Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * proc/fs/generic.c --- generic routines for the proc-fs
4 *
5 * This file contains generic proc-fs routines for handling
6 * directories and files.
7 *
8 * Copyright (C) 1991, 1992 Linus Torvalds.
9 * Copyright (C) 1997 Theodore Ts'o
10 */
11
12#include <linux/cache.h>
13#include <linux/errno.h>
14#include <linux/time.h>
15#include <linux/proc_fs.h>
16#include <linux/stat.h>
17#include <linux/mm.h>
18#include <linux/module.h>
19#include <linux/namei.h>
20#include <linux/slab.h>
21#include <linux/printk.h>
22#include <linux/mount.h>
23#include <linux/init.h>
24#include <linux/idr.h>
25#include <linux/bitops.h>
26#include <linux/spinlock.h>
27#include <linux/completion.h>
28#include <linux/uaccess.h>
29#include <linux/seq_file.h>
30
31#include "internal.h"
32
33static DEFINE_RWLOCK(proc_subdir_lock);
34
35struct kmem_cache *proc_dir_entry_cache __ro_after_init;
36
37void pde_free(struct proc_dir_entry *pde)
38{
39 if (S_ISLNK(pde->mode))
40 kfree(pde->data);
41 if (pde->name != pde->inline_name)
42 kfree(pde->name);
43 kmem_cache_free(proc_dir_entry_cache, pde);
44}
45
46static int proc_match(const char *name, struct proc_dir_entry *de, unsigned int len)
47{
48 if (len < de->namelen)
49 return -1;
50 if (len > de->namelen)
51 return 1;
52
53 return memcmp(name, de->name, len);
54}
55
56static struct proc_dir_entry *pde_subdir_first(struct proc_dir_entry *dir)
57{
58 return rb_entry_safe(rb_first(&dir->subdir), struct proc_dir_entry,
59 subdir_node);
60}
61
62static struct proc_dir_entry *pde_subdir_next(struct proc_dir_entry *dir)
63{
64 return rb_entry_safe(rb_next(&dir->subdir_node), struct proc_dir_entry,
65 subdir_node);
66}
67
68static struct proc_dir_entry *pde_subdir_find(struct proc_dir_entry *dir,
69 const char *name,
70 unsigned int len)
71{
72 struct rb_node *node = dir->subdir.rb_node;
73
74 while (node) {
75 struct proc_dir_entry *de = rb_entry(node,
76 struct proc_dir_entry,
77 subdir_node);
78 int result = proc_match(name, de, len);
79
80 if (result < 0)
81 node = node->rb_left;
82 else if (result > 0)
83 node = node->rb_right;
84 else
85 return de;
86 }
87 return NULL;
88}
89
90static bool pde_subdir_insert(struct proc_dir_entry *dir,
91 struct proc_dir_entry *de)
92{
93 struct rb_root *root = &dir->subdir;
94 struct rb_node **new = &root->rb_node, *parent = NULL;
95
96 /* Figure out where to put new node */
97 while (*new) {
98 struct proc_dir_entry *this = rb_entry(*new,
99 struct proc_dir_entry,
100 subdir_node);
101 int result = proc_match(de->name, this, de->namelen);
102
103 parent = *new;
104 if (result < 0)
105 new = &(*new)->rb_left;
106 else if (result > 0)
107 new = &(*new)->rb_right;
108 else
109 return false;
110 }
111
112 /* Add new node and rebalance tree. */
113 rb_link_node(&de->subdir_node, parent, new);
114 rb_insert_color(&de->subdir_node, root);
115 return true;
116}
117
118static int proc_notify_change(struct user_namespace *mnt_userns,
119 struct dentry *dentry, struct iattr *iattr)
120{
121 struct inode *inode = d_inode(dentry);
122 struct proc_dir_entry *de = PDE(inode);
123 int error;
124
125 error = setattr_prepare(&init_user_ns, dentry, iattr);
126 if (error)
127 return error;
128
129 setattr_copy(&init_user_ns, inode, iattr);
130 mark_inode_dirty(inode);
131
132 proc_set_user(de, inode->i_uid, inode->i_gid);
133 de->mode = inode->i_mode;
134 return 0;
135}
136
137static int proc_getattr(struct user_namespace *mnt_userns,
138 const struct path *path, struct kstat *stat,
139 u32 request_mask, unsigned int query_flags)
140{
141 struct inode *inode = d_inode(path->dentry);
142 struct proc_dir_entry *de = PDE(inode);
143 if (de) {
144 nlink_t nlink = READ_ONCE(de->nlink);
145 if (nlink > 0) {
146 set_nlink(inode, nlink);
147 }
148 }
149
150 generic_fillattr(&init_user_ns, inode, stat);
151 return 0;
152}
153
154static const struct inode_operations proc_file_inode_operations = {
155 .setattr = proc_notify_change,
156};
157
158/*
159 * This function parses a name such as "tty/driver/serial", and
160 * returns the struct proc_dir_entry for "/proc/tty/driver", and
161 * returns "serial" in residual.
162 */
163static int __xlate_proc_name(const char *name, struct proc_dir_entry **ret,
164 const char **residual)
165{
166 const char *cp = name, *next;
167 struct proc_dir_entry *de;
168
169 de = *ret ?: &proc_root;
170 while ((next = strchr(cp, '/')) != NULL) {
171 de = pde_subdir_find(de, cp, next - cp);
172 if (!de) {
173 WARN(1, "name '%s'\n", name);
174 return -ENOENT;
175 }
176 cp = next + 1;
177 }
178 *residual = cp;
179 *ret = de;
180 return 0;
181}
182
183static int xlate_proc_name(const char *name, struct proc_dir_entry **ret,
184 const char **residual)
185{
186 int rv;
187
188 read_lock(&proc_subdir_lock);
189 rv = __xlate_proc_name(name, ret, residual);
190 read_unlock(&proc_subdir_lock);
191 return rv;
192}
193
194static DEFINE_IDA(proc_inum_ida);
195
196#define PROC_DYNAMIC_FIRST 0xF0000000U
197
198/*
199 * Return an inode number between PROC_DYNAMIC_FIRST and
200 * 0xffffffff, or zero on failure.
201 */
202int proc_alloc_inum(unsigned int *inum)
203{
204 int i;
205
206 i = ida_simple_get(&proc_inum_ida, 0, UINT_MAX - PROC_DYNAMIC_FIRST + 1,
207 GFP_KERNEL);
208 if (i < 0)
209 return i;
210
211 *inum = PROC_DYNAMIC_FIRST + (unsigned int)i;
212 return 0;
213}
214
215void proc_free_inum(unsigned int inum)
216{
217 ida_simple_remove(&proc_inum_ida, inum - PROC_DYNAMIC_FIRST);
218}
219
220static int proc_misc_d_revalidate(struct dentry *dentry, unsigned int flags)
221{
222 if (flags & LOOKUP_RCU)
223 return -ECHILD;
224
225 if (atomic_read(&PDE(d_inode(dentry))->in_use) < 0)
226 return 0; /* revalidate */
227 return 1;
228}
229
230static int proc_misc_d_delete(const struct dentry *dentry)
231{
232 return atomic_read(&PDE(d_inode(dentry))->in_use) < 0;
233}
234
235static const struct dentry_operations proc_misc_dentry_ops = {
236 .d_revalidate = proc_misc_d_revalidate,
237 .d_delete = proc_misc_d_delete,
238};
239
240/*
241 * Don't create negative dentries here, return -ENOENT by hand
242 * instead.
243 */
244struct dentry *proc_lookup_de(struct inode *dir, struct dentry *dentry,
245 struct proc_dir_entry *de)
246{
247 struct inode *inode;
248
249 read_lock(&proc_subdir_lock);
250 de = pde_subdir_find(de, dentry->d_name.name, dentry->d_name.len);
251 if (de) {
252 pde_get(de);
253 read_unlock(&proc_subdir_lock);
254 inode = proc_get_inode(dir->i_sb, de);
255 if (!inode)
256 return ERR_PTR(-ENOMEM);
257 d_set_d_op(dentry, de->proc_dops);
258 return d_splice_alias(inode, dentry);
259 }
260 read_unlock(&proc_subdir_lock);
261 return ERR_PTR(-ENOENT);
262}
263
264struct dentry *proc_lookup(struct inode *dir, struct dentry *dentry,
265 unsigned int flags)
266{
267 struct proc_fs_info *fs_info = proc_sb_info(dir->i_sb);
268
269 if (fs_info->pidonly == PROC_PIDONLY_ON)
270 return ERR_PTR(-ENOENT);
271
272 return proc_lookup_de(dir, dentry, PDE(dir));
273}
274
275/*
276 * This returns non-zero if at EOF, so that the /proc
277 * root directory can use this and check if it should
278 * continue with the <pid> entries..
279 *
280 * Note that the VFS-layer doesn't care about the return
281 * value of the readdir() call, as long as it's non-negative
282 * for success..
283 */
284int proc_readdir_de(struct file *file, struct dir_context *ctx,
285 struct proc_dir_entry *de)
286{
287 int i;
288
289 if (!dir_emit_dots(file, ctx))
290 return 0;
291
292 i = ctx->pos - 2;
293 read_lock(&proc_subdir_lock);
294 de = pde_subdir_first(de);
295 for (;;) {
296 if (!de) {
297 read_unlock(&proc_subdir_lock);
298 return 0;
299 }
300 if (!i)
301 break;
302 de = pde_subdir_next(de);
303 i--;
304 }
305
306 do {
307 struct proc_dir_entry *next;
308 pde_get(de);
309 read_unlock(&proc_subdir_lock);
310 if (!dir_emit(ctx, de->name, de->namelen,
311 de->low_ino, de->mode >> 12)) {
312 pde_put(de);
313 return 0;
314 }
315 ctx->pos++;
316 read_lock(&proc_subdir_lock);
317 next = pde_subdir_next(de);
318 pde_put(de);
319 de = next;
320 } while (de);
321 read_unlock(&proc_subdir_lock);
322 return 1;
323}
324
325int proc_readdir(struct file *file, struct dir_context *ctx)
326{
327 struct inode *inode = file_inode(file);
328 struct proc_fs_info *fs_info = proc_sb_info(inode->i_sb);
329
330 if (fs_info->pidonly == PROC_PIDONLY_ON)
331 return 1;
332
333 return proc_readdir_de(file, ctx, PDE(inode));
334}
335
336/*
337 * These are the generic /proc directory operations. They
338 * use the in-memory "struct proc_dir_entry" tree to parse
339 * the /proc directory.
340 */
341static const struct file_operations proc_dir_operations = {
342 .llseek = generic_file_llseek,
343 .read = generic_read_dir,
344 .iterate_shared = proc_readdir,
345};
346
347static int proc_net_d_revalidate(struct dentry *dentry, unsigned int flags)
348{
349 return 0;
350}
351
352const struct dentry_operations proc_net_dentry_ops = {
353 .d_revalidate = proc_net_d_revalidate,
354 .d_delete = always_delete_dentry,
355};
356
357/*
358 * proc directories can do almost nothing..
359 */
360static const struct inode_operations proc_dir_inode_operations = {
361 .lookup = proc_lookup,
362 .getattr = proc_getattr,
363 .setattr = proc_notify_change,
364};
365
366/* returns the registered entry, or frees dp and returns NULL on failure */
367struct proc_dir_entry *proc_register(struct proc_dir_entry *dir,
368 struct proc_dir_entry *dp)
369{
370 if (proc_alloc_inum(&dp->low_ino))
371 goto out_free_entry;
372
373 write_lock(&proc_subdir_lock);
374 dp->parent = dir;
375 if (pde_subdir_insert(dir, dp) == false) {
376 WARN(1, "proc_dir_entry '%s/%s' already registered\n",
377 dir->name, dp->name);
378 write_unlock(&proc_subdir_lock);
379 goto out_free_inum;
380 }
381 dir->nlink++;
382 write_unlock(&proc_subdir_lock);
383
384 return dp;
385out_free_inum:
386 proc_free_inum(dp->low_ino);
387out_free_entry:
388 pde_free(dp);
389 return NULL;
390}
391
392static struct proc_dir_entry *__proc_create(struct proc_dir_entry **parent,
393 const char *name,
394 umode_t mode,
395 nlink_t nlink)
396{
397 struct proc_dir_entry *ent = NULL;
398 const char *fn;
399 struct qstr qstr;
400
401 if (xlate_proc_name(name, parent, &fn) != 0)
402 goto out;
403 qstr.name = fn;
404 qstr.len = strlen(fn);
405 if (qstr.len == 0 || qstr.len >= 256) {
406 WARN(1, "name len %u\n", qstr.len);
407 return NULL;
408 }
409 if (qstr.len == 1 && fn[0] == '.') {
410 WARN(1, "name '.'\n");
411 return NULL;
412 }
413 if (qstr.len == 2 && fn[0] == '.' && fn[1] == '.') {
414 WARN(1, "name '..'\n");
415 return NULL;
416 }
417 if (*parent == &proc_root && name_to_int(&qstr) != ~0U) {
418 WARN(1, "create '/proc/%s' by hand\n", qstr.name);
419 return NULL;
420 }
421 if (is_empty_pde(*parent)) {
422 WARN(1, "attempt to add to permanently empty directory");
423 return NULL;
424 }
425
426 ent = kmem_cache_zalloc(proc_dir_entry_cache, GFP_KERNEL);
427 if (!ent)
428 goto out;
429
430 if (qstr.len + 1 <= SIZEOF_PDE_INLINE_NAME) {
431 ent->name = ent->inline_name;
432 } else {
433 ent->name = kmalloc(qstr.len + 1, GFP_KERNEL);
434 if (!ent->name) {
435 pde_free(ent);
436 return NULL;
437 }
438 }
439
440 memcpy(ent->name, fn, qstr.len + 1);
441 ent->namelen = qstr.len;
442 ent->mode = mode;
443 ent->nlink = nlink;
444 ent->subdir = RB_ROOT;
445 refcount_set(&ent->refcnt, 1);
446 spin_lock_init(&ent->pde_unload_lock);
447 INIT_LIST_HEAD(&ent->pde_openers);
448 proc_set_user(ent, (*parent)->uid, (*parent)->gid);
449
450 ent->proc_dops = &proc_misc_dentry_ops;
451 /* Revalidate everything under /proc/${pid}/net */
452 if ((*parent)->proc_dops == &proc_net_dentry_ops)
453 pde_force_lookup(ent);
454
455out:
456 return ent;
457}
458
459struct proc_dir_entry *proc_symlink(const char *name,
460 struct proc_dir_entry *parent, const char *dest)
461{
462 struct proc_dir_entry *ent;
463
464 ent = __proc_create(&parent, name,
465 (S_IFLNK | S_IRUGO | S_IWUGO | S_IXUGO),1);
466
467 if (ent) {
468 ent->data = kmalloc((ent->size=strlen(dest))+1, GFP_KERNEL);
469 if (ent->data) {
470 strcpy((char*)ent->data,dest);
471 ent->proc_iops = &proc_link_inode_operations;
472 ent = proc_register(parent, ent);
473 } else {
474 pde_free(ent);
475 ent = NULL;
476 }
477 }
478 return ent;
479}
480EXPORT_SYMBOL(proc_symlink);
481
482struct proc_dir_entry *_proc_mkdir(const char *name, umode_t mode,
483 struct proc_dir_entry *parent, void *data, bool force_lookup)
484{
485 struct proc_dir_entry *ent;
486
487 if (mode == 0)
488 mode = S_IRUGO | S_IXUGO;
489
490 ent = __proc_create(&parent, name, S_IFDIR | mode, 2);
491 if (ent) {
492 ent->data = data;
493 ent->proc_dir_ops = &proc_dir_operations;
494 ent->proc_iops = &proc_dir_inode_operations;
495 if (force_lookup) {
496 pde_force_lookup(ent);
497 }
498 ent = proc_register(parent, ent);
499 }
500 return ent;
501}
502EXPORT_SYMBOL_GPL(_proc_mkdir);
503
504struct proc_dir_entry *proc_mkdir_data(const char *name, umode_t mode,
505 struct proc_dir_entry *parent, void *data)
506{
507 return _proc_mkdir(name, mode, parent, data, false);
508}
509EXPORT_SYMBOL_GPL(proc_mkdir_data);
510
511struct proc_dir_entry *proc_mkdir_mode(const char *name, umode_t mode,
512 struct proc_dir_entry *parent)
513{
514 return proc_mkdir_data(name, mode, parent, NULL);
515}
516EXPORT_SYMBOL(proc_mkdir_mode);
517
518struct proc_dir_entry *proc_mkdir(const char *name,
519 struct proc_dir_entry *parent)
520{
521 return proc_mkdir_data(name, 0, parent, NULL);
522}
523EXPORT_SYMBOL(proc_mkdir);
524
525struct proc_dir_entry *proc_create_mount_point(const char *name)
526{
527 umode_t mode = S_IFDIR | S_IRUGO | S_IXUGO;
528 struct proc_dir_entry *ent, *parent = NULL;
529
530 ent = __proc_create(&parent, name, mode, 2);
531 if (ent) {
532 ent->data = NULL;
533 ent->proc_dir_ops = NULL;
534 ent->proc_iops = NULL;
535 ent = proc_register(parent, ent);
536 }
537 return ent;
538}
539EXPORT_SYMBOL(proc_create_mount_point);
540
541struct proc_dir_entry *proc_create_reg(const char *name, umode_t mode,
542 struct proc_dir_entry **parent, void *data)
543{
544 struct proc_dir_entry *p;
545
546 if ((mode & S_IFMT) == 0)
547 mode |= S_IFREG;
548 if ((mode & S_IALLUGO) == 0)
549 mode |= S_IRUGO;
550 if (WARN_ON_ONCE(!S_ISREG(mode)))
551 return NULL;
552
553 p = __proc_create(parent, name, mode, 1);
554 if (p) {
555 p->proc_iops = &proc_file_inode_operations;
556 p->data = data;
557 }
558 return p;
559}
560
561static inline void pde_set_flags(struct proc_dir_entry *pde)
562{
563 if (pde->proc_ops->proc_flags & PROC_ENTRY_PERMANENT)
564 pde->flags |= PROC_ENTRY_PERMANENT;
565}
566
567struct proc_dir_entry *proc_create_data(const char *name, umode_t mode,
568 struct proc_dir_entry *parent,
569 const struct proc_ops *proc_ops, void *data)
570{
571 struct proc_dir_entry *p;
572
573 p = proc_create_reg(name, mode, &parent, data);
574 if (!p)
575 return NULL;
576 p->proc_ops = proc_ops;
577 pde_set_flags(p);
578 return proc_register(parent, p);
579}
580EXPORT_SYMBOL(proc_create_data);
581
582struct proc_dir_entry *proc_create(const char *name, umode_t mode,
583 struct proc_dir_entry *parent,
584 const struct proc_ops *proc_ops)
585{
586 return proc_create_data(name, mode, parent, proc_ops, NULL);
587}
588EXPORT_SYMBOL(proc_create);
589
590static int proc_seq_open(struct inode *inode, struct file *file)
591{
592 struct proc_dir_entry *de = PDE(inode);
593
594 if (de->state_size)
595 return seq_open_private(file, de->seq_ops, de->state_size);
596 return seq_open(file, de->seq_ops);
597}
598
599static int proc_seq_release(struct inode *inode, struct file *file)
600{
601 struct proc_dir_entry *de = PDE(inode);
602
603 if (de->state_size)
604 return seq_release_private(inode, file);
605 return seq_release(inode, file);
606}
607
608static const struct proc_ops proc_seq_ops = {
609 /* not permanent -- can call into arbitrary seq_operations */
610 .proc_open = proc_seq_open,
611 .proc_read_iter = seq_read_iter,
612 .proc_lseek = seq_lseek,
613 .proc_release = proc_seq_release,
614};
615
616struct proc_dir_entry *proc_create_seq_private(const char *name, umode_t mode,
617 struct proc_dir_entry *parent, const struct seq_operations *ops,
618 unsigned int state_size, void *data)
619{
620 struct proc_dir_entry *p;
621
622 p = proc_create_reg(name, mode, &parent, data);
623 if (!p)
624 return NULL;
625 p->proc_ops = &proc_seq_ops;
626 p->seq_ops = ops;
627 p->state_size = state_size;
628 return proc_register(parent, p);
629}
630EXPORT_SYMBOL(proc_create_seq_private);
631
632static int proc_single_open(struct inode *inode, struct file *file)
633{
634 struct proc_dir_entry *de = PDE(inode);
635
636 return single_open(file, de->single_show, de->data);
637}
638
639static const struct proc_ops proc_single_ops = {
640 /* not permanent -- can call into arbitrary ->single_show */
641 .proc_open = proc_single_open,
642 .proc_read_iter = seq_read_iter,
643 .proc_lseek = seq_lseek,
644 .proc_release = single_release,
645};
646
647struct proc_dir_entry *proc_create_single_data(const char *name, umode_t mode,
648 struct proc_dir_entry *parent,
649 int (*show)(struct seq_file *, void *), void *data)
650{
651 struct proc_dir_entry *p;
652
653 p = proc_create_reg(name, mode, &parent, data);
654 if (!p)
655 return NULL;
656 p->proc_ops = &proc_single_ops;
657 p->single_show = show;
658 return proc_register(parent, p);
659}
660EXPORT_SYMBOL(proc_create_single_data);
661
662void proc_set_size(struct proc_dir_entry *de, loff_t size)
663{
664 de->size = size;
665}
666EXPORT_SYMBOL(proc_set_size);
667
668void proc_set_user(struct proc_dir_entry *de, kuid_t uid, kgid_t gid)
669{
670 de->uid = uid;
671 de->gid = gid;
672}
673EXPORT_SYMBOL(proc_set_user);
674
675void pde_put(struct proc_dir_entry *pde)
676{
677 if (refcount_dec_and_test(&pde->refcnt)) {
678 proc_free_inum(pde->low_ino);
679 pde_free(pde);
680 }
681}
682
683/*
684 * Remove a /proc entry and free it if it's not currently in use.
685 */
686void remove_proc_entry(const char *name, struct proc_dir_entry *parent)
687{
688 struct proc_dir_entry *de = NULL;
689 const char *fn = name;
690 unsigned int len;
691
692 write_lock(&proc_subdir_lock);
693 if (__xlate_proc_name(name, &parent, &fn) != 0) {
694 write_unlock(&proc_subdir_lock);
695 return;
696 }
697 len = strlen(fn);
698
699 de = pde_subdir_find(parent, fn, len);
700 if (de) {
701 if (unlikely(pde_is_permanent(de))) {
702 WARN(1, "removing permanent /proc entry '%s'", de->name);
703 de = NULL;
704 } else {
705 rb_erase(&de->subdir_node, &parent->subdir);
706 if (S_ISDIR(de->mode))
707 parent->nlink--;
708 }
709 }
710 write_unlock(&proc_subdir_lock);
711 if (!de) {
712 WARN(1, "name '%s'\n", name);
713 return;
714 }
715
716 proc_entry_rundown(de);
717
718 WARN(pde_subdir_first(de),
719 "%s: removing non-empty directory '%s/%s', leaking at least '%s'\n",
720 __func__, de->parent->name, de->name, pde_subdir_first(de)->name);
721 pde_put(de);
722}
723EXPORT_SYMBOL(remove_proc_entry);
724
725int remove_proc_subtree(const char *name, struct proc_dir_entry *parent)
726{
727 struct proc_dir_entry *root = NULL, *de, *next;
728 const char *fn = name;
729 unsigned int len;
730
731 write_lock(&proc_subdir_lock);
732 if (__xlate_proc_name(name, &parent, &fn) != 0) {
733 write_unlock(&proc_subdir_lock);
734 return -ENOENT;
735 }
736 len = strlen(fn);
737
738 root = pde_subdir_find(parent, fn, len);
739 if (!root) {
740 write_unlock(&proc_subdir_lock);
741 return -ENOENT;
742 }
743 if (unlikely(pde_is_permanent(root))) {
744 write_unlock(&proc_subdir_lock);
745 WARN(1, "removing permanent /proc entry '%s/%s'",
746 root->parent->name, root->name);
747 return -EINVAL;
748 }
749 rb_erase(&root->subdir_node, &parent->subdir);
750
751 de = root;
752 while (1) {
753 next = pde_subdir_first(de);
754 if (next) {
755 if (unlikely(pde_is_permanent(next))) {
756 write_unlock(&proc_subdir_lock);
757 WARN(1, "removing permanent /proc entry '%s/%s'",
758 next->parent->name, next->name);
759 return -EINVAL;
760 }
761 rb_erase(&next->subdir_node, &de->subdir);
762 de = next;
763 continue;
764 }
765 next = de->parent;
766 if (S_ISDIR(de->mode))
767 next->nlink--;
768 write_unlock(&proc_subdir_lock);
769
770 proc_entry_rundown(de);
771 if (de == root)
772 break;
773 pde_put(de);
774
775 write_lock(&proc_subdir_lock);
776 de = next;
777 }
778 pde_put(root);
779 return 0;
780}
781EXPORT_SYMBOL(remove_proc_subtree);
782
783void *proc_get_parent_data(const struct inode *inode)
784{
785 struct proc_dir_entry *de = PDE(inode);
786 return de->parent->data;
787}
788EXPORT_SYMBOL_GPL(proc_get_parent_data);
789
790void proc_remove(struct proc_dir_entry *de)
791{
792 if (de)
793 remove_proc_subtree(de->name, de->parent);
794}
795EXPORT_SYMBOL(proc_remove);
796
797/*
798 * Pull a user buffer into memory and pass it to the file's write handler if
799 * one is supplied. The ->write() method is permitted to modify the
800 * kernel-side buffer.
801 */
802ssize_t proc_simple_write(struct file *f, const char __user *ubuf, size_t size,
803 loff_t *_pos)
804{
805 struct proc_dir_entry *pde = PDE(file_inode(f));
806 char *buf;
807 int ret;
808
809 if (!pde->write)
810 return -EACCES;
811 if (size == 0 || size > PAGE_SIZE - 1)
812 return -EINVAL;
813 buf = memdup_user_nul(ubuf, size);
814 if (IS_ERR(buf))
815 return PTR_ERR(buf);
816 ret = pde->write(f, buf, size);
817 kfree(buf);
818 return ret == 0 ? size : ret;
819}
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/printk.h>
19#include <linux/mount.h>
20#include <linux/init.h>
21#include <linux/idr.h>
22#include <linux/bitops.h>
23#include <linux/spinlock.h>
24#include <linux/completion.h>
25#include <linux/uaccess.h>
26
27#include "internal.h"
28
29static DEFINE_RWLOCK(proc_subdir_lock);
30
31static int proc_match(unsigned int len, const char *name, struct proc_dir_entry *de)
32{
33 if (len < de->namelen)
34 return -1;
35 if (len > de->namelen)
36 return 1;
37
38 return memcmp(name, de->name, len);
39}
40
41static struct proc_dir_entry *pde_subdir_first(struct proc_dir_entry *dir)
42{
43 return rb_entry_safe(rb_first(&dir->subdir), struct proc_dir_entry,
44 subdir_node);
45}
46
47static struct proc_dir_entry *pde_subdir_next(struct proc_dir_entry *dir)
48{
49 return rb_entry_safe(rb_next(&dir->subdir_node), struct proc_dir_entry,
50 subdir_node);
51}
52
53static struct proc_dir_entry *pde_subdir_find(struct proc_dir_entry *dir,
54 const char *name,
55 unsigned int len)
56{
57 struct rb_node *node = dir->subdir.rb_node;
58
59 while (node) {
60 struct proc_dir_entry *de = container_of(node,
61 struct proc_dir_entry,
62 subdir_node);
63 int result = proc_match(len, name, de);
64
65 if (result < 0)
66 node = node->rb_left;
67 else if (result > 0)
68 node = node->rb_right;
69 else
70 return de;
71 }
72 return NULL;
73}
74
75static bool pde_subdir_insert(struct proc_dir_entry *dir,
76 struct proc_dir_entry *de)
77{
78 struct rb_root *root = &dir->subdir;
79 struct rb_node **new = &root->rb_node, *parent = NULL;
80
81 /* Figure out where to put new node */
82 while (*new) {
83 struct proc_dir_entry *this =
84 container_of(*new, struct proc_dir_entry, subdir_node);
85 int result = proc_match(de->namelen, de->name, this);
86
87 parent = *new;
88 if (result < 0)
89 new = &(*new)->rb_left;
90 else if (result > 0)
91 new = &(*new)->rb_right;
92 else
93 return false;
94 }
95
96 /* Add new node and rebalance tree. */
97 rb_link_node(&de->subdir_node, parent, new);
98 rb_insert_color(&de->subdir_node, root);
99 return true;
100}
101
102static int proc_notify_change(struct dentry *dentry, struct iattr *iattr)
103{
104 struct inode *inode = d_inode(dentry);
105 struct proc_dir_entry *de = PDE(inode);
106 int error;
107
108 error = setattr_prepare(dentry, iattr);
109 if (error)
110 return error;
111
112 setattr_copy(inode, iattr);
113 mark_inode_dirty(inode);
114
115 proc_set_user(de, inode->i_uid, inode->i_gid);
116 de->mode = inode->i_mode;
117 return 0;
118}
119
120static int proc_getattr(struct vfsmount *mnt, struct dentry *dentry,
121 struct kstat *stat)
122{
123 struct inode *inode = d_inode(dentry);
124 struct proc_dir_entry *de = PDE(inode);
125 if (de && de->nlink)
126 set_nlink(inode, de->nlink);
127
128 generic_fillattr(inode, stat);
129 return 0;
130}
131
132static const struct inode_operations proc_file_inode_operations = {
133 .setattr = proc_notify_change,
134};
135
136/*
137 * This function parses a name such as "tty/driver/serial", and
138 * returns the struct proc_dir_entry for "/proc/tty/driver", and
139 * returns "serial" in residual.
140 */
141static int __xlate_proc_name(const char *name, struct proc_dir_entry **ret,
142 const char **residual)
143{
144 const char *cp = name, *next;
145 struct proc_dir_entry *de;
146 unsigned int len;
147
148 de = *ret;
149 if (!de)
150 de = &proc_root;
151
152 while (1) {
153 next = strchr(cp, '/');
154 if (!next)
155 break;
156
157 len = next - cp;
158 de = pde_subdir_find(de, cp, len);
159 if (!de) {
160 WARN(1, "name '%s'\n", name);
161 return -ENOENT;
162 }
163 cp += len + 1;
164 }
165 *residual = cp;
166 *ret = de;
167 return 0;
168}
169
170static int xlate_proc_name(const char *name, struct proc_dir_entry **ret,
171 const char **residual)
172{
173 int rv;
174
175 read_lock(&proc_subdir_lock);
176 rv = __xlate_proc_name(name, ret, residual);
177 read_unlock(&proc_subdir_lock);
178 return rv;
179}
180
181static DEFINE_IDA(proc_inum_ida);
182static DEFINE_SPINLOCK(proc_inum_lock); /* protects the above */
183
184#define PROC_DYNAMIC_FIRST 0xF0000000U
185
186/*
187 * Return an inode number between PROC_DYNAMIC_FIRST and
188 * 0xffffffff, or zero on failure.
189 */
190int proc_alloc_inum(unsigned int *inum)
191{
192 unsigned int i;
193 int error;
194
195retry:
196 if (!ida_pre_get(&proc_inum_ida, GFP_KERNEL))
197 return -ENOMEM;
198
199 spin_lock_irq(&proc_inum_lock);
200 error = ida_get_new(&proc_inum_ida, &i);
201 spin_unlock_irq(&proc_inum_lock);
202 if (error == -EAGAIN)
203 goto retry;
204 else if (error)
205 return error;
206
207 if (i > UINT_MAX - PROC_DYNAMIC_FIRST) {
208 spin_lock_irq(&proc_inum_lock);
209 ida_remove(&proc_inum_ida, i);
210 spin_unlock_irq(&proc_inum_lock);
211 return -ENOSPC;
212 }
213 *inum = PROC_DYNAMIC_FIRST + i;
214 return 0;
215}
216
217void proc_free_inum(unsigned int inum)
218{
219 unsigned long flags;
220 spin_lock_irqsave(&proc_inum_lock, flags);
221 ida_remove(&proc_inum_ida, inum - PROC_DYNAMIC_FIRST);
222 spin_unlock_irqrestore(&proc_inum_lock, flags);
223}
224
225/*
226 * Don't create negative dentries here, return -ENOENT by hand
227 * instead.
228 */
229struct dentry *proc_lookup_de(struct proc_dir_entry *de, struct inode *dir,
230 struct dentry *dentry)
231{
232 struct inode *inode;
233
234 read_lock(&proc_subdir_lock);
235 de = pde_subdir_find(de, dentry->d_name.name, dentry->d_name.len);
236 if (de) {
237 pde_get(de);
238 read_unlock(&proc_subdir_lock);
239 inode = proc_get_inode(dir->i_sb, de);
240 if (!inode)
241 return ERR_PTR(-ENOMEM);
242 d_set_d_op(dentry, &simple_dentry_operations);
243 d_add(dentry, inode);
244 return NULL;
245 }
246 read_unlock(&proc_subdir_lock);
247 return ERR_PTR(-ENOENT);
248}
249
250struct dentry *proc_lookup(struct inode *dir, struct dentry *dentry,
251 unsigned int flags)
252{
253 return proc_lookup_de(PDE(dir), dir, dentry);
254}
255
256/*
257 * This returns non-zero if at EOF, so that the /proc
258 * root directory can use this and check if it should
259 * continue with the <pid> entries..
260 *
261 * Note that the VFS-layer doesn't care about the return
262 * value of the readdir() call, as long as it's non-negative
263 * for success..
264 */
265int proc_readdir_de(struct proc_dir_entry *de, struct file *file,
266 struct dir_context *ctx)
267{
268 int i;
269
270 if (!dir_emit_dots(file, ctx))
271 return 0;
272
273 read_lock(&proc_subdir_lock);
274 de = pde_subdir_first(de);
275 i = ctx->pos - 2;
276 for (;;) {
277 if (!de) {
278 read_unlock(&proc_subdir_lock);
279 return 0;
280 }
281 if (!i)
282 break;
283 de = pde_subdir_next(de);
284 i--;
285 }
286
287 do {
288 struct proc_dir_entry *next;
289 pde_get(de);
290 read_unlock(&proc_subdir_lock);
291 if (!dir_emit(ctx, de->name, de->namelen,
292 de->low_ino, de->mode >> 12)) {
293 pde_put(de);
294 return 0;
295 }
296 read_lock(&proc_subdir_lock);
297 ctx->pos++;
298 next = pde_subdir_next(de);
299 pde_put(de);
300 de = next;
301 } while (de);
302 read_unlock(&proc_subdir_lock);
303 return 1;
304}
305
306int proc_readdir(struct file *file, struct dir_context *ctx)
307{
308 struct inode *inode = file_inode(file);
309
310 return proc_readdir_de(PDE(inode), file, ctx);
311}
312
313/*
314 * These are the generic /proc directory operations. They
315 * use the in-memory "struct proc_dir_entry" tree to parse
316 * the /proc directory.
317 */
318static const struct file_operations proc_dir_operations = {
319 .llseek = generic_file_llseek,
320 .read = generic_read_dir,
321 .iterate_shared = proc_readdir,
322};
323
324/*
325 * proc directories can do almost nothing..
326 */
327static const struct inode_operations proc_dir_inode_operations = {
328 .lookup = proc_lookup,
329 .getattr = proc_getattr,
330 .setattr = proc_notify_change,
331};
332
333static int proc_register(struct proc_dir_entry * dir, struct proc_dir_entry * dp)
334{
335 int ret;
336
337 ret = proc_alloc_inum(&dp->low_ino);
338 if (ret)
339 return ret;
340
341 write_lock(&proc_subdir_lock);
342 dp->parent = dir;
343 if (pde_subdir_insert(dir, dp) == false) {
344 WARN(1, "proc_dir_entry '%s/%s' already registered\n",
345 dir->name, dp->name);
346 write_unlock(&proc_subdir_lock);
347 proc_free_inum(dp->low_ino);
348 return -EEXIST;
349 }
350 write_unlock(&proc_subdir_lock);
351
352 return 0;
353}
354
355static struct proc_dir_entry *__proc_create(struct proc_dir_entry **parent,
356 const char *name,
357 umode_t mode,
358 nlink_t nlink)
359{
360 struct proc_dir_entry *ent = NULL;
361 const char *fn;
362 struct qstr qstr;
363
364 if (xlate_proc_name(name, parent, &fn) != 0)
365 goto out;
366 qstr.name = fn;
367 qstr.len = strlen(fn);
368 if (qstr.len == 0 || qstr.len >= 256) {
369 WARN(1, "name len %u\n", qstr.len);
370 return NULL;
371 }
372 if (*parent == &proc_root && name_to_int(&qstr) != ~0U) {
373 WARN(1, "create '/proc/%s' by hand\n", qstr.name);
374 return NULL;
375 }
376 if (is_empty_pde(*parent)) {
377 WARN(1, "attempt to add to permanently empty directory");
378 return NULL;
379 }
380
381 ent = kzalloc(sizeof(struct proc_dir_entry) + qstr.len + 1, GFP_KERNEL);
382 if (!ent)
383 goto out;
384
385 memcpy(ent->name, fn, qstr.len + 1);
386 ent->namelen = qstr.len;
387 ent->mode = mode;
388 ent->nlink = nlink;
389 ent->subdir = RB_ROOT;
390 atomic_set(&ent->count, 1);
391 spin_lock_init(&ent->pde_unload_lock);
392 INIT_LIST_HEAD(&ent->pde_openers);
393 proc_set_user(ent, (*parent)->uid, (*parent)->gid);
394
395out:
396 return ent;
397}
398
399struct proc_dir_entry *proc_symlink(const char *name,
400 struct proc_dir_entry *parent, const char *dest)
401{
402 struct proc_dir_entry *ent;
403
404 ent = __proc_create(&parent, name,
405 (S_IFLNK | S_IRUGO | S_IWUGO | S_IXUGO),1);
406
407 if (ent) {
408 ent->data = kmalloc((ent->size=strlen(dest))+1, GFP_KERNEL);
409 if (ent->data) {
410 strcpy((char*)ent->data,dest);
411 ent->proc_iops = &proc_link_inode_operations;
412 if (proc_register(parent, ent) < 0) {
413 kfree(ent->data);
414 kfree(ent);
415 ent = NULL;
416 }
417 } else {
418 kfree(ent);
419 ent = NULL;
420 }
421 }
422 return ent;
423}
424EXPORT_SYMBOL(proc_symlink);
425
426struct proc_dir_entry *proc_mkdir_data(const char *name, umode_t mode,
427 struct proc_dir_entry *parent, void *data)
428{
429 struct proc_dir_entry *ent;
430
431 if (mode == 0)
432 mode = S_IRUGO | S_IXUGO;
433
434 ent = __proc_create(&parent, name, S_IFDIR | mode, 2);
435 if (ent) {
436 ent->data = data;
437 ent->proc_fops = &proc_dir_operations;
438 ent->proc_iops = &proc_dir_inode_operations;
439 parent->nlink++;
440 if (proc_register(parent, ent) < 0) {
441 kfree(ent);
442 parent->nlink--;
443 ent = NULL;
444 }
445 }
446 return ent;
447}
448EXPORT_SYMBOL_GPL(proc_mkdir_data);
449
450struct proc_dir_entry *proc_mkdir_mode(const char *name, umode_t mode,
451 struct proc_dir_entry *parent)
452{
453 return proc_mkdir_data(name, mode, parent, NULL);
454}
455EXPORT_SYMBOL(proc_mkdir_mode);
456
457struct proc_dir_entry *proc_mkdir(const char *name,
458 struct proc_dir_entry *parent)
459{
460 return proc_mkdir_data(name, 0, parent, NULL);
461}
462EXPORT_SYMBOL(proc_mkdir);
463
464struct proc_dir_entry *proc_create_mount_point(const char *name)
465{
466 umode_t mode = S_IFDIR | S_IRUGO | S_IXUGO;
467 struct proc_dir_entry *ent, *parent = NULL;
468
469 ent = __proc_create(&parent, name, mode, 2);
470 if (ent) {
471 ent->data = NULL;
472 ent->proc_fops = NULL;
473 ent->proc_iops = NULL;
474 if (proc_register(parent, ent) < 0) {
475 kfree(ent);
476 parent->nlink--;
477 ent = NULL;
478 }
479 }
480 return ent;
481}
482EXPORT_SYMBOL(proc_create_mount_point);
483
484struct proc_dir_entry *proc_create_data(const char *name, umode_t mode,
485 struct proc_dir_entry *parent,
486 const struct file_operations *proc_fops,
487 void *data)
488{
489 struct proc_dir_entry *pde;
490 if ((mode & S_IFMT) == 0)
491 mode |= S_IFREG;
492
493 if (!S_ISREG(mode)) {
494 WARN_ON(1); /* use proc_mkdir() */
495 return NULL;
496 }
497
498 BUG_ON(proc_fops == NULL);
499
500 if ((mode & S_IALLUGO) == 0)
501 mode |= S_IRUGO;
502 pde = __proc_create(&parent, name, mode, 1);
503 if (!pde)
504 goto out;
505 pde->proc_fops = proc_fops;
506 pde->data = data;
507 pde->proc_iops = &proc_file_inode_operations;
508 if (proc_register(parent, pde) < 0)
509 goto out_free;
510 return pde;
511out_free:
512 kfree(pde);
513out:
514 return NULL;
515}
516EXPORT_SYMBOL(proc_create_data);
517
518void proc_set_size(struct proc_dir_entry *de, loff_t size)
519{
520 de->size = size;
521}
522EXPORT_SYMBOL(proc_set_size);
523
524void proc_set_user(struct proc_dir_entry *de, kuid_t uid, kgid_t gid)
525{
526 de->uid = uid;
527 de->gid = gid;
528}
529EXPORT_SYMBOL(proc_set_user);
530
531static void free_proc_entry(struct proc_dir_entry *de)
532{
533 proc_free_inum(de->low_ino);
534
535 if (S_ISLNK(de->mode))
536 kfree(de->data);
537 kfree(de);
538}
539
540void pde_put(struct proc_dir_entry *pde)
541{
542 if (atomic_dec_and_test(&pde->count))
543 free_proc_entry(pde);
544}
545
546/*
547 * Remove a /proc entry and free it if it's not currently in use.
548 */
549void remove_proc_entry(const char *name, struct proc_dir_entry *parent)
550{
551 struct proc_dir_entry *de = NULL;
552 const char *fn = name;
553 unsigned int len;
554
555 write_lock(&proc_subdir_lock);
556 if (__xlate_proc_name(name, &parent, &fn) != 0) {
557 write_unlock(&proc_subdir_lock);
558 return;
559 }
560 len = strlen(fn);
561
562 de = pde_subdir_find(parent, fn, len);
563 if (de)
564 rb_erase(&de->subdir_node, &parent->subdir);
565 write_unlock(&proc_subdir_lock);
566 if (!de) {
567 WARN(1, "name '%s'\n", name);
568 return;
569 }
570
571 proc_entry_rundown(de);
572
573 if (S_ISDIR(de->mode))
574 parent->nlink--;
575 de->nlink = 0;
576 WARN(pde_subdir_first(de),
577 "%s: removing non-empty directory '%s/%s', leaking at least '%s'\n",
578 __func__, de->parent->name, de->name, pde_subdir_first(de)->name);
579 pde_put(de);
580}
581EXPORT_SYMBOL(remove_proc_entry);
582
583int remove_proc_subtree(const char *name, struct proc_dir_entry *parent)
584{
585 struct proc_dir_entry *root = NULL, *de, *next;
586 const char *fn = name;
587 unsigned int len;
588
589 write_lock(&proc_subdir_lock);
590 if (__xlate_proc_name(name, &parent, &fn) != 0) {
591 write_unlock(&proc_subdir_lock);
592 return -ENOENT;
593 }
594 len = strlen(fn);
595
596 root = pde_subdir_find(parent, fn, len);
597 if (!root) {
598 write_unlock(&proc_subdir_lock);
599 return -ENOENT;
600 }
601 rb_erase(&root->subdir_node, &parent->subdir);
602
603 de = root;
604 while (1) {
605 next = pde_subdir_first(de);
606 if (next) {
607 rb_erase(&next->subdir_node, &de->subdir);
608 de = next;
609 continue;
610 }
611 write_unlock(&proc_subdir_lock);
612
613 proc_entry_rundown(de);
614 next = de->parent;
615 if (S_ISDIR(de->mode))
616 next->nlink--;
617 de->nlink = 0;
618 if (de == root)
619 break;
620 pde_put(de);
621
622 write_lock(&proc_subdir_lock);
623 de = next;
624 }
625 pde_put(root);
626 return 0;
627}
628EXPORT_SYMBOL(remove_proc_subtree);
629
630void *proc_get_parent_data(const struct inode *inode)
631{
632 struct proc_dir_entry *de = PDE(inode);
633 return de->parent->data;
634}
635EXPORT_SYMBOL_GPL(proc_get_parent_data);
636
637void proc_remove(struct proc_dir_entry *de)
638{
639 if (de)
640 remove_proc_subtree(de->name, de->parent);
641}
642EXPORT_SYMBOL(proc_remove);
643
644void *PDE_DATA(const struct inode *inode)
645{
646 return __PDE_DATA(inode);
647}
648EXPORT_SYMBOL(PDE_DATA);