xref: /openbmc/linux/fs/proc/generic.c (revision b6dcefde)
1 /*
2  * proc/fs/generic.c --- generic routines for the proc-fs
3  *
4  * This file contains generic proc-fs routines for handling
5  * directories and files.
6  *
7  * Copyright (C) 1991, 1992 Linus Torvalds.
8  * Copyright (C) 1997 Theodore Ts'o
9  */
10 
11 #include <linux/errno.h>
12 #include <linux/time.h>
13 #include <linux/proc_fs.h>
14 #include <linux/stat.h>
15 #include <linux/module.h>
16 #include <linux/mount.h>
17 #include <linux/init.h>
18 #include <linux/idr.h>
19 #include <linux/namei.h>
20 #include <linux/bitops.h>
21 #include <linux/spinlock.h>
22 #include <linux/completion.h>
23 #include <asm/uaccess.h>
24 
25 #include "internal.h"
26 
27 DEFINE_SPINLOCK(proc_subdir_lock);
28 
29 static int proc_match(int len, const char *name, struct proc_dir_entry *de)
30 {
31 	if (de->namelen != len)
32 		return 0;
33 	return !memcmp(name, de->name, len);
34 }
35 
36 /* buffer size is one page but our output routines use some slack for overruns */
37 #define PROC_BLOCK_SIZE	(PAGE_SIZE - 1024)
38 
39 static ssize_t
40 __proc_file_read(struct file *file, char __user *buf, size_t nbytes,
41 	       loff_t *ppos)
42 {
43 	struct inode * inode = file->f_path.dentry->d_inode;
44 	char 	*page;
45 	ssize_t	retval=0;
46 	int	eof=0;
47 	ssize_t	n, count;
48 	char	*start;
49 	struct proc_dir_entry * dp;
50 	unsigned long long pos;
51 
52 	/*
53 	 * Gaah, please just use "seq_file" instead. The legacy /proc
54 	 * interfaces cut loff_t down to off_t for reads, and ignore
55 	 * the offset entirely for writes..
56 	 */
57 	pos = *ppos;
58 	if (pos > MAX_NON_LFS)
59 		return 0;
60 	if (nbytes > MAX_NON_LFS - pos)
61 		nbytes = MAX_NON_LFS - pos;
62 
63 	dp = PDE(inode);
64 	if (!(page = (char*) __get_free_page(GFP_TEMPORARY)))
65 		return -ENOMEM;
66 
67 	while ((nbytes > 0) && !eof) {
68 		count = min_t(size_t, PROC_BLOCK_SIZE, nbytes);
69 
70 		start = NULL;
71 		if (dp->read_proc) {
72 			/*
73 			 * How to be a proc read function
74 			 * ------------------------------
75 			 * Prototype:
76 			 *    int f(char *buffer, char **start, off_t offset,
77 			 *          int count, int *peof, void *dat)
78 			 *
79 			 * Assume that the buffer is "count" bytes in size.
80 			 *
81 			 * If you know you have supplied all the data you
82 			 * have, set *peof.
83 			 *
84 			 * You have three ways to return data:
85 			 * 0) Leave *start = NULL.  (This is the default.)
86 			 *    Put the data of the requested offset at that
87 			 *    offset within the buffer.  Return the number (n)
88 			 *    of bytes there are from the beginning of the
89 			 *    buffer up to the last byte of data.  If the
90 			 *    number of supplied bytes (= n - offset) is
91 			 *    greater than zero and you didn't signal eof
92 			 *    and the reader is prepared to take more data
93 			 *    you will be called again with the requested
94 			 *    offset advanced by the number of bytes
95 			 *    absorbed.  This interface is useful for files
96 			 *    no larger than the buffer.
97 			 * 1) Set *start = an unsigned long value less than
98 			 *    the buffer address but greater than zero.
99 			 *    Put the data of the requested offset at the
100 			 *    beginning of the buffer.  Return the number of
101 			 *    bytes of data placed there.  If this number is
102 			 *    greater than zero and you didn't signal eof
103 			 *    and the reader is prepared to take more data
104 			 *    you will be called again with the requested
105 			 *    offset advanced by *start.  This interface is
106 			 *    useful when you have a large file consisting
107 			 *    of a series of blocks which you want to count
108 			 *    and return as wholes.
109 			 *    (Hack by Paul.Russell@rustcorp.com.au)
110 			 * 2) Set *start = an address within the buffer.
111 			 *    Put the data of the requested offset at *start.
112 			 *    Return the number of bytes of data placed there.
113 			 *    If this number is greater than zero and you
114 			 *    didn't signal eof and the reader is prepared to
115 			 *    take more data you will be called again with the
116 			 *    requested offset advanced by the number of bytes
117 			 *    absorbed.
118 			 */
119 			n = dp->read_proc(page, &start, *ppos,
120 					  count, &eof, dp->data);
121 		} else
122 			break;
123 
124 		if (n == 0)   /* end of file */
125 			break;
126 		if (n < 0) {  /* error */
127 			if (retval == 0)
128 				retval = n;
129 			break;
130 		}
131 
132 		if (start == NULL) {
133 			if (n > PAGE_SIZE) {
134 				printk(KERN_ERR
135 				       "proc_file_read: Apparent buffer overflow!\n");
136 				n = PAGE_SIZE;
137 			}
138 			n -= *ppos;
139 			if (n <= 0)
140 				break;
141 			if (n > count)
142 				n = count;
143 			start = page + *ppos;
144 		} else if (start < page) {
145 			if (n > PAGE_SIZE) {
146 				printk(KERN_ERR
147 				       "proc_file_read: Apparent buffer overflow!\n");
148 				n = PAGE_SIZE;
149 			}
150 			if (n > count) {
151 				/*
152 				 * Don't reduce n because doing so might
153 				 * cut off part of a data block.
154 				 */
155 				printk(KERN_WARNING
156 				       "proc_file_read: Read count exceeded\n");
157 			}
158 		} else /* start >= page */ {
159 			unsigned long startoff = (unsigned long)(start - page);
160 			if (n > (PAGE_SIZE - startoff)) {
161 				printk(KERN_ERR
162 				       "proc_file_read: Apparent buffer overflow!\n");
163 				n = PAGE_SIZE - startoff;
164 			}
165 			if (n > count)
166 				n = count;
167 		}
168 
169  		n -= copy_to_user(buf, start < page ? page : start, n);
170 		if (n == 0) {
171 			if (retval == 0)
172 				retval = -EFAULT;
173 			break;
174 		}
175 
176 		*ppos += start < page ? (unsigned long)start : n;
177 		nbytes -= n;
178 		buf += n;
179 		retval += n;
180 	}
181 	free_page((unsigned long) page);
182 	return retval;
183 }
184 
185 static ssize_t
186 proc_file_read(struct file *file, char __user *buf, size_t nbytes,
187 	       loff_t *ppos)
188 {
189 	struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
190 	ssize_t rv = -EIO;
191 
192 	spin_lock(&pde->pde_unload_lock);
193 	if (!pde->proc_fops) {
194 		spin_unlock(&pde->pde_unload_lock);
195 		return rv;
196 	}
197 	pde->pde_users++;
198 	spin_unlock(&pde->pde_unload_lock);
199 
200 	rv = __proc_file_read(file, buf, nbytes, ppos);
201 
202 	pde_users_dec(pde);
203 	return rv;
204 }
205 
206 static ssize_t
207 proc_file_write(struct file *file, const char __user *buffer,
208 		size_t count, loff_t *ppos)
209 {
210 	struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
211 	ssize_t rv = -EIO;
212 
213 	if (pde->write_proc) {
214 		spin_lock(&pde->pde_unload_lock);
215 		if (!pde->proc_fops) {
216 			spin_unlock(&pde->pde_unload_lock);
217 			return rv;
218 		}
219 		pde->pde_users++;
220 		spin_unlock(&pde->pde_unload_lock);
221 
222 		/* FIXME: does this routine need ppos?  probably... */
223 		rv = pde->write_proc(file, buffer, count, pde->data);
224 		pde_users_dec(pde);
225 	}
226 	return rv;
227 }
228 
229 
230 static loff_t
231 proc_file_lseek(struct file *file, loff_t offset, int orig)
232 {
233 	loff_t retval = -EINVAL;
234 	switch (orig) {
235 	case 1:
236 		offset += file->f_pos;
237 	/* fallthrough */
238 	case 0:
239 		if (offset < 0 || offset > MAX_NON_LFS)
240 			break;
241 		file->f_pos = retval = offset;
242 	}
243 	return retval;
244 }
245 
246 static const struct file_operations proc_file_operations = {
247 	.llseek		= proc_file_lseek,
248 	.read		= proc_file_read,
249 	.write		= proc_file_write,
250 };
251 
252 static int proc_notify_change(struct dentry *dentry, struct iattr *iattr)
253 {
254 	struct inode *inode = dentry->d_inode;
255 	struct proc_dir_entry *de = PDE(inode);
256 	int error;
257 
258 	error = inode_change_ok(inode, iattr);
259 	if (error)
260 		goto out;
261 
262 	error = inode_setattr(inode, iattr);
263 	if (error)
264 		goto out;
265 
266 	de->uid = inode->i_uid;
267 	de->gid = inode->i_gid;
268 	de->mode = inode->i_mode;
269 out:
270 	return error;
271 }
272 
273 static int proc_getattr(struct vfsmount *mnt, struct dentry *dentry,
274 			struct kstat *stat)
275 {
276 	struct inode *inode = dentry->d_inode;
277 	struct proc_dir_entry *de = PROC_I(inode)->pde;
278 	if (de && de->nlink)
279 		inode->i_nlink = de->nlink;
280 
281 	generic_fillattr(inode, stat);
282 	return 0;
283 }
284 
285 static const struct inode_operations proc_file_inode_operations = {
286 	.setattr	= proc_notify_change,
287 };
288 
289 /*
290  * This function parses a name such as "tty/driver/serial", and
291  * returns the struct proc_dir_entry for "/proc/tty/driver", and
292  * returns "serial" in residual.
293  */
294 static int xlate_proc_name(const char *name,
295 			   struct proc_dir_entry **ret, const char **residual)
296 {
297 	const char     		*cp = name, *next;
298 	struct proc_dir_entry	*de;
299 	int			len;
300 	int 			rtn = 0;
301 
302 	de = *ret;
303 	if (!de)
304 		de = &proc_root;
305 
306 	spin_lock(&proc_subdir_lock);
307 	while (1) {
308 		next = strchr(cp, '/');
309 		if (!next)
310 			break;
311 
312 		len = next - cp;
313 		for (de = de->subdir; de ; de = de->next) {
314 			if (proc_match(len, cp, de))
315 				break;
316 		}
317 		if (!de) {
318 			rtn = -ENOENT;
319 			goto out;
320 		}
321 		cp += len + 1;
322 	}
323 	*residual = cp;
324 	*ret = de;
325 out:
326 	spin_unlock(&proc_subdir_lock);
327 	return rtn;
328 }
329 
330 static DEFINE_IDA(proc_inum_ida);
331 static DEFINE_SPINLOCK(proc_inum_lock); /* protects the above */
332 
333 #define PROC_DYNAMIC_FIRST 0xF0000000U
334 
335 /*
336  * Return an inode number between PROC_DYNAMIC_FIRST and
337  * 0xffffffff, or zero on failure.
338  *
339  * Current inode allocations in the proc-fs (hex-numbers):
340  *
341  * 00000000		reserved
342  * 00000001-00000fff	static entries	(goners)
343  *      001		root-ino
344  *
345  * 00001000-00001fff	unused
346  * 0001xxxx-7fffxxxx	pid-dir entries for pid 1-7fff
347  * 80000000-efffffff	unused
348  * f0000000-ffffffff	dynamic entries
349  *
350  * Goal:
351  *	Once we split the thing into several virtual filesystems,
352  *	we will get rid of magical ranges (and this comment, BTW).
353  */
354 static unsigned int get_inode_number(void)
355 {
356 	unsigned int i;
357 	int error;
358 
359 retry:
360 	if (ida_pre_get(&proc_inum_ida, GFP_KERNEL) == 0)
361 		return 0;
362 
363 	spin_lock(&proc_inum_lock);
364 	error = ida_get_new(&proc_inum_ida, &i);
365 	spin_unlock(&proc_inum_lock);
366 	if (error == -EAGAIN)
367 		goto retry;
368 	else if (error)
369 		return 0;
370 
371 	if (i > UINT_MAX - PROC_DYNAMIC_FIRST) {
372 		spin_lock(&proc_inum_lock);
373 		ida_remove(&proc_inum_ida, i);
374 		spin_unlock(&proc_inum_lock);
375 		return 0;
376 	}
377 	return PROC_DYNAMIC_FIRST + i;
378 }
379 
380 static void release_inode_number(unsigned int inum)
381 {
382 	spin_lock(&proc_inum_lock);
383 	ida_remove(&proc_inum_ida, inum - PROC_DYNAMIC_FIRST);
384 	spin_unlock(&proc_inum_lock);
385 }
386 
387 static void *proc_follow_link(struct dentry *dentry, struct nameidata *nd)
388 {
389 	nd_set_link(nd, PDE(dentry->d_inode)->data);
390 	return NULL;
391 }
392 
393 static const struct inode_operations proc_link_inode_operations = {
394 	.readlink	= generic_readlink,
395 	.follow_link	= proc_follow_link,
396 };
397 
398 /*
399  * As some entries in /proc are volatile, we want to
400  * get rid of unused dentries.  This could be made
401  * smarter: we could keep a "volatile" flag in the
402  * inode to indicate which ones to keep.
403  */
404 static int proc_delete_dentry(struct dentry * dentry)
405 {
406 	return 1;
407 }
408 
409 static const struct dentry_operations proc_dentry_operations =
410 {
411 	.d_delete	= proc_delete_dentry,
412 };
413 
414 /*
415  * Don't create negative dentries here, return -ENOENT by hand
416  * instead.
417  */
418 struct dentry *proc_lookup_de(struct proc_dir_entry *de, struct inode *dir,
419 		struct dentry *dentry)
420 {
421 	struct inode *inode = NULL;
422 	int error = -ENOENT;
423 
424 	spin_lock(&proc_subdir_lock);
425 	for (de = de->subdir; de ; de = de->next) {
426 		if (de->namelen != dentry->d_name.len)
427 			continue;
428 		if (!memcmp(dentry->d_name.name, de->name, de->namelen)) {
429 			unsigned int ino;
430 
431 			ino = de->low_ino;
432 			pde_get(de);
433 			spin_unlock(&proc_subdir_lock);
434 			error = -EINVAL;
435 			inode = proc_get_inode(dir->i_sb, ino, de);
436 			goto out_unlock;
437 		}
438 	}
439 	spin_unlock(&proc_subdir_lock);
440 out_unlock:
441 
442 	if (inode) {
443 		dentry->d_op = &proc_dentry_operations;
444 		d_add(dentry, inode);
445 		return NULL;
446 	}
447 	if (de)
448 		pde_put(de);
449 	return ERR_PTR(error);
450 }
451 
452 struct dentry *proc_lookup(struct inode *dir, struct dentry *dentry,
453 		struct nameidata *nd)
454 {
455 	return proc_lookup_de(PDE(dir), dir, dentry);
456 }
457 
458 /*
459  * This returns non-zero if at EOF, so that the /proc
460  * root directory can use this and check if it should
461  * continue with the <pid> entries..
462  *
463  * Note that the VFS-layer doesn't care about the return
464  * value of the readdir() call, as long as it's non-negative
465  * for success..
466  */
467 int proc_readdir_de(struct proc_dir_entry *de, struct file *filp, void *dirent,
468 		filldir_t filldir)
469 {
470 	unsigned int ino;
471 	int i;
472 	struct inode *inode = filp->f_path.dentry->d_inode;
473 	int ret = 0;
474 
475 	ino = inode->i_ino;
476 	i = filp->f_pos;
477 	switch (i) {
478 		case 0:
479 			if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
480 				goto out;
481 			i++;
482 			filp->f_pos++;
483 			/* fall through */
484 		case 1:
485 			if (filldir(dirent, "..", 2, i,
486 				    parent_ino(filp->f_path.dentry),
487 				    DT_DIR) < 0)
488 				goto out;
489 			i++;
490 			filp->f_pos++;
491 			/* fall through */
492 		default:
493 			spin_lock(&proc_subdir_lock);
494 			de = de->subdir;
495 			i -= 2;
496 			for (;;) {
497 				if (!de) {
498 					ret = 1;
499 					spin_unlock(&proc_subdir_lock);
500 					goto out;
501 				}
502 				if (!i)
503 					break;
504 				de = de->next;
505 				i--;
506 			}
507 
508 			do {
509 				struct proc_dir_entry *next;
510 
511 				/* filldir passes info to user space */
512 				pde_get(de);
513 				spin_unlock(&proc_subdir_lock);
514 				if (filldir(dirent, de->name, de->namelen, filp->f_pos,
515 					    de->low_ino, de->mode >> 12) < 0) {
516 					pde_put(de);
517 					goto out;
518 				}
519 				spin_lock(&proc_subdir_lock);
520 				filp->f_pos++;
521 				next = de->next;
522 				pde_put(de);
523 				de = next;
524 			} while (de);
525 			spin_unlock(&proc_subdir_lock);
526 	}
527 	ret = 1;
528 out:
529 	return ret;
530 }
531 
532 int proc_readdir(struct file *filp, void *dirent, filldir_t filldir)
533 {
534 	struct inode *inode = filp->f_path.dentry->d_inode;
535 
536 	return proc_readdir_de(PDE(inode), filp, dirent, filldir);
537 }
538 
539 /*
540  * These are the generic /proc directory operations. They
541  * use the in-memory "struct proc_dir_entry" tree to parse
542  * the /proc directory.
543  */
544 static const struct file_operations proc_dir_operations = {
545 	.llseek			= generic_file_llseek,
546 	.read			= generic_read_dir,
547 	.readdir		= proc_readdir,
548 };
549 
550 /*
551  * proc directories can do almost nothing..
552  */
553 static const struct inode_operations proc_dir_inode_operations = {
554 	.lookup		= proc_lookup,
555 	.getattr	= proc_getattr,
556 	.setattr	= proc_notify_change,
557 };
558 
559 static int proc_register(struct proc_dir_entry * dir, struct proc_dir_entry * dp)
560 {
561 	unsigned int i;
562 	struct proc_dir_entry *tmp;
563 
564 	i = get_inode_number();
565 	if (i == 0)
566 		return -EAGAIN;
567 	dp->low_ino = i;
568 
569 	if (S_ISDIR(dp->mode)) {
570 		if (dp->proc_iops == NULL) {
571 			dp->proc_fops = &proc_dir_operations;
572 			dp->proc_iops = &proc_dir_inode_operations;
573 		}
574 		dir->nlink++;
575 	} else if (S_ISLNK(dp->mode)) {
576 		if (dp->proc_iops == NULL)
577 			dp->proc_iops = &proc_link_inode_operations;
578 	} else if (S_ISREG(dp->mode)) {
579 		if (dp->proc_fops == NULL)
580 			dp->proc_fops = &proc_file_operations;
581 		if (dp->proc_iops == NULL)
582 			dp->proc_iops = &proc_file_inode_operations;
583 	}
584 
585 	spin_lock(&proc_subdir_lock);
586 
587 	for (tmp = dir->subdir; tmp; tmp = tmp->next)
588 		if (strcmp(tmp->name, dp->name) == 0) {
589 			WARN(1, KERN_WARNING "proc_dir_entry '%s/%s' already registered\n",
590 				dir->name, dp->name);
591 			break;
592 		}
593 
594 	dp->next = dir->subdir;
595 	dp->parent = dir;
596 	dir->subdir = dp;
597 	spin_unlock(&proc_subdir_lock);
598 
599 	return 0;
600 }
601 
602 static struct proc_dir_entry *__proc_create(struct proc_dir_entry **parent,
603 					  const char *name,
604 					  mode_t mode,
605 					  nlink_t nlink)
606 {
607 	struct proc_dir_entry *ent = NULL;
608 	const char *fn = name;
609 	int len;
610 
611 	/* make sure name is valid */
612 	if (!name || !strlen(name)) goto out;
613 
614 	if (xlate_proc_name(name, parent, &fn) != 0)
615 		goto out;
616 
617 	/* At this point there must not be any '/' characters beyond *fn */
618 	if (strchr(fn, '/'))
619 		goto out;
620 
621 	len = strlen(fn);
622 
623 	ent = kmalloc(sizeof(struct proc_dir_entry) + len + 1, GFP_KERNEL);
624 	if (!ent) goto out;
625 
626 	memset(ent, 0, sizeof(struct proc_dir_entry));
627 	memcpy(((char *) ent) + sizeof(struct proc_dir_entry), fn, len + 1);
628 	ent->name = ((char *) ent) + sizeof(*ent);
629 	ent->namelen = len;
630 	ent->mode = mode;
631 	ent->nlink = nlink;
632 	atomic_set(&ent->count, 1);
633 	ent->pde_users = 0;
634 	spin_lock_init(&ent->pde_unload_lock);
635 	ent->pde_unload_completion = NULL;
636 	INIT_LIST_HEAD(&ent->pde_openers);
637  out:
638 	return ent;
639 }
640 
641 struct proc_dir_entry *proc_symlink(const char *name,
642 		struct proc_dir_entry *parent, const char *dest)
643 {
644 	struct proc_dir_entry *ent;
645 
646 	ent = __proc_create(&parent, name,
647 			  (S_IFLNK | S_IRUGO | S_IWUGO | S_IXUGO),1);
648 
649 	if (ent) {
650 		ent->data = kmalloc((ent->size=strlen(dest))+1, GFP_KERNEL);
651 		if (ent->data) {
652 			strcpy((char*)ent->data,dest);
653 			if (proc_register(parent, ent) < 0) {
654 				kfree(ent->data);
655 				kfree(ent);
656 				ent = NULL;
657 			}
658 		} else {
659 			kfree(ent);
660 			ent = NULL;
661 		}
662 	}
663 	return ent;
664 }
665 
666 struct proc_dir_entry *proc_mkdir_mode(const char *name, mode_t mode,
667 		struct proc_dir_entry *parent)
668 {
669 	struct proc_dir_entry *ent;
670 
671 	ent = __proc_create(&parent, name, S_IFDIR | mode, 2);
672 	if (ent) {
673 		if (proc_register(parent, ent) < 0) {
674 			kfree(ent);
675 			ent = NULL;
676 		}
677 	}
678 	return ent;
679 }
680 
681 struct proc_dir_entry *proc_net_mkdir(struct net *net, const char *name,
682 		struct proc_dir_entry *parent)
683 {
684 	struct proc_dir_entry *ent;
685 
686 	ent = __proc_create(&parent, name, S_IFDIR | S_IRUGO | S_IXUGO, 2);
687 	if (ent) {
688 		ent->data = net;
689 		if (proc_register(parent, ent) < 0) {
690 			kfree(ent);
691 			ent = NULL;
692 		}
693 	}
694 	return ent;
695 }
696 EXPORT_SYMBOL_GPL(proc_net_mkdir);
697 
698 struct proc_dir_entry *proc_mkdir(const char *name,
699 		struct proc_dir_entry *parent)
700 {
701 	return proc_mkdir_mode(name, S_IRUGO | S_IXUGO, parent);
702 }
703 
704 struct proc_dir_entry *create_proc_entry(const char *name, mode_t mode,
705 					 struct proc_dir_entry *parent)
706 {
707 	struct proc_dir_entry *ent;
708 	nlink_t nlink;
709 
710 	if (S_ISDIR(mode)) {
711 		if ((mode & S_IALLUGO) == 0)
712 			mode |= S_IRUGO | S_IXUGO;
713 		nlink = 2;
714 	} else {
715 		if ((mode & S_IFMT) == 0)
716 			mode |= S_IFREG;
717 		if ((mode & S_IALLUGO) == 0)
718 			mode |= S_IRUGO;
719 		nlink = 1;
720 	}
721 
722 	ent = __proc_create(&parent, name, mode, nlink);
723 	if (ent) {
724 		if (proc_register(parent, ent) < 0) {
725 			kfree(ent);
726 			ent = NULL;
727 		}
728 	}
729 	return ent;
730 }
731 
732 struct proc_dir_entry *proc_create_data(const char *name, mode_t mode,
733 					struct proc_dir_entry *parent,
734 					const struct file_operations *proc_fops,
735 					void *data)
736 {
737 	struct proc_dir_entry *pde;
738 	nlink_t nlink;
739 
740 	if (S_ISDIR(mode)) {
741 		if ((mode & S_IALLUGO) == 0)
742 			mode |= S_IRUGO | S_IXUGO;
743 		nlink = 2;
744 	} else {
745 		if ((mode & S_IFMT) == 0)
746 			mode |= S_IFREG;
747 		if ((mode & S_IALLUGO) == 0)
748 			mode |= S_IRUGO;
749 		nlink = 1;
750 	}
751 
752 	pde = __proc_create(&parent, name, mode, nlink);
753 	if (!pde)
754 		goto out;
755 	pde->proc_fops = proc_fops;
756 	pde->data = data;
757 	if (proc_register(parent, pde) < 0)
758 		goto out_free;
759 	return pde;
760 out_free:
761 	kfree(pde);
762 out:
763 	return NULL;
764 }
765 
766 static void free_proc_entry(struct proc_dir_entry *de)
767 {
768 	unsigned int ino = de->low_ino;
769 
770 	if (ino < PROC_DYNAMIC_FIRST)
771 		return;
772 
773 	release_inode_number(ino);
774 
775 	if (S_ISLNK(de->mode))
776 		kfree(de->data);
777 	kfree(de);
778 }
779 
780 void pde_put(struct proc_dir_entry *pde)
781 {
782 	if (atomic_dec_and_test(&pde->count))
783 		free_proc_entry(pde);
784 }
785 
786 /*
787  * Remove a /proc entry and free it if it's not currently in use.
788  */
789 void remove_proc_entry(const char *name, struct proc_dir_entry *parent)
790 {
791 	struct proc_dir_entry **p;
792 	struct proc_dir_entry *de = NULL;
793 	const char *fn = name;
794 	int len;
795 
796 	if (xlate_proc_name(name, &parent, &fn) != 0)
797 		return;
798 	len = strlen(fn);
799 
800 	spin_lock(&proc_subdir_lock);
801 	for (p = &parent->subdir; *p; p=&(*p)->next ) {
802 		if (proc_match(len, fn, *p)) {
803 			de = *p;
804 			*p = de->next;
805 			de->next = NULL;
806 			break;
807 		}
808 	}
809 	spin_unlock(&proc_subdir_lock);
810 	if (!de)
811 		return;
812 
813 	spin_lock(&de->pde_unload_lock);
814 	/*
815 	 * Stop accepting new callers into module. If you're
816 	 * dynamically allocating ->proc_fops, save a pointer somewhere.
817 	 */
818 	de->proc_fops = NULL;
819 	/* Wait until all existing callers into module are done. */
820 	if (de->pde_users > 0) {
821 		DECLARE_COMPLETION_ONSTACK(c);
822 
823 		if (!de->pde_unload_completion)
824 			de->pde_unload_completion = &c;
825 
826 		spin_unlock(&de->pde_unload_lock);
827 
828 		wait_for_completion(de->pde_unload_completion);
829 
830 		goto continue_removing;
831 	}
832 	spin_unlock(&de->pde_unload_lock);
833 
834 continue_removing:
835 	spin_lock(&de->pde_unload_lock);
836 	while (!list_empty(&de->pde_openers)) {
837 		struct pde_opener *pdeo;
838 
839 		pdeo = list_first_entry(&de->pde_openers, struct pde_opener, lh);
840 		list_del(&pdeo->lh);
841 		spin_unlock(&de->pde_unload_lock);
842 		pdeo->release(pdeo->inode, pdeo->file);
843 		kfree(pdeo);
844 		spin_lock(&de->pde_unload_lock);
845 	}
846 	spin_unlock(&de->pde_unload_lock);
847 
848 	if (S_ISDIR(de->mode))
849 		parent->nlink--;
850 	de->nlink = 0;
851 	WARN(de->subdir, KERN_WARNING "%s: removing non-empty directory "
852 			"'%s/%s', leaking at least '%s'\n", __func__,
853 			de->parent->name, de->name, de->subdir->name);
854 	pde_put(de);
855 }
856