xref: /linux/net/ipv4/raw.c (revision a8b70ccf10e38775785d9cb12ead916474549f99)
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 <linux/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/kernel.h>
50 #include <linux/export.h>
51 #include <linux/spinlock.h>
52 #include <linux/sockios.h>
53 #include <linux/socket.h>
54 #include <linux/in.h>
55 #include <linux/mroute.h>
56 #include <linux/netdevice.h>
57 #include <linux/in_route.h>
58 #include <linux/route.h>
59 #include <linux/skbuff.h>
60 #include <linux/igmp.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 #include <linux/uio.h>
82 
83 struct raw_frag_vec {
84 	struct msghdr *msg;
85 	union {
86 		struct icmphdr icmph;
87 		char c[1];
88 	} hdr;
89 	int hlen;
90 };
91 
92 struct raw_hashinfo raw_v4_hashinfo = {
93 	.lock = __RW_LOCK_UNLOCKED(raw_v4_hashinfo.lock),
94 };
95 EXPORT_SYMBOL_GPL(raw_v4_hashinfo);
96 
97 int raw_hash_sk(struct sock *sk)
98 {
99 	struct raw_hashinfo *h = sk->sk_prot->h.raw_hash;
100 	struct hlist_head *head;
101 
102 	head = &h->ht[inet_sk(sk)->inet_num & (RAW_HTABLE_SIZE - 1)];
103 
104 	write_lock_bh(&h->lock);
105 	sk_add_node(sk, head);
106 	sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
107 	write_unlock_bh(&h->lock);
108 
109 	return 0;
110 }
111 EXPORT_SYMBOL_GPL(raw_hash_sk);
112 
113 void raw_unhash_sk(struct sock *sk)
114 {
115 	struct raw_hashinfo *h = sk->sk_prot->h.raw_hash;
116 
117 	write_lock_bh(&h->lock);
118 	if (sk_del_node_init(sk))
119 		sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
120 	write_unlock_bh(&h->lock);
121 }
122 EXPORT_SYMBOL_GPL(raw_unhash_sk);
123 
124 struct sock *__raw_v4_lookup(struct net *net, struct sock *sk,
125 			     unsigned short num, __be32 raddr, __be32 laddr,
126 			     int dif, int sdif)
127 {
128 	sk_for_each_from(sk) {
129 		struct inet_sock *inet = inet_sk(sk);
130 
131 		if (net_eq(sock_net(sk), net) && inet->inet_num == num	&&
132 		    !(inet->inet_daddr && inet->inet_daddr != raddr) 	&&
133 		    !(inet->inet_rcv_saddr && inet->inet_rcv_saddr != laddr) &&
134 		    !(sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif &&
135 		      sk->sk_bound_dev_if != sdif))
136 			goto found; /* gotcha */
137 	}
138 	sk = NULL;
139 found:
140 	return sk;
141 }
142 EXPORT_SYMBOL_GPL(__raw_v4_lookup);
143 
144 /*
145  *	0 - deliver
146  *	1 - block
147  */
148 static int icmp_filter(const struct sock *sk, const struct sk_buff *skb)
149 {
150 	struct icmphdr _hdr;
151 	const struct icmphdr *hdr;
152 
153 	hdr = skb_header_pointer(skb, skb_transport_offset(skb),
154 				 sizeof(_hdr), &_hdr);
155 	if (!hdr)
156 		return 1;
157 
158 	if (hdr->type < 32) {
159 		__u32 data = raw_sk(sk)->filter.data;
160 
161 		return ((1U << hdr->type) & data) != 0;
162 	}
163 
164 	/* Do not block unknown ICMP types */
165 	return 0;
166 }
167 
168 /* IP input processing comes here for RAW socket delivery.
169  * Caller owns SKB, so we must make clones.
170  *
171  * RFC 1122: SHOULD pass TOS value up to the transport layer.
172  * -> It does. And not only TOS, but all IP header.
173  */
174 static int raw_v4_input(struct sk_buff *skb, const struct iphdr *iph, int hash)
175 {
176 	int sdif = inet_sdif(skb);
177 	struct sock *sk;
178 	struct hlist_head *head;
179 	int delivered = 0;
180 	struct net *net;
181 
182 	read_lock(&raw_v4_hashinfo.lock);
183 	head = &raw_v4_hashinfo.ht[hash];
184 	if (hlist_empty(head))
185 		goto out;
186 
187 	net = dev_net(skb->dev);
188 	sk = __raw_v4_lookup(net, __sk_head(head), iph->protocol,
189 			     iph->saddr, iph->daddr,
190 			     skb->dev->ifindex, sdif);
191 
192 	while (sk) {
193 		delivered = 1;
194 		if ((iph->protocol != IPPROTO_ICMP || !icmp_filter(sk, skb)) &&
195 		    ip_mc_sf_allow(sk, iph->daddr, iph->saddr,
196 				   skb->dev->ifindex, sdif)) {
197 			struct sk_buff *clone = skb_clone(skb, GFP_ATOMIC);
198 
199 			/* Not releasing hash table! */
200 			if (clone)
201 				raw_rcv(sk, clone);
202 		}
203 		sk = __raw_v4_lookup(net, sk_next(sk), iph->protocol,
204 				     iph->saddr, iph->daddr,
205 				     skb->dev->ifindex, sdif);
206 	}
207 out:
208 	read_unlock(&raw_v4_hashinfo.lock);
209 	return delivered;
210 }
211 
212 int raw_local_deliver(struct sk_buff *skb, int protocol)
213 {
214 	int hash;
215 	struct sock *raw_sk;
216 
217 	hash = protocol & (RAW_HTABLE_SIZE - 1);
218 	raw_sk = sk_head(&raw_v4_hashinfo.ht[hash]);
219 
220 	/* If there maybe a raw socket we must check - if not we
221 	 * don't care less
222 	 */
223 	if (raw_sk && !raw_v4_input(skb, ip_hdr(skb), hash))
224 		raw_sk = NULL;
225 
226 	return raw_sk != NULL;
227 
228 }
229 
230 static void raw_err(struct sock *sk, struct sk_buff *skb, u32 info)
231 {
232 	struct inet_sock *inet = inet_sk(sk);
233 	const int type = icmp_hdr(skb)->type;
234 	const int code = icmp_hdr(skb)->code;
235 	int err = 0;
236 	int harderr = 0;
237 
238 	if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
239 		ipv4_sk_update_pmtu(skb, sk, info);
240 	else if (type == ICMP_REDIRECT) {
241 		ipv4_sk_redirect(skb, sk);
242 		return;
243 	}
244 
245 	/* Report error on raw socket, if:
246 	   1. User requested ip_recverr.
247 	   2. Socket is connected (otherwise the error indication
248 	      is useless without ip_recverr and error is hard.
249 	 */
250 	if (!inet->recverr && sk->sk_state != TCP_ESTABLISHED)
251 		return;
252 
253 	switch (type) {
254 	default:
255 	case ICMP_TIME_EXCEEDED:
256 		err = EHOSTUNREACH;
257 		break;
258 	case ICMP_SOURCE_QUENCH:
259 		return;
260 	case ICMP_PARAMETERPROB:
261 		err = EPROTO;
262 		harderr = 1;
263 		break;
264 	case ICMP_DEST_UNREACH:
265 		err = EHOSTUNREACH;
266 		if (code > NR_ICMP_UNREACH)
267 			break;
268 		err = icmp_err_convert[code].errno;
269 		harderr = icmp_err_convert[code].fatal;
270 		if (code == ICMP_FRAG_NEEDED) {
271 			harderr = inet->pmtudisc != IP_PMTUDISC_DONT;
272 			err = EMSGSIZE;
273 		}
274 	}
275 
276 	if (inet->recverr) {
277 		const struct iphdr *iph = (const struct iphdr *)skb->data;
278 		u8 *payload = skb->data + (iph->ihl << 2);
279 
280 		if (inet->hdrincl)
281 			payload = skb->data;
282 		ip_icmp_error(sk, skb, err, 0, info, payload);
283 	}
284 
285 	if (inet->recverr || harderr) {
286 		sk->sk_err = err;
287 		sk->sk_error_report(sk);
288 	}
289 }
290 
291 void raw_icmp_error(struct sk_buff *skb, int protocol, u32 info)
292 {
293 	int hash;
294 	struct sock *raw_sk;
295 	const struct iphdr *iph;
296 	struct net *net;
297 
298 	hash = protocol & (RAW_HTABLE_SIZE - 1);
299 
300 	read_lock(&raw_v4_hashinfo.lock);
301 	raw_sk = sk_head(&raw_v4_hashinfo.ht[hash]);
302 	if (raw_sk) {
303 		int dif = skb->dev->ifindex;
304 		int sdif = inet_sdif(skb);
305 
306 		iph = (const struct iphdr *)skb->data;
307 		net = dev_net(skb->dev);
308 
309 		while ((raw_sk = __raw_v4_lookup(net, raw_sk, protocol,
310 						iph->daddr, iph->saddr,
311 						dif, sdif)) != NULL) {
312 			raw_err(raw_sk, skb, info);
313 			raw_sk = sk_next(raw_sk);
314 			iph = (const struct iphdr *)skb->data;
315 		}
316 	}
317 	read_unlock(&raw_v4_hashinfo.lock);
318 }
319 
320 static int raw_rcv_skb(struct sock *sk, struct sk_buff *skb)
321 {
322 	/* Charge it to the socket. */
323 
324 	ipv4_pktinfo_prepare(sk, skb);
325 	if (sock_queue_rcv_skb(sk, skb) < 0) {
326 		kfree_skb(skb);
327 		return NET_RX_DROP;
328 	}
329 
330 	return NET_RX_SUCCESS;
331 }
332 
333 int raw_rcv(struct sock *sk, struct sk_buff *skb)
334 {
335 	if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) {
336 		atomic_inc(&sk->sk_drops);
337 		kfree_skb(skb);
338 		return NET_RX_DROP;
339 	}
340 	nf_reset(skb);
341 
342 	skb_push(skb, skb->data - skb_network_header(skb));
343 
344 	raw_rcv_skb(sk, skb);
345 	return 0;
346 }
347 
348 static int raw_send_hdrinc(struct sock *sk, struct flowi4 *fl4,
349 			   struct msghdr *msg, size_t length,
350 			   struct rtable **rtp, unsigned int flags,
351 			   const struct sockcm_cookie *sockc)
352 {
353 	struct inet_sock *inet = inet_sk(sk);
354 	struct net *net = sock_net(sk);
355 	struct iphdr *iph;
356 	struct sk_buff *skb;
357 	unsigned int iphlen;
358 	int err;
359 	struct rtable *rt = *rtp;
360 	int hlen, tlen;
361 
362 	if (length > rt->dst.dev->mtu) {
363 		ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport,
364 			       rt->dst.dev->mtu);
365 		return -EMSGSIZE;
366 	}
367 	if (length < sizeof(struct iphdr))
368 		return -EINVAL;
369 
370 	if (flags&MSG_PROBE)
371 		goto out;
372 
373 	hlen = LL_RESERVED_SPACE(rt->dst.dev);
374 	tlen = rt->dst.dev->needed_tailroom;
375 	skb = sock_alloc_send_skb(sk,
376 				  length + hlen + tlen + 15,
377 				  flags & MSG_DONTWAIT, &err);
378 	if (!skb)
379 		goto error;
380 	skb_reserve(skb, hlen);
381 
382 	skb->priority = sk->sk_priority;
383 	skb->mark = sk->sk_mark;
384 	skb_dst_set(skb, &rt->dst);
385 	*rtp = NULL;
386 
387 	skb_reset_network_header(skb);
388 	iph = ip_hdr(skb);
389 	skb_put(skb, length);
390 
391 	skb->ip_summed = CHECKSUM_NONE;
392 
393 	sock_tx_timestamp(sk, sockc->tsflags, &skb_shinfo(skb)->tx_flags);
394 
395 	if (flags & MSG_CONFIRM)
396 		skb_set_dst_pending_confirm(skb, 1);
397 
398 	skb->transport_header = skb->network_header;
399 	err = -EFAULT;
400 	if (memcpy_from_msg(iph, msg, length))
401 		goto error_free;
402 
403 	iphlen = iph->ihl * 4;
404 
405 	/*
406 	 * We don't want to modify the ip header, but we do need to
407 	 * be sure that it won't cause problems later along the network
408 	 * stack.  Specifically we want to make sure that iph->ihl is a
409 	 * sane value.  If ihl points beyond the length of the buffer passed
410 	 * in, reject the frame as invalid
411 	 */
412 	err = -EINVAL;
413 	if (iphlen > length)
414 		goto error_free;
415 
416 	if (iphlen >= sizeof(*iph)) {
417 		if (!iph->saddr)
418 			iph->saddr = fl4->saddr;
419 		iph->check   = 0;
420 		iph->tot_len = htons(length);
421 		if (!iph->id)
422 			ip_select_ident(net, skb, NULL);
423 
424 		iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
425 		skb->transport_header += iphlen;
426 		if (iph->protocol == IPPROTO_ICMP &&
427 		    length >= iphlen + sizeof(struct icmphdr))
428 			icmp_out_count(net, ((struct icmphdr *)
429 				skb_transport_header(skb))->type);
430 	}
431 
432 	err = NF_HOOK(NFPROTO_IPV4, NF_INET_LOCAL_OUT,
433 		      net, sk, skb, NULL, rt->dst.dev,
434 		      dst_output);
435 	if (err > 0)
436 		err = net_xmit_errno(err);
437 	if (err)
438 		goto error;
439 out:
440 	return 0;
441 
442 error_free:
443 	kfree_skb(skb);
444 error:
445 	IP_INC_STATS(net, IPSTATS_MIB_OUTDISCARDS);
446 	if (err == -ENOBUFS && !inet->recverr)
447 		err = 0;
448 	return err;
449 }
450 
451 static int raw_probe_proto_opt(struct raw_frag_vec *rfv, struct flowi4 *fl4)
452 {
453 	int err;
454 
455 	if (fl4->flowi4_proto != IPPROTO_ICMP)
456 		return 0;
457 
458 	/* We only need the first two bytes. */
459 	rfv->hlen = 2;
460 
461 	err = memcpy_from_msg(rfv->hdr.c, rfv->msg, rfv->hlen);
462 	if (err)
463 		return err;
464 
465 	fl4->fl4_icmp_type = rfv->hdr.icmph.type;
466 	fl4->fl4_icmp_code = rfv->hdr.icmph.code;
467 
468 	return 0;
469 }
470 
471 static int raw_getfrag(void *from, char *to, int offset, int len, int odd,
472 		       struct sk_buff *skb)
473 {
474 	struct raw_frag_vec *rfv = from;
475 
476 	if (offset < rfv->hlen) {
477 		int copy = min(rfv->hlen - offset, len);
478 
479 		if (skb->ip_summed == CHECKSUM_PARTIAL)
480 			memcpy(to, rfv->hdr.c + offset, copy);
481 		else
482 			skb->csum = csum_block_add(
483 				skb->csum,
484 				csum_partial_copy_nocheck(rfv->hdr.c + offset,
485 							  to, copy, 0),
486 				odd);
487 
488 		odd = 0;
489 		offset += copy;
490 		to += copy;
491 		len -= copy;
492 
493 		if (!len)
494 			return 0;
495 	}
496 
497 	offset -= rfv->hlen;
498 
499 	return ip_generic_getfrag(rfv->msg, to, offset, len, odd, skb);
500 }
501 
502 static int raw_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
503 {
504 	struct inet_sock *inet = inet_sk(sk);
505 	struct net *net = sock_net(sk);
506 	struct ipcm_cookie ipc;
507 	struct rtable *rt = NULL;
508 	struct flowi4 fl4;
509 	int free = 0;
510 	__be32 daddr;
511 	__be32 saddr;
512 	u8  tos;
513 	int err;
514 	struct ip_options_data opt_copy;
515 	struct raw_frag_vec rfv;
516 	int hdrincl;
517 
518 	err = -EMSGSIZE;
519 	if (len > 0xFFFF)
520 		goto out;
521 
522 	/* hdrincl should be READ_ONCE(inet->hdrincl)
523 	 * but READ_ONCE() doesn't work with bit fields.
524 	 * Doing this indirectly yields the same result.
525 	 */
526 	hdrincl = inet->hdrincl;
527 	hdrincl = READ_ONCE(hdrincl);
528 	/*
529 	 *	Check the flags.
530 	 */
531 
532 	err = -EOPNOTSUPP;
533 	if (msg->msg_flags & MSG_OOB)	/* Mirror BSD error message */
534 		goto out;               /* compatibility */
535 
536 	/*
537 	 *	Get and verify the address.
538 	 */
539 
540 	if (msg->msg_namelen) {
541 		DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
542 		err = -EINVAL;
543 		if (msg->msg_namelen < sizeof(*usin))
544 			goto out;
545 		if (usin->sin_family != AF_INET) {
546 			pr_info_once("%s: %s forgot to set AF_INET. Fix it!\n",
547 				     __func__, current->comm);
548 			err = -EAFNOSUPPORT;
549 			if (usin->sin_family)
550 				goto out;
551 		}
552 		daddr = usin->sin_addr.s_addr;
553 		/* ANK: I did not forget to get protocol from port field.
554 		 * I just do not know, who uses this weirdness.
555 		 * IP_HDRINCL is much more convenient.
556 		 */
557 	} else {
558 		err = -EDESTADDRREQ;
559 		if (sk->sk_state != TCP_ESTABLISHED)
560 			goto out;
561 		daddr = inet->inet_daddr;
562 	}
563 
564 	ipc.sockc.tsflags = sk->sk_tsflags;
565 	ipc.addr = inet->inet_saddr;
566 	ipc.opt = NULL;
567 	ipc.tx_flags = 0;
568 	ipc.ttl = 0;
569 	ipc.tos = -1;
570 	ipc.oif = sk->sk_bound_dev_if;
571 
572 	if (msg->msg_controllen) {
573 		err = ip_cmsg_send(sk, msg, &ipc, false);
574 		if (unlikely(err)) {
575 			kfree(ipc.opt);
576 			goto out;
577 		}
578 		if (ipc.opt)
579 			free = 1;
580 	}
581 
582 	saddr = ipc.addr;
583 	ipc.addr = daddr;
584 
585 	if (!ipc.opt) {
586 		struct ip_options_rcu *inet_opt;
587 
588 		rcu_read_lock();
589 		inet_opt = rcu_dereference(inet->inet_opt);
590 		if (inet_opt) {
591 			memcpy(&opt_copy, inet_opt,
592 			       sizeof(*inet_opt) + inet_opt->opt.optlen);
593 			ipc.opt = &opt_copy.opt;
594 		}
595 		rcu_read_unlock();
596 	}
597 
598 	if (ipc.opt) {
599 		err = -EINVAL;
600 		/* Linux does not mangle headers on raw sockets,
601 		 * so that IP options + IP_HDRINCL is non-sense.
602 		 */
603 		if (hdrincl)
604 			goto done;
605 		if (ipc.opt->opt.srr) {
606 			if (!daddr)
607 				goto done;
608 			daddr = ipc.opt->opt.faddr;
609 		}
610 	}
611 	tos = get_rtconn_flags(&ipc, sk);
612 	if (msg->msg_flags & MSG_DONTROUTE)
613 		tos |= RTO_ONLINK;
614 
615 	if (ipv4_is_multicast(daddr)) {
616 		if (!ipc.oif)
617 			ipc.oif = inet->mc_index;
618 		if (!saddr)
619 			saddr = inet->mc_addr;
620 	} else if (!ipc.oif) {
621 		ipc.oif = inet->uc_index;
622 	} else if (ipv4_is_lbcast(daddr) && inet->uc_index) {
623 		/* oif is set, packet is to local broadcast and
624 		 * and uc_index is set. oif is most likely set
625 		 * by sk_bound_dev_if. If uc_index != oif check if the
626 		 * oif is an L3 master and uc_index is an L3 slave.
627 		 * If so, we want to allow the send using the uc_index.
628 		 */
629 		if (ipc.oif != inet->uc_index &&
630 		    ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk),
631 							      inet->uc_index)) {
632 			ipc.oif = inet->uc_index;
633 		}
634 	}
635 
636 	flowi4_init_output(&fl4, ipc.oif, sk->sk_mark, tos,
637 			   RT_SCOPE_UNIVERSE,
638 			   hdrincl ? IPPROTO_RAW : sk->sk_protocol,
639 			   inet_sk_flowi_flags(sk) |
640 			    (hdrincl ? FLOWI_FLAG_KNOWN_NH : 0),
641 			   daddr, saddr, 0, 0, sk->sk_uid);
642 
643 	if (!hdrincl) {
644 		rfv.msg = msg;
645 		rfv.hlen = 0;
646 
647 		err = raw_probe_proto_opt(&rfv, &fl4);
648 		if (err)
649 			goto done;
650 	}
651 
652 	security_sk_classify_flow(sk, flowi4_to_flowi(&fl4));
653 	rt = ip_route_output_flow(net, &fl4, sk);
654 	if (IS_ERR(rt)) {
655 		err = PTR_ERR(rt);
656 		rt = NULL;
657 		goto done;
658 	}
659 
660 	err = -EACCES;
661 	if (rt->rt_flags & RTCF_BROADCAST && !sock_flag(sk, SOCK_BROADCAST))
662 		goto done;
663 
664 	if (msg->msg_flags & MSG_CONFIRM)
665 		goto do_confirm;
666 back_from_confirm:
667 
668 	if (hdrincl)
669 		err = raw_send_hdrinc(sk, &fl4, msg, len,
670 				      &rt, msg->msg_flags, &ipc.sockc);
671 
672 	 else {
673 		sock_tx_timestamp(sk, ipc.sockc.tsflags, &ipc.tx_flags);
674 
675 		if (!ipc.addr)
676 			ipc.addr = fl4.daddr;
677 		lock_sock(sk);
678 		err = ip_append_data(sk, &fl4, raw_getfrag,
679 				     &rfv, len, 0,
680 				     &ipc, &rt, msg->msg_flags);
681 		if (err)
682 			ip_flush_pending_frames(sk);
683 		else if (!(msg->msg_flags & MSG_MORE)) {
684 			err = ip_push_pending_frames(sk, &fl4);
685 			if (err == -ENOBUFS && !inet->recverr)
686 				err = 0;
687 		}
688 		release_sock(sk);
689 	}
690 done:
691 	if (free)
692 		kfree(ipc.opt);
693 	ip_rt_put(rt);
694 
695 out:
696 	if (err < 0)
697 		return err;
698 	return len;
699 
700 do_confirm:
701 	if (msg->msg_flags & MSG_PROBE)
702 		dst_confirm_neigh(&rt->dst, &fl4.daddr);
703 	if (!(msg->msg_flags & MSG_PROBE) || len)
704 		goto back_from_confirm;
705 	err = 0;
706 	goto done;
707 }
708 
709 static void raw_close(struct sock *sk, long timeout)
710 {
711 	/*
712 	 * Raw sockets may have direct kernel references. Kill them.
713 	 */
714 	ip_ra_control(sk, 0, NULL);
715 
716 	sk_common_release(sk);
717 }
718 
719 static void raw_destroy(struct sock *sk)
720 {
721 	lock_sock(sk);
722 	ip_flush_pending_frames(sk);
723 	release_sock(sk);
724 }
725 
726 /* This gets rid of all the nasties in af_inet. -DaveM */
727 static int raw_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len)
728 {
729 	struct inet_sock *inet = inet_sk(sk);
730 	struct sockaddr_in *addr = (struct sockaddr_in *) uaddr;
731 	u32 tb_id = RT_TABLE_LOCAL;
732 	int ret = -EINVAL;
733 	int chk_addr_ret;
734 
735 	if (sk->sk_state != TCP_CLOSE || addr_len < sizeof(struct sockaddr_in))
736 		goto out;
737 
738 	if (sk->sk_bound_dev_if)
739 		tb_id = l3mdev_fib_table_by_index(sock_net(sk),
740 						 sk->sk_bound_dev_if) ? : tb_id;
741 
742 	chk_addr_ret = inet_addr_type_table(sock_net(sk), addr->sin_addr.s_addr,
743 					    tb_id);
744 
745 	ret = -EADDRNOTAVAIL;
746 	if (addr->sin_addr.s_addr && chk_addr_ret != RTN_LOCAL &&
747 	    chk_addr_ret != RTN_MULTICAST && chk_addr_ret != RTN_BROADCAST)
748 		goto out;
749 	inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
750 	if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
751 		inet->inet_saddr = 0;  /* Use device */
752 	sk_dst_reset(sk);
753 	ret = 0;
754 out:	return ret;
755 }
756 
757 /*
758  *	This should be easy, if there is something there
759  *	we return it, otherwise we block.
760  */
761 
762 static int raw_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
763 		       int noblock, int flags, int *addr_len)
764 {
765 	struct inet_sock *inet = inet_sk(sk);
766 	size_t copied = 0;
767 	int err = -EOPNOTSUPP;
768 	DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
769 	struct sk_buff *skb;
770 
771 	if (flags & MSG_OOB)
772 		goto out;
773 
774 	if (flags & MSG_ERRQUEUE) {
775 		err = ip_recv_error(sk, msg, len, addr_len);
776 		goto out;
777 	}
778 
779 	skb = skb_recv_datagram(sk, flags, noblock, &err);
780 	if (!skb)
781 		goto out;
782 
783 	copied = skb->len;
784 	if (len < copied) {
785 		msg->msg_flags |= MSG_TRUNC;
786 		copied = len;
787 	}
788 
789 	err = skb_copy_datagram_msg(skb, 0, msg, copied);
790 	if (err)
791 		goto done;
792 
793 	sock_recv_ts_and_drops(msg, sk, skb);
794 
795 	/* Copy the address. */
796 	if (sin) {
797 		sin->sin_family = AF_INET;
798 		sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
799 		sin->sin_port = 0;
800 		memset(&sin->sin_zero, 0, sizeof(sin->sin_zero));
801 		*addr_len = sizeof(*sin);
802 	}
803 	if (inet->cmsg_flags)
804 		ip_cmsg_recv(msg, skb);
805 	if (flags & MSG_TRUNC)
806 		copied = skb->len;
807 done:
808 	skb_free_datagram(sk, skb);
809 out:
810 	if (err)
811 		return err;
812 	return copied;
813 }
814 
815 static int raw_init(struct sock *sk)
816 {
817 	struct raw_sock *rp = raw_sk(sk);
818 
819 	if (inet_sk(sk)->inet_num == IPPROTO_ICMP)
820 		memset(&rp->filter, 0, sizeof(rp->filter));
821 	return 0;
822 }
823 
824 static int raw_seticmpfilter(struct sock *sk, char __user *optval, int optlen)
825 {
826 	if (optlen > sizeof(struct icmp_filter))
827 		optlen = sizeof(struct icmp_filter);
828 	if (copy_from_user(&raw_sk(sk)->filter, optval, optlen))
829 		return -EFAULT;
830 	return 0;
831 }
832 
833 static int raw_geticmpfilter(struct sock *sk, char __user *optval, int __user *optlen)
834 {
835 	int len, ret = -EFAULT;
836 
837 	if (get_user(len, optlen))
838 		goto out;
839 	ret = -EINVAL;
840 	if (len < 0)
841 		goto out;
842 	if (len > sizeof(struct icmp_filter))
843 		len = sizeof(struct icmp_filter);
844 	ret = -EFAULT;
845 	if (put_user(len, optlen) ||
846 	    copy_to_user(optval, &raw_sk(sk)->filter, len))
847 		goto out;
848 	ret = 0;
849 out:	return ret;
850 }
851 
852 static int do_raw_setsockopt(struct sock *sk, int level, int optname,
853 			  char __user *optval, unsigned int optlen)
854 {
855 	if (optname == ICMP_FILTER) {
856 		if (inet_sk(sk)->inet_num != IPPROTO_ICMP)
857 			return -EOPNOTSUPP;
858 		else
859 			return raw_seticmpfilter(sk, optval, optlen);
860 	}
861 	return -ENOPROTOOPT;
862 }
863 
864 static int raw_setsockopt(struct sock *sk, int level, int optname,
865 			  char __user *optval, unsigned int optlen)
866 {
867 	if (level != SOL_RAW)
868 		return ip_setsockopt(sk, level, optname, optval, optlen);
869 	return do_raw_setsockopt(sk, level, optname, optval, optlen);
870 }
871 
872 #ifdef CONFIG_COMPAT
873 static int compat_raw_setsockopt(struct sock *sk, int level, int optname,
874 				 char __user *optval, unsigned int optlen)
875 {
876 	if (level != SOL_RAW)
877 		return compat_ip_setsockopt(sk, level, optname, optval, optlen);
878 	return do_raw_setsockopt(sk, level, optname, optval, optlen);
879 }
880 #endif
881 
882 static int do_raw_getsockopt(struct sock *sk, int level, int optname,
883 			  char __user *optval, int __user *optlen)
884 {
885 	if (optname == ICMP_FILTER) {
886 		if (inet_sk(sk)->inet_num != IPPROTO_ICMP)
887 			return -EOPNOTSUPP;
888 		else
889 			return raw_geticmpfilter(sk, optval, optlen);
890 	}
891 	return -ENOPROTOOPT;
892 }
893 
894 static int raw_getsockopt(struct sock *sk, int level, int optname,
895 			  char __user *optval, int __user *optlen)
896 {
897 	if (level != SOL_RAW)
898 		return ip_getsockopt(sk, level, optname, optval, optlen);
899 	return do_raw_getsockopt(sk, level, optname, optval, optlen);
900 }
901 
902 #ifdef CONFIG_COMPAT
903 static int compat_raw_getsockopt(struct sock *sk, int level, int optname,
904 				 char __user *optval, int __user *optlen)
905 {
906 	if (level != SOL_RAW)
907 		return compat_ip_getsockopt(sk, level, optname, optval, optlen);
908 	return do_raw_getsockopt(sk, level, optname, optval, optlen);
909 }
910 #endif
911 
912 static int raw_ioctl(struct sock *sk, int cmd, unsigned long arg)
913 {
914 	switch (cmd) {
915 	case SIOCOUTQ: {
916 		int amount = sk_wmem_alloc_get(sk);
917 
918 		return put_user(amount, (int __user *)arg);
919 	}
920 	case SIOCINQ: {
921 		struct sk_buff *skb;
922 		int amount = 0;
923 
924 		spin_lock_bh(&sk->sk_receive_queue.lock);
925 		skb = skb_peek(&sk->sk_receive_queue);
926 		if (skb)
927 			amount = skb->len;
928 		spin_unlock_bh(&sk->sk_receive_queue.lock);
929 		return put_user(amount, (int __user *)arg);
930 	}
931 
932 	default:
933 #ifdef CONFIG_IP_MROUTE
934 		return ipmr_ioctl(sk, cmd, (void __user *)arg);
935 #else
936 		return -ENOIOCTLCMD;
937 #endif
938 	}
939 }
940 
941 #ifdef CONFIG_COMPAT
942 static int compat_raw_ioctl(struct sock *sk, unsigned int cmd, unsigned long arg)
943 {
944 	switch (cmd) {
945 	case SIOCOUTQ:
946 	case SIOCINQ:
947 		return -ENOIOCTLCMD;
948 	default:
949 #ifdef CONFIG_IP_MROUTE
950 		return ipmr_compat_ioctl(sk, cmd, compat_ptr(arg));
951 #else
952 		return -ENOIOCTLCMD;
953 #endif
954 	}
955 }
956 #endif
957 
958 int raw_abort(struct sock *sk, int err)
959 {
960 	lock_sock(sk);
961 
962 	sk->sk_err = err;
963 	sk->sk_error_report(sk);
964 	__udp_disconnect(sk, 0);
965 
966 	release_sock(sk);
967 
968 	return 0;
969 }
970 EXPORT_SYMBOL_GPL(raw_abort);
971 
972 struct proto raw_prot = {
973 	.name		   = "RAW",
974 	.owner		   = THIS_MODULE,
975 	.close		   = raw_close,
976 	.destroy	   = raw_destroy,
977 	.connect	   = ip4_datagram_connect,
978 	.disconnect	   = __udp_disconnect,
979 	.ioctl		   = raw_ioctl,
980 	.init		   = raw_init,
981 	.setsockopt	   = raw_setsockopt,
982 	.getsockopt	   = raw_getsockopt,
983 	.sendmsg	   = raw_sendmsg,
984 	.recvmsg	   = raw_recvmsg,
985 	.bind		   = raw_bind,
986 	.backlog_rcv	   = raw_rcv_skb,
987 	.release_cb	   = ip4_datagram_release_cb,
988 	.hash		   = raw_hash_sk,
989 	.unhash		   = raw_unhash_sk,
990 	.obj_size	   = sizeof(struct raw_sock),
991 	.useroffset	   = offsetof(struct raw_sock, filter),
992 	.usersize	   = sizeof_field(struct raw_sock, filter),
993 	.h.raw_hash	   = &raw_v4_hashinfo,
994 #ifdef CONFIG_COMPAT
995 	.compat_setsockopt = compat_raw_setsockopt,
996 	.compat_getsockopt = compat_raw_getsockopt,
997 	.compat_ioctl	   = compat_raw_ioctl,
998 #endif
999 	.diag_destroy	   = raw_abort,
1000 };
1001 
1002 #ifdef CONFIG_PROC_FS
1003 static struct sock *raw_get_first(struct seq_file *seq)
1004 {
1005 	struct sock *sk;
1006 	struct raw_iter_state *state = raw_seq_private(seq);
1007 
1008 	for (state->bucket = 0; state->bucket < RAW_HTABLE_SIZE;
1009 			++state->bucket) {
1010 		sk_for_each(sk, &state->h->ht[state->bucket])
1011 			if (sock_net(sk) == seq_file_net(seq))
1012 				goto found;
1013 	}
1014 	sk = NULL;
1015 found:
1016 	return sk;
1017 }
1018 
1019 static struct sock *raw_get_next(struct seq_file *seq, struct sock *sk)
1020 {
1021 	struct raw_iter_state *state = raw_seq_private(seq);
1022 
1023 	do {
1024 		sk = sk_next(sk);
1025 try_again:
1026 		;
1027 	} while (sk && sock_net(sk) != seq_file_net(seq));
1028 
1029 	if (!sk && ++state->bucket < RAW_HTABLE_SIZE) {
1030 		sk = sk_head(&state->h->ht[state->bucket]);
1031 		goto try_again;
1032 	}
1033 	return sk;
1034 }
1035 
1036 static struct sock *raw_get_idx(struct seq_file *seq, loff_t pos)
1037 {
1038 	struct sock *sk = raw_get_first(seq);
1039 
1040 	if (sk)
1041 		while (pos && (sk = raw_get_next(seq, sk)) != NULL)
1042 			--pos;
1043 	return pos ? NULL : sk;
1044 }
1045 
1046 void *raw_seq_start(struct seq_file *seq, loff_t *pos)
1047 {
1048 	struct raw_iter_state *state = raw_seq_private(seq);
1049 
1050 	read_lock(&state->h->lock);
1051 	return *pos ? raw_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
1052 }
1053 EXPORT_SYMBOL_GPL(raw_seq_start);
1054 
1055 void *raw_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1056 {
1057 	struct sock *sk;
1058 
1059 	if (v == SEQ_START_TOKEN)
1060 		sk = raw_get_first(seq);
1061 	else
1062 		sk = raw_get_next(seq, v);
1063 	++*pos;
1064 	return sk;
1065 }
1066 EXPORT_SYMBOL_GPL(raw_seq_next);
1067 
1068 void raw_seq_stop(struct seq_file *seq, void *v)
1069 {
1070 	struct raw_iter_state *state = raw_seq_private(seq);
1071 
1072 	read_unlock(&state->h->lock);
1073 }
1074 EXPORT_SYMBOL_GPL(raw_seq_stop);
1075 
1076 static void raw_sock_seq_show(struct seq_file *seq, struct sock *sp, int i)
1077 {
1078 	struct inet_sock *inet = inet_sk(sp);
1079 	__be32 dest = inet->inet_daddr,
1080 	       src = inet->inet_rcv_saddr;
1081 	__u16 destp = 0,
1082 	      srcp  = inet->inet_num;
1083 
1084 	seq_printf(seq, "%4d: %08X:%04X %08X:%04X"
1085 		" %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %d\n",
1086 		i, src, srcp, dest, destp, sp->sk_state,
1087 		sk_wmem_alloc_get(sp),
1088 		sk_rmem_alloc_get(sp),
1089 		0, 0L, 0,
1090 		from_kuid_munged(seq_user_ns(seq), sock_i_uid(sp)),
1091 		0, sock_i_ino(sp),
1092 		refcount_read(&sp->sk_refcnt), sp, atomic_read(&sp->sk_drops));
1093 }
1094 
1095 static int raw_seq_show(struct seq_file *seq, void *v)
1096 {
1097 	if (v == SEQ_START_TOKEN)
1098 		seq_printf(seq, "  sl  local_address rem_address   st tx_queue "
1099 				"rx_queue tr tm->when retrnsmt   uid  timeout "
1100 				"inode ref pointer drops\n");
1101 	else
1102 		raw_sock_seq_show(seq, v, raw_seq_private(seq)->bucket);
1103 	return 0;
1104 }
1105 
1106 static const struct seq_operations raw_seq_ops = {
1107 	.start = raw_seq_start,
1108 	.next  = raw_seq_next,
1109 	.stop  = raw_seq_stop,
1110 	.show  = raw_seq_show,
1111 };
1112 
1113 int raw_seq_open(struct inode *ino, struct file *file,
1114 		 struct raw_hashinfo *h, const struct seq_operations *ops)
1115 {
1116 	int err;
1117 	struct raw_iter_state *i;
1118 
1119 	err = seq_open_net(ino, file, ops, sizeof(struct raw_iter_state));
1120 	if (err < 0)
1121 		return err;
1122 
1123 	i = raw_seq_private((struct seq_file *)file->private_data);
1124 	i->h = h;
1125 	return 0;
1126 }
1127 EXPORT_SYMBOL_GPL(raw_seq_open);
1128 
1129 static int raw_v4_seq_open(struct inode *inode, struct file *file)
1130 {
1131 	return raw_seq_open(inode, file, &raw_v4_hashinfo, &raw_seq_ops);
1132 }
1133 
1134 static const struct file_operations raw_seq_fops = {
1135 	.open	 = raw_v4_seq_open,
1136 	.read	 = seq_read,
1137 	.llseek	 = seq_lseek,
1138 	.release = seq_release_net,
1139 };
1140 
1141 static __net_init int raw_init_net(struct net *net)
1142 {
1143 	if (!proc_create("raw", 0444, net->proc_net, &raw_seq_fops))
1144 		return -ENOMEM;
1145 
1146 	return 0;
1147 }
1148 
1149 static __net_exit void raw_exit_net(struct net *net)
1150 {
1151 	remove_proc_entry("raw", net->proc_net);
1152 }
1153 
1154 static __net_initdata struct pernet_operations raw_net_ops = {
1155 	.init = raw_init_net,
1156 	.exit = raw_exit_net,
1157 };
1158 
1159 int __init raw_proc_init(void)
1160 {
1161 	return register_pernet_subsys(&raw_net_ops);
1162 }
1163 
1164 void __init raw_proc_exit(void)
1165 {
1166 	unregister_pernet_subsys(&raw_net_ops);
1167 }
1168 #endif /* CONFIG_PROC_FS */
1169