xref: /linux/net/dccp/ipv4.c (revision 2fe05e1139a555ae91f00a812cb9520e7d3022ab)
1 /*
2  *  net/dccp/ipv4.c
3  *
4  *  An implementation of the DCCP protocol
5  *  Arnaldo Carvalho de Melo <acme@conectiva.com.br>
6  *
7  *	This program is free software; you can redistribute it and/or
8  *	modify it under the terms of the GNU General Public License
9  *	as published by the Free Software Foundation; either version
10  *	2 of the License, or (at your option) any later version.
11  */
12 
13 #include <linux/dccp.h>
14 #include <linux/icmp.h>
15 #include <linux/slab.h>
16 #include <linux/module.h>
17 #include <linux/skbuff.h>
18 #include <linux/random.h>
19 
20 #include <net/icmp.h>
21 #include <net/inet_common.h>
22 #include <net/inet_hashtables.h>
23 #include <net/inet_sock.h>
24 #include <net/protocol.h>
25 #include <net/sock.h>
26 #include <net/timewait_sock.h>
27 #include <net/tcp_states.h>
28 #include <net/xfrm.h>
29 #include <net/secure_seq.h>
30 
31 #include "ackvec.h"
32 #include "ccid.h"
33 #include "dccp.h"
34 #include "feat.h"
35 
36 /*
37  * The per-net dccp.v4_ctl_sk socket is used for responding to
38  * the Out-of-the-blue (OOTB) packets. A control sock will be created
39  * for this socket at the initialization time.
40  */
41 
42 int dccp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
43 {
44 	const struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
45 	struct inet_sock *inet = inet_sk(sk);
46 	struct dccp_sock *dp = dccp_sk(sk);
47 	__be16 orig_sport, orig_dport;
48 	__be32 daddr, nexthop;
49 	struct flowi4 *fl4;
50 	struct rtable *rt;
51 	int err;
52 	struct ip_options_rcu *inet_opt;
53 
54 	dp->dccps_role = DCCP_ROLE_CLIENT;
55 
56 	if (addr_len < sizeof(struct sockaddr_in))
57 		return -EINVAL;
58 
59 	if (usin->sin_family != AF_INET)
60 		return -EAFNOSUPPORT;
61 
62 	nexthop = daddr = usin->sin_addr.s_addr;
63 
64 	inet_opt = rcu_dereference_protected(inet->inet_opt,
65 					     lockdep_sock_is_held(sk));
66 	if (inet_opt != NULL && inet_opt->opt.srr) {
67 		if (daddr == 0)
68 			return -EINVAL;
69 		nexthop = inet_opt->opt.faddr;
70 	}
71 
72 	orig_sport = inet->inet_sport;
73 	orig_dport = usin->sin_port;
74 	fl4 = &inet->cork.fl.u.ip4;
75 	rt = ip_route_connect(fl4, nexthop, inet->inet_saddr,
76 			      RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
77 			      IPPROTO_DCCP,
78 			      orig_sport, orig_dport, sk);
79 	if (IS_ERR(rt))
80 		return PTR_ERR(rt);
81 
82 	if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
83 		ip_rt_put(rt);
84 		return -ENETUNREACH;
85 	}
86 
87 	if (inet_opt == NULL || !inet_opt->opt.srr)
88 		daddr = fl4->daddr;
89 
90 	if (inet->inet_saddr == 0)
91 		inet->inet_saddr = fl4->saddr;
92 	sk_rcv_saddr_set(sk, inet->inet_saddr);
93 	inet->inet_dport = usin->sin_port;
94 	sk_daddr_set(sk, daddr);
95 
96 	inet_csk(sk)->icsk_ext_hdr_len = 0;
97 	if (inet_opt)
98 		inet_csk(sk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
99 	/*
100 	 * Socket identity is still unknown (sport may be zero).
101 	 * However we set state to DCCP_REQUESTING and not releasing socket
102 	 * lock select source port, enter ourselves into the hash tables and
103 	 * complete initialization after this.
104 	 */
105 	dccp_set_state(sk, DCCP_REQUESTING);
106 	err = inet_hash_connect(&dccp_death_row, sk);
107 	if (err != 0)
108 		goto failure;
109 
110 	rt = ip_route_newports(fl4, rt, orig_sport, orig_dport,
111 			       inet->inet_sport, inet->inet_dport, sk);
112 	if (IS_ERR(rt)) {
113 		err = PTR_ERR(rt);
114 		rt = NULL;
115 		goto failure;
116 	}
117 	/* OK, now commit destination to socket.  */
118 	sk_setup_caps(sk, &rt->dst);
119 
120 	dp->dccps_iss = secure_dccp_sequence_number(inet->inet_saddr,
121 						    inet->inet_daddr,
122 						    inet->inet_sport,
123 						    inet->inet_dport);
124 	inet->inet_id = dp->dccps_iss ^ jiffies;
125 
126 	err = dccp_connect(sk);
127 	rt = NULL;
128 	if (err != 0)
129 		goto failure;
130 out:
131 	return err;
132 failure:
133 	/*
134 	 * This unhashes the socket and releases the local port, if necessary.
135 	 */
136 	dccp_set_state(sk, DCCP_CLOSED);
137 	ip_rt_put(rt);
138 	sk->sk_route_caps = 0;
139 	inet->inet_dport = 0;
140 	goto out;
141 }
142 EXPORT_SYMBOL_GPL(dccp_v4_connect);
143 
144 /*
145  * This routine does path mtu discovery as defined in RFC1191.
146  */
147 static inline void dccp_do_pmtu_discovery(struct sock *sk,
148 					  const struct iphdr *iph,
149 					  u32 mtu)
150 {
151 	struct dst_entry *dst;
152 	const struct inet_sock *inet = inet_sk(sk);
153 	const struct dccp_sock *dp = dccp_sk(sk);
154 
155 	/* We are not interested in DCCP_LISTEN and request_socks (RESPONSEs
156 	 * send out by Linux are always < 576bytes so they should go through
157 	 * unfragmented).
158 	 */
159 	if (sk->sk_state == DCCP_LISTEN)
160 		return;
161 
162 	dst = inet_csk_update_pmtu(sk, mtu);
163 	if (!dst)
164 		return;
165 
166 	/* Something is about to be wrong... Remember soft error
167 	 * for the case, if this connection will not able to recover.
168 	 */
169 	if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
170 		sk->sk_err_soft = EMSGSIZE;
171 
172 	mtu = dst_mtu(dst);
173 
174 	if (inet->pmtudisc != IP_PMTUDISC_DONT &&
175 	    ip_sk_accept_pmtu(sk) &&
176 	    inet_csk(sk)->icsk_pmtu_cookie > mtu) {
177 		dccp_sync_mss(sk, mtu);
178 
179 		/*
180 		 * From RFC 4340, sec. 14.1:
181 		 *
182 		 *	DCCP-Sync packets are the best choice for upward
183 		 *	probing, since DCCP-Sync probes do not risk application
184 		 *	data loss.
185 		 */
186 		dccp_send_sync(sk, dp->dccps_gsr, DCCP_PKT_SYNC);
187 	} /* else let the usual retransmit timer handle it */
188 }
189 
190 static void dccp_do_redirect(struct sk_buff *skb, struct sock *sk)
191 {
192 	struct dst_entry *dst = __sk_dst_check(sk, 0);
193 
194 	if (dst)
195 		dst->ops->redirect(dst, sk, skb);
196 }
197 
198 void dccp_req_err(struct sock *sk, u64 seq)
199 	{
200 	struct request_sock *req = inet_reqsk(sk);
201 	struct net *net = sock_net(sk);
202 
203 	/*
204 	 * ICMPs are not backlogged, hence we cannot get an established
205 	 * socket here.
206 	 */
207 	if (!between48(seq, dccp_rsk(req)->dreq_iss, dccp_rsk(req)->dreq_gss)) {
208 		__NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
209 	} else {
210 		/*
211 		 * Still in RESPOND, just remove it silently.
212 		 * There is no good way to pass the error to the newly
213 		 * created socket, and POSIX does not want network
214 		 * errors returned from accept().
215 		 */
216 		inet_csk_reqsk_queue_drop(req->rsk_listener, req);
217 	}
218 	reqsk_put(req);
219 }
220 EXPORT_SYMBOL(dccp_req_err);
221 
222 /*
223  * This routine is called by the ICMP module when it gets some sort of error
224  * condition. If err < 0 then the socket should be closed and the error
225  * returned to the user. If err > 0 it's just the icmp type << 8 | icmp code.
226  * After adjustment header points to the first 8 bytes of the tcp header. We
227  * need to find the appropriate port.
228  *
229  * The locking strategy used here is very "optimistic". When someone else
230  * accesses the socket the ICMP is just dropped and for some paths there is no
231  * check at all. A more general error queue to queue errors for later handling
232  * is probably better.
233  */
234 static void dccp_v4_err(struct sk_buff *skb, u32 info)
235 {
236 	const struct iphdr *iph = (struct iphdr *)skb->data;
237 	const u8 offset = iph->ihl << 2;
238 	const struct dccp_hdr *dh;
239 	struct dccp_sock *dp;
240 	struct inet_sock *inet;
241 	const int type = icmp_hdr(skb)->type;
242 	const int code = icmp_hdr(skb)->code;
243 	struct sock *sk;
244 	__u64 seq;
245 	int err;
246 	struct net *net = dev_net(skb->dev);
247 
248 	/* Only need dccph_dport & dccph_sport which are the first
249 	 * 4 bytes in dccp header.
250 	 * Our caller (icmp_socket_deliver()) already pulled 8 bytes for us.
251 	 */
252 	BUILD_BUG_ON(offsetofend(struct dccp_hdr, dccph_sport) > 8);
253 	BUILD_BUG_ON(offsetofend(struct dccp_hdr, dccph_dport) > 8);
254 	dh = (struct dccp_hdr *)(skb->data + offset);
255 
256 	sk = __inet_lookup_established(net, &dccp_hashinfo,
257 				       iph->daddr, dh->dccph_dport,
258 				       iph->saddr, ntohs(dh->dccph_sport),
259 				       inet_iif(skb));
260 	if (!sk) {
261 		__ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
262 		return;
263 	}
264 
265 	if (sk->sk_state == DCCP_TIME_WAIT) {
266 		inet_twsk_put(inet_twsk(sk));
267 		return;
268 	}
269 	seq = dccp_hdr_seq(dh);
270 	if (sk->sk_state == DCCP_NEW_SYN_RECV)
271 		return dccp_req_err(sk, seq);
272 
273 	bh_lock_sock(sk);
274 	/* If too many ICMPs get dropped on busy
275 	 * servers this needs to be solved differently.
276 	 */
277 	if (sock_owned_by_user(sk))
278 		__NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS);
279 
280 	if (sk->sk_state == DCCP_CLOSED)
281 		goto out;
282 
283 	dp = dccp_sk(sk);
284 	if ((1 << sk->sk_state) & ~(DCCPF_REQUESTING | DCCPF_LISTEN) &&
285 	    !between48(seq, dp->dccps_awl, dp->dccps_awh)) {
286 		__NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
287 		goto out;
288 	}
289 
290 	switch (type) {
291 	case ICMP_REDIRECT:
292 		if (!sock_owned_by_user(sk))
293 			dccp_do_redirect(skb, sk);
294 		goto out;
295 	case ICMP_SOURCE_QUENCH:
296 		/* Just silently ignore these. */
297 		goto out;
298 	case ICMP_PARAMETERPROB:
299 		err = EPROTO;
300 		break;
301 	case ICMP_DEST_UNREACH:
302 		if (code > NR_ICMP_UNREACH)
303 			goto out;
304 
305 		if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
306 			if (!sock_owned_by_user(sk))
307 				dccp_do_pmtu_discovery(sk, iph, info);
308 			goto out;
309 		}
310 
311 		err = icmp_err_convert[code].errno;
312 		break;
313 	case ICMP_TIME_EXCEEDED:
314 		err = EHOSTUNREACH;
315 		break;
316 	default:
317 		goto out;
318 	}
319 
320 	switch (sk->sk_state) {
321 	case DCCP_REQUESTING:
322 	case DCCP_RESPOND:
323 		if (!sock_owned_by_user(sk)) {
324 			__DCCP_INC_STATS(DCCP_MIB_ATTEMPTFAILS);
325 			sk->sk_err = err;
326 
327 			sk->sk_error_report(sk);
328 
329 			dccp_done(sk);
330 		} else
331 			sk->sk_err_soft = err;
332 		goto out;
333 	}
334 
335 	/* If we've already connected we will keep trying
336 	 * until we time out, or the user gives up.
337 	 *
338 	 * rfc1122 4.2.3.9 allows to consider as hard errors
339 	 * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
340 	 * but it is obsoleted by pmtu discovery).
341 	 *
342 	 * Note, that in modern internet, where routing is unreliable
343 	 * and in each dark corner broken firewalls sit, sending random
344 	 * errors ordered by their masters even this two messages finally lose
345 	 * their original sense (even Linux sends invalid PORT_UNREACHs)
346 	 *
347 	 * Now we are in compliance with RFCs.
348 	 *							--ANK (980905)
349 	 */
350 
351 	inet = inet_sk(sk);
352 	if (!sock_owned_by_user(sk) && inet->recverr) {
353 		sk->sk_err = err;
354 		sk->sk_error_report(sk);
355 	} else /* Only an error on timeout */
356 		sk->sk_err_soft = err;
357 out:
358 	bh_unlock_sock(sk);
359 	sock_put(sk);
360 }
361 
362 static inline __sum16 dccp_v4_csum_finish(struct sk_buff *skb,
363 				      __be32 src, __be32 dst)
364 {
365 	return csum_tcpudp_magic(src, dst, skb->len, IPPROTO_DCCP, skb->csum);
366 }
367 
368 void dccp_v4_send_check(struct sock *sk, struct sk_buff *skb)
369 {
370 	const struct inet_sock *inet = inet_sk(sk);
371 	struct dccp_hdr *dh = dccp_hdr(skb);
372 
373 	dccp_csum_outgoing(skb);
374 	dh->dccph_checksum = dccp_v4_csum_finish(skb,
375 						 inet->inet_saddr,
376 						 inet->inet_daddr);
377 }
378 EXPORT_SYMBOL_GPL(dccp_v4_send_check);
379 
380 static inline u64 dccp_v4_init_sequence(const struct sk_buff *skb)
381 {
382 	return secure_dccp_sequence_number(ip_hdr(skb)->daddr,
383 					   ip_hdr(skb)->saddr,
384 					   dccp_hdr(skb)->dccph_dport,
385 					   dccp_hdr(skb)->dccph_sport);
386 }
387 
388 /*
389  * The three way handshake has completed - we got a valid ACK or DATAACK -
390  * now create the new socket.
391  *
392  * This is the equivalent of TCP's tcp_v4_syn_recv_sock
393  */
394 struct sock *dccp_v4_request_recv_sock(const struct sock *sk,
395 				       struct sk_buff *skb,
396 				       struct request_sock *req,
397 				       struct dst_entry *dst,
398 				       struct request_sock *req_unhash,
399 				       bool *own_req)
400 {
401 	struct inet_request_sock *ireq;
402 	struct inet_sock *newinet;
403 	struct sock *newsk;
404 
405 	if (sk_acceptq_is_full(sk))
406 		goto exit_overflow;
407 
408 	newsk = dccp_create_openreq_child(sk, req, skb);
409 	if (newsk == NULL)
410 		goto exit_nonewsk;
411 
412 	newinet		   = inet_sk(newsk);
413 	ireq		   = inet_rsk(req);
414 	sk_daddr_set(newsk, ireq->ir_rmt_addr);
415 	sk_rcv_saddr_set(newsk, ireq->ir_loc_addr);
416 	newinet->inet_saddr	= ireq->ir_loc_addr;
417 	newinet->inet_opt	= ireq->opt;
418 	ireq->opt	   = NULL;
419 	newinet->mc_index  = inet_iif(skb);
420 	newinet->mc_ttl	   = ip_hdr(skb)->ttl;
421 	newinet->inet_id   = jiffies;
422 
423 	if (dst == NULL && (dst = inet_csk_route_child_sock(sk, newsk, req)) == NULL)
424 		goto put_and_exit;
425 
426 	sk_setup_caps(newsk, dst);
427 
428 	dccp_sync_mss(newsk, dst_mtu(dst));
429 
430 	if (__inet_inherit_port(sk, newsk) < 0)
431 		goto put_and_exit;
432 	*own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash));
433 
434 	return newsk;
435 
436 exit_overflow:
437 	__NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
438 exit_nonewsk:
439 	dst_release(dst);
440 exit:
441 	__NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS);
442 	return NULL;
443 put_and_exit:
444 	inet_csk_prepare_forced_close(newsk);
445 	dccp_done(newsk);
446 	goto exit;
447 }
448 EXPORT_SYMBOL_GPL(dccp_v4_request_recv_sock);
449 
450 static struct dst_entry* dccp_v4_route_skb(struct net *net, struct sock *sk,
451 					   struct sk_buff *skb)
452 {
453 	struct rtable *rt;
454 	const struct iphdr *iph = ip_hdr(skb);
455 	struct flowi4 fl4 = {
456 		.flowi4_oif = inet_iif(skb),
457 		.daddr = iph->saddr,
458 		.saddr = iph->daddr,
459 		.flowi4_tos = RT_CONN_FLAGS(sk),
460 		.flowi4_proto = sk->sk_protocol,
461 		.fl4_sport = dccp_hdr(skb)->dccph_dport,
462 		.fl4_dport = dccp_hdr(skb)->dccph_sport,
463 	};
464 
465 	security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
466 	rt = ip_route_output_flow(net, &fl4, sk);
467 	if (IS_ERR(rt)) {
468 		IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
469 		return NULL;
470 	}
471 
472 	return &rt->dst;
473 }
474 
475 static int dccp_v4_send_response(const struct sock *sk, struct request_sock *req)
476 {
477 	int err = -1;
478 	struct sk_buff *skb;
479 	struct dst_entry *dst;
480 	struct flowi4 fl4;
481 
482 	dst = inet_csk_route_req(sk, &fl4, req);
483 	if (dst == NULL)
484 		goto out;
485 
486 	skb = dccp_make_response(sk, dst, req);
487 	if (skb != NULL) {
488 		const struct inet_request_sock *ireq = inet_rsk(req);
489 		struct dccp_hdr *dh = dccp_hdr(skb);
490 
491 		dh->dccph_checksum = dccp_v4_csum_finish(skb, ireq->ir_loc_addr,
492 							      ireq->ir_rmt_addr);
493 		err = ip_build_and_send_pkt(skb, sk, ireq->ir_loc_addr,
494 					    ireq->ir_rmt_addr,
495 					    ireq->opt);
496 		err = net_xmit_eval(err);
497 	}
498 
499 out:
500 	dst_release(dst);
501 	return err;
502 }
503 
504 static void dccp_v4_ctl_send_reset(const struct sock *sk, struct sk_buff *rxskb)
505 {
506 	int err;
507 	const struct iphdr *rxiph;
508 	struct sk_buff *skb;
509 	struct dst_entry *dst;
510 	struct net *net = dev_net(skb_dst(rxskb)->dev);
511 	struct sock *ctl_sk = net->dccp.v4_ctl_sk;
512 
513 	/* Never send a reset in response to a reset. */
514 	if (dccp_hdr(rxskb)->dccph_type == DCCP_PKT_RESET)
515 		return;
516 
517 	if (skb_rtable(rxskb)->rt_type != RTN_LOCAL)
518 		return;
519 
520 	dst = dccp_v4_route_skb(net, ctl_sk, rxskb);
521 	if (dst == NULL)
522 		return;
523 
524 	skb = dccp_ctl_make_reset(ctl_sk, rxskb);
525 	if (skb == NULL)
526 		goto out;
527 
528 	rxiph = ip_hdr(rxskb);
529 	dccp_hdr(skb)->dccph_checksum = dccp_v4_csum_finish(skb, rxiph->saddr,
530 								 rxiph->daddr);
531 	skb_dst_set(skb, dst_clone(dst));
532 
533 	local_bh_disable();
534 	bh_lock_sock(ctl_sk);
535 	err = ip_build_and_send_pkt(skb, ctl_sk,
536 				    rxiph->daddr, rxiph->saddr, NULL);
537 	bh_unlock_sock(ctl_sk);
538 
539 	if (net_xmit_eval(err) == 0) {
540 		__DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
541 		__DCCP_INC_STATS(DCCP_MIB_OUTRSTS);
542 	}
543 	local_bh_enable();
544 out:
545 	dst_release(dst);
546 }
547 
548 static void dccp_v4_reqsk_destructor(struct request_sock *req)
549 {
550 	dccp_feat_list_purge(&dccp_rsk(req)->dreq_featneg);
551 	kfree(inet_rsk(req)->opt);
552 }
553 
554 void dccp_syn_ack_timeout(const struct request_sock *req)
555 {
556 }
557 EXPORT_SYMBOL(dccp_syn_ack_timeout);
558 
559 static struct request_sock_ops dccp_request_sock_ops __read_mostly = {
560 	.family		= PF_INET,
561 	.obj_size	= sizeof(struct dccp_request_sock),
562 	.rtx_syn_ack	= dccp_v4_send_response,
563 	.send_ack	= dccp_reqsk_send_ack,
564 	.destructor	= dccp_v4_reqsk_destructor,
565 	.send_reset	= dccp_v4_ctl_send_reset,
566 	.syn_ack_timeout = dccp_syn_ack_timeout,
567 };
568 
569 int dccp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
570 {
571 	struct inet_request_sock *ireq;
572 	struct request_sock *req;
573 	struct dccp_request_sock *dreq;
574 	const __be32 service = dccp_hdr_request(skb)->dccph_req_service;
575 	struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
576 
577 	/* Never answer to DCCP_PKT_REQUESTs send to broadcast or multicast */
578 	if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
579 		return 0;	/* discard, don't send a reset here */
580 
581 	if (dccp_bad_service_code(sk, service)) {
582 		dcb->dccpd_reset_code = DCCP_RESET_CODE_BAD_SERVICE_CODE;
583 		goto drop;
584 	}
585 	/*
586 	 * TW buckets are converted to open requests without
587 	 * limitations, they conserve resources and peer is
588 	 * evidently real one.
589 	 */
590 	dcb->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY;
591 	if (inet_csk_reqsk_queue_is_full(sk))
592 		goto drop;
593 
594 	if (sk_acceptq_is_full(sk))
595 		goto drop;
596 
597 	req = inet_reqsk_alloc(&dccp_request_sock_ops, sk, true);
598 	if (req == NULL)
599 		goto drop;
600 
601 	if (dccp_reqsk_init(req, dccp_sk(sk), skb))
602 		goto drop_and_free;
603 
604 	dreq = dccp_rsk(req);
605 	if (dccp_parse_options(sk, dreq, skb))
606 		goto drop_and_free;
607 
608 	if (security_inet_conn_request(sk, skb, req))
609 		goto drop_and_free;
610 
611 	ireq = inet_rsk(req);
612 	sk_rcv_saddr_set(req_to_sk(req), ip_hdr(skb)->daddr);
613 	sk_daddr_set(req_to_sk(req), ip_hdr(skb)->saddr);
614 	ireq->ireq_family = AF_INET;
615 	ireq->ir_iif = sk->sk_bound_dev_if;
616 
617 	/*
618 	 * Step 3: Process LISTEN state
619 	 *
620 	 * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
621 	 *
622 	 * Setting S.SWL/S.SWH to is deferred to dccp_create_openreq_child().
623 	 */
624 	dreq->dreq_isr	   = dcb->dccpd_seq;
625 	dreq->dreq_gsr	   = dreq->dreq_isr;
626 	dreq->dreq_iss	   = dccp_v4_init_sequence(skb);
627 	dreq->dreq_gss     = dreq->dreq_iss;
628 	dreq->dreq_service = service;
629 
630 	if (dccp_v4_send_response(sk, req))
631 		goto drop_and_free;
632 
633 	inet_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT);
634 	return 0;
635 
636 drop_and_free:
637 	reqsk_free(req);
638 drop:
639 	__DCCP_INC_STATS(DCCP_MIB_ATTEMPTFAILS);
640 	return -1;
641 }
642 EXPORT_SYMBOL_GPL(dccp_v4_conn_request);
643 
644 int dccp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
645 {
646 	struct dccp_hdr *dh = dccp_hdr(skb);
647 
648 	if (sk->sk_state == DCCP_OPEN) { /* Fast path */
649 		if (dccp_rcv_established(sk, skb, dh, skb->len))
650 			goto reset;
651 		return 0;
652 	}
653 
654 	/*
655 	 *  Step 3: Process LISTEN state
656 	 *	 If P.type == Request or P contains a valid Init Cookie option,
657 	 *	      (* Must scan the packet's options to check for Init
658 	 *		 Cookies.  Only Init Cookies are processed here,
659 	 *		 however; other options are processed in Step 8.  This
660 	 *		 scan need only be performed if the endpoint uses Init
661 	 *		 Cookies *)
662 	 *	      (* Generate a new socket and switch to that socket *)
663 	 *	      Set S := new socket for this port pair
664 	 *	      S.state = RESPOND
665 	 *	      Choose S.ISS (initial seqno) or set from Init Cookies
666 	 *	      Initialize S.GAR := S.ISS
667 	 *	      Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookies
668 	 *	      Continue with S.state == RESPOND
669 	 *	      (* A Response packet will be generated in Step 11 *)
670 	 *	 Otherwise,
671 	 *	      Generate Reset(No Connection) unless P.type == Reset
672 	 *	      Drop packet and return
673 	 *
674 	 * NOTE: the check for the packet types is done in
675 	 *	 dccp_rcv_state_process
676 	 */
677 
678 	if (dccp_rcv_state_process(sk, skb, dh, skb->len))
679 		goto reset;
680 	return 0;
681 
682 reset:
683 	dccp_v4_ctl_send_reset(sk, skb);
684 	kfree_skb(skb);
685 	return 0;
686 }
687 EXPORT_SYMBOL_GPL(dccp_v4_do_rcv);
688 
689 /**
690  *	dccp_invalid_packet  -  check for malformed packets
691  *	Implements RFC 4340, 8.5:  Step 1: Check header basics
692  *	Packets that fail these checks are ignored and do not receive Resets.
693  */
694 int dccp_invalid_packet(struct sk_buff *skb)
695 {
696 	const struct dccp_hdr *dh;
697 	unsigned int cscov;
698 	u8 dccph_doff;
699 
700 	if (skb->pkt_type != PACKET_HOST)
701 		return 1;
702 
703 	/* If the packet is shorter than 12 bytes, drop packet and return */
704 	if (!pskb_may_pull(skb, sizeof(struct dccp_hdr))) {
705 		DCCP_WARN("pskb_may_pull failed\n");
706 		return 1;
707 	}
708 
709 	dh = dccp_hdr(skb);
710 
711 	/* If P.type is not understood, drop packet and return */
712 	if (dh->dccph_type >= DCCP_PKT_INVALID) {
713 		DCCP_WARN("invalid packet type\n");
714 		return 1;
715 	}
716 
717 	/*
718 	 * If P.Data Offset is too small for packet type, drop packet and return
719 	 */
720 	dccph_doff = dh->dccph_doff;
721 	if (dccph_doff < dccp_hdr_len(skb) / sizeof(u32)) {
722 		DCCP_WARN("P.Data Offset(%u) too small\n", dccph_doff);
723 		return 1;
724 	}
725 	/*
726 	 * If P.Data Offset is too too large for packet, drop packet and return
727 	 */
728 	if (!pskb_may_pull(skb, dccph_doff * sizeof(u32))) {
729 		DCCP_WARN("P.Data Offset(%u) too large\n", dccph_doff);
730 		return 1;
731 	}
732 	dh = dccp_hdr(skb);
733 	/*
734 	 * If P.type is not Data, Ack, or DataAck and P.X == 0 (the packet
735 	 * has short sequence numbers), drop packet and return
736 	 */
737 	if ((dh->dccph_type < DCCP_PKT_DATA    ||
738 	    dh->dccph_type > DCCP_PKT_DATAACK) && dh->dccph_x == 0)  {
739 		DCCP_WARN("P.type (%s) not Data || [Data]Ack, while P.X == 0\n",
740 			  dccp_packet_name(dh->dccph_type));
741 		return 1;
742 	}
743 
744 	/*
745 	 * If P.CsCov is too large for the packet size, drop packet and return.
746 	 * This must come _before_ checksumming (not as RFC 4340 suggests).
747 	 */
748 	cscov = dccp_csum_coverage(skb);
749 	if (cscov > skb->len) {
750 		DCCP_WARN("P.CsCov %u exceeds packet length %d\n",
751 			  dh->dccph_cscov, skb->len);
752 		return 1;
753 	}
754 
755 	/* If header checksum is incorrect, drop packet and return.
756 	 * (This step is completed in the AF-dependent functions.) */
757 	skb->csum = skb_checksum(skb, 0, cscov, 0);
758 
759 	return 0;
760 }
761 EXPORT_SYMBOL_GPL(dccp_invalid_packet);
762 
763 /* this is called when real data arrives */
764 static int dccp_v4_rcv(struct sk_buff *skb)
765 {
766 	const struct dccp_hdr *dh;
767 	const struct iphdr *iph;
768 	bool refcounted;
769 	struct sock *sk;
770 	int min_cov;
771 
772 	/* Step 1: Check header basics */
773 
774 	if (dccp_invalid_packet(skb))
775 		goto discard_it;
776 
777 	iph = ip_hdr(skb);
778 	/* Step 1: If header checksum is incorrect, drop packet and return */
779 	if (dccp_v4_csum_finish(skb, iph->saddr, iph->daddr)) {
780 		DCCP_WARN("dropped packet with invalid checksum\n");
781 		goto discard_it;
782 	}
783 
784 	dh = dccp_hdr(skb);
785 
786 	DCCP_SKB_CB(skb)->dccpd_seq  = dccp_hdr_seq(dh);
787 	DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type;
788 
789 	dccp_pr_debug("%8.8s src=%pI4@%-5d dst=%pI4@%-5d seq=%llu",
790 		      dccp_packet_name(dh->dccph_type),
791 		      &iph->saddr, ntohs(dh->dccph_sport),
792 		      &iph->daddr, ntohs(dh->dccph_dport),
793 		      (unsigned long long) DCCP_SKB_CB(skb)->dccpd_seq);
794 
795 	if (dccp_packet_without_ack(skb)) {
796 		DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ;
797 		dccp_pr_debug_cat("\n");
798 	} else {
799 		DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb);
800 		dccp_pr_debug_cat(", ack=%llu\n", (unsigned long long)
801 				  DCCP_SKB_CB(skb)->dccpd_ack_seq);
802 	}
803 
804 lookup:
805 	sk = __inet_lookup_skb(&dccp_hashinfo, skb, __dccp_hdr_len(dh),
806 			       dh->dccph_sport, dh->dccph_dport, &refcounted);
807 	if (!sk) {
808 		dccp_pr_debug("failed to look up flow ID in table and "
809 			      "get corresponding socket\n");
810 		goto no_dccp_socket;
811 	}
812 
813 	/*
814 	 * Step 2:
815 	 *	... or S.state == TIMEWAIT,
816 	 *		Generate Reset(No Connection) unless P.type == Reset
817 	 *		Drop packet and return
818 	 */
819 	if (sk->sk_state == DCCP_TIME_WAIT) {
820 		dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: do_time_wait\n");
821 		inet_twsk_put(inet_twsk(sk));
822 		goto no_dccp_socket;
823 	}
824 
825 	if (sk->sk_state == DCCP_NEW_SYN_RECV) {
826 		struct request_sock *req = inet_reqsk(sk);
827 		struct sock *nsk;
828 
829 		sk = req->rsk_listener;
830 		if (unlikely(sk->sk_state != DCCP_LISTEN)) {
831 			inet_csk_reqsk_queue_drop_and_put(sk, req);
832 			goto lookup;
833 		}
834 		sock_hold(sk);
835 		refcounted = true;
836 		nsk = dccp_check_req(sk, skb, req);
837 		if (!nsk) {
838 			reqsk_put(req);
839 			goto discard_and_relse;
840 		}
841 		if (nsk == sk) {
842 			reqsk_put(req);
843 		} else if (dccp_child_process(sk, nsk, skb)) {
844 			dccp_v4_ctl_send_reset(sk, skb);
845 			goto discard_and_relse;
846 		} else {
847 			sock_put(sk);
848 			return 0;
849 		}
850 	}
851 	/*
852 	 * RFC 4340, sec. 9.2.1: Minimum Checksum Coverage
853 	 *	o if MinCsCov = 0, only packets with CsCov = 0 are accepted
854 	 *	o if MinCsCov > 0, also accept packets with CsCov >= MinCsCov
855 	 */
856 	min_cov = dccp_sk(sk)->dccps_pcrlen;
857 	if (dh->dccph_cscov && (min_cov == 0 || dh->dccph_cscov < min_cov))  {
858 		dccp_pr_debug("Packet CsCov %d does not satisfy MinCsCov %d\n",
859 			      dh->dccph_cscov, min_cov);
860 		/* FIXME: "Such packets SHOULD be reported using Data Dropped
861 		 *         options (Section 11.7) with Drop Code 0, Protocol
862 		 *         Constraints."                                     */
863 		goto discard_and_relse;
864 	}
865 
866 	if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
867 		goto discard_and_relse;
868 	nf_reset(skb);
869 
870 	return __sk_receive_skb(sk, skb, 1, dh->dccph_doff * 4, refcounted);
871 
872 no_dccp_socket:
873 	if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
874 		goto discard_it;
875 	/*
876 	 * Step 2:
877 	 *	If no socket ...
878 	 *		Generate Reset(No Connection) unless P.type == Reset
879 	 *		Drop packet and return
880 	 */
881 	if (dh->dccph_type != DCCP_PKT_RESET) {
882 		DCCP_SKB_CB(skb)->dccpd_reset_code =
883 					DCCP_RESET_CODE_NO_CONNECTION;
884 		dccp_v4_ctl_send_reset(sk, skb);
885 	}
886 
887 discard_it:
888 	kfree_skb(skb);
889 	return 0;
890 
891 discard_and_relse:
892 	if (refcounted)
893 		sock_put(sk);
894 	goto discard_it;
895 }
896 
897 static const struct inet_connection_sock_af_ops dccp_ipv4_af_ops = {
898 	.queue_xmit	   = ip_queue_xmit,
899 	.send_check	   = dccp_v4_send_check,
900 	.rebuild_header	   = inet_sk_rebuild_header,
901 	.conn_request	   = dccp_v4_conn_request,
902 	.syn_recv_sock	   = dccp_v4_request_recv_sock,
903 	.net_header_len	   = sizeof(struct iphdr),
904 	.setsockopt	   = ip_setsockopt,
905 	.getsockopt	   = ip_getsockopt,
906 	.addr2sockaddr	   = inet_csk_addr2sockaddr,
907 	.sockaddr_len	   = sizeof(struct sockaddr_in),
908 #ifdef CONFIG_COMPAT
909 	.compat_setsockopt = compat_ip_setsockopt,
910 	.compat_getsockopt = compat_ip_getsockopt,
911 #endif
912 };
913 
914 static int dccp_v4_init_sock(struct sock *sk)
915 {
916 	static __u8 dccp_v4_ctl_sock_initialized;
917 	int err = dccp_init_sock(sk, dccp_v4_ctl_sock_initialized);
918 
919 	if (err == 0) {
920 		if (unlikely(!dccp_v4_ctl_sock_initialized))
921 			dccp_v4_ctl_sock_initialized = 1;
922 		inet_csk(sk)->icsk_af_ops = &dccp_ipv4_af_ops;
923 	}
924 
925 	return err;
926 }
927 
928 static struct timewait_sock_ops dccp_timewait_sock_ops = {
929 	.twsk_obj_size	= sizeof(struct inet_timewait_sock),
930 };
931 
932 static struct proto dccp_v4_prot = {
933 	.name			= "DCCP",
934 	.owner			= THIS_MODULE,
935 	.close			= dccp_close,
936 	.connect		= dccp_v4_connect,
937 	.disconnect		= dccp_disconnect,
938 	.ioctl			= dccp_ioctl,
939 	.init			= dccp_v4_init_sock,
940 	.setsockopt		= dccp_setsockopt,
941 	.getsockopt		= dccp_getsockopt,
942 	.sendmsg		= dccp_sendmsg,
943 	.recvmsg		= dccp_recvmsg,
944 	.backlog_rcv		= dccp_v4_do_rcv,
945 	.hash			= inet_hash,
946 	.unhash			= inet_unhash,
947 	.accept			= inet_csk_accept,
948 	.get_port		= inet_csk_get_port,
949 	.shutdown		= dccp_shutdown,
950 	.destroy		= dccp_destroy_sock,
951 	.orphan_count		= &dccp_orphan_count,
952 	.max_header		= MAX_DCCP_HEADER,
953 	.obj_size		= sizeof(struct dccp_sock),
954 	.slab_flags		= SLAB_TYPESAFE_BY_RCU,
955 	.rsk_prot		= &dccp_request_sock_ops,
956 	.twsk_prot		= &dccp_timewait_sock_ops,
957 	.h.hashinfo		= &dccp_hashinfo,
958 #ifdef CONFIG_COMPAT
959 	.compat_setsockopt	= compat_dccp_setsockopt,
960 	.compat_getsockopt	= compat_dccp_getsockopt,
961 #endif
962 };
963 
964 static const struct net_protocol dccp_v4_protocol = {
965 	.handler	= dccp_v4_rcv,
966 	.err_handler	= dccp_v4_err,
967 	.no_policy	= 1,
968 	.netns_ok	= 1,
969 	.icmp_strict_tag_validation = 1,
970 };
971 
972 static const struct proto_ops inet_dccp_ops = {
973 	.family		   = PF_INET,
974 	.owner		   = THIS_MODULE,
975 	.release	   = inet_release,
976 	.bind		   = inet_bind,
977 	.connect	   = inet_stream_connect,
978 	.socketpair	   = sock_no_socketpair,
979 	.accept		   = inet_accept,
980 	.getname	   = inet_getname,
981 	/* FIXME: work on tcp_poll to rename it to inet_csk_poll */
982 	.poll		   = dccp_poll,
983 	.ioctl		   = inet_ioctl,
984 	/* FIXME: work on inet_listen to rename it to sock_common_listen */
985 	.listen		   = inet_dccp_listen,
986 	.shutdown	   = inet_shutdown,
987 	.setsockopt	   = sock_common_setsockopt,
988 	.getsockopt	   = sock_common_getsockopt,
989 	.sendmsg	   = inet_sendmsg,
990 	.recvmsg	   = sock_common_recvmsg,
991 	.mmap		   = sock_no_mmap,
992 	.sendpage	   = sock_no_sendpage,
993 #ifdef CONFIG_COMPAT
994 	.compat_setsockopt = compat_sock_common_setsockopt,
995 	.compat_getsockopt = compat_sock_common_getsockopt,
996 #endif
997 };
998 
999 static struct inet_protosw dccp_v4_protosw = {
1000 	.type		= SOCK_DCCP,
1001 	.protocol	= IPPROTO_DCCP,
1002 	.prot		= &dccp_v4_prot,
1003 	.ops		= &inet_dccp_ops,
1004 	.flags		= INET_PROTOSW_ICSK,
1005 };
1006 
1007 static int __net_init dccp_v4_init_net(struct net *net)
1008 {
1009 	if (dccp_hashinfo.bhash == NULL)
1010 		return -ESOCKTNOSUPPORT;
1011 
1012 	return inet_ctl_sock_create(&net->dccp.v4_ctl_sk, PF_INET,
1013 				    SOCK_DCCP, IPPROTO_DCCP, net);
1014 }
1015 
1016 static void __net_exit dccp_v4_exit_net(struct net *net)
1017 {
1018 	inet_ctl_sock_destroy(net->dccp.v4_ctl_sk);
1019 }
1020 
1021 static void __net_exit dccp_v4_exit_batch(struct list_head *net_exit_list)
1022 {
1023 	inet_twsk_purge(&dccp_hashinfo, AF_INET);
1024 }
1025 
1026 static struct pernet_operations dccp_v4_ops = {
1027 	.init	= dccp_v4_init_net,
1028 	.exit	= dccp_v4_exit_net,
1029 	.exit_batch = dccp_v4_exit_batch,
1030 };
1031 
1032 static int __init dccp_v4_init(void)
1033 {
1034 	int err = proto_register(&dccp_v4_prot, 1);
1035 
1036 	if (err)
1037 		goto out;
1038 
1039 	inet_register_protosw(&dccp_v4_protosw);
1040 
1041 	err = register_pernet_subsys(&dccp_v4_ops);
1042 	if (err)
1043 		goto out_destroy_ctl_sock;
1044 
1045 	err = inet_add_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1046 	if (err)
1047 		goto out_proto_unregister;
1048 
1049 out:
1050 	return err;
1051 out_proto_unregister:
1052 	unregister_pernet_subsys(&dccp_v4_ops);
1053 out_destroy_ctl_sock:
1054 	inet_unregister_protosw(&dccp_v4_protosw);
1055 	proto_unregister(&dccp_v4_prot);
1056 	goto out;
1057 }
1058 
1059 static void __exit dccp_v4_exit(void)
1060 {
1061 	inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1062 	unregister_pernet_subsys(&dccp_v4_ops);
1063 	inet_unregister_protosw(&dccp_v4_protosw);
1064 	proto_unregister(&dccp_v4_prot);
1065 }
1066 
1067 module_init(dccp_v4_init);
1068 module_exit(dccp_v4_exit);
1069 
1070 /*
1071  * __stringify doesn't likes enums, so use SOCK_DCCP (6) and IPPROTO_DCCP (33)
1072  * values directly, Also cover the case where the protocol is not specified,
1073  * i.e. net-pf-PF_INET-proto-0-type-SOCK_DCCP
1074  */
1075 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 33, 6);
1076 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 0, 6);
1077 MODULE_LICENSE("GPL");
1078 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@mandriva.com>");
1079 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");
1080