240 	struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
241 	long rv = -ENOTTY;
242 	long (*unlocked_ioctl)(struct file *, unsigned int, unsigned long);
243 	int (*ioctl)(struct inode *, struct file *, unsigned int, unsigned long);
244 
245 	spin_lock(&pde->pde_unload_lock);
246 	if (!pde->proc_fops) {
247 		spin_unlock(&pde->pde_unload_lock);
248 		return rv;
249 	}
250 	pde->pde_users++;
251 	unlocked_ioctl = pde->proc_fops->unlocked_ioctl;
252 	ioctl = pde->proc_fops->ioctl;
253 	spin_unlock(&pde->pde_unload_lock);
254 
255 	if (unlocked_ioctl) {
256 		rv = unlocked_ioctl(file, cmd, arg);
257 		if (rv == -ENOIOCTLCMD)
258 			rv = -EINVAL;
259 	} else if (ioctl) {
260 		lock_kernel();
261 		rv = ioctl(file->f_path.dentry->d_inode, file, cmd, arg);
262 		unlock_kernel();
263 	}
264 
265 	pde_users_dec(pde);
266 	return rv;
267 }
268 
269 #ifdef CONFIG_COMPAT
270 static long proc_reg_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
271 {
272 	struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
273 	long rv = -ENOTTY;
274 	long (*compat_ioctl)(struct file *, unsigned int, unsigned long);
275 
276 	spin_lock(&pde->pde_unload_lock);
277 	if (!pde->proc_fops) {
278 		spin_unlock(&pde->pde_unload_lock);
279 		return rv;
280 	}
281 	pde->pde_users++;
282 	compat_ioctl = pde->proc_fops->compat_ioctl;
283 	spin_unlock(&pde->pde_unload_lock);
284 
285 	if (compat_ioctl)
286 		rv = compat_ioctl(file, cmd, arg);
287 
288 	pde_users_dec(pde);
289 	return rv;
290 }
291 #endif
292 
293 static int proc_reg_mmap(struct file *file, struct vm_area_struct *vma)
294 {
295 	struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
296 	int rv = -EIO;
297 	int (*mmap)(struct file *, struct vm_area_struct *);
298 
299 	spin_lock(&pde->pde_unload_lock);
300 	if (!pde->proc_fops) {
301 		spin_unlock(&pde->pde_unload_lock);
302 		return rv;
303 	}
304 	pde->pde_users++;
305 	mmap = pde->proc_fops->mmap;
306 	spin_unlock(&pde->pde_unload_lock);
307 
308 	if (mmap)
309 		rv = mmap(file, vma);
310 
311 	pde_users_dec(pde);
312 	return rv;
313 }
314 
315 static int proc_reg_open(struct inode *inode, struct file *file)
316 {
317 	struct proc_dir_entry *pde = PDE(inode);
318 	int rv = 0;
319 	int (*open)(struct inode *, struct file *);
320 	int (*release)(struct inode *, struct file *);
321 	struct pde_opener *pdeo;
322 
323 	/*
324 	 * What for, you ask? Well, we can have open, rmmod, remove_proc_entry
325 	 * sequence. ->release won't be called because ->proc_fops will be
326 	 * cleared. Depending on complexity of ->release, consequences vary.
327 	 *
328 	 * We can't wait for mercy when close will be done for real, it's
329 	 * deadlockable: rmmod foo </proc/foo . So, we're going to do ->release
330 	 * by hand in remove_proc_entry(). For this, save opener's credentials
331 	 * for later.
332 	 */
333 	pdeo = kmalloc(sizeof(struct pde_opener), GFP_KERNEL);
334 	if (!pdeo)
335 		return -ENOMEM;
336 
337 	spin_lock(&pde->pde_unload_lock);
338 	if (!pde->proc_fops) {
339 		spin_unlock(&pde->pde_unload_lock);
340 		kfree(pdeo);
341 		return -EINVAL;
342 	}
343 	pde->pde_users++;
344 	open = pde->proc_fops->open;
345 	release = pde->proc_fops->release;
346 	spin_unlock(&pde->pde_unload_lock);
347 
348 	if (open)
349 		rv = open(inode, file);
350 
351 	spin_lock(&pde->pde_unload_lock);
352 	if (rv == 0 && release) {
353 		/* To know what to release. */
354 		pdeo->inode = inode;
355 		pdeo->file = file;
356 		/* Strictly for "too late" ->release in proc_reg_release(). */
357 		pdeo->release = release;
358 		list_add(&pdeo->lh, &pde->pde_openers);
359 	} else
360 		kfree(pdeo);
361 	__pde_users_dec(pde);
362 	spin_unlock(&pde->pde_unload_lock);
363 	return rv;
364 }
365 
366 static struct pde_opener *find_pde_opener(struct proc_dir_entry *pde,
367 					struct inode *inode, struct file *file)
368 {
369 	struct pde_opener *pdeo;
370 
371 	list_for_each_entry(pdeo, &pde->pde_openers, lh) {
372 		if (pdeo->inode == inode && pdeo->file == file)
373 			return pdeo;
374 	}
375 	return NULL;
376 }
377 
378 static int proc_reg_release(struct inode *inode, struct file *file)
379 {
380 	struct proc_dir_entry *pde = PDE(inode);
381 	int rv = 0;
382 	int (*release)(struct inode *, struct file *);
383 	struct pde_opener *pdeo;
384 
385 	spin_lock(&pde->pde_unload_lock);
386 	pdeo = find_pde_opener(pde, inode, file);
387 	if (!pde->proc_fops) {
388 		/*
389 		 * Can't simply exit, __fput() will think that everything is OK,
390 		 * and move on to freeing struct file. remove_proc_entry() will
391 		 * find slacker in opener's list and will try to do non-trivial
392 		 * things with struct file. Therefore, remove opener from list.
393 		 *
394 		 * But if opener is removed from list, who will ->release it?
395 		 */
396 		if (pdeo) {
397 			list_del(&pdeo->lh);
398 			spin_unlock(&pde->pde_unload_lock);
399 			rv = pdeo->release(inode, file);
400 			kfree(pdeo);
401 		} else
402 			spin_unlock(&pde->pde_unload_lock);
403 		return rv;
404 	}
405 	pde->pde_users++;
406 	release = pde->proc_fops->release;
407 	if (pdeo) {
408 		list_del(&pdeo->lh);
409 		kfree(pdeo);
410 	}
411 	spin_unlock(&pde->pde_unload_lock);
412 
413 	if (release)
414 		rv = release(inode, file);
415 
416 	pde_users_dec(pde);
417 	return rv;
418 }
419 
420 static const struct file_operations proc_reg_file_ops = {
421 	.llseek		= proc_reg_llseek,
422 	.read		= proc_reg_read,
423 	.write		= proc_reg_write,
424 	.poll		= proc_reg_poll,
425 	.unlocked_ioctl	= proc_reg_unlocked_ioctl,
426 #ifdef CONFIG_COMPAT
427 	.compat_ioctl	= proc_reg_compat_ioctl,
428 #endif
429 	.mmap		= proc_reg_mmap,
430 	.open		= proc_reg_open,
431 	.release	= proc_reg_release,
432 };
433 
434 #ifdef CONFIG_COMPAT
435 static const struct file_operations proc_reg_file_ops_no_compat = {
436 	.llseek		= proc_reg_llseek,
437 	.read		= proc_reg_read,
438 	.write		= proc_reg_write,
439 	.poll		= proc_reg_poll,
440 	.unlocked_ioctl	= proc_reg_unlocked_ioctl,
441 	.mmap		= proc_reg_mmap,
442 	.open		= proc_reg_open,
443 	.release	= proc_reg_release,
444 };
445 #endif
446 
447 struct inode *proc_get_inode(struct super_block *sb, unsigned int ino,
448 				struct proc_dir_entry *de)
449 {
450 	struct inode * inode;
451 
452 	if (!try_module_get(de->owner))
453 		goto out_mod;
454 
455 	inode = iget_locked(sb, ino);
456 	if (!inode)
457 		goto out_ino;
458 	if (inode->i_state & I_NEW) {
459 		inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
460 		PROC_I(inode)->fd = 0;
461 		PROC_I(inode)->pde = de;
462 
463 		if (de->mode) {
464 			inode->i_mode = de->mode;
465 			inode->i_uid = de->uid;
466 			inode->i_gid = de->gid;
467 		}
468 		if (de->size)
469 			inode->i_size = de->size;
470 		if (de->nlink)
471 			inode->i_nlink = de->nlink;
472 		if (de->proc_iops)
473 			inode->i_op = de->proc_iops;
474 		if (de->proc_fops) {
475 			if (S_ISREG(inode->i_mode)) {
476 #ifdef CONFIG_COMPAT
477 				if (!de->proc_fops->compat_ioctl)
478 					inode->i_fop =
479 						&proc_reg_file_ops_no_compat;
480 				else
481 #endif
482 					inode->i_fop = &proc_reg_file_ops;
483 			} else {
484 				inode->i_fop = de->proc_fops;
485 			}
486 		}
487 		unlock_new_inode(inode);
488 	} else {
489 	       module_put(de->owner);
490 	       de_put(de);
491 	}
492 	return inode;
493 
494 out_ino:
495 	module_put(de->owner);
496 out_mod:
497 	return NULL;
498 }
499 
500 int proc_fill_super(struct super_block *s)
501 {
502 	struct inode * root_inode;
503 
504 	s->s_flags |= MS_NODIRATIME | MS_NOSUID | MS_NOEXEC;
505 	s->s_blocksize = 1024;
506 	s->s_blocksize_bits = 10;
507 	s->s_magic = PROC_SUPER_MAGIC;
508 	s->s_op = &proc_sops;
509 	s->s_time_gran = 1;
510 
511 	de_get(&proc_root);
512 	root_inode = proc_get_inode(s, PROC_ROOT_INO, &proc_root);
513 	if (!root_inode)
514 		goto out_no_root;
515 	root_inode->i_uid = 0;
516 	root_inode->i_gid = 0;
517 	s->s_root = d_alloc_root(root_inode);
518 	if (!s->s_root)
519 		goto out_no_root;
520 	return 0;
521 
522 out_no_root:
523 	printk("proc_read_super: get root inode failed\n");
524 	iput(root_inode);
525 	de_put(&proc_root);
526 	return -ENOMEM;
527 }
528