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