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/uio.h>
16#include <linux/module.h>
17#include <linux/bpf-cgroup.h>
18#include <linux/mount.h>
19#include "internal.h"
20
21static const struct dentry_operations proc_sys_dentry_operations;
22static const struct file_operations proc_sys_file_operations;
23static const struct inode_operations proc_sys_inode_operations;
24static const struct file_operations proc_sys_dir_file_operations;
25static const struct inode_operations proc_sys_dir_operations;
26
27/* shared constants to be used in various sysctls */
28const int sysctl_vals[] = { 0, 1, INT_MAX };
29EXPORT_SYMBOL(sysctl_vals);
30
31/* Support for permanently empty directories */
32
33struct ctl_table sysctl_mount_point[] = {
34 { }
35};
36
37static bool is_empty_dir(struct ctl_table_header *head)
38{
39 return head->ctl_table[0].child == sysctl_mount_point;
40}
41
42static void set_empty_dir(struct ctl_dir *dir)
43{
44 dir->header.ctl_table[0].child = sysctl_mount_point;
45}
46
47static void clear_empty_dir(struct ctl_dir *dir)
48
49{
50 dir->header.ctl_table[0].child = NULL;
51}
52
53void proc_sys_poll_notify(struct ctl_table_poll *poll)
54{
55 if (!poll)
56 return;
57
58 atomic_inc(&poll->event);
59 wake_up_interruptible(&poll->wait);
60}
61
62static struct ctl_table root_table[] = {
63 {
64 .procname = "",
65 .mode = S_IFDIR|S_IRUGO|S_IXUGO,
66 },
67 { }
68};
69static struct ctl_table_root sysctl_table_root = {
70 .default_set.dir.header = {
71 {{.count = 1,
72 .nreg = 1,
73 .ctl_table = root_table }},
74 .ctl_table_arg = root_table,
75 .root = &sysctl_table_root,
76 .set = &sysctl_table_root.default_set,
77 },
78};
79
80static DEFINE_SPINLOCK(sysctl_lock);
81
82static void drop_sysctl_table(struct ctl_table_header *header);
83static int sysctl_follow_link(struct ctl_table_header **phead,
84 struct ctl_table **pentry);
85static int insert_links(struct ctl_table_header *head);
86static void put_links(struct ctl_table_header *header);
87
88static void sysctl_print_dir(struct ctl_dir *dir)
89{
90 if (dir->header.parent)
91 sysctl_print_dir(dir->header.parent);
92 pr_cont("%s/", dir->header.ctl_table[0].procname);
93}
94
95static int namecmp(const char *name1, int len1, const char *name2, int len2)
96{
97 int cmp;
98
99 cmp = memcmp(name1, name2, min(len1, len2));
100 if (cmp == 0)
101 cmp = len1 - len2;
102 return cmp;
103}
104
105/* Called under sysctl_lock */
106static struct ctl_table *find_entry(struct ctl_table_header **phead,
107 struct ctl_dir *dir, const char *name, int namelen)
108{
109 struct ctl_table_header *head;
110 struct ctl_table *entry;
111 struct rb_node *node = dir->root.rb_node;
112
113 while (node)
114 {
115 struct ctl_node *ctl_node;
116 const char *procname;
117 int cmp;
118
119 ctl_node = rb_entry(node, struct ctl_node, node);
120 head = ctl_node->header;
121 entry = &head->ctl_table[ctl_node - head->node];
122 procname = entry->procname;
123
124 cmp = namecmp(name, namelen, procname, strlen(procname));
125 if (cmp < 0)
126 node = node->rb_left;
127 else if (cmp > 0)
128 node = node->rb_right;
129 else {
130 *phead = head;
131 return entry;
132 }
133 }
134 return NULL;
135}
136
137static int insert_entry(struct ctl_table_header *head, struct ctl_table *entry)
138{
139 struct rb_node *node = &head->node[entry - head->ctl_table].node;
140 struct rb_node **p = &head->parent->root.rb_node;
141 struct rb_node *parent = NULL;
142 const char *name = entry->procname;
143 int namelen = strlen(name);
144
145 while (*p) {
146 struct ctl_table_header *parent_head;
147 struct ctl_table *parent_entry;
148 struct ctl_node *parent_node;
149 const char *parent_name;
150 int cmp;
151
152 parent = *p;
153 parent_node = rb_entry(parent, struct ctl_node, node);
154 parent_head = parent_node->header;
155 parent_entry = &parent_head->ctl_table[parent_node - parent_head->node];
156 parent_name = parent_entry->procname;
157
158 cmp = namecmp(name, namelen, parent_name, strlen(parent_name));
159 if (cmp < 0)
160 p = &(*p)->rb_left;
161 else if (cmp > 0)
162 p = &(*p)->rb_right;
163 else {
164 pr_err("sysctl duplicate entry: ");
165 sysctl_print_dir(head->parent);
166 pr_cont("/%s\n", entry->procname);
167 return -EEXIST;
168 }
169 }
170
171 rb_link_node(node, parent, p);
172 rb_insert_color(node, &head->parent->root);
173 return 0;
174}
175
176static void erase_entry(struct ctl_table_header *head, struct ctl_table *entry)
177{
178 struct rb_node *node = &head->node[entry - head->ctl_table].node;
179
180 rb_erase(node, &head->parent->root);
181}
182
183static void init_header(struct ctl_table_header *head,
184 struct ctl_table_root *root, struct ctl_table_set *set,
185 struct ctl_node *node, struct ctl_table *table)
186{
187 head->ctl_table = table;
188 head->ctl_table_arg = table;
189 head->used = 0;
190 head->count = 1;
191 head->nreg = 1;
192 head->unregistering = NULL;
193 head->root = root;
194 head->set = set;
195 head->parent = NULL;
196 head->node = node;
197 INIT_HLIST_HEAD(&head->inodes);
198 if (node) {
199 struct ctl_table *entry;
200 for (entry = table; entry->procname; entry++, node++)
201 node->header = head;
202 }
203}
204
205static void erase_header(struct ctl_table_header *head)
206{
207 struct ctl_table *entry;
208 for (entry = head->ctl_table; entry->procname; entry++)
209 erase_entry(head, entry);
210}
211
212static int insert_header(struct ctl_dir *dir, struct ctl_table_header *header)
213{
214 struct ctl_table *entry;
215 int err;
216
217 /* Is this a permanently empty directory? */
218 if (is_empty_dir(&dir->header))
219 return -EROFS;
220
221 /* Am I creating a permanently empty directory? */
222 if (header->ctl_table == sysctl_mount_point) {
223 if (!RB_EMPTY_ROOT(&dir->root))
224 return -EINVAL;
225 set_empty_dir(dir);
226 }
227
228 dir->header.nreg++;
229 header->parent = dir;
230 err = insert_links(header);
231 if (err)
232 goto fail_links;
233 for (entry = header->ctl_table; entry->procname; entry++) {
234 err = insert_entry(header, entry);
235 if (err)
236 goto fail;
237 }
238 return 0;
239fail:
240 erase_header(header);
241 put_links(header);
242fail_links:
243 if (header->ctl_table == sysctl_mount_point)
244 clear_empty_dir(dir);
245 header->parent = NULL;
246 drop_sysctl_table(&dir->header);
247 return err;
248}
249
250/* called under sysctl_lock */
251static int use_table(struct ctl_table_header *p)
252{
253 if (unlikely(p->unregistering))
254 return 0;
255 p->used++;
256 return 1;
257}
258
259/* called under sysctl_lock */
260static void unuse_table(struct ctl_table_header *p)
261{
262 if (!--p->used)
263 if (unlikely(p->unregistering))
264 complete(p->unregistering);
265}
266
267static void proc_sys_invalidate_dcache(struct ctl_table_header *head)
268{
269 proc_invalidate_siblings_dcache(&head->inodes, &sysctl_lock);
270}
271
272/* called under sysctl_lock, will reacquire if has to wait */
273static void start_unregistering(struct ctl_table_header *p)
274{
275 /*
276 * if p->used is 0, nobody will ever touch that entry again;
277 * we'll eliminate all paths to it before dropping sysctl_lock
278 */
279 if (unlikely(p->used)) {
280 struct completion wait;
281 init_completion(&wait);
282 p->unregistering = &wait;
283 spin_unlock(&sysctl_lock);
284 wait_for_completion(&wait);
285 } else {
286 /* anything non-NULL; we'll never dereference it */
287 p->unregistering = ERR_PTR(-EINVAL);
288 spin_unlock(&sysctl_lock);
289 }
290 /*
291 * Invalidate dentries for unregistered sysctls: namespaced sysctls
292 * can have duplicate names and contaminate dcache very badly.
293 */
294 proc_sys_invalidate_dcache(p);
295 /*
296 * do not remove from the list until nobody holds it; walking the
297 * list in do_sysctl() relies on that.
298 */
299 spin_lock(&sysctl_lock);
300 erase_header(p);
301}
302
303static struct ctl_table_header *sysctl_head_grab(struct ctl_table_header *head)
304{
305 BUG_ON(!head);
306 spin_lock(&sysctl_lock);
307 if (!use_table(head))
308 head = ERR_PTR(-ENOENT);
309 spin_unlock(&sysctl_lock);
310 return head;
311}
312
313static void sysctl_head_finish(struct ctl_table_header *head)
314{
315 if (!head)
316 return;
317 spin_lock(&sysctl_lock);
318 unuse_table(head);
319 spin_unlock(&sysctl_lock);
320}
321
322static struct ctl_table_set *
323lookup_header_set(struct ctl_table_root *root)
324{
325 struct ctl_table_set *set = &root->default_set;
326 if (root->lookup)
327 set = root->lookup(root);
328 return set;
329}
330
331static struct ctl_table *lookup_entry(struct ctl_table_header **phead,
332 struct ctl_dir *dir,
333 const char *name, int namelen)
334{
335 struct ctl_table_header *head;
336 struct ctl_table *entry;
337
338 spin_lock(&sysctl_lock);
339 entry = find_entry(&head, dir, name, namelen);
340 if (entry && use_table(head))
341 *phead = head;
342 else
343 entry = NULL;
344 spin_unlock(&sysctl_lock);
345 return entry;
346}
347
348static struct ctl_node *first_usable_entry(struct rb_node *node)
349{
350 struct ctl_node *ctl_node;
351
352 for (;node; node = rb_next(node)) {
353 ctl_node = rb_entry(node, struct ctl_node, node);
354 if (use_table(ctl_node->header))
355 return ctl_node;
356 }
357 return NULL;
358}
359
360static void first_entry(struct ctl_dir *dir,
361 struct ctl_table_header **phead, struct ctl_table **pentry)
362{
363 struct ctl_table_header *head = NULL;
364 struct ctl_table *entry = NULL;
365 struct ctl_node *ctl_node;
366
367 spin_lock(&sysctl_lock);
368 ctl_node = first_usable_entry(rb_first(&dir->root));
369 spin_unlock(&sysctl_lock);
370 if (ctl_node) {
371 head = ctl_node->header;
372 entry = &head->ctl_table[ctl_node - head->node];
373 }
374 *phead = head;
375 *pentry = entry;
376}
377
378static void next_entry(struct ctl_table_header **phead, struct ctl_table **pentry)
379{
380 struct ctl_table_header *head = *phead;
381 struct ctl_table *entry = *pentry;
382 struct ctl_node *ctl_node = &head->node[entry - head->ctl_table];
383
384 spin_lock(&sysctl_lock);
385 unuse_table(head);
386
387 ctl_node = first_usable_entry(rb_next(&ctl_node->node));
388 spin_unlock(&sysctl_lock);
389 head = NULL;
390 if (ctl_node) {
391 head = ctl_node->header;
392 entry = &head->ctl_table[ctl_node - head->node];
393 }
394 *phead = head;
395 *pentry = entry;
396}
397
398/*
399 * sysctl_perm does NOT grant the superuser all rights automatically, because
400 * some sysctl variables are readonly even to root.
401 */
402
403static int test_perm(int mode, int op)
404{
405 if (uid_eq(current_euid(), GLOBAL_ROOT_UID))
406 mode >>= 6;
407 else if (in_egroup_p(GLOBAL_ROOT_GID))
408 mode >>= 3;
409 if ((op & ~mode & (MAY_READ|MAY_WRITE|MAY_EXEC)) == 0)
410 return 0;
411 return -EACCES;
412}
413
414static int sysctl_perm(struct ctl_table_header *head, struct ctl_table *table, int op)
415{
416 struct ctl_table_root *root = head->root;
417 int mode;
418
419 if (root->permissions)
420 mode = root->permissions(head, table);
421 else
422 mode = table->mode;
423
424 return test_perm(mode, op);
425}
426
427static struct inode *proc_sys_make_inode(struct super_block *sb,
428 struct ctl_table_header *head, struct ctl_table *table)
429{
430 struct ctl_table_root *root = head->root;
431 struct inode *inode;
432 struct proc_inode *ei;
433
434 inode = new_inode(sb);
435 if (!inode)
436 return ERR_PTR(-ENOMEM);
437
438 inode->i_ino = get_next_ino();
439
440 ei = PROC_I(inode);
441
442 spin_lock(&sysctl_lock);
443 if (unlikely(head->unregistering)) {
444 spin_unlock(&sysctl_lock);
445 iput(inode);
446 return ERR_PTR(-ENOENT);
447 }
448 ei->sysctl = head;
449 ei->sysctl_entry = table;
450 hlist_add_head_rcu(&ei->sibling_inodes, &head->inodes);
451 head->count++;
452 spin_unlock(&sysctl_lock);
453
454 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
455 inode->i_mode = table->mode;
456 if (!S_ISDIR(table->mode)) {
457 inode->i_mode |= S_IFREG;
458 inode->i_op = &proc_sys_inode_operations;
459 inode->i_fop = &proc_sys_file_operations;
460 } else {
461 inode->i_mode |= S_IFDIR;
462 inode->i_op = &proc_sys_dir_operations;
463 inode->i_fop = &proc_sys_dir_file_operations;
464 if (is_empty_dir(head))
465 make_empty_dir_inode(inode);
466 }
467
468 if (root->set_ownership)
469 root->set_ownership(head, table, &inode->i_uid, &inode->i_gid);
470 else {
471 inode->i_uid = GLOBAL_ROOT_UID;
472 inode->i_gid = GLOBAL_ROOT_GID;
473 }
474
475 return inode;
476}
477
478void proc_sys_evict_inode(struct inode *inode, struct ctl_table_header *head)
479{
480 spin_lock(&sysctl_lock);
481 hlist_del_init_rcu(&PROC_I(inode)->sibling_inodes);
482 if (!--head->count)
483 kfree_rcu(head, rcu);
484 spin_unlock(&sysctl_lock);
485}
486
487static struct ctl_table_header *grab_header(struct inode *inode)
488{
489 struct ctl_table_header *head = PROC_I(inode)->sysctl;
490 if (!head)
491 head = &sysctl_table_root.default_set.dir.header;
492 return sysctl_head_grab(head);
493}
494
495static struct dentry *proc_sys_lookup(struct inode *dir, struct dentry *dentry,
496 unsigned int flags)
497{
498 struct ctl_table_header *head = grab_header(dir);
499 struct ctl_table_header *h = NULL;
500 const struct qstr *name = &dentry->d_name;
501 struct ctl_table *p;
502 struct inode *inode;
503 struct dentry *err = ERR_PTR(-ENOENT);
504 struct ctl_dir *ctl_dir;
505 int ret;
506
507 if (IS_ERR(head))
508 return ERR_CAST(head);
509
510 ctl_dir = container_of(head, struct ctl_dir, header);
511
512 p = lookup_entry(&h, ctl_dir, name->name, name->len);
513 if (!p)
514 goto out;
515
516 if (S_ISLNK(p->mode)) {
517 ret = sysctl_follow_link(&h, &p);
518 err = ERR_PTR(ret);
519 if (ret)
520 goto out;
521 }
522
523 inode = proc_sys_make_inode(dir->i_sb, h ? h : head, p);
524 if (IS_ERR(inode)) {
525 err = ERR_CAST(inode);
526 goto out;
527 }
528
529 d_set_d_op(dentry, &proc_sys_dentry_operations);
530 err = d_splice_alias(inode, dentry);
531
532out:
533 if (h)
534 sysctl_head_finish(h);
535 sysctl_head_finish(head);
536 return err;
537}
538
539static ssize_t proc_sys_call_handler(struct kiocb *iocb, struct iov_iter *iter,
540 int write)
541{
542 struct inode *inode = file_inode(iocb->ki_filp);
543 struct ctl_table_header *head = grab_header(inode);
544 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
545 size_t count = iov_iter_count(iter);
546 char *kbuf;
547 ssize_t error;
548
549 if (IS_ERR(head))
550 return PTR_ERR(head);
551
552 /*
553 * At this point we know that the sysctl was not unregistered
554 * and won't be until we finish.
555 */
556 error = -EPERM;
557 if (sysctl_perm(head, table, write ? MAY_WRITE : MAY_READ))
558 goto out;
559
560 /* if that can happen at all, it should be -EINVAL, not -EISDIR */
561 error = -EINVAL;
562 if (!table->proc_handler)
563 goto out;
564
565 /* don't even try if the size is too large */
566 error = -ENOMEM;
567 if (count >= KMALLOC_MAX_SIZE)
568 goto out;
569 kbuf = kvzalloc(count + 1, GFP_KERNEL);
570 if (!kbuf)
571 goto out;
572
573 if (write) {
574 error = -EFAULT;
575 if (!copy_from_iter_full(kbuf, count, iter))
576 goto out_free_buf;
577 kbuf[count] = '\0';
578 }
579
580 error = BPF_CGROUP_RUN_PROG_SYSCTL(head, table, write, &kbuf, &count,
581 &iocb->ki_pos);
582 if (error)
583 goto out_free_buf;
584
585 /* careful: calling conventions are nasty here */
586 error = table->proc_handler(table, write, kbuf, &count, &iocb->ki_pos);
587 if (error)
588 goto out_free_buf;
589
590 if (!write) {
591 error = -EFAULT;
592 if (copy_to_iter(kbuf, count, iter) < count)
593 goto out_free_buf;
594 }
595
596 error = count;
597out_free_buf:
598 kvfree(kbuf);
599out:
600 sysctl_head_finish(head);
601
602 return error;
603}
604
605static ssize_t proc_sys_read(struct kiocb *iocb, struct iov_iter *iter)
606{
607 return proc_sys_call_handler(iocb, iter, 0);
608}
609
610static ssize_t proc_sys_write(struct kiocb *iocb, struct iov_iter *iter)
611{
612 return proc_sys_call_handler(iocb, iter, 1);
613}
614
615static int proc_sys_open(struct inode *inode, struct file *filp)
616{
617 struct ctl_table_header *head = grab_header(inode);
618 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
619
620 /* sysctl was unregistered */
621 if (IS_ERR(head))
622 return PTR_ERR(head);
623
624 if (table->poll)
625 filp->private_data = proc_sys_poll_event(table->poll);
626
627 sysctl_head_finish(head);
628
629 return 0;
630}
631
632static __poll_t proc_sys_poll(struct file *filp, poll_table *wait)
633{
634 struct inode *inode = file_inode(filp);
635 struct ctl_table_header *head = grab_header(inode);
636 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
637 __poll_t ret = DEFAULT_POLLMASK;
638 unsigned long event;
639
640 /* sysctl was unregistered */
641 if (IS_ERR(head))
642 return EPOLLERR | EPOLLHUP;
643
644 if (!table->proc_handler)
645 goto out;
646
647 if (!table->poll)
648 goto out;
649
650 event = (unsigned long)filp->private_data;
651 poll_wait(filp, &table->poll->wait, wait);
652
653 if (event != atomic_read(&table->poll->event)) {
654 filp->private_data = proc_sys_poll_event(table->poll);
655 ret = EPOLLIN | EPOLLRDNORM | EPOLLERR | EPOLLPRI;
656 }
657
658out:
659 sysctl_head_finish(head);
660
661 return ret;
662}
663
664static bool proc_sys_fill_cache(struct file *file,
665 struct dir_context *ctx,
666 struct ctl_table_header *head,
667 struct ctl_table *table)
668{
669 struct dentry *child, *dir = file->f_path.dentry;
670 struct inode *inode;
671 struct qstr qname;
672 ino_t ino = 0;
673 unsigned type = DT_UNKNOWN;
674
675 qname.name = table->procname;
676 qname.len = strlen(table->procname);
677 qname.hash = full_name_hash(dir, qname.name, qname.len);
678
679 child = d_lookup(dir, &qname);
680 if (!child) {
681 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
682 child = d_alloc_parallel(dir, &qname, &wq);
683 if (IS_ERR(child))
684 return false;
685 if (d_in_lookup(child)) {
686 struct dentry *res;
687 inode = proc_sys_make_inode(dir->d_sb, head, table);
688 if (IS_ERR(inode)) {
689 d_lookup_done(child);
690 dput(child);
691 return false;
692 }
693 d_set_d_op(child, &proc_sys_dentry_operations);
694 res = d_splice_alias(inode, child);
695 d_lookup_done(child);
696 if (unlikely(res)) {
697 if (IS_ERR(res)) {
698 dput(child);
699 return false;
700 }
701 dput(child);
702 child = res;
703 }
704 }
705 }
706 inode = d_inode(child);
707 ino = inode->i_ino;
708 type = inode->i_mode >> 12;
709 dput(child);
710 return dir_emit(ctx, qname.name, qname.len, ino, type);
711}
712
713static bool proc_sys_link_fill_cache(struct file *file,
714 struct dir_context *ctx,
715 struct ctl_table_header *head,
716 struct ctl_table *table)
717{
718 bool ret = true;
719
720 head = sysctl_head_grab(head);
721 if (IS_ERR(head))
722 return false;
723
724 /* It is not an error if we can not follow the link ignore it */
725 if (sysctl_follow_link(&head, &table))
726 goto out;
727
728 ret = proc_sys_fill_cache(file, ctx, head, table);
729out:
730 sysctl_head_finish(head);
731 return ret;
732}
733
734static int scan(struct ctl_table_header *head, struct ctl_table *table,
735 unsigned long *pos, struct file *file,
736 struct dir_context *ctx)
737{
738 bool res;
739
740 if ((*pos)++ < ctx->pos)
741 return true;
742
743 if (unlikely(S_ISLNK(table->mode)))
744 res = proc_sys_link_fill_cache(file, ctx, head, table);
745 else
746 res = proc_sys_fill_cache(file, ctx, head, table);
747
748 if (res)
749 ctx->pos = *pos;
750
751 return res;
752}
753
754static int proc_sys_readdir(struct file *file, struct dir_context *ctx)
755{
756 struct ctl_table_header *head = grab_header(file_inode(file));
757 struct ctl_table_header *h = NULL;
758 struct ctl_table *entry;
759 struct ctl_dir *ctl_dir;
760 unsigned long pos;
761
762 if (IS_ERR(head))
763 return PTR_ERR(head);
764
765 ctl_dir = container_of(head, struct ctl_dir, header);
766
767 if (!dir_emit_dots(file, ctx))
768 goto out;
769
770 pos = 2;
771
772 for (first_entry(ctl_dir, &h, &entry); h; next_entry(&h, &entry)) {
773 if (!scan(h, entry, &pos, file, ctx)) {
774 sysctl_head_finish(h);
775 break;
776 }
777 }
778out:
779 sysctl_head_finish(head);
780 return 0;
781}
782
783static int proc_sys_permission(struct user_namespace *mnt_userns,
784 struct inode *inode, int mask)
785{
786 /*
787 * sysctl entries that are not writeable,
788 * are _NOT_ writeable, capabilities or not.
789 */
790 struct ctl_table_header *head;
791 struct ctl_table *table;
792 int error;
793
794 /* Executable files are not allowed under /proc/sys/ */
795 if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode))
796 return -EACCES;
797
798 head = grab_header(inode);
799 if (IS_ERR(head))
800 return PTR_ERR(head);
801
802 table = PROC_I(inode)->sysctl_entry;
803 if (!table) /* global root - r-xr-xr-x */
804 error = mask & MAY_WRITE ? -EACCES : 0;
805 else /* Use the permissions on the sysctl table entry */
806 error = sysctl_perm(head, table, mask & ~MAY_NOT_BLOCK);
807
808 sysctl_head_finish(head);
809 return error;
810}
811
812static int proc_sys_setattr(struct user_namespace *mnt_userns,
813 struct dentry *dentry, struct iattr *attr)
814{
815 struct inode *inode = d_inode(dentry);
816 int error;
817
818 if (attr->ia_valid & (ATTR_MODE | ATTR_UID | ATTR_GID))
819 return -EPERM;
820
821 error = setattr_prepare(&init_user_ns, dentry, attr);
822 if (error)
823 return error;
824
825 setattr_copy(&init_user_ns, inode, attr);
826 mark_inode_dirty(inode);
827 return 0;
828}
829
830static int proc_sys_getattr(struct user_namespace *mnt_userns,
831 const struct path *path, struct kstat *stat,
832 u32 request_mask, unsigned int query_flags)
833{
834 struct inode *inode = d_inode(path->dentry);
835 struct ctl_table_header *head = grab_header(inode);
836 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
837
838 if (IS_ERR(head))
839 return PTR_ERR(head);
840
841 generic_fillattr(&init_user_ns, inode, stat);
842 if (table)
843 stat->mode = (stat->mode & S_IFMT) | table->mode;
844
845 sysctl_head_finish(head);
846 return 0;
847}
848
849static const struct file_operations proc_sys_file_operations = {
850 .open = proc_sys_open,
851 .poll = proc_sys_poll,
852 .read_iter = proc_sys_read,
853 .write_iter = proc_sys_write,
854 .splice_read = generic_file_splice_read,
855 .splice_write = iter_file_splice_write,
856 .llseek = default_llseek,
857};
858
859static const struct file_operations proc_sys_dir_file_operations = {
860 .read = generic_read_dir,
861 .iterate_shared = proc_sys_readdir,
862 .llseek = generic_file_llseek,
863};
864
865static const struct inode_operations proc_sys_inode_operations = {
866 .permission = proc_sys_permission,
867 .setattr = proc_sys_setattr,
868 .getattr = proc_sys_getattr,
869};
870
871static const struct inode_operations proc_sys_dir_operations = {
872 .lookup = proc_sys_lookup,
873 .permission = proc_sys_permission,
874 .setattr = proc_sys_setattr,
875 .getattr = proc_sys_getattr,
876};
877
878static int proc_sys_revalidate(struct dentry *dentry, unsigned int flags)
879{
880 if (flags & LOOKUP_RCU)
881 return -ECHILD;
882 return !PROC_I(d_inode(dentry))->sysctl->unregistering;
883}
884
885static int proc_sys_delete(const struct dentry *dentry)
886{
887 return !!PROC_I(d_inode(dentry))->sysctl->unregistering;
888}
889
890static int sysctl_is_seen(struct ctl_table_header *p)
891{
892 struct ctl_table_set *set = p->set;
893 int res;
894 spin_lock(&sysctl_lock);
895 if (p->unregistering)
896 res = 0;
897 else if (!set->is_seen)
898 res = 1;
899 else
900 res = set->is_seen(set);
901 spin_unlock(&sysctl_lock);
902 return res;
903}
904
905static int proc_sys_compare(const struct dentry *dentry,
906 unsigned int len, const char *str, const struct qstr *name)
907{
908 struct ctl_table_header *head;
909 struct inode *inode;
910
911 /* Although proc doesn't have negative dentries, rcu-walk means
912 * that inode here can be NULL */
913 /* AV: can it, indeed? */
914 inode = d_inode_rcu(dentry);
915 if (!inode)
916 return 1;
917 if (name->len != len)
918 return 1;
919 if (memcmp(name->name, str, len))
920 return 1;
921 head = rcu_dereference(PROC_I(inode)->sysctl);
922 return !head || !sysctl_is_seen(head);
923}
924
925static const struct dentry_operations proc_sys_dentry_operations = {
926 .d_revalidate = proc_sys_revalidate,
927 .d_delete = proc_sys_delete,
928 .d_compare = proc_sys_compare,
929};
930
931static struct ctl_dir *find_subdir(struct ctl_dir *dir,
932 const char *name, int namelen)
933{
934 struct ctl_table_header *head;
935 struct ctl_table *entry;
936
937 entry = find_entry(&head, dir, name, namelen);
938 if (!entry)
939 return ERR_PTR(-ENOENT);
940 if (!S_ISDIR(entry->mode))
941 return ERR_PTR(-ENOTDIR);
942 return container_of(head, struct ctl_dir, header);
943}
944
945static struct ctl_dir *new_dir(struct ctl_table_set *set,
946 const char *name, int namelen)
947{
948 struct ctl_table *table;
949 struct ctl_dir *new;
950 struct ctl_node *node;
951 char *new_name;
952
953 new = kzalloc(sizeof(*new) + sizeof(struct ctl_node) +
954 sizeof(struct ctl_table)*2 + namelen + 1,
955 GFP_KERNEL);
956 if (!new)
957 return NULL;
958
959 node = (struct ctl_node *)(new + 1);
960 table = (struct ctl_table *)(node + 1);
961 new_name = (char *)(table + 2);
962 memcpy(new_name, name, namelen);
963 new_name[namelen] = '\0';
964 table[0].procname = new_name;
965 table[0].mode = S_IFDIR|S_IRUGO|S_IXUGO;
966 init_header(&new->header, set->dir.header.root, set, node, table);
967
968 return new;
969}
970
971/**
972 * get_subdir - find or create a subdir with the specified name.
973 * @dir: Directory to create the subdirectory in
974 * @name: The name of the subdirectory to find or create
975 * @namelen: The length of name
976 *
977 * Takes a directory with an elevated reference count so we know that
978 * if we drop the lock the directory will not go away. Upon success
979 * the reference is moved from @dir to the returned subdirectory.
980 * Upon error an error code is returned and the reference on @dir is
981 * simply dropped.
982 */
983static struct ctl_dir *get_subdir(struct ctl_dir *dir,
984 const char *name, int namelen)
985{
986 struct ctl_table_set *set = dir->header.set;
987 struct ctl_dir *subdir, *new = NULL;
988 int err;
989
990 spin_lock(&sysctl_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 spin_unlock(&sysctl_lock);
998 new = new_dir(set, name, namelen);
999 spin_lock(&sysctl_lock);
1000 subdir = ERR_PTR(-ENOMEM);
1001 if (!new)
1002 goto failed;
1003
1004 /* Was the subdir added while we dropped the lock? */
1005 subdir = find_subdir(dir, name, namelen);
1006 if (!IS_ERR(subdir))
1007 goto found;
1008 if (PTR_ERR(subdir) != -ENOENT)
1009 goto failed;
1010
1011 /* Nope. Use the our freshly made directory entry. */
1012 err = insert_header(dir, &new->header);
1013 subdir = ERR_PTR(err);
1014 if (err)
1015 goto failed;
1016 subdir = new;
1017found:
1018 subdir->header.nreg++;
1019failed:
1020 if (IS_ERR(subdir)) {
1021 pr_err("sysctl could not get directory: ");
1022 sysctl_print_dir(dir);
1023 pr_cont("/%*.*s %ld\n",
1024 namelen, namelen, name, PTR_ERR(subdir));
1025 }
1026 drop_sysctl_table(&dir->header);
1027 if (new)
1028 drop_sysctl_table(&new->header);
1029 spin_unlock(&sysctl_lock);
1030 return subdir;
1031}
1032
1033static struct ctl_dir *xlate_dir(struct ctl_table_set *set, struct ctl_dir *dir)
1034{
1035 struct ctl_dir *parent;
1036 const char *procname;
1037 if (!dir->header.parent)
1038 return &set->dir;
1039 parent = xlate_dir(set, dir->header.parent);
1040 if (IS_ERR(parent))
1041 return parent;
1042 procname = dir->header.ctl_table[0].procname;
1043 return find_subdir(parent, procname, strlen(procname));
1044}
1045
1046static int sysctl_follow_link(struct ctl_table_header **phead,
1047 struct ctl_table **pentry)
1048{
1049 struct ctl_table_header *head;
1050 struct ctl_table_root *root;
1051 struct ctl_table_set *set;
1052 struct ctl_table *entry;
1053 struct ctl_dir *dir;
1054 int ret;
1055
1056 ret = 0;
1057 spin_lock(&sysctl_lock);
1058 root = (*pentry)->data;
1059 set = lookup_header_set(root);
1060 dir = xlate_dir(set, (*phead)->parent);
1061 if (IS_ERR(dir))
1062 ret = PTR_ERR(dir);
1063 else {
1064 const char *procname = (*pentry)->procname;
1065 head = NULL;
1066 entry = find_entry(&head, dir, procname, strlen(procname));
1067 ret = -ENOENT;
1068 if (entry && use_table(head)) {
1069 unuse_table(*phead);
1070 *phead = head;
1071 *pentry = entry;
1072 ret = 0;
1073 }
1074 }
1075
1076 spin_unlock(&sysctl_lock);
1077 return ret;
1078}
1079
1080static int sysctl_err(const char *path, struct ctl_table *table, char *fmt, ...)
1081{
1082 struct va_format vaf;
1083 va_list args;
1084
1085 va_start(args, fmt);
1086 vaf.fmt = fmt;
1087 vaf.va = &args;
1088
1089 pr_err("sysctl table check failed: %s/%s %pV\n",
1090 path, table->procname, &vaf);
1091
1092 va_end(args);
1093 return -EINVAL;
1094}
1095
1096static int sysctl_check_table_array(const char *path, struct ctl_table *table)
1097{
1098 int err = 0;
1099
1100 if ((table->proc_handler == proc_douintvec) ||
1101 (table->proc_handler == proc_douintvec_minmax)) {
1102 if (table->maxlen != sizeof(unsigned int))
1103 err |= sysctl_err(path, table, "array not allowed");
1104 }
1105
1106 if (table->proc_handler == proc_dou8vec_minmax) {
1107 if (table->maxlen != sizeof(u8))
1108 err |= sysctl_err(path, table, "array not allowed");
1109 }
1110
1111 return err;
1112}
1113
1114static int sysctl_check_table(const char *path, struct ctl_table *table)
1115{
1116 int err = 0;
1117 for (; table->procname; table++) {
1118 if (table->child)
1119 err |= sysctl_err(path, table, "Not a file");
1120
1121 if ((table->proc_handler == proc_dostring) ||
1122 (table->proc_handler == proc_dointvec) ||
1123 (table->proc_handler == proc_douintvec) ||
1124 (table->proc_handler == proc_douintvec_minmax) ||
1125 (table->proc_handler == proc_dointvec_minmax) ||
1126 (table->proc_handler == proc_dou8vec_minmax) ||
1127 (table->proc_handler == proc_dointvec_jiffies) ||
1128 (table->proc_handler == proc_dointvec_userhz_jiffies) ||
1129 (table->proc_handler == proc_dointvec_ms_jiffies) ||
1130 (table->proc_handler == proc_doulongvec_minmax) ||
1131 (table->proc_handler == proc_doulongvec_ms_jiffies_minmax)) {
1132 if (!table->data)
1133 err |= sysctl_err(path, table, "No data");
1134 if (!table->maxlen)
1135 err |= sysctl_err(path, table, "No maxlen");
1136 else
1137 err |= sysctl_check_table_array(path, table);
1138 }
1139 if (!table->proc_handler)
1140 err |= sysctl_err(path, table, "No proc_handler");
1141
1142 if ((table->mode & (S_IRUGO|S_IWUGO)) != table->mode)
1143 err |= sysctl_err(path, table, "bogus .mode 0%o",
1144 table->mode);
1145 }
1146 return err;
1147}
1148
1149static struct ctl_table_header *new_links(struct ctl_dir *dir, struct ctl_table *table,
1150 struct ctl_table_root *link_root)
1151{
1152 struct ctl_table *link_table, *entry, *link;
1153 struct ctl_table_header *links;
1154 struct ctl_node *node;
1155 char *link_name;
1156 int nr_entries, name_bytes;
1157
1158 name_bytes = 0;
1159 nr_entries = 0;
1160 for (entry = table; entry->procname; entry++) {
1161 nr_entries++;
1162 name_bytes += strlen(entry->procname) + 1;
1163 }
1164
1165 links = kzalloc(sizeof(struct ctl_table_header) +
1166 sizeof(struct ctl_node)*nr_entries +
1167 sizeof(struct ctl_table)*(nr_entries + 1) +
1168 name_bytes,
1169 GFP_KERNEL);
1170
1171 if (!links)
1172 return NULL;
1173
1174 node = (struct ctl_node *)(links + 1);
1175 link_table = (struct ctl_table *)(node + nr_entries);
1176 link_name = (char *)&link_table[nr_entries + 1];
1177
1178 for (link = link_table, entry = table; entry->procname; link++, entry++) {
1179 int len = strlen(entry->procname) + 1;
1180 memcpy(link_name, entry->procname, len);
1181 link->procname = link_name;
1182 link->mode = S_IFLNK|S_IRWXUGO;
1183 link->data = link_root;
1184 link_name += len;
1185 }
1186 init_header(links, dir->header.root, dir->header.set, node, link_table);
1187 links->nreg = nr_entries;
1188
1189 return links;
1190}
1191
1192static bool get_links(struct ctl_dir *dir,
1193 struct ctl_table *table, struct ctl_table_root *link_root)
1194{
1195 struct ctl_table_header *head;
1196 struct ctl_table *entry, *link;
1197
1198 /* Are there links available for every entry in table? */
1199 for (entry = table; entry->procname; entry++) {
1200 const char *procname = entry->procname;
1201 link = find_entry(&head, dir, procname, strlen(procname));
1202 if (!link)
1203 return false;
1204 if (S_ISDIR(link->mode) && S_ISDIR(entry->mode))
1205 continue;
1206 if (S_ISLNK(link->mode) && (link->data == link_root))
1207 continue;
1208 return false;
1209 }
1210
1211 /* The checks passed. Increase the registration count on the links */
1212 for (entry = table; entry->procname; entry++) {
1213 const char *procname = entry->procname;
1214 link = find_entry(&head, dir, procname, strlen(procname));
1215 head->nreg++;
1216 }
1217 return true;
1218}
1219
1220static int insert_links(struct ctl_table_header *head)
1221{
1222 struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1223 struct ctl_dir *core_parent = NULL;
1224 struct ctl_table_header *links;
1225 int err;
1226
1227 if (head->set == root_set)
1228 return 0;
1229
1230 core_parent = xlate_dir(root_set, head->parent);
1231 if (IS_ERR(core_parent))
1232 return 0;
1233
1234 if (get_links(core_parent, head->ctl_table, head->root))
1235 return 0;
1236
1237 core_parent->header.nreg++;
1238 spin_unlock(&sysctl_lock);
1239
1240 links = new_links(core_parent, head->ctl_table, head->root);
1241
1242 spin_lock(&sysctl_lock);
1243 err = -ENOMEM;
1244 if (!links)
1245 goto out;
1246
1247 err = 0;
1248 if (get_links(core_parent, head->ctl_table, head->root)) {
1249 kfree(links);
1250 goto out;
1251 }
1252
1253 err = insert_header(core_parent, links);
1254 if (err)
1255 kfree(links);
1256out:
1257 drop_sysctl_table(&core_parent->header);
1258 return err;
1259}
1260
1261/**
1262 * __register_sysctl_table - register a leaf sysctl table
1263 * @set: Sysctl tree to register on
1264 * @path: The path to the directory the sysctl table is in.
1265 * @table: the top-level table structure
1266 *
1267 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1268 * array. A completely 0 filled entry terminates the table.
1269 *
1270 * The members of the &struct ctl_table structure are used as follows:
1271 *
1272 * procname - the name of the sysctl file under /proc/sys. Set to %NULL to not
1273 * enter a sysctl file
1274 *
1275 * data - a pointer to data for use by proc_handler
1276 *
1277 * maxlen - the maximum size in bytes of the data
1278 *
1279 * mode - the file permissions for the /proc/sys file
1280 *
1281 * child - must be %NULL.
1282 *
1283 * proc_handler - the text handler routine (described below)
1284 *
1285 * extra1, extra2 - extra pointers usable by the proc handler routines
1286 *
1287 * Leaf nodes in the sysctl tree will be represented by a single file
1288 * under /proc; non-leaf nodes will be represented by directories.
1289 *
1290 * There must be a proc_handler routine for any terminal nodes.
1291 * Several default handlers are available to cover common cases -
1292 *
1293 * proc_dostring(), proc_dointvec(), proc_dointvec_jiffies(),
1294 * proc_dointvec_userhz_jiffies(), proc_dointvec_minmax(),
1295 * proc_doulongvec_ms_jiffies_minmax(), proc_doulongvec_minmax()
1296 *
1297 * It is the handler's job to read the input buffer from user memory
1298 * and process it. The handler should return 0 on success.
1299 *
1300 * This routine returns %NULL on a failure to register, and a pointer
1301 * to the table header on success.
1302 */
1303struct ctl_table_header *__register_sysctl_table(
1304 struct ctl_table_set *set,
1305 const char *path, struct ctl_table *table)
1306{
1307 struct ctl_table_root *root = set->dir.header.root;
1308 struct ctl_table_header *header;
1309 const char *name, *nextname;
1310 struct ctl_dir *dir;
1311 struct ctl_table *entry;
1312 struct ctl_node *node;
1313 int nr_entries = 0;
1314
1315 for (entry = table; entry->procname; entry++)
1316 nr_entries++;
1317
1318 header = kzalloc(sizeof(struct ctl_table_header) +
1319 sizeof(struct ctl_node)*nr_entries, GFP_KERNEL);
1320 if (!header)
1321 return NULL;
1322
1323 node = (struct ctl_node *)(header + 1);
1324 init_header(header, root, set, node, table);
1325 if (sysctl_check_table(path, table))
1326 goto fail;
1327
1328 spin_lock(&sysctl_lock);
1329 dir = &set->dir;
1330 /* Reference moved down the diretory tree get_subdir */
1331 dir->header.nreg++;
1332 spin_unlock(&sysctl_lock);
1333
1334 /* Find the directory for the ctl_table */
1335 for (name = path; name; name = nextname) {
1336 int namelen;
1337 nextname = strchr(name, '/');
1338 if (nextname) {
1339 namelen = nextname - name;
1340 nextname++;
1341 } else {
1342 namelen = strlen(name);
1343 }
1344 if (namelen == 0)
1345 continue;
1346
1347 dir = get_subdir(dir, name, namelen);
1348 if (IS_ERR(dir))
1349 goto fail;
1350 }
1351
1352 spin_lock(&sysctl_lock);
1353 if (insert_header(dir, header))
1354 goto fail_put_dir_locked;
1355
1356 drop_sysctl_table(&dir->header);
1357 spin_unlock(&sysctl_lock);
1358
1359 return header;
1360
1361fail_put_dir_locked:
1362 drop_sysctl_table(&dir->header);
1363 spin_unlock(&sysctl_lock);
1364fail:
1365 kfree(header);
1366 dump_stack();
1367 return NULL;
1368}
1369
1370/**
1371 * register_sysctl - register a sysctl table
1372 * @path: The path to the directory the sysctl table is in.
1373 * @table: the table structure
1374 *
1375 * Register a sysctl table. @table should be a filled in ctl_table
1376 * array. A completely 0 filled entry terminates the table.
1377 *
1378 * See __register_sysctl_table for more details.
1379 */
1380struct ctl_table_header *register_sysctl(const char *path, struct ctl_table *table)
1381{
1382 return __register_sysctl_table(&sysctl_table_root.default_set,
1383 path, table);
1384}
1385EXPORT_SYMBOL(register_sysctl);
1386
1387static char *append_path(const char *path, char *pos, const char *name)
1388{
1389 int namelen;
1390 namelen = strlen(name);
1391 if (((pos - path) + namelen + 2) >= PATH_MAX)
1392 return NULL;
1393 memcpy(pos, name, namelen);
1394 pos[namelen] = '/';
1395 pos[namelen + 1] = '\0';
1396 pos += namelen + 1;
1397 return pos;
1398}
1399
1400static int count_subheaders(struct ctl_table *table)
1401{
1402 int has_files = 0;
1403 int nr_subheaders = 0;
1404 struct ctl_table *entry;
1405
1406 /* special case: no directory and empty directory */
1407 if (!table || !table->procname)
1408 return 1;
1409
1410 for (entry = table; entry->procname; entry++) {
1411 if (entry->child)
1412 nr_subheaders += count_subheaders(entry->child);
1413 else
1414 has_files = 1;
1415 }
1416 return nr_subheaders + has_files;
1417}
1418
1419static int register_leaf_sysctl_tables(const char *path, char *pos,
1420 struct ctl_table_header ***subheader, struct ctl_table_set *set,
1421 struct ctl_table *table)
1422{
1423 struct ctl_table *ctl_table_arg = NULL;
1424 struct ctl_table *entry, *files;
1425 int nr_files = 0;
1426 int nr_dirs = 0;
1427 int err = -ENOMEM;
1428
1429 for (entry = table; entry->procname; entry++) {
1430 if (entry->child)
1431 nr_dirs++;
1432 else
1433 nr_files++;
1434 }
1435
1436 files = table;
1437 /* If there are mixed files and directories we need a new table */
1438 if (nr_dirs && nr_files) {
1439 struct ctl_table *new;
1440 files = kcalloc(nr_files + 1, sizeof(struct ctl_table),
1441 GFP_KERNEL);
1442 if (!files)
1443 goto out;
1444
1445 ctl_table_arg = files;
1446 for (new = files, entry = table; entry->procname; entry++) {
1447 if (entry->child)
1448 continue;
1449 *new = *entry;
1450 new++;
1451 }
1452 }
1453
1454 /* Register everything except a directory full of subdirectories */
1455 if (nr_files || !nr_dirs) {
1456 struct ctl_table_header *header;
1457 header = __register_sysctl_table(set, path, files);
1458 if (!header) {
1459 kfree(ctl_table_arg);
1460 goto out;
1461 }
1462
1463 /* Remember if we need to free the file table */
1464 header->ctl_table_arg = ctl_table_arg;
1465 **subheader = header;
1466 (*subheader)++;
1467 }
1468
1469 /* Recurse into the subdirectories. */
1470 for (entry = table; entry->procname; entry++) {
1471 char *child_pos;
1472
1473 if (!entry->child)
1474 continue;
1475
1476 err = -ENAMETOOLONG;
1477 child_pos = append_path(path, pos, entry->procname);
1478 if (!child_pos)
1479 goto out;
1480
1481 err = register_leaf_sysctl_tables(path, child_pos, subheader,
1482 set, entry->child);
1483 pos[0] = '\0';
1484 if (err)
1485 goto out;
1486 }
1487 err = 0;
1488out:
1489 /* On failure our caller will unregister all registered subheaders */
1490 return err;
1491}
1492
1493/**
1494 * __register_sysctl_paths - register a sysctl table hierarchy
1495 * @set: Sysctl tree to register on
1496 * @path: The path to the directory the sysctl table is in.
1497 * @table: the top-level table structure
1498 *
1499 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1500 * array. A completely 0 filled entry terminates the table.
1501 *
1502 * See __register_sysctl_table for more details.
1503 */
1504struct ctl_table_header *__register_sysctl_paths(
1505 struct ctl_table_set *set,
1506 const struct ctl_path *path, struct ctl_table *table)
1507{
1508 struct ctl_table *ctl_table_arg = table;
1509 int nr_subheaders = count_subheaders(table);
1510 struct ctl_table_header *header = NULL, **subheaders, **subheader;
1511 const struct ctl_path *component;
1512 char *new_path, *pos;
1513
1514 pos = new_path = kmalloc(PATH_MAX, GFP_KERNEL);
1515 if (!new_path)
1516 return NULL;
1517
1518 pos[0] = '\0';
1519 for (component = path; component->procname; component++) {
1520 pos = append_path(new_path, pos, component->procname);
1521 if (!pos)
1522 goto out;
1523 }
1524 while (table->procname && table->child && !table[1].procname) {
1525 pos = append_path(new_path, pos, table->procname);
1526 if (!pos)
1527 goto out;
1528 table = table->child;
1529 }
1530 if (nr_subheaders == 1) {
1531 header = __register_sysctl_table(set, new_path, table);
1532 if (header)
1533 header->ctl_table_arg = ctl_table_arg;
1534 } else {
1535 header = kzalloc(sizeof(*header) +
1536 sizeof(*subheaders)*nr_subheaders, GFP_KERNEL);
1537 if (!header)
1538 goto out;
1539
1540 subheaders = (struct ctl_table_header **) (header + 1);
1541 subheader = subheaders;
1542 header->ctl_table_arg = ctl_table_arg;
1543
1544 if (register_leaf_sysctl_tables(new_path, pos, &subheader,
1545 set, table))
1546 goto err_register_leaves;
1547 }
1548
1549out:
1550 kfree(new_path);
1551 return header;
1552
1553err_register_leaves:
1554 while (subheader > subheaders) {
1555 struct ctl_table_header *subh = *(--subheader);
1556 struct ctl_table *table = subh->ctl_table_arg;
1557 unregister_sysctl_table(subh);
1558 kfree(table);
1559 }
1560 kfree(header);
1561 header = NULL;
1562 goto out;
1563}
1564
1565/**
1566 * register_sysctl_paths - register a sysctl table hierarchy
1567 * @path: The path to the directory the sysctl table is in.
1568 * @table: the top-level table structure
1569 *
1570 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1571 * array. A completely 0 filled entry terminates the table.
1572 *
1573 * See __register_sysctl_paths for more details.
1574 */
1575struct ctl_table_header *register_sysctl_paths(const struct ctl_path *path,
1576 struct ctl_table *table)
1577{
1578 return __register_sysctl_paths(&sysctl_table_root.default_set,
1579 path, table);
1580}
1581EXPORT_SYMBOL(register_sysctl_paths);
1582
1583/**
1584 * register_sysctl_table - register a sysctl table hierarchy
1585 * @table: the top-level table structure
1586 *
1587 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1588 * array. A completely 0 filled entry terminates the table.
1589 *
1590 * See register_sysctl_paths for more details.
1591 */
1592struct ctl_table_header *register_sysctl_table(struct ctl_table *table)
1593{
1594 static const struct ctl_path null_path[] = { {} };
1595
1596 return register_sysctl_paths(null_path, table);
1597}
1598EXPORT_SYMBOL(register_sysctl_table);
1599
1600static void put_links(struct ctl_table_header *header)
1601{
1602 struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1603 struct ctl_table_root *root = header->root;
1604 struct ctl_dir *parent = header->parent;
1605 struct ctl_dir *core_parent;
1606 struct ctl_table *entry;
1607
1608 if (header->set == root_set)
1609 return;
1610
1611 core_parent = xlate_dir(root_set, parent);
1612 if (IS_ERR(core_parent))
1613 return;
1614
1615 for (entry = header->ctl_table; entry->procname; entry++) {
1616 struct ctl_table_header *link_head;
1617 struct ctl_table *link;
1618 const char *name = entry->procname;
1619
1620 link = find_entry(&link_head, core_parent, name, strlen(name));
1621 if (link &&
1622 ((S_ISDIR(link->mode) && S_ISDIR(entry->mode)) ||
1623 (S_ISLNK(link->mode) && (link->data == root)))) {
1624 drop_sysctl_table(link_head);
1625 }
1626 else {
1627 pr_err("sysctl link missing during unregister: ");
1628 sysctl_print_dir(parent);
1629 pr_cont("/%s\n", name);
1630 }
1631 }
1632}
1633
1634static void drop_sysctl_table(struct ctl_table_header *header)
1635{
1636 struct ctl_dir *parent = header->parent;
1637
1638 if (--header->nreg)
1639 return;
1640
1641 if (parent) {
1642 put_links(header);
1643 start_unregistering(header);
1644 }
1645
1646 if (!--header->count)
1647 kfree_rcu(header, rcu);
1648
1649 if (parent)
1650 drop_sysctl_table(&parent->header);
1651}
1652
1653/**
1654 * unregister_sysctl_table - unregister a sysctl table hierarchy
1655 * @header: the header returned from register_sysctl_table
1656 *
1657 * Unregisters the sysctl table and all children. proc entries may not
1658 * actually be removed until they are no longer used by anyone.
1659 */
1660void unregister_sysctl_table(struct ctl_table_header * header)
1661{
1662 int nr_subheaders;
1663 might_sleep();
1664
1665 if (header == NULL)
1666 return;
1667
1668 nr_subheaders = count_subheaders(header->ctl_table_arg);
1669 if (unlikely(nr_subheaders > 1)) {
1670 struct ctl_table_header **subheaders;
1671 int i;
1672
1673 subheaders = (struct ctl_table_header **)(header + 1);
1674 for (i = nr_subheaders -1; i >= 0; i--) {
1675 struct ctl_table_header *subh = subheaders[i];
1676 struct ctl_table *table = subh->ctl_table_arg;
1677 unregister_sysctl_table(subh);
1678 kfree(table);
1679 }
1680 kfree(header);
1681 return;
1682 }
1683
1684 spin_lock(&sysctl_lock);
1685 drop_sysctl_table(header);
1686 spin_unlock(&sysctl_lock);
1687}
1688EXPORT_SYMBOL(unregister_sysctl_table);
1689
1690void setup_sysctl_set(struct ctl_table_set *set,
1691 struct ctl_table_root *root,
1692 int (*is_seen)(struct ctl_table_set *))
1693{
1694 memset(set, 0, sizeof(*set));
1695 set->is_seen = is_seen;
1696 init_header(&set->dir.header, root, set, NULL, root_table);
1697}
1698
1699void retire_sysctl_set(struct ctl_table_set *set)
1700{
1701 WARN_ON(!RB_EMPTY_ROOT(&set->dir.root));
1702}
1703
1704int __init proc_sys_init(void)
1705{
1706 struct proc_dir_entry *proc_sys_root;
1707
1708 proc_sys_root = proc_mkdir("sys", NULL);
1709 proc_sys_root->proc_iops = &proc_sys_dir_operations;
1710 proc_sys_root->proc_dir_ops = &proc_sys_dir_file_operations;
1711 proc_sys_root->nlink = 0;
1712
1713 return sysctl_init();
1714}
1715
1716struct sysctl_alias {
1717 const char *kernel_param;
1718 const char *sysctl_param;
1719};
1720
1721/*
1722 * Historically some settings had both sysctl and a command line parameter.
1723 * With the generic sysctl. parameter support, we can handle them at a single
1724 * place and only keep the historical name for compatibility. This is not meant
1725 * to add brand new aliases. When adding existing aliases, consider whether
1726 * the possibly different moment of changing the value (e.g. from early_param
1727 * to the moment do_sysctl_args() is called) is an issue for the specific
1728 * parameter.
1729 */
1730static const struct sysctl_alias sysctl_aliases[] = {
1731 {"hardlockup_all_cpu_backtrace", "kernel.hardlockup_all_cpu_backtrace" },
1732 {"hung_task_panic", "kernel.hung_task_panic" },
1733 {"numa_zonelist_order", "vm.numa_zonelist_order" },
1734 {"softlockup_all_cpu_backtrace", "kernel.softlockup_all_cpu_backtrace" },
1735 {"softlockup_panic", "kernel.softlockup_panic" },
1736 { }
1737};
1738
1739static const char *sysctl_find_alias(char *param)
1740{
1741 const struct sysctl_alias *alias;
1742
1743 for (alias = &sysctl_aliases[0]; alias->kernel_param != NULL; alias++) {
1744 if (strcmp(alias->kernel_param, param) == 0)
1745 return alias->sysctl_param;
1746 }
1747
1748 return NULL;
1749}
1750
1751/* Set sysctl value passed on kernel command line. */
1752static int process_sysctl_arg(char *param, char *val,
1753 const char *unused, void *arg)
1754{
1755 char *path;
1756 struct vfsmount **proc_mnt = arg;
1757 struct file_system_type *proc_fs_type;
1758 struct file *file;
1759 int len;
1760 int err;
1761 loff_t pos = 0;
1762 ssize_t wret;
1763
1764 if (strncmp(param, "sysctl", sizeof("sysctl") - 1) == 0) {
1765 param += sizeof("sysctl") - 1;
1766
1767 if (param[0] != '/' && param[0] != '.')
1768 return 0;
1769
1770 param++;
1771 } else {
1772 param = (char *) sysctl_find_alias(param);
1773 if (!param)
1774 return 0;
1775 }
1776
1777 if (!val)
1778 return -EINVAL;
1779 len = strlen(val);
1780 if (len == 0)
1781 return -EINVAL;
1782
1783 /*
1784 * To set sysctl options, we use a temporary mount of proc, look up the
1785 * respective sys/ file and write to it. To avoid mounting it when no
1786 * options were given, we mount it only when the first sysctl option is
1787 * found. Why not a persistent mount? There are problems with a
1788 * persistent mount of proc in that it forces userspace not to use any
1789 * proc mount options.
1790 */
1791 if (!*proc_mnt) {
1792 proc_fs_type = get_fs_type("proc");
1793 if (!proc_fs_type) {
1794 pr_err("Failed to find procfs to set sysctl from command line\n");
1795 return 0;
1796 }
1797 *proc_mnt = kern_mount(proc_fs_type);
1798 put_filesystem(proc_fs_type);
1799 if (IS_ERR(*proc_mnt)) {
1800 pr_err("Failed to mount procfs to set sysctl from command line\n");
1801 return 0;
1802 }
1803 }
1804
1805 path = kasprintf(GFP_KERNEL, "sys/%s", param);
1806 if (!path)
1807 panic("%s: Failed to allocate path for %s\n", __func__, param);
1808 strreplace(path, '.', '/');
1809
1810 file = file_open_root_mnt(*proc_mnt, path, O_WRONLY, 0);
1811 if (IS_ERR(file)) {
1812 err = PTR_ERR(file);
1813 if (err == -ENOENT)
1814 pr_err("Failed to set sysctl parameter '%s=%s': parameter not found\n",
1815 param, val);
1816 else if (err == -EACCES)
1817 pr_err("Failed to set sysctl parameter '%s=%s': permission denied (read-only?)\n",
1818 param, val);
1819 else
1820 pr_err("Error %pe opening proc file to set sysctl parameter '%s=%s'\n",
1821 file, param, val);
1822 goto out;
1823 }
1824 wret = kernel_write(file, val, len, &pos);
1825 if (wret < 0) {
1826 err = wret;
1827 if (err == -EINVAL)
1828 pr_err("Failed to set sysctl parameter '%s=%s': invalid value\n",
1829 param, val);
1830 else
1831 pr_err("Error %pe writing to proc file to set sysctl parameter '%s=%s'\n",
1832 ERR_PTR(err), param, val);
1833 } else if (wret != len) {
1834 pr_err("Wrote only %zd bytes of %d writing to proc file %s to set sysctl parameter '%s=%s\n",
1835 wret, len, path, param, val);
1836 }
1837
1838 err = filp_close(file, NULL);
1839 if (err)
1840 pr_err("Error %pe closing proc file to set sysctl parameter '%s=%s\n",
1841 ERR_PTR(err), param, val);
1842out:
1843 kfree(path);
1844 return 0;
1845}
1846
1847void do_sysctl_args(void)
1848{
1849 char *command_line;
1850 struct vfsmount *proc_mnt = NULL;
1851
1852 command_line = kstrdup(saved_command_line, GFP_KERNEL);
1853 if (!command_line)
1854 panic("%s: Failed to allocate copy of command line\n", __func__);
1855
1856 parse_args("Setting sysctl args", command_line,
1857 NULL, 0, -1, -1, &proc_mnt, process_sysctl_arg);
1858
1859 if (proc_mnt)
1860 kern_unmount(proc_mnt);
1861
1862 kfree(command_line);
1863}
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}