xref: /openbmc/linux/net/netlink/af_netlink.c (revision c819e2cf)
1 /*
2  * NETLINK      Kernel-user communication protocol.
3  *
4  * 		Authors:	Alan Cox <alan@lxorguk.ukuu.org.uk>
5  * 				Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
6  * 				Patrick McHardy <kaber@trash.net>
7  *
8  *		This program is free software; you can redistribute it and/or
9  *		modify it under the terms of the GNU General Public License
10  *		as published by the Free Software Foundation; either version
11  *		2 of the License, or (at your option) any later version.
12  *
13  * Tue Jun 26 14:36:48 MEST 2001 Herbert "herp" Rosmanith
14  *                               added netlink_proto_exit
15  * Tue Jan 22 18:32:44 BRST 2002 Arnaldo C. de Melo <acme@conectiva.com.br>
16  * 				 use nlk_sk, as sk->protinfo is on a diet 8)
17  * Fri Jul 22 19:51:12 MEST 2005 Harald Welte <laforge@gnumonks.org>
18  * 				 - inc module use count of module that owns
19  * 				   the kernel socket in case userspace opens
20  * 				   socket of same protocol
21  * 				 - remove all module support, since netlink is
22  * 				   mandatory if CONFIG_NET=y these days
23  */
24 
25 #include <linux/module.h>
26 
27 #include <linux/capability.h>
28 #include <linux/kernel.h>
29 #include <linux/init.h>
30 #include <linux/signal.h>
31 #include <linux/sched.h>
32 #include <linux/errno.h>
33 #include <linux/string.h>
34 #include <linux/stat.h>
35 #include <linux/socket.h>
36 #include <linux/un.h>
37 #include <linux/fcntl.h>
38 #include <linux/termios.h>
39 #include <linux/sockios.h>
40 #include <linux/net.h>
41 #include <linux/fs.h>
42 #include <linux/slab.h>
43 #include <asm/uaccess.h>
44 #include <linux/skbuff.h>
45 #include <linux/netdevice.h>
46 #include <linux/rtnetlink.h>
47 #include <linux/proc_fs.h>
48 #include <linux/seq_file.h>
49 #include <linux/notifier.h>
50 #include <linux/security.h>
51 #include <linux/jhash.h>
52 #include <linux/jiffies.h>
53 #include <linux/random.h>
54 #include <linux/bitops.h>
55 #include <linux/mm.h>
56 #include <linux/types.h>
57 #include <linux/audit.h>
58 #include <linux/mutex.h>
59 #include <linux/vmalloc.h>
60 #include <linux/if_arp.h>
61 #include <linux/rhashtable.h>
62 #include <asm/cacheflush.h>
63 #include <linux/hash.h>
64 #include <linux/genetlink.h>
65 
66 #include <net/net_namespace.h>
67 #include <net/sock.h>
68 #include <net/scm.h>
69 #include <net/netlink.h>
70 
71 #include "af_netlink.h"
72 
73 struct listeners {
74 	struct rcu_head		rcu;
75 	unsigned long		masks[0];
76 };
77 
78 /* state bits */
79 #define NETLINK_CONGESTED	0x0
80 
81 /* flags */
82 #define NETLINK_KERNEL_SOCKET	0x1
83 #define NETLINK_RECV_PKTINFO	0x2
84 #define NETLINK_BROADCAST_SEND_ERROR	0x4
85 #define NETLINK_RECV_NO_ENOBUFS	0x8
86 
87 static inline int netlink_is_kernel(struct sock *sk)
88 {
89 	return nlk_sk(sk)->flags & NETLINK_KERNEL_SOCKET;
90 }
91 
92 struct netlink_table *nl_table;
93 EXPORT_SYMBOL_GPL(nl_table);
94 
95 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
96 
97 static int netlink_dump(struct sock *sk);
98 static void netlink_skb_destructor(struct sk_buff *skb);
99 
100 /* nl_table locking explained:
101  * Lookup and traversal are protected with nl_sk_hash_lock or nl_table_lock
102  * combined with an RCU read-side lock. Insertion and removal are protected
103  * with nl_sk_hash_lock while using RCU list modification primitives and may
104  * run in parallel to nl_table_lock protected lookups. Destruction of the
105  * Netlink socket may only occur *after* nl_table_lock has been acquired
106  * either during or after the socket has been removed from the list.
107  */
108 DEFINE_RWLOCK(nl_table_lock);
109 EXPORT_SYMBOL_GPL(nl_table_lock);
110 static atomic_t nl_table_users = ATOMIC_INIT(0);
111 
112 #define nl_deref_protected(X) rcu_dereference_protected(X, lockdep_is_held(&nl_table_lock));
113 
114 /* Protects netlink socket hash table mutations */
115 DEFINE_MUTEX(nl_sk_hash_lock);
116 EXPORT_SYMBOL_GPL(nl_sk_hash_lock);
117 
118 #ifdef CONFIG_PROVE_LOCKING
119 static int lockdep_nl_sk_hash_is_held(void *parent)
120 {
121 	if (debug_locks)
122 		return lockdep_is_held(&nl_sk_hash_lock) || lockdep_is_held(&nl_table_lock);
123 	return 1;
124 }
125 #endif
126 
127 static ATOMIC_NOTIFIER_HEAD(netlink_chain);
128 
129 static DEFINE_SPINLOCK(netlink_tap_lock);
130 static struct list_head netlink_tap_all __read_mostly;
131 
132 static inline u32 netlink_group_mask(u32 group)
133 {
134 	return group ? 1 << (group - 1) : 0;
135 }
136 
137 int netlink_add_tap(struct netlink_tap *nt)
138 {
139 	if (unlikely(nt->dev->type != ARPHRD_NETLINK))
140 		return -EINVAL;
141 
142 	spin_lock(&netlink_tap_lock);
143 	list_add_rcu(&nt->list, &netlink_tap_all);
144 	spin_unlock(&netlink_tap_lock);
145 
146 	__module_get(nt->module);
147 
148 	return 0;
149 }
150 EXPORT_SYMBOL_GPL(netlink_add_tap);
151 
152 static int __netlink_remove_tap(struct netlink_tap *nt)
153 {
154 	bool found = false;
155 	struct netlink_tap *tmp;
156 
157 	spin_lock(&netlink_tap_lock);
158 
159 	list_for_each_entry(tmp, &netlink_tap_all, list) {
160 		if (nt == tmp) {
161 			list_del_rcu(&nt->list);
162 			found = true;
163 			goto out;
164 		}
165 	}
166 
167 	pr_warn("__netlink_remove_tap: %p not found\n", nt);
168 out:
169 	spin_unlock(&netlink_tap_lock);
170 
171 	if (found && nt->module)
172 		module_put(nt->module);
173 
174 	return found ? 0 : -ENODEV;
175 }
176 
177 int netlink_remove_tap(struct netlink_tap *nt)
178 {
179 	int ret;
180 
181 	ret = __netlink_remove_tap(nt);
182 	synchronize_net();
183 
184 	return ret;
185 }
186 EXPORT_SYMBOL_GPL(netlink_remove_tap);
187 
188 static bool netlink_filter_tap(const struct sk_buff *skb)
189 {
190 	struct sock *sk = skb->sk;
191 
192 	/* We take the more conservative approach and
193 	 * whitelist socket protocols that may pass.
194 	 */
195 	switch (sk->sk_protocol) {
196 	case NETLINK_ROUTE:
197 	case NETLINK_USERSOCK:
198 	case NETLINK_SOCK_DIAG:
199 	case NETLINK_NFLOG:
200 	case NETLINK_XFRM:
201 	case NETLINK_FIB_LOOKUP:
202 	case NETLINK_NETFILTER:
203 	case NETLINK_GENERIC:
204 		return true;
205 	}
206 
207 	return false;
208 }
209 
210 static int __netlink_deliver_tap_skb(struct sk_buff *skb,
211 				     struct net_device *dev)
212 {
213 	struct sk_buff *nskb;
214 	struct sock *sk = skb->sk;
215 	int ret = -ENOMEM;
216 
217 	dev_hold(dev);
218 	nskb = skb_clone(skb, GFP_ATOMIC);
219 	if (nskb) {
220 		nskb->dev = dev;
221 		nskb->protocol = htons((u16) sk->sk_protocol);
222 		nskb->pkt_type = netlink_is_kernel(sk) ?
223 				 PACKET_KERNEL : PACKET_USER;
224 		skb_reset_network_header(nskb);
225 		ret = dev_queue_xmit(nskb);
226 		if (unlikely(ret > 0))
227 			ret = net_xmit_errno(ret);
228 	}
229 
230 	dev_put(dev);
231 	return ret;
232 }
233 
234 static void __netlink_deliver_tap(struct sk_buff *skb)
235 {
236 	int ret;
237 	struct netlink_tap *tmp;
238 
239 	if (!netlink_filter_tap(skb))
240 		return;
241 
242 	list_for_each_entry_rcu(tmp, &netlink_tap_all, list) {
243 		ret = __netlink_deliver_tap_skb(skb, tmp->dev);
244 		if (unlikely(ret))
245 			break;
246 	}
247 }
248 
249 static void netlink_deliver_tap(struct sk_buff *skb)
250 {
251 	rcu_read_lock();
252 
253 	if (unlikely(!list_empty(&netlink_tap_all)))
254 		__netlink_deliver_tap(skb);
255 
256 	rcu_read_unlock();
257 }
258 
259 static void netlink_deliver_tap_kernel(struct sock *dst, struct sock *src,
260 				       struct sk_buff *skb)
261 {
262 	if (!(netlink_is_kernel(dst) && netlink_is_kernel(src)))
263 		netlink_deliver_tap(skb);
264 }
265 
266 static void netlink_overrun(struct sock *sk)
267 {
268 	struct netlink_sock *nlk = nlk_sk(sk);
269 
270 	if (!(nlk->flags & NETLINK_RECV_NO_ENOBUFS)) {
271 		if (!test_and_set_bit(NETLINK_CONGESTED, &nlk_sk(sk)->state)) {
272 			sk->sk_err = ENOBUFS;
273 			sk->sk_error_report(sk);
274 		}
275 	}
276 	atomic_inc(&sk->sk_drops);
277 }
278 
279 static void netlink_rcv_wake(struct sock *sk)
280 {
281 	struct netlink_sock *nlk = nlk_sk(sk);
282 
283 	if (skb_queue_empty(&sk->sk_receive_queue))
284 		clear_bit(NETLINK_CONGESTED, &nlk->state);
285 	if (!test_bit(NETLINK_CONGESTED, &nlk->state))
286 		wake_up_interruptible(&nlk->wait);
287 }
288 
289 #ifdef CONFIG_NETLINK_MMAP
290 static bool netlink_skb_is_mmaped(const struct sk_buff *skb)
291 {
292 	return NETLINK_CB(skb).flags & NETLINK_SKB_MMAPED;
293 }
294 
295 static bool netlink_rx_is_mmaped(struct sock *sk)
296 {
297 	return nlk_sk(sk)->rx_ring.pg_vec != NULL;
298 }
299 
300 static bool netlink_tx_is_mmaped(struct sock *sk)
301 {
302 	return nlk_sk(sk)->tx_ring.pg_vec != NULL;
303 }
304 
305 static __pure struct page *pgvec_to_page(const void *addr)
306 {
307 	if (is_vmalloc_addr(addr))
308 		return vmalloc_to_page(addr);
309 	else
310 		return virt_to_page(addr);
311 }
312 
313 static void free_pg_vec(void **pg_vec, unsigned int order, unsigned int len)
314 {
315 	unsigned int i;
316 
317 	for (i = 0; i < len; i++) {
318 		if (pg_vec[i] != NULL) {
319 			if (is_vmalloc_addr(pg_vec[i]))
320 				vfree(pg_vec[i]);
321 			else
322 				free_pages((unsigned long)pg_vec[i], order);
323 		}
324 	}
325 	kfree(pg_vec);
326 }
327 
328 static void *alloc_one_pg_vec_page(unsigned long order)
329 {
330 	void *buffer;
331 	gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP | __GFP_ZERO |
332 			  __GFP_NOWARN | __GFP_NORETRY;
333 
334 	buffer = (void *)__get_free_pages(gfp_flags, order);
335 	if (buffer != NULL)
336 		return buffer;
337 
338 	buffer = vzalloc((1 << order) * PAGE_SIZE);
339 	if (buffer != NULL)
340 		return buffer;
341 
342 	gfp_flags &= ~__GFP_NORETRY;
343 	return (void *)__get_free_pages(gfp_flags, order);
344 }
345 
346 static void **alloc_pg_vec(struct netlink_sock *nlk,
347 			   struct nl_mmap_req *req, unsigned int order)
348 {
349 	unsigned int block_nr = req->nm_block_nr;
350 	unsigned int i;
351 	void **pg_vec;
352 
353 	pg_vec = kcalloc(block_nr, sizeof(void *), GFP_KERNEL);
354 	if (pg_vec == NULL)
355 		return NULL;
356 
357 	for (i = 0; i < block_nr; i++) {
358 		pg_vec[i] = alloc_one_pg_vec_page(order);
359 		if (pg_vec[i] == NULL)
360 			goto err1;
361 	}
362 
363 	return pg_vec;
364 err1:
365 	free_pg_vec(pg_vec, order, block_nr);
366 	return NULL;
367 }
368 
369 static int netlink_set_ring(struct sock *sk, struct nl_mmap_req *req,
370 			    bool closing, bool tx_ring)
371 {
372 	struct netlink_sock *nlk = nlk_sk(sk);
373 	struct netlink_ring *ring;
374 	struct sk_buff_head *queue;
375 	void **pg_vec = NULL;
376 	unsigned int order = 0;
377 	int err;
378 
379 	ring  = tx_ring ? &nlk->tx_ring : &nlk->rx_ring;
380 	queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
381 
382 	if (!closing) {
383 		if (atomic_read(&nlk->mapped))
384 			return -EBUSY;
385 		if (atomic_read(&ring->pending))
386 			return -EBUSY;
387 	}
388 
389 	if (req->nm_block_nr) {
390 		if (ring->pg_vec != NULL)
391 			return -EBUSY;
392 
393 		if ((int)req->nm_block_size <= 0)
394 			return -EINVAL;
395 		if (!PAGE_ALIGNED(req->nm_block_size))
396 			return -EINVAL;
397 		if (req->nm_frame_size < NL_MMAP_HDRLEN)
398 			return -EINVAL;
399 		if (!IS_ALIGNED(req->nm_frame_size, NL_MMAP_MSG_ALIGNMENT))
400 			return -EINVAL;
401 
402 		ring->frames_per_block = req->nm_block_size /
403 					 req->nm_frame_size;
404 		if (ring->frames_per_block == 0)
405 			return -EINVAL;
406 		if (ring->frames_per_block * req->nm_block_nr !=
407 		    req->nm_frame_nr)
408 			return -EINVAL;
409 
410 		order = get_order(req->nm_block_size);
411 		pg_vec = alloc_pg_vec(nlk, req, order);
412 		if (pg_vec == NULL)
413 			return -ENOMEM;
414 	} else {
415 		if (req->nm_frame_nr)
416 			return -EINVAL;
417 	}
418 
419 	err = -EBUSY;
420 	mutex_lock(&nlk->pg_vec_lock);
421 	if (closing || atomic_read(&nlk->mapped) == 0) {
422 		err = 0;
423 		spin_lock_bh(&queue->lock);
424 
425 		ring->frame_max		= req->nm_frame_nr - 1;
426 		ring->head		= 0;
427 		ring->frame_size	= req->nm_frame_size;
428 		ring->pg_vec_pages	= req->nm_block_size / PAGE_SIZE;
429 
430 		swap(ring->pg_vec_len, req->nm_block_nr);
431 		swap(ring->pg_vec_order, order);
432 		swap(ring->pg_vec, pg_vec);
433 
434 		__skb_queue_purge(queue);
435 		spin_unlock_bh(&queue->lock);
436 
437 		WARN_ON(atomic_read(&nlk->mapped));
438 	}
439 	mutex_unlock(&nlk->pg_vec_lock);
440 
441 	if (pg_vec)
442 		free_pg_vec(pg_vec, order, req->nm_block_nr);
443 	return err;
444 }
445 
446 static void netlink_mm_open(struct vm_area_struct *vma)
447 {
448 	struct file *file = vma->vm_file;
449 	struct socket *sock = file->private_data;
450 	struct sock *sk = sock->sk;
451 
452 	if (sk)
453 		atomic_inc(&nlk_sk(sk)->mapped);
454 }
455 
456 static void netlink_mm_close(struct vm_area_struct *vma)
457 {
458 	struct file *file = vma->vm_file;
459 	struct socket *sock = file->private_data;
460 	struct sock *sk = sock->sk;
461 
462 	if (sk)
463 		atomic_dec(&nlk_sk(sk)->mapped);
464 }
465 
466 static const struct vm_operations_struct netlink_mmap_ops = {
467 	.open	= netlink_mm_open,
468 	.close	= netlink_mm_close,
469 };
470 
471 static int netlink_mmap(struct file *file, struct socket *sock,
472 			struct vm_area_struct *vma)
473 {
474 	struct sock *sk = sock->sk;
475 	struct netlink_sock *nlk = nlk_sk(sk);
476 	struct netlink_ring *ring;
477 	unsigned long start, size, expected;
478 	unsigned int i;
479 	int err = -EINVAL;
480 
481 	if (vma->vm_pgoff)
482 		return -EINVAL;
483 
484 	mutex_lock(&nlk->pg_vec_lock);
485 
486 	expected = 0;
487 	for (ring = &nlk->rx_ring; ring <= &nlk->tx_ring; ring++) {
488 		if (ring->pg_vec == NULL)
489 			continue;
490 		expected += ring->pg_vec_len * ring->pg_vec_pages * PAGE_SIZE;
491 	}
492 
493 	if (expected == 0)
494 		goto out;
495 
496 	size = vma->vm_end - vma->vm_start;
497 	if (size != expected)
498 		goto out;
499 
500 	start = vma->vm_start;
501 	for (ring = &nlk->rx_ring; ring <= &nlk->tx_ring; ring++) {
502 		if (ring->pg_vec == NULL)
503 			continue;
504 
505 		for (i = 0; i < ring->pg_vec_len; i++) {
506 			struct page *page;
507 			void *kaddr = ring->pg_vec[i];
508 			unsigned int pg_num;
509 
510 			for (pg_num = 0; pg_num < ring->pg_vec_pages; pg_num++) {
511 				page = pgvec_to_page(kaddr);
512 				err = vm_insert_page(vma, start, page);
513 				if (err < 0)
514 					goto out;
515 				start += PAGE_SIZE;
516 				kaddr += PAGE_SIZE;
517 			}
518 		}
519 	}
520 
521 	atomic_inc(&nlk->mapped);
522 	vma->vm_ops = &netlink_mmap_ops;
523 	err = 0;
524 out:
525 	mutex_unlock(&nlk->pg_vec_lock);
526 	return err;
527 }
528 
529 static void netlink_frame_flush_dcache(const struct nl_mmap_hdr *hdr, unsigned int nm_len)
530 {
531 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
532 	struct page *p_start, *p_end;
533 
534 	/* First page is flushed through netlink_{get,set}_status */
535 	p_start = pgvec_to_page(hdr + PAGE_SIZE);
536 	p_end   = pgvec_to_page((void *)hdr + NL_MMAP_HDRLEN + nm_len - 1);
537 	while (p_start <= p_end) {
538 		flush_dcache_page(p_start);
539 		p_start++;
540 	}
541 #endif
542 }
543 
544 static enum nl_mmap_status netlink_get_status(const struct nl_mmap_hdr *hdr)
545 {
546 	smp_rmb();
547 	flush_dcache_page(pgvec_to_page(hdr));
548 	return hdr->nm_status;
549 }
550 
551 static void netlink_set_status(struct nl_mmap_hdr *hdr,
552 			       enum nl_mmap_status status)
553 {
554 	smp_mb();
555 	hdr->nm_status = status;
556 	flush_dcache_page(pgvec_to_page(hdr));
557 }
558 
559 static struct nl_mmap_hdr *
560 __netlink_lookup_frame(const struct netlink_ring *ring, unsigned int pos)
561 {
562 	unsigned int pg_vec_pos, frame_off;
563 
564 	pg_vec_pos = pos / ring->frames_per_block;
565 	frame_off  = pos % ring->frames_per_block;
566 
567 	return ring->pg_vec[pg_vec_pos] + (frame_off * ring->frame_size);
568 }
569 
570 static struct nl_mmap_hdr *
571 netlink_lookup_frame(const struct netlink_ring *ring, unsigned int pos,
572 		     enum nl_mmap_status status)
573 {
574 	struct nl_mmap_hdr *hdr;
575 
576 	hdr = __netlink_lookup_frame(ring, pos);
577 	if (netlink_get_status(hdr) != status)
578 		return NULL;
579 
580 	return hdr;
581 }
582 
583 static struct nl_mmap_hdr *
584 netlink_current_frame(const struct netlink_ring *ring,
585 		      enum nl_mmap_status status)
586 {
587 	return netlink_lookup_frame(ring, ring->head, status);
588 }
589 
590 static struct nl_mmap_hdr *
591 netlink_previous_frame(const struct netlink_ring *ring,
592 		       enum nl_mmap_status status)
593 {
594 	unsigned int prev;
595 
596 	prev = ring->head ? ring->head - 1 : ring->frame_max;
597 	return netlink_lookup_frame(ring, prev, status);
598 }
599 
600 static void netlink_increment_head(struct netlink_ring *ring)
601 {
602 	ring->head = ring->head != ring->frame_max ? ring->head + 1 : 0;
603 }
604 
605 static void netlink_forward_ring(struct netlink_ring *ring)
606 {
607 	unsigned int head = ring->head, pos = head;
608 	const struct nl_mmap_hdr *hdr;
609 
610 	do {
611 		hdr = __netlink_lookup_frame(ring, pos);
612 		if (hdr->nm_status == NL_MMAP_STATUS_UNUSED)
613 			break;
614 		if (hdr->nm_status != NL_MMAP_STATUS_SKIP)
615 			break;
616 		netlink_increment_head(ring);
617 	} while (ring->head != head);
618 }
619 
620 static bool netlink_dump_space(struct netlink_sock *nlk)
621 {
622 	struct netlink_ring *ring = &nlk->rx_ring;
623 	struct nl_mmap_hdr *hdr;
624 	unsigned int n;
625 
626 	hdr = netlink_current_frame(ring, NL_MMAP_STATUS_UNUSED);
627 	if (hdr == NULL)
628 		return false;
629 
630 	n = ring->head + ring->frame_max / 2;
631 	if (n > ring->frame_max)
632 		n -= ring->frame_max;
633 
634 	hdr = __netlink_lookup_frame(ring, n);
635 
636 	return hdr->nm_status == NL_MMAP_STATUS_UNUSED;
637 }
638 
639 static unsigned int netlink_poll(struct file *file, struct socket *sock,
640 				 poll_table *wait)
641 {
642 	struct sock *sk = sock->sk;
643 	struct netlink_sock *nlk = nlk_sk(sk);
644 	unsigned int mask;
645 	int err;
646 
647 	if (nlk->rx_ring.pg_vec != NULL) {
648 		/* Memory mapped sockets don't call recvmsg(), so flow control
649 		 * for dumps is performed here. A dump is allowed to continue
650 		 * if at least half the ring is unused.
651 		 */
652 		while (nlk->cb_running && netlink_dump_space(nlk)) {
653 			err = netlink_dump(sk);
654 			if (err < 0) {
655 				sk->sk_err = -err;
656 				sk->sk_error_report(sk);
657 				break;
658 			}
659 		}
660 		netlink_rcv_wake(sk);
661 	}
662 
663 	mask = datagram_poll(file, sock, wait);
664 
665 	spin_lock_bh(&sk->sk_receive_queue.lock);
666 	if (nlk->rx_ring.pg_vec) {
667 		netlink_forward_ring(&nlk->rx_ring);
668 		if (!netlink_previous_frame(&nlk->rx_ring, NL_MMAP_STATUS_UNUSED))
669 			mask |= POLLIN | POLLRDNORM;
670 	}
671 	spin_unlock_bh(&sk->sk_receive_queue.lock);
672 
673 	spin_lock_bh(&sk->sk_write_queue.lock);
674 	if (nlk->tx_ring.pg_vec) {
675 		if (netlink_current_frame(&nlk->tx_ring, NL_MMAP_STATUS_UNUSED))
676 			mask |= POLLOUT | POLLWRNORM;
677 	}
678 	spin_unlock_bh(&sk->sk_write_queue.lock);
679 
680 	return mask;
681 }
682 
683 static struct nl_mmap_hdr *netlink_mmap_hdr(struct sk_buff *skb)
684 {
685 	return (struct nl_mmap_hdr *)(skb->head - NL_MMAP_HDRLEN);
686 }
687 
688 static void netlink_ring_setup_skb(struct sk_buff *skb, struct sock *sk,
689 				   struct netlink_ring *ring,
690 				   struct nl_mmap_hdr *hdr)
691 {
692 	unsigned int size;
693 	void *data;
694 
695 	size = ring->frame_size - NL_MMAP_HDRLEN;
696 	data = (void *)hdr + NL_MMAP_HDRLEN;
697 
698 	skb->head	= data;
699 	skb->data	= data;
700 	skb_reset_tail_pointer(skb);
701 	skb->end	= skb->tail + size;
702 	skb->len	= 0;
703 
704 	skb->destructor	= netlink_skb_destructor;
705 	NETLINK_CB(skb).flags |= NETLINK_SKB_MMAPED;
706 	NETLINK_CB(skb).sk = sk;
707 }
708 
709 static int netlink_mmap_sendmsg(struct sock *sk, struct msghdr *msg,
710 				u32 dst_portid, u32 dst_group,
711 				struct sock_iocb *siocb)
712 {
713 	struct netlink_sock *nlk = nlk_sk(sk);
714 	struct netlink_ring *ring;
715 	struct nl_mmap_hdr *hdr;
716 	struct sk_buff *skb;
717 	unsigned int maxlen;
718 	int err = 0, len = 0;
719 
720 	mutex_lock(&nlk->pg_vec_lock);
721 
722 	ring   = &nlk->tx_ring;
723 	maxlen = ring->frame_size - NL_MMAP_HDRLEN;
724 
725 	do {
726 		unsigned int nm_len;
727 
728 		hdr = netlink_current_frame(ring, NL_MMAP_STATUS_VALID);
729 		if (hdr == NULL) {
730 			if (!(msg->msg_flags & MSG_DONTWAIT) &&
731 			    atomic_read(&nlk->tx_ring.pending))
732 				schedule();
733 			continue;
734 		}
735 
736 		nm_len = ACCESS_ONCE(hdr->nm_len);
737 		if (nm_len > maxlen) {
738 			err = -EINVAL;
739 			goto out;
740 		}
741 
742 		netlink_frame_flush_dcache(hdr, nm_len);
743 
744 		skb = alloc_skb(nm_len, GFP_KERNEL);
745 		if (skb == NULL) {
746 			err = -ENOBUFS;
747 			goto out;
748 		}
749 		__skb_put(skb, nm_len);
750 		memcpy(skb->data, (void *)hdr + NL_MMAP_HDRLEN, nm_len);
751 		netlink_set_status(hdr, NL_MMAP_STATUS_UNUSED);
752 
753 		netlink_increment_head(ring);
754 
755 		NETLINK_CB(skb).portid	  = nlk->portid;
756 		NETLINK_CB(skb).dst_group = dst_group;
757 		NETLINK_CB(skb).creds	  = siocb->scm->creds;
758 
759 		err = security_netlink_send(sk, skb);
760 		if (err) {
761 			kfree_skb(skb);
762 			goto out;
763 		}
764 
765 		if (unlikely(dst_group)) {
766 			atomic_inc(&skb->users);
767 			netlink_broadcast(sk, skb, dst_portid, dst_group,
768 					  GFP_KERNEL);
769 		}
770 		err = netlink_unicast(sk, skb, dst_portid,
771 				      msg->msg_flags & MSG_DONTWAIT);
772 		if (err < 0)
773 			goto out;
774 		len += err;
775 
776 	} while (hdr != NULL ||
777 		 (!(msg->msg_flags & MSG_DONTWAIT) &&
778 		  atomic_read(&nlk->tx_ring.pending)));
779 
780 	if (len > 0)
781 		err = len;
782 out:
783 	mutex_unlock(&nlk->pg_vec_lock);
784 	return err;
785 }
786 
787 static void netlink_queue_mmaped_skb(struct sock *sk, struct sk_buff *skb)
788 {
789 	struct nl_mmap_hdr *hdr;
790 
791 	hdr = netlink_mmap_hdr(skb);
792 	hdr->nm_len	= skb->len;
793 	hdr->nm_group	= NETLINK_CB(skb).dst_group;
794 	hdr->nm_pid	= NETLINK_CB(skb).creds.pid;
795 	hdr->nm_uid	= from_kuid(sk_user_ns(sk), NETLINK_CB(skb).creds.uid);
796 	hdr->nm_gid	= from_kgid(sk_user_ns(sk), NETLINK_CB(skb).creds.gid);
797 	netlink_frame_flush_dcache(hdr, hdr->nm_len);
798 	netlink_set_status(hdr, NL_MMAP_STATUS_VALID);
799 
800 	NETLINK_CB(skb).flags |= NETLINK_SKB_DELIVERED;
801 	kfree_skb(skb);
802 }
803 
804 static void netlink_ring_set_copied(struct sock *sk, struct sk_buff *skb)
805 {
806 	struct netlink_sock *nlk = nlk_sk(sk);
807 	struct netlink_ring *ring = &nlk->rx_ring;
808 	struct nl_mmap_hdr *hdr;
809 
810 	spin_lock_bh(&sk->sk_receive_queue.lock);
811 	hdr = netlink_current_frame(ring, NL_MMAP_STATUS_UNUSED);
812 	if (hdr == NULL) {
813 		spin_unlock_bh(&sk->sk_receive_queue.lock);
814 		kfree_skb(skb);
815 		netlink_overrun(sk);
816 		return;
817 	}
818 	netlink_increment_head(ring);
819 	__skb_queue_tail(&sk->sk_receive_queue, skb);
820 	spin_unlock_bh(&sk->sk_receive_queue.lock);
821 
822 	hdr->nm_len	= skb->len;
823 	hdr->nm_group	= NETLINK_CB(skb).dst_group;
824 	hdr->nm_pid	= NETLINK_CB(skb).creds.pid;
825 	hdr->nm_uid	= from_kuid(sk_user_ns(sk), NETLINK_CB(skb).creds.uid);
826 	hdr->nm_gid	= from_kgid(sk_user_ns(sk), NETLINK_CB(skb).creds.gid);
827 	netlink_set_status(hdr, NL_MMAP_STATUS_COPY);
828 }
829 
830 #else /* CONFIG_NETLINK_MMAP */
831 #define netlink_skb_is_mmaped(skb)	false
832 #define netlink_rx_is_mmaped(sk)	false
833 #define netlink_tx_is_mmaped(sk)	false
834 #define netlink_mmap			sock_no_mmap
835 #define netlink_poll			datagram_poll
836 #define netlink_mmap_sendmsg(sk, msg, dst_portid, dst_group, siocb)	0
837 #endif /* CONFIG_NETLINK_MMAP */
838 
839 static void netlink_skb_destructor(struct sk_buff *skb)
840 {
841 #ifdef CONFIG_NETLINK_MMAP
842 	struct nl_mmap_hdr *hdr;
843 	struct netlink_ring *ring;
844 	struct sock *sk;
845 
846 	/* If a packet from the kernel to userspace was freed because of an
847 	 * error without being delivered to userspace, the kernel must reset
848 	 * the status. In the direction userspace to kernel, the status is
849 	 * always reset here after the packet was processed and freed.
850 	 */
851 	if (netlink_skb_is_mmaped(skb)) {
852 		hdr = netlink_mmap_hdr(skb);
853 		sk = NETLINK_CB(skb).sk;
854 
855 		if (NETLINK_CB(skb).flags & NETLINK_SKB_TX) {
856 			netlink_set_status(hdr, NL_MMAP_STATUS_UNUSED);
857 			ring = &nlk_sk(sk)->tx_ring;
858 		} else {
859 			if (!(NETLINK_CB(skb).flags & NETLINK_SKB_DELIVERED)) {
860 				hdr->nm_len = 0;
861 				netlink_set_status(hdr, NL_MMAP_STATUS_VALID);
862 			}
863 			ring = &nlk_sk(sk)->rx_ring;
864 		}
865 
866 		WARN_ON(atomic_read(&ring->pending) == 0);
867 		atomic_dec(&ring->pending);
868 		sock_put(sk);
869 
870 		skb->head = NULL;
871 	}
872 #endif
873 	if (is_vmalloc_addr(skb->head)) {
874 		if (!skb->cloned ||
875 		    !atomic_dec_return(&(skb_shinfo(skb)->dataref)))
876 			vfree(skb->head);
877 
878 		skb->head = NULL;
879 	}
880 	if (skb->sk != NULL)
881 		sock_rfree(skb);
882 }
883 
884 static void netlink_skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
885 {
886 	WARN_ON(skb->sk != NULL);
887 	skb->sk = sk;
888 	skb->destructor = netlink_skb_destructor;
889 	atomic_add(skb->truesize, &sk->sk_rmem_alloc);
890 	sk_mem_charge(sk, skb->truesize);
891 }
892 
893 static void netlink_sock_destruct(struct sock *sk)
894 {
895 	struct netlink_sock *nlk = nlk_sk(sk);
896 
897 	if (nlk->cb_running) {
898 		if (nlk->cb.done)
899 			nlk->cb.done(&nlk->cb);
900 
901 		module_put(nlk->cb.module);
902 		kfree_skb(nlk->cb.skb);
903 	}
904 
905 	skb_queue_purge(&sk->sk_receive_queue);
906 #ifdef CONFIG_NETLINK_MMAP
907 	if (1) {
908 		struct nl_mmap_req req;
909 
910 		memset(&req, 0, sizeof(req));
911 		if (nlk->rx_ring.pg_vec)
912 			netlink_set_ring(sk, &req, true, false);
913 		memset(&req, 0, sizeof(req));
914 		if (nlk->tx_ring.pg_vec)
915 			netlink_set_ring(sk, &req, true, true);
916 	}
917 #endif /* CONFIG_NETLINK_MMAP */
918 
919 	if (!sock_flag(sk, SOCK_DEAD)) {
920 		printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
921 		return;
922 	}
923 
924 	WARN_ON(atomic_read(&sk->sk_rmem_alloc));
925 	WARN_ON(atomic_read(&sk->sk_wmem_alloc));
926 	WARN_ON(nlk_sk(sk)->groups);
927 }
928 
929 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
930  * SMP. Look, when several writers sleep and reader wakes them up, all but one
931  * immediately hit write lock and grab all the cpus. Exclusive sleep solves
932  * this, _but_ remember, it adds useless work on UP machines.
933  */
934 
935 void netlink_table_grab(void)
936 	__acquires(nl_table_lock)
937 {
938 	might_sleep();
939 
940 	write_lock_irq(&nl_table_lock);
941 
942 	if (atomic_read(&nl_table_users)) {
943 		DECLARE_WAITQUEUE(wait, current);
944 
945 		add_wait_queue_exclusive(&nl_table_wait, &wait);
946 		for (;;) {
947 			set_current_state(TASK_UNINTERRUPTIBLE);
948 			if (atomic_read(&nl_table_users) == 0)
949 				break;
950 			write_unlock_irq(&nl_table_lock);
951 			schedule();
952 			write_lock_irq(&nl_table_lock);
953 		}
954 
955 		__set_current_state(TASK_RUNNING);
956 		remove_wait_queue(&nl_table_wait, &wait);
957 	}
958 }
959 
960 void netlink_table_ungrab(void)
961 	__releases(nl_table_lock)
962 {
963 	write_unlock_irq(&nl_table_lock);
964 	wake_up(&nl_table_wait);
965 }
966 
967 static inline void
968 netlink_lock_table(void)
969 {
970 	/* read_lock() synchronizes us to netlink_table_grab */
971 
972 	read_lock(&nl_table_lock);
973 	atomic_inc(&nl_table_users);
974 	read_unlock(&nl_table_lock);
975 }
976 
977 static inline void
978 netlink_unlock_table(void)
979 {
980 	if (atomic_dec_and_test(&nl_table_users))
981 		wake_up(&nl_table_wait);
982 }
983 
984 struct netlink_compare_arg
985 {
986 	struct net *net;
987 	u32 portid;
988 };
989 
990 static bool netlink_compare(void *ptr, void *arg)
991 {
992 	struct netlink_compare_arg *x = arg;
993 	struct sock *sk = ptr;
994 
995 	return nlk_sk(sk)->portid == x->portid &&
996 	       net_eq(sock_net(sk), x->net);
997 }
998 
999 static struct sock *__netlink_lookup(struct netlink_table *table, u32 portid,
1000 				     struct net *net)
1001 {
1002 	struct netlink_compare_arg arg = {
1003 		.net = net,
1004 		.portid = portid,
1005 	};
1006 	u32 hash;
1007 
1008 	hash = rhashtable_hashfn(&table->hash, &portid, sizeof(portid));
1009 
1010 	return rhashtable_lookup_compare(&table->hash, hash,
1011 					 &netlink_compare, &arg);
1012 }
1013 
1014 static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
1015 {
1016 	struct netlink_table *table = &nl_table[protocol];
1017 	struct sock *sk;
1018 
1019 	read_lock(&nl_table_lock);
1020 	rcu_read_lock();
1021 	sk = __netlink_lookup(table, portid, net);
1022 	if (sk)
1023 		sock_hold(sk);
1024 	rcu_read_unlock();
1025 	read_unlock(&nl_table_lock);
1026 
1027 	return sk;
1028 }
1029 
1030 static const struct proto_ops netlink_ops;
1031 
1032 static void
1033 netlink_update_listeners(struct sock *sk)
1034 {
1035 	struct netlink_table *tbl = &nl_table[sk->sk_protocol];
1036 	unsigned long mask;
1037 	unsigned int i;
1038 	struct listeners *listeners;
1039 
1040 	listeners = nl_deref_protected(tbl->listeners);
1041 	if (!listeners)
1042 		return;
1043 
1044 	for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
1045 		mask = 0;
1046 		sk_for_each_bound(sk, &tbl->mc_list) {
1047 			if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
1048 				mask |= nlk_sk(sk)->groups[i];
1049 		}
1050 		listeners->masks[i] = mask;
1051 	}
1052 	/* this function is only called with the netlink table "grabbed", which
1053 	 * makes sure updates are visible before bind or setsockopt return. */
1054 }
1055 
1056 static int netlink_insert(struct sock *sk, struct net *net, u32 portid)
1057 {
1058 	struct netlink_table *table = &nl_table[sk->sk_protocol];
1059 	int err = -EADDRINUSE;
1060 
1061 	mutex_lock(&nl_sk_hash_lock);
1062 	if (__netlink_lookup(table, portid, net))
1063 		goto err;
1064 
1065 	err = -EBUSY;
1066 	if (nlk_sk(sk)->portid)
1067 		goto err;
1068 
1069 	err = -ENOMEM;
1070 	if (BITS_PER_LONG > 32 && unlikely(table->hash.nelems >= UINT_MAX))
1071 		goto err;
1072 
1073 	nlk_sk(sk)->portid = portid;
1074 	sock_hold(sk);
1075 	rhashtable_insert(&table->hash, &nlk_sk(sk)->node);
1076 	err = 0;
1077 err:
1078 	mutex_unlock(&nl_sk_hash_lock);
1079 	return err;
1080 }
1081 
1082 static void netlink_remove(struct sock *sk)
1083 {
1084 	struct netlink_table *table;
1085 
1086 	mutex_lock(&nl_sk_hash_lock);
1087 	table = &nl_table[sk->sk_protocol];
1088 	if (rhashtable_remove(&table->hash, &nlk_sk(sk)->node)) {
1089 		WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
1090 		__sock_put(sk);
1091 	}
1092 	mutex_unlock(&nl_sk_hash_lock);
1093 
1094 	netlink_table_grab();
1095 	if (nlk_sk(sk)->subscriptions) {
1096 		__sk_del_bind_node(sk);
1097 		netlink_update_listeners(sk);
1098 	}
1099 	if (sk->sk_protocol == NETLINK_GENERIC)
1100 		atomic_inc(&genl_sk_destructing_cnt);
1101 	netlink_table_ungrab();
1102 }
1103 
1104 static struct proto netlink_proto = {
1105 	.name	  = "NETLINK",
1106 	.owner	  = THIS_MODULE,
1107 	.obj_size = sizeof(struct netlink_sock),
1108 };
1109 
1110 static int __netlink_create(struct net *net, struct socket *sock,
1111 			    struct mutex *cb_mutex, int protocol)
1112 {
1113 	struct sock *sk;
1114 	struct netlink_sock *nlk;
1115 
1116 	sock->ops = &netlink_ops;
1117 
1118 	sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto);
1119 	if (!sk)
1120 		return -ENOMEM;
1121 
1122 	sock_init_data(sock, sk);
1123 
1124 	nlk = nlk_sk(sk);
1125 	if (cb_mutex) {
1126 		nlk->cb_mutex = cb_mutex;
1127 	} else {
1128 		nlk->cb_mutex = &nlk->cb_def_mutex;
1129 		mutex_init(nlk->cb_mutex);
1130 	}
1131 	init_waitqueue_head(&nlk->wait);
1132 #ifdef CONFIG_NETLINK_MMAP
1133 	mutex_init(&nlk->pg_vec_lock);
1134 #endif
1135 
1136 	sk->sk_destruct = netlink_sock_destruct;
1137 	sk->sk_protocol = protocol;
1138 	return 0;
1139 }
1140 
1141 static int netlink_create(struct net *net, struct socket *sock, int protocol,
1142 			  int kern)
1143 {
1144 	struct module *module = NULL;
1145 	struct mutex *cb_mutex;
1146 	struct netlink_sock *nlk;
1147 	int (*bind)(struct net *net, int group);
1148 	void (*unbind)(struct net *net, int group);
1149 	int err = 0;
1150 
1151 	sock->state = SS_UNCONNECTED;
1152 
1153 	if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
1154 		return -ESOCKTNOSUPPORT;
1155 
1156 	if (protocol < 0 || protocol >= MAX_LINKS)
1157 		return -EPROTONOSUPPORT;
1158 
1159 	netlink_lock_table();
1160 #ifdef CONFIG_MODULES
1161 	if (!nl_table[protocol].registered) {
1162 		netlink_unlock_table();
1163 		request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
1164 		netlink_lock_table();
1165 	}
1166 #endif
1167 	if (nl_table[protocol].registered &&
1168 	    try_module_get(nl_table[protocol].module))
1169 		module = nl_table[protocol].module;
1170 	else
1171 		err = -EPROTONOSUPPORT;
1172 	cb_mutex = nl_table[protocol].cb_mutex;
1173 	bind = nl_table[protocol].bind;
1174 	unbind = nl_table[protocol].unbind;
1175 	netlink_unlock_table();
1176 
1177 	if (err < 0)
1178 		goto out;
1179 
1180 	err = __netlink_create(net, sock, cb_mutex, protocol);
1181 	if (err < 0)
1182 		goto out_module;
1183 
1184 	local_bh_disable();
1185 	sock_prot_inuse_add(net, &netlink_proto, 1);
1186 	local_bh_enable();
1187 
1188 	nlk = nlk_sk(sock->sk);
1189 	nlk->module = module;
1190 	nlk->netlink_bind = bind;
1191 	nlk->netlink_unbind = unbind;
1192 out:
1193 	return err;
1194 
1195 out_module:
1196 	module_put(module);
1197 	goto out;
1198 }
1199 
1200 static int netlink_release(struct socket *sock)
1201 {
1202 	struct sock *sk = sock->sk;
1203 	struct netlink_sock *nlk;
1204 
1205 	if (!sk)
1206 		return 0;
1207 
1208 	netlink_remove(sk);
1209 	sock_orphan(sk);
1210 	nlk = nlk_sk(sk);
1211 
1212 	/*
1213 	 * OK. Socket is unlinked, any packets that arrive now
1214 	 * will be purged.
1215 	 */
1216 
1217 	/* must not acquire netlink_table_lock in any way again before unbind
1218 	 * and notifying genetlink is done as otherwise it might deadlock
1219 	 */
1220 	if (nlk->netlink_unbind) {
1221 		int i;
1222 
1223 		for (i = 0; i < nlk->ngroups; i++)
1224 			if (test_bit(i, nlk->groups))
1225 				nlk->netlink_unbind(sock_net(sk), i + 1);
1226 	}
1227 	if (sk->sk_protocol == NETLINK_GENERIC &&
1228 	    atomic_dec_return(&genl_sk_destructing_cnt) == 0)
1229 		wake_up(&genl_sk_destructing_waitq);
1230 
1231 	sock->sk = NULL;
1232 	wake_up_interruptible_all(&nlk->wait);
1233 
1234 	skb_queue_purge(&sk->sk_write_queue);
1235 
1236 	if (nlk->portid) {
1237 		struct netlink_notify n = {
1238 						.net = sock_net(sk),
1239 						.protocol = sk->sk_protocol,
1240 						.portid = nlk->portid,
1241 					  };
1242 		atomic_notifier_call_chain(&netlink_chain,
1243 				NETLINK_URELEASE, &n);
1244 	}
1245 
1246 	module_put(nlk->module);
1247 
1248 	if (netlink_is_kernel(sk)) {
1249 		netlink_table_grab();
1250 		BUG_ON(nl_table[sk->sk_protocol].registered == 0);
1251 		if (--nl_table[sk->sk_protocol].registered == 0) {
1252 			struct listeners *old;
1253 
1254 			old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
1255 			RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
1256 			kfree_rcu(old, rcu);
1257 			nl_table[sk->sk_protocol].module = NULL;
1258 			nl_table[sk->sk_protocol].bind = NULL;
1259 			nl_table[sk->sk_protocol].unbind = NULL;
1260 			nl_table[sk->sk_protocol].flags = 0;
1261 			nl_table[sk->sk_protocol].registered = 0;
1262 		}
1263 		netlink_table_ungrab();
1264 	}
1265 
1266 	kfree(nlk->groups);
1267 	nlk->groups = NULL;
1268 
1269 	local_bh_disable();
1270 	sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
1271 	local_bh_enable();
1272 	sock_put(sk);
1273 	return 0;
1274 }
1275 
1276 static int netlink_autobind(struct socket *sock)
1277 {
1278 	struct sock *sk = sock->sk;
1279 	struct net *net = sock_net(sk);
1280 	struct netlink_table *table = &nl_table[sk->sk_protocol];
1281 	s32 portid = task_tgid_vnr(current);
1282 	int err;
1283 	static s32 rover = -4097;
1284 
1285 retry:
1286 	cond_resched();
1287 	netlink_table_grab();
1288 	rcu_read_lock();
1289 	if (__netlink_lookup(table, portid, net)) {
1290 		/* Bind collision, search negative portid values. */
1291 		portid = rover--;
1292 		if (rover > -4097)
1293 			rover = -4097;
1294 		rcu_read_unlock();
1295 		netlink_table_ungrab();
1296 		goto retry;
1297 	}
1298 	rcu_read_unlock();
1299 	netlink_table_ungrab();
1300 
1301 	err = netlink_insert(sk, net, portid);
1302 	if (err == -EADDRINUSE)
1303 		goto retry;
1304 
1305 	/* If 2 threads race to autobind, that is fine.  */
1306 	if (err == -EBUSY)
1307 		err = 0;
1308 
1309 	return err;
1310 }
1311 
1312 /**
1313  * __netlink_ns_capable - General netlink message capability test
1314  * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
1315  * @user_ns: The user namespace of the capability to use
1316  * @cap: The capability to use
1317  *
1318  * Test to see if the opener of the socket we received the message
1319  * from had when the netlink socket was created and the sender of the
1320  * message has has the capability @cap in the user namespace @user_ns.
1321  */
1322 bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
1323 			struct user_namespace *user_ns, int cap)
1324 {
1325 	return ((nsp->flags & NETLINK_SKB_DST) ||
1326 		file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
1327 		ns_capable(user_ns, cap);
1328 }
1329 EXPORT_SYMBOL(__netlink_ns_capable);
1330 
1331 /**
1332  * netlink_ns_capable - General netlink message capability test
1333  * @skb: socket buffer holding a netlink command from userspace
1334  * @user_ns: The user namespace of the capability to use
1335  * @cap: The capability to use
1336  *
1337  * Test to see if the opener of the socket we received the message
1338  * from had when the netlink socket was created and the sender of the
1339  * message has has the capability @cap in the user namespace @user_ns.
1340  */
1341 bool netlink_ns_capable(const struct sk_buff *skb,
1342 			struct user_namespace *user_ns, int cap)
1343 {
1344 	return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
1345 }
1346 EXPORT_SYMBOL(netlink_ns_capable);
1347 
1348 /**
1349  * netlink_capable - Netlink global message capability test
1350  * @skb: socket buffer holding a netlink command from userspace
1351  * @cap: The capability to use
1352  *
1353  * Test to see if the opener of the socket we received the message
1354  * from had when the netlink socket was created and the sender of the
1355  * message has has the capability @cap in all user namespaces.
1356  */
1357 bool netlink_capable(const struct sk_buff *skb, int cap)
1358 {
1359 	return netlink_ns_capable(skb, &init_user_ns, cap);
1360 }
1361 EXPORT_SYMBOL(netlink_capable);
1362 
1363 /**
1364  * netlink_net_capable - Netlink network namespace message capability test
1365  * @skb: socket buffer holding a netlink command from userspace
1366  * @cap: The capability to use
1367  *
1368  * Test to see if the opener of the socket we received the message
1369  * from had when the netlink socket was created and the sender of the
1370  * message has has the capability @cap over the network namespace of
1371  * the socket we received the message from.
1372  */
1373 bool netlink_net_capable(const struct sk_buff *skb, int cap)
1374 {
1375 	return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
1376 }
1377 EXPORT_SYMBOL(netlink_net_capable);
1378 
1379 static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
1380 {
1381 	return (nl_table[sock->sk->sk_protocol].flags & flag) ||
1382 		ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
1383 }
1384 
1385 static void
1386 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
1387 {
1388 	struct netlink_sock *nlk = nlk_sk(sk);
1389 
1390 	if (nlk->subscriptions && !subscriptions)
1391 		__sk_del_bind_node(sk);
1392 	else if (!nlk->subscriptions && subscriptions)
1393 		sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
1394 	nlk->subscriptions = subscriptions;
1395 }
1396 
1397 static int netlink_realloc_groups(struct sock *sk)
1398 {
1399 	struct netlink_sock *nlk = nlk_sk(sk);
1400 	unsigned int groups;
1401 	unsigned long *new_groups;
1402 	int err = 0;
1403 
1404 	netlink_table_grab();
1405 
1406 	groups = nl_table[sk->sk_protocol].groups;
1407 	if (!nl_table[sk->sk_protocol].registered) {
1408 		err = -ENOENT;
1409 		goto out_unlock;
1410 	}
1411 
1412 	if (nlk->ngroups >= groups)
1413 		goto out_unlock;
1414 
1415 	new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
1416 	if (new_groups == NULL) {
1417 		err = -ENOMEM;
1418 		goto out_unlock;
1419 	}
1420 	memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
1421 	       NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
1422 
1423 	nlk->groups = new_groups;
1424 	nlk->ngroups = groups;
1425  out_unlock:
1426 	netlink_table_ungrab();
1427 	return err;
1428 }
1429 
1430 static void netlink_undo_bind(int group, long unsigned int groups,
1431 			      struct sock *sk)
1432 {
1433 	struct netlink_sock *nlk = nlk_sk(sk);
1434 	int undo;
1435 
1436 	if (!nlk->netlink_unbind)
1437 		return;
1438 
1439 	for (undo = 0; undo < group; undo++)
1440 		if (test_bit(undo, &groups))
1441 			nlk->netlink_unbind(sock_net(sk), undo);
1442 }
1443 
1444 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
1445 			int addr_len)
1446 {
1447 	struct sock *sk = sock->sk;
1448 	struct net *net = sock_net(sk);
1449 	struct netlink_sock *nlk = nlk_sk(sk);
1450 	struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1451 	int err;
1452 	long unsigned int groups = nladdr->nl_groups;
1453 
1454 	if (addr_len < sizeof(struct sockaddr_nl))
1455 		return -EINVAL;
1456 
1457 	if (nladdr->nl_family != AF_NETLINK)
1458 		return -EINVAL;
1459 
1460 	/* Only superuser is allowed to listen multicasts */
1461 	if (groups) {
1462 		if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1463 			return -EPERM;
1464 		err = netlink_realloc_groups(sk);
1465 		if (err)
1466 			return err;
1467 	}
1468 
1469 	if (nlk->portid)
1470 		if (nladdr->nl_pid != nlk->portid)
1471 			return -EINVAL;
1472 
1473 	if (nlk->netlink_bind && groups) {
1474 		int group;
1475 
1476 		for (group = 0; group < nlk->ngroups; group++) {
1477 			if (!test_bit(group, &groups))
1478 				continue;
1479 			err = nlk->netlink_bind(net, group);
1480 			if (!err)
1481 				continue;
1482 			netlink_undo_bind(group, groups, sk);
1483 			return err;
1484 		}
1485 	}
1486 
1487 	if (!nlk->portid) {
1488 		err = nladdr->nl_pid ?
1489 			netlink_insert(sk, net, nladdr->nl_pid) :
1490 			netlink_autobind(sock);
1491 		if (err) {
1492 			netlink_undo_bind(nlk->ngroups, groups, sk);
1493 			return err;
1494 		}
1495 	}
1496 
1497 	if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1498 		return 0;
1499 
1500 	netlink_table_grab();
1501 	netlink_update_subscriptions(sk, nlk->subscriptions +
1502 					 hweight32(groups) -
1503 					 hweight32(nlk->groups[0]));
1504 	nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
1505 	netlink_update_listeners(sk);
1506 	netlink_table_ungrab();
1507 
1508 	return 0;
1509 }
1510 
1511 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1512 			   int alen, int flags)
1513 {
1514 	int err = 0;
1515 	struct sock *sk = sock->sk;
1516 	struct netlink_sock *nlk = nlk_sk(sk);
1517 	struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1518 
1519 	if (alen < sizeof(addr->sa_family))
1520 		return -EINVAL;
1521 
1522 	if (addr->sa_family == AF_UNSPEC) {
1523 		sk->sk_state	= NETLINK_UNCONNECTED;
1524 		nlk->dst_portid	= 0;
1525 		nlk->dst_group  = 0;
1526 		return 0;
1527 	}
1528 	if (addr->sa_family != AF_NETLINK)
1529 		return -EINVAL;
1530 
1531 	if ((nladdr->nl_groups || nladdr->nl_pid) &&
1532 	    !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1533 		return -EPERM;
1534 
1535 	if (!nlk->portid)
1536 		err = netlink_autobind(sock);
1537 
1538 	if (err == 0) {
1539 		sk->sk_state	= NETLINK_CONNECTED;
1540 		nlk->dst_portid = nladdr->nl_pid;
1541 		nlk->dst_group  = ffs(nladdr->nl_groups);
1542 	}
1543 
1544 	return err;
1545 }
1546 
1547 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1548 			   int *addr_len, int peer)
1549 {
1550 	struct sock *sk = sock->sk;
1551 	struct netlink_sock *nlk = nlk_sk(sk);
1552 	DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1553 
1554 	nladdr->nl_family = AF_NETLINK;
1555 	nladdr->nl_pad = 0;
1556 	*addr_len = sizeof(*nladdr);
1557 
1558 	if (peer) {
1559 		nladdr->nl_pid = nlk->dst_portid;
1560 		nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
1561 	} else {
1562 		nladdr->nl_pid = nlk->portid;
1563 		nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1564 	}
1565 	return 0;
1566 }
1567 
1568 static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1569 {
1570 	struct sock *sock;
1571 	struct netlink_sock *nlk;
1572 
1573 	sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1574 	if (!sock)
1575 		return ERR_PTR(-ECONNREFUSED);
1576 
1577 	/* Don't bother queuing skb if kernel socket has no input function */
1578 	nlk = nlk_sk(sock);
1579 	if (sock->sk_state == NETLINK_CONNECTED &&
1580 	    nlk->dst_portid != nlk_sk(ssk)->portid) {
1581 		sock_put(sock);
1582 		return ERR_PTR(-ECONNREFUSED);
1583 	}
1584 	return sock;
1585 }
1586 
1587 struct sock *netlink_getsockbyfilp(struct file *filp)
1588 {
1589 	struct inode *inode = file_inode(filp);
1590 	struct sock *sock;
1591 
1592 	if (!S_ISSOCK(inode->i_mode))
1593 		return ERR_PTR(-ENOTSOCK);
1594 
1595 	sock = SOCKET_I(inode)->sk;
1596 	if (sock->sk_family != AF_NETLINK)
1597 		return ERR_PTR(-EINVAL);
1598 
1599 	sock_hold(sock);
1600 	return sock;
1601 }
1602 
1603 static struct sk_buff *netlink_alloc_large_skb(unsigned int size,
1604 					       int broadcast)
1605 {
1606 	struct sk_buff *skb;
1607 	void *data;
1608 
1609 	if (size <= NLMSG_GOODSIZE || broadcast)
1610 		return alloc_skb(size, GFP_KERNEL);
1611 
1612 	size = SKB_DATA_ALIGN(size) +
1613 	       SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1614 
1615 	data = vmalloc(size);
1616 	if (data == NULL)
1617 		return NULL;
1618 
1619 	skb = build_skb(data, size);
1620 	if (skb == NULL)
1621 		vfree(data);
1622 	else {
1623 		skb->head_frag = 0;
1624 		skb->destructor = netlink_skb_destructor;
1625 	}
1626 
1627 	return skb;
1628 }
1629 
1630 /*
1631  * Attach a skb to a netlink socket.
1632  * The caller must hold a reference to the destination socket. On error, the
1633  * reference is dropped. The skb is not send to the destination, just all
1634  * all error checks are performed and memory in the queue is reserved.
1635  * Return values:
1636  * < 0: error. skb freed, reference to sock dropped.
1637  * 0: continue
1638  * 1: repeat lookup - reference dropped while waiting for socket memory.
1639  */
1640 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1641 		      long *timeo, struct sock *ssk)
1642 {
1643 	struct netlink_sock *nlk;
1644 
1645 	nlk = nlk_sk(sk);
1646 
1647 	if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1648 	     test_bit(NETLINK_CONGESTED, &nlk->state)) &&
1649 	    !netlink_skb_is_mmaped(skb)) {
1650 		DECLARE_WAITQUEUE(wait, current);
1651 		if (!*timeo) {
1652 			if (!ssk || netlink_is_kernel(ssk))
1653 				netlink_overrun(sk);
1654 			sock_put(sk);
1655 			kfree_skb(skb);
1656 			return -EAGAIN;
1657 		}
1658 
1659 		__set_current_state(TASK_INTERRUPTIBLE);
1660 		add_wait_queue(&nlk->wait, &wait);
1661 
1662 		if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1663 		     test_bit(NETLINK_CONGESTED, &nlk->state)) &&
1664 		    !sock_flag(sk, SOCK_DEAD))
1665 			*timeo = schedule_timeout(*timeo);
1666 
1667 		__set_current_state(TASK_RUNNING);
1668 		remove_wait_queue(&nlk->wait, &wait);
1669 		sock_put(sk);
1670 
1671 		if (signal_pending(current)) {
1672 			kfree_skb(skb);
1673 			return sock_intr_errno(*timeo);
1674 		}
1675 		return 1;
1676 	}
1677 	netlink_skb_set_owner_r(skb, sk);
1678 	return 0;
1679 }
1680 
1681 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1682 {
1683 	int len = skb->len;
1684 
1685 	netlink_deliver_tap(skb);
1686 
1687 #ifdef CONFIG_NETLINK_MMAP
1688 	if (netlink_skb_is_mmaped(skb))
1689 		netlink_queue_mmaped_skb(sk, skb);
1690 	else if (netlink_rx_is_mmaped(sk))
1691 		netlink_ring_set_copied(sk, skb);
1692 	else
1693 #endif /* CONFIG_NETLINK_MMAP */
1694 		skb_queue_tail(&sk->sk_receive_queue, skb);
1695 	sk->sk_data_ready(sk);
1696 	return len;
1697 }
1698 
1699 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1700 {
1701 	int len = __netlink_sendskb(sk, skb);
1702 
1703 	sock_put(sk);
1704 	return len;
1705 }
1706 
1707 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1708 {
1709 	kfree_skb(skb);
1710 	sock_put(sk);
1711 }
1712 
1713 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1714 {
1715 	int delta;
1716 
1717 	WARN_ON(skb->sk != NULL);
1718 	if (netlink_skb_is_mmaped(skb))
1719 		return skb;
1720 
1721 	delta = skb->end - skb->tail;
1722 	if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1723 		return skb;
1724 
1725 	if (skb_shared(skb)) {
1726 		struct sk_buff *nskb = skb_clone(skb, allocation);
1727 		if (!nskb)
1728 			return skb;
1729 		consume_skb(skb);
1730 		skb = nskb;
1731 	}
1732 
1733 	if (!pskb_expand_head(skb, 0, -delta, allocation))
1734 		skb->truesize -= delta;
1735 
1736 	return skb;
1737 }
1738 
1739 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1740 				  struct sock *ssk)
1741 {
1742 	int ret;
1743 	struct netlink_sock *nlk = nlk_sk(sk);
1744 
1745 	ret = -ECONNREFUSED;
1746 	if (nlk->netlink_rcv != NULL) {
1747 		ret = skb->len;
1748 		netlink_skb_set_owner_r(skb, sk);
1749 		NETLINK_CB(skb).sk = ssk;
1750 		netlink_deliver_tap_kernel(sk, ssk, skb);
1751 		nlk->netlink_rcv(skb);
1752 		consume_skb(skb);
1753 	} else {
1754 		kfree_skb(skb);
1755 	}
1756 	sock_put(sk);
1757 	return ret;
1758 }
1759 
1760 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1761 		    u32 portid, int nonblock)
1762 {
1763 	struct sock *sk;
1764 	int err;
1765 	long timeo;
1766 
1767 	skb = netlink_trim(skb, gfp_any());
1768 
1769 	timeo = sock_sndtimeo(ssk, nonblock);
1770 retry:
1771 	sk = netlink_getsockbyportid(ssk, portid);
1772 	if (IS_ERR(sk)) {
1773 		kfree_skb(skb);
1774 		return PTR_ERR(sk);
1775 	}
1776 	if (netlink_is_kernel(sk))
1777 		return netlink_unicast_kernel(sk, skb, ssk);
1778 
1779 	if (sk_filter(sk, skb)) {
1780 		err = skb->len;
1781 		kfree_skb(skb);
1782 		sock_put(sk);
1783 		return err;
1784 	}
1785 
1786 	err = netlink_attachskb(sk, skb, &timeo, ssk);
1787 	if (err == 1)
1788 		goto retry;
1789 	if (err)
1790 		return err;
1791 
1792 	return netlink_sendskb(sk, skb);
1793 }
1794 EXPORT_SYMBOL(netlink_unicast);
1795 
1796 struct sk_buff *netlink_alloc_skb(struct sock *ssk, unsigned int size,
1797 				  u32 dst_portid, gfp_t gfp_mask)
1798 {
1799 #ifdef CONFIG_NETLINK_MMAP
1800 	struct sock *sk = NULL;
1801 	struct sk_buff *skb;
1802 	struct netlink_ring *ring;
1803 	struct nl_mmap_hdr *hdr;
1804 	unsigned int maxlen;
1805 
1806 	sk = netlink_getsockbyportid(ssk, dst_portid);
1807 	if (IS_ERR(sk))
1808 		goto out;
1809 
1810 	ring = &nlk_sk(sk)->rx_ring;
1811 	/* fast-path without atomic ops for common case: non-mmaped receiver */
1812 	if (ring->pg_vec == NULL)
1813 		goto out_put;
1814 
1815 	if (ring->frame_size - NL_MMAP_HDRLEN < size)
1816 		goto out_put;
1817 
1818 	skb = alloc_skb_head(gfp_mask);
1819 	if (skb == NULL)
1820 		goto err1;
1821 
1822 	spin_lock_bh(&sk->sk_receive_queue.lock);
1823 	/* check again under lock */
1824 	if (ring->pg_vec == NULL)
1825 		goto out_free;
1826 
1827 	/* check again under lock */
1828 	maxlen = ring->frame_size - NL_MMAP_HDRLEN;
1829 	if (maxlen < size)
1830 		goto out_free;
1831 
1832 	netlink_forward_ring(ring);
1833 	hdr = netlink_current_frame(ring, NL_MMAP_STATUS_UNUSED);
1834 	if (hdr == NULL)
1835 		goto err2;
1836 	netlink_ring_setup_skb(skb, sk, ring, hdr);
1837 	netlink_set_status(hdr, NL_MMAP_STATUS_RESERVED);
1838 	atomic_inc(&ring->pending);
1839 	netlink_increment_head(ring);
1840 
1841 	spin_unlock_bh(&sk->sk_receive_queue.lock);
1842 	return skb;
1843 
1844 err2:
1845 	kfree_skb(skb);
1846 	spin_unlock_bh(&sk->sk_receive_queue.lock);
1847 	netlink_overrun(sk);
1848 err1:
1849 	sock_put(sk);
1850 	return NULL;
1851 
1852 out_free:
1853 	kfree_skb(skb);
1854 	spin_unlock_bh(&sk->sk_receive_queue.lock);
1855 out_put:
1856 	sock_put(sk);
1857 out:
1858 #endif
1859 	return alloc_skb(size, gfp_mask);
1860 }
1861 EXPORT_SYMBOL_GPL(netlink_alloc_skb);
1862 
1863 int netlink_has_listeners(struct sock *sk, unsigned int group)
1864 {
1865 	int res = 0;
1866 	struct listeners *listeners;
1867 
1868 	BUG_ON(!netlink_is_kernel(sk));
1869 
1870 	rcu_read_lock();
1871 	listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1872 
1873 	if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1874 		res = test_bit(group - 1, listeners->masks);
1875 
1876 	rcu_read_unlock();
1877 
1878 	return res;
1879 }
1880 EXPORT_SYMBOL_GPL(netlink_has_listeners);
1881 
1882 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1883 {
1884 	struct netlink_sock *nlk = nlk_sk(sk);
1885 
1886 	if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1887 	    !test_bit(NETLINK_CONGESTED, &nlk->state)) {
1888 		netlink_skb_set_owner_r(skb, sk);
1889 		__netlink_sendskb(sk, skb);
1890 		return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1891 	}
1892 	return -1;
1893 }
1894 
1895 struct netlink_broadcast_data {
1896 	struct sock *exclude_sk;
1897 	struct net *net;
1898 	u32 portid;
1899 	u32 group;
1900 	int failure;
1901 	int delivery_failure;
1902 	int congested;
1903 	int delivered;
1904 	gfp_t allocation;
1905 	struct sk_buff *skb, *skb2;
1906 	int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1907 	void *tx_data;
1908 };
1909 
1910 static void do_one_broadcast(struct sock *sk,
1911 				    struct netlink_broadcast_data *p)
1912 {
1913 	struct netlink_sock *nlk = nlk_sk(sk);
1914 	int val;
1915 
1916 	if (p->exclude_sk == sk)
1917 		return;
1918 
1919 	if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1920 	    !test_bit(p->group - 1, nlk->groups))
1921 		return;
1922 
1923 	if (!net_eq(sock_net(sk), p->net))
1924 		return;
1925 
1926 	if (p->failure) {
1927 		netlink_overrun(sk);
1928 		return;
1929 	}
1930 
1931 	sock_hold(sk);
1932 	if (p->skb2 == NULL) {
1933 		if (skb_shared(p->skb)) {
1934 			p->skb2 = skb_clone(p->skb, p->allocation);
1935 		} else {
1936 			p->skb2 = skb_get(p->skb);
1937 			/*
1938 			 * skb ownership may have been set when
1939 			 * delivered to a previous socket.
1940 			 */
1941 			skb_orphan(p->skb2);
1942 		}
1943 	}
1944 	if (p->skb2 == NULL) {
1945 		netlink_overrun(sk);
1946 		/* Clone failed. Notify ALL listeners. */
1947 		p->failure = 1;
1948 		if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR)
1949 			p->delivery_failure = 1;
1950 	} else if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1951 		kfree_skb(p->skb2);
1952 		p->skb2 = NULL;
1953 	} else if (sk_filter(sk, p->skb2)) {
1954 		kfree_skb(p->skb2);
1955 		p->skb2 = NULL;
1956 	} else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) {
1957 		netlink_overrun(sk);
1958 		if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR)
1959 			p->delivery_failure = 1;
1960 	} else {
1961 		p->congested |= val;
1962 		p->delivered = 1;
1963 		p->skb2 = NULL;
1964 	}
1965 	sock_put(sk);
1966 }
1967 
1968 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 portid,
1969 	u32 group, gfp_t allocation,
1970 	int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
1971 	void *filter_data)
1972 {
1973 	struct net *net = sock_net(ssk);
1974 	struct netlink_broadcast_data info;
1975 	struct sock *sk;
1976 
1977 	skb = netlink_trim(skb, allocation);
1978 
1979 	info.exclude_sk = ssk;
1980 	info.net = net;
1981 	info.portid = portid;
1982 	info.group = group;
1983 	info.failure = 0;
1984 	info.delivery_failure = 0;
1985 	info.congested = 0;
1986 	info.delivered = 0;
1987 	info.allocation = allocation;
1988 	info.skb = skb;
1989 	info.skb2 = NULL;
1990 	info.tx_filter = filter;
1991 	info.tx_data = filter_data;
1992 
1993 	/* While we sleep in clone, do not allow to change socket list */
1994 
1995 	netlink_lock_table();
1996 
1997 	sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1998 		do_one_broadcast(sk, &info);
1999 
2000 	consume_skb(skb);
2001 
2002 	netlink_unlock_table();
2003 
2004 	if (info.delivery_failure) {
2005 		kfree_skb(info.skb2);
2006 		return -ENOBUFS;
2007 	}
2008 	consume_skb(info.skb2);
2009 
2010 	if (info.delivered) {
2011 		if (info.congested && (allocation & __GFP_WAIT))
2012 			yield();
2013 		return 0;
2014 	}
2015 	return -ESRCH;
2016 }
2017 EXPORT_SYMBOL(netlink_broadcast_filtered);
2018 
2019 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
2020 		      u32 group, gfp_t allocation)
2021 {
2022 	return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
2023 		NULL, NULL);
2024 }
2025 EXPORT_SYMBOL(netlink_broadcast);
2026 
2027 struct netlink_set_err_data {
2028 	struct sock *exclude_sk;
2029 	u32 portid;
2030 	u32 group;
2031 	int code;
2032 };
2033 
2034 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
2035 {
2036 	struct netlink_sock *nlk = nlk_sk(sk);
2037 	int ret = 0;
2038 
2039 	if (sk == p->exclude_sk)
2040 		goto out;
2041 
2042 	if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
2043 		goto out;
2044 
2045 	if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
2046 	    !test_bit(p->group - 1, nlk->groups))
2047 		goto out;
2048 
2049 	if (p->code == ENOBUFS && nlk->flags & NETLINK_RECV_NO_ENOBUFS) {
2050 		ret = 1;
2051 		goto out;
2052 	}
2053 
2054 	sk->sk_err = p->code;
2055 	sk->sk_error_report(sk);
2056 out:
2057 	return ret;
2058 }
2059 
2060 /**
2061  * netlink_set_err - report error to broadcast listeners
2062  * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
2063  * @portid: the PORTID of a process that we want to skip (if any)
2064  * @group: the broadcast group that will notice the error
2065  * @code: error code, must be negative (as usual in kernelspace)
2066  *
2067  * This function returns the number of broadcast listeners that have set the
2068  * NETLINK_RECV_NO_ENOBUFS socket option.
2069  */
2070 int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
2071 {
2072 	struct netlink_set_err_data info;
2073 	struct sock *sk;
2074 	int ret = 0;
2075 
2076 	info.exclude_sk = ssk;
2077 	info.portid = portid;
2078 	info.group = group;
2079 	/* sk->sk_err wants a positive error value */
2080 	info.code = -code;
2081 
2082 	read_lock(&nl_table_lock);
2083 
2084 	sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
2085 		ret += do_one_set_err(sk, &info);
2086 
2087 	read_unlock(&nl_table_lock);
2088 	return ret;
2089 }
2090 EXPORT_SYMBOL(netlink_set_err);
2091 
2092 /* must be called with netlink table grabbed */
2093 static void netlink_update_socket_mc(struct netlink_sock *nlk,
2094 				     unsigned int group,
2095 				     int is_new)
2096 {
2097 	int old, new = !!is_new, subscriptions;
2098 
2099 	old = test_bit(group - 1, nlk->groups);
2100 	subscriptions = nlk->subscriptions - old + new;
2101 	if (new)
2102 		__set_bit(group - 1, nlk->groups);
2103 	else
2104 		__clear_bit(group - 1, nlk->groups);
2105 	netlink_update_subscriptions(&nlk->sk, subscriptions);
2106 	netlink_update_listeners(&nlk->sk);
2107 }
2108 
2109 static int netlink_setsockopt(struct socket *sock, int level, int optname,
2110 			      char __user *optval, unsigned int optlen)
2111 {
2112 	struct sock *sk = sock->sk;
2113 	struct netlink_sock *nlk = nlk_sk(sk);
2114 	unsigned int val = 0;
2115 	int err;
2116 
2117 	if (level != SOL_NETLINK)
2118 		return -ENOPROTOOPT;
2119 
2120 	if (optname != NETLINK_RX_RING && optname != NETLINK_TX_RING &&
2121 	    optlen >= sizeof(int) &&
2122 	    get_user(val, (unsigned int __user *)optval))
2123 		return -EFAULT;
2124 
2125 	switch (optname) {
2126 	case NETLINK_PKTINFO:
2127 		if (val)
2128 			nlk->flags |= NETLINK_RECV_PKTINFO;
2129 		else
2130 			nlk->flags &= ~NETLINK_RECV_PKTINFO;
2131 		err = 0;
2132 		break;
2133 	case NETLINK_ADD_MEMBERSHIP:
2134 	case NETLINK_DROP_MEMBERSHIP: {
2135 		if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
2136 			return -EPERM;
2137 		err = netlink_realloc_groups(sk);
2138 		if (err)
2139 			return err;
2140 		if (!val || val - 1 >= nlk->ngroups)
2141 			return -EINVAL;
2142 		if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
2143 			err = nlk->netlink_bind(sock_net(sk), val);
2144 			if (err)
2145 				return err;
2146 		}
2147 		netlink_table_grab();
2148 		netlink_update_socket_mc(nlk, val,
2149 					 optname == NETLINK_ADD_MEMBERSHIP);
2150 		netlink_table_ungrab();
2151 		if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
2152 			nlk->netlink_unbind(sock_net(sk), val);
2153 
2154 		err = 0;
2155 		break;
2156 	}
2157 	case NETLINK_BROADCAST_ERROR:
2158 		if (val)
2159 			nlk->flags |= NETLINK_BROADCAST_SEND_ERROR;
2160 		else
2161 			nlk->flags &= ~NETLINK_BROADCAST_SEND_ERROR;
2162 		err = 0;
2163 		break;
2164 	case NETLINK_NO_ENOBUFS:
2165 		if (val) {
2166 			nlk->flags |= NETLINK_RECV_NO_ENOBUFS;
2167 			clear_bit(NETLINK_CONGESTED, &nlk->state);
2168 			wake_up_interruptible(&nlk->wait);
2169 		} else {
2170 			nlk->flags &= ~NETLINK_RECV_NO_ENOBUFS;
2171 		}
2172 		err = 0;
2173 		break;
2174 #ifdef CONFIG_NETLINK_MMAP
2175 	case NETLINK_RX_RING:
2176 	case NETLINK_TX_RING: {
2177 		struct nl_mmap_req req;
2178 
2179 		/* Rings might consume more memory than queue limits, require
2180 		 * CAP_NET_ADMIN.
2181 		 */
2182 		if (!capable(CAP_NET_ADMIN))
2183 			return -EPERM;
2184 		if (optlen < sizeof(req))
2185 			return -EINVAL;
2186 		if (copy_from_user(&req, optval, sizeof(req)))
2187 			return -EFAULT;
2188 		err = netlink_set_ring(sk, &req, false,
2189 				       optname == NETLINK_TX_RING);
2190 		break;
2191 	}
2192 #endif /* CONFIG_NETLINK_MMAP */
2193 	default:
2194 		err = -ENOPROTOOPT;
2195 	}
2196 	return err;
2197 }
2198 
2199 static int netlink_getsockopt(struct socket *sock, int level, int optname,
2200 			      char __user *optval, int __user *optlen)
2201 {
2202 	struct sock *sk = sock->sk;
2203 	struct netlink_sock *nlk = nlk_sk(sk);
2204 	int len, val, err;
2205 
2206 	if (level != SOL_NETLINK)
2207 		return -ENOPROTOOPT;
2208 
2209 	if (get_user(len, optlen))
2210 		return -EFAULT;
2211 	if (len < 0)
2212 		return -EINVAL;
2213 
2214 	switch (optname) {
2215 	case NETLINK_PKTINFO:
2216 		if (len < sizeof(int))
2217 			return -EINVAL;
2218 		len = sizeof(int);
2219 		val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0;
2220 		if (put_user(len, optlen) ||
2221 		    put_user(val, optval))
2222 			return -EFAULT;
2223 		err = 0;
2224 		break;
2225 	case NETLINK_BROADCAST_ERROR:
2226 		if (len < sizeof(int))
2227 			return -EINVAL;
2228 		len = sizeof(int);
2229 		val = nlk->flags & NETLINK_BROADCAST_SEND_ERROR ? 1 : 0;
2230 		if (put_user(len, optlen) ||
2231 		    put_user(val, optval))
2232 			return -EFAULT;
2233 		err = 0;
2234 		break;
2235 	case NETLINK_NO_ENOBUFS:
2236 		if (len < sizeof(int))
2237 			return -EINVAL;
2238 		len = sizeof(int);
2239 		val = nlk->flags & NETLINK_RECV_NO_ENOBUFS ? 1 : 0;
2240 		if (put_user(len, optlen) ||
2241 		    put_user(val, optval))
2242 			return -EFAULT;
2243 		err = 0;
2244 		break;
2245 	default:
2246 		err = -ENOPROTOOPT;
2247 	}
2248 	return err;
2249 }
2250 
2251 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
2252 {
2253 	struct nl_pktinfo info;
2254 
2255 	info.group = NETLINK_CB(skb).dst_group;
2256 	put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
2257 }
2258 
2259 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock,
2260 			   struct msghdr *msg, size_t len)
2261 {
2262 	struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
2263 	struct sock *sk = sock->sk;
2264 	struct netlink_sock *nlk = nlk_sk(sk);
2265 	DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
2266 	u32 dst_portid;
2267 	u32 dst_group;
2268 	struct sk_buff *skb;
2269 	int err;
2270 	struct scm_cookie scm;
2271 	u32 netlink_skb_flags = 0;
2272 
2273 	if (msg->msg_flags&MSG_OOB)
2274 		return -EOPNOTSUPP;
2275 
2276 	if (NULL == siocb->scm)
2277 		siocb->scm = &scm;
2278 
2279 	err = scm_send(sock, msg, siocb->scm, true);
2280 	if (err < 0)
2281 		return err;
2282 
2283 	if (msg->msg_namelen) {
2284 		err = -EINVAL;
2285 		if (addr->nl_family != AF_NETLINK)
2286 			goto out;
2287 		dst_portid = addr->nl_pid;
2288 		dst_group = ffs(addr->nl_groups);
2289 		err =  -EPERM;
2290 		if ((dst_group || dst_portid) &&
2291 		    !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
2292 			goto out;
2293 		netlink_skb_flags |= NETLINK_SKB_DST;
2294 	} else {
2295 		dst_portid = nlk->dst_portid;
2296 		dst_group = nlk->dst_group;
2297 	}
2298 
2299 	if (!nlk->portid) {
2300 		err = netlink_autobind(sock);
2301 		if (err)
2302 			goto out;
2303 	}
2304 
2305 	if (netlink_tx_is_mmaped(sk) &&
2306 	    msg->msg_iter.iov->iov_base == NULL) {
2307 		err = netlink_mmap_sendmsg(sk, msg, dst_portid, dst_group,
2308 					   siocb);
2309 		goto out;
2310 	}
2311 
2312 	err = -EMSGSIZE;
2313 	if (len > sk->sk_sndbuf - 32)
2314 		goto out;
2315 	err = -ENOBUFS;
2316 	skb = netlink_alloc_large_skb(len, dst_group);
2317 	if (skb == NULL)
2318 		goto out;
2319 
2320 	NETLINK_CB(skb).portid	= nlk->portid;
2321 	NETLINK_CB(skb).dst_group = dst_group;
2322 	NETLINK_CB(skb).creds	= siocb->scm->creds;
2323 	NETLINK_CB(skb).flags	= netlink_skb_flags;
2324 
2325 	err = -EFAULT;
2326 	if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
2327 		kfree_skb(skb);
2328 		goto out;
2329 	}
2330 
2331 	err = security_netlink_send(sk, skb);
2332 	if (err) {
2333 		kfree_skb(skb);
2334 		goto out;
2335 	}
2336 
2337 	if (dst_group) {
2338 		atomic_inc(&skb->users);
2339 		netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
2340 	}
2341 	err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags&MSG_DONTWAIT);
2342 
2343 out:
2344 	scm_destroy(siocb->scm);
2345 	return err;
2346 }
2347 
2348 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock,
2349 			   struct msghdr *msg, size_t len,
2350 			   int flags)
2351 {
2352 	struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
2353 	struct scm_cookie scm;
2354 	struct sock *sk = sock->sk;
2355 	struct netlink_sock *nlk = nlk_sk(sk);
2356 	int noblock = flags&MSG_DONTWAIT;
2357 	size_t copied;
2358 	struct sk_buff *skb, *data_skb;
2359 	int err, ret;
2360 
2361 	if (flags&MSG_OOB)
2362 		return -EOPNOTSUPP;
2363 
2364 	copied = 0;
2365 
2366 	skb = skb_recv_datagram(sk, flags, noblock, &err);
2367 	if (skb == NULL)
2368 		goto out;
2369 
2370 	data_skb = skb;
2371 
2372 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
2373 	if (unlikely(skb_shinfo(skb)->frag_list)) {
2374 		/*
2375 		 * If this skb has a frag_list, then here that means that we
2376 		 * will have to use the frag_list skb's data for compat tasks
2377 		 * and the regular skb's data for normal (non-compat) tasks.
2378 		 *
2379 		 * If we need to send the compat skb, assign it to the
2380 		 * 'data_skb' variable so that it will be used below for data
2381 		 * copying. We keep 'skb' for everything else, including
2382 		 * freeing both later.
2383 		 */
2384 		if (flags & MSG_CMSG_COMPAT)
2385 			data_skb = skb_shinfo(skb)->frag_list;
2386 	}
2387 #endif
2388 
2389 	/* Record the max length of recvmsg() calls for future allocations */
2390 	nlk->max_recvmsg_len = max(nlk->max_recvmsg_len, len);
2391 	nlk->max_recvmsg_len = min_t(size_t, nlk->max_recvmsg_len,
2392 				     16384);
2393 
2394 	copied = data_skb->len;
2395 	if (len < copied) {
2396 		msg->msg_flags |= MSG_TRUNC;
2397 		copied = len;
2398 	}
2399 
2400 	skb_reset_transport_header(data_skb);
2401 	err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
2402 
2403 	if (msg->msg_name) {
2404 		DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
2405 		addr->nl_family = AF_NETLINK;
2406 		addr->nl_pad    = 0;
2407 		addr->nl_pid	= NETLINK_CB(skb).portid;
2408 		addr->nl_groups	= netlink_group_mask(NETLINK_CB(skb).dst_group);
2409 		msg->msg_namelen = sizeof(*addr);
2410 	}
2411 
2412 	if (nlk->flags & NETLINK_RECV_PKTINFO)
2413 		netlink_cmsg_recv_pktinfo(msg, skb);
2414 
2415 	if (NULL == siocb->scm) {
2416 		memset(&scm, 0, sizeof(scm));
2417 		siocb->scm = &scm;
2418 	}
2419 	siocb->scm->creds = *NETLINK_CREDS(skb);
2420 	if (flags & MSG_TRUNC)
2421 		copied = data_skb->len;
2422 
2423 	skb_free_datagram(sk, skb);
2424 
2425 	if (nlk->cb_running &&
2426 	    atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
2427 		ret = netlink_dump(sk);
2428 		if (ret) {
2429 			sk->sk_err = -ret;
2430 			sk->sk_error_report(sk);
2431 		}
2432 	}
2433 
2434 	scm_recv(sock, msg, siocb->scm, flags);
2435 out:
2436 	netlink_rcv_wake(sk);
2437 	return err ? : copied;
2438 }
2439 
2440 static void netlink_data_ready(struct sock *sk)
2441 {
2442 	BUG();
2443 }
2444 
2445 /*
2446  *	We export these functions to other modules. They provide a
2447  *	complete set of kernel non-blocking support for message
2448  *	queueing.
2449  */
2450 
2451 struct sock *
2452 __netlink_kernel_create(struct net *net, int unit, struct module *module,
2453 			struct netlink_kernel_cfg *cfg)
2454 {
2455 	struct socket *sock;
2456 	struct sock *sk;
2457 	struct netlink_sock *nlk;
2458 	struct listeners *listeners = NULL;
2459 	struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
2460 	unsigned int groups;
2461 
2462 	BUG_ON(!nl_table);
2463 
2464 	if (unit < 0 || unit >= MAX_LINKS)
2465 		return NULL;
2466 
2467 	if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
2468 		return NULL;
2469 
2470 	/*
2471 	 * We have to just have a reference on the net from sk, but don't
2472 	 * get_net it. Besides, we cannot get and then put the net here.
2473 	 * So we create one inside init_net and the move it to net.
2474 	 */
2475 
2476 	if (__netlink_create(&init_net, sock, cb_mutex, unit) < 0)
2477 		goto out_sock_release_nosk;
2478 
2479 	sk = sock->sk;
2480 	sk_change_net(sk, net);
2481 
2482 	if (!cfg || cfg->groups < 32)
2483 		groups = 32;
2484 	else
2485 		groups = cfg->groups;
2486 
2487 	listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2488 	if (!listeners)
2489 		goto out_sock_release;
2490 
2491 	sk->sk_data_ready = netlink_data_ready;
2492 	if (cfg && cfg->input)
2493 		nlk_sk(sk)->netlink_rcv = cfg->input;
2494 
2495 	if (netlink_insert(sk, net, 0))
2496 		goto out_sock_release;
2497 
2498 	nlk = nlk_sk(sk);
2499 	nlk->flags |= NETLINK_KERNEL_SOCKET;
2500 
2501 	netlink_table_grab();
2502 	if (!nl_table[unit].registered) {
2503 		nl_table[unit].groups = groups;
2504 		rcu_assign_pointer(nl_table[unit].listeners, listeners);
2505 		nl_table[unit].cb_mutex = cb_mutex;
2506 		nl_table[unit].module = module;
2507 		if (cfg) {
2508 			nl_table[unit].bind = cfg->bind;
2509 			nl_table[unit].unbind = cfg->unbind;
2510 			nl_table[unit].flags = cfg->flags;
2511 			if (cfg->compare)
2512 				nl_table[unit].compare = cfg->compare;
2513 		}
2514 		nl_table[unit].registered = 1;
2515 	} else {
2516 		kfree(listeners);
2517 		nl_table[unit].registered++;
2518 	}
2519 	netlink_table_ungrab();
2520 	return sk;
2521 
2522 out_sock_release:
2523 	kfree(listeners);
2524 	netlink_kernel_release(sk);
2525 	return NULL;
2526 
2527 out_sock_release_nosk:
2528 	sock_release(sock);
2529 	return NULL;
2530 }
2531 EXPORT_SYMBOL(__netlink_kernel_create);
2532 
2533 void
2534 netlink_kernel_release(struct sock *sk)
2535 {
2536 	sk_release_kernel(sk);
2537 }
2538 EXPORT_SYMBOL(netlink_kernel_release);
2539 
2540 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2541 {
2542 	struct listeners *new, *old;
2543 	struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2544 
2545 	if (groups < 32)
2546 		groups = 32;
2547 
2548 	if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2549 		new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2550 		if (!new)
2551 			return -ENOMEM;
2552 		old = nl_deref_protected(tbl->listeners);
2553 		memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2554 		rcu_assign_pointer(tbl->listeners, new);
2555 
2556 		kfree_rcu(old, rcu);
2557 	}
2558 	tbl->groups = groups;
2559 
2560 	return 0;
2561 }
2562 
2563 /**
2564  * netlink_change_ngroups - change number of multicast groups
2565  *
2566  * This changes the number of multicast groups that are available
2567  * on a certain netlink family. Note that it is not possible to
2568  * change the number of groups to below 32. Also note that it does
2569  * not implicitly call netlink_clear_multicast_users() when the
2570  * number of groups is reduced.
2571  *
2572  * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2573  * @groups: The new number of groups.
2574  */
2575 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2576 {
2577 	int err;
2578 
2579 	netlink_table_grab();
2580 	err = __netlink_change_ngroups(sk, groups);
2581 	netlink_table_ungrab();
2582 
2583 	return err;
2584 }
2585 
2586 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2587 {
2588 	struct sock *sk;
2589 	struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2590 
2591 	sk_for_each_bound(sk, &tbl->mc_list)
2592 		netlink_update_socket_mc(nlk_sk(sk), group, 0);
2593 }
2594 
2595 struct nlmsghdr *
2596 __nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2597 {
2598 	struct nlmsghdr *nlh;
2599 	int size = nlmsg_msg_size(len);
2600 
2601 	nlh = (struct nlmsghdr *)skb_put(skb, NLMSG_ALIGN(size));
2602 	nlh->nlmsg_type = type;
2603 	nlh->nlmsg_len = size;
2604 	nlh->nlmsg_flags = flags;
2605 	nlh->nlmsg_pid = portid;
2606 	nlh->nlmsg_seq = seq;
2607 	if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2608 		memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2609 	return nlh;
2610 }
2611 EXPORT_SYMBOL(__nlmsg_put);
2612 
2613 /*
2614  * It looks a bit ugly.
2615  * It would be better to create kernel thread.
2616  */
2617 
2618 static int netlink_dump(struct sock *sk)
2619 {
2620 	struct netlink_sock *nlk = nlk_sk(sk);
2621 	struct netlink_callback *cb;
2622 	struct sk_buff *skb = NULL;
2623 	struct nlmsghdr *nlh;
2624 	int len, err = -ENOBUFS;
2625 	int alloc_size;
2626 
2627 	mutex_lock(nlk->cb_mutex);
2628 	if (!nlk->cb_running) {
2629 		err = -EINVAL;
2630 		goto errout_skb;
2631 	}
2632 
2633 	cb = &nlk->cb;
2634 	alloc_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2635 
2636 	if (!netlink_rx_is_mmaped(sk) &&
2637 	    atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2638 		goto errout_skb;
2639 
2640 	/* NLMSG_GOODSIZE is small to avoid high order allocations being
2641 	 * required, but it makes sense to _attempt_ a 16K bytes allocation
2642 	 * to reduce number of system calls on dump operations, if user
2643 	 * ever provided a big enough buffer.
2644 	 */
2645 	if (alloc_size < nlk->max_recvmsg_len) {
2646 		skb = netlink_alloc_skb(sk,
2647 					nlk->max_recvmsg_len,
2648 					nlk->portid,
2649 					GFP_KERNEL |
2650 					__GFP_NOWARN |
2651 					__GFP_NORETRY);
2652 		/* available room should be exact amount to avoid MSG_TRUNC */
2653 		if (skb)
2654 			skb_reserve(skb, skb_tailroom(skb) -
2655 					 nlk->max_recvmsg_len);
2656 	}
2657 	if (!skb)
2658 		skb = netlink_alloc_skb(sk, alloc_size, nlk->portid,
2659 					GFP_KERNEL);
2660 	if (!skb)
2661 		goto errout_skb;
2662 	netlink_skb_set_owner_r(skb, sk);
2663 
2664 	len = cb->dump(skb, cb);
2665 
2666 	if (len > 0) {
2667 		mutex_unlock(nlk->cb_mutex);
2668 
2669 		if (sk_filter(sk, skb))
2670 			kfree_skb(skb);
2671 		else
2672 			__netlink_sendskb(sk, skb);
2673 		return 0;
2674 	}
2675 
2676 	nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
2677 	if (!nlh)
2678 		goto errout_skb;
2679 
2680 	nl_dump_check_consistent(cb, nlh);
2681 
2682 	memcpy(nlmsg_data(nlh), &len, sizeof(len));
2683 
2684 	if (sk_filter(sk, skb))
2685 		kfree_skb(skb);
2686 	else
2687 		__netlink_sendskb(sk, skb);
2688 
2689 	if (cb->done)
2690 		cb->done(cb);
2691 
2692 	nlk->cb_running = false;
2693 	mutex_unlock(nlk->cb_mutex);
2694 	module_put(cb->module);
2695 	consume_skb(cb->skb);
2696 	return 0;
2697 
2698 errout_skb:
2699 	mutex_unlock(nlk->cb_mutex);
2700 	kfree_skb(skb);
2701 	return err;
2702 }
2703 
2704 int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2705 			 const struct nlmsghdr *nlh,
2706 			 struct netlink_dump_control *control)
2707 {
2708 	struct netlink_callback *cb;
2709 	struct sock *sk;
2710 	struct netlink_sock *nlk;
2711 	int ret;
2712 
2713 	/* Memory mapped dump requests need to be copied to avoid looping
2714 	 * on the pending state in netlink_mmap_sendmsg() while the CB hold
2715 	 * a reference to the skb.
2716 	 */
2717 	if (netlink_skb_is_mmaped(skb)) {
2718 		skb = skb_copy(skb, GFP_KERNEL);
2719 		if (skb == NULL)
2720 			return -ENOBUFS;
2721 	} else
2722 		atomic_inc(&skb->users);
2723 
2724 	sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2725 	if (sk == NULL) {
2726 		ret = -ECONNREFUSED;
2727 		goto error_free;
2728 	}
2729 
2730 	nlk = nlk_sk(sk);
2731 	mutex_lock(nlk->cb_mutex);
2732 	/* A dump is in progress... */
2733 	if (nlk->cb_running) {
2734 		ret = -EBUSY;
2735 		goto error_unlock;
2736 	}
2737 	/* add reference of module which cb->dump belongs to */
2738 	if (!try_module_get(control->module)) {
2739 		ret = -EPROTONOSUPPORT;
2740 		goto error_unlock;
2741 	}
2742 
2743 	cb = &nlk->cb;
2744 	memset(cb, 0, sizeof(*cb));
2745 	cb->dump = control->dump;
2746 	cb->done = control->done;
2747 	cb->nlh = nlh;
2748 	cb->data = control->data;
2749 	cb->module = control->module;
2750 	cb->min_dump_alloc = control->min_dump_alloc;
2751 	cb->skb = skb;
2752 
2753 	nlk->cb_running = true;
2754 
2755 	mutex_unlock(nlk->cb_mutex);
2756 
2757 	ret = netlink_dump(sk);
2758 	sock_put(sk);
2759 
2760 	if (ret)
2761 		return ret;
2762 
2763 	/* We successfully started a dump, by returning -EINTR we
2764 	 * signal not to send ACK even if it was requested.
2765 	 */
2766 	return -EINTR;
2767 
2768 error_unlock:
2769 	sock_put(sk);
2770 	mutex_unlock(nlk->cb_mutex);
2771 error_free:
2772 	kfree_skb(skb);
2773 	return ret;
2774 }
2775 EXPORT_SYMBOL(__netlink_dump_start);
2776 
2777 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
2778 {
2779 	struct sk_buff *skb;
2780 	struct nlmsghdr *rep;
2781 	struct nlmsgerr *errmsg;
2782 	size_t payload = sizeof(*errmsg);
2783 
2784 	/* error messages get the original request appened */
2785 	if (err)
2786 		payload += nlmsg_len(nlh);
2787 
2788 	skb = netlink_alloc_skb(in_skb->sk, nlmsg_total_size(payload),
2789 				NETLINK_CB(in_skb).portid, GFP_KERNEL);
2790 	if (!skb) {
2791 		struct sock *sk;
2792 
2793 		sk = netlink_lookup(sock_net(in_skb->sk),
2794 				    in_skb->sk->sk_protocol,
2795 				    NETLINK_CB(in_skb).portid);
2796 		if (sk) {
2797 			sk->sk_err = ENOBUFS;
2798 			sk->sk_error_report(sk);
2799 			sock_put(sk);
2800 		}
2801 		return;
2802 	}
2803 
2804 	rep = __nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2805 			  NLMSG_ERROR, payload, 0);
2806 	errmsg = nlmsg_data(rep);
2807 	errmsg->error = err;
2808 	memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(*nlh));
2809 	netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid, MSG_DONTWAIT);
2810 }
2811 EXPORT_SYMBOL(netlink_ack);
2812 
2813 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2814 						     struct nlmsghdr *))
2815 {
2816 	struct nlmsghdr *nlh;
2817 	int err;
2818 
2819 	while (skb->len >= nlmsg_total_size(0)) {
2820 		int msglen;
2821 
2822 		nlh = nlmsg_hdr(skb);
2823 		err = 0;
2824 
2825 		if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2826 			return 0;
2827 
2828 		/* Only requests are handled by the kernel */
2829 		if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2830 			goto ack;
2831 
2832 		/* Skip control messages */
2833 		if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2834 			goto ack;
2835 
2836 		err = cb(skb, nlh);
2837 		if (err == -EINTR)
2838 			goto skip;
2839 
2840 ack:
2841 		if (nlh->nlmsg_flags & NLM_F_ACK || err)
2842 			netlink_ack(skb, nlh, err);
2843 
2844 skip:
2845 		msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2846 		if (msglen > skb->len)
2847 			msglen = skb->len;
2848 		skb_pull(skb, msglen);
2849 	}
2850 
2851 	return 0;
2852 }
2853 EXPORT_SYMBOL(netlink_rcv_skb);
2854 
2855 /**
2856  * nlmsg_notify - send a notification netlink message
2857  * @sk: netlink socket to use
2858  * @skb: notification message
2859  * @portid: destination netlink portid for reports or 0
2860  * @group: destination multicast group or 0
2861  * @report: 1 to report back, 0 to disable
2862  * @flags: allocation flags
2863  */
2864 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2865 		 unsigned int group, int report, gfp_t flags)
2866 {
2867 	int err = 0;
2868 
2869 	if (group) {
2870 		int exclude_portid = 0;
2871 
2872 		if (report) {
2873 			atomic_inc(&skb->users);
2874 			exclude_portid = portid;
2875 		}
2876 
2877 		/* errors reported via destination sk->sk_err, but propagate
2878 		 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2879 		err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2880 	}
2881 
2882 	if (report) {
2883 		int err2;
2884 
2885 		err2 = nlmsg_unicast(sk, skb, portid);
2886 		if (!err || err == -ESRCH)
2887 			err = err2;
2888 	}
2889 
2890 	return err;
2891 }
2892 EXPORT_SYMBOL(nlmsg_notify);
2893 
2894 #ifdef CONFIG_PROC_FS
2895 struct nl_seq_iter {
2896 	struct seq_net_private p;
2897 	int link;
2898 	int hash_idx;
2899 };
2900 
2901 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos)
2902 {
2903 	struct nl_seq_iter *iter = seq->private;
2904 	int i, j;
2905 	struct netlink_sock *nlk;
2906 	struct sock *s;
2907 	loff_t off = 0;
2908 
2909 	for (i = 0; i < MAX_LINKS; i++) {
2910 		struct rhashtable *ht = &nl_table[i].hash;
2911 		const struct bucket_table *tbl = rht_dereference_rcu(ht->tbl, ht);
2912 
2913 		for (j = 0; j < tbl->size; j++) {
2914 			rht_for_each_entry_rcu(nlk, tbl->buckets[j], node) {
2915 				s = (struct sock *)nlk;
2916 
2917 				if (sock_net(s) != seq_file_net(seq))
2918 					continue;
2919 				if (off == pos) {
2920 					iter->link = i;
2921 					iter->hash_idx = j;
2922 					return s;
2923 				}
2924 				++off;
2925 			}
2926 		}
2927 	}
2928 	return NULL;
2929 }
2930 
2931 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos)
2932 	__acquires(nl_table_lock) __acquires(RCU)
2933 {
2934 	read_lock(&nl_table_lock);
2935 	rcu_read_lock();
2936 	return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2937 }
2938 
2939 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2940 {
2941 	struct rhashtable *ht;
2942 	struct netlink_sock *nlk;
2943 	struct nl_seq_iter *iter;
2944 	struct net *net;
2945 	int i, j;
2946 
2947 	++*pos;
2948 
2949 	if (v == SEQ_START_TOKEN)
2950 		return netlink_seq_socket_idx(seq, 0);
2951 
2952 	net = seq_file_net(seq);
2953 	iter = seq->private;
2954 	nlk = v;
2955 
2956 	i = iter->link;
2957 	ht = &nl_table[i].hash;
2958 	rht_for_each_entry(nlk, nlk->node.next, ht, node)
2959 		if (net_eq(sock_net((struct sock *)nlk), net))
2960 			return nlk;
2961 
2962 	j = iter->hash_idx + 1;
2963 
2964 	do {
2965 		const struct bucket_table *tbl = rht_dereference_rcu(ht->tbl, ht);
2966 
2967 		for (; j < tbl->size; j++) {
2968 			rht_for_each_entry(nlk, tbl->buckets[j], ht, node) {
2969 				if (net_eq(sock_net((struct sock *)nlk), net)) {
2970 					iter->link = i;
2971 					iter->hash_idx = j;
2972 					return nlk;
2973 				}
2974 			}
2975 		}
2976 
2977 		j = 0;
2978 	} while (++i < MAX_LINKS);
2979 
2980 	return NULL;
2981 }
2982 
2983 static void netlink_seq_stop(struct seq_file *seq, void *v)
2984 	__releases(RCU) __releases(nl_table_lock)
2985 {
2986 	rcu_read_unlock();
2987 	read_unlock(&nl_table_lock);
2988 }
2989 
2990 
2991 static int netlink_seq_show(struct seq_file *seq, void *v)
2992 {
2993 	if (v == SEQ_START_TOKEN) {
2994 		seq_puts(seq,
2995 			 "sk       Eth Pid    Groups   "
2996 			 "Rmem     Wmem     Dump     Locks     Drops     Inode\n");
2997 	} else {
2998 		struct sock *s = v;
2999 		struct netlink_sock *nlk = nlk_sk(s);
3000 
3001 		seq_printf(seq, "%pK %-3d %-6u %08x %-8d %-8d %d %-8d %-8d %-8lu\n",
3002 			   s,
3003 			   s->sk_protocol,
3004 			   nlk->portid,
3005 			   nlk->groups ? (u32)nlk->groups[0] : 0,
3006 			   sk_rmem_alloc_get(s),
3007 			   sk_wmem_alloc_get(s),
3008 			   nlk->cb_running,
3009 			   atomic_read(&s->sk_refcnt),
3010 			   atomic_read(&s->sk_drops),
3011 			   sock_i_ino(s)
3012 			);
3013 
3014 	}
3015 	return 0;
3016 }
3017 
3018 static const struct seq_operations netlink_seq_ops = {
3019 	.start  = netlink_seq_start,
3020 	.next   = netlink_seq_next,
3021 	.stop   = netlink_seq_stop,
3022 	.show   = netlink_seq_show,
3023 };
3024 
3025 
3026 static int netlink_seq_open(struct inode *inode, struct file *file)
3027 {
3028 	return seq_open_net(inode, file, &netlink_seq_ops,
3029 				sizeof(struct nl_seq_iter));
3030 }
3031 
3032 static const struct file_operations netlink_seq_fops = {
3033 	.owner		= THIS_MODULE,
3034 	.open		= netlink_seq_open,
3035 	.read		= seq_read,
3036 	.llseek		= seq_lseek,
3037 	.release	= seq_release_net,
3038 };
3039 
3040 #endif
3041 
3042 int netlink_register_notifier(struct notifier_block *nb)
3043 {
3044 	return atomic_notifier_chain_register(&netlink_chain, nb);
3045 }
3046 EXPORT_SYMBOL(netlink_register_notifier);
3047 
3048 int netlink_unregister_notifier(struct notifier_block *nb)
3049 {
3050 	return atomic_notifier_chain_unregister(&netlink_chain, nb);
3051 }
3052 EXPORT_SYMBOL(netlink_unregister_notifier);
3053 
3054 static const struct proto_ops netlink_ops = {
3055 	.family =	PF_NETLINK,
3056 	.owner =	THIS_MODULE,
3057 	.release =	netlink_release,
3058 	.bind =		netlink_bind,
3059 	.connect =	netlink_connect,
3060 	.socketpair =	sock_no_socketpair,
3061 	.accept =	sock_no_accept,
3062 	.getname =	netlink_getname,
3063 	.poll =		netlink_poll,
3064 	.ioctl =	sock_no_ioctl,
3065 	.listen =	sock_no_listen,
3066 	.shutdown =	sock_no_shutdown,
3067 	.setsockopt =	netlink_setsockopt,
3068 	.getsockopt =	netlink_getsockopt,
3069 	.sendmsg =	netlink_sendmsg,
3070 	.recvmsg =	netlink_recvmsg,
3071 	.mmap =		netlink_mmap,
3072 	.sendpage =	sock_no_sendpage,
3073 };
3074 
3075 static const struct net_proto_family netlink_family_ops = {
3076 	.family = PF_NETLINK,
3077 	.create = netlink_create,
3078 	.owner	= THIS_MODULE,	/* for consistency 8) */
3079 };
3080 
3081 static int __net_init netlink_net_init(struct net *net)
3082 {
3083 #ifdef CONFIG_PROC_FS
3084 	if (!proc_create("netlink", 0, net->proc_net, &netlink_seq_fops))
3085 		return -ENOMEM;
3086 #endif
3087 	return 0;
3088 }
3089 
3090 static void __net_exit netlink_net_exit(struct net *net)
3091 {
3092 #ifdef CONFIG_PROC_FS
3093 	remove_proc_entry("netlink", net->proc_net);
3094 #endif
3095 }
3096 
3097 static void __init netlink_add_usersock_entry(void)
3098 {
3099 	struct listeners *listeners;
3100 	int groups = 32;
3101 
3102 	listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
3103 	if (!listeners)
3104 		panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
3105 
3106 	netlink_table_grab();
3107 
3108 	nl_table[NETLINK_USERSOCK].groups = groups;
3109 	rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
3110 	nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
3111 	nl_table[NETLINK_USERSOCK].registered = 1;
3112 	nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
3113 
3114 	netlink_table_ungrab();
3115 }
3116 
3117 static struct pernet_operations __net_initdata netlink_net_ops = {
3118 	.init = netlink_net_init,
3119 	.exit = netlink_net_exit,
3120 };
3121 
3122 static int __init netlink_proto_init(void)
3123 {
3124 	int i;
3125 	int err = proto_register(&netlink_proto, 0);
3126 	struct rhashtable_params ht_params = {
3127 		.head_offset = offsetof(struct netlink_sock, node),
3128 		.key_offset = offsetof(struct netlink_sock, portid),
3129 		.key_len = sizeof(u32), /* portid */
3130 		.hashfn = jhash,
3131 		.max_shift = 16, /* 64K */
3132 		.grow_decision = rht_grow_above_75,
3133 		.shrink_decision = rht_shrink_below_30,
3134 #ifdef CONFIG_PROVE_LOCKING
3135 		.mutex_is_held = lockdep_nl_sk_hash_is_held,
3136 #endif
3137 	};
3138 
3139 	if (err != 0)
3140 		goto out;
3141 
3142 	BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb));
3143 
3144 	nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
3145 	if (!nl_table)
3146 		goto panic;
3147 
3148 	for (i = 0; i < MAX_LINKS; i++) {
3149 		if (rhashtable_init(&nl_table[i].hash, &ht_params) < 0) {
3150 			while (--i > 0)
3151 				rhashtable_destroy(&nl_table[i].hash);
3152 			kfree(nl_table);
3153 			goto panic;
3154 		}
3155 	}
3156 
3157 	INIT_LIST_HEAD(&netlink_tap_all);
3158 
3159 	netlink_add_usersock_entry();
3160 
3161 	sock_register(&netlink_family_ops);
3162 	register_pernet_subsys(&netlink_net_ops);
3163 	/* The netlink device handler may be needed early. */
3164 	rtnetlink_init();
3165 out:
3166 	return err;
3167 panic:
3168 	panic("netlink_init: Cannot allocate nl_table\n");
3169 }
3170 
3171 core_initcall(netlink_proto_init);
3172