Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * /proc/sys support
4 */
5#include <linux/init.h>
6#include <linux/sysctl.h>
7#include <linux/poll.h>
8#include <linux/proc_fs.h>
9#include <linux/printk.h>
10#include <linux/security.h>
11#include <linux/sched.h>
12#include <linux/cred.h>
13#include <linux/namei.h>
14#include <linux/mm.h>
15#include <linux/module.h>
16#include "internal.h"
17
18static const struct dentry_operations proc_sys_dentry_operations;
19static const struct file_operations proc_sys_file_operations;
20static const struct inode_operations proc_sys_inode_operations;
21static const struct file_operations proc_sys_dir_file_operations;
22static const struct inode_operations proc_sys_dir_operations;
23
24/* Support for permanently empty directories */
25
26struct ctl_table sysctl_mount_point[] = {
27 { }
28};
29
30static bool is_empty_dir(struct ctl_table_header *head)
31{
32 return head->ctl_table[0].child == sysctl_mount_point;
33}
34
35static void set_empty_dir(struct ctl_dir *dir)
36{
37 dir->header.ctl_table[0].child = sysctl_mount_point;
38}
39
40static void clear_empty_dir(struct ctl_dir *dir)
41
42{
43 dir->header.ctl_table[0].child = NULL;
44}
45
46void proc_sys_poll_notify(struct ctl_table_poll *poll)
47{
48 if (!poll)
49 return;
50
51 atomic_inc(&poll->event);
52 wake_up_interruptible(&poll->wait);
53}
54
55static struct ctl_table root_table[] = {
56 {
57 .procname = "",
58 .mode = S_IFDIR|S_IRUGO|S_IXUGO,
59 },
60 { }
61};
62static struct ctl_table_root sysctl_table_root = {
63 .default_set.dir.header = {
64 {{.count = 1,
65 .nreg = 1,
66 .ctl_table = root_table }},
67 .ctl_table_arg = root_table,
68 .root = &sysctl_table_root,
69 .set = &sysctl_table_root.default_set,
70 },
71};
72
73static DEFINE_SPINLOCK(sysctl_lock);
74
75static void drop_sysctl_table(struct ctl_table_header *header);
76static int sysctl_follow_link(struct ctl_table_header **phead,
77 struct ctl_table **pentry);
78static int insert_links(struct ctl_table_header *head);
79static void put_links(struct ctl_table_header *header);
80
81static void sysctl_print_dir(struct ctl_dir *dir)
82{
83 if (dir->header.parent)
84 sysctl_print_dir(dir->header.parent);
85 pr_cont("%s/", dir->header.ctl_table[0].procname);
86}
87
88static int namecmp(const char *name1, int len1, const char *name2, int len2)
89{
90 int minlen;
91 int cmp;
92
93 minlen = len1;
94 if (minlen > len2)
95 minlen = len2;
96
97 cmp = memcmp(name1, name2, minlen);
98 if (cmp == 0)
99 cmp = len1 - len2;
100 return cmp;
101}
102
103/* Called under sysctl_lock */
104static struct ctl_table *find_entry(struct ctl_table_header **phead,
105 struct ctl_dir *dir, const char *name, int namelen)
106{
107 struct ctl_table_header *head;
108 struct ctl_table *entry;
109 struct rb_node *node = dir->root.rb_node;
110
111 while (node)
112 {
113 struct ctl_node *ctl_node;
114 const char *procname;
115 int cmp;
116
117 ctl_node = rb_entry(node, struct ctl_node, node);
118 head = ctl_node->header;
119 entry = &head->ctl_table[ctl_node - head->node];
120 procname = entry->procname;
121
122 cmp = namecmp(name, namelen, procname, strlen(procname));
123 if (cmp < 0)
124 node = node->rb_left;
125 else if (cmp > 0)
126 node = node->rb_right;
127 else {
128 *phead = head;
129 return entry;
130 }
131 }
132 return NULL;
133}
134
135static int insert_entry(struct ctl_table_header *head, struct ctl_table *entry)
136{
137 struct rb_node *node = &head->node[entry - head->ctl_table].node;
138 struct rb_node **p = &head->parent->root.rb_node;
139 struct rb_node *parent = NULL;
140 const char *name = entry->procname;
141 int namelen = strlen(name);
142
143 while (*p) {
144 struct ctl_table_header *parent_head;
145 struct ctl_table *parent_entry;
146 struct ctl_node *parent_node;
147 const char *parent_name;
148 int cmp;
149
150 parent = *p;
151 parent_node = rb_entry(parent, struct ctl_node, node);
152 parent_head = parent_node->header;
153 parent_entry = &parent_head->ctl_table[parent_node - parent_head->node];
154 parent_name = parent_entry->procname;
155
156 cmp = namecmp(name, namelen, parent_name, strlen(parent_name));
157 if (cmp < 0)
158 p = &(*p)->rb_left;
159 else if (cmp > 0)
160 p = &(*p)->rb_right;
161 else {
162 pr_err("sysctl duplicate entry: ");
163 sysctl_print_dir(head->parent);
164 pr_cont("/%s\n", entry->procname);
165 return -EEXIST;
166 }
167 }
168
169 rb_link_node(node, parent, p);
170 rb_insert_color(node, &head->parent->root);
171 return 0;
172}
173
174static void erase_entry(struct ctl_table_header *head, struct ctl_table *entry)
175{
176 struct rb_node *node = &head->node[entry - head->ctl_table].node;
177
178 rb_erase(node, &head->parent->root);
179}
180
181static void init_header(struct ctl_table_header *head,
182 struct ctl_table_root *root, struct ctl_table_set *set,
183 struct ctl_node *node, struct ctl_table *table)
184{
185 head->ctl_table = table;
186 head->ctl_table_arg = table;
187 head->used = 0;
188 head->count = 1;
189 head->nreg = 1;
190 head->unregistering = NULL;
191 head->root = root;
192 head->set = set;
193 head->parent = NULL;
194 head->node = node;
195 INIT_HLIST_HEAD(&head->inodes);
196 if (node) {
197 struct ctl_table *entry;
198 for (entry = table; entry->procname; entry++, node++)
199 node->header = head;
200 }
201}
202
203static void erase_header(struct ctl_table_header *head)
204{
205 struct ctl_table *entry;
206 for (entry = head->ctl_table; entry->procname; entry++)
207 erase_entry(head, entry);
208}
209
210static int insert_header(struct ctl_dir *dir, struct ctl_table_header *header)
211{
212 struct ctl_table *entry;
213 int err;
214
215 /* Is this a permanently empty directory? */
216 if (is_empty_dir(&dir->header))
217 return -EROFS;
218
219 /* Am I creating a permanently empty directory? */
220 if (header->ctl_table == sysctl_mount_point) {
221 if (!RB_EMPTY_ROOT(&dir->root))
222 return -EINVAL;
223 set_empty_dir(dir);
224 }
225
226 dir->header.nreg++;
227 header->parent = dir;
228 err = insert_links(header);
229 if (err)
230 goto fail_links;
231 for (entry = header->ctl_table; entry->procname; entry++) {
232 err = insert_entry(header, entry);
233 if (err)
234 goto fail;
235 }
236 return 0;
237fail:
238 erase_header(header);
239 put_links(header);
240fail_links:
241 if (header->ctl_table == sysctl_mount_point)
242 clear_empty_dir(dir);
243 header->parent = NULL;
244 drop_sysctl_table(&dir->header);
245 return err;
246}
247
248/* called under sysctl_lock */
249static int use_table(struct ctl_table_header *p)
250{
251 if (unlikely(p->unregistering))
252 return 0;
253 p->used++;
254 return 1;
255}
256
257/* called under sysctl_lock */
258static void unuse_table(struct ctl_table_header *p)
259{
260 if (!--p->used)
261 if (unlikely(p->unregistering))
262 complete(p->unregistering);
263}
264
265static void proc_sys_prune_dcache(struct ctl_table_header *head)
266{
267 struct inode *inode;
268 struct proc_inode *ei;
269 struct hlist_node *node;
270 struct super_block *sb;
271
272 rcu_read_lock();
273 for (;;) {
274 node = hlist_first_rcu(&head->inodes);
275 if (!node)
276 break;
277 ei = hlist_entry(node, struct proc_inode, sysctl_inodes);
278 spin_lock(&sysctl_lock);
279 hlist_del_init_rcu(&ei->sysctl_inodes);
280 spin_unlock(&sysctl_lock);
281
282 inode = &ei->vfs_inode;
283 sb = inode->i_sb;
284 if (!atomic_inc_not_zero(&sb->s_active))
285 continue;
286 inode = igrab(inode);
287 rcu_read_unlock();
288 if (unlikely(!inode)) {
289 deactivate_super(sb);
290 rcu_read_lock();
291 continue;
292 }
293
294 d_prune_aliases(inode);
295 iput(inode);
296 deactivate_super(sb);
297
298 rcu_read_lock();
299 }
300 rcu_read_unlock();
301}
302
303/* called under sysctl_lock, will reacquire if has to wait */
304static void start_unregistering(struct ctl_table_header *p)
305{
306 /*
307 * if p->used is 0, nobody will ever touch that entry again;
308 * we'll eliminate all paths to it before dropping sysctl_lock
309 */
310 if (unlikely(p->used)) {
311 struct completion wait;
312 init_completion(&wait);
313 p->unregistering = &wait;
314 spin_unlock(&sysctl_lock);
315 wait_for_completion(&wait);
316 } else {
317 /* anything non-NULL; we'll never dereference it */
318 p->unregistering = ERR_PTR(-EINVAL);
319 spin_unlock(&sysctl_lock);
320 }
321 /*
322 * Prune dentries for unregistered sysctls: namespaced sysctls
323 * can have duplicate names and contaminate dcache very badly.
324 */
325 proc_sys_prune_dcache(p);
326 /*
327 * do not remove from the list until nobody holds it; walking the
328 * list in do_sysctl() relies on that.
329 */
330 spin_lock(&sysctl_lock);
331 erase_header(p);
332}
333
334static struct ctl_table_header *sysctl_head_grab(struct ctl_table_header *head)
335{
336 BUG_ON(!head);
337 spin_lock(&sysctl_lock);
338 if (!use_table(head))
339 head = ERR_PTR(-ENOENT);
340 spin_unlock(&sysctl_lock);
341 return head;
342}
343
344static void sysctl_head_finish(struct ctl_table_header *head)
345{
346 if (!head)
347 return;
348 spin_lock(&sysctl_lock);
349 unuse_table(head);
350 spin_unlock(&sysctl_lock);
351}
352
353static struct ctl_table_set *
354lookup_header_set(struct ctl_table_root *root)
355{
356 struct ctl_table_set *set = &root->default_set;
357 if (root->lookup)
358 set = root->lookup(root);
359 return set;
360}
361
362static struct ctl_table *lookup_entry(struct ctl_table_header **phead,
363 struct ctl_dir *dir,
364 const char *name, int namelen)
365{
366 struct ctl_table_header *head;
367 struct ctl_table *entry;
368
369 spin_lock(&sysctl_lock);
370 entry = find_entry(&head, dir, name, namelen);
371 if (entry && use_table(head))
372 *phead = head;
373 else
374 entry = NULL;
375 spin_unlock(&sysctl_lock);
376 return entry;
377}
378
379static struct ctl_node *first_usable_entry(struct rb_node *node)
380{
381 struct ctl_node *ctl_node;
382
383 for (;node; node = rb_next(node)) {
384 ctl_node = rb_entry(node, struct ctl_node, node);
385 if (use_table(ctl_node->header))
386 return ctl_node;
387 }
388 return NULL;
389}
390
391static void first_entry(struct ctl_dir *dir,
392 struct ctl_table_header **phead, struct ctl_table **pentry)
393{
394 struct ctl_table_header *head = NULL;
395 struct ctl_table *entry = NULL;
396 struct ctl_node *ctl_node;
397
398 spin_lock(&sysctl_lock);
399 ctl_node = first_usable_entry(rb_first(&dir->root));
400 spin_unlock(&sysctl_lock);
401 if (ctl_node) {
402 head = ctl_node->header;
403 entry = &head->ctl_table[ctl_node - head->node];
404 }
405 *phead = head;
406 *pentry = entry;
407}
408
409static void next_entry(struct ctl_table_header **phead, struct ctl_table **pentry)
410{
411 struct ctl_table_header *head = *phead;
412 struct ctl_table *entry = *pentry;
413 struct ctl_node *ctl_node = &head->node[entry - head->ctl_table];
414
415 spin_lock(&sysctl_lock);
416 unuse_table(head);
417
418 ctl_node = first_usable_entry(rb_next(&ctl_node->node));
419 spin_unlock(&sysctl_lock);
420 head = NULL;
421 if (ctl_node) {
422 head = ctl_node->header;
423 entry = &head->ctl_table[ctl_node - head->node];
424 }
425 *phead = head;
426 *pentry = entry;
427}
428
429/*
430 * sysctl_perm does NOT grant the superuser all rights automatically, because
431 * some sysctl variables are readonly even to root.
432 */
433
434static int test_perm(int mode, int op)
435{
436 if (uid_eq(current_euid(), GLOBAL_ROOT_UID))
437 mode >>= 6;
438 else if (in_egroup_p(GLOBAL_ROOT_GID))
439 mode >>= 3;
440 if ((op & ~mode & (MAY_READ|MAY_WRITE|MAY_EXEC)) == 0)
441 return 0;
442 return -EACCES;
443}
444
445static int sysctl_perm(struct ctl_table_header *head, struct ctl_table *table, int op)
446{
447 struct ctl_table_root *root = head->root;
448 int mode;
449
450 if (root->permissions)
451 mode = root->permissions(head, table);
452 else
453 mode = table->mode;
454
455 return test_perm(mode, op);
456}
457
458static struct inode *proc_sys_make_inode(struct super_block *sb,
459 struct ctl_table_header *head, struct ctl_table *table)
460{
461 struct ctl_table_root *root = head->root;
462 struct inode *inode;
463 struct proc_inode *ei;
464
465 inode = new_inode(sb);
466 if (!inode)
467 goto out;
468
469 inode->i_ino = get_next_ino();
470
471 ei = PROC_I(inode);
472
473 spin_lock(&sysctl_lock);
474 if (unlikely(head->unregistering)) {
475 spin_unlock(&sysctl_lock);
476 iput(inode);
477 inode = NULL;
478 goto out;
479 }
480 ei->sysctl = head;
481 ei->sysctl_entry = table;
482 hlist_add_head_rcu(&ei->sysctl_inodes, &head->inodes);
483 head->count++;
484 spin_unlock(&sysctl_lock);
485
486 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
487 inode->i_mode = table->mode;
488 if (!S_ISDIR(table->mode)) {
489 inode->i_mode |= S_IFREG;
490 inode->i_op = &proc_sys_inode_operations;
491 inode->i_fop = &proc_sys_file_operations;
492 } else {
493 inode->i_mode |= S_IFDIR;
494 inode->i_op = &proc_sys_dir_operations;
495 inode->i_fop = &proc_sys_dir_file_operations;
496 if (is_empty_dir(head))
497 make_empty_dir_inode(inode);
498 }
499
500 if (root->set_ownership)
501 root->set_ownership(head, table, &inode->i_uid, &inode->i_gid);
502
503out:
504 return inode;
505}
506
507void proc_sys_evict_inode(struct inode *inode, struct ctl_table_header *head)
508{
509 spin_lock(&sysctl_lock);
510 hlist_del_init_rcu(&PROC_I(inode)->sysctl_inodes);
511 if (!--head->count)
512 kfree_rcu(head, rcu);
513 spin_unlock(&sysctl_lock);
514}
515
516static struct ctl_table_header *grab_header(struct inode *inode)
517{
518 struct ctl_table_header *head = PROC_I(inode)->sysctl;
519 if (!head)
520 head = &sysctl_table_root.default_set.dir.header;
521 return sysctl_head_grab(head);
522}
523
524static struct dentry *proc_sys_lookup(struct inode *dir, struct dentry *dentry,
525 unsigned int flags)
526{
527 struct ctl_table_header *head = grab_header(dir);
528 struct ctl_table_header *h = NULL;
529 const struct qstr *name = &dentry->d_name;
530 struct ctl_table *p;
531 struct inode *inode;
532 struct dentry *err = ERR_PTR(-ENOENT);
533 struct ctl_dir *ctl_dir;
534 int ret;
535
536 if (IS_ERR(head))
537 return ERR_CAST(head);
538
539 ctl_dir = container_of(head, struct ctl_dir, header);
540
541 p = lookup_entry(&h, ctl_dir, name->name, name->len);
542 if (!p)
543 goto out;
544
545 if (S_ISLNK(p->mode)) {
546 ret = sysctl_follow_link(&h, &p);
547 err = ERR_PTR(ret);
548 if (ret)
549 goto out;
550 }
551
552 err = ERR_PTR(-ENOMEM);
553 inode = proc_sys_make_inode(dir->i_sb, h ? h : head, p);
554 if (!inode)
555 goto out;
556
557 err = NULL;
558 d_set_d_op(dentry, &proc_sys_dentry_operations);
559 d_add(dentry, inode);
560
561out:
562 if (h)
563 sysctl_head_finish(h);
564 sysctl_head_finish(head);
565 return err;
566}
567
568static ssize_t proc_sys_call_handler(struct file *filp, void __user *buf,
569 size_t count, loff_t *ppos, int write)
570{
571 struct inode *inode = file_inode(filp);
572 struct ctl_table_header *head = grab_header(inode);
573 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
574 ssize_t error;
575 size_t res;
576
577 if (IS_ERR(head))
578 return PTR_ERR(head);
579
580 /*
581 * At this point we know that the sysctl was not unregistered
582 * and won't be until we finish.
583 */
584 error = -EPERM;
585 if (sysctl_perm(head, table, write ? MAY_WRITE : MAY_READ))
586 goto out;
587
588 /* if that can happen at all, it should be -EINVAL, not -EISDIR */
589 error = -EINVAL;
590 if (!table->proc_handler)
591 goto out;
592
593 /* careful: calling conventions are nasty here */
594 res = count;
595 error = table->proc_handler(table, write, buf, &res, ppos);
596 if (!error)
597 error = res;
598out:
599 sysctl_head_finish(head);
600
601 return error;
602}
603
604static ssize_t proc_sys_read(struct file *filp, char __user *buf,
605 size_t count, loff_t *ppos)
606{
607 return proc_sys_call_handler(filp, (void __user *)buf, count, ppos, 0);
608}
609
610static ssize_t proc_sys_write(struct file *filp, const char __user *buf,
611 size_t count, loff_t *ppos)
612{
613 return proc_sys_call_handler(filp, (void __user *)buf, count, ppos, 1);
614}
615
616static int proc_sys_open(struct inode *inode, struct file *filp)
617{
618 struct ctl_table_header *head = grab_header(inode);
619 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
620
621 /* sysctl was unregistered */
622 if (IS_ERR(head))
623 return PTR_ERR(head);
624
625 if (table->poll)
626 filp->private_data = proc_sys_poll_event(table->poll);
627
628 sysctl_head_finish(head);
629
630 return 0;
631}
632
633static __poll_t proc_sys_poll(struct file *filp, poll_table *wait)
634{
635 struct inode *inode = file_inode(filp);
636 struct ctl_table_header *head = grab_header(inode);
637 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
638 __poll_t ret = DEFAULT_POLLMASK;
639 unsigned long event;
640
641 /* sysctl was unregistered */
642 if (IS_ERR(head))
643 return EPOLLERR | EPOLLHUP;
644
645 if (!table->proc_handler)
646 goto out;
647
648 if (!table->poll)
649 goto out;
650
651 event = (unsigned long)filp->private_data;
652 poll_wait(filp, &table->poll->wait, wait);
653
654 if (event != atomic_read(&table->poll->event)) {
655 filp->private_data = proc_sys_poll_event(table->poll);
656 ret = EPOLLIN | EPOLLRDNORM | EPOLLERR | EPOLLPRI;
657 }
658
659out:
660 sysctl_head_finish(head);
661
662 return ret;
663}
664
665static bool proc_sys_fill_cache(struct file *file,
666 struct dir_context *ctx,
667 struct ctl_table_header *head,
668 struct ctl_table *table)
669{
670 struct dentry *child, *dir = file->f_path.dentry;
671 struct inode *inode;
672 struct qstr qname;
673 ino_t ino = 0;
674 unsigned type = DT_UNKNOWN;
675
676 qname.name = table->procname;
677 qname.len = strlen(table->procname);
678 qname.hash = full_name_hash(dir, qname.name, qname.len);
679
680 child = d_lookup(dir, &qname);
681 if (!child) {
682 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
683 child = d_alloc_parallel(dir, &qname, &wq);
684 if (IS_ERR(child))
685 return false;
686 if (d_in_lookup(child)) {
687 inode = proc_sys_make_inode(dir->d_sb, head, table);
688 if (!inode) {
689 d_lookup_done(child);
690 dput(child);
691 return false;
692 }
693 d_set_d_op(child, &proc_sys_dentry_operations);
694 d_add(child, inode);
695 }
696 }
697 inode = d_inode(child);
698 ino = inode->i_ino;
699 type = inode->i_mode >> 12;
700 dput(child);
701 return dir_emit(ctx, qname.name, qname.len, ino, type);
702}
703
704static bool proc_sys_link_fill_cache(struct file *file,
705 struct dir_context *ctx,
706 struct ctl_table_header *head,
707 struct ctl_table *table)
708{
709 bool ret = true;
710
711 head = sysctl_head_grab(head);
712 if (IS_ERR(head))
713 return false;
714
715 /* It is not an error if we can not follow the link ignore it */
716 if (sysctl_follow_link(&head, &table))
717 goto out;
718
719 ret = proc_sys_fill_cache(file, ctx, head, table);
720out:
721 sysctl_head_finish(head);
722 return ret;
723}
724
725static int scan(struct ctl_table_header *head, struct ctl_table *table,
726 unsigned long *pos, struct file *file,
727 struct dir_context *ctx)
728{
729 bool res;
730
731 if ((*pos)++ < ctx->pos)
732 return true;
733
734 if (unlikely(S_ISLNK(table->mode)))
735 res = proc_sys_link_fill_cache(file, ctx, head, table);
736 else
737 res = proc_sys_fill_cache(file, ctx, head, table);
738
739 if (res)
740 ctx->pos = *pos;
741
742 return res;
743}
744
745static int proc_sys_readdir(struct file *file, struct dir_context *ctx)
746{
747 struct ctl_table_header *head = grab_header(file_inode(file));
748 struct ctl_table_header *h = NULL;
749 struct ctl_table *entry;
750 struct ctl_dir *ctl_dir;
751 unsigned long pos;
752
753 if (IS_ERR(head))
754 return PTR_ERR(head);
755
756 ctl_dir = container_of(head, struct ctl_dir, header);
757
758 if (!dir_emit_dots(file, ctx))
759 goto out;
760
761 pos = 2;
762
763 for (first_entry(ctl_dir, &h, &entry); h; next_entry(&h, &entry)) {
764 if (!scan(h, entry, &pos, file, ctx)) {
765 sysctl_head_finish(h);
766 break;
767 }
768 }
769out:
770 sysctl_head_finish(head);
771 return 0;
772}
773
774static int proc_sys_permission(struct inode *inode, int mask)
775{
776 /*
777 * sysctl entries that are not writeable,
778 * are _NOT_ writeable, capabilities or not.
779 */
780 struct ctl_table_header *head;
781 struct ctl_table *table;
782 int error;
783
784 /* Executable files are not allowed under /proc/sys/ */
785 if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode))
786 return -EACCES;
787
788 head = grab_header(inode);
789 if (IS_ERR(head))
790 return PTR_ERR(head);
791
792 table = PROC_I(inode)->sysctl_entry;
793 if (!table) /* global root - r-xr-xr-x */
794 error = mask & MAY_WRITE ? -EACCES : 0;
795 else /* Use the permissions on the sysctl table entry */
796 error = sysctl_perm(head, table, mask & ~MAY_NOT_BLOCK);
797
798 sysctl_head_finish(head);
799 return error;
800}
801
802static int proc_sys_setattr(struct dentry *dentry, struct iattr *attr)
803{
804 struct inode *inode = d_inode(dentry);
805 int error;
806
807 if (attr->ia_valid & (ATTR_MODE | ATTR_UID | ATTR_GID))
808 return -EPERM;
809
810 error = setattr_prepare(dentry, attr);
811 if (error)
812 return error;
813
814 setattr_copy(inode, attr);
815 mark_inode_dirty(inode);
816 return 0;
817}
818
819static int proc_sys_getattr(const struct path *path, struct kstat *stat,
820 u32 request_mask, unsigned int query_flags)
821{
822 struct inode *inode = d_inode(path->dentry);
823 struct ctl_table_header *head = grab_header(inode);
824 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
825
826 if (IS_ERR(head))
827 return PTR_ERR(head);
828
829 generic_fillattr(inode, stat);
830 if (table)
831 stat->mode = (stat->mode & S_IFMT) | table->mode;
832
833 sysctl_head_finish(head);
834 return 0;
835}
836
837static const struct file_operations proc_sys_file_operations = {
838 .open = proc_sys_open,
839 .poll = proc_sys_poll,
840 .read = proc_sys_read,
841 .write = proc_sys_write,
842 .llseek = default_llseek,
843};
844
845static const struct file_operations proc_sys_dir_file_operations = {
846 .read = generic_read_dir,
847 .iterate_shared = proc_sys_readdir,
848 .llseek = generic_file_llseek,
849};
850
851static const struct inode_operations proc_sys_inode_operations = {
852 .permission = proc_sys_permission,
853 .setattr = proc_sys_setattr,
854 .getattr = proc_sys_getattr,
855};
856
857static const struct inode_operations proc_sys_dir_operations = {
858 .lookup = proc_sys_lookup,
859 .permission = proc_sys_permission,
860 .setattr = proc_sys_setattr,
861 .getattr = proc_sys_getattr,
862};
863
864static int proc_sys_revalidate(struct dentry *dentry, unsigned int flags)
865{
866 if (flags & LOOKUP_RCU)
867 return -ECHILD;
868 return !PROC_I(d_inode(dentry))->sysctl->unregistering;
869}
870
871static int proc_sys_delete(const struct dentry *dentry)
872{
873 return !!PROC_I(d_inode(dentry))->sysctl->unregistering;
874}
875
876static int sysctl_is_seen(struct ctl_table_header *p)
877{
878 struct ctl_table_set *set = p->set;
879 int res;
880 spin_lock(&sysctl_lock);
881 if (p->unregistering)
882 res = 0;
883 else if (!set->is_seen)
884 res = 1;
885 else
886 res = set->is_seen(set);
887 spin_unlock(&sysctl_lock);
888 return res;
889}
890
891static int proc_sys_compare(const struct dentry *dentry,
892 unsigned int len, const char *str, const struct qstr *name)
893{
894 struct ctl_table_header *head;
895 struct inode *inode;
896
897 /* Although proc doesn't have negative dentries, rcu-walk means
898 * that inode here can be NULL */
899 /* AV: can it, indeed? */
900 inode = d_inode_rcu(dentry);
901 if (!inode)
902 return 1;
903 if (name->len != len)
904 return 1;
905 if (memcmp(name->name, str, len))
906 return 1;
907 head = rcu_dereference(PROC_I(inode)->sysctl);
908 return !head || !sysctl_is_seen(head);
909}
910
911static const struct dentry_operations proc_sys_dentry_operations = {
912 .d_revalidate = proc_sys_revalidate,
913 .d_delete = proc_sys_delete,
914 .d_compare = proc_sys_compare,
915};
916
917static struct ctl_dir *find_subdir(struct ctl_dir *dir,
918 const char *name, int namelen)
919{
920 struct ctl_table_header *head;
921 struct ctl_table *entry;
922
923 entry = find_entry(&head, dir, name, namelen);
924 if (!entry)
925 return ERR_PTR(-ENOENT);
926 if (!S_ISDIR(entry->mode))
927 return ERR_PTR(-ENOTDIR);
928 return container_of(head, struct ctl_dir, header);
929}
930
931static struct ctl_dir *new_dir(struct ctl_table_set *set,
932 const char *name, int namelen)
933{
934 struct ctl_table *table;
935 struct ctl_dir *new;
936 struct ctl_node *node;
937 char *new_name;
938
939 new = kzalloc(sizeof(*new) + sizeof(struct ctl_node) +
940 sizeof(struct ctl_table)*2 + namelen + 1,
941 GFP_KERNEL);
942 if (!new)
943 return NULL;
944
945 node = (struct ctl_node *)(new + 1);
946 table = (struct ctl_table *)(node + 1);
947 new_name = (char *)(table + 2);
948 memcpy(new_name, name, namelen);
949 new_name[namelen] = '\0';
950 table[0].procname = new_name;
951 table[0].mode = S_IFDIR|S_IRUGO|S_IXUGO;
952 init_header(&new->header, set->dir.header.root, set, node, table);
953
954 return new;
955}
956
957/**
958 * get_subdir - find or create a subdir with the specified name.
959 * @dir: Directory to create the subdirectory in
960 * @name: The name of the subdirectory to find or create
961 * @namelen: The length of name
962 *
963 * Takes a directory with an elevated reference count so we know that
964 * if we drop the lock the directory will not go away. Upon success
965 * the reference is moved from @dir to the returned subdirectory.
966 * Upon error an error code is returned and the reference on @dir is
967 * simply dropped.
968 */
969static struct ctl_dir *get_subdir(struct ctl_dir *dir,
970 const char *name, int namelen)
971{
972 struct ctl_table_set *set = dir->header.set;
973 struct ctl_dir *subdir, *new = NULL;
974 int err;
975
976 spin_lock(&sysctl_lock);
977 subdir = find_subdir(dir, name, namelen);
978 if (!IS_ERR(subdir))
979 goto found;
980 if (PTR_ERR(subdir) != -ENOENT)
981 goto failed;
982
983 spin_unlock(&sysctl_lock);
984 new = new_dir(set, name, namelen);
985 spin_lock(&sysctl_lock);
986 subdir = ERR_PTR(-ENOMEM);
987 if (!new)
988 goto failed;
989
990 /* Was the subdir added while we dropped the lock? */
991 subdir = find_subdir(dir, name, namelen);
992 if (!IS_ERR(subdir))
993 goto found;
994 if (PTR_ERR(subdir) != -ENOENT)
995 goto failed;
996
997 /* Nope. Use the our freshly made directory entry. */
998 err = insert_header(dir, &new->header);
999 subdir = ERR_PTR(err);
1000 if (err)
1001 goto failed;
1002 subdir = new;
1003found:
1004 subdir->header.nreg++;
1005failed:
1006 if (IS_ERR(subdir)) {
1007 pr_err("sysctl could not get directory: ");
1008 sysctl_print_dir(dir);
1009 pr_cont("/%*.*s %ld\n",
1010 namelen, namelen, name, PTR_ERR(subdir));
1011 }
1012 drop_sysctl_table(&dir->header);
1013 if (new)
1014 drop_sysctl_table(&new->header);
1015 spin_unlock(&sysctl_lock);
1016 return subdir;
1017}
1018
1019static struct ctl_dir *xlate_dir(struct ctl_table_set *set, struct ctl_dir *dir)
1020{
1021 struct ctl_dir *parent;
1022 const char *procname;
1023 if (!dir->header.parent)
1024 return &set->dir;
1025 parent = xlate_dir(set, dir->header.parent);
1026 if (IS_ERR(parent))
1027 return parent;
1028 procname = dir->header.ctl_table[0].procname;
1029 return find_subdir(parent, procname, strlen(procname));
1030}
1031
1032static int sysctl_follow_link(struct ctl_table_header **phead,
1033 struct ctl_table **pentry)
1034{
1035 struct ctl_table_header *head;
1036 struct ctl_table_root *root;
1037 struct ctl_table_set *set;
1038 struct ctl_table *entry;
1039 struct ctl_dir *dir;
1040 int ret;
1041
1042 ret = 0;
1043 spin_lock(&sysctl_lock);
1044 root = (*pentry)->data;
1045 set = lookup_header_set(root);
1046 dir = xlate_dir(set, (*phead)->parent);
1047 if (IS_ERR(dir))
1048 ret = PTR_ERR(dir);
1049 else {
1050 const char *procname = (*pentry)->procname;
1051 head = NULL;
1052 entry = find_entry(&head, dir, procname, strlen(procname));
1053 ret = -ENOENT;
1054 if (entry && use_table(head)) {
1055 unuse_table(*phead);
1056 *phead = head;
1057 *pentry = entry;
1058 ret = 0;
1059 }
1060 }
1061
1062 spin_unlock(&sysctl_lock);
1063 return ret;
1064}
1065
1066static int sysctl_err(const char *path, struct ctl_table *table, char *fmt, ...)
1067{
1068 struct va_format vaf;
1069 va_list args;
1070
1071 va_start(args, fmt);
1072 vaf.fmt = fmt;
1073 vaf.va = &args;
1074
1075 pr_err("sysctl table check failed: %s/%s %pV\n",
1076 path, table->procname, &vaf);
1077
1078 va_end(args);
1079 return -EINVAL;
1080}
1081
1082static int sysctl_check_table_array(const char *path, struct ctl_table *table)
1083{
1084 int err = 0;
1085
1086 if ((table->proc_handler == proc_douintvec) ||
1087 (table->proc_handler == proc_douintvec_minmax)) {
1088 if (table->maxlen != sizeof(unsigned int))
1089 err |= sysctl_err(path, table, "array not allowed");
1090 }
1091
1092 return err;
1093}
1094
1095static int sysctl_check_table(const char *path, struct ctl_table *table)
1096{
1097 int err = 0;
1098 for (; table->procname; table++) {
1099 if (table->child)
1100 err |= sysctl_err(path, table, "Not a file");
1101
1102 if ((table->proc_handler == proc_dostring) ||
1103 (table->proc_handler == proc_dointvec) ||
1104 (table->proc_handler == proc_douintvec) ||
1105 (table->proc_handler == proc_douintvec_minmax) ||
1106 (table->proc_handler == proc_dointvec_minmax) ||
1107 (table->proc_handler == proc_dointvec_jiffies) ||
1108 (table->proc_handler == proc_dointvec_userhz_jiffies) ||
1109 (table->proc_handler == proc_dointvec_ms_jiffies) ||
1110 (table->proc_handler == proc_doulongvec_minmax) ||
1111 (table->proc_handler == proc_doulongvec_ms_jiffies_minmax)) {
1112 if (!table->data)
1113 err |= sysctl_err(path, table, "No data");
1114 if (!table->maxlen)
1115 err |= sysctl_err(path, table, "No maxlen");
1116 else
1117 err |= sysctl_check_table_array(path, table);
1118 }
1119 if (!table->proc_handler)
1120 err |= sysctl_err(path, table, "No proc_handler");
1121
1122 if ((table->mode & (S_IRUGO|S_IWUGO)) != table->mode)
1123 err |= sysctl_err(path, table, "bogus .mode 0%o",
1124 table->mode);
1125 }
1126 return err;
1127}
1128
1129static struct ctl_table_header *new_links(struct ctl_dir *dir, struct ctl_table *table,
1130 struct ctl_table_root *link_root)
1131{
1132 struct ctl_table *link_table, *entry, *link;
1133 struct ctl_table_header *links;
1134 struct ctl_node *node;
1135 char *link_name;
1136 int nr_entries, name_bytes;
1137
1138 name_bytes = 0;
1139 nr_entries = 0;
1140 for (entry = table; entry->procname; entry++) {
1141 nr_entries++;
1142 name_bytes += strlen(entry->procname) + 1;
1143 }
1144
1145 links = kzalloc(sizeof(struct ctl_table_header) +
1146 sizeof(struct ctl_node)*nr_entries +
1147 sizeof(struct ctl_table)*(nr_entries + 1) +
1148 name_bytes,
1149 GFP_KERNEL);
1150
1151 if (!links)
1152 return NULL;
1153
1154 node = (struct ctl_node *)(links + 1);
1155 link_table = (struct ctl_table *)(node + nr_entries);
1156 link_name = (char *)&link_table[nr_entries + 1];
1157
1158 for (link = link_table, entry = table; entry->procname; link++, entry++) {
1159 int len = strlen(entry->procname) + 1;
1160 memcpy(link_name, entry->procname, len);
1161 link->procname = link_name;
1162 link->mode = S_IFLNK|S_IRWXUGO;
1163 link->data = link_root;
1164 link_name += len;
1165 }
1166 init_header(links, dir->header.root, dir->header.set, node, link_table);
1167 links->nreg = nr_entries;
1168
1169 return links;
1170}
1171
1172static bool get_links(struct ctl_dir *dir,
1173 struct ctl_table *table, struct ctl_table_root *link_root)
1174{
1175 struct ctl_table_header *head;
1176 struct ctl_table *entry, *link;
1177
1178 /* Are there links available for every entry in table? */
1179 for (entry = table; entry->procname; entry++) {
1180 const char *procname = entry->procname;
1181 link = find_entry(&head, dir, procname, strlen(procname));
1182 if (!link)
1183 return false;
1184 if (S_ISDIR(link->mode) && S_ISDIR(entry->mode))
1185 continue;
1186 if (S_ISLNK(link->mode) && (link->data == link_root))
1187 continue;
1188 return false;
1189 }
1190
1191 /* The checks passed. Increase the registration count on the links */
1192 for (entry = table; entry->procname; entry++) {
1193 const char *procname = entry->procname;
1194 link = find_entry(&head, dir, procname, strlen(procname));
1195 head->nreg++;
1196 }
1197 return true;
1198}
1199
1200static int insert_links(struct ctl_table_header *head)
1201{
1202 struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1203 struct ctl_dir *core_parent = NULL;
1204 struct ctl_table_header *links;
1205 int err;
1206
1207 if (head->set == root_set)
1208 return 0;
1209
1210 core_parent = xlate_dir(root_set, head->parent);
1211 if (IS_ERR(core_parent))
1212 return 0;
1213
1214 if (get_links(core_parent, head->ctl_table, head->root))
1215 return 0;
1216
1217 core_parent->header.nreg++;
1218 spin_unlock(&sysctl_lock);
1219
1220 links = new_links(core_parent, head->ctl_table, head->root);
1221
1222 spin_lock(&sysctl_lock);
1223 err = -ENOMEM;
1224 if (!links)
1225 goto out;
1226
1227 err = 0;
1228 if (get_links(core_parent, head->ctl_table, head->root)) {
1229 kfree(links);
1230 goto out;
1231 }
1232
1233 err = insert_header(core_parent, links);
1234 if (err)
1235 kfree(links);
1236out:
1237 drop_sysctl_table(&core_parent->header);
1238 return err;
1239}
1240
1241/**
1242 * __register_sysctl_table - register a leaf sysctl table
1243 * @set: Sysctl tree to register on
1244 * @path: The path to the directory the sysctl table is in.
1245 * @table: the top-level table structure
1246 *
1247 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1248 * array. A completely 0 filled entry terminates the table.
1249 *
1250 * The members of the &struct ctl_table structure are used as follows:
1251 *
1252 * procname - the name of the sysctl file under /proc/sys. Set to %NULL to not
1253 * enter a sysctl file
1254 *
1255 * data - a pointer to data for use by proc_handler
1256 *
1257 * maxlen - the maximum size in bytes of the data
1258 *
1259 * mode - the file permissions for the /proc/sys file
1260 *
1261 * child - must be %NULL.
1262 *
1263 * proc_handler - the text handler routine (described below)
1264 *
1265 * extra1, extra2 - extra pointers usable by the proc handler routines
1266 *
1267 * Leaf nodes in the sysctl tree will be represented by a single file
1268 * under /proc; non-leaf nodes will be represented by directories.
1269 *
1270 * There must be a proc_handler routine for any terminal nodes.
1271 * Several default handlers are available to cover common cases -
1272 *
1273 * proc_dostring(), proc_dointvec(), proc_dointvec_jiffies(),
1274 * proc_dointvec_userhz_jiffies(), proc_dointvec_minmax(),
1275 * proc_doulongvec_ms_jiffies_minmax(), proc_doulongvec_minmax()
1276 *
1277 * It is the handler's job to read the input buffer from user memory
1278 * and process it. The handler should return 0 on success.
1279 *
1280 * This routine returns %NULL on a failure to register, and a pointer
1281 * to the table header on success.
1282 */
1283struct ctl_table_header *__register_sysctl_table(
1284 struct ctl_table_set *set,
1285 const char *path, struct ctl_table *table)
1286{
1287 struct ctl_table_root *root = set->dir.header.root;
1288 struct ctl_table_header *header;
1289 const char *name, *nextname;
1290 struct ctl_dir *dir;
1291 struct ctl_table *entry;
1292 struct ctl_node *node;
1293 int nr_entries = 0;
1294
1295 for (entry = table; entry->procname; entry++)
1296 nr_entries++;
1297
1298 header = kzalloc(sizeof(struct ctl_table_header) +
1299 sizeof(struct ctl_node)*nr_entries, GFP_KERNEL);
1300 if (!header)
1301 return NULL;
1302
1303 node = (struct ctl_node *)(header + 1);
1304 init_header(header, root, set, node, table);
1305 if (sysctl_check_table(path, table))
1306 goto fail;
1307
1308 spin_lock(&sysctl_lock);
1309 dir = &set->dir;
1310 /* Reference moved down the diretory tree get_subdir */
1311 dir->header.nreg++;
1312 spin_unlock(&sysctl_lock);
1313
1314 /* Find the directory for the ctl_table */
1315 for (name = path; name; name = nextname) {
1316 int namelen;
1317 nextname = strchr(name, '/');
1318 if (nextname) {
1319 namelen = nextname - name;
1320 nextname++;
1321 } else {
1322 namelen = strlen(name);
1323 }
1324 if (namelen == 0)
1325 continue;
1326
1327 dir = get_subdir(dir, name, namelen);
1328 if (IS_ERR(dir))
1329 goto fail;
1330 }
1331
1332 spin_lock(&sysctl_lock);
1333 if (insert_header(dir, header))
1334 goto fail_put_dir_locked;
1335
1336 drop_sysctl_table(&dir->header);
1337 spin_unlock(&sysctl_lock);
1338
1339 return header;
1340
1341fail_put_dir_locked:
1342 drop_sysctl_table(&dir->header);
1343 spin_unlock(&sysctl_lock);
1344fail:
1345 kfree(header);
1346 dump_stack();
1347 return NULL;
1348}
1349
1350/**
1351 * register_sysctl - register a sysctl table
1352 * @path: The path to the directory the sysctl table is in.
1353 * @table: the table structure
1354 *
1355 * Register a sysctl table. @table should be a filled in ctl_table
1356 * array. A completely 0 filled entry terminates the table.
1357 *
1358 * See __register_sysctl_table for more details.
1359 */
1360struct ctl_table_header *register_sysctl(const char *path, struct ctl_table *table)
1361{
1362 return __register_sysctl_table(&sysctl_table_root.default_set,
1363 path, table);
1364}
1365EXPORT_SYMBOL(register_sysctl);
1366
1367static char *append_path(const char *path, char *pos, const char *name)
1368{
1369 int namelen;
1370 namelen = strlen(name);
1371 if (((pos - path) + namelen + 2) >= PATH_MAX)
1372 return NULL;
1373 memcpy(pos, name, namelen);
1374 pos[namelen] = '/';
1375 pos[namelen + 1] = '\0';
1376 pos += namelen + 1;
1377 return pos;
1378}
1379
1380static int count_subheaders(struct ctl_table *table)
1381{
1382 int has_files = 0;
1383 int nr_subheaders = 0;
1384 struct ctl_table *entry;
1385
1386 /* special case: no directory and empty directory */
1387 if (!table || !table->procname)
1388 return 1;
1389
1390 for (entry = table; entry->procname; entry++) {
1391 if (entry->child)
1392 nr_subheaders += count_subheaders(entry->child);
1393 else
1394 has_files = 1;
1395 }
1396 return nr_subheaders + has_files;
1397}
1398
1399static int register_leaf_sysctl_tables(const char *path, char *pos,
1400 struct ctl_table_header ***subheader, struct ctl_table_set *set,
1401 struct ctl_table *table)
1402{
1403 struct ctl_table *ctl_table_arg = NULL;
1404 struct ctl_table *entry, *files;
1405 int nr_files = 0;
1406 int nr_dirs = 0;
1407 int err = -ENOMEM;
1408
1409 for (entry = table; entry->procname; entry++) {
1410 if (entry->child)
1411 nr_dirs++;
1412 else
1413 nr_files++;
1414 }
1415
1416 files = table;
1417 /* If there are mixed files and directories we need a new table */
1418 if (nr_dirs && nr_files) {
1419 struct ctl_table *new;
1420 files = kzalloc(sizeof(struct ctl_table) * (nr_files + 1),
1421 GFP_KERNEL);
1422 if (!files)
1423 goto out;
1424
1425 ctl_table_arg = files;
1426 for (new = files, entry = table; entry->procname; entry++) {
1427 if (entry->child)
1428 continue;
1429 *new = *entry;
1430 new++;
1431 }
1432 }
1433
1434 /* Register everything except a directory full of subdirectories */
1435 if (nr_files || !nr_dirs) {
1436 struct ctl_table_header *header;
1437 header = __register_sysctl_table(set, path, files);
1438 if (!header) {
1439 kfree(ctl_table_arg);
1440 goto out;
1441 }
1442
1443 /* Remember if we need to free the file table */
1444 header->ctl_table_arg = ctl_table_arg;
1445 **subheader = header;
1446 (*subheader)++;
1447 }
1448
1449 /* Recurse into the subdirectories. */
1450 for (entry = table; entry->procname; entry++) {
1451 char *child_pos;
1452
1453 if (!entry->child)
1454 continue;
1455
1456 err = -ENAMETOOLONG;
1457 child_pos = append_path(path, pos, entry->procname);
1458 if (!child_pos)
1459 goto out;
1460
1461 err = register_leaf_sysctl_tables(path, child_pos, subheader,
1462 set, entry->child);
1463 pos[0] = '\0';
1464 if (err)
1465 goto out;
1466 }
1467 err = 0;
1468out:
1469 /* On failure our caller will unregister all registered subheaders */
1470 return err;
1471}
1472
1473/**
1474 * __register_sysctl_paths - register a sysctl table hierarchy
1475 * @set: Sysctl tree to register on
1476 * @path: The path to the directory the sysctl table is in.
1477 * @table: the top-level table structure
1478 *
1479 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1480 * array. A completely 0 filled entry terminates the table.
1481 *
1482 * See __register_sysctl_table for more details.
1483 */
1484struct ctl_table_header *__register_sysctl_paths(
1485 struct ctl_table_set *set,
1486 const struct ctl_path *path, struct ctl_table *table)
1487{
1488 struct ctl_table *ctl_table_arg = table;
1489 int nr_subheaders = count_subheaders(table);
1490 struct ctl_table_header *header = NULL, **subheaders, **subheader;
1491 const struct ctl_path *component;
1492 char *new_path, *pos;
1493
1494 pos = new_path = kmalloc(PATH_MAX, GFP_KERNEL);
1495 if (!new_path)
1496 return NULL;
1497
1498 pos[0] = '\0';
1499 for (component = path; component->procname; component++) {
1500 pos = append_path(new_path, pos, component->procname);
1501 if (!pos)
1502 goto out;
1503 }
1504 while (table->procname && table->child && !table[1].procname) {
1505 pos = append_path(new_path, pos, table->procname);
1506 if (!pos)
1507 goto out;
1508 table = table->child;
1509 }
1510 if (nr_subheaders == 1) {
1511 header = __register_sysctl_table(set, new_path, table);
1512 if (header)
1513 header->ctl_table_arg = ctl_table_arg;
1514 } else {
1515 header = kzalloc(sizeof(*header) +
1516 sizeof(*subheaders)*nr_subheaders, GFP_KERNEL);
1517 if (!header)
1518 goto out;
1519
1520 subheaders = (struct ctl_table_header **) (header + 1);
1521 subheader = subheaders;
1522 header->ctl_table_arg = ctl_table_arg;
1523
1524 if (register_leaf_sysctl_tables(new_path, pos, &subheader,
1525 set, table))
1526 goto err_register_leaves;
1527 }
1528
1529out:
1530 kfree(new_path);
1531 return header;
1532
1533err_register_leaves:
1534 while (subheader > subheaders) {
1535 struct ctl_table_header *subh = *(--subheader);
1536 struct ctl_table *table = subh->ctl_table_arg;
1537 unregister_sysctl_table(subh);
1538 kfree(table);
1539 }
1540 kfree(header);
1541 header = NULL;
1542 goto out;
1543}
1544
1545/**
1546 * register_sysctl_table_path - register a sysctl table hierarchy
1547 * @path: The path to the directory the sysctl table is in.
1548 * @table: the top-level table structure
1549 *
1550 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1551 * array. A completely 0 filled entry terminates the table.
1552 *
1553 * See __register_sysctl_paths for more details.
1554 */
1555struct ctl_table_header *register_sysctl_paths(const struct ctl_path *path,
1556 struct ctl_table *table)
1557{
1558 return __register_sysctl_paths(&sysctl_table_root.default_set,
1559 path, table);
1560}
1561EXPORT_SYMBOL(register_sysctl_paths);
1562
1563/**
1564 * register_sysctl_table - register a sysctl table hierarchy
1565 * @table: the top-level table structure
1566 *
1567 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1568 * array. A completely 0 filled entry terminates the table.
1569 *
1570 * See register_sysctl_paths for more details.
1571 */
1572struct ctl_table_header *register_sysctl_table(struct ctl_table *table)
1573{
1574 static const struct ctl_path null_path[] = { {} };
1575
1576 return register_sysctl_paths(null_path, table);
1577}
1578EXPORT_SYMBOL(register_sysctl_table);
1579
1580static void put_links(struct ctl_table_header *header)
1581{
1582 struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1583 struct ctl_table_root *root = header->root;
1584 struct ctl_dir *parent = header->parent;
1585 struct ctl_dir *core_parent;
1586 struct ctl_table *entry;
1587
1588 if (header->set == root_set)
1589 return;
1590
1591 core_parent = xlate_dir(root_set, parent);
1592 if (IS_ERR(core_parent))
1593 return;
1594
1595 for (entry = header->ctl_table; entry->procname; entry++) {
1596 struct ctl_table_header *link_head;
1597 struct ctl_table *link;
1598 const char *name = entry->procname;
1599
1600 link = find_entry(&link_head, core_parent, name, strlen(name));
1601 if (link &&
1602 ((S_ISDIR(link->mode) && S_ISDIR(entry->mode)) ||
1603 (S_ISLNK(link->mode) && (link->data == root)))) {
1604 drop_sysctl_table(link_head);
1605 }
1606 else {
1607 pr_err("sysctl link missing during unregister: ");
1608 sysctl_print_dir(parent);
1609 pr_cont("/%s\n", name);
1610 }
1611 }
1612}
1613
1614static void drop_sysctl_table(struct ctl_table_header *header)
1615{
1616 struct ctl_dir *parent = header->parent;
1617
1618 if (--header->nreg)
1619 return;
1620
1621 put_links(header);
1622 start_unregistering(header);
1623 if (!--header->count)
1624 kfree_rcu(header, rcu);
1625
1626 if (parent)
1627 drop_sysctl_table(&parent->header);
1628}
1629
1630/**
1631 * unregister_sysctl_table - unregister a sysctl table hierarchy
1632 * @header: the header returned from register_sysctl_table
1633 *
1634 * Unregisters the sysctl table and all children. proc entries may not
1635 * actually be removed until they are no longer used by anyone.
1636 */
1637void unregister_sysctl_table(struct ctl_table_header * header)
1638{
1639 int nr_subheaders;
1640 might_sleep();
1641
1642 if (header == NULL)
1643 return;
1644
1645 nr_subheaders = count_subheaders(header->ctl_table_arg);
1646 if (unlikely(nr_subheaders > 1)) {
1647 struct ctl_table_header **subheaders;
1648 int i;
1649
1650 subheaders = (struct ctl_table_header **)(header + 1);
1651 for (i = nr_subheaders -1; i >= 0; i--) {
1652 struct ctl_table_header *subh = subheaders[i];
1653 struct ctl_table *table = subh->ctl_table_arg;
1654 unregister_sysctl_table(subh);
1655 kfree(table);
1656 }
1657 kfree(header);
1658 return;
1659 }
1660
1661 spin_lock(&sysctl_lock);
1662 drop_sysctl_table(header);
1663 spin_unlock(&sysctl_lock);
1664}
1665EXPORT_SYMBOL(unregister_sysctl_table);
1666
1667void setup_sysctl_set(struct ctl_table_set *set,
1668 struct ctl_table_root *root,
1669 int (*is_seen)(struct ctl_table_set *))
1670{
1671 memset(set, 0, sizeof(*set));
1672 set->is_seen = is_seen;
1673 init_header(&set->dir.header, root, set, NULL, root_table);
1674}
1675
1676void retire_sysctl_set(struct ctl_table_set *set)
1677{
1678 WARN_ON(!RB_EMPTY_ROOT(&set->dir.root));
1679}
1680
1681int __init proc_sys_init(void)
1682{
1683 struct proc_dir_entry *proc_sys_root;
1684
1685 proc_sys_root = proc_mkdir("sys", NULL);
1686 proc_sys_root->proc_iops = &proc_sys_dir_operations;
1687 proc_sys_root->proc_fops = &proc_sys_dir_file_operations;
1688 proc_sys_root->nlink = 0;
1689
1690 return sysctl_init();
1691}
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * /proc/sys support
4 */
5#include <linux/init.h>
6#include <linux/sysctl.h>
7#include <linux/poll.h>
8#include <linux/proc_fs.h>
9#include <linux/printk.h>
10#include <linux/security.h>
11#include <linux/sched.h>
12#include <linux/cred.h>
13#include <linux/namei.h>
14#include <linux/mm.h>
15#include <linux/module.h>
16#include <linux/bpf-cgroup.h>
17#include "internal.h"
18
19static const struct dentry_operations proc_sys_dentry_operations;
20static const struct file_operations proc_sys_file_operations;
21static const struct inode_operations proc_sys_inode_operations;
22static const struct file_operations proc_sys_dir_file_operations;
23static const struct inode_operations proc_sys_dir_operations;
24
25/* shared constants to be used in various sysctls */
26const int sysctl_vals[] = { 0, 1, INT_MAX };
27EXPORT_SYMBOL(sysctl_vals);
28
29/* Support for permanently empty directories */
30
31struct ctl_table sysctl_mount_point[] = {
32 { }
33};
34
35static bool is_empty_dir(struct ctl_table_header *head)
36{
37 return head->ctl_table[0].child == sysctl_mount_point;
38}
39
40static void set_empty_dir(struct ctl_dir *dir)
41{
42 dir->header.ctl_table[0].child = sysctl_mount_point;
43}
44
45static void clear_empty_dir(struct ctl_dir *dir)
46
47{
48 dir->header.ctl_table[0].child = NULL;
49}
50
51void proc_sys_poll_notify(struct ctl_table_poll *poll)
52{
53 if (!poll)
54 return;
55
56 atomic_inc(&poll->event);
57 wake_up_interruptible(&poll->wait);
58}
59
60static struct ctl_table root_table[] = {
61 {
62 .procname = "",
63 .mode = S_IFDIR|S_IRUGO|S_IXUGO,
64 },
65 { }
66};
67static struct ctl_table_root sysctl_table_root = {
68 .default_set.dir.header = {
69 {{.count = 1,
70 .nreg = 1,
71 .ctl_table = root_table }},
72 .ctl_table_arg = root_table,
73 .root = &sysctl_table_root,
74 .set = &sysctl_table_root.default_set,
75 },
76};
77
78static DEFINE_SPINLOCK(sysctl_lock);
79
80static void drop_sysctl_table(struct ctl_table_header *header);
81static int sysctl_follow_link(struct ctl_table_header **phead,
82 struct ctl_table **pentry);
83static int insert_links(struct ctl_table_header *head);
84static void put_links(struct ctl_table_header *header);
85
86static void sysctl_print_dir(struct ctl_dir *dir)
87{
88 if (dir->header.parent)
89 sysctl_print_dir(dir->header.parent);
90 pr_cont("%s/", dir->header.ctl_table[0].procname);
91}
92
93static int namecmp(const char *name1, int len1, const char *name2, int len2)
94{
95 int minlen;
96 int cmp;
97
98 minlen = len1;
99 if (minlen > len2)
100 minlen = len2;
101
102 cmp = memcmp(name1, name2, minlen);
103 if (cmp == 0)
104 cmp = len1 - len2;
105 return cmp;
106}
107
108/* Called under sysctl_lock */
109static struct ctl_table *find_entry(struct ctl_table_header **phead,
110 struct ctl_dir *dir, const char *name, int namelen)
111{
112 struct ctl_table_header *head;
113 struct ctl_table *entry;
114 struct rb_node *node = dir->root.rb_node;
115
116 while (node)
117 {
118 struct ctl_node *ctl_node;
119 const char *procname;
120 int cmp;
121
122 ctl_node = rb_entry(node, struct ctl_node, node);
123 head = ctl_node->header;
124 entry = &head->ctl_table[ctl_node - head->node];
125 procname = entry->procname;
126
127 cmp = namecmp(name, namelen, procname, strlen(procname));
128 if (cmp < 0)
129 node = node->rb_left;
130 else if (cmp > 0)
131 node = node->rb_right;
132 else {
133 *phead = head;
134 return entry;
135 }
136 }
137 return NULL;
138}
139
140static int insert_entry(struct ctl_table_header *head, struct ctl_table *entry)
141{
142 struct rb_node *node = &head->node[entry - head->ctl_table].node;
143 struct rb_node **p = &head->parent->root.rb_node;
144 struct rb_node *parent = NULL;
145 const char *name = entry->procname;
146 int namelen = strlen(name);
147
148 while (*p) {
149 struct ctl_table_header *parent_head;
150 struct ctl_table *parent_entry;
151 struct ctl_node *parent_node;
152 const char *parent_name;
153 int cmp;
154
155 parent = *p;
156 parent_node = rb_entry(parent, struct ctl_node, node);
157 parent_head = parent_node->header;
158 parent_entry = &parent_head->ctl_table[parent_node - parent_head->node];
159 parent_name = parent_entry->procname;
160
161 cmp = namecmp(name, namelen, parent_name, strlen(parent_name));
162 if (cmp < 0)
163 p = &(*p)->rb_left;
164 else if (cmp > 0)
165 p = &(*p)->rb_right;
166 else {
167 pr_err("sysctl duplicate entry: ");
168 sysctl_print_dir(head->parent);
169 pr_cont("/%s\n", entry->procname);
170 return -EEXIST;
171 }
172 }
173
174 rb_link_node(node, parent, p);
175 rb_insert_color(node, &head->parent->root);
176 return 0;
177}
178
179static void erase_entry(struct ctl_table_header *head, struct ctl_table *entry)
180{
181 struct rb_node *node = &head->node[entry - head->ctl_table].node;
182
183 rb_erase(node, &head->parent->root);
184}
185
186static void init_header(struct ctl_table_header *head,
187 struct ctl_table_root *root, struct ctl_table_set *set,
188 struct ctl_node *node, struct ctl_table *table)
189{
190 head->ctl_table = table;
191 head->ctl_table_arg = table;
192 head->used = 0;
193 head->count = 1;
194 head->nreg = 1;
195 head->unregistering = NULL;
196 head->root = root;
197 head->set = set;
198 head->parent = NULL;
199 head->node = node;
200 INIT_HLIST_HEAD(&head->inodes);
201 if (node) {
202 struct ctl_table *entry;
203 for (entry = table; entry->procname; entry++, node++)
204 node->header = head;
205 }
206}
207
208static void erase_header(struct ctl_table_header *head)
209{
210 struct ctl_table *entry;
211 for (entry = head->ctl_table; entry->procname; entry++)
212 erase_entry(head, entry);
213}
214
215static int insert_header(struct ctl_dir *dir, struct ctl_table_header *header)
216{
217 struct ctl_table *entry;
218 int err;
219
220 /* Is this a permanently empty directory? */
221 if (is_empty_dir(&dir->header))
222 return -EROFS;
223
224 /* Am I creating a permanently empty directory? */
225 if (header->ctl_table == sysctl_mount_point) {
226 if (!RB_EMPTY_ROOT(&dir->root))
227 return -EINVAL;
228 set_empty_dir(dir);
229 }
230
231 dir->header.nreg++;
232 header->parent = dir;
233 err = insert_links(header);
234 if (err)
235 goto fail_links;
236 for (entry = header->ctl_table; entry->procname; entry++) {
237 err = insert_entry(header, entry);
238 if (err)
239 goto fail;
240 }
241 return 0;
242fail:
243 erase_header(header);
244 put_links(header);
245fail_links:
246 if (header->ctl_table == sysctl_mount_point)
247 clear_empty_dir(dir);
248 header->parent = NULL;
249 drop_sysctl_table(&dir->header);
250 return err;
251}
252
253/* called under sysctl_lock */
254static int use_table(struct ctl_table_header *p)
255{
256 if (unlikely(p->unregistering))
257 return 0;
258 p->used++;
259 return 1;
260}
261
262/* called under sysctl_lock */
263static void unuse_table(struct ctl_table_header *p)
264{
265 if (!--p->used)
266 if (unlikely(p->unregistering))
267 complete(p->unregistering);
268}
269
270static void proc_sys_prune_dcache(struct ctl_table_header *head)
271{
272 struct inode *inode;
273 struct proc_inode *ei;
274 struct hlist_node *node;
275 struct super_block *sb;
276
277 rcu_read_lock();
278 for (;;) {
279 node = hlist_first_rcu(&head->inodes);
280 if (!node)
281 break;
282 ei = hlist_entry(node, struct proc_inode, sysctl_inodes);
283 spin_lock(&sysctl_lock);
284 hlist_del_init_rcu(&ei->sysctl_inodes);
285 spin_unlock(&sysctl_lock);
286
287 inode = &ei->vfs_inode;
288 sb = inode->i_sb;
289 if (!atomic_inc_not_zero(&sb->s_active))
290 continue;
291 inode = igrab(inode);
292 rcu_read_unlock();
293 if (unlikely(!inode)) {
294 deactivate_super(sb);
295 rcu_read_lock();
296 continue;
297 }
298
299 d_prune_aliases(inode);
300 iput(inode);
301 deactivate_super(sb);
302
303 rcu_read_lock();
304 }
305 rcu_read_unlock();
306}
307
308/* called under sysctl_lock, will reacquire if has to wait */
309static void start_unregistering(struct ctl_table_header *p)
310{
311 /*
312 * if p->used is 0, nobody will ever touch that entry again;
313 * we'll eliminate all paths to it before dropping sysctl_lock
314 */
315 if (unlikely(p->used)) {
316 struct completion wait;
317 init_completion(&wait);
318 p->unregistering = &wait;
319 spin_unlock(&sysctl_lock);
320 wait_for_completion(&wait);
321 } else {
322 /* anything non-NULL; we'll never dereference it */
323 p->unregistering = ERR_PTR(-EINVAL);
324 spin_unlock(&sysctl_lock);
325 }
326 /*
327 * Prune dentries for unregistered sysctls: namespaced sysctls
328 * can have duplicate names and contaminate dcache very badly.
329 */
330 proc_sys_prune_dcache(p);
331 /*
332 * do not remove from the list until nobody holds it; walking the
333 * list in do_sysctl() relies on that.
334 */
335 spin_lock(&sysctl_lock);
336 erase_header(p);
337}
338
339static struct ctl_table_header *sysctl_head_grab(struct ctl_table_header *head)
340{
341 BUG_ON(!head);
342 spin_lock(&sysctl_lock);
343 if (!use_table(head))
344 head = ERR_PTR(-ENOENT);
345 spin_unlock(&sysctl_lock);
346 return head;
347}
348
349static void sysctl_head_finish(struct ctl_table_header *head)
350{
351 if (!head)
352 return;
353 spin_lock(&sysctl_lock);
354 unuse_table(head);
355 spin_unlock(&sysctl_lock);
356}
357
358static struct ctl_table_set *
359lookup_header_set(struct ctl_table_root *root)
360{
361 struct ctl_table_set *set = &root->default_set;
362 if (root->lookup)
363 set = root->lookup(root);
364 return set;
365}
366
367static struct ctl_table *lookup_entry(struct ctl_table_header **phead,
368 struct ctl_dir *dir,
369 const char *name, int namelen)
370{
371 struct ctl_table_header *head;
372 struct ctl_table *entry;
373
374 spin_lock(&sysctl_lock);
375 entry = find_entry(&head, dir, name, namelen);
376 if (entry && use_table(head))
377 *phead = head;
378 else
379 entry = NULL;
380 spin_unlock(&sysctl_lock);
381 return entry;
382}
383
384static struct ctl_node *first_usable_entry(struct rb_node *node)
385{
386 struct ctl_node *ctl_node;
387
388 for (;node; node = rb_next(node)) {
389 ctl_node = rb_entry(node, struct ctl_node, node);
390 if (use_table(ctl_node->header))
391 return ctl_node;
392 }
393 return NULL;
394}
395
396static void first_entry(struct ctl_dir *dir,
397 struct ctl_table_header **phead, struct ctl_table **pentry)
398{
399 struct ctl_table_header *head = NULL;
400 struct ctl_table *entry = NULL;
401 struct ctl_node *ctl_node;
402
403 spin_lock(&sysctl_lock);
404 ctl_node = first_usable_entry(rb_first(&dir->root));
405 spin_unlock(&sysctl_lock);
406 if (ctl_node) {
407 head = ctl_node->header;
408 entry = &head->ctl_table[ctl_node - head->node];
409 }
410 *phead = head;
411 *pentry = entry;
412}
413
414static void next_entry(struct ctl_table_header **phead, struct ctl_table **pentry)
415{
416 struct ctl_table_header *head = *phead;
417 struct ctl_table *entry = *pentry;
418 struct ctl_node *ctl_node = &head->node[entry - head->ctl_table];
419
420 spin_lock(&sysctl_lock);
421 unuse_table(head);
422
423 ctl_node = first_usable_entry(rb_next(&ctl_node->node));
424 spin_unlock(&sysctl_lock);
425 head = NULL;
426 if (ctl_node) {
427 head = ctl_node->header;
428 entry = &head->ctl_table[ctl_node - head->node];
429 }
430 *phead = head;
431 *pentry = entry;
432}
433
434/*
435 * sysctl_perm does NOT grant the superuser all rights automatically, because
436 * some sysctl variables are readonly even to root.
437 */
438
439static int test_perm(int mode, int op)
440{
441 if (uid_eq(current_euid(), GLOBAL_ROOT_UID))
442 mode >>= 6;
443 else if (in_egroup_p(GLOBAL_ROOT_GID))
444 mode >>= 3;
445 if ((op & ~mode & (MAY_READ|MAY_WRITE|MAY_EXEC)) == 0)
446 return 0;
447 return -EACCES;
448}
449
450static int sysctl_perm(struct ctl_table_header *head, struct ctl_table *table, int op)
451{
452 struct ctl_table_root *root = head->root;
453 int mode;
454
455 if (root->permissions)
456 mode = root->permissions(head, table);
457 else
458 mode = table->mode;
459
460 return test_perm(mode, op);
461}
462
463static struct inode *proc_sys_make_inode(struct super_block *sb,
464 struct ctl_table_header *head, struct ctl_table *table)
465{
466 struct ctl_table_root *root = head->root;
467 struct inode *inode;
468 struct proc_inode *ei;
469
470 inode = new_inode(sb);
471 if (!inode)
472 return ERR_PTR(-ENOMEM);
473
474 inode->i_ino = get_next_ino();
475
476 ei = PROC_I(inode);
477
478 spin_lock(&sysctl_lock);
479 if (unlikely(head->unregistering)) {
480 spin_unlock(&sysctl_lock);
481 iput(inode);
482 return ERR_PTR(-ENOENT);
483 }
484 ei->sysctl = head;
485 ei->sysctl_entry = table;
486 hlist_add_head_rcu(&ei->sysctl_inodes, &head->inodes);
487 head->count++;
488 spin_unlock(&sysctl_lock);
489
490 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
491 inode->i_mode = table->mode;
492 if (!S_ISDIR(table->mode)) {
493 inode->i_mode |= S_IFREG;
494 inode->i_op = &proc_sys_inode_operations;
495 inode->i_fop = &proc_sys_file_operations;
496 } else {
497 inode->i_mode |= S_IFDIR;
498 inode->i_op = &proc_sys_dir_operations;
499 inode->i_fop = &proc_sys_dir_file_operations;
500 if (is_empty_dir(head))
501 make_empty_dir_inode(inode);
502 }
503
504 if (root->set_ownership)
505 root->set_ownership(head, table, &inode->i_uid, &inode->i_gid);
506 else {
507 inode->i_uid = GLOBAL_ROOT_UID;
508 inode->i_gid = GLOBAL_ROOT_GID;
509 }
510
511 return inode;
512}
513
514void proc_sys_evict_inode(struct inode *inode, struct ctl_table_header *head)
515{
516 spin_lock(&sysctl_lock);
517 hlist_del_init_rcu(&PROC_I(inode)->sysctl_inodes);
518 if (!--head->count)
519 kfree_rcu(head, rcu);
520 spin_unlock(&sysctl_lock);
521}
522
523static struct ctl_table_header *grab_header(struct inode *inode)
524{
525 struct ctl_table_header *head = PROC_I(inode)->sysctl;
526 if (!head)
527 head = &sysctl_table_root.default_set.dir.header;
528 return sysctl_head_grab(head);
529}
530
531static struct dentry *proc_sys_lookup(struct inode *dir, struct dentry *dentry,
532 unsigned int flags)
533{
534 struct ctl_table_header *head = grab_header(dir);
535 struct ctl_table_header *h = NULL;
536 const struct qstr *name = &dentry->d_name;
537 struct ctl_table *p;
538 struct inode *inode;
539 struct dentry *err = ERR_PTR(-ENOENT);
540 struct ctl_dir *ctl_dir;
541 int ret;
542
543 if (IS_ERR(head))
544 return ERR_CAST(head);
545
546 ctl_dir = container_of(head, struct ctl_dir, header);
547
548 p = lookup_entry(&h, ctl_dir, name->name, name->len);
549 if (!p)
550 goto out;
551
552 if (S_ISLNK(p->mode)) {
553 ret = sysctl_follow_link(&h, &p);
554 err = ERR_PTR(ret);
555 if (ret)
556 goto out;
557 }
558
559 inode = proc_sys_make_inode(dir->i_sb, h ? h : head, p);
560 if (IS_ERR(inode)) {
561 err = ERR_CAST(inode);
562 goto out;
563 }
564
565 d_set_d_op(dentry, &proc_sys_dentry_operations);
566 err = d_splice_alias(inode, dentry);
567
568out:
569 if (h)
570 sysctl_head_finish(h);
571 sysctl_head_finish(head);
572 return err;
573}
574
575static ssize_t proc_sys_call_handler(struct file *filp, void __user *buf,
576 size_t count, loff_t *ppos, int write)
577{
578 struct inode *inode = file_inode(filp);
579 struct ctl_table_header *head = grab_header(inode);
580 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
581 void *new_buf = NULL;
582 ssize_t error;
583
584 if (IS_ERR(head))
585 return PTR_ERR(head);
586
587 /*
588 * At this point we know that the sysctl was not unregistered
589 * and won't be until we finish.
590 */
591 error = -EPERM;
592 if (sysctl_perm(head, table, write ? MAY_WRITE : MAY_READ))
593 goto out;
594
595 /* if that can happen at all, it should be -EINVAL, not -EISDIR */
596 error = -EINVAL;
597 if (!table->proc_handler)
598 goto out;
599
600 error = BPF_CGROUP_RUN_PROG_SYSCTL(head, table, write, buf, &count,
601 ppos, &new_buf);
602 if (error)
603 goto out;
604
605 /* careful: calling conventions are nasty here */
606 if (new_buf) {
607 mm_segment_t old_fs;
608
609 old_fs = get_fs();
610 set_fs(KERNEL_DS);
611 error = table->proc_handler(table, write, (void __user *)new_buf,
612 &count, ppos);
613 set_fs(old_fs);
614 kfree(new_buf);
615 } else {
616 error = table->proc_handler(table, write, buf, &count, ppos);
617 }
618
619 if (!error)
620 error = count;
621out:
622 sysctl_head_finish(head);
623
624 return error;
625}
626
627static ssize_t proc_sys_read(struct file *filp, char __user *buf,
628 size_t count, loff_t *ppos)
629{
630 return proc_sys_call_handler(filp, (void __user *)buf, count, ppos, 0);
631}
632
633static ssize_t proc_sys_write(struct file *filp, const char __user *buf,
634 size_t count, loff_t *ppos)
635{
636 return proc_sys_call_handler(filp, (void __user *)buf, count, ppos, 1);
637}
638
639static int proc_sys_open(struct inode *inode, struct file *filp)
640{
641 struct ctl_table_header *head = grab_header(inode);
642 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
643
644 /* sysctl was unregistered */
645 if (IS_ERR(head))
646 return PTR_ERR(head);
647
648 if (table->poll)
649 filp->private_data = proc_sys_poll_event(table->poll);
650
651 sysctl_head_finish(head);
652
653 return 0;
654}
655
656static __poll_t proc_sys_poll(struct file *filp, poll_table *wait)
657{
658 struct inode *inode = file_inode(filp);
659 struct ctl_table_header *head = grab_header(inode);
660 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
661 __poll_t ret = DEFAULT_POLLMASK;
662 unsigned long event;
663
664 /* sysctl was unregistered */
665 if (IS_ERR(head))
666 return EPOLLERR | EPOLLHUP;
667
668 if (!table->proc_handler)
669 goto out;
670
671 if (!table->poll)
672 goto out;
673
674 event = (unsigned long)filp->private_data;
675 poll_wait(filp, &table->poll->wait, wait);
676
677 if (event != atomic_read(&table->poll->event)) {
678 filp->private_data = proc_sys_poll_event(table->poll);
679 ret = EPOLLIN | EPOLLRDNORM | EPOLLERR | EPOLLPRI;
680 }
681
682out:
683 sysctl_head_finish(head);
684
685 return ret;
686}
687
688static bool proc_sys_fill_cache(struct file *file,
689 struct dir_context *ctx,
690 struct ctl_table_header *head,
691 struct ctl_table *table)
692{
693 struct dentry *child, *dir = file->f_path.dentry;
694 struct inode *inode;
695 struct qstr qname;
696 ino_t ino = 0;
697 unsigned type = DT_UNKNOWN;
698
699 qname.name = table->procname;
700 qname.len = strlen(table->procname);
701 qname.hash = full_name_hash(dir, qname.name, qname.len);
702
703 child = d_lookup(dir, &qname);
704 if (!child) {
705 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
706 child = d_alloc_parallel(dir, &qname, &wq);
707 if (IS_ERR(child))
708 return false;
709 if (d_in_lookup(child)) {
710 struct dentry *res;
711 inode = proc_sys_make_inode(dir->d_sb, head, table);
712 if (IS_ERR(inode)) {
713 d_lookup_done(child);
714 dput(child);
715 return false;
716 }
717 d_set_d_op(child, &proc_sys_dentry_operations);
718 res = d_splice_alias(inode, child);
719 d_lookup_done(child);
720 if (unlikely(res)) {
721 if (IS_ERR(res)) {
722 dput(child);
723 return false;
724 }
725 dput(child);
726 child = res;
727 }
728 }
729 }
730 inode = d_inode(child);
731 ino = inode->i_ino;
732 type = inode->i_mode >> 12;
733 dput(child);
734 return dir_emit(ctx, qname.name, qname.len, ino, type);
735}
736
737static bool proc_sys_link_fill_cache(struct file *file,
738 struct dir_context *ctx,
739 struct ctl_table_header *head,
740 struct ctl_table *table)
741{
742 bool ret = true;
743
744 head = sysctl_head_grab(head);
745 if (IS_ERR(head))
746 return false;
747
748 /* It is not an error if we can not follow the link ignore it */
749 if (sysctl_follow_link(&head, &table))
750 goto out;
751
752 ret = proc_sys_fill_cache(file, ctx, head, table);
753out:
754 sysctl_head_finish(head);
755 return ret;
756}
757
758static int scan(struct ctl_table_header *head, struct ctl_table *table,
759 unsigned long *pos, struct file *file,
760 struct dir_context *ctx)
761{
762 bool res;
763
764 if ((*pos)++ < ctx->pos)
765 return true;
766
767 if (unlikely(S_ISLNK(table->mode)))
768 res = proc_sys_link_fill_cache(file, ctx, head, table);
769 else
770 res = proc_sys_fill_cache(file, ctx, head, table);
771
772 if (res)
773 ctx->pos = *pos;
774
775 return res;
776}
777
778static int proc_sys_readdir(struct file *file, struct dir_context *ctx)
779{
780 struct ctl_table_header *head = grab_header(file_inode(file));
781 struct ctl_table_header *h = NULL;
782 struct ctl_table *entry;
783 struct ctl_dir *ctl_dir;
784 unsigned long pos;
785
786 if (IS_ERR(head))
787 return PTR_ERR(head);
788
789 ctl_dir = container_of(head, struct ctl_dir, header);
790
791 if (!dir_emit_dots(file, ctx))
792 goto out;
793
794 pos = 2;
795
796 for (first_entry(ctl_dir, &h, &entry); h; next_entry(&h, &entry)) {
797 if (!scan(h, entry, &pos, file, ctx)) {
798 sysctl_head_finish(h);
799 break;
800 }
801 }
802out:
803 sysctl_head_finish(head);
804 return 0;
805}
806
807static int proc_sys_permission(struct inode *inode, int mask)
808{
809 /*
810 * sysctl entries that are not writeable,
811 * are _NOT_ writeable, capabilities or not.
812 */
813 struct ctl_table_header *head;
814 struct ctl_table *table;
815 int error;
816
817 /* Executable files are not allowed under /proc/sys/ */
818 if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode))
819 return -EACCES;
820
821 head = grab_header(inode);
822 if (IS_ERR(head))
823 return PTR_ERR(head);
824
825 table = PROC_I(inode)->sysctl_entry;
826 if (!table) /* global root - r-xr-xr-x */
827 error = mask & MAY_WRITE ? -EACCES : 0;
828 else /* Use the permissions on the sysctl table entry */
829 error = sysctl_perm(head, table, mask & ~MAY_NOT_BLOCK);
830
831 sysctl_head_finish(head);
832 return error;
833}
834
835static int proc_sys_setattr(struct dentry *dentry, struct iattr *attr)
836{
837 struct inode *inode = d_inode(dentry);
838 int error;
839
840 if (attr->ia_valid & (ATTR_MODE | ATTR_UID | ATTR_GID))
841 return -EPERM;
842
843 error = setattr_prepare(dentry, attr);
844 if (error)
845 return error;
846
847 setattr_copy(inode, attr);
848 mark_inode_dirty(inode);
849 return 0;
850}
851
852static int proc_sys_getattr(const struct path *path, struct kstat *stat,
853 u32 request_mask, unsigned int query_flags)
854{
855 struct inode *inode = d_inode(path->dentry);
856 struct ctl_table_header *head = grab_header(inode);
857 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
858
859 if (IS_ERR(head))
860 return PTR_ERR(head);
861
862 generic_fillattr(inode, stat);
863 if (table)
864 stat->mode = (stat->mode & S_IFMT) | table->mode;
865
866 sysctl_head_finish(head);
867 return 0;
868}
869
870static const struct file_operations proc_sys_file_operations = {
871 .open = proc_sys_open,
872 .poll = proc_sys_poll,
873 .read = proc_sys_read,
874 .write = proc_sys_write,
875 .llseek = default_llseek,
876};
877
878static const struct file_operations proc_sys_dir_file_operations = {
879 .read = generic_read_dir,
880 .iterate_shared = proc_sys_readdir,
881 .llseek = generic_file_llseek,
882};
883
884static const struct inode_operations proc_sys_inode_operations = {
885 .permission = proc_sys_permission,
886 .setattr = proc_sys_setattr,
887 .getattr = proc_sys_getattr,
888};
889
890static const struct inode_operations proc_sys_dir_operations = {
891 .lookup = proc_sys_lookup,
892 .permission = proc_sys_permission,
893 .setattr = proc_sys_setattr,
894 .getattr = proc_sys_getattr,
895};
896
897static int proc_sys_revalidate(struct dentry *dentry, unsigned int flags)
898{
899 if (flags & LOOKUP_RCU)
900 return -ECHILD;
901 return !PROC_I(d_inode(dentry))->sysctl->unregistering;
902}
903
904static int proc_sys_delete(const struct dentry *dentry)
905{
906 return !!PROC_I(d_inode(dentry))->sysctl->unregistering;
907}
908
909static int sysctl_is_seen(struct ctl_table_header *p)
910{
911 struct ctl_table_set *set = p->set;
912 int res;
913 spin_lock(&sysctl_lock);
914 if (p->unregistering)
915 res = 0;
916 else if (!set->is_seen)
917 res = 1;
918 else
919 res = set->is_seen(set);
920 spin_unlock(&sysctl_lock);
921 return res;
922}
923
924static int proc_sys_compare(const struct dentry *dentry,
925 unsigned int len, const char *str, const struct qstr *name)
926{
927 struct ctl_table_header *head;
928 struct inode *inode;
929
930 /* Although proc doesn't have negative dentries, rcu-walk means
931 * that inode here can be NULL */
932 /* AV: can it, indeed? */
933 inode = d_inode_rcu(dentry);
934 if (!inode)
935 return 1;
936 if (name->len != len)
937 return 1;
938 if (memcmp(name->name, str, len))
939 return 1;
940 head = rcu_dereference(PROC_I(inode)->sysctl);
941 return !head || !sysctl_is_seen(head);
942}
943
944static const struct dentry_operations proc_sys_dentry_operations = {
945 .d_revalidate = proc_sys_revalidate,
946 .d_delete = proc_sys_delete,
947 .d_compare = proc_sys_compare,
948};
949
950static struct ctl_dir *find_subdir(struct ctl_dir *dir,
951 const char *name, int namelen)
952{
953 struct ctl_table_header *head;
954 struct ctl_table *entry;
955
956 entry = find_entry(&head, dir, name, namelen);
957 if (!entry)
958 return ERR_PTR(-ENOENT);
959 if (!S_ISDIR(entry->mode))
960 return ERR_PTR(-ENOTDIR);
961 return container_of(head, struct ctl_dir, header);
962}
963
964static struct ctl_dir *new_dir(struct ctl_table_set *set,
965 const char *name, int namelen)
966{
967 struct ctl_table *table;
968 struct ctl_dir *new;
969 struct ctl_node *node;
970 char *new_name;
971
972 new = kzalloc(sizeof(*new) + sizeof(struct ctl_node) +
973 sizeof(struct ctl_table)*2 + namelen + 1,
974 GFP_KERNEL);
975 if (!new)
976 return NULL;
977
978 node = (struct ctl_node *)(new + 1);
979 table = (struct ctl_table *)(node + 1);
980 new_name = (char *)(table + 2);
981 memcpy(new_name, name, namelen);
982 new_name[namelen] = '\0';
983 table[0].procname = new_name;
984 table[0].mode = S_IFDIR|S_IRUGO|S_IXUGO;
985 init_header(&new->header, set->dir.header.root, set, node, table);
986
987 return new;
988}
989
990/**
991 * get_subdir - find or create a subdir with the specified name.
992 * @dir: Directory to create the subdirectory in
993 * @name: The name of the subdirectory to find or create
994 * @namelen: The length of name
995 *
996 * Takes a directory with an elevated reference count so we know that
997 * if we drop the lock the directory will not go away. Upon success
998 * the reference is moved from @dir to the returned subdirectory.
999 * Upon error an error code is returned and the reference on @dir is
1000 * simply dropped.
1001 */
1002static struct ctl_dir *get_subdir(struct ctl_dir *dir,
1003 const char *name, int namelen)
1004{
1005 struct ctl_table_set *set = dir->header.set;
1006 struct ctl_dir *subdir, *new = NULL;
1007 int err;
1008
1009 spin_lock(&sysctl_lock);
1010 subdir = find_subdir(dir, name, namelen);
1011 if (!IS_ERR(subdir))
1012 goto found;
1013 if (PTR_ERR(subdir) != -ENOENT)
1014 goto failed;
1015
1016 spin_unlock(&sysctl_lock);
1017 new = new_dir(set, name, namelen);
1018 spin_lock(&sysctl_lock);
1019 subdir = ERR_PTR(-ENOMEM);
1020 if (!new)
1021 goto failed;
1022
1023 /* Was the subdir added while we dropped the lock? */
1024 subdir = find_subdir(dir, name, namelen);
1025 if (!IS_ERR(subdir))
1026 goto found;
1027 if (PTR_ERR(subdir) != -ENOENT)
1028 goto failed;
1029
1030 /* Nope. Use the our freshly made directory entry. */
1031 err = insert_header(dir, &new->header);
1032 subdir = ERR_PTR(err);
1033 if (err)
1034 goto failed;
1035 subdir = new;
1036found:
1037 subdir->header.nreg++;
1038failed:
1039 if (IS_ERR(subdir)) {
1040 pr_err("sysctl could not get directory: ");
1041 sysctl_print_dir(dir);
1042 pr_cont("/%*.*s %ld\n",
1043 namelen, namelen, name, PTR_ERR(subdir));
1044 }
1045 drop_sysctl_table(&dir->header);
1046 if (new)
1047 drop_sysctl_table(&new->header);
1048 spin_unlock(&sysctl_lock);
1049 return subdir;
1050}
1051
1052static struct ctl_dir *xlate_dir(struct ctl_table_set *set, struct ctl_dir *dir)
1053{
1054 struct ctl_dir *parent;
1055 const char *procname;
1056 if (!dir->header.parent)
1057 return &set->dir;
1058 parent = xlate_dir(set, dir->header.parent);
1059 if (IS_ERR(parent))
1060 return parent;
1061 procname = dir->header.ctl_table[0].procname;
1062 return find_subdir(parent, procname, strlen(procname));
1063}
1064
1065static int sysctl_follow_link(struct ctl_table_header **phead,
1066 struct ctl_table **pentry)
1067{
1068 struct ctl_table_header *head;
1069 struct ctl_table_root *root;
1070 struct ctl_table_set *set;
1071 struct ctl_table *entry;
1072 struct ctl_dir *dir;
1073 int ret;
1074
1075 ret = 0;
1076 spin_lock(&sysctl_lock);
1077 root = (*pentry)->data;
1078 set = lookup_header_set(root);
1079 dir = xlate_dir(set, (*phead)->parent);
1080 if (IS_ERR(dir))
1081 ret = PTR_ERR(dir);
1082 else {
1083 const char *procname = (*pentry)->procname;
1084 head = NULL;
1085 entry = find_entry(&head, dir, procname, strlen(procname));
1086 ret = -ENOENT;
1087 if (entry && use_table(head)) {
1088 unuse_table(*phead);
1089 *phead = head;
1090 *pentry = entry;
1091 ret = 0;
1092 }
1093 }
1094
1095 spin_unlock(&sysctl_lock);
1096 return ret;
1097}
1098
1099static int sysctl_err(const char *path, struct ctl_table *table, char *fmt, ...)
1100{
1101 struct va_format vaf;
1102 va_list args;
1103
1104 va_start(args, fmt);
1105 vaf.fmt = fmt;
1106 vaf.va = &args;
1107
1108 pr_err("sysctl table check failed: %s/%s %pV\n",
1109 path, table->procname, &vaf);
1110
1111 va_end(args);
1112 return -EINVAL;
1113}
1114
1115static int sysctl_check_table_array(const char *path, struct ctl_table *table)
1116{
1117 int err = 0;
1118
1119 if ((table->proc_handler == proc_douintvec) ||
1120 (table->proc_handler == proc_douintvec_minmax)) {
1121 if (table->maxlen != sizeof(unsigned int))
1122 err |= sysctl_err(path, table, "array not allowed");
1123 }
1124
1125 return err;
1126}
1127
1128static int sysctl_check_table(const char *path, struct ctl_table *table)
1129{
1130 int err = 0;
1131 for (; table->procname; table++) {
1132 if (table->child)
1133 err |= sysctl_err(path, table, "Not a file");
1134
1135 if ((table->proc_handler == proc_dostring) ||
1136 (table->proc_handler == proc_dointvec) ||
1137 (table->proc_handler == proc_douintvec) ||
1138 (table->proc_handler == proc_douintvec_minmax) ||
1139 (table->proc_handler == proc_dointvec_minmax) ||
1140 (table->proc_handler == proc_dointvec_jiffies) ||
1141 (table->proc_handler == proc_dointvec_userhz_jiffies) ||
1142 (table->proc_handler == proc_dointvec_ms_jiffies) ||
1143 (table->proc_handler == proc_doulongvec_minmax) ||
1144 (table->proc_handler == proc_doulongvec_ms_jiffies_minmax)) {
1145 if (!table->data)
1146 err |= sysctl_err(path, table, "No data");
1147 if (!table->maxlen)
1148 err |= sysctl_err(path, table, "No maxlen");
1149 else
1150 err |= sysctl_check_table_array(path, table);
1151 }
1152 if (!table->proc_handler)
1153 err |= sysctl_err(path, table, "No proc_handler");
1154
1155 if ((table->mode & (S_IRUGO|S_IWUGO)) != table->mode)
1156 err |= sysctl_err(path, table, "bogus .mode 0%o",
1157 table->mode);
1158 }
1159 return err;
1160}
1161
1162static struct ctl_table_header *new_links(struct ctl_dir *dir, struct ctl_table *table,
1163 struct ctl_table_root *link_root)
1164{
1165 struct ctl_table *link_table, *entry, *link;
1166 struct ctl_table_header *links;
1167 struct ctl_node *node;
1168 char *link_name;
1169 int nr_entries, name_bytes;
1170
1171 name_bytes = 0;
1172 nr_entries = 0;
1173 for (entry = table; entry->procname; entry++) {
1174 nr_entries++;
1175 name_bytes += strlen(entry->procname) + 1;
1176 }
1177
1178 links = kzalloc(sizeof(struct ctl_table_header) +
1179 sizeof(struct ctl_node)*nr_entries +
1180 sizeof(struct ctl_table)*(nr_entries + 1) +
1181 name_bytes,
1182 GFP_KERNEL);
1183
1184 if (!links)
1185 return NULL;
1186
1187 node = (struct ctl_node *)(links + 1);
1188 link_table = (struct ctl_table *)(node + nr_entries);
1189 link_name = (char *)&link_table[nr_entries + 1];
1190
1191 for (link = link_table, entry = table; entry->procname; link++, entry++) {
1192 int len = strlen(entry->procname) + 1;
1193 memcpy(link_name, entry->procname, len);
1194 link->procname = link_name;
1195 link->mode = S_IFLNK|S_IRWXUGO;
1196 link->data = link_root;
1197 link_name += len;
1198 }
1199 init_header(links, dir->header.root, dir->header.set, node, link_table);
1200 links->nreg = nr_entries;
1201
1202 return links;
1203}
1204
1205static bool get_links(struct ctl_dir *dir,
1206 struct ctl_table *table, struct ctl_table_root *link_root)
1207{
1208 struct ctl_table_header *head;
1209 struct ctl_table *entry, *link;
1210
1211 /* Are there links available for every entry in table? */
1212 for (entry = table; entry->procname; entry++) {
1213 const char *procname = entry->procname;
1214 link = find_entry(&head, dir, procname, strlen(procname));
1215 if (!link)
1216 return false;
1217 if (S_ISDIR(link->mode) && S_ISDIR(entry->mode))
1218 continue;
1219 if (S_ISLNK(link->mode) && (link->data == link_root))
1220 continue;
1221 return false;
1222 }
1223
1224 /* The checks passed. Increase the registration count on the links */
1225 for (entry = table; entry->procname; entry++) {
1226 const char *procname = entry->procname;
1227 link = find_entry(&head, dir, procname, strlen(procname));
1228 head->nreg++;
1229 }
1230 return true;
1231}
1232
1233static int insert_links(struct ctl_table_header *head)
1234{
1235 struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1236 struct ctl_dir *core_parent = NULL;
1237 struct ctl_table_header *links;
1238 int err;
1239
1240 if (head->set == root_set)
1241 return 0;
1242
1243 core_parent = xlate_dir(root_set, head->parent);
1244 if (IS_ERR(core_parent))
1245 return 0;
1246
1247 if (get_links(core_parent, head->ctl_table, head->root))
1248 return 0;
1249
1250 core_parent->header.nreg++;
1251 spin_unlock(&sysctl_lock);
1252
1253 links = new_links(core_parent, head->ctl_table, head->root);
1254
1255 spin_lock(&sysctl_lock);
1256 err = -ENOMEM;
1257 if (!links)
1258 goto out;
1259
1260 err = 0;
1261 if (get_links(core_parent, head->ctl_table, head->root)) {
1262 kfree(links);
1263 goto out;
1264 }
1265
1266 err = insert_header(core_parent, links);
1267 if (err)
1268 kfree(links);
1269out:
1270 drop_sysctl_table(&core_parent->header);
1271 return err;
1272}
1273
1274/**
1275 * __register_sysctl_table - register a leaf sysctl table
1276 * @set: Sysctl tree to register on
1277 * @path: The path to the directory the sysctl table is in.
1278 * @table: the top-level table structure
1279 *
1280 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1281 * array. A completely 0 filled entry terminates the table.
1282 *
1283 * The members of the &struct ctl_table structure are used as follows:
1284 *
1285 * procname - the name of the sysctl file under /proc/sys. Set to %NULL to not
1286 * enter a sysctl file
1287 *
1288 * data - a pointer to data for use by proc_handler
1289 *
1290 * maxlen - the maximum size in bytes of the data
1291 *
1292 * mode - the file permissions for the /proc/sys file
1293 *
1294 * child - must be %NULL.
1295 *
1296 * proc_handler - the text handler routine (described below)
1297 *
1298 * extra1, extra2 - extra pointers usable by the proc handler routines
1299 *
1300 * Leaf nodes in the sysctl tree will be represented by a single file
1301 * under /proc; non-leaf nodes will be represented by directories.
1302 *
1303 * There must be a proc_handler routine for any terminal nodes.
1304 * Several default handlers are available to cover common cases -
1305 *
1306 * proc_dostring(), proc_dointvec(), proc_dointvec_jiffies(),
1307 * proc_dointvec_userhz_jiffies(), proc_dointvec_minmax(),
1308 * proc_doulongvec_ms_jiffies_minmax(), proc_doulongvec_minmax()
1309 *
1310 * It is the handler's job to read the input buffer from user memory
1311 * and process it. The handler should return 0 on success.
1312 *
1313 * This routine returns %NULL on a failure to register, and a pointer
1314 * to the table header on success.
1315 */
1316struct ctl_table_header *__register_sysctl_table(
1317 struct ctl_table_set *set,
1318 const char *path, struct ctl_table *table)
1319{
1320 struct ctl_table_root *root = set->dir.header.root;
1321 struct ctl_table_header *header;
1322 const char *name, *nextname;
1323 struct ctl_dir *dir;
1324 struct ctl_table *entry;
1325 struct ctl_node *node;
1326 int nr_entries = 0;
1327
1328 for (entry = table; entry->procname; entry++)
1329 nr_entries++;
1330
1331 header = kzalloc(sizeof(struct ctl_table_header) +
1332 sizeof(struct ctl_node)*nr_entries, GFP_KERNEL);
1333 if (!header)
1334 return NULL;
1335
1336 node = (struct ctl_node *)(header + 1);
1337 init_header(header, root, set, node, table);
1338 if (sysctl_check_table(path, table))
1339 goto fail;
1340
1341 spin_lock(&sysctl_lock);
1342 dir = &set->dir;
1343 /* Reference moved down the diretory tree get_subdir */
1344 dir->header.nreg++;
1345 spin_unlock(&sysctl_lock);
1346
1347 /* Find the directory for the ctl_table */
1348 for (name = path; name; name = nextname) {
1349 int namelen;
1350 nextname = strchr(name, '/');
1351 if (nextname) {
1352 namelen = nextname - name;
1353 nextname++;
1354 } else {
1355 namelen = strlen(name);
1356 }
1357 if (namelen == 0)
1358 continue;
1359
1360 dir = get_subdir(dir, name, namelen);
1361 if (IS_ERR(dir))
1362 goto fail;
1363 }
1364
1365 spin_lock(&sysctl_lock);
1366 if (insert_header(dir, header))
1367 goto fail_put_dir_locked;
1368
1369 drop_sysctl_table(&dir->header);
1370 spin_unlock(&sysctl_lock);
1371
1372 return header;
1373
1374fail_put_dir_locked:
1375 drop_sysctl_table(&dir->header);
1376 spin_unlock(&sysctl_lock);
1377fail:
1378 kfree(header);
1379 dump_stack();
1380 return NULL;
1381}
1382
1383/**
1384 * register_sysctl - register a sysctl table
1385 * @path: The path to the directory the sysctl table is in.
1386 * @table: the table structure
1387 *
1388 * Register a sysctl table. @table should be a filled in ctl_table
1389 * array. A completely 0 filled entry terminates the table.
1390 *
1391 * See __register_sysctl_table for more details.
1392 */
1393struct ctl_table_header *register_sysctl(const char *path, struct ctl_table *table)
1394{
1395 return __register_sysctl_table(&sysctl_table_root.default_set,
1396 path, table);
1397}
1398EXPORT_SYMBOL(register_sysctl);
1399
1400static char *append_path(const char *path, char *pos, const char *name)
1401{
1402 int namelen;
1403 namelen = strlen(name);
1404 if (((pos - path) + namelen + 2) >= PATH_MAX)
1405 return NULL;
1406 memcpy(pos, name, namelen);
1407 pos[namelen] = '/';
1408 pos[namelen + 1] = '\0';
1409 pos += namelen + 1;
1410 return pos;
1411}
1412
1413static int count_subheaders(struct ctl_table *table)
1414{
1415 int has_files = 0;
1416 int nr_subheaders = 0;
1417 struct ctl_table *entry;
1418
1419 /* special case: no directory and empty directory */
1420 if (!table || !table->procname)
1421 return 1;
1422
1423 for (entry = table; entry->procname; entry++) {
1424 if (entry->child)
1425 nr_subheaders += count_subheaders(entry->child);
1426 else
1427 has_files = 1;
1428 }
1429 return nr_subheaders + has_files;
1430}
1431
1432static int register_leaf_sysctl_tables(const char *path, char *pos,
1433 struct ctl_table_header ***subheader, struct ctl_table_set *set,
1434 struct ctl_table *table)
1435{
1436 struct ctl_table *ctl_table_arg = NULL;
1437 struct ctl_table *entry, *files;
1438 int nr_files = 0;
1439 int nr_dirs = 0;
1440 int err = -ENOMEM;
1441
1442 for (entry = table; entry->procname; entry++) {
1443 if (entry->child)
1444 nr_dirs++;
1445 else
1446 nr_files++;
1447 }
1448
1449 files = table;
1450 /* If there are mixed files and directories we need a new table */
1451 if (nr_dirs && nr_files) {
1452 struct ctl_table *new;
1453 files = kcalloc(nr_files + 1, sizeof(struct ctl_table),
1454 GFP_KERNEL);
1455 if (!files)
1456 goto out;
1457
1458 ctl_table_arg = files;
1459 for (new = files, entry = table; entry->procname; entry++) {
1460 if (entry->child)
1461 continue;
1462 *new = *entry;
1463 new++;
1464 }
1465 }
1466
1467 /* Register everything except a directory full of subdirectories */
1468 if (nr_files || !nr_dirs) {
1469 struct ctl_table_header *header;
1470 header = __register_sysctl_table(set, path, files);
1471 if (!header) {
1472 kfree(ctl_table_arg);
1473 goto out;
1474 }
1475
1476 /* Remember if we need to free the file table */
1477 header->ctl_table_arg = ctl_table_arg;
1478 **subheader = header;
1479 (*subheader)++;
1480 }
1481
1482 /* Recurse into the subdirectories. */
1483 for (entry = table; entry->procname; entry++) {
1484 char *child_pos;
1485
1486 if (!entry->child)
1487 continue;
1488
1489 err = -ENAMETOOLONG;
1490 child_pos = append_path(path, pos, entry->procname);
1491 if (!child_pos)
1492 goto out;
1493
1494 err = register_leaf_sysctl_tables(path, child_pos, subheader,
1495 set, entry->child);
1496 pos[0] = '\0';
1497 if (err)
1498 goto out;
1499 }
1500 err = 0;
1501out:
1502 /* On failure our caller will unregister all registered subheaders */
1503 return err;
1504}
1505
1506/**
1507 * __register_sysctl_paths - register a sysctl table hierarchy
1508 * @set: Sysctl tree to register on
1509 * @path: The path to the directory the sysctl table is in.
1510 * @table: the top-level table structure
1511 *
1512 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1513 * array. A completely 0 filled entry terminates the table.
1514 *
1515 * See __register_sysctl_table for more details.
1516 */
1517struct ctl_table_header *__register_sysctl_paths(
1518 struct ctl_table_set *set,
1519 const struct ctl_path *path, struct ctl_table *table)
1520{
1521 struct ctl_table *ctl_table_arg = table;
1522 int nr_subheaders = count_subheaders(table);
1523 struct ctl_table_header *header = NULL, **subheaders, **subheader;
1524 const struct ctl_path *component;
1525 char *new_path, *pos;
1526
1527 pos = new_path = kmalloc(PATH_MAX, GFP_KERNEL);
1528 if (!new_path)
1529 return NULL;
1530
1531 pos[0] = '\0';
1532 for (component = path; component->procname; component++) {
1533 pos = append_path(new_path, pos, component->procname);
1534 if (!pos)
1535 goto out;
1536 }
1537 while (table->procname && table->child && !table[1].procname) {
1538 pos = append_path(new_path, pos, table->procname);
1539 if (!pos)
1540 goto out;
1541 table = table->child;
1542 }
1543 if (nr_subheaders == 1) {
1544 header = __register_sysctl_table(set, new_path, table);
1545 if (header)
1546 header->ctl_table_arg = ctl_table_arg;
1547 } else {
1548 header = kzalloc(sizeof(*header) +
1549 sizeof(*subheaders)*nr_subheaders, GFP_KERNEL);
1550 if (!header)
1551 goto out;
1552
1553 subheaders = (struct ctl_table_header **) (header + 1);
1554 subheader = subheaders;
1555 header->ctl_table_arg = ctl_table_arg;
1556
1557 if (register_leaf_sysctl_tables(new_path, pos, &subheader,
1558 set, table))
1559 goto err_register_leaves;
1560 }
1561
1562out:
1563 kfree(new_path);
1564 return header;
1565
1566err_register_leaves:
1567 while (subheader > subheaders) {
1568 struct ctl_table_header *subh = *(--subheader);
1569 struct ctl_table *table = subh->ctl_table_arg;
1570 unregister_sysctl_table(subh);
1571 kfree(table);
1572 }
1573 kfree(header);
1574 header = NULL;
1575 goto out;
1576}
1577
1578/**
1579 * register_sysctl_table_path - register a sysctl table hierarchy
1580 * @path: The path to the directory the sysctl table is in.
1581 * @table: the top-level table structure
1582 *
1583 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1584 * array. A completely 0 filled entry terminates the table.
1585 *
1586 * See __register_sysctl_paths for more details.
1587 */
1588struct ctl_table_header *register_sysctl_paths(const struct ctl_path *path,
1589 struct ctl_table *table)
1590{
1591 return __register_sysctl_paths(&sysctl_table_root.default_set,
1592 path, table);
1593}
1594EXPORT_SYMBOL(register_sysctl_paths);
1595
1596/**
1597 * register_sysctl_table - register a sysctl table hierarchy
1598 * @table: the top-level table structure
1599 *
1600 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1601 * array. A completely 0 filled entry terminates the table.
1602 *
1603 * See register_sysctl_paths for more details.
1604 */
1605struct ctl_table_header *register_sysctl_table(struct ctl_table *table)
1606{
1607 static const struct ctl_path null_path[] = { {} };
1608
1609 return register_sysctl_paths(null_path, table);
1610}
1611EXPORT_SYMBOL(register_sysctl_table);
1612
1613static void put_links(struct ctl_table_header *header)
1614{
1615 struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1616 struct ctl_table_root *root = header->root;
1617 struct ctl_dir *parent = header->parent;
1618 struct ctl_dir *core_parent;
1619 struct ctl_table *entry;
1620
1621 if (header->set == root_set)
1622 return;
1623
1624 core_parent = xlate_dir(root_set, parent);
1625 if (IS_ERR(core_parent))
1626 return;
1627
1628 for (entry = header->ctl_table; entry->procname; entry++) {
1629 struct ctl_table_header *link_head;
1630 struct ctl_table *link;
1631 const char *name = entry->procname;
1632
1633 link = find_entry(&link_head, core_parent, name, strlen(name));
1634 if (link &&
1635 ((S_ISDIR(link->mode) && S_ISDIR(entry->mode)) ||
1636 (S_ISLNK(link->mode) && (link->data == root)))) {
1637 drop_sysctl_table(link_head);
1638 }
1639 else {
1640 pr_err("sysctl link missing during unregister: ");
1641 sysctl_print_dir(parent);
1642 pr_cont("/%s\n", name);
1643 }
1644 }
1645}
1646
1647static void drop_sysctl_table(struct ctl_table_header *header)
1648{
1649 struct ctl_dir *parent = header->parent;
1650
1651 if (--header->nreg)
1652 return;
1653
1654 if (parent) {
1655 put_links(header);
1656 start_unregistering(header);
1657 }
1658
1659 if (!--header->count)
1660 kfree_rcu(header, rcu);
1661
1662 if (parent)
1663 drop_sysctl_table(&parent->header);
1664}
1665
1666/**
1667 * unregister_sysctl_table - unregister a sysctl table hierarchy
1668 * @header: the header returned from register_sysctl_table
1669 *
1670 * Unregisters the sysctl table and all children. proc entries may not
1671 * actually be removed until they are no longer used by anyone.
1672 */
1673void unregister_sysctl_table(struct ctl_table_header * header)
1674{
1675 int nr_subheaders;
1676 might_sleep();
1677
1678 if (header == NULL)
1679 return;
1680
1681 nr_subheaders = count_subheaders(header->ctl_table_arg);
1682 if (unlikely(nr_subheaders > 1)) {
1683 struct ctl_table_header **subheaders;
1684 int i;
1685
1686 subheaders = (struct ctl_table_header **)(header + 1);
1687 for (i = nr_subheaders -1; i >= 0; i--) {
1688 struct ctl_table_header *subh = subheaders[i];
1689 struct ctl_table *table = subh->ctl_table_arg;
1690 unregister_sysctl_table(subh);
1691 kfree(table);
1692 }
1693 kfree(header);
1694 return;
1695 }
1696
1697 spin_lock(&sysctl_lock);
1698 drop_sysctl_table(header);
1699 spin_unlock(&sysctl_lock);
1700}
1701EXPORT_SYMBOL(unregister_sysctl_table);
1702
1703void setup_sysctl_set(struct ctl_table_set *set,
1704 struct ctl_table_root *root,
1705 int (*is_seen)(struct ctl_table_set *))
1706{
1707 memset(set, 0, sizeof(*set));
1708 set->is_seen = is_seen;
1709 init_header(&set->dir.header, root, set, NULL, root_table);
1710}
1711
1712void retire_sysctl_set(struct ctl_table_set *set)
1713{
1714 WARN_ON(!RB_EMPTY_ROOT(&set->dir.root));
1715}
1716
1717int __init proc_sys_init(void)
1718{
1719 struct proc_dir_entry *proc_sys_root;
1720
1721 proc_sys_root = proc_mkdir("sys", NULL);
1722 proc_sys_root->proc_iops = &proc_sys_dir_operations;
1723 proc_sys_root->proc_fops = &proc_sys_dir_file_operations;
1724 proc_sys_root->nlink = 0;
1725
1726 return sysctl_init();
1727}