xref: /linux/net/ipv4/raw.c (revision fe833e4397fbdc3ae13a60202dfc7c335b032499)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * INET		An implementation of the TCP/IP protocol suite for the LINUX
4  *		operating system.  INET is implemented using the  BSD Socket
5  *		interface as the means of communication with the user level.
6  *
7  *		RAW - implementation of IP "raw" sockets.
8  *
9  * Authors:	Ross Biro
10  *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
11  *
12  * Fixes:
13  *		Alan Cox	:	verify_area() fixed up
14  *		Alan Cox	:	ICMP error handling
15  *		Alan Cox	:	EMSGSIZE if you send too big a packet
16  *		Alan Cox	: 	Now uses generic datagrams and shared
17  *					skbuff library. No more peek crashes,
18  *					no more backlogs
19  *		Alan Cox	:	Checks sk->broadcast.
20  *		Alan Cox	:	Uses skb_free_datagram/skb_copy_datagram
21  *		Alan Cox	:	Raw passes ip options too
22  *		Alan Cox	:	Setsocketopt added
23  *		Alan Cox	:	Fixed error return for broadcasts
24  *		Alan Cox	:	Removed wake_up calls
25  *		Alan Cox	:	Use ttl/tos
26  *		Alan Cox	:	Cleaned up old debugging
27  *		Alan Cox	:	Use new kernel side addresses
28  *	Arnt Gulbrandsen	:	Fixed MSG_DONTROUTE in raw sockets.
29  *		Alan Cox	:	BSD style RAW socket demultiplexing.
30  *		Alan Cox	:	Beginnings of mrouted support.
31  *		Alan Cox	:	Added IP_HDRINCL option.
32  *		Alan Cox	:	Skip broadcast check if BSDism set.
33  *		David S. Miller	:	New socket lookup architecture.
34  */
35 
36 #include <linux/types.h>
37 #include <linux/atomic.h>
38 #include <asm/byteorder.h>
39 #include <asm/current.h>
40 #include <linux/uaccess.h>
41 #include <asm/ioctls.h>
42 #include <linux/stddef.h>
43 #include <linux/slab.h>
44 #include <linux/errno.h>
45 #include <linux/kernel.h>
46 #include <linux/export.h>
47 #include <linux/spinlock.h>
48 #include <linux/sockios.h>
49 #include <linux/socket.h>
50 #include <linux/in.h>
51 #include <linux/mroute.h>
52 #include <linux/netdevice.h>
53 #include <linux/in_route.h>
54 #include <linux/route.h>
55 #include <linux/skbuff.h>
56 #include <linux/igmp.h>
57 #include <net/net_namespace.h>
58 #include <net/dst.h>
59 #include <net/sock.h>
60 #include <linux/ip.h>
61 #include <linux/net.h>
62 #include <net/ip.h>
63 #include <net/icmp.h>
64 #include <net/udp.h>
65 #include <net/raw.h>
66 #include <net/snmp.h>
67 #include <net/tcp_states.h>
68 #include <net/inet_common.h>
69 #include <net/checksum.h>
70 #include <net/xfrm.h>
71 #include <linux/rtnetlink.h>
72 #include <linux/proc_fs.h>
73 #include <linux/seq_file.h>
74 #include <linux/netfilter.h>
75 #include <linux/netfilter_ipv4.h>
76 #include <linux/compat.h>
77 #include <linux/uio.h>
78 
79 struct raw_frag_vec {
80 	struct msghdr *msg;
81 	union {
82 		struct icmphdr icmph;
83 		char c[1];
84 	} hdr;
85 	int hlen;
86 };
87 
88 struct raw_hashinfo raw_v4_hashinfo;
89 EXPORT_SYMBOL_GPL(raw_v4_hashinfo);
90 
91 int raw_hash_sk(struct sock *sk)
92 {
93 	struct raw_hashinfo *h = sk->sk_prot->h.raw_hash;
94 	struct hlist_head *hlist;
95 
96 	hlist = &h->ht[raw_hashfunc(sock_net(sk), inet_sk(sk)->inet_num)];
97 
98 	spin_lock(&h->lock);
99 	sk_add_node_rcu(sk, hlist);
100 	sock_set_flag(sk, SOCK_RCU_FREE);
101 	spin_unlock(&h->lock);
102 	sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
103 
104 	return 0;
105 }
106 EXPORT_SYMBOL_GPL(raw_hash_sk);
107 
108 void raw_unhash_sk(struct sock *sk)
109 {
110 	struct raw_hashinfo *h = sk->sk_prot->h.raw_hash;
111 
112 	spin_lock(&h->lock);
113 	if (sk_del_node_init_rcu(sk))
114 		sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
115 	spin_unlock(&h->lock);
116 }
117 EXPORT_SYMBOL_GPL(raw_unhash_sk);
118 
119 bool raw_v4_match(struct net *net, const struct sock *sk, unsigned short num,
120 		  __be32 raddr, __be32 laddr, int dif, int sdif)
121 {
122 	const struct inet_sock *inet = inet_sk(sk);
123 
124 	if (net_eq(sock_net(sk), net) && inet->inet_num == num	&&
125 	    !(inet->inet_daddr && inet->inet_daddr != raddr) 	&&
126 	    !(inet->inet_rcv_saddr && inet->inet_rcv_saddr != laddr) &&
127 	    raw_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif))
128 		return true;
129 	return false;
130 }
131 EXPORT_SYMBOL_GPL(raw_v4_match);
132 
133 /*
134  *	0 - deliver
135  *	1 - block
136  */
137 static int icmp_filter(const struct sock *sk, const struct sk_buff *skb)
138 {
139 	struct icmphdr _hdr;
140 	const struct icmphdr *hdr;
141 
142 	hdr = skb_header_pointer(skb, skb_transport_offset(skb),
143 				 sizeof(_hdr), &_hdr);
144 	if (!hdr)
145 		return 1;
146 
147 	if (hdr->type < 32) {
148 		__u32 data = raw_sk(sk)->filter.data;
149 
150 		return ((1U << hdr->type) & data) != 0;
151 	}
152 
153 	/* Do not block unknown ICMP types */
154 	return 0;
155 }
156 
157 /* IP input processing comes here for RAW socket delivery.
158  * Caller owns SKB, so we must make clones.
159  *
160  * RFC 1122: SHOULD pass TOS value up to the transport layer.
161  * -> It does. And not only TOS, but all IP header.
162  */
163 static int raw_v4_input(struct net *net, struct sk_buff *skb,
164 			const struct iphdr *iph, int hash)
165 {
166 	int sdif = inet_sdif(skb);
167 	struct hlist_head *hlist;
168 	int dif = inet_iif(skb);
169 	int delivered = 0;
170 	struct sock *sk;
171 
172 	hlist = &raw_v4_hashinfo.ht[hash];
173 	rcu_read_lock();
174 	sk_for_each_rcu(sk, hlist) {
175 		if (!raw_v4_match(net, sk, iph->protocol,
176 				  iph->saddr, iph->daddr, dif, sdif))
177 			continue;
178 
179 		if (atomic_read(&sk->sk_rmem_alloc) >=
180 		    READ_ONCE(sk->sk_rcvbuf)) {
181 			atomic_inc(&sk->sk_drops);
182 			continue;
183 		}
184 
185 		delivered = 1;
186 		if ((iph->protocol != IPPROTO_ICMP || !icmp_filter(sk, skb)) &&
187 		    ip_mc_sf_allow(sk, iph->daddr, iph->saddr,
188 				   skb->dev->ifindex, sdif)) {
189 			struct sk_buff *clone = skb_clone(skb, GFP_ATOMIC);
190 
191 			/* Not releasing hash table! */
192 			if (clone)
193 				raw_rcv(sk, clone);
194 		}
195 	}
196 	rcu_read_unlock();
197 	return delivered;
198 }
199 
200 int raw_local_deliver(struct sk_buff *skb, int protocol)
201 {
202 	struct net *net = dev_net(skb->dev);
203 
204 	return raw_v4_input(net, skb, ip_hdr(skb),
205 			    raw_hashfunc(net, protocol));
206 }
207 
208 static void raw_err(struct sock *sk, struct sk_buff *skb, u32 info)
209 {
210 	struct inet_sock *inet = inet_sk(sk);
211 	const int type = icmp_hdr(skb)->type;
212 	const int code = icmp_hdr(skb)->code;
213 	int harderr = 0;
214 	bool recverr;
215 	int err = 0;
216 
217 	if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
218 		ipv4_sk_update_pmtu(skb, sk, info);
219 	else if (type == ICMP_REDIRECT) {
220 		ipv4_sk_redirect(skb, sk);
221 		return;
222 	}
223 
224 	/* Report error on raw socket, if:
225 	   1. User requested ip_recverr.
226 	   2. Socket is connected (otherwise the error indication
227 	      is useless without ip_recverr and error is hard.
228 	 */
229 	recverr = inet_test_bit(RECVERR, sk);
230 	if (!recverr && sk->sk_state != TCP_ESTABLISHED)
231 		return;
232 
233 	switch (type) {
234 	default:
235 	case ICMP_TIME_EXCEEDED:
236 		err = EHOSTUNREACH;
237 		break;
238 	case ICMP_SOURCE_QUENCH:
239 		return;
240 	case ICMP_PARAMETERPROB:
241 		err = EPROTO;
242 		harderr = 1;
243 		break;
244 	case ICMP_DEST_UNREACH:
245 		err = EHOSTUNREACH;
246 		if (code > NR_ICMP_UNREACH)
247 			break;
248 		if (code == ICMP_FRAG_NEEDED) {
249 			harderr = READ_ONCE(inet->pmtudisc) != IP_PMTUDISC_DONT;
250 			err = EMSGSIZE;
251 		} else {
252 			err = icmp_err_convert[code].errno;
253 			harderr = icmp_err_convert[code].fatal;
254 		}
255 	}
256 
257 	if (recverr) {
258 		const struct iphdr *iph = (const struct iphdr *)skb->data;
259 		u8 *payload = skb->data + (iph->ihl << 2);
260 
261 		if (inet_test_bit(HDRINCL, sk))
262 			payload = skb->data;
263 		ip_icmp_error(sk, skb, err, 0, info, payload);
264 	}
265 
266 	if (recverr || harderr) {
267 		sk->sk_err = err;
268 		sk_error_report(sk);
269 	}
270 }
271 
272 void raw_icmp_error(struct sk_buff *skb, int protocol, u32 info)
273 {
274 	struct net *net = dev_net(skb->dev);
275 	int dif = skb->dev->ifindex;
276 	int sdif = inet_sdif(skb);
277 	struct hlist_head *hlist;
278 	const struct iphdr *iph;
279 	struct sock *sk;
280 	int hash;
281 
282 	hash = raw_hashfunc(net, protocol);
283 	hlist = &raw_v4_hashinfo.ht[hash];
284 
285 	rcu_read_lock();
286 	sk_for_each_rcu(sk, hlist) {
287 		iph = (const struct iphdr *)skb->data;
288 		if (!raw_v4_match(net, sk, iph->protocol,
289 				  iph->daddr, iph->saddr, dif, sdif))
290 			continue;
291 		raw_err(sk, skb, info);
292 	}
293 	rcu_read_unlock();
294 }
295 
296 static int raw_rcv_skb(struct sock *sk, struct sk_buff *skb)
297 {
298 	enum skb_drop_reason reason;
299 
300 	/* Charge it to the socket. */
301 
302 	ipv4_pktinfo_prepare(sk, skb, true);
303 	if (sock_queue_rcv_skb_reason(sk, skb, &reason) < 0) {
304 		kfree_skb_reason(skb, reason);
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_reason(skb, SKB_DROP_REASON_XFRM_POLICY);
316 		return NET_RX_DROP;
317 	}
318 	nf_reset_ct(skb);
319 
320 	skb_push(skb, -skb_network_offset(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 			   struct msghdr *msg, size_t length,
328 			   struct rtable **rtp, unsigned int flags,
329 			   const struct sockcm_cookie *sockc)
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 (length < sizeof(struct iphdr))
346 		return -EINVAL;
347 
348 	if (flags&MSG_PROBE)
349 		goto out;
350 
351 	hlen = LL_RESERVED_SPACE(rt->dst.dev);
352 	tlen = rt->dst.dev->needed_tailroom;
353 	skb = sock_alloc_send_skb(sk,
354 				  length + hlen + tlen + 15,
355 				  flags & MSG_DONTWAIT, &err);
356 	if (!skb)
357 		goto error;
358 	skb_reserve(skb, hlen);
359 
360 	skb->protocol = htons(ETH_P_IP);
361 	skb->priority = READ_ONCE(sk->sk_priority);
362 	skb->mark = sockc->mark;
363 	skb->tstamp = sockc->transmit_time;
364 	skb_dst_set(skb, &rt->dst);
365 	*rtp = NULL;
366 
367 	skb_reset_network_header(skb);
368 	iph = ip_hdr(skb);
369 	skb_put(skb, length);
370 
371 	skb->ip_summed = CHECKSUM_NONE;
372 
373 	skb_setup_tx_timestamp(skb, sockc->tsflags);
374 
375 	if (flags & MSG_CONFIRM)
376 		skb_set_dst_pending_confirm(skb, 1);
377 
378 	skb->transport_header = skb->network_header;
379 	err = -EFAULT;
380 	if (memcpy_from_msg(iph, msg, length))
381 		goto error_free;
382 
383 	iphlen = iph->ihl * 4;
384 
385 	/*
386 	 * We don't want to modify the ip header, but we do need to
387 	 * be sure that it won't cause problems later along the network
388 	 * stack.  Specifically we want to make sure that iph->ihl is a
389 	 * sane value.  If ihl points beyond the length of the buffer passed
390 	 * in, reject the frame as invalid
391 	 */
392 	err = -EINVAL;
393 	if (iphlen > length)
394 		goto error_free;
395 
396 	if (iphlen >= sizeof(*iph)) {
397 		if (!iph->saddr)
398 			iph->saddr = fl4->saddr;
399 		iph->check   = 0;
400 		iph->tot_len = htons(length);
401 		if (!iph->id)
402 			ip_select_ident(net, skb, NULL);
403 
404 		iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
405 		skb->transport_header += iphlen;
406 		if (iph->protocol == IPPROTO_ICMP &&
407 		    length >= iphlen + sizeof(struct icmphdr))
408 			icmp_out_count(net, ((struct icmphdr *)
409 				skb_transport_header(skb))->type);
410 	}
411 
412 	err = NF_HOOK(NFPROTO_IPV4, NF_INET_LOCAL_OUT,
413 		      net, sk, skb, NULL, rt->dst.dev,
414 		      dst_output);
415 	if (err > 0)
416 		err = net_xmit_errno(err);
417 	if (err)
418 		goto error;
419 out:
420 	return 0;
421 
422 error_free:
423 	kfree_skb(skb);
424 error:
425 	IP_INC_STATS(net, IPSTATS_MIB_OUTDISCARDS);
426 	if (err == -ENOBUFS && !inet_test_bit(RECVERR, sk))
427 		err = 0;
428 	return err;
429 }
430 
431 static int raw_probe_proto_opt(struct raw_frag_vec *rfv, struct flowi4 *fl4)
432 {
433 	int err;
434 
435 	if (fl4->flowi4_proto != IPPROTO_ICMP)
436 		return 0;
437 
438 	/* We only need the first two bytes. */
439 	rfv->hlen = 2;
440 
441 	err = memcpy_from_msg(rfv->hdr.c, rfv->msg, rfv->hlen);
442 	if (err)
443 		return err;
444 
445 	fl4->fl4_icmp_type = rfv->hdr.icmph.type;
446 	fl4->fl4_icmp_code = rfv->hdr.icmph.code;
447 
448 	return 0;
449 }
450 
451 static int raw_getfrag(void *from, char *to, int offset, int len, int odd,
452 		       struct sk_buff *skb)
453 {
454 	struct raw_frag_vec *rfv = from;
455 
456 	if (offset < rfv->hlen) {
457 		int copy = min(rfv->hlen - offset, len);
458 
459 		if (skb->ip_summed == CHECKSUM_PARTIAL)
460 			memcpy(to, rfv->hdr.c + offset, copy);
461 		else
462 			skb->csum = csum_block_add(
463 				skb->csum,
464 				csum_partial_copy_nocheck(rfv->hdr.c + offset,
465 							  to, copy),
466 				odd);
467 
468 		odd = 0;
469 		offset += copy;
470 		to += copy;
471 		len -= copy;
472 
473 		if (!len)
474 			return 0;
475 	}
476 
477 	offset -= rfv->hlen;
478 
479 	return ip_generic_getfrag(rfv->msg, to, offset, len, odd, skb);
480 }
481 
482 static int raw_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
483 {
484 	struct inet_sock *inet = inet_sk(sk);
485 	struct net *net = sock_net(sk);
486 	struct ipcm_cookie ipc;
487 	struct rtable *rt = NULL;
488 	struct flowi4 fl4;
489 	u8 tos, scope;
490 	int free = 0;
491 	__be32 daddr;
492 	__be32 saddr;
493 	int uc_index, err;
494 	struct ip_options_data opt_copy;
495 	struct raw_frag_vec rfv;
496 	int hdrincl;
497 
498 	err = -EMSGSIZE;
499 	if (len > 0xFFFF)
500 		goto out;
501 
502 	hdrincl = inet_test_bit(HDRINCL, sk);
503 
504 	/*
505 	 *	Check the flags.
506 	 */
507 
508 	err = -EOPNOTSUPP;
509 	if (msg->msg_flags & MSG_OOB)	/* Mirror BSD error message */
510 		goto out;               /* compatibility */
511 
512 	/*
513 	 *	Get and verify the address.
514 	 */
515 
516 	if (msg->msg_namelen) {
517 		DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
518 		err = -EINVAL;
519 		if (msg->msg_namelen < sizeof(*usin))
520 			goto out;
521 		if (usin->sin_family != AF_INET) {
522 			pr_info_once("%s: %s forgot to set AF_INET. Fix it!\n",
523 				     __func__, current->comm);
524 			err = -EAFNOSUPPORT;
525 			if (usin->sin_family)
526 				goto out;
527 		}
528 		daddr = usin->sin_addr.s_addr;
529 		/* ANK: I did not forget to get protocol from port field.
530 		 * I just do not know, who uses this weirdness.
531 		 * IP_HDRINCL is much more convenient.
532 		 */
533 	} else {
534 		err = -EDESTADDRREQ;
535 		if (sk->sk_state != TCP_ESTABLISHED)
536 			goto out;
537 		daddr = inet->inet_daddr;
538 	}
539 
540 	ipcm_init_sk(&ipc, inet);
541 	/* Keep backward compat */
542 	if (hdrincl)
543 		ipc.protocol = IPPROTO_RAW;
544 
545 	if (msg->msg_controllen) {
546 		err = ip_cmsg_send(sk, msg, &ipc, false);
547 		if (unlikely(err)) {
548 			kfree(ipc.opt);
549 			goto out;
550 		}
551 		if (ipc.opt)
552 			free = 1;
553 	}
554 
555 	saddr = ipc.addr;
556 	ipc.addr = daddr;
557 
558 	if (!ipc.opt) {
559 		struct ip_options_rcu *inet_opt;
560 
561 		rcu_read_lock();
562 		inet_opt = rcu_dereference(inet->inet_opt);
563 		if (inet_opt) {
564 			memcpy(&opt_copy, inet_opt,
565 			       sizeof(*inet_opt) + inet_opt->opt.optlen);
566 			ipc.opt = &opt_copy.opt;
567 		}
568 		rcu_read_unlock();
569 	}
570 
571 	if (ipc.opt) {
572 		err = -EINVAL;
573 		/* Linux does not mangle headers on raw sockets,
574 		 * so that IP options + IP_HDRINCL is non-sense.
575 		 */
576 		if (hdrincl)
577 			goto done;
578 		if (ipc.opt->opt.srr) {
579 			if (!daddr)
580 				goto done;
581 			daddr = ipc.opt->opt.faddr;
582 		}
583 	}
584 	tos = get_rttos(&ipc, inet);
585 	scope = ip_sendmsg_scope(inet, &ipc, msg);
586 
587 	uc_index = READ_ONCE(inet->uc_index);
588 	if (ipv4_is_multicast(daddr)) {
589 		if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif))
590 			ipc.oif = READ_ONCE(inet->mc_index);
591 		if (!saddr)
592 			saddr = READ_ONCE(inet->mc_addr);
593 	} else if (!ipc.oif) {
594 		ipc.oif = uc_index;
595 	} else if (ipv4_is_lbcast(daddr) && uc_index) {
596 		/* oif is set, packet is to local broadcast
597 		 * and uc_index is set. oif is most likely set
598 		 * by sk_bound_dev_if. If uc_index != oif check if the
599 		 * oif is an L3 master and uc_index is an L3 slave.
600 		 * If so, we want to allow the send using the uc_index.
601 		 */
602 		if (ipc.oif != uc_index &&
603 		    ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk),
604 							      uc_index)) {
605 			ipc.oif = uc_index;
606 		}
607 	}
608 
609 	flowi4_init_output(&fl4, ipc.oif, ipc.sockc.mark, tos, scope,
610 			   hdrincl ? ipc.protocol : sk->sk_protocol,
611 			   inet_sk_flowi_flags(sk) |
612 			    (hdrincl ? FLOWI_FLAG_KNOWN_NH : 0),
613 			   daddr, saddr, 0, 0, sk->sk_uid);
614 
615 	fl4.fl4_icmp_type = 0;
616 	fl4.fl4_icmp_code = 0;
617 
618 	if (!hdrincl) {
619 		rfv.msg = msg;
620 		rfv.hlen = 0;
621 
622 		err = raw_probe_proto_opt(&rfv, &fl4);
623 		if (err)
624 			goto done;
625 	}
626 
627 	security_sk_classify_flow(sk, flowi4_to_flowi_common(&fl4));
628 	rt = ip_route_output_flow(net, &fl4, sk);
629 	if (IS_ERR(rt)) {
630 		err = PTR_ERR(rt);
631 		rt = NULL;
632 		goto done;
633 	}
634 
635 	err = -EACCES;
636 	if (rt->rt_flags & RTCF_BROADCAST && !sock_flag(sk, SOCK_BROADCAST))
637 		goto done;
638 
639 	if (msg->msg_flags & MSG_CONFIRM)
640 		goto do_confirm;
641 back_from_confirm:
642 
643 	if (hdrincl)
644 		err = raw_send_hdrinc(sk, &fl4, msg, len,
645 				      &rt, msg->msg_flags, &ipc.sockc);
646 
647 	 else {
648 		if (!ipc.addr)
649 			ipc.addr = fl4.daddr;
650 		lock_sock(sk);
651 		err = ip_append_data(sk, &fl4, raw_getfrag,
652 				     &rfv, len, 0,
653 				     &ipc, &rt, msg->msg_flags);
654 		if (err)
655 			ip_flush_pending_frames(sk);
656 		else if (!(msg->msg_flags & MSG_MORE)) {
657 			err = ip_push_pending_frames(sk, &fl4);
658 			if (err == -ENOBUFS && !inet_test_bit(RECVERR, sk))
659 				err = 0;
660 		}
661 		release_sock(sk);
662 	}
663 done:
664 	if (free)
665 		kfree(ipc.opt);
666 	ip_rt_put(rt);
667 
668 out:
669 	if (err < 0)
670 		return err;
671 	return len;
672 
673 do_confirm:
674 	if (msg->msg_flags & MSG_PROBE)
675 		dst_confirm_neigh(&rt->dst, &fl4.daddr);
676 	if (!(msg->msg_flags & MSG_PROBE) || len)
677 		goto back_from_confirm;
678 	err = 0;
679 	goto done;
680 }
681 
682 static void raw_close(struct sock *sk, long timeout)
683 {
684 	/*
685 	 * Raw sockets may have direct kernel references. Kill them.
686 	 */
687 	ip_ra_control(sk, 0, NULL);
688 
689 	sk_common_release(sk);
690 }
691 
692 static void raw_destroy(struct sock *sk)
693 {
694 	lock_sock(sk);
695 	ip_flush_pending_frames(sk);
696 	release_sock(sk);
697 }
698 
699 /* This gets rid of all the nasties in af_inet. -DaveM */
700 static int raw_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len)
701 {
702 	struct inet_sock *inet = inet_sk(sk);
703 	struct sockaddr_in *addr = (struct sockaddr_in *) uaddr;
704 	struct net *net = sock_net(sk);
705 	u32 tb_id = RT_TABLE_LOCAL;
706 	int ret = -EINVAL;
707 	int chk_addr_ret;
708 
709 	lock_sock(sk);
710 	if (sk->sk_state != TCP_CLOSE || addr_len < sizeof(struct sockaddr_in))
711 		goto out;
712 
713 	if (sk->sk_bound_dev_if)
714 		tb_id = l3mdev_fib_table_by_index(net,
715 						  sk->sk_bound_dev_if) ? : tb_id;
716 
717 	chk_addr_ret = inet_addr_type_table(net, addr->sin_addr.s_addr, tb_id);
718 
719 	ret = -EADDRNOTAVAIL;
720 	if (!inet_addr_valid_or_nonlocal(net, inet, addr->sin_addr.s_addr,
721 					 chk_addr_ret))
722 		goto out;
723 
724 	inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
725 	if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
726 		inet->inet_saddr = 0;  /* Use device */
727 	sk_dst_reset(sk);
728 	ret = 0;
729 out:
730 	release_sock(sk);
731 	return ret;
732 }
733 
734 /*
735  *	This should be easy, if there is something there
736  *	we return it, otherwise we block.
737  */
738 
739 static int raw_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
740 		       int flags, int *addr_len)
741 {
742 	struct inet_sock *inet = inet_sk(sk);
743 	size_t copied = 0;
744 	int err = -EOPNOTSUPP;
745 	DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
746 	struct sk_buff *skb;
747 
748 	if (flags & MSG_OOB)
749 		goto out;
750 
751 	if (flags & MSG_ERRQUEUE) {
752 		err = ip_recv_error(sk, msg, len, addr_len);
753 		goto out;
754 	}
755 
756 	skb = skb_recv_datagram(sk, flags, &err);
757 	if (!skb)
758 		goto out;
759 
760 	copied = skb->len;
761 	if (len < copied) {
762 		msg->msg_flags |= MSG_TRUNC;
763 		copied = len;
764 	}
765 
766 	err = skb_copy_datagram_msg(skb, 0, msg, copied);
767 	if (err)
768 		goto done;
769 
770 	sock_recv_cmsgs(msg, sk, skb);
771 
772 	/* Copy the address. */
773 	if (sin) {
774 		sin->sin_family = AF_INET;
775 		sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
776 		sin->sin_port = 0;
777 		memset(&sin->sin_zero, 0, sizeof(sin->sin_zero));
778 		*addr_len = sizeof(*sin);
779 	}
780 	if (inet_cmsg_flags(inet))
781 		ip_cmsg_recv(msg, skb);
782 	if (flags & MSG_TRUNC)
783 		copied = skb->len;
784 done:
785 	skb_free_datagram(sk, skb);
786 out:
787 	if (err)
788 		return err;
789 	return copied;
790 }
791 
792 static int raw_sk_init(struct sock *sk)
793 {
794 	struct raw_sock *rp = raw_sk(sk);
795 
796 	if (inet_sk(sk)->inet_num == IPPROTO_ICMP)
797 		memset(&rp->filter, 0, sizeof(rp->filter));
798 	return 0;
799 }
800 
801 static int raw_seticmpfilter(struct sock *sk, sockptr_t optval, int optlen)
802 {
803 	if (optlen > sizeof(struct icmp_filter))
804 		optlen = sizeof(struct icmp_filter);
805 	if (copy_from_sockptr(&raw_sk(sk)->filter, optval, optlen))
806 		return -EFAULT;
807 	return 0;
808 }
809 
810 static int raw_geticmpfilter(struct sock *sk, char __user *optval, int __user *optlen)
811 {
812 	int len, ret = -EFAULT;
813 
814 	if (get_user(len, optlen))
815 		goto out;
816 	ret = -EINVAL;
817 	if (len < 0)
818 		goto out;
819 	if (len > sizeof(struct icmp_filter))
820 		len = sizeof(struct icmp_filter);
821 	ret = -EFAULT;
822 	if (put_user(len, optlen) ||
823 	    copy_to_user(optval, &raw_sk(sk)->filter, len))
824 		goto out;
825 	ret = 0;
826 out:	return ret;
827 }
828 
829 static int do_raw_setsockopt(struct sock *sk, int optname,
830 			     sockptr_t optval, unsigned int optlen)
831 {
832 	if (optname == ICMP_FILTER) {
833 		if (inet_sk(sk)->inet_num != IPPROTO_ICMP)
834 			return -EOPNOTSUPP;
835 		else
836 			return raw_seticmpfilter(sk, optval, optlen);
837 	}
838 	return -ENOPROTOOPT;
839 }
840 
841 static int raw_setsockopt(struct sock *sk, int level, int optname,
842 			  sockptr_t optval, unsigned int optlen)
843 {
844 	if (level != SOL_RAW)
845 		return ip_setsockopt(sk, level, optname, optval, optlen);
846 	return do_raw_setsockopt(sk, optname, optval, optlen);
847 }
848 
849 static int do_raw_getsockopt(struct sock *sk, int optname,
850 			     char __user *optval, int __user *optlen)
851 {
852 	if (optname == ICMP_FILTER) {
853 		if (inet_sk(sk)->inet_num != IPPROTO_ICMP)
854 			return -EOPNOTSUPP;
855 		else
856 			return raw_geticmpfilter(sk, optval, optlen);
857 	}
858 	return -ENOPROTOOPT;
859 }
860 
861 static int raw_getsockopt(struct sock *sk, int level, int optname,
862 			  char __user *optval, int __user *optlen)
863 {
864 	if (level != SOL_RAW)
865 		return ip_getsockopt(sk, level, optname, optval, optlen);
866 	return do_raw_getsockopt(sk, optname, optval, optlen);
867 }
868 
869 static int raw_ioctl(struct sock *sk, int cmd, int *karg)
870 {
871 	switch (cmd) {
872 	case SIOCOUTQ: {
873 		*karg = sk_wmem_alloc_get(sk);
874 		return 0;
875 	}
876 	case SIOCINQ: {
877 		struct sk_buff *skb;
878 
879 		spin_lock_bh(&sk->sk_receive_queue.lock);
880 		skb = skb_peek(&sk->sk_receive_queue);
881 		if (skb)
882 			*karg = skb->len;
883 		else
884 			*karg = 0;
885 		spin_unlock_bh(&sk->sk_receive_queue.lock);
886 		return 0;
887 	}
888 
889 	default:
890 #ifdef CONFIG_IP_MROUTE
891 		return ipmr_ioctl(sk, cmd, karg);
892 #else
893 		return -ENOIOCTLCMD;
894 #endif
895 	}
896 }
897 
898 #ifdef CONFIG_COMPAT
899 static int compat_raw_ioctl(struct sock *sk, unsigned int cmd, unsigned long arg)
900 {
901 	switch (cmd) {
902 	case SIOCOUTQ:
903 	case SIOCINQ:
904 		return -ENOIOCTLCMD;
905 	default:
906 #ifdef CONFIG_IP_MROUTE
907 		return ipmr_compat_ioctl(sk, cmd, compat_ptr(arg));
908 #else
909 		return -ENOIOCTLCMD;
910 #endif
911 	}
912 }
913 #endif
914 
915 int raw_abort(struct sock *sk, int err)
916 {
917 	lock_sock(sk);
918 
919 	sk->sk_err = err;
920 	sk_error_report(sk);
921 	__udp_disconnect(sk, 0);
922 
923 	release_sock(sk);
924 
925 	return 0;
926 }
927 EXPORT_SYMBOL_GPL(raw_abort);
928 
929 struct proto raw_prot = {
930 	.name		   = "RAW",
931 	.owner		   = THIS_MODULE,
932 	.close		   = raw_close,
933 	.destroy	   = raw_destroy,
934 	.connect	   = ip4_datagram_connect,
935 	.disconnect	   = __udp_disconnect,
936 	.ioctl		   = raw_ioctl,
937 	.init		   = raw_sk_init,
938 	.setsockopt	   = raw_setsockopt,
939 	.getsockopt	   = raw_getsockopt,
940 	.sendmsg	   = raw_sendmsg,
941 	.recvmsg	   = raw_recvmsg,
942 	.bind		   = raw_bind,
943 	.backlog_rcv	   = raw_rcv_skb,
944 	.release_cb	   = ip4_datagram_release_cb,
945 	.hash		   = raw_hash_sk,
946 	.unhash		   = raw_unhash_sk,
947 	.obj_size	   = sizeof(struct raw_sock),
948 	.useroffset	   = offsetof(struct raw_sock, filter),
949 	.usersize	   = sizeof_field(struct raw_sock, filter),
950 	.h.raw_hash	   = &raw_v4_hashinfo,
951 #ifdef CONFIG_COMPAT
952 	.compat_ioctl	   = compat_raw_ioctl,
953 #endif
954 	.diag_destroy	   = raw_abort,
955 };
956 
957 #ifdef CONFIG_PROC_FS
958 static struct sock *raw_get_first(struct seq_file *seq, int bucket)
959 {
960 	struct raw_hashinfo *h = pde_data(file_inode(seq->file));
961 	struct raw_iter_state *state = raw_seq_private(seq);
962 	struct hlist_head *hlist;
963 	struct sock *sk;
964 
965 	for (state->bucket = bucket; state->bucket < RAW_HTABLE_SIZE;
966 			++state->bucket) {
967 		hlist = &h->ht[state->bucket];
968 		sk_for_each(sk, hlist) {
969 			if (sock_net(sk) == seq_file_net(seq))
970 				return sk;
971 		}
972 	}
973 	return NULL;
974 }
975 
976 static struct sock *raw_get_next(struct seq_file *seq, struct sock *sk)
977 {
978 	struct raw_iter_state *state = raw_seq_private(seq);
979 
980 	do {
981 		sk = sk_next(sk);
982 	} while (sk && sock_net(sk) != seq_file_net(seq));
983 
984 	if (!sk)
985 		return raw_get_first(seq, state->bucket + 1);
986 	return sk;
987 }
988 
989 static struct sock *raw_get_idx(struct seq_file *seq, loff_t pos)
990 {
991 	struct sock *sk = raw_get_first(seq, 0);
992 
993 	if (sk)
994 		while (pos && (sk = raw_get_next(seq, sk)) != NULL)
995 			--pos;
996 	return pos ? NULL : sk;
997 }
998 
999 void *raw_seq_start(struct seq_file *seq, loff_t *pos)
1000 	__acquires(&h->lock)
1001 {
1002 	struct raw_hashinfo *h = pde_data(file_inode(seq->file));
1003 
1004 	spin_lock(&h->lock);
1005 
1006 	return *pos ? raw_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
1007 }
1008 EXPORT_SYMBOL_GPL(raw_seq_start);
1009 
1010 void *raw_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1011 {
1012 	struct sock *sk;
1013 
1014 	if (v == SEQ_START_TOKEN)
1015 		sk = raw_get_first(seq, 0);
1016 	else
1017 		sk = raw_get_next(seq, v);
1018 	++*pos;
1019 	return sk;
1020 }
1021 EXPORT_SYMBOL_GPL(raw_seq_next);
1022 
1023 void raw_seq_stop(struct seq_file *seq, void *v)
1024 	__releases(&h->lock)
1025 {
1026 	struct raw_hashinfo *h = pde_data(file_inode(seq->file));
1027 
1028 	spin_unlock(&h->lock);
1029 }
1030 EXPORT_SYMBOL_GPL(raw_seq_stop);
1031 
1032 static void raw_sock_seq_show(struct seq_file *seq, struct sock *sp, int i)
1033 {
1034 	struct inet_sock *inet = inet_sk(sp);
1035 	__be32 dest = inet->inet_daddr,
1036 	       src = inet->inet_rcv_saddr;
1037 	__u16 destp = 0,
1038 	      srcp  = inet->inet_num;
1039 
1040 	seq_printf(seq, "%4d: %08X:%04X %08X:%04X"
1041 		" %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %u\n",
1042 		i, src, srcp, dest, destp, sp->sk_state,
1043 		sk_wmem_alloc_get(sp),
1044 		sk_rmem_alloc_get(sp),
1045 		0, 0L, 0,
1046 		from_kuid_munged(seq_user_ns(seq), sock_i_uid(sp)),
1047 		0, sock_i_ino(sp),
1048 		refcount_read(&sp->sk_refcnt), sp, atomic_read(&sp->sk_drops));
1049 }
1050 
1051 static int raw_seq_show(struct seq_file *seq, void *v)
1052 {
1053 	if (v == SEQ_START_TOKEN)
1054 		seq_printf(seq, "  sl  local_address rem_address   st tx_queue "
1055 				"rx_queue tr tm->when retrnsmt   uid  timeout "
1056 				"inode ref pointer drops\n");
1057 	else
1058 		raw_sock_seq_show(seq, v, raw_seq_private(seq)->bucket);
1059 	return 0;
1060 }
1061 
1062 static const struct seq_operations raw_seq_ops = {
1063 	.start = raw_seq_start,
1064 	.next  = raw_seq_next,
1065 	.stop  = raw_seq_stop,
1066 	.show  = raw_seq_show,
1067 };
1068 
1069 static __net_init int raw_init_net(struct net *net)
1070 {
1071 	if (!proc_create_net_data("raw", 0444, net->proc_net, &raw_seq_ops,
1072 			sizeof(struct raw_iter_state), &raw_v4_hashinfo))
1073 		return -ENOMEM;
1074 
1075 	return 0;
1076 }
1077 
1078 static __net_exit void raw_exit_net(struct net *net)
1079 {
1080 	remove_proc_entry("raw", net->proc_net);
1081 }
1082 
1083 static __net_initdata struct pernet_operations raw_net_ops = {
1084 	.init = raw_init_net,
1085 	.exit = raw_exit_net,
1086 };
1087 
1088 int __init raw_proc_init(void)
1089 {
1090 
1091 	return register_pernet_subsys(&raw_net_ops);
1092 }
1093 
1094 void __init raw_proc_exit(void)
1095 {
1096 	unregister_pernet_subsys(&raw_net_ops);
1097 }
1098 #endif /* CONFIG_PROC_FS */
1099 
1100 static void raw_sysctl_init_net(struct net *net)
1101 {
1102 #ifdef CONFIG_NET_L3_MASTER_DEV
1103 	net->ipv4.sysctl_raw_l3mdev_accept = 1;
1104 #endif
1105 }
1106 
1107 static int __net_init raw_sysctl_init(struct net *net)
1108 {
1109 	raw_sysctl_init_net(net);
1110 	return 0;
1111 }
1112 
1113 static struct pernet_operations __net_initdata raw_sysctl_ops = {
1114 	.init	= raw_sysctl_init,
1115 };
1116 
1117 void __init raw_init(void)
1118 {
1119 	raw_sysctl_init_net(&init_net);
1120 	if (register_pernet_subsys(&raw_sysctl_ops))
1121 		panic("RAW: failed to init sysctl parameters.\n");
1122 }
1123