xref: /openbmc/linux/fs/proc/generic.c (revision e868d61272caa648214046a096e5a6bfc068dc8c)
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/smp_lock.h>
18 #include <linux/init.h>
19 #include <linux/idr.h>
20 #include <linux/namei.h>
21 #include <linux/bitops.h>
22 #include <linux/spinlock.h>
23 #include <asm/uaccess.h>
24 
25 #include "internal.h"
26 
27 static ssize_t proc_file_read(struct file *file, char __user *buf,
28 			      size_t nbytes, loff_t *ppos);
29 static ssize_t proc_file_write(struct file *file, const char __user *buffer,
30 			       size_t count, loff_t *ppos);
31 static loff_t proc_file_lseek(struct file *, loff_t, int);
32 
33 DEFINE_SPINLOCK(proc_subdir_lock);
34 
35 static int proc_match(int len, const char *name, struct proc_dir_entry *de)
36 {
37 	if (de->namelen != len)
38 		return 0;
39 	return !memcmp(name, de->name, len);
40 }
41 
42 static const struct file_operations proc_file_operations = {
43 	.llseek		= proc_file_lseek,
44 	.read		= proc_file_read,
45 	.write		= proc_file_write,
46 };
47 
48 /* buffer size is one page but our output routines use some slack for overruns */
49 #define PROC_BLOCK_SIZE	(PAGE_SIZE - 1024)
50 
51 static ssize_t
52 proc_file_read(struct file *file, char __user *buf, size_t nbytes,
53 	       loff_t *ppos)
54 {
55 	struct inode * inode = file->f_path.dentry->d_inode;
56 	char 	*page;
57 	ssize_t	retval=0;
58 	int	eof=0;
59 	ssize_t	n, count;
60 	char	*start;
61 	struct proc_dir_entry * dp;
62 	unsigned long long pos;
63 
64 	/*
65 	 * Gaah, please just use "seq_file" instead. The legacy /proc
66 	 * interfaces cut loff_t down to off_t for reads, and ignore
67 	 * the offset entirely for writes..
68 	 */
69 	pos = *ppos;
70 	if (pos > MAX_NON_LFS)
71 		return 0;
72 	if (nbytes > MAX_NON_LFS - pos)
73 		nbytes = MAX_NON_LFS - pos;
74 
75 	dp = PDE(inode);
76 	if (!(page = (char*) __get_free_page(GFP_KERNEL)))
77 		return -ENOMEM;
78 
79 	while ((nbytes > 0) && !eof) {
80 		count = min_t(size_t, PROC_BLOCK_SIZE, nbytes);
81 
82 		start = NULL;
83 		if (dp->get_info) {
84 			/* Handle old net routines */
85 			n = dp->get_info(page, &start, *ppos, count);
86 			if (n < count)
87 				eof = 1;
88 		} else if (dp->read_proc) {
89 			/*
90 			 * How to be a proc read function
91 			 * ------------------------------
92 			 * Prototype:
93 			 *    int f(char *buffer, char **start, off_t offset,
94 			 *          int count, int *peof, void *dat)
95 			 *
96 			 * Assume that the buffer is "count" bytes in size.
97 			 *
98 			 * If you know you have supplied all the data you
99 			 * have, set *peof.
100 			 *
101 			 * You have three ways to return data:
102 			 * 0) Leave *start = NULL.  (This is the default.)
103 			 *    Put the data of the requested offset at that
104 			 *    offset within the buffer.  Return the number (n)
105 			 *    of bytes there are from the beginning of the
106 			 *    buffer up to the last byte of data.  If the
107 			 *    number of supplied bytes (= n - offset) is
108 			 *    greater than zero and you didn't signal eof
109 			 *    and the reader is prepared to take more data
110 			 *    you will be called again with the requested
111 			 *    offset advanced by the number of bytes
112 			 *    absorbed.  This interface is useful for files
113 			 *    no larger than the buffer.
114 			 * 1) Set *start = an unsigned long value less than
115 			 *    the buffer address but greater than zero.
116 			 *    Put the data of the requested offset at the
117 			 *    beginning of the buffer.  Return the number of
118 			 *    bytes of data placed there.  If this number is
119 			 *    greater than zero and you didn't signal eof
120 			 *    and the reader is prepared to take more data
121 			 *    you will be called again with the requested
122 			 *    offset advanced by *start.  This interface is
123 			 *    useful when you have a large file consisting
124 			 *    of a series of blocks which you want to count
125 			 *    and return as wholes.
126 			 *    (Hack by Paul.Russell@rustcorp.com.au)
127 			 * 2) Set *start = an address within the buffer.
128 			 *    Put the data of the requested offset at *start.
129 			 *    Return the number of bytes of data placed there.
130 			 *    If this number is greater than zero and you
131 			 *    didn't signal eof and the reader is prepared to
132 			 *    take more data you will be called again with the
133 			 *    requested offset advanced by the number of bytes
134 			 *    absorbed.
135 			 */
136 			n = dp->read_proc(page, &start, *ppos,
137 					  count, &eof, dp->data);
138 		} else
139 			break;
140 
141 		if (n == 0)   /* end of file */
142 			break;
143 		if (n < 0) {  /* error */
144 			if (retval == 0)
145 				retval = n;
146 			break;
147 		}
148 
149 		if (start == NULL) {
150 			if (n > PAGE_SIZE) {
151 				printk(KERN_ERR
152 				       "proc_file_read: Apparent buffer overflow!\n");
153 				n = PAGE_SIZE;
154 			}
155 			n -= *ppos;
156 			if (n <= 0)
157 				break;
158 			if (n > count)
159 				n = count;
160 			start = page + *ppos;
161 		} else if (start < page) {
162 			if (n > PAGE_SIZE) {
163 				printk(KERN_ERR
164 				       "proc_file_read: Apparent buffer overflow!\n");
165 				n = PAGE_SIZE;
166 			}
167 			if (n > count) {
168 				/*
169 				 * Don't reduce n because doing so might
170 				 * cut off part of a data block.
171 				 */
172 				printk(KERN_WARNING
173 				       "proc_file_read: Read count exceeded\n");
174 			}
175 		} else /* start >= page */ {
176 			unsigned long startoff = (unsigned long)(start - page);
177 			if (n > (PAGE_SIZE - startoff)) {
178 				printk(KERN_ERR
179 				       "proc_file_read: Apparent buffer overflow!\n");
180 				n = PAGE_SIZE - startoff;
181 			}
182 			if (n > count)
183 				n = count;
184 		}
185 
186  		n -= copy_to_user(buf, start < page ? page : start, n);
187 		if (n == 0) {
188 			if (retval == 0)
189 				retval = -EFAULT;
190 			break;
191 		}
192 
193 		*ppos += start < page ? (unsigned long)start : n;
194 		nbytes -= n;
195 		buf += n;
196 		retval += n;
197 	}
198 	free_page((unsigned long) page);
199 	return retval;
200 }
201 
202 static ssize_t
203 proc_file_write(struct file *file, const char __user *buffer,
204 		size_t count, loff_t *ppos)
205 {
206 	struct inode *inode = file->f_path.dentry->d_inode;
207 	struct proc_dir_entry * dp;
208 
209 	dp = PDE(inode);
210 
211 	if (!dp->write_proc)
212 		return -EIO;
213 
214 	/* FIXME: does this routine need ppos?  probably... */
215 	return dp->write_proc(file, buffer, count, dp->data);
216 }
217 
218 
219 static loff_t
220 proc_file_lseek(struct file *file, loff_t offset, int orig)
221 {
222 	loff_t retval = -EINVAL;
223 	switch (orig) {
224 	case 1:
225 		offset += file->f_pos;
226 	/* fallthrough */
227 	case 0:
228 		if (offset < 0 || offset > MAX_NON_LFS)
229 			break;
230 		file->f_pos = retval = offset;
231 	}
232 	return retval;
233 }
234 
235 static int proc_notify_change(struct dentry *dentry, struct iattr *iattr)
236 {
237 	struct inode *inode = dentry->d_inode;
238 	struct proc_dir_entry *de = PDE(inode);
239 	int error;
240 
241 	error = inode_change_ok(inode, iattr);
242 	if (error)
243 		goto out;
244 
245 	error = inode_setattr(inode, iattr);
246 	if (error)
247 		goto out;
248 
249 	de->uid = inode->i_uid;
250 	de->gid = inode->i_gid;
251 	de->mode = inode->i_mode;
252 out:
253 	return error;
254 }
255 
256 static int proc_getattr(struct vfsmount *mnt, struct dentry *dentry,
257 			struct kstat *stat)
258 {
259 	struct inode *inode = dentry->d_inode;
260 	struct proc_dir_entry *de = PROC_I(inode)->pde;
261 	if (de && de->nlink)
262 		inode->i_nlink = de->nlink;
263 
264 	generic_fillattr(inode, stat);
265 	return 0;
266 }
267 
268 static const struct inode_operations proc_file_inode_operations = {
269 	.setattr	= proc_notify_change,
270 };
271 
272 /*
273  * This function parses a name such as "tty/driver/serial", and
274  * returns the struct proc_dir_entry for "/proc/tty/driver", and
275  * returns "serial" in residual.
276  */
277 static int xlate_proc_name(const char *name,
278 			   struct proc_dir_entry **ret, const char **residual)
279 {
280 	const char     		*cp = name, *next;
281 	struct proc_dir_entry	*de;
282 	int			len;
283 	int 			rtn = 0;
284 
285 	spin_lock(&proc_subdir_lock);
286 	de = &proc_root;
287 	while (1) {
288 		next = strchr(cp, '/');
289 		if (!next)
290 			break;
291 
292 		len = next - cp;
293 		for (de = de->subdir; de ; de = de->next) {
294 			if (proc_match(len, cp, de))
295 				break;
296 		}
297 		if (!de) {
298 			rtn = -ENOENT;
299 			goto out;
300 		}
301 		cp += len + 1;
302 	}
303 	*residual = cp;
304 	*ret = de;
305 out:
306 	spin_unlock(&proc_subdir_lock);
307 	return rtn;
308 }
309 
310 static DEFINE_IDR(proc_inum_idr);
311 static DEFINE_SPINLOCK(proc_inum_lock); /* protects the above */
312 
313 #define PROC_DYNAMIC_FIRST 0xF0000000UL
314 
315 /*
316  * Return an inode number between PROC_DYNAMIC_FIRST and
317  * 0xffffffff, or zero on failure.
318  */
319 static unsigned int get_inode_number(void)
320 {
321 	int i, inum = 0;
322 	int error;
323 
324 retry:
325 	if (idr_pre_get(&proc_inum_idr, GFP_KERNEL) == 0)
326 		return 0;
327 
328 	spin_lock(&proc_inum_lock);
329 	error = idr_get_new(&proc_inum_idr, NULL, &i);
330 	spin_unlock(&proc_inum_lock);
331 	if (error == -EAGAIN)
332 		goto retry;
333 	else if (error)
334 		return 0;
335 
336 	inum = (i & MAX_ID_MASK) + PROC_DYNAMIC_FIRST;
337 
338 	/* inum will never be more than 0xf0ffffff, so no check
339 	 * for overflow.
340 	 */
341 
342 	return inum;
343 }
344 
345 static void release_inode_number(unsigned int inum)
346 {
347 	int id = (inum - PROC_DYNAMIC_FIRST) | ~MAX_ID_MASK;
348 
349 	spin_lock(&proc_inum_lock);
350 	idr_remove(&proc_inum_idr, id);
351 	spin_unlock(&proc_inum_lock);
352 }
353 
354 static void *proc_follow_link(struct dentry *dentry, struct nameidata *nd)
355 {
356 	nd_set_link(nd, PDE(dentry->d_inode)->data);
357 	return NULL;
358 }
359 
360 static const struct inode_operations proc_link_inode_operations = {
361 	.readlink	= generic_readlink,
362 	.follow_link	= proc_follow_link,
363 };
364 
365 /*
366  * As some entries in /proc are volatile, we want to
367  * get rid of unused dentries.  This could be made
368  * smarter: we could keep a "volatile" flag in the
369  * inode to indicate which ones to keep.
370  */
371 static int proc_delete_dentry(struct dentry * dentry)
372 {
373 	return 1;
374 }
375 
376 static struct dentry_operations proc_dentry_operations =
377 {
378 	.d_delete	= proc_delete_dentry,
379 };
380 
381 /*
382  * Don't create negative dentries here, return -ENOENT by hand
383  * instead.
384  */
385 struct dentry *proc_lookup(struct inode * dir, struct dentry *dentry, struct nameidata *nd)
386 {
387 	struct inode *inode = NULL;
388 	struct proc_dir_entry * de;
389 	int error = -ENOENT;
390 
391 	lock_kernel();
392 	spin_lock(&proc_subdir_lock);
393 	de = PDE(dir);
394 	if (de) {
395 		for (de = de->subdir; de ; de = de->next) {
396 			if (de->namelen != dentry->d_name.len)
397 				continue;
398 			if (!memcmp(dentry->d_name.name, de->name, de->namelen)) {
399 				unsigned int ino = de->low_ino;
400 
401 				de_get(de);
402 				spin_unlock(&proc_subdir_lock);
403 				error = -EINVAL;
404 				inode = proc_get_inode(dir->i_sb, ino, de);
405 				spin_lock(&proc_subdir_lock);
406 				break;
407 			}
408 		}
409 	}
410 	spin_unlock(&proc_subdir_lock);
411 	unlock_kernel();
412 
413 	if (inode) {
414 		dentry->d_op = &proc_dentry_operations;
415 		d_add(dentry, inode);
416 		return NULL;
417 	}
418 	de_put(de);
419 	return ERR_PTR(error);
420 }
421 
422 /*
423  * This returns non-zero if at EOF, so that the /proc
424  * root directory can use this and check if it should
425  * continue with the <pid> entries..
426  *
427  * Note that the VFS-layer doesn't care about the return
428  * value of the readdir() call, as long as it's non-negative
429  * for success..
430  */
431 int proc_readdir(struct file * filp,
432 	void * dirent, filldir_t filldir)
433 {
434 	struct proc_dir_entry * de;
435 	unsigned int ino;
436 	int i;
437 	struct inode *inode = filp->f_path.dentry->d_inode;
438 	int ret = 0;
439 
440 	lock_kernel();
441 
442 	ino = inode->i_ino;
443 	de = PDE(inode);
444 	if (!de) {
445 		ret = -EINVAL;
446 		goto out;
447 	}
448 	i = filp->f_pos;
449 	switch (i) {
450 		case 0:
451 			if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
452 				goto out;
453 			i++;
454 			filp->f_pos++;
455 			/* fall through */
456 		case 1:
457 			if (filldir(dirent, "..", 2, i,
458 				    parent_ino(filp->f_path.dentry),
459 				    DT_DIR) < 0)
460 				goto out;
461 			i++;
462 			filp->f_pos++;
463 			/* fall through */
464 		default:
465 			spin_lock(&proc_subdir_lock);
466 			de = de->subdir;
467 			i -= 2;
468 			for (;;) {
469 				if (!de) {
470 					ret = 1;
471 					spin_unlock(&proc_subdir_lock);
472 					goto out;
473 				}
474 				if (!i)
475 					break;
476 				de = de->next;
477 				i--;
478 			}
479 
480 			do {
481 				struct proc_dir_entry *next;
482 
483 				/* filldir passes info to user space */
484 				de_get(de);
485 				spin_unlock(&proc_subdir_lock);
486 				if (filldir(dirent, de->name, de->namelen, filp->f_pos,
487 					    de->low_ino, de->mode >> 12) < 0) {
488 					de_put(de);
489 					goto out;
490 				}
491 				spin_lock(&proc_subdir_lock);
492 				filp->f_pos++;
493 				next = de->next;
494 				de_put(de);
495 				de = next;
496 			} while (de);
497 			spin_unlock(&proc_subdir_lock);
498 	}
499 	ret = 1;
500 out:	unlock_kernel();
501 	return ret;
502 }
503 
504 /*
505  * These are the generic /proc directory operations. They
506  * use the in-memory "struct proc_dir_entry" tree to parse
507  * the /proc directory.
508  */
509 static const struct file_operations proc_dir_operations = {
510 	.read			= generic_read_dir,
511 	.readdir		= proc_readdir,
512 };
513 
514 /*
515  * proc directories can do almost nothing..
516  */
517 static const struct inode_operations proc_dir_inode_operations = {
518 	.lookup		= proc_lookup,
519 	.getattr	= proc_getattr,
520 	.setattr	= proc_notify_change,
521 };
522 
523 static int proc_register(struct proc_dir_entry * dir, struct proc_dir_entry * dp)
524 {
525 	unsigned int i;
526 
527 	i = get_inode_number();
528 	if (i == 0)
529 		return -EAGAIN;
530 	dp->low_ino = i;
531 
532 	spin_lock(&proc_subdir_lock);
533 	dp->next = dir->subdir;
534 	dp->parent = dir;
535 	dir->subdir = dp;
536 	spin_unlock(&proc_subdir_lock);
537 
538 	if (S_ISDIR(dp->mode)) {
539 		if (dp->proc_iops == NULL) {
540 			dp->proc_fops = &proc_dir_operations;
541 			dp->proc_iops = &proc_dir_inode_operations;
542 		}
543 		dir->nlink++;
544 	} else if (S_ISLNK(dp->mode)) {
545 		if (dp->proc_iops == NULL)
546 			dp->proc_iops = &proc_link_inode_operations;
547 	} else if (S_ISREG(dp->mode)) {
548 		if (dp->proc_fops == NULL)
549 			dp->proc_fops = &proc_file_operations;
550 		if (dp->proc_iops == NULL)
551 			dp->proc_iops = &proc_file_inode_operations;
552 	}
553 	return 0;
554 }
555 
556 /*
557  * Kill an inode that got unregistered..
558  */
559 static void proc_kill_inodes(struct proc_dir_entry *de)
560 {
561 	struct list_head *p;
562 	struct super_block *sb = proc_mnt->mnt_sb;
563 
564 	/*
565 	 * Actually it's a partial revoke().
566 	 */
567 	file_list_lock();
568 	list_for_each(p, &sb->s_files) {
569 		struct file * filp = list_entry(p, struct file, f_u.fu_list);
570 		struct dentry * dentry = filp->f_path.dentry;
571 		struct inode * inode;
572 		const struct file_operations *fops;
573 
574 		if (dentry->d_op != &proc_dentry_operations)
575 			continue;
576 		inode = dentry->d_inode;
577 		if (PDE(inode) != de)
578 			continue;
579 		fops = filp->f_op;
580 		filp->f_op = NULL;
581 		fops_put(fops);
582 	}
583 	file_list_unlock();
584 }
585 
586 static struct proc_dir_entry *proc_create(struct proc_dir_entry **parent,
587 					  const char *name,
588 					  mode_t mode,
589 					  nlink_t nlink)
590 {
591 	struct proc_dir_entry *ent = NULL;
592 	const char *fn = name;
593 	int len;
594 
595 	/* make sure name is valid */
596 	if (!name || !strlen(name)) goto out;
597 
598 	if (!(*parent) && xlate_proc_name(name, parent, &fn) != 0)
599 		goto out;
600 
601 	/* At this point there must not be any '/' characters beyond *fn */
602 	if (strchr(fn, '/'))
603 		goto out;
604 
605 	len = strlen(fn);
606 
607 	ent = kmalloc(sizeof(struct proc_dir_entry) + len + 1, GFP_KERNEL);
608 	if (!ent) goto out;
609 
610 	memset(ent, 0, sizeof(struct proc_dir_entry));
611 	memcpy(((char *) ent) + sizeof(struct proc_dir_entry), fn, len + 1);
612 	ent->name = ((char *) ent) + sizeof(*ent);
613 	ent->namelen = len;
614 	ent->mode = mode;
615 	ent->nlink = nlink;
616  out:
617 	return ent;
618 }
619 
620 struct proc_dir_entry *proc_symlink(const char *name,
621 		struct proc_dir_entry *parent, const char *dest)
622 {
623 	struct proc_dir_entry *ent;
624 
625 	ent = proc_create(&parent,name,
626 			  (S_IFLNK | S_IRUGO | S_IWUGO | S_IXUGO),1);
627 
628 	if (ent) {
629 		ent->data = kmalloc((ent->size=strlen(dest))+1, GFP_KERNEL);
630 		if (ent->data) {
631 			strcpy((char*)ent->data,dest);
632 			if (proc_register(parent, ent) < 0) {
633 				kfree(ent->data);
634 				kfree(ent);
635 				ent = NULL;
636 			}
637 		} else {
638 			kfree(ent);
639 			ent = NULL;
640 		}
641 	}
642 	return ent;
643 }
644 
645 struct proc_dir_entry *proc_mkdir_mode(const char *name, mode_t mode,
646 		struct proc_dir_entry *parent)
647 {
648 	struct proc_dir_entry *ent;
649 
650 	ent = proc_create(&parent, name, S_IFDIR | mode, 2);
651 	if (ent) {
652 		ent->proc_fops = &proc_dir_operations;
653 		ent->proc_iops = &proc_dir_inode_operations;
654 
655 		if (proc_register(parent, ent) < 0) {
656 			kfree(ent);
657 			ent = NULL;
658 		}
659 	}
660 	return ent;
661 }
662 
663 struct proc_dir_entry *proc_mkdir(const char *name,
664 		struct proc_dir_entry *parent)
665 {
666 	return proc_mkdir_mode(name, S_IRUGO | S_IXUGO, parent);
667 }
668 
669 struct proc_dir_entry *create_proc_entry(const char *name, mode_t mode,
670 					 struct proc_dir_entry *parent)
671 {
672 	struct proc_dir_entry *ent;
673 	nlink_t nlink;
674 
675 	if (S_ISDIR(mode)) {
676 		if ((mode & S_IALLUGO) == 0)
677 			mode |= S_IRUGO | S_IXUGO;
678 		nlink = 2;
679 	} else {
680 		if ((mode & S_IFMT) == 0)
681 			mode |= S_IFREG;
682 		if ((mode & S_IALLUGO) == 0)
683 			mode |= S_IRUGO;
684 		nlink = 1;
685 	}
686 
687 	ent = proc_create(&parent,name,mode,nlink);
688 	if (ent) {
689 		if (S_ISDIR(mode)) {
690 			ent->proc_fops = &proc_dir_operations;
691 			ent->proc_iops = &proc_dir_inode_operations;
692 		}
693 		if (proc_register(parent, ent) < 0) {
694 			kfree(ent);
695 			ent = NULL;
696 		}
697 	}
698 	return ent;
699 }
700 
701 void free_proc_entry(struct proc_dir_entry *de)
702 {
703 	unsigned int ino = de->low_ino;
704 
705 	if (ino < PROC_DYNAMIC_FIRST)
706 		return;
707 
708 	release_inode_number(ino);
709 
710 	if (S_ISLNK(de->mode) && de->data)
711 		kfree(de->data);
712 	kfree(de);
713 }
714 
715 /*
716  * Remove a /proc entry and free it if it's not currently in use.
717  * If it is in use, we set the 'deleted' flag.
718  */
719 void remove_proc_entry(const char *name, struct proc_dir_entry *parent)
720 {
721 	struct proc_dir_entry **p;
722 	struct proc_dir_entry *de;
723 	const char *fn = name;
724 	int len;
725 
726 	if (!parent && xlate_proc_name(name, &parent, &fn) != 0)
727 		goto out;
728 	len = strlen(fn);
729 
730 	spin_lock(&proc_subdir_lock);
731 	for (p = &parent->subdir; *p; p=&(*p)->next ) {
732 		if (!proc_match(len, fn, *p))
733 			continue;
734 		de = *p;
735 		*p = de->next;
736 		de->next = NULL;
737 		if (S_ISDIR(de->mode))
738 			parent->nlink--;
739 		proc_kill_inodes(de);
740 		de->nlink = 0;
741 		WARN_ON(de->subdir);
742 		if (!atomic_read(&de->count))
743 			free_proc_entry(de);
744 		else {
745 			de->deleted = 1;
746 			printk("remove_proc_entry: %s/%s busy, count=%d\n",
747 				parent->name, de->name, atomic_read(&de->count));
748 		}
749 		break;
750 	}
751 	spin_unlock(&proc_subdir_lock);
752 out:
753 	return;
754 }
755