xref: /linux/fs/proc/inode.c (revision cc4589ebfae6f8dbb5cf880a0a67eedab3416492)
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 #include <linux/slab.h>
22 
23 #include <asm/system.h>
24 #include <asm/uaccess.h>
25 
26 #include "internal.h"
27 
28 static void proc_evict_inode(struct inode *inode)
29 {
30 	struct proc_dir_entry *de;
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 	if (PROC_I(inode)->sysctl)
43 		sysctl_head_put(PROC_I(inode)->sysctl);
44 }
45 
46 struct vfsmount *proc_mnt;
47 
48 static struct kmem_cache * proc_inode_cachep;
49 
50 static struct inode *proc_alloc_inode(struct super_block *sb)
51 {
52 	struct proc_inode *ei;
53 	struct inode *inode;
54 
55 	ei = (struct proc_inode *)kmem_cache_alloc(proc_inode_cachep, GFP_KERNEL);
56 	if (!ei)
57 		return NULL;
58 	ei->pid = NULL;
59 	ei->fd = 0;
60 	ei->op.proc_get_link = NULL;
61 	ei->pde = NULL;
62 	ei->sysctl = NULL;
63 	ei->sysctl_entry = NULL;
64 	inode = &ei->vfs_inode;
65 	inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
66 	return inode;
67 }
68 
69 static void proc_destroy_inode(struct inode *inode)
70 {
71 	kmem_cache_free(proc_inode_cachep, PROC_I(inode));
72 }
73 
74 static void init_once(void *foo)
75 {
76 	struct proc_inode *ei = (struct proc_inode *) foo;
77 
78 	inode_init_once(&ei->vfs_inode);
79 }
80 
81 void __init proc_init_inodecache(void)
82 {
83 	proc_inode_cachep = kmem_cache_create("proc_inode_cache",
84 					     sizeof(struct proc_inode),
85 					     0, (SLAB_RECLAIM_ACCOUNT|
86 						SLAB_MEM_SPREAD|SLAB_PANIC),
87 					     init_once);
88 }
89 
90 static const struct super_operations proc_sops = {
91 	.alloc_inode	= proc_alloc_inode,
92 	.destroy_inode	= proc_destroy_inode,
93 	.drop_inode	= generic_delete_inode,
94 	.evict_inode	= proc_evict_inode,
95 	.statfs		= simple_statfs,
96 };
97 
98 static void __pde_users_dec(struct proc_dir_entry *pde)
99 {
100 	pde->pde_users--;
101 	if (pde->pde_unload_completion && pde->pde_users == 0)
102 		complete(pde->pde_unload_completion);
103 }
104 
105 void pde_users_dec(struct proc_dir_entry *pde)
106 {
107 	spin_lock(&pde->pde_unload_lock);
108 	__pde_users_dec(pde);
109 	spin_unlock(&pde->pde_unload_lock);
110 }
111 
112 static loff_t proc_reg_llseek(struct file *file, loff_t offset, int whence)
113 {
114 	struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
115 	loff_t rv = -EINVAL;
116 	loff_t (*llseek)(struct file *, loff_t, int);
117 
118 	spin_lock(&pde->pde_unload_lock);
119 	/*
120 	 * remove_proc_entry() is going to delete PDE (as part of module
121 	 * cleanup sequence). No new callers into module allowed.
122 	 */
123 	if (!pde->proc_fops) {
124 		spin_unlock(&pde->pde_unload_lock);
125 		return rv;
126 	}
127 	/*
128 	 * Bump refcount so that remove_proc_entry will wail for ->llseek to
129 	 * complete.
130 	 */
131 	pde->pde_users++;
132 	/*
133 	 * Save function pointer under lock, to protect against ->proc_fops
134 	 * NULL'ifying right after ->pde_unload_lock is dropped.
135 	 */
136 	llseek = pde->proc_fops->llseek;
137 	spin_unlock(&pde->pde_unload_lock);
138 
139 	if (!llseek)
140 		llseek = default_llseek;
141 	rv = llseek(file, offset, whence);
142 
143 	pde_users_dec(pde);
144 	return rv;
145 }
146 
147 static ssize_t proc_reg_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
148 {
149 	struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
150 	ssize_t rv = -EIO;
151 	ssize_t (*read)(struct file *, char __user *, size_t, loff_t *);
152 
153 	spin_lock(&pde->pde_unload_lock);
154 	if (!pde->proc_fops) {
155 		spin_unlock(&pde->pde_unload_lock);
156 		return rv;
157 	}
158 	pde->pde_users++;
159 	read = pde->proc_fops->read;
160 	spin_unlock(&pde->pde_unload_lock);
161 
162 	if (read)
163 		rv = read(file, buf, count, ppos);
164 
165 	pde_users_dec(pde);
166 	return rv;
167 }
168 
169 static ssize_t proc_reg_write(struct file *file, const 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 (*write)(struct file *, const 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 	write = pde->proc_fops->write;
182 	spin_unlock(&pde->pde_unload_lock);
183 
184 	if (write)
185 		rv = write(file, buf, count, ppos);
186 
187 	pde_users_dec(pde);
188 	return rv;
189 }
190 
191 static unsigned int proc_reg_poll(struct file *file, struct poll_table_struct *pts)
192 {
193 	struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
194 	unsigned int rv = DEFAULT_POLLMASK;
195 	unsigned int (*poll)(struct file *, struct poll_table_struct *);
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 	poll = pde->proc_fops->poll;
204 	spin_unlock(&pde->pde_unload_lock);
205 
206 	if (poll)
207 		rv = poll(file, pts);
208 
209 	pde_users_dec(pde);
210 	return rv;
211 }
212 
213 static long proc_reg_unlocked_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
214 {
215 	struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
216 	long rv = -ENOTTY;
217 	long (*unlocked_ioctl)(struct file *, unsigned int, unsigned long);
218 	int (*ioctl)(struct inode *, struct file *, unsigned int, unsigned long);
219 
220 	spin_lock(&pde->pde_unload_lock);
221 	if (!pde->proc_fops) {
222 		spin_unlock(&pde->pde_unload_lock);
223 		return rv;
224 	}
225 	pde->pde_users++;
226 	unlocked_ioctl = pde->proc_fops->unlocked_ioctl;
227 	ioctl = pde->proc_fops->ioctl;
228 	spin_unlock(&pde->pde_unload_lock);
229 
230 	if (unlocked_ioctl) {
231 		rv = unlocked_ioctl(file, cmd, arg);
232 		if (rv == -ENOIOCTLCMD)
233 			rv = -EINVAL;
234 	} else if (ioctl) {
235 		WARN_ONCE(1, "Procfs ioctl handlers must use unlocked_ioctl, "
236 			  "%pf will be called without the Bkl held\n", ioctl);
237 		rv = ioctl(file->f_path.dentry->d_inode, file, cmd, arg);
238 	}
239 
240 	pde_users_dec(pde);
241 	return rv;
242 }
243 
244 #ifdef CONFIG_COMPAT
245 static long proc_reg_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
246 {
247 	struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
248 	long rv = -ENOTTY;
249 	long (*compat_ioctl)(struct file *, unsigned int, unsigned long);
250 
251 	spin_lock(&pde->pde_unload_lock);
252 	if (!pde->proc_fops) {
253 		spin_unlock(&pde->pde_unload_lock);
254 		return rv;
255 	}
256 	pde->pde_users++;
257 	compat_ioctl = pde->proc_fops->compat_ioctl;
258 	spin_unlock(&pde->pde_unload_lock);
259 
260 	if (compat_ioctl)
261 		rv = compat_ioctl(file, cmd, arg);
262 
263 	pde_users_dec(pde);
264 	return rv;
265 }
266 #endif
267 
268 static int proc_reg_mmap(struct file *file, struct vm_area_struct *vma)
269 {
270 	struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
271 	int rv = -EIO;
272 	int (*mmap)(struct file *, struct vm_area_struct *);
273 
274 	spin_lock(&pde->pde_unload_lock);
275 	if (!pde->proc_fops) {
276 		spin_unlock(&pde->pde_unload_lock);
277 		return rv;
278 	}
279 	pde->pde_users++;
280 	mmap = pde->proc_fops->mmap;
281 	spin_unlock(&pde->pde_unload_lock);
282 
283 	if (mmap)
284 		rv = mmap(file, vma);
285 
286 	pde_users_dec(pde);
287 	return rv;
288 }
289 
290 static int proc_reg_open(struct inode *inode, struct file *file)
291 {
292 	struct proc_dir_entry *pde = PDE(inode);
293 	int rv = 0;
294 	int (*open)(struct inode *, struct file *);
295 	int (*release)(struct inode *, struct file *);
296 	struct pde_opener *pdeo;
297 
298 	/*
299 	 * What for, you ask? Well, we can have open, rmmod, remove_proc_entry
300 	 * sequence. ->release won't be called because ->proc_fops will be
301 	 * cleared. Depending on complexity of ->release, consequences vary.
302 	 *
303 	 * We can't wait for mercy when close will be done for real, it's
304 	 * deadlockable: rmmod foo </proc/foo . So, we're going to do ->release
305 	 * by hand in remove_proc_entry(). For this, save opener's credentials
306 	 * for later.
307 	 */
308 	pdeo = kmalloc(sizeof(struct pde_opener), GFP_KERNEL);
309 	if (!pdeo)
310 		return -ENOMEM;
311 
312 	spin_lock(&pde->pde_unload_lock);
313 	if (!pde->proc_fops) {
314 		spin_unlock(&pde->pde_unload_lock);
315 		kfree(pdeo);
316 		return -EINVAL;
317 	}
318 	pde->pde_users++;
319 	open = pde->proc_fops->open;
320 	release = pde->proc_fops->release;
321 	spin_unlock(&pde->pde_unload_lock);
322 
323 	if (open)
324 		rv = open(inode, file);
325 
326 	spin_lock(&pde->pde_unload_lock);
327 	if (rv == 0 && release) {
328 		/* To know what to release. */
329 		pdeo->inode = inode;
330 		pdeo->file = file;
331 		/* Strictly for "too late" ->release in proc_reg_release(). */
332 		pdeo->release = release;
333 		list_add(&pdeo->lh, &pde->pde_openers);
334 	} else
335 		kfree(pdeo);
336 	__pde_users_dec(pde);
337 	spin_unlock(&pde->pde_unload_lock);
338 	return rv;
339 }
340 
341 static struct pde_opener *find_pde_opener(struct proc_dir_entry *pde,
342 					struct inode *inode, struct file *file)
343 {
344 	struct pde_opener *pdeo;
345 
346 	list_for_each_entry(pdeo, &pde->pde_openers, lh) {
347 		if (pdeo->inode == inode && pdeo->file == file)
348 			return pdeo;
349 	}
350 	return NULL;
351 }
352 
353 static int proc_reg_release(struct inode *inode, struct file *file)
354 {
355 	struct proc_dir_entry *pde = PDE(inode);
356 	int rv = 0;
357 	int (*release)(struct inode *, struct file *);
358 	struct pde_opener *pdeo;
359 
360 	spin_lock(&pde->pde_unload_lock);
361 	pdeo = find_pde_opener(pde, inode, file);
362 	if (!pde->proc_fops) {
363 		/*
364 		 * Can't simply exit, __fput() will think that everything is OK,
365 		 * and move on to freeing struct file. remove_proc_entry() will
366 		 * find slacker in opener's list and will try to do non-trivial
367 		 * things with struct file. Therefore, remove opener from list.
368 		 *
369 		 * But if opener is removed from list, who will ->release it?
370 		 */
371 		if (pdeo) {
372 			list_del(&pdeo->lh);
373 			spin_unlock(&pde->pde_unload_lock);
374 			rv = pdeo->release(inode, file);
375 			kfree(pdeo);
376 		} else
377 			spin_unlock(&pde->pde_unload_lock);
378 		return rv;
379 	}
380 	pde->pde_users++;
381 	release = pde->proc_fops->release;
382 	if (pdeo) {
383 		list_del(&pdeo->lh);
384 		kfree(pdeo);
385 	}
386 	spin_unlock(&pde->pde_unload_lock);
387 
388 	if (release)
389 		rv = release(inode, file);
390 
391 	pde_users_dec(pde);
392 	return rv;
393 }
394 
395 static const struct file_operations proc_reg_file_ops = {
396 	.llseek		= proc_reg_llseek,
397 	.read		= proc_reg_read,
398 	.write		= proc_reg_write,
399 	.poll		= proc_reg_poll,
400 	.unlocked_ioctl	= proc_reg_unlocked_ioctl,
401 #ifdef CONFIG_COMPAT
402 	.compat_ioctl	= proc_reg_compat_ioctl,
403 #endif
404 	.mmap		= proc_reg_mmap,
405 	.open		= proc_reg_open,
406 	.release	= proc_reg_release,
407 };
408 
409 #ifdef CONFIG_COMPAT
410 static const struct file_operations proc_reg_file_ops_no_compat = {
411 	.llseek		= proc_reg_llseek,
412 	.read		= proc_reg_read,
413 	.write		= proc_reg_write,
414 	.poll		= proc_reg_poll,
415 	.unlocked_ioctl	= proc_reg_unlocked_ioctl,
416 	.mmap		= proc_reg_mmap,
417 	.open		= proc_reg_open,
418 	.release	= proc_reg_release,
419 };
420 #endif
421 
422 struct inode *proc_get_inode(struct super_block *sb, unsigned int ino,
423 				struct proc_dir_entry *de)
424 {
425 	struct inode * inode;
426 
427 	inode = iget_locked(sb, 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_INO, &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