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 <linux/kmemleak.h>
20 #include "internal.h"
21
22 #define list_for_each_table_entry(entry, header) \
23 entry = header->ctl_table; \
24 for (size_t i = 0 ; i < header->ctl_table_size; ++i, entry++)
25
26 static const struct dentry_operations proc_sys_dentry_operations;
27 static const struct file_operations proc_sys_file_operations;
28 static const struct inode_operations proc_sys_inode_operations;
29 static const struct file_operations proc_sys_dir_file_operations;
30 static const struct inode_operations proc_sys_dir_operations;
31
32 /*
33 * Support for permanently empty directories.
34 * Must be non-empty to avoid sharing an address with other tables.
35 */
36 static struct ctl_table sysctl_mount_point[] = {
37 { }
38 };
39
40 /**
41 * register_sysctl_mount_point() - registers a sysctl mount point
42 * @path: path for the mount point
43 *
44 * Used to create a permanently empty directory to serve as mount point.
45 * There are some subtle but important permission checks this allows in the
46 * case of unprivileged mounts.
47 */
48 struct ctl_table_header *register_sysctl_mount_point(const char *path)
49 {
50 return register_sysctl_sz(path, sysctl_mount_point, 0);
51 }
52 EXPORT_SYMBOL(register_sysctl_mount_point);
53
54 #define sysctl_is_perm_empty_ctl_header(hptr) \
55 (hptr->type == SYSCTL_TABLE_TYPE_PERMANENTLY_EMPTY)
56 #define sysctl_set_perm_empty_ctl_header(hptr) \
57 (hptr->type = SYSCTL_TABLE_TYPE_PERMANENTLY_EMPTY)
58 #define sysctl_clear_perm_empty_ctl_header(hptr) \
59 (hptr->type = SYSCTL_TABLE_TYPE_DEFAULT)
60
61 void proc_sys_poll_notify(struct ctl_table_poll *poll)
62 {
63 if (!poll)
64 return;
65
66 atomic_inc(&poll->event);
67 wake_up_interruptible(&poll->wait);
68 }
69
70 static struct ctl_table root_table[] = {
71 {
72 .procname = "",
73 .mode = S_IFDIR|S_IRUGO|S_IXUGO,
74 },
75 };
76 static struct ctl_table_root sysctl_table_root = {
77 .default_set.dir.header = {
78 {{.count = 1,
79 .nreg = 1,
80 .ctl_table = root_table }},
81 .ctl_table_arg = root_table,
82 .root = &sysctl_table_root,
83 .set = &sysctl_table_root.default_set,
84 },
85 };
86
87 static DEFINE_SPINLOCK(sysctl_lock);
88
89 static void drop_sysctl_table(struct ctl_table_header *header);
90 static int sysctl_follow_link(struct ctl_table_header **phead,
91 struct ctl_table **pentry);
92 static int insert_links(struct ctl_table_header *head);
93 static void put_links(struct ctl_table_header *header);
94
95 static void sysctl_print_dir(struct ctl_dir *dir)
96 {
97 if (dir->header.parent)
98 sysctl_print_dir(dir->header.parent);
99 pr_cont("%s/", dir->header.ctl_table[0].procname);
100 }
101
102 static int namecmp(const char *name1, int len1, const char *name2, int len2)
103 {
104 int cmp;
105
106 cmp = memcmp(name1, name2, min(len1, len2));
107 if (cmp == 0)
108 cmp = len1 - len2;
109 return cmp;
110 }
111
112 /* Called under sysctl_lock */
113 static struct ctl_table *find_entry(struct ctl_table_header **phead,
114 struct ctl_dir *dir, const char *name, int namelen)
115 {
116 struct ctl_table_header *head;
117 struct ctl_table *entry;
118 struct rb_node *node = dir->root.rb_node;
119
120 while (node)
121 {
122 struct ctl_node *ctl_node;
123 const char *procname;
124 int cmp;
125
126 ctl_node = rb_entry(node, struct ctl_node, node);
127 head = ctl_node->header;
128 entry = &head->ctl_table[ctl_node - head->node];
129 procname = entry->procname;
130
131 cmp = namecmp(name, namelen, procname, strlen(procname));
132 if (cmp < 0)
133 node = node->rb_left;
134 else if (cmp > 0)
135 node = node->rb_right;
136 else {
137 *phead = head;
138 return entry;
139 }
140 }
141 return NULL;
142 }
143
144 static int insert_entry(struct ctl_table_header *head, struct ctl_table *entry)
145 {
146 struct rb_node *node = &head->node[entry - head->ctl_table].node;
147 struct rb_node **p = &head->parent->root.rb_node;
148 struct rb_node *parent = NULL;
149 const char *name = entry->procname;
150 int namelen = strlen(name);
151
152 while (*p) {
153 struct ctl_table_header *parent_head;
154 struct ctl_table *parent_entry;
155 struct ctl_node *parent_node;
156 const char *parent_name;
157 int cmp;
158
159 parent = *p;
160 parent_node = rb_entry(parent, struct ctl_node, node);
161 parent_head = parent_node->header;
162 parent_entry = &parent_head->ctl_table[parent_node - parent_head->node];
163 parent_name = parent_entry->procname;
164
165 cmp = namecmp(name, namelen, parent_name, strlen(parent_name));
166 if (cmp < 0)
167 p = &(*p)->rb_left;
168 else if (cmp > 0)
169 p = &(*p)->rb_right;
170 else {
171 pr_err("sysctl duplicate entry: ");
172 sysctl_print_dir(head->parent);
173 pr_cont("%s\n", entry->procname);
174 return -EEXIST;
175 }
176 }
177
178 rb_link_node(node, parent, p);
179 rb_insert_color(node, &head->parent->root);
180 return 0;
181 }
182
183 static void erase_entry(struct ctl_table_header *head, struct ctl_table *entry)
184 {
185 struct rb_node *node = &head->node[entry - head->ctl_table].node;
186
187 rb_erase(node, &head->parent->root);
188 }
189
190 static void init_header(struct ctl_table_header *head,
191 struct ctl_table_root *root, struct ctl_table_set *set,
192 struct ctl_node *node, struct ctl_table *table, size_t table_size)
193 {
194 head->ctl_table = table;
195 head->ctl_table_size = table_size;
196 head->ctl_table_arg = table;
197 head->used = 0;
198 head->count = 1;
199 head->nreg = 1;
200 head->unregistering = NULL;
201 head->root = root;
202 head->set = set;
203 head->parent = NULL;
204 head->node = node;
205 INIT_HLIST_HEAD(&head->inodes);
206 if (node) {
207 struct ctl_table *entry;
208
209 list_for_each_table_entry(entry, head) {
210 node->header = head;
211 node++;
212 }
213 }
214 if (table == sysctl_mount_point)
215 sysctl_set_perm_empty_ctl_header(head);
216 }
217
218 static void erase_header(struct ctl_table_header *head)
219 {
220 struct ctl_table *entry;
221
222 list_for_each_table_entry(entry, head)
223 erase_entry(head, entry);
224 }
225
226 static int insert_header(struct ctl_dir *dir, struct ctl_table_header *header)
227 {
228 struct ctl_table *entry;
229 struct ctl_table_header *dir_h = &dir->header;
230 int err;
231
232
233 /* Is this a permanently empty directory? */
234 if (sysctl_is_perm_empty_ctl_header(dir_h))
235 return -EROFS;
236
237 /* Am I creating a permanently empty directory? */
238 if (sysctl_is_perm_empty_ctl_header(header)) {
239 if (!RB_EMPTY_ROOT(&dir->root))
240 return -EINVAL;
241 sysctl_set_perm_empty_ctl_header(dir_h);
242 }
243
244 dir_h->nreg++;
245 header->parent = dir;
246 err = insert_links(header);
247 if (err)
248 goto fail_links;
249 list_for_each_table_entry(entry, header) {
250 err = insert_entry(header, entry);
251 if (err)
252 goto fail;
253 }
254 return 0;
255 fail:
256 erase_header(header);
257 put_links(header);
258 fail_links:
259 if (header->ctl_table == sysctl_mount_point)
260 sysctl_clear_perm_empty_ctl_header(dir_h);
261 header->parent = NULL;
262 drop_sysctl_table(dir_h);
263 return err;
264 }
265
266 /* called under sysctl_lock */
267 static int use_table(struct ctl_table_header *p)
268 {
269 if (unlikely(p->unregistering))
270 return 0;
271 p->used++;
272 return 1;
273 }
274
275 /* called under sysctl_lock */
276 static void unuse_table(struct ctl_table_header *p)
277 {
278 if (!--p->used)
279 if (unlikely(p->unregistering))
280 complete(p->unregistering);
281 }
282
283 static void proc_sys_invalidate_dcache(struct ctl_table_header *head)
284 {
285 proc_invalidate_siblings_dcache(&head->inodes, &sysctl_lock);
286 }
287
288 /* called under sysctl_lock, will reacquire if has to wait */
289 static void start_unregistering(struct ctl_table_header *p)
290 {
291 /*
292 * if p->used is 0, nobody will ever touch that entry again;
293 * we'll eliminate all paths to it before dropping sysctl_lock
294 */
295 if (unlikely(p->used)) {
296 struct completion wait;
297 init_completion(&wait);
298 p->unregistering = &wait;
299 spin_unlock(&sysctl_lock);
300 wait_for_completion(&wait);
301 } else {
302 /* anything non-NULL; we'll never dereference it */
303 p->unregistering = ERR_PTR(-EINVAL);
304 spin_unlock(&sysctl_lock);
305 }
306 /*
307 * Invalidate dentries for unregistered sysctls: namespaced sysctls
308 * can have duplicate names and contaminate dcache very badly.
309 */
310 proc_sys_invalidate_dcache(p);
311 /*
312 * do not remove from the list until nobody holds it; walking the
313 * list in do_sysctl() relies on that.
314 */
315 spin_lock(&sysctl_lock);
316 erase_header(p);
317 }
318
319 static struct ctl_table_header *sysctl_head_grab(struct ctl_table_header *head)
320 {
321 BUG_ON(!head);
322 spin_lock(&sysctl_lock);
323 if (!use_table(head))
324 head = ERR_PTR(-ENOENT);
325 spin_unlock(&sysctl_lock);
326 return head;
327 }
328
329 static void sysctl_head_finish(struct ctl_table_header *head)
330 {
331 if (!head)
332 return;
333 spin_lock(&sysctl_lock);
334 unuse_table(head);
335 spin_unlock(&sysctl_lock);
336 }
337
338 static struct ctl_table_set *
339 lookup_header_set(struct ctl_table_root *root)
340 {
341 struct ctl_table_set *set = &root->default_set;
342 if (root->lookup)
343 set = root->lookup(root);
344 return set;
345 }
346
347 static struct ctl_table *lookup_entry(struct ctl_table_header **phead,
348 struct ctl_dir *dir,
349 const char *name, int namelen)
350 {
351 struct ctl_table_header *head;
352 struct ctl_table *entry;
353
354 spin_lock(&sysctl_lock);
355 entry = find_entry(&head, dir, name, namelen);
356 if (entry && use_table(head))
357 *phead = head;
358 else
359 entry = NULL;
360 spin_unlock(&sysctl_lock);
361 return entry;
362 }
363
364 static struct ctl_node *first_usable_entry(struct rb_node *node)
365 {
366 struct ctl_node *ctl_node;
367
368 for (;node; node = rb_next(node)) {
369 ctl_node = rb_entry(node, struct ctl_node, node);
370 if (use_table(ctl_node->header))
371 return ctl_node;
372 }
373 return NULL;
374 }
375
376 static void first_entry(struct ctl_dir *dir,
377 struct ctl_table_header **phead, struct ctl_table **pentry)
378 {
379 struct ctl_table_header *head = NULL;
380 struct ctl_table *entry = NULL;
381 struct ctl_node *ctl_node;
382
383 spin_lock(&sysctl_lock);
384 ctl_node = first_usable_entry(rb_first(&dir->root));
385 spin_unlock(&sysctl_lock);
386 if (ctl_node) {
387 head = ctl_node->header;
388 entry = &head->ctl_table[ctl_node - head->node];
389 }
390 *phead = head;
391 *pentry = entry;
392 }
393
394 static void next_entry(struct ctl_table_header **phead, struct ctl_table **pentry)
395 {
396 struct ctl_table_header *head = *phead;
397 struct ctl_table *entry = *pentry;
398 struct ctl_node *ctl_node = &head->node[entry - head->ctl_table];
399
400 spin_lock(&sysctl_lock);
401 unuse_table(head);
402
403 ctl_node = first_usable_entry(rb_next(&ctl_node->node));
404 spin_unlock(&sysctl_lock);
405 head = NULL;
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
414 /*
415 * sysctl_perm does NOT grant the superuser all rights automatically, because
416 * some sysctl variables are readonly even to root.
417 */
418
419 static int test_perm(int mode, int op)
420 {
421 if (uid_eq(current_euid(), GLOBAL_ROOT_UID))
422 mode >>= 6;
423 else if (in_egroup_p(GLOBAL_ROOT_GID))
424 mode >>= 3;
425 if ((op & ~mode & (MAY_READ|MAY_WRITE|MAY_EXEC)) == 0)
426 return 0;
427 return -EACCES;
428 }
429
430 static int sysctl_perm(struct ctl_table_header *head, struct ctl_table *table, int op)
431 {
432 struct ctl_table_root *root = head->root;
433 int mode;
434
435 if (root->permissions)
436 mode = root->permissions(head, table);
437 else
438 mode = table->mode;
439
440 return test_perm(mode, op);
441 }
442
443 static struct inode *proc_sys_make_inode(struct super_block *sb,
444 struct ctl_table_header *head, struct ctl_table *table)
445 {
446 struct ctl_table_root *root = head->root;
447 struct inode *inode;
448 struct proc_inode *ei;
449
450 inode = new_inode(sb);
451 if (!inode)
452 return ERR_PTR(-ENOMEM);
453
454 inode->i_ino = get_next_ino();
455
456 ei = PROC_I(inode);
457
458 spin_lock(&sysctl_lock);
459 if (unlikely(head->unregistering)) {
460 spin_unlock(&sysctl_lock);
461 iput(inode);
462 return ERR_PTR(-ENOENT);
463 }
464 ei->sysctl = head;
465 ei->sysctl_entry = table;
466 hlist_add_head_rcu(&ei->sibling_inodes, &head->inodes);
467 head->count++;
468 spin_unlock(&sysctl_lock);
469
470 simple_inode_init_ts(inode);
471 inode->i_mode = table->mode;
472 if (!S_ISDIR(table->mode)) {
473 inode->i_mode |= S_IFREG;
474 inode->i_op = &proc_sys_inode_operations;
475 inode->i_fop = &proc_sys_file_operations;
476 } else {
477 inode->i_mode |= S_IFDIR;
478 inode->i_op = &proc_sys_dir_operations;
479 inode->i_fop = &proc_sys_dir_file_operations;
480 if (sysctl_is_perm_empty_ctl_header(head))
481 make_empty_dir_inode(inode);
482 }
483
484 inode->i_uid = GLOBAL_ROOT_UID;
485 inode->i_gid = GLOBAL_ROOT_GID;
486 if (root->set_ownership)
487 root->set_ownership(head, &inode->i_uid, &inode->i_gid);
488
489 return inode;
490 }
491
492 void proc_sys_evict_inode(struct inode *inode, struct ctl_table_header *head)
493 {
494 spin_lock(&sysctl_lock);
495 hlist_del_init_rcu(&PROC_I(inode)->sibling_inodes);
496 if (!--head->count)
497 kfree_rcu(head, rcu);
498 spin_unlock(&sysctl_lock);
499 }
500
501 static struct ctl_table_header *grab_header(struct inode *inode)
502 {
503 struct ctl_table_header *head = PROC_I(inode)->sysctl;
504 if (!head)
505 head = &sysctl_table_root.default_set.dir.header;
506 return sysctl_head_grab(head);
507 }
508
509 static struct dentry *proc_sys_lookup(struct inode *dir, struct dentry *dentry,
510 unsigned int flags)
511 {
512 struct ctl_table_header *head = grab_header(dir);
513 struct ctl_table_header *h = NULL;
514 const struct qstr *name = &dentry->d_name;
515 struct ctl_table *p;
516 struct inode *inode;
517 struct dentry *err = ERR_PTR(-ENOENT);
518 struct ctl_dir *ctl_dir;
519 int ret;
520
521 if (IS_ERR(head))
522 return ERR_CAST(head);
523
524 ctl_dir = container_of(head, struct ctl_dir, header);
525
526 p = lookup_entry(&h, ctl_dir, name->name, name->len);
527 if (!p)
528 goto out;
529
530 if (S_ISLNK(p->mode)) {
531 ret = sysctl_follow_link(&h, &p);
532 err = ERR_PTR(ret);
533 if (ret)
534 goto out;
535 }
536
537 d_set_d_op(dentry, &proc_sys_dentry_operations);
538 inode = proc_sys_make_inode(dir->i_sb, h ? h : head, p);
539 err = d_splice_alias(inode, dentry);
540
541 out:
542 if (h)
543 sysctl_head_finish(h);
544 sysctl_head_finish(head);
545 return err;
546 }
547
548 static ssize_t proc_sys_call_handler(struct kiocb *iocb, struct iov_iter *iter,
549 int write)
550 {
551 struct inode *inode = file_inode(iocb->ki_filp);
552 struct ctl_table_header *head = grab_header(inode);
553 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
554 size_t count = iov_iter_count(iter);
555 char *kbuf;
556 ssize_t error;
557
558 if (IS_ERR(head))
559 return PTR_ERR(head);
560
561 /*
562 * At this point we know that the sysctl was not unregistered
563 * and won't be until we finish.
564 */
565 error = -EPERM;
566 if (sysctl_perm(head, table, write ? MAY_WRITE : MAY_READ))
567 goto out;
568
569 /* if that can happen at all, it should be -EINVAL, not -EISDIR */
570 error = -EINVAL;
571 if (!table->proc_handler)
572 goto out;
573
574 /* don't even try if the size is too large */
575 error = -ENOMEM;
576 if (count >= KMALLOC_MAX_SIZE)
577 goto out;
578 kbuf = kvzalloc(count + 1, GFP_KERNEL);
579 if (!kbuf)
580 goto out;
581
582 if (write) {
583 error = -EFAULT;
584 if (!copy_from_iter_full(kbuf, count, iter))
585 goto out_free_buf;
586 kbuf[count] = '\0';
587 }
588
589 error = BPF_CGROUP_RUN_PROG_SYSCTL(head, table, write, &kbuf, &count,
590 &iocb->ki_pos);
591 if (error)
592 goto out_free_buf;
593
594 /* careful: calling conventions are nasty here */
595 error = table->proc_handler(table, write, kbuf, &count, &iocb->ki_pos);
596 if (error)
597 goto out_free_buf;
598
599 if (!write) {
600 error = -EFAULT;
601 if (copy_to_iter(kbuf, count, iter) < count)
602 goto out_free_buf;
603 }
604
605 error = count;
606 out_free_buf:
607 kvfree(kbuf);
608 out:
609 sysctl_head_finish(head);
610
611 return error;
612 }
613
614 static ssize_t proc_sys_read(struct kiocb *iocb, struct iov_iter *iter)
615 {
616 return proc_sys_call_handler(iocb, iter, 0);
617 }
618
619 static ssize_t proc_sys_write(struct kiocb *iocb, struct iov_iter *iter)
620 {
621 return proc_sys_call_handler(iocb, iter, 1);
622 }
623
624 static int proc_sys_open(struct inode *inode, struct file *filp)
625 {
626 struct ctl_table_header *head = grab_header(inode);
627 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
628
629 /* sysctl was unregistered */
630 if (IS_ERR(head))
631 return PTR_ERR(head);
632
633 if (table->poll)
634 filp->private_data = proc_sys_poll_event(table->poll);
635
636 sysctl_head_finish(head);
637
638 return 0;
639 }
640
641 static __poll_t proc_sys_poll(struct file *filp, poll_table *wait)
642 {
643 struct inode *inode = file_inode(filp);
644 struct ctl_table_header *head = grab_header(inode);
645 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
646 __poll_t ret = DEFAULT_POLLMASK;
647 unsigned long event;
648
649 /* sysctl was unregistered */
650 if (IS_ERR(head))
651 return EPOLLERR | EPOLLHUP;
652
653 if (!table->proc_handler)
654 goto out;
655
656 if (!table->poll)
657 goto out;
658
659 event = (unsigned long)filp->private_data;
660 poll_wait(filp, &table->poll->wait, wait);
661
662 if (event != atomic_read(&table->poll->event)) {
663 filp->private_data = proc_sys_poll_event(table->poll);
664 ret = EPOLLIN | EPOLLRDNORM | EPOLLERR | EPOLLPRI;
665 }
666
667 out:
668 sysctl_head_finish(head);
669
670 return ret;
671 }
672
673 static bool proc_sys_fill_cache(struct file *file,
674 struct dir_context *ctx,
675 struct ctl_table_header *head,
676 struct ctl_table *table)
677 {
678 struct dentry *child, *dir = file->f_path.dentry;
679 struct inode *inode;
680 struct qstr qname;
681 ino_t ino = 0;
682 unsigned type = DT_UNKNOWN;
683
684 qname.name = table->procname;
685 qname.len = strlen(table->procname);
686 qname.hash = full_name_hash(dir, qname.name, qname.len);
687
688 child = d_lookup(dir, &qname);
689 if (!child) {
690 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
691 child = d_alloc_parallel(dir, &qname, &wq);
692 if (IS_ERR(child))
693 return false;
694 if (d_in_lookup(child)) {
695 struct dentry *res;
696 d_set_d_op(child, &proc_sys_dentry_operations);
697 inode = proc_sys_make_inode(dir->d_sb, head, table);
698 res = d_splice_alias(inode, child);
699 d_lookup_done(child);
700 if (unlikely(res)) {
701 if (IS_ERR(res)) {
702 dput(child);
703 return false;
704 }
705 dput(child);
706 child = res;
707 }
708 }
709 }
710 inode = d_inode(child);
711 ino = inode->i_ino;
712 type = inode->i_mode >> 12;
713 dput(child);
714 return dir_emit(ctx, qname.name, qname.len, ino, type);
715 }
716
717 static bool proc_sys_link_fill_cache(struct file *file,
718 struct dir_context *ctx,
719 struct ctl_table_header *head,
720 struct ctl_table *table)
721 {
722 bool ret = true;
723
724 head = sysctl_head_grab(head);
725 if (IS_ERR(head))
726 return false;
727
728 /* It is not an error if we can not follow the link ignore it */
729 if (sysctl_follow_link(&head, &table))
730 goto out;
731
732 ret = proc_sys_fill_cache(file, ctx, head, table);
733 out:
734 sysctl_head_finish(head);
735 return ret;
736 }
737
738 static int scan(struct ctl_table_header *head, struct ctl_table *table,
739 unsigned long *pos, struct file *file,
740 struct dir_context *ctx)
741 {
742 bool res;
743
744 if ((*pos)++ < ctx->pos)
745 return true;
746
747 if (unlikely(S_ISLNK(table->mode)))
748 res = proc_sys_link_fill_cache(file, ctx, head, table);
749 else
750 res = proc_sys_fill_cache(file, ctx, head, table);
751
752 if (res)
753 ctx->pos = *pos;
754
755 return res;
756 }
757
758 static int proc_sys_readdir(struct file *file, struct dir_context *ctx)
759 {
760 struct ctl_table_header *head = grab_header(file_inode(file));
761 struct ctl_table_header *h = NULL;
762 struct ctl_table *entry;
763 struct ctl_dir *ctl_dir;
764 unsigned long pos;
765
766 if (IS_ERR(head))
767 return PTR_ERR(head);
768
769 ctl_dir = container_of(head, struct ctl_dir, header);
770
771 if (!dir_emit_dots(file, ctx))
772 goto out;
773
774 pos = 2;
775
776 for (first_entry(ctl_dir, &h, &entry); h; next_entry(&h, &entry)) {
777 if (!scan(h, entry, &pos, file, ctx)) {
778 sysctl_head_finish(h);
779 break;
780 }
781 }
782 out:
783 sysctl_head_finish(head);
784 return 0;
785 }
786
787 static int proc_sys_permission(struct mnt_idmap *idmap,
788 struct inode *inode, int mask)
789 {
790 /*
791 * sysctl entries that are not writeable,
792 * are _NOT_ writeable, capabilities or not.
793 */
794 struct ctl_table_header *head;
795 struct ctl_table *table;
796 int error;
797
798 /* Executable files are not allowed under /proc/sys/ */
799 if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode))
800 return -EACCES;
801
802 head = grab_header(inode);
803 if (IS_ERR(head))
804 return PTR_ERR(head);
805
806 table = PROC_I(inode)->sysctl_entry;
807 if (!table) /* global root - r-xr-xr-x */
808 error = mask & MAY_WRITE ? -EACCES : 0;
809 else /* Use the permissions on the sysctl table entry */
810 error = sysctl_perm(head, table, mask & ~MAY_NOT_BLOCK);
811
812 sysctl_head_finish(head);
813 return error;
814 }
815
816 static int proc_sys_setattr(struct mnt_idmap *idmap,
817 struct dentry *dentry, struct iattr *attr)
818 {
819 struct inode *inode = d_inode(dentry);
820 int error;
821
822 if (attr->ia_valid & (ATTR_MODE | ATTR_UID | ATTR_GID))
823 return -EPERM;
824
825 error = setattr_prepare(&nop_mnt_idmap, dentry, attr);
826 if (error)
827 return error;
828
829 setattr_copy(&nop_mnt_idmap, inode, attr);
830 return 0;
831 }
832
833 static int proc_sys_getattr(struct mnt_idmap *idmap,
834 const struct path *path, struct kstat *stat,
835 u32 request_mask, unsigned int query_flags)
836 {
837 struct inode *inode = d_inode(path->dentry);
838 struct ctl_table_header *head = grab_header(inode);
839 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
840
841 if (IS_ERR(head))
842 return PTR_ERR(head);
843
844 generic_fillattr(&nop_mnt_idmap, request_mask, inode, stat);
845 if (table)
846 stat->mode = (stat->mode & S_IFMT) | table->mode;
847
848 sysctl_head_finish(head);
849 return 0;
850 }
851
852 static const struct file_operations proc_sys_file_operations = {
853 .open = proc_sys_open,
854 .poll = proc_sys_poll,
855 .read_iter = proc_sys_read,
856 .write_iter = proc_sys_write,
857 .splice_read = copy_splice_read,
858 .splice_write = iter_file_splice_write,
859 .llseek = default_llseek,
860 };
861
862 static const struct file_operations proc_sys_dir_file_operations = {
863 .read = generic_read_dir,
864 .iterate_shared = proc_sys_readdir,
865 .llseek = generic_file_llseek,
866 };
867
868 static const struct inode_operations proc_sys_inode_operations = {
869 .permission = proc_sys_permission,
870 .setattr = proc_sys_setattr,
871 .getattr = proc_sys_getattr,
872 };
873
874 static const struct inode_operations proc_sys_dir_operations = {
875 .lookup = proc_sys_lookup,
876 .permission = proc_sys_permission,
877 .setattr = proc_sys_setattr,
878 .getattr = proc_sys_getattr,
879 };
880
881 static int proc_sys_revalidate(struct dentry *dentry, unsigned int flags)
882 {
883 if (flags & LOOKUP_RCU)
884 return -ECHILD;
885 return !PROC_I(d_inode(dentry))->sysctl->unregistering;
886 }
887
888 static int proc_sys_delete(const struct dentry *dentry)
889 {
890 return !!PROC_I(d_inode(dentry))->sysctl->unregistering;
891 }
892
893 static int sysctl_is_seen(struct ctl_table_header *p)
894 {
895 struct ctl_table_set *set = p->set;
896 int res;
897 spin_lock(&sysctl_lock);
898 if (p->unregistering)
899 res = 0;
900 else if (!set->is_seen)
901 res = 1;
902 else
903 res = set->is_seen(set);
904 spin_unlock(&sysctl_lock);
905 return res;
906 }
907
908 static int proc_sys_compare(const struct dentry *dentry,
909 unsigned int len, const char *str, const struct qstr *name)
910 {
911 struct ctl_table_header *head;
912 struct inode *inode;
913
914 /* Although proc doesn't have negative dentries, rcu-walk means
915 * that inode here can be NULL */
916 /* AV: can it, indeed? */
917 inode = d_inode_rcu(dentry);
918 if (!inode)
919 return 1;
920 if (name->len != len)
921 return 1;
922 if (memcmp(name->name, str, len))
923 return 1;
924 head = rcu_dereference(PROC_I(inode)->sysctl);
925 return !head || !sysctl_is_seen(head);
926 }
927
928 static const struct dentry_operations proc_sys_dentry_operations = {
929 .d_revalidate = proc_sys_revalidate,
930 .d_delete = proc_sys_delete,
931 .d_compare = proc_sys_compare,
932 };
933
934 static struct ctl_dir *find_subdir(struct ctl_dir *dir,
935 const char *name, int namelen)
936 {
937 struct ctl_table_header *head;
938 struct ctl_table *entry;
939
940 entry = find_entry(&head, dir, name, namelen);
941 if (!entry)
942 return ERR_PTR(-ENOENT);
943 if (!S_ISDIR(entry->mode))
944 return ERR_PTR(-ENOTDIR);
945 return container_of(head, struct ctl_dir, header);
946 }
947
948 static struct ctl_dir *new_dir(struct ctl_table_set *set,
949 const char *name, int namelen)
950 {
951 struct ctl_table *table;
952 struct ctl_dir *new;
953 struct ctl_node *node;
954 char *new_name;
955
956 new = kzalloc(sizeof(*new) + sizeof(struct ctl_node) +
957 sizeof(struct ctl_table) + namelen + 1,
958 GFP_KERNEL);
959 if (!new)
960 return NULL;
961
962 node = (struct ctl_node *)(new + 1);
963 table = (struct ctl_table *)(node + 1);
964 new_name = (char *)(table + 1);
965 memcpy(new_name, name, namelen);
966 table[0].procname = new_name;
967 table[0].mode = S_IFDIR|S_IRUGO|S_IXUGO;
968 init_header(&new->header, set->dir.header.root, set, node, table, 1);
969
970 return new;
971 }
972
973 /**
974 * get_subdir - find or create a subdir with the specified name.
975 * @dir: Directory to create the subdirectory in
976 * @name: The name of the subdirectory to find or create
977 * @namelen: The length of name
978 *
979 * Takes a directory with an elevated reference count so we know that
980 * if we drop the lock the directory will not go away. Upon success
981 * the reference is moved from @dir to the returned subdirectory.
982 * Upon error an error code is returned and the reference on @dir is
983 * simply dropped.
984 */
985 static struct ctl_dir *get_subdir(struct ctl_dir *dir,
986 const char *name, int namelen)
987 {
988 struct ctl_table_set *set = dir->header.set;
989 struct ctl_dir *subdir, *new = NULL;
990 int err;
991
992 spin_lock(&sysctl_lock);
993 subdir = find_subdir(dir, name, namelen);
994 if (!IS_ERR(subdir))
995 goto found;
996 if (PTR_ERR(subdir) != -ENOENT)
997 goto failed;
998
999 spin_unlock(&sysctl_lock);
1000 new = new_dir(set, name, namelen);
1001 spin_lock(&sysctl_lock);
1002 subdir = ERR_PTR(-ENOMEM);
1003 if (!new)
1004 goto failed;
1005
1006 /* Was the subdir added while we dropped the lock? */
1007 subdir = find_subdir(dir, name, namelen);
1008 if (!IS_ERR(subdir))
1009 goto found;
1010 if (PTR_ERR(subdir) != -ENOENT)
1011 goto failed;
1012
1013 /* Nope. Use the our freshly made directory entry. */
1014 err = insert_header(dir, &new->header);
1015 subdir = ERR_PTR(err);
1016 if (err)
1017 goto failed;
1018 subdir = new;
1019 found:
1020 subdir->header.nreg++;
1021 failed:
1022 if (IS_ERR(subdir)) {
1023 pr_err("sysctl could not get directory: ");
1024 sysctl_print_dir(dir);
1025 pr_cont("%*.*s %ld\n", namelen, namelen, name,
1026 PTR_ERR(subdir));
1027 }
1028 drop_sysctl_table(&dir->header);
1029 if (new)
1030 drop_sysctl_table(&new->header);
1031 spin_unlock(&sysctl_lock);
1032 return subdir;
1033 }
1034
1035 static struct ctl_dir *xlate_dir(struct ctl_table_set *set, struct ctl_dir *dir)
1036 {
1037 struct ctl_dir *parent;
1038 const char *procname;
1039 if (!dir->header.parent)
1040 return &set->dir;
1041 parent = xlate_dir(set, dir->header.parent);
1042 if (IS_ERR(parent))
1043 return parent;
1044 procname = dir->header.ctl_table[0].procname;
1045 return find_subdir(parent, procname, strlen(procname));
1046 }
1047
1048 static int sysctl_follow_link(struct ctl_table_header **phead,
1049 struct ctl_table **pentry)
1050 {
1051 struct ctl_table_header *head;
1052 struct ctl_table_root *root;
1053 struct ctl_table_set *set;
1054 struct ctl_table *entry;
1055 struct ctl_dir *dir;
1056 int ret;
1057
1058 spin_lock(&sysctl_lock);
1059 root = (*pentry)->data;
1060 set = lookup_header_set(root);
1061 dir = xlate_dir(set, (*phead)->parent);
1062 if (IS_ERR(dir))
1063 ret = PTR_ERR(dir);
1064 else {
1065 const char *procname = (*pentry)->procname;
1066 head = NULL;
1067 entry = find_entry(&head, dir, procname, strlen(procname));
1068 ret = -ENOENT;
1069 if (entry && use_table(head)) {
1070 unuse_table(*phead);
1071 *phead = head;
1072 *pentry = entry;
1073 ret = 0;
1074 }
1075 }
1076
1077 spin_unlock(&sysctl_lock);
1078 return ret;
1079 }
1080
1081 static int sysctl_err(const char *path, struct ctl_table *table, char *fmt, ...)
1082 {
1083 struct va_format vaf;
1084 va_list args;
1085
1086 va_start(args, fmt);
1087 vaf.fmt = fmt;
1088 vaf.va = &args;
1089
1090 pr_err("sysctl table check failed: %s/%s %pV\n",
1091 path, table->procname, &vaf);
1092
1093 va_end(args);
1094 return -EINVAL;
1095 }
1096
1097 static int sysctl_check_table_array(const char *path, struct ctl_table *table)
1098 {
1099 unsigned int extra;
1100 int err = 0;
1101
1102 if ((table->proc_handler == proc_douintvec) ||
1103 (table->proc_handler == proc_douintvec_minmax)) {
1104 if (table->maxlen != sizeof(unsigned int))
1105 err |= sysctl_err(path, table, "array not allowed");
1106 }
1107
1108 if (table->proc_handler == proc_dou8vec_minmax) {
1109 if (table->maxlen != sizeof(u8))
1110 err |= sysctl_err(path, table, "array not allowed");
1111
1112 if (table->extra1) {
1113 extra = *(unsigned int *) table->extra1;
1114 if (extra > 255U)
1115 err |= sysctl_err(path, table,
1116 "range value too large for proc_dou8vec_minmax");
1117 }
1118 if (table->extra2) {
1119 extra = *(unsigned int *) table->extra2;
1120 if (extra > 255U)
1121 err |= sysctl_err(path, table,
1122 "range value too large for proc_dou8vec_minmax");
1123 }
1124 }
1125
1126 if (table->proc_handler == proc_dobool) {
1127 if (table->maxlen != sizeof(bool))
1128 err |= sysctl_err(path, table, "array not allowed");
1129 }
1130
1131 return err;
1132 }
1133
1134 static int sysctl_check_table(const char *path, struct ctl_table_header *header)
1135 {
1136 struct ctl_table *entry;
1137 int err = 0;
1138 list_for_each_table_entry(entry, header) {
1139 if (!entry->procname)
1140 err |= sysctl_err(path, entry, "procname is null");
1141 if ((entry->proc_handler == proc_dostring) ||
1142 (entry->proc_handler == proc_dobool) ||
1143 (entry->proc_handler == proc_dointvec) ||
1144 (entry->proc_handler == proc_douintvec) ||
1145 (entry->proc_handler == proc_douintvec_minmax) ||
1146 (entry->proc_handler == proc_dointvec_minmax) ||
1147 (entry->proc_handler == proc_dou8vec_minmax) ||
1148 (entry->proc_handler == proc_dointvec_jiffies) ||
1149 (entry->proc_handler == proc_dointvec_userhz_jiffies) ||
1150 (entry->proc_handler == proc_dointvec_ms_jiffies) ||
1151 (entry->proc_handler == proc_doulongvec_minmax) ||
1152 (entry->proc_handler == proc_doulongvec_ms_jiffies_minmax)) {
1153 if (!entry->data)
1154 err |= sysctl_err(path, entry, "No data");
1155 if (!entry->maxlen)
1156 err |= sysctl_err(path, entry, "No maxlen");
1157 else
1158 err |= sysctl_check_table_array(path, entry);
1159 }
1160 if (!entry->proc_handler)
1161 err |= sysctl_err(path, entry, "No proc_handler");
1162
1163 if ((entry->mode & (S_IRUGO|S_IWUGO)) != entry->mode)
1164 err |= sysctl_err(path, entry, "bogus .mode 0%o",
1165 entry->mode);
1166 }
1167 return err;
1168 }
1169
1170 static struct ctl_table_header *new_links(struct ctl_dir *dir, struct ctl_table_header *head)
1171 {
1172 struct ctl_table *link_table, *entry, *link;
1173 struct ctl_table_header *links;
1174 struct ctl_node *node;
1175 char *link_name;
1176 int name_bytes;
1177
1178 name_bytes = 0;
1179 list_for_each_table_entry(entry, head) {
1180 name_bytes += strlen(entry->procname) + 1;
1181 }
1182
1183 links = kzalloc(sizeof(struct ctl_table_header) +
1184 sizeof(struct ctl_node)*head->ctl_table_size +
1185 sizeof(struct ctl_table)*head->ctl_table_size +
1186 name_bytes,
1187 GFP_KERNEL);
1188
1189 if (!links)
1190 return NULL;
1191
1192 node = (struct ctl_node *)(links + 1);
1193 link_table = (struct ctl_table *)(node + head->ctl_table_size);
1194 link_name = (char *)(link_table + head->ctl_table_size);
1195 link = link_table;
1196
1197 list_for_each_table_entry(entry, head) {
1198 int len = strlen(entry->procname) + 1;
1199 memcpy(link_name, entry->procname, len);
1200 link->procname = link_name;
1201 link->mode = S_IFLNK|S_IRWXUGO;
1202 link->data = head->root;
1203 link_name += len;
1204 link++;
1205 }
1206 init_header(links, dir->header.root, dir->header.set, node, link_table,
1207 head->ctl_table_size);
1208 links->nreg = head->ctl_table_size;
1209
1210 return links;
1211 }
1212
1213 static bool get_links(struct ctl_dir *dir,
1214 struct ctl_table_header *header,
1215 struct ctl_table_root *link_root)
1216 {
1217 struct ctl_table_header *tmp_head;
1218 struct ctl_table *entry, *link;
1219
1220 if (header->ctl_table_size == 0 ||
1221 sysctl_is_perm_empty_ctl_header(header))
1222 return true;
1223
1224 /* Are there links available for every entry in table? */
1225 list_for_each_table_entry(entry, header) {
1226 const char *procname = entry->procname;
1227 link = find_entry(&tmp_head, dir, procname, strlen(procname));
1228 if (!link)
1229 return false;
1230 if (S_ISDIR(link->mode) && S_ISDIR(entry->mode))
1231 continue;
1232 if (S_ISLNK(link->mode) && (link->data == link_root))
1233 continue;
1234 return false;
1235 }
1236
1237 /* The checks passed. Increase the registration count on the links */
1238 list_for_each_table_entry(entry, header) {
1239 const char *procname = entry->procname;
1240 link = find_entry(&tmp_head, dir, procname, strlen(procname));
1241 tmp_head->nreg++;
1242 }
1243 return true;
1244 }
1245
1246 static int insert_links(struct ctl_table_header *head)
1247 {
1248 struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1249 struct ctl_dir *core_parent;
1250 struct ctl_table_header *links;
1251 int err;
1252
1253 if (head->set == root_set)
1254 return 0;
1255
1256 core_parent = xlate_dir(root_set, head->parent);
1257 if (IS_ERR(core_parent))
1258 return 0;
1259
1260 if (get_links(core_parent, head, head->root))
1261 return 0;
1262
1263 core_parent->header.nreg++;
1264 spin_unlock(&sysctl_lock);
1265
1266 links = new_links(core_parent, head);
1267
1268 spin_lock(&sysctl_lock);
1269 err = -ENOMEM;
1270 if (!links)
1271 goto out;
1272
1273 err = 0;
1274 if (get_links(core_parent, head, head->root)) {
1275 kfree(links);
1276 goto out;
1277 }
1278
1279 err = insert_header(core_parent, links);
1280 if (err)
1281 kfree(links);
1282 out:
1283 drop_sysctl_table(&core_parent->header);
1284 return err;
1285 }
1286
1287 /* Find the directory for the ctl_table. If one is not found create it. */
1288 static struct ctl_dir *sysctl_mkdir_p(struct ctl_dir *dir, const char *path)
1289 {
1290 const char *name, *nextname;
1291
1292 for (name = path; name; name = nextname) {
1293 int namelen;
1294 nextname = strchr(name, '/');
1295 if (nextname) {
1296 namelen = nextname - name;
1297 nextname++;
1298 } else {
1299 namelen = strlen(name);
1300 }
1301 if (namelen == 0)
1302 continue;
1303
1304 /*
1305 * namelen ensures if name is "foo/bar/yay" only foo is
1306 * registered first. We traverse as if using mkdir -p and
1307 * return a ctl_dir for the last directory entry.
1308 */
1309 dir = get_subdir(dir, name, namelen);
1310 if (IS_ERR(dir))
1311 break;
1312 }
1313 return dir;
1314 }
1315
1316 /**
1317 * __register_sysctl_table - register a leaf sysctl table
1318 * @set: Sysctl tree to register on
1319 * @path: The path to the directory the sysctl table is in.
1320 *
1321 * @table: the top-level table structure. This table should not be free'd
1322 * after registration. So it should not be used on stack. It can either
1323 * be a global or dynamically allocated by the caller and free'd later
1324 * after sysctl unregistration.
1325 * @table_size : The number of elements in table
1326 *
1327 * Register a sysctl table hierarchy. @table should be a filled in ctl_table
1328 * array.
1329 *
1330 * The members of the &struct ctl_table structure are used as follows:
1331 * procname - the name of the sysctl file under /proc/sys. Set to %NULL to not
1332 * enter a sysctl file
1333 * data - a pointer to data for use by proc_handler
1334 * maxlen - the maximum size in bytes of the data
1335 * mode - the file permissions for the /proc/sys file
1336 * type - Defines the target type (described in struct definition)
1337 * proc_handler - the text handler routine (described below)
1338 *
1339 * extra1, extra2 - extra pointers usable by the proc handler routines
1340 * XXX: we should eventually modify these to use long min / max [0]
1341 * [0] https://lkml.kernel.org/87zgpte9o4.fsf@email.froward.int.ebiederm.org
1342 *
1343 * Leaf nodes in the sysctl tree will be represented by a single file
1344 * under /proc; non-leaf nodes are not allowed.
1345 *
1346 * There must be a proc_handler routine for any terminal nodes.
1347 * Several default handlers are available to cover common cases -
1348 *
1349 * proc_dostring(), proc_dointvec(), proc_dointvec_jiffies(),
1350 * proc_dointvec_userhz_jiffies(), proc_dointvec_minmax(),
1351 * proc_doulongvec_ms_jiffies_minmax(), proc_doulongvec_minmax()
1352 *
1353 * It is the handler's job to read the input buffer from user memory
1354 * and process it. The handler should return 0 on success.
1355 *
1356 * This routine returns %NULL on a failure to register, and a pointer
1357 * to the table header on success.
1358 */
1359 struct ctl_table_header *__register_sysctl_table(
1360 struct ctl_table_set *set,
1361 const char *path, struct ctl_table *table, size_t table_size)
1362 {
1363 struct ctl_table_root *root = set->dir.header.root;
1364 struct ctl_table_header *header;
1365 struct ctl_dir *dir;
1366 struct ctl_node *node;
1367
1368 header = kzalloc(sizeof(struct ctl_table_header) +
1369 sizeof(struct ctl_node)*table_size, GFP_KERNEL_ACCOUNT);
1370 if (!header)
1371 return NULL;
1372
1373 node = (struct ctl_node *)(header + 1);
1374 init_header(header, root, set, node, table, table_size);
1375 if (sysctl_check_table(path, header))
1376 goto fail;
1377
1378 spin_lock(&sysctl_lock);
1379 dir = &set->dir;
1380 /* Reference moved down the directory tree get_subdir */
1381 dir->header.nreg++;
1382 spin_unlock(&sysctl_lock);
1383
1384 dir = sysctl_mkdir_p(dir, path);
1385 if (IS_ERR(dir))
1386 goto fail;
1387 spin_lock(&sysctl_lock);
1388 if (insert_header(dir, header))
1389 goto fail_put_dir_locked;
1390
1391 drop_sysctl_table(&dir->header);
1392 spin_unlock(&sysctl_lock);
1393
1394 return header;
1395
1396 fail_put_dir_locked:
1397 drop_sysctl_table(&dir->header);
1398 spin_unlock(&sysctl_lock);
1399 fail:
1400 kfree(header);
1401 return NULL;
1402 }
1403
1404 /**
1405 * register_sysctl_sz - register a sysctl table
1406 * @path: The path to the directory the sysctl table is in. If the path
1407 * doesn't exist we will create it for you.
1408 * @table: the table structure. The calller must ensure the life of the @table
1409 * will be kept during the lifetime use of the syctl. It must not be freed
1410 * until unregister_sysctl_table() is called with the given returned table
1411 * with this registration. If your code is non modular then you don't need
1412 * to call unregister_sysctl_table() and can instead use something like
1413 * register_sysctl_init() which does not care for the result of the syctl
1414 * registration.
1415 * @table_size: The number of elements in table.
1416 *
1417 * Register a sysctl table. @table should be a filled in ctl_table
1418 * array. A completely 0 filled entry terminates the table.
1419 *
1420 * See __register_sysctl_table for more details.
1421 */
1422 struct ctl_table_header *register_sysctl_sz(const char *path, struct ctl_table *table,
1423 size_t table_size)
1424 {
1425 return __register_sysctl_table(&sysctl_table_root.default_set,
1426 path, table, table_size);
1427 }
1428 EXPORT_SYMBOL(register_sysctl_sz);
1429
1430 /**
1431 * __register_sysctl_init() - register sysctl table to path
1432 * @path: path name for sysctl base. If that path doesn't exist we will create
1433 * it for you.
1434 * @table: This is the sysctl table that needs to be registered to the path.
1435 * The caller must ensure the life of the @table will be kept during the
1436 * lifetime use of the sysctl.
1437 * @table_name: The name of sysctl table, only used for log printing when
1438 * registration fails
1439 * @table_size: The number of elements in table
1440 *
1441 * The sysctl interface is used by userspace to query or modify at runtime
1442 * a predefined value set on a variable. These variables however have default
1443 * values pre-set. Code which depends on these variables will always work even
1444 * if register_sysctl() fails. If register_sysctl() fails you'd just loose the
1445 * ability to query or modify the sysctls dynamically at run time. Chances of
1446 * register_sysctl() failing on init are extremely low, and so for both reasons
1447 * this function does not return any error as it is used by initialization code.
1448 *
1449 * Context: if your base directory does not exist it will be created for you.
1450 */
1451 void __init __register_sysctl_init(const char *path, struct ctl_table *table,
1452 const char *table_name, size_t table_size)
1453 {
1454 struct ctl_table_header *hdr = register_sysctl_sz(path, table, table_size);
1455
1456 if (unlikely(!hdr)) {
1457 pr_err("failed when register_sysctl_sz %s to %s\n", table_name, path);
1458 return;
1459 }
1460 kmemleak_not_leak(hdr);
1461 }
1462
1463 static void put_links(struct ctl_table_header *header)
1464 {
1465 struct ctl_table_set *root_set = &sysctl_table_root.default_set;
1466 struct ctl_table_root *root = header->root;
1467 struct ctl_dir *parent = header->parent;
1468 struct ctl_dir *core_parent;
1469 struct ctl_table *entry;
1470
1471 if (header->set == root_set)
1472 return;
1473
1474 core_parent = xlate_dir(root_set, parent);
1475 if (IS_ERR(core_parent))
1476 return;
1477
1478 list_for_each_table_entry(entry, header) {
1479 struct ctl_table_header *link_head;
1480 struct ctl_table *link;
1481 const char *name = entry->procname;
1482
1483 link = find_entry(&link_head, core_parent, name, strlen(name));
1484 if (link &&
1485 ((S_ISDIR(link->mode) && S_ISDIR(entry->mode)) ||
1486 (S_ISLNK(link->mode) && (link->data == root)))) {
1487 drop_sysctl_table(link_head);
1488 }
1489 else {
1490 pr_err("sysctl link missing during unregister: ");
1491 sysctl_print_dir(parent);
1492 pr_cont("%s\n", name);
1493 }
1494 }
1495 }
1496
1497 static void drop_sysctl_table(struct ctl_table_header *header)
1498 {
1499 struct ctl_dir *parent = header->parent;
1500
1501 if (--header->nreg)
1502 return;
1503
1504 if (parent) {
1505 put_links(header);
1506 start_unregistering(header);
1507 }
1508
1509 if (!--header->count)
1510 kfree_rcu(header, rcu);
1511
1512 if (parent)
1513 drop_sysctl_table(&parent->header);
1514 }
1515
1516 /**
1517 * unregister_sysctl_table - unregister a sysctl table hierarchy
1518 * @header: the header returned from register_sysctl or __register_sysctl_table
1519 *
1520 * Unregisters the sysctl table and all children. proc entries may not
1521 * actually be removed until they are no longer used by anyone.
1522 */
1523 void unregister_sysctl_table(struct ctl_table_header * header)
1524 {
1525 might_sleep();
1526
1527 if (header == NULL)
1528 return;
1529
1530 spin_lock(&sysctl_lock);
1531 drop_sysctl_table(header);
1532 spin_unlock(&sysctl_lock);
1533 }
1534 EXPORT_SYMBOL(unregister_sysctl_table);
1535
1536 void setup_sysctl_set(struct ctl_table_set *set,
1537 struct ctl_table_root *root,
1538 int (*is_seen)(struct ctl_table_set *))
1539 {
1540 memset(set, 0, sizeof(*set));
1541 set->is_seen = is_seen;
1542 init_header(&set->dir.header, root, set, NULL, root_table, 1);
1543 }
1544
1545 void retire_sysctl_set(struct ctl_table_set *set)
1546 {
1547 WARN_ON(!RB_EMPTY_ROOT(&set->dir.root));
1548 }
1549
1550 int __init proc_sys_init(void)
1551 {
1552 struct proc_dir_entry *proc_sys_root;
1553
1554 proc_sys_root = proc_mkdir("sys", NULL);
1555 proc_sys_root->proc_iops = &proc_sys_dir_operations;
1556 proc_sys_root->proc_dir_ops = &proc_sys_dir_file_operations;
1557 proc_sys_root->nlink = 0;
1558
1559 return sysctl_init_bases();
1560 }
1561
1562 struct sysctl_alias {
1563 const char *kernel_param;
1564 const char *sysctl_param;
1565 };
1566
1567 /*
1568 * Historically some settings had both sysctl and a command line parameter.
1569 * With the generic sysctl. parameter support, we can handle them at a single
1570 * place and only keep the historical name for compatibility. This is not meant
1571 * to add brand new aliases. When adding existing aliases, consider whether
1572 * the possibly different moment of changing the value (e.g. from early_param
1573 * to the moment do_sysctl_args() is called) is an issue for the specific
1574 * parameter.
1575 */
1576 static const struct sysctl_alias sysctl_aliases[] = {
1577 {"hardlockup_all_cpu_backtrace", "kernel.hardlockup_all_cpu_backtrace" },
1578 {"hung_task_panic", "kernel.hung_task_panic" },
1579 {"numa_zonelist_order", "vm.numa_zonelist_order" },
1580 {"softlockup_all_cpu_backtrace", "kernel.softlockup_all_cpu_backtrace" },
1581 { }
1582 };
1583
1584 static const char *sysctl_find_alias(char *param)
1585 {
1586 const struct sysctl_alias *alias;
1587
1588 for (alias = &sysctl_aliases[0]; alias->kernel_param != NULL; alias++) {
1589 if (strcmp(alias->kernel_param, param) == 0)
1590 return alias->sysctl_param;
1591 }
1592
1593 return NULL;
1594 }
1595
1596 bool sysctl_is_alias(char *param)
1597 {
1598 const char *alias = sysctl_find_alias(param);
1599
1600 return alias != NULL;
1601 }
1602
1603 /* Set sysctl value passed on kernel command line. */
1604 static int process_sysctl_arg(char *param, char *val,
1605 const char *unused, void *arg)
1606 {
1607 char *path;
1608 struct vfsmount **proc_mnt = arg;
1609 struct file_system_type *proc_fs_type;
1610 struct file *file;
1611 int len;
1612 int err;
1613 loff_t pos = 0;
1614 ssize_t wret;
1615
1616 if (strncmp(param, "sysctl", sizeof("sysctl") - 1) == 0) {
1617 param += sizeof("sysctl") - 1;
1618
1619 if (param[0] != '/' && param[0] != '.')
1620 return 0;
1621
1622 param++;
1623 } else {
1624 param = (char *) sysctl_find_alias(param);
1625 if (!param)
1626 return 0;
1627 }
1628
1629 if (!val)
1630 return -EINVAL;
1631 len = strlen(val);
1632 if (len == 0)
1633 return -EINVAL;
1634
1635 /*
1636 * To set sysctl options, we use a temporary mount of proc, look up the
1637 * respective sys/ file and write to it. To avoid mounting it when no
1638 * options were given, we mount it only when the first sysctl option is
1639 * found. Why not a persistent mount? There are problems with a
1640 * persistent mount of proc in that it forces userspace not to use any
1641 * proc mount options.
1642 */
1643 if (!*proc_mnt) {
1644 proc_fs_type = get_fs_type("proc");
1645 if (!proc_fs_type) {
1646 pr_err("Failed to find procfs to set sysctl from command line\n");
1647 return 0;
1648 }
1649 *proc_mnt = kern_mount(proc_fs_type);
1650 put_filesystem(proc_fs_type);
1651 if (IS_ERR(*proc_mnt)) {
1652 pr_err("Failed to mount procfs to set sysctl from command line\n");
1653 return 0;
1654 }
1655 }
1656
1657 path = kasprintf(GFP_KERNEL, "sys/%s", param);
1658 if (!path)
1659 panic("%s: Failed to allocate path for %s\n", __func__, param);
1660 strreplace(path, '.', '/');
1661
1662 file = file_open_root_mnt(*proc_mnt, path, O_WRONLY, 0);
1663 if (IS_ERR(file)) {
1664 err = PTR_ERR(file);
1665 if (err == -ENOENT)
1666 pr_err("Failed to set sysctl parameter '%s=%s': parameter not found\n",
1667 param, val);
1668 else if (err == -EACCES)
1669 pr_err("Failed to set sysctl parameter '%s=%s': permission denied (read-only?)\n",
1670 param, val);
1671 else
1672 pr_err("Error %pe opening proc file to set sysctl parameter '%s=%s'\n",
1673 file, param, val);
1674 goto out;
1675 }
1676 wret = kernel_write(file, val, len, &pos);
1677 if (wret < 0) {
1678 err = wret;
1679 if (err == -EINVAL)
1680 pr_err("Failed to set sysctl parameter '%s=%s': invalid value\n",
1681 param, val);
1682 else
1683 pr_err("Error %pe writing to proc file to set sysctl parameter '%s=%s'\n",
1684 ERR_PTR(err), param, val);
1685 } else if (wret != len) {
1686 pr_err("Wrote only %zd bytes of %d writing to proc file %s to set sysctl parameter '%s=%s\n",
1687 wret, len, path, param, val);
1688 }
1689
1690 err = filp_close(file, NULL);
1691 if (err)
1692 pr_err("Error %pe closing proc file to set sysctl parameter '%s=%s\n",
1693 ERR_PTR(err), param, val);
1694 out:
1695 kfree(path);
1696 return 0;
1697 }
1698
1699 void do_sysctl_args(void)
1700 {
1701 char *command_line;
1702 struct vfsmount *proc_mnt = NULL;
1703
1704 command_line = kstrdup(saved_command_line, GFP_KERNEL);
1705 if (!command_line)
1706 panic("%s: Failed to allocate copy of command line\n", __func__);
1707
1708 parse_args("Setting sysctl args", command_line,
1709 NULL, 0, -1, -1, &proc_mnt, process_sysctl_arg);
1710
1711 if (proc_mnt)
1712 kern_unmount(proc_mnt);
1713
1714 kfree(command_line);
1715 }
1716
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.