xref: /linux/fs/proc/inode.c (revision b8bb76713ec50df2f11efee386e16f93d51e1076)
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 		de_put(de);
63 	if (PROC_I(inode)->sysctl)
64 		sysctl_head_put(PROC_I(inode)->sysctl);
65 	clear_inode(inode);
66 }
67 
68 struct vfsmount *proc_mnt;
69 
70 static struct kmem_cache * proc_inode_cachep;
71 
72 static struct inode *proc_alloc_inode(struct super_block *sb)
73 {
74 	struct proc_inode *ei;
75 	struct inode *inode;
76 
77 	ei = (struct proc_inode *)kmem_cache_alloc(proc_inode_cachep, GFP_KERNEL);
78 	if (!ei)
79 		return NULL;
80 	ei->pid = NULL;
81 	ei->fd = 0;
82 	ei->op.proc_get_link = NULL;
83 	ei->pde = NULL;
84 	ei->sysctl = NULL;
85 	ei->sysctl_entry = NULL;
86 	inode = &ei->vfs_inode;
87 	inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
88 	return inode;
89 }
90 
91 static void proc_destroy_inode(struct inode *inode)
92 {
93 	kmem_cache_free(proc_inode_cachep, PROC_I(inode));
94 }
95 
96 static void init_once(void *foo)
97 {
98 	struct proc_inode *ei = (struct proc_inode *) foo;
99 
100 	inode_init_once(&ei->vfs_inode);
101 }
102 
103 void __init proc_init_inodecache(void)
104 {
105 	proc_inode_cachep = kmem_cache_create("proc_inode_cache",
106 					     sizeof(struct proc_inode),
107 					     0, (SLAB_RECLAIM_ACCOUNT|
108 						SLAB_MEM_SPREAD|SLAB_PANIC),
109 					     init_once);
110 }
111 
112 static const struct super_operations proc_sops = {
113 	.alloc_inode	= proc_alloc_inode,
114 	.destroy_inode	= proc_destroy_inode,
115 	.drop_inode	= generic_delete_inode,
116 	.delete_inode	= proc_delete_inode,
117 	.statfs		= simple_statfs,
118 };
119 
120 static void __pde_users_dec(struct proc_dir_entry *pde)
121 {
122 	pde->pde_users--;
123 	if (pde->pde_unload_completion && pde->pde_users == 0)
124 		complete(pde->pde_unload_completion);
125 }
126 
127 void pde_users_dec(struct proc_dir_entry *pde)
128 {
129 	spin_lock(&pde->pde_unload_lock);
130 	__pde_users_dec(pde);
131 	spin_unlock(&pde->pde_unload_lock);
132 }
133 
134 static loff_t proc_reg_llseek(struct file *file, loff_t offset, int whence)
135 {
136 	struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
137 	loff_t rv = -EINVAL;
138 	loff_t (*llseek)(struct file *, loff_t, int);
139 
140 	spin_lock(&pde->pde_unload_lock);
141 	/*
142 	 * remove_proc_entry() is going to delete PDE (as part of module
143 	 * cleanup sequence). No new callers into module allowed.
144 	 */
145 	if (!pde->proc_fops) {
146 		spin_unlock(&pde->pde_unload_lock);
147 		return rv;
148 	}
149 	/*
150 	 * Bump refcount so that remove_proc_entry will wail for ->llseek to
151 	 * complete.
152 	 */
153 	pde->pde_users++;
154 	/*
155 	 * Save function pointer under lock, to protect against ->proc_fops
156 	 * NULL'ifying right after ->pde_unload_lock is dropped.
157 	 */
158 	llseek = pde->proc_fops->llseek;
159 	spin_unlock(&pde->pde_unload_lock);
160 
161 	if (!llseek)
162 		llseek = default_llseek;
163 	rv = llseek(file, offset, whence);
164 
165 	pde_users_dec(pde);
166 	return rv;
167 }
168 
169 static ssize_t proc_reg_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
170 {
171 	struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
172 	ssize_t rv = -EIO;
173 	ssize_t (*read)(struct file *, char __user *, size_t, loff_t *);
174 
175 	spin_lock(&pde->pde_unload_lock);
176 	if (!pde->proc_fops) {
177 		spin_unlock(&pde->pde_unload_lock);
178 		return rv;
179 	}
180 	pde->pde_users++;
181 	read = pde->proc_fops->read;
182 	spin_unlock(&pde->pde_unload_lock);
183 
184 	if (read)
185 		rv = read(file, buf, count, ppos);
186 
187 	pde_users_dec(pde);
188 	return rv;
189 }
190 
191 static ssize_t proc_reg_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
192 {
193 	struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
194 	ssize_t rv = -EIO;
195 	ssize_t (*write)(struct file *, const char __user *, size_t, loff_t *);
196 
197 	spin_lock(&pde->pde_unload_lock);
198 	if (!pde->proc_fops) {
199 		spin_unlock(&pde->pde_unload_lock);
200 		return rv;
201 	}
202 	pde->pde_users++;
203 	write = pde->proc_fops->write;
204 	spin_unlock(&pde->pde_unload_lock);
205 
206 	if (write)
207 		rv = write(file, buf, count, ppos);
208 
209 	pde_users_dec(pde);
210 	return rv;
211 }
212 
213 static unsigned int proc_reg_poll(struct file *file, struct poll_table_struct *pts)
214 {
215 	struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
216 	unsigned int rv = DEFAULT_POLLMASK;
217 	unsigned int (*poll)(struct file *, struct poll_table_struct *);
218 
219 	spin_lock(&pde->pde_unload_lock);
220 	if (!pde->proc_fops) {
221 		spin_unlock(&pde->pde_unload_lock);
222 		return rv;
223 	}
224 	pde->pde_users++;
225 	poll = pde->proc_fops->poll;
226 	spin_unlock(&pde->pde_unload_lock);
227 
228 	if (poll)
229 		rv = poll(file, pts);
230 
231 	pde_users_dec(pde);
232 	return rv;
233 }
234 
235 static long proc_reg_unlocked_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
236 {
237 	struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
238 	long rv = -ENOTTY;
239 	long (*unlocked_ioctl)(struct file *, unsigned int, unsigned long);
240 	int (*ioctl)(struct inode *, struct file *, unsigned int, unsigned long);
241 
242 	spin_lock(&pde->pde_unload_lock);
243 	if (!pde->proc_fops) {
244 		spin_unlock(&pde->pde_unload_lock);
245 		return rv;
246 	}
247 	pde->pde_users++;
248 	unlocked_ioctl = pde->proc_fops->unlocked_ioctl;
249 	ioctl = pde->proc_fops->ioctl;
250 	spin_unlock(&pde->pde_unload_lock);
251 
252 	if (unlocked_ioctl) {
253 		rv = unlocked_ioctl(file, cmd, arg);
254 		if (rv == -ENOIOCTLCMD)
255 			rv = -EINVAL;
256 	} else if (ioctl) {
257 		lock_kernel();
258 		rv = ioctl(file->f_path.dentry->d_inode, file, cmd, arg);
259 		unlock_kernel();
260 	}
261 
262 	pde_users_dec(pde);
263 	return rv;
264 }
265 
266 #ifdef CONFIG_COMPAT
267 static long proc_reg_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
268 {
269 	struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
270 	long rv = -ENOTTY;
271 	long (*compat_ioctl)(struct file *, unsigned int, unsigned long);
272 
273 	spin_lock(&pde->pde_unload_lock);
274 	if (!pde->proc_fops) {
275 		spin_unlock(&pde->pde_unload_lock);
276 		return rv;
277 	}
278 	pde->pde_users++;
279 	compat_ioctl = pde->proc_fops->compat_ioctl;
280 	spin_unlock(&pde->pde_unload_lock);
281 
282 	if (compat_ioctl)
283 		rv = compat_ioctl(file, cmd, arg);
284 
285 	pde_users_dec(pde);
286 	return rv;
287 }
288 #endif
289 
290 static int proc_reg_mmap(struct file *file, struct vm_area_struct *vma)
291 {
292 	struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
293 	int rv = -EIO;
294 	int (*mmap)(struct file *, struct vm_area_struct *);
295 
296 	spin_lock(&pde->pde_unload_lock);
297 	if (!pde->proc_fops) {
298 		spin_unlock(&pde->pde_unload_lock);
299 		return rv;
300 	}
301 	pde->pde_users++;
302 	mmap = pde->proc_fops->mmap;
303 	spin_unlock(&pde->pde_unload_lock);
304 
305 	if (mmap)
306 		rv = mmap(file, vma);
307 
308 	pde_users_dec(pde);
309 	return rv;
310 }
311 
312 static int proc_reg_open(struct inode *inode, struct file *file)
313 {
314 	struct proc_dir_entry *pde = PDE(inode);
315 	int rv = 0;
316 	int (*open)(struct inode *, struct file *);
317 	int (*release)(struct inode *, struct file *);
318 	struct pde_opener *pdeo;
319 
320 	/*
321 	 * What for, you ask? Well, we can have open, rmmod, remove_proc_entry
322 	 * sequence. ->release won't be called because ->proc_fops will be
323 	 * cleared. Depending on complexity of ->release, consequences vary.
324 	 *
325 	 * We can't wait for mercy when close will be done for real, it's
326 	 * deadlockable: rmmod foo </proc/foo . So, we're going to do ->release
327 	 * by hand in remove_proc_entry(). For this, save opener's credentials
328 	 * for later.
329 	 */
330 	pdeo = kmalloc(sizeof(struct pde_opener), GFP_KERNEL);
331 	if (!pdeo)
332 		return -ENOMEM;
333 
334 	spin_lock(&pde->pde_unload_lock);
335 	if (!pde->proc_fops) {
336 		spin_unlock(&pde->pde_unload_lock);
337 		kfree(pdeo);
338 		return -EINVAL;
339 	}
340 	pde->pde_users++;
341 	open = pde->proc_fops->open;
342 	release = pde->proc_fops->release;
343 	spin_unlock(&pde->pde_unload_lock);
344 
345 	if (open)
346 		rv = open(inode, file);
347 
348 	spin_lock(&pde->pde_unload_lock);
349 	if (rv == 0 && release) {
350 		/* To know what to release. */
351 		pdeo->inode = inode;
352 		pdeo->file = file;
353 		/* Strictly for "too late" ->release in proc_reg_release(). */
354 		pdeo->release = release;
355 		list_add(&pdeo->lh, &pde->pde_openers);
356 	} else
357 		kfree(pdeo);
358 	__pde_users_dec(pde);
359 	spin_unlock(&pde->pde_unload_lock);
360 	return rv;
361 }
362 
363 static struct pde_opener *find_pde_opener(struct proc_dir_entry *pde,
364 					struct inode *inode, struct file *file)
365 {
366 	struct pde_opener *pdeo;
367 
368 	list_for_each_entry(pdeo, &pde->pde_openers, lh) {
369 		if (pdeo->inode == inode && pdeo->file == file)
370 			return pdeo;
371 	}
372 	return NULL;
373 }
374 
375 static int proc_reg_release(struct inode *inode, struct file *file)
376 {
377 	struct proc_dir_entry *pde = PDE(inode);
378 	int rv = 0;
379 	int (*release)(struct inode *, struct file *);
380 	struct pde_opener *pdeo;
381 
382 	spin_lock(&pde->pde_unload_lock);
383 	pdeo = find_pde_opener(pde, inode, file);
384 	if (!pde->proc_fops) {
385 		/*
386 		 * Can't simply exit, __fput() will think that everything is OK,
387 		 * and move on to freeing struct file. remove_proc_entry() will
388 		 * find slacker in opener's list and will try to do non-trivial
389 		 * things with struct file. Therefore, remove opener from list.
390 		 *
391 		 * But if opener is removed from list, who will ->release it?
392 		 */
393 		if (pdeo) {
394 			list_del(&pdeo->lh);
395 			spin_unlock(&pde->pde_unload_lock);
396 			rv = pdeo->release(inode, file);
397 			kfree(pdeo);
398 		} else
399 			spin_unlock(&pde->pde_unload_lock);
400 		return rv;
401 	}
402 	pde->pde_users++;
403 	release = pde->proc_fops->release;
404 	if (pdeo) {
405 		list_del(&pdeo->lh);
406 		kfree(pdeo);
407 	}
408 	spin_unlock(&pde->pde_unload_lock);
409 
410 	if (release)
411 		rv = release(inode, file);
412 
413 	pde_users_dec(pde);
414 	return rv;
415 }
416 
417 static const struct file_operations proc_reg_file_ops = {
418 	.llseek		= proc_reg_llseek,
419 	.read		= proc_reg_read,
420 	.write		= proc_reg_write,
421 	.poll		= proc_reg_poll,
422 	.unlocked_ioctl	= proc_reg_unlocked_ioctl,
423 #ifdef CONFIG_COMPAT
424 	.compat_ioctl	= proc_reg_compat_ioctl,
425 #endif
426 	.mmap		= proc_reg_mmap,
427 	.open		= proc_reg_open,
428 	.release	= proc_reg_release,
429 };
430 
431 #ifdef CONFIG_COMPAT
432 static const struct file_operations proc_reg_file_ops_no_compat = {
433 	.llseek		= proc_reg_llseek,
434 	.read		= proc_reg_read,
435 	.write		= proc_reg_write,
436 	.poll		= proc_reg_poll,
437 	.unlocked_ioctl	= proc_reg_unlocked_ioctl,
438 	.mmap		= proc_reg_mmap,
439 	.open		= proc_reg_open,
440 	.release	= proc_reg_release,
441 };
442 #endif
443 
444 struct inode *proc_get_inode(struct super_block *sb, unsigned int ino,
445 				struct proc_dir_entry *de)
446 {
447 	struct inode * inode;
448 
449 	inode = iget_locked(sb, ino);
450 	if (!inode)
451 		return NULL;
452 	if (inode->i_state & I_NEW) {
453 		inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
454 		PROC_I(inode)->fd = 0;
455 		PROC_I(inode)->pde = de;
456 
457 		if (de->mode) {
458 			inode->i_mode = de->mode;
459 			inode->i_uid = de->uid;
460 			inode->i_gid = de->gid;
461 		}
462 		if (de->size)
463 			inode->i_size = de->size;
464 		if (de->nlink)
465 			inode->i_nlink = de->nlink;
466 		if (de->proc_iops)
467 			inode->i_op = de->proc_iops;
468 		if (de->proc_fops) {
469 			if (S_ISREG(inode->i_mode)) {
470 #ifdef CONFIG_COMPAT
471 				if (!de->proc_fops->compat_ioctl)
472 					inode->i_fop =
473 						&proc_reg_file_ops_no_compat;
474 				else
475 #endif
476 					inode->i_fop = &proc_reg_file_ops;
477 			} else {
478 				inode->i_fop = de->proc_fops;
479 			}
480 		}
481 		unlock_new_inode(inode);
482 	} else
483 	       de_put(de);
484 	return inode;
485 }
486 
487 int proc_fill_super(struct super_block *s)
488 {
489 	struct inode * root_inode;
490 
491 	s->s_flags |= MS_NODIRATIME | MS_NOSUID | MS_NOEXEC;
492 	s->s_blocksize = 1024;
493 	s->s_blocksize_bits = 10;
494 	s->s_magic = PROC_SUPER_MAGIC;
495 	s->s_op = &proc_sops;
496 	s->s_time_gran = 1;
497 
498 	de_get(&proc_root);
499 	root_inode = proc_get_inode(s, PROC_ROOT_INO, &proc_root);
500 	if (!root_inode)
501 		goto out_no_root;
502 	root_inode->i_uid = 0;
503 	root_inode->i_gid = 0;
504 	s->s_root = d_alloc_root(root_inode);
505 	if (!s->s_root)
506 		goto out_no_root;
507 	return 0;
508 
509 out_no_root:
510 	printk("proc_read_super: get root inode failed\n");
511 	iput(root_inode);
512 	de_put(&proc_root);
513 	return -ENOMEM;
514 }
515