xref: /linux/net/ipv4/raw.c (revision 7effbd18e496a954540fb6319db0d87d28a70496)
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_nulls_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_nulls_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_nulls_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, struct sock *sk, unsigned short num,
120 		  __be32 raddr, __be32 laddr, int dif, int sdif)
121 {
122 	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 	struct hlist_nulls_head *hlist;
167 	struct hlist_nulls_node *hnode;
168 	int sdif = inet_sdif(skb);
169 	int dif = inet_iif(skb);
170 	int delivered = 0;
171 	struct sock *sk;
172 
173 	hlist = &raw_v4_hashinfo.ht[hash];
174 	rcu_read_lock();
175 	sk_nulls_for_each(sk, hnode, hlist) {
176 		if (!raw_v4_match(net, sk, iph->protocol,
177 				  iph->saddr, iph->daddr, dif, sdif))
178 			continue;
179 		delivered = 1;
180 		if ((iph->protocol != IPPROTO_ICMP || !icmp_filter(sk, skb)) &&
181 		    ip_mc_sf_allow(sk, iph->daddr, iph->saddr,
182 				   skb->dev->ifindex, sdif)) {
183 			struct sk_buff *clone = skb_clone(skb, GFP_ATOMIC);
184 
185 			/* Not releasing hash table! */
186 			if (clone)
187 				raw_rcv(sk, clone);
188 		}
189 	}
190 	rcu_read_unlock();
191 	return delivered;
192 }
193 
194 int raw_local_deliver(struct sk_buff *skb, int protocol)
195 {
196 	struct net *net = dev_net(skb->dev);
197 
198 	return raw_v4_input(net, skb, ip_hdr(skb),
199 			    raw_hashfunc(net, protocol));
200 }
201 
202 static void raw_err(struct sock *sk, struct sk_buff *skb, u32 info)
203 {
204 	struct inet_sock *inet = inet_sk(sk);
205 	const int type = icmp_hdr(skb)->type;
206 	const int code = icmp_hdr(skb)->code;
207 	int err = 0;
208 	int harderr = 0;
209 
210 	if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
211 		ipv4_sk_update_pmtu(skb, sk, info);
212 	else if (type == ICMP_REDIRECT) {
213 		ipv4_sk_redirect(skb, sk);
214 		return;
215 	}
216 
217 	/* Report error on raw socket, if:
218 	   1. User requested ip_recverr.
219 	   2. Socket is connected (otherwise the error indication
220 	      is useless without ip_recverr and error is hard.
221 	 */
222 	if (!inet->recverr && sk->sk_state != TCP_ESTABLISHED)
223 		return;
224 
225 	switch (type) {
226 	default:
227 	case ICMP_TIME_EXCEEDED:
228 		err = EHOSTUNREACH;
229 		break;
230 	case ICMP_SOURCE_QUENCH:
231 		return;
232 	case ICMP_PARAMETERPROB:
233 		err = EPROTO;
234 		harderr = 1;
235 		break;
236 	case ICMP_DEST_UNREACH:
237 		err = EHOSTUNREACH;
238 		if (code > NR_ICMP_UNREACH)
239 			break;
240 		if (code == ICMP_FRAG_NEEDED) {
241 			harderr = inet->pmtudisc != IP_PMTUDISC_DONT;
242 			err = EMSGSIZE;
243 		} else {
244 			err = icmp_err_convert[code].errno;
245 			harderr = icmp_err_convert[code].fatal;
246 		}
247 	}
248 
249 	if (inet->recverr) {
250 		const struct iphdr *iph = (const struct iphdr *)skb->data;
251 		u8 *payload = skb->data + (iph->ihl << 2);
252 
253 		if (inet->hdrincl)
254 			payload = skb->data;
255 		ip_icmp_error(sk, skb, err, 0, info, payload);
256 	}
257 
258 	if (inet->recverr || harderr) {
259 		sk->sk_err = err;
260 		sk_error_report(sk);
261 	}
262 }
263 
264 void raw_icmp_error(struct sk_buff *skb, int protocol, u32 info)
265 {
266 	struct net *net = dev_net(skb->dev);
267 	struct hlist_nulls_head *hlist;
268 	struct hlist_nulls_node *hnode;
269 	int dif = skb->dev->ifindex;
270 	int sdif = inet_sdif(skb);
271 	const struct iphdr *iph;
272 	struct sock *sk;
273 	int hash;
274 
275 	hash = raw_hashfunc(net, protocol);
276 	hlist = &raw_v4_hashinfo.ht[hash];
277 
278 	rcu_read_lock();
279 	sk_nulls_for_each(sk, hnode, hlist) {
280 		iph = (const struct iphdr *)skb->data;
281 		if (!raw_v4_match(net, sk, iph->protocol,
282 				  iph->daddr, iph->saddr, dif, sdif))
283 			continue;
284 		raw_err(sk, skb, info);
285 	}
286 	rcu_read_unlock();
287 }
288 
289 static int raw_rcv_skb(struct sock *sk, struct sk_buff *skb)
290 {
291 	enum skb_drop_reason reason;
292 
293 	/* Charge it to the socket. */
294 
295 	ipv4_pktinfo_prepare(sk, skb);
296 	if (sock_queue_rcv_skb_reason(sk, skb, &reason) < 0) {
297 		kfree_skb_reason(skb, reason);
298 		return NET_RX_DROP;
299 	}
300 
301 	return NET_RX_SUCCESS;
302 }
303 
304 int raw_rcv(struct sock *sk, struct sk_buff *skb)
305 {
306 	if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) {
307 		atomic_inc(&sk->sk_drops);
308 		kfree_skb_reason(skb, SKB_DROP_REASON_XFRM_POLICY);
309 		return NET_RX_DROP;
310 	}
311 	nf_reset_ct(skb);
312 
313 	skb_push(skb, skb->data - skb_network_header(skb));
314 
315 	raw_rcv_skb(sk, skb);
316 	return 0;
317 }
318 
319 static int raw_send_hdrinc(struct sock *sk, struct flowi4 *fl4,
320 			   struct msghdr *msg, size_t length,
321 			   struct rtable **rtp, unsigned int flags,
322 			   const struct sockcm_cookie *sockc)
323 {
324 	struct inet_sock *inet = inet_sk(sk);
325 	struct net *net = sock_net(sk);
326 	struct iphdr *iph;
327 	struct sk_buff *skb;
328 	unsigned int iphlen;
329 	int err;
330 	struct rtable *rt = *rtp;
331 	int hlen, tlen;
332 
333 	if (length > rt->dst.dev->mtu) {
334 		ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport,
335 			       rt->dst.dev->mtu);
336 		return -EMSGSIZE;
337 	}
338 	if (length < sizeof(struct iphdr))
339 		return -EINVAL;
340 
341 	if (flags&MSG_PROBE)
342 		goto out;
343 
344 	hlen = LL_RESERVED_SPACE(rt->dst.dev);
345 	tlen = rt->dst.dev->needed_tailroom;
346 	skb = sock_alloc_send_skb(sk,
347 				  length + hlen + tlen + 15,
348 				  flags & MSG_DONTWAIT, &err);
349 	if (!skb)
350 		goto error;
351 	skb_reserve(skb, hlen);
352 
353 	skb->priority = sk->sk_priority;
354 	skb->mark = sockc->mark;
355 	skb->tstamp = sockc->transmit_time;
356 	skb_dst_set(skb, &rt->dst);
357 	*rtp = NULL;
358 
359 	skb_reset_network_header(skb);
360 	iph = ip_hdr(skb);
361 	skb_put(skb, length);
362 
363 	skb->ip_summed = CHECKSUM_NONE;
364 
365 	skb_setup_tx_timestamp(skb, sockc->tsflags);
366 
367 	if (flags & MSG_CONFIRM)
368 		skb_set_dst_pending_confirm(skb, 1);
369 
370 	skb->transport_header = skb->network_header;
371 	err = -EFAULT;
372 	if (memcpy_from_msg(iph, msg, length))
373 		goto error_free;
374 
375 	iphlen = iph->ihl * 4;
376 
377 	/*
378 	 * We don't want to modify the ip header, but we do need to
379 	 * be sure that it won't cause problems later along the network
380 	 * stack.  Specifically we want to make sure that iph->ihl is a
381 	 * sane value.  If ihl points beyond the length of the buffer passed
382 	 * in, reject the frame as invalid
383 	 */
384 	err = -EINVAL;
385 	if (iphlen > length)
386 		goto error_free;
387 
388 	if (iphlen >= sizeof(*iph)) {
389 		if (!iph->saddr)
390 			iph->saddr = fl4->saddr;
391 		iph->check   = 0;
392 		iph->tot_len = htons(length);
393 		if (!iph->id)
394 			ip_select_ident(net, skb, NULL);
395 
396 		iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
397 		skb->transport_header += iphlen;
398 		if (iph->protocol == IPPROTO_ICMP &&
399 		    length >= iphlen + sizeof(struct icmphdr))
400 			icmp_out_count(net, ((struct icmphdr *)
401 				skb_transport_header(skb))->type);
402 	}
403 
404 	err = NF_HOOK(NFPROTO_IPV4, NF_INET_LOCAL_OUT,
405 		      net, sk, skb, NULL, rt->dst.dev,
406 		      dst_output);
407 	if (err > 0)
408 		err = net_xmit_errno(err);
409 	if (err)
410 		goto error;
411 out:
412 	return 0;
413 
414 error_free:
415 	kfree_skb(skb);
416 error:
417 	IP_INC_STATS(net, IPSTATS_MIB_OUTDISCARDS);
418 	if (err == -ENOBUFS && !inet->recverr)
419 		err = 0;
420 	return err;
421 }
422 
423 static int raw_probe_proto_opt(struct raw_frag_vec *rfv, struct flowi4 *fl4)
424 {
425 	int err;
426 
427 	if (fl4->flowi4_proto != IPPROTO_ICMP)
428 		return 0;
429 
430 	/* We only need the first two bytes. */
431 	rfv->hlen = 2;
432 
433 	err = memcpy_from_msg(rfv->hdr.c, rfv->msg, rfv->hlen);
434 	if (err)
435 		return err;
436 
437 	fl4->fl4_icmp_type = rfv->hdr.icmph.type;
438 	fl4->fl4_icmp_code = rfv->hdr.icmph.code;
439 
440 	return 0;
441 }
442 
443 static int raw_getfrag(void *from, char *to, int offset, int len, int odd,
444 		       struct sk_buff *skb)
445 {
446 	struct raw_frag_vec *rfv = from;
447 
448 	if (offset < rfv->hlen) {
449 		int copy = min(rfv->hlen - offset, len);
450 
451 		if (skb->ip_summed == CHECKSUM_PARTIAL)
452 			memcpy(to, rfv->hdr.c + offset, copy);
453 		else
454 			skb->csum = csum_block_add(
455 				skb->csum,
456 				csum_partial_copy_nocheck(rfv->hdr.c + offset,
457 							  to, copy),
458 				odd);
459 
460 		odd = 0;
461 		offset += copy;
462 		to += copy;
463 		len -= copy;
464 
465 		if (!len)
466 			return 0;
467 	}
468 
469 	offset -= rfv->hlen;
470 
471 	return ip_generic_getfrag(rfv->msg, to, offset, len, odd, skb);
472 }
473 
474 static int raw_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
475 {
476 	struct inet_sock *inet = inet_sk(sk);
477 	struct net *net = sock_net(sk);
478 	struct ipcm_cookie ipc;
479 	struct rtable *rt = NULL;
480 	struct flowi4 fl4;
481 	int free = 0;
482 	__be32 daddr;
483 	__be32 saddr;
484 	u8  tos;
485 	int err;
486 	struct ip_options_data opt_copy;
487 	struct raw_frag_vec rfv;
488 	int hdrincl;
489 
490 	err = -EMSGSIZE;
491 	if (len > 0xFFFF)
492 		goto out;
493 
494 	/* hdrincl should be READ_ONCE(inet->hdrincl)
495 	 * but READ_ONCE() doesn't work with bit fields.
496 	 * Doing this indirectly yields the same result.
497 	 */
498 	hdrincl = inet->hdrincl;
499 	hdrincl = READ_ONCE(hdrincl);
500 	/*
501 	 *	Check the flags.
502 	 */
503 
504 	err = -EOPNOTSUPP;
505 	if (msg->msg_flags & MSG_OOB)	/* Mirror BSD error message */
506 		goto out;               /* compatibility */
507 
508 	/*
509 	 *	Get and verify the address.
510 	 */
511 
512 	if (msg->msg_namelen) {
513 		DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
514 		err = -EINVAL;
515 		if (msg->msg_namelen < sizeof(*usin))
516 			goto out;
517 		if (usin->sin_family != AF_INET) {
518 			pr_info_once("%s: %s forgot to set AF_INET. Fix it!\n",
519 				     __func__, current->comm);
520 			err = -EAFNOSUPPORT;
521 			if (usin->sin_family)
522 				goto out;
523 		}
524 		daddr = usin->sin_addr.s_addr;
525 		/* ANK: I did not forget to get protocol from port field.
526 		 * I just do not know, who uses this weirdness.
527 		 * IP_HDRINCL is much more convenient.
528 		 */
529 	} else {
530 		err = -EDESTADDRREQ;
531 		if (sk->sk_state != TCP_ESTABLISHED)
532 			goto out;
533 		daddr = inet->inet_daddr;
534 	}
535 
536 	ipcm_init_sk(&ipc, inet);
537 
538 	if (msg->msg_controllen) {
539 		err = ip_cmsg_send(sk, msg, &ipc, false);
540 		if (unlikely(err)) {
541 			kfree(ipc.opt);
542 			goto out;
543 		}
544 		if (ipc.opt)
545 			free = 1;
546 	}
547 
548 	saddr = ipc.addr;
549 	ipc.addr = daddr;
550 
551 	if (!ipc.opt) {
552 		struct ip_options_rcu *inet_opt;
553 
554 		rcu_read_lock();
555 		inet_opt = rcu_dereference(inet->inet_opt);
556 		if (inet_opt) {
557 			memcpy(&opt_copy, inet_opt,
558 			       sizeof(*inet_opt) + inet_opt->opt.optlen);
559 			ipc.opt = &opt_copy.opt;
560 		}
561 		rcu_read_unlock();
562 	}
563 
564 	if (ipc.opt) {
565 		err = -EINVAL;
566 		/* Linux does not mangle headers on raw sockets,
567 		 * so that IP options + IP_HDRINCL is non-sense.
568 		 */
569 		if (hdrincl)
570 			goto done;
571 		if (ipc.opt->opt.srr) {
572 			if (!daddr)
573 				goto done;
574 			daddr = ipc.opt->opt.faddr;
575 		}
576 	}
577 	tos = get_rtconn_flags(&ipc, sk);
578 	if (msg->msg_flags & MSG_DONTROUTE)
579 		tos |= RTO_ONLINK;
580 
581 	if (ipv4_is_multicast(daddr)) {
582 		if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif))
583 			ipc.oif = inet->mc_index;
584 		if (!saddr)
585 			saddr = inet->mc_addr;
586 	} else if (!ipc.oif) {
587 		ipc.oif = inet->uc_index;
588 	} else if (ipv4_is_lbcast(daddr) && inet->uc_index) {
589 		/* oif is set, packet is to local broadcast
590 		 * and uc_index is set. oif is most likely set
591 		 * by sk_bound_dev_if. If uc_index != oif check if the
592 		 * oif is an L3 master and uc_index is an L3 slave.
593 		 * If so, we want to allow the send using the uc_index.
594 		 */
595 		if (ipc.oif != inet->uc_index &&
596 		    ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk),
597 							      inet->uc_index)) {
598 			ipc.oif = inet->uc_index;
599 		}
600 	}
601 
602 	flowi4_init_output(&fl4, ipc.oif, ipc.sockc.mark, tos,
603 			   RT_SCOPE_UNIVERSE,
604 			   hdrincl ? IPPROTO_RAW : sk->sk_protocol,
605 			   inet_sk_flowi_flags(sk) |
606 			    (hdrincl ? FLOWI_FLAG_KNOWN_NH : 0),
607 			   daddr, saddr, 0, 0, sk->sk_uid);
608 
609 	if (!hdrincl) {
610 		rfv.msg = msg;
611 		rfv.hlen = 0;
612 
613 		err = raw_probe_proto_opt(&rfv, &fl4);
614 		if (err)
615 			goto done;
616 	}
617 
618 	security_sk_classify_flow(sk, flowi4_to_flowi_common(&fl4));
619 	rt = ip_route_output_flow(net, &fl4, sk);
620 	if (IS_ERR(rt)) {
621 		err = PTR_ERR(rt);
622 		rt = NULL;
623 		goto done;
624 	}
625 
626 	err = -EACCES;
627 	if (rt->rt_flags & RTCF_BROADCAST && !sock_flag(sk, SOCK_BROADCAST))
628 		goto done;
629 
630 	if (msg->msg_flags & MSG_CONFIRM)
631 		goto do_confirm;
632 back_from_confirm:
633 
634 	if (hdrincl)
635 		err = raw_send_hdrinc(sk, &fl4, msg, len,
636 				      &rt, msg->msg_flags, &ipc.sockc);
637 
638 	 else {
639 		if (!ipc.addr)
640 			ipc.addr = fl4.daddr;
641 		lock_sock(sk);
642 		err = ip_append_data(sk, &fl4, raw_getfrag,
643 				     &rfv, len, 0,
644 				     &ipc, &rt, msg->msg_flags);
645 		if (err)
646 			ip_flush_pending_frames(sk);
647 		else if (!(msg->msg_flags & MSG_MORE)) {
648 			err = ip_push_pending_frames(sk, &fl4);
649 			if (err == -ENOBUFS && !inet->recverr)
650 				err = 0;
651 		}
652 		release_sock(sk);
653 	}
654 done:
655 	if (free)
656 		kfree(ipc.opt);
657 	ip_rt_put(rt);
658 
659 out:
660 	if (err < 0)
661 		return err;
662 	return len;
663 
664 do_confirm:
665 	if (msg->msg_flags & MSG_PROBE)
666 		dst_confirm_neigh(&rt->dst, &fl4.daddr);
667 	if (!(msg->msg_flags & MSG_PROBE) || len)
668 		goto back_from_confirm;
669 	err = 0;
670 	goto done;
671 }
672 
673 static void raw_close(struct sock *sk, long timeout)
674 {
675 	/*
676 	 * Raw sockets may have direct kernel references. Kill them.
677 	 */
678 	ip_ra_control(sk, 0, NULL);
679 
680 	sk_common_release(sk);
681 }
682 
683 static void raw_destroy(struct sock *sk)
684 {
685 	lock_sock(sk);
686 	ip_flush_pending_frames(sk);
687 	release_sock(sk);
688 }
689 
690 /* This gets rid of all the nasties in af_inet. -DaveM */
691 static int raw_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len)
692 {
693 	struct inet_sock *inet = inet_sk(sk);
694 	struct sockaddr_in *addr = (struct sockaddr_in *) uaddr;
695 	struct net *net = sock_net(sk);
696 	u32 tb_id = RT_TABLE_LOCAL;
697 	int ret = -EINVAL;
698 	int chk_addr_ret;
699 
700 	lock_sock(sk);
701 	if (sk->sk_state != TCP_CLOSE || addr_len < sizeof(struct sockaddr_in))
702 		goto out;
703 
704 	if (sk->sk_bound_dev_if)
705 		tb_id = l3mdev_fib_table_by_index(net,
706 						  sk->sk_bound_dev_if) ? : tb_id;
707 
708 	chk_addr_ret = inet_addr_type_table(net, addr->sin_addr.s_addr, tb_id);
709 
710 	ret = -EADDRNOTAVAIL;
711 	if (!inet_addr_valid_or_nonlocal(net, inet, addr->sin_addr.s_addr,
712 					 chk_addr_ret))
713 		goto out;
714 
715 	inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
716 	if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
717 		inet->inet_saddr = 0;  /* Use device */
718 	sk_dst_reset(sk);
719 	ret = 0;
720 out:
721 	release_sock(sk);
722 	return ret;
723 }
724 
725 /*
726  *	This should be easy, if there is something there
727  *	we return it, otherwise we block.
728  */
729 
730 static int raw_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
731 		       int flags, int *addr_len)
732 {
733 	struct inet_sock *inet = inet_sk(sk);
734 	size_t copied = 0;
735 	int err = -EOPNOTSUPP;
736 	DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
737 	struct sk_buff *skb;
738 
739 	if (flags & MSG_OOB)
740 		goto out;
741 
742 	if (flags & MSG_ERRQUEUE) {
743 		err = ip_recv_error(sk, msg, len, addr_len);
744 		goto out;
745 	}
746 
747 	skb = skb_recv_datagram(sk, flags, &err);
748 	if (!skb)
749 		goto out;
750 
751 	copied = skb->len;
752 	if (len < copied) {
753 		msg->msg_flags |= MSG_TRUNC;
754 		copied = len;
755 	}
756 
757 	err = skb_copy_datagram_msg(skb, 0, msg, copied);
758 	if (err)
759 		goto done;
760 
761 	sock_recv_cmsgs(msg, sk, skb);
762 
763 	/* Copy the address. */
764 	if (sin) {
765 		sin->sin_family = AF_INET;
766 		sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
767 		sin->sin_port = 0;
768 		memset(&sin->sin_zero, 0, sizeof(sin->sin_zero));
769 		*addr_len = sizeof(*sin);
770 	}
771 	if (inet->cmsg_flags)
772 		ip_cmsg_recv(msg, skb);
773 	if (flags & MSG_TRUNC)
774 		copied = skb->len;
775 done:
776 	skb_free_datagram(sk, skb);
777 out:
778 	if (err)
779 		return err;
780 	return copied;
781 }
782 
783 static int raw_sk_init(struct sock *sk)
784 {
785 	struct raw_sock *rp = raw_sk(sk);
786 
787 	if (inet_sk(sk)->inet_num == IPPROTO_ICMP)
788 		memset(&rp->filter, 0, sizeof(rp->filter));
789 	return 0;
790 }
791 
792 static int raw_seticmpfilter(struct sock *sk, sockptr_t optval, int optlen)
793 {
794 	if (optlen > sizeof(struct icmp_filter))
795 		optlen = sizeof(struct icmp_filter);
796 	if (copy_from_sockptr(&raw_sk(sk)->filter, optval, optlen))
797 		return -EFAULT;
798 	return 0;
799 }
800 
801 static int raw_geticmpfilter(struct sock *sk, char __user *optval, int __user *optlen)
802 {
803 	int len, ret = -EFAULT;
804 
805 	if (get_user(len, optlen))
806 		goto out;
807 	ret = -EINVAL;
808 	if (len < 0)
809 		goto out;
810 	if (len > sizeof(struct icmp_filter))
811 		len = sizeof(struct icmp_filter);
812 	ret = -EFAULT;
813 	if (put_user(len, optlen) ||
814 	    copy_to_user(optval, &raw_sk(sk)->filter, len))
815 		goto out;
816 	ret = 0;
817 out:	return ret;
818 }
819 
820 static int do_raw_setsockopt(struct sock *sk, int level, int optname,
821 			     sockptr_t optval, unsigned int optlen)
822 {
823 	if (optname == ICMP_FILTER) {
824 		if (inet_sk(sk)->inet_num != IPPROTO_ICMP)
825 			return -EOPNOTSUPP;
826 		else
827 			return raw_seticmpfilter(sk, optval, optlen);
828 	}
829 	return -ENOPROTOOPT;
830 }
831 
832 static int raw_setsockopt(struct sock *sk, int level, int optname,
833 			  sockptr_t optval, unsigned int optlen)
834 {
835 	if (level != SOL_RAW)
836 		return ip_setsockopt(sk, level, optname, optval, optlen);
837 	return do_raw_setsockopt(sk, level, optname, optval, optlen);
838 }
839 
840 static int do_raw_getsockopt(struct sock *sk, int level, int optname,
841 			  char __user *optval, int __user *optlen)
842 {
843 	if (optname == ICMP_FILTER) {
844 		if (inet_sk(sk)->inet_num != IPPROTO_ICMP)
845 			return -EOPNOTSUPP;
846 		else
847 			return raw_geticmpfilter(sk, optval, optlen);
848 	}
849 	return -ENOPROTOOPT;
850 }
851 
852 static int raw_getsockopt(struct sock *sk, int level, int optname,
853 			  char __user *optval, int __user *optlen)
854 {
855 	if (level != SOL_RAW)
856 		return ip_getsockopt(sk, level, optname, optval, optlen);
857 	return do_raw_getsockopt(sk, level, optname, optval, optlen);
858 }
859 
860 static int raw_ioctl(struct sock *sk, int cmd, unsigned long arg)
861 {
862 	switch (cmd) {
863 	case SIOCOUTQ: {
864 		int amount = sk_wmem_alloc_get(sk);
865 
866 		return put_user(amount, (int __user *)arg);
867 	}
868 	case SIOCINQ: {
869 		struct sk_buff *skb;
870 		int amount = 0;
871 
872 		spin_lock_bh(&sk->sk_receive_queue.lock);
873 		skb = skb_peek(&sk->sk_receive_queue);
874 		if (skb)
875 			amount = skb->len;
876 		spin_unlock_bh(&sk->sk_receive_queue.lock);
877 		return put_user(amount, (int __user *)arg);
878 	}
879 
880 	default:
881 #ifdef CONFIG_IP_MROUTE
882 		return ipmr_ioctl(sk, cmd, (void __user *)arg);
883 #else
884 		return -ENOIOCTLCMD;
885 #endif
886 	}
887 }
888 
889 #ifdef CONFIG_COMPAT
890 static int compat_raw_ioctl(struct sock *sk, unsigned int cmd, unsigned long arg)
891 {
892 	switch (cmd) {
893 	case SIOCOUTQ:
894 	case SIOCINQ:
895 		return -ENOIOCTLCMD;
896 	default:
897 #ifdef CONFIG_IP_MROUTE
898 		return ipmr_compat_ioctl(sk, cmd, compat_ptr(arg));
899 #else
900 		return -ENOIOCTLCMD;
901 #endif
902 	}
903 }
904 #endif
905 
906 int raw_abort(struct sock *sk, int err)
907 {
908 	lock_sock(sk);
909 
910 	sk->sk_err = err;
911 	sk_error_report(sk);
912 	__udp_disconnect(sk, 0);
913 
914 	release_sock(sk);
915 
916 	return 0;
917 }
918 EXPORT_SYMBOL_GPL(raw_abort);
919 
920 struct proto raw_prot = {
921 	.name		   = "RAW",
922 	.owner		   = THIS_MODULE,
923 	.close		   = raw_close,
924 	.destroy	   = raw_destroy,
925 	.connect	   = ip4_datagram_connect,
926 	.disconnect	   = __udp_disconnect,
927 	.ioctl		   = raw_ioctl,
928 	.init		   = raw_sk_init,
929 	.setsockopt	   = raw_setsockopt,
930 	.getsockopt	   = raw_getsockopt,
931 	.sendmsg	   = raw_sendmsg,
932 	.recvmsg	   = raw_recvmsg,
933 	.bind		   = raw_bind,
934 	.backlog_rcv	   = raw_rcv_skb,
935 	.release_cb	   = ip4_datagram_release_cb,
936 	.hash		   = raw_hash_sk,
937 	.unhash		   = raw_unhash_sk,
938 	.obj_size	   = sizeof(struct raw_sock),
939 	.useroffset	   = offsetof(struct raw_sock, filter),
940 	.usersize	   = sizeof_field(struct raw_sock, filter),
941 	.h.raw_hash	   = &raw_v4_hashinfo,
942 #ifdef CONFIG_COMPAT
943 	.compat_ioctl	   = compat_raw_ioctl,
944 #endif
945 	.diag_destroy	   = raw_abort,
946 };
947 
948 #ifdef CONFIG_PROC_FS
949 static struct sock *raw_get_first(struct seq_file *seq, int bucket)
950 {
951 	struct raw_hashinfo *h = pde_data(file_inode(seq->file));
952 	struct raw_iter_state *state = raw_seq_private(seq);
953 	struct hlist_nulls_head *hlist;
954 	struct hlist_nulls_node *hnode;
955 	struct sock *sk;
956 
957 	for (state->bucket = bucket; state->bucket < RAW_HTABLE_SIZE;
958 			++state->bucket) {
959 		hlist = &h->ht[state->bucket];
960 		sk_nulls_for_each(sk, hnode, hlist) {
961 			if (sock_net(sk) == seq_file_net(seq))
962 				return sk;
963 		}
964 	}
965 	return NULL;
966 }
967 
968 static struct sock *raw_get_next(struct seq_file *seq, struct sock *sk)
969 {
970 	struct raw_iter_state *state = raw_seq_private(seq);
971 
972 	do {
973 		sk = sk_nulls_next(sk);
974 	} while (sk && sock_net(sk) != seq_file_net(seq));
975 
976 	if (!sk)
977 		return raw_get_first(seq, state->bucket + 1);
978 	return sk;
979 }
980 
981 static struct sock *raw_get_idx(struct seq_file *seq, loff_t pos)
982 {
983 	struct sock *sk = raw_get_first(seq, 0);
984 
985 	if (sk)
986 		while (pos && (sk = raw_get_next(seq, sk)) != NULL)
987 			--pos;
988 	return pos ? NULL : sk;
989 }
990 
991 void *raw_seq_start(struct seq_file *seq, loff_t *pos)
992 	__acquires(RCU)
993 {
994 	rcu_read_lock();
995 	return *pos ? raw_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
996 }
997 EXPORT_SYMBOL_GPL(raw_seq_start);
998 
999 void *raw_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1000 {
1001 	struct sock *sk;
1002 
1003 	if (v == SEQ_START_TOKEN)
1004 		sk = raw_get_first(seq, 0);
1005 	else
1006 		sk = raw_get_next(seq, v);
1007 	++*pos;
1008 	return sk;
1009 }
1010 EXPORT_SYMBOL_GPL(raw_seq_next);
1011 
1012 void raw_seq_stop(struct seq_file *seq, void *v)
1013 	__releases(RCU)
1014 {
1015 	rcu_read_unlock();
1016 }
1017 EXPORT_SYMBOL_GPL(raw_seq_stop);
1018 
1019 static void raw_sock_seq_show(struct seq_file *seq, struct sock *sp, int i)
1020 {
1021 	struct inet_sock *inet = inet_sk(sp);
1022 	__be32 dest = inet->inet_daddr,
1023 	       src = inet->inet_rcv_saddr;
1024 	__u16 destp = 0,
1025 	      srcp  = inet->inet_num;
1026 
1027 	seq_printf(seq, "%4d: %08X:%04X %08X:%04X"
1028 		" %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %u\n",
1029 		i, src, srcp, dest, destp, sp->sk_state,
1030 		sk_wmem_alloc_get(sp),
1031 		sk_rmem_alloc_get(sp),
1032 		0, 0L, 0,
1033 		from_kuid_munged(seq_user_ns(seq), sock_i_uid(sp)),
1034 		0, sock_i_ino(sp),
1035 		refcount_read(&sp->sk_refcnt), sp, atomic_read(&sp->sk_drops));
1036 }
1037 
1038 static int raw_seq_show(struct seq_file *seq, void *v)
1039 {
1040 	if (v == SEQ_START_TOKEN)
1041 		seq_printf(seq, "  sl  local_address rem_address   st tx_queue "
1042 				"rx_queue tr tm->when retrnsmt   uid  timeout "
1043 				"inode ref pointer drops\n");
1044 	else
1045 		raw_sock_seq_show(seq, v, raw_seq_private(seq)->bucket);
1046 	return 0;
1047 }
1048 
1049 static const struct seq_operations raw_seq_ops = {
1050 	.start = raw_seq_start,
1051 	.next  = raw_seq_next,
1052 	.stop  = raw_seq_stop,
1053 	.show  = raw_seq_show,
1054 };
1055 
1056 static __net_init int raw_init_net(struct net *net)
1057 {
1058 	if (!proc_create_net_data("raw", 0444, net->proc_net, &raw_seq_ops,
1059 			sizeof(struct raw_iter_state), &raw_v4_hashinfo))
1060 		return -ENOMEM;
1061 
1062 	return 0;
1063 }
1064 
1065 static __net_exit void raw_exit_net(struct net *net)
1066 {
1067 	remove_proc_entry("raw", net->proc_net);
1068 }
1069 
1070 static __net_initdata struct pernet_operations raw_net_ops = {
1071 	.init = raw_init_net,
1072 	.exit = raw_exit_net,
1073 };
1074 
1075 int __init raw_proc_init(void)
1076 {
1077 
1078 	return register_pernet_subsys(&raw_net_ops);
1079 }
1080 
1081 void __init raw_proc_exit(void)
1082 {
1083 	unregister_pernet_subsys(&raw_net_ops);
1084 }
1085 #endif /* CONFIG_PROC_FS */
1086 
1087 static void raw_sysctl_init_net(struct net *net)
1088 {
1089 #ifdef CONFIG_NET_L3_MASTER_DEV
1090 	net->ipv4.sysctl_raw_l3mdev_accept = 1;
1091 #endif
1092 }
1093 
1094 static int __net_init raw_sysctl_init(struct net *net)
1095 {
1096 	raw_sysctl_init_net(net);
1097 	return 0;
1098 }
1099 
1100 static struct pernet_operations __net_initdata raw_sysctl_ops = {
1101 	.init	= raw_sysctl_init,
1102 };
1103 
1104 void __init raw_init(void)
1105 {
1106 	raw_sysctl_init_net(&init_net);
1107 	if (register_pernet_subsys(&raw_sysctl_ops))
1108 		panic("RAW: failed to init sysctl parameters.\n");
1109 }
1110