xref: /linux/fs/proc/inode.c (revision 8ec3b8432e4fe8d452f88f1ed9a3450e715bb797) 
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 
32 	truncate_inode_pages(&inode->i_data, 0);
33 	end_writeback(inode);
34 
35 	/* Stop tracking associated processes */
36 	put_pid(PROC_I(inode)->pid);
37 
38 	/* Let go of any associated proc directory entry */
39 	de = PROC_I(inode)->pde;
40 	if (de)
41 		pde_put(de);
42 	head = PROC_I(inode)->sysctl;
43 	if (head) {
44 		rcu_assign_pointer(PROC_I(inode)->sysctl, NULL);
45 		sysctl_head_put(head);
46 	}
47 }
48 
49 struct vfsmount *proc_mnt;
50 
51 static struct kmem_cache * proc_inode_cachep;
52 
53 static struct inode *proc_alloc_inode(struct super_block *sb)
54 {
55 	struct proc_inode *ei;
56 	struct inode *inode;
57 
58 	ei = (struct proc_inode *)kmem_cache_alloc(proc_inode_cachep, GFP_KERNEL);
59 	if (!ei)
60 		return NULL;
61 	ei->pid = NULL;
62 	ei->fd = 0;
63 	ei->op.proc_get_link = NULL;
64 	ei->pde = NULL;
65 	ei->sysctl = NULL;
66 	ei->sysctl_entry = NULL;
67 	inode = &ei->vfs_inode;
68 	inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
69 	return inode;
70 }
71 
72 static void proc_i_callback(struct rcu_head *head)
73 {
74 	struct inode *inode = container_of(head, struct inode, i_rcu);
75 	INIT_LIST_HEAD(&inode->i_dentry);
76 	kmem_cache_free(proc_inode_cachep, PROC_I(inode));
77 }
78 
79 static void proc_destroy_inode(struct inode *inode)
80 {
81 	call_rcu(&inode->i_rcu, proc_i_callback);
82 }
83 
84 static void init_once(void *foo)
85 {
86 	struct proc_inode *ei = (struct proc_inode *) foo;
87 
88 	inode_init_once(&ei->vfs_inode);
89 }
90 
91 void __init proc_init_inodecache(void)
92 {
93 	proc_inode_cachep = kmem_cache_create("proc_inode_cache",
94 					     sizeof(struct proc_inode),
95 					     0, (SLAB_RECLAIM_ACCOUNT|
96 						SLAB_MEM_SPREAD|SLAB_PANIC),
97 					     init_once);
98 }
99 
100 static const struct super_operations proc_sops = {
101 	.alloc_inode	= proc_alloc_inode,
102 	.destroy_inode	= proc_destroy_inode,
103 	.drop_inode	= generic_delete_inode,
104 	.evict_inode	= proc_evict_inode,
105 	.statfs		= simple_statfs,
106 };
107 
108 static void __pde_users_dec(struct proc_dir_entry *pde)
109 {
110 	pde->pde_users--;
111 	if (pde->pde_unload_completion && pde->pde_users == 0)
112 		complete(pde->pde_unload_completion);
113 }
114 
115 void pde_users_dec(struct proc_dir_entry *pde)
116 {
117 	spin_lock(&pde->pde_unload_lock);
118 	__pde_users_dec(pde);
119 	spin_unlock(&pde->pde_unload_lock);
120 }
121 
122 static loff_t proc_reg_llseek(struct file *file, loff_t offset, int whence)
123 {
124 	struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
125 	loff_t rv = -EINVAL;
126 	loff_t (*llseek)(struct file *, loff_t, int);
127 
128 	spin_lock(&pde->pde_unload_lock);
129 	/*
130 	 * remove_proc_entry() is going to delete PDE (as part of module
131 	 * cleanup sequence). No new callers into module allowed.
132 	 */
133 	if (!pde->proc_fops) {
134 		spin_unlock(&pde->pde_unload_lock);
135 		return rv;
136 	}
137 	/*
138 	 * Bump refcount so that remove_proc_entry will wail for ->llseek to
139 	 * complete.
140 	 */
141 	pde->pde_users++;
142 	/*
143 	 * Save function pointer under lock, to protect against ->proc_fops
144 	 * NULL'ifying right after ->pde_unload_lock is dropped.
145 	 */
146 	llseek = pde->proc_fops->llseek;
147 	spin_unlock(&pde->pde_unload_lock);
148 
149 	if (!llseek)
150 		llseek = default_llseek;
151 	rv = llseek(file, offset, whence);
152 
153 	pde_users_dec(pde);
154 	return rv;
155 }
156 
157 static ssize_t proc_reg_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
158 {
159 	struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
160 	ssize_t rv = -EIO;
161 	ssize_t (*read)(struct file *, char __user *, size_t, loff_t *);
162 
163 	spin_lock(&pde->pde_unload_lock);
164 	if (!pde->proc_fops) {
165 		spin_unlock(&pde->pde_unload_lock);
166 		return rv;
167 	}
168 	pde->pde_users++;
169 	read = pde->proc_fops->read;
170 	spin_unlock(&pde->pde_unload_lock);
171 
172 	if (read)
173 		rv = read(file, buf, count, ppos);
174 
175 	pde_users_dec(pde);
176 	return rv;
177 }
178 
179 static ssize_t proc_reg_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
180 {
181 	struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
182 	ssize_t rv = -EIO;
183 	ssize_t (*write)(struct file *, const char __user *, size_t, loff_t *);
184 
185 	spin_lock(&pde->pde_unload_lock);
186 	if (!pde->proc_fops) {
187 		spin_unlock(&pde->pde_unload_lock);
188 		return rv;
189 	}
190 	pde->pde_users++;
191 	write = pde->proc_fops->write;
192 	spin_unlock(&pde->pde_unload_lock);
193 
194 	if (write)
195 		rv = write(file, buf, count, ppos);
196 
197 	pde_users_dec(pde);
198 	return rv;
199 }
200 
201 static unsigned int proc_reg_poll(struct file *file, struct poll_table_struct *pts)
202 {
203 	struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
204 	unsigned int rv = DEFAULT_POLLMASK;
205 	unsigned int (*poll)(struct file *, struct poll_table_struct *);
206 
207 	spin_lock(&pde->pde_unload_lock);
208 	if (!pde->proc_fops) {
209 		spin_unlock(&pde->pde_unload_lock);
210 		return rv;
211 	}
212 	pde->pde_users++;
213 	poll = pde->proc_fops->poll;
214 	spin_unlock(&pde->pde_unload_lock);
215 
216 	if (poll)
217 		rv = poll(file, pts);
218 
219 	pde_users_dec(pde);
220 	return rv;
221 }
222 
223 static long proc_reg_unlocked_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
224 {
225 	struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
226 	long rv = -ENOTTY;
227 	long (*ioctl)(struct file *, unsigned int, unsigned long);
228 
229 	spin_lock(&pde->pde_unload_lock);
230 	if (!pde->proc_fops) {
231 		spin_unlock(&pde->pde_unload_lock);
232 		return rv;
233 	}
234 	pde->pde_users++;
235 	ioctl = pde->proc_fops->unlocked_ioctl;
236 	spin_unlock(&pde->pde_unload_lock);
237 
238 	if (ioctl)
239 		rv = ioctl(file, cmd, arg);
240 
241 	pde_users_dec(pde);
242 	return rv;
243 }
244 
245 #ifdef CONFIG_COMPAT
246 static long proc_reg_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
247 {
248 	struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
249 	long rv = -ENOTTY;
250 	long (*compat_ioctl)(struct file *, unsigned int, unsigned long);
251 
252 	spin_lock(&pde->pde_unload_lock);
253 	if (!pde->proc_fops) {
254 		spin_unlock(&pde->pde_unload_lock);
255 		return rv;
256 	}
257 	pde->pde_users++;
258 	compat_ioctl = pde->proc_fops->compat_ioctl;
259 	spin_unlock(&pde->pde_unload_lock);
260 
261 	if (compat_ioctl)
262 		rv = compat_ioctl(file, cmd, arg);
263 
264 	pde_users_dec(pde);
265 	return rv;
266 }
267 #endif
268 
269 static int proc_reg_mmap(struct file *file, struct vm_area_struct *vma)
270 {
271 	struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
272 	int rv = -EIO;
273 	int (*mmap)(struct file *, struct vm_area_struct *);
274 
275 	spin_lock(&pde->pde_unload_lock);
276 	if (!pde->proc_fops) {
277 		spin_unlock(&pde->pde_unload_lock);
278 		return rv;
279 	}
280 	pde->pde_users++;
281 	mmap = pde->proc_fops->mmap;
282 	spin_unlock(&pde->pde_unload_lock);
283 
284 	if (mmap)
285 		rv = mmap(file, vma);
286 
287 	pde_users_dec(pde);
288 	return rv;
289 }
290 
291 static int proc_reg_open(struct inode *inode, struct file *file)
292 {
293 	struct proc_dir_entry *pde = PDE(inode);
294 	int rv = 0;
295 	int (*open)(struct inode *, struct file *);
296 	int (*release)(struct inode *, struct file *);
297 	struct pde_opener *pdeo;
298 
299 	/*
300 	 * What for, you ask? Well, we can have open, rmmod, remove_proc_entry
301 	 * sequence. ->release won't be called because ->proc_fops will be
302 	 * cleared. Depending on complexity of ->release, consequences vary.
303 	 *
304 	 * We can't wait for mercy when close will be done for real, it's
305 	 * deadlockable: rmmod foo </proc/foo . So, we're going to do ->release
306 	 * by hand in remove_proc_entry(). For this, save opener's credentials
307 	 * for later.
308 	 */
309 	pdeo = kmalloc(sizeof(struct pde_opener), GFP_KERNEL);
310 	if (!pdeo)
311 		return -ENOMEM;
312 
313 	spin_lock(&pde->pde_unload_lock);
314 	if (!pde->proc_fops) {
315 		spin_unlock(&pde->pde_unload_lock);
316 		kfree(pdeo);
317 		return -EINVAL;
318 	}
319 	pde->pde_users++;
320 	open = pde->proc_fops->open;
321 	release = pde->proc_fops->release;
322 	spin_unlock(&pde->pde_unload_lock);
323 
324 	if (open)
325 		rv = open(inode, file);
326 
327 	spin_lock(&pde->pde_unload_lock);
328 	if (rv == 0 && release) {
329 		/* To know what to release. */
330 		pdeo->inode = inode;
331 		pdeo->file = file;
332 		/* Strictly for "too late" ->release in proc_reg_release(). */
333 		pdeo->release = release;
334 		list_add(&pdeo->lh, &pde->pde_openers);
335 	} else
336 		kfree(pdeo);
337 	__pde_users_dec(pde);
338 	spin_unlock(&pde->pde_unload_lock);
339 	return rv;
340 }
341 
342 static struct pde_opener *find_pde_opener(struct proc_dir_entry *pde,
343 					struct inode *inode, struct file *file)
344 {
345 	struct pde_opener *pdeo;
346 
347 	list_for_each_entry(pdeo, &pde->pde_openers, lh) {
348 		if (pdeo->inode == inode && pdeo->file == file)
349 			return pdeo;
350 	}
351 	return NULL;
352 }
353 
354 static int proc_reg_release(struct inode *inode, struct file *file)
355 {
356 	struct proc_dir_entry *pde = PDE(inode);
357 	int rv = 0;
358 	int (*release)(struct inode *, struct file *);
359 	struct pde_opener *pdeo;
360 
361 	spin_lock(&pde->pde_unload_lock);
362 	pdeo = find_pde_opener(pde, inode, file);
363 	if (!pde->proc_fops) {
364 		/*
365 		 * Can't simply exit, __fput() will think that everything is OK,
366 		 * and move on to freeing struct file. remove_proc_entry() will
367 		 * find slacker in opener's list and will try to do non-trivial
368 		 * things with struct file. Therefore, remove opener from list.
369 		 *
370 		 * But if opener is removed from list, who will ->release it?
371 		 */
372 		if (pdeo) {
373 			list_del(&pdeo->lh);
374 			spin_unlock(&pde->pde_unload_lock);
375 			rv = pdeo->release(inode, file);
376 			kfree(pdeo);
377 		} else
378 			spin_unlock(&pde->pde_unload_lock);
379 		return rv;
380 	}
381 	pde->pde_users++;
382 	release = pde->proc_fops->release;
383 	if (pdeo) {
384 		list_del(&pdeo->lh);
385 		kfree(pdeo);
386 	}
387 	spin_unlock(&pde->pde_unload_lock);
388 
389 	if (release)
390 		rv = release(inode, file);
391 
392 	pde_users_dec(pde);
393 	return rv;
394 }
395 
396 static const struct file_operations proc_reg_file_ops = {
397 	.llseek		= proc_reg_llseek,
398 	.read		= proc_reg_read,
399 	.write		= proc_reg_write,
400 	.poll		= proc_reg_poll,
401 	.unlocked_ioctl	= proc_reg_unlocked_ioctl,
402 #ifdef CONFIG_COMPAT
403 	.compat_ioctl	= proc_reg_compat_ioctl,
404 #endif
405 	.mmap		= proc_reg_mmap,
406 	.open		= proc_reg_open,
407 	.release	= proc_reg_release,
408 };
409 
410 #ifdef CONFIG_COMPAT
411 static const struct file_operations proc_reg_file_ops_no_compat = {
412 	.llseek		= proc_reg_llseek,
413 	.read		= proc_reg_read,
414 	.write		= proc_reg_write,
415 	.poll		= proc_reg_poll,
416 	.unlocked_ioctl	= proc_reg_unlocked_ioctl,
417 	.mmap		= proc_reg_mmap,
418 	.open		= proc_reg_open,
419 	.release	= proc_reg_release,
420 };
421 #endif
422 
423 struct inode *proc_get_inode(struct super_block *sb, struct proc_dir_entry *de)
424 {
425 	struct inode * inode;
426 
427 	inode = iget_locked(sb, de->low_ino);
428 	if (!inode)
429 		return NULL;
430 	if (inode->i_state & I_NEW) {
431 		inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
432 		PROC_I(inode)->fd = 0;
433 		PROC_I(inode)->pde = de;
434 
435 		if (de->mode) {
436 			inode->i_mode = de->mode;
437 			inode->i_uid = de->uid;
438 			inode->i_gid = de->gid;
439 		}
440 		if (de->size)
441 			inode->i_size = de->size;
442 		if (de->nlink)
443 			inode->i_nlink = de->nlink;
444 		if (de->proc_iops)
445 			inode->i_op = de->proc_iops;
446 		if (de->proc_fops) {
447 			if (S_ISREG(inode->i_mode)) {
448 #ifdef CONFIG_COMPAT
449 				if (!de->proc_fops->compat_ioctl)
450 					inode->i_fop =
451 						&proc_reg_file_ops_no_compat;
452 				else
453 #endif
454 					inode->i_fop = &proc_reg_file_ops;
455 			} else {
456 				inode->i_fop = de->proc_fops;
457 			}
458 		}
459 		unlock_new_inode(inode);
460 	} else
461 	       pde_put(de);
462 	return inode;
463 }
464 
465 int proc_fill_super(struct super_block *s)
466 {
467 	struct inode * root_inode;
468 
469 	s->s_flags |= MS_NODIRATIME | MS_NOSUID | MS_NOEXEC;
470 	s->s_blocksize = 1024;
471 	s->s_blocksize_bits = 10;
472 	s->s_magic = PROC_SUPER_MAGIC;
473 	s->s_op = &proc_sops;
474 	s->s_time_gran = 1;
475 
476 	pde_get(&proc_root);
477 	root_inode = proc_get_inode(s, &proc_root);
478 	if (!root_inode)
479 		goto out_no_root;
480 	root_inode->i_uid = 0;
481 	root_inode->i_gid = 0;
482 	s->s_root = d_alloc_root(root_inode);
483 	if (!s->s_root)
484 		goto out_no_root;
485 	return 0;
486 
487 out_no_root:
488 	printk("proc_read_super: get root inode failed\n");
489 	iput(root_inode);
490 	pde_put(&proc_root);
491 	return -ENOMEM;
492 }
493