xref: /linux/net/ipv4/raw.c (revision b77e0ce62d63a761ffb7f7245a215a49f5921c2f)
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 	.lock = __RW_LOCK_UNLOCKED(raw_v4_hashinfo.lock),
90 };
91 EXPORT_SYMBOL_GPL(raw_v4_hashinfo);
92 
93 int raw_hash_sk(struct sock *sk)
94 {
95 	struct raw_hashinfo *h = sk->sk_prot->h.raw_hash;
96 	struct hlist_head *head;
97 
98 	head = &h->ht[inet_sk(sk)->inet_num & (RAW_HTABLE_SIZE - 1)];
99 
100 	write_lock_bh(&h->lock);
101 	sk_add_node(sk, head);
102 	sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
103 	write_unlock_bh(&h->lock);
104 
105 	return 0;
106 }
107 EXPORT_SYMBOL_GPL(raw_hash_sk);
108 
109 void raw_unhash_sk(struct sock *sk)
110 {
111 	struct raw_hashinfo *h = sk->sk_prot->h.raw_hash;
112 
113 	write_lock_bh(&h->lock);
114 	if (sk_del_node_init(sk))
115 		sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
116 	write_unlock_bh(&h->lock);
117 }
118 EXPORT_SYMBOL_GPL(raw_unhash_sk);
119 
120 struct sock *__raw_v4_lookup(struct net *net, struct sock *sk,
121 			     unsigned short num, __be32 raddr, __be32 laddr,
122 			     int dif, int sdif)
123 {
124 	sk_for_each_from(sk) {
125 		struct inet_sock *inet = inet_sk(sk);
126 
127 		if (net_eq(sock_net(sk), net) && inet->inet_num == num	&&
128 		    !(inet->inet_daddr && inet->inet_daddr != raddr) 	&&
129 		    !(inet->inet_rcv_saddr && inet->inet_rcv_saddr != laddr) &&
130 		    raw_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif))
131 			goto found; /* gotcha */
132 	}
133 	sk = NULL;
134 found:
135 	return sk;
136 }
137 EXPORT_SYMBOL_GPL(__raw_v4_lookup);
138 
139 /*
140  *	0 - deliver
141  *	1 - block
142  */
143 static int icmp_filter(const struct sock *sk, const struct sk_buff *skb)
144 {
145 	struct icmphdr _hdr;
146 	const struct icmphdr *hdr;
147 
148 	hdr = skb_header_pointer(skb, skb_transport_offset(skb),
149 				 sizeof(_hdr), &_hdr);
150 	if (!hdr)
151 		return 1;
152 
153 	if (hdr->type < 32) {
154 		__u32 data = raw_sk(sk)->filter.data;
155 
156 		return ((1U << hdr->type) & data) != 0;
157 	}
158 
159 	/* Do not block unknown ICMP types */
160 	return 0;
161 }
162 
163 /* IP input processing comes here for RAW socket delivery.
164  * Caller owns SKB, so we must make clones.
165  *
166  * RFC 1122: SHOULD pass TOS value up to the transport layer.
167  * -> It does. And not only TOS, but all IP header.
168  */
169 static int raw_v4_input(struct sk_buff *skb, const struct iphdr *iph, int hash)
170 {
171 	int sdif = inet_sdif(skb);
172 	int dif = inet_iif(skb);
173 	struct sock *sk;
174 	struct hlist_head *head;
175 	int delivered = 0;
176 	struct net *net;
177 
178 	read_lock(&raw_v4_hashinfo.lock);
179 	head = &raw_v4_hashinfo.ht[hash];
180 	if (hlist_empty(head))
181 		goto out;
182 
183 	net = dev_net(skb->dev);
184 	sk = __raw_v4_lookup(net, __sk_head(head), iph->protocol,
185 			     iph->saddr, iph->daddr, dif, sdif);
186 
187 	while (sk) {
188 		delivered = 1;
189 		if ((iph->protocol != IPPROTO_ICMP || !icmp_filter(sk, skb)) &&
190 		    ip_mc_sf_allow(sk, iph->daddr, iph->saddr,
191 				   skb->dev->ifindex, sdif)) {
192 			struct sk_buff *clone = skb_clone(skb, GFP_ATOMIC);
193 
194 			/* Not releasing hash table! */
195 			if (clone)
196 				raw_rcv(sk, clone);
197 		}
198 		sk = __raw_v4_lookup(net, sk_next(sk), iph->protocol,
199 				     iph->saddr, iph->daddr,
200 				     dif, sdif);
201 	}
202 out:
203 	read_unlock(&raw_v4_hashinfo.lock);
204 	return delivered;
205 }
206 
207 int raw_local_deliver(struct sk_buff *skb, int protocol)
208 {
209 	int hash;
210 	struct sock *raw_sk;
211 
212 	hash = protocol & (RAW_HTABLE_SIZE - 1);
213 	raw_sk = sk_head(&raw_v4_hashinfo.ht[hash]);
214 
215 	/* If there maybe a raw socket we must check - if not we
216 	 * don't care less
217 	 */
218 	if (raw_sk && !raw_v4_input(skb, ip_hdr(skb), hash))
219 		raw_sk = NULL;
220 
221 	return raw_sk != NULL;
222 
223 }
224 
225 static void raw_err(struct sock *sk, struct sk_buff *skb, u32 info)
226 {
227 	struct inet_sock *inet = inet_sk(sk);
228 	const int type = icmp_hdr(skb)->type;
229 	const int code = icmp_hdr(skb)->code;
230 	int err = 0;
231 	int harderr = 0;
232 
233 	if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
234 		ipv4_sk_update_pmtu(skb, sk, info);
235 	else if (type == ICMP_REDIRECT) {
236 		ipv4_sk_redirect(skb, sk);
237 		return;
238 	}
239 
240 	/* Report error on raw socket, if:
241 	   1. User requested ip_recverr.
242 	   2. Socket is connected (otherwise the error indication
243 	      is useless without ip_recverr and error is hard.
244 	 */
245 	if (!inet->recverr && sk->sk_state != TCP_ESTABLISHED)
246 		return;
247 
248 	switch (type) {
249 	default:
250 	case ICMP_TIME_EXCEEDED:
251 		err = EHOSTUNREACH;
252 		break;
253 	case ICMP_SOURCE_QUENCH:
254 		return;
255 	case ICMP_PARAMETERPROB:
256 		err = EPROTO;
257 		harderr = 1;
258 		break;
259 	case ICMP_DEST_UNREACH:
260 		err = EHOSTUNREACH;
261 		if (code > NR_ICMP_UNREACH)
262 			break;
263 		if (code == ICMP_FRAG_NEEDED) {
264 			harderr = inet->pmtudisc != IP_PMTUDISC_DONT;
265 			err = EMSGSIZE;
266 		} else {
267 			err = icmp_err_convert[code].errno;
268 			harderr = icmp_err_convert[code].fatal;
269 		}
270 	}
271 
272 	if (inet->recverr) {
273 		const struct iphdr *iph = (const struct iphdr *)skb->data;
274 		u8 *payload = skb->data + (iph->ihl << 2);
275 
276 		if (inet->hdrincl)
277 			payload = skb->data;
278 		ip_icmp_error(sk, skb, err, 0, info, payload);
279 	}
280 
281 	if (inet->recverr || harderr) {
282 		sk->sk_err = err;
283 		sk->sk_error_report(sk);
284 	}
285 }
286 
287 void raw_icmp_error(struct sk_buff *skb, int protocol, u32 info)
288 {
289 	int hash;
290 	struct sock *raw_sk;
291 	const struct iphdr *iph;
292 	struct net *net;
293 
294 	hash = protocol & (RAW_HTABLE_SIZE - 1);
295 
296 	read_lock(&raw_v4_hashinfo.lock);
297 	raw_sk = sk_head(&raw_v4_hashinfo.ht[hash]);
298 	if (raw_sk) {
299 		int dif = skb->dev->ifindex;
300 		int sdif = inet_sdif(skb);
301 
302 		iph = (const struct iphdr *)skb->data;
303 		net = dev_net(skb->dev);
304 
305 		while ((raw_sk = __raw_v4_lookup(net, raw_sk, protocol,
306 						iph->daddr, iph->saddr,
307 						dif, sdif)) != NULL) {
308 			raw_err(raw_sk, skb, info);
309 			raw_sk = sk_next(raw_sk);
310 			iph = (const struct iphdr *)skb->data;
311 		}
312 	}
313 	read_unlock(&raw_v4_hashinfo.lock);
314 }
315 
316 static int raw_rcv_skb(struct sock *sk, struct sk_buff *skb)
317 {
318 	/* Charge it to the socket. */
319 
320 	ipv4_pktinfo_prepare(sk, skb);
321 	if (sock_queue_rcv_skb(sk, skb) < 0) {
322 		kfree_skb(skb);
323 		return NET_RX_DROP;
324 	}
325 
326 	return NET_RX_SUCCESS;
327 }
328 
329 int raw_rcv(struct sock *sk, struct sk_buff *skb)
330 {
331 	if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) {
332 		atomic_inc(&sk->sk_drops);
333 		kfree_skb(skb);
334 		return NET_RX_DROP;
335 	}
336 	nf_reset_ct(skb);
337 
338 	skb_push(skb, skb->data - skb_network_header(skb));
339 
340 	raw_rcv_skb(sk, skb);
341 	return 0;
342 }
343 
344 static int raw_send_hdrinc(struct sock *sk, struct flowi4 *fl4,
345 			   struct msghdr *msg, size_t length,
346 			   struct rtable **rtp, unsigned int flags,
347 			   const struct sockcm_cookie *sockc)
348 {
349 	struct inet_sock *inet = inet_sk(sk);
350 	struct net *net = sock_net(sk);
351 	struct iphdr *iph;
352 	struct sk_buff *skb;
353 	unsigned int iphlen;
354 	int err;
355 	struct rtable *rt = *rtp;
356 	int hlen, tlen;
357 
358 	if (length > rt->dst.dev->mtu) {
359 		ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport,
360 			       rt->dst.dev->mtu);
361 		return -EMSGSIZE;
362 	}
363 	if (length < sizeof(struct iphdr))
364 		return -EINVAL;
365 
366 	if (flags&MSG_PROBE)
367 		goto out;
368 
369 	hlen = LL_RESERVED_SPACE(rt->dst.dev);
370 	tlen = rt->dst.dev->needed_tailroom;
371 	skb = sock_alloc_send_skb(sk,
372 				  length + hlen + tlen + 15,
373 				  flags & MSG_DONTWAIT, &err);
374 	if (!skb)
375 		goto error;
376 	skb_reserve(skb, hlen);
377 
378 	skb->priority = sk->sk_priority;
379 	skb->mark = sockc->mark;
380 	skb->tstamp = sockc->transmit_time;
381 	skb_dst_set(skb, &rt->dst);
382 	*rtp = NULL;
383 
384 	skb_reset_network_header(skb);
385 	iph = ip_hdr(skb);
386 	skb_put(skb, length);
387 
388 	skb->ip_summed = CHECKSUM_NONE;
389 
390 	skb_setup_tx_timestamp(skb, sockc->tsflags);
391 
392 	if (flags & MSG_CONFIRM)
393 		skb_set_dst_pending_confirm(skb, 1);
394 
395 	skb->transport_header = skb->network_header;
396 	err = -EFAULT;
397 	if (memcpy_from_msg(iph, msg, length))
398 		goto error_free;
399 
400 	iphlen = iph->ihl * 4;
401 
402 	/*
403 	 * We don't want to modify the ip header, but we do need to
404 	 * be sure that it won't cause problems later along the network
405 	 * stack.  Specifically we want to make sure that iph->ihl is a
406 	 * sane value.  If ihl points beyond the length of the buffer passed
407 	 * in, reject the frame as invalid
408 	 */
409 	err = -EINVAL;
410 	if (iphlen > length)
411 		goto error_free;
412 
413 	if (iphlen >= sizeof(*iph)) {
414 		if (!iph->saddr)
415 			iph->saddr = fl4->saddr;
416 		iph->check   = 0;
417 		iph->tot_len = htons(length);
418 		if (!iph->id)
419 			ip_select_ident(net, skb, NULL);
420 
421 		iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
422 		skb->transport_header += iphlen;
423 		if (iph->protocol == IPPROTO_ICMP &&
424 		    length >= iphlen + sizeof(struct icmphdr))
425 			icmp_out_count(net, ((struct icmphdr *)
426 				skb_transport_header(skb))->type);
427 	}
428 
429 	err = NF_HOOK(NFPROTO_IPV4, NF_INET_LOCAL_OUT,
430 		      net, sk, skb, NULL, rt->dst.dev,
431 		      dst_output);
432 	if (err > 0)
433 		err = net_xmit_errno(err);
434 	if (err)
435 		goto error;
436 out:
437 	return 0;
438 
439 error_free:
440 	kfree_skb(skb);
441 error:
442 	IP_INC_STATS(net, IPSTATS_MIB_OUTDISCARDS);
443 	if (err == -ENOBUFS && !inet->recverr)
444 		err = 0;
445 	return err;
446 }
447 
448 static int raw_probe_proto_opt(struct raw_frag_vec *rfv, struct flowi4 *fl4)
449 {
450 	int err;
451 
452 	if (fl4->flowi4_proto != IPPROTO_ICMP)
453 		return 0;
454 
455 	/* We only need the first two bytes. */
456 	rfv->hlen = 2;
457 
458 	err = memcpy_from_msg(rfv->hdr.c, rfv->msg, rfv->hlen);
459 	if (err)
460 		return err;
461 
462 	fl4->fl4_icmp_type = rfv->hdr.icmph.type;
463 	fl4->fl4_icmp_code = rfv->hdr.icmph.code;
464 
465 	return 0;
466 }
467 
468 static int raw_getfrag(void *from, char *to, int offset, int len, int odd,
469 		       struct sk_buff *skb)
470 {
471 	struct raw_frag_vec *rfv = from;
472 
473 	if (offset < rfv->hlen) {
474 		int copy = min(rfv->hlen - offset, len);
475 
476 		if (skb->ip_summed == CHECKSUM_PARTIAL)
477 			memcpy(to, rfv->hdr.c + offset, copy);
478 		else
479 			skb->csum = csum_block_add(
480 				skb->csum,
481 				csum_partial_copy_nocheck(rfv->hdr.c + offset,
482 							  to, copy),
483 				odd);
484 
485 		odd = 0;
486 		offset += copy;
487 		to += copy;
488 		len -= copy;
489 
490 		if (!len)
491 			return 0;
492 	}
493 
494 	offset -= rfv->hlen;
495 
496 	return ip_generic_getfrag(rfv->msg, to, offset, len, odd, skb);
497 }
498 
499 static int raw_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
500 {
501 	struct inet_sock *inet = inet_sk(sk);
502 	struct net *net = sock_net(sk);
503 	struct ipcm_cookie ipc;
504 	struct rtable *rt = NULL;
505 	struct flowi4 fl4;
506 	int free = 0;
507 	__be32 daddr;
508 	__be32 saddr;
509 	u8  tos;
510 	int err;
511 	struct ip_options_data opt_copy;
512 	struct raw_frag_vec rfv;
513 	int hdrincl;
514 
515 	err = -EMSGSIZE;
516 	if (len > 0xFFFF)
517 		goto out;
518 
519 	/* hdrincl should be READ_ONCE(inet->hdrincl)
520 	 * but READ_ONCE() doesn't work with bit fields.
521 	 * Doing this indirectly yields the same result.
522 	 */
523 	hdrincl = inet->hdrincl;
524 	hdrincl = READ_ONCE(hdrincl);
525 	/*
526 	 *	Check the flags.
527 	 */
528 
529 	err = -EOPNOTSUPP;
530 	if (msg->msg_flags & MSG_OOB)	/* Mirror BSD error message */
531 		goto out;               /* compatibility */
532 
533 	/*
534 	 *	Get and verify the address.
535 	 */
536 
537 	if (msg->msg_namelen) {
538 		DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
539 		err = -EINVAL;
540 		if (msg->msg_namelen < sizeof(*usin))
541 			goto out;
542 		if (usin->sin_family != AF_INET) {
543 			pr_info_once("%s: %s forgot to set AF_INET. Fix it!\n",
544 				     __func__, current->comm);
545 			err = -EAFNOSUPPORT;
546 			if (usin->sin_family)
547 				goto out;
548 		}
549 		daddr = usin->sin_addr.s_addr;
550 		/* ANK: I did not forget to get protocol from port field.
551 		 * I just do not know, who uses this weirdness.
552 		 * IP_HDRINCL is much more convenient.
553 		 */
554 	} else {
555 		err = -EDESTADDRREQ;
556 		if (sk->sk_state != TCP_ESTABLISHED)
557 			goto out;
558 		daddr = inet->inet_daddr;
559 	}
560 
561 	ipcm_init_sk(&ipc, inet);
562 
563 	if (msg->msg_controllen) {
564 		err = ip_cmsg_send(sk, msg, &ipc, false);
565 		if (unlikely(err)) {
566 			kfree(ipc.opt);
567 			goto out;
568 		}
569 		if (ipc.opt)
570 			free = 1;
571 	}
572 
573 	saddr = ipc.addr;
574 	ipc.addr = daddr;
575 
576 	if (!ipc.opt) {
577 		struct ip_options_rcu *inet_opt;
578 
579 		rcu_read_lock();
580 		inet_opt = rcu_dereference(inet->inet_opt);
581 		if (inet_opt) {
582 			memcpy(&opt_copy, inet_opt,
583 			       sizeof(*inet_opt) + inet_opt->opt.optlen);
584 			ipc.opt = &opt_copy.opt;
585 		}
586 		rcu_read_unlock();
587 	}
588 
589 	if (ipc.opt) {
590 		err = -EINVAL;
591 		/* Linux does not mangle headers on raw sockets,
592 		 * so that IP options + IP_HDRINCL is non-sense.
593 		 */
594 		if (hdrincl)
595 			goto done;
596 		if (ipc.opt->opt.srr) {
597 			if (!daddr)
598 				goto done;
599 			daddr = ipc.opt->opt.faddr;
600 		}
601 	}
602 	tos = get_rtconn_flags(&ipc, sk);
603 	if (msg->msg_flags & MSG_DONTROUTE)
604 		tos |= RTO_ONLINK;
605 
606 	if (ipv4_is_multicast(daddr)) {
607 		if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif))
608 			ipc.oif = inet->mc_index;
609 		if (!saddr)
610 			saddr = inet->mc_addr;
611 	} else if (!ipc.oif) {
612 		ipc.oif = inet->uc_index;
613 	} else if (ipv4_is_lbcast(daddr) && inet->uc_index) {
614 		/* oif is set, packet is to local broadcast
615 		 * and uc_index is set. oif is most likely set
616 		 * by sk_bound_dev_if. If uc_index != oif check if the
617 		 * oif is an L3 master and uc_index is an L3 slave.
618 		 * If so, we want to allow the send using the uc_index.
619 		 */
620 		if (ipc.oif != inet->uc_index &&
621 		    ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk),
622 							      inet->uc_index)) {
623 			ipc.oif = inet->uc_index;
624 		}
625 	}
626 
627 	flowi4_init_output(&fl4, ipc.oif, ipc.sockc.mark, tos,
628 			   RT_SCOPE_UNIVERSE,
629 			   hdrincl ? IPPROTO_RAW : sk->sk_protocol,
630 			   inet_sk_flowi_flags(sk) |
631 			    (hdrincl ? FLOWI_FLAG_KNOWN_NH : 0),
632 			   daddr, saddr, 0, 0, sk->sk_uid);
633 
634 	if (!hdrincl) {
635 		rfv.msg = msg;
636 		rfv.hlen = 0;
637 
638 		err = raw_probe_proto_opt(&rfv, &fl4);
639 		if (err)
640 			goto done;
641 	}
642 
643 	security_sk_classify_flow(sk, flowi4_to_flowi_common(&fl4));
644 	rt = ip_route_output_flow(net, &fl4, sk);
645 	if (IS_ERR(rt)) {
646 		err = PTR_ERR(rt);
647 		rt = NULL;
648 		goto done;
649 	}
650 
651 	err = -EACCES;
652 	if (rt->rt_flags & RTCF_BROADCAST && !sock_flag(sk, SOCK_BROADCAST))
653 		goto done;
654 
655 	if (msg->msg_flags & MSG_CONFIRM)
656 		goto do_confirm;
657 back_from_confirm:
658 
659 	if (hdrincl)
660 		err = raw_send_hdrinc(sk, &fl4, msg, len,
661 				      &rt, msg->msg_flags, &ipc.sockc);
662 
663 	 else {
664 		if (!ipc.addr)
665 			ipc.addr = fl4.daddr;
666 		lock_sock(sk);
667 		err = ip_append_data(sk, &fl4, raw_getfrag,
668 				     &rfv, len, 0,
669 				     &ipc, &rt, msg->msg_flags);
670 		if (err)
671 			ip_flush_pending_frames(sk);
672 		else if (!(msg->msg_flags & MSG_MORE)) {
673 			err = ip_push_pending_frames(sk, &fl4);
674 			if (err == -ENOBUFS && !inet->recverr)
675 				err = 0;
676 		}
677 		release_sock(sk);
678 	}
679 done:
680 	if (free)
681 		kfree(ipc.opt);
682 	ip_rt_put(rt);
683 
684 out:
685 	if (err < 0)
686 		return err;
687 	return len;
688 
689 do_confirm:
690 	if (msg->msg_flags & MSG_PROBE)
691 		dst_confirm_neigh(&rt->dst, &fl4.daddr);
692 	if (!(msg->msg_flags & MSG_PROBE) || len)
693 		goto back_from_confirm;
694 	err = 0;
695 	goto done;
696 }
697 
698 static void raw_close(struct sock *sk, long timeout)
699 {
700 	/*
701 	 * Raw sockets may have direct kernel references. Kill them.
702 	 */
703 	ip_ra_control(sk, 0, NULL);
704 
705 	sk_common_release(sk);
706 }
707 
708 static void raw_destroy(struct sock *sk)
709 {
710 	lock_sock(sk);
711 	ip_flush_pending_frames(sk);
712 	release_sock(sk);
713 }
714 
715 /* This gets rid of all the nasties in af_inet. -DaveM */
716 static int raw_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len)
717 {
718 	struct inet_sock *inet = inet_sk(sk);
719 	struct sockaddr_in *addr = (struct sockaddr_in *) uaddr;
720 	u32 tb_id = RT_TABLE_LOCAL;
721 	int ret = -EINVAL;
722 	int chk_addr_ret;
723 
724 	if (sk->sk_state != TCP_CLOSE || addr_len < sizeof(struct sockaddr_in))
725 		goto out;
726 
727 	if (sk->sk_bound_dev_if)
728 		tb_id = l3mdev_fib_table_by_index(sock_net(sk),
729 						 sk->sk_bound_dev_if) ? : tb_id;
730 
731 	chk_addr_ret = inet_addr_type_table(sock_net(sk), addr->sin_addr.s_addr,
732 					    tb_id);
733 
734 	ret = -EADDRNOTAVAIL;
735 	if (addr->sin_addr.s_addr && chk_addr_ret != RTN_LOCAL &&
736 	    chk_addr_ret != RTN_MULTICAST && chk_addr_ret != RTN_BROADCAST)
737 		goto out;
738 	inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
739 	if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
740 		inet->inet_saddr = 0;  /* Use device */
741 	sk_dst_reset(sk);
742 	ret = 0;
743 out:	return ret;
744 }
745 
746 /*
747  *	This should be easy, if there is something there
748  *	we return it, otherwise we block.
749  */
750 
751 static int raw_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
752 		       int noblock, int flags, int *addr_len)
753 {
754 	struct inet_sock *inet = inet_sk(sk);
755 	size_t copied = 0;
756 	int err = -EOPNOTSUPP;
757 	DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
758 	struct sk_buff *skb;
759 
760 	if (flags & MSG_OOB)
761 		goto out;
762 
763 	if (flags & MSG_ERRQUEUE) {
764 		err = ip_recv_error(sk, msg, len, addr_len);
765 		goto out;
766 	}
767 
768 	skb = skb_recv_datagram(sk, flags, noblock, &err);
769 	if (!skb)
770 		goto out;
771 
772 	copied = skb->len;
773 	if (len < copied) {
774 		msg->msg_flags |= MSG_TRUNC;
775 		copied = len;
776 	}
777 
778 	err = skb_copy_datagram_msg(skb, 0, msg, copied);
779 	if (err)
780 		goto done;
781 
782 	sock_recv_ts_and_drops(msg, sk, skb);
783 
784 	/* Copy the address. */
785 	if (sin) {
786 		sin->sin_family = AF_INET;
787 		sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
788 		sin->sin_port = 0;
789 		memset(&sin->sin_zero, 0, sizeof(sin->sin_zero));
790 		*addr_len = sizeof(*sin);
791 	}
792 	if (inet->cmsg_flags)
793 		ip_cmsg_recv(msg, skb);
794 	if (flags & MSG_TRUNC)
795 		copied = skb->len;
796 done:
797 	skb_free_datagram(sk, skb);
798 out:
799 	if (err)
800 		return err;
801 	return copied;
802 }
803 
804 static int raw_sk_init(struct sock *sk)
805 {
806 	struct raw_sock *rp = raw_sk(sk);
807 
808 	if (inet_sk(sk)->inet_num == IPPROTO_ICMP)
809 		memset(&rp->filter, 0, sizeof(rp->filter));
810 	return 0;
811 }
812 
813 static int raw_seticmpfilter(struct sock *sk, sockptr_t optval, int optlen)
814 {
815 	if (optlen > sizeof(struct icmp_filter))
816 		optlen = sizeof(struct icmp_filter);
817 	if (copy_from_sockptr(&raw_sk(sk)->filter, optval, optlen))
818 		return -EFAULT;
819 	return 0;
820 }
821 
822 static int raw_geticmpfilter(struct sock *sk, char __user *optval, int __user *optlen)
823 {
824 	int len, ret = -EFAULT;
825 
826 	if (get_user(len, optlen))
827 		goto out;
828 	ret = -EINVAL;
829 	if (len < 0)
830 		goto out;
831 	if (len > sizeof(struct icmp_filter))
832 		len = sizeof(struct icmp_filter);
833 	ret = -EFAULT;
834 	if (put_user(len, optlen) ||
835 	    copy_to_user(optval, &raw_sk(sk)->filter, len))
836 		goto out;
837 	ret = 0;
838 out:	return ret;
839 }
840 
841 static int do_raw_setsockopt(struct sock *sk, int level, int optname,
842 			     sockptr_t optval, unsigned int optlen)
843 {
844 	if (optname == ICMP_FILTER) {
845 		if (inet_sk(sk)->inet_num != IPPROTO_ICMP)
846 			return -EOPNOTSUPP;
847 		else
848 			return raw_seticmpfilter(sk, optval, optlen);
849 	}
850 	return -ENOPROTOOPT;
851 }
852 
853 static int raw_setsockopt(struct sock *sk, int level, int optname,
854 			  sockptr_t optval, unsigned int optlen)
855 {
856 	if (level != SOL_RAW)
857 		return ip_setsockopt(sk, level, optname, optval, optlen);
858 	return do_raw_setsockopt(sk, level, optname, optval, optlen);
859 }
860 
861 static int do_raw_getsockopt(struct sock *sk, int level, int optname,
862 			  char __user *optval, int __user *optlen)
863 {
864 	if (optname == ICMP_FILTER) {
865 		if (inet_sk(sk)->inet_num != IPPROTO_ICMP)
866 			return -EOPNOTSUPP;
867 		else
868 			return raw_geticmpfilter(sk, optval, optlen);
869 	}
870 	return -ENOPROTOOPT;
871 }
872 
873 static int raw_getsockopt(struct sock *sk, int level, int optname,
874 			  char __user *optval, int __user *optlen)
875 {
876 	if (level != SOL_RAW)
877 		return ip_getsockopt(sk, level, optname, optval, optlen);
878 	return do_raw_getsockopt(sk, level, optname, optval, optlen);
879 }
880 
881 static int raw_ioctl(struct sock *sk, int cmd, unsigned long arg)
882 {
883 	switch (cmd) {
884 	case SIOCOUTQ: {
885 		int amount = sk_wmem_alloc_get(sk);
886 
887 		return put_user(amount, (int __user *)arg);
888 	}
889 	case SIOCINQ: {
890 		struct sk_buff *skb;
891 		int amount = 0;
892 
893 		spin_lock_bh(&sk->sk_receive_queue.lock);
894 		skb = skb_peek(&sk->sk_receive_queue);
895 		if (skb)
896 			amount = skb->len;
897 		spin_unlock_bh(&sk->sk_receive_queue.lock);
898 		return put_user(amount, (int __user *)arg);
899 	}
900 
901 	default:
902 #ifdef CONFIG_IP_MROUTE
903 		return ipmr_ioctl(sk, cmd, (void __user *)arg);
904 #else
905 		return -ENOIOCTLCMD;
906 #endif
907 	}
908 }
909 
910 #ifdef CONFIG_COMPAT
911 static int compat_raw_ioctl(struct sock *sk, unsigned int cmd, unsigned long arg)
912 {
913 	switch (cmd) {
914 	case SIOCOUTQ:
915 	case SIOCINQ:
916 		return -ENOIOCTLCMD;
917 	default:
918 #ifdef CONFIG_IP_MROUTE
919 		return ipmr_compat_ioctl(sk, cmd, compat_ptr(arg));
920 #else
921 		return -ENOIOCTLCMD;
922 #endif
923 	}
924 }
925 #endif
926 
927 int raw_abort(struct sock *sk, int err)
928 {
929 	lock_sock(sk);
930 
931 	sk->sk_err = err;
932 	sk->sk_error_report(sk);
933 	__udp_disconnect(sk, 0);
934 
935 	release_sock(sk);
936 
937 	return 0;
938 }
939 EXPORT_SYMBOL_GPL(raw_abort);
940 
941 struct proto raw_prot = {
942 	.name		   = "RAW",
943 	.owner		   = THIS_MODULE,
944 	.close		   = raw_close,
945 	.destroy	   = raw_destroy,
946 	.connect	   = ip4_datagram_connect,
947 	.disconnect	   = __udp_disconnect,
948 	.ioctl		   = raw_ioctl,
949 	.init		   = raw_sk_init,
950 	.setsockopt	   = raw_setsockopt,
951 	.getsockopt	   = raw_getsockopt,
952 	.sendmsg	   = raw_sendmsg,
953 	.recvmsg	   = raw_recvmsg,
954 	.bind		   = raw_bind,
955 	.backlog_rcv	   = raw_rcv_skb,
956 	.release_cb	   = ip4_datagram_release_cb,
957 	.hash		   = raw_hash_sk,
958 	.unhash		   = raw_unhash_sk,
959 	.obj_size	   = sizeof(struct raw_sock),
960 	.useroffset	   = offsetof(struct raw_sock, filter),
961 	.usersize	   = sizeof_field(struct raw_sock, filter),
962 	.h.raw_hash	   = &raw_v4_hashinfo,
963 #ifdef CONFIG_COMPAT
964 	.compat_ioctl	   = compat_raw_ioctl,
965 #endif
966 	.diag_destroy	   = raw_abort,
967 };
968 
969 #ifdef CONFIG_PROC_FS
970 static struct sock *raw_get_first(struct seq_file *seq)
971 {
972 	struct sock *sk;
973 	struct raw_hashinfo *h = PDE_DATA(file_inode(seq->file));
974 	struct raw_iter_state *state = raw_seq_private(seq);
975 
976 	for (state->bucket = 0; state->bucket < RAW_HTABLE_SIZE;
977 			++state->bucket) {
978 		sk_for_each(sk, &h->ht[state->bucket])
979 			if (sock_net(sk) == seq_file_net(seq))
980 				goto found;
981 	}
982 	sk = NULL;
983 found:
984 	return sk;
985 }
986 
987 static struct sock *raw_get_next(struct seq_file *seq, struct sock *sk)
988 {
989 	struct raw_hashinfo *h = PDE_DATA(file_inode(seq->file));
990 	struct raw_iter_state *state = raw_seq_private(seq);
991 
992 	do {
993 		sk = sk_next(sk);
994 try_again:
995 		;
996 	} while (sk && sock_net(sk) != seq_file_net(seq));
997 
998 	if (!sk && ++state->bucket < RAW_HTABLE_SIZE) {
999 		sk = sk_head(&h->ht[state->bucket]);
1000 		goto try_again;
1001 	}
1002 	return sk;
1003 }
1004 
1005 static struct sock *raw_get_idx(struct seq_file *seq, loff_t pos)
1006 {
1007 	struct sock *sk = raw_get_first(seq);
1008 
1009 	if (sk)
1010 		while (pos && (sk = raw_get_next(seq, sk)) != NULL)
1011 			--pos;
1012 	return pos ? NULL : sk;
1013 }
1014 
1015 void *raw_seq_start(struct seq_file *seq, loff_t *pos)
1016 	__acquires(&h->lock)
1017 {
1018 	struct raw_hashinfo *h = PDE_DATA(file_inode(seq->file));
1019 
1020 	read_lock(&h->lock);
1021 	return *pos ? raw_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
1022 }
1023 EXPORT_SYMBOL_GPL(raw_seq_start);
1024 
1025 void *raw_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1026 {
1027 	struct sock *sk;
1028 
1029 	if (v == SEQ_START_TOKEN)
1030 		sk = raw_get_first(seq);
1031 	else
1032 		sk = raw_get_next(seq, v);
1033 	++*pos;
1034 	return sk;
1035 }
1036 EXPORT_SYMBOL_GPL(raw_seq_next);
1037 
1038 void raw_seq_stop(struct seq_file *seq, void *v)
1039 	__releases(&h->lock)
1040 {
1041 	struct raw_hashinfo *h = PDE_DATA(file_inode(seq->file));
1042 
1043 	read_unlock(&h->lock);
1044 }
1045 EXPORT_SYMBOL_GPL(raw_seq_stop);
1046 
1047 static void raw_sock_seq_show(struct seq_file *seq, struct sock *sp, int i)
1048 {
1049 	struct inet_sock *inet = inet_sk(sp);
1050 	__be32 dest = inet->inet_daddr,
1051 	       src = inet->inet_rcv_saddr;
1052 	__u16 destp = 0,
1053 	      srcp  = inet->inet_num;
1054 
1055 	seq_printf(seq, "%4d: %08X:%04X %08X:%04X"
1056 		" %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %u\n",
1057 		i, src, srcp, dest, destp, sp->sk_state,
1058 		sk_wmem_alloc_get(sp),
1059 		sk_rmem_alloc_get(sp),
1060 		0, 0L, 0,
1061 		from_kuid_munged(seq_user_ns(seq), sock_i_uid(sp)),
1062 		0, sock_i_ino(sp),
1063 		refcount_read(&sp->sk_refcnt), sp, atomic_read(&sp->sk_drops));
1064 }
1065 
1066 static int raw_seq_show(struct seq_file *seq, void *v)
1067 {
1068 	if (v == SEQ_START_TOKEN)
1069 		seq_printf(seq, "  sl  local_address rem_address   st tx_queue "
1070 				"rx_queue tr tm->when retrnsmt   uid  timeout "
1071 				"inode ref pointer drops\n");
1072 	else
1073 		raw_sock_seq_show(seq, v, raw_seq_private(seq)->bucket);
1074 	return 0;
1075 }
1076 
1077 static const struct seq_operations raw_seq_ops = {
1078 	.start = raw_seq_start,
1079 	.next  = raw_seq_next,
1080 	.stop  = raw_seq_stop,
1081 	.show  = raw_seq_show,
1082 };
1083 
1084 static __net_init int raw_init_net(struct net *net)
1085 {
1086 	if (!proc_create_net_data("raw", 0444, net->proc_net, &raw_seq_ops,
1087 			sizeof(struct raw_iter_state), &raw_v4_hashinfo))
1088 		return -ENOMEM;
1089 
1090 	return 0;
1091 }
1092 
1093 static __net_exit void raw_exit_net(struct net *net)
1094 {
1095 	remove_proc_entry("raw", net->proc_net);
1096 }
1097 
1098 static __net_initdata struct pernet_operations raw_net_ops = {
1099 	.init = raw_init_net,
1100 	.exit = raw_exit_net,
1101 };
1102 
1103 int __init raw_proc_init(void)
1104 {
1105 	return register_pernet_subsys(&raw_net_ops);
1106 }
1107 
1108 void __init raw_proc_exit(void)
1109 {
1110 	unregister_pernet_subsys(&raw_net_ops);
1111 }
1112 #endif /* CONFIG_PROC_FS */
1113 
1114 static void raw_sysctl_init_net(struct net *net)
1115 {
1116 #ifdef CONFIG_NET_L3_MASTER_DEV
1117 	net->ipv4.sysctl_raw_l3mdev_accept = 1;
1118 #endif
1119 }
1120 
1121 static int __net_init raw_sysctl_init(struct net *net)
1122 {
1123 	raw_sysctl_init_net(net);
1124 	return 0;
1125 }
1126 
1127 static struct pernet_operations __net_initdata raw_sysctl_ops = {
1128 	.init	= raw_sysctl_init,
1129 };
1130 
1131 void __init raw_init(void)
1132 {
1133 	raw_sysctl_init_net(&init_net);
1134 	if (register_pernet_subsys(&raw_sysctl_ops))
1135 		panic("RAW: failed to init sysctl parameters.\n");
1136 }
1137