xref: /linux/net/ipv4/raw.c (revision 0d456bad36d42d16022be045c8a53ddbb59ee478)
1 /*
2  * INET		An implementation of the TCP/IP protocol suite for the LINUX
3  *		operating system.  INET is implemented using the  BSD Socket
4  *		interface as the means of communication with the user level.
5  *
6  *		RAW - implementation of IP "raw" sockets.
7  *
8  * Authors:	Ross Biro
9  *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10  *
11  * Fixes:
12  *		Alan Cox	:	verify_area() fixed up
13  *		Alan Cox	:	ICMP error handling
14  *		Alan Cox	:	EMSGSIZE if you send too big a packet
15  *		Alan Cox	: 	Now uses generic datagrams and shared
16  *					skbuff library. No more peek crashes,
17  *					no more backlogs
18  *		Alan Cox	:	Checks sk->broadcast.
19  *		Alan Cox	:	Uses skb_free_datagram/skb_copy_datagram
20  *		Alan Cox	:	Raw passes ip options too
21  *		Alan Cox	:	Setsocketopt added
22  *		Alan Cox	:	Fixed error return for broadcasts
23  *		Alan Cox	:	Removed wake_up calls
24  *		Alan Cox	:	Use ttl/tos
25  *		Alan Cox	:	Cleaned up old debugging
26  *		Alan Cox	:	Use new kernel side addresses
27  *	Arnt Gulbrandsen	:	Fixed MSG_DONTROUTE in raw sockets.
28  *		Alan Cox	:	BSD style RAW socket demultiplexing.
29  *		Alan Cox	:	Beginnings of mrouted support.
30  *		Alan Cox	:	Added IP_HDRINCL option.
31  *		Alan Cox	:	Skip broadcast check if BSDism set.
32  *		David S. Miller	:	New socket lookup architecture.
33  *
34  *		This program is free software; you can redistribute it and/or
35  *		modify it under the terms of the GNU General Public License
36  *		as published by the Free Software Foundation; either version
37  *		2 of the License, or (at your option) any later version.
38  */
39 
40 #include <linux/types.h>
41 #include <linux/atomic.h>
42 #include <asm/byteorder.h>
43 #include <asm/current.h>
44 #include <asm/uaccess.h>
45 #include <asm/ioctls.h>
46 #include <linux/stddef.h>
47 #include <linux/slab.h>
48 #include <linux/errno.h>
49 #include <linux/aio.h>
50 #include <linux/kernel.h>
51 #include <linux/export.h>
52 #include <linux/spinlock.h>
53 #include <linux/sockios.h>
54 #include <linux/socket.h>
55 #include <linux/in.h>
56 #include <linux/mroute.h>
57 #include <linux/netdevice.h>
58 #include <linux/in_route.h>
59 #include <linux/route.h>
60 #include <linux/skbuff.h>
61 #include <net/net_namespace.h>
62 #include <net/dst.h>
63 #include <net/sock.h>
64 #include <linux/ip.h>
65 #include <linux/net.h>
66 #include <net/ip.h>
67 #include <net/icmp.h>
68 #include <net/udp.h>
69 #include <net/raw.h>
70 #include <net/snmp.h>
71 #include <net/tcp_states.h>
72 #include <net/inet_common.h>
73 #include <net/checksum.h>
74 #include <net/xfrm.h>
75 #include <linux/rtnetlink.h>
76 #include <linux/proc_fs.h>
77 #include <linux/seq_file.h>
78 #include <linux/netfilter.h>
79 #include <linux/netfilter_ipv4.h>
80 #include <linux/compat.h>
81 
82 static struct raw_hashinfo raw_v4_hashinfo = {
83 	.lock = __RW_LOCK_UNLOCKED(raw_v4_hashinfo.lock),
84 };
85 
86 void raw_hash_sk(struct sock *sk)
87 {
88 	struct raw_hashinfo *h = sk->sk_prot->h.raw_hash;
89 	struct hlist_head *head;
90 
91 	head = &h->ht[inet_sk(sk)->inet_num & (RAW_HTABLE_SIZE - 1)];
92 
93 	write_lock_bh(&h->lock);
94 	sk_add_node(sk, head);
95 	sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
96 	write_unlock_bh(&h->lock);
97 }
98 EXPORT_SYMBOL_GPL(raw_hash_sk);
99 
100 void raw_unhash_sk(struct sock *sk)
101 {
102 	struct raw_hashinfo *h = sk->sk_prot->h.raw_hash;
103 
104 	write_lock_bh(&h->lock);
105 	if (sk_del_node_init(sk))
106 		sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
107 	write_unlock_bh(&h->lock);
108 }
109 EXPORT_SYMBOL_GPL(raw_unhash_sk);
110 
111 static struct sock *__raw_v4_lookup(struct net *net, struct sock *sk,
112 		unsigned short num, __be32 raddr, __be32 laddr, int dif)
113 {
114 	struct hlist_node *node;
115 
116 	sk_for_each_from(sk, node) {
117 		struct inet_sock *inet = inet_sk(sk);
118 
119 		if (net_eq(sock_net(sk), net) && inet->inet_num == num	&&
120 		    !(inet->inet_daddr && inet->inet_daddr != raddr) 	&&
121 		    !(inet->inet_rcv_saddr && inet->inet_rcv_saddr != laddr) &&
122 		    !(sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif))
123 			goto found; /* gotcha */
124 	}
125 	sk = NULL;
126 found:
127 	return sk;
128 }
129 
130 /*
131  *	0 - deliver
132  *	1 - block
133  */
134 static int icmp_filter(const struct sock *sk, const struct sk_buff *skb)
135 {
136 	struct icmphdr _hdr;
137 	const struct icmphdr *hdr;
138 
139 	hdr = skb_header_pointer(skb, skb_transport_offset(skb),
140 				 sizeof(_hdr), &_hdr);
141 	if (!hdr)
142 		return 1;
143 
144 	if (hdr->type < 32) {
145 		__u32 data = raw_sk(sk)->filter.data;
146 
147 		return ((1U << hdr->type) & data) != 0;
148 	}
149 
150 	/* Do not block unknown ICMP types */
151 	return 0;
152 }
153 
154 /* IP input processing comes here for RAW socket delivery.
155  * Caller owns SKB, so we must make clones.
156  *
157  * RFC 1122: SHOULD pass TOS value up to the transport layer.
158  * -> It does. And not only TOS, but all IP header.
159  */
160 static int raw_v4_input(struct sk_buff *skb, const struct iphdr *iph, int hash)
161 {
162 	struct sock *sk;
163 	struct hlist_head *head;
164 	int delivered = 0;
165 	struct net *net;
166 
167 	read_lock(&raw_v4_hashinfo.lock);
168 	head = &raw_v4_hashinfo.ht[hash];
169 	if (hlist_empty(head))
170 		goto out;
171 
172 	net = dev_net(skb->dev);
173 	sk = __raw_v4_lookup(net, __sk_head(head), iph->protocol,
174 			     iph->saddr, iph->daddr,
175 			     skb->dev->ifindex);
176 
177 	while (sk) {
178 		delivered = 1;
179 		if (iph->protocol != IPPROTO_ICMP || !icmp_filter(sk, skb)) {
180 			struct sk_buff *clone = skb_clone(skb, GFP_ATOMIC);
181 
182 			/* Not releasing hash table! */
183 			if (clone)
184 				raw_rcv(sk, clone);
185 		}
186 		sk = __raw_v4_lookup(net, sk_next(sk), iph->protocol,
187 				     iph->saddr, iph->daddr,
188 				     skb->dev->ifindex);
189 	}
190 out:
191 	read_unlock(&raw_v4_hashinfo.lock);
192 	return delivered;
193 }
194 
195 int raw_local_deliver(struct sk_buff *skb, int protocol)
196 {
197 	int hash;
198 	struct sock *raw_sk;
199 
200 	hash = protocol & (RAW_HTABLE_SIZE - 1);
201 	raw_sk = sk_head(&raw_v4_hashinfo.ht[hash]);
202 
203 	/* If there maybe a raw socket we must check - if not we
204 	 * don't care less
205 	 */
206 	if (raw_sk && !raw_v4_input(skb, ip_hdr(skb), hash))
207 		raw_sk = NULL;
208 
209 	return raw_sk != NULL;
210 
211 }
212 
213 static void raw_err(struct sock *sk, struct sk_buff *skb, u32 info)
214 {
215 	struct inet_sock *inet = inet_sk(sk);
216 	const int type = icmp_hdr(skb)->type;
217 	const int code = icmp_hdr(skb)->code;
218 	int err = 0;
219 	int harderr = 0;
220 
221 	if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
222 		ipv4_sk_update_pmtu(skb, sk, info);
223 	else if (type == ICMP_REDIRECT)
224 		ipv4_sk_redirect(skb, sk);
225 
226 	/* Report error on raw socket, if:
227 	   1. User requested ip_recverr.
228 	   2. Socket is connected (otherwise the error indication
229 	      is useless without ip_recverr and error is hard.
230 	 */
231 	if (!inet->recverr && sk->sk_state != TCP_ESTABLISHED)
232 		return;
233 
234 	switch (type) {
235 	default:
236 	case ICMP_TIME_EXCEEDED:
237 		err = EHOSTUNREACH;
238 		break;
239 	case ICMP_SOURCE_QUENCH:
240 		return;
241 	case ICMP_PARAMETERPROB:
242 		err = EPROTO;
243 		harderr = 1;
244 		break;
245 	case ICMP_DEST_UNREACH:
246 		err = EHOSTUNREACH;
247 		if (code > NR_ICMP_UNREACH)
248 			break;
249 		err = icmp_err_convert[code].errno;
250 		harderr = icmp_err_convert[code].fatal;
251 		if (code == ICMP_FRAG_NEEDED) {
252 			harderr = inet->pmtudisc != IP_PMTUDISC_DONT;
253 			err = EMSGSIZE;
254 		}
255 	}
256 
257 	if (inet->recverr) {
258 		const struct iphdr *iph = (const struct iphdr *)skb->data;
259 		u8 *payload = skb->data + (iph->ihl << 2);
260 
261 		if (inet->hdrincl)
262 			payload = skb->data;
263 		ip_icmp_error(sk, skb, err, 0, info, payload);
264 	}
265 
266 	if (inet->recverr || harderr) {
267 		sk->sk_err = err;
268 		sk->sk_error_report(sk);
269 	}
270 }
271 
272 void raw_icmp_error(struct sk_buff *skb, int protocol, u32 info)
273 {
274 	int hash;
275 	struct sock *raw_sk;
276 	const struct iphdr *iph;
277 	struct net *net;
278 
279 	hash = protocol & (RAW_HTABLE_SIZE - 1);
280 
281 	read_lock(&raw_v4_hashinfo.lock);
282 	raw_sk = sk_head(&raw_v4_hashinfo.ht[hash]);
283 	if (raw_sk != NULL) {
284 		iph = (const struct iphdr *)skb->data;
285 		net = dev_net(skb->dev);
286 
287 		while ((raw_sk = __raw_v4_lookup(net, raw_sk, protocol,
288 						iph->daddr, iph->saddr,
289 						skb->dev->ifindex)) != NULL) {
290 			raw_err(raw_sk, skb, info);
291 			raw_sk = sk_next(raw_sk);
292 			iph = (const struct iphdr *)skb->data;
293 		}
294 	}
295 	read_unlock(&raw_v4_hashinfo.lock);
296 }
297 
298 static int raw_rcv_skb(struct sock *sk, struct sk_buff *skb)
299 {
300 	/* Charge it to the socket. */
301 
302 	ipv4_pktinfo_prepare(skb);
303 	if (sock_queue_rcv_skb(sk, skb) < 0) {
304 		kfree_skb(skb);
305 		return NET_RX_DROP;
306 	}
307 
308 	return NET_RX_SUCCESS;
309 }
310 
311 int raw_rcv(struct sock *sk, struct sk_buff *skb)
312 {
313 	if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) {
314 		atomic_inc(&sk->sk_drops);
315 		kfree_skb(skb);
316 		return NET_RX_DROP;
317 	}
318 	nf_reset(skb);
319 
320 	skb_push(skb, skb->data - skb_network_header(skb));
321 
322 	raw_rcv_skb(sk, skb);
323 	return 0;
324 }
325 
326 static int raw_send_hdrinc(struct sock *sk, struct flowi4 *fl4,
327 			   void *from, size_t length,
328 			   struct rtable **rtp,
329 			   unsigned int flags)
330 {
331 	struct inet_sock *inet = inet_sk(sk);
332 	struct net *net = sock_net(sk);
333 	struct iphdr *iph;
334 	struct sk_buff *skb;
335 	unsigned int iphlen;
336 	int err;
337 	struct rtable *rt = *rtp;
338 	int hlen, tlen;
339 
340 	if (length > rt->dst.dev->mtu) {
341 		ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport,
342 			       rt->dst.dev->mtu);
343 		return -EMSGSIZE;
344 	}
345 	if (flags&MSG_PROBE)
346 		goto out;
347 
348 	hlen = LL_RESERVED_SPACE(rt->dst.dev);
349 	tlen = rt->dst.dev->needed_tailroom;
350 	skb = sock_alloc_send_skb(sk,
351 				  length + hlen + tlen + 15,
352 				  flags & MSG_DONTWAIT, &err);
353 	if (skb == NULL)
354 		goto error;
355 	skb_reserve(skb, hlen);
356 
357 	skb->priority = sk->sk_priority;
358 	skb->mark = sk->sk_mark;
359 	skb_dst_set(skb, &rt->dst);
360 	*rtp = NULL;
361 
362 	skb_reset_network_header(skb);
363 	iph = ip_hdr(skb);
364 	skb_put(skb, length);
365 
366 	skb->ip_summed = CHECKSUM_NONE;
367 
368 	skb->transport_header = skb->network_header;
369 	err = -EFAULT;
370 	if (memcpy_fromiovecend((void *)iph, from, 0, length))
371 		goto error_free;
372 
373 	iphlen = iph->ihl * 4;
374 
375 	/*
376 	 * We don't want to modify the ip header, but we do need to
377 	 * be sure that it won't cause problems later along the network
378 	 * stack.  Specifically we want to make sure that iph->ihl is a
379 	 * sane value.  If ihl points beyond the length of the buffer passed
380 	 * in, reject the frame as invalid
381 	 */
382 	err = -EINVAL;
383 	if (iphlen > length)
384 		goto error_free;
385 
386 	if (iphlen >= sizeof(*iph)) {
387 		if (!iph->saddr)
388 			iph->saddr = fl4->saddr;
389 		iph->check   = 0;
390 		iph->tot_len = htons(length);
391 		if (!iph->id)
392 			ip_select_ident(iph, &rt->dst, NULL);
393 
394 		iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
395 	}
396 	if (iph->protocol == IPPROTO_ICMP)
397 		icmp_out_count(net, ((struct icmphdr *)
398 			skb_transport_header(skb))->type);
399 
400 	err = NF_HOOK(NFPROTO_IPV4, NF_INET_LOCAL_OUT, skb, NULL,
401 		      rt->dst.dev, dst_output);
402 	if (err > 0)
403 		err = net_xmit_errno(err);
404 	if (err)
405 		goto error;
406 out:
407 	return 0;
408 
409 error_free:
410 	kfree_skb(skb);
411 error:
412 	IP_INC_STATS(net, IPSTATS_MIB_OUTDISCARDS);
413 	if (err == -ENOBUFS && !inet->recverr)
414 		err = 0;
415 	return err;
416 }
417 
418 static int raw_probe_proto_opt(struct flowi4 *fl4, struct msghdr *msg)
419 {
420 	struct iovec *iov;
421 	u8 __user *type = NULL;
422 	u8 __user *code = NULL;
423 	int probed = 0;
424 	unsigned int i;
425 
426 	if (!msg->msg_iov)
427 		return 0;
428 
429 	for (i = 0; i < msg->msg_iovlen; i++) {
430 		iov = &msg->msg_iov[i];
431 		if (!iov)
432 			continue;
433 
434 		switch (fl4->flowi4_proto) {
435 		case IPPROTO_ICMP:
436 			/* check if one-byte field is readable or not. */
437 			if (iov->iov_base && iov->iov_len < 1)
438 				break;
439 
440 			if (!type) {
441 				type = iov->iov_base;
442 				/* check if code field is readable or not. */
443 				if (iov->iov_len > 1)
444 					code = type + 1;
445 			} else if (!code)
446 				code = iov->iov_base;
447 
448 			if (type && code) {
449 				if (get_user(fl4->fl4_icmp_type, type) ||
450 				    get_user(fl4->fl4_icmp_code, code))
451 					return -EFAULT;
452 				probed = 1;
453 			}
454 			break;
455 		default:
456 			probed = 1;
457 			break;
458 		}
459 		if (probed)
460 			break;
461 	}
462 	return 0;
463 }
464 
465 static int raw_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
466 		       size_t len)
467 {
468 	struct inet_sock *inet = inet_sk(sk);
469 	struct ipcm_cookie ipc;
470 	struct rtable *rt = NULL;
471 	struct flowi4 fl4;
472 	int free = 0;
473 	__be32 daddr;
474 	__be32 saddr;
475 	u8  tos;
476 	int err;
477 	struct ip_options_data opt_copy;
478 
479 	err = -EMSGSIZE;
480 	if (len > 0xFFFF)
481 		goto out;
482 
483 	/*
484 	 *	Check the flags.
485 	 */
486 
487 	err = -EOPNOTSUPP;
488 	if (msg->msg_flags & MSG_OOB)	/* Mirror BSD error message */
489 		goto out;               /* compatibility */
490 
491 	/*
492 	 *	Get and verify the address.
493 	 */
494 
495 	if (msg->msg_namelen) {
496 		struct sockaddr_in *usin = (struct sockaddr_in *)msg->msg_name;
497 		err = -EINVAL;
498 		if (msg->msg_namelen < sizeof(*usin))
499 			goto out;
500 		if (usin->sin_family != AF_INET) {
501 			pr_info_once("%s: %s forgot to set AF_INET. Fix it!\n",
502 				     __func__, current->comm);
503 			err = -EAFNOSUPPORT;
504 			if (usin->sin_family)
505 				goto out;
506 		}
507 		daddr = usin->sin_addr.s_addr;
508 		/* ANK: I did not forget to get protocol from port field.
509 		 * I just do not know, who uses this weirdness.
510 		 * IP_HDRINCL is much more convenient.
511 		 */
512 	} else {
513 		err = -EDESTADDRREQ;
514 		if (sk->sk_state != TCP_ESTABLISHED)
515 			goto out;
516 		daddr = inet->inet_daddr;
517 	}
518 
519 	ipc.addr = inet->inet_saddr;
520 	ipc.opt = NULL;
521 	ipc.tx_flags = 0;
522 	ipc.oif = sk->sk_bound_dev_if;
523 
524 	if (msg->msg_controllen) {
525 		err = ip_cmsg_send(sock_net(sk), msg, &ipc);
526 		if (err)
527 			goto out;
528 		if (ipc.opt)
529 			free = 1;
530 	}
531 
532 	saddr = ipc.addr;
533 	ipc.addr = daddr;
534 
535 	if (!ipc.opt) {
536 		struct ip_options_rcu *inet_opt;
537 
538 		rcu_read_lock();
539 		inet_opt = rcu_dereference(inet->inet_opt);
540 		if (inet_opt) {
541 			memcpy(&opt_copy, inet_opt,
542 			       sizeof(*inet_opt) + inet_opt->opt.optlen);
543 			ipc.opt = &opt_copy.opt;
544 		}
545 		rcu_read_unlock();
546 	}
547 
548 	if (ipc.opt) {
549 		err = -EINVAL;
550 		/* Linux does not mangle headers on raw sockets,
551 		 * so that IP options + IP_HDRINCL is non-sense.
552 		 */
553 		if (inet->hdrincl)
554 			goto done;
555 		if (ipc.opt->opt.srr) {
556 			if (!daddr)
557 				goto done;
558 			daddr = ipc.opt->opt.faddr;
559 		}
560 	}
561 	tos = RT_CONN_FLAGS(sk);
562 	if (msg->msg_flags & MSG_DONTROUTE)
563 		tos |= RTO_ONLINK;
564 
565 	if (ipv4_is_multicast(daddr)) {
566 		if (!ipc.oif)
567 			ipc.oif = inet->mc_index;
568 		if (!saddr)
569 			saddr = inet->mc_addr;
570 	} else if (!ipc.oif)
571 		ipc.oif = inet->uc_index;
572 
573 	flowi4_init_output(&fl4, ipc.oif, sk->sk_mark, tos,
574 			   RT_SCOPE_UNIVERSE,
575 			   inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol,
576 			   inet_sk_flowi_flags(sk) | FLOWI_FLAG_CAN_SLEEP,
577 			   daddr, saddr, 0, 0);
578 
579 	if (!inet->hdrincl) {
580 		err = raw_probe_proto_opt(&fl4, msg);
581 		if (err)
582 			goto done;
583 	}
584 
585 	security_sk_classify_flow(sk, flowi4_to_flowi(&fl4));
586 	rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
587 	if (IS_ERR(rt)) {
588 		err = PTR_ERR(rt);
589 		rt = NULL;
590 		goto done;
591 	}
592 
593 	err = -EACCES;
594 	if (rt->rt_flags & RTCF_BROADCAST && !sock_flag(sk, SOCK_BROADCAST))
595 		goto done;
596 
597 	if (msg->msg_flags & MSG_CONFIRM)
598 		goto do_confirm;
599 back_from_confirm:
600 
601 	if (inet->hdrincl)
602 		err = raw_send_hdrinc(sk, &fl4, msg->msg_iov, len,
603 				      &rt, msg->msg_flags);
604 
605 	 else {
606 		if (!ipc.addr)
607 			ipc.addr = fl4.daddr;
608 		lock_sock(sk);
609 		err = ip_append_data(sk, &fl4, ip_generic_getfrag,
610 				     msg->msg_iov, len, 0,
611 				     &ipc, &rt, msg->msg_flags);
612 		if (err)
613 			ip_flush_pending_frames(sk);
614 		else if (!(msg->msg_flags & MSG_MORE)) {
615 			err = ip_push_pending_frames(sk, &fl4);
616 			if (err == -ENOBUFS && !inet->recverr)
617 				err = 0;
618 		}
619 		release_sock(sk);
620 	}
621 done:
622 	if (free)
623 		kfree(ipc.opt);
624 	ip_rt_put(rt);
625 
626 out:
627 	if (err < 0)
628 		return err;
629 	return len;
630 
631 do_confirm:
632 	dst_confirm(&rt->dst);
633 	if (!(msg->msg_flags & MSG_PROBE) || len)
634 		goto back_from_confirm;
635 	err = 0;
636 	goto done;
637 }
638 
639 static void raw_close(struct sock *sk, long timeout)
640 {
641 	/*
642 	 * Raw sockets may have direct kernel references. Kill them.
643 	 */
644 	ip_ra_control(sk, 0, NULL);
645 
646 	sk_common_release(sk);
647 }
648 
649 static void raw_destroy(struct sock *sk)
650 {
651 	lock_sock(sk);
652 	ip_flush_pending_frames(sk);
653 	release_sock(sk);
654 }
655 
656 /* This gets rid of all the nasties in af_inet. -DaveM */
657 static int raw_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len)
658 {
659 	struct inet_sock *inet = inet_sk(sk);
660 	struct sockaddr_in *addr = (struct sockaddr_in *) uaddr;
661 	int ret = -EINVAL;
662 	int chk_addr_ret;
663 
664 	if (sk->sk_state != TCP_CLOSE || addr_len < sizeof(struct sockaddr_in))
665 		goto out;
666 	chk_addr_ret = inet_addr_type(sock_net(sk), addr->sin_addr.s_addr);
667 	ret = -EADDRNOTAVAIL;
668 	if (addr->sin_addr.s_addr && chk_addr_ret != RTN_LOCAL &&
669 	    chk_addr_ret != RTN_MULTICAST && chk_addr_ret != RTN_BROADCAST)
670 		goto out;
671 	inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
672 	if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
673 		inet->inet_saddr = 0;  /* Use device */
674 	sk_dst_reset(sk);
675 	ret = 0;
676 out:	return ret;
677 }
678 
679 /*
680  *	This should be easy, if there is something there
681  *	we return it, otherwise we block.
682  */
683 
684 static int raw_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
685 		       size_t len, int noblock, int flags, int *addr_len)
686 {
687 	struct inet_sock *inet = inet_sk(sk);
688 	size_t copied = 0;
689 	int err = -EOPNOTSUPP;
690 	struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name;
691 	struct sk_buff *skb;
692 
693 	if (flags & MSG_OOB)
694 		goto out;
695 
696 	if (addr_len)
697 		*addr_len = sizeof(*sin);
698 
699 	if (flags & MSG_ERRQUEUE) {
700 		err = ip_recv_error(sk, msg, len);
701 		goto out;
702 	}
703 
704 	skb = skb_recv_datagram(sk, flags, noblock, &err);
705 	if (!skb)
706 		goto out;
707 
708 	copied = skb->len;
709 	if (len < copied) {
710 		msg->msg_flags |= MSG_TRUNC;
711 		copied = len;
712 	}
713 
714 	err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
715 	if (err)
716 		goto done;
717 
718 	sock_recv_ts_and_drops(msg, sk, skb);
719 
720 	/* Copy the address. */
721 	if (sin) {
722 		sin->sin_family = AF_INET;
723 		sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
724 		sin->sin_port = 0;
725 		memset(&sin->sin_zero, 0, sizeof(sin->sin_zero));
726 	}
727 	if (inet->cmsg_flags)
728 		ip_cmsg_recv(msg, skb);
729 	if (flags & MSG_TRUNC)
730 		copied = skb->len;
731 done:
732 	skb_free_datagram(sk, skb);
733 out:
734 	if (err)
735 		return err;
736 	return copied;
737 }
738 
739 static int raw_init(struct sock *sk)
740 {
741 	struct raw_sock *rp = raw_sk(sk);
742 
743 	if (inet_sk(sk)->inet_num == IPPROTO_ICMP)
744 		memset(&rp->filter, 0, sizeof(rp->filter));
745 	return 0;
746 }
747 
748 static int raw_seticmpfilter(struct sock *sk, char __user *optval, int optlen)
749 {
750 	if (optlen > sizeof(struct icmp_filter))
751 		optlen = sizeof(struct icmp_filter);
752 	if (copy_from_user(&raw_sk(sk)->filter, optval, optlen))
753 		return -EFAULT;
754 	return 0;
755 }
756 
757 static int raw_geticmpfilter(struct sock *sk, char __user *optval, int __user *optlen)
758 {
759 	int len, ret = -EFAULT;
760 
761 	if (get_user(len, optlen))
762 		goto out;
763 	ret = -EINVAL;
764 	if (len < 0)
765 		goto out;
766 	if (len > sizeof(struct icmp_filter))
767 		len = sizeof(struct icmp_filter);
768 	ret = -EFAULT;
769 	if (put_user(len, optlen) ||
770 	    copy_to_user(optval, &raw_sk(sk)->filter, len))
771 		goto out;
772 	ret = 0;
773 out:	return ret;
774 }
775 
776 static int do_raw_setsockopt(struct sock *sk, int level, int optname,
777 			  char __user *optval, unsigned int optlen)
778 {
779 	if (optname == ICMP_FILTER) {
780 		if (inet_sk(sk)->inet_num != IPPROTO_ICMP)
781 			return -EOPNOTSUPP;
782 		else
783 			return raw_seticmpfilter(sk, optval, optlen);
784 	}
785 	return -ENOPROTOOPT;
786 }
787 
788 static int raw_setsockopt(struct sock *sk, int level, int optname,
789 			  char __user *optval, unsigned int optlen)
790 {
791 	if (level != SOL_RAW)
792 		return ip_setsockopt(sk, level, optname, optval, optlen);
793 	return do_raw_setsockopt(sk, level, optname, optval, optlen);
794 }
795 
796 #ifdef CONFIG_COMPAT
797 static int compat_raw_setsockopt(struct sock *sk, int level, int optname,
798 				 char __user *optval, unsigned int optlen)
799 {
800 	if (level != SOL_RAW)
801 		return compat_ip_setsockopt(sk, level, optname, optval, optlen);
802 	return do_raw_setsockopt(sk, level, optname, optval, optlen);
803 }
804 #endif
805 
806 static int do_raw_getsockopt(struct sock *sk, int level, int optname,
807 			  char __user *optval, int __user *optlen)
808 {
809 	if (optname == ICMP_FILTER) {
810 		if (inet_sk(sk)->inet_num != IPPROTO_ICMP)
811 			return -EOPNOTSUPP;
812 		else
813 			return raw_geticmpfilter(sk, optval, optlen);
814 	}
815 	return -ENOPROTOOPT;
816 }
817 
818 static int raw_getsockopt(struct sock *sk, int level, int optname,
819 			  char __user *optval, int __user *optlen)
820 {
821 	if (level != SOL_RAW)
822 		return ip_getsockopt(sk, level, optname, optval, optlen);
823 	return do_raw_getsockopt(sk, level, optname, optval, optlen);
824 }
825 
826 #ifdef CONFIG_COMPAT
827 static int compat_raw_getsockopt(struct sock *sk, int level, int optname,
828 				 char __user *optval, int __user *optlen)
829 {
830 	if (level != SOL_RAW)
831 		return compat_ip_getsockopt(sk, level, optname, optval, optlen);
832 	return do_raw_getsockopt(sk, level, optname, optval, optlen);
833 }
834 #endif
835 
836 static int raw_ioctl(struct sock *sk, int cmd, unsigned long arg)
837 {
838 	switch (cmd) {
839 	case SIOCOUTQ: {
840 		int amount = sk_wmem_alloc_get(sk);
841 
842 		return put_user(amount, (int __user *)arg);
843 	}
844 	case SIOCINQ: {
845 		struct sk_buff *skb;
846 		int amount = 0;
847 
848 		spin_lock_bh(&sk->sk_receive_queue.lock);
849 		skb = skb_peek(&sk->sk_receive_queue);
850 		if (skb != NULL)
851 			amount = skb->len;
852 		spin_unlock_bh(&sk->sk_receive_queue.lock);
853 		return put_user(amount, (int __user *)arg);
854 	}
855 
856 	default:
857 #ifdef CONFIG_IP_MROUTE
858 		return ipmr_ioctl(sk, cmd, (void __user *)arg);
859 #else
860 		return -ENOIOCTLCMD;
861 #endif
862 	}
863 }
864 
865 #ifdef CONFIG_COMPAT
866 static int compat_raw_ioctl(struct sock *sk, unsigned int cmd, unsigned long arg)
867 {
868 	switch (cmd) {
869 	case SIOCOUTQ:
870 	case SIOCINQ:
871 		return -ENOIOCTLCMD;
872 	default:
873 #ifdef CONFIG_IP_MROUTE
874 		return ipmr_compat_ioctl(sk, cmd, compat_ptr(arg));
875 #else
876 		return -ENOIOCTLCMD;
877 #endif
878 	}
879 }
880 #endif
881 
882 struct proto raw_prot = {
883 	.name		   = "RAW",
884 	.owner		   = THIS_MODULE,
885 	.close		   = raw_close,
886 	.destroy	   = raw_destroy,
887 	.connect	   = ip4_datagram_connect,
888 	.disconnect	   = udp_disconnect,
889 	.ioctl		   = raw_ioctl,
890 	.init		   = raw_init,
891 	.setsockopt	   = raw_setsockopt,
892 	.getsockopt	   = raw_getsockopt,
893 	.sendmsg	   = raw_sendmsg,
894 	.recvmsg	   = raw_recvmsg,
895 	.bind		   = raw_bind,
896 	.backlog_rcv	   = raw_rcv_skb,
897 	.hash		   = raw_hash_sk,
898 	.unhash		   = raw_unhash_sk,
899 	.obj_size	   = sizeof(struct raw_sock),
900 	.h.raw_hash	   = &raw_v4_hashinfo,
901 #ifdef CONFIG_COMPAT
902 	.compat_setsockopt = compat_raw_setsockopt,
903 	.compat_getsockopt = compat_raw_getsockopt,
904 	.compat_ioctl	   = compat_raw_ioctl,
905 #endif
906 };
907 
908 #ifdef CONFIG_PROC_FS
909 static struct sock *raw_get_first(struct seq_file *seq)
910 {
911 	struct sock *sk;
912 	struct raw_iter_state *state = raw_seq_private(seq);
913 
914 	for (state->bucket = 0; state->bucket < RAW_HTABLE_SIZE;
915 			++state->bucket) {
916 		struct hlist_node *node;
917 
918 		sk_for_each(sk, node, &state->h->ht[state->bucket])
919 			if (sock_net(sk) == seq_file_net(seq))
920 				goto found;
921 	}
922 	sk = NULL;
923 found:
924 	return sk;
925 }
926 
927 static struct sock *raw_get_next(struct seq_file *seq, struct sock *sk)
928 {
929 	struct raw_iter_state *state = raw_seq_private(seq);
930 
931 	do {
932 		sk = sk_next(sk);
933 try_again:
934 		;
935 	} while (sk && sock_net(sk) != seq_file_net(seq));
936 
937 	if (!sk && ++state->bucket < RAW_HTABLE_SIZE) {
938 		sk = sk_head(&state->h->ht[state->bucket]);
939 		goto try_again;
940 	}
941 	return sk;
942 }
943 
944 static struct sock *raw_get_idx(struct seq_file *seq, loff_t pos)
945 {
946 	struct sock *sk = raw_get_first(seq);
947 
948 	if (sk)
949 		while (pos && (sk = raw_get_next(seq, sk)) != NULL)
950 			--pos;
951 	return pos ? NULL : sk;
952 }
953 
954 void *raw_seq_start(struct seq_file *seq, loff_t *pos)
955 {
956 	struct raw_iter_state *state = raw_seq_private(seq);
957 
958 	read_lock(&state->h->lock);
959 	return *pos ? raw_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
960 }
961 EXPORT_SYMBOL_GPL(raw_seq_start);
962 
963 void *raw_seq_next(struct seq_file *seq, void *v, loff_t *pos)
964 {
965 	struct sock *sk;
966 
967 	if (v == SEQ_START_TOKEN)
968 		sk = raw_get_first(seq);
969 	else
970 		sk = raw_get_next(seq, v);
971 	++*pos;
972 	return sk;
973 }
974 EXPORT_SYMBOL_GPL(raw_seq_next);
975 
976 void raw_seq_stop(struct seq_file *seq, void *v)
977 {
978 	struct raw_iter_state *state = raw_seq_private(seq);
979 
980 	read_unlock(&state->h->lock);
981 }
982 EXPORT_SYMBOL_GPL(raw_seq_stop);
983 
984 static void raw_sock_seq_show(struct seq_file *seq, struct sock *sp, int i)
985 {
986 	struct inet_sock *inet = inet_sk(sp);
987 	__be32 dest = inet->inet_daddr,
988 	       src = inet->inet_rcv_saddr;
989 	__u16 destp = 0,
990 	      srcp  = inet->inet_num;
991 
992 	seq_printf(seq, "%4d: %08X:%04X %08X:%04X"
993 		" %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %pK %d\n",
994 		i, src, srcp, dest, destp, sp->sk_state,
995 		sk_wmem_alloc_get(sp),
996 		sk_rmem_alloc_get(sp),
997 		0, 0L, 0,
998 		from_kuid_munged(seq_user_ns(seq), sock_i_uid(sp)),
999 		0, sock_i_ino(sp),
1000 		atomic_read(&sp->sk_refcnt), sp, atomic_read(&sp->sk_drops));
1001 }
1002 
1003 static int raw_seq_show(struct seq_file *seq, void *v)
1004 {
1005 	if (v == SEQ_START_TOKEN)
1006 		seq_printf(seq, "  sl  local_address rem_address   st tx_queue "
1007 				"rx_queue tr tm->when retrnsmt   uid  timeout "
1008 				"inode ref pointer drops\n");
1009 	else
1010 		raw_sock_seq_show(seq, v, raw_seq_private(seq)->bucket);
1011 	return 0;
1012 }
1013 
1014 static const struct seq_operations raw_seq_ops = {
1015 	.start = raw_seq_start,
1016 	.next  = raw_seq_next,
1017 	.stop  = raw_seq_stop,
1018 	.show  = raw_seq_show,
1019 };
1020 
1021 int raw_seq_open(struct inode *ino, struct file *file,
1022 		 struct raw_hashinfo *h, const struct seq_operations *ops)
1023 {
1024 	int err;
1025 	struct raw_iter_state *i;
1026 
1027 	err = seq_open_net(ino, file, ops, sizeof(struct raw_iter_state));
1028 	if (err < 0)
1029 		return err;
1030 
1031 	i = raw_seq_private((struct seq_file *)file->private_data);
1032 	i->h = h;
1033 	return 0;
1034 }
1035 EXPORT_SYMBOL_GPL(raw_seq_open);
1036 
1037 static int raw_v4_seq_open(struct inode *inode, struct file *file)
1038 {
1039 	return raw_seq_open(inode, file, &raw_v4_hashinfo, &raw_seq_ops);
1040 }
1041 
1042 static const struct file_operations raw_seq_fops = {
1043 	.owner	 = THIS_MODULE,
1044 	.open	 = raw_v4_seq_open,
1045 	.read	 = seq_read,
1046 	.llseek	 = seq_lseek,
1047 	.release = seq_release_net,
1048 };
1049 
1050 static __net_init int raw_init_net(struct net *net)
1051 {
1052 	if (!proc_net_fops_create(net, "raw", S_IRUGO, &raw_seq_fops))
1053 		return -ENOMEM;
1054 
1055 	return 0;
1056 }
1057 
1058 static __net_exit void raw_exit_net(struct net *net)
1059 {
1060 	proc_net_remove(net, "raw");
1061 }
1062 
1063 static __net_initdata struct pernet_operations raw_net_ops = {
1064 	.init = raw_init_net,
1065 	.exit = raw_exit_net,
1066 };
1067 
1068 int __init raw_proc_init(void)
1069 {
1070 	return register_pernet_subsys(&raw_net_ops);
1071 }
1072 
1073 void __init raw_proc_exit(void)
1074 {
1075 	unregister_pernet_subsys(&raw_net_ops);
1076 }
1077 #endif /* CONFIG_PROC_FS */
1078