Loading...
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/mm.h>
16#include <linux/module.h>
17#include <linux/slab.h>
18#include <linux/mount.h>
19#include <linux/init.h>
20#include <linux/idr.h>
21#include <linux/namei.h>
22#include <linux/bitops.h>
23#include <linux/spinlock.h>
24#include <linux/completion.h>
25#include <asm/uaccess.h>
26
27#include "internal.h"
28
29DEFINE_SPINLOCK(proc_subdir_lock);
30
31static int proc_match(unsigned int len, const char *name, struct proc_dir_entry *de)
32{
33 if (de->namelen != len)
34 return 0;
35 return !memcmp(name, de->name, len);
36}
37
38/* buffer size is one page but our output routines use some slack for overruns */
39#define PROC_BLOCK_SIZE (PAGE_SIZE - 1024)
40
41static ssize_t
42__proc_file_read(struct file *file, char __user *buf, size_t nbytes,
43 loff_t *ppos)
44{
45 struct inode * inode = file->f_path.dentry->d_inode;
46 char *page;
47 ssize_t retval=0;
48 int eof=0;
49 ssize_t n, count;
50 char *start;
51 struct proc_dir_entry * dp;
52 unsigned long long pos;
53
54 /*
55 * Gaah, please just use "seq_file" instead. The legacy /proc
56 * interfaces cut loff_t down to off_t for reads, and ignore
57 * the offset entirely for writes..
58 */
59 pos = *ppos;
60 if (pos > MAX_NON_LFS)
61 return 0;
62 if (nbytes > MAX_NON_LFS - pos)
63 nbytes = MAX_NON_LFS - pos;
64
65 dp = PDE(inode);
66 if (!(page = (char*) __get_free_page(GFP_TEMPORARY)))
67 return -ENOMEM;
68
69 while ((nbytes > 0) && !eof) {
70 count = min_t(size_t, PROC_BLOCK_SIZE, nbytes);
71
72 start = NULL;
73 if (dp->read_proc) {
74 /*
75 * How to be a proc read function
76 * ------------------------------
77 * Prototype:
78 * int f(char *buffer, char **start, off_t offset,
79 * int count, int *peof, void *dat)
80 *
81 * Assume that the buffer is "count" bytes in size.
82 *
83 * If you know you have supplied all the data you
84 * have, set *peof.
85 *
86 * You have three ways to return data:
87 * 0) Leave *start = NULL. (This is the default.)
88 * Put the data of the requested offset at that
89 * offset within the buffer. Return the number (n)
90 * of bytes there are from the beginning of the
91 * buffer up to the last byte of data. If the
92 * number of supplied bytes (= n - offset) is
93 * greater than zero and you didn't signal eof
94 * and the reader is prepared to take more data
95 * you will be called again with the requested
96 * offset advanced by the number of bytes
97 * absorbed. This interface is useful for files
98 * no larger than the buffer.
99 * 1) Set *start = an unsigned long value less than
100 * the buffer address but greater than zero.
101 * Put the data of the requested offset at the
102 * beginning of the buffer. Return the number of
103 * bytes of data placed there. If this number is
104 * greater than zero and you didn't signal eof
105 * and the reader is prepared to take more data
106 * you will be called again with the requested
107 * offset advanced by *start. This interface is
108 * useful when you have a large file consisting
109 * of a series of blocks which you want to count
110 * and return as wholes.
111 * (Hack by Paul.Russell@rustcorp.com.au)
112 * 2) Set *start = an address within the buffer.
113 * Put the data of the requested offset at *start.
114 * Return the number of bytes of data placed there.
115 * If this number is greater than zero and you
116 * didn't signal eof and the reader is prepared to
117 * take more data you will be called again with the
118 * requested offset advanced by the number of bytes
119 * absorbed.
120 */
121 n = dp->read_proc(page, &start, *ppos,
122 count, &eof, dp->data);
123 } else
124 break;
125
126 if (n == 0) /* end of file */
127 break;
128 if (n < 0) { /* error */
129 if (retval == 0)
130 retval = n;
131 break;
132 }
133
134 if (start == NULL) {
135 if (n > PAGE_SIZE) {
136 printk(KERN_ERR
137 "proc_file_read: Apparent buffer overflow!\n");
138 n = PAGE_SIZE;
139 }
140 n -= *ppos;
141 if (n <= 0)
142 break;
143 if (n > count)
144 n = count;
145 start = page + *ppos;
146 } else if (start < page) {
147 if (n > PAGE_SIZE) {
148 printk(KERN_ERR
149 "proc_file_read: Apparent buffer overflow!\n");
150 n = PAGE_SIZE;
151 }
152 if (n > count) {
153 /*
154 * Don't reduce n because doing so might
155 * cut off part of a data block.
156 */
157 printk(KERN_WARNING
158 "proc_file_read: Read count exceeded\n");
159 }
160 } else /* start >= page */ {
161 unsigned long startoff = (unsigned long)(start - page);
162 if (n > (PAGE_SIZE - startoff)) {
163 printk(KERN_ERR
164 "proc_file_read: Apparent buffer overflow!\n");
165 n = PAGE_SIZE - startoff;
166 }
167 if (n > count)
168 n = count;
169 }
170
171 n -= copy_to_user(buf, start < page ? page : start, n);
172 if (n == 0) {
173 if (retval == 0)
174 retval = -EFAULT;
175 break;
176 }
177
178 *ppos += start < page ? (unsigned long)start : n;
179 nbytes -= n;
180 buf += n;
181 retval += n;
182 }
183 free_page((unsigned long) page);
184 return retval;
185}
186
187static ssize_t
188proc_file_read(struct file *file, char __user *buf, size_t nbytes,
189 loff_t *ppos)
190{
191 struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
192 ssize_t rv = -EIO;
193
194 spin_lock(&pde->pde_unload_lock);
195 if (!pde->proc_fops) {
196 spin_unlock(&pde->pde_unload_lock);
197 return rv;
198 }
199 pde->pde_users++;
200 spin_unlock(&pde->pde_unload_lock);
201
202 rv = __proc_file_read(file, buf, nbytes, ppos);
203
204 pde_users_dec(pde);
205 return rv;
206}
207
208static ssize_t
209proc_file_write(struct file *file, const char __user *buffer,
210 size_t count, loff_t *ppos)
211{
212 struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
213 ssize_t rv = -EIO;
214
215 if (pde->write_proc) {
216 spin_lock(&pde->pde_unload_lock);
217 if (!pde->proc_fops) {
218 spin_unlock(&pde->pde_unload_lock);
219 return rv;
220 }
221 pde->pde_users++;
222 spin_unlock(&pde->pde_unload_lock);
223
224 /* FIXME: does this routine need ppos? probably... */
225 rv = pde->write_proc(file, buffer, count, pde->data);
226 pde_users_dec(pde);
227 }
228 return rv;
229}
230
231
232static loff_t
233proc_file_lseek(struct file *file, loff_t offset, int orig)
234{
235 loff_t retval = -EINVAL;
236 switch (orig) {
237 case 1:
238 offset += file->f_pos;
239 /* fallthrough */
240 case 0:
241 if (offset < 0 || offset > MAX_NON_LFS)
242 break;
243 file->f_pos = retval = offset;
244 }
245 return retval;
246}
247
248static const struct file_operations proc_file_operations = {
249 .llseek = proc_file_lseek,
250 .read = proc_file_read,
251 .write = proc_file_write,
252};
253
254static int proc_notify_change(struct dentry *dentry, struct iattr *iattr)
255{
256 struct inode *inode = dentry->d_inode;
257 struct proc_dir_entry *de = PDE(inode);
258 int error;
259
260 error = inode_change_ok(inode, iattr);
261 if (error)
262 return error;
263
264 if ((iattr->ia_valid & ATTR_SIZE) &&
265 iattr->ia_size != i_size_read(inode)) {
266 error = vmtruncate(inode, iattr->ia_size);
267 if (error)
268 return error;
269 }
270
271 setattr_copy(inode, iattr);
272 mark_inode_dirty(inode);
273
274 de->uid = inode->i_uid;
275 de->gid = inode->i_gid;
276 de->mode = inode->i_mode;
277 return 0;
278}
279
280static int proc_getattr(struct vfsmount *mnt, struct dentry *dentry,
281 struct kstat *stat)
282{
283 struct inode *inode = dentry->d_inode;
284 struct proc_dir_entry *de = PROC_I(inode)->pde;
285 if (de && de->nlink)
286 inode->i_nlink = de->nlink;
287
288 generic_fillattr(inode, stat);
289 return 0;
290}
291
292static const struct inode_operations proc_file_inode_operations = {
293 .setattr = proc_notify_change,
294};
295
296/*
297 * This function parses a name such as "tty/driver/serial", and
298 * returns the struct proc_dir_entry for "/proc/tty/driver", and
299 * returns "serial" in residual.
300 */
301static int __xlate_proc_name(const char *name, struct proc_dir_entry **ret,
302 const char **residual)
303{
304 const char *cp = name, *next;
305 struct proc_dir_entry *de;
306 unsigned int len;
307
308 de = *ret;
309 if (!de)
310 de = &proc_root;
311
312 while (1) {
313 next = strchr(cp, '/');
314 if (!next)
315 break;
316
317 len = next - cp;
318 for (de = de->subdir; de ; de = de->next) {
319 if (proc_match(len, cp, de))
320 break;
321 }
322 if (!de) {
323 WARN(1, "name '%s'\n", name);
324 return -ENOENT;
325 }
326 cp += len + 1;
327 }
328 *residual = cp;
329 *ret = de;
330 return 0;
331}
332
333static int xlate_proc_name(const char *name, struct proc_dir_entry **ret,
334 const char **residual)
335{
336 int rv;
337
338 spin_lock(&proc_subdir_lock);
339 rv = __xlate_proc_name(name, ret, residual);
340 spin_unlock(&proc_subdir_lock);
341 return rv;
342}
343
344static DEFINE_IDA(proc_inum_ida);
345static DEFINE_SPINLOCK(proc_inum_lock); /* protects the above */
346
347#define PROC_DYNAMIC_FIRST 0xF0000000U
348
349/*
350 * Return an inode number between PROC_DYNAMIC_FIRST and
351 * 0xffffffff, or zero on failure.
352 */
353static unsigned int get_inode_number(void)
354{
355 unsigned int i;
356 int error;
357
358retry:
359 if (ida_pre_get(&proc_inum_ida, GFP_KERNEL) == 0)
360 return 0;
361
362 spin_lock(&proc_inum_lock);
363 error = ida_get_new(&proc_inum_ida, &i);
364 spin_unlock(&proc_inum_lock);
365 if (error == -EAGAIN)
366 goto retry;
367 else if (error)
368 return 0;
369
370 if (i > UINT_MAX - PROC_DYNAMIC_FIRST) {
371 spin_lock(&proc_inum_lock);
372 ida_remove(&proc_inum_ida, i);
373 spin_unlock(&proc_inum_lock);
374 return 0;
375 }
376 return PROC_DYNAMIC_FIRST + i;
377}
378
379static void release_inode_number(unsigned int inum)
380{
381 spin_lock(&proc_inum_lock);
382 ida_remove(&proc_inum_ida, inum - PROC_DYNAMIC_FIRST);
383 spin_unlock(&proc_inum_lock);
384}
385
386static void *proc_follow_link(struct dentry *dentry, struct nameidata *nd)
387{
388 nd_set_link(nd, PDE(dentry->d_inode)->data);
389 return NULL;
390}
391
392static const struct inode_operations proc_link_inode_operations = {
393 .readlink = generic_readlink,
394 .follow_link = proc_follow_link,
395};
396
397/*
398 * As some entries in /proc are volatile, we want to
399 * get rid of unused dentries. This could be made
400 * smarter: we could keep a "volatile" flag in the
401 * inode to indicate which ones to keep.
402 */
403static int proc_delete_dentry(const struct dentry * dentry)
404{
405 return 1;
406}
407
408static const struct dentry_operations proc_dentry_operations =
409{
410 .d_delete = proc_delete_dentry,
411};
412
413/*
414 * Don't create negative dentries here, return -ENOENT by hand
415 * instead.
416 */
417struct dentry *proc_lookup_de(struct proc_dir_entry *de, struct inode *dir,
418 struct dentry *dentry)
419{
420 struct inode *inode = NULL;
421 int error = -ENOENT;
422
423 spin_lock(&proc_subdir_lock);
424 for (de = de->subdir; de ; de = de->next) {
425 if (de->namelen != dentry->d_name.len)
426 continue;
427 if (!memcmp(dentry->d_name.name, de->name, de->namelen)) {
428 pde_get(de);
429 spin_unlock(&proc_subdir_lock);
430 error = -EINVAL;
431 inode = proc_get_inode(dir->i_sb, de);
432 goto out_unlock;
433 }
434 }
435 spin_unlock(&proc_subdir_lock);
436out_unlock:
437
438 if (inode) {
439 d_set_d_op(dentry, &proc_dentry_operations);
440 d_add(dentry, inode);
441 return NULL;
442 }
443 if (de)
444 pde_put(de);
445 return ERR_PTR(error);
446}
447
448struct dentry *proc_lookup(struct inode *dir, struct dentry *dentry,
449 struct nameidata *nd)
450{
451 return proc_lookup_de(PDE(dir), dir, dentry);
452}
453
454/*
455 * This returns non-zero if at EOF, so that the /proc
456 * root directory can use this and check if it should
457 * continue with the <pid> entries..
458 *
459 * Note that the VFS-layer doesn't care about the return
460 * value of the readdir() call, as long as it's non-negative
461 * for success..
462 */
463int proc_readdir_de(struct proc_dir_entry *de, struct file *filp, void *dirent,
464 filldir_t filldir)
465{
466 unsigned int ino;
467 int i;
468 struct inode *inode = filp->f_path.dentry->d_inode;
469 int ret = 0;
470
471 ino = inode->i_ino;
472 i = filp->f_pos;
473 switch (i) {
474 case 0:
475 if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
476 goto out;
477 i++;
478 filp->f_pos++;
479 /* fall through */
480 case 1:
481 if (filldir(dirent, "..", 2, i,
482 parent_ino(filp->f_path.dentry),
483 DT_DIR) < 0)
484 goto out;
485 i++;
486 filp->f_pos++;
487 /* fall through */
488 default:
489 spin_lock(&proc_subdir_lock);
490 de = de->subdir;
491 i -= 2;
492 for (;;) {
493 if (!de) {
494 ret = 1;
495 spin_unlock(&proc_subdir_lock);
496 goto out;
497 }
498 if (!i)
499 break;
500 de = de->next;
501 i--;
502 }
503
504 do {
505 struct proc_dir_entry *next;
506
507 /* filldir passes info to user space */
508 pde_get(de);
509 spin_unlock(&proc_subdir_lock);
510 if (filldir(dirent, de->name, de->namelen, filp->f_pos,
511 de->low_ino, de->mode >> 12) < 0) {
512 pde_put(de);
513 goto out;
514 }
515 spin_lock(&proc_subdir_lock);
516 filp->f_pos++;
517 next = de->next;
518 pde_put(de);
519 de = next;
520 } while (de);
521 spin_unlock(&proc_subdir_lock);
522 }
523 ret = 1;
524out:
525 return ret;
526}
527
528int proc_readdir(struct file *filp, void *dirent, filldir_t filldir)
529{
530 struct inode *inode = filp->f_path.dentry->d_inode;
531
532 return proc_readdir_de(PDE(inode), filp, dirent, filldir);
533}
534
535/*
536 * These are the generic /proc directory operations. They
537 * use the in-memory "struct proc_dir_entry" tree to parse
538 * the /proc directory.
539 */
540static const struct file_operations proc_dir_operations = {
541 .llseek = generic_file_llseek,
542 .read = generic_read_dir,
543 .readdir = proc_readdir,
544};
545
546/*
547 * proc directories can do almost nothing..
548 */
549static const struct inode_operations proc_dir_inode_operations = {
550 .lookup = proc_lookup,
551 .getattr = proc_getattr,
552 .setattr = proc_notify_change,
553};
554
555static int proc_register(struct proc_dir_entry * dir, struct proc_dir_entry * dp)
556{
557 unsigned int i;
558 struct proc_dir_entry *tmp;
559
560 i = get_inode_number();
561 if (i == 0)
562 return -EAGAIN;
563 dp->low_ino = i;
564
565 if (S_ISDIR(dp->mode)) {
566 if (dp->proc_iops == NULL) {
567 dp->proc_fops = &proc_dir_operations;
568 dp->proc_iops = &proc_dir_inode_operations;
569 }
570 dir->nlink++;
571 } else if (S_ISLNK(dp->mode)) {
572 if (dp->proc_iops == NULL)
573 dp->proc_iops = &proc_link_inode_operations;
574 } else if (S_ISREG(dp->mode)) {
575 if (dp->proc_fops == NULL)
576 dp->proc_fops = &proc_file_operations;
577 if (dp->proc_iops == NULL)
578 dp->proc_iops = &proc_file_inode_operations;
579 }
580
581 spin_lock(&proc_subdir_lock);
582
583 for (tmp = dir->subdir; tmp; tmp = tmp->next)
584 if (strcmp(tmp->name, dp->name) == 0) {
585 WARN(1, KERN_WARNING "proc_dir_entry '%s/%s' already registered\n",
586 dir->name, dp->name);
587 break;
588 }
589
590 dp->next = dir->subdir;
591 dp->parent = dir;
592 dir->subdir = dp;
593 spin_unlock(&proc_subdir_lock);
594
595 return 0;
596}
597
598static struct proc_dir_entry *__proc_create(struct proc_dir_entry **parent,
599 const char *name,
600 mode_t mode,
601 nlink_t nlink)
602{
603 struct proc_dir_entry *ent = NULL;
604 const char *fn = name;
605 unsigned int len;
606
607 /* make sure name is valid */
608 if (!name || !strlen(name)) goto out;
609
610 if (xlate_proc_name(name, parent, &fn) != 0)
611 goto out;
612
613 /* At this point there must not be any '/' characters beyond *fn */
614 if (strchr(fn, '/'))
615 goto out;
616
617 len = strlen(fn);
618
619 ent = kmalloc(sizeof(struct proc_dir_entry) + len + 1, GFP_KERNEL);
620 if (!ent) goto out;
621
622 memset(ent, 0, sizeof(struct proc_dir_entry));
623 memcpy(ent->name, fn, len + 1);
624 ent->namelen = len;
625 ent->mode = mode;
626 ent->nlink = nlink;
627 atomic_set(&ent->count, 1);
628 ent->pde_users = 0;
629 spin_lock_init(&ent->pde_unload_lock);
630 ent->pde_unload_completion = NULL;
631 INIT_LIST_HEAD(&ent->pde_openers);
632 out:
633 return ent;
634}
635
636struct proc_dir_entry *proc_symlink(const char *name,
637 struct proc_dir_entry *parent, const char *dest)
638{
639 struct proc_dir_entry *ent;
640
641 ent = __proc_create(&parent, name,
642 (S_IFLNK | S_IRUGO | S_IWUGO | S_IXUGO),1);
643
644 if (ent) {
645 ent->data = kmalloc((ent->size=strlen(dest))+1, GFP_KERNEL);
646 if (ent->data) {
647 strcpy((char*)ent->data,dest);
648 if (proc_register(parent, ent) < 0) {
649 kfree(ent->data);
650 kfree(ent);
651 ent = NULL;
652 }
653 } else {
654 kfree(ent);
655 ent = NULL;
656 }
657 }
658 return ent;
659}
660EXPORT_SYMBOL(proc_symlink);
661
662struct proc_dir_entry *proc_mkdir_mode(const char *name, mode_t mode,
663 struct proc_dir_entry *parent)
664{
665 struct proc_dir_entry *ent;
666
667 ent = __proc_create(&parent, name, S_IFDIR | mode, 2);
668 if (ent) {
669 if (proc_register(parent, ent) < 0) {
670 kfree(ent);
671 ent = NULL;
672 }
673 }
674 return ent;
675}
676EXPORT_SYMBOL(proc_mkdir_mode);
677
678struct proc_dir_entry *proc_net_mkdir(struct net *net, const char *name,
679 struct proc_dir_entry *parent)
680{
681 struct proc_dir_entry *ent;
682
683 ent = __proc_create(&parent, name, S_IFDIR | S_IRUGO | S_IXUGO, 2);
684 if (ent) {
685 ent->data = net;
686 if (proc_register(parent, ent) < 0) {
687 kfree(ent);
688 ent = NULL;
689 }
690 }
691 return ent;
692}
693EXPORT_SYMBOL_GPL(proc_net_mkdir);
694
695struct proc_dir_entry *proc_mkdir(const char *name,
696 struct proc_dir_entry *parent)
697{
698 return proc_mkdir_mode(name, S_IRUGO | S_IXUGO, parent);
699}
700EXPORT_SYMBOL(proc_mkdir);
701
702struct proc_dir_entry *create_proc_entry(const char *name, mode_t mode,
703 struct proc_dir_entry *parent)
704{
705 struct proc_dir_entry *ent;
706 nlink_t nlink;
707
708 if (S_ISDIR(mode)) {
709 if ((mode & S_IALLUGO) == 0)
710 mode |= S_IRUGO | S_IXUGO;
711 nlink = 2;
712 } else {
713 if ((mode & S_IFMT) == 0)
714 mode |= S_IFREG;
715 if ((mode & S_IALLUGO) == 0)
716 mode |= S_IRUGO;
717 nlink = 1;
718 }
719
720 ent = __proc_create(&parent, name, mode, nlink);
721 if (ent) {
722 if (proc_register(parent, ent) < 0) {
723 kfree(ent);
724 ent = NULL;
725 }
726 }
727 return ent;
728}
729EXPORT_SYMBOL(create_proc_entry);
730
731struct proc_dir_entry *proc_create_data(const char *name, mode_t mode,
732 struct proc_dir_entry *parent,
733 const struct file_operations *proc_fops,
734 void *data)
735{
736 struct proc_dir_entry *pde;
737 nlink_t nlink;
738
739 if (S_ISDIR(mode)) {
740 if ((mode & S_IALLUGO) == 0)
741 mode |= S_IRUGO | S_IXUGO;
742 nlink = 2;
743 } else {
744 if ((mode & S_IFMT) == 0)
745 mode |= S_IFREG;
746 if ((mode & S_IALLUGO) == 0)
747 mode |= S_IRUGO;
748 nlink = 1;
749 }
750
751 pde = __proc_create(&parent, name, mode, nlink);
752 if (!pde)
753 goto out;
754 pde->proc_fops = proc_fops;
755 pde->data = data;
756 if (proc_register(parent, pde) < 0)
757 goto out_free;
758 return pde;
759out_free:
760 kfree(pde);
761out:
762 return NULL;
763}
764EXPORT_SYMBOL(proc_create_data);
765
766static void free_proc_entry(struct proc_dir_entry *de)
767{
768 release_inode_number(de->low_ino);
769
770 if (S_ISLNK(de->mode))
771 kfree(de->data);
772 kfree(de);
773}
774
775void pde_put(struct proc_dir_entry *pde)
776{
777 if (atomic_dec_and_test(&pde->count))
778 free_proc_entry(pde);
779}
780
781/*
782 * Remove a /proc entry and free it if it's not currently in use.
783 */
784void remove_proc_entry(const char *name, struct proc_dir_entry *parent)
785{
786 struct proc_dir_entry **p;
787 struct proc_dir_entry *de = NULL;
788 const char *fn = name;
789 unsigned int len;
790
791 spin_lock(&proc_subdir_lock);
792 if (__xlate_proc_name(name, &parent, &fn) != 0) {
793 spin_unlock(&proc_subdir_lock);
794 return;
795 }
796 len = strlen(fn);
797
798 for (p = &parent->subdir; *p; p=&(*p)->next ) {
799 if (proc_match(len, fn, *p)) {
800 de = *p;
801 *p = de->next;
802 de->next = NULL;
803 break;
804 }
805 }
806 spin_unlock(&proc_subdir_lock);
807 if (!de) {
808 WARN(1, "name '%s'\n", name);
809 return;
810 }
811
812 spin_lock(&de->pde_unload_lock);
813 /*
814 * Stop accepting new callers into module. If you're
815 * dynamically allocating ->proc_fops, save a pointer somewhere.
816 */
817 de->proc_fops = NULL;
818 /* Wait until all existing callers into module are done. */
819 if (de->pde_users > 0) {
820 DECLARE_COMPLETION_ONSTACK(c);
821
822 if (!de->pde_unload_completion)
823 de->pde_unload_completion = &c;
824
825 spin_unlock(&de->pde_unload_lock);
826
827 wait_for_completion(de->pde_unload_completion);
828
829 spin_lock(&de->pde_unload_lock);
830 }
831
832 while (!list_empty(&de->pde_openers)) {
833 struct pde_opener *pdeo;
834
835 pdeo = list_first_entry(&de->pde_openers, struct pde_opener, lh);
836 list_del(&pdeo->lh);
837 spin_unlock(&de->pde_unload_lock);
838 pdeo->release(pdeo->inode, pdeo->file);
839 kfree(pdeo);
840 spin_lock(&de->pde_unload_lock);
841 }
842 spin_unlock(&de->pde_unload_lock);
843
844 if (S_ISDIR(de->mode))
845 parent->nlink--;
846 de->nlink = 0;
847 WARN(de->subdir, KERN_WARNING "%s: removing non-empty directory "
848 "'%s/%s', leaking at least '%s'\n", __func__,
849 de->parent->name, de->name, de->subdir->name);
850 pde_put(de);
851}
852EXPORT_SYMBOL(remove_proc_entry);
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/cache.h>
12#include <linux/errno.h>
13#include <linux/time.h>
14#include <linux/proc_fs.h>
15#include <linux/stat.h>
16#include <linux/mm.h>
17#include <linux/module.h>
18#include <linux/namei.h>
19#include <linux/slab.h>
20#include <linux/printk.h>
21#include <linux/mount.h>
22#include <linux/init.h>
23#include <linux/idr.h>
24#include <linux/bitops.h>
25#include <linux/spinlock.h>
26#include <linux/completion.h>
27#include <linux/uaccess.h>
28
29#include "internal.h"
30
31static DEFINE_RWLOCK(proc_subdir_lock);
32
33struct kmem_cache *proc_dir_entry_cache __ro_after_init;
34
35void pde_free(struct proc_dir_entry *pde)
36{
37 if (S_ISLNK(pde->mode))
38 kfree(pde->data);
39 if (pde->name != pde->inline_name)
40 kfree(pde->name);
41 kmem_cache_free(proc_dir_entry_cache, pde);
42}
43
44static int proc_match(const char *name, struct proc_dir_entry *de, unsigned int len)
45{
46 if (len < de->namelen)
47 return -1;
48 if (len > de->namelen)
49 return 1;
50
51 return memcmp(name, de->name, len);
52}
53
54static struct proc_dir_entry *pde_subdir_first(struct proc_dir_entry *dir)
55{
56 return rb_entry_safe(rb_first(&dir->subdir), struct proc_dir_entry,
57 subdir_node);
58}
59
60static struct proc_dir_entry *pde_subdir_next(struct proc_dir_entry *dir)
61{
62 return rb_entry_safe(rb_next(&dir->subdir_node), struct proc_dir_entry,
63 subdir_node);
64}
65
66static struct proc_dir_entry *pde_subdir_find(struct proc_dir_entry *dir,
67 const char *name,
68 unsigned int len)
69{
70 struct rb_node *node = dir->subdir.rb_node;
71
72 while (node) {
73 struct proc_dir_entry *de = rb_entry(node,
74 struct proc_dir_entry,
75 subdir_node);
76 int result = proc_match(name, de, len);
77
78 if (result < 0)
79 node = node->rb_left;
80 else if (result > 0)
81 node = node->rb_right;
82 else
83 return de;
84 }
85 return NULL;
86}
87
88static bool pde_subdir_insert(struct proc_dir_entry *dir,
89 struct proc_dir_entry *de)
90{
91 struct rb_root *root = &dir->subdir;
92 struct rb_node **new = &root->rb_node, *parent = NULL;
93
94 /* Figure out where to put new node */
95 while (*new) {
96 struct proc_dir_entry *this = rb_entry(*new,
97 struct proc_dir_entry,
98 subdir_node);
99 int result = proc_match(de->name, this, de->namelen);
100
101 parent = *new;
102 if (result < 0)
103 new = &(*new)->rb_left;
104 else if (result > 0)
105 new = &(*new)->rb_right;
106 else
107 return false;
108 }
109
110 /* Add new node and rebalance tree. */
111 rb_link_node(&de->subdir_node, parent, new);
112 rb_insert_color(&de->subdir_node, root);
113 return true;
114}
115
116static int proc_notify_change(struct dentry *dentry, struct iattr *iattr)
117{
118 struct inode *inode = d_inode(dentry);
119 struct proc_dir_entry *de = PDE(inode);
120 int error;
121
122 error = setattr_prepare(dentry, iattr);
123 if (error)
124 return error;
125
126 setattr_copy(inode, iattr);
127 mark_inode_dirty(inode);
128
129 proc_set_user(de, inode->i_uid, inode->i_gid);
130 de->mode = inode->i_mode;
131 return 0;
132}
133
134static int proc_getattr(const struct path *path, struct kstat *stat,
135 u32 request_mask, unsigned int query_flags)
136{
137 struct inode *inode = d_inode(path->dentry);
138 struct proc_dir_entry *de = PDE(inode);
139 if (de && de->nlink)
140 set_nlink(inode, de->nlink);
141
142 generic_fillattr(inode, stat);
143 return 0;
144}
145
146static const struct inode_operations proc_file_inode_operations = {
147 .setattr = proc_notify_change,
148};
149
150/*
151 * This function parses a name such as "tty/driver/serial", and
152 * returns the struct proc_dir_entry for "/proc/tty/driver", and
153 * returns "serial" in residual.
154 */
155static int __xlate_proc_name(const char *name, struct proc_dir_entry **ret,
156 const char **residual)
157{
158 const char *cp = name, *next;
159 struct proc_dir_entry *de;
160 unsigned int len;
161
162 de = *ret;
163 if (!de)
164 de = &proc_root;
165
166 while (1) {
167 next = strchr(cp, '/');
168 if (!next)
169 break;
170
171 len = next - cp;
172 de = pde_subdir_find(de, cp, len);
173 if (!de) {
174 WARN(1, "name '%s'\n", name);
175 return -ENOENT;
176 }
177 cp += len + 1;
178 }
179 *residual = cp;
180 *ret = de;
181 return 0;
182}
183
184static int xlate_proc_name(const char *name, struct proc_dir_entry **ret,
185 const char **residual)
186{
187 int rv;
188
189 read_lock(&proc_subdir_lock);
190 rv = __xlate_proc_name(name, ret, residual);
191 read_unlock(&proc_subdir_lock);
192 return rv;
193}
194
195static DEFINE_IDA(proc_inum_ida);
196
197#define PROC_DYNAMIC_FIRST 0xF0000000U
198
199/*
200 * Return an inode number between PROC_DYNAMIC_FIRST and
201 * 0xffffffff, or zero on failure.
202 */
203int proc_alloc_inum(unsigned int *inum)
204{
205 int i;
206
207 i = ida_simple_get(&proc_inum_ida, 0, UINT_MAX - PROC_DYNAMIC_FIRST + 1,
208 GFP_KERNEL);
209 if (i < 0)
210 return i;
211
212 *inum = PROC_DYNAMIC_FIRST + (unsigned int)i;
213 return 0;
214}
215
216void proc_free_inum(unsigned int inum)
217{
218 ida_simple_remove(&proc_inum_ida, inum - PROC_DYNAMIC_FIRST);
219}
220
221static int proc_misc_d_revalidate(struct dentry *dentry, unsigned int flags)
222{
223 if (flags & LOOKUP_RCU)
224 return -ECHILD;
225
226 if (atomic_read(&PDE(d_inode(dentry))->in_use) < 0)
227 return 0; /* revalidate */
228 return 1;
229}
230
231static int proc_misc_d_delete(const struct dentry *dentry)
232{
233 return atomic_read(&PDE(d_inode(dentry))->in_use) < 0;
234}
235
236static const struct dentry_operations proc_misc_dentry_ops = {
237 .d_revalidate = proc_misc_d_revalidate,
238 .d_delete = proc_misc_d_delete,
239};
240
241/*
242 * Don't create negative dentries here, return -ENOENT by hand
243 * instead.
244 */
245struct dentry *proc_lookup_de(struct inode *dir, struct dentry *dentry,
246 struct proc_dir_entry *de)
247{
248 struct inode *inode;
249
250 read_lock(&proc_subdir_lock);
251 de = pde_subdir_find(de, dentry->d_name.name, dentry->d_name.len);
252 if (de) {
253 pde_get(de);
254 read_unlock(&proc_subdir_lock);
255 inode = proc_get_inode(dir->i_sb, de);
256 if (!inode)
257 return ERR_PTR(-ENOMEM);
258 d_set_d_op(dentry, &proc_misc_dentry_ops);
259 d_add(dentry, inode);
260 return NULL;
261 }
262 read_unlock(&proc_subdir_lock);
263 return ERR_PTR(-ENOENT);
264}
265
266struct dentry *proc_lookup(struct inode *dir, struct dentry *dentry,
267 unsigned int flags)
268{
269 return proc_lookup_de(dir, dentry, PDE(dir));
270}
271
272/*
273 * This returns non-zero if at EOF, so that the /proc
274 * root directory can use this and check if it should
275 * continue with the <pid> entries..
276 *
277 * Note that the VFS-layer doesn't care about the return
278 * value of the readdir() call, as long as it's non-negative
279 * for success..
280 */
281int proc_readdir_de(struct file *file, struct dir_context *ctx,
282 struct proc_dir_entry *de)
283{
284 int i;
285
286 if (!dir_emit_dots(file, ctx))
287 return 0;
288
289 read_lock(&proc_subdir_lock);
290 de = pde_subdir_first(de);
291 i = ctx->pos - 2;
292 for (;;) {
293 if (!de) {
294 read_unlock(&proc_subdir_lock);
295 return 0;
296 }
297 if (!i)
298 break;
299 de = pde_subdir_next(de);
300 i--;
301 }
302
303 do {
304 struct proc_dir_entry *next;
305 pde_get(de);
306 read_unlock(&proc_subdir_lock);
307 if (!dir_emit(ctx, de->name, de->namelen,
308 de->low_ino, de->mode >> 12)) {
309 pde_put(de);
310 return 0;
311 }
312 read_lock(&proc_subdir_lock);
313 ctx->pos++;
314 next = pde_subdir_next(de);
315 pde_put(de);
316 de = next;
317 } while (de);
318 read_unlock(&proc_subdir_lock);
319 return 1;
320}
321
322int proc_readdir(struct file *file, struct dir_context *ctx)
323{
324 struct inode *inode = file_inode(file);
325
326 return proc_readdir_de(file, ctx, PDE(inode));
327}
328
329/*
330 * These are the generic /proc directory operations. They
331 * use the in-memory "struct proc_dir_entry" tree to parse
332 * the /proc directory.
333 */
334static const struct file_operations proc_dir_operations = {
335 .llseek = generic_file_llseek,
336 .read = generic_read_dir,
337 .iterate_shared = proc_readdir,
338};
339
340/*
341 * proc directories can do almost nothing..
342 */
343static const struct inode_operations proc_dir_inode_operations = {
344 .lookup = proc_lookup,
345 .getattr = proc_getattr,
346 .setattr = proc_notify_change,
347};
348
349static int proc_register(struct proc_dir_entry * dir, struct proc_dir_entry * dp)
350{
351 int ret;
352
353 ret = proc_alloc_inum(&dp->low_ino);
354 if (ret)
355 return ret;
356
357 write_lock(&proc_subdir_lock);
358 dp->parent = dir;
359 if (pde_subdir_insert(dir, dp) == false) {
360 WARN(1, "proc_dir_entry '%s/%s' already registered\n",
361 dir->name, dp->name);
362 write_unlock(&proc_subdir_lock);
363 proc_free_inum(dp->low_ino);
364 return -EEXIST;
365 }
366 write_unlock(&proc_subdir_lock);
367
368 return 0;
369}
370
371static struct proc_dir_entry *__proc_create(struct proc_dir_entry **parent,
372 const char *name,
373 umode_t mode,
374 nlink_t nlink)
375{
376 struct proc_dir_entry *ent = NULL;
377 const char *fn;
378 struct qstr qstr;
379
380 if (xlate_proc_name(name, parent, &fn) != 0)
381 goto out;
382 qstr.name = fn;
383 qstr.len = strlen(fn);
384 if (qstr.len == 0 || qstr.len >= 256) {
385 WARN(1, "name len %u\n", qstr.len);
386 return NULL;
387 }
388 if (qstr.len == 1 && fn[0] == '.') {
389 WARN(1, "name '.'\n");
390 return NULL;
391 }
392 if (qstr.len == 2 && fn[0] == '.' && fn[1] == '.') {
393 WARN(1, "name '..'\n");
394 return NULL;
395 }
396 if (*parent == &proc_root && name_to_int(&qstr) != ~0U) {
397 WARN(1, "create '/proc/%s' by hand\n", qstr.name);
398 return NULL;
399 }
400 if (is_empty_pde(*parent)) {
401 WARN(1, "attempt to add to permanently empty directory");
402 return NULL;
403 }
404
405 ent = kmem_cache_zalloc(proc_dir_entry_cache, GFP_KERNEL);
406 if (!ent)
407 goto out;
408
409 if (qstr.len + 1 <= sizeof(ent->inline_name)) {
410 ent->name = ent->inline_name;
411 } else {
412 ent->name = kmalloc(qstr.len + 1, GFP_KERNEL);
413 if (!ent->name) {
414 pde_free(ent);
415 return NULL;
416 }
417 }
418
419 memcpy(ent->name, fn, qstr.len + 1);
420 ent->namelen = qstr.len;
421 ent->mode = mode;
422 ent->nlink = nlink;
423 ent->subdir = RB_ROOT;
424 refcount_set(&ent->refcnt, 1);
425 spin_lock_init(&ent->pde_unload_lock);
426 INIT_LIST_HEAD(&ent->pde_openers);
427 proc_set_user(ent, (*parent)->uid, (*parent)->gid);
428
429out:
430 return ent;
431}
432
433struct proc_dir_entry *proc_symlink(const char *name,
434 struct proc_dir_entry *parent, const char *dest)
435{
436 struct proc_dir_entry *ent;
437
438 ent = __proc_create(&parent, name,
439 (S_IFLNK | S_IRUGO | S_IWUGO | S_IXUGO),1);
440
441 if (ent) {
442 ent->data = kmalloc((ent->size=strlen(dest))+1, GFP_KERNEL);
443 if (ent->data) {
444 strcpy((char*)ent->data,dest);
445 ent->proc_iops = &proc_link_inode_operations;
446 if (proc_register(parent, ent) < 0) {
447 pde_free(ent);
448 ent = NULL;
449 }
450 } else {
451 pde_free(ent);
452 ent = NULL;
453 }
454 }
455 return ent;
456}
457EXPORT_SYMBOL(proc_symlink);
458
459struct proc_dir_entry *proc_mkdir_data(const char *name, umode_t mode,
460 struct proc_dir_entry *parent, void *data)
461{
462 struct proc_dir_entry *ent;
463
464 if (mode == 0)
465 mode = S_IRUGO | S_IXUGO;
466
467 ent = __proc_create(&parent, name, S_IFDIR | mode, 2);
468 if (ent) {
469 ent->data = data;
470 ent->proc_fops = &proc_dir_operations;
471 ent->proc_iops = &proc_dir_inode_operations;
472 parent->nlink++;
473 if (proc_register(parent, ent) < 0) {
474 pde_free(ent);
475 parent->nlink--;
476 ent = NULL;
477 }
478 }
479 return ent;
480}
481EXPORT_SYMBOL_GPL(proc_mkdir_data);
482
483struct proc_dir_entry *proc_mkdir_mode(const char *name, umode_t mode,
484 struct proc_dir_entry *parent)
485{
486 return proc_mkdir_data(name, mode, parent, NULL);
487}
488EXPORT_SYMBOL(proc_mkdir_mode);
489
490struct proc_dir_entry *proc_mkdir(const char *name,
491 struct proc_dir_entry *parent)
492{
493 return proc_mkdir_data(name, 0, parent, NULL);
494}
495EXPORT_SYMBOL(proc_mkdir);
496
497struct proc_dir_entry *proc_create_mount_point(const char *name)
498{
499 umode_t mode = S_IFDIR | S_IRUGO | S_IXUGO;
500 struct proc_dir_entry *ent, *parent = NULL;
501
502 ent = __proc_create(&parent, name, mode, 2);
503 if (ent) {
504 ent->data = NULL;
505 ent->proc_fops = NULL;
506 ent->proc_iops = NULL;
507 parent->nlink++;
508 if (proc_register(parent, ent) < 0) {
509 pde_free(ent);
510 parent->nlink--;
511 ent = NULL;
512 }
513 }
514 return ent;
515}
516EXPORT_SYMBOL(proc_create_mount_point);
517
518struct proc_dir_entry *proc_create_data(const char *name, umode_t mode,
519 struct proc_dir_entry *parent,
520 const struct file_operations *proc_fops,
521 void *data)
522{
523 struct proc_dir_entry *pde;
524 if ((mode & S_IFMT) == 0)
525 mode |= S_IFREG;
526
527 if (!S_ISREG(mode)) {
528 WARN_ON(1); /* use proc_mkdir() */
529 return NULL;
530 }
531
532 BUG_ON(proc_fops == NULL);
533
534 if ((mode & S_IALLUGO) == 0)
535 mode |= S_IRUGO;
536 pde = __proc_create(&parent, name, mode, 1);
537 if (!pde)
538 goto out;
539 pde->proc_fops = proc_fops;
540 pde->data = data;
541 pde->proc_iops = &proc_file_inode_operations;
542 if (proc_register(parent, pde) < 0)
543 goto out_free;
544 return pde;
545out_free:
546 pde_free(pde);
547out:
548 return NULL;
549}
550EXPORT_SYMBOL(proc_create_data);
551
552struct proc_dir_entry *proc_create(const char *name, umode_t mode,
553 struct proc_dir_entry *parent,
554 const struct file_operations *proc_fops)
555{
556 return proc_create_data(name, mode, parent, proc_fops, NULL);
557}
558EXPORT_SYMBOL(proc_create);
559
560void proc_set_size(struct proc_dir_entry *de, loff_t size)
561{
562 de->size = size;
563}
564EXPORT_SYMBOL(proc_set_size);
565
566void proc_set_user(struct proc_dir_entry *de, kuid_t uid, kgid_t gid)
567{
568 de->uid = uid;
569 de->gid = gid;
570}
571EXPORT_SYMBOL(proc_set_user);
572
573void pde_put(struct proc_dir_entry *pde)
574{
575 if (refcount_dec_and_test(&pde->refcnt)) {
576 proc_free_inum(pde->low_ino);
577 pde_free(pde);
578 }
579}
580
581/*
582 * Remove a /proc entry and free it if it's not currently in use.
583 */
584void remove_proc_entry(const char *name, struct proc_dir_entry *parent)
585{
586 struct proc_dir_entry *de = NULL;
587 const char *fn = name;
588 unsigned int len;
589
590 write_lock(&proc_subdir_lock);
591 if (__xlate_proc_name(name, &parent, &fn) != 0) {
592 write_unlock(&proc_subdir_lock);
593 return;
594 }
595 len = strlen(fn);
596
597 de = pde_subdir_find(parent, fn, len);
598 if (de)
599 rb_erase(&de->subdir_node, &parent->subdir);
600 write_unlock(&proc_subdir_lock);
601 if (!de) {
602 WARN(1, "name '%s'\n", name);
603 return;
604 }
605
606 proc_entry_rundown(de);
607
608 if (S_ISDIR(de->mode))
609 parent->nlink--;
610 de->nlink = 0;
611 WARN(pde_subdir_first(de),
612 "%s: removing non-empty directory '%s/%s', leaking at least '%s'\n",
613 __func__, de->parent->name, de->name, pde_subdir_first(de)->name);
614 pde_put(de);
615}
616EXPORT_SYMBOL(remove_proc_entry);
617
618int remove_proc_subtree(const char *name, struct proc_dir_entry *parent)
619{
620 struct proc_dir_entry *root = NULL, *de, *next;
621 const char *fn = name;
622 unsigned int len;
623
624 write_lock(&proc_subdir_lock);
625 if (__xlate_proc_name(name, &parent, &fn) != 0) {
626 write_unlock(&proc_subdir_lock);
627 return -ENOENT;
628 }
629 len = strlen(fn);
630
631 root = pde_subdir_find(parent, fn, len);
632 if (!root) {
633 write_unlock(&proc_subdir_lock);
634 return -ENOENT;
635 }
636 rb_erase(&root->subdir_node, &parent->subdir);
637
638 de = root;
639 while (1) {
640 next = pde_subdir_first(de);
641 if (next) {
642 rb_erase(&next->subdir_node, &de->subdir);
643 de = next;
644 continue;
645 }
646 write_unlock(&proc_subdir_lock);
647
648 proc_entry_rundown(de);
649 next = de->parent;
650 if (S_ISDIR(de->mode))
651 next->nlink--;
652 de->nlink = 0;
653 if (de == root)
654 break;
655 pde_put(de);
656
657 write_lock(&proc_subdir_lock);
658 de = next;
659 }
660 pde_put(root);
661 return 0;
662}
663EXPORT_SYMBOL(remove_proc_subtree);
664
665void *proc_get_parent_data(const struct inode *inode)
666{
667 struct proc_dir_entry *de = PDE(inode);
668 return de->parent->data;
669}
670EXPORT_SYMBOL_GPL(proc_get_parent_data);
671
672void proc_remove(struct proc_dir_entry *de)
673{
674 if (de)
675 remove_proc_subtree(de->name, de->parent);
676}
677EXPORT_SYMBOL(proc_remove);
678
679void *PDE_DATA(const struct inode *inode)
680{
681 return __PDE_DATA(inode);
682}
683EXPORT_SYMBOL(PDE_DATA);